text
stringlengths 4
2.78M
| meta
dict |
|---|---|
---
abstract: |
This document contains additional information for the paper: J. Quer. [*Fields of definition of building blocks*]{}, Math. Comp. [**78**]{} (2009), no. 265, pp. 537–554.
It contains a description of the implementation in [Magma]{} of the algorithms and computational techniques described in that paper, and also several tables with data that were elaborated using this implementation.
author:
- Jordi Quer
date: 'May, 31 2006'
title: |
Package description and tables for the paper\
Fields of definition of building blocks
---
Background: definitions and notation
====================================
Let $f=\sum_{n=1}^\infty a_n q^n\in S_2^{\operatorname{new}}(N,\varepsilon)$ be a newform of weight $2$, level $N$ and Nebentypus $\varepsilon$. Let $$E=\Q(\{a_n\}_{n\geq 1})\qquad\text{and}\qquad F=\Q(\{a_p^2/\varepsilon(p)\}_{p\nmid N})$$ be the number fields generated, respectively, by all the Fourier coefficients of $f$ and by the numbers $$\mu_p=a_p^2/\varepsilon(p),$$ for all primes $p$ not dividing the level $N$.
The abelian variety $A_f/\Q$ attached by Shimura to $f$ is a variety of $\GL_2$-type with $\End^0(A_f)\simeq E$; the isomorphism is given by the identification of (the restriction to $A_f$ of) the Hecke operator $T_p$ acting on the jacobian $J_1(N)$ with the algebraic integer $a_p$. The variety $A_f$ factors up to isogeny over $\Qb$ as a power $B_f^r$ of a simple variety $B_f$.
The variety $A_f$ has complex multiplication if, and only if, $B_f$ is a CM elliptic curve. This happens exactly when the form $f$ has complex multiplication by a quadratic field, meaning that there exists a nontrivial primitive Dirichlet character $\chi$ (which is necessarily unique, quadratic and odd) such that $$a_p=\chi(p)\,a_p\qquad\text{for all}\quad p\nmid N.$$ In that case, the curve $B_f$ is just the curve with complex multiplication by (an order of) that quadratic field (which is unique up to isogeny).
From now on we assume that $f$ has no complex multiplication; in this case the varieties $B_f$ are called *building blocks*. Now, the field $F$ is totally real and the field $E$ is an abelian extension of it that is either a totally real field when $\varepsilon$ is trivial (which is equivalent to $A_f$ being a factor of $J_0(N)$ up to $\Q$-isogeny), or a CM field otherwise. The endomorphism algebra $\D=\End^0(B_f)$ is a division algebra; it can be either equal to the totally real field $F$, with $\dim(B_f)=[F:\Q]$ (and $r=[E:F]$ in this case) or to a quaternion algebra over the totally real field $F$ (which is necessarily totally indefinite), with $\dim(B_f)=2[F:\Q]$ (and $r=\frac12[E:F]$ in this case).
For every element $s\in G_F$ there exists a unique primitive Dirichlet character $\chi_s$, which only depends on the action of $s$ on the field $E$, such that $$\label{definition-phi}
{}^s a_p =\chi_s(p)\, a_p\qquad\text{for all}\quad p\nmid N.$$ The character $\chi_s$, and sometimes also the pair $(s,\chi_s)$, is called an *inner twist* of the form $f$.
Let $F_\delta=F(\{\sqrt{\mu_p}\}_{p\nmid N})$ be the extension of $F$ generated by the square roots of the elements $\mu_p\in F$, which is a polyquadratic extension of $F$. For every element $s\in G_F$ there exists a unique primitive quadratic Dirichlet character $\psi_s$, which only depends on the action of $s$ on the field $F_\delta$, such that $$\label{definition-psi}
{}^s \sqrt{\mu_p}=\psi_s(p)\,\sqrt{\mu_p}\qquad\text{for all}\quad p\nmid N.$$ The character $\psi_s$, and also the pair $(s,\psi_s)$, will be called a *quadratic degree character* of the form $f$. The group $\Psi$ of all the quadratic degree characters is an abelian group of exponent $2$, isomorphic to $\Gal(F_\delta/F)$. It depends only on the $\Qb$-isogeny class of the building block $B_f$.
The inner twists and quadratic degree characters are related by the following identity $$\label{relation-chi-psi}
\chi_s(p)=\psi_s(p)\sqrt{\varepsilon(p)}^{s-1}\qquad\text{for all}\quad p\nmid N.$$ We observe that since the $\chi_s$ depend only on $s|_E$ and the $\psi_s$ depend on $s|_{F_\delta}$, everything in this identity depends only on the restriction of $s$ to the field $E\cdot F_\delta=E(\sqrt{\varepsilon})$, where $E(\sqrt{\varepsilon})$ denotes the quadratic field obtained by adjoining to $E$ the square roots of the values of the character $\varepsilon$, which is either equal to $E$ or to a quadratic extension of it. Using this relation we see that the quadratic degree characters can easily be computed from the knowledge of the inner twists and viceversa.
Let $K_P$ denote the polyquadratic extension of $\Q$ which is the fixed field of the intersection of all the quadratic degree characters. It coincides with the kernel of the map $\delta\colon G_\Q\to F^*/F^{*2}$ defined by putting $\delta(\Frob_p)=\mu_p\pmod{F^{*2}}$ for all primes $p\nmid N$ with $a_p\neq0$ (it is well defined by this condition because the primes $p$ with $a_p\neq0$ are of density one in the set of all prime numbers for non-CM newforms, and hence every element of $\Gal(K_P/\Q)$ is a Frobenius for some prime $p$ with $a_p\neq0$).
Let $\sigma_1,\dots,\sigma_r\in G_\Q$ be elements whose restriction to the field $K_P$ is a basis of the Galois group $\Gal(K_P/\Q)$, and let $\psi_1,\dots,\psi_r\colon G_\Q\to\Z/2\Z$ be (additive versions of) quadratic degree characters that are a basis of $\Psi$, and assume that these bases are dual of each other, i.e. such that $$\psi_i(\sigma_j)=0\quad\text{for}\quad j\neq i\qquad\text{and}\qquad
\psi_i(\sigma_i)=1,$$ for every $1\leq i,j\leq r$. Let $t_i\in\Q^*$ be rational numbers such that $$\Q(\sqrt{t_i})=\Qb^{\ker\psi_i}$$ and let $$\delta_i=\delta(\sigma_i)\in F^*$$ for $i=1,\dots,r$.
We denote by $\gamma_\varepsilon$ the element of the two-torsion $\Br(\Q)[2]\simeq H^2(G_\Q,\{\pm1\})$ of the Brauer group of the rational numbers that is the cohomology class of the two-cocycle defined by the formula $$(\sigma,\tau)\mapsto\sqrt{\varepsilon(\sigma)}\,\sqrt{\varepsilon(\tau)}\,
{\sqrt{\varepsilon(\sigma\tau)}\,}^{-1}.$$
Then, the Brauer class of $\D=\End^0(B_f)$ is $$\label{brauer-class}
[\D]=\Res_{\Q}^F(\gamma_\varepsilon)\prod(t_i,\delta_i)\in\Br(F)[2],$$ and the obstruction to descend the building block over the field $K_P$ is $$\label{obstruction}
\Res_{\Q}^{K_P}(\gamma_\varepsilon)\in\Br(K_P)[2].$$
The package
===========
The functions have been implemented in [Magma]{}, and are based in the packages written by W. Stein for computing with modular symbols, modular forms, and modular abelian varieties. In the following we assume that the reader is familiar with [Magma]{} and especially with the functions of Stein’s packages.
One of the main practical problems in determining the inner twists of a modular form (which in practice is given by a finite truncation of its Fourier series) is to check the identity (\[definition-psi\]) for sufficiently many primes $p$ to guarantee that it holds for all primes. Since the coefficients for primes smaller than $\frac{k}{12}\psi(N)$ completely determine a cusp form of level $N$ and weight $k$, with $\psi$ being the function $\psi(n)=n\prod_{p\mid n}(1+1/p)$, and the twist of a modular form by a Dirichlet character has level dividing the least common multiple of the level of the form and the square of the conductor of the character, we see that it is enough to use the bound $\frac k6\psi(N^2)$. In practice this number becomes too large for computations even for relatively small levels, and must be replaced by smaller bounds that should be enough, but for which there is no proof. The functions computing the inner twists in W. Stein’s package have an optional parameter to decide which bound to use. In our implementation we use the large proved bound for levels $N\leq100$ and we use the unproved bound $15+N/2$ for larger levels. In particular we introduced small modifications in W. Stein’s file [inner\_twist.m]{}.
All the intrinsics we programmed are in a file named [building\_blocks.m]{}; for using them, just copy that file into the [Magma]{} packages directory [ModAbVar]{}, and add the name of the file to [ModAbVar.spec]{}. It seems that all these intrinsics will be part of the standard distribution of [Magma]{} in the future. The intrinsics implemented are the following:
- [DegreeMap.]{} Its input is a space of modular symbols that is new and irreducible, corresponding to a modular form $f$. The output consists of a sequence (possibly empty) of pairs $(t_i,\delta_i)$, with the $t_i$ being nonzero rational numbers only determined modulo squares, that in fact are normalised to be rational integers that are discriminants of quadratic fields, and the $\delta_i$ are nonzero numbers of a number field $F$ only determined modulo squares. These numbers represent the degree map $\delta$ and using them everything else can easily be computed.
- [BrauerClass.]{} Its input is a space of modular symbols that is new and irreducible. The output is the Brauer class of the endomorphism algebra $\D=\End^0(B_f)$, or equivalently, that of $\End^0(A_f)$, as an element of $\Br(F)[2]$, computed using the formula (\[brauer-class\]). It is given as a sequence with an even number (possibly zero) of the places of the field $F$, (necessarily finite in weight $2$) in which that quaternion algebra is ramified. Thus, the variety $B_f$ is a RM-building block (real multiplication) with $\End^0(B_f)=F$ when this element is trivial, in which case $A_f\sim B_f^{[E:F]}$ and $\dim B_f=[F:\Q]$; when this element is nontrivial, then $B_f$ is a QM-building block (quaternionic multiplication) with $\End^0(B_f)=\D$ a quaternion division algebra over $F$, in which case $A_f=B_f^{[E:F]/2}$ and $\dim B_f=2[F:\Q]$.
- [ObstructionDescentBuildingBlock.]{} Its input is a space of modular symbols that is new and irreducible. The output is the obstruction to the existence of a building block over the field $K_P$ that is isogenous to $B_f$. This obstruction is the element of $\Br(K_P)[2]$ given by (\[obstruction\]) and is given as the list of the places of the field for which the local component is nontrivial. When the list is empty then the field $K_P$ is the smallest possible such field; otherwise the fields having this property are the extensions of $K_P$ that are splitting fields for this element of the Brauer group, the smallest possibility being quadratic extensions.
- [BoundedFSubspace.]{} Its input is a Dirichlet character $\varepsilon$, a weight $k$, and a range of positive integers. The output is the list of modular symbols spaces of that Nebentypus and weight corresponding to non-CM newforms, and for which the corresponding field $F$ has degree over Q belonging by the numbers in the given range.
The computation avoids to split the full modular symbols space into newform subspaces, which is very time-consuming, by the computation of the characteristic polynomials of the operators $T_p^2/\varepsilon(p)$ for a few primes $p$ not dividing the level, and the restriction to the modular symbols subspaces that are in the kernel of the operators obtained from factors of this characteristic polynomial of small degree.
The table
=========
Using the functions described in the previous paragraph we elaborated a table containing information for all non-CM newforms $f$ of weight $2$, level up to $500$, Nebentypus character $\varepsilon$ of not too large order, and for which the field $F$ has degree over $\Q$ bounded by $4$. The restrictions introduced in the Nebentypus character are due to the fact that the Nebentypus modular symbols computations take place in the field generated by the values of the character, and consequently depend very much on the degree of that field. So we considered only those characters whose values generate a field of degree up to $12$.
In fact we are quite confident that our table contains all the newforms in the given level range $N\leq500$, with every Nebentypus character, and whose corresponding field $F$ has degree up to $4$, although for checking the nonexistence of such forms with such small $F$ for characters of larger orders, we replaced the computations of modular symbols over number fields, that were too slow and required too much memory to be performed, by computations based in modular symbols over finite fields and we do not have a full theoretical justification for this reduction process to be correct.
The total number of forms obtained is $5609$. We give now eight tables containing information for every such form, corresponding to the eight possible endomorphism algebras structures of $B_f$: either a totally real field of degree $1,2,3$ or $4$, or a quaternion division algebra over a totally real field of dimension $4,8,12$ or $16$ over $\Q$. For every form we give the following information: the level $N$, the Nebentypus character $\varepsilon$ (the format is the output of the function [Eltseq]{}) and its order; the dimension of the variety $A_f$, equal to the degree of the field $E$ over $\Q$; information of the field $F$ (when it is not $\Q$) given by a list of coefficients of a defining polynomial (when it is not quadratic) and its discriminant; information about the degree map, given by the numbers $t_i$ and the numbers $N_{F/\Q}(\delta_i)$; the Brauer class of $A_f$ (if it is nontrivial), given as a list of integers containing the norms over $\Q$ of the prime ideals of $F$ at which the corresponding quaternion algebra is ramified, and finally the obstruction to descend $B_f$ to the field $K_P$ (except in the RM-case of even dimension in which we know it is trivial by a theorem of Ribet), given in the same format but for the field $K_P$.
Tables of newforms corresponding to RM-building blocks
======================================================
$\Q$-curves
-----------
$N$ $\varepsilon$ $\operatorname{ord}(\varepsilon)$ $\dim(A_f)$ $[t_i]$ $[\delta_i]$
------- ------------------ ----------------------------------- ------------- -------------------- ---------------
$11$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$13$ $[ 2 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$14$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$15$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$16$ $[ 0, 1 ]$ $4$ $2$ $[ 8 ]$ $[ 2 ]$
$17$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$18$ $[ 2 ]$ $3$ $2$ $[]$ $[]$
$19$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$20$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$21$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$21$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$24$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$24$ $[ 0, 1, 0 ]$ $2$ $2$ $[]$ $[]$
$26$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$26$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$26$ $[ 6 ]$ $2$ $2$ $[]$ $[]$
$26$ $[ 4 ]$ $3$ $2$ $[]$ $[]$
$28$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$29$ $[ 14 ]$ $2$ $2$ $[ 29 ]$ $[ 5 ]$
$30$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$30$ $[ 0, 2 ]$ $2$ $2$ $[]$ $[]$
$30$ $[ 1, 1 ]$ $4$ $4$ $[ -15 ]$ $[ 2 ]$
$33$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$34$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$34$ $[ 8 ]$ $2$ $2$ $[ 17 ]$ $[ 2 ]$
$34$ $[ 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$34$ $[ 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$35$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$35$ $[ 2, 0 ]$ $2$ $2$ $[]$ $[]$
$35$ $[ 2, 2 ]$ $6$ $4$ $[]$ $[]$
$36$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$37$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$37$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$37$ $[ 18 ]$ $2$ $2$ $[]$ $[]$
$37$ $[ 12 ]$ $3$ $2$ $[]$ $[]$
$38$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$38$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$38$ $[ 6 ]$ $3$ $2$ $[]$ $[]$
$39$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$39$ $[ 0, 6 ]$ $2$ $2$ $[ 13 ]$ $[ 3 ]$
$39$ $[ 0, 4 ]$ $3$ $2$ $[]$ $[]$
$39$ $[ 1, 3 ]$ $4$ $4$ $[ -39 ]$ $[ 2 ]$
$39$ $[ 0, 2 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$40$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$40$ $[ 0, 0, 2 ]$ $2$ $2$ $[]$ $[]$
$40$ $[ 0, 1, 2 ]$ $2$ $4$ $[ 5, 40 ]$ $[ 2, 3 ]$
$41$ $[ 20 ]$ $2$ $2$ $[ 41 ]$ $[ 2 ]$
$42$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$42$ $[ 1, 3 ]$ $2$ $4$ $[ -7 ]$ $[ 6 ]$
$42$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$42$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$42$ $[ 1, 1 ]$ $6$ $4$ $[ -7 ]$ $[ 3 ]$
$43$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$43$ $[ 14 ]$ $3$ $2$ $[]$ $[]$
$44$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$45$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$45$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$45$ $[ 3, 1 ]$ $4$ $4$ $[ -15 ]$ $[ 2 ]$
$46$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$48$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$50$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$50$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$50$ $[ 10 ]$ $2$ $2$ $[]$ $[]$
$51$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$51$ $[ 0, 8 ]$ $2$ $2$ $[]$ $[]$
$51$ $[ 0, 8 ]$ $2$ $2$ $[]$ $[]$
$52$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$52$ $[ 0, 4 ]$ $3$ $2$ $[]$ $[]$
$52$ $[ 0, 4 ]$ $3$ $2$ $[]$ $[]$
$52$ $[ 0, 2 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$53$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$54$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$54$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$54$ $[ 6 ]$ $3$ $2$ $[]$ $[]$
$55$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$55$ $[ 1, 5 ]$ $4$ $4$ $[ -55, -11 ]$ $[ 2, 5 ]$
$56$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$56$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$56$ $[ 0, 1, 0 ]$ $2$ $2$ $[ 8 ]$ $[ 2 ]$
$56$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$56$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$57$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$57$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$57$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$57$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$58$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$58$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$58$ $[ 14 ]$ $2$ $2$ $[]$ $[]$
$60$ $[ 0, 0, 2 ]$ $2$ $2$ $[]$ $[]$
$61$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$61$ $[ 10 ]$ $6$ $2$ $[ 61 ]$ $[ 3 ]$
$62$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$62$ $[ 10 ]$ $3$ $2$ $[]$ $[]$
$62$ $[ 10 ]$ $3$ $2$ $[]$ $[]$
$63$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$63$ $[ 0, 0 ]$ $1$ $2$ $[ -3 ]$ $[ 3 ]$
$63$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$63$ $[ 2, 2 ]$ $3$ $2$ $[]$ $[]$
$63$ $[ 4, 2 ]$ $3$ $2$ $[]$ $[]$
$63$ $[ 1, 1 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$63$ $[ 5, 1 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$64$ $[ 0, 4 ]$ $4$ $2$ $[ 8 ]$ $[ 2 ]$
$65$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$65$ $[ 2, 6 ]$ $2$ $2$ $[]$ $[]$
$65$ $[ 2, 6 ]$ $2$ $2$ $[]$ $[]$
$65$ $[ 1, 3 ]$ $4$ $2$ $[ 65 ]$ $[ 2 ]$
$65$ $[ 3, 3 ]$ $4$ $2$ $[ 65 ]$ $[ 2 ]$
$66$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$66$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$66$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$66$ $[ 1, 5 ]$ $2$ $2$ $[ 33 ]$ $[ 2 ]$
$66$ $[ 1, 5 ]$ $2$ $2$ $[ 33 ]$ $[ 2 ]$
$67$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$68$ $[ 0, 8 ]$ $2$ $2$ $[ 17 ]$ $[ 2 ]$
$69$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$70$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$70$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$70$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$70$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$70$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$70$ $[ 2, 2 ]$ $6$ $4$ $[]$ $[]$
$70$ $[ 2, 2 ]$ $6$ $4$ $[]$ $[]$
$72$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$72$ $[ 0, 1, 0 ]$ $2$ $2$ $[]$ $[]$
$72$ $[ 1, 1, 3 ]$ $2$ $4$ $[ -3, -8 ]$ $[ 2, 3 ]$
$72$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$72$ $[ 0, 1, 2 ]$ $6$ $4$ $[]$ $[]$
$73$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$73$ $[ 18 ]$ $4$ $2$ $[ 73 ]$ $[ 2 ]$
$74$ $[ 12 ]$ $3$ $2$ $[]$ $[]$
$74$ $[ 12 ]$ $3$ $2$ $[]$ $[]$
$75$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$75$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$75$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$75$ $[ 0, 10 ]$ $2$ $2$ $[]$ $[]$
$75$ $[ 0, 10 ]$ $2$ $2$ $[]$ $[]$
$76$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$76$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$77$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$77$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$77$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$78$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$78$ $[ 0, 6 ]$ $2$ $2$ $[]$ $[]$
$78$ $[ 0, 4 ]$ $3$ $2$ $[]$ $[]$
$78$ $[ 0, 4 ]$ $3$ $2$ $[]$ $[]$
$79$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$80$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$80$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$80$ $[ 0, 0, 2 ]$ $2$ $2$ $[]$ $[]$
$80$ $[ 0, 1, 2 ]$ $4$ $2$ $[ 8 ]$ $[ 2 ]$
$80$ $[ 0, 1, 2 ]$ $4$ $2$ $[ 8 ]$ $[ 2 ]$
$80$ $[ 1, 1, 1 ]$ $4$ $2$ $[ 40 ]$ $[ 2 ]$
$80$ $[ 1, 3, 1 ]$ $4$ $2$ $[ 40 ]$ $[ 2 ]$
$81$ $[ 0 ]$ $1$ $2$ $[ -3 ]$ $[ 3 ]$
$81$ $[ 18 ]$ $3$ $4$ $[ -3 ]$ $[ 3 ]$
$82$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$82$ $[ 20 ]$ $2$ $2$ $[]$ $[]$
$82$ $[ 20 ]$ $2$ $2$ $[ 41 ]$ $[ 2 ]$
$82$ $[ 10 ]$ $4$ $2$ $[ 41 ]$ $[ 2 ]$
$82$ $[ 10 ]$ $4$ $2$ $[ 41 ]$ $[ 2 ]$
$82$ $[ 10 ]$ $4$ $2$ $[ 41 ]$ $[ 2 ]$
$83$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$84$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$84$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$84$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$84$ $[ 0, 1, 1 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$84$ $[ 0, 1, 1 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$85$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$85$ $[ 2, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$85$ $[ 2, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$86$ $[ 14 ]$ $3$ $2$ $[]$ $[]$
$88$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$89$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$89$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$90$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$90$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$90$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$90$ $[ 0, 2 ]$ $2$ $2$ $[]$ $[]$
$90$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$90$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$90$ $[ 3, 1 ]$ $4$ $4$ $[ -15 ]$ $[ 2 ]$
$90$ $[ 2, 2 ]$ $6$ $4$ $[]$ $[]$
$91$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$91$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$91$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$91$ $[ 2, 4 ]$ $3$ $2$ $[]$ $[]$
$91$ $[ 4, 4 ]$ $3$ $2$ $[]$ $[]$
$91$ $[ 2, 2 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$91$ $[ 4, 2 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$92$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$92$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$93$ $[ 1, 5 ]$ $6$ $2$ $[ 93 ]$ $[ 3 ]$
$93$ $[ 1, 5 ]$ $6$ $2$ $[ 93 ]$ $[ 3 ]$
$94$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$95$ $[ 2, 0 ]$ $2$ $2$ $[]$ $[]$
$95$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$95$ $[ 2, 6 ]$ $6$ $4$ $[]$ $[]$
$96$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$96$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$96$ $[ 0, 4, 0 ]$ $2$ $2$ $[]$ $[]$
$96$ $[ 1, 0, 1 ]$ $2$ $4$ $[ -3 ]$ $[ 2 ]$
$98$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$98$ $[ 0 ]$ $1$ $2$ $[ -7 ]$ $[ 2 ]$
$98$ $[ 14 ]$ $3$ $2$ $[]$ $[]$
$98$ $[ 14 ]$ $3$ $2$ $[]$ $[]$
$98$ $[ 14 ]$ $3$ $4$ $[ -7 ]$ $[ 2 ]$
$99$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$99$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$99$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$99$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$99$ $[ 3, 5 ]$ $2$ $4$ $[ 33, -11 ]$ $[ 2, 3 ]$
$99$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$99$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$99$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$100$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$100$ $[ 0, 10 ]$ $2$ $2$ $[]$ $[]$
$100$ $[ 1, 5 ]$ $4$ $8$ $[ -20, -4 ]$ $[ 3, 5 ]$
$101$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$102$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$102$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$102$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$102$ $[ 0, 8 ]$ $2$ $2$ $[]$ $[]$
$104$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$104$ $[ 0, 1, 0 ]$ $2$ $2$ $[]$ $[]$
$104$ $[ 0, 1, 6 ]$ $2$ $2$ $[]$ $[]$
$104$ $[ 0, 1, 6 ]$ $2$ $2$ $[]$ $[]$
$104$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$105$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$105$ $[ 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$105$ $[ 1, 0, 3 ]$ $2$ $2$ $[ 21 ]$ $[ 3 ]$
$105$ $[ 1, 0, 3 ]$ $2$ $2$ $[ 21 ]$ $[ 3 ]$
$105$ $[ 1, 2, 3 ]$ $2$ $4$ $[ 105, -35 ]$ $[ 2, 3 ]$
$105$ $[ 1, 2, 3 ]$ $2$ $4$ $[ 105, -35 ]$ $[ 2, 3 ]$
$105$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$105$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$105$ $[ 1, 0, 1 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$105$ $[ 1, 0, 1 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$106$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$106$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$106$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$106$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$106$ $[ 26 ]$ $2$ $2$ $[]$ $[]$
$106$ $[ 26 ]$ $2$ $2$ $[]$ $[]$
$108$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$109$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$109$ $[ 54 ]$ $2$ $2$ $[ 109 ]$ $[ 3 ]$
$109$ $[ 18 ]$ $6$ $2$ $[ 109 ]$ $[ 3 ]$
$110$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$110$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$110$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$110$ $[ 2, 0 ]$ $2$ $2$ $[]$ $[]$
$110$ $[ 2, 0 ]$ $2$ $2$ $[]$ $[]$
$110$ $[ 2, 0 ]$ $2$ $2$ $[]$ $[]$
$110$ $[ 1, 5 ]$ $4$ $4$ $[ -55 ]$ $[ 2 ]$
$111$ $[ 0, 18 ]$ $2$ $2$ $[]$ $[]$
$111$ $[ 0, 6 ]$ $6$ $2$ $[ 37 ]$ $[ 3 ]$
$112$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$112$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$112$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$112$ $[ 1, 0, 3 ]$ $2$ $2$ $[ 28 ]$ $[ 3 ]$
$112$ $[ 1, 0, 3 ]$ $2$ $2$ $[ 28 ]$ $[ 3 ]$
$112$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$112$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$112$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$112$ $[ 0, 1, 0 ]$ $4$ $2$ $[ 8 ]$ $[ 2 ]$
$112$ $[ 0, 1, 0 ]$ $4$ $2$ $[ 8 ]$ $[ 2 ]$
$112$ $[ 1, 0, 1 ]$ $6$ $2$ $[ 28 ]$ $[ 3 ]$
$112$ $[ 1, 0, 1 ]$ $6$ $2$ $[ 28 ]$ $[ 3 ]$
$113$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$113$ $[ 28 ]$ $4$ $2$ $[ 113 ]$ $[ 2 ]$
$114$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$114$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$114$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$114$ $[ 1, 9 ]$ $2$ $2$ $[ 57 ]$ $[ 2 ]$
$114$ $[ 1, 9 ]$ $2$ $2$ $[ 57 ]$ $[ 3 ]$
$114$ $[ 1, 9 ]$ $2$ $2$ $[ 57 ]$ $[ 2 ]$
$114$ $[ 1, 9 ]$ $2$ $2$ $[ 57 ]$ $[ 3 ]$
$114$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$114$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$114$ $[ 1, 3 ]$ $6$ $2$ $[ 57 ]$ $[ 3 ]$
$114$ $[ 1, 3 ]$ $6$ $2$ $[ 57 ]$ $[ 3 ]$
$114$ $[ 1, 3 ]$ $6$ $2$ $[ 57 ]$ $[ 3 ]$
$114$ $[ 1, 3 ]$ $6$ $2$ $[ 57 ]$ $[ 3 ]$
$115$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$115$ $[ 2, 0 ]$ $2$ $2$ $[]$ $[]$
$116$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$116$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$116$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$116$ $[ 0, 14 ]$ $2$ $2$ $[ 29 ]$ $[ 7 ]$
$117$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$117$ $[ 0, 0 ]$ $1$ $2$ $[ -3 ]$ $[ 3 ]$
$117$ $[ 0, 6 ]$ $2$ $2$ $[ 13 ]$ $[ 3 ]$
$117$ $[ 0, 4 ]$ $3$ $2$ $[]$ $[]$
$117$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$117$ $[ 0, 2 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$117$ $[ 0, 2 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$117$ $[ 2, 2 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$117$ $[ 2, 6 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$117$ $[ 2, 6 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$117$ $[ 4, 2 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$118$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$118$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$118$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$118$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$120$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$120$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$120$ $[ 0, 0, 0, 2 ]$ $2$ $2$ $[]$ $[]$
$120$ $[ 0, 1, 0, 0 ]$ $2$ $2$ $[]$ $[]$
$120$ $[ 0, 0, 1, 1 ]$ $4$ $4$ $[ -15 ]$ $[ 2 ]$
$121$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$121$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$121$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$121$ $[ 22 ]$ $5$ $4$ $[]$ $[]$
$121$ $[ 22 ]$ $5$ $4$ $[]$ $[]$
$121$ $[ 22 ]$ $5$ $4$ $[]$ $[]$
$121$ $[ 22 ]$ $5$ $4$ $[]$ $[]$
$122$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$122$ $[ 30 ]$ $2$ $2$ $[]$ $[]$
$123$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$123$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$124$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$124$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$126$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$126$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$126$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$126$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$126$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$126$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$126$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$126$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$126$ $[ 2, 2 ]$ $3$ $2$ $[]$ $[]$
$126$ $[ 2, 2 ]$ $3$ $2$ $[]$ $[]$
$126$ $[ 4, 2 ]$ $3$ $2$ $[]$ $[]$
$126$ $[ 4, 2 ]$ $3$ $2$ $[]$ $[]$
$128$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$128$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$128$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$128$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$128$ $[ 0, 8 ]$ $4$ $2$ $[ 8 ]$ $[ 2 ]$
$128$ $[ 0, 8 ]$ $4$ $2$ $[ 8 ]$ $[ 2 ]$
$129$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$129$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$129$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$130$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$130$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$130$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$130$ $[ 0, 6 ]$ $2$ $2$ $[]$ $[]$
$130$ $[ 0, 4 ]$ $3$ $2$ $[]$ $[]$
$130$ $[ 0, 4 ]$ $3$ $2$ $[]$ $[]$
$130$ $[ 1, 3 ]$ $4$ $2$ $[ 65 ]$ $[ 2 ]$
$130$ $[ 1, 3 ]$ $4$ $2$ $[ 65 ]$ $[ 2 ]$
$130$ $[ 1, 3 ]$ $4$ $2$ $[ 65 ]$ $[ 2 ]$
$130$ $[ 3, 3 ]$ $4$ $2$ $[ 65 ]$ $[ 2 ]$
$130$ $[ 3, 3 ]$ $4$ $2$ $[ 65 ]$ $[ 2 ]$
$130$ $[ 3, 3 ]$ $4$ $2$ $[ 65 ]$ $[ 2 ]$
$131$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$132$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$132$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$132$ $[ 1, 1, 0 ]$ $2$ $2$ $[ 12 ]$ $[ 2 ]$
$132$ $[ 1, 1, 0 ]$ $2$ $2$ $[ 12 ]$ $[ 2 ]$
$133$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$133$ $[ 1, 3 ]$ $6$ $2$ $[ 133 ]$ $[ 3 ]$
$133$ $[ 1, 3 ]$ $6$ $2$ $[ 133 ]$ $[ 3 ]$
$133$ $[ 5, 3 ]$ $6$ $2$ $[ 133 ]$ $[ 3 ]$
$133$ $[ 5, 3 ]$ $6$ $2$ $[ 133 ]$ $[ 3 ]$
$134$ $[ 22 ]$ $3$ $2$ $[]$ $[]$
$135$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$135$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$135$ $[ 6, 0 ]$ $3$ $2$ $[]$ $[]$
$136$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$136$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$136$ $[ 0, 0, 8 ]$ $2$ $2$ $[ 17 ]$ $[ 2 ]$
$136$ $[ 0, 0, 8 ]$ $2$ $2$ $[]$ $[]$
$136$ $[ 0, 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$136$ $[ 0, 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$136$ $[ 0, 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$136$ $[ 0, 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$138$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$138$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$138$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$139$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$140$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$140$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$140$ $[ 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$140$ $[ 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$140$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$140$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$141$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$141$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$141$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$141$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$141$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$142$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$142$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$142$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$142$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$142$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$143$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$144$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$144$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$144$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$144$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$144$ $[ 0, 1, 0 ]$ $4$ $2$ $[ 8 ]$ $[ 2 ]$
$144$ $[ 1, 0, 1 ]$ $6$ $2$ $[ 12 ]$ $[ 3 ]$
$144$ $[ 1, 0, 1 ]$ $6$ $2$ $[ 12 ]$ $[ 3 ]$
$144$ $[ 1, 0, 1 ]$ $6$ $2$ $[ 12 ]$ $[ 3 ]$
$144$ $[ 1, 0, 1 ]$ $6$ $2$ $[ 12 ]$ $[ 3 ]$
$145$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$145$ $[ 2, 14 ]$ $2$ $4$ $[ 29 ]$ $[ 5 ]$
$146$ $[ 36 ]$ $2$ $2$ $[ 73 ]$ $[ 6 ]$
$146$ $[ 36 ]$ $2$ $2$ $[ 73 ]$ $[ 2 ]$
$146$ $[ 18 ]$ $4$ $2$ $[ 73 ]$ $[ 2 ]$
$146$ $[ 12 ]$ $6$ $2$ $[ 73 ]$ $[ 3 ]$
$147$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$147$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$147$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$147$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$147$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$147$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$148$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$148$ $[ 0, 18 ]$ $2$ $2$ $[ 37 ]$ $[ 3 ]$
$148$ $[ 0, 6 ]$ $6$ $2$ $[ 37 ]$ $[ 3 ]$
$148$ $[ 0, 6 ]$ $6$ $2$ $[ 37 ]$ $[ 3 ]$
$150$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$150$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$150$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$150$ $[ 0, 10 ]$ $2$ $2$ $[]$ $[]$
$150$ $[ 1, 5 ]$ $4$ $4$ $[ -15 ]$ $[ 2 ]$
$150$ $[ 1, 5 ]$ $4$ $8$ $[ -15 ]$ $[ 3 ]$
$152$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$152$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$152$ $[ 0, 1, 0 ]$ $2$ $2$ $[]$ $[]$
$152$ $[ 0, 0, 6 ]$ $3$ $2$ $[]$ $[]$
$152$ $[ 0, 0, 6 ]$ $3$ $2$ $[]$ $[]$
$153$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$153$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$153$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$153$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$153$ $[ 0, 8 ]$ $2$ $2$ $[]$ $[]$
$153$ $[ 0, 8 ]$ $2$ $2$ $[]$ $[]$
$154$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$154$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$154$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$154$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$154$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$154$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$154$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$155$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$155$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$155$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$155$ $[ 0, 10 ]$ $3$ $2$ $[]$ $[]$
$155$ $[ 0, 10 ]$ $3$ $2$ $[]$ $[]$
$156$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$156$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$156$ $[ 0, 0, 6 ]$ $2$ $2$ $[]$ $[]$
$156$ $[ 0, 0, 6 ]$ $2$ $2$ $[ 13 ]$ $[ 3 ]$
$156$ $[ 1, 1, 0 ]$ $2$ $2$ $[ 12 ]$ $[ 2 ]$
$156$ $[ 1, 1, 0 ]$ $2$ $2$ $[ 12 ]$ $[ 2 ]$
$156$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$156$ $[ 1, 0, 3 ]$ $4$ $2$ $[ 13 ]$ $[ 2 ]$
$156$ $[ 1, 0, 3 ]$ $4$ $2$ $[ 13 ]$ $[ 2 ]$
$156$ $[ 1, 0, 3 ]$ $4$ $2$ $[ 13 ]$ $[ 2 ]$
$156$ $[ 1, 0, 3 ]$ $4$ $2$ $[ 13 ]$ $[ 2 ]$
$156$ $[ 0, 0, 2 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$157$ $[ 78 ]$ $2$ $2$ $[]$ $[]$
$157$ $[ 26 ]$ $6$ $2$ $[ 157 ]$ $[ 3 ]$
$158$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$158$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$158$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$158$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$158$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$158$ $[ 26 ]$ $3$ $2$ $[]$ $[]$
$158$ $[ 26 ]$ $3$ $2$ $[]$ $[]$
$160$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$160$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$160$ $[ 0, 0, 0 ]$ $1$ $2$ $[ -4 ]$ $[ 2 ]$
$160$ $[ 0, 4, 2 ]$ $2$ $4$ $[ 5, 40 ]$ $[ 2, 3 ]$
$160$ $[ 1, 0, 1 ]$ $4$ $2$ $[ 5 ]$ $[ 2 ]$
$160$ $[ 1, 0, 1 ]$ $4$ $2$ $[ 5 ]$ $[ 2 ]$
$160$ $[ 1, 0, 1 ]$ $4$ $2$ $[ 5 ]$ $[ 2 ]$
$160$ $[ 1, 0, 1 ]$ $4$ $2$ $[ 5 ]$ $[ 2 ]$
$160$ $[ 1, 0, 1 ]$ $4$ $2$ $[ 5 ]$ $[ 2 ]$
$160$ $[ 1, 0, 1 ]$ $4$ $2$ $[ 5 ]$ $[ 2 ]$
$161$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$162$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$162$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$162$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$162$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$162$ $[ 18 ]$ $3$ $2$ $[]$ $[]$
$162$ $[ 18 ]$ $3$ $2$ $[]$ $[]$
$162$ $[ 18 ]$ $3$ $2$ $[]$ $[]$
$162$ $[ 18 ]$ $3$ $2$ $[]$ $[]$
$163$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$166$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$168$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$168$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$168$ $[ 0, 1, 1, 3 ]$ $2$ $4$ $[ -3, -56 ]$ $[ 2, 3 ]$
$168$ $[ 0, 1, 1, 3 ]$ $2$ $4$ $[ -3, -56 ]$ $[ 2, 3 ]$
$168$ $[ 0, 1, 1, 3 ]$ $2$ $8$ $[ -7, -56, -24 ]$ $[ 2, 3, 5 ]$
$168$ $[ 1, 1, 1, 0 ]$ $2$ $4$ $[ -3 ]$ $[ 2 ]$
$168$ $[ 0, 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$168$ $[ 0, 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$169$ $[ 0 ]$ $1$ $2$ $[ 13 ]$ $[ 3 ]$
$169$ $[ 78 ]$ $2$ $2$ $[ 13 ]$ $[ 3 ]$
$169$ $[ 52 ]$ $3$ $4$ $[ 13 ]$ $[ 3 ]$
$169$ $[ 26 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$170$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$170$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$170$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$170$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$170$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$170$ $[ 0, 8 ]$ $2$ $2$ $[]$ $[]$
$170$ $[ 0, 8 ]$ $2$ $2$ $[]$ $[]$
$170$ $[ 0, 8 ]$ $2$ $2$ $[]$ $[]$
$170$ $[ 2, 0 ]$ $2$ $2$ $[]$ $[]$
$170$ $[ 2, 8 ]$ $2$ $2$ $[]$ $[]$
$170$ $[ 2, 8 ]$ $2$ $2$ $[]$ $[]$
$170$ $[ 2, 8 ]$ $2$ $4$ $[ 17 ]$ $[ 2 ]$
$170$ $[ 2, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$170$ $[ 2, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$170$ $[ 2, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$170$ $[ 2, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$171$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$171$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$171$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$171$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$171$ $[ 3, 9 ]$ $2$ $4$ $[ 57, -19 ]$ $[ 2, 5 ]$
$171$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$171$ $[ 2, 6 ]$ $3$ $2$ $[]$ $[]$
$171$ $[ 2, 6 ]$ $3$ $2$ $[]$ $[]$
$171$ $[ 4, 6 ]$ $3$ $2$ $[]$ $[]$
$171$ $[ 4, 6 ]$ $3$ $2$ $[]$ $[]$
$172$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$174$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$174$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$174$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$174$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$174$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$174$ $[ 0, 14 ]$ $2$ $2$ $[]$ $[]$
$174$ $[ 1, 7 ]$ $4$ $4$ $[ -87 ]$ $[ 2 ]$
$175$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$175$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$175$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$175$ $[ 10, 0 ]$ $2$ $2$ $[]$ $[]$
$175$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$175$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$176$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$176$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$176$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$178$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$178$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$178$ $[ 44 ]$ $2$ $2$ $[ 89 ]$ $[ 5 ]$
$178$ $[ 44 ]$ $2$ $2$ $[ 89 ]$ $[ 5 ]$
$178$ $[ 22 ]$ $4$ $2$ $[ 89 ]$ $[ 2 ]$
$178$ $[ 22 ]$ $4$ $2$ $[ 89 ]$ $[ 2 ]$
$179$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$180$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$180$ $[ 0, 0, 2 ]$ $2$ $2$ $[]$ $[]$
$180$ $[ 1, 3, 2 ]$ $2$ $8$ $[ 12, -4, -20 ]$ $[ 2, 5, 6 ]$
$180$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$180$ $[ 0, 3, 1 ]$ $4$ $4$ $[ -15 ]$ $[ 2 ]$
$181$ $[ 30 ]$ $6$ $2$ $[ 181 ]$ $[ 3 ]$
$182$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$182$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$182$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$182$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$182$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$182$ $[ 0, 6 ]$ $2$ $2$ $[]$ $[]$
$182$ $[ 0, 4 ]$ $3$ $2$ $[]$ $[]$
$182$ $[ 0, 4 ]$ $3$ $2$ $[]$ $[]$
$182$ $[ 0, 4 ]$ $3$ $2$ $[]$ $[]$
$182$ $[ 0, 4 ]$ $3$ $2$ $[]$ $[]$
$182$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$182$ $[ 2, 4 ]$ $3$ $2$ $[]$ $[]$
$182$ $[ 2, 4 ]$ $3$ $2$ $[]$ $[]$
$182$ $[ 4, 4 ]$ $3$ $2$ $[]$ $[]$
$182$ $[ 4, 4 ]$ $3$ $2$ $[]$ $[]$
$182$ $[ 2, 6 ]$ $6$ $4$ $[]$ $[]$
$183$ $[ 0, 30 ]$ $2$ $2$ $[]$ $[]$
$183$ $[ 0, 30 ]$ $2$ $2$ $[]$ $[]$
$183$ $[ 0, 20 ]$ $3$ $2$ $[]$ $[]$
$183$ $[ 0, 10 ]$ $6$ $2$ $[ 61 ]$ $[ 3 ]$
$184$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$184$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$184$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$184$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$184$ $[ 0, 1, 0 ]$ $2$ $2$ $[ 8 ]$ $[ 2 ]$
$185$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$185$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$185$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$185$ $[ 0, 18 ]$ $2$ $2$ $[]$ $[]$
$185$ $[ 1, 9 ]$ $4$ $2$ $[ 185 ]$ $[ 2 ]$
$185$ $[ 1, 9 ]$ $4$ $2$ $[ 185 ]$ $[ 2 ]$
$185$ $[ 3, 9 ]$ $4$ $2$ $[ 185 ]$ $[ 2 ]$
$185$ $[ 3, 9 ]$ $4$ $2$ $[ 185 ]$ $[ 2 ]$
$186$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$186$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$186$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$186$ $[ 1, 15 ]$ $2$ $4$ $[ -31 ]$ $[ 3 ]$
$186$ $[ 0, 10 ]$ $3$ $2$ $[]$ $[]$
$186$ $[ 0, 10 ]$ $3$ $2$ $[]$ $[]$
$187$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$187$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$189$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$189$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$189$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$189$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$189$ $[ 0, 0 ]$ $1$ $2$ $[ -3 ]$ $[ 3 ]$
$189$ $[ 0, 0 ]$ $1$ $2$ $[ -3 ]$ $[ 7 ]$
$189$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$189$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$189$ $[ 6, 2 ]$ $3$ $2$ $[]$ $[]$
$189$ $[ 12, 2 ]$ $3$ $2$ $[]$ $[]$
$189$ $[ 3, 1 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$189$ $[ 15, 1 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$190$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$190$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$190$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$190$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$190$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$190$ $[ 1, 9 ]$ $4$ $4$ $[ -95 ]$ $[ 2 ]$
$192$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$192$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$192$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$192$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$192$ $[ 0, 8, 0 ]$ $2$ $4$ $[ -8 ]$ $[ 3 ]$
$192$ $[ 1, 0, 1 ]$ $2$ $4$ $[ -3 ]$ $[ 2 ]$
$194$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$194$ $[ 32 ]$ $3$ $2$ $[]$ $[]$
$194$ $[ 16 ]$ $6$ $2$ $[ 97 ]$ $[ 3 ]$
$194$ $[ 16 ]$ $6$ $2$ $[ 97 ]$ $[ 3 ]$
$195$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$195$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$195$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$195$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$195$ $[ 0, 0, 6 ]$ $2$ $2$ $[]$ $[]$
$195$ $[ 0, 0, 6 ]$ $2$ $2$ $[]$ $[]$
$195$ $[ 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$195$ $[ 0, 2, 6 ]$ $2$ $2$ $[]$ $[]$
$195$ $[ 0, 2, 6 ]$ $2$ $2$ $[]$ $[]$
$195$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$195$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$196$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$196$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$196$ $[ 0, 0 ]$ $1$ $2$ $[ -7 ]$ $[ 2 ]$
$196$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$196$ $[ 0, 14 ]$ $3$ $4$ $[ -7 ]$ $[ 2 ]$
$197$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$198$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$198$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$198$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$198$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$198$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$198$ $[ 3, 5 ]$ $2$ $2$ $[ 33 ]$ $[ 2 ]$
$198$ $[ 3, 5 ]$ $2$ $2$ $[ 33 ]$ $[ 2 ]$
$198$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$198$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$200$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$200$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$200$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$200$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$200$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$200$ $[ 0, 0, 10 ]$ $2$ $2$ $[]$ $[]$
$200$ $[ 0, 0, 10 ]$ $2$ $2$ $[]$ $[]$
$200$ $[ 0, 1, 0 ]$ $2$ $2$ $[ 8 ]$ $[ 7 ]$
$200$ $[ 0, 1, 0 ]$ $2$ $2$ $[ 8 ]$ $[ 7 ]$
$200$ $[ 0, 1, 0 ]$ $2$ $2$ $[]$ $[]$
$200$ $[ 0, 1, 0 ]$ $2$ $2$ $[]$ $[]$
$200$ $[ 0, 1, 0 ]$ $2$ $4$ $[ 5, 40 ]$ $[ 2, 3 ]$
$200$ $[ 0, 1, 10 ]$ $2$ $2$ $[]$ $[]$
$200$ $[ 0, 1, 10 ]$ $2$ $2$ $[]$ $[]$
$200$ $[ 0, 1, 10 ]$ $2$ $4$ $[ 8 ]$ $[ 7 ]$
$201$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$201$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$201$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$201$ $[ 0, 22 ]$ $3$ $2$ $[]$ $[]$
$202$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$203$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$203$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$203$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$203$ $[ 0, 14 ]$ $2$ $2$ $[ 29 ]$ $[ 5 ]$
$203$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$203$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$204$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$204$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$204$ $[ 0, 0, 8 ]$ $2$ $2$ $[]$ $[]$
$205$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$205$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$205$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$205$ $[ 0, 20 ]$ $2$ $2$ $[ 41 ]$ $[ 2 ]$
$205$ $[ 0, 10 ]$ $4$ $2$ $[ 41 ]$ $[ 2 ]$
$205$ $[ 2, 10 ]$ $4$ $4$ $[ 41, 5 ]$ $[ 2, 5 ]$
$206$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$206$ $[ 34 ]$ $3$ $2$ $[]$ $[]$
$207$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$208$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$208$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$208$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$208$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$208$ $[ 0, 0, 6 ]$ $2$ $2$ $[]$ $[]$
$208$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$208$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$208$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$208$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$208$ $[ 1, 1, 3 ]$ $4$ $2$ $[ 104 ]$ $[ 2 ]$
$208$ $[ 1, 3, 3 ]$ $4$ $2$ $[ 104 ]$ $[ 2 ]$
$208$ $[ 0, 0, 2 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$208$ $[ 0, 0, 2 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$209$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$210$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$210$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$210$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$210$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$210$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$210$ $[ 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$210$ $[ 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$210$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$210$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$210$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$210$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$210$ $[ 0, 2, 2 ]$ $6$ $4$ $[]$ $[]$
$212$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$212$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$213$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$214$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$214$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$214$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$214$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$215$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$215$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$216$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$216$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$216$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$216$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$216$ $[ 0, 1, 0 ]$ $2$ $4$ $[ -24 ]$ $[ 7 ]$
$216$ $[ 0, 0, 6 ]$ $3$ $2$ $[]$ $[]$
$216$ $[ 0, 1, 6 ]$ $6$ $4$ $[]$ $[]$
$218$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$218$ $[ 36 ]$ $3$ $2$ $[]$ $[]$
$219$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$219$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$219$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$219$ $[ 0, 36 ]$ $2$ $2$ $[ 73 ]$ $[ 2 ]$
$219$ $[ 0, 36 ]$ $2$ $2$ $[]$ $[]$
$219$ $[ 0, 24 ]$ $3$ $2$ $[]$ $[]$
$219$ $[ 0, 18 ]$ $4$ $2$ $[ 73 ]$ $[ 2 ]$
$220$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$220$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$220$ $[ 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$220$ $[ 1, 0, 5 ]$ $2$ $2$ $[ 44 ]$ $[ 2 ]$
$220$ $[ 1, 0, 5 ]$ $2$ $2$ $[ 44 ]$ $[ 2 ]$
$220$ $[ 1, 1, 0 ]$ $4$ $2$ $[ 5 ]$ $[ 2 ]$
$220$ $[ 1, 1, 0 ]$ $4$ $2$ $[ 5 ]$ $[ 2 ]$
$221$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$221$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$221$ $[ 0, 8 ]$ $2$ $2$ $[]$ $[]$
$221$ $[ 6, 0 ]$ $2$ $2$ $[ 13 ]$ $[ 3 ]$
$221$ $[ 6, 8 ]$ $2$ $4$ $[ 17, 13 ]$ $[ 2, 3 ]$
$221$ $[ 6, 8 ]$ $2$ $4$ $[ 17 ]$ $[ 2 ]$
$221$ $[ 2, 0 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$222$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$222$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$222$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$222$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$222$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$222$ $[ 0, 18 ]$ $2$ $2$ $[]$ $[]$
$222$ $[ 0, 12 ]$ $3$ $2$ $[]$ $[]$
$222$ $[ 0, 12 ]$ $3$ $2$ $[]$ $[]$
$224$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$224$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$224$ $[ 0, 4, 0 ]$ $2$ $2$ $[ 8 ]$ $[ 2 ]$
$225$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$225$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$225$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$225$ $[ 0, 10 ]$ $2$ $2$ $[]$ $[]$
$225$ $[ 0, 10 ]$ $2$ $2$ $[]$ $[]$
$225$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$225$ $[ 3, 5 ]$ $4$ $4$ $[ -15 ]$ $[ 2 ]$
$225$ $[ 2, 10 ]$ $6$ $4$ $[]$ $[]$
$226$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$226$ $[ 56 ]$ $2$ $2$ $[ 113 ]$ $[ 2 ]$
$226$ $[ 56 ]$ $2$ $2$ $[ 113 ]$ $[ 2 ]$
$226$ $[ 56 ]$ $2$ $2$ $[ 113 ]$ $[ 2 ]$
$226$ $[ 28 ]$ $4$ $2$ $[ 113 ]$ $[ 2 ]$
$226$ $[ 28 ]$ $4$ $2$ $[ 113 ]$ $[ 2 ]$
$226$ $[ 28 ]$ $4$ $2$ $[ 113 ]$ $[ 2 ]$
$228$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$228$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$229$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$229$ $[ 114 ]$ $2$ $2$ $[ 229 ]$ $[ 5 ]$
$229$ $[ 38 ]$ $6$ $2$ $[ 229 ]$ $[ 3 ]$
$231$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$231$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$231$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$231$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$232$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$232$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$232$ $[ 0, 1, 14 ]$ $2$ $4$ $[ 29, 232 ]$ $[ 2, 3 ]$
$233$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$234$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$234$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$234$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$234$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$234$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$234$ $[ 0, 6 ]$ $2$ $2$ $[]$ $[]$
$234$ $[ 0, 6 ]$ $2$ $2$ $[]$ $[]$
$234$ $[ 0, 4 ]$ $3$ $2$ $[]$ $[]$
$234$ $[ 0, 4 ]$ $3$ $2$ $[]$ $[]$
$234$ $[ 0, 4 ]$ $3$ $2$ $[]$ $[]$
$234$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$234$ $[ 2, 4 ]$ $3$ $2$ $[]$ $[]$
$234$ $[ 2, 4 ]$ $3$ $2$ $[]$ $[]$
$234$ $[ 4, 4 ]$ $3$ $2$ $[]$ $[]$
$234$ $[ 4, 4 ]$ $3$ $2$ $[]$ $[]$
$234$ $[ 3, 3 ]$ $4$ $4$ $[ -39 ]$ $[ 2 ]$
$234$ $[ 0, 2 ]$ $6$ $4$ $[ -39 ]$ $[ 3 ]$
$235$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$235$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$235$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$236$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$236$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$238$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$238$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$238$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$238$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$238$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$238$ $[ 0, 8 ]$ $2$ $2$ $[]$ $[]$
$238$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$238$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$238$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$238$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$238$ $[ 2, 8 ]$ $6$ $4$ $[]$ $[]$
$240$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$240$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$240$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$240$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$240$ $[ 0, 0, 0, 2 ]$ $2$ $2$ $[]$ $[]$
$240$ $[ 0, 0, 0, 2 ]$ $2$ $2$ $[]$ $[]$
$240$ $[ 0, 0, 0, 2 ]$ $2$ $2$ $[]$ $[]$
$240$ $[ 1, 0, 1, 0 ]$ $2$ $4$ $[ -4 ]$ $[ 6 ]$
$240$ $[ 1, 0, 1, 0 ]$ $2$ $4$ $[ -4 ]$ $[ 3 ]$
$240$ $[ 1, 0, 1, 2 ]$ $2$ $8$ $[ -15, -20 ]$ $[ 2, 3 ]$
$240$ $[ 0, 0, 1, 1 ]$ $4$ $4$ $[ -15 ]$ $[ 2 ]$
$240$ $[ 0, 0, 1, 1 ]$ $4$ $4$ $[ -15 ]$ $[ 2 ]$
$240$ $[ 1, 0, 0, 1 ]$ $4$ $4$ $[ -20 ]$ $[ 2 ]$
$240$ $[ 1, 1, 0, 1 ]$ $4$ $2$ $[ 40 ]$ $[ 2 ]$
$240$ $[ 1, 1, 0, 1 ]$ $4$ $2$ $[ 40 ]$ $[ 2 ]$
$240$ $[ 1, 1, 0, 1 ]$ $4$ $2$ $[ 40 ]$ $[ 2 ]$
$240$ $[ 1, 1, 1, 0 ]$ $4$ $4$ $[ -24 ]$ $[ 2 ]$
$240$ $[ 1, 3, 0, 1 ]$ $4$ $2$ $[ 40 ]$ $[ 2 ]$
$240$ $[ 1, 3, 0, 1 ]$ $4$ $2$ $[ 40 ]$ $[ 2 ]$
$240$ $[ 1, 3, 0, 1 ]$ $4$ $2$ $[ 40 ]$ $[ 2 ]$
$242$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$242$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$242$ $[ 22 ]$ $5$ $4$ $[]$ $[]$
$242$ $[ 22 ]$ $5$ $4$ $[]$ $[]$
$243$ $[ 0 ]$ $1$ $2$ $[ -3 ]$ $[ 3 ]$
$243$ $[ 54 ]$ $3$ $4$ $[ -3 ]$ $[ 3 ]$
$244$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$244$ $[ 0, 20 ]$ $3$ $2$ $[]$ $[]$
$244$ $[ 0, 20 ]$ $3$ $2$ $[]$ $[]$
$245$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$245$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$245$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$245$ $[ 2, 0 ]$ $2$ $2$ $[]$ $[]$
$245$ $[ 2, 0 ]$ $2$ $2$ $[]$ $[]$
$245$ $[ 2, 0 ]$ $2$ $2$ $[]$ $[]$
$245$ $[ 2, 0 ]$ $2$ $4$ $[ -7 ]$ $[ 2 ]$
$245$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$245$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$245$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$245$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$245$ $[ 2, 14 ]$ $6$ $4$ $[]$ $[]$
$245$ $[ 2, 14 ]$ $6$ $4$ $[]$ $[]$
$245$ $[ 2, 14 ]$ $6$ $4$ $[]$ $[]$
$245$ $[ 2, 14 ]$ $6$ $8$ $[ -7 ]$ $[ 2 ]$
$246$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$246$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$246$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$246$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$246$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$246$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$246$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$246$ $[ 0, 20 ]$ $2$ $2$ $[]$ $[]$
$248$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$248$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$248$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$248$ $[ 0, 0, 10 ]$ $3$ $2$ $[]$ $[]$
$248$ $[ 0, 0, 10 ]$ $3$ $2$ $[]$ $[]$
$248$ $[ 0, 0, 10 ]$ $3$ $2$ $[]$ $[]$
$248$ $[ 0, 0, 10 ]$ $3$ $2$ $[]$ $[]$
$248$ $[ 0, 0, 10 ]$ $3$ $2$ $[]$ $[]$
$248$ $[ 0, 1, 10 ]$ $6$ $4$ $[ 8 ]$ $[ 7 ]$
$249$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$249$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$252$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$252$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$252$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$252$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$252$ $[ 0, 2, 2 ]$ $3$ $2$ $[]$ $[]$
$252$ $[ 0, 4, 2 ]$ $3$ $2$ $[]$ $[]$
$254$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$254$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$254$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$254$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$254$ $[ 42 ]$ $3$ $2$ $[]$ $[]$
$255$ $[ 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$256$ $[ 0, 32 ]$ $2$ $2$ $[]$ $[]$
$256$ $[ 0, 32 ]$ $2$ $2$ $[]$ $[]$
$257$ $[ 128 ]$ $2$ $2$ $[ 257 ]$ $[ 2 ]$
$257$ $[ 64 ]$ $4$ $2$ $[ 257 ]$ $[ 2 ]$
$258$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$258$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$258$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$258$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$258$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$258$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$258$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$258$ $[ 1, 21 ]$ $2$ $2$ $[ 129 ]$ $[ 3 ]$
$258$ $[ 1, 21 ]$ $2$ $2$ $[ 129 ]$ $[ 3 ]$
$258$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$258$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$258$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$258$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$258$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$258$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$258$ $[ 1, 7 ]$ $6$ $2$ $[ 129 ]$ $[ 3 ]$
$258$ $[ 1, 7 ]$ $6$ $2$ $[ 129 ]$ $[ 3 ]$
$259$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$259$ $[ 0, 18 ]$ $2$ $2$ $[]$ $[]$
$259$ $[ 0, 6 ]$ $6$ $2$ $[ 37 ]$ $[ 3 ]$
$260$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$260$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$260$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$260$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$260$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$260$ $[ 0, 1, 3 ]$ $4$ $2$ $[ 65 ]$ $[ 2 ]$
$260$ $[ 0, 3, 3 ]$ $4$ $2$ $[ 65 ]$ $[ 2 ]$
$261$ $[ 0, 14 ]$ $2$ $2$ $[ 29 ]$ $[ 5 ]$
$261$ $[ 3, 7 ]$ $4$ $4$ $[ -87 ]$ $[ 2 ]$
$262$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$262$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$264$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$264$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$264$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$264$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$264$ $[ 0, 1, 0, 0 ]$ $2$ $2$ $[]$ $[]$
$264$ $[ 0, 1, 0, 0 ]$ $2$ $2$ $[]$ $[]$
$264$ $[ 0, 1, 0, 0 ]$ $2$ $2$ $[]$ $[]$
$264$ $[ 1, 1, 0, 5 ]$ $2$ $2$ $[ 88 ]$ $[ 7 ]$
$264$ $[ 1, 1, 0, 5 ]$ $2$ $2$ $[ 88 ]$ $[ 7 ]$
$264$ $[ 1, 1, 0, 5 ]$ $2$ $4$ $[ -8, 88 ]$ $[ 2, 7 ]$
$264$ $[ 1, 1, 0, 5 ]$ $2$ $4$ $[ -11, -8 ]$ $[ 2, 3 ]$
$265$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$265$ $[ 1, 13 ]$ $4$ $2$ $[ 265 ]$ $[ 2 ]$
$265$ $[ 3, 13 ]$ $4$ $2$ $[ 265 ]$ $[ 2 ]$
$266$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$266$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$266$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$266$ $[ 2, 6 ]$ $3$ $2$ $[]$ $[]$
$266$ $[ 4, 6 ]$ $3$ $2$ $[]$ $[]$
$267$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$267$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$268$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$269$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$270$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$270$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$270$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$270$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$270$ $[ 0, 2 ]$ $2$ $2$ $[]$ $[]$
$270$ $[ 0, 2 ]$ $2$ $2$ $[]$ $[]$
$270$ $[ 0, 2 ]$ $2$ $4$ $[ -15 ]$ $[ 19 ]$
$270$ $[ 6, 0 ]$ $3$ $2$ $[]$ $[]$
$270$ $[ 6, 0 ]$ $3$ $2$ $[]$ $[]$
$270$ $[ 6, 2 ]$ $6$ $4$ $[]$ $[]$
$272$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$272$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$272$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$272$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$272$ $[ 0, 0, 8 ]$ $2$ $2$ $[ 17 ]$ $[ 2 ]$
$272$ $[ 0, 0, 8 ]$ $2$ $2$ $[ 17 ]$ $[ 2 ]$
$272$ $[ 0, 0, 8 ]$ $2$ $2$ $[]$ $[]$
$272$ $[ 0, 0, 8 ]$ $2$ $2$ $[ 17 ]$ $[ 2 ]$
$272$ $[ 0, 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$272$ $[ 0, 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$272$ $[ 0, 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$272$ $[ 0, 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$272$ $[ 0, 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$272$ $[ 0, 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$272$ $[ 0, 1, 0 ]$ $4$ $2$ $[ 8 ]$ $[ 2 ]$
$273$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$273$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$273$ $[ 0, 0, 6 ]$ $2$ $2$ $[]$ $[]$
$273$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$273$ $[ 0, 2, 4 ]$ $3$ $2$ $[]$ $[]$
$273$ $[ 0, 4, 4 ]$ $3$ $2$ $[]$ $[]$
$273$ $[ 0, 2, 2 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$273$ $[ 0, 2, 6 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$273$ $[ 0, 2, 6 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$273$ $[ 0, 4, 2 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$274$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$274$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$274$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$274$ $[ 68 ]$ $2$ $2$ $[ 137 ]$ $[ 2 ]$
$274$ $[ 34 ]$ $4$ $2$ $[ 137 ]$ $[ 2 ]$
$275$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$275$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$275$ $[ 10, 0 ]$ $2$ $2$ $[]$ $[]$
$275$ $[ 10, 0 ]$ $2$ $2$ $[]$ $[]$
$275$ $[ 5, 5 ]$ $4$ $4$ $[ -55, -11 ]$ $[ 2, 5 ]$
$277$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$277$ $[ 92 ]$ $3$ $2$ $[]$ $[]$
$277$ $[ 46 ]$ $6$ $2$ $[ 277 ]$ $[ 3 ]$
$278$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$278$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$278$ $[ 46 ]$ $3$ $2$ $[]$ $[]$
$280$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$280$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$280$ $[ 0, 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$280$ $[ 0, 1, 0, 0 ]$ $2$ $2$ $[]$ $[]$
$280$ $[ 0, 1, 0, 0 ]$ $2$ $2$ $[]$ $[]$
$280$ $[ 0, 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$280$ $[ 0, 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$280$ $[ 0, 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$280$ $[ 0, 1, 0, 2 ]$ $6$ $4$ $[]$ $[]$
$281$ $[ 70 ]$ $4$ $2$ $[ 281 ]$ $[ 2 ]$
$282$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$282$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$285$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$285$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$285$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$285$ $[ 0, 0, 6 ]$ $3$ $2$ $[]$ $[]$
$285$ $[ 0, 0, 6 ]$ $3$ $2$ $[]$ $[]$
$285$ $[ 0, 0, 6 ]$ $3$ $2$ $[]$ $[]$
$285$ $[ 0, 1, 9 ]$ $4$ $4$ $[ -95 ]$ $[ 2 ]$
$285$ $[ 1, 0, 3 ]$ $6$ $4$ $[ -19 ]$ $[ 3 ]$
$286$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$286$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$286$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$286$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$286$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$286$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$286$ $[ 0, 6 ]$ $2$ $2$ $[]$ $[]$
$286$ $[ 0, 6 ]$ $2$ $2$ $[]$ $[]$
$286$ $[ 0, 6 ]$ $2$ $2$ $[]$ $[]$
$286$ $[ 0, 4 ]$ $3$ $2$ $[]$ $[]$
$286$ $[ 0, 4 ]$ $3$ $2$ $[]$ $[]$
$286$ $[ 0, 4 ]$ $3$ $2$ $[]$ $[]$
$286$ $[ 0, 4 ]$ $3$ $2$ $[]$ $[]$
$288$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$288$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$288$ $[ 0, 4, 0 ]$ $2$ $2$ $[]$ $[]$
$288$ $[ 1, 0, 3 ]$ $2$ $4$ $[ -3 ]$ $[ 2 ]$
$288$ $[ 1, 4, 3 ]$ $2$ $4$ $[ -3, -8 ]$ $[ 2, 3 ]$
$288$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$288$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$288$ $[ 0, 4, 2 ]$ $6$ $4$ $[]$ $[]$
$289$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$289$ $[ 136 ]$ $2$ $2$ $[]$ $[]$
$289$ $[ 68 ]$ $4$ $4$ $[]$ $[]$
$289$ $[ 34 ]$ $8$ $8$ $[]$ $[]$
$290$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$290$ $[ 0, 14 ]$ $2$ $2$ $[]$ $[]$
$290$ $[ 1, 7 ]$ $4$ $2$ $[ 145 ]$ $[ 2 ]$
$290$ $[ 1, 7 ]$ $4$ $2$ $[ 145 ]$ $[ 2 ]$
$290$ $[ 1, 7 ]$ $4$ $2$ $[ 145 ]$ $[ 2 ]$
$290$ $[ 3, 7 ]$ $4$ $2$ $[ 145 ]$ $[ 2 ]$
$290$ $[ 3, 7 ]$ $4$ $2$ $[ 145 ]$ $[ 2 ]$
$290$ $[ 3, 7 ]$ $4$ $2$ $[ 145 ]$ $[ 2 ]$
$291$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$291$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$291$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$291$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$291$ $[ 0, 48 ]$ $2$ $2$ $[ 97 ]$ $[ 3 ]$
$291$ $[ 0, 48 ]$ $2$ $2$ $[ 97 ]$ $[ 2 ]$
$291$ $[ 0, 48 ]$ $2$ $2$ $[ 97 ]$ $[ 2 ]$
$291$ $[ 0, 32 ]$ $3$ $2$ $[]$ $[]$
$291$ $[ 0, 32 ]$ $3$ $2$ $[]$ $[]$
$291$ $[ 0, 32 ]$ $3$ $2$ $[]$ $[]$
$291$ $[ 0, 32 ]$ $3$ $2$ $[]$ $[]$
$291$ $[ 0, 24 ]$ $4$ $2$ $[ 97 ]$ $[ 2 ]$
$292$ $[ 0, 18 ]$ $4$ $2$ $[ 73 ]$ $[ 2 ]$
$292$ $[ 0, 12 ]$ $6$ $2$ $[ 73 ]$ $[ 3 ]$
$294$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$294$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$294$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$294$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$294$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$294$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$294$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$294$ $[ 1, 21 ]$ $2$ $4$ $[ -7 ]$ $[ 3 ]$
$294$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$294$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$294$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$294$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$294$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$294$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$294$ $[ 1, 7 ]$ $6$ $4$ $[ -7 ]$ $[ 3 ]$
$294$ $[ 1, 7 ]$ $6$ $8$ $[ -7 ]$ $[ 6 ]$
$295$ $[ 2, 0 ]$ $2$ $2$ $[]$ $[]$
$296$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$296$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$296$ $[ 0, 0, 12 ]$ $3$ $2$ $[]$ $[]$
$297$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$297$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$297$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$297$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$297$ $[ 6, 0 ]$ $3$ $2$ $[]$ $[]$
$297$ $[ 6, 0 ]$ $3$ $2$ $[]$ $[]$
$297$ $[ 6, 0 ]$ $3$ $2$ $[]$ $[]$
$298$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$298$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$299$ $[ 6, 0 ]$ $2$ $2$ $[ 13 ]$ $[ 3 ]$
$299$ $[ 4, 0 ]$ $3$ $2$ $[]$ $[]$
$300$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$300$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$300$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$300$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$300$ $[ 0, 0, 10 ]$ $2$ $2$ $[]$ $[]$
$300$ $[ 1, 0, 5 ]$ $4$ $8$ $[ -20 ]$ $[ 7 ]$
$301$ $[ 3, 7 ]$ $6$ $2$ $[ 301 ]$ $[ 3 ]$
$301$ $[ 3, 7 ]$ $6$ $2$ $[ 301 ]$ $[ 3 ]$
$302$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$302$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$302$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$303$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$303$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$303$ $[ 0, 50 ]$ $2$ $2$ $[]$ $[]$
$304$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$304$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$304$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$304$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$304$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$304$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$304$ $[ 1, 0, 9 ]$ $2$ $4$ $[ 76, -19 ]$ $[ 3, 7 ]$
$304$ $[ 0, 0, 6 ]$ $3$ $2$ $[]$ $[]$
$304$ $[ 0, 0, 6 ]$ $3$ $2$ $[]$ $[]$
$304$ $[ 0, 0, 6 ]$ $3$ $2$ $[]$ $[]$
$304$ $[ 0, 0, 6 ]$ $3$ $2$ $[]$ $[]$
$304$ $[ 1, 0, 3 ]$ $6$ $2$ $[ 76 ]$ $[ 3 ]$
$304$ $[ 1, 0, 3 ]$ $6$ $2$ $[ 76 ]$ $[ 3 ]$
$305$ $[ 0, 30 ]$ $2$ $2$ $[]$ $[]$
$305$ $[ 2, 0 ]$ $2$ $2$ $[]$ $[]$
$305$ $[ 0, 20 ]$ $3$ $2$ $[]$ $[]$
$305$ $[ 1, 15 ]$ $4$ $2$ $[ 305 ]$ $[ 2 ]$
$305$ $[ 3, 15 ]$ $4$ $2$ $[ 305 ]$ $[ 2 ]$
$305$ $[ 0, 10 ]$ $6$ $2$ $[ 61 ]$ $[ 3 ]$
$305$ $[ 0, 10 ]$ $6$ $2$ $[ 61 ]$ $[ 3 ]$
$305$ $[ 2, 20 ]$ $6$ $4$ $[]$ $[]$
$306$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$306$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$306$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$306$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$306$ $[ 0, 8 ]$ $2$ $2$ $[ 17 ]$ $[ 2 ]$
$306$ $[ 0, 8 ]$ $2$ $2$ $[ 17 ]$ $[ 2 ]$
$306$ $[ 0, 8 ]$ $2$ $2$ $[ 17 ]$ $[ 2 ]$
$306$ $[ 0, 8 ]$ $2$ $2$ $[]$ $[]$
$306$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$306$ $[ 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$306$ $[ 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$306$ $[ 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$306$ $[ 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$306$ $[ 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$306$ $[ 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$306$ $[ 2, 8 ]$ $6$ $4$ $[]$ $[]$
$307$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$307$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$307$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$307$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$307$ $[ 102 ]$ $3$ $2$ $[]$ $[]$
$308$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$308$ $[ 0, 3, 5 ]$ $2$ $4$ $[ -7, -11 ]$ $[ 2, 7 ]$
$308$ $[ 1, 2, 5 ]$ $6$ $4$ $[ 44 ]$ $[ 7 ]$
$308$ $[ 1, 2, 5 ]$ $6$ $4$ $[ 44 ]$ $[ 7 ]$
$309$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$309$ $[ 0, 34 ]$ $3$ $2$ $[]$ $[]$
$310$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$310$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$312$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$312$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$312$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$312$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$312$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$312$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$312$ $[ 0, 0, 0, 6 ]$ $2$ $2$ $[]$ $[]$
$312$ $[ 0, 0, 0, 6 ]$ $2$ $2$ $[]$ $[]$
$312$ $[ 0, 0, 0, 6 ]$ $2$ $2$ $[]$ $[]$
$312$ $[ 0, 1, 0, 6 ]$ $2$ $2$ $[]$ $[]$
$312$ $[ 0, 1, 0, 6 ]$ $2$ $2$ $[]$ $[]$
$312$ $[ 0, 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$312$ $[ 0, 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$312$ $[ 0, 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$312$ $[ 0, 0, 1, 3 ]$ $4$ $4$ $[ -39 ]$ $[ 2 ]$
$312$ $[ 1, 1, 0, 3 ]$ $4$ $2$ $[ 13 ]$ $[ 2 ]$
$312$ $[ 1, 1, 0, 3 ]$ $4$ $2$ $[ 13 ]$ $[ 2 ]$
$312$ $[ 1, 1, 0, 3 ]$ $4$ $2$ $[ 13 ]$ $[ 2 ]$
$312$ $[ 1, 1, 0, 3 ]$ $4$ $2$ $[ 13 ]$ $[ 2 ]$
$314$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$314$ $[ 52 ]$ $3$ $2$ $[]$ $[]$
$315$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$315$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$315$ $[ 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$315$ $[ 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$315$ $[ 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$315$ $[ 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$315$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$315$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$315$ $[ 2, 0, 0 ]$ $3$ $2$ $[]$ $[]$
$315$ $[ 2, 0, 0 ]$ $3$ $2$ $[]$ $[]$
$315$ $[ 0, 1, 3 ]$ $4$ $4$ $[ 105, 21 ]$ $[ 2, 5 ]$
$315$ $[ 0, 1, 3 ]$ $4$ $4$ $[ 105, 21 ]$ $[ 2, 5 ]$
$315$ $[ 0, 2, 2 ]$ $6$ $4$ $[]$ $[]$
$315$ $[ 1, 0, 1 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$315$ $[ 1, 0, 3 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$315$ $[ 1, 0, 3 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$315$ $[ 1, 0, 3 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$315$ $[ 1, 0, 3 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$315$ $[ 1, 0, 3 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$315$ $[ 1, 0, 3 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$315$ $[ 1, 0, 3 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$315$ $[ 1, 0, 3 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$315$ $[ 2, 2, 0 ]$ $6$ $4$ $[]$ $[]$
$315$ $[ 2, 2, 0 ]$ $6$ $4$ $[]$ $[]$
$315$ $[ 2, 2, 2 ]$ $6$ $4$ $[]$ $[]$
$315$ $[ 4, 2, 2 ]$ $6$ $4$ $[]$ $[]$
$315$ $[ 5, 0, 1 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$316$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$316$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$316$ $[ 0, 26 ]$ $3$ $2$ $[]$ $[]$
$318$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$318$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$318$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$318$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$318$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$318$ $[ 0, 26 ]$ $2$ $2$ $[]$ $[]$
$318$ $[ 0, 26 ]$ $2$ $2$ $[]$ $[]$
$318$ $[ 0, 26 ]$ $2$ $2$ $[]$ $[]$
$319$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$320$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$320$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$320$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$320$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$320$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$320$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$320$ $[ 0, 0, 0 ]$ $1$ $2$ $[ -4 ]$ $[ 2 ]$
$320$ $[ 0, 0, 2 ]$ $2$ $2$ $[]$ $[]$
$320$ $[ 0, 0, 2 ]$ $2$ $2$ $[]$ $[]$
$320$ $[ 0, 4, 2 ]$ $4$ $2$ $[ 8 ]$ $[ 2 ]$
$320$ $[ 0, 4, 2 ]$ $4$ $2$ $[ 8 ]$ $[ 2 ]$
$320$ $[ 1, 0, 1 ]$ $4$ $2$ $[ 5 ]$ $[ 2 ]$
$320$ $[ 1, 0, 1 ]$ $4$ $2$ $[ 5 ]$ $[ 2 ]$
$320$ $[ 1, 0, 1 ]$ $4$ $2$ $[ 5 ]$ $[ 2 ]$
$320$ $[ 1, 0, 1 ]$ $4$ $2$ $[ 5 ]$ $[ 2 ]$
$320$ $[ 1, 0, 1 ]$ $4$ $2$ $[ 5 ]$ $[ 2 ]$
$320$ $[ 1, 0, 1 ]$ $4$ $2$ $[ 5 ]$ $[ 2 ]$
$320$ $[ 1, 4, 1 ]$ $4$ $2$ $[ 40 ]$ $[ 2 ]$
$320$ $[ 1, 8, 1 ]$ $4$ $2$ $[ 5 ]$ $[ 2 ]$
$320$ $[ 1, 8, 1 ]$ $4$ $2$ $[ 5 ]$ $[ 2 ]$
$320$ $[ 1, 8, 1 ]$ $4$ $2$ $[ 5 ]$ $[ 2 ]$
$320$ $[ 1, 8, 1 ]$ $4$ $2$ $[ 5 ]$ $[ 2 ]$
$320$ $[ 1, 12, 1 ]$ $4$ $2$ $[ 40 ]$ $[ 2 ]$
$322$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$322$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$322$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$322$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$322$ $[ 3, 11 ]$ $2$ $4$ $[ 161, -23 ]$ $[ 2, 7 ]$
$322$ $[ 3, 11 ]$ $2$ $4$ $[ -7, -23 ]$ $[ 2, 7 ]$
$323$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$323$ $[ 8, 0 ]$ $2$ $2$ $[ 17 ]$ $[ 2 ]$
$323$ $[ 8, 0 ]$ $2$ $2$ $[]$ $[]$
$323$ $[ 8, 0 ]$ $2$ $2$ $[]$ $[]$
$323$ $[ 4, 0 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$323$ $[ 4, 0 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$323$ $[ 4, 0 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$324$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$324$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$324$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$324$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$324$ $[ 0, 18 ]$ $3$ $2$ $[]$ $[]$
$324$ $[ 0, 18 ]$ $3$ $2$ $[]$ $[]$
$325$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$325$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$325$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$325$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$325$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$325$ $[ 0, 6 ]$ $2$ $2$ $[]$ $[]$
$325$ $[ 0, 6 ]$ $2$ $2$ $[]$ $[]$
$325$ $[ 0, 6 ]$ $2$ $2$ $[]$ $[]$
$325$ $[ 0, 6 ]$ $2$ $2$ $[]$ $[]$
$325$ $[ 0, 6 ]$ $2$ $2$ $[]$ $[]$
$325$ $[ 0, 6 ]$ $2$ $2$ $[]$ $[]$
$325$ $[ 10, 0 ]$ $2$ $2$ $[]$ $[]$
$325$ $[ 10, 0 ]$ $2$ $2$ $[]$ $[]$
$325$ $[ 10, 0 ]$ $2$ $2$ $[]$ $[]$
$325$ $[ 10, 6 ]$ $2$ $2$ $[]$ $[]$
$325$ $[ 10, 6 ]$ $2$ $2$ $[]$ $[]$
$325$ $[ 10, 6 ]$ $2$ $2$ $[]$ $[]$
$325$ $[ 10, 6 ]$ $2$ $2$ $[]$ $[]$
$325$ $[ 5, 3 ]$ $4$ $2$ $[ 65 ]$ $[ 2 ]$
$325$ $[ 15, 3 ]$ $4$ $2$ $[ 65 ]$ $[ 2 ]$
$325$ $[ 0, 2 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$325$ $[ 10, 2 ]$ $6$ $4$ $[ 13 ]$ $[ 3 ]$
$326$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$326$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$326$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$326$ $[ 54 ]$ $3$ $2$ $[]$ $[]$
$327$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$327$ $[ 0, 36 ]$ $3$ $2$ $[]$ $[]$
$327$ $[ 1, 27 ]$ $4$ $4$ $[ -327 ]$ $[ 2 ]$
$327$ $[ 0, 18 ]$ $6$ $2$ $[ 109 ]$ $[ 3 ]$
$328$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$328$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$328$ $[ 0, 0, 10 ]$ $4$ $2$ $[ 41 ]$ $[ 2 ]$
$328$ $[ 0, 0, 10 ]$ $4$ $2$ $[ 41 ]$ $[ 2 ]$
$329$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$330$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$330$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$330$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$330$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$330$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$331$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$333$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$333$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$333$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$333$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$333$ $[ 0, 18 ]$ $2$ $2$ $[]$ $[]$
$333$ $[ 0, 18 ]$ $2$ $2$ $[]$ $[]$
$333$ $[ 0, 12 ]$ $3$ $2$ $[]$ $[]$
$333$ $[ 0, 6 ]$ $6$ $2$ $[ 37 ]$ $[ 3 ]$
$333$ $[ 2, 6 ]$ $6$ $2$ $[ 37 ]$ $[ 3 ]$
$333$ $[ 4, 6 ]$ $6$ $2$ $[ 37 ]$ $[ 3 ]$
$334$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$335$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$336$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$336$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$336$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$336$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$336$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$336$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$336$ $[ 0, 0, 1, 3 ]$ $2$ $4$ $[ -7 ]$ $[ 6 ]$
$336$ $[ 1, 0, 0, 3 ]$ $2$ $2$ $[ 28 ]$ $[ 6 ]$
$336$ $[ 1, 0, 0, 3 ]$ $2$ $2$ $[ 28 ]$ $[ 3 ]$
$336$ $[ 1, 0, 0, 3 ]$ $2$ $2$ $[ 28 ]$ $[ 6 ]$
$336$ $[ 1, 0, 0, 3 ]$ $2$ $2$ $[ 28 ]$ $[ 3 ]$
$336$ $[ 1, 0, 1, 0 ]$ $2$ $4$ $[ -3 ]$ $[ 2 ]$
$336$ $[ 0, 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$336$ $[ 0, 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$336$ $[ 0, 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$336$ $[ 0, 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$336$ $[ 0, 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$336$ $[ 0, 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$336$ $[ 0, 0, 1, 1 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$336$ $[ 0, 0, 1, 1 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$336$ $[ 0, 0, 1, 1 ]$ $6$ $4$ $[ -7 ]$ $[ 3 ]$
$336$ $[ 1, 0, 0, 1 ]$ $6$ $2$ $[ 28 ]$ $[ 3 ]$
$336$ $[ 1, 0, 0, 1 ]$ $6$ $2$ $[ 28 ]$ $[ 3 ]$
$336$ $[ 1, 0, 0, 1 ]$ $6$ $2$ $[ 28 ]$ $[ 3 ]$
$336$ $[ 1, 0, 0, 1 ]$ $6$ $2$ $[ 28 ]$ $[ 3 ]$
$336$ $[ 1, 0, 0, 1 ]$ $6$ $2$ $[ 28 ]$ $[ 3 ]$
$336$ $[ 1, 0, 0, 1 ]$ $6$ $2$ $[ 28 ]$ $[ 3 ]$
$336$ $[ 1, 0, 0, 1 ]$ $6$ $2$ $[ 28 ]$ $[ 3 ]$
$336$ $[ 1, 0, 0, 1 ]$ $6$ $2$ $[ 28 ]$ $[ 3 ]$
$337$ $[ 168 ]$ $2$ $2$ $[ 337 ]$ $[ 2 ]$
$337$ $[ 84 ]$ $4$ $2$ $[ 337 ]$ $[ 2 ]$
$338$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$338$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$338$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$338$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$338$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$338$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$338$ $[ 78 ]$ $2$ $2$ $[]$ $[]$
$338$ $[ 78 ]$ $2$ $2$ $[]$ $[]$
$338$ $[ 78 ]$ $2$ $2$ $[]$ $[]$
$338$ $[ 52 ]$ $3$ $2$ $[]$ $[]$
$338$ $[ 52 ]$ $3$ $2$ $[]$ $[]$
$338$ $[ 52 ]$ $3$ $2$ $[]$ $[]$
$338$ $[ 52 ]$ $3$ $2$ $[]$ $[]$
$338$ $[ 52 ]$ $3$ $2$ $[]$ $[]$
$338$ $[ 52 ]$ $3$ $2$ $[]$ $[]$
$338$ $[ 52 ]$ $3$ $2$ $[]$ $[]$
$338$ $[ 26 ]$ $6$ $4$ $[]$ $[]$
$338$ $[ 26 ]$ $6$ $4$ $[]$ $[]$
$338$ $[ 26 ]$ $6$ $4$ $[]$ $[]$
$338$ $[ 26 ]$ $6$ $4$ $[]$ $[]$
$339$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$339$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$339$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$340$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$340$ $[ 0, 2, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$340$ $[ 0, 2, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$342$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$342$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$342$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$342$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$342$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$342$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$342$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$342$ $[ 3, 9 ]$ $2$ $2$ $[ 57 ]$ $[ 2 ]$
$342$ $[ 3, 9 ]$ $2$ $2$ $[ 57 ]$ $[ 2 ]$
$342$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$342$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$342$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$342$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$342$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$342$ $[ 2, 6 ]$ $3$ $2$ $[]$ $[]$
$342$ $[ 2, 6 ]$ $3$ $2$ $[]$ $[]$
$342$ $[ 2, 6 ]$ $3$ $2$ $[]$ $[]$
$342$ $[ 2, 6 ]$ $3$ $2$ $[]$ $[]$
$342$ $[ 4, 6 ]$ $3$ $2$ $[]$ $[]$
$342$ $[ 4, 6 ]$ $3$ $2$ $[]$ $[]$
$342$ $[ 4, 6 ]$ $3$ $2$ $[]$ $[]$
$342$ $[ 4, 6 ]$ $3$ $2$ $[]$ $[]$
$342$ $[ 1, 3 ]$ $6$ $2$ $[ 57 ]$ $[ 3 ]$
$342$ $[ 1, 3 ]$ $6$ $2$ $[ 57 ]$ $[ 3 ]$
$342$ $[ 1, 3 ]$ $6$ $2$ $[ 57 ]$ $[ 3 ]$
$342$ $[ 5, 3 ]$ $6$ $2$ $[ 57 ]$ $[ 3 ]$
$342$ $[ 5, 3 ]$ $6$ $2$ $[ 57 ]$ $[ 3 ]$
$342$ $[ 5, 3 ]$ $6$ $2$ $[ 57 ]$ $[ 3 ]$
$344$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$344$ $[ 0, 0, 14 ]$ $3$ $2$ $[]$ $[]$
$344$ $[ 0, 0, 14 ]$ $3$ $2$ $[]$ $[]$
$345$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$345$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$345$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$345$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$345$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$345$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$345$ $[ 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$345$ $[ 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$346$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$346$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$346$ $[ 86 ]$ $2$ $2$ $[]$ $[]$
$347$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$348$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$348$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$348$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$348$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$348$ $[ 0, 0, 14 ]$ $2$ $2$ $[]$ $[]$
$348$ $[ 0, 0, 14 ]$ $2$ $2$ $[]$ $[]$
$348$ $[ 0, 0, 14 ]$ $2$ $2$ $[]$ $[]$
$348$ $[ 1, 1, 14 ]$ $2$ $8$ $[ -4, -116 ]$ $[ 3, 5 ]$
$349$ $[ 174 ]$ $2$ $2$ $[ 349 ]$ $[ 5 ]$
$350$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$350$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$350$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$350$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$350$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$350$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$350$ $[ 10, 0 ]$ $2$ $2$ $[]$ $[]$
$350$ $[ 10, 0 ]$ $2$ $2$ $[]$ $[]$
$350$ $[ 10, 0 ]$ $2$ $2$ $[]$ $[]$
$350$ $[ 10, 0 ]$ $2$ $2$ $[]$ $[]$
$350$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$350$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$350$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$350$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$350$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$350$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$350$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$350$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$350$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$350$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$350$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$350$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$350$ $[ 10, 2 ]$ $6$ $4$ $[]$ $[]$
$350$ $[ 10, 2 ]$ $6$ $4$ $[]$ $[]$
$350$ $[ 10, 2 ]$ $6$ $4$ $[]$ $[]$
$350$ $[ 10, 2 ]$ $6$ $4$ $[]$ $[]$
$350$ $[ 10, 2 ]$ $6$ $4$ $[]$ $[]$
$350$ $[ 10, 2 ]$ $6$ $4$ $[]$ $[]$
$351$ $[ 0, 4 ]$ $3$ $2$ $[]$ $[]$
$351$ $[ 0, 4 ]$ $3$ $2$ $[]$ $[]$
$351$ $[ 6, 0 ]$ $3$ $2$ $[]$ $[]$
$351$ $[ 0, 2 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$351$ $[ 0, 2 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$351$ $[ 0, 2 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$351$ $[ 0, 2 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$351$ $[ 6, 2 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$351$ $[ 6, 6 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$351$ $[ 6, 6 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$351$ $[ 12, 2 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$352$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$352$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$352$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$352$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$352$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$352$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$353$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$353$ $[ 88 ]$ $4$ $2$ $[ 353 ]$ $[ 2 ]$
$354$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$354$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$354$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$354$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$354$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$354$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$355$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$355$ $[ 2, 0 ]$ $2$ $2$ $[]$ $[]$
$356$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$356$ $[ 0, 22 ]$ $4$ $2$ $[ 89 ]$ $[ 2 ]$
$357$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$357$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$357$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$357$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$357$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$357$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$357$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$357$ $[ 1, 3, 4 ]$ $4$ $4$ $[ 17, 21 ]$ $[ 2, 5 ]$
$357$ $[ 1, 3, 4 ]$ $4$ $4$ $[ 17, 21 ]$ $[ 2, 5 ]$
$357$ $[ 1, 1, 0 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$357$ $[ 1, 1, 0 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$357$ $[ 1, 1, 0 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$357$ $[ 1, 1, 0 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$358$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$358$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$359$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$359$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$360$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$360$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$360$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$360$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$360$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$360$ $[ 0, 0, 0, 2 ]$ $2$ $2$ $[]$ $[]$
$360$ $[ 0, 0, 0, 2 ]$ $2$ $2$ $[]$ $[]$
$360$ $[ 0, 0, 0, 2 ]$ $2$ $2$ $[]$ $[]$
$360$ $[ 0, 0, 0, 2 ]$ $2$ $2$ $[]$ $[]$
$360$ $[ 0, 1, 0, 0 ]$ $2$ $2$ $[]$ $[]$
$360$ $[ 0, 1, 0, 0 ]$ $2$ $2$ $[]$ $[]$
$360$ $[ 0, 1, 0, 0 ]$ $2$ $2$ $[]$ $[]$
$360$ $[ 0, 1, 0, 0 ]$ $2$ $4$ $[ -24 ]$ $[ 7 ]$
$360$ $[ 0, 1, 0, 2 ]$ $2$ $4$ $[ 5, 40 ]$ $[ 2, 3 ]$
$360$ $[ 1, 1, 3, 0 ]$ $2$ $2$ $[ 24 ]$ $[ 2 ]$
$360$ $[ 1, 1, 3, 0 ]$ $2$ $2$ $[ 24 ]$ $[ 2 ]$
$360$ $[ 0, 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$360$ $[ 0, 0, 3, 1 ]$ $4$ $4$ $[ -15 ]$ $[ 2 ]$
$360$ $[ 0, 0, 2, 2 ]$ $6$ $4$ $[]$ $[]$
$360$ $[ 0, 1, 2, 0 ]$ $6$ $4$ $[]$ $[]$
$361$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$361$ $[ 114 ]$ $3$ $2$ $[]$ $[]$
$361$ $[ 114 ]$ $3$ $2$ $[]$ $[]$
$361$ $[ 38 ]$ $9$ $6$ $[]$ $[]$
$361$ $[ 38 ]$ $9$ $6$ $[]$ $[]$
$362$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$362$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$362$ $[ 90 ]$ $2$ $2$ $[]$ $[]$
$363$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$363$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$363$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$363$ $[ 0, 0 ]$ $1$ $2$ $[ -11 ]$ $[ 3 ]$
$363$ $[ 0, 0 ]$ $1$ $2$ $[ -11 ]$ $[ 5 ]$
$363$ $[ 1, 55 ]$ $2$ $2$ $[ 33 ]$ $[ 2 ]$
$363$ $[ 1, 55 ]$ $2$ $2$ $[ 33 ]$ $[ 2 ]$
$363$ $[ 0, 22 ]$ $5$ $4$ $[]$ $[]$
$363$ $[ 0, 22 ]$ $5$ $4$ $[]$ $[]$
$363$ $[ 0, 22 ]$ $5$ $4$ $[]$ $[]$
$363$ $[ 0, 22 ]$ $5$ $4$ $[]$ $[]$
$363$ $[ 0, 22 ]$ $5$ $4$ $[ -11 ]$ $[ 5 ]$
$363$ $[ 0, 22 ]$ $5$ $4$ $[ -11 ]$ $[ 5 ]$
$363$ $[ 0, 22 ]$ $5$ $8$ $[ -11 ]$ $[ 3 ]$
$363$ $[ 1, 11 ]$ $10$ $8$ $[ 33 ]$ $[ 2 ]$
$363$ $[ 1, 11 ]$ $10$ $8$ $[ 33 ]$ $[ 2 ]$
$364$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$364$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$364$ $[ 1, 3, 0 ]$ $2$ $4$ $[ -4 ]$ $[ 7 ]$
$364$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$364$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$364$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$364$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$364$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$364$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$364$ $[ 1, 3, 4 ]$ $6$ $8$ $[ -4 ]$ $[ 7 ]$
$366$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$366$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$366$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$366$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$366$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$366$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$366$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$366$ $[ 0, 20 ]$ $3$ $2$ $[]$ $[]$
$366$ $[ 0, 20 ]$ $3$ $2$ $[]$ $[]$
$366$ $[ 1, 15 ]$ $4$ $4$ $[ -183 ]$ $[ 2 ]$
$368$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$368$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$368$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$368$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$368$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$368$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$368$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$368$ $[ 0, 1, 0 ]$ $4$ $2$ $[ 8 ]$ $[ 2 ]$
$369$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$369$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$369$ $[ 0, 20 ]$ $2$ $2$ $[ 41 ]$ $[ 2 ]$
$369$ $[ 0, 20 ]$ $2$ $2$ $[ 41 ]$ $[ 5 ]$
$369$ $[ 0, 20 ]$ $2$ $2$ $[ 41 ]$ $[ 5 ]$
$369$ $[ 0, 10 ]$ $4$ $2$ $[ 41 ]$ $[ 2 ]$
$369$ $[ 0, 10 ]$ $4$ $2$ $[ 41 ]$ $[ 2 ]$
$370$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$370$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$370$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$370$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$370$ $[ 0, 18 ]$ $2$ $2$ $[]$ $[]$
$370$ $[ 0, 18 ]$ $2$ $2$ $[]$ $[]$
$370$ $[ 2, 0 ]$ $2$ $2$ $[]$ $[]$
$370$ $[ 2, 0 ]$ $2$ $2$ $[]$ $[]$
$370$ $[ 0, 12 ]$ $3$ $2$ $[]$ $[]$
$370$ $[ 0, 12 ]$ $3$ $2$ $[]$ $[]$
$370$ $[ 0, 12 ]$ $3$ $2$ $[]$ $[]$
$370$ $[ 1, 9 ]$ $4$ $2$ $[ 185 ]$ $[ 2 ]$
$370$ $[ 1, 9 ]$ $4$ $2$ $[ 185 ]$ $[ 2 ]$
$370$ $[ 3, 9 ]$ $4$ $2$ $[ 185 ]$ $[ 2 ]$
$370$ $[ 3, 9 ]$ $4$ $2$ $[ 185 ]$ $[ 2 ]$
$370$ $[ 2, 12 ]$ $6$ $4$ $[]$ $[]$
$370$ $[ 2, 12 ]$ $6$ $4$ $[]$ $[]$
$371$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$371$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$371$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$372$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$372$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$372$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$372$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$372$ $[ 1, 1, 0 ]$ $2$ $4$ $[ -3 ]$ $[ 2 ]$
$373$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$374$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$377$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$377$ $[ 0, 14 ]$ $2$ $2$ $[ 29 ]$ $[ 5 ]$
$377$ $[ 6, 0 ]$ $2$ $2$ $[ 13 ]$ $[ 3 ]$
$377$ $[ 6, 0 ]$ $2$ $2$ $[ 13 ]$ $[ 3 ]$
$377$ $[ 4, 0 ]$ $3$ $2$ $[]$ $[]$
$377$ $[ 3, 7 ]$ $4$ $2$ $[ 377 ]$ $[ 2 ]$
$377$ $[ 9, 7 ]$ $4$ $2$ $[ 377 ]$ $[ 2 ]$
$378$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$378$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$378$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$378$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$378$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$378$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$378$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$378$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$378$ $[ 9, 3 ]$ $2$ $4$ $[ -7 ]$ $[ 3 ]$
$378$ $[ 9, 3 ]$ $2$ $4$ $[ -7 ]$ $[ 3 ]$
$378$ $[ 9, 3 ]$ $2$ $4$ $[ -7 ]$ $[ 3 ]$
$378$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$378$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$378$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$378$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$378$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$378$ $[ 0, 2 ]$ $3$ $2$ $[]$ $[]$
$378$ $[ 6, 0 ]$ $3$ $2$ $[]$ $[]$
$378$ $[ 6, 0 ]$ $3$ $2$ $[]$ $[]$
$378$ $[ 6, 2 ]$ $3$ $2$ $[]$ $[]$
$378$ $[ 6, 2 ]$ $3$ $2$ $[]$ $[]$
$378$ $[ 12, 2 ]$ $3$ $2$ $[]$ $[]$
$378$ $[ 12, 2 ]$ $3$ $2$ $[]$ $[]$
$378$ $[ 9, 1 ]$ $6$ $4$ $[ -7 ]$ $[ 3 ]$
$378$ $[ 9, 1 ]$ $6$ $4$ $[ -7 ]$ $[ 3 ]$
$378$ $[ 9, 1 ]$ $6$ $4$ $[ -7 ]$ $[ 3 ]$
$380$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$380$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$380$ $[ 0, 0, 6 ]$ $3$ $2$ $[]$ $[]$
$380$ $[ 1, 1, 0 ]$ $4$ $2$ $[ 5 ]$ $[ 2 ]$
$380$ $[ 1, 1, 0 ]$ $4$ $2$ $[ 5 ]$ $[ 2 ]$
$381$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$381$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$384$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$384$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$384$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$384$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$384$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$384$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$384$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$384$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$384$ $[ 0, 16, 0 ]$ $2$ $2$ $[]$ $[]$
$384$ $[ 0, 16, 0 ]$ $2$ $2$ $[]$ $[]$
$384$ $[ 0, 16, 0 ]$ $2$ $4$ $[ -8 ]$ $[ 2 ]$
$384$ $[ 1, 16, 1 ]$ $2$ $4$ $[ 12 ]$ $[ 2 ]$
$384$ $[ 1, 16, 1 ]$ $2$ $4$ $[ 12 ]$ $[ 2 ]$
$385$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$385$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$385$ $[ 2, 0, 0 ]$ $2$ $2$ $[]$ $[]$
$385$ $[ 2, 0, 0 ]$ $2$ $2$ $[]$ $[]$
$385$ $[ 2, 2, 0 ]$ $6$ $4$ $[]$ $[]$
$386$ $[ 48 ]$ $4$ $2$ $[ 193 ]$ $[ 2 ]$
$386$ $[ 32 ]$ $6$ $2$ $[ 193 ]$ $[ 3 ]$
$387$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$387$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$387$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$387$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$387$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$387$ $[ 0, 0 ]$ $1$ $2$ $[ -3 ]$ $[ 3 ]$
$387$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$387$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$387$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$388$ $[ 0, 48 ]$ $2$ $2$ $[ 97 ]$ $[ 2 ]$
$388$ $[ 0, 16 ]$ $6$ $2$ $[ 97 ]$ $[ 3 ]$
$389$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$390$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$390$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$390$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$390$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$390$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$390$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$390$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$390$ $[ 0, 0, 6 ]$ $2$ $2$ $[]$ $[]$
$390$ $[ 0, 0, 6 ]$ $2$ $2$ $[]$ $[]$
$390$ $[ 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$390$ $[ 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$390$ $[ 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$390$ $[ 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$390$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$390$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$390$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$390$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$390$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$390$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$390$ $[ 1, 0, 3 ]$ $4$ $4$ $[ -39 ]$ $[ 2 ]$
$390$ $[ 1, 0, 3 ]$ $4$ $4$ $[ -39 ]$ $[ 2 ]$
$390$ $[ 1, 0, 3 ]$ $4$ $4$ $[ -39 ]$ $[ 2 ]$
$390$ $[ 1, 0, 3 ]$ $4$ $4$ $[ -39 ]$ $[ 2 ]$
$390$ $[ 1, 1, 6 ]$ $4$ $8$ $[ 65, -39 ]$ $[ 2, 5 ]$
$390$ $[ 0, 2, 4 ]$ $6$ $4$ $[]$ $[]$
$390$ $[ 0, 2, 4 ]$ $6$ $4$ $[]$ $[]$
$390$ $[ 0, 2, 4 ]$ $6$ $4$ $[]$ $[]$
$390$ $[ 0, 2, 4 ]$ $6$ $4$ $[]$ $[]$
$392$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$392$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$392$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$392$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$392$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$392$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$392$ $[ 0, 0, 0 ]$ $1$ $2$ $[ -7 ]$ $[ 2 ]$
$392$ $[ 0, 0, 0 ]$ $1$ $2$ $[ -7 ]$ $[ 2 ]$
$392$ $[ 0, 1, 0 ]$ $2$ $2$ $[ 8 ]$ $[ 2 ]$
$392$ $[ 0, 0, 14 ]$ $3$ $2$ $[]$ $[]$
$392$ $[ 0, 0, 14 ]$ $3$ $2$ $[]$ $[]$
$392$ $[ 0, 0, 14 ]$ $3$ $2$ $[]$ $[]$
$392$ $[ 0, 0, 14 ]$ $3$ $2$ $[]$ $[]$
$392$ $[ 0, 0, 14 ]$ $3$ $2$ $[]$ $[]$
$392$ $[ 0, 0, 14 ]$ $3$ $2$ $[]$ $[]$
$392$ $[ 0, 0, 14 ]$ $3$ $4$ $[ -7 ]$ $[ 2 ]$
$392$ $[ 0, 0, 14 ]$ $3$ $4$ $[ -7 ]$ $[ 2 ]$
$392$ $[ 0, 1, 14 ]$ $6$ $4$ $[ 8 ]$ $[ 2 ]$
$392$ $[ 0, 1, 14 ]$ $6$ $4$ $[ 8 ]$ $[ 2 ]$
$395$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$395$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$395$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$395$ $[ 2, 0 ]$ $2$ $2$ $[]$ $[]$
$395$ $[ 0, 26 ]$ $3$ $2$ $[]$ $[]$
$396$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$396$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$396$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$396$ $[ 0, 3, 5 ]$ $2$ $2$ $[ 33 ]$ $[ 2 ]$
$396$ $[ 0, 3, 5 ]$ $2$ $2$ $[ 33 ]$ $[ 2 ]$
$396$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$396$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$396$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$397$ $[ 198 ]$ $2$ $2$ $[]$ $[]$
$397$ $[ 132 ]$ $3$ $2$ $[]$ $[]$
$397$ $[ 66 ]$ $6$ $2$ $[ 397 ]$ $[ 3 ]$
$398$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$399$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$399$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$399$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$399$ $[ 0, 3, 9 ]$ $2$ $2$ $[ 133 ]$ $[ 3 ]$
$399$ $[ 0, 3, 9 ]$ $2$ $2$ $[ 133 ]$ $[ 3 ]$
$399$ $[ 1, 0, 9 ]$ $2$ $4$ $[ 57, -3 ]$ $[ 2, 3 ]$
$399$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$399$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$399$ $[ 0, 2, 6 ]$ $3$ $2$ $[]$ $[]$
$399$ $[ 0, 4, 6 ]$ $3$ $2$ $[]$ $[]$
$399$ $[ 0, 1, 3 ]$ $6$ $2$ $[ 133 ]$ $[ 3 ]$
$399$ $[ 0, 1, 9 ]$ $6$ $2$ $[ 133 ]$ $[ 3 ]$
$399$ $[ 0, 1, 9 ]$ $6$ $2$ $[ 133 ]$ $[ 3 ]$
$399$ $[ 0, 3, 3 ]$ $6$ $2$ $[ 133 ]$ $[ 3 ]$
$399$ $[ 0, 3, 3 ]$ $6$ $2$ $[ 133 ]$ $[ 3 ]$
$399$ $[ 0, 3, 3 ]$ $6$ $2$ $[ 133 ]$ $[ 3 ]$
$399$ $[ 0, 3, 3 ]$ $6$ $2$ $[ 133 ]$ $[ 3 ]$
$399$ $[ 0, 5, 3 ]$ $6$ $2$ $[ 133 ]$ $[ 3 ]$
$400$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$400$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$400$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$400$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$400$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$400$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$400$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$400$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$400$ $[ 0, 0, 10 ]$ $2$ $2$ $[]$ $[]$
$400$ $[ 0, 0, 10 ]$ $2$ $2$ $[]$ $[]$
$400$ $[ 0, 0, 10 ]$ $2$ $2$ $[]$ $[]$
$400$ $[ 0, 0, 10 ]$ $2$ $2$ $[]$ $[]$
$400$ $[ 0, 1, 0 ]$ $4$ $2$ $[ 8 ]$ $[ 2 ]$
$400$ $[ 0, 1, 0 ]$ $4$ $2$ $[ 8 ]$ $[ 2 ]$
$400$ $[ 0, 1, 0 ]$ $4$ $2$ $[ 8 ]$ $[ 2 ]$
$400$ $[ 0, 1, 10 ]$ $4$ $2$ $[ 8 ]$ $[ 2 ]$
$400$ $[ 0, 1, 10 ]$ $4$ $2$ $[ 8 ]$ $[ 2 ]$
$400$ $[ 1, 0, 5 ]$ $4$ $8$ $[ -20 ]$ $[ 3 ]$
$400$ $[ 1, 1, 5 ]$ $4$ $2$ $[ 40 ]$ $[ 2 ]$
$400$ $[ 1, 3, 5 ]$ $4$ $2$ $[ 40 ]$ $[ 2 ]$
$402$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$402$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$402$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$402$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$403$ $[ 6, 0 ]$ $2$ $2$ $[]$ $[]$
$403$ $[ 4, 0 ]$ $3$ $2$ $[]$ $[]$
$403$ $[ 6, 10 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$403$ $[ 6, 10 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$404$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$404$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$404$ $[ 0, 50 ]$ $2$ $2$ $[]$ $[]$
$405$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$405$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$405$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$405$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$405$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$405$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$405$ $[ 18, 0 ]$ $3$ $2$ $[]$ $[]$
$405$ $[ 18, 0 ]$ $3$ $2$ $[]$ $[]$
$405$ $[ 18, 0 ]$ $3$ $2$ $[]$ $[]$
$405$ $[ 18, 0 ]$ $3$ $2$ $[]$ $[]$
$405$ $[ 18, 0 ]$ $3$ $2$ $[]$ $[]$
$405$ $[ 18, 0 ]$ $3$ $2$ $[]$ $[]$
$405$ $[ 18, 0 ]$ $3$ $2$ $[]$ $[]$
$405$ $[ 18, 0 ]$ $3$ $2$ $[]$ $[]$
$405$ $[ 9, 1 ]$ $12$ $8$ $[ -15 ]$ $[ 2 ]$
$406$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$406$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$406$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$406$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$406$ $[ 0, 14 ]$ $2$ $2$ $[]$ $[]$
$407$ $[ 0, 12 ]$ $3$ $2$ $[]$ $[]$
$407$ $[ 0, 12 ]$ $3$ $2$ $[]$ $[]$
$407$ $[ 0, 6 ]$ $6$ $2$ $[ 37 ]$ $[ 3 ]$
$408$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$408$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$408$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$408$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$408$ $[ 0, 1, 0, 0 ]$ $2$ $2$ $[]$ $[]$
$408$ $[ 0, 1, 0, 4 ]$ $4$ $4$ $[ 136 ]$ $[ 2 ]$
$410$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$410$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$410$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$410$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$410$ $[ 0, 20 ]$ $2$ $2$ $[ 41 ]$ $[ 2 ]$
$410$ $[ 0, 20 ]$ $2$ $2$ $[]$ $[]$
$410$ $[ 0, 20 ]$ $2$ $2$ $[ 41 ]$ $[ 2 ]$
$410$ $[ 0, 10 ]$ $4$ $2$ $[ 41 ]$ $[ 2 ]$
$410$ $[ 0, 10 ]$ $4$ $2$ $[ 41 ]$ $[ 2 ]$
$410$ $[ 0, 10 ]$ $4$ $2$ $[ 41 ]$ $[ 2 ]$
$410$ $[ 0, 10 ]$ $4$ $2$ $[ 41 ]$ $[ 2 ]$
$410$ $[ 0, 10 ]$ $4$ $2$ $[ 41 ]$ $[ 2 ]$
$410$ $[ 0, 10 ]$ $4$ $2$ $[ 41 ]$ $[ 2 ]$
$410$ $[ 2, 10 ]$ $4$ $2$ $[ 41 ]$ $[ 2 ]$
$410$ $[ 2, 10 ]$ $4$ $2$ $[ 41 ]$ $[ 2 ]$
$410$ $[ 2, 10 ]$ $4$ $2$ $[ 41 ]$ $[ 2 ]$
$410$ $[ 2, 10 ]$ $4$ $2$ $[ 41 ]$ $[ 2 ]$
$413$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$414$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$414$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$414$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$414$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$414$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$415$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$416$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$416$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$416$ $[ 0, 0, 0 ]$ $1$ $2$ $[ -4 ]$ $[ 5 ]$
$416$ $[ 0, 0, 6 ]$ $2$ $2$ $[]$ $[]$
$416$ $[ 0, 0, 6 ]$ $2$ $2$ $[]$ $[]$
$416$ $[ 0, 0, 6 ]$ $2$ $4$ $[ -4 ]$ $[ 5 ]$
$416$ $[ 0, 4, 0 ]$ $2$ $2$ $[]$ $[]$
$416$ $[ 0, 4, 6 ]$ $2$ $2$ $[]$ $[]$
$416$ $[ 0, 4, 6 ]$ $2$ $2$ $[]$ $[]$
$416$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$416$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$416$ $[ 1, 0, 3 ]$ $4$ $2$ $[ 13 ]$ $[ 2 ]$
$416$ $[ 1, 0, 3 ]$ $4$ $2$ $[ 13 ]$ $[ 2 ]$
$417$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$417$ $[ 0, 46 ]$ $3$ $2$ $[]$ $[]$
$418$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$418$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$418$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$418$ $[ 5, 9 ]$ $2$ $2$ $[ 209 ]$ $[ 10 ]$
$418$ $[ 5, 9 ]$ $2$ $2$ $[ 209 ]$ $[ 10 ]$
$418$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$418$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$418$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$418$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$418$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$418$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$420$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$420$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$420$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$420$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$420$ $[ 0, 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$420$ $[ 0, 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$420$ $[ 0, 1, 0, 3 ]$ $2$ $2$ $[ 21 ]$ $[ 3 ]$
$420$ $[ 0, 1, 0, 3 ]$ $2$ $2$ $[ 21 ]$ $[ 3 ]$
$420$ $[ 1, 1, 2, 0 ]$ $2$ $4$ $[ -3 ]$ $[ 2 ]$
$420$ $[ 1, 1, 2, 0 ]$ $2$ $4$ $[ -3 ]$ $[ 2 ]$
$420$ $[ 0, 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$420$ $[ 0, 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$420$ $[ 1, 0, 1, 0 ]$ $4$ $4$ $[ -20 ]$ $[ 2 ]$
$420$ $[ 1, 0, 1, 0 ]$ $4$ $4$ $[ -20 ]$ $[ 2 ]$
$421$ $[ 70 ]$ $6$ $2$ $[ 421 ]$ $[ 3 ]$
$422$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$423$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$423$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$423$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$423$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$423$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$423$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$423$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$423$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$423$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$423$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$424$ $[ 0, 1, 0 ]$ $2$ $2$ $[]$ $[]$
$425$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$425$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$425$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$425$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$425$ $[ 10, 0 ]$ $2$ $2$ $[]$ $[]$
$425$ $[ 10, 0 ]$ $2$ $2$ $[]$ $[]$
$425$ $[ 10, 0 ]$ $2$ $2$ $[]$ $[]$
$425$ $[ 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$425$ $[ 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$426$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$426$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$426$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$427$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$427$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$427$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$427$ $[ 0, 20 ]$ $3$ $2$ $[]$ $[]$
$428$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$428$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$429$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$429$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$429$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$429$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$430$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$430$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$430$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$430$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$430$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$430$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$430$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$430$ $[ 1, 21 ]$ $4$ $4$ $[ -215 ]$ $[ 2 ]$
$431$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$431$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$432$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$432$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$432$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$432$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$432$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$432$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$432$ $[ 1, 0, 9 ]$ $2$ $4$ $[ -4 ]$ $[ 3 ]$
$432$ $[ 0, 0, 6 ]$ $3$ $2$ $[]$ $[]$
$432$ $[ 0, 0, 6 ]$ $3$ $2$ $[]$ $[]$
$432$ $[ 0, 0, 6 ]$ $3$ $2$ $[]$ $[]$
$432$ $[ 1, 0, 3 ]$ $6$ $2$ $[ 12 ]$ $[ 3 ]$
$432$ $[ 1, 0, 3 ]$ $6$ $2$ $[ 12 ]$ $[ 3 ]$
$432$ $[ 1, 0, 3 ]$ $6$ $2$ $[ 12 ]$ $[ 3 ]$
$432$ $[ 1, 0, 3 ]$ $6$ $2$ $[ 12 ]$ $[ 3 ]$
$433$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$433$ $[ 144 ]$ $3$ $2$ $[]$ $[]$
$434$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$434$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$434$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$434$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$434$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$434$ $[ 3, 15 ]$ $2$ $2$ $[ 217 ]$ $[ 3 ]$
$434$ $[ 3, 15 ]$ $2$ $2$ $[ 217 ]$ $[ 3 ]$
$434$ $[ 3, 15 ]$ $2$ $2$ $[ 217 ]$ $[ 3 ]$
$434$ $[ 3, 15 ]$ $2$ $2$ $[ 217 ]$ $[ 3 ]$
$434$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$434$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$434$ $[ 2, 10 ]$ $3$ $2$ $[]$ $[]$
$434$ $[ 2, 10 ]$ $3$ $2$ $[]$ $[]$
$434$ $[ 4, 10 ]$ $3$ $2$ $[]$ $[]$
$434$ $[ 4, 10 ]$ $3$ $2$ $[]$ $[]$
$434$ $[ 1, 5 ]$ $6$ $2$ $[ 217 ]$ $[ 3 ]$
$434$ $[ 1, 5 ]$ $6$ $2$ $[ 217 ]$ $[ 3 ]$
$434$ $[ 1, 15 ]$ $6$ $4$ $[ 217, -31 ]$ $[ 3, 7 ]$
$434$ $[ 3, 5 ]$ $6$ $2$ $[ 217 ]$ $[ 3 ]$
$434$ $[ 3, 5 ]$ $6$ $2$ $[ 217 ]$ $[ 3 ]$
$434$ $[ 3, 5 ]$ $6$ $2$ $[ 217 ]$ $[ 3 ]$
$434$ $[ 3, 5 ]$ $6$ $2$ $[ 217 ]$ $[ 3 ]$
$434$ $[ 5, 5 ]$ $6$ $2$ $[ 217 ]$ $[ 3 ]$
$434$ $[ 5, 5 ]$ $6$ $2$ $[ 217 ]$ $[ 3 ]$
$435$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$435$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$435$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$435$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$435$ $[ 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$435$ $[ 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$435$ $[ 0, 2, 14 ]$ $2$ $2$ $[]$ $[]$
$435$ $[ 0, 2, 14 ]$ $2$ $2$ $[]$ $[]$
$435$ $[ 0, 2, 14 ]$ $2$ $2$ $[]$ $[]$
$435$ $[ 0, 2, 14 ]$ $2$ $2$ $[]$ $[]$
$435$ $[ 0, 1, 7 ]$ $4$ $2$ $[ 145 ]$ $[ 2 ]$
$435$ $[ 0, 3, 7 ]$ $4$ $2$ $[ 145 ]$ $[ 2 ]$
$436$ $[ 0, 54 ]$ $2$ $2$ $[]$ $[]$
$437$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$437$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$438$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$438$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$438$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$438$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$438$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$438$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$438$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$438$ $[ 0, 36 ]$ $2$ $2$ $[ 73 ]$ $[ 3 ]$
$438$ $[ 0, 36 ]$ $2$ $2$ $[ 73 ]$ $[ 2 ]$
$438$ $[ 0, 36 ]$ $2$ $2$ $[ 73 ]$ $[ 6 ]$
$438$ $[ 0, 24 ]$ $3$ $2$ $[]$ $[]$
$438$ $[ 0, 24 ]$ $3$ $2$ $[]$ $[]$
$438$ $[ 0, 24 ]$ $3$ $2$ $[]$ $[]$
$438$ $[ 0, 24 ]$ $3$ $2$ $[]$ $[]$
$438$ $[ 0, 18 ]$ $4$ $2$ $[ 73 ]$ $[ 2 ]$
$438$ $[ 0, 18 ]$ $4$ $2$ $[ 73 ]$ $[ 2 ]$
$438$ $[ 0, 18 ]$ $4$ $2$ $[ 73 ]$ $[ 2 ]$
$438$ $[ 0, 18 ]$ $4$ $2$ $[ 73 ]$ $[ 2 ]$
$438$ $[ 0, 18 ]$ $4$ $2$ $[ 73 ]$ $[ 2 ]$
$438$ $[ 0, 18 ]$ $4$ $2$ $[ 73 ]$ $[ 2 ]$
$438$ $[ 0, 18 ]$ $4$ $2$ $[ 73 ]$ $[ 2 ]$
$438$ $[ 0, 12 ]$ $6$ $2$ $[ 73 ]$ $[ 3 ]$
$438$ $[ 0, 12 ]$ $6$ $2$ $[ 73 ]$ $[ 3 ]$
$438$ $[ 0, 12 ]$ $6$ $2$ $[ 73 ]$ $[ 3 ]$
$438$ $[ 0, 12 ]$ $6$ $2$ $[ 73 ]$ $[ 3 ]$
$440$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$440$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$440$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$440$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$440$ $[ 0, 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$440$ $[ 0, 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$440$ $[ 0, 0, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$440$ $[ 0, 1, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$440$ $[ 0, 1, 2, 0 ]$ $2$ $2$ $[]$ $[]$
$440$ $[ 0, 0, 1, 5 ]$ $4$ $4$ $[ -55 ]$ $[ 2 ]$
$440$ $[ 1, 1, 1, 0 ]$ $4$ $2$ $[ 5 ]$ $[ 2 ]$
$440$ $[ 1, 1, 1, 0 ]$ $4$ $2$ $[ 5 ]$ $[ 2 ]$
$440$ $[ 1, 1, 1, 0 ]$ $4$ $2$ $[ 5 ]$ $[ 2 ]$
$440$ $[ 1, 1, 1, 0 ]$ $4$ $2$ $[ 5 ]$ $[ 2 ]$
$441$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$441$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$441$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$441$ $[ 0, 0 ]$ $1$ $2$ $[ -3 ]$ $[ 3 ]$
$441$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$441$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$441$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$441$ $[ 0, 14 ]$ $3$ $4$ $[ -3 ]$ $[ 3 ]$
$441$ $[ 0, 14 ]$ $3$ $4$ $[ -3 ]$ $[ 3 ]$
$441$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$441$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$441$ $[ 2, 14 ]$ $3$ $2$ $[]$ $[]$
$441$ $[ 4, 14 ]$ $3$ $2$ $[]$ $[]$
$441$ $[ 1, 7 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$441$ $[ 1, 21 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$441$ $[ 1, 21 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$441$ $[ 5, 7 ]$ $6$ $2$ $[ 21 ]$ $[ 3 ]$
$442$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$442$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$442$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$442$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$442$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$442$ $[ 0, 8 ]$ $2$ $2$ $[]$ $[]$
$442$ $[ 0, 8 ]$ $2$ $2$ $[ 17 ]$ $[ 2 ]$
$442$ $[ 6, 8 ]$ $2$ $4$ $[ 17 ]$ $[ 2 ]$
$442$ $[ 4, 0 ]$ $3$ $2$ $[]$ $[]$
$442$ $[ 4, 0 ]$ $3$ $2$ $[]$ $[]$
$442$ $[ 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$442$ $[ 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$442$ $[ 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$442$ $[ 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$442$ $[ 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$442$ $[ 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$442$ $[ 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$442$ $[ 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$442$ $[ 0, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$442$ $[ 6, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$442$ $[ 6, 4 ]$ $4$ $2$ $[ 17 ]$ $[ 2 ]$
$443$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$443$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$443$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$444$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$444$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$444$ $[ 0, 0, 12 ]$ $3$ $2$ $[]$ $[]$
$444$ $[ 1, 0, 9 ]$ $4$ $2$ $[ 37 ]$ $[ 2 ]$
$444$ $[ 1, 0, 9 ]$ $4$ $2$ $[ 37 ]$ $[ 2 ]$
$444$ $[ 0, 0, 6 ]$ $6$ $2$ $[ 37 ]$ $[ 3 ]$
$444$ $[ 0, 0, 6 ]$ $6$ $2$ $[ 37 ]$ $[ 3 ]$
$445$ $[ 0, 44 ]$ $2$ $2$ $[]$ $[]$
$445$ $[ 0, 22 ]$ $4$ $2$ $[ 89 ]$ $[ 2 ]$
$445$ $[ 0, 22 ]$ $4$ $2$ $[ 89 ]$ $[ 2 ]$
$446$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$446$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$446$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$446$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$446$ $[ 74 ]$ $3$ $2$ $[]$ $[]$
$446$ $[ 74 ]$ $3$ $2$ $[]$ $[]$
$446$ $[ 74 ]$ $3$ $2$ $[]$ $[]$
$448$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$448$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$448$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$448$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$448$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$448$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$448$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$448$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$448$ $[ 0, 8, 0 ]$ $2$ $2$ $[]$ $[]$
$448$ $[ 0, 8, 0 ]$ $2$ $2$ $[]$ $[]$
$448$ $[ 1, 0, 3 ]$ $2$ $2$ $[ 28 ]$ $[ 3 ]$
$448$ $[ 1, 0, 3 ]$ $2$ $2$ $[ 28 ]$ $[ 3 ]$
$448$ $[ 1, 8, 3 ]$ $2$ $8$ $[ -7, -56 ]$ $[ 2, 3 ]$
$448$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$448$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$448$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$448$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$448$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$448$ $[ 0, 0, 2 ]$ $3$ $2$ $[]$ $[]$
$448$ $[ 0, 4, 0 ]$ $4$ $2$ $[ 8 ]$ $[ 2 ]$
$448$ $[ 0, 4, 0 ]$ $4$ $2$ $[ 8 ]$ $[ 2 ]$
$448$ $[ 1, 0, 1 ]$ $6$ $2$ $[ 28 ]$ $[ 3 ]$
$448$ $[ 1, 0, 1 ]$ $6$ $2$ $[ 28 ]$ $[ 3 ]$
$450$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$450$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$450$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$450$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$450$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$450$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$450$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$450$ $[ 0, 10 ]$ $2$ $2$ $[]$ $[]$
$450$ $[ 0, 10 ]$ $2$ $2$ $[]$ $[]$
$450$ $[ 0, 10 ]$ $2$ $2$ $[]$ $[]$
$450$ $[ 0, 10 ]$ $2$ $2$ $[]$ $[]$
$450$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$450$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$450$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$450$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$450$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$450$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$450$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$450$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$450$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$450$ $[ 3, 5 ]$ $4$ $4$ $[ -15 ]$ $[ 2 ]$
$450$ $[ 3, 5 ]$ $4$ $4$ $[ -15 ]$ $[ 2 ]$
$450$ $[ 3, 5 ]$ $4$ $4$ $[ -15 ]$ $[ 2 ]$
$450$ $[ 2, 10 ]$ $6$ $4$ $[]$ $[]$
$450$ $[ 2, 10 ]$ $6$ $4$ $[]$ $[]$
$450$ $[ 2, 10 ]$ $6$ $4$ $[]$ $[]$
$450$ $[ 2, 10 ]$ $6$ $4$ $[]$ $[]$
$450$ $[ 2, 10 ]$ $6$ $4$ $[]$ $[]$
$450$ $[ 1, 5 ]$ $12$ $8$ $[ -15 ]$ $[ 3 ]$
$450$ $[ 1, 5 ]$ $12$ $8$ $[ -15 ]$ $[ 3 ]$
$450$ $[ 1, 5 ]$ $12$ $8$ $[ -15 ]$ $[ 3 ]$
$450$ $[ 1, 5 ]$ $12$ $8$ $[ -15 ]$ $[ 3 ]$
$451$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$455$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$455$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$455$ $[ 2, 0, 0 ]$ $2$ $2$ $[]$ $[]$
$455$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$455$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$455$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$455$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$455$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$455$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$455$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$455$ $[ 2, 0, 4 ]$ $6$ $4$ $[]$ $[]$
$456$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$456$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$456$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$456$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$456$ $[ 1, 1, 1, 0 ]$ $2$ $4$ $[ -3, -8 ]$ $[ 2, 3 ]$
$456$ $[ 0, 0, 0, 6 ]$ $3$ $2$ $[]$ $[]$
$456$ $[ 0, 0, 0, 6 ]$ $3$ $2$ $[]$ $[]$
$456$ $[ 0, 0, 0, 6 ]$ $3$ $2$ $[]$ $[]$
$458$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$458$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$458$ $[ 114 ]$ $2$ $2$ $[]$ $[]$
$458$ $[ 114 ]$ $2$ $2$ $[]$ $[]$
$458$ $[ 76 ]$ $3$ $2$ $[]$ $[]$
$458$ $[ 76 ]$ $3$ $2$ $[]$ $[]$
$458$ $[ 76 ]$ $3$ $2$ $[]$ $[]$
$458$ $[ 76 ]$ $3$ $2$ $[]$ $[]$
$459$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$459$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$459$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$459$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$459$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$459$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$459$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$459$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$460$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$460$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$460$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$460$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$461$ $[ 230 ]$ $2$ $2$ $[ 461 ]$ $[ 5 ]$
$462$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$462$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$462$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$462$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$462$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$462$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$462$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$462$ $[ 1, 3, 0 ]$ $2$ $4$ $[ -7 ]$ $[ 6 ]$
$462$ $[ 1, 3, 0 ]$ $2$ $4$ $[ -7 ]$ $[ 3 ]$
$462$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$462$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$462$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$462$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$462$ $[ 1, 1, 0 ]$ $6$ $4$ $[ -7 ]$ $[ 3 ]$
$462$ $[ 1, 1, 0 ]$ $6$ $4$ $[ -7 ]$ $[ 3 ]$
$462$ $[ 1, 1, 0 ]$ $6$ $4$ $[ -7 ]$ $[ 3 ]$
$464$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$464$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$464$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$464$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$464$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$464$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$464$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$464$ $[ 0, 0, 14 ]$ $2$ $2$ $[]$ $[]$
$464$ $[ 0, 0, 14 ]$ $2$ $2$ $[ 29 ]$ $[ 7 ]$
$464$ $[ 0, 0, 14 ]$ $2$ $2$ $[ 29 ]$ $[ 5 ]$
$464$ $[ 0, 1, 14 ]$ $4$ $2$ $[ 8 ]$ $[ 2 ]$
$464$ $[ 0, 1, 14 ]$ $4$ $2$ $[ 8 ]$ $[ 2 ]$
$465$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$465$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$465$ $[ 1, 2, 15 ]$ $2$ $8$ $[ -15, -155 ]$ $[ 2, 3 ]$
$465$ $[ 0, 0, 10 ]$ $3$ $2$ $[]$ $[]$
$465$ $[ 0, 2, 10 ]$ $6$ $4$ $[]$ $[]$
$466$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$466$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$466$ $[ 116 ]$ $2$ $2$ $[ 233 ]$ $[ 14 ]$
$466$ $[ 116 ]$ $2$ $2$ $[ 233 ]$ $[ 7 ]$
$466$ $[ 58 ]$ $4$ $2$ $[ 233 ]$ $[ 2 ]$
$467$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$468$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$468$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$468$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$468$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$468$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$468$ $[ 0, 0, 6 ]$ $2$ $2$ $[ 13 ]$ $[ 3 ]$
$468$ $[ 0, 0, 6 ]$ $2$ $2$ $[]$ $[]$
$468$ $[ 1, 3, 0 ]$ $2$ $4$ $[ -3 ]$ $[ 2 ]$
$468$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$468$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$468$ $[ 0, 0, 4 ]$ $3$ $2$ $[]$ $[]$
$468$ $[ 0, 0, 4 ]$ $3$ $4$ $[ -3 ]$ $[ 7 ]$
$468$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$468$ $[ 1, 0, 3 ]$ $4$ $2$ $[ 13 ]$ $[ 2 ]$
$468$ $[ 1, 0, 3 ]$ $4$ $2$ $[ 13 ]$ $[ 2 ]$
$468$ $[ 1, 0, 3 ]$ $4$ $2$ $[ 13 ]$ $[ 2 ]$
$468$ $[ 1, 0, 3 ]$ $4$ $2$ $[ 13 ]$ $[ 2 ]$
$468$ $[ 0, 0, 2 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$468$ $[ 0, 0, 2 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$468$ $[ 0, 2, 2 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$468$ $[ 0, 2, 6 ]$ $6$ $4$ $[]$ $[]$
$468$ $[ 0, 4, 2 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$469$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$469$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$469$ $[ 0, 22 ]$ $3$ $2$ $[]$ $[]$
$469$ $[ 1, 11 ]$ $6$ $2$ $[ 469 ]$ $[ 3 ]$
$469$ $[ 3, 11 ]$ $6$ $2$ $[ 469 ]$ $[ 3 ]$
$469$ $[ 3, 11 ]$ $6$ $2$ $[ 469 ]$ $[ 3 ]$
$469$ $[ 5, 11 ]$ $6$ $2$ $[ 469 ]$ $[ 3 ]$
$470$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$470$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$470$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$470$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$470$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$470$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$470$ $[ 2, 0 ]$ $2$ $2$ $[]$ $[]$
$470$ $[ 2, 0 ]$ $2$ $2$ $[]$ $[]$
$471$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$472$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$472$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$472$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$472$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$472$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$473$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$473$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$473$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$474$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$474$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$474$ $[ 0, 26 ]$ $3$ $2$ $[]$ $[]$
$474$ $[ 0, 26 ]$ $3$ $2$ $[]$ $[]$
$474$ $[ 0, 26 ]$ $3$ $2$ $[]$ $[]$
$474$ $[ 0, 26 ]$ $3$ $2$ $[]$ $[]$
$475$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$475$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$475$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$475$ $[ 10, 0 ]$ $2$ $2$ $[]$ $[]$
$475$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$475$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$475$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$475$ $[ 10, 6 ]$ $6$ $4$ $[]$ $[]$
$476$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$476$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$476$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$477$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$480$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$480$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$480$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$480$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$480$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$480$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$480$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$480$ $[ 0, 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$480$ $[ 0, 0, 0, 2 ]$ $2$ $2$ $[]$ $[]$
$480$ $[ 0, 0, 0, 2 ]$ $2$ $2$ $[]$ $[]$
$480$ $[ 0, 0, 0, 2 ]$ $2$ $2$ $[]$ $[]$
$480$ $[ 0, 0, 0, 2 ]$ $2$ $2$ $[]$ $[]$
$480$ $[ 0, 0, 0, 2 ]$ $2$ $4$ $[ -20 ]$ $[ 5 ]$
$480$ $[ 0, 4, 0, 0 ]$ $2$ $2$ $[]$ $[]$
$480$ $[ 0, 0, 1, 1 ]$ $4$ $4$ $[ -15 ]$ $[ 2 ]$
$480$ $[ 0, 0, 1, 1 ]$ $4$ $4$ $[ -15 ]$ $[ 2 ]$
$481$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$481$ $[ 0, 12 ]$ $3$ $2$ $[]$ $[]$
$481$ $[ 4, 12 ]$ $3$ $2$ $[]$ $[]$
$481$ $[ 8, 12 ]$ $3$ $2$ $[]$ $[]$
$481$ $[ 3, 9 ]$ $4$ $2$ $[ 481 ]$ $[ 2 ]$
$481$ $[ 9, 9 ]$ $4$ $2$ $[ 481 ]$ $[ 2 ]$
$481$ $[ 2, 12 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$481$ $[ 10, 12 ]$ $6$ $2$ $[ 13 ]$ $[ 3 ]$
$482$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$482$ $[ 80 ]$ $3$ $2$ $[]$ $[]$
$482$ $[ 40 ]$ $6$ $2$ $[ 241 ]$ $[ 3 ]$
$482$ $[ 40 ]$ $6$ $2$ $[ 241 ]$ $[ 3 ]$
$483$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$483$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$483$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$483$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$483$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$483$ $[ 0, 2, 0 ]$ $3$ $2$ $[]$ $[]$
$484$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$484$ $[ 0, 0 ]$ $1$ $2$ $[ -11 ]$ $[ 3 ]$
$484$ $[ 0, 22 ]$ $5$ $4$ $[]$ $[]$
$484$ $[ 0, 22 ]$ $5$ $4$ $[]$ $[]$
$484$ $[ 0, 22 ]$ $5$ $8$ $[ -11 ]$ $[ 3 ]$
$485$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$485$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$485$ $[ 2, 48 ]$ $2$ $2$ $[]$ $[]$
$485$ $[ 2, 48 ]$ $2$ $2$ $[]$ $[]$
$486$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$486$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$486$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$486$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$486$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$486$ $[ 0 ]$ $1$ $1$ $[]$ $[]$
$486$ $[ 54 ]$ $3$ $2$ $[]$ $[]$
$486$ $[ 54 ]$ $3$ $2$ $[]$ $[]$
$486$ $[ 54 ]$ $3$ $2$ $[]$ $[]$
$486$ $[ 54 ]$ $3$ $2$ $[]$ $[]$
$486$ $[ 54 ]$ $3$ $2$ $[]$ $[]$
$486$ $[ 54 ]$ $3$ $2$ $[]$ $[]$
$490$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$490$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$490$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$490$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$490$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$490$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$490$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$490$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$490$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$490$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$490$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$490$ $[ 2, 0 ]$ $2$ $2$ $[]$ $[]$
$490$ $[ 2, 0 ]$ $2$ $2$ $[]$ $[]$
$490$ $[ 2, 0 ]$ $2$ $2$ $[]$ $[]$
$490$ $[ 2, 0 ]$ $2$ $2$ $[]$ $[]$
$490$ $[ 2, 0 ]$ $2$ $4$ $[ -7 ]$ $[ 2 ]$
$490$ $[ 2, 0 ]$ $2$ $4$ $[ -7 ]$ $[ 2 ]$
$490$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$490$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$490$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$490$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$490$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$490$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$490$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$490$ $[ 0, 14 ]$ $3$ $2$ $[]$ $[]$
$490$ $[ 2, 14 ]$ $6$ $4$ $[]$ $[]$
$490$ $[ 2, 14 ]$ $6$ $4$ $[]$ $[]$
$490$ $[ 2, 14 ]$ $6$ $8$ $[ -7 ]$ $[ 2 ]$
$490$ $[ 2, 14 ]$ $6$ $8$ $[ -7 ]$ $[ 2 ]$
$492$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$492$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$493$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$493$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$494$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$494$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$494$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$494$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$494$ $[ 6, 0 ]$ $2$ $2$ $[]$ $[]$
$494$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$494$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$494$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$494$ $[ 0, 6 ]$ $3$ $2$ $[]$ $[]$
$494$ $[ 4, 0 ]$ $3$ $2$ $[]$ $[]$
$494$ $[ 4, 6 ]$ $3$ $2$ $[]$ $[]$
$494$ $[ 4, 6 ]$ $3$ $2$ $[]$ $[]$
$494$ $[ 8, 6 ]$ $3$ $2$ $[]$ $[]$
$494$ $[ 8, 6 ]$ $3$ $2$ $[]$ $[]$
$495$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$495$ $[ 3, 2, 5 ]$ $2$ $4$ $[ 33 ]$ $[ 2 ]$
$495$ $[ 3, 2, 5 ]$ $2$ $4$ $[ 33 ]$ $[ 2 ]$
$495$ $[ 2, 0, 0 ]$ $3$ $2$ $[]$ $[]$
$495$ $[ 2, 0, 0 ]$ $3$ $2$ $[]$ $[]$
$495$ $[ 0, 1, 5 ]$ $4$ $4$ $[ -55, -11 ]$ $[ 2, 5 ]$
$496$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$496$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$496$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$496$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$496$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$496$ $[ 0, 0, 0 ]$ $1$ $1$ $[]$ $[]$
$496$ $[ 1, 0, 15 ]$ $2$ $2$ $[ 124 ]$ $[ 3 ]$
$496$ $[ 1, 0, 15 ]$ $2$ $2$ $[ 124 ]$ $[ 3 ]$
$496$ $[ 0, 0, 10 ]$ $3$ $2$ $[]$ $[]$
$496$ $[ 0, 0, 10 ]$ $3$ $2$ $[]$ $[]$
$496$ $[ 0, 0, 10 ]$ $3$ $2$ $[]$ $[]$
$496$ $[ 0, 0, 10 ]$ $3$ $2$ $[]$ $[]$
$496$ $[ 0, 0, 10 ]$ $3$ $2$ $[]$ $[]$
$496$ $[ 0, 0, 10 ]$ $3$ $2$ $[]$ $[]$
$496$ $[ 0, 0, 10 ]$ $3$ $2$ $[]$ $[]$
$496$ $[ 0, 1, 0 ]$ $4$ $2$ $[ 8 ]$ $[ 2 ]$
$496$ $[ 1, 0, 5 ]$ $6$ $2$ $[ 124 ]$ $[ 3 ]$
$496$ $[ 1, 0, 5 ]$ $6$ $2$ $[ 124 ]$ $[ 3 ]$
$497$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$497$ $[ 2, 0 ]$ $3$ $2$ $[]$ $[]$
$498$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
$498$ $[ 0, 0 ]$ $1$ $1$ $[]$ $[]$
RM-surfaces
-----------
$N$ $\varepsilon$ $\operatorname{ord}(\varepsilon)$ $\dim(A_f)$ $\Disc(F)$ $[t_i]$ $[N_{F/\Q}(\delta_i)]$ $\gamma_\varepsilon\in\Br(K_P)[2]$
------- ------------------ ----------------------------------- ------------- ------------ --------------------- ------------------------ ------------------------------------
$17$ $[ 2 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$22$ $[ 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$23$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$25$ $[ 4 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$29$ $[ 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$31$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$31$ $[ 10 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$31$ $[ 6 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$32$ $[ 0, 1 ]$ $8$ $4$ $8$ $[ 8 ]$ $[ 2 ]$ $[]$
$33$ $[ 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$33$ $[ 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$33$ $[ 1, 1 ]$ $10$ $8$ $5$ $[ -11 ]$ $[ 5 ]$ $[ 3, 3 ]$
$34$ $[ 2 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$35$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$35$ $[ 0, 2 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$35$ $[ 1, 1 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$35$ $[ 1, 1 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$37$ $[ 6 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$38$ $[ 6 ]$ $3$ $4$ $28$ $[]$ $[]$ $[]$
$39$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$39$ $[ 0, 4 ]$ $3$ $4$ $17$ $[]$ $[]$ $[]$
$39$ $[ 1, 1 ]$ $12$ $8$ $12$ $[ -39, -3 ]$ $[ 4, 13 ]$ $[]$
$40$ $[ 0, 1, 0 ]$ $2$ $4$ $12$ $[]$ $[]$ $[]$
$43$ $[ 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$43$ $[ 14 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$44$ $[ 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$45$ $[ 2, 2 ]$ $6$ $8$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$50$ $[ 4 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$51$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$52$ $[ 1, 3 ]$ $4$ $8$ $8$ $[ -4 ]$ $[ 17 ]$ $[ 13, 13 ]$
$53$ $[ 26 ]$ $2$ $4$ $8$ $[ 53 ]$ $[ 7 ]$ $[]$
$55$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$55$ $[ 2, 0 ]$ $2$ $4$ $33$ $[ 5 ]$ $[ 4 ]$ $[]$
$56$ $[ 0, 1, 0 ]$ $2$ $4$ $17$ $[ 8 ]$ $[ 8 ]$ $[]$
$57$ $[ 1, 3 ]$ $6$ $8$ $12$ $[ -3 ]$ $[ 1 ]$ $[ 19, 19 ]$
$60$ $[ 1, 1, 0 ]$ $2$ $8$ $17$ $[ -4 ]$ $[ 8 ]$ $[]$
$61$ $[ 30 ]$ $2$ $4$ $12$ $[ 61 ]$ $[ 13 ]$ $[]$
$62$ $[ 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$65$ $[ 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$65$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$65$ $[ 0, 4 ]$ $3$ $4$ $13$ $[]$ $[]$ $[]$
$65$ $[ 0, 4 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$65$ $[ 2, 2 ]$ $6$ $8$ $21$ $[ 13 ]$ $[ 1 ]$ $[]$
$66$ $[ 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$66$ $[ 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$67$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$67$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$68$ $[ 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$68$ $[ 0, 4 ]$ $4$ $4$ $13$ $[ 17 ]$ $[ 4 ]$ $[]$
$68$ $[ 0, 2 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$69$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$70$ $[ 2, 0 ]$ $2$ $4$ $24$ $[]$ $[]$ $[]$
$70$ $[ 1, 3 ]$ $4$ $8$ $8$ $[ -7 ]$ $[ 2 ]$ $[]$
$72$ $[ 0, 0, 2 ]$ $3$ $4$ $33$ $[]$ $[]$ $[]$
$73$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$73$ $[ 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$73$ $[ 36 ]$ $2$ $4$ $5$ $[ 73 ]$ $[ 89 ]$ $[]$
$74$ $[ 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$74$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$74$ $[ 18 ]$ $2$ $4$ $21$ $[]$ $[]$ $[]$
$74$ $[ 6 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$74$ $[ 6 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$75$ $[ 0, 4 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$77$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$77$ $[ 2, 2 ]$ $15$ $8$ $5$ $[]$ $[]$ $[]$
$78$ $[ 0, 2 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$78$ $[ 0, 2 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$82$ $[ 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$82$ $[ 8 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$82$ $[ 8 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$84$ $[ 1, 0, 3 ]$ $2$ $4$ $17$ $[ 28 ]$ $[ 8 ]$ $[]$
$84$ $[ 1, 0, 3 ]$ $2$ $4$ $17$ $[ 28 ]$ $[ 8 ]$ $[]$
$85$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$85$ $[ 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$85$ $[ 2, 8 ]$ $2$ $8$ $28$ $[ 17, 85 ]$ $[ 2, 9 ]$ $[]$
$86$ $[ 0 ]$ $1$ $2$ $21$ $[]$ $[]$ $[]$
$86$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$86$ $[ 14 ]$ $3$ $4$ $17$ $[]$ $[]$ $[]$
$87$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$87$ $[ 0, 14 ]$ $2$ $4$ $5$ $[]$ $[]$ $[]$
$87$ $[ 1, 7 ]$ $4$ $4$ $12$ $[ 29 ]$ $[ 1 ]$ $[]$
$87$ $[ 1, 7 ]$ $4$ $4$ $12$ $[ 29 ]$ $[ 1 ]$ $[]$
$87$ $[ 1, 7 ]$ $4$ $8$ $13$ $[ -87 ]$ $[ 4 ]$ $[]$
$88$ $[ 0, 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$88$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$90$ $[ 2, 0 ]$ $3$ $4$ $33$ $[]$ $[]$ $[]$
$90$ $[ 2, 2 ]$ $6$ $8$ $24$ $[]$ $[]$ $[]$
$90$ $[ 1, 1 ]$ $12$ $8$ $24$ $[ -15 ]$ $[ 1 ]$ $[]$
$91$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$91$ $[ 0, 4 ]$ $3$ $4$ $12$ $[]$ $[]$ $[]$
$91$ $[ 0, 4 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$91$ $[ 2, 0 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$93$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$93$ $[ 1, 15 ]$ $2$ $8$ $13$ $[ 93, -31 ]$ $[ 3, 108 ]$ $[]$
$93$ $[ 0, 10 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$93$ $[ 1, 5 ]$ $6$ $4$ $33$ $[ 93 ]$ $[ 4 ]$ $[]$
$93$ $[ 1, 5 ]$ $6$ $4$ $33$ $[ 93 ]$ $[ 4 ]$ $[]$
$94$ $[ 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$97$ $[ 8 ]$ $12$ $4$ $12$ $[ 97 ]$ $[ 1 ]$ $[]$
$99$ $[ 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$99$ $[ 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$99$ $[ 2, 2 ]$ $15$ $8$ $5$ $[]$ $[]$ $[]$
$101$ $[ 10 ]$ $10$ $4$ $5$ $[ 101 ]$ $[ 5 ]$ $[]$
$102$ $[ 0, 4 ]$ $4$ $4$ $8$ $[]$ $[]$ $[]$
$103$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$104$ $[ 0, 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$104$ $[ 0, 0, 6 ]$ $2$ $4$ $17$ $[]$ $[]$ $[]$
$104$ $[ 0, 1, 0 ]$ $2$ $4$ $12$ $[]$ $[]$ $[]$
$104$ $[ 0, 1, 6 ]$ $2$ $8$ $41$ $[ 13, 8 ]$ $[ 10, 20 ]$ $[]$
$104$ $[ 0, 0, 4 ]$ $3$ $4$ $17$ $[]$ $[]$ $[]$
$104$ $[ 0, 1, 2 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$104$ $[ 0, 1, 2 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$105$ $[ 0, 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$105$ $[ 1, 0, 3 ]$ $2$ $4$ $33$ $[ 21 ]$ $[ 4 ]$ $[]$
$105$ $[ 1, 0, 3 ]$ $2$ $4$ $33$ $[ 21 ]$ $[ 4 ]$ $[]$
$105$ $[ 0, 0, 2 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$105$ $[ 0, 0, 2 ]$ $3$ $4$ $12$ $[]$ $[]$ $[]$
$107$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$110$ $[ 0, 0 ]$ $1$ $2$ $33$ $[]$ $[]$ $[]$
$110$ $[ 1, 5 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 5, 11 ]$
$110$ $[ 1, 5 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 5, 11 ]$
$110$ $[ 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$110$ $[ 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$110$ $[ 2, 2 ]$ $10$ $8$ $5$ $[]$ $[]$ $[]$
$111$ $[ 0, 18 ]$ $2$ $4$ $24$ $[ 37 ]$ $[ 3 ]$ $[]$
$111$ $[ 0, 6 ]$ $6$ $4$ $21$ $[ 37 ]$ $[ 1 ]$ $[]$
$112$ $[ 0, 1, 2 ]$ $12$ $4$ $12$ $[ 8 ]$ $[ 1 ]$ $[]$
$112$ $[ 0, 1, 2 ]$ $12$ $4$ $12$ $[ 8 ]$ $[ 1 ]$ $[]$
$113$ $[ 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$114$ $[ 1, 3 ]$ $6$ $4$ $24$ $[ 57 ]$ $[ 1 ]$ $[]$
$114$ $[ 1, 3 ]$ $6$ $4$ $24$ $[ 57 ]$ $[ 1 ]$ $[]$
$115$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$117$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$117$ $[ 0, 4 ]$ $3$ $4$ $17$ $[]$ $[]$ $[]$
$117$ $[ 3, 1 ]$ $12$ $8$ $12$ $[ -39 ]$ $[ 1 ]$ $[]$
$117$ $[ 3, 1 ]$ $12$ $8$ $12$ $[ -39 ]$ $[ 9 ]$ $[]$
$119$ $[ 2, 4 ]$ $12$ $8$ $44$ $[ 17 ]$ $[ 1 ]$ $[]$
$120$ $[ 0, 0, 1, 1 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 5 ]$
$120$ $[ 0, 0, 1, 1 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 5 ]$
$122$ $[ 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$122$ $[ 30 ]$ $2$ $4$ $5$ $[]$ $[]$ $[]$
$122$ $[ 20 ]$ $3$ $4$ $13$ $[]$ $[]$ $[]$
$123$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$123$ $[ 0, 8 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$124$ $[ 0, 6 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$124$ $[ 0, 6 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$124$ $[ 1, 3 ]$ $10$ $8$ $5$ $[ -31 ]$ $[ 5 ]$ $[ 2, 2 ]$
$125$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$125$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$125$ $[ 0 ]$ $1$ $4$ $5$ $[ 5 ]$ $[ 11 ]$ $[]$
$125$ $[ 50 ]$ $2$ $4$ $5$ $[ 5 ]$ $[ 11 ]$ $[]$
$125$ $[ 50 ]$ $2$ $4$ $5$ $[]$ $[]$ $[]$
$125$ $[ 20 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$125$ $[ 10 ]$ $10$ $8$ $5$ $[]$ $[]$ $[]$
$126$ $[ 2, 0 ]$ $3$ $4$ $33$ $[]$ $[]$ $[]$
$126$ $[ 2, 0 ]$ $3$ $4$ $24$ $[]$ $[]$ $[]$
$126$ $[ 3, 1 ]$ $6$ $8$ $8$ $[ -7 ]$ $[ 9 ]$ $[]$
$128$ $[ 0, 4 ]$ $8$ $4$ $8$ $[ 8 ]$ $[ 2 ]$ $[]$
$129$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$130$ $[ 2, 6 ]$ $2$ $4$ $33$ $[ 65 ]$ $[ 4 ]$ $[]$
$130$ $[ 2, 6 ]$ $2$ $4$ $33$ $[ 65 ]$ $[ 4 ]$ $[]$
$130$ $[ 0, 4 ]$ $3$ $4$ $12$ $[]$ $[]$ $[]$
$130$ $[ 0, 4 ]$ $3$ $4$ $40$ $[]$ $[]$ $[]$
$130$ $[ 1, 3 ]$ $4$ $4$ $44$ $[ 65 ]$ $[ 1 ]$ $[]$
$130$ $[ 1, 3 ]$ $4$ $4$ $12$ $[ 65 ]$ $[ 1 ]$ $[]$
$130$ $[ 3, 3 ]$ $4$ $4$ $12$ $[ 65 ]$ $[ 1 ]$ $[]$
$130$ $[ 3, 3 ]$ $4$ $4$ $44$ $[ 65 ]$ $[ 1 ]$ $[]$
$130$ $[ 0, 2 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$131$ $[ 26 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$131$ $[ 26 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$132$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$132$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$132$ $[ 1, 1, 2 ]$ $10$ $8$ $5$ $[]$ $[]$ $[ 2, 3 ]$
$132$ $[ 1, 1, 2 ]$ $10$ $8$ $5$ $[]$ $[]$ $[ 2, 3 ]$
$133$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$133$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$133$ $[ 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$133$ $[ 3, 9 ]$ $2$ $4$ $13$ $[]$ $[]$ $[ 7, 19 ]$
$133$ $[ 3, 9 ]$ $2$ $4$ $13$ $[]$ $[]$ $[ 7, 19 ]$
$133$ $[ 0, 6 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$133$ $[ 2, 0 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$133$ $[ 2, 0 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$133$ $[ 1, 3 ]$ $6$ $4$ $21$ $[ 133 ]$ $[ 1 ]$ $[]$
$133$ $[ 1, 9 ]$ $6$ $4$ $12$ $[]$ $[]$ $[ 7, 19 ]$
$133$ $[ 1, 9 ]$ $6$ $4$ $12$ $[]$ $[]$ $[ 7, 19 ]$
$133$ $[ 3, 3 ]$ $6$ $4$ $21$ $[ 133 ]$ $[ 1 ]$ $[]$
$133$ $[ 3, 3 ]$ $6$ $4$ $21$ $[ 133 ]$ $[ 1 ]$ $[]$
$133$ $[ 5, 3 ]$ $6$ $4$ $21$ $[ 133 ]$ $[ 1 ]$ $[]$
$134$ $[ 22 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$134$ $[ 22 ]$ $3$ $4$ $21$ $[]$ $[]$ $[]$
$135$ $[ 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$135$ $[ 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$135$ $[ 9, 1 ]$ $4$ $8$ $21$ $[ 5, -15 ]$ $[ 1, 100 ]$ $[]$
$135$ $[ 6, 2 ]$ $6$ $8$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$136$ $[ 0, 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$136$ $[ 0, 1, 4 ]$ $4$ $8$ $17$ $[ 8, 136 ]$ $[ 8, 64 ]$ $[]$
$136$ $[ 0, 0, 2 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$136$ $[ 0, 0, 2 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$138$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$138$ $[ 1, 11 ]$ $2$ $8$ $5$ $[ -23 ]$ $[ 116 ]$ $[ 3, 3 ]$
$140$ $[ 1, 0, 3 ]$ $2$ $4$ $12$ $[]$ $[]$ $[ 2, 7 ]$
$140$ $[ 1, 0, 3 ]$ $2$ $4$ $12$ $[]$ $[]$ $[ 2, 7 ]$
$140$ $[ 1, 0, 3 ]$ $2$ $8$ $17$ $[ -4 ]$ $[ 8 ]$ $[]$
$140$ $[ 0, 2, 2 ]$ $6$ $4$ $57$ $[ 5 ]$ $[ 4 ]$ $[]$
$140$ $[ 0, 2, 2 ]$ $6$ $4$ $57$ $[ 5 ]$ $[ 4 ]$ $[]$
$141$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$142$ $[ 14 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$143$ $[ 0, 2 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$143$ $[ 2, 0 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$144$ $[ 0, 0, 2 ]$ $3$ $4$ $33$ $[]$ $[]$ $[]$
$144$ $[ 0, 1, 2 ]$ $12$ $4$ $12$ $[ 8 ]$ $[ 1 ]$ $[]$
$144$ $[ 0, 1, 2 ]$ $12$ $4$ $12$ $[ 8 ]$ $[ 9 ]$ $[]$
$144$ $[ 0, 1, 2 ]$ $12$ $4$ $12$ $[ 8 ]$ $[ 9 ]$ $[]$
$144$ $[ 0, 1, 2 ]$ $12$ $4$ $12$ $[ 8 ]$ $[ 1 ]$ $[]$
$145$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$145$ $[ 0, 14 ]$ $2$ $4$ $12$ $[ 29 ]$ $[ 1 ]$ $[]$
$145$ $[ 2, 0 ]$ $2$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$145$ $[ 2, 0 ]$ $2$ $4$ $33$ $[ 5 ]$ $[ 4 ]$ $[]$
$145$ $[ 2, 14 ]$ $2$ $4$ $17$ $[ 145 ]$ $[ 36 ]$ $[]$
$145$ $[ 2, 14 ]$ $2$ $4$ $17$ $[ 145 ]$ $[ 36 ]$ $[]$
$146$ $[ 36 ]$ $2$ $4$ $12$ $[ 73 ]$ $[ 6 ]$ $[]$
$147$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$147$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$147$ $[ 1, 21 ]$ $2$ $8$ $8$ $[ -7, -3 ]$ $[ 2, 7 ]$ $[]$
$147$ $[ 0, 14 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$147$ $[ 0, 14 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$147$ $[ 1, 7 ]$ $6$ $16$ $8$ $[ -7, -3 ]$ $[ 2, 7 ]$ $[]$
$148$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$150$ $[ 0, 4 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$150$ $[ 0, 4 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$153$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$153$ $[ 2, 0 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$153$ $[ 2, 8 ]$ $6$ $8$ $12$ $[ 17 ]$ $[ 9 ]$ $[]$
$153$ $[ 0, 2 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$153$ $[ 0, 2 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$153$ $[ 0, 2 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$154$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$154$ $[ 3, 5 ]$ $2$ $8$ $21$ $[ -7 ]$ $[ 4 ]$ $[ 11, 11 ]$
$154$ $[ 2, 0 ]$ $3$ $4$ $28$ $[]$ $[]$ $[]$
$154$ $[ 2, 0 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$154$ $[ 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$154$ $[ 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$154$ $[ 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$154$ $[ 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$154$ $[ 2, 2 ]$ $15$ $8$ $5$ $[]$ $[]$ $[]$
$155$ $[ 2, 0 ]$ $2$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$156$ $[ 1, 1, 0 ]$ $2$ $8$ $105$ $[ 12, -3 ]$ $[ 4, 16 ]$ $[]$
$156$ $[ 0, 0, 2 ]$ $6$ $4$ $129$ $[ 13 ]$ $[ 16 ]$ $[]$
$156$ $[ 1, 0, 1 ]$ $12$ $4$ $12$ $[ 13 ]$ $[ 4 ]$ $[]$
$156$ $[ 1, 0, 1 ]$ $12$ $4$ $12$ $[ 13 ]$ $[ 4 ]$ $[]$
$157$ $[ 26 ]$ $6$ $4$ $21$ $[ 157 ]$ $[ 1 ]$ $[]$
$158$ $[ 0 ]$ $1$ $2$ $24$ $[]$ $[]$ $[]$
$158$ $[ 26 ]$ $3$ $4$ $13$ $[]$ $[]$ $[]$
$158$ $[ 26 ]$ $3$ $4$ $21$ $[]$ $[]$ $[]$
$160$ $[ 0, 4, 0 ]$ $2$ $4$ $12$ $[]$ $[]$ $[]$
$161$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$164$ $[ 0, 20 ]$ $2$ $4$ $28$ $[ 41 ]$ $[ 2 ]$ $[]$
$165$ $[ 0, 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$165$ $[ 0, 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$165$ $[ 1, 1, 0 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 5 ]$
$165$ $[ 1, 1, 0 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 5 ]$
$166$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$167$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$168$ $[ 0, 0, 1, 3 ]$ $2$ $8$ $17$ $[ -7 ]$ $[ 8 ]$ $[]$
$168$ $[ 0, 1, 0, 0 ]$ $2$ $4$ $12$ $[]$ $[]$ $[]$
$168$ $[ 1, 1, 1, 0 ]$ $2$ $4$ $5$ $[]$ $[]$ $[ 2, 3 ]$
$168$ $[ 1, 1, 1, 0 ]$ $2$ $4$ $5$ $[]$ $[]$ $[ 2, 3 ]$
$168$ $[ 0, 0, 0, 2 ]$ $3$ $4$ $57$ $[]$ $[]$ $[]$
$170$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$170$ $[ 0, 4 ]$ $4$ $4$ $8$ $[]$ $[]$ $[]$
$170$ $[ 2, 4 ]$ $4$ $4$ $8$ $[]$ $[]$ $[]$
$170$ $[ 2, 4 ]$ $4$ $4$ $8$ $[]$ $[]$ $[]$
$171$ $[ 0, 0 ]$ $1$ $4$ $33$ $[ -3 ]$ $[ 12 ]$ $[]$
$171$ $[ 0, 6 ]$ $3$ $8$ $24$ $[ -3 ]$ $[ 3 ]$ $[]$
$172$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$174$ $[ 0, 14 ]$ $2$ $4$ $33$ $[]$ $[]$ $[]$
$174$ $[ 1, 7 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 29 ]$
$174$ $[ 1, 7 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 29 ]$
$174$ $[ 1, 7 ]$ $4$ $8$ $24$ $[ -87 ]$ $[ 1 ]$ $[]$
$175$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$175$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$175$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$175$ $[ 10, 0 ]$ $2$ $4$ $17$ $[]$ $[]$ $[]$
$175$ $[ 10, 0 ]$ $2$ $4$ $5$ $[]$ $[]$ $[]$
$175$ $[ 0, 2 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$175$ $[ 10, 2 ]$ $6$ $8$ $8$ $[]$ $[]$ $[]$
$175$ $[ 4, 0 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$175$ $[ 5, 1 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$175$ $[ 5, 1 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$176$ $[ 0, 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$176$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$176$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$176$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$177$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$177$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$177$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$178$ $[ 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$178$ $[ 44 ]$ $2$ $4$ $17$ $[ 89 ]$ $[ 2 ]$ $[]$
$178$ $[ 22 ]$ $4$ $4$ $12$ $[ 89 ]$ $[ 1 ]$ $[]$
$180$ $[ 1, 3, 0 ]$ $2$ $8$ $8$ $[ -4 ]$ $[ 68 ]$ $[]$
$181$ $[ 36 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$181$ $[ 18 ]$ $10$ $4$ $5$ $[ 181 ]$ $[ 5 ]$ $[]$
$182$ $[ 0, 4 ]$ $3$ $4$ $12$ $[]$ $[]$ $[]$
$182$ $[ 0, 2 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$183$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$183$ $[ 0, 20 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$183$ $[ 0, 20 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$183$ $[ 1, 15 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 61 ]$
$183$ $[ 1, 15 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 61 ]$
$183$ $[ 0, 10 ]$ $6$ $4$ $21$ $[ 61 ]$ $[ 1 ]$ $[]$
$183$ $[ 0, 10 ]$ $6$ $4$ $21$ $[ 61 ]$ $[ 1 ]$ $[]$
$184$ $[ 0, 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$186$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$186$ $[ 1, 15 ]$ $2$ $8$ $56$ $[ -31 ]$ $[ 2 ]$ $[]$
$186$ $[ 0, 10 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$186$ $[ 0, 10 ]$ $3$ $4$ $76$ $[]$ $[]$ $[]$
$186$ $[ 0, 6 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$186$ $[ 0, 6 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$187$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$187$ $[ 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$187$ $[ 0, 4 ]$ $4$ $4$ $8$ $[]$ $[]$ $[]$
$187$ $[ 2, 0 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$187$ $[ 2, 0 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$187$ $[ 2, 0 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$188$ $[ 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$188$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$190$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$190$ $[ 2, 0 ]$ $2$ $4$ $8$ $[]$ $[]$ $[]$
$190$ $[ 0, 6 ]$ $3$ $4$ $17$ $[]$ $[]$ $[]$
$190$ $[ 1, 3 ]$ $12$ $8$ $24$ $[ -95 ]$ $[ 9 ]$ $[]$
$191$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$191$ $[ 38 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$193$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$193$ $[ 16 ]$ $12$ $4$ $12$ $[ 193 ]$ $[ 1 ]$ $[]$
$194$ $[ 48 ]$ $2$ $4$ $13$ $[ 97 ]$ $[ 3 ]$ $[]$
$194$ $[ 16 ]$ $6$ $4$ $24$ $[ 97 ]$ $[ 1 ]$ $[]$
$194$ $[ 16 ]$ $6$ $4$ $8$ $[ 97 ]$ $[ 9 ]$ $[]$
$194$ $[ 16 ]$ $6$ $4$ $60$ $[ 97 ]$ $[ 4 ]$ $[]$
$195$ $[ 0, 0, 6 ]$ $2$ $4$ $17$ $[]$ $[]$ $[]$
$195$ $[ 0, 0, 4 ]$ $3$ $4$ $12$ $[]$ $[]$ $[]$
$195$ $[ 0, 0, 2 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$196$ $[ 1, 21 ]$ $2$ $8$ $8$ $[ 28, -4 ]$ $[ 2, 34 ]$ $[]$
$196$ $[ 1, 7 ]$ $6$ $16$ $8$ $[ 28, -4 ]$ $[ 2, 34 ]$ $[]$
$198$ $[ 2, 0 ]$ $3$ $4$ $24$ $[]$ $[]$ $[]$
$198$ $[ 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$198$ $[ 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$198$ $[ 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$198$ $[ 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$198$ $[ 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$199$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$200$ $[ 0, 1, 0 ]$ $2$ $4$ $12$ $[]$ $[]$ $[]$
$200$ $[ 0, 1, 10 ]$ $2$ $4$ $12$ $[]$ $[]$ $[]$
$200$ $[ 0, 1, 10 ]$ $2$ $4$ $12$ $[]$ $[]$ $[]$
$200$ $[ 0, 0, 4 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$203$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$203$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$204$ $[ 0, 0, 4 ]$ $4$ $4$ $8$ $[]$ $[]$ $[]$
$205$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$205$ $[ 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$205$ $[ 1, 5 ]$ $8$ $4$ $8$ $[ 41 ]$ $[ 2 ]$ $[ 5, 5 ]$
$205$ $[ 1, 5 ]$ $8$ $4$ $8$ $[ 41 ]$ $[ 2 ]$ $[ 5, 5 ]$
$205$ $[ 3, 5 ]$ $8$ $4$ $8$ $[ 41 ]$ $[ 2 ]$ $[ 5, 5 ]$
$205$ $[ 3, 5 ]$ $8$ $4$ $8$ $[ 41 ]$ $[ 2 ]$ $[ 5, 5 ]$
$206$ $[ 0 ]$ $1$ $2$ $29$ $[]$ $[]$ $[]$
$206$ $[ 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$207$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$207$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$207$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$207$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$207$ $[ 3, 11 ]$ $2$ $8$ $12$ $[ -3, -23 ]$ $[ 1, 54 ]$ $[]$
$208$ $[ 0, 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$208$ $[ 0, 0, 6 ]$ $2$ $4$ $17$ $[]$ $[]$ $[]$
$208$ $[ 0, 0, 4 ]$ $3$ $4$ $17$ $[]$ $[]$ $[]$
$208$ $[ 0, 1, 6 ]$ $4$ $8$ $44$ $[ 104 ]$ $[ 1 ]$ $[]$
$208$ $[ 1, 0, 1 ]$ $12$ $4$ $12$ $[ 13 ]$ $[ 4 ]$ $[]$
$208$ $[ 1, 0, 1 ]$ $12$ $4$ $12$ $[ 13 ]$ $[ 4 ]$ $[]$
$208$ $[ 1, 0, 1 ]$ $12$ $4$ $12$ $[ 13 ]$ $[ 4 ]$ $[]$
$208$ $[ 1, 0, 1 ]$ $12$ $4$ $12$ $[ 13 ]$ $[ 4 ]$ $[]$
$208$ $[ 1, 0, 1 ]$ $12$ $8$ $12$ $[ -52, -4 ]$ $[ 36, 37 ]$ $[]$
$209$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$209$ $[ 5, 9 ]$ $2$ $8$ $5$ $[ 209, -19 ]$ $[ 5, 11 ]$ $[]$
$209$ $[ 2, 0 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$209$ $[ 1, 9 ]$ $10$ $8$ $5$ $[ 209, -19 ]$ $[ 5, 11 ]$ $[]$
$210$ $[ 1, 0, 3 ]$ $2$ $4$ $5$ $[]$ $[]$ $[ 3, 7 ]$
$210$ $[ 1, 0, 3 ]$ $2$ $4$ $5$ $[]$ $[]$ $[ 3, 7 ]$
$210$ $[ 1, 2, 3 ]$ $2$ $8$ $28$ $[ -7, 105 ]$ $[ 8, 16 ]$ $[]$
$210$ $[ 1, 2, 3 ]$ $2$ $8$ $28$ $[ -7, 105 ]$ $[ 8, 16 ]$ $[]$
$210$ $[ 1, 2, 1 ]$ $6$ $4$ $33$ $[ 105 ]$ $[ 16 ]$ $[]$
$210$ $[ 1, 2, 1 ]$ $6$ $4$ $33$ $[ 105 ]$ $[ 16 ]$ $[]$
$210$ $[ 1, 2, 1 ]$ $6$ $4$ $33$ $[ 105 ]$ $[ 16 ]$ $[]$
$210$ $[ 1, 2, 1 ]$ $6$ $4$ $33$ $[ 105 ]$ $[ 16 ]$ $[]$
$210$ $[ 1, 2, 1 ]$ $6$ $8$ $120$ $[ -15, -7 ]$ $[ 1, 4 ]$ $[]$
$210$ $[ 1, 2, 1 ]$ $6$ $8$ $120$ $[ -15, -7 ]$ $[ 1, 4 ]$ $[]$
$211$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$211$ $[ 42 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$212$ $[ 0, 26 ]$ $2$ $4$ $12$ $[ 53 ]$ $[ 13 ]$ $[]$
$213$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$213$ $[ 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$213$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$213$ $[ 0, 14 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$213$ $[ 1, 7 ]$ $10$ $8$ $5$ $[ -71 ]$ $[ 5 ]$ $[ 3, 3 ]$
$214$ $[ 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$214$ $[ 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$215$ $[ 0, 14 ]$ $3$ $4$ $13$ $[]$ $[]$ $[]$
$215$ $[ 1, 7 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$215$ $[ 1, 7 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$216$ $[ 1, 1, 9 ]$ $2$ $8$ $33$ $[ -3, 24 ]$ $[ 3, 3 ]$ $[]$
$216$ $[ 0, 0, 6 ]$ $3$ $4$ $33$ $[]$ $[]$ $[]$
$217$ $[ 0, 6 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$218$ $[ 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$218$ $[ 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$218$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$219$ $[ 0, 36 ]$ $2$ $4$ $24$ $[ 73 ]$ $[ 12 ]$ $[]$
$219$ $[ 0, 12 ]$ $6$ $4$ $201$ $[ 73 ]$ $[ 4 ]$ $[]$
$219$ $[ 0, 6 ]$ $12$ $4$ $12$ $[ 73 ]$ $[ 1 ]$ $[]$
$220$ $[ 0, 2, 0 ]$ $2$ $4$ $57$ $[ 5 ]$ $[ 16 ]$ $[]$
$220$ $[ 1, 0, 5 ]$ $2$ $8$ $105$ $[ -4 ]$ $[ 4 ]$ $[]$
$221$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$221$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$221$ $[ 0, 0 ]$ $1$ $2$ $21$ $[]$ $[]$ $[]$
$221$ $[ 0, 8 ]$ $2$ $4$ $17$ $[ 17 ]$ $[ 8 ]$ $[]$
$221$ $[ 6, 0 ]$ $2$ $4$ $8$ $[]$ $[]$ $[]$
$221$ $[ 6, 0 ]$ $2$ $4$ $13$ $[ 13 ]$ $[ 3 ]$ $[]$
$221$ $[ 0, 4 ]$ $4$ $4$ $12$ $[ 17 ]$ $[ 4 ]$ $[]$
$221$ $[ 0, 4 ]$ $4$ $4$ $8$ $[]$ $[]$ $[]$
$221$ $[ 0, 4 ]$ $4$ $4$ $8$ $[]$ $[]$ $[]$
$221$ $[ 6, 2 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$221$ $[ 6, 2 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$222$ $[ 0, 18 ]$ $2$ $4$ $65$ $[]$ $[]$ $[]$
$222$ $[ 0, 12 ]$ $3$ $4$ $33$ $[]$ $[]$ $[]$
$222$ $[ 0, 6 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$223$ $[ 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$224$ $[ 0, 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$224$ $[ 0, 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$224$ $[ 0, 4, 0 ]$ $2$ $4$ $17$ $[ 8 ]$ $[ 8 ]$ $[]$
$224$ $[ 1, 0, 3 ]$ $2$ $8$ $8$ $[ -7 ]$ $[ 8 ]$ $[ 2, 2 ]$
$224$ $[ 0, 0, 2 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$224$ $[ 0, 0, 2 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$224$ $[ 0, 1, 0 ]$ $8$ $4$ $8$ $[ 8 ]$ $[ 2 ]$ $[]$
$225$ $[ 2, 0 ]$ $3$ $8$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$225$ $[ 0, 4 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$225$ $[ 0, 4 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$226$ $[ 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$226$ $[ 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$226$ $[ 56 ]$ $2$ $4$ $13$ $[ 113 ]$ $[ 4 ]$ $[]$
$227$ $[ 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$227$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$227$ $[ 0 ]$ $1$ $2$ $29$ $[]$ $[]$ $[]$
$228$ $[ 0, 0, 0 ]$ $1$ $2$ $33$ $[]$ $[]$ $[]$
$228$ $[ 0, 0, 6 ]$ $3$ $4$ $28$ $[]$ $[]$ $[]$
$228$ $[ 0, 0, 6 ]$ $3$ $4$ $24$ $[]$ $[]$ $[]$
$228$ $[ 0, 1, 3 ]$ $6$ $4$ $33$ $[ 57 ]$ $[ 9 ]$ $[]$
$228$ $[ 0, 1, 3 ]$ $6$ $4$ $33$ $[ 57 ]$ $[ 9 ]$ $[]$
$230$ $[ 0, 0 ]$ $1$ $2$ $21$ $[]$ $[]$ $[]$
$230$ $[ 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$230$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$230$ $[ 2, 0 ]$ $2$ $4$ $5$ $[]$ $[]$ $[]$
$230$ $[ 1, 11 ]$ $4$ $8$ $8$ $[ -23 ]$ $[ 2 ]$ $[]$
$231$ $[ 0, 0, 0 ]$ $1$ $2$ $21$ $[]$ $[]$ $[]$
$231$ $[ 0, 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$231$ $[ 0, 2, 0 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$231$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$231$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$231$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$231$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$231$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$232$ $[ 0, 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$234$ $[ 0, 4 ]$ $3$ $4$ $40$ $[]$ $[]$ $[]$
$234$ $[ 0, 4 ]$ $3$ $4$ $40$ $[]$ $[]$ $[]$
$234$ $[ 2, 0 ]$ $3$ $4$ $12$ $[]$ $[]$ $[]$
$234$ $[ 2, 0 ]$ $3$ $4$ $24$ $[]$ $[]$ $[]$
$234$ $[ 0, 2 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$234$ $[ 0, 2 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$234$ $[ 3, 1 ]$ $12$ $8$ $12$ $[ -39 ]$ $[ 1 ]$ $[]$
$237$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$238$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$238$ $[ 2, 8 ]$ $6$ $8$ $60$ $[ 17 ]$ $[ 4 ]$ $[]$
$238$ $[ 2, 4 ]$ $12$ $4$ $12$ $[ 17 ]$ $[ 1 ]$ $[]$
$238$ $[ 2, 4 ]$ $12$ $4$ $12$ $[ 17 ]$ $[ 1 ]$ $[]$
$240$ $[ 0, 0, 1, 1 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 5 ]$
$240$ $[ 0, 0, 1, 1 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 5 ]$
$240$ $[ 0, 1, 0, 0 ]$ $4$ $4$ $8$ $[]$ $[]$ $[]$
$240$ $[ 1, 0, 0, 1 ]$ $4$ $8$ $24$ $[ -20 ]$ $[ 4 ]$ $[]$
$242$ $[ 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$242$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$242$ $[ 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$242$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$242$ $[ 22 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$242$ $[ 22 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$242$ $[ 22 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$242$ $[ 22 ]$ $5$ $8$ $12$ $[]$ $[]$ $[]$
$242$ $[ 22 ]$ $5$ $8$ $12$ $[]$ $[]$ $[]$
$244$ $[ 0, 30 ]$ $2$ $4$ $8$ $[ 61 ]$ $[ 127 ]$ $[]$
$245$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$245$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$245$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$245$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$245$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$245$ $[ 0, 14 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$245$ $[ 0, 14 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$245$ $[ 0, 14 ]$ $3$ $4$ $17$ $[]$ $[]$ $[]$
$245$ $[ 0, 14 ]$ $3$ $4$ $17$ $[]$ $[]$ $[]$
$245$ $[ 0, 14 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$245$ $[ 1, 21 ]$ $4$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$245$ $[ 1, 21 ]$ $4$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$245$ $[ 1, 7 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$245$ $[ 1, 7 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$246$ $[ 0, 20 ]$ $2$ $4$ $57$ $[]$ $[]$ $[]$
$246$ $[ 0, 8 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$246$ $[ 0, 8 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$246$ $[ 0, 8 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$246$ $[ 0, 8 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$246$ $[ 1, 5 ]$ $8$ $4$ $8$ $[ 41 ]$ $[ 2 ]$ $[]$
$246$ $[ 1, 5 ]$ $8$ $4$ $8$ $[ 41 ]$ $[ 2 ]$ $[]$
$246$ $[ 1, 5 ]$ $8$ $4$ $8$ $[ 41 ]$ $[ 2 ]$ $[]$
$246$ $[ 1, 5 ]$ $8$ $4$ $8$ $[ 41 ]$ $[ 2 ]$ $[]$
$247$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$247$ $[ 3, 3 ]$ $12$ $4$ $12$ $[ 13 ]$ $[ 1 ]$ $[]$
$247$ $[ 3, 3 ]$ $12$ $4$ $12$ $[ 13 ]$ $[ 1 ]$ $[]$
$248$ $[ 0, 0, 0 ]$ $1$ $2$ $33$ $[]$ $[]$ $[]$
$248$ $[ 1, 1, 5 ]$ $6$ $4$ $24$ $[ 248 ]$ $[ 1 ]$ $[]$
$248$ $[ 1, 1, 5 ]$ $6$ $4$ $24$ $[ 248 ]$ $[ 1 ]$ $[]$
$249$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$250$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$250$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$250$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$250$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$250$ $[ 50 ]$ $2$ $4$ $5$ $[]$ $[]$ $[]$
$250$ $[ 50 ]$ $2$ $4$ $5$ $[]$ $[]$ $[]$
$250$ $[ 20 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$250$ $[ 10 ]$ $10$ $8$ $5$ $[]$ $[]$ $[]$
$252$ $[ 1, 0, 3 ]$ $2$ $4$ $17$ $[ 28 ]$ $[ 8 ]$ $[]$
$252$ $[ 1, 0, 3 ]$ $2$ $4$ $17$ $[ 28 ]$ $[ 8 ]$ $[]$
$253$ $[ 5, 11 ]$ $2$ $8$ $21$ $[ -23, -11 ]$ $[ 4, 15 ]$ $[]$
$253$ $[ 2, 0 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$254$ $[ 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$254$ $[ 42 ]$ $3$ $4$ $76$ $[]$ $[]$ $[]$
$255$ $[ 0, 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$255$ $[ 0, 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$255$ $[ 0, 0, 8 ]$ $2$ $4$ $13$ $[]$ $[]$ $[]$
$255$ $[ 0, 2, 0 ]$ $2$ $4$ $5$ $[]$ $[]$ $[]$
$255$ $[ 1, 1, 4 ]$ $4$ $8$ $13$ $[ -255 ]$ $[ 4 ]$ $[]$
$255$ $[ 1, 3, 4 ]$ $4$ $8$ $13$ $[ -255 ]$ $[ 4 ]$ $[]$
$256$ $[ 0, 16 ]$ $4$ $8$ $12$ $[ -8 ]$ $[ 4 ]$ $[]$
$256$ $[ 0, 16 ]$ $4$ $8$ $12$ $[ -8 ]$ $[ 4 ]$ $[]$
$256$ $[ 0, 8 ]$ $8$ $4$ $8$ $[ 8 ]$ $[ 2 ]$ $[]$
$256$ $[ 0, 8 ]$ $8$ $4$ $8$ $[ 8 ]$ $[ 2 ]$ $[]$
$257$ $[ 64 ]$ $4$ $4$ $5$ $[ 257 ]$ $[ 4 ]$ $[]$
$258$ $[ 0, 14 ]$ $3$ $4$ $13$ $[]$ $[]$ $[]$
$258$ $[ 1, 7 ]$ $6$ $4$ $24$ $[ 129 ]$ $[ 1 ]$ $[]$
$258$ $[ 1, 7 ]$ $6$ $4$ $24$ $[ 129 ]$ $[ 1 ]$ $[]$
$259$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$259$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$259$ $[ 2, 12 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$259$ $[ 4, 12 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$260$ $[ 0, 2, 6 ]$ $2$ $8$ $21$ $[ 5, 13 ]$ $[ 4, 4 ]$ $[]$
$260$ $[ 0, 1, 3 ]$ $4$ $4$ $5$ $[ 65 ]$ $[ 4 ]$ $[]$
$260$ $[ 0, 3, 3 ]$ $4$ $4$ $5$ $[ 65 ]$ $[ 4 ]$ $[]$
$260$ $[ 0, 1, 1 ]$ $12$ $4$ $12$ $[ 65 ]$ $[ 1 ]$ $[]$
$260$ $[ 0, 1, 1 ]$ $12$ $4$ $12$ $[ 65 ]$ $[ 1 ]$ $[]$
$260$ $[ 0, 3, 1 ]$ $12$ $4$ $12$ $[ 65 ]$ $[ 1 ]$ $[]$
$260$ $[ 0, 3, 1 ]$ $12$ $4$ $12$ $[ 65 ]$ $[ 1 ]$ $[]$
$261$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$261$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$261$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$261$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$261$ $[ 0, 14 ]$ $2$ $4$ $5$ $[]$ $[]$ $[]$
$261$ $[ 3, 7 ]$ $4$ $8$ $5$ $[ -87 ]$ $[ 4 ]$ $[]$
$261$ $[ 3, 7 ]$ $4$ $8$ $13$ $[ -87 ]$ $[ 4 ]$ $[]$
$262$ $[ 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$262$ $[ 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$262$ $[ 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$262$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$262$ $[ 26 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$264$ $[ 0, 1, 0, 0 ]$ $2$ $4$ $8$ $[]$ $[]$ $[]$
$264$ $[ 0, 1, 0, 0 ]$ $2$ $4$ $28$ $[]$ $[]$ $[]$
$264$ $[ 1, 1, 1, 0 ]$ $2$ $8$ $8$ $[ -8 ]$ $[ 68 ]$ $[ 3, 3 ]$
$264$ $[ 0, 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$264$ $[ 0, 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$264$ $[ 0, 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$264$ $[ 0, 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$264$ $[ 1, 1, 1, 2 ]$ $10$ $16$ $5$ $[ -8 ]$ $[ 41 ]$ $[ 3, 3 ]$
$265$ $[ 0, 0 ]$ $1$ $2$ $21$ $[]$ $[]$ $[]$
$265$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$265$ $[ 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$265$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$265$ $[ 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$265$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$265$ $[ 2, 26 ]$ $2$ $4$ $5$ $[ 265 ]$ $[ 11 ]$ $[]$
$265$ $[ 2, 26 ]$ $2$ $4$ $5$ $[ 265 ]$ $[ 11 ]$ $[]$
$266$ $[ 0, 0 ]$ $1$ $2$ $29$ $[]$ $[]$ $[]$
$266$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$266$ $[ 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$266$ $[ 0, 6 ]$ $3$ $4$ $17$ $[]$ $[]$ $[]$
$266$ $[ 2, 0 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$266$ $[ 1, 3 ]$ $6$ $4$ $12$ $[]$ $[]$ $[ 7, 19 ]$
$266$ $[ 5, 3 ]$ $6$ $4$ $12$ $[]$ $[]$ $[ 7, 19 ]$
$267$ $[ 1, 11 ]$ $8$ $4$ $8$ $[ 89 ]$ $[ 2 ]$ $[]$
$267$ $[ 1, 11 ]$ $8$ $4$ $8$ $[ 89 ]$ $[ 2 ]$ $[]$
$268$ $[ 0, 0 ]$ $1$ $2$ $21$ $[]$ $[]$ $[]$
$268$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$270$ $[ 6, 0 ]$ $3$ $4$ $33$ $[]$ $[]$ $[]$
$270$ $[ 9, 1 ]$ $4$ $8$ $12$ $[ -15 ]$ $[ 1 ]$ $[]$
$270$ $[ 9, 1 ]$ $4$ $8$ $24$ $[ -15 ]$ $[ 1 ]$ $[]$
$270$ $[ 6, 2 ]$ $6$ $8$ $24$ $[]$ $[]$ $[]$
$270$ $[ 3, 1 ]$ $12$ $8$ $24$ $[ -15 ]$ $[ 1 ]$ $[]$
$272$ $[ 0, 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$272$ $[ 0, 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$272$ $[ 0, 0, 4 ]$ $4$ $4$ $13$ $[ 17 ]$ $[ 4 ]$ $[]$
$272$ $[ 0, 0, 2 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$272$ $[ 0, 0, 2 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$272$ $[ 0, 0, 2 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$272$ $[ 0, 0, 2 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$272$ $[ 0, 0, 2 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$272$ $[ 0, 1, 2 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$272$ $[ 0, 3, 2 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$273$ $[ 0, 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$273$ $[ 0, 3, 3 ]$ $4$ $4$ $40$ $[ 13 ]$ $[ 9 ]$ $[]$
$273$ $[ 0, 3, 3 ]$ $4$ $4$ $40$ $[ 13 ]$ $[ 9 ]$ $[]$
$273$ $[ 0, 3, 3 ]$ $4$ $4$ $24$ $[ 13 ]$ $[ 9 ]$ $[]$
$273$ $[ 0, 3, 3 ]$ $4$ $4$ $24$ $[ 13 ]$ $[ 9 ]$ $[]$
$273$ $[ 1, 0, 3 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 13 ]$
$273$ $[ 1, 0, 3 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 13 ]$
$273$ $[ 0, 2, 2 ]$ $6$ $4$ $21$ $[ 13 ]$ $[ 1 ]$ $[]$
$273$ $[ 0, 4, 2 ]$ $6$ $4$ $21$ $[ 13 ]$ $[ 1 ]$ $[]$
$273$ $[ 0, 3, 1 ]$ $12$ $4$ $12$ $[ 13 ]$ $[ 1 ]$ $[]$
$273$ $[ 0, 3, 1 ]$ $12$ $4$ $12$ $[ 13 ]$ $[ 1 ]$ $[]$
$273$ $[ 1, 2, 3 ]$ $12$ $8$ $8$ $[]$ $[]$ $[ 3, 13 ]$
$273$ $[ 1, 2, 3 ]$ $12$ $8$ $8$ $[]$ $[]$ $[ 3, 13 ]$
$274$ $[ 68 ]$ $2$ $4$ $60$ $[ 137 ]$ $[ 4 ]$ $[]$
$275$ $[ 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$275$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$275$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$275$ $[ 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$275$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$275$ $[ 0, 0 ]$ $1$ $4$ $33$ $[ 5 ]$ $[ 4 ]$ $[]$
$275$ $[ 10, 0 ]$ $2$ $4$ $8$ $[]$ $[]$ $[]$
$275$ $[ 10, 0 ]$ $2$ $4$ $5$ $[]$ $[]$ $[]$
$275$ $[ 10, 0 ]$ $2$ $4$ $13$ $[]$ $[]$ $[]$
$275$ $[ 5, 5 ]$ $4$ $16$ $5$ $[ -11, -55 ]$ $[ 5, 9 ]$ $[]$
$275$ $[ 4, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$275$ $[ 4, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$275$ $[ 4, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$275$ $[ 16, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$275$ $[ 16, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$275$ $[ 16, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$276$ $[ 0, 0, 0 ]$ $1$ $2$ $40$ $[]$ $[]$ $[]$
$276$ $[ 0, 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$276$ $[ 0, 1, 11 ]$ $2$ $8$ $17$ $[ -23, 69 ]$ $[ 8, 52 ]$ $[]$
$278$ $[ 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$279$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$279$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$279$ $[ 0, 10 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$279$ $[ 0, 10 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$279$ $[ 0, 10 ]$ $3$ $8$ $5$ $[ -3 ]$ $[ 4 ]$ $[]$
$279$ $[ 0, 6 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$280$ $[ 0, 0, 0, 0 ]$ $1$ $2$ $33$ $[]$ $[]$ $[]$
$280$ $[ 0, 0, 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$280$ $[ 0, 0, 0, 2 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$280$ $[ 1, 1, 0, 1 ]$ $6$ $4$ $24$ $[ 56 ]$ $[ 1 ]$ $[]$
$280$ $[ 1, 1, 0, 1 ]$ $6$ $4$ $24$ $[ 56 ]$ $[ 1 ]$ $[]$
$280$ $[ 1, 1, 0, 1 ]$ $6$ $4$ $12$ $[]$ $[]$ $[ 2, 7 ]$
$280$ $[ 1, 1, 0, 1 ]$ $6$ $4$ $12$ $[]$ $[]$ $[ 2, 7 ]$
$280$ $[ 0, 1, 1, 1 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$280$ $[ 0, 1, 1, 1 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$280$ $[ 0, 1, 1, 1 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$280$ $[ 0, 1, 1, 1 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$282$ $[ 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$282$ $[ 0, 0 ]$ $1$ $2$ $24$ $[]$ $[]$ $[]$
$285$ $[ 0, 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$285$ $[ 0, 0, 0 ]$ $1$ $2$ $28$ $[]$ $[]$ $[]$
$285$ $[ 0, 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$285$ $[ 0, 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$285$ $[ 0, 0, 6 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$285$ $[ 0, 0, 6 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$285$ $[ 1, 1, 0 ]$ $4$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$285$ $[ 1, 1, 0 ]$ $4$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$286$ $[ 0, 6 ]$ $2$ $4$ $17$ $[]$ $[]$ $[]$
$286$ $[ 0, 4 ]$ $3$ $4$ $12$ $[]$ $[]$ $[]$
$286$ $[ 2, 0 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$286$ $[ 2, 6 ]$ $10$ $8$ $5$ $[]$ $[]$ $[]$
$287$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$287$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$287$ $[ 2, 0 ]$ $3$ $4$ $13$ $[]$ $[]$ $[]$
$287$ $[ 2, 0 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$287$ $[ 0, 8 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$287$ $[ 0, 8 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$287$ $[ 2, 10 ]$ $12$ $4$ $12$ $[ 41 ]$ $[ 1 ]$ $[]$
$287$ $[ 2, 10 ]$ $12$ $4$ $12$ $[ 41 ]$ $[ 1 ]$ $[]$
$288$ $[ 0, 0, 2 ]$ $3$ $4$ $24$ $[]$ $[]$ $[]$
$288$ $[ 0, 0, 2 ]$ $3$ $4$ $24$ $[]$ $[]$ $[]$
$288$ $[ 0, 1, 0 ]$ $8$ $4$ $8$ $[ 8 ]$ $[ 2 ]$ $[]$
$289$ $[ 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$289$ $[ 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$289$ $[ 0 ]$ $1$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$289$ $[ 136 ]$ $2$ $4$ $13$ $[]$ $[]$ $[]$
$289$ $[ 136 ]$ $2$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$289$ $[ 68 ]$ $4$ $8$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$289$ $[ 68 ]$ $4$ $8$ $13$ $[]$ $[]$ $[]$
$289$ $[ 34 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$289$ $[ 34 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$289$ $[ 34 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$289$ $[ 34 ]$ $8$ $16$ $13$ $[]$ $[]$ $[]$
$290$ $[ 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$290$ $[ 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$290$ $[ 0, 14 ]$ $2$ $4$ $5$ $[]$ $[]$ $[]$
$290$ $[ 0, 14 ]$ $2$ $4$ $29$ $[]$ $[]$ $[]$
$290$ $[ 2, 0 ]$ $2$ $4$ $5$ $[]$ $[]$ $[]$
$290$ $[ 1, 7 ]$ $4$ $4$ $76$ $[ 145 ]$ $[ 9 ]$ $[]$
$290$ $[ 3, 7 ]$ $4$ $4$ $76$ $[ 145 ]$ $[ 9 ]$ $[]$
$291$ $[ 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$291$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$291$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$291$ $[ 0, 16 ]$ $6$ $4$ $105$ $[ 97 ]$ $[ 4 ]$ $[]$
$291$ $[ 0, 16 ]$ $6$ $4$ $33$ $[ 97 ]$ $[ 16 ]$ $[]$
$291$ $[ 0, 12 ]$ $8$ $4$ $8$ $[ 97 ]$ $[ 2 ]$ $[]$
$291$ $[ 0, 8 ]$ $12$ $4$ $12$ $[ 97 ]$ $[ 1 ]$ $[]$
$292$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$294$ $[ 1, 21 ]$ $2$ $8$ $8$ $[ -7 ]$ $[ 2 ]$ $[]$
$294$ $[ 1, 7 ]$ $6$ $16$ $8$ $[ -7 ]$ $[ 2 ]$ $[]$
$296$ $[ 0, 1, 0 ]$ $2$ $4$ $5$ $[]$ $[]$ $[]$
$296$ $[ 0, 1, 0 ]$ $2$ $4$ $29$ $[]$ $[]$ $[]$
$296$ $[ 1, 1, 3 ]$ $12$ $4$ $12$ $[ 37 ]$ $[ 1 ]$ $[]$
$296$ $[ 1, 1, 3 ]$ $12$ $4$ $12$ $[ 37 ]$ $[ 1 ]$ $[]$
$297$ $[ 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$297$ $[ 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$297$ $[ 9, 5 ]$ $2$ $8$ $24$ $[ -11, 33 ]$ $[ 3, 4 ]$ $[]$
$297$ $[ 6, 2 ]$ $15$ $8$ $5$ $[]$ $[]$ $[]$
$298$ $[ 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$299$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$299$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$299$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$299$ $[ 0, 0 ]$ $1$ $2$ $21$ $[]$ $[]$ $[]$
$299$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$299$ $[ 3, 11 ]$ $4$ $8$ $28$ $[ -23, -299 ]$ $[ 8, 9 ]$ $[]$
$300$ $[ 1, 1, 0 ]$ $2$ $8$ $17$ $[ -4 ]$ $[ 8 ]$ $[]$
$300$ $[ 1, 1, 10 ]$ $2$ $8$ $17$ $[ -4 ]$ $[ 8 ]$ $[]$
$300$ $[ 1, 1, 10 ]$ $2$ $8$ $17$ $[ -4 ]$ $[ 8 ]$ $[]$
$300$ $[ 1, 0, 5 ]$ $4$ $8$ $12$ $[]$ $[]$ $[ 2, 5 ]$
$300$ $[ 1, 0, 5 ]$ $4$ $8$ $12$ $[]$ $[]$ $[ 2, 5 ]$
$301$ $[ 2, 0 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$301$ $[ 1, 7 ]$ $6$ $4$ $21$ $[ 301 ]$ $[ 1 ]$ $[]$
$301$ $[ 3, 7 ]$ $6$ $4$ $21$ $[ 301 ]$ $[ 1 ]$ $[]$
$301$ $[ 3, 7 ]$ $6$ $4$ $21$ $[ 301 ]$ $[ 1 ]$ $[]$
$301$ $[ 5, 7 ]$ $6$ $4$ $21$ $[ 301 ]$ $[ 1 ]$ $[]$
$302$ $[ 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$302$ $[ 50 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$303$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$303$ $[ 1, 25 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 101 ]$
$303$ $[ 1, 25 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 101 ]$
$304$ $[ 0, 0, 6 ]$ $3$ $4$ $28$ $[]$ $[]$ $[]$
$304$ $[ 0, 1, 0 ]$ $4$ $4$ $8$ $[]$ $[]$ $[]$
$304$ $[ 1, 1, 3 ]$ $12$ $4$ $12$ $[ 8 ]$ $[ 1 ]$ $[]$
$304$ $[ 1, 1, 3 ]$ $12$ $4$ $12$ $[ 8 ]$ $[ 1 ]$ $[]$
$305$ $[ 2, 0 ]$ $2$ $4$ $33$ $[ 5 ]$ $[ 4 ]$ $[]$
$305$ $[ 0, 20 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$305$ $[ 0, 10 ]$ $6$ $4$ $24$ $[ 61 ]$ $[ 4 ]$ $[]$
$305$ $[ 0, 10 ]$ $6$ $4$ $21$ $[ 61 ]$ $[ 1 ]$ $[]$
$305$ $[ 0, 10 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$305$ $[ 2, 20 ]$ $6$ $8$ $13$ $[]$ $[]$ $[]$
$305$ $[ 0, 12 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$306$ $[ 0, 0 ]$ $1$ $2$ $24$ $[]$ $[]$ $[]$
$306$ $[ 0, 0 ]$ $1$ $2$ $24$ $[]$ $[]$ $[]$
$306$ $[ 2, 0 ]$ $3$ $4$ $24$ $[]$ $[]$ $[]$
$306$ $[ 0, 4 ]$ $4$ $4$ $8$ $[]$ $[]$ $[]$
$306$ $[ 0, 2 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$306$ $[ 0, 2 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$306$ $[ 0, 2 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$306$ $[ 2, 4 ]$ $12$ $4$ $12$ $[ 17 ]$ $[ 16 ]$ $[]$
$306$ $[ 2, 4 ]$ $12$ $4$ $12$ $[ 17 ]$ $[ 1 ]$ $[]$
$306$ $[ 2, 4 ]$ $12$ $4$ $12$ $[ 17 ]$ $[ 1 ]$ $[]$
$306$ $[ 2, 4 ]$ $12$ $4$ $12$ $[ 17 ]$ $[ 16 ]$ $[]$
$307$ $[ 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$308$ $[ 0, 0, 0 ]$ $1$ $2$ $24$ $[]$ $[]$ $[]$
$308$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$310$ $[ 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$310$ $[ 0, 0 ]$ $1$ $2$ $24$ $[]$ $[]$ $[]$
$310$ $[ 0, 6 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$310$ $[ 0, 6 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$312$ $[ 0, 0, 0, 4 ]$ $3$ $4$ $13$ $[]$ $[]$ $[]$
$312$ $[ 0, 0, 1, 3 ]$ $4$ $8$ $12$ $[ -39 ]$ $[ 4 ]$ $[]$
$312$ $[ 0, 0, 0, 2 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$312$ $[ 0, 1, 0, 2 ]$ $6$ $8$ $60$ $[ 13 ]$ $[ 4 ]$ $[]$
$312$ $[ 1, 1, 0, 1 ]$ $12$ $4$ $12$ $[ 13 ]$ $[ 1 ]$ $[]$
$312$ $[ 1, 1, 0, 1 ]$ $12$ $4$ $12$ $[ 13 ]$ $[ 1 ]$ $[]$
$313$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$313$ $[ 104 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$313$ $[ 52 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$313$ $[ 26 ]$ $12$ $4$ $12$ $[ 313 ]$ $[ 4 ]$ $[]$
$313$ $[ 26 ]$ $12$ $4$ $12$ $[ 313 ]$ $[ 1 ]$ $[]$
$315$ $[ 0, 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$315$ $[ 0, 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$315$ $[ 0, 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$315$ $[ 0, 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$315$ $[ 3, 0, 3 ]$ $2$ $4$ $12$ $[ 21 ]$ $[ 1 ]$ $[]$
$315$ $[ 3, 0, 3 ]$ $2$ $4$ $12$ $[ 21 ]$ $[ 1 ]$ $[]$
$315$ $[ 3, 2, 3 ]$ $2$ $16$ $24$ $[ -35, 105, -7 ]$ $[ 3, 4, 40 ]$ $[]$
$315$ $[ 0, 0, 2 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$315$ $[ 0, 0, 2 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$315$ $[ 0, 0, 2 ]$ $3$ $4$ $12$ $[]$ $[]$ $[]$
$315$ $[ 2, 0, 2 ]$ $3$ $4$ $13$ $[]$ $[]$ $[]$
$315$ $[ 4, 0, 2 ]$ $3$ $4$ $13$ $[]$ $[]$ $[]$
$315$ $[ 0, 1, 1 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$315$ $[ 0, 1, 1 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$315$ $[ 2, 1, 1 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$315$ $[ 2, 1, 1 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$315$ $[ 2, 1, 1 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$315$ $[ 2, 1, 1 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$315$ $[ 4, 1, 1 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$315$ $[ 4, 1, 1 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$315$ $[ 4, 1, 1 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$315$ $[ 4, 1, 1 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$316$ $[ 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$316$ $[ 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$318$ $[ 0, 0 ]$ $1$ $2$ $41$ $[]$ $[]$ $[]$
$318$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$318$ $[ 0, 26 ]$ $2$ $4$ $17$ $[]$ $[]$ $[]$
$318$ $[ 1, 13 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 53 ]$
$318$ $[ 1, 13 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 53 ]$
$319$ $[ 2, 0 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$320$ $[ 0, 8, 0 ]$ $2$ $4$ $12$ $[]$ $[]$ $[]$
$320$ $[ 0, 8, 0 ]$ $2$ $4$ $12$ $[]$ $[]$ $[]$
$320$ $[ 1, 8, 1 ]$ $4$ $8$ $21$ $[ 5, 40 ]$ $[ 4, 16 ]$ $[]$
$320$ $[ 1, 8, 1 ]$ $4$ $8$ $21$ $[ 5, 40 ]$ $[ 4, 16 ]$ $[]$
$321$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$321$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$321$ $[ 1, 53 ]$ $2$ $4$ $13$ $[ 321 ]$ $[ 4 ]$ $[]$
$321$ $[ 1, 53 ]$ $2$ $4$ $13$ $[ 321 ]$ $[ 4 ]$ $[]$
$322$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$322$ $[ 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$322$ $[ 3, 11 ]$ $2$ $4$ $17$ $[ 161 ]$ $[ 8 ]$ $[]$
$322$ $[ 3, 11 ]$ $2$ $4$ $17$ $[ 161 ]$ $[ 8 ]$ $[]$
$323$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$323$ $[ 4, 0 ]$ $4$ $4$ $24$ $[ 17 ]$ $[ 1 ]$ $[]$
$323$ $[ 2, 0 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$325$ $[ 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$325$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$325$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$325$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$325$ $[ 10, 0 ]$ $2$ $4$ $8$ $[]$ $[]$ $[]$
$325$ $[ 10, 0 ]$ $2$ $4$ $12$ $[]$ $[]$ $[]$
$325$ $[ 10, 0 ]$ $2$ $4$ $8$ $[]$ $[]$ $[]$
$325$ $[ 0, 4 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$325$ $[ 0, 4 ]$ $3$ $4$ $13$ $[]$ $[]$ $[]$
$325$ $[ 0, 2 ]$ $6$ $4$ $21$ $[ 13 ]$ $[ 1 ]$ $[]$
$325$ $[ 0, 2 ]$ $6$ $4$ $21$ $[ 13 ]$ $[ 1 ]$ $[]$
$325$ $[ 10, 4 ]$ $6$ $8$ $13$ $[]$ $[]$ $[]$
$325$ $[ 10, 4 ]$ $6$ $8$ $5$ $[]$ $[]$ $[]$
$325$ $[ 4, 0 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$327$ $[ 0, 54 ]$ $2$ $4$ $24$ $[ 109 ]$ $[ 3 ]$ $[]$
$328$ $[ 0, 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$328$ $[ 0, 0, 20 ]$ $2$ $4$ $5$ $[ 41 ]$ $[ 4 ]$ $[]$
$329$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$329$ $[ 2, 0 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$330$ $[ 0, 2, 0 ]$ $2$ $4$ $40$ $[]$ $[]$ $[]$
$330$ $[ 0, 2, 0 ]$ $2$ $4$ $8$ $[]$ $[]$ $[]$
$330$ $[ 0, 1, 5 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 5, 11 ]$
$330$ $[ 0, 1, 5 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 5, 11 ]$
$330$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$330$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$330$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$330$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$330$ $[ 0, 2, 2 ]$ $10$ $8$ $5$ $[]$ $[]$ $[]$
$330$ $[ 0, 2, 2 ]$ $10$ $8$ $5$ $[]$ $[]$ $[]$
$332$ $[ 0, 0 ]$ $1$ $2$ $28$ $[]$ $[]$ $[]$
$332$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$333$ $[ 0, 0 ]$ $1$ $4$ $24$ $[ -3 ]$ $[ 3 ]$ $[]$
$333$ $[ 0, 18 ]$ $2$ $4$ $24$ $[ 37 ]$ $[ 3 ]$ $[]$
$333$ $[ 0, 6 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$333$ $[ 0, 6 ]$ $6$ $4$ $21$ $[ 37 ]$ $[ 1 ]$ $[]$
$333$ $[ 0, 6 ]$ $6$ $8$ $33$ $[ -111 ]$ $[ 9 ]$ $[]$
$333$ $[ 2, 6 ]$ $6$ $4$ $21$ $[ 37 ]$ $[ 1 ]$ $[]$
$333$ $[ 2, 18 ]$ $6$ $8$ $13$ $[ 37 ]$ $[ 3 ]$ $[]$
$333$ $[ 4, 6 ]$ $6$ $4$ $21$ $[ 37 ]$ $[ 1 ]$ $[]$
$334$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$334$ $[ 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$334$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$335$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$335$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$336$ $[ 0, 0, 1, 3 ]$ $2$ $8$ $17$ $[ -7 ]$ $[ 8 ]$ $[]$
$336$ $[ 1, 0, 1, 0 ]$ $2$ $8$ $12$ $[ -3 ]$ $[ 52 ]$ $[]$
$336$ $[ 0, 0, 0, 2 ]$ $3$ $4$ $57$ $[]$ $[]$ $[]$
$336$ $[ 1, 1, 1, 0 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 2, 3 ]$
$336$ $[ 1, 1, 1, 0 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 2, 3 ]$
$336$ $[ 1, 0, 1, 2 ]$ $6$ $8$ $57$ $[ 12, -4 ]$ $[ 7, 7 ]$ $[]$
$336$ $[ 1, 0, 1, 2 ]$ $6$ $8$ $57$ $[ 12, -4 ]$ $[ 7, 7 ]$ $[]$
$336$ $[ 0, 1, 0, 2 ]$ $12$ $8$ $8$ $[]$ $[]$ $[]$
$339$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$339$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$339$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$339$ $[ 0, 28 ]$ $4$ $4$ $8$ $[]$ $[]$ $[]$
$340$ $[ 0, 2, 8 ]$ $2$ $8$ $28$ $[ 17, 85 ]$ $[ 2, 84 ]$ $[]$
$340$ $[ 1, 1, 0 ]$ $4$ $8$ $8$ $[ -4 ]$ $[ 17 ]$ $[ 5, 5 ]$
$341$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$341$ $[ 5, 15 ]$ $2$ $8$ $13$ $[ -31 ]$ $[ 4 ]$ $[]$
$341$ $[ 4, 6 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$341$ $[ 4, 6 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$341$ $[ 6, 6 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$341$ $[ 6, 6 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$342$ $[ 0, 6 ]$ $3$ $4$ $28$ $[]$ $[]$ $[]$
$342$ $[ 2, 6 ]$ $3$ $4$ $24$ $[]$ $[]$ $[]$
$342$ $[ 4, 6 ]$ $3$ $4$ $24$ $[]$ $[]$ $[]$
$342$ $[ 3, 3 ]$ $6$ $4$ $24$ $[ 57 ]$ $[ 4 ]$ $[]$
$342$ $[ 3, 3 ]$ $6$ $4$ $24$ $[ 57 ]$ $[ 4 ]$ $[]$
$344$ $[ 0, 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$345$ $[ 0, 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$345$ $[ 0, 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$345$ $[ 0, 0, 0 ]$ $1$ $2$ $24$ $[]$ $[]$ $[]$
$345$ $[ 1, 1, 0 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 5 ]$
$345$ $[ 1, 1, 0 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 5 ]$
$347$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$350$ $[ 0, 0 ]$ $1$ $2$ $24$ $[]$ $[]$ $[]$
$350$ $[ 0, 0 ]$ $1$ $2$ $24$ $[]$ $[]$ $[]$
$350$ $[ 5, 3 ]$ $4$ $8$ $8$ $[ -7 ]$ $[ 2 ]$ $[]$
$350$ $[ 5, 3 ]$ $4$ $16$ $12$ $[ -7 ]$ $[ 37 ]$ $[]$
$350$ $[ 5, 1 ]$ $12$ $8$ $12$ $[]$ $[]$ $[ 5, 7 ]$
$350$ $[ 5, 1 ]$ $12$ $8$ $12$ $[]$ $[]$ $[ 5, 7 ]$
$350$ $[ 5, 1 ]$ $12$ $16$ $57$ $[ -7 ]$ $[ 4 ]$ $[]$
$350$ $[ 4, 2 ]$ $15$ $8$ $5$ $[]$ $[]$ $[]$
$351$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$351$ $[ 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$351$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$351$ $[ 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$351$ $[ 0, 0 ]$ $1$ $4$ $37$ $[ -3 ]$ $[ 3 ]$ $[]$
$351$ $[ 0, 0 ]$ $1$ $4$ $5$ $[ -3 ]$ $[ 19 ]$ $[]$
$351$ $[ 0, 6 ]$ $2$ $4$ $13$ $[ 13 ]$ $[ 3 ]$ $[]$
$351$ $[ 0, 6 ]$ $2$ $4$ $13$ $[ 13 ]$ $[ 3 ]$ $[]$
$352$ $[ 0, 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$352$ $[ 0, 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$352$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$352$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$354$ $[ 0, 0 ]$ $1$ $2$ $44$ $[]$ $[]$ $[]$
$355$ $[ 0, 14 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$355$ $[ 0, 14 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$357$ $[ 0, 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$357$ $[ 0, 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$357$ $[ 1, 1, 0 ]$ $6$ $4$ $33$ $[ 21 ]$ $[ 1 ]$ $[]$
$357$ $[ 1, 1, 0 ]$ $6$ $4$ $33$ $[ 21 ]$ $[ 1 ]$ $[]$
$358$ $[ 0 ]$ $1$ $2$ $21$ $[]$ $[]$ $[]$
$358$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$358$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$358$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$360$ $[ 0, 1, 0, 0 ]$ $2$ $4$ $12$ $[]$ $[]$ $[]$
$360$ $[ 1, 1, 3, 2 ]$ $2$ $16$ $17$ $[ -15, 120, -40 ]$ $[ 8, 32, 104 ]$ $[]$
$360$ $[ 0, 0, 2, 0 ]$ $3$ $4$ $12$ $[]$ $[]$ $[]$
$360$ $[ 0, 0, 2, 0 ]$ $3$ $4$ $24$ $[]$ $[]$ $[]$
$360$ $[ 0, 0, 3, 1 ]$ $4$ $8$ $5$ $[ -15 ]$ $[ 4 ]$ $[]$
$360$ $[ 1, 1, 0, 1 ]$ $4$ $16$ $41$ $[ 120, -8, -40 ]$ $[ 40, 80, 160 ]$ $[]$
$360$ $[ 0, 1, 1, 1 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 9 ]$ $[]$
$360$ $[ 0, 1, 1, 1 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 9 ]$ $[]$
$360$ $[ 0, 1, 1, 1 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 9 ]$ $[]$
$360$ $[ 0, 1, 1, 1 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 9 ]$ $[]$
$361$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$361$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$361$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$361$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$361$ $[ 0 ]$ $1$ $4$ $5$ $[ -19 ]$ $[ 5 ]$ $[]$
$361$ $[ 114 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$361$ $[ 114 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$361$ $[ 114 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$361$ $[ 114 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$361$ $[ 114 ]$ $3$ $8$ $5$ $[ -19 ]$ $[ 5 ]$ $[]$
$361$ $[ 38 ]$ $9$ $12$ $5$ $[]$ $[]$ $[]$
$361$ $[ 38 ]$ $9$ $12$ $5$ $[]$ $[]$ $[]$
$361$ $[ 38 ]$ $9$ $12$ $5$ $[]$ $[]$ $[]$
$361$ $[ 38 ]$ $9$ $12$ $5$ $[]$ $[]$ $[]$
$361$ $[ 38 ]$ $9$ $24$ $5$ $[ -19 ]$ $[ 5 ]$ $[]$
$362$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$362$ $[ 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$362$ $[ 36 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$362$ $[ 36 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$362$ $[ 36 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$363$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$363$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$363$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$363$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$363$ $[ 0, 0 ]$ $1$ $4$ $33$ $[ -11 ]$ $[ 4 ]$ $[]$
$363$ $[ 1, 55 ]$ $2$ $8$ $5$ $[ -11 ]$ $[ 5 ]$ $[ 3, 3 ]$
$363$ $[ 1, 55 ]$ $2$ $8$ $12$ $[ 33, -3 ]$ $[ 9, 13 ]$ $[]$
$363$ $[ 0, 22 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$363$ $[ 0, 22 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$363$ $[ 0, 22 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$363$ $[ 0, 22 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$363$ $[ 0, 22 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$363$ $[ 0, 22 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$363$ $[ 0, 22 ]$ $5$ $16$ $33$ $[ -11 ]$ $[ 4 ]$ $[]$
$363$ $[ 1, 11 ]$ $10$ $8$ $5$ $[ -11 ]$ $[ 5 ]$ $[ 3, 3 ]$
$363$ $[ 1, 11 ]$ $10$ $8$ $5$ $[ -11 ]$ $[ 5 ]$ $[ 3, 3 ]$
$363$ $[ 1, 11 ]$ $10$ $8$ $5$ $[ -11 ]$ $[ 5 ]$ $[ 3, 3 ]$
$363$ $[ 1, 11 ]$ $10$ $32$ $12$ $[ 33, -3 ]$ $[ 9, 13 ]$ $[]$
$364$ $[ 0, 0, 0 ]$ $1$ $2$ $24$ $[]$ $[]$ $[]$
$364$ $[ 0, 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$364$ $[ 0, 0, 4 ]$ $3$ $4$ $21$ $[]$ $[]$ $[]$
$364$ $[ 0, 0, 4 ]$ $3$ $4$ $13$ $[]$ $[]$ $[]$
$364$ $[ 0, 2, 6 ]$ $6$ $8$ $37$ $[ 13 ]$ $[ 27 ]$ $[]$
$364$ $[ 1, 3, 4 ]$ $6$ $8$ $13$ $[]$ $[]$ $[ 2, 7 ]$
$364$ $[ 1, 3, 4 ]$ $6$ $8$ $13$ $[]$ $[]$ $[ 2, 7 ]$
$364$ $[ 1, 0, 1 ]$ $12$ $8$ $12$ $[ -52 ]$ $[ 1 ]$ $[]$
$365$ $[ 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$365$ $[ 0, 12 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$366$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$366$ $[ 0, 30 ]$ $2$ $4$ $41$ $[]$ $[]$ $[]$
$366$ $[ 0, 20 ]$ $3$ $4$ $60$ $[]$ $[]$ $[]$
$366$ $[ 0, 12 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$368$ $[ 0, 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$368$ $[ 0, 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$368$ $[ 1, 0, 11 ]$ $2$ $8$ $33$ $[ -4 ]$ $[ 4 ]$ $[]$
$368$ $[ 0, 1, 0 ]$ $4$ $4$ $12$ $[ 8 ]$ $[ 4 ]$ $[]$
$369$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$369$ $[ 0, 8 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$369$ $[ 0, 4 ]$ $10$ $4$ $5$ $[ 41 ]$ $[ 5 ]$ $[]$
$369$ $[ 0, 4 ]$ $10$ $4$ $5$ $[ 41 ]$ $[ 5 ]$ $[]$
$369$ $[ 2, 10 ]$ $12$ $4$ $12$ $[ 41 ]$ $[ 9 ]$ $[]$
$369$ $[ 2, 10 ]$ $12$ $4$ $12$ $[ 41 ]$ $[ 9 ]$ $[]$
$370$ $[ 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$370$ $[ 0, 0 ]$ $1$ $2$ $33$ $[]$ $[]$ $[]$
$370$ $[ 2, 0 ]$ $2$ $4$ $24$ $[]$ $[]$ $[]$
$370$ $[ 0, 12 ]$ $3$ $4$ $40$ $[]$ $[]$ $[]$
$370$ $[ 0, 12 ]$ $3$ $4$ $12$ $[]$ $[]$ $[]$
$370$ $[ 1, 9 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 5, 37 ]$
$370$ $[ 3, 9 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 5, 37 ]$
$370$ $[ 0, 6 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$370$ $[ 0, 6 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$370$ $[ 2, 6 ]$ $6$ $4$ $33$ $[ 185 ]$ $[ 4 ]$ $[]$
$370$ $[ 2, 6 ]$ $6$ $4$ $33$ $[ 185 ]$ $[ 4 ]$ $[]$
$370$ $[ 2, 12 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$370$ $[ 2, 12 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$370$ $[ 2, 12 ]$ $6$ $8$ $44$ $[]$ $[]$ $[]$
$370$ $[ 1, 3 ]$ $12$ $4$ $12$ $[ 185 ]$ $[ 1 ]$ $[]$
$370$ $[ 1, 3 ]$ $12$ $4$ $12$ $[ 185 ]$ $[ 1 ]$ $[]$
$370$ $[ 3, 3 ]$ $12$ $4$ $12$ $[ 185 ]$ $[ 1 ]$ $[]$
$370$ $[ 3, 3 ]$ $12$ $4$ $12$ $[ 185 ]$ $[ 1 ]$ $[]$
$371$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$371$ $[ 0, 26 ]$ $2$ $4$ $33$ $[ 53 ]$ $[ 4 ]$ $[]$
$371$ $[ 3, 13 ]$ $4$ $4$ $40$ $[ 53 ]$ $[ 4 ]$ $[ 7, 7 ]$
$371$ $[ 3, 13 ]$ $4$ $4$ $40$ $[ 53 ]$ $[ 4 ]$ $[ 7, 7 ]$
$371$ $[ 1, 13 ]$ $12$ $4$ $12$ $[ 53 ]$ $[ 1 ]$ $[ 7, 7 ]$
$371$ $[ 1, 13 ]$ $12$ $4$ $12$ $[ 53 ]$ $[ 1 ]$ $[ 7, 7 ]$
$372$ $[ 0, 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$372$ $[ 1, 1, 0 ]$ $2$ $4$ $8$ $[]$ $[]$ $[ 2, 3 ]$
$372$ $[ 1, 1, 0 ]$ $2$ $4$ $8$ $[]$ $[]$ $[ 2, 3 ]$
$372$ $[ 0, 0, 10 ]$ $3$ $4$ $28$ $[]$ $[]$ $[]$
$372$ $[ 0, 0, 6 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$373$ $[ 186 ]$ $2$ $4$ $93$ $[]$ $[]$ $[]$
$374$ $[ 2, 0 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$374$ $[ 2, 0 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$374$ $[ 2, 0 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$375$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$375$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$375$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$375$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$375$ $[ 0, 50 ]$ $2$ $4$ $5$ $[]$ $[]$ $[]$
$375$ $[ 0, 50 ]$ $2$ $4$ $5$ $[]$ $[]$ $[]$
$375$ $[ 0, 20 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$375$ $[ 0, 10 ]$ $10$ $8$ $5$ $[]$ $[]$ $[]$
$376$ $[ 0, 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$376$ $[ 0, 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$377$ $[ 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$377$ $[ 6, 0 ]$ $2$ $4$ $24$ $[ 13 ]$ $[ 9 ]$ $[]$
$378$ $[ 0, 2 ]$ $3$ $4$ $28$ $[]$ $[]$ $[]$
$378$ $[ 0, 2 ]$ $3$ $4$ $28$ $[]$ $[]$ $[]$
$378$ $[ 6, 0 ]$ $3$ $4$ $24$ $[]$ $[]$ $[]$
$378$ $[ 6, 0 ]$ $3$ $4$ $33$ $[]$ $[]$ $[]$
$378$ $[ 9, 1 ]$ $6$ $8$ $8$ $[ -7 ]$ $[ 9 ]$ $[]$
$380$ $[ 0, 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$380$ $[ 0, 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$380$ $[ 0, 2, 0 ]$ $2$ $4$ $5$ $[ 5 ]$ $[ 4 ]$ $[]$
$381$ $[ 0, 42 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$381$ $[ 1, 21 ]$ $6$ $4$ $33$ $[ 381 ]$ $[ 4 ]$ $[]$
$381$ $[ 1, 21 ]$ $6$ $4$ $33$ $[ 381 ]$ $[ 4 ]$ $[]$
$382$ $[ 38 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$382$ $[ 38 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$383$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$384$ $[ 1, 0, 1 ]$ $2$ $4$ $5$ $[]$ $[]$ $[ 2, 3 ]$
$384$ $[ 1, 0, 1 ]$ $2$ $4$ $5$ $[]$ $[]$ $[ 2, 3 ]$
$384$ $[ 1, 0, 1 ]$ $2$ $4$ $5$ $[]$ $[]$ $[ 2, 3 ]$
$384$ $[ 1, 0, 1 ]$ $2$ $4$ $5$ $[]$ $[]$ $[ 2, 3 ]$
$385$ $[ 0, 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$385$ $[ 0, 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$385$ $[ 1, 3, 0 ]$ $4$ $4$ $12$ $[ 5 ]$ $[ 4 ]$ $[]$
$385$ $[ 1, 3, 0 ]$ $4$ $4$ $12$ $[ 5 ]$ $[ 4 ]$ $[]$
$385$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$385$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$386$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$386$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$386$ $[ 64 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$386$ $[ 32 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$386$ $[ 32 ]$ $6$ $4$ $249$ $[ 193 ]$ $[ 9 ]$ $[]$
$386$ $[ 32 ]$ $6$ $4$ $24$ $[ 193 ]$ $[ 1 ]$ $[]$
$386$ $[ 32 ]$ $6$ $4$ $21$ $[ 193 ]$ $[ 1 ]$ $[]$
$386$ $[ 16 ]$ $12$ $4$ $12$ $[ 193 ]$ $[ 1 ]$ $[]$
$387$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$387$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$387$ $[ 0, 0 ]$ $1$ $4$ $65$ $[ -3 ]$ $[ 4 ]$ $[]$
$387$ $[ 0, 14 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$387$ $[ 2, 0 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$388$ $[ 0, 12 ]$ $8$ $4$ $8$ $[ 97 ]$ $[ 2 ]$ $[]$
$389$ $[ 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$390$ $[ 0, 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$390$ $[ 0, 0, 6 ]$ $2$ $4$ $17$ $[]$ $[]$ $[]$
$390$ $[ 0, 0, 6 ]$ $2$ $4$ $13$ $[]$ $[]$ $[]$
$390$ $[ 0, 2, 0 ]$ $2$ $4$ $5$ $[]$ $[]$ $[]$
$390$ $[ 0, 0, 4 ]$ $3$ $4$ $17$ $[]$ $[]$ $[]$
$390$ $[ 1, 0, 3 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 13 ]$
$390$ $[ 1, 0, 3 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 13 ]$
$390$ $[ 1, 0, 3 ]$ $4$ $8$ $40$ $[ -39 ]$ $[ 4 ]$ $[]$
$390$ $[ 1, 1, 0 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 5 ]$
$390$ $[ 1, 1, 0 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 5 ]$
$390$ $[ 0, 0, 2 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$390$ $[ 0, 0, 2 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$390$ $[ 0, 2, 4 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$390$ $[ 0, 2, 4 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$390$ $[ 0, 2, 4 ]$ $6$ $8$ $44$ $[]$ $[]$ $[]$
$390$ $[ 1, 0, 1 ]$ $12$ $8$ $12$ $[ -39 ]$ $[ 4 ]$ $[]$
$390$ $[ 1, 1, 2 ]$ $12$ $16$ $120$ $[ -15, 65 ]$ $[ 1, 4 ]$ $[]$
$391$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$392$ $[ 0, 1, 0 ]$ $2$ $4$ $17$ $[ 8 ]$ $[ 8 ]$ $[]$
$392$ $[ 1, 1, 21 ]$ $2$ $8$ $8$ $[ 56, -8 ]$ $[ 2, 18 ]$ $[]$
$392$ $[ 1, 1, 21 ]$ $2$ $8$ $8$ $[ 56, -8 ]$ $[ 2, 136 ]$ $[]$
$392$ $[ 0, 1, 14 ]$ $6$ $8$ $17$ $[ 8 ]$ $[ 8 ]$ $[]$
$392$ $[ 0, 1, 14 ]$ $6$ $8$ $17$ $[ 8 ]$ $[ 8 ]$ $[]$
$392$ $[ 1, 1, 7 ]$ $6$ $8$ $8$ $[ 56, -8 ]$ $[ 2, 18 ]$ $[]$
$392$ $[ 1, 1, 7 ]$ $6$ $8$ $8$ $[ 56, -8 ]$ $[ 2, 18 ]$ $[]$
$392$ $[ 1, 1, 7 ]$ $6$ $16$ $8$ $[ 56, -8 ]$ $[ 2, 136 ]$ $[]$
$393$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$394$ $[ 0 ]$ $1$ $2$ $29$ $[]$ $[]$ $[]$
$394$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$394$ $[ 0 ]$ $1$ $2$ $21$ $[]$ $[]$ $[]$
$394$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$394$ $[ 98 ]$ $2$ $4$ $29$ $[]$ $[]$ $[]$
$395$ $[ 2, 26 ]$ $6$ $8$ $13$ $[]$ $[]$ $[]$
$396$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$396$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$396$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$397$ $[ 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$397$ $[ 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$398$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$398$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$398$ $[ 66 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$399$ $[ 0, 2, 0 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$399$ $[ 0, 1, 3 ]$ $6$ $4$ $12$ $[]$ $[]$ $[ 7, 19 ]$
$399$ $[ 0, 1, 9 ]$ $6$ $4$ $12$ $[]$ $[]$ $[ 7, 19 ]$
$399$ $[ 0, 1, 9 ]$ $6$ $4$ $12$ $[]$ $[]$ $[ 7, 19 ]$
$399$ $[ 0, 5, 3 ]$ $6$ $4$ $12$ $[]$ $[]$ $[ 7, 19 ]$
$400$ $[ 0, 1, 0 ]$ $4$ $4$ $44$ $[ 8 ]$ $[ 1 ]$ $[]$
$400$ $[ 0, 1, 0 ]$ $4$ $4$ $44$ $[ 8 ]$ $[ 1 ]$ $[]$
$400$ $[ 0, 1, 10 ]$ $4$ $4$ $44$ $[ 8 ]$ $[ 1 ]$ $[]$
$400$ $[ 0, 1, 10 ]$ $4$ $4$ $44$ $[ 8 ]$ $[ 1 ]$ $[]$
$400$ $[ 1, 1, 5 ]$ $4$ $8$ $12$ $[ 8 ]$ $[ 1 ]$ $[]$
$400$ $[ 1, 3, 5 ]$ $4$ $8$ $12$ $[ 8 ]$ $[ 1 ]$ $[]$
$400$ $[ 0, 0, 4 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$400$ $[ 0, 0, 4 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$400$ $[ 0, 0, 4 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$401$ $[ 80 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$401$ $[ 50 ]$ $8$ $4$ $8$ $[ 401 ]$ $[ 8 ]$ $[]$
$402$ $[ 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$402$ $[ 0, 0 ]$ $1$ $2$ $41$ $[]$ $[]$ $[]$
$403$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$403$ $[ 0, 6 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$403$ $[ 0, 6 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$403$ $[ 0, 6 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$403$ $[ 3, 5 ]$ $12$ $4$ $12$ $[ 13 ]$ $[ 1 ]$ $[]$
$403$ $[ 3, 5 ]$ $12$ $4$ $12$ $[ 13 ]$ $[ 1 ]$ $[]$
$403$ $[ 4, 6 ]$ $15$ $8$ $5$ $[]$ $[]$ $[]$
$405$ $[ 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$405$ $[ 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$405$ $[ 0, 2 ]$ $2$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$405$ $[ 0, 2 ]$ $2$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$405$ $[ 0, 2 ]$ $2$ $4$ $33$ $[ 5 ]$ $[ 4 ]$ $[]$
$405$ $[ 0, 2 ]$ $2$ $4$ $33$ $[ 5 ]$ $[ 4 ]$ $[]$
$405$ $[ 18, 0 ]$ $3$ $4$ $13$ $[]$ $[]$ $[]$
$405$ $[ 18, 0 ]$ $3$ $4$ $13$ $[]$ $[]$ $[]$
$405$ $[ 18, 0 ]$ $3$ $4$ $12$ $[]$ $[]$ $[]$
$405$ $[ 18, 0 ]$ $3$ $4$ $12$ $[]$ $[]$ $[]$
$405$ $[ 18, 2 ]$ $6$ $4$ $33$ $[ 5 ]$ $[ 4 ]$ $[]$
$405$ $[ 18, 2 ]$ $6$ $4$ $33$ $[ 5 ]$ $[ 4 ]$ $[]$
$405$ $[ 18, 2 ]$ $6$ $4$ $33$ $[ 5 ]$ $[ 4 ]$ $[]$
$405$ $[ 18, 2 ]$ $6$ $4$ $33$ $[ 5 ]$ $[ 4 ]$ $[]$
$405$ $[ 9, 1 ]$ $12$ $16$ $21$ $[ 5, -15 ]$ $[ 1, 100 ]$ $[]$
$406$ $[ 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$407$ $[ 0, 12 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$407$ $[ 0, 6 ]$ $6$ $4$ $21$ $[ 37 ]$ $[ 1 ]$ $[]$
$407$ $[ 0, 6 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$408$ $[ 0, 0, 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$408$ $[ 0, 0, 0, 0 ]$ $1$ $2$ $57$ $[]$ $[]$ $[]$
$408$ $[ 0, 0, 0, 8 ]$ $2$ $4$ $33$ $[]$ $[]$ $[]$
$408$ $[ 0, 1, 0, 0 ]$ $2$ $4$ $5$ $[]$ $[]$ $[]$
$408$ $[ 1, 1, 1, 0 ]$ $2$ $4$ $24$ $[]$ $[]$ $[ 2, 3 ]$
$408$ $[ 1, 1, 1, 0 ]$ $2$ $4$ $8$ $[]$ $[]$ $[ 2, 3 ]$
$408$ $[ 1, 1, 1, 0 ]$ $2$ $4$ $24$ $[]$ $[]$ $[ 2, 3 ]$
$408$ $[ 1, 1, 1, 0 ]$ $2$ $4$ $8$ $[]$ $[]$ $[ 2, 3 ]$
$408$ $[ 0, 0, 0, 4 ]$ $4$ $4$ $8$ $[]$ $[]$ $[]$
$408$ $[ 0, 0, 0, 4 ]$ $4$ $4$ $8$ $[]$ $[]$ $[]$
$409$ $[ 34 ]$ $12$ $4$ $12$ $[ 409 ]$ $[ 1 ]$ $[]$
$410$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$410$ $[ 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$410$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$410$ $[ 0, 0 ]$ $1$ $2$ $24$ $[]$ $[]$ $[]$
$410$ $[ 0, 20 ]$ $2$ $4$ $17$ $[ 41 ]$ $[ 8 ]$ $[]$
$410$ $[ 0, 20 ]$ $2$ $4$ $28$ $[ 41 ]$ $[ 8 ]$ $[]$
$410$ $[ 0, 10 ]$ $4$ $4$ $33$ $[ 41 ]$ $[ 4 ]$ $[]$
$410$ $[ 0, 10 ]$ $4$ $4$ $5$ $[ 41 ]$ $[ 4 ]$ $[]$
$411$ $[ 0, 34 ]$ $4$ $4$ $8$ $[]$ $[]$ $[]$
$412$ $[ 0, 0 ]$ $1$ $2$ $21$ $[]$ $[]$ $[]$
$412$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$413$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$414$ $[ 0, 0 ]$ $1$ $2$ $28$ $[]$ $[]$ $[]$
$414$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$414$ $[ 0, 0 ]$ $1$ $2$ $28$ $[]$ $[]$ $[]$
$414$ $[ 3, 11 ]$ $2$ $8$ $8$ $[ -23 ]$ $[ 2 ]$ $[ 3, 3 ]$
$415$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$416$ $[ 0, 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$416$ $[ 0, 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$416$ $[ 0, 0, 0 ]$ $1$ $4$ $41$ $[ -4 ]$ $[ 32 ]$ $[]$
$416$ $[ 0, 4, 0 ]$ $2$ $4$ $12$ $[]$ $[]$ $[]$
$416$ $[ 0, 4, 6 ]$ $2$ $8$ $41$ $[ 13, 8 ]$ $[ 10, 20 ]$ $[]$
$416$ $[ 0, 0, 4 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$416$ $[ 0, 0, 4 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$416$ $[ 1, 0, 3 ]$ $4$ $4$ $40$ $[ 13 ]$ $[ 9 ]$ $[]$
$416$ $[ 1, 0, 3 ]$ $4$ $4$ $40$ $[ 13 ]$ $[ 9 ]$ $[]$
$416$ $[ 0, 0, 2 ]$ $6$ $8$ $12$ $[ -4 ]$ $[ 13 ]$ $[]$
$416$ $[ 0, 4, 2 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$416$ $[ 0, 4, 2 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$416$ $[ 1, 0, 1 ]$ $12$ $4$ $12$ $[ 13 ]$ $[ 1 ]$ $[]$
$416$ $[ 1, 0, 1 ]$ $12$ $4$ $12$ $[ 13 ]$ $[ 1 ]$ $[]$
$417$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$418$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$418$ $[ 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$418$ $[ 0, 0 ]$ $1$ $2$ $21$ $[]$ $[]$ $[]$
$418$ $[ 0, 6 ]$ $3$ $4$ $24$ $[]$ $[]$ $[]$
$418$ $[ 2, 0 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$418$ $[ 2, 0 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$418$ $[ 2, 0 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$418$ $[ 2, 0 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$418$ $[ 2, 0 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$418$ $[ 2, 6 ]$ $15$ $8$ $5$ $[]$ $[]$ $[]$
$418$ $[ 2, 6 ]$ $15$ $8$ $5$ $[]$ $[]$ $[]$
$418$ $[ 2, 6 ]$ $15$ $8$ $5$ $[]$ $[]$ $[]$
$420$ $[ 0, 1, 0, 3 ]$ $2$ $4$ $5$ $[]$ $[]$ $[ 3, 7 ]$
$420$ $[ 0, 1, 0, 3 ]$ $2$ $4$ $5$ $[]$ $[]$ $[ 3, 7 ]$
$420$ $[ 0, 0, 0, 2 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$420$ $[ 0, 0, 0, 2 ]$ $3$ $4$ $28$ $[]$ $[]$ $[]$
$420$ $[ 1, 0, 0, 1 ]$ $6$ $4$ $12$ $[]$ $[]$ $[ 2, 7 ]$
$420$ $[ 1, 0, 0, 1 ]$ $6$ $4$ $12$ $[]$ $[]$ $[ 2, 7 ]$
$420$ $[ 0, 1, 1, 2 ]$ $12$ $8$ $44$ $[ 5 ]$ $[ 25 ]$ $[]$
$420$ $[ 0, 1, 1, 2 ]$ $12$ $8$ $44$ $[ 5 ]$ $[ 25 ]$ $[]$
$421$ $[ 210 ]$ $2$ $4$ $5$ $[ 421 ]$ $[ 11 ]$ $[]$
$422$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$422$ $[ 70 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$422$ $[ 70 ]$ $3$ $4$ $29$ $[]$ $[]$ $[]$
$423$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$423$ $[ 3, 23 ]$ $2$ $4$ $5$ $[ 141 ]$ $[ 4 ]$ $[]$
$423$ $[ 3, 23 ]$ $2$ $4$ $5$ $[ 141 ]$ $[ 4 ]$ $[]$
$423$ $[ 3, 23 ]$ $2$ $8$ $12$ $[ 141, -47 ]$ $[ 1, 6 ]$ $[]$
$424$ $[ 0, 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$425$ $[ 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$425$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$425$ $[ 0, 8 ]$ $2$ $8$ $28$ $[ 17, 85 ]$ $[ 2, 9 ]$ $[]$
$425$ $[ 10, 0 ]$ $2$ $4$ $12$ $[]$ $[]$ $[]$
$425$ $[ 10, 0 ]$ $2$ $4$ $8$ $[]$ $[]$ $[]$
$425$ $[ 4, 0 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$425$ $[ 0, 2 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$425$ $[ 10, 2 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$425$ $[ 10, 2 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$426$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$426$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$426$ $[ 1, 35 ]$ $2$ $8$ $5$ $[ -71 ]$ $[ 20 ]$ $[ 3, 3 ]$
$426$ $[ 0, 14 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$426$ $[ 0, 14 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$426$ $[ 1, 7 ]$ $10$ $8$ $5$ $[ -71 ]$ $[ 5 ]$ $[ 3, 3 ]$
$427$ $[ 0, 12 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$427$ $[ 0, 12 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$427$ $[ 0, 12 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$428$ $[ 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$429$ $[ 0, 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$429$ $[ 0, 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$430$ $[ 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$430$ $[ 0, 0 ]$ $1$ $2$ $24$ $[]$ $[]$ $[]$
$430$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$430$ $[ 0, 14 ]$ $3$ $4$ $40$ $[]$ $[]$ $[]$
$432$ $[ 0, 0, 6 ]$ $3$ $4$ $33$ $[]$ $[]$ $[]$
$432$ $[ 0, 1, 6 ]$ $12$ $4$ $12$ $[ 8 ]$ $[ 4 ]$ $[]$
$432$ $[ 0, 1, 6 ]$ $12$ $4$ $12$ $[ 8 ]$ $[ 4 ]$ $[]$
$432$ $[ 0, 1, 6 ]$ $12$ $4$ $12$ $[ 8 ]$ $[ 1 ]$ $[]$
$432$ $[ 0, 1, 6 ]$ $12$ $4$ $12$ $[ 8 ]$ $[ 1 ]$ $[]$
$434$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$434$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$434$ $[ 3, 15 ]$ $2$ $8$ $5$ $[ -31 ]$ $[ 20 ]$ $[ 7, 7 ]$
$434$ $[ 3, 15 ]$ $2$ $8$ $105$ $[ -31, 217 ]$ $[ 4, 4 ]$ $[]$
$434$ $[ 0, 10 ]$ $3$ $4$ $13$ $[]$ $[]$ $[]$
$434$ $[ 0, 10 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$434$ $[ 2, 0 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$434$ $[ 0, 6 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$434$ $[ 2, 6 ]$ $15$ $8$ $5$ $[]$ $[]$ $[]$
$435$ $[ 0, 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$435$ $[ 0, 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$435$ $[ 0, 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$435$ $[ 0, 0, 0 ]$ $1$ $2$ $21$ $[]$ $[]$ $[]$
$435$ $[ 0, 2, 0 ]$ $2$ $4$ $8$ $[]$ $[]$ $[]$
$435$ $[ 0, 1, 7 ]$ $4$ $4$ $12$ $[ 145 ]$ $[ 4 ]$ $[]$
$435$ $[ 0, 3, 7 ]$ $4$ $4$ $12$ $[ 145 ]$ $[ 4 ]$ $[]$
$435$ $[ 1, 0, 7 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 29 ]$
$435$ $[ 1, 0, 7 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 29 ]$
$435$ $[ 1, 1, 0 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 5 ]$
$435$ $[ 1, 1, 0 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 3, 5 ]$
$436$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$437$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$437$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$437$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$438$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$438$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$438$ $[ 0, 36 ]$ $2$ $4$ $12$ $[ 73 ]$ $[ 16 ]$ $[]$
$438$ $[ 0, 18 ]$ $4$ $4$ $12$ $[ 73 ]$ $[ 4 ]$ $[]$
$438$ $[ 0, 12 ]$ $6$ $4$ $60$ $[ 73 ]$ $[ 4 ]$ $[]$
$438$ $[ 0, 12 ]$ $6$ $4$ $24$ $[ 73 ]$ $[ 4 ]$ $[]$
$438$ $[ 0, 12 ]$ $6$ $4$ $105$ $[ 73 ]$ $[ 4 ]$ $[]$
$438$ $[ 0, 6 ]$ $12$ $4$ $12$ $[ 73 ]$ $[ 1 ]$ $[]$
$438$ $[ 0, 6 ]$ $12$ $4$ $12$ $[ 73 ]$ $[ 1 ]$ $[]$
$438$ $[ 0, 6 ]$ $12$ $4$ $12$ $[ 73 ]$ $[ 9 ]$ $[]$
$439$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$439$ $[ 146 ]$ $3$ $4$ $104$ $[]$ $[]$ $[]$
$440$ $[ 0, 0, 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$440$ $[ 0, 0, 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$440$ $[ 0, 0, 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$440$ $[ 0, 1, 0, 0 ]$ $2$ $4$ $8$ $[]$ $[]$ $[]$
$440$ $[ 1, 1, 1, 0 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 2, 5 ]$
$440$ $[ 1, 1, 1, 0 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 2, 5 ]$
$440$ $[ 0, 1, 0, 2 ]$ $10$ $8$ $5$ $[]$ $[]$ $[]$
$440$ $[ 0, 1, 0, 2 ]$ $10$ $8$ $5$ $[]$ $[]$ $[]$
$441$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$441$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$441$ $[ 3, 21 ]$ $2$ $8$ $8$ $[ -7 ]$ $[ 98 ]$ $[]$
$441$ $[ 0, 14 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$441$ $[ 0, 14 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$441$ $[ 3, 7 ]$ $6$ $16$ $8$ $[ -7 ]$ $[ 98 ]$ $[]$
$442$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$442$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$442$ $[ 0, 8 ]$ $2$ $4$ $73$ $[ 17 ]$ $[ 2 ]$ $[]$
$442$ $[ 0, 8 ]$ $2$ $4$ $8$ $[ 17 ]$ $[ 8 ]$ $[]$
$442$ $[ 6, 0 ]$ $2$ $4$ $12$ $[]$ $[]$ $[]$
$442$ $[ 0, 4 ]$ $4$ $4$ $12$ $[ 17 ]$ $[ 1 ]$ $[]$
$442$ $[ 0, 4 ]$ $4$ $4$ $13$ $[ 17 ]$ $[ 36 ]$ $[]$
$442$ $[ 0, 2 ]$ $8$ $4$ $8$ $[ 17 ]$ $[ 2 ]$ $[]$
$442$ $[ 4, 4 ]$ $12$ $8$ $40$ $[ 17 ]$ $[ 4 ]$ $[]$
$444$ $[ 0, 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$444$ $[ 0, 0, 0 ]$ $1$ $2$ $24$ $[]$ $[]$ $[]$
$444$ $[ 0, 0, 18 ]$ $2$ $4$ $13$ $[ 37 ]$ $[ 12 ]$ $[]$
$444$ $[ 0, 0, 18 ]$ $2$ $4$ $5$ $[ 37 ]$ $[ 4 ]$ $[]$
$444$ $[ 0, 0, 12 ]$ $3$ $4$ $37$ $[]$ $[]$ $[]$
$444$ $[ 0, 0, 6 ]$ $6$ $4$ $21$ $[ 37 ]$ $[ 1 ]$ $[]$
$445$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$445$ $[ 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$445$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$448$ $[ 0, 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$448$ $[ 0, 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$448$ $[ 0, 8, 0 ]$ $2$ $4$ $12$ $[]$ $[]$ $[]$
$448$ $[ 0, 8, 0 ]$ $2$ $4$ $12$ $[]$ $[]$ $[]$
$448$ $[ 1, 0, 3 ]$ $2$ $8$ $8$ $[ -7 ]$ $[ 8 ]$ $[ 2, 2 ]$
$448$ $[ 0, 0, 2 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$448$ $[ 0, 0, 2 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$448$ $[ 0, 4, 2 ]$ $12$ $4$ $12$ $[ 8 ]$ $[ 1 ]$ $[]$
$448$ $[ 0, 4, 2 ]$ $12$ $4$ $12$ $[ 8 ]$ $[ 1 ]$ $[]$
$450$ $[ 2, 0 ]$ $3$ $4$ $24$ $[]$ $[]$ $[]$
$450$ $[ 2, 0 ]$ $3$ $4$ $24$ $[]$ $[]$ $[]$
$450$ $[ 2, 0 ]$ $3$ $4$ $24$ $[]$ $[]$ $[]$
$450$ $[ 2, 0 ]$ $3$ $4$ $33$ $[]$ $[]$ $[]$
$450$ $[ 2, 0 ]$ $3$ $4$ $24$ $[]$ $[]$ $[]$
$450$ $[ 2, 10 ]$ $6$ $8$ $24$ $[]$ $[]$ $[]$
$450$ $[ 2, 10 ]$ $6$ $8$ $33$ $[]$ $[]$ $[]$
$450$ $[ 0, 4 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$450$ $[ 0, 4 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$450$ $[ 0, 4 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$450$ $[ 1, 5 ]$ $12$ $8$ $12$ $[]$ $[]$ $[ 3, 5 ]$
$450$ $[ 1, 5 ]$ $12$ $8$ $12$ $[]$ $[]$ $[ 3, 5 ]$
$450$ $[ 1, 5 ]$ $12$ $8$ $24$ $[ -15 ]$ $[ 1 ]$ $[]$
$450$ $[ 2, 4 ]$ $15$ $8$ $5$ $[]$ $[]$ $[]$
$453$ $[ 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$453$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$453$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$453$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$453$ $[ 0, 50 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$454$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$455$ $[ 1, 0, 3 ]$ $4$ $4$ $12$ $[ 65 ]$ $[ 4 ]$ $[]$
$455$ $[ 1, 3, 0 ]$ $4$ $8$ $12$ $[ -35 ]$ $[ 1 ]$ $[]$
$455$ $[ 1, 3, 6 ]$ $4$ $8$ $5$ $[ 65 ]$ $[ 4 ]$ $[ 7, 7 ]$
$455$ $[ 1, 3, 6 ]$ $4$ $8$ $5$ $[ 65 ]$ $[ 4 ]$ $[ 7, 7 ]$
$455$ $[ 3, 0, 3 ]$ $4$ $4$ $12$ $[ 65 ]$ $[ 4 ]$ $[]$
$455$ $[ 1, 0, 1 ]$ $12$ $4$ $12$ $[ 65 ]$ $[ 1 ]$ $[]$
$455$ $[ 3, 0, 1 ]$ $12$ $4$ $12$ $[ 65 ]$ $[ 1 ]$ $[]$
$456$ $[ 0, 0, 0, 0 ]$ $1$ $2$ $41$ $[]$ $[]$ $[]$
$456$ $[ 0, 0, 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$456$ $[ 1, 1, 1, 0 ]$ $2$ $8$ $33$ $[ -3, -8 ]$ $[ 4, 12 ]$ $[]$
$456$ $[ 0, 0, 0, 6 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$456$ $[ 0, 0, 0, 6 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$456$ $[ 0, 0, 1, 3 ]$ $6$ $4$ $33$ $[ 57 ]$ $[ 4 ]$ $[]$
$456$ $[ 0, 0, 1, 3 ]$ $6$ $4$ $33$ $[ 57 ]$ $[ 4 ]$ $[]$
$457$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$458$ $[ 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$458$ $[ 38 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$458$ $[ 38 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$458$ $[ 38 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$459$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$459$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$459$ $[ 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$459$ $[ 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$459$ $[ 0, 8 ]$ $2$ $8$ $12$ $[ -51 ]$ $[ 13 ]$ $[]$
$459$ $[ 6, 0 ]$ $3$ $4$ $5$ $[]$ $[]$ $[]$
$459$ $[ 6, 8 ]$ $6$ $8$ $12$ $[ 17 ]$ $[ 9 ]$ $[]$
$460$ $[ 0, 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$461$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$461$ $[ 46 ]$ $10$ $4$ $5$ $[ 461 ]$ $[ 5 ]$ $[]$
$462$ $[ 0, 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$462$ $[ 1, 3, 0 ]$ $2$ $4$ $5$ $[]$ $[]$ $[ 3, 7 ]$
$462$ $[ 1, 3, 0 ]$ $2$ $4$ $5$ $[]$ $[]$ $[ 3, 7 ]$
$462$ $[ 1, 3, 0 ]$ $2$ $8$ $17$ $[ -7 ]$ $[ 32 ]$ $[]$
$462$ $[ 0, 2, 0 ]$ $3$ $4$ $28$ $[]$ $[]$ $[]$
$462$ $[ 1, 1, 0 ]$ $6$ $8$ $8$ $[ -7 ]$ $[ 9 ]$ $[]$
$462$ $[ 1, 2, 5 ]$ $6$ $4$ $57$ $[ 33 ]$ $[ 4 ]$ $[]$
$462$ $[ 1, 2, 5 ]$ $6$ $4$ $57$ $[ 33 ]$ $[ 4 ]$ $[]$
$462$ $[ 1, 2, 5 ]$ $6$ $4$ $57$ $[ 33 ]$ $[ 4 ]$ $[]$
$462$ $[ 1, 2, 5 ]$ $6$ $4$ $57$ $[ 33 ]$ $[ 4 ]$ $[]$
$462$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$462$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$462$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$462$ $[ 0, 0, 2 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$464$ $[ 0, 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$464$ $[ 0, 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$464$ $[ 1, 0, 7 ]$ $4$ $8$ $5$ $[ 29, -116 ]$ $[ 5, 20 ]$ $[]$
$465$ $[ 0, 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$465$ $[ 0, 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$465$ $[ 0, 0, 10 ]$ $3$ $4$ $97$ $[]$ $[]$ $[]$
$465$ $[ 0, 0, 6 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$465$ $[ 0, 0, 6 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$466$ $[ 58 ]$ $4$ $4$ $12$ $[ 233 ]$ $[ 1 ]$ $[]$
$468$ $[ 1, 3, 0 ]$ $2$ $8$ $28$ $[ -3 ]$ $[ 4 ]$ $[]$
$468$ $[ 1, 0, 3 ]$ $4$ $8$ $8$ $[ -4 ]$ $[ 17 ]$ $[ 13, 13 ]$
$468$ $[ 1, 0, 3 ]$ $4$ $16$ $17$ $[ 156, -4 ]$ $[ 4, 8 ]$ $[]$
$468$ $[ 0, 0, 2 ]$ $6$ $4$ $129$ $[ 13 ]$ $[ 16 ]$ $[]$
$468$ $[ 1, 0, 1 ]$ $12$ $4$ $12$ $[ 13 ]$ $[ 1 ]$ $[]$
$468$ $[ 1, 0, 1 ]$ $12$ $4$ $12$ $[ 13 ]$ $[ 1 ]$ $[]$
$469$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$469$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$470$ $[ 0, 0 ]$ $1$ $2$ $21$ $[]$ $[]$ $[]$
$470$ $[ 2, 0 ]$ $2$ $4$ $5$ $[]$ $[]$ $[]$
$470$ $[ 2, 0 ]$ $2$ $4$ $5$ $[]$ $[]$ $[]$
$471$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$471$ $[ 0, 52 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$473$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$473$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$473$ $[ 5, 21 ]$ $2$ $4$ $13$ $[ 473 ]$ $[ 4 ]$ $[]$
$473$ $[ 5, 21 ]$ $2$ $4$ $13$ $[ 473 ]$ $[ 4 ]$ $[]$
$474$ $[ 0, 0 ]$ $1$ $2$ $29$ $[]$ $[]$ $[]$
$474$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$474$ $[ 0, 26 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$474$ $[ 0, 26 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$475$ $[ 5, 3 ]$ $12$ $16$ $5$ $[ -95 ]$ $[ 9 ]$ $[]$
$475$ $[ 5, 3 ]$ $12$ $16$ $29$ $[ -95 ]$ $[ 9 ]$ $[]$
$475$ $[ 5, 3 ]$ $12$ $16$ $5$ $[ -95 ]$ $[ 9 ]$ $[]$
$476$ $[ 0, 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$476$ $[ 0, 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$476$ $[ 0, 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$476$ $[ 0, 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$477$ $[ 0, 26 ]$ $2$ $4$ $8$ $[ 53 ]$ $[ 7 ]$ $[]$
$480$ $[ 1, 0, 1, 0 ]$ $2$ $4$ $8$ $[]$ $[]$ $[ 2, 3 ]$
$480$ $[ 1, 0, 1, 0 ]$ $2$ $4$ $8$ $[]$ $[]$ $[ 2, 3 ]$
$480$ $[ 1, 0, 1, 0 ]$ $2$ $4$ $8$ $[]$ $[]$ $[ 2, 3 ]$
$480$ $[ 1, 0, 1, 0 ]$ $2$ $4$ $8$ $[]$ $[]$ $[ 2, 3 ]$
$480$ $[ 1, 0, 0, 1 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 2, 5 ]$
$480$ $[ 1, 0, 0, 1 ]$ $4$ $4$ $8$ $[]$ $[]$ $[ 2, 5 ]$
$481$ $[ 4, 12 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$481$ $[ 8, 12 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$481$ $[ 4, 6 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$481$ $[ 8, 6 ]$ $6$ $4$ $12$ $[]$ $[]$ $[]$
$482$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$482$ $[ 120 ]$ $2$ $4$ $13$ $[ 241 ]$ $[ 27 ]$ $[]$
$482$ $[ 60 ]$ $4$ $4$ $12$ $[ 241 ]$ $[ 1 ]$ $[]$
$482$ $[ 20 ]$ $12$ $4$ $12$ $[ 241 ]$ $[ 4 ]$ $[]$
$482$ $[ 20 ]$ $12$ $4$ $12$ $[ 241 ]$ $[ 1 ]$ $[]$
$482$ $[ 20 ]$ $12$ $4$ $12$ $[ 241 ]$ $[ 1 ]$ $[]$
$483$ $[ 0, 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$483$ $[ 0, 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$483$ $[ 0, 0, 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$483$ $[ 0, 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$483$ $[ 0, 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$483$ $[ 0, 3, 11 ]$ $2$ $8$ $17$ $[ -23 ]$ $[ 8 ]$ $[]$
$483$ $[ 1, 3, 0 ]$ $2$ $4$ $5$ $[]$ $[]$ $[ 3, 7 ]$
$483$ $[ 1, 3, 0 ]$ $2$ $4$ $5$ $[]$ $[]$ $[ 3, 7 ]$
$483$ $[ 1, 3, 0 ]$ $2$ $8$ $8$ $[ -7 ]$ $[ 2 ]$ $[]$
$483$ $[ 0, 2, 0 ]$ $3$ $4$ $17$ $[]$ $[]$ $[]$
$484$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$484$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$484$ $[ 1, 55 ]$ $2$ $8$ $5$ $[ 44, -11 ]$ $[ 5, 20 ]$ $[]$
$484$ $[ 0, 22 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$484$ $[ 0, 22 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$484$ $[ 0, 22 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$484$ $[ 1, 11 ]$ $10$ $8$ $5$ $[ 44, -11 ]$ $[ 5, 20 ]$ $[]$
$484$ $[ 1, 11 ]$ $10$ $8$ $5$ $[ 44, -11 ]$ $[ 5, 20 ]$ $[]$
$484$ $[ 1, 11 ]$ $10$ $8$ $5$ $[ 44, -11 ]$ $[ 5, 20 ]$ $[]$
$484$ $[ 1, 11 ]$ $10$ $8$ $5$ $[ 44, -11 ]$ $[ 5, 20 ]$ $[]$
$485$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$485$ $[ 0, 0 ]$ $1$ $2$ $29$ $[]$ $[]$ $[]$
$487$ $[ 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$487$ $[ 0 ]$ $1$ $2$ $13$ $[]$ $[]$ $[]$
$488$ $[ 0, 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$488$ $[ 0, 0, 20 ]$ $3$ $4$ $65$ $[]$ $[]$ $[]$
$490$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$490$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$490$ $[ 2, 0 ]$ $2$ $4$ $24$ $[]$ $[]$ $[]$
$490$ $[ 0, 14 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$490$ $[ 0, 14 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$490$ $[ 1, 21 ]$ $4$ $8$ $8$ $[ -7 ]$ $[ 2 ]$ $[]$
$490$ $[ 2, 14 ]$ $6$ $8$ $24$ $[]$ $[]$ $[]$
$490$ $[ 2, 14 ]$ $6$ $8$ $24$ $[]$ $[]$ $[]$
$490$ $[ 1, 7 ]$ $12$ $16$ $8$ $[ -7 ]$ $[ 2 ]$ $[]$
$490$ $[ 1, 7 ]$ $12$ $16$ $8$ $[ -7 ]$ $[ 2 ]$ $[]$
$491$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$492$ $[ 0, 0, 0 ]$ $1$ $2$ $24$ $[]$ $[]$ $[]$
$492$ $[ 0, 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$492$ $[ 1, 1, 20 ]$ $2$ $16$ $17$ $[ -4, -164 ]$ $[ 2, 98 ]$ $[]$
$492$ $[ 0, 0, 8 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$492$ $[ 0, 0, 8 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$492$ $[ 0, 0, 8 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$492$ $[ 0, 0, 8 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$493$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$494$ $[ 0, 6 ]$ $3$ $4$ $24$ $[]$ $[]$ $[]$
$494$ $[ 0, 6 ]$ $3$ $4$ $24$ $[]$ $[]$ $[]$
$494$ $[ 0, 6 ]$ $3$ $4$ $12$ $[]$ $[]$ $[]$
$495$ $[ 0, 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$495$ $[ 0, 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$495$ $[ 0, 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$495$ $[ 0, 2, 0 ]$ $2$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$495$ $[ 0, 2, 0 ]$ $2$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$495$ $[ 0, 2, 0 ]$ $2$ $4$ $33$ $[ 5 ]$ $[ 4 ]$ $[]$
$495$ $[ 1, 1, 0 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$495$ $[ 1, 1, 0 ]$ $12$ $4$ $12$ $[ 5 ]$ $[ 1 ]$ $[]$
$496$ $[ 0, 0, 0 ]$ $1$ $2$ $33$ $[]$ $[]$ $[]$
$496$ $[ 0, 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$496$ $[ 0, 0, 0 ]$ $1$ $2$ $12$ $[]$ $[]$ $[]$
$496$ $[ 1, 0, 15 ]$ $2$ $8$ $21$ $[ -4, 124 ]$ $[ 4, 4 ]$ $[]$
$496$ $[ 0, 0, 10 ]$ $3$ $4$ $8$ $[]$ $[]$ $[]$
$496$ $[ 0, 1, 0 ]$ $4$ $4$ $12$ $[ 8 ]$ $[ 4 ]$ $[]$
$496$ $[ 0, 0, 6 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$496$ $[ 0, 0, 6 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$496$ $[ 0, 0, 6 ]$ $5$ $4$ $5$ $[]$ $[]$ $[]$
$497$ $[ 0, 0 ]$ $1$ $2$ $8$ $[]$ $[]$ $[]$
$497$ $[ 2, 14 ]$ $15$ $8$ $5$ $[]$ $[]$ $[]$
$498$ $[ 0, 0 ]$ $1$ $2$ $21$ $[]$ $[]$ $[]$
$498$ $[ 0, 0 ]$ $1$ $2$ $17$ $[]$ $[]$ $[]$
$498$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$499$ $[ 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$500$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$500$ $[ 0, 0 ]$ $1$ $2$ $5$ $[]$ $[]$ $[]$
$500$ $[ 0, 0 ]$ $1$ $4$ $5$ $[ 5 ]$ $[ 31 ]$ $[]$
$500$ $[ 0, 50 ]$ $2$ $4$ $5$ $[ 5 ]$ $[ 31 ]$ $[]$
$500$ $[ 0, 50 ]$ $2$ $4$ $5$ $[]$ $[]$ $[]$
3-dimensional RM-varieties
--------------------------
$N$ $\varepsilon$ $\operatorname{ord}(\varepsilon)$ $\dim(A_f)$ $F$ $\Disc(F)$ $[t_i]$ $[N_{F/\Q}(\delta_i)]$
------- ------------------ ----------------------------------- ------------- --------------------- ------------ ---------------- ------------------------
$19$ $[ 2 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$29$ $[ 4 ]$ $7$ $6$ $[ -5, 6, -1 ]$ $49$ $[]$ $[]$
$37$ $[ 4 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$37$ $[ 4 ]$ $9$ $6$ $[ -6, 9, -1 ]$ $81$ $[]$ $[]$
$38$ $[ 2 ]$ $9$ $6$ $[ -6, 9, -1 ]$ $81$ $[]$ $[]$
$41$ $[ 0 ]$ $1$ $3$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$41$ $[ 10 ]$ $4$ $6$ $[ -1, -5, 3 ]$ $564$ $[ 41 ]$ $[ 2 ]$
$43$ $[ 6 ]$ $7$ $6$ $[ -6, 5, -1 ]$ $49$ $[]$ $[]$
$43$ $[ 6 ]$ $7$ $6$ $[ -5, 6, -1 ]$ $49$ $[]$ $[]$
$45$ $[ 2, 0 ]$ $3$ $6$ $[ -1, -5, 3 ]$ $564$ $[]$ $[]$
$49$ $[ 6 ]$ $7$ $6$ $[ -6, 5, -1 ]$ $49$ $[]$ $[]$
$53$ $[ 0 ]$ $1$ $3$ $[ 1, -3, -1 ]$ $148$ $[]$ $[]$
$54$ $[ 2 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$56$ $[ 0, 1, 2 ]$ $6$ $12$ $[ 1, -5, 1 ]$ $404$ $[ 8 ]$ $[ 7 ]$
$57$ $[ 0, 6 ]$ $3$ $6$ $[ -1, -5, 3 ]$ $564$ $[]$ $[]$
$57$ $[ 0, 2 ]$ $9$ $6$ $[ -6, 9, -1 ]$ $81$ $[]$ $[]$
$58$ $[ 4 ]$ $7$ $6$ $[ -6, 5, -1 ]$ $49$ $[]$ $[]$
$60$ $[ 1, 0, 1 ]$ $4$ $12$ $[ 1, -6, 2 ]$ $568$ $[ -20 ]$ $[ 32 ]$
$61$ $[ 0 ]$ $1$ $3$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$63$ $[ 2, 0 ]$ $3$ $6$ $[ -1, -4, 1 ]$ $321$ $[]$ $[]$
$63$ $[ 2, 0 ]$ $3$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$65$ $[ 0, 6 ]$ $2$ $6$ $[ -1, -5, 3 ]$ $564$ $[]$ $[]$
$65$ $[ 2, 0 ]$ $2$ $6$ $[ 1, -3, -1 ]$ $148$ $[]$ $[]$
$65$ $[ 2, 4 ]$ $6$ $12$ $[ -8, 10, -1 ]$ $1765$ $[ 5 ]$ $[ 1 ]$
$71$ $[ 0 ]$ $1$ $3$ $[ 0, -5, 3 ]$ $257$ $[]$ $[]$
$71$ $[ 0 ]$ $1$ $3$ $[ 1, -4, -3 ]$ $257$ $[]$ $[]$
$74$ $[ 12 ]$ $3$ $6$ $[ 1, -5, 1 ]$ $404$ $[]$ $[]$
$74$ $[ 4 ]$ $9$ $6$ $[ -6, 9, -1 ]$ $81$ $[]$ $[]$
$77$ $[ 2, 0 ]$ $3$ $6$ $[ 1, -4, -3 ]$ $257$ $[]$ $[]$
$77$ $[ 2, 0 ]$ $3$ $6$ $[ -6, 9, -1 ]$ $81$ $[]$ $[]$
$78$ $[ 1, 3 ]$ $4$ $12$ $[ 0, -9, -6 ]$ $1944$ $[ -39 ]$ $[ 288 ]$
$84$ $[ 1, 1, 0 ]$ $2$ $12$ $[ -1, -8, 10 ]$ $892$ $[ -3 ]$ $[ 8 ]$
$85$ $[ 0, 8 ]$ $2$ $6$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$85$ $[ 2, 4 ]$ $4$ $12$ $[ -9, 19, -1 ]$ $1940$ $[ 85, 17 ]$ $[ 1, 2 ]$
$87$ $[ 0, 0 ]$ $1$ $3$ $[ -2, -4, 7 ]$ $229$ $[]$ $[]$
$89$ $[ 44 ]$ $2$ $6$ $[ -1, -3, 1 ]$ $148$ $[ 89 ]$ $[ 125 ]$
$91$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -4, 2 ]$ $316$ $[]$ $[]$
$91$ $[ 0, 6 ]$ $2$ $6$ $[ 1, -3, -1 ]$ $148$ $[]$ $[]$
$93$ $[ 0, 0 ]$ $1$ $3$ $[ 0, -4, 1 ]$ $229$ $[]$ $[]$
$93$ $[ 0, 10 ]$ $3$ $6$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$95$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$95$ $[ 2, 0 ]$ $2$ $6$ $[ -10, 27, -16 ]$ $1016$ $[ 5 ]$ $[ 16 ]$
$95$ $[ 0, 6 ]$ $3$ $6$ $[ -13, 50, -49 ]$ $361$ $[]$ $[]$
$95$ $[ 1, 9 ]$ $4$ $12$ $[ -9, 23, -17 ]$ $148$ $[ -19, -95 ]$ $[ 17, 200 ]$
$95$ $[ 2, 6 ]$ $6$ $12$ $[ -7, 6, -1 ]$ $257$ $[ 5 ]$ $[ 1 ]$
$97$ $[ 0 ]$ $1$ $3$ $[ 4, 3, -1 ]$ $49$ $[]$ $[]$
$97$ $[ 48 ]$ $2$ $6$ $[ -6, 9, -1 ]$ $81$ $[ 97 ]$ $[ 467 ]$
$99$ $[ 2, 0 ]$ $3$ $6$ $[ -6, 9, -1 ]$ $81$ $[]$ $[]$
$104$ $[ 0, 1, 0 ]$ $2$ $6$ $[ -1, -4, 2 ]$ $316$ $[]$ $[]$
$105$ $[ 0, 2, 0 ]$ $2$ $6$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$109$ $[ 0 ]$ $1$ $3$ $[ 2, -1, -1 ]$ $49$ $[]$ $[]$
$109$ $[ 54 ]$ $2$ $6$ $[ -1, -4, 1 ]$ $321$ $[ 109 ]$ $[ 29 ]$
$111$ $[ 0, 0 ]$ $1$ $3$ $[ -3, -1, 5 ]$ $148$ $[]$ $[]$
$111$ $[ 0, 12 ]$ $3$ $6$ $[ -8, 16, -1 ]$ $229$ $[]$ $[]$
$111$ $[ 0, 6 ]$ $6$ $6$ $[ -12, 36, -3 ]$ $2349$ $[ 37 ]$ $[ 3 ]$
$113$ $[ 0 ]$ $1$ $3$ $[ 2, -1, -1 ]$ $49$ $[]$ $[]$
$113$ $[ 0 ]$ $1$ $3$ $[ 2, -5, -9 ]$ $321$ $[]$ $[]$
$114$ $[ 0, 2 ]$ $9$ $6$ $[ 0, -3, 1 ]$ $81$ $[]$ $[]$
$114$ $[ 0, 2 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$114$ $[ 0, 2 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$114$ $[ 0, 2 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$116$ $[ 0, 4 ]$ $7$ $6$ $[ 1, -2, -1 ]$ $49$ $[]$ $[]$
$116$ $[ 0, 2 ]$ $14$ $6$ $[ -7, 14, -7 ]$ $49$ $[ 29 ]$ $[ 7 ]$
$116$ $[ 0, 2 ]$ $14$ $6$ $[ -7, 14, -7 ]$ $49$ $[ 29 ]$ $[ 7 ]$
$117$ $[ 3, 3 ]$ $4$ $12$ $[ -1, -5, 3 ]$ $564$ $[ -39 ]$ $[ 8 ]$
$119$ $[ 2, 0 ]$ $3$ $6$ $[ -5, 6, -1 ]$ $49$ $[]$ $[]$
$120$ $[ 0, 1, 0, 0 ]$ $2$ $6$ $[ -1, -4, 2 ]$ $316$ $[]$ $[]$
$120$ $[ 0, 1, 0, 2 ]$ $2$ $6$ $[ 0, -4, -1 ]$ $229$ $[ 40 ]$ $[ 64 ]$
$120$ $[ 0, 1, 0, 2 ]$ $2$ $6$ $[ 0, -4, -1 ]$ $229$ $[ 40 ]$ $[ 64 ]$
$122$ $[ 0 ]$ $1$ $3$ $[ 1, -5, 2 ]$ $229$ $[]$ $[]$
$122$ $[ 20 ]$ $3$ $6$ $[ -11, 33, -16 ]$ $469$ $[]$ $[]$
$123$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -4, 2 ]$ $316$ $[]$ $[]$
$124$ $[ 0, 10 ]$ $3$ $6$ $[ -1, -5, 3 ]$ $564$ $[]$ $[]$
$126$ $[ 2, 2 ]$ $3$ $6$ $[ 1, -4, -1 ]$ $321$ $[]$ $[]$
$126$ $[ 2, 2 ]$ $3$ $6$ $[ 1, -4, -1 ]$ $321$ $[]$ $[]$
$126$ $[ 4, 2 ]$ $3$ $6$ $[ 1, -4, -1 ]$ $321$ $[]$ $[]$
$126$ $[ 4, 2 ]$ $3$ $6$ $[ 1, -4, -1 ]$ $321$ $[]$ $[]$
$127$ $[ 0 ]$ $1$ $3$ $[ 3, 0, -3 ]$ $81$ $[]$ $[]$
$129$ $[ 0, 0 ]$ $1$ $3$ $[ 2, -5, -8 ]$ $568$ $[]$ $[]$
$129$ $[ 0, 14 ]$ $3$ $6$ $[ -5, 6, -1 ]$ $49$ $[]$ $[]$
$129$ $[ 0, 14 ]$ $3$ $6$ $[ -9, 22, -9 ]$ $257$ $[]$ $[]$
$129$ $[ 0, 6 ]$ $7$ $6$ $[ -5, 6, -1 ]$ $49$ $[]$ $[]$
$130$ $[ 2, 0 ]$ $2$ $6$ $[ 0, -7, -4 ]$ $940$ $[]$ $[]$
$130$ $[ 2, 4 ]$ $6$ $12$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$132$ $[ 1, 0, 5 ]$ $2$ $12$ $[ -1, -8, 10 ]$ $892$ $[ -11 ]$ $[ 8 ]$
$133$ $[ 0, 0 ]$ $1$ $3$ $[ -2, -4, 7 ]$ $229$ $[]$ $[]$
$133$ $[ 0, 6 ]$ $3$ $6$ $[ 1, -4, -1 ]$ $321$ $[]$ $[]$
$133$ $[ 1, 9 ]$ $6$ $6$ $[ 1, -7, -1 ]$ $1524$ $[ 133 ]$ $[ 3 ]$
$133$ $[ 1, 9 ]$ $6$ $6$ $[ 1, -7, -1 ]$ $1524$ $[ 133 ]$ $[ 3 ]$
$134$ $[ 0 ]$ $1$ $3$ $[ -1, -8, 11 ]$ $473$ $[]$ $[]$
$134$ $[ 0 ]$ $1$ $3$ $[ -3, 0, 1 ]$ $81$ $[]$ $[]$
$135$ $[ 6, 0 ]$ $3$ $6$ $[ -1, -5, 3 ]$ $564$ $[]$ $[]$
$139$ $[ 0 ]$ $1$ $3$ $[ 2, -1, -1 ]$ $49$ $[]$ $[]$
$142$ $[ 10 ]$ $7$ $6$ $[ -6, 5, -1 ]$ $49$ $[]$ $[]$
$143$ $[ 0, 4 ]$ $3$ $6$ $[ -13, 50, -49 ]$ $361$ $[]$ $[]$
$143$ $[ 0, 4 ]$ $3$ $6$ $[ 0, -6, -2 ]$ $756$ $[]$ $[]$
$145$ $[ 0, 0 ]$ $1$ $3$ $[ -3, -1, 5 ]$ $148$ $[]$ $[]$
$145$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$145$ $[ 0, 14 ]$ $2$ $6$ $[ 1, -6, -2 ]$ $1016$ $[ 29 ]$ $[ 1 ]$
$145$ $[ 2, 0 ]$ $2$ $6$ $[ 1, -14, 14 ]$ $2296$ $[ 5 ]$ $[ 1 ]$
$146$ $[ 0 ]$ $1$ $3$ $[ 1, -5, 1 ]$ $404$ $[]$ $[]$
$146$ $[ 24 ]$ $3$ $6$ $[ -10, 25, -1 ]$ $473$ $[]$ $[]$
$146$ $[ 18 ]$ $4$ $6$ $[ 1, -5, -3 ]$ $564$ $[ 73 ]$ $[ 2 ]$
$146$ $[ 18 ]$ $4$ $6$ $[ 0, -8, -6 ]$ $1076$ $[ 73 ]$ $[ 18 ]$
$146$ $[ 12 ]$ $6$ $6$ $[ 0, -9, -7 ]$ $1593$ $[ 73 ]$ $[ 3 ]$
$148$ $[ 0, 12 ]$ $3$ $6$ $[ -1, -6, 2 ]$ $1016$ $[]$ $[]$
$148$ $[ 0, 2 ]$ $18$ $6$ $[ 0, -3, -1 ]$ $81$ $[ 37 ]$ $[ 27 ]$
$149$ $[ 0 ]$ $1$ $3$ $[ 1, -2, -1 ]$ $49$ $[]$ $[]$
$151$ $[ 0 ]$ $1$ $3$ $[ 2, -1, -1 ]$ $49$ $[]$ $[]$
$151$ $[ 0 ]$ $1$ $3$ $[ 0, -5, 3 ]$ $257$ $[]$ $[]$
$152$ $[ 0, 0, 0 ]$ $1$ $3$ $[ -1, -10, 8 ]$ $961$ $[]$ $[]$
$152$ $[ 0, 0, 6 ]$ $3$ $6$ $[ -1, -4, 2 ]$ $316$ $[]$ $[]$
$152$ $[ 0, 0, 2 ]$ $9$ $6$ $[ -6, 9, -1 ]$ $81$ $[]$ $[]$
$152$ $[ 0, 0, 2 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$156$ $[ 1, 1, 0 ]$ $2$ $12$ $[ -13, 22, -8 ]$ $2089$ $[ -3, -4 ]$ $[ 8, 8 ]$
$161$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$161$ $[ 3, 11 ]$ $2$ $12$ $[ -8, 16, -1 ]$ $229$ $[ -23, 161 ]$ $[ 112, 116 ]$
$162$ $[ 6 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$164$ $[ 0, 10 ]$ $4$ $6$ $[ -10, 28, -18 ]$ $564$ $[ 41 ]$ $[ 2 ]$
$165$ $[ 0, 0, 0 ]$ $1$ $3$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$165$ $[ 0, 2, 0 ]$ $2$ $6$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$165$ $[ 0, 2, 0 ]$ $2$ $6$ $[ 1, -3, -1 ]$ $148$ $[]$ $[]$
$166$ $[ 0 ]$ $1$ $3$ $[ 0, -4, 1 ]$ $229$ $[]$ $[]$
$168$ $[ 1, 1, 1, 0 ]$ $2$ $12$ $[ 1, -8, -10 ]$ $892$ $[ -3 ]$ $[ 32 ]$
$169$ $[ 0 ]$ $1$ $3$ $[ 2, -1, -1 ]$ $49$ $[]$ $[]$
$169$ $[ 0 ]$ $1$ $3$ $[ -2, -1, 1 ]$ $49$ $[]$ $[]$
$169$ $[ 78 ]$ $2$ $6$ $[ -6, 5, -1 ]$ $49$ $[]$ $[]$
$169$ $[ 52 ]$ $3$ $6$ $[ -6, 5, -1 ]$ $49$ $[]$ $[]$
$169$ $[ 52 ]$ $3$ $6$ $[ -6, 5, -1 ]$ $49$ $[]$ $[]$
$169$ $[ 26 ]$ $6$ $12$ $[ -6, 5, -1 ]$ $49$ $[]$ $[]$
$170$ $[ 2, 0 ]$ $2$ $6$ $[ 1, -6, 2 ]$ $568$ $[]$ $[]$
$171$ $[ 0, 6 ]$ $3$ $6$ $[ -1, -5, 3 ]$ $564$ $[]$ $[]$
$171$ $[ 0, 2 ]$ $9$ $6$ $[ -6, 9, -1 ]$ $81$ $[]$ $[]$
$171$ $[ 0, 2 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$172$ $[ 0, 6 ]$ $7$ $6$ $[ -17, 94, -169 ]$ $49$ $[]$ $[]$
$174$ $[ 0, 4 ]$ $7$ $6$ $[ 1, -2, -1 ]$ $49$ $[]$ $[]$
$174$ $[ 0, 4 ]$ $7$ $6$ $[ -17, 94, -169 ]$ $49$ $[]$ $[]$
$175$ $[ 0, 2 ]$ $3$ $6$ $[ -9, 22, -9 ]$ $257$ $[]$ $[]$
$175$ $[ 0, 2 ]$ $3$ $6$ $[ -9, 22, -9 ]$ $257$ $[]$ $[]$
$175$ $[ 10, 2 ]$ $6$ $12$ $[ -9, 22, -9 ]$ $257$ $[]$ $[]$
$175$ $[ 5, 1 ]$ $12$ $24$ $[ -11, 34, -27 ]$ $993$ $[ -35 ]$ $[ 27 ]$
$177$ $[ 0, 0 ]$ $1$ $3$ $[ 0, -4, -1 ]$ $229$ $[]$ $[]$
$177$ $[ 1, 29 ]$ $2$ $6$ $[ -12, 44, -49 ]$ $229$ $[ 177 ]$ $[ 23 ]$
$177$ $[ 1, 29 ]$ $2$ $6$ $[ -12, 44, -49 ]$ $229$ $[ 177 ]$ $[ 23 ]$
$178$ $[ 0 ]$ $1$ $3$ $[ 1, -6, 2 ]$ $568$ $[]$ $[]$
$178$ $[ 22 ]$ $4$ $6$ $[ -18, 97, -162 ]$ $1436$ $[ 89 ]$ $[ 8 ]$
$179$ $[ 0 ]$ $1$ $3$ $[ 1, -2, -1 ]$ $49$ $[]$ $[]$
$180$ $[ 0, 2, 0 ]$ $3$ $6$ $[ 1, -5, -3 ]$ $564$ $[]$ $[]$
$180$ $[ 1, 0, 1 ]$ $4$ $12$ $[ 1, -6, 2 ]$ $568$ $[ -20 ]$ $[ 32 ]$
$180$ $[ 0, 2, 2 ]$ $6$ $12$ $[ -1, -22, 46 ]$ $4344$ $[ 5 ]$ $[ 64 ]$
$182$ $[ 0, 6 ]$ $2$ $6$ $[ 1, -10, -14 ]$ $1384$ $[]$ $[]$
$182$ $[ 2, 0 ]$ $3$ $6$ $[ -1, -4, 1 ]$ $321$ $[]$ $[]$
$182$ $[ 2, 4 ]$ $3$ $6$ $[ -17, 82, -81 ]$ $1257$ $[]$ $[]$
$182$ $[ 4, 4 ]$ $3$ $6$ $[ -17, 82, -81 ]$ $1257$ $[]$ $[]$
$182$ $[ 2, 6 ]$ $6$ $12$ $[ -6, 5, -1 ]$ $49$ $[]$ $[]$
$183$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$183$ $[ 0, 30 ]$ $2$ $6$ $[ 1, -30, -74 ]$ $1304$ $[ 61 ]$ $[ 3 ]$
$183$ $[ 0, 20 ]$ $3$ $6$ $[ -11, 33, -16 ]$ $469$ $[]$ $[]$
$183$ $[ 0, 20 ]$ $3$ $6$ $[ -16, 80, -121 ]$ $469$ $[]$ $[]$
$184$ $[ 1, 1, 11 ]$ $2$ $12$ $[ -1, -3, 1 ]$ $148$ $[ -8 ]$ $[ 23 ]$
$185$ $[ 1, 9 ]$ $4$ $6$ $[ -1, -3, 1 ]$ $148$ $[ 185 ]$ $[ 2 ]$
$185$ $[ 3, 9 ]$ $4$ $6$ $[ -1, -3, 1 ]$ $148$ $[ 185 ]$ $[ 2 ]$
$187$ $[ 0, 0 ]$ $1$ $3$ $[ 2, -2, -2 ]$ $148$ $[]$ $[]$
$189$ $[ 0, 2 ]$ $3$ $6$ $[ 1, -4, -1 ]$ $321$ $[]$ $[]$
$189$ $[ 0, 2 ]$ $3$ $6$ $[ 1, -4, -1 ]$ $321$ $[]$ $[]$
$189$ $[ 6, 0 ]$ $3$ $6$ $[ -1, -4, 1 ]$ $321$ $[]$ $[]$
$189$ $[ 6, 0 ]$ $3$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$189$ $[ 9, 1 ]$ $6$ $12$ $[ 1, -4, -3 ]$ $257$ $[ -7 ]$ $[ 243 ]$
$190$ $[ 2, 0 ]$ $2$ $6$ $[ 1, -6, 2 ]$ $568$ $[]$ $[]$
$190$ $[ 0, 2 ]$ $9$ $6$ $[ -6, 9, -1 ]$ $81$ $[]$ $[]$
$194$ $[ 48 ]$ $2$ $6$ $[ -1, -7, 4 ]$ $1509$ $[ 97 ]$ $[ 24 ]$
$194$ $[ 32 ]$ $3$ $6$ $[ 1, -3, -1 ]$ $148$ $[]$ $[]$
$195$ $[ 0, 0, 0 ]$ $1$ $3$ $[ -1, -4, 2 ]$ $316$ $[]$ $[]$
$195$ $[ 0, 2, 6 ]$ $2$ $12$ $[ -1, -8, 10 ]$ $892$ $[ 13 ]$ $[ 1 ]$
$195$ $[ 0, 0, 4 ]$ $3$ $6$ $[ 0, -6, -2 ]$ $756$ $[]$ $[]$
$195$ $[ 0, 0, 4 ]$ $3$ $6$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$197$ $[ 14 ]$ $14$ $6$ $[ -7, 14, -7 ]$ $49$ $[ 197 ]$ $[ 7 ]$
$198$ $[ 2, 0 ]$ $3$ $6$ $[ -1, -5, 3 ]$ $564$ $[]$ $[]$
$198$ $[ 2, 0 ]$ $3$ $6$ $[ 1, -5, -3 ]$ $564$ $[]$ $[]$
$199$ $[ 66 ]$ $3$ $6$ $[ -1, -8, 5 ]$ $2177$ $[]$ $[]$
$201$ $[ 0, 0 ]$ $1$ $3$ $[ -3, -1, 5 ]$ $148$ $[]$ $[]$
$202$ $[ 0 ]$ $1$ $3$ $[ 3, 0, -1 ]$ $81$ $[]$ $[]$
$203$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -3, -1 ]$ $148$ $[]$ $[]$
$203$ $[ 0, 14 ]$ $2$ $6$ $[ -10, 27, -16 ]$ $1016$ $[ 29 ]$ $[ 1 ]$
$203$ $[ 0, 14 ]$ $2$ $6$ $[ -1, -6, -2 ]$ $568$ $[ 29 ]$ $[ 5 ]$
$205$ $[ 0, 0 ]$ $1$ $3$ $[ -2, -4, 7 ]$ $229$ $[]$ $[]$
$205$ $[ 0, 0 ]$ $1$ $3$ $[ 0, -4, -1 ]$ $229$ $[]$ $[]$
$205$ $[ 0, 20 ]$ $2$ $6$ $[ -8, 16, -1 ]$ $229$ $[ 41 ]$ $[ 4 ]$
$205$ $[ 0, 20 ]$ $2$ $6$ $[ 0, -6, -3 ]$ $621$ $[ 41 ]$ $[ 32 ]$
$205$ $[ 2, 0 ]$ $2$ $6$ $[ -6, 8, -1 ]$ $229$ $[ 5 ]$ $[ 1 ]$
$206$ $[ 34 ]$ $3$ $6$ $[ -13, 46, -49 ]$ $257$ $[]$ $[]$
$209$ $[ 0, 2 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$210$ $[ 0, 2, 2 ]$ $6$ $12$ $[ 0, -15, -20 ]$ $2700$ $[]$ $[]$
$211$ $[ 0 ]$ $1$ $3$ $[ 0, -4, 1 ]$ $229$ $[]$ $[]$
$211$ $[ 0 ]$ $1$ $3$ $[ 2, -1, -1 ]$ $49$ $[]$ $[]$
$212$ $[ 0, 0 ]$ $1$ $3$ $[ 3, -3, -7 ]$ $756$ $[]$ $[]$
$215$ $[ 0, 0 ]$ $1$ $3$ $[ 2, -3, -3 ]$ $321$ $[]$ $[]$
$215$ $[ 2, 0 ]$ $2$ $6$ $[ -1, -4, -1 ]$ $169$ $[]$ $[]$
$217$ $[ 0, 0 ]$ $1$ $3$ $[ 3, 0, -3 ]$ $81$ $[]$ $[]$
$217$ $[ 0, 0 ]$ $1$ $3$ $[ 3, 0, -3 ]$ $81$ $[]$ $[]$
$217$ $[ 3, 15 ]$ $2$ $12$ $[ -5, 6, -1 ]$ $49$ $[ 217, -31 ]$ $[ 83, 97 ]$
$218$ $[ 0 ]$ $1$ $3$ $[ -3, -3, 8 ]$ $621$ $[]$ $[]$
$218$ $[ 12 ]$ $9$ $6$ $[ -6, 9, -1 ]$ $81$ $[]$ $[]$
$219$ $[ 0, 36 ]$ $2$ $6$ $[ -1, -4, 2 ]$ $316$ $[ 73 ]$ $[ 8 ]$
$219$ $[ 0, 12 ]$ $6$ $6$ $[ 0, -6, -3 ]$ $621$ $[ 73 ]$ $[ 12 ]$
$219$ $[ 0, 12 ]$ $6$ $6$ $[ -14, 61, -81 ]$ $321$ $[ 73 ]$ $[ 27 ]$
$220$ $[ 1, 0, 5 ]$ $2$ $12$ $[ -13, 42, -8 ]$ $2089$ $[ -4, -11 ]$ $[ 8, 32 ]$
$221$ $[ 0, 0 ]$ $1$ $3$ $[ 0, -4, 1 ]$ $229$ $[]$ $[]$
$221$ $[ 0, 8 ]$ $2$ $6$ $[ 0, -6, -3 ]$ $621$ $[ 17 ]$ $[ 2 ]$
$221$ $[ 0, 8 ]$ $2$ $6$ $[ -8, 16, -1 ]$ $229$ $[ 17 ]$ $[ 4 ]$
$221$ $[ 6, 8 ]$ $2$ $6$ $[ -12, 38, -13 ]$ $733$ $[ 221 ]$ $[ 13 ]$
$221$ $[ 6, 8 ]$ $2$ $6$ $[ -12, 38, -13 ]$ $733$ $[ 221 ]$ $[ 13 ]$
$222$ $[ 0, 4 ]$ $9$ $6$ $[ 0, -3, 1 ]$ $81$ $[]$ $[]$
$222$ $[ 0, 4 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$222$ $[ 0, 4 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$222$ $[ 0, 4 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$224$ $[ 0, 4, 2 ]$ $6$ $12$ $[ 1, -5, 1 ]$ $404$ $[ 8 ]$ $[ 7 ]$
$225$ $[ 2, 0 ]$ $3$ $6$ $[ -1, -5, 3 ]$ $564$ $[]$ $[]$
$225$ $[ 2, 10 ]$ $6$ $12$ $[ -1, -5, 3 ]$ $564$ $[]$ $[]$
$226$ $[ 28 ]$ $4$ $6$ $[ 1, -2, -1 ]$ $49$ $[ 113 ]$ $[ 8 ]$
$226$ $[ 16 ]$ $7$ $6$ $[ -6, 5, -1 ]$ $49$ $[]$ $[]$
$227$ $[ 0 ]$ $1$ $3$ $[ 2, -1, -1 ]$ $49$ $[]$ $[]$
$228$ $[ 0, 0, 2 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$230$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -9, 12 ]$ $1101$ $[]$ $[]$
$231$ $[ 0, 0, 0 ]$ $1$ $3$ $[ 0, -6, -1 ]$ $837$ $[]$ $[]$
$231$ $[ 0, 0, 0 ]$ $1$ $3$ $[ -2, -4, 7 ]$ $229$ $[]$ $[]$
$232$ $[ 0, 0, 0 ]$ $1$ $3$ $[ -2, -5, 8 ]$ $568$ $[]$ $[]$
$232$ $[ 0, 0, 4 ]$ $7$ $6$ $[ -17, 94, -169 ]$ $49$ $[]$ $[]$
$232$ $[ 0, 0, 4 ]$ $7$ $6$ $[ -6, 5, -1 ]$ $49$ $[]$ $[]$
$232$ $[ 0, 0, 4 ]$ $7$ $6$ $[ -5, 6, -1 ]$ $49$ $[]$ $[]$
$234$ $[ 2, 0 ]$ $3$ $6$ $[ -1, -4, 1 ]$ $321$ $[]$ $[]$
$236$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -4, 1 ]$ $321$ $[]$ $[]$
$238$ $[ 0, 8 ]$ $2$ $6$ $[ 1, -3, -1 ]$ $148$ $[]$ $[]$
$238$ $[ 2, 0 ]$ $3$ $6$ $[ 1, -5, 1 ]$ $404$ $[]$ $[]$
$238$ $[ 2, 8 ]$ $6$ $12$ $[ -1, -14, -14 ]$ $2296$ $[ 17 ]$ $[ 4 ]$
$239$ $[ 0 ]$ $1$ $3$ $[ 1, -2, -1 ]$ $49$ $[]$ $[]$
$239$ $[ 34 ]$ $7$ $6$ $[ -5, 6, -1 ]$ $49$ $[]$ $[]$
$240$ $[ 1, 1, 0, 1 ]$ $4$ $6$ $[ -1, -4, 2 ]$ $316$ $[ 40 ]$ $[ 8 ]$
$240$ $[ 1, 3, 0, 1 ]$ $4$ $6$ $[ -1, -4, 2 ]$ $316$ $[ 40 ]$ $[ 8 ]$
$243$ $[ 0 ]$ $1$ $3$ $[ 3, 0, -3 ]$ $81$ $[]$ $[]$
$243$ $[ 0 ]$ $1$ $3$ $[ -3, 0, 3 ]$ $81$ $[]$ $[]$
$243$ $[ 54 ]$ $3$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$243$ $[ 54 ]$ $3$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$245$ $[ 2, 6 ]$ $14$ $12$ $[ -17, 94, -169 ]$ $49$ $[]$ $[]$
$247$ $[ 0, 0 ]$ $1$ $3$ $[ 3, 0, -3 ]$ $81$ $[]$ $[]$
$248$ $[ 0, 0, 0 ]$ $1$ $3$ $[ 1, -4, -2 ]$ $316$ $[]$ $[]$
$248$ $[ 0, 0, 10 ]$ $3$ $6$ $[ 1, -3, -1 ]$ $148$ $[]$ $[]$
$252$ $[ 1, 3, 0 ]$ $2$ $12$ $[ -1, -4, 2 ]$ $316$ $[ -3 ]$ $[ 8 ]$
$252$ $[ 0, 2, 0 ]$ $3$ $6$ $[ 1, -4, -1 ]$ $321$ $[]$ $[]$
$252$ $[ 0, 2, 0 ]$ $3$ $6$ $[ -1, -4, 1 ]$ $321$ $[]$ $[]$
$253$ $[ 0, 0 ]$ $1$ $3$ $[ -3, 0, 3 ]$ $81$ $[]$ $[]$
$253$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -4, 1 ]$ $169$ $[]$ $[]$
$254$ $[ 14 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$254$ $[ 14 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$255$ $[ 0, 0, 0 ]$ $1$ $3$ $[ 0, -4, 1 ]$ $229$ $[]$ $[]$
$258$ $[ 1, 21 ]$ $2$ $6$ $[ 0, -6, 1 ]$ $837$ $[ 129 ]$ $[ 2 ]$
$258$ $[ 1, 21 ]$ $2$ $6$ $[ 0, -6, 1 ]$ $837$ $[ 129 ]$ $[ 2 ]$
$258$ $[ 0, 6 ]$ $7$ $6$ $[ 1, -2, -1 ]$ $49$ $[]$ $[]$
$258$ $[ 0, 6 ]$ $7$ $6$ $[ 1, -2, -1 ]$ $49$ $[]$ $[]$
$258$ $[ 0, 6 ]$ $7$ $6$ $[ -5, 6, -1 ]$ $49$ $[]$ $[]$
$258$ $[ 0, 6 ]$ $7$ $6$ $[ -5, 6, -1 ]$ $49$ $[]$ $[]$
$259$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -2, 1 ]$ $49$ $[]$ $[]$
$259$ $[ 0, 0 ]$ $1$ $3$ $[ 3, 0, -3 ]$ $81$ $[]$ $[]$
$260$ $[ 0, 0, 0 ]$ $1$ $3$ $[ -1, -5, 3 ]$ $564$ $[]$ $[]$
$260$ $[ 0, 0, 6 ]$ $2$ $6$ $[ 0, -6, -2 ]$ $756$ $[]$ $[]$
$260$ $[ 0, 2, 0 ]$ $2$ $6$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$260$ $[ 0, 2, 4 ]$ $6$ $12$ $[ 0, -4, -1 ]$ $229$ $[ 5 ]$ $[ 1 ]$
$261$ $[ 0, 0 ]$ $1$ $3$ $[ 2, -4, -7 ]$ $229$ $[]$ $[]$
$261$ $[ 0, 4 ]$ $7$ $6$ $[ -5, 6, -1 ]$ $49$ $[]$ $[]$
$264$ $[ 0, 0, 1, 5 ]$ $2$ $6$ $[ -13, 46, -32 ]$ $961$ $[ 33 ]$ $[ 32 ]$
$264$ $[ 0, 0, 1, 5 ]$ $2$ $6$ $[ -13, 46, -32 ]$ $961$ $[ 33 ]$ $[ 32 ]$
$264$ $[ 0, 1, 0, 0 ]$ $2$ $6$ $[ 1, -4, -2 ]$ $316$ $[]$ $[]$
$264$ $[ 1, 1, 0, 5 ]$ $2$ $12$ $[ 1, -10, -8 ]$ $961$ $[ -8, 88 ]$ $[ 16, 32 ]$
$266$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -7, 4 ]$ $469$ $[]$ $[]$
$266$ $[ 0, 6 ]$ $3$ $6$ $[ 1, -3, -1 ]$ $148$ $[]$ $[]$
$266$ $[ 2, 0 ]$ $3$ $6$ $[ 1, -4, -3 ]$ $257$ $[]$ $[]$
$266$ $[ 0, 2 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$267$ $[ 0, 0 ]$ $1$ $3$ $[ 0, -3, 1 ]$ $81$ $[]$ $[]$
$267$ $[ 0, 0 ]$ $1$ $3$ $[ -2, -3, 5 ]$ $169$ $[]$ $[]$
$267$ $[ 0, 0 ]$ $1$ $3$ $[ 4, 3, -1 ]$ $49$ $[]$ $[]$
$270$ $[ 2, 0 ]$ $9$ $6$ $[ 0, -3, 1 ]$ $81$ $[]$ $[]$
$273$ $[ 0, 0, 0 ]$ $1$ $3$ $[ 2, -3, -2 ]$ $316$ $[]$ $[]$
$273$ $[ 0, 0, 6 ]$ $2$ $6$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$273$ $[ 0, 0, 4 ]$ $3$ $6$ $[ -10, 25, -1 ]$ $473$ $[]$ $[]$
$273$ $[ 0, 0, 4 ]$ $3$ $6$ $[ -6, 9, -1 ]$ $81$ $[]$ $[]$
$273$ $[ 0, 0, 4 ]$ $3$ $6$ $[ -10, 29, -25 ]$ $169$ $[]$ $[]$
$273$ $[ 0, 0, 4 ]$ $3$ $6$ $[ -6, 5, -1 ]$ $49$ $[]$ $[]$
$273$ $[ 0, 2, 0 ]$ $3$ $6$ $[ -6, 9, -1 ]$ $81$ $[]$ $[]$
$273$ $[ 0, 2, 0 ]$ $3$ $6$ $[ -6, 5, -1 ]$ $49$ $[]$ $[]$
$274$ $[ 0 ]$ $1$ $3$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$274$ $[ 68 ]$ $2$ $6$ $[ 1, -6, -2 ]$ $1016$ $[ 137 ]$ $[ 16 ]$
$277$ $[ 0 ]$ $1$ $3$ $[ 1, -3, -1 ]$ $148$ $[]$ $[]$
$278$ $[ 0 ]$ $1$ $3$ $[ -3, 0, 3 ]$ $81$ $[]$ $[]$
$279$ $[ 0, 0 ]$ $1$ $3$ $[ 0, -4, -1 ]$ $229$ $[]$ $[]$
$279$ $[ 0, 0 ]$ $1$ $6$ $[ -12, 40, -27 ]$ $1373$ $[ -3 ]$ $[ 12 ]$
$279$ $[ 0, 10 ]$ $3$ $6$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$280$ $[ 0, 0, 2, 0 ]$ $2$ $6$ $[ -1, -6, -2 ]$ $568$ $[]$ $[]$
$280$ $[ 0, 0, 0, 2 ]$ $3$ $6$ $[ 0, -9, -6 ]$ $1944$ $[]$ $[]$
$282$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -3, -1 ]$ $148$ $[]$ $[]$
$284$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -4, 1 ]$ $321$ $[]$ $[]$
$284$ $[ 0, 0 ]$ $1$ $3$ $[ 3, 0, -3 ]$ $81$ $[]$ $[]$
$284$ $[ 1, 5 ]$ $14$ $6$ $[ 1, -2, -1 ]$ $49$ $[ 284 ]$ $[ 7 ]$
$284$ $[ 1, 5 ]$ $14$ $6$ $[ 1, -2, -1 ]$ $49$ $[ 284 ]$ $[ 7 ]$
$285$ $[ 0, 2, 0 ]$ $2$ $6$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$286$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -10, 8 ]$ $961$ $[]$ $[]$
$286$ $[ 0, 4 ]$ $3$ $6$ $[ 1, -5, -3 ]$ $564$ $[]$ $[]$
$286$ $[ 0, 4 ]$ $3$ $6$ $[ 1, -3, -1 ]$ $148$ $[]$ $[]$
$287$ $[ 0, 0 ]$ $1$ $3$ $[ -4, 3, 1 ]$ $49$ $[]$ $[]$
$287$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -4, 3 ]$ $257$ $[]$ $[]$
$289$ $[ 0 ]$ $1$ $3$ $[ 0, -3, 1 ]$ $81$ $[]$ $[]$
$289$ $[ 0 ]$ $1$ $3$ $[ 0, -3, 1 ]$ $81$ $[]$ $[]$
$289$ $[ 136 ]$ $2$ $6$ $[ -6, 9, -1 ]$ $81$ $[]$ $[]$
$289$ $[ 68 ]$ $4$ $12$ $[ -6, 9, -1 ]$ $81$ $[]$ $[]$
$289$ $[ 34 ]$ $8$ $24$ $[ -6, 9, -1 ]$ $81$ $[]$ $[]$
$290$ $[ 0, 0 ]$ $1$ $3$ $[ -3, -3, 8 ]$ $621$ $[]$ $[]$
$290$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -7, 4 ]$ $469$ $[]$ $[]$
$291$ $[ 0, 48 ]$ $2$ $6$ $[ -11, 33, -16 ]$ $469$ $[]$ $[]$
$291$ $[ 0, 48 ]$ $2$ $6$ $[ -16, 80, -121 ]$ $469$ $[]$ $[]$
$292$ $[ 0, 36 ]$ $2$ $6$ $[ -1, -7, 4 ]$ $1509$ $[ 73 ]$ $[ 6 ]$
$294$ $[ 0, 6 ]$ $7$ $6$ $[ -5, 6, -1 ]$ $49$ $[]$ $[]$
$295$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -2, -1 ]$ $49$ $[]$ $[]$
$295$ $[ 0, 0 ]$ $1$ $3$ $[ 3, 0, -3 ]$ $81$ $[]$ $[]$
$295$ $[ 2, 0 ]$ $2$ $6$ $[ 1, -6, -2 ]$ $1016$ $[ 5 ]$ $[ 1 ]$
$295$ $[ 2, 0 ]$ $2$ $6$ $[ -7, 12, -1 ]$ $257$ $[ 5 ]$ $[ 1 ]$
$296$ $[ 0, 0, 0 ]$ $1$ $3$ $[ -2, -4, 7 ]$ $229$ $[]$ $[]$
$296$ $[ 0, 0, 12 ]$ $3$ $6$ $[ 1, -3, -1 ]$ $148$ $[]$ $[]$
$297$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -5, -3 ]$ $564$ $[]$ $[]$
$297$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -5, 3 ]$ $564$ $[]$ $[]$
$297$ $[ 6, 0 ]$ $3$ $6$ $[ -6, 9, -1 ]$ $81$ $[]$ $[]$
$297$ $[ 2, 0 ]$ $9$ $6$ $[ -6, 9, -1 ]$ $81$ $[]$ $[]$
$298$ $[ 0 ]$ $1$ $3$ $[ 5, 4, -5 ]$ $169$ $[]$ $[]$
$299$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -7, 3 ]$ $788$ $[]$ $[]$
$300$ $[ 1, 0, 5 ]$ $4$ $12$ $[ 1, -6, 2 ]$ $568$ $[ -20 ]$ $[ 32 ]$
$301$ $[ 0, 6 ]$ $7$ $6$ $[ -5, 6, -1 ]$ $49$ $[]$ $[]$
$304$ $[ 0, 0, 0 ]$ $1$ $3$ $[ 1, -10, -8 ]$ $961$ $[]$ $[]$
$304$ $[ 0, 0, 6 ]$ $3$ $6$ $[ -1, -4, 2 ]$ $316$ $[]$ $[]$
$304$ $[ 1, 0, 3 ]$ $6$ $6$ $[ -1, -9, 3 ]$ $3252$ $[ 76 ]$ $[ 3 ]$
$304$ $[ 1, 0, 3 ]$ $6$ $6$ $[ -1, -9, 3 ]$ $3252$ $[ 76 ]$ $[ 3 ]$
$304$ $[ 0, 0, 2 ]$ $9$ $6$ $[ -6, 9, -1 ]$ $81$ $[]$ $[]$
$304$ $[ 0, 0, 2 ]$ $9$ $6$ $[ -9, 6, -1 ]$ $81$ $[]$ $[]$
$304$ $[ 0, 0, 2 ]$ $9$ $6$ $[ -6, 9, -1 ]$ $81$ $[]$ $[]$
$304$ $[ 0, 0, 2 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$305$ $[ 0, 0 ]$ $1$ $3$ $[ 0, -3, 1 ]$ $81$ $[]$ $[]$
$305$ $[ 0, 20 ]$ $3$ $6$ $[ -12, 36, -1 ]$ $837$ $[]$ $[]$
$305$ $[ 0, 10 ]$ $6$ $6$ $[ -10, 25, -3 ]$ $1257$ $[ 61 ]$ $[ 3 ]$
$306$ $[ 2, 0 ]$ $3$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$306$ $[ 2, 0 ]$ $3$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$306$ $[ 2, 0 ]$ $3$ $6$ $[ -1, -4, 1 ]$ $321$ $[]$ $[]$
$308$ $[ 0, 0, 0 ]$ $1$ $3$ $[ 1, -6, -2 ]$ $1016$ $[]$ $[]$
$308$ $[ 0, 2, 0 ]$ $3$ $6$ $[ 1, -4, -1 ]$ $321$ $[]$ $[]$
$308$ $[ 0, 2, 0 ]$ $3$ $6$ $[ -13, 46, -49 ]$ $257$ $[]$ $[]$
$309$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$310$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$310$ $[ 0, 10 ]$ $3$ $6$ $[ -1, -4, -1 ]$ $169$ $[]$ $[]$
$310$ $[ 0, 10 ]$ $3$ $6$ $[ -1, -4, 1 ]$ $321$ $[]$ $[]$
$310$ $[ 0, 10 ]$ $3$ $6$ $[ -1, -10, 7 ]$ $4065$ $[]$ $[]$
$310$ $[ 0, 10 ]$ $3$ $6$ $[ 1, -4, -1 ]$ $321$ $[]$ $[]$
$312$ $[ 0, 0, 0, 4 ]$ $3$ $6$ $[ 0, -6, 1 ]$ $837$ $[]$ $[]$
$315$ $[ 0, 2, 0 ]$ $2$ $6$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$315$ $[ 0, 0, 2 ]$ $3$ $6$ $[ 1, -5, 1 ]$ $404$ $[]$ $[]$
$315$ $[ 0, 0, 2 ]$ $3$ $6$ $[ 1, -5, 1 ]$ $404$ $[]$ $[]$
$319$ $[ 0, 0 ]$ $1$ $3$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$322$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -4, -2 ]$ $316$ $[]$ $[]$
$325$ $[ 0, 0 ]$ $1$ $3$ $[ -3, -1, 5 ]$ $148$ $[]$ $[]$
$325$ $[ 0, 0 ]$ $1$ $3$ $[ 3, -1, -5 ]$ $148$ $[]$ $[]$
$325$ $[ 0, 6 ]$ $2$ $6$ $[ -1, -5, 3 ]$ $564$ $[]$ $[]$
$325$ $[ 10, 6 ]$ $2$ $6$ $[ -1, -5, 3 ]$ $564$ $[]$ $[]$
$325$ $[ 10, 6 ]$ $2$ $6$ $[ -1, -5, 3 ]$ $564$ $[]$ $[]$
$325$ $[ 0, 4 ]$ $3$ $12$ $[ -8, 10, -1 ]$ $1765$ $[ 5 ]$ $[ 1 ]$
$325$ $[ 5, 3 ]$ $4$ $12$ $[ -11, 29, -9 ]$ $5780$ $[ 5, 65 ]$ $[ 4, 18 ]$
$325$ $[ 15, 3 ]$ $4$ $12$ $[ -11, 29, -9 ]$ $5780$ $[ 5, 65 ]$ $[ 4, 18 ]$
$327$ $[ 0, 0 ]$ $1$ $3$ $[ 3, -1, -5 ]$ $148$ $[]$ $[]$
$327$ $[ 0, 54 ]$ $2$ $6$ $[ -8, 10, -1 ]$ $1765$ $[ 109 ]$ $[ 1 ]$
$328$ $[ 0, 0, 0 ]$ $1$ $3$ $[ 2, -6, -10 ]$ $788$ $[]$ $[]$
$328$ $[ 0, 0, 0 ]$ $1$ $3$ $[ 4, 2, -2 ]$ $148$ $[]$ $[]$
$328$ $[ 0, 0, 20 ]$ $2$ $6$ $[ 1, -2, -1 ]$ $49$ $[ 41 ]$ $[ 8 ]$
$329$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -2, -1 ]$ $49$ $[]$ $[]$
$329$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$329$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -2, -1 ]$ $49$ $[]$ $[]$
$330$ $[ 1, 2, 5 ]$ $2$ $24$ $[ 1, -18, 22 ]$ $3368$ $[ -15, -55 ]$ $[ 72, 288 ]$
$331$ $[ 0 ]$ $1$ $3$ $[ 2, -4, -7 ]$ $229$ $[]$ $[]$
$332$ $[ 0, 0 ]$ $1$ $3$ $[ -4, 3, 1 ]$ $49$ $[]$ $[]$
$333$ $[ 0, 0 ]$ $1$ $3$ $[ 3, -1, -5 ]$ $148$ $[]$ $[]$
$333$ $[ 0, 12 ]$ $3$ $6$ $[ -8, 16, -1 ]$ $229$ $[]$ $[]$
$333$ $[ 0, 12 ]$ $3$ $12$ $[ -10, 24, -3 ]$ $3021$ $[ -3 ]$ $[ 3 ]$
$333$ $[ 0, 6 ]$ $6$ $6$ $[ -12, 36, -3 ]$ $2349$ $[ 37 ]$ $[ 3 ]$
$333$ $[ 2, 18 ]$ $6$ $12$ $[ -11, 30, -21 ]$ $1929$ $[ 37 ]$ $[ 21 ]$
$333$ $[ 0, 4 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$333$ $[ 0, 4 ]$ $9$ $6$ $[ -6, 9, -1 ]$ $81$ $[]$ $[]$
$334$ $[ 0 ]$ $1$ $3$ $[ 1, -5, -4 ]$ $469$ $[]$ $[]$
$334$ $[ 0 ]$ $1$ $3$ $[ -1, -7, 8 ]$ $733$ $[]$ $[]$
$338$ $[ 0 ]$ $1$ $3$ $[ -3, -4, 13 ]$ $49$ $[]$ $[]$
$338$ $[ 0 ]$ $1$ $3$ $[ -3, -4, 13 ]$ $49$ $[]$ $[]$
$338$ $[ 78 ]$ $2$ $6$ $[ -17, 94, -169 ]$ $49$ $[]$ $[]$
$338$ $[ 52 ]$ $3$ $6$ $[ -17, 94, -169 ]$ $49$ $[]$ $[]$
$338$ $[ 52 ]$ $3$ $6$ $[ -17, 94, -169 ]$ $49$ $[]$ $[]$
$338$ $[ 26 ]$ $6$ $12$ $[ -17, 94, -169 ]$ $49$ $[]$ $[]$
$340$ $[ 0, 0, 0 ]$ $1$ $3$ $[ 1, -5, 1 ]$ $404$ $[]$ $[]$
$340$ $[ 0, 0, 8 ]$ $2$ $6$ $[ 1, -7, -1 ]$ $1524$ $[]$ $[]$
$340$ $[ 0, 2, 4 ]$ $4$ $12$ $[ -8, 10, -2 ]$ $1076$ $[ 5, 85 ]$ $[ 2, 18 ]$
$342$ $[ 2, 0 ]$ $3$ $6$ $[ 0, -3, 1 ]$ $81$ $[]$ $[]$
$342$ $[ 2, 0 ]$ $3$ $6$ $[ 1, -4, -1 ]$ $321$ $[]$ $[]$
$342$ $[ 0, 2 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$342$ $[ 0, 2 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$342$ $[ 0, 2 ]$ $9$ $6$ $[ 0, -3, 1 ]$ $81$ $[]$ $[]$
$342$ $[ 0, 2 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$342$ $[ 0, 2 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$343$ $[ 0 ]$ $1$ $6$ $[ -6, 5, -1 ]$ $49$ $[ -7 ]$ $[ 232 ]$
$343$ $[ 98 ]$ $3$ $12$ $[ -6, 5, -1 ]$ $49$ $[ -7 ]$ $[ 232 ]$
$343$ $[ 42 ]$ $7$ $6$ $[ -6, 5, -1 ]$ $49$ $[]$ $[]$
$343$ $[ 14 ]$ $21$ $12$ $[ -6, 5, -1 ]$ $49$ $[]$ $[]$
$343$ $[ 14 ]$ $21$ $12$ $[ -6, 5, -1 ]$ $49$ $[]$ $[]$
$344$ $[ 0, 0, 0 ]$ $1$ $3$ $[ -3, -1, 4 ]$ $229$ $[]$ $[]$
$345$ $[ 0, 0, 0 ]$ $1$ $3$ $[ 1, -4, -2 ]$ $316$ $[]$ $[]$
$345$ $[ 0, 2, 0 ]$ $2$ $6$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$346$ $[ 0 ]$ $1$ $3$ $[ 0, -4, 1 ]$ $229$ $[]$ $[]$
$351$ $[ 2, 0 ]$ $9$ $6$ $[ -6, 9, -1 ]$ $81$ $[]$ $[]$
$352$ $[ 1, 0, 5 ]$ $2$ $12$ $[ 1, -4, -2 ]$ $316$ $[ -11 ]$ $[ 128 ]$
$353$ $[ 0 ]$ $1$ $3$ $[ 0, -4, 1 ]$ $229$ $[]$ $[]$
$354$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -4, -2 ]$ $316$ $[]$ $[]$
$357$ $[ 0, 0, 0 ]$ $1$ $3$ $[ -1, -4, 2 ]$ $316$ $[]$ $[]$
$357$ $[ 0, 0, 8 ]$ $2$ $6$ $[ 1, -3, -1 ]$ $148$ $[]$ $[]$
$360$ $[ 0, 1, 0, 0 ]$ $2$ $6$ $[ -1, -4, 2 ]$ $316$ $[]$ $[]$
$360$ $[ 0, 1, 0, 2 ]$ $2$ $6$ $[ 0, -4, -1 ]$ $229$ $[ 40 ]$ $[ 64 ]$
$360$ $[ 0, 1, 0, 2 ]$ $2$ $6$ $[ 0, -4, -1 ]$ $229$ $[ 40 ]$ $[ 64 ]$
$360$ $[ 1, 1, 3, 0 ]$ $2$ $6$ $[ 1, -6, 2 ]$ $568$ $[ 24 ]$ $[ 8 ]$
$360$ $[ 1, 1, 3, 0 ]$ $2$ $6$ $[ 1, -6, 2 ]$ $568$ $[ 24 ]$ $[ 8 ]$
$360$ $[ 0, 0, 2, 0 ]$ $3$ $6$ $[ -1, -5, 3 ]$ $564$ $[]$ $[]$
$361$ $[ 0 ]$ $1$ $3$ $[ 3, 0, -3 ]$ $81$ $[]$ $[]$
$361$ $[ 0 ]$ $1$ $3$ $[ -3, 0, 3 ]$ $81$ $[]$ $[]$
$361$ $[ 114 ]$ $3$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$361$ $[ 114 ]$ $3$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$361$ $[ 38 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$361$ $[ 38 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$361$ $[ 38 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$361$ $[ 38 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$361$ $[ 38 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$364$ $[ 0, 2, 6 ]$ $6$ $12$ $[ 1, -10, -14 ]$ $1384$ $[ 13 ]$ $[ 16 ]$
$365$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -2, -1 ]$ $49$ $[]$ $[]$
$366$ $[ 0, 30 ]$ $2$ $6$ $[ 1, -10, -8 ]$ $961$ $[]$ $[]$
$366$ $[ 0, 20 ]$ $3$ $6$ $[ 1, -5, -3 ]$ $564$ $[]$ $[]$
$366$ $[ 0, 20 ]$ $3$ $6$ $[ -1, -9, 11 ]$ $1556$ $[]$ $[]$
$369$ $[ 0, 0 ]$ $1$ $3$ $[ 2, -2, -2 ]$ $148$ $[]$ $[]$
$369$ $[ 0, 0 ]$ $1$ $3$ $[ -2, -2, 2 ]$ $148$ $[]$ $[]$
$369$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$369$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -4, -2 ]$ $316$ $[]$ $[]$
$369$ $[ 0, 10 ]$ $4$ $6$ $[ -1, -5, 3 ]$ $564$ $[ 41 ]$ $[ 2 ]$
$370$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -8, 10 ]$ $892$ $[]$ $[]$
$370$ $[ 0, 18 ]$ $2$ $6$ $[ 1, -4, -2 ]$ $316$ $[]$ $[]$
$370$ $[ 0, 12 ]$ $3$ $6$ $[ 1, -4, -2 ]$ $316$ $[]$ $[]$
$370$ $[ 2, 12 ]$ $6$ $12$ $[ 1, -3, -1 ]$ $148$ $[]$ $[]$
$371$ $[ 0, 0 ]$ $1$ $3$ $[ 0, -4, -1 ]$ $229$ $[]$ $[]$
$372$ $[ 0, 0, 10 ]$ $3$ $6$ $[ 1, -9, -11 ]$ $1556$ $[]$ $[]$
$374$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -6, -5 ]$ $785$ $[]$ $[]$
$374$ $[ 0, 0 ]$ $1$ $3$ $[ -3, -2, 7 ]$ $257$ $[]$ $[]$
$374$ $[ 0, 8 ]$ $2$ $6$ $[ -9, 6, -1 ]$ $81$ $[]$ $[]$
$377$ $[ 4, 2 ]$ $42$ $12$ $[ -7, 14, -7 ]$ $49$ $[ 29 ]$ $[ 7 ]$
$378$ $[ 6, 2 ]$ $3$ $6$ $[ 1, -4, -1 ]$ $321$ $[]$ $[]$
$378$ $[ 6, 2 ]$ $3$ $6$ $[ 1, -4, -1 ]$ $321$ $[]$ $[]$
$378$ $[ 12, 2 ]$ $3$ $6$ $[ 1, -4, -1 ]$ $321$ $[]$ $[]$
$378$ $[ 12, 2 ]$ $3$ $6$ $[ 1, -4, -1 ]$ $321$ $[]$ $[]$
$378$ $[ 2, 0 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$380$ $[ 0, 2, 0 ]$ $2$ $6$ $[ -1, -5, 4 ]$ $469$ $[ 5 ]$ $[ 16 ]$
$380$ $[ 0, 0, 6 ]$ $3$ $6$ $[ 1, -4, -3 ]$ $257$ $[]$ $[]$
$380$ $[ 0, 1, 9 ]$ $4$ $12$ $[ 1, -3, -1 ]$ $148$ $[ -95, -19 ]$ $[ 8, 244 ]$
$382$ $[ 0 ]$ $1$ $3$ $[ 1, -4, 1 ]$ $169$ $[]$ $[]$
$382$ $[ 0 ]$ $1$ $3$ $[ 5, 6, 1 ]$ $49$ $[]$ $[]$
$385$ $[ 0, 0, 0 ]$ $1$ $3$ $[ 3, -1, -5 ]$ $148$ $[]$ $[]$
$385$ $[ 0, 0, 0 ]$ $1$ $3$ $[ 3, -1, -5 ]$ $148$ $[]$ $[]$
$385$ $[ 0, 0, 0 ]$ $1$ $3$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$386$ $[ 64 ]$ $3$ $6$ $[ -11, 25, -16 ]$ $229$ $[]$ $[]$
$386$ $[ 64 ]$ $3$ $6$ $[ -11, 13, -4 ]$ $229$ $[]$ $[]$
$387$ $[ 0, 0 ]$ $1$ $3$ $[ -2, -5, 8 ]$ $568$ $[]$ $[]$
$387$ $[ 3, 21 ]$ $2$ $12$ $[ 1, -4, -2 ]$ $316$ $[ 129, -43 ]$ $[ 32, 53 ]$
$387$ $[ 0, 14 ]$ $3$ $6$ $[ -9, 22, -9 ]$ $257$ $[]$ $[]$
$387$ $[ 0, 14 ]$ $3$ $6$ $[ -5, 6, -1 ]$ $49$ $[]$ $[]$
$387$ $[ 0, 14 ]$ $3$ $12$ $[ 1, -4, -3 ]$ $257$ $[ -3 ]$ $[ 19 ]$
$387$ $[ 0, 6 ]$ $7$ $6$ $[ -6, 5, -1 ]$ $49$ $[]$ $[]$
$387$ $[ 0, 6 ]$ $7$ $6$ $[ -5, 6, -1 ]$ $49$ $[]$ $[]$
$387$ $[ 0, 6 ]$ $7$ $6$ $[ -5, 6, -1 ]$ $49$ $[]$ $[]$
$388$ $[ 0, 0 ]$ $1$ $3$ $[ 2, -1, -1 ]$ $49$ $[]$ $[]$
$388$ $[ 0, 48 ]$ $2$ $6$ $[ 1, -8, 3 ]$ $1425$ $[ 97 ]$ $[ 3 ]$
$389$ $[ 0 ]$ $1$ $3$ $[ 0, -4, -2 ]$ $148$ $[]$ $[]$
$390$ $[ 0, 2, 6 ]$ $2$ $6$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$390$ $[ 0, 2, 6 ]$ $2$ $6$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$391$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -4, -3 ]$ $257$ $[]$ $[]$
$391$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -4, 1 ]$ $169$ $[]$ $[]$
$392$ $[ 0, 1, 0 ]$ $2$ $6$ $[ 1, -5, 1 ]$ $404$ $[ 8 ]$ $[ 7 ]$
$392$ $[ 0, 1, 0 ]$ $2$ $6$ $[ 1, -5, 1 ]$ $404$ $[ 8 ]$ $[ 7 ]$
$392$ $[ 0, 1, 0 ]$ $2$ $12$ $[ 0, -4, -1 ]$ $229$ $[ -7, -56 ]$ $[ 8, 256 ]$
$392$ $[ 0, 1, 14 ]$ $6$ $12$ $[ 1, -5, 1 ]$ $404$ $[ 8 ]$ $[ 7 ]$
$392$ $[ 0, 1, 14 ]$ $6$ $24$ $[ 0, -4, -1 ]$ $229$ $[ -7, -56 ]$ $[ 8, 256 ]$
$395$ $[ 0, 0 ]$ $1$ $3$ $[ 0, -3, 1 ]$ $81$ $[]$ $[]$
$395$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -5, 3 ]$ $564$ $[]$ $[]$
$395$ $[ 0, 0 ]$ $1$ $3$ $[ 2, -1, -1 ]$ $49$ $[]$ $[]$
$396$ $[ 1, 0, 5 ]$ $2$ $12$ $[ -1, -8, 10 ]$ $892$ $[ -11 ]$ $[ 8 ]$
$396$ $[ 0, 2, 0 ]$ $3$ $6$ $[ 1, -4, -1 ]$ $321$ $[]$ $[]$
$397$ $[ 44 ]$ $9$ $6$ $[ -9, 6, -1 ]$ $81$ $[]$ $[]$
$398$ $[ 22 ]$ $9$ $6$ $[ -9, 6, -1 ]$ $81$ $[]$ $[]$
$399$ $[ 0, 0, 0 ]$ $1$ $3$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$399$ $[ 0, 0, 0 ]$ $1$ $3$ $[ -1, -7, 9 ]$ $404$ $[]$ $[]$
$399$ $[ 0, 0, 2 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$402$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -4, 2 ]$ $316$ $[]$ $[]$
$402$ $[ 0, 22 ]$ $3$ $6$ $[ 1, -10, 5 ]$ $2505$ $[]$ $[]$
$402$ $[ 0, 22 ]$ $3$ $6$ $[ 1, -4, -1 ]$ $321$ $[]$ $[]$
$402$ $[ 0, 22 ]$ $3$ $6$ $[ -14, 49, -1 ]$ $1345$ $[]$ $[]$
$402$ $[ 0, 22 ]$ $3$ $6$ $[ 1, -6, -5 ]$ $785$ $[]$ $[]$
$404$ $[ 0, 50 ]$ $2$ $6$ $[ 0, -5, -1 ]$ $473$ $[ 101 ]$ $[ 25 ]$
$405$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -5, 3 ]$ $564$ $[]$ $[]$
$405$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -5, -3 ]$ $564$ $[]$ $[]$
$406$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -6, 2 ]$ $568$ $[]$ $[]$
$406$ $[ 0, 14 ]$ $2$ $6$ $[ 1, -5, 1 ]$ $404$ $[]$ $[]$
$406$ $[ 2, 4 ]$ $21$ $12$ $[ -1, -2, 1 ]$ $49$ $[]$ $[]$
$407$ $[ 0, 6 ]$ $6$ $6$ $[ -9, 21, -12 ]$ $621$ $[ 37 ]$ $[ 12 ]$
$408$ $[ 0, 0, 0, 8 ]$ $2$ $6$ $[ 1, -4, -2 ]$ $316$ $[]$ $[]$
$410$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -5, 1 ]$ $404$ $[]$ $[]$
$410$ $[ 2, 0 ]$ $2$ $6$ $[ 1, -3, -1 ]$ $148$ $[]$ $[]$
$411$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -2, 1 ]$ $49$ $[]$ $[]$
$411$ $[ 0, 0 ]$ $1$ $3$ $[ -2, -3, 5 ]$ $169$ $[]$ $[]$
$411$ $[ 0, 0 ]$ $1$ $3$ $[ 3, 0, -3 ]$ $81$ $[]$ $[]$
$413$ $[ 0, 0 ]$ $1$ $3$ $[ -3, -1, 4 ]$ $229$ $[]$ $[]$
$416$ $[ 0, 4, 0 ]$ $2$ $6$ $[ -1, -4, 2 ]$ $316$ $[]$ $[]$
$417$ $[ 0, 0 ]$ $1$ $3$ $[ 0, -4, -1 ]$ $229$ $[]$ $[]$
$417$ $[ 0, 0 ]$ $1$ $3$ $[ 0, -4, -1 ]$ $229$ $[]$ $[]$
$418$ $[ 0, 0 ]$ $1$ $3$ $[ 0, -6, -3 ]$ $621$ $[]$ $[]$
$418$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -5, 4 ]$ $469$ $[]$ $[]$
$418$ $[ 0, 6 ]$ $3$ $6$ $[ 0, -8, -2 ]$ $1940$ $[]$ $[]$
$418$ $[ 0, 6 ]$ $3$ $6$ $[ 1, -3, -1 ]$ $148$ $[]$ $[]$
$422$ $[ 0 ]$ $1$ $3$ $[ 1, -4, -3 ]$ $257$ $[]$ $[]$
$422$ $[ 0 ]$ $1$ $3$ $[ 1, -6, -5 ]$ $785$ $[]$ $[]$
$422$ $[ 0 ]$ $1$ $3$ $[ 5, 6, 1 ]$ $49$ $[]$ $[]$
$423$ $[ 0, 0 ]$ $1$ $3$ $[ 2, -3, -2 ]$ $316$ $[]$ $[]$
$423$ $[ 0, 0 ]$ $1$ $3$ $[ -2, -3, 2 ]$ $316$ $[]$ $[]$
$424$ $[ 0, 0, 0 ]$ $1$ $3$ $[ -2, -3, 2 ]$ $316$ $[]$ $[]$
$424$ $[ 0, 0, 0 ]$ $1$ $3$ $[ 1, -3, -1 ]$ $148$ $[]$ $[]$
$425$ $[ 0, 8 ]$ $2$ $6$ $[ -1, -3, 1 ]$ $148$ $[ 17 ]$ $[ 67 ]$
$425$ $[ 0, 8 ]$ $2$ $6$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$425$ $[ 0, 8 ]$ $2$ $6$ $[ -1, -3, 1 ]$ $148$ $[ 17 ]$ $[ 67 ]$
$425$ $[ 10, 8 ]$ $2$ $6$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$425$ $[ 10, 8 ]$ $2$ $6$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$425$ $[ 10, 8 ]$ $2$ $12$ $[ -1, -3, 1 ]$ $148$ $[ 17 ]$ $[ 67 ]$
$425$ $[ 0, 4 ]$ $4$ $12$ $[ -9, 19, -1 ]$ $1940$ $[ 85, 17 ]$ $[ 1, 2 ]$
$426$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -6, 2 ]$ $568$ $[]$ $[]$
$426$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -5, -4 ]$ $469$ $[]$ $[]$
$427$ $[ 2, 0 ]$ $3$ $6$ $[ -6, 5, -1 ]$ $49$ $[]$ $[]$
$429$ $[ 0, 0, 0 ]$ $1$ $3$ $[ 1, -5, -3 ]$ $564$ $[]$ $[]$
$429$ $[ 0, 0, 0 ]$ $1$ $3$ $[ -3, -1, 5 ]$ $148$ $[]$ $[]$
$429$ $[ 0, 0, 0 ]$ $1$ $3$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$430$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -4, -2 ]$ $316$ $[]$ $[]$
$430$ $[ 2, 0 ]$ $2$ $6$ $[ 1, -3, -1 ]$ $148$ $[]$ $[]$
$430$ $[ 0, 14 ]$ $3$ $6$ $[ -1, -5, 3 ]$ $564$ $[]$ $[]$
$430$ $[ 0, 14 ]$ $3$ $6$ $[ 1, -6, -2 ]$ $1016$ $[]$ $[]$
$431$ $[ 0 ]$ $1$ $3$ $[ -1, -4, 3 ]$ $257$ $[]$ $[]$
$431$ $[ 0 ]$ $1$ $3$ $[ 0, -5, 1 ]$ $473$ $[]$ $[]$
$432$ $[ 0, 0, 2 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$433$ $[ 0 ]$ $1$ $3$ $[ 1, -5, 1 ]$ $404$ $[]$ $[]$
$434$ $[ 0, 0 ]$ $1$ $3$ $[ 2, -5, -8 ]$ $568$ $[]$ $[]$
$434$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -6, 2 ]$ $568$ $[]$ $[]$
$434$ $[ 2, 0 ]$ $3$ $6$ $[ -9, 22, -9 ]$ $257$ $[]$ $[]$
$434$ $[ 2, 0 ]$ $3$ $6$ $[ -1, -6, 3 ]$ $993$ $[]$ $[]$
$434$ $[ 2, 0 ]$ $3$ $6$ $[ -5, 6, -1 ]$ $49$ $[]$ $[]$
$435$ $[ 0, 0, 0 ]$ $1$ $3$ $[ -1, -5, 4 ]$ $469$ $[]$ $[]$
$436$ $[ 0, 0 ]$ $1$ $3$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$436$ $[ 0, 54 ]$ $2$ $6$ $[ -6, 5, -1 ]$ $49$ $[ 109 ]$ $[ 5677 ]$
$438$ $[ 0, 24 ]$ $3$ $6$ $[ -1, -12, 8 ]$ $1772$ $[]$ $[]$
$438$ $[ 0, 24 ]$ $3$ $6$ $[ -1, -6, 2 ]$ $1016$ $[]$ $[]$
$438$ $[ 0, 18 ]$ $4$ $6$ $[ -1, -4, 2 ]$ $316$ $[ 73 ]$ $[ 8 ]$
$438$ $[ 0, 8 ]$ $9$ $6$ $[ 0, -3, 1 ]$ $81$ $[]$ $[]$
$438$ $[ 0, 8 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$438$ $[ 0, 8 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$438$ $[ 0, 4 ]$ $18$ $6$ $[ 0, -3, -1 ]$ $81$ $[ 73 ]$ $[ 3 ]$
$441$ $[ 2, 0 ]$ $3$ $6$ $[ -1, -4, 1 ]$ $321$ $[]$ $[]$
$441$ $[ 2, 0 ]$ $3$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$441$ $[ 2, 14 ]$ $3$ $6$ $[ -1, -4, 1 ]$ $321$ $[]$ $[]$
$441$ $[ 2, 14 ]$ $3$ $6$ $[ -1, -4, 1 ]$ $321$ $[]$ $[]$
$441$ $[ 2, 14 ]$ $3$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$441$ $[ 2, 14 ]$ $3$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$441$ $[ 4, 14 ]$ $3$ $6$ $[ -1, -4, 1 ]$ $321$ $[]$ $[]$
$441$ $[ 4, 14 ]$ $3$ $6$ $[ -1, -4, 1 ]$ $321$ $[]$ $[]$
$441$ $[ 4, 14 ]$ $3$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$441$ $[ 4, 14 ]$ $3$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$441$ $[ 0, 6 ]$ $7$ $6$ $[ -6, 5, -1 ]$ $49$ $[]$ $[]$
$442$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$442$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -4, -2 ]$ $316$ $[]$ $[]$
$442$ $[ 0, 8 ]$ $2$ $6$ $[ 1, -10, -8 ]$ $961$ $[ 17 ]$ $[ 32 ]$
$442$ $[ 6, 0 ]$ $2$ $6$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$442$ $[ 6, 0 ]$ $2$ $6$ $[ 0, -6, -2 ]$ $756$ $[]$ $[]$
$444$ $[ 0, 0, 12 ]$ $3$ $6$ $[ -1, -9, 6 ]$ $3021$ $[]$ $[]$
$444$ $[ 0, 1, 9 ]$ $4$ $12$ $[ 0, -14, -18 ]$ $2228$ $[ -111 ]$ $[ 128 ]$
$445$ $[ 2, 0 ]$ $2$ $6$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$447$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -2, -1 ]$ $49$ $[]$ $[]$
$447$ $[ 0, 0 ]$ $1$ $3$ $[ 3, 0, -3 ]$ $81$ $[]$ $[]$
$448$ $[ 0, 8, 2 ]$ $6$ $12$ $[ -1, -7, 1 ]$ $1524$ $[ -8 ]$ $[ 27 ]$
$448$ $[ 0, 8, 2 ]$ $6$ $12$ $[ -1, -7, 1 ]$ $1524$ $[ -8 ]$ $[ 27 ]$
$448$ $[ 1, 8, 1 ]$ $6$ $12$ $[ -1, -5, -1 ]$ $404$ $[ -56 ]$ $[ 27 ]$
$448$ $[ 1, 8, 1 ]$ $6$ $12$ $[ -1, -5, -1 ]$ $404$ $[ -56 ]$ $[ 27 ]$
$452$ $[ 0, 0 ]$ $1$ $3$ $[ 3, 0, -1 ]$ $81$ $[]$ $[]$
$453$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -2, -1 ]$ $49$ $[]$ $[]$
$456$ $[ 1, 1, 1, 0 ]$ $2$ $12$ $[ -1, -10, 8 ]$ $961$ $[ -3, -8 ]$ $[ 128, 128 ]$
$456$ $[ 0, 0, 0, 6 ]$ $3$ $6$ $[ 0, -3, 1 ]$ $81$ $[]$ $[]$
$459$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -7, -9 ]$ $404$ $[]$ $[]$
$459$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -7, 9 ]$ $404$ $[]$ $[]$
$461$ $[ 0 ]$ $1$ $3$ $[ 2, -1, -1 ]$ $49$ $[]$ $[]$
$462$ $[ 0, 2, 0 ]$ $3$ $6$ $[ 0, -15, -20 ]$ $2700$ $[]$ $[]$
$462$ $[ 0, 2, 0 ]$ $3$ $6$ $[ 0, -6, 3 ]$ $621$ $[]$ $[]$
$464$ $[ 0, 0, 0 ]$ $1$ $3$ $[ 2, -5, -8 ]$ $568$ $[]$ $[]$
$464$ $[ 0, 0, 4 ]$ $7$ $6$ $[ -17, 94, -169 ]$ $49$ $[]$ $[]$
$464$ $[ 0, 0, 4 ]$ $7$ $6$ $[ -6, 5, -1 ]$ $49$ $[]$ $[]$
$464$ $[ 0, 0, 4 ]$ $7$ $6$ $[ 1, -2, -1 ]$ $49$ $[]$ $[]$
$464$ $[ 0, 0, 4 ]$ $7$ $6$ $[ -6, 5, -1 ]$ $49$ $[]$ $[]$
$464$ $[ 0, 0, 4 ]$ $7$ $6$ $[ -5, 6, -1 ]$ $49$ $[]$ $[]$
$464$ $[ 0, 0, 4 ]$ $7$ $6$ $[ -5, 6, -1 ]$ $49$ $[]$ $[]$
$464$ $[ 0, 0, 2 ]$ $14$ $6$ $[ -7, 14, -7 ]$ $49$ $[ 29 ]$ $[ 7 ]$
$464$ $[ 0, 0, 2 ]$ $14$ $6$ $[ -7, 14, -7 ]$ $49$ $[ 29 ]$ $[ 7 ]$
$465$ $[ 0, 0, 0 ]$ $1$ $3$ $[ -1, -5, 3 ]$ $564$ $[]$ $[]$
$465$ $[ 0, 0, 0 ]$ $1$ $3$ $[ -3, -1, 5 ]$ $148$ $[]$ $[]$
$465$ $[ 0, 0, 0 ]$ $1$ $3$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$465$ $[ 0, 0, 10 ]$ $3$ $6$ $[ -14, 61, -81 ]$ $321$ $[]$ $[]$
$466$ $[ 0 ]$ $1$ $3$ $[ 2, -3, -5 ]$ $169$ $[]$ $[]$
$466$ $[ 0 ]$ $1$ $3$ $[ 4, 3, -1 ]$ $49$ $[]$ $[]$
$466$ $[ 116 ]$ $2$ $6$ $[ -1, -4, 3 ]$ $257$ $[]$ $[]$
$468$ $[ 1, 3, 0 ]$ $2$ $12$ $[ 0, -7, -4 ]$ $940$ $[ -3, 12 ]$ $[ 40, 80 ]$
$468$ $[ 0, 3, 3 ]$ $4$ $12$ $[ 0, -6, 2 ]$ $756$ $[ -39 ]$ $[ 32 ]$
$468$ $[ 0, 2, 2 ]$ $6$ $6$ $[ -1, -5, 3 ]$ $564$ $[ 13 ]$ $[ 27 ]$
$468$ $[ 0, 4, 2 ]$ $6$ $6$ $[ -1, -5, 3 ]$ $564$ $[ 13 ]$ $[ 27 ]$
$469$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -3, -1 ]$ $148$ $[]$ $[]$
$469$ $[ 0, 0 ]$ $1$ $3$ $[ 3, 0, -3 ]$ $81$ $[]$ $[]$
$470$ $[ 0, 0 ]$ $1$ $3$ $[ 0, -6, -1 ]$ $837$ $[]$ $[]$
$470$ $[ 0, 0 ]$ $1$ $3$ $[ -3, -5, 12 ]$ $1373$ $[]$ $[]$
$470$ $[ 0, 0 ]$ $1$ $3$ $[ -2, -4, 7 ]$ $229$ $[]$ $[]$
$471$ $[ 0, 0 ]$ $1$ $3$ $[ 0, -4, 1 ]$ $229$ $[]$ $[]$
$474$ $[ 0, 0 ]$ $1$ $3$ $[ -3, -1, 2 ]$ $229$ $[]$ $[]$
$474$ $[ 0, 26 ]$ $3$ $6$ $[ 0, -7, 4 ]$ $940$ $[]$ $[]$
$474$ $[ 0, 26 ]$ $3$ $6$ $[ -1, -8, 10 ]$ $892$ $[]$ $[]$
$475$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -3, -1 ]$ $148$ $[]$ $[]$
$475$ $[ 0, 0 ]$ $1$ $3$ $[ 4, 3, -1 ]$ $49$ $[]$ $[]$
$475$ $[ 0, 0 ]$ $1$ $3$ $[ -2, -3, 5 ]$ $169$ $[]$ $[]$
$475$ $[ 0, 0 ]$ $1$ $3$ $[ -4, 3, 1 ]$ $49$ $[]$ $[]$
$475$ $[ 0, 0 ]$ $1$ $3$ $[ 2, -3, -5 ]$ $169$ $[]$ $[]$
$475$ $[ 0, 0 ]$ $1$ $6$ $[ -10, 27, -16 ]$ $1016$ $[ 5 ]$ $[ 16 ]$
$475$ $[ 10, 0 ]$ $2$ $6$ $[ -1, -2, 1 ]$ $49$ $[]$ $[]$
$475$ $[ 10, 0 ]$ $2$ $6$ $[ -10, 29, -25 ]$ $169$ $[]$ $[]$
$475$ $[ 10, 0 ]$ $2$ $6$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$475$ $[ 0, 6 ]$ $3$ $6$ $[ -13, 50, -49 ]$ $361$ $[]$ $[]$
$475$ $[ 0, 6 ]$ $3$ $12$ $[ -7, 6, -1 ]$ $257$ $[ 5 ]$ $[ 1 ]$
$475$ $[ 5, 9 ]$ $4$ $12$ $[ -9, 23, -17 ]$ $148$ $[ -19, -95 ]$ $[ 17, 200 ]$
$475$ $[ 5, 9 ]$ $4$ $24$ $[ 0, -5, 1 ]$ $473$ $[ -19, -95 ]$ $[ 5, 27 ]$
$475$ $[ 10, 6 ]$ $6$ $12$ $[ -13, 50, -49 ]$ $361$ $[]$ $[]$
$475$ $[ 5, 3 ]$ $12$ $24$ $[ 1, -6, 1 ]$ $761$ $[ 5, -95 ]$ $[ 1, 27 ]$
$475$ $[ 0, 2 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$475$ $[ 10, 2 ]$ $18$ $12$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$476$ $[ 0, 2, 0 ]$ $3$ $6$ $[ -9, 22, -9 ]$ $257$ $[]$ $[]$
$477$ $[ 0, 0 ]$ $1$ $3$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$480$ $[ 0, 4, 0, 0 ]$ $2$ $6$ $[ -1, -4, 2 ]$ $316$ $[]$ $[]$
$480$ $[ 0, 4, 0, 2 ]$ $2$ $6$ $[ 0, -4, -1 ]$ $229$ $[ 40 ]$ $[ 64 ]$
$480$ $[ 0, 4, 0, 2 ]$ $2$ $6$ $[ 0, -4, -1 ]$ $229$ $[ 40 ]$ $[ 64 ]$
$480$ $[ 1, 0, 1, 2 ]$ $2$ $24$ $[ 1, -4, -2 ]$ $316$ $[ -3, -15 ]$ $[ 32, 64 ]$
$480$ $[ 0, 0, 1, 1 ]$ $4$ $24$ $[ -1, -8, 10 ]$ $892$ $[ 12, -20 ]$ $[ 8, 800 ]$
$481$ $[ 4, 0 ]$ $3$ $6$ $[ -6, 5, -1 ]$ $49$ $[]$ $[]$
$482$ $[ 0 ]$ $1$ $3$ $[ 0, -4, 1 ]$ $229$ $[]$ $[]$
$483$ $[ 0, 0, 0 ]$ $1$ $3$ $[ 0, -6, -1 ]$ $837$ $[]$ $[]$
$483$ $[ 0, 3, 11 ]$ $2$ $12$ $[ -8, 16, -1 ]$ $229$ $[ -23 ]$ $[ 14 ]$
$483$ $[ 0, 3, 11 ]$ $2$ $12$ $[ -15, 65, -64 ]$ $733$ $[ -23 ]$ $[ 7 ]$
$485$ $[ 0, 0 ]$ $1$ $3$ $[ 2, -5, -8 ]$ $568$ $[]$ $[]$
$486$ $[ 0 ]$ $1$ $3$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$486$ $[ 0 ]$ $1$ $3$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$486$ $[ 54 ]$ $3$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$486$ $[ 54 ]$ $3$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$486$ $[ 18 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$486$ $[ 18 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$486$ $[ 18 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$486$ $[ 18 ]$ $9$ $6$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$487$ $[ 0 ]$ $1$ $3$ $[ 0, -5, 3 ]$ $257$ $[]$ $[]$
$488$ $[ 0, 0, 0 ]$ $1$ $3$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$490$ $[ 0, 6 ]$ $7$ $6$ $[ -17, 94, -169 ]$ $49$ $[]$ $[]$
$490$ $[ 0, 6 ]$ $7$ $6$ $[ -1, -2, 1 ]$ $49$ $[]$ $[]$
$490$ $[ 0, 6 ]$ $7$ $6$ $[ -1, -2, 1 ]$ $49$ $[]$ $[]$
$492$ $[ 0, 0, 20 ]$ $2$ $6$ $[ 1, -5, -3 ]$ $564$ $[]$ $[]$
$494$ $[ 0, 0 ]$ $1$ $3$ $[ 0, -3, -1 ]$ $81$ $[]$ $[]$
$494$ $[ 0, 0 ]$ $1$ $3$ $[ 1, -6, -7 ]$ $361$ $[]$ $[]$
$494$ $[ 0, 0 ]$ $1$ $3$ $[ -5, 6, -1 ]$ $49$ $[]$ $[]$
$494$ $[ 6, 0 ]$ $2$ $6$ $[ -17, 94, -169 ]$ $49$ $[]$ $[]$
$494$ $[ 0, 6 ]$ $3$ $6$ $[ 1, -3, -1 ]$ $148$ $[]$ $[]$
$494$ $[ 0, 6 ]$ $3$ $6$ $[ 1, -4, -2 ]$ $316$ $[]$ $[]$
$494$ $[ 4, 0 ]$ $3$ $6$ $[ 0, -7, -3 ]$ $1129$ $[]$ $[]$
$494$ $[ 4, 0 ]$ $3$ $6$ $[ 0, -7, -3 ]$ $1129$ $[]$ $[]$
$494$ $[ 4, 0 ]$ $3$ $6$ $[ -10, 29, -25 ]$ $169$ $[]$ $[]$
$494$ $[ 0, 2 ]$ $9$ $6$ $[ -6, 9, -1 ]$ $81$ $[]$ $[]$
$495$ $[ 0, 0, 0 ]$ $1$ $3$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$495$ $[ 0, 2, 0 ]$ $2$ $6$ $[ 1, -3, -1 ]$ $148$ $[]$ $[]$
$495$ $[ 0, 2, 0 ]$ $2$ $6$ $[ -1, -3, 1 ]$ $148$ $[]$ $[]$
$496$ $[ 0, 0, 0 ]$ $1$ $3$ $[ 1, -4, -2 ]$ $316$ $[]$ $[]$
$496$ $[ 0, 0, 10 ]$ $3$ $6$ $[ -1, -5, 3 ]$ $564$ $[]$ $[]$
$496$ $[ 0, 0, 10 ]$ $3$ $6$ $[ 1, -3, -1 ]$ $148$ $[]$ $[]$
$498$ $[ 0, 0 ]$ $1$ $3$ $[ 0, -6, -3 ]$ $621$ $[]$ $[]$
4-dimensional RM-varieties
--------------------------
$N$ $\varepsilon$ $\operatorname{ord}(\varepsilon)$ $\dim(A_f)$ $F$ $\Disc(F)$ $[t_i]$ $[N_{F/\Q}(\delta_i)]$ $\gamma_\varepsilon\in\Br(K_P)[2]$
------- ------------------ ----------------------------------- ------------- --------------------------- ------------ --------------------- ------------------------ ------------------------------------
$25$ $[ 2 ]$ $10$ $8$ $[ -11, 36, -31, 1 ]$ $7625$ $[ 5 ]$ $[ 1 ]$ $[]$
$32$ $[ 0, 1 ]$ $8$ $8$ $[ 0, -6, -4, 2 ]$ $4352$ $[ 8 ]$ $[ 4 ]$ $[]$
$41$ $[ 8 ]$ $5$ $8$ $[ -1, -3, 1, 1 ]$ $725$ $[]$ $[]$ $[]$
$41$ $[ 4 ]$ $10$ $8$ $[ -9, 26, -24, 1 ]$ $1125$ $[ 41 ]$ $[ 1 ]$ $[]$
$45$ $[ 1, 1 ]$ $12$ $16$ $[ -10, 30, -28, 1 ]$ $54864$ $[ -3, -15 ]$ $[ 1, 4 ]$ $[]$
$47$ $[ 0 ]$ $1$ $4$ $[ -1, -5, 5, -1 ]$ $1957$ $[]$ $[]$ $[]$
$48$ $[ 0, 1, 0 ]$ $4$ $8$ $[ 0, -6, -4, 2 ]$ $4352$ $[]$ $[]$ $[]$
$50$ $[ 4 ]$ $5$ $8$ $[ 1, -9, -4, 16 ]$ $7625$ $[]$ $[]$ $[]$
$50$ $[ 2 ]$ $10$ $8$ $[ 0, -5, 0, 5 ]$ $2000$ $[]$ $[]$ $[]$
$51$ $[ 0, 4 ]$ $4$ $8$ $[ 2, -15, -48, -34 ]$ $7232$ $[]$ $[]$ $[]$
$51$ $[ 0, 2 ]$ $8$ $8$ $[ 0, -4, 0, 2 ]$ $2048$ $[]$ $[]$ $[]$
$52$ $[ 1, 1 ]$ $12$ $16$ $[ 2, -13, -38, -26 ]$ $13968$ $[ 13, -4 ]$ $[ 1, 52 ]$ $[]$
$55$ $[ 0, 2 ]$ $5$ $8$ $[ -7, 13, -7, 1 ]$ $725$ $[]$ $[]$ $[]$
$55$ $[ 0, 2 ]$ $5$ $8$ $[ -2, -4, 5, 5 ]$ $2525$ $[]$ $[]$ $[]$
$55$ $[ 2, 2 ]$ $10$ $16$ $[ -9, 27, -31, 11 ]$ $725$ $[ 5 ]$ $[ 11 ]$ $[]$
$61$ $[ 20 ]$ $3$ $8$ $[ -8, 18, -9, 1 ]$ $1957$ $[]$ $[]$ $[]$
$61$ $[ 10 ]$ $6$ $8$ $[ 1, -6, -2, 1 ]$ $23301$ $[ 61 ]$ $[ 1 ]$ $[]$
$62$ $[ 6 ]$ $5$ $8$ $[ 1, -10, -2, 19 ]$ $8725$ $[]$ $[]$ $[]$
$62$ $[ 6 ]$ $5$ $8$ $[ -1, -7, 3, 9 ]$ $4525$ $[]$ $[]$ $[]$
$62$ $[ 2 ]$ $15$ $8$ $[ 1, -4, -4, 1 ]$ $1125$ $[]$ $[]$ $[]$
$62$ $[ 2 ]$ $15$ $8$ $[ 1, -4, -4, 1 ]$ $1125$ $[]$ $[]$ $[]$
$65$ $[ 1, 3 ]$ $4$ $8$ $[ 2, -4, -2, 2 ]$ $6224$ $[ 65 ]$ $[ 4 ]$ $[]$
$65$ $[ 3, 3 ]$ $4$ $8$ $[ 2, -4, -2, 2 ]$ $6224$ $[ 65 ]$ $[ 4 ]$ $[]$
$65$ $[ 0, 2 ]$ $6$ $8$ $[ 2, -3, -4, 1 ]$ $4752$ $[]$ $[]$ $[]$
$66$ $[ 1, 1 ]$ $10$ $8$ $[ -13, 49, -52, 16 ]$ $13625$ $[ 33 ]$ $[ 16 ]$ $[]$
$66$ $[ 1, 1 ]$ $10$ $8$ $[ -13, 49, -52, 16 ]$ $13625$ $[ 33 ]$ $[ 16 ]$ $[]$
$70$ $[ 1, 1 ]$ $12$ $16$ $[ 2, -15, -16, 46 ]$ $51264$ $[ -7 ]$ $[ 4 ]$ $[]$
$72$ $[ 0, 1, 2 ]$ $6$ $16$ $[ 1, -8, -6, 6 ]$ $177300$ $[ 8 ]$ $[ 9 ]$ $[]$
$72$ $[ 1, 1, 1 ]$ $6$ $16$ $[ 1, -12, 10, -2 ]$ $29268$ $[ 24, -8 ]$ $[ 1, 516 ]$ $[]$
$73$ $[ 24 ]$ $3$ $8$ $[ -1, -4, 2, 1 ]$ $3981$ $[]$ $[]$ $[]$
$73$ $[ 18 ]$ $4$ $8$ $[ 2, -3, -6, -1 ]$ $2624$ $[]$ $[]$ $[]$
$73$ $[ 12 ]$ $6$ $8$ $[ -9, 26, -24, 1 ]$ $1125$ $[ 73 ]$ $[ 1 ]$ $[]$
$75$ $[ 0, 4 ]$ $5$ $8$ $[ -2, -6, 7, 11 ]$ $5125$ $[]$ $[]$ $[]$
$77$ $[ 0, 2 ]$ $5$ $8$ $[ -2, -4, 5, 5 ]$ $2525$ $[]$ $[]$ $[]$
$78$ $[ 1, 1 ]$ $12$ $16$ $[ 0, -6, 0, 6 ]$ $13824$ $[ -39 ]$ $[ 4 ]$ $[]$
$80$ $[ 0, 1, 2 ]$ $4$ $16$ $[ 2, -3, -4, 1 ]$ $4752$ $[ 8, 40 ]$ $[ 4, 4 ]$ $[]$
$82$ $[ 8 ]$ $5$ $8$ $[ -1, -5, 2, 4 ]$ $2225$ $[]$ $[]$ $[]$
$82$ $[ 4 ]$ $10$ $8$ $[ -1, -14, 9, 41 ]$ $18625$ $[ 41 ]$ $[ 1 ]$ $[]$
$82$ $[ 4 ]$ $10$ $8$ $[ -21, 141, -336, 256 ]$ $42625$ $[ 41 ]$ $[ 1 ]$ $[]$
$82$ $[ 2 ]$ $20$ $8$ $[ -8, 19, -12, 1 ]$ $2000$ $[ 41 ]$ $[ 1 ]$ $[]$
$84$ $[ 1, 0, 1 ]$ $6$ $8$ $[ 1, -7, -4, 10 ]$ $23724$ $[ 28 ]$ $[ 16 ]$ $[]$
$84$ $[ 1, 0, 1 ]$ $6$ $8$ $[ 1, -7, -4, 10 ]$ $23724$ $[ 28 ]$ $[ 16 ]$ $[]$
$85$ $[ 2, 0 ]$ $2$ $8$ $[ 2, -4, -2, 2 ]$ $6224$ $[]$ $[]$ $[]$
$88$ $[ 0, 0, 2 ]$ $5$ $8$ $[ -1, -8, 1, 11 ]$ $5225$ $[]$ $[]$ $[]$
$90$ $[ 1, 1 ]$ $12$ $16$ $[ 0, -6, 0, 3 ]$ $27648$ $[ -15 ]$ $[ 25 ]$ $[]$
$91$ $[ 0, 4 ]$ $3$ $8$ $[ 1, -6, -3, 1 ]$ $27004$ $[]$ $[]$ $[]$
$91$ $[ 2, 6 ]$ $6$ $16$ $[ -11, 36, -31, 3 ]$ $50908$ $[ 13 ]$ $[ 3 ]$ $[]$
$93$ $[ 0, 6 ]$ $5$ $8$ $[ -9, 26, -24, 1 ]$ $1125$ $[]$ $[]$ $[]$
$95$ $[ 0, 0 ]$ $1$ $4$ $[ 2, -6, -8, 9 ]$ $11344$ $[]$ $[]$ $[]$
$95$ $[ 0, 6 ]$ $3$ $8$ $[ -13, 50, -73, 36 ]$ $7537$ $[]$ $[]$ $[]$
$97$ $[ 0 ]$ $1$ $4$ $[ -3, -1, 6, -1 ]$ $2777$ $[]$ $[]$ $[]$
$99$ $[ 2, 0 ]$ $3$ $8$ $[ -1, -6, 5, 4 ]$ $22545$ $[]$ $[]$ $[]$
$99$ $[ 1, 5 ]$ $6$ $16$ $[ -15, 75, -144, 93 ]$ $22545$ $[ 33, -11 ]$ $[ 1, 93 ]$ $[]$
$99$ $[ 3, 1 ]$ $10$ $16$ $[ -12, 39, -18, 1 ]$ $22000$ $[ -11, 33 ]$ $[ 1, 1 ]$ $[]$
$100$ $[ 0, 2 ]$ $10$ $8$ $[ -11, 36, -31, 1 ]$ $7625$ $[ 5 ]$ $[ 1 ]$ $[]$
$101$ $[ 50 ]$ $2$ $8$ $[ 1, -6, 1, 2 ]$ $15529$ $[ 101 ]$ $[ 20 ]$ $[]$
$102$ $[ 0, 2 ]$ $8$ $8$ $[ 0, -4, 0, 2 ]$ $2048$ $[]$ $[]$ $[]$
$102$ $[ 0, 2 ]$ $8$ $8$ $[ 0, -4, 0, 2 ]$ $2048$ $[]$ $[]$ $[]$
$104$ $[ 0, 0, 2 ]$ $6$ $8$ $[ 2, -13, -38, -26 ]$ $13968$ $[]$ $[]$ $[]$
$104$ $[ 0, 1, 2 ]$ $6$ $16$ $[ -1, -43, 109, 64 ]$ $304596$ $[ 8, 13 ]$ $[ 4, 4 ]$ $[]$
$105$ $[ 0, 1, 3 ]$ $4$ $16$ $[ 2, -4, -4, 2 ]$ $14656$ $[ -35 ]$ $[ 1 ]$ $[]$
$105$ $[ 0, 2, 2 ]$ $6$ $16$ $[ 2, -4, -4, 3 ]$ $11344$ $[]$ $[]$ $[]$
$105$ $[ 1, 0, 1 ]$ $6$ $8$ $[ -11, 36, -32, 4 ]$ $29268$ $[ 21 ]$ $[ 4 ]$ $[]$
$105$ $[ 1, 0, 1 ]$ $6$ $8$ $[ -11, 36, -32, 4 ]$ $29268$ $[ 21 ]$ $[ 4 ]$ $[]$
$109$ $[ 0 ]$ $1$ $4$ $[ 1, -5, -4, 3 ]$ $7537$ $[]$ $[]$ $[]$
$110$ $[ 0, 2 ]$ $5$ $8$ $[ -1, -8, 1, 11 ]$ $5225$ $[]$ $[]$ $[]$
$110$ $[ 2, 2 ]$ $10$ $16$ $[ 0, -17, 0, 71 ]$ $28400$ $[]$ $[]$ $[]$
$111$ $[ 0, 0 ]$ $1$ $4$ $[ 0, -6, 2, 5 ]$ $6224$ $[]$ $[]$ $[]$
$112$ $[ 0, 1, 0 ]$ $4$ $8$ $[ 2, -4, -4, 2 ]$ $14656$ $[ 8 ]$ $[ 4 ]$ $[]$
$112$ $[ 1, 1, 3 ]$ $4$ $16$ $[ 2, -4, -6, 1 ]$ $14656$ $[ -56 ]$ $[ 256 ]$ $[]$
$113$ $[ 56 ]$ $2$ $8$ $[ -11, 39, -46, 9 ]$ $7537$ $[ 113 ]$ $[ 194 ]$ $[]$
$115$ $[ 0, 0 ]$ $1$ $4$ $[ -2, -4, 5, 2 ]$ $15317$ $[]$ $[]$ $[]$
$115$ $[ 2, 0 ]$ $2$ $8$ $[ 0, -5, -2, 1 ]$ $5744$ $[]$ $[]$ $[]$
$119$ $[ 0, 0 ]$ $1$ $4$ $[ 1, -5, -1, 3 ]$ $9301$ $[]$ $[]$ $[]$
$120$ $[ 1, 1, 1, 0 ]$ $2$ $8$ $[ 1, -9, -9, -2 ]$ $20308$ $[ 24 ]$ $[ 256 ]$ $[]$
$120$ $[ 1, 1, 1, 0 ]$ $2$ $8$ $[ 1, -9, -9, -2 ]$ $20308$ $[ 24 ]$ $[ 256 ]$ $[]$
$120$ $[ 0, 1, 1, 1 ]$ $4$ $32$ $[ 0, -5, -1, 1 ]$ $6809$ $[ -120, -3, -24 ]$ $[ 16, 128, 1856 ]$ $[]$
$122$ $[ 10 ]$ $6$ $8$ $[ 0, -7, -6, 1 ]$ $10512$ $[]$ $[]$ $[]$
$123$ $[ 0, 20 ]$ $2$ $8$ $[ 1, -6, -6, 2 ]$ $22676$ $[]$ $[]$ $[]$
$125$ $[ 20 ]$ $5$ $16$ $[ -11, 36, -31, 1 ]$ $7625$ $[ 5 ]$ $[ 1 ]$ $[]$
$125$ $[ 10 ]$ $10$ $8$ $[ -11, 36, -31, 1 ]$ $7625$ $[ 5 ]$ $[ 1 ]$ $[]$
$126$ $[ 1, 3 ]$ $6$ $16$ $[ 2, -13, -38, -26 ]$ $13968$ $[ -7 ]$ $[ 36 ]$ $[]$
$128$ $[ 0, 4 ]$ $8$ $8$ $[ 0, -6, -4, 2 ]$ $4352$ $[ 8 ]$ $[ 4 ]$ $[]$
$130$ $[ 0, 2 ]$ $6$ $8$ $[ 2, -3, -4, 1 ]$ $4752$ $[]$ $[]$ $[]$
$130$ $[ 2, 2 ]$ $6$ $8$ $[ -1, -39, 149, -152 ]$ $223668$ $[ 65 ]$ $[ 4 ]$ $[]$
$130$ $[ 2, 2 ]$ $6$ $8$ $[ -1, -39, 149, -152 ]$ $223668$ $[ 65 ]$ $[ 4 ]$ $[]$
$132$ $[ 1, 1, 0 ]$ $2$ $8$ $[ 2, -5, -2, 2 ]$ $19664$ $[]$ $[]$ $[ 2, 3 ]$
$132$ $[ 1, 1, 0 ]$ $2$ $8$ $[ 2, -5, -2, 2 ]$ $19664$ $[]$ $[]$ $[ 2, 3 ]$
$135$ $[ 3, 1 ]$ $12$ $16$ $[ -10, 30, -28, 1 ]$ $54864$ $[ -3, -15 ]$ $[ 1, 4 ]$ $[]$
$136$ $[ 0, 1, 0 ]$ $2$ $8$ $[ -1, -9, 9, -2 ]$ $20308$ $[ 8 ]$ $[ 64 ]$ $[]$
$136$ $[ 0, 1, 0 ]$ $2$ $8$ $[ -1, -8, 8, 4 ]$ $33428$ $[ 8 ]$ $[ 16 ]$ $[]$
$136$ $[ 0, 1, 8 ]$ $2$ $16$ $[ 1, -5, -3, 4 ]$ $8468$ $[ 8, 136 ]$ $[ 32, 2176 ]$ $[]$
$137$ $[ 0 ]$ $1$ $4$ $[ 3, 0, -4, -1 ]$ $725$ $[]$ $[]$ $[]$
$140$ $[ 0, 1, 1 ]$ $12$ $16$ $[ 0, -13, 6, 10 ]$ $567504$ $[ -7, -35 ]$ $[ 16, 49 ]$ $[]$
$142$ $[ 14 ]$ $5$ $8$ $[ -2, -4, 5, 5 ]$ $2525$ $[]$ $[]$ $[]$
$143$ $[ 0, 0 ]$ $1$ $4$ $[ -3, -1, 5, 1 ]$ $1957$ $[]$ $[]$ $[]$
$144$ $[ 0, 1, 0 ]$ $4$ $8$ $[ 0, -6, -4, 2 ]$ $4352$ $[]$ $[]$ $[]$
$144$ $[ 1, 1, 3 ]$ $4$ $16$ $[ 2, -4, -4, 2 ]$ $14656$ $[ -24 ]$ $[ 144 ]$ $[]$
$146$ $[ 0 ]$ $1$ $4$ $[ -2, -4, 8, -1 ]$ $6224$ $[]$ $[]$ $[]$
$146$ $[ 12 ]$ $6$ $8$ $[ -1, -8, 3, 12 ]$ $72153$ $[ 73 ]$ $[ 9 ]$ $[]$
$150$ $[ 0, 4 ]$ $5$ $8$ $[ 1, -19, -4, 76 ]$ $32625$ $[]$ $[]$ $[]$
$150$ $[ 0, 2 ]$ $10$ $8$ $[ 0, -5, 0, 5 ]$ $2000$ $[]$ $[]$ $[]$
$153$ $[ 2, 0 ]$ $3$ $8$ $[ -1, -5, 3, 3 ]$ $12357$ $[]$ $[]$ $[]$
$153$ $[ 0, 4 ]$ $4$ $8$ $[ 2, -15, -48, -34 ]$ $7232$ $[]$ $[]$ $[]$
$153$ $[ 0, 2 ]$ $8$ $8$ $[ 0, -4, 0, 2 ]$ $2048$ $[]$ $[]$ $[]$
$154$ $[ 1, 5 ]$ $6$ $16$ $[ 2, -6, -4, 1 ]$ $54864$ $[ -7 ]$ $[ 4 ]$ $[ 11, 11 ]$
$154$ $[ 0, 2 ]$ $5$ $8$ $[ -1, -9, 4, 16 ]$ $7625$ $[]$ $[]$ $[]$
$155$ $[ 0, 0 ]$ $1$ $4$ $[ -1, -9, 9, -2 ]$ $20308$ $[]$ $[]$ $[]$
$155$ $[ 0, 0 ]$ $1$ $4$ $[ 1, -5, -3, 4 ]$ $8468$ $[]$ $[]$ $[]$
$155$ $[ 0, 10 ]$ $3$ $8$ $[ -7, 13, -7, 1 ]$ $725$ $[]$ $[]$ $[]$
$155$ $[ 0, 10 ]$ $3$ $8$ $[ -11, 33, -31, 9 ]$ $9301$ $[]$ $[]$ $[]$
$155$ $[ 1, 15 ]$ $4$ $16$ $[ 0, -6, 2, 5 ]$ $6224$ $[ -155, -31 ]$ $[ 4, 1249 ]$ $[]$
$156$ $[ 0, 1, 1 ]$ $12$ $16$ $[ 0, -9, -6, 6 ]$ $25488$ $[ 13, -3 ]$ $[ 16, 333 ]$ $[]$
$159$ $[ 0, 0 ]$ $1$ $4$ $[ -3, -1, 7, -3 ]$ $1957$ $[]$ $[]$ $[]$
$159$ $[ 0, 26 ]$ $2$ $8$ $[ -1, -5, 3, 1 ]$ $12197$ $[]$ $[]$ $[]$
$164$ $[ 0, 0 ]$ $1$ $4$ $[ -2, -12, 34, -18 ]$ $25808$ $[]$ $[]$ $[]$
$165$ $[ 1, 0, 5 ]$ $2$ $8$ $[ 2, -4, -8, -1 ]$ $6224$ $[]$ $[]$ $[ 3, 11 ]$
$165$ $[ 1, 0, 5 ]$ $2$ $8$ $[ 2, -4, -8, -1 ]$ $6224$ $[]$ $[]$ $[ 3, 11 ]$
$165$ $[ 0, 0, 2 ]$ $5$ $8$ $[ -1, -3, 1, 1 ]$ $725$ $[]$ $[]$ $[]$
$165$ $[ 0, 0, 2 ]$ $5$ $8$ $[ -17, 98, -212, 121 ]$ $5725$ $[]$ $[]$ $[]$
$165$ $[ 0, 0, 2 ]$ $5$ $8$ $[ -9, 26, -24, 1 ]$ $1125$ $[]$ $[]$ $[]$
$165$ $[ 0, 0, 2 ]$ $5$ $8$ $[ -1, -3, 1, 1 ]$ $725$ $[]$ $[]$ $[]$
$165$ $[ 1, 0, 1 ]$ $10$ $16$ $[ 1, -3, -1, 1 ]$ $725$ $[ -11 ]$ $[ 25 ]$ $[ 3, 3 ]$
$168$ $[ 0, 1, 0, 0 ]$ $2$ $8$ $[ 2, -5, -10, -2 ]$ $19664$ $[]$ $[]$ $[]$
$168$ $[ 1, 1, 0, 3 ]$ $2$ $16$ $[ -1, -8, 8, 4 ]$ $33428$ $[ -8 ]$ $[ 64 ]$ $[]$
$168$ $[ 0, 0, 1, 1 ]$ $6$ $16$ $[ 1, -10, -10, 4 ]$ $136908$ $[ -7 ]$ $[ 16 ]$ $[]$
$170$ $[ 0, 4 ]$ $4$ $8$ $[ 2, -11, -12, 34 ]$ $37952$ $[]$ $[]$ $[]$
$170$ $[ 0, 2 ]$ $8$ $8$ $[ 0, -4, 0, 2 ]$ $2048$ $[]$ $[]$ $[]$
$172$ $[ 0, 14 ]$ $3$ $8$ $[ -1, -7, 0, 6 ]$ $30972$ $[]$ $[]$ $[]$
$173$ $[ 0 ]$ $1$ $4$ $[ 1, -3, -1, 1 ]$ $725$ $[]$ $[]$ $[]$
$176$ $[ 0, 0, 2 ]$ $5$ $8$ $[ -1, -8, 1, 11 ]$ $5225$ $[]$ $[]$ $[]$
$176$ $[ 1, 0, 1 ]$ $10$ $16$ $[ 1, -10, -7, 19 ]$ $9225$ $[ -4, -11 ]$ $[ 1, 400 ]$ $[]$
$182$ $[ 2, 0 ]$ $3$ $8$ $[ 0, -9, -3, 12 ]$ $93177$ $[]$ $[]$ $[]$
$183$ $[ 0, 10 ]$ $6$ $8$ $[ -1, -4, 4, 1 ]$ $1125$ $[ 61 ]$ $[ 1 ]$ $[]$
$184$ $[ 0, 1, 0 ]$ $2$ $8$ $[ 1, -8, 2, 6 ]$ $25492$ $[ 8 ]$ $[ 8 ]$ $[]$
$185$ $[ 2, 18 ]$ $2$ $16$ $[ 1, -7, -10, -2 ]$ $20808$ $[ 37, 185 ]$ $[ 4, 34 ]$ $[]$
$186$ $[ 0, 6 ]$ $5$ $8$ $[ 1, -15, -2, 44 ]$ $19225$ $[]$ $[]$ $[]$
$186$ $[ 0, 2 ]$ $15$ $8$ $[ 1, -4, -4, 1 ]$ $1125$ $[]$ $[]$ $[]$
$186$ $[ 0, 2 ]$ $15$ $8$ $[ -1, -4, 4, 1 ]$ $1125$ $[]$ $[]$ $[]$
$186$ $[ 0, 2 ]$ $15$ $8$ $[ 1, -4, -4, 1 ]$ $1125$ $[]$ $[]$ $[]$
$186$ $[ 0, 2 ]$ $15$ $8$ $[ 1, -4, -4, 1 ]$ $1125$ $[]$ $[]$ $[]$
$187$ $[ 0, 0 ]$ $1$ $4$ $[ -1, -6, 2, 2 ]$ $33844$ $[]$ $[]$ $[]$
$187$ $[ 2, 0 ]$ $5$ $8$ $[ -8, 18, -13, 1 ]$ $725$ $[]$ $[]$ $[]$
$187$ $[ 2, 0 ]$ $5$ $8$ $[ 2, -10, -26, -11 ]$ $3600$ $[]$ $[]$ $[]$
$190$ $[ 1, 9 ]$ $4$ $16$ $[ 0, -10, -4, 15 ]$ $95488$ $[ -95 ]$ $[ 25 ]$ $[]$
$192$ $[ 0, 4, 0 ]$ $4$ $8$ $[ 0, -6, -4, 2 ]$ $4352$ $[]$ $[]$ $[]$
$194$ $[ 0 ]$ $1$ $4$ $[ 2, -5, -2, 3 ]$ $14272$ $[]$ $[]$ $[]$
$194$ $[ 0 ]$ $1$ $4$ $[ -2, -5, 6, 7 ]$ $8768$ $[]$ $[]$ $[]$
$194$ $[ 32 ]$ $3$ $8$ $[ -1, -8, 6, 10 ]$ $81908$ $[]$ $[]$ $[]$
$194$ $[ 24 ]$ $4$ $8$ $[ 2, -7, -4, 5 ]$ $152272$ $[ 97 ]$ $[ 9 ]$ $[]$
$195$ $[ 0, 0, 2 ]$ $6$ $8$ $[ 0, -6, 0, 6 ]$ $13824$ $[]$ $[]$ $[]$
$195$ $[ 0, 0, 2 ]$ $6$ $8$ $[ 2, -4, -8, -2 ]$ $7488$ $[]$ $[]$ $[]$
$198$ $[ 3, 1 ]$ $10$ $8$ $[ 0, -10, 0, 20 ]$ $8000$ $[ 33 ]$ $[ 16 ]$ $[]$
$198$ $[ 3, 1 ]$ $10$ $8$ $[ 0, -10, 0, 20 ]$ $8000$ $[ 33 ]$ $[ 16 ]$ $[]$
$199$ $[ 0 ]$ $1$ $4$ $[ 3, 0, -4, -1 ]$ $725$ $[]$ $[]$ $[]$
$200$ $[ 0, 0, 4 ]$ $5$ $8$ $[ 1, -9, -4, 16 ]$ $7625$ $[]$ $[]$ $[]$
$200$ $[ 1, 1, 1 ]$ $20$ $8$ $[ 0, -5, 0, 5 ]$ $2000$ $[ 5 ]$ $[ 16 ]$ $[]$
$200$ $[ 1, 1, 1 ]$ $20$ $8$ $[ 0, -5, 0, 5 ]$ $2000$ $[ 5 ]$ $[ 16 ]$ $[]$
$202$ $[ 0 ]$ $1$ $4$ $[ 1, -8, 1, 8 ]$ $10273$ $[]$ $[]$ $[]$
$202$ $[ 50 ]$ $2$ $8$ $[ 1, -7, -2, 1 ]$ $43449$ $[]$ $[]$ $[]$
$205$ $[ 2, 10 ]$ $4$ $8$ $[ 0, -7, 0, 11 ]$ $4400$ $[ 41 ]$ $[ 1 ]$ $[]$
$205$ $[ 2, 10 ]$ $4$ $8$ $[ 0, -7, 0, 11 ]$ $4400$ $[ 41 ]$ $[ 1 ]$ $[]$
$205$ $[ 2, 10 ]$ $4$ $16$ $[ 1, -8, -4, 12 ]$ $20308$ $[ 205, 41 ]$ $[ 1, 4 ]$ $[]$
$206$ $[ 0 ]$ $1$ $4$ $[ -2, -5, 12, -5 ]$ $5744$ $[]$ $[]$ $[]$
$206$ $[ 34 ]$ $3$ $8$ $[ 1, -8, -9, -2 ]$ $26873$ $[]$ $[]$ $[]$
$208$ $[ 0, 0, 2 ]$ $6$ $8$ $[ 2, -13, -38, -26 ]$ $13968$ $[]$ $[]$ $[]$
$209$ $[ 5, 3 ]$ $6$ $16$ $[ -13, 44, -17, 1 ]$ $32625$ $[ -11, 209 ]$ $[ 1, 16 ]$ $[]$
$210$ $[ 0, 1, 3 ]$ $4$ $8$ $[ 2, -9, -20, -2 ]$ $10304$ $[]$ $[]$ $[ 5, 7 ]$
$210$ $[ 0, 1, 3 ]$ $4$ $8$ $[ 2, -9, -20, -2 ]$ $10304$ $[]$ $[]$ $[ 5, 7 ]$
$213$ $[ 0, 0 ]$ $1$ $4$ $[ -3, -2, 7, 1 ]$ $2225$ $[]$ $[]$ $[]$
$216$ $[ 0, 1, 6 ]$ $6$ $16$ $[ 1, -8, -6, 6 ]$ $177300$ $[ 8 ]$ $[ 9 ]$ $[]$
$216$ $[ 1, 1, 3 ]$ $6$ $16$ $[ 1, -12, 10, -2 ]$ $29268$ $[ 24, -8 ]$ $[ 1, 516 ]$ $[]$
$217$ $[ 0, 0 ]$ $1$ $4$ $[ 0, -5, 1, 1 ]$ $6809$ $[]$ $[]$ $[]$
$219$ $[ 0, 0 ]$ $1$ $4$ $[ 1, -5, -3, 4 ]$ $8468$ $[]$ $[]$ $[]$
$219$ $[ 0, 18 ]$ $4$ $8$ $[ 0, -6, -2, 2 ]$ $21200$ $[ 73 ]$ $[ 4 ]$ $[]$
$220$ $[ 0, 0, 2 ]$ $5$ $8$ $[ -2, -6, 7, 11 ]$ $5125$ $[]$ $[]$ $[]$
$220$ $[ 0, 0, 2 ]$ $5$ $8$ $[ -2, -4, 5, 5 ]$ $2525$ $[]$ $[]$ $[]$
$220$ $[ 1, 0, 1 ]$ $10$ $8$ $[ -8, 14, -7, 1 ]$ $1125$ $[ 44 ]$ $[ 1 ]$ $[]$
$220$ $[ 1, 0, 1 ]$ $10$ $8$ $[ -8, 14, -7, 1 ]$ $1125$ $[ 44 ]$ $[ 1 ]$ $[]$
$221$ $[ 0, 4 ]$ $4$ $8$ $[ 0, -5, 0, 3 ]$ $8112$ $[ 17 ]$ $[ 1 ]$ $[]$
$222$ $[ 0, 6 ]$ $6$ $8$ $[ 0, -7, 0, 4 ]$ $17424$ $[]$ $[]$ $[]$
$223$ $[ 0 ]$ $1$ $4$ $[ 4, 2, -5, -3 ]$ $1957$ $[]$ $[]$ $[]$
$224$ $[ 1, 0, 1 ]$ $6$ $16$ $[ 2, -7, -14, 1 ]$ $9792$ $[ -7 ]$ $[ 1 ]$ $[ 2, 2 ]$
$225$ $[ 2, 0 ]$ $3$ $8$ $[ -2, -4, 5, 3 ]$ $11661$ $[]$ $[]$ $[]$
$225$ $[ 2, 0 ]$ $3$ $8$ $[ -2, -4, 5, 3 ]$ $11661$ $[]$ $[]$ $[]$
$225$ $[ 2, 10 ]$ $6$ $16$ $[ -2, -4, 5, 3 ]$ $11661$ $[]$ $[]$ $[]$
$225$ $[ 0, 4 ]$ $5$ $8$ $[ -2, -6, 7, 11 ]$ $5125$ $[]$ $[]$ $[]$
$225$ $[ 0, 2 ]$ $10$ $8$ $[ -11, 36, -31, 1 ]$ $7625$ $[ 5 ]$ $[ 1 ]$ $[]$
$225$ $[ 1, 5 ]$ $12$ $16$ $[ -10, 30, -28, 1 ]$ $54864$ $[ -3, -15 ]$ $[ 1, 4 ]$ $[]$
$225$ $[ 1, 5 ]$ $12$ $32$ $[ 0, -6, -3, 3 ]$ $9909$ $[ -15 ]$ $[ 81 ]$ $[]$
$226$ $[ 0 ]$ $1$ $4$ $[ 0, -5, 0, 5 ]$ $2000$ $[]$ $[]$ $[]$
$226$ $[ 28 ]$ $4$ $8$ $[ 1, -4, -2, 1 ]$ $3981$ $[ 113 ]$ $[ 16 ]$ $[]$
$230$ $[ 2, 0 ]$ $2$ $8$ $[ 0, -7, 0, 2 ]$ $13448$ $[]$ $[]$ $[]$
$230$ $[ 1, 11 ]$ $4$ $8$ $[ 2, -3, -2, 1 ]$ $2624$ $[]$ $[]$ $[ 5, 23 ]$
$230$ $[ 1, 11 ]$ $4$ $8$ $[ 2, -3, -2, 1 ]$ $2624$ $[]$ $[]$ $[ 5, 23 ]$
$231$ $[ 0, 3, 5 ]$ $2$ $16$ $[ -11, 36, -31, 1 ]$ $7625$ $[ -11 ]$ $[ 1 ]$ $[]$
$231$ $[ 0, 2, 0 ]$ $3$ $8$ $[ 2, -4, -6, 2 ]$ $11344$ $[]$ $[]$ $[]$
$231$ $[ 0, 0, 2 ]$ $5$ $8$ $[ 1, -3, -1, 1 ]$ $725$ $[]$ $[]$ $[]$
$232$ $[ 0, 0, 14 ]$ $2$ $8$ $[ -1, -5, 3, 4 ]$ $8468$ $[]$ $[]$ $[]$
$234$ $[ 2, 0 ]$ $3$ $8$ $[ 1, -6, -5, 4 ]$ $22545$ $[]$ $[]$ $[]$
$234$ $[ 3, 1 ]$ $12$ $16$ $[ 2, -12, -28, -2 ]$ $39744$ $[ -39 ]$ $[ 9 ]$ $[]$
$237$ $[ 0, 0 ]$ $1$ $4$ $[ 3, -1, -5, 1 ]$ $1957$ $[]$ $[]$ $[]$
$238$ $[ 0, 4 ]$ $4$ $8$ $[ 0, -4, 0, 2 ]$ $2048$ $[]$ $[]$ $[]$
$238$ $[ 0, 4 ]$ $4$ $8$ $[ 2, -3, -6, -1 ]$ $2624$ $[]$ $[]$ $[]$
$238$ $[ 2, 4 ]$ $12$ $16$ $[ 0, -6, 2, 5 ]$ $6224$ $[ 17 ]$ $[ 4 ]$ $[]$
$238$ $[ 2, 2 ]$ $24$ $16$ $[ -8, 16, -8, 1 ]$ $4352$ $[ 17 ]$ $[ 1 ]$ $[]$
$240$ $[ 0, 1, 0, 0 ]$ $4$ $8$ $[ 0, -6, 4, 2 ]$ $4352$ $[]$ $[]$ $[]$
$244$ $[ 0, 0 ]$ $1$ $4$ $[ -1, -9, 9, -2 ]$ $20308$ $[]$ $[]$ $[]$
$244$ $[ 0, 20 ]$ $3$ $8$ $[ -1, -9, -5, 3 ]$ $56749$ $[]$ $[]$ $[]$
$246$ $[ 0, 10 ]$ $4$ $8$ $[ 2, -5, -8, 2 ]$ $7232$ $[]$ $[]$ $[]$
$246$ $[ 0, 10 ]$ $4$ $8$ $[ 2, -7, -8, 14 ]$ $16448$ $[]$ $[]$ $[]$
$246$ $[ 0, 8 ]$ $5$ $8$ $[ 4, -27, -62, 164 ]$ $93025$ $[]$ $[]$ $[]$
$246$ $[ 1, 5 ]$ $8$ $8$ $[ 0, -6, 0, 7 ]$ $7168$ $[ 41 ]$ $[ 4 ]$ $[]$
$246$ $[ 1, 5 ]$ $8$ $8$ $[ 0, -6, 0, 7 ]$ $7168$ $[ 41 ]$ $[ 4 ]$ $[]$
$246$ $[ 0, 4 ]$ $10$ $8$ $[ 0, -5, 0, 5 ]$ $2000$ $[]$ $[]$ $[]$
$247$ $[ 0, 0 ]$ $1$ $4$ $[ 3, -2, -9, -4 ]$ $6809$ $[]$ $[]$ $[]$
$247$ $[ 2, 6 ]$ $6$ $8$ $[ -1, -7, -2, 1 ]$ $24417$ $[ 13 ]$ $[ 1 ]$ $[]$
$247$ $[ 10, 6 ]$ $6$ $8$ $[ -1, -7, -2, 1 ]$ $24417$ $[ 13 ]$ $[ 1 ]$ $[]$
$249$ $[ 0, 0 ]$ $1$ $4$ $[ -2, -4, 8, -1 ]$ $6224$ $[]$ $[]$ $[]$
$250$ $[ 20 ]$ $5$ $8$ $[ 0, -5, 0, 5 ]$ $2000$ $[]$ $[]$ $[]$
$250$ $[ 20 ]$ $5$ $8$ $[ 1, -9, -4, 16 ]$ $7625$ $[]$ $[]$ $[]$
$250$ $[ 20 ]$ $5$ $8$ $[ 0, -5, 0, 5 ]$ $2000$ $[]$ $[]$ $[]$
$250$ $[ 10 ]$ $10$ $8$ $[ 0, -5, 0, 5 ]$ $2000$ $[]$ $[]$ $[]$
$250$ $[ 10 ]$ $10$ $16$ $[ 1, -9, -4, 16 ]$ $7625$ $[]$ $[]$ $[]$
$251$ $[ 0 ]$ $1$ $4$ $[ 2, -2, -3, 1 ]$ $725$ $[]$ $[]$ $[]$
$252$ $[ 0, 1, 3 ]$ $6$ $16$ $[ -1, -7, 4, 7 ]$ $69777$ $[ -3, 21 ]$ $[ 9, 9 ]$ $[]$
$252$ $[ 1, 0, 1 ]$ $6$ $8$ $[ 1, -7, -4, 10 ]$ $23724$ $[ 28 ]$ $[ 16 ]$ $[]$
$252$ $[ 1, 0, 1 ]$ $6$ $8$ $[ 1, -7, -4, 10 ]$ $23724$ $[ 28 ]$ $[ 16 ]$ $[]$
$254$ $[ 42 ]$ $3$ $8$ $[ 1, -5, -1, 1 ]$ $8957$ $[]$ $[]$ $[]$
$255$ $[ 0, 0, 0 ]$ $1$ $4$ $[ -1, -5, 2, 2 ]$ $13768$ $[]$ $[]$ $[]$
$255$ $[ 0, 0, 8 ]$ $2$ $8$ $[ 1, -6, -3, 1 ]$ $27004$ $[]$ $[]$ $[]$
$255$ $[ 0, 2, 8 ]$ $2$ $8$ $[ -10, 30, -27, 4 ]$ $68557$ $[ 85 ]$ $[ 4 ]$ $[]$
$255$ $[ 0, 2, 8 ]$ $2$ $8$ $[ -10, 30, -27, 4 ]$ $68557$ $[ 85 ]$ $[ 4 ]$ $[]$
$255$ $[ 0, 0, 4 ]$ $4$ $8$ $[ 2, -3, -2, 1 ]$ $2624$ $[]$ $[]$ $[]$
$256$ $[ 0, 8 ]$ $8$ $8$ $[ 0, -6, -4, 2 ]$ $4352$ $[ 8 ]$ $[ 4 ]$ $[]$
$256$ $[ 0, 8 ]$ $8$ $8$ $[ 0, -6, -4, 2 ]$ $4352$ $[ 8 ]$ $[ 4 ]$ $[]$
$258$ $[ 1, 7 ]$ $6$ $8$ $[ -1, -12, -10, -2 ]$ $29268$ $[ 129 ]$ $[ 4 ]$ $[]$
$258$ $[ 1, 7 ]$ $6$ $8$ $[ -1, -12, -10, -2 ]$ $29268$ $[ 129 ]$ $[ 4 ]$ $[]$
$259$ $[ 0, 0 ]$ $1$ $4$ $[ 0, -9, 1, 17 ]$ $26825$ $[]$ $[]$ $[]$
$259$ $[ 0, 0 ]$ $1$ $4$ $[ -1, -6, 5, 4 ]$ $22545$ $[]$ $[]$ $[]$
$259$ $[ 0, 18 ]$ $2$ $8$ $[ 0, -7, -1, 2 ]$ $53121$ $[ 37 ]$ $[ 16 ]$ $[]$
$259$ $[ 0, 18 ]$ $2$ $8$ $[ -12, 45, -61, 25 ]$ $94441$ $[ 37 ]$ $[ 16 ]$ $[]$
$259$ $[ 2, 6 ]$ $6$ $8$ $[ -12, 46, -63, 25 ]$ $88821$ $[ 37 ]$ $[ 1 ]$ $[]$
$259$ $[ 4, 6 ]$ $6$ $8$ $[ -12, 46, -63, 25 ]$ $88821$ $[ 37 ]$ $[ 1 ]$ $[]$
$260$ $[ 0, 1, 3 ]$ $4$ $8$ $[ -2, -6, 2, 2 ]$ $44496$ $[ 65 ]$ $[ 4 ]$ $[]$
$260$ $[ 0, 3, 3 ]$ $4$ $8$ $[ -2, -6, 2, 2 ]$ $44496$ $[ 65 ]$ $[ 4 ]$ $[]$
$260$ $[ 0, 0, 2 ]$ $6$ $8$ $[ 2, -3, -4, 1 ]$ $4752$ $[]$ $[]$ $[]$
$260$ $[ 0, 2, 2 ]$ $6$ $16$ $[ -1, -15, 26, -8 ]$ $321597$ $[ 65, 13 ]$ $[ 4, 1024 ]$ $[]$
$264$ $[ 0, 0, 0, 2 ]$ $5$ $8$ $[ -2, -11, 12, 31 ]$ $13625$ $[]$ $[]$ $[]$
$265$ $[ 0, 0 ]$ $1$ $4$ $[ 2, -15, -48, -34 ]$ $7232$ $[]$ $[]$ $[]$
$265$ $[ 2, 26 ]$ $2$ $16$ $[ 1, -12, -6, 34 ]$ $111028$ $[ 53, 265 ]$ $[ 1, 1104 ]$ $[]$
$266$ $[ 3, 9 ]$ $2$ $16$ $[ -1, -26, -19, -3 ]$ $759989$ $[ -7 ]$ $[ 18 ]$ $[]$
$266$ $[ 0, 6 ]$ $3$ $8$ $[ -1, -10, 8, 4 ]$ $151572$ $[]$ $[]$ $[]$
$267$ $[ 0, 0 ]$ $1$ $4$ $[ -1, -7, 6, 7 ]$ $23377$ $[]$ $[]$ $[]$
$270$ $[ 3, 1 ]$ $12$ $16$ $[ 0, -6, 0, 3 ]$ $27648$ $[ -15 ]$ $[ 25 ]$ $[]$
$271$ $[ 18 ]$ $15$ $8$ $[ -9, 26, -24, 1 ]$ $1125$ $[]$ $[]$ $[]$
$272$ $[ 1, 0, 1 ]$ $16$ $16$ $[ 0, -4, 0, 2 ]$ $2048$ $[ -4, -68 ]$ $[ 514, 578 ]$ $[]$
$273$ $[ 0, 0, 0 ]$ $1$ $4$ $[ -1, -6, 4, 6 ]$ $17428$ $[]$ $[]$ $[]$
$273$ $[ 0, 0, 6 ]$ $2$ $8$ $[ 1, -7, -7, 2 ]$ $64436$ $[]$ $[]$ $[]$
$273$ $[ 0, 2, 0 ]$ $3$ $8$ $[ -1, -7, 4, 7 ]$ $69777$ $[]$ $[]$ $[]$
$273$ $[ 0, 2, 6 ]$ $6$ $16$ $[ -12, 37, -23, 4 ]$ $25961$ $[ 13 ]$ $[ 4 ]$ $[]$
$273$ $[ 0, 2, 6 ]$ $6$ $16$ $[ -11, 33, -30, 1 ]$ $169537$ $[ 13 ]$ $[ 1 ]$ $[]$
$274$ $[ 34 ]$ $4$ $8$ $[ 2, -16, -18, 39 ]$ $186576$ $[ 137 ]$ $[ 81 ]$ $[]$
$275$ $[ 0, 2 ]$ $5$ $8$ $[ -7, 13, -7, 1 ]$ $725$ $[]$ $[]$ $[]$
$275$ $[ 0, 2 ]$ $5$ $8$ $[ -2, -4, 5, 5 ]$ $2525$ $[]$ $[]$ $[]$
$275$ $[ 0, 2 ]$ $5$ $16$ $[ -9, 27, -31, 11 ]$ $725$ $[ 5 ]$ $[ 11 ]$ $[]$
$275$ $[ 10, 2 ]$ $10$ $16$ $[ -7, 13, -7, 1 ]$ $725$ $[]$ $[]$ $[]$
$275$ $[ 10, 2 ]$ $10$ $16$ $[ -2, -4, 5, 5 ]$ $2525$ $[]$ $[]$ $[]$
$275$ $[ 5, 1 ]$ $20$ $32$ $[ -11, 39, -45, 5 ]$ $2525$ $[ 5, -55 ]$ $[ 5, 25 ]$ $[]$
$278$ $[ 46 ]$ $3$ $8$ $[ 0, -5, 0, 3 ]$ $8112$ $[]$ $[]$ $[]$
$279$ $[ 0, 6 ]$ $5$ $8$ $[ -9, 26, -24, 1 ]$ $1125$ $[]$ $[]$ $[]$
$280$ $[ 0, 1, 0, 0 ]$ $2$ $8$ $[ 0, -6, 4, 2 ]$ $4352$ $[]$ $[]$ $[]$
$282$ $[ 1, 23 ]$ $2$ $16$ $[ 2, -4, -2, 2 ]$ $6224$ $[ -47 ]$ $[ 2084 ]$ $[ 3, 3 ]$
$286$ $[ 2, 0 ]$ $5$ $8$ $[ -2, -4, 5, 5 ]$ $2525$ $[]$ $[]$ $[]$
$288$ $[ 0, 4, 2 ]$ $6$ $16$ $[ 1, -8, -6, 6 ]$ $177300$ $[ 8 ]$ $[ 9 ]$ $[]$
$288$ $[ 1, 4, 1 ]$ $6$ $16$ $[ 1, -12, 10, -2 ]$ $29268$ $[ 24, -8 ]$ $[ 1, 516 ]$ $[]$
$288$ $[ 0, 1, 0 ]$ $8$ $8$ $[ 0, -6, -4, 2 ]$ $4352$ $[ 8 ]$ $[ 4 ]$ $[]$
$290$ $[ 2, 14 ]$ $2$ $8$ $[ -2, -6, 9, 4 ]$ $68557$ $[ 145 ]$ $[ 64 ]$ $[]$
$290$ $[ 2, 14 ]$ $2$ $8$ $[ -2, -6, 9, 4 ]$ $68557$ $[ 145 ]$ $[ 64 ]$ $[]$
$290$ $[ 1, 7 ]$ $4$ $8$ $[ 2, -9, -10, 17 ]$ $20032$ $[]$ $[]$ $[ 5, 29 ]$
$290$ $[ 3, 7 ]$ $4$ $8$ $[ 2, -9, -10, 17 ]$ $20032$ $[]$ $[]$ $[ 5, 29 ]$
$291$ $[ 0, 32 ]$ $3$ $8$ $[ 1, -8, -8, 6 ]$ $87092$ $[]$ $[]$ $[]$
$291$ $[ 0, 8 ]$ $12$ $8$ $[ 0, -4, 0, 1 ]$ $2304$ $[]$ $[]$ $[]$
$292$ $[ 0, 0 ]$ $1$ $4$ $[ -3, -5, 16, -8 ]$ $13448$ $[]$ $[]$ $[]$
$292$ $[ 0, 18 ]$ $4$ $8$ $[ 2, -7, -10, 7 ]$ $13888$ $[]$ $[]$ $[]$
$296$ $[ 0, 0, 0 ]$ $1$ $4$ $[ -2, -8, 15, 4 ]$ $48389$ $[]$ $[]$ $[]$
$297$ $[ 6, 0 ]$ $3$ $8$ $[ -1, -6, 5, 4 ]$ $22545$ $[]$ $[]$ $[]$
$297$ $[ 3, 5 ]$ $6$ $16$ $[ -15, 75, -144, 93 ]$ $22545$ $[ 33, -11 ]$ $[ 1, 93 ]$ $[]$
$299$ $[ 6, 0 ]$ $2$ $8$ $[ -14, 66, -119, 64 ]$ $101277$ $[ 13 ]$ $[ 4 ]$ $[]$
$299$ $[ 4, 0 ]$ $3$ $8$ $[ 1, -7, -3, 9 ]$ $4525$ $[]$ $[]$ $[]$
$300$ $[ 0, 0, 4 ]$ $5$ $8$ $[ -2, -6, 7, 11 ]$ $5125$ $[]$ $[]$ $[]$
$300$ $[ 0, 0, 4 ]$ $5$ $8$ $[ -24, 26, -9, 1 ]$ $1125$ $[]$ $[]$ $[]$
$301$ $[ 0, 0 ]$ $1$ $4$ $[ 4, 2, -5, -3 ]$ $1957$ $[]$ $[]$ $[]$
$302$ $[ 0 ]$ $1$ $4$ $[ -2, -12, -8, 3 ]$ $25808$ $[]$ $[]$ $[]$
$302$ $[ 0 ]$ $1$ $4$ $[ -2, -4, 8, -1 ]$ $6224$ $[]$ $[]$ $[]$
$302$ $[ 50 ]$ $3$ $8$ $[ 1, -8, -6, 11 ]$ $5725$ $[]$ $[]$ $[]$
$302$ $[ 10 ]$ $15$ $8$ $[ -9, 26, -24, 1 ]$ $1125$ $[]$ $[]$ $[]$
$305$ $[ 0, 0 ]$ $1$ $4$ $[ 3, -1, -6, -1 ]$ $2777$ $[]$ $[]$ $[]$
$305$ $[ 2, 0 ]$ $2$ $8$ $[ -9, 17, -8, 1 ]$ $7537$ $[ 5 ]$ $[ 1 ]$ $[]$
$305$ $[ 0, 20 ]$ $3$ $8$ $[ -10, 25, -21, 4 ]$ $2777$ $[]$ $[]$ $[]$
$305$ $[ 0, 2 ]$ $30$ $8$ $[ -7, 14, -8, 1 ]$ $1125$ $[ 61 ]$ $[ 1 ]$ $[]$
$306$ $[ 2, 0 ]$ $3$ $8$ $[ -1, -8, 3, 12 ]$ $72153$ $[]$ $[]$ $[]$
$306$ $[ 2, 8 ]$ $6$ $16$ $[ 0, -23, 51, -23 ]$ $142929$ $[ 17 ]$ $[ 16 ]$ $[]$
$306$ $[ 2, 8 ]$ $6$ $16$ $[ 1, -23, -12, 123 ]$ $508401$ $[ 17 ]$ $[ 16 ]$ $[]$
$306$ $[ 0, 2 ]$ $8$ $8$ $[ 0, -4, 0, 2 ]$ $2048$ $[]$ $[]$ $[]$
$306$ $[ 0, 2 ]$ $8$ $8$ $[ 0, -4, 0, 2 ]$ $2048$ $[]$ $[]$ $[]$
$306$ $[ 3, 1 ]$ $16$ $8$ $[ 0, -4, 0, 2 ]$ $2048$ $[ 17 ]$ $[ 2 ]$ $[]$
$306$ $[ 3, 1 ]$ $16$ $8$ $[ 0, -4, 0, 2 ]$ $2048$ $[ 17 ]$ $[ 2 ]$ $[]$
$306$ $[ 3, 1 ]$ $16$ $8$ $[ 0, -4, 0, 2 ]$ $2048$ $[ 17 ]$ $[ 32 ]$ $[]$
$306$ $[ 3, 1 ]$ $16$ $8$ $[ 0, -4, 0, 2 ]$ $2048$ $[ 17 ]$ $[ 2 ]$ $[]$
$306$ $[ 3, 1 ]$ $16$ $8$ $[ 0, -4, 0, 2 ]$ $2048$ $[ 17 ]$ $[ 32 ]$ $[]$
$306$ $[ 3, 1 ]$ $16$ $8$ $[ 0, -4, 0, 2 ]$ $2048$ $[ 17 ]$ $[ 2 ]$ $[]$
$308$ $[ 0, 1, 5 ]$ $6$ $16$ $[ -14, 51, -17, 1 ]$ $299241$ $[ -7, -11 ]$ $[ 1, 9604 ]$ $[]$
$308$ $[ 0, 0, 2 ]$ $5$ $8$ $[ -2, -18, 19, 89 ]$ $37525$ $[]$ $[]$ $[]$
$308$ $[ 0, 2, 2 ]$ $15$ $16$ $[ -1, -3, 1, 1 ]$ $725$ $[]$ $[]$ $[]$
$310$ $[ 2, 0 ]$ $2$ $8$ $[ 2, -4, -4, 3 ]$ $11344$ $[]$ $[]$ $[]$
$310$ $[ 2, 0 ]$ $2$ $8$ $[ 0, -6, 2, 5 ]$ $6224$ $[]$ $[]$ $[]$
$310$ $[ 0, 6 ]$ $5$ $8$ $[ 2, -5, -6, -1 ]$ $1600$ $[]$ $[]$ $[]$
$310$ $[ 0, 6 ]$ $5$ $8$ $[ 0, -7, 0, 11 ]$ $4400$ $[]$ $[]$ $[]$
$310$ $[ 0, 6 ]$ $5$ $8$ $[ 2, -5, -6, -1 ]$ $1600$ $[]$ $[]$ $[]$
$310$ $[ 0, 2 ]$ $15$ $8$ $[ -1, -4, 4, 1 ]$ $1125$ $[]$ $[]$ $[]$
$310$ $[ 0, 2 ]$ $15$ $8$ $[ -24, 206, -729, 841 ]$ $1125$ $[]$ $[]$ $[]$
$310$ $[ 0, 2 ]$ $15$ $8$ $[ 1, -4, -4, 1 ]$ $1125$ $[]$ $[]$ $[]$
$310$ $[ 0, 2 ]$ $15$ $8$ $[ 1, -4, -4, 1 ]$ $1125$ $[]$ $[]$ $[]$
$310$ $[ 0, 2 ]$ $15$ $8$ $[ -24, 206, -729, 841 ]$ $1125$ $[]$ $[]$ $[]$
$310$ $[ 0, 2 ]$ $15$ $8$ $[ 1, -4, -4, 1 ]$ $1125$ $[]$ $[]$ $[]$
$311$ $[ 0 ]$ $1$ $4$ $[ 1, -3, -1, 1 ]$ $725$ $[]$ $[]$ $[]$
$312$ $[ 0, 1, 0, 0 ]$ $2$ $8$ $[ 0, -5, 0, 5 ]$ $2000$ $[]$ $[]$ $[]$
$312$ $[ 0, 0, 1, 3 ]$ $4$ $16$ $[ 2, -7, 0, 2 ]$ $46912$ $[ -39 ]$ $[ 16 ]$ $[]$
$312$ $[ 0, 0, 0, 2 ]$ $6$ $8$ $[ 0, -9, 6, 6 ]$ $25488$ $[]$ $[]$ $[]$
$315$ $[ 2, 0, 0 ]$ $3$ $8$ $[ -1, -4, 2, 1 ]$ $3981$ $[]$ $[]$ $[]$
$315$ $[ 2, 0, 0 ]$ $3$ $8$ $[ -9, 26, -24, 1 ]$ $1125$ $[]$ $[]$ $[]$
$315$ $[ 0, 1, 3 ]$ $4$ $16$ $[ 2, -4, -4, 2 ]$ $14656$ $[ -35 ]$ $[ 1 ]$ $[]$
$315$ $[ 0, 2, 2 ]$ $6$ $16$ $[ 2, -4, -4, 3 ]$ $11344$ $[]$ $[]$ $[]$
$319$ $[ 0, 0 ]$ $1$ $4$ $[ 2, -3, -3, 2 ]$ $2777$ $[]$ $[]$ $[]$
$320$ $[ 0, 4, 2 ]$ $4$ $16$ $[ 2, -3, -4, 1 ]$ $4752$ $[ 8, 40 ]$ $[ 4, 4 ]$ $[]$
$322$ $[ 2, 0 ]$ $3$ $8$ $[ -1, -6, 6, 3 ]$ $14013$ $[]$ $[]$ $[]$
$322$ $[ 2, 0 ]$ $3$ $8$ $[ -1, -5, 1, 1 ]$ $8957$ $[]$ $[]$ $[]$
$322$ $[ 2, 0 ]$ $3$ $8$ $[ 0, -4, -1, 1 ]$ $1957$ $[]$ $[]$ $[]$
$322$ $[ 2, 0 ]$ $3$ $8$ $[ 1, -6, -6, 3 ]$ $14013$ $[]$ $[]$ $[]$
$322$ $[ 1, 11 ]$ $6$ $16$ $[ -1, -17, 39, -15 ]$ $140949$ $[ -7, -23 ]$ $[ 1, 9 ]$ $[]$
$322$ $[ 1, 11 ]$ $6$ $16$ $[ -17, 77, -97, 25 ]$ $1432341$ $[ 161, -23 ]$ $[ 25, 49 ]$ $[]$
$323$ $[ 0, 0 ]$ $1$ $4$ $[ 0, -6, -1, 7 ]$ $1957$ $[]$ $[]$ $[]$
$325$ $[ 5, 3 ]$ $4$ $8$ $[ 2, -4, -2, 2 ]$ $6224$ $[ 65 ]$ $[ 4 ]$ $[]$
$325$ $[ 5, 3 ]$ $4$ $16$ $[ 1, -7, -4, 10 ]$ $23724$ $[ 13, 65 ]$ $[ 4, 4 ]$ $[]$
$325$ $[ 15, 3 ]$ $4$ $8$ $[ 2, -4, -2, 2 ]$ $6224$ $[ 65 ]$ $[ 4 ]$ $[]$
$325$ $[ 15, 3 ]$ $4$ $16$ $[ 1, -7, -4, 10 ]$ $23724$ $[ 13, 65 ]$ $[ 4, 4 ]$ $[]$
$325$ $[ 0, 2 ]$ $6$ $8$ $[ 2, -3, -4, 1 ]$ $4752$ $[]$ $[]$ $[]$
$325$ $[ 10, 2 ]$ $6$ $8$ $[ 2, -3, -4, 1 ]$ $4752$ $[]$ $[]$ $[]$
$325$ $[ 10, 2 ]$ $6$ $8$ $[ 2, -3, -4, 1 ]$ $4752$ $[]$ $[]$ $[]$
$325$ $[ 5, 1 ]$ $12$ $16$ $[ -10, 29, -20, 1 ]$ $34704$ $[ 13, 65 ]$ $[ 1, 1 ]$ $[]$
$325$ $[ 15, 1 ]$ $12$ $16$ $[ -10, 29, -20, 1 ]$ $34704$ $[ 13, 65 ]$ $[ 1, 1 ]$ $[]$
$325$ $[ 1, 3 ]$ $20$ $8$ $[ 0, -5, 0, 5 ]$ $2000$ $[ 65 ]$ $[ 1 ]$ $[]$
$325$ $[ 3, 3 ]$ $20$ $8$ $[ 0, -5, 0, 5 ]$ $2000$ $[ 65 ]$ $[ 1 ]$ $[]$
$327$ $[ 0, 54 ]$ $2$ $8$ $[ -9, 24, -19, 1 ]$ $31288$ $[ 109 ]$ $[ 1 ]$ $[]$
$328$ $[ 0, 0, 10 ]$ $4$ $8$ $[ -1, -6, 4, 6 ]$ $17428$ $[ 41 ]$ $[ 4 ]$ $[]$
$328$ $[ 0, 0, 8 ]$ $5$ $8$ $[ 1, -12, -3, 29 ]$ $13025$ $[]$ $[]$ $[]$
$329$ $[ 3, 23 ]$ $2$ $16$ $[ -10, 31, -31, 1 ]$ $2777$ $[ 329, -47 ]$ $[ 53, 1867 ]$ $[]$
$330$ $[ 1, 0, 5 ]$ $2$ $8$ $[ 1, -5, -4, 2 ]$ $11324$ $[]$ $[]$ $[ 3, 11 ]$
$330$ $[ 1, 0, 5 ]$ $2$ $8$ $[ 1, -5, -4, 2 ]$ $11324$ $[]$ $[]$ $[ 3, 11 ]$
$330$ $[ 0, 1, 5 ]$ $4$ $8$ $[ 0, -8, 0, 14 ]$ $14336$ $[]$ $[]$ $[ 5, 11 ]$
$330$ $[ 0, 1, 5 ]$ $4$ $8$ $[ 0, -8, 0, 14 ]$ $14336$ $[]$ $[]$ $[ 5, 11 ]$
$330$ $[ 0, 0, 2 ]$ $5$ $8$ $[ 1, -24, -19, 121 ]$ $52625$ $[]$ $[]$ $[]$
$330$ $[ 0, 0, 2 ]$ $5$ $8$ $[ 1, -21, -10, 100 ]$ $43025$ $[]$ $[]$ $[]$
$333$ $[ 0, 0 ]$ $1$ $4$ $[ 0, -6, -2, 5 ]$ $6224$ $[]$ $[]$ $[]$
$336$ $[ 0, 0, 1, 1 ]$ $6$ $16$ $[ 1, -10, -10, 4 ]$ $136908$ $[ -7 ]$ $[ 16 ]$ $[]$
$340$ $[ 0, 2, 0 ]$ $2$ $8$ $[ 2, -4, -4, 3 ]$ $11344$ $[]$ $[]$ $[]$
$341$ $[ 0, 0 ]$ $1$ $4$ $[ 1, -5, -2, 4 ]$ $2225$ $[]$ $[]$ $[]$
$342$ $[ 1, 3 ]$ $6$ $8$ $[ 0, -6, 0, 3 ]$ $27648$ $[ 57 ]$ $[ 1 ]$ $[]$
$342$ $[ 1, 3 ]$ $6$ $8$ $[ -1, -5, 3, 3 ]$ $12357$ $[ 57 ]$ $[ 9 ]$ $[]$
$342$ $[ 5, 3 ]$ $6$ $8$ $[ 0, -6, 0, 3 ]$ $27648$ $[ 57 ]$ $[ 1 ]$ $[]$
$342$ $[ 5, 3 ]$ $6$ $8$ $[ -1, -5, 3, 3 ]$ $12357$ $[ 57 ]$ $[ 9 ]$ $[]$
$344$ $[ 0, 0, 14 ]$ $3$ $8$ $[ 1, -10, 7, 3 ]$ $47032$ $[]$ $[]$ $[]$
$346$ $[ 0 ]$ $1$ $4$ $[ 2, -5, -5, -1 ]$ $2777$ $[]$ $[]$ $[]$
$348$ $[ 1, 1, 0 ]$ $2$ $16$ $[ 1, -9, -11, 2 ]$ $49708$ $[ -4 ]$ $[ 4 ]$ $[]$
$350$ $[ 4, 0 ]$ $5$ $8$ $[ -9, 26, -24, 1 ]$ $1125$ $[]$ $[]$ $[]$
$350$ $[ 5, 1 ]$ $12$ $16$ $[ 2, -15, -16, 46 ]$ $51264$ $[ -7 ]$ $[ 4 ]$ $[]$
$351$ $[ 9, 3 ]$ $4$ $16$ $[ -10, 27, -16, 1 ]$ $171504$ $[ -3, -39 ]$ $[ 1, 1 ]$ $[]$
$351$ $[ 9, 3 ]$ $4$ $16$ $[ -16, 87, -196, 157 ]$ $4752$ $[ -39, -3 ]$ $[ 16, 157 ]$ $[]$
$351$ $[ 0, 2 ]$ $6$ $8$ $[ -1, -4, 2, 1 ]$ $3981$ $[ 13 ]$ $[ 9 ]$ $[]$
$351$ $[ 0, 2 ]$ $6$ $8$ $[ -1, -4, 2, 1 ]$ $3981$ $[ 13 ]$ $[ 9 ]$ $[]$
$351$ $[ 9, 1 ]$ $12$ $16$ $[ 2, -3, -4, 1 ]$ $4752$ $[ -3, -39 ]$ $[ 4, 169 ]$ $[]$
$353$ $[ 88 ]$ $4$ $8$ $[ 2, -14, -8, 4 ]$ $226544$ $[ 353 ]$ $[ 1 ]$ $[]$
$355$ $[ 0, 0 ]$ $1$ $4$ $[ 2, -2, -3, 1 ]$ $725$ $[]$ $[]$ $[]$
$355$ $[ 0, 0 ]$ $1$ $4$ $[ 4, 2, -5, -3 ]$ $1957$ $[]$ $[]$ $[]$
$355$ $[ 2, 0 ]$ $2$ $8$ $[ -8, 16, -9, 1 ]$ $8069$ $[ 5 ]$ $[ 1 ]$ $[]$
$356$ $[ 0, 44 ]$ $2$ $8$ $[ -1, -11, 5, 18 ]$ $223604$ $[ 89 ]$ $[ 18 ]$ $[]$
$357$ $[ 0, 0, 0 ]$ $1$ $4$ $[ -2, -15, -16, -2 ]$ $7232$ $[]$ $[]$ $[]$
$357$ $[ 0, 2, 0 ]$ $3$ $8$ $[ 1, -6, -6, 3 ]$ $14013$ $[]$ $[]$ $[]$
$357$ $[ 0, 2, 0 ]$ $3$ $8$ $[ 1, -5, -1, 3 ]$ $9301$ $[]$ $[]$ $[]$
$358$ $[ 0 ]$ $1$ $4$ $[ 1, -6, -1, 1 ]$ $4913$ $[]$ $[]$ $[]$
$359$ $[ 0 ]$ $1$ $4$ $[ 2, -3, -5, 1 ]$ $2777$ $[]$ $[]$ $[]$
$360$ $[ 0, 0, 2, 0 ]$ $3$ $8$ $[ 1, -9, -5, 16 ]$ $29268$ $[]$ $[]$ $[]$
$362$ $[ 12 ]$ $15$ $8$ $[ 1, -4, -4, 1 ]$ $1125$ $[]$ $[]$ $[]$
$364$ $[ 0, 0, 6 ]$ $2$ $8$ $[ 1, -8, 2, 6 ]$ $25492$ $[]$ $[]$ $[]$
$364$ $[ 1, 3, 0 ]$ $2$ $16$ $[ -1, -8, 8, 4 ]$ $33428$ $[ -4 ]$ $[ 16 ]$ $[]$
$364$ $[ 0, 2, 0 ]$ $3$ $8$ $[ -1, -12, -10, -2 ]$ $29268$ $[]$ $[]$ $[]$
$366$ $[ 0, 12 ]$ $5$ $8$ $[ 1, -11, -10, 20 ]$ $10025$ $[]$ $[]$ $[]$
$366$ $[ 0, 12 ]$ $5$ $8$ $[ -2, -13, 14, 44 ]$ $19025$ $[]$ $[]$ $[]$
$366$ $[ 0, 6 ]$ $10$ $8$ $[ 0, -5, 0, 5 ]$ $2000$ $[]$ $[]$ $[]$
$366$ $[ 0, 6 ]$ $10$ $8$ $[ 0, -5, 0, 5 ]$ $2000$ $[]$ $[]$ $[]$
$369$ $[ 0, 20 ]$ $2$ $8$ $[ 1, -6, -6, 2 ]$ $22676$ $[]$ $[]$ $[]$
$369$ $[ 0, 8 ]$ $5$ $8$ $[ -1, -3, 1, 1 ]$ $725$ $[]$ $[]$ $[]$
$369$ $[ 0, 4 ]$ $10$ $8$ $[ -9, 26, -24, 1 ]$ $1125$ $[ 41 ]$ $[ 1 ]$ $[]$
$370$ $[ 1, 3 ]$ $12$ $8$ $[ 0, -4, 0, 1 ]$ $2304$ $[]$ $[]$ $[ 5, 37 ]$
$370$ $[ 3, 3 ]$ $12$ $8$ $[ 0, -4, 0, 1 ]$ $2304$ $[]$ $[]$ $[ 5, 37 ]$
$372$ $[ 0, 0, 6 ]$ $5$ $8$ $[ -1, -10, 2, 19 ]$ $8725$ $[]$ $[]$ $[]$
$374$ $[ 0, 0 ]$ $1$ $4$ $[ -1, -10, 9, 16 ]$ $55585$ $[]$ $[]$ $[]$
$374$ $[ 0, 0 ]$ $1$ $4$ $[ -1, -10, 13, -4 ]$ $17417$ $[]$ $[]$ $[]$
$374$ $[ 0, 8 ]$ $2$ $8$ $[ 2, -5, -1, 2 ]$ $10273$ $[]$ $[]$ $[]$
$374$ $[ 2, 0 ]$ $5$ $8$ $[ -2, -4, 5, 5 ]$ $2525$ $[]$ $[]$ $[]$
$374$ $[ 2, 0 ]$ $5$ $8$ $[ -1, -3, 1, 1 ]$ $725$ $[]$ $[]$ $[]$
$374$ $[ 0, 2 ]$ $8$ $8$ $[ 0, -4, 0, 2 ]$ $2048$ $[]$ $[]$ $[]$
$375$ $[ 0, 0 ]$ $1$ $4$ $[ 3, -3, -11, -1 ]$ $2525$ $[]$ $[]$ $[]$
$375$ $[ 0, 0 ]$ $1$ $4$ $[ -3, -3, 11, -1 ]$ $2525$ $[]$ $[]$ $[]$
$375$ $[ 0, 50 ]$ $2$ $8$ $[ -2, -4, 5, 5 ]$ $2525$ $[]$ $[]$ $[]$
$375$ $[ 0, 20 ]$ $5$ $8$ $[ -2, -6, 7, 11 ]$ $5125$ $[]$ $[]$ $[]$
$375$ $[ 0, 10 ]$ $10$ $16$ $[ -2, -6, 7, 11 ]$ $5125$ $[]$ $[]$ $[]$
$376$ $[ 0, 0, 0 ]$ $1$ $4$ $[ -3, -5, 16, -8 ]$ $13448$ $[]$ $[]$ $[]$
$376$ $[ 0, 0, 0 ]$ $1$ $4$ $[ 1, -9, -4, 16 ]$ $7625$ $[]$ $[]$ $[]$
$378$ $[ 3, 3 ]$ $6$ $16$ $[ 2, -13, -38, -26 ]$ $13968$ $[ -7 ]$ $[ 36 ]$ $[]$
$380$ $[ 0, 0, 6 ]$ $3$ $8$ $[ -1, -8, 3, 2 ]$ $133593$ $[]$ $[]$ $[]$
$382$ $[ 0 ]$ $1$ $4$ $[ -3, -2, 9, -4 ]$ $6809$ $[]$ $[]$ $[]$
$384$ $[ 0, 8, 0 ]$ $4$ $8$ $[ 0, -6, -4, 2 ]$ $4352$ $[]$ $[]$ $[]$
$384$ $[ 0, 8, 0 ]$ $4$ $8$ $[ 0, -6, -4, 2 ]$ $4352$ $[]$ $[]$ $[]$
$385$ $[ 0, 0, 0 ]$ $1$ $4$ $[ 1, -5, -1, 2 ]$ $11348$ $[]$ $[]$ $[]$
$385$ $[ 0, 2, 0 ]$ $3$ $8$ $[ 0, -4, -1, 1 ]$ $1957$ $[]$ $[]$ $[]$
$385$ $[ 0, 0, 2 ]$ $5$ $8$ $[ 1, -3, -1, 1 ]$ $725$ $[]$ $[]$ $[]$
$386$ $[ 96 ]$ $2$ $8$ $[ 0, -7, 0, 3 ]$ $65712$ $[ 193 ]$ $[ 48 ]$ $[]$
$386$ $[ 64 ]$ $3$ $8$ $[ 0, -6, 0, 7 ]$ $7168$ $[]$ $[]$ $[]$
$386$ $[ 64 ]$ $3$ $8$ $[ 0, -6, -2, 2 ]$ $21200$ $[]$ $[]$ $[]$
$390$ $[ 0, 0, 2 ]$ $6$ $8$ $[ -2, -34, 8, 208 ]$ $131472$ $[]$ $[]$ $[]$
$390$ $[ 0, 1, 1 ]$ $12$ $8$ $[ 0, -4, 0, 1 ]$ $2304$ $[]$ $[]$ $[ 5, 13 ]$
$390$ $[ 0, 3, 1 ]$ $12$ $8$ $[ 0, -4, 0, 1 ]$ $2304$ $[]$ $[]$ $[ 5, 13 ]$
$393$ $[ 0, 0 ]$ $1$ $4$ $[ 1, -4, -2, 3 ]$ $1957$ $[]$ $[]$ $[]$
$393$ $[ 0, 0 ]$ $1$ $4$ $[ 3, 0, -4, -1 ]$ $725$ $[]$ $[]$ $[]$
$393$ $[ 0, 26 ]$ $5$ $8$ $[ -7, 13, -7, 1 ]$ $725$ $[]$ $[]$ $[]$
$394$ $[ 0 ]$ $1$ $4$ $[ 3, -2, -7, 1 ]$ $2225$ $[]$ $[]$ $[]$
$394$ $[ 0 ]$ $1$ $4$ $[ 1, -6, -1, 1 ]$ $4913$ $[]$ $[]$ $[]$
$395$ $[ 0, 0 ]$ $1$ $4$ $[ -1, -7, 6, -1 ]$ $10273$ $[]$ $[]$ $[]$
$396$ $[ 0, 2, 0 ]$ $3$ $8$ $[ 1, -8, -3, 12 ]$ $72153$ $[]$ $[]$ $[]$
$396$ $[ 0, 1, 5 ]$ $6$ $16$ $[ 1, -6, -5, 4 ]$ $22545$ $[ 33, -11 ]$ $[ 1, 6975 ]$ $[]$
$396$ $[ 0, 3, 1 ]$ $10$ $16$ $[ 0, -20, 0, 95 ]$ $38000$ $[ -11, -3 ]$ $[ 121, 361 ]$ $[]$
$399$ $[ 1, 0, 9 ]$ $2$ $16$ $[ -1, -7, 0, 2 ]$ $52396$ $[ -3 ]$ $[ 1 ]$ $[ 19, 19 ]$
$399$ $[ 0, 0, 6 ]$ $3$ $8$ $[ -11, 33, -15, 1 ]$ $1957$ $[]$ $[]$ $[]$
$399$ $[ 0, 0, 6 ]$ $3$ $8$ $[ -1, -5, 3, 1 ]$ $12197$ $[]$ $[]$ $[]$
$399$ $[ 0, 2, 0 ]$ $3$ $8$ $[ -8, 18, -9, 1 ]$ $1957$ $[]$ $[]$ $[]$
$399$ $[ 0, 2, 0 ]$ $3$ $8$ $[ -8, 18, -13, 1 ]$ $725$ $[]$ $[]$ $[]$
$399$ $[ 0, 2, 0 ]$ $3$ $8$ $[ -1, -6, 2, 1 ]$ $23301$ $[]$ $[]$ $[]$
$400$ $[ 0, 1, 0 ]$ $4$ $16$ $[ 2, -3, -4, 1 ]$ $4752$ $[ 8, 40 ]$ $[ 4, 4 ]$ $[]$
$400$ $[ 1, 1, 5 ]$ $4$ $16$ $[ 0, -7, 2, 1 ]$ $33424$ $[ 8, 40 ]$ $[ 4, 100 ]$ $[]$
$400$ $[ 1, 3, 5 ]$ $4$ $16$ $[ 0, -7, 2, 1 ]$ $33424$ $[ 8, 40 ]$ $[ 4, 100 ]$ $[]$
$400$ $[ 0, 0, 4 ]$ $5$ $8$ $[ 1, -9, -4, 16 ]$ $7625$ $[]$ $[]$ $[]$
$400$ $[ 0, 0, 4 ]$ $5$ $8$ $[ 1, -9, -4, 16 ]$ $7625$ $[]$ $[]$ $[]$
$400$ $[ 0, 0, 2 ]$ $10$ $8$ $[ -11, 36, -31, 1 ]$ $7625$ $[ 5 ]$ $[ 1 ]$ $[]$
$400$ $[ 0, 0, 2 ]$ $10$ $8$ $[ 0, -5, 0, 5 ]$ $2000$ $[]$ $[]$ $[]$
$400$ $[ 0, 0, 2 ]$ $10$ $8$ $[ -1, -9, 4, 16 ]$ $7625$ $[ 5 ]$ $[ 1 ]$ $[]$
$400$ $[ 1, 0, 1 ]$ $20$ $16$ $[ 0, -5, 0, 5 ]$ $2000$ $[ -20, -4 ]$ $[ 25, 205 ]$ $[]$
$400$ $[ 1, 0, 1 ]$ $20$ $16$ $[ 0, -5, 0, 5 ]$ $2000$ $[ -4, -20 ]$ $[ 305, 400 ]$ $[]$
$404$ $[ 0, 10 ]$ $10$ $8$ $[ 1, -3, -1, 1 ]$ $725$ $[ 101 ]$ $[ 25 ]$ $[]$
$404$ $[ 0, 10 ]$ $10$ $8$ $[ 1, -14, -14, 31 ]$ $15125$ $[ 101 ]$ $[ 1 ]$ $[]$
$405$ $[ 27, 1 ]$ $4$ $16$ $[ -10, 30, -28, 1 ]$ $54864$ $[ -3, -15 ]$ $[ 1, 4 ]$ $[]$
$406$ $[ 0, 0 ]$ $1$ $4$ $[ -1, -5, 1, 2 ]$ $11348$ $[]$ $[]$ $[]$
$406$ $[ 0, 14 ]$ $2$ $8$ $[ -1, -10, 8, 4 ]$ $151572$ $[]$ $[]$ $[]$
$407$ $[ 0, 0 ]$ $1$ $4$ $[ 1, -4, 0, 1 ]$ $1957$ $[]$ $[]$ $[]$
$407$ $[ 0, 0 ]$ $1$ $4$ $[ -1, -4, 2, 3 ]$ $1957$ $[]$ $[]$ $[]$
$407$ $[ 5, 3 ]$ $12$ $8$ $[ 2, -3, -4, 1 ]$ $4752$ $[ 37 ]$ $[ 1 ]$ $[ 11, 11 ]$
$407$ $[ 5, 3 ]$ $12$ $8$ $[ 2, -3, -4, 1 ]$ $4752$ $[ 37 ]$ $[ 1 ]$ $[ 11, 11 ]$
$410$ $[ 0, 10 ]$ $4$ $8$ $[ -1, -4, 1, 2 ]$ $2777$ $[ 41 ]$ $[ 16 ]$ $[]$
$410$ $[ 0, 8 ]$ $5$ $8$ $[ -2, -4, 5, 5 ]$ $2525$ $[]$ $[]$ $[]$
$410$ $[ 0, 4 ]$ $10$ $8$ $[ -8, 14, -7, 1 ]$ $1125$ $[ 41 ]$ $[ 1 ]$ $[]$
$412$ $[ 0, 0 ]$ $1$ $4$ $[ -1, -5, 2, 4 ]$ $2225$ $[]$ $[]$ $[]$
$416$ $[ 1, 0, 3 ]$ $4$ $8$ $[ -1, -8, 6, 4 ]$ $46952$ $[ 13 ]$ $[ 16 ]$ $[]$
$416$ $[ 1, 0, 3 ]$ $4$ $8$ $[ -1, -8, 6, 4 ]$ $46952$ $[ 13 ]$ $[ 16 ]$ $[]$
$416$ $[ 0, 4, 2 ]$ $6$ $16$ $[ -1, -43, 109, 64 ]$ $304596$ $[ 8, 13 ]$ $[ 4, 4 ]$ $[]$
$416$ $[ 1, 0, 1 ]$ $12$ $8$ $[ 2, -3, -4, 1 ]$ $4752$ $[ 13 ]$ $[ 9 ]$ $[]$
$416$ $[ 1, 0, 1 ]$ $12$ $8$ $[ 2, -3, -4, 1 ]$ $4752$ $[ 13 ]$ $[ 9 ]$ $[]$
$418$ $[ 5, 9 ]$ $2$ $8$ $[ 1, -6, -3, 1 ]$ $27004$ $[ 209 ]$ $[ 45 ]$ $[]$
$418$ $[ 5, 9 ]$ $2$ $8$ $[ 1, -6, -3, 1 ]$ $27004$ $[ 209 ]$ $[ 45 ]$ $[]$
$420$ $[ 1, 1, 2, 0 ]$ $2$ $8$ $[ 2, -5, -2, 2 ]$ $19664$ $[]$ $[]$ $[ 2, 3 ]$
$420$ $[ 1, 1, 2, 0 ]$ $2$ $8$ $[ 2, -5, -2, 2 ]$ $19664$ $[]$ $[]$ $[ 2, 3 ]$
$420$ $[ 1, 1, 2, 0 ]$ $2$ $8$ $[ 2, -5, -2, 2 ]$ $19664$ $[]$ $[]$ $[ 2, 3 ]$
$420$ $[ 1, 1, 2, 0 ]$ $2$ $8$ $[ 2, -5, -2, 2 ]$ $19664$ $[]$ $[]$ $[ 2, 3 ]$
$420$ $[ 1, 1, 2, 0 ]$ $2$ $16$ $[ -1, -8, 4, 12 ]$ $20308$ $[ -3, -20 ]$ $[ 256, 864 ]$ $[]$
$420$ $[ 1, 1, 2, 0 ]$ $2$ $16$ $[ -1, -8, 4, 12 ]$ $20308$ $[ -3, -20 ]$ $[ 256, 864 ]$ $[]$
$420$ $[ 0, 0, 1, 3 ]$ $4$ $16$ $[ 2, -6, -14, -5 ]$ $16448$ $[ -35 ]$ $[ 16 ]$ $[]$
$420$ $[ 0, 0, 2, 2 ]$ $6$ $16$ $[ 2, -6, -12, -3 ]$ $44496$ $[]$ $[]$ $[]$
$422$ $[ 70 ]$ $3$ $8$ $[ 2, -4, -3, 3 ]$ $7053$ $[]$ $[]$ $[]$
$423$ $[ 0, 0 ]$ $1$ $4$ $[ 1, -5, -5, -1 ]$ $1957$ $[]$ $[]$ $[]$
$425$ $[ 0, 0 ]$ $1$ $4$ $[ -2, -4, 8, -1 ]$ $6224$ $[]$ $[]$ $[]$
$425$ $[ 0, 0 ]$ $1$ $4$ $[ 2, -4, -8, -1 ]$ $6224$ $[]$ $[]$ $[]$
$426$ $[ 1, 35 ]$ $2$ $16$ $[ 2, -5, -2, 2 ]$ $19664$ $[ -71 ]$ $[ 1657 ]$ $[ 3, 3 ]$
$426$ $[ 0, 14 ]$ $5$ $8$ $[ 1, -8, -1, 11 ]$ $5225$ $[]$ $[]$ $[]$
$426$ $[ 0, 14 ]$ $5$ $8$ $[ 2, -10, -26, -11 ]$ $3600$ $[]$ $[]$ $[]$
$426$ $[ 1, 7 ]$ $10$ $16$ $[ 0, -15, 0, 55 ]$ $22000$ $[ -71 ]$ $[ 1 ]$ $[ 3, 3 ]$
$429$ $[ 0, 0, 0 ]$ $1$ $4$ $[ 1, -5, -3, 4 ]$ $8468$ $[]$ $[]$ $[]$
$431$ $[ 0 ]$ $1$ $4$ $[ 1, -3, -1, 1 ]$ $725$ $[]$ $[]$ $[]$
$434$ $[ 0, 10 ]$ $3$ $8$ $[ 0, -7, -2, 7 ]$ $26224$ $[]$ $[]$ $[]$
$434$ $[ 0, 10 ]$ $3$ $8$ $[ -2, -5, 6, -1 ]$ $1600$ $[]$ $[]$ $[]$
$434$ $[ 0, 10 ]$ $3$ $8$ $[ 2, -5, -4, 5 ]$ $18736$ $[]$ $[]$ $[]$
$434$ $[ 2, 10 ]$ $3$ $8$ $[ 2, -4, -4, 3 ]$ $11344$ $[]$ $[]$ $[]$
$434$ $[ 4, 10 ]$ $3$ $8$ $[ 2, -4, -4, 3 ]$ $11344$ $[]$ $[]$ $[]$
$434$ $[ 1, 15 ]$ $6$ $16$ $[ -14, 53, -37, 7 ]$ $227241$ $[ 217, -31 ]$ $[ 4, 7 ]$ $[]$
$434$ $[ 3, 5 ]$ $6$ $16$ $[ -14, 47, -16, 1 ]$ $39600$ $[ -31, 217 ]$ $[ 1, 16 ]$ $[]$
$434$ $[ 3, 5 ]$ $6$ $16$ $[ 0, -22, -12, 76 ]$ $643824$ $[ 217, -31 ]$ $[ 49, 361 ]$ $[]$
$434$ $[ 0, 2 ]$ $15$ $8$ $[ -9, 26, -24, 1 ]$ $1125$ $[]$ $[]$ $[]$
$434$ $[ 3, 1 ]$ $30$ $16$ $[ -23, 194, -712, 961 ]$ $1125$ $[ -31 ]$ $[ 1 ]$ $[ 7, 7 ]$
$435$ $[ 0, 0, 0 ]$ $1$ $4$ $[ 3, -2, -7, 1 ]$ $2225$ $[]$ $[]$ $[]$
$436$ $[ 0, 0 ]$ $1$ $4$ $[ 0, -7, -1, 8 ]$ $30273$ $[]$ $[]$ $[]$
$438$ $[ 0, 6 ]$ $12$ $8$ $[ -2, -15, 16, 52 ]$ $129168$ $[ 73 ]$ $[ 16 ]$ $[]$
$440$ $[ 0, 0, 2, 0 ]$ $2$ $8$ $[ -1, -9, 3, 2 ]$ $54764$ $[]$ $[]$ $[]$
$440$ $[ 0, 0, 0, 2 ]$ $5$ $8$ $[ -8, 18, -13, 1 ]$ $725$ $[]$ $[]$ $[]$
$442$ $[ 6, 8 ]$ $2$ $16$ $[ -1, -7, 7, 2 ]$ $64436$ $[ 17 ]$ $[ 4 ]$ $[]$
$442$ $[ 4, 0 ]$ $3$ $8$ $[ -1, -5, 1, 1 ]$ $8957$ $[]$ $[]$ $[]$
$442$ $[ 4, 0 ]$ $3$ $8$ $[ -19, 125, -319, 225 ]$ $33709$ $[]$ $[]$ $[]$
$442$ $[ 4, 0 ]$ $3$ $8$ $[ 1, -6, -6, 3 ]$ $14013$ $[]$ $[]$ $[]$
$442$ $[ 0, 2 ]$ $8$ $8$ $[ 2, -3, -6, -1 ]$ $2624$ $[ 17 ]$ $[ 4 ]$ $[]$
$444$ $[ 1, 1, 0 ]$ $2$ $8$ $[ -2, -5, 10, -2 ]$ $19664$ $[]$ $[]$ $[ 2, 3 ]$
$444$ $[ 1, 1, 0 ]$ $2$ $8$ $[ -2, -5, 10, -2 ]$ $19664$ $[]$ $[]$ $[ 2, 3 ]$
$444$ $[ 0, 0, 6 ]$ $6$ $8$ $[ 1, -25, -28, 100 ]$ $384381$ $[ 37 ]$ $[ 64 ]$ $[]$
$444$ $[ 1, 0, 3 ]$ $12$ $8$ $[ 2, -19, 16, 4 ]$ $58896$ $[ 37 ]$ $[ 4 ]$ $[]$
$444$ $[ 1, 0, 3 ]$ $12$ $8$ $[ 2, -19, 16, 4 ]$ $58896$ $[ 37 ]$ $[ 4 ]$ $[]$
$445$ $[ 0, 0 ]$ $1$ $4$ $[ -1, -5, 7, -1 ]$ $725$ $[]$ $[]$ $[]$
$445$ $[ 0, 0 ]$ $1$ $4$ $[ -1, -5, 5, 1 ]$ $8069$ $[]$ $[]$ $[]$
$448$ $[ 0, 4, 0 ]$ $4$ $8$ $[ 2, -4, -4, 2 ]$ $14656$ $[ 8 ]$ $[ 4 ]$ $[]$
$448$ $[ 1, 4, 3 ]$ $4$ $16$ $[ 2, -4, -6, 1 ]$ $14656$ $[ -56 ]$ $[ 256 ]$ $[]$
$448$ $[ 1, 0, 1 ]$ $6$ $16$ $[ 2, -7, -14, 1 ]$ $9792$ $[ -7 ]$ $[ 1 ]$ $[ 2, 2 ]$
$448$ $[ 0, 1, 0 ]$ $16$ $8$ $[ 0, -4, 0, 2 ]$ $2048$ $[ 8 ]$ $[ 2 ]$ $[]$
$449$ $[ 112 ]$ $4$ $8$ $[ 2, -3, -2, 1 ]$ $2624$ $[]$ $[]$ $[]$
$450$ $[ 0, 4 ]$ $5$ $8$ $[ 1, -9, -4, 16 ]$ $7625$ $[]$ $[]$ $[]$
$450$ $[ 0, 4 ]$ $5$ $8$ $[ 1, -19, -4, 76 ]$ $32625$ $[]$ $[]$ $[]$
$450$ $[ 0, 2 ]$ $10$ $8$ $[ 0, -5, 0, 5 ]$ $2000$ $[]$ $[]$ $[]$
$450$ $[ 0, 2 ]$ $10$ $8$ $[ 0, -5, 0, 5 ]$ $2000$ $[]$ $[]$ $[]$
$450$ $[ 0, 2 ]$ $10$ $16$ $[ -1, -9, 4, 16 ]$ $7625$ $[ -15 ]$ $[ 16 ]$ $[]$
$450$ $[ 1, 5 ]$ $12$ $16$ $[ 0, -6, 0, 3 ]$ $27648$ $[ -15 ]$ $[ 25 ]$ $[]$
$450$ $[ 3, 1 ]$ $20$ $16$ $[ 0, -5, 0, 5 ]$ $2000$ $[ -15 ]$ $[ 1 ]$ $[]$
$450$ $[ 3, 1 ]$ $20$ $16$ $[ 0, -5, 0, 5 ]$ $2000$ $[ -15 ]$ $[ 256 ]$ $[]$
$450$ $[ 3, 1 ]$ $20$ $16$ $[ 0, -5, 0, 5 ]$ $2000$ $[ -15 ]$ $[ 1 ]$ $[]$
$454$ $[ 0 ]$ $1$ $4$ $[ 2, -3, -2, 1 ]$ $2624$ $[]$ $[]$ $[]$
$455$ $[ 0, 0, 0 ]$ $1$ $4$ $[ 1, -5, -3, 1 ]$ $12197$ $[]$ $[]$ $[]$
$455$ $[ 0, 0, 0 ]$ $1$ $4$ $[ -3, -1, 5, 1 ]$ $1957$ $[]$ $[]$ $[]$
$455$ $[ 0, 2, 0 ]$ $3$ $8$ $[ 1, -3, -1, 1 ]$ $725$ $[]$ $[]$ $[]$
$459$ $[ 6, 0 ]$ $3$ $8$ $[ -1, -5, 3, 3 ]$ $12357$ $[]$ $[]$ $[]$
$460$ $[ 1, 0, 11 ]$ $2$ $16$ $[ -1, -6, -1, 1 ]$ $13068$ $[ -4 ]$ $[ 1 ]$ $[]$
$460$ $[ 0, 1, 11 ]$ $4$ $16$ $[ 0, -11, 4, 19 ]$ $212336$ $[ -23, -115 ]$ $[ 4, 25 ]$ $[]$
$462$ $[ 0, 3, 5 ]$ $2$ $8$ $[ 0, -7, 2, 6 ]$ $40864$ $[]$ $[]$ $[ 7, 11 ]$
$462$ $[ 0, 3, 5 ]$ $2$ $8$ $[ 0, -7, 2, 6 ]$ $40864$ $[]$ $[]$ $[ 7, 11 ]$
$462$ $[ 1, 1, 0 ]$ $6$ $8$ $[ 0, -4, 0, 1 ]$ $2304$ $[]$ $[]$ $[ 3, 7 ]$
$462$ $[ 1, 1, 0 ]$ $6$ $8$ $[ 0, -4, 0, 1 ]$ $2304$ $[]$ $[]$ $[ 3, 7 ]$
$462$ $[ 0, 0, 2 ]$ $5$ $8$ $[ 0, -17, 0, 71 ]$ $28400$ $[]$ $[]$ $[]$
$462$ $[ 0, 0, 2 ]$ $5$ $8$ $[ 2, -10, 4, 4 ]$ $3600$ $[]$ $[]$ $[]$
$462$ $[ 0, 0, 2 ]$ $5$ $8$ $[ 0, -10, 0, 20 ]$ $8000$ $[]$ $[]$ $[]$
$462$ $[ 0, 0, 2 ]$ $5$ $8$ $[ 0, -7, 0, 11 ]$ $4400$ $[]$ $[]$ $[]$
$464$ $[ 0, 0, 14 ]$ $2$ $8$ $[ -1, -5, 3, 4 ]$ $8468$ $[]$ $[]$ $[]$
$465$ $[ 0, 0, 0 ]$ $1$ $4$ $[ -1, -5, 3, 4 ]$ $8468$ $[]$ $[]$ $[]$
$465$ $[ 0, 0, 10 ]$ $3$ $8$ $[ -14, 57, -57, 16 ]$ $67009$ $[]$ $[]$ $[]$
$465$ $[ 0, 0, 6 ]$ $5$ $8$ $[ 1, -3, -1, 1 ]$ $725$ $[]$ $[]$ $[]$
$468$ $[ 0, 3, 1 ]$ $12$ $16$ $[ 0, -9, -6, 6 ]$ $25488$ $[ -39 ]$ $[ 16 ]$ $[]$
$468$ $[ 1, 0, 1 ]$ $12$ $16$ $[ 2, -13, 10, -2 ]$ $13968$ $[ -52, -4 ]$ $[ 1, 208 ]$ $[]$
$469$ $[ 3, 33 ]$ $2$ $8$ $[ 1, -7, -5, 8 ]$ $37108$ $[ 469 ]$ $[ 1 ]$ $[]$
$469$ $[ 3, 33 ]$ $2$ $8$ $[ 1, -7, -5, 8 ]$ $37108$ $[ 469 ]$ $[ 1 ]$ $[]$
$469$ $[ 2, 0 ]$ $3$ $8$ $[ -1, -5, 1, 1 ]$ $8957$ $[]$ $[]$ $[]$
$472$ $[ 0, 0, 0 ]$ $1$ $4$ $[ 1, -5, 0, 1 ]$ $6809$ $[]$ $[]$ $[]$
$474$ $[ 0, 0 ]$ $1$ $4$ $[ 0, -18, -23, 16 ]$ $151717$ $[]$ $[]$ $[]$
$475$ $[ 0, 0 ]$ $1$ $4$ $[ -2, -6, 8, 9 ]$ $11344$ $[]$ $[]$ $[]$
$475$ $[ 10, 0 ]$ $2$ $8$ $[ 2, -4, -6, 2 ]$ $11344$ $[]$ $[]$ $[]$
$475$ $[ 0, 6 ]$ $3$ $8$ $[ -13, 50, -73, 36 ]$ $7537$ $[]$ $[]$ $[]$
$475$ $[ 10, 6 ]$ $6$ $16$ $[ -13, 50, -73, 36 ]$ $7537$ $[]$ $[]$ $[]$
$476$ $[ 0, 0, 8 ]$ $2$ $8$ $[ 0, -4, -1, 1 ]$ $1957$ $[]$ $[]$ $[]$
$476$ $[ 0, 2, 0 ]$ $3$ $8$ $[ -2, -9, 3, 12 ]$ $113481$ $[]$ $[]$ $[]$
$477$ $[ 0, 0 ]$ $1$ $4$ $[ 3, -1, -5, 1 ]$ $1957$ $[]$ $[]$ $[]$
$477$ $[ 0, 0 ]$ $1$ $4$ $[ -3, -1, 5, 1 ]$ $1957$ $[]$ $[]$ $[]$
$477$ $[ 0, 0 ]$ $1$ $4$ $[ 3, -1, -7, -3 ]$ $1957$ $[]$ $[]$ $[]$
$477$ $[ 0, 26 ]$ $2$ $8$ $[ -1, -5, 3, 1 ]$ $12197$ $[]$ $[]$ $[]$
$478$ $[ 0 ]$ $1$ $4$ $[ 2, -4, -5, -1 ]$ $4205$ $[]$ $[]$ $[]$
$478$ $[ 0 ]$ $1$ $4$ $[ 6, 10, 3, -1 ]$ $725$ $[]$ $[]$ $[]$
$480$ $[ 1, 4, 1, 0 ]$ $2$ $8$ $[ 1, -9, -9, -2 ]$ $20308$ $[ 24 ]$ $[ 256 ]$ $[]$
$480$ $[ 1, 4, 1, 0 ]$ $2$ $8$ $[ 1, -9, -9, -2 ]$ $20308$ $[ 24 ]$ $[ 256 ]$ $[]$
$480$ $[ 0, 4, 1, 1 ]$ $4$ $32$ $[ 0, -5, -1, 1 ]$ $6809$ $[ -120, -3, -24 ]$ $[ 16, 128, 1856 ]$ $[]$
$480$ $[ 1, 0, 0, 1 ]$ $4$ $8$ $[ -2, -9, 20, -2 ]$ $10304$ $[]$ $[]$ $[ 2, 5 ]$
$480$ $[ 1, 0, 0, 1 ]$ $4$ $8$ $[ -2, -9, 20, -2 ]$ $10304$ $[]$ $[]$ $[ 2, 5 ]$
$482$ $[ 120 ]$ $2$ $8$ $[ -2, -18, 19, 86 ]$ $419917$ $[ 241 ]$ $[ 128 ]$ $[]$
$482$ $[ 60 ]$ $4$ $8$ $[ 0, -13, -18, -3 ]$ $260016$ $[ 241 ]$ $[ 1 ]$ $[]$
$483$ $[ 0, 0, 0 ]$ $1$ $4$ $[ -2, -4, 5, 2 ]$ $15317$ $[]$ $[]$ $[]$
$483$ $[ 0, 0, 0 ]$ $1$ $4$ $[ 0, -6, 1, 2 ]$ $24197$ $[]$ $[]$ $[]$
$484$ $[ 1, 55 ]$ $2$ $16$ $[ 2, -4, -8, -2 ]$ $7488$ $[ 44, -4 ]$ $[ 1, 388 ]$ $[]$
$484$ $[ 1, 11 ]$ $10$ $64$ $[ 2, -4, -8, -2 ]$ $7488$ $[ 44, -4 ]$ $[ 1, 388 ]$ $[]$
$485$ $[ 0, 0 ]$ $1$ $4$ $[ 1, -4, -2, 3 ]$ $1957$ $[]$ $[]$ $[]$
$488$ $[ 0, 0, 0 ]$ $1$ $4$ $[ 1, -6, -7, 1 ]$ $13676$ $[]$ $[]$ $[]$
$489$ $[ 0, 0 ]$ $1$ $4$ $[ 2, -2, -3, 1 ]$ $725$ $[]$ $[]$ $[]$
$490$ $[ 1, 21 ]$ $4$ $16$ $[ 2, -3, -6, -1 ]$ $2624$ $[ -7 ]$ $[ 4 ]$ $[]$
$490$ $[ 1, 21 ]$ $4$ $16$ $[ 2, -15, -16, 46 ]$ $51264$ $[ -7 ]$ $[ 4 ]$ $[]$
$490$ $[ 1, 7 ]$ $12$ $16$ $[ 2, -15, -16, 46 ]$ $51264$ $[ -7 ]$ $[ 4 ]$ $[]$
$490$ $[ 1, 7 ]$ $12$ $32$ $[ 2, -3, -6, -1 ]$ $2624$ $[ -7 ]$ $[ 4 ]$ $[]$
$492$ $[ 0, 0, 8 ]$ $5$ $8$ $[ 1, -10, -7, 19 ]$ $9225$ $[]$ $[]$ $[]$
$493$ $[ 0, 0 ]$ $1$ $4$ $[ -1, -5, 3, 4 ]$ $8468$ $[]$ $[]$ $[]$
$494$ $[ 0, 0 ]$ $1$ $4$ $[ -2, -5, 7, 5 ]$ $16609$ $[]$ $[]$ $[]$
$494$ $[ 0, 6 ]$ $3$ $8$ $[ 2, -4, -4, 3 ]$ $11344$ $[]$ $[]$ $[]$
$494$ $[ 4, 0 ]$ $3$ $8$ $[ -14, 33, -21, 4 ]$ $10273$ $[]$ $[]$ $[]$
$495$ $[ 0, 0, 0 ]$ $1$ $4$ $[ 2, -6, -10, 3 ]$ $48704$ $[]$ $[]$ $[]$
$495$ $[ 0, 0, 0 ]$ $1$ $4$ $[ -2, -6, 10, 3 ]$ $48704$ $[]$ $[]$ $[]$
$495$ $[ 3, 0, 5 ]$ $2$ $16$ $[ -12, 40, -28, 1 ]$ $41216$ $[ -11 ]$ $[ 1 ]$ $[ 3, 3 ]$
$495$ $[ 0, 0, 2 ]$ $5$ $8$ $[ -1, -3, 1, 1 ]$ $725$ $[]$ $[]$ $[]$
$495$ $[ 0, 0, 2 ]$ $5$ $8$ $[ -7, 13, -7, 1 ]$ $725$ $[]$ $[]$ $[]$
$495$ $[ 0, 0, 2 ]$ $5$ $8$ $[ -9, 26, -24, 1 ]$ $1125$ $[]$ $[]$ $[]$
$495$ $[ 0, 0, 2 ]$ $5$ $8$ $[ -17, 98, -212, 121 ]$ $5725$ $[]$ $[]$ $[]$
$495$ $[ 0, 0, 2 ]$ $5$ $8$ $[ -2, -4, 5, 5 ]$ $2525$ $[]$ $[]$ $[]$
$495$ $[ 0, 0, 2 ]$ $5$ $8$ $[ -1, -3, 1, 1 ]$ $725$ $[]$ $[]$ $[]$
$495$ $[ 0, 2, 2 ]$ $10$ $16$ $[ -9, 27, -31, 11 ]$ $725$ $[ 5 ]$ $[ 11 ]$ $[]$
$496$ $[ 1, 0, 5 ]$ $6$ $8$ $[ -1, -10, 4, 22 ]$ $81684$ $[ 124 ]$ $[ 1 ]$ $[]$
$496$ $[ 1, 0, 5 ]$ $6$ $8$ $[ -1, -10, 4, 22 ]$ $81684$ $[ 124 ]$ $[ 1 ]$ $[]$
$496$ $[ 0, 0, 6 ]$ $5$ $8$ $[ 1, -10, -2, 19 ]$ $8725$ $[]$ $[]$ $[]$
$496$ $[ 0, 0, 6 ]$ $5$ $8$ $[ -1, -7, 3, 9 ]$ $4525$ $[]$ $[]$ $[]$
$496$ $[ 0, 0, 2 ]$ $15$ $8$ $[ 1, -4, -4, 1 ]$ $1125$ $[]$ $[]$ $[]$
$496$ $[ 0, 0, 2 ]$ $15$ $8$ $[ 1, -4, -4, 1 ]$ $1125$ $[]$ $[]$ $[]$
$498$ $[ 0, 0 ]$ $1$ $4$ $[ -2, -8, 11, 14 ]$ $40709$ $[]$ $[]$ $[]$
$500$ $[ 1, 25 ]$ $4$ $32$ $[ -10, 25, -20, 5 ]$ $2000$ $[ -4, -20 ]$ $[ 5, 10576 ]$ $[]$
$500$ $[ 0, 20 ]$ $5$ $16$ $[ -11, 36, -31, 1 ]$ $7625$ $[ 5 ]$ $[ 1 ]$ $[]$
$500$ $[ 0, 10 ]$ $10$ $8$ $[ -11, 36, -31, 1 ]$ $7625$ $[ 5 ]$ $[ 1 ]$ $[]$
Tables of newforms corresponding to QM-building blocks
======================================================
QM-surfaces
-----------
$N$ $\varepsilon$ $\operatorname{ord}(\varepsilon)$ $\dim(A_f)$ $[t_i]$ $[\delta_i]$ $[\D]\in\Br(\Q)[2]$ $\gamma_\varepsilon\in\Br(K_P)[2]$
------- ------------------ ----------------------------------- ------------- ---------------------- --------------- --------------------- ------------------------------------
$28$ $[ 1, 1 ]$ $6$ $4$ $[ -7 ]$ $[ 3 ]$ $[ 2, 3 ]$ $[ 2, 2 ]$
$35$ $[ 1, 3 ]$ $4$ $4$ $[ -35, -7 ]$ $[ 2, 5 ]$ $[ 2, 5 ]$ $[]$
$44$ $[ 1, 5 ]$ $2$ $4$ $[ -11, 44 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$56$ $[ 1, 1, 3 ]$ $2$ $4$ $[ -7 ]$ $[ 6 ]$ $[ 2, 3 ]$ $[ 2, 2 ]$
$57$ $[ 1, 9 ]$ $2$ $4$ $[ -19, 57 ]$ $[ 2, 5 ]$ $[ 2, 5 ]$ $[]$
$60$ $[ 0, 1, 1 ]$ $4$ $4$ $[ 5, -3 ]$ $[ 2, 5 ]$ $[ 2, 5 ]$ $[]$
$63$ $[ 3, 1 ]$ $6$ $4$ $[ -3, -7 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$77$ $[ 3, 5 ]$ $2$ $4$ $[ -11, -7 ]$ $[ 2, 5 ]$ $[ 2, 5 ]$ $[]$
$80$ $[ 1, 0, 1 ]$ $4$ $4$ $[ 5, -4 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$92$ $[ 1, 11 ]$ $2$ $4$ $[ -23 ]$ $[ 14 ]$ $[ 2, 7 ]$ $[ 2, 2 ]$
$93$ $[ 1, 5 ]$ $6$ $4$ $[ -3, 93 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$95$ $[ 1, 3 ]$ $12$ $4$ $[ 5, -19 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$95$ $[ 1, 3 ]$ $12$ $4$ $[ 5, -19 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$104$ $[ 1, 1, 3 ]$ $4$ $4$ $[ -104, -8 ]$ $[ 2, 13 ]$ $[ 2, 13 ]$ $[]$
$108$ $[ 1, 9 ]$ $2$ $4$ $[ -3, -4 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$108$ $[ 1, 9 ]$ $2$ $4$ $[ -4, 12 ]$ $[ 3, 5 ]$ $[ 3, 5 ]$ $[]$
$112$ $[ 1, 1, 3 ]$ $4$ $4$ $[ 8, -7 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$112$ $[ 1, 1, 3 ]$ $4$ $4$ $[ -56, -7 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$112$ $[ 1, 0, 1 ]$ $6$ $4$ $[ 28, -4 ]$ $[ 3, 7 ]$ $[ 2, 7 ]$ $[]$
$116$ $[ 1, 7 ]$ $4$ $4$ $[ -116, -4 ]$ $[ 2, 7 ]$ $[ 2, 7 ]$ $[]$
$117$ $[ 0, 2 ]$ $6$ $4$ $[ -39, -3 ]$ $[ 3, 5 ]$ $[ 3, 5 ]$ $[]$
$117$ $[ 1, 3 ]$ $12$ $4$ $[ 13, -3 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$124$ $[ 1, 15 ]$ $2$ $4$ $[ -31, 124 ]$ $[ 6, 7 ]$ $[ 3, 7 ]$ $[]$
$124$ $[ 1, 15 ]$ $2$ $4$ $[ -31 ]$ $[ 6 ]$ $[ 2, 3 ]$ $[ 2, 2 ]$
$135$ $[ 0, 2 ]$ $2$ $4$ $[ -3 ]$ $[ 2 ]$ $[ 2, 3 ]$ $[]$
$140$ $[ 1, 1, 2 ]$ $12$ $8$ $[ -20, -4 ]$ $[ 2, 7 ]$ $[ 2, 7 ]$ $[]$
$141$ $[ 1, 23 ]$ $2$ $4$ $[ 141, -47 ]$ $[ 2, 15 ]$ $[ 2, 5 ]$ $[]$
$144$ $[ 1, 0, 1 ]$ $6$ $4$ $[ -3 ]$ $[ 3 ]$ $[ 2, 3 ]$ $[]$
$148$ $[ 1, 9 ]$ $4$ $4$ $[ 37, -4 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$152$ $[ 1, 1, 9 ]$ $2$ $4$ $[ 152, -8 ]$ $[ 3, 5 ]$ $[ 3, 5 ]$ $[]$
$152$ $[ 1, 1, 3 ]$ $6$ $4$ $[ -8, 152 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$153$ $[ 0, 4 ]$ $4$ $4$ $[ -51 ]$ $[ 2 ]$ $[ 2, 3 ]$ $[]$
$155$ $[ 1, 5 ]$ $12$ $4$ $[ 5, -31 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$155$ $[ 1, 5 ]$ $12$ $4$ $[ 5, -31 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$156$ $[ 0, 1, 3 ]$ $4$ $4$ $[ 13, -3 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$156$ $[ 1, 1, 2 ]$ $6$ $4$ $[ -3, 156 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$156$ $[ 1, 1, 2 ]$ $6$ $4$ $[ -3, 156 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$156$ $[ 1, 1, 4 ]$ $6$ $8$ $[ -4, 12 ]$ $[ 2, 5 ]$ $[ 2, 5 ]$ $[]$
$175$ $[ 5, 3 ]$ $4$ $4$ $[ -35, -7 ]$ $[ 2, 5 ]$ $[ 2, 5 ]$ $[]$
$175$ $[ 5, 1 ]$ $12$ $8$ $[ -7 ]$ $[ 3 ]$ $[ 3, 5 ]$ $[]$
$180$ $[ 1, 0, 1 ]$ $4$ $8$ $[ -15, -4 ]$ $[ 2, 7 ]$ $[ 3, 7 ]$ $[]$
$180$ $[ 1, 1, 2 ]$ $6$ $4$ $[ -20, 60 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$180$ $[ 1, 1, 2 ]$ $6$ $4$ $[ -20, 60 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$184$ $[ 1, 1, 11 ]$ $2$ $4$ $[ -23 ]$ $[ 14 ]$ $[ 2, 7 ]$ $[ 2, 2 ]$
$189$ $[ 9, 3 ]$ $2$ $4$ $[ 21, -7 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$189$ $[ 9, 3 ]$ $2$ $4$ $[ -7, -3 ]$ $[ 3, 5 ]$ $[ 3, 5 ]$ $[]$
$189$ $[ 0, 2 ]$ $3$ $4$ $[ -3 ]$ $[ 6 ]$ $[ 2, 3 ]$ $[]$
$189$ $[ 9, 1 ]$ $6$ $4$ $[ -7, -3 ]$ $[ 3, 5 ]$ $[ 3, 5 ]$ $[]$
$189$ $[ 9, 1 ]$ $6$ $4$ $[ -3, -7 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$196$ $[ 1, 21 ]$ $2$ $4$ $[ -7 ]$ $[ 3 ]$ $[ 2, 3 ]$ $[ 2, 2 ]$
$196$ $[ 1, 7 ]$ $6$ $4$ $[ -7 ]$ $[ 3 ]$ $[ 2, 3 ]$ $[ 2, 2 ]$
$200$ $[ 1, 1, 5 ]$ $4$ $8$ $[ -40, -8 ]$ $[ 3, 5 ]$ $[ 3, 5 ]$ $[]$
$203$ $[ 1, 7 ]$ $12$ $4$ $[ 29, -7 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$220$ $[ 0, 1, 5 ]$ $4$ $4$ $[ 5, -11 ]$ $[ 2, 11 ]$ $[ 2, 11 ]$ $[]$
$224$ $[ 1, 4, 3 ]$ $2$ $4$ $[ -7 ]$ $[ 6 ]$ $[ 2, 3 ]$ $[ 2, 2 ]$
$224$ $[ 0, 0, 2 ]$ $3$ $4$ $[ -4 ]$ $[ 3 ]$ $[ 2, 3 ]$ $[]$
$224$ $[ 0, 0, 2 ]$ $3$ $4$ $[ -4 ]$ $[ 7 ]$ $[ 2, 7 ]$ $[]$
$225$ $[ 3, 5 ]$ $4$ $8$ $[ -3, 5 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$228$ $[ 0, 1, 9 ]$ $2$ $4$ $[ -19, -3 ]$ $[ 2, 5 ]$ $[ 2, 5 ]$ $[]$
$231$ $[ 1, 2, 5 ]$ $6$ $8$ $[ -3, -11 ]$ $[ 2, 21 ]$ $[ 2, 7 ]$ $[]$
$235$ $[ 1, 23 ]$ $4$ $4$ $[ -235, -47 ]$ $[ 2, 5 ]$ $[ 2, 5 ]$ $[]$
$240$ $[ 0, 0, 1, 1 ]$ $4$ $4$ $[ 5, -3 ]$ $[ 2, 5 ]$ $[ 2, 5 ]$ $[]$
$243$ $[ 0 ]$ $1$ $2$ $[ -3 ]$ $[ 6 ]$ $[ 2, 3 ]$ $[]$
$243$ $[ 54 ]$ $3$ $4$ $[ -3 ]$ $[ 6 ]$ $[ 2, 3 ]$ $[]$
$245$ $[ 2, 0 ]$ $2$ $4$ $[ -35, 5 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$245$ $[ 2, 14 ]$ $6$ $4$ $[ -35, 5 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$245$ $[ 2, 14 ]$ $6$ $4$ $[ -35, 5 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$245$ $[ 1, 7 ]$ $12$ $8$ $[ -35, -7 ]$ $[ 2, 5 ]$ $[ 2, 5 ]$ $[]$
$248$ $[ 1, 1, 5 ]$ $6$ $4$ $[ -31 ]$ $[ 3 ]$ $[ 2, 3 ]$ $[ 2, 2 ]$
$252$ $[ 0, 3, 3 ]$ $2$ $4$ $[ 21, -7 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$252$ $[ 0, 3, 1 ]$ $6$ $4$ $[ -3, -7 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$252$ $[ 1, 0, 1 ]$ $6$ $4$ $[ -3, -84 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$252$ $[ 1, 0, 1 ]$ $6$ $4$ $[ -3, -84 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$252$ $[ 1, 0, 1 ]$ $6$ $4$ $[ -7, -4 ]$ $[ 3, 7 ]$ $[ 3, 7 ]$ $[]$
$252$ $[ 1, 0, 1 ]$ $6$ $4$ $[ -7, -4 ]$ $[ 3, 7 ]$ $[ 3, 7 ]$ $[]$
$252$ $[ 1, 0, 1 ]$ $6$ $4$ $[ -7 ]$ $[ 3 ]$ $[ 2, 3 ]$ $[ 2, 2 ]$
$252$ $[ 1, 2, 1 ]$ $6$ $4$ $[ -7 ]$ $[ 3 ]$ $[ 2, 3 ]$ $[ 2, 2 ]$
$252$ $[ 1, 2, 3 ]$ $6$ $4$ $[ -7 ]$ $[ 3 ]$ $[ 2, 3 ]$ $[ 2, 2 ]$
$252$ $[ 1, 2, 3 ]$ $6$ $4$ $[ -7 ]$ $[ 3 ]$ $[ 2, 3 ]$ $[ 2, 2 ]$
$252$ $[ 1, 4, 1 ]$ $6$ $4$ $[ -7 ]$ $[ 3 ]$ $[ 2, 3 ]$ $[ 2, 2 ]$
$259$ $[ 3, 9 ]$ $4$ $4$ $[ -259, -7 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$259$ $[ 3, 9 ]$ $4$ $4$ $[ 37, -7 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$260$ $[ 1, 1, 6 ]$ $4$ $8$ $[ 65, -4, -52 ]$ $[ 2, 3, 5 ]$ $[ 3, 5 ]$ $[]$
$279$ $[ 3, 5 ]$ $6$ $4$ $[ -3, -31 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$279$ $[ 3, 5 ]$ $6$ $4$ $[ -3, -31 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$285$ $[ 1, 2, 3 ]$ $6$ $8$ $[ 5, -95, -3 ]$ $[ 2, 3, 5 ]$ $[ 2, 5 ]$ $[]$
$288$ $[ 0, 0, 2 ]$ $3$ $4$ $[ -4 ]$ $[ 3 ]$ $[ 2, 3 ]$ $[]$
$299$ $[ 3, 11 ]$ $4$ $4$ $[ -299, -23 ]$ $[ 2, 7 ]$ $[ 2, 7 ]$ $[]$
$300$ $[ 0, 1, 5 ]$ $4$ $4$ $[ 5, -3 ]$ $[ 2, 5 ]$ $[ 2, 5 ]$ $[]$
$304$ $[ 1, 0, 3 ]$ $6$ $4$ $[ -19 ]$ $[ 3 ]$ $[ 2, 3 ]$ $[]$
$308$ $[ 0, 3, 5 ]$ $2$ $4$ $[ -11, 77 ]$ $[ 2, 5 ]$ $[ 2, 5 ]$ $[]$
$315$ $[ 0, 1, 3 ]$ $4$ $4$ $[ -35, -7 ]$ $[ 2, 5 ]$ $[ 2, 5 ]$ $[]$
$315$ $[ 0, 1, 3 ]$ $4$ $8$ $[ 105, -7 ]$ $[ 2, 5 ]$ $[ 3, 5 ]$ $[]$
$315$ $[ 3, 2, 1 ]$ $6$ $8$ $[ -3, -35, 5 ]$ $[ 2, 3, 7 ]$ $[ 2, 7 ]$ $[]$
$316$ $[ 1, 39 ]$ $2$ $4$ $[ 316, -79 ]$ $[ 2, 7 ]$ $[ 2, 7 ]$ $[]$
$316$ $[ 1, 13 ]$ $6$ $4$ $[ -79 ]$ $[ 3 ]$ $[ 2, 3 ]$ $[ 2, 2 ]$
$320$ $[ 0, 8, 2 ]$ $2$ $8$ $[ 5, -40 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$320$ $[ 1, 0, 1 ]$ $4$ $4$ $[ 5, -4 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$324$ $[ 1, 9 ]$ $6$ $4$ $[ -3, -4 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$324$ $[ 1, 9 ]$ $6$ $4$ $[ -3, -4 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$324$ $[ 1, 9 ]$ $6$ $8$ $[ -4, 12 ]$ $[ 3, 5 ]$ $[ 3, 5 ]$ $[]$
$329$ $[ 3, 23 ]$ $2$ $4$ $[ 329, -47 ]$ $[ 3, 5 ]$ $[ 3, 5 ]$ $[]$
$336$ $[ 1, 0, 1, 2 ]$ $6$ $8$ $[ -3 ]$ $[ 11 ]$ $[ 2, 11 ]$ $[]$
$344$ $[ 1, 1, 7 ]$ $6$ $4$ $[ 344, -8 ]$ $[ 3, 5 ]$ $[ 3, 5 ]$ $[]$
$351$ $[ 0, 2 ]$ $6$ $4$ $[ -3, -39 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$351$ $[ 3, 3 ]$ $12$ $4$ $[ 13, -3 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$355$ $[ 1, 35 ]$ $4$ $4$ $[ -355, -71 ]$ $[ 2, 11 ]$ $[ 2, 11 ]$ $[]$
$360$ $[ 1, 1, 0, 1 ]$ $4$ $4$ $[ -15 ]$ $[ 2 ]$ $[ 2, 3 ]$ $[ 2, 2 ]$
$360$ $[ 1, 1, 0, 1 ]$ $4$ $4$ $[ -15 ]$ $[ 2 ]$ $[ 2, 3 ]$ $[ 2, 2 ]$
$360$ $[ 1, 1, 1, 2 ]$ $6$ $8$ $[ -8, 24 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$363$ $[ 1, 55 ]$ $2$ $4$ $[ -11, -3 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$363$ $[ 1, 11 ]$ $10$ $16$ $[ -11, -3 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$364$ $[ 1, 3, 4 ]$ $6$ $4$ $[ -7 ]$ $[ 3 ]$ $[ 2, 3 ]$ $[ 2, 2 ]$
$364$ $[ 1, 3, 4 ]$ $6$ $4$ $[ -7 ]$ $[ 3 ]$ $[ 2, 3 ]$ $[ 2, 2 ]$
$368$ $[ 1, 0, 11 ]$ $2$ $4$ $[ -23, 92 ]$ $[ 2, 5 ]$ $[ 2, 5 ]$ $[]$
$371$ $[ 3, 13 ]$ $4$ $4$ $[ -371, -7 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$380$ $[ 0, 1, 3 ]$ $12$ $4$ $[ 5, -19 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$380$ $[ 1, 1, 6 ]$ $12$ $4$ $[ 5, -4 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$380$ $[ 1, 1, 6 ]$ $12$ $4$ $[ 5, -4 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$392$ $[ 1, 1, 7 ]$ $6$ $8$ $[ -7 ]$ $[ 6 ]$ $[ 2, 3 ]$ $[ 2, 2 ]$
$395$ $[ 1, 13 ]$ $12$ $4$ $[ 5, -79 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$395$ $[ 1, 13 ]$ $12$ $4$ $[ 5, -79 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$396$ $[ 1, 0, 5 ]$ $2$ $4$ $[ -11, 44 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$396$ $[ 1, 0, 5 ]$ $2$ $8$ $[ 33, 44, -132 ]$ $[ 2, 3, 5 ]$ $[ 3, 5 ]$ $[]$
$396$ $[ 1, 2, 5 ]$ $6$ $4$ $[ -11, 44 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$396$ $[ 1, 2, 5 ]$ $6$ $4$ $[ -11, 44 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$396$ $[ 1, 2, 5 ]$ $6$ $4$ $[ -11, 44 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$396$ $[ 1, 2, 5 ]$ $6$ $4$ $[ -11, 44 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$399$ $[ 1, 2, 9 ]$ $6$ $8$ $[ -19, -3 ]$ $[ 2, 5 ]$ $[ 2, 5 ]$ $[]$
$400$ $[ 1, 0, 5 ]$ $4$ $4$ $[ 5, -4 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$405$ $[ 0, 2 ]$ $2$ $4$ $[ 5, -15 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$405$ $[ 18, 2 ]$ $6$ $8$ $[ -3 ]$ $[ 2 ]$ $[ 2, 3 ]$ $[]$
$405$ $[ 18, 2 ]$ $6$ $8$ $[ 5, -15 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$407$ $[ 5, 9 ]$ $4$ $4$ $[ 37, -11 ]$ $[ 2, 19 ]$ $[ 2, 19 ]$ $[]$
$416$ $[ 0, 0, 4 ]$ $3$ $4$ $[ -4 ]$ $[ 11 ]$ $[ 2, 11 ]$ $[]$
$416$ $[ 1, 4, 3 ]$ $4$ $4$ $[ -104, -8 ]$ $[ 2, 13 ]$ $[ 2, 13 ]$ $[]$
$416$ $[ 0, 0, 2 ]$ $6$ $4$ $[ -52 ]$ $[ 3 ]$ $[ 2, 3 ]$ $[]$
$420$ $[ 0, 1, 2, 3 ]$ $2$ $8$ $[ 21, -35 ]$ $[ 2, 5 ]$ $[ 2, 5 ]$ $[]$
$429$ $[ 1, 5, 6 ]$ $2$ $8$ $[ -39, -143 ]$ $[ 3, 10 ]$ $[ 3, 5 ]$ $[]$
$429$ $[ 1, 5, 4 ]$ $6$ $8$ $[ -3, -11 ]$ $[ 2, 13 ]$ $[ 2, 13 ]$ $[]$
$432$ $[ 0, 1, 0 ]$ $4$ $4$ $[ -24 ]$ $[ 2 ]$ $[ 2, 3 ]$ $[]$
$432$ $[ 0, 1, 0 ]$ $4$ $4$ $[ -24 ]$ $[ 2 ]$ $[ 2, 3 ]$ $[]$
$432$ $[ 1, 0, 3 ]$ $6$ $4$ $[ -3 ]$ $[ 3 ]$ $[ 2, 3 ]$ $[]$
$436$ $[ 1, 27 ]$ $4$ $4$ $[ -436, -4 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$440$ $[ 0, 1, 1, 5 ]$ $4$ $4$ $[ -440, -88 ]$ $[ 2, 5 ]$ $[ 2, 5 ]$ $[]$
$440$ $[ 0, 1, 1, 5 ]$ $4$ $4$ $[ -440, -88 ]$ $[ 2, 5 ]$ $[ 2, 5 ]$ $[]$
$441$ $[ 3, 21 ]$ $2$ $4$ $[ -3, -7 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$441$ $[ 3, 7 ]$ $6$ $4$ $[ -3, -7 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$444$ $[ 1, 1, 18 ]$ $2$ $4$ $[ -111, 444 ]$ $[ 3, 7 ]$ $[ 3, 7 ]$ $[]$
$444$ $[ 1, 1, 18 ]$ $2$ $4$ $[ -111, 444 ]$ $[ 3, 7 ]$ $[ 3, 7 ]$ $[]$
$444$ $[ 0, 1, 9 ]$ $4$ $4$ $[ 37, -3 ]$ $[ 2, 11 ]$ $[ 2, 11 ]$ $[]$
$444$ $[ 0, 1, 9 ]$ $4$ $4$ $[ 37, -3 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$444$ $[ 1, 1, 6 ]$ $6$ $4$ $[ -3, 444 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$444$ $[ 1, 1, 6 ]$ $6$ $4$ $[ -3, 444 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$448$ $[ 0, 0, 2 ]$ $3$ $4$ $[ -4 ]$ $[ 3 ]$ $[ 2, 3 ]$ $[]$
$448$ $[ 0, 0, 2 ]$ $3$ $4$ $[ -4 ]$ $[ 7 ]$ $[ 2, 7 ]$ $[]$
$448$ $[ 1, 4, 3 ]$ $4$ $4$ $[ -56, -7 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$448$ $[ 1, 4, 3 ]$ $4$ $4$ $[ 8, -7 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$448$ $[ 0, 8, 2 ]$ $6$ $8$ $[ -8 ]$ $[ 7 ]$ $[ 2, 7 ]$ $[]$
$448$ $[ 1, 0, 1 ]$ $6$ $4$ $[ -7 ]$ $[ 3 ]$ $[ 2, 3 ]$ $[ 2, 2 ]$
$448$ $[ 1, 0, 1 ]$ $6$ $4$ $[ 28, -4 ]$ $[ 3, 7 ]$ $[ 2, 7 ]$ $[]$
$448$ $[ 1, 8, 1 ]$ $6$ $8$ $[ -56, 8 ]$ $[ 3, 5 ]$ $[ 2, 5 ]$ $[]$
$455$ $[ 1, 3, 6 ]$ $4$ $8$ $[ -35, -91 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$455$ $[ 1, 3, 2 ]$ $12$ $8$ $[ -35, -91 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$455$ $[ 1, 3, 2 ]$ $12$ $8$ $[ -35, -91 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$456$ $[ 1, 1, 1, 6 ]$ $6$ $8$ $[ -8, 24 ]$ $[ 2, 11 ]$ $[ 2, 11 ]$ $[]$
$459$ $[ 0, 4 ]$ $4$ $4$ $[ -51 ]$ $[ 2 ]$ $[ 2, 3 ]$ $[]$
$460$ $[ 1, 2, 11 ]$ $2$ $8$ $[ -115, 460, -23 ]$ $[ 2, 3, 5 ]$ $[ 2, 5 ]$ $[]$
$465$ $[ 1, 2, 15 ]$ $2$ $4$ $[ -155, 465 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$465$ $[ 1, 2, 15 ]$ $2$ $4$ $[ -155, 465 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$465$ $[ 1, 2, 5 ]$ $6$ $8$ $[ 5, -155, -15 ]$ $[ 2, 3, 5 ]$ $[ 2, 5 ]$ $[]$
$468$ $[ 1, 0, 3 ]$ $4$ $8$ $[ -39, -4 ]$ $[ 2, 7 ]$ $[ 3, 7 ]$ $[]$
$468$ $[ 1, 3, 2 ]$ $6$ $4$ $[ -3, -52 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$468$ $[ 1, 3, 2 ]$ $6$ $4$ $[ -3, -52 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$468$ $[ 1, 3, 4 ]$ $6$ $4$ $[ -3, -4 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$468$ $[ 1, 3, 4 ]$ $6$ $4$ $[ -3, -4 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$469$ $[ 3, 33 ]$ $2$ $4$ $[ -67, -7 ]$ $[ 2, 5 ]$ $[ 2, 5 ]$ $[]$
$473$ $[ 5, 21 ]$ $2$ $4$ $[ -11, -43 ]$ $[ 2, 5 ]$ $[ 2, 5 ]$ $[]$
$475$ $[ 5, 3 ]$ $12$ $4$ $[ 5, -19 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$475$ $[ 5, 3 ]$ $12$ $4$ $[ 5, -95 ]$ $[ 2, 3 ]$ $[ 2, 5 ]$ $[]$
$475$ $[ 5, 3 ]$ $12$ $4$ $[ 5, -95 ]$ $[ 2, 3 ]$ $[ 2, 5 ]$ $[]$
$475$ $[ 5, 3 ]$ $12$ $4$ $[ 5, -19 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$484$ $[ 1, 11 ]$ $10$ $16$ $[ -11, 44 ]$ $[ 2, 3 ]$ $[ 2, 3 ]$ $[]$
$496$ $[ 1, 0, 15 ]$ $2$ $4$ $[ -31, 124 ]$ $[ 2, 13 ]$ $[ 2, 13 ]$ $[]$
4-dimensional QM-varieties
--------------------------
$N$ $\varepsilon$ $\operatorname{ord}(\varepsilon)$ $\dim(A_f)$ $\Disc(F)$ $[t_i]$ $[N_{F/\Q}(\delta_i)]$ $[\D]\in\Br(F)[2]$ $\gamma_\varepsilon\in\Br(K_P)[2]$
------- ------------------ ----------------------------------- ------------- ------------ --------------------- ------------------------ -------------------- ------------------------------------
$36$ $[ 1, 1 ]$ $6$ $8$ $33$ $[ 12, -4 ]$ $[ 1, 3 ]$ $[ 2, 3 ]$ $[]$
$40$ $[ 1, 1, 1 ]$ $4$ $8$ $5$ $[ -40, -8 ]$ $[ 4, 20 ]$ $[ 4, 5 ]$ $[]$
$76$ $[ 1, 9 ]$ $2$ $8$ $17$ $[ -19, 76 ]$ $[ 8, 13 ]$ $[ 2, 13 ]$ $[]$
$88$ $[ 1, 1, 5 ]$ $2$ $8$ $8$ $[ 88, -11 ]$ $[ 49, 56 ]$ $[ 2, 7 ]$ $[]$
$99$ $[ 0, 2 ]$ $5$ $8$ $5$ $[ -3 ]$ $[ 11 ]$ $[ 9, 11 ]$ $[]$
$108$ $[ 1, 3 ]$ $6$ $8$ $33$ $[ 12, -4 ]$ $[ 1, 3 ]$ $[ 2, 3 ]$ $[]$
$120$ $[ 1, 1, 1, 2 ]$ $2$ $16$ $8$ $[ -3, -15 ]$ $[ 8, 56 ]$ $[ 2, 7 ]$ $[]$
$135$ $[ 9, 1 ]$ $4$ $8$ $24$ $[ -15, -3 ]$ $[ 1, 10 ]$ $[ 2, 5 ]$ $[]$
$140$ $[ 1, 2, 3 ]$ $2$ $16$ $17$ $[ -4, -35, -20 ]$ $[ 2, 16, 26 ]$ $[ 2, 13 ]$ $[]$
$140$ $[ 0, 1, 3 ]$ $4$ $8$ $41$ $[ -7, -35 ]$ $[ 2, 400 ]$ $[ 2, 2 ]$ $[]$
$144$ $[ 0, 1, 0 ]$ $4$ $8$ $28$ $[ -3, -24 ]$ $[ 8, 16 ]$ $[ 2, 3 ]$ $[]$
$153$ $[ 0, 2 ]$ $8$ $8$ $8$ $[ -51, -3 ]$ $[ 2, 17 ]$ $[ 2, 17 ]$ $[]$
$156$ $[ 1, 1, 6 ]$ $2$ $16$ $17$ $[ -52, -39 ]$ $[ 2, 34 ]$ $[ 2, 17 ]$ $[]$
$160$ $[ 1, 4, 1 ]$ $4$ $8$ $5$ $[ -40, -8 ]$ $[ 4, 20 ]$ $[ 4, 5 ]$ $[]$
$165$ $[ 1, 2, 5 ]$ $2$ $16$ $12$ $[ 165, -55, -15 ]$ $[ 13, 24, 52 ]$ $[ 3, 13 ]$ $[]$
$172$ $[ 1, 7 ]$ $6$ $8$ $24$ $[ 172, -4 ]$ $[ 1, 43 ]$ $[ 2, 43 ]$ $[]$
$176$ $[ 1, 0, 1 ]$ $10$ $8$ $5$ $[ -11, -4 ]$ $[ 5, 11 ]$ $[ 4, 11 ]$ $[]$
$200$ $[ 1, 1, 5 ]$ $4$ $8$ $5$ $[ -40, -8 ]$ $[ 4, 20 ]$ $[ 4, 5 ]$ $[]$
$209$ $[ 5, 9 ]$ $2$ $8$ $8$ $[ -11, 209 ]$ $[ 2, 7 ]$ $[ 2, 7 ]$ $[]$
$213$ $[ 1, 35 ]$ $2$ $8$ $5$ $[ -71 ]$ $[ 341 ]$ $[ 11, 31 ]$ $[ 3, 3 ]$
$216$ $[ 0, 1, 0 ]$ $2$ $8$ $28$ $[ -3 ]$ $[ 8 ]$ $[ 2, 3 ]$ $[]$
$216$ $[ 1, 1, 9 ]$ $2$ $8$ $12$ $[ -3, 24 ]$ $[ 24, 33 ]$ $[ 2, 11 ]$ $[]$
$225$ $[ 1, 5 ]$ $12$ $16$ $24$ $[ -3, -15 ]$ $[ 3, 27 ]$ $[ 3, 5 ]$ $[]$
$247$ $[ 3, 9 ]$ $4$ $8$ $40$ $[ 13, -19 ]$ $[ 9, 26 ]$ $[ 2, 13 ]$ $[]$
$259$ $[ 3, 9 ]$ $4$ $8$ $73$ $[ -7, 37 ]$ $[ 2, 16 ]$ $[ 2, 2 ]$ $[]$
$285$ $[ 1, 2, 9 ]$ $2$ $16$ $17$ $[ -19, -95 ]$ $[ 2, 236 ]$ $[ 2, 59 ]$ $[]$
$288$ $[ 0, 0, 2 ]$ $3$ $8$ $33$ $[ -4 ]$ $[ 27 ]$ $[ 2, 3 ]$ $[]$
$296$ $[ 1, 1, 9 ]$ $4$ $8$ $21$ $[ -296, -8 ]$ $[ 4, 37 ]$ $[ 4, 37 ]$ $[]$
$297$ $[ 9, 5 ]$ $2$ $8$ $12$ $[ -11, 33 ]$ $[ 6, 94 ]$ $[ 2, 47 ]$ $[]$
$300$ $[ 1, 1, 0 ]$ $2$ $8$ $41$ $[ 12, -3 ]$ $[ 5, 32 ]$ $[ 2, 5 ]$ $[]$
$300$ $[ 1, 1, 0 ]$ $2$ $8$ $41$ $[ 12, -3 ]$ $[ 5, 32 ]$ $[ 2, 5 ]$ $[]$
$300$ $[ 1, 1, 10 ]$ $2$ $16$ $41$ $[ 12, -3 ]$ $[ 5, 32 ]$ $[ 2, 5 ]$ $[]$
$309$ $[ 1, 17 ]$ $6$ $8$ $5$ $[ -103, -3 ]$ $[ 9, 11 ]$ $[ 9, 11 ]$ $[]$
$315$ $[ 0, 2, 2 ]$ $6$ $16$ $28$ $[ -3, -15 ]$ $[ 2, 21 ]$ $[ 2, 7 ]$ $[]$
$324$ $[ 1, 27 ]$ $2$ $8$ $33$ $[ 12, -4 ]$ $[ 1, 3 ]$ $[ 2, 3 ]$ $[]$
$324$ $[ 1, 27 ]$ $2$ $8$ $12$ $[ -3, -4 ]$ $[ 1, 132 ]$ $[ 3, 11 ]$ $[]$
$324$ $[ 1, 9 ]$ $6$ $16$ $12$ $[ -3, -4 ]$ $[ 1, 132 ]$ $[ 3, 11 ]$ $[]$
$333$ $[ 0, 6 ]$ $6$ $8$ $13$ $[ -3, -111 ]$ $[ 3, 9 ]$ $[ 3, 3 ]$ $[]$
$340$ $[ 1, 1, 2 ]$ $8$ $8$ $8$ $[ -68, -4 ]$ $[ 2, 56 ]$ $[ 2, 7 ]$ $[]$
$340$ $[ 1, 3, 2 ]$ $8$ $8$ $8$ $[ -68, -4 ]$ $[ 2, 56 ]$ $[ 2, 7 ]$ $[]$
$345$ $[ 1, 2, 11 ]$ $2$ $8$ $12$ $[ -3, 345 ]$ $[ 6, 6 ]$ $[ 2, 3 ]$ $[]$
$345$ $[ 1, 2, 11 ]$ $2$ $8$ $12$ $[ -3, 345 ]$ $[ 6, 6 ]$ $[ 2, 3 ]$ $[]$
$351$ $[ 0, 2 ]$ $6$ $8$ $40$ $[ -3, -39 ]$ $[ 6, 9 ]$ $[ 2, 3 ]$ $[]$
$352$ $[ 1, 4, 5 ]$ $2$ $8$ $8$ $[ 88, -11 ]$ $[ 49, 56 ]$ $[ 2, 7 ]$ $[]$
$352$ $[ 0, 0, 2 ]$ $5$ $8$ $5$ $[ -4 ]$ $[ 11 ]$ $[ 4, 11 ]$ $[]$
$352$ $[ 0, 0, 2 ]$ $5$ $8$ $5$ $[ -4 ]$ $[ 55 ]$ $[ 4, 11 ]$ $[]$
$360$ $[ 1, 1, 0, 1 ]$ $4$ $8$ $5$ $[ -40, -8 ]$ $[ 4, 20 ]$ $[ 4, 5 ]$ $[]$
$372$ $[ 0, 1, 15 ]$ $2$ $8$ $5$ $[ 93, -3 ]$ $[ 20, 44 ]$ $[ 5, 11 ]$ $[]$
$372$ $[ 0, 1, 3 ]$ $10$ $8$ $5$ $[ 93, -3 ]$ $[ 5, 89 ]$ $[ 5, 89 ]$ $[]$
$375$ $[ 1, 25 ]$ $4$ $16$ $5$ $[ -3, -15 ]$ $[ 5, 89 ]$ $[ 5, 89 ]$ $[]$
$396$ $[ 0, 0, 2 ]$ $5$ $8$ $5$ $[ -3 ]$ $[ 55 ]$ $[ 5, 11 ]$ $[]$
$405$ $[ 9, 1 ]$ $12$ $16$ $24$ $[ -15, -3 ]$ $[ 1, 10 ]$ $[ 2, 5 ]$ $[]$
$416$ $[ 0, 0, 4 ]$ $3$ $8$ $17$ $[ -4 ]$ $[ 8 ]$ $[ 2, 2 ]$ $[]$
$429$ $[ 1, 5, 6 ]$ $2$ $16$ $17$ $[ -11, 33 ]$ $[ 2, 52 ]$ $[ 2, 13 ]$ $[]$
$460$ $[ 0, 1, 11 ]$ $4$ $8$ $41$ $[ -23, -115 ]$ $[ 2, 4 ]$ $[ 2, 2 ]$ $[]$
$468$ $[ 1, 3, 6 ]$ $2$ $16$ $8$ $[ -4, -52 ]$ $[ 28, 92 ]$ $[ 7, 23 ]$ $[]$
$480$ $[ 1, 4, 1, 2 ]$ $2$ $16$ $8$ $[ -3, -15 ]$ $[ 8, 56 ]$ $[ 2, 7 ]$ $[]$
$480$ $[ 0, 0, 1, 1 ]$ $4$ $16$ $24$ $[ 60, -4 ]$ $[ 4, 12 ]$ $[ 2, 3 ]$ $[]$
6-dimensional QM-varieties
--------------------------
$N$ $\varepsilon$ $\operatorname{ord}(\varepsilon)$ $\dim(A_f)$ $F$ $\Disc(F)$ $[t_i]$ $[N_{F/\Q}(\delta_i)]$ $[\D]\in\Br(F)[2]$ $\gamma_\varepsilon\in\Br(K_P)[2]$
------- ---------------- ----------------------------------- ------------- -------------------- ------------ --------------------- ------------------------ -------------------- ------------------------------------
$48$ $[ 1, 1, 1 ]$ $4$ $12$ $[ -1, -4, 2 ]$ $316$ $[ -24, -3 ]$ $[ 8, 16 ]$ $[ 2, 2 ]$ $[]$
$56$ $[ 1, 1, 1 ]$ $6$ $12$ $[ 0, -6, -2 ]$ $756$ $[ 56, -8 ]$ $[ 3, 21 ]$ $[ 3, 7 ]$ $[]$
$63$ $[ 1, 3 ]$ $6$ $12$ $[ -7, 10, -3 ]$ $321$ $[ -3, -7 ]$ $[ 3, 27 ]$ $[ 3, 3 ]$ $[]$
$77$ $[ 1, 5 ]$ $6$ $12$ $[ -8, 17, -7 ]$ $321$ $[ -7, -11 ]$ $[ 3, 7 ]$ $[ 3, 7 ]$ $[]$
$84$ $[ 1, 1, 2 ]$ $6$ $24$ $[ -12, 35, -8 ]$ $2089$ $[ -3, -4 ]$ $[ 8, 64 ]$ $[ 2, 2 ]$ $[]$
$91$ $[ 3, 3 ]$ $4$ $12$ $[ -6, 8, -2 ]$ $148$ $[ -91, -7 ]$ $[ 2, 13 ]$ $[ 2, 13 ]$ $[]$
$105$ $[ 1, 2, 1 ]$ $6$ $24$ $[ -9, 18, -6 ]$ $1944$ $[ -3, 105, -15 ]$ $[ 6, 12, 28 ]$ $[ 2, 7 ]$ $[]$
$129$ $[ 1, 21 ]$ $2$ $12$ $[ -7, 10, -2 ]$ $568$ $[ -3, 129 ]$ $[ 2, 106 ]$ $[ 2, 53 ]$ $[]$
$133$ $[ 3, 3 ]$ $6$ $12$ $[ -6, 9, -3 ]$ $81$ $[ -19, -7 ]$ $[ 3, 19 ]$ $[ 3, 19 ]$ $[]$
$152$ $[ 1, 1, 9 ]$ $2$ $12$ $[ 0, -7, 4 ]$ $940$ $[ 152, -8 ]$ $[ 8, 40 ]$ $[ 2, 5 ]$ $[]$
$171$ $[ 0, 2 ]$ $9$ $12$ $[ -12, 45, -51 ]$ $81$ $[ -3 ]$ $[ 51 ]$ $[ 3, 17 ]$ $[]$
$188$ $[ 1, 23 ]$ $2$ $12$ $[ -7, 11, -4 ]$ $229$ $[ -4, -47 ]$ $[ 4, 1016 ]$ $[ 4, 127 ]$ $[]$
$189$ $[ 3, 3 ]$ $6$ $12$ $[ -7, 10, -3 ]$ $321$ $[ -3, -7 ]$ $[ 3, 27 ]$ $[ 3, 3 ]$ $[]$
$192$ $[ 1, 4, 1 ]$ $4$ $12$ $[ -1, -4, 2 ]$ $316$ $[ -24, -3 ]$ $[ 8, 16 ]$ $[ 2, 2 ]$ $[]$
$204$ $[ 1, 1, 8 ]$ $2$ $24$ $[ 0, -4, -1 ]$ $229$ $[ 204, -51, -68 ]$ $[ 28, 32, 148 ]$ $[ 2, 37 ]$ $[]$
$208$ $[ 1, 0, 3 ]$ $4$ $12$ $[ -14, 49, -12 ]$ $3144$ $[ -4, 13 ]$ $[ 12, 18 ]$ $[ 2, 3 ]$ $[]$
$220$ $[ 1, 2, 5 ]$ $2$ $24$ $[ -14, 55, -58 ]$ $961$ $[ -20, 220, -55 ]$ $[ 16, 32, 58 ]$ $[ 2, 29 ]$ $[]$
$224$ $[ 1, 4, 1 ]$ $6$ $12$ $[ 0, -6, -2 ]$ $756$ $[ 56, -8 ]$ $[ 3, 21 ]$ $[ 3, 7 ]$ $[]$
$253$ $[ 5, 11 ]$ $2$ $12$ $[ -9, 22, -13 ]$ $473$ $[ -23, 253 ]$ $[ 11, 13 ]$ $[ 11, 13 ]$ $[]$
$279$ $[ 3, 5 ]$ $6$ $12$ $[ -10, 28, -18 ]$ $564$ $[ 93, -31 ]$ $[ 18, 54 ]$ $[ 2, 3 ]$ $[]$
$301$ $[ 1, 3 ]$ $42$ $12$ $[ -7, 14, -7 ]$ $49$ $[ -43, -7 ]$ $[ 7, 27 ]$ $[ 7, 27 ]$ $[]$
$304$ $[ 1, 0, 1 ]$ $18$ $12$ $[ 0, -3, -1 ]$ $81$ $[ 76, -4 ]$ $[ 3, 19 ]$ $[ 8, 19 ]$ $[]$
$304$ $[ 1, 0, 1 ]$ $18$ $12$ $[ -15, 54, -57 ]$ $81$ $[ 76, -19 ]$ $[ 19, 57 ]$ $[ 3, 19 ]$ $[]$
$315$ $[ 3, 2, 1 ]$ $6$ $24$ $[ -12, 42, -42 ]$ $756$ $[ -7, 21, -15 ]$ $[ 14, 42, 189 ]$ $[ 2, 7 ]$ $[]$
$345$ $[ 1, 2, 11 ]$ $2$ $24$ $[ -8, 10, -1 ]$ $1765$ $[ -3, -23, 345 ]$ $[ 1, 160, 416 ]$ $[ 5, 13 ]$ $[]$
$348$ $[ 1, 1, 14 ]$ $2$ $24$ $[ 1, -8, -10 ]$ $892$ $[ 29, -87 ]$ $[ 1, 62 ]$ $[ 2, 31 ]$ $[]$
$351$ $[ 0, 4 ]$ $3$ $12$ $[ -9, 16, -6 ]$ $1436$ $[ -3 ]$ $[ 6 ]$ $[ 2, 3 ]$ $[]$
$357$ $[ 1, 3, 8 ]$ $2$ $24$ $[ -7, 12, -2 ]$ $316$ $[ -51, -7, -119 ]$ $[ 2, 8, 1538 ]$ $[ 2, 769 ]$ $[]$
$369$ $[ 0, 10 ]$ $4$ $12$ $[ -12, 38, -18 ]$ $3028$ $[ 41, -123 ]$ $[ 1, 18 ]$ $[ 2, 3 ]$ $[]$
$384$ $[ 1, 8, 1 ]$ $4$ $12$ $[ -1, -4, 2 ]$ $316$ $[ -24, -3 ]$ $[ 8, 16 ]$ $[ 2, 2 ]$ $[]$
$384$ $[ 1, 8, 1 ]$ $4$ $12$ $[ -1, -4, 2 ]$ $316$ $[ -24, -3 ]$ $[ 8, 16 ]$ $[ 2, 2 ]$ $[]$
$392$ $[ 1, 1, 21 ]$ $2$ $12$ $[ 0, -6, -2 ]$ $756$ $[ 56, -8 ]$ $[ 3, 21 ]$ $[ 3, 7 ]$ $[]$
$392$ $[ 1, 1, 7 ]$ $6$ $12$ $[ 0, -6, -2 ]$ $756$ $[ 56, -8 ]$ $[ 3, 21 ]$ $[ 3, 7 ]$ $[]$
$395$ $[ 1, 39 ]$ $4$ $12$ $[ -10, 25, -2 ]$ $892$ $[ -395, -79 ]$ $[ 2, 188 ]$ $[ 2, 47 ]$ $[]$
$416$ $[ 0, 0, 2 ]$ $6$ $12$ $[ 0, -6, 1 ]$ $837$ $[ -52 ]$ $[ 12 ]$ $[ 2, 3 ]$ $[]$
$441$ $[ 2, 0 ]$ $3$ $12$ $[ -1, -4, 1 ]$ $321$ $[ -7 ]$ $[ 81 ]$ $[ 3, 3 ]$ $[]$
$441$ $[ 2, 14 ]$ $3$ $12$ $[ -1, -4, 1 ]$ $321$ $[ -7 ]$ $[ 81 ]$ $[ 3, 3 ]$ $[]$
$441$ $[ 4, 14 ]$ $3$ $12$ $[ -1, -4, 1 ]$ $321$ $[ -7 ]$ $[ 81 ]$ $[ 3, 3 ]$ $[]$
$441$ $[ 1, 7 ]$ $6$ $12$ $[ -7, 10, -3 ]$ $321$ $[ -3, -7 ]$ $[ 3, 27 ]$ $[ 3, 3 ]$ $[]$
$441$ $[ 5, 7 ]$ $6$ $12$ $[ -7, 10, -3 ]$ $321$ $[ -3, -7 ]$ $[ 3, 27 ]$ $[ 3, 3 ]$ $[]$
$453$ $[ 1, 25 ]$ $6$ $12$ $[ -11, 29, -2 ]$ $4933$ $[ -3, -151 ]$ $[ 2, 27 ]$ $[ 2, 27 ]$ $[]$
$459$ $[ 0, 8 ]$ $2$ $12$ $[ -11, 31, -18 ]$ $3021$ $[ -3 ]$ $[ 8 ]$ $[ 2, 3 ]$ $[]$
$495$ $[ 0, 1, 5 ]$ $4$ $24$ $[ -9, 17, -3 ]$ $3028$ $[ 165, -15, -55 ]$ $[ 3, 12, 200 ]$ $[ 3, 3 ]$ $[]$
$496$ $[ 1, 0, 5 ]$ $6$ $12$ $[ -11, 35, -31 ]$ $564$ $[ 124, -4 ]$ $[ 3, 31 ]$ $[ 2, 31 ]$ $[]$
8-dimensional QM-varieties
--------------------------
$N$ $\varepsilon$ $\operatorname{ord}(\varepsilon)$ $\dim(A_f)$ $F$ $\Disc(F)$ $[t_i]$ $[N_{F/\Q}(\delta_i)]$ $[\D]\in\Br(F)[2]$ $\gamma_\varepsilon\in\Br(K_P)[2]$
------- ------------------ ----------------------------------- ------------- ------------------------- ------------ ---------------------- ------------------------ -------------------- ------------------------------------
$44$ $[ 1, 1 ]$ $10$ $16$ $[ -12, 39, -38, 11 ]$ $7625$ $[ -11, -4 ]$ $[ 1, 11 ]$ $[ 4, 11 ]$ $[]$
$76$ $[ 1, 3 ]$ $6$ $16$ $[ -1, -10, 10, -2 ]$ $70164$ $[ -19, 76 ]$ $[ 19, 304 ]$ $[ 2, 19 ]$ $[]$
$77$ $[ 3, 1 ]$ $10$ $16$ $[ -8, 14, -7, 1 ]$ $1125$ $[ -11, -7 ]$ $[ 1, 295 ]$ $[ 5, 59 ]$ $[]$
$132$ $[ 0, 1, 1 ]$ $10$ $16$ $[ 1, -8, -1, 11 ]$ $5225$ $[ -3, -11 ]$ $[ 121, 3025 ]$ $[ 11, 11 ]$ $[]$
$140$ $[ 1, 2, 1 ]$ $6$ $32$ $[ -13, 53, -77, 28 ]$ $90972$ $[ 5, 28, -20 ]$ $[ 28, 324, 1708 ]$ $[ 2, 7 ]$ $[]$
$171$ $[ 3, 3 ]$ $6$ $16$ $[ -16, 82, -150, 81 ]$ $248016$ $[ 57, -19 ]$ $[ 36, 81 ]$ $[ 3, 3 ]$ $[]$
$195$ $[ 1, 1, 6 ]$ $4$ $32$ $[ -9, 23, -15, 2 ]$ $19796$ $[ 5, -15, -39 ]$ $[ 2, 8, 2300 ]$ $[ 2, 23 ]$ $[]$
$225$ $[ 0, 4 ]$ $5$ $16$ $[ -10, 30, -25, 5 ]$ $5125$ $[ -3 ]$ $[ 5 ]$ $[ 5, 9 ]$ $[]$
$252$ $[ 1, 3, 2 ]$ $6$ $32$ $[ -1, -10, -7, 3 ]$ $47032$ $[ -3, -4 ]$ $[ 64, 756 ]$ $[ 3, 7 ]$ $[]$
$273$ $[ 1, 3, 6 ]$ $2$ $32$ $[ -10, 29, -24, 2 ]$ $137152$ $[ 21, 273, -39 ]$ $[ 2, 44, 112 ]$ $[ 2, 7 ]$ $[]$
$297$ $[ 0, 2 ]$ $5$ $16$ $[ -13, 51, -55, 5 ]$ $13525$ $[ -3 ]$ $[ 5 ]$ $[ 5, 9 ]$ $[]$
$297$ $[ 0, 2 ]$ $5$ $16$ $[ -11, 37, -39, 11 ]$ $5725$ $[ -3 ]$ $[ 11 ]$ $[ 9, 11 ]$ $[]$
$312$ $[ 1, 1, 1, 6 ]$ $2$ $32$ $[ 1, -4, -1, 2 ]$ $2777$ $[ 312, -39, -104 ]$ $[ 8, 296, 1952 ]$ $[ 37, 61 ]$ $[]$
$315$ $[ 0, 1, 1 ]$ $12$ $32$ $[ -12, 40, -42, 10 ]$ $658512$ $[ -7, 105, -3 ]$ $[ 1, 9, 10 ]$ $[ 2, 5 ]$ $[]$
$351$ $[ 9, 1 ]$ $12$ $16$ $[ -16, 87, -178, 94 ]$ $25488$ $[ -3, -39 ]$ $[ 94, 169 ]$ $[ 2, 47 ]$ $[]$
$364$ $[ 0, 3, 3 ]$ $4$ $16$ $[ -14, 54, -66, 8 ]$ $252692$ $[ 13, -91 ]$ $[ 2, 8 ]$ $[ 2, 2 ]$ $[]$
$369$ $[ 0, 4 ]$ $10$ $16$ $[ -13, 51, -55, 5 ]$ $13525$ $[ -3, 41 ]$ $[ 5, 25 ]$ $[ 5, 9 ]$ $[]$
$375$ $[ 1, 25 ]$ $4$ $32$ $[ -11, 34, -25, 5 ]$ $10025$ $[ 5, -15 ]$ $[ 5, 625 ]$ $[ 5, 5 ]$ $[]$
$385$ $[ 2, 3, 5 ]$ $2$ $32$ $[ -11, 39, -49, 18 ]$ $17428$ $[ -35, -55, 385 ]$ $[ 18, 86, 5328 ]$ $[ 2, 37 ]$ $[]$
$396$ $[ 1, 0, 1 ]$ $10$ $16$ $[ 1, -9, -4, 16 ]$ $7625$ $[ 44, -4 ]$ $[ 1, 176 ]$ $[ 4, 11 ]$ $[]$
$417$ $[ 1, 69 ]$ $2$ $16$ $[ -1, -5, 4, 3 ]$ $7537$ $[ -139, 417 ]$ $[ 24, 36 ]$ $[ 3, 8 ]$ $[]$
$420$ $[ 0, 1, 2, 1 ]$ $6$ $32$ $[ 2, -26, -36, 96 ]$ $2433384$ $[ -3, -7, -35 ]$ $[ 640, 1764, 9216 ]$ $[ 2, 5 ]$ $[]$
$465$ $[ 1, 2, 15 ]$ $2$ $32$ $[ 2, -6, -3, 2 ]$ $58397$ $[ -155, -31, -15 ]$ $[ 11, 736, 736 ]$ $[ 11, 23 ]$ $[]$
$484$ $[ 1, 55 ]$ $2$ $16$ $[ -12, 39, -38, 11 ]$ $7625$ $[ -11, -4 ]$ $[ 1, 11 ]$ $[ 4, 11 ]$ $[]$
$484$ $[ 1, 11 ]$ $10$ $16$ $[ -12, 39, -38, 11 ]$ $7625$ $[ -11, -4 ]$ $[ 1, 11 ]$ $[ 4, 11 ]$ $[]$
$484$ $[ 1, 11 ]$ $10$ $16$ $[ -12, 39, -38, 11 ]$ $7625$ $[ -11, -4 ]$ $[ 1, 11 ]$ $[ 4, 11 ]$ $[]$
$484$ $[ 1, 11 ]$ $10$ $16$ $[ -12, 39, -38, 11 ]$ $7625$ $[ -11, -4 ]$ $[ 1, 11 ]$ $[ 4, 11 ]$ $[]$
$496$ $[ 1, 0, 3 ]$ $10$ $16$ $[ -13, 54, -77, 31 ]$ $7625$ $[ -31, -4 ]$ $[ 1, 31 ]$ $[ 4, 31 ]$ $[]$
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Metastasis is a process of cell migration that can be collective and guided by chemical cues. Viewing metastasis in this way, as a physical phenomenon, allows one to draw upon insights from other studies of collective sensing and migration in cell biology. Here we review recent progress in the study of cell sensing and migration as collective phenomena, including in the context of metastatic cells. We describe simple physical models that yield the limits to the precision of cell sensing, and we review experimental evidence that cells operate near these limits. Models of collective migration are surveyed in order understand how collective metastatic invasion can occur. We conclude by contrasting cells’ sensory abilities with their sensitivity to drugs, and suggesting potential alternatives to cell-death-based cancer therapies.'
author:
- Julien Varennes
- Andrew Mugler
title: ' Sense and sensitivity: physical limits to multicellular sensing, migration and drug response'
---
Metastasis is one of the most intensely studied stages of cancer progression because it is the most deadly stage of cancer. The first step of metastasis is invasion, wherein cells break away from the tumor and invade the surrounding tissue. Our understanding of metastatic invasion has benefited tremendously from genetic and biochemical approaches [@leber2009molecular; @hanahan2000hallmarks; @hanahan2011hallmarks]. However, the physical aspects of metastatic invasion are still unclear [@hanahan2011hallmarks]. We know that at a fundamental level, metastatic invasion is a physical process. Tumor cells sense and respond to chemical gradients provided by surrounding cells [@bhowmick2004stromal; @condeelis2006macrophages; @shields2007autologous; @puliafito2015three] or other features of the tumor environment [@shields2007autologous; @polacheck2011interstitial; @shieh2011regulation] (Fig. \[overview\]A). Indeed, tumor cells are highly sensitive, able to detect a $1\%$ difference in concentration across the cell length [@shields2007autologous]. Sensing is ultimately a physical phenomenon. Therefore, can we build a simple physical theory to understand the sensory behavior of tumor cells, and can this physical theory inform treatment options?
Metastatic invasion involves coordinated migration of tumor cells away from the tumor site. In many types of cancer, migration is collective and highly organized, involving the coherent motion of connected groups of cells [@cheung2013collective; @friedl2012classifying; @aceto2014circulating; @puliafito2015three] (Fig. \[overview\]B). Collective migration is ultimately a physical phenomenon, since it relies on mechanical coupling and can often be understood as emerging from simple physical interactions at the cell-to-cell level. Can we understand the collective migration of tumor cells with simple physical models?
Here we review recent progress on modeling sensing and migration in cells and cell collectives. We discuss metastatic cells explicitly, and emphasize that physical insights gained from other cellular systems can inform our understanding of metastatic invasion. We focus on simple physical models and order-of-magnitude numerical estimates in order to quantitatively probe the extent of, and the limits to, cell sensory and migratory behavior. Our hope is that a more quantitative understanding of metastatic invasion will inform treatment protocols, and to that end we conclude by discussing drug sensitivity and potential treatment strategies (Fig. \[overview\]C).
![Metastatic invasion is guided by chemical attractants and can occur via (A) single cells or (B) multicellular groups. (C) Drugs are delivered to the tumor environment in order to prevent tumor growth and metastasis. Drugs may cause cell death (orange), block cell-to-cell communication (purple), or prevent cell migration (blue).[]{data-label="overview"}](fig1.png){width="0.9\columnwidth"}
Physical limits to sensory precision
====================================
Tumor cells sense very small concentration gradients[@shields2007autologous] and act in a collective manner[@cheung2013collective; @friedl2012classifying; @aceto2014circulating; @puliafito2015three]. Here we review the basic theory of concentration and gradient sensing by cells and cell collectives. This theory places physical bounds on sensory precision and allows us to quantitatively compare the capabilities of tumor cells to other cell types.
Single-cell concentration sensing
---------------------------------
Theoretical limits to the precision of concentration sensing were first introduced by Berg and Purcell almost 40 years ago[@berg1977physics]. Berg and Purcell began by considering an idealized cell that acts as a perfect counting instrument. Their simplest model assumed that the cell is a sphere in which molecules can freely diffuse in and out (Fig. \[sensing\]A). The concentration of these molecules is uniform in space, and the cell derives all its information about the concentration by counting each molecule inside its spherical body. The expected count is $\bar{n} = \bar{c}V$ where $\bar{c}$ is the mean concentration and $V$ is the cell volume. However, since molecules arrive and leave via diffusion, there will be fluctuations around this expected value. Diffusion is a Poisson process, meaning that the variance in this count $\sigma_n^2$ equals the mean $\bar{n}$. Therefore the relative error in the cell’s concentration estimate is $\sigma_c/\bar{c} = \sigma_n/\bar{n} = 1/\sqrt{\bar{c} V}$.
{width=".75\textwidth"}
The cell can improve upon the relative error in its concentration estimate by time-averaging over multiple measurements. However, consecutive measurements are only statistically independent if they are separated by a sufficient amount of time such that the molecules inside the cell volume are refreshed. The amount of time required is characterized by the diffusion time, $\tau \sim V^{2/3}/D \sim a^2/D$, where $D$ is the diffusion constant and $a$ is the cell diameter. In a time period $T$ the cell makes $\nu = T/\tau$ independent measurements, and the variance is reduced by the factor $1/\nu$. This gives the long-standing lower limit $$\label{eq:singleConc}
\frac{ \sigma_c }{\bar{c}} = \frac{ \sigma_n}{\bar{n}} \sim \frac{1}{\sqrt{a\bar{c}DT}}$$ for the cell’s relative error in estimating a uniform concentration. The relative error decreases with $a$ and $\bar{c}$, since the molecule count is larger, and also with $D$ and $T$, since more independent measurements can be made. Berg and Purcell derived this limit more rigorously[@berg1977physics], and the problem has been revisited more recently to account for binding kinetics, spatiotemporal correlations, and spatial confinement [@bialek2005physical; @kaizu2014berg; @bicknell2015limits]. In all cases a term of the form in Eq. \[eq:singleConc\] emerges as the fundamental limit for three-dimensional diffusion.
Do cells reach this limit? Berg and Purcell themselves asked this question in the context of several single-celled organisms, including the *Escheria coli* bacterium [@berg1977physics]. Motility of *E. coli* has two distinct phases: the run phase in which a cell swims in a fixed direction, and the tumble phase in which the cell erratically rotates in order to begin a new run in a different direction. The bacterium biases its motion by continually measuring the chemoattractant concentration, and extending the time of runs for which the change in concentration is positive [@celani2010bacterial; @tu2013quantitative; @berg1977physics; @dahlquist1976studies]. The change in concentration $\Delta \bar{c} = Tv\bar{g}$ over a run time $T$ depends on the concentration gradient $\bar{g} = \partial \bar{c} / \partial x$ and the bacterium’s velocity $v$. Berg and Purcell argued that for a change in concentration to be detectable, it must be larger than the measurement uncertainty, $\Delta \bar{c} > \sigma_c $. Together with Eq. \[eq:singleConc\], this places a lower limit on the run time, $T > [ \bar{c}/(aDv^2\bar{g}^2)]^{1/3}$. Using typical values [@berg1977physics] for the sensory threshold of *E. coli* of $\bar{c} = 1$ mM, $\partial \bar{c} / \partial x = 1$ mM/cm, $a=1$ $\mu$m, $v=15$ $\mu$m/s, and $D=10^{-5}$ cm$^2$/s, we find $T > 0.1$ s. Actual run times are on the order of $1$ s. Thus we see that *E. coli* chemotaxis is consistent with this physical bound. Although the end goal of concentration sensing in *E. coli* is chemotaxis by temporally sampling changes in the chemical concentration, we would like to focus the reader’s attention on the remarkable fact that the bacterium’s concentration sensing machinery operates very near the predicted physical limits. If *E. coli* were to use any shorter run times, chemotaxis would be physically impossible. Consequently, the time period for measuring the chemical concentration, $T$ in Eq. \[eq:singleConc\], would be so short that the bacterium would be unable to make an accurate measurement of the chemical concentration.
Single-cell gradient sensing
----------------------------
Cells are not only able to detect chemical concentrations, they are also able to measure spatial concentration gradients. Many cells, including amoeba, epithelial cells, neutrophils, and neurons, sense gradients by comparing concentrations between compartments in different locations of the cell body [@jilkine2011comparison]. These compartments are typically receptors or groups of receptors on the cell surface, but in a simple model we may treat these compartments as idealized counting volumes as we did before (Fig. \[sensing\]B). The difference in counts between two such compartments provides the cell with an estimate of the gradient. What is the relative error in this estimate?
Consider two compartments of linear size $s$ on either side of a cell with diameter $a$ (Fig. \[sensing\]B). If the compartments are aligned with the gradient $\bar{g}$ of a linear concentration profile, then the mean concentrations at each compartment are $\bar{c}_1$ and $\bar{c}_2 = \bar{c}_1 + a\bar{g}$. The mean molecule counts in the two compartments are roughly $\bar{n}_1 = \bar{c}_1s^3$ and $\bar{n}_2 = \bar{c}_2s^3$, and the difference is $\Delta\bar{n} = \bar{n}_2 - \bar{n}_1 = a\bar{g}s^3$. The variance in this difference is $\sigma_{\Delta n}^2 = \sigma_{n_1}^2 + \sigma_{n_2}^2 \sim \bar{n}_1^2/(s\bar{c}_1DT) + \bar{n}_2^2/(s\bar{c}_2DT)$, where the first step assumes the two compartments are independent, and the second step uses Eq. \[eq:singleConc\] for the variance in each compartment’s measurement. For shallow gradients, where the limits on sensing are generally probed, we have $a\bar{g} \ll \bar{c}_1$, and therefore we may assume $\bar{c_1} \approx \bar{c_2} \approx \bar{c}$, where $\bar{c}$ is the mean concentration at the center of the cell. Thus $\sigma_{\Delta n}^2 \sim 2(\bar{c}s^3)^2/(s\bar{c}DT)$, and the relative error in the cell’s estimate of the gradient is then $$\label{eq:g}
\frac{\sigma_g}{\bar{g}} = \frac{\sigma_{\Delta n}}{\Delta \bar{n}} \sim \sqrt{\frac{\bar{c}}{s(a\bar{g})^2DT}},$$ where the factor of $2$ is neglected in this simple scaling estimate. As in Eq. \[eq:singleConc\], we see that the relative error decreases with $s$, since the molecule counts in each compartment are larger, and also with $D$ and $T$, since more independent measurements can be made. Additionally, the relative error decreases with $a\bar{g}$, since the concentrations measured by the two compartments are more different from each other. However, we see that unlike in Eq. \[eq:singleConc\], the relative error increases with the background concentration $\bar{c}$. The reason is that the cell is not measuring a concentration, but rather a *difference* in concentrations, and it is more difficult to measure a small difference on a larger background than on a smaller background [@ellison2016cell]. Eq. \[eq:g\] has been derived more rigorously[@endres2009accuracy], and the problem has been extended to describe rings of receptors [@endres2009accuracy] or detectors distributed over the surface of a circle [@hu2010physical] or a sphere [@endres2008accuracy]. In all cases a term of the form in Eq. \[eq:g\] emerges as the fundamental limit, with the lengthscale $s$ dictated by the particular sensory mechanism and geometry. It is clear that the optimal mechanism would result in an effective compartment size that is roughly half of the cell volume, in which case $s\sim a$.
Do cells reach this limit on gradient sensing? This question has been directly addressed for the amoeba *Dictyostelium discoideum*. Experiments[@van2007biased] have shown that *Dictyostelium* cells exhibit biased movement when exposed to gradients of cyclic adenosine monophosphate as small as $\bar{g} = 10$ nM/mm, on top of a background concentration of $\bar{c} = 7$ nM. Bias is typically quantified in terms of the chemotactic index (CI), which is the cosine of the angle between the gradient direction and the direction of a cell’s actual motion. By relating the error in gradient sensing (a term of the form in Eq. \[eq:g\] with $s = a$) to the error in this angle, Endres and Wingreen [@endres2008accuracy] obtained an expression for the optimal CI, which they then fit to the experimental data with one free parameter, the integration time $T$. The inferred value of $T = 3.2$ s serves as the physical lower bound on the response time required to perform chemotaxis. Actual response times of *Dictyostelium* cells, as measured by the time from the addition of a chemoattractant to the peak activity of an observable signaling pathway associated with cell motility [@postma2003uniform; @parent2004making], are about $5$$-$$10$ s. Taken together, these results imply that *Dictyostelium* operates remarkably close to the physical limit to sensory precision set by the physics of molecule counting.
Relative changes vs. absolute molecule numbers
----------------------------------------------
The precision of gradient sensing is often reported in terms of percent concentration change across a cell body. For example, both amoeba [@van2007biased] and tumor cells [@shields2007autologous] are sensitive to a roughly $1\%$ change in concentration across the cell body. However, this method of reporting sensitivity may be misleading. Experiments imply very different sensory thresholds for these cells in terms of absolute molecule numbers, as we will now see.
The key is that it takes two numbers to specify the conditions for gradient sensing: the mean gradient $\bar{g}$ and the mean background concentration $\bar{c}$. For the amoeba *Dictyostelium*, these numbers are $\bar{g} = 10$ nM/mm and $\bar{c} = 7$ nM at the sensory threshold [@van2007biased]. Given a typical cell size of $a = 10$ $\mu$m, these values imply a mean percent concentration change of $\bar{p} = a\bar{g}/\bar{c} = 1.4\%$ (Table \[sense\_table\]). However, we may also compute from these values the mean molecule number difference $\Delta\bar{n} = a\bar{g}s^3$ from one side of the cell to the other, within the effective compartments of size $s$. Taking $s\sim a$ gives the maximal molecule number difference of $\Delta\bar{n} = a^4\bar{g} = 60$ for *Dictyostelium* (Table \[sense\_table\]). Together $\bar{p}$ and $\Delta\bar{n}$ specify the sensing conditions as completely as $\bar{g}$ and $\bar{c}$ do.
Experiments [@shields2007autologous] have shown that breast cancer tumor cells exhibit a chemotactic response in a gradient $\bar{g} = 550$ nM/mm of the cytokine CCL21, on top of a background concentration of $\bar{c} = 1100$ nM. Given a typical cell size of $a = 20$ $\mu$m, this corresponds to a percent difference of $\bar{p} = a\bar{g}/\bar{c} = 1\%$, similar to *Dictyostelium*. Yet, this also corresponds to a maximal molecule number difference of $\Delta\bar{n} = a^4\bar{g} = 53,$$000$, which is much higher than that of *Dictyostelium* (Table \[sense\_table\]). Even though the sensitivities are similar in terms of percent change, they are very different in terms of absolute molecule number.
Lower molecule numbers correspond to higher relative error. We can see this explicitly by writing Eq. \[eq:g\] in terms of the percent change $\bar{p} = a\bar{g}/\bar{c}$. Defining $\epsilon = \sigma_g/\bar{g}$ and taking $s \sim a$, we have $\epsilon \sim 1/\sqrt{\bar{p}^2a\bar{c}DT}$. Accounting for the fact that tumor cells (TC) have roughly twice the diameter as *Dictyostelium* cells (DC), this expression implies that the sensitivities of the two cell types over the same integration time $T$ to chemoattractants with the same diffusion constant $D$ satisfy $\epsilon_{\rm DC}/\epsilon_{\rm TC} = \sqrt{2\bar{c}_{\rm TC}/\bar{c}_{\rm DC}} \approx 18$. We see that because the *Dictyostelium* experiments were performed at lower background concentration, corresponding to lower absolute molecule numbers, the relative error in gradient sensing is $18$ times that of the tumor cells, despite the fact that both cell types are responsive to $1\%$ concentration gradients. Therefore, it is important to take note of the background concentration when studying the precision of gradient sensing. These data imply that *Dictyostelium* cells can sense noisier gradients than tumor cells. However, *Dictyostelium* cells have been studied more extensively than tumor cells as exemplars of gradient detection. It remains an interesting open question what is the minimum gradient that tumor cells can detect, not only in terms of percent concentration change, but also in terms of absolute molecule number differences.
We see that although cancerous cells and *Dictyostelium* cells are of similar size, their sensory responses to absolute molecule numbers can be very different (Table \[sense\_table\]). This difference is also reflected in their migration speeds: carcinoma and epithelial cells migrate [@wolf2003compensation; @wang2004differential; @gilles1999vimentin; @legrand1999airway] at $\sim 0.5 \mu\text{m}/\text{s}$ whereas *Dictyostelium* can migrate [@mccann2010cell; @song2006dictyostelium] at speeds of $\sim 10 \mu\text{m}/\text{s}$.
------------------------------------------ --------------------------- --------------------------------- -------------------------- ------------------------------
*Dictyostelium* Breast Neurons [@rosoff2004new] Mammary
(Amoeba) [@van2007biased] Cancer [@shields2007autologous] Epithelia [@ellison2016cell]
Cell Length 10 $\mu$m 20 $\mu$m 10 $\mu$m 10 $\mu$m
Scale, $a$
Background 7 nM 1100 nM 1 nM 2.5 nM
Concentration, $\bar{c}$
Concentration 10 nM/mm 550 nM/mm $0.1$ nM/mm $0.5$ nM/mm
Gradient, $\bar{g}$
Percent Concentration 1.4% 1.0% 0.1% 0.2%
Difference, $\bar{p} = \bar{g}a/\bar{c}$
Molecule Number 60 53,000 0.6 3
Difference, $\Delta\bar{n} = \bar{g}a^4$
------------------------------------------ --------------------------- --------------------------------- -------------------------- ------------------------------
Multicellular gradient sensing
------------------------------
In many cancer types, tumor cells invade the surrounding tissue in a collective manner [@cheung2013collective; @friedl2012classifying]. Cell collectives can sense shallower gradients than single cells [@ellison2016cell; @malet2015collective], both in terms of percent concentration changes and absolute molecule numbers (Table \[sense\_table\]). Indeed, groups of neurons respond to gradients equivalent to a difference of less than one molecule across an individual neuron’s growth cone [@rosoff2004new]. This raises the possibility that during the invasion process tumor cell collectives benefit from higher sensory precision than single tumor cells.
We can understand immediately from Eq. \[eq:g\] why a multicellular collective would have lower sensory error than a single cell: a collective is larger than a single cell. Therefore, the collective covers a larger portion of the concentration profile, which leads to a larger difference between the concentration measurements on either end, and a lower relative error. In terms of Eq. \[eq:g\], if we consider that cells on the ends act as the molecule-counting compartments, $s \to a$, and that the entire collective acts as the detector, $a \to Na$, where $N$ is the number of cells in the gradient direction, then we have [@mugler2016limits] $$\label{eq:G1}
\frac{\sigma_g}{\bar{g}} \sim \sqrt{\frac{\bar{c}}{a(Na\bar{g})^2DT}}.$$ We see that, as expected, the relative error goes down with the size $Na$ of the multicellular collective.
However, there is a crucial point that is overlooked in formulating Eq. \[eq:G1\]: the larger the group of cells, the more difficult it is for cells on either end to communicate the measurement information. This fact is not accounted for in Eq. \[eq:G1\]. Instead, we see that the relative error decreases with the separation $Na$ between the end cells without bound, which is unrealistic. For a single cell it may be a reasonable approximation to assume that compartments quickly and reliably communicate information across the cell body, but for a multicellular collective, the communication process cannot be overlooked. Importantly, the communication mechanism of multicellular collectives may introduce additional noise into the gradient sensing process. Therefore, it is imperative when considering collective sensing to properly account for the effects of communication.
Recently, the physical limits to collective gradient sensing including communication effects were derived [@ellison2016cell; @mugler2016limits]. Communication was modeled using a multicellular version of the local excitation–global inhibition (LEGI) paradigm [@levchenko2002models], in which each cell produces a “local” and a “global” molecular species in response to the chemoattractant, and the global species is exchanged between cells to provide the communication (Fig. \[sensing\]C). The difference between local and global molecule numbers in a given cell provides the readout. A positive difference informs the cell that its detected chemoattractant concentration is above the spatial average among its neighbors, and therefore that the cell is located up the gradient, not down. In this model, the relative error of gradient sensing was shown [@mugler2016limits] to be limited from below by $$\label{eq:G2}
\frac{\sigma_g}{\bar{g}} \sim \sqrt{\frac{\bar{c}}{a(n_0a\bar{g})^2DT}},$$ where $n_0^2$ is the ratio of the cell-to-cell exchange rate to the degradation rate of the global species. Comparing Eq. \[eq:G1\] to Eq. \[eq:G2\], we see that without communication the error decreases indefinitely with the size $Na$ of the collective, whereas with communication the error is bounded by that of a collective with effective size $n_0a$. Evidently, communication defines an effective number of cells $n_0$ over which information can be reliably conveyed, and a collective that grows beyond this size no longer improves its sensory precision.
These theoretical predictions were tested experimentally in collectives of epithelial cells [@ellison2016cell]. Mouse mammary epithelial cells were grown in organotypic culture and subjected to very shallow gradients of epidermal growth factor (Table \[sense\_table\]). It was shown that while single cells did not respond to these gradients, the multicellular collectives did: they exhibited a biased cell-branching response. Importantly, the response of large collectives was no more biased than the response of small collectives, supporting the theory with communication (Eq. \[eq:G2\]) over the theory without communication (Eq. \[eq:G1\]). The effective detector size was inferred to be $n_0 \approx 3.5$ cells, which is consistent with the size of these collectives in their natural context (the “end buds” of growing mammary ducts) [@lu2008genetic]. Interestingly, when the gap junctions between cells, which mediate the molecular communication, were blocked with each of several drugs, the biased responses were abolished [@ellison2016cell], demonstrating that the collective response was critically dependent on the cell-to-cell communication. Taken together, these results indicate that cell-to-cell communication is a necessary but imperfect enabler of collective gradient sensing. The results also speak to the power of simple physical theory to quantitatively explain collective cellular capabilities. Since epithelial cancers are known to invade collectively [@cheung2013collective], it remains an important open question whether this theory also describes the sensory behavior of tumor cell collectives.
Physical models of collective migration
=======================================
Metastatic invasion is a process of cell migration. Collective invasion, in turn, is a process of collective migration. Therefore, it is important to understand not only the collective sensing capabilities of tumor cells, but also the properties of their collective migration—and ideally the relation between the two. From a physical modeling perspective, describing collective cell dynamics is an interesting problem, because often rich and unexpected behavior can emerge from a few simple interaction rules between cells. Even in the absence of sensing, simple models have successfully explained observed collective behaviors such as cell streaming, cell sorting, cell sheet migration, wound healing, and cell aggregation [@kabla2012collective; @szabo2010collective; @basan2013alignment; @janulevicius2015short]. Here we focus on the collective dynamics that emerge when sensing plays a key role. In this case, a sensory cue results in polarization of a cell or cell collective via one of a variety of mechanisms [@jilkine2011comparison], and the dynamics are directed, i.e. migratory.
Mechanisms of collective migration
----------------------------------
Broadly speaking, the mechanisms of collective migration can be divided into three categories. First, cells may exhibit individual sensing and individual migration (Fig. \[models\]A). Here, each cell can perform gradient sensing and migration individually, although the precision may be low. When many such cells are placed in a group, the group migration can be enhanced and focused by local interactions between the cells. Even if each individual cell has low sensory and migratory precision, the precision of the group as a whole is high due to the interactions. Collisions act to average over the errors in individual cells’ noisy measurements, thereby decoupling group behavior from single-cell properties. This mechanism is often termed “many wrongs,” and it is successful at explaining how group migratory behavior emerges from individual agents that act independently [@simons2004many; @coburn2013tactile]. As discussed later, the failure of a communication-blocking therapy could act as proof that a “many wrongs” method of collective migration is at work in tumor cell invasion.
{width=".75\textwidth"}
Second, cells may exhibit individual sensing but collective migration (Fig. \[models\]B). In this mechanism, each individual cell senses its own local environment, and tight mechanical interactions result in the emergent directed motion of the entire group. This mechanism is applicable to the collective migration of connected clusters of cells. For example, models of this type were recently developed by Camley et al. [@camley2015emergent] and by Malet-Engra et al. [@malet2015collective] to describe behavior seen in clusters of neural crest cells and lymphocytes, respectively. In this model, cells are tightly connected but are polarized away from neighboring cells due to contact inhibition of locomotion (CIL), the physical phenomenon of cells ceasing motion in the direction of cell-cell contact [@mayor2010keeping]. Individual cells sense a local chemoattractant concentration and attempt to migrate away from the group with a strength proportional to this concentration. However, the mechanical coupling keeps them together. In the presence of a concentration gradient, the imbalance in their migration strengths results in net directed motion (Fig. \[models\]B). Notably, this mechanism results in directed motion of a cluster even though individual cells cannot execute directed motion alone, since without other cells, there is no CIL to bias the motion.
Third, cells may exhibit collective sensing and collective migration (Fig. \[models\]C). As discussed above, multicellular groups exploit cell-to-cell communication to sense gradients collectively, thereby enhancing the precision of sensing. A feature of this collective sensing, e.g. via the multicellular LEGI mechanism discussed above [@ellison2016cell; @mugler2016limits], is that each cell has information on the extent to which it is up or down the gradient. Through CIL or other contact-mediated interactions, this information can translate directly into cell polarity, leading to more coherent collective migration than in the previous mechanism (Fig. \[models\]C vs. B). In fact, the multicellular LEGI model was used by Camley et al. [@camley2015emergent] to explore a model of this type. Adding collective sensing to their model of CIL-dependent migration gave the advantage that the repulsive tension on a cell cluster was adaptive and therefore remained constant as the cluster migrated to regions of higher chemical concentration.
Model implementations
---------------------
To study the above mechanisms quantitatively and compare predictions with experiments, one must turn to mathematical and computational modeling. Models of cell dynamics range from continuum or semi-continuum descriptions, which describe groups of cells as continuous tissues, to individual-based models, which describe cells as individual interacting entities [@maclaren2015models]. Physics-driven individual-based models generally fall into two categories: force-based models and energy-based models.
Force-based models (Fig. \[models\]D) typically represent cells as centers of mass or as collections of vertices. Cell dynamics evolve from forces acting on individual cells, which can be stochastic, and arise from internal features such as cell polarity, and external features such as mechanical interactions with other cells [@maclaren2015models]. Force-based models are able to reproduce multicellular behavior such as chemotaxis, wound healing, and cell aggregation [@camley2015emergent; @basan2013alignment; @janulevicius2015short]. Parameters are often directly relatable to experimental measurements, and the simplest models are often amenable to exact mathematical analysis [@camley2015emergent].
Energy-based models (Fig. \[models\]E) allow cell dynamics to emerge from the minimization of a potential energy with thermal noise (the so-called Monte Carlo scheme). A widely used example is the cellular Potts model (CPM) [@graner1992simulation; @swat2012multi], in which cells are represented as collections of co-aligned “spins” on a lattice (Fig. \[models\]E). Cells remain contiguous because it is energetically favorable for neighboring spins to be co-aligned. Biophysical features such as cell shape, cell-cell adhesion, and cell protrusions into the environment are modeled by introducing corresponding terms into the global potential energy. The CPM has successfully reproduced experimental observations of cell sorting, streaming, chemotaxis, and collective migration [@kabla2012collective; @maree2007cellular; @szabo2010collective]. In energy-based models, the parameters are set by calibrating emergent features, such as cell diffusion coefficients or average speeds, with experimental measurements [@szabo2010collective].
Although the physical limits to multicellular sensing are becoming better understood, the physical limits constraining multicellular migration are less clear. This remains an interesting open question, and answering it will require integrating the theories of sensing and communication with the models of collective migration described herein. For tumor cells in particular, an integrated physical theory of sensing and migration would prove immensely useful for identifying the key determinants of invasive capabilities. Identifying these determinants would help pinpoint the ways that these capabilities could be disrupted, using drugs and other therapies, as described next.
Drug sensitivity and implications for therapy
=============================================
We have seen that cells, including tumor cells, are remarkably precise sensors of molecules in their environment. This raises the question of how sensitive tumor cells are to drug molecules in their environment. What is the minimum drug concentration required not just for precise detection by a cell, but for causing a phenotypic change, such as cell death?
Experiments have shown that cancer cells are sensitive to very small drug concentrations. For example, lung carcinoma cells were exposed *in vitro* to various concentrations of the anti-cancer drug paclitaxel, also known as taxol, which acts to block mitosis in order to achieve cell death by disrupting microtubule regulation [@torres1998mechanisms]. Paclitaxel concentrations as low as $1$ nM were shown to affect microtubule dynamics of the cells. This concentration is commensurate with the smallest background concentrations in which cells can perform gradient sensing (Table \[sense\_table\]). Assuming a cell length of $20$ $\mu$m, which is typical of carcinoma cells [@leber2009molecular], this concentration corresponds to only a few thousand drug molecules in the volume of a cell (Table \[drug\_table\]). Evidently lung cancer cells are affected by drug concentrations that are near the fundamental limits of what can be sensed.
Drug Concentration Molecules per Cell
--------------------------------------------------------- -------------------- --------------------
Physical change [@torres1998mechanisms] $1$ nM $5$,$000$
Cell death [@grantab2006penetration] $10^4$ nM $5\times 10^7$
Cell death, nanoparticle delivery [@malam2009liposomes] $100$ nM $500$,$000$
Communication blockage [@ellison2016cell] $50$ nM $30$,$000$
Although cancer cells may be very sensitive to small drug concentrations, that does not translate to successful treatment. In order to achieve cell death, much larger drug concentrations are required. In the same study on lung carcinoma cells, cell death was observed for drug concentrations on the order of $10$ nM and greater. More typical drug concentrations required for cell death are on the order of micromolars. For instance, it has been shown *in vitro* that anticancer drug concentrations on the order of $10$ $\mu$M are required to kill at least $90\%$ of tumor cells [@grantab2006penetration]. With a cell length of $20$ $\mu$m, $10$ $\mu$M corresponds to tens of millions of drug molecules in the volume of a cell, four orders of magnitude greater than drug concentrations required to affect cell functionality (Table \[drug\_table\]). In order to effectively kill a solid tumor, very high drug doses are required.
Complicating matters is the fact that the tumor and its surrounding microenvironment comprise a complex and heterogeneous system. Although most cells in the human body are naturally within a few cell lengths of a blood vessel, due to high proliferation tumor cells may be upwards of tens of cell lengths away from a vessel [@minchinton2006drug]. This makes it difficult for drugs to reach the tumor. Moreover, the high density of many solid tumors causes gradients of drug concentration to form as a function of tumor radius [@kwon2012analysis]. This results in a reduced drug concentration at the center of the tumor and makes innermost tumor cells the most difficult to kill. A promising way to overcome this difficulty is through the use of nanoparticle drug delivery systems, which increase both the specificity and penetration of drugs to the tumor. Nanoparticle delivery has been shown [@malam2009liposomes] to achieve cell death with concentrations as low as $100$ nM. Although this concentration is lower than delivery without nanoparticles, it is still two orders of magnitude higher than the minimum concentration that causes physical changes in the cell (Table \[drug\_table\]). Even with targeted delivery, achieving drug-induced tumor cell death remains a challenging task.
Given this challenge, we hope to draw upon the physical insights reviewed herein to devise therapeutic strategies that are alternative or complementary to comprehensive cell death. Specifically, we imagine focusing on the metastatic invasion phase, and targeting the functions of invading tumor cells, including communication and migration, in addition to targeting cells’ overall viability, to produce better treatment (Fig. \[overview\]C). Communication is a particularly promising candidate, since it has recently been shown that cell-to-cell communication makes cancer cells more resistant to therapy and helps sustain tumor growth [@boelens2014exosome]. Indeed, the exchange of extracellular vesicles, which is a form of communication observed between tumor cells and stromal cells, has been linked to immune suppression, tumor development, angiogensis, and metastasis [@vader2014extracellular]. This suggests that disrupting cell-to-cell communication could be an effective strategy for stopping tumor progression or curbing metastatic invasion. Disrupting communication may not require concentrations as large as those necessary for cell death, which are difficult to maintain *in vivo* across the whole tumor. For example, as little as $50$ nM of the gap-junction-blocking drug Endothelin-1 is sufficient to remove collective responses in epithelial cells [@ellison2016cell]. This concentration is several orders of magnitude smaller than that required for comprehensive cell death, and it is on the order of concentrations that are effective with targeted nanoparticle delivery (Table \[drug\_table\]). Therefore, it is tempting to suggest that managing metastatic invasion by blocking communication or other cell functions is a more accessible therapeutic strategy than eradicating a tumor outright.
Although blocking intercellular communication pathways could curb the invasive potential of metastatic cells it is also important to address the ulterior consequences of this strategy. Gap junction intercellular communication (GJIC) is an important way for the environment to affect change on cells, maintaining tissue homeostasis and balancing cell growth and death [@krysko2005gap]. In cancerous cells GJIC is reduced, causing unregulated cell growth [@salameh2005pharmacology]. Interestingly, many existing cancer-combatting drugs are small enough to pass through cell gap junctions which permit molecules of sizes up to 1000 Dalton, but there is a lack of *in vivo* studies concerning the benefits and effects of gap junctions on cancer treatment [@salameh2005pharmacology]. It has been shown *in vitro* that GJIC can propagate cell-death signals through cancerous cells and that high connexin expression, the proteins that compose gap junctions, corresponds to high anticancer drug sensitivity [@krutovskikh2002gap; @krysko2005gap]. Therefore, it is important to consider the potential negative consequences of blocking intercellular communication in reducing metastatic invasion. It may be sufficient to administer an anticancer drug and a communication-blocking drug at different times in order to avoid negative side-effects. Although this puts a caveat on our proposal of communication-blocking drugs as a viable option for treating metastatic invasion it is important to recall that GJIC is not the only communication pathway available to cancerous cells: extracellular vesicle-meditated signaling pathways are potential alternates which could be targeted in place of GJIC [@boelens2014exosome; @vader2014extracellular].
Outlook
=======
In this review, we have taken a quantitative look at metastatic invasion as a sensing-and-migration process, which has allowed us to compare metastatic cells to other cell types in terms of their physical capabilities. We have seen that tumor cells can sense very shallow chemoattractant gradients, which may help guide metastatic invasion, but it remains unclear whether tumor cells operate near fundamental sensing limits, as bacteria and amoeba do. Recognizing that metastatic invasion can be collective, we have reviewed recent results on the physical limits to collective sensing, and we have identified the overarching mechanisms of collective migration. A key insight that emerges is that collective capabilities rely critically on cell-to-cell communication. This insight opens up alternative strategies for therapy that target specific cell capabilities such as communication, in addition to strategies that aim for comprehensive cell death.
A detailed presentation of the underlying physical mechanics for cell motility and chemotaxis are outside the scope of this review. Readers interested in these topics are referred to these excellent resources [@lauffenburger1996cell; @van2004chemotaxis; @firtel2000molecular; @chung2001signaling]. It is also important to note that in deriving the limits to concentration sensing we have assumed that the molecules of interest diffuse normally with fixed, space-independent diffusion coefficients. However, this may not always be the case in the tumor environment, where molecules can also experience anomalous diffusion [@chauhan2011delivery; @dix2008crowding].
Moving forward, it will be important to identify whether the physical theory of sensing reviewed herein can be applied in a predictive manner to tumor cells, and whether gradient sensing plays a dominant role during metastatic invasion. More generally, it will be necessary to integrate the theory of sensing with models of collective migration to predict quantitatively what groups of migratory cells can and cannot do. Finally, controlled experiments with metastatic cells are required to validate these predictions, and to assess the viability of alternative therapies that target specific cell functions in order to combat metastatic invasion. Our hope is that the integrated, physics-based perspective presented herein will help generate innovative solutions to the pervasive problem of metastatic disease.
Acknowledgments
===============
This work was supported by the Ralph W. and Grace M. Showalter Research Trust and the Purdue Research Foundation.
[10]{}
Nicola Aceto, Aditya Bardia, David T Miyamoto, Maria C Donaldson, Ben S Wittner, Joel A Spencer, Min Yu, Adam Pely, Amanda Engstrom, Huili Zhu, Brian W Brannigan, Ravi Kapur, Shannon L Stott, Toshi Shioda, Sridhar Ramaswamy, David T Ting, Charles P Lin, Mehmet Toner, Daniel A Haber, and Shyamala Maheswaran. Circulating tumor cell clusters are oligoclonal precursors of breast cancer metastasis. , 158(5):1110–1122, 2014.
Markus Basan, Jens Elgeti, Edouard Hannezo, Wouter-Jan Rappel, and Herbert Levine. Alignment of cellular motility forces with tissue flow as a mechanism for efficient wound healing. , 110(7):2452–2459, 2013.
Howard C Berg and Edward M Purcell. Physics of chemoreception. , 20(2):193, 1977.
Neil A Bhowmick, Eric G Neilson, and Harold L Moses. Stromal fibroblasts in cancer initiation and progression. , 432(7015):332–337, 2004.
William Bialek and Sima Setayeshgar. Physical limits to biochemical signaling. , 102(29):10040–10045, 2005.
Brendan A Bicknell, Peter Dayan, and Geoffrey J Goodhill. The limits of chemosensation vary across dimensions. , 6, 2015.
Mirjam C Boelens, Tony J Wu, Barzin Y Nabet, Bihui Xu, Yu Qiu, Taewon Yoon, Diana J Azzam, Christina Twyman-Saint Victor, Brianne Z Wiemann, Hemant Ishwaran, Petra J ter Brugge, Jos Jonkers, Joyce Slingerland, and Andy J Minn. Exosome transfer from stromal to breast cancer cells regulates therapy resistance pathways. , 159(3):499–513, 2014.
Brian A Camley, Juliane Zimmermann, Herbert Levine, and Wouter-Jan Rappel. Emergent collective chemotaxis without single-cell gradient sensing. , 2015.
Antonio Celani and Massimo Vergassola. Bacterial strategies for chemotaxis response. , 107(4):1391–1396, 2010.
Vikash P Chauhan, Triantafyllos Stylianopoulos, Yves Boucher, and Rakesh K Jain. Delivery of molecular and nanoscale medicine to tumors: transport barriers and strategies. , 2:281–298, 2011.
Kevin J Cheung, Edward Gabrielson, Zena Werb, and Andrew J Ewald. Collective invasion in breast cancer requires a conserved basal epithelial program. , 155(7):1639–1651, 2013.
Chang Y Chung, Satoru Funamoto, and Richard A Firtel. Signaling pathways controlling cell polarity and chemotaxis. , 26(9):557–566, 2001.
Luke Coburn, Luca Cerone, Colin Torney, Iain D Couzin, and Zoltan Neufeld. Tactile interactions lead to coherent motion and enhanced chemotaxis of migrating cells. , 10(4):046002, 2013.
John Condeelis and Jeffrey W Pollard. Macrophages: obligate partners for tumor cell migration, invasion, and metastasis. , 124(2):263–266, 2006.
FW Dahlquist, RA Elwell, and Peter S Lovely. Studies of bacterial chemotaxis in defined concentration gradients. a model for chemotaxis toward l-serine. , 4(3):329–342, 1976.
James A Dix and AS Verkman. Crowding effects on diffusion in solutions and cells. , 37:247–263, 2008.
David Ellison, Andrew Mugler, Matthew D Brennan, Sung Hoon Lee, Robert J Huebner, Eliah R Shamir, Laura A Woo, Joseph Kim, Patrick Amar, Ilya Nemenman, et al. Cell–cell communication enhances the capacity of cell ensembles to sense shallow gradients during morphogenesis. , page 201516503, 2016.
Robert G Endres and Ned S Wingreen. Accuracy of direct gradient sensing by single cells. , 105(41):15749–15754, 2008.
Robert G Endres and Ned S Wingreen. Accuracy of direct gradient sensing by cell-surface receptors. , 100(1):33–39, 2009.
Richard A Firtel and Chang Y Chung. The molecular genetics of chemotaxis: sensing and responding to chemoattractant gradients. , 22(7):603–615, 2000.
Peter Friedl, Joseph Locker, Erik Sahai, and Jeffrey E Segall. Classifying collective cancer cell invasion. , 14(8):777–783, 2012.
Christine Gilles, Myriam Polette, Jean-Marie Zahm, Jean-Marie Tournier, Laure Volders, Jean-Michel Foidart, and Philippe Birembaut. Vimentin contributes to human mammary epithelial cell migration. , 112(24):4615–4625, 1999.
Fran[ç]{}ois Graner and James A Glazier. Simulation of biological cell sorting using a two-dimensional extended potts model. , 69(13):2013, 1992.
Rama Grantab, Shankar Sivananthan, and Ian F Tannock. The penetration of anticancer drugs through tumor tissue as a function of cellular adhesion and packing density of tumor cells. , 66(2):1033–1039, 2006.
Douglas Hanahan and Robert A Weinberg. The hallmarks of cancer. , 100(1):57–70, 2000.
Douglas Hanahan and Robert A Weinberg. Hallmarks of cancer: the next generation. , 144(5):646–674, 2011.
Bo Hu, Wen Chen, Wouter-Jan Rappel, and Herbert Levine. Physical limits on cellular sensing of spatial gradients. , 105(4):048104, 2010.
Albertas Janulevicius, Mark van Loosdrecht, and Cristian Picioreanu. Short-range guiding can result in the formation of circular aggregates in myxobacteria populations. , 11, 2015.
Alexandra Jilkine and Leah Edelstein-Keshet. A comparison of mathematical models for polarization of single eukaryotic cells in response to guided cues. , 7(4):e1001121–e1001121, 2011.
Alexandre J Kabla. Collective cell migration: leadership, invasion and segregation. , page rsif20120448, 2012.
Kazunari Kaizu, Wiet de Ronde, Joris Paijmans, Koichi Takahashi, Filipe Tostevin, and Pieter Rein ten Wolde. The berg-purcell limit revisited. , 106(4):976–985, 2014.
Vladimir A Krutovskikh, Colette Piccoli, and Horashi Yamasaki. Gap junction intercellular communication propagates cell death in cancerous cells. , 21(13):1989–1999, 2002.
DV Krysko, Luc Leybaert, Peter Vandenabeele, and Katharina D’Herde. Gap junctions and the propagation of cell survival and cell death signals. , 10(3):459–469, 2005.
Il Keun Kwon, Sang Cheon Lee, Bumsoo Han, and Kinam Park. Analysis on the current status of targeted drug delivery to tumors. , 164(2):108–114, 2012.
Douglas A Lauffenburger and Alan F Horwitz. Cell migration: a physically integrated molecular process. , 84(3):359–369, 1996.
Mathias Felix Leber and Thomas Efferth. Molecular principles of cancer invasion and metastasis (review). , 34(4):881–895, 2009.
Claire Legrand, Christine Gilles, Jean-Marie Zahm, Myriam Polette, Anne-C[é]{}cile Buisson, Herv[é]{} Kaplan, Philippe Birembaut, and Jean-Marie Tournier. Airway epithelial cell migration dynamics: Mmp-9 role in cell–extracellular matrix remodeling. , 146(2):517–529, 1999.
Andre Levchenko and Pablo A Iglesias. Models of eukaryotic gradient sensing: application to chemotaxis of amoebae and neutrophils. , 82(1):50–63, 2002.
Pengfei Lu, Andrew J Ewald, Gail R Martin, and Zena Werb. Genetic mosaic analysis reveals fgf receptor 2 function in terminal end buds during mammary gland branching morphogenesis. , 321(1):77–87, 2008.
Oliver J Maclaren, AG Fletcher, HM Byrne, and Philip K Maini. Models, measurement and inference in epithelial tissue dynamics. , 2015.
Yogeshkumar Malam, Marilena Loizidou, and Alexander M Seifalian. Liposomes and nanoparticles: nanosized vehicles for drug delivery in cancer. , 30(11):592–599, 2009.
Gema Malet-Engra, Weimiao Yu, Amanda Oldani, Javier Rey-Barroso, Nir S Gov, Giorgio Scita, and Lo[ï]{}c Dupr[é]{}. Collective cell motility promotes chemotactic prowess and resistance to chemorepulsion. , 25(2):242–250, 2015.
Athanasius F. M. Mar[é]{}e, Ver[ô]{}nica A. Grieneisen, and Paulien Hogeweg. The [Cellular]{} [Potts]{} [Model]{} and [Biophysical]{} [Properties]{} of [Cells]{}, [Tissues]{} and [Morphogenesis]{}. In Alexander R. A. Anderson, Mark A. J. Chaplain, and Katarzyna A. Rejniak, editors, [*Single-[Cell]{}-[Based]{} [Models]{} in [Biology]{} and [Medicine]{}*]{}, Mathematics and [Biosciences]{} in [Interaction]{}. Birkh[ä]{}user Basel, 2007. DOI: 10.1007/978-3-7643-8123-3\_5.
Roberto Mayor and Carlos Carmona-Fontaine. Keeping in touch with contact inhibition of locomotion. , 20(6):319–328, 2010.
Colin P McCann, Paul W Kriebel, Carole A Parent, and Wolfgang Losert. Cell speed, persistence and information transmission during signal relay and collective migration. , 123(10):1724–1731, 2010.
Andrew I Minchinton and Ian F Tannock. Drug penetration in solid tumours. , 6(8):583–592, 2006.
Andrew Mugler, Andre Levchenko, and Ilya Nemenman. Limits to the precision of gradient sensing with spatial communication and temporal integration. , page 201509597, 2016.
Carole A Parent. Making all the right moves: chemotaxis in neutrophils and dictyostelium. , 16(1):4–13, 2004.
William J Polacheck, Joseph L Charest, and Roger D Kamm. Interstitial flow influences direction of tumor cell migration through competing mechanisms. , 108(27):11115–11120, 2011.
Marten Postma, Jeroen Roelofs, Joachim Goedhart, Theodorus WJ Gadella, Antonie JWG Visser, and Peter JM Van Haastert. Uniform camp stimulation of dictyostelium cells induces localized patches of signal transduction and pseudopodia. , 14(12):5019–5027, 2003.
Alberto Puliafito, Alessandro De Simone, Giorgio Seano, Paolo Armando Gagliardi, Laura Di Blasio, Federica Chianale, Andrea Gamba, Luca Primo, and Antonio Celani. Three-dimensional chemotaxis-driven aggregation of tumor cells. , 5, 2015.
William J Rosoff, Jeffrey S Urbach, Mark A Esrick, Ryan G McAllister, Linda J Richards, and Geoffrey J Goodhill. A new chemotaxis assay shows the extreme sensitivity of axons to molecular gradients. , 7(6):678–682, 2004.
Aida Salameh and Stefan Dhein. Pharmacology of gap junctions. new pharmacological targets for treatment of arrhythmia, seizure and cancer? , 1719(1):36–58, 2005.
Adrian C Shieh and Melody A Swartz. Regulation of tumor invasion by interstitial fluid flow. , 8(1):015012, 2011.
Jacqueline D Shields, Mark E Fleury, Carolyn Yong, Alice A Tomei, Gwendalyn J Randolph, and Melody A Swartz. Autologous chemotaxis as a mechanism of tumor cell homing to lymphatics via interstitial flow and autocrine ccr7 signaling. , 11(6):526–538, 2007.
Andrew M Simons. Many wrongs: the advantage of group navigation. , 19(9):453–455, 2004.
Loling Song, Sharvari M Nadkarni, Hendrik U B[ö]{}deker, Carsten Beta, Albert Bae, Carl Franck, Wouter-Jan Rappel, William F Loomis, and Eberhard Bodenschatz. Dictyostelium discoideum chemotaxis: threshold for directed motion. , 85(9):981–989, 2006.
Maciej H Swat, Gilberto L Thomas, Julio M Belmonte, Abbas Shirinifard, Dimitrij Hmeljak, and James A Glazier. Multi-scale modeling of tissues using compucell3d. , 110:325, 2012.
A Szab[ó]{}, R [Ü]{}nnep, E M[é]{}hes, WO Twal, WS Argraves, Y Cao, and A Czir[ó]{}k. Collective cell motion in endothelial monolayers. , 7(4):046007, 2010.
Keila Torres and Susan Band Horwitz. Mechanisms of taxol-induced cell death are concentration dependent. , 58(16):3620–3626, 1998.
Yuhai Tu. Quantitative modeling of bacterial chemotaxis: signal amplification and accurate adaptation. , 42:337, 2013.
Pieter Vader, Xandra O Breakefield, and Matthew JA Wood. Extracellular vesicles: emerging targets for cancer therapy. , 20(7):385–393, 2014.
Peter JM Van Haastert and Peter N Devreotes. Chemotaxis: signalling the way forward. , 5(8):626–634, 2004.
Peter JM Van Haastert and Marten Postma. Biased random walk by stochastic fluctuations of chemoattractant-receptor interactions at the lower limit of detection. , 93(5):1787–1796, 2007.
Shur-Jen Wang, Wajeeh Saadi, Francis Lin, Connie Minh-Canh Nguyen, and Noo Li Jeon. Differential effects of egf gradient profiles on mda-mb-231 breast cancer cell chemotaxis. , 300(1):180–189, 2004.
Katarina Wolf, Irina Mazo, Harry Leung, Katharina Engelke, Ulrich H Von Andrian, Elena I Deryugina, Alex Y Strongin, Eva-B Br[ö]{}cker, and Peter Friedl. Compensation mechanism in tumor cell migration mesenchymal–amoeboid transition after blocking of pericellular proteolysis. , 160(2):267–277, 2003.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Feynman’s path integral is generalized to quantum mechanics on $p$-adic space and time. Such $p$-adic path integral is analytically evaluated for quadratic Lagrangians. Obtained result has the same form as that one in ordinary quantum mechanics.'
---
-50pt
[**On $p$-adic path integral** ]{}
[*Institute of Physics, P.O.Box 57, 11001 Belgrade, Yugoslavia*]{}
[Dedicated to the memory]{}
[of N.N.Bogolyubov]{}
[**1.**]{} It is well known that dynamical evolution of any one-dimensional quantum-mechanical system, described by a wave function $\Psi (x,t)$, is given by $$\Psi (x'',t'') = \int {\cal K}(x'',t'';x',t')\Psi(x',t')dx' , \eqno(1)$$ where ${\cal K}(x'',t'';x',t')$ is the kernel of the corresponding unitary operator acting as follows: $$\Psi (t'') = {\cal U}(t'',t')\Psi (t') . \eqno(2)$$
$ {\cal K}(x'',t'';x',t')$ is also called Green’s function, or the quantum-mechanical propagator, and the probability amplitude to go a particle from a point $ (x',t')$ to a point $(x'',t'')$. One can easily deduce the following three general properties: $$\int{\cal K}(x'',t'';x,t) {\cal K}(x,t;x',t') dx = {\cal K}(x'',t'';x't') , \eqno(3)$$ $$\int {\cal K}^{*}(x'',t'';x',t') {\cal K}(x'',t'';x,t') dx'' = \delta (x'-x) , \eqno(4)$$ $${\cal K}(x'',t'';x',t'') = lim_{t'\to t''} {\cal K}(x'',t'';x',t') = \delta (x''-x') . \eqno(5)$$ Since all information on quantum dynamics can be deduced from the propagator ${\cal K}(x'',t'';x',t')$ it can be regarded as the basic ingredient of quantum theory. In Feynman’s formulation [@feynman] of quantum mechanics, ${\cal K}(x'',t'';x',t')$ was postulated to be the path integral $${\cal K}(x'',t'';x',t') = \int \exp{\left(\frac{2\pi i}{h} \int_{t'}^{t''}
L(\dot{q},q,t)dt\right)} {\cal D}q , \eqno(6)$$ where $x'' = q(t'')$ and $x' = q(t')$, and $h$ is the Planck constant.
In its original form, the path integral (6) is the limit of the corresponding multiple integral of $n-1$ variables $q_i = q(t_i), \ \ (i=1,2,...,n-1),$ when $n\to\infty$. For the half of century of its history, the path integral has been a subject of permanent interest in theoretical and mathematical physics. At present days (see, e.g. [@firenca]) it is one of the most profound and promising approaches to foundations of quantum theory (in particular, quantum field theory and superstring theory). Feynman’s path integral is inevitable in formulation of $p$-adic [@vlad-vol] and adelic [@branko1] quantum mechanics. It is worth noting that just Feynman’s path integral approach enables natural foundation of quantum theory on $p$-adic and adelic spaces.
[**2.**]{} Recall that the set of rational numbers $Q$ plays an important role in mathematics as well as in physics. From algebraic point of view, $Q$ is the simplest number field of characteristic $0$. The usual absolute value and $p$-adic valuation ($p$ is any of prime numbers) exhaust all possible non-trivial norms on $Q$ [@schik]. Completion of $Q$ with respect to metrics induced by these norms leads to the field of real numbers $R$ and the fields of $p$-adic numbers $Q_p$, $(p=2,3,5,...)$. Thus $Q$ is dense in $R$ and all $Q_p$. From physical point of view, all numerical results of measurements are rational numbers. However, theoretical models of physical systems are traditionally constructed using real and complex numbers. One can ask the following question: Why real (and complex) numbers are so good in description of usual physical phenomena, and, is there any aspect of physical reality which has to be described by $p$-adic numbers. Construction of $p$-adic models and their appropriate interpretation can gradually give answer to this question. Since 1987, there have been many publications (for a review, see, e.g. [@freund; @vvz; @andrei] ) on possible applications of $p$-adic numbers in modern theoretical and mathematical physics. For a systematic approach to this subject, $p$-adic [@vlad-vol] and adelic [@branko1] quantum mechanics have been formulated.
Recall also that any $p$-adic number $x\in Q_p$ can be presented as the following infinite expansion $$x = p^\nu (x_0 + x_1p + x_2p^2 + \cdots ) , \ \ \nu\in Z ,$$ where $x_i = 0,1,...,p-1$ are digits. We will use the Gauss integral [@vvz] $$\int_{Q_p} \chi_p (\alpha x^2 + \beta x)dx =
\lambda_p(\alpha) \mid 2\alpha \mid_p^{-\frac{1}{2}}
\chi_p\left( - \frac{\beta^2}{4\alpha}\right) , \ \ \alpha\neq 0 ,$$ where $\chi_p (a) = \exp (2\pi i \{ a\}_p)$ is the additive character, and $\{ a\}_p$ is the fractional part of $a\in Q_p$. $\lambda_p(x)$ is a complex-valued arithmetic function (for a definition, see [@vvz]) with the following properties: $$\lambda_p(0) = 1 ,\ \lambda_p(a^2x) = \lambda_p(x) ,
\ \lambda_p(x)\lambda_p(y) = \lambda_p(x+y)\lambda_p(x^{-1}+y^{-1}) ,
\ \lambda_p^{*}(x)\lambda_p(x) = 1 .$$
[**3.**]{} $p$-Adic quantum mechanics, we are interested in, contains complex-valued functions of $p$-adic arguments. There is not the corresponding Schr$\ddot{o}$dinger equation, but Feynman’s path integral approach seems to be quite natural. Feynman’s path integral for $p$-adic propagator ${\cal K}_p(x'',t'';x',t')$, where ${\cal K}_p$ is complex-valued and $x'',x',t'',t'$ are $p$-adic variables, is a direct $p$-adic generalization of (6), i.e. $${\cal K}_p (x'',t'';x',t') = \int \chi_p \left( -\frac{1}{h}
\int_{t'}^{t''} L(\dot{q},q,t)dt\right) {\cal D}q , \eqno(7)$$ where $ \chi_p(a)$ is $p$-adic additive character. The Planck constant $h$ in (6) and (7) is the same rational number. Integral $ \int_{t'}^{t''}L(\dot{q},q,t)dt$ we consider as the difference of antiderivative (without pseudoconstants) of $L(\dot{q},q,t) $ in final $(t'')$ and initial $(t')$ times. ${\cal D}q = \prod_{i=1}^{n-1}dq(t_i)$, where $dq(t_i)$ is the $p$-adic Haar measure. Thus, $p$-adic path integral is the limit of the multiple Haar integral when $n\to\infty$. To calculate (7) in this way one has to introduce some order on $t\in Q_p$, and it is successfully done in Ref. [@zelenov]. On previous investigations of $p$-adic path integral one can see [@branko2]. Our main task here is derivation of the exact result for $p$-adic Feynman’s path integral (7) for the general case of Lagrangians $L(\dot{q},q,t)$, which are quadratic polynomials in $\dot{q}$ and $q$, without making time discretization.
A general quadratic Lagrangian can be written as follows: $$L(\dot{q},q,t) = \frac{1}{2}\frac{\partial^2 L_0}{\partial\dot{q}^2}\dot{q}^2
+ \frac{\partial L_0}{\partial\dot{q}} \dot{q} +
\frac{\partial^2 L_0}{\partial\dot{q}\partial q}\dot{q}q + L_0
+ \frac{\partial L_0}{\partial q} q +
\frac{1}{2}\frac{\partial^2 L_0}{\partial q^2} q^2 ,
\eqno(8)$$ where index $0$ denotes that the Taylor expansion of $L(\dot{q},q,t)$ is around $\dot{q}=q=0$. The Euler-Lagrange equation of motion is $$\frac{\partial^2 L_0}{\partial\dot{q}^2} \ddot{q} +
\frac{d}{dt} \left( \frac{\partial^2 L_0}{\partial\dot{q}^2} \right) \dot{q} +
\left[ \frac{d}{dt} \left( \frac{\partial^2 L_0}{\partial\dot{q}\partial q} \right) -
\frac{\partial^2 L_0}{\partial q^2} \right] q = \frac{\partial L_0}{\partial q}
- \frac{d}{dt}\left( \frac{\partial L_0}{\partial\dot{q}} \right) .
\eqno(9)$$ General solution of (9) is $$q \equiv x(t) = C_1 x_1(t) + C_2 x_2(t) + w(t) , \eqno(10)$$ where $x_1(t)$ and $x_2(t)$ are two linearly independent solutions of the corresponding homogeneous equation, and $w(t)$ is a particular solution of the complete equation (9). Note that $x(t)$ denotes the classical trajectory. Imposing the boundary conditions $x'=x(t')$ and $x''=x(t'')$, constants of integration $C_1$ and $C_2$ become: $$C_1 \equiv C_1 (t'',t') = \frac{(x'-w')x''_2 - (x''-w'')x'_2}{x''_2x'_1 - x''_1x'_2}, \eqno(11a)$$ $$C_2 \equiv C_2 (t'',t') = \frac{(x''-w'')x'_1 - (x'-w')x''_1}{x''_2x'_1 - x''_1x'_2}. \eqno(11b)$$ Since $C_1(t'',t')$ and $C_2(t'',t')$ are linear in $x''$ and $x'$, the corresponding classical action $\bar{S}(x'',t'';x',t') = \int_{t'}^{t''} L(\dot{x},x,t)dt$ is quadratic in $x''$ and $x'$. Note that the above expressions have the same form in $R$ and $Q_p$.
Quantum fluctuations lead to deviations of classical trajectory and any quantum path may be presented as $q(t) = x(t) + y(t)$, where $y'=y(t')=0$ and $y''=y(t'')=0$. The corresponding Taylor expansion of $S[q]$ around classical path $x(t)$ is $$S[q] = S[x+y] = S[x] + \frac{1}{2!} \delta^2 S[x] = S[x] +
\frac{1}{2}\int_{t'}^{t''} \left( \dot{y}\frac{\partial}{\partial \dot{q}} +
y\frac{\partial}{\partial q} \right)^{(2)} L(\dot{q},q,t)dt , \eqno(12)$$ where we used $\delta S[x] = 0$. We have now $${\cal K}_p (x'',t'';x',t') = \chi_p \left( -\frac{1}{h} S[x] \right) \int
\chi_p \left(- \frac{1}{2h} \int_{t'}^{t''} \left( \dot{y}\frac{\partial}{\partial\dot{q}} +
y \frac{\partial}{\partial q} \right)^{(2)}L(\dot{q},q,t)dt \right) {\cal D}y
\eqno(13)$$ with $ y''=y'=0 $ and $S[x] = \bar{S}(x'',t'';x',t')$.
Note that ${\cal K}_p(x'',t'';x',t')$ has the form $${\cal K}_p(x'',t'';x',t') = N_p(t'',t') \chi_p \left( -\frac{1}{h}
\bar{S}(x'',t'';x',t')\right) , \eqno(14)$$ where $N_p(t'',t')$ does not depend on end points $x''$ and $x'$. To calculate $N_p(t'',t')$ we use conditions (3) and (4). Substituting ${\cal K}_p(x'',t'';x',t')$ into (4) we obtain (for details, see [@branko2]): $$N_p(t'',t') = \left\vert\frac{1}{h} \frac{\partial^2\bar{S}}
{\partial x''\partial x'}(x'',t'';x',t') \right\vert_p^{\frac{1}{2}}
A_p(t'',t') ,\eqno(15)$$ where $\vert A_p(t'',t')\vert_\infty =1$, ($ \vert \cdot\vert_p$ and $\vert\cdot\vert_\infty$ denote $p$-adic and absolute value, respectively). Replacing (15) in equation (3) we get conditions: $$A_p(t'',t) A_p(t,t')\lambda_p (\alpha) = A_p(t'',t') , \eqno(16)$$ $$\left\vert \frac{1}{h}\frac{\partial^2\bar{S}}{\partial x'' \partial x}(x'',t'';x,t)
\right\vert_p^{\frac{1}{2}} \left\vert \frac{1}{h} \frac{\partial^2\bar{S}}{\partial x\partial x'}
(x,t;x',t')\right\vert_p^{\frac{1}{2}} \vert 2\alpha \vert_p^{-\frac{1}{2}}
= \left\vert \frac{1}{h} \frac{\partial^2\bar{S}}{\partial x'' \partial x'}
(x'',t'';x',t')\right\vert_p^{\frac{1}{2}}, \eqno(17)$$ where $$\alpha = - \frac{1}{2h} \left[ \frac{\partial^2\bar{S}}{\partial x^2}(x'',t'';x,t)
+\frac{\partial^2\bar{S}}{\partial x^2}(x,t;x',t') \right] . \eqno(18)$$ Analysing the above formulae we obtain [@branko2] $$A_p(t'',t') = \lambda_p\left( -\frac{1}{2h} \frac{\partial^2\bar{S}}{\partial x''\partial x'}
(x'',t'';x',t')\right) . \eqno(19)$$ For details of a quite rigorous derivation of (19), see [@branko3].
As the final result we have $${\cal K}_p(x'',t'';x',t') = \lambda_p\left( -\frac{1}{2h} \frac{\partial^2\bar{S}}
{\partial x'' \partial x'} \right) \left\vert \frac{1}{h} \frac{\partial^2\bar{S}}
{\partial x'' \partial x'} \right\vert_p^{\frac{1}{2}}
\chi_p\left( -\frac{1}{h} \bar{S}(x'',t'';x',t') \right)
\eqno(20)$$ which is the $p$-adic Feynman path integral for quadratic Lagrangians. The corresponding path integral of ordinary quantum mechanics [@branko2] can be transformed into the same form as (20), i.e. in such case index $p$ is replaced by index $\infty$. This supports Volovich’s conjecture [@volovich] that fundamental physical laws should be invariant under interchange of number fields $Q_p$ and $R$.
. Author wishes to thank the organizers of the Bogolyubov Conference: Problems of Theoretical and Mathematical Physics, for invitation to participate in Moscow and Dubna parts of the Conference.
[99]{} R.H.Feynman, [*Rev. Mod. Phys.*]{} [**20**]{} (1948)367. ”Path Integrals from peV to TeV: 50 Years after Feynman’s Paper”, Eds. R.Casalbuoni et al. (World Scientific, Singapore, 1999). V.S.Vladimirov and I.V.Volovich, [*Commun. Math. Phys.*]{} [**123**]{} (1989)659. B.Dragovich, [*Teor. Mat. Fiz.*]{} [**101**]{} (1994)349; [*Int. J. Mod. Phys.*]{} [**A10**]{} (1995)2349. W.H.Schikhof, [*Ultrametric Calculus*]{} (Cambridge Univ. Press, 1984). L.Brekke and P.G.O.Freund, [*Phys. Reports*]{} [**233**]{} (1993)1. V.S.Vladimirov, I.V.Volovich and E.I.Zelenov, [*$p$-Adic Analysis and Mathematical Physics*]{} (World Scientific, Singapore, 1994). A.Khrennikov, [*$p$-Adic Valued Distributions in Mathematical Physics*]{} (Kluwer, 1994). E.I.Zelenov, [*J. Math. Phys.*]{} [**32**]{} (1991)147. G.S.Djordjević and B.Dragovich, [*Mod. Phys. Lett.*]{} [**A12**]{} (1997)1455. G.S.Djordjević, B.Dragovich and Lj.Nešić, [*$p$-Adic Generalization of Feynman’s Path Integral*]{}, in preparation. I.V.Volovich, [*Number Theory as the Ultimate Physical Theory*]{}, preprint CERN-Th. 4781/87, July 1987.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A Linear-quadratic optimal control problem is considered for mean-field stochastic differential equations with deterministic coefficients. By a variational method, the optimality system is derived, which turns out to be a linear mean-field forward-backward stochastic differential equation. Using a decoupling technique, two Riccati differential equations are obtained, which are uniquely solvable under certain conditions. Then a feedback representation is obtained for the optimal control.'
author:
- |
Jiongmin Yong\
Department of Mathematics, University of Central Florida, Orlando, FL 32816, USA.
title: |
**A Linear-Quadratic Optimal Control Problem\
for Mean-Field Stochastic Differential Equations[^1]**
---
**Keywords. Mean-field stochastic differential equation, linear-quadratic optimal control, Riccati differential equation, feedback representation.**
**AMS Mathematics subject classification. 49N10, 49N35, 93E20.**
Introduction.
=============
Let $(\O,\cF,\dbP,\dbF)$ be a complete filtered probability space, on which a one-dimensional standard Brownian motion $W(\cd)$ is defined with $\dbF\equiv\{\cF_t\}_{t\ge0}$ being its natural filtration augmented by all the $\dbP$-null sets. Consider the following controlled linear stochastic differential equation (SDE, for short): {n dX(s)={A(s)X(s)+B(s)u(s)+A(s)+B(s)}ds\
+{A\_1(s)X(s)n+n B\_1(s)u(s)n+nA\_1(s)n+nB\_1(s)}dW(s),s,\
X(0)=x,.where $A(\cd),B(\cd),\h A(\cd),\h B(\cd),A_1(\cd),B_1(\cd),\h
A_1(\cd),\h B_1(\cd)$ are given deterministic matrix valued functions. In the above, $X(\cd)$, valued in $\dbR^n$, is the [*state process*]{}, and $u(\cd)$, valued in $\dbR^m$, is the [*control process*]{}.
We note that $\dbE[X(\cd)]$ and $\dbE[u(\cd)]$ appear in the state equation. Such an equation is referred to as a linear controlled mean-field (forward) SDE (MF-FSDE, for short). MF-FSDEs can be used to describe particle systems at mesoscopic level, which is of great importance in applications. Historically, the later-called McKean–Vlasov SDE, a kind of MF-FSDE, was suggested by Kac [@Kac; @1956] in 1956 as a stochastic toy model for the Vlasov kinetic equation of plasma and the study of which was initiated by McKean [@McKean; @1966] in 1966. Since then, many authors made contributions on McKean–Vlasov type SDEs and applications, see, for examples, Dawson [@Dawson; @1983], Dawson–Gärtner [@Dawson-Gartner; @1987], Gártner [@Gartner; @1988], Scheutzow [@Scheutzow; @1987], Graham [@Graham; @1992], Chan [@Chan; @1994], Chiang [@Chiang; @1994], Ahmed–Ding [@Ahmed-Ding; @1995], to mention a few. In recent years, related topics and problems have attracted more and more researchers’ attentions, see, for examples, Veretennikov [@Veretennikov; @2003], Huang–Malhamé–Caines [@Huang-Malhame-Caines; @2006], Mahmudov–McKibben [@Mahmudov-Mckibben; @2007], Buckdahn-Djehiche-Li-Peng [@Buckdahn-Djehiche-Li-Peng; @2009], Buckdahn-Li-Peng [@Buckdahn-Li-Peng; @2009], Borkar–Kumar [@Borkar-Kumar; @2010], Crisan–Xiong [@Crisan-Xiong; @2010], Kotelenez–Kurtz [@Kotelenez-Kurtz; @2010], Kloeden–Lorenz [@Kloeden-Lorenz; @2010], and so on. More interestingly, control problems of McKean–Vlasov equation or MF-FSDEs were investigated by Ahmed–Ding [@Ahmed-Ding; @2001], Ahmed [@Ahmed; @2007], Buckdahn–Djehiche–Li [@Buckdahn-Djehiche-Li; @2010], Park–Balasubramaniam–Kang [@Park-Balasubramaniam-Kang; @2008], Andersson–Djehiche [@Andersson-Djehche; @2011], Meyer-Brandis–Oksendal–Zhou [@Meyer-Brandis-Oksandal-Zhou; @2011], and so on. This paper can be regarded as an addition to the study of optimal control for MF-FSDEs.
For the state equation (\[1.1\]), we introduce the following: $$\cU[0,T]=L^2_\dbF(0,T;\dbR^m)\deq\Big\{u:[0,T]\times\O\to\dbR^m\bigm|\dbE\int_0^T|u(s)|^2ds<\infty\Big\}.$$ Any $u(\cd)\in\cU[0,T]$ is called an [*admissible control*]{}. Under mild conditions, one can show that (see below) for any $(x,u(\cd))\in\dbR^n\times\cU[0,T]$, (\[1.1\]) admits a unique solution $X(\cd)=X(\cd\,;x,u(\cd))$. We introduce the following cost functional: J(x;u())={\_0\^T$$\lan Q(s)X(s),X(s)\ran+\lan\h
Q(s)\dbE[X(s)],\dbE[X(s)]\ran+\lan R(s)u(s),u(s)\ran\\
\ns\ds\qq\qq\qq\qq\q+\lan\h R(s)\dbE[u(s)],\dbE[u(s)]\ran$$ds+GX(T),X(T)+G,},with $Q(\cd),R(\cd),\h Q(\cd),\h R(\cd)$ being suitable deterministic symmetric matrix-valued functions, and $G,\h G$ being symmetric matrices. Our optimal control problem can be stated as follows:
**Problem (MF). For given $x\in\dbR^n$, find a $\bar
u(\cd)\in\cU[0,T]$ such that J(x;|u())=\_[u()]{}J(x;u()).**
Any $\bar u(\cd)\in\cU[0,T]$ satisfying the above is called an [*optimal control*]{} and the corresponding state process $\bar
X(\cd)\equiv X(\cd\,;x,\bar u(\cd))$ is called an [*optimal state process*]{}; the pair $(\bar X(\cd),\bar u(\cd))$ is called an [*optimal pair*]{}.
From the above-listed literature, one has some motivations for the inclusion of $\dbE[X(\cd)]$ and $\dbE[u(\cd)]$ in the state equation. We now briefly explain a motivation of including $\dbE[X(\cd)]$ and $\dbE[u(\cd)]$ in the cost functional. We recall that for a classical LQ problem with the state equation {n dX(s)=$$A(s)X(s)+B(s)u(s)$$ds+$$A_1(s)X(s)+B_1(s)u(s)$$dW(s),s,\
X(0)=x,.one has the following cost functional J\_0(x;u())={\_0\^T$$\lan
Q_0(s)X(s),X(s)\ran+\lan R_0(s)u(s),u(s)\ran$$ds+G\_0X(T),X(T)}.For such a corresponding optimal control problem, it is natural to hope that the optimal state process and/or control process are not too sensitive with respect to the possible variation of the random events. One way to achieve this is try to make the variations $\var[X(\cd)]$ and $\var[u(\cd)]$ small. Therefore, one could include the $\var[X(\cd)]$ and $\var[u(\cd)]$ in the cost functional. Consequently, one might want to replace (\[1.5\]) by the following: J\_0(x;u())={\_0\^T$$\lan
Q_0(s)X(s),X(s)\ran+q(s)\var[X(s)]+\lan
R_0(s)u(s),u(s)\ran+\rho(s)\var[u(s)]$$ds\
+G\_0X(T),X(T)+g},for some (positive) weighting factors $q(\cd)$, $\rho(\cd)$, and $g$. Since $$\var[X(s)]=\dbE|X(s)|^2-\(\dbE[X(s)]\)^2,$$ and similar things hold for $\var[X(T)]$ and $\var[u(s)]$, we see that $$\ba{ll}
\ns\ds\h J_0(x;u(\cd))=\dbE\Big\{\int_0^T\[\lan
[Q_0(s)+q(s)I]X(s),X(s)\ran-q(s)\(\dbE[X(s)]\)^2+\lan
[R_0(s)+\rho(s)I]u(s),u(s)\ran\\
\ns\ds\qq\qq\qq\qq\qq-\rho(s)\(\dbE[u(s)]\)^2\]ds+\lan
[G_0+gI]X(T),X(T)\ran-g\(\dbE[X(T)]\)^2\Big\}.\ea$$ Clearly, the above is a special case of (\[1.2\]) with $$\ba{ll}
\ns\ds Q(\cd)=Q_0(\cd)+q(\cd)I,\q R(\cd)=R_0(\cd)+\rho(\cd)I,\q G=G_0+gI,\\
\ns\ds\h Q(\cd)=-q(\cd)I,\q\h R(\cd)=-\rho(\cd)I,\q\h G=-gI.\ea$$ Note that in the above case, $\h Q(\cd)$, $\h R(\cd)$, and $\h G$ are not positive semi-definite. Because of this, we do not assume the positive semi-definiteness for $\h Q(\cd)$, $\h R(\cd)$, and $\h
G$.
The purpose of this paper is to study Problem (MF). We will begin with the well-posedness of the state equation and the solvability of Problem (MF) in Section 2. Then, in Section 3, we will establish necessary and sufficient conditions for optimal pairs. Naturally, a linear backward stochastic differential equation of mean-filed type (MF-BSDE, for short) will be derived. Consequently, the optimality system turns out to be a coupled mean-field type forward-backward stochastic differential equation (MF-FBSDE, for short). Inspired by the invariant imbedding [@Bellman; @1960], and the Four-Step Scheme [@MY; @1999], we derive two Riccati differential equations in Section 4, so that the optimal control is represented as a state feedback form. Well-posedness of these Riccati equations will be established. We also present a direct verification of optimality for the state feedback control by means of completing squares. In Section 5, we will look at a modified LQ problem which is one motivation of the current paper.
Preliminaries.
==============
In this section, we make some preliminaries. First of all, for any Euclidean space $H=\dbR^n,\dbR^{n\times m},\cS^n$ (with $\cS^n$ being the set of all $(n\times n)$ symmetric matrices), we let $L^p(0,t;H)$ be the set of all $H$-valued functions that are $L^p$-integrable, $p\in[1,\infty]$. Next, we introduce the following spaces: $$\ba{ll}
\ns\ds\cX_t\equiv
L^2_{\cF_t}(\O;\dbR^n)=\Big\{\xi:\O\to\dbR^n\bigm|\xi\hb{ is
$\cF_t$-measurable, }\dbE|\xi|^2<\infty\Big\},\\
\ns\ds\cU_t\equiv
L^2_{\cF_t}(\O;\dbR^m)=\Big\{\eta:\O\to\dbR^m\bigm|\eta\hb{ is
$\cF_t$-measurable, }\dbE|\eta|^2<\infty\Big\},\\
\ns\ds
L^2_\dbF(0,t;\dbR^n)=\Big\{X:[0,t]\times\O\to\dbR^n\bigm|X(\cd)\hb{
is $\dbF$-adapted,
}\dbE\int_0^t|X(s)|^2ds<\infty\Big\},\\
\ns\ds\cX[0,t]\equiv
C_\dbF([0,t];\dbR^n)=\Big\{X:[0,t]\to\cX_t\bigm|X(\cd)\hb{ is
$\dbF$-adapted,}\\
\ns\ds\qq\qq\qq\qq\qq\qq\q\hb{ $s\mapsto X(s)$ is continuous, }
\sup_{s\in[0,t]}\dbE|X(s)|^2<\infty\Big\},\\
\ns\ds\h\cX[0,t]\equiv
L^2_\dbF(\O;C([0,t];\dbR^n))=\Big\{X:[0,t]\times\O\to\dbR^n\bigm|X(\cd)\hb{
is $\dbF$-adapted, }\\
\ns\ds\qq\qq\qq\qq\qq\qq\qq\q X(\cd)\hb{ has continuous paths,
}\dbE\[\sup_{s\in[0,t]}|X(s)|^2\]<\infty\Big\}.\ea$$ Note that in the definition of $\cX[0,t]$, the continuity of $s\mapsto X(s)$ means that as a map from $[0,t]$ to $\cX_t$, it is continuous. Whereas, in the definition of $\h\cX[0,t]$, $X(\cd)$ has continuous paths means that for almost sure $\o\in\O$, $s\mapsto
X(s,\o)$ is continuous. It is known that $$\ba{ll}
\ns\ds\h\cX[0,t]\subseteq\cX[0,t]\subseteq L^2_\dbF(0,t;\dbR^n),\qq
\h\cX[0,t]\ne\cX[0,t]\ne L^2_\dbF(0,t;\dbR^n).\ea$$ We now introduce the following assumptions for the coefficients of the state equation.
[**(H1)**]{} The following hold: { A(),A(),A\_1(),A\_1()L\^(0,T;\^[nn]{}),\
B(),B(),B\_1(),B\_1()L\^(0,T;\^[nm]{})..
[**(H1)$'$**]{} The following hold: { A(),A()L\^2(0,T;\^[nn]{}),A\_1(),A\_1()L\^(0,T;\^[nn]{}),\
B(),B()L\^2(0,T;\^[nm]{}),B\_1(),B\_1()L\^(0,T;\^[nm]{})..
[**(H1)$''$**]{} The following hold: { A(),A()L\^1(0,T;\^[nn]{}),A\_1(),A\_1()L\^2(0,T;\^[nn]{}),\
B(),B()L\^2(0,T;\^[nm]{}),B\_1(),B\_1()L\^(0,T;\^[nm]{})..
Clearly, (H1) implies (H1)$'$ which further implies (H1)$''$. Namely, (H1)$''$ is the weakest assumption among the above three. Whereas, (H1) is the most common assumption. For the weighting matrices in the cost functional, we introduce the following assumption.
[**(H2)**]{} The following hold: Q(),Q()L\^(0,T;\^n),R(),R()L\^(0,T;\^m),G,G\^n.
[**(H2)$'$**]{} In addition to (H2), the following holds: { Q(s),Q(s)+Q(s)0,R(s),R(s)+R(s)0,s,\
G,G+G0..
[**(H2)$''$**]{} In addition to (H2), the following holds: For some $\d>0$, { Q(s),Q(s)+Q(s)0,R(s),R(s)+R(s)I,s,\
G,G+G0..
From (\[1.4\]), we see that $\h Q(\cd)$, $\h R(\cd)$, and $\h G$ are not necessarily positive semi-definite. Therefore, in (H2), we do not mention positive-definiteness of the involved matrices and matrix-valued functions.
Now, for any $X(\cd)\in L^2_\dbF(0,T;\dbR^n)$ and any $u(\cd)\in\cU[0,T]$, we define { (t)=\_0\^t$A(s)X(s)+\h
A(s)\dbE[X(s)]$ds\
+\_0\^t$A_1(s)X(s)+\h
A_1(s)\dbE[X(s)]$dW(s),t,\
(t)=\_0\^t$B(s)u(s)+\h
B(s)\dbE[u(s)]$ds\
+\_0\^t$B_1(s)u(s)+\h
B_1(s)\dbE[u(s)]$dW(s),t..The following result is concerned with operators $\cA$ and $\cB$.
**Lemma 2.1. *The following estimates hold as long as the involved norms on the right hand sides are meaningful: For any $t\in[0,T]$, X()\^2\_C$$\|A(\cd)\|_{L^2(0,t;\dbR^{n\times n})}^2+\|\h
A(\cd)\|_{L^2(0,t;\dbR^{n\times
n})}^2\\
\ns\ds\qq\qq\qq\qq+\|A_1(\cd)\|_{L^\infty(0,t;\dbR^{n\times
n})}^2+\|\h A_1(\cd)\|_{L^\infty(0,t;\dbR^{n\times
n})}^2$$X()\^2\_[L\^2\_(0,t;\^n)]{}, X()\_\^2C$$\|A(\cd)\|_{L^1(0,t;\dbR^{n\times n})}^2+\|\h
A(\cd)\|_{L^1(0,t;\dbR^{n\times
n})}^2\\
\ns\ds\qq\qq\qq\qq+\|A_1(\cd)\|_{L^2(0,t;\dbR^{n\times n})}^2+\|\h
A_1(\cd)\|_{L^2(0,t;\dbR^{n\times
n})}^2$$X()\_\^2,and u()\^2\_C$$\|B(\cd)\|_{L^2(0,t;\dbR^{n\times n})}^2+\|\h
B(\cd)\|_{L^2(0,t;\dbR^{n\times
n})}^2\\
\ns\ds\qq\qq\qq\qq+\|B_1(\cd)\|_{L^\infty(0,t;\dbR^{n\times
n})}^2+\|\h B_1(\cd)\|_{L^\infty(0,t;\dbR^{n\times
n})}^2$$u()\^2\_.Hereafter, $C>0$ represents a generic constant which can be different from line to line.***
*Proof. For any $t\in(0,T]$, and any $X(\cd)\in
L^2_\dbF(0,t;\dbR^n)$, $$\ba{ll}
\ns\ds\dbE\(\sup_{s\in[0,t]}|[\cA X(\cd)](s)|\)\le
C\Big\{\dbE\(\int_0^t|A(s)||X(s)|ds\)^2+\(\int_0^t|\h
A(s)||\dbE[X(s)]|ds\)^2\\
\ns\ds\qq+\dbE\int_0^t|A_1(s)|^2|X(s)|^2ds+\int_0^t|\h
A_1(s)|^2|\dbE[X(s)]|^2ds\Big\}\\
\ns\ds\le
C\Big\{\dbE\[\(\int_0^t|A(s)|^2ds\)\(\int_0^t|X(s)|^2ds\)\]+\(\int_0^t|\h
A(s)|^2ds\)\(\int_0^t|\dbE[X(s)]|^2ds\)\\
\ns\ds\qq+\(\sup_{s\in[0,t]}|A_1(s)|\)^2\int_0^t\dbE|X(s)|^2ds+\(\sup_{s\in[0,t]}|\h
A_1(s)|^2\)\int_0^t\dbE|X(s)|^2ds\Big\}\\
\ns\ds\le C\Big\{\(\int_0^t|A(s)|^2ds\)+\(\int_0^t|\h
A(s)|^2ds\)+\sup_{s\in[0,t]}|A_1(s)|^2+\sup_{s\in[0,t]}|\h
A_1(s)|^2\Big\}\int_0^t\dbE|X(s)|^2ds.\ea$$ Thus, estimate (\[estimate1\]) holds. Next, for any $X(\cd)\in
\cX[0,t]$, making use of Burkholder–Davis–Gundy’s inequality ([@Yong-Zhou; @1999]), we have $$\ba{ll}
\ns\ds\dbE\(\sup_{s\in[0,t]}|[\cA X(\cd)](s)|^2\)\le
C\Big\{\dbE\(\int_0^t|A(s)||X(s)|ds\)^2+\(\int_0^t|\h
A(s)||\dbE[X(s)]|ds\)^2\\
\ns\ds\qq\qq\qq\qq\qq+\dbE\int_0^t|A_1(s)|^2|X(s)|^2ds+\int_0^t|\h
A_1(s)|^2|\dbE[X(s)]|^2ds\Big\}\\
\ns\ds\le
C\Big\{\(\int_0^t|A(s)|ds\)\(\int_0^t|A(s)|\dbE|X(s)|^2ds\)+\(\int_0^t|\h
A(s)|ds\)^2\(\sup_{s\in[0,t]}|\dbE[X(s)]|^2\)\\
\ns\ds\qq\qq\qq\qq\qq+\int_0^t|A_1(s)|^2\dbE|X(s)|^2ds+\int_0^t|\h
A_1(s)|^2\dbE|X(s)|^2ds\Big\}\\
\ns\ds\le C\[\(\int_0^t|A(s)|ds\)^2+\(\int_0^t|\h
A(s)|ds\)^2+\int_0^t|A_1(s)|^2ds+\int_0^t|\h
A_1(s)|^2ds\]\(\sup_{s\in[0,t]}\dbE|X(s)|^2\).\ea$$ Hence, (\[estimate2\]) follows.*
Similar to the proof of (\[estimate1\]), we can prove (\[estimate3\]).
The above lemma leads to the following corollary.
**Corollary 2.2. *If [(H1)$''$]{} holds, then $\cA:\cX[0,T]\to\h\cX[0,T]$ and $\cB:\cU[0,T]\to\h\cX[0,T]$ are bounded. Further, if [(H1)$'$]{} holds, then $\cA:L^2_\dbF(0,T;\dbR^n)\to\h\cX[0,T]$ is also bounded. In particular, all the above hold if [(H1)]{} holds.***
Next, we define { I\_TX()=X(T),\
\_TX()=I\_TX()(T)\
=\_0\^T$A(s)X(s)+\h
A(s)\dbE[X(s)]$ds+\_0\^T$A_1(s)X(s)+\h
A_1(s)\dbE[X(s)]$dW(s),\
\_Tu()=I\_Tu()(T)\
=\_0\^T$B(s)u(s)+\h
B(s)\dbE[u(s)]$ds+\_0\^T$B_1(s)u(s)+\h
B_1(s)\dbE[u(s)]$dW(s)..It is easy to see that $$I_T:\cX[0,T]\to\cX_T$$ is bounded. According to Lemma 2.1, we have the following result.
**Corollary 2.3. *If [(H1)$''$]{} holds, then $\cA_T:\cX[0,T]\to\cX_T$ and $\cB_T:\cU[0,T]\to\cX_T$ are bounded. Further, if [(H1)$'$]{} holds, then $\cA_T:L^2_\dbF(0,T;\dbR^n)\to\cX_T$ is also bounded. In particular, all the above hold if [(H1)]{} holds.***
Recall that if $\xi\in\cX_T$, then there exists a unique $\z(\cd)\in
L^2_\dbF(0,T;\dbR^n)$ such that $$\xi=\dbE\xi+\int_0^T\z(s)dW(s).$$ We denote $$\z(s)=D_s\xi,\qq s\in[0,T],$$ and call it the [*Malliavin derivative*]{} of $\xi$ ([@Nualart]). Next, we have the following results which give representation of the adjoint operators of $\cA$, $\cB$, $\cA_T$, and $\cB_T$.
**Proposition 2.4. *The following hold: { (\^\*Y)(s)=\_s\^Tn$A(s)^TY(t)+\h
A(s)^T\dbE[Y(t)]+A_1(s)^TD_sY(t)+\h A_1(s)^T\dbE[D_sY(t)]$dt,\
(\^\*Y)(s)=\_s\^Tn$B(s)^TY(t)+\h
B(s)^T\dbE[Y(t)]+B_1(s)^TD_sY(t)+\h B_1(s)^T\dbE[D_sY(t)]$dt,\
(\_T\^\*)(s)=A(s)\^T+A(s)\^T+A\_1(s)\^TD\_s+A\_1(s)\^T,\
(\_T\^\*)(s)=B(s)\^T+B(s)\^T+B\_1(s)\^TD\_s+B\_1(s)\^T..***
*Proof. For any $Y(\cd)\in L^2_\dbF(0,T;\dbR^n)$, $$\ba{ll}
\ns\ds\lan X,\cA^*Y\ran=\lan\cA X,Y\ran=\dbE\int_0^T\lan[\cA
X](t),Y(t)\ran dt\\
\ns\ds=\1n\dbE\2n\int_0^T\3n\lan\2n\int_0^t\3n\(A(s)X(s)\1n+\1n\h
A(s)\dbE[X(s)]\)ds\2n+\2n\int_0^t\2n\(A_1(s)X(s)\1n+\1n\h
A_1(s)\dbE[X(s)]\)dW(s),Y(t)\ran dt\\
\ns\ds=\1n\dbE\2n\int_0^T\3n\int_s^T\3n\lan A(s)X(s)\1n+\1n\h
A(s)\dbE[X(s)],Y(t)\ran dtds\1n+\1n\dbE\2n\int_0^T\3n\int_s^T\3n\lan
A_1(s)X(s)\1n+\1n\h
A_1(s)\dbE[X(s)],D_sY(t)\ran dtds\\
\ns\ds=\dbE\2n\int_0^T\3n\lan X(s),\int_s^T\3n\(A(s)^TY(t)\1n+\2n\h
A(s)^T\dbE[Y(t)]\1n+\1n A_1(s)^TD_sY(t)\1n+\1n\h
A_1(s)^T\dbE[D_sY(t)]\)dt\ran ds.\ea$$ Thus, the representation of $\cA^*$ follows. Similarly, we can obtain the representation of $\cB^*$.*
Next, for any $\xi\in\cX_T$, $$\ba{ll}
\ns\ds\lan
X,\cA_T^*\xi\ran=\lan\cA_TX,\xi\ran\\
\ns\ds=\dbE\lan\int_0^T\(A(s)X(s)+\h
A(s)\dbE[X(s)]\)ds,\xi\ran+\dbE\lan\int_0^T\(A_1(s)X(s)+\h
A_1(s)\dbE[X(s)]\)dW(s),\xi\ran\\
\ns\ds=\dbE\int_0^T\lan X(s),A(s)^T\xi+\h
A(s)^T\dbE\xi+A_1(s)^TD_s\xi+\h A_1(s)^T\dbE[D_s\xi]\ran ds.\ea$$ Therefore, the representation of $\cA_T^*$ follows. Similarly, we can obtain the representation of $\cB_T^*$.
For completeness, let us also prove the following result.
**Proposition 2.5. It holds I\_T\^\*=\_[{T}]{},\_T,where $\d_{\{T\}}$ is the Dirac measure at $T$, and \^\*x=x\^T,x\^n.**
*Proof. First of all, since $I_T:\cX[0,T]\to\cX_T$ is bounded, we have $I_T^*:\cX_T^*\equiv\cX_T\to\cX[0,T]^*$. For any $\xi\in\cX_T$, and any $Y(\cd)\in\cX[0,T]$, we have $$\lan
I_T^*\xi,Y(\cd)\ran=\lan\xi,I_TY(\cd)\ran=\dbE\lan\xi,Y(T)\ran=\dbE\int_0^T\lan
Y(s),\xi\ran\d_{T}(ds).$$ Next, since $\dbE:\cX_T\to\dbR$ is bounded, we have $\dbE^*:\dbR\to\cX_T$. For any $\xi\in\cX_T$ and $x\in\dbR$, $$\lan\dbE^*x,\xi\ran=\lan x,\dbE\xi\ran=\dbE\lan x,\xi\ran=x^T\dbE\xi.$$ This completes the proof.*
With operators $\cA$ and $\cB$, we can write the state equation (\[1.1\]) as follows: X=x+X+u.We now have the following result for the well-posedness of the state equaton.
**Proposition 2.6. *Let [(H1)]{} hold. Then for any $(x,u(\cd))\in\dbR^n\times\cU[0,T]$, state equation $(\ref{1.1})$ admits a unique solution $X(\cd)\equiv
X(\cd\,;x,u(\cd))\in\h\cX[0,T]$.***
*Proof. For any $X(\cd)\in\cX[0,T]$ and $u(\cd)\in\cU[0,T]$, by (\[estimate2\]), we have $$\dbE\[\sup_{s\in[0,t]}|(\cA X)(s)|^2\]\le
\a(t)\dbE\[\sup_{s\in[0,t]}|X(s)|^2\],$$ with $\a(t)\in(0,1)$ when $t>0$ is small. Hence, by contraction mapping theorem, we obtain well-posedness of the state equation on $[0,t]$. Then by a usual continuation argument, we obtain the well-posedness of the state equation on $[0,T]$.*
From (\[estimate2\]), we see that if for some $\e>0$, { A(),A()L\^[1+]{}(0,T;\^[nn]{}),A\_1(),A\_1()L\^[2+]{}(0,T;\^[nn]{}),\
B(),B()L\^2(0,T;\^[nm]{}),B\_1(),B\_1()L\^(0,T;\^[nm]{}),.then the result of Proposition 2.6 holds. It is ready to see that (\[2.14\]) is stronger than (H1)$''$ and weaker than (H1)$'$. In what follows, for convenient, we will assume (H1). However, we keep in mind that (H1) can actually be relaxed.
Proposition 2.6 tells us that under, say, (H1), the operator $I-\cA:\h\cX[0,T]\to\h\cX[0,T]$ is invertible and the solution $X$ to the state equation corresponding to $(x,u(\cd))\in\dbR^n\times\cU[0,T]$ is given by $$X=(I-\cA)^{-1}x+(I-\cA)^{-1}\cB u.$$ Note that $$\ba{ll}
\ns\ds I_T\[(I-\cA)^{-1}x+(I-\cA)^{-1}\cB
u\]=I_TX=X(T)=x+\cA_TX+\cB_Tu\\
\ns\ds=\[I+\cA_T(I-\cA)^{-1}\]x+\[\cA_T(I-\cA)^{-1}\cB+\cB_T\]u.\ea$$ Therefore, $$I_T(I-\cA)^{-1}=I+\cA_T(I-\cA)^{-1},\qq
I_T(I-\cA)^{-1}\cB=\cA_T(I-\cA)^{-1}\cB+\cB_T.$$ Now, let $$\left\{\ba{ll}
\ns\ds[\cQ X(\cd)](s)=Q(s)X(s),\q s\in[0,T],\qq\forall X(\cd)\in L^2_\dbF(0,T;\dbR^n),\\
\ns\ds[\h\cQ\f(\cd)](s)=\h Q(s)\f(s),\q
s\in[0,T],\qq\forall\f(\cd)\in L^2(0,T;\dbR^n),\\
\ns\ds[\cR u(\cd)](s)=R(s)u(s),\q s\in[0,T],\qq\forall u(\cd)\in\cU[0,T],\\
\ns\ds[\h\cR\psi(\cd)](s)=\h R(s)\psi(s),\q
s\in[0,T],\qq\forall\psi(\cd)\in L^2(0,T;\dbR^m),\\
\ns\ds\cG\xi=G\xi,\qq\forall\xi\in\cX_T,\qq\h\cG x=\h Gx,\qq\forall
x\in\dbR^n.\ea\right.$$ Then the cost functional can be written as n J(x;u())n=nnX,Xnn+nnX,Xnn+nnu,unn+nnu,unn+nnX(T),X(T)nn+nnX(T),X(T)n\
=,(I-)\^[-1]{}x+(I-)\^[-1]{}u\
+,+u,u+u,u\
+n{\[In+n\_T(In-n)\^[-1]{}\]xn+n\[\_T(In-n)\^[-1]{}n+n\_T\]u}, \[In+n\_T(In-n)\^[-1]{}\]xn+n\[\_T(In-n)\^[-1]{}n+n\_T\]u\
+nn{\[In+n\_T(In-n)\^[-1]{}\]xn+n\[\_T(In-n)\^[-1]{}n+n \_T\]u},{\[In+n\_T(In-n)\^[-1]{}\]xn+n\[\_T(In-n)\^[-1]{}n+n\_T\]u}n\
\_2u,u+2\_1x,u+\_0x,x,where $$\ba{ll}
\ns\ds\Th_2=\cR+\dbE^*\h\cR\dbE+\cB^*(I-\cA^*)^{-1}\cQ(I-\cA)^{-1}\cB+\cB^*(I-\cA^*)^{-1}\dbE^*\h\cQ
\dbE(I-\cA)^{-1}\cB\\
\ns\ds\qq+\1n[\cB^*(I\1n-\1n\cA^*)^{-1}\cA_T^*\1n+\1n\cB_T^*]\cG[\cA_T(I\1n-\1n\cA)^{-1}\cB\1n+\1n\cB_T]
\1n+\1n[\cB^*(I\1n-\1n\cA^*)^{-1}\cA_T^*\1n+\1n\cB_T^*]\dbE^*\h\cG\dbE[\cA_T(I\1n-\1n\cA)^{-1}
\cB\1n+\1n\cB_T]\\
\ns\ds\q=\cR+\dbE^*\h\cR\dbE+\cB^*(I-\cA^*)^{-1}(\cQ+\dbE^*\h\cQ\dbE)(I-\cA)^{-1}\cB\\
\ns\ds\qq\q+[\cB^*(I-\cA^*)^{-1}\cA_T^*+\cB_T^*](\cG+\dbE^*\h\cG\dbE)[\cA_T(I-\cA)^{-1}\cB+\cB_T],\\
\ns\ds\Th_1=\cB^*(I-\cA^*)^{-1}\cQ(I-\cA)^{-1}+\cB^*(I-\cA^*)^{-1}\dbE^*\h\cQ\dbE(I-\cA)^{-1}\\
\ns\ds\qq\q+[\cB^*(I-\cA^*)^{-1}\cA_T^*+\cB_T^*]\cG[I+\cA_T(I-\cA)^{-1}]\\
\ns\ds\qq\q+[\cB^*(I-\cA^*)^{-1}\cA_T^*+\cB_T^*]\dbE^*\h\cG\dbE[I+\cA_T(I-\cA)^{-1}]\\
\ns\ds\q=\cB^*(I-\cA^*)^{-1}(\cQ+\dbE^*\h\cQ\dbE)(I-\cA)^{-1}
+[\cB^*(I-\cA^*)^{-1}\cA_T^*+\cB_T^*](\cG+\dbE^*\h\cG\dbE)[I+\cA_T(I-\cA)^{-1}],\\
\ns\ds\Th_0=(I-\cA^*)^{-1}\cQ(I-\cA)^{-1}+(I-\cA^*)^{-1}\dbE^*\h\cQ
\dbE(I-\cA)^{-1}\\
\ns\ds\qq\q+[I+(I-\cA^*)^{-1}\cA_T^*]\cG[I+\cA_T(I-\cA)^{-1}]
+[I+(I-\cA^*)^{-1}\cA_T^*]\dbE^*\h\cG\dbE[I+\cA_T(I-\cA)^{-1}]\\
\ns\ds\q=(I-\cA^*)^{-1}(\cQ+\dbE^*\h\cQ\dbE)(I-\cA)^{-1}+[I+(I-\cA^*)^{-1}\cA_T^*]
(\cG+\dbE^*\h\cG\dbE)[I+\cA_T(I-\cA)^{-1}].\ea$$ Consequently, for any $u(\cd),v(\cd)\in\cU[0,T]$, and $x\in\dbR^n$, J(x;v())=J(x;u()+\[v()-u()\])\
=\_2\[u+(v-u)\],u+(v-u)+2\_1x,u+(v-u)+\_0x,x\
=\_2u,u+2\_1x,u+\_0x,x+2\_2u+\_1x,v-u+\_2(v-u),v-u\
=J(x;u())+2\_2u+\_1x,v-u+\_2(v-u),v-u.We now present the following result whose proof is standard, making use of the above (see [@Mou-Yong; @2006] for details).
**Proposition 2.7. *If $u(\cd)\mapsto J(x;u(\cd))$ admits a minimum, then \_20.Conversely, if in addition to $(\ref{Th2ge0})$, one has \_1x\_2$\cU[0,T]$,then $u(\cd)\mapsto J(x;u(\cd))$ admits a minimum $\bar
u(\cd)\in\cU[0,T]$. Further, if \_2I,for some $\d>0$, then for any given $x\in\dbR^n$, $u(\cd)\mapsto
J(x;u(\cd))$ admits a unique minimum.***
By the definition of $\Th_2$, we see that (\[Th2ge0\]) is implied by the following: +\^\*0,+\^\*0,+\^\*0,and (\[Th2>0\]) is implied by +\^\*I,+\^\*0,+\^\*0,for some $\d>0$. Now, we would like to present more direct conditions under which (\[Th2ge0\]) and (\[Th2>0\]) hold, respectively.
**Proposition 2.8. *Let [(H1)]{} and [(H2)$'$]{} hold. Then $(\ref{Th2ge0})$ holds. Further, if [(H2)$''$]{} holds for some $\d>0$, then $(\ref{Th2>0})$ holds and Problem [(MF)]{} admits a unique solution.***
*Proof. For any $\xi\in\cX_T$, $$\ba{ll}
\ns\ds\dbE\[\lan G\xi,\xi\ran+\lan\h
G\dbE[\xi],\dbE[\xi]\ran\]=\dbE\[\lan
G(\xi-\dbE[\xi]),\xi-\dbE[\xi]\ran +\lan(G+\h
G)\dbE[\xi],\dbE[\xi]\ran\]\ge0,\\
\ns\ds\dbE\[\lan Q(s)\xi,\xi\ran+\lan\h
Q(s)\dbE[\xi],\dbE[\xi]\ran\]=\dbE\[\lan
Q(s)(\xi-\dbE[\xi]),\xi-\dbE[\xi]\ran+\lan[Q(s)+\h
Q(s)]\dbE[\xi],\dbE[\xi]\ran\]\ge0,\ea$$ and for any $\eta\in\cU_T$, $$\dbE\[\lan R(s)\eta,\eta\ran+\lan\h
R(s)\dbE[\eta],\dbE[\eta]\ran\]=\dbE\[\lan
R(s)(\eta-\dbE[\eta]),\eta-\dbE[\eta]\ran+\lan[R(s)+\h
R(s)]\dbE[\eta],\dbE[\eta]\ran\]\ge0.$$ Thus, (\[Th2ge0\]) holds. Next, if (H2)$''$ holds, then $$\ba{ll}
\ns\ds\dbE\[\lan R(s)\eta,\eta\ran+\lan\h
R(s)\dbE[\eta],\dbE[\eta]\ran\]=\dbE\[\lan
R(s)(\eta-\dbE[\eta]),\eta-\dbE[\eta]\ran+\lan[R(s)+\h
R(s)]\dbE[\eta],\dbE[\eta]\ran\]\\
\ns\ds\qq\ge\d\dbE\[|\eta-\dbE[\eta]|^2+|\dbE[\eta]|^2\]=\d\dbE|\eta|^2.\ea$$ Hence, (\[Th2>0\]) holds.*
Optimality Conditions.
======================
In this section, we first derive a necessary condition for optimal pair of Problem (MF).
**Theorem 3.1. *Let [(H1)]{} and [(H2)]{} hold. Let $(\bar
X(\cd),\bar u(\cd))$ be an optimal pair of Problem [(MF)]{}. Then the following mean-field backward stochastic differential equation (MF-BSDE, for short) admits a unique adapted solution $(Y(\cd),Z(\cd))$: { dY(s)=-$A(s)^TY(s)+A_1(s)^TZ(s)+\h A(s)^T\dbE[Y(s)]+\h
A_1(s)^T\dbE[Z(s)]\\
\ns\ds\qq\qq\qq+Q(s)\bar X(s)+\h Q(s)\dbE[\bar X(s)]$ds+Z(s)dW(s),s,\
Y(T)=G|X(T)+G,.such that R(s)|u(s)+B(s)\^TY(s)+B\_1(s)\^TZ(s)+R(s)+B(s)\^T+B\_1(s)\^T=0,\
s,***
*Proof. Let $(\bar X(\cd),\bar u(\cd))$ be an optimal pair of Problem (MF). For any $u(\cd)\in\cU[0,T]$, let $X(\cd)$ be the state process corresponding to the zero initial condition and the control $u(\cd)$. Then we have $$\ba{ll}
\ns\ds0=\dbE\Big\{\int_0^T\[\lan Q(s)\bar X(s),X(s)\ran+\lan\h
Q(s)\dbE[\bar X(s)],\dbE[X(s)]\ran+\lan R(s)\bar
u(s),u(s)\ran\\
\ns\ds\qq\qq\qq+\lan\h R(s)\dbE[\bar u(s)],\dbE[u(s)]\ran\]ds+\lan
G\bar X(T),X(T)\ran+\lan\h G\dbE[\bar
X(T)],\dbE[X(T)]\ran\Big\}\\
\ns\ds=\dbE\Big\{\int_0^T\[\lan Q(s)\bar X(s)+\h Q(s)\dbE[\bar
X(s)],X(s)\ran+\lan R(s)\bar u(s)+\h R(s)\dbE[\bar
u(s)],u(s)\ran\]ds\\
\ns\ds\qq\qq\qq+\lan G\bar X(T)+\h G\dbE[\bar
X(T)],X(T)\ran\Big\},\ea$$ On the other hand, by [@Buckdahn-Li-Peng; @2009], we know that (\[MF-BSDE1\]) admits a unique adapted solution $(Y(\cd),Z(\cd))$. Then $$\ba{ll}
\ns\ds\dbE\lan X(T),G\bar X(T)+\h G\dbE[\bar X(T)]\ran=\dbE\lan
X(T),Y(T)\ran\\
\ns\ds=\dbE\Big\{\int_0^T\(\lan A(s)X(s)+B(s)u(s)+\h
A(s)\dbE[X(s)]+\h B(s)\dbE[u(s)],Y(s)\ran\\
\ns\ds\qq-\lan X(s),A(s)^TY(s)+A_1(s)^TZ(s)+\h A(s)^T\dbE[Y(s)]+\h
A_1(s)^T\dbE[Z(s)]+Q(s)\bar X(s)+\h Q(s)\dbE[\bar X(s)]\ran\\
\ns\ds\qq+\lan A_1(s)X(s)+B_1(s)u(s)+\h
A_1(s)\dbE[X(s)]+\h B_1(s)\dbE[u(s)],Z(s)\ran\)ds\Big\}\\
\ns\ds=\dbE\Big\{\int_0^T\(-\lan X(s),Q(s)\bar X(s)+\h Q(s)\dbE[\bar
X(s)]\ran\\
\ns\ds\qq\qq+\lan u(s),B(s)^TY(s)+B_1(s)^TZ(s)+\h
B(s)^T\dbE[Y(s)]+\h B_1(s)^T\dbE[Z(s)]\ran\)ds\Big\}.\ea$$ Hence, $$\ba{ll}
\ns\ds\dbE\2n\int_0^T\2n\3n\(\2n\lan u(s),B(s)^TY(s)\1n+\1n
B_1(s)^TZ(s)\1n+\1n\h B(s)^T\dbE[Y(s)]\1n+\1n\h
B_1(s)^T\dbE[Z(s)]\1n+\1n R(s)\bar u(s)\1n+\1n\h R(s)\dbE[\bar
u(s)]\ran\2n\)ds=0,\ea$$ which leads to $$\ba{ll}
\ns\ds R(s)\bar u(s)+B(s)^TY(s)+B_1(s)^TZ(s)+\h R(s)\dbE[\bar
u(s)]+\h B(s)^T\dbE[Y(s)]+\h B_1(s)^T\dbE[Z(s)]=0.\ea$$ This completes the proof.*
From the above, we end up with the following optimality system: (with $s$ suppressed) { d|X=$A\bar X+B\bar u+\h A\dbE[\bar X]+\h B\dbE[\bar
u]$ds+$A_1\bar X+B_1\bar u+\h A_1\dbE[\bar X]+\h
B_1\dbE[\bar u]$dW(s),\
dY=-$A^TY+A_1^TZ+Q\bar X+\h
A^T\dbE[Y]+\h A_1^T\dbE[Z]+\h Q\dbE[\bar X]$ds+ZdW(s),\
X(0)=x,Y(T)=G|X(T)+G,\
R|u+R+B\^TY+B\_1\^TZ+B\^T+B\_1\^T=0..This is called a (coupled) forward-backward stochastic differential equations of mean-field type (MF-FBSDE, for short). Note that the coupling comes from the last relation (which is essentially the maximum condition in the usual Pontryagin type maximum principle). The 4-tuple $(\bar X(\cd),\bar u(\cd),Y(\cd),Z(\cd))$ of $\dbF$-adapted processes satisfying the above is called an [*adapted solution*]{} of (\[MF-FBSDE1\]). We now look at the sufficiency of the above result.
**Theorem 3.2. *Let [(H1)]{}, [(H2)]{}, and $(\ref{Th2ge0})$ hold. Suppose $(\bar X(\cd),\bar u(\cd),Y(\cd),Z(\cd))$ is an adapted solution to the MF-FBSDE $(\ref{MF-FBSDE1})$. Then $(\bar
X(\cd),\bar u(\cd))$ is an optimal pair.***
*Proof. Let $(\bar X(\cd),\bar u(\cd),Y(\cd),Z(\cd))$ be an adapted solution to the MF-FBSDE. For any $u(\cd)\in\cU[0,T]$, let $$X_1(\cd)\equiv X(\cd\,;0,u(\cd)-\bar u(\cd)).$$ Then $$X(s;x,u(\cd))=\bar X(s)+X_1(s),\qq s\in[0,T].$$ Hence, (suppressing $s$) $$\ba{ll}
\ns\ds J(x;u(\cd))-J(x;\bar u(\cd))\\
\ns\ds=2\dbE\Big\{\int_0^T\[\lan Q\bar X,X_1\ran+\lan\h
Q\dbE[\bar X],\dbE[X_1]\ran+\lan R\bar u,u-\bar u\ran+\lan\h R\dbE[\bar u],\dbE[u-\bar u]\ran\]ds\\
\ns\ds\qq+\lan G\bar X(T),X_1(T)\ran+\lan\h G\dbE[\bar X(T)],\dbE[X_1(T)]\ran\Big\}\\
\ns\ds\qq+\dbE\Big\{\int_0^T\[\lan QX_1,X_1\ran+\lan\h
Q\dbE[X_1],\dbE[X_1]\ran+\lan R(u-\bar u),u-\bar u\ran+\lan\h R\dbE[u-\bar u],\dbE[u-\bar u]\ran\]ds\\
\ns\ds\qq+\lan GX_1(T),X_1(T)\ran+\lan\h G\dbE[X_1(T)],\dbE[X_1(T)]\ran\Big\}\\
\ns\ds=2\dbE\Big\{\int_0^T\[\lan X_1,Q\bar X+\h Q\dbE[\bar
X]\ran+\lan u-\bar u,R\bar u+\h R\dbE[\bar u]\ran\]ds+\lan
X_1(T),G\bar X(T)+\h
G\dbE[\bar X(T)]\ran\Big\}\\
\ns\ds\qq+J(0;u(\cd)-\bar u(\cd)).\ea$$ Note that $$\ba{ll}
\ns\ds\dbE\lan X_1(T),G\bar X(T)+\h G\dbE[\bar X(T)]\ran=\dbE\lan
X_1(T),Y(T)\ran\\
\ns\ds=\dbE\Big\{\int_0^T\(\lan AX_1+B(u-\bar u)+\h
A\dbE[X_1]+\h B\dbE[u-\bar u],Y\ran\\
\ns\ds\qq-\lan X_1,A^TY+A_1^TZ+\h A^T\dbE[Y]+\h
A_1^T\dbE[Z]+Q\bar X+\h Q\dbE[\bar X]\ran\\
\ns\ds\qq+\lan A_1X_1+B_1(u-\bar u)+\h
A_1\dbE[X_1]+\h B_1\dbE[u-\bar u],Z\ran\)ds\Big\}\\
\ns\ds=\dbE\Big\{\int_0^T\(-\lan X_1,Q\bar X+\h Q\dbE[\bar
X]\ran+\lan u-\bar u,B^TY+B_1^TZ+\h B^T\dbE[Y]+\h
B_1^T\dbE[Z]\ran\)ds\Big\}.\ea$$ Thus, $$\ba{ll}
\ns\ds J(x;u(\cd))-J(x;\bar u(\cd))\\
\ns\ds=2\dbE\int_0^T\lan u-\bar u,R\bar u+\h R\dbE[\bar
u]+B^TY+B_1^TZ+\h B^T\dbE[Y]+\h B_1^T\dbE[Z]\ran
ds+J(0;u(\cd)-\bar u(\cd))\\
\ns\ds=J(0;u(\cd)-\bar u(\cd))=\lan\Th_2(u-\bar u),u-\bar
u\ran\ge0.\ea$$ Hence, $(\bar X(\cd),\bar u(\cd))$ is optimal.*
We have the following corollary.
**Corollary 3.3. *Let [(H1)]{} and [(H2)$''$]{} hold. Then MF-FBSDE $(\ref{MF-FBSDE1})$ admits a unique adapted solution $(\bar
X(\cd),\bar u(\cd),Y(\cd),Z(\cd))$ of which $(\bar X(\cd),\bar
u(\cd))$ is the unique optimal pair of Problem [(MF)]{}.***
*Proof. We know from Proposition 2.8 that under (H1) and (H2)$''$, Problem (MF) admits a unique optimal pair $(\bar
X(\cd),\bar u(\cd))$. Then by Theorem 3.1, for some $(Y(\cd),Z(\cd))$, the 4-tuple $(\bar X(\cd),\bar
u(\cd),Y(\cd),Z(\cd))$ is an adapted solution to MF-FBSDE (\[MF-FBSDE1\]). Next, if (\[MF-FBSDE1\]) has another adapted solution $(\wt X(\cd),\wt u(\cd),\wt Y(\cd),\wt Z(\cd))$. Then by Theorem 3.2, $(\wt X(\cd),\wt u(\cd))$ must be an optimal pair of Problem (MF). Hence, by the uniqueness of the optimal pair of Problem (MF), we must have $$\wt X(\cd)=\bar X(\cd),\qq\wt u(\cd)=\bar u(\cd).$$ Then by the uniqueness of MF-BSDE (\[MF-BSDE1\]), one must have $$\wt Y(\cd)=Y(\cd),\qq\wt Z(\cd)=Z(\cd),$$ proving the corollary.*
Decoupling the MF-FBSDE and Riccati Equations
=============================================
From Corollary 3.3, we see that under (H1) and (H2)$''$, to solve Problem (MF), we need only to solve MF-FBSDE (\[MF-FBSDE1\]). To solve (\[MF-FBSDE1\]), we use the idea of decoupling inspired by the Four-Step Scheme ([@MPY; @1994; @MY; @1999]). More precisely, we have the following result. For simplicity, we have suppressed the time variable $s$ below.
**Theorem 4.1. *Let [(H1)]{} and [(H2)$''$]{} hold. Then the unique adapted solution $(\bar X(\cd),\bar u(\cd),Y(\cd),Z(\cd))$ of MF-FBSDE $(\ref{MF-FBSDE1})$ admits the following representation: { |u=-K\_0\^[-1]{}(B\^TP+B\_1\^TPA\_1)$\bar X-\dbE[\bar
X]$-K\_1\^[-1]{}$$(B+\h B)^T\Pi+(B_1+\h B_1)^TP(A_1+\h A_1)$$,\
Y=P$\bar X-\dbE[\bar X]$+,\
Z=$$PA_1-PB_1K_0^{-1}(B^TP+B_1^TPA_1)$$$\bar X-\dbE[\bar
X]$\
+$$P(A_1+\h A_1)-P(B_1+\h B_1)K_1^{-1}\((B+\h
B)^T\Pi+(B_1+\h B_1)^TP(A_1+\h A_1)\)$$..where K\_0=R+B\_1\^TPB\_1,K\_1=R+R+(B\_1+B\_1)\^TP(B\_1+B\_1).and $P(\cd)$ and $\Pi(\cd)$ are solutions to the following Riccati equations, respectively: {n P’+PA+A\^TP+A\_1\^TPA\_1+Q-(PBn+n A\_1\^TPB\_1)K\_0\^[-1]{} (B\^TP+B\_1\^TPA\_1)=0,s,\
P(T)=G,.and {n ’+$$(A+\h A)-(B+\h B)K_1^{-1}(B_1+\h B_1)^TP(A_1+\h A_1)$$\
+$$(A+\h A)^T-(A_1+\h A_1)^TP(B_1+\h B_1)K_1^{-1}(B+\h
B)^T$$-(B+B)K\_1\^[-1]{} (B+B)\^T\
+(A\_1+A\_1)\^T$$P-P(B_1+\h B_1)K_1^{-1}(B_1+\h
B_1)^TP$$(A\_1+A\_1)+Q+Q=0,s,\
(T)=G+G..Finally, $\bar X(\cd)$ solves the following closed-loop system { d|X={$$A-BK_0^{-1}(B^TP+B_1^TPA_1)$$$\bar X-\dbE[\bar X]$\
+$$(A+\h A)-(B+\h B)K_1^{-1}\((B+\h B)^T\Pi+(B_1+\h
B_1)^TP(A_1+\h A_1)\)$$}dt\
+{$$A_1-B_1K_0^{-1}(B^TP+B_1^TPA_1)$$$\bar X-\dbE[\bar X]$\
+$$(A_1+\h A_1)-(B_1+\h B_1)K_1^{-1}\((B+\h B)^T\Pi+(B_1+\h
B_1)^TP(A_1+\h A_1)\)$$}dW(t),\
|X(0)=x..***
*Proof. First of all, under (H1) and (H2)$''$, Riccati equation (\[Riccati P\]) admits a unique solution $P(\cd)$ which is positive definite matrix valued ([@Yong-Zhou; @1999]). With such a function $P(\cd)$, $K_1$ defined in (\[K\]) is positive definite. Next, we note that $$\ba{ll}
\ns\ds P-P(B_1+\h B_1)K_1^{-1}(B_1+\h B_1)^TP\\
\ns\ds\q=P\1n-\1n P(B_1\1n+\1n\h B_1)\[R\1n+\1n\h
R\1n+\1n(B_1\1n+\1n\h B_1)^TP(B_1\1n+\1n\h
B_1)\]^{-1}(B_1\1n+\1n\h B_1)^TP\equiv P\1n-\1n P\wt B(\wt R\1n+\1n\wt B^TP\wt B)^{-1}\wt B^TP\\
\ns\ds\q=P^{1\over2}\[I\1n-\1n P^{1\over2}\wt B\wt
R^{-{1\over2}}\(I\1n+\1n\wt R^{-{1\over2}}\wt
B^TP^{1\over2}P^{1\over2}\wt B\wt R^{-{1\over2}}\)^{-1}\wt
R^{-{1\over2}}\wt B^TP^{1\over2}\]P^{1\over2}\equiv
P^{1\over2}\[I\1n-\1n\G(I\1n+\1n\G^T\G)^{-1}\G^T\]P^{1\over2}\\
\ns\ds=\1n P^{1\over2}\1n(I\1n+\1n\G\G^T\1n)^{-1}\1n
P^{1\over2}\1n\equiv\1n P^{1\over2}(I\1n+\1n P^{1\over2}\wt B\wt
R^{-1}\wt B^T\1n P^{1\over2})^{-1}\1n P^{1\over2}\1n\equiv\1n
P^{1\over2}\1n[I\2n+\2n P^{1\over2}\1n(B_1\2n+\1n\h B_1)(R\1n+\1n\h
R)^{-1}\1n(B_1\2n+\1n\h B_1)^T\1n P^{1\over2}]^{-1}\1n
P^{1\over2}\1n\ge\1n0.\ea$$ In the above, we have denoted $$\wt B=B_1+\h B_1,\q\wt R=R+\h R,\q\G=P^{1\over2}\wt B\wt R^{1\over2},$$ and used the fact $$I-\G(I+\G^T\G)^{-1}\G^T=(I+\G^T\G)^{-1}.$$ Hence, Riccati equation (\[Riccati Pi\]) is equivalent to the following: { ’+$$(A+\h A)-(B+\h B)K_1^{-1}(B_1+\h B_1)^TP(A_1+\h A_1)$$\
+$$(A\1n+\1n\h A)^T\1n-\1n(A_1\1n+\1n\h A_1)^TP(B_1\1n+\1n\h
B_1)K_1^{-1}(B\1n+\1n\h B)^T$$n-n(Bn+nB)K\_1\^[-1]{} (Bn+nB)\^T\
+n(A\_1n+nA\_1)\^Tn P\^[12]{}$$I\1n+\1n
P^{1\over2}(B_1\2n+\1n\h B_1)(R\1n+\1n\h R)^{-1}(B_1\2n+\1n\h
B_1)^T\1n P^{1\over2}$$\^[-1]{}n P\^[12]{}(A\_1n+nA\_1)n+n Q+nQ=0,\
(T)=G+G..Since $$\ba{ll}
\ns\ds G+\h G\ge0,\\
\ns\ds(A_1\1n+\1n\h A_1)^TP^{1\over2}\[I\1n+\1n
P^{1\over2}(B_1\1n+\1n\h B_1)(R\1n+\1n\h R)^{-1}(B_1\1n+\1n\h
B_1)^TP^{1\over2}\]^{-1}P^{1\over2}(A_1\1n+\1n\h A_1)\1n+\1n Q+\1n\h
Q\ge0,\ea$$ according to [@Yong-Zhou; @1999], Riccati equation (\[Riccati Pi\]) admits a unique solution $\Pi(\cd)$ which is also positive definite matrix valued.*
Now, suppose $(\bar X(\cd),\bar u(\cd),Y(\cd),Z(\cd))$ is the adapted solution to (\[MF-FBSDE1\]). Assume that Y(s)=P(s)X(s)+P(s),s,for some deterministic and differentiable functions $P(\cd)$ and $\h
P(\cd)$ such that P(T)=G,P(T)=G.For the time being, we do not assume that $P(\cd)$ is the solution to (\[Riccati P\]). Then (suppressing $s$) $-A^TY-A_1^TZ-Q\bar
X-\h A^T\dbE[Y]-\h A_1^T\dbE[Z]-\h Q\dbE[\bar X]$ds+ZdW(s)\
=dY=d$P\bar X+\h P\dbE[\bar X]$\
=$$P'\bar X+P\(A\bar X+B\bar u+\h A\dbE[\bar X]+\h B\dbE[\bar
u]\)+\h P'\dbE[\bar X]+\h P\((A+\h A)\dbE[\bar
X]+(B+\h B)\dbE[\bar u]\)$$ds\
+P$A_1\bar X+B_1\bar u+\h A_1\dbE[\bar X]+\h B_1\dbE[\bar u]$dW(s)\
=$$(P'+PA)\bar X+PB\bar u+\(\h P'+\h P(A+\h A)+P\h
A\)\dbE[\bar X]+\(\h P(B+\h B)+P\h B\)\dbE[\bar u]$$ds\
+P$A_1\bar X+B_1\bar u+\h A_1\dbE[\bar X]+\h
B_1\dbE[\bar u]$dW(s)Comparing the diffusion terms, we should have Z=P$A_1\bar X+B_1\bar u+\h A_1\dbE[\bar X]+\h B_1\dbE[\bar
u]$.This yields from (\[bar u\]) that $$\ba{ll}
\ns\ds0=R\bar u+\h R\dbE[\bar u]+B^TY+\h B^T\dbE[Y]+B_1^TZ+\h B_1^T\dbE[Z]\\
\ns\ds\q=R\bar u+\h R\dbE[\bar u]+B^T(P\bar X+\h P\dbE[\bar X])+\h
B^T(P+\h P)\dbE[\bar X]\\
\ns\ds\qq+B_1^TP\(A_1\bar X+B_1\bar u+\h A_1\dbE[\bar X]+\h
B_1\dbE[\bar u]\)+\h B_1^TP\((A_1+\h A_1)\dbE[\bar X]+(B_1
+\h B_1)\dbE[\bar u]\)\\
\ns\ds\q=(R+B_1^TPB_1)\bar u+(\h R+B_1^TP\h B_1+\h B_1^TPB_1+\h
B_1^TP\h B_1)\dbE[\bar u]+(B^TP+B_1^TPA_1)\bar X\\
\ns\ds\qq+\(B^T\h P+\h B^T(P+\h P)+B_1^TP\h A_1+\h B_1^TP(A_1+\h
A_1)\)\dbE[\bar X].\ea$$ Taking $\dbE$, we obtain $$\ba{ll}
\ns\ds0=\[R\1n+\1n\h R\1n+\1n(B_1\1n+\1n\h B_1)^TP(B_1\1n+\1n\h
B_1)\]\dbE[\bar u]\1n+\1n\[(B\1n+\1n\h B)^T(P\1n+\1n\h
P)\1n+\1n(B_1\1n+\1n\h B_1)^TP(A_1\1n+\1n\h A_1)\]\dbE[\bar X]\\
\ns\ds\q\equiv K_1\dbE[\bar u]\1n+\1n\[(B\1n+\1n\h B)^T(P\1n+\1n\h
P)\1n+\1n(B_1\1n+\1n\h B_1)^TP(A_1\1n+\1n\h A_1)\]\dbE[\bar X].\ea$$ Assuming $$K_1\equiv K_1(P)\deq R+\h R+(B_1+\h B_1)^TP(B_1+\h B_1)$$ to be invertible, one gets $$\dbE[\bar u]=-K_1^{-1}\[(B+\h B)^T(P+\h P)+(B_1+\h B_1)^TP(A_1+\h
A_1)\]\dbE[\bar X].$$ Then we have $$\ba{ll}
\ns\ds0=(R+B_1^TPB_1)\bar u+\(\h R+B_1^TP\h B_1+\h B_1^TPB_1+\h
B_1^TP\h B_1\)\dbE[\bar u]+(B^TP+B_1^TPA_1)\bar X\\
\ns\ds\qq+\(B^T\h P+\h B^T(P+\h P)+B_1^TP\h A_1+\h B_1^TP(A_1+\h
A_1)\)\dbE[\bar X]\\
\ns\ds\q\equiv K_0\bar u+(B^TP+B_1^TPA_1)\bar X++\(B^T\h P+\h
B^T(P+\h P)+B_1^TP\h A_1+\h B_1^TP(A_1+\h
A_1)\)\dbE[\bar X]\\
\ns\ds\qq-\(\h R\1n+\1n B_1^TP\h B_1\1n+\1n\h B_1^TPB_1\1n+\1n\h
B_1^TP\h B_1\)K_1^{-1}\[(B\1n+\1n\h B)^T(P\1n+\1n\h
P)\1n+\1n(B_1\1n+\1n\h B_1)^TP(A_1\1n+\1n\h A_1)\]\dbE[\bar X].\ea$$ Consequently, by assuming $$K_0\equiv K_0(P)\deq R+B_1^TPB_1$$ to be invertible, we obtain |u=-K\_0\^[-1]{}(B\^TP+B\_1\^TPA\_1)|X-K\_0\^[-1]{}$B^T\h P+\h
B^T(P+\h P)+B_1^TP\h A_1+\h
B_1^TP(A_1+\h A_1)$\
+K\_0\^[-1]{}$\1n\h R\1n+\1n B_1^T\1n P\h B_1\1n+\1n\h
B_1^TPB_1\1n+\1n\h B_1^T\1n P\h B_1\1n$K\_1\^[-1]{}$$(B\1n+\1n\h
B)^T\1n(P\1n+\1n\h P)\1n+\1n(B_1\1n+\1n\h B_1)^T\1n P(A_1\1n+\1n\h
A_1)$$\
=-K\_0\^[-1]{}(B\^TP+B\_1\^TPA\_1)|X-K\_0\^[-1]{}$B^T\h P+\h B^T(P+\h
P)+B_1^TP\h A_1+\h B_1^TP(A_1+\h A_1)$\
+K\_0\^[-1]{}$K_1-K_0$K\_1\^[-1]{}$$(B+\h B)^T(P+\h P)+(B_1+\h
B_1)^TP(A_1+\h A_1)$$\
=-K\_0\^[-1]{}(B\^TP+B\_1\^TPA\_1)|X-K\_0\^[-1]{}$B^T\h P+\h B^T(P+\h
P)+B_1^TP\h A_1+\h B_1^TP(A_1+\h A_1)$\
+K\_0\^[-1]{}$$(B+\h B)^T(P+\h P)+(B_1+\h B_1)^TP(A_1+\h
A_1)$$\
-K\_1\^[-1]{}$$(B+\h B)^T(P+\h P)+(B_1+\h B_1)^TP(A_1+\h
A_1)$$\
=-K\_0\^[-1]{}(B\^TP+B\_1\^TPA\_1)|X+K\_0\^[-1]{}$$B^TP+B_1^TPA_1$$\
-K\_1\^[-1]{}$$(B+\h B)^T(P+\h P)+(B_1+\h B_1)^TP(A_1+\h
A_1)$$.Here, we note that $$\ba{ll}
\ns\ds K_1-K_0=R+\h R+(B_1+\h B_1)^TP(B_1+\h B_1)-R-B_1^TPB_1\\
\ns\ds\qq\qq=\h R+B_1^TP\h B_1+\h B_1^TPB_1+\h B_1^TP\h B_1.\ea$$ Hence, comparing the drift terms in (\[4.9\]), we have $$\ba{ll}
\ns\ds0=(P'+PA)\bar X+PB\bar u+\(\h P'+\h P(A+\h A)+P\h
A\)\dbE[\bar X]+\(\h P(B+\h B)+P\h B\)\dbE[\bar u]\\
\ns\ds\qq+A^T\(P\bar X+\h P\dbE[\bar X]\)+A_1^TP\(A_1\bar X+B_1\bar
u+\h A_1\dbE[\bar X]+\h B_1\dbE[\bar u]\)\\
\ns\ds\qq+\h A^T(P+\h P)\dbE[\bar X]+\h A_1^TP\[(A_1+\h
A_1)\dbE[\bar X]+(B_1+\h B_1)\dbE[\bar u]\] +Q\bar X+\h Q\dbE[\bar X]\\
\ns\ds=\[P'+PA+A^TP+A_1^TPA_1+Q\]\bar X+(PB+A_1^TPB_1)\bar u\\
\ns\ds\qq+\[\h P'\1n+\1n\h P(A\1n+\1n\h A)\1n+\1n(A\1n+\1n\h A)^T\h
P\1n+\1n P\h A\1n+\1n\h
A^TP\1n+\1n(A_1\1n+\1n\h A_1)^TP(A_1\1n+\1n\h A_1)\1n-\1n A_1^TPA_1\1n+\1n\h Q\]\dbE[\bar X]\\
\ns\ds\qq+\[\h P(B+\h B)+P\h B+A_1^TP\h B_1+\h A_1^TP(B_1+\h
B_1)\]\dbE[\bar u]\\
\ns\ds=\[P'+PA+A^TP+A_1^TPA_1+Q\]\bar X\\
\ns\ds\qq+(PB+A_1^TPB_1)\Big\{-K_0^{-1}(B^TP+B_1^TPA_1)\bar X+K_0^{-1}\[B^TP+B_1^TPA_1\]\dbE[\bar X]\\
\ns\ds\qq-K_1^{-1}\[(B+\h B)^T(P+\h P)+(B_1+\h B_1)^TP(A_1+\h
A_1)\]\dbE[\bar X]\Big\}\\
\ns\ds\qq+\[\h P'\1n+\1n\h P(A\1n+\1n\h A)\1n+\1n(A\1n+\1n\h A)^T\h
P\1n+\1n P\h A\1n+\1n\h
A^TP\1n+\1n(A_1\1n+\1n\h A_1)^TP(A_1\1n+\1n\h A_1)\1n-\1n A_1^TPA_1\1n+\1n\h Q\]\dbE[\bar X]\\
\ns\ds\qq+\[\h P(B+\h B)+P\h B+A_1^TP\h B_1+\h A_1^TP(B_1+\h
B_1)\]\\
\ns\ds\qq\qq\cd\Big\{-K_1^{-1}\[(B+\h B)^T(P+\h P)+(B_1+\h
B_1)^TP(A_1+\h
A_1)\]\dbE[\bar X]\Big\}\\
\ns\ds=\[P'+PA+A^TP+A_1^TPA_1+Q-(PB+A_1^TPB_1)K_0^{-1}(B^TP+B_1^TPA_1)\]\bar X\\
\ns\ds\qq+\Big\{\h P'+\h P(A+\h A)+(A+\h A)^T\h P+P\h A+\h
A^TP+(A_1+\h A_1)^TP(A_1+\h A_1)-A_1^TPA_1+\h Q\\
\ns\ds\qq+(PB+A_1^TPB_1)K_0^{-1}(B^TP+B_1^TPA_1)\\
\ns\ds\qq-\[(P\1n+\1n\h P)(B\1n+\1n\h B)\1n+\1n(A_1\1n+\1n\h
A_1)^TP(B_1\1n+\1n\h B_1)\]K_1^{-1}\[(B\1n+\1n\h B)^T(P\1n+\1n\h
P)\1n+\1n(B_1\1n+\1n\h B_1)^TP(A_1\1n+\1n\h A_1)\]\1n\Big\}\dbE[\bar
X].\ea$$ Therefore, by choosing $P(\cd)$ to be the solution to Riccati equation (\[Riccati P\]), we have that $K_0$ and $K_1$ are positive definite, and the above leads to $$\ba{ll}
\ns\ds\Big\{\h P'+\h P(A+\h A)+(A+\h A)^T\h P+P\h A+\h
A^TP+(A_1+\h A_1)^TP(A_1+\h A_1)-A_1^TPA_1+\h Q\\
\ns\ds+(PB+A_1^TPB_1)K_0^{-1}(B^TP+B_1^TPA_1)\\
\ns\ds-\[(P\1n+\1n\h P)(B\1n+\1n\h B)\1n+\1n(A_1\1n+\1n\h
A_1)^TP(B_1\1n+\1n\h B_1)\]K_1^{-1}\[(B\1n+\1n\h B)^T(P\1n+\1n\h
P)\1n+\1n(B_1\1n+\1n\h B_1)^TP(A_1\1n+\1n\h A_1)\]\1n\Big\}\dbE[\bar
X]=0.\ea$$ Now, if $\h P(\cd)$ satisfies the following: $$\ba{ll}
\ns\ds0=\h P'+\h P(A+\h A)+(A+\h A)^T\h P+P\h A+\h
A^TP+(A_1+\h A_1)^TP(A_1+\h A_1)-A_1^TPA_1+\h Q\\
\ns\ds\qq+(PB+A_1^TPB_1)K_0^{-1}(B^TP+B_1^TPA_1)\\
\ns\ds\qq-\[(P+\h P)(B+\h B)+(A_1+\h A_1)^TP(B_1+\h
B_1)\]K_1^{-1}\[(B+\h B)^T(P+\h P)+(B_1+\h B_1)^TP(A_1+\h
A_1)\]\\
\ns\ds\q=\h P'+\h P\[A+\h A-(B+\h B)K_1^{-1}\((B+\h B)^TP+(B_1+\h B_1)^TP(A_1+\h A_1)\)\]\\
\ns\ds\qq+\[(A+\h A)^T-\(P(B+\h B)+(A_1+\h A_1)^TP(B_1+\h B_1)\)K_1^{-1}(B+\h B)^T\]\h P
-\h P(B+\h B)K_1^{-1}(B+\h B)^T\h P\\
\ns\ds\qq+P\h A+\h
A^TP+(A_1+\h A_1)^TP(A_1+\h A_1)-A_1^TPA_1+\h Q+(PB+A_1^TPB_1)K_0^{-1}(B^TP+B_1^TPA_1)\\
\ns\ds\qq-\[P(B+\h B)+(A_1+\h A_1)^TP(B_1+\h B_1)\]K_1^{-1}\[(B+\h
B)^TP+(B_1+\h B_1)^TP(A_1+\h A_1)\],\ea$$ then $(Y(\cd),Z(\cd))$ defined by (\[Y\]) and (\[Z\]) with $\bar
u(\cd)$ given by (\[u\]) satisfies the MF-BSDE in (\[MF-FBSDE1\]). Hence, we introduce the following Riccati equation for $\h P(\cd)$: { P’+P$$A+\h A-(B+\h B)K_1^{-1}\((B+\h B)^TP+(B_1+\h B_1)^TP(A_1+\h A_1)\)$$\
+$$(A+\h A)^T-\(P(B+\h B)+(A_1+\h A_1)^TP(B_1+\h
B_1)\)K_1^{-1}(B+\h B)^T$$P\
-P(B+B)K\_1\^[-1]{}(B+B)\^TP+PA+A\^TP+(A\_1+A\_1)\^TP(A\_1+A\_1)-A\_1\^TPA\_1+Q\
+(PB+A\_1\^TPB\_1)K\_0\^[-1]{}(B\^TP+B\_1\^TPA\_1)\
-$$P(B+\h B)+(A_1+\h A_1)^TP(B_1+\h B_1)$$K\_1\^[-1]{}$$(B+\h
B)^TP+(B_1+\h B_1)^TP(A_1+\h A_1)$$=0,\
P(T)=G..The solvability of this Riccati equation is not obvious since $\h G$ is just assumed to be symmetric, and $$\ba{ll}
\ns\ds\wt Q\equiv P\h A+\h
A^TP+(A_1+\h A_1)^TP(A_1+\h A_1)-A_1^TPA_1+\h Q+(PB+A_1^TPB_1)K_0^{-1}(B^TP+B_1^TPA_1)\\
\ns\ds\qq-\[P(B+\h B)+(A_1+\h A_1)^TP(B_1+\h B_1)\]K_1^{-1}\[(B+\h
B)^TP+(B_1+\h B_1)^TP(A_1+\h A_1)\]\ea$$ is also just symmetric. To look at the solvability of such a Riccati equation, we let $$\Pi=P+\h P.$$ Then $$\Pi(T)=G+\h G\ge0,$$ and consider the following $$\ba{ll}
\ns\ds0=P'+PA+A^TP+A_1^TPA_1+Q-(PB+A_1^TPB_1)K_0^{-1}(B^TP+B_1^TPA_1)\\
\ns\ds\qq+\h P'+\h P(A+\h A)+(A+\h A)^T\h P+P\h A+\h
A^TP+(A_1+\h A_1)^TP(A_1+\h A_1)-A_1^TPA_1+\h Q\\
\ns\ds\qq+(PB+A_1^TPB_1)K_0^{-1}(B^TP+B_1^TPA_1)\\
\ns\ds\qq-\[(P\1n+\1n\h P)(B\1n+\1n\h B)\1n+\1n(A_1\1n+\1n\h
A_1)^TP(B_1\1n+\1n\h B_1)\]K_1^{-1}\[(B\1n+\1n\h B)^T(P\1n+\1n\h
P)\1n+\1n(B_1\1n+\1n\h B_1)^TP(A_1\1n+\1n\h
A_1)\]\\
\ns\ds\q=\Pi'\1n+\1n P(A\1n+\1n\h A)\1n+\1n(A\1n+\1n\h A)^TP\1n+\1n Q\1n+\1n\h Q\1n+\1n(\Pi\1n-\1n P)(A\1n+\1n
\h A)\1n+\1n(A\1n+\1n\h A)^T(\Pi\1n-\1n P)\1n+\1n(A_1\1n+\1n\h A_1)^TP(A_1\1n+\1n\h A_1)\\
\ns\ds\qq-\[\Pi(B+\h B)+(A_1+\h A_1)^TP(B_1+\h B_1)\]K_1^{-1}\[(B+\h
B)^T\Pi+(B_1+\h B_1)^TP(A_1+\h A_1)\]\\
\ns\ds\q=\Pi'+\Pi\[(A+\h A)-(B+\h B)K_1^{-1}(B_1+\h B_1)^TP(A_1+\h A_1)\]\\
\ns\ds\qq+\[(A+\h A)^T-(A_1+\h A_1)^TP(B_1+\h B_1)K_1^{-1}(B+\h
B)^T\]\Pi-\Pi(B+\h B)K_1^{-1}
(B+\h B)^T\Pi\\
\ns\ds\qq+(A_1+\h A_1)^T\[P-P(B_1+\h B_1)K_1^{-1}(B_1+\h
B_1)^TP\](A_1+\h A_1)+Q+\h Q.\ea$$ Thus, $\Pi(\cd)$ is the solution to Riccati equation (\[Riccati Pi\]). Consequently, Riccati equation (\[Riccati hP\]) admits a unique solution $\h P(\cd)=\Pi(\cd)-P(\cd)$. Then we obtain (from (\[Y\]) and (\[u\])) $$Y=P\(\bar X-\dbE[\bar X]\)+\Pi\dbE[\bar X],$$ and $$\bar u=-K_0^{-1}(B^TP+B_1^TPA_1)\(\bar X-\dbE[\bar
X]\)-K_1^{-1}\[(B+\h B)^T\Pi+(B_1+\h B_1)^TP(A_1+\h A_1)\]\dbE[\bar
X].$$ Also, from (\[Z\]), it follows that $$\3n\ba{ll}
\ns\ds Z=P\(A_1\bar X+B_1\bar u+\h A_1\dbE[\bar X]+\h B_1\dbE[\bar
u]\)\\
\ns\ds\q=\1n PA_1\bar X\1n-\1n PB_1\Big\{\1n K_0^{-1}(B^T\1n
P\1n+\1n B_1^T\1n PA_1)\(\bar X\1n-\1n\dbE[\bar X]\)\2n+\1n
K_1^{-1}\[(B\2n+\1n\h B)^T\Pi\1n+\1n(B_1\1n+\1n\h B_1)^T\1n
P(A_1\2n+\1n\h
A_1)\]\dbE[\bar X]\1n\Big\}\\
\ns\ds\qq+P\h A_1\dbE[\bar X]\1n-\1nP\h B_1K_1^{-1}\[(B\1n+\1n\h
B)^T\Pi\1n+\1n(B_1\1n+\1n\h
B_1)^TP(A_1\1n+\1n\h A_1)\]\dbE[\bar X]\\
\ns\ds\q=\[PA_1-PB_1K_0^{-1}(B^TP+B_1^TPA_1)\]\(\bar X-\dbE[\bar
X]\)\\
\ns\ds\qq+\[P(A_1+\h A_1)-P(B_1+\h B_1)K_1^{-1}\((B+\h
B)^T\Pi+(B_1+\h B_1)^TP(A_1+\h A_1)\)\]\dbE[\bar X].\ea$$ Hence, (\[representation\]) follows. Plugging the above representations into the state equation, we obtain $$\ba{ll}
\ns\ds d\bar X\1n=\1n\Big\{A\bar X\1n-\1n B\[K_0^{-1}(B^T\1n
P\1n+\1n B_1^TPA_1)\(\bar X\1n-\1n\dbE[\bar X]\)\1n-\1n
K_1^{-1}\((B\1n+\1n\h B)^T\Pi\1n+\1n(B_1\1n+\1n\h
B_1)^TP(A_1\1n+\1n\h A_1)\)\dbE[\bar X]\]\\
\ns\ds\qq+\h A\dbE[\bar X]-\h BK_1^{-1}\[(B+\h B)^T\Pi+(B_1+\h
B_1)^TP(A_1+\h A_1)\]\dbE[\bar X]\Big\}dt\\
\ns\ds\qq+\Big\{A_1\bar X\1n-\1n B_1\[K_0^{-1}(B^T\1n P\1n+\1n
B_1^T\1n PA_1)\(\bar X\1n-\1n\dbE[\bar X]\)\1n-\1n
K_1^{-1}\((B\1n+\1n\h B)^T\Pi\1n+\1n(B_1\1n+\1n\h B_1)^T\1n
P(A_1\1n+\1n\h A_1)\)\dbE[\bar X]\]\\
\ns\ds\qq+\h A_1\dbE[\bar X]-\h B_1K_1^{-1}\[(B+\h B)^T\Pi+(B_1+\h
B_1)^TP(A_1+\h A_1)\]\dbE[\bar X]\Big\}dW(t)\\
\ns\ds\q=\Big\{\[A-BK_0^{-1}(B^TP+B_1^TPA_1)\]\(\bar X-\dbE[\bar X]\)\\
\ns\ds\qq+\[(A+\h A)-(B+\h B)K_1^{-1}\((B+\h B)^T\Pi+(B_1+\h
B_1)^TP(A_1+\h A_1)\)\]\dbE[\bar X]\Big\}dt\\
\ns\ds\qq+\Big\{\[A_1-B_1K_0^{-1}(B^TP+B_1^TPA_1)\]\(\bar X-\dbE[\bar X]\)\\
\ns\ds\qq+\[(A_1+\h A_1)-(B_1+\h B_1)K_1^{-1}\((B+\h B)^T\Pi+(B_1+\h
B_1)^TP(A_1+\h A_1)\)\]\dbE[\bar X]\Big\}dW(t).\ea$$ This gives the closed-loop system (\[closed-loop\]). From the above derivation, we see that if $\bar X(\cd)$ is a solution to (\[closed-loop\]), and $(\bar u(\cd),Y(\cd),Z(\cd))$ are given by (\[representation\]), then $(\bar X(\cd),\bar
u(\cd),Y(\cd),Z(\cd))$ is an adapted solution to MF-FBSDE (\[MF-FBSDE1\]). By Corollary 3.3, we know that such a constructed 4-tuple $(\bar X(\cd),\bar u(\cd),Y(\cd),Z(\cd))$ is the unique solution to (\[MF-FBSDE1\]).
The following is a direct verification of optimality of state feedback control.
**Theorem 4.2. *Let [(H1)]{} and [(H2)$''$]{} hold. Let $P(\cd)$ and $\Pi(\cd)$ be the solutions to the Riccati equations $(\ref{Riccati P})$ and $(\ref{Riccati Pi})$, respectively. Then the state feedback control $\bar u(\cd)$ given in $(\ref{representation})$ is the optimal control of Problem [(MF)]{}. Moreover, the optimal value of the cost is given by \_[u()]{}J(x;u())=(0)x,x,x\^n.***
*Proof. Let $P(\cd)$ and $\Pi(\cd)$ be the solutions to the Riccati equations (\[Riccati P\]) and (\[Riccati Pi\]), respectively and denote $K_0$ and $K_1$ as in (\[K\]), which are positive definite. Let $\h P(\cd)=\Pi(\cd)-P(\cd)$. Then $\h P(\cd)$ solves (\[Riccati hP\]). Now, we observe $$\ba{ll}
\ns\ds J(x;u(\cd))-\lan\Pi(0)x,x\ran=J(x;u(\cd))-\lan[P(0)+\h
P(0)]x,x\ran\\
\ns\ds=\dbE\int_0^T\Big\{\lan QX,X\ran+\lan\h
Q\dbE[X],\dbE[X]\ran+\lan Ru,u\ran+\lan
\h R\dbE[u],\dbE[u]\ran\\
\ns\ds\qq+\lan P'X,X\ran+2\lan P(AX+Bu+\h A\dbE[X]+\h
B\dbE[u]),X\ran\\
\ns\ds\qq+\lan
P(A_1X+B_1u+\h A_1\dbE[X]+\h B_1\dbE[u]),A_1X+B_1u+\h A_1\dbE[X]+\h B_1\dbE[u]\ran\\
\ns\ds\qq+\lan\h P'\dbE[X],\dbE[X]\ran+2\lan\h
P\dbE[X],(A+\h A)\dbE[X]+(B+\h B)\dbE[u]\ran\Big\}ds\\
\ns\ds=\dbE\2n\int_0^T\2n\Big\{\2n\lan(P'\1n+\1n PA\1n+\1n A^T\1n
P\1n+\1n A_1^T\1n PA_1\2n+\1n Q)X,X\ran\1n+2\lan u,(B^T\1n P\1n+\1n
B_1^T\1n PA_1)X\ran\1n+\1n\lan
(R\1n+\1n B_1^T\1n PB_1)u,u\ran\\
\ns\ds\qq+\lan\h Q\dbE[X],\dbE[X]\ran+\lan\h
R\dbE[u],\dbE[u]\ran+2\lan P(\h A\dbE[X]+\h
B\dbE[u]),\dbE[X]\ran\\
\ns\ds\qq+2\lan\1n P(A_1\dbE[X]\1n+\1n B_1\dbE[u]),\h
A_1\dbE[X]\1n+\1n\h B_1\dbE[u]\1n\ran\1n+\1n\lan\1n P(\h
A_1\dbE[X]\1n+\1n\h B_1\dbE[u]),\h A_1\dbE[X]\1n+\1n\h
B_1\dbE[u]\1n\ran\\
\ns\ds\qq+\lan\h P'\dbE[X],\dbE[X]\ran+2\lan\h
P\dbE[X],(A+\h A)\dbE[X]+(B+\h B)\dbE[u]\ran\Big\}ds\\
\ns\ds=\dbE\int_0^T\3n\Big\{\2n\lan(P'\1n+\1n PA\1n+\1n A^TP\1n+\1n
A_1^TPA_1\1n+\1n Q)X,X\ran\1n+2\lan
u,(B^TP\1n+\1n B_1^TPA_1)X\ran\1n+\1n\lan K_0u,u\ran\\
\ns\ds\qq+\lan[\h P'\1n+\1n P\h A\1n+\1n\h A^T\1n P\1n+\1n\h
A_1^T\1n PA_1\1n+\1n A_1^T\1n P\h A_1\1n+\1n\h A_1^T\1n P\h
A_1\1n+\1n\h P(A\1n+\1n\h A)\1n+\1n(A\1n+\1n\h A)^T\1n\h P\1n+\1n\h
Q]\dbE[X],\dbE[X]\ran\\
\ns\ds\qq+\lan(\h R+\h B_1^TPB_1+B_1^TP\h B_1+\h B_1^TP\h
B_1)\dbE[u],\dbE[u]\ran\\
\ns\ds\qq+2\lan\dbE[u],[\h B^TP+\h B_1^TPA_1+(B_1+\h B_1)^TP\h
A_1+(B+\h B)^T\h P]\dbE[X]\ran
\Big\}ds\\
\ns\ds=\dbE\int_0^T\3n\Big\{\2n\lan(P'\1n+\1n PA\1n+\1n A^TP\1n+\1n
A_1^TPA_1\1n+\1n Q)X,X\ran\1n+2\lan
u\1n-\1n\dbE[u],(B^TP\1n+\1n B_1^TPA_1)(X\1n-\1n\dbE[X])\ran\\
\ns\ds\qq\qq+\lan K_0(u-\dbE[u]),u-\dbE[u]\ran\\
\ns\ds\qq+\lan\(\h P'+P\h A+\h A^TP+\h A_1^TPA_1+A_1^TP\h A_1+\h
A_1^TP\h A_1\\
\ns\ds\qq\qq\qq+\h P(A+\h A)+(A+\h A)^T\h P+\h
Q\)\dbE[X],\dbE[X]\ran+\lan K_1\dbE[u],\dbE[u]\ran\\
\ns\ds\qq+2\lan\dbE[u],(B+\h B)^T(P+\h P)+(B_1+\h B_1)^TPA_1+(B_1+\h
B_1)^TP\h A_1\)\dbE[X]\ran \Big\}ds\ea$$ $$\ba{ll}
\ns\ds=\dbE\int_0^T\2n\Big\{\1n\lan[P'\1n+\1n PA\1n+\1n A^TP\1n+\1n
A_1^TPA_1\1n+\1n Q\1n-\1n
(PB\1n+\1n A_1PB_1)K_0^{-1}(B^TP\1n+\1n B_1^TPA_1)]X,X\ran\\
\ns\ds\qq+\Big|K_0^{1\over2}\[u-\dbE[u]+K_0^{-1}(B^TP+B_1^TPA_1)(X-\dbE[X])\]\Big|^2\\
\ns\ds\qq+\lan\[\h P'+P\h A+\h A^TP+\h A_1^TPA_1+A_1^TP\h A_1+\h
A_1^TP\h A_1+\h P(A+\h A)+(A+\h A)^T\h P+\h Q\\
\ns\ds\qq+(PB+A_1PB_1)K_0^{-1}(B^TP+B_1^TPA_1)\\
\ns\ds\qq-\((P+\h P)(B+\h B)+A_1^TP(B_1+\h B_1)+\h A_1^TP(B_1+\h
B_1)\)K_1^{-1}\\
\ns\ds\qq\q\cd\((B+\h B)^T(P+\h P)+(B_1+\h B_1)^TPA_1+(B_1+\h
B_1)^TP\h A_1\)\]\dbE[X],\dbE[X]\ran\\
\ns\ds\qq+\Big|K_1^{1\over2}\[\dbE[u]\1n+\1n K_1^{-1}\((B\1n+\1n\h
B)^T(P\1n+\1n\h P)\1n+\1n(B_1\1n+\1n\h
B_1)^TPA_1\1n+\1n(B_1\1n+\1n\h B_1)^T P\h
A_1\)\dbE[X]\]\Big|^2\Big\}ds\\
\ns\ds=\dbE\int_0^T\Big\{\Big|K_0^{1\over2}\[u-\dbE[u]+K_0^{-1}(B^TP+B_1^TPA_1)(X-\dbE[X])\]\Big|^2\\
\ns\ds\qq+\Big|K_1^{1\over2}\[\dbE[u]+K_1^{-1}\((B+\h
B)^T\Pi+(B_1+\h B_1)^TPA_1+(B_1+\h B_1)^TP\h
A_1\)\dbE[X]\]\Big|^2\Big\}ds\ge0.\ea$$ Then our claim follows.*
We see that Riccati equation (\[Riccati Pi\]) can be written as follows: {n ’+(A+A)+(A+A)\^T+(A\_1+A\_1)\^TP(A\_1+A\_1)+Q+Q\
-$$\Pi(B+\h B)+(A_1+\h A_1)^TP(B_1+\h B_1)$$K\_1\^[-1]{} $$(B+\h B)^T\Pi+(B_1+\h B_1)^TP(A_1+\h A_1)$$=0,\
s,\
(T)=G+G..When $$\h A=\h A_1=0,\qq\h B=\h B_1=0,\qq\h Q=0,\qq\h R=0,\qq\h G=0,$$ we have $$K_0=K_1=R+B_1^TPB_1,$$ and the Riccati equation for $\Pi(\cd)$ can be written as $$\left\{\ba{ll}
\ns\ds\Pi'+\Pi A+A^T\Pi-(\Pi
B+A_1^TPB_1)K_0^{-1}(B^T\Pi+B_1^TPA_1)+A_1^TPA_1+Q=0,\\
\ns\ds\Pi(T)=G.\ea\right.$$ Then $$\ba{ll}
\ns\ds0=(\Pi-P)'+(\Pi-P)A+A^T(\Pi-P)-(\Pi
B+A_1^TPB_1)K_0^{-1}(B^T\Pi+B_1^TPA_1)\\
\ns\ds\qq+(PB+A_1^TPB_1)K_0^{-1}(B^TP+B_1^TPA_1)\\
\ns\ds\q=(\Pi-P)'+(\Pi-P)A+A^T(\Pi-P)-(\Pi-P)
BK_0^{-1}(B^T\Pi+B_1^TPA_1)\\
\ns\ds\qq-(PB+A_1^TPB_1)K_0^{-1}B^T(\Pi-P).\ea$$ Therefore, by uniqueness, we have $$\Pi=P.$$ Consequently, the feedback control can be written as $$\ba{ll}
\ns\ds\bar u=-K_0^{-1}(B^TP+B_1^TPA_1)\(\bar X-\dbE[\bar
X]\)-K_1^{-1}\[(B+\h B)^T\Pi+(B_1+\h B_1)^TP(A_1+\h A_1)\]\dbE[\bar
X],\\
\ns\ds\q=-K_0^{-1}(B^TP+B_1^TPA_1)\(\bar X-\dbE[\bar
X]\)-K_0^{-1}\(B^TP+B_1^TPA_1\)\dbE[\bar X]\\
\ns\ds\q=-K_0^{-1}(B^TP+B_1^TPA_1)\bar X.\ea$$ This recovers the result for classical LQ problem ([@Yong-Zhou; @1999]).
A Modification of Standard LQ Problems.
=======================================
In this section, we are going to look at a special case which was mentioned in the introduction. For convenience, let us rewrite the state equation here: { dX(s)=$$A(s)X(s)+B(s)u(s)$$ds+$$A_1(s)X(s)+B_1(s)u(s)$$dW(s),\
X(0)=x,.with the cost functional: J\_0(x;u())=$$\int_0^T\(\lan Q_0(s)X(s),X(s)\ran+\lan
R_0(s)u(s),u(s)\ran\)ds+\lan G_0X(T),X(T)\ran$$.Classical LQ problem can be stated as follows.
**Problem (LQ). For any given $x\in\dbR^n$, find a $\bar
u(\cd)\in\cU[0,T]$ such that J\_0(x;|u())=\_[u()]{}J\_0(x;u()).**
The following result is standard (see [@Yong-Zhou; @1999]).
**Theorem 5.1. *Let [(H1)]{} hold and Q\_0(s)0,R\_0(s)I,s;G\_00.Then Problem [(LQ)]{} admits a unique optimal pair $(\bar
X_0(\cd),\bar u_0(\cd))$. Moreover, the following holds: |u\_0(s)=-$$R_0(s)+B_1(s)^TP_0(s)B(s)$$\^[-1]{}$$B(s)^TP_0(s)+B_1(s)^TP_0(s)A_1(s)$$|X\_0(s),s,where $P_0(\cd)$ is the solution to the following Riccati equation: {n P\_0’+P\_0A+A\^TP\_0+A\_1\^TP\_0A\_1+Q\_0\
-(P\_0B+A\_1\^TP\_0B\_1)(R\_0+B\_1\^TP\_0B\_1)\^[-1]{} (B\^TP\_0+B\_1\^TP\_0A\_1)=0,s,\
P\_0(T)=G\_0..and $\bar X(\cd)$ is the solution to the following closed-loop system: {n d|X\_0(s)=$$A-B(R_0+B_1^TP_0B_1)^{-1}(B^TP_0+B_1^TP_0A)$$|X(s)ds\
+$$A_1-B_1(R_0+B_1^TP_0B_1)(B^TP_0+B_1^TP_0A_1)$$|X(s)dW(s),s,\
|X\_0(0)=x..***
We now introduce the following modified cost functional: J\_0(x;u())=$$\int_0^T\(\lan Q_0(s)X(s),X(s)\ran+\lan
R_0(s)u(s),u(s)\ran\)ds+\lan
G_0X(T),X(T)\ran$$\
+$$\int_0^T\(q(s)\var[X(s)]+\rho(s)\var[u(s)]\)ds
+g\var[X(T)$$\
={\_0\^T$$\lan\(Q_0(s)+q(s)I\)X(s),X(s)\ran-q(s)\Big|\dbE[X(s)]\Big|^2+\lan
\(R_0(s)+\rho(s)I\)u(s),u(s)\ran\\
\ns\ds\qq\qq-\rho(s)\Big|\dbE[u(s)]\Big|^2$$ds+$G_0+gI$X(T),X(T)-g||\^2},with $q(\cd),\rho(\cd)\in L^\infty(0,T)$, $g\in[0,\infty)$ such that q(s),(s)0,s.Also, of course, we assume that \_0\^T\[q(s)+(s)\]ds+g>0.We want to compare the above Problem (LQ) with the following problem:
**Problem (LQ)$'$. For any given $x\in\dbR^n$, find a $\bar
u(\cd)\in\cU[0,T]$ such that J\_0(x;|u())=\_[u()]{}J\_0(x;u()).We refer to the above Problem (LQ)$'$ as a [*modified*]{} LQ problem. This is a special case of Problem (MF) with $$\left\{\ba{ll}
\ns\ds\h A=\h A_1=0,\q\h B=\h B_1=0,\\
\ns\ds Q=Q_0+qI,\q\h Q=-qI,\q R=R_0+\rho I,\q\h R=-\rho I,\q
G=G_0+gI,\q\h G=-gI.\ea\right.$$ Then the Riccati equations are {n P’+PA+A\^TP+A\_1\^TPA\_1+Q\_0+qI\
-(PB+A\_1\^TPB\_1)(R\_0+I+B\_1\^TPB\_1)\^[-1]{} (B\^TP+B\_1\^TPA\_1)=0,s,\
P(T)=G\_0+gI,.and {n ’n+nAn+n A\^Tn+n A\_1\^TPA\_1n+n Q\_0n-n(Bn+n A\_1\^TPB\_1)(R\_0n+n B\_1\^TPB\_1)\^[-1]{} (B\^Tn+n B\_1\^TPA\_1)n=n0,s,\
(T)=G\_0..The optimal control is given by |u=-(R\_0+I+B\_1\^TPB\_1)\^[-1]{}(B\^TP+B\_1\^TPA\_1)$\bar
X-\dbE[\bar X]$\
-(R\_0+B\_1\^TPB\_1)\^[-1]{}(B\^T+B\_1\^TPA\_1),and the closed-loop system reads { d|X={$$A-B(R_0+\rho I+B_1^TPB_1)^{-1}(B^TP+B_1^TPA_1)$$$\bar X-\dbE[\bar X]$\
+$$A-B(R_0+B_1^TPB_1)^{-1}\(B^T\Pi+B_1^TPA_1\)$$}ds\
+{$$A_1-B_1K_0^{-1}(B^TP+B_1^TPA_1)$$$\bar X-\dbE[\bar X]$\
+$$A_1-B_1K_1^{-1}\(B^T\Pi+B_1^TPA_1\)$$}dW(s),\
|X(0)=x..By the optimality of $\bar u_0(\cd)$ and $\bar u(\cd)$, we have that P\_0(0)x,x=J\_0(x;|u\_0())J\_0(x;|u()),and (0)x,x=J\_0(x;|u())+$$\int_0^T\(q(s)\var[\bar X(s)]+\rho(s)\var[\bar
u(s)]\)ds+g\var[\bar
X(T)]$$\
=J\_0(x;|u())J\_0(x;|u\_0())\
=J\_0(x;|u\_0())+$$\int_0^T\(q(s)\var[\bar X_0(s)]+\rho(s)\var[\bar u_0(s)]\)ds+g\var[\bar X_0(T)]$$\
J\_0(x;|u())+$$\int_0^T\(q(s)\var[\bar
X_0(s)]+\rho(s)\var[\bar u_0(s)]\)ds+g\var[\bar X_0(T)]$$.This implies $$\int_0^T\(q(s)\var[\bar X(s)]+\rho(s)\var[\bar u(s)]\)ds+g\var[\bar X(T)]$$\
$$\int_0^T\(q(s)\var[\bar X_0(s)]+\rho(s)\var[\bar
u_0(s)]\)ds+g\var[\bar X_0(T)]$$.Hence, $$J_0(x;\bar u(\cd))-J_0(x;\bar u_0(\cd))$$ is the price for the decrease $$\ba{ll}
\ns\ds\dbE\[\int_0^T\(q(s)\var[\bar X_0(s)]+\rho(s)\var[\bar u_0(s)]\)ds+g\var[\bar X_0(T)]\]\\
\ns\ds-\dbE\[\int_0^T\(q(s)\var[\bar X(s)]+\rho(s)\var[\bar
u(s)]\)ds+g\var[\bar X(T)]\ea$$ of the (weighted) variances of the optimal state-control pair $(\bar
X_0(\cd),\bar u_0(\cd))$. Moreover, (\[5.17\]) further implies that $$\ba{ll}
\ns\ds J_0(x;\bar u(\cd))-J_0(x;\bar u_0(\cd))\\
\ns\ds\le\dbE\[\int_0^T\(q(s)\var[\bar X_0(s)]+\rho(s)\var[\bar u_0(s)]\)ds+g\var[\bar X_0(T)]\]\\
\ns\ds\qq-\dbE\[\int_0^T\(q(s)\var[\bar X(s)]+\rho(s)\var[\bar
u(s)]\)ds+g\var[\bar X(T)].\ea$$ The above roughly means that the amount increased in the cost is “covered” by the amount decreased in the weighted variance of the optimal state-control pair.**
We now look at a simple case to illustrate the above. Let us look at a one-dimensional controlled linear SDE: { dX(s)=bu(s)ds+X(s)dW(s),\
X(0)=x,.with cost functionals: J\_0(x;u())=$$\int_0^T|u(s)|^2ds+g_0|X(T)|^2$$,and J\_0(x;u())={\_0\^T|u(s)|\^2ds+g\_0|X(T)|\^2+g}\
={\_0\^T|u(s)|\^2ds+(g\_0+g)|X(T)|\^2-g$\dbE[X(T)]$\^2},where $g_0\ge0$ and $g>0$. As above, we refer to the optimal control problem associated with (\[5.15\]) and (\[5.16\]) as the [*standard LQ problem*]{}, and to that associated with (\[5.15\]) and (\[5.17\]) as the [*modified LQ problem*]{}. The Riccati equation for the standard LQ problem is { p\_0’(s)+p\_0(s)-b\^2p\_0(s)\^2=0,s,\
p\_0(T)=g\_0..A straightforward calculation shows that p\_0(s)=[e\^[T-s]{}g\_0(e\^[T-s]{}-1)b\^2g\_0+1]{}>0,s.The optimal control is |u\_0(s)=-bp\_0(s)|X\_0(s),s,and the closed-loop system is { d|X\_0(s)=-b\^2p\_0(s)|X\_0(s)ds+|X\_0(s)dW(s),s,\
|X\_0(0)=x..Thus, |X\_0(s)=xe\^[-b\^2\_0\^sp\_0()d-[12]{}s+W(s)]{},s.Consequently, =xe\^[-b\^2\_0\^Tp\_0()d-[12]{}T+[12]{}T]{}=xe\^[-b\^2\_0\^Tp\_0()d]{},and =x\^2e\^[-2b\^2\_0\^Tp\_0()d-T+2T]{}=x\^2e\^[-2b\^2\_0\^Tp\_0()d+T]{}.Hence, =-$\dbE[\bar
X_0(T)]$\^2=x\^2e\^[-2b\^2\_0\^Tp\_0()d]{}(e\^T-1).Also, the optimal expected cost is J\_0(x;|u\_0())=p\_0(0)x\^2=[e\^Tg\_0(e\^T-1)b\^2g\_0+1]{}x\^2.Next, for the modified LQ peoblem, the Riccati equations are: { p’(s)+p(s)-b\^2p(s)\^2=0,s,\
p(T)=g\_0+g,.and { ’(s)-b\^2\^2(s)+p(s)=0,s,\
(T)=g\_0..Clearly, p(s)=[e\^[T-s]{}(g\_0+g)(e\^[T-s]{}-1)b\^2(g\_0+g)+1]{}>[e\^[T-s]{}g\_0(e\^[T-s]{}-1)b\^2g\_0+1]{}=p\_0(s)>0,s.We now show that p\_0(s)<(s)<p(s),s.In fact, $$\left\{\ba{ll}
\ns\ds{d\over
ds}[\pi(s)-p_0(s)]-b^2[\pi(s)+p_0(s)][\pi(s)-p_0(s)]+p(s)-p_0(s)=0,\qq
s\in[0,T],\\
\ns\ds\pi(T)-p_0(T)=0.\ea\right.$$ Thus, $$\pi(s)-p_0(s)=\int_s^Te^{-\int_s^tb^2[\pi(\t)+p_0(\t)]d\t}[p(t)-p_0(t)]dt>0,\qq s\in[0,T).$$ Next, $$\left\{\ba{ll}
\ns\ds{d\over ds}[p(s)-\pi(s)]-b^2[p(s)+\pi(s)][p(s)-\pi(s)]=0,\qq
s\in[0,T],\\
\ns\ds p(T)-\pi(T)=g.\ea\right.$$ Hence, $$p(s)-\pi(s)=e^{-\int_s^Tb^2[p(\t)+\pi(\t)]d\t}g>0,\qq s\in[0,T].$$ This proves (\[p0<pi<p\]). Note that the optimal control of modified LQ problem is given by $$\bar u(s)=-bp(s)\(\bar X(s)-\dbE[\bar
X(s)]\)-b\pi(s)\dbE[\bar X(s)],\qq s\in[0,T],$$ and the closed-loop system is { d|X(s)=-$$b^2p(s)\(\bar X(s)-\dbE[\bar
X(s)]\)+b^2\pi(s)\dbE[\bar X(s)]$$ds+|X(s)dW(s),s,\
|X(0)=x..Thus, $${d\over ds}\(\dbE[\bar X(s)]\)=-b^2\pi(s)\dbE[\bar X(s)],$$ which leads to =e\^[-b\^2\_0\^s()d]{}x,s.On the other hand, by Itô’s formula, $$d[\bar X^2]=\Big\{-2\[b^2p\bar X\(\bar
X-\dbE[\bar X]\)+b^2\pi\bar X\dbE[\bar X]\]+\bar
X^2\Big\}ds+[\cds]dW.$$ Then $$\ba{ll}
\ns\ds d\(\dbE[\bar X^2]\)=\Big\{-2b^2p\[\dbE[\bar X^2]-\(\dbE[\bar
X]\)^2\]-2b^2\pi\(\dbE[\bar X]\)^2+\dbE[\bar
X^2]\Big\}ds\\
\ns\ds\qq=\Big\{(1-2b^2p)\dbE[\bar X^2]+2b^2(p-\pi)\(\dbE[\bar
X]\)^2\Big\}ds\\
\ns\ds\qq=\Big\{(1-2b^2p)\dbE[\bar
X^2]+2b^2(p-\pi)e^{-2b^2\int_0^s\pi(\t)d\t}x^2\Big\}ds.\ea$$ Hence, $$\ba{ll}
\ns\ds\dbE[\bar
X^2(s)]=e^{s-2b^2\int_0^sp(\t)d\t}x^2\[1+\int_0^se^{-t+2\int_0^tp(\t)d\t}2b^2[p(t)-\pi(t)]
e^{-2b^2\int_0^t\pi(\t)d\t}dt\]\\
\ns\ds=e^{s-2b^2\int_0^sp(\t)d\t}x^2\[1+\int_0^se^{-t+2b^2\int_0^t[p(\t)-\pi(\t)]d\t}2b^2[p(t)-\pi(t)]
dt\]\\
\ns\ds=e^{s-2b^2\int_0^sp(\t)d\t}x^2\[e^{-s+2b^2\int_0^s[p(\t)-\pi(\t)]d\t}+\int_0^se^{-t+2b^2
\int_0^t[p(\t)-\pi(\t)]d\t}dt\]\\
\ns\ds=\[e^{-2b^2\int_0^s\pi(\t)d\t}+\int_0^se^{s-t-2b^2\int_t^sp(\t)d\t-2b^2\int_0^t\pi(\t)d\t}dt\]x^2.\ea$$ It follows that $$\ba{ll}
\ns\ds\var[\bar
X(s)]=\[e^{-2b^2\int_0^s\pi(\t)d\t}+\int_0^se^{s-t-2b^2\int_t^sp(\t)d\t-2b^2\int_0^s\pi(\t)d\t}dt\]x^2-\(e^{-2b^2\int_0^t\pi(\t)d\t}\)x^2\\
\ns\ds\qq\qq\;=\[\int_0^se^{s-t-2b^2\int_t^sp(\t)d\t-2b^2\int_0^t\pi(\t)d\t}dt\]x^2.\ea$$ Consequently, noting $p_0(s)<p(s)$, we have n=n$$\1n\int_0^T\2ne^{T-t-2b^2\int_t^Tp(\t)d\t-2b^2\int_0^tp_0(\t)d\t}dt$$x\^2 n<n$$\1n\int_0^T\2ne^{T-t-2b^2\int_t^Tp_0(\t)d\t-2b^2\int_0^tp_0(\t)d\t}dt$$x\^2\
=$$\int_0^Te^{T-t-2b^2\int_0^Tp_0(\t)d\t}dt$$x\^2=$$e^{-2b^2\int_0^Tp_0(\t)d\t}
\int_0^Te^{T-t}dt$$x\^2=.On the other hand, we claim that p\_0(s)>[g\_0g\_0+g]{}p(s),s\[0,T).In fact, by letting $\wt p={g_0\over g_0+g}p$, we have $$\left\{\ba{ll}
\ns\ds\wt p'(s)+\wt p(s)-b^2\wt p(s)^2-{g_0g\over(g_0+g)^2}p(s)^2=0,\\
\ns\ds\wt p(T)=g_0.\ea\right.$$ Then $$\left\{\ba{ll}
\ns\ds[p_0'(s)-\wt p'(s)]+[p_0(s)-\wt p(s)]-b^2[p_0(s)+\wt
p(s)][p_0(s)-\wt p(s)]+{g_0g\over(g_0+g)^2}p(s)^2=0,\\
\ns\ds p_0(T)-\wt p(T)=0.\ea\right.$$ This leads to (\[wt p>p\]). Consequently, $$\ba{ll}
\ns\ds J_0(x;\bar u(\cd))+\var[\bar X(T)]=\h J_0(x;\bar
u(\cd))=\pi(0)x^2\le p(0)x^2\le{g_0+g\over g_0}p_0(0)x^2={g_0+g\over
g_0}J(x;\bar u_0(\cd)).\ea$$ Therefore, $$0\le J_0(x;\bar u(\cd))-J_0(x;\bar u_0(\cd))+g\var[\bar X(T)]\le{g\over
g_0}J(x;\bar u_0(\cd)).$$ The above gives an upper bound for the cost increase in order to have a smaller $\var[X(T)]$. Taking into account of (\[5.37\]), we see that it is a very good trade-off to consider the modified LQ problem if one wishes to have a smaller $\var[X(T)]$. It is possible to more carefully calculate the price difference $J_0(x;u(\cd))-J_0(x;\bar u_0(\cd))$. We omit the details here.
Also, it is possible to calculate the situation of including $\var[u(s)]$ and/or $\var[X(s)]$ in the integrand of the modified cost functional. The details are omitted here as well.
To conclude this paper, let us make some remarks. First of all, we have presented some results on the LQ problem for MF-SDEs with deterministic coefficients. Optimal control is represented by a state feedback form involving both $X(\cd)$ and $\dbE[X(\cd)]$, via the solutions of two Riccati equations. Apparently, there are many problems left unsolved (and some of them might be challenging). To mention a few, one may consider the case of infinite-horizon problem, following the idea of [@Wu-Zhou; @2001], and more interestingly, the case of random coefficients (for which one might have to introduce some other techniques since the approach used in this paper will not work). We will continue our study and report new results in our future publications.
[9]{}
N. U. Ahmed and X. Ding, *A semilinear McKean-Vlasov stochastic evolution equation in Hilbert space, *Stoch. Proc. Appl., 60 (1995), 65–85.**
N. U. Ahmed and X. Ding, *Controlled McKean-Vlasov equations, *Comm. Appl. Anal., 5 (2001), 183–206.**
N. U. Ahmed, *Nonlinear diffusion governed by McKean-Vlasov equation on Hilbert space and optimal control, *SIAM J. Control Optim., 46 (2007), 356–378.**
D. Andersson and B. Djehiche, *A maximum principle for SDEs of mean-field type, *Appl. Math. Optim., 63 (2011), 341–356.**
R. Bellman, R. Kalaba, and G. M. Wing, *Invariant imbedding and the reduction of two-point boundary value problems to initial value problems, *Proc Natl Acad Sci USA, 46 (1960), 1646–-1649.**
V. S. Borkar and K. S. Kumar, *McKean-Vlasov limit in portfolio optimization, *Stoch. Anal. Appl., 28 (2010), 884–906.**
R. Buckdahn, B. Djehiche, and J. Li, *A general maximum principle for SDEs of mean-field type, preprint.*
R. Buckdahn, B. Djehiche, J. Li, and S. Peng, *Mean-field backward stochastic differential equations: a limit approach, *Ann. Probab., 37 (2009), 1524-–1565.**
R. Buckdahn, J. Li, and S. Peng, *Mean-field backward stochastic differential equations and related partial differential equations, *Stoch. Process. Appl., 119 (2009), 3133-–3154,**
T. Chan, *Dynamics of the McKean-Vlasov equation, *Ann. Probab. 22 (1994), 431–441.**
T. Chiang, *McKean-Vlasov equations with discontinuous coefficients, *Soochow J. Math., 20 (1994), 507–526.**
D. Crisan and J. Xiong, *Approximate McKean-Vlasov representations for a class of SPDEs, *Stochastics, 82 (2010), 53–68.**
D. A. Dawson, *Critical dynamics and fluctuations for a mean-field model of cooperative behavior, *J. Statist. Phys., 31 (1983), 29–85.**
D. A. Dawson and J. Gärtner, *Large deviations from the McKean-Vlasov limit for weakly interacting diffusions, *Stochastics, 20 (1987), 247–308.**
J. Gärtner, *On the Mckean-Vlasov limit for interacting diffusions, *Math. Nachr., 137 (1988), 197–248.**
C. Graham, *McKean-Vlasov Ito-Skorohod equations, and nonlinear diffusions with discrete jump sets, *Stoch. Proc. Appl., 40 (1992), 69–82.**
M. Huang, R. P. Malhamé, and P. E. Caines, *Large population stochastic dynamic games: closed-loop McKean-Vlasov systems and the Nash certainty equivalence principle, *Comm. Inform. Systems, 6 (2006), 221–252.**
M. Kac, *Foundations of kinetic theory, *Proc. 3rd Berkeley Sympos. Math. Statist. Prob. 3 (1956), 171–197.**
P. E. Kloeden and T. Lorenz, *Stochastic differential equations with nonlocal sample dependence, *Stoch. Anal. Appl., 28 (2010), 937–945.**
P. M. Kotelenez and T. G. Kurtz, *Macroscopic limit for stochastic partial differential equations of McKean-Vlasov type, *Prob. Theory Rel. Fields, 146 (2010), 189–222.**
J. Ma, P. Protter, and J. Yong, *Solving forward-backward stochastic differential equations explicitly — a four-step scheme, *Probab. Theory & Related Fields, 98 (1994), 339–359.**
J. Ma and J. Yong, *Forward-Backward Stochastic Differential Equations and Their Applications, Lecture Notes in Math., Vol. 1702, Springer-Verlag, 1999.*
N. I. Mahmudov and M. A. McKibben, *On a class of backward McKean-Vlasov stochastic equations in Hilbert space: existence and convergence properties, *Dynamic Systems Appl., 16 (2007), 643–664.**
H. P. McKean, *A class of Markov processes associated with nonlinear parabolic equations, *Proc. Natl. Acad. Sci. USA, 56 (1966), 1907–1911.**
T. Meyer-Brandis, B. Oksendal, and X. Zhou, *A mean-field stochastic maximum principle via Malliavin calculus, *A special issue for Mark Davis’ Festschrift, to appear in Stochastics.**
L. Mou and J. Yong, *Two-person zero-sum linear quadratic stochastic differential games by a Hilbert space method, *J. Industrial Management Optim., 2 (2006), 95–117.**
D. Nualart, *The Malliavin Calculus and Related Topics, 2nd Ed., Springer-Verlag, Berlin, 2006.*
J. Y. Park, P. Balasubramaniam, and Y. H. Kang, *Controllability of McKean-Vlasov stochastic integrodifferential evolution equation in Hilbert spaces, *Numer. Funct. Anal. Optim., 29 (2008), 1328–1346.**
M. Scheutzow, *Uniqueness and non-uniqueness of solutions of Vlasov-McKean equations, *J. Austral. Math. Soc., Ser. A, 43 (1987), 246–256.**
S. Tang, *General linear quadratic optimal stochastic control problems with random coefficients: linear stochastic Hamilton systems and backward stochastic Riccati equations, *SIAM J. Control Optim., 42 (2003), 53-–75.**
H. Wu and X. Y. Zhou, *Stochastic frenquency characteristics, *SIAM J. Control Optim., 40 (2001), 557–576.**
A. Yu. Veretennikov, *On ergodic measures for McKean–Vlasov stochastic equations, *From Stochastic Calculus to Mathematical Finance, 623–633, Springer, Berline, 2006.**
J. Yong, *Stochastic optimal control and forward-backward stochastic differential equations, *Computational & Appl. Math., 21 (2002), 369–403.**
J. Yong and X. Y. Zhou, *Stochastic Controls: Hamiltonian Systems and HJB Equations, Springer-Verlag, 1999.*
[^1]: This work is supported in part by NSF Grant DMS-1007514.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We describe Monte Carlo approximation to the maximum likelihood estimator in models with intractable norming constants and explanatory variables. We consider both sources of randomness (due to the initial sample and to Monte Carlo simulations) and prove asymptotical normality of the estimator.'
author:
- |
B[ł]{}ażej Miasojedow [^1], Wojciech Niemiro [^2],\
Jan Palczewski [^3] and Wojciech Rejchel [^4]
title: 'Asymptotics of Monte Carlo maximum likelihood estimators [^5]'
---
[**Keywords:**]{} asymptotic statistics, empirical process, importance sampling, maximum likelihood estimation, Monte Carlo method
Introduction {#Sec:Introduction}
============
Maximum likelihood (ML) is a well-known and often used method in estimation of parameters in statistical models. However, for many complex models exact calculation of such estimators is very difficult or impossible. Such problems arise if considered densities are known only up to intractable norming constants, for instance in Markov random fields or spatial statistics. The wide range of applications of models with unknown norming constants is discussed e.g. in [@MPRB2006]. Methods proposed to overcome the problems with computing ML estimates in such models include, among others, maximum pseudolikelihood (MPL) [@Besag1974] or Monte Carlo maximum likelihood (MCML) [@Cappe2002; @GeyerThom1992; @Geyer1994; @GeyerSung2007]. MPL estimators are easy to compute but not efficient. This is demonstrated e.g. in [@WuHu1997] for an important autologistic spatial model via a simulation study. Comparison of MLP or ,,coding method” with MCML is also discussed in [@HuWu1998]. In our paper we focus on MCML.
In influential papers [@GeyerThom1992; @Geyer1994] the authors prove consistency and asymptotic normality of MCML estimators under the assumption that the initial sample is fixed, and only the Monte Carlo sample size tends to infinity. Both sources of randomness (one due to the initial sample and the other due to Monte Carlo simulations) are considered in [@Cappe2002; @GeyerSung2007; @Imput2010]. Authors of the first mentioned paper apply the general importance sampling recipe. They show that for their scheme of simulations, the Monte Carlo sample size has to grow exponentially fast to ensure consistency of the estimator. As the remedy for this problem they propose to use a preliminary estimator which is consistent. Another possibility to overcome this problem is proposed in [@GeyerSung2007]. The log-likelihood is first decomposed into independent summands and then importance sampling is applied. Papers [@Cappe2002; @GeyerSung2007] describe asymptotic properties of MCML estimators for models with missing data. In our paper we consider models with intractable norming constants and explanatory variables. We apply argumentation similar to [@GeyerSung2007] in our setting.
We consider a parametric model with covariates $$\nonumber
p(y|x,\th)=\frac{1}{\Z(x,\th)}f(y|x,\th),$$ where $y \in \mY \subset \mathbb{R}^d$ is a response variable, $x \in \mX \subset
\mathbb{R}^l$ is a covariate or “explanatory” variable (random or deterministic), $\th \in \mathbb{R} ^p$ is a parameter describing the relation between $y$ and $x$. The norming constant, $$\nonumber
\Z(x,\th) = \int f(y|x,\th) dy,$$ is difficult or intractable.
Assume that the data consist of $n$ independent observations $(Y_1,X_1),\ldots,$ $(Y_n,X_n).$ If we regard covariates as random, then we assume that these pairs form an i.i.d. sample from a joint distribution with a density $g(y,x)$. Alternatively, $x_i$ can be regarded as deterministic and then we assume that random variable $Y_i$ has a probability distribution $g_i$ which depends on $x_i$. Both cases can be analysed very similarly. For simplicity we focus attention on the model with random covariates. It is not necessary to assume that $g(y | x)=p(y|x,\th_0)$ for some $\th_0$. The case when no such $\th_0$ exists, i.e.the model is misspecified, makes the considerations only slightly more difficult. Thus, let us consider the following log-likelihood $$\begin{aligned}
\nonumber \label{eq:likn}
\lln(\th) &=&\log p(Y_1,\ldots,Y_n|X_1,\ldots,X_n,\th)\\
&=&\sum_{i=1}^{n} \log f(Y_i|X_i,\th)- \sum_{i=1}^{n} \log {\Z(X_i,\th)} .\end{aligned}$$ The first term in (\[eq:likn\]) is easy to compute while the second one is approximated by Monte Carlo (MC). Let $h(y)$ be an importance sampling (instrumental) distribution and note that $$\nonumber
{\Z(x,\th)}=\int f(y|x,\th)\d y=\int \frac{f(y|x,\th)}{h(y)}h(y)\d y=\Ex_{Y\sim h}\frac{f(Y|x,\th)}{h(Y)}.$$ Thus, an MC approximation of the log-likelihood $\lln(\th)$ is $$\label{eq:fr}
\llnm(\th) =\sum_{i=1}^{n }\log f(Y_i|X_i, \th)-\sum_{i=1}^{n}\log C_m(X_i, \th) ,$$ where $$C_m(x,\th) = \frac{1}{m}\sum_{k=1}^{m}\frac{f(Y^k|x, \th)}{h(Y^k)}$$ and $Y^1,\ldots, Y^m$ is a sample drawn form $h$.
Let us note that the general Monte Carlo recipe can also lead to approximation schemes different from . For instance, we could generate $n$ independent MC samples instead of one, i.e. $Y_i^1,\ldots,Y_i^m \sim h_i, i=1,\ldots,n$ and use $i$th sample to approximate $\Z (x_i, \theta).$ Using this scenario one can obtain estimators with better convergence rates, but at the cost of increased computational complexity. Another scheme, proposed in [@Cappe2002], approximates the log-likelihood by $$\label{Cappe}
\sum_{i=1}^{n }\log f(Y_i|X_i, \th)-\log \frac{1}{m}\sum_{k=1}^{m}\prod_{i=1}^{n}\frac{f(Y_i^k|X_i, \th)}{h_i(Y_i^k).
}$$
However, this scheme leads to estimators with unsatisfactory asymptotics unless a preliminary estimator is used. Thus, we focus our attention only on $(\ref{eq:fr}).$
Let $\thn$ be a maximizer of $\lln(\th)$ (a genuine maximum likelihood estimator). It is well-known that under some regularity assumptions [@Pollard1984; @vaart1998] $$\nonumber %\label{eq:mle}
\thn \sim\app \mathcal{N} \left(\tht,\frac{1}{n}D^{-1}VD^{-1}\right),$$ where $\tht$ is a maximizer of $\Ex_{(Y,X)\sim g} \log p(Y|X,\th)$, i.e. Kullback-Leibler projection, $D=\Ex_{(Y,X)\sim g} \hess \log p(Y|X,\tht)$ and $V=\VAR_{(Y,X)\sim g}\grad \log p(Y|X,\tht)$. Symbols $\grad$ and $\hess$ denote derivatives with respect to $\th$ and $\VAR$ stands for the variance-covariance matrix. In Theorem \[with\_cov\] we will prove that the maximizer of (\[eq:fr\]), denoted by $\thnm$, satisfies $$\label{asymp}
\thnm
\sim\app \mathcal{N} \left(\tht,D^{-1}\left(\frac{V}{n}+\frac{W}{m}\right)D^{-1}\right),$$ where the matrix $W$ will be given later. Formula (\[asymp\]) means that the estimator $\thnm$ behaves like a normal vector with the mean $\tht$ when both the initial sample size $n$ and the Monte Carlo sample size $m$ are large. Note that the first component of the asymptotic variance in (\[asymp\]), $D^{-1}VD^{-1}/n$, is the same as the asymptotic variance of the maximum likelihood estimator $\thn.$ The second component, $D^{-1}WD^{-1}/m$, is due to Monte Carlo randomness. Furthermore, if $m$ is large, then asymptotic behaviour of $\thnm$ and $\thn$ is similar. If the model is correctly specified, that is $g(y|x) = p(y|x , \th_0)$ for some $\th_0,$ then $\tht = \th_0$ and $D=-V$ (under standard assumptions on passing the derivative under the integral sign).
The choice of the instrumental distribution $h$ affects $W$ and thus the asymptotic efficiency of MCML. In [@MNPR2014b Equation (2.17)] a formula for optimal $h$ is derived (this $h$ minimizes the trace of $W$ in a model without covariates). This result may be of some theoretical interest but has a limited practical value, because the optimal $h$ can be very difficult to sample from. On the other hand, a more practical approach, suggested by several authors, e.g. [@Cappe2002; @Imput2010], is to select some distribution in the underlying parametric family, i.e. to put $$\nonumber
h(y)=p(y|\psi)=\frac{1}{\Z(\psi)}f(y|\psi),$$ for some fixed $\psi\in\Rl^p$ (here we restrict attention to models without covariates). It is natural to guess that a “good choice” of $\psi$ should be close to the target, $\tht$. Since $\tht$ is unknown, one can use a preliminary estimator. Such a choice of $h$ is recommended in [@Cappe2002; @Imput2010]. In the first of the cited papers, theoretical results are given which justify using a consistent preliminary estimate of $\tht$ as $\psi$, compare [@Cappe2002 Theorems 4 and 7]. However, the results are about sampling scheme . In [@Imput2010], sampling scheme $(\ref{eq:fr})$ is considered and the choice of $\psi$ near $\tht$ is recommended on heuristical grounds. In fact the intuition behind this choice turns out to be wrong, as demonstrated by the following toy example.
Let $\Y=\{0,1\}$ and $f(y|\th)=\e^{\th y}$ for $\th\in\Rl$. Of course, the norming constant $C(\th)=1+\e^\th$ is easy and there is no need to apply MCML, but the simplicity of this model will allow us to clearly illustrate our point. Assume we have an i.i.d. sample $Y_1,\ldots,Y_n$ from $f(\cdot|\tht)/C(\tht)$. The MLE is $\thn=\log(\yn/(1-\yn))$, where $\yn=n^{-1}\sum_{i=1}^n Y_i$. Now suppose that we use MCML approximation with $h(y)=f(y|\psi)/{\Z(\psi)}$. It can be easily shown that the asymptotic variance $W$ (now a scalar) is minimum for $\psi_\star=0$ – and not for $\psi=\tht$! The following direct derivation explains this fact. The formula now assumes the form $$\nonumber
\llnm(\th)=n\th\yn-n\log\left(\frac{1}{m}\sum_{k=1}^m\e^{(\th-\psi)Y^k}\right)-n\log\Z(\psi).$$ On noting that $$\nonumber
\frac{1}{m}\sum_{k=1}^mY^k\e^{(\th-\psi)Y^k}=\ym\e^{\th-\psi},\quad
\frac{1}{m}\sum_{k=1}^m\e^{(\th-\psi)Y^k}=\ym\e^{\th-\psi}+(1-\ym)$$ we see that the equation $\grad \llnm(\th)=0$ is equivalent to $$\nonumber
\yn-\dfrac{\ym\e^{\th-\psi}}{\ym\e^{\th-\psi}+(1-\ym)}=0.$$ After elementary computations we obtain that the solution $\thnm$ of this equation is $$\nonumber
\thnm=\log\dfrac{\yn}{1-\yn}+\psi-\log\dfrac{\ym}{1-\ym}.$$ Let us rewrite this expression as follows: $$\nonumber
\thnm=\thn+\psi-\hat\psi^m,$$ where $\hat\psi^m$ is an ML estimate of $\psi$ based on the MC sample. It is clear that $\sqrt{m} (\psi-\hat\psi^m)\to\N(0,\e^{-\psi}(1+ \e^\psi)^2)$, independently of $\th$. The asymptotic variance of the MC error is minimum for $\psi_\star=0$. The overall error of MCML is the sum of two independent terms $(\thn-\tht)+(\hat\psi^m-\psi_\star)$.
Asymptotic properties of MCML estimator (consistency, rates of convergence, asymptotic normality) can be obtained using standard statistical methods from the empirical processes theory [@Pollard1984; @vaart1998]. However, these tools should be adjusted to the model with double randomness when both sample sizes $n$ and $m$ tend to infinity simultaneously. This adaptation makes our proofs very arduous and technical despite the fact that the main ideas are rather clear. Therefore, to make the paper more transparent we present only the proof of asymptotic normality. This result is the most important from a practical point of view. Moreover, the argumentation used in proving this property illustrates well how to adapt standard methods to the double randomness setup. Similar adaptation can be used to obtain consistency and the rate of convergence of the MCML estimator. Since the proof of (\[asymp\]) for the model with covariates is rather complicated, we begin in Section 2 with a model without covariates. It is extended to the general case in Section 3.
As we have already mentioned, related results on MCML for missing data models can be found in [@Cappe2002; @GeyerSung2007]. In particular, our theorems are of similar form as those in [@GeyerSung2007]. However, models with intractable norming constants and observable covariates, considered in our paper, are more difficult to analyse. Let us also mention that for the missing data models there exists another powerful tool for computing maximum likelihood estimates, namely the EM (expectation-maximization) algorithm [@DempsterLairdRubin1977]. The expectation step (E-step) can be implemented using MC computations resulting in Monte Carlo EM (MCEM) algorithm which has been examined in several papers [@WeiTanner1990; @LevineCasella2001; @FortMoulines2003]. MCEM cannot be applied to models with intractable norming constants and observable covariates. This points to particular importance of MCML in this setting and motivates examination of its behaviour.
Model without covariates
========================
First, we consider a model without covariates $$\nonumber
p(y|\th)=\frac{1}{\Z(\th)}f(y|\th)$$ with an intractable norming constant $\Z(\th ) = \int f(y|\th) dy.$ Assume we have an i.i.d. sample $Y_1,\ldots,Y_n\sim g(y)$. Similarly to the general case, we allow for misspecification of the model, i.e.we do not assume $g(y)=p(y|\th_0)$ for some $\th_0$. In what follows, $\tht$ is a maximizer of $\Ex_{Y\sim g} \log p(Y|\th)$, i.e.the Kullback-Leibler projection. The MC approximation (\[eq:fr\]) multiplied by $\frac{1}{n}$ is denoted by $$\nonumber %\label{eq:likm}
\begin{split}
\lanm(\th) &=\frac{1}{n}\sum_{i=1}^{n }\log f(Y_i|\th)-\log \frac{1}{m}\sum_{k=1}^{m}\frac{f(Y^k|\th)}{h(Y^k)}
=\lan(\th) - \rem(\th),
\end{split}$$ where $$\begin{aligned}
\lan(\th) &=& \frac{1}{n} \sum_{i=1}^{n}\left[ \log f(Y_i|\th)-\log {\Z(\th)}\right], \\
\rem(\th) &=&\log \frac{1}{m}\sum_{k=1}^{m}\frac{f(Y^k|\th)}{h(Y^k)}-\log {\Z(\th).}\end{aligned}$$ Now we can state the main result of this section.
\[without\_cov\] For some $\delta >0$ let $U= \{\th : |\th - \tht| \leq \delta\} $ be a neighbourhood of $\tht.$ If the following assumptions are satisfied:
1. second partial derivatives of $f(y |\th)$ with respect to $\th$ exist and are continuous for all $y$, and can be passed under the integral sign in $\int f(y| \th) dy,$
2. $\sqrt{\min(n,m)} \left( \thnm - \tht \right) = O_p (1),$
3. matrices $$\begin{aligned}
V&=&\VAR_{Y\sim g}\grad \log p(Y|\tht), \\
D&=&
\Ex_{Y\sim g} \hess \log p(Y|\tht)\\
W &=& \frac{1}{\Z^2(\tht)} \VAR_{Y \sim h}
\left[\dfrac{\grad f(Y|\tht)}{h(Y)}-\dfrac{\grad \Z(\tht)}{\Z(\tht)}\dfrac{ f(Y|\tht)}{h(Y)}\right]\end{aligned}$$ exist and $D$ is negative definite,
4. function $D(\th) = \Ex_{Y\sim g} \hess \log p(Y|\th) $ is continuous at $\tht$,
5. $
\sup\limits_{\th \in U} | \hess \lan (\th) - D(\th)| \topr 0 , \quad n
\rightarrow \infty,
$
6. $
\sup\limits_{\th \in U} | \hess C_m (\th) - \hess C(\th)| \topr 0, \quad
m \rightarrow \infty,
$
then $$\nonumber
\left( \frac{V}{n} + \frac{W}{m} \right)^{- \frac{1}{2}}\: D \left(\thnm -
\tht
\right) \tod \mathcal{N} (0,I), \quad n,m \rightarrow \infty.$$
Note that 1 and 3 are rather standard regularity assumptions. Condition 2 stipulates the square root consistency of the MCML estimator. If the MC approximation $\lanm(\th)$ is concave (as in the example studied below), then assumption 2 is automatically fulfilled [@Niemiro92]. Otherwise, it can be deduced from more explicit assumptions by adapting standard methods from the empirical processes theory [@Pollard1984; @vaart1998] to the double randomness problem. For simplicity, we do not explore this topic. We just choose condition 2 as a starting point of our argumentation (which is by itself quite complicated).
We shall show that conditions 4 - 6 are satisfied for exponential families, i.e. if $$f(y|\th)= \exp( \th\t W(y))$$ with $W(y) = \left( W_1(y), \ldots, W_p(y)\right).$ We can easily verify that $\hess \log p(y|\th)= - \hess \log C(\th), $ so assumptions 4 and 5 are obviously fulfilled. Thus, condition 6 is the last one to establish. Function $ \hess
C_m(\th)$ is matrix-valued, so it is enough to prove that for each component (that is for each $r,s=1, \ldots, p$) $$\label{VC}
\sup\limits_{\th \in U} \left| \left[ \hess C_m(\th) \right]_{rs} - \left[ \hess C(\th) \right]_{rs} \right| \to_p 0, \quad m \rightarrow \infty.$$ Consider a family of functions $$\label{VCfam}
\left\{ \left[ \frac{\hess f(y | \th) }{h(y)} \right]_{rs } =
\exp( \th\t W(y)) \, \frac{W_r(y) W_s(y)}{h(y) } : \th \in U
\right\}.$$ The set $U$ is compact, so to obtain it is sufficient to assume functions
in are dominated by an integrable function (see [@Ferguson1996 Theorem 16(a)], [@vaart1998 Example 19.8]), i.e. for each $r,s$ there is a function $\eta$ such that $\Ex_{Y\sim h} \eta(Y) < \infty$ and $\big|
[ \hess f(y | \th) / h(y) ]_{rs } \big| \le \eta(y)$ for each $\th, y$.
Without loss of generality we can assume that $\tht = 0.$ First we assume that $\frac{n}{n+m}\to a$ and consider three cases corresponding to rates at which $n$ and $m$ go to infinity: $0<a<1$, $a=0$ and $a=1$. Once our theorem is proved in these three special cases, standard application of the subsequence principle shows that it is valid in general (for $n\to \infty$ and $m\to \infty$ at arbirary rates).
We begin with the case $0<a<1$. It is well-known (see [@Pollard1984 Theorem VII.5]) that we need to prove $$\label{as_norm1}
\left( \frac{V}{n} + \frac{W}{m} \right)^{- \frac{1}{2}} \grad \lanm(0)
\rightarrow_d \mathcal{N} (0,I), \quad n,m \rightarrow \infty$$ and for every $M>0$ $$\label{sup1}
(n+m) \sup_{|\th|\leq \frac{M}{\sqrt{n+m}}} \left|\lanm(\th)-\lanm(0)- \th\t \grad\lanm(0)-\frac{1}{2}\th\t D\th\right| \topr 0, \quad n,m \rightarrow \infty.$$
To obtain (\[as\_norm1\]) notice that $$\label{as_norm11}
\sqrt{n+m} \grad \lanm(0) = \sqrt{\frac{n+m}{n}} \: \sqrt{n} \grad \lan(0)
- \sqrt{\frac{n+m}{m}} \: \sqrt{m} \grad \rem(0)$$ and the terms on the right hand side in (\[as\_norm11\]) are independent. We can calculate the gradient $$\nonumber
\grad\rem(0) =\frac{\dfrac{1}{m}\sum\limits_{k=1}^{m}\left[\dfrac{\grad f(Y^k|0)}{h(Y^k)}-
\dfrac{\grad\Z(0)}{\Z(0)} \dfrac{f(Y^k|0)}{h(Y^k)}\right]}
{\dfrac{1}{m}\sum\limits_{k=1}^{m} \dfrac{f(Y^k|0)}{h(Y^k)}}.$$ Therefore, by LLN, CLT and Slutsky’s theorem we have that $\sqrt{m} \grad \rem(0) \tod \mN (0,W) $ and $\sqrt{n} \grad \lan(0) \tod \mN(0,V) $ which implies $$\nonumber
\sqrt{n+m} \grad \lanm(0) \tod \mN \left(0, V/a + W/ (1-a)\right), \quad n,m
\rightarrow \infty.$$ Thus, we obtain (\[as\_norm1\]) since $$\sqrt{n+m} \left(V/a + W/ (1-a)\right)^{-\frac{1}{2}}
\left(V/n + W/ m \right)^{\frac{1}{2}} \rightarrow I \quad n,m \rightarrow \infty.$$
Now we focus on (\[sup1\]). Using Taylor expansion it can be bounded by $$\label{sup11}
\frac{M^2}{2}\!\! \left( \sup_{\th \in U^m_n} \left|\hess \lan(\th)-D(\th)\right| + \sup_{\th \in U_n^m}\left|D(\th)-D(0)\right| +
\sup_{\th \in U^m_n} \left|\hess \rem(\th) \right| \right)$$ for $U^m_n = \{ \th : |\th| \leq \frac{M}{\sqrt{n+m}} \}.$ First two terms in (\[sup11\]) tend to zero in probability by assumptions 4 and 5. We prove that assumption 6 implies convergence to zero in probability of the third term in (\[sup11\]). Calculating the second derivative we get $$\nonumber
\hess \rem(\th)=\frac{ \hess C_m(\th)}{ C_m(\th)}-\frac{\grad C_m(\th) \grad\t C_m(\th)}{C_m^2(\th)}
-\frac{\hess \Z(\th)}{\Z(\th)}+\frac{\grad \Z(\th)\grad\t \Z(\th)}{\Z^2(\th)}.$$ Therefore $$\begin{aligned}
&&\sup_{\th \in U} \left|\hess \rem(\th) \right| \leq \sup_{\th \in U}
\frac{ |\hess C_m(\th)| \, |C_m(\th) - C(\th)|}{ C_m(\th) C(\th)} +
\sup_{\th \in U} \frac{ | \hess C_m(\th) - \hess C(\th)|}{ C(\th)}\\
&&+ \sup_{\th \in U}
\frac{ |\grad C_m(\th)|^2 \, |C_m^2(\th) - C^2(\th)|}{ C_m^2(\th) C^2(\th)}+
\sup_{\th \in U} \frac{ | \grad C_m(\th) \grad\t C_m(\th)-
\grad C(\th)\grad\t C(\th)|}{ C^2(\th)}.\end{aligned}$$ Note that continuous functions $C(\th), |\grad C (\th)|, |\hess C (\th)|$ are bounded on the compact set $U$, in particular function $C(\th)$ is separated from zero. Therefore, all we need is assumption 6 and $$\begin{aligned}
\label{sup12}
\sup_{\th \in U} | C_m (\th) - C(\th)| \topr 0, \quad
m \rightarrow \infty, \\ \label{sup125}
\sup_{\th \in U} | \grad C_m (\th) - \grad C(\th)| \topr 0, \quad
m \rightarrow \infty.\end{aligned}$$ However, uniform convergence in (\[sup12\]) and (\[sup125\]) easily follows from Taylor expansion, LLN and assumption 6. For instance, for some $\th' \in (0,\th)$ $$\nonumber
\grad C_m(\th)-\grad C( \th) =\grad C_m(0) - \grad C( 0) +
\left[ \hess C_m (\th') - \hess C(\th')\right] \th,$$ so $$\nonumber
\sup_{\th \in U } |\grad C_m(\th)-\grad C( \th) | \leq|\grad C_m(0) - \grad C( 0)| +
\delta \, \sup_{\th \in U } \left| \hess C_m (\th) - \hess C(\th)\right|.$$ Thus, the proof in the case $0<a<1$ is finished. For $a=0$ or $a=1$ we proceed similarly. For example, if $a=0$, then we should prove an analog of (\[sup1\]), namely for every $M>0$ $$\label{a0}
n \sup_{|\th|\leq \frac{M}{\sqrt{n}}} \left|\lanm(\th)-\lanm(0)- \th\t \grad\lanm(0)-\frac{1}{2}\th\t D\th\right| \topr 0, \quad
n,m \rightarrow \infty.$$ Argumentation is almost the same as in the proof of (\[sup1\]). To obtain (\[as\_norm1\]) in this case note that $$\label{as_norma0}
\sqrt{n} \grad \lanm(0) = \sqrt{n} \grad \lan(0)
- \sqrt{\frac{n}{m}} \: \sqrt{m} \grad \rem(0).$$ Therefore, expression (\[as\_norma0\]) tends in distribution to $\mathcal{N} (0, V) .$ Moreover, $$\sqrt{n} V^{-\frac{1}{2}}
\left(V/n + W/ m \right)^{\frac{1}{2}} \rightarrow I, \quad n,m \rightarrow \infty.$$
Model with covariates
=====================
Let us return to the general case and state the main theorem of the paper. We need new notation: $$\begin{aligned}
\phi(y|x) &=&\left[\dfrac{\grad f(y|x,\tht)}{h(y)}-\dfrac{\grad \Z(x,\tht)}{\Z(x,\tht)}\dfrac{ f(y|x,\tht)}{h(y)}\right] \frac{1}{\Z(x,\tht)} \;,\\
\rem_n(\th) &=& \frac{1}{n} \sum_{i=1}^n \left[ \log \frac{1}{m} \sum_{k=1}^m
\frac{f(Y^k|X_i,\th)}{ h(Y^k)} - \log \Z (X_i, \th)
\right]\;.\end{aligned}$$ Then $\lanm (\th) = \lan(\th) - \rem_n (\th).$
\[with\_cov\] For some $\delta >0$ let $U= \{\th : |\th - \tht| \leq \delta\} $ be a neighbourhood of $\tht.$ Suppose the following assumptions are satisfied:
1. second partial derivatives of $f(y |x,\th)$ with respect to $\th$ exist and are continuous for all $y$ and $x$, and may be passed under the integral sign in $\int f(y|x, \th) dy$ for fixed $x$,
2. $\sqrt{\min(n,m)} \left( \thnm - \tht \right) = O_p (1),$
3. matrices $$\begin{aligned}
V&=&\VAR_{(Y,X)\sim g}\grad \log p(Y|X,\tht), \\
D&=&
\Ex_{(Y,X)\sim g} \hess \log p(Y|X,\tht)\end{aligned}$$ and the expectation $\tilde{W} = \Ex_{Y \sim h, X\sim g}
|\phi(Y|X)|^2$ exist and $D$ is negative definite,
4. function $D(\th) = \Ex_{(Y,X)\sim g} \hess \log p(Y|X,\theta) $ is continuous at $\tht$,
5. $\sup_{\th \in U} | \hess \lan (\th) - D(\th)| \rightarrow_P 0 , \quad n
\rightarrow \infty,$
6. 1. $\sup_{x \in \mX} |C_m (x,\tht) - C(x,\tht)| \topr 0, \quad
m \rightarrow \infty,$
2. $\sup_{x \in \mX} |\nabla C_m (x,\tht) - \nabla C(x,\tht)| \topr 0, \quad
m \rightarrow \infty,$
3. $\sup\limits_{\th \in U ,x \in \mX} |\hess C_m (x,\th) - \hess C(x,\th)| \topr 0, \quad
m \rightarrow \infty,$
7. there exist constants $\alpha > 0,$ $K > 0$ such that for each $x \in \mX$ and $\th \in U$ $$\alpha \leq C(x,\th) \leq K, \quad
| \grad C(x,\th)| \leq K, \quad
| \hess C(x,\th)| \leq K.$$
Then matrix $$\nonumber
W=\VAR_{Y \sim h} \: \Ex_{X\sim g}
\; \phi(Y|X)$$ is finite and $$\nonumber
\left( \frac{V}{n} + \frac{W}{m} \right)^{- \frac{1}{2}}\: D \left(\thnm -
\tht
\right) \tod \mathcal{N} (0,I), \quad n,m \rightarrow \infty.$$
We discuss assumptions in Theorem \[with\_cov\] for functions $f(y|x,\th)$ belonging to the exponential family at the end of this section.
Without loss of generality we can assume that $\tht = 0.$
Similarly to the proof of Theorem \[without\_cov\] we consider three cases: $0<a<1$, $a=0$ and $a=1$, where $\frac{n}{n+m}\to a$. Finally, we complete the proof by using the subsequence principle.
We focus on the case $0<a<1$, because for $a=0$ or $a=1$ we proceed in a similar way (cf. the proof of Theorem \[without\_cov\]). It is well-known (see [@Pollard1984 Theorem VII.5]) that we need to prove that for every $M>0$ $$\label{sup2}
(n+m) \sup_{|\th|\leq \frac{M}{\sqrt{n+m}}} \left|\lanm(\th)-\lanm(0)- \th\t \grad\lanm(0)-\frac{1}{2}\th\t D\th\right| \topr 0, \quad n,m \rightarrow \infty,$$ and $$\label{asnorm2}
\left( \frac{V}{n} + \frac{W}{m} \right)^{- \frac{1}{2}} \grad \lanm(0)
\rightarrow_d \mathcal{N} (0,I), \quad n,m \rightarrow \infty.$$ We start with . Using Taylor expansion the left hand side of (\[sup2\]) can be bounded by $$\label{sup21}
\frac{M^2}{2}\!\! \left( \sup_{\th \in U^m_n} \left|\hess \lan(\th)-D(\th)\right| + \sup_{\th \in U_n^m}\left|D(\th)-D(0)\right| +
\sup_{\th \in U^m_n} \left|\hess \rem_n(\th) \right| \right)$$ for $U^m_n = \{ \th : |\th| \leq \frac{M}{\sqrt{n+m}} \}.$ First two terms in (\[sup21\]) tend to zero in probability by assumptions 4 and 5. We prove that assumptions 6 and 7 imply convergence to zero in probability of the third term in (\[sup21\]). Calculating the second derivative of $\rem_n(\th) $ we get $$\begin{aligned}
\nonumber
\hess \rem_n(\th)&=&\frac{1}{n} \sum_{i=1}^n \left[ \frac{ \hess C_m(X_i,\th)}{ C_m(X_i,\th)}-\frac{\grad C_m(X_i,\th) \grad\t C_m(X_i,\th)}{C_m^2(X_i,\th)} \right. \\
&-& \left. \frac{\hess \Z(X_i,\th)}{\Z(X_i, \th)}+\frac{\grad \Z(X_i, \th)\grad\t \Z(X_i, \th)}{\Z^2(X_i, \th) }\right].\end{aligned}$$ Therefore $$\begin{aligned}
\label{sup22}
&&\sup_{\th \in U} \left|\hess \rem_n(\th) \right| \leq \sup_{\th \in U, x \in \mX}
\frac{ |\hess C_m(x,\th)| \, |C_m(x,\th) - C(x,\th)|}{ C_m(x,\th) C(x,\th)} \\
&+&
\sup_{\th \in U, x \in \mX} \frac{ | \hess C_m(x, \th) - \hess C(x, \th)|}{ C(x, \th)}
\nonumber \\
&+& \sup_{\th \in U, x \in \mX}
\frac{ |\grad C_m(x, \th)|^2 \, |C_m^2(x, \th) - C^2(x, \th)|}{ C_m^2(x,\th) C^2(x,\th) }
\nonumber \\
&+&
\sup_{\th \in U, x \in \mX} \frac{ | \grad C_m(x,\th) \grad\t C_m(x, \th)-
\grad C(x,\th)\grad\t C(x, \th)|}{ C^2(x,\th)} . \nonumber\end{aligned}$$ The convergence in assumptions 6(a) and 6(b) can be strengthened to be uniform over $\th \in U$ in the similar way as in the proof of Theorem \[without\_cov\]. Using these arguments and assumption 7 we obtain that for arbitrary $\eta >0$ and sufficiently large $m$ with probability at least $ 1 - \eta$ for each $x \in \mX, \theta \in U$ $$\alpha/2\leq C_m(x,\th)\leq K+\alpha/2.$$ Hence, every term on the right side of (\[sup22\]) tends to zero in probability.
The last step is proving (\[asnorm2\]). First, notice that $$\begin{aligned}
\sqrt{n+m} \grad \lanm(0) &=& \sqrt{\frac{n+m}{n}} \: \sqrt{n} \grad \lan(0)
- \sqrt{\frac{n+m}{m}} \: \sqrt{m} \grad \renm(0) \nonumber \\ \label{asymp1}
&=& \left[ \sqrt{\frac{n+m}{n}} \: \sqrt{n} \grad \lan(0) - \sqrt{\frac{n+m}{m}} \frac{1}{ \sqrt{m}} \sum_{k=1}^m \bar{\phi} (Y^k)
\right] \\ \label{asymp111}
&+& \sqrt{\frac{n+m}{m}} \left[ \frac{1}{ \sqrt{m}} \sum_{k=1}^m \bar{\phi} (Y^k)
-\sqrt{m} \grad \renm(0) \right],\end{aligned}$$ where $\bar{\phi} (y) = \Ex_{X \sim g} \phi(y|X).$ By CLT, the expression (\[asymp1\]) tends in distribution to $\mN(0, V/a + W/(1-a)), $ since the Monte Carlo sample is independent of the observation. To show that the term (\[asymp111\]) tends to zero in probability, we prove that $$\label{asymp12}
\sqrt{m} \grad \renm(0) - \frac{1}{n}\sum_{i=1}^n \frac{1}{ \sqrt{m}} \sum_{k=1}^m
\phi(Y^k| X_i)$$ and $$\label{asymp13}
\frac{1}{n}\sum_{i=1}^n \frac{1}{ \sqrt{m}} \sum_{k=1}^m
\phi(Y^k| X_i) -\frac{1}{ \sqrt{m}} \sum_{k=1}^m \bar{\phi} (Y^k)$$ tends to zero in probability. We start with (\[asymp12\]) and calculate $$\nonumber
\grad\rem_n(0)
=\frac{1}{n}\sum_{i=1}^{n }\frac{\dfrac{1}{m}\sum\limits_{k=1}^{m}
\phi(Y^k |X_i) \, \Z (X_i,0)}
{C_m(X_i,0)}\;.$$ Therefore, by Cauchy-Schwarz inequality, expression (\[asymp12\]) is bounded by $$\label{asymp121}
\sqrt{\frac{1}{n} \sum_{i=1}^n \frac{[C_m(X_i,0) - C(X_i,0)]^2}{C^2_m(X_i,0)}}\:
\sqrt{\frac{1}{n} \sum_{i=1}^n \left| \frac{1}{ \sqrt{m}} \sum_{k=1}^m
\phi(Y^k| X_i) \right|^2}.$$ By assumptions 6(a) and 7 we again obtain that for arbitrary $\varepsilon>0, \eta >0$ and sufficiently large $m$ with probability at least $1-\eta$ for every $x \in \mX$ $$\nonumber
|\Z_m (x,0) -\Z(x,0)|\leq \varepsilon \quad {\rm and } \quad C_m(x,0)
\geq \alpha/2.$$
Therefore, the term under the first square root in (\[asymp121\]) tends in probability to zero, because with probability at least $1-\eta$ $$\nonumber
\frac{1}{n} \sum_{i=1}^n \frac{[C_m(X_i,0) - C(X_i,0)]^2}{C^2_m(X_i,0)} \leq \sup_{x \in \mX} \frac{[C_m(x,0) - C(x,0)]^2}{C^2_m(x,0)} \leq \frac{4 \varepsilon^2}{\alpha ^2} \:.$$ Using Markov’s inequality and assumption 3 the second square root is bounded in probability, since $$\begin{aligned}
\Ex_{X_i \sim g,Y^k \sim h} \frac{1}{n} \sum_{i=1}^n \left| \frac{1}{ \sqrt{m}} \sum_{k=1}^m
\phi(Y^k| X_i) \right|^2 &=& \Ex_{X \sim g ,Y^k \sim h} \left| \frac{1}{ \sqrt{m}} \sum_{k=1}^m
\phi(Y^k| X) \right|^2 \\
&=& \Ex_{X \sim g ,Y \sim h} \left|
\phi(Y| X) \right|^2 = \tilde W < \infty,\end{aligned}$$ where we use the fact that $\Ex_{Y \sim h} \phi (Y|x)=0$ for fixed $x.$ Now consider (\[asymp13\]). Change the order of summation and notice that $$\begin{aligned}
&&\Ex_{X_i \sim g ,Y^k \sim h} \left|\frac{1}{ \sqrt{m}} \sum_{k=1}^m \left[\frac{1}{n}\sum_{i=1}^n
\phi(Y^k| X_i) - \bar{\phi}(Y^k)\right]\right|^2 \\
&=& \Ex_{X_i \sim g,Y \sim h} \left|\frac{1}{n}\sum_{i=1}^n
\phi(Y| X_i) - \bar{\phi}(Y)\right|^2
=\frac{1}{n} \Ex_{X \sim g ,Y \sim h} \left|\phi(Y| X) - \bar{\phi}(Y)\right|^2,\end{aligned}$$ so (\[asymp13\]) tends to zero in $L^2$, hence, in probability.
Finally, we discuss assumptions in Theorem \[with\_cov\]. Note that conditions 1-3 are similar to their analogs in Theorem \[without\_cov\]. Therefore, we briefly comment on the others. Consider the exponential family $$f(y|x, \th)= \exp( \th\t W(y,x))$$ where $W(y,x) = \left( W_1(y,x), \ldots, W_p(y,x)\right),$ the set $ \mX$ is compact and the function $W(y,x)$ is continuous with respect to the variable $x$. For simplicity we restrict attention to finite (but very large) space $\mathcal{Y}$, so that $$\Z(x,\th) = \sum_{y \in \mathcal{Y}} \exp( \th\t W(y,x)).$$ The autologistic model [@HuWu1998] that is very popular in spatial statistics belongs to this family. We can calculate that $$\begin{aligned}
\grad \Z(x,\th) &=& \sum_{y \in \mathcal{Y}} \exp( \th\t W(y,x)) W(y,x)\\
\hess \Z(x,\th) &=& \sum_{y \in \mathcal{Y}} \exp( \th\t W(y,x)) W(y,x) W^T(y,x)\\
\hess \log p(y|x,\th)&=& -\hess \log \Z (x,\th)= -\frac{\hess \Z (x,\th)}{\Z(x,\th)}
+ \frac{\grad \Z (x,\th) \grad ^T \Z (x,\th)}{\Z ^2(x,\th)}\:.\end{aligned}$$ Since the function $W(y,x)$ is continuous with respect to $x$ functions $\Z(x,\th),$ $\grad \Z(x,\th),\hess \Z(x,\th)$ are continuous with respect to both variables on the compact set $\mathcal{X} \times U,$ therefore assumption 7 is satisfied. Besides, the function $\hess \log p(y|x,\th)$ is also continuous that implies condition 4. The uniform convergence in assumption 5 and 6 follows from [@Ferguson1996 Theorem 16(a)] or [@vaart1998 Example 19.8] if we again use compactness of sets $\mathcal{X}, U $ and continuity of considered functions.
[8]{}
Besag J. (1974). Spatial interaction and the statistical analysis of lattice systems. *J.R. Statist. Soc. B*, 36, 192–236.
Cappe O., Douc R., Moulines E. (2002). On the convergence of the Monte Carlo maximum likelihhod method for latent lariable models, *Scand. J. of Stat.*, 29, 615–635.
Dempster A.P., Laird N.M. and Rubin D.B. (1977). Likelihood from Incomplete Data via the EM Algorithm. *J.R. Statist. Soc. B*, 39, 1–38.
Ferguson T.S. (1996). *A course in large sample theory*, Chapman and Hall, London.
Fort G. and Moulines, E. (2003). Convergence of the Monte Carlo EM for curved exponential families. *Ann. Statist.* 31, 1033–1391.
Geyer C.J. and Thompson E.A. (1992). Constrained Monte Carlo maximum likelihood for dependent data. *J.R. Statist. Soc. B*, 54, 657–699.
Geyer C.J. (1994). On the convergence of Monte Carlo maximum likelihood calculations, *J. R. Statist. Soc. B*, 56, 261–274.
Geyer C. J., Sung Y.J. (2007). Monte Carlo likelihood inference for missing data models, *Ann. Statist.*, 35, 990–1011.
Huffer F.W., Wu H. (1998). Markov chain Monte Carlo for autologistic regression models with application to the distribution of plant species. *Biometrics*, 54, 509–524.
Levine R.A. and Casella G. (2001). Implementations of the Monte Carlo EM algorithm. *J. Computational and Graphical Statistics*, 10, 422–439.
Miasojedow, B., Niemiro, W., Palczewski, J. and Rejchel, W. (2014) Adaptive MCML. *Preprint*.
Møller B.J., Pettitt A.N., Reeves R. and Berthelsen, K.K. (2006). An efficient Markov chain Monte Carlo method for distributions with intractable normalising constants. *Biometrika*, 93, 451–458.
Niemiro, W. (1992). Asymptotics for M-estimators defined by convex minimization. *Ann. Statist.*, 20, 1514–1533.
Pollard D. (1984). *Convergence of stochastic processes*, Springer, New York.
van der Vaart A.W. (1998). *Asymptotic Statistics*. Cambridge University Press.
Wei G.C.G. and Tanner M.A. (1990). A Monte Carlo imlementation of the EM algorithm and the poor man’s data augumentation algorithms. *J. Am. Statist. Assoc.* 85, 699–704.
Wu, H. and Huffer, F. W. (1997). Modeling the distribution of plant species using the autologistic regression model. *Environmental and Ecological Statistics* 4, 49–64.
Zalewska M., Niemiro W. and Samoliński B. (2010). MCMC imputation in autologistic model. *Monte Carlo Methods Appl.* 16, 421–438.
[^1]: Institute of Applied Mathematics and Mechanics, University of Warsaw, Banacha 2, 02-097 Warsaw, Poland, [[email protected]]{}
[^2]: Faculty of Mathematics and Computer Science, Nicolaus Copernicus University, Chopina 12/18, 87-100, Toruń, Poland [[email protected]]{} and Institute of Applied Mathematics and Mechanics, University of Warsaw, Banacha 2, 02-097 Warsaw, Poland, [[email protected]]{}
[^3]: School of Mathematics, University of Leeds, Woodhouse Lane, Leeds LS2 9JT , UK, [[email protected]]{}
[^4]: corresponding author, Faculty of Mathematics and Computer Science, Nicolaus Copernicus University, Chopina 12/18, 87-100, Toruń, Poland, +48 566112943, [[email protected]]{}
[^5]: Work partially supported by Polish National Science Center No. N N201 608 740
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In the framework of the theory of open systems based on completely positive quantum dynamical semigroups, we give a description of continuous variable quantum entanglement and quantum discord for a system consisting of two non-interacting non-resonant bosonic modes embedded in a thermal environment. We study the time evolution of logarithmic negativity, which characterizes the degree of entanglement, and show that in the case of an entangled initial squeezed thermal state, entanglement suppression takes place for all temperatures of the environment, including zero temperature. We analyze the time evolution of the Gaussian quantum discord, which is a measure of all quantum correlations in the bipartite state, including entanglement, and show that discord decays asymptotically in time under the effect of the thermal bath. We describe also the time evolution of classical correlations and quantum mutual information, which measures the total correlations of the quantum system.'
address: 'National Institute of Physics and Nuclear Engineering, P.O.Box MG-6, Bucharest-Magurele, Romania'
author:
- Aurelian Isar
title: 'Quantum Correlations of Two-Mode Gaussian Systems in a Thermal Environment'
---
Introduction
============
In recent years there is an increasing interest in using non-classical entangled states of continuous variable systems in applications of quantum information processing, communication and computation [@bra1]. In this respect, Gaussian states, in particular two-mode Gaussian states, play a key role since they can be easily created and controlled experimentally. Due to the unavoidable interaction with the environment, in order to describe realistically quantum information processes it is necessary to take decoherence and dissipation into consideration. Decoherence and dynamics of quantum entanglement in continuous variable open systems have been intensively studied in the last years [@oli; @ser3; @pra; @dod1; @ser4; @avd; @ben1; @mch; @man; @jan; @aphysa; @aeur; @arus1; @paz2; @arus2; @ascri1].
In this paper we study, in the framework of the theory of open systems based on completely positive quantum dynamical semigroups, the dynamics of continuous variable quantum entanglement and quantum discord of a subsystem consisting of two uncoupled bosonic modes (harmonic oscillators) interacting with a common thermal environment. We are interested in discussing the correlation effect of the environment, therefore we assume that the two modes are independent, i.e. they do not interact directly. The initial state of the subsystem is taken of Gaussian form and the evolution under the quantum dynamical semigroup assures the preservation in time of the Gaussian form of the state. In Refs. [@ascri4; @aosid1; @ascri2] we studied the evolution of entanglement and quantum discord of two identical harmonic oscillators interacting with a thermal environment for initial symmetric squeezed vacuum and squeezed thermal states of the subsystem. In this work we extend the previous analysis to the case of non-resonant bosonic modes and take non-symmetric squeezed thermal states as initial states. We show that in the case of an entangled initial squeezed thermal state, entanglement suppression (entanglement sudden death) takes place for all temperatures of the environment, including zero temperature. We analyze the time evolution of Gaussian quantum discord, which is a measure of all quantum correlations in the bipartite state, including entanglement, and show that discord decays asymptotically in time under the effect of the thermal bath. Before the suppression of the entanglement, the qualitative evolution of discord is very similar to that of the entanglement. We describe also the time evolution of classical correlations and quantum mutual information, which measures the total correlations of quantum system.
Equations of motion for two modes interacting with an environment
=================================================================
We study the dynamics of a subsystem composed of two non-interacting bosonic modes in weak interaction with a thermal environment. In the axiomatic formalism based on completely positive quantum dynamical semigroups, the Markovian irreversible time evolution of an open system is described by the Kossakowski-Lindblad master equation [@rev; @san]. We are interested in the set of Gaussian states, therefore we introduce such quantum dynamical semigroups that preserve this set during time evolution of the system. The Hamiltonian of the two uncoupled non-resonant harmonic oscillators of identical mass $m$ and frequencies $\omega_1$ and $\omega_2$ is $$\begin{aligned}
H={1\over 2m}(p_x^2+p_y^2)+\frac{m}{2}(\omega_1^2 x^2+\omega_2^2 y^2),\end{aligned}$$ where $x,y$ are the coordinates and $p_x,p_y$ are the momenta of the two quantum oscillators.
The equations of motion for the quantum correlations of the canonical observables $x,y$ and $p_x,p_y$ are the following ($\rm T$ denotes the transposed matrix) [@san]: $$\begin{aligned}
{d \sigma(t)\over
dt} = Y \sigma(t) + \sigma(t) Y^{\rm T}+2 D,\label{vareq}\end{aligned}$$ where $$Y=\left(\matrix{ -\lambda&1/m&0 &0\cr -m\omega_1^2&-\lambda&0&
0\cr 0&0&-\lambda&1/m \cr 0&0&-m\omega_2^2&-\lambda}\right),~~
D=\left(\matrix{
D_{xx}& D_{xp_x} &D_{xy}& D_{xp_y} \cr D_{xp_x}&D_{p_x p_x}&
D_{yp_x}&D_{p_x p_y} \cr D_{xy}& D_{y p_x}&D_{yy}& D_{y p_y}
\cr D_{xp_y} &D_{p_x p_y}& D_{yp_y} &D_{p_y p_y}} \right),$$ and the diffusion coefficients $D_{xx}, D_{xp_x},$... and the dissipation constant $\lambda$ are real quantities. We introduced the following $4\times 4$ bimodal covariance matrix: $$\begin{aligned}
\sigma(t)=\left(\matrix{\sigma_{xx}(t)&\sigma_{xp_x}(t) &\sigma_{xy}(t)&
\sigma_{xp_y}(t)\cr \sigma_{xp_x}(t)&\sigma_{p_xp_x}(t)&\sigma_{yp_x}(t)
&\sigma_{p_xp_y}(t)\cr \sigma_{xy}(t)&\sigma_{yp_x}(t)&\sigma_{yy}(t)
&\sigma_{yp_y}(t)\cr \sigma_{xp_y}(t)&\sigma_{p_xp_y}(t)&\sigma_{yp_y}(t)
&\sigma_{p_yp_y}(t)}\right)=\left(\begin{array}{cc}A&C\\
C^{\rm T}&B \end{array}\right),\label{covar}\end{aligned}$$ where $A$, $B$ and $C$ are $2\times 2$ Hermitian matrices. $A$ and $B$ denote the symmetric covariance matrices for the individual reduced one-mode states, while the matrix $C$ contains the cross-correlations between modes. The elements of the covariance matrix are defined as $\sigma_{ij}=<R_iR_j+R_jR_i>/2, i,j=1,..,4,$ with ${\bf R}=\{x,p_x,y,p_y\},$ which up to local displacements fully characterize any Gaussian state of a bipartite system.
The time-dependent solution of Eq. (\[vareq\]) is given by [@san] $$\begin{aligned}
\sigma(t)= M(t)[\sigma(0)-\sigma(\infty)] M^{\rm T}(t)+\sigma(\infty),\label{covart}\end{aligned}$$ where the matrix $M(t)=\exp(Yt)$ has to fulfill the condition $\lim_{t\to\infty} M(t) = 0.$ The values at infinity are obtained from the equation $$\begin{aligned}
Y\sigma(\infty)+\sigma(\infty) Y^{\rm T}=-2 D.\label{covarinf}\end{aligned}$$
Dynamics of quantum correlations
================================
Time evolution of entanglement
------------------------------
In order to quantify the degree of entanglement of the two-mode states it is appropriate to use the logarithmic negativity. For a Gaussian density operator, the logarithmic negativity is completely defined by the symplectic spectrum of the partial transpose of the covariance matrix. It is given by $E_N={\rm max}\{0,-\log_2 2\tilde\nu_-\},$ where $\tilde\nu_-$ is the smallest of the two symplectic eigenvalues of the partial transpose $\tilde{{\sigma}}$ of the two-mode covariance matrix $\sigma$ [@ser4]: $$\begin{aligned}
2\tilde{\nu}_{\mp}^2 = \tilde{\Delta}\mp\sqrt{\tilde{\Delta}^2
-4\det\sigma}\end{aligned}$$ and $ \tilde\Delta$ is the symplectic invariant (seralian), given by $ \tilde\Delta=\det A+\det B-2\det C.$
In our model, the logarithmic negativity is calculated as [@aijqi; @aosid] $$\begin{aligned}
E_N(t)={\rm max}\{0,-\frac{1}{2}\log_2[4g(\sigma(t))]\}, \end{aligned}$$ where $$\begin{aligned}
g(\sigma(t))=\frac{1}{2}(\det A +\det
B)-\det C\nonumber\\
-\left({\left[\frac{1}{2}(\det A+\det B)-\det
C\right]^2-\det\sigma(t)}\right)^{1/2}.\end{aligned}$$ It determines the strength of entanglement for $E_N(t)>0,$ and if $E_N(t)=0,$ then the state is separable.
We assume that the initial Gaussian state is a two-mode squeezed thermal state, with the covariance matrix of the form [@mar] $$\begin{aligned}
\sigma_{st}(0)=\frac{1}{2}\left(\matrix{a&0&c&0\cr
0&a&0&-c\cr
c&0&b&0\cr
0&-c&0&b}\right),\label{ini1} \end{aligned}$$ with the matrix elements given by $$\begin{aligned}
a=n_1 \cosh^2 r + n_2 \sinh^2 r + \frac{1}{2} \cosh 2r,\\
b=n_1 \sinh^2 r + n_2 \cosh^2 r + \frac{1}{2} \cosh 2r,\\
c=\frac{1}{2}(n_1 + n_2 + 1) \sinh 2r,\label{ini2}\end{aligned}$$ where $n_1,n_2$ are the average number of thermal photons associated with the two modes and $r$ denotes the squeezing parameter. In the particular case $n_1=0$ and $n_2=0$, (\[ini1\]) becomes the covariance matrix of the two-mode squeezed vacuum state. A two-mode squeezed thermal state is entangled when the squeezing parameter $r$ satisfies the inequality $r>r_s$ [@mar], where $$\begin{aligned}
\cosh^2 r_s=\frac{(n_1+1)(n_2+1)}{ n_1+n_2+1}. \end{aligned}$$
We suppose that the asymptotic state of the considered open system is a Gibbs state corresponding to two independent bosonic modes in thermal equilibrium at temperature $T.$ Then the quantum diffusion coefficients have the following form (we put from now on $\hbar=1$) [@rev]: $$\begin{aligned}
m\omega_1 D_{xx}=\frac{D_{p_xp_x}}{m\omega_1}=\frac{\lambda}{2}\coth\frac{\omega_1}{2kT},\nonumber\\
m\omega_2 D_{yy}=\frac{D_{p_yp_y}}{m\omega_2}=\frac{\lambda}{2}\coth\frac{\omega_2}{2kT},\label{envcoe}\\
D_{xp_x}=D_{yp_y}=D_{xy}=D_{p_xp_y}=D_{xp_y}=D_{yp_x}=0.\nonumber\end{aligned}$$
The evolution of entangled initial squeezed thermal states with the covariance matrix given by Eq. (\[ini1\]) is illustrated in Fig. 1, where we represent the dependence of the logarithmic negativity $E_N(t)$ on time $t$ and temperature $T$ for the case of an initial non-symmetric Gaussian state ($a\neq b$). For all temperatures $T,$ including zero temperature, at certain finite moment of time, which depends on $T,$ $E_N(t)$ becomes zero and therefore the state becomes separable. This is the so-called phenomenon of entanglement sudden death. It is in contrast to the quantum decoherence, during which the loss of quantum coherence is usually gradual [@aphysa; @arus]. One can also show that the dissipation favors the phenomenon of entanglement sudden death – with increasing the dissipation parameter $\lambda,$ the entanglement suppression happens earlier. The same qualitative behaviour of the time evolution of entanglement was obtained previously [@aosid1; @ascri3] in the particular case $n_1=0$ and $n_2=0$ corresponding to an initial two-mode squeezed vacuum state and in the case of symmetric initial squeezed thermal states.
Comparing the present results with those obtained in the previous paper [@ascri2] one can assert that the asymmetry ($a\neq b$) of the initial Gaussian state favors the suppression of entanglement. The most robust under the influence of the environment is the entanglement of symmetric ($a=b$) initial squeezed thermal states. An even stronger influence on the entanglement has the non-resonant character of the two modes: by increasing the ratio of the frequencies of the two modes, the entanglement sudden death happens earlier in time. The longest surviving entanglement takes place when the modes are resonant ($\omega_1=\omega_2$). This effect due to the non-resonance of the modes is stronger for small values of the frequencies, and it diminishes, for the same ratio of frequencies, by increasing the values of frequencies.
In our model, in which we suppose that the asymptotic state of the considered open system is a Gibbs state corresponding to two independent bosonic modes in thermal equilibrium, an separable initial state remains separable in time, and it is not possible to generate entanglement. This is in contrast with the possibility of entanglement generation starting, for instance, with a separable state in the case of two non-interacting two-level systems immersed in a common bath [@ben2]. At the same time we remind that we have studied previously [@aosid; @ascri] the evolution of the entanglement of two identical harmonic oscillators interacting with a general environment, characterized by general diffusion and dissipation coefficients, and we obtained that, for separable initial states and for definite values of these coefficients, entanglement generation or a periodic generation and collapse of entanglement take place. In discussing the entanglement decay, it is interesting to mention that models have been elaborated to realize quantum feedback control of continuous variable entanglement for a system consisting of two interacting bosonic modes plunged into an environment, based on a local technique [@man1], or on a nonlocal homodyne measurement [@man2].
Gaussian quantum discord
------------------------
Recent studies have shown that separable states, usually considered as being classically correlated, might also contain quantum correlations. Quantum discord was introduced [@zur; @oll] as a measure of all quantum correlations in a bipartite state, including – but not restricted to – entanglement. Quantum discord has been defined as the difference between two quantum analogues of classically equivalent expressions of the mutual information, which is a measure of total correlations in a quantum state. For an arbitrary bipartite state $\rho_{12},$ the total correlations are expressed by quantum mutual information $$\begin{aligned}
I(\rho_{12})=\sum_{i=1,2} S(\rho_{i})-S(\rho_{12}),\end{aligned}$$ where $\rho_i$ represents the reduced density matrix of subsystem $i$ and $S(\rho)= - {\rm Tr}(\rho \ln \rho)$ is the von Neumann entropy. A measure of bipartite classical correlations $C(\rho_{12})$ in the bipartite quantum state $\rho_{12}$ based on a complete set of local projectors $\{\Pi_{2}^k\}$ on the subsystem 2 can be given by $$\begin{aligned}
C(\rho_{12})=S(\rho_{1})-{\inf}_{\{\Pi_{2}^k\}}\{S(\rho_{1|2})\},\end{aligned}$$ where $S(\rho_{1|2}) =\sum_{k}p^k S(\rho_{1}^k)$ is the conditional entropy of subsystem [1]{} and $\inf\{S(\rho_{1|2})\}$ represents the minimal value of the entropy with respect to a complete set of local measurements $\{\Pi_{2}^k\}.$ Here, $p^k$ is the measurement probability for the $k$th local projector and $\rho_{1}^k$ denotes the reduced state of subsystem $1$ after the local measurements. Then the quantum discord is defined by $$\begin{aligned}
D(\rho_{12})=I(\rho_{12})-C(\rho_{12}).\end{aligned}$$
Originally the quantum discord was defined and evaluated mainly for finite dimensional systems. Recently [@par; @ade] the notion of discord has been extended to the domain of continuous variable systems, in particular to the analysis of bipartite systems described by two-mode Gaussian states. Closed formulas have been derived for bipartite thermal squeezed states [@par] and for all two-mode Gaussian states [@ade].
The Gaussian quantum discord of a general two-mode Gaussian state $\rho_{12}$ can be defined as the quantum discord where the conditional entropy is restricted to generalized Gaussian positive operator valued measurements (POVM) on the mode 2, and in terms of symplectic invariants it is given by (the symmetry between the two modes 1 and 2 is broken) [@ade] $$\begin{aligned}
D=f(\sqrt{\beta})-f(\nu_-) - f(\nu_+) + f(\sqrt{\varepsilon}),
\label{disc}\end{aligned}$$ where $$\begin{aligned}
f(x) =\frac{x+1}{2} \log\frac{x+1}{2} -\frac{x-1}{2} \log\frac{x-1}{2},\end{aligned}$$ $$\begin{aligned}
\label{infdet}
\varepsilon=
& &
\hspace*{-.1cm}
\left\{ \hspace*{-.5cm} \begin{array}{rcl}& &\begin{array}{c}\displaystyle{\frac{{2 \gamma^2+(\beta-1)(\delta-\alpha)
+2 |\gamma| \sqrt{\gamma^2+(\beta-1) (\delta-\alpha)}}}{{(\beta-1){}^2}}}\end{array},\\& &\qquad
\hbox{if}~~(\delta-\alpha\beta)^2 \le (\beta+1)\gamma^2 (\alpha +\delta)\\ \\& &
\begin{array}{c}\displaystyle{\frac{{\alpha\beta-\gamma^2+\delta-\sqrt{\gamma^4+(\delta-\alpha\beta){}^2-
2\gamma^2(\delta+\alpha\beta)}}}{{2\beta}}}\end{array}, \\& & \qquad
\hbox{otherwise,} \end{array} \right.
$$ $$\begin{aligned}
\alpha=4\det A,~~~\beta=4\det B,~~~\gamma=4\det C,~~~\delta=16\det\sigma,\end{aligned}$$ and $\nu_\mp$ are the symplectic eigenvalues of the state, given by $$\begin{aligned}
2{\nu}_{\mp}^2 ={\Delta}\mp\sqrt{{\Delta}^2
-4\det\sigma},\end{aligned}$$ where $\Delta=\det A+\det B+2\det C.$
The evolution of the Gaussian quantum discord $D$ is illustrated in Figure 2, where we represent the dependence of $D$ on time $t$ and temperature $T$ for an entangled initial non-symmetric Gaussian state, taken of the form of a two-mode squeezed thermal state (\[ini1\]), for such values of the parameters which satisfy for all times the first condition in formula (\[infdet\]). The Gaussian discord has nonzero values for all finite times and this fact certifies the existence of non-classical correlations in two-mode Gaussian states, either separable or entangled. Gaussian discord asymptotically decreases in time, compared to the case of logarithmic negativity, which has an evolution leading to a sudden suppression of entanglement. For entangled initial states the Gaussian discord remains strictly positive in time and in the limit of infinite time it tends asymptotically to zero, corresponding to the thermal product (separable) state, with no correlations at all.
From Fig. 2 we notice that, in agreement with the general properties of the Gaussian quantum discord [@ade], the states can be either separable or entangled for $D\le 1$ and all the states above the threshold $D=1$ are entangled. We also notice that the decay of quantum discord is stronger when the temperature $T$ is increasing. It should be remarked that the decay of quantum discord is very similar to that of the entanglement before the time of the sudden death of entanglement. Near the threshold of zero logarithmic negativity ($E_N = 0$), the nonzero values of the discord can quantify the non-classical correlations for separable mixed states and one considers that this fact could make possible some tasks in quantum computation [@yut]. The discord is increasing with the squeezing parameter $r$ and it is decreasing with increasing the ratio of the frequencies $\omega_1$ and $\omega_2$ of the two modes and the difference of parameters $a$ and $b.$
Classical corellations and quantum mutual information
-----------------------------------------------------
The measure of classical correlations for a general two-mode Gaussian state $\rho_{12}$ can also be calculated and it is given by [@ade] $$\begin{aligned}
C=f(\sqrt{\alpha}) - f(\sqrt{\varepsilon}),
\label{clas}\end{aligned}$$ while the expression of the quantum mutual information, which measures the total correlations, is given by $$\begin{aligned}
I=f(\sqrt{\alpha}) + f(\sqrt{\beta}) -f(\nu_-) - f(\nu_+).
\label{mut}\end{aligned}$$
In Figs. 3 and 4 we illustrate the evolution of classical correlations $C$ and, respectively, quantum mutual information $I,$ as functions of time $t$ and temperature $T$ for an entangled initial Gaussian state, taken of the form of a two-mode squeezed thermal state (\[ini1\]). These two quantities manifest a qualitative behaviour similar to that one of the Gaussian discord: they have nonzero values for all finite times and in the limit of infinite time they tend asymptotically to zero, corresponding to the thermal product (separable) state, with no correlations at all. One can also see that the classical correlations and quantum mutual information decrease with increasing the temperature of the thermal bath. One can show that the classical correlations and quantum mutual information increase with increasing the squeezing parameter $r$ and the difference of parameters $a$ and $b.$ At the same time classical correlations increase with the ratio of the frequencies $\omega_1$ and $\omega_2$ of the two modes, while quantum mutual information is decreasing with increasing this ratio.
For comparison all these quantities are represented also on the same graphic in Fig. 4. In the considered case the value of classical correlations is larger than that of quantum correlations, represented by the Gaussian quantum discord.
Summary
=======
We investigated the Markovian dynamics of quantum correlations for a subsystem composed of two non-interacting bosonic modes embedded in a thermal bath. We have analyzed the influence of the environment on the dynamics of quantum entanglement and quantum discord for Gaussian initial states. We have described the time evolution of the logarithmic negativity in terms of the covariance matrix for non-symmetric squeezed thermal states, for the case when the asymptotic state of the considered open system is a Gibbs state corresponding to two independent quantum harmonic oscillators in thermal equilibrium. The dynamics of the quantum entanglement strongly depends on the initial states and the parameters characterizing the environment (dissipation coefficient and temperature). For an entangled initial squeezed thermal state, entanglement suppression (entanglement sudden death) takes place for all values of the temperatures of the environment, including zero temperature. The time when the entanglement is suppressed decreases with increasing the temperature and dissipation.
We described also the time evolution of Gaussian quantum discord, which is a measure of all quantum correlations in the bipartite state, including entanglement. The values of quantum discord decrease asymptotically in time. This is in contrast to the sudden death of entanglement. The time evolution of quantum discord is very similar to that of entanglement before the sudden suppression of the entanglement. Quantum discord is decreasing with increasing the temperature. After the sudden death of entanglement the nonzero values of discord manifest the existence of quantum correlations for separable mixed states. We described also the time evolution of classical correlations and quantum mutual information, which measures the total correlations of the quantum system.
The author thanks the referee for his useful suggestions and recommendations. The author acknowledges the financial support received from the Romanian Ministry of Education and Research, through the Projects CNCS-UEFISCDI PN-II-ID-PCE-2011-3-0083 and PN 09 37 01 02/2010.
References {#references .unnumbered}
==========
[99]{}
Braunstein S L and van Loock P 2005 [*Rev. Mod. Phys.*]{} **77** 513
Olivares S, Paris M G A and Rossi A R 2003 [*Phys. Lett.*]{} A **319** 32
Serafini A, Illuminati F, Paris M G A and de Siena S 2004 [*Phys. Rev. A*]{} [**69**]{} 022318
Prauzner-Bechcicki J S 2004 [*J. Phys. A: Math. Gen.*]{} **37** L173
Dodd P J and Halliwell J J 2004 [*Phys. Rev. A*]{} **69** 052105
Adesso G, Serafini A and Illuminati F 2004 [*Phys. Rev. A*]{} [**70**]{} 022318
Dodonov A V, Dodonov V V and Mizrahi S S 2005 [*J. Phys. A: Math. Gen.*]{} **38** 683
Benatti F and Floreanini R 2006 [*J. Phys. A: Math. Gen.*]{} **39** 2689
McHugh D, Ziman M and Buzek V 2006 [*Phys. Rev. A*]{} **74** 042303
Maniscalco S, Olivares S and Paris M G A 2007 [*Phys. Rev. A*]{} **75** 062119
An J H and Zhang W M 2007 [*Phys. Rev. A*]{} **76** 042127
Isar A and Scheid W 2007 [*Physica*]{} A **373** 298
Isar A 2008 [*Eur. J. Phys. Special Topics*]{} **160** 225
Isar A 2009 [*J. Russ. Laser Res.*]{} **30** 458
Paz J P and Roncaglia A J 2009 [*Phys. Rev. A*]{} **79** 032102
Isar A 2010 [*J. Russ. Laser Res.*]{} **31** 182
Isar A 2010 [*Phys. Scr.*]{} [**82**]{} 038116
Isar A 2010 [*Phys. Scr., Topical Issue*]{} **140** 014023
Isar A 2011 [*Open Sys. Inf. Dynamics*]{} **18** 175
Isar A 2012 [*Phys. Scr., Topical Issue*]{} **147** 014015
Isar A, Sandulescu A, Scutaru H, Stefanescu E and Scheid W 1994 [*Int. J. Mod. Phys.*]{} E **3** 635
Sandulescu A, Scutaru H and Scheid W 1987 [*J. Phys. A: Math. Gen.*]{} **20** 2121
Isar A 2008 [*Int. J. Quantum Inf.*]{} **6** 689
Isar A 2009 [*Open Sys. Inf. Dynamics*]{} **16** 205
Marian P, Marian TA and Scutaru H 2003 [*Phys. Rev. A*]{} **68** 062309
Isar A 2007 [*J. Russ. Laser Res.*]{} **28** 439
Isar A 2011 [*Phys. Scr., Topical Issue*]{} **143** 014012
Benatti F, Floreanini R and Piani M 2003 [*Phys. Rev. Lett.*]{} **91** 070402
Isar A 2009 [*Phys. Scr., Topical Issue*]{} **135** 014033
Mancini S 2006 [*Phys. Rev. A*]{} **73** 010304
Mancini S and Wiseman H M 2007 [*Phys. Rev. A*]{} **75** 012330
Zurek W H 2000 [*Annalen der Physik*]{} (Leipzig) [**9**]{} 853
Ollivier H and Zurek W H 2001 [*Phys. Rev. Lett.*]{} [**88**]{} 017901
Giorda P and Paris M G A 2010 [*Phys. Rev. Lett.*]{} [**105**]{} 020503
Adesso G and Datta A 2010 [*Phys. Rev. Lett.*]{} [**105**]{} 030501
Yu T and Eberly J H 2004 [*Phys. Rev. Lett.*]{} [**93**]{} 140404
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'After reviewing the problematic behavior of some previously suggested finite interval spatial operators of the symmetric Riesz type, we create a wish list leading toward a new spatial operator suitable to use in the space-time fractional differential equation of anomalous diffusion when the transport of material is strictly restricted to a bounded domain. Based on recent studies of wall effects, we introduce a new definition of the spatial operator and illustrate its favorable characteristics. We provide two numerical methods to solve the modified space-time fractional differential equation and show particular results illustrating compliance to our established list of requirements, most important to the conservation principle and the second law of thermodynamics.'
address:
- 'Texas A&M University'
- Department of Petroleum Engineering
- Department of Physics
author:
- 'P. P. Valkó'
- 'X. H. Zhang'
title: Finite Domain Anomalous Spreading Consistent with First and Second Law
---
Space-time fractional differential equation ,Caputo derivative ,Riesz derivative ,Laplace transform ,collocation ,finite differences 05.40.-a ,05.60.k
Introduction {#sec1}
============
In practical sense anomalous diffusion can be detected by heavy tails of the resulting density distribution (at a given time) and by the departure from linearly evolving mean-square displacement for an initially concentrated plume [@Schneider89; @Shlesinger93; @Mainardi97]. Such a behavior is well documented for instance for the spreading of contaminants in heterogeneous porous media, where shortcut pathways may be present between two points in space (causing a departure from Fick’s law) and/or particles can be trapped, hindered at various locations (causing memory effects)[@Berkowitz95; @Benson00a; @Benson00b; @Metzler00; @Metzler04]. With a common term, the behavior may be non-local both in space and time [@Zaslavsky02; @Chechkin06; @Zhang09].
Continuous time random walks (CTRW) serve as a small-scale conceptual model for describing anomalous diffusion. Of practical interest is the evolution of the density of a cloud of walkers on the macroscopic scale that is ultimately determined by the statistical characteristics of the jump lengths and waiting times on the microscopic level. The so called Lévy flights are dominated by rare but large jumps and can reproduce the power-law tails of the spatial distributions at a given time. Allowing rare but long waiting times can also lead to marked departure from the scaling law of diffusion. The flux computed on the scale of particle motion, depends on the parameters of the random walk and on the density at the considered time (in the Markovian case) or on the complete density history (in the case of memory effects included). The passage from microscopic to macroscopic scale is performed by letting the characteristic length and time of the particle motion tend to zero [@Montroll65; @Cushman91; @Barkai02]. On unbounded domain, the resulting macroscopic behavior is conveniently described by the space-time fractional diffusion equation [@Gorenflo02]. The passage also results in the generalization of Fick’s law [@Paradisi01; @Neel07].
Brief summary of known results for the unbounded domain case
------------------------------------------------------------
The one-dimensional space-time fractional diffusion equation is written as $$\begin{aligned}
\label{equ1}
\frac{\partial ^\beta}{\partial t^\beta} u(x,t) = a \frac{\partial ^\alpha}{\partial |x|^\alpha} u(x,t)\;\;\; - \infty < x < \infty, \; t > 0\end{aligned}$$ where $a$ is positive constant with dimension $[L^\alpha / T^{\beta}]$. (The dimension of $u$ is $[1/L]$ because it is understood as one dimensional density of a countable quantity.)
The fractional time derivative is taken in the Caputo sense [@Caputo67]. The notation for the operator $\frac{\partial ^\alpha}{\partial |x|^\alpha} $ was introduced by Saichev and Zaslavsky [@Saichev97]. It is understood as the application of the (symmetric) Riesz derivative $\frac{d ^\alpha}{d |x|^\alpha} $ operator with respect to the space variable $x$. The Riesz derivative is defined through the Liouville-Weyl fractional derivatives: $$\begin{aligned}
\frac{d ^\alpha}{d |x|^\alpha} f(x) &=&
\begin{Cases}
-\frac{1}{2 \cos(\pi \alpha /2 )} [ D^\alpha_+ f + D^\alpha_- f ],\;\;\; 0 < \alpha \leq 2,\;\alpha \neq 1\\
-\frac {d}{dx}\textbf{H} (f ; x),\;\;\;\alpha=1\\
- f,\;\;\; \alpha=0
\end{Cases}\end{aligned}$$ where $D^\alpha_\pm$ are called the left- and right Liouville-Weyl derivatives: $$\begin{aligned}
\label{LW1}
D^\alpha_+ &=& \frac{1}{\Gamma(m-\alpha)} \frac{d^m}{dx^m} \int_{-\infty}^{x} \frac{f(\xi)d\xi}{(x-\xi)^{\alpha-m+1}},\;\;\; m = \lceil \alpha \rceil\\
\label{LW2}
D^\alpha_- &=& \frac{(-1)^m}{\Gamma(m-\alpha)} \frac{d^m}{dx^m} \int_{x}^{\infty} \frac{f(\xi)d\xi} {(\xi-x)^{\alpha-m+1}}.\end{aligned}$$ and **H** denotes Hilbert transform. (See a more detailed discussion, for instance, by Chechkin *at al.* [@Chechkin08] )
For the unbounded case the following Cauchy problem can be stated: solve (\[equ1\]) for a given parameter set $\{ 0<\alpha\leq2,\; 0<\beta\leq1,\; a > 0 \}$ augmented with the initial condition $u(x,0)=f_i(x)$, where $f_i$ is a probability density function. The solution of the Cauchy problem can be obtained by the Laplace-Fourier approach, probably first used in this context by Montroll and Weiss [@Montrol73]. The transforms serve double purpose: they provide a better understanding of the operators involved and also lead to the solution for particular cases. In fact, the $\beta$-order Caputo derivative ($0<\beta\leq 1$) is the generalization of the *first derivative* via Laplace transform: $$\begin{aligned}
\textbf{L}\left( \frac{d ^\beta f }{d t^\beta} ; s \right) = s^{\beta-1} \left[ s \textbf{L}(f; s) -f(0)\right]\end{aligned}$$ and the $\alpha$ order Riesz derivative ($0\leq\alpha\leq 2$) is the generalization of the *second derivative* via Fourier transform: $$\begin{aligned}
\textbf{F}\left( \frac{d ^\alpha f }{d |x|^\alpha} ;\omega \right) = -|\omega|^\alpha \textbf{F}(f ;\omega)\end{aligned}$$
The fundamental solution (spreading of an initial Dirac delta) can be obtained by the Laplace-Fourier method and can be given in terms of well investigated special functions. Though various representations are available, their equivalence has been well established [@Mainardi01; @Kilbas06] . The remarkable scaling property of the fundamental solution can be stated as: $$\begin{aligned}
\label{fundsol}
u(x,t) = (a t)^{-\frac{\beta}{\alpha}}\textit{M}\left( \frac{|x|}{(a t)^\frac{\beta}{\alpha} } ; \frac{\beta}{\alpha} \right)\end{aligned}$$ where $\textit{M}$ denotes the Mainardi (or $\textit{M}$) -function given by $$\begin{aligned}
\textit{M}(\xi; \mu)= \frac{1}{2} \sum_{n=0}^\infty \frac{\left(-\xi\right)^n}{n! \; \Gamma\left[-\mu n +(1-\mu)\right]}\end{aligned}$$ (By including the factor $1/2$ we ensure that the integral of (\[fundsol\]) over the $x$-axis is unity.) While computability of the $\textit{M}$-function is far from trivial, it has been basically resolved. For instance, *Mathematica* can calculate it *with any desired accuracy* from its definition above, provided the $\mu$ parameter is passed as a rational fraction. Some known special cases, see e.g. [@Piryatinska05] can be reproduced symbolically with $Mathematica$:
$$\begin{aligned}
\textit{M}(\xi ; 1/2) &=& \frac{1}{2 \sqrt{\pi}} \exp(-\xi^2 /4)\\
\label{M13}
\textit{M}(\xi ; 1/3) &=& \frac{3^{2/3}}{2} Ai(\frac{\xi}{\sqrt[3]{3}}) \\
\textit{M}(\xi ; 2/3) &=& \frac{1}{2 \; 3^{2/3}} \exp(-\frac{2 \xi^3}{27}) \left[
\sqrt[3]{3} \; \xi \; Ai(\frac{ \xi^2}{3 \; 3^{1/3}}) - 3 Ai'(\frac{ \xi^2}{3 \; 3^{1/3}})\right]\end{aligned}$$
where $Ai$ stands for the Airy function and and $Ai'$ for the Airy-prime function.
Anomalous spreading on a finite interval
----------------------------------------
For Fickian diffusion the finite domain model consistent with the first law (conservation principle) is obtained by requiring zero flux at the two endpoints of the considered interval. This translates to the well-known homogenous Neumann boundary conditions for the traditional $\{\beta=1, \; \alpha=2\}$ diffusion equation. Mapping the same physical requirement to mathematically treatable objects for the space-time fractional partial differential equation has turned out to be extremely challenging.
Numerical experimentation followed two complementary approaches: Monte-Carlo (Langevin) simulation of random walk and finite difference approximation to the solution of a space-time fractional differential equation on bounded domain. The first is easier to conduct and always results in physically meaningful results (including conservation, if reflective walls are applied), but those results are difficult to use in a practical sense. The second approach has grown mature during the years. Here we will rely on the well documented “matrix approach” suite by Podlubny *et al.* [@Podlubny09] that provides a general framework to the numerical solution of partial fractional differential equations.
However, the problem is deeper than purely finding an adequate numerical method. In the presence of wall(s) of various properties the spatial operator itself needs to be modified, see e.g. [@Chechkin03; @Metzler07].
In this work we are looking for the solution of the Cauchy problem for the fractional partial differential equation $$\begin{aligned}
\label{problem}
\frac{\partial ^\beta}{\partial t^\beta} u(x,t) = a \frac{\partial ^\alpha}{\partial_{mod} |x|^\alpha} u(x,t)\;,\;\;\;\; 0 \leq x \leq 1 , \;\;\;\; t > 0\end{aligned}$$ when $a > 0$ and the initial condition $u(x,0) = f_i(x), \;\; 0 \leq x \leq 1 $ is a probability density function. The fractional time derivative of order ($0<\beta\leq1$) is in the Caputo sense. We added a subscript “mod” to the spatial operator of order ($0<\alpha\leq2$), because the definitions (\[LW1\]-\[LW2\]) require the extension of the $u(x,t)$ function defined on the interval $[0,1]$ to a left function $f_l(x)$ defined on $(-\infty,1]$ (or, strictly speaking, at least on $(-\infty,x]$,) and to a right function $f_r(x)$ defined on $[0,\infty)$. Notice that these auxiliary functions need not be probability distributions. For brevity, we will call the set of choices we make in creating these extensions a “prescription”. Various prescriptions will give rise to various finite domain Riesz operators and will ultimately define the characteristics of the solution of (\[problem\]).
In the following section (\[sec2\]) we illustrate the problematic behavior of some previously suggested prescriptions and create a “wish list” leading toward a new spatial operator. The subsequent section (\[sec3\]) introduces the new prescription suitable to treat anomalous spreading on a bounded domain and describes its main characteristics. The (\[sec4\]) provides two numerical methods to solve the modified space-time fractional differential equation and shows particular results illustrating the key issues. We finish the paper with summary and conclusions.
Finite domain approaches {#sec2}
========================
The mathematically straightforward prescription to create $f_l(x)$ and $f_r(x)$ is padding the function with zero from both sides: $$\begin{aligned}
\label{prescription 1}
\nonumber
f_l(x)&=& \begin{Cases}
u(x,t),\;\;\; 0 \leq x \leq 1 \\
0,\;\;\; -\infty < x < 0
\end{Cases} \\
f_r(x)&=& \begin{Cases}
u(x,t),\;\;\; 0 \leq x \leq 1 \\
0,\;\;\; 1 < x < \infty
\end{Cases}\end{aligned}$$ With such a prescription, the operator $D^\alpha_+$ will yield the left finite Riemann-Liouville fractional derivative $ _0D^\alpha_x u(x,t)$ and the operator $D^\alpha_-$ will yield the right finite Riemann-Liouville fractional derivative $ _xD^\alpha_1 u(x,t)$, where the integrals are defined already only over the appropriate part of the interval $[0,1]$, see [@Podlubny99]. In most numerical calculations published so far such a prescription has been used. It is also the default in the matrix approach. Starting from an $f_i(x)$ probability distribution obeying $f_i(0)=f_i(1)=0$, it seems reasonable to augment the problem with the two Dirichlet boundary conditions $u(0,t)=u(1,t)=0$ and solve it numerically by finite differences.
It is well known however, that the solution with Dirichlet conditions will not satisfy the first law, even in the case of $\beta=1$ and $\alpha=2$. Fig 1. shows the solution of a well documented problem with data $\{\beta=1/2,\;\alpha=3/2,\;a=1,\; f_i(x) = 6 x (1-x)\}$ obtained via the matrix approach suite. (Notice that the actual a parameter passed on should be $a' = \sqrt2$ instead of $a=1$ because in that suite the Riesz derivative is understood without the factor $\cos(\pi \alpha /2 )$.) In the illustration we also show the evolution of the first integral of the spatial density that is the fraction of substance remaining in the finite domain. As it is obvious, material is lost during the process. Replacing the two boundary conditions by “fractional derivative of order $\alpha -1$ equal to zero” condition – motivated by some interpretation of Fick’s law – does not help either. The problematic behavior of prescription (\[prescription 1\]) has been repeatedly discussed, for instance, in the groundwater literature [@Zhang06].
Here we show, that no boundary condition can be found to reconcile the non-physical nature of prescription (\[prescription 1\]). To this end we introduce a small change into the problem specification. Instead of requiring something at the two boundaries, we require that the numerical solution preserve the two important characteristics of the initial distribution: the first integral over $[0,1]$ is equal to unity and it is symmetric. To satisfy these conditions instead of the two Dirichlet boundary conditions we require $\int^1_0 u(x,t) dx = 1$ and $u(0,t)=u(1,t)$. (The second one is obviously necessary, but it turns out to be sufficient as well.) The two conditions are easily passed on to any finite-difference method, in this case to the matrix approach suite [@Podlubny09]. Summarized in Fig. 2 are the results for $f_i(x)= 6 x (1-x)$. Forcing the model to satisfy the first law, we lost compliance to the second law. The normalized entropy ($\sum_{i=1}^n{u_i \ln{u_i}}/\ln{\frac{1}{n}}$, where $n$ is the number of spatial mesh points) is *decreasing* with time.
Turning to Caputo’s idea
------------------------
Similar experiences known for practitioners have led to various ideas. For instance, del-Castillo-Negrete *et al.* [@Del-Castillo-Negrete08] suggested to use a modification of the Riesz derivative operator, following the recipe of Caputo being so successful for the time derivative. In our terminology, the prescription (explicitly given here only for $ 1 < \alpha \leq 2$) $$\begin{aligned}
\label{prescription 2}
\nonumber
f_l(x)&=& \begin{Cases}
u(x,t) -u(0,t)- x \: \left[\frac{\partial}{\partial x}u(x,t)\right]_{\& \; x=0},\;\;\; 0 \leq x \leq 1 \\
0,\;\;\; -\infty < x < 0
\end{Cases} \\
f_r(x)&=& \begin{Cases}
u(x,t) -u(1,t) + (1-x) \: \left[\frac{\partial}{\partial x}u(x,t)\right]_{\& \; x=1},\;\;\; 0 \leq x \leq 1 \\
0,\;\;\; 1 < x < \infty
\end{Cases}\end{aligned}$$ results in the Caputo form of the finite Riesz derivative. (Notice that the prescription introduces a jump discontinuity between the two functions $f_l$ and $f_r$.) Initiating the spreading process from the uniform distribution will leave the initial state at rest, since the right-hand side of (\[problem\]) will be identically zero. Thus the prescription resolved a contradiction, but now we have to face another one: starting the process from a triangular distribution will also leave the system at rest.
Constructing the flux expression (Fick’s law) directly with the Caputo derivative, Zhang *et al.* [@Zhang07] introduced another variant without explicit use of the first spatial derivative. In our notation, the prescription will take the form: $$\begin{aligned}
\label{prescription Zhang}
\nonumber
f_l(x)&=& \begin{Cases}
u(x,t) -u(0,t),\;\;\; 0 \leq x \leq 1 \\
0,\;\;\; -\infty < x < 0
\end{Cases} \\
f_r(x)&=& \begin{Cases}
u(x,t) -u(1,t),\;\;\; 0 \leq x \leq 1 \\
0,\;\;\; 1 < x < \infty
\end{Cases}\end{aligned}$$ The advantage of prescription (\[prescription Zhang\]) is that an initial triangular distribution will not be a steady-state solution of problem (\[problem\]) any more. However, it is still easy to find an initial condition (e.g. a box function non-zero only over a part of the \[0;1\] interval) that will also be a steady-state solution – in contrast to physical intuition. Fig 3 illustrates the problematic behavior of prescriptions (\[prescription 1\]), (\[prescription 2\]), and (\[prescription Zhang\]).
Summary of desired characteristics for a finite domain model
------------------------------------------------------------
From the introductory numerical experiments we glean a wish list. Sought is a macroscopic model in the form of fractional partial differential equation (\[problem\]) with the “hard” requirements (1-5) and “soft” ones (6-10):
i) If the initial state is a probability distribution (non-negative and with unit area under the curve), this property should be preserved for any time;
ii) If the initial probability distribution is symmetric around $x=0.5$, this property should be preserved for any time;
iii) The only stable steady state should be the uniform distribution;
iv) For any non-uniform initial distribution, the entropy should monotonically increase with time;
v) For integer orders, the model should reduce to known results;
vi) Starting from a Dirac delta distribution, the solution should follow the unbounded fundamental solution for short times;
vii) We still want to preserve the deep correspondence with Caputo fractional derivative in time and LW fractional derivative in space (allowing, however, some liberty in the selection of the $f_l$ and $f_r$ functions);
viii) Motivation should stem from a microscopic CTRW concept;
ix) It is desirable to have analytic solution for special cases;
x) It is desirable to have a numerical solution method within the general framework of discrete fractional calculus [@Podlubny09] .
Notice that *item i)* is sometimes stated as the probability preserving property, or the conservation principle. In this work we take the liberty to refer to it as “first law”. *Item ii)* expresses the invariance with respect to directing the coordinate axis, that is we consider only the symmetric Riesz derivative. *Items iii-iv* are obviously related to the second law (of thermodynamics). In some applications (see e.g. stock prices, [@Scalas06]) the listed requirements can be relaxed, but for description of spreading of material they are obviously necessary.
The effects of walls {#sec3}
====================
A prescription by Krepysheva *et al.*
-------------------------------------
Our starting point is the work of Krepysheva *et al.* [@Krepysheva06a] who visualized a “reflecting” wall at location $x=0$ and showed that, due to its non-local character, the kernel of the fractional space derivative has to be modified. The rule for the hopping particle was that if its jump interacts with the wall, it would continue to move in the mirror direction, preserving the overall length of the “initially intended” jump. In the macroscopic limit, the modified Riesz kernel turned out to be markedly different from the standard one based on finite interval left and right Riemann-Luiville derivatives. In our terminology, the works Krepysheva [@Krepysheva06a; @Krepysheva06b] derived the following specific prescription: Starting from an $u(x,t)$ function available over the non-negative $x$-axis, construct: $$\begin{aligned}
\label{prescription 3}
\nonumber
f_l(x)&=& \begin{Cases}
u(x,t),\;\;\; 0 \leq x \leq 1 \\
u(-x,t),\;\;\; -\infty < x < 0 \\
\end{Cases} \\
f_r(x)&=&u(x,t), \;\;\; 0 \leq x < \infty\end{aligned}$$ Calvo *at al.* [@Calvo07] have developed the idea further, for a rather specific geometric situation, when the random walk is along the perimeter of a circle. Building on these results, van Milligen *at al.* [@vanMilligen08] recently suggested a modification of the spatial operator for our problem (\[problem\]) that involves the Hurwitz zeta function. Another extension of the idea – based on the so-called Kolwankar-Gangal derivative – was proposed by Néel *et al.* [@Neel07]. Recently, Zoia *et al.* [@Zoia07] also discussed the effect of walls, although not from a first law point of view.
A new prescription
------------------
This work suggests another turn in the development for the isolating two-wall case. We recall that the hydrodynamic limit procedure makes use of the fact that in general, measurements correspond to time and length scales much larger than those of particle motions. In our opinion, it follows that even the “extremely long” jumps of a particle must be shorter than the domain size. Our main idea is therefore to limit the jump size to one, and hence allow zero or one particle-wall interaction, but never more than one. This suggests a new prescription: Starting from an $u(x,t)$ function available on $0 \leq x \leq 1$, construct $$\begin{aligned}
\nonumber
f_l(\xi) &=& \begin{Cases}
u(\xi,t),\;\;\; 0 \leq \xi \leq x \\
u(-\xi,t),\;\;\; x-1 \leq \xi \leq 0 \\
0,\;\;\; -\infty < \xi < 0
\end{Cases}
\\
\label{prescription new}
f_r(\xi)&=& \begin{Cases}
u(\xi,t),\;\;\; x \leq \xi \leq 1\\
u(2-\xi,t),\;\;\; 1 < \xi \leq 1 +x\\
0,\;\;\; 1+x < \xi < \infty
\end{Cases}\end{aligned}$$ The total support of both $f_l$ and $f_r$ is always two units long but it moves with the location of $x$. The $f_l$ and $f_r$ functions combined contain every function value from the original $u(x,t)$ twice (one corresponding to direct jump and the other bumped from the wall.) Fig. 4 shows the construction of the extensions $f_l$ and $f_r$ for two specific functions and a *specific location* $x=1/3$.
The modified Riemann-Luiville-Riesz derivative
----------------------------------------------
For comparison purposes, we can cast prescription (\[prescription new\]) into a more familiar form, using the concept of modified left and right finite-interval Riemann-Luiville derivatives: $$\begin{aligned}
_{x-1}D^\alpha_{x,mod} f &=& \frac{1}{\Gamma(m-\alpha)} \frac{d^m}{dx^m} \int_{x-1}^{x} \frac{f_{mod}(\xi)d\xi}{(x-\xi)^{\alpha-m+1}}, \;\;\; m = \lceil \alpha \rceil\\
_{x}D^\alpha_{x+1,mod} f &=& \frac{(-1)^m}{\Gamma(m-\alpha)} \frac{d^m}{dx^m} \int_{x}^{x+1} \frac{f_{mod}(\xi)d\xi} {(\xi-x)^{\alpha-m+1}}.\end{aligned}$$ where $$\begin{aligned}
f_{mod}(\xi) &=& f(2-\xi) \Pi \left(\xi-3/2\right)+f(\xi) \Pi \left(\xi-1/2\right)+
f(-\xi) \Pi \left(\xi+1/2\right) \;\;\;\;\;\;\;\;\;\;\end{aligned}$$ with the Heaviside box function defined as $\Pi \left(\xi\right)=1$ for $|\xi| \leq 1/2$ and zero otherwise. These finite Riemann-Luiville derivatives are based on an interval of total length 2, always centered at the location $x$ where we are interested in the derivative. Therefore, for non-integer $\alpha$ the modified finite-interval Riemann-Luiville-Riesz derivative takes the form $$\begin{aligned}
\label{modRLRiesz}
\frac{d ^\alpha}{d_{mod} |x|^\alpha} f(x) =
-\frac{1}{2 \cos(\pi \alpha /2 )} [ _{x-1}D^\alpha_{x,mod} f +\; _xD^\alpha_{x+1,mod} f]\end{aligned}$$ Definition (\[modRLRiesz\]) and prescription (\[prescription new\]) are equivalent.
It is illuminating to compare the regularly used spatial operator (\[prescription 1\]) and the modified one (\[modRLRiesz\]) for some simple functions defined over $[0;1]$. Fig 5 shows the comparison for $f(x)=x-0.5$, $f(x)=(x-0.5)^2$, and $f(x)=\sin(\pi x)$. Somewhat disappointingly, the modified Riesz derivative (\[modRLRiesz\]) does not eliminate singularity at the end points of the interval for these functions and – by and large – behaves similarly to the commonly used definition (\[prescription 1\]). However, for one function family investigated in the next sub-section, the difference is dramatic.
Eigenfunctions and eigenvectors
-------------------------------
Of particular interest is the the family of functions $\cos( j \pi x)$, $j=0,1, \ldots $. Also shown in Fig. 5 the comparison for $f(x)=\cos(3 \pi x)$. We see that prescription (\[prescription 1\]) leads to singular behavior at the endpoints – as usual. On the other hand, (\[modRLRiesz\]) is not only non-singular, but it *almost* coincides with the Fourier-Riesz derivative of the entire $\cos(3 \pi x)$ function over the interval $[0,1]$, differing only in a constant factor $1.01328\ldots$. Using $Mathematica$ one can show that $$\begin{aligned}
\label{eigenfunctions}
\frac{d ^\alpha}{d_{mod} |x|^\alpha} \cos( j \pi x) &=& c_{\alpha,j} \cos( j \pi x) \;\;\; j=0,1,\ldots\end{aligned}$$ where $c_{\alpha,j}$ is constant. In other words, $\cos( j \pi x)$ is an eigenfunction of the operator (\[modRLRiesz\]) with eigenvalue $c_{\alpha,j}$. Moreover, $c_{\alpha,0}=0$, implying that the uniform distribution has a modified Riesz derivative equal to zero. This property will ensure that the uniform distribution will be a steady state solution of (\[problem\]). We managed to obtain closed form expression for specific $\alpha$ parameters (with repeated help from $Mathematica$) as follows: $$\begin{aligned}
\label{eigenvalues}
c_{2,j} &=& - {(j \pi)}^2, \;\;\; j=1,2, \ldots \\
c_{3/2,j} &=& -2 {(j \pi)}^{3/2} C_F(\sqrt{2 j}), \;\;\; j=1,2, \ldots \\
c_{1/2,j} &=& -2 \sqrt{j \pi } S_F(\sqrt{2 j} ), \;\;\; j=1,2, \ldots\end{aligned}$$ where $C_F$ and $S_F$ denote the $\cos$ and $\sin$ Fresnel integrals, respectively. For other $\alpha$ values we could not obtain an explicit expression, but could still develop a simple code in $Mathematica$ that calculates the eigenvalue *with any required number of digit* accuracy for an $\alpha$ given as a rational fraction (see Appendix).
The solution of the space-time fractional differential equation on the interval \[0,1\] {#sec4}
=======================================================================================
We introduce two approaches. The first one uses Laplace transform in time and collocation in space. The second one is based on finite differences.
The Laplace Transform – Collocation method: an example
------------------------------------------------------
We illustrate this method on a simple example: $\{\beta = 1/2,\;\alpha = 3/2,\; a = 1,\; f_i(x)=1 - \cos(2 \pi x)\}$. Using 6 collocation points: $\textbf{x}_c = \{0, \frac15,\frac25,\frac35,\frac45,1\}$ we seek the Laplace transform of the solution at an arbitrary point, $x$. Introducing the vector notation $$\begin{aligned}
\textbf{v}(s)&=&\{ c_0(s),c_1(s),c_2(s),c_3(s),c_4(s),c_5(s)\} \\
\textbf{g}(x)&=&\{1,\cos (\pi x),\cos (2 \pi x),\cos (3 \pi x),\cos (4 \pi x),\cos (5 \pi x)\}\\
\nonumber
\textbf{e}&=&
\{
0,-2 \pi^{3/2} C_F(\sqrt{2}),-4 \sqrt{2} \pi^{3/2} C_F(2),-6 \sqrt{3} \pi ^{3/2} C_F(\sqrt{6}), \\
&&-16\pi^{3/2} C_F(2\sqrt{2}), -10\sqrt {5}\pi^{3/2} C_F(\sqrt{10} ) \}\end{aligned}$$ we represent the Laplace space solution at $x$ as $$\label{Uxs}
U(x,s) =\textbf{v}(s) . \textbf{g}(x)$$ Then its modified Riesz derivative of order $3/2$ takes the form $$\frac{\partial ^{3/2}}{\partial_{mod} |x|^{3/2}} U(x,s) = \textbf{v}(s).\left[ \textbf{e} \, \textbf{g}(x)\right]$$ (with component by component multiplication between $\textbf{e}$ and $\textbf{g}$) and its Caputo derivative of order $1/2$ is written as $$\textbf{L} \left( \frac{\partial ^{1/2}}{\partial t^{1/2}} u(x,t),\;s \right) = s^{-1/2} \left[s \left( \textbf{v}(s) . \textbf{g}(x)\right) - f_i(x)\right]$$ Writing the partial differential equation $$\label{collocation}
s^{-1/2} \left[s \left( \textbf{v}(s) . \textbf{g}(x)\right) - f_i(x)\right] \;-\; \textbf{v}(s).\left[ \textbf{e} \, \textbf{g}(x)\right] = 0$$ at the 6 collocation points $\textbf{x}_c$, we obtain a system of linear equations in the 6 unknown variables, $\textbf{v}(s)$. The solution turns out to be $$\textbf{v}(s)=\{\frac{1}{s},0,-\frac{1}{4 \sqrt{2} \pi ^{3/2} C_F(2) \sqrt{s}+s},0,0,0 \}$$ and hence we obtain $$U(x,s) = \frac{1}{s}-\frac{\cos (2 \pi x)}{4 \sqrt{2} \pi ^{3/2} C_F(2) \sqrt{s}+s}$$ This can be inverted on $s$ resulting in $$\label{inverted}
u(x,t) = 1 + \exp\left({32 \pi ^3 C_F(2)^2 t}\right) \left[ \textrm{erf} \left(4 \sqrt{2} \pi ^{3/2} C_F(2) \sqrt{t} \right)-1 \right]\cos (2 \pi x)$$ Increasing the number of collocation points does not change the solution (\[inverted\]), it is already exact.
The fundamental solution for $\beta = 1/2,\;\alpha = 3/2$
---------------------------------------------------------
We can repeat the above procedure for $f_i(x)=\delta(x-\frac12) = 1-2 \cos(2 \pi x) + 2 \cos(4 \pi x)-2 \cos(6 \pi x)+ \; \ldots$ and obtain the fundamental solution: $$\begin{aligned}
\nonumber
u(x,t)&=&
1+2 \exp\left({32 \pi ^3 C_F(2)^2 t}\right) \cos (2 \pi x) \left[\textrm{erf}\left(4 \sqrt{2} \pi ^{3/2} C_F(2) \sqrt{t}\right)-1\right]-\\
\nonumber
&&2 \exp\left({256 \pi ^3 C_F(2 \sqrt{2})^2 t}\right) \cos (4 \pi x) \left[\textrm{erf}\left(16 \pi ^{3/2} C_F(2 \sqrt{2}) \sqrt{t}\right)-1\right]+\\\nonumber
&&2 \exp\left({864 \pi ^3 C_F(2 \sqrt{3})^2 t}\right) \cos (6 \pi x) \left[\textrm{erf}\left(12 \sqrt{6} \pi ^{3/2} C_F(2 \sqrt{3}) \sqrt{t}\right)-1\right]-\\
\label{fundsol3}
&&\ldots\end{aligned}$$ Looking back at our wish list, we would like to check *item vi)* requiring the correspondence of the unbounded and bounded solutions at early times. The analytic solution of the problem stated on unbounded domain would take the form: $$\begin{aligned}
\label{fundsol2}
u_\infty(x,t) = t^{-\frac13}\textit{M}\left( |x|\, t^{-1/3} ; 1/3 \right)\end{aligned}$$ where $\textit{M}(\xi ; 1/3)$ is given by (\[M13\]). Fig. 6 compares, at an early time, $t=10^{-3}$, the new fundamental solution (\[fundsol3\]) computed with 50 terms and the well-know unbounded solution (\[fundsol2\]). We see that the effect of the boundary has just started to show. (Unfortunately, the numerical evaluation of the fundamental solution is not trivial – even with $Mathematica$ – because of the extremely large exponents and hence we could not increase the number of terms to get rid of the small oscillations of the curve.)
Pursuing the “analytic” solution further has other drawbacks too. The fundamental solution would contain Mittag-Leffler functions (in addition to the eigenvalues, $c_{\alpha,j}$ and eigenfunctions $\cos(j\: \pi \:x)$ ) in the case of a general $\alpha$ and ${\beta}$. Moreover, any other (non Dirac delta) initial condition would necessitate further numerical convolution.
One can, however, easily construct the system of linear equations for a given set of $\alpha$, $\beta$, $a$ and $f_i(.)$ at any selected value of the Laplace variable $s$, solve the system numerically by Gauss elimination and substitute $\textbf{v}(s)$ into (\[Uxs\]). Therefore, we have a way to calculate (at any specified $x$) the Laplace transform of the solution numerically, and hence we can use a numerical inversion technique [@Valko04]. The procedure is robust, if care is taken to do the Gauss elimination with multiple precision – large enough with respect to the number of terms in the GWR algorithm [@Abate04]. We will call the method LT–collocation with numerical inversion. A $Mathematica$ realization of the algorithm is provided in the Appendix.
We emphasize that the procedure does not require any boundary conditions, rather $x=0$ and $x=1$ are included in the set of collocation points. (The physical “boundary conditions” are taken care of within the spatial operator itself.)
Shown in Fig. 7 is the summary of the results of the procedure for $\{\beta = 1/2,\;\alpha = 3/2,\; a = 1,\; f_i(x)=\delta (x-\frac12)\}$, using $51$ collocation points. Since the LT–collocation method with numerical inversion cannot be started from the Dirac delta “function”, we pass on the solution of the unbounded problem (\[fundsol2\]) at a very early time, $t = 0.00001$ as a new “initial” condition. Then we do the numerical inversion for $t-0.00001$ where $t$ is the time we are interested in. The “fraction of substance still in the domain” is not shown in Fig. 7, because at the “initial” state it is unity (for all practical purposes) and hence it remains unity during all times.
Next we compare traditional $\{\beta = 1,\;\alpha = 2,\; a = 1\}$ and space-time fractional spreading $\{\beta = 1/2,\;\alpha = 3/2,\; a = 1\}$ starting from a non-symmetric initial distribution $f_i(x) = 12 x(1-x)^2$, as illustrated by Figs 8 and 9. The number of collocation points is kept at $51$. We find that the spreading is initially faster for the space-time fractional case, but at late time traditional dispersion becomes faster. There is also a remarkable difference in the “overshoot” of the density at $x=0$ – at least for the studied time points. (We note, however, that such comparisons are of limited value, because the parameter $a$ is not the same in the two cases, in spite of looking the same.)
Solution by the method of finite differences
--------------------------------------------
While the LT–collocation with numerical inversion works effectively, it is still illuminating to solve the problem within the framework of discrete fractional calculus. The form of the modified Riesz operator (\[modRLRiesz\]) suggests that a relatively small modification to the established matrix approach will suffice. Indeed, one has to use prescription (\[prescription new\]) to pad the list of available function values and add a small correction to assure that for a constant function the modified Riesz derivatives yield zero.
We introduced this modification into the matrix approach suite (see Appendix for some details). When working with the suite, we do not use the concept of “mathematical boundary conditions” at all, rather we write the equations for the endpoints $x=0$ and $x=1$ as well. Therefore, the total number of equations remains the same as the number of unknowns. Shown in Fig. 10 is the summary of results for our previous example $\{\beta = 1/2,\;\alpha = 3/2,\; a = 1\}$ (that is $a' = \sqrt2$) when the initial condition is $f_i(x) = 12 x(1-x)^2$ and the finite difference step sizes are $\Delta x=0.02$ and $\Delta t=0.01$. (Notice that the number of unknowns in the matrix approach was 5000 and the system matrix had $2.5$ million elements, hence we were limited by computer memory.) Regarding accuracy, the results are still behind the ones depicted in Fig. 9, but the overall correspondence is remarkable. In particular, the fraction of substance still in the domain has less than 2 % error. Not only the modified Riesz derivative is “probability preserving” but also is our finite difference representation of it.
Summary and conclusions
=======================
We have introduced a new version of the finite Riesz derivative. The new spatial operator – combined with the Caputo derivative in time – results in a space-time fractional differential equation that is well suited to describe anomalous spreading of substance in a finite domain. The space-time fractional differential equation satisfies our postulated requirements. If the initial state is a probability distribution (non-negative and with unit area under the curve), this property is preserved for any later time; if the initial probability density is symmetric around $x=0.5$, this property is also preserved. We could not prove rigorously that the only stable steady state is the uniform distribution, but all specific analytic formulaes and numerical examples indicated so. In all our examples, starting from a non-uniform initial distribution, the entropy monotonically increased with time. For integer orders, the model reproduces the known results of Fickian-Markovian diffusion over a finite domain. Starting from a Dirac delta distribution, the solution follows the unbounded fundamental solution of the space-time fractional differential equation for short times. We could preserve the deep correspondence with Liouville-Weyl fractional derivative in space by creating the appropriate prescription, however we are not sure that this is the only (or even the best) way to do it. The new approach arose from a microscopic CTRW concept and we could manage to provide analytic solution for special cases. While our preferred solution method is the LT–collocation with numerical inversion, we could also extend the matrix approach and hence fit our operator into the mainstream framework of discrete fractional calculus. In this work we focused on “pure” diffusion but we do not envisage any difficulty in considering simultaneous advection or other external potential field.
[00]{} Schneider W.R. and Wyss W. Fractional diffusion and wave equations. J. Math. Phys. (1989) 30, 134-144.
Shlesinger M.F., Zaslavsky G.M. and Klafter J. Strange kinetics, Nature (London) (1993) 363, 31-37.
Mainardi F. Fractional relaxation-oscillation and fractional diffusion-wave phenomena, Chaos, Solitons and Fractals (1997) 7 1461-1477.
Berkowitz B., Scher H. On characterization of anomalous dispersion in porous media. Water Resour Res (1995) 31, 1461-1466.
Benson D.A., Wheatcraft S.W. and Meerschaert M.M. The fractional-order governing equation of Levy motion. Water Resour Res (2000) 36, 1403-12.
Benson D.A., Wheatcraft S.W., Meerschaert M.M. Application of a fractional advection-dispersion equation. Water Resour. Res. (2000) 36, 1413-23.
Metzler R. and Klafter J. The random walk’s guide to anomalous diffusion: a fractional dynamics approach. Phys. Rep. (2000) 339, 1.
Metzler R. and Klafter J. The restaurant at the end of the random walk. Recent developments in the description of anomalous transport by fractional dynamics. J. Phys. A (2004) 37, R161-R208.
Zaslavsky G.M., Chaos, fractional kinetics, and anomalous transport. Physics Report (2002) 371, (6) 461-580.
Chechkin A.V., Gonchar V.Y., Metzler R. and Klafter J. Fundamentals of Lévy flight processes. Advances in Chemical Physics (2006) 133, 439-496.
Zhang Y., Benson D.A. and Reeves D.M. Time and space nonlocalities underlying fractional-derivative models: Distinction and literature review of field applications. Advances in Water Resources (2009) 32, (4) 561-581.
Montroll E.W. and Weiss G.H. Random walks on lattices II, J. Math. Phys. (1965) 6, 167-181.
Cushman J.H. On diffusion in fractal media. Water Resour. Res. (1991) 27, (4) 643-644.
Barkai E. CTRW pathways to the fractional diffusion equation. Chem. Phys. (2002) 284, 13-27.
Gorenflo R., Mainardi F., Moretti D., Pagnini G. and Paradisi P. Discrete random walk models for space-time fractional diffusion, Chemical Physics (2002) 284, (1-2) 521-541.
Paradisi P., Cesari R., Mainardi F. and Tampieri F. Fractional Fick’s law for non-local transport processes. Physica A: Statistical Mechanics and its Applications (2001) 293, (1-2) 130-142.
Néel M-C., Abdennadher A. and Joelson M. Fractional Fick’s law: The direct way. Journal of Physics A: Mathematical and Theoretical (2007) 40, (29), art. 007 8299-8314.
Caputo M. Linear models of dissipation whose Q is almost frequency independent-II. Geophy. J. R. Astro. Soc. (1967) 13, 529-539.
Saichev A.I. and Zaslavsky G.M. Fractional kinetic equations: solutions and applications. Chaos (1997) 7 753-764.
Chechkin A.V., Gonchar V.Yu., Gorenflo R. Korabel N. and Sokolov I.M. Generalized fractional diffusion equations for accelerating subdiffusion and truncated Lévy flights. Phys. Rev. E (2008) 78, art. 021111 1-13.
Montroll E.W. and Scher H. Random walks on lattices. IV. Continuous-time walks and influence of absorbing boundaries. J. Stat. Phys. (1973) 9 (2) 101-135.
Mainardi F., Luchko Yu. and Pagnini G. The fundamental solution of the space-time fractional diffusion equation. Fractional Calculus Appl. Anal. (2001), 4 153-192.
Kilbas A.A., Srivastava H.M. and Trujillo J.J. Theory and Applications of Fractional Differential Equations (North-Holland Mathematical Studies vol 204) ed J Mill North-Holland, Amsterdam 2006.
Piryatinska A., Saichev A.I. and Woyczynski W.A. Models of anomalous diffusion: the subdiffusive case. Physica A (2005) 349, 375-420.
Podlubny I., Chechkin A., Skovranek T., Chen Y. and Vinagre Jara B.M. Matrix approach to discrete fractional calculus II: Partial fractional differential equations. J. Comp. Phys. (2009) 228, (8) 3137-3153. (doi:10.1016/j.jcp.2009.01.014)
Chechkin A.V., Metzler R., Gonchar V.Y., Klafter J and Tanatarov L.V. First passage and arrival time densities for Levy flights and the failure of the method of images. J. Phys. A - Math General (2003) 36, L537-544.
Metzler R., Chechkin A.V., Gonchar V.Yu. and Klafter J. Some fundamental aspects of Levy flights. Chaos, Soliton and Fractals (2007) 34, 129-142. (doi:10.1016/j.chaos.2007.01.055)
Podlubny I. Fractional differential equation, Academic Press, New York 1999.
Zhang Y., Benson D.A., Meerschaert M.M. and LaBolle E.M. Space-fractional advection-dispersion equations with variable parameters: Diverse formulas, numerical solutions, and application to the Macrodispersion Experiment site data. Water Resources Research (2007) 43, (5), art. W05439.
Del-Castillo-Negrete D. Fractional diffusion models of nonlocal transport. Physics of Plasmas (2006) 13, (8), art. 082308 1-17.
del-Castillo-Negrete D., Gonchar V.Yu. and Chechkin A.V. Fluctuation-driven directed transport in the presence of Lévy flights. Physica A: Statistical Mechanics and its Applications (2008) 387, (27) 6693-6704.
Zhang X., Lv M., Crawford J.W. and Young I.M. The impact of boundary on the fractional advection-dispersion equation for solute transport in soil: Defining the fractional dispersive flux with the Caputo derivatives. Advances in Water Resources (2007) 30, 1205-1217.
Scalas E. The application of continuous-time random walks in finance and economics. Physica A: Statistical Mechanics and its Applications (2006) 362, (2) 225-239.
Krepysheva N., Pietro L.Di. and Néel M-C. Fractional diffusion and reflective boundary condition. Physica A (2006) 368, 355-361.
Krepysheva N., Pietro L.Di. and Néel M-C. Space fractional advection-diffusion and reflective boundary conditions. Phys. Rev. E (2006) 73, 021104 1-9.
Calvo I., Carreras B.A., S´anchez R. and van Milligen B.Ph., Continuous time random walks in periodic systems: fluid limit and fractional differential equations on the circle. J. Phys. A: Math. Theor. (2007) 40, 13511.
van Milligen B.Ph., Calvo I. and Sánchez R. Continuous time random walks in finite domains and general boundary conditions: Some formal considerations. Journal of Physics A: Mathematical and Theoretical (2008) 41, (21), art. 215004.
Zoia A., Rosso A. and Kardar M. Fractional Laplacian in bounded domains. Physical Review E - Statistical, Nonlinear, and Soft Matter Physics (2007) 76, (2), art. 021116 1-13.
Valkó P.P. and Abate J. Comparison of Sequence Accelerators for the Gaver Method of Numerical Laplace Transform Inversion. Comp. & Math. with Appl. (2004) 48, 629-636. (doi:10.1016/j.camwa.2002.10.017)
Abate J. and Valkó P.P. Multi-Precision Laplace Transform Inversion. Int. J. Numer. Meth. Engng. (2004) 60, 979-993. (doi:10.1002/nme.995)
$Mathematica$ v7.0 $\textrm{http://www.wolfram.com/}$ (Last visited: Aug. 15, 2009)
Appendix
========
Calculations were done in $Mathematica$. Here we show some code snippets and results in order to ease reproduction of our results.
The code snippet for the LT-collocation method with numerical inversion is the following:
$$\begin{array}{lll}
\textsf{vari}[\textsf{s$\_$}]&=&\textsf{Table}[c[j][s],\{j,0,\textsf{nm}\}];\\
\textsf{cosi}[\textsf{x$\_$}]&=&\textsf{Table}[\textsf{Cos}[ (j \pi )\textsf{ }x],\{j,0,\textsf{nm}\}];\\
\textsf{symi}[\textsf{x$\_$}]&=&\textsf{Table}[\textsf{scos}[\alpha ][j][x],\{j,0,\textsf{nm}\}];\\
\textsf{equl}[\textsf{s$\_$}][\textsf{x$\_$}]&=&s{}^{\wedge} {(\textsf{$\beta $}-1)} \; (s\textsf{ }\textsf{vari}[s].\textsf{cosi}[x]\textsf{ }-\textsf{fi}[x]) - a \; \textsf{vari}[s].\textsf{symi}[x] ;\\
\textsf{xc}&=&N[\textsf{Range}[0,1,1/\textsf{nm}],\textsf{maxprec}];\\
\{\textsf{mb}[\textsf{s$\_$}],\textsf{As}[\textsf{s$\_$}]\}&=&\textsf{Map}[\textsf{Normal},
\textsf{CoefficientArrays}[\textsf{equl}[s][\textsf{xc}],\textsf{vari}[s]]];\\
\textsf{sols}[\textsf{s$\_$}]&:=&\textsf{sols}[s]=\textsf{LinearSolve}[\textsf{As}[s],-\textsf{mb}[s]];\\
\textsf{Uxs}[\textsf{x$\_$}][\textsf{s$\_$}]&:=&\textsf{sols}[s].\textsf{cosi}[x];\\
\textsf{uxt}[\textsf{M$\_$}][\textsf{x$\_$}][\textsf{t$\_$}]&:=&\textsf{GWR}[\textsf{Uxs}[x],t,\textsf{M}];\\
\end{array}$$
The user has to provide $\beta, \alpha, a$, the function $fi[x\_]$ in addition to the integer $nm$ (the number of collocation points minus one.) The number of terms in the GWR algorithm $M$ [@Abate04] and the maximum used precision $maxprec$ (we used 200) are also required. Once defined, the $\textsf{uxt}[\textsf{M}][\textsf{x}][\textsf{t}]$ function can be used to calculate the solution at a specific $x$ and $t$. The $\textsf{scos}[\alpha ][j][x]$ expression should evaluate to $-2 j^{3/2} \pi^{3/2} FresnelC[\sqrt{2 j}] Cos[j \pi x]$ when $\alpha =3/2$. In general, it will be the product of an eigenvalue $cei[\alpha][j]$ and the appropriate $Cos[j \pi x]$. While no general formula is currently available for the eigenvalue, we can calculate it by the following code snippet for a given rational fraction $\alpha$ and positive integer $j$:
$$\begin{array}{lcl}
\textsf{eug}=\textsf{EulerGamma};\\
h[j][x\_]=\textsf{Cos}[j \: \pi x];
\\
\textsf{cei}[\alpha][j]=N[ \textsf{symRmod}[h[j],\textsf{eug},\alpha]/\textsf{h}[j][\textsf{eug}], \textsf{maxprec}];
\end{array}$$ where
$$\textsf{symRmod}[f\_,x\_,\alpha\_]:=-\frac{1}{2 Cos[\alpha \pi /2]} (\textsf{lRmod}[f,x,\alpha]+
\textsf{rRmod}[f,x,\alpha]);$$
In the above code the left modified Rieman-Luiville-Riesz derivative is calculated from
$$\begin{array}{l}
\textsf{lRmod}[\textsf{f$\_$},\textsf{t$\_$},\alpha \_] := \textsf{Module}[\{\textsf{fext},m=\textsf{Ceiling}[\alpha ]\},
\\
\textsf{fext}[\textsf{x$\_$}]=f[x]\;\textsf{UnitBox}[x-1/2]+f[2-x]\;\textsf{UnitBox}[x-3/2]+
f[-x]\;\textsf{UnitBox}[x+1/2];\;\;\;\;
\\
\textsf{If} [ \textsf{IntegerQ}[\alpha ],\;
D[f[t],\{t,\alpha \}],
\\
\frac{1}{
\textsf{Gamma[m - $\alpha$ ]}
}\;
D[(\textsf{Integrate}[\frac{\textsf{fext}[\tau ]}{(\textsf{eug} -\tau ){}^{\wedge}(\alpha +1-m)},
\{\tau ,\textsf{eug}-1,\textsf{eug}\}])\textsf{/.}\textsf{eug} \rightarrow t,
\{t,m \}]]]
\end{array}$$ and the right modified finite-interval Rieman-Luiville-Riesz derivative is calculated from $$\begin{array}{l}
\textsf{rRmod}[\textsf{f$\_$},\textsf{t$\_$},\alpha \_] := \textsf{Module}[\{\textsf{fext},m=\textsf{Ceiling}[\alpha ]\},
\\
\textsf{fext}[\textsf{x$\_$}]=f[x]\;\textsf{UnitBox}[x-1/2]+f[2-x]\;\textsf{UnitBox}[x-3/2]+
f[-x]\;\textsf{UnitBox}[x+1/2];\;\;\;\;
\\
\textsf{If} [ \textsf{IntegerQ}[\alpha ],\;
D[f[t],\{t,\alpha \}],
\\
\frac{(-1)^m}{
\textsf{Gamma[m - $\alpha$ ]}
}\;
D[(\textsf{Integrate}[\frac{\textsf{fext}[\tau ]}{(\tau-\textsf{eug} ){}^{\wedge}(\alpha +1-m)},
\{\tau ,\textsf{eug},\textsf{eug}+1 \}])\textsf{/.}\textsf{eug} \rightarrow t,
\{t,m \}]]]
\end{array}$$ (The extensive use of “EulerGamma” is somewhat arbitrary but proved useful in the context of the current version – v7.0 – of $Mathematica$ [@MMA].)
Without going into details of the derivation, here we illustrate the concept of extending the matrix approach suite of Podlubny *et al.* [@Podlubny09]. For $1<\alpha\leq2$, the current symmetric Riesz function of the suit (called $ransym$) would hypothetically provide the following array $SR$ of symmetric Riesz derivatives at points $\{0,1h,2h,3h,4h\}$ where the function values $\{y_0,y_1,y_2,y_3,y_4\}$ are known:
$$\begin{array}{lll}
SR_0&=&
2^{-1+2 \alpha } (1-\alpha ) y_0+\\&&2^{-1+2 \alpha } y_1 \\
SR_1&=& \frac{1}{3} 4^{-2+\alpha } (24+12 (-1+\alpha ) \alpha ) y_0-\\&&4^{\alpha } \alpha y_1+\\
&&
\frac{1}{3} 4^{-2+\alpha } (24+12 (-1+\alpha ) \alpha ) y_2-\\&&\frac{1}{3} 4^{-1+\alpha } (-2+\alpha ) (-1+\alpha ) \alpha y_3+\\
&&
\frac{1}{3} 4^{-2+\alpha } (-3+\alpha ) (-2+\alpha ) (-1+\alpha ) \alpha y_4 \\
SR_2&=& \frac{1}{3} 4^{-1+\alpha } (2-\alpha ) (-1+\alpha ) \alpha y_0+\\&&\frac{1}{3} 4^{-1+\alpha } (6+3 (-1+\alpha ) \alpha ) y_1-\\
&&4^{\alpha } \alpha y_2+\\&&
\frac{1}{3} 4^{-1+\alpha } (6+3 (-1+\alpha ) \alpha ) y_3+\\
&&\frac{1}{3} 4^{-1+\alpha } (2-\alpha ) (-1+\alpha ) \alpha y_4 \\
SR_3&=& \frac{1}{3} 4^{-2+\alpha } \alpha \big(-6+11 \alpha -6 \alpha ^2+\alpha ^3\big) y_0-\\&&\frac{1}{3} 4^{-1+\alpha } \alpha \big(2-3 \alpha +\alpha ^2\big) y_1+\\
&&4^{-1+\alpha } \big(2-\alpha +\alpha ^2\big) y_2-\\&&
4^{\alpha } \alpha y_3+\\
&&4^{-1+\alpha } \big(2-\alpha +\alpha ^2\big) y_4 \\
SR_4&=& 2^{-1+2 \alpha } y_3+\\&&2^{-1+2 \alpha } (1-\alpha ) y_4\\
\end{array}$$
A symmetric Riesz function modified according to prescription (\[prescription new\]) would hypothetically provide the following array $SRM$ for the same input:
$$\begin{array}{lll}
SRM_0&=&
-2^{2 \alpha } \alpha y_0+\\&&2^{-1+2 \alpha } \big(2-\alpha +\alpha ^2\big) y_1-\\&&
\frac{1}{3} 2^{-1+2 \alpha } \alpha \big(2-3 \alpha +\alpha ^2\big) y_2+\\&&\frac{1}{3} 2^{-3+2 \alpha } (-3+\alpha ) \alpha \big(2-3 \alpha +\alpha ^2\big) y_3+\\&&
\frac{1}{3} 2^{-3+2 \alpha } (4-\alpha ) (-3+\alpha ) \big(2-3 \alpha +\alpha ^2\big) y_4 \\
SRM_0&=&
4^{-1+\alpha } \big(2-\alpha +\alpha ^2\big) y_0-\\&&\frac{1}{3} 4^{-1+\alpha } \alpha \big(14-3 \alpha +\alpha ^2\big) y_1+\\&&
\frac{1}{3} 4^{-2+\alpha } \big(24-18 \alpha +23 \alpha ^2-6 \alpha ^3+\alpha ^4\big) y_2+\\&&\frac{1}{3} 4^{-2+\alpha } \big(-48+92 \alpha -58 \alpha ^2+16 \alpha ^3-2 \alpha ^4\big) y_3+\\&&
\frac{1}{3} 4^{-2+\alpha } \big(-6 \alpha +11 \alpha ^2-6 \alpha ^3+\alpha ^4\big) y_4 \\
SRM_0&=&
-\frac{1}{3} 4^{-1+\alpha } \alpha \big(2-3 \alpha +\alpha ^2\big) y_0+\\&&\frac{1}{3} 4^{-2+\alpha } \big(24-18 \alpha +23 \alpha ^2-6 \alpha ^3+\alpha ^4\big) y_1+\\&&
\frac{1}{3} 4^{-2+\alpha } \big(-48+52 \alpha -70 \alpha ^2+20 \alpha ^3-2 \alpha ^4\big) y_2+\\&&\frac{1}{3} 4^{-2+\alpha } \big(24-18 \alpha +23 \alpha ^2-6 \alpha ^3+\alpha ^4\big) y_3+\\&&
\frac{1}{3} 4^{-2+\alpha } \big(-8 \alpha +12 \alpha ^2-4 \alpha ^3\big) y_4 \\
SRM_0&=&
\frac{1}{3} 4^{-2+\alpha } \alpha \big(-6+11 \alpha -6 \alpha ^2+\alpha ^3\big) y_0-\\&&\frac{1}{3} 2^{-3+2 \alpha } \big(24-46 \alpha +29 \alpha ^2-8 \alpha ^3+\alpha ^4\big) y_1+\\&&
\frac{1}{3} 4^{-2+\alpha } \big(24-18 \alpha +23 \alpha ^2-6 \alpha ^3+\alpha ^4\big) y_2+\\&&\frac{1}{3} 4^{-2+\alpha } \big(-56 \alpha +12 \alpha ^2-4 \alpha ^3\big) y_3+\\&&
\frac{1}{3} 4^{-2+\alpha } \big(24-12 \alpha +12 \alpha ^2\big) y_4 \\
SRM_0&=&
\frac{1}{3} 2^{-3+2 \alpha } \big(-24+50 \alpha -35 \alpha ^2+10 \alpha ^3-\alpha ^4\big) y_0+\\&&\frac{1}{3} 2^{-3+2 \alpha } \alpha \big(-6+11 \alpha -6 \alpha ^2+\alpha ^3\big) y_1\\&&
-\frac{1}{3} 2^{-1+2 \alpha } \alpha \big(2-3 \alpha +\alpha ^2\big) y_2+\\&&2^{-1+2 \alpha } \big(2-\alpha +\alpha ^2\big) y_3\\&&
-2^{2 \alpha } \alpha y_4
\end{array}$$
One can easily check by substitution, that when $y_0=y_1=y_2=y_3=y_4$, each derivative is zero $SRM_0=SRM_1=SRM_2=SRM_3=SRM_4=0$, and when $y_0=1, y_1=0.5, y_2=0,y_3=-0.5,y_4=-1$, the sum is zero $SRM_0+SRM_1+SRM_2+SRM_3+SRM_4\;=\;0$.
Figure captions
===============
{width="85.00000%"}
{width="85.00000%"}
{width="85.00000%"}
{width="50.00000%"} {width="50.00000%"}
{width="50.00000%"} {width="50.00000%"} {width="50.00000%"} {width="50.00000%"}
{width="85.00000%"}
{width="85.00000%"}
{width="85.00000%"}
{width="85.00000%"}
{width="85.00000%"}
| {
"pile_set_name": "ArXiv"
} |
Tiling a unit square with 8 squares
Iwan Praton\
Franklin and Marshall College\
Lancaster, PA 17604\
`[email protected]`
Suppose a unit square is packed with $n$ squares of side lengths $s_1, s_2,\ldots, s_n$. We define $\psi_1(n)=\max \sum_{i=1}^n s_i$, where the maximum is taken over all possible packings of the unit square. Not a lot is known about the function $\psi_1$. Erd[ő]{}s \[1\] asked whether $\psi_1(k^2+1)=k$. More generally, Erd[ő]{}s and Soifer \[2\] presented explicit packings that provided lower bounds of $\psi_1(n)$ for all (nonsquare) $n$; they mentioned that these lower bounds appear to be good. Thus we have tentative values for $\psi_1(n)$.
In \[3\] Staton and Tyler introduced two modifications of $\psi_1$ as follows. Define a *right packing* to be a packing by squares whose sides are parallel to the sides of the unit square. Then $\psi_2(n)$ is defined to be $\max \sum s_i$ where the maximum is taken over all right packings with $n$ squares. Also, $\psi_3(n)$, for $n\neq 2, 3, 5$, is defined to be $\max \sum s_i$, where the maximum is now taken over all right *tilings* with $n$ squares. (A tiling is a packing where the unit square is completely filled. The unit square can be tiled with $n$ squares for all values of $n$ except for $n=2,3,5$, thus the restriction on $n$ in the definition of $\psi_3$.) It is clear that $\psi_1(n)\geq \psi_2(n)\geq \psi_3(n)$. Staton and Tyler asked for what values of $n$ we have $\psi_1(n)=\psi_2(n)=\psi_3(n)$.
There are some reasons to suspect that the three functions might be identical. The packings constructed by Erd[ő]{}s and Soifer in \[2\] are actually tilings, except when $n$ differs by 1 from a square integer. Staton and Tyler in \[3\] took care of the case when $n$ is one more than a perfect square by constructing tilings whose sums of edge lengths are the same as the Erd[ő]{}s-Soifer lower bounds. Thus if the Erd[ő]{}s-Soifer conjecture is correct, then $\psi_1(n)=\psi_2(n)=\psi_3(n)$ for all values of $n$ except possibly when $n$ is one less than a perfect square. In this note we show that, alas, $\psi_2(n)\neq
\psi_3(n)$ when $n=8$; more precisely, we show that $\psi_3(8)=2.6$.
We first define our terminology and notation. All packings and tilings in this paper of the unit square, so we will omit the phrase “of the unit square” in what follows. If $A$ is a square, its side length is denoted by $s_A$. If $\mathcal{C}=\{A_1,\ldots,A_n\}$ is a collection of squares, we write $\sigma(\mathcal{C})=
\sigma(A_1,\dots,A_n)$ for $\sum s_{A_i}$.
Here is an upper bound due to Erdös; the proof below appeared in Erdös and Soifer \[2\].
\[CS\] If $\mathcal{C}$ is a collection of $n$ squares with total area $A$, then $\sigma(\mathcal{C})
\leq \sqrt{nA}$, with equality only if the $n$ squares are the same size.
Let $s_1,\dots, s_n$ be the side lengths of the $n$ squares. Apply the Cauchy-Schwarz inequality to the $n$-component vectors $(1,1,\dots,1)$ and $(s_1,\dots,s_n)$.
As an immediate consequence, we see that $\psi_3(8)\leq \sqrt{8}$. We also get a lower bound from an explicit construction: the tiling
0.5mm
(50.00,50.00)(0,0)
(0.00,0.00)[(1,0)[50.00]{}]{} (0.00,0.00)[(0,1)[50.00]{}]{} (50.00,0.00)[(0,1)[50.00]{}]{} (0.00,50.00)[(1,0)[50.00]{}]{} (40.00,40.00)[(1,0)[10.00]{}]{} (40.00,40.00)[(0,1)[10.00]{}]{} (50.00,40.00)[(0,1)[10.00]{}]{} (40.00,50.00)[(1,0)[10.00]{}]{} (40.00,30.00)[(1,0)[10.00]{}]{} (40.00,30.00)[(0,1)[10.00]{}]{} (50.00,30.00)[(0,1)[10.00]{}]{} (40.00,40.00)[(1,0)[10.00]{}]{} (30.00,20.00)[(1,0)[10.00]{}]{} (30.00,20.00)[(0,1)[10.00]{}]{} (40.00,20.00)[(0,1)[10.00]{}]{} (40.00,30.00)[(1,0)[10.00]{}]{} (40.00,20.00)[(1,0)[10.00]{}]{} (40.00,20.00)[(0,1)[10.00]{}]{} (50.00,20.00)[(0,1)[10.00]{}]{} (40.00,30.00)[(1,0)[10.00]{}]{} (0.00,30.00)[(1,0)[20.00]{}]{} (0.00,30.00)[(0,1)[20.00]{}]{} (20.00,30.00)[(0,1)[20.00]{}]{} (0.00,50.00)[(1,0)[20.00]{}]{} (20.00,30.00)[(1,0)[20.00]{}]{} (20.00,30.00)[(0,1)[20.00]{}]{} (40.00,30.00)[(0,1)[20.00]{}]{} (20.00,50.00)[(1,0)[20.00]{}]{} (30.00,0.00)[(1,0)[20.00]{}]{} (30.00,0.00)[(0,1)[20.00]{}]{} (50.00,0.00)[(0,1)[20.00]{}]{} (30.00,20.00)[(1,0)[20.00]{}]{} (0.00,0.00)[(1,0)[30.00]{}]{} (0.00,0.00)[(0,1)[30.00]{}]{} (30.00,0.00)[(0,1)[30.00]{}]{} (0.00,30.00)[(1,0)[30.00]{}]{}
shows that $\psi_3(8)\geq 2.6$. To show that $\psi_3(8)=2.6$, we need to investigate the actual tiling in more detail.
Put our unit square so its corners are at $(0,0)$, $(1,0)$, $(0,1)$, and $(1,1)$. Let ${\mathcal{C}}$ be any tiling of this square with 8 tiles. For any $c$ where $0< c < 1$, we define ${\mathcal{C}}_c$ to be the set of tiles whose interior intersect the vertical line $x=c$. We want to avoid the case where there is a tile with a vertical edge on the line $x=c$ (such a line is called *ambiguous* by Staton and Tyler \[2\]), so we will assume forthwith that the vertical line $x=c$ is not ambiguous. Thus $\sigma({\mathcal{C}}_c)=1$. Note that there is an unambiguous line as close as we want to an ambiguous line.
The values $c=0$ and $c=1$ are special. We call the line $x=0$ the *left coast* and the line $x=1$ the *right coast*. The *left coastal tiles* ${\mathcal{C}}_0$ are the tiles that have an edge on the left coast. Similarly, the *right coastal tiles* ${\mathcal{C}}_1$ are those tiles with an edge on the right coast. Their union is the set of *coastal tiles*. Tiles that are not coastal tiles are called *inland tiles*. There are not too many of these.
\[inlandtiles\] The sum of the side lengths of all inland tiles is less than 1.
For any tiling ${\mathcal{C}}$, we know that $\sigma({\mathcal{C}})\leq \sqrt{8}<3$. We have $\sigma({\mathcal{C}}_0)=\sigma(C_1)=1$. If the sum of the side lengths of inland tiles is 1 or more, then $\sigma({\mathcal{C}})\geq 1+1+1
=3$, a contradiction.
\[noinlines\] For any $0<c<1$, the set ${\mathcal{C}}_c$ contains at least one coastal tile.
Otherwise ${\mathcal{C}}_c$ contains only inland tiles. Since $\sigma(C_c)=1$, this contradicts Lemma \[inlandtiles\].
\[twobigtiles\] There is a tile $A\in {\mathcal{C}}_0$ and $B\in {\mathcal{C}}_1$ such that $s_A+s_B=1$.
Let $a$ denote the maximum side lengths of all left coastal tiles; similarly, let $b$ denote the maximum side length of all right coastal tiles. If $a+b<1$, then there exists a value $x_0$ (where $a<x_0<1-b$) such that the line $x=x_0$ does not intersect any coastal tiles. This contradicts Lemma \[noinlines\]. Thus $a+b=1$, which is what we want.
**Note**: the proof works just as well when we turn the tiling 90 degrees. Thus there exist two tiles, one with an edge on the line $y=0$, and one with an edge on the line $y=1$, such that the total edge lengths of these two tiles is 1.
Suppose as in Lemma \[twobigtiles\] we have tiles $A\in {\mathcal{C}}_0$ and $B\in {\mathcal{C}}_1$ with $s_A+s_B=1$.
One (or both) of $A$ and $B$ is a corner tile.
Suppose not.
0.5mm
(50.00,50.00)(0,0)
(0.00,0.00)[(1,0)[50.00]{}]{} (0.00,0.00)[(0,1)[50.00]{}]{} (50.00,0.00)[(0,1)[50.00]{}]{} (0.00,50.00)[(1,0)[50.00]{}]{} (30.00,17.00)[(1,0)[20.00]{}]{} (30.00,17.00)[(0,1)[20.00]{}]{} (50.00,17.00)[(0,1)[20.00]{}]{} (30.00,37.00)[(1,0)[20.00]{}]{} (00.00,10.00)[(1,0)[30.00]{}]{} (00.00,10.00)[(0,1)[30.00]{}]{} (30.00,10.00)[(0,1)[30.00]{}]{} (00.00,40.00)[(1,0)[30.00]{}]{} (15.00,25.00)[(0,0)\[cc\][A]{}]{}
(40.00,26.00)[(0,0)\[cc\][B]{}]{}
0.5mm
(50.00,50.00)(0,0)
(0.00,0.00)[(1,0)[50.00]{}]{} (0.00,0.00)[(0,1)[50.00]{}]{} (50.00,0.00)[(0,1)[50.00]{}]{} (0.00,50.00)[(1,0)[50.00]{}]{} (12.00,30.00)[(1,0)[20.00]{}]{} (12.00,30.00)[(0,1)[20.00]{}]{} (32.00,30.00)[(0,1)[20.00]{}]{} (12.00,50.00)[(1,0)[20.00]{}]{} (10.00,0.00)[(1,0)[30.00]{}]{} (10.00,0.00)[(0,1)[30.00]{}]{} (40.00,00.00)[(0,1)[30.00]{}]{} (10.00,30.00)[(1,0)[30.00]{}]{} (25.00,15.00)[(0,0)\[cc\][A]{}]{}
(22.00,40.00)[(0,0)\[cc\][B]{}]{}
Then rotating the tiling 90 degrees produces two inland tiles whose side lengths add up to 1, contradicting Lemma \[inlandtiles\].
From now on we will assume, without loss of generality, that the left coastal tile $A$ is a corner tile, with a corner at $(0,0)$.
Note that we can further assume, without loss of generality, that $B$ is also a corner tile, with a corner at $(1,0)$. For any tiling where $B$ has a corner at $(1,b)$, with $b>0$, there is a similar tiling, with the same total edge length, where $B$ has a corner at $(1,0)$.
0.5mm
(30.00,50.00)(0,0)
(0.00,0.00)[(1,0)[30.00]{}]{} (0.00,0.00)[(0,1)[50.00]{}]{} (30.00,0.00)[(0,1)[50.00]{}]{} (0.00,50.00)[(1,0)[30.00]{}]{} (0.00,20.00)[(1,0)[30.00]{}]{} (0.00,20.00)[(0,1)[30.00]{}]{} (30.00,20.00)[(0,1)[30.00]{}]{} (0.00,50.00)[(1,0)[30.00]{}]{} (15.00,15.00)[(0,0)\[cc\]]{}
(15.00,10.00)[(0,0)\[cc\]]{}
(15.00,35.00)[(0,0)\[cc\][B]{}]{}
(30.00,50.00)(0,0)
(0.00,0.00)[(1,0)[30.00]{}]{} (0.00,0.00)[(0,1)[50.00]{}]{} (30.00,0.00)[(0,1)[50.00]{}]{} (0.00,50.00)[(1,0)[30.00]{}]{} (0.00,0.00)[(1,0)[30.00]{}]{} (0.00,0.00)[(0,1)[30.00]{}]{} (30.00,0.00)[(0,1)[30.00]{}]{} (0.00,30.00)[(1,0)[30.00]{}]{} (15.00,45.00)[(0,0)\[cc\]]{}
(15.00,40.00)[(0,0)\[cc\]]{}
(15.00,15.00)[(0,0)\[cc\][B]{}]{}
Clearly it does no harm to assume that $s_A\geq s_B$. (Simply reflect the tiling across the line $x=1/2$ if necessary.) Thus our tiling contains a big tile $A$, with a corner at $(0,0)$, where $s_A\geq 1/2$. There is also a tile $B$, with a corner at $(1,0)$, where $s_B=1-s_A$. Similarly (see the note after Lemma \[twobigtiles\]) there is tile $B'$, with $s_{B'}=1-s_A$, which we can assume has a corner at $(0,1)$. This is enough to show that $\psi_3(8)$ is not equal to $\psi_2(8)$.
$\psi_2(8)>\psi_3(8)$.
In the standard $3\times 3$ tiling, remove one tile. We then have a packing with 8 squares with total edge length $\frac83$. Thus $\psi_2(8)\geq \frac83$, so all we need to show is that $\psi_3(8)<\frac83$.
Let $t=s_B$. The three tiles $A$, $B$, $B'$ have total area $2t^2+(1-t)^2=1-2t+3t^2$, leaving an area of $2t-3t^2$ for the remaining 5 tiles. By Lemma \[CS\], the total side lengths of these 5 tiles is at most $\sqrt{5(2t-3t^2)}=\sqrt{10t-15t^2}$. Thus the total side lengths of all 8 tiles is at most $1+t+\sqrt{10t-5t^2}$. It is straightforward to verify that this function has a maximum at $t=5/12$, with a maximum value of $8/3$. Thus we get that $\psi_3(8)\leq 8/3$.
Equality is achieved only if $t=5/12$ and the 5 tiles are all the same size. Let us figure out what this size is. The 5 tiles have a total area of $2t-3t^2=2\cdot (5/12)-3\cdot (5/12)^2
=45/144$, so each tile has area $9/144$, i.e., each tile has side length $3/12$. Now $B$ (and $B'$) must have an edge on the border of the unit square, so the remaining $7/12$ must be covered by tiles of side length $3/12$, i.e., an integer multiple of $3/12$ must be equal to $7/12$. This is impossible. Thus the optimal tiling must either have $t\neq 5/12$ or it must have 5 remaining tiles of different sizes. In either case, the total side length will be smaller than $8/3$. Hence $\psi_3(8)<8/3$, as claimed.
We will now proceed with the proof that $\psi_3(8)=2.6$. Suppose $P$ is an optimal tiling, i.e., $\sigma(P)$ is maximal. We know that $\sigma(P)\geq 2.6$. As always, we assume without harm that $P$ contains a corner tile $A$ with a corner at $(0,0)$; there are also at least two tiles $B$ and $B'$ with edge lengths $s_B=s_{B'}=1-s_A$.
\[atmost3\] There are at most 3 tiles with edge lengths $s_B$.
Suppose there are 4 tiles with edge lengths $t=s_B$. Then these 4 tiles, together with $A$, have total area $4t^2+(1-t)^2$, leaving an area of $2t-5t^2$ for the remaining 3 tiles. The edge lengths of these 3 tiles sum up to at most $\sqrt{3(2t-5t^2)}=
\sqrt{6t-15t^2}$, so the total edge length of all 8 tiles is at most $4t+(1-t)+\sqrt{6t-15t^2}=
1+3t+\sqrt{6t-15t^2}$. It is straightforward to calculate that this function has a maximum value of $\frac{8+2\sqrt{6}}{5}
<2.58$ (at $t=\frac{4+\sqrt{6}}{20}$). Since $\sigma(P)\geq 2.6$, any tiling with 4 tiles of edge length $s_B$ cannot be optimal. The situation is even worse if the tiling has more than 4 tiles of edge length $s_B$.
\[exactly3\] In an optimal tiling, there are exactly 3 tiles with edge lengths $s_B$.
By Lemma \[atmost3\] we need to show that there are at least 3 tiles with edge lengths $s_B$. We already know that there are 2 tiles, $B$ and $B'$, with $s_{B'}=s_B$. Suppose there are no other tiles with edge length $s_B$; we will derive a contradiction.
Recall that $B$ can be assumed to be a right coastal tile with a corner at $(1,0)$ and that $B'$ can be assumed to have a corner at $(0,1)$. Thus we have the following configuration.
0.5mm
(50.00,50.00)(0,0)
(0.00,0.00)[(1,0)[50.00]{}]{} (0.00,0.00)[(0,1)[50.00]{}]{} (50.00,0.00)[(0,1)[50.00]{}]{} (0.00,50.00)[(1,0)[50.00]{}]{} (0.00,30.00)[(1,0)[20.00]{}]{} (0,30.00)[(0,1)[20.00]{}]{} (20.00,30.00)[(0,1)[20.00]{}]{} (0.00,50.00)[(1,0)[20.00]{}]{} (30.00,0.00)[(1,0)[20.00]{}]{} (30.00,0.00)[(0,1)[20.00]{}]{} (50.00,0.00)[(0,1)[20.00]{}]{} (30.00,20.00)[(1,0)[20.00]{}]{} (0.00,0.00)[(1,0)[30.00]{}]{} (0.00,0.00)[(0,1)[30.00]{}]{} (30.00,0.00)[(0,1)[30.00]{}]{} (0.00,30.00)[(1,0)[30.00]{}]{} (15.00,15.00)[(0,0)\[cc\][A]{}]{}
(10.00,40.00)[(0,0)\[cc\][B$'$]{}]{}
(40.00,10.00)[(0,0)\[cc\][B]{}]{}
If $s_A=s_B=1/2$, then the remaining empty square of size $1/2$-by-$1/2$ needs to be tiled by 5 squares. This is impossible. It follows that $s_A>1/2$ (and so $s_B<1/2$).
Let $C$ denote the tile with a corner at $(1,1)$.
0.5mm
(50.00,50.00)(0,0)
(0.00,0.00)[(1,0)[50.00]{}]{} (0.00,0.00)[(0,1)[50.00]{}]{} (50.00,0.00)[(0,1)[50.00]{}]{} (0.00,50.00)[(1,0)[50.00]{}]{} (35.00,35.00)[(1,0)[15.00]{}]{} (35.00,35.00)[(0,1)[15.00]{}]{} (50.00,35.00)[(0,1)[15.00]{}]{} (35.00,50.00)[(1,0)[15.00]{}]{} (0.00,30.00)[(1,0)[20.00]{}]{} (0,30.00)[(0,1)[20.00]{}]{} (20.00,30.00)[(0,1)[20.00]{}]{} (0.00,50.00)[(1,0)[20.00]{}]{} (30.00,0.00)[(1,0)[20.00]{}]{} (30.00,0.00)[(0,1)[20.00]{}]{} (50.00,0.00)[(0,1)[20.00]{}]{} (30.00,20.00)[(1,0)[20.00]{}]{} (0.00,0.00)[(1,0)[30.00]{}]{} (0.00,0.00)[(0,1)[30.00]{}]{} (30.00,0.00)[(0,1)[30.00]{}]{} (0.00,30.00)[(1,0)[30.00]{}]{} (15.00,15.00)[(0,0)\[cc\][A]{}]{}
(42.50,42.50)[(0,0)\[cc\][C]{}]{}
(10.00,40.00)[(0,0)\[cc\][B$'$]{}]{}
(40.00,10.00)[(0,0)\[cc\][B]{}]{}
There are 4 tiles that remain to be placed. At least 2 must share a border on the line $y=s_B$ with $B$ (if there were only 1, then it must have edge length $s_B$); similarly, at least 2 must share a border on the line $x=s_B$ with $B'$. Thus there are exactly 2 tiles on top of $B$: one a right coastal tile (call it $E$) and one an inland tile, with a corner at $(s_A, s_B)$ (call it $D$). Similarly, there are 2 tiles to the right of $B'$, one on the north border (call it $E'$) and one with a corner at $(s_B,s_A)$ (call it $D'$).
Note that $s_E=s_{E'}=1-s_B-s_C$; also, $s_D=s_{D'}=s_B-s_E$. Thus the tiling is symmetric with respect to the main diagonal $y=x$.
Now consider the line connecting the northwest corner of $A$ to the southeast corner of $C$. Since this is a diagonal line, it must intersect the interior of a tile, either $D$ or $E$ or $D'$ or $E'$. But the symmetry of the tiling indicates that the aforementioned line must intersect the interior of *two* tiles, contradicting our requirement that the tiles don’t overlap.
We note here a by-product of the proof: it cannot be the case that $s_A=s_B=1/2$. Thus we have $s_A>s_B$.
Denote by $B''$ the third tile whose side length is equal to $s_B$. As above, we can assume that $B''$ lies adjacent to $B'$.
$\psi_3(8)=\frac{13}{5}$.
Recall that $C$ is the tile with a corner at $(1,1)$. We have a configuration similar to the following.
0.5mm
(50.00,50.00)(0,0)
(0.00,0.00)[(1,0)[50.00]{}]{} (0.00,0.00)[(0,1)[50.00]{}]{} (50.00,0.00)[(0,1)[50.00]{}]{} (0.00,50.00)[(1,0)[50.00]{}]{} (43.00,43.00)[(1,0)[7.00]{}]{} (43.00,43.00)[(0,1)[7.00]{}]{} (50.00,43.00)[(0,1)[7.00]{}]{} (43.00,50.00)[(1,0)[7.00]{}]{} (0.00,30.00)[(1,0)[20.00]{}]{} (0.00,30.00)[(0,1)[20.00]{}]{} (20.00,30.00)[(0,1)[20.00]{}]{} (0.00,50.00)[(1,0)[20.00]{}]{} (20.00,30.00)[(1,0)[20.00]{}]{} (20.00,30.00)[(0,1)[20.00]{}]{} (40.00,30.00)[(0,1)[20.00]{}]{} (20.00,50.00)[(1,0)[20.00]{}]{} (30.00,0.00)[(1,0)[20.00]{}]{} (30.00,0.00)[(0,1)[20.00]{}]{} (50.00,0.00)[(0,1)[20.00]{}]{} (30.00,20.00)[(1,0)[20.00]{}]{} (0.00,0.00)[(1,0)[30.00]{}]{} (0.00,0.00)[(0,1)[30.00]{}]{} (30.00,0.00)[(0,1)[30.00]{}]{} (0.00,30.00)[(1,0)[30.00]{}]{} (15.00,15.00)[(0,0)\[cc\][A]{}]{}
(46.50,46.50)[(0,0)\[cc\]]{}
(40.00,10.00)[(0,0)\[cc\][B]{}]{}
(10.00,40.00)[(0,0)\[cc\][B$'$]{}]{}
(30.00,40.00)[(0,0)\[cc\][B$''$]{}]{}
Three tiles remain to be placed. Two of them are right coastal tiles and one—the one with a corner at $(s_A,s_B)$, which as before we call $D$—is an inland tile. Thus there are two inland tiles, $B''$ and $D$, and the total edge length of this tiling is $2+s_{B}+s_D$.
If $s_D>\frac12 s_B$, then the two right coastal tiles besides $B$ and $C$ must each have edge length $s_B-s_D
<\frac12 s_B$; thus $\sigma({\mathcal{C}}_1)<
s_B+\frac12 s_B+\frac12 s_B + s_C=
2s_B+s_C$. But looking at the north border we see that $2s_B+s_C\leq 1$, so $\sigma({\mathcal{C}}_1)<1$, a contradiction. Thus we must have $s_D\leq \frac12s_B$, so the total length of the tiling is at most $2+s_B+\frac12 s_B=2+\frac32 s_B$. Therefore $2+\frac32 s_B\geq \frac{13}{5}$, i.e., $s_B\geq \frac25$.
Now consider just the tiles $A$, $B$, $B'$, and $B''$. Let $t=s_B$ as before. These tiles have total area $3t^2+(1-t)^2$, leaving an area of $2t-4t^2$ to be covered with 4 tiles. By Lemma \[CS\], the total edge lengths of these 4 tiles is at most $\sqrt{4(2t-4t^2)}$; thus the total edge length of all tiles is at most $3t+(1-t)+\sqrt{4(2t-4t^2)}$. For $t\geq \frac25$, this function has a maximum value of $\frac{13}{5}$ (which occurs at $t=\frac25$); thus $\sigma({\mathcal{C}})\leq \frac{13}{5}$, as required.
**Note**: I do not know the value of $\psi_3(k^2-1)$ for $k>3$. It is possible to show that $\psi_3(k^2-1)\geq
k-\frac{1}{k-1}$ as follows. Start with a standard $(k+1)\times (k+1)$ tiling, and replace a $k\times k$ subsquare with a standard $(k-1)\times (k-1)$ tiling. We now have a tiling with $(k+1)^2-k^2+(k-1)^2 = k^2+2$ tiles. There are (a) $2k+1$ tiles with edge length $1/(k+1)$, and (b) $(k-1)^2$ tiles with edge length $\frac{k}{k^2-1}$. Pick any $2\times 2$ subsquare in (b) and replace it with one big square; we now have a tiling with $k^2-1$ tiles. The total edge length of this tiling is $\frac{2k+1}{k+1} + \frac{k(k-1)^2}{k^2-1} - \frac{2k}{k^2-1}
= k-\frac{1}{k-1}$.
[9]{}
P. Erd[ő]{}s, Some of my favorite problems in number theory. Comb. Week. Resenhas 2 (1995), no. 2, 165–186.
P. Erd[ő]{}s and A. Soifer, Squares in a Square. Geombinatorics, vol. IV, issue 4 (1995), 110–114.
W. Staton and B. Tyler, On the Erdös Square-Packing Conjecture. Geombinatorics, vol. XVII, issue 2 (2007), 88–94.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The Cygnus region is a very bright and complex portion of the TeV sky, host to unidentified sources and a diffuse excess with respect to conventional cosmic-ray propagation models. Two of the brightest TeV sources, MGRO J2019+37 and MGRO J2031+41, are analyzed using Milagro data with a new technique, and their emission is tested under two different spectral assumptions: a power law and a power law with an exponential cutoff. The new analysis technique is based on an energy estimator that uses the fraction of photomultiplier tubes in the observatory that detect the extensive air shower. The photon spectrum is measured in the range 1 to 200 TeV using the last 3 years of Milagro data (2005-2008), with the detector in its final configuration. MGRO J2019+37 is detected with a significance of 12.3 standard deviations ($\sigma$), and is better fit by a power law with an exponential cutoff than by a simple power law, with a probability $>98$% (F-test). The best-fitting parameters for the power law with exponential cutoff model are a normalization at 10 TeV of $7^{+5}_{-2}\times10^{-10}$ $\mathrm{s^{-1}\: m^{-2}\: TeV^{-1}}$, a spectral index of $2.0^{+0.5}_{-1.0}$ and a cutoff energy of $29^{+50}_{-16}$ TeV. MGRO J2031+41 is detected with a significance of 7.3$\sigma$, with no evidence of a cutoff. The best-fitting parameters for a power law are a normalization of $2.4^{+0.6}_{-0.5}\times10^{-10}$ $\mathrm{s^{-1}\: m^{-2}\: TeV^{-1}}$ and a spectral index of $3.08^{+0.19}_{-0.17}$. The overall flux is subject to an $\sim$30% systematic uncertainty. The systematic uncertainty on the power law indices is $\sim$0.1. A comparison with previous results from TeV J2032+4130, MGRO J2031+41 and MGRO J2019+37 is also presented.'
author:
- 'A. A. Abdo, B. T. Allen, T. Aune, D. Berley, E. Bonamente, G. E. Christopher, T. DeYoung, B. L. Dingus, R. W. Ellsworth, J. G. Galbraith-Frew, M. M. Gonzalez, J. A. Goodman, C. M. Hoffman, P. H. Hüntemeyer, B. E. Kolterman, J. T. Linnemann, J. E. McEnery, A. I. Mincer, T. Morgan, P. Nemethy, J. Pretz, J. M. Ryan, P. M. Saz Parkinson, G. Sinnis, A. J. Smith, V. Vasileiou, G. P. Walker, D. A. Williams and G. B. Yodh'
title: 'Spectrum and Morphology of the Two Brightest Milagro Sources in the Cygnus Region: MGRO J2019+37 and MGRO J2031+41.'
---
Introduction
============
The Cygnus region is a part of our Galaxy of active massive star formation and destruction, and has been studied over a broad range of wavelengths, including radio, X-ray, GeV and TeV gamma-ray, as well as in cosmic rays. From GeV up to multi-TeV energies the Cygnus region is the brightest diffuse gamma-ray source in the northern hemisphere [@egret1997].
One of the challenges in analyzing the Cygnus region at TeV energies, is the proper separation of the gamma-ray flux that is attributed to the point or extended sources in the region or to the diffuse emission. Previous Milagro analyses computed the diffuse emission from the region using two alternative methods to isolate the contribution from the resolved sources [@milagro2007a; @milagro2008], and found that at TeV energies the flux is still in excess with respect to the predicted flux from the GALPROP model [@galprop]. Milagro also published the discovery of two unidentified TeV sources in the region, MGRO J2019+37 and MGRO J2031+41 [@milagro2007b; @milagro2009]. The location of MGRO J2019+37 was found to be consistent with two EGRET sources, while the best fit position for MGRO J2031+41 was near two EGRET sources and the HEGRA unidentified source TeV J2032+4130. In a correlation study connecting the TeV sources discovered by Milagro with sources detected above 10$\sigma$ (the so-called Bright Source List, BSL) by the Fermi Large Area Telescope (LAT), the two aforementioned brightest Milagro sources in the region were associated with GeV pulsars [@fermi2009c; @milagro2009]. MGRO J2019+37 and MGRO J2031+41 are currently associated with two and one pulsars identified by the Fermi LAT, respectively [@fermi2011c]. Milagro also detected two candidate sources in the Cygnus region ($l\in[65^\circ, 85^\circ]$ and $b\in[-2^\circ,+2^\circ]$) [@milagro2009].
Recently, several collaborations have presented new surveys of the Cygnus region, confirming the complexity of the region and showing the highly structured diffuse emission. The TeV emission from the position of MGRO J2019+37 has been confirmed by the VERITAS experiment [@veritas2011]. The VERITAS spatial counts map shows a clear structure associated with MGRO J2019+37, a region of extended emission which seems to be produced by previously unresolved sources. At lower energies, the Fermi LAT collaboration has also recently published new results on the diffuse emission from the Cygnus region. An extended source, the so-called Cocoon, overlapping with MGRO J2031+41, has been found and its emission has been explained by freshly accelerated cosmic rays, trapped in a shell of photon-dominated emission formed by stellar winds and supernovae, as shown by mid-infrared observations [@fermi2011b]. The spectrum from this region is hard, extending up to 100 GeV with no evidence of softening, and could be explained as the result of pulsar-accelerated particles or as an active super-bubble. The average diffuse emission from the region has also been analyzed at MeV to GeV energies [@fermi2011a]. Despite the very rich source population, this emission is similar to that of the local interstellar space, and there is no evidence that it is necessary to include an extra contribution in the model, resembling the diffuse excess previously measured by Milagro [@milagro2007a; @milagro2008]. Most recently, the ARGO-YBJ collaboration presented the results of a survey of the Cygnus region in the energy range of 600 GeV to 7 TeV. MGRO J2031+41 is observed with a significance of 6.4$\sigma$ and a flux consistent with previous Milagro results, but there is no significant detection of MGRO J2019+37 [@argo2012].
Here, we present a new analysis of the last three years of data collected with the Milagro experiment (2005-2008). An improved gamma-hadron separation and a newly-developed technique are exploited to reconstruct the energy spectra of gamma rays from the sources in the Cygnus region. The best fits for the spectra of the two brightest Milagro sources are presented: MGRO J2019+37, a source with a post-trials significance in excess of 12$\sigma$ between 1 and 100 TeV, and MGRO J2031+41 a source with a post-trials significance in excess of 7$\sigma$ [@milagro2007b]. We compare our spectra with results from the HEGRA, MAGIC, Whipple and ARGO-YBJ experiments [@hegra2005; @magic2008; @whipple2004; @argo2012].
Analysis Technique {#s.analysis}
==================
The Milagro detector was located in the Jemez Mountains in New Mexico, at an altitude of 2,630 m a.l.s. It was operated from 2001 to 2008, and, in its final configuration, consisted of two components: (1) a central pond ($60\times80$ $\mathrm{m^2}$, 8 m deep) with two photomultiplier tube (PMT) layers, a shallow [*air shower*]{} layer consisting of 450 PMTs, and a deep [*muon layer*]{} consisting of 273 PMTs and (2) an array of 175 single-PMT tank detectors surrounding the pond covering an area of 40,000 $\mathrm{m^2}$ [@milagro2001; @milagro2003].
A detailed description of the analysis method and the parameters used here can be found in the Milagro paper on the spectral measurement of the Crab Nebula [@milagro2011]. Systematic uncertainties of this method are estimated to be $\sim$30% on the overall flux and $\sim$0.1 on the spectral index.
The new element in this analysis, with respect to the previous approach [@milagro2007b; @milagro2009], is the introduction of an estimator, $\mathcal{F}$, for the energy of the primary particle initiating the extensive air shower, based on the number of PMTs hit for each event. The $\mathcal{F}$ parameter is defined as the sum of two fractions: $$\mathcal{F}=\frac{N_{A.S.}}{N_{A.S.}^{live}}+\frac{N_{T.A.}}{N_{T.A.}^{live}},
\label{F}$$ where $N_{A.S.}$ and $N_{T.A.}$ are the number of PMTs detecting the event, while $N^{live}_{A.S.}$ and $N^{live}_{T.A.}$ are the number of functional PMTs in the air shower layer and in the tank array, respectively. Because the typical energy resolution of Milagro is quite broad (50-100%), and the median energy associated with a given $\mathcal{F}$ value is dependent on the spectral assumption for the observed source, the fit is performed in the measured $\mathcal{F}$ space rather than in the energy space.
The reconstructed Milagro data contain information about the direction and $\mathcal{F}$ value of air shower events. According to their $\mathcal{F}$ value, events are sorted into 9 bins ($0.2\le \mathcal{F} \le 2$), resulting in a total of 9 signal skymaps, binned in $0.1^{\circ}\times0.1^{\circ}$ pixels. For each $\mathcal{F}$ bin, background maps are calculated using the Direct Integration method [@milagro2011]. The two brightest regions of interest in the sky, a $2^{\circ}\times2^{\circ}$ area around the position of the Crab Nebula and a band around the Galactic plane ($-2.5^{\circ}<b<2.5^{\circ}$) are excluded when calculating the background [@milagro2011]. Rather than discriminating between gamma-ray and cosmic-ray initiated air showers, a weight is applied to all measured events, in both signal and background maps, where gamma-ray like events are assigned a higher weight than cosmic-ray like events [@milagro2011]. The gamma-ray excess with respect to the estimated background is calculated as the difference between signal weights and background weights, and is filled into the so-called excess map. Based on these excess maps we compute the $\mathcal{F}$ distributions used in the energy fit.
Excess maps are smoothed according to the detector angular resolution (or point spread function, PSF), which is a function of $\mathcal{F}$ and varies between $0.3^{\circ}$ and $0.7^{\circ}$. The measured $\mathcal{F}$ distribution for the target source in a $0.1^{\circ}\times0.1^{\circ}$ bin, as a result, is the average excess coming from a PSF-wide region around the nominal source position. Since the background is measured with the direct integration method removing the Galactic plane ($b>2.5^{\circ}$ and $b<-2.5^{\circ}$), the measured excess from sources in the Galactic plane includes any other diffuse or extended emission possibly present in the vicinity of the source. The Galactic diffuse background was estimated to contribute up to 15% of the flux at 35 TeV for the weakest Galactic BSL source [@milagro2009].
We test the emission from the two sources for two different spectral hypotheses: a power law, defined as $$\frac{dN}{dE}=N_{10\mathrm{TeV}}\left( \frac{E}{10\;\mathrm{TeV}}\right)^{-\alpha},
\label{e.PL}$$ where $N_{10\mathrm{TeV}}$ is the normalization scale set at 10 TeV and $\alpha$ the spectral index, and a power law with an exponential cutoff, defined as $$\frac{dN}{dE}=N_{10\mathrm{TeV}}\left( \frac{E}{10\;\mathrm{TeV}}\right)^{-\alpha}\mathrm{exp}\left( -\frac{E}{E_c}\right),
\label{e.CO}$$ where $E_c$ is the cutoff energy.
The fit to the data is performed comparing the measured excess to simulations. First, a set of simulated data is generated varying sensitive parameters (i.e. $N_{10\mathrm{TeV}}$, $\alpha$ and $E_c$). The best spectral parameters and the corresponding fit probability are then found using a $\chi^2$ minimization, comparing the measured and the simulated $\mathcal{F}$ distributions. Uncertainties on fit parameters are computed using 1-$\sigma$ contours of $\chi^{2}$ histograms (as discussed in sections \[s.2019\] and \[s.2031\]). The uncertainty is defined as the distance between the best fit value and the lower and upper edges of the 1-$\sigma$ contour.
Results and Discussion {#s.results}
======================
The results presented here are obtained from the analysis of the Milagro data taken with the detector in its final configuration. The start date is 2005-10-22 20:39:16 GMT and the stop date is 2008-04-15 00:02:53 GMT, corresponding to a total of 906 days of observation, resulting in 832 integrated days after data quality cuts.
The positions of MGRO J2019+37 and MGRO J2031+41 are obtained fitting Milagro excess maps with a 2-dimensional Gaussian function with equal widths in right ascension and declination ($\sigma_{RA}=\sigma_{Dec}$), plus a constant base to take into account contribution from diffuse emission. The fit is performed in a square region of $\pm 3^{\circ}$ around the excess. Table \[t.positions\] shows the positions used in this paper and a comparison to other surveys.
MGRO J2019+37 spectrum {#s.2019}
----------------------
In Fig. \[f.2019frasor\] the $\mathcal{F}$ distribution of data is shown with the two simulated $\mathcal{F}$ distributions for the best fitting power law and power law with exponential cutoff models. These distributions show the weighted excess coming from the source position after background subtraction. The excess from MGRO J2019+37 is compared to the excess measured from the entire Cygnus region ($65^{\circ}<l< 85{^\circ}$ and $-2^{\circ}<b<2^{\circ}$). We find that the emission from MGRO J2019+37 accounts for $\sim$12$\%$ of the excess from the entire region.
Figs. \[f.2019contPL\] and \[f.2019contCO\] show the 1 and 2-$\sigma$ contours from the $\chi^2$ fit for the two tested hypotheses. 1-$\sigma$ contours are used to compute uncertainty on the parameters as described in section \[s.analysis\].
The best fit spectra for the two tested hypotheses (Eqs. \[e.PL\] and \[e.CO\]) are shown in Fig. \[f.2019spectra\]. The best-fit spectral parameters are summarized in Table \[t.results\].
With the power law hypothesis, the best-fit parameters for the normalization and the spectral index are $N_{10\mathrm{TeV}} = (4.1\pm0.6) \cdot 10^{-10}$ $\mathrm{s^{-1}\: m^{-2}\: TeV^{-1}}$ and $\alpha=2.78\pm0.10$ respectively. The $\chi^{2}$ is 16.12 for 7 degrees of freedom (d.o.f.), which gives a $\chi^{2}$ fit probability of 0.024.
In the case of the power law with cutoff hypothesis, the best fit parameters are $N_{10\mathrm{TeV}}=7^{+5}_{-2}\cdot10^{-10}$ $\mathrm{s^{-1}\: m^{-2}\: TeV^{-1}}$, $\alpha=2.0^{+0.5}_{-1.0}$ and $E_c=29^{+50}_{-16}$ TeV. The $\chi^{2}$ is 1.924 (6 d.o.f.), which gives a $\chi^{2}$ fit probability of 0.93.
The $\chi^{2}$ fit probabilities suggest that the power law with a cutoff model ($\chi^{2}$ probability = 93%) fits the data better than a simple power law ($\chi^{2}$ probability = 2.4%). We use the F-test [@ftest] to compute if the improvement in the fit is significant. For MGRO J2019+37, the simple power law is disfavoured at a C.L. $>98$%.
Previous Milagro analyses quoted the flux of MGRO J2019+37 at 20 and 35 TeV respectively [@milagro2007b; @milagro2009]. Those values, using a different analysis technique, are in agreement with the results presented here. An independent analysis [@allen2007], which used a different energy estimator for the particle initiating the air shower and a different parameter to distinguish between gamma and cosmic rays, also is consistent with our results. ARGO-YBJ [@argo2012] has not observed any significant emission from MGRO J2019+37 in the range 600 GeV to 7 TeV. The explanations put forward in [@argo2012] are an insufficient exposure above 5 TeV or a possible time variability of MGRO J2019+37. The 90% C.L. upper limit from ARGO-YBJ is consistent with our best fitting model, a power law with an exponential cutoff (see Fig. \[f.2019spectra\]). On the other hand, the simple power law model ($\alpha=2.78\pm0.10$), already disfavored by the F-test, does not agree with the ARGO-YBJ results.
![MGRO J2019+37: distributions of the parameter used to estimate the photon energy, $\mathcal{F}$. The spectrum fit is performed in the $\mathcal{F}$ space. Black squares represent the data. Red triangles and blue circles represent the simulated distributions for the best fit assuming a power law with a cutoff and a simple power law, respectively. The unit on the y axis is the weighted excess per day, the unit on the x axis is the $\mathcal{F}$ value. \[f.2019frasor\] ](J2019frasor.eps){width="8cm"}
![MGRO J2019+37: 1 and 2-$\sigma$ C.L. contours for the power law fit. The point indicates the best fit result (minimum $\chi^2$). The $\chi^2$ increments for the 1 and 2-$\sigma$ contours are 2.30 and 6.18 respectively. \[f.2019contPL\]](J2019contoursPL.eps){width="5cm"}
![MGRO J2019+37: 1 and 2-$\sigma$ C.L. contours for the power law with cutoff fit. *left panel* - $N_{10\mathrm{TeV}}$ vs $\alpha$ projection. *right panel* - $E_c$ vs $\alpha$ projection. The points indicate the best fit result (minimum $\chi^2$). The $\chi^2$ increments for the 1 and 2-$\sigma$ contours are 3.53 and 8.02 respectively. \[f.2019contCO\]](J2019contoursCO.eps){width="7.5cm"}
![MGRO J2019+37: energy spectra. The best fit is obtained for a power law with cutoff model (in red). The power law model is also shown (in blue). The shadowed area represents the 1-$\sigma$ band, obtained by varying the parameters within the 1-$\sigma$ contour. ARGO-YBJ 90% C.L. upper limits for MGRO J2019+37 are shown in black [@argo2012]. \[f.2019spectra\] ](J2019spectra.eps){width="8cm"}
MGRO J2031+41 spectrum {#s.2031}
----------------------
The $\mathcal{F}$ distribution from data and those simulated for the two best fit models are shown in Fig. \[f.2031frasor\]. MGRO J2031+41 accounts for $\sim$6$\%$ of the excess from the entire Cygnus region.
Fig. \[f.2031contPL\] and \[f.2031contCO\] show the 1 and 2-$\sigma$ contours from the $\chi^2$ fit for the two tested hypotheses.
Best-fit spectral parameters are summarized in Table \[t.results\].
For the power law hypothesis, the best fit parameters are $N_{10\mathrm{TeV}} = 2.4^{+0.6}_{-0.5} \cdot 10^{-10}$ $\mathrm{s^{-1}\: m^{-2}\: TeV^{-1}}$ and $\alpha=3.08^{+0.19}_{-0.17}$. The $\chi^{2}$ is 6.98 (7 d.o.f.), with a $\chi^{2}$ fit probability of 0.43.
For the power law with cutoff hypothesis, the best fit parameters are $N_{10\mathrm{TeV}}=3.7^{+90.0}_{-1.9}\cdot10^{-10}$ $\mathrm{s^{-1}\: m^{-2}\: TeV^{-1}}$, $\alpha=2.7^{+0.5}_{-2.6}$ and $E_c=44^{+\infty}_{-40}$ TeV. In this case we are not able to constrain the upper limit of the 1-$\sigma$ uncertainty for the cutoff energy (see right panel of Fig. \[f.2031contCO\]). The maximum $E_c$ simulated value is 1000 TeV, almost an order of magnitude above the highest detected energy. The $\chi^{2}$ is 5.80 (6 d.o.f.), with a $\chi^{2}$ fit probability of 0.45. The best fit spectra for the two models are shown in Fig. \[f.2031spectra\].
The improvement in the fit obtained with the more complex model is not significant, according to the F-test, and the simple power law is to be preferred over the power law with the exponential cutoff.
The flux of MGRO J2031+41 at 20 and 35 TeV, measured by previous Milagro analyses [@milagro2007b; @milagro2009], is in agreement with the results presented here. Results from ARGO-YBJ, MAGIC, HEGRA and Whipple [@argo2012; @magic2008; @hegra2005; @whipple2004] are shown in Fig. \[f.2031spectra\]. The HEGRA and MAGIC models are mutually consistent, but they disagree with our best fit in terms of integral flux in the overlapping energy range. Between 1 and 10 TeV the flux measured by MAGIC accounts for only $\sim$3% of the flux measured by Milagro. The spectrum as measured by the air Cherenkov telescopes (ACTs) is much harder ($\alpha\sim$2) than the Milagro power law best fit ($\alpha\sim$3). This discrepancy can be explained by the following two facts. First, the angular resolution of HEGRA and MAGIC ($<0.1^{\circ}$) is much better than that of Milagro ($0.3^{\circ}$ to $0.7^{\circ}$). The Whipple flux was measured at 0.6 TeV, with a PSF of $0.21^{\circ}$, and it agrees better with the extrapolation of the Milagro result at lower energies. The same is true for ARGO-YBJ, which, with an angular resolution between $0.47^{\circ}$ and $2.8^{\circ}$, measured an emission consistent with the results presented here. The measurement presented in this paper accounts for photons coming from a larger region around the nominal source position compared to ACTs. Second, the way the background is computed is also different. Whipple used the ON/OFF method and observed the source with a significance of 3.3$\sigma$, while MAGIC primarily was operated in wobble mode (5.6$\sigma$). A possible explanation of the presented results is that TeV J2032+4130, whose extension is slightly larger than the HEGRA and MAGIC angular resolution, is surrounded by an extended emission, $\sim$$4^{\circ}\times3^{\circ}$ according to the Milagro map (see Fig. \[f.maps\] discussed in Sec. \[s.morphology\]). Therefore, Milagro, Whipple and ARGO-YBJ are not able to disentangle the extended emission from the central source and observe a higher flux. MAGIC and HEGRA, on the other hand, take into account the extended emission as a background. As a result, the ACTs measure a fainter emission and a harder spectrum for TeV J2032+4130.
![MGRO J2031+41: distributions of the parameter used to estimate the photon energy, $\mathcal{F}$. The spectrum fit is performed in the $\mathcal{F}$ space. Black squares represent the data. Red triangles and blue circles represent the simulated distributions for the best fit assuming a power law with a cutoff and a simple power law, respectively. The unit on the y axis is the weighted excess per day, the unit on the x axis is the $\mathcal{F}$ value. \[f.2031frasor\] ](J2031frasor.eps){width="8cm"}
![MGRO J2031+41: 1 and 2-$\sigma$ C.L. contours for the power law fit. The point indicates the best fit result (minimum $\chi^2$). \[f.2031contPL\]](J2031contoursPL.eps){width="5cm"}
![MGRO J2031+41: 1 and 2-$\sigma$ C.L. contours for the power law with cutoff fit. *left panel* - $N_{10\mathrm{TeV}}$ vs $\alpha$ projection. *right panel* - $E_c$ vs $\alpha$ projection. The points indicate the best fit result (minimum $\chi^2$). The right panel shows that the present analysis is unable to constrain the upper limit of the energy cutoff within 1$\sigma$. \[f.2031contCO\]](J2031contoursCO.eps){width="7.5cm"}
![MGRO J2031+41: energy spectra. The power law model is shown in blue and the power law with cutoff model is shown in red. These two hypotheses give the same $\chi^{2}$ fit probability. The pink cross is the Whipple flux at 0.6 TeV [@whipple2004]. The fine-dashed cyan line and the dashed yellow line are the HEGRA and MAGIC best fits [@hegra2005; @magic2008], respectively. The dot-dash black line is the ARGO-YBJ best fit [@argo2012]. The shadowed area represents the 1-$\sigma$ band. \[f.2031spectra\] ](J2031spectra.eps){width="8cm"}
Morphology {#s.morphology}
----------
The significance map of the Cygnus region for the last three years of Milagro data (2005-2008) is shown in Fig. \[f.maps\]. An energy-dependent PSF smoothing is applied. The top panel shows the entire region in Galactic coordinates. Milagro positions for MGRO J2019+37 and MGRO J2031+41 are shown, as well as the high-significance Fermi LAT sources ($>10\sigma$) and the extended Fermi Cocoon.
The Fermi counts map for photon energies above 10 GeV overlaid with the Milagro significance contours in the Cygnus region is shown in Fig. \[f.fermi\]. Milagro detects an excess consistent with the position of the two pulsars PSR J2021+3651 and PSR J2032+4127, associated with MGRO J2019+37 and J2031+41 respectively, but no detection of significant emission coincident with PSR J2021+4026.
MGRO J2019+37 is observed with a significance of 12.3$\sigma$. Its extension is obtained with the 2-dimensional Gaussian fit discussed in Sec. \[s.results\]. The fit results in $\sigma=0.75^{\circ}$, slightly larger then the angular resolution of the detector in its final configuration. There is evidence from VERITAS that the extended source Milagro observes is the result of the superimposition of close point-like sources [@veritas2011].
MGRO J2031+41, observed with a significance of 7.3$\sigma$, is an extended source, with $\sigma=1.1^{\circ}$. It shows a central core with a higher significance ($>6\sigma$) surrounded by a broader region (approximately $3^{\circ}\times4^{\circ}$) with a lower significance ($>4\sigma$). TeV J2032+4130 lies within the central core, and the Fermi Cocoon 1-$\sigma$ radius ($\sigma=2^{\circ}$) completely includes MGRO J2031+41. However, the Fermi analysis [@fermi2011b] shows that the Cocoon has an elongated shape, poorly correlating with the Milagro contours, and the possible association between the Cocoon and the extended Milagro excess remains unclear.
The spectral fit results do not change if the maximum significance position is chosen instead of the 2-D Gaussian fit position.
{width="10cm"}
{width="13.cm"}
Conclusions {#s.conclusions}
===========
We present the spectra of the two brightest Milagro sources in the Cygnus region using a new analysis technique applied to the last 3 years of data collected by the Milagro experiment.
MGRO J2019+37 is observed with a significance over 12$\sigma$. Its emission is well fitted by a power law with an exponential cutoff ($E_c=29^{+50}_{-16}$ TeV) and a hard asymptotic spectral index ($\alpha=2.0^{+0.5}_{-1.0}$). The simple power law hypothesis is disfavoured at a C.L. $>98$%. The TeV excess measured by Milagro from MGRO J2019+37, spatially associated with the Fermi LAT pulsars J2018.0+3626 and J2021.0+3651, has been confirmed by VERITAS, and it is likely produced by nearby unresolved sources. ARGO-YBJ does not detect a significant emission from MGRO J2019+37, but the 90% C.L. upper limits do not conflict with the Milagro best fitting model.
The emission from MGRO J2031+41 (7.3$\sigma$ significance), is well modeled by a power law with $\alpha=3.08^{+0.19}_{-0.17}$. Our result, in particular the integral flux in the overlapping energy range, is consistent with previous measurements by ARGO-YBJ and Whipple, but it disagrees with HEGRA and MAGIC results for TeV J2032+4130, most likely because of the different PSF of the instruments and different background subtraction methods. MGRO J2031+41 appears to show a structured morphology, produced by the superimposition of a central point-like source, coincident with TeV J2032+4130, and an extended emission, possibly produced either by unresolved sources or interactions of cosmic rays with the local interstellar medium. The correlation between the TeV extended emission and the overlapping Fermi Cocoon is unclear and needs further studies.
HAWC, the next generation water Cherenkov observatory, will be able to produce a more accurate analysis of the TeV emission from the Cygnus region, with its improved sensitivity (10-15 times better than Milagro) and angular resolution.
We gratefully acknowledge Scott Delay and Michael Schneider for their dedicated efforts in the construction and maintenance of the Milagro experiment. This work has been supported by the National Science Foundation (under grants PHY-0245234, -0302000, -0400424, -0504201, -0601080, -1002445, and ATM-0002744), the US Department of Energy (Office of High-Energy Physics and Office of Nuclear Physics), Los Alamos National Laboratory, the University of California, and the Institute of Geophysics and Planetary Physics.
Abdo, A. A. et al., 2007, , 658, L33 Abdo, A. A. et al., 2007a, , 664, L91 Abdo, A. A. et al., 2008, , 688, 1078 Abdo, A. A. et al., 2009, , 700, L127 Abdo, A. A. et al., 2009a, Science, 326, 1512 Abdo, A. A. et al., 2009b, Physical Review Letters, 103, 251101 Abdo, A. A. et al., 2009c, , 183, 46 Abdo, A. A. et al., 2011, arXiv:1110.0409v1 Abdo, A. A. et al., 2011a, arXiv:1110.6123v1 Abdo, A. A. et al., 2011b, Science, 334, 1103 Abdo, A. A. et al., 2011c, arXiv:1108.1435v1 Aharonian, F. et al., 2005, , 431, 197 Albert, J. et al., 2008, , 675, L25 Aliu, E. for the VERITAS Collaboration, 2011, in Proc. of 32nd ICRC, arxiv:1110.4656v2 Allen, B. 2007, PhD thesis, University of California, Irvine Atkins, R. et al., 2001, in Proc. 27th ICRC, arXiv:astro-ph/0110513 Atkins, R. et al., 2003, , 595, 803 Bartoli, B. et al., 2012, , 754, L22 Bevington R. P. and Robinson D. K., 2003, ISBN: 0-07-247227-8 GALPROP, http://galprop.stanford.edu Hunter, S. D. et al., 1997, , 481, 205 Lang, M. J. et al., 2004, A&A, 423, 415 LAT Data Server, http://fermi.gsfc.nasa.gov/ssc/data/access/
[lccc]{} MGRO J2019+37 &&&\
2-D Gaussian with base & 304.646$\pm$0.008 & 36.847$\pm$0.006 & 0.75\
Fermi position (0FGL J2020.8+3649) & 305.22 & 36.83 & —\
Milagro 2007 & 304.98 & 36.66 & 1.1\
\
MGRO J2031+41 &&&\
2-D Gaussian with base & 307.41$\pm$0.02 & 41.190$\pm$0.015 & 1.1\
Fermi position (0FGL J2032.2+4122) & 308.06 & 41.38 & —\
Milagro 2007 & 307.98 & 41.51 & 3.0\
\[t.positions\]
[lcccccccccc]{} MGRO J2019+37 & $4.1\pm0.6$ & $2.78\pm0.10$ & $\infty$ & 16.12/7 & $0.024$\
MGRO J2019+37 & $7^{+5}_{-2}$ & $2.0^{+0.5}_{-1.0}$ & $29^{+50}_{-16}$ & 1.924/6 & $0.93$\
\
MGRO J2031+41 & $2.4^{+0.6}_{-0.5}$ & $3.08^{+0.19}_{-0.17}$ & $\infty$ & 6.980/7 & $0.43$\
MGRO J2031+41 & $3.7^{+90.0}_{-1.9}$ & $2.7^{+0.5}_{-2.6}$ & $44^{+\infty}_{-40}$ & 5.796/6 & $0.45$\
\[t.results\]
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'J. Jurčák'
- 'N. Bello González'
- 'R. Schlichenmaier'
- 'R. Rezaei'
bibliography:
- 'manuscript.bib'
date: 'Received 11 December, 2014; accepted '
subtitle: 'II. Formation of a penumbra at the expense of a pore'
title: A distinct magnetic property of the inner penumbral boundary
---
[We recently presented evidence that stable umbra-penumbra boundaries are characterised by a distinct canonical value of the vertical component of the magnetic field, $B^{\rm stable}_{\rm ver}$. In order to trigger the formation of a penumbra, large inclinations in the magnetic field are necessary. In sunspots, the penumbra develops and establishes by colonising both umbral areas and granulation, that is, penumbral magneto-convection takes over in umbral regions with $B_{\rm ver} < B^{\rm stable}_{\rm ver}$, as well as in granular convective areas. Eventually, a stable umbra-penumbra boundary settles at $B^{\rm stable}_{\rm ver}$.]{} [Here, we aim to study the development of a penumbra initiated at the boundary of a pore, where the penumbra colonises the entire pore ultimately.]{} [We have used Hinode/SOT G-band images to study the evolution of the penumbra. Hinode/SOT spectropolarimetric data were used to infer the magnetic field properties in the studied region.]{} [The penumbra forms at the boundary of a pore located close to the polarity inversion line of NOAA10960. As the penumbral bright grains protrude into the pore, the magnetic flux in the forming penumbra increases at the expense of the pore magnetic flux. Consequently, the pore disappears completely giving rise to an orphan penumbra. At all times, the vertical component of the magnetic field in the pore is smaller than $B^{\rm stable}_{\rm ver} \approx 1.8$ kG.]{} [Our findings are in an agreement with the need of $B^{\rm stable}_{\rm ver}$ for establishing a stable umbra-penumbra boundary: while $B_{\rm ver}$ in the pore is smaller than $B^{\rm stable}_{\rm ver}$, the protrusion of penumbral grains into the pore area is not blocked, a stable pore-penumbra boundary does not establish, and the pore is fully overtaken by the penumbral magneto-convective mode. This scenario could also be one of the mechanisms giving rise to orphan penumbrae.]{}
Introduction
============
Penumbra formation is not yet fully understood. @Rucklidge:1995 showed that the efficiency of energy transport across the magnetopause increases dramatically when the magnetic field inclination exceeds the critical value of 45$^\circ$ ($\gamma_{\rm crit}$). Despite the simplicity of the model used, the $\gamma_{\rm crit}$ of 45$^\circ$ for the penumbral formation was confirmed by recent MHD simulations of @Rempel:2009b. @jurcak:2014a confirmed observationally the importance of magnetic field inclination in triggering penumbra formation.
The magnetic field inclination on a protospot boundary is related to the total magnetic flux as the magnetic field inclination increases with the increasing flux. According to theoretical prediction of @Rucklidge:1995, the necessary magnetic flux for a penumbra formation is between $1-7 \times 10^{20}$ Mx. These values are in agreement with observed values of $5 \times 10^{20}$ Mx [@Zwaan:1987], $1 \times 10^{20}$ Mx [rudimentary penumbra, @Leka:1998], and $4 \times 10^{20}$ Mx [protospot with forming penumbral segments, @Rezaei:2012].
Observations of orphan penumbrae [@Zirin:1991; @Kuckein:2012; @Lim:2013; @jurcak:2014; @Zuccarello:2014] show that they are comparable in all aspects to regular sunspot penumbrae. These observations prove that sufficient magnetic flux is not a necessary condition for a penumbra formation. A favourable configuration of magnetic field strength and inclination can result in the penumbra formation.
Once the penumbra formation is triggered, the penumbra extends mostly at the expense of granular regions. @Jurcak:2015 show that the inner penumbral end extends also to the umbral area where the umbra-penumbra (UP) boundary settles at a distinct canonical value of the vertical component of the magnetic field, $B^{\rm stable}_{\rm ver}$ of 1.8 kG. This confirms the results of @Jurcak:2011 who found constant values of $B_{\rm ver}$ along UP boundaries of stable sunspots, where the actual value of $B^{\rm stable}_{\rm ver}$ seems to be weakly dependent on the sunspot size. Larger samples of sunspots must be investigated to clarify whether or not such a dependence is real.
{width="0.95\linewidth"}
In this paper, we present the Hinode observations of a small pore near the polarity inversion line of AR 10960. We follow the formation and development of a penumbra forming at the expense of the magnetic flux of the pore while colonising it. Six spectro-polarimetric scans of the region are used to infer the maps of the magnetic field vector. Details of the studied data are described in Sect. \[observations\] along with the analysis methods. We describe and discuss the results in Sect. \[results\], and summarise them in Sect. \[discussion\].
Observations and data analysis {#observations}
==============================
Our analysis is based on data observed with the Solar Optical Telescope [SOT, @Tsuneta:2008] aboard the Hinode satellite [@Kosugi:2007]. We made use of spectro-polarimetric (SP) data in the lines at 630.15 and 630.25 nm, and images taken through a G-band filter centred at 430.5 nm with a bandpass of 0.8 nm.
We study the evolution of a pore in AR 10960 between 19:00 UT on 6 June 2007 and 8:00 UT on 8 June 2007. The pore was located 6$^\circ$S and between 14$^\circ$E and 6$^\circ$W during the analysed time period. The cadence of G-band images vary between 60 sec and 100 sec (see Fig. \[pore\_area\]). At all times, the spatial sampling is 011 resulting in a diffraction-limited spatial resolution of 022. During the analysed time period, six Hinode/SP scans of the region were acquired. The SP observations were taken in a fast mode with a pixel sampling of 032 and a noise level of $10^{-3}I_{\rm c}$. These data were calibrated using the available routines in the Hinode SolarSoft package.
In order to retrieve the physical parameters, the SP data were inverted with the SIR code [Stokes Inversion based on Response function, @Cobo:1992]. We used a simple atmosphere model supposing all free parameters of the inversion, except for temperature, to be constant with height. We took into account the spectral point spread function of the Hinode SP in the inversion process. We assumed the magnetic filling factor to be unity and assumed no stray light. The macro-turbulence was set to zero while micro-turbulence was a free parameter of the inversion. We applied the code AMBIG [@Leka:2009] to solve the 180$^\circ$ ambiguity of the LOS azimuth. The disambiguated LOS vector magnetic field was then transformed to the local reference frame (LRF) with the help of routines from the AZAM code [@Lites:1995].
The apparent motions of photospheric structures were determined from the G-band images using local correlation tracking [LCT, @November:1988] with a Gaussian tracking window of FWHM 1. We first aligned the images and removed the p-mode oscillations by applying a $k-\omega$ filter with a cut-off of 5 km s$^{-1}$.
Results
=======
![Evolution of the pore area with $I_\mathrm{c} < 0.5 \times I^\mathrm{QS}_\mathrm{c}$ (black $+$ symbols). Blue $\triangle$ symbols connected by the dashed line show the total positive magnetic flux ($\Phi$) in the pore. Red $\Box$ symbols connected by the dash-doted line show $\Phi$ in the penumbra. The horizontal blue lines at the top of the plot indicate the times of G-band observations, where the light blue correspond to the imaging cadence of 60 sec. The vertical red lines correspond to the times of SP scans and the vertical black lines labelled a–f show the times of images shown in Figs. \[G-band\], \[lct\], and \[sp\_maps\].[]{data-label="pore_area"}](figure_2.eps){width="\linewidth"}
{width="12cm"}
{width="12cm"}
In Fig. \[G-band\], we show the pore and penumbra evolution in AR 10960 at selected stages. A video attached to Fig. \[G-band\] is available in the online edition. At the beginning of the studied period (a), the pore with a small light bridge and no penumbra is observed. Approximately two hours later (b), a penumbra segment west from the pore can be observed; the penumbral filaments, which are not directly adjacent to the pore, have south-north (S-N, cf. Fig. \[G-band\]f) orientation (heads of the filaments are located on the southern end). The animation on the evolution shows that this penumbral segment in not coupled to the investigated pore. Later (c), new penumbral filaments appear with their heads - penumbral grains - connected to the pore boundary. Initially, these filaments have a N-S orientation (c), and at later stages (d, e, f), they show an E-W orientation.
The evolution of the pore area (pixels with $I_\mathrm{c} < 0.5 \times I^\mathrm{QS}_\mathrm{c}$) shown in Fig. \[G-band\] is shown in Fig. \[pore\_area\]. The pore area increases until 4:00 UT on 7 June (Fig. \[G-band\]c). Afterwards, the pore area decreases. The evolution of the pore size is closely related to the evolution of penumbral filaments that are directly adjacent to the pore. These filaments appear around 2:00 UT on 7 June (see the animation), and we immediately observe proper motions of penumbral bright grains into the pore [as observed in stable and forming penumbrae, @Wang:1992; @Sobotka:1999; @Marquez:2006; @Jurcak:2015]. In Fig. \[lct\], we show the proper motions in the close surrounding of the studied pore at the beginning of the penumbra formation (left) and towards the end of the pore lifetime (right). A characteristic property of a penumbra is clearly depicted: the apparent inward (outward) motion of the inner (outer) penumbra.
In Fig. \[sp\_maps\], we show maps of the magnetic field configuration in the area of the forming penumbra for the three SP scans that captured the evolving pore-penumbra system. The first SP scan of the region where the penumbra is attached to the pore is shown in the upper row (c). The penumbral filaments are directed away from the polarity inversion line and the field lines (shown in the inclination map) do not connect the pore-penumbra system with the pores of opposite polarity located north and west of it. Fig. 4, middle row (e), shows the magnetic field configuration around 17:10 UT. At that time, a transient second penumbral segment develops oriented along the polarity inversion line, partially connected by the field lines to the west pore of opposite polarity around (12, 12). Yet, three hours later (bottom row of Fig. 4, f), the penumbral filaments are no longer oriented westwards and the magnetic field creating the penumbra is not longer connected with the west pore of opposite polarity.
At all times, the pore-penumbra system under study belongs to the same polarity in this bipolar region. The positive magnetic flux in the penumbra increases in time (Fig. 2), while the magnetic flux in the pore diminishes. This indicates that the penumbra grows at the expense of the pore magnetic flux as signatures of emerging flux are not seen during the relevant time span. The maximum flux found in the penumbra reaches 84% of the maximum flux in the pore, that is, even if some magnetic flux of the pore was cancelled out by the nearby regions of opposite polarity, the majority is transformed into the penumbral magnetic flux. Therefore, the fundamental cause of the pore disappearance is its transformation into the penumbra.
The process of pore disappearance is not straightforward. At some stages there are almost no regions with $I_\mathrm{c} < 0.5 \times I^\mathrm{QS}_\mathrm{c}$ (Fig. \[G-band\]d), while at later times these regions revive again (Fig. \[G-band\]e). During the whole sequence, we observe proper motions of penumbral bright grains into the pore (Fig. \[lct\]). As shown in @Jurcak:2015, such motions ‘displace’ the UP boundary towards the umbral core in the case of a forming sunspot penumbra, until the B$_{ver}$ reaches the B$_{ver}^{stable}$ value. Analogously, the area of the studied pore decreases at the expense of the evolving penumbra. Since the strongest vertical component of the magnetic field reaches only 1.4 kG in the studied pore, the stable boundary between the pore and the forming penumbra cannot establish and the penumbral magneto-convective mode takes over in the pore area.
The pore disappearance at the expense of the evolving penumbra can be also understood as a mechanism for an orphan penumbra formation. The predominantly vertical magnetic field of the pore becomes horizontal and we observe an orphan penumbra for a certain time before its disappearance. These horizontal fields creating the orphan penumbra submerge in time, possibly due to downward pumping as described by @Tobias:1998. This is an alternative to the formation of an orphan penumbra by a flux emergence that is blocked by an overlaying magnetic field [@Zuccarello:2014] or by the relation of an orphan penumbra to an active region filament [@Kuckein:2012; @Buehler:2016].
Conclusions {#discussion}
===========
We describe the evolution of a penumbra at the boundary of a small pore. The penumbra eventually colonises the pore area leading to its extinction. We link these observations with previous studies that found constant values of the vertical component of magnetic field strength on UP boundaries of stable sunspots [@Jurcak:2011] and how this canonical $B_{\rm ver}^{\rm stable}$ value establishes a demarcation line in the forming UP boundary [@Jurcak:2015].
The studied pore has the vertical component of the magnetic field around 1.4 kG at its maximum (Fig. \[sp\_maps\]) which is smaller than the $B_{\rm ver}^{\rm stable}$ of about 1.8 kG [@Jurcak:2011; @Jurcak:2015]. Therefore, the protrusion of penumbral magneto-convection into the pore area is not hindered, a stable pore-penumbra boundary does not establish, and the pore is eventually colonised by the penumbra. As the penumbral bright grains protrude into the pore, the magnetic flux of the pore is transformed into penumbral magnetic flux, and the predominantly vertical field becomes horizontal. This scenario describes a mechanism by which an orphan penumbra forms.
The support from GA CR 14-04338S and RVO:67985815 is gratefully acknowledged. N.B.G. acknowledges financial support by the Senatsausschuss of the Leibniz-Gemeinschaft, Ref.-No. SAW-2012-KIS-5. R.R. acknowledges financial support by the DFG grant RE 3282/1-1 and by the Spanish Ministry of Economy and Competitiveness through project AYA2014-60476-P. Hinode is a Japanese mission developed and launched by ISAS/JAXA, with NAOJ as domestic partner and NASA and STFC (UK) as international partners. It is operated by these agencies in cooperation with ESA and NSC (Norway).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'This work addresses the identification of zonal flows in fusion plasmas. Zonal flows are large scale phenomena, hence multipoint measurements taken at remote locations are required for their identification. Given such data, the Biorthogonal Decomposition (or Singular Value Decomposition) is capable of extracting the globally correlated component of the multipoint fluctuations. By using a novel quadrature technique based on the Hilbert transform, propagating global modes (such as MHD modes) can be distinguished from the non-propagating, synchronous (zonal flow-like) global component. The combination of these techniques with further information such as the spectrogram and the spatial structure then allows an unambiguous identification of the zonal flow component of the fluctuations. The technique is tested using gyro-kinetic simulations. The first unambiguous identification of a zonal flow at the TJ-II stellarator is presented, based on multipoint Langmuir probe measurements.'
address: 'Laboratorio Nacional de Fusi[ó]{}n, CIEMAT, Avda. Complutense 40, 28040 Madrid, Spain'
author:
- 'B.Ph. van Milligen, E. Sánchez, A. Alonso, M.A. Pedrosa, C. Hidalgo, A. Martín de Aguilera, A. López Fraguas'
title: 'The use of the Biorthogonal Decomposition for the identification of zonal flows at TJ-II'
---
Introduction
============
Spontaneous confinement transitions occurring in magnetically confined plasmas are vital for the development of fusion as a viable future energy source. It is generally assumed that Zonal Flows are an essential ingredient for understanding these transitions [@Wagner:2007]. These flows can be generated spontaneously from turbulence via Reynolds stress [@Burrell:1997]. However, the experimental identification of zonal flows is a hard problem.
Zonal flows are low-frequency phenomena with a global character, i.e., with long wavelengths in the toroidal and poloidal directions [@Calvo:2009]. More specifically, they are associated with a potential perturbation with toroidal mode number $n=0$, poloidal mode number $m = 0$ and finite radial wavelength $k_r$. The detection of these flows has been difficult, partly due to the fact that fusion-grade plasmas are characterized by the presence of many instabilities and waves, making it hard to isolate the low-frequency, long-wavelength zonal flow in the data, containing other fluctuating and/or oscillating contributions [@Fujisawa:2007b; @Fujisawa:2009]. Several techniques have been [used]{}, including the measurement of spectra and (long-range) correlation [@McKee:2003; @Pedrosa:2005; @Pedrosa:2005c; @Bencze:2006; @Zhao:2006; @Pedrosa:2007; @Pedrosa:2008; @Xu:2009; @Liu:2009], the Hilbert-Huang transform [@Carreras:2011], bicoherence [@Diamond:2000; @Tynan:2001; @Milligen:2008c] and non-linear energy transfer [@Xu:2012].
The fact that zonal flows are ‘global’ in the sense that they are long-wavelength phenomena should facilitate their unambiguous detection provided simultaneous measurements at several remote points inside the plasma are available, at the (radial) location where the zonal flow exists.
Given such multipoint data, a technique is needed to isolate and extract the ‘global’ component from the fluctuation data. This task is ideally suited for the BiOrthogonal Decomposition (BOD) [@Dudok:1994], also known as Proper Orthogonal Decomposition (POD) or Singular Value Decomposition (SVD), among others [@Kerschen:2005]. However, there may be other global oscillations affecting the fluctuation data, different from Zonal Flows. Therefore, in the present work, the BOD technique is complemented with additional techniques to quantify the long-range character of the biorthogonal modes (cf. [@Alonso:2012]) or their propagating nature, thus facilitating the distinction between global modes associated with Magneto-HydroDynamic (MHD) activity or other oscillations, on the one hand, and proper Zonal Flows, on the other.
The biorthogonal decomposition {#Method}
==============================
The details of the biorthogonal decomposition are well-known, and the reader is referred to the references for detailed information [@Dudok:1994; @Kerschen:2005]. Here, we will summarize its main features.
The multipoint measurements constitute a data matrix $Y(i,j)$, where the index $i = 1,...,N$ labels the time and $j = 1,...,M$ the detector. Typically, the time corresponding to the time index, $t(i)$, is equally spaced, since measurements are typically taken at a fixed sampling rate, although this is not strictly necessary. The physical location of the detectors, $\vec x(j)$, however, is often dictated by practical convenience or space limitations and will not necessarily correspond to a regular grid. Finally, to facilitate the physical interpretation of the results, the measurements $Y$ performed at the various detectors $j$ should ideally be of the same physical quantity (in our case, an electric potential) and be given in the same units.
The BOD method decomposes the data matrix $Y(i,j)$ as follows: $$\label{biortho}
Y(i,j) = \sum_k \lambda_k \psi_k(i) \phi_k(j)$$ where $\psi_k$ is a ‘chrono’ (a temporal function) and $\phi_k$ a ‘topo’ (a spatial or detector-dependent function), such that the chronos and topos satisfy the following orthogonality relation [@Dudok:1994]: $$\label{norm}
\sum_i{\psi_k(i)\psi_l(i)} = \sum_j{\phi_k(j)\phi_l(j)} = \delta_{kl}.$$ An alternative normalization (indicated by the superscript $r$) is $$\label{norm_rms}
\frac{1}{N}\sum_i{\psi^r_k(i)\psi^r_l(i)} = \frac{1}{M}\sum_j{\phi^r_k(j)\phi^r_l(j)} = \delta_{kl}.$$ such that the root-mean-square (RMS) value of the topos and chronos equals 1. In this case, $$\label{biortho_rms}
Y(i,j) = \sum_k \lambda^r_k \psi^r_k(i) \phi^r_k(j)$$ with $\lambda^r_k = \lambda_k/\sqrt{NM}$. With this normalization, the number $\lambda_k^r$ represents the contribution of mode $k$ to the total RMS ‘fluctuation amplitude’. Thus, the fractional contribution of a given mode $k$ to the total ‘fluctuation energy’ (proportional to the square of the RMS) can be computed from $$\label{mode_energy}
e_k^2 = \frac{\lambda_k^2}{ \sum_{k'} \lambda_{k'}^2 }$$ The superscript ‘$r$’ is irrelevant in this expression.
The combination chrono/topo at a given $k$, $\psi_k(i) \phi_k(j)$, is called a spatio-temporal ‘mode’ of the fluctuating system, and is constructed from the data matrix without any prejudice regarding the mode shape.
The decomposition is performed by computing the Singular Value Decomposition (SVD) of the data matrix $Y(i,j)$. Note that this is always possible for any real-valued rectangular matrix $Y$, and standard software packages are available for this purpose. The $\lambda_k \ge 0$ are the eigenvalues (sorted in decreasing order), where $k=1,...,\min(N,M)$.
A threshold may be set for cutting off the expansion in BOD modes (based on, e.g., a noise level) and so keeping only the dominant contribution of modes to the data matrix for further analysis and ignoring noisy or minor contributions. When the BOD expansion is cut off at a certain mode $k_{\rm max}$, the reconstructed data, $Y_{\rm rec}$, are defined by Eq. (\[biortho\]) while restricting the sum to $k \le k_{\rm max}$. In this case, the reconstructed data $Y_{\rm rec}$ minimize the least squares error with respect to the actual data $Y$ among all possible sets of $k_{\rm max}$ spatiotemporal modes.
If the physical system under study contains normal mode oscillations and is sufficiently well-sampled (spatially and temporally), the probability that the BOD modes correspond closely to the said normal modes is high; the BOD is particularly sensitive to resonant linear normal modes [@Kerschen:2005]. On the other hand, the BOD technique is perhaps less suited for the analysis of non-linear systems in which no normal modes exist, as the BOD modes are unlikely to correspond to any meaningful physical modes in this case. An advantage of the BOD analysis is that no [*a priori*]{} assumption is made regarding the mode shape or spectral properties, unlike standard analysis techniques such as Fourier decomposition. This has the mentioned advantage that the BOD modes concentrate a maximum of fluctuation power in the lowest modes, but the disadvantage that the modes are not guaranteed to coincide exactly with the normal modes of the system (if known).
Furthermore, it should be noted that the SVD decomposition is not unique. First, the sign of the chrono and topo of a given mode $k$ can trivially be inverted without affecting their product, i.e., without affecting the reconstruction according to Eq. (\[biortho\]). Second, if two SVD modes $k_1$ and $k_2$ have the same eigenvalues, a rotation in the two-dimensional vector space spanned up by the chronos $k_1$ and $k_2$ and a compensating rotation in the space spanned up by the corresponding topos can be defined so that the reconstructed data, according to Eq. (\[biortho\]), are [unchanged]{}, giving rise to a (rotational) indeterminacy. As will be clarified below, this may occur in the case of propagating modes. As such, it does not imply a great disadvantage, since such modes will be considered pairwise anyway.
Covariance and the biorthogonal decomposition
---------------------------------------------
The covariance between a signal $Y(i,j_1)$ and a signal $Y(i,j_2)$ ($i$ being the temporal and $j$ the spatial index) is defined as: $$\label{covariance}
{\rm Cov}(j_1,j_2) = \frac{1}{N}\sum_i{Y(i,j_1)Y(i,j_2)}$$ Since the signals are expanded according to Eq. (\[biortho\]), this expression can be simplified, using Eq. (\[norm\]), to: $$\begin{aligned}
\label{covariance_topos}
{\rm Cov}(j_1,j_2) &=& \frac{1}{N} \sum_k{\lambda_k^2\phi_k(j_1)\phi_k(j_2)} \nonumber \\
&=& \sum_k{(\lambda_k^r)^2\phi_k^r(j_1)\phi_k^r(j_2)},\end{aligned}$$ i.e., the covariance between two signals is simply the sum of the products of the corresponding topos, weighed by the square of the eigenvalues. In other words, the topos reflect the covariance between the measurement signals; and the contribution of each mode $k$ to the covariance ${\rm Cov}(j_1,j_2)$ is simply $(\lambda_k^r)^2\phi_k^r(j_1)\phi_k^r(j_2)$.
From Eq. (\[covariance\_topos\]) one immediately obtains, again using Eq. (\[norm\]): $$\sum_{j_2}{{\rm Cov}(j_1,j_2)\phi_k(j_2)} = \frac{\lambda_k^2}{N}\phi_k(j_1).$$ Thus, the topos $\phi_k$ are the eigenvectors of the covariance matrix ${\rm Cov}$, and $\lambda_k^2/N$ the corresponding eigenvalues. This suggests that the topos $\phi_k$ and eigenvalues $\lambda_k$ can be found by computing the eigenvectors and eigenvalues of the covariance matrix ${\rm Cov}$, after which the corresponding chronos $\psi_k$ can be obtained by simple matrix multiplication [@Kerschen:2005]: $$\lambda_k\psi_k(i) = \sum_j{Y(i,j)\phi_k(j)}$$ This view of the BOD mode decomposition may be helpful when interpreting the meaning of the BOD modes.
The fact that the topos are the eigenvectors of the covariance matrix immediately suggests an alternative approach: instead of analyzing the raw data $Y(i,j)$, one could first normalize the data to their RMS value $$y(i,j) = \frac{Y(i,j)}{\sqrt{{\rm Cov}(j,j)}}$$ When computing the SVD of $y(i,j)$ instead of $Y(i,j)$, the resulting topos $\phi_k$ will then be the eigenvectors of the [*correlation*]{} matrix $C$.
Correlation
-----------
By definition, the correlation between measurements $j_1$ and $j_2$ is obtained by normalizing the covariance, Eq. (\[covariance\]), to the RMS of each signal: $$C(j_1,j_2) = \frac{{\rm Cov}(j_1,j_2)}{\sqrt{{\rm Cov}(j_1,j_1){\rm Cov}(j_2,j_2)}}$$ From Eq. (\[covariance\_topos\]), it is clear that this quantity only depends on the eigenvalues and the topos. The contribution of a given mode $k$ to the correlation is (cf. [@Alonso:2012]) $$\label{correlation}
C_k(j_1,j_2) = \frac{(\lambda_k^r)^2\phi_k^r(j_1)\phi_k^r(j_2)}{\sqrt{{\rm Cov}(j_1,j_1){\rm Cov}(j_2,j_2)}}$$
Thus, it is possible to quantify the contribution of a specific BOD mode $k$ to the so-called [*Long Range Correlation*]{} (LRC). To achieve this, assume we may subdivide the set of measurements $j=1,\dots,M$ into two complementary sets: $\{j \in S_1\}$, with $M_1$ elements, and $\{j \in S_2\}$, with $M_2$ elements, such that measurements in set $S_1$ are taken at a remote location from measurements in set $S_2$. The Long Range Correlation is then defined as the correlation between any pair of measurements taken at remote locations. We define the contribution to the [*mean*]{} Long Range Correlation of a given mode $k$ by averaging the correlation due to this mode among any two remote measurement points: $$\label{LRC}
C_k^{\rm LR} = \frac{1}{M_1M_2}\sum_{\{j_1 \in S_1\}}\sum_{\{j_2 \in S_2\}}{C_k(j_1,j_2)}$$ This definition may lead to low values in the case of partial cancellations due to the existence of a mixture of correlations and anti-correlations between individual measurements. To measure the overall correlation ‘intensity’ regardless of sign, we also define the mean absolute Long Range Correlation: $$\label{LRCabs}
C_k^{\rm LR, abs} = \frac{1}{M_1M_2}\sum_{\{j_1 \in S_1\}}\sum_{\{j_2 \in S_2\}}{\left | C_k(j_1,j_2) \right |}$$
Mode identification
-------------------
One important aspect of the BOD method is that it assumes that all ’modes’ are separable in space and time. Standing waves are of this type. However, one often has to contend with propagating structures, such as running waves. It is important to be able to recognize and distinguish these different structures.
A typical example of a running wave is $Y(t,x) = \cos(kx-\omega t)$. This can trivially be decomposed into a sum of two standing wave patterns: $$\cos(kx-\omega t) = \cos(kx)\cos(\omega t)-\sin(kx)\sin(\omega t)$$ By Fourier’s Theorem, this result can be generalized to any linearly propagating spatiotemporal structure of argument $kx-\omega t$. Thus, linearly propagating or ‘traveling’ waves or structures generate a pair of BOD modes, with similar eigenvalues and a mutual phase difference (in space and time) of around 90$^\circ$ [@Dudok:1994].
Consequently, a method is needed to identify such mode pairs. If the propagating mode has a clearly defined frequency $\omega$, the Fourier power spectra $|\hat \psi_k|^2$ of the concerned chronos will have a peak at this frequency (here, the circumflex indicates the Fourier transform), and the cross phase between the chronos $k_1,k_2$ is easily determined as the phase of the complex cross spectrum $\hat \psi_{k_1} \hat \psi_{k_2}^*$ at the frequency $\omega$, where the star signifies complex conjugate.
As the low-frequency modes of interest do not always have a clearly defined spectral peak, we will use a technique for quadrature detection based on the Hilbert transform [@Hahn:1996]. In this framework, we must assume that the BOD topos and chronos fluctuate symmetrically about zero, which can be achieved by subtracting their mean or applying a suitable high-pass frequency filter prior to analysis. This is necessary as it is a well-known fact that the Hilbert transform does not produce a reasonable estimate of the quadrature if the mean of the signal is not zero or is drifting [@Hahn:1996]. With this additional assumption, it is feasible to compute the quadrature of a given topo or chrono by means of the Hilbert transform.
Thus, we can use the full set of topos and chronos and compute the following spatial and temporal quadrature matrices: $$\begin{aligned}
Q_x(k_1,k_2) = \sum_j \tilde \phi_{k_1}(j) H(\tilde \phi_{k_2}(j))\nonumber \\
Q_t(k_1,k_2) = \sum_i \tilde \psi_{k_1}(i) H(\tilde \psi_{k_2}(i)) \end{aligned}$$ where $H$ is the Hilbert transform and the tilde refers to the removal of the mean (or lowest frequencies) mentioned above. The elements of these quadrature matrices are restricted to the range $[-1,1]$; the absolute value of a given element $Q_{x,t}(k_1,k_2)$ will differ significantly from 0 when the corresponding modes are in approximate quadrature. By means of this technique, it becomes possible to identify linearly propagating structures by finding pairs of modes with similar eigenvalues, $\lambda_{k_1} \simeq \lambda_{k_2}$, such that $Q_x(k_1,k_2)$ and $Q_t(k_1,k_2)$ are significantly different from zero. By default, modes that do not occur in pairs then correspond to standing wave structures.
Tests using gyrokinetic simulations {#Euterpe}
===================================
To test whether the analysis methods described in the preceding section are capable of delivering the promised results, in this section we will apply these techniques to gyrokinetic simulations with known properties, carried out with the code EUTERPE.
EUTERPE is a global gyrokinetic particle-in-cell Monte Carlo code developed originally at CRPP Lausanne [@Jost:2001] with the aim of simulating plasma turbulence in arbitrary threedimensional geometries. It is being developed at the Max Plank IPP (Greifswald, Germany) [@Kornilov:2005; @Kauffmann:2010; @Kleiber:2011; @Kleiber:2012] and is presently used at the National Fusion Laboratory (CIEMAT, Spain) for the simulation of TJ-II plasmas. For more [information]{} about the code the reader is referred to the cited references.
Several linear simulations have been run in TJ-II geometry: linear simulations of zonal flow relaxation [@Sanchez:2013] and simulations of Ion Temperature Gradient (ITG) instabilities [@Sanchez:2011]. Results from these two kinds of simulations are used in this work to test the zonal flow detection technique.
Simulation of Zonal Flow relaxation
-----------------------------------
In this kind of simulation, the relaxation of an initial zonal perturbation to the density is studied following the seminal work by Rosenbluth and Hinton [@Rosenbluth:1998]. The simulation is initiated with a perturbation to the density of the form $\delta n_i \propto n_0 \langle k_r^2 \rho_i^2/T_i \rangle \cos (4 \pi s)$, where $s$ is the normalized toroidal flux used as radial coordinate in EUTERPE, $n_0$ the equilibrium density, $k_r= 4 \pi |\nabla s |$ the radial wavenumber of the perturbation, $\rho_i=\sqrt{2m_i T_i }/eB$ the ion Larmor radius, $m_i$ and $e$ the ion mass and charge, respectively, $T_i$ the equilibrium ion temperature, $B$ the magnetic field and $\langle \rangle$ means flux surface average. The simulation is carried out in the standard magnetic configuration of TJ-II (labelled 100\_44\_64) under plasma conditions in which a neoclassical root confinement transition (the so-called low density transition) occurs at the plasma edge. The simulation corresponds to a numeric experiment in which this root transition is simulated by slowly evolving density and temperature profiles [@Velasco:2013]. The density and temperature profiles and also the neoclassical equilibrium electric field are included in the simulation.
The initial perturbation is allowed to evolve linearly without taking into account collisions and the plasma potential is monitored at several positions emulating several multi-pin Langmuir probes measuring plasma potential at a set of radial positions, as shown in the Fig. \[Euterpe\_probe\]: probe 1 is located at $\phi=0^\circ$ and probe 2 at half the period ($45^\circ$). The probe pins (30 for each probe) are ordered from large to small radius and cover most of the minor radius.
In this simulation, the zonal component of the potential ($m=0, n=0$) is observed to relax via a low frequency damped oscillation, typical for stellarators [@Mishchenko:2008], and also higher frequency oscillations identified as Geodesic Acoustic Modes (GAM) appear (see [@Sanchez:2013] and references therein). The frequency of the slow oscillation is in the range of 5–10 kHz and the GAM, only noticeable at the very beginning of the simulation, is around $50-100$ kHz. The GAM oscillation is associated with low-$m$ modes having much lower amplitude than the $m=0$ component. Raw simulation data are shown in Fig. \[Euterpe\_ZF\_data\].
The BOD eigenvalues and the long range correlation contribution of each mode, $C^{\rm LR}$, are shown in Fig. \[Euterpe\_ZF\_LRC\]. Clearly, nearly all LRC is concentrated in the first BOD mode.
Fig. \[Euterpe\_ZF\_topos\] shows the first four topos, Fig. \[Euterpe\_ZF\_chronos\] the first four chronos, and Fig. \[Euterpe\_ZF\_spectra\] the spectra of the first four chronos. Note that the spectrum of the first chrono exhibits the expected temporal behavior of a Zonal Flow: it peaks at very low frequency. Comparing the spectrum calculated for the whole time window and for $t\ge 0.1$ ms, it is clear that the peak at about 50 kHz corresponds mainly to the GAM appearing in the initial part of the time window ($t<0.1$ ms). In this respect, the quadrature analysis, shown in Fig. \[Euterpe\_ZF\_quadrature\], shows that modes 3 and 4, most clearly showing the mentioned spectral peak (Fig. \[Euterpe\_ZF\_spectra\]), form a propagating mode pair, as one might have suspected on the basis of their similar eigenvalues (Fig. \[Euterpe\_ZF\_LRC\]) and spectra. [This propagation would be [*radial*]{}, not poloidal/toroidal, in view of the near identity of the topos 3 and 4 in the two remote probes, which is consistent with the GAM nature of the fluctuations (constancy of potential on flux surfaces).]{}
Thus having identified modes 1 and modes 3+4, mode 2 remains. Note that this radially oscillating mode with very similar shape in both probes (Fig. \[Euterpe\_ZF\_topos\]) hardly oscillates in time but mainly decays in amplitude (Fig. \[Euterpe\_ZF\_chronos\]), while exhibiting a very similar spectrum as mode 1 for $t > 0.1$ ms (Fig. \[Euterpe\_ZF\_spectra\]).
We conclude that in this case, the Zonal Flow is captured by the first 2 BOD modes, with interesting properties: mode 1 has no radial sign changes and oscillates in time, while mode 2 has no temporal sign changes and oscillates in space. Note that in view of the orthogonality requirements of the BOD modes, Eq. (\[norm\]), there can be only one topo that does not change sign and only one chrono that does not change sign.
Fig. \[reconstruction\] shows a comparison between the $\phi_{00}$, the flux surface averaged component of the simulated potential, and the reconstructed potential oscillations using only the first and the first two BOD modes, respectively. Very close agreement is observed when both ZF BOD modes are included.
Linear simulation of an ITG instability
---------------------------------------
In this case, a linear simulation of ideal ITG instability is studied in the TJ-II standard configuration [@Sanchez:2011; @Mishchenko:2008].
The ion and electron density profiles and the electron temperature profiles are [ taken to be]{} flat while the ion temperature profile [ is defined to have a typical tanh shape, such that it]{} has a large gradient at [mid-radius]{}. This [renders]{} the ITG modes unstable in this region of the plasma. The electron temperature, which is the same as the ion temperature at the maximum gradient position, is $T_e=550$ eV. In this kind of simulation, the energy of the unstable modes increases with time, as no saturation mechanism is included in the simulation. The potential shows typical structures with resonant mode numbers ($m,n$) such that $n/m \approx \iota/2\pi$ and centered around $k_\theta \rho_i \approx \frac{m}{\langle r \rangle} \rho_i \approx 0.5$, $\langle r \rangle$ being the average minor radius of the flux surface at half radius. In this case, $15<m<30$ and $15<n<55$. The unstable modes propagate in the ion diamagnetic direction as corresponds to an ion drift wave. In this simulation, no zonal flow is generated by these modes, as it is linear and no mode interaction is taken into account. Raw simulation data of potential at the locations of the synthetic probes are shown in Fig. \[Euterpe\_ITG\_data\].
The BOD eigenvalues and the long range correlation contribution of each mode, $C^{\rm LR}$, are shown in Fig. \[Euterpe\_ITG\_LRC\]. By contrast with the preceding section, the LRC is very small here. On the other hand, the first two BOD eigenvalues have a similar amplitude, suggesting a possible mode pair (corresponding to a propagating mode).
Indeed, the quadrature analysis shown in Fig. \[Euterpe\_ITG\_QxQt\] confirms this suspicion: clearly, modes 1 and 2 form a (propagating) pair, as do modes 3 and 4.
Fig. \[Euterpe\_ITG\_topos\] shows the topos, reflecting the radial mode structure of these propagating modes.
\
Fig. \[Euterpe\_ITG\_chronos\] shows the corresponding chronos, which exhibit the expected exponential growth of the oscillation amplitude.
These observations are in accordance with the theoretical mode structure obtained from the distribution of the mode shapes and amplitudes in the simulation.
Application to TJ-II {#TJ-II}
====================
In this section, we will apply the techniques described in Section \[Method\] and tested on gyro-kinetic simulations in Section \[Euterpe\] to data obtained from a toroidally confined plasma in TJ-II. The goal is to see whether it is actually possible to identify a Zonal Flow with some degree of confidence, using appropriately located potential measurements.
TJ-II disposes of a double reciprocating probe system, allowing the simultaneous measurement of the floating potential on various radially separated probe pins on two probes with a large toroidal separation [@Pedrosa:1999]. Fluctuating structures with a high level of correlation between the toroidally separated probes (‘Long Range Correlation’ or LRC) can possibly be identified with Zonal Flow (ZF) structures [@Alonso:2012]. The radially spaced pins on each of the two probes provide information about the radial wave number of the detected structures.
However, Zonal Flows are not the only phenomena that can give rise to LRCs. In particular, it has been shown that other modes (e.g., drift waves or MagnetoHydroDynamic (MHD) oscillations) may also produce LRCs [@Milligen:2013b]. The question, therefore, is whether one can distinguish between different origins of observed LRCs: on the one hand, ZF-like behavior and on the other, MHD or other (rotating/propagating) oscillations. ZFs are electrostatic potential fluctuations having toroidal mode number $n=0$, poloidal wave number $m=0$, and a finite radial wavenumber [@Diamond:2005]. Due to this symmetry, the ZF potential structure may fluctuate – relatively slowly – but [*does not rotate*]{}. The latter property allows one to distinguish a ZF from, e.g., a low-frequency MHD mode, which (with very rare exceptions) usually does rotate.
The two techniques discussed in Section \[Method\] (i.e., the quantification of the contribution of each mode to LRCs and the quadrature detection technique) should enable one to make this distinction based on the experimental data: ZFs are long-range correlated ‘standing wave structures’, while MHD, rotating or propagating modes are ‘traveling wave structures’.
TJ-II is a Heliac type stellarator with four field periods, having major radius $R = 1.5$ m, minor radius $a < 0.2$ m, and toroidal magnetic field $B_0 < 1$ T. [The plasmas considered here]{} are heated by the electron cyclotron resonant heating (ECRH) system, consisting of two beam lines with an injected power of up to 400 kW each. Electron density is controlled by means of a gas puffing system and the line average electron density typically reaches values of about $\overline{n_e} \simeq 0.6 \cdot 10^{19}$ m$^{-3}$. The central electron temperature is typically $T_e(0) \simeq 300-800$ eV, and the central ion temperature [is]{} around $T_i(0) \simeq 120 - 150$ eV.
Langmuir probe D is located in a top port at toroidal position $\phi \simeq 35^\circ$, while probe B is located in a bottom port at $\phi \simeq 195^\circ$. Thus, the two probe systems are remote, separated toroidally by a distance of $\sim 5$ m (about half a toroidal turn) [@Pedrosa:2010]. In the experiments analyzed here, each probe is fitted with a ‘rake’ probe head measuring floating potential at $\sim 10$ radially spaced pins. Probe pins are separated radially by 3 mm in the D probe, and by 1.7 mm in the B probe; [probe D covers a radial range of about 3.5 cm, while probe B covers about 1.5 cm.]{}
We analyze shot 36012 in the time interval $1090 \le t \le 1250$ ms. [ In this discharge, the rotational transform, $\iota/2\pi$, has a value of around 1.455 at the magnetic axis and 1.55 at the edge, so that the 3/2 rational surface is located at about $\rho \simeq 0.73$. In the mentioned]{} time interval, both the line average electron density and the plasma energy content are fairly constant; $\overline{n_e} \simeq 0.63 \cdot 10^{19}$ m$^{-3}$, near the critical density of the electron to ion root confinement transition at TJ-II [@Hidalgo:2006b; @Velasco:2012]. [Fig. \[36012\_spectrum\] shows the mean spectrum of all probe pins. Several spectral peaks are visible, which we will identify in the following.]{}
Fig. \[36012\_lambda\] shows the BOD eigenvalues and the LRC. To calculate the LRC, Eqs. (\[LRC\]) and (\[LRCabs\]), the probe pins have been subdivided into two sets: namely, pins corresponding to probe D or B, respectively. Fig. \[36012\_spectra\] shows the spectra of the first 3 modes. The first BOD mode suggests a strong long range [*anti*]{}-correlation. The second and third BOD modes form a propagating pair (confirmed by quadrature analysis, not shown); the spectrum shows that this propagating mode has a clear frequency peak at $5-6$ kHz. Regarding the negative value of $C^{\rm LR}_1$, it should be noted that topo 1 exhibits a complex radial structure, cf. Fig. \[36012\_topos\]. Thus, the global character of the definition of $C^{\rm LR}$ may not capture the details of this structure. The two peaks appearing in topo 1 at $\rho \simeq 0.86$ are in fact [*correlated*]{} and correspond to a zonal flow-like structure, indicated schematically by a grey area in Fig. \[36012\_topos\]. This structure is considered zonal flow-like [for]{} the following reasons: (1) it is Long Range Correlated (recall Eq. (\[covariance\_topos\])); (2) the structure has the same value of floating potential in the two remote probes, suggesting an $m=0, n=0$ global structure: (3) the spectral characteristics of the structure show no clear peak and spectral power is concentrated at very low frequency, cf. Fig. \[36012\_spectra\]; (4) the radial position and radial width are very similar for the two remote probe systems (as indicated by the grey area in Fig. \[36012\_topos\]). The radial width of the ZF-like structure is rather small, namely about one centimeter. The four points mentioned above constitute the most unambiguous identification of a ZF-like structure reported yet in literature.
Further outward ([ $\rho \gtrsim 0.9$]{}), the floating potentials in the two probes are predominantly [*anti-*]{} correlated such that topo 1 has an opposite sign for the two probes. The temporal variation of this anti-correlated spatial structure is linked to the ZF fluctuations, as is evident from its inclusion in the same chrono. We speculate that the fluctuations of the ZF intensity (at $\rho \simeq 0.86$) cause a variation of outward transport (in the range [ $\rho \gtrsim 0.9$]{}) with some toroidal/poloidal asymmetry – probably associated with the different local curvature and flux expansion at the two probe locations.
At this point, we would like to emphasize that the radial structure revealed by this technique is very hard to extract using traditional methods based on, e.g., two-point correlation functions, due to the fact that the signals contain contributions from both the ZF-like structure and the propagating modes. Traditional techniques cannot separate these contributions clearly, unless a hypothesis is made about the spectral distribution and frequency filters are applied. Here, no hypotheses are needed and the results follow directly from the multipoint analysis of the raw data.
Discussion and conclusions
==========================
In this work, we propose a novel analysis method to identify ‘global’, long range correlated (LRC) components from multipoint fluctuation measurements. The analysis is based on the well-known Biorthogonal Decomposition (BOD), which is extended by defining a quantity that measures the contribution of each BOD mode to the overall LRC, cf. Eqs. (\[LRC\]-\[LRCabs\]). In addition, a quadrature detection technique is introduced, based on the Hilbert transform, to determine whether two BOD modes (with similar eigenvalues) are in quadrature and thus are likely to correspond to a propagating mode. The combination of these various items of information (mode contribution to LRC and quadrature among modes, as well as the spectral and spatial characteristics of the BOD modes) allows distinguishing Zonal Flow-like (‘global’), low frequency oscillations from propagating oscillations – provided sufficient and adequately placed multipoint data are available.
The method was tested on gyrokinetic simulations with the global code EUTERPE, using data from synthetic probes. Two simulations were analyzed, using synthetic probes to generate signals. In the first simulation, the linear, collisionless relaxation of an initial zonal perturbation to the density was studied. The BOD analysis successfully extracted the ZF mode from the generated signals. The GAM-like oscillations appearing at the beginning of the simulation were identified as associated to a “[radially]{} propagating structure”. In the second simulation, the linear ITG instability was simulated, and no ZF was produced. In this case, the BOD analysis technique detected propagating modes and yielded their structure, and did not detect any ZF (as indeed it shouldn’t).
Then, the method was applied to TJ-II Langmuir probe data. In a discharge near the electron-ion root confinement transition, the method was shown to be able to separate ZF-like fluctuations from propagating contributions to the LRC. Unambiguous identification of the ZF-like structure was possible, using four elements: (1) the structure exhibits (positive) Long Range Correlation; (2) it has the same value of floating potential in the two remote probes, suggesting an $m=0, n=0$ global structure; (3) the spectrum shows no clear peak and spectral power is concentrated at very low frequency; (4) the radial position and radial width are very similar in the two remote probes. By contrast, the propagating structure was recognized by the quadrature of two topos and chronos having similar BOD eigenvalues, and a peak in the corresponding spectra.
It is noted that with only two toroidally/poloidally separated multi-pin probes, as used in this work, the capacity to identify modes propagating in the various directions is somewhat limited. More sophisticated probes, providing more detailed spatial information in the toroidal/poloidal directions, or the combination of information from various different types of diagnostics might overcome such limitations and allow a better identification of propagating modes.
While bearing this in mind, it is shown that the BOD is a powerful technique to separate ZF-like, global oscillations from other (propagating) oscillations and to extract their mode structure. In this framework, it has been speculated that microscale turbulence might not only interact with Zonal Flows, but also with MHD modes [@Ishizawa:2007; @Ishizawa:2008]. This would imply the existence of an (indirect) interaction between ZFs and MHD modes. Indeed, in the past it has been observed that the growth of ZFs may modify MHD activity (e.g., [@Toi:1989]). It is expected that the methodology presented here could help clarifying this issue, or more generally, yield fruitful results for a wide range of multipoint (and remote) measurements in systems with long range correlated structures.
Acknowledgements {#acknowledgements .unnumbered}
================
[The authors thankfully acknowledge the computer resources, technical expertise and assistance provided by the Barcelona Supercomputing Center – Centro Nacional de Supercomputación and the CIEMAT Computing Center. ]{} Research sponsored in part by the Ministerio de Economía y Competitividad of Spain under project Nr. ENE2012-30832. This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement number 633053. The views and opinions expressed herein do not necessarily reflect those of the European Commission.
References {#references .unnumbered}
==========
[10]{}
F. Wagner. A quarter-century of [H]{}-mode studies. , 49:B1, 2007.
K.H. Burrell. Effects of [$E \times B$]{} velocity shear and magnetic shear on turbulence and transport in magnetic confinement devices. , 4:1499, 1997.
I. Calvo, B.A. Carreras, L. Garc[í]{}a, M.A. Pedrosa, and C. Hidalgo. Zonal flows and long-distance correlations during the formation of the edge shear layer in the [TJ-II]{} stellarator. , 51:065007, 2009.
A. Fujisawa, T. Ido, A. Shimizu, S. Okamura, K. Matsuoka, H. Iguchi, Y. Hamada, H. Nakano, S. Ohshima, K. Itoh, K. Hoshino, K. Shinohara, Y. Miura, Y. Nagashima, S.-I. Itoh, M. Shats, H. Xia, J.Q. Dong, L.W. Yan, K.J. Zhao, G.D. Conway, U. Stroth, A.V. Melnikov, L.G. Eliseev, S.E. Lysenko, S.V. Perfilov, C. Hidalgo, G.R. Tynan, C. Holland, P.H. Diamond, G.R. McKee, R.J. Fonck, D.K. Gupta, and P.M. Schoch. Experimental progress on zonal flow physics in toroidal plasmas. , 47(10):S718, 2007.
A. Fujisawa. A review of zonal flow experiments. , 49:013001, 2009.
[ G.R. McKee, R.J. Fonck, M. Jakubowski, K.H. Burrell, K. Hallatschek, R.A. Moyer, W. Nevins, D.L. Rudakov, and X. Xu. Observation and characterization of radially sheared zonal flows in [DIII-D]{}. , 45:A477, 2003. ]{}
M.A. Pedrosa, C. Hidalgo, E. Calder[ó]{}n, T. Estrada, A. Fern[á]{}ndez, J. Herranz, I. Pastor, and [the TJ-II Team]{}. Threshold for sheared flow and turbulence development in the [TJ-II]{} stellarator. , 47:777, 2005.
M.A. Pedrosa, C. Hidalgo, E. Calder[ó]{}n, A. Alonso, R.O. Orozco, J.L. de Pablos, [the TJ-II Team]{}, and P. Balan. Spontaneous edge [$E \times B$]{} sheared flow development studies in the [TJ-II]{} stellarator. , 55(12):1579, 2005.
A. Bencze, M. Berta, S. Zoletnik, J. Stockel, J. Ad[á]{}mek, and M. Hron. Observation of zonal flow-like structures using the autocorrelation-width technique. , 48:S137, 2006.
K.J. Zhao, T. Lan, J.Q. Dong, L.W. Yan, W.Y. Hong, C.X. Yu, A.D. Liu, J. Qian, J. Cheng, D.L. Yu, Q.W. Yang, X.T. Ding, Y. Liu, and C.H. Pan. Toroidal symmetry of the geodesic acoustic mode zonal flow in a tokamak plasma. , 96:255004, 2006.
M.A. Pedrosa, B.A. Carreras, C. Hidalgo, C. Silva, M. Hron, L. Garc[í]{}a, J.A. Alonso, I. Calvo, J.L. de Pablos, and J. St[ö]{}ckel. Sheared flows and turbulence in fusion plasmas. , 49:B303, 2007.
M.A. Pedrosa, C. Silva, C. Hidalgo, B.A. Carreras, R.O. Orozco, D. Carralero, and [the TJ-II Team]{}. Evidence of long-distance correlation of fluctuations during edge transitions to improved-confinement regimes in the [TJ-II]{} stellarator. , 100:215003, 2008.
Y. Xu, S. Jachmich, R.R. Weynants, M. Van Schoor, M. Vergote, A. Kr[ä]{}mer-Flecken, O. Schmitz, B. Unterberg, C. Hidalgo, and [TEXTOR Team]{}. Long-distance correlation and zonal flow structures induced by mean [$E \times B$]{} shear flows in the biasing [H]{}-mode at [TEXTOR]{}. , 16:110704, 2009.
[ A.D. Liu, T. Lan, C.X. Yu, H.L. Zhao, L.W. Yan, W.Y. Hong, J.Q. Dong, K.J. Zhao, J. Qian, J. Cheng, X.R. Duan, and Y. Liu. Characterizations of low-frequency zonal flow in the edge plasma of the [HL-2A]{} tokamak. , 103:095002, 2009. ]{}
B.A. Carreras, B.Ph. van Milligen, R.B. Perez, M.A. Pedrosa, C. Hidalgo, and C. Silva. Reconstruction of intermittent waveforms associated with the zonal flow at the transition leading to the edge shear flow layer. , 51:053022, 2011.
P.H. Diamond, M.N. Rosenbluth, E. S[á]{}nchez, C. Hidalgo, B. van Milligen, T. Estrada, B. Bra[ñ]{}as, M. Hirsch, H.J. Hartfuss, and B.A. Carreras. In search of the elusive zonal flow using cross-bicoherence analysis. , 84(21):4842, 2000.
G.R. Tynan, R.A. Moyer, M.J. Burin, and C. Holland. On the nonlinear turbulent dynamics of shear-flow decorrelation and zonal flow generation. , 8(6):2691, 2001.
B.Ph. van Milligen, T. Kalhoff, M.A. Pedrosa, and C. Hidalgo. Bicoherence during confinement transitions in the [TJ-II]{} stellarator. , 48:115003, 2008.
M. Xu, G. R. Tynan, P. H. Diamond, P. Manz, C. Holland, N. Fedorczak, S. Chakraborty Thakur, J. H. Yu, K. J. Zhao, J. Q. Dong, J. Cheng, W. Y. Hong, L. W. Yan, Q. W. Yang, X. M. Song, Y. Huang, L. Z. Cai, W. L. Zhong, Z. B. Shi, X. T. Ding, X. R. Duan, and Y. Liu. Frequency-resolved nonlinear turbulent energy transfer into zonal flows in strongly heated [L]{}-mode plasmas in the [HL-2A]{} tokamak. , 108:245001, 2012.
T. Dudok de Wit, A.-L. Pecquet, J.-C. Vallet, and R. Lima. The biorthogonal decomposition as a tool for investigating fluctuations in plasmas. , 1(10):3288, 1994.
G. Kerschen, J.-C. Golinval, A.F. Vakakis, and L.A. Bergman. The method of proper orthogonal decomposition for dynamical characterization and order reduction of mechanical systems: An overview. , 41:147, 2005.
J.A. Alonso, C. Hidalgo, M.A. Pedrosa, B.Ph. van Milligen, D. Carralero, and C. Silva. Dynamic transport regulation by zonal flow-like structures in the [TJ-II]{} stellarator. , 52:063010, 2012.
S.L. Hahn. . Artech House Signal Processing Library. Artech Print on Demand, 1996.
G. Jost, T.M. Tran, W.A. Cooper, L. Villard, and K. Appert. Global linear gyrokinetic simulations in quasi-symmetric configurations. , 8:3321, 2001.
V. Kornilov, R. Kleiber, and R. Hatzky. Gyrokinetic global electrostatic ion-temperature-gradient modes in finite $\beta$ equilibria of [Wendelstein 7-X]{}. , 45:238, 2005.
K. Kauffmann, R. Kleiber, R. Hatzky, and M. Borchardt. Global linear gyrokinetic simulations for [LHD]{} including collisions. , 260:012014, 2010.
R. Kleiber, R. Hatzky, A. K[ö]{}nies, K. Kauffmann, and P. Helander. An improved control-variate scheme for particle-in-cell simulations with collisions. , 182:1005, 2011.
R. Kleiber and R. Hatzky. A partly matrix-free solver for the gyrokinetic field equation in three-dimensional geometry. , 183(2):305, 2012.
E. S[á]{}nchez, R. Kleiber, R. Hatzky, M. Borchardt, P. Monreal, F. Castej[ó]{}n, A. L[ó]{}pez-Fraguas, X. S[á]{}ez, J.L. Velasco, I. Calvo, A. Alonso, and D. L[ó]{}pez-Bruna. Collisionless damping of flows in the tj-ii stellarator. , 55:014015, 2013.
E. S[á]{}nchez, R. KLeiber, R. Hatzky, M. Borchardt, P. Monreal, F. Castej[ó]{}n, A. Soba, Z. S[á]{}ez, and J.M. Cela. Simulaciones girocin[é]{}ticas de turbulencia en plasmas de fusi[ó]{}n con geometr[í]{}a tridimensional. In Real Sociedad Espa[ñ]{}ola de F[í]{}sica, editor, [*Proc. 33$^{\rm rd}$ Biennial Meeting of the Royal Spanish Physics Soc., Santander*]{}, volume IV, page 114, 2011.
M.N. Rosenbluth and F.L. Hinton. Poloidal flow driven by ion-temperature-gradient turbulence in tokamaks. , 80:724, 1998.
J.L. Velasco, J.A. Alonso, I. Calvo, J. Ar[é]{}valo, E. S[á]{}nchez, L. Eliseev, S. Perfilov, T. Estrada, A. L[ó]{}pez-Fraguas, C. Hidalgo, and [the TJ-II Team]{}. Damping of radial electric field fluctuations in the [TJ-II]{} stellarator. , 55:124044, 2013.
A. Mishchenko, P. Helander, and A. K[ö]{}nies. Collisionless dynamics of zonal flows in stellarator geometry. , 15:072309, 2008.
M.A. Pedrosa, A. L[ó]{}pez-S[á]{}nchez, C. Hidalgo, A. Montoro, A. Gabriel, J. Encabo, J. de la Gama, L.M. Mart[í]{}nez, E. S[á]{}nchez, R. P[é]{}rez, and C. Sierra. Fast movable remotely controlled [Langmuir]{} probe system. , 70(1):415, 1999.
B.Ph. van Milligen, A. Lopez Fraguas, M.A. Pedrosa, C. Hidalgo, A. Martin de Aguilera, and E. Ascas[í]{}bar. Parallel and perpendicular turbulence correlation length in the [TJ-II]{} stellarator. , 53:093025, 2013.
P.H. Diamond, S.-I. Itoh, K. Itoh, and T.S. Hahm. Zonal flows in plasma - a review. , 47(5):R35, 2005.
M.A. Pedrosa, C. Hidalgo, C. Silva, B.A. Carreras, D. Carralero, I. Calvo, and [the TJ-II Team]{}. Long-range correlations during plasma transitions in the [TJ-II]{} stellarator. , 50(6-7):507, 2010.
C. Hidalgo, M.A. Pedrosa, E. S[á]{}nchez, B. Gon[ç]{}alves, J.A. Alonso, E. Calder[ó]{}n, A.A. Chmyga, N.B. Dreval, L. Eliseev, T. Estrada, L. Krupnik, A.V. Melnikov, R.O. Orozco, J.L. de Pablos, and C. Silva. Physics of sheared flow development in the boundary of fusion plasmas. , 48:S169, 2006.
J.L. Velasco, J.A. Alonso, I. Calvo, and J. Ar[é]{}valo. Vanishing neoclassical viscosity and physics of the shear layer in stellarators. , 109(13):135003, 2012.
A. Ishizawa and N. Nakajima. Excitation of macromagnetohydrodynamic mode due to multiscale interaction in a quasi-steady equilibrium formed by a balance between microturbulence and zonal flow. , 14:040702, 2007.
A. Ishizawa and N. Nakajima. Effect of zonal flow caused by microturbulence on the double tearing mode. , 15:084504, 2008.
K. Toi, J. Gernhardt, O. Klüber, and M. Kornherr. Observation of precursor magnetic oscillations to the [H]{}-mode transition of the [ASDEX]{} tokamak. , 62:430, 1989.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'At filling factor $\nu=1/m$, $m$ odd integer, I present variational ground-state and excited-state wave functions, of two-dimensional electron system with homogeneous ion background, that show the condensation into a liquid-crystal state. For $m=1, 3, 5$, the ground-state energy per electron is substantially lower than the Laughlin one, for uniform liquid state.'
author:
- 'O. G. Balev'
title: 'Liquid-Crystal State of $\nu=1/m$ Quantum Hall Effects'
---
Discovery of the fractional quantum Hall effect [@tsui1982] in two-dimensional electron systems (2DES) of GaAs-based samples and the Laughlin seminal theory of this effect [@laughlin1983] have generated strong interest to properties of fractional quantum Hall states at $\nu=1/m$, especially for $m=3$ and $5$, . Current understanding is that for $m=3,\; 5$ the Laughlin wave function [@laughlin1983] gives the best known analytical approximation of exact many-body ground-state wave function . For $m=1$ the Laughlin wave function coincides with the Hartree-Fock approximation (HFA) one [@laughlin1983; @mikhailov2001], built from the symmetric gauge single-electron wave functions of the lowest Landau level, and leads to the total energy per electron [@laughlin1983] $\epsilon_{HF}=-\sqrt{\pi/8}
e^{2}/(\varepsilon \ell_{0})$; $\ell_{0}=\sqrt{\hbar c/|e|B}$ is the magnetic length and $\varepsilon$ the background dielectric constant. In present study strong many-body effects are essentially related as with $N$ electrons of 2DES so with $N$ ions. I treat the ions on more equal footing with 2DES than previously . In addition, I use more adequate sets of single-body wave functions; they are localized mainly (or exactly) within the unit cell $L_{x}^{\square} \times L_{x}^{\square}$, $(L_{x}^{\square})^{2}=L_{x} L_{y}/N$. These wave functions help better reflect the tendency: i) of an ion to be mainly localized within its own unit cell, and ii) of an electron to be present mainly within any such unit cell, with equal probability.
In this Letter, at filling factors $\nu=1/m$ with odd integer $m$, I present many-body variational ground-state and excited-state wave functions for electron-ion system, with homogeneous ion density, that have strong correlations between 2DES and ions. The former wave function result in: i) substantially lower ground-state energies for $\nu=1,\; 1/3, \; 1/5$ than obtained in Ref. [@laughlin1983]; ii) the electron density, Eq. (\[38d\]), periodic along one direction with period $\sqrt{2\pi/m} \; \ell_{0}$ is typically very weakly modulated; iii) fractionally quantized Hall conductance, for $m=3,\; 5,\ldots$. I obtain finite excitation gaps along with fractional quasielectron, $e/m$, and quasihole, $|e|/m$, charges, for $m \geq 3$.
We consider a zero-thickness 2DES of width $L_{y}$ ($L_{y}/2> y >-L_{y}/2$) and of length $L_{x}$ ($L_{x}> x >0$) in the presence of a magnetic field, $\mathbf{B}= B \hat{\mathbf{z}}$. The Landau gauge for the vector potential, $\mathbf{A}(\mathbf{r})=(-By,0,0)$, is used; $N$ electrons of a 2DES and $N$ ions are located in the main region, $L_{x} \times L_{y}$. As ions are very heavy, their kinetic energy can be neglected [@laughlin1983; @mikhailov2001]. Then the many-body Hamiltonian $\hat{H}=\hat{H}_{0}+V_{ee}+V_{ei}+V_{ii}$ , where the kinetic energy of electrons $\hat{H}_{0}=\sum_{i=1}^{N} \hat{h}%
_{0i}$, $\hat{h}_{0j}=[i\hbar \mathbf{\nabla}_{j}+ e\mathbf{A}(\mathbf{r}%
_{j})/c]^{2}/2m^{\ast}$; the electron-electron potential $V_{ee}=\frac{1}{2}\sum_{i=1}^{N} \sum_{j=1, j \neq
i}^{N} V(\mathbf{r}_{i}-\mathbf{r}_{j})$, $\mathbf{r}_{i}=(x_{i},y_{i})$; the electron-ion potential $V_{ei}=-\sum_{i=1}^{N} \sum_{j=1}^{N}
V(\mathbf{r}_{i}-\mathbf{R}_{j})$; and the ion-ion potential $V_{ii}=\frac{1}{2}\sum_{i=1}^{N} \sum_{j=1,
j \neq i}^{N} V(\mathbf{R}_{i}-\mathbf{R}_{j})$, $\mathbf{R}_{i}=(X_{i},Y_{i})$; $V(\mathbf{r})=e^{2}/\varepsilon |\mathbf{r}|$. Let us for a $x-$stripe region, given by $x \in(L_{x}^{\square}(n_{xs}^{\alpha}-1); L_{x}^{\square} n_{xs}^{\alpha})$ and $y \in(-\infty,\infty)$, introduce normalized solutions of the single-electron Schrödinger ($\omega_{c}=|e|B/m^{\ast}
c$, $k_{x\alpha}=2\pi n_{ys}^{\alpha}/L_{x}^{\square}$) equation $$\hat{h}_{0} \psi_{n_{\alpha};n_{xs}^{\alpha}, k_{x\alpha}}^{L_{x}^{\square}}(%
\mathbf{r})= \hbar \omega_{c} (n_{\alpha}+1/2)
\psi_{n_{\alpha};n_{xs}^{\alpha},k_{x\alpha}}^{L_{x}^{\square}}(\mathbf{r})
\label{8d}$$of ($\Psi_{n}(y)$ is the harmonic oscillator function) the form $$\psi_{n_{\alpha};n_{xs}^{\alpha},k_{x\alpha}}^{L_{x}^{\square}}(\mathbf{r})
= \frac{e^{ik_{x\alpha} x}}{(L_{x}^{\square})^{1/2}} \Psi_{n_{%
\alpha}}(y-y_{0}(k_{x\alpha})) , \label{9d}$$where $y_{0}(k_{x\alpha})=\ell_{0}^{2} k_{x\alpha}$, $n_{ys}^{\alpha}=0,\pm
1,\dots, \pm(n_{ys}^{\max,t}-1)/2$; $n_{ys}^{\max,t}$ is the odd integer such that $(2\pi/L_{x}^{\square}) n_{ys}^{\max,t} \ell_{0}^{2}=L_{y}$ . For $x > L_{x}^{\square} n_{xs}^{\alpha}$ or $x <L_{x}^{\square}(n_{xs}^{\alpha}-1)$, $\psi_{n_{\alpha};n_{xs}^{\alpha},k_{x\alpha}}^{L_{x}^{\square}}(\mathbf{r})
\equiv 0$. Here $n_{xs}^{\alpha}=1, 2,\ldots, n_{xs}^{\max}$ gives the number to the $x-$stripe region and $L_{x}^{\square} n_{xs}^{\max} =L_{x}$. Then the total number of states of the wave functions Eq. (\[9d\]), on the $n_{\alpha}-$th Landau level in the main region, is $n_{xs}^{\max} n_{ys}^{\max,t} =L_{x}L_{y}/(2\pi
\ell_{0}^{2})=N_{L}$, which is equal to the number of states of “usual” wave functions [@landau1965] $\psi_{n_{\alpha};1,k_{x\alpha}}^{L_{x}}(%
\mathbf{r})$. Wave functions Eq. (\[9d\]) are orthonormal as $$\begin{aligned}
\int_{0}^{L_{x}} dx \int_{-\infty}^{\infty} &&dy \;
\psi_{n_{\beta};n_{xs}^{\beta},k_{x\beta}}^{L_{x}^{\square} \ast}(\mathbf{r}%
) \psi_{n_{\alpha};n_{xs}^{\alpha},k_{x\alpha}}^{L_{x}^{\square}}(\mathbf{r})
\notag \\
&&= \delta_{n_{\beta},n_{\alpha}} \delta_{n_{xs}^{\beta},n_{xs}^{\alpha}}
\delta_{k_{x\beta},k_{x\alpha}} . \label{13d}\end{aligned}$$It can be shown that the set of single-electron wave functions Eq. (\[9d\]) is complete. Eq. (\[13d\]) reduces to well known result [@landau1965] for $\psi_{n_{\alpha};1,k_{x\alpha}}^{L_{x}}(\mathbf{r})$ if to change $%
L_{x}^{\square}$ on $L_{x}$. Further, $(L_{x}^{\square})^{2}=L_{x} L_{y}/N=L_{x} L_{y}/\nu N_{L}$, where $N$ is fixed for a given sample; hence, $L_{x}^{\square}$ is also fixed. Then we obtain $$L_{x}^{\square}/\Delta y_{0}=1/\nu . \label{15d}$$From (\[15d\]) it is seen that within each unit cell can appear only an odd integer number, $m=1, 3,\ldots$, of quantized oscillator centres, $%
y_{0}(k_{x\alpha})$; even $m$, not treated here, is a special case. Then Eq. (\[15d\]) gives, $\ell=0,\; 1,\ldots$, that $$1/\nu=m , \label{16d}$$where $m=2\ell+1$. From Eqs. (\[15d\]), (\[16d\]) it follows $$L_{x}^{\square }=\sqrt{2\pi m}\;\ell _{0}. \label{18d}$$As for $\nu=1/m$ there are $m$ quantized values of $y_{0}(k_{x\alpha i})$ within an $i-$th unit cell and each of them has a particular position within the unit cell, we separate all $N_{L}$ states Eq. (\[9d\]), of a $%
n_{\alpha}-$th Landau level, into the $m$ sets of wave functions. Within any such $n-$th set of states $[y_{0}(k_{x\alpha j}^{(n)})- y_{0}(k_{x\alpha
i}^{(n)})]=k L_{x}^{\square}$, where $k$ is an integer. Here $j$($i$) is the number of a unit cell; it can be any integer from $1$ to $N$. This $i-$number exactly defines the $i-$th unit cell. The superscript in $k_{x\alpha
i}^{(n)}$ is given to distinguish the $k_{x\alpha i}$ pertinent to the $n-$th set of states; the subscript (superscript), $i$, in $k_{x\alpha i}$, etc. ($n_{xs}^{(i)}$, etc.) indicates belonging to the $i-$th unit cell. We choose the values of $n$ as $n=0,\ldots,\pm \ell$ and define $k_{x i}^{(n)}
\equiv k_{x\alpha i}^{(n)}$ as $$k_{x i}^{(0)}=(2\pi m/L_{x}^{\square}) n_{ys}^{(i)},\ldots, k_{x i}^{(\pm
\ell)}=k_{x i}^{(0)} \pm 2\pi \ell/L_{x}^{\square} , \label{19d}$$where $n_{ys}^{(i)}=0, \pm 1,\ldots, \pm (n_{ys}^{\max}-1)/2$; $%
n_{ys}^{\max}=n_{ys}^{\max,t}/m$ is an odd integer. Wave functions Eq. ([9d]{}) of the $n_{\alpha }=0$ Landau level we denote, at $\nu =1/m$, as well as $$\varphi _{n_{xs}^{(i)},k_{xi}^{(n)}}^{i,(m)}(\mathbf{r})\equiv \psi
_{0;n_{xs}^{(i)},k_{xi}^{(n)}}^{L_{x}^{\square}}(\mathbf{r}). \label{20d}$$
I assume the ground-state wave function of electron-ion system, $\Psi
_{N,N}^{(m)}(\mathbf{r}_{1},\ldots ,\mathbf{r}_{N};\mathbf{R}_{1},\ldots ,%
\mathbf{R}_{N})$, in the form ($C_{n} \equiv C_{n}(m)$) $$\Psi _{N,N}^{(m)}=\left[ \sum_{n=-\ell }^{\ell }C_{n}\Psi _{N}^{n,(m)}\right]
\;\dprod\limits_{i=1}^{N}\phi _{n_{xs}^{(i)},n_{ys}^{(i)}}^{(i)}(\mathbf{R}%
_{i}), \label{45d}$$where $|C_{n}|^{2}=1/m$, and the partial many-electron wave function, $\Psi _{N}^{n,(m)}(%
\mathbf{r}_{1},\mathbf{r}_{2},\ldots ,\mathbf{r}_{N})$, is $N-$dimensional Slater determinant of wave functions Eq. (\[20d\]). Here single-ionwave functions $\phi _{n_{xs}^{(i)},n_{ys}^{(i)}}^{(i)}(\mathbf{R})$ are introduced as: $|\phi
_{n_{xs}^{(i)},n_{ys}^{(i)}}^{(i)}(\mathbf{R})|^{2}=1/(L_{x}^{\square })^{2}$, if both $X\in (L_{x}^{\square }(n_{xs}^{(i)}-1),L_{x}^{\square
}n_{xs}^{(i)})$ and $Y\in (L_{x}^{\square }(n_{ys}^{(i)}-1/2),L_{x}^{\square
}(n_{ys}^{(i)}+1/2))$; if $X$ or/and $Y$ is outside of this $i-$th unit cell then $\phi _{n_{xs}^{(i)},n_{ys}^{(i)}}^{(i)}(\mathbf{R})\equiv 0$. The set of these single-body wave functions is orthonormal; then, $\langle \Psi
_{N,N}^{(m)}|\Psi _{N,N}^{(m)}\rangle =1$. The electron density, $n(\mathbf{r%
})=\langle \Psi _{N,N}^{(m)}|\sum_{j=1}^{N}\delta (\mathbf{r}-\mathbf{r}%
_{j})|\Psi _{N,N}^{(m)}\rangle $, in the main region, for $n_{ys}^{\max
}\ggg 1$ (then $N\ggg 1$), is $$n(y)=\frac{\ell _{0}^{-2}}{2\pi m}[1+2\sum_{k=1}^{\infty }e^{-\pi
mk^{2}/2}\cos (\frac{\sqrt{2\pi m}}{\ell _{0}}ky)], \label{38d}$$after using the Fourier transformations and the Poisson’s summation formula [@hilbert]. Eq. (\[38d\]) gives that a unit cell is dressed by electron charge, $e$. The ion density $n_{io}(\mathbf{r})=n_{io}$, where $n_{io}=1/(2\pi m\ell
_{0}^{2})$. Point out, in a good approximation of experimental conditions, $\Psi _{N,N}^{(m)}$ gives that each ion is located in its own unit cell.
The total energy in the ground-state Eq. (\[45d\]) is $$E_{N}^{(m)}=\langle \Psi _{N,N}^{(m)}|\hat{H} |\Psi _{N,N}^{(m)}\rangle ,
\label{80d}$$where the kinetic energy term gives $m^{-1} \sum_{n=-\ell }^{\ell }\langle
\Psi _{N}^{n,(m)}|\hat{H}_{0}|\Psi_{N}^{n,(m)}\rangle =\hbar \omega_{c} N/2$, cf. with Ref. [@yoshioka83i84]; details will be published elsewhere [@balev2004]. In Eq. (\[80d\]) the term $\langle \Psi _{N,N}^{(m)}|V_{ii}|\Psi
_{N,N}^{(m)}\rangle$ obtains the form $$\begin{aligned}
&&\frac{1}{2} \sum_{i=1}^{N} \sum_{j=1, j \neq i}^{N}
\int_{-\infty}^{\infty} d \mathbf{R} \int_{-\infty}^{\infty} d \mathbf{R}%
^{\prime} \frac{e^{2}}{\varepsilon |\mathbf{R}-\mathbf{R}^{\prime}|} \notag
\\
&&\;\;\; \times |\phi_{n_{xs}^{(i)},n_{ys}^{(i)}}^{(i)}(\mathbf{R})|^{2} \;
\; |\phi_{n_{xs}^{(j)},n_{ys}^{(j)}}^{(j)}(\mathbf{R}^{\prime})|^{2} .
\label{82d}\end{aligned}$$$\langle \Psi_{N,N}^{(m)}|V_{ei}|\Psi _{N,N}^{(m)}\rangle$ in Eq. (\[80d\]) has the form $$\begin{aligned}
&&-\; \frac{1}{m} \sum_{n=-\ell }^{\ell } \sum_{i=1}^{N} \sum_{j=1}^{N}
\int_{-\infty}^{\infty} d \mathbf{r} \int_{-\infty}^{\infty} d \mathbf{R}
\frac{e^{2}}{\varepsilon |\mathbf{r}-\mathbf{R}|} \notag \\
&&\;\;\; \;\;\;\; \times |\varphi_{n_{xs}^{(i)},k_{xi}^{(n)}}^{i,(m)}(%
\mathbf{r})|^{2} \; |\phi_{n_{xs}^{(j)},n_{ys}^{(j)}}^{(j)}(\mathbf{R})|^{2}
. \label{83d}\end{aligned}$$$\langle \Psi_{N,N}^{(m),eh}|V_{ee}|\Psi _{N,N}^{(m),eh}\rangle$ in Eq. ([80d]{}) we can rewrite as $$\begin{aligned}
\frac{1}{2m} \sum_{n=-\ell}^{\ell} \sum_{i=1}^{N} \sum_{j=1,j \neq i}^{N}
\langle \Psi_{N}^{n,(m)}| \frac{e^{2}}{\varepsilon |\mathbf{r}_{i}-\mathbf{r}%
_{j}|} |\Psi _{N}^{n,(m)} \rangle , \label{54d}\end{aligned}$$where the matrix elements are calculated as in HFA [@madelung1981]. Then Eq. (\[80d\]) is written, $\tilde{E}_{N}^{(m)}=E_{N}^{(m)}-\hbar
\omega_{c} N/2$, as $$\tilde{E}_{N}^{(m)}= \frac{e^{2} N}{\varepsilon \ell_{0}}
\left[ F_{2}^{A}(m)+F_{1}^{C}(m)+\Delta F_{1}^{C}(m) \right] ,
\label{84d}$$where $F_{2}^{A}(m)$ is exchange-alike term from Eq. (\[54d\]), $$\begin{aligned}
F_{2}^{A}(m) &=& -\frac{1}{\pi } \sum_{k=-\infty; k \neq 0}^{\infty }
e^{-\pi m\;k^{2}} \int_{0}^{\infty }d\xi \int_{0}^{\infty }d\eta \notag \\
&& \times e^{-\eta ^{2}/2} \; S_{m}^{2}(\xi) \; G_{m}(\xi,\eta;k) ,
\label{65d}\end{aligned}$$ $G_{m}(\xi,\eta;k)=[(\xi -\sqrt{2\pi m}\;k)^{2}+\eta^{2}]^{-1/2}$. Further, from the “diagonal” part, $i=j$, of Eq. (\[83d\]) it follows the term $$\begin{aligned}
F^{C}_{1}(m)&=&-\;\frac{2}{\pi} \int_{0}^{\infty} d \xi \int_{0}^{\infty} d
\eta e^{-\eta^{2}/4} \; (\xi^{2}+\eta^{2})^{-1/2} \notag \\
&&\times f_{m}(\eta) \; S_{m}(\eta) \; S_{m}^{2}(\xi) , \label{87d}\end{aligned}$$where $S_{m}(x)=\sin(\sqrt{\pi m/2}\; x)/(\sqrt{\pi m/2}\; x)$, $%
f_{1}(\eta)=1$; for $m=3, 5,\ldots$, $f_{m}(\eta)=m^{-1} \left[%
1+2\sum_{n=1}^{\ell} \cos(\sqrt{2\pi/m} \; n \; \eta ) \right]$. The sum of i) “nondiagonal” part, $i
\neq j$, of Eq. (\[83d\]), ii) Eq. (\[82d\]), and iii) direct-alike contribution from Eq. (\[54d\]) gives $$\begin{aligned}
\Delta F^{C}_{1}(m)&=&-\frac{1}{\pi} \int_{0}^{\infty} d \xi
\int_{0}^{\infty} d \eta \; g_{m}(\eta) S_{m}^{2}(\xi)/\sqrt{\xi^{2}+\eta^{2}%
} \notag \\
&&+(1/\sqrt{2 \pi \; m}) \sum_{k=1}^{\infty} k^{-1} \; e^{-\pi \; k^{2}/m} ,
\label{89d}\end{aligned}$$where $g_{m}(\eta)=S_{m}^{2}(\eta)+e^{-\eta^{2}/2} -2 e^{-\eta^{2}/4} \;
f_{m}(\eta)\; S_{m}(\eta)$. We can rewrite Eq. (\[84d\]) as $\tilde{E}%
_{N}^{(m)}/[e^{2} N/(\varepsilon \ell_{0})]=U^{C}(m)$, where $%
U^{C}(m)=[F^{C}_{1}(m)+\Delta F^{C}_{1}(m)+F_{2}^{A}(m)]$ gives lowering of the total energy per electron in the units of $e^{2}/\varepsilon \ell_{0}$. I calculate numerically that $U^{C}(1) \approx -1.202775$, $U^{C}(3) \approx -0.712971$, $U^{C}(5) \approx -0.552704$, and $%
U^{C}(7) \approx -0.466528$. Here $U^{C}(1)$, $U^{C}(3)$, and $U^{C}(5)$ are substantially lower than pertinent total lowering at $\nu=1, \;1/3$, and $1/5
$ for the Laughlin variational wave function [@laughlin1983] $\; -\;
\sqrt{\pi/8} \approx -0.6267$, $-0.4156 \pm 0.0012$, and $-0.3340 \pm 0.0028$, respectively. Notice, for $m=1$, if in Eqs. (\[80d\])-(\[54d\]) formally to change both single-electron and “single-ion” wave functions on “usual” single-particle wave function $\psi_{0;1,k_{x\alpha}}^{L_{x}}$, then $\tilde{E}_{N}^{(m)}=N \epsilon_{HF}$, for $L_{x} \rightarrow \infty$.
I assume, $\Psi _{N,N;(m)}^{i_{0};j_{0}}(\mathbf{r}_{1},\ldots ,\mathbf{r}%
_{N};\mathbf{R}_{1},\ldots ,\mathbf{R}_{N})$, excited-state wave function of the ground-state Eq. (\[45d\]) as $$\begin{aligned}
&&\Psi _{N,N;(m)}^{i_{0};j_{0}}=\dprod\limits_{i=1}^{N}\phi
_{n_{xs}^{(i)},n_{ys}^{(i)}}^{(i)}(\mathbf{R}_{i})\sum_{n=-\ell }^{\ell }%
\tilde{C}_{n}[(1-\delta _{n,0}) \notag \\
&&\;\;\;\;\;\times \Psi _{N}^{n,(m)}(\mathbf{r}_{1},\ldots)+\delta
_{n,0}\Phi _{N;(m)}^{i_{0};j_{0}}(\mathbf{r}_{1},\ldots)],
\label{115d}\end{aligned}$$where $\tilde{C}_{n}=C_{n}$, for $n\geq 0$, and $\tilde{C}_{n}=-\;C_{n}$, for $n<0$. An excited partialmany-electron wave function $\Phi _{N;(m)}^{i_{0};j_{0}}$ it follows from the $\Psi _{N}^{0,(m)}$ after changing of the $i_{0}-$th row, $\varphi _{n_{xs}^{(i_{0})},k_{xi_{0}}^{(0)}}^{i_{0},(m)}(\mathbf{r}%
_{1}),\ldots ,\varphi _{n_{xs}^{(i_{0})},k_{xi_{0}}^{(0)}}^{i_{0},(m)}(%
\mathbf{r}_{N})$, by the determinant row of the, for $m \geq 3$, form $%
\varphi _{n_{xs}^{(j_{0})},k_{xj_{0}}^{(\tilde{n})}}^{j_{0},(m)}(\mathbf{r}%
_{1}),\ldots ,\varphi _{n_{xs}^{(j_{0})},k_{xj_{0}}^{(\tilde{n}%
)}}^{j_{0},(m)}(\mathbf{r}_{N})$, where $\tilde{n}\neq 0$; Eq. (\[19d\]) gives $k_{xj_{0}}^{(\tilde{n})}=(2\pi \;m/L_{x}^{\square })\left[
n_{ys}^{(j_{0})}+\tilde{n}/m\right] $, $\tilde{n}=\pm 1,\ldots ,\pm \ell $; i.e., $k_{xj_{0}}^{(\tilde{n})}\neq k_{xi}^{(0)}$, where $i=1,\ldots ,N$. For $m=1$ in new determinant row: i) the implicit spin up wave function $%
|1>=\psi _{1}(\sigma _{j})$ should be substituted by spin down one, $%
|-1>=\psi _{-1}(\sigma _{j})$; ii) $\tilde{n}=0$. We assume that the $i_{0}$-th unit cell, where the quasihole appears, as well as the $j_{0}-$th unit cell, where the quasielectron is mainly localized, there are well inside of the main region. $\Psi_{N,N;(m)}^{i_{0};j_{0}}$ describes excitation of a quasiexciton type . It is seen that at any separation between the quasielectron and the quasihole their charges are given (details will be published elsewhere [@balev2004]) as $e/m$ and $|e|/m$, respectively. We need the energy gap, $\Delta ^{(m)}$, for the creation of one quasielectron and one quasihole, infinitely spatially separated[laughlin1983,prang1990, chakraborty1995,sarma1997,morf1986,macdonald1986,willett1988,dorozhkin1995]{}; notice, $\langle \Psi _{N,N;(m)}^{i_{0};j_{0}}|\Psi
_{N,N}^{(m)}\rangle =0$ and $\langle \Psi _{N,N;(m)}^{i_{0};j_{0}}|\Psi
_{N,N;(m)}^{i_{0};j_{0}}\rangle =1$.
With infinitely spatially separated quasielectron and quasihole, $\Delta ^{(m)}=\langle \Psi_{N,N;(m)}^{i_{0};j_{0}}|\hat{H}|
\Psi_{N,N;(m)}^{i_{0};j_{0}}\rangle -E_{N}^{(m)}$ is given, for $m=3, 5, 7$, as $\tilde{\Delta}^{(m)}=\left[|F_{1}^{C}(m)| +\Delta F^{C}(m,1) +
2 |F_{2}^{A}(m)|-F^{(m)}_{1} \right]/m$, $\tilde{\Delta}^{(m)}=\Delta^{(m)}/(e^{2}/\varepsilon \ell_{0})$, where $$\begin{aligned}
F^{(m)}_{\tilde{n}}&=&(1/\pi) \sum_{k=-\infty}^{\infty }e^{-\pi m\;(k-\tilde{%
n}/m)^{2}}\int_{-\infty}^{\infty }d\xi \int_{0}^{\infty }d\eta \notag \\
&&\times e^{-\eta ^{2}/2} \; S_{m}^{2}(\xi) \; G_{m}(\xi,\eta;k-\tilde{n}/m)
, \label{A.4}\end{aligned}$$ $$\begin{aligned}
&&\Delta F^{C}(m,\tilde{n})=-\; \frac{2^{3/2}}{\sqrt{\pi} m^{3/2}}
\sum_{k=1}^{\infty} \frac{\exp(-\frac{\pi k^{2}}{m})}{k} \sin^{2}\left(\frac{%
\pi k \tilde{n}}{m}\right) \notag \\
&&+\frac{2}{\pi m} \int_{0}^{\infty} \int_{0}^{\infty} \frac{ d \xi d \eta}{%
\sqrt{\xi^{2}+\eta^{2}}} \; S_{m}^{2}(\xi) \; G_{m}(\eta) , \label{123d}\end{aligned}$$$G_{m}(\eta)=\exp(-\eta^{2}/4) \; [ \exp(-\eta^{2}/4) -S_{m}(\eta) ]$, and it is taken into account that only $\tilde{n}=1$ corresponds to $\Delta^{(m)}$. For $m=1$, $\Delta^{(1)}-|g_{0}| \mu_{B}
B=(e^{2}/\varepsilon \ell_{0}) [|F_{1}^{C}(1)| +\Delta F^{C}(1,0)+2
|F_{2}^{A}(1)|]$, where $g_{0}$ is the bare Landé g-factor. I calculate numerically that $\tilde{\Delta}^{(1)}- |g_{0}| \mu_{B} B/(e^{2}/\varepsilon
\ell_{0}) \approx 1.253895$ (i.e., very close to $\sqrt{\pi/2} \approx
1.253314$), $\tilde{\Delta}^{(3)} \approx 0.170657$, $\tilde{\Delta}%
^{(5)} \approx 0.069867$, and $\tilde{\Delta}^{(7)} \approx 0.036086$.
Point out that the ground-state Eq. (\[45d\]) shows broken symmetry “liquid-crystal” behavior of 2DES as the electron density, Eq. (\[38d\]), is periodic along $y$-direction, with period $\ell_{0} \sqrt{2\pi/m}$. We can make electron density much more homogeneous, however, the latter state has much higher energy than $U^{C}(m)$. For the ground-state Eq. (\[45d\]), at $\nu=1/m$, I calculate (details will be published elsewhere [@balev2004]) that the Hall conductance $\sigma_{H}=e^{2}/( 2m \pi \hbar)$; i.e., it is properly quantized. Similar to Refs. [@laughlin1983; @niu1985], we can speculate that for a weak disorder if the Fermi level still lies in a gap or mobility gap the Hall conductance should be quantized in a finite range of $B$.
Present energy gap $\tilde{\Delta}^{(3)}$ is about $1.6$ times larger than typically calculated for the Laughlin liquid pertinent excitation gap [@haldane1985; @morf1986; @macdonald1986]. For detailed comparison of the gap with experiment it is known that a finite thickness of 2DES should be taken into account as well as effects of disorder [@willett1988; @morf2002]. In addition, we can speculate that many-body effects similar to those studied in [@balev2001] (for traditional $\nu =1$ state) and related with edge states here, maybe, also will lead to highly asymmetric pinning of the Fermi level within the energy gap. Then, similar to [@balev2001], actual activation gap can be much smaller than $\tilde{\Delta}^{(m)}/2$. In summary, I have presented, at $\nu=1/m$, the theory of liquid-crystal ground-state with periodic, along one direction, density of 2DES and uniform density of ions. The ground-state has strong correlations between 2DES and ions. The Hall conductance is properly quantized. Excitation gap, for $m=1, \; 3, \; 5, \; 7$, is finite; quasielectron and quasihole charges are fractional, $\pm e/m$, for $m \geq 3$.
This work was supported in part by Universidade Federal de São Carlos.
[99]{} D. C. Tsui, H. L. Stormer, and A. C. Gossard, Phys. Rev. Lett. **48**, 1559 (1982).
R. B. Laughlin, Phys. Rev. Lett. **50**, 1395 (1983).
*The Quantum Hall Effect*, edited by R. E. Prang and S. M. Girvin, 2nd ed. (Springer-Verlag, New York, 1990).
T. Chakraborty and P. Pietilainen, *The Quantum Hall Effects-Fractional and Integral*, 2nd ed. (Springer-Verlag, Berlin, 1995).
*Perspectives in Quantum Hall Effects*, edited by S. Das Sarma and A. Pinczuk (John Wiley & Sons, New York, 1997).
D. Yoshioka and P. A. Lee, Phys. Rev. B **27**, 4986 (1983); A. H. MacDonald, Phys. Rev. B **30**, 4392 (1984).
K. Maki and X. Zotos, Phys. Rev. B **28**, 4349 (1983); D. J. Thouless, Phys. Rev. B **31**, R8305 (1985). D. Yoshioka, B. I. Halperin, and P. A. Lee, Phys. Rev. Lett. **50**, 1219 (1983); D. Yoshioka, Phys. Rev. B **29**, 6833 (1984).
F. D. M. Haldane and E. H. Rezayi, Phys. Rev. Lett. **54**, 237 (1985).
S. Kivelson, C. Kallin, D. P. Arovas, and J. R. Schrieffer, Phys. Rev. Lett. **56**, 873 (1986).
R. Morf and B. I. Halperin, Phys. Rev. B **33**, 2221 (1986).
A. H. MacDonald and S. M. Girvin, Phys. Rev. B **34**, 5639 (1986).
F. Claro, Phys. Rev. B **35**, 7980 (1987).
R. L. Willett, H. L. Stormer, D. C. Tsui, A. C. Gossard, and J. H. English, Phys. Rev. B **37**, R8476 (1988).
J. K. Jain, Phys. Rev. Lett. **63**, 199 (1989).
S. I. Dorozhkin, R. J. Haug, K. von Klitzing, and K. Ploog, Phys. Rev. B **51**, 14729 (1995).
S. A. Mikhailov, Physica B **299**, 6 (2001).
O. Ciftja, C. M. Lapilli, and C. Wexler, Phys. Rev. B **69**, 125320 (2004).
A. Cabo and F. Claro, Phys. Rev. B **70**, 235320 (2004).
L. D. Landau and E. M. Lifshitz, *Quantum Mechanics* (Pergamon Press, Oxford, 1965).
R. Courant and D. Hilbert, *Methods of Mathematical Physics*, vol. I (John Wiley & Sons, New York, 1989).
O. G. Balev (unpublished).
O. Madelung, *Introduction to Solid-State Theory*, 2nd ed. (Springer-Verlag, Berlin, 1981).
Q. Niu, D. J. Thouless, and Y. S. Wu, Phys. Rev. B **31**, 3372 (1985).
R. H. Morf, N. d’Ambrumenil, and S. Das Sarma, Phys. Rev. B **66**, 075408 (2002).
O. G. Balev and N. Studart, Phys. Rev. B **64**, 115309 (2001).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study the invariant-mass distributions of dileptons produced in ultrarelativistic heavy-ion collisions at very low pair transverse momenta, $P_T\leq 0.15$GeV. Specifically, we investigate the interplay of thermal radiation with initial photon annihilation processes, $\gamma \gamma \to l^+ l^-$, triggered by the coherent electromagnetic fields of the incoming nuclei. For the thermal radiation, we employ the emission from the QGP and hadronic phases with in-medium vector spectral functions which describes the inclusive excess radiation observed over a wide range of collision energies. For the coherent photon fusion processes, whose spectrum is much softer than for thermal radiation, we employ initial fluxes from the Fourier transform of charge distributions of the colliding nuclei in the equivalent-photon approximation. We first verify that the combination of photon fusion, thermal radiation and final-state hadron decays gives a fair description of the low-$P_T$ dilepton mass spectra as recently measured by the STAR collaboration in $\sqrt{s_{NN}}$=200GeV Au+Au collisions for different centrality classes, including experimental acceptance cuts. The coherent contribution dominates in peripheral collisions, while thermal radiation shows a markedly stronger increase with centrality. We perform similar calculations at lower collision energies ($\sqrt{s_{NN}}$=17.3GeV) and compare to the acceptance-corrected dimuon excess spectra measured by the NA60 experiment at the CERN SPS; here, the contribution from photoproduction is subleading. We also provide predictions for the ALICE experiment at the LHC; the pertinent excitation function from SPS to LHC energies reveals a nontrivial interplay of photoproduction and thermal radiation.'
author:
- 'Mariola K[ł]{}usek-Gawenda'
- Ralf Rapp
- Wolfgang Schäfer
- 'Antoni Szczurek [^1]'
title: |
Dilepton Radiation in Heavy-Ion Collisions\
at Small Transverse Momentum
---
Introduction {#sec:intro}
============
Dilepton production in ultra-relativistic heavy-ion collisions (URHICs) has a long history as a probe of the hot QCD medium produced in these collisions [@Feinberg:1976ua; @Shuryak:1978ij]. To date, the observed excess radiation over final-state hadron decays has been firmly established as thermal radiation from the interacting fireball [@Tserruya:2009zt; @Specht:2010xu]. In the low-mass region, at invariant mass $M\lesssim 1\, \rm GeV$, hadronic radiation dominates revealing the melting of the $\rho$ resonance [@Rapp:1997fs], which indicates a transition to partonic degrees of freedom and is consistent with chiral symmetry restoration [@Hohler:2013eba]. In the intermediate-mass region, quark-gluon plasma (QGP) radiation dominates [@Rapp:1999zw; @Ruppert:2007cr], opening the possibility for direct measurements of its temperature [@Rapp:2014hha].
On the other hand, recent measurement of dileptons in ultraperipheral heavy-ion collisions (UPCs) [@PHENIX; @ALICE; @ATLAS], where the incoming nuclei do not touch and thus no fireball is created, have also revealed a substantial amount of dilepton radiation at both low and intermediate masses. This radiation is characterized by a very soft slope in pair transverse momentum, $P_T$, much steeper than for thermal radiation emitted from strongly interacting fireballs. A good description of the ultraperipheral dilepton data can be achieved with photon fusion reactions using realistic fluxes generated by the electromagnetic (EM) field of the highly relativistic incoming nuclei [@KS2017]. The coherent EM fields are expected to also give a contribution in impact parameter configurations where the two ions collide.
The question then arises how the interplay of these two processes works out in peripheral heavy-ion collisions, where thermal radiation is much suppressed compared to central collisions while the coherent photon emission is still appreciable. Recently, two of us have shown that $J/\psi$ photoproduction in UPCs gives a significant contribution to the low-$P_T$ production yield in semi-central collisions at the LHC [@Klusek-Gawenda:2015hja], in agreement with the ALICE results for different centralities [@ALICE-jpsi]. Similar findings have been reported in Ref. [@Shi]. Recently, the STAR collaboration released new data for low-$P_T$ $e^+ e^-$ production in 200GeV Au+Au and U+U collisions over a large range of invariant mass and for different centralities [@Adam:2018tdm]. In that work, initial model comparisons were conducted using contributions from thermal radiation [@Rapp:2014hha] plus hadronic final-state decays plus photon fusion contributions from two different approaches [@Klein:2018cjh; @Zha:2018ywo]. The general trend was that photon fusion processes dominate, and can explain, the low-$P_T$ yield ($P_T$$<$0.15GeV) for peripheral collisions, although significant differences in the predicted yields are present. On the other hand, thermal radiation plus the hadronic decay “cocktail" increase much more rapidly with centrality and dominate the yield for $P_T$$>$0.2GeV at all centralities. A recent transport calculation [@Song:2018dvf] confirmed that hadronic sources (thermal emission plus cocktail) cannot explain the low-$P_T$ excess observed by the STAR in peripheral Au-Au collisions. No photoproduction processes were considered in their study.
In the present paper we follow up on the above question by combining thermal radiation with photon fusion processes for low-$P_T$ dilepton production. Quantitatively disentangling the two contributions in this regime provides important tests for either one, especially if systematic centrality and collision energy dependences can be established. After benchmarking our combined results against the STAR data at 200GeV, we therefore expand our analysis to both lower and higher energies. For the former, we focus on In-In collisions at SPS energy ($\sqrt{s_{NN}}$=17.3GeV) where we can test our predictions against the high-precision NA60 data [@Arnaldi:2008fw; @Specht:2010xu]; in fact, a low-$P_T$ excess was observed in these data which has not been fully explained to date.
The paper is organized as follows. In Sec. \[sec\_sources\] we briefly review the mechanisms for coherent photon fusion (Sec. \[ssec\_gamgam\]) and thermal radiation (Sec. \[ssec\_thermal\]). In Sec. \[sec\_spec\] we apply them to low-$P_T$ dilepton invariant-mass spectra as measured at RHIC (Sec. \[ssec\_rhic\]) and the SPS (Sec. \[ssec\_sps\]), make predictions for the LHC (Sec. \[ssec\_sps\]), and compute an excitation function (Sec. \[ssec\_excit\]). In Sec \[sec\_concl\] we conclude.
Dilepton Sources {#sec_sources}
================
In this section we recall the main ingredients to the two main dilepton sources considered in this work, [[*[i.e.]{}*]{}]{}, the coherent initial photon fusion reactions in Sec. \[ssec\_gamgam\] and thermal radiation in Sec. \[ssec\_thermal\].
Initial Photon-Photon Fusion Mechanism {#ssec_gamgam}
--------------------------------------
The main ingredient for the photon-photon fusion mechanism is the Weizsäcker-Williams flux of photons for an ion of charge $Z$ moving along impact parameter ${\mbox{\boldmath $b$}}$ ($b = |{\mbox{\boldmath $b$}}|$) with the Lorentz-boost parameter $\gamma$. With the nuclear charge form factor $F_{\rm em}$ as an input the flux can be calculated as [@Bertulani:1987tz; @Baur:2001jj] $$\begin{aligned}
N(\omega,b)
&&= {Z^2 \alpha_{\rm EM} \over \pi^2}
\Big| \int_0^\infty dq_T {q_T^2 F_{\rm em}(q_T^2 + {\omega^2 \over \gamma^2} )
\over q_T^2 + {\omega^2 \over \gamma^2} } J_1(b q_T) \Big|^2\, ,
\label{eq:WW-flux}\end{aligned}$$ where $J_1$ is a Bessel function. We calculate the formfactor from the Fourier transform of the nuclear charge density, for which parameterizations are available in Ref. [@DeJager:1987qc].
The differential cross section for dilepton ($l^+ l^-$) production via $\gamma \gamma$ fusion at fixed impact parameter ${\mbox{\boldmath $b$}}$ of a nucleus nucleus collision can then be written as $$\begin{aligned}
{d \sigma_{ll} \over d\xi d^2{\mbox{\boldmath $b$}}} =
\int d^2{\mbox{\boldmath $b$}}_1 d^2{\mbox{\boldmath $b$}}_2 \, \delta^{(2)}({\mbox{\boldmath $b$}}- {\mbox{\boldmath $b$}}_1 - {\mbox{\boldmath $b$}}_2) N(\omega_1,b_1) N(\omega_2,b_2)
{d \sigma(\gamma \gamma \to l^+ l^-; \hat s) \over d (-\hat t)} \ ,\end{aligned}$$ where the phase space element is $d\xi = dy_+ dy_- dp_T^2$. Here, $y_\pm$, $p_T$ and $m_l$ are the single-lepton rapidities, transverse momentum and mass, respectively, and $$\begin{aligned}
\omega_1 = {\sqrt{p_T^2 + m_l^2} \over 2} \, ( e^{y_+} + e^{y_-} ) \ , \
\omega_2 = {\sqrt{p_T^2 + m_l^2} \over 2} \, ( e^{-y_+} + e^{-y_-} ) \ , \ \hat{s} = 4 \omega_1 \omega_2 \ .\end{aligned}$$ As can be seen from Eq.(\[eq:WW-flux\]), the transverse momenta, $q_T$, of the photons have been integrated out, and dileptons are produced back-to-back in the transverse plane, [[*[i.e.]{}*]{}]{}, the transverse momentum $P_T$ of the pair is neglected.
In UPCs the incoming nuclei do not touch, [[*[i.e.]{}*]{}]{}, no strong interactions occur between them. In this case one usually imposes the constraint $b > 2 R_A$ when integrating over impact parameter $b = |{\mbox{\boldmath $b$}}|$. Here we lift this restriction allowing the nuclei to collide. Then the final state will no longer contain the intact nuclei but the dileptons will be produced on top of the hadronic nuclear event characterized by an impact parameter ${\mbox{\boldmath $b$}}$ (or range thereof). Note that even for overlapping nuclei, $b < 2 R_A$, leptons are predominantly produced outside the overlap region, for $ b_{1,2} > R_A$. This situation is very different from the photoproduction heavy vector mesons [@Klusek-Gawenda:2015hja] which tend to be produced “inside" of one of the nuclei.
Here we are interested in the dependence on centrality. The mass-differential dilepton yield from coherent photons in a centrality class ${\cal C}$ corresponding to an impact parameter range of $[b_{\rm min}, b_{\rm max}]$ is given by $$\begin{aligned}
{dN_{ll}[{\cal C}] \over dM} = {1 \over f_{\cal C} \cdot \sigma^{\rm in}_{\rm AA}}
\int_{b_{\rm min}}^{b_{\rm max}} db \,
\int d\xi \,
\delta(M - 2 \sqrt{\omega_1 \omega_2}) \,
{d \sigma_{ll} \over d\xi db }\Big|_{\rm cuts} \ , \end{aligned}$$ where we have indicated kinematic cuts on single-lepton variables as applied in experiment, and $f_{\cal C}$ is the fraction of inelastic hadronic events contained in the centrality class ${\cal C}$, $$f_{\cal C} = { 1 \over \sigma^{\rm in}_{\rm AA} }
\int_{b_{\rm min}}^{b_{\rm max}} db \frac{d \sigma^{\rm in}_{\rm AA}}{d b} \, .$$ We determine $[b_{\rm min}, b_{\rm max}]$ and $\sigma^{\rm in}_{\rm AA}$ by using the optical Glauber model as $$\frac{d \sigma^{\rm in}_{\rm AA}}{d b} = 2 \pi b (1 - e^{-\sigma^{\rm in}_{\rm NN} T_{\rm AA}(b)}) \; .$$ The nuclear thickness function, $T_{\rm AA}(b)$, is obtained from the convolution of nuclear density distributions for which we use standard Woods-Saxon profiles, $n_{\rm A}(r)$, $$T_{\rm AA}(b) = \int d^3 \vec{r}_1 d^3\vec{r}_2 \,
\delta^{(2)}({\mbox{\boldmath $b$}}- {\mbox{\boldmath $r$}}_{1 \perp} - {\mbox{\boldmath $r$}}_{2 \perp}) \, n_{\rm A}(r_1) n_{\rm A}(r_2) \, .$$
Thermal Dileptons {#ssec_thermal}
-----------------
Thermal dilepton radiation in URHICs is based on the idea that the abundant production of hadrons, together with strong re-interactions, leads to the formation of a locally equilibrated medium whose expansion can be described by relativistic hydrodynamics. This idea is by now well established, on the one hand by the success of hydrodynamic modelling in reproducing the transverse-momentum spectra of the produced hadrons [@Shuryak:2008eq; @Heinz:2013th; @Gale:2013da], and, on the other hand, by the observation and theoretical description of dilepton radiation that goes well beyond the final-state decays of the produced hadrons [@Tserruya:2009zt; @Specht:2010xu; @Rapp:2013nxa; @Rapp:2014hha]. The basic equation to compute dilepton invariant-mass spectra involves an integration of the 8-differential emission rate over the space-time evolution of the expanding fireball, $$\frac{dN_{ll}}{dM} = \int d^4x \ \frac{Md^3P}{P_0} \
\frac{dN_{ll}}{d^4xd^4P} \ .$$ where $(P_0,\vec P)$ and $M=\sqrt{P_0^2-P^2}$ are the 4-vector ($P=|\vec P|$) and invariant mass of the lepton pair, respectively. The thermal emission rate can be expressed as $$\frac{dN_{ll}}{d^4xd^4P} = -\frac{\alpha_{\rm EM}^2 L(M)}{\pi^3 M^2} \
f^B(P_0;T) \ {\rm Im}\Pi_{\rm EM}(M,P;\mu_B,T) \ ,
\label{rate}$$ in terms of the Bose distribution function, $f^B$, and the EM spectral function, Im$\Pi_{\rm EM}$, depending on the local temperature, $T$, and baryon chemical potential, $\mu_B$, of the medium ($L(M)$ is a lepton phase-space factor which approaches one for $M\gg m_l$). The fireball medium generally goes through both QGP and hadronic phases; for the respective spectral functions we employ in-medium quark-antiquark annihilation constrained by lattice-QCD [@Rapp:2013nxa] and in-medium vector spectral functions in the hadronic sector [@Urban:1999im]; the vector meson resonances strongly broaden in the medium and essentially melt at temperatures close to the pseudocritical one, providing a nearly smooth transition to the QGP rates. Different centrality classes for different colliding systems are characterized by the measured hadron multiplicities and appropriate initial conditions for the fireball. Its expansion is modelled by a simple volume parameterization guided by hydrodynamic models, and the underlying equation-of-state is based on $\mu_B$=0 lattice-QCD results for the QGP smoothly matched to a hadron resonance gas [@Rapp:2014hha]. This approach is consistent with available dilepton data from SIS [@Galatyuk:2017ack] via SPS [@Arnaldi:2008fw; @Agakichiev:2005ai] to RHIC [@Huck:2014mfa; @Adamczyk:2015lme; @Adare:2015ila] energies.
Low-$P_T$ Dilepton Invariant-Mass Spectra {#sec_spec}
=========================================
We are now in position to combine the two sources described above and compare their sum to low-$P_T$ measurements at RHIC (\[ssec\_rhic\]) and SPS (\[ssec\_sps\]) and make predictions for the LHC (\[ssec\_lhc\]) and their excitation function (\[ssec\_excit\]).
Au-Au Collisions at RHIC {#ssec_rhic}
------------------------
\(a) ![ (a) Dielectron invariant-mass spectra for pair $P_T$$<$0.15GeV in Au+Au ($\sqrt{s_{NN}}$=200GeV) collisions for three centrality classes calculated (including experimental acceptance cuts, $p_{T}$$>$0.2GeV, $|\eta_e|$$<$1 and $|y_{e^+e^-}|$$<$1) for $\gamma \gamma$ fusion (solid lines), thermal radiation (dotted lines) and the hadronic cocktail (dashed lines). (b) Comparison of the total sum (solid lines) to STAR data [@Adam:2018tdm]. []{data-label="fig:photoproduction_vs_STAR"}](dN_dMee_gg_rho_coc_new_STAR.eps "fig:") (b) ![ (a) Dielectron invariant-mass spectra for pair $P_T$$<$0.15GeV in Au+Au ($\sqrt{s_{NN}}$=200GeV) collisions for three centrality classes calculated (including experimental acceptance cuts, $p_{T}$$>$0.2GeV, $|\eta_e|$$<$1 and $|y_{e^+e^-}|$$<$1) for $\gamma \gamma$ fusion (solid lines), thermal radiation (dotted lines) and the hadronic cocktail (dashed lines). (b) Comparison of the total sum (solid lines) to STAR data [@Adam:2018tdm]. []{data-label="fig:photoproduction_vs_STAR"}](dN_dMee_gg_rho_coc_new_sum_STAR.eps "fig:")
In Fig. \[fig:photoproduction\_vs\_STAR\] we show dielectron invariant-mass spectra for small pair $P_{T} <$ 0.15 GeV and three different centrality classes as selected in the STAR data: peripheral (60-80%), semi-peripheral (40-60%) and semi-central (10-40%) collisions. We also include the experimental acceptance cuts on the single-lepton tracks as applied by STAR, and take the cocktail contribution as provided by STAR [@Adam:2018tdm] representing the final-state decays of the produced hadrons. In peripheral collisions the photon-photon contribution dominates while in semi-central collisions all three contributions are of similar magnitude. Their sum yields a rather good agreement with the STAR data, except for the $J/\psi$ peak region. Our calculations only contain incoherent $J/\psi$ production, from binary nucleon-nucleon collisions; we conjecture that the missing contribution is due to a coherent contribution discussed, [[*[e.g]{}*]{}]{}, in Ref. [@Klusek-Gawenda:2015hja], which we do not further pursue here, as our focus is on the interplay with thermal radiation.
In+In Collisions at the SPS {#ssec_sps}
---------------------------
After benchmarking our approach with the STAR data, we now turn to the high-precision NA60 data from In-In ($\sqrt{s_{NN}}$=17.3GeV) collisions at the SPS, for which no calculations of the coherent contribution are available to our knowledge. In addition, the dimuon $P_T$ spectra of the NA60 collaboration show a distinct hint for an enhancement at very low $P_T$ in various mass bins up to $M$$\simeq$1GeV that could not fully be explained by calculations of thermal radiation [@Arnaldi:2008fw; @Specht:2010xu].
In Fig. \[fig:NA60\] we show the results of our calculations for thermal radiation and the coherent $\gamma \gamma$ mechanism for minimum bias (MB) In-In collisions, in comparison to the acceptance corrected NA60 excess data (from which the cocktail has also been subtracted). Unlike for RHIC energies the $\gamma \gamma$ contribution is rather small and only plays some role at small dimuon invariant masses where the NA60 data run out of precision.
![Low-$P_T$ (<0.2GeV) acceptance-corrected dimuon invariant mass excess spectra in the rapidity range 3.3<$Y_{\mu^+\mu^-,LAB}$<4.2 for MB In+In ($\sqrt{s_{NN}}$=17.3 GeV) collisions at the SPS. Calculations for coherent $\gamma \gamma$ fusion (solid line) and thermal radiation (dashed line) are compared to NA60 data [@Arnaldi:2008fw; @Specht:2010xu]. []{data-label="fig:NA60"}](dN_dMmumu_rho_QGP_NA60_corrected.eps)
Pb+Pb Collisions at the LHC {#ssec_lhc}
---------------------------
The rather different relative importance of low-$P_T$ dilepton emission from coherent $\gamma\gamma$ and thermal radiation sources found at RHIC and SPS energies calls for their calculation at LHC energies. This is shown in Fig. \[fig:ALICE\] where we compare our predictions for the two sources for Pb+Pb collisions at $\sqrt{s_{NN}}$=5.02TeV for the same centrality classes and single-lepton acceptance cuts as for our RHIC calculations above. Compared to the latter, the picture is qualitatively similar, although the strength of thermal contribution is relatively stronger, especially in semi-peripheral and central collisions where it is comparable and even larger, respectively, than the $\gamma\gamma$ yield at low mass.
![Our predictions for low-$P_T$ dilepton radiation in Pb+Pb ($\sqrt{s_{NN}}$=5.02TeV) collisions from coherent $\gamma\gamma$ fusion (solid lines) and thermal radiation (dashed lines) for three centrality classes and acceptance cuts as specified in the figures. []{data-label="fig:ALICE"}](dN_dMee_gg_rho_LHC_c10_40_new.eps "fig:") ![Our predictions for low-$P_T$ dilepton radiation in Pb+Pb ($\sqrt{s_{NN}}$=5.02TeV) collisions from coherent $\gamma\gamma$ fusion (solid lines) and thermal radiation (dashed lines) for three centrality classes and acceptance cuts as specified in the figures. []{data-label="fig:ALICE"}](dN_dMee_gg_rho_LHC_c40_60_new.eps "fig:") ![Our predictions for low-$P_T$ dilepton radiation in Pb+Pb ($\sqrt{s_{NN}}$=5.02TeV) collisions from coherent $\gamma\gamma$ fusion (solid lines) and thermal radiation (dashed lines) for three centrality classes and acceptance cuts as specified in the figures. []{data-label="fig:ALICE"}](dN_dMee_gg_rho_LHC_c60_80_new.eps "fig:")
Excitation Function {#ssec_excit}
-------------------
The results reported for the three different collision energies at the SPS, RHIC and the LHC in the previous three sections are now generalized into a systematic excitation function. In Fig. \[fig:sig\_tot\_sqrts\] we show the invariant-mass integrated low-$P_T$ (<0.15GeV) dilepton yields for the $\gamma \gamma$ and thermal components as a function of collision energy for the 3 centralities as used at RHIC and the LHC above, and including the same single-electron acceptance cuts. The photoproduction yields can be straightforwardly obtained from a direct calculation. For the thermal radiation, this would be much more involved. Instead we make use of the power-law like $N_{\rm ch}$ dependence of thermal radiation [@Rapp:2013nxa] together with a power-law behavior of $dN_{\rm ch}/dy$ on collision energy [@Basu:2016dmo] with the ansatz for $N_{ee}^{\rm th} = N_0 s^\beta$. We then fit the two parameters, $N_0$ and $\beta$, independently for each of the three centrality classes considered above, including the experimental acceptance cuts.
The $\gamma \gamma$ fusion yield rises rather strongly in the tens of GeV collision energy regime, followed by a saturation above $\sqrt{s_{NN}}$ = 100 GeV, while thermal radiation shows a much more gradual increase, cf. Fig. \[fig:sig\_tot\_sqrts\]. The latter therefore dominates above the former at low energies, $\sqrt{s_{NN}}\lesssim 20$GeV (as found in Sec. \[ssec\_sps\]) at all centralities, and then increasingly so again in the TeV energy range for more central collisions. On the other hand, the $\gamma \gamma$ fusion contribution is most significant around the regime where it levels off, where its yield is (much) larger than the one from thermal radiation for semi-/peripheral collisions, and comparable for semi-central collisions. Our analysis therefore identifies the RHIC energy regime as the most promising ground to investigate this production mechanism.
![Excitation function of low-$P_T$ ($<$0.15GeV) dilepton yields from $\gamma \gamma$ fusion (solid lines) and thermal radiation (dashed lines) in collisions of heavy nuclei (A$\simeq$200) around midrapidity in three centrality classes, including single-$e^\pm$ acceptance cuts. []{data-label="fig:sig_tot_sqrts"}](N_tot_c_2ap.eps)
Conclusions {#sec_concl}
===========
We have studied low-$P_T$ dilepton production in ultrarelativistic heavy-ion collisions, by conducting systematic comparisons of the two sources that are believed to be prevalent in this regime, [[*[i.e.]{}*]{}]{}, thermal radiation and photon-photon fusion within the coherent fields of the incoming nuclei. The former was taken from a well-tested model including in-medium hadronic and QGP emission rates, while the latter was calculated utilizing photon fluxes with realistic nuclear form factors including the case of nuclear overlap. We first reconfirmed the finding of a recent STAR analysis that the combination of the two sources (augmented by a contribution from the hadronic final-state decay cocktail) gives a fair description of low-$P_T$ dilepton data in Au-Au ($\sqrt{s_{NN}}$=200GeV) collisions in three centrality classes for invariant masses from threshold to 4GeV (with the exception of the $J/\psi$ peak, indicating an additional production mechanism not included here). The coherent emission was found to be dominant for the two peripheral samples, and comparable to the cocktail and thermal radiation yields in semi-central collisions.
At the lower SPS energies ($\sqrt{s_{NN}}$=17.3GeV) we found that the $\gamma \gamma$ contribution is subleading. Specifically, for acceptance-corrected low-$P_T$ dimuon spectra as measured by NA60 in MB In-In collisions, it reaches up to ten percent of the thermal radiation for masses near the dimuon threshold, rapidly falling off with increasing mass. On the other hand, at the high-energy frontier, the situation turned out to be similar to RHIC energies, although the role of thermal radiation relative to the coherent mechanism is somewhat more pronounced in the multiple-TeV range. The interplay of these processes at the LHC is of particular interest in view of plans by ALICE [@Antinori:2018] to lower the single-electron $p_T$ cuts and measure very-low mass spectra to possibly extract the EM conductivity from thermal emission.
We have summarized our results in an excitation function of low-$P_T$ radiation covering three orders of magnitude in collision energy. While coherent production increases rather sharply, and then levels off, near $\sqrt{s_{NN}}$$\simeq$100GeV, thermal radiation increases more gradually with $s_{NN}$. This explains why the latter is dominant at the SPS, the former dominates at RHIC, and the latter becomes more important again at the LHC.
This work has been supported by the US National Science Foundation under grant no. PHY-1614484 (RR) and by the Polish National Science Center grant DEC-2014/15/B/ST2/02528 (MKG, WS, AS).
[9]{}
E. L. Feinberg, Nuovo Cim. A **34** (1976) 391.
E. V. Shuryak, Phys. Lett. B **78** (1978) 150.
I. Tserruya, in [*Relativistic Heavy-Ion Physics*]{}, edited by R. Stock and Landolt Börnstein (Springer), New Series [**I/23A**]{} (2010) 4-2 \[arXiv:0903.0415\[nucl-ex\]\]. H.J. Specht \[for the NA60 Collaboration\], AIP Conf. Proc. [**1322**]{} (2010) 1. R. Rapp, G. Chanfray and J. Wambach, Nucl. Phys. [**A617**]{} (1997) 472. P. M. Hohler and R. Rapp, Phys. Lett. B [**731**]{} (2014) 103. R. Rapp and E. V. Shuryak, Phys. Lett. B [**473**]{} (2000) 13. J. Ruppert, C. Gale, T. Renk, P. Lichard and J. I. Kapusta, Phys. Rev. Lett. [**100**]{} (2008) 162301. R. Rapp and H. van Hees, Phys. Lett. B [**753**]{} (2016) 586. S. Afanasiev [*et al.*]{} \[PHENIX Collaboration\], Phys. Lett. **B679** (2009) 321.
E.L. Kryshen for the collaboration \[ALICE Collaboration\], Nucl. Phys. **A967** (2017) 273.
ATLAS Collaboration, ATLAS-CONF-2016-025.
A. van Hameren, M. K[ł]{}usek-Gawenda and A. Szczurek Phys. Lett. **B776** (2018) 84.
M. Klusek-Gawenda and A. Szczurek, Phys. Rev. C [**93**]{} (2016) 044912. J. Adam [*et al.*]{} \[ALICE Collaboration\] Phys. Rev. Lett. **116** (2016) 222301.
W. Shi, W. Zha and B. Chen, Phys. Lett. **B777** (2018) 399.
J. Adam [*et al.*]{} \[STAR Collaboration\], arXiv:1806.02295 \[hep-ex\].
S. R. Klein, Phys. Rev. C [**97**]{} (2018) 054903. W. Zha, L. Ruan, Z. Tang, Z. Xu and S. Yang, Phys. Lett. B [**781**]{} (2018) 182. T. Song, W. Cassing, P. Moreau and E. Bratkovskaya, arXiv:1806.09377 \[nucl-th\].
R. Arnaldi [*et al.*]{} \[NA60 Collaboration\], Eur. Phys. J. C [**61**]{} (2009) 711. C. A. Bertulani and G. Baur, Phys. Rept. [**163**]{} (1988) 299. G. Baur, K. Hencken, D. Trautmann, S. Sadovsky and Y. Kharlov, Phys. Rept. [**364**]{} (2002) 359. H. De Vries, C. W. De Jager and C. De Vries, Atom. Data Nucl. Data Tabl. [**36**]{} (1987) 495. E. Shuryak, Prog. Part. Nucl. Phys. [**62**]{} (2009) 48. U. Heinz and R. Snellings, Ann. Rev. Nucl. Part. Sci. [**63**]{} (2013) 123. C. Gale, S. Jeon and B. Schenke, Int. J. Mod. Phys. A [**28**]{} (2013) 1340011. R. Rapp, Adv. High Energy Phys. [**2013**]{} (2013) 148253. M. Urban, M. Buballa, R. Rapp and J. Wambach, Nucl. Phys. A [**673**]{} (2000) 357. T. Galatyuk \[HADES Collaboration\], Nucl. Phys. A [**967**]{} (2017) 680. G. Agakichiev [*et al.*]{} \[CERES Collaboration\], Eur. Phys. J. C [**41**]{} (2005) 475. P. Huck \[STAR Collaboration\], Nucl. Phys. A [**931**]{} (2014) 659. L. Adamczyk [*et al.*]{} \[STAR Collaboration\], Phys. Rev. C [**92**]{} (2015) 024912. A. Adare [*et al.*]{} \[PHENIX Collaboration\], Phys. Rev. C [**93**]{} (2016) 014904. S. Basu, T. K. Nayak and K. Datta, Phys. Rev. C [**93**]{} (2016) 064902. F. Antinori, P. Braun-Munzinger and S. Flörchinger, priv. comm. (2018).
[^1]: Also at *Faculty of Mathematics and Natural Sciences, University of Rzeszów, Pigonia 1, PL-35-310 Rzeszów, Poland*.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
The impact of the strongly attractive electromagnetic field of heavy nuclei on electrons in quasi-elastic $(e,e')$ scattering is often accounted for by the effective momentum approximation. This method is a plane wave Born approximation which takes the twofold effect of the attractive nucleus on initial and final state electrons into account, namely the modification of the electron momentum in the vicinity of the nucleus, and the focusing of electrons towards the nuclear region leading to an enhancement of the corresponding wave function amplitudes. The focusing effect due to the attractive Coulomb field of a homogeneously charged sphere on a classical ensemble of charged particles incident on the field is calculated in the highly relativistic limit and compared to results obtained from exact solutions of the Dirac equation. The result is relevant for the theoretical foundation of the effective momentum approximation and describes the high energy behavior of the amplitude of continuum Dirac waves in the potential of a homogeneously charged sphere. Our findings indicate that the effective momentum approximation is a useful approximation for the calculation of Coulomb corrections in $(e,e')$ scattering off heavy nuclei for sufficiently high electron energies and momentum transfer.
0.1 cm [**Keywords**]{}: Coulomb corrections, quasi-elastic electron scattering, effective momentum approximation. 0.1 cm [**PACS**]{}: 11.80.-m Relativistic scattering theory; 11.15.Kc Classical and semiclassical techniques; 11.80.Fv Approximations; 13.40.-f Electromagnetic processes and properties; 25.30.Fj Inelastic electron scattering to continuum; 25.30.Bf Elastic electron scattering; 25.70.Bc Elastic and quasielastic scattering
author:
- |
Andreas Aste, Dirk Trautmann\
Department of Physics and Astronomy, University of Basel,\
Klingelbergstrasse 82, 4056 Basel, Switzerland
date: 'July 4, 2007'
title: |
Focusing of high-energy particles\
in the electrostatic field of a homogeneously charged sphere\
and the effective momentum approximation
---
Introduction
============
Scattering experiments can be viewed as one of the very important tools of experimental particle physics since the famous Lord Ernest Rutherford scattering experiment of $\alpha$-particles off the nuclei within a gold foil in 1911 [@Rutherford]. To explore the structure of the nucleus, the main tool used today is electron scattering due to the transparency of the nuclear volume for electrons. E.g., inclusive $(e,e')$ scattering, where only the final electron is observed, provides information about the nuclear Fermi momentum by measuring the width of the quasi-elastic peak [@Whitney], or the high-momentum components of nucleon wave functions when the tail of the quasi-elastic peak is investigated [@Benhar95; @Rohe]. Information about infinite nuclear matter is obtained by extrapolating the mass number $A \rightarrow \infty$ [@Day89], and possible modifications of the nucleon form factors inside a nucleus are related to the Coulomb sum rules [@MezianiCoulomb]. However, although electrons with energies of typically some hundred MeV are used in the experiments, the distortion of the electron wave functions due to the strongly attractive electrostatic field of heavy nuclei can no longer be neglected, such that calculations in the plane wave Born approximation (PWBA) are no longer reliable.
Calculations using exact Dirac wave functions are feasible but cumbersome and difficult compared to the PWBA calculations. As a consequence, various approximate methods have been proposed in the past for the treatment of Coulomb distortions [@Lenz; @Knoll; @Giusti; @Rosenfelder; @Giusti88; @Traini88; @Traini95; @Rosenfelder80; @Kosik], and there is an extensive literature on the so-called eikonal approximation [@Yennie64; @Sucher; @Blankenbecler; @Wallace1; @Wallace2; @Abarbanel; @Aste0; @Aste1; @Aste2].
In this paper, we give a concise classical derivation of the effective momentum approximation (EMA), which has the advantage that one works with plane waves and which plays an important role in experimental data analysis. The classical high-energy results are compared to results obtained from exact solutions of the Dirac equation. Our findings concerning the correct use of the EMA are of actual importance, since there is now considerable theoretical and experimental interest in extracting longitudinal and transverse structure functions as a function of energy loss for fixed three-momentum transfer for a range of nuclei. Recently, a Thomas Jefferson National Accelerator Facility (TJNAF) proposal for quasi-elastic electron scattering measurement in the momentum transfer range $0.55$ GeV/c $\le$ $|\vec{q}|$ $\le$ $1.0$ GeV/c was approved such that the experiments will be performed in the near future using $^4$He, $^{12}$C, $^{56}$Fe and $^{208}$Pb as target nuclei [@jlab].
We shortly comment qualitatively on the connection between the distorted wave Born approximation (DWBA) and the EMA and its correct application. In DWBA, one calculates matrix elements with exact initial and final state electron wave functions. Unlike the plane waves with constant amplitude used in the PWBA, these wave functions are focused towards the nuclear region, and the local electron momenta are enhanced there due to the attractive positively charged nucleus. In the EMA, the focusing and the momentum transfer in the relevant nuclear region, where the nucleons get knocked, are accounted for by effective (average) values. It is important to mention that one must base EMA calculations indeed on average values, although in the literature, the use of effective values for the focusing and the effective momenta valid only in the center of the nucleus is widespread. However, the choice of such values is not appropriate, as will be explained in detail in this paper.
There are two equivalent methods for the correct application of the EMA. First, one may calculate the cross section from the corresponding theoretical PWBA expression for the $(e,e')$ scattering cross section with effective momenta. This introduces an artificially enhanced phase space for the final state electron, since also in the DWBA, the phase space is given by the undistorted asymptotic momenta of the final state particles. However, this enhanced phase space accidentally accounts for the focusing effect on the final state electron with a high level of accuracy. Because the initial focusing has not yet been taken into account, one has to multiply the cross section calculated so far additionally by the effective focusing factor of the initial state electron.
Another equivalent approach is to factorize the theoretical expression for the $(e,e')$ cross section into the Mott cross section given by eq. (\[Mott2\]) and a response function according to eq. (\[Mott1\]). The interesting point is that the impact of the focusing cancels against the modification of the momentum transfer in the Mott cross section. Accordingly, one may calculate the EMA cross section by leaving the Mott part unchanged and by evaluating the response function with the momenta replaced by their effective values.
A critical overwiew on the history of the effective momentum approximation and its correct and incorrect application can be found in [@Traini2001].
Quasi-elastic scattering
========================
![Quasi-elastic $(e,e')$ scattering cross section data taken at Saclay for initial electron energy $\epsilon_i=485$ MeV and electron scattering angle $\Theta_e=60^o$.[]{data-label="quasi"}](quasi.eps){width="10.0"}
In order to illustrate the importance of Coulomb corrections for quasi-elastic $(e,e')$ scattering, we shortly review the basic properties of this scattering process. For this purpose, we envisage an electron with initial and final asymptotic four-momenta $k^\mu_{i,f}=(\epsilon_{i,f},\vec{k}_{i,f})$, which scatters off a nucleon. We will always set $\hbar=c=1$ in the following, and for highly relativistic electrons we have $\epsilon_{i,f}=|\vec{k}_{i,f}|$. Additionally, we assume that the nucleon inside the nucleus is at rest. Neglecting interactions on the nucleon with its surrounding such that the nucleon can be considered quasi-free, the initial and final momenta of the nucleon are given by $p^\mu_i=(m_n,\vec{0})$ and $p^\mu_f=(E_f,\vec{p}_f)=(m_n+\omega,\vec{k}_i-\vec{k}_f)$, where $\omega=k^0_i-k^0_f$ is the energy transfer and $\vec{q}=\vec{k}_i-\vec{k}_f$ the three-momentum transfer of the electron to the nucleon. From four-momentum conservation $$q^\mu=(k_i^\mu-k_f^\mu)=(p_f^\mu-p_i^\mu)$$ we obtain from the four-momentum transfer squared $Q^2$ $$-Q^2=q_\mu q^\mu=2m_n^2-2m_n E_f,$$ and consequently $\omega=(E_f-m_n)=\frac{Q^2}{2 m_n}$. Therefore, under the simplifying assumptions made above, the $(e,e')$ scattering cross section as a function of the energy transfer for fixed electron scattering angle $\Theta_e$ should exhibit a peak where $$\omega=\frac{Q^2}{2 m_n}. \label{peakempirical}$$
Fig. \[quasi\] shows such a typical experimental curve from measurements taken at Saclay [@Saclay]. First, one observes that the peak has a width which is basically due to the Fermi motion of the nucleons. Second, the peak is shifted with respect to the empirical formula eq. (\[peakempirical\]), which predicts $\omega_{peak} \simeq 100$ MeV, to a value of nearly $140$ MeV. A phenomenological description of this observation could be given within the Fermi gas model by the observation that eq. (\[peakempirical\]) does not take into account that an average removal energy $\bar{E}_{rem}$ is necessary to remove a nucleon from the nucleus which is larger than the average binding energy $\bar{E}_{bind}$ of a nucleon inside the nucleus. E.g., for $^{208}$Pb with $\bar{E}_{bind} \simeq 20$ MeV, a two-parameter fit for the Fermi momentum $k_F$ and the removal energy $\bar{E}_{rem}$ leads to $k_F \simeq 265$ MeV and $\bar{E}_{rem} \simeq 44$ MeV [@Whitney]. The higher value of the removal energy also incorporates correlation effects due to the short range interaction of the nucleons [@Benhar94]. However, there is a significant non-quasielastic background present in Fig. \[quasi\], which if removed would make the peak appear around $130$ MeV, a value which is not so different from what one would expect from the binding energy, putting the observations made above into perspective. Furthermore, the momentum of the electron in the nuclear vicinity is enhanced due to the attraction of the nucleus, which induces an additional positive shift of the peak. This leads us to the idea of effective momenta. From a classical point of view, the momentum of a highly relativistic electron which moves virtually on a straight line is locally dependent and given by $$\vec{k}_{i,f} (\vec{r}) = (k_{i,f}-V(\vec{r})) \hat{k}_{i,f},$$ where $\hat{k}_{i,f}$ is the unit vector in direction of $\vec{k}_{i,f}$ $k_{i,f}=|\vec{k}_{i,f}|$, and $V(\vec{r})$ is the potential energy of the electron in the electrostatic field of the nucleus. This local change of the momentum of, e.g., the incoming particle with momentum $\vec{k}_i=k_i \hat{k}_i$ is taken into account by the eikonal approximation through a modification of the plane wave part of the free wave function describing the initial state of the particle. Defining the relativistic eikonal phase $$\chi_i(\vec{r})=-\int \limits_{-\infty}^{0} V(\vec{r}+
\hat{k}_i s) ds=
-\int \limits_{-\infty}^{z} V(x,y,z') dz' \,$$ if we choose $\vec{k}_i=k_z^i \hat{{\bf{e}}}_z$, the free electron spinor used in PWBA calculations $$\Psi_i (\vec{r})=u_{s_i}(\vec{k_i}) e^{i \vec{k}_i\vec{r}}$$ is replaced by $$\Psi_i (\vec{r})=u_{s_i}(\vec{k_i}) e^{i \vec{k}_i\vec{r}+i\chi_i(\vec{r})}$$ in the corresponding eikonal distorted wave Born approximation (EDWBA). $u_{s_i}(\vec{k_i})$ is the constant spinor which depends on the spin (helicity) and momentum of the particle. As desired, the dominant longitudinal $z$-component of the momentum $p_z$ is then recovered via $$p_z e^{i k_z^i z+i\chi_i}=-i \partial_z e^{i k_z^i z+i\chi_i}=
(k_z^i-V)e^{i k_z^i z+i\chi_i}.$$ The final state wave function is constructed analogously by the replacement $e^{i \vec{k}_f\vec{r}} \rightarrow e^{i \vec{k}_f\vec{r}-i\chi_f(\vec{r})}$, where $$\chi_f(\vec{r})=-\int \limits_{0}^{\infty} V(\vec{r}+
\hat{k}_f s') ds' \, .$$ However, this approximation does not yet include the fact that also the amplitude of the electron wave function corresponding to initial and final asymptotic momenta $\vec{k}_{i,f}$ is modified by the attractive nucleus. An improved version of the eikonal approximation thus should read $$\Psi_{i,f}(\vec{r})=f_{i,f}^{1/2} (\vec{r})
u_{s_{i,f}}(\vec{k}_{i,f}) e^{i \vec{k}_{i,f}\vec{r}+i\chi_{i,f}(\vec{r})} ,$$ such that the electron probability density is locally enhanced by focusing factors $f_{i,f} (\vec{r})$.
A simpler strategy than the eikonal approximation, which will eventually lead to the EMA and which avoids the introduction of non-planar wave functions, is to average the locally dependent momentum over the nuclear volume, such that effective momenta $\vec{k}_{i,f}^{eff}$ are obtained $$\vec{k}_{i,f}^{eff}=\langle \vec{k}_{i,f} (\vec{r}) \rangle=
\frac{\int \vec{k}_{i,f} (\vec{r})
\rho(\vec{r}) d^3 r}{ \int \rho(\vec{r}) d^3 r},$$ with $\rho(\vec{r})$ representing a reasonable nuclear density profile. If both the charge and the nuclear density are approximated by a homogeneous distribution inside a sphere with radius $R$ $$\rho(\vec{r}) = \left\{ \begin{array}{ccc}
const. & : & |\vec{r} \, | \leq R \\
0 & : & |\vec{r} \, |>R
\end{array} \right. \quad ,$$ then it is straightforward to show that the effective momenta are given by $$\vec{k}^{eff}_{i,f} = \Bigl( k_{i,f}-\frac{4}{5} V(0) \Bigr) \hat{k}_{i,f} =
(k_{i,f}-V_{eff}) \hat{k}_{i,f}, \label{effmom}$$ and the potential energy of an electron $V(0)$ in the center of the nucleus is given by $V(0)=-\frac{3 \alpha Z}{2 R}$, where $\alpha=e^2/4 \pi$ is the fine structure constant and $e$ the elemental charge. Accordingly, one can define now an effective four-momentum transfer squared $Q^2_{eff}$. The effective potential $V_{eff}=4 V(0)/5$ is indeed the average value of the potential $V$ generated by the homogeneous charge distribution inside the sphere itself. Eventually, since the present discussion has a phenomenological character due to the complex and partially uncertain structure of the nuclear current, one may modify eq. (\[peakempirical\]) to an even more general form $$\omega=\frac{Q_{eff}^2}{2 {\tilde{m}}_n} + {\tilde{E}}_{rem},$$ where ${\tilde{m}_n}$ and ${\tilde{E}}_{rem}$ are a phenomenological (momentum-dependent) nucleon mass and a phenomenological removal energy, respectively. Replacing $Q^2$ by $Q^2_{eff}$ leads to an additional peak shift of $\sim 8$ MeV in the present example.
![Quasi-elastic electron scattering off a heavy nucleus. Within a strongly simplified picture, the process can be viewed as scattering of the electron off the constituents (mainly nucleons) of the nucleus via exchange of a ‘hard short-range’ photon.[]{data-label="ee"}](ee.eps){width="10"}
Viewing quasi-elastic scattering as a nucleon knock-out process provides only a poor picture of the actual physical processes taking place inside the nucleus and for details we refer to the literature [@Benhar]. What is important for the forthcoming section is the fact that the electron interacts with the nuclear medium by exchange of photons, and that the hard scattering process can be viewed as a [*quasi-local*]{} process. E.g., for $\epsilon_i=485$ MeV and $\omega=160$ MeV, the four-momentum transfer is $Q^2=(397 \, \mbox{MeV})^2$. Taking into account that $\hbar c = 197
\, \mbox{MeV fm}$, the virtuality $Q^2$ of the exchanged photon corresponds to a typical space-time length scale of $0.5$ fm, which is much smaller than the size of the nucleus, as depicted in Fig. \[ee\].
Effective momentum approximation
================================
The differential cross section for single nucleon knockout is given by [@Udias] $$\frac{d^4 \sigma}{d \epsilon_f d\Omega_f dE_f d\Omega_f}=
\frac{4 \alpha^2}{(2 \pi)^9} \epsilon_f^2 E_f p_f
\delta(\epsilon_i \! + \! E_A \! - \! \epsilon_f \!- \! E_f
\! - \! E_{A-1}) \sum \limits_{}^{ - \! \! -} \, |W_{if}|^2 \label{crosssection},$$ with the matrix element $$W_{if}=\int \! d^3 x \! \int d^3 y \! \int \! d^3 q \, \Bigl[
j_\mu^e(\vec{x}) \, \frac{e^{-i\vec{q} (\vec{x}-\vec{y})}}{q_\mu^2} \,
J^\mu_N(\vec{y}) \Bigr], \label{matrixelement}$$ where $J^\mu_N(\vec{y})$ is the nucleon current obtained from some suitable nuclear model, the $\sum \limits_{}^{ - \! \! -}$ in eq. (\[crosssection\]) indicates the sum (average) over final (initial) polarizations, and $E_A$, $E_{A-1}$ is the energy of the initial and final nucleus, respectively. In the PWBA, the electron current is given by $$j^\mu(\vec{x})=
\bar{u}_{s_f}(\vec{k}_f) \gamma^\mu u_{s_i}(\vec{k}_i)
e^{i \vec{k}_i \vec{r}-i \vec{k}_f \vec{r}},$$ where ${u}_{s_i},{u}_{s_f}$ are initial/final state plane wave electron spinors corresponding to the initial/final electron momentum $\vec{k}_{i,f}$ and spin $s_{i,f}$. In the DWBA, exact solutions of the Dirac equation are used for electrons instead of plane waves. The usual procedure to calculate the inclusive $(e,e')$ cross section is to sum over all the individual nucleon knockout cross sections for all protons and neutrons in the nucleus under consideration.
The basic idea of the effective momentum approximation (EMA) is to describe the electron wave functions by modified plane waves $$e^{i \vec{k}_{i,f} \vec{r}} \rightarrow \frac{{k}_{i,f}'}
{k_{i,f}} e^{i \vec{k}_{i,f}'' \vec{r}} \, , \label{EMA}$$ which account for the enhanced electron density and momentum in the nuclear region. Here, $k_{i,f}'$ and $k_{i,f}''$ denote effective momenta which need not necessarily be identical. It will be one of the main results of the forthcoming section, that for high electron energies and the electrostatic potential of a homogeneously charged sphere, $k_{i,f}'=k_{i,f}''=k_{i,f}^{eff}$ is indeed fulfilled, i.e., at high energies, the effective (average) focusing factor is given by $$f_{i,f}=\frac{\int f_{i,f} (\vec{r})
\rho(\vec{r}) d^3 r}{ \int \rho(\vec{r}) d^3 r}=\Biggl(\frac{k_{i,f}^{eff}}{k_{i,f}}
\Biggr)^2.$$ We will therefore identify the $k_{i,f}'=k_{i,f}''=k_{i,f}^{eff}$ in the sequel. When the exact wave functions appearing in the matrix element eq. (\[matrixelement\]) are replaced by the corresponding effective wave functions, the momentum integral in eq. (\[matrixelement\]) can be trivially performed and replaced basically by a constant factor $1/q_{\mu,eff}^2=-1/Q_{eff}^2$. This expresses the fact that the virtual photon emitted by the electron is actually harder than if no attractive potential were present, since the electron is accelerated to higher momenta in the nuclear vicinity, and $1/q_{\mu,eff}^2$ is the photon virtuality averaged over the nuclear volume.
Note that the reason why the replacement of the locally dependent wave function amplitudes and momenta by effective values makes sense is rooted in the local character of the scattering process mentioned above. E.g., if the virtual photon would propagate over distances comparable to the size of the nucleus, then nucleons could also be knocked by photons which were emitted outside the nucleus, such that an averaging of the focusing and the local momenta inside the nuclear interior would not make sense. The correct mathematical counterpart of this pictorial description can be found in [@Knoll]. For the EMA to hold, it is mandatory that the wave lengths of the electron and the virtual photon are significantly smaller than the nuclear radius, i.e. $\epsilon_f > 200$ MeV and $Q^2>(200 \, \mbox{MeV})^2$ corresponding to a length scale of $1$ fm should be required for $^{208}$Pb [@Jin; @Aste3]. Note also that the enhancement of the wave function amplitudes is not very large at high energies. One can write $$f^{1/2}_{i,f}(\vec{r})=1+\delta_{i,f}(\vec{r}), \quad |\delta(\vec{r})| \ll 1,$$ and we may therefore neglect higher order terms in the $\delta$’s in formal expressions, like $$\langle f_i^{1/2}(\vec{r}) f_f^{1/2}(\vec{r}) \rangle =
\langle (1+\delta_{i,f}(\vec{r})) (1+\delta_{i,f}(\vec{r})) \rangle \simeq
\langle f_i^{1/2} (\vec{r})\rangle \langle f_f^{1/2}(\vec{r}) \rangle,$$ and one may equate expressions like $$f_{i,f} (\vec{r}) = (1 + \delta_{i,f}(\vec{r}))^2 \simeq 1+ 2 \delta_{i,f}(\vec{r}).$$
It is instructive to calculate the impact of the focusing factors on the size of the cross section for a typical example. E.g., if we consider electron scattering off $^{208}$Pb for $|k_{i}|=485$ MeV/c and $\omega=100$ MeV, we have $V(0)=-25$ MeV and $4V(0)/5=-20$ MeV such that $k_{i}^{eff}$ is given by $(485+20)$ MeV, and $k_{f}^{eff}=(385+20)$ MeV. The focusing factors enter the cross section both linearly via the matrix element squared, enhancing the cross section by a factor of $(k_{i}^{eff}/k_i)^2(k_{f}^{eff}/k_f)^2=1.2$.
There are two different, but equivalent strategies to calculate cross sections in the EMA framework. First, the EMA cross section can be calculated by replacing the electron momenta $\vec{k}_{i,f}$ by the effective momenta $\vec{k}_{i,f}^{eff}$ in the (theoretical) expression for the quasi-elastic scattering cross section (accordingly, the energies $\epsilon_{i,f}$ must be replaced by $|\vec{k}^{eff}_{i,f}|$). The cross section obtained this way must be multiplied subsequently by the factor $(k_{i}'/k_i)^2$ which accounts for the focusing of the incoming electron wave in the nuclear center. The focusing factor $(k_{f}'/k_f)^2$ for the scattered electron is already contained in the artificially enhanced phase space factor of the final state electron, if $k'_f=k''_f$ is presumed. Second, the cross section for inclusive quasi-elastic electron scattering can also be written by the help of the total response function $S_{tot}$ as $$\frac{d^2 \sigma_{_{PWBA}}}{ d \Omega_f d\epsilon_{f}}=
\sigma_{Mott} \times S_{tot}(|\vec{q} \, |,\omega,\Theta_e),
\label{Mott1}$$ where the Mott cross section is given by ($q_\mu^4=Q^4$) $$\sigma_{Mott}=4 \alpha^2 \cos^2(\Theta_e/2) \epsilon_f^2/q_\mu^4.
\label{Mott2}$$ The Mott cross section remains unchanged when it gets multiplied by the EMA focusing factors and the momentum transfer $q_\mu^4$ is replaced by its corresponding effective value. A short calculation shows indeed that ($\epsilon_{i,f} \gg m$) $$\frac{Q_{eff}^2}{Q^2}=\frac{k_i^{eff} k_f^{eff}}{k_i k_f}, \quad
\frac{f_i (\epsilon_{f}^{eff})^2}{Q_{eff}^4}=\frac{\epsilon_f^2}{Q^4}.$$ Therefore, the EMA cross section can also be obtained from (\[Mott1\]) by leaving the Mott cross section unchanged and by replacing $S_{tot}(|\vec{q} \, |,\omega,\Theta_e)$ by the effective value $$S_{tot}(|\vec{q}_{eff}|, \omega,\Theta_e)=
S_{tot}(|\vec{k}_i^{eff}-\vec{k}^{eff}_f|, \omega,\Theta_e),$$ since the effect of replacing $q_\mu^4$ by its effective value in the Mott cross section is to exactly divide away the initial state focusing factor and the final state focusing factor which is generated by the replacement $\epsilon_f \rightarrow \epsilon_f^{eff}=|k_f^{eff}|$.
The classical focusing factor
=============================
![Electrons incident on an nucleus with impact parameter $b_0$.[]{data-label="figfoc"}](focus.eps){width="10.8cm"}
The focusing factor can be derived approximately from a classical toy model according to Figs. \[figfoc\] and \[regions\]. We consider the trajectories of an ensemble of highly relativistic particles approaching a nucleus located in the center of the $(b,z)$-coordinate system. The particles shall move in $z$-direction with equal velocity and with an asymptotic impact parameter in the range between $b_0$ and $b_0+db_0$. The longitudinal velocity of the particles can be taken as the speed of light, since the particles are highly relativistic and changes of the velocity in transverse direction and the kinetic energy cause a negligible second-order effect to the longitudinal component. Therefore, one may adopt the straight-line approximation by setting $z(t)=ct=t$, $r(t)=\sqrt{b_0^2+t^2}$, and the impact parameter will be considered constant at certain stages of our calculation. Due to the attractive nucleus, the original impact parameter $b_0$ is reduced to $b(b_0,z)$ as a particle moves along its trajectory, such that the particle density at $z$ is increased by a focusing factor $f$ which is given by the ratio of the area of two annuli with radii $b_0, b_0+db_0$ and $b(b_0,z), b(b_0+db_0,z)$: $$f^{-1}(b_0,z) \simeq f^{-1}(b(b_0,z),z)=\frac{\partial b(b_0,z)}{\partial b_0}
\frac{b(b_0,z)}{b_0}. \label{focusing_general}$$
In the following, we will calculate $b(b_0,z)$ for an electron in the potential of a homogeneously charged sphere with radius $R$ and charge $eZ$, given by $$V_{hom}(r) = \left\{ \begin{array}{ccc}
-\frac{\alpha Z}{R} \Bigl( \frac{3}{2}- \frac{r^2}{2 R^2} \Bigr) & : & r \leq R \\
-\frac{\alpha Z}{r} & : & r>R
\end{array} \right. \quad .$$ since the potential of a homogeneously charged sphere provides a simple but quite realistic model for the electromagnetic field of a heavy nucleus like, e.g., $^{208}$Pb, where one has $R \simeq 7.1 \, \mbox{fm}$ and $Z=82$.
The force $f_t$ acting on the particle in transverse direction is given by $$f_t = -\frac{b}{r} \frac{\partial V_{hom}(r)}{\partial r} \sim
-\frac{b_0}{r} \frac{\partial V_{hom}(r)}{\partial r}, \label{transverse_force}$$ In eqns. (\[focusing\_general\]) and (\[transverse\_force\]), we made use of the straight-line assumption by replacing $b$ by $b_0$. Also in the forthcoming, we will sometimes replace $b$ by $b_0$ or use these two quantities synonymously where such a substitution is adequate.
![The three different regions according to the case distinction in the text.[]{data-label="regions"}](regions.eps){width="7.8cm"}
To calculate the transverse acceleration of the particle due to the attractive Coulomb field, we distinct three cases (see Fig. \[regions\]). In the first case, we consider the region where the particle moves solely in the $1/r$-field according to the straight-line approximation, i.e. where we have $b_0>R$ (or $b_0<-R$, if we formally allow negative impact parameters), plus the region with $b_0<R$ and $z<-\sqrt{R^2-b_0^2}$. The transverse force is then given by $$f_t^I=-\frac{\alpha Z b_0}{r^3}.$$ Correspondingly, we obtain for the transverse acceleration, taking into account that the “transverse mass” of a relativistic particle is given by its energy $E$, which is also considered as constant: $$a_t^I=\dot{v}_t^I=
-\frac{\alpha Z}{E}\frac{b_0}{r^3}=
-\frac{\alpha Z}{E}\frac{b_0}{\sqrt{t^2+b_0^2}^3},$$ where $r$ is the distance of the particle from the nuclear center, and $$v_t^I(b_0,t)=\int \limits_{-\infty}^{t} a_t^I (b_0,t') dt'=
-\frac{\alpha Z}{E} \frac{t+\sqrt{t^2+b_0^2}}
{b_0 \sqrt{t^2+b_0^2}}=
-\frac{\alpha Z}{E} \frac{t+r}
{b_0 r}. \label{velocity}$$ Note that from $v_t^I(t \rightarrow \infty)=-\frac{2\alpha Z}{E}$ we obtain for a pure $1/r$-Coulomb field the well-known transverse momentum transfer $$\Delta k_t = \frac{2 \alpha Z}{b_0}.$$ Furthermore, we obtain from (\[velocity\]) $$b^I(b_0,z)=b_0+\int \limits_{-\infty}^{z} v_t^I(b_0,t') dt'
=b_0-\frac{\alpha Z}{E} \frac{z+r}{b_0}. \label{classbI}$$ A short calculation yields the focusing factor $$f^I=\Biggl[ 1-\frac{\alpha Z}{E} \Biggl( \frac{1}{r}+\frac{z+r}{b_0^2}
\Biggr) \Biggr]^{-1} \Biggl[ 1-\frac{\alpha Z}{E} \Biggl( \frac{z+r}{b_0^2}
\Biggr) \Biggr]^{-1}.
\label{focus2}$$ Since we are interested in the high energy behavior, we keep in eq. (\[focus2\]) only the relevant zeroth- and first-order terms in $\alpha Z/E$. For the focusing factor in region I we obtain the simple result $$f^I(b_0,z)=1+\frac{\alpha Z}{E} \frac{1}{r}. \label{classicalfocus1}$$ The calculation for regions II and III are a bit more involved, but can be performed along the same lines as above.
We calculate now the focusing inside the charged sphere (region II), which gets traversed by particles with $b_0<R$. Inside the sphere, the transverse acceleration of the particles is due to the harmonic oscillator potential generated by the homogeneous charge distribution. Correspondingly, we have $$a_t^{II} = -\frac{b}{r} \frac{\alpha Z}{E} \frac{r}{R^3}
\sim -\frac{\alpha Z}{E} \frac{b_0}{R^3},$$ and for the transverse distance from the $z$-axis we obtain after a short calculation $$b^{II} (b_0,z)=b_0-\frac{\alpha Z}{E} \Biggl[
\frac{R-\tilde{R}}{b_0} + \frac{\frac{b_0}{2} \Bigl(
z+\tilde{R} \Bigr)^2}{R^3} +
\frac{\Bigl( R - \tilde{R} \Bigr) \Bigl( \tilde{R}+z \Bigr)}
{b_0 R} \Biggr].$$ Above, we have introduced the abbreviation $\tilde{R}=\sqrt{R^2-b_0^2}$. Note that the first term in the bracket above describes the transverse shift of the particles when they arrive on the surface of the charged sphere according to eq. (\[classbI\]), where $z=-\tilde{R}$. The second term is due to the transverse acceleration of the particles inside the sphere, and the last term in the expression above is generated by the transverse velocity $v_t^I (b_0,-\tilde{R})$ which is reached by the particles when they cross the border of the sphere. A straightforward calculation leads to the following result for the focusing factor at first order in $\alpha Z/E$: $$f^{II}(b_0,z)=1 +\frac{\alpha Z}{ER} \Bigl(3+ 3 \frac{z \tilde{R}}{R^2} +
\frac{z^2}{R^2}-2b_0^2/R^2 \Bigr).$$ In the center of the nucleus, the particle density is enhanced by a factor $$f^{II}(0)=f^{II}(b_0=0,z=0)=1+\frac{3 \alpha Z}{ER},$$ however, the average focusing factor inside the sphere is given by the volume integral $$\bar{f}^{II}= \frac{\int \limits_{b_0^2+z^2<R^2} f^{II}(b_0,z) dV}{\frac{4 \pi}{3} R^3}
=\frac{4}{5} f^{II}(0),$$ i.e. one obtains the focusing factor used in the effective momentum approximation, where the increased particle probability density near the nucleus is taken into account by multiplying the particle’s Dirac wave function by a suitable factor $f^{1/2} \simeq (\bar{f}^{II})^{1/2}$.
Finally, we consider the ‘shadow region’ of the nucleus (region III). We calculate first the transverse velocity of the particle when it arrives in region III in three steps. First, the particle moves inside the $1/r$-field and reaches a transverse velocity $$v_1=v_t^I(b_0,-\tilde{R})=-\frac{\alpha Z}{E} \frac{R-\tilde{R}}{b_0 R}$$ at the surface where it enters the sphere. Inside the sphere, the particle undergoes a constant transverse acceleration $a_t^{II}$ for $t \! \in \! [-\tilde{R},\tilde{R}]$. Therefore, the particle gains an additional velocity $$v_2=-\frac{\alpha Z}{E} \frac{2 b_0 \tilde{R}}{R^3}.$$ In the downstream region III, the particle is again moving in the field of a point-like charge, and the transverse acceleration is given by $$a_t^{III}=-\frac{\alpha Z}{E} \frac{b_0}{\sqrt{t^2+b_0^2}^3},$$ such that we end up with ($r=\sqrt(t^2+b_0^2)$) $$v_t^{III}=v_1+v_2+\int \limits_{\tilde{R}}^{t} a_t^{III}(b_0,t') dt'=$$ $$-\frac{\alpha Z}{E} \Biggl[
-\frac{\tilde{R}}{b_0 R} +\frac{1}{b_0} +\frac{2 b_0 \tilde{R}}{R^3}
\Biggr]-\frac{\alpha Z}{E} \Biggl[ \frac{t+r}{b_0 r}
-\frac{\tilde{R}}{b_0 R} - \frac{1}{b_0} \Biggr]=$$ $$-\frac{\alpha Z}{E} \Biggl[ \frac{t+r}{b_0 r}
-\frac{2 \tilde{R}}{b_0 R} + \frac{2 b_0 \tilde{R}}{R^3} \Biggr].$$ The transverse distance of the particles from the $z$-axis in region III is therefore given by $$b^{III}(b_0,z)=b^{II}(b_0,\tilde{R})+\int \limits_{\tilde{R}}^{z}
v_t^{III}(b_0,t) dt=$$ $$b_0-\frac{\alpha Z}{E} \Biggl[ -\frac{R}{b_0} + \frac{4 b_0}{R} +
\frac{\tilde{R}}{b_0}-\frac{2 b_0^3}{R^3} \Biggr]+
\int \limits_{\tilde{R}}^{z}
v_t^{III}(b_0,t) dt=$$ $$b_0-\frac{\alpha Z}{E} \Biggl[ \frac{z}{b_0}+\frac{2 b_0 z \tilde{R}}{R^3}
+\frac{\sqrt{z^2+b_0^2}}{b_0}-\frac{2 z \tilde{R}}{b_0 R} \Biggr].
\label{bregionIII}$$ For the focusing factor we obtain from eq. (\[bregionIII\]) $$f^{III}(b_0,z)=1+\frac{\alpha Z}{E} \Biggl[ \frac{6z}{R \tilde{R}}
+\frac{1}{\sqrt{z^2+b_0^2}}-\frac{6z b_0^2}{R^3 \tilde{R}} \Biggr].$$
We finally summarize the results as follows. An attractive nucleus modeled by a homogeneously charged sphere acts like a focusing lens on an ensemble of classical particles incident on the nucleus with impact parameter $b_0$ on quasi-straight trajectories parallel to the $z$-axis. For highly relativistic particles, the particle density is enhanced by a focusing factor $$f(b_0,z)=1+ \frac{\alpha Z}{E} \Phi(b_0,z)$$ with ($r=\sqrt(b_0^2+z^2)$) $$\Phi(b_0,z) = \left\{ \begin{array}{ccc}
\frac{1}{r} & : & I \\
\frac{3}{R}+ 3 \frac{z \tilde{R}}{R^3} +
\frac{z^2}{R^3}-2b_0^2/R^3 & : & II \\
\frac{1}{r}+\frac{6z}{R \tilde{R}}
-\frac{6z b_0^2}{R^3 \tilde{R}} & : & III
\end{array} \right. \quad . \label{fullfoc}$$ The typical deviation of the focusing from unity in the nuclear interior is of the order of $3 \alpha Z/ER$ or $V_{hom}(0)/E$; this ratio should be considered as the expansion parameter for higher order corrections to the focusing, which become irrelevant at high energies.
Fig. \[classical\] shows a surface plot of the universal function $\Phi(b_0,z)$.
In order to compare the classical focusing to the results obtained by solving the Dirac equation exactly [@Aste3; @TBR83; @Pauli], we define $$\Phi(E;b,z):=\frac{E}{\alpha Z} (\rho(E;b,z)-1),$$ where $\rho(E;b,z)=\Psi(E;\vec{r})^\dagger \Psi(E;\vec{r})$ is the axially symmetric probability density of the Dirac wave function $\Psi(E;\vec{r})$ of an electron incident with asymptotic momentum $\vec{k}=
\sqrt{E^2-m^2} \hat{z} \simeq E \hat{z}$ and spin parallel to the $z$-axis. Results are shown in Figs. \[classical\] and \[longitrans\]. The Dirac density indeed approaches the classical limit for high electron energies. Note that the focusing of the Dirac wave function is clearly underestimated by the classical high-energy approximation in region III, where the straight-line assumption starts to break down. However, in the case of quasi-elastic electron scattering, the relevant region is the interior of the nucleus, where the focusing is described in a satisfactory way for electron energies above $200$ MeV as shown in Figs. \[classical\], \[ratio\] and \[longitrans\].
![The ratio $\Phi(E,b,z)/
\Phi(b,z)$ for $E=200 \, \mbox{MeV}$ in region II.[]{data-label="ratio"}](ratio.eps){width="8.5cm"}
We note that Knoll [@Knoll] derived the focusing effect from a high energy partial wave expansion, following previous results given by Lenz and Rosenfelder [@Lenz; @Rosenfelder]. For the incoming particle wave expanded around the center of the nucleus he obtained $$\Psi_i(\vec{r})=e^{i \delta_i} (\tilde{k}_i/k_i) e^{i \vec{\tilde{k}}_i \vec{r}} \times$$ $$\{1+a_1r^2-2a_2\vec{\tilde{k}}_i\vec{r}+i a_1 r^2 \vec{\tilde{k}}_i \vec{r}+ia_2 [
(\vec{\tilde{k}}_i \times \vec{r})^2+
\vec{\sigma} (\vec{\tilde{k}}_i \times \vec{r})] \} u_{s_i}(\vec{k}_i),
\label{expknoll}$$ where $\delta_i$ is a phase, $\vec{\tilde{k}}_i$ is an effective momentum parallel to $\vec{k}_i$ calculated by using the central potential value $\tilde{k}_i=k_i-V_{hom}(0)$, and $\vec{\sigma}$ acts on the spinor $u_{s_i}(\vec{k}_i)$ to describe spin dependent effects, which are negligible in our cases of interest with definite helicity. An analogous equation holds for the distortion of the outgoing wave. The parameters $a_{1,2}$ depend on the shape of the potential. For a homogeneously charged sphere with radius $R$ they are given by $$a_1=-\frac{\alpha Z}{6 \tilde{k_i} R^3} , \quad
a_2=-\frac{3 \alpha Z}{4 {\tilde{k}_i}^2 R^2} \label{para} ,$$ and the central potential value is given by $V_{hom} (0)=-\frac{3}{2} \frac{\alpha Z}{R}$. The increase of the amplitude of the wave while passing through the nucleus is described mainly by the $-2a_2\vec{\tilde{k}}_i \vec{r}$-term, the $a_1 r^2$-term accounts for a decrease of the focusing also in transverse direction. Performing the replacements $r^2=b^2+z^2$, $\vec{\tilde{k}}_i \vec{r}= \tilde{k}_i z$ and $|\vec{\tilde{k}}_i \times \vec{r}|= \tilde{k}_i b$, one obtains from eq. (\[expknoll\]) for a particle with spin parallel to the momentum and energy $E=k_i$ up to second order in $b$ and $z$ and first order in $\alpha Z$ $$\Psi_i(\vec{r})^\dagger \Psi_i(\vec{r}) \simeq \Biggl[
1+\frac{\alpha Z}{ER} \Biggl( 3 + \frac{3z}{R}+\frac{z^2+b^2}{3 R^2} \Biggr) \Biggr]
u_{s_i}(\vec{k}_i)^\dagger u_{s_i}(\vec{k}_i),$$ i.e. the correct linear term in $z$ in eq. (\[fullfoc\]) is recovered, however, the transverse decay of the focusing is strongly suppressed in the $V_{eff}=4 V(0)/5$ expansion given by Knoll, which is therefore not suited to describe the focusing for $b \neq 0$. The use of the expansion eq. (\[expknoll\]) is the reason why the $(e,e')$ cross sections in [@Aste1] are overestimated.
Calculations using exact Dirac wave function show that the effective electron momenta are very well described by an effective potential $V_{eff}=4 V(0)/5$ [@Aste3]. Furthermore, our findings indicate that also the focusing can be described in an accurate way by the same effective potential value. This demonstrates the validity of the EMA as a valuable tool for the description of Coulomb distortion effects. One may observe from Fig. \[longitrans\] that the exact focusing for finite energies is slightly larger than in the high energy limit. This leads to a minor amplification of the DWBA cross section compared to the EMA result. Exact calculations, which will be presented in a forthcoming paper, show that this effect is only of the order of $2$% in the region of the quasielastic peak for typical kinematical values used in experiments, e.g. for an initial electron energy of $\epsilon_i=485$ MeV and scattering angle $\Theta_e=60^o$, or $\epsilon_i=310$ MeV and $\Theta_e=143^o$.
Final remarks and conclusions
=============================
The high energy trajectory of a charged classical particle moving in the field of a homogeneously charged field was investigated and related through the quantum-classical correspondence principle to the probability density of exact continuum wave functions obtained as solutions of the Dirac equation. As a result, a universal function $\Phi$ was found which allows to describe the high energy behavior of the amplitude of Dirac electron wave functions with definite helicity. The focusing in the downstream side of the charge distribution converges slowly towards the high-energy limit described by $\Phi$, however, the universal function $\Phi$ provides an accurate description of the focusing inside the charged space region, which can be considered as a model for the charge distribution of a heavy nucleus. As a consequence, it is found that both the effective (average) momenta and the average focusing can be described by a common effective potential $V_{eff}=4 V(0)/5$ in the case of a homogeneously charged sphere, despite the fact that the local classical momenta exhibit the same spherical symmetry as the electrostatic potential, whereas the axially symmetric focusing is smaller in the upstream side and larger in the downstream side of the nucleus.
Our findings establish the role of the EMA as a valuable semiclassical method for the analysis of Coulomb corrections in $(e,e')$ scattering. They also indicate that the analysis of experimental data based on calculations of Kim [*et al.*]{} should be revisited [@Kosik; @JourdanCoulomb], and support the strategy in previous works concerning the extraction of the longitudinal and transverse response functions in medium-weight and heavy nuclei [@MezianiCoulomb]. However, it is also advisable to await new experimental data which will hopefully be accessible in the near future [@jlab].
We finally remark that Baker investigated also the second-order eikonal approximation for potential scattering in the non-relativistic case [@Baker], finding thereby an expression for the focusing factor of continuum Schrödinger wave functions. For the focusing in the center of a spherically symmetric potential, one finds (see eq. (23) in [@Baker]) $$f^{1/2}(0)=f^{1/2}(b=0,z=0) \simeq 1-\frac{V(0)}{2 k v}, \quad \mbox{or} \quad
f(0) \simeq 1-\frac{V(0)}{k v}$$ where $k$ is the asymptotic momentum and $v$ the velocity of the particle. Roughly speaking, the approximation is valid if the kinetic energy of the particle is larger than the depth of the disturbing potential $m \gg E_{kin}=E-m \gg V(0)$, and the wave length of the particle $\sim 2 \pi/k$ should be significantly smaller than the extension of the potential. For the classical particle momentum in the center of the potential $k(0)$ one has non-relativistically $$k(0) = \sqrt{2m(E_{kin}-V(0))}=\sqrt{2 m E_{kin}} \sqrt{1-\frac{V(0)}{E_{kin}}}
\simeq k \Biggl( 1 - \frac{V(0)}{2 E_{kin}} \Biggr),$$ such that $$f(0) \simeq 1 - \frac{V(0)}{k v} \simeq \frac{k(0)}{k},$$ i.e. it is found that the probability density is enhanced by the ratio of the central and asymptotic momenta $k(0)/k$, instead of $(k(0)/k)^2 \simeq
((E-V(0))/E)^2$ in the highly relativistic case. One may ask how the non-relativistic and the highly relativistic regime are connected. A classical relativistic analysis of the particle trajectories shows that the central focusing is given by the expression $$f(0)=\frac{k(0)}{k} \frac{E-V(0)}{E},$$ which interpolates between the non-relativistic and relativistic regime and which is given here, for the sake of brevity, as a result without proof.
[99]{}
E. Rutherford, Philosophical Magazine [**[6]{}**]{}, 669-688 (1911).
R.R. Whitney, I. Sick, J.R. Ficenec, R.D. Kephart, W.P. Trower, Phys. Rev. [**[C9]{}**]{}, 2230-2235 (1974).
O. Benhar, A. Fabrocini, S. Fantoni, I. Sick, Phys. Lett. [**[B343]{}**]{}, 47-52 (1995).
D. Rohe et al., Phys. Rev. Lett. [**[93]{}**]{}, 182501 (2004).
D. B. Day, J. S. McCarthy, Z. E. Meziani, R. C. Minehart, R. M. Sealock, S. T. Thornton, J. Jourdan, I. Sick, B. W. Filippone, R. D. McKeown, R. G. Milner, D. H. Potterveld, Z. Szalata, Phys. Rev. [**[C40]{}**]{}, 1011-1024 (1989).
J. Morgenstern, Z.E. Meziani, Phys. Lett. [**[B515]{}**]{}, 269-275 (2001).
F. Lenz, Ph.D. thesis, Freiburg, Germany (1971).
J. Knoll, Nucl. Phys. [**[A223]{}**]{}, 462-476 (1974).
C. Giusti, F.D. Pacati, Nucl. Phys. [**[A473]{}**]{}, 717-735 (1987).
F. Lenz, R. Rosenfelder, Nucl. Phys. [**[A176]{}**]{}, 513-525 (1971).
C. Giusti, F. D. Pacati, Nucl. Phys. [**[A485]{}**]{}, 461-480 (1988).
M. Traini, S. Turck-Chieze, A. Zghiche, Phys. Rev. [**[C38]{}**]{}, 2799-2812 (1988).
M. Traini, M. Covi, Nuovo Cim. [**[A108]{}**]{}, 723-736 (1995).
R. Rosenfelder, Annals Phys. [**[128]{}**]{}, 188-240 (1980).
K. S. Kim, L. E. Wright, Y. Jin, D. W. Kosik, Phys. Rev. [**[C54]{}**]{}, 2515-2524 (1996).
D.R. Yennie, F.L. Boos, D.G. Ravenhall, Phys. Rev. [**[137]{}**]{}, B882-903 (1965).
M. Levy, J. Sucher, Phys. Rev. [**[186]{}**]{}, 1656-1670 (1969).
R.L. Sugar, R. Blankenbecler, Phys. Rev. [**[183]{}**]{}, 1387-1396 (1969).
S.J. Wallace, Annals Phys. [**[78]{}**]{}, 190-257 (1973).
S.J. Wallace, J.A. McNeil, Phys. Rev. [**[D16]{}**]{}, 3565-3580 (1977).
H. Abarbanel, C. Itzykson, Phys. Rev. Lett. [**[23]{}**]{}, 53-56 (1969).
A. Aste, K. Hencken, D. Trautmann, Eur. Phys. J. [**[A21]{}**]{}, 161-167 (2004).
A. Aste, K. Hencken, J. Jourdan, I. Sick, D. Trautmann, Nucl. Phys. [**[A743]{}**]{}, 259-282 (2004).
A. Aste, J. Jourdan, Europhys. Lett. [**[67]{}**]{}, 753-759 (2004).
S. Choi, J.P. Chen, Z.-E. Meziani, [*Precision measurement of longitudinal and transverse response functions of quasi-elastic electron scattering in the momentum transfer range $0.55$ GeV $\le$ $|\vec{q}|$ $\le$ $1.0$ GeV*]{}, TJNAF Proposal E01-016 (2005).
M. Traini, Nucl. Phys. [**[A694]{}**]{}, 325-336 (2001).
A. Zghiche, J.F. Danel, M. Bernheim, M.K. Brussel, G.P. Capitani, E. De Sanctis, S. Frullani, F. Garibaldi, A. Gerard, J.M. Le Goff, A. Magnon, C. Marchand, Z.E. Meziani, J. Morgenstern, J. Picard, D. Reffay-Pikeroen, M. Traini, S. Turck-Chieze, P. Vernin, Nucl. Phys. [**[A572]{}**]{}, 513-559 (1994), Erratum ibid. [**[A584]{}**]{}, 757 (1995).
O. Benhar, A. Fabrocini, S. Fantoni, I. Sick, Nucl. Phys. [**[A579]{}**]{}, 493-517 (1994).
O. Benhar, D. Day, I. Sick, nucl-ex/0603029 (2006).
J.M. Udias, P. Sarriguren, E. Moya de Guerra, E. Garrido, J.A. Caballero, Phys. Rev. [**[C48]{}**]{}, 2731-2739 (1993).
K.S. Kim, L.E. Wright, Yanhe Jin, Phys. Rev. [**[C54]{}**]{}, 2515-2524 (1996).
A. Aste, C. von Arx, D. Trautmann, Eur. Phys. J. [**[A26]{}**]{}, 167-178 (2005).
D. Trautmann, G. Baur, F. Rösel, J. Phys. B: At. Mol. Phys. [**[16]{}**]{}, 3005-3013 (1983).
H. C. Pauli, U. Raff, Comp. Phys. Commun. [**[9]{}**]{}, 392-407 (1975).
J. Jourdan, Nucl. Phys. [**[A603]{}**]{}, 117-160 (1996).
A. Baker, Phys. Rev. [**[D6]{}**]{}, 3462-3469 (1972).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'For a given space-time and for an arbitrary time-like geodesic, we analyze the conditions for the construction of Fermi coordinates so that they are also rigid covariant. We then apply these conditions to linear plane gravitational waves.'
author:
- 'Xavier Jaén[^1]'
title: 'Rigid covariance, equivalence principle and Fermi rigid coordinates: gravitational waves'
---
*Keywords:* Rigid motion, Fermi coordinates, Equivalence principle, Linear plane gravitational wave
Introduction {#s_1}
============
In a series of recent papers, [@01-jaen_1; @02-jaen_2; @03-jaen_3; @04-jaen_4], we have presented a formulation of the general theory of relativity with a covariance group that is smaller than usual. General covariance implies having ten potentials to describe gravitation; but four of them can be eliminated by means of coordinate transformations. So, only six potentials are really necessary. In those papers we have reduced the covariance group to bring it as close as possible to the usual group of rigid motions, in a formulation with six potentials. It is not a novelty to use the concept of rigidity in relativity. In 1909 Born naturally extended this concept to relativity, but it turned out to be inconsistent in too many cases [@Born1909]. In spite of the difficulties, some authors [@Bona83; @Bel90; @Bel96; @Llosa04; @Coll07] think that rigidity could be the missing piece so that relativity acquires a state of maturity comparable to Newtonian mechanics. The differences between these authors are related to the difficulty in carrying out this natural extension. In some aspects, our rigid formulation of general relativity is remarkably close to the corresponding formulation of Newtonian gravitation. This allows us to analyze the topics mentioned in the title of this paper. We will see that, at the Newtonian level, there are gauge transformations that, although they do not entail any relevant physical changes, allows us to implement a Newtonian equivalence principle, in which rigid motions are those that lead us to the locally inertial reference system, where the effects of gravitation are locally canceled out.
When we analyze the same topic in general relativity we see how new coordinate transformations appear that we can still say are rigid in the sense that they leave the space-time metric shape invariant. We will see how we can implement the equivalence principle using these transformations together with the gauge transformations, already seen at the Newtonian level, which in general relativity induce temporal transformations.
As a consequence of this study, we can conjecture an implementation of the (relativistic) equivalence principle where rigid motions play a leading role that is very close to the Newtonian case. We will see how the theoretical framework that arises can be applied to some relevant space-times; the most interesting being that of linear plane gravitational waves. These have already been analyzed in [@04-jaen_4], where we found a set of rigid reference systems with which to express the metric. In the present study, we will see how we can choose a system from that previously determined set that defines a rigid locally inertial system. In other words, for linear plane gravitational waves, we will find rigid coordinates which in turn are Fermi coordinates attached to a geodesic.
In 1922 Fermi [@05-fermi] defined the so-called Fermi coordinates (FC), a construction of a coordinate system [^2] ${x^\mu } = \left\{ {t,\vec x} \right\}$, that near a given time like curve, $\vec x=0$, the space-time looks like a flat one, that is $$\label{eq_0}
{\left. {d{s^2}} \right|_{\vec x = 0}} = {\eta _{\mu \nu }}d{x^\mu }d{x^\nu }\;\;,{\left. {\Gamma _{\mu \nu }^\sigma }
\right|_{\vec x = 0}} = 0$$ which is equivalent to: [^3] $$d{s^2} = {\eta _{\mu \nu }}d{x^\mu }d{x^\nu } + O\left( {{x^2}} \right)$$ Manasse and Misner [@06-manasse] specialized Fermi’s coordinates for a given time-like geodesic and, in order to find a physical meaning for Fermi’s proposal, they chose a particular set of coordinates from those satisfying (\[eq\_0\]), which they called Fermi normal coordinates (FNC). The name was suggested after Riemann [@07-riemann] introduced the so-called Riemann normal coordinates, any coordinate system $\left\{ {t ,\vec x} \right\}$ for which the metric can be written in a neighborhood of a given space-time point $P$ as: $$d{s^2} = \left\{ {{\eta _{\mu \nu }} + \frac{1}{3}{{\left. {{R_{\alpha \beta \mu \nu }}} \right|}_P}{x^\alpha }{x^\beta
}} \right\}d{x^\mu }d{x^\nu } + O\left( {{{({x^\mu })}^3}} \right)$$ where ${{R_{\alpha \beta \mu \nu }}}$ is the Riemann tensor and $O\left( {{{({x^\mu })}^3}} \right)$ labels the terms of third order or higher in the space-time coordinates. Following the Fermi and Riemann construction, Manasse and Misner showed that in their FNC, the metric coefficients $g_{\mu\nu}$ fulfill:
$$\label{eq_2}
\begin{array}{*{20}{l}}
{{g_{00}} = - 1 + {{\left. {{R_{0\ell 0m}}} \right|}_{\vec x = 0}}\;{x^\ell }{x^m} + O\left( {{x^3}} \right)}\\
{{g_{0i}} = \frac{2}{3}{{\left. {{R_{0\ell im}}} \right|}_{\vec x = 0}}\;{x^\ell }{x^m} + O\left( {{x^3}} \right)}\\
{{g_{ij}} = {\delta _{ij}} + \frac{1}{3}{{\left. {{R_{i\ell jm}}} \right|}_{\vec x = 0}}\;{x^\ell }{x^m} + O\left(
{{x^3}} \right)}
\end{array}$$
Ni and Zimmermann [@08-ni] generalized the work of Manasse and Misner for any initial time-like non-geodesic world-lines and included arbitrary rotation. They studied accelerated reference systems with rotation. Li and Ni in [@09-li_1; @10-li_2], Nesterov [@11-nesterov], Marzlin [@12-marzlin_1], have all made contributions in the same direction, calculating increasing approximation orders of the metric using FNC.
In the present work, our focus is on the way FNC are constructed. Manasse and Misner based FNC on geodesics in order to be close to a reasonable interpretation. From a geometrical point of view, it is totally understandable to use geodesics. However, the physical meaning is not satisfactory, or at least it does not seem entirely clear to all authors; see, for example, the comments in [@13-rakhmanov; @14-marzlin_2] and especially in [@15-delva].
In this context, it seems reasonable to look for Fermi-like coordinates based on some alternative procedure that can be compared with the previous constructions. The idea of not basing the spatial Fermi coordinates on the geodesic distance is not new [@14-marzlin_2; @15-delva].
As we have a rigid covariant formulation of general relativity, it seems reasonable to try to use the rigid concept and the rigid covariance in order to find FC that are as close as possible to rigid ones. We will see that this approach is fruitful and we will call the resultant coordinates Fermi rigid coordinates (FRC). Of course, we demand that these coordinates satisfy (\[eq\_0\]) near the geodesic $x^i=0$ but not necessarily (\[eq\_2\]). Up to first order, the only difference between FNC and FRC will be the way they are constructed.
In [@04-jaen_4], we already derived a rigid formulation for linear plane gravitational waves. In fact, we found a broad set of possible rigid coordinates without having a physical criterion for choosing one of them. The present work can therefore be seen as the continuation of [@04-jaen_4]. This means that our purpose is not to reach a high order in the calculation. Rather, it is to prove that the concept of rigidity can be useful in the construction of reference systems and coordinates with physical significance, at least to the order in which physical significance can be given, and this means that we will explore the framework of Fermi’s coordinates.
The paper is organized as follows: Section $\S$\[s\_2\] is devoted to reviewing the rigid covariant formulation of Newtonian gravitation. In Section \[s\_3\]$\S$, we look at the gauge invariance of this formulation; while in Section $\S$\[s\_4\], we offer a complete version for the implementation of the equivalence principle for Newtonian gravitation. In Section $\S$\[s\_5\], within the context of general relativity, we study Painlevé-Gullstrand space-times for which we give the conditions for FRC and we find them up to first order in the metric. Then, in Section $\S$\[s\_6\], we deal with a general space-time in the covariant rigid formulation. We analyze, beyond the usual rigid motions, the transformations that leave the metric shape invariant and define what we mean by an isochronous geodesic congruence with respect to a given geodesic. In Section $\S$\[s\_7\], we use what we have learned in the previous section and present a protocol for finding FRC that we solve completely up to first order in the metric. Finally, in Section $\S$\[s\_8\], we apply what we have learned to find FRC for linear plane gravitational waves.
Newtonian gravitation {#s_2}
=====================
Over the years and for various reasons several authors have been interested in the structure of the Maxwell-Lorentz equations beyond their application to electromagnetism. One such example of this interest that particularly fascinates us is that expressed by R. Feynmann and reported by F.J. Dyson [@16-dyson]. We are interested in a particular family of gauge transformations that are especially suited to gravity. Let us look at them first in a simple and generic way, and then we will consider the more specific gravitational case.
Given a Lagrangian for a particle of the form: $$\label{eq_3}
L = \frac{1}{2}{\dot {\vec x}^2} + \vec A\cdot\dot {\vec x} + B$$ where $\vec A$ i $B$ are functions of $(\vec x, t)$, we can always write it in the form: $$\label{eq_4}
L = \frac{1}{2}{\left( {\dot {\vec x} - \vec V} \right)^2} + \frac{{d\tau }}{{dt}}$$ where $\vec V$ and $\tau$ are functions of $(\vec x, t)$. The proof is very simple. From the equality between (\[eq\_3\]) and (\[eq\_4\]), we get the conditions: $$\label{eq_5}
\vec V = - \vec A + {\partial _{\vec x}}\tau$$ and: $$\label{eq_6}
{\partial _t}\tau = B - \frac{1}{2}{\left( {\vec A - {\partial _{\vec x}}\tau } \right)^2}$$ What is not so well known, is that if $H$ is the Hamiltonian associated with (\[eq\_3\]), (\[eq\_6\]) is equivalent to the Hamilton-Jacobi equation with ${\tau }$ playing the role of the action, ${\partial _t}\tau + H(\vec x,\vec p = {\partial _{\vec x}}\tau ,t) = 0$, and (\[eq\_5\]) defines the velocity field, which is equivalent to $\vec V =\frac{{\partial H}}{{\partial \vec p}}(\vec x,\vec p = {\partial _{\vec x}}\tau ,t)$. For each ${\tau }$ solution of the Hamilton-Jacobi equation (\[eq\_6\]), we have a potential $\vec V$ from (\[eq\_5\]) and, with it, a Lagrangian of the form (\[eq\_4\]).
The equations of motion corresponding to (\[eq\_4\]) can be written: $$\label{eq_7}
\ddot {\vec x} = \vec g + \dot {\vec x} \times \vec \beta$$ where the relationships between the fields, ${\vec g}$ and ${\vec \beta }$, and the potential ${\vec V}$ are: $$\label{eq_8}
\begin{array}{*{20}{l}}
{\vec g = {\partial _{\vec x}}\left( {\frac{{{V^2}}}{2}} \right) + {\partial _t}\vec V}\\
{\vec \beta = - {\partial _{\vec x}} \times \vec V}
\end{array}$$ This result is applicable to gravitation. As we saw in [@01-jaen_1], in any rigid reference system $S$, the gravitational field can be represented by a potential $\vec V(\vec x,t )$ where $\left\{ {t ,\vec x} \right\}$ are the rigid time and space Euclidean coordinates. The equation of motion is of the Lorentz type (\[eq\_7\]) and (\[eq\_8\]). We will call a geodesic trajectory of the Newtonian gravitational field (\[eq\_8\]) any solution of the equation of motion (\[eq\_7\]). The gravitational field equations for ${\vec V}$ can be written in terms of ${\vec g}$ and ${\vec \beta}$ in the form: $$\label{eq_9}
\begin{array}{*{20}{l}}
{{\partial _{\vec x}}\vec g - \frac{1}{2}{\beta ^2} = - 4\pi G\rho }\\
{{\partial _{\vec x}} \times \vec \beta = 0}
\end{array}$$ Under a rigid motion transformation, ${\vec x = \vec R + \vec y}$, we will go from the reference system $S$, $\left\{ {t ,\vec x} \right\}$, to the reference system $S'$, $\left\{ {t ,\vec y} \right\}$. A translation will be represented by the vector $\vec R \equiv {R^m}(t)\;{\hat X_m}$, and a rotation by the matrix $R_i^m(t)$, so that ${\hat Y_i} = R_i^m(t){\hat X_m}$ where ${{{\hat X}_i}}$ and ${{{\hat Y}_i}}$ are the orthonormal basis of $S$ and $S'$ respectively. In the $S'$ system, the equations of motion and the field equations have the same shape as in the $S$ system, but now the gravitational potential is $\vec V'$, and the relation with ${\vec V}$ is given by: [^4] $$\label{eq_10}
\vec V = \dot {\vec R} + \vec V' + \vec \Omega \times \vec y$$ where $\dot {\vec R} \equiv {\dot {R}^m}{\hat X_m}$ and $\vec \Omega \equiv \frac{1}{2}\sum\limits_s {R_s^m\dot {R}_s^n{\delta _{mn}}^i} {\hat X_i}$. That is, the potential of the gravitational field transforms as a velocity field.
The equation of motion (\[eq\_7\]) includes pure gravitational effects in addition to Coriolis, centrifugal and Euler forces, and those due to translation. The inertial forces are due to inertial fields that are source-free solutions of the field equations (\[eq\_9\]).
The trajectories that are solutions of $\dot {\vec x} = \vec V$ are solutions of the equation of motion (\[eq\_7\]). That is, the gravitational potential ${\vec V}$ is itself a geodesic congruence.
The equation of motion (\[eq\_7\]) admits a very simple Lagrangian formulation, already suggested at the beginning of this section, with a scalar covariant lagrangian under rigid motion transformations: $$L = \frac{1}{2}{\left( {\dot {\vec x} - \vec V} \right)^2} = \frac{1}{2}{\left( {\dot {\vec y} - \vec V'} \right)^2}$$
Gauge transformations in Newtonian gravitation {#s_3}
==============================================
Because the physical fields are ${\vec g}$ and ${\vec \beta }$, we can use any potential ${\vec V^*}$, instead of ${\vec V}$, that gives, via the same relationships, the same physical fields. An elegant way to analyze a gauge transformation is that described in [@03-jaen_3], where the physical meaning of the potential ${\vec V}$ as geodesic congruence plays an important role, which means that gauge invariance becomes a requirement: if the potential ${\vec V}$ is a geodesic congruence, it seems quite natural to require that we can use any other geodesic congruence ${\vec V^*}$ as a potential. This requirement is met: given the Lagrangian $L = \frac{1}{2}\left( {\dot {\vec x} - \vec V} \right)^2$ and the corresponding Hamiltonian $H$, we define the gauge transformation $\vec V \to {\vec V^*}$ through the relations: $$\label{eq_12}
{\vec V^*} \equiv \frac{{\partial H}}{{\partial \vec p}}(\vec x,\vec p = {\partial _{\vec x}}{\tau ^*},t)$$ where ${\tau ^*}$, the action, is any solution of the Hamilton-Jacobi equation: $$\label{eq_13}
{\partial _t}{\tau ^*} + H(\vec x,\vec p = {\partial _{\vec x}}{\tau ^*},t) = 0$$ Under these conditions, it is easy to see that: $${L^*} + \frac{{d{\tau ^*}}}{{dt}} = \frac{1}{2}{\left( {\dot {\vec x} - {{\vec V}^*}} \right)^2} + \frac{{d{\tau
^*}}}{{dt}} = \frac{1}{2}{\left( {\dot {\vec x} - \vec V} \right)^2} = L$$ $L^*$ and $L$ differ by a total derivative. Using the Lagrangian $L^*$ we obtain the same equation of motion as we obtain with $L$. The equation of motion associated with ${\vec V^*}$ is the same as that associated with ${\vec V}$. The conclusion is that the fields ${\vec g}$ and ${\vec \beta }$ are gauge invariant under the transformation $\vec V \to {\vec V^*}$.
Newtonian equivalence principle {#s_4}
===============================
In [@01-jaen_1], we saw an unfinished version of the implementation of the Newtonian equivalence principle. Here we complete that implementation and emphasize some of its characteristics that will be useful at the relativistic level.
Given a gravitational potential ${\vec V}$ in a rigid reference system $S$, $\left\{ {t ,\vec x} \right\}$, for each geodesic $G$ , $\vec x(t)$, of the geodesic congruence ${\vec V}$ we define a rigid reference system $S'$,$\left\{ {t ,\vec y} \right\}$, as that which is rigidly related to $S$ according to the rigid transformation: $${\vec x = \vec x(t) + \vec y\;}$$ where ${\vec x(t) = {x^m}(t){\mkern 1mu} {{\hat X}_m}}$ is the geodesic $G$, ${\vec y = {y^m}\,{{\hat Y}_m}}$, and ${{{\hat Y}_i} = R_i^m(t){\mkern 1mu} \,{{\hat X}_m}}$ with $R_i^m(t)$ being any matrix rotation solution of the equation: $$\label{eq_16}
\vec \Omega (t) \equiv \sum\limits_s {R_s^m\dot {R}_s^n{\delta _{mn}}^i} {{\hat X}_i} = {\left[ {{\partial _{\vec x}}
\times \vec V} \right]_{\vec x = \vec x(t)}}$$ That is to say, the origin of the reference system $S'$ moves along the geodesic trajectory $G$, $\vec x(t)$, belonging to the congruence ${\vec V}$. The rotation of $S'$, with respect to $S$, $\vec \Omega (t)$, is equal to the vorticity of the congruence ${\vec V}$ evaluated on the geodesic $G$.
With these conditions, the transformed potential, $\vec V'$, is, according to (\[eq\_10\]) expressed in $\vec y$ coordinates and developed up to order $y$: $$\label{eq_17}
\vec V'\left( {\vec y,t} \right) = {\left[ {\left( {\vec y\cdot{\partial _{\vec x}}} \right)\vec V + \frac{1}{2}\vec y
\times \left( {{\partial _{\vec x}} \times \vec V} \right)} \right]_{\vec x = \vec x(t)}} + O\left( {{y^2}} \right)$$ In this expression, we make extensive use of Euclidean vector notation. To expand it, it is necessary to keep in mind the relationship between the basis of $S$ and $S'$, ${\hat Y_i} = R_i^m{\hat X_m}$, when resolving some products. For example, in ${\vec y\cdot{\partial _{\vec x}}}$, the two factors must be expressed in their defining basis and the change must be made. The notation is consistent because the rotation $R_i^j$ depends only on $t$.
A direct calculation, using (\[eq\_17\]), shows that the gravitational fields in the reference system $S'$, $\vec g'$ and $\vec\beta '$, that we obtain from the potential $\vec V'$, are null at the origin of $S'$: $$\label{eq_18}
\begin{array}{*{20}{l}}
{\vec g'\left( {\vec y = 0,t} \right) = {{\left[ {{\partial _{\vec y}}\left( {\frac{{{{V'}^2}}}{2}} \right) + {\partial
_t}\vec V'} \right]}_{\vec y = 0}} = 0}\\
{\;\vec \beta '\left( {\vec y = 0,t} \right) = - {{\left[ {{\partial _{\vec y}} \times \vec V'} \right]}_{\vec y = 0}}
= 0}
\end{array}$$ So, $S'$ is actually a rigid locally inertial reference system attached to $G$: $S_{IG}$.
This result is in accordance with the equivalence principle. If we take into account gauge transformations $\vec V \to{\vec V^*}$ we can extend this result to any reference systems $S'$ attached on an arbitrary geodesic, not only a geodesic solutions of $\dot {\vec x} = \vec V$. In this context, the implementation of the Newtonian equivalence principle can be stated thus:
> ***Newtonian equivalence principle**: if, in a rigid reference system $S$, we have a gravitational field, for any geodesic $G$ we can always associate a geodesic congruence $\vec V$ and use it to build a rigid locally inertial reference system $S_{IG}$ such that its origin moves along the geodesic $G$ and rotates according to the vorticity of $\vec V$ evaluated at $G$. At the origin of $S_{IG}$, the gravitational field is null, $\vec g'\left( {0,t } \right) = 0\;$ and $\vec \beta '\left( {0,t } \right) = 0$, and consequently, test particles move freely.*
It should be noted that it is the gauge invariance of the theory that allows it to be in accordance with the equivalence principle. Moreover, it is remarkable that, given a geodesic $G$, there may be more than one associated congruence $\vec V$. The vorticity of these congruences, due to the gauge invariance, will be equal and therefore all of them will define the same rotation $\vec \Omega $ (see (\[eq\_16\]) ), that is, the same rigid locally inertial reference system $S_{IG}$.
We can give a Lagrangian version of this result. What is going to be interesting now is not the result but how we arrive at it. We start in the rigid reference system $S$, with the Lagrangian $L = \frac{1}{2}{\left( {\dot{\vec x} - \vec V} \right)^2}$ where, using the gauge freedom, we take $\vec V$ in such a way that it include the given geodesic $G$ to build the system $S_{IG}$. The Lagrangian in $S_{IG}$, with coordinates $\vec y$, can be written, taking into account expression (\[eq\_17\]) for $\vec V'$ and developing it around $\vec y = 0$, in the form: $$\label{eq_19}
L = \frac{1}{2}{\left( {\dot {\vec y} - \vec V'} \right)^2} = \frac{1}{2}{\dot {\vec y}^2} - \frac{{dF}}{{dt}} + O\left(
{{y^2}} \right)$$ where $F = \frac{1}{2}\vec y\cdot\vec V' $. The equation of motion related to (\[eq\_19\]) at the origin of the system $S_{IG}$, $\vec y = 0$, is $\ddot {\vec y} = 0$.
It is worth commenting that the gauge transformations (\[eq\_12\]) and (\[eq\_13\]) have the effect of adding a total derivative of a function to the original Lagrangian, which that turns out to be the action. This suggests that we can proceed in an alternative way to reproduce (\[eq\_19\]). Let us start with the Lagrangian $L = \frac{1}{2}{\left( {\dot {\vec y} - \vec V'} \right)^2}$ and look for a gauge transformation $\vec V' \to {\vec V'^*}$, described by (\[eq\_12\]) and (\[eq\_13\]), but now using $\vec y$ coordinates, which conforms to ${\vec V'^*} = O\left( {{y^2}} \right)$. The result is, as expected, ${\tau ^*} = - \frac{1}{2}\vec y\cdot\vec V'$, and we reproduce expression (\[eq\_19\]).
We should point out here that at the Newtonian level, the action ${\tau ^*}$ does not have any relevant meaning.
Painlevé-Gullstrand space-times {#s_5}
================================
Let us now turn to general relativity. We will begin by studying a space-time that in the rigid reference system $S$ has a Painlevé-Gullstrand metric form [@01-jaen_1], based on the form of the metric that Painlevé [@17-painleve] and, independently, Gullstrand [@18-gullstrand] proposed for Schwarzschild space-time. On $S$, we will take the space-time coordinates $\left\{ {\lambda ,\vec x} \right\}$ and, from now on in this paper, we will take $c=1$. The metric is thus: $$\label{eq_20}
d{s^2} = - d{\lambda ^2} + {\left( {d\vec x - \vec Vd\lambda } \right)^2}$$ where $\vec V$ is a function of $(\vec x, \lambda)$. If we adapt the factor so that $Ld\lambda $ becomes the proper time, the associated Lagrangian, $L$, is: $$\label{eq_21}
L = \sqrt {1 - {{\left( {\dot {\vec x} - \vec V} \right)}^2}}$$ These space-times are rigid covariants, defining rigid transformations exactly in the same way as in the Newtonian case, only now using the time $\lambda$. The potential ${\vec V}$ again transforms as a velocity (the relation (\[eq\_10\]) remains valid).
As we have seen in [@03-jaen_3], the metric (\[eq\_20\]) belongs to a broad class of rigid covariant metrics of the form: $$\label{eq_22}
d{s^2} = - d{\tau ^2} + \left( {{\tau _{,{x^i}}}{\tau _{,{x^j}}} + {\delta _{ij}}} \right)(d{x^i} - {V^i}d\lambda
)(d{x^j} - {V^j}d\lambda )$$ where $\tau = \tau \left( {\vec x,\lambda } \right)$. For the metric (\[eq\_20\]), $\tau \left( {\vec x,\lambda } \right) = \lambda $.
From (\[eq\_22\]), it is easy to see that $U = {\partial _\lambda } + {V^i}\left( {\vec x,\lambda } \right){\partial_{{x^i}}}$ is a time-like geodesic congruence with proper time $\tau \left( {\vec x,\lambda } \right)$.
The metric (\[eq\_20\]), as a member of the metric class (\[eq\_22\]), is invariant under gauge transformations $\left\{ \tau = \lambda ,\vec V \right\}\to\left\{ {\tau ^*},{{\vec V}^*} \right\}$ as described in [@03-jaen_3]. The difference with the Newtonian case is that now the action ${\tau ^*}$, as a solution of the Hamilton-Jacobi equation, is also a potential. This potential has its own meaning as the proper time of the geodesic congruence $U^*$.
If $H$ is the Hamiltonian associated with the Lagrangian (\[eq\_21\]), the gauge transformation $\left\{ {\tau = \lambda ,\vec V} \right\} \to \left\{ {{\tau ^*},{{\vec V}^*}} \right\}$ is defined in a similar way to the Newtonian case: $${\vec V^*} \equiv \frac{{\partial H}}{{\partial \vec p}}(\vec x,\vec p = {\partial _{\vec x}}{\tau ^*},\lambda )$$ and ${\tau ^*}$ is any solution of the Hamilton-Jacobi equation: $${\partial _\lambda }{\tau ^*} + H(\vec x,\vec p = {\partial _{\vec x}}{\tau ^*},\lambda ) = 0$$ For each geodesic $G$, ${\vec x}(\lambda)$, of the congruence $\vec V$ of (\[eq\_20\]), we can define a reference system $S'$ by performing the same rigid transformation, $\vec x\to\vec y$, as we did in the Newtonian equivalence principle in Section $
\S$\[s\_4\]. The result is: $$\label{eq_25}
d{s^2} = - d{\lambda ^2} + {\left( {d\vec y - \vec V'\;d\lambda } \right)^2}$$ where $\vec V'\left( {\vec y,\lambda } \right)$, as in the Newtonian case, can be expressed in $\vec y$ coordinates, developed up to order $y$ and giving the same result as in (\[eq\_17\]). Now we look for a gauge transformation of the form $\left\{ {\tau = \lambda ,\vec V'} \right\} \to \left\{ \tau^*,\vec {V'}^* = O\left( y^2 \right) \right\}$. The Hamiltonian associated with (\[eq\_25\]) is: $$H\left( {\vec y,\vec p,\lambda } \right) = \vec V'\cdot\vec p - \sqrt {1 + {p^2}}$$ and the Hamilton-Jacobi equation: $$\label{eq_27}
{\partial _\lambda }{\tau ^*} + \vec V'\cdot{\partial _{\vec y}}{\tau ^*} - \sqrt {1 + {{\left( {{\partial _{\vec
y}}{\tau ^*}} \right)}^2}} = 0$$ We look for ${\tau ^*}$ solutions fulfilling:
$$\label{eq_28}
{\vec V'^*} = O\left( {{y^2}} \right) = \frac{{\partial H}}{{\partial \vec p}}(\vec y,\vec p = {\partial _{\vec y}}{\tau
^*},\lambda )$$
If we write (\[eq\_28\]) except for the $O\left( {{y^2}} \right)$ terms, we have: $$\label{eq_29}
\frac{{{\partial _{\vec y}}{\tau ^*}}}{{\sqrt {1 + {{\left( {{\partial _{\vec y}}{\tau ^*}} \right)}^2}} }} = \vec V'$$ As $\vec V'$ is $O\left( y \right)$, (\[eq\_29\]) means that ${\partial _{\vec y}}{\tau ^*} = O(y)$. So we can write (\[eq\_27\]) and (\[eq\_29\]) as: $$\label{eq_30}
\begin{array}{*{20}{l}}
{{\partial _\lambda }{\tau ^*} = 1 + O({y^2})}\\
{{\partial _{\vec y}}{\tau ^*} = \vec V' + O({y^2})}
\end{array}$$ The integrability conditions are guaranteed by (\[eq\_18\]), that is, ${\partial _{\vec y}} \times \vec V' = O({y^2})$, and by the fact that we are neglecting terms of order $O({y^2})$. We can check by direct calculation, in which expression (\[eq\_17\]) for $\vec V'$ is used, that a solution satisfying (\[eq\_30\]) is: $$\label{eq_31}
{\tau ^*} = \lambda + \frac{1}{2}\vec y\cdot\vec V' + O({y^3})$$ which is the relativistic version of the function $F$ in the Newtonian case: expression (\[eq\_19\]). We now define the time coordinate $t = {\tau ^*}(\vec y,\lambda )$. The explicit expression of $t$ can be found using (\[eq\_31\]) and (\[eq\_17\]): $$\label{eq_32}
t = \lambda + \frac{1}{2}\vec y\cdot{\left[ {\left( {\vec y\cdot{\partial _{\vec x}}} \right)\vec V + \frac{1}{2}\vec y
\times \left( {{\partial _{\vec x}} \times \vec V} \right)} \right]_{\vec x = \vec R}} + O\left( {{y^3}} \right)$$ The metric in $\left\{ {t,\vec y} \right\}$ coordinates takes the form: $$\label{eq_33}
d{s^2} = - d{t^2} + {\delta _{ij}}d{y^i}d{y^j} + O({y^2})$$ that is, the $\left\{ {t,\vec y} \right\}$ coordinates are FC and the system $S'$ is a rigid locally inertial reference system attached to the geodesic $G$, $S_{IG}$. These coordinates are those we announce in the Introduction above and name: Fermi rigid coordinates (FRC).
We should note that of the coordinates $\left\{ {t,\vec y} \right\}$ only $\vec y$ can be labeled as *rigid* at any order, in the sense that the $\left\{ {\lambda ,\vec y} \right\}$ coordinates are rigid at any order and from these we only perform a change of time. That is, we could write the metric in the system $S_{IG}$ in the form: $$\label{eq_34}
d{s^2} = - d{\tau ^{*2}} + {\delta _{ij}}d{y^i}d{y^j} + O({y^2})$$ understanding that we use $\left\{ {\lambda ,\vec y} \right\}$ coordinates and with ${\tau ^*}(\vec y,\lambda )$ being the potential.
Regarding the terms grouped under the symbol $O({y^2})$ (terms of order ${y^2}$ or higher), we observe that we could obtain a more accurate expression for the metric by solving (\[eq\_27\]) for ${\tau ^*}(\vec z,\lambda )$, expanding (\[eq\_27\]) in a power series around $\vec y = 0$ and using the condition (\[eq\_31\]). As mentioned in the Introduction, these second-order terms would allow us to distinguish between FRC and FNC. We will not develop this line of work here because our current objective is to prove the feasibility of FRC and to do this it suffices to analyze the metric up to the first order.
We can say that the $S_{IG}$ system supports rigid coordinates $\left\{ {\lambda ,\vec y} \right\}$ and in this sense is defined for all orders in $y$. $\lambda$ is the rigid time of the $S_{IG}$ system. In $S_{IG}$, we can define a new time $t$ and use the metric form (\[eq\_33\]) or just use the metric form (\[eq\_34\]).
If we take into account what we have seen so far, we can offer an implementation of the equivalence principle in the following terms:
> *If we have a space-time that, in a rigid reference system $S$, can be described by (\[eq\_20\]), with a potential $\vec V$, then for each geodesic $G$ of the congruence $\vec V$ we can always define a rigid locally inertial reference system $S_{IG}$ such that its origin moves along the geodesic $G$ and rotates according to the vorticity of $\vec V$ evaluated at $G$. In the neighborhood of the origin of $S_{IG}$, it is possible to define a time with respect to which test particles move freely.*
The extension of this result to any geodesic, unlike the Newtonian case, is not at all evident. In Section $\S$\[s\_6\] we deal with the case of a general space-time. First, we will give a simple example to illustrate how things work.
Fermi rigid coordinates for radial escape geodesics in Schwarzschild space-time {#s_5_1}
-------------------------------------------------------------------------------
Schwarzschild space-time in a rigid reference system $S$ can be written in the Painlevé-Gullstrand form (\[eq\_20\]), with: $$\label{eq_35}
\vec V = \sqrt {\frac{K}{{\left| {\vec x} \right|}}} \;\frac{{\vec x}}{{\left| {\vec x} \right|}}$$ Under these coordinates, all radial geodesics $\vec x(\lambda )$ that are solutions of $\dot {\vec x} = \vec V$, that is to say the family of geodesics: $$\label{eq_36}
\vec x(\lambda ) = {\left[ {\frac{3}{2}\sqrt K \lambda + {{\left| {{{\vec x}_0}} \right|}^{\frac{3}{2}}}} \right]^{\frac{2}{3}}}\frac{{{{\vec x}_0}}}{{\left| {{{\vec x}_0}} \right|}}$$ has the same proper time $\lambda$. They are escape trajectories, i.e., outgoing trajectories with null speed at infinity. We note that if we want to study falling trajectories, we would need to consider Schwarzschild space-time with different rigid space-time coordinates. The *good* expression for the metric would be the one we obtained with the potential $\vec V = - \sqrt {\frac{K}{{\left| {\vec x} \right|}}} \;\frac{{\vec x}}{{\left| {\vec x} \right|}}$. In Section $\S$\[s\_7\_5\], we discuss this kind of problem in more detail.
We can find FRC for each of the geodesics $G$ in (\[eq\_36\]). First, we make a rigid transformation $\vec x \to \vec y$ to give us the rigid locally inertial reference system $S_{IG}$, based on the congruence (\[eq\_35\]). We do not need to make a rotation because the congruence (\[eq\_35\]) is irrotational. We just need to perform the translation $\vec x = \vec y + \vec x(\lambda )$. We will then have: $$\begin{array}{l}
d{s^2} = - d{\lambda ^2} + {\left( {d\vec y - \vec V'd\lambda } \right)^2}\\
\vec V' = \sqrt {\frac{K}{{\left| {\vec y + \vec x(\lambda )} \right|}}} \;\frac{{\vec y + \vec x(\lambda )}}{{\left|
{\vec y + \vec x(\lambda )} \right|}} - \dot {\vec x}(\lambda )
\end{array}$$ Now, we make a gauge transformation to eliminate the potential ${\vec V'}$ up to order $y^2$. According to (\[eq\_32\]), this will define the time $t$ in $S_{IG}$: $$t = \lambda - \frac{{{{\vec y}^2}}}{4}\sqrt {\frac{K}{{{{\left| {\dot {\vec x}(\lambda )} \right|}^3}}}} + O({y^2})$$ We can verify that the $\left\{ {t,\vec y} \right\}$ coordinates are FC for the chosen geodesic (\[eq\_36\]).
General space-time {#s_6}
==================
We will consider a general space-time which admits rigid covariant coordinates in the sense explained in [@04-jaen_4]. The metric of such a space-time in a rigid reference system $S$ can be expressed as [@04-jaen_4]: $$\label{eq_39}
d{s^2} = - {\Phi ^2}d{\lambda ^2} + 2{K_i}\;d{x^i}\;d\lambda + {\gamma _{ij}}d{x^i}d{x^j}$$ with: $$\label{eq_40}
{\gamma _{ij}} = \frac{1}{{{{\cal H}^2}}}{\delta _{ij}} - {\sigma _{,{x^i}}}\;{\sigma _{,{x^j}}}$$ The six gravitational potentials in this metric expression are $\left\{ {{\Phi},{K_i},{{\cal H}},\sigma } \right\}$.
If $U = {\partial _\lambda } + {V^i}\left( {\vec x,\lambda } \right){\partial _{{x^i}}}$ is a time-like geodesic congruence of the space-time (\[eq\_39\]), with proper time $\tau \left( {\vec x,\lambda } \right)$, we can express ${\Phi}$ and $K_i$ in terms of the potentials $\tau$ and $\vec V$ in the form: $$\begin{array}{*{20}{l}}
{{\Phi ^2} = - {\gamma _{ij}}{V^i}{V^j} + \tau _{,\lambda }^2 - {{\left( {{\tau _{,{x^i}}}{V^i}} \right)}^2}}\\
{{K_i}\; = - {\gamma _{ij}}{V^j} - \left( {{\tau _{,\lambda }} + {\tau _{,{x^j}}}{V^j}} \right){\tau _{,{x^i}}}}
\end{array}$$ The manifestly rigid covariant potentials of general relativity are $\left\{ {\tau ,\vec V,\;{\cal H},\sigma \;} \right\}$ and we can write (\[eq\_39\]) in the manifestly rigid covariant form as: $$\label{eq_42}
d{s^2} = - d{\tau ^2} + \left( {{\tau _{,{x^i}}}{\tau _{,{x^j}}} + {\gamma _{ij}}} \right)(d{x^i} - {V^i}d\lambda
)(d{x^j} - {V^j}d\lambda )$$ where $d\tau = d\tau (\vec x,\lambda )$, even if we do not specify this. (\[eq\_39\]) (or (\[eq\_42\]) ) is shape invariant under the usual rigid transformations, such as those performed in the Newtonian case. The potential $\vec V$ transforms as a velocity ((\[eq\_10\]) remains valid) and the potentials $\tau ,{\cal H}$ and $\sigma$ as scalar fields.
The Lagrangian associated with (\[eq\_39\]) is, if we adapt the factor so that $Ld\lambda $ becomes the proper time: $$L = \sqrt {{\Phi ^2} - 2{K_i}\;{{\dot {x}}^i}\; - {\gamma _{ij}}{{\dot {x}}^i}{{\dot {x}}^j}}$$ and the corresponding Hamiltonian is: $$H(\vec x,\vec p,\lambda ) = - {K_i}\;{\gamma ^{ij}}{p_j} - \sqrt {\left[ {1 + {\gamma ^{ij}}{p_i}{p_j}} \right]\left[
{{\gamma ^{ij}}{K_i}\;{K_j} + {\Phi ^2}} \right]}$$ where ${\gamma ^{ij}}$ is the inverse matrix of ${\gamma _{ij}}$, i.e., ${\gamma _{im}}{\gamma ^{mj}} = \delta _i^{\;j}$.
The metric (\[eq\_39\]) can be constructed from any geodesic congruence $U$. That is, (\[eq\_39\]) is invariant under gauge transformations $\left\{ {\vec V,\tau } \right\} \to \left\{ {{{\vec V}^*},{\tau ^*}} \right\}$ with $\tau^*$ being any solution of the Hamilton-Jacobi equation: $$\label{eq_45}
{\partial _\lambda }{\tau ^*} + H(\vec x,\vec p = {\partial _{\vec x}}{\tau ^*},\lambda ) = 0$$ and ${\vec V^*}$ defined by: $$\label{eq_46}
{\vec V^*} = \frac{{\partial H}}{{\partial \vec p}}(\vec x,\vec p = {\partial _{\vec x}}{\tau ^*},\lambda )$$
General rigid transformations {#s_6_1}
-----------------------------
Given a metric of the form (\[eq\_39\]), with (\[eq\_40\]), we will say that a transformation of the space-time coordinates, $\left\{ {\lambda ,\vec x} \right\} \to \left\{ {\zeta ,\vec y} \right\}$, is a general rigid transformation if the transformed metric has the same rigid form as the original, but it may have a different rigid time $\zeta$. In Section $\S$4 of the reference [@04-jaen_4], we saw how to find rigid coordinates from a metric expressed in arbitrary space-time coordinates. If the starting metric in Section $\S$4 of reference [@04-jaen_4] is already in rigid coordinates, the procedure described there will define a general rigid transformation.
Given all the general rigid transformations, we distinguish the set that transform time, which we will call $\lambda$-rigid transformations, from the rest that do not transform time, $\zeta = \lambda $.
The $\lambda$-rigid transformations are characterized by the procedure mentioned above, in Section 4 of reference [@04-jaen_4], when the starting metric, expression (\[eq\_12\]) of [@04-jaen_4], is already in rigid coordinates. That is, the coordinates $\{ {T,\vec X} \}$ of the metric (\[eq\_12\]) of [@04-jaen_4] are rigid coordinates so ${C_{ij}} = {\delta _{ij}} - {\sigma _{,{X^i}}}\;{\sigma _{,{X^j}}}$ and the time $T$ is the starting rigid time. Non-trivial solutions for $T( {\lambda ,\vec X} )$ will define $\lambda$-rigid transformations. Once a solution has been found for $T( {\lambda ,\vec X} )$, it will be necessary to complete the work by searching for rigid Euclidean spatial coordinates $\vec x$ for the new time $\lambda$, as described in [@04-jaen_4]. For any $\lambda$-rigid transformation $T( {\lambda ,\vec X} )$, there always exists a broad set of rigid Euclidean spatial coordinates $\vec x$.
Within the second set, that of the transformations that do not change time, we find the subset of the usual rigid motions: rotations and translations. They are characterized by the fact that they leave invariant the element: $$\bar d{s^2} = {\delta _{ij}}\bar d{x^i}\bar d{x^j}$$ where $\bar d$ is the restriction of $d$ at the hypersurface $d\lambda=0$. To distinguish these transformations from those explained in the following section, we will call them $\delta$-rigid transformations.
$\gamma$-rigid transformations {#s_6_2}
-------------------------------
Another subset of the general rigid transformations, also belonging to the set that do not transform time, are those that leave invariant the element: $$\bar d{s^2} = \left( {\frac{1}{{{{\cal H}^2}}}{\delta _{ij}} - {\sigma _{,{x^i}}}{\sigma _{,{x^j}}}} \right)\bar
d{x^i}\bar d{x^j}$$ and are not $\delta$-rigid transformations.
We will restrict our study to the case in which ${\cal H} = 1$. The transformations we are looking for must leave invariant the element: $$\bar d{\vec x^2} - \bar d{\sigma ^2} = {\eta _{\alpha \beta }}\bar d{z^\alpha }\bar d{z^\beta }$$ where ${z^\alpha } = \left\{ {\sigma ,\vec x} \right\}$ and ${\eta _{\alpha \beta }} = {\rm{diag}}( - 1,1,1,1)$. This defines a kind of Poincaré space transformation that we can write as: $${z'^\alpha } = L_\beta ^\alpha (\lambda )\left( {{z^\beta } - {Z^\beta }(\lambda )} \right)$$ where $z{'^\alpha } = \left\{ {\sigma ',\vec y} \right\}$. The transformation is characterized by the functions $\{{{Z^\beta }(\lambda ),L_\beta ^\alpha (\lambda )} \}$ where $L_\beta ^\alpha (\lambda )$ is a Lorentz matrix: $L_\alpha ^\mu L_\beta ^\nu {\eta _{\mu \nu }} = {\eta _{\alpha \beta }}$. The transformation has the explicit appearance: $$\label{eq_51}
\bar d{\vec x^2} - \bar d{\sigma ^2} = \bar d{\vec y^2} - \bar d\sigma {'^2}$$ with: $$\label{eq_52}
\begin{array}{*{20}{l}}
{{y^i} = L_j^i(\lambda )\left( {{x^j} - {Z^j}(\lambda )} \right) + L_0^i(\lambda )\left( {\sigma (\vec x,\lambda ) -
{Z^0}(\lambda )} \right)}\\
{\sigma '(\vec y,\lambda ) = L_j^0(\lambda )\left( {{x^j} - {Z^j}(\lambda )} \right) + L_0^0(\lambda )\left( {\sigma
(\vec x,\lambda ) - {Z^0}(\lambda )} \right)}
\end{array}$$ The first expression in (\[eq\_52\]) is the transformation of the coordinates we are looking for. The second expression is the transformation law of the potential $\sigma$ so that (\[eq\_51\]) is satisfied. If, from (\[eq\_52\]), we rule out $\delta$-rigid transformations (rotations and translations), then what we have left are the transformations that we can call $\gamma$-rigid transformations: $$\label{eq_53}
\begin{array}{l}
\vec y = \vec x - \Upsilon \vec \beta \sigma + \frac{{\Upsilon - 1}}{{{\beta ^2}}}\left( {\vec \beta \cdot\vec x}
\right)\vec \beta \\
\sigma ' = \Upsilon \left( {\sigma - \vec \beta \cdot\vec x} \right)
\end{array}$$ where $\vec \beta $ is a function of $\lambda$ and $\Upsilon \equiv \frac{1}{{\sqrt {1 - {\beta ^2}}}}$. Note that $\sigma (\vec x,\lambda )$, which is involved in the $\gamma$-rigid transformation, is set by the metric.
Under an arbitrary transformation of space coordinates, $\vec x\; = \vec x(\vec y,\lambda )$, the metric (\[eq\_42\]) becomes: $$d{s^2} = - d{\tau ^2} + \left( {{\tau _{,{y^i}}}{\tau _{,{y^j}}} + {{\gamma '}_{ij}}} \right)(d{y^i} - {V'^i}d\lambda
)(d{y^j} - {V'^j}d\lambda )$$ where the potential ${{V'}^i}$ is: $$\label{eq_55}
{V'^i} = \frac{{\partial {y^i}}}{{\partial {x^m}}}\left( {{V^m} - \frac{{\partial {x^m}}}{{\partial \lambda }}} \right)$$ The potential $\tau$ behaves like an scalar. If in addition the transformation is of the type (\[eq\_53\]), then we will have: $${\gamma '_{ij}} = {\delta _{ij}} - {\sigma '_{,{y^i}}}{\sigma '_{,{y^j}}}$$ where $\sigma ' = \frac{1}{{\sqrt {1 - {\beta ^2}} }}( {\sigma - \vec \beta \cdot\vec x} )$.
Minkowski space-time {#s_6_3}
--------------------
Now, we want to study the ordinary Poincaré transformations in the context of rigid general relativity. The decomposition of those transformations derived from this study is not a novelty and was already analyzed by Möller [@19-moller] and used by Bel [@20-bel].
A standard observer $S$ in Minkowski space-time will use coordinates $\left\{ {\lambda ,\vec x} \right\}$: $$\label{eq_57}
d{s^2} = - d{\lambda ^2} + d{\vec x^2}$$ First, we perform a $\delta$-rigid transformation $\vec x = \vec y + \vec v\lambda $ with $\vec v = $constant. Using the $\left\{ {\lambda ,\vec y} \right\}$ coordinates, the metric becomes: $$\label{eq_58}
d{s^2} = - d{\lambda ^2} + {\left( {d\vec y + \vec v\;d\lambda } \right)^2}$$ that is to say, in $\left\{ {\lambda ,\vec y} \right\}$ coordinates, we have a non-null potential $\vec V = - \vec v$.
Second, from (\[eq\_58\]), we perform a gauge transformation of the type $\{ {\tau = \lambda ,\vec V = - \vec v}\} \to \{ {{\tau ^*},{{\vec V}^*} = 0} \}$. The Lagrangian associated with (\[eq\_57\]) is $L = \sqrt {1 - {{(\dot {\vec y} + \vec v)}^2}} $ and the Hamiltonian $H = \vec p\cdot\vec v - \sqrt {1 + {p^2}} $. The equations that must be satisfied are: $$\begin{array}{*{20}{l}}
{{\partial _\lambda }{\tau ^*} + \vec v\cdot{\partial _{\vec y}}{\tau ^*} - \sqrt {1 + {{\left( {{\partial _{\vec
y}}{\tau ^*}} \right)}^2}} = 0}\\
{0 = \vec v - \frac{{{\partial _{\vec y}}{\tau ^*}}}{{\sqrt {1 + {{\left( {{\partial _{\vec y}}{\tau ^*}} \right)}^2}}
}}}
\end{array}$$ and the solution is: $$\label{eq_60}
{\tau ^*} = \sqrt {1 - {v^2}} \;\lambda + \frac{{\vec v\cdot\vec y}}{{\sqrt {1 - {v^2}} \;}}$$ In this gauge, the metric can be written using the potential ${{\tau ^*}}$: $$\label{eq_61}
d{s^2} = - d{\tau ^*}^2 + \left( {\tau _{,i}^*\tau _{,j}^* + {\delta _{ij}}} \right)d{y^i}d{y^j}$$ This expression suggests performing a $\gamma$-rigid transformation, as described in Section $\S$\[s\_6\_2\], in such a way that we pass from $\sigma=0$ to $\sigma ' = - \frac{{\vec v\cdot\vec y}}{{\sqrt {1 - {v^2}} \;}}$. Taking into account (\[eq\_53\]), this is an $\vec y \to \vec z$ transformation with $\vec \beta = \vec v$: $$\label{eq_62}
\vec z = \vec y + \frac{{\Upsilon - 1}}{{{v^2}}}\left( {\vec v\cdot\vec y} \right)\vec v$$ where $\Upsilon \equiv \frac{1}{{\sqrt {1 - {v^2}} }}$. In this way, we cancel the term ${\tau_{,i}^*\tau _{,j}^*}$ from (\[eq\_61\]) with the new terms $ - {{\sigma '}_{,i}}{{\sigma '}_{,j}}$. In addition, being a transformation that does not depend on time $\lambda$, and since the original potential $\vec V$ is null, according to (\[eq\_55\]), the transformed potential $\vec V'$ will also be null. That is to say, in $\left\{ {\lambda ,\vec z} \right\}$ coordinates, the metric is written: $$d{s^2} = - d{\tau ^*}^2 + d{{\vec z}^2}$$ where ${\tau ^*}$ is now a function of $\left\{ {\lambda ,\vec z} \right\}$, ${\tau ^*}(\vec z,\lambda )$, according to (\[eq\_60\]) and (\[eq\_62\]).
Third and finally, we perform the transformation $t = {\tau ^*}(\vec z,\lambda )$, thus obtaining the metric in Minkowski coordinates $\left\{ {t,\vec z} \right\}$. The composite transformation $\left\{ {\lambda ,\vec x} \right\} \to \left\{ {t,\vec z} \right\} $ is, of course, an ordinary Lorentz transformation. Thus, we see that we can understand a Lorentz transformation as a composition of general rigid transformations: a $\delta$-, a $\gamma$- and finally a $\lambda$-rigid transformation induced by a gauge transformation.
$G$-isochronous geodesic congruence {#s_6_4}
-----------------------------------
Given a metric expressed in a rigid reference system $S$ with coordinates $\left\{ {\lambda ,\vec x} \right\}$, which we can write as (\[eq\_39\]) with the potentials ${\Phi}$, $K_i$, $\cal {H}$ and $\sigma$ known, as functions of $\left\{ {\lambda ,\vec x} \right\}$, any time-like geodesic congruence $\{ {\tau\left( {\vec x,\lambda } \right),\vec V\left( {\vec x,\lambda } \right)} \}$ satisfies equations (\[eq\_45\]) and (\[eq\_46\]): $$\label{eq_64}
\left\{ {\begin{array}{*{20}{l}}
{{\partial _\lambda }\tau \left( {\vec x,\lambda } \right) + H\left( {\vec x,\vec p = {\partial _{\vec x}}\tau (\vec
x,\lambda ),\lambda } \right) = 0}\\
{\vec V\left( {\vec x,\lambda } \right) = \frac{{\partial H}}{{\partial \vec p}}\left( {\vec x,\vec p = {\partial _{\vec
x}}\tau (\vec x,\lambda ),\lambda } \right)}
\end{array}} \right.$$ Given a time-like geodesic $G$, $\left\{ {\vec x(\lambda ),\tau (\lambda )} \right\}$, we will say that the geodesic congruence $\{ {\tau \left( {\vec x,\lambda } \right),\vec V\left( {\vec x,\lambda } \right)} \}$ is $G$-isochronous if $G$ belongs to the congruence, that is to say: $$\begin{array}{*{20}{l}}
{\dot {\vec x}(\lambda ) = \vec V\left( {\vec x(\lambda ),\lambda } \right)\;}\\
{\tau (\lambda ) = \tau \left( {\vec x(\lambda ),\lambda } \right)}
\end{array}$$ and in addition it fulfills: $${\left. {\bar d\tau } \right|_G} = 0$$ which explicitly is ${\partial _{\vec x}}\tau(\vec x(\lambda ),\lambda ) = 0$. We will also add the simple condition $\tau (\vec x(\lambda ),\lambda ) = \lambda $ so that we can say that $\lambda$ is the proper time of the geodesic $G$.
It should be noted that if $G$ is $\left\{ {\tau (\lambda ),\vec x(\lambda )} \right\}$, and the congruence $\{ {\tau \left( {\vec x,\lambda } \right),\vec V\left( {\vec x,\lambda } \right)} \}$ is $G$-isochronous, then the following will be satisfied: $$\label{eq_67}
\dot {\vec x}(\lambda ) = \frac{{\partial H}}{{\partial \vec p}}(\vec x(\lambda ),\vec p = 0,\lambda )$$ It is worth noting that, given the metric, this equation is determined by the choice of the time coordinate $\lambda$. Thus, once time $\lambda$ is set, only the geodesics $G$ that meet (\[eq\_67\]) can belong to a $G$-isochronous congruence. In general, given a geodesic $G$, in order to find its $G$-isochronous congruence, it will be necessary to find a suitable time $\lambda$. It will be necessary to make $\lambda$-rigid transformations.
Fermi rigid coordinates for a general space-time {#s_7}
=================================================
Given a space-time and, in it, a specific time-like geodesic, we want to implement the equivalence principle by finding rigid coordinates that are, in turn, FC for the given geodesic.
First step. Fixing the rigid time: $\lambda$-rigid transformation {#s_7_1}
-----------------------------------------------------------------
We consider an arbitrary space-time and in it, an arbitrary time-like geodesic $G$. The first and most difficult step is to set the rigid time coordinate $\lambda$ for which there exists a $G$-isochronous geodesic congruence. It may be the case that:
1. From the beginning of the process we have a rigid system of coordinates. We will check whether the geodesic $G$ satisfies the corresponding equation (\[eq\_67\]). If it does, then it will be possible to find the $G$-isochronous congruence by solving (\[eq\_64\]). If equation (\[eq\_67\]) is not satisfied, then it will be necessary to perform $\lambda$-rigid transformations.
2. We do not have a rigid coordinate system to express the metrics of our space-time. In this case, we will need to solve the problem to find a set of rigid coordinates, as described in Section $\S$4 of the reference [@04-jaen_4], but restricted to the condition that the geodesic congruence potential must be $G$-isochronous.
We have no evidence that it is always possible to meet the above requirement, nor do we have examples to the contrary. However, the final purpose of this work is to construct a sufficiently broad theoretical framework whereby Fermi and rigid coordinates can be related for cases such as linear gravitational waves. We will work on this topic successfully in Section $\S$\[s\_8\]
In what follows, we will assume that we are able to write the space-time in a system of coordinates $\left\{ {\lambda,\vec x} \right\}$ in such a way that the geodesic $G$ is $\vec x(\lambda )$ and the potentials $\left\{ {\tau \left( {\vec x,\lambda } \right),\vec V\left( {\vec x,\lambda } \right)}\right\}$ are a $G$-isochronous congruence. This means that $\tau (\lambda ) = \tau(\lambda ,\vec x(\lambda )) = \lambda $, ${\partial _{\vec x}}\tau (\lambda ,\vec x(\lambda )) = 0$ and $\dot {\vec x}(\lambda ) = \vec V(\vec x(\lambda ),\lambda ) = \frac{{\partial H}}{{\partial \vec p}}(\vec x(\lambda ),\vec p = 0,\lambda )$.
As a consequence, the metric expressed in the starting rigid system $S$ can be written in the form: $$\label{eq_68}
d{s^2} = - d{\tau ^2} + {\gamma _{ij}}(d{x^i} - {V^i}d\lambda )(d{x^j} - {V^j}d\lambda ) + O\left( {{{\left| {\vec x -
\vec x(\lambda )} \right|}^2}} \right)$$ with ${\gamma _{ij}} = {\delta _{ij}} - {\sigma _{,{x^i}}}\;{\sigma _{,{x^j}}}$. That is to say, in linear order, we can eliminate the terms ${\tau _{,{x^i}}}{\tau _{,{x^j}}}$. In this expression, there may be terms $O\left( {{{\left| {\vec x - \vec x(\lambda )} \right|}^2}} \right)$ in both ${V^i}$ and $\tau$.
With this first step, undoubtedly the most difficult, we have defined a suitable rigid time $\lambda$. Consequently, the rigid motions and $\gamma$-rigid transformations will be defined using this time.
Second step. $\gamma$-rigid transformation {#s_7_2}
------------------------------------------
From a metric of the form (\[eq\_68\]) we make a $\gamma$-rigid transformation, $\vec x \to \vec y$, with the condition $\sigma ' = f(\lambda ) + O( {{{\left| {\vec x - \vec x(\lambda )} \right|}^2}} )$. This is achieved by a transformation such as (\[eq\_53\]) and choosing: $$\label{eq_69}
\vec \beta (\lambda ) = {\partial _{\vec x}}\sigma \left( {\vec x(\lambda ),\lambda } \right)$$ It is important to note that this transformation, although it does not leave the Euclidean metric unchanged, meets $O({{{\left| {\vec x - \vec x(\lambda )} \right|}^2}} ) = O( {{{\left| {\vec y - \vec y(\lambda )} \right|}^2}} )$ where $\vec y(\lambda )$ is the geodesic $\vec x(\lambda )$ in the new coordinates.
The result of this transformation on the metric (\[eq\_53\]), will be a metric of the form: $$\label{eq_70}
d{s^2} = - d{\tau ^2} + {\left( {d\vec y - \vec V'd\lambda } \right)^2} + O\left( {{{\left| {\vec y - \vec y(\lambda )}
\right|}^2}} \right)$$ where $\vec V'$ is defined by (\[eq\_55\]), according to the transformation $\vec x \to \vec y$ (\[eq\_53\]) with (\[eq\_69\]).
Third step. $\delta$-rigid transformation {#s_7_3}
-----------------------------------------
Starting from (\[eq\_70\]), we perform a $\delta$-rigid transformation (ordinary translation and rotation) $\vec y \to \vec z$ defined according to the geodesic $\vec y(\lambda )$ and the congruence $\vec V'$. That is, defined as in the Newtonian case but now using the congruence $\vec V'$ derived from performing the $\gamma$-rigid transformation above. That is, given by: $$\vec R(\lambda ) = \vec y(\lambda )\;\;;\;\;\;\vec \Omega (\lambda ) = \frac{1}{2}\left[ {{\partial _{\vec y}} \times
\vec V'} \right](\vec R(\lambda ),\lambda )$$ In $\vec z$ coordinates the geodesic will be simply $\vec z = 0$ and $O\left( {{{\left| {\vec y - \vec y(\lambda )}\right|}^n}} \right) = O\left( {{z^n}} \right)$. We will arrive at the metric form: $$\label{eq_72}
d{s^2} = - d{\tau ^2} + {\left( {d\vec z - \vec V''d\lambda } \right)^2} + O\left( {{z^2}} \right)$$ where, as in the non-relativistic case: $$\label{eq_73}
\vec V''\left( {\vec z,\lambda } \right) = {\left[ {\left( {\vec z\cdot{\partial _{\vec y}}} \right)\vec V' +
\frac{1}{2}\vec z \times \left( {{\partial _{\vec y}} \times \vec V'} \right)} \right]_{\vec y = \vec R}} + O\left(
{{z^2}} \right)$$ which does not include zero order but $z$ order. As in the Newtonian case, the property ${\partial _{\vec z}}\times \vec V'' = O({z^2})$ is met.
Fourth step. The proper time coordinate $t$ {#s_7_4}
-------------------------------------------
As the congruence is $G$-isochronous, the potential $\tau $ meets ${\left. {\bar d\tau }
\right|_G} = 0$ which is now ${\partial _{\vec z}}\tau \left( {0,\lambda } \right) = 0$. That is to say, until now we have a known $\tau \left( {\vec z,\lambda } \right)$ that has the form $\tau \left( {\vec z,\lambda } \right) = \lambda + O({z^2})$. We can still make $O({z^2})$ gauge transformations. This fact will allow us, starting from the metric (\[eq\_72\]), to perform a gauge transformation: $$\left\{ {\vec V'' = O\left( z \right),\tau = O\left( {{z^2}} \right)} \right\} \to {\rm{ }}\left\{ {{\vec V''^*} = O\left( {{z^2}} \right),{\tau ^*} = O\left( {{z^2}} \right)} \right\}{\rm{ }}$$ The equations that must be satisfied are: $$\label{eq_74}
{\partial _\lambda }{\tau ^*} + H\left( {\vec z,\vec p = {\partial _{\vec z}}{\tau ^*},\lambda } \right) = 0$$ $$\label{eq_75}
{\vec V''^*} = \frac{{\partial H}}{{\partial \vec p}}\left( {\vec z,\vec p = {\partial _{\vec z}}{\tau ^*},\lambda }
\right) = O\left( {{z^2}} \right)$$ We will take into account that the geodesic congruence is $G$-isochronous, ${\partial _{\vec z}}\tau\left( {\vec z = 0,\lambda } \right) = 0$, and we also require ${\partial _{\vec z}}{\tau^*}\left( {\vec z = 0,\lambda } \right) = 0$. Up to order $O({z^2})$ (\[eq\_74\]) and (\[eq\_75\]) become: $$\left\{ {\begin{array}{*{20}{l}}
{{\partial _\lambda }{\tau ^*} = 1 + O({z^2})}\\
{{\partial _{\vec z}}{\tau ^*} = {\partial _{\vec z}}\tau \;\; + \vec V'' + O({z^2})}
\end{array}} \right.$$ We note that the integrability conditions are again guaranteed because ${\partial _{\vec z}} \times \vec V'' = O({z^2})$ and by the fact that we are neglecting terms of order $O({z^2})$. As can be seen by direct calculation (explicitly using expression (\[eq\_73\]) for $\vec V''$), the solution is: $$\label{eq_77}
{\tau ^*}\left( {\vec z,\lambda } \right) = \tau \left( {\vec z,\lambda } \right)\;\; + \frac{1}{2}\vec z\cdot\vec V'' +
O\left( {{z^3}} \right)$$ Using the space-time coordinates $\left\{ {\lambda ,\vec z} \right\}$ and this gauge, the metric can be written as: $$d{s^2} = - d{\tau ^{*2}} + d{\vec z^2} + O\left( {{z^2}} \right)$$ which means that these rigid coordinates define the rigid locally inertial reference system attached to the geodesic $G$: $S_{IG}$.
We define the time coordinate $t$ of the system $S_{IG}$ according to $t = {\tau ^*}\left( {\vec z,\lambda} \right)$. Making the expression explicit, the result is: $$\label{eq_78}
t = \tau \;(\vec z,\lambda ) + \frac{1}{2}\vec z\cdot{\left[ {\left( {\vec z\cdot{\partial _{\vec y}}} \right)\vec V' +
\frac{1}{2}\vec z \times \left( {{\partial _{\vec y}} \times \vec V'} \right)} \right]_{\vec y = \vec R}} + O\left(
{{{\left| {\vec z} \right|}^3}} \right)$$ We should recall that in $\vec z$ coordinates, the geodesic is $\vec z = 0$ and the congruence $G$-isochronous, i.e., ${\partial _{\vec z}}\tau \left( {\vec z = 0,\lambda } \right) = 0$.
In $\left\{ {t,\vec z} \right\}$ coordinates the metric takes the form: $$d{s^2} = - d{t^2} + d{\vec z^2} + O\left( {{z^2}} \right)$$ which means that the $\left\{ {t,\vec z} \right\}$ coordinates are FRC.
Regarding the terms grouped under the symbol $O\left( {{z^2}} \right)$ (terms of order $z^2$ or higher), we observe that we could obtain a more accurate expression for the metric by solving (\[eq\_74\]) for ${\tau ^*}(\vec z,\lambda )$ with the condition (\[eq\_77\]). The same comments that are made above after equation (\[eq\_34\]) apply here.
In fact, beyond second order (first order in the metric), time $t$ and the metric no longer have clear physical meaning, and it may be preferable to work with the rigid time $\lambda$. In particular the transformation (\[eq\_78\]) is a $\lambda$-rigid transformation as long as we remain in second order. What we can say, at any order, is that the $S_{IG}$ system supports rigid coordinates $\left\{{\lambda ,\vec z} \right\}$ and in this sense $S_{IG}$ is defined for all orders in $z$.
As far as the principle of equivalence is concerned, we can conjecture the following statement, which we call the rigid equivalence principle.
> ***Rigid equivalence principle**: in any space-time and for any time-like geodesic $G$, it is always possible to construct a locally inertial rigid reference system $S_{IG}$ whose origin moves with the geodesic G and whose rotation coincides with the vorticity of a $G$-isochronous geodesic congruence at the origin. In the neighborhood of the origin of $S_{IG}$, it is possible to define a time with respect to which test particles move freely.*
Fermi rigid coordinates for radial time-like geodesics in Schwarzschild space-time {#s_7_5}
----------------------------------------------------------------------------------
We can apply what we have learned and find $S_{IG}$, thus finding FRC, for a Schwarzschild radial time-like geodesic $G$, beyond the escape geodesics. In order not to lengthen the study, we will only detail the first step, which is the only one that is not automatic. Once this first step has been carried out, the others can be followed without any more explanation than that given in the previous section, which we only mention here.
To cover the first step, we begin with the following stationary rigid covariant spherical symmetry form of the metric: $$d{s^2} = - d{\lambda ^2} + \left( {1 - \sigma _{,r}^2} \right){\left( {d{r^2} - Vd\lambda } \right)^2} + {r^2}d{\Omega
^2}$$ where $d{\Omega ^2} = d{\theta ^2} + {\sin ^2}\theta \;d{\phi ^2}$, the potentials $\sigma$ and $V$ are functions of the radial coordinate $r$ and the potential $\tau \;(r,\lambda )$ is simply $\tau \;(r,\lambda ) = \lambda $. Solving Einstein’s field equations in empty space we find: $$V =\pm \sqrt {\frac{K}{r} + V_0^2} \;\;\;;\;\;\;\sigma = \;\sqrt {\frac{{V_0^2}}{{1 + V_0^2}}} \;r$$ If $K=0 $ then $V=V_0$ and the space-time is Minkowski, written using a constant velocity $V_0$ radial congruence. If $V_0=0$ we recover the usual Painlevé-Gullstrand form for the Schwarzschild metric. However, the present form is not simply a rigid or gauge transformation from $V_0=0$ to try to include the required geodesics. These transformations would not lead us to the $G$-isochronous condition ${\left. {\bar d\tau } \right|_G} = 0$ for radial non-escape geodesics. What is important here is that we have found a solution that meets the requirements for implementing the rest of the steps outlined in the previous section, ${\left. {\bar d\tau } \right|_G} = 0$ because in fact $\tau = \lambda $. We have found a rigid covariant form for the Schwarzschild metric with a suitable time coordinate $\lambda$ for the required geodesic radial non-escape geodesic. For each family of radial geodesics characterized by a different value of $V_0$ and a different sign $\pm$ of $V$, we have a different coordinate system with a different time, $\lambda$. These systems are related by $\lambda$-rigid transformations.
By setting a value of $V_0$ and a sign of $V$, say $+$, and using Euclidean space coordinates, we have: $$\vec V = \sqrt {\frac{K}{{\left| {\vec x} \right|}} + V_0^2} \;\frac{{\vec x}}{{\left| {\vec x} \right|}}\;;\;\;\;\sigma
= \;\sqrt {\frac{{V_0^2}}{{1 + V_0^2}}} \;\left| {\vec x} \right|\;$$ The second step will be to perform a $\gamma$-rigid transformation, $\vec x \to \vec y$, with $\vec \beta (\lambda ) = {\partial _{\vec x}}\sigma \left( {\vec x(\lambda ),\lambda } \right)$, where $\vec x(\lambda )$ is the chosen geodesic $G$ solution of $\dot {\vec x}(\lambda ) = \vec V\left( {\vec x(\lambda )} \right)$. The new potential $\vec V'$ will be given according to (\[eq\_55\]).
The third step will consist of a $\delta$-rigid transformation $\vec y \to \vec z$ towards the locally inertial reference system $S_{IG}$ based on the $\vec V'$ potential that emerged in the previous step. The new potential will be $\vec V''$.
The fourth step will consist of eliminating the first order in $z$ of $\vec V''$ by a gauge transformation with $ {\tau ^*}\left( {\vec z,\lambda } \right)$ defined by (\[eq\_77\]), and finally defining time $t$ according to $t = {\tau ^*}\left( {\vec z,\lambda } \right)$ or (\[eq\_78\]).
Fermi rigid coordinates for linear plane gravitational waves {#s_8}
============================================================
In this section, we apply what we have learned in order to find FRC for a linear plane gravitational wave or, in other words, to find a rigid locally inertial reference system for these waves.
In the reference [@04-jaen_4] we find a set of rigid coordinates $\left\{ {\lambda ,\vec x}\right\}$ for a linear plane gravitational wave originally written in Gaussian coordinates $\{ {T,\vec X} \}$ as: $$\label{eq_83}
d{s^2} = - d{T^2} + d{\vec X^2} + {\varepsilon ^2}\left\{ {2{h_ \times }dX\;dY + {h_ + }(d{X^2} - d{Y^2})} \right\}$$ The $\{ {T,\vec X} \}$ coordinates are Gaussian so that the points at rest $\vec X = {\vec X_0}$ form a family of time-like geodesics of the proper time $T$. If the wave “disappears”, i.e. if ${h_\times } = {h_ + } = 0$, the geodesic $\vec X = {\vec X_0}$ will consist of points at rest in Minkowski space-time with standard coordinates. This is why we are interested in these geodesics.
We want to find FRC for one of the geodesics, say $G$, of this family. We can always adjust the coordinates so that the required geodesic becomes $\vec X = 0$.
In this section, and as far as wave linearity is concerned, we will work up to order $\varepsilon ^2$, even if we do not explicitly indicate it.
First step {#s_8_1}
----------
We have to find rigid coordinates for (\[eq\_83\]) such that the geodesic congruence is $G$-isochronous. Part of this work was already performed in reference [@04-jaen_4] for the same kind of waves. Equations (17) of [@04-jaen_4] were solved with $\tau (\vec X,\lambda ) = T(\vec X,\lambda ) = \lambda + \varepsilon (X{T_{(x)}} + Y{T_{(y)}})$, $\sigma (\vec X,\lambda ) = \varepsilon(X{\sigma _{(x)}} + Y{\sigma _{(y)}})$ where $\{T_{(x)},T_{(y)},\sigma _{(x)},\sigma _{(y)}\}$ are functions of $Z-\lambda$, and the equations that must be satisfied are:[^5]
$$\label{eq_84}
\begin{array}{l}
{\sigma_{(x)}}'^2 + {\sigma_{(y)}}'^2 = {T_{(x)}}'^2 + {T_{(y)}}'^2 \\
2{\sigma _{(x)}}'{\sigma_{(y)}}' - 2{T_{(x)}}' {T_{(y)}}' + {h_\times}''= 0 \\
2{\sigma_{(x)}}'^2 - 2 {T_{(x)}}'^2 + {h_+}'' = 0 \\
2 {T_{(y)}}'^2 - 2{\sigma_{(y)}}'^2 + {h_+}'' = 0 \\
\end{array}$$
If these conditions are met, we arrive at a metric form for (\[eq\_83\]), given by (\[eq\_39\]) with (\[eq\_40\]), but instead of having $\delta_{ij}$ in (\[eq\_40\]) we have $\Delta_{ij}$, which is also a flat space metric but not in the Euclidean form. Because of this, we have to solve $\Delta _{ij}dX^{i}dX^{j}=\delta_{ij}dx^{i}dx^{j}$ for a coordinate change $\vec{X}=\{X,Y,Z\}\to \vec{x}=\{x,y,z\}$. This change depends on $\lambda$, which in the space $\vec{X}$ acts as a parameter. This work was also completed in the reference [@04-jaen_4]. Linking the two transformations, $\{X,Y,Z,T\}\to \{x,y,z,\lambda\}$ and up to order $\varepsilon ^{2}$, we have the explicit result for the cross mode ($h_+=0$): $$\label{eq_85}
\begin{array}{l}
X = x + {\varepsilon ^2}y\left[ {{T_{(x)}}{T_{(y)}} - \frac{{{h_ \times }}}{2} + I + f(\lambda )} \right]\\
Y = y + {\varepsilon ^2}x\left[ {{T_{(x)}}{T_{(y)}} - \frac{{{h_ \times }}}{2} - I - f(\lambda )} \right]\\
Z = z + \frac{{{\varepsilon ^2}}}{2}xy{h_ \times }^\prime \\
T = \lambda + \varepsilon \left( {x{T_{(x)}} + y{T_{(y)}}} \right)
\end{array}$$ where $I \equiv {\rm{ }}\int {\left( {{T_{(x)}}{T_{(y)}}^\prime - {T_{(y)}}{T_{(x)}}^\prime } \right)dz} $ will be understood as a primitive, without an integration constant, because we have already taken it into account through the arbitrary function ${f(\lambda )}$. We want to clarify that in equation (27) of reference [@04-jaen_4], we made a small error in considering that each integral of (24)-[@04-jaen_4] provided an independent function when writing (27)-[@04-jaen_4]. Only if ${f_x}(\lambda ) = - {f_y}(\lambda ) = f(\lambda )$ (27)-[@04-jaen_4] is there a change towards rigid coordinates. It should be noted that, in [@04-jaen_4], the subsequent election of these functions fulfilled the condition ${f_x}(\lambda ) = - {f_y}(\lambda )$. This is why here, in (\[eq\_85\]), we express (24)-[@04-jaen_4] making the arbitrary functions explicit.
Unlike [@04-jaen_4], where the objective was to find some rigid coordinate system and therefore the arbitrary functions of $\lambda$ were fixed without a specific criterion, we now impose conditions with physical meaning that allow us to fix these functions. The main condition is to make the congruence $\vec X = {\vec X_0}$ $G$-isochronous, with $G$ being the geodesic selected: $\vec X = 0$. Because the proper time of the congruence is $T$ and the geodesic $G$ is $\vec x =0$, in the rigid coordinate system $\left\{ {\lambda ,\vec x} \right\}$ the $G$-isochronous condition is: $${\partial _{\vec x}}T(\vec x = 0,\lambda ) = 0$$ There is a secondary condition which, if not imposed, we could overcome with a $\gamma$-rigid transformation when we are carrying out the second step: the change (\[eq\_85\]) in the metric (\[eq\_83\]) will define the rigid potentials and in particular the potential $\sigma$. Of that potential, we require that: $$\label{eq_87}
{\partial _{\vec x}}\sigma (\vec x = 0,\lambda ) = 0$$ Taking into account these conditions, together with (\[eq\_84\]), we obtain the conditions on the functions ${T_{(x)}},{T_{(y)}}$ and $f(\lambda)$: $$\label{eq_88}
{T_{(x)}}\left( {0 - \lambda } \right) = 0\;;\;{T_{(y)}}\left( {0 - \lambda } \right) = 0\;;f(\lambda ) = 0$$ where $0-\lambda$ means that we have taken $z=0$.
The fulfillment of these conditions ensures that (\[eq\_85\]) is a change towards rigid coordinates with the geodesic congruence $\vec X = {\vec X_0}$ being $G$-isochronous, where $G$ is the geodesic $\vec X = 0$ and also satisfies (\[eq\_87\]).
A particularly interesting case is that of the cross mode of a monochromatic linear plane wave, with frequency $\omega$. This corresponds to considering (\[eq\_83\]) with $h_+=0$ and: $$\varepsilon ^{2}\;h_{\times}(Z,T)=A^{2}\sin (\omega (Z-T))$$ i.e., $\varepsilon =A$ and $h_{\times}=\sin [\omega (Z-\lambda )]$.
As a solutions of (\[eq\_84\]), also satisfying (\[eq\_88\]), we choose: $$\begin{array}{l}
{T_{(x)}} = \sqrt 2 \sin [\frac{\omega }{2}(Z - \lambda )]\; + \sqrt 2 \sin [\frac{\omega }{2}\lambda ]\;\\
{T_{(y)}} = \sqrt 2 \cos [\frac{\omega }{2}(Z - \lambda )] - \sqrt 2 \cos [\frac{\omega }{2}\lambda ]
\end{array}$$ From (\[eq\_85\]) and recalling that, according to (\[eq\_88\]), we set $f(\lambda ) = 0$: $$\label{eq_91}
\begin{array}{*{20}{l}}
{X = x + {A^2}\frac{1}{2}y\left\{ {\sin \left[ {\omega (z - \lambda )} \right] + 4\sin \left[ {\frac{\omega }{2}z}
\right] - 4\sin \left[ {\omega (\frac{z}{2} - \lambda )} \right] - 2\sin \left[ {\omega \lambda } \right] - 2\omega z}
\right\}}\\
{Y = y + {A^2}\frac{1}{2}x\left\{ {\sin \left[ {\omega (z - \lambda )} \right] + 4\sin \left[ {\frac{\omega }{2}z}
\right] - 4\sin \left[ {\omega (\frac{z}{2} - \lambda )} \right] - 2\sin \left[ {\omega \lambda } \right] - 2\omega z}
\right\}}\\
{Z = z + {A^2}\frac{1}{2}\omega xy\cos \left[ {\omega (z - \lambda )} \right]}\\
{T = \lambda + \sqrt 2 A\left\{ {x\left( {\sin \left[ {\frac{\omega }{2}(z - \lambda )} \right] + \sin \left[
{\frac{\omega }{2}\lambda } \right]} \right) + y\left( {\cos \left[ {\frac{\omega }{2}(z - \lambda )} \right] - \cos
\left[ {\frac{\omega }{2}\lambda } \right]} \right)} \right\}}
\end{array}$$
Second step. $\gamma$-rigid transformation {#s_8_2}
------------------------------------------
This step will not be necessary as we have been able to set ${\partial _{\vec x}}\sigma (\vec x = 0,\lambda ) =
0$ during the first step. However, simply for consistency of notation we will set: $\vec y = \vec x$.
Third step. $\delta$-rigid transformation {#s_8_3}
-----------------------------------------
From (\[eq\_91\]) we can calculate the potential $\vec V$ associated with the geodesics $\vec X =$ constant. We obtain: $$\label{eq_92}
\begin{array}{*{20}{l}}
{\vec V = {A^2}y\left\{ {\frac{1}{2}\cos \left[ {\omega (z - \lambda )} \right] - 2\cos \left[ {\omega (\frac{z}{2} -
\lambda )} \right] + \omega \cos \left[ {\omega \lambda } \right]} \right\}{\partial _x} + }\\
{{A^2}x\left\{ {\frac{1}{2}\cos \left[ {\omega (z - \lambda )} \right] - 2\cos \left[ {\omega (\frac{z}{2} - \lambda )}
\right] + \omega \cos \left[ {\omega \lambda } \right]} \right\}{\partial _y} - }\\
{{A^2}\frac{1}{2}{\omega ^2}xy\sin \left[ {\omega (z - \lambda )} \right]{\partial _z}}
\end{array}$$ which together with $\tau (\vec x,\lambda ) = T(\vec x,\lambda )$ defines the geodesic congruence as $G$-isochronous.
From (\[eq\_91\]), $\vec X=0$ implies $\vec x=0$, so it is not necessary to make a rigid translation: $\vec R(\lambda ) = 0$. Direct calculation of (\[eq\_92\]) shows us that $\vec \Omega(\lambda ) = \frac{1}{2}{\left[ {{\partial _{\vec x}} \times \vec V} \right]_{\vec x = 0}} = 0$ so that we will not need to make a rigid rotation either. Once again, only for consistency of notation we will set: $\vec z = \vec y = \vec x$.
The difference between the change we have now, expression (\[eq\_91\]), and that found in (28)-[@04-jaen_4] is that the latter leads us to possible rigid coordinates. Now, from all possible rigid coordinates, we select those in which the metric takes the form (\[eq\_72\]), with a potential $\vec V'' = \vec V$, given by (\[eq\_92\]), that fulfills the conditions of being null and irrotational on the geodesic. It still contains linear terms in the space coordinates.
We can say that we have now found the rigid locally inertial reference system attached to the geodesic $\vec X=0$, $S_{IG}$, with coordinates $\left\{ {\lambda ,\vec x} \right\}$. The change (\[eq\_91\]) globally defines $S_{IG}$, i.e., it is not limited to any order in $x$.
Fourth step. The proper time coordinate $t$ {#s_8_4}
-------------------------------------------
In the final step, we perform the gauge transformation $\{ \vec V,\tau \} \to \{ {{\vec V}^*} = O\left( {{x^2}} \right),{\tau ^*}\}$, defined in (\[eq\_74\]) (\[eq\_75\]), and define the time coordinate $t$ of $S_{IG}$ as $t = {\tau^*}\left( {\vec z,\lambda } \right)$.
Taking into account (\[eq\_78\]), recalling that $\vec y = \vec x$ and maintaining, for ease of calculation, $\vec z \ne \vec x$, we can write: $$t = {\left\{ {T(\vec z,\lambda ) + \frac{1}{2}\vec z\cdot{{\left[ {\left( {\vec z\cdot{\partial _{\vec x}}} \right)\vec
V} \right]}_{\vec x = 0}}} \right\}_{\vec z = \vec x}} + O\left( {{{\left| {\vec x} \right|}^3}} \right)$$ where, only after the derivation and the substitution of the geodesic $\vec x = 0$, do we make the substitution $\vec z = \vec x$. The result of this operation, taking into account expression (\[eq\_92\]) for ${\vec V}$, is: $$t = T(\vec x,\lambda ) - \frac{1}{2}{A^2}\omega xy\cos \left[ {\omega \lambda } \right] + O\left( {{{\left| {\vec x}
\right|}^3}} \right)$$ With this information, we obtain the change from the original $\{ {T,\vec X} \}$ coordinates to FRC attached to the geodesic $\vec X=0$, $\{t,\vec x\} $: $$\label{eq_95}
\begin{array}{*{20}{l}}
{X = x + {A^2}\frac{1}{2}y\left\{ {\sin \left[ {\omega (z - t)} \right] + 4\sin \left[ {\frac{\omega }{2}z} \right] -
4\sin \left[ {\omega (\frac{z}{2} - t)} \right] - 2\sin \left[ {\omega t} \right] - 2\omega z} \right\}}\\
{Y = y + {A^2}\frac{1}{2}x\left\{ {\sin \left[ {\omega (z - t)} \right] + 4\sin \left[ {\frac{\omega }{2}z} \right] -
4\sin \left[ {\omega (\frac{z}{2} - \lambda )} \right] - 2\sin \left[ {\omega t} \right] - 2\omega z} \right\}}\\
{Z = z + {A^2}\frac{1}{2}\omega xy\cos \left[ {\omega (z - t)} \right]}\\
{T = t + \frac{1}{2}{A^2}\omega xy\cos \left[ {\omega t} \right] + O\left( {{{\left| {\vec x} \right|}^3}} \right)}
\end{array}$$ Finally, we can now study the geodesics $\vec X = {\vec X_0}$. We can find the relationship between $\left\{ {{X_0},{Y_0},{Z_0}} \right\}$ and $\left\{ {t = 0,{x_0},{y_0},{z_0}} \right\}$ by making $t=0$ in expression (\[eq\_95\]). For the geodesics $\left\{ {t = 0,{x_0},{y_0},{z_0} = 0} \right\}$ we find the expressions: $$\begin{array}{*{20}{l}}
{x = {x_0} - {A^2}\frac{1}{2}{y_0}\sin \left[ {\omega \;t} \right]}\\
{y = {y_0} - {A^2}\frac{1}{2}{x_0}\sin \left[ {\omega t} \right]}\\
{z = {A^2}\frac{1}{2}\omega {x_0}{y_0}\left( {1 - \cos \left[ {\omega \;t} \right]} \right)}\\
{T = t + \frac{1}{2}{A^2}\omega {x_0}{y_0}\cos \left[ {\omega t} \right] + O\left( {{{\left| {{{\vec x}_0}} \right|}^3}}
\right)}
\end{array}$$ Up to order $O\left( {\left| {{{\vec x}_0}} \right|} \right)$, this coincides with the usual result. Of course, up to order $O\left( {\left| {{{\vec x}_0}} \right|} \right)$, the metric will have the standard Minkowski form. This result is also valid up to order $O( {{{\left| {{{\vec x}_0}} \right|}^2}} )$. In contrast, up to order $O( {{{\left| {{{\vec x}_0}} \right|}^2}} )$, the metric will no longer has a Minkowski form, although it will still have a rigid form.
Conclusions
===========
In this paper, we have progressed further and moved deeper into the previous study [@04-jaen_4], where we found a set of rigid coordinate systems for linear plane gravitational waves. Two related issues remained unanswered: how to arrive at a rigid coordinate system from among the set found; and what relation those systems have with the commonly used Fermi coordinates.
Our goal was to find coordinates that were both rigid and Fermi. To do this, we have analyzed rigid covariant transformations beyond the usual rigid motions.
Given any space-time and, in it, any time-like geodesic, we have identified four steps that allow us to construct a rigid coordinate system that is also a Fermi coordinate system. We call them: Fermi rigid coordinates (FRC), as opposed to Fermi normal coordinates (FNC).
The first step consists of representing the space-time in rigid coordinates that fulfill the condition of being $G$-isochronous for the chosen geodesic. This step is important because it provides us with a time coordinate with which to define generalized rigid transformations and thereby to arrive at the locally inertial system. Although we have been able to apply this step to several space-times, particularly to linear plane gravitational waves in the cross mode, and we do not have any counterexample, it has not been shown that it is always possible to carry this step out. That is why we present an implementation of the principle of equivalence, which we call the principle of rigid equivalence, in the form of a conjecture.
We have succeeded in finding Fermi rigid coordinates for a type of linear gravitational wave, and this could be an interesting and useful alternative when trying to invest Fermi normal coordinates with physical meaning. It seems that, using Fermi rigid coordinates, we are free to interpret space rigid coordinates as those that label, at least approximately, points on a body made of rigid material, in the sense that the cohesive forces of this material are much more intense than the tensions caused by the passage of the gravitational wave. This feature may be interesting for those experimenters who need good correspondence between the mathematical symbols used in the theory and the laboratory tools that they use to design experimental devices. Whatever the case, we believe that the rigid formulation proposed herein can assist in clarifying some open questions related to the correspondence between the mathematical formulation of general relativity and how to perform laboratory experiments.
The Fermi coordinate condition does not determine which ones are the “good coordinates”. It would be interesting to study the relationship between the different proposals that incorporate the rigidity in relativity [@Bona83; @Bel90; @Llosa04; @Coll07] with that presented here and with the Fermi condition. This study is beyond the scope of this work and we hope to deal with it in detail in the future.
Acknowledgments {#acknowledgments .unnumbered}
---------------
I want to thank Alfred Molina for carefully reading a previous draft of the paper and providing useful criticism that led to improvements; and Lluís Bel, without whose inspiration and encouragement, hardly any of this series of papers would have occurred to me.
[99]{}
Jaén, X. and Molina, A. (2013). *Rigid motions and generalized Newtonian gravitation*, General Relativity and Gravitation 45 : 1531-1546.
Jaén, X. and Molina, A. (2014). *Homothetic motions and Newtonian cosmology*, General Relativity and Gravitation 46 : 1-14.
Jaén, X. and Molina, A. (2015). *On the meaning of Painlevé–Gullstrand synchronization*, General Relativity and Gravitation 47 : 1-16.
Jaén, X. and Molina, A. (2017). *Rigid covariance as a natural extension of Painlevé-Gullstrand space-times: gravitational waves*, General Relativity and Gravitation 49 : 108.
Born, M. *Über die Dynamik des Elektrons in der Kinematik des Relativitätsprinzips*. Phys. Zeitschr, 10, 814-817,1909.
Bona, C. *Rigid-motion conditions in special relativity*. Physical Review D, 27(6), 1243, 1983.
Bel, L. *Rigid motion invariance of Newtonian and Einstein’s theories of general relativity*. Verdaguer, E., Cespedes, J., Jaume, G. (Eds.) Recent developments in gravitation-Proceedings of the Relativity Meeting-89. World Scientific (1990).
Bel, L. *Static elastic deformations in general relativity*. (1996) arXiv preprint gr-qc/9609045.
Llosa, J., and Soler, D. *Reference frames and rigid motions in relativity*. Classical and Quantum Gravity, 21(13), 3067, (2004).
Coll, B. Coll, B. *About deformation and rigidity in relativity*. J. Phys. Conf. Ser. 66(1), 1-17. https://doi.org/10.1088/1742-6596/66/1/012001 (2007).
Fermi, E. (1922). *Sopra i Fenomeni che Avvengono in Vicinanza di Una Linea Oraria.*, Rendiconti dell’Accademia Nazionale dei Lincei 31 : 21-23, 51-52, 101-103.
Manasse, F. and Misner, C. W. (1963). *Fermi normal coordinates and some basic concepts in differential geometry*, Journal of mathematical physics 4 : 735-745.
Riemann, B. (1869). *Sur les hypothèses qui servent de fondement à la Gèomètrie*, Annali di Matematica Pura ed Applicata (1867-1897) 3 : 309-326.
Ni, W.-T. and Zimmermann, M. (1978). *Inertial and gravitational effects in the proper reference frame of an accelerated, rotating observer*, Physical Review D 17 : 1473.
Li, W.-Q. and Ni, W.-T. (1979). *Coupled inertial and gravitational effects in the proper reference frame of an accelerated, rotating observer*, Journal of Mathematical Physics 20 : 1473-1480.
Li, W.-Q. and Ni, W.-T. (1979). *Expansions of the affinity, metric and geodesic equations in Fermi normal coordinates about a geodesic*, Journal of Mathematical Physics 20 : 1925-1929.
Nesterov, A. I. (1999). *Riemann normal coordinates, Fermi reference system and the geodesic deviation equation*, Classical and Quantum Gravity 16 : 465.
Marzlin, K.-P. (1994). *Fermi coordinates for weak gravitational fields*, Physical Review D 50 : 888.
Rakhmanov, M. (2005). *Response of test masses to gravitational waves in the local Lorentz gauge*, Physical Review D 71 : 084003.
Marzlin, K.-P. (1994). *The physical meaning of Fermi coordinates*, General relativity and gravitation 26 : 619-636.
Delva, P. and Angonin, M.-C. (2012). *Extended Fermi coordinates*, General Relativity and Gravitation 44 : 1-19.
Dyson, F. J. (1990). *Feynman’s proof of the Maxwell equations*, Am. J. Phys 58 : 209-211.
Painlevé, P. (1922). *Le Mecanique Classique et la Theorie de la Relativite., L’Astronomie* 36 : 6-9.
Gullstrand, A., 1922. *Allgemeine ösung des statischen einkörperproblems in der Einsteinschen gravitationstheorie*. Almqvist & Wiksell, .
Möller, C. (1952). *The theory of relativity*, Oxford: Clarendon Press.
Bel, L. *Eppur si muove!*, Rizzi, G., Ruggiero, M.L. (Eds.) Relativity in rotating frames, relativistic physics in rotating reference frames. Kluwer Academic Publisher, Dordrecht (2004).
[^1]: Dept. de Física, Universitat Politènica de Catalunya, Barcelona, Spain, e-mail address: [email protected]
[^2]: Greek indices are used for space-time $\mu ,\nu = 0,1,2,3$. $ {\eta _{\mu \nu }} = {\rm{diag}}( - 1,1,1,1)$. Latin indices are used for space $i,j = 1,2,3$. We use the usual Euclidean vector notation: scalar products, “$\cdot$“ , norms, $x \equiv \left| {\vec x} \right|$, vector products “$ \times $“ and the operation of raising and lowering indices are always performed using the Euclidean metric with Euclidean components ${\delta _{ij}}$ and the corresponding totally antisymmetric volume form ${\delta _{ijk}}$ with ${\delta _{123}}=1$.
[^3]: The notation $A + O\left( {{x^n}} \right)$ means that $O\left( {{x^n}} \right)$ is an object of the same kind as $A$ (scalar, vector, tensor, etc.) whose components in the natural basis associated with $\vec x$ are of the order ${\left| {\vec x} \right|^n}$ or higher.
[^4]: The total derivative of a scalar with respect to time will be indicated as $\dot {f} $. In $S$, with $\vec x = {x^m}{\hat X_m}$, we will use the notation $\dot {\vec x} \equiv {\dot {x}^m}{\hat X_m}$ and $\ddot {\vec x} \equiv {\ddot {x}^m}{\hat X_m}$; in $S'$, with $\vec y = {y^m}{\hat Y_m}$, $\dot {\vec y} \equiv {\dot {y}^m}{\hat Y_m}$ and $\ddot {\vec y} \equiv {\ddot {y}^m}{\hat Y_m}$.
[^5]: In this section we will use a prime, $f' $, to indicate the derivation of a function with respect to its argument.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'It is shown that the quasi-normal modes arise, in a natural way, when considering the oscillations in unbounded regions by imposing the radiation condition at spatial infinity with a complex wave vector $k$. Hence quasi-normal modes are not peculiarities of gravitation problems only (black holes and relativistic stars). It is proposed to consider the space form of the quasi-normal modes with allowance for their time dependence. As a result, the problem of their unbounded increase when $r\to \infty$ is not encountered more. The properties of quasi-normal modes of a compact dielectric sphere are discussed in detail. It is argued that the spatial form of these modes (especially so-called surface modes) should be taken into account, for example, when estimating the potential health hazards due to the use of portable telephones.'
author:
- 'V. V. Nesterenko'
- 'A. Feoli'
- 'G. Lambiase'
- 'G. Scarpetta'
title: |
Quasi-normal modes of a dielectric sphere\
and some their implications
---
Introduction
============
Quasi-normal modes (qnm) are widely used now in black hole physics and in relativistic theory of stellar structure (see, for example, Refs. [@Nollert; @FN; @KS]). The corresponding eigenfrequencies are complex numbers, however it is not due to the dissipative processes but it is a consequence of unbounded region occupied by the oscillating system. The latter naturally leads to the energy loss due to the wave emission (for example, gravity waves).
In this paper we would like to show that the quasi-normal modes are not the peculiarities of the gravitational problems only. Actually they appear, in a natural way, when considering the oscillating systems unbounded in space. The necessary condition for emergence of such modes is imposing the radiation condition at spatial infinity on the field functions. It is this condition that leads to the characteristic behaviour of the quasi-normal modes, namely, these solutions to the relevant equations exponentially decay in time when $t\to
\infty$ and simultaneously they exponentially rise at spatial infinity $r\to \infty$. An interesting and physically motivated example is provided here by the oscillations of electromagnetic field connected with a compact dielectric sphere placed in unrestricted homogeneous media with a different refraction index or in vacuum. Taking here the formal limit $\varepsilon \to \infty$ one passes to a perfectly conducting sphere ($\varepsilon $ is the refraction index of the sphere material). In this case the quasi-normal modes describing the electromagnetic oscillations outside the sphere are tractable analytically. We propose to consider the spatial form of a quasi-normal modes with allowance for their time dependence. Doing in this way one can escape the exponential rise of quasi-normal modes at spatial infinity.
The eigenfrequencies of a dielectric sphere are complex $\omega
=\omega'-i\, \omega''$, where $\omega '$ is the free oscillation (radian) frequency and $\omega ''$ is its relaxation time. These modes can be classified as the interior and exterior ones and, at the same time, as volume modes and surface modes.
In physical applications the surface modes turn out to be important, for example, when estimating the health hazards due to the use of portable telephones. The point is the eigenfrequencies of a dielectric sphere with physical characteristics close to those of a human head lay in the GSM 400 MHz frequency band which has been used in a first generation of mobile phone systems and now is considered for using again. In this situation one can assume that the surface modes excited by a cellular phone will lead to higher heat generation in the tissues close to a head surface as compared with the predictions of routine calculations in this field.
The layout of the paper is as follows. In Sec. II we show that the quasi-normal modes are the eigenfunctions of unbounded oscillating regions. As a simple example the quasi-normal modes of a perfectly conducting sphere are considered. The Sec. III is devoted to the consideration of the quasi-normal modes of a dielectric sphere. The main features of these modes are revealed and their classification is presented. The implication of these quasi-normal modes for estimation of the health hazards of portable telephones is considered in Sec. IV. In Conclusion (Sec. V) the main results are formulated and their relation to the general theory of open systems is discussed.
Quasi-normal modes as the eigenfunctions of unbounded oscillating regions
=========================================================================
Here we show in the general case in what way complex frequencies and quasi-normal modes appear when considering harmonic oscillations in unbounded regions. Let a closed smooth surface $S$ divides the $d$-dimensional Euclidean space $\mathbb{R}^d$ into a compact internal region $D_{\text{ in}}$ and noncompact external region $D_{\text{ex}}$. We consider here a simple scalar wave equation $$\label{2-1} \left ( \Delta -\frac{1}{c^2} \frac{\partial
^2}{\partial t^2} \right )u(t, {\textbf x})=0\,{,}$$ where $c$ is the velocity of oscillation propagation and $\Delta $ is the Laplace operator in $\mathbb{R}^d$. For harmonic oscillations $$\label{2-2} u(t,{\textbf x})=e^{-i\omega
t}u({\textbf x})$$ the wave equation (\[2-1\]) is reduced to the Helmholtz equation $$\label{2-3}
(\Delta +k^2)\,
u(\textbf{x}) =0, \quad
k={\omega}/{c}\,{.}$$
The oscillations in the internal region $D_{\text{in}}$ are described by an infinite countable set of normal modes $$\label{2-4}
u_n(t,\textbf{x})=e^{-i\omega _n t}u_n(\textbf{x}), \quad
n=1,2,\ldots .$$ The spatial form of the normal modes (the functions $u_n(\textbf{x})$) is determined by the boundary conditions which are imposed upon the function $u(\textbf{x})$ on the internal side of the surface $S$. These conditions should fit the physical content of the problem under study. The set of normal modes is a complete one. Hence any solution of (\[2-3\]) obeying relevant boundary conditions can be expanded in terms of the normal modes $u_n(\textbf{x})$.
When considering the oscillations in the external domain $D_{\rm
ex }$ one imposes, in addition to the conditions on the compact surface $S$, a special requirement concerning the behavior of the function $u(\textbf{x})$ at large $r\equiv |\textbf{x}|$. In the classical mathematical physics [@RR] the radiation conditions, proposed by Sommerfeld [@Sommerfeld; @FM], are used here $$\label{2-5} \lim_{r\to \infty} r^{\frac{d-1}{2}}u(r)=
\textrm{const}\,{,}\qquad \lim_{r\to\infty}r^{\frac{d-1}{2}}\left
( \frac{\partial u}{\partial r} -i ku\right )= 0\,{.}$$ For real values of the wave vector $k$ (for real frequencies $\omega$) the solution to Eq. (\[2-3\]), which obeys the radiation conditions (\[2-5\]) and reasonable boundary condition on a compact surface $S$, identically vanishes. In this case the Laplace operator entering the Helmholtz equation (\[2-3\]) has no eigenfunctions with real eigenvalues.
The physical content of the radiation conditions is very clear. They select only the oscillations with real frequencies driven by external sources which are situated in a compact spatial area. From the mathematical standpoint, these conditions ensure the [*uniqueness*]{} of the solutions to [*inhomogeneous*]{} wave or Helmholtz equations with external sources on the right-hand side, when these solutions are considered in the external region $D_{\text{ex}}$ or in the whole space $D_{\text{in}}+D_{\text{ex}}$ in the case of compound media.
Here the question arises, how to change minimally the conditions in the [*homogeneous*]{} problem at hand in order to get nonzero solutions, i.e., eigenfunctions in unbounded regions. The energy conservation law prompts a simple way to construct nonzero solutions to the homogeneous wave equation (\[2-1\]) or to the Helmholtz equation (\[2-3\]) describing the outgoing waves at spatial infinity, namely, one has to introduce [*complex*]{} frequencies $\omega =
\omega'- i\,\omega '',\quad \omega ''>0$. We may hope that in this case the factor $e^{-\omega ''t}$ will describe the decay of the initial solutions in time accounting the fact that outgoing waves take away the energy. In other words, we are dealing here with the radiation of the energy with the amplitude decaying in time.
Indeed, if we remove the requirement of reality of the wave vector $k$, then the homogeneous wave equation (2.1) and the Helmholtz equation (2.3) will have nonzero solutions with [*complex frequencies*]{}, these solutions obeying the radiation conditions (\[2-5\]) and a common boundary condition on a compact surface $S$ (for instance, Dirichlet or Neumann conditions). In quantum mechanics the radiation condition with a complex wave number $k$ is known as the Gamov condition which singles out the resonance states in the spectrum of the Hamiltonian [@Gamov; @G1; @G2].
When introducing the radiation conditions and proving the respective uniqueness theorem in the text books [@RR] only the real wave vector $k$ is considered. The possibility of existence of quasi-normal modes with complex frequencies satisfying the radiation conditions at spatial infinity with a complex $k$ is not mentioned usually. We are aware only of alone textbook where the eigenfunctions with complex frequencies are noted in this context. It is the article written by Sommerfeld in the book [@FM], where it is emphasized that the uniqueness in this problem is only up to the eigenfunctions with complex frequencies, i.e., up to the quasi-normal modes.
Thus imposing the radiation conditions with real $k$ we remove the quasi-normal modes from our consideration only. However this cannot prevent the excitation of these modes in the real physical problem. Hence, when dealing with systems unbounded in space (open systems) one has always to investigate the consequences of qnm excitation. An example of such a problem will be considered in Sec. III.
As a very simple and physically motivated example of quasi-normal modes we consider here the oscillations of electromagnetic field outside a perfectly conducting sphere of radius $a$. In this case the electric and magnetic fields are expressed in terms of two scalar functions $f_{kl}^{\rm{TE}}(r)$ and $f_{kl}^{\rm{TM}}(r)$ (Debye potentials [@Stratton]) which are the radial parts of the solutions to the scalar wave equation (\[2-1\]). Outside the perfectly conducting sphere placed in vacuum the solution to the Helmholtz equation (\[2-3\]) obeying the radiation conditions (\[2-5\]) has the form $(d=3)$ $$\label{2-6} f_{kl}(r)=C\, h^{(1)}_l\left ( \frac{\omega}{c}\,r
\right ){,}\quad r>a\,{,}$$ where $ h^{(1)}_l(z)$ is the spherical Hankel function of the first kind [@AS]. At the surface of perfectly conducting sphere the tangential component of the electric field should vanish. This leads to the following frequency equation for TE-modes $$\label{2-7} h^{(1)}_l\left ( \frac{\omega}{c}\,a \right
)=0{,}\quad l\geq 1$$ and for TM-modes $$\label{2-8} \frac{d}{d r}\left ( r \,h^{(1)}_l\left (
\frac{\omega}{c}\,r \right ) \right )=0, \quad r=a,\quad l\geq
1\,{.}$$ The spherical Hankel function $h^{(1)}_l(z)$ is $e ^{i z}$ multiplied by the polynomial in $1/z$ of a finite order [@AS]. Hence frequency equations (\[2-7\]) and (\[2-8\]) have a finite number of roots which are in the general case complex numbers. For $l=1$ (the lowest oscillations) Eqs.(\[2-7\]) and (\[2-8\]) assume the form $(z=a\,\omega/c)$ $$\begin{aligned}
h^{(1)}_l(z)&=&-\frac{1}{z}\,e^{i z}\left ( 1+\frac{i}{z} \right
)=0\quad (\text{TE modes}),
\label{2-8a} \\
\frac{d }{d z}\left (z\, h^{(1)}_l(z)\right
)&=&-\frac{i}{z^2}\,e^{i z}\left ( z^2+i z-1 \right )=0\quad
(\text{TM modes})\,{.}
\label{2-9}\end{aligned}$$ Thus the lowest eigenfrequencies are $$\begin{aligned}
\frac{\omega}{c}&=&-\frac{i}{a}\quad (\mbox{TE modes})\,{,}
\label{2-10}
\\
\frac{\omega}{c}&=&-\frac{1}{2a}(i\pm\sqrt 3)\quad (\text{TM
modes})\,{.} \label{2-11}\end{aligned}$$
The complex eigenfrequencies lead to a specific time and spatial dependence of the respective natural modes and ultimately of the electromagnetic fields. So, with allowance of (\[2-10\]), we obtain $$\label{2-12} e^{-i \omega t}f_{k1}^{\text{TE}}(r)=-i
\,C\,\frac{a}{r}\,e^{(r-ct)/a}\left ( 1-\frac{a}{r} \right ),
\quad r\geq a\,{.}$$ Thus, the eigenfunctions are exponentially going down in time and exponentially going up when $r$ increases. Such a time and spatial behaviour is a direct consequence of radiation conditions (\[2-5\]) and it is typical for eigenfunctions describing oscillations in external unbounded regions, the physical content and details of oscillation process being irrelevant. The eigenfunctions corresponding to complex eigenvalues are called [*quasi-normal modes*]{} keeping in mind their unusual properties [@Nollert]. The physical origin of such features is obvious, in fact we are dealing here with [*open systems*]{} in which the energy can be radiated to infinity. Therefore in open systems field cannot acquire a stationary state.
Quasi-normal modes do not obey the standard completeness condition and the notion of norm cannot be defined for them [@Nollert]. Therefore these eigenfunctions cannot be used for expansion of the classical field with the aim to quantize it and to introduce the relevant Fock operators. The treatment of these problems can be found, for example, in [@Ching; @Leung; @Chang].
It is worthy to investigate the spatial form of the of quasi-normal modes with allowance for their time dependence. Indeed, these solutions have the character of propagating waves that are eventually going to spatial infinity. Let us take the point which is sufficiently far from the region with nontrivial dynamics in the system under study. Obviously, it has sense to say about the value of the quasi-normal mode at a given point only after arrival at this point of the wave described by this mode. The maximal value of the quasi-normal mode is observed just at the moment of its arrival at this point. At the later moments the quasi-normal mode is dumping due to its characteristic time dependence. Indeed, taking into account all this we obtain for the maximal observed value of the quasi-normal mode (\[2-12\]) the following physically acceptable expression $$\label{2-13}\left ( e^{-i \omega t}f_{k1}^{\text{TE}}(r)\right
)_{\text{max-obs}}=-i \,C\,\frac{a}{r}\left ( 1-\frac{a}{r} \right
), \quad r\geq a\,{.}$$ Thus, in our consideration the problem of unbounded (exponential) rising of quasi-normal modes, when $r\to \infty$, does not arise.
In order to associate with resonance phenomenon a single square integrable eigenfunction, rather sophisticated methods are used, for example, complex scaling [@scaling] (known also as the complex-coordinate method or as the complex-rotational method).
The necessary condition for appearing the quasi-normal modes is imposing the radiation conditions at spatial infinity on the field functions. When other conditions are used at spatial infinity, the quasi-normal modes do not arise. For example, if we demand that the solution to the wave equation (\[2-1\]) becomes the sum of incoming and outgoing waves when $r\to \infty$ then the spectrum of the variable $k^2$ in the Helmholtz equation (\[2-3\]) will be positive and continuous.
Quasi-normal modes of a dielectric sphere
=========================================
The same situation, with regard to quasi-normal modes, takes place when we consider the oscillations of compound unbounded media. In this case in both the regions $D_{\text{in}}$ and $D_{\text{ex}}$ the wave equations are defined $$\begin{aligned}
\left ( \Delta -\frac{1}{c_{\text{in}}^2}\frac{\partial }{\partial
t^2} \right ) u_{\text{in}}(t,\mathbf{x}) &=& 0, \quad
\mathbf{x}\in D_{\text{in}}\,{,} \label{2-14}\\ \left ( \Delta
-\frac{1}{c_{\text{ex}}^2}\frac{\partial }{\partial t^2} \right )
u_{\text{ex} }(t,\mathbf{x}) &=& 0, \quad \mathbf{x}\in D_{\text{
ex}} \label{2-15}\end{aligned}$$ with the matching conditions at the interface $S$, for example, of the following kind $$\begin{aligned}
u_{\text{ in}}(t,\mathbf{x})&=&u_{\text{ex}}(t,
\mathbf{x}){,} \label{2-16}
\\
\lambda_{\text{ in}}\frac{\partial u_{\text{in
}}(t,\mathbf{x})}{\partial n_{\text{in
}}(\mathbf{x})}&=&\lambda_{\text{ ex}}\frac{\partial u_{\rm
ex}(t,\mathbf{x})}{\partial n_{\text{ ex }}(\mathbf{x})}, \quad
\mathbf{x}\in S\,{,} \label{2-17}\end{aligned}$$ where $n_{\text{in}}(\textbf{x})$ and $n_{\text{ ex}}(\textbf{x})$ are the normals to the surface $S$ at the point $\textbf{x}$ for the regions $D_{\text{in}}$ and $D_{\text{ex}}$, respectively. The parameters $c_{\text{in}}$, $c_{\text{ex}}$, $\lambda
_{\text{in}}$, and $\lambda _{\text{ex}}$ specify the material characteristics of the media. At the spatial infinity the solution $u_{\text{ex}}(t, \mathbf{x})$ should satisfy the radiation conditions (\[2-5\]). For real $k$ we again have only zero solution in this problem, both functions $u_{\text{in}}(t,
\mathbf{x})$ and $u_{\text{ex}}(t, \mathbf{x})$ vanishing. However, the wave equations (\[2-14\]) and (\[2-15\]) have nonzero solutions with complex frequencies, i.e. quasi-normal modes, which satisfy the matching conditions (\[2-16\]) and (\[2-17\]) at the interface $S$ and radiation conditions (\[2-5\]) at spatial infinity. It is important, that the frequencies of oscillations in internal ($D_{\text{in}}$) and external ($D_{\text{ex}}$) regions are the same. A typical example here is the complex eigenfrequencies of a dielectric sphere. This problem has been investigated by Debye in his PhD thesis concerned with the light pressure on a material particles [@Debye].
Let us consider a sphere of radius $a$, consisting of a material which is characterized by permittivity $\varepsilon_1 $ and permeability $\mu_1$. The sphere is assumed to be placed in an infinite medium with permittivity $\varepsilon_2 $ and permeability $\mu_2$. In the case of spherical symmetry the solutions to Maxwell equations are expressed in terms of two scalar Debye potentials $\psi$ (see, for example, textbooks [@Stratton; @Jackson]): $$\begin{aligned}
\mathbf{E}^{\text{TM}}_{lm}&=&\bm{\nabla} \times
\bm{\nabla}\times(\mathbf{r}\psi^{\text{TM}}_{lm}),\quad
\mathbf{H}^{\text{TM}}_{lm}=-i\,\omega \,\bm{\nabla} \times
(\mathbf{r}\psi^{\text{TM}}_{lm})\quad (\text{E-modes}), \nonumber
\\ \mathbf{E}^{\text{TE}}_{lm}&=&i\,\omega \,\bm{\nabla} \times
(\mathbf{r}\psi^{\text{TE}}_{lm}),\quad
\mathbf{H}^{\text{TE}}_{lm}=\bm{\nabla} \times
\bm{\nabla}\times(\mathbf{r}\psi^{\text{TE}}_{lm})\quad
(\text{H-modes})\,{.} \label{3-1}\end{aligned}$$ These potentials obey the Helmholtz equation and have the indicated angular dependence $$\label{3-2} \left ( \mathbf{\nabla}^2+k^2_i\right
)\psi_{lm}=0,\quad
k_i^2=\varepsilon_i\,\mu_i\,\frac{\omega^2}{c^2}, \quad i=1,2\quad
(r\neq a); \quad \psi_{lm}(\mathbf{r})=f_l(r)Y_{lm}(\Omega)\,{.}$$
Equations (\[3-2\]) should be supplemented by the boundary conditions at the origin, at the sphere surface, and at infinity. In order for the fields to be finite at $r=0$ the Debye potentials should be regular here. At the spatial infinity we impose the radiation conditions with the goal to find the spectrum of eigenfunctions with complex frequencies (quasi-normal modes in the problem at hand). At the sphere surface the standard matching conditions for electric and magnetic fields should be satisfied [@Stratton].
In view of all this the Helmholtz equation (\[3-2\]) becomes now the spectral problem for the Laplace operator multiplied by the discontinuous factor $-1/(\varepsilon
(r)\,\mu(r))$ $$\label{3-2a}-\, \frac{1}{\varepsilon (r)\,\mu(r)}\Delta
\,\psi_{\omega l m}(r)=\frac{\omega^2}{c^2}\,\psi_{\omega l m}(r),
\quad r\neq a\,{,}$$ where $$\varepsilon (r)\,\mu(r)=
\begin{cases}
\varepsilon_1\,\mu_1\,{,}&r<a \,{,}\\
\varepsilon_2\,\mu_2\,{,}&r>a \,{.}
\end{cases}$$ In this problem the spectral parameter is $\omega^2/c^2$.
In order to obey the boundary conditions at the origin and at spatial infinity formulated above, the solution to the spectral problem (\[3-2a\]) should have the form $$\label{3-2b} f_{\omega l}(r)=C_1\,j_l(k_1r)\,{,} \quad r<a,\quad
f_{\omega l}(r)=C_2\,h^{(1)}_l(k_2r)\,{,} \quad r>a\,{,}$$ where $j_l(z)$ is the spherical Bessel function and $h_l^{(1)}(z)$ is the spherical Hankel function of the first kind [@AS], the latter obeys the radiation conditions (\[2-5\]).
Now we address the matching conditions at the sphere surface. By making use of Eqs. (\[3-1\]) we can write, in an explicit form, the radial (r) and tangential (t) components of electric and magnetic fields in the case of spherical symmetry. For TE-modes these equations read $$\begin{aligned}
E^{\text{TE}}_{klm,\text{r}}&=&0\,{,} \label{a-20}\\
E^{\text{TE}}_{klm,\text{t}}&=&a_{lm}(k)\,f_{kl}^{\text{TE}}(r)\,\mathbf{X}_{lm}\,{,}
\label{a-21}\\ H^{\text{TE}}_{klm,\text{r}}&=&\frac{1}{kr}\left (
\frac{\varepsilon}{\mu} \right
)^{1/2}\sqrt{l(l+1)}\,a_{lm}(k)f_{kl}^{\text{TE}}(r)\,Y_{lm} \,{,}
\label{a-22}\\ H^{\text{TE}}_{klm,\text{t}}&=&\frac{i}{kr}\left (
\frac{\varepsilon}{\mu} \right
)^{1/2}a_{lm}(k)\,\frac{d}{dr}\left( rf_{kl}^{\text{TE}}(r)
\right)\,\mathbf{X}_{lm}^\perp\,{,} \label{a-23}\end{aligned}$$ and the same for the TM-modes $$\begin{aligned}
E^{\text{TM}}_{klm,\text{r}}&=&-\frac{1}{kr}\left (
\frac{\mu}{\varepsilon} \right
)^{1/2}\sqrt{l(l+1)}\,b_{lm}(k)f_{kl}^{\text{TM}}(r)\,Y_{lm} \,{,}
\\ E^{\text{TM}}_{klm,\text{t}}&=&-\frac{i}{kr}\left (
\frac{\mu}{\varepsilon} \right
)^{1/2}b_{lm}(k)\,\frac{d}{dr}\left( rf_{kl}^{\text{TM}}(r)
\right)\,\mathbf{X}_{lm}^\perp\,{,} \label{a-25}\\
H^{\text{TM}}_{klm,\text{r}}&=&0\,{,} \label{a-26}\\
H^{\text{TM}}_{klm,\text{t}}&=&b_{lm}(k)\,f_{kl}^{\text{TM}}(r)\,\mathbf{X}_{lm}\,{.}
\label{a-27}\end{aligned}$$ Here $\mathbf{X}_{lm}$ are the vector spherical harmonics [@Jackson] $$\label{vsh}
\mathbf{X}_{lm}(\theta,\phi)=\frac{\mathbf{L}\,Y_{lm}(\theta,\phi)}{\sqrt{l(l+1)}}\,{,}\quad
l\geq1\,{,}$$ where $\mathbf{L}$ is the angular momentum operator $$\mathbf{L}=-i\,(\mathbf{r}\times \bm{\nabla})\,{.}$$ The vector spherical harmonic $\mathbf{X}_{lm}^{\perp}$ is obtained from $\mathbf{X}_{lm}$ after rotation by the angle $\pi/2$ around the normal $\mathbf{n}=\mathbf{r}/r$. From Eqs.(\[a-20\]) – (\[a-27\]) it follows, in particular, that the tangential components of electric field in TE- and TM-modes are orthogonal each other and the same holds for the magnetic field. It implies that the matching conditions on the sphere surface do not couple TE- and TM-modes.
At the sphere surface the tangential components of electric and magnetic fields are continuous (see Eqs. (\[a-20\]) – (\[a-27\])). As a result, the eigenfrequencies of electromagnetic field for this configuration are determined [@Stratton] by the frequency equation for the TE-modes $$\label{3-3} \Delta^{\text{TE}}_l(a\omega)\equiv
\sqrt{\varepsilon_1\mu_2}\,\hat j_l'(k_1a)\,\hat h_l(k_2a)-
\sqrt{\varepsilon_2\mu_1}\,\hat j_l(k_1a)\,{\hat h_l}'(k_2a)=0\,{}$$ and by the analogous equation for the TM-modes $$\label{3-4} \Delta^{\text{TE}}_l(a\omega)\equiv
\sqrt{\varepsilon_2\mu_1}\,\hat j_l'(k_1a)\,\hat h_l(k_2a)-
\sqrt{\varepsilon_1\mu_2}\,\hat j_l(k_1a)\,{\hat
h_l}'(k_2a)=0\,{,}$$ where $k_i=\sqrt{\varepsilon_i\mu_i}\,\omega/c,\quad i=1,2$ are the wave numbers inside and outside the sphere, respectively, and $\hat j_l(z)$ and $\hat h_l(z)$ are the Riccati-Bessel functions [@AS] $$\label{3-5} \hat j_l(z)=z\,j_l(z)=\sqrt{\frac{\pi
z}{2}}\,J_{l+1/2}(z)\,{,}\quad \hat
h_l(z)=z\,h_l^{(1)}(z)=\sqrt{\frac{\pi
z}{2}}\,H^{(1)}_{l+1/2}(z)\,{.}$$ In Eqs. (\[3-3\]) and (\[3-4\]) the orbital momentum $l$ assumes the values $1,2,\ldots $, and prime stands for the differentiation with respect of the arguments $k_1a$ and $k_2a$ of the Riccati-Bessel functions.
The frequency equations for a dielectric sphere of permittivity $\varepsilon $ placed in vacuum follow from Eqs. (\[3-3\]) and (\[3-4\]) after putting there $$\label{3-12} \varepsilon_1=\varepsilon, \quad
\varepsilon_2=\mu_1=\mu_2=1\,{.}$$ The roots of these equations have been studied in the Debye paper [@Debye] by making use of an approximate method. As the starting solution the eigenfrequencies of a perfectly conducting sphere were used. These frequencies are different for electromagnetic oscillations inside and outside sphere. Namely, inside sphere they are given by the roots of the following equations $(l\geq 1)$ $$\begin{aligned}
j_l\left ( \frac{\omega}{c}\,a \right)&=&0 \quad
(\text{TE-modes})\,{,} \label{3-13}\\ \frac{d}{dr}
\left(r\,j_l\left ( \frac{\omega}{c}\,a \right)\right)&=&0\,{,}
\quad r=a \quad (\text{TM-modes})\,{,} \label{3-14}\end{aligned}$$ while outside sphere they are determined by Eqs. (\[2-7\]) and (\[2-8\]). The frequency equations for perfectly conducting sphere (\[2-7\]), (\[2-8\]) and (\[3-13\]), (\[3-14\]) can be formally derived by substituting (\[3-12\]) into frequency equations (\[3-3\]) and (\[3-4\]) and taking there the limit $\varepsilon \to \infty$.
Approximate calculation of the eigenfrequencies of a dielectric sphere without using computer [@Debye] didn’t allow one to reveal the characteristic features of the respective eigenfunctions (quasi-normal modes). The computer analysis of this spectral problem was accomplished in the work [@Gastine] where the experimental verification of the calculated frequencies was accomplished also by making use of radio engineering measurements.
These studies enable one to separate all the dielectric sphere modes into the [*interior*]{} and [*exterior*]{} modes and, at the same time, into the [*volume*]{} and [*surface*]{} modes. It is worth noting that all the eigenfrequencies are complex $$\label{3-15} \omega=\omega '-i\,\omega''\,{.}$$ Thus we are dealing with “leaky modes”.
The classification of the modes as the interior and exterior ones relies on the investigation of the behaviour of a given eigenfrequency in the limit $\varepsilon \to \infty$. The modes are called ”interior” when the product $k\,a= \sqrt
{\varepsilon}\,\omega \,a/c$ remains finite in the limit $\varepsilon \to \infty$, provided the imaginary part of the frequency ($\omega ''$) tends to zero. The modes are referred to as ”exterior” when the product $k\,a/\sqrt{\varepsilon}=
\omega\,a/c$ remains finite with growing $\omega''$. In the first case the frequency equations for a dielectric sphere (\[3-3\]) and (\[3-4\]) tend to Eqs. (\[3-13\]) and (\[3-14\]) and in the second case they tend to Eqs. (\[2-7\]) and (\[2-8\]). The order of the root obtained will be denoted by the index $r$ for interior modes and by $r'$ for exterior modes. Thus $\text{TE}_{\,lr}$ and $\text{TM}_{\,lr}$ denote the interior TE- and TM-modes, respectively, while $\text{TE}_{\,lr'}$ and $\text{TM}_{\,lr'}$ stand for the exterior TE- and TM-modes.
For fixed $l$ the number of the modes of exterior type is limited because the frequency equations for exterior oscillations of a perfectly conducting sphere (\[2-7\]) and (\[2-8\]) have finite number of solutions (see the preceding Section). In view of this, the number of exterior TE- and TM-modes is given by the following rule. For even $l$ there are $l/2$ exterior TE-modes and $l/2$ exterior TM-modes, for odd $l$ the number of the modes $\text{TE}_{\,l\,r'}$ is $(l+1)/2$ and the number of the modes $\text{TM}_{\,l\,r'}$ equals $(l-1)/2$.
An important parameter is the $Q$ factor $$\label{3-16} Q_{\text{rad}}=\frac{\omega'}{2\omega''}=
2\,\pi\,\frac{\text{stored energy}}{\text{radiated energy per
cycle}}\,{.}$$ For exterior modes the value of $Q_{\text{rad}}$ is always less than 1, hence these modes can never be observed as sharp resonances. At the same time for $\varepsilon $ greater than 5, the $Q_{\text{rad}}$ for interior modes is greater than 10 and it can reach very high values when $\varepsilon \to \infty$.
For physical implications more important is the classification in terms of [*volume*]{} or [*surface*]{} modes according to whether $r> l$ or $l> r$. For volume modes the electromagnetic energy is distributed in the whole volume of the sphere while in the case of surface modes the energy is concentrated in the proximity of the sphere surface. The exterior modes are the first roots of the characteristic equations and it can be shown that they are always surface modes.
![Electric energy density $r^2\,E_{\text{t}}^2$ for the surface (A) and volume (B) TE-modes of a dielectric sphere with $\varepsilon =40$ placed in vacuum.[]{data-label="Plot:s-v-modes"}](pic1.eps "fig:"){width="75mm"} ![Electric energy density $r^2\,E_{\text{t}}^2$ for the surface (A) and volume (B) TE-modes of a dielectric sphere with $\varepsilon =40$ placed in vacuum.[]{data-label="Plot:s-v-modes"}](pic2.eps "fig:"){width="75mm"}
Figure 1 shows a typical spatial behaviour of the surface and volume modes of a dielectric sphere.
Thus a substantial part of the sphere modes (about one half) belong to the interior surface modes. It is important that respective frequencies are the [*first*]{} roots of the characteristic equations.
In order to escape the confusion, it is worth noting here that the surface modes in the problem in question obey the same boundary conditions at the sphere surface and when $r\to \infty$ as the volume modes do. Hence, these surface modes cannot be classified as the evanescent surface waves propagating along the interface between two media (propagating waves along dielectric waveguides [@Jackson], surface plasmon waves on the interface between metal bulk and adjacent vacuum [@Raether; @BPN] and so on). When describing the evanescent waves one imposes the requirement of their exponential decaying away from interface between two media. In this respect the evanescent surface wave differ from the modes in the bulk.
Implication of QNM of a dielectric sphere for estimation of the health hazard of portable telephones
====================================================================================================
Here we shall argue that the features of the quasi-normal modes of a dielectric sphere (namely, existence of surface and volume modes) should be taken into account, in particular, when estimating the potential health hazards due to the use of the cellular phones. The safety guidelines in this field [@Health] are based on the findings from animal experiments that the biological hazards due to radio waves result mainly from the temperature rise in tissues[^1] and a whole-body-averaged specific absorption rate (SAR) below 0.4 W/kg is not hazardous to human health. This corresponds to a limits on the output from the cellular phones (0.6 W at 900 MHz frequency band and 0.27 W at 1.5 GHz frequency band). Obviously, the [*local*]{} absorption rate should be also considered especially in a human head [@WF].
In such studies the following point should be taken into account. The parts of human body (for example, head) posse the eigenfrequencies of electromagnetic oscillations like any compact body. In particular, one can anticipate that the eigenfrequencies of human head are close to those of a dielectric sphere with radius $a\approx 8$ cm and permittivity $\varepsilon
\approx 40$ (for human brain $\varepsilon =44.1$ for 900 MHz and $\varepsilon =42.8$ for 1.5 GHz [@WF]). Certainly, our model is very rough, however for the evaluation of the order of the effect anticipated (see below) it is sufficient. By making use of the results of calculations conducted in the work [@Gastine] one can easily obtain the eigenfrequencies of a dialectic sphere with the parameters mentioned above. For $\text{TE}_{l1}$ modes with $l=1,2,3$ we have, respectively, the following frequencies: 280 MHz, 420 MHz, and 545 MHz. For $\text{TM}_{l1}$ modes with $l=1,2,3$ the resonance frequencies are 425 MHz, 540 MHz, and 665 MHz. The imaginary parts of these eigenfrequencies are very small so the $Q$ factor in Eq. (\[3-16\]) responsible for radiation is greater than 100.
These eigenfrequencies belong to a new GSM 400 MHz frequency band which is now being standardized by the European Telecommunications Standards Institute. This band was primarily used in Nordic countries, Eastern Europe, and Russia in a first generation of mobile phone system prior to the introduction of GSM.
Due to the Ohmic losses the resonances of a dielectric sphere in question are in fact broad, overlapping and, as the result, they cannot be manifested separately. Indeed, the electric conductance $\sigma$ of the human brain is rather substantial. According to the data presented in Ref. [@WF] $\sigma\simeq 1.0$ S/m. The eigenfrequencies of a dielectric dissipative sphere with allowance for a finite conductance $\sigma $ can be found in the following way. As known [@LL] the effects of $\sigma $ on electromagnetic processes in a media possessing a common real dielectric constant $\varepsilon$ are described by a complex dielectric constant $\varepsilon_{\text{diss}}$ depending on frequency $$\label{4-1}
\varepsilon_{\text{diss}} =\varepsilon +i\frac{4\pi \sigma}{\omega}\,{.}$$ The eigenfrequencies $\omega$, calculated for a real $\varepsilon $, are related to eigenfrequencies $\omega_{\text{diss}}$ for $\varepsilon_{\text{diss}}$ by the formula [@LL] $$\label{4-2}
\omega_{\text{diss}} =\frac{\omega}{\sqrt{\varepsilon_{\text{diss}} }}\simeq \omega
-2\pi\, i \,\frac{\sigma}{\varepsilon}\,{.}$$ The corresponding factor $Q_{\text{diss}}$ is $$\label{4-3} Q_{\text{diss}}=\frac{\omega'_{\text{diss}}}{2
\omega''_{\text{diss}}}\simeq
\frac{\varepsilon \, \omega}{4\pi\, \sigma}\,{.}$$ Substituting in this equation the values $\omega /2\pi =0.5\cdot 10^9\;\text{Hz}, \quad \varepsilon =40,
\quad \sigma = 1\,\text{S/m}=9\cdot 10^9\; \text{s}^{-1}$ one finds $$\label{4-4}
Q_{\text{diss}}\simeq\frac{20}{18}\simeq 1\,{.}$$
Thus the real spectrum of electromagnetic oscillations in the problem under study is practically a continuous band around the frequency 400 MHz. The radiation of the cellular telephone with frequency laying in this band, will excite (practically with the same amplitudes) all the neighbouring modes of a dissipative dielectric sphere. In order to get the upper bound for the anticipated effect (see below) we assume that the number of excited modes is sufficiently large so that the half of these modes are surface modes.[^2] Thus one can expect that the resulting spatial configuration of electric and magnetic fields inside a dielectric sphere with Ohm losses will follow, to some extent, the spatial behaviour of the relevant natural modes of the sphere, volume and surface ones. It is obvious that due to excitation of surface modes the maximum values of electric and magnetic fields inside dissipative sphere will be shifted to its outer part $r> a/2$.
When assuming the total number of the surface modes to be the same as those for the volume modes and consequently it is equal to a half of all the dielectric sphere modes, then one can anticipate that the temperature rise in the head tissues close to head surface may be by a factor 1.5 higher in comparison with the standard calculations using the numerical methods without special allowance for the spatial behaviour of the relevant natural modes.
However the numerical methods used for estimation of the temperature rise in human tissues due to the radio frequency irradiation do not take into account this effect. Indeed, such calculations (see, for example, paper [@WF]) are carried out in two steps. First the electric and magnetic fields inside the human body are calculated by solving the Maxwell equations with a given source (antenna of a portable telephone). The electric field gives rise to conduction currents with the energy dissipation rate $\sigma \,E^2/2$, where $\sigma $ is a conduction constant. In turn it leads to the temperature rising. The second step is the solution of the respective heat conduction equation (or more precisely, bioheat equation [@WF]) with found local heat sources $\sigma \,E^2/2$ and with allowance for all the possible heat currents. Hence, for this method the distribution of electric field inside the head is of primary importance. The spatial behaviour of the eigenfunctions characterize the system as a whole, and these properties cannot be taken into account by local methods for calculating the solution to partial differential equations (in our case, to the Maxwell equations).
Conclusion
==========
We have shown that such different, at first glance, notions as quasi-normal modes in black hole physics, Gamov states in quantum mechanics, and quasi-bound states in the theory of open electromagnetic resonators [@Vaynstayn] have the same origin, namely, all these are the eigenmodes of oscillating unbounded domains. By making use of a simple but physically motivated example of electromagnetic oscillations outside a perfectly conducting sphere, which admits analytical treatment, we have easily shown the main features of such oscillations, namely, their exponential decaying in time and, simultaneously, their exponential grows at spatial infinity. These properties are a direct consequence of the radiation condition which is met by qnm at spatial infinity. It is shown also that the exponential rising of qnm at infinity is not observable because the time dependence of qnm should be taken into account here. In the considered example the qnm are the outgoing spherical waves for $r>a$. This point is disregarded in all the attempts to treat qnm mathematically, in particular, to formulate the completeness condition and the relevant expansions in terms of qnm. We have considered the qnm modes in the problem without inhomogeneous potential (instead of the potential the boundary conditions at the surface of the sphere are introduced). It enables us to infer the conclusion stated above clear and easy.
It is argued also that imposing the standard radiation conditions with a real wave vector $k=\omega /c$ does not prevent us from the necessity to investigate the eigenmodes with complex $\omega'$s, i.e., qnm in a given problem.
The importance of this is demonstrated by investigating the role of the qnm in the problem of estimating the potential health hazards due to the use of portable telephones. The general analysis of the qnm spectra of a dielectric sphere with allowance for the dissipative processes enables us to estimate quantitatively the expected effect, namely, a possible temperature rise in the tissues laying in the outer part of the human head may be 1.5 times greater as compared with the inner part of the brain.
The predicted effect is, in some sense, analogous to the usual (but weakly manifested here) skin-effect. It is not surprising because the substantial conductivity of the sphere material plays a principal part in our consideration. Due to this conductivity the individual resonances of a sphere become very overlapping and, being not observed separately, many of them (volume and surface ones) are excited by the cellular telephone irradiation with the frequency in this band. In view of different spatial behaviour, the surface and volume modes will lead to different spatial distributions of the net heat inside the sphere (and also inside a human head).
When we have the quasi-normal modes instead the usual normal modes it implies that we are dealing with the open systems [@Open]. Open systems admit a dual description: on the one hand, they can be considered from the “inside” point of view, treating the coupling to the environment as a – not necessarily small – perturbation. From this point of view, one can study the (discreet) eigenvalues of the system, the width of resonances and the resulting decay properties [@Ching; @Leung]. On the other hand, open systems allow to take the “outside” point of view, considering the system as a perturbation of the environment. The typical quantity to be investigated from this point of view is the [*scattering matrix*]{} ($S$-matrix), i.e. the amplitude for passing from a given incoming field configuration to a certain outgoing configuration as a function of energy [@Vaynstayn].
The registration of the quasi-normal modes of a black hole is considered now as a possible way to detect this object [@Nollert; @KS]. In this connection it is surpassing that till now the quasi-normal modes have not been used in an analogous acoustic problem, i.e., for description of sound generation by ringing body [@Raylaygh; @Lamb; @Morse]. At the same time one can find in literature the statement (without proving) that we hear ringing quasi-normal modes of a bell when we hear the bell sound [@Moss].
This paper was completed during the visit of on of the authors (VVN) to Salerno University. It is his pleasant duty to thank G.Scarpetta and G. Lambiase for the kind hospitality extended to him. VVN was supported in part by the Russian Foundation for Basic Research (Grant No. 03-01-00025). The financial support of INFN is acknowledged. The authors are indebted to A.V. Nesterenko for preparing the figure.
[99]{} H.-P. Nollert, Class. Quantum Grav. [**16**]{}, R159 (1999). V. P. Frolov and I. D. Novikov, [*Black Hole Physics*]{} (Kluwer Academic Publishers, Dordrecht, 1998). K. D. Kokkotas and B. G. Schmidt, Living Rev.Relativity, [**2**]{}, 1 (1999); arXiv:gr-qc/9909058. I. Rubinstein, L. Rubinstein, [*Partial Differential Equations in Classical Mathematical Physics*]{} (Cambridge University Press, Cambridge, UK, 1998). A. Sommerfeld, [*Partial Differential Equations of Physics*]{} (Academic Press, New York, 1949). P. Frank and R. Mises, [*Die Differential- und Integralgleihungen der Mechanik und Physik*]{}, II Physikalischer Teil (Dover Publications, New York, 1961). G. Gamov, Z. Phys. [**51**]{}, 2004 (1928). E. U. Condon and R. W. Gurney, Phys. Rev. [**33**]{}, 127 (1929). R. de la Madrid, Lect. Notes Phys. [**622**]{}, 327 (2003); arXiv:. J. A. Stratton, [*Electromagnetic Theory*]{} (McGraw-Hill, New York, 1941). M. Abramowitz and I. Stegun, eds., [*Handbook of Mathematical Functions*]{} (Dover Publications, New York, 1972). E. S. C. Ching, P. T. Leung, W. M. Suen,, and K. Young, Phys. Rev. Lett. [**74**]{}, 4588 (1995). P. T. Leung, S. Y. Liu, S. S. Tong, and K. Young, Phys. Rev. A [**49**]{}, 3068 (1994). R. K. Chang and A. J. Campillo, [*Optical Processes in Microcavities*]{} (1996). Y. K. Ho, Phys. Reports [**99**]{}, 1 (1983). P. Debye, [*Ann. Phys. (Leipzig)*]{} [**30**]{}, 57 (1909); G. Mie, [*Ann. Phys. (Leipzig)*]{} [**25**]{}, 377 (1908). J. D. Jackson, [*Classical Electrodynamics*]{}, 3rd ed. (J. Wiley & Sons, New York, 1999). M. Gastine, L. Courtois, and J. L. Dormann, IEEE Trans. Microwave Theory and Techniques,vol. MTT-15, No. 12, pp. 694–700, December 1967. H. Raether, [*Surface Plasmons*]{} (Springer, Berlin, 1988). M. Bordag, I. G. Pirozhenko, V. V. Nesterenko, J Phys. A [**38**]{}, 11027 (2005). International Commission on Non-Ionizing Radiation Protection, ”IC-NIRP statement — Health issues related to the use of hand-held radiotelephones and base transmitters,” [*Health Phys.*]{} [**70**]{}, No. 4, pp.587–593, Apr. 1996. B. E. Sernelius, [*Europhys. Lett.*]{} [**60**]{}, 643 (2002). J. Wang and O. Fujiwara, IEEE Transactions on Microwave Theory and Techniques, [**47**]{}, No. 8, 1528 (1999). L. D. Landau and E. M. Lifshitz, [*Electrodynamics of Continuouis Media*]{}, Vol. 8 of [*Course of Theoretical Physics*]{} (Pergamon, New York, london, 1960). L. A. Vaynstayn, [*Theory of open resonators and open waveguides*]{} (Sovetskoe Radio, Moscow, 1966; English transaltion 1969). K.-H. Li, Phys. Reports, [**134**]{}, 1–85 (1986). J. W. Strutt, Baron Rayleigh, [*The theory of sound*]{}, vols. I, II (Macmillan, London,1926). H. Lamb, [*The dynamical theory of sound*]{}, 2nd ed. (Adward Arnold, London, 1931). F. Morse, [*Vibraions and sound*]{} (Gostechizdat, Moscow, Leningrad, 1949). I. G. Moss, [*Can you hear the shape of a bell?: Asymptotics of quasinormal modes,*]{} Nucl. Phys. B (Proc. Suppl.) [**104**]{}, 181 – 184 (2002).
[^1]: In principle, non-ionizing radiation can lead also to other effects in biological tissues [@Sernelius].
[^2]: As it was shown in preceding Section this relation between the number of the surface and volume modes holds only for the spectra as a whole.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We propose a simple method to prove non-smoothness of a black hole horizon. The existence of a $C^1$ extension across the horizon implies that there is no $C^{N + 2}$ extension across the horizon if some components of $N$-th covariant derivative of Riemann tensor diverge at the horizon in the coordinates of the $C^1$ extension. In particular, the divergence of a component of the Riemann tensor at the horizon directly indicates the presence of a curvature singularity. By using this method, we can confirm the existence of a curvature singularity for several cases where the scalar invariants constructed from the Riemann tensor, [*e.g.*]{}, the Ricci scalar and the Kretschmann invariant, take finite values at the horizon. As a concrete example of the application, we show that the Kaluza-Klein black holes constructed by Myers have a curvature singularity at the horizon if the spacetime dimension is higher than five.'
author:
- '${}^{1}$Masashi Kimura[^1], ${}^{2}$Hideki Ishihara[^2], ${}^{2}$Ken Matsuno[^3] and ${}^{3,4}$Takahiro Tanaka[^4],'
title: |
A simple diagnosis of non-smoothness of black hole horizon\
[*Curvature singularity at horizons in extremal Kaluza-Klein black holes* ]{}
---
Introduction
============
To test whether our world is the higher-dimensional spacetime, we need to identify phenomena which clearly indicate the existence of extra dimensions. Recently, the study of higher dimensional black holes has attracted much attention under the expectation that they may have characteristic features of extra dimensions. For example, in the higher dimensional scenarios based on the TeV gravity mini black holes might be produced in a linear collider [@Banks:1999gd; @Dimopoulos:2001hw; @Giddings:2001bu; @Ida:2002ez; @Ida:2005ax; @Ida:2006tf] or in cosmic ray events [@Argyres:1998qn; @Feng:2001ib; @Anchordoqui:2001cg] unlike the case of four dimensional gravity.
Since the sizes of extra dimensions should be compact from a realistic point of view, we should study higher dimensional spacetime with compactified extra dimensions, [*i.e.*]{}, Kaluza-Klein (KK) spacetime. In this paper, we focus on higher dimensional black holes in KK spacetime (KK black holes). As a first step, it would be important to investigate exact solutions of KK black holes to understand their qualitative feature. In the five dimensional case, recent studies showed that there exist a variety of KK black holes called squashed KK black holes [@Dobiasch:1981vh; @Gibbons:1985ac; @Gauntlett:2002nw; @Gaiotto:2005gf; @Ishihara:2005dp; @Wang:2006nw; @Yazadjiev:2006iv; @Nakagawa:2008rm; @Tomizawa:2008hw; @Matsuno:2008fn; @Tomizawa:2008rh; @Stelea:2008tt; @Tomizawa:2008qr; @Gal'tsov:2008sh; @Bena:2009ev; @Tomizawa:2010xq; @Mizoguchi:2011zj; @Chen:2010ih; @Stelea:2011fj; @Nedkova:2011hx; @Tatsuoka:2011tx; @Nedkova:2011aa; @Mizoguchi:2012vg]. However, in general, to construct an exact solution of KK black hole is difficult because of the less symmetry except for special cases. In fact, if $D \ge 6$ and the number of the non-compact dimensions is four, the KK black holes constructed from multi black holes solutions [@Myers:1986rx] are the only family of exact solution of KK black holes with spherical topology.
Though one might think that we cannot construct an exact solution of multi black holes since the gravitational force is only attractive, it is possible if each black hole has the same mass and charge[^5], where the gravitational attractive force is balanced with the Coulomb force. Such exact solutions are known as Majumdar-Papapetrou solutions [@Majumdar:1947eu; @Papaetrou:1947ib; @Hartle:1972ya], and then higher-dimensional generalization was considered by Myers [@Myers:1986rx]. In Ref. [@Myers:1986rx], Myers constructed KK black holes by placing an infinite number of black holes in a lattice configuration, which is equivalent to placing a single black hole with an appropriate periodic identification of space.
In Ref. [@Candlish:2007fh] Candlish and Reall showed that higher dimensional multi black holes have a non-smooth event horizon[^6] unlike the case of four dimensional multi black holes, which have an analytic horizon [@Hartle:1972ya]. In $D=5$ we can find a $C^2$ (but not $C^3$) extension across the horizon. By contrast, in $D \ge 6$ the metric is not $C^2$ but $C^1$ at the horizon since some components of the second derivatives of the metric always diverge at the horizon for any extension across the horizon. This means the existence of a curvature singularity at the horizon in $D \ge 6$.
The result in Ref. [@Candlish:2007fh] seems to indicate that KK black holes constructed by Myers [@Myers:1986rx] also have a curvature singularity at the horizon because they are constructed from higher dimensional multi black holes. However it is non trivial whether this expectation is correct or not since we consider an infinite number of black holes in the case of KK black holes, which might be qualitatively different from the case of a finite number of black holes. In fact, as shown in Ref. [@Candlish:2007fh], five dimensional KK black holes with a $S^1$ compactified extra dimension have an analytic event horizon[^7] in contrast to the case of a finite number of black holes, whose horizon is not $C^3$. Extrapolating the results in $D=4$ and $D=5$, we can expect that the horizon might be analytic if the number of the non-compact dimensions is four. Therefore, we would like to investigate whether or not KK black holes with $T^{D-4}$ compactified extra dimensions can have a smooth horizon in $D \ge 6$. However, since the methods used in previous works are restricted to the axi-symmetric case [@Candlish:2007fh] or coplanar case [@Gowdigere:2014aca], we need to develop a new tool to investigate the smoothness of KK black holes with less symmetry.
In this paper, as an approach to this issue, we propose a simple method to prove non-smoothness of a black hole horizon which applies to less-symmetric cases. Our claim is that there is no $C^{N + 2}$ extension across the horizon if some components of $N$-th covariant derivative of Riemann tensor diverge at the horizon in the coordinates of a $C^1$ extension across the horizon. Furthermore, the divergence of a component of the Riemann tensor means that a curvature singularity appears on the horizon. Using this method, we can identify a curvature singularity even when the scalar invariants constructed from the Riemann tensor, [*e.g.*]{}, the Ricci scalar and the Kretschmann invariant, are finite at the horizon. As an application of this method, we show that the Kaluza-Klein black holes constructed by Myers have a curvature singularity at the horizon in $D \ge 6$.
This paper is organized as follows. In Sec. II, we develop a method to prove non-smoothness of horizon. In Sec. III, we apply our method to a spherically symmetric toy model and show our method works well. In Sec. IV, we discuss the case of multi black holes and show our method can reproduce the result of previous works. We discuss the case of Kaluza-Klein black holes and show that there exists curvature singularity on the horizon if $D \ge 6$ in Sec. V. Sec. VI is devoted to the summary and discussion. We use the units in which $c = G = 1$.
method to prove non-smoothness of horizon {#formalism}
=========================================
Black hole solutions are usually constructed in a single coordinate system which does not cover the event horizon. If we want to discuss the global structure of such solutions, we need to find an extension across coordinate boundaries, such as an event horizon. If the spacetime admits an analytic extension, we can find a unique and natural extension of the original spacetime. However, as shown in the previous works [@Gibbons:1994vm; @Welch:1995dh; @Candlish:2007fh; @Candlish:2009vy; @Gowdigere:2014aca; @Chrusciel:1992tj], some black hole spacetimes are not smooth at the horizon, but in general it is not easy to prove the non-smoothness of the horizon. In this section, we develop a method which can be applied to prove non-smoothness of black hole horizon.
Let $({\cal M},g_{\mu \nu})$ be a $D$-dimensional $C^\infty$ manifold with a $C^\infty$ metric tensor with Lorentzian signature and $({\cal M^\prime}, g_{\mu \nu}^\prime)$ be a $D$-dimensional extension of $({\cal M},g_{\mu \nu})$ with an isometric imbedding $\mu^{\prime} : {\cal M} \to {\cal M^\prime}$. Here ${\cal M^\prime}$ is a $C^2$ manifold [^8] and the metric $g_{\mu \nu}^\prime$ is $C^1$ on this manifold.[^9] We assume that the boundary of $\mu^{\prime}({\cal M})$ contains a smooth hypersurface ${\cal H^\prime}$ in ${\cal M^\prime}$.[^10] (see Fig.\[fig1:extension\])
![ $C^1$ extension $({\cal M^\prime}, g_{\mu \nu}^\prime)$ of $({\cal M},g_{\mu \nu})$. []{data-label="fig1:extension"}](fig1_extension.eps){width="0.4\linewidth"}
At first, we define [*an extension of $({\cal M}, g_{\mu \nu})$ across the “same” boundary*]{} for later convenience. Let $({\cal \bar{M}}, \bar{g}_{\mu \nu})$ be an another $D$-dimensional $C^1$ extension of $({\cal M},g_{\mu \nu})$ with an isometric imbedding $\bar{\mu} : {\cal M} \to {\cal \bar{M}}$. We assume the boundary of $\bar{\mu}({\cal M})$ also contains a smooth hypersurface $\bar{\cal H}$ in $\bar{\cal M}$. Let $\gamma(\lambda) : (\lambda_i, \lambda_f) \to {\cal M}$ be an incomplete geodesic with affine parameter $\lambda$ in ${\cal M}$ such that $\gamma(\lambda \to \lambda_f) \notin {\cal M}$ and $\mu^\prime(\gamma(\lambda))$ reaches a point in ${\cal H}^\prime$, [*i.e.*]{}, there exists a limit point $p^\prime = \mu^\prime(\gamma(\lambda \to \lambda_f)) \in {\cal H}^\prime$. If $\bar{\mu}(\gamma(\lambda))$ also reaches a point in ${\cal \bar{H}}$, [*i.e.*]{}, if there exists a limit point $\bar{p} = \bar{\mu}(\gamma(\lambda \to \lambda_f)) \in \bar{\cal H}$, for all such $\gamma(\lambda)$, $({\cal M^{\prime}}, g_{\mu \nu}^{\prime})$ and $({\cal \bar{M}}, \bar{g}_{\mu \nu})$ are called the extensions of $({\cal M}, g_{\mu \nu})$ across the “[*same*]{}” boundary.\
In this setup we would like to prove the following theorem.\
[*Theorem.1*]{} [*Let $({\cal M},g_{\mu \nu})$ be a $D$-dimensional $C^\infty$ manifold with a $C^\infty$ metric tensor with Lorentzian signature and $({\cal M^\prime}, g_{\mu \nu}^\prime)$ be a $D$-dimensional extension of $({\cal M},g_{\mu \nu})$ with an isometric imbedding $\mu^{\prime} : {\cal M} \to {\cal M^\prime}$. We assume that ${\cal M^\prime}$ is a $C^2$ manifold, the metric $g_{\mu \nu}^\prime$ is $C^1$ on this manifold and the boundary of $\mu^{\prime}({\cal M})$ contains a smooth hypersurface ${\cal H^\prime}$ in ${\cal M^\prime}$. If at least one component of the Riemann tensor diverges at a point on ${\cal H}^\prime$ independently of the approaching direction from $\mu^\prime({\cal M})$ in the coordinates of the $C^1$ extension $({\cal M^\prime}, g_{\mu \nu}^\prime)$, there is no $C^2$ extension of $({\cal M}, g_{\mu \nu})$ across the “same” boundary.*]{}\
We divide the proof of this theorem into two steps, lemmas. 1 and 2. From these lemmas, we can immediately prove the above theorem. Roughly speaking, the lemma. 1 implies that if there exists a parallelly propagated (p.p.) curvature singularity at the boundary of $({\cal M},g_{\mu \nu})$ along a time like geodesic, then there is no $C^2$ extension in which the geodesic is extendible across the boundary. The lemma. 2 claims the existence of p.p. curvature singularity at the boundary of $\mu^\prime({\cal M})$ along a time like geodesic under the same assumption as theorem. 1.\
[*Lemma.1*]{} [ *Let $({\cal M},g_{\mu \nu})$ be a $D$-dimensional $C^\infty$ manifold with a $C^\infty$ metric tensor with Lorentzian signature. If $({\cal M},g_{\mu \nu})$ has a p.p. curvature singularity at the boundary of $({\cal M},g_{\mu \nu})$ along a time like geodesic with finite affine parameter, [*i.e.*]{}, there exists an incomplete time like geodesic $\gamma(\lambda):(\lambda_{i}, \lambda_{f}) \to {\cal M}$ with an affine parameter $\lambda$ in ${\cal M}$ such that $\gamma(\lambda \to \lambda_f) \notin {\cal M}$ and at least one component of the Riemann tensor measured by a parallelly propagated frame along $\gamma(\lambda)$ diverges in the limit $\lambda \to \lambda_{f}$, then there is no $D$-dimensional $C^2$ extension of $({\cal M}, g_{\mu \nu})$ such that the boundary of the isometric imbedding map of ${\cal M}$ contains a smooth hypersurface and the map of $\gamma(\lambda)$ has a limit point on the hypersurface in the limit $\lambda \to \lambda_{f}$.*]{}\
[*Proof.*]{} Suppose that there also existed a $D$-dimensional extension $({\cal M^{\prime \prime}}, g_{\mu \nu}^{\prime \prime})$ of $({\cal M}, g_{\mu \nu})$ with an isometric imbedding $\mu^{\prime \prime} : {\cal M} \to {\cal M^{\prime \prime}}$ such that ${\cal M^{\prime \prime}}$ is a $C^2$ manifold and the metric $g_{\mu \nu}^{\prime \prime}$ is $C^2$ on ${\cal M^{\prime \prime}}$, and the time like geodesic $\gamma^{\prime \prime}(\lambda) := \mu^{\prime \prime}(\gamma(\lambda))$ reaches a point on the smooth hypersurface ${\cal H^{\prime \prime}}$ which is contained in the boundary of $\mu^{\prime \prime}({\cal M})$. Let $p^{\prime \prime}=\gamma^{\prime \prime}(\lambda \to \lambda_f)
\in \partial(\mu^{\prime \prime}({\cal M}))$ be the point on ${\cal H^{\prime \prime}}$. Using local coordinates $\{y^\mu \}$ around $p^{\prime \prime}$, we denote the geodesic $\gamma^{\prime \prime}(\lambda)$ in ${\cal M^{\prime \prime}}$ by $y^\mu(\lambda)$. The geodesic equation for $y^\mu(\lambda)$ becomes $$\begin{aligned}
\frac{d^2y^\mu}{d\lambda^2} &=&
-\Gamma^\mu{}_{\alpha \beta} \frac{dy^\alpha}{d\lambda}\frac{dy^\beta}{d\lambda},
\label{geoeqc2extension}\end{aligned}$$ with the Christoffel symbols $\Gamma^\mu{}_{\alpha \beta}$. We can show that the tangent $dy^\mu/d\lambda$ does not diverge at the point $p^{\prime \prime}$ if $\gamma^{\prime \prime}(\lambda)$ is a time like geodesic as discussed in Appendix \[appendixA\]. Thus, we can uniquely extend the geodesic across ${\cal H^{\prime \prime}}$.
Let $e_{(\alpha)}^\mu$ denote the linearly independent parallelly propagated vectors along $\gamma^{\prime \prime}(\lambda)$, where the index $(\alpha)$ distinguishes different vectors, and we assume $e_{(0)}^\mu$ is tangent to $\gamma^{\prime \prime}(\lambda)$. Then, the components of $e_{(\alpha)}^\mu$ measured by the coordinate basis of $\{y^\mu \}$ take finite value at $p^{\prime \prime}$ as shown in Appendix \[appendixB\].
On the other hand, we have a relation $$\begin{aligned}
R_{(\alpha) (\beta) (\gamma) (\delta)} =
e_{(\alpha)}^\mu e_{(\beta)}^\nu e_{(\gamma)}^\rho e_{(\delta)}^\sigma
R_{\mu \nu \rho \sigma}^{(y^\mu)},
\label{tetradcomponent0}\end{aligned}$$ where $R_{(\alpha) (\beta) (\gamma) (\delta)}$ and $R_{\mu \nu \rho \sigma}^{(y^\mu)}$ are the components of the Riemann tensor in the basis of $e_{(\alpha)}^\mu$ and in the coordinate basis of $\{y^\mu\}$, respectively. From the assumption, at least one component of $R_{(\alpha) (\beta) (\gamma) (\delta)}$ diverges at ${\cal H^{\prime \prime}}$, while $R_{\alpha \beta \gamma \delta}^{(y^\mu)}$ remains finite there. Thus at least one component of $e_{(\alpha)}^\mu$ must diverge at ${\cal H^{\prime \prime}}$. However, this contradicts the result established in the preceding paragraph that $e_{(\alpha)}^\mu$ are finite. $\Box$\
[*Lemma.2*]{} [*Let $({\cal M},g_{\mu \nu})$ be a $D$-dimensional $C^\infty$ manifold with a $C^\infty$ metric tensor with Lorentzian signature and $({\cal M^\prime}, g_{\mu \nu}^\prime)$ be a $D$-dimensional extension of $({\cal M},g_{\mu \nu})$ with an isometric imbedding $\mu^{\prime} : {\cal M} \to {\cal M^\prime}$. We assume that ${\cal M^\prime}$ is a $C^2$ manifold and the metric $g_{\mu \nu}^\prime$ is $C^1$ on this manifold and the boundary of $\mu^{\prime}({\cal M})$ contains a smooth hypersurface ${\cal H^\prime}$ in ${\cal M^\prime}$. If at least one component of the Riemann tensor diverges at a point on ${\cal H}^\prime$ independently of the approaching direction from $\mu^\prime({\cal M})$ in the coordinates of the $C^1$ extension $({\cal M^\prime}, g_{\mu \nu}^\prime)$, there exists a p.p. curvature singularity at a point $p^\prime$ on ${\cal H^\prime}$ along a time like geodesic reaching from $\mu^\prime({\cal M})$ with finite affine parameter.*]{}\
[*Proof.*]{} Since the Christoffel symbols are finite, we can always move to coordinates in which the metric is apparently locally flat at the point $p^\prime$ on ${\cal H^\prime}$. Then, we can prepare linearly independent orthonormal basis vectors $v^{\mu}_{(\alpha)}$ at $p^\prime$ such that satisfy $$\begin{aligned}
v^\mu_{(\alpha)} v_{\mu(\beta)} = \eta_{(\alpha)(\beta)} = {\rm diag}[-1,1,\cdots,1],\end{aligned}$$ and we assume that $v^\mu_{(0)}$ is not parallel to ${\cal H^\prime}$. Solving the geodesic equation from ${\cal H^\prime}$ with the initial velocity $v^\mu_{(0)}$, and considering parallel transport of $v^\mu_{(\alpha)}$ along this geodesic, we obtain an orthonormal frame spanned by $e_{(\alpha)}^\mu$ along the geodesic. Without loss of generality, we can assume that $v^\mu_{(0)}$ is a past-directed time like vector such that the geodesic stays in $\mu^\prime({\cal M})$ as long as a sufficiently short geodesic is concerned.[^11] (See Appendices \[appendixC\] and \[appendixB\] for the existence of the solution of geodesic equation and the orthonormal vectors $e_{(\alpha)}^\mu$, respectively)
A component of the Riemann tensor in the vielbein frame spanned by $e_{(\alpha)}^\mu$ becomes $$\begin{aligned}
R_{(\alpha) (\beta) (\gamma) (\delta)} =
e_{(\alpha)}^\mu e_{(\beta)}^\nu e_{(\gamma)}^\rho e_{(\delta)}^\sigma
R_{\mu \nu \rho \sigma}.
\label{tetradcomponent}\end{aligned}$$ Since the vielbein frame is linearly independent, $e_{(\alpha)}^\mu$ has its inverse $e^{-1}{}_\mu^{(\alpha)}$ defined by $$\begin{aligned}
\sum_\alpha e_{(\alpha)}^\mu e^{-1}{}_\nu^{(\alpha)} &=& \delta^\mu_{\nu},\end{aligned}$$ we can rewrite Eq. (\[tetradcomponent\]) as $$\begin{aligned}
\sum_{\alpha,\beta,\gamma,\delta}
e^{-1}{}_\mu^{(\alpha)} e^{-1}{}_\nu^{(\beta)} e^{-1}{}_\rho^{(\gamma)} e^{-1}{}_\sigma^{(\delta)}
R_{(\alpha) (\beta) (\gamma) (\delta)} =
R_{\mu \nu \rho \sigma}.
\label{tetradcomponent2}\end{aligned}$$ From the assumption, at least one component of $R_{\mu \nu \rho \sigma}$ diverges at $p^\prime$. As shown in Appendix \[appendixB\], $e_{(\alpha)}^\mu$ and $e^{-1}{}_\mu^{(\alpha)}$ take finite values at $p^\prime$, and hence $R_{(\alpha) (\beta) (\gamma) (\delta)}$ must diverge at $p^\prime$ along the geodesic. $\Box$\
We can easily generalize the above theorem to be able to prove no existence of $C^{N+2}$ extension across the same boundary.\
[*Theorem.2*]{} [*Let $({\cal M},g_{\mu \nu})$ be a $D$-dimensional $C^\infty$ manifold with a $C^\infty$ metric tensor with Lorentzian signature and $({\cal M^\prime}, g_{\mu \nu}^\prime)$ be a $D$-dimensional extension of $({\cal M},g_{\mu \nu})$ with an isometric imbedding $\mu^{\prime} : {\cal M} \to {\cal M^\prime}$. We assume that ${\cal M^\prime}$ is a $C^2$ manifold, the metric $g_{\mu \nu}^\prime$ is $C^1$ on this manifold and the boundary of $\mu^{\prime}({\cal M})$ contains a smooth hypersurface ${\cal H^\prime}$ in ${\cal M^\prime}$. If at least one component of the $N$-th covariant derivative of the Riemann tensor diverges at a point on ${\cal H}^\prime$ independently of the approaching direction from $\mu^\prime({\cal M})$ in the coordinates of the $C^1$ extension $({\cal M^\prime}, g_{\mu \nu}^\prime)$, there is no $C^{N+2}$ extension of $({\cal M}, g_{\mu \nu})$ across the “same” boundary.*]{}\
[*Proof.*]{} Suppose that there also existed a $C^{N + 2}$ extension $({\cal M^{\prime \prime}}, g_{\mu \nu}^{\prime \prime})$ of $({\cal M}, g_{\mu \nu})$. First, similarly to the discussion in the proof of the lemma. 2, we can say that the divergence of the $N$-th covariant derivative of the Riemann tensor at a point on ${\cal H}^\prime$ in the coordinates of a $C^1$ extension independently of the approaching direction implies that at least one component of the $N$-th covariant derivative of the Riemann tensor measured by a parallelly propagated frame along a time like geodesic diverges at a point on ${\cal H}^\prime$. Next, similarly to the discussion in the proof of the lemma. 1, we can say that such a divergence of the $N$-th covariant derivative of the Riemann tensor measured by a parallelly propagated frame implies the divergence of the $N$-th covariant derivative of the Riemann tensor in the coordinate basis of $C^1$ extension. However this contradicts our assumption. $\Box$\
Application I : spherically-symmetric toy model
===============================================
In this section, as an example, we consider a deformed Schwarzschild spacetime in four dimensions $$\begin{aligned}
ds^2 &=& -f dt^2 + f^{-1}dr^2 +
r^2 \left(1 + \frac{m^{1/2}(r-2m)^{3/2}}{r^2}\right)(d\theta^2 + \sin^2\theta d\phi^2),
\\
f &=& 1 - \frac{2m}{r},\end{aligned}$$ and we apply our method to this spherically-symmetric toy model, and show that there exists a curvature singularity at the horizon $r = 2m$.
First, to obtain $C^1$ extension across the horizon, we introduce a new coordinate $u$ as $$\begin{aligned}
dt = du - f^{-1}dr.\end{aligned}$$ Then, the metric becomes $$\begin{aligned}
ds^2 = -f du^2
+2 dudr + r^2 \left(1 + \frac{m^{1/2}(r-2m)^{3/2}}{r^2}\right)(d\theta^2 + \sin^2\theta d\phi^2).
\label{toymodelefcoord}\end{aligned}$$ We can easily check that this is a $C^1$ extension but not a $C^2$ extension across the horizon because of the existence of the factor $(r-2m)^{3/2}$.
Though the Ricci scalar $R$ and the Kretschmann invariant $R^{\mu \nu \rho \sigma}R_{\mu \nu \rho \sigma}$ take finite values at the horizon, for this metric, we can show that a component of the Riemann tensor in the coordinates of $C^1$ extension (\[toymodelefcoord\]) behaves as $$\begin{aligned}
R_{r\theta r\theta} \simeq -\frac{3}{8}\frac{\sqrt{m}}{\sqrt{r-2m}} \to -\infty.\end{aligned}$$ From the theorem. 1, the divergence of Riemann tensor in the coordinate of $C^1$ extension implies that there exists no $C^2$ extension across the horizon at $r = 2m$ and that there always exists a curvature singularity at the horizon.
In this symmetric case, we can also show that there exists the p.p. curvature singularity in the usual way. We prepare the orthogonal vielbein bases $e^{(\mu)} = e^{(\mu)}_{\alpha} dx^\alpha$ as $$\begin{aligned}
e^{(0)} &=& -dt - \frac{\sqrt{2m r}}{r -2 m} dr,
\notag\\
e^{(1)} &=& - \frac{\sqrt{2mr}}{r}dt - \frac{r}{r - 2m} dr,
\notag\\
e^{(2)} &=&
r \left(1 + \frac{m^{1/2}(r-2m)^{3/2}}{r^2}\right)^{1/2}d\theta,
\notag\\
e^{(3)} &=&
r \left(1 + \frac{m^{1/2}(r-2m)^{3/2}}{r^2}\right)^{1/2} \sin \theta d\phi,\end{aligned}$$ such that $e^{(\mu)}_\alpha$ satisfy $g^{\alpha \beta} e^{(\mu)}_\alpha e^{(\nu)}_\beta
=\eta^{\mu \nu} = {\rm diag}[-1,1,1,1]$ and $ g^{\alpha \beta} e^{(0)}_{\alpha} \nabla_\beta e^{(0)}_{\gamma} = 0$. Then we obtain $$\begin{aligned}
e_{\alpha}^{(1)}
e_{\beta}^{(2)}
e_{\gamma}^{(1)}
e_{\delta}^{(2)}
R^{\alpha \beta \gamma \delta} \simeq - \frac{3}{32 m^{3/2}\sqrt{r-2m}} \to - \infty.\end{aligned}$$ We can see that the p.p. curvature singularity exists at the horizon $r = 2m$ and the behavior of the Riemann tensor in the coordinates of the $C^1$ extension is basically the same as the vielbein component of the Riemann tensor for a free fall observer.
Application II : case of multi black holes
==========================================
In this section we apply our method to the case of multi black hole solutions to reproduce the results of previous works [@Welch:1995dh; @Candlish:2007fh].
construction of multi black holes
---------------------------------
We consider $D$-dimensional Einstein-Maxwell system described by the action $$\begin{aligned}
S = \frac{1}{16\pi G_D}\int d^Dx \sqrt{-g}(R - F_{\mu \nu} F^{\mu \nu}),\end{aligned}$$ where $R$ is the Ricci scalar, $F_{\mu \nu} = \partial_\mu A_\nu - \partial_\nu A_\mu$ is the Maxwell field strength, and $G_D$ is $D$ dimensional gravitational constant. From this action we obtain the Einstein equations and the Maxwell equations as $$\begin{aligned}
R_{\mu \nu} - \frac{1}{2}R g_{\mu \nu}
&=& 2\left(
F_{\mu \lambda}F_{\nu}{}^{\lambda}
-\frac{1}{4}g_{\mu \nu}F_{\rho \sigma}F^{\rho \sigma}
\right),
\label{einsteineq}
\\
\nabla_\nu F^{\mu \nu} &=& 0.
\label{maxwelleq}\end{aligned}$$ In this paper, as a solution of Eqs (\[einsteineq\]) and (\[maxwelleq\]), we focus on $D$-dimensional Majumdar-Papapetrou solution [@Majumdar:1947eu; @Papaetrou:1947ib; @Myers:1986rx], whose metric and gauge 1-form are given by $$\begin{aligned}
ds^2 &=& -H^{-2} dt^2 + H^{2/(D-3)} \sum_{i,j=1}^{D-1} \delta_{ij}dx^i dx^j,
\label{MPsol}
\\
A_{\mu}dx^\mu &=& \pm \sqrt{\frac{D-2}{2(D-3)}} H^{-1} dt,
\label{MPsolgaugepotential}
\\
H &=& 1 + \sum_{n} \frac{m_n}{|\bm{x} - \bm{a}_n|^{D-3}},\end{aligned}$$ where $\bm{x}$ and $\bm{a}_n$ denote the position vector and the location of the horizon of the $n$-th black hole in $D-1$ dimensional Euclid space, respectively. At $\bm{x} = \bm{a}_n$ the lapse $g_{tt}$ vanishes. The mass parameter of the $n$-th black hole is denoted by $m_n$.
In this section, for simplicity, we focus on the case of two black holes. The metric becomes $$\begin{aligned}
ds^2 &=& -H^{-2} dt^2 + H^{2/(D-3)} (dr^2 + r^2 d\theta^2 + r^2 \sin^2\theta d\Omega_{S^{D-3}}^2),
\label{twobhmetric1}
\\
H &=& 1 + \frac{m_1}{r^{D-3}}
+
\frac{m_2}{(r^2 - 2 ar \cos{\theta} + a^2)^{(D-3)/2}},\end{aligned}$$ where $d\Omega_{S^{D-3}}^2$ is the metric of an unit $D-3$ sphere and $a$ specifies the separation between the two black holes. We set the horizon of one black hole to the origin of Euclidean space.
$C^1$ extension across the horizon
----------------------------------
At the horizon $r=0$, the metric component $g_{rr}$ in Eq. (\[twobhmetric1\]) diverges. To remove this divergence, we first introduce the Eddington-Finkelstein coordinate $u$ as $$\begin{aligned}
dt &=:& du - H^{(D-2)/(D-3)}dr + Y d\theta,
\label{EFcoordtwobh1}
\\
Y &=& - \int dr \partial_\theta H^{(D-2)/(D-3)}.\end{aligned}$$ The last term in Eq. (\[EFcoordtwobh1\]) is needed to satisfy integrability condition. Near the horizon, the metric behaves as $$\begin{aligned}
ds^2 &\simeq &
2 m_1^{-(D-4)/(D-3)} r^{(D-4)} dr du
+
m_1^{2/(D-3)} (d\theta^2 + {\sin}^2 \theta d\Omega_{S^{D-3}}^2).
\label{EFcoordtwobh2}\end{aligned}$$ If $D > 4$, the metric in this coordinate degenerates at the horizon since $g_{ru}$ vanishes at $r = 0$. We can remove this coordinate singularity by introducing new radial coordinate $\rho$ as $$\begin{aligned}
\rho = r^{D-3},\end{aligned}$$ However, since the function $H$ becomes $$\begin{aligned}
H &=& 1 + \frac{m_1}{\rho}
+
\frac{m_2}{(\rho^{2/(D-3)} - 2 a\rho^{1/(D-3)} \cos{\theta} + a^2 )^{(D-3)/2}},\end{aligned}$$ we find that the last term in the denominator contains fractional powers of $\rho$. Thus this extension is not an analytic extension across the horizon. Even worse, since the metric behaves as $$\begin{aligned}
ds^2 &=&
2 m_1^{-(D-4)/(D-3)} (D-3)^{-1}
du d\rho + 2F d\rho d\theta
\notag\\ &&
+m_1^{2/(D-3)} (d\theta^2 + {\sin}^2 \theta d\Omega_{S^{D-3}}^2)
+
{\cal O}(\rho),
\\
F &=&
(D-3)^{-1} m_1^{-(D-4)/(D-3)} Y
\notag\\&= &
(D-3)^{-1}(D-2) a^{-(D-2)} m_1^{-(D-5)/(D-3)}
m_2
\sin \theta \left(
\rho^{1/(D-3)}
+
{\cal O}(\rho^{2/(D-3)})
\right),\end{aligned}$$ the first derivative of $g_{\rho \theta}$ diverges at the horizon because of the fractional power of $\rho$ in the form of $F$. Then, this is only $C^0$ extension. To obtain a $C^1$ extension, we finally introduce new coordinates $\bar{u}$ and $\bar{\theta}$ as $$\begin{aligned}
d\bar{\theta} &:=& d\theta + \frac{1}{m_1^{2/(D-3)}} F d\rho
+\frac{1}{m_1^{2/(D-3)}}d\theta \int d\rho \partial_\theta F,
\label{thetabartwobh}
\\
d\bar{u} &:=& m_1^{-(D-4)/(D-3)} (D-3)^{-1}du - \frac{1}{m_1^{2/(D-3)}} \frac{F^2}{2} d\rho
- \frac{1}{m_1^{2/(D-3)}} d\theta \int d\rho \partial_\theta \frac{F^2}{2}.
\label{ubartwobh}\end{aligned}$$ Notice that the leading terms of the last terms in the Eqs. (\[thetabartwobh\]) and (\[ubartwobh\]) have positive powers of $\rho$ higher than unity because their integrands are proportional to positive fractional powers of $\rho$ near $\rho = 0$. Then the metric behaves as $$\begin{aligned}
ds^2 =
2 d\bar{u} d\rho
+ m_1^{2/(D-3)}(d\bar{\theta}^2 + {\sin}^2 \bar{\theta} d\Omega_{S^{D-3}}^2)
+{\cal O }(\rho^1).\end{aligned}$$ We can see that the derivatives of all the metric component w.r.t. $\rho$ take finite values at the horizon, namely, this is a $C^1$ extension across the horizon.
Divergence of Riemann tensors in the coordinate of $C^1$ extension
------------------------------------------------------------------
We should comment that two coordinate bases $(du, d\rho, d\theta)$ and $(d\bar{u}, d\rho, d\bar{\theta})$ are linearly related with non-degenerate finite coefficients at $\rho = 0$. Thus, if a component of a tensor diverges at $\rho = 0$ in the coordinates $(du, d\rho, d\theta)$, at least one component of the tensor diverges at $\rho = 0$ also in the coordinates $(d\bar{u}, d\rho, d\bar{\theta})$. For this reason, it is sufficient to confirm the divergence of the Riemann tensor in the coordinates $(du, d\rho, d\theta)$ in order to conclude its divergence in the coordinates of the $C^1$ extension $(d\bar{u}, d\rho, d\bar{\theta})$.
In five dimensional case, we can easily verify that all the components of the Riemann tensor in the coordinates $(du, d\rho, d\theta)$ take finite values at the horizon. We can also show a component of the first covariant derivatives behaves as $$\begin{aligned}
\nabla_{\rho} R_{\rho u \theta \rho}
\propto \frac{1}{\rho^{1/2}}
\to \infty.\end{aligned}$$ Thus, we conclude that there is no $C^3$ extension in $D=5$ from the theorem. 2. As for $C^2$ extension, we can construct such an extension by using the Riemann normal coordinates.[^12]
If the dimension is higher than five, the Riemann tensor behaves as $$\begin{aligned}
R_{\rho \theta \rho \theta} \propto \frac{1}{\rho^{(D-5)/(D-3)}}
\to \infty.\end{aligned}$$ Thus, we conclude that there is no $C^2$ extension across the horizon from the theorem. 1. and that we cannot remove the curvature singularity on the horizon by considering any extension across the horizon.
Application III : case of Kaluza-Klein black holes
==================================================
In the case of $D \ge 5$, if we superpose black holes with the same mass $m$ periodically, we can obtain a toy model of a Kaluza-Klein black hole with $T^N$ compactified extra dimensions as constructed by Myers [@Myers:1986rx] by using multi black hole solutions Eqs. (\[MPsol\]) and (\[MPsolgaugepotential\]).
In this section we apply our method to this Kaluza-Klein black hole solution. One might think that the curvature singularity should exist at the horizon since it exists even in the two black hole case. However, whether this expectation is correct or not is not so obvious. Infinite superposition of black holes may have qualitatively different feature of spacetime than the case of finite number superposition.
In fact, the previous works show that a five dimensional Kaluza-Klein black hole admits analytic extension across the horizon [@Candlish:2007fh] in contrast with the case of two black holes where the horizon is not $C^3$ but $C^2$. The four dimensional multi black hole solutions also admit analytic extension across the horizon. Hence, we can also expect that the Kaluza-Klein black hole admits smooth extension across the horizon when the number of non-compact dimensions is four. In general, such a Kaluza-Klein black hole spacetime becomes less-symmetric. While the method used in previous works [@Candlish:2007fh; @Gowdigere:2014aca] can be applied only to axi-symmetric or plane symmetric case, our method is applicable to this case.
In this section, we clarify whether the Kaluza-Klein black hole admits smoother extension compared to the case of two black hole when the number of non-compact dimensions is four.
metric form of Kaluza-Klein black holes
---------------------------------------
The explicit form of the metric of Kaluza-Klein black hole is given by $$\begin{aligned}
ds^2 &=& -H^{-2}dt^2
+H^{2/(D-3)} \left[
\sum_{I,J=1}^{N}\delta_{IJ}dx^Idx^J
+
\sum_{A,B=N+1}^{D-1}\delta_{AB}dx^Adx^B
\right],
\label{metkkbh}
\\
H &=& 1 + \sum_{ n_1,n_2,\cdots,n_N =-\infty}^\infty \frac{m}{
\left[
\displaystyle\sum_{I,J = 1}^{N}\delta_{IJ}(x^I - n_I \ell^{(I)})(x^J - n_J \ell^{(J)})
+
\displaystyle\sum_{A,B = N+1}^{D-1} \delta_{AB}x^A x^B
\right]^{(D-3)/2}},
\notag\end{aligned}$$ where $\ell^{(I)}$ denotes the size of extra dimension in the $x^I$ direction, and the number of the compactified extra dimensions $N$ satisfies $1 \le N \le D-4$. (see Fig.\[fig:t2extradim\] in the case of $N=2$ for example).
![ Periodic identification in the case of $T^2$ compactified extra dimensions. []{data-label="fig:t2extradim"}](extradim.eps){width="0.4\linewidth"}
$C^1$ extension across the horizon
----------------------------------
We can place the black hole at the origin of Euclidean coordinates without loss of generality. Let us introduce polar coordinates as $$\begin{aligned}
x^i &=& r
\left[\prod_{j=1}^{i-1} \sin \theta^j \right] \cos \theta^i
~(=: r \Theta^i )~~(1 \le i \le D-2),
\\
x^{D-1} &=& r
\left[\prod_{j=1}^{D-2} \sin \theta^j \right]
(=: r \Theta^{D-1} ).\end{aligned}$$ Then, the metric (\[metkkbh\]) becomes $$\begin{aligned}
ds^2 &=& -H^{-2} dt^2 + H^{2/(D-3)}(dr^2 + r^2 d\Omega_{{\rm S}^{D-2}}^2),
\\
d\Omega_{{\rm S}^{D-2}}^2 &=& \sum_{i=1}^{D-2} \left[
\prod_{j=1}^{i-1} \sin^2 \theta^j
\right]
(d\theta^{i})^2,
\\
H &=& 1 + \frac{m}{r^{D-3}}
+ m \sum_{n_1,n_2,\cdots,n_N \neq \{0,0,\cdots,0\}} (H_{n_1,n_2,\cdots,n_N})^{-(D-3)/2},
\\
H_{n_1,n_2,\cdots,n_N} &=&
\sum_{I,J = 1}^{N}\delta_{IJ}(r \Theta^I - n_I \ell^{(I)})(r \Theta^J - n_J \ell^{(J)})
+
\sum_{A,B = N+1}^{D-1} \delta_{AB}r^2 \Theta^A \Theta^B.\end{aligned}$$ In these coordinates, the metric component $g_{rr}$ diverges at the horizon $r=0$. To remove this divergence, we introduce a coordinate $u$ as $$\begin{aligned}
dt
&=&
du
-
H^{(D-2)/(D-3)} dr
+
\sum_{i = 1}^{D-2}Y_i d\theta^i,
\label{EFcoord1}
\\
Y_i &:=& -\int dr \frac{\partial H^{(D-2)/(D-3)}}{\partial \theta^i}.\end{aligned}$$ Note that the last term in Eq. (\[EFcoord1\]) is needed to satisfy the integrability condition. Near $r = 0$, we can show that $$\begin{aligned}
Y_i &\simeq & W_i(\theta) r^2,
\label{eqyi}\end{aligned}$$ where $W_i(\theta)$ is a finite function of the angular coordinates. To derive Eq. (\[eqyi\]), we used the condition that the function $\sum (H_{n_1,n_2,\cdots,n_N})^{-(D-3)/2} $ is a function of $r^2$ and angular coordinates $$\begin{aligned}
\sum_{n_1,n_2,\cdots,n_N \neq \{0,0,\cdots,0\}} (H_{n_1,n_2,\cdots,n_N})^{-(D-3)/2} = C_0 + C_2(\theta) r^2 + C_4(\theta) r^4 + \cdots\end{aligned}$$ since the function $\sum (H_{n_1,n_2,\cdots,n_N})^{-(D-3)/2} $ has formally $r \leftrightarrow -r$ symmetry. Then, the metric becomes $$\begin{aligned}
ds^2 &=&
-H^{-2} du^2
-
H^{-2}\sum_{i,j}^{D-2}Y_i Y_j d\theta^i d\theta^j
+
2 H^{-(D-4)/(D-3)} du dr
-
2 H^{-2} \sum_{i}^{D-2}Y_i du d\theta^i
\notag\\ & &
+
2
H^{-(D-4)/(D-3)} \sum_{i}^{D-2} Y_i dr d\theta^i
+
H^{2/(D-3)}
r^2 d\Omega^2_{S^{D-2}}.\end{aligned}$$ However, in these coordinates, all the metric components except for the coefficient of $d\Omega^2$ become zero at the horizon because the functions $H, Y_i$ behaves as $$\begin{aligned}
H & \sim & r^{-(D-3)},
\\
Y_i &\sim & r^{2},\end{aligned}$$ near $r=0$. Thus, the metric in these coordinates degenerates at the horizon $r=0$. We can remove this coordinate singularity by further introducing a new radial coordinate $\rho$ as $$\begin{aligned}
\rho = r^{D-3}.\end{aligned}$$ Then, the metric behaves as $$\begin{aligned}
ds^2 &=&
2(D-3)^{-1} m^{-(D-4)/(D-3)}
du d\rho
+
2 \sum_{i}^{D-2} F_i d\rho d\theta^i
+
m^{2/(D-3)} d\Omega^2_{S^{D-2}} + {\cal O}(\rho),
\label{rthetacomp}
\\
F_i &=&
(D-3)^{-1} m^{-(D-4)/(D-3)} Y_i
\notag\\&=&
(D-3)^{-1} m^{-(D-4)/(D-3)} \left(
W_i(\theta) \rho^{2/(D-3)} + {\cal O}(\rho^{4/(D-3)})
\right).\end{aligned}$$ From Eq. (\[rthetacomp\]), we can see the function $\partial_\rho g_{\rho \theta^i}$ diverges at the horizon, thus this extension is only a $C^0$ extension, not being a $C^1$ extension, if $D \ge 6$. To get a $C^1$ extension, we introduce new coordinates $\bar{u}$ and $\bar{\theta}^i$ as $$\begin{aligned}
d \bar{\theta}^i &:= &
d\theta^i +
\frac{1}{m^{2/(D-3)}}
{\cal G}^i d\rho +
\frac{1}{m^{2/(D-3)}}
\sum_{j} d\theta^j \int d\rho \partial_{\theta^j}{\cal G}^i,
\label{thetabar}
\\
d\bar{u} &:= &
(D-3)^{-1} m^{-(D-4)/(D-3)} du
-
\frac{1}{m^{2/(D-3)}}
\sum_{i=1}^{D-2}
\frac{\left({\cal G}^i \right)^2}{2} \prod_{j=1}^{i-1} \sin^2 \theta^j d\rho
\notag\\&&-
\frac{1}{m^{2/(D-3)}}
\sum_{i,j=1}^{D-2} d\theta^j \left(\int d\rho \partial_{\theta^j}
\left[
\frac{\left({\cal G}^i \right)^2}{2} \prod_{j=1}^{i-1} \sin^2 \theta^j
\right] \right),
\label{ubar}
\\
{\cal G}^i &:=& F_i
\left[
\prod_{j=1}^{i-1} \sin^2 \theta^j
\right]^{-1}.\end{aligned}$$ The dominant parts of the last terms in Eqs. (\[thetabar\]) and (\[ubar\]) have positive powers of $\rho$ higher than unity because their integrands are proportional to positive fractional powers of $\rho$ near $\rho = 0$. Then, the metric behaves as $$\begin{aligned}
ds^2 =
2 d\bar{u} d\rho
+ m^{2/(D-3)}(d\bar{\theta}^2 + {\sin}^2 \bar{\theta} d\Omega_{S^{D-3}}^2)
+{\cal O }(\rho^1).\end{aligned}$$ As we find that the derivatives of all the metric components w.r.t. $\rho$ take finite values at the horizon, this is a $C^1$ extension across the horizon.
Divergence of Riemann tensor in the coordinates of the $C^1$ extension
----------------------------------------------------------------------
We consider following two 1-forms $$\begin{aligned}
e^{(0)} &=& -dt - \sqrt{-1 + H^2} H^{1/(D-3)} dr,
\\
e^{(1)} &=& d\theta^1.\end{aligned}$$ If we express these 1-forms by using $(d\bar{u}, d\rho, d\bar{\theta}^i)$, they become $$\begin{aligned}
e^{(0)} &=& -(D-3)m^{(D-4)/(D-3)}d\bar{u}
+
\frac{1}{2}m^{-(D-4)/(D-3)}(D-3)^{-1}d\rho
+
{\cal O}(\rho^{2/(D-3)}),
\\
e^{(1)} &=& d\bar{\theta}^1 + {\cal O}(\rho^{2/(D-3)}).\end{aligned}$$ Thus, if a component of a tensor projected to $e^{(0)}$ and $e^{(1)}$ diverges, we can say that some component of such a tensor also diverges at $\rho = 0$ in the coordinate $(d\bar{u}, d\rho, d\bar{\theta}^i)$. For this reason, it is sufficient to confirm the divergence of the Riemann tensor projected to $e^{(0)}$ and $e^{(1)}$. After some calculations, for $D >5$, we obtain $$\begin{aligned}
e^{(0)}_{\mu}
e^{(1)}_{\nu}
e^{(0)}_{\rho}
e^{(1)}_{\sigma}
R^{\mu \nu \rho \sigma} &\propto & \frac{1}{\rho^{(D-5)/(D-3)}}
\to \infty.\end{aligned}$$ Thus, we conclude that there is no $C^2$ extension across the horizon from the theorem. 1. By contrast, for $D = 5$, the horizon becomes analytic as shown in Ref. [@Candlish:2007fh].
summary and discussion
======================
In this paper, we have proposed a simple method to prove the non-smoothness of the horizon and applied it to several black hole spacetimes for which the scalar invariants constructed from the Riemann tensor, [*e.g.*]{}, the Ricci scalar and the Kretschmann invariant, take finite values at the horizon. In Secs. III and IV, we have shown that our method works well for a toy model and the multi black holes, reproducing the results in Refs. [@Welch:1995dh; @Candlish:2007fh]. We have shown that the Kaluza-Klein black holes constructed by Myers have a curvature singularity at the horizon if $D \ge 6$ in Sec. V.
Though one may think that the existence of the curvature singularities immediately means the breakdown of the classical theory, in fact, it depends on the strength of the curvature singularities. Using our method, one can also discuss the strength of the curvature singularities. In the case of Kaluza-Klein black holes, the Riemann tensor diverges as $\rho^{-(D-5)/(D-3)}$ where $\rho$ is approximately the proper length from the horizon. In this case, since the singularity is relatively mild, [*i.e.*]{}, the second integral of the Riemann tensor is finite, the tidal force on a finite-sized body is not divergent across the horizon.
One of the advantage of our method is that it applies to less-symmetric spacetimes, but it has a merit even in the case of symmetric spacetimes. Even if it is shown that there exists no smooth extension across the horizon which maintains the spacetime symmetry, there is a possibility that we may find a smoother extension by considering an extension which breaks the symmetry of the spacetime, like in the case of AdS Poincaré horizon where we need to introduce a new coordinate system across the horizon which does not have the same Killing coordinate of the Poincaré chart. Our method can be used to prove that such a possibility is excluded.
We comment on the restriction that we have only focused on $C^2$ extensions across the “[*same*]{}” boundary in this paper. First, we should emphasize that our method is a natural extension of the previous works [@Candlish:2007fh; @Candlish:2009vy; @Gowdigere:2014aca]. Since the discussion in the previous works is based on the explicit construction of extensions across the horizon by using the same coordinate system for the outside and on the horizon, what were discussed are in fact extensions across the “[*same*]{}” boundary. Secondly, from the definition of the extensions across the “[*same*]{}” boundary, if there exists any other $C^2$ extension across the boundary, some geodesic cannot reach the boundary[^13] and is inextensible while the affine parameter is finite there. This implies that there exists a singularity in the same sence as used in the singularity theorem [@Hawking:1973uf]. One may think that the divergence of the Riemann tensor only in a $C^1$ extension does not have a covariant meaning. However, even in that case, we can say that there exists some singularity at least in any extension from our method.
As far as we know, there is no discussion on the connection between the existence of p.p. curvature singularity and the no existence of the $C^2$ extension for the case of Lorentzian signature in literature. In lemma. 1, we have discussed it when there exists a p.p. curvature singularity along a time like geodesic.
Finally, we should note that we need a $C^1$ extension across the boundary to give a criteria for non-existence of $C^{N+2}$ extension in our theorem. There are possible cases that spacetimes do not admit $C^1$ extension across the boundary. It is also important to study the existence and construction of $C^1$ extension for general spacetime. We leave this problem for future work.
While this paper was being prepared for submission, an interesting paper [@Gowdigere:2014cqa] appeared, in which the smoothness of horizons in the most generic multi center black hole and membrane solutions were discussed.\
\
Acknowledgements {#acknowledgements .unnumbered}
================
The authors thank P. Chrusciel, T. Houri, H. Kodama, K. -i. Nakao, H. Reall, R. Saito, T. Shiromizu, K. Tanabe, N. Tanahashi, T. Tatsuoka, B. Way, Y. Yasui and C. -M. Yoo for very helpful comments and suggestions. MK is supported by a grant for research abroad from JSPS. HI is supported by the Grant-in-Aid for Scientific Research No. 19540305 and 24540282. TT is supported by the Grant-in-Aid for Scientific Research No. 26287044, 24103006 and 24103001.
behavior of $dy^\mu/d\lambda$ near ${\cal H^{\prime \prime}}$ {#appendixA}
=============================================================
In this section, we study the behavior of the tangent vector $dy^\mu/d\lambda$ in Eq. (\[geoeqc2extension\]) in the limit to ${\cal H^{\prime \prime}}$, [*i.e.*]{}, $\lambda \to \lambda_f$. Firstly, to gain an intuitive understanding, we show that $dy^\mu/d\lambda$ cannot diverge as a power low of the affine parameter. Later, we treat the general case.
the case of power low divergence
--------------------------------
We assume the leading behavior of the most divergent component of the tangent vector $dy^\mu/d\lambda$ near $p^{\prime \prime}$ as $$\begin{aligned}
\frac{dy^\mu}{d\lambda} \sim \frac{c}{|\lambda - \lambda_f|^{m}},
\label{appendix_D_1_01}\end{aligned}$$ where $c~(\neq 0)$ and $m~(> 0)$ are constants. Since the coordinates $\{y^\mu \}$ cover the point $p^{\prime \prime}$, the values of $y^\mu$ at $p^{\prime \prime}$ should be finite. Then, the power $m$ in Eq. (\[appendix\_D\_1\_01\]) should be less than unity $$\begin{aligned}
m < 1.\end{aligned}$$ Substituting this to Eq. (\[geoeqc2extension\]), we find that the leading behaviors of the left-hand side (LHS) and right-hand side (RHS) become $$\begin{aligned}
{\rm LHS} &\sim& \frac{c}{|\lambda - \lambda_f|^{m+1}},
\\
|{\rm RHS}| & \le & |\Gamma^\mu{}_{\alpha \beta}| \frac{c^2}{|\lambda - \lambda_f|^{2m}}.\end{aligned}$$ Since $m <1$ and the Christoffel symbols take finite values, the LHS cannot be balanced with RHS in Eq. (\[geoeqc2extension\]). This contradiction shows that our assumption (\[appendix\_D\_1\_01\]) cannot be true.
the general case
----------------
Without loss of generality, we can assume that the tangent of the geodesic is future-directed.[^14] From Appendix. \[appendixD\] we can assume that all vectors normal to the constant surfaces of the coordinate functions $y^\mu$ are timelike and future directed, at least, in the vicinity of a point $p^{\prime \prime}$ on ${\cal H}^{\prime \prime}$. In this case, the tangent of a time like geodesic ending at $p^{\prime \prime}$ satisfies $$\begin{aligned}
\frac{dy^\mu}{d\lambda} > 0,\end{aligned}$$ as long as a sufficiently short geodesic is concerned. We introduce a non-affine parameter $\zeta$ for the geodesic $y^\mu(\lambda)$ increasing toward the future satisfying $$\begin{aligned}
\delta_{\mu \nu}\frac{dy^\mu}{d\zeta}\frac{dy^\nu}{d\zeta} = 1,
\label{zetaparameter}\end{aligned}$$ then, we also have $$\begin{aligned}
\frac{dy^\mu}{d\zeta} > 0.\end{aligned}$$ From these equations we can show that the value of the parameter $\zeta$ at $p^{\prime \prime}$ on ${\cal H^{\prime \prime}}$ has a definite value as $$\begin{aligned}
\zeta(p^{\prime \prime})
-
\zeta|_{\lambda = \lambda_i}
&=&
\int_{\lambda_i}^{\lambda_f} d\lambda
\sqrt{\delta_{\mu \nu}\left|\frac{dy^\mu}{d\lambda}\right|\left|\frac{dy^\nu}{d\lambda}\right|}
\notag\\&< &
\int_{\lambda_i}^{\lambda_f} d\lambda
\sum_\mu \left|\frac{dy^\mu}{d\lambda}\right|
\notag\\&= &
\int_{\lambda_i}^{\lambda_f} d\lambda
\sum_\mu \frac{dy^\mu}{d\lambda}
\notag\\&= &
\sum_\mu
(y^\mu|_{p^{\prime \prime}} - y^\mu|_{\lambda = \lambda_i})
\notag\\ &<& \infty.\end{aligned}$$ For any finite value of $\zeta|_{\lambda = \lambda_i}$, $\zeta(p^{\prime \prime})$ is finite. The geodesic equation in term of the parameter $\zeta$ becomes $$\begin{aligned}
\frac{d^2y^\mu}{d\zeta^2}
=
-
\Gamma^\mu{}_{\alpha \beta}
\frac{dy^\alpha}{d\zeta}
\frac{dy^\beta}{d\zeta}
+
\delta_{\rho \nu}
\Gamma^\rho{}_{\alpha \beta}
\frac{dy^\alpha}{d\zeta}
\frac{dy^\beta}{d\zeta}
\frac{dy^\nu}{d\zeta}
\frac{dy^\mu}{d\zeta}.\end{aligned}$$ Since $dy^\mu/d\zeta$ does not diverge at $p^{\prime \prime}$ from the definition of the parameter $\zeta$ in Eq. (\[zetaparameter\]), $d^2y^\mu/d\zeta^2$ also does not diverge at $p^{\prime \prime}$. For this reason, all components of $dy^\mu/d\zeta$ have definite values $w^\mu$ at $p^{\prime \prime}$.[^15]
On the other hand, when we consider the solution $\bar{y}^\mu(\bar{\lambda})$ of the equation $$\begin{aligned}
\frac{d}{d\bar{\lambda}} \frac{d\bar{y}^\mu}{d\bar{\lambda}} + \Gamma^\mu{}_{\alpha \beta}
\frac{d\bar{y}^\alpha}{d\bar{\lambda}}
\frac{d\bar{y}^\beta}{d\bar{\lambda}}
=0,\end{aligned}$$ with the initial conditions $$\begin{aligned}
\frac{d\bar{y}^\mu}{d\bar{\lambda}}
=
w^\mu,\end{aligned}$$ at the point $p^{\prime \prime}$, in a similar manner, with the parameter $\bar{\zeta}$ satisfying $$\begin{aligned}
\delta_{\mu \nu}\frac{d\bar{y}^\mu}{d\bar{\zeta}}\frac{d\bar{y}^\nu}{d\bar{\zeta}} = 1,
\label{barzetaparameter}\end{aligned}$$ we can show $d\bar{y}^\mu / d\bar{\zeta} $ take definite values $\bar{w}^\mu$ at the point $p^{\prime \prime}$. From the relation $$\begin{aligned}
\frac{d\bar{y}^\mu}{d\bar{\lambda}} =
\frac{d\bar{\zeta}}{d\bar{\lambda}}
\frac{d\bar{y}^\mu}{d\bar{\zeta}},\end{aligned}$$ we have a relation $$\begin{aligned}
w^{\mu} =
\frac{d\bar{\zeta}}{d\bar{\lambda}}\bigg|_{p^{\prime \prime}}
\bar{w}^\mu.\end{aligned}$$ From the definition of $w^{\mu}$ and $\bar{w}^\mu$, we have relations $$\begin{aligned}
\delta_{\mu \nu}w^{\mu}w^{\nu} &=& 1,
\\
\delta_{\mu \nu}\bar{w}^\mu \bar{w}^\nu &=& 1.\end{aligned}$$ Thus, we conclude $d\bar{\zeta}/d\bar{\lambda}|_{p^{\prime \prime}} = 1$, and it is clear that the orbit $y^\mu = \bar{y}^\mu(\bar{\lambda}) $ is the same as $y^\mu(\lambda)$ or $y^\mu(\zeta)$.
Two affine parameters of the same orbit must be related by an affine transformation $$\begin{aligned}
\bar{\lambda} = \alpha \lambda + \beta,\end{aligned}$$ and the tangent vectors are related as $$\begin{aligned}
\frac{dy^\mu}{d\lambda}
=
\alpha
\frac{dy^\mu}{d\bar{\lambda}}
\stackrel{p^{\prime \prime}}{\to} \alpha w^\mu.\end{aligned}$$ If a component of $dy^\mu/d\lambda$ diverges at the point $p^{\prime \prime}$, the only possibility is $\alpha \to \infty$. However, in that case, it diverges everywhere on the curve. This contradicts the assumption that $dy^\mu/d\lambda$ does not diverge in $\mu^{\prime \prime}({\cal M})$.
parallel transport of vector along a geodesic in $C^1$ spacetime {#appendixB}
================================================================
In this section, let the spacetime $(M^\prime, g_{\mu \nu}^\prime)$ be a $C^2$ manifold with a $C^1$ metric tensor. We focus on a chart described by local coordinates $\{x^\mu \}$ on ${\cal M^\prime}$, and we assume that there exists a geodesic $\gamma^\prime(\lambda)$ on this chart. Denoting the geodesic $\gamma^\prime(\lambda)$ by $x^\mu(\lambda)$, we find that the function $x^\mu(\lambda)$ is twice differentiable owing to the geodesic equation. Let us consider a vector $e^\mu_0$ at a point $\gamma^\prime(\lambda_0)$ on the geodesic and its parallel transport $e^\mu$ along the geodesic. To find the parallelly transported vector field $e^\mu$ on $\gamma^\prime(\lambda)$, we should solve the differential equation $$\begin{aligned}
\frac{d e^\nu}{d\lambda}
+
\Gamma^\nu{}_{\rho \sigma}
\frac{dx^\rho}{d\lambda}
e^\sigma
= 0.\end{aligned}$$ for $e^\mu$ with the initial value $e^\mu = e^\mu_0$ at $\lambda = \lambda_0$. Since all components of the Christoffel symbol, which take finite values on ${\cal M^\prime}$, can be considered as functions of $\lambda$ along the curve $x^\mu(\lambda)$, this equation is a system of linear ordinary differential equations with finite coefficients. Thus, the solution for a given initial data is unique and finite as long as $\lambda$ is finite.
geodesics starting with ${\cal H^\prime}$ {#appendixC}
=========================================
In this section we discuss the existence of a solution of the geodesic equation emanating from a point on ${\cal H^\prime}$ in $({\cal M^\prime}, g_{\mu \nu}^\prime)$, where ${\cal M^\prime}$ is a $C^2$ manifold and the metric $g_{\mu \nu}^\prime$ is $C^1$. The geodesic equation is given by $$\begin{aligned}
\frac{dx^\mu}{d\lambda} &=& v^\mu,
\label{appendix_B_01}
\\
\frac{d v^\mu}{d\lambda} &=&
- \Gamma^\mu{}_{\alpha \beta}
v^\alpha
v^\beta.
\label{appendix_B_02}\end{aligned}$$ Since the Christoffel symbol is continuous, we can say that there exists at least one solution for any initial condition from the Peano existence theorem.
We can find a solution iteratively for an initial values $x^\mu = x^\mu_{\rm ini}, v^\mu = v^\mu_{\rm ini}$ unless all the components of $v^\mu_{\rm ini}$ vanish. Firstly, we solve the Eqs. (\[appendix\_B\_01\]) and (\[appendix\_B\_02\]) approximately as $$\begin{aligned}
x^\mu(\lambda) &\simeq & x_{\rm 1st}^\mu(\lambda)
\notag\\
&:=& x^\mu_{\rm ini} + v^\mu_{\rm ini} \lambda,
\label{appendix_B_03}
\\
v^\mu(\lambda)
& \simeq & v^\mu_{\rm 1st}(\lambda)
\notag\\
&:= &
v^\mu_{\rm ini}
- \Gamma^\mu{}_{\alpha \beta}(x^\nu_{\rm ini} )
v^\alpha_{\rm ini}
v^\beta_{\rm ini} \lambda.
\label{appendix_B_04}\end{aligned}$$ Substituting these to the right hand side of the Eqs. (\[appendix\_B\_01\]) and (\[appendix\_B\_02\]) again, we obtain the next order approximation as $$\begin{aligned}
x^\mu(\lambda) &\simeq & x_{\rm 1st}^\mu(\lambda)
+
x_{\rm 2nd}^\mu(\lambda),
\notag\\
&:=& x^\mu_{\rm ini} + v^\mu_{\rm ini} \lambda
-\frac{1}{2} \Gamma^\mu{}_{\alpha \beta}(x^\nu_{\rm ini} )
v^\mu_{\rm ini}
v^\mu_{\rm ini} \lambda^2,
\label{appendix_B_05}
\\
v^\mu(\lambda)
& \simeq & v^\mu_{\rm 1st}(\lambda)
+
v^\mu_{\rm 2nd}(\lambda)
\notag\\
&:= &
v^\mu_{\rm ini}
- \Gamma^\mu{}_{\alpha \beta}(x^\nu_{\rm ini} )
v^\mu_{\rm ini}
v^\mu_{\rm ini} \lambda
+
\int_0^\lambda d\lambda^\prime
\bigg[
2
\Gamma^\mu{}_{\alpha \beta}(x^\nu_{\rm 1st}(\lambda^\prime))
\Gamma^\beta{}_{\rho \sigma}(x^\nu_{\rm ini} )
v^\alpha_{\rm ini}
v^\rho_{\rm ini}
v^\sigma_{\rm ini} \lambda^\prime
\notag
\\&& -
\Gamma^\mu{}_{\alpha \beta}(x^\nu_{\rm 1st}(\lambda^\prime))
\Gamma^\alpha{}_{\rho \sigma}(x^\nu_{\rm ini} )
\Gamma^\beta{}_{\kappa \mu}(x^\nu_{\rm ini} )
v^\rho_{\rm ini}
v^\sigma_{\rm ini}
v^\kappa_{\rm ini}
v^\mu_{\rm ini} \lambda^{\prime 2}
\bigg].
\label{appendix_B_06}\end{aligned}$$ Repeating this process, we can obtain a solution of geodesic equation locally.
coordinate system in which all coordinates are time coordinates {#appendixD}
===============================================================
If both the manifold and metric are $C^2$, we can introduce Riemann normal coordinates around any point $p$. In these coordinates the metric becomes like $$\begin{aligned}
ds^2 = -dt^2 + \sum_{i,j}\delta_{ij}dx^idx^j
+
\frac{1}{3} R_{\mu \alpha \beta \nu} x^\alpha x^\beta dx^\mu dx^\nu
+
{\cal O}(x^3),\end{aligned}$$ where we choose the point $p$ as the origin of coordinates. If we introduce a new coordinate system $\{y^\mu \}$ as $$\begin{aligned}
y^0 &=& t,
\\
y^i &=& t + \epsilon x^i,\end{aligned}$$ where $\epsilon$ is a constant, the norm of the normal vector of this coordinate $y^i$ become $$\begin{aligned}
|dy^i|^2 &=& |dt + \epsilon dx^i |^2
\notag\\&=&
-1 + \epsilon^2 + \epsilon
\left[-\frac{1}{3}R_{0 \alpha \beta i}x^\alpha x^\beta + {\cal O}(x^3)
\right].\end{aligned}$$ Then, if we choose $\epsilon \ll 1 $, the norm of $dy^i$ becomes negative near the point $p$. Restricting the region of the coordinate system to the neighborhood in which all the norms of $dy^i$ take negative values, we can obtain a local coordinate system around any point $p$ in which all normal vectors of the coordinate functions $y^\mu$ are timelike and future directed.
[99]{}
T. Banks and W. Fischler, hep-th/9906038. S. Dimopoulos and G. L. Landsberg, Phys. Rev. Lett. [**87**]{}, 161602 (2001) \[hep-ph/0106295\]. S. B. Giddings and S. D. Thomas, Phys. Rev. D [**65**]{}, 056010 (2002) \[hep-ph/0106219\]. D. Ida, K. -y. Oda and S. C. Park, Phys. Rev. D [**67**]{}, 064025 (2003) \[Erratum-ibid. D [**69**]{}, 049901 (2004)\] \[hep-th/0212108\]. D. Ida, K. -y. Oda and S. C. Park, Phys. Rev. D [**71**]{}, 124039 (2005) \[hep-th/0503052\]. D. Ida, K. -y. Oda and S. C. Park, Phys. Rev. D [**73**]{}, 124022 (2006) \[hep-th/0602188\].
P. C. Argyres, S. Dimopoulos and J. March-Russell, Phys. Lett. B [**441**]{}, 96 (1998) \[hep-th/9808138\]. J. L. Feng and A. D. Shapere, Phys. Rev. Lett. [**88**]{}, 021303 (2002) \[hep-ph/0109106\]. L. A. Anchordoqui, J. L. Feng, H. Goldberg and A. D. Shapere, Phys. Rev. D [**65**]{}, 124027 (2002) \[hep-ph/0112247\].
P. Dobiasch and D. Maison, Gen. Rel. Grav. [**14**]{}, 231 (1982). G. W. Gibbons and D. L. Wiltshire, Annals Phys. [**167**]{}, 201 (1986) \[Erratum-ibid. [**176**]{}, 393 (1987)\]. J. P. Gauntlett, J. B. Gutowski, C. M. Hull, S. Pakis and H. S. Reall, Class. Quant. Grav. [**20**]{}, 4587 (2003) \[hep-th/0209114\]. D. Gaiotto, A. Strominger and X. Yin, JHEP [**0602**]{}, 024 (2006) \[hep-th/0503217\]. H. Ishihara and K. Matsuno, Prog. Theor. Phys. [**116**]{}, 417 (2006) \[hep-th/0510094\]. T. Wang, Nucl. Phys. B [**756**]{}, 86 (2006) \[hep-th/0605048\]. S. S. Yazadjiev, Phys. Rev. D [**74**]{}, 024022 (2006) \[hep-th/0605271\]. T. Nakagawa, H. Ishihara, K. Matsuno and S. Tomizawa, Phys. Rev. D [**77**]{}, 044040 (2008) \[arXiv:0801.0164 \[hep-th\]\]. S. Tomizawa, H. Ishihara, K. Matsuno and T. Nakagawa, Prog. Theor. Phys. [**121**]{}, 823 (2009) \[arXiv:0803.3873 \[hep-th\]\]. K. Matsuno, H. Ishihara, T. Nakagawa and S. Tomizawa, Phys. Rev. D [**78**]{}, 064016 (2008) \[arXiv:0806.3316 \[hep-th\]\]. S. Tomizawa and A. Ishibashi, Class. Quant. Grav. [**25**]{}, 245007 (2008) \[arXiv:0807.1564 \[hep-th\]\]. C. Stelea, K. Schleich and D. Witt, Phys. Rev. D [**78**]{}, 124006 (2008) \[arXiv:0807.4338 \[hep-th\]\]. S. Tomizawa, Y. Yasui and Y. Morisawa, Class. Quant. Grav. [**26**]{}, 145006 (2009) \[arXiv:0809.2001 \[hep-th\]\]. D. V. Gal’tsov and N. G. Scherbluk, Phys. Rev. D [**79**]{}, 064020 (2009) \[arXiv:0812.2336 \[hep-th\]\]. I. Bena, G. Dall’Agata, S. Giusto, C. Ruef and N. P. Warner, JHEP [**0906**]{}, 015 (2009) \[arXiv:0902.4526 \[hep-th\]\]. S. Tomizawa, arXiv:1009.3568 \[hep-th\]. S. ’y. Mizoguchi and S. Tomizawa, Phys. Rev. D [**84**]{}, 104009 (2011) \[arXiv:1106.3165 \[hep-th\]\]. Y. Chen and E. Teo, Nucl. Phys. B [**850**]{}, 253 (2011) \[arXiv:1011.6464 \[hep-th\]\]. C. Stelea, K. Schleich and D. Witt, arXiv:1108.5145 \[gr-qc\]. P. G. Nedkova and S. S. Yazadjiev, Phys. Rev. D [**84**]{}, 124040 (2011) \[arXiv:1109.2838 \[hep-th\]\]. T. Tatsuoka, H. Ishihara, M. Kimura and K. Matsuno, Phys. Rev. D [**85**]{}, 044006 (2012) \[arXiv:1110.6731 \[hep-th\]\]. P. G. Nedkova and S. S. Yazadjiev, Phys. Rev. D [**85**]{}, 064021 (2012) \[arXiv:1112.3326 \[hep-th\]\]. S. ’y. Mizoguchi and S. Tomizawa, Phys. Rev. D [**86**]{}, 024022 (2012) \[arXiv:1201.3063 \[hep-th\]\].
R. C. Myers, Phys. Rev. D [**35**]{}, 455 (1987).
S. D. Majumdar, Phys. Rev. [**72**]{}, 390 (1947). A. Papaetrou, Proc. Roy. Irish Acad. (Sect. A) A [**51**]{} (1947) 191. J. B. Hartle and S. W. Hawking, Commun. Math. Phys. [**26**]{}, 87 (1972). G. N. Candlish and H. S. Reall, Class. Quant. Grav. [**24**]{}, 6025 (2007) \[arXiv:0707.4420 \[gr-qc\]\]. G. W. Gibbons, G. T. Horowitz and P. K. Townsend, Class. Quant. Grav. [**12**]{}, 297 (1995) \[hep-th/9410073\]. D. L. Welch, Phys. Rev. D [**52**]{}, 985 (1995) \[arXiv:hep-th/9502146\]. G. N. Candlish, arXiv:0904.3885 \[hep-th\]. M. Kimura, Phys. Rev. D [**78**]{}, 047504 (2008) \[arXiv:0805.1125 \[gr-qc\]\]. H. Ishihara, M. Kimura, K. Matsuno and S. Tomizawa, Class. Quant. Grav. [**23**]{}, 6919 (2006) \[arXiv:hep-th/0605030\]. H. Ishihara, M. Kimura, K. Matsuno and S. Tomizawa, Phys. Rev. D [**74**]{}, 047501 (2006) \[arXiv:hep-th/0607035\]. C. N. Gowdigere, A. Kumar, H. Raj and Y. K. Srivastava, arXiv:1401.5189 \[hep-th\]. P. T. Chrusciel and D. B. Singleton, Commun. Math. Phys. [**147**]{}, 137 (1992).
S. W. Hawking and G. F. R. Ellis, “The Large scale structure of space-time,” Cambridge University Press, Cambridge, 1973
C. N. Gowdigere, arXiv:1407.5338 \[hep-th\].
[^1]: E-mail:[email protected]
[^2]: E-mail:[email protected]
[^3]: E-mail:[email protected]
[^4]: E-mail:[email protected]
[^5]: In this case, the black hole horizon becomes extremal.
[^6]: The non-smoothness of the horizon was firstly investigated in [@Gibbons:1994vm; @Welch:1995dh], and re-investigated in detail [@Candlish:2007fh]. In [@Candlish:2009vy] and [@Kimura:2008cq], the case of rotating black hole and non trivial topology [@Ishihara:2006iv; @Ishihara:2006pb] were studied, respectively. Recently, the case of multi-center coplanar black hole and membrane horizons were studied in [@Gowdigere:2014aca].
[^7]: In $D\ge 6$, KK black holes with a $S^1$ compactified extra dimension do not have $C^2$ horizon [@Candlish:2007fh].
[^8]: In this paper, we assume all manifolds are at least $C^2$ so that we can consider the second derivative of a geodesic curve w.r.t. affine parameter.
[^9]: The submanifold $\mu^{\prime}({\cal M})$ can be considered as $C^\infty$ manifold with $C^\infty$ metric because of the existence of isometric imbedding $\mu^{\prime} : {\cal M} \to {\cal M^\prime}$.
[^10]: A hypersurface ${\cal H^\prime}$ is smooth if it is described by an equation $\phi = 0$ where $\phi$ is a $C^1$ function on ${\cal M^\prime}$ and it satisfies $d \phi \neq 0$ on ${\cal H^{\prime}}$. For example, in the case of four-dimensional Schwarzschild spacetime $ds^2 = -(1-2M/r)dt^2 + (1-2M/r)^{-1}dr^2 + r^2 d\Omega^2$, we can consider an extension as $ds^2 = -(1-2M/r)du^2 + 2dudr + r^2 d\Omega^2$. In this case, ${\cal H^\prime}$ corresponds to the horizon $r = 2M$. If we choose $\phi$ as $\phi = r- 2M$, $\phi$ satisfies that $\phi = 0$ and $d\phi \neq 0$ at the horizon.
[^11]: If the geodesic does not stay in $\mu^\prime({\cal M})$ even for a sufficiently short geodesic, we only have to change $v^\mu_{(0)}$ into $-v^\mu_{(0)}$ and interchange the past and the future in the following discussion.
[^12]: Although the metric has only $C^1$ on the horizon, we can still construct a Riemann normal coordinate by using a solution of the geodesic equation in Appendix. \[appendixB\].
[^13]: For example, there is a possibility that a geodesic oscillates infinitely many times near the boundary and does not have a limiting point on the boundary.
[^14]: If the tangent of the geodesic is past-directed, we only have to interchange the past and the future in the following discussion.
[^15]: Since the integrand is bounded above and below, $dy^\mu/d\zeta = \int d\zeta d^2y^\mu/d\zeta^2$ takes definite value. From the Eq. (\[zetaparameter\]), we can say that $w^\mu$ cannot be zero vector.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'P. Serra , S.C. Trager , J.M. van der Hulst , T.A. Oosterloo R. Morganti'
date: 'Received ...; accepted ...'
title: 'IC 4200: a gas-rich early-type galaxy formed via a major merger'
---
Introduction
============
Observations during the last decades have been changing the traditional view of early-type galaxies. Once thought to be old, passively evolving and kinematically-relaxed systems devoid of gas, early-type galaxies have been discovered to host subtle but frequent (hence important) deviations from this simple picture: for example, morphological fine structure, intermediate-age stellar populations and a multi-phase interstellar medium. Explaining how these features come to be is an essential requirement for any theory of galaxy formation and evolution.
Among the observed features, neutral hydrogen gas has been found in a significant fraction of early-type galaxies. The detection rate varies from 5 to 45% depending on the sample and the depth of the observations (Sadler et al. 2002) and seems to be higher in the field (Sadler et al. 2001; for individual examples of gas-rich galaxies see Schiminovich et al. 1994, 1995, 1997; Oosterloo et al. 2002). This is usually interpreted in terms of recent accretion of a gas-rich satellite galaxy or equal-mass merger. In particular, major mergers are likely to be the formation mechanism of the very -rich early-type galaxies observed recently, with extremely massive ($>$10$^9$ [$M_{\odot}$]{}), extended (tens of kpc) and quite regular structures surrounding the stellar body (Véron-Cetty et al. 1995; Morganti et al. 1997; Sadler et al. 2002; see also the case of NGC 3656 discussed in Balcells et al. 2001). Simulations have confirmed that such systems can form during the merger of similarly sized gas-rich galaxies as a result of the re-accretion of high-angular-momentum (Hibbard & Mihos 1995; Barnes 2002). However, other scenarios must be taken into account. Theoretical works showed that galaxies can accrete gas from the IGM via a cold mode (Binney 1977; Fardal et al. 2001; Keres et al. 2005). This could provide the supply of atomic gas necessary to build structures without requiring any galaxy-galaxy interaction. It is not clear which of these processes plays the dominant role in the formation of -rich E/S0’s.
Optical observations can shed light on this picture. Unlike accretion from the IGM, merging should produce long-lived signatures in the stellar remnant (Mihos 1995) along with the gas features. During a merger part of the of the progenitors is driven toward the centre of the newly formed galaxy, triggering star formation. Moreover, the incomplete relaxation of the stellar body can result in morphological disturbances visible long after the formation event (Hernquist & Spergel 1992). Such signatures of recent hierarchical assembling have already been observed in the optical images and spectra of early-type galaxies (e.g., Malin & Carter 1983; Trager et al. 2000). There are also suggestions that the morphological disturbances are associated with intermediate-age stellar populations (Schweizer et al. 1990; Schweizer & Seitzer 1992; de Jong & Davies 1997).
On the basis of these results, an analysis of stellar features in gas-rich E/S0’s appears to be a promising way of unveiling their formation history. If these galaxies acquired their via mergers rather than accretion from the IGM, stellar and atomic-gas features, being caused by the same event, should be correlated and give the same indications of the particular process that originated each galaxy and of the time passed since then. Despite observations suggesting that galaxies with optical fine structure are more likely to be detected at 21 cm (Van Gorkom & Schiminovich 1997), other studies indicate that the relation between stellar and properties is not straightforward and other factors (for example environment) play a role (Sansom et al. 2000, Hibbard & Sansom 2003). On the other hand, very little is known about the correlation between and stellar populations, as galaxies with good data have not been well-studied in their stellar content and vice-versa.
To shed light on this picture we have started an observational project aimed to study the -gas content, stellar populations and optical morphology of a sample of gas-rich early-type galaxies. To test the connection between gas and stellar populations/morphology, we will also analyse the optical spectra and images of a control sample of gas-poor galaxies.
We note that optical data will also enable us to investigate the properties of the warm gas (10$^4$ K) in early-type galaxies. More than half of the E/S0’s show ionised gas emission lines in their optical spectra (Phillips et al. 1986; Sarzi et al. 2005). The kinematics of this gas is not always consistent with simple coplanar rotation and is often decoupled from the stellar motion (e.g. Falcón-Barroso et al. 2005), providing clues on the recent history of these systems and on the processes that bring gas into them (Sarzi et al. 2005). Linking ionised and atomic gas observations could provide more complete and detailed indications in this sense.
Here we present the results of optical imaging, long-slit spectroscopy and radio observations of the galaxy IC 4200. This is the first galaxy of the sample for which we obtained both optical and radio data. Our aim is to show the technique that we will apply to all the galaxies of our sample and stress its potential. The main observational results are a very massive ($8.5\times{10^9}$ [$M_{\odot}$]{}) and extended ($\sim$60 kpc) warped disk around the stellar body, an intermediate-age stellar population $\sim$1.5 Gyr old in the centre of the galaxy, optical shells, and ionised gas kinematically decoupled from the stars and characterised by LINER-like line ratios. The combination of these results strongly suggests that IC 4200 formed between 1 and 3 Gyr ago as a result of a major merger rather than by gas accretion from the IGM or accretion of a satellite galaxy.
The paper is structured as follows: in Section 2 the general properties of IC 4200 are discussed; in Section 3 radio, optical spectroscopy and optical imaging observations and data reduction are described; the results of these observations are presented in Section 4; a discussion of the global picture that we derive from these results is given in Section 5; and final conclusions are drawn in Section 6. Appendixes \[Lick/IDScal\] and \[vdispcorr\] give details of the calibration of the absorption-line strengths used in the stellar population analysis. Appendix \[eltables\] contains two tables that give informations on the kinematics of stars and ionised gas and on the line-strength indices across the galaxy. The tables are available electronically at the Centre de Données astronomiques de Strasbourg (CDS).
Basic Properties of IC 4200 {#basicprop}
===========================
[c c c]{} RA (J2000) & 13h 09m 34.7s & c\
Dec (J2000) & –51d 58m 07s & c\
type & S0 & b\
stellar heliocentric velocity & 3850 km/s & a\
heliocentric velocity & 3890 km/s & a\
$z$ & 0.0132 &\
distance & 55 Mpc &\
angular scale & 16 kpc/arcmin &\
$M_B$ & –21.33 & c\
$L_B$ & 5.30$\times{10^{10}}$L$_{\odot}$ &\
$\sigma_{\rm stars,centre}$ & 240 km/s & a\
$M_{\rm {\ion{H}{i}}}$ & 8.54$\times{10^9}$[$M_{\odot}$]{}& a\
$F_{\rm 60 \mu m}$ & 0.17 Jy & d\
F$_{\rm 1.4 GHz}$ & 11.5 mJy & a\
References: (a) this paper, (b) de Vaucouleurs et al. (1991), (c) HyperLeda, (d) Thronson et al. 1989.
Table \[generalproperties\] summarises the main properties of IC 4200. The galaxy is a high-luminosity S0. Its heliocentric stellar systemic velocity derived from our optical long-slit spectroscopy is 3850 km/s, which corresponds to a distance of 55 Mpc and a scale of 16 kpc/arcmin [^1]. The heliocentric systemic velocity of the gas is $\sim3890$ km/s [^2].
IC 4200 was first detected in by the HIPASS[^3] survey (Meyer et al. 2004). emission was measured in the velocity range 3647-4129 km/s resulting in a total flux of 13.3 Jy$\times$km/s within a beam of 14.3 arcmin FWHM (230 kpc). This detection was confirmed by Oosterloo et al. (2006, in preparation) with the ATCA (Australian Telescope Compact Array), as a part of follow-up observations of $\sim$40 early-type galaxies detected by HIPASS. To establish how many of these detections are caused by confusion with independent gas clouds, they observed each galaxy for three hours. They found that in roughly half of the cases gas is associated to the early-type galaxy. IC 4200 is one of these: a very extended gas disk was observed around the stellar body and its mass was estimated as 5.9$\times{10^9}$ [$M_{\odot}$]{}. On the basis of these observations, we included IC 4200 in our sample for the study of the relation between stellar and properties in early-type galaxies.
We characterised the environment of IC 4200 by looking for extragalactic objects within 100 arcmin ($\sim$1.6 Mpc at this redshift) and 300 km/s from the galaxy; for this search we used the NED database. The only object with measured redshift that we found is ESO 219-20, an Sc galaxy at a projected distance of 83 arcmin (1.3 Mpc) from IC 4200 and at a velocity of 3958 km/s. Within 20 arcmin (320 kpc), we found eleven sources without measured redshift, ten of which appear only in the catalogue of 2MASS[^4] detections. All of them have small optical size compared to IC 4200. Using the image derived from our radio observations (Sect.\[hisection\]), we found that only one of the 2MASS objects, 2MASX J13090029–5153544, contains detectable atomic gas at the same velocity of IC 4200: this is an Sc galaxy 6.8 arcmin (110 kpc) northwest of IC 4200 and is the closest object to our target. In addition, our image shows two more sources at $\sim$3900 km/s. These are 15 and 17 arcmin (240 and 270 kpc) distant from IC 4200, but we could not identify their optical counterparts with the available catalogues and DSS[^5] images. Although we cannot conclude that IC 4200 is an isolated galaxy, it appears that the few possible nearby galaxies are much less massive than IC 4200 itself.
[c c c c]{} & & Baselines &\
Dates & Configuration & range(m) & Integration\
5,6 March 2004 & 750A & 77-735 & 2$\times{12}$ h\
11 April 2004 & EW367 & 46-367 & 12 h\
[c c]{}
Centre RA (J2000) & 13:09:34.699\
Centre Dec (J2000) & –51:58:06.99\
Frequency & 1402 MHz\
Bandwidth & 16 MHz\
Number of channels & 512\
Velocity resolution$^1$ & 13 km/s\
*21-cm emission* & *(natural weighting)*\
Synthesised beam & 78"$\times{67"}$\
& $PA$=6.8 deg\
RMS noise & 0.9 mJy/beam\
Maximum signal & 10.5 mJy/beam\
*1.4 GHz continuum* & *(robust=0.5)*\
Synthesised beam & 68"$\times{59"}$\
& $PA$=5.3 deg\
RMS noise & 0.24 mJy/beam\
Maximum signal & 37.21 mJy/beam\
\(1) After Hanning smoothing.
IC 4200 has been observed by many authors at different wavelengths. Malin & Carter (1983) included it in their catalogue of shell galaxies identified by un-sharp masking. According to these authors, shells are observable in the southern part of the galaxy optical image. Carter et al. (1988) followed up this study by observing the optical spectra of a large fraction of the shell galaxies. They measured an intense \[\] emission line in the spectrum of IC 4200.
$UBVRI$ aperture photometry was obtained by Winkler (1997) as a part of his study of a sample of Seyfert galaxies. Within a 20-arcsec-diameter aperture IC 4200 was found to have $V$=13.82, $U-B$=0.77, $B-V$=1.33 and $V-R$=0.78 before correction for the galactic extinction $E(B-V)$=0.30.
Thronson et al. (1989) studied the mid- and far-infrared emission from early-type shell galaxies; they measured a flux density of $(170\pm30)$ mJy at 60 $\mu$m from IC 4200. A flux density of 0.24 Jy was measured at 100 $\mu$m but this measure was flagged as questionable. Finally, IC 4200 has not been detected by any X-ray survey.
Observations and Data Reduction
===============================
Radio
-----
To study the morphology and kinematics of the gas around IC 4200, radio observations were carried out with the ATCA in March and April 2004. Our aim is to obtain resolution high enough to study the velocity structure of the gas, as well as to detect faint extended emission. To achieve this, the galaxy was observed for 3$\times{12}$ h combining the configurations 750A (2$\times{12}$ h) and the very compact EW367 (12 h). Details of the observations can be found in Table \[baselines\] and \[atca\].
In all three 12-h observing runs, we used a 16-MHz bandwidth with 512 channels centred on a frequency of 1402 MHz. PKS 0823–500 and PKS 1934–638 were observed respectively at the beginning and end of each run to determine the bandpass and the flux-density scale; gain changes were monitored by observing the secondary calibrator PKS 1215–457 for 3 minutes for every 40 minutes integration on IC 4200.
Data were reduced using MIRAD software (Sault et al. 1995) in a standard way. 1.4 GHz continuum and 21-cm emission line were separated by continuum linear fitting through the line-free channels in the UV plane.
### {#hisection}
The final emission data cube was obtained by combining the continuum-free data from each run and inverting with natural weighting to maximise the sensitivity to low-column-density, extended regions. Hanning smoothing was applied resulting in a velocity resolution of 13 km/s. The cube consists of 110 7-km/s-wide channels and is characterised by a rms noise of $\sim$0.9 mJy/beam in each channel. The beam has axes of 78 and 67 arcsec ($PA$=6.8 deg) which at the redshift of IC 4200 correspond to 20.8 and 17.8 kpc respectively.
gas is detected around IC 4200 in the velocity range 3590-4130 km/s, consistent with HIPASS detection except for very faint emission at the lower velocities. 21-cm line emission is detected also NW of IC 4200 between 3670 and 3900 km/s; the latter is associated to the 6.8 arcmin (110 kpc) distant galaxy 2MASX J13090029-5153544, which we will call IC 4200-A[^6].
The total image was built using a masked version of the data cube; the mask was built to select pixels above 3 $\sigma$ in either the 60-arcsec or the 180-arcsec smoothed versions of the data cube and above 2 $\sigma$ in the original cube. The masses detected around IC 4200 and IC 4200-A are respectively $8.54\times{10^9}$[$M_{\odot}$]{} and $7.3\times{10^8}$ [$M_{\odot}$]{} and extend out to 60 kpc and 22 kpc from their centres. As mentioned in Sect.\[basicprop\], less massive systems without any visible optical counterpart are detected in the same velocity range $\sim$17 arcmin (270 kpc) south and $\sim$15 arcmin (240 kpc) west of IC 4200. The masses of these systems are 1.5 and $1.7\times{10^8}$ [$M_{\odot}$]{}.
### Continuum
The 1.4-GHz continuum image was made by combining the 21-cm emission-free data from each observing run and inverting them with $robust$=0.5 (for details on the $robust$ parameter see Briggs 1995). The beam of the image has axes of 68 and 59 arcsec ($PA$=5.3 deg). The rms noise is of 0.24 mJy/beam. A flux density of 11.5 mJy is detected within one beam centred on IC 4200. IC 4200-A is also detected at a 6 $\sigma$ level, with a total unresolved flux of 1.5 mJy.
Optical Spectroscopy {#genoptspec}
--------------------
[c c]{} Date & 12 July 2004\
Period & ESO-073.A\
Telescope & NTT(3.6 m)\
Instrument & EMMI (red arm)\
Dispersing medium & Grism 5\
Spectral range & 3800-7020 Å\
Dispersion & 1.66 Å/pixel\
Resolution & 5 Å FWHM\
Detector & $2\times{2048}\times{4096}$ MIT/LL CCD\
Binning & 2$\times{2}$\
Gain & 1.32 e$^-$/ADU\
Read-out-noise & 3.825 e$^-$\
Scale & 0.332 arcsec/pixel\
Slit length & 8 arcmin\
Slit width & 1 and 5 arcsec\
Dispersion, scale, gain and $RON$ refer to 2$\times{2}$ binning.
Long-slit spectroscopy of IC 4200 was carried out on 12 July 2004 at ESO La Silla Observatory using EMMI mounted on the NTT. The main purpose of these observations is to study the stellar content and kinematics of the galaxy. To achieve this target our spectra cover the range 4000 to 7000 Å with dispersion 1.66 Å/pixel (with 2$\times{2}$ binning) and resolution of $\sim$5 Å. Observational details can be found in Table \[spec\].
The time spent on the galaxy was divided between two perpendicular slit positions along the optical major and minor axes (respectively at $PA$=150 and 60 deg). Because of bad weather, the total exposure times are 3$\times{1350}$ sec along the major axis and only 2$\times{1350}$ sec along minor axis. The galaxy was observed through a 1-arcsec-wide and 8-arcmin-long slit with an average seeing of $\sim$1.7 arcsec. During the same observing run, we observed the spectra of 3 spectrophotometric standard stars for flux calibration purposes through a 5-arcsec-wide slit: the observed stars are EG 21, LTT 4816 and LTT 6248. Eight more stars (which we will call Lick/IDS stars) were observed through the 1-arcsec-wide slit to calibrate our line-strength indices to the Lick/IDS system (Worthey et al. 1994) in order to study the stellar populations; Table \[LICKlist\] lists the stars with their spectral type and exposure times.
[c c c c]{} Star & Spectral & Stellar & Exposure\
& Type & Library & Time (sec)\
HD 114113 & K3III & INDO-US & 42\
HD 124850 & F7IV & MILES, INDO-US & 40\
HD 125454 & G8III & INDO-US,Lick/IDS & 120\
HD 126218 & K0III & MILES & 120\
HD 131977 & K4V & MILES,Lick/IDS & 240\
HD 137052 & F5IV & INDO-US & 120\
HD 138716 & K1IV & INDO-US & 120\
HD 139446 & G8III/IV & MILES,INDO-US & 130\
[c c c c]{} Index & Index band & Blue continuum & Red continuum\
H$\beta$ & 4847.875 & 4827.875 & 4876.625\
& 4876.625 & 4847.875 & 4891.625\
Mg$b$ & 5160.125 & 5142.625 & 5191.375\
& 5192.625 & 5161.375 & 5206.375\
Fe5270 & 5245.650 & 5233.150 & 5285.650\
& 5285.650 & 5248.150 & 5318.150\
Fe5335 & 5312.125 & 5304.625 & 5353.375\
& 5352.125 & 5315.875 & 5363.375\
H$\gamma_{\rm A}$ & 4319.750 & 4283.500 & 4367.250\
& 4363.500 & 4319.750 & 4419.750\
H$\delta_{\rm A}$ & 4083.500 & 4041.600 & 4128.500\
& 4122.250 & 4079.750 & 4161.000\
H$\gamma_{\rm F}$ & 4331.250 & 4283.500 & 4354.750\
& 4352.250 & 4319.750 & 4384.750\
H$\delta_{\rm F}$ & 4091.000 & 4057.250 & 4114.750\
& 4112.250 & 4088.500 & 4137.250\
\(1) Wavelengths are in angstroms.
[c c c c]{} Index & Slope ($m$) & Intercept ($q$) & $\chi^2$\
H$\beta$ & 1.05 & –0.23 & 1.7\
Mg$b$ & 1.15 & –0.16 & 12.6\
Fe5270 & 1.03 & –0.12 & 22.7\
Fe5335 & 0.77 & 0.40 & 0.8\
H$\gamma_{\rm A}$ & 1.08 & –1.45 & 7.5\
H$\delta_{\rm A}$ & 1.03 & –1.09 & 1.4\
H$\gamma_{\rm F}$ & 1.00 & –0.51 & 3.3\
H$\delta_{\rm F}$ & 1.05 & –0.45 & 2.5\
The measured value of a given index $i$ is: $i_{\rm meas}=m\cdot i_{\rm ref}+q$, where $i_{\rm ref}$ is the value on the Lick/IDS system.
Data were reduced using a variety of suitable programs written by Daniel Kelson, Scott Trager and Paolo Serra in the Python programming language. Reduction was performed in a standard way. The rms of the wavelength calibration of the frames is on average 0.05 Å and never exceeds 0.1 Å. Sky emission was modelled and subtracted before re-binning the frames to correct for distortion and wavelength calibration; this allows one to better sample the sky lines profiles by exploiting the distortion and therefore reduces the residuals at the line edges (Kelson 2003).
One-dimensional spectra of the calibration stars were extracted. A flux calibration curve was derived from the spectrophotometric standard stars and applied to the Lick/IDS stars. The latter were then used to calibrate the measured line-strength indices to the Lick/IDS system. Each index was calibrated by determining the linear relation between its measured value and the one on the Lick/IDS system (see Appendix \[Lick/IDScal\] for details). To efficiently disentangle age from metallicity in the spectrum of the galaxy, we use the indices H$\beta$, Mg$b$, Fe5270 and Fe5335. We test the result against indices of higher-order Balmer-lines: H$\gamma_{\rm A}$, H$\delta_{\rm A}$, H$\gamma_{\rm F}$, and H$\delta_{\rm F}$, defined in Worthey & Ottaviani (1997). All the indices used are defined in Table \[ind\_def\]. Table \[Lick\_cal\] contains the coefficients of the relations that calibrate these indices to the Lick/IDS system.
The flux calibration was finally applied to the two-dimensional spectra along the major and minor axes of IC 4200. Any information of interest can be derived from these calibrated 2D spectra as a function of the radial coordinate. To do so we built 1D spectral bins of appropriate S/N. Higher S/N is needed for line-strength indices measurement than for kinematics analysis in order to obtain reasonable errors; therefore, we extracted bins of S/N$\sim$30/pixel and of S/N$\sim$80/pixel at 5000 Å for these two different purposes. In the rest of the paper low- and high-S/N 1D bins will be referred to as $B_{30}$’s and $B_{80}$’s.
### Disentangling stars from ionised gas {#fit}
![Central spectrum $B_{80}$ of IC 4200 along the optical major axis in arbitrary flux units. Ionised gas emission lines and the main stellar absorption features are highlighted.[]{data-label="B80MJcen"}](centralspec.eps){width="8.5cm"}
Fig.\[B80MJcen\] shows the central $B_{80}$ along the major axis. \[\], \[\], \[\], H$\alpha$ and \[\] emission lines are easily recognised and the H$\beta$ absorption feature is completely filled by emission; even $H\gamma$ absorption is contaminated by the ionised gas. The same features are observed in all the $B_{80}$’s. On one hand emission lines enable us to study the physical processes and kinematics of the gas within the stellar body. However, they make the stellar population analysis more complicated by contaminating or even hiding important absorption features (e.g. H$\beta$, crucial for determining the age of stellar populations). Before measuring line-strength indices from the $B_{80}$’s it is therefore necessary to remove the ionised gas emission from these spectra. We separate gas and stellar spectra also along the $B_{30}$’s to study the kinematics of the gas and of the stars separately.
To disentangle the ionised gas and the stellar contribution to the spectrum of IC 4200, we fit Bruzual & Charlot (2003, hereafter BC03) stellar population synthesis models to the $B_{80}$’s and $B_{30}$’s in their emission-free wavelength ranges. We use the high-resolution (3 Å) models built with the Padova 1994-isochrones and a Salpeter IMF, selecting metallicities $Z$=0.004, 0.008, 0.02, 0.05 ($Z_{\odot}$=0.02) and ages between 1 and 20 Gyr, evenly spaced in logarithmic steps of 0.1. We apply the direct fitting method described in Kelson et al. (2000). In this method the model spectra, smoothed to the instrumental resolution of our data, are convolved with a parameterised line-of-sight velocity distribution (LOSVD); this is assumed to be a Gaussian and has two free parameters, the velocity $v$ and the velocity dispersion $\sigma$. Continuum matching is obtained by an additive and a multiplicative correction of the model spectra, both dependent on wavelength via polynomial functions. The former is necessary because the overall shape of the galaxy spectrum might be not well represented by the models; the latter removes any mismatch caused by the instrumental response and provides the necessary normalisation. We have tested the effect of using the additive correction during the fitting procedure. We found that while applying this correction gives a better quality of the fit, it does not change the best fitting spectrum within 1.5 $\sigma$ at any wavelength.
A best fitting model $M_k$ is associated to each bin $B_k$. To separate gas emission from the underlying stellar absorption spectrum we use the technique described in Emsellem et al. (2002). The residual $G_k=B_k-M_k$ represents an estimate to the emission from ionised gas. We then build a model emission spectrum $\widetilde{G_k}$ by fitting each emission line in $G_k$ with a Gaussian curve. The model emission spectrum is finally subtracted from the original bin, leaving us with a pure stellar absorption spectrum $S_k=B_k-\widetilde{G_k}$.
As a part of the fitting procedure of the $B_{30}$’s we obtain the stellar LOSVD along the two axis. Furthermore, ionised gas velocities are determined from the central wavelengths of the single-Gaussian fit to the emission lines performed to build the $\widetilde{G_{30}}$’s. Finally, we use the flux of the ionised gas emission lines – measured when modelling the $G_{80}$’s – to examine the ionisation mechanism in the galaxy. Informations on the stellar and ionised-gas kinematics along the $B_{30}$’s are listed in Table C.1.
### Absorption-line indices {#indices}
We use the $S_{80}$’s to study the stellar populations of IC 4200. For each bin $S_k$ five main steps have to be performed in order to properly determine the Simple-Stellar-Populations-equivalent (SSP-equivalent) age, metallicity and abundance ratio:
- broaden $S_k$ to match the Lick/IDS resolution
- measure the indices $i_n$ ($n=1,...,N_{\rm indices}$)
- correct the indices for the effect of the stellar velocity dispersion $\sigma_k$
- bring the corrected indices to the Lick/IDS system
- compare the final values of the indices to the models
The first two steps are performed by feeding the stellar velocity $v_k$ to a Python code that broadens $S_k$ to the Lick/IDS resolution (Worthey & Ottaviani 1997) and measures the indices placing the bandpasses of Table \[ind\_def\] in the appropriate wavelength range according to the stellar velocity.
Velocity dispersion broadens the spectral feature of galaxies with respect to the ones of stellar spectra, altering the line strengths by transporting flux in and out of the bandpasses that define each index. Therefore, before comparing the measured indices to the models (which are built with zero velocity dispersion) we need to compensate for this effect. To perform this correction we use the result of BC03-models fitting of Sect.\[fit\]. We measure each index $i$ from the best fitting model $M_k$ and from its zero-velocity-dispersion equivalent $M_{0,k}$, obtaining $i_{k}$ and $i_{0,k}$ respectively. We then calculate $f=i_{k}/i_{0,k}$. Finally, we divide the index measured from $S_k$ by $f$, so correcting it for the effect of the stellar velocity dispersion in $S_k$. Plots of the correction coefficient $f$ as a function of the velocity dispersion are given for each index used in this paper in Appendix \[vdispcorr\].
The linear transformations that bring the measured indices onto the Lick/IDS system (i.e. the inverse of the transformations of Table \[Lick\_cal\]) are then applied. Table C.2 lists the value of all the Lick/IDS indices fully corrected and calibrated. The result of the comparison of the index values to the models is showed in Sect.\[spopandfrac\].
Optical Imaging {#optobs}
---------------
$V$ and $R$ broad-band imaging of IC 4200 was obtained on 4 February 2005 at La Silla Observatory, ESO. As for the optical spectroscopy, we used the red arm of EMMI on the NTT which is characterised by a field of view of 22.7$\times$22.7 arcmin$^2$. Detector parameters are the same of table \[spec\] (2$\times{2}$ binning was applied). During the observations the seeing was on average $\sim$0.8 arcsec.
The galaxy was observed for 12$\times{50}$ sec and 12$\times{25}$ sec respectively in $V$ and $R$ band. Each frame was bias subtracted. Flat fielding was performed using $V$ and $R$ median sky flats obtained from multiple twilight exposures. The sky emission was assumed to be equal to the mode of the pixel distribution in each science frame and then subtracted. The calibration onto the magnitude scale was obtained by use of five and seven standard stars in $V$ and $R$ band respectively.
To analyse the image of IC 4200 without contamination from the stars in the field, we built a stellar mask in each band. This was done by applying a 41$\times{41}$ pixel$^2$ median filter that smooths away the stars, dividing the original image by the smoothed one and imposing a threshold on the resulting frame to find the pixels that belong to the stars. We used the mask when measuring the total flux of IC 4200 in the two observed bands. Total $V$ and $R$ magnitude and the $V-R$ colour are $V$=12.65, $R$=11.86 and $V-R$=0.79. The colour is in good agreement with the 0.77 found by Winkler (1997); on the contrary, our magnitude is brighter, but this is a consequence of his narrow aperture [^7]. We corrected magnitudes and $V-R$ colour for galactic reddening using the corrections 1.06 mag in $V$ and 0.85 mag in $R$ given by NED. Corrected magnitudes and colour are $V$=11.59, $R$=11.01, $V-R$=0.58.
Results
=======
{#higasres}
![Total contours on top of the DSS optical image of IC 4200. Contour levels are (0.2, 0.4, 0.6, 0.8, 1.2, 1.6, 2.0, 2.6, 3.2)$\times{10^{20}}$ cm$^{-2}$. The maximum column density is 3.32$\times{10^{20}}$ cm$^{-2}$. The beam has axes of 78 and 67 arcsec ($PA$=6.8 deg). For a grey scale representation of the total image see Fig.\[model\].[]{data-label="hitot"}](himap.eps){width="8cm"}
![Position-velocity diagrams along $PA$=30 deg (left) and $PA$=125 deg (right). The left diagram is aligned along the kinematic major axis of the inner gas region. The right diagram is only approximately aligned along the kinematic major axis of the outer gas region ($PA$=120 deg) to better show the connection between IC 4200 and IC 4200-A (lower right; see the discussion in the text).[]{data-label="pa_mos"}](kpvslice.eps){width="9cm"}
Fig.\[hitot\] shows the total emission image derived from our ATCA observations. The resolution of the data is low: the whole gas structure around IC 4200 is sampled by only 6 beams along its maximum extension; the one around IC 4200-A is contained within 2 beams. Nevertheless, it is possible to extract a significant amount of useful information.
We detect $8.54\times{10^9}$ [$M_{\odot}$]{} of gas around IC 4200. The gas goes as far as 60 kpc from the centre of the galaxy and shows a fairly regular configuration. The major axis of the isophotes rotates from $\sim$20 to $\sim$80 deg when moving out from the centre. The isophotes are quite regular and symmetric around the centre in the inner $\sim$3 beams. The regular inner appearance vanishes in the asymmetric outer regions, where the east side is more extended than the west one and presents a feature that could be a spiral arm. On the northwest the structure is elongated toward the companion.
The velocity structure of the cube shows a certain order, indicating that the gas must have already been in place for some time. Position-velocity diagrams (Fig.\[pa\_mos\]) suggest that the kinematic major axis is aligned along $\sim$30 deg in the inner region and along $\sim$120 deg in the outer part of the gas structure, following the trend of the isophotal major axis with radius. The kinematic major axis of the outer regions is therefore parallel to the photometric minor axis of the inner region, indicating that this is not a regular disk. The strong rotation of the kinematic major axis with radius also rules out simple coplanar rotation as a model for the motion of the gas around IC 4200.
![On the left: edge-on (top) and sky-projected (bottom) view of the warped gas-disk model. The model is not meant to reproduce the gas density distribution of the in IC 4200 but rather its velocity structure (see Fig.\[modelandcube\]). However, the figure shows that projection effects caused by the warp can produce a spiral pattern like the one observed in the total- image of the galaxy, showed on the right.[]{data-label="model"}](modelviews2.eps){width="9cm"}
We model the observed system in a simple way, finding that a strongly-warped disk satisfactorily reproduces the velocity structure of the cube. The model has a radius of 4 arcmin ($\sim$60 kpc) and a flat rotation curve with $v_{rot}=320$ km/s. We keep the density constant with radius as we are interested in reproducing the kinematics of the disk rather than its detailed density distribution. The inclination is also kept constant as a function of the radius and equal to 45 deg; finally, the major axis $PA$ increases linearly from 30 deg in the centre to 120 deg at the outer radius of 4 arcmin, producing a 90 deg warp.
![Constant-Dec (top six panels) and -RA (bottom six panels) position-velocity diagrams of the observed (grey scale) and modelled (contours) gas system. Both Dec and RA increase from left to right, top to bottom with increments of 1 arcmin.[]{data-label="modelandcube"}](dmoslabels.eps "fig:"){width="8.75cm"} ![Constant-Dec (top six panels) and -RA (bottom six panels) position-velocity diagrams of the observed (grey scale) and modelled (contours) gas system. Both Dec and RA increase from left to right, top to bottom with increments of 1 arcmin.[]{data-label="modelandcube"}](rmoslabels.eps "fig:"){width="8.75cm"}
In Fig.\[model\] the disk model is shown edge-on and projected on the sky. The strong warp produces projection effects that can simulate spiral arms like the observed one. Fig.\[modelandcube\] shows the good agreement between data cube and model when looking at constant-RA and -Dec slices. Although the model is not the result of a best fit, only minor changes are allowed in order to reproduce the observed velocity structure. Finely tuning the model parameters is not worth the effort given the resolution of our data. We can therefore conclude that the gas orbits around IC 4200 forming a $\sim$90 deg warped disk with a flat rotation curve with $v_{rot}$=320 km/s. This is the main conclusion of our 21-cm line emission observations.
The content of IC 4200 is very high, 1.7 times the one of the Milky Way [^8]. However, the gas is spread over an extremely wide area; it is therefore very dilute, its column density being below a few times 10$^{20}$ cm$^{-2}$ across most of the system. As a result, no star formation is expected on the scale of the beam size. To quantify this point we estimate the critical surface density for star formation proposed by Kennicutt (1989). Star formation occurs at densities higher than $$\Sigma_{\rm critical}(R)=6.2 \frac{\sigma(\rm km/s) \ v(km/s)}{R(\rm kpc)} 10^{18}\rm cm^{-2},$$ where $v$ and $\sigma$ are the rotational velocity and the local velocity dispersion of the gas at the radius $R$, and flat rotation has been assumed. Using the model rotational velocity of 320 km/s and a standard velocity dispersion of 7 km/s, the critical column density ranges from 2.5$\times{10^{20}}$ cm$^{-2}$ at the outer edge of the disk to $\sim$10$^{21}$ cm$^{-2}$ in the centre. These values are always higher than the measured column density, which ranges from 2.0$\times{10^{19}}$ to 3.3$\times{10^{20}}$ cm$^{-2}$. Therefore, no star formation is expected on a scale of the order of the beam size. The same conclusion is reached using the criterion proposed by Schaye (2004), where the cooling of the gas triggers gravitational instability and therefore star formation at a fixed surface density of $\sim$6$\times{10^{20}}$ cm$^{-2}$. However, because of the large beam smearing, we cannot exclude that star formation is occurring in unresolved clumps of cold gas across the disk or in its central regions.
gas is detected also around the nearby galaxy IC 4200-A, resulting in a mass of $7.3\times{10^8}$ [$M_{\odot}$]{} and a radius of $\sim$22 kpc. Although the two galaxies do not show any interaction at optical wavelength, the 21-cm picture is perhaps different. The detected gas systems are extremely close and a faint plume of appears to be connecting IC 4200 to the companion, suggesting that the two gas disks might indeed be interacting. However, the forming the connection fits well in the velocity structure of IC 4200, as can be seen in the right panel in Fig.\[pa\_mos\]. This suggests that the gas of the plume belongs to IC 4200 and is not being significantly disturbed by the passage of IC 4200-A, which is much less massive than IC 4200 itself. The spectacular appearance of the plume of gas in Fig.\[hitot\] could be caused by missing gas with density lower than our detection limit in the neighbouring regions. Our data are therefore not conclusive on the actual interaction between the two galaxies.
Radio continuum {#cont_res}
---------------
![ Plot of the stellar and ionised-gas velocity measured from the $B_{30}$’s along the optical major (top panel) and minor (bottom panel) axis. The velocity of the ionised gas is measured from H$\alpha$ emission line. The circles represent the ionised gas points. The horizontal bars represent the width of each bin. In order to gain S/N, the outer bins are very wide, going as far as 20 arcsec from the centre along the major axis and 14 arcsec along the minor axis. In the last bin along the major axis no reliable measurement of the gas velocity was possible. The origin of the velocity axis is the stellar systemic velocity.[]{data-label="kin"}](kinMJnew.eps){width="7cm"}
![ Plot of the stellar and ionised-gas velocity measured from the $B_{30}$’s along the optical major (top panel) and minor (bottom panel) axis. The velocity of the ionised gas is measured from H$\alpha$ emission line. The circles represent the ionised gas points. The horizontal bars represent the width of each bin. In order to gain S/N, the outer bins are very wide, going as far as 20 arcsec from the centre along the major axis and 14 arcsec along the minor axis. In the last bin along the major axis no reliable measurement of the gas velocity was possible. The origin of the velocity axis is the stellar systemic velocity.[]{data-label="kin"}](kinminew.eps){width="7cm"}
A unresolved source with 1.4 GHz continuum flux density of 11.5 mJy is associated to IC 4200. Previous results (strong \[\] emission line and 60 $\mu$m detection) suggest that IC 4200 hosts some activity (star formation, nuclear activity) and therefore a non-zero 1.4 GHz flux is not surprising. To understand its origin we compare it with the emission at 60 $\mu$m (which has a flux density of 0.17 Jy). In star forming galaxies radio and FIR luminosities are tightly correlated. Using the radio-FIR relation of Yun et al. (2001), we find that the 1.4 GHz flux density expected from the 60 $\mu$m emission is 1.23 mJy. There is therefore a radio excess of a factor of ten, $\sim$6 times the scatter of the radio-FIR relation. This result strongly suggests that physical processes other than star formation contribute to the radio continuum (e.g. a weak AGN); however, it still allows some star formation to be occurring. We derive an upper limit on the star formation rate (SFR) using the FIR luminosity of IC 4200 and applying the result of Kewley et al. (2002). The resulting estimate is SFR=0.26 [$M_{\odot}$]{}/yr.
As mentioned, IC 4200-A is also detected in the radio continuum, its flux density being 1.5 mJy. The 60 $\mu$m flux density of IC 4200-A is 0.4 Jy, which implies a radio excess of a factor of two (only 1.5 times the scatter of the relation). IC 4200-A seems to lie on the normal radio-FIR relation for star forming galaxies, with a SFR of $\sim$0.34 [$M_{\odot}$]{}/yr.
Stellar and ionised gas kinematics
----------------------------------
![Plot of the stellar velocity dispersion measured from the $B_{30}$’s. Open circles correspond to points on the major axis, filled squares correspond to the minor axis. The horizontal bars represent the width of each bin.[]{data-label="sigma"}](sigmapaper2.eps){width="8cm"}
We sampled the kinematics of stars and ionised gas within 20 and 14 arcsec along the major and minor axis respectively [^9]. Fig. \[kin\] shows stellar and gas velocities. The gas velocity is the one derived from the H$\alpha$ emission line. The obvious and main result is that stars and ionised gas have decoupled kinematics. Along the major axis the stars show $\sim$180 km/s rotation at the outer radius. In the inner 1.5-2 arcsec ($\sim$500 pc) the gas is counter-rotating compared to the stars, with the velocity steeply rising to $\sim$100 km/s. Along the minor axis, very little rotation is observed in the stellar phase, for which the velocity gradient changes sign from the inner $\sim$2 arcsec to the outer regions, and becomes again negative when looking at the outer bins $\sim$10 arcsec (2.7 kpc) from the centre. On the other hand, the gas is clearly rotating around a systemic velocity of $\sim$50 km/s and with a maximum velocity of $\sim$150 km/s.
The kinematics of the gas is certainly not consistent with coplanar circular motion, but the available data do not allow a clear interpretation. In any case, it is interesting to note that the behaviour of the ionised gas along the optical minor axis could be seen as a continuation of the velocity profile of the atomic gas in the inner region of the disk (the left panel of Fig.\[pa\_mos\] gives a good representation of the velocity curve along the optical minor axis). However, the difference in resolution (the kinematics of the ionised gas is sampled in a region which is roughly one tenth of the radio beam) does not allow us to further investigate this connection.
Fig.\[sigma\] shows the stellar velocity dispersion profiles along the two optical axes. The central dispersion is of $\sim$240 km/s and decreases to 150 km/s in the outer bins along the major axis. It is interesting to compare this result with the rotation of the disk. The latter could be modelled with a flat rotation of 320 km/s, implying a rising mass-to-light ratio and, with the assumption of spherical symmetry, an isothermal density profile ($\rho \propto r^{-2}$). For an isothermal sphere the ratio between the rotational velocity and the central velocity dispersion is $\sqrt{2}$. Interestingly, the ratio between the modelled rotational velocity and the observed line-of-sight velocity dispersion is 320/240=1.33, in reasonable agreement with the isothermal case.
Gas ionisation
--------------
![Diagnostic diagram for the gas ionisation derived from the $B_{80}$’s. Open circles correspond to points on the major axis, filled squares correspond to the minor axis. Taking into account the error bars, all the points are roughly coincident; therefore individual bins were not labelled. Galaxies characterised by different ionisation mechanisms fall in different regions of the plot, as schematically indicated in this diagram (see for example Kauffmann et al. 2003 and references therein).[]{data-label="bpt"}](BPT_marks2.eps){width="8cm"}
Scaling the H$\alpha$ luminosity measured along the major and minor axis, we estimate a total H$\alpha$ luminosity of 2.2$\times{10^{41}}$ ergs/s within one effective radius. To understand what causes the ionisation of the gas we use the diagnostic diagram introduced by Baldwin, Phillips & Terlevich (1981), as shown in Fig.\[bpt\]. Points belong to both major and minor axis and correspond to the radial bins listed in Table \[stellarpopresult\]. The line ratios fall at all radii in the region typical of LINERs. LINERs spectra are not well understood and are commonly believed to be the product of different processes at work simultaneously: e.g., star formation, nuclear activity, shocks, old hot AGB stars. Given our 1.4 GHz continuum result, it is not surprising to find that the ionisation of the gas cannot be explained in terms of star formation only. However, it is hard to understand which processes are going on in IC 4200.
It is interesting that the nature of the ionisation remains the one of a LINER across the whole radial range sampled ($\sim$1.5 kpc from the centre). This suggests that diffuse shocks might be playing an important role (shocks were proposed as an explanation of LINERs spectra for example by Dopita & Sutherland 1995). Shocks could be caused in IC 4200 by continuing gas inflow from the extended disk or by the propagation of a jet produced by a weak AGN in the centre of the galaxy. The AGN could contribute to the observed 1.4 GHz emission, but a radio jet should produce a difference in the emission-line ratios along the two axes, which is not observed. The kinematics of the ionised gas could in principle provide a clue on whether a gas inflow or jets are present. However, although some line profiles are suspected of being composed by more than just one Gaussian, no clear evidence was observed in this sense. On the whole, we cannot reach a conclusion on the ionisation mechanism other than there is an important component not related to star formation.
Stellar populations {#spopandfrac}
-------------------
![Plot of the spectral indices (derived from the $B_{80}$’s) on top of the model grid of W94 (solar abundance ratios). In the plot the age runs along the $\sim$horizontal, constant-H$\beta$ solid lines, the metallicity in the perpendicular direction (dashed lines). The age of the constant-age lines is, from top to bottom, 1.5, 2, 3, 5, 8, 10, 12, 15 and 18 Gyr; metallicities \[Z/H\] are, –0.25, 0.0, 0.25, 0.5 ordered along increasing $<$MgFe$>$. Open circles correspond to points on the major axis and are connected to each other by a dotted line; filled squares correspond to the minor axis and are connected to each other by a dash-dotted line.[]{data-label="gridpoints"}](gridnarrow_label.eps){width="8.5cm"}
[c c c c c]{} r (arcsec) & age (Gyr) & \[Z/H\] & \[E/Fe\] & $\chi^2$\
Major axis & & & &\
-5.5,-1.8 & 3.5$\pm$1.4 & 0.22$\pm$0.12 & 0.19$\pm$0.05 & 0.30\
-1.8,-0.5 & 1.6$\pm$0.3 & 0.70$\pm$0.17 & 0.28$\pm$0.05 & 1.67\
-0.5,+0.5 & 1.6$\pm$0.3 & 0.79$\pm$0.16 & 0.29$\pm$0.05 & 1.73\
+0.5,+1.8 & 1.7$\pm$0.3 & 0.69$\pm$0.16 & 0.29$\pm$0.05 & 1.93\
+1.8,+5.5 & 2.0$\pm$0.5 & 0.50$\pm$0.15 & 0.15$\pm$0.05 & 1.74\
Minor axis & & & &\
-5.8,-0.8 & 5.4$\pm$2.3 & 0.05$\pm$0.12 & 0.20$\pm$0.05 & 0.04\
-0.8,+0.8 & 1.8$\pm$0.3 & 0.55$\pm$0.15 & 0.30$\pm$0.05 & 0.38\
+0.8,+5.8 & 2.2$\pm$0.6 & 0.50$\pm$0.14 & 0.24$\pm$0.05 & 0.83\
The upper part of the table corresponds to the optical major axis, the lower part to the minor axis.
Fig.\[gridpoints\] shows the comparison of the measured Lick/IDS line-strength indices with Worthey (1994, hereafter W94) models on the plane \[H$\beta$,$<$MgFe$>$\][^10]. Points correspond to the bins listed in Table \[stellarpopresult\]. On the plane of Fig.\[gridpoints\] age and metallicity of the models are efficiently decoupled (especially at metallicities around solar or higher), the age being well traced by H$\beta$ and the metallicity by $<$MgFe$>$. From this figure it appears that the SSP-equivalent age is $\sim$1.5 Gyr in the centre of the galaxy and increases to $\sim$3 Gyr in the outer bins.
H$\beta$, Mg$b$, Fe5270 and Fe5335 line-strength indices were used in a 4-dimensional space to more precisely derive the most likely SSP-equivalent age, metallicity and abundance ratios by comparison with W94 models extended to non-solar abundance ratios (see Trager et al. 2005 and references therein). The result of this analysis confirms the indications derived from Fig.\[gridpoints\] (see Table \[stellarpopresult\]). This implies that there is a substantial stellar component 1.5 Gyr old or younger in the centre of the galaxy.
As emphasised by Fig.\[gridpoints\], H$\beta$-absorption-line strength plays a crucial role when measuring the SSP-equivalent age of stellar populations. In our case, this line was studied after removing the ionised gas emission. To test the reliability of this method we compared these results with another method used by previous authors. Namely, we derived H$\beta$ absorption strength by measuring it from the original $B_{80}$’s (i.e. without subtracting the fitted ionised gas emission spectrum) and then correcting it by an additive term proportional to the equivalent width of \[\]5007 emission line by a factor $g$. According to Trager et al. (2000), $g$ varies from galaxy to galaxy within the range 0.33-1.25 and is on average 0.6. In our case the best match between the two different techniques is obtained for $g$=0.65, which fits well with previous studies.
We performed a further check on our results by studying the stellar populations with higher-order Balmer lines, namely H$\gamma$ and H$\delta$. These lines are less affected than H$\beta$ by emission contamination, and therefore provide a good test on the robustness of our emission-correction. We used the indices H$\gamma_{\rm A}$, H$\gamma_{\rm F}$, H$\delta_{\rm A}$ and H$\delta_{\rm F}$ as defined in Table \[ind\_def\]. We find that ages determined from H$\gamma_{\rm A}$ and H$\delta_{\rm A}$ (measured from the $S_{80}$’s) are in agreement with the ones determined using H$\beta$. There are systematic deviations, H$\gamma_{\rm A}$- and H$\delta_{\rm A}$-based ages being respectively lower and higher than the H$\beta$-based ones; however, these differences are comprised within the errors. On the other hand, the ages determined from H$\gamma_{\rm F}$ and H$\delta_{\rm F}$ are always higher than the ones determined from H$\beta$ by 2 Gyr or more. This discrepancy was noted already by Thomas, Maraston & Korn (2004). In their Fig.2 one can see that ages determined from H$\gamma_{\rm A}$ are systematically lower than the ones determined from H$\gamma_{\rm F}$ and that H$\beta$-based measurements lie somewhere in between them. We finally note that we obtain the same relative ages, metallicities, and abundance ratios (for all the Balmer lines mentioned) by comparing our line strength indices to the Thomas, Maraston & Bender (2003) models instead of the W94 ones.
We can therefore robustly conclude that the stellar populations in IC 4200 has a SSP-equivalent age of 1.5 Gyr in the centre, and that this rises to 3 Gyr in the outer region where it could be reliably determined (between $\sim r_e$/4 and $\sim r_e$/3).
Optical Imaging {#imag_res}
---------------
![$V-R$ colour image of the central 66$\times$66 arcsec$^2$. In the image north is up, east is left. The grey scale is in magnitudes. Axis units are in pixels, with 1 pixel=0.332 arcsec.[]{data-label="colimage"}](spiral.eps){width="8cm"}
[c c c c c c c]{} & & scale & Sersic & Axis &\
Model & Magnitude & (arcsec) & Exponent & Ratio & $PA$\
Sersic & 11.77 & 19.2 & 4.03 & 0.696 &–23.9\
S+d: bulge & 12.88 & 4.6 & 1.98 & 0.746 &–22.6\
S+d: disk & 12.35 & 21.5 & 1 & 0.590 &–25.5\
The $\chi^2$ of the fit is $\sim$5 for the Sersic model, $\sim$2 for the Sersic+disk model.
Fig.\[colimage\] shows the $V-R$ colour image of IC 4200. This was derived after smoothing the $V$ and $R$ images with the same Gaussian filter of width $\sim$0.7 arcsec. The central region shows the presence of dust in a spiral-like pattern. This implies the existence of a small disk in the inner 5-10 arcsec. Although less clearly, the same pattern can be seen in the un-sharp-masked images in $V$ and $R$ band.
The $V$- and $R$-band morphology of IC 4200 was studied by fitting the images in the two bands with a Sersic and a Sersic+disk profile using GALFIT (Peng et al. 2002). The fit was performed masking out the stars as described in Sect.\[optobs\]. In both bands the Sersic+disk profile gives a slightly better fit, but the two models are very similar within the noise of the image. Relevant fitted quantities are listed for the $R$ band in Table \[galfit\] (the result is the same in $V$ band). The total magnitudes and colour derived from the models are in agreement with what determined manually (V=11.59, R=11.01, $V-R$=0.58 after correction for the reddening).
![$R$-band residual image after fitting a Sersic+disk model to IC 4200. In the image, north is up, east is left. The grey scale is in units of the rms noise. Axis units are in pixels, with 1 pixel=0.332 arcsec.[]{data-label="Rres"}](Rsersdisk_ressig.eps){width="9cm"}
Fig.\[Rres\] shows the $R$-band residual image obtained after subtraction of the best fitting Sersic+disk model. The general morphology of the galaxy clearly deviates from the one of a simple model. Excess light is visible $\sim$30 arcsec from the centre and is distributed in a ring-like shape. Furthermore, a faint broad shell is seen in the south part of the image. This is very likely to be the shell found by Malin & Carter (1983). The same shell is observed in the $V$ band. The residual obtained by subtracting the best-fitting Sersic model is similar to the one of Fig.\[Rres\]. The main difference is that the ring-like feature is much brighter and broader, especially along the optical major axis; on the other hand, the fit is slightly better in the region outside the ring.
Discussion
==========
The results described in the last section reveal the complexity of IC 4200. The galaxy hosts a huge reservoir of gas spread over a very extended, warped disk. This gas seems to be triggering important processes in the stellar body: optical spectroscopy suggests recent star formation episodes and reveals ubiquitous LINER-like activity, and 1.4 GHz continuum excess is observed. Furthermore, ionised gas and stars are kinematically decoupled and a disk is possibly present in the inner part of the galaxy. Finally, a stellar ring and shells are detected in the outer regions. In this section we attempt to derive a global picture of the nature and recent history of the galaxy that can explain all of the observed properties at the same time.
Origin of the disk: merger or IGM accretion? {#discussiongas}
---------------------------------------------
The morphology and velocity structure can help placing a constraint on the age of the gas disk. The kinematics is that of a fairly regularly rotating, warped disk. This regularity suggests that the gas had time to complete at least 1-2 orbits all the way to the outer regions, where the rotational period is $\sim$1 Gyr, obtained using the model rotational velocity of 320 km/s at a radius of 60 kpc. In this time the gas can complete twice as many orbits in the inner 30 kpc, justifying the regular and symmetric appearance of the isophotes within the inner $\sim$3 beams. Therefore we conclude that the disk has been in place for at least 1-2 Gyr. No major encounters can have occurred more recently than this time.
IC 4200 contains 1.7 times more atomic gas than the Milky Way although its $B$-luminosity is only twice that of our galaxy and it belongs to an intrinsically gas-poor morphological type [^11]. It is natural to ask where this exceptionally large amount of gas comes from. Three scenarios are possible: merger origin, accretion from a satellite galaxy, or accretion from the IGM.
A merger seems a plausible explanation. Barnes (2002, hereafter B02) showed that up to 60% of the total gas of two similarly-sized merging galaxies can survive to form a very extended disk around the merger remnant. The gas disk results from re-accretion of gas that conserved its angular momentum, forming the tidal bridges and tails observed in many merging systems, and later settled into an extended but dilute rotationally supported disk. Only a fraction of the inital gas is available for this process, while the remaining part loses its angular momentum because of shocks and gravitational torques, falling toward the centre of the remnant and there triggering star formation and probably nuclear activity.
We investigate whether it is possible to form IC 4200 within the framework of B02 simulations. The first step is to constrain the properties of the progenitor galaxies. We derive the stellar mass of IC 4200 by estimating the stellar mass-to-light ratio in $V$ band and applying the result to the observed $V$ magnitude. $M/L$=1.8 was obtained by comparison of the $V-R$ colour of IC 4200 ($V-R$=0.58) with BC03 SSP models of metallicity around solar and higher. This value implies a stellar mass of 1.1$\times$10$^{11}$[$M_{\odot}$]{}, 1.9 times the stellar mass of the Milky Way [^12]. Furthermore, assuming a 60% efficency in gas disk formation, the mass of IC 4200 (8.5$\times{10^9}$[$M_{\odot}$]{}) is explained if 1.4$\times{10^{10}}$ [$M_{\odot}$]{} of atomic gas are available during the merger – 3 times more than the mass of the Milky Way. In conclusion, to build the stellar and mass of IC 4200 via an equal-mass merger, two galaxies each with 90% of the stellar mass of the Milky Way and 1.5 times its atomic gas content are needed.
According to the simulations of B02, such systems would indeed be able to produce a very extended disk: scaling from his results, two Milky Way disk galaxies (containing $5\times{10^9}$ [$M_{\odot}$]{} of each in the simulations) could merge and produce a disk extended out to 30 kpc from the centre of the remnant. This result is obtained inspecting the remnant $\sim$1 Gyr after the first encounter; however, as the time passes, more gas is re-accreted and settles into the outer parts of the disk, making it grow from the inside out and hence increasing its size. Combining this fact with the slightly higher initial gas mass required by our observations (7 rather than 5$\times{10^9}$ [$M_{\odot}$]{} of atomic gas in each of the progenitors), it is possible to produce a disk of the mass and size of the one observed. Furthermore, in the simulations of B02 the gas of the progenitors follows the same distribution as the disk stars. In real spiral galaxies the gas reaches larger radii than stars do; including this effect would produce even larger disks around the merger remnants.
The warp of the disk can also be accommodated in the merger picture . The returning gas comes from all directions with different angular momenta and it is therefore not surprising that it does not settle into a plane. The simulations of B02 confirm that warped disks are a very common result. Our data are therefore in a good agreement with a merger scenario for the formation of IC 4200 and would date the merger back to *at least* 1-2 Gyr ago.
Are the radio data consistent with other scenarios? Accretion from another galaxy is not very likely: any accretion process capable of stripping 8.5$\times{10^9}$ [$M_{\odot}$]{} from a passing galaxy would probably result in a major interaction. IC 4200-A itself is unlikely to be the source of all of the gas now around IC 4200; this would require that 90% of its gas was stripped off while leaving IC 4200-A itself unperturbed. Furthermore, IC 4200-A has an -mass typical of its morphological type and luminosity (derived from the Tully-Fisher relation using the observed rotation of its gas disk), so it would have been exceptionally -rich if all the gas now around IC 4200 came from it.
Accretion from the IGM is another possible way of forming gas structures around galaxies. Keres et al. (2005; hereafter KKWD05) showed that halos can partially accrete gas from the IGM via a cold mode that does not shock heat the gas to the virial temperature of the halo but leaves it constantly below 10$^5$ K. Gas accreted via the conventional hot mode could not cool quickly enough at such low densities, and therefore could not form dilute and extended atomic gas structures like the one described in this paper. On the contrary, cold accretion would have a better chance to provide the atomic gas necessary to explain the existence of these systems.
The results of KKWD05 place important constraints on the amount of cold accretion as a function of the halo mass. We use this estimate to understand whether cold accretion can build the disk observed around IC 4200. If the warped disk model is a fair representation of the gas structure around IC 4200, the velocity curve implies a halo mass $M_{halo}=Rv^2/G=1.7\times{10^{12}}$ [$M_{\odot}$]{} within 60 kpc. For such a halo the fraction of gas accreted via the cold mode is quite small, going from 30% of the accreted gas at $z$=3 to less than 5% at $z$=0 (see Fig.6 in KKWD05). However, accretion proceeds at high rates in the young Universe, making it possible to accumulate significant masses of gas at T$<10^5$ K in an early phase. Using the results of KKWD05 for a halo of 10$^{12}$ [$M_{\odot}$]{}, the average mass of gas accreted via cold mode around the central galaxy is of $1.2\times{10^{10}}$ [$M_{\odot}$]{}. We obtain this result on the basis of Fig.20 in KKWD05, where the average gas accretion rate onto the central galaxy of a halo is shown for different halo masses as a function of redshift. We scale the accretion rate at each $z$ by the fraction of gas accreted via cold mode at that redshift (this fraction is obtained from Fig.6 of KKWD05). We then integrate the total mass accreted via cold mode, using the transformation $t=9.3$ Gyr/(1+z)$^{1.5}$. The result of $1.2\times{10^{10}}$ [$M_{\odot}$]{} is close to our mass estimate for the gas disk ($8.5\times{10^{9}}$ [$M_{\odot}$]{}) [^13].
The gas accreted via cold mode has a temperature just below $10^5$ K and therefore is not in the form of . However, at least a fraction of it can cool to the atomic state. Assuming a $<$10%-in-mass enrichment from the stellar processes within the galaxy (where the metallicity is as high as 5 times solar), the accreted gas can reach $Z\sim0.1$ $Z_{\odot}$. At this metallicity and assuming a density of $\sim$10$^{-2}$ cm$^{-3}$ and no significant heating (as in IC 4200, especially at radii of tens of kpc), the cooling time is $\sim$1 Gyr. As a consequence, a significant fraction of the mass accreted via cold mode can cool to 100 K and be observed in the form of atomic gas. On the whole, considering also that KKWD05 conclusions are statistical and do not necessarily hold for any give halo, accretion from the IGM seems to be another possible way to form the observed system.
According to our calculations based on KKWD05 simulations, the largest fraction of accreted mass accumulates at higher redshifts. Assuming that accretion starts at $z$=4, 90% of the total mass accreted via cold mode is in place at $z$=1 (this number may vary slightly depending on the environment). This implies that no substantial accretion occurred in the last $\sim$6 Gyr. If the gas around IC 4200 comes from IGM accretion, most of it has been in place for at least 6 Gyr. This is an important conclusion that has to be compared to other observational results.
Clues from the stellar properties
---------------------------------
Optical observations can help to further distinguish between the merger and the IGM accretion scenarios. One result of our observations is that ionised gas and stars have different kinematics. The gas rotates in a way that may be seen as a continuation of the velocity field, even though the spatial resolution of radio and optical data are completely different. The independent motion of gas and stars is consistent with a merger hypothesis. If hydrodynamic interactions take place gas and stars (which interact only gravitationally) can follow different behaviours and therefore end up having different kinematics. B02 simulations confirm that inner gas disks rotationally supported and decoupled from the stellar remnant can be formed in some mergers. However, the same could be said of gas accreted from the IGM, as the gas would fall in and settle in the potential independently of the underlying stellar motion.
Our results on the nature of the gas ionisation do not seem to depend on the formation scenario. LINER emission lines ratios and the 1.4 GHz radio continuum excess strongly suggest that, perhaps along with star formation, some other activity is taking place. However, both the merger and the IGM-accretion formation scenarios could produce this situation, the relevant fact being only that the gas is present within the optical body of the galaxy, and that it might be fuelling nuclear activity and star formation, as well as producing shocks.
Two crucial results effective in understanding the formation of IC 4200 are the age of the stellar populations and the detection of shells. Star formation has occurred in the centre of the galaxy within the past 1.5 Gyr, as demonstrated by the line-strength indices presented in Sect.\[spopandfrac\]. Assuming a two-populations model, with a young stellar population on top of a much older one, we can estimate the mass fraction of the star-formation burst that formed the young population as a function of its age.
The equivalent width EW of a given spectral feature measured within a band of width $w$ can be written as: $$\rm EW=\it w\times \left(1-\frac{F_1+F_2}{F_{\rm c1}+F_{\rm c2}}\right),$$ where $F_i$ ($i$=1,2) is the total flux coming from population $i$ and $F_{\rm c \it i}$ is the flux coming from its continuum, which is assumed to be a linear function of wavelength. Writing the flux $F_i$ of each populations in terms of its mass $M_i$ and its flux per unit mass $f_i$, and dividing and multiplying the right-hand side of the equation by the mass $M_1$ of population 1 one obtains: $$\rm EW=\it w\times \left(1-\frac{f_1+\mu_{21}f_2}{f_{\rm c1}+\mu_{21}f_{\rm c2}}\right),$$ where $\mu_{21}$ is the fraction $M_2$/$M_1$. From this expression $\mu_{21}$ can be derived as a function of the measured EW, of $w$ and of the modelled quantities $f_1$, $f_2$, EW$_1$ and EW$_2$ (EW$_1$ and EW$_2$ are the equivalent widths that would be measured from the individual spectra of population 1 and 2 respectively): $$\mu_{21}=-\frac{f_1}{f_2}\times\frac{\rm EW_1-EW}{\rm EW_2-EW}\times\frac{w-\rm EW_2}{w-\rm EW_1}.$$ From this $\mu_{2}$=$M_2$/($M_1$+$M_2$) can be easily derived.
To study the case of IC 4200 we measured the H$\beta$ line-strength index from two spectra of radial aperture equal to the $V$-band effective radius, one along the major axis and one along the minor axis (for the latter we scaled $r_e$ to the axis ratio). The index was measured in the same way as described in Sect.\[indices\]. The average value of H$\beta$ obtained from the two axes is 2.13 Å. The bandwidth $w$ is of 28.75 Å (Table \[ind\_def\]). The other quantities that appear in the last equation were taken from the models available at Guy Worthey’s web page http://astro.wsu.edu/worthey/.
We assume that population 1 and 2 are respectively 15 and 1.5 Gyr old and that both have solar metallicities. The age of the young stellar populations is the maximum allowed by the SSP-equivalent age in the centre of the galaxy (see Fig.\[gridpoints\]). As a result we obtain $\mu_{2}$=4.1%. Changing the assumption on the age of the old stellar populations will not change these fractions much while increasing the metallicity to \[Fe/H\]=0.25 brings $\mu_2$ up to 15%. These values of $\mu_2$ are derived using 1 $r_e$ radial aperture and therefore they are fractions of the mass enclosed in one effective radius which by definition, assuming constant M/L throughout the galaxy, contains half of its stellar mass. Our result therefore implies that if the observed line strengths are caused by a burst of star formation occurred 1.5 Gyr ago, between 2.3 and 8.3$\times{10^{9}}$ [$M_{\odot}$]{} of gas were turned into stars during this burst.
Interestingly, the mass of the star formation burst is of the order of the mass of the disk. This is an extremely important consistency argument in favour of the merger scenario. In that picture roughly the 60% of the gas of the progenitors had to be conserved to form the observed disk. To have a correct balance of mass, it is necessary that the remaining 40% is observable in IC 4200. Our stellar populations analysis is consistent with this gas mass being turned into stars during the merger. The SSP-equivalent age in the center of the galaxy provides an upper limit on the age of most massive burst of star formation induced by the merger. Such burst could occur at the time of the first encounter but also 1-2 Gyr later, during the second encounter, depending on the impact parameters (e.g. Kapferer et al. 2005). Therefore, the upper limit of 1.5 Gyr on the age of the burst of star formation provides an upper limit on the beginning of the merging process of $\sim$3 Gyr. Matching this indication with the lower limit on the age of the disk (1-2 Gyr), we can conclude that if IC 4200 formed via a major merger, this must have occurred between 1 and 3 Gyr ago.
Along with the SSP-equivalent age, our analysis of line-strength indices provide estimate for the SSP-equivalent abundance ratio $\rm [E/Fe]$. We find $\rm [E/Fe]\sim$0.30 in the centre of the galaxy. Within the picture of the two-populations model, the SSP-equivalent $\rm [E/Fe]$ is determined mostly by the old and more massive population. This is a consequence of $\rm [E/Fe]$ being measured primarily on the basis of the Mgb/$\rm<Fe>$, to which cooler stars (i.e., RGB stars) contribute the most. For example, given an old population of $\rm [E/Fe]\sim$0.4 and a population 1.5 Gyr old with $\rm[E/Fe]\sim$0.1 and ten times less massive, the SSP-equivalent $\rm[E/Fe]$ is slightly above 0.3 (this result was obtained using W94 models). Therefore, our result suggests that the less massive young population lies on top of an older population formed in a short massive burst of star formation and hence characterised by high $\rm [E/Fe]$.
Can IGM accretion match the requirements imposed by stellar populations results? As mentioned, IGM accretion implies that the disk formed more than 6 Gyr ago. The accreted atomic gas might have fuelled star formation at that time, but it is not clear why such a massive burst of residual star formation (that would have consumed roughly half of the mass accreted as cold gas) should have occurred so much later, in the last 1.5 Gyr. IGM accretion is therefore hard to reconcile with the stellar population properties of IC 4200.
Some of the morphological features (the inner disk and the ring-like feature) can be explained by both merger and IGM accretion. However, the detection of shells in the images of the galaxy points toward a merger origin. It is believed that shells can form as a result of accretion of smaller satellites by bigger galaxies (e.g. Quinn 1984) or by equal mass mergers (Hernquist & Spergel 1992). On the contrary, IGM accretion alone cannot explain shell formation.
In conclusion, the only formation process that can by itself explain all the observed properties is a major merger. According to the observational evidence, a merger of two galaxies each with 90% of the Milky Way stellar mass and containing 1.5 times its gas mass occurred between 1 and 3 Gyr ago. During the merger roughly half of the atomic gas of the progenitors retained its angular momentum and was later re-accreted, settling into a 90-deg warped disk extended out to 60 kpc from the center of the galaxy. The loss of angular momentum drove the remaining gas toward the centre of the newly formed galaxy, where it triggered star formation. This resulted also in the formation of a decoupled central component visible as a spiral-like dust pattern. The dynamically violent process produced a merger remnant with morphological disturbances that are still visible in the outer regions. Finally, ionised gas is present in the stellar body and, as a result of the merger, is kinematically decoupled from the stars. It is also fuelling some activity, possibly a central AGN or shocks, and maybe star formation.
We note that the non-detection of IC 4200 in the X-ray supports our conclusion on the origin of the galaxy. As suggested by Sansom et al. (2000), it probably takes several gigayears for hot gas halos to build up after a merger or a strong interaction. They reached this conclusion on the basis of the anti-correlation between X-ray excess and morphological fine structure. Given its $L_B$ (and following the mean relation between $L_B$ and $L_X$ of Sansom et al. 2000), IC 4200 should have been detected by X-ray surveys (e.g., $ROSAT$). The same is true if we use our estimate of $L_X$ based on the accretion of hot gas from the IGM (Sec. \[discussiongas\]). The fact that IC 4200 has not been detected in the X-ray is therefore consistent with a major merger origin of the galaxy.
The fate of IC 4200
-------------------
A final interesting question is what the future evolution of IC 4200 will be. Given the amount of gas and its configuration, it is very unlikely that it will ever become a more typical gas-poor member of its morphological type. The gas seems to be in a quite regular and stable configuration, having been in place already more than 1 Gyr. If, on the other hand, the gas density increases just enough to start star formation, it is possible that IC 4200 will grow a stellar disk, becoming an early-type spiral. In this case it would be an exceptionally bulge-dominated galaxy, with M$_{bulge}$/M$_{disk}\sim$15 or more and a total stellar mass of $\sim$1.2$\times$10$^{11}$ [$M_{\odot}$]{}. The mass-to-light ratio of the galaxy is currently $\sim$30 in the $V$ band and out to 60 kpc and would maintain roughly the same value in case of the formation of a stellar disk.
Conclusions
===========
[c|c c c]{} & & Satellite & IGM\
& Merger & Accretion & Accretion\
& $\surd$ & $\times$ & $\surd$\
stellar populations & $\surd$ & $\times$ & $\times$\
shells & $\surd$ & $\surd$ & $\times$\
We have presented the result of radio (21-cm-line emission and 1.4 GHz continuum) and optical (long-slit spectroscopy and imaging) observations of the early-type galaxy IC 4200. A total of $8.54\times{10^9}$ [$M_{\odot}$]{} of is detected around this object and $7.3\times{10^8}$ [$M_{\odot}$]{} around its companion galaxy IC 4200-A. The gas around IC 4200 can be modelled as a warped disk and its structure suggests that it has been in place for at least 1-2 Gyr. The stellar populations of IC 4200 have a 1.5 Gyr SSP-equivalent age. Ionised gas is present across the whole stellar body and is kinematically decoupled from the stars. This gas shows a spectrum characteristic of LINERs across the inner $r_e$/3. The activity is confirmed by a 1.4 GHz continuum excess compared to the 60 $\mu$m emission. An optical shell have been detected around the galaxy next to a ring-like feature of radius 30 arcsec.
Table \[conclusion\] summarises the conclusions we derive from these observational results. The merger scenario is the only one that can explain at once all our results, while IGM or satellite accretion fail to do so. From data we conclude that IC 4200 formation event must have occurred more than 1-2 Gyr ago. From the SSP-equivalent age of the stellar populations we derive that the formation of the galaxy has occurred via a merger less than $\sim$3 Gyr ago (taking into account a delayed burst of star formation). Matching these two results we can claim that IC 4200 formed via a merger between 1 and 3 Gyr ago. Within this interpretation of the data, two Milky-Way-like progenitors are required in order to produced the observed stellar and -gas masses.
Table \[conclusion\] stresses the power of our approach in determining the past history of galaxies. Only the combination of 21-cm data, optical imaging and stellar populations analysis allowed us to obtain a robust result. Following papers will present a similar study of a larger number of galaxies of our sample.
The authors wish to thank the anonymous referee for the constructive comments. We would like to thank Daniel Kelson for his help during the reduction of the long-slit spectroscopy data. We would also like to thank Guy Worthey for providing the material necessary for the calibration of the line-strength indices onto the Lick/IDS system. Finally, we thank Jacqueline van Gorkom and Renzo Sancisi for stimulating discussions on the origin of gas in early-type galaxies. This research has made use of HyperLeda and of the NASA/IPAC Extragalactic Database (NED) which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration. This work is based on observation with the Australia Telescope Compact Array, which is operated by the CSIRO Australia Telescope National Facility, and on observations collected at the European Southern Observatory, La Silla, Chile.
Calibration of the line-strength indices to the Lick/IDS system {#Lick/IDScal}
===============================================================
[c c c c]{} Index & Slope ($m$) & Intercept ($q$) & RMS\
H$\beta$ & 0.97 & 0.12 & 0.27\
Mg$b$ & 0.88 & 0.31 & 0.41\
Fe5270 & 0.91 & 0.16 & 0.32\
Fe5335 & 0.94 & 0.23 & 0.34\
H$\gamma_{\rm A}$ & 0.95 & –0.30 & 0.80\
H$\delta_{\rm A}$ & 0.94 & –0.17 & 0.85\
H$\gamma_{\rm F}$ & 0.96 & –0.15 & 0.46\
H$\delta_{\rm F}$ & 0.95 & 0.02 & 0.49\
The calibrated index is $i_{\rm Lick/IDS}=m\times{i_{\rm MILES}}+q$; the RMS is calculated on the basis of $\sim$240 stellar spectra.
[c c c c]{} Index & Slope ($m$) & Intercept ($q$) & RMS\
H$\beta$ & 1.01 & 0.13 & 0.25\
Mg$b$ & 0.89 & 0.32 & 0.35\
Fe5270 & 0.97 & 0.12 & 0.29\
Fe5335 & 1.01 & 0.08 & 0.32\
H$\gamma_{\rm A}$ & 1.01 & –0.15 & 0.71\
H$\delta_{\rm A}$ & 0.98 & –0.02 & 0.83\
H$\gamma_{\rm F}$ & 1.01 & –0.10 & 0.43\
H$\delta_{\rm F}$ & 0.98 & 0.14 & 0.51\
The calibrated index is $i_{\rm Lick/IDS}=m\times{i_{\rm INDO-US}}+q$; the RMS is calculated on the basis of $\sim$150 stellar spectra.
Our analysis of the stellar populations of IC 4200 relies on the comparison of its spectral indices with the ones predicted by the stellar populations models of W94. These models are built to match age and metallicity of the stars included in the Lick/IDS library (Worthey et al. 1994), which have been observed in well defined instrumental conditions and are not flux-calibrated. Therefore, any comparison of the spectral indices of IC 4200 to these models makes sense only if the indices themselves are onto the Lick/IDS system. A practical way to satisfy this requirement is to observe with the same instrumental setup used for IC 4200 a sample of stars that already have indices on the Lick/IDS system. We can then measure the spectral indices of these stars from our spectra and calculate the transformation necessary to make them match their Lick/IDS-reference values. The same transformation will then be applied to the indices measured from the spectrum of IC 4200. Once this is done, the spectral indices of IC 4200 can be compared to W94 models.
![Calibration to the Lick/IDS system. For each index the points in the plane ($i_{\rm ref},i_{\rm meas}$) are showed together with the best fitting straight line. Each point represents one of our Lick/IDS stars. The fitting was performed weighting points according to their errors in both coordinates.[]{data-label="lickcalib"}](indices_cal.eps){width="8.75cm"}
The stars observed in order to perform the calibration are listed in Table \[LICKlist\]. Guy Worthey kindly provided us the reference values of the spectral indices for these stars. However as indicated in Table \[LICKlist\], only two of them belong to the Lick/IDS library and therefore have indices already on the Lick/IDS system. Indices measured from spectra in either the MILES (Sánchez-Blázquez et al. 2006, in preparation) or the INDO-US library (Valdes et al. 2004) are not calibrated and had to be brought to the Lick/IDS system. We did this by means of a set of transformations also kindly provided by Guy Worthey. For each index, a linear transformation brings from the “MILES value” to the Lick/IDS system and another one brings from the “INDO-US value” to the Lick/IDS system. The coefficient of these transformations for the indices used in this paper can be found in Table \[M2L\] and Table \[I2L\].
![Correction of the line-strength indices for the effect of the stellar velocity dispersion. The ratio $f$ between the measured and the corrected value is plotted versus the velocity dispersion. Each point correspond to one of the S$_{80}$’s used to study the stellar populations in IC 4200.[]{data-label="fplots"}](sigmacorr_tot.eps){width="8.75cm"}
Once all the stars have Lick/IDS-reference indices values, the calibration of the generic index $i$ was done by:
- broadening our stellar spectra to match the Lick/IDS resolution
- measuring $i$ for each of our Lick/IDS stars obtaining the values $i_{j,\rm meas}$, where $j$=$1,...,N_{\rm stars}$
- comparing $i_{j,\rm meas}$ to the respective reference values $i_{j,\rm ref}$; by fitting a straight line to the N$_{\rm stars}$ points in the plane ($i_{\rm ref},i_{\rm meas}$) we could obtain for each index of interest the linear relation between the measured and the reference value.
Fig.\[lickcalib\] shows the planes ($i_{\rm ref},i_{\rm meas}$) with the data points and the linear fit. The coefficient of the linear relation $i_{\rm meas}=m\cdot i_{\rm ref}+q$ are given in Table \[Lick\_cal\].
Velocity dispersion corrections of the line-strength indices {#vdispcorr}
============================================================
Fig.\[fplots\] shows the factor $f$ used to correct line-strength indices for the stellar velocity dispersion as a function of the velocity dispersion for the indices used in this paper. $f$ is defined so that the corrected line-strength index is $i_{0,k}=i_k/f$ where $i_k$ is the measured value. For more details see Sect.\[indices\].
Tables {#eltables}
======
Tables C.1 and C.2 are available electronically at the CDS. Table C.1 contains informations on the stellar and ionised-gas kinematics along the $B_{30}$’s. Column 1 specifies the axis (either major or minor axis), Columns 2 and 3 give the lower and upper radial coordinate of the bins, Columns 4 and 5 list the stellar line-of-sight velocity and its error, Columns 5 and 6 give the stellar line-of-sight velocity dispersion ($\sigma$) and its error, Columns 7 and 8 provide the central wavelength of H$\alpha$ emission line and its error, and finally Columns 9 and 10 give the FWHM of H$\alpha$ emission line and its error.
Table C.2 lists the value of all the Lick/IDS indices fully corrected and calibrated. In this table Column 1 specifies the axis (either major or minor axis), Columns 2 and 3 give the lower and upper radial coordinate of the bins from which the indices were measured, and the following 50 columns give the value of the 25 Lick/IDS indices, each followed by its error.
Bahcall, J. N. & Soneira, R.M. 1980, ApJS, 44, 73
Balcells, M., van Gorkom, J. H., Sancisi, R., del Burgo, C. 2001, AJ, 122, 1758
Barnes, J. E. 2002, MNRAS 333, 481
Binney, J. 1977, ApJ, 215, 483
Briggs D.S. 1995, Ph.D. Thesis, New Mexico Institute of Mining and Technology, New Mexico
Bruzual A. G. & Charlot, S. 2003, MNRAS, 344, 1000
Carter, D., Prieur, J. L., Wilkinson, A., Sparks, W. B., Malin, D. F. 1988, MNRAS, 235, 813
de Vaucouleurs, G., de Vaucouleurs, A., Corwin, H. G., Jr., Buta, R. J., Paturel, G. & Fouqué, P. 1991, Third Reference Catalogue of Bright Galaxies (New York: Springer)
de Jong, R. S., Davies, R. L. 1997, MNRAS, 285, L1
Dopita M. A. & Sutherland, R. S. 1995, ApJ, 455, 468
Emsellem, E., Davies, R., McDermid, R., Kuntschner, H., Peletier, R., Bacon, R., Bureau, M., Cappellari, M., Copin, Y., Miller, B., Verolme, E., de Zeeuw, T. 2002, in ASP Conf.. Ser. 282, Galaxies: The Third Dimension, 189
Falcón-Barroso, J., Sarzi, M., Bacon, R., Bureau, M., Cappellari, M., Davies, R. L., Emsellem, E., Fathi, K., Krajnovic, D., Kuntschner, H., McDermid, R. M., Peletier, R. F., de Zeeuw, P. T. 2005, astro-ph/0508199
Fardal, M. A., Katz, N., Gardner, J.P., Hernquist, L., Weinberg, D. H. & Davé, R. 2001, ApJ, 562, 605
Henderson, A. P., Jackson, P. D., Kerr, F. J. 1982, ApJ, 263, 116
Hernquist, L. & Spergel, D. N. 1992, ApJ, 399, L117
Hibbard, J. E. & Sansom, A. E. 2003, AJ, 125, 667
Kapferer, W., Knapp, A., Schindler, S., Kimeswenger, S., van Kampen, E., 2005, A&A, 438, 87
Kauffmann, G., Heckman, T. M., Tremonti, C., Brinchmann, J., Charlot, S., White, S. D. M., Ridgway, S. E., Brinkmann, J., Fukugita, M., Hall, P. B., Ivezíc, Z., Richards, G. T., Schneider, D. P. 2003, MNRAS, 346, 1055
Kelson, D. D. 2003, PASP, 115, 688
Kelson, D. D., Illingworth G. D., van Dokkum, P. G., Franx, M. 2000, ApJ, 531, 159
Kennicutt R. C., Jr., 1989 ApJ, 344, 685
Kennicutt R. C., Jr., 1998 ARA&A 36, 189
Keres, D., Katz, N., Weinberg, D. H., Dave, R. 2005, MNRAS, 363, 2
Kewley, L. J., Geller, M. J., Jansen, R. A., Dopita, M. A. 2002, AJ, 124, 3135
Malin, D. F. & Carter, D. 1983, ApJ, 274, 534
Meyer, M. J. 2004, MNRAS, 350, 1195
Mihos, J. C. 1995, ApJL, 438, L75
Morganti, R., Sadler, E. M., Oosterloo, T., Pizzella, A. & Bertola, F. 1997, AJ, 1113, 937
Oosterloo, T. A., Sadler, E. M., Morganti, R., van der Hulst, J. M. 2006, in preparation
Oosterloo, T. A., Morganti, R., Sadler, E. M., Vergani, D. & Caldwell, N. 2002, AJ, 123, 729
Peng, C. Y., Ho, L. C., Impey, C. D., Rix, H. -W. 2002, AJ, 124, 266
Phillips, M. M., Jenkins, C. R., Dopita, M. A., Sadler, E. M., Binette, L. 1986, AJ, 91, 1066
Quinn, P. J. 1983, ApJ, 279, 596
Sadler, E. M., Oosterloo, T. & Morganti, R. 2001, in ASP Conf. Ser. 230: Galaxy Disks and Disk Galaxies, 285
Sadler, E. M., Oosterloo, T. & Morganti, R. 2002, in ASP Conf. Ser. 273: The Dynamics, Structure & History of Galaxies: A Workshop in Honour of Professor Ken Freeman, 215
Sánchez-Blázquez et al. 2006, in preparation
Sansom, A. E., Hibbard, J. E., Schweizer, F. 2000, AJ, 120, 1946
Sarzi, M., Falcón-Barroso, J., Davies, R. L., Bacon, R., Bureau, M., Cappellari, M., de Zeeuw, P. T., Emsellem, E., Fathi, K., Krajnovic, D., Kuntschner, H., McDermid, R. M., Peletier 2005, astro-ph/0511307
Sault, R. J., Teuben, P. J. & Wright, M. C. H 1995, in ASP Conf. Ser. 77, Astronomical Data Analysis Software and Systems IV, ed. R. Shaw, H. E. Payne & E. Hayes (San Francisco: ASP), 433
Schaye, J. 2004, ApJ, 609, 667
Schiminovich, D., van Gorkom, J. H., van der Hulst, J. M. & Kasow, S. 1994, ApJL, 423, L101
Schiminovich, D., van Gorkom, J. H., van der Hulst, J. M. & Malin, D. F. 1995, ApJL, 444, L77
Schiminovich, D., van Gorkom, J., van der Hulst, T., Oosterloo, T. & Wilkinson, A. 1997, in ASP Conf. Ser. 116: The Nature of Elliptical Galaxies; 2nd Stromlo Symposium, 362
Schweizer, F., Seitzer, P., Faber, S. M., Burstein, D., Dalle Ore, C. M. & Gonz[á]{}lez, J. J. 1990, ApJL, 364, L33
Schweizer, F. & Seitzer, P. 1992, AJ, 104, 1039
Thomas, D., Maraston, C. & Bender, R. 2003, MNRAS, 339, 897
Thomas, D., Maraston, C. & Korn, A. 2004, MNRAS, 351, L19
Thronson, H. A., Jr., Bally, J., Hacking, P. 1989, AJ, 97, 363
Trager, S. C., Faber, S. M., Worthey, G. & Gonz[á]{}lez, J. J. 2000, AJ, 120, 165
Trager, S. C., Worthey, G., Faber, S. M., Dressler, A. 2005, MNRAS, 362, 2
Valdes, F., Gupta, R., Rose, J. A., Singh, H. P., Bell, D. J. 2004, ApJS, 152, 251
van den Bergh, S. 1988, ComAp, 12, 131
van Gorkom, J. & Schiminovich, D. 1997, in ASP Conf. Ser. 116: The Nature of Elliptical Galaxies; 2nd Stromlo Symposium, 310
Véron-Cetty, M.-P., Woltjer, L., Ekers, R.D. & Staveley-Smith, L. 1995, A&A, 297, L79
Winkler, H. 1997, MNRAS, 292, 273
Worthey, G. 1994, ApJS, 95, 107
Worthey, G., Faber, S. M., Gonz[á]{}lez, J. J., Burstein, D. 1994, ApJS, 94, 687
Worthey, G. & Ottaviani, D. L. 1997, ApJS, 111, 377
Yun M. S., Naveen, A. R. & Condon, J. J. 2001, ApJ, 554, 803
[^1]: $H_0$ is assumed to be 70 km s$^{-1}$ Mpc$^{-1}$ throughout this paper.
[^2]: This is the systemic velocity of the gas model that reproduces our 21-cm wavelength observations; the model is described in Sect.\[higasres\].
[^3]: Parkes All Sky Survey.
[^4]: 2 Micron All Sky Survey.
[^5]: Digitised Sky Survey.
[^6]: As mentioned in Sect.\[basicprop\], 2MASX J13090029–5153544 appears only in the catalogue of the 2MASS survey and is the closest object to IC 4200.
[^7]: Indeed, the $V$-magnitude measured within a 20-arcsec-diameter aperture is in agreement with the result of Winkler (1997).
[^8]: $M_{\rm {\ion{H}{i}},MW}=5\times{10^{9}}$ [$M_{\odot}$]{} (Henderson et al 1982).
[^9]: These are the $V$-band half-light radii along major and minor axis, obtained as a part of our analysis of the morphology of IC 4200 described in Sect.\[imag\_res\].
[^10]: $<$MgFe$>$ is defined as $<$MgFe$>$=$\sqrt{ \rm Mgb\times{<Fe>}}$, where $<$Fe$>$= =(Fe5270+Fe5335)/2
[^11]: The Milky Way has $L_B=2.3\times{10^{10}} L_{\odot}$ (Van den Bergh 1988).
[^12]: The stellar mass of the Milky Way is $6\times{10^{10}}$ [$M_{\odot}$]{} (Bahcall & Soneira 1980).
[^13]: Repeating this calculation without distinguishing between cold and hot accretion, we find that the total gaseous mass accreted is 7.5$\times{10^{10}}$ [$M_{\odot}$]{}. Assuming that the hot gas is at the virial temperature of the halo ($\sim3.7\times{10^6}$ K) and that it is distributed within 60 kpc from the center, we estimate that its X-ray luminosity is $\sim5\times{10^{41}}$ ergs/s.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We show that for two quivers without oriented cycles related by a BGP reflection, the posets of their cluster tilting objects are related by a simple combinatorial construction, which we call a flip-flop.
We deduce that the posets of cluster tilting objects of derived equivalent path algebras of quivers without oriented cycles are universally derived equivalent. In particular, all Cambrian lattices corresponding to the various orientations of the same Dynkin diagram are universally derived equivalent.
address: 'Einstein Institute of Mathematics, The Hebrew University of Jerusalem, Jerusalem 91904, Israel'
author:
- Sefi Ladkani
title: Universal derived equivalences of posets of cluster tilting objects
---
Introduction
============
In this note we investigate the combinatorial relations between the posets of cluster tilting objects of derived equivalent path algebras, continuing our work [@Ladkani07t] on the posets of tilting modules of such algebras.
Throughout this note, we fix an algebraically closed field $k$. Let $Q$ be a finite quiver without oriented cycles and let $\operatorname{rep}Q$ denote the category of finite dimensional representations of $Q$ over $k$. The associated cluster category ${\mathcal{C}}_Q$ was introduced in [@BMRRT06] (and in [@CCS06] for the $A_n$ case) as a representation theoretic approach to the cluster algebras introduced and studied by Fomin and Zelevinsky [@FominZelevinsky02]. It is defined as the orbit category [@Keller05] of the bounded derived category ${\mathcal{D}}^b(Q)$ of $\operatorname{rep}Q$ by the functor $S \cdot [-2]$ where $S : {\mathcal{D}}^b(Q) \to
{\mathcal{D}}^b(Q)$ is the Serre functor and $[1]$ is the suspension. The indecomposables of ${\mathcal{C}}_Q$ can be represented by the indecomposables of ${\mathcal{D}}^b(Q)$ in the fundamental domain of $S \cdot [-2]$, hence $\operatorname{ind}{\mathcal{C}}_Q = \operatorname{ind}\operatorname{rep}Q \cup \left\{ P_y[1] \,:\, y \in Q \right\}$ where $P_y$ are the indecomposable projectives in $\operatorname{rep}Q$.
Cluster tilting theory was investigated in [@BMRRT06]. A basic object $T \in {\mathcal{C}}_Q$ is a *cluster tilting object* if $\operatorname{Ext}^1_{{\mathcal{C}}_Q}(T,T) = 0$ and $T$ is maximal with respect to this property, or equivalently, the number of indecomposable summands of $T$ equals the number of vertices of $Q$. If $T = M \oplus U$ is cluster tilting and $M$ is indecomposable, then there exist a unique indecomposable $M' \neq M$ such that $T' = M' \oplus U$ is cluster tilting. $T'$ is called the *mutation* of $T$ with respect to $M$.
Denote by ${\mathcal{T}}_{{\mathcal{C}}_Q}$ the set of all cluster tilting objects. In [@IngallsThomas06], a partial order on ${\mathcal{T}}_{{\mathcal{C}}_Q}$, extending the partial order on tilting modules introduced in [@RiedtmannSchofield91], is defined by $T \leq T'$ if $\operatorname{fac}T
\supseteq \operatorname{fac}T'$. Here, for $M \in \operatorname{rep}Q$, $\operatorname{fac}M$ denotes the full subcategory of $\operatorname{rep}Q$ consisting of all quotients of finite sums of copies of $M$, and for $T \in {\mathcal{T}}_{{\mathcal{C}}_Q}$, $\operatorname{fac}T = \operatorname{fac}{\widehat}{T}$ where ${\widehat}{T} \in \operatorname{rep}Q$ is the sum of all indecomposable summands of $T$ which are not shifted projectives.
As shown in [@IngallsThomas06], the map $T \mapsto \operatorname{fac}T$ induces an order preserving bijection between $({\mathcal{T}}_{{\mathcal{C}}_Q}, \leq)$ and the set of finitely generated torsion classes in $\operatorname{rep}Q$ ordered by reverse inclusion. Moreover, it is also shown that when $Q$ is Dynkin, $({\mathcal{T}}_{{\mathcal{C}}_Q}, \leq)$ is isomorphic to the corresponding Cambrian lattice defined in [@Reading06] as a certain lattice quotient of the weak order on the Coxeter group associated with $Q$.
For two partially ordered sets $(X, \leq_X)$, $(Y, \leq_Y)$ and an order preserving function $f : X \to Y$, define two partial orders $\leq^f_+$ and $\leq^f_-$ on the disjoint union $X \sqcup Y$ by keeping the original partial orders inside $X$ and $Y$ and setting $$\begin{aligned}
x \leq^f_{+} y \Longleftrightarrow f(x) \leq_Y y && y \leq^f_{-} x
\Longleftrightarrow y \leq_Y f(x)\end{aligned}$$ with no other additional order relations. We say that two posets $Z$ and $Z'$ are related via a *flip-flop* if there exist $X$, $Y$ and $f: X \to Y$ as above such that $Z \simeq (X \sqcup Y, \leq^f_{+})$ and $Z' \simeq (X \sqcup Y, \leq^f_{-})$.
Let $x$ be a sink of $Q$ and let $Q'$ be the quiver obtained from $Q$ by a BGP reflection [@BGP73] at $x$, that is, by reverting all the arrows ending at $x$. Our main result is the following.
\[t:clff\] The posets ${\mathcal{T}}_{{\mathcal{C}}_Q}$ and ${\mathcal{T}}_{{\mathcal{C}}_{Q'}}$ are related via a flip-flop.
We give a brief outline of the proof. Let ${\mathcal{T}}^x_{{\mathcal{C}}_Q}$ denote the subset of cluster tilting objects containing the simple projective $S_x$ at $x$ as direct summand. Given $T \in {\mathcal{T}}^x_{{\mathcal{C}}_Q}$, let $f(T)$ be the mutation of $T$ with respect to $S_x$. In Section \[sec:clsink\] we prove that the function $f : {\mathcal{T}}^x_{{\mathcal{C}}_Q}
\to {\mathcal{T}}_{{\mathcal{C}}_Q} \setminus {\mathcal{T}}^x_{{\mathcal{C}}_Q}$ is order preserving and moreover $$\label{e:ffp}
{\mathcal{T}}_{{\mathcal{C}}_Q} \simeq \bigl( {\mathcal{T}}^x_{{\mathcal{C}}_Q} \sqcup ({\mathcal{T}}_{{\mathcal{C}}_Q} \setminus
{\mathcal{T}}^x_{{\mathcal{C}}_Q}), \leq^f_{+} \bigr)$$
Similarly, let ${\mathcal{T}}^{x[1]}_{{\mathcal{C}}_{Q'}}$ be the subset of cluster tilting objects containing the shifted projective $P'_x[1]$ at $x$ as direct summand. Given $T \in {\mathcal{T}}^{x[1]}_{{\mathcal{C}}_{Q'}}$, let $g(T)$ be the mutation of $T$ with respect to $P'_x[1]$. In Section \[sec:clsource\] we prove that the function $g :
{\mathcal{T}}^{x[1]}_{{\mathcal{C}}_{Q'}} \to {\mathcal{T}}_{{\mathcal{C}}_{Q'}} \setminus
{\mathcal{T}}^{x[1]}_{{\mathcal{C}}_{Q'}}$ is order preserving and moreover $$\label{e:ffm}
{\mathcal{T}}_{{\mathcal{C}}_{Q'}} \simeq \bigl( {\mathcal{T}}^{x[1]}_{{\mathcal{C}}_{Q'}} \sqcup
({\mathcal{T}}_{{\mathcal{C}}_{Q'}} \setminus {\mathcal{T}}^{x[1]}_{{\mathcal{C}}_{Q'}}), \leq^g_{-} \bigr)$$
In Section \[sec:clBGP\] we relate the two isomorphisms given in and by considering, following [@Zhu07], the action of the BGP reflection functor on the cluster tilting objects. We prove the existence of the following commutative diagram with horizontal isomorphisms of posets $$\xymatrix{ {\mathcal{T}}^x_{{\mathcal{C}}_Q} \ar[r]^{\simeq} \ar[d]^{f} &
{\mathcal{T}}^{x[1]}_{{\mathcal{C}}_{Q'}} \ar[d]^{g} \\
{\mathcal{T}}_{{\mathcal{C}}_Q} \setminus {\mathcal{T}}^x_{{\mathcal{C}}_Q} \ar[r]^{\simeq} & {\mathcal{T}}_{{\mathcal{C}}_{Q'}}
\setminus {\mathcal{T}}^{x[1]}_{{\mathcal{C}}_{Q'}} }$$ from which we deduce Theorem \[t:clff\]. An example demonstrating the theorem and its proof is given in Section \[sec:clexample\].
In our previous work [@Ladkani07t], we have shown a result analogous to Theorem \[t:clff\] for the posets ${\mathcal{T}}_Q$ and ${\mathcal{T}}_{Q'}$ of tilting modules, following a similar strategy of proof. However, there are some important differences.
First, the situation in the cluster tilting case is asymmetric, as the partition for a sink involves the subset of cluster tilting objects containing the corresponding simple, while the corresponding partition of at a source involves the subset of cluster tilting objects containing the shifted projective. In contrast, both partitions for the tilting case involve the subset of tilting modules containing the simple, either at a source or sink. This asymmetry is inherent in the proof of , which is not the dual of that of , and also in the analysis of the effect of the BGP reflection functor.
Second, in the cluster tilting case, the order preserving maps occurring in the flip-flop construction are from the set containing the simple (or shifted projective) to its complement, while in the tilting case, they are in the opposite direction, into the set containing the simple. As a consequence, a partition with respect to a sink in the cluster tilting case yields an order of the form $\leq_{+}$, while for the tilting case it gives $\leq_{-}$.
While two posets $Z$ and $Z'$ related via a flip-flop are in general not isomorphic, they are *universally derived equivalent* in the following sense; for any abelian category ${\mathcal{A}}$, the derived categories of the categories of functors $Z \to {\mathcal{A}}$ and $Z' \to {\mathcal{A}}$ are equivalent as triangulated categories, see [@Ladkani07u].
It is known [@Happel88 (I.5.7)] that the path algebras of two quivers $Q$, $Q'$ without oriented cycles are derived equivalent if and only if $Q'$ can be obtained from $Q$ by a sequence of BGP reflections (at sources or sinks). We therefore deduce the following theorem.
\[t:clunider\] Let $Q$ and $Q'$ be two quivers without oriented cycles whose path algebras are derived equivalent. Then the posets ${\mathcal{T}}_{{\mathcal{C}}_Q}$ and ${\mathcal{T}}_{{\mathcal{C}}_{Q'}}$ are universally derived equivalent.
Since for a Dynkin quiver $Q$, the poset ${\mathcal{T}}_{{\mathcal{C}}_Q}$ is isomorphic to the corresponding Cambrian lattice, the above theorem can be restated as follows.
All Cambrian lattices corresponding to the various orientations of the same Dynkin diagram are universally derived equivalent.
In particular, the incidence algebras of the Cambrian lattices corresponding to the various orientations the same Dynkin diagram are derived equivalent, as the universal derived equivalence of two finite posets implies the derived equivalence of their incidence algebras.
Acknowledgement {#acknowledgement .unnumbered}
---------------
I would like to thank Frédéric Chapoton for suggesting a conjectural version of Theorem \[t:clunider\] in the case of Dynkin quivers and for many helpful discussions.
Cluster tilting objects containing $P_x$ {#sec:clsink}
========================================
Let $x \in Q$ be a vertex, and denote by ${\mathcal{T}}^x_{{\mathcal{C}}_Q}$ the subset of cluster tilting objects containing $P_x$ as direct summand.
Let $M \in \operatorname{rep}Q$. Then $P_x \in \operatorname{fac}M$ if and only if $M$ contains $P_x$ as a direct summand.
Assume that $P_x \in \operatorname{fac}M$, and let $q : M^n \twoheadrightarrow P_x$ be a surjection, for some $n \geq 1$. Since $P_x$ is projective, there exists $j : P_x \to M^n$ such that $qj = 1_{P_x}$. Let $N = \operatorname{Im}j =
\operatorname{Im}jq$. As $(jq)^2 = jq$, we deduce that $N$ is a direct summand of $M^n$ and that $j : P_x \to N$ is an isomorphism. Since $P_x$ is indecomposable, it is also a summand of $M$.
\[c:cTxPx\] Let $T \in {\mathcal{T}}_{{\mathcal{C}}_Q}$. Then $T \in {\mathcal{T}}^x_{{\mathcal{C}}_Q}$ if and only if $P_x
\in \operatorname{fac}T$.
\[c:cTxclose\] If $T \in {\mathcal{T}}^x_{{\mathcal{C}}_Q}$ and $T' \leq T$, then $T' \in {\mathcal{T}}^x_{{\mathcal{C}}_Q}$,
Let $T \in {\mathcal{T}}^x_{{\mathcal{C}}_Q}$ and $T' \in {\mathcal{T}}_{{\mathcal{C}}_Q}$. If $T' \leq T$, then $P_x \in \operatorname{fac}T \subseteq \operatorname{fac}T'$, hence $T' \in {\mathcal{T}}^x_{{\mathcal{C}}_Q}$.
Define a map $f : {\mathcal{T}}^x_{{\mathcal{C}}_Q} \to {\mathcal{T}}_{{\mathcal{C}}_Q} \setminus
{\mathcal{T}}^x_{{\mathcal{C}}_Q}$ as follows. Given $T \in {\mathcal{T}}^x_{{\mathcal{C}}_Q}$, write $T = P_x
\oplus U$ and set $f(T) = M \oplus U$ where $M$ is the unique other indecomposable complement of $U$ such that $M \oplus U$ is a cluster tilting object.
Recall that for a tilting module $T \in \operatorname{rep}Q$, we have $\operatorname{fac}T =
T^{\perp}$ where $$T^{\perp} = \left\{ M \in \operatorname{rep}Q \,:\, \operatorname{Ext}^1_Q(T, M) = 0 \right\}$$
\[l:fTgeT\] Let $T \in {\mathcal{T}}^x_{{\mathcal{C}}_Q}$. Then $f(T) > T$.
One could deduce the claim from Lemma 2.32 of [@IngallsThomas06]. Instead, we shall give a direct proof. Write $T = P_x \oplus U$ and $f(T) = M \oplus U$. If $M$ is a shifted projective, the claim is clear. Otherwise, by deleting the vertices of $Q$ corresponding to the shifted projective summands of $U$, we may and will assume that $P_x
\oplus U$ and $M \oplus U$ are tilting modules. Therefore $$\operatorname{fac}(P_x \oplus U) = (P_x \oplus U)^{\perp} = U^{\perp}$$ where the last equality follows since $P_x$ is projective. As $M \in
U^{\perp}$, we get that $M \in \operatorname{fac}(P_x \oplus U)$, hence $\operatorname{fac}(M
\oplus U) \subseteq \operatorname{fac}(P_x \oplus U)$.
For the rest of this section, *we assume that the vertex $x$ is a sink in $Q$*. In this case, $P_x = S_x$ and $\operatorname{ind}\operatorname{fac}S_x = \{S_x\}$. Moreover, $S_x \not \in \operatorname{fac}M$ for any other indecomposable $M \neq
S_x$, since $\operatorname{Hom}_Q(M, S_x) = 0$.
\[l:facfT\] Let $T \in {\mathcal{T}}^x_{{\mathcal{C}}_Q}$. Then $\operatorname{ind}\operatorname{fac}f(T) = \operatorname{ind}\operatorname{fac}T \setminus
\{S_x\}$.
Write $T = S_x \oplus U$ and $f(T) = M \oplus U$. By the preceding remarks, $$\operatorname{ind}\operatorname{fac}T = \operatorname{ind}\operatorname{fac}(S_x \oplus U) = \operatorname{ind}\operatorname{fac}S_x \cup \operatorname{ind}\operatorname{fac}U$$ is a disjoint union $\{S_x\} \amalg \operatorname{ind}\operatorname{fac}U$. By Lemma \[l:fTgeT\], $$\operatorname{ind}\operatorname{fac}f(T) = \operatorname{ind}\operatorname{fac}(M \oplus U) \subseteq \operatorname{ind}\operatorname{fac}(S_x \oplus
U) = \{S_x\} \amalg \operatorname{ind}\operatorname{fac}U ,$$ therefore $\operatorname{ind}\operatorname{fac}(M \oplus U) = \operatorname{ind}\operatorname{fac}U$, as $S_x \not \in \operatorname{fac}M$.
\[c:fTgeuniq\] Let $T \in {\mathcal{T}}^x_{{\mathcal{C}}_Q}$ and $T' \in {\mathcal{T}}_{{\mathcal{C}}_Q} \setminus
{\mathcal{T}}^x_{{\mathcal{C}}_Q}$ be such that $T' > T$. Then $T' \geq f(T)$.
By assumption, $\operatorname{fac}T' \subseteq \operatorname{fac}T$. Moreover, $S_x \not \in \operatorname{fac}T'$, since $T' \not \in {\mathcal{T}}^x_{{\mathcal{C}}_Q}$. Hence by Lemma \[l:facfT\], $\operatorname{ind}\operatorname{fac}T' \subseteq \operatorname{ind}\operatorname{fac}f(T)$, thus $T' \geq f(T)$.
\[c:ffclsink\] The map $f : {\mathcal{T}}^x_{{\mathcal{C}}_Q} \to {\mathcal{T}}_{{\mathcal{C}}_Q} \setminus {\mathcal{T}}^x_{{\mathcal{C}}_Q}$ is order preserving and $${\mathcal{T}}_{{\mathcal{C}}_Q} \simeq \bigl( {\mathcal{T}}^x_{{\mathcal{C}}_Q} \sqcup ({\mathcal{T}}_{{\mathcal{C}}_Q} \setminus
{\mathcal{T}}^x_{{\mathcal{C}}_Q}), \leq^f_{+} \bigr)$$
If $T, T' \in {\mathcal{T}}^x_{{\mathcal{C}}_Q}$ are such that $T' \geq T$, then by Lemma \[l:fTgeT\], $f(T') > T' \geq T$, hence by Corollary \[c:fTgeuniq\], $f(T') \geq f(T)$, therefore $f$ is order preserving. The other assertion follows from Corollaries \[c:cTxclose\], \[c:fTgeuniq\] and Lemma \[l:fTgeT\].
Cluster tilting objects containing $P_x[1]$ {#sec:clsource}
===========================================
For $M \in \operatorname{rep}Q$ and $y \in Q$, let $M(y)$ denote the vector space corresponding to $y$, and let $\operatorname{supp}M = \{y \in Q \,:\, M(y) \neq 0
\}$ be the *support* of $M$.
Let $x \in Q$ be a vertex, and denote by ${\mathcal{T}}^{x[1]}_{{\mathcal{C}}_Q}$ the subset of cluster tilting objects containing the shifted indecomposable projective $P_x[1]$ as direct summand.
\[l:cTxopen\] If $T \in {\mathcal{T}}^{x[1]}_{{\mathcal{C}}_Q}$ and $T' \geq T$, then $T' \in
{\mathcal{T}}^{x[1]}_{{\mathcal{C}}_Q}$.
Since $T$ contains $P_x[1]$ as summand, we have $\operatorname{Ext}^1_{{\mathcal{C}}_Q}(P_x[1],
T) = 0$, that is, $\operatorname{Hom}_Q(P_x, {\widehat}{T}) = 0$, or equivalently $x \not
\in \operatorname{supp}{\widehat}{T}$.
Now let $T' \geq T$. Then all the modules in $\operatorname{fac}T' \subseteq \operatorname{fac}T$ are not supported on $x$, and in particular $\operatorname{Hom}_Q(P_x, {\widehat}{T'}) = 0$, thus $\operatorname{Ext}^1_{{\mathcal{C}}_Q}(P_x[1], T') = 0$. The maximality of $T'$ implies that it contains $P_x[1]$ as summand.
Similarly to the previous section, define a map $g : {\mathcal{T}}^{x[1]}_{{\mathcal{C}}_Q}
\to {\mathcal{T}}_{{\mathcal{C}}_Q} \setminus {\mathcal{T}}^{x[1]}_{{\mathcal{C}}_Q}$ as follows. Given $T \in
{\mathcal{T}}^{x[1]}_{{\mathcal{C}}_Q}$, write $T = P_x[1] \oplus U$ and set $g(T) = M
\oplus U$ where $M$ is the unique other indecomposable complement of $U$ such that $M \oplus U$ is a cluster tilting object.
\[l:gTltT\] Let $T \in {\mathcal{T}}^{x[1]}_{{\mathcal{C}}_Q}$. Then $g(T) < T$.
This is obvious. Indeed, write $T = P_x[1] \oplus U$ and $g(T) = M
\oplus U$. Then $\operatorname{fac}g(T) = \operatorname{fac}(M \oplus U) \supseteq \operatorname{fac}U = \operatorname{fac}T$.
For the rest of this section, *we assume that the vertex $x$ is a source*. In this case, for any module $M \in \operatorname{rep}Q$, we have that $S_x
\in \operatorname{fac}M$ if and only if $M$ is supported at $x$. Therefore we deduce the following lemma, which can be viewed as an analogue of Corollary \[c:cTxPx\].
Let $T \in {\mathcal{T}}_{{\mathcal{C}}_Q}$. Then $T \in {\mathcal{T}}^{x[1]}_{{\mathcal{C}}_Q}$ if and only if $S_x \not \in \operatorname{fac}T$.
Recall that a basic module $U \in \operatorname{rep}Q$ is an *almost complete tilting module* if $\operatorname{Ext}^1_Q(U,U)=0$ and the number of indecomposable summands of $U$ equals the number of vertices of $Q$ less one. A *complement* to $U$ is an indecomposable $M$ such that $M \oplus
U$ is a tilting module. It is known [@HappelUnger89] that an almost complete tilting module $U$ has at most two complements, and exactly two if and only if $U$ is sincere, that is, $\operatorname{supp}U = Q$.
\[p:UMtorsion\] Let $U$ be an almost complete tilting module of $\operatorname{rep}Q$ not supported on $x$, and let $M$ be its unique indecomposable complement to a tilting module. Let ${\mathcal{X}}$ be a torsion class in $\operatorname{rep}Q$ satisfying $\operatorname{fac}U \subseteq {\mathcal{X}}$ and $S_x \in {\mathcal{X}}$. Then $M \in {\mathcal{X}}$.
The natural inclusion $j : {Q \setminus \{x\}}\to Q$ induces a pair $({j_{!}}, {j^{-1}})$ of exact functors $$\begin{aligned}
{j^{-1}}: \operatorname{rep}Q \to \operatorname{rep}({Q \setminus \{x\}}) && {j_{!}}: \operatorname{rep}({Q \setminus \{x\}}) \to \operatorname{rep}Q\end{aligned}$$ where ${j^{-1}}$ is the natural restriction and ${j_{!}}$ is its left adjoint, defined as the extension of a representation of ${Q \setminus \{x\}}$ by zero at $x$.
Now ${j_{!}}{j^{-1}}U \simeq U$ since $U$ is not supported on $x$. By adjunction and exactness, $$\operatorname{Ext}^1_{{Q \setminus \{x\}}}({j^{-1}}U, {j^{-1}}U) \simeq \operatorname{Ext}^1_Q({j_{!}}{j^{-1}}U, U) = \operatorname{Ext}^1_Q(U,
U)$$ thus ${j^{-1}}U$ is a (basic) tilting module of $\operatorname{rep}({Q \setminus \{x\}})$. However, by [@Ladkani07t Proposition 2.6], ${j^{-1}}(M \oplus U)$ is also a tilting module of $\operatorname{rep}({Q \setminus \{x\}})$, but not necessarily basic. It follows that ${j^{-1}}M \in \operatorname{add}{j^{-1}}U$, hence ${j_{!}}{j^{-1}}M \in \operatorname{add}{j_{!}}{j^{-1}}U =
\operatorname{add}U$.
The adjunction morphism ${j_{!}}{j^{-1}}M \to M$ is injective, and we have an exact sequence $$0 \to {j_{!}}{j^{-1}}M \to M \to S_x^n \to 0$$ for some $n \geq 0$. Now $S_x \in {\mathcal{X}}$ by assumption and ${j_{!}}{j^{-1}}M
\in \operatorname{add}U \subseteq {\mathcal{X}}$, hence $M \in {\mathcal{X}}$ as ${\mathcal{X}}$ is closed under extensions.
\[c:gTleuniq\] Let $T \in {\mathcal{T}}^{x[1]}_{{\mathcal{C}}_Q}$ and $T' \in {\mathcal{T}}_{{\mathcal{C}}_Q} \setminus
{\mathcal{T}}^{x[1]}_{{\mathcal{C}}_Q}$ be such that $T' < T$. Then $T' \leq g(T)$.
Write $T = P_x[1] \oplus U$ and $g(T) = M \oplus U$. The assumptions on $T'$ imply that $S_x \in \operatorname{fac}T'$ and $\operatorname{fac}U = \operatorname{fac}T \subseteq \operatorname{fac}T'$.
By deleting the vertices of $Q$ corresponding to the shifted projective summands of $U$, we may and will assume that $M \oplus U$ is a tilting module, so that $U$ is an almost complete tilting module. Applying Proposition \[p:UMtorsion\] for ${\mathcal{X}}= \operatorname{fac}T'$, we deduce that $M
\in \operatorname{fac}T'$, hence $\operatorname{fac}g(T) = \operatorname{fac}(M \oplus U) \subseteq \operatorname{fac}T'$.
\[c:ffclsource\] The map $g : {\mathcal{T}}^{x[1]}_{{\mathcal{C}}_Q} \to {\mathcal{T}}_{{\mathcal{C}}_Q} \setminus
{\mathcal{T}}^{x[1]}_{{\mathcal{C}}_Q}$ is order preserving and $${\mathcal{T}}_{{\mathcal{C}}_Q} \simeq \bigl( {\mathcal{T}}^{x[1]}_{{\mathcal{C}}_Q} \sqcup ({\mathcal{T}}_{{\mathcal{C}}_Q}
\setminus {\mathcal{T}}^{x[1]}_{{\mathcal{C}}_Q}), \leq^g_{-} \bigr)$$
The claim follows from Lemmas \[l:cTxopen\], \[l:gTltT\] and Corollary \[c:gTleuniq\] as in the proof of Corollary \[c:ffclsink\].
The effect of a BGP reflection {#sec:clBGP}
==============================
Let $Q$ be a quiver without oriented cycles and let $x$ be a sink. Let $y_1, \dots, y_m$ be the endpoints of the arrows ending at $x$, and denote by $Q'$ the quiver obtained from $Q$ by reflection at $x$. For a vertex $y \in Q$, denote by $S_y$, $S'_y$ the simple modules corresponding to $y$ in $\operatorname{rep}Q$, $\operatorname{rep}Q'$ and by $P_y$, $P'_y$ their projective covers.
The categories $\operatorname{rep}Q$ and $\operatorname{rep}Q'$ are related by the BGP reflection functors, introduced in [@BGP73]. We recollect here the basic facts on these functors that will be needed in the sequel.
The BGP reflection functors are the functors $$\begin{aligned}
{F^{+}}: \operatorname{rep}Q \to \operatorname{rep}Q' && {F^{-}}: \operatorname{rep}Q' \to \operatorname{rep}Q\end{aligned}$$ defined by $$\begin{aligned}
\label{e:FpFm}
({F^{+}}M)(x) &= \ker \Bigl(\bigoplus_{i=1}^{m} M(y_i) \to M(x) \Bigr) &&
({F^{+}}M)(y) = M(y)
\\ \notag
({F^{-}}M')(x) &= \operatorname{coker}\Bigl(M'(x) \to \bigoplus_{i=1}^{m} M'(y_i)
\Bigr) && ({F^{-}}M')(y) = M'(y)\end{aligned}$$ for $M \in \operatorname{rep}Q$, $M' \in \operatorname{rep}Q'$ and $y \in {Q \setminus \{x\}}$, where the maps $({F^{+}}M)(x) \to ({F^{+}}M)(y_i)$ and $({F^{-}}M)(y_i) \to ({F^{-}}M)(x)$ are induced by the natural projection and inclusion.
It is clear that ${F^{+}}$ is left exact and ${F^{-}}$ is right exact. The classical right derived functor of ${F^{+}}$ takes the form $$\begin{aligned}
\label{e:R1Fp}
(R^1 {F^{+}}M)(x) &= \operatorname{coker}\Bigl(\bigoplus_{i=1}^{m} M(y_i) \to M(x)
\Bigr) && (R^1 {F^{+}}M)(y) = 0\end{aligned}$$ hence $R^1 {F^{+}}$ vanishes for modules not containing $S_x$ as direct summand.
The total derived functors $$\begin{aligned}
R{F^{+}}: {\mathcal{D}}^b(Q) \to {\mathcal{D}}^b(Q') && L{F^{-}}: {\mathcal{D}}^b(Q') \to {\mathcal{D}}^b(Q)\end{aligned}$$ are triangulated equivalences, and their effect on the corresponding cluster categories has been analyzed in [@Zhu07], where it is shown that $R{F^{+}}$ induces a triangulated equivalence ${\mathcal{C}}_Q
\xrightarrow{\simeq} {\mathcal{C}}_{Q'}$ whose action on the indecomposables of ${\mathcal{C}}_Q$ is given by $$\begin{aligned}
\label{e:CFpindec}
S_x \mapsto P'_x[1] && M \mapsto {F^{+}}M && P_x[1] \mapsto S'_x && P_y[1]
\mapsto P'_y[1]\end{aligned}$$ with an inverse given by $$\begin{aligned}
\label{e:CFmindec}
S'_x \mapsto P_x[1] && M' \mapsto {F^{-}}M' && P'_x[1] \mapsto S_x &&
P'_y[1] \mapsto P_y[1]\end{aligned}$$ for $M \neq S_x$, $M' \neq S'_x$ and $y \in {Q \setminus \{x\}}$. Moreover, this equivalence induces a bijection $\rho : {\mathcal{T}}_{{\mathcal{C}}_Q} \to {\mathcal{T}}_{{\mathcal{C}}_{Q'}}$ preserving the mutation graph [@Zhu07 Proposition 3.2].
\[l:rhorder\] Let $T, T' \in {\mathcal{T}}_{{\mathcal{C}}_Q}$. If $\rho(T) \leq \rho(T')$, then $T \leq
T'$.
By , $\operatorname{fac}T = \operatorname{fac}{F^{-}}{\widehat}{\rho(T)}$ if $P'_x[1]$ is not a summand of $\rho(T)$, and $\operatorname{fac}T = \operatorname{fac}(S_x \oplus {F^{-}}{\widehat}{\rho(T)})$ if $P'_x[1]$ is a summand of $\rho(T)$. Note that by Lemma \[l:cTxopen\], the latter case implies that $P'_x[1]$ is also a summand of $\rho(T')$, hence in any case it is enough to verify that if $M, N \in \operatorname{rep}Q'$ satisfy $\operatorname{fac}N \subseteq \operatorname{fac}M$, then $\operatorname{fac}F^{-}N
\subseteq \operatorname{fac}F^{-}M$. Indeed, since $F^{-}$ is right exact, it takes an exact sequence $M^n \to N \to 0$ to an exact sequence $(F^{-} M)^n
\to F^{-} N \to 0$.
\[p:rhocTx\] $\rho$ induces an isomorphism of posets ${\mathcal{T}}^x_{{\mathcal{C}}_Q}
\xrightarrow{\simeq} {\mathcal{T}}^{x[1]}_{{\mathcal{C}}_{Q'}}$.
Note that by , $\rho({\mathcal{T}}^x_{{\mathcal{C}}_Q}) =
{\mathcal{T}}^{x[1]}_{{\mathcal{C}}_{Q'}}$. In view of Lemma \[l:rhorder\], it remains to show that if $T, T' \in {\mathcal{T}}^x_{{\mathcal{C}}_Q}$ satisfy $T \leq T'$, then $\rho(T) \leq \rho(T')$.
Write ${\widehat}{T} = S_x \oplus U$ and ${\widehat}{T'} = S_x \oplus U'$. Then $\operatorname{fac}\rho(T) = \operatorname{fac}{F^{+}}U$ and $\operatorname{fac}\rho(T') = \operatorname{fac}{F^{+}}U'$, and we need to show that ${F^{+}}U' \in \operatorname{fac}{F^{+}}U$.
Indeed, since $\operatorname{fac}T' \subseteq \operatorname{fac}T$, the proof of Lemma \[l:facfT\] shows that $U' \in \operatorname{fac}U$, hence there exists a short exact sequence $$0 \to K \to U^n \xrightarrow{{\varphi}} U' \to 0$$ for some $n > 0$ and $K \in \operatorname{rep}Q$. Applying $\operatorname{Hom}_Q(-,S_x)$ to this sequence, noting that $\operatorname{Ext}^1_Q(U', S_x) = 0$ since $T' \in
{\mathcal{T}}^x_{{\mathcal{C}}_Q}$, we get that $\operatorname{Hom}_Q(U^n, S_x) \to \operatorname{Hom}_Q(K, S_x)$ is surjective, hence $K$ does not contain $S_x$ as summand (otherwise $U^n$ would contain $S_x$ as summand). Therefore the exact sequence $${F^{+}}U^n \to {F^{+}}U' \to R^1 {F^{+}}K = 0$$ shows that ${F^{+}}U' \in \operatorname{fac}{F^{+}}U$.
\[p:rhocTxcomp\] $\rho$ induces an isomorphism of posets ${\mathcal{T}}_{{\mathcal{C}}_Q} \setminus
{\mathcal{T}}^x_{{\mathcal{C}}_Q} \xrightarrow{\simeq} {\mathcal{T}}_{{\mathcal{C}}_{Q'}} \setminus
{\mathcal{T}}^{x[1]}_{{\mathcal{C}}_{Q'}}$.
For a representation $M \in \operatorname{rep}Q$, let $Q_M$ and $Q'_M$ be the subquivers of $Q$ and $Q'$ obtained by deleting the vertices outside $\operatorname{supp}M \cup \{x\}$. The quivers $Q'_M$ and $Q_M$ are related via a BGP reflection at $x$, and we denote by $F^{+}_{Q_M} : \operatorname{rep}Q_M \to
\operatorname{rep}Q'_M$ the corresponding reflection functor. The restriction functors $i^{-1} : \operatorname{rep}Q \to \operatorname{rep}Q_M$ and $j^{-1} : \operatorname{rep}Q' \to \operatorname{rep}Q'_M$ induced by the natural embeddings $i : Q_M \to Q$ and $j : Q'_M
\to Q'$ satisfy $$j^{-1} {F^{+}}M = F^{+}_{Q_M} i^{-1} M ,$$ as can be easily verified using .
As in the proof of Proposition \[p:rhocTx\], it is enough to show that if $T, T' \in {\mathcal{T}}_{{\mathcal{C}}_Q} \setminus {\mathcal{T}}^x_{{\mathcal{C}}_Q}$ satisfy $T \leq
T'$, then $\rho(T) \leq \rho(T')$. In view of the preceding paragraph, we may assume that $Q = \operatorname{supp}{\widehat}{T} \cup \{x\}$.
We consider two cases. First, assume that $x \in \operatorname{supp}{\widehat}{T}$. Then $T
= {\widehat}{T}$ is a tilting module, $\rho(T) = {F^{+}}T$ and $\operatorname{fac}\rho(T') =
\operatorname{fac}{F^{+}}{\widehat}{T'}$ or $\operatorname{fac}\rho(T') = \operatorname{fac}(S'_x \oplus {F^{+}}{\widehat}{T'})$ according to whether $x \in \operatorname{supp}{\widehat}{T'}$ or not, hence it is enough to show that $S'_x \oplus {F^{+}}{\widehat}{T'} \in \operatorname{fac}{F^{+}}T$.
By assumption, ${\widehat}{T'} \in \operatorname{fac}T = T^{\perp}$. Since $T$ does not contain $S_x$ as summand, ${F^{+}}T$ is a tilting module and ${F^{+}}{\widehat}{T'}
\in ({F^{+}}T)^{\perp} = \operatorname{fac}{F^{+}}T$ [@Ladkani07t Corollary 4.3]. Moreover, $S'_x \in \operatorname{fac}{F^{+}}T$, as ${F^{+}}T$ is sincere.
For the second case, assume that $x \not \in \operatorname{supp}{\widehat}{T}$. Then $T =
P_x[1] \oplus {\widehat}{T}$ and by Lemma \[l:cTxopen\], $T' = P_x[1] \oplus
{\widehat}{T'} \oplus P[1]$ where $P$ is a sum of projectives other than $P_x$. By , $\rho(T) = S'_x \oplus {F^{+}}{\widehat}{T}$ and $\rho(T') = S'_x \oplus {F^{+}}{\widehat}{T'} \oplus P'[1]$, hence it is enough to show that ${F^{+}}{\widehat}{T'} \in \operatorname{fac}(S'_x \oplus {F^{+}}{\widehat}{T})$.
Indeed, since ${\widehat}{T'} \in \operatorname{fac}{\widehat}{T}$, there exists a short exact sequence $$0 \to K \to {\widehat}{T}^n \to {\widehat}{T'} \to 0$$ for some $n > 0$ and $K \in \operatorname{rep}Q$. Applying the functor ${F^{+}}$, noting that ${\widehat}{T}$ does not contain $S_x$ as summand, we get $$0 \to {F^{+}}K \to {F^{+}}{\widehat}{T}^n \to {F^{+}}{\widehat}{T'} \to R^1 {F^{+}}K \to R^1 {F^{+}}{\widehat}{T}^n = 0 .$$ By , $R^1 {F^{+}}K = {S'_x}^{n'}$ for some $n' \geq 0$, hence ${F^{+}}{\widehat}{T'}$ is an extension of ${S'_x}^{n'}$ with a quotient of ${F^{+}}{\widehat}{T}^n$. The result now follows, as $\operatorname{fac}(S'_x \oplus {F^{+}}{\widehat}{T})$ is closed under extensions.
\[c:clcomm\] We have a commutative diagram $$\xymatrix{
{\mathcal{T}}^x_{{\mathcal{C}}_Q} \ar[r]^{\rho}_{\simeq} \ar[d]^{f} &
{\mathcal{T}}^{x[1]}_{{\mathcal{C}}_{Q'}} \ar[d]^{g} \\
{\mathcal{T}}_{{\mathcal{C}}_Q} \setminus {\mathcal{T}}^x_{{\mathcal{C}}_Q} \ar[r]^{\rho}_{\simeq} &
{\mathcal{T}}_{{\mathcal{C}}_{Q'}} \setminus {\mathcal{T}}^{x[1]}_{{\mathcal{C}}_{Q'}}
}$$
By Propositions \[p:rhocTx\] and \[p:rhocTxcomp\], $\rho$ induces the two horizontal isomorphisms. For $T \in {\mathcal{T}}^x_{{\mathcal{C}}_Q}$, $f(T)$ is defined as the mutation of $T$ with respect to $S_x$ and $g(\rho(T))$ is defined as the mutation of $\rho(T)$ with respect to $P'_x[1]$, which is, by , the image of $S_x$ under the triangulated equivalence ${\mathcal{C}}_Q \to {\mathcal{C}}_{Q'}$. Therefore the commutativity of the diagram follows by the fact that $\rho$ preserves the mutation graph [@Zhu07 Proposition 3.2].
The posets ${\mathcal{T}}_{{\mathcal{C}}_Q}$ and ${\mathcal{T}}_{{\mathcal{C}}_{Q'}}$ are related via a flip-flop.
Use Corollaries \[c:ffclsink\], \[c:ffclsource\] and \[c:clcomm\].
Example {#sec:clexample}
=======
Consider the following two quivers $Q$ and $Q'$ whose underlying graph is the Dynkin diagram $A_3$. The quiver $Q'$ is obtained from $Q$ by a BGP reflection at the sink $3$. $$\begin{aligned}
Q : \xymatrix{
{\bullet_1} \ar[r] & {\bullet_2} \ar[r] & {\bullet_3}
} &&
Q' : \xymatrix{
{\bullet_1} \ar[r] & {\bullet_2} & {\bullet_3} \ar[l]
}\end{aligned}$$
We denote the indecomposables of the cluster categories ${\mathcal{C}}_Q$ and ${\mathcal{C}}_{Q'}$ by specifying their dimension vectors. These consist of the positive roots of $A_3$, which correspond to the indecomposable representations of the quivers, together with the negative simple roots $-e_1, -e_2, -e_3$ which correspond to the shifted projectives.
$$\xymatrix@=1.5pc{
& &
{\begin{smallmatrix} 0 & 1 & 0 \\ 0 & 1 & 1 \\ 1 & 1 & 1
\end{smallmatrix}} \ar[d] \ar[dddrrr] \\
& & {\begin{smallmatrix} 0 & 1 & 0 \\ 1 & 1 & 0 \\ 1 & 1 & 1
\end{smallmatrix}} \ar[dl] \ar[dr] \\
{\begin{smallmatrix} \mathbf{0} & \mathbf{0} & \mathbf{1} \\
\mathbf{0} & \mathbf{1} & \mathbf{1} \\
\mathbf{1} & \mathbf{1} & \mathbf{1}
\end{smallmatrix}} \ar[uurr] \pmb{\ar[d]} \pmb{\ar[dddrrr]} &
{\begin{smallmatrix} 1 & 0 & 0 \\ 1 & 1 & 0 \\ 1 & 1 & 1
\end{smallmatrix}} \ar[dr] & &
{\begin{smallmatrix} 0 & 1 & 0 \\ 1 & 1 & 0 \\ 0 & 0 & -1
\end{smallmatrix}} \ar[dl] \ar[ddrr] \\
{\begin{smallmatrix} \mathbf{0} & \mathbf{0} & \mathbf{1} \\
\mathbf{1} & \mathbf{0} & \mathbf{0} \\
\mathbf{1} & \mathbf{1} & \mathbf{1}
\end{smallmatrix}} \ar[ur] \pmb{\ar[ddrr]} & &
{\begin{smallmatrix} 1 & 0 & 0 \\ 1 & 1 & 0 \\ 0 & 0 & -1
\end{smallmatrix}} \ar[dr] & & &
{\begin{smallmatrix} 0 & 1 & 0 \\ 0 & 1 & 1 \\ -1 & 0 & 0
\end{smallmatrix}} \ar[d] \\
& & &
{\begin{smallmatrix} 1 & 0 & 0 \\ 0 & -1 & 0 \\ 0 & 0 & -1
\end{smallmatrix}} \ar[dr] & &
{\begin{smallmatrix} 0 & 1 & 0 \\ 0 & 0 & -1 \\ -1 & 0 & 0
\end{smallmatrix}} \ar[dl] \\
& &
{\begin{smallmatrix} \mathbf{0} & \mathbf{0} & \mathbf{1} \\
\mathbf{1} & \mathbf{0} & \mathbf{0} \\
\mathbf{0} & \mathbf{-1} & \mathbf{0}
\end{smallmatrix}} \ar[ur] \pmb{\ar[dr]} &
{\begin{smallmatrix} \mathbf{0} & \mathbf{0} & \mathbf{1} \\
\mathbf{0} & \mathbf{1} & \mathbf{1} \\
\mathbf{-1} & \mathbf{0} & \mathbf{0}
\end{smallmatrix}} \ar[uurr] \pmb{\ar[d]} &
{\begin{smallmatrix} -1 & 0 & 0 \\ 0 & -1 & 0 \\ 0 & 0 & -1
\end{smallmatrix}} \\
& & &
{\begin{smallmatrix} \mathbf{0} & \mathbf{0} & \mathbf{1} \\
\mathbf{0} & \mathbf{-1} & \mathbf{0} \\
\mathbf{-1} & \mathbf{0} & \mathbf{0}
\end{smallmatrix}} \ar[ur]
}$$
$$\xymatrix@=1.5pc{
& &
{\begin{smallmatrix} 0 & 1 & 1 \\ 0 & 1 & 0 \\ 1 & 1 & 0
\end{smallmatrix}} \ar[ddll] \ar[d] \ar[dddrrr] \\
& &
{\begin{smallmatrix} 0 & 1 & 1 \\ 1 & 1 & 1 \\ 1 & 1 & 0
\end{smallmatrix}} \ar[dl] \ar[dr] \\
{\begin{smallmatrix} \mathbf{0} & \mathbf{0} & \mathbf{-1} \\
\mathbf{0} & \mathbf{1} & \mathbf{0} \\
\mathbf{1} & \mathbf{1} & \mathbf{0}
\end{smallmatrix}} \pmb{\ar[d]} \pmb{\ar[dddrrr]} &
{\begin{smallmatrix} 1 & 0 & 0 \\ 1 & 1 & 1 \\ 1 & 1 & 0
\end{smallmatrix}} \ar[dl] \ar[dr]
& &
{\begin{smallmatrix} 0 & 1 & 1 \\ 1 & 1 & 1 \\ 0 & 0 & 1
\end{smallmatrix}} \ar[dl] \ar[ddrr] \\
{\begin{smallmatrix} \mathbf{0} & \mathbf{0} & \mathbf{-1} \\
\mathbf{1} & \mathbf{0} & \mathbf{0} \\
\mathbf{1} & \mathbf{1} & \mathbf{0}
\end{smallmatrix}} \pmb{\ar[ddrr]} & &
{\begin{smallmatrix} 1 & 0 & 0 \\ 1 & 1 & 1 \\ 0 & 0 & 1
\end{smallmatrix}} \ar[dr] & & &
{\begin{smallmatrix} 0 & 1 & 1 \\ 0 & 1 & 0 \\ -1 & 0 & 0
\end{smallmatrix}} \ar[ddll] \ar[d] \\
& & &
{\begin{smallmatrix} 1 & 0 & 0 \\ 0 & -1 & 0 \\ 0 & 0 & 1
\end{smallmatrix}} \ar[dl] \ar[dr] & &
{\begin{smallmatrix} 0 & 1 & 1 \\ 0 & 0 & 1 \\ -1 & 0 & 0
\end{smallmatrix}} \ar[dl] \\
& &
{\begin{smallmatrix} \mathbf{0} & \mathbf{0} & \mathbf{-1} \\
\mathbf{1} & \mathbf{0} & \mathbf{0} \\
\mathbf{0} & \mathbf{-1} & \mathbf{0}
\end{smallmatrix}} \pmb{\ar[dr]} &
{\begin{smallmatrix} \mathbf{0} & \mathbf{0} & \mathbf{-1} \\
\mathbf{0} & \mathbf{1} & \mathbf{0} \\
\mathbf{-1} & \mathbf{0} & \mathbf{0}
\end{smallmatrix}} \pmb{\ar[d]} &
{\begin{smallmatrix} -1 & 0 & 0 \\ 0 & -1 & 0 \\ 0 & 0 & 1
\end{smallmatrix}} \ar[dl] \\
& & &
{\begin{smallmatrix} \mathbf{0} & \mathbf{0} & \mathbf{-1} \\
\mathbf{0} & \mathbf{-1} & \mathbf{0} \\
\mathbf{-1} & \mathbf{0} & \mathbf{0}
\end{smallmatrix}}
}$$
Figure \[fig:cltilting\] shows the Hasse diagrams of the posets ${\mathcal{T}}_{{\mathcal{C}}_Q}$ and ${\mathcal{T}}_{{\mathcal{C}}_{Q'}}$, where we used bold font to indicate the subsets ${\mathcal{T}}^3_{{\mathcal{C}}_Q}$ and ${\mathcal{T}}^{3[1]}_{{\mathcal{C}}_{Q'}}$ of cluster tilting objects containing the simple $S_3$ and the shifted projective $P'_3[1]$ as summand, respectively.
The posets ${\mathcal{T}}_{{\mathcal{C}}_Q}$ and ${\mathcal{T}}_{{\mathcal{C}}_{Q'}}$ are Cambrian lattices, and can be realized as sublattices of the weak order on the group of permutations on $4$ letters, see [@Reading06 Section 6]. Moreover, the underlying graph of their Hasse diagrams is the $1$-skeleton of the three-dimensional Stasheff associhedron. ${\mathcal{T}}_{{\mathcal{C}}_Q}$ is a Tamari lattice, corresponding to the linear orientation on $A_3$.
The BGP reflection at the vertex $3$, whose action on the dimension vectors is given by $$v \mapsto \begin{cases} v & \text{if $v \in \{-e_1, -e_2\}$} \\
s_3(v) & \text{otherwise}
\end{cases}$$ where $s_3$ is the linear transformation specified by $$s_3(v) = v \cdot
\begin{pmatrix}
1 & 0 & 0 \\
0 & 1 & 1 \\
0 & 0 & -1 \\
\end{pmatrix} ,$$ induces isomorphisms ${\mathcal{T}}^3_{{\mathcal{C}}_Q} \xrightarrow{\simeq}
{\mathcal{T}}^{3[1]}_{{\mathcal{C}}_{Q'}}$ and ${\mathcal{T}}_{{\mathcal{C}}_Q} \setminus {\mathcal{T}}^3_{{\mathcal{C}}_Q}
\xrightarrow{\simeq} {\mathcal{T}}_{{\mathcal{C}}_{Q'}} \setminus {\mathcal{T}}^{3[1]}_{{\mathcal{C}}_{Q'}}$ compatible with the mutations at $S_3$ and $P'_3[1]$.
[10]{}
I.N. Bernstein, I.M. Gel’fand, and V.A. Ponomarev, *[Coxeter functors and Gabriel’s theorem.]{}*, Russ. Math. Surv. **28** (1973), no. 2, 17–32 (English).
Aslak Bakke Buan, Robert Marsh, Markus Reineke, Idun Reiten, and Gordana Todorov, *Tilting theory and cluster combinatorics*, Adv. Math. **204** (2006), no. 2, 572–618.
P. Caldero, F. Chapoton, and R. Schiffler, *Quivers with relations arising from clusters ([$A\sb n$]{} case)*, Trans. Amer. Math. Soc. **358** (2006), no. 3, 1347–1364 (electronic).
Sergey Fomin and Andrei Zelevinsky, *Cluster algebras. [I]{}. [F]{}oundations*, J. Amer. Math. Soc. **15** (2002), no. 2, 497–529 (electronic).
Dieter Happel, *Triangulated categories in the representation theory of finite-dimensional algebras*, London Mathematical Society Lecture Note Series, vol. 119, Cambridge University Press, Cambridge, 1988.
Dieter Happel and Luise Unger, *Almost complete tilting modules*, Proc. Amer. Math. Soc. **107** (1989), no. 3, 603–610.
Colin Ingalls and Hugh Thomas, *Noncrossing partitions and representations of quivers*, [arXiv:math/0612219v2]{}.
Bernhard Keller, *On triangulated orbit categories*, Doc. Math. **10** (2005), 551–581 (electronic).
Sefi Ladkani, *Universal derived equivalences of posets*, [ arXiv:0705.0946v2]{}.
[to3em]{}, *Universal derived equivalences of posets of tilting modules*, [arXiv:0708.1287]{}.
Nathan Reading, *Cambrian lattices*, Adv. Math. **205** (2006), no. 2, 313–353.
Christine Riedtmann and Aidan Schofield, *On a simplicial complex associated with tilting modules*, Comment. Math. Helv. **66** (1991), no. 1, 70–78.
Bin Zhu, *B[GP]{}-reflection functors and cluster combinatorics*, J. Pure Appl. Algebra **209** (2007), no. 2, 497–506.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We show that the intuitionistic first-order theory of equality has continuum many complete extensions. We also study the Vitali equivalence relation and show there are many intuitionistically precise versions of it.'
address: 'Institute for Mathematics, Astrophysics and Particle Physics, Faculty of Science, Radboud University Nijmegen, Postbus 9010, 6500 GL Nijmegen, the Netherlands'
author:
- Wim Veldman
title: 'Equality and equivalence, intuitionistically'
---
Introduction
============
We want to contribute to L.E.J. Brouwer’s program of doing mathematics\
*intuitionistically*.
We follow his advice to interpret the logical constants constructively.
A conjunction $A\;\wedge \;B$ is considered proven if and only if one has a proof of $A$ and also a proof of $B$.
A disjunction $A\vee B$ is considered proven if and only if either $A$ or $B$ is proven.
An implication $A\rightarrow B$ is considered proven if and only if there is a proof of $B$ using the assumption $A$.
A negation $\neg A$ is considered proven if and only if there is a proof of $A\rightarrow 0=1$.
An existential statement $\exists x \in V[P(x)]$ is considered proven if and only an element $x_0$ is produced together with a proof of the associated statement $P(x_0)$.
A universal statement $\forall x\in V[P(x)]$ is considered proven if and only if a method is given that produces, given any $x$ in $V$, a proof of the associated statement $P(x)$.
We also use some axioms proposed by Brouwer: his *Continuity Principle*, our Axiom \[ax:bcp\], a slightly stronger version of it, the *First Axiom of Continuous Choice*, our Axiom \[ax:fcc\], and his *Thesis on Bars in $\mathcal{N}$*, our Axiom \[A:barinduction\].
In some of our proofs, we use an Axiom of Countable Choice, our Axiom \[ax:countable choice\]. Intuitionistic mathematicians, who accept infinite step-by-step constructions not determined by a rule, consider this axiom a reasonable proposal.
Finally, we believe that generalized inductive definitions, like our Definition \[D:perhapsiveextvitali\], fall within the compass of intuitionistic mathematics.
Our subject is the (intuitionistic) first-order theory of equality. By considering structures $(\mathcal{X}, =)$ where $\mathcal{X}$ is a subset of Baire space $\mathcal{N}=\omega^\omega$ and $=$ the usual equality relation on $\mathcal{N}$, we find that the theory has an uncountable and therefore astonishing[^1] variety of elementarily different infinite models and, as a consequence, an astonishing variety of complete extensions, see Theorem \[T:uncountablymany\]. The key observation[^2] leading to this result is the recognition that, in a *spread*[^3], an *isolated* point is the same as a *decidable* point.[^4] It follows that the set of the non-isolated points of a spread is a definable subset of the spread. In spreads that are *transparent*[^5], the set of the non-isolated points of the spread coincides with the *coherence* of the spread[^6], and the coherence itself is spread. It may happen that the coherence of a transparent spread is transparent itself and then the coherence of the coherence also is a definable subset of the spread. And so on.
Any structure $(\mathcal{N}, R)$, where $R$ is an equivalence relation on $\mathcal{N}$, is a model of the theory of equality. We study the *Vitali equivalence relation*, see Section \[S:vitali\], as an example. This equivalence relation, in contrast to the equality relation on $\mathcal{N}$, is not *stable*[^7], see Theorem \[T:vitaliunstable\].
There is a host of binary relations on $\mathcal{N}$ that, from a classical point of view, all would be the same as the Vitali equivalence relation, see Sections \[S:vitalivar\] and \[S:morevitali\], and especially Definition \[D:perhapsiveextvitali\], Corollary \[C:hierarchy\] and Definition \[D:almost\]. It turned out to be difficult to find differences between them that are first-order expressible. We did find some such differences, however, by studying structures $(\mathcal{N}, =, R)$, where $R$ is an intuitionistic version of the Vitali equivalence relation and $=$ the usual equality, see Section \[S: equequiv\].
The paper is divided into 13 Sections and consists roughly of two parts. Sections 2-8 lead up to the result that the theory of equality has continuum many complete extensions, see Theorem \[T:uncountablymany\]. Sections 9-12 treat the Vitali equivalence relations. Section 13 lists some notations and conventions and may be used by the reader as a reference.
Intuitionistic model theory
===========================
Given a relational structure $\mathfrak{A}=(A, R_0, R_1, \ldots, R_{n-1})$, we construct a first-order language $\mathcal{L}$ with basic formulas $\mathsf{R}_i(\mathsf{x}_0, \mathsf{x}_1,\ldots, \mathsf{x}_{l_i-1})$, where $i<n$ and $l_i$ is the arity of $R_i$. The formulas of $\mathcal{L}$ are obtained from the basic formulas by using $\wedge, \vee, \rightarrow, \neg, \exists, \forall$ in the usual way.
For every formula $\varphi=\varphi(\mathsf{x}_0, \mathsf{x}_1, \ldots, \mathsf{x}_{m-1})$ of $\mathcal{L}$, for all $a_0, a_1, \ldots, a_{m-1}$ in $A$, we define the statement: $$\mathfrak{A}\models \varphi[a_0, a_1, \ldots, a_{m-1}]$$ (*$\mathfrak{A}$ realizes $\varphi$ if $\mathsf{x}_0, \mathsf{x}_1, \ldots, \mathsf{x}_{m-1}$ are interpreted by $a_0, a_1, \ldots, a_{m-1}$, respectively*), as Tarski did it, with the proviso that connectives and quantifiers are interpreted intuitionistically.
A formula $\varphi$ of $\mathcal{L}$ without free variables will be called a *sentence*.
A *theory (in $\mathcal{L}$)* is a set of sentences of $\mathcal{L}$.
Given a theory $\Gamma$ in $\mathcal{L}$ and a structure $\mathfrak{A}$, we define: $\mathfrak{A}$ *realizes* $\Gamma$ if and only if, for every $\varphi$ in $\Gamma$, $\mathfrak{A}\models \varphi$.
Given a structure $\mathfrak{B}$ that has the same signature as $\mathfrak{A}$, so that the formulas of $\mathcal{L}$ may be interpreted in $\mathfrak{B}$ as well as in $\mathfrak{A}$, we let $Th(\mathfrak{B})$, the *theory of $\mathfrak{B}$*, be the set of all sentences $\varphi$ of $\mathcal{L}$ such that $\mathfrak{B}\models \varphi$.
A theory $\Gamma$ in $\mathcal{L}$ will be called a *complete theory* if and only if there exists a structure $\mathfrak{B}$ such that $\Gamma=Th(\mathfrak{B})$.
This agrees with one of the uses of the expression ‘*complete theory*’ in classical, that is: usual, non-intuitionistic, model theory, see [@hodges p. 43]. Note that one may be unable to decide, for a given sentence $\varphi$ and a given structure $\mathfrak{B}$, whether or not $\mathfrak{B}\models \varphi$. Intuitionistically, it is not true that, for every complete theory $\Gamma$ and every sentence $\varphi$, *either* $\varphi\in \Gamma$ *or* $\neg\varphi\in\Gamma$.
Complete theories $\Gamma, \Delta$ are *positively different* if one may point out a sentence $\psi$ such that $\psi \in \Gamma$ and $\neg\psi \in \Delta$.[^8]
Structures $\mathfrak{A}, \mathfrak{B}$ are *elementarily equivalent* if and only if $Th(\mathfrak{A})=Th(\mathfrak{B})$ and *(positively) elementarily different* if $Th(\mathfrak{A})$ is positively different from $ Th(\mathfrak{B})$.
Let $\Gamma$ be a theory in $\mathcal{L}$. A good question is the following:
> *How many complete theories $\Delta$ can one find extending $\Gamma$?*
We will say: *$\Gamma$ admits countably many complete extensions* if and only if there exists an infinite sequence $\Delta_0, \Delta_1, \ldots$ of complete theories extending $\Gamma$ such that, for all $m,n$, if $m\neq n$, then $\Delta_m,\Delta_n$ are (positively) different, and
*$\Gamma$ admits continuum many complete extensions* if and only if there exists a function $\alpha\mapsto \Delta_\alpha$ associating to every element $\alpha$ of $\mathcal{C}=2^\omega$ a complete theory extending $\Gamma$ such that for all $\alpha, \beta$, if[^9] $\alpha\;\#\;\beta$, then $\Delta_\alpha,\Delta_\beta$ are (positively) different.
A main result of this paper is that the first-order theory of equality admits continuum many complete extensions.
Equality may be undecidable
===========================
The first-order theory $EQ$ of equality consists of the following three axioms:
1. $\mathsf{\forall x[x=x]}$,
2. $\mathsf{\forall x\forall y[x=y\rightarrow y=x]}$ and
3. $\mathsf{\forall x\forall y\forall z[(x=y\;\wedge\;y=z)\rightarrow x=z]}$.
A model of $EQ$ is a structure of the form $(V,R)$, where $V$ is a set and $R$ is an equivalence relation on $V$, possibly, but not necessarily, the equality relation belonging to $V$.
Classically, every complete extension of $EQ$ is realized in one of the structures from the list: $(\{0\},=)$, $(\{0,1\},=)$, $(\{0,1,2\},=)$, $\dots$ and $(\omega,=)$. This shows that, classically, $EQ$ admits of (no more than) countably many complete extensions.
Intuitionistically, however, we have to observe that all structures on this list satisfy the sentence
> $ \mathsf{\forall x\forall y[x=y\;\vee\;\neg(x=y)]}$,
that is: the equality relation, on each of these sets, is a *decidable* relation.
It is well-known, however, that the equality relation on the set $\mathcal{N}$ is *not* a decidable relation. Let us recall why.
If we define an element $\alpha$ of $\mathcal{N}$ by stipulating: $$\forall n[\alpha(n)\neq 0\leftrightarrow \forall i<99[d(n+i)=9]],$$ where $d:\mathbb{N}\rightarrow \{0,1,\ldots,9\}$ is the decimal expansion of $\pi$, we are unable to decide: $\alpha=\underline 0\;\vee\neg(\alpha =\underline 0)$.
This is because, if $\alpha =\underline 0$, then $\neg\exists n\forall i<99[\alpha(n+i)=9]$, and, if $\neg(\alpha=\underline 0)$, then\
$\neg\neg\exists n\forall i<99[d(n+i)=9]$, and we have no proof of either alternative.
This example shows us that the statement $\forall \alpha[\alpha =\underline 0\;\vee\;\neg(\alpha =\underline 0)]$, for a constructive mathematician, who interprets the disjunction strongly, is a *reckless* statement.[^10]
The following axiom, used by Brouwer[^11], implies that the statement\
$\forall \alpha[\alpha =\underline 0\;\vee\;\neg(\alpha =\underline 0)]$ even leads to a contradiction.
\[ax:bcp\] $\;\;$\
For all $R\subseteq \mathcal{N}\times\omega$, if $\forall \alpha\exists n[\alpha Rn]$, then $\forall \alpha\exists m \exists n\forall \beta[\overline\alpha m \sqsubset\beta\rightarrow \beta Rn]$.
An immediate consequence is:
\[L:bcpdisj\] $\;\;$\
For all $P_0, P_1\subseteq \mathcal{N}$, if $\forall \alpha [\alpha\in P_0\;\vee\;\alpha \in P_1]$, then\
$\forall \alpha\exists m[\forall \beta[\overline \alpha m\sqsubset \beta\rightarrow \beta \in P_0]\;\vee\; \forall \beta[\overline \alpha m\sqsubset \beta\rightarrow \beta \in P_1]]$.
Define $R:=\{(\alpha, n)\mid \alpha \in P_n]$ and apply Axiom \[ax:bcp\].
1. $(\mathcal{N}, =)\models \mathsf{\forall x\neg\forall y[x=y\;\vee\;\neg(x=y)]}$.
2. $(\mathcal{N}, =)\models \mathsf{\neg\forall x\forall y[x=y\;\vee\;\neg(x=y)]}$.
\(i) Let $\alpha$ be given and assume: $\forall \beta[\alpha=\beta\;\vee\;\neg(\alpha=\beta)]$.\
Using Lemma \[L:bcpdisj\], find $m$ such that\
*either* $\forall \beta[\overline \alpha m\sqsubset\beta \rightarrow \alpha = \beta]$ *or* $\forall \beta[\overline \alpha m\sqsubset\beta \rightarrow \neg(\alpha = \beta)]$.\
Consider $\beta:=\overline\alpha m\ast\langle \alpha(m)+1\rangle\ast\underline 0$ (for the first alternative) and $\beta:=\alpha$ (for the second one) and conclude that both alternatives are false.
\(ii) This is an immediate consequence of (i).
\[D:Tinf\]$\;$
For each $n$, we let $\psi_n$ be the sentence $\mathsf{\exists x_0\exists x_1\ldots \exists x_n}[\bigwedge_{i<j<n}\neg (\mathsf{x}_i=\mathsf{x}_j)]$.
$T_{inf}:=EQ\cup\{\psi_n\mid n\in \omega\}$.
$\psi_n$ expresses that a set has at least $n+1$ elements.
Note that, in classical mathematics, $T_{inf}$ has only one complete extension.
Intuitionistically, however, $T_{inf}$ has (at least) two positively different complete extensions, $Th\bigl((\mathcal{N},=)\bigr)$ and $Th\bigl((\omega, =)\bigr)$.
The next Theorem reflects the fact that, in classical model theory, all models of $T_{inf}$ are elementarily equivalent.
\[T:classicalcompleteness\] The theory $T_{inf}\cup \{\forall\mathsf{x\forall y[x=y\;\vee\;\neg(x=y)]}\}$ has only one complete extension.
For each $n$, consider the first $n$ variables of our language: $\mathsf{x}_0, \mathsf{x}_1, \ldots ,\mathsf{x}_{n-1}$. A formula $\varepsilon=\varepsilon(\mathsf{x}_0, \mathsf{x}_1, \ldots ,\mathsf{x}_{n-1})$ is called an *equality type* if and only if it is of the form $\bigwedge_{i<j<n} \sigma_{ij}$ where each $\sigma_{ij}$ either is the formula $\mathsf{x}_i=\mathsf{x}_j$ or the formula $\neg(\mathsf{x}_i=\mathsf{x}_j)$.[^12] One may prove: for all structures $(V_0, R_0), (V_1, R_1)$, both realizing $T_{inf}\cup \{\forall\mathsf{x\forall y[x=y\;\vee\;\neg(x=y)]}\}$, for each formula $\varphi=\varphi(\mathsf{x}_0, \mathsf{x}_1, \ldots ,\mathsf{x}_{n-1})$, for each equality type $\varepsilon=\varepsilon(\mathsf{x}_0, \mathsf{x}_1, \ldots ,\mathsf{x}_{n-1})$, $(V_0, R_0)\models \forall \mathsf{x}_0\forall\mathsf{x}_1 \ldots\forall\mathsf{x}_{n-1}[\varepsilon \rightarrow \varphi]$ if and only if $(V_1, R_1)\models \forall \mathsf{x}_0\forall\mathsf{x}_1 \ldots\forall\mathsf{x}_{n-1}[\varepsilon \rightarrow \varphi]$. The proof is by induction on the complexity of the formula $\varphi$.
It follows that any two models $(V_0, R_0), (V_1, R_1)$, both realizing\
$T_{inf}\cup \{\forall\mathsf{x\forall y[x=y\;\vee\;\neg(x=y)]}\}$, are elementarily equivalent.
From here on, we restrict attention to infinite models of $EQ$, that is, to models of $T_{inf}$. The hackneyed question to make a survey of models that are *finite*, or at least *not infinite*, and of models for which one can not decide if they are finite or infinite, is left for another occasion. That the job is not an easy one will be clear to readers of [@veldman1995].
Spreads {#S:spreads}
=======
Let $\beta$ be given. $\beta$ is called a *spread-law*, $Spr(\beta)$, if and only if $\forall s[\beta(s)=0\leftrightarrow \exists n[\beta(s\ast\langle n \rangle)=0]]$.
For every $\beta$, we define: $\mathcal{F}_\beta:=\{\alpha\mid\forall n[\beta(\overline \alpha n)=0]\}$.
$\mathcal{X}\subseteq \mathcal{N}$ is *closed* if and only if $\exists \beta[\mathcal{X}=\mathcal{F}_\beta]$.
$\mathcal{X}\subseteq \mathcal{N}$ is a *spread* if and only if $\exists \beta[Spr(\beta)\;\wedge\;\mathcal{X}=\mathcal{F}_\beta]$.
If $Spr(\beta)$ and $\beta(\langle \;\rangle)\neq 0$, then $\mathcal{F}_\beta=\emptyset$.
If $Spr(\beta)$ and $\beta(\langle\;\rangle)=0$, then $\mathcal{F}_\beta$ is inhabited[^13]. One may define $\alpha$ such that\
$\forall n[\alpha(n)=\mu p[\beta(\overline\alpha n\ast \langle p \rangle)=0]]$ and observe: $\forall n[\beta(\overline \alpha n)=0]$, that is: $\alpha \in \mathcal{F}_\beta$.
Is every closed set a spread?
Define $\beta$ such that $\forall s[\beta(s) =0\leftrightarrow \neg\forall i<99[d(n+i)=9]],$ where\
$d:\mathbb{N}\rightarrow \{0,1,\ldots, 9\}$ is the decimal expansion of $\pi$.
If $\mathcal{F}_\beta$ is a spread, that is $\exists\gamma[Spr(\gamma)\;\wedge\;\mathcal{F}_\gamma=\mathcal{F}_\beta]$, then *either* $\mathcal{F}_\beta$ is inhabited and $\neg \exists s\forall i<99[d(s+i)=9]$ *or* $\mathcal{F}_\beta = \emptyset$ and $\neg \neg\exists s\forall i<99[d(s+i)=9]$.
For this $\beta$, the statement ‘$\mathcal{F}_\beta$ is a spread’ thus turns out to be reckless.
Brouwer’s Continuity Principle enables one to obtain a stronger conclusion.
$\neg\forall \beta\exists \gamma[Spr(\gamma)\;\wedge\;\mathcal{F}_\gamma=\mathcal{F}_\beta]$.
Assume: $\forall \beta\exists \gamma[Spr(\gamma)\;\wedge\;\mathcal{F}_\gamma=\mathcal{F}_\beta]$. Then $\forall \beta[\exists\alpha[\alpha \in \mathcal{F}_\beta]\;\vee\;\neg\exists\alpha[\alpha\in \mathcal{F}_\beta]]$. Using Lemma \[L:bcpdisj\], find $m$ such that *either* $\forall \beta[\overline{\underline 0}m\sqsubset \beta\rightarrow \exists\alpha[\alpha\in \mathcal{F}_\beta]]$ *or*\
$\forall \beta[\overline{\underline 0}m\sqsubset \beta\rightarrow \neg\exists\alpha[\alpha\in \mathcal{F}_\beta]]$. Both alternatives are false, as we see by considering $\beta = \underline{\overline0}m\ast\underline 1$ (for the first alternative), and $\beta=\underline 0$ (for the second one).
\[L:bcpspreads\] $\;\;$\
Let $\beta$ be given such that $Spr(\beta)$. Then, for all $R\subseteq \mathcal{N}\times\omega$,\
if $\forall \alpha\in\mathcal{F}_\beta\exists n[\alpha Rn]$, then $\forall \alpha\in\mathcal{F}_\beta\exists m \exists n\forall \gamma\in\mathcal{F}_\beta[\overline\alpha m \sqsubset\gamma\rightarrow \gamma Rn]$.
Assume: $Spr(\beta)$. If $\beta(\langle\;\rangle)\neq 0$, then $\mathcal{F}_\beta=\emptyset$ and there is nothing to prove.
Assume $\beta(\langle\;\rangle)=0$. Define $\sigma$ such that $\sigma(\langle\;\rangle)=\langle\;\rangle$ and, for all $s$, for all $n$,
1. if $\beta(s\ast\langle n\rangle)=0$, then $\sigma(s\ast\langle n\rangle)=s\ast\langle n\rangle$, and,
2. if $\beta(s\ast\langle n \rangle)\neq 0$, then $\sigma(s\ast\langle n \rangle)=\sigma(s)\ast\langle \mu p[\beta\bigl(\sigma(s)\ast\langle p\rangle\bigr)=0]\rangle$.
Note: $\forall s[\beta\bigl(\sigma(s)\bigr)=0]$ and $\forall s\forall t[s\sqsubset t\rightarrow \sigma(s)\sqsubset\sigma(t)]$.
Define $\rho:\mathcal{N}\rightarrow\mathcal{N}$ such that $\forall \alpha\forall n[ \sigma(\overline \alpha n)\sqsubset \rho|\alpha]$.
Note: $\forall \alpha[\rho|\alpha\in\mathcal{F}_\beta]\;\wedge\; \forall \alpha \in \mathcal{F}_\beta[\rho|\alpha=\alpha]$.
The function $\rho$ is called a *retraction* of $\mathcal{N}$ onto $\mathcal{F}_\beta$.
Now assume: $\forall \alpha\in\mathcal{F}_\beta\exists n[\alpha Rn]$. Conclude: $\forall \alpha\exists n[(\rho|\alpha)R n]$.
Let $\alpha$ in $\mathcal{F}_\beta$ be given. Using Axiom \[ax:bcp\], find $m,n$ such that\
$\forall \gamma[\overline \alpha m\sqsubset \gamma\rightarrow (\rho|\gamma)Rn]$. Conclude: $\forall \gamma\in\mathcal{F}_\beta[\overline \alpha m\sqsubset \gamma\rightarrow \gamma R n]$.
We thus see: $\forall \alpha \in \mathcal{F}_\beta\exists m\exists n\forall \gamma \in \mathcal{F}_\beta[\overline \gamma m\sqsubset \alpha\rightarrow \gamma Rn]$.
Recall that, for all $\alpha, \beta$, $\alpha\;\#\;\beta\leftrightarrow \alpha\perp\beta\leftrightarrow \exists n[\alpha(n)\neq \beta(n)],$ and\
$\alpha=\beta\leftrightarrow \forall n[\alpha(n)=\beta(n)]\leftrightarrow \neg(\alpha\;\#\;\beta)$, and $\alpha\neq \beta \leftrightarrow \neg \forall n[\alpha(n)=\beta(n)].$
The constructive *apartness relation* $\#$ is more useful than the negative *inequality relation* $\neq$.
*Markov’s Principle*, in the form: $\forall \alpha[\neg\neg\exists n[\alpha(n)=0]\rightarrow\exists n[\alpha(n)=0]]$[^14], is equivalent to the statement that the two relations coincide: $\forall \alpha\forall \beta[\alpha\neq \beta\rightarrow \alpha\;\#\;\beta]$.
The intuitionistic mathematician does not accept Markov’s Principle.
We let $AP=AP(\mathsf{x,y})$ be the formula $\mathsf{\forall z[ \neg(z=x)\;\vee\;\neg(z=y)]}$.
The following theorem reformulates a well-known fact.
\[T:apartdef\] For all $ \beta$ such that $Spr(\beta)$,\
for all $\alpha, \delta$ in $\mathcal{F}_\beta$, $\alpha\;\#\;\delta$ if and only if $(\mathcal{F}_\beta,=)\models AP[\alpha, \delta]$.
First, assume $\alpha\;\#\;\delta$. Find $n$ such that $\overline \alpha n \neq\overline \delta n$. Note: for every $\gamma$ in $\mathcal{F}_\beta$, *either*: $\overline \gamma n \neq \overline \alpha n$ and $\gamma\;\#\;\alpha$, *or*: $\overline \gamma n\neq\delta n$ and $\gamma\;\#\;\delta$. Conclude: $(\mathcal{F}_\beta,=)\models AP[\alpha, \delta]$.
Next, assume $(\mathcal{F}_\beta,=)\models AP[\alpha, \delta]$, that is $\forall \gamma\in \mathcal{F}_\beta[\gamma \neq \alpha\;\vee\;\gamma\neq \delta]$.
Applying Lemma \[L:bcpspreads\], find $m$ such that *either* $\forall \gamma \in \mathcal{F}_\beta[\overline \alpha m\sqsubset \gamma \rightarrow \gamma \neq \alpha]$ *or* $\forall \gamma \in \mathcal{F}_\beta[\overline \alpha m\sqsubset \gamma \rightarrow \gamma \neq \delta]$. The first alternative is clearly wrong (take $\gamma:=\alpha$). The second alternative implies: $\overline\alpha m \perp \delta$ (if $\overline \alpha m \sqsubset \delta$, one could take $\gamma:=\delta$), and thus: $\alpha\;\#\;\delta$.
\[D:Tinfplus\] $\;$
For each $n$, we let $\psi_n^+$ be the sentence $\mathsf{\exists x_0\exists x_1\ldots \exists x_n}[\bigwedge_{i<j<n}AP(\mathsf{x}_i,\mathsf{x}_j)]$.
$T_{inf}^+:=EQ\cup\{\psi_n^+\mid n\in \omega\}$.
$\psi_n^+$ expresses that a set has at least $n+1$ elements that are mutually apart.
Every model of $T^+_{inf}$ realizes $T_{inf}$. In the second part of the paper we will meet a structure that realizes $T_{inf}$ but not $T^+_{inf}$, see Theorem \[T:vitalinoapartness\] in Section \[S:vitali\].
The theory $T^+_{inf}\cup \{\forall\mathsf{x\forall y[x=y\;\vee\;\neg(x=y)]}\}$ has only one complete extension, the same as the one and only complete extension of $T_{inf}\cup \{\forall\mathsf{x\forall y[x=y\;\vee\;\neg(x=y)]}\}$, see Theorem \[T:classicalcompleteness\].
Spreads with a decidable equality
=================================
We let $D=D(\mathsf{x})$ be the formula: $\mathsf{\forall y[x=y\;\vee\;\neg(x=y)]}$.
Assume $Spr(\beta)$ and $\alpha\in \mathcal{F}_\beta$.
$\alpha$ is an *isolated* point of $\mathcal{F}_\beta$ if and only if $\exists n\forall \gamma \in \mathcal{F}_\beta[\overline \alpha n \sqsubset \gamma\rightarrow \alpha = \gamma]$, or, equivalently, $\exists n\forall s[\bigl(\overline \alpha n\sqsubset s\;\wedge\;\beta(s)=0\bigr)\rightarrow s\sqsubset \alpha]$.
$\alpha$ is a *decidable* point of $\mathcal{F}_\beta$ if and only if $\forall \gamma\in\mathcal{F}_\beta[\alpha =\gamma\;\vee\;\neg(\alpha =\gamma)]$, or, equivalently, $(\mathcal{F}_\beta,=)\models D[\alpha]$.
$\mathcal{I}(\mathcal{F}_\beta)$ is the set of the isolated points of $\mathcal{F}_\beta$.
Cantor called $\mathcal{I}(\mathcal{F}_\beta)$ the *adherence* of $\mathcal{F}_\beta$.
\[L:defisol\] Assume $Spr(\beta)$.
1. For each $\alpha$ in $\mathcal{F}_\beta$, $\alpha$ is an isolated point of $\mathcal{F}_\beta$ if and only if $\alpha$ is a decidable point of $\mathcal{F}_\beta$.
2. $\mathcal{I}(\mathcal{F}_\beta)$ is a definable subset of $\mathcal{F}_\beta$.
\(i) Let $\alpha$ be an isolated point of $\mathcal{F}_\beta$.
Find $n$ such that $\forall \gamma \in \mathcal{F}_\beta[\overline \alpha n \sqsubset \gamma\rightarrow \alpha = \gamma]$.
Note: for each $\gamma$ in $\mathcal{F}_\beta$, *either* $\overline \alpha n\sqsubset \gamma$ and $\alpha =\gamma$, *or* $\overline \alpha n\perp \gamma$ and $\alpha\neq \gamma$.
Conclude: $\forall \gamma \in \mathcal{F}_\beta[\alpha=\gamma\;\vee\;\neg(\alpha=\gamma)]$, that is: $\alpha$ is a decidable point of $\mathcal{F}_\beta$.
Now assume: $\alpha$ is a decidable point of $\mathcal{F}_\beta$, that is: $\forall \gamma \in \mathcal{F}_\beta[\alpha=\gamma\;\vee\;\neg(\alpha=\gamma)]$.
Apply Lemma \[L:bcpspreads\] and find $m$ such that *either* $\forall \gamma \in \mathcal{F}_\beta[\overline \alpha m\sqsubset \gamma\rightarrow \alpha =\gamma]$ *or* $\forall \gamma \in \mathcal{F}_\beta[\overline \alpha m\sqsubset \gamma\rightarrow \neg( \alpha =\gamma)]$. As the second alternative does not hold (take $\gamma =\alpha$), conclude: $\forall \gamma \in \mathcal{F}_\beta[\overline \alpha m\sqsubset \gamma\rightarrow \alpha =\gamma]$, and: $\alpha$ is an isolated point of $ \mathcal{F}_\beta$.
\(ii) Using (i), note: $\mathcal{I}(\mathcal{F}_\beta) =\{\alpha\in \mathcal{F}_\beta\mid (\mathcal{F}_\beta, =)\models D[\alpha]\}$.
\[D:limitpoint\] Assume $Spr(\beta)$ and $\alpha \in \mathcal{F}_\beta$.
$\alpha$ is a *limit point* of $\mathcal{F}_\beta$ if and only if $\forall n\exists \delta \in \mathcal{F}_\beta[ \overline \alpha n\sqsubset \delta\;\wedge\;\alpha\perp\delta]$, or, equivalently, $\forall n\exists s[\overline \alpha n \sqsubset s\;\wedge\; \beta(s) =0\;\wedge\;\overline \alpha n\perp s]$.
$\mathcal{L}(\mathcal{F}_\beta)$ is the set of the limit points of $\mathcal{F}_\beta$.
Cantor called $\mathcal{L}(\mathcal{F}_\beta)$ the *coherence* of $\mathcal{F}_\beta$.
\[L:limit\] $\forall \beta[Spr(\beta)\rightarrow \mathcal{L}(\mathcal{F}_\beta)\subseteq\mathcal{F}_\beta\setminus \mathcal{I}(\mathcal{F}_\beta) ]$, that is:\
in all spreads, every limit point is a non-isolated point.
Obvious.
\[T:equivmarkov\] The following are equivalent:
1. Markov’s Principle: $\forall\alpha[\neg\neg\exists n[\alpha(n)=0]\rightarrow \exists n[\alpha(n)=0]]$.
2. $\forall \beta[Spr(\beta)\rightarrow \mathcal{F}_\beta\setminus \mathcal{I}(\mathcal{F}_\beta)\subseteq \mathcal{L}(\mathcal{F}_\beta)]$, that is:\
in all spreads, every non-isolated point is a limit point.
\(i) $\Rightarrow$ (ii). Let $\beta$ be given such that $Spr(\beta)$. Assume $\alpha$ is not an isolated point of $\mathcal{F}_\beta$, that is: $\neg \exists n\forall s[\bigl(\overline\alpha n\sqsubset s\;\wedge\;\beta(s)=0\bigr)\rightarrow s\sqsubset\alpha]$.
Let $n$ be given. Define $\delta$ such that $\forall s[\delta(s)=0 \leftrightarrow (\overline \alpha n\sqsubset s\;\wedge\;\beta(s) =0 \;\wedge\; s\perp \alpha)]$. Then $\neg\forall s[\delta(s)\neq 0]$ and: $\neg\neg\exists s[\delta(s)=0]$.
Using *Markov’s Principle*, we conclude: $\exists s[\delta(s)=0]$.
We thus see: $\forall n \exists s[\overline \alpha s\sqsubset s \;\wedge\;\beta(s)=0 \;\wedge\; s\perp\alpha]$, and: $\alpha$ is a limit point of $\mathcal{F}_\beta$.
\(ii) $\Rightarrow$ (i). Let us assume: $\forall \beta[Spr(\beta)\rightarrow \mathcal{F}_\beta\setminus \mathcal{I}(\mathcal{F}_\beta)\subseteq \mathcal{L}(\mathcal{F}_\beta)]$,
Let $\alpha$ be given such that $\neg\neg\exists n[\alpha(n)=0]$.\
Define $\beta$ such that $\forall s[\beta(s)=0\leftrightarrow \forall m<length(s)[s(m)\neq 0\rightarrow \exists n\le m[\alpha(n)=0]]]$.
Note: $Spr(\beta)$ and $\underline 0 \in \mathcal{F}_\beta$, and: if $\exists n[\alpha(n)=0]$, then $\underline 0$ is a limit point of $\mathcal{F}_\beta$.
Conclude: if $\underline 0$ is an isolated point of $\mathcal{F}_\beta$, then $\neg\exists n[\alpha(n)=0]$.
As $\neg\neg\exists n[\alpha(n)=0]$, conclude: $\underline 0$ is not an isolated point of $\mathcal{F}_\beta$.
By our assumption, $\underline 0$ thus is a limit point of $\mathcal{F}_\beta$.
Find $s$ such that $\beta(s)=0$ and $s \perp \underline 0$. Conclude: $\exists n\le length(s)[\alpha(n)=0]$.
Conclude: $\forall\alpha[\neg\neg\exists n[\alpha(n)=0]\rightarrow\exists n[\alpha(n)=0]]$, that is: Markov’s Principle.
We thus see that the converse of Lemma \[L:limit\], being equivalent to Markov’s Principle, is not an intuitionistic theorem.
We could not answer the question if, in general, $\mathcal{L}(\mathcal{F}_\beta)$ is a definable subset of $(\mathcal{F}_\beta, =)$. In some special cases, however, it is, and the following definition is useful.
\[D:transparent\] Assume $Spr(\beta)$. $\mathcal{F}_\beta$ is called *transparent* if and only if there exists $\gamma$ such that $Spr(\gamma)$ and $\mathcal{F}_\gamma = \mathcal{L}(\mathcal{F}_\beta)$ and $\forall \alpha \in \mathcal{F}_\beta[\exists n[\gamma(\overline \alpha n)\neq 0]\rightarrow \alpha \in \mathcal{I}(\mathcal{F}_\beta)]$.
Note that, for each $\beta$ such that $Spr(\beta)$, if $\mathcal{F}_\beta$ is transparent, then\
$\mathcal{F}_\beta\setminus \mathcal{I}(\mathcal{F}_\beta)\subseteq \mathcal{L}(\mathcal{F}_\beta)$. The statement that every spread $\mathcal{F}_\beta$ is transparent thus is seen to imply Markov’s Principle.
In Section \[S:moreundec\] we will see many examples of transparent spreads.
The fact that not every spread is a transparent spread is one of the reasons that Brouwer did not succeed in finding a nice intuitionistic version of Cantor’s Main Theorem[^15], see [@brouwer19].
\[D:functionspreads\] Let $\beta$ satisfy $Spr(\beta)$ and let $\varphi$ be given.
We define: $\varphi:\mathcal{F}_\beta\rightarrow \omega$ if and only if $\forall \alpha \in \mathcal{F}_\beta \exists p[\varphi(\overline\alpha p)\neq 0]$.
If $\varphi:\mathcal{F}_\beta \rightarrow \omega$, then we define, for each $\alpha$ in $\mathcal{F}_\beta$, $\varphi(\alpha)$ as the number $z$ such that $\varphi(\overline \alpha q)=z+1$, where $q=\mu p[\varphi(\overline \alpha p)\neq 0]$.
We define: $\varphi$ is an injective map from $\mathcal{F}_\beta$ into $\omega$, notation: $\varphi:\mathcal{F}_\beta \hookrightarrow\omega$,\
if and only if $\varphi:\mathcal{F}_\alpha\rightarrow \omega$ and $\forall \alpha\in\mathcal{F}_\beta\forall \delta\in\mathcal{F}_\beta[\alpha\;\#\;\delta\rightarrow \varphi(\alpha)\neq\varphi(\delta)]$.
We define: $\varphi: \mathcal{F}_\beta\rightarrow \mathcal{N}$ if and only if $\forall n[\varphi^n:\mathcal{F}_\beta\rightarrow \omega]$.
If $\varphi:\mathcal{F}_\beta\rightarrow \mathcal{N}$, then we define, for each $\alpha$ in $\mathcal{F}_\beta$, $\varphi|\alpha$ as the element $\delta$ of $\mathcal{N}$ such that $\forall n[\delta(n)=\varphi^n(\alpha)]$.
We define: $\varphi$ is an injective map from $\mathcal{F}_\beta$ into $\mathcal{N}$, notation: $\varphi:\mathcal{F}_\beta \hookrightarrow\mathcal{N}$,\
if and only if $\varphi:\mathcal{F}_\alpha\rightarrow \mathcal{N}$ and $\forall \alpha\in\mathcal{F}_\beta\forall \delta\in\mathcal{F}_\beta[\alpha\;\#\;\delta\rightarrow \varphi|\alpha\;\#\;\varphi|\delta]$.
For every $\mathcal{X}\subseteq \mathcal{N}$, $\mathcal{F}_\beta$ *embeds into* $\mathcal{X}$ if and only if there exists an injective map from $\mathcal{F}_\beta$ into $\mathcal{X}$.
The following axiom is, at least at first sight, a little bit stronger than Brouwer’s Continuity Principle.
\[ax:fcc\] For all $R\subseteq \mathcal{N}\times\omega$,\
if $\forall \alpha \exists n[\alpha Rn]$, then $\exists \varphi:\mathcal{N}\rightarrow\omega\forall \alpha[\alpha R\varphi(\alpha)]$.
\[L:fccspreads\]$\;\;$\
Let $\beta$ be given such that $Spr(\beta)$. Then, for all $R\subseteq \mathcal{F}_\beta\times \omega$,\
if $\forall \alpha \in \mathcal{F}_\beta\exists n[\alpha R n]$, then $\exists\varphi: \mathcal{F}_\beta\rightarrow \omega\forall\alpha\in\mathcal{F}_\beta[\alpha R\varphi(\alpha)]$.
Assume: $Spr(\beta)$ and $\beta(\langle\;\rangle)=0$. As in the proof of Lemma \[L:bcpspreads\], define $\rho:\mathcal{N}\rightarrow\mathcal{F}_\beta$ such that $\forall \alpha[\rho|\alpha\in\mathcal{F}_\beta]\;\wedge\; \forall \alpha \in \mathcal{F}_\beta[\rho|\alpha=\alpha]$.
Now assume $\forall \alpha\in\mathcal{F}_\beta\exists n[\alpha Rn]$. Conclude: $\forall \alpha\exists n[(\rho|\alpha)R n]$.
Applying Axiom \[ax:fcc\], find $\varphi:\mathcal{N}\rightarrow\omega$ such that $\forall \gamma[ (\rho|\gamma)R\varphi(\gamma)]$.
Conclude: $\varphi:\mathcal{F}_\beta\rightarrow \omega$ and $\forall \gamma\in\mathcal{F}_\beta[\gamma R\varphi(\gamma)]$.
\[T:decsprintoomega\] Assume $Spr(\beta)$. $(\mathcal{F}_\beta,=)\models \mathsf{\forall x}[D(\mathsf{x})]$ if and only if $\exists \varphi[\varphi:\mathcal{F}_\beta\hookrightarrow \omega]$.
First assume: $(\mathcal{F}_\beta,=)\models\mathsf{\forall x}[D(\mathsf{x})]$. Then, by Lemma \[L:defisol\],\
$\forall \alpha\in\mathcal{F}_\beta\exists n\forall \gamma\in\mathcal{F}_\beta[\overline\alpha n\sqsubset \gamma\rightarrow \alpha =\gamma]$. Using Lemma \[L:fccspreads\], find $\varphi:\mathcal{F}_\beta\rightarrow \omega$ such that $\forall \alpha\in\mathcal{F}_\beta\forall \gamma\in\mathcal{F}_\beta[\overline\alpha \varphi(\alpha)\sqsubset \gamma\rightarrow \alpha =\gamma]$. Define $\psi:\mathcal{F}_\beta\rightarrow \omega$ such that $\forall \alpha \in \mathcal{F}_\beta[\psi(\alpha)=\overline \alpha \varphi(\alpha)]$. Clearly, $\psi:\mathcal{F_\beta}\hookrightarrow \omega$.
Now assume: $\varphi:\mathcal{F}_\beta\hookrightarrow \omega$. Note: $\forall \alpha\in\mathcal{F}_\beta\forall \delta\in \mathcal{F}_\beta[\alpha = \delta\leftrightarrow \varphi(\alpha)=\varphi(\delta)]$.\
Also: $\forall \alpha\in\mathcal{F}_\beta\forall \delta\in \mathcal{F}_\beta[\varphi(\alpha)=\varphi(\delta)\;\vee\;\neg\bigl(\varphi(\alpha)=\varphi(\delta)\bigr)]$.\
Therefore: $\forall \alpha\in\mathcal{F}_\beta\forall \delta\in \mathcal{F}_\beta[\alpha = \delta\;\vee\; \neg(\alpha =\delta)]$. Conclude: $(\mathcal{F}_\beta,=)\models \mathsf{\forall x}[D(\mathsf{x})]$.
Assume $Spr(\beta)$. $\mathcal{F}_\beta$ is *enumerable* if and only if either $\mathcal{F}_\beta=\emptyset$ or $\exists \delta[\forall n[\delta^n \in \mathcal{F}_\beta]\;\wedge\;\forall \alpha \in \mathcal{F}_\beta\exists n[\alpha =\delta^n]]$.
\[L:enumer\] Assume $Spr(\beta)$. $\mathcal{F}_\beta$ is enumerable if and only if $\exists \varphi[\varphi:\mathcal{F}_\beta\hookrightarrow \omega]$.
Assume $\mathcal{F}_\beta$ is enumerable and $\beta(\langle\;\rangle)=0$.\
Find $\delta$ such that $\forall n[\delta^n\in \mathcal{F}_\beta]$ and $\forall \alpha \in \mathcal{F}_\beta\exists n[\alpha =\delta^n]$.\
Using Lemma \[L:fccspreads\], find $\varphi:\mathcal{F}_\beta\rightarrow \omega$ such that $\forall \alpha \in \mathcal{F}_\beta[\alpha=\delta^{\varphi(\alpha)}]$.\
Note: $\varphi:\mathcal{F}_\beta \hookrightarrow \omega$.
Now assume: $\varphi:\mathcal{F}_\beta \hookrightarrow \omega$.
We make a preliminary observation.
Let $s, n$ be given such that $\beta(s) =0$ and $\varphi(s)=n+1$ and $\forall t\sqsubset s[\varphi(t)=0]$.\
Note: $\forall \alpha \in \mathcal{F}_\beta[s\sqsubset \alpha\rightarrow \varphi(\alpha)=n]$ and, therefore:\
$\forall \alpha \in \mathcal{F}_\beta\forall \delta \in \mathcal{F}_\beta[(s\sqsubset \alpha\;\wedge\; s\sqsubset \delta)\rightarrow \alpha =\delta]$.
Now let $\gamma$ be the element of $\mathcal{F}_\beta$ satisfying $\forall n[\gamma(n):=\mu p[\beta(\overline \gamma n\ast \langle p\rangle)=0]]$.\
Define $\delta$ such that, for all $s$, *if* $\beta(s) =0$ and $\varphi(s)\neq 0$ and $\forall t\sqsubset s[\varphi(t)=0]$, then $s\sqsubset \delta^s$ and $\delta^s \in \mathcal{F}_\beta$, and *if not*, then $\delta^s=\gamma$.
Note: $\forall s[\delta^s \in \mathcal{F}_\beta]$ and $\forall \alpha \in \mathcal{F}_\beta\exists s[\alpha =\delta^s]$.
\[C:decenum\] Assume $Spr(\beta)$.\
$(\mathcal{F}_\beta, =)\models \forall\mathsf{x}[D(\mathsf{x})]$ if and only if $\mathcal{F}_\beta$ is enumerable.
Use Theorem \[T:decsprintoomega\] and Lemma \[L:enumer\].
Spreads with exactly one undecidable point
==========================================
We let $\tau_2$ be the element of $\mathcal{C}$ satisfying:\
$\forall s[\tau_2(s)=0\leftrightarrow \forall i<length(s)[s(i)<2\;\wedge\; \bigl(i+1<length(s)\rightarrow s(i)\le s(i+1)\bigr)]]$.\
We define: $\mathcal{T}_2:=\mathcal{F}_{\tau_2}$.
Note: $\tau_2$ is a spread-law and $\mathcal{T}_2$ is a spread.
Let us take a closer look at $\mathcal{T}_2$.
Observe: $\forall \alpha[\alpha\in\mathcal{T}_2\leftrightarrow \forall i[\alpha(i)\le\alpha(i+1)<2]]$.
For each $n$, we define $n^\ast:= \underline{\overline 0}n\ast\underline 1$.
The infinite sequence $\underline 0, 0^\ast, 1^\ast, 2^\ast, \ldots$ is a list of elements of $\mathcal{T}_2$ and a classical mathematician might think it is the list of all elements of $\mathcal{T}_2$. The intuitionistic mathematician knows better. He defines $\alpha$ in $\mathcal{T}_2$ such that $$\forall n[\alpha(n) =1\leftrightarrow \exists k\le n\forall i<99[d(k+i)=9]],$$ where $d:\mathbb{N}\rightarrow \{0,1,\ldots,9\}$ is the decimal expansion of $\pi$. As yet, one has no proof of the statement ‘$\alpha =\underline 0$’, as this statement implies: $\forall k\exists i<99]d(k+i)=9]$. As yet, one also has no proof of the statement: ‘$\exists n[\alpha= n^\ast]$’ as this statement implies: $\exists n\forall i<99[d(n+i)=9]$. The statement that $\alpha$ occurs in the above list is a reckless one.
For each $n$, $n^\ast$ is an isolated and a decidable point of $\mathcal{T}_2$, and $\underline 0$ is a non-isolated and an undecidable point of $\mathcal{T}_2$. It follows, by Lemma \[L:defisol\] and Corollary \[C:decenum\], that $\mathcal{T}_2$ is not an enumerable spread. In particular, the statement that the list $\underline 0, 0^\ast, 1^\ast, 2^\ast, \ldots$ is a complete list of the elements of $\mathcal{T}_2$, leads to a contradiction, as appears again from the following Theorem.
\[T:tau2\]
1. $\neg\forall\alpha\in\mathcal{T}_2[\alpha =\underline 0\;\vee\;\exists n[\alpha = n^\ast]]$.
2. $\forall\alpha\in\mathcal{T}_2[\alpha \;\#\;\underline 0\rightarrow \exists n[\alpha = n^\ast]]$.
\(i) Assume $\forall\alpha\in\mathcal{T}_2[\alpha =\underline 0\;\vee\;\exists n[\alpha = n^\ast]]$. Using Lemma \[L:bcpspreads\], find $m,n$ such that *either* $\forall\alpha\in \mathcal{T}_2[\underline{\overline 0}m\sqsubset\alpha\rightarrow \alpha=\underline 0]$ *or* $\forall\alpha\in \mathcal{T}_2[\underline{\overline 0}m\sqsubset\alpha\rightarrow \alpha=n^\ast]$. Note that both alternatives are false.
Conclude: $\neg\forall\alpha\in\mathcal{T}_2[\alpha =\underline 0\;\vee\;\exists n[\alpha = n^\ast]]$.
\(ii) Let $\alpha$ in $\mathcal{T}_2$ be given such that $\alpha\;\#\;\underline 0$. Define $n:=\mu m[\overline \alpha(m+1)\perp \underline 0]$. Note: $\overline \alpha(n+1)=\overline{\underline 0}n\ast\langle 1 \rangle$ and $\alpha =n^\ast$.
Assume $Spr(\beta)$. $\mathcal{F}_\beta$ is *almost-enumerable* if and only if either $\mathcal{F}_\beta=\emptyset$ or $\exists \delta[\forall n[\delta^n\in\mathcal{F}_\beta]\;\wedge\;\forall \alpha\in \mathcal{F}_\beta\forall \varepsilon \exists n[\overline \alpha \varepsilon(n) = \overline{\delta^n}\varepsilon(n)]]$.
This definition deserves some explanation. If $\mathcal{F}_\beta$ is almost-enumerable and inhabited, we are able to come forward with an infinite sequence $\delta^0, \delta^1, \ldots$ of elements of $\mathcal{F}_\beta$ such that, for every $\alpha$ in $\mathcal{F}_\beta$, every attempt $\varepsilon$ to prove that $\alpha$ is apart from all elements of the infinite sequence $\delta^0, \delta^1, \ldots$, ($\varepsilon$ expresses the guess: $\forall n[\overline \alpha\varepsilon(n)\perp \overline{\delta^n}\varepsilon(n)]$), will positively fail.
Almost-enumerable spreads are studied in [@veldman2019 Section 9], where they are called *almost-countable located and closed subsets of $\mathcal{N}$.*
$\mathcal{T}_2$ is almost-enumerable.
Define $\delta$ such that $\delta^0=\underline 0$ and, for each $n$. $\delta^{n+1}=n^\ast =\underline{\overline 0}n\ast\underline 1$. Note: $\forall n[\delta^n \in \mathcal{T}_2]$. Let $\varepsilon$ be given. If $\overline\alpha\varepsilon(0)=\overline{\delta^0}\varepsilon(0)$, we are done. If not, then $\alpha\perp\underline 0$ and we may determine $n$ such that $\alpha=\delta^{n+1}$ and $\overline \alpha \varepsilon(n+1)=\overline{\delta^{n+1}}\varepsilon(n+1)$.
\[ax:countable choice\]$\;\;$\
For every $R\subseteq \mathbb{N}\times \mathcal{N}$, if $\forall n\exists \alpha[nR\alpha]$, then $\exists \alpha\forall n[n R \alpha^n]$.
1. $(\mathcal{T}_2,=)\models\exists\mathsf{x}[\neg D(\mathsf{x})\;\wedge\;\forall \mathsf{y}[AP(\mathsf{x, y})\rightarrow D(\mathsf{y})]]$.
2. For all $\beta$ such that $Spr(\beta)$,\
if $(\mathcal{F}_\beta,=)\models \exists\mathsf{x}[\neg D(\mathsf{x})\;\wedge\;\forall \mathsf{y}[AP(\mathsf{x, y})\rightarrow D(\mathsf{y})]]$, then $\mathcal{F}_\beta$ embeds into $\mathcal{T}_2$.
\(i) $\underline 0$ is not an isolated point of $\mathcal{T}_2$, and, therefore, not a decidable point of $\mathcal{T}_2$. Also, by Theorem \[T:tau2\](ii), $\forall \alpha \in \mathcal{T}_2[\alpha\;\#\;\underline 0\rightarrow \exists n[\alpha= n^\ast]]$, and, for each $n$, for each $\alpha$ in $\mathcal{T}_2$, $\alpha = n^\ast\leftrightarrow \overline{\underline 0}n\ast\langle 1\rangle\sqsubset \alpha$, so one may decide: $\alpha = n^\ast$ or $\neg(\alpha = n^\ast)$, and: $n^\ast$ is a decidable point of $\mathcal{T}_2$. We thus see: $(\mathcal{T}_2,=)\models\neg D(\mathsf{x})\;\wedge\;\forall \mathsf{y}[AP(\mathsf{x, y})\rightarrow D(\mathsf{y})][\underline 0]$, and are done.
\(ii) Assume: $Spr(\beta)$ and $(\mathcal{F}_\beta,=)\models \exists\mathsf{x}[\neg D(\mathsf{x})\;\wedge\;\forall \mathsf{y}[AP(\mathsf{x, y})\rightarrow D(\mathsf{y})]]$.
Find $\alpha$ in $\mathcal{F}_\beta$ such that $\alpha$ is not an isolated point of $\mathcal{F}_\beta$.
Note: for each $s$ such that $\beta(s)=0$, the set $\mathcal{F}_\beta\cap s:=\{\delta \in \mathcal{F}_\beta\mid s\sqsubset \delta\}$ is a spread, and, if $s\perp\alpha$, then $\mathcal{F}_\beta\cap s$ consists of isolated points of $\mathcal{F}_\beta\cap s$ only, and thus, by Theorem \[T:decsprintoomega\], embeds into $\omega$.
Using Axiom \[ax:countable choice\], we find $\varphi$ such that, for each $s$, if $\beta(s)=0$ and there exist $n, i$ such that $s=\overline \alpha n\ast\langle i\rangle$ and $i\neq \alpha(n)$, then $\varphi^s:\mathcal{F}_\beta\cap s\hookrightarrow \omega$.
We now define $\psi:\mathcal{F}_\beta\rightarrow \mathcal{T}_2$ such that $\psi|\alpha =\underline 0$ and, for each $\delta$ in $\mathcal{F}_\beta$, if $\delta\;\#\;\alpha$, then $\psi|\delta=\overline{\underline 0}\bigl(\overline \delta n, \varphi^{\overline \delta n}(\delta)\bigr)\ast \underline 1$ where $n:=\mu i[\overline \delta i \perp \alpha]$.
More and more undecidable points: the toy spreads {#S:moreundec}
=================================================
For each $n$, we let $\tau_n$ be the element of $\mathcal{C}$ satisfying:\
$\forall s[\tau_n(s)=0\leftrightarrow \forall i<length(s)[s(i)<n\;\wedge\; \bigl(i+1<length(s)\rightarrow s(i)\le s(i+1)\bigr)]]$.\
We also define: $\mathcal{T}_n:=\mathcal{F}_{\tau_n}$.
For each $n$, $\tau_n$ is a spread-law and $\mathcal{T}_n$ and $\mathcal{T}_n=\{\alpha\mid \forall i[\alpha(i)\le\alpha(i+1)<n]\}$ is a spread.
In this paper, the spreads $\mathcal{T}_0, \mathcal{T}_1, \ldots$ will be called the *toy spreads*.
Note: $\mathcal{T}_0=\emptyset$ and $\mathcal{T}_1=\{\underline 0\}$.
For each $s\neq \langle\;\rangle$, we let $s^\dag$ be the element of $\mathcal{N}$ satisfying $s\sqsubset s^\dag$ and $\forall i\ge
length(s)[s^\dag(i)=s^\dag(i-1)]$.
Note that, for each $n$, for each $s$, if $s\neq\langle\;\rangle$ and $\tau_n(s)=0$, then $s^\dag \in \mathcal{T}_n$.
For each $n>0$, $\mathcal{T}_n$ is almost-enumerable.
Let $n>0$ be given. Define $\delta$ such that, for each $s$, *if* $s\neq\langle \;\rangle$ and $\tau_n(s)=0$, then $\delta^s=s^\dag$, and *if not*, then $\delta^s = \underline 0$.
We claim: $\forall \alpha\in\mathcal{T}_n\forall \varepsilon \exists s[\overline \alpha\varepsilon(s)=\overline{\delta^s}\varepsilon(s)]$.
We establish this claim by proving, for each $k<n$,\
$\forall \alpha\in\mathcal{T}_n[\exists i[\alpha(i)\ge k]\rightarrow\forall \varepsilon \exists s[\overline \alpha\varepsilon(s)=\overline{\delta^s}\varepsilon(s)]]$, and we do so by backwards induction, starting with the case $k=n-1$.
The case $k=n-1$ is treated as follows. If $\exists i[\alpha(i)=n-1]$, find\
$i_0:=\mu i[\alpha(i)=n-1]$ and consider $s:=\overline \alpha(i_0+1)$.\
Note: $\alpha=s^\dag=\delta^s$ and, therefore, for every $\varepsilon$: $\overline \alpha\varepsilon(s)=\overline{\delta^s}\varepsilon(s)$.
Now assume $k<n-1$ is given such that\
$\forall \alpha\in\mathcal{T}_n[\exists i[\alpha(i)\ge k+1]\rightarrow\forall \varepsilon \exists s[\overline \alpha\varepsilon(s)=\overline{\delta^s}\varepsilon(s)]]$.
We have to prove: $\forall \alpha\in\mathcal{T}_n[\exists i[\alpha(i)=k]\rightarrow\forall \varepsilon \exists s[\overline \alpha\varepsilon(s)=\overline{\delta^s}\varepsilon(s)]]$.
Let $\alpha$ be given such that $\exists i[\alpha(i)=k]$. Let also $\varepsilon$ be given.
Define $i_0:=\mu i[\alpha(i)=k]$ and define $s:=\overline\alpha(i_0+1)$.
There are two cases to consider.
*Case (i)*: $\overline \alpha\varepsilon(s)=\overline{s^\dag}\varepsilon(s)=\overline{\delta^s}\varepsilon(s)$. We are done.
*Case (ii)*: $\overline \alpha\varepsilon(s)\perp\overline{s^\dag}\varepsilon(s)$. Then $\exists i<\varepsilon(s) [\alpha(i)\ge k+1]$.
Using the induction hypothesis, we conclude: $\exists s[\overline \alpha\varepsilon(s)=\overline{\delta^s}\varepsilon(s)]$.
\[T:taun\] $\;$
1. For each $n$, for all $\alpha$ in $\mathcal{T}_n$, $\alpha \in \mathcal{I}(\mathcal{T}_n)$ if and only if $\exists m[\alpha(m)+1=n]$.
2. For each $n$, $\mathcal{T}_{n+1}\setminus \mathcal{I}(\mathcal{T}_{n+1}) = \mathcal{T}_n=\mathcal{L}(\mathcal{T}_{n+1})$.
3. For each $n$, $\mathcal{T}_n=\{\alpha \in \mathcal{T}_{n+1}\mid (\mathcal{T}_{n+1},=)\models \neg D[\alpha]\}$.
The proof uses Lemma \[L:defisol\] and is left to the reader.
We define an infinite sequence $D_0, D_1, \ldots$ of formulas, as follows.
$D_0:= \mathsf{\forall y[x=y\;\vee\;\neg(x=y)}]$,
$D_1:=\neg D_0(\mathsf{x}) \;\wedge\; \forall \mathsf{y}[\neg D_0(\mathsf{y})\rightarrow \bigl(\mathsf{x=y\;\vee\;\neg(x=y)\bigr)}]$,
$D_2:=\neg D_0(\mathsf{x})\;\wedge\; \neg D_1(\mathsf{x})\;\wedge\; \forall \mathsf{y}[\bigl(\neg D_0(\mathsf{y})\;\wedge\;\neg D_1(\mathsf{y})\bigr)\rightarrow \bigl(\mathsf{x=y\;\vee\;\neg(x=y)\bigr)}]$,
and, more generally for each $m>0$,
$D_{m}:=\bigwedge_{i<m}\neg D_i(\mathsf{x}) \;\wedge\; \forall \mathsf{y}[\bigl(\bigwedge_{i<m}\neg D_i(\mathsf{y})\bigr)\rightarrow \bigl(\mathsf{x=y\;\vee\;\neg(x=y)\bigr)}]$.
We also define, for each $m>0$, sentences $\psi_m$ and $\rho_m$, as follows:
$\psi_{m}:=\exists\mathsf{x}[ D_{m}(\mathsf{x})]$ and $\rho_m:=\exists\mathsf{x}[D_m(\mathsf{x})\;\wedge\;\forall \mathsf{y}[D_m(\mathsf{y})\rightarrow \mathsf{y=x}]]$.
Assume $Spr(\beta)$. $\alpha$ in $\mathcal{F}_\beta$ is a *limit point of order $0$* of $\mathcal{F}_\beta$ if and only if $\alpha$ is an isolated point of $\mathcal{F}_\beta$.
For each $m$, $\alpha$ is a *limit point of order $m+1$* of $\mathcal{F}_\beta$ if and only if, for each $p$, there exists a limit point $\gamma$ of order $m$ such that $\overline \alpha p\sqsubset \gamma$ and $\alpha \perp \gamma$.
Assume $n>0$ and $\alpha\in\mathcal{T}_n$. Note the following:
1. $(\mathcal{T}_n, =)\models D_0[\alpha]$ if and only if $\alpha$ is an isolated point of $\mathcal{T}_n$ if and only if either: $n=1$ or: $n>1$ and $\exists p[\alpha(p) = n-1]$.
2. $(\mathcal{T}_n, =)\models \neg D_0[\alpha]$ if and only if $\alpha$ is a limit point (of order 1) of $\mathcal{T}_n$ if and only if $n>1$ and $\alpha \in \mathcal{T}_{n-1}$.
3. $(\mathcal{T}_n, =)\models D_1[\alpha]$ if and only if $\alpha$ is an isolated point among the limit points (of order 1) of $\mathcal{T}_n$ if and only if $n>1$ and $\alpha\in \mathcal{T}_{n-1}$ and $\exists p[\alpha(p) =n-2]$.
4. $(\mathcal{T}_n, =)\models \neg D_0\;\wedge\;\neg D_1[\alpha]$, if and only if $\alpha$ is a limit point of order $2$ of $\mathcal{T}_n$ if and only if $n>2$ and $\alpha \in \mathcal{T}_{n-2}$.
5. For each $m>0$, $(\mathcal{T}_n, =)\models D_{2}[\alpha]$ if and only if $\alpha$ is an isolated point among the limit points of order $2$ if and only if $n>2$ and $\alpha\in \mathcal{T}_{n-2}$ and $\exists p[\alpha(p)= n-3]$.
6. For each $m>0$, $(\mathcal{T}_n, =)\models \bigwedge_{i<m} \neg D_i[\alpha]$ if and only if $\alpha$ is a limit point of order $m$ of $\mathcal{T}_n$ if and only if $n>m$ and $\alpha \in \mathcal{T}_{n-m}$.
7. For each $m>0$, $(\mathcal{T}_n, =)\models D_{m}[\alpha]$ if and only if $\alpha$ is an isolated point among the limit points of order $m$ if and only if $n>m$ and $\alpha\in \mathcal{T}_{n-m}$ and $\exists p[\alpha(p)= n-m-1]$.
8. For each $m>0$, $\mathcal{T}_n\models \psi_m$ if and only if $\mathcal{T}_n$ contains an isolated point of $\mathcal{T}_{n-m}$ if and only if $n>m$.
9. For each $m>0$, $\mathcal{T}_n\models \rho_m$ if and only if $\mathcal{T}_n$ contains exactly one isolated point of $\mathcal{T}_{n-m}$ if and only if $\mathcal{T}_{n-m}=\{\underline 0\}$ if and only if $n=m+1$.
After these preliminary observations, the following Theorem is easy to understand:
\[T:psirho\]$\;$
1. For each $n$, $\mathcal{T}_n$ is a transparent[^16] spread and,\
if $n>0$, then $\mathcal{I}(\mathcal{T}_n)= \{\alpha \in \mathcal{T}_n\mid\exists p[\alpha(p)+1=n]\}$ and $\mathcal{L}(\mathcal{T}_n)=\mathcal{T}_{n-1}$.
2. For all $n$, for all $m>0$, $\mathcal{T}_n=\{\alpha \in \mathcal{T}_{n+m}\mid(\mathcal{T}_{n+m},=)\models \bigwedge_{i<m} \neg D_i[\alpha]\}$.
3. For all $m$, $\{\underline 0\}=\mathcal{T}_1=\{\alpha \in \mathcal{T}_{m+1}\mid(\mathcal{T}_{m+1},=)\models \bigwedge_{i<m} \neg D_i[\alpha]\}$.
4. For all $n>0$, for all $m>0$, $(\mathcal{T}_n, =)\models \psi_m$ if and only if $m+1\le n$.
5. For all $n>0$, for all $m>0$, $(\mathcal{T}_n, =)\models \rho_m$ if and only if $m+1=n$.
Use the preliminary observations preceding this Theorem.
For all $n, m$, if $n\neq m$, then there exists a sentence $\psi$ such that $(\mathcal{T}_m,=)\models \psi$ and $(\mathcal{T}_n, =)\models\neg \psi$.
Finite and infinite sums of toy spreads
=======================================
A main result
-------------
Assume $Spr(\beta), Spr(\gamma)$.
We define: $\mathcal{F}_\beta\uplus\mathcal{F}_\gamma:=\{\langle 0\rangle\ast\delta\mid\delta\in\mathcal{F}_\beta\}\cup\{\langle 1\rangle\ast\delta\mid\delta\in\mathcal{F}_\gamma\}$.
For each $m$, we define: $m\otimes\mathcal{F}_\beta:=\{\langle i\rangle\ast\delta\mid i<m, \delta\in\mathcal{F}_\beta\}$.
We also define: $\omega\otimes\mathcal{F}_\beta:=\{\langle i\rangle\ast\delta\mid i\in\omega, \delta\in\mathcal{F}_\beta\}$.
Note that $\mathcal{F}_\beta\uplus\mathcal{F}_\gamma$, $m\otimes\mathcal{F}_\beta$ and $\omega\otimes\mathcal{F}_\beta$ are spreads again.
We also define, for all $m,n>0$, sentences $\psi_m^n$ and $\rho_m^n$, as follows:
$\psi^n_{m}:=\exists\mathsf{x}_0\exists\mathsf{x_1}\ldots \exists\mathsf{x}_{n-1}[\bigwedge_{i<j<n}[AP(\mathsf{x}_i, \mathsf{x}_j) \;\wedge\;\bigwedge_{i<n}\bigwedge_{j<m}\neg D(\mathsf{x}_j)]$.
and $\rho^n_m:=\exists\mathsf{x}_0\exists\mathsf{x_1}\ldots \exists\mathsf{x}_{n-1}[\bigwedge_{i<j<n}[AP(\mathsf{x}_i, \mathsf{x}_j) \;\wedge\;\bigwedge_{i<n}\bigwedge_{j<m}\neg D_j(\mathsf{x}_i)\;\wedge\\\forall \mathsf{z}[\bigwedge_{j<m} \neg D_j(\mathsf{z})\rightarrow \bigvee_{i<n} \mathsf{z}=\mathsf{x}_i] ]$.
The sentence $\psi^n_m$ expresses: ‘*there exist (at least) $n$ limit points of order $m$ that are mutually apart*’.
The sentence $\rho^n_m$ expresses: ‘*there exist exactly $n$ limit points of order $m$ that are mutually apart*’.
\[T:tauplustau\]
1. For all $m,n,p,q>0$,\
$(n\otimes \mathcal{T}_m,=)\models \psi^q_p$ if and only if either: $p+1< m$ or: $p+1=m$ and $q\le n $.
2. For all $m,n,p,q>0$, $(n\otimes \mathcal{T}_m,=)\models \rho_p^q$ if and only if $p+1=m$ and $n=q$.
3. For all $m,p,q>0$, $(\omega\otimes \mathcal{T}_m,=)\models \psi^q_p$ if and only if $p< m$.
\(i) Note the following:
If $p+1<m$ and $n>0$, then $\mathcal{T}_m$ and also $n\otimes\mathcal{T}_m$ contain infinitely many limit points of order $p$ that are mutually apart.
If $p+1=m$ and $n>0$, then $n \otimes \mathcal{T}_m$ contains exactly $n$ limit points of order $p$ that are mutually apart: the points $\langle i\rangle\ast \underline 0$, where $i<n$, so $(n\otimes \mathcal{T}_m,=)\models \psi^q_p$ if and only if $q\le n$.
If $p< m$, then $\omega \times \mathcal{T}_m$ contains infinitely many limit points of order $p$ that are mutually apart.
The proofs of (i), (ii) and (iii) follow easily from these observations.
$\;$
For each $k$, for each $s$ in $\omega^k$, we define: $\mathcal{T}_s=\bigcup_{i<k}\{\langle i\rangle \ast\delta\mid \delta \in \mathcal{T}_{s(i)}\}$.
For each $\alpha$, we define: $\mathcal{T_\alpha}:=\bigcup_i \{\langle i \rangle\ast \delta\mid\delta \in \mathcal{T}_{\alpha(i)}\}$.
Let $\mathcal{F}_0, \mathcal{F}_1\subseteq\mathcal{N}$ and assume $\varphi:\mathcal{F}_0\rightarrow\mathcal{F}_1$.
$\varphi$ is a *(surjective)* map from $\mathcal{F}_0$ *onto* $\mathcal{F}_1$ if and only if $\forall \beta \in \mathcal{F}_1\exists \alpha\in \mathcal{F}_0[\varphi|\alpha =\beta]$.
$\mathcal{F}_0$ *is equivalent to* $\mathcal{F}_1$, notation: $\mathcal{F}_0\sim\mathcal{F}_1$, if and only if there exists $\varphi:\mathcal{F}_0\rightarrow\mathcal{F}_1$ that is both injective[^17] and surjective.
\[T:finitesumstoys\]
1. For each $m$, $\mathcal{T}_m\oplus\mathcal{T}_{m+1}\sim \mathcal{T}_{m+1}$.
2. For all $m,n$, if $m<n$, then $\mathcal{T}_m\oplus\mathcal{T}_n\sim\mathcal{T}_{n}$.
3. For all $k$, for all $s$ in $\omega^k$, there exist $m,n$ such that $\mathcal{T}_s\sim n\otimes \mathcal{T}_m$.
\(i) Let $m$ be given. Define $\varphi:\mathcal{T}_m\oplus \mathcal{T}_{m+1}\rightarrow \mathcal{T}_{m+1}$ such that, for all $\delta$ in $\mathcal{T}_m$, $\varphi|\langle 0\rangle \ast\delta = \langle 1\rangle\ast S\circ\delta$, and, for each $\delta$ in $\mathcal{T}_{m+1}$, $\varphi|\langle 1\rangle \ast \delta = \langle 0 \rangle \ast \delta$. Clearly, $\varphi$ is a one-to-one function mapping $\mathcal{T}_m\oplus \mathcal{T}_{m+1}$ onto $\mathcal{T}_{m+1}$.
\(ii) Let $m$ be given. We use induction on $n$. The case $n=m+1$ has been treated in (i). Now let $n$ be given such that $m<n$ and $\mathcal{T}_m\oplus\mathcal{T}_n\sim\mathcal{T}_n$. Then $\mathcal{T}_m\oplus \mathcal{T}_{n+1}\sim \mathcal{T}_m\oplus (\mathcal{T}_n\oplus \mathcal{T}_{n+1})\sim(\mathcal{T}_m\oplus \mathcal{T}_n)\oplus \mathcal{T}_{n+1} \sim \mathcal{T}_n\oplus \mathcal{T}_{n+1}\sim \mathcal{T}_{n+1}$.
\(iii) We use induction on $k$. If $s\in \omega^0$, then $s=\langle\;\rangle$ and $\emptyset=\mathcal{T}_s = 0\otimes \mathcal{T}_1$.
Now let $k$ be given such that for all $s$ in $\omega^k$ there exist $m,n$ such that $\mathcal{T}_s= n \otimes \mathcal{T}_m$.
Let $s=t\ast\langle p\rangle$ in $\omega^{k+1}$ be given. Find $m,n$ such that $\mathcal{T}_t=n\otimes \mathcal{T}_m$. Note: $\mathcal{T}_s\sim\mathcal{T}_t\oplus \mathcal{T}_{p}$ and consider several cases.
*Case (1)*: $t=\langle\;\rangle$. Then $\mathcal{T}_s=1\otimes \mathcal{T}_{p}$.
*Case (2)*: $t \neq \langle\;\rangle$ and $p<m$. Then, by (ii): $\mathcal{T}_s \sim \mathcal{T}_t\sim n\otimes \mathcal{T}_m$.
*Case (3)*: $t \neq \langle\;\rangle$ and $p=m$. Then: $\mathcal{T}_s \sim \mathcal{T}_t\oplus \mathcal{T}_m\sim (n+1)\otimes \mathcal{T}_m$.
*Case (4)*: $t \neq \langle\;\rangle$ and $p>m$. Then, by (ii): $$\mathcal{T}_s \sim\mathcal{T}_t\oplus \mathcal{T}_{p
}\sim\underbrace{\mathcal{T}_m\oplus\ldots\oplus\mathcal{T}_m}_{n}\oplus \mathcal{T}_{p
}\sim\mathcal{T}_p\sim 1\otimes \mathcal{T}_{p}.$$
\[T:uncountablymany\]$\;$
1. For each $\alpha$, $\mathcal{I}(\mathcal{T}_\alpha)=\bigcup_i\{\langle i\rangle \ast\delta\mid \delta \in \mathcal{T}_{\alpha(i)}\;\wedge\;\exists p[\delta(p)+1=\alpha(i)]\}$.
2. For all $\alpha$, for all $n$, $(\mathcal{T}_\alpha,=)\models \psi_n$ if and only if $\exists i[\alpha(i)>n]$.
3. For all $\alpha$, for all $n$, $(\mathcal{T}_\alpha,=)\models \rho_n$ if and only if\
$\exists i[\alpha(i)=n+1\;\wedge\;\forall j[\alpha(j)=n+1\rightarrow i=j]]$.
4. For all $\zeta, \eta$ in $[\omega]^\omega$, if $\zeta\perp\eta$ and $\zeta(0)=\eta(0)=2$,\
then there exists a sentence $\psi$ such that $(\mathcal{T}_\zeta,=)\models \psi$ and $(\mathcal{T}_\eta,=)\models \neg\psi$.
\(i) Use Theorem \[T:psirho\](i).
\(ii) Note that, for each $\alpha$, for each $n$, $(\mathcal{T}_\alpha,=)\models \psi_n$ if and only if $\mathcal{T}_\alpha$ contains a limit point of order $n$ if and only if $\exists i[\alpha(i)>n]$.
\(iii) Note that, for each $\alpha$, for each $n$, $(\mathcal{T}_\alpha,=)\models \rho_n$ if and only if $\mathcal{T}_\alpha$ contains exactly one limit point of order $n$ if and only if\
$\exists i[\alpha(i)=n+1\;\wedge\;\forall j[\alpha(j)=n+1\rightarrow i=j]]$.
\(iv) Using (iii), note that, for all $\zeta$ in $[\omega]^\omega$, if $\zeta(0)>1$, then $\forall n[(\mathcal{T}_\zeta,=) \models \rho_n$ if and only if $\exists p[\zeta(p)= n+1]$.
Conclude that, for all $\zeta,\eta$ in $[\omega]^\omega$, for all $p$, if $\zeta(0)=\eta(0)=2$ and $\zeta \perp\eta$ and\
$p:=\mu i[\zeta(i)\neq \eta(i)]$ and $\zeta(p)<\eta(p)$, then $\neg \exists i[\eta(i)=\zeta(p)]$, and, therefore,\
$(\mathcal{T}_\zeta,=)\models \psi_{\zeta(p)-1}$ and $(\mathcal{T}_\eta,=)\models \neg \psi_{\zeta(p)-1}$.
Finitary spreads suffice
------------------------
Assume $Spr(\beta)$. $\beta$ is called a *finitary spread-law* or a *fan-law* if and only if $\exists \gamma\forall s[\beta(s)=0\rightarrow \forall n[\beta(s\ast\langle n \rangle)=0\rightarrow n\le \gamma(s)]]$.
$\mathcal{X}\subseteq \mathcal{N}$ is a *fan* if and only if there exists a fan-law $\beta$ such that $\mathcal{X}=\mathcal{F}_\beta$.
Note that the toy spreads $\mathcal{T}_0, \mathcal{T}_1, \ldots$ are fans.
The set $\mathcal{T}_\alpha$, however, is a spread but, in general, not a fan.
Define, for each $\alpha$, $\mathcal{T}^\ast_\alpha:= \overline{\bigcup_n \underline{\overline 0} n\ast\langle 1\rangle\ast \mathcal{T}_{\alpha(n)}}$.[^18]
Note that, for each $\alpha$, $\mathcal{T}_\alpha^\ast$ is a fan.
One may prove a statement very similar to Theorem \[T:uncountablymany\](iv):
> *For all $\zeta, \eta$ in $[\omega]^\omega$, if $\zeta\perp\eta$ and $\zeta(0)=\eta(0)=2$, then there exists a sentence $\psi$ such that $(\mathcal{T}^\ast_\zeta, =)\models \psi$ and $(\mathcal{T}^\ast_\eta, =)\models\neg \psi$.*
Comparison with an older theorem
--------------------------------
The first-order theory $DLO$ of *dense linear orderings without endpoints* is formulated in a first-order language with binary predicate symbols $=$ and $\sqsubset$ and consists of the following axioms:
1. $\mathsf{\forall x[x\sqsubset x]}$,
2. $\mathsf{\forall x\forall y\forall z[(x\sqsubset y\;\wedge\;y\sqsubset z)\rightarrow x\sqsubset z]}$,
3. $\mathsf{\forall x\forall y[
\bigl(\neg (x\sqsubset y)\;\wedge\;\neg (y\sqsubset x)\bigr)\rightarrow x=y]}$.
4. $\mathsf{\forall x\forall y[x\sqsubset y\rightarrow\forall z[x\sqsubset z\;\vee\;z\sqsubset y]]}$,
5. $\mathsf{\forall x\exists y[x\sqsubset y]\;\wedge\;\forall x\exists y[y\sqsubset x]}$,
6. $\mathsf{\forall x\forall y[x\sqsubset y\rightarrow\exists z[x\sqsubset z\;\wedge\;z\sqsubset y]]}$, and
7. axioms of equality.
$(\mathcal{R}, =_\mathcal{R}, <_\mathcal{R})$ realizes $DLO$.
Let $DLO^-$ be the theory one obtains from $DLO$ by leaving out axiom (4). If one defines a relation $<'_\mathcal{R}$ on $\mathcal{R}$ by: $\forall x\forall y[x<'_\mathcal{R} y\leftrightarrow \neg\neg(x<_\mathcal{R}y)]$, then $(\mathcal{R}, =_\mathcal{R}, <'_\mathcal{R})$ realizes $DLO^-$ but not $DLO$.
In [@veldmanjanssen90 Theorem 2.4] one constructs a function $\alpha\mapsto A_\alpha$ associating to each element $\alpha$ of $2^\omega=\mathcal{C}$ a subset $A_\alpha$ of the set $\mathcal{R}$ of the real numbers such that, for each $\alpha $ in $\mathcal{C}$, $A_\alpha$ is dense in $(\mathcal{R}, <_\mathcal{R})$, and, for all $\alpha, \beta$ in $\mathcal{C}$, if $\alpha\perp \beta$, then there exists a sentence $\psi$ such that $(A_\alpha, <_\mathcal{R})\models \psi$ and $(A_\beta, <_\mathcal{R})\models \neg \psi$.
Note: each structure $(A_\alpha, <_\mathcal{R})$ realizes $DLO$. The (intuitionistic) theory $DLO$ thus has continuum many complete extensions. [^19]
Theorem \[T:uncountablymany\](iii) strengthens this result.
One may obtain the result of \[T:uncountablymany\](iii) with subsets of $\mathcal{R}$ as well as with subsets of $\mathcal{N}$. Define an infinite sequence $\mathcal{U}_0, \mathcal{U}_1, \ldots$ of subsets of $\mathcal{R}$ by:
$\mathcal{U}_0:=\emptyset$ and $\mathcal{U}_1:=\{0_\mathcal{R}\}$, and for each $m>0$, $\mathcal{U}_{m+1}= \overline{\bigcup_n \frac{1}{2^{n+1}}+\frac{1}{2^{n+2}}\cdot_\mathcal{R} \mathcal{U}_m}.$[^20]
For each $m$, one may define $\varphi:\mathcal{T}_m\rightarrow \mathcal{U}_m$ such that $\varphi$ is surjective and satisfies: $\forall \delta\in\mathcal{T}_m\forall \zeta \in \mathcal{T}_m[\delta\perp\zeta\leftrightarrow \varphi|\delta\;\#_\mathcal{R}\;\varphi|\zeta]$.
It follows that, for each $m$, the structures $(\mathcal{T}_m,=)$ and $(\mathcal{U}_m, =_\mathcal{R})$ are elementarily equivalent.
Define, for each $\alpha$ in $[\omega]^\omega$, $A_\alpha:=\bigcup_n n+_\mathcal{R}\mathcal{U}_{\alpha(n)}$.
Note: for all $\alpha,\beta$ in $[\omega]^\omega$, if $\alpha\perp \beta$, then there exists a sentence $\psi$ such that $(A_\alpha, =_\mathcal{R})\models \psi$ and $(A_\beta, =_\mathcal{R})\models \neg\psi$.
We thus obtain from Theorem \[T:uncountablymany\] a result similar to [@veldmanjanssen90 Theorem 2.4], this time using not the ordering relation $<_\mathcal{R}$ but only the equality relation $=_\mathcal{R}$.
Note that the relation $=_\mathcal{R}$ is definable in the structure $(\mathcal{R},<_\mathcal{R})$ as\
$\forall x\in\mathcal{R}\forall y \in \mathcal{R}[x=_\mathcal{R}y\leftrightarrow \bigl(\neg(x<_\mathcal{R}y)\;\wedge\;\neg(y<_\mathcal{R}x)\bigr)]$.
The Vitali equivalence relation {#S:vitali}
===============================
For all $\alpha, \beta$, we define $$\alpha\sim_V\beta\leftrightarrow \exists n\forall m>n[\alpha(m)=\beta(m)].$$
The relation $\sim_V$ will be called the *Vitali equivalence relation*.
This is because the relation $\sim_V$ on $\mathcal{N}$ resembles the relation $\sim_\mathbb{Q}$ on the set $\mathcal{R}$ of the real numbers defined by: $$x\sim_\mathbb{Q}y\leftrightarrow \exists q\in\mathbb{Q}[x-_\mathcal{R}y=q].$$
The relation $\sim_\mathbb{Q}$ has played an important rôle in classical set theory.
If one constructs, using the axiom of choice, within the interval $[0,1]$, a *transversal* for this equivalence relation, that is: a complete set of mutually inequivalent representatives, one obtains a set that is not Lebesgue measurable. This discovery is due to G. Vitali.
Note: $(\mathcal{N}, \sim_V)\models EQ$.
The following theorem brings to light an important difference between $(\mathcal{N},=)$ and $(\mathcal{N}, \sim_V)$.
\[D:stable\] A proposition $P$ is *stable* if and only if $\neg\neg P\rightarrow P$.
A binary relation $\sim$ on $\mathcal{N}$ is *stable* if and and only if\
$\forall\alpha\forall\beta[\neg\neg(\alpha\sim\beta)\rightarrow \alpha\sim\beta]$[^21].
\[T:vitaliunstable\]
1. $(\mathcal{N}, =)\models \mathsf{\forall x \forall y[\neg\neg(x=y)\rightarrow x=y]}$.
2. $(\mathcal{N}, \sim_V)\models \mathsf{\forall x\neg \forall y[\neg\neg(x=y)\rightarrow x=y]}$.
\(i) Note: for all $\alpha, \beta$, $\alpha=\beta\leftrightarrow \neg (\alpha\;\#\;\beta)$, and, therefore:\
$\neg\neg(\alpha=\beta)\leftrightarrow \neg \neg\neg(\alpha\;\#\;\beta)\leftrightarrow \neg(\alpha\;\#\; \beta)\leftrightarrow \alpha=\beta$.
\(ii) Let $\gamma$ be given.
Consider $\mathcal{F}^\gamma:=\{\alpha\mid\forall m\forall n[\bigl(\alpha(m)\neq\gamma(m)\;\wedge\;\alpha(n)\neq \gamma(n)\bigr)\rightarrow m=n]$.
$\mathcal{F}^\gamma$ is the set of all $\alpha$ that differ at at most one place from $\gamma$.
Note that $\mathcal{F}^\gamma$ is a spread.
We have two claims.
*First claim:* $\forall \alpha \in \mathcal{F}^\gamma[\neg\neg(\alpha\sim_V \gamma]$.
The proof is as follows. Let $\alpha$ in $\mathcal{F}^\gamma$ be given. Distinguish two cases.
*Case (1)*. $\exists n[\alpha(n)\neq \gamma(n)]$. Find $n$ such that $\alpha(n)\neq \gamma(n)$ and conclude:\
$\forall m>n[\alpha(m)=\gamma(m)]$ and $\alpha\sim_V\gamma$.
*Case (2)*. $\neg\exists n[\alpha(n)\neq \gamma(n)]$. Conclude: $\forall n[\alpha(n)=\gamma(n)]$ and $\alpha\sim_V\gamma$.
We thus see: if $\exists n[\alpha(n)\neq \gamma(n)]\;\vee\;\neg\exists n[\alpha(n)\neq \gamma(n)]$, then $\alpha\sim_V\underline \gamma$.
As $\neg\neg(\exists n[\alpha(n)\neq \gamma(n)]\;\vee\;\neg\exists n[\alpha(n)\neq \gamma(n)])$, also $\neg\neg(\alpha\sim_V\gamma)$.
*Second claim:* $\neg\forall \alpha \in \mathcal{F}^\gamma[\alpha \sim \gamma]$.
In order to see this, assume: $\forall \alpha \in \mathcal{F}^\gamma[\alpha \sim \gamma]$, that is:\
$\forall \alpha \in \mathcal{F}\exists n\forall m>n[\alpha(m)=\gamma(m)]$. Using Lemma \[L:bcpspreads\], find $p,n$ such that\
$\forall \alpha \in \mathcal{F}^\gamma[\overline{\gamma}p\sqsubset \alpha\rightarrow \forall m>n[\alpha(m)=\gamma(m)]]$. Define $m:=\max(p,n+1)$ and define $\alpha$ such that $\forall n[\alpha(n)\neq\gamma(n)\leftrightarrow n=m]$. Note: $\overline{\gamma} p\sqsubset \alpha$ and $m>n$ and $\alpha(m)\neq \gamma(m)$. Contradiction.
Combining our two claims, we see:\
not: for all $\alpha$, if $\neg\neg(\alpha\sim_V \gamma)$ then $\alpha\sim_V \gamma$.
Conclude: $(\mathcal{N}, \sim_V)\models \mathsf{\forall x\neg\forall y[\neg\neg(x=y)\rightarrow x=y]}$.
It follows from Theorem \[T:vitaliunstable\] that there is no relation $\#_V$ on $\mathcal{N}$ satisfying the requirements of an apartness relation[^22] with respect to $\sim_V$:
1. $\forall \alpha\forall \beta[\neg (\alpha\;\#_V\;\beta)\leftrightarrow \alpha \sim_V\beta]$
2. $\forall\alpha\forall\beta[\alpha\;\#_V\;\beta\rightarrow \beta\;\#_V\;\alpha]$
3. $\forall \alpha\forall \beta[\alpha\;\#_V\;\beta \rightarrow\forall \gamma[\alpha\;\#_V\;\gamma\;\vee\;\gamma\;\#_V\;\beta]]$.
The existence of an apartness $\#_V$ would imply, by the first one of these requirements, that $\sim_V$ is a stable relation, as, for any proposition $P$, $\neg\neg\neg P\leftrightarrow\neg P$.
The next Theorem now is no surprise:
\[T:vitalinoapartness\] $(\mathcal{N}, \sim_V)\models \mathsf{\forall x \forall y[\neg}AP(\mathsf{x,y})]$.
Let $\alpha, \beta$ be given.
Assume $(\mathcal{N},\sim_V)\models AP[\alpha, \beta]$, that is, $\forall \gamma[\gamma\nsim_V \alpha\;\vee\;\gamma\nsim_V \beta]$.
Applying Lemma \[L:bcpdisj\], find $p$ such that\
either $\forall \gamma [\overline{\alpha}p\sqsubset \gamma \rightarrow \gamma \nsim_V \alpha]$ or $\forall \gamma [\overline{\alpha}p\sqsubset \gamma \rightarrow \gamma \nsim_V \beta]$.
The first of these two alternatives is wrong, as $\overline \alpha p\sqsubset \alpha \;\wedge\; \alpha \sim_V \alpha$.
Conclude: $\forall \gamma [\overline{\alpha}p\sqsubset \gamma \rightarrow \gamma \nsim_V \beta]$.
Define $\gamma$ such that $\overline\alpha p\sqsubset \gamma $ and $\forall i>p[\gamma(i)=\beta(i)]$.
Note: $\overline \alpha p\sqsubset \gamma \;\wedge\; \gamma \sim_V \beta $.
Contradiction.
Conclude: $(\mathcal{N}, =_V)\models \neg AP[\alpha, \beta]$.
We thus see: $(\mathcal{N}, =_V)\models \mathsf{\forall x \forall y[\neg}AP(\mathsf{x,y})]$.
Clearly, the relation defined by the formula $AP$ in the structure $(\mathcal{N}, \sim_V)$ fails to satisfy the first requirement for an apartness relation with respect to $\sim_V$.
It follows from Theorem \[T:vitalinoapartness\] that $(\mathcal{N}, \sim_V)$, while realizing $T_{inf}$, does not realize $T^+_{inf}$, see Definitions \[D:Tinf\] and \[D:Tinfplus\].
A first Vitali variation {#S:vitalivar}
========================
There are many intuitionistic versions of the classical Vitali equivalence relation. This is obvious to someone who knows that there are many variations upon the notion of a finite and decidable subset of $\mathbb{N}$, see [@veldman1995] and [@veldman2005 Section 3].
We define an infinite sequence $\sim^0_V, \sim^1_V, \ldots$ of relations on $\mathcal{N}$ such that $\sim^0_V\;=\;\sim_V$ and, for each $i$,
$$\alpha\sim_V^{i+1}\beta\leftrightarrow \exists n\forall m>n[\alpha(m)\neq\beta(m)\rightarrow \alpha\sim^i_V\beta].$$ We also define: $$\alpha\sim^\omega_V\beta\leftrightarrow \exists i[\alpha\sim^i_V\beta].$$
\[T:vitalivariations\]$\;$
1. $\forall i \forall n\forall s \in \omega^n\forall t \in \omega^n\forall \alpha\forall \beta[s\ast \alpha\sim^i_V t\ast\beta \leftrightarrow \alpha\sim_V^i \beta]$.
2. $\forall i\forall \alpha\forall \beta [\alpha\sim_V^{i}\beta\rightarrow \alpha \sim_V^{i+1}\beta]$.
3. $\forall i\forall\gamma\neg\forall \alpha[\alpha\sim_V^{i+1}\gamma\rightarrow \alpha \sim_V^i\gamma]$.
4. $\forall i\forall j\forall \alpha\forall \beta\forall \gamma[(\alpha \sim_V^i \beta\;\wedge\;\beta \sim_V^j \gamma)\rightarrow \alpha\sim_V^{i+j}\gamma]$.
5. $\sim_V^\omega$ is an equivalence relation on $\mathcal{N}$.
\(i) One proves this easily by induction.
\(ii) Obvious.
\(iii) Let $\gamma$ be given.
For each $i$, define $\mathcal{F}^\gamma_i:=\{\alpha\mid \forall s \in [\omega]^{i+1}\exists j<i+1[\alpha\circ s(j)=\gamma\circ s(j)]\}$.
Note: for each $i$, $\mathcal{F}^\gamma_i$ is a spread, and $\mathcal{F}^\gamma_i \subsetneq \mathcal{F}^\gamma_{i+1}$.
For each $i$, $\mathcal{F}^\gamma_i$ consists of all $\alpha$ that assume at most $i$ times a value different from the value assumed by $\gamma$. In particular, $\mathcal{F}^0_\gamma=\{\gamma\}$.
Note: for all $i, m, \alpha, \beta$,\
if $m=\mu n[\alpha(n)\neq\gamma(n)]$ and $\alpha =\overline{\alpha}(m+1)\ast\beta$, then $\alpha \in \mathcal{F}^\gamma_{i+1}\leftrightarrow \beta \in \mathcal{F}^\gamma_i$.
We have two claims.
*First claim:* $\forall i\forall \alpha\in \mathcal{F}^\gamma_i[\alpha\sim_V^i\gamma]$.
We prove this claim by induction.
The starting point of the induction is the observation:\
$\forall \alpha \in \mathcal{F}_0^\gamma[\alpha =\gamma]$, so $\forall \alpha \in \mathcal{F}^\gamma_0[\alpha \sim_V^0\gamma]$.
Now assume $i$ is given such that $\forall \alpha \in \mathcal{F}_i^\gamma[\alpha \sim^i_V\gamma]$.
Assume $\alpha\in\mathcal{F}^\gamma_{i+1}$ and $\exists n[\alpha(n)\neq \gamma(n)]$. Find $n$ such that $\alpha(n)\neq \gamma(n)$. Find $\beta$ such that $\alpha=\overline\alpha(n+1)\ast\beta$, and note: $\beta \in \mathcal{F}^\gamma_i$ and thus, by the induction hypothesis, $\beta\sim_V^i \gamma$. Conclude, using (i): $\alpha \sim_V^i \gamma$.
We thus see:\
$\forall \alpha \in \mathcal{F}^\gamma_{i+1}[\exists n[\alpha(n)\neq \gamma(n)]\rightarrow \alpha \sim^i_V\gamma]$, that is: $\forall \alpha \in \mathcal{F}^\gamma_{i+1}[\alpha\sim^{i+1}_V \gamma]$.
This completes the proof of the induction step.
*Second claim:* $\forall i \neg \forall \alpha \in \mathcal{F}^\gamma_{i+1}[\alpha\sim^i_V \gamma]$.
We again use induction.
We first prove: $\neg \forall \alpha \in \mathcal{F}^\gamma_1[\alpha\sim_V \gamma]$.
Assume $\forall \alpha \in \mathcal{F}^\gamma_1[\alpha\sim_V \gamma]$, that is: $\forall \alpha \in \mathcal{F}_1\exists n\forall m>n[\alpha(m) =\gamma(m)]$.
Note: $\gamma \in \mathcal{F}_1^\gamma$ and $\mathcal{F}^1_\gamma$ is a spread.
Using Lemma \[L:bcpspreads\], find $p, n$ such that $\forall \alpha \in \mathcal{F}_1[\overline \gamma p\sqsubset\alpha\rightarrow \forall m>n[\alpha(m)=\gamma(m)]$. Define $m:=\max(n+1,p)$ and define $\alpha$ such that $\forall n[\alpha(n)=\gamma(n)\leftrightarrow n\neq m]$. Note: $\alpha \in \mathcal{F}_1$ and $\overline \gamma p\sqsubset \alpha$ and $\alpha(m)\neq \gamma(m)$ and $m>n$. Contradiction.
Conclude: $\neg \forall \alpha \in \mathcal{F}^\gamma_1[\alpha\sim_V \gamma]$.
Now let $i$ be given such that $ \neg \forall \alpha \in \mathcal{F}^\gamma_{i+1}[\alpha\sim_V^i \gamma]$.
We want to prove: $ \neg \forall \alpha \in \mathcal{F}^\gamma_{i+2}[\alpha\sim_V^{i+1} \gamma]$.
Assume: $\forall \alpha \in \mathcal{F}^\gamma_{i+2}[\alpha\sim_V^{i+1} \gamma]$, that is:\
$\forall \alpha \in \mathcal{F}^\gamma_{i+2}\exists n\forall m>n[\alpha(m)\neq \gamma(m)\rightarrow \alpha\sim^{i+1}_V\gamma]$. Using Lemma \[L:bcpspreads\], find $p,n$ such that $\forall \alpha \in \mathcal{F}^\gamma_{i+2}[(\overline \gamma p\sqsubset \alpha \;\wedge\;m>n\;\wedge\;\alpha(m)\neq \gamma(m))\rightarrow \alpha\sim^i_V\gamma]$. Define $m:=\max(n+1, p)$. Let $\beta$ in $\mathcal{F}^\gamma_{i+1}$ be given. Define $\alpha$ such that $m=\mu n[\alpha(n)\neq \gamma(n)]$ and $\forall n> m[\alpha(n)=\beta(n)]$. Note: $\alpha \in \mathcal{F}^\gamma_{i+2}$ and $\alpha(m)\neq \gamma(m)$ and $m>n$, so $\alpha \sim_V^{i}\gamma$, and, therefore, by (i), $\beta\sim_V^{i}\gamma$. We thus see: $\forall \beta \in \mathcal{F}^\gamma_{i+1}[\beta\sim_V^i\gamma]$ and, by the induction hypothesis, obtain a contradiction.
This completes the proof of the induction step.
Taking our first and second claim together, we obtain the conclusion:\
$\forall\gamma\forall i\neg \forall \alpha[\alpha\sim^{i+1}_V \gamma\rightarrow \alpha\sim^{i}_V \gamma]$.
\(iv) We have to prove:
for all $i$, for all $j$, $\forall \alpha\forall \beta\forall \gamma[(\alpha \sim_V^i \beta\;\wedge\;\beta \sim_V^j \gamma)\rightarrow \alpha\sim_V^{i+j}\gamma]$.
We use induction on $i+j$ and distinguish four cases.
*Case (1):* $i=j=0$. Assume $\alpha\sim^0_V\beta$ and $\beta\sim^0_V\gamma$. Find $n,p$ such that $\forall m>n[\alpha(m)=\beta(m)]$ and $\forall m>p[\beta(m)=\gamma(m)$. Define $q:=\max(n, p)$ and note: $\forall m>q[\alpha(m)=\gamma(m)]$. Conclude: $\alpha\sim^0_V\gamma$.
*Case (2):* $i=0$ and $j>0$. Assume $\alpha\sim^0_V\beta$ and $\beta\sim^j_V\gamma$. Find $n,p$ such that $\forall m>n[\alpha(m)=\beta(m)]$ and $\forall m>p[\beta(m)\neq\gamma(m)\rightarrow \beta\sim^{j-1}_V\gamma]$. Define $q:=\max(n,p)$.
Assume $m>q$ and note: if $\alpha(m)\neq \gamma(m)$, then $\beta(m) \neq \gamma(m)$ and $\beta\sim^{j-1}_V\gamma$. Using the induction hypothesis, conclude: $\alpha\sim^{j-1}_V \gamma$.
We thus see: $\forall m>q[\alpha(m)\neq \gamma(m)\rightarrow \alpha\sim^{j-1}_V\gamma]$, that is: $\alpha\sim^j_V\gamma$.
*Case (3):* $i>0$ and $j=0$. Assume $\alpha\sim^i_V\beta$ and $\beta\sim^0_V\gamma$. Find $n,p$ such that $\forall m>n[\alpha(m)\neq\beta(m)\rightarrow \alpha\sim^{i-1}_V \beta]$ and $\forall m>p[\beta(m)=\gamma(m)]$. Define $q:=\max(n,p)$.
Assume $m>q$ and note: if $\alpha(m)\neq \gamma(m)$, then $\alpha(m) \neq \beta(m)$ and $\alpha\sim^{i-1}_V\beta$. Using the induction hypothesis, conclude: $\alpha\sim^{i-1}_V \gamma$.
We thus see: $\forall m>q[\alpha(m)\neq \gamma(m)\rightarrow \alpha\sim^{i-1}_V\gamma]$, that is: $\alpha\sim^i_V\gamma$.
*Case (4):* $i>0$ and $j>0$. Assume $\alpha\sim^i_V\beta$ and $\beta\sim^j_V\gamma$. Find $n,p$ such that $\forall m>n[\alpha(m)\neq\beta(m)\rightarrow \alpha\sim^{i-1}_V \beta]$ and $\forall m>p[\beta(m)\neq\gamma(m)\rightarrow \beta\sim^{j-1}_V\gamma]$. Define $q:=\max(n,p)$.
Assume $m>q$ and $\alpha(m)\neq \gamma(m)$. Then *either:* $\alpha(m)\neq\beta(m)$ and $\alpha\sim^{i-1}\beta$, and, by the induction hypothesis, $\alpha\sim^{i+j-1}_V\gamma$, *or:* $\beta(m)\neq\gamma(m)$ and $\beta\sim^{j-1}\gamma$ and, by the induction hypothesis, $\alpha\sim^{i+j-1}\gamma$.
We thus see: $\forall m>q[\alpha(m)\neq\gamma(m)\rightarrow \alpha\sim^{i+j-1}\gamma]$. Conclude: $\alpha\sim^{i+j}\gamma$.
\(v) is an easy consequence of (iv).
The next Theorem shows that the structures $(\mathcal{N}, \sim_V)$ and $(\mathcal{N}, \sim^\omega_V)$ have a property in common.
\[T:omegaunstable\] $\;$
$(\mathcal{N}, \sim_V^\omega)\models \mathsf{\forall x\neg\forall y[\neg\neg(x=y)\rightarrow x =y]}$.
Let $\gamma$ be given.
We repeat a definition we gave in the proof of Theorem \[T:vitalivariations\](iii).
For each $i$, $\mathcal{F}^\gamma_i:=\{\alpha\mid\forall s \in [\omega]^{i+1}\exists j<i+1[\alpha\circ s(j)=\gamma\circ s(j)]\}$.
In the proof of Theorem \[T:vitalivariations\](iii), we saw: $\forall i\forall \alpha \in \mathcal{F}^\gamma_i[\alpha\sim^i_V \gamma]$.
Conclude: $\forall i\forall \alpha \in \mathcal{F}^\gamma_i[\alpha\sim^\omega_V \gamma]$.
We now define: $\mathcal{F}^\gamma_\omega:=\{\alpha\mid \forall i[i=\mu n[\alpha(n)\neq \gamma(n)]\rightarrow\alpha\in \mathcal{F}_{i+1}]\}$.
Like each $\mathcal{F}^\gamma_i$, $\mathcal{F}^\gamma_\omega$ is a spread, and $\gamma \in \mathcal{F}^\gamma_\omega$.
We have two claims.
*First claim:* $\forall \alpha \in \mathcal{F}^\gamma_\omega[\neg\neg(\alpha\sim_V^\omega \gamma)]$.
The argument is as follows. Let $\alpha$ in $\mathcal{F}^\gamma_\omega$ be given and distinguish two cases.
*Case (1):* $\neg\exists n[\alpha(n)\neq \gamma(n)]$. Then $\alpha=\gamma$ and $\alpha\sim_V^\omega \gamma$.
*Case (2):* $\exists n[\alpha(n)\neq \gamma(n)]$. Find $i:=\mu n[\alpha(n)\neq\gamma(n)]$. Note: $\alpha \in \mathcal{F}_{i+1}^\gamma$ and $\alpha\sim^\omega_V \gamma$.\
As $\neg\neg (\exists n[\alpha(n)\neq \gamma(n)]\;\vee\;\neg\exists n[\alpha(n)\neq \gamma(n)])$, also $\neg\neg(\alpha\sim_V^\omega \gamma)$.
*Second claim:* $\neg \forall \alpha \in \mathcal{F}^\gamma_\omega[\alpha\sim_V^\omega \gamma]$.
In order to see this, assume: $\forall \alpha \in \mathcal{F}^\gamma_\omega[\alpha\sim_V^\omega \gamma]$, that is: $\forall \alpha \in \mathcal{F}_\omega\exists i[\alpha\sim^i_V \gamma]$.
Using Lemma \[L:bcpspreads\], find $p,i$ such that $\forall \alpha \in \mathcal{F}^\gamma_\omega[\overline \gamma p\sqsubset \alpha\rightarrow \alpha\sim^i_V \gamma]$.
Define $q:=\max(p, i+1)$. Let $\alpha$ in $\mathcal{F}^\gamma_q$ be given. Define $\beta$ such that\
$\forall n<q[\beta(n)=\gamma(n)]$ and $\beta(q)\neq \gamma(q)$ and $\forall n>q[\beta(n)=\alpha(n)]$.
Note: $\beta\in \mathcal{F}_{q+1}$ and $q=\mu n[\beta(n)\neq\gamma(n)]$, and, therefore, $\beta\in \mathcal{F}_\omega^\gamma$.
As $\overline \gamma q\sqsubset \beta$, we conclude: $\beta \sim_V^i\gamma$.
As $\beta\sim^0_V\alpha$, also $\alpha\sim_V^i\gamma$.
We thus see: $\forall \alpha \in \mathcal{F}_q[\alpha\sim_V^i \gamma]$.
As $q>i$, this contradicts the Second claim in the proof of Theorem \[T:vitalivariations\](iii).
Taking our two claims together, we conclude:
$\forall \gamma\neg \forall\alpha\in\mathcal{F}^\gamma_\omega[\neg\neg(\alpha\sim^\omega_V\gamma)\rightarrow\alpha\sim^\omega_V\gamma]$.
Conclude: $(\mathcal{N}, \sim^\omega_V)\models \mathsf{\forall x\neg\forall y[\neg\neg(x=y)\rightarrow x =y]}$.
We did not succeed in finding a sentence $\psi$ such that $(\mathcal{N}, \sim_V)\models \psi$ and\
$(\mathcal{N}, \sim^\omega_V)\models \neg \psi$.
More and more Vitali relations {#S:morevitali}
==============================
In [@veldman1995], [@veldman1999] and [@veldman2005 Section 3], one studies the set $$\mathbf{Fin}:=\{\alpha\mid\alpha\sim_V\underline 0\}=\{\alpha\mid\exists n \forall m>n[\alpha(m)=0]\}.$$ For each $\alpha$, $\alpha\in\mathbf{Fin}$ if and only if $D_\alpha:=\{m\mid\alpha(m)\neq 0\}$ is a *finite* subset of $\mathbb{N}$.
For each $i$, the set $\{\alpha\mid\alpha\sim_V^i \underline 0\}$ is called, in [@veldman1999] and [@veldman2005], the $i$-th *perhapsive extension* of the set $\mathbf{Fin}$. It is shown, in [@veldman1995], [@veldman1999] and [@veldman2005], that the process of building perhapsive extensions of $\mathbf{Fin}$ can be carried on into the transfinite.
In a similar way, the Vitali equivalence relation $\sim_V$ admits of transfinitely many extensions.
The relation $\sim_V^\omega$ is only a *first* extension of $\sim_V$. Let us consider a second one.
Recall: $\forall\alpha\forall\beta[\alpha\sim_V^\omega\beta\sim\exists i[\alpha\sim_V^ i\beta]]$.
We define an infinite sequence $\sim_V^{\omega+0}=\sim_V^\omega, \sim^{\omega+1}_V, \sim^{\omega+2}_V, \ldots$ of relations on $\mathcal{N}$, such that, for each $i>0$, $$\alpha\sim_V^{\omega+i+1}\beta\leftrightarrow \exists n\forall m>n[\alpha(m)\neq\beta(m)\rightarrow \alpha\sim^{\omega+i}_V\beta].$$ We also define: $$\alpha\sim^{\omega+\omega}_V\beta\leftrightarrow \exists i[\alpha\sim^{\omega+i}_V\beta].$$
One may prove analogues of Theorems \[T:vitalivariations\] and \[T:omegaunstable\] and conclude:
$\sim_V^{\omega+\omega}$ is an equivalence relation on $\mathcal{N}$, properly extending $\sim_V^\omega$, that, like $\sim_V$ and $\sim_V^\omega$, is not stable in the sense of Theorem \[T:omegaunstable\].
One may continue and define $\sim_V^{\omega+\omega+\omega}$, and $\sim_V^{\omega+\omega+\omega+\omega}$ and so on.
The process of building such extensions leads further into the transfinite, as follows.
\[D:perhapsiveextvitali\] Let $R$ be binary relation on $\mathcal{N}$.
We define a binary relation $R^+$ on $\mathcal{N}$ by: $$\alpha R^+\beta\leftrightarrow \exists n\forall m>n[\alpha(m)\neq \beta(m)\rightarrow \alpha R \beta].$$
We let $\mathcal{E}$ be the least class of binary relations on $\mathcal{N}$ such that
1. the Vitali equivalence relation $\sim_V$ belongs to $\mathcal{E}$, and,
2. for every $R$ in $\mathcal{E}$, also $R^+ \in \mathcal{E}$, and,
3. for every infinite sequence $R_0, R_1, \ldots$ of elements of $\mathcal{E}$, also $\bigcup_i R_i \in \mathcal{E}$.
The elements of $\mathcal{E}$ are the *extensions of the Vitali equivalence relation*.
Note that $<_V^\omega$ and $<_V^{\omega+\omega}$ are in $\mathcal{E}$.
In general, a relation $R$ in $\mathcal{E}$ is not transitive. One may prove, for instance, that the relation $<_V^1$, while belonging to $\mathcal{E}$, is not transitive.
The next Theorem shows that $\mathcal{E}$ contains many transitive relations.
\[T:transitive\]
1. $\sim_V$ is transitive.
2. Given any transitive $R$ in $\mathcal{E}$, there exists a transitive $T$ in $\mathcal{E}$ such that $R^+\subseteq T$.
3. Given any infinite and increasing sequence $R_0\subseteq R_1\subseteq \ldots$ of transitive relations in $\mathcal{E}$, also $\bigcup_i R_i$ is a transitive relation in $\mathcal{E}$.
\(i) Obvious.
\(ii) We take our inspiration from Theorem \[T:vitalivariations\] (iv) and (v).
Let a transitive $R$ in $\mathcal{E}$ be given.
Define an infinite sequence $R^0, R^1, \ldots$ of elements of $\mathcal{E}$ such that $R^0=R$ and, for each $i$, $R^{i+1}= (R^ i)^+$.
One may prove: for all $i$, for all $j$, $\forall\alpha\forall\beta\forall \gamma[(\alpha R^i\beta\;\wedge\;\beta R^i\gamma)\rightarrow \alpha R^{i+j} \gamma]$, as it is done for the special case $R=\sim_V$ in the proof of Theorem \[T:vitalivariations\](iv).
Define $T:=\bigcup_i R^i$ and note: $T\in \mathcal{E}$, $R^+\subseteq T$ and $T$ is transitive.
\(iii) Note: for every increasing sequence $R_0\subseteq R_1\subseteq \ldots$ of transitive relations on $\mathcal{N}$, also $\bigcup_i R_i$ is transitive.
Theorem \[T:transitive\] will gain significance after Corollary \[C:hierarchy\], which shows that, for every $R$ in $\mathcal{E}$, $R\subseteq R^+$ and $\neg(R^+\subseteq R)$.
We did not succeed in proving that every $R$ in $\mathcal{E}$ extends to a transitive $T$ in $\mathcal{E}$.
\[D:shiftinv\] A binary relation $R$ on $\mathcal{N}$ is *shift-invariant* if and only if\
$\forall \alpha\forall \beta[\alpha R \beta \leftrightarrow (\alpha\circ S) R (\beta\circ S)]$.
\[L:shiftinvariant\] Every $R$ in $\mathcal{E}$ is shift-invariant.
The proof is a straightforward exercise in induction on $\mathcal{E}$. Note:
\(I) $\sim_V$ is shift-invariant.
\(II) For every binary relation $R$ on $\mathcal{N}$, if $R$ is shift-invariant, then $R^+$ is shift-invariant.
\(III) For every infinite sequence $R_0, R_1, \ldots$ of binary relations on $\mathcal{N}$, if each $R_n$ is shift-invariant, then $\bigcup_iR_i$ is shift-invariant.
Conclude: every $R$ in $\mathcal{E}$ is shift-invariant.
\[D:Estar\] We let $\mathcal{E}^\ast$ be the least class of binary relations on $\mathcal{N}$ such that
1. the Vitali equivalence relation $\sim_V$ belongs to $\mathcal{E}^\ast$, and
2. for every infinite sequence $R_0, R_1, \ldots $ of elements of $\mathcal{E}^\ast$, also $(\bigcup_i R_i)^+\in \mathcal{E}^\ast.$
\[L:eestar\] $\mathcal{E}^\ast\subseteq \mathcal{E}$ and, for all $R$ in $\mathcal{E}$, there exists $T$ in $\mathcal{E}^\ast$ such that $R\subseteq T$.
The proofs of the two statements are straightforward, by induction on $\mathcal{E}^\ast$ and $\mathcal{E}$, respectively.
\[T:hierarchy\] For each $R$ in $\mathcal{E}^\ast$, $R\subseteq R^+$ and $\neg(R^+\subseteq R)$.
For each $R$ in $\mathcal{E}$, we define $Fin_R:=\{\alpha\mid \alpha R\underline 0\}$. [^23]
We prove for each $R$ in $\mathcal{E}^\ast$ there exists a fan $\mathcal{F}$ such that $\mathcal{F}\subseteq Fin_{R^+}$ and $\neg(\mathcal{F}\subseteq Fin_R)$.
We do so by induction on $\mathcal{E}^\ast$.
\(I) Define $\mathcal{F}:=\{\alpha\mid\forall m\forall n[(\alpha(m)\neq 0 \;\wedge\;\alpha(n)\neq 0)\rightarrow m=n]\}$.
Note that $\mathcal{F}$ is a fan.
For each $\alpha$ in $\mathcal{F}$, for each $n$, if $\alpha(n)\neq 0$ then: $\forall m>n[\alpha(m)=0]$ and $\alpha \in Fin_{\sim_V}$. Conclude: for each $\alpha \in \mathcal{F}$, if $\exists n[\alpha(n)\neq 0]$, then $\alpha\in Fin_{\sim_V}$, that is: $\alpha \in Fin_{(\sim_V)^+}$. Conclude: $\mathcal{F}\subseteq Fin_{(\sim_V)^+}$.
Now assume $\mathcal{F}\subseteq Fin_{\sim_V}$, that is: $\forall \alpha \in \mathcal{F}\exists n\forall m>n[\alpha(m)=0]$. Using Lemma \[L:bcpspreads\], find $p,n$ such that $\forall \alpha \in \mathcal{F}[\underline{\overline 0}p\sqsubset \alpha\rightarrow \forall m>n[\alpha(m)=0]]$.\
Define $q:=\max(p, n+1)$ and consider $\alpha:=\underline{\overline 0}q\ast\langle 1 \rangle\ast \underline 0$. Contradiction.
Conclude: $\neg(\mathcal{F}\subseteq Fin_{\sim_V})$.
\(II) Let $R_0, R_1, \ldots$ be an infinite sequence of elements of $\mathcal{E}$.
Let $\mathcal{F}_0, \mathcal{F}_1,\ldots $ be an infinite sequence of fans such that,\
for each $n$, $\mathcal{F}_n\subseteq Fin_{(R_n)^+}$ and $\neg(\mathcal{F}_n\subseteq Fin_{ R_n})$.
Consider $R:=(\bigcup_i R_i)^+$.
Define $\mathcal{F}:=\{\alpha\mid \forall n[n=\mu i[\alpha(i)\neq 0]\rightarrow \exists \beta \in \mathcal{F}_{n'}[\alpha=\overline\alpha(n+1)\ast \beta]\}$.[^24]
Note that $\mathcal{F}$ is a fan.
We now prove: $\mathcal{F}\subseteq Fin_{R^+}$ and $\neg(\mathcal{F}\subseteq Fin_R)$.
Note that, for each $\alpha \in \mathcal{F}$, for each $n$, if $n=\mu i[\alpha(i)\neq 0]$, then there exists $\beta$ in $\mathcal{F}_{n'}$ such that $\alpha=\overline\alpha (n+1)\ast\beta$.
As, for each $n$, $\mathcal{F}_n\subseteq Fin_{(R_n)^+}\subseteq Fin_{\bigcup_i (R_i)^+}$, and $\bigcup_i(R_i)^+\subseteq \bigl(\bigcup_i R_i\bigr)^+=R$ and $R$ is shift-invariant, conclude: $\forall\alpha\in\mathcal{F}[\exists n[\alpha(n)\neq 0]\rightarrow \alpha \in Fin_R]$, that is: $\mathcal{F}\subseteq Fin_{R^+}$.
Now assume $\mathcal{F}\subseteq Fin_R$, that is: $\forall \alpha \in \mathcal{F}\exists n\forall m>n[\alpha(m)\neq 0]\rightarrow \exists i[\alpha \in Fin_{R_i}]]$. Using Lemma \[L:bcpspreads\], find $p,n$ such that\
$\forall \alpha \in \mathcal{F}[\overline{\underline 0} p\sqsubset \alpha \rightarrow \forall m >n[\alpha(m)\neq 0\rightarrow \exists i[\alpha \in Fin_{R_i}]]$.
Define $q:=\max(p, n+1)$ and note: $\forall \alpha \in \mathcal{F}[\overline{\underline 0}q\ast\langle 1\rangle\sqsubset \alpha\rightarrow \exists i[\alpha\in \mathcal{F}_i]]$.
Using Lemma \[L:bcpspreads\] again, find $r,i$ such that $\forall\alpha \in \mathcal{F}[\overline{\underline 0}q\ast\langle 1 \rangle\ast\overline{\underline 0}r\sqsubset \alpha\rightarrow \alpha \in \mathcal{F}_i]$.
Find $n\ge q+r+1$ such that $n' =i$ and define $t:=n-(q+1)$.
Note: $t\ge r$ and conclude: $\forall \beta \in \mathcal{F}_i[\overline{\underline 0}q \ast \langle 1\rangle\ast \overline{\underline 0}t\ast\langle 1 \rangle\ast \beta \in Fin_{R_i}]$.
As $R_i$ is shift-invariant, conclude: $\mathcal{F}_i\subseteq Fin_{R_i}$.
Contradiction, as $\neg(\mathcal{F}_i\subseteq Fin_{R_i})$.
Conclude: $\neg(\mathcal{F} \subseteq Fin_{R})$.
\[C:hierarchy\] For each $R$ in $\mathcal{E}$, $R\subseteq R^+$ and $\neg(R^+\subseteq R)$.
Assume we find $R$ in $\mathcal{E}$ such that $R=R^+$.
Conclude, by induction on $\mathcal{E}$: for all $U$ in $\mathcal{E}$, $U\subseteq R$.
Using Lemma \[L:eestar\], find $T$ in $\mathcal{E}^\ast$ such that $R\subseteq T$.
By Theorem \[T:hierarchy\], $T\subseteq T^+$ and $\neg(T^+\subseteq T)$.
On the other hand, $T^+\subseteq R\subseteq T$.
Contradiction.
\[D:almost\] We define binary relations $\sim_V^{\neg\neg}$ and $\sim_V^{almost}$ on $\mathcal{N}$, as follows.
For all $\alpha, \beta$, $\alpha\sim_V^{\neg\neg}\beta \leftrightarrow \neg\neg\exists n\forall m>n[\alpha(n)=\beta(n)]\leftrightarrow \neg\neg (\alpha\sim_V\beta)$, and
$\alpha \sim^{almost}_V\beta \leftrightarrow \forall \zeta \in [\omega]^\omega\exists n[\alpha\circ \zeta(n)=\beta\circ\zeta(n)]$.
$\alpha\sim_V^{almost}\beta$ if and only if the set $\{n\mid \alpha(n)\neq \beta(n)\}$ is *almost$^\ast$-finite* in the sense used in [@veldman2005 Section 0.8.2].
The following axiom is a form of Brouwer’s famous *Thesis on bars in $\mathcal{N}$*, see [@veldman2006b].
\[A:barinduction\] $\;$
For all $B,C\subseteq \mathbb{N}$, if $\forall \alpha \exists n[\overline \alpha n \in B]$ and $B\subseteq C$ and $\forall s[s\in C\leftrightarrow \forall n[s\ast\langle n\rangle \in C]]$, then $\langle \; \rangle \in C$,
or, equivalently,
for all $B,C\subseteq [\omega]^{<\omega}$, if $\forall \zeta \in [\omega]^\omega \exists n[\overline \zeta n \in B]$ and $B\subseteq C$ and\
$\forall s\in [\omega]^{<\omega}[s\in C\leftrightarrow \forall n[s\ast\langle n\rangle \in [\omega]^{<\omega}\rightarrow s\ast\langle n\rangle\in C]]$, then $\langle \; \rangle \in C$.
\[T:vitalistable\]$\:$
1. $\sim_V^{\neg\neg}$ and $\sim_V^{almost}$ are equivalence relations on $\mathcal{N}$.
2. For all $R$ in $\mathcal{E}$, $\sim_V\;\subseteq R\subseteq \;\sim_V^{\neg\neg}$.
3. For all $R$ in $\mathcal{E}$, $R\;\subseteq\; \sim^{almost}_V $.
4. $\forall\alpha\forall\beta[ \alpha\sim_V^{almost}\beta\rightarrow\exists R\in\mathcal{E}[\alpha R\; \beta]$.
5. $\forall \alpha\forall \beta[\alpha\sim_V^{almost}\beta\rightarrow \alpha \sim_V^{\neg\neg}\beta]$.
\(i) One easily proves that $\sim_V^{\neg\neg}$ is an equivalence relation. One needs the fact that, for all propositions $P,Q$, $(\neg\neg P\;\wedge\;\neg\neg Q)\rightarrow \neg\neg(P\;\wedge\; Q)$.
We prove that $\sim_V^{almost}$ is a transitive relation.
Let $\alpha, \beta, \gamma$ be given such that $\alpha \sim_V^{almost}\beta$ and $\beta\sim_V^{almost} \gamma$.
Let $\zeta$ in $[\omega]^\omega$ be given. Find $\eta$ in $[\omega]^\omega$ such that $\forall n[\alpha\circ\zeta\circ\eta(n)=\beta\circ\zeta\circ\eta(n)]$. Find $p$ such that $\beta\circ\zeta\circ\eta(p)=\gamma\circ\zeta\circ\eta(p)$. Define $n:=\eta(p)$ and note: $\alpha\circ\zeta(n)=\gamma\circ\zeta(n)$.
We thus see: $\forall \zeta\in [\omega]^\omega\exists n[\alpha\circ\zeta(n)=\gamma\circ\zeta(n)]$, that is: $\alpha \sim^{almost}_V\gamma$.
\(ii) The proof is by (transfinite) induction on $\mathcal{E}$. We only prove: for all $R$ in $\mathcal{E}$, $R\subseteq \;\sim_V^{\neg\neg}$ as the statement: for all $R$ in $\mathcal{E}$, $\sim_V\;\subseteq R$ is very easy to prove.
\(I) Our starting point is the trivial observation: $\forall\alpha\forall\beta[\alpha \sim_V\beta\rightarrow \neg\neg (\alpha\sim_V \beta)]$.
\(II) Now let $R$ in $\mathcal{E}$ be given such that $\forall\alpha\forall\beta[\alpha R\beta\rightarrow \neg\neg(\alpha\sim_V \beta)]$.
We have to prove: $\forall\alpha\forall\beta[\alpha R^+\beta\rightarrow \neg\neg(\alpha\sim_V \beta)]$.
We do so as follows.
Let $\alpha, \beta$ be given such that $\alpha R^+\beta$.\
Find $n$ such that $\forall m>n[\alpha(m)\neq \beta(m) \rightarrow \alpha R\beta]$ and consider two special cases.
*Case (1):* $\exists m>n[\alpha(m)\neq \beta(m)$. Then $\alpha R\;\beta$, and, therefore: $\neg\neg(\alpha\sim_V\beta)$.
*Case (2):* $\neg\exists m>n[\alpha(m) \neq \beta(m)$. Then $\forall m>n[\alpha(m)=\beta(m)]$ and $\alpha\sim_V\beta$.
In both cases, we find: $\neg\neg(\alpha\sim_V\beta)$.
Conclude[^25]: $\neg\neg(\alpha\sim_V\beta)$.
\(III) Now let $R_0, R_1, \ldots$ be an infinite sequence of elements of $\mathcal{E}$ such that, for all $n$, $\forall \alpha\forall\beta[\alpha R_n\beta \rightarrow \neg\neg(\alpha\sim_V\beta)]$.
Define $R:=\bigcup_n R_n$ and note: $\forall \alpha\forall\beta[\alpha R\beta \rightarrow \neg\neg(\alpha\sim_V\beta)]$.
\(iii) The proof is by (transfinite) induction on $\mathcal{E}$.
\(I) Our starting point is the observation: $\forall\alpha\forall\beta[a\sim^0_V\beta\rightarrow \alpha\sim^{almost}_V \beta]$.\
We prove this as follows:
Let $\alpha,\beta$ be given such that $\alpha\sim^0_V \beta$. Find $n$ such that $\forall m>n[\alpha(m)=\beta(m)]$. Note: $\forall \zeta \in [\omega]^\omega][\zeta(n+1)>n \;\wedge\;\alpha\circ\zeta(n+1)=\beta\circ\zeta(n+1)]$.\
Conclude: $\alpha\sim_V^{almost}\beta$.
\(II) Now let $R$ in $\mathcal{E}$ be given such that $\forall\alpha\forall\beta[\alpha R\beta\rightarrow \alpha\sim^{almost}_V \beta]$.
We have to prove: $\forall\alpha\forall\beta[aR^+\beta\rightarrow \alpha\sim^{almost}_V \beta]$.
We do so as follows.
Let $\alpha, \beta$ be given such that $\alpha R^+\beta$.\
Find $n$ such that $\forall m>n[\alpha(m)\neq \beta(m) \rightarrow \alpha R\beta]$. Let $\zeta$ in $[\omega]^\omega$ be given. Consider $\zeta(n+1)$ and note $\zeta(n+1)>n$. There now are two cases.
*Either* $\alpha\circ\zeta(n+1)=\beta\circ\zeta(n+1)$ *or* $\alpha\circ\zeta(n+1)\neq\beta\circ\zeta(n+1)$.
In the first case we are done, and in the second case we conclude $\alpha R \beta$, and, using the induction hypothesis, find $p$ such that $\alpha\circ\zeta(p)=\beta\circ\zeta(p)$.
In both cases we conclude: $\exists q[\alpha\circ\zeta(q)=\beta\circ\zeta(q)]$.
We thus see: $\forall \zeta \in [\omega]^\omega\exists q[\alpha\circ\zeta(q)=\beta\circ\zeta(q)]$, that is $\alpha\sim^{almost}_V\beta$.
Clearly then: $\forall \alpha\forall\beta[[\alpha R^+\beta\rightarrow \alpha\sim^{almost}_V \beta]$.
\(III) Now let $R_0, R_1, \ldots$ be an infinite sequence of elements of $\mathcal{E}$ such that, for all $n$, $\forall \alpha\forall\beta[\alpha R_n\beta \rightarrow \alpha\sim_V^{almost}\beta]$.
Define $R:=\bigcup_n R_n$ and note: $\forall \alpha\forall\beta[\alpha R\beta \rightarrow \alpha\sim_V^{almost}\beta]$.
\(iv) Let $\alpha,\beta$ be given such that $\alpha\sim^{almost}\beta$, that is:\
$\forall \zeta \in [\omega]^\omega\exists n[\alpha\circ\zeta(n)=\beta\circ\zeta(n)]$.\
Using Axiom \[A:barinduction\], we shall prove: there exists $R$ in $\mathcal{E}$ such that $\alpha R\beta$.
Define $B:=\bigcup_k\{s\in [\omega]^{k+1}\mid \alpha\circ s(k)=\beta\circ s(k)\}$ and note: $B$ is a bar in $[\omega]^\omega$, that is: $\forall \zeta \in [\omega]^\omega\exists n[\overline \zeta n \in B]$.
Define $C:= \bigcup_k\{ s \in [\omega]^{k}\mid \exists n<k[\alpha\circ s(n)=\beta\circ s(n)] \;\vee\;\exists R \in \mathcal{E}[\alpha R \beta]\}$.
Note: $C= \bigcup_k\{ s \in [\omega]^{k}\mid \forall n<k[\alpha\circ s(n)\neq\beta\circ s(n)] \rightarrow\exists R \in \mathcal{E}[\alpha R \beta]\}$.
Note: $B\subseteq C$ and: $C$ is *monotone*, that is:\
$\forall s\in[\omega]^{<\omega}[s \in C \rightarrow \forall n[s\ast\langle n \rangle \in [\omega]^{<\omega}\rightarrow s\ast\langle n \rangle \in C]]$.
We still have to prove that $C$ is what one calls *inductive* or *hereditary*.
Let $s$ in $[\omega]^{< \omega}$ be given such that $\forall n[s\ast\langle n \rangle \in [\omega]^{<\omega}\rightarrow s\ast\langle n \rangle \in C]$.\
We want to prove: $s\in C$.
Find $k$ such that $s\in [\omega]^k$. In case $\exists n<k[\alpha\circ s(n) =\beta\circ s(n)]$, $s \in C$ and we are done, so we assume: $\forall n<k[\alpha\circ s(n)\neq\beta\circ s(n)]$.
Find a sequence[^26] $R_0, R_1, \ldots$ of elements of $\mathcal{E}$ such that, for each $n$, if $s\ast\langle n \rangle \in [\omega]^\omega$ and $\alpha(n)\neq \beta (n)$, then $\alpha R_n \beta$.
Define $R:=(\bigcup_i R_i)^+$ and note: $R\in \mathcal{E}$.
We claim: $\alpha R\beta$.
We establish this claim as follows.
Define $p$ such that, if $k=0$, then $p:=0$ and, if $k>0$, then $p:=s(k-1)+1$.
Assume: $\exists n\ge p[s\ast \langle \alpha(n)\neq\beta(n)]$ and find $n\ge p$ such that $\alpha(n)\neq \beta(n)$.
Note: $s\ast\langle n \rangle \in [\omega]^{k+1}$ and $\forall i<k+1[\alpha\circ(s\ast\langle n \rangle)(i)\neq \beta\circ(s\ast\langle n \rangle)(i)]$ and $s\ast\langle n \rangle \in C$. Conclude: $\alpha R_n\beta$ and $\alpha( \bigcup_i R_i )\beta$.
We thus see: $\forall n\ge p[\alpha(n)\neq \beta(n)\rightarrow\alpha( \bigcup_i R_i )\beta ]$.
Conclude: $\alpha ( \bigcup_i R_i )^+ \beta$, that is: $\alpha R \beta$, and, therefore: $s\in C$.
We thus see that $C$ is inductive.
Using Axiom \[A:barinduction\], we conclude: $\langle\;\rangle\in C$, that is: $\exists R\in \mathcal{E}[\alpha R \beta]$.
\(v) Let $\alpha,\beta$ be given such that $\alpha\sim^{almost}\beta$, that is:\
$\forall \zeta \in [\omega]^\omega\exists n[\alpha\circ\zeta(n)=\beta\circ\zeta(n)]$.\
Using Axiom \[A:barinduction\], we prove: $\neg\neg\exists p\forall n>p[\alpha(n)=\beta(n)]$.
Define $B:=\bigcup_{k}\{s\in [\omega]^{k+1}\mid \alpha\circ s(k)=\beta\circ s(k)\}$ and note: $B$ is a bar in $[\omega]^\omega$, that is: $\forall \zeta \in [\omega]^\omega\exists n[\overline \zeta n \in B]$. Define\
$C:=\bigcup_k \{ s \in [\omega]^k\mid \exists n<k[\alpha\circ s(n)=\beta\circ s(n)]\;\vee\;\neg\neg\exists p\forall n>p[\alpha(n)=\beta(n)]\}$. Note: $C=\bigcup_k \{ s \in [\omega]^k\mid \forall n<k[\alpha\circ s(n)\neq\beta\circ s(n)]\rightarrow\neg\neg\exists p\forall n>p[\alpha(n)=\beta(n)]\}$.
Note: $B\subseteq C$ and $C$ is monotone, that is:\
$\forall s\in[\omega]^{<\omega}[s \in C \rightarrow \forall n[s\ast\langle n \rangle \in [\omega]^{<\omega}\rightarrow s\ast\langle n \rangle \in C]]$.
We still have to prove that $C$ is inductive.
Let $s$ in $[\omega]^{< \omega}$ be given such that $\forall n[s\ast\langle n \rangle \in [\omega]^{<\omega}\rightarrow s\ast\langle n \rangle \in C]]$.\
We want to prove: $s\in C$.
Find $k$ such that $s\in [\omega]^k$. In case $\exists n<k[\alpha\circ s(n)=\beta\circ s(n)]$, $s\in C$, and we are done, so we assume $\forall n<k[\alpha\circ s(n)\neq\beta\circ s(n)]$.
Define $q$ such that $q:=0$ if $k=0$ and $q:=s(k-1)$ if $k>0$.
Consider two special cases:
*Case (1):* $\exists n>q[\alpha (n) \neq\beta(n)]$.\
Find such $n$, note: $s\ast\langle n \rangle \in [\omega]^\omega$ and $\forall i<k+1[\alpha\circ (s\ast\langle n \rangle)(i)\neq\beta\circ (s\ast\langle n \rangle)(i)]$ and $s\ast\langle n \rangle \in C$, and conclude: $\neg\neg\exists p\forall n>p[\alpha(n)=\beta(n)]$.
*Case (2):* $\neg\exists n>q[\alpha (n) \neq\beta(n)]$, and, therefore, $\forall n>q[\alpha(n)=\beta(n)]$.
In both cases, we find: $\neg\neg\exists p\forall n>p[\alpha(n)=\beta(n)]$.
Conclude[^27]: $\neg\neg\exists p\forall n>p[\alpha(n)=\beta(n)]$, and: $s\in C$.
We thus see that $C$ is inductive.
Using Axiom \[A:barinduction\], we conclude: $\langle\;\rangle \in C$, and, therefore,
$\neg\neg\exists p\forall n>p[\alpha(n)=\beta(n)]$, that is: $\neg\neg(\alpha\sim_V \beta)$.
\[C:unstableextensions\]
1. $(\mathcal{N}, \sim_V^{\neg\neg})\models \mathsf{\forall x\forall y[\neg\neg(x=y)\rightarrow x=y]}$.
2. For each $R$ in $\mathcal{E}$, $(\mathcal{N}, R)\models \mathsf{\forall x\neg\forall y[\neg\neg(x=y)\rightarrow x=y]}$.
\(i) Obvious, as, for any proposition $P$, $\neg\neg\neg\neg P\leftrightarrow \neg\neg P$.
\(ii) Assume $R\in \mathcal{E}$.
We first prove: $(\mathcal{N}, R)\models \mathsf{\neg\forall x\forall y[\neg\neg(x=y)\rightarrow x=y]}$.
Assume $\forall \alpha\forall \beta[\neg\neg(\alpha R \beta)\rightarrow \alpha R \beta]$.
Note: $\forall\alpha\forall \beta[\alpha \sim_V\beta\rightarrow\alpha R\beta]$ and, therefore: $\forall\alpha\forall \beta[\neg\neg(\alpha \sim_V\beta)\rightarrow\neg\neg(\alpha R\beta)]$.
Conclude: $\sim_V^{\neg\neg}\; \subseteq R$.
By Theorem \[T:vitalistable\](ii), $R^+\subseteq \sim_V^{\neg\neg}$, so $R^+\subseteq R$. This contradicts Corollary \[C:hierarchy\].
The stronger statement announced in the Theorem may be proven in a similar way. Inspection of he proof of Theorem \[T:vitalistable\] enables one to conclude:
$(\mathcal{N}, R)\models \mathsf{\neg \forall y[\neg\neg(x=y)\rightarrow x=y]}[\underline 0]$. One easily generalizes this conclusion to:\
for each $\alpha$, $(\mathcal{N}, R)\models \mathsf{\neg \forall y[\neg\neg(x=y)\rightarrow x=y]}[\alpha]$.
Conclude: $(\mathcal{N}, R)\models \mathsf{\forall x\neg\forall y[\neg\neg(x=y)\rightarrow x=y]}$.
Markov’s Principle has been mentioned in Section \[S:spreads\]. Markov’s Principle is not accepted in intuitionistic mathematics, but the following observation still is of interest.
The following are equivalent.
1. Markov’s Principle: $\forall\alpha[\neg\neg\exists n[\alpha(n)=0]\rightarrow \exists n[\alpha(n)=0]]$.
2. $\sim_V^{\neg\neg}\;\subseteq \;\sim_V^{almost}$.
3. $\sim_V^{almost}$ is stable.
\(i) $\Rightarrow$ (ii). Assume $\neg\neg (\alpha\sim_V\beta)$, that is $\neg\neg\exists n\forall m>n[\alpha(m)=\beta(m)]$.
Let $\zeta\in [\omega]^\omega$ be given.
Assume: $\neg \exists n[\alpha\circ\zeta(n)=\beta\circ\zeta(n)]$.
Then $\forall n[\zeta(n+1)>n\;\wedge\;\alpha\circ\zeta(n)\neq \beta\circ\zeta(n)]$, so $\forall n\exists m>n[\alpha(m)\neq\beta(m)]$. Contradiction.
Conclude: $\neg\neg\exists n[\alpha\circ \zeta(n)=\beta\circ\zeta(n)]$ and, by Markov’s Principle,\
$\exists n[\alpha\circ\zeta(n)=\beta\circ\zeta(n)]$.
We thus see $\forall\zeta\in [\omega]^\omega \exists n[\alpha\circ\zeta(n)=\beta\circ\zeta(n)]$, that is: $\alpha\sim_V^{almost} \beta$.
\(ii) $\Rightarrow$ (iii). By Theorem \[T:vitalistable\](v), $\sim_V^{almost}\;\subseteq\; \sim_V^{\neg\neg}$. Therefore: $(\sim_V^{almost})^{\neg\neg}\;\subseteq\; \sim_V^{\neg\neg}$.
Using (ii), we conclude: $(\sim_V^{almost})^{\neg\neg}\;\subseteq \;\sim_V^{almost}$, that is: $\sim_V^{almost}$ is stable.
\(iii) $\Rightarrow$ (i). Let $\alpha$ be given such that $\neg\neg\exists n[\alpha(n)\neq 0]$.
Define $\beta$ such that $\forall m[\beta(m)=0\leftrightarrow \exists n\le m[\alpha(n)=0]]$.
Note: $\neg\neg(\beta\sim_V\;\underline 0)$ and, therefore: $\neg\neg (\beta\sim_V^{almost}\underline 0)$.
Conclude, using (iii), $\beta\sim_V^{almost} \underline 0$.
Define $\zeta$ such that $\forall n[\zeta(n)=n]$.
Find $m$ such that $\beta \circ\zeta(m)=\beta(m)=0$ and, therefore, $\exists n\le m[\alpha(n) =0]$.
We thus see: $\forall\alpha[\neg\neg\exists n[\alpha(n)=0]\rightarrow \exists n[\alpha(n)=0]]$, that is: Markov’s Principle.
Equality and equivalence {#S: equequiv}
========================
We did not succeed in finding a sentence $\psi$ such that $(\mathcal{N}, \sim_V)\models\psi$ and\
$(\mathcal{N}, \sim^\omega_V)\models \neg \psi$. We now want to compare the structures $(\mathcal{N}, =, \sim_V)$ and\
$(\mathcal{N}, =, \sim_V^\omega)$. We need a first order language with two binary relation symbols: $=$ and $\sim$. The symbol $=$ will denote the equality relation and the symbol $\sim$ will denote, in the first structure, the relation $\sim_V$ and, in the second structure, the relation $\sim_V^\omega$. The reader hopefully will not be confused by the fact that, in the earlier sections, where we used the first order language with a single binary relation symbol, $=$, the symbol $=$ denoted the relations $\sim_V$ and $\sim_V^\omega$.
The next Theorem makes us see that equality is decidable on each equivalence class of $\sim_V$ whereas, on each equivalence class of $\sim_V^\omega$, it is not decidable.
\[T:vitaliomegavitali\]$\;$
1. $(\mathcal{N}, =, \sim_V)\models \mathsf{\forall x\forall y[x\sim y\rightarrow (x=y\;\vee\;\neg(x=y))]}$.
2. $(\mathcal{N}, =, \sim_V^\omega)\models \mathsf{\forall x\neg\forall y[x\sim y\rightarrow (x=y\;\vee\;\neg(x=y)]}$.
\(i) Let $\gamma, \alpha$ be given such that $\gamma\sim_V\alpha$.\
Find $n$ such that $\forall m>n[\gamma(m)=\alpha(m)]$ and distinguish two cases.\
*Either* $\overline \gamma (m+1)=\overline \alpha(m+1)$ and $\gamma=\alpha$, *or* $\overline \gamma (m+1)\neq\overline \alpha(m+1)$ and $\neg(\gamma =\alpha)$.
Conclude: $\forall \gamma\forall \alpha[\gamma\sim_V\alpha\rightarrow (\gamma =\alpha \;\vee\;\neg(\gamma=\alpha)]$.
\(ii) Let $\gamma$ be given.
Consider $\mathcal{F}_1^\gamma := \{\alpha\mid \forall m\forall n[(\alpha(m)\neq\gamma(m)\;\wedge\;\alpha(n)\neq\gamma(n)\rightarrow m=n]\}$.
Note: $\mathcal{F}_1^\gamma$ is a spread.
Also: $\forall \alpha \in \mathcal{F}_1^\gamma[\gamma\sim_V^1\alpha]$ [^28] and, therefore, $\forall \alpha \in \mathcal{F}_1^\gamma[\gamma\sim_V^\omega \alpha]$.
Assume $\forall \alpha\in \mathcal{F}_1^\gamma[\gamma=\alpha\;\vee\;\neg(\gamma=\alpha)]$. Applying Lemma \[L:bcpdisj\], find $p$ such that *either* $\forall \alpha\in\mathcal{F}_1^\gamma[ \overline \gamma p\sqsubset \alpha\rightarrow \gamma =\alpha]$ *or* $\forall \alpha[ \overline \gamma p\sqsubset \alpha\rightarrow \neg(\gamma =\alpha)]$, and note that both alternatives are false.
Conclude: $\forall \gamma\neg\forall \alpha[\gamma\sim_V^\omega \alpha\;\vee\neg(\gamma=\alpha)]$.
\[L:almostperhapsive\] $(\sim_V^{\neg\neg})^+\;\subseteq\;\;\sim_V^{\neg\neg}$ and $(\sim_V^{almost})^+\;\subseteq\;\;\sim_V^{almost}$.[^29]
Assume $\alpha (\sim_V^{\neg\neg})^+ \beta$.
Find $n$ such that $\forall m>n[\alpha(m)\neq \beta(m)\rightarrow \alpha\sim_V^{\neg\neg}\beta]$.
Note: if $\exists m>n[\alpha(m)\neq \beta(m)]$, then $\alpha\sim_V^{\neg\neg} \beta$, and if $\neg\exists m>n[\alpha(m)\neq \beta(m)]$, then $\forall m>n[\alpha(m)=\beta(m)]$ and $\alpha\sim_V\beta$ and also $\alpha\sim_V^{\neg\neg}\beta$.
Conclude: $\neg\neg(\alpha\sim_V^{\neg\neg}\beta)$, and, therefore, $\alpha\sim_V^{\neg\neg}\beta$.
Assume $\alpha (\sim_V^{almost})^+ \beta$.
Find $n$ such that $\forall m>n[\alpha(m)\neq \beta(m)\rightarrow \alpha\sim_V^{almost}\beta]$.
Let $\zeta$ in $[\omega]^\omega$ be given. Note: $\zeta(n+1)>n$.
*Either*: $\alpha\circ\zeta(n+1)=\beta\circ\zeta(n+1)$ *or*: $\alpha \sim_V^{almost}\beta$ and $\exists p[\alpha\circ\zeta(p)=\beta\circ\zeta(p)]$.
We thus see: $\forall \zeta\in[\omega]^\omega\exists n[\alpha\circ\zeta(n)=\beta\circ\zeta(n)]$, that is: $\alpha\sim_V^{almost}\beta$.
\[L:lastformula\] For every shift-invariant binary relation $R$ on $\mathcal{N}$,
$R^+\subseteq R$ if and only if $(\mathcal{N},R)\models \mathsf{\forall x \forall y[}\bigl(AP(\mathsf{x,y)\rightarrow x\sim y\bigr)\rightarrow x\sim y]}$.
First assume $R^+\subseteq R$.
Assume $\alpha\;\#\;\beta\rightarrow \alpha R \beta$.
Then: $\forall m>0[\alpha(m)\neq\beta(m)\rightarrow \alpha R\beta]$, so: $\alpha R^+\beta$, and, therefore: $\alpha R\beta$.
We thus see: $(\mathcal{N},R)\models \mathsf{\forall x \forall y[}\bigl(AP(\mathsf{x,y)\rightarrow x\sim y\bigr)\rightarrow x\sim y]}$.
Now assume $(\mathcal{N},R)\models \mathsf{\forall x \forall y[}\bigl(AP(\mathsf{x,y)\rightarrow x\sim y\bigr)\rightarrow x\sim y]}$.
Assume $\alpha R^+\beta$. Find $n$ such that $\forall m>n[\alpha(m)\neq \beta(m)\rightarrow \alpha R \beta]$.
Define $\gamma, \delta$ such that $\forall m[\gamma (m)=\alpha(n+1+m)\;\wedge\;\delta(m) =\beta(n+1+m)]$.
Note: $\gamma \;\#\;\delta \rightarrow \alpha R \beta$, and, as $R$ is shift-invariant, also: $\gamma\;\#\;\delta \rightarrow \gamma R \delta$, and, therefore: $\gamma R \delta$, and also: $\alpha R\beta$.
We thus see: $R^+\subseteq R$.
\[C:lastform\]
1. $(\mathcal{N}, \sim_V^{\neg\neg})\models \mathsf{\forall x \forall y[}\bigl(AP(\mathsf{x,y)\rightarrow x\sim y\bigr)\rightarrow x\sim y]}$.
2. $(\mathcal{N}, \sim_V^{almost})\models \mathsf{\forall x \forall y[}\bigl(AP(\mathsf{x,y)\rightarrow x\sim y\bigr)\rightarrow x\sim y]}$.
3. For each $R$ in $\mathcal{E}$, $(\mathcal{N},R)\models \mathsf{\neg\forall x \forall y[}\bigl(AP(\mathsf{x,y)\rightarrow x\sim y\bigr)\rightarrow x\sim y]}$.
Use Lemmas \[L:almostperhapsive\] and \[L:lastformula\] and Corollary \[C:hierarchy\].
Notations and conventions {#S:notations}
=========================
We use $m,n, \ldots$ as variables over the set $\omega=\mathbb{N}$ of the natural numbers.
For every $P\subseteq \mathbb{N}$ such that $\forall n[P(n)\;\vee\;\neg P(n)]$, for all $m$,
$m=\mu n[P(n)]\leftrightarrow \bigl(P(m)\;\wedge\;\forall n<m[\neg P(n)]\bigr)$.
$(m,n)\mapsto J(m,n)$ is a one-to-one surjective mapping from $\omega\times\omega$ onto $\omega$.
$K,L:\omega\times \omega$ are its inverse functions, so $\forall n[J\bigl(K(n), L(n)\bigr)=n]$.
For each $n$, $n':=K(n)$ and $n'':=L(n)$.
$(n_0, n_1, \ldots, n_{k-1})\mapsto \langle n_0, n_1, \ldots, n_{k-1}\rangle$ is a one-to-one surjective mapping from the set of finite sequences of natural numbers to the set of the natural numbers.
$\langle n_0, n_1, \ldots, n_{k-1}\rangle$ is the *code* of the finite sequence $(n_0, n_1, \ldots, n_{k-1})$.
$s\mapsto length(s)$ is is the function that, for each $s$, gives the length of the finite sequence coded by $s$.
$s,n\mapsto s(n)$ is the function that, for all $s,n$, gives the value of the finite sequence coded by $s$ at $n$. If $n\ge length(s)$, then $s(n)=0$.
For all $s,k$, if $length(s)=k$, then $s=\langle s(0), s(1), \ldots s(k-1)\rangle$.
$0=\langle\;\rangle$ codes the empty sequence of natural numbers,\
the unique finite sequence $s$ such that $length(s)=0$.
$\omega^k:=\{s\mid length(s)=k\}$.
$[\omega]^k:=\{s\in \omega^k\mid\forall i[i+1<k\rightarrow s(i)<s(i+1)]\}$.
$[\omega]^{<\omega}:=\bigcup_k[\omega]^k$.
For all $s,k,t,l$, if $s\in \omega^k $ and $t\in \omega^l$, then $s\ast t$ is the element $u$ of $\omega^{k+l}$ such that $\forall i<k[u(i)=s(i)]$ and $\forall j<l[u(k+j)=t(j)]$.
$s\sqsubseteq t\leftrightarrow \exists u[s\ast u =t]$.
$s\sqsubset t\leftrightarrow (s\sqsubseteq t \;\wedge\;s\neq t)$.
We use $\alpha, \beta, \ldots$ as variables over *Baire space*, the set $\omega^\omega:=\mathcal{N}$ of functions from $\mathbb{N}$ to $\mathbb{N}$.
$(\alpha, n)\mapsto \alpha(n)$ is the function that associates to all $\alpha, n$, the value of $\alpha$ at $n$.
For all $\alpha, \beta$, $\alpha\circ\beta$ is the element $\gamma$ of $\mathcal{N}$ such that $\forall n[\gamma(n)=\alpha\bigl(\beta(n)\bigr)]$.
$2^\omega:=\mathcal{C}:=\{\alpha\mid\forall n[\alpha(n)<2]\}$ is *Cantor space*.
For all $\alpha$, for all $k$, for all $s$ in $\omega^k$, $\alpha\circ s$ is the element $t$ of $\omega^k$ satisfying\
$\forall n<k[t(k)=\alpha\bigl(s(k)\bigr)]$.
For each $s, k$, if $s \in \omega^k$, then, for each $\alpha$, $s\ast\alpha$ is the element $\beta$ of $\mathcal{N}$ such that $\forall i <k[\beta(i)=s(i)]$ and $\forall i[\beta(k+i)=\alpha(i)]$.
For each $s$, for each $\mathcal{X}\subseteq \mathcal{N}$, $s\ast\mathcal{X}:=\{s\ast\alpha\mid\alpha \in \mathcal{X}\}$.
For each $\alpha$, for each $n$, $\alpha^n$ is the element of $\mathcal{N}$ satisfying $\forall m[\alpha^n(m)=\alpha\bigl(J(n,m)\bigr)]$.
For each $m$, $\underline m \in \mathcal{N}$ is the element of $\mathcal{N}$ satisfying $\forall n[\underline m(n)=m]$.
$S$ is the element of $\mathcal{N}$ satisfying $\forall n[S(n)=n+1]$.
$\forall n[\alpha'(n)=\bigl(\alpha(n)\bigr)'\;\wedge\;\alpha''(n)=\bigl(\alpha(n)\bigr)'']$.
$ \overline \alpha n :=\langle \alpha(0),\alpha(1), \ldots \alpha(n-1)\rangle$.
$s\sqsubset\alpha\leftrightarrow \exists n[\overline \alpha n =s]$.
$\alpha\perp\beta \leftrightarrow \alpha\;\#\;\beta\leftrightarrow \exists n[\alpha(n)\neq\beta(n)]$.
$[\omega]^\omega:=\{\zeta\in\mathcal{N}\mid \forall i[\zeta(i)<\zeta(i+1)]\}$.
$\mathbb{Q}$, the set of the rationals, may be defined as a subset of $\omega$, with accompanying relations $=_\mathbb{Q}$, $<_\mathbb{Q}$, $\le_\mathbb{Q}$ and operations $+_\mathbb{Q}, -_\mathbb{Q}, \cdot_\mathbb{Q}$.
$\mathcal{R}:=\{\alpha\mid \forall n[\alpha' (n)\in \mathbb{Q}\;\wedge\; \alpha''(n)\in \mathbb{Q}]\;\wedge\\\forall n[\alpha'(n)\le_\mathbb{Q}\alpha' (n+1)\le_\mathbb{Q} \alpha''(n+1)\le_\mathbb{Q}\alpha''(n)]\;\wedge\;\forall m\exists n[\alpha''(n)-_\mathbb{Q} \alpha'(n)<_\mathbb{Q} \frac{1}{2^m}]\}$.
For all $\alpha, \beta$ in $\mathcal{R}$,
$\alpha<_\mathcal{R}\beta \leftrightarrow \exists n[\alpha''(n)<_\mathbb{Q}\beta'(n)]$ and $\alpha=_\mathcal{R}\beta \leftrightarrow \bigl(\neg (\alpha<_\mathcal{R} \beta)\;\wedge\;\neg(\beta <_\mathcal{R}\alpha)\bigr)$.
Operations $+_\mathcal{R}, -_\mathcal{R}$ are defined straightforwardly.
[10]{} L. E. J. Brouwer, Begründung der Mengenlehre unabhängig vom logischen Satz vom ausgeschlossenem Dritten. Zweiter Teil: Theorie der Punktmengen, *Koninklijke Nederlandse Akademie van Wetenschappen Verhandelingen, 1e Sectie 12* (7), 1919; 33 pp.; also in: [@brouwer2], pp. 191-221.
L. E. J. Brouwer, *Collected Works 1, Philosophy and Foundations of Mathematics*, (edited by A. Heyting), North-Holland Publishing Company, Amsterdam 1975. D. van Dantzig, On the principles of intuitonistic and affirmative mathematics, *Proceedings Koninklijke Nederlandse Akademie van Wetenschappen* 50(1947)918-929 and 1092-1103 = *Indagationes Mathematicae* 9(1947)429-440 and 506-517. W. Hodges, *Model Theory*, Encyclopedia of Mathematics and its Applications, v. 42, Cambridge University Press, 1993. A.S. Troelstra, D. van Dalen, *Constructivism in Mathematics, an Introduction, Vol. I*, Studies in Logic and the Foundations of Mathematics, vol. 121, North-Holland Publ. Co., Amsterdam etc., 1988. W. Veldman, M. Janssen, Some Observations on Intuitionistically Elementary Properties of Linear Orderings, *Archive for Mathematical Logic*, 29(1990)171-187. W. Veldman, F. Waaldijk, Some Elementary Results in Intuitionistic Model Theory, *The Journal of Symbolic Logic*, 61(1996)745-767.
W. Veldman, Some intuitionistic variations on the notion of a finite set of natural numbers, in: H.C.M. de Swart, L.J.M. Bergmans (ed.), [*Perspectives on Negation, essays in honour of Johan J. de Iongh on the occasion of his 80th birthday*]{}, Tilburg University Press, Tilburg, 1995, pp. 177-202.
W. Veldman, On sets enclosed between a set and its double complement, in: A. Cantini e.a.(ed.), [*Logic and Foundations of Mathematics*]{}, Proceedings Xth International Congress on Logic, Methodology and Philosophy of Science, Florence 1995, Volume III, Kluwer Academic Publishers, Dordrecht, 1999, pp. 143-154. W. Veldman, Understanding and using Brouwer’s Continuity Principle, in: U. Berger, H. Osswald, P. Schuster (ed.), *Reuniting the Antipodes, constructive and nonstandard views of the continuum, Proceedings of a Symposium held in San Servolo/Venice, 1999*, Kluwer, Dordrecht, 2001, pp. 285-302. W. Veldman, Two simple sets that are not positively Borel, [*Annals of Pure and Applied Logic*]{} 135(2005)151-2009. W. Veldman, Brouwer’s Real Thesis on Bars, in: G. Heinzmann, G. Ronzitti, eds., [*Constructivism: Mathematics, Logic, Philosophy and Linguistics, Philosophia Scientiae, Cahier Spécial 6*]{}, 2006, pp. 21-39. W. Veldman, Projective sets, intuitionistically, submitted to: *The Review of Symbolic Logic*, arXiv:1104.3077.
[^1]: Classically, all infinite models of the first-order theory of equality are elementarily equivalent.
[^2]: This observation has been made earlier in [@veldman2001 Section 5]. The first part of the present paper elaborates part of [@veldman2001 Section 5].
[^3]: Every *spread* is a closed subset of $\mathcal{N}$, see Section \[S:spreads\].
[^4]: See Lemma \[L:defisol\]. $\alpha\in \mathcal{X}\subseteq\mathcal{N}$ is a *decidable* point of $\mathcal{X}$ if and only if $\forall \beta \in \mathcal{X}[\alpha=\beta\;\vee\;\neg(\alpha=\beta)]$.
[^5]: see Definition \[D:transparent\].
[^6]: The *coherence* of a closed set is the set of its limit points, see Definition \[D:limitpoint\].
[^7]: $R\subseteq \mathcal{N}\times\mathcal{N}$ is called *stable* if $\forall \alpha\forall\beta[\neg\neg\alpha R\beta \rightarrow \alpha R\beta]$, see Definition \[D:stable\].
[^8]: If $\psi \in \Gamma$ and $\neg\psi \in \Delta$, then $\neg \psi \in \Delta$ and $\neg\neg\psi \in \Gamma$: the relation *positively different* is symmetric.
[^9]: $\alpha\;\#\;\beta\leftrightarrow\alpha\perp\beta\leftrightarrow \exists n[\alpha(n)\neq\beta(n)]$, see Section \[S:notations\].
[^10]: A statement is *reckless* if one might think it is true while the intuitionistic mathematician understands there is no proof for his constructive reading of it.
[^11]: see [@veldman2001].
[^12]: *Inconsistent* equality types may be annoying but do not cause difficulties.
[^13]: $\mathcal{X}\subseteq\mathcal{N}$ is *inhabited* if and only if $\exists \alpha[\alpha\in \mathcal{X}]$.
[^14]: A.A. Markov enuntiated this principle for *primitive recursive* $\alpha$ only.
[^15]: Cantor’s Main Theorem nowadays is called the Perfect Set Theorem: *every closed subset of $\mathcal{N}$ is the union of a perfect set and an at most countable set.*
[^16]: See Definition \[D:transparent\].
[^17]: See Definition \[D:functionspreads\].
[^18]: For each $\mathcal{X}\subseteq \mathcal{N}$, $\overline{\mathcal{X}}:=\{\alpha\mid\forall n\exists \beta \in \mathcal{X}[\overline \alpha n\sqsubset \beta]\}$ is the *closure* of $\mathcal{X}$.\
$\bigcup_n \underline{\overline 0} n\ast\langle 1\rangle\ast \mathcal{T}_{\alpha(n)}$, in general, is not a spread, but its closure is.
[^19]: Classically, $Th\bigl((\mathbb{Q}, <)\bigr)$ is the one and only complete extension of $DLO$.
[^20]: For each $\mathcal{X}\subseteq \mathcal{R}$, $\overline{\mathcal{X}}:=\{x\in \mathcal{R}\mid \forall n\exists y\in \mathcal{X}[|x-y|< \frac{1}{2^n}]\}$ is the *closure* of $\mathcal{X}$.
[^21]: The term ‘*stable*’ has been introduced by D. Van Dantzig, who hoped to be able to reconstruct ‘classical’, non-intuitionistic mathematics within the stable part of intuitionistic mathematics, see [@dantzig].
[^22]: See [@troelstra p. 256]
[^23]: In [@veldman1995], $\mathcal{X}\subseteq \mathcal{N}$ is called a *notion of finiteness* if $\mathbf{Fin}\subseteq \mathcal{X}\subseteq \mathbf{Fin}^{\neg\neg}$. For every $R$ in $\mathcal{E}$, $Fin_R$ is a notion a finiteness.
[^24]: For each $n$, $n=(n',n'')$, see Section \[S:notations\].
[^25]: using the scheme: if $P\rightarrow Q$ and $\neg P\rightarrow Q$, then $\neg\neg Q$.
[^26]: This application of countable choice may be reduced to Axiom \[ax:countable choice\]. One may define $\mathcal{B}\subseteq \mathcal{N}$ and a *coding* mapping $\alpha\mapsto R_\alpha$ such that $\mathcal{E}=\{R_\alpha\mid\alpha\in \mathcal{B}\}$.
[^27]: Using the scheme: If $P\rightarrow Q$ and $\neg P \rightarrow Q$, then $\neg\neg Q$.
[^28]: See the proof of Theorem \[T:vitalivariations\](iii)
[^29]: Following the terminology in [@veldman1995], a binary relation $R$ on $\mathcal{N}$ should be called *perhapsive* if $R^+\subseteq R$.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Radical recombination has been proposed to lead to the formation of complex organic molecules (COMs) in CO-rich ices in the early stages of star formation. These COMs can then undergo hydrogen addition and abstraction reactions leading to a higher or lower degree of saturation. Here, we have studied [14]{} hydrogen transfer reactions for the molecules glyoxal, glycoaldehyde, ethylene glycol, and methylformate [and an additional three reactions where fragments are involved.]{} Over-the-barrier reactions are possible only if tunneling is invoked in the description at low temperature. Therefore the rate constants for the studied reactions are calculated using instanton theory that takes quantum effects into account inherently. [[The reactions were characterized in the gas phase, but this is expected to yield meaningful results for CO-rich ices due to the minimal alteration of reaction landscapes by the CO molecules.]{}]{}
We found that rate constants should not be extrapolated based on the height of the barrier alone, since the shape of the barrier plays an increasingly larger role at decreasing temperature. It is neither possible to predict rate constants based only on considering the type of reaction, the specific reactants and functional groups play a crucial role. Within a single molecule, though, hydrogen abstraction from an aldehyde group seems to be always faster than hydrogen addition to the same carbon atom. Reactions that involve heavy-atom tunneling, *e.g.*, breaking or forming a C–C or C–O bond, have rate constants that are much lower than those where H transfer is involved.
author:
- |
S. Álvarez-Barcia,$^{1}$ P. Russ,$^{1}$ J. Kästner,$^{1}$ and T. Lamberts$^{1,2}$[^1]\
$^{1}$[Institute for Theoretical Chemistry, University of Stuttgart, Pfaffenwaldring 55, 70569 Stuttgart, Germany]{}\
$^{2}$[Current Address: Leiden Institute of Chemistry, Gorlaeus Laboratories, Leiden University, P.O. Box 9502, 2300 RA Leiden, The Netherlands]{}
bibliography:
- 'text.bib'
- 'rsc.bib'
date: 'Accepted XXX. Received YYY; in original form ZZZ'
title: Hydrogen transfer reactions of interstellar Complex Organic Molecules
---
\[firstpage\]
astrochemistry – methods: laboratory: solid state – ISM: molecules
Introduction
============
Thanks to the unprecedented sensitivity of ALMA the detection and quantification of interstellar complex organic molecules (COM) has become more and more within reach. A complex organic molecule in the context of astrochemistry is loosely defined as a molecule consisting of more than 6 H, C, O, and/or, N atoms. Typical gas-phase abundances of such molecules are only of the order of $< 10^{-8}$ with respect to [@Jorgensen:2012; @Halfen:2015; @Taquet:2015; @Lopez-Sepulcre:2017] with even lower abundances for deuterated species [@Belloche:2016]. These molecules are currently thought to find their origins in the CO-rich top layers of the grain ice mantle [@Boogert:2015] where the reaction network has been shown to lead to the formation of the parent species formaldehyde, , and methanol, . Furthermore, besides hydrogen addition reactions, also hydrogen abstraction reactions can take place that decrease the number of H atoms on the CO backbone [[@Nagaoka:2005; @Nagaoka:2007]]{}. Although it has been suggested that formaldehyde and methanol may desorb from the grain surface and subsequently react in the gas phase to yield more complex species [@Bottinelli:2004; @Balucani:2015; @Taquet:2017], a variety of COMs have been detected in cold interstellar regions [@Bacmann:2012; @Oberg:2010; @Vastel:2014]. This indicates that low-temperature surface chemistry can play an important role in the formation of larger species.
In fact the reaction network has evolved into a network where carbon-carbon bonds can be formed via radical-radical reactions between the ‘fundamental’ radicals that are created as intermediates, *i.e.*, HCO, , and . Most of these reactions have been studied experimentally in various ways [@fed15; @but15; @chu16; @fed17; @chu17; @but17] and they have also been proposed by and are included in a number of astrochemical model studies [@Garrod:2008; @Woods:2012; @Coutens:2018]. Similar conclusions are also supported by observational work for specific species [@Rivilla:2017; @Li:2017]. Despite this significant amount of investigations, relatively little is known about the reaction rate constants at low temperature, while these are the crucial parameters needed to constrain modeling studies.
Here we focus on hydrogen addition and abstraction reactions of species with two carbon atoms and two oxygen atoms, *i.e.*, methylformate, glyoxal, glycoaldehyde, and ethylene glycol (Section \[sec:network-new\]). [Several other reactions are discussed as well where a C–C or C–O bond is formed via an over-the-barrier reaction between an radical and formaldehyde (Section \[sec:FA-reactions\]).]{} [Finally, we]{} provide an overview of reaction rate constants previously calculated for the network [involving both formaldehyde and methanol]{} [@and11; @Goumans2011a; @gou11; @Song2017] (Section \[sec:prev\]). Low-temperature reaction rate constants have been calculated for the first time using instanton theory and serve as an order of magniture estimate implementation in astrochemical models. We will also comment on the possibility to generalize rate constants based only on the type of reaction.
Computational Details {#sec:computation}
=====================
Two different levels of theory have been used throughout this study in order to balance the computational cost and chemical accuracy. All calculated activation and reaction energies, as well as the rate constants have been calculated with density functional theory (DFT). In particular, the functional MPWB1K combined with the basis set def2-TZVP has been used. The accuracy of the activation energies or barrier heights is ensured by benchmarking these values to a better level of theory, namely CCSD(T)-F12/VTZ-F12.
Optimizations of the stationary points, corresponding energies, and spin densities were computed at the MPWB1K/def2-TZVP level [@MPWB1K; @Weigend2005; @Weigend2006]. Geometry optimizations (minima and transition states) were done with DL-FIND [@dlfind] in ChemShell [@she03; @met14]. For the electronic structure computations (energies, gradients, and Hessians) Gaussian 09 [@gaussian09] has been employed. SCF cycles were stopped when the convergence, as defined in G09, reached $\num{1E-9}$ Hartree. A pruned (99 590) grid (ultrafine grid) was employed, having 99 radial shells and 590 angular points per shell.
The MPWB1K functional has been previously benchmarked in order to predict the correct bond dissociation energy of methyl formate (MF), for which accurate results were obtained [@Truhlar_mf_2016]. Furthermore, MPWB1K was developed to take into account weak interactions such as those found in the pre-reactive complexes treated here. In order to confirm the use of this functional for the current study, single point energy calculations at the RHF-UCCSD(T)-F12/VTZ-F12//MPWB1K/def2-TZVP level [@Knowles:1993; @Knowles:err; @Deegan:1994; @Adler:2007; @Knizia:2009; @Peterson:2008] were carried out and are discussed in Appendix \[benchmark\].
The instanton method based on Feynman path integral theory using the semiclassical approximation was used to compute the reaction rate constants [@lan67; @lan69; @mil75; @col77; @cal77; @gil77; @aff81; @col88; @han90; @ben94; @mes95; @ric09; @kry11; @alt11; @rom11; @rom11b; @kry14; @ric16]. For a given temperature, it provides the most probable tunnelling path, the instanton, which connects the reactant and product valleys of the potential energy surface. Instanton theory is applicable whenever the temperature is low enough for the instanton to spread out. At higher temperatures, the instanton collapses to a point which renders the theory inapplicable. For most barriershapes this collapse happens at the crossover temperature $T_\text{c}$ [@gil87; @alv14], $$T_\mathrm{c}=\frac{\hbar\ \Omega}{2\pi k_\text{B}}$$ with $\Omega$ being the the absolute value of the imaginary frequency corresponding to the transition mode and $k_\text{B}$ corresponding to Boltzmann’s constant. $T_\text{c}$ qualitatively indicates at which temperature the reaction is dominated by tunneling ($T$ $<$ $T_\text{c}$ ) or by the thermal activation ($T$ $>$ $T_\text{c}$).
Instanton paths were optimized via a quasi Newton–Raphson method [@rom11; @rom11b]. Energies, gradients, and Hessians were provided by Gaussian 09, but instanton optimizations are done in DL-FIND. The instanton path was discretised using 80 images, except for reactions MF3 and MF4 where 158 images were employed at T $\leq$ 100 K and 314 images for MF4 at 75 K.
This study focuses on unimolecular rate constants, *i.e.*, on the Langmuir–Hinshelwood mechanism. Both reactants are adsorbed on the surface, approach each other via diffusion and form a pre-reactive complex (PRC) on the surface. This PRC can then decay to yield the reaction products via a unimolecular process. It has been shown in the recent literature that often gas-phase calculations of stationary points offer a reasonably accurate approach for representing the very same reactions on an ice surface. This even holds for ices composed of water molecules as typical changes of the activation energy are roughly only 1–2 kJ/mol [@Rimola:2014; @Song2017; @Lamberts:2018]. However, in particular cases, larger energy differences may be found [@Lamberts:2017] and to which extent surface molecules may affect the binding orientation is currently unclear. Finally, adsorption on a surface is simulated by keeping the rotational partition function constant between the reactant and transition state. For more information regarding this approach the reader is referred to @Meisner:2017 and @Lamberts:2017B.
{width="71.00000%"}
Results
=======
[We simulated a total of 14 reactions revolving around the molecules glyoxal (GX), glycoaldehyde (GA), ethylene glycol (EG), and methyl formate (MF), an additional three reactions where reactions of fragments with are involved (FAR), and discuss the results in the light of the 6 previously studied reactions with carbon monoxide (CO), formaldehyde (FA), and methanol (ME).]{} To structure the analysis, the reactions are labeled according to their type, [except for the FARn series]{}:
1) [ H addition to aldehyde carbon]{} [– MF, GX, GA, CO, FA]{}
2) [ H addition to aldehyde oxygen]{} [– MF, GX, GA, FA]{}
3) [ H abstraction from aldehyde carbon]{} [– MF, GX, GA, FA]{}
4) [ H abstraction from methyl group]{} [– MF, EG, ME]{}
5) [ H addition to [etheric]{} oxygen]{} [– MF]{}
6) [ H abstraction from alcohol oxygen]{} [– GA, EG, ME]{}
[Note that we expect all reactions studied and discussed here to take place in an environment where carbon monoxide, CO, is the main component of the ice mantle. Due to the general weak interactions of this molecule, we expect that the activation energies calculated here in the gas phase will be similar to those in the presence of a CO environment. For instance for the reactions and this has been confirmed by @Rimola:2014. Therefore, the values presented here are thought to be a good representation of the situation in the interstellar medium.]{}
[Reactions with MF, GX, GA, and EG]{} {#sec:network-new}
--------------------------------------
Activation energies for the reactions described in this Section and in Section \[sec:FA-reactions\] can be found in Table \[tbl:energies\]. Crossover temperatures ($T_\text{c}$) as an indication of the importance of tunneling, and the calculated low-temperature rate constants are presented as well. A schematic representation of the reactions with methylformate, glyoxal, glycoaldehyde, and ethylene glycol is given in Figure \[fig:structures\]. [The temperature-dependence of the calculated rate constants is depicted in Figs. \[fig:mf\]–\[fig:IRC\].]{} [The reaction of the hydrogen atom with methylformate has been studied in order to determine if it is an efficient destruction channel. ]{} [The addition and abstraction reactions of H with glyoxal, glycoaldehyde, and ethylene glycol serve to study the sequential hydrogenation steps. In this way the same reaction type (see above) can be compared between various molecules, and it can be determined whether or not addition is faster than abstraction. Note that]{} the reaction abstracting a hydrogen atom from glyoxal (\[GX3\]) could not be studied, because according to the benchmark study the most accurate value for the activation energy cannot be validated: the difference between the DFT and CCSD(T)-F12 value is too large and moreover multireference effects prevent the CCSD(T)-F12 value from being trusted.
### Reactions with methylformate (MF) {#reactions-with-methylformate-mf .unnumbered}
$$\begin{aligned}
&\ce{H + MF -> CH3OCOH2} \tag{MF1}\label{MF1} \\
&\ce{H + MF -> CH3OCHOH} \tag{MF2}\label{MF2} \\
&\ce{H + MF -> CH3OCO + H2} \tag{MF3}\label{MF3} \\
&\ce{H + MF -> CH2OCOH + H2} \tag{MF4}\label{MF4}\\
&\ce{H + MF -> HCO + CH3OH} \tag{MF5}\label{MF5} \end{aligned}$$
### Reactions with glyoxal (GX) {#reactions-with-glyoxal-gx .unnumbered}
$$\begin{aligned}
&\ce{H + GX -> (H2CO)CHO } \tag{GX1}\label{GX1} \\
&\ce{H + GX -> (HCOH)CHO } \tag{GX2}\label{GX2} \\
&\ce{H + GX -> (CO)CHO + H2} \tag{GX3}\label{GX3} \end{aligned}$$
### Reactions with glycoaldehyde (GA) {#reactions-with-glycoaldehyde-ga .unnumbered}
$$\begin{aligned}
&\ce{H + GA -> (H2CO)CH2OH } \tag{GA1}\label{GA1} \\
&\ce{H + GA -> (HCOH)CH2OH } \tag{GA2}\label{GA2} \\
&\ce{H + GA -> (CO)CH2OH + H2 } \tag{GA3}\label{GA3} \\
&\ce{H + GA -> (HCO)CHOH + H2 } \tag{GA4}\label{GA4} \\
&\ce{H + GA -> (HCO)CH2O + H2 } \tag{GA6}\label{GA6} \end{aligned}$$
### Reactions with ethylene glycol (EG) {#reactions-with-ethylene-glycol-eg .unnumbered}
$$\begin{aligned}
&\ce{H + EG -> (HOCH)CH2OH + H2 } \tag{EG4}\label{EG4} \\
&\ce{H + EG -> (OCH2)CH2OH + H2 } \tag{EG6}\label{EG6} \end{aligned}$$
[lrrrl]{} & $\Delta E^{\ddagger}$ & $\Delta E^{0,\ddagger}$ & $T_\text{c}$ & $k$ (75 K)\
& [(kJ mol$^{-1}$)]{} & [(kJ mol$^{-1}$)]{} & (K) & s$^{-1}$\
& & & &\
\[MF1\] & 38.1 & 41.2 & 262.2 &\
\[MF2\] & 59.0 & 59.9 & 400.4 &\
\[MF3\] & 46.6 & 38.1 & 365.8 & $^{a}$\
\[MF4\] & 51.2 & 42.8 & 345.1 &\
\[MF5\] & 146.0 & 149.9 & 496.7 &\
& & & &\
[H + GX]{} & & & &\
\[GX1\] & 15.1 & 15.1 & 179.9 &\
\[GX2\] & 29.8 & 31.7 & 298.9 &\
& & & &\
& & & &\
[H + GA]{} & & & &\
\[GA1\] & 19.0 & 20.8 & 203.4 &\
\[GA2\] & 38.5 & 39.8 & 342.5 &\
\[GA3\] & 24.3 & 14.6 & 317.6 &\
\[GA4\] & 27.3 & 20.6 & 333.5 &\
\[GA6\] & 55.8 & 46.2 & 405.2 &\
& & & &\
[H + EG]{} & & & &\
\[EG4\] & 28.4 & 19.3 & 303.3 &\
\[EG6\] & 54.1 & 42.2 & 406.3 & $^{b}$\
& & & &\
[FARn]{} & & & &\
\[FAR1\] & 30.6 & 22.2 & 336.6 & $^{c}$\
\[FAR2\] & 44.8 & 48.5 & 151.9 & $^{d}$\
\[FAR3\] & 19.9 & 24.5 & 58.3 & $^{c}$\
\
\
\
Reactions between FA and {#sec:FA-reactions}
-------------------------
[Although the COMs discussed above have been proposed to be formed mainly through radical-radical reactions, reactions between a neutral and radical species may also lead to the formation of a C–C or C–O bond. The reactions between and or [@but17] are therefore studied as well in order to compare their efficiency to other radical-neutral reactions as well as to fast barrierless radical-radical reactions.]{}
Reactions \[FAR1\] and \[FAR2\] [are in direct competition with each other]{}, see also Fig. \[fig:fa\]. $$\begin{aligned}
&\ce{H2CO + CH3O -> CH3OH + HCO} \tag{FAR1}\label{FAR1} \\
&\ce{H2CO + CH3O -> CH3OCH2O} \tag{FAR2}\label{FAR2} \\
&\ce{H2CO + HCO -> (HCO)CH2O} \tag{FAR3}\label{FAR3} \end{aligned}$$
Reactions with CO, FA, and ME {#sec:prev}
-----------------------------
[Prior to discussing hydrogen transfer reactions in COMs]{}, this Section summarizes previous theoretical studies related to the reaction network for cases where calcualtions have also been performed with instanton theory. The main results, in terms of activation energy and reaction rate constant from those studies are listed in Table \[tbl:lit\].
[lrrl]{} & $\Delta E^{0,\ddagger}$ & $k$\
& (kJ/mol) & s$^{-1}$\
& & &\
\[CO1\] & 12.4 + $\sim$1.2$^{a}$& at 5 K & \[1\]\
& & &\
[H + FA]{} & & &\
\[FA1\] & 15.8 - 17.9 & - at 70 K & \[2\]\
\[FA2\] & 43.3 - 47.1 & - at 75 K & \[2\]\
\[FA3\] & 20.5 - 25.2 & - at 70 K & \[2\]\
& & &\
[H + ME]{} & & &\
\[ME4\] & 30.2 & – & \[3\]\
\[ME6\] & 46.4 & – & \[3\]\
\
\
![Unimolecular rate constants (in $s^{-1}$) calculated with instanton theory for the [methylformate (MF)]{} + H reactions.[]{data-label="fig:mf"}](MF2_new.pdf){width="44.00000%"}
![Unimolecular rate constants (in $s^{-1}$) calculated with instanton theory for the [glyoxal (GX)]{} + H reactions.[]{data-label="fig:gx"}](GX1.pdf){width="44.00000%"}
![Unimolecular rate constants (in $s^{-1}$) calculated with instanton theory for the [glycoaldehyde (GA)]{} + H reactions.[]{data-label="fig:ga"}](GA1.pdf){width="44.00000%"}
![Unimolecular rate constants (in $s^{-1}$) calculated with instanton theory for the [ethylene glycol (EG)]{} + H reactions.[]{data-label="fig:eg"}](EG1.pdf){width="44.00000%"}
![Intrinsic reaction coordinates for reactions \[MF1\] - \[MF4\].[]{data-label="fig:IRC"}](IRC1.pdf){width="44.00000%"}
![Unimolecular rate constants (in $s^{-1}$) calculated with instanton theory for the [reactions of formaldehyde with fragments (FARn)]{}.[]{data-label="fig:fa"}](FA21.pdf){width="44.00000%"}
### Reaction with CO {#reaction-with-co .unnumbered}
Unimolecular rate constants for the system have been theoretically calculated by @Andersson2011, using a PES previously obtainded by @Keller1996: $$\begin{aligned}
&\ce{H + CO -> HCO} \tag{CO1}\label{CO1}\end{aligned}$$
### Reactions with FA {#reactions-with-fa .unnumbered}
The reaction of H and () has been theoretically studied by both @gou11b and @Song2017. $$\begin{aligned}
&\ce{H + FA -> CH3O} \tag{FA1}\label{FA1} \\
&\ce{H + FA -> CH2OH} \tag{FA2}\label{FA2} \\
&\ce{H + FA -> H2 + HCO} \tag{FA3}\label{FA3} \end{aligned}$$
### Reactions with ME {#reactions-with-me .unnumbered}
The abstraction of H on the methanol (ME) mehtyl group has been studied theoretically by @Goumans2011a. $$\begin{aligned}
&\ce{H + ME -> CH2OH + H2} \tag{ME4}\label{ME4} \\
&\ce{H + ME -> CH3O + H2} \tag{ME6}\label{ME6} \end{aligned}$$
Discussion
==========
[Hydrogen addition and abstraction reactions]{}
------------------------------------------------
[For most of the reactions, the rate constants level off at a given temperature, especially below 80 K.]{}
The reactions of the hydrogen atom with methylformate overall have the highest activation energies, indicating that MF is quite stable with respect to attack by a H radical. Reaction \[MF5\] is a special case, where a C-O bond is being broken while an O-H bond is formed, which is most likely the reason for the corresponding activation energy, or high barrier.
Comparing between the various reaction types, Figures \[fig:mf\] and \[fig:ga\] show that the $\ce{H}$ abstractions from the H of the HC=O group (type 3) occur with high reaction rate constants. The $\ce{H}$ abstraction from the -OH group (type 6), on the other hand, appears to be very unfavourable with barriers larger than $\sim$50 kJ/mol (Figures \[fig:ga\] and \[fig:eg\]). This is consistent with previous results obtained for the reaction between hydrogen and methanol. The barrier [between reactions \[ME4\] and \[ME6\]]{} differs by 16 kJ/mol, in line with the experimental work of @chu16 and @Nagaoka:2007.
Generalizing reaction types 1, 2, and 4 is not trivial. The activation energies for type 1 are always lower than those for type 2, when compared within the same molecule (FA, MF, GX, and GA). For instance the formation of is preferred over the formation of , contrary to the findings of @but15, but in line with those of @chu16. The rate constants for type 1 are indeed higher than those for type 2 for reactions with FA, GX, and GA, but MF is a special case. Similarly for type 4, where the activation energies are higher than for type 2 and consequently the rate constants are lower for GA, but again reaction \[MF2\] deviates. At temperatures below 200 K the reaction rate constant for \[MF2\] crosses first that of \[MF4\] and later that of \[MF1\] even though the barrier is higher. The origin of this behavior lies in the barrier width. Tunneling namely depends both on the barrier height and width as well as on the effective mass of the system. The narrower the barrier at low-energy incidence, the more tunneling may be expected. This can be visualized with the help of intrinsic reaction coordinates (IRCs). The IRC curves are presented in Fig. \[fig:IRC\], note that these do not include ZPE corrections and therefore relate to the barrier height $\Delta E^{\ddagger}$ of Table \[tbl:energies\].
[Finally, it cannot be said that in general addition is more efficient than abstraction or vice versa, *e.g.*, compare reaction types 1 and 2 against 3, 4, and 6.]{}
[Reactions between FA and ]{} {#reactions-between-fa-and}
------------------------------
[At decreasing temperatures tunneling dominates a reaction more and more. This can explain the large difference between the low temperature value for the rate constant of reaction \[FAR2\] compared to \[FAR1\]. For \[FAR2\] a C–O bond is formed and as heavy-atom tunneling is less efficient than hydrogen atom tunneling, the rate constant is much lower than what would be expected from the barrier height only (compare for instance \[MF4\], \[GA6\], and \[FAR2\]). Similarly for the reaction of FA with , where a C–C bond is formed (\[FAR3\]), again the low-temperature rate constant is very low. Note also the lower values for the crossover temperatures of \[FAR2\] and \[FAR3\] compared to the hydrogen transfer reactions, indicating that tunneling also sets in at lower temperatures. Comparing the rate constant to the typical value assumed for radical-radical barrierless reactions, $\sim10^{12}$ s$^{-1}$, it is clear that these reactions are much less likely to contribute to COM formation. On that note we do wish to stress, however, to keep in mind that although radical-radical reactions may be able to proceed without a barrier, this does not mean that all reaction pathways are open, see for instance @Lamberts:2018]{}.
Astrochemically relevant conclusions
====================================
Unimolecular reaction rate constants have been calculated and are provided for hydrogen addition and abstraction reactions from methylformate, glyoxal, glycoaldehyde, and ethylene glycol and are thus available to be implemented in both rate-equation and kinetic Monte Carlo models aimed at studying the formation of COMs at low temperatures.
Our results are generally in agreement with experimental work, although some discrepancies exist on the efficiency of specific reaction paths, such as the formation of or after hydrogen abstraction from methanol, which impacts on the ease of methylformate formation (for which is needed) or ethylene glycol formation (for which is required). A microscopic model aiming to reproduce experiments may be able to provide a clear picture of how the reactions are intertwined with each other.
The reaction could not be studied and thus deserves further attention.
We found that one cannot predict average rate constants solely based on the type of the reaction. The spread in the low-temperature rate constant can be roughly 7 orders of magnitude for a single reaction type (*e.g.*, hydrogen addition to an aldehyde carbon) showing a strong dependence on the other functional groups that are attached to the carbon backbone.
Within a single molecule, on the other hand, one can loosely say that hydrogen abstraction from an aldehyde group is faster than hydrogen addition to the same carbon. Both of these have a rate constant that is larger than hydrogen abstraction from a methyl group.
Care should be taken with extrapolating rate constants based on the height of the barrier alone, as calculations show that reactions with narrow barriers can have rate constants at low temperature that are higher than those with a lower activation energy.
Reactions that include the breakage or formation of a bond between two heavy atoms generally have low-temperature rate constants that are much lower than those for hydrogen addition or abstraction reactions as a result of the low efficiency of tunneling when heavy atoms are involved.
Acknowledgements {#acknowledgements .unnumbered}
================
The authors acknowledges support for computer time by the state of Baden-Württemberg through bwHPC and the Germany Research Foundation (DFG) through grant no. INST 40/467-1FUGG and SFB 716/C.6. This project was financially supported by the European Union’s Horizon 2020 research and innovation programme (grant agreement No. 646717, TUNNELCHEM), the Alexander von Humboldt Foundation, the Netherlands Organisation for Scientific Research (NWO) via a VENI fellowship (722.017.008) and the COST Action CM1401 via an STSM travel grant.
Benchmark calculations {#benchmark}
======================
RHF-UCCSD(T)-F12/VTZ-F12//MPWB1K/def2-TZVP single-point energy calculations were performed in order to check if MPWB1K provides a suitable description of the energy landscape for the reactions studied here. In general, the CCSD(T)-F12 method can be seen as the gold standard for obtaining relative energies for systems that are well-described by a single reference wavefunction. This is typically assumed to be the case when the so-called T1 and D1 diagnostics are smaller than the commonly used threshold values ($T1 \leq 0.02$ and $D1 \leq 0.05$) [@Lee2003]. Here, this is the case for reactions MF2, MF4, GX1, GA1, GA4, and EG4. These reactions are included in Table \[tbl:benchmark\] and the deviation in the activation energy ranges between 0.4 and 3.3 kJ/mol, *i.e.*, within chemical accuracy.
Furthermore, the extent of the multireference character for reaction type 2 (MF2, GX2 and GA2) was tested via MRCI-F12/VTZ-F12//MPWB1K/def2-TZVP calculations [@Shiozaki:2011; @Shiozaki:2011a; @Peterson2008] for a reaction of the same type, but with a smaller reactant: . These single-point energy calculations indicate that the reaction does not have a large multireference character. Firstly, the CI coefficients for the reference wavefuntion of the transition state structure correspond to 0.934, -0.124, 0.074 and -0.051. In addition, the activation energies at DFT, CCSD(T)-F12 and MRCI-F12 level are similar (38.8, 40.6 and 36.4 kJ/mol, respectively). Therefore, here the CCSD(T)-F12 method is considered to be a reasonable reference method for these specific three reactions as well, *i.e.*, for a H addition to an aldehyde oxygen.
The MPWB1K functional has been shown to provide a good description for 9 out of the 17 reactions dealt with here, with reaction types 1 to 4 being included in this benchmark. Therefore, we assume that the other 8 reactions, including reaction type 5 and 6, can also be described with the same functional and basis set combination. As a double-check we have tested several functionals suggested by @Truhlar_mf_2016 to make sure that the activation energies obtained are of the correct magnitude, see Table \[tbl:dft\_comp\].
[lll]{} & $\Delta E^{\ddagger}$ DFT & $\Delta E^{\ddagger}$ CC\
MF2 & 58.8 & 59.2\
MF3 & 46.5 & *48.3*$^{a}$\
MF4 & 51.1 & 52.8\
GX1 & 14.7 & 15.4\
GX2 & 29.7 & *31.9*$^{a}$\
GA1 & 18.5 & 17.5\
GA2 & 38.2 & *37.7*$^{a}$\
GA4 & 26.5 & 29.5\
EG4 & 27.9 & 31.2\
\
MPWB1K M06-2X MPW1B95 MN12-SX N12-SX SOGGA11-X
------------ -------- -------- --------- --------- -------- -----------
[H + MF]{}
MF1 38.0 39.2 35.0 33.7 43.1 39.1
MF2 58.8 57.1 52.0 56.7 59.3 61.3
MF3 46.5 50.5 38.0 43.3 43.5 46.2
MF4 51.1 55.1 44.3 50.1 48.7 52.5
MF5 145.8 138.0 131.5 142.3 137.6 148.1
[H + GX]{}
GX1 14.7 17.0 12.9 8.3 21.9 15.7
GX2 29.7 33.1 24.5 24.3 32.2 30.1
[H + GA]{}
GA1 18.5 18.0 16.8 11.3 27.4 20.0
GA2 38.2 36.9 32.4 35.8 42.3 39.8
GA3 23.8 28.4 16.1 20.9 24.8 23.1
GA4 26.5 32.4 20.1 26.7 28.3 27.0
GA6 55.4 59.3 45.7 50.5 51.9 52.8
[H + EG]{}
EG4 27.9 33.2 21.0 29.0 28.1 29.2
EG6 53.5 57.8 44.4 49.6 50.7 51.3
Rate Constants
==============
Tables \[tbl:k\_MF\]-\[tbl:k\_FA\] give the values for the unimolecular reaction rate constants as calculated with instanton theory and corresponding to Figs. \[fig:mf\]-\[fig:eg\] and \[fig:fa\] of the main manuscript.
[llllll]{} T(K) & MF1 & MF2 & MF3 & MF4 & MF5\
75 & 3.60E-01 & 1.75E+00 & & 1.13E-01$^{b}$ & 3.77E-33\
80 & & & 3.37E+00$^{a}$ & 1.38E-01$^{a}$ &\
85 & 4.80E-01 & 1.86E+00 & 4.16E+00$^{a}$ & 1.61E-01$^{a}$ & 1.70E-32\
90 & 5.72E-01 & 1.90E+00 & 5.20E+00$^{a}$ & 1.87E-01$^{a}$ & 8.08E-32\
95 & 6.91E-01 & 2.08E+00 & 6.47E+00$^{a}$ & 2.25E-01$^{a}$ & 1.78E-31\
100 & & & 8.23E+00$^{a}$ & 2.79E-01 &\
105 & 1.02E+00 & 2.54E+00 & 1.12E+01 & 3.76E-01 & 1.95E-30\
110 & & & 1.41E+01 & 4.66E-01 &\
120 & & & 2.29E+01 & 7.50E-01 &\
130 & & & 3.77E+01 & 1.24E+00 &\
140 & 5.56E+00 & 5.60E+00 & 6.22E+01 & 2.08E+00 & 1.31E-26\
150 & & & 1.03E+02 & &\
160 & & & 1.70E+02 & &\
170 & 2.80E+01 & 1.66E+01 & 2.78E+02 & 1.04E+01 & 2.73E-23\
200 & 1.61E+02 & 5.11E+01 & 1.15E+03 & 4.93E+01 & 3.13E-20\
250 & 3.80E+03 & 3.53E+02 & 9.31E+03 & 5.45E+02 & 6.43E-16\
300 & & 2.66E+03 & 6.22E+04 & 5.35E+03 & 1.72E-12\
350 & & 2.03E+04 & 3.70E+05 & & 9.58E-10\
\
T(K) GX1 GX2
------ ---------- ----------
75 9.62E+06 1.81E+03
85 1.02E+07 2.27E+03
90 1.16E+07 2.39E+03
95 1.33E+07 2.56E+03
105 1.76E+07 3.45E+03
140 6.97E+07 9.85E+03
170 2.88E+08 2.69E+04
200 7.55E+04
250 4.44E+05
T(K) GA1 GA2 GA3 GA4 GA6
------ ---------- ---------- ---------- ---------- ----------
75 2.78E+05 2.77E+02 6.83E+07 2.56E+04 9.61E-01
85 3.00E+05 1.90E+02 7.84E+07 3.23E+04 1.37E+00
90 3.22E+05 1.97E+02 8.66E+07 3.63E+04 1.36E+00
95 3.56E+05 2.24E+02 9.74E+07 4.13E+04 1.45E+00
105 4.62E+05 3.04E+02 1.26E+08 5.48E+04 1.85E+00
140 1.91E+06 1.07E+03 3.70E+08 1.57E+05 7.40E+00
170 7.83E+06 2.68E+03 9.96E+08 4.04E+05 2.77E+01
200 7.12E+03 2.72E+09 1.02E+06 9.41E+01
250 4.37E+04 1.26E+10 4.34E+06 6.25E+02
300 2.96E+05 3.96E+10 1.54E+07 4.16E+03
350 2.74E+04
T(K) EG4 EG6
------ ---------- ----------
75 3.47E+06
85 4.60E+06
90 5.46E+06 2.24E+03
95 6.20E+06 2.13E+03
105 7.48E+06 2.25E+03
140 2.73E+07 6.30E+03
170 7.54E+07 1.72E+04
200 2.28E+08 4.63E+04
250 1.02E+09 2.15E+05
300 1.11E+06
350 5.98E+06
T(K) FAR1 FAR2
------ ---------- ----------
50 1.99E+03
55 2.29E+03
65 3.16E+03 4.59E-09
75 4.89E+03 1.03E-08
85 7.87E+03 3.51E-08
90 9.93E+03 2.58E-07
95 1.26E+04 1.88E-07
105 2.13E+04 1.26E-06
140 1.24E+05 6.05E-04
170 5.77E+05
200 2.63E+06
250 2.99E+07
300 2.62E+08
\[lastpage\]
[^1]: E-mail: [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Bi-Hamiltonian structures involving Hamiltonian operators of degree 2 are studied. Firstly, pairs of degree 2 operators are considered in terms of an algebra structure on the space of 1-forms, related to so-called Fermionic Novikov algebras. Then, degree 2 operators are considered as deformations of hydrodynamic type Poisson brackets.'
address: |
Department of Mathematics\
University of Glasgow\
Glasgow G12 8QW\
U.K.
author:
- 'James T. Ferguson'
date: 'April 11, 2007'
title: Flat Pencils of Symplectic Connections and Hamiltonian Operators of Degree 2
---
Introduction
============
Hamilton’s equations for a finite-dimensional system with position coordinates $q^i$ and associated momenta $p_i$, $$\begin{aligned}
\frac{dq^i}{dt} &=& {\frac{\partial H}{\partial p_i}}\,,\\
\frac{dp_i}{dt} &=& -{\frac{\partial H}{\partial q^i}}\,,
\end{aligned}$$ are understood geometrically as describing the flow of a vector field $X_H$ which is associated with the Hamiltonian function $H(q^1,\dots,q^n,p_1\dots,p_n)$ by the formula $X_H(f)=\{f,H\}$, where $\{\cdot,\cdot\}$ is the Poisson bracket: $$\label{eqn:canpoisson}
\{f,g\}=\sum_{i=1}^n\left({\frac{\partial f}{\partial q^i}}{\frac{\partial g}{\partial p_i}}
-{\frac{\partial f}{\partial p_i}}{\frac{\partial g}{\partial q^i}}\right)\,.$$
More generally, one defines a Poisson bracket on an n-dimensional manifold $M$ as a map $C^{\infty}(M)\times C^{\infty}(M)\rightarrow
C^{\infty}(M)$, $(f,g)\mapsto\{f,g\}$, satisfying, for any functions $f,g,h$ on $M$:
1. antisymmetry: $\{f,g\}=-\{g,f\}$,
2. linearity: $\{af+bg,h\}=a\{f,h\}+b\{g,h\}$ for any constants $a,b$,
3. product rule: $\{fg,h\}=f\{g,h\}+g\{f,h\}$,
4. Jacobi identity: $\{\{f,g\},h\}+\{\{g,h\},f\}+\{\{h,f\},g\}=0$.
The conditions 1-3 identify $\{\cdot,\cdot\}$ as a bivector: a rank two, antisymmetric, contravariant tensor field $\omega$ on $M$. It can therefore be represented, by introducing coordinates $\{u^i\}$ on $M$, as a matrix of coefficients $\omega^{ij}$, giving $$\omega =\omega^{ij}{\frac{\partial~}{\partial u^i}}\otimes{\frac{\partial~}{\partial u^j}}
=\frac{1}{2}\omega^{ij}{\frac{\partial~}{\partial u^i}}\wedge{\frac{\partial~}{\partial u^j}}\,,$$ and $$\label{eqn:fdpbop}
\{f,g\} =\omega^{ij}{\frac{\partial f}{\partial u^i}}{\frac{\partial g}{\partial u^j}}\,.$$
The Jacobi identity places the following constraint on the components of $\omega$: $$\label{eqn:fdjac}
\omega^{ir}{\frac{\partial \omega^{jk}}{\partial u^r}}
+\omega^{jr}{\frac{\partial \omega^{ki}}{\partial u^r}}
+\omega^{kr}{\frac{\partial \omega^{ij}}{\partial u^r}}=0\,.$$
If the matrix $\omega^{ij}$ is non-degenerate, we may introduce its inverse $\omega_{ij}$, satisfying $\omega_{ir}\omega^{rj}=\delta^j_i$. The Jacobi identity for $\omega^{ij}$ is equivalent to the closedness of $\omega_{ij}$. We refer to a closed non-degenerate two-form as a symplectic form, and a manifold equipped with one as a symplectic manifold. Darboux’s theorem asserts that on any $2n$-dimensional symplectic manifold there exists a set of local coordinates $\{q^1,\dots,q^n,p_1\dots,p_n\}$ in which the Poisson bracket takes the form (\[eqn:canpoisson\]); i.e. the components of $\omega^{ij}$, and so those of $\omega_{ij}$, are constant.
One may also introduce Poisson brackets on infinite-dimensional manifolds. The loop space of a finite-dimensional manifold $M$, $L(M)$, is the space of smooth maps $u:S^1\rightarrow M$. Poisson brackets relating Hamiltonians to flows in $L(M)$ will therefore act on functionals mapping $L(M)\rightarrow\mathbb{R}$. In [@dubnov84],[@dubnov89] Dubrovin and Novikov studied the so-called Poisson brackets of differential-geometric type, which are of the form $$\label{def:pb}
\{f,g\}= \int\frac{\delta f}{\delta u^i}
P^{ij}\left(\frac{\delta g}{\delta u^j}\right) dx$$ where $u^{i}$ are coordinates on the target space $M$, and $x$ is the coordinate on $S^1$. $P^{ij}$ is a matrix of differential operators (in $\frac{d}{dx}$), with no explicit dependence on $x$, which is assumed to be polynomial in the derivatives $u^i_x,u^i_{xx},\dots$. If $\{\cdot,\cdot\}$ defines a Poisson bracket on the loop space then $P$ is referred to as a Hamiltonian operator.
There is a grading on such operators, preserved by diffeomorphisms of $M$, given by assigning degree 1 to $\frac{d}{dx}$, and degree $n$ to the $n^{th}$ $x$-derivative of each field $u^i$. An important class is the hydrodynamic type Poisson brackets, which are homogeneous of degree 1: $$P^{ij}=g^{ij}(u)\frac{d}{dx}+\Gamma^{ij}_k(u)u^k_x\,.$$
According to the programme set out by Novikov [@novikov85], differential-geometric type Poisson brackets on $L(M)$ should be studied in terms of finite-dimensional differential geometry on the target space $M$. When expanded as a polynomial in $\frac{d~}{dx}$ and the field derivatives, the coefficients, which are functions of the fields $u^i$ alone, can often be naturally related to known objects of differential geometry, or else used to define new ones. In the hydrodynamic case, for instance, with $g^{ij}$ non-degenerate, $P$ is Hamiltonian if and only if $g^{ij}$ is a flat metric on $M$ and $\Gamma^k_{ij}=-g_{ir}\Gamma^{rk}_j$ are the Christoffel symbols of its Levi-Civita connection.
In [@dubflatpencil] Dubrovin considered the geometry of bi-Hamiltonian structures of Hydrodynamic operators, that is pairs of such operators compatible in the sense of [@magribiham], that every linear combination of them also determines a Poisson bracket. In particular, he introduced a multiplication of covectors on $M$ and expressed the compatibility of the operators in terms of a quadratic relations on this algebra.
This paper is principally concerned with Hamiltonian operators which are homogeneous of degree 2. Section \[section:H2O\] presents the differential geometry of such operators, and in particular relates the subclass which can be put into a constant form by a change of coordinates on $M$ to symplectic connections. Section \[section:pencils\] then considers pairs of operators from this subclass, and the algebraic constraints their compatibility places upon the associated multiplication. In section \[section:bhd12\] inhomogeneous bi-Hamiltonian structures consisting of a degree 1 and a degree 2 operator are studied.
Hamiltonian Operators of Degree 2 {#section:H2O}
=================================
We begin with a review of known results on Hamiltonian operators of degree 2: $$\label{def:h20}
P^{ij}=a^{ij}\left(\frac{d}{dx}\right)^{2}+b^{ij}_{k}u^{k}_{x}\frac{d}{dx}+c^{ij}_{kl}u^{k}_{x}u^{l}_{x}+c^{ij}_{k}u^{k}_{xx},$$ in which the matrix $a^{ij}$ is assumed to be non-degenerate. Such operators have been considered already in, for example, [@potemin86], [@mokhov98], [@doyle93], [@novikov85], in which the (conditional) Darboux theorem has been discussed. In preparation for the bi-Hamiltonian theory we present these results without the use of special coordinates.
Under the change of coordinates $\tilde{u}^i=\tilde{u}^i(u^p)$ the coefficients in $P^{ij}$ transform as $$\begin{aligned}
\tilde{a}^{ij}&=&{\frac{\partial\tilde{u}^{i}}{\partial
u^{p}}}{\frac{\partial\tilde{u}^{j}}{\partial
u^{q}}}a^{pq}\nonumber\,,\\
\tilde{b}^{ij}_{k}&=&{\frac{\partial\tilde{u}^{i}}{\partial
u^{p}}}{\frac{\partial\tilde{u}^{j}}{\partial
u^{q}}}{\frac{\partial u^{r}}{\partial\tilde{u}^{k}}}b^{pq}_{r}
-2{\frac{\partial\tilde{u}^{i}}{\partial
u^{p}}}{\frac{\partial\tilde{u}^{s}}{\partial
u^{q}}}{\frac{\partial\tilde{u}^{j}}{\partial
u^{r}}}{\frac{\partial^2 u^{r}}{\partial\tilde{u}^{k}
\partial\tilde{u}^{s}}}a^{pq}\nonumber\,,\\
\tilde{c}^{ij}_{k}&=&{\frac{\partial\tilde{u}^{i}}{\partial
u^{p}}}{\frac{\partial\tilde{u}^{j}}{\partial
u^{q}}}{\frac{\partial u^{r}}{\partial\tilde{u}^{k}}}c^{pq}_{r}
-{\frac{\partial\tilde{u}^{i}}{\partial
u^{p}}}{\frac{\partial\tilde{u}^{s}}{\partial
u^{q}}}{\frac{\partial\tilde{u}^{j}}{\partial
u^{r}}}{\frac{\partial^2 u^{r}}{\partial\tilde{u}^{k}
\partial\tilde{u}^{s}}}a^{pq}\nonumber\,,\\
\tilde{c}^{ij}_{kl}&=&{\frac{\partial\tilde{u}^{i}}{\partial
u^{p}}}{\frac{\partial\tilde{u}^{j}}{\partial
u^{q}}}{\frac{\partial u^{r}}{\partial\tilde{u}^{k}}}{\frac{\partial u^{s}}{\partial\tilde{u}^{l}}}c^{pq}_{rs}
+{\frac{\partial\tilde{u}^{i}}{\partial
u^{p}}}{\frac{\partial\tilde{u}^{j}}{\partial
u^{q}}}{\frac{\partial^2 u^{r}}{\partial\tilde{u}^{k}
\partial\tilde{u}^{l}}}c^{pq}_{r}\nonumber\\
&& +
{\frac{\partial\tilde{u}^{i}}{\partial
u^{p}}}{\frac{\partial^2 \tilde{u}^{j}}{\partial u^{q}
\partial u^{s}}}{\frac{\partial u^{r}}{\partial\tilde{u}^{(k}}}{\frac{\partial u^{s}}{\partial\tilde{u}^{l)}}}b^{pq}_{r} +
{\frac{\partial\tilde{u}^{i}}{\partial
u^{p}}}\frac{\partial^3 \tilde{u}^j}{\partial u^q \partial
u^r \partial u^s}{\frac{\partial u^{r}}{\partial\tilde{u}^{k}}}{\frac{\partial u^{s}}{\partial\tilde{u}^{l}}}a^{pq}\nonumber\\&& +
{\frac{\partial\tilde{u}^{i}}{\partial
u^{p}}}{\frac{\partial^2 \tilde{u}^{j}}{\partial u^{q}
\partial u^{r}}}{\frac{\partial^2 u^{r}}{\partial\tilde{u}^{k}
\partial\tilde{u}^{l}}}a^{pq}\,,\label{eqn:trrules}
\end{aligned}$$ where the brackets denote symmetrisation. So in particular $a^{ij}$ transforms as a rank 2 contravariant tensor on the target space and $b^{ij}_{k}$ and $c^{ij}_{k}$ are related to Christoffel symbols of connections by $b^{ij}_{k}=-2a^{ir}\bar{\Gamma}^{j}_{rk}$ and $c^{ij}_{k}=-a^{ir}\Gamma^{j}_{rk}$. Call these connections $\bar{\nabla}$ and $\nabla$ respectively.
The transformation rules for $c^{ij}_{kl}$ are not determined uniquely by those for $P$, since (\[def:h20\]) sees only the part symmetric in $k$ and $l$. To fix $c^{ij}_{kl}$, we always assume the antisymmetric part is zero. Denote by $a_{ij}$ the inverse of $a^{ij}$ defined by $a_{ir}a^{rj}=\delta^{j}_{i}$.
The condition that the operation defined in (\[def:pb\]) is skew-symmetric and satisfies the Jacobi identity places constraints on the coefficients appearing in (\[def:h20\]).
\[thm:hamcond\] The operator $P$ in equation (\[def:h20\]) defines a Poisson bracket by equation (\[def:pb\]) if and only if
1. $a^{ij}=-a^{ji}$\[eqn:hamcond:antisym\],
2. $\nabla_{k}a^{ij}=b^{ij}_{k}-2c^{ij}_{k}$\[eqn:hamcond:deriv\],
3. $a^{ir}\left(b^{jk}_{r}-2c^{jk}_{r}\right)=a^{kr}\left(b^{ij}_{r}-2c^{ij}_{r}\right)$ \[eqn:hamcond:cyclic\],
4. $\nabla$ is flat (zero torsion, zero curvature)\[eqn:hamcond:flat\],
5. $c^{ij}_{kl}=c^{ij}_{(k,l)}-a_{pr}c^{ri}_{(k}c^{pj}_{l)}$\[eqn:hamcond:c4reln\].
[@mokhov98] states that, by virtue of being Hamiltonian, the operator (\[def:h20\]) can be put in the form $$\label{def:shortho}
P^{ij}=a^{ij}\left(\frac{d~}{dx}\right)^{2}+b^{ij}_{k}u^{k}_{x}\frac{d}{dx}\,,$$ by a change of coordinates $u^i=u^i({\tilde u})$, and that for an operator of this shorter form to be Hamiltonian is equivalent to the three conditions
1. \[eqn:mokhov:asymm\] $a^{ij}=-a^{ji}$,
2. \[eqn:mokhov:deriv\] $a^{ij},_k=b^{ij}_k$,
3. \[eqn:mokhov:cyclic\] $a^{ir}b^{jk}_r=a^{jr}b^{ki}_r$.
We first assume that $P$ is a Poisson bracket, so there exists the special coordinates in which $P$ takes the form (\[def:shortho\]) and \[eqn:mokhov:asymm\]-\[eqn:mokhov:cyclic\] hold. By reversing the change of variables as ${\tilde u}^i={\tilde
u^i}(u)$, conditions \[eqn:hamcond:antisym\]-\[eqn:hamcond:cyclic\] of Theorem \[thm:hamcond\] are Mokhov’s three conditions converted to tensorial identities. That $\nabla$ is flat follows from its Christoffel symbols, $\Gamma^k_{ij}=-a_{ir}c^{rk}_j$, being zero in the $u$ coordinates.
The formula in condition \[eqn:hamcond:c4reln\] is derived from the transformation rules above. In changing from flat coordinates $u^i$ to coordinates ${\tilde
u}^i$ they give: $$\begin{aligned}
\tilde{c}^{ij}_{kl}&=&
{\frac{\partial\tilde{u}^{i}}{\partial
u^{p}}}{\frac{\partial^2 \tilde{u}^{j}}{\partial u^{q}
\partial u^{s}}}{\frac{\partial u^{s}}{\partial\tilde{u}^{(k}}}{\frac{\partial u^{s}}{\partial\tilde{u}^{l)}}}b^{pq}_{r} +
{\frac{\partial\tilde{u}^{i}}{\partial
u^{p}}}\frac{\partial^3 \tilde{u}^j}{\partial u^q \partial
u^r \partial u^s}{\frac{\partial u^{s}}{\partial\tilde{u}^{k}}}{\frac{\partial u^{s}}{\partial\tilde{u}^{l}}}a^{pq}\\&& +
{\frac{\partial\tilde{u}^{i}}{\partial
u^{p}}}{\frac{\partial^2 \tilde{u}^{j}}{\partial u^{q}
\partial u^{r}}}{\frac{\partial^2 u^{r}}{\partial\tilde{u}^{k}
\partial\tilde{u}^{l}}}a^{pq}\,,
\end{aligned}$$ and $$\begin{aligned}
\tilde{c}^{ij}_{k}&=&
-{\frac{\partial\tilde{u}^{i}}{\partial
u^{p}}}{\frac{\partial\tilde{u}^{s}}{\partial
u^{q}}}{\frac{\partial\tilde{u}^{j}}{\partial
u^{r}}}{\frac{\partial^2 u^{r}}{\partial\tilde{u}^{k}
\partial\tilde{u}^{s}}}a^{pq}\,,\\
&=& {\frac{\partial\tilde{u}^{i}}{\partial
u^{p}}}{\frac{\partial^2 \tilde{u}^{j}}{\partial u^{q}
\partial u^{r}}}{\frac{\partial u^{r}}{\partial\tilde{u}^{k}}}a^{pq}\,,
\end{aligned}$$ where the last line has used the identity $${\frac{\partial^2 \tilde{u}^{i}}{\partial u^{r}
\partial u^{s}}}{\frac{\partial u^{r}}{\partial\tilde{u}^{j}}}{\frac{\partial u^{s}}{\partial\tilde{u}^{k}}} +
{\frac{\partial\tilde{u}^{i}}{\partial
u^{r}}}{\frac{\partial^2 u^{r}}{\partial\tilde{u}^{j}
\partial\tilde{u}^{k}}}=0\,,$$ which is a differential consequence of ${\frac{\partial\tilde{u}^{i}}{\partial
u^{r}}}{\frac{\partial u^{r}}{\partial\tilde{u}^{j}}}=\delta^i_j\,.$ $$\begin{aligned}
{\tilde c}^{ij}_{k,l} &=& {\frac{\partial {\tilde c}^{ij}_k}{\partial {\tilde u}_l}}\\
&=& {\frac{\partial^2 \tilde{u}^{i}}{\partial u^{p}
\partial u^{s}}}{\frac{\partial u^{s}}{\partial\tilde{u}^{l}}}{\frac{\partial^2 \tilde{u}^{j}}{\partial u^{r}
\partial u^{q}}}{\frac{\partial u^{r}}{\partial\tilde{u}^{k}}}a^{pq}\\
&&+{\frac{\partial\tilde{u}^{i}}{\partial
u^{p}}}\frac{\partial^3 {\tilde u}^j}{\partial u^q\partial
u^r\partial u^s}{\frac{\partial u^{r}}{\partial\tilde{u}^{k}}}{\frac{\partial u^{s}}{\partial\tilde{u}^{l}}}a^{pq}\\
&&+{\frac{\partial\tilde{u}^{i}}{\partial
u^{p}}}{\frac{\partial^2 \tilde{u}^{j}}{\partial u^{q}
\partial u^{r}}}{\frac{\partial^2 u^{r}}{\partial\tilde{u}^{k}
\partial\tilde{u}^{l}}}a^{pq}\\
&&+{\frac{\partial\tilde{u}^{i}}{\partial
u^{p}}}{\frac{\partial^2 \tilde{u}^{j}}{\partial u^{q}
\partial u^{r}}}{\frac{\partial u^{r}}{\partial\tilde{u}^{k}}}{\frac{\partial u^{s}}{\partial\tilde{u}^{l}}}b^{pq}_s\,,
\end{aligned}$$ from which we see $${\tilde c}^{ij}_{kl} ={\tilde c}^{ij}_{(k,l)}
-{\frac{\partial^2 \tilde{u}^{i}}{\partial u^{p}
\partial u^{s}}}{\frac{\partial^2 \tilde{u}^{j}}{\partial u^{r}
\partial u^{q}}}{\frac{\partial u^{s}}{\partial\tilde{u}^{(l}}}{\frac{\partial u^{r}}{\partial\tilde{u}^{k)}}}a^{pq}\,.$$ This last term can be seen to be $${\tilde a}_{pr}{\tilde
c}^{ri}_{(k}{\tilde c}^{pj}_{l)}\,.$$
Conversely, if \[eqn:hamcond:antisym\]-\[eqn:hamcond:c4reln\] hold, the flatness of $\nabla$ asserts the existence of coordinates in which $c^{ij}_k=0$, and condition \[eqn:hamcond:c4reln\] then asserts that $c^{ij}_{kl}=0$ in these coordinates.
\[sec:withlinh20\] If we take, as a simple case, an operator $P$ as in (\[def:h20\]) with $b^{ij}_k=2c^{ij}_k$ constants, and assume $c^{ij}_{kl}$ to be defined by \[eqn:hamcond:c4reln\], then $P$ is Hamiltonian if and only if $a^{ij}=A^{ij}_ku^k+A^{ij}_0$ where $A^{ij}_k,A^{ij}_0$ are constants with $A^{ij}_k=c^{ij}_k-c^{ji}_k$, $A^{ir}_lc^{jk}_r=A^{jr}_lc^{ik}_r$, $A^{ir}_0c^{jk}_r=A^{jr}_0c^{ik}_r$ and $c^{ij}_rc^{rk}+c^{ik}_rc^{rj}_l=0$.
If we take an algebra $\mathcal{A}$ with basis $\{e^1,\dots,e^n\}$, $n=\rm{dim}M$, and use $c^{ij}_k$ and $A^{ij}_0$ to define a multiplication, $\circ$, and skew-symmetric bilinear form, $\langle\cdot,\cdot\rangle$, by $e^i\circ e^j=c^{ij}_re^r$ and $\langle e^i,e^j\rangle=A^{ij}_0$, then we may rewrite these conditions as $$\begin{aligned}
e^i\circ e^j -e^j\circ e^i &=& A^{ij}_re^r\,,\nonumber\\
(I\circ J)\circ K &=& -(I\circ K)\circ J\,, \label{eqn:circfna}\\
\Lambda(I,J,K) &=& \Lambda(J,I,K)\,, \label{eqn:vindberg}\\
\text{and}\qquad \langle I,J\circ K\rangle &=& \langle J,I\circ K\rangle\,,\nonumber
\end{aligned}$$ for all $I,J,K\in\mathcal{A}$, where $\Lambda$ is the associator of $\circ$: $\Lambda(I,J,K)=(I\circ J)\circ K -I\circ(J\circ K)$.
Algebras satisfying conditions (\[eqn:circfna\]) and (\[eqn:vindberg\]) have appeared before in [@xufermion], in the context of linear hydrodynamic Hamiltonian operators taking values in a completely odd superspace, where the following definition was proposed:
\[def:fna\] An algebra $(\mathcal{A},\circ)$ satisfying conditions (\[eqn:circfna\]) and (\[eqn:vindberg\]) is called a Fermionic Novikov algebra.
In [@baimenghe02] Fermionic Novikov algebras in dimensions 2-5 were studied, and the listing therein provides a source of examples of Hamiltonian operators of degree two.
$$\begin{aligned}
P &=& \left(\begin{array}{cccc} 0&0&0&a\\ 0&0&-a&-b-(t-1)u^1\\
0&a&0&c-u^2\\ -a&b+(t-1)u^1&-c+u^2&0 \end{array}\right)
\left(\frac{d~}{dx}\right)^2\\
&& + 2\left(\begin{array}{cccc} 0&0&0&0\\
0&0&0&u^1_x\\ 0&0&-u^1_x&0\\ 0&\tau u^1_x&u^2_x&u^3_x
\end{array}\right) \left(\frac{d~}{dx}\right)\\
&& +\left(\frac{1}{a}\right)\left(\begin{array}{cccc} 0&0&0&0\\
0&0&0&0\\ 0&0&0&(u^1_x)^2\\ 0&0&-(u^1_x)^2&0\end{array}\right)
+\left(\begin{array}{cccc} 0&0&0&0\\
0&0&0&u^1_x\\ 0&0&-u^1_{xx}&0\\ 0&\tau u^1_{xx}&u^2_{xx}&u^3_{xx}
\end{array}\right)
\end{aligned}$$
is Hamiltonian for all values of the constants $a,b,c$ and $\tau$ with $a\neq 0$. This is the most general Hamiltonian operator associated in the manner discussed above to the algebra designated $(44)_\tau$ in [@baimenghe02].
Returning to the general Hamiltonian operator (\[def:h20\]), it can be seen from conditions \[eqn:hamcond:deriv\] and \[eqn:hamcond:c4reln\] in Theorem \[thm:hamcond\] that the coefficients $b^{ij}_k$ and $c^{ij}_{kl}$ in (\[def:h20\]) are completely determined by $a^{ij}$ and $c^{ij}_k$. Thus the Hamiltonian operator on $L(M)$ is represented uniquely on $M$ by only these latter two objects.
\[thm:adela\] There is a one-to-one correspondence between Hamiltonian operators of the form (\[def:h20\]) on $L(M)$ and pairs $(a,\nabla)$ on $M$ consisting of a non-degenerate bivector $a^{ij}$ and a torsion-free connection $\nabla$ satisfying two conditions: firstly, that the curvature of $\nabla$ vanishes, and secondly, $$\label{eqn:bivectcompat}
a^{ir}\nabla_ra^{jk}=a^{jr}\nabla_ra^{ki}\,.$$ The Christoffel symbols, $\Gamma_{ij}^k$, of $\nabla$ are related to $c^{ij}_k$ by $c^{ij}_k=-a^{ir}\Gamma_{rk}^j$. We then have $$\begin{aligned}
b^{ij}_k &=& \nabla_ka^{ij}+2c^{ij}_k\,,\\
c^{ij}_{kl} &=& c^{ij}_{k,l}-a_{pr}c^{ri}_{(k}c^{pj}_{l)}\,.
\end{aligned}$$
With this, we may verify the following facts [@potemin86],[@mokhov98]:
\[cor:symbit\] For P in (\[def:h20\]) a Hamiltonian operator we have
1. $\Gamma$ is the symmetric part of $\bar{\Gamma}$,
2. Let $\bar{T}^{k}_{ij}=\bar{\Gamma}^{k}_{ij}-\bar{\Gamma}^{k}_{ji}$ be the torsion of $\bar{\nabla}$. Then $\bar{T}_{ijk}=a_{ir}\bar{T}^{r}_{jk}$ is skew symmetric and the forms $\bar{T}=\frac{1}{6}\bar{T}_{ijk}du^{i}\wedge
du^{j}\wedge du^{k}$ and $a=\frac{1}{2}a_{ij}du^{i}\wedge
du^{j}$ are related by $3\bar{T}=da$.
We begin by noting that equation (\[eqn:bivectcompat\]) is equivalent to the condition $$\label{eqn:2formcompat}
\nabla_ka_{ij}=\nabla_ia_{jk}$$ on the two-form $a_{ij}$.
In terms of covariant Christoffel symbols, Theorem \[thm:adela\] gives $$\label{eqn:covbfromc}
\bar{\Gamma}_{ij}^k =
\frac{1}{2}a^{kr}\nabla_ra_{ij}+\Gamma_{ij}^k\,,$$ from which it is clear that $\bar{\Gamma}_{(ij)}^k=\Gamma_{ij}^k$.
We therefore also have $$\frac{1}{2}\nabla_ka_{ij}=\bar{\Gamma}_{ijk}-\Gamma_{ijk}\,,$$ where $\bar{\Gamma}_{ijk}=a_{ir}\bar{\Gamma}^{r}_{jk}$ and $\Gamma_{ijk}=a_{ir}\Gamma^{r}_{jk}$. Because $\nabla$ is torsion-free we have $$\begin{aligned}
\bar{T}_{ijk} &=& \bar{\Gamma}_{ijk}-\bar{\Gamma}_{ikj}\,,\\
&=&
\bar{\Gamma}_{ijk}-\Gamma_{ijk}-\bar{\Gamma}_{ikj}+\Gamma_{ikj}\,,\\
&=& \frac{1}{2}\nabla_ka_{ij} -\frac{1}{2}\nabla_ja_{ik}\,,\\
&=& \nabla_ka_{ij}\,,\\
&=& \nabla_{[k}a_{ij]}\,,\\
&=& \frac{1}{3}(da)_{ijk}\,.
\end{aligned}$$
\[lem:3equiv\] For a Hamiltonian operator of the form (\[def:h20\]), the following three statements, presented in both covariant and contravariant forms, are equivalent:
1. The 2-form $a$ is closed (and so symplectic), or equivalently $a^{ij}$ satisfies equation (\[eqn:fdjac\]) (and so defines a Poisson bracket on $M$ by equation (\[eqn:fdpbop\]));
2. $\nabla_{k}a^{ij}=0$, i.e. $\nabla_{k}a_{ij}=0$;
3. $b^{ij}_{k}=2c^{ij}_{k}$, i.e. $\Gamma^{k}_{ij}=\bar{\Gamma}^{k}_{ij}$.
We see, from the characterisation of Hamiltonian operators given in Theorem \[thm:adela\], $$\begin{aligned}
a^{ij}\text{ is Poisson} &\Longleftrightarrow&
a^{ir}a^{jk}_{,r}+a^{jr}a^{ki}_{,r}+a^{kr}a^{ij}_{,r}=0\\
&\Longleftrightarrow&
a^{ir}\nabla_ra^{jk}+a^{jr}\nabla_ra^{ki}+a^{kr}\nabla_ra^{ij}=0\\
&\Longleftrightarrow& 3a^{kr}\nabla_ra^{ij}=0\\
&\Longleftrightarrow& \nabla_ka^{ij}=0\,,\\
&\Longleftrightarrow& b^{ij}_k=2c^{ij}_k\,.
\end{aligned}$$
Lemma \[lem:3equiv\] therefore tells us that in the special case where the leading coefficient in $P$ is the inverse of a symplectic form, the pair $(a,\nabla)$ defining $P$ can be thought of as containing the symplectic form $a_{ij}$, and a torsionless connection compatible with it (in the sense that $\nabla a=0$); that is, a symplectic connection. More precisely (see e.g. [@bieliavsky05]):
\[def:sympcon\] A symplectic connection on a symplectic manifold $(M,\omega)$ is a smooth connection $\nabla$ which is torsion-free and compatible with the symplectic form $\omega$, i.e. $$\nabla_XY-\nabla_YX-[X,Y]=0$$ and $$\left(\nabla\omega\right)(X,Y,Z)=X(\omega(Y,Z))-\omega(\nabla_XY,Z)-\omega(Y,\nabla_YZ)=0\,,$$ where $X,Y$ and $Z$ are vector fields on $M$.
In local coordinates $\{x^i\}$, introducing Christoffel symbols $\Gamma_{ij}^k$ for $\nabla$ and writing $\omega=\frac{1}{2}\omega_{ij}dx^i\wedge dx^j$, the conditions for $\nabla$ to be a symplectic connection read $\Gamma_{ij}^k=\Gamma_{ji}^k$, as usual, and $$\label{eqn:symconcompat}
\nabla_k\omega_{ij} = \frac{\partial\omega_{ij}}{\partial x^r}
- \Gamma^r_{ki}\omega_{rj} - \Gamma^r_{kj}\omega_{ir} = 0\,.$$
This definition is analogous to that of the Levi-Civita connection of a pseudo-Riemannian metric, however there is an important difference in that the Levi-Civita connection is uniquely specified by its metric. From the compatibility condition (\[eqn:symconcompat\]) it can be seen that if $\Gamma_{ij}^k$ are the Christoffel symbols of a symplectic connection for $\omega$, then the connection with Christoffel symbols $\tilde{\Gamma}_{ij}^k=\Gamma_{ij}^k+\omega^{kr}S_{rij}$ is a symplectic connection if and only if the tensor $S_{ijk}$ is completely symmetric. In [@gelfandfedosov] a symplectic manifold with a specified symplectic connection is called, in light of [@fedosov94], a Fedosov manifold. Here we call the pair $(\omega,\nabla)$ of a symplectic form and a symplectic connection a Fedosov structure on M, and call the structure flat if $\nabla$ is flat.
In the discussion of Hamiltonian operators it is convenient to work with contravariant quantities. We call $$\Gamma^{ij}_k=-\omega^{ir}\Gamma^j_{rk}$$ the contravariant Christoffel symbols of the symplectic connection.
\[res:compat\] The compatibility of $\nabla$ and $\omega$ is equivalent to $$\frac{\partial\omega^{ij}}{\partial x^k} =
\Gamma^{ij}_k-\Gamma^{ji}_k\,.$$
\[res:torsion\] $\nabla$ being torsion-free is equivalent to $\omega^{ir}\Gamma^{jk}_r=\omega^{jr}\Gamma^{ik}_r\,.$
The curvature of $\nabla$, $$R^{k}_{slt}=\partial_{s}\Gamma^{k}_{lt}-\partial_{l}\Gamma^{k}_{st}+\Gamma^{k}_{sr}\Gamma^{r}_{lt}-\Gamma^{k}_{lr}\Gamma^{r}_{st}\,,$$ can be expressed in terms of contravariant quantities by raising indices as $$R^{ijk}_l=\omega^{is}\omega^{jt}R^k_{slt}\,.$$ This gives
\[res:curv\] $$R^{ijk}_l=
\omega^{ir}\left(\partial_l\Gamma^{jk}_r-\partial_r\Gamma^{jk}_l\right)
+\Gamma^{ij}_r\Gamma^{rk}_l +\Gamma^{ik}_r\Gamma^{rj}_l\,.$$
Having introduced symplectic connections, we are now in a position to interpret the following Darboux theorem for Hamiltonian operators of degree 2:
\[thm:H2ODarboux\] [@potemin86] Given a Hamiltonian operator $$P^{ij}=a^{ij}\left(\frac{d}{dx}\right)^{2}+b^{ij}_{k}u^{k}_{x}\frac{d}{dx}+c^{ij}_{kl}u^{k}_{x}u^{l}_{x}+c^{ij}_{k}u^{k}_{xx}$$ where $a^{ij}$ is non-degenerate, then $P$ can be put in the constant form $P^{ij}=\omega^{ij}\left(\frac{d}{dx}\right)^2$ (where $\omega$ is a constant matrix) by a change of target space coordinates $\{u^i\}$ if and only if $a_{ij}$ is closed. The coordinates in which this happens are flat coordinates for the connection $\Gamma_{ij}^k=-g_{ir}c^{rk}_j$ which can be chosen, using a linear substitution, to be canonical coordinates for the symplectic form $a_{ij}=\omega_{ij}$.
In arbitrary coordinates operators satisfying the conditions of Theorem \[thm:H2ODarboux\] have the form $$\label{def:cpb}
P^{ij}=\omega^{ij}\left(\frac{d}{dx}\right)^2
+2\Gamma^{ij}_ku^k_x\frac{d}{dx} +c^{ij}_{kl}u^k_xu^l_x
+\Gamma^{ij}_ku^k_{xx}$$ where $\omega^{ij}$ is the inverse of a symplectic form, $c^{ij}_{kl}=\Gamma^{ij}_{(k,l)}-\omega_{pr}\Gamma^{ri}_{(k}\Gamma^{pj}_{l)}$, and $\Gamma^{ij}_k$ are the contravariant Christoffel symbols of a flat symplectic connection compatible with $\omega$. This class of operators on $L(M)$ is therefore in one-to-one correspondence with flat Fedosov structures on $M$.
Flat Pencils of Fedosov Structures {#section:pencils}
==================================
In this section we consider pairs of Hamiltonian operators of the form (\[def:cpb\]): $$\begin{aligned}
P_1^{ij}&=&\omega_1^{ij}\left(\frac{d}{dx}\right)^2
+2{\Gamma_1}^{ij}_ku^k_x\frac{d}{dx} +{c_1}^{ij}_{kl}u^k_xu^l_x
+{\Gamma_1}^{ij}_ku^k_{xx}\,,\\
P_2^{ij}&=&\omega_2^{ij}\left(\frac{d}{dx}\right)^2
+2{\Gamma_2}^{ij}_ku^k_x\frac{d}{dx} +{c_2}^{ij}_{kl}u^k_xu^l_x
+{\Gamma_2}^{ij}_ku^k_{xx}\,.
\end{aligned}$$
The first fact to establish is that if $P_1$ and $P_2$ are compatible then all elements of the pencil, $P_\lambda=P_1+\lambda P_2$, remain in the class (\[def:cpb\]).
\[thm:fdbh\] If $P_1$ and $P_2$ are compatible then $\omega_1^{ij}$ and $\omega_2^{ij}$ form a finite-dimensional bi-Hamiltonian structure on the target space.
$P_\lambda$ could have the general form $$P_\lambda^{ij}=a_\lambda^{ij}\left(\frac{d}{dx}\right)^{2}
+{b_\lambda}^{ij}_{k}u^{k}_{x}\frac{d}{dx}
+{c_\lambda}^{ij}_{kl}u^{k}_{x}u^{l}_{x}
+{c_\lambda}^{ij}_{k}u^{k}_{xx}\,,$$ but clearly ${b_\lambda}^{ij}_k=2{\Gamma_1}^{ij}_k+2\lambda{\Gamma_2}^{ij}_k$ and ${c_\lambda}^{ij}_k={\Gamma_1}^{ij}_k+\lambda{\Gamma_2}^{ij}_k$, so ${b_\lambda}^{ij}_k=2{c_\lambda}^{ij}_k$, and hence, by Lemma \[lem:3equiv\], $a_\lambda^{ij}$ satisfies the Jacobi identity (\[eqn:fdjac\]) for all $\lambda$.
So we write $$P_\lambda^{ij}=\omega_\lambda^{ij}\left(\frac{d}{dx}\right)^2
+2{\Gamma_\lambda}^{ij}_ku^k_x\frac{d}{dx}
+{c_\lambda}^{ij}_{kl}u^k_xu^l_x
+{\Gamma_\lambda}^{ij}_ku^k_{xx}\,.$$
An immediate corollary of Theorem \[thm:fdbh\] is that the tensor $L^i_j=\omega_1^{ir}\omega_{2rj}$ has vanishing Nijenhuis torsion.
Multiplication of covectors {#subsec:multiplication}
---------------------------
As in [@dubflatpencil], we proceed to understand the compatibility conditions on $P_1$ and $P_2$ in terms of the algebraic properties of a tensorial multiplication of covectors on $M$.
\[def:mult\] Using the tensors $$\begin{aligned}
\Delta^{sjk} &=&
\omega_2^{jr}{\Gamma_1}^{sk}_r-\omega_1^{sr}{\Gamma_2}^{jk}_r\,,\\
\Delta^{jk}_i &=& \omega_{2is}\Delta^{sjk}\,,
\end{aligned}$$ we define a multiplication $\circ$ of covectors on $M$ by $$(\alpha\circ\beta)_i=\alpha_j\beta_k\Delta^{jk}_i\,.$$
\[thm:circmult\] The compatibility of $P_1$ and $P_2$ is equivalent to $$\begin{aligned}
(I,J\circ K)_2 &=& (J,I\circ K)_2\,,\\
\text{and}\quad (I\circ J)\circ K &=& 0\,,\label{eqn:circfbna}
\end{aligned}$$ for all covectors $I,J,K$ on $M$. Here $(\cdot,\cdot)_2$ is the skew-symmetric bilinear form on $T^*M$ induced by $\omega_2^{ij}$, i.e. $(I,J)_2=I_rJ_s\omega_2^{rs}$. The compatibility also implies $$\nabla^2_l\Delta^{ij}_k = \nabla^2_k\Delta^{ij}_l\,.$$
Because of Theorem \[thm:fdbh\], we phrase the compatibility of $P_1$ and $P_2$ in terms of Fedosov structures on $M$, and break the above theorem into stages:
\[def:comfed\] Two flat Fedosov structures $(\omega_1,\nabla^1)$ and $(\omega_2,\nabla^2)$, where $\nabla^1$ and $\nabla^2$ have contravariant Christoffel symbols ${\Gamma_1}^{ij}_k$ and ${\Gamma_2}^{ij}_k$ respectively, are said to be
1. [*almost compatible*]{} if and only if $(\omega_\lambda,\nabla^\lambda)$ is a Fedosov structure for all $\lambda$, where the connection $\nabla^\lambda$ is given by ${\Gamma_\lambda}^{ij}_k={\Gamma_1}^{ij}_k+\lambda{\Gamma_2}^{ij}_k$.
2. [*almost compatible and flat*]{} if and only if they are almost compatible, and in addition the curvature of $\nabla^\lambda$ vanishes for all $\lambda$.
3. [*compatible*]{} if and only if they are almost compatible and flat, and ${c_\lambda}^{ij}_{kl}={\Gamma_\lambda}^{ij}_{(k,l)}
-\omega_{\lambda pr}{\Gamma_\lambda}^{ri}_{(k}
{\Gamma_\lambda}^{pj}_{l)}$ satisfies ${c_\lambda}^{ij}_{kl}={c_1}^{ij}_{kl}+\lambda{c_2}^{ij}_{kl}$ for all $\lambda$.
The compatibility of two flat Fedosov structures on $M$ is equivalent to the compatibility of the associated Poisson brackets on $L(M)$.
We now turn to the two Fedosov strucutres defined by $P_1$ and $P_2$, and to the pair $(\omega_\lambda,\nabla^\lambda)$ defined by $P_\lambda$. From the linearity of Result \[res:compat\] in the contravariant symbols it can be seen that $\omega_\lambda$ is automatically $\nabla^\lambda$-constant, so the almost compatibility of $(\omega_1,\nabla^1)$ and $(\omega_2,\nabla^2)$ is equivalent to $\nabla^\lambda$ being torsion free, i.e. to $$\omega_\lambda^{ir}{\Gamma_\lambda}^{jk}_l =
\omega_\lambda^{jr}{\Gamma_\lambda}^{ik}_l\,.$$ In flat coordinates for $\nabla^2$, this condition reduces to $$\label{eqn:ptf}
\omega_2^{ir}{\Gamma_1}^{jk}_r
=\omega_2^{jr}{\Gamma_1}^{ik}_r\,.$$ Note that we already have $$\label{eqn:1tf}
\omega_1^{ir}{\Gamma_1}^{jk}_r
=\omega_1^{jr}{\Gamma_1}^{ik}_r\,.$$
\[lem:fedacf\] If $(\omega_1,\nabla^1)$ and $(\omega_2,\nabla^2)$ are almost compatible, then the flatness of $\nabla^\lambda$ is equivalent to either, and hence both, of $$\begin{aligned}
\partial_l{\Gamma_1}^{jk}_s-\partial_s{\Gamma_1}^{jk}_l &=& 0
\label{eqn:pzclin}\\
\text{and}\quad
{\Gamma_1}^{ij}_{r}{\Gamma_1}^{rk}_{l}+{\Gamma_1}^{ik}_{r}{\Gamma_1}^{rj}_{l}
&=& 0\label{eqn:pzcquad}
\end{aligned}$$ in the flat coordinates for $\nabla^2$.
The contravariant curvature of $\Gamma_\lambda$ is $$\begin{aligned}
{R_\lambda}^{ijk}_{l} &=& \omega_\lambda^{ir}\left(\partial_{l}{\Gamma_\lambda}^{jk}_{r}-\partial_{s}{\Gamma_\lambda}^{jk}_{l}\right)+{\Gamma_\lambda}^{ij}_{r}{\Gamma_\lambda}^{rk}_{l}+{\Gamma_\lambda}^{ik}_{r}{\Gamma_\lambda}^{rj}_{l}\\
&=&{R_1}^{ijk}_l\\
&&+\lambda\left\{ \omega_2^{is}\left(\partial_l{\Gamma_1}^{jk}_s-\partial_s{\Gamma_1}^{jk}_l\right)
+\omega_1^{is}\left(\partial_l{\Gamma_2}^{jk}_s-\partial_s{\Gamma_2}^{jk}_l\right)\phantom{\frac{1}{2}}\right.\\
&&\left.\phantom{+\lambda\frac{1}{2}}+{\Gamma_2}^{ij}_r{\Gamma_1}^{rk}_l
+{\Gamma_1}^{ij}_r{\Gamma_2}^{rk}_l
+{\Gamma_1}^{ik}_r{\Gamma_2}^{rj}_l
+{\Gamma_2}^{ik}_r{\Gamma_1}^{rj}_l \right\}\\
&&+\lambda^2{R_2}^{ijk}_l\,,
\end{aligned}$$ which in flat coordinates for ${\Gamma_2}^{ij}_k$ reads $$\begin{aligned}
{R_\lambda}^{ijk}_{l} &=&
\omega_1^{ir}\left(\partial_l{\Gamma_1}^{jk}_r-\partial_r{\Gamma_1}^{jk}_l\right)
+{\Gamma_1}^{ij}_r{\Gamma_1}^{rk}_l
+{\Gamma_1}^{ik}_r{\Gamma_1}^{rj}_l\\
&&
+\lambda\omega_2^{is}\left(\partial_l{\Gamma_1}^{jk}_s-\partial_s{\Gamma_1}^{jk}_l\right)\,.
\end{aligned}$$ The vanishing of the order $\lambda$ term is equivalent to equation (\[eqn:pzclin\]), and with this the vanishing of the $\lambda$-independent term is equivalent to (\[eqn:pzcquad\]).
\[lem:c4pencil\] If $(\omega_1,\nabla^1)$ and $(\omega_2,\nabla^2)$ are almost compatible then the condition ${c_\lambda}^{ij}_{kl}={\Gamma_\lambda}^{ij}_{(k,l)}
-\omega_{\lambda pr}{\Gamma_\lambda}^{ri}_{(k}
{\Gamma_\lambda}^{pj}_{l)}$ reads, in the flat coordinates for $\nabla^2$, $$\label{eqn:pc4t}
{\Gamma_1}^{ij}_{r}{\Gamma_1}^{rk}_{l}-{\Gamma_1}^{ik}_{r}{\Gamma_1}^{rj}_{l}=0\,.$$
For an arbitrary Fedosov structure $(\omega,\nabla)$ the object $c^{ij}_{kl}=\Gamma^{ij}_{(k,l)}-\omega_{pr}\Gamma^{ri}_{(k}\Gamma^{pj}_{l)}$ can be converted into a quadratic expression in contravariant quantities as $$\label{eqn:c4c}
\omega^{sk}c^{ij}_{kl}=
\omega^{sk}\Gamma^{ij}_{(k,l)}-\frac{1}{2}\Gamma^{si}_p\Gamma^{pj}_l
+\frac{1}{2}\Gamma^{pi}_l\Gamma^{sj}_p\,.$$ This has similarities to the formula for covariant curvature obtained in Result \[res:curv\]; only certain signs have changed. Indeed, if we define a quantity $c^j_{rkl}$ by $$c^j_{rkl}dx^r =\frac{1}{2}\left(\nabla_{\partial_k}\nabla_{\partial_l}+\nabla_{\partial_l}\nabla_{\partial_k}\right)dx^j\,,$$ then $c^{ij}_{kl}=\omega^{ir}c^j_{rkl}$.
We have two ways of expanding $\omega_\lambda^{sk}{c_\lambda}^{ij}_{kl}$, corresponding to whether we choose first to substitute it into equation (\[eqn:c4c\]), or to expand the pencil quantities. We work in flat coordinates for $\nabla^2$; in these, ${c_2}^{ij}_{kl}$ also vanishes. First expanding the pencil we have $$\begin{aligned}
\omega_\lambda^{sk}{c_\lambda}^{ij}_{kl} &=&
\left(\omega_1^{sk}+\lambda\omega_2^{sk}\right){c_1}^{ij}_{kl}\,,\\
&=& \omega_1^{sk}{c_1}^{ij}_{kl}
+\lambda\omega_2^{sk}{c_1}^{ij}_{kl}\,,
\end{aligned}$$ whilst (\[eqn:c4c\]) gives $$\begin{aligned}
\omega_\lambda^{sk}{c_\lambda}^{ij}_{kl} &=& \omega_\lambda^{sk}{\Gamma_\lambda}^{ij}_{(k,l)}
-\frac{1}{2}{\Gamma_\lambda}^{si}_p{\Gamma_\lambda}^{pj}_l
+\frac{1}{2}{\Gamma_\lambda}^{pi}_l{\Gamma_\lambda}^{sj}_p\,,\\
&=&
\left(\omega_1^{sk}+\lambda\omega_2^{sk}\right){\Gamma_1}^{ij}_{(k,l)}
-\frac{1}{2}{\Gamma_1}^{si}_p{\Gamma_1}^{pj}_l
+\frac{1}{2}{\Gamma_1}^{pi}_l{\Gamma_1}^{sj}_p\,.
\end{aligned}$$
The order $1$ terms merely express equation (\[eqn:c4c\]) for $P_1$. Equality of the order $\lambda$ terms is equivalent to ${\Gamma_1}^{ij}_{(k,l)} = {c_1}^{ij}_{kl}$ and so to $$\begin{aligned}
\omega_1^{sk}{\Gamma_1}^{ij}_{(k,l)} &=&
\omega_1^{sk}{c_1}^{ij}_{kl}\,,\\
&=& \omega_1^{sk}{\Gamma_1}^{ij}_{(k,l)}-\frac{1}{2}{\Gamma_1}^{si}_p{\Gamma_1}^{pj}_l
+\frac{1}{2}{\Gamma_1}^{pi}_l{\Gamma_1}^{sj}_p\,.
\end{aligned}$$
Using equation (\[eqn:ptf\]) in Definition \[def:mult\] it can be seen that in the flat coordinates for $\nabla^2$ we have $\Delta^{ij}_k={\Gamma_1}^{ij}_k$. Thus we may regard equations (\[eqn:ptf\]),(\[eqn:pzclin\]),(\[eqn:pzcquad\]) and (\[eqn:pc4t\]) as identities on $\Delta^{ij}_k$; the result is Theorem \[thm:circmult\].
The condition imposed by equation (\[eqn:pzcquad\]) for an almost compatible and flat pair of Fedosov structures on the mutliplication $\circ$ is $(I\circ J)\circ K=-(I\circ K)\circ J$, i.e. the first condition (\[eqn:circfna\]) satisfied by the multiplication of a Fermionic Novikov algebra. In general (\[eqn:vindberg\]) is not satisfied even for compatible Fedosov structures, however we do have, for two flat Fedosov structures, $(\omega_1,\nabla^1)$, $(\omega_2,\nabla^2)$, which are almost compatible, $$\begin{aligned}
\lefteqn{\omega_1^{ir}\nabla^2_r\Delta^{jk}_l
-\omega_1^{jr}\nabla^2_r\Delta^{ik}_l}\\
&\qquad=& \Delta^{ij}_r\Delta^{rk}_l
-\Delta^{ir}_l\Delta^{jk}_r -\Delta^{ji}_r\Delta^{rk}_l
+\Delta^{jr}_k\Delta^{ik}_r\,.
\end{aligned}$$ So, in particular, if $\Delta^{ij}_k$ is constant in the flat coordinates for $\nabla^2$, almost compatible and flat Fedosov structures will define a Fermionic Novikov algebra structure on the covectors of $M$.
In [@baimenghe02] it emerged that examples of such algebras which do not also satisfy the [Bosonic]{} relation $(I\circ
J)\circ K=(I\circ K)\circ J$, and hence $(I\circ J)\circ K=0$, are relatively rare. $\nabla^2$-constant multiplications arising from pairs of Fedosov structures which are almost compatible and flat, but not compatible, such as that given in Example \[ex:acflat\] below, are in this class.
The pencil in flat coordinates {#subsec:flatcoord}
------------------------------
We now turn our consideration to the form the pencil takes in the flat coordinates for $\nabla^2$. From the elements of the proof of Theorem \[thm:circmult\] we have $$\label{eqn:pbfc}
P_\lambda^{ij}=
\left(\omega_1^{ij}+\lambda\omega_2^{ij}\right)\left(\frac{d}{dx}\right)^2
+2{\Gamma_1}^{ij}_ku^k_x\frac{d}{dx}
+{\Gamma_1}^{ij}_{k,l}u^k_xu^l_x +{\Gamma_1}^{ij}_ku^k_{xx}\,.$$ The Jacobi identity for $P_\lambda$ (without assuming $P_1$ and $P_2$ are Hamiltonian themselves) is equivalent to the constraints
1. \[eqn:pbfc:w2const\] $\omega_2^{ij}$ is constant and antisymmetric,
2. \[eqn:pbfc:w1anti\] $\omega_1^{ij}$ is antisymmetric,
3. \[eqn:pbfc:g1tf\] $\omega_1^{ir}{\Gamma_1}^{jk}_r=\omega_1^{jr}{\Gamma_1}^{ik}_r$,
4. \[eqn:pbfc:compat\] ${\omega^{ij}_1}_{,k}={\Gamma_1}^{ij}_k-{\Gamma_1}^{ji}_k$,
5. \[eqn:pbfc:ptf\] $\omega_2^{ir}{\Gamma_1}^{jk}_r=\omega_2^{jr}{\Gamma_1}^{ik}_r$,
6. \[eqn:pbfc:g1lin\] ${\Gamma_1}^{ij}_{k,l}={\Gamma_1}^{ij}_{l,k}$
7. \[eqn:pbfc:g1quad\] ${\Gamma_1}^{ij}_r{\Gamma_1}^{rk}_l=0$.
\[prop:bvector\] In a fixed coordinate system $\{u^i\}$ (the flat coordinates for $\Gamma_2$), given a constant non-degenerate 2-form $\omega_2^{ij}$ and a vector field $B=B^r\partial_r$ satisfying $$\label{eqn:bg1tf}
\left(\omega_2^{is}B^r_{,s}-\omega_2^{rs}B^i_{,s}\right)\omega_2^{jp}B^k_{,pr}
=\left(\omega_2^{js}B^r_{,s}-\omega_2^{rs}B^j_{,s}\right)\omega_2^{ip}B^k_{,pr}$$ and $$\label{eqn:bzcquad}
B^j_{,ir}\omega_2^{rs}B^k_{,sl}=0$$ then the prescription $$\begin{aligned}
\omega_1^{ij} &=& -(\mathcal{L}_B\omega_2)^{ij}\, =
\,\omega_2^{ir}B^j_{,r}-\omega_2^{jr}B^i_{,r} \,,\\
{\Gamma_1}^{ij}_k &=& \omega_2^{ir}B^j_{,rk}
\end{aligned}$$ satisfies the constraints [*\[eqn:pbfc:w2const\]-\[eqn:pbfc:g1quad\]*]{}. Further, all solutions of [*\[eqn:pbfc:w2const\]-\[eqn:pbfc:g1quad\]*]{} have this form.
Equations (\[eqn:bg1tf\]) and (\[eqn:bzcquad\]) are the quadratic constraints, $\omega_1^{ir}{\Gamma_1}^{jk}_r=\omega_1^{jr}{\Gamma_1}^{ik}_r$ and ${\Gamma_1}^{ij}_r{\Gamma_1}^{rk}_l=0$ respectively. That $\omega_1$ and ${\Gamma_1}$ satisfy the (linear) constraints \[eqn:pbfc:compat\], \[eqn:pbfc:ptf\] and \[eqn:pbfc:g1lin\] is an immediate consequence of their definition.
Using the Poincare lemma together with the symmetries expressed in conditions \[eqn:pbfc:g1lin\] and \[eqn:pbfc:ptf\], we have the existence of a vector field satisfying ${\Gamma_1}^{ij}_k=\omega_2^{ir}A^j,_{rk}\,$. With this condition \[eqn:pbfc:compat\] gives $\omega_1^{ij}=-(\mathcal{L}_A\omega_2)^{ij}+c^{ij}$, where $c^{ij}$ is a constant antisymmetric matrix. We may now introduce a vector field $B$ with $B^i=A^i+\frac{1}{2}x^sw_{2sr}c^{ri}$ which satisfies $\omega_1^{ij}=-\mathcal{L}_B\omega_2^{ij}$ and ${\Gamma_1}^{ij}_k=\omega_2^{ir}B^j_{,rk}$.
Since $\omega_2$ is a symplectic form, its symmetries are precisely (locally) Hamiltonian vector fields. Therefore, if $\omega_2$ and $\omega_1$ are given, the requirement that $\omega_1^{ij}=-\mathcal{L}_B\omega_2^{ij}$ fixes the non-Hamiltonian part of $B$. Then the condition ${\Gamma_1}^{ij}_k=\omega_2^{ir}B^j_{,rk}$ fixes the Hamiltonian to within a quadratic function. From the point of view of the multiplication of covectors from Section \[subsec:multiplication\], the Hamiltonian affects only the commutative part of $\circ$, thus the anti-commutative part is fixed by $\omega_1^{ij}$ and $\omega_2^{ij}$.
With consideration of the transformation rules (\[eqn:trrules\]), one can phrase Proposition \[prop:bvector\] as the existence of a vector field $B$ such that $$\begin{aligned}
\omega^{ij}_1 &=& -\mathcal{L}_B\omega^{ij}_2\,,\nonumber\\
{\Gamma_1}^{ij}_k &=&
-\mathcal{L}_B{\Gamma_2}^{ij}_k\,.\label{eqn:liederiv1}
\end{aligned}$$ We can also calculate from (\[eqn:trrules\]) the correct interpretation of the Lie derivative for an object of type $c^{ij}_{kl}$, namely: $$\begin{aligned}
\mathcal{L}_Xc^{ij}_{kl} &=& X^rc^{ij}_{kl,r} -X^i_{,r}c^{rj}_{kl}
-X^j_{,r}c^{ir}_{kl} +X^r_{,k}c^{ij}_{rl} +X^r_{,l}c^{ij}_{kr}\\
&&+X^r_{,kl}c^{ij}_r -\frac{1}{2}X^j_{rl}b^{ir}_k
-\frac{1}{2}X^j_{,rk}b^{ir}_l -X^j_{,rkl}a^{ir}\,.
\end{aligned}$$
If we work in the flat coordinates for ${\Gamma_2}$, so that the components ${c_2}^{ij}_{kl}=0$, we have for our pencil $$\begin{aligned}
-\mathcal{L}_B{c_2}^{ij}_{kl} &=& +\omega_2^{ir}B^j_{,rkl}\,,\\
&=& (\omega_2^{ir}B^j_{,rk})_{,l}\,,\\
&=& {\Gamma_1}^{ij}_{k,l}\,.
\end{aligned}$$ Now, in the flat coordinates for $\nabla^2$ we have the relation ${c_1}^{ij}_{kl}={\Gamma_1}^{ij}_{k,l}$. The linearity of the transformation rules shows that the Lie derivative of ${c_2}^{ij}_{kl}$ should be an object of the same type as ${c_1}^{ij}_{kl}$. Thus we have, in addition to (\[eqn:liederiv1\]), $${c_1}^{ij}_{kl}=-\mathcal{L}_B{c_2}^{ij}_{kl}\,.$$
One may understand these three infinitesimal relations between the coefficients of $P_1$ and $P_2$ as averring the existence on $L(M)$ of an evolutionary vector field $$\hat{B}=B^i(u(x)){\frac{\partial~}{\partial u^i(x)}}+\dots$$ such that $$P_1^{ij}=-\mathcal{L}_{\hat{B}}P_2^{ij}\,.$$
We now turn our attention to some examples of pairs of Fedosov structures, using the framework of Proposition \[prop:bvector\].
\[ex:2dma\][Two-dimensional pencils]{}. Without loss of generality we take $$\omega_2={\frac{\partial~}{\partial u^1}}\wedge{\frac{\partial~}{\partial u^2}}\,,$$ where $u^1$ and $u^2$ are a flat coordinate system for $\nabla^2$.
We take $$B=f(u^1,u^2){\frac{\partial~}{\partial u^1}}+g(u^1,u^2){\frac{\partial~}{\partial u^2}}$$ and from it calculate $\omega_1$ and $\Gamma_1$ according to (\[eqn:liederiv1\]). In particular $$\omega_1=(f_{,1}+g_{,2})\omega_2\,,$$ from which it follows immediately that $(\omega_1,\nabla^1)$ and $(\omega_2,\nabla^2)$ are almost compatible.
They are almost compatible and flat if and only if $h=f+\lambda g$ satisfies the homogeneous Monge-Ampere Equation $h_{12}^2-h_{11}h_{22}=0$ for all $\lambda$.
They are compatible if and only if $a=f+\lambda g$ and $b=f+\mu g$ satisfy $$a_{12}b_{12}-a_{11}b_{22}=0$$ for all $\lambda$, $\mu$.
For instance, one may recover the three two-dimensional Fermionic Novikov algebras of [@baimenghe02] as constant multiplications via
1. \[T1\] $f=u^1$, $g=0$,
2. \[T2\] $f=u^1$, $g=(u^1)^2$,
3. $f=(u^1)^2$, $g=0$.
[Commutative algebras.]{} In the case in which $\omega_1$ is constant in the flat coordinates for $\nabla^2$, we have, by condition \[eqn:pbfc:compat\], $${\Gamma_1}^{ij}_k={\Gamma_1}^{ji}_k\,,$$ so that the multiplication $\circ$ is commutative.
In particular if $$\omega_1=\omega_2=\omega=\sum_{i=1}^{n}{\frac{\partial~}{\partial q^i}}\wedge{\frac{\partial~}{\partial p_i}}\,,$$ then the non-Hamiltonian part of $B$ is $$\sum_{i=1}^{n}q^i{\frac{\partial~}{\partial q^i}}\,.$$
To this we may add a Hamiltonian vector field, giving $$B=\sum_{i=1}^n\left(\left[q^i+\frac{\partial H}{\partial
p_i}\right]{\frac{\partial~}{\partial q^i}} -\frac{\partial H}{\partial
q^i}{\frac{\partial~}{\partial p_i}}\right)\,.$$
Since $\omega_1=\omega_2$, equation (\[eqn:bg1tf\]) is immediate. Equation (\[eqn:bzcquad\]) becomes $$H,_{ijr}\omega^{rs}H,_{skl}=0\,,$$ where the indices $i,j,k,l,r,s$ account for both $q$ and $p$ variables.
A solution to this is $H=f(x^1,x^2,\dots,x^n)$, where each $x^i$ is either $p_i$ or $q^i$; only one from each pair of conjugate variables features in $H$.
It is not hard to see that Proposition \[prop:bvector\] can be modified to describe almost compatible and flat pairs of Fedosov structures. Specifically, we replace equation (\[eqn:bzcquad\]) by the expression corresponding to ${\Gamma_1}^{ij}_r{\Gamma_1}^{rk}_l={\Gamma_1}^{ik}_r{\Gamma_1}^{rj}_l$, namely: $$\label{eqn:bacflat}
B^j_{,ir}\omega_2^{rs}B^k_{,sl} =
B^j_{,lr}\omega_2^{rs}B^k_{,si}\,.$$
\[ex:acflat\] The Fedosov structures specified by $$\begin{aligned}
\omega_2 &=&
{\frac{\partial~}{\partial q_1}}\wedge{\frac{\partial~}{\partial p_1}}+{\frac{\partial~}{\partial q_2}}\wedge{\frac{\partial~}{\partial p_2}}\,,\\
{\Gamma_2}^{ij}_k &=& 0\,,\\
B &=& \frac{3}{2}q_1^2{\frac{\partial~}{\partial q_1}} +2q_1q_2{\frac{\partial~}{\partial q_2}}
+q_1p_2{\frac{\partial~}{\partial p_2}}\,,\\
\end{aligned}$$ and $\omega_1^{ij}=-\mathcal{L}_B\omega_2^{ij}$ and ${\Gamma_1}^{ij}_k=-\mathcal{L}_B{\Gamma_2}^{ij}_k$ are almost compatible and flat, but not compatible.
The non-zero components of $\omega_1$ and $\circ$ are $$\begin{aligned}
\{q_1,p_1\}_1 = \{q_2,p_2\}_1 &=& 3q_1\,,\\
\{q_2,p_1\}_1 &=& 2q_2\,,\\
\{p_2,p_1\}_1 &=& p_2\,,
\end{aligned}$$ and $$\begin{aligned}
dq_2\circ dp_2 &=& dq_1\,,\\
dp_1\circ dq_1 &=& -3dq_1\,,\\
dp_1\circ dq_2 &=& -2dq_2\,,\\
dp_1\circ dp_2 &=& -dp_2\,,\\
dp_2\circ dq_2 &=& -2dq_1\,.
\end{aligned}$$
Thus, the products $$\begin{aligned}
(dp_1\circ dq_2)\circ dp_2 &=& -2dq_1\\
\text{and}\qquad (dp_1\circ dp_2)\circ dq_2 &=& 2dq_1
\end{aligned}$$ violate equation (\[eqn:circfbna\]) but not (\[eqn:circfna\]). Note that $\circ$ also satisfies (\[eqn:vindberg\]) and thus defines a Fermionic Novikov algebra which is not [Bosonic]{}.
$\omega N$ manifold with Potential {#subsec:wN}
----------------------------------
The tangent bundle $T^*Q$ of a manifold $Q$ is naturally equipped with a symplectic form, and thus cotangent bundles form the basic set of examples of symplectic manifolds. One may hope to find examples of finite-dimensional bi-Hamiltonian structures on cotangent bundles by exploiting the existence of additional structures on the underlying manifolds. The main object used to do this is a $(1,1)$-tensor $L^i_j$ on $Q$ whose Nijenhuis torsion is zero. Such an object was utilised by Benenti [@benenti92] to demonstrate the separability of the geodesic equations on a class of Riemannian manifolds. This result was later interpreted in [@ibortmagrimarmo] in terms of a bi-Hamiltonian structure on $T^*Q$ which was extended to a degenerate Poisson pencil on $T^*Q\times\mathbb{R}$.
To obtain Fedosov structures we require more than just a tensor $L^i_j$ on $Q$ with vanishing Nijenhuis torsion; we also need a means of specifying the connections. If $Q$ is equipped with a torsion-free connection ${\widetilde{\nabla}}$, then the Nijenhuis torsion of a $(1,1)$-tensor $L^i_j$ can be written as $$N^i_{jk}=L^s_j{\widetilde{\nabla}}_sL^i_k -L^s_k{\widetilde{\nabla}}_sL^i_j
-L^i_s{\widetilde{\nabla}}_jL^s_k +L^i_s{\widetilde{\nabla}}_kL^s_j\,.$$ If there exists a vector field, $A$, on Q such that $L^i_j={\widetilde{\nabla}}_jA^i$ then $$N^i_{jk}=({\widetilde{\nabla}}_jA^s)({\widetilde{\nabla}}_s{\widetilde{\nabla}}_kA^i)
-({\widetilde{\nabla}}_kA^s)({\widetilde{\nabla}}_s{\widetilde{\nabla}}_jA^i)
-({\widetilde{\nabla}}_sA^i)(R^s_{jkr}A^r)\,,$$ where $R^i_{jkl}$ is the curvature tensor of ${\widetilde{\nabla}}$.
So, if ${\widetilde{\nabla}}$ is flat then the vanishing of the Nijenhuis tensor of $L={\widetilde{\nabla}}A$ is equivalent to the identity $$\label{eqn:Aznt}
({\widetilde{\nabla}}_jA^s)({\widetilde{\nabla}}_s{\widetilde{\nabla}}_kA^i)
=({\widetilde{\nabla}}_kA^s)({\widetilde{\nabla}}_s{\widetilde{\nabla}}_jA^i)\,.$$
Given a manifold $Q$ endowed with a flat connection ${\widetilde{\nabla}}$ and a vector field $A$ satisfying (\[eqn:Aznt\]), the cotangent bundle $T^*Q$ is endowed with a compatible pair of Fedosov structures, $(\omega_1,\nabla^1)$ and $(\omega_2,\nabla^2)$, as follows: $\omega_2$ is the canonical Poisson bracket on $T^*Q$.
The connection $\nabla^2$ on $T^*Q$ is the horizontal lift [@yanokentarobook] of the connection ${\widetilde{\nabla}}$ on $Q$; i.e. the Christoffel symbols ${\Gamma_2}_{ij}^k$ of $\nabla^2$ are zero in the coordinates induced on $T^*Q$ by the flat coordinates for ${\widetilde{\nabla}}$.
$(\omega_1,\nabla^1)$ is calculated from $(\omega_2,\nabla^2)$ according to the prescription of Proposition \[prop:bvector\], where the vector field $B$ is the horizontal lift of $A$ to $T^*Q$.
Let $\{q^1,\dots,q^n\}$ be flat coordinates for ${\widetilde{\nabla}}$ on $Q$, and $\mathcal{C}=\{q^1,\dots,q^n,p_1,\dots,p_n\}$ be the induced coordinates on $T^*Q$. Then $$\omega_2=\sum_{r=1}^n{\frac{\partial~}{\partial q^r}}\wedge{\frac{\partial~}{\partial p_r}}$$ and $$B=\sum_{r=1}^nA^i{\frac{\partial~}{\partial q^i}}\,.$$ The space of sections of the cotangent bundle of $T^*Q$, $\Omega$, naturally splits into $\mathcal{P}={\rm span}\{dp_i\}$ and $\mathcal{Q}={\rm
span}\{dq^i\}$. For ${\Gamma_1}^{ij}_k=\omega_2^{ir}B^j_{,rk}$ to be non-zero requires $k$ to represent a variable $q^k$, and $i$ to represent a $p_i$ variable. Thus $\Omega\circ\Omega\subseteq\mathcal{Q}$ and $\mathcal{Q}\circ\Omega=\{0\}$, meaning that $(\Omega\circ\Omega)\circ\Omega=\{0\}$. So the relation (\[eqn:bzcquad\]), ${\Gamma_1}^{ij}_r{\Gamma_1}^{rk}_l=0$, is satisfied.
$\omega_1^{ij}$ has only one kind of non-zero component, $\omega^{p_iq^j}=A^j_{,i}$, so the expression $\omega_1^{ir}{\Gamma_1}^{jk}_r$ has only one non-zero case: $$\sum_{x^r\in\mathcal{C}}\omega_1^{p_ix^r}{\Gamma_1}^{p_jq^k}_{x^r}
=\sum_{r=1}^n\omega_1^{p_iq^r}{\Gamma_1}^{p_jq^k}_{q^r} =A^r,_iA^k,_{rj}\,,$$ which is seen to be symmetric in $i$ and $j$ by condition (\[eqn:Aznt\]), which in the flat coordinates $q^i$ reads $$A^s,_jA^i,_{sk}=A^s,_kA^i,_{sj}\,.$$
If the eigenvalues of $L:TQ\rightarrow TQ$ are functionally independent in some neighbourhood then they may be used as coordinates, and $L$ takes the form $$L=\sum_{i=1}^{n}u^i{\frac{\partial~}{\partial u^i}}\otimes du^i\,.$$
In this case we may set $A=\sum_{i=1}^n\frac{1}{2}(u^i)^2{\frac{\partial~}{\partial u^i}}$, and have ${\widetilde{\nabla}}$ defined by vanishing Christoffel symbols in these coordinates.
This gives, writing $v_i$ as the conjugate coordinate to $u^i$ on $T^*Q$, $$\begin{aligned}
\omega_2 &=& \sum_{i=1}^n{\frac{\partial~}{\partial u^i}}\wedge{\frac{\partial~}{\partial v_i}}\,,\\
\omega_1 &=& \sum_{i=1}^nu^i{\frac{\partial~}{\partial u^i}}\wedge{\frac{\partial~}{\partial v_i}}\,,\\
{\Gamma_2}^{ij}_{k} &=& 0\\
{\Gamma_1}^{v_iu^i}_{u^i} &=& -1\,,\\
\end{aligned}$$ and all other Christoffel symbols zero.
Bi-Hamiltonian Structures in Degrees 1 and 2 {#section:bhd12}
============================================
We now consider a pair of operators, $P_1$ and $P_2$ in which $P_1$ is a Hamiltonian operator of hydrodynamic type and $P_2$ is of second order, i.e.: $$\begin{aligned}
P_1^{ij} &=& g^{ij}(u)\frac{d}{dx}+\Gamma^{ij}_k(u)u^k_x\,,\\
P_2^{ij} &=& a^{ij}\left(\frac{d}{dx}\right)^{2}+b^{ij}_{k}u^{k}_{x}\frac{d}{dx}+c^{ij}_{kl}u^{k}_{x}u^{l}_{x}+c^{ij}_{k}u^{k}_{xx}\,,
\end{aligned}$$ where $g^{ij}$ is the inverse of a flat metric $g_{ij}$ on $M$ and $\Gamma^{ij}_k=-g^{ir}\Gamma^j_{rk}$ where the $\Gamma^k_{ij}$ are the Christoffel symbols of the Levi-Civita connection of $g$. We also assume that $P_2^{ij}$ is antisymmetric, so that $a^{ij}=-a^{ji}$, $b^{ij}_k=a^{ij}_{,k}+c^{ij}_k+c^{ji}_k$ and $c^{(ij)}_{kl}=c^{(ij)}_{(k,l)}$.
The motivation [@dubzhangnorm] for studying such pairs of operators comes not from regarding them as separate Hamiltonian operators, but from thinking of $P_2^{ij}$ as a first order (dispersive) deformation of $P_1^{ij}$ into some non-homogeneous Hamiltonian operator $P^{ij}=P_1^{ij}+\varepsilon P_2^{ij}
+O(\varepsilon^2)$. Thus, in such a pair, it is sensible to regard the geometry of $P_1^{ij}$ as being more intrinsic than any associated to $P_2^{ij}$.
We choose to work in flat coordinates for $g$ so that $g^{ij}$ is constant and $\Gamma^{ij}_k=0$. Direct calculation of the Jacobi identity for $P^{ij}$ in these coordinates yields
\[thm:d12compat\] $P_2$ is an infinitesimal deformation of $P_1$, i.e. $P^{ij}=P_1^{ij}+\varepsilon P_2^{ij}+O(\varepsilon^2)$ satisfies the Jacobi identity to order $\varepsilon$, if and only if
1. \[eqn:d12compat:1\] $g^{ir}c^{jk}_r+g^{jr}c^{ik}_r=0$,
2. \[eqn:d12compat:2\] $c^{ij}_{kl}=c^{ij}_{(k,l)}$,
3. \[eqn:d12compat:3\] $g^{ir}c^{jk}_{l,r}=g^{jr}(c^{ik}_{l,r}-c^{ik}_{r,l})$,
4. \[eqn:d12compat:4\] $g^{ir}(a^{jk}_{,r}-c^{jk}_r)+g^{jr}(a^{ki}_{,r}-c^{ki}_r)+g^{kr}(a^{ij}_{,r}-c^{ij}_r)=0$
in the flat coordinates for $g^{ij}$.
By introducing the tensor $T^{ij}_k=a^{ir}\Gamma^j_{rk}+c^{ij}_k$ is it easy to convert conditions \[eqn:d12compat:1\], \[eqn:d12compat:3\] and \[eqn:d12compat:4\] to arbitrary coordinates, whilst condition \[eqn:d12compat:2\] becomes $$2c^{ij}_{kl}=c^{ij}_{k,l}+c^{ij}_{l,k} -c^{ri}_k\Gamma^j_{rl}
-c^{ri}_l\Gamma^j_{rk} +T^{ij}_r\Gamma^r_{kl}
+T^{rj}_k\Gamma^i_{rl} +T^{rj}_l\Gamma^i_{rk}\,.$$
To consider a bi-Hamiltonian structure involving operators $P_1^{ij}$ and $P_2^{ij}$ one need only add conditions \[eqn:hamcond:cyclic\], \[eqn:hamcond:flat\] and \[eqn:hamcond:c4reln\] of Theorem \[thm:hamcond\] to Theorem \[thm:d12compat\], however, condition \[eqn:d12compat:2\] above allows \[eqn:hamcond:c4reln\] to be replaced by $c^{ij}_rc^{rk}_l=c^{ik}_rc^{rj}_l$.
As discussed in section \[sec:withlinh20\], $P_2$ with $b^{ij}_k=2c^{ij}_k$ constant and $a^{ij}$ non-degenerate is Hamiltonian if and only if $a^{ij}=A^{ij}_ku^k+A^{ij}_0$ with $A^{ij}_k=c^{ij}_k=c^{ji}_k$, $A^{ij}_0$ is constant, $c^{ij}_k$ are the structure constants of a Fermionic Novikov algebra $(\mathcal{A},\circ)$, and $A^{ij}_0$ defines a skew-symmetric bilinear form on $\mathcal{A}$ satisfying $\langle I,J\circ K\rangle=\langle J,I\circ K\rangle$.
If we ask that $P_2$ satisfies the above constancy conditions in the flat coordinates for $g^{ij}$, then, defining an inner product on $\mathcal{A}$ by $(e^i,e^j)=g^{ij}$, we have that the compatibility of $P_1$ and $P_2$ is equivalent to the additional constraints: $$\begin{aligned}
(I\circ J)\circ K &=& (I\circ K)\circ J\,,\\
(I,J\circ K) &=& -(J,I\circ K)
\end{aligned}$$ and $$\left(I,[J,K]\right)+ \left(J,[K,I]\right)
+\left(K,[I,J]\right)=0\,,$$ where $[I,J]=I\circ J-J\circ I$ is the commutator of $\circ$, which is a Lie bracket by equation (\[eqn:vindberg\]).
For example, if we take the algebra $(\mathcal{A}={\rm
span}\{e^1,e^2,e^3,e^4\},\circ)$ where the only non-zero products are $e^3\circ e^3=e^1$ and $e^4\circ e^3=e^2$ then we may take as our symplectic form and metric $$[\omega^{ij}] =
\left(\begin{array}{cccc}0&0&a&b\\0&0&b&c\\-a&-b&0&d-u^2\\-b&-c&-d+u^2&0\end{array}\right)$$ and $$[g^{ij}] =
\left(\begin{array}{cccc}0&0&0&e\\0&0&-e&0\\0&-e&f&g\\e&0&g&h\end{array}\right)\,,$$ for any choice of the constants $a, b, c, d, e, f, g, h$ such that $e\neq 0$ and $b^2\neq ac$.
This algebra, essentially $(57)_{-1}$, is the only algebra in [@baimenghe02] of dimension 2 or 4 which admits non-degenerate forms $(\cdot,\cdot)$ and $\langle\cdot,\cdot\rangle$ satisfying the above compatibility conditions with $\circ$, other than the trivial case in which all products are zero, i.e. in which the Hamiltonian operators share the same flat connection, and so are simultaneously constant.
\[prop:atensor\] If $P_2$ is an infinitesimal deformation of $P_1$ then there exists a tensor field $A^i_j$ such that $$\begin{aligned}
a^{ij} &=& g^{ir}A^j_r-g^{jr}A^i_r\,,\nonumber\\
b^{ij}_k &=&
2g^{is}A^j_{s,k}-g^{jr}A^i_{k,r}-g^{is}A^j_{k,s}\,,\nonumber\\
c^{ij}_{kl} &=& g^{is}A^j_{s,kl}-g^{is}A^j_{(k,l)s}\,,\nonumber\\
c^{ij}_k &=& g^{is}A^j_{s,k}-g^{is}A^j_{k,s}
\label{eqn:p1p2A}
\end{aligned}$$ in flat coordinates for $g^{ij}$. Further, any (1,1)-tensor field $A^i_j$ produces an infinitesimal deformation of $P_1$ by the above formulae.
Using the non-degeneracy of $g^{ij}$, we introduce objects $\theta^k_{ij}$ and $\phi_{ij}$ by $$\begin{aligned}
c^{ij}_k &=& g^{ir}\theta^j_{rk}\,,\\
a^{ij} &=& g^{ir}g^{js}\phi_{rs}\,.
\end{aligned}$$ Then condition \[eqn:d12compat:1\] of Theorem \[thm:d12compat\] is equivalent to $\theta^k_{ij}=-\theta^k_{ji}$, and so we regard $\theta^k_{ij}$ as a family of 2-forms $\theta^k$ indexed by $k$.
Condition \[eqn:d12compat:3\] is equivalent to $\theta^k_{jl,i}=\theta^k_{il,j}-\theta^k_{ij,l}$, so that $d\theta^k=0$ for each $k$. This allows us to introduce a family of 1-forms $\psi^k$ such that $$\theta^k_{ij}=(d\psi^k)_{ij}=\psi^k_{i,j}-\psi^k_{j,i}\,.$$ Each $\psi^k$ can be adjusted by the addition of the exterior derivative, $df^k$, of some function $f^k$ without affecting the value of $\theta^k_{ij}$.
Writing $\alpha_{ij}=\phi_{ij}-g_{jr}\psi^r_i+g_{jr}\psi^r_k$, we find that condition \[eqn:d12compat:4\] is equivalent to the closedness of the 2-form $\alpha_{ij}$, upon substituting $\phi_{ij}$ and $\psi^i_j$ for $a^{ij}$ and $c^{ij}_k$. Thus we may introduce a 1-form $h$ with components $h_i$ such that $\alpha_{ij}=h_{i,j}-h_{j,i}$, and so $$\phi_{ij}=g_{jr}\psi^r_i-g_{jr}\psi^r_j+h_{i,j}-h_{j,i}\,.$$
If we now let $A^i_j=\psi^i_j+(g^{ir}h_r)_{,j}$ then we have $\theta^k_{ij}=A^k_{i,j}-A^k_{j,i}$ and $\phi_{ij}=g_{jr}\psi^r_i-g_{ir}\psi^r_j$, so that the two equations $a^{ij}=g^{ir}A^j_r-g^{jr}A^i_r$ and $c^{ij}_k=g^{ir}A^j_{r,k}-g^{jr}A^j_{k,r}$ are satisfied. The remaining to equations follow easily from $c^{ij}_{kl}=c^{ij}_{k,l}$ and $b^{ij}_k=a^{ij}_k+c^{ij}_k+c^{ji}_k$.
For the converse, it is easy to check that conditions \[eqn:d12compat:1\]-\[eqn:d12compat:4\] of Theorem \[thm:d12compat\] follow from (\[eqn:p1p2A\]) for any tensor field $A^i_j$.
As with Proposition \[prop:bvector\], Proposition \[prop:atensor\] may be understood as asserting the existence of an evolutionary vector field $$e=A^i_j\left(u(x)\right)u^j_x(x){\frac{\partial~}{\partial u^i(x)}}+\dots$$ satisfying $P_2=-\mathcal{L}_eP_1$ whenever $P_2$ is an infinitesimal deformation of $P_1$. This is therefore not a surprising result; in [@getzlerdarboux] Getzler showed the triviality of infinitesimal deformations of Hydrodynamic type Poisson brackets. With this, Proposition \[prop:atensor\] can be looked upon as a proof of Theorem \[thm:d12compat\].
There is a freedom in $A^i_j$ of $A^i_j\mapsto
A^i_j+g^{ir}f_{,rj}$ for some function $f$, which does not affect the coefficients of $P_2$. This corresponds to adjusting $e$ by a Hamiltonian vector field, $e\mapsto e+P_1(\delta f)$.
If, with reference to Lemma \[lem:3equiv\], we impose the additional constraint on (\[eqn:p1p2A\]) that $b^{ij}_k=2c^{ij}_k$ then we have the potentiality condition $g_{jr}A^r_{k,i}=g_{ir}A^r_{k,j}$, so that there exists a 1-form $B_k$ such that $$\label{eqn:miuraAB}
A^i_j=g^{ir}B_{j,r}\,.$$ In this case $a^{ij}=g^{ir}g^{jr}(B_{r,s}-B_{s,r})=g^{ir}g^{jr}(dB)_{rs}$ and the freedom $A^i_j\mapsto A^i_j+g^{ir}f_{,rj}$ is $B\mapsto B+df$. This means that $B$ can be determined purely from $g^{ij}$ and $a^{ij}$, and thus there is no freedom in the choice of $c^{ij}_k$ and $c^{ij}_{kl}$. In fact we may write explicitly $$\label{eqn:uniinfdef}
c^{ij}_k=g^{js}g_{kr}{\frac{\partial a^{ir}}{\partial u^s}}\,,\quad
c^{ij}_{kl}=c^{ij}_{(k,l)}\,,$$ and with this, $P_2$ is an infinitesimal deformation of $P_1$ if and only if $$\label{eqn:uniinfdefcond}
g^{ir}a^{jk}_{,r}+g^{jr}a^{ki}_{,r}+g^{kr}a^{ij}_{,r}=0\,.$$
Given a flat metric $g$ and a symplectic form $\omega$, there is at most one choice of flat symplectic connection $\nabla$ such that the degree 2 Hamiltonian operator specified by $(\omega,\nabla)$ is compatible with the hydrodynamic operator specified by $g$.
Clearly, if this connection exists it is given by(\[eqn:uniinfdef\]), so this definition must be checked against Theorem \[thm:hamcond\] to verify $$P_2^{ij}=\omega^{ij}\left(\frac{d~}{dx}\right)^2
+2c^{ij}_ku_x^k\frac{d~}{dx} +c^{ij}_{kl}u_x^ku_x^l
+c^{ij}_ku_{xx}^k$$ is Hamiltonian. Since equation (\[eqn:uniinfdefcond\]) is a consequence of the antisymmetry of $P_2$, compatibility with the Hydrodynamic operator follows immediately.
We conclude this section with an example of this type.
The Kaup-Broer system [@Oevellaxpoisson], $$\left(\begin{array}{c}u^1_t\\u^2_t\end{array}\right) =
\left(\begin{array}{c}u^1_{xx}+2u^2_x+2u^1u^1_x\\-u^2_{xx}+2(u^1u^2)_x\end{array}\right)\,,$$ is described by the pair of compatible Hamiltonian operators $$\begin{aligned}
P_1 &=&
\left(\begin{array}{cc}0&1\\1&0\end{array}\right)\frac{d}{dx}\,,\\
P_2 &=&
\left(\begin{array}{cc}0&1\\-1&0\end{array}\right)\left(\frac{d}{dx}\right)^2
+\left(\begin{array}{cc}2&u^1\\u^1&2u^2\end{array}\right)\frac{d}{dx}
+\left(\begin{array}{cc}0&u^1_x\\0&u^2_x\end{array}\right)\,.
\end{aligned}$$
Scaling $x\mapsto\varepsilon x$, $t\mapsto\varepsilon t$ splits $P_2$ into $P_2^{(1)}+\varepsilon P_2^{(2)}$ where $$\begin{aligned}
P_2^{(1)} &=&
\left(\begin{array}{cc}2&u^1\\u^1&2u^2\end{array}\right)\frac{d}{dx}
+\left(\begin{array}{cc}0&u^1_x\\0&u^2_x\end{array}\right)\,,\\
P_2^{(2)} &=& \left(\begin{array}{cc}0&1\\-1&0\end{array}\right)
\left(\frac{d}{dx}\right)^2\,.
\end{aligned}$$ Since $P_2=P_2^{(1)}+\varepsilon P_2^{(2)}$ is Hamiltonian for all $\varepsilon$, $P_2^{(1)}$ and $P_2^{(2)}$ constitute a bi-Hamiltonian structure of the type considered above. A set of flat coordinates for the metric in $P_2^{(1)}$ is $$\begin{aligned}
{\tilde u}^1 &=& u^1 \,,\\
{\tilde u}^2 &=& \sqrt{4u^2-(u^1)^2}\,,
\end{aligned}$$ in which $$\begin{aligned}
{\tilde P}_2^{(1)} &=&
\left(\begin{array}{cc}2&0\\0&2\end{array}\right)\frac{d}{dx}\,,\\
{\tilde P}_2^{(2)} &=& \frac{2}{{\tilde
u}^2}\left(\begin{array}{cc}0&1\\-1&0\end{array}\right)\left(\frac{d}{dx}\right)^2
+\frac{4}{({\tilde u}^2)^2}
\left(\begin{array}{cc}0&-{\tilde u}^2_x\\0&{\tilde
u}^1_x\end{array}\right)\frac{d}{dx}\\
&& +\frac{4}{({\tilde u}^2)^3}
\left(\begin{array}{cc}0&({\tilde u}^2_x)^2\\0&-{\tilde u}^1_x{\tilde u}^2_x\end{array}\right)
+\frac{2}{({\tilde u}^2)^2} \left(\begin{array}{cc}0&-{\tilde u}^2_{xx}\\0&{\tilde
u}^1_{xx}\end{array}\right)\,.
\end{aligned}$$
So in this situation we have, for the 1-form in (\[eqn:miuraAB\]), $$B=\frac{{\tilde u}^1}{2{\tilde u}^2}d{\tilde u}^2\,.$$
Conclusions
===========
In section \[section:pencils\] an approach was taken based upon the methods of [@dubflatpencil] to study compatible pairs of Hamiltonian operators of degree 2 which satisfy the conditions of the relevant Darboux theorem, Theorem \[thm:H2ODarboux\]. As for Hydrodynamic Poisson pencils, the compatibility could be reduced to algebraic constraints on a multiplication of covectors. Driving this was the ability to reduce a given Hamiltonian operator on $L(M)$ to a flat Fedosov structure $(\omega,\nabla)$ on $M$, which are natural symplectic analogues of the pair consisting of a flat metric and its Levi-Civita connection which determines a Hydrodynamic Poisson bracket.
To extend such a results to pairs of arbitrary degree 2 Hamiltonian operators, one must consider the pair $(a,\nabla)$ of Theorem \[thm:adela\]. The condition (\[eqn:bivectcompat\]), whilst atypical, expresses a familiar concept; in almost-symplectic geometry, it is common to consider connections such that the covariant derivative of the almost-symplectic form is zero, but which have torsion; if the torsion of such a connection is skew-symmetric then its symmetric part satisfies (\[eqn:bivectcompat\]). Equation (\[eqn:covbfromc\]) provides the means of going from the symmetric connection to the compatible connection with skew-torsion. The only formula missing above necessary to the study of arbitrary bi-Hamiltonian structures of degree 2 is an expression for the contravariant curvature of the connection defined by $c^{ij}_k$, which is, in the presence of Theorem \[thm:hamcond\]’s condition \[eqn:hamcond:deriv\], $$R^{ijk}_l=a^{ir}(c^{jk}_{r,l}-c^{jk}_{l,r}) +c^{ij}_rc^{rk}_l
+c^{ik}_rc^{rj}_l -(b^{ij}_r-2c^{ij}_r)c^{rk}_l
+c^{ik}_r(b^{rj}_l-2c^{rj}_l)\,.$$ One may use \[eqn:hamcond:deriv\] to replace the components of $b^{ij}_k$ in this expression with those of $c^{ij}_k$ and the derivatives of $a^{ij}$. However, one sees that the compatibility conditions do not naturally become algebraic constraints on $\Delta^{ij}_k$, and the relevancy of such an approach is undermined. It is interesting to note, however, that equation (\[eqn:c4c\]) still holds (with $\Gamma^{ij}_k=c^{ij}_k$), so that $\circ$ defined by $\Delta^{ij}_k$ still satisfies $(I\circ J)\circ K=(I\circ
K)\circ J$, and that it is the [Fermionic]{} condition $(I\circ J)\circ K=-(I\circ K)\circ J$ which is altered.
The proof of Proposition \[prop:bvector\] is easily adapted to confirm the existence of a vector field $B$ realising $P_1=-\mathcal{L}_BP_2$ whenever $P_1$, of the form (\[def:h20\]) is an infinitesimal deformation of $P_2$ as a Hamiltonian operator, provided ${b_1}^{ij}_k=2{c_1}^{ij}_k$. A simple calculation of $\mathcal{L}_BP_2$ for arbitrary $B$ shows that ${b_1}^{ij}_k=2{c_1}^{ij}_k$ is also a necessary condition. Thus we have determined the trivial deformations of a degree 2 Hamiltonian operator admitting a constant form, which are themselves of degree 2. Clearly a different approach is necessary to understand deformations of higher degrees. For the case of operators not satisfying the constraints of Theorem \[thm:H2ODarboux\], it is not immediately obvious what conditions, if any, will guarantee the triviality of a deformation; owing to the different form the contravariant curvature tensor takes, the condition ${c_1}^{ij}_{k,l}={c_1}^{ij}_{l,k}$ is absent. Owing to the lack of a constant form, the methods of [@dubzhangnorm] in ascertaining the triviality of higher degree deformations, if applicable, will be somewhat more complicated.
Finally, there is a certain artificiality to the examples of compatible Fedosov structures presented in section \[section:pencils\]. Given Theorem \[thm:fdbh\]’s assertion that underlying a pair of compatible Fedosov structures is a finite-dimensional bi-Hamiltonian structure, the question is raised asking which finite-dimensional bi-Hamiltonian structures admit symplectic connections forming almost compatible, almost compatible and flat, or compatible Fedosov structures? It would be interesting to exhibit a pair of compatible Fedosov structures in which the flat coordinates for one of the connections are in some sense physical.
Acknowledgements {#acknowledgements .unnumbered}
================
The author would like to thank Ian Strachan for suggesting this project, and the Carnegie Trust for the Universities of Scotland for the scholarship under which this work was conducted.
[10]{}
Chengming Bai, Daoji Meng, and Liguo He. On [F]{}ermionic [N]{}ovikov algebras. , 35(47):10053–10063, 2002.
S. Benenti. Inertia tensors and [S]{}täckel systems in the [E]{}uclidean spaces. , 50(4):315–341, 1992.
Pierre Bieliavsky, Michel Cahen, Simone Gutt, John Rawnsley, and Lorenz Schwachh[ö]{}fer. Symplectic connections. , 3(3):375–420, 2006.
Philip W. Doyle. Differential geometric [P]{}oisson bivectors in one space variable. , 34(4):1314–1338, 1993.
B. A. Dubrovin and S. P. Novikov. Poisson brackets of hydrodynamic type. , 279(2):294–297, 1984.
B. A. Dubrovin and S. P. Novikov. Hydrodynamics of weakly deformed soliton lattices. [D]{}ifferential geometry and [H]{}amiltonian theory. , 44(6(270)):29–98, 203, 1989.
Boris Dubrovin. Flat pencils of metrics and [F]{}robenius manifolds. In [*Integrable systems and algebraic geometry (Kobe/Kyoto, 1997)*]{}, pages 47–72. World Sci. Publ., River Edge, NJ, 1998.
Boris Dubrovin and Youjin Zhang. Normal forms of hierarchies of integrable pde’s, [F]{}robenius manifolds and [G]{}romov-[W]{}itten invariants. , 2001.
Boris V. Fedosov. A simple geometrical construction of deformation quantization. , 40(2):213–238, 1994.
I.M. Gel’fand, V. Retakh, and M. Shubin. Fedosov manifolds. , 136(1):104–140, 1998.
Ezra Getzler. A [D]{}arboux theorem for [H]{}amiltonian operators in the formal calculus of variations. , 111(3):535–560, 2002.
A. Ibort, F. Magri, and G. Marmo. Bihamiltonian structures and [S]{}täckel separability. , 33(3-4):210–228, 2000.
Franco Magri. A simple model of the integrable [H]{}amiltonian equation. , 19(5):1156–1162, 1978.
O.I. Mokhov. Symplectic and [P]{}oisson structures on loops spaces of smooth manifolds, and integrable systems. , 53(3):515–622, 1998.
S.P. Novikov. The geometry of conservative systems of hydrodynamic type. the method of averaging for field-theoretical systems. , 40(4):85–98, 1985.
W. Oevel. A note on the [P]{}oisson brackets associated with [L]{}ax operators. , 186(1-2):79–86, 1994.
G.V. Potemin. On [P]{}oisson brackets of differential geometric type. , 33(1):30–33, 1986.
X Xu. Variational calculus of supervariables and related algebraic structures. , 223(2):396–437, 2000.
Kentaro Yano and Shigeru Ishihara. . Marcel Dekker Inc., New York, 1973. Pure and Applied Mathematics, No. 16.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Using the planar Temperley-Lieb algebra, critical bond percolation on the square lattice can be reformulated as a loop model. In this form, it is incorporated as ${\cal LM}(2,3)$ in the Yang-Baxter integrable family of logarithmic minimal models ${\cal LM}(p,p')$. We consider this model of percolation in the presence of boundaries and with periodic boundary conditions. Inspired by Kuniba, Sakai and Suzuki, we rewrite the recently obtained infinite $Y$-system of functional equations. In this way, we obtain nonlinear integral equations in the form of a closed finite set of TBA equations described by a $D_3$ Dynkin diagram. Following the methods of Klümper and Pearce, we solve the TBA equations for the conformal finite-size corrections. For the ground states of the standard modules on the strip, these agree with the known central charge $c=0$ and conformal weights $\Delta_{1,s}$ for $s\in {\mathbb Z_{{\geqslant}1}}$ with $\Delta_{r,s}=\big((3r-2s)^2-1\big)/24$. For the periodic case, the finite-size corrections agree with the conformal weights $\Delta_{0,s}$, $\Delta_{1,s}$ with $s\in\frac{1}{2}{\mathbb Z_{{\geqslant}0}}$. These are obtained analytically using Rogers dilogarithm identities. We incorporate all finite excitations by formulating empirical selection rules for the patterns of zeros of all the eigenvalues of the standard modules. We thus obtain the conformal partition functions on the cylinder and the modular invariant partition function (MIPF) on the torus. By applying $q$-binomial and $q$-Narayana identities, it is shown that our refined finitized characters on the strip agree with those of Pearce, Rasmussen and Zuber. For percolation on the torus, the MIPF is a non-diagonal sesquilinear form in affine $u(1)$ characters given by the $u(1)$ partition function $Z_{2,3}(q)=Z_{2,3}^{\text{Circ}}(q)$. The $u(1)$ operator content is ${\cal N}_{\Delta,\bar\Delta}=1$ for $\Delta=\bar\Delta=-\frac{1}{24}, \frac{35}{24}$ and ${\cal N}_{\Delta,\bar\Delta}=2$ for $\Delta=\bar\Delta=\frac{1}{8}, \frac{1}{3}, \frac{5}{8}$ and $(\Delta,\bar\Delta)=(0,1), (1,0)$. This result is compatible with the general conjecture of Pearce and Rasmussen, namely $Z_{p,p'}(q)=Z^{\text{Proj}}_{p,p'}(q)+n_{p,p'} Z^{\text{Min}}_{p,p'}(q)$ with $n_{p,p'}\in {\mathbb Z}$, where the minimal partition function is $Z^{\text{Min}}_{2,3}(q)=1$ and the lattice derivation fixes $n_{2,3}=-1$.
Keywords: percolation, solvable lattice models, conformal field theory.
title: |
**Conformal partition functions of critical percolation\
from $\boldsymbol{D_3}$ Thermodynamic Bethe Ansatz equations**
---
-15mm
[Alexi Morin-Duchesne$^\ast$, Andreas Klümper$^\dagger$, Paul A. Pearce$^\ddagger$]{}\
[*$^\ast$Institut de Recherche en Mathématique et Physique\
Université catholique de Louvain, Louvain-la-Neuve, B-1348, Belgium*]{}\
[*$^\dagger$Fachbereich C Physik, Bergische Universität Wuppertal, 42097 Wuppertal, Germany*]{}\
[*$^\dagger$School of Mathematics and Statistics, University of Melbourne\
Parkville, Victoria 3010, Australia*]{}\
[[email protected]]{}
Introduction
============
In 1957, Broadbent and Hammersley [@BroadHamm57] introduced a lattice percolation model as a mathematical model of the physical process of a fluid flowing through a random medium. Most importantly, they showed that their model exhibits a phase transition characterized by a critical probability threshold $p=p_c$. Comprehensive reviews of percolation theory can be found in [@KestonPerc82; @Stauffer92; @Grimmet97; @Saberi16].
One of the challenging goals in percolation is to understand the precise thermodynamic behavior of the model in the vicinity of the critical point. This behavior is believed to be [*conformally invariant*]{} and [*universal*]{}. Invariance under conformal maps implies invariance under translation, rotation and local scaling transformations. Universality implies that the critical behavior, characterized by critical exponents, depends on the lattice dimensionality but is otherwise insensitive to the details of the lattice model (for example the lattice structure or the choice of site versus bond percolation). In this paper, we view critical bond percolation on the two-dimensional square lattice as a Yang-Baxter integrable [@BaxBook] loop model and solve exactly the associated $D_3$ $Y$-system for the conformal spectra to establish how it fits into the framework of logarithmic conformal field theory [@FMS; @Gurarie; @SpecialIssue].
In bond percolation on the square lattice, the bonds $j$ of the lattice are [*open or occupied*]{} ($\sigma_j=1$) with a probability $p$ and [*closed or empty*]{} ($\sigma_j=0$) with a probability $1-p$. A typical bond configuration is shown in \[fig:percoconf\]. In this description, the configuration space representing the local degrees of freedom is $\Omega=\{0,1\}^{{\mathbb Z}^2}$. The “spins" $\sigma_j$ are independent identically distributed random variables. As a consequence, the usual observables given by the correlations of these spins factorize and are trivial. Accordingly, the statistical weight $W(\sigma)$ of a configuration $\sigma$ is $$W(\sigma)=p^\text{\#\! bonds}(1-p)^\text{\#\! empty bonds},\qquad Z=\sum_\sigma W(\sigma)=1$$ and the partition function $Z$ is trivial.
In fact, the interesting physical behavior resides in the properties of connected clusters. The probability $P(p)$ that the origin is part of an infinite connected cluster is called the [*percolation*]{} probability. For $p<p_c$, the open bonds are sparsely distributed at random throughout the lattice with no large clusters and $P(p)=0$. For $p>p_c$, the percolation probability is strictly positive: $P(p)>0$. The percolation probability is thus the order parameter for an order-disorder phase transition. For bond percolation on the square lattice, it has been proved that $P(p_c)=0$ [@Harris] and that the critical threshold is precisely $p_c=1/2$ [@Kesten80]. More generally, the interesting physical observables [@VJS12; @DPSV13] include the probabilities $P(j_1,j_2,\ldots,j_n)$ that the bonds $j_1,j_2,\ldots,j_n$ all lie in the same connected cluster.
The behavior of connected clusters is captured by introducing degrees of freedom in the form of planar non-crossing loop segments representing non-local [*connectivities*]{}. Mathematically, the local properties of loop segments are encoded in the planar Temperley-Lieb algebra [@TL71; @Jones]. In critical percolation, the loop segments can close to form loops with an assigned statistical weight or fugacity $\beta=1$. This description of percolation is sometimes referred to as hull percolation [@HullPerc88; @DuplantierHull]. On the square lattice, there is a one-to-one mapping between bond configurations and loop configurations. This is illustrated by an example in \[fig:percoconf\]. On rectangular lattices, each connected cluster is surrounded by loop segments, and crucially for crossing probabilities [@Cardy92; @LPSA94], a connected cluster spans the lattice if and only if the surrounding loop segments also span the lattice.
The critical point $p=p_c$ of percolation marks a second order phase transition. From the viewpoint of statistical mechanics, the universality class of such a phase transition is characterized by critical exponents. The first few critical exponents considered for percolation are related to the number of clusters per site, the percolation probability, the truncated mean cluster size, the cluster volume and the correlation length respectively:
$$\begin{aligned}
{3}
\alpha&=\frac{2\Delta_t-1}{\Delta_t-1}=-\frac{2}{3},\qquad&&
\beta=\frac{\Delta_h}{1-\Delta_t}=\frac{5}{36},\qquad
&&\gamma=\frac{2\Delta_h-1}{\Delta_t-1}=\frac{43}{18},\\[0.2cm]
\delta&=\frac{1-\Delta_h}{\Delta_h}=\frac{91}{5},\qquad&&
\nu=\frac{1}{2(1-\Delta_t)}=\frac{4}{3}.\end{aligned}$$
For example, as $p\to p_c^+$, the percolation probability behaves as $P(p)\sim (p-p_c)^\beta$. Only two of these exponents are independent. The others are related by scaling relations [@CardyRGScaling] to the thermal and magnetic conformal weights $$\Delta_t=\Delta_{2,1}=\frac{5}{8},\qquad \Delta_h=\Delta_{\frac{1}{2},0}=\frac{5}{96}.$$ The values of these critical exponents were originally conjectured by den Nijs [@denNijs79] and Nienhuis, Riedel and Schick [@NienhuisRS80] based on Coulomb gas arguments [@CoulombGas1; @CoulombGas2; @FSZ87] by viewing percolation as the $Q\to 1$ limit of the critical $Q$-state Potts model. The $Q$-state Potts model with $Q=1$ is indeed trivial. It has a unique frozen state and the partition function is trivially $Z=1$, so a $Q\to 1$ limit is needed [@Delfino1; @Delfino2] to recover the critical exponents. In general, the critical line of the Coulomb gas maps onto the critical line of the six-vertex model and is parameterised by the crossing parameter $\lambda\in (0,\pi)$. It is related to the loop fugacity by $\beta=\sqrt{Q}=2\cos\lambda$, with percolation corresponding to $\lambda=\frac \pi 3$.
The statistical behavior of percolation shares many commonalities with the model of critical dense polymers, which has a loop fugacity $\beta=0$. In polymers, the non-local degrees of freedom are extended segments of polymer chains which are not allowed to form closed loops. The study of polymers and percolation as Conformal Field Theories (CFTs) began with Saleur and Duplantier [@Saleur86; @Saleur87a; @Saleur87b; @Duplantier86; @DupSaleur86; @SaleurDup87; @Saleur92] in the mid-eighties. A CFT is a continuum theory that describes directly the [*universal*]{} properties of a critical statistical system (characterized by a linear system size $N$, a lattice spacing $a$ and a continuum coordinate $R$) in the continuum scaling limit ($N\to\infty$, $a\to 0$, $Na\to R$). The conformal symmetry of percolation is described by a Virasoro algebra with central charge $c=0$. As a CFT, the Coulomb critical line is an $su(2)$ affine Wess-Zumino-Witten CFT with effective central charge $c_\text{eff}=1$. The nature of a critical point on this line is very different if $\lambda/\pi$ is rational compared to generic points where $\lambda/\pi$ is irrational. If $\frac{\lambda}{\pi}=\frac{p'-p}{p'}$ is rational, the theory admits a higher symmetry algebra described by the $sl_2$ loop algebra [@DeguchiFabMcCoy; @Deguchi]. These points are characterized by two integers $p,p'$ satisfying \[eq:ppp\] 1p<p’,(p,p’)=1 and are dense along the critical line, with each point representing a different CFT. At each of these points, there are additional eigenvalue degeneracies and the theory is logarithmic. Percolation corresponds to the point with $(p,p')=(2,3)$.
To set it in context, the loop model of critical bond percolation is the $(p,p')=(2,3)$ member of the family of logarithmic minimal models ${\cal LM}(p,p')$ [@PRZ06] with conformal data consisting of the central charge $c$ and Virasoro Kac conformal weights $\Delta_{r,s}$ \[logMin\] c= 1 - , \_[r,s]{}=. Since these loop models are defined in terms of the diagrammatic action of local operators on a vector space of link states, the logarithmic minimal models are intrinsically quantum in nature. The choice of the vector space of link states is an integral part of the definition of the model. But this space of states is not a Hilbert space since the inner product is not positive-definite. The infinitely extended Virasoro Kac table of conformal weights for percolation is shown in \[fig:VirKac\] for $r, s \in \mathbb Z_{> 0}$. In this paper, we will encounter conformal weights $\Delta_{r,s}$ with $r=0,1$ and $s \in \frac12 \mathbb Z_{>0}$. Additional physical conformal weights are given by allowing $r$ or $s$ or both in this Kac formula to be half-integers [@PRT14; @BPT2016] or even possibly to take values in $\mathbb Q$ [@CJV17]. The central charge and conformal weights, given by the Kac formula, vary continuously with the parameter $\lambda\in (0,\pi)$.
In analogy to the rational minimal models ${\cal M}(m,m')$, the logarithmic minimal models ${\cal LM}(p,p')$ are coset CFTs [@PRcosetBranch13]. This analogy is the origin of the name but, in contradistinction to the unitary minimal models ${\cal M}(m'\!-\!1,m')$, the logarithmic minimal models are all nonunitary. In particular, ${\cal LM}(2,3)$ is a nonunitary coset CFT with $$c=0,\qquad \Delta_\text{min}=-1/24,\qquad c_\text{eff}=c-24\Delta_\text{min}=1,\qquad \Delta_{r,s}=\frac{(3r-2s)^2-1}{24}.$$ Tellingly, since $\frac{\lambda}{\pi}=\tfrac{1}{3}$, critical percolation is a [*logarithmic*]{} CFT [@MathieuRidout] and not a [*rational*]{} CFT [@MooreSeiberg]. The infinitely extended Virasoro Kac table of percolation in \[fig:VirKac\] displays the conformal weights of an infinite number of Virasoro scaling operators. If a theory is rational, there can only be a finite number of conformal weights associated with a finite number of scaling operators and the associated Virasoro (or extended) representations must be irreducible and close among themselves under fusion. In contrast, logarithmic CFTs are characterized [@Gurarie; @MDSA2011; @MDSA13no2] by the existence of reducible yet indecomposable representations of the Virasoro algebra. On the strip, there is a single copy of the Virasoro algebra but, on a torus, there are two chiral copies of the Virasoro algebra and conformal invariance extends [@CardyModInv] to include invariance under the modular group. For simple rational CFTs, such as the $c<1$ $A$-$D$-$E$ models [@CIZ], conformal and modular invariance together suffice to uniquely determine the conformal torus partition function. This is not the case for general logarithmic minimal models.
Strikingly, conformality was only rigorously established in 2001 by Smirnov [@Smirnov01] for critical site percolation on the triangular lattice. This mathematical approach, which is built on random conformally invariant fractal curves, entails the identification of the models with $\mbox{SLE}_\kappa$ (Schramm-Loewner Evolution) with $\kappa = \frac{4p'}{p}$. For percolation, with $(p,p')=(2,3)$ and $\kappa=6$, the fractal dimensions $d=2(1-\Delta)$ of various fractal geometric curves are known [@Saleur87a; @SaleurDup87; @AAMRH; @JS2005; @JS2006; @SAPR2009] including those of chordal SLE paths, hulls (H), cluster mass (C), external perimeter (EP) and red bonds (RB):
$$\begin{aligned}
(\Delta_\text{path}^\text{SLE},\Delta_\text{H},\Delta_\text{C},\Delta_\text{EP},\Delta_\text{RB})
&=(\Delta_{p,p'\pm1},\Delta_{p,p'\pm 1},\Delta_{\frac12(p\pm1),\frac12p'},\Delta_{p\pm1,p'},\Delta_{p,p'\pm2})
=(\tfrac{1}{8},\tfrac{1}{8},\tfrac{5}{96},\tfrac{1}{3},\tfrac{5}{8}),\\[4pt]
&(d_\text{path}^\text{SLE},d_\text{H},d_\text{C},d_\text{EP},d_\text{RB})
=(\tfrac{7}{4},\tfrac{7}{4},\tfrac{91}{48},\tfrac{4}{3},\tfrac{3}{4}).\end{aligned}$$
The value $d_\text{EP}=\frac{4}{3}$ was conjectured by Mandelbrot [@Mandelbrot] and much later proved by Lawler, Schramm and Werner [@LSW01]. The value $d_\text{path}^\text{SLE}=\frac{7}{4}$ was proved by Beffara [@Beffara].
The incorporation of critical dense polymers ${\cal LM}(1,2)$ [@PR07; @PRVcyl2010; @MD11; @PRV1210; @MDPR13; @PRT14] and critical percolation ${\cal LM}(2,3)$ into the framework of the family of logarithmic minimal models ${\cal LM}(p,p')$ [@PRZ06; @PRcosetBranch13; @PRcosetGraphs11; @MDPR14; @PTC2015; @MDRD2015; @BPT2016] establishes that these models are Yang-Baxter integrable. The transfer matrices of the logarithmic minimal models are built from so called [*transfer tangles*]{} of the planar Temperley-Lieb algebra [@TL71; @Jones], which we respectively denote by $\Db(u)$ and $\Tb(u)$ for the boundary and the periodic cases. The finite-size corrections to the eigenvalues $D(u)$ and $T(u)$ of the transfer matrices provide a direct way to access the central charge and conformal weights analytically. Indeed, for large horizontal system size $N$, the leading eigenvalues of the transfer matrices behave as
$$\begin{aligned}
{2}
\hspace{-0.2cm}-\ln D(u) &= 2N f_{\text{b}}(u) + f_{\text{s}}(u) + \tfrac{2 \pi}{N} \sin(\tfrac{\pi u}{\lambda}) \big(\!-\tfrac c {24} + \Delta + k \big) + o(\tfrac1N)\label{eq:Dexpansion},\\
\hspace{-0.2cm}-\ln T(u) &= N f_{\text{b}}(u) + \tfrac{2 \pi }{N} \Big(\sin(\tfrac{\pi u}{\lambda})(\Delta + \bar \Delta + k + \bar k- \tfrac c {12}) + \ir \cos(\tfrac{\pi u}{\lambda}) (\Delta - \bar \Delta + k - \bar k) \Big) + o(\tfrac1N),
\label{eq:Texpansion}\end{aligned}$$
where $f_{\text{b}}(u)$ and $f_{\text{s}}(u)$ are the [*non-universal*]{} bulk and surface free energies, and $k,\bar k$ are integers. Yang-Baxter integrability on the lattice means that $f_{\text{b}}(u)$ and $f_{\text{s}}(u)$ can be calculated exactly. In addition, by solving $T$- and $Y$-systems [@BR1989; @Zam1991a; @Zam1991b; @PK91; @KP92; @KunibaNS9309; @KunibaNS9310; @KunibaNS1010] satisfied by the commuting transfer matrices, one can calculate analytically the $\frac1N$ term to obtain [*universal*]{} quantities such as the central charge, conformal weights and conformal partition functions. The $T$-system takes [@KLWZ1997] the form of a bilinear Hirota equation and is the master equation of integrability. Two key steps [@PK91; @KP92] in the process of solving the system are, first, to derive the $Y$-system from the $T$-system and, second, to use analyticity properties of the eigenvalues to convert the $Y$-system of functional equations into non-linear integral equations in the form of Thermodynamic Bethe Ansatz (TBA) equations [@YangYang; @T71; @G71; @Zam90; @Zam91]. These latter works on TBA focused on the ground state. The approach of Klümper and Pearce [@PK91; @KP92], which we follow closely here, applies to all finite excitations and enables the analytic calculation of conformal partition functions. The $T$-system is non-universal but the $Y$-system, which relates to the conformal spectra, is universal [@CMP02] in the sense that it holds for all boundary conditions and topologies. So these calculations can be carried out with periodic boundary conditions or in the presence of boundaries on the strip [@Sklyanin; @BPO96]. This program has been carried to completion [@PK91; @KP92; @KP91; @BPO96; @OPW97; @BDP15] for prototypical $c<1$ $A$-type rational minimal models.
Within the lattice approach, our longer term goal is to extend these calculations, based on functional equations and TBA, to the general logarithmic minimal models ${\cal LM}(p,p')$. The $T$- and $Y$-systems for the general ${\cal LM}(p,p')$ models were obtained recently in [@MDPR14]. These hierarchies of functional equations are infinite but the $Y$-system can be truncated to a finite $D$-type $Y$-system following the methods of [@KSS98]. In this paper, we start with critical percolation ${\cal LM}(2,3)$ as a prototypical example with $p>1$. This model admits a set of TBA equations encoded by a $D_3\simeq A_3$ Dynkin diagram. Our specific goals are to calculate analytically, for critical percolation, the following quantities:
1. the central charge and conformal weights using dilogarithm identities;\
2. the finitized characters on the strip for half-arc boundary conditions and an arbitrary number of defects;\
3. the cylinder conformal partition functions with half-arc boundary conditions;\
4. the modular invariant partition function (MIPF) on the torus.
This program has been completed [@PR07; @PRVcyl2010; @MD11; @PRV1210; @MDPR13; @PRT14] for critical dense polymers ${\cal LM}(1,2)$. In this case, the task was simplified because the transfer matrices satisfy a trivial $Y$-system in the form of an inversion identity similar to that of the (free-fermionic) Ising model [@BaxBook; @OPW1996]. The analysis of critical dense polymers introduced combinatorial constructs to enumerate patterns of zeros, namely single- and double-column diagrams and $q$-Narayana polynomials. Remarkably, these reappear in generalizing the calculations to critical percolation. Similarly, because of the occurrence of non-contractible loops and winding on the cylinder, a modified trace [@MDPR13] (analogous to the Markov trace [@Jones83] on the strip) is needed to obtain the MIPF of critical percolation as was the case for critical dense polymers. We also stress that, as for critical dense polymers, the MIPF that we find for $\mathcal{LM}(2,3)$ is obtained from the scaling limit of the loop model on a torus of size $M\times N$, with $M$ and $N$ even, where each non-contractible loop is weighted by a fugacity $\alpha = 2$.
\#1 \#1
(0,-.3)(1,3) (0,0) (0,2)
(-0.5,-.3)(7.5,4) (0,0) (1,0) (2,0) (3,0) (4,0) (5,0) (6,0) (0,1) (1,1) (2,1) (3,1) (4,1) (5,1) (6,1) (0,2) (1,2) (2,2) (3,2) (4,2) (5,2) (6,2) (0,3) (1,3) (2,3) (3,3) (4,3) (5,3) (6,3) (0,1)[.5]{}[90]{}[270]{} (0,3)[.5]{}[90]{}[270]{} (7,1)[.5]{}[270]{}[90]{} (7,3)[.5]{}[270]{}[90]{} (0,0)(2,0)[3]{}[(1,0)[.5]{}[180]{}[360]{}]{} (0,0)(2,0)[3]{}[(2,4)[.5]{}[0]{}[180]{}]{} (.5,4)(.5,4.4) (6.5,-.4)(6.5,0)
$$\begin{pspicture}(0,-.3)(7,11)
\psframe[linewidth=0pt,fillstyle=solid,fillcolor=lightestblue](0,0)(7,11)
\psframe[linewidth=0pt,fillstyle=solid,fillcolor=lightlightblue](1,0)(2,11)
\psframe[linewidth=0pt,fillstyle=solid,fillcolor=lightlightblue](3,0)(4,11)
\psframe[linewidth=0pt,fillstyle=solid,fillcolor=lightlightblue](5,0)(6,11)
\psframe[linewidth=0pt,fillstyle=solid,fillcolor=lightlightblue](0,2)(7,3)
\psframe[linewidth=0pt,fillstyle=solid,fillcolor=lightlightblue](0,5)(7,6)
\psframe[linewidth=0pt,fillstyle=solid,fillcolor=lightlightblue](0,8)(7,9)
\multiput(0,0)(0,3){3}{\multiput(0,0)(2,0){3}{\psframe[linewidth=0pt,fillstyle=solid,fillcolor=midblue](1,2)(2,3)}}
\psgrid[gridlabels=0pt,subgriddiv=1]
\rput(.5,10.6){$\vdots$}\rput(1.5,10.6){$\vdots$}\rput(2.5,10.6){$\vdots$}
\rput(3.5,10.6){$\vdots$}\rput(4.5,10.6){$\vdots$}\rput(5.5,10.6){$\vdots$}
\rput(6.5,10.5){$\vvdots$}
\rput(.5,9.5){$12$}\rput(1.5,9.5){$\frac{65}8$}\rput(2.5,9.5){$5$}
\rput(3.5,9.5){$\frac{21}8$}\rput(4.5,9.5){$1$}\rput(5.5,9.5){$\frac{1}8$}\rput(6.525,9.5){$\cdots$}
\rput(.5,8.5){$\frac{28}3$}\rput(1.5,8.5){$\frac{143}{24}$}\rput(2.5,8.5){$\frac{10}3$}
\rput(3.5,8.5){$\frac{35}{24}$}\rput(4.5,8.5){$\frac 13$}\rput(5.5,8.5){$-\frac{1}{24}$}
\rput(6.525,8.5){$\cdots$}
\rput(.5,7.5){$7$}\rput(1.5,7.5){$\frac {33}8$}\rput(2.5,7.5){$2$}
\rput(3.5,7.5){$\frac{5}8$}\rput(4.5,7.5){$0$}\rput(5.5,7.5){$\frac{1}8$}
\rput(6.525,7.5){$\cdots$}
\rput(.5,6.5){$5$}\rput(1.5,6.5){$\frac {21}8$}\rput(2.5,6.5){$1$}
\rput(3.5,6.5){$\frac{1}8$}\rput(4.5,6.5){$0$}\rput(5.5,6.5){$\frac{5}8$}\rput(6.525,6.5){$\cdots$}
\rput(.5,5.5){$\frac{10}3$}\rput(1.5,5.5){$\frac {35}{24}$}\rput(2.5,5.5){$\frac 13$}
\rput(3.5,5.5){$-\frac{1}{24}$}\rput(4.5,5.5){$\frac 13$}\rput(5.5,5.5){$\frac{35}{24}$}
\rput(6.525,5.5){$\cdots$}
\rput(.5,4.5){$2$}\rput(1.5,4.5){$\frac 58$}\rput(2.5,4.5){$0$}\rput(3.5,4.5){$\frac{1}8$}
\rput(4.5,4.5){$1$}\rput(5.5,4.5){$\frac{21}8$}\rput(6.525,4.5){$\cdots$}
\rput(.5,3.5){$1$}\rput(1.5,3.5){$\frac 18$}\rput(2.5,3.5){$0$}\rput(3.5,3.5){$\frac{5}8$}
\rput(4.5,3.5){$2$}\rput(5.5,3.5){$\frac{33}8$}\rput(6.525,3.5){$\cdots$}
\rput(.5,2.5){$\frac 13$}\rput(1.5,2.5){$-\frac 1{24}$}\rput(2.5,2.5){$\frac 13$}
\rput(3.5,2.5){$\frac{35}{24}$}\rput(4.5,2.5){$\frac{10}3$}\rput(5.5,2.5){$\frac{143}{24}$}
\rput(6.525,2.5){$\cdots$}
\rput(.5,1.5){$0$}\rput(1.5,1.5){$\frac 18$}\rput(2.5,1.5){$1$}\rput(3.5,1.5){$\frac{21}8$}
\rput(4.5,1.5){$5$}\rput(5.5,1.5){$\frac{65}8$}\rput(6.525,1.5){$\cdots$}
\rput(.5,.5){$0$}\rput(1.5,.5){$\frac 58$}\rput(2.5,.5){$2$}\rput(3.5,.5){$\frac{33}8$}\rput(4.5,.5){$7$}
\rput(5.5,.5){$\frac{85}8$}\rput(6.525,.5){$\cdots$}
{\color{blue}
\rput(.5,-.5){$1$}
\rput(1.5,-.5){$2$}
\rput(2.5,-.5){$3$}
\rput(3.5,-.5){$4$}
\rput(4.5,-.5){$5$}
\rput(5.5,-.5){$6$}
\rput(6.5,-.5){$r$}
\rput(-.5,.5){$1$}
\rput(-.5,1.5){$2$}
\rput(-.5,2.5){$3$}
\rput(-.5,3.5){$4$}
\rput(-.5,4.5){$5$}
\rput(-.5,5.5){$6$}
\rput(-.5,6.5){$7$}
\rput(-.5,7.5){$8$}
\rput(-.5,8.5){$9$}
\rput(-.5,9.5){$10$}
\rput(-.5,10.5){$s$}}
\end{pspicture}
\hspace{3cm}
\begin{pspicture}(-0.5,-0.3)(5,15)
\psframe[fillstyle=solid,fillcolor=white,linewidth=0pt](0,0)(5,15)
\multiput(1,2)(0,2){7}{\psframe[fillstyle=solid,fillcolor=lightlightblue,linewidth=0pt](0,0)(1,1)}
\multiput(2,2)(0,2){7}{\psframe[fillstyle=solid,fillcolor=lightlightblue,linewidth=0pt](0,0)(1,1)}
\multiput(3,2)(0,2){7}{\psframe[fillstyle=solid,fillcolor=lightlightblue,linewidth=0pt](0,0)(1,1)}
\multiput(4,2)(0,2){7}{\psframe[fillstyle=solid,fillcolor=lightlightblue,linewidth=0pt](0,0)(1,1)}
\psgrid[gridlabels=0pt,subgriddiv=1](0,0)(5,15)
{\color{blue}
\rput(0.5,-0.5){$0$}
\rput(1.5,-0.5){$1$}
\rput(2.5,-0.5){$2$}
\rput(3.5,-0.5){$3$}
\rput(4.5,-0.5){$r$}
\rput(-0.5,0.5){$0$}
\rput(-0.5,1.5){$\frac12$}
\rput(-0.5,2.5){$1$}
\rput(-0.5,3.5){$\frac32$}
\rput(-0.5,4.5){$2$}
\rput(-0.5,5.5){$\frac52$}
\rput(-0.5,6.5){$3$}
\rput(-0.5,7.5){$\frac72$}
\rput(-0.5,8.5){$4$}
\rput(-0.5,9.5){$\frac92$}
\rput(-0.5,10.5){$5$}
\rput(-0.5,11.5){$\frac{11}2$}
\rput(-0.5,12.5){$6$}
\rput(-0.5,13.5){$\frac{13}2$}
\rput(-0.5,14.5){$s$}
}
\rput(0.5,0.5){$-\frac1{24}$}
\rput(0.5,1.5){$0$}
\rput(0.5,2.5){$\frac18$}
\rput(0.5,3.5){$\frac13$}
\rput(0.5,4.5){$\frac58$}
\rput(0.5,5.5){$1$}
\rput(0.5,6.5){$\frac{35}{24}$}
\rput(0.5,7.5){$2$}
\rput(0.5,8.5){$\frac{21}{8}$}
\rput(0.5,9.5){$\frac{10}{3}$}
\rput(0.5,10.5){$\frac{33}{8}$}
\rput(0.5,11.5){$5$}
\rput(0.5,12.5){$\frac{143}{24}$}
\rput(0.5,13.5){$7$}
\rput(0.5,14.60){$\vdots$}
\rput(1.5,0.5){$\frac13$}
\rput(1.5,1.5){$\frac18$}
\rput(1.5,2.5){$0$}
\rput(1.5,3.5){$-\frac1{24}$}
\rput(1.5,4.5){$0$}
\rput(1.5,5.5){$\frac18$}
\rput(1.5,6.5){$\frac13$}
\rput(1.5,7.5){$\frac58$}
\rput(1.5,8.5){$1$}
\rput(1.5,9.5){$\frac{35}{24}$}
\rput(1.5,10.5){$2$}
\rput(1.5,11.5){$\frac{21}{8}$}
\rput(1.5,12.5){$\frac{10}{3}$}
\rput(1.5,13.5){$\frac{33}{8}$}
\rput(1.5,14.60){$\vdots$}
\rput(2.5,0.5){$\frac{35}{24}$}
\rput(2.5,1.5){$1$}
\rput(2.5,2.5){$\frac{5}{8}$}
\rput(2.5,3.5){$\frac{1}{3}$}
\rput(2.5,4.5){$\frac{1}{8}$}
\rput(2.5,5.5){$0$}
\rput(2.5,6.5){$-\frac{1}{24}$}
\rput(2.5,7.5){$0$}
\rput(2.5,8.5){$\frac{1}{8}$}
\rput(2.5,9.5){$\frac{1}{3}$}
\rput(2.5,10.5){$\frac{5}{8}$}
\rput(2.5,11.5){$1$}
\rput(2.5,12.5){$\frac{35}{24}$}
\rput(2.5,13.5){$2$}
\rput(2.5,14.60){$\vdots$}
\rput(3.5,0.5){$\frac{10}{3}$}
\rput(3.5,1.5){$\frac{21}{8}$}
\rput(3.5,2.5){$2$}
\rput(3.5,3.5){$\frac{35}{24}$}
\rput(3.5,4.5){$1$}
\rput(3.5,5.5){$\frac{5}{8}$}
\rput(3.5,6.5){$\frac{1}{3}$}
\rput(3.5,7.5){$\frac{1}{8}$}
\rput(3.5,8.5){$0$}
\rput(3.5,9.5){$-\frac{1}{24}$}
\rput(3.5,10.5){$0$}
\rput(3.5,11.5){$\frac{1}{8}$}
\rput(3.5,12.5){$\frac{1}{3}$}
\rput(3.5,13.5){$\frac{5}{8}$}
\rput(3.5,14.60){$\vdots$}
\multiput(0,0)(0,1){14}{\rput(4.5,0.525){$\cdots$}}
\rput(4.5,14.50){$\vvdots$}
\end{pspicture}$$
For logarithmic minimal models ${\cal LM}(p,p')$ with $p>1$, the MIPF is not uniquely determined by conformal and modular invariance. The conjectured form [@PRcosetGraphs11] for these MIPFs is
$$\begin{aligned}
{2}
Z_{p,p'}(q)&=Z^{\text{Proj}}_{p,p'}(q)+n_{p,p'} Z^{\text{Min}}_{p,p'}(q),\label{Zppa}\\
&=\tfrac{1}{2}(1+n_{p,p'})Z_{1,pp'}^{\text{Circ}}(q)+\tfrac{1}{2}(1-n_{p,p'})Z_{p,p'}^{\text{Circ}}(q),\qquad n_{p,p'}\in {\mathbb Z},
\label{Zpp}\end{aligned}$$
where the integer $n_{p,p'}$ is undetermined and the projective partition function $Z^{\text{Proj}}_{p,p'}(q)$ is defined in [@PRcosetGraphs11]. The $u(1)$ modular invariant partition functions, corresponding to a compactified boson on $S^1$ with radius $R=\sqrt\frac{2p'}{p}$, are Z\^\_[p,p’]{}(q)=\_[j=0]{}\^[2n-1]{} \^n\_j(q)\^n\_[\_0 j]{}(|q) \[ZCirc\] where $n=pp'$ and the $u(1)$ characters $\varkappa_j^n(q)$ are given by . The Bezout number $\omega_0$ is defined by \_0=r\_0p’+s\_0p() in terms of the Bezout pair $(r_0,s_0)$ which is uniquely determined by the conditions r\_0p’-s\_0p=1,1r\_0p-1,1s\_0p’-1,p s\_0<p’r\_0. For $p=1$, $Z_{1,p'}(q)=Z^{\text{Circ}}_{1,p'}(q)$ is the diagonal $u(1)$ partition function
$$Z^{\text{Circ}}_{1,p'}(q) = \sum_{j=0}^{p'} d^{p'}_j |\varkappa_{j}^{p'}(q)|^2,\qquad
\label{eq:dj}
d_j^{n}=
\left\{\begin{array}{ll}
1,&j=0, \,n,\\
2,&\mbox{otherwise}
\end{array}\right.$$
and $Z^{\text{Min}}_{p,p'}(q)=\tfrac{1}{2}(Z^{\text{Circ}}_{1,pp'}(q)-Z^{\text{Circ}}_{p,p'}(q))$ implies $Z^{\text{Min}}_{1,p'}(q)=0$. For $p>1$, $Z^{\text{Circ}}_{p,p'}(q)$ is a non-diagonal $u(1)$ partition function. For critical percolation, $Z^{\text{Min}}_{2,3}(q)=1$ and our analytic derivation of the MIPF from the lattice model shows that $n_{2,3} = -1$. The MIPF of critical percolation is therefore given by the non-diagonal $u(1)$ partition function $$Z_{2,3}(q)=Z^{\text{Circ}}_{2,3}(q)$$ where the Bezout number giving the Bezout conjugation is $\omega_0=5$.
The layout of the paper is as follows. \[sec:CFTdata\] recalls the conformal data for bond percolation which is referred to in the rest of the paper. \[sec:boundarycase\] contains our computations and results for bond percolation on the strip with vacuum boundary conditions. We recall the definition of the Temperley-Lieb algebra $\tl_N$ and the transfer tangle $\Db(u)$ in \[sec:TLandD\] and review the standard modules over this algebra in \[sec:stanmod\]. In \[sec:fusion\], we give the definition of the fused transfer matrices and present the fusion hierarchy relations. We write down the corresponding $T$- and $Y$-systems in \[sec:functional\]. In \[sec:props\], we analyse the analyticity properties of the eigenvalues of the transfer matrices in terms of their patterns of zeros. In \[sec:fsc\], we transform the $T$- and $Y$-systems into TBA equations and solve for the finite-size corrections of the finite excitations characterized by their patterns of zeros. The results are expressed in terms of sums of Rogers dilogarithms which are evaluated in \[sec:Rogers\]. In \[sec:cfgs\], we specialise the result to the ground states of the standard modules and reproduce the conformal weights in the $(1,s)$ column of the Kac table. In \[sec:colconfs\], we review the construction of single- and double-column diagrams which were previously introduced in the analysis of critical dense polymers. In \[sec:excitations\], we formulate a set of empirical selection rules which describe, in terms of column diagrams, the patterns of zeros for the full set of eigenvalues of the standard modules. We use these to write down explicit expressions for the finite-size characters. These are simplified to the known finitized Kac characters in \[sec:charids\] using identities derived in \[sec:qids\]. In \[sec:cylPF\], we combine the partition functions of the standard modules using the Markov trace to obtain the conformal cylinder partition function.
In \[sec:periodiccase\], we present our results for periodic boundary conditions. \[sec:PTLandT,sec:Pstanmod,sec:Pfusion,sec:Pfunctional,sec:Pprops,sec:Pfsc,sec:Pcfgs\] follow the same presentation as \[sec:TLandD,sec:stanmod,sec:fusion,sec:functional,sec:props,sec:fsc,sec:cfgs\], presenting the corresponding results for the periodic case. In \[sec:fsgf\], we write down empirical selection rules that describe the full set of finite excitations in the standard modules over the periodic Temperley-Lieb algebra $\eptl_N(\alpha, \beta)$. These allow us to write down explicit expressions for the spectrum generating functions, which are collected in \[sec:Pcpf\]. Also in \[sec:Pcpf\], the behavior of these generating functions in the scaling limit is extracted using the identities derived in \[sec:qids\]. In \[sec:mipf\], we combine the previous results using the equivalent of the Markov trace for the torus, compute the modular invariant/covariant partition functions and write the result in terms of $u(1)$ characters. \[sec:conclusion\] presents a discussion of our results and an overview of future avenues to be explored. The torus partition functions for critical dense polymers and sample patterns of zeros for critical percolation are collected in Appendices C and D respectively.
Conformal data of critical percolation {#sec:CFTdata}
======================================
For critical percolation, the central charge $c$ and the Virasoro Kac conformal weights $\Delta_{r,s}$ are given by c = 0, \_[r,s]{} = , with $r,s \in \mathbb Z_{>0}$. These are organised in the infinitely extended Kac table in the left panel of \[fig:VirKac\]. The conformal weights with $r=0,1$ and $s$ taking half-integer are given in the right panel.
In terms of the lattice data, the modular nome is given by \[eq:qbdy\] q = (- 2(3u)) for the boundary case and by \[eq:qper\] q = (- 2 \^[-3u]{}) for the periodic case, where the aspect ratio is = \_[M,N]{} MN. The finitized Kac characters are given by \[eq:finchar\] \_[r,s]{}(q) = q\^[\_[r,s]{}-c/24]{} (-q\^[rs]{} ) and yield the conformal Kac characters in the scaling limit: \_[r,s]{}(q) = \_[N]{}\_[r,s]{}(q) = q\^[\_[r,s]{}-c/24]{} where \[eq:Poch\] (q)\_= [\_[i=1]{}\^(1-q\^i)]{}. The $u(1)$ characters are given by $$\varkappa_j^{n}(q) = \varkappa_j^{n}(q,1), \qquad \varkappa_j^{n}(q,z)=\frac{\Theta_{j,n}(q,z)}{q^{1/24}(q)_\infty}=\frac{q^{-1/24}}{(q)_\infty} \sum_{k\in{\mathbb Z}} z^kq^{(j+2kn)^2/4n},
\label{eq:u1chars}$$ with $n = pp' = 6$ for percolation.
Critical percolation with strip boundary conditions {#sec:boundarycase}
===================================================
The transfer tangle and the Temperley-Lieb algebra {#sec:TLandD}
--------------------------------------------------
The dense loop model of critical percolation is a Temperley-Lieb model described in terms of the elementary face operator
(1,1) (0,0)[0.16]{}[0]{}[90]{} (.5,.5)[$u$]{}
= s\_1(-u)
(1,1) (0,0)
+s\_0(u)
(1,1) (0,0)
\[eq:faceop\] where s\_k(u)=. Here $\lambda$ is the crossing parameter and is related to the loop fugacity $\beta$ by $\beta = 2 \cos \lambda$. For critical percolation, $\lambda = \frac \pi 3$ and therefore $\beta =1$. The double-row transfer tangle is defined as (u)= (-1)\^N
(-0.5,-0.8)(5.5,2.0) (0,0)[0.16]{}[0]{}[90]{} (0,1)[0.16]{}[0]{}[90]{} (1,0)[0.16]{}[0]{}[90]{} (1,1)[0.16]{}[0]{}[90]{} (4,0)[0.16]{}[0]{}[90]{} (4,1)[0.16]{}[0]{}[90]{} (2.5,0.5)[$\ldots$]{} (2.5,1.5)[$\ldots$]{} (3.5,0.5)[$\ldots$]{} (3.5,1.5)[$\ldots$]{} (0,1)[0.5]{}[90]{}[-90]{} (5,1)[0.5]{}[-90]{}[90]{} (0.5,.5)[$u$]{} (0.5,1.5)[$\lambda\!-\!u$]{} (1.5,.5)[$u$]{} (1.5,1.5)[$\lambda\!-\!u$]{} (4.5,.5)[$u$]{} (4.5,1.5)[$\lambda\!-\!u$]{} (2.5,-0.5)[$\underbrace{\qquad \qquad \qquad \qquad \qquad \qquad}_N$]{}
\[eq:Du\] where $u$ is the spectral parameter. The tangle $\Db(u)$ is a linear combination of connectivity diagrams and therefore an element of the Temperley-Lieb algebra [@TL71] $\tl_N(\beta)$ at $\beta =1$: \[eq:TLdiag\] \_N()=I,e\_j ;j=1,…,N-1, I=
(0.0,-0.65)(2.0,0.45) (0,-0.35)(2.0,-0.35)(2.0,0.35)(0,0.35) (1.4,0.0)[$...$]{} (0.2,0.35)(0.2,-0.35)(0.2,-0.55)[$_1$]{} (0.6,0.35)(0.6,-0.35)(0.6,-0.55)[$_2$]{} (1.0,0.35)(1.0,-0.35)(1.0,-0.55)[$_3$]{} (1.8,0.35)(1.8,-0.35)(1.8,-0.55)[$_N$]{}
, e\_j=
(0.0,-0.65)(3.2,0.45) (0,-0.35)(3.2,-0.35)(3.2,0.35)(0,0.35) (0.6,0.0)[$...$]{} (2.6,0.0)[$...$]{} (0.2,0.35)(0.2,-0.35)(0.2,-0.55)[$_1$]{} (1.0,0.35)(1.0,-0.35) (2.2,0.35)(2.2,-0.35) (3.0,0.35)(3.0,-0.35)(3.0,-0.55)[$_{N}$]{} (1.6,0.35)[0.2]{}[180]{}[0]{}(1.35,-0.55)[$_j$]{} (1.6,-0.35)[0.2]{}[0]{}[180]{}(1.85,-0.55)[$_{j+1}$]{}
. The algebra $\tl_N(\beta)$ is a unital associative algebra whose defining relations are e\_j\^2=e\_j, e\_j e\_[j1]{} e\_j = e\_j, e\_i e\_j = e\_j e\_i (|i-j|>1). \[eq:TLdef\] The transfer tangle satisfies a number of relations, in particular the crossing symmetry $\Db(\lambda - u) = \Db(u)$, the periodicity symmetry $\Db(u+\pi) = \Db(u)$, the commutativity property $[\Db(u),\Db(v)] = 0$ and the initial condition $\Db(u=0) = (-1)^N \Ib$. We sometimes denote the identity connectivity using the bold letter $\Ib$.
The braid transfer matrix is also an element of $\tl_N(\beta)$. It is defined by \[eq:braidD\] \_=
(-0.5,-0.8)(5.5,2.0) (2.5,0.5)[$\ldots$]{} (2.5,1.5)[$\ldots$]{} (3.5,0.5)[$\ldots$]{} (3.5,1.5)[$\ldots$]{} (0,1)[0.5]{}[90]{}[-90]{} (5,1)[0.5]{}[-90]{}[90]{} (0,0)[(0.0,0.5)(1,0.5) (0.5,0)(0.5,0.35) (0.5,0.65)(0.5,1)]{} (1,0)[(0.0,0.5)(1,0.5) (0.5,0)(0.5,0.35) (0.5,0.65)(0.5,1)]{} (4,0)[(0.0,0.5)(1,0.5) (0.5,0)(0.5,0.35) (0.5,0.65)(0.5,1)]{} (0,1)[(0.5,0.0)(0.5,1) (0,0.5)(0.35,0.5) (0.65,0.5)(1,0.5)]{} (1,1)[(0.5,0.0)(0.5,1) (0,0.5)(0.35,0.5) (0.65,0.5)(1,0.5)]{} (4,1)[(0.5,0.0)(0.5,1) (0,0.5)(0.35,0.5) (0.65,0.5)(1,0.5)]{} (2.5,-0.5)[$\underbrace{\qquad \qquad \qquad \qquad \qquad \qquad}_N$]{}
where the elementary braid operators are given by \[eq:braidops\]
(1,1) (0,0.5)(1,0.5) (0.5,0)(0.5,0.35) (0.5,0.65)(0.5,1)
= e\^[-2]{}
(1,1) (0,0)
+e\^[2]{}
(1,1) (0,0)
,
(1,1) (0,0.5)(0.35,0.5) (0.65,0.5)(1,0.5) (0.5,0)(0.5,1)
=e\^[2]{}
(1,1) (0,0)
+e\^[-2]{}
(1,1) (0,0)
. The braid transfer matrix is obtained as the $u \rightarrow \ir \infty$ limit of $\Db(u)$: \_= \_[u ]{} ()\^N (-1)\^N(u). We note that $\Db_\infty$ is also obtained by taking the limit $u\to -\ir \infty$ of $\Db(u)$.
Standard modules {#sec:stanmod}
----------------
The representation theory of the Temperley-Lieb algebra was investigated by Jones [@Jones83], Martin [@M91], Goodman and Wenzl [@GW93] and Westbury [@W95] and was recently reviewed by Ridout and Saint-Aubin [@RSA14]. In the following, we study the action of $\Db(u)$ on a family of finite-dimensional modules over $\tl_N(\beta)$: the standard modules $\stan_N^d$. These modules are constructed on the vector spaces generated from link states with $d$ defects, with $0 {\leqslant}d {\leqslant}N$ and $d \equiv N \text{ mod }2$, and have dimension \_N\^d = -. For example, for $N = 6$ and $d=2$, there are nine link states:
[c]{}
(-0.0,0)(2.4,0.5) (0,0)(2.4,0) (0.4,0)[0.2]{}[0]{}[180]{} (1.2,0)[0.2]{}[0]{}[180]{} (1.8,0)(1.8,0.5) (2.2,0)(2.2,0.5)
,
(-0.0,0)(2.4,0.5) (0,0)(2.4,0) (0.4,0)[0.2]{}[0]{}[180]{} (1.6,0)[0.2]{}[0]{}[180]{} (1.0,0)(1.0,0.5) (2.2,0)(2.2,0.5)
,
(-0.0,0)(2.4,0.5) (0,0)(2.4,0) (0.4,0)[0.2]{}[0]{}[180]{} (2.0,0)[0.2]{}[0]{}[180]{} (1.0,0)(1.0,0.5) (1.4,0)(1.4,0.5)
,
(-0.0,0)(2.4,0.5) (0,0)(2.4,0) (0.8,0)[0.2]{}[0]{}[180]{} (1.6,0)[0.2]{}[0]{}[180]{} (0.2,0)(0.2,0.5) (2.2,0)(2.2,0.5)
,
(-0.0,0)(2.4,0.5) (0,0)(2.4,0) (0.8,0)[0.2]{}[0]{}[180]{} (2.0,0)[0.2]{}[0]{}[180]{} (0.2,0)(0.2,0.5) (1.4,0)(1.4,0.5)
,\
(-0.0,0)(2.4,0.5) (0,0)(2.4,0) (0.2,0)(0.2,0.5) (0.6,0)(0.6,0.5) (1.2,0)[0.2]{}[0]{}[180]{} (2.0,0)[0.2]{}[0]{}[180]{}
,
(-0.0,0)(2.4,0.5) (0,0)(2.4,0) (0.8,0)[0.2]{}[0]{}[180]{} (0.2,0)(0.2,0.7)(1.4,0.7)(1.4,0) (1.8,0)(1.8,0.5) (2.2,0)(2.2,0.5)
,
(-0.0,0)(2.4,0.5) (0,0)(2.4,0) (1.2,0)[0.2]{}[0]{}[180]{} (0.6,0)(0.6,0.7)(1.8,0.7)(1.8,0) (0.2,0)(0.2,0.5) (2.2,0)(2.2,0.5)
,
(-0.0,0)(2.4,0.5) (0,0)(2.4,0) (1.6,0)[0.2]{}[0]{}[180]{} (1.0,0)(1.0,0.7)(2.2,0.7)(2.2,0) (0.2,0)(0.2,0.5) (0.6,0)(0.6,0.5)
.
The standard modules are defined by the defect-preserving action of the Temperley-Lieb connectivity diagrams on the link patterns. To compute this action, one draws the link state above the connectivity diagram and reads the new link state from the bottom nodes. A multiplicative factor of $\beta$ is then added for each closed loop. The result is set to zero if the number of defects of the new link state is smaller than that of the original link state. Here are examples to illustrate:
(0,-0.65)(1.6,0.95) (0,-0.35)(1.6,-0.35)(1.6,0.35)(0,0.35) (0.2,0.35)(0.2,-0.35) (0.6,0.35)(0.6,-0.35) (1.2,0.35)[0.2]{}[180]{}[0]{} (1.2,-0.35)[0.2]{}[0]{}[180]{} (0,0.35)(1.6,0.35) (0.8,0.35)[0.2]{}[0]{}[180]{} (0.2,0.35)(0.2,1.05)(1.4,1.05)(1.4,0.35)
=
(0,0.35)(1.6,0.95) (0,0.35)(1.6,0.35) (0.4,0.35)[0.2]{}[0]{}[180]{} (1.2,0.35)[0.2]{}[0]{}[180]{}
,
(0,-0.65)(1.6,0.95) (0,-0.35)(1.6,-0.35)(1.6,0.35)(0,0.35) (0.2,0.35)(0.2,-0.35) (0.6,0.35)(0.6,-0.35) (1.2,0.35)[0.2]{}[180]{}[0]{} (1.2,-0.35)[0.2]{}[0]{}[180]{} (0,0.35)(1.6,0.35) (1.2,0.35)[0.2]{}[0]{}[180]{} (0.2,0.35)(0.2,0.85) (0.6,0.35)(0.6,0.85)
=
(0,0.35)(1.6,0.95) (0,0.35)(1.6,0.35) (1.2,0.35)[0.2]{}[0]{}[180]{} (0.2,0.35)(0.2,0.85) (0.6,0.35)(0.6,0.85)
,
(0,-0.65)(1.6,0.95) (0,-0.35)(1.6,-0.35)(1.6,0.35)(0,0.35) (0.2,0.35)(0.2,-0.35) (0.6,0.35)(0.6,-0.35) (1.2,0.35)[0.2]{}[180]{}[0]{} (1.2,-0.35)[0.2]{}[0]{}[180]{} (0,0.35)(1.6,0.35) (0.4,0.35)[0.2]{}[0]{}[180]{} (1.0,0.35)(1.0,0.85) (1.4,0.35)(1.4,0.85)
= 0. The standard modules play a key role in the representation theory of $\tl_N(\beta)$. In particular, they generate a complete set of irreducible modules for generic values of $\beta$.
The case $\beta = 1$ is however not generic: The standard modules remain indecomposable, but depending on $d$ some of them are reducible. Let us refer to the integers $d \equiv 2 \text{ mod } 3$ in the set $\{0, \dots, N\}$ as [*critical integers*]{} and to the maximal such integer as $\hat d$. For $d$ critical, the standard module $\stan_N^d$ is irreducible: $\stan_N^d \simeq \irre_N^d$. For $d \equiv 0,1 \text{ mod } 3$, $\stan_N^d$ typically has two composition factors: an irreducible submodule $R_N^d$ and an irreducible quotient $\irre_N^d$. The submodule $R_N^d$ is isomorphic to $\irre_N^{d'}$ where $d' = d+4$ for $d \equiv 0 \text{ mod } 3$ and $d' = d+2$ for $d \equiv 1 \text{ mod } 3$. In other words, $d'$ is the integer obtained by reflecting $d$ with respect to the next critical integer. The structure of $\stan_N^d$ in this case is easily understood from its Loewy diagram, which we write as $\stan_N^d \simeq \irre_N^d \rightarrow \irre_N^{d'}$. The arrow indicates that the states in $\irre_N^{d'}$ can be obtained from those in $\irre_N^d$ by the action of $\tl_N(\beta = 1)$, but not the other way around. If $d'>\hat d$, then $R_N^d$ is trivial and the corresponding standard module is irreducible: $\stan_N^d\simeq \irre_N^d$.
The modules $\irre_N^d$, with $0 {\leqslant}d {\leqslant}N$ and $d \equiv N \text{ mod } 2$, form a complete list of non-isomorphic irreducible modules of $\tl_N(\beta =1)$. Their dimensions are given by \_N\^d = {
[ll]{}\_N\^d & d 2 3,\
\_[k0]{} \_N\^[d+6k]{} - \_[k0]{} \_N\^[d’+6k]{} & d 0,1 3,
. where it is understood that $\dim \stan_N^d = 0$ for $d>N$. These dimensions are displayed in \[tab:dim\] for $1{\leqslant}N {\leqslant}10$. We note that $\irre_N^0$ and $\irre_N^1$ are always one-dimensional.
$$\begin{pspicture}(0,0)(0.1,0.1)
\psline[linecolor=gray,linestyle=dashed,dash=1.5pt 2pt]{-}(4.91,-3.2)(4.91,-2.95)
\psline[linecolor=gray,linestyle=dashed,dash=1.5pt 2pt]{-}(4.91,-2.55)(4.91,-1.8)
\psline[linecolor=gray,linestyle=dashed,dash=1.5pt 2pt]{-}(4.91,-1.4)(4.91,-0.65)
\psline[linecolor=gray,linestyle=dashed,dash=1.5pt 2pt]{-}(4.91,-0.25)(4.91,0.5)
\psline[linecolor=gray,linestyle=dashed,dash=1.5pt 2pt]{-}(4.91,0.9)(4.91,1.65)
\psline[linecolor=gray,linestyle=dashed,dash=1.5pt 2pt]{-}(4.91,2.1)(4.91,2.6)
\psline[linecolor=gray,linestyle=dashed,dash=1.5pt 2pt]{-}(9.505,-3.2)(9.505,-2.4)
\psline[linecolor=gray,linestyle=dashed,dash=1.5pt 2pt]{-}(9.505,-1.95)(9.505,-1.25)
\psline[linecolor=gray,linestyle=dashed,dash=1.5pt 2pt]{-}(9.505,-0.8)(9.505,-0.10)
\psline[linecolor=gray,linestyle=dashed,dash=1.5pt 2pt]{-}(9.505,0.35)(9.505,2.6)
\psline[linecolor=gray,linestyle=dashed,dash=1.5pt 2pt]{-}(13.685,-3.2)(13.685,-2.95)
\psline[linecolor=gray,linestyle=dashed,dash=1.5pt 2pt]{-}(13.685,-2.55)(13.685,-1.8)
\psline[linecolor=gray,linestyle=dashed,dash=1.5pt 2pt]{-}(13.685,-1.4)(13.685,2.6)
\end{pspicture}
\begin{array}{c | cccccccccccccc}
N\backslash d & 0 & 1 & 2 & 3 & 4 & 5 & 6 & 7 & 8 & 9 & 10
\\
\hline\\[-0.3cm]
1 & & 1 & \phantom{1 \rightarrow 1} &\phantom{1 \rightarrow 1} &\phantom{1 \rightarrow 1} & \phantom{1 \rightarrow 1} &\phantom{1 \rightarrow 1} &\phantom{1 \rightarrow 1} & \phantom{1 \rightarrow 1} & \phantom{1 \rightarrow 1}& \phantom{1 \rightarrow 1}\\[0.1cm]
2 & 1 && 1 & & & & & & & \\[0.1cm]
3 && 1 \rightarrow 1&& 1 && && && \\[0.1cm]
4 & 1\rightarrow 1 && 3 && 1 && && && \\[0.1cm]
5 && 1\rightarrow 4&& 4 && 1 && && \\[0.1cm]
6 & 1\rightarrow 4 && 9 && 4\rightarrow 1&& 1&& && \\[0.1cm]
7 && 1\rightarrow 13&& 13\rightarrow 1&& 6 && 1 && \\[0.1cm]
8 & 1\rightarrow 13&& 28 && 13\rightarrow 7&& 7 && 1 && \\[0.1cm]
9 && 1\rightarrow 41&& 41\rightarrow 7&& 27 && 7 \rightarrow 1&& 1\\[0.1cm]
10 & 1\rightarrow 41 && 90 && 41\rightarrow 34 && 34 \rightarrow 1&& 9 && 1 \\[0.1cm]
\end{array}$$
\[tab:dim\]
Fused transfer matrices and the fusion hierarchy {#sec:fusion}
------------------------------------------------
Starting from the transfer tangle $\Db^1(u) = \Db(u)$, one can construct a family of fused transfer tangles $\Db^n(u)$ satisfying the fusion hierarchy relations $$\Db^n_0\Db^{1}_n=\frac{s_{n-3}(2u)s_{2n}(2u)}{s_{n-2}(2u)s_{2n-1}(2u)} f_n \Db^{n-1}_0 + \frac{s_{n-1}(2u)s_{2n-2}(2u)}{s_{n-2}(2u)s_{2n-1}(2u)}f_{n-1}\, \Db^{n+1}_0, \qquad n {\geqslant}0,
\label{eq:fushier}$$ where \_k\^n = \^n(u + k ), \_0\^0 = f\_[-1]{} , \^[-1]{}\_k = 0, f\_k = (-1)\^[N]{}s\_[k]{}(u)\^[2N]{}. These transfer tangles were constructed in terms of fused face operators in [@MDPR14]. They commute as elements of $\tl_N(\beta)$: $[\Db^m(u),\Db^n(v)] = 0$. The fusion hierarchy relations were proven using the diagrammatic calculus of the Temperley-Lieb algebra. Of particular relevance for our investigation in later sections is the fused transfer tangle with fusion label $n=2$. For $\beta\ne 0$, it is constructed from the Wenzl-Jones projector on two sites,
(0,-0.15)(1.0,0.15) (0,-0.15)(1.0,-0.15)(1.0,0.15)(0,0.15)(0,0.15) (0.5,0)[$_{2}$]{}
=
(-0.0,-0.35)(0.8,0.35) (0.2,0.35)(0.2,-0.35) (0.6,0.35)(0.6,-0.35)
-
(-0.,-0.35)(0.8,0.35) (0.4,0.35)[0.2]{}[180]{}[0]{} (0.4,-0.35)[0.2]{}[0]{}[180]{}
, and the $1 \times 2$ fused face operator \[eq:1x2face\]
(0,0)(1,1) (0,0)(1,0)(1,1)(0,1)(0,0) (0,0)[0.16]{}[0]{}[90]{} (0.5,0.75) (0.5,0.5)[$u$]{}
= 1[s\_0(u)]{}
(-0.3,0)(1.3,2) (0.5,0.55)[$u$]{} (0.5,1.55)[$u\!+\!\lambda$]{} (0,0.1)(0,1.9)(-0.3,1.9)(-0.3,0.1)(0,0.1)(-0.15,1)[$_2$]{} (1,0.1)(1,1.9)(1.3,1.9)(1.3,0.1)(1,0.1)(1.15,1)[$_2$]{} (0,0)[0.16]{}[0]{}[90]{} (0,1)[0.16]{}[0]{}[90]{}
= s\_1(-u)
(-0.3,0)(1.3,2) (0,0) (0,1) (0,0.1)(0,1.9)(-0.3,1.9)(-0.3,0.1)(0,0.1)(-0.15,1)[$_2$]{} (1,0.1)(1,1.9)(1.3,1.9)(1.3,0.1)(1,0.1)(1.15,1)[$_2$]{}
+ s\_1(u)
(-0.3,0)(1.3,2) (0,0) (0,1) (0,0.1)(0,1.9)(-0.3,1.9)(-0.3,0.1)(0,0.1)(-0.15,1)[$_2$]{} (1,0.1)(1,1.9)(1.3,1.9)(1.3,0.1)(1,0.1)(1.15,1)[$_2$]{}
. The normalisation is different from that appearing in [@MDPR14] and is instead chosen such that the fusion hierarchy relations are identical to those of the rational models, see for instance [@CMP02]. The $1\times 2$ fused transfer tangle is defined as \^[2]{}(u)=
(-0.7,-0.7)(5.7,2) (0,1)[0.5]{}[90]{}[-90]{}(0,1)[0.5]{}[90]{}[-90]{} (5,1)[0.5]{}[-90]{}[90]{}(5,1)[0.5]{}[-90]{}[90]{} (0,0)[0.16]{}[0]{}[90]{} (0,1)[0.16]{}[0]{}[90]{} (1,0)[0.16]{}[0]{}[90]{} (1,1)[0.16]{}[0]{}[90]{} (4,0)[0.16]{}[0]{}[90]{} (4,1)[0.16]{}[0]{}[90]{} (0.5,0.75)(0.5,1.75) (1.5,0.75)(1.5,1.75) (4.5,0.75)(4.5,1.75) (2.5,0.5)[$\ldots$]{} (2.5,1.5)[$\ldots$]{} (3.5,0.5)[$\ldots$]{} (3.5,1.5)[$\ldots$]{} (0.5,.5)[$u$]{} (0.52,1.45)[$-u$]{} (1.5,.5)[$u$]{} (1.52,1.45)[$-u$]{} (4.5,.5)[$u$]{} (4.52,1.45)[$-u$]{} (2.5,-0.5)[$\underbrace{\qquad \qquad \qquad \qquad \qquad \qquad}_N$]{}
. \[eq:D2u\] It satisfies the relation $\Db^2(-u)=\Db^2(u)$, the periodicity property $\Db(u+\pi)=\Db(u)$, the commutativity property $[\Db^2(u),\Db^n(v)] = 0$, $n=1,2$, and most importantly the fusion hierarchy relation with $n=1$. The construction of $\Db^n(u)$ for $n>2$ in terms of diagrams uses similar ideas.
For rational values of the crossing parameter, that is for $\lambda =\frac{(p'-p)\pi}{p'}$ with $(p,p')$ a pair of integers satisfying , it was shown in [@MDPR14] that the fused transfer tangles satisfy the closure relation \^[p’]{}\_0 = \^[p’-2]{}\_1 + 2(-1)\^[p’-p]{} f\_[-1]{} . \[eq:clo\] Critical percolation corresponds to $(p,p') = (2,3)$, in which case this is just \^[3]{}\_0 = \^[1]{}\_1 - 2 f\_[-1]{} . \[eq:cloperco\] The proof given in [@MDPR14] relies on diagrammatic manipulations performed on the tangles $\Db^n(u)$ as elements of $\tl_N(\beta)$. If follows that holds for all modules, and in particular for $\stan_N^d$.
The $\boldsymbol T$-system and the $\boldsymbol Y$-system {#sec:functional}
---------------------------------------------------------
Using the fusion hierarchy relations and a recursive argument [@KP92], one can show that the transfer tangles satisfy a set of functional relations known as the $T$-system: $$\Db^{n}_0\Db^{n}_1 = \frac{s_{-2}(2u)s_{2n}(2u)}{s_{n-2}(2u)s_{n}(2u)} f_{-1} f_n \Ib + \frac{s_{n-1}(2u)^2}{s_{n-2}(2u)s_{n}(2u)} \Db^{n+1}_0\Db^{n-1}_1, \qquad n {\geqslant}0.
\label{eq:tsys}$$ The $T$-system holds for generic values of $\beta$. By defining \^n(u) = , n 0, and $\db^n_k = \db^n(u+k \lambda)$, one finds that these tangles satisfy a set of non-linear equations known as the (universal) $Y$-system: \^[n]{}\_0 \^[n]{}\_1=(+ \^[n-1]{}\_1)(+ \^[n+1]{}\_0), n 1, \[eq:ysys\] where $\db^0_k = 0$.
For rational values of $\lambda$, this set of non-linear relations closes finitely. Indeed, it was found in [@MDPR14] that the tangles $\db^n_k$ satisfy a linear, four-term closure relation. After a careful analysis, we find that this closure relation is not convenient for extracting eigenvalue solutions for $\Db(u)$. Inspired by ideas applied to vertex models [@KSS98], one instead defines the tangle \_0 = \_1\^[p’-2]{} \[eq:Ktang\] and, using and , finds the following alternative closure relations + \^[p’-1]{}\_0 = (+\_0)\^2, \_0\_1 = +\_1\^[p’-2]{}. \[eq:yclo\] The closed $Y$-system thus consists of the relations for $n = 1, \dots, p'-2$ along with the relations . This truncates the initial infinite $Y$-system, which corresponds to a one-sided $A_\infty$ Dynkin diagram, to a finite $Y$-system described by a Dynkin diagram of type $D_{p'}$, see \[fig:Dynkin\].
------------------------------------------------------------------------------------------------------------------------------------------------------------- -- --------------------------------------------------------------------------------------------------------------------------------------
$ $
A_\infty: \quad D_{p'}: \quad
\begin{pspicture}[shift=-0.9](0,-1)(6,1) \begin{pspicture}[shift=-0.9](0,-1)(5,1)
\multiput(0,0)(1,0){4}{\psline{-}(0,0)(1,0)} \multiput(0,0)(1,0){3}{\psline{-}(0,0)(1,0)}
\rput(4,0){\psline{-}(0,0)(0.5,0)} \psline{-}(3,0)(3.7,0.7)\psline{-}(3,0)(3.7,-0.7)
\multiput(0,0)(1,0){5}{\pscircle[linewidth=1.5pt,linecolor=black,fillstyle=solid,fillcolor=white](0,0){.175}} \multiput(0,0)(1,0){4}{\pscircle[linewidth=1.5pt,linecolor=black,fillstyle=solid,fillcolor=white](0,0){.175}}
\rput(0,0.375){\scriptsize$1$}\rput(1,0.375){\scriptsize$2$}\rput(2,0.375){\scriptsize$3$}\rput(3,0.375){\scriptsize$4$}\rput(4,0.375){\scriptsize$\cdots$} \rput(0,0.35){\scriptsize$1$}\rput(1,0.375){\scriptsize$2$}\rput(2,0.375){\scriptsize$\dots$}\rput(2.9,0.375){\scriptsize$p'\!-\!2$}
\rput(5,0){$\cdots$} \pscircle[linewidth=1.5pt,linecolor=black,fillstyle=solid,fillcolor=white](3.7,0.7){.175}
\end{pspicture} \pscircle[linewidth=1.5pt,linecolor=black,fillstyle=solid,fillcolor=white](3.7,-0.7){.175}
$ \rput(3.7,1.06){\scriptsize$p'\!-\!1$}
\rput(3.7,-1.10){\scriptsize$p'$}
\rput(3.7,0.7){\tiny$+$}
\rput(3.7,-0.7){\tiny$-$}
\end{pspicture}
$
------------------------------------------------------------------------------------------------------------------------------------------------------------- -- --------------------------------------------------------------------------------------------------------------------------------------
For $(p,p')=(2,3)$, the Dynkin diagram is $D_3 \simeq A_3$ and the finite $Y$-system only involves three tangles: the tangles $\Kb$ and $\db^1$, which are scalar multiples of $\Db^1(u)$ and $\Db^2(u)$ respectively, and the identity $\Ib$. It takes the form $$\db_0^1\db_1^1 = (\Ib+\Kb_0)^2, \qquad \Kb_0\Kb_1 = \Ib + \db_1^1.
\label{eq:ysysperco1}$$ By defining
\[eq:aU\] $$\begin{aligned}
{3}
&\ab^1(x) = \db_0^1(\tfrac {\ir x} 3), \qquad &&\ab^2(x) = \Kb_0(\tfrac {\ir x} 3- \tfrac \pi 6), \\ &\Ab^1(x) = \Ib+\ab^1(x), \qquad &&\Ab^2(x) = \big(\Ib+\ab^2(x)\big)^2,\end{aligned}$$
the $Y$-system is written in a symmetric form: $$\ab^1(x-\tfrac{\ir \pi}2)\ab^1(x+\tfrac{\ir \pi}2) = \Ab^2(x),\qquad
\ab^2(x-\tfrac{\ir \pi}2)\ab^2(x+\tfrac{\ir \pi}2) = \Ab^1(x).
\label{eq:finaly}$$
Properties of the eigenvalues {#sec:props}
-----------------------------
The functional relations given in the previous section were derived quite generally in [@MDPR14] using the diagrammatic calculus of the Temperley-Lieb algebra. The eigenvalues of the corresponding tangles are solutions to these relations in any given representation. In \[sec:fsc\], we use these relations to extract the finite-size corrections for the eigenvalues of $\Db(u)$ in the modules $\stan_N^d$. The analysis is based on some properties of the eigenvalues of $\Db^n(u)$, $\db^1(u)$ and $\Kb(u)$ in $\stan_N^d$, which we respectively denote by $D^n(u)$, $d^1(u)$ and $K(u)$. This section details these properties.
#### Razumov-Stroganov eigenvalues.
A simple solution to is $K_0 = -1$ and $d_0^1 = 0$, which corresponds to D\^1(u) = f\_[-2]{} = (-1)\^N ()\^[2N]{}, D\^2(u) = 0. \[eq:RSeig\] This solution appears once, as the ground state, in the spectrum of the standard module $\stan_N^d$ for $d=0$ if $N$ is even, and for $d=1$ if $N$ is odd. This is the celebrated Razumov-Stroganov eigenvalue for the loop model with strip boundary conditions [@BGN01; @MBNM04]. It corresponds to the unique eigenvalue of the trivial rational model of percolation, or alternatively to the one-dimensional irreducible representations of $\tl_N(\beta = 1)$, $\irre_N^0$ or $\irre_N^1$, see \[sec:stanmod\]. In this case, the expansion can be computed exactly and the $\frac1N$ finite-size corrections are exactly zero, consistent with $c= \Delta = 0$.
#### Patterns of zeros and analyticity strips.
For the other eigenvalues, the exact solutions to the $Y$-system are unknown, but as shown in \[sec:fsc\], it is possible to compute the finite-size corrections. This requires knowledge about the analytic behavior of the eigenvalues, which we formulate empirically based on exact computations for small system sizes. Our computer implementation produces the zeros and analyticity data of the eigenvalues in the standard modules up to $N=12$. We find that the leading eigenvalues have the following analyticity strips: D\^1(u): -6< (u) < 2, D\^2(u): -3< (u) < 3. Let us be more precise as to what this means. From the definitions and , $D^1(u)$ and $D^2(u)$ are Laurent polynomials in the variable $z = \eE^{\ir u}$, with minimal and maximal power $-2N$ and $2N$. Their eigenvalues share this property and thus as functions of $z$ have at most $4N$ zeros, and no poles except at $z = 0$. Due to the property $\Db^{n}(u+\pi)=\Db^{n}(u)$, in the complex $u$ plane, there are at most $2N$ zeros in any vertical strip of width $\pi$ and the pattern is repeated periodically. For $\Db^1(u)$, the leading eigenvalues in each $\stan_N^d$ have a finite number of zeros inside the analyticity strip. The other zeros are located either on the edges of this strip, that is for $\text{Re}(u) = -\frac\pi6$ and $\frac\pi2$, or outside the analyticity strip at $\text{Re}(u) =-\frac\pi3,\frac{2\pi}3$. As $N$ grows, the number of zeros of these leading eigenvalues inside the analyticity strip of $\Db^1(u)$ remains unchanged, whereas the number of zeros on the edges increases. For $\Db^2(u)$, the analyticity strip also contains finitely many zeros, and as $N$ increases, the extra zeros accumulate outside the analyticity strip on the lines $\text{Re}(u) = \pm\frac\pi2$. Two examples of patterns of zeros are given in \[fig:eigpatterns,fig:eigpatterns2\].
We observe that all the eigenvalues of $D^2(u)$ share single real zeros at $u = \pm \frac\pi 6,\pm \frac\pi 3$. These zeros can be understood from the fusion hierarchy equation for $n=1$. We also see, for instance in \[fig:eigpatterns,fig:eigpatterns2\], that $D^1(u)$ has a zero near $u = \frac \pi 2$. Its location is however not exactly at $u = \frac \pi 2$. Indeed, by specializing to $n=1$ and $u = \frac\pi2$, we find \^1(2)\^2 = ()\^[2N]{} . Using a diagrammatic argument, one can show that $\Db^1(\tfrac \pi 2) =(-\tfrac13)^N \Ib$. Thus, $D^1(u)$ has a zero near but not directly at $u = \frac \pi 2$. Its location in fact varies slightly for each eigenvalue.
The other zeros of $D^1(u)$ and $D^2(u)$ are not common to all the eigenvalues and come in complex conjugate pairs. From this observation, we infer that these eigenvalues are real for $\text{Im}(u) = 0$. Empirically, we also find that pairs of complex zeros inside the analyticity strips all lie on the central vertical line, that is respectively at $\text{Re}(u) = \frac \pi 6$ and $\text{Re}(u) = 0$ for $D^1(u)$ and $D^2(u)$. The degeneracy of these zeros is always one for $D^1(u)$, but can be one or two for $D^2(u)$. For instance, the eigenvalue whose zeros are shown in the right panel of \[fig:eigpatterns2\] has one pair of double zeros with $\text{Re}(u) = 0$. In all cases, the patterns of zeros are symmetric with respect to a reflection about the central vertical line of the analyticity strip. It follows that the eigenvalues are real on the central line of the analyticity strips.
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
$D^1(u):$ $D^2(u):$
\[0.1cm\] ![The location of the zeros for the ground state in $\stan_8^2$. The analyticity strips are shaded in gray. There is one pair of complex conjugate zeros in the analyticity strip for $D^1(u)$ and four real zeros on the real line for $D^2(u)$.[]{data-label="fig:eigpatterns"}](eig1d1.eps "fig:"){width=".45\textwidth"} ![The location of the zeros for the ground state in $\stan_8^2$. The analyticity strips are shaded in gray. There is one pair of complex conjugate zeros in the analyticity strip for $D^1(u)$ and four real zeros on the real line for $D^2(u)$.[]{data-label="fig:eigpatterns"}](eig1d2.eps "fig:"){width=".45\textwidth"}
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
$D^1(u):$ $D^2(u):$
\[0.1cm\] ![The location of the zeros for the fourth excited state in $\stan_8^2$. There are three pairs of complex zeros in the analyticity strip for $D^1(u)$. For $D^2(u)$, there are four real zeros, as well as one pair of complex conjugate zeros, each of order $2$, in the center of the analyticity strip.[]{data-label="fig:eigpatterns2"}](eig2d1.eps "fig:"){width=".45\textwidth"} ![The location of the zeros for the fourth excited state in $\stan_8^2$. There are three pairs of complex zeros in the analyticity strip for $D^1(u)$. For $D^2(u)$, there are four real zeros, as well as one pair of complex conjugate zeros, each of order $2$, in the center of the analyticity strip.[]{data-label="fig:eigpatterns2"}](eig2d2.eps "fig:"){width=".45\textwidth"}
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Because the zeros are symmetric under a reflection about the real axis, we depict an eigenvalue with a pattern diagram that includes only the zeros of the lower-half plane. In these diagrams, we omit the zeros on the real axis as they are common to all eigenvalues. For instance, the eigenvalues in \[fig:eigpatterns,fig:eigpatterns2\] are represented by the patterns
[rl]{}
(-0.4,-2.4)(1.4,0.5) (0,-1.9)(0,0)(1,0)(1,-1.9) (0,-0.3)(0,-0.6)(0,-0.9) (0.5,-1.2) (1,-0.3)(1,-0.6)(1,-0.9)
(-0.4,-2.4)(1.4,0.5) (0,-1.9)(0,0)(1,0)(1,-1.9)
[rl]{}
(-0.4,-2.4)(1.4,0.5) (0,-1.9)(0,0)(1,0)(1,-1.9) (0,-0.3)(0,-0.9) (0.5,-0.6)(0.5,-1.2)(0.5,-1.5) (1,-0.3)(1,-0.9)
(-0.4,-2.4)(1.4,0.5) (0,-1.9)(0,0)(1,0)(1,-1.9) (0.5,-0.95)
. Black and gray circles respectively denote single and double zeros, and the analyticity strip is delimited by the vertical segments. More examples are given in \[sec:pats\].
The analyticity properties of $K(u)$ and $d^1(u)$ follow readily from those of $D^1(u)$ and $D^2(u)$. From the definition , we see that $K(u)$ has a pole of order $2N$ at $u = \frac \pi3$, whereas $d^1(u)$ has poles of order $2N$ at $u =-\frac\pi3, \frac\pi 3$, a zero of order $2N+2$ at $u=0$ and neither poles nor zeros at $u = \pm \frac \pi 6$. The resulting analyticity strips K(u): -2< (u) < 6, d\^1(u): -3< (u) < 3 have width $2 \lambda$. A similar analysis on small system sizes reveals that the functions $1+K(u)$ and $1+d^1(u)$ are analytic and non-zero in the following strips of width $\lambda$: 1+K(u): -3< (u) < 0, 1+d\^1(u): -6< (u) < 6. In these cases, the analyticity strips are entirely free of zeros and poles.
Finally, in terms of the variables defined in , the analyticity strips take the following elegant forms:
$$\begin{aligned}
{4}
&\mathfrak a^1(x): &&\quad - \pi < \text{Im}(x) < \pi , \qquad
&&\mathfrak a^2(x): &&\quad -\pi < \text{Im}(x) < \pi, \\[0.15cm]
&\mathfrak A^1(x): &&\quad -\frac \pi 2< \text{Im}(x) < \frac \pi 2, \qquad
&&\mathfrak A^2(x): &&\quad -\frac \pi 2< \text{Im}(x) < \frac {\pi} 2.\end{aligned}$$
In terms of $x$, the patterns of zeros are rotated by 90 degrees, the central line of the analyticity strip coincides with the real axis and the pairs of complex zeros inside the analyticity strips lie on this axis.
#### Braid limit.
The braid limit $\Db^{n}_\infty$ is obtained by multiplying $\Db^n(u)$ by a suitable trigonometric function and taking the limit $u\rightarrow \ir \infty$. For $n=1$, $\Db^1_\infty\equiv \Db_\infty$ is defined in . For $n=2$, \^2\_= \_[u ]{} ()\^N \^2(u) = (\_\^1)\^2 - . The last equality is obtained by applying the braid limit to for $n=1$. Likewise, the braid tangles $\db^1_\infty$, $\Kb_\infty$, $\ab^1_\infty$ and $\ab^2_\infty$ are defined as \^1\_= \_[u ]{} \^1(u) = \_\^2, \_= \_[u ]{} (u) = - \_, \^n\_= \_[x ]{} \^n(x). For generic $\beta$, the matrix representatives of the braid transfer tangles on $\stan_N^d$ are scalar multiples of the identity matrix [@MDPR14], and the corresponding scalars depend only on $d$. On a given standard module $\stan_N^d$, each eigenvalue of $\db^1(u)$ has the same braid limit, and likewise for $\Kb(u)$. For $\beta = 1$, the braid behavior is
$$\begin{aligned}
{3}
&d \equiv 0,1 \text{ mod } 3:\qquad && D^1_\infty = 1, \qquad && D^2_\infty = 0, \\
&d \equiv 2 \text{ mod } 3:\qquad && D^1_\infty = -2, \qquad && D^2_\infty = 3,\end{aligned}$$
which implies that
$$\begin{aligned}
{3}
&d \equiv 0,1 \text{ mod } 3:\qquad && d^1_\infty = \mathfrak a^1_\infty= 0, \qquad && K_\infty = \mathfrak a^2_\infty= -1, \label{eq:braid01}\\
&d \equiv 2 \text{ mod } 3:\qquad && d^1_\infty =\mathfrak a^1_\infty= 3, \qquad && K_\infty = \mathfrak a^2_\infty= 2.\label{eq:braid02}\end{aligned}$$
One can readily check that these satisfy .
The analysis of the finite-size corrections in \[sec:fsc\] in the case $d \equiv 0,1 \Mod 3$ requires a refinement of to two subcases, characterised by the rate of convergence of $1+\mathfrak a^2(x)$ to zero as $x\rightarrow \pm \infty$. From our numerical investigation, we find the following subcases:[^1] : \_[x ]{} \^[3]{} (1+ a\^2(x)) = , : \_[x ]{} \^[3]{} (1+ a\^2(x)) = ’, \[eq:subcases\] where $\kappa,\kappa'$ are non-zero real constants. We argue that $\kappa,\kappa'>0$. Indeed, as noted earlier, on the real $x$ axis, the function $1+ \mathfrak a^2(x)$ is real but never zero. Our numerics reveal that there are always one or more values $x=x^j$ where $\mathfrak a^2(x)$ vanishes, implying that $1+\mathfrak a^2(x^j) = 1$. From the previous observations, $1+\mathfrak a^2(x)$ is positive everywhere on the real $x$ axis, so $\kappa,\kappa' >0$.
Together, and determine the rate of convergence of $\mathfrak a^1(x)$ to zero as $x\rightarrow \pm \infty$: : \_[x ]{} \^[3]{} a\^1(x) = -, : \_[x ]{} \^[3]{} a\^1(x) = ’. \[eq:subcasesd1\] For $d\equiv 0,1 \Mod 3$, we include the letter A or B at the bottom of the patterns of zeros to indicate which subcase the corresponding eigenvalue belongs to, see for example \[fig:GSpatterns,fig:pats84\].
#### Bulk behavior.
The bulk limit of the eigenvalues is obtained by increasing the system size $N$ while keeping the spectral parameter $u$ finite and within the analyticity strip. Importantly, it comes into play in \[sec:fsc\] as the asymptotic behavior at $x = -\infty$ of the scaling functions, defined in . In this limit, $d^1(u)$ and $K(u)$ converge to constants: d\^1\_(u) = 0, K\_(u) = -1. \[eq:bulkb\] Indeed, for finite values of $u$ near the origin, the behavior of $d^1(u)$ is governed by the zero of order $2N+2$ at $u = 0$, so the function is approximately zero for large $N$. In the bulk limit, becomes $K_{\text{bulk}}(u)^2 = 1$, so $K_{\text{bulk}}(u) \in \{+1, -1\}$. For a given pattern of zeros, one can deduce the values of $K_{\text{bulk}}(u)$ from the braid value $K_\infty$ and the number of zeros in the central line of the analyticity strip. For instance, let us consider the eigenvalue of \[fig:eigpatterns2\]. The braid limit is $K_\infty = 2$. On the central line of the analyticity strip (namely $\text{Re}(u) = \frac\pi6$ for $D^1(u)$, corresponding to $\text{Re}(u) =-\frac \pi 6$ for $K_0(u)$), there are three zeros with $\text{Im}(u)>0$. We conclude that $K(-\frac \pi 6)$ is negative and that $K_{\text{bulk}}(u) = -1$. Applying the same logic to the pattern of zeros of \[fig:eigpatterns\] would also yield $K_{\text{bulk}}(u) = -1$.
The same reasoning can be repeated for each eigenvalue. Empirically, we find that every pattern of zeros has the same asymptotic behavior: $K_{\text{bulk}}(u) =
-1$.
Finite-size corrections {#sec:fsc}
-----------------------
The eigenvalues $D(u)$ can be factored into a bulk, a surface and a finite-size correction as \[eq:Dseparation\] D(u) = D\_(u) D\_(u) D\_(u). From , the bulk and surface contributions satisfy the inversion relations D\_(u)D\_(u+) = f\_[-1]{}f\_1, D\_(u)D\_(u+) = , \[eq:inv.identities\] whereas the finite-size correction satisfies D\_(u)D\_(u+) = 1+d\^1(u). \[eq:finitepart\] The solutions to give the bulk and surface free energies and were obtained in [@PRZ06] for generic $\beta$. For $\beta =1$, the right-hand side of the inversion identity for $D_{\text{s}}(u)$ equals $1$ and the solution is $D_{\text{s}}(u)=1$. In this section, we derive the finite-size corrections of $D_{\text{f}}(u)$ in the standard modules $\stan_N^d$. We do so by using the methods developed in [@PK91; @KP92]. The solution holds for any finite excitation and works for all three cases, namely $d \equiv 0,1 \textrm{ mod } 3$ (A), $d \equiv 0,1 \textrm{ mod } 3$(B) and $d \equiv 2 \textrm{ mod } 3$. For convenience, we work with the functions $\mathfrak a^1(x)$, $\mathfrak a^2(x)$, $\mathfrak A^1(x)$ and $\mathfrak A^2(x)$ in terms of which the $Y$-system is symmetric. We also define b(x) = D\_( 3+ 6), so that becomes b(x - 2) b(x + 2) = A\^1(x). \[eq:bbU\] We note that defines $\mathfrak b(x)$ up to a sign. We choose this sign to be such that b\_= D\^1\_.
#### TBA equations.
Let us denote by $w^1, w^2, \dots, w^{t^1}$ the values of $x$ where $\mathfrak a^1(x)=0$ is zero on the half-line $\text{Im}(x)=0$, $\text{Re}(x)>0$. The same pattern is repeated symmetrically on the negative part of the real axis: $\mathfrak a^1(\pm w^i) = 0$, $i = 1, \dots, t^1$. Likewise for $\mathfrak a^2(x)$, we denote the positions of its zeros on the positive real $x$ axis by $x^1, x^2, \dots, x^{t^2}$. The pattern is repeated symmetrically on the negative part of the real axis: $\mathfrak a^2(\pm x^j) = 0$, $j = 1, \dots, t^2$. For convenience, we label the zeros such that $w^1{\geqslant}w^2{\geqslant}\dots
{\geqslant}w^{t^1}$ and $x^1> x^2 > \dots> x^{t^2}$, recalling that the zeros of $\mathfrak a^2(x)$ are all distinct whereas those of $\mathfrak a^1(x)$ can be twofold degenerate.
We introduce the finite-size correction functions $\ell^1(x)$ and $\ell^2(x)$ by writing
$$\begin{aligned}
{2}
\mathfrak a^1(x) &= \tanh^{2N+2}(\tfrac x 2)\cdot \bigg(\prod_{i = 1}^{t^1} \tanh(\tfrac{x-w^i}2)\tanh(\tfrac{x+w^i}2) \bigg)\cdot \ell^1(x),\\
\mathfrak a^2(x) &= \bigg(\prod_{j = 1}^{t^2} \tanh(\tfrac{x-x^j}2)\tanh(\tfrac{x+x^j}2) \bigg)\cdot \ell^2(x).\end{aligned}$$
The functions $\ell^1(x)$, $\ell^2(x)$, $\mathfrak A^1(x)$ and $\mathfrak A^2(x)$ are analytic and non-zero in their corresponding analyticity strips, and satisfy \[eq:Lsystem\] \^1(x-2)\^1(x+2) = A\^2(x), \^2(x-2)\^2(x+2) = A\^1(x). From , $\ell^1(x)$, $\ell^2(x)$, $\mathfrak A^1(x)$ and $\mathfrak A^2(x)$ have constant asymptotics for $d \equiv 2 \textrm{ mod } 3$. One can therefore define the Fourier transforms of their logarithmic derivatives. From , for $d \equiv 0,1 \textrm{ mod }3$, $\ell^2(x)$, $\mathfrak A^1(x)$ and $\mathfrak A^2(x)$ also have non-zero asymptotics, but $\ell^1(x)$ does not. According to , $\ell^1(x)$ behaves respectively as $\eE^{-2|x|/3}$ and $\eE^{-4|x|/3}$ as $x\rightarrow \pm \infty$ for the subcases A and B, so we instead define the Fourier transform of the second logarithmic derivative. To treat $d \equiv 0,1,2 \textrm{ mod } 3$ simultaneously, we consider the second logarithmic derivative of $\ell^1(x)$ in all cases:
$$\begin{aligned}
{2}
&\displaystyle L^1(k) = \frac{1}{2 \pi} \int_{-\infty}^\infty \dd x\, \big(\ln \ell^1(x)\big)''\eE^{-\ir k x}, \qquad &&\displaystyle \big(\ln \ell^1(x)\big)'' = \int_{-\infty}^\infty \dd k\, L^1(k)\, \eE^{\ir k x},\\[0.15cm]
&\displaystyle L^2(k) = \frac{1}{2 \pi} \int_{-\infty}^\infty \dd x\, \big(\ln \ell^2(x)\big)'\eE^{-\ir k x}, \qquad &&\displaystyle \big(\ln \ell^2(x)\big)' = \int_{-\infty}^\infty \dd k\, L^2(k)\, \eE^{\ir k x},\\[0.15cm]
&\displaystyle A^n(k) = \frac{1}{2 \pi} \int_{-\infty}^\infty \dd x\, \big(\ln \mathfrak A^n(x)\big)'\eE^{-\ir k x}, \qquad &&\displaystyle \big(\ln \mathfrak A^n(x)\big)' = \int_{-\infty}^\infty \dd k\, A^n(k)\, \eE^{\ir k x}, \qquad n=1,2.\end{aligned}$$
Applying the Fourier transform to , the reverse transform and then integrating with respect to $x$, we obtain the TBA equations:
$$\begin{aligned}
{2}
\ln \mathfrak a^1(x) &= \ln \tanh^{2N+2}(\tfrac x 2) + \sum_{i = 1}^{t^1} \ln \Big(\!-\!\tanh(\tfrac{x-w^i}2)\tanh(\tfrac{x+w^i}2) \Big) + K \ast \ln \mathfrak A^2 + \phi\, x + \phi^1,\\
\ln \mathfrak a^2(x) &= \sum_{j = 1}^{t^2} \ln \Big(\!-\!\tanh(\tfrac{x-x^j}2)\tanh(\tfrac{x+x^j}2) \Big) + K \ast \ln \mathfrak A^1 + \phi^2,\end{aligned}$$
where $\phi$, $\phi^1$ and $\phi^2$ are the integration constants. The kernel $K(x)$ is given by \[eq:kernel\] K(x) = and the convolution of two functions is defined as \[eq:convolution\] (f g) (x) = \_[-]{}\^y f(x-y)g(y) = \_[-]{}\^y f(y)g(x-y). The constants $\phi$, $\phi^1$ and $\phi^2$ are evaluated below from the braid limits.
#### Scaling TBA equations.
In the non-linear integral equations, the dependence on $N$ appears only algebraically, in the function $\ln \tanh^{2N+2}(\frac x 2)$. If $x$ is of order $\ln N$ and $N$ is large, this function has the following behavior: \_[N ]{} \^[2N+2]{}(12(x + N)) = (-4 \^[-x]{}). To compute the finite-size corrections, we assume that the following scaling limits also exist: \[eq:scalingfunctions\] a\^n(x) = \_[N ]{} a\^n( (x + N)), A\^n(x) = \_[N ]{} A\^n( (x + N)), n = 1,2. Because the patterns of zeros are symmetric in the upper and lower parts of the complex $u$ plane, the functions $\mathfrak a^n(x)$ and $\mathfrak A^n(x)$ are even in $x$. In , the limits are therefore independent of the choice of the signs $\pm$. Let us denote the zeros of $\mathsf a^1(x)$ and $\mathsf a^2(x)$ by $z^i$ and $y^j$, namely z\^i = w\^i - N, y\^j = x\^j - N. In the scaling limit, the non-linear equations become
\[eq:scalNLIE\] $$\begin{aligned}
{2}
\ln \mathsf a^1(x) &= -4 \,\eE^{-x} + \sum_{i=1}^{t^1} \ln\big(\!-\!\tanh (\tfrac{x-z^i}2) \big) + K \ast \ln \mathsf A^2 + \phi\, x + \phi^1,\\
\ln \mathsf a^2(x) &= \sum_{j=1}^{t^2} \ln\big(\!-\!\tanh (\tfrac{x-y^j}2) \big)+ K \ast \ln \mathsf A^1 + \phi^2.\end{aligned}$$
#### Evaluation of the constants.
For $d \equiv 2 \textrm{ mod } 3$, we fix the branch cuts of the logarithms by fixing a\^1(x) 3, a\^2(x) 2, (-(2)) . With this choice, the constants $\phi^1$ and $\phi^2$ are evaluated using the braid limit. Indeed, using 1[2]{}\_[-]{}\^ = 12, we find $K \ast \ln \mathsf A^1 \xrightarrow{x \rightarrow \infty} \ln 2$, $K \ast \ln \mathsf A^2 \xrightarrow{x \rightarrow \infty} \ln 3$ and d 2 3: = 0, \^1 = -t\^1, \^2 = -t\^2. \[eq:const2\]
For $d \equiv 0,1 \textrm{ mod } 3$(A), according to , the large $x$ behavior of $\mathfrak a^1(x)$ and $\mathfrak a^2(x)$ is given by a\^1(x) - \^[-3]{}, 1+a\^2(x) \^[-3]{}, \[eq:largexbehavior1\] with $\kappa>0$. The branches of the logarithms are fixed using the convention a\^1(x) -3 + + , a\^2(x) . The braid behavior of $K \ast \ln \mathsf A^2$ can be evaluated explicitly: K A\^2 \_[-]{}\^y = - 3. We find that the constants are d 0,1 3 ():= 0, \^1 = -(t\^1-1), \^2 = -(t\^2-1). \[eq:const01A\]
Finally, for $d \equiv 0,1 \textrm{ mod } 3$ subcase B, according to , the large $x$ behavior of the eigenvalues is a\^1(x) ’ \^[-3]{}, 1+a\^2(x) ’ \^[-3]{}, with $\kappa'>0$. The branches of the logarithms are fixed using the conventions a\^1(x) -3 + ’, a\^2(x) , the braid behavior of $K \ast \ln \mathsf A^2$ is K A\^2 \_[-]{}\^y = ’ - 3 and the integration constants are evaluated to d 0,1 3 ():= 0, \^1 = -t\^1, \^2 = -(t\^2-1). \[eq:const01B\]
#### Finite-size corrections.
To apply the Fourier transform and its inverse to , one needs to remove the zeros of $\mathfrak b(x)$ on the real $x$ axis by dividing by $\prod_{j = 1}^{t^2} \tanh(\tfrac{x-x^j}2)\tanh(\tfrac{x+x^j}2)$. Because $\mathfrak b(x)$ has constant asymptotics, a single derivative of its logarithm is required. The result is b(x) = \_[j = 1]{}\^[t\^2]{} (-(2)(2) ) + K A\^1 + where $\psi$ is the integration constant. With the branch choices b(x) {
[cl]{} 0 &d 0,1 3,\
2 + &d 2 3,
. the constant $\psi$ is evaluated using the braid limit and found to be $-\ir \pi t^2$ and $-\ir\pi (t^2-1)$ for $d \equiv 0,1 \textrm{ mod } 3$ and $d \equiv 2 \textrm{ mod } 3$ respectively. Because $\psi$ is independent of $N$, it does not contribute to the finite-size corrections. These are written in terms of the scaling functions as follows: $$\begin{aligned}
{2}
\ln \mathfrak b(x) - \psi &= \sum_{j = 1}^{t^2} \ln \Big(\!-\!\tanh(\tfrac{x-x^j}2)\tanh(\tfrac{x+x^j}2) \Big)
\nonumber\\&\hspace{0.8cm}
+ \frac{1}{2\pi} \int_{-\ln N}^\infty \dd y \Big( \frac{\ln\mathfrak A^1(y+\ln N)}{\cosh(x-y-\ln N)}+\frac{\ln\mathfrak A^1(-y-\ln N)}{\cosh(x+y+\ln N)}\Big) \nonumber\\
& \simeq -\frac{2 \cosh x}N \bigg(2 \sum_{j=1}^{t^2}\eE^{-y^j}- \frac1\pi \int_{-\infty}^\infty \dd y\, \eE^{-y} \ln \mathsf A^1(y)\bigg).\label{eq:FSC1}\end{aligned}$$
#### Zeros of $\boldsymbol{\mathfrak a^1(x)}$ and $\boldsymbol{\mathfrak a^2(x)}$.
The next step is to rewrite $\eE^{-y^j}$ in terms of integrals involving the scaling functions. From , we find that a\^2(x\^j) = 0 a\^1(x\^j-2) = -1 a\^1(y\^j-2) = -1. Taking the logarithm and using , we find $$\begin{aligned}
{2}
(2 k^j-1)\pi &= \ir \ln \mathsf a^1 (y^j - \tfrac{\ir \pi }2) \nonumber \\&
= 4\, \eE^{-y^j} + \ir \sum_{i=1}^{t^1} \ln\big(\!-\!\tanh \tfrac12(y^j-z^i- \tfrac{\ir\pi}2 )\big) + \frac1{2\pi} \int_{-\infty}^\infty \dd y\, \frac{\ln \mathsf A^2(y)}{\sinh(y-y^j)} + \ir \phi^1 \label{eq:integerK}\end{aligned}$$ where the $k^j$ are integers. Isolating $\eE^{-y^j}$ from this equation and replacing it in produces an expression for the finite-size corrections wherein the zeros of $\mathsf a^1(x)$ and $\mathsf a^2(x)$ appear in terms of the expression $\ln\big(\!-\!\tanh \tfrac12(y^j-z^i- \tfrac{\ir\pi}2 )\big)$. We can also write these in integral form: a\^1(w\^i) = 0 a\^2(w\^i-2) = -1 a\^2( z\^i-2) = -1 and $$\begin{aligned}
{2}
(2 \ell^i-1)\pi &= \ir \ln \mathsf a^2 (z^i - \tfrac{\ir \pi}2)
= \ir \sum_{j=1}^{t^2} \ln \big(\!-\!\tanh \tfrac12(z^i-y^j -\tfrac{\ir \pi}2)\big) + \frac{1}{2 \pi} \int_{-\infty}^\infty \dd y \frac{\ln \mathsf A^1(y)}{\sinh(y-z^i)}+ \ir\phi^2
\label{eq:integerL}\end{aligned}$$ where the $\ell^i$ are integers. From our choice of branches for the logarithms, we have \[eq:tanh.id\] (-12(z\^i-y\^j - 2) ) = - - (-12(y\^j-z\^i - 2) ) which we use to isolate the expression $\ln\big(\!-\!\tanh \tfrac12(y^j-z^i- \tfrac{\ir\pi}2 )\big)$ in . We obtain $$\begin{aligned}
{2}
\ln \mathfrak b(x) - \psi \simeq -\frac{2 \pi \cosh x}{N} \bigg( &\sum_{j=1}^{t^2}(k^j-\tfrac12- \tfrac{\ir\phi^1}{2 \pi}) + \sum_{i=1}^{t^1}(\ell^i-\tfrac12 + \tfrac {t^2}2- \tfrac{\ir\phi^2}{2 \pi}) - \frac 1 {\pi^2} \int_{-\infty}^\infty \dd y \,\eE^{-y} \ln \mathsf A^1(y) \nonumber\\
& - \frac{1}{4\pi^2} \sum_{i=1}^{t^1}\int_{-\infty}^{\infty} \dd y\, \frac{\ln \mathsf A^1(y)}{\sinh(y-z^i)} - \frac{1}{4\pi^2} \sum_{j=1}^{t^2}\int_{-\infty}^{\infty} \dd y\, \frac{\ln \mathsf A^2(y)}{\sinh(y-y^j)}\bigg).\label{eq:FSC2pap}\end{aligned}$$
#### Dilogarithm technique.
To evaluate , we consider the integral J = \_[-]{}\^y ((a\^1)’ A\^1 -|a\^1| (A\^1)’ ) + \_[-]{}\^y ((a\^2)’ A\^2 -|a\^2| (A\^2)’ ) where $\ln |\mathsf a^1|$ and $\ln |\mathsf a^2|$ are real for all $x$ and are thus given by \[eq:ln|a|\] |a\^n|(x) = a\^n(x) + \^n(x), n = 1,2. Here, the $\theta^n(x)$ are step functions defined for $x \in \mathbb R$. Starting from $x=+\infty$ and moving to the left on the $x$ axis, the $\theta^n(x)$ decrease by $\ir \pi$ each time a zero of the corresponding type is crossed ($z^i$ for $\theta^1(x)$ and $y^j$ for $\theta^2(x)$). The values at the right endpoints are $\theta^1(x) = 0$ for $x > z_1$ and $\theta^2(x) = 0$ for $x > y_1$, consistent with our choice of branches for the logarithms.
The integral $\mathcal J$ can be evaluated in two ways. For the first, one uses the non-linear integral equations and the symmetries of $K(x)$ to obtain $$\begin{aligned}
{2}
\mathcal J &= 4\int_{-\infty}^\infty \dd y \, \eE^{-y} \big(\ln \mathsf A^1 + (\ln \mathsf A^1)'\big) + \int_{-\infty}^\infty \dd y \sum_{i=1}^{t^1} \Big[\ln\big(\!-\! \tanh (\tfrac{y-z^i}2)\big) \Big]' \ln \mathsf A^1 \nonumber\\
& - \int_{-\infty}^\infty \dd y \Big(\sum_{i=1}^{t^1} \ln \big(\!-\!\tanh (\tfrac{y-z^i}2)\big) + \phi^1 - \theta^1(y) \Big) \big(\ln \mathsf A^1\big)' + \int_{-\infty}^\infty \dd y \sum_{j=1}^{t^2} \Big[\ln \big(\!-\!\tanh (\tfrac{y-y^j}2)\big) \Big]' \ln \mathsf A^2 \nonumber\\
& - \int_{-\infty}^\infty \dd y \Big(\sum_{j=1}^{t^2} \ln \big(\!-\!\tanh (\tfrac{y-y^j}2)\big) + \phi^2- \theta^2(y) \Big) \big(\ln \mathsf A^2\big)'.\end{aligned}$$ The integrals involving derivatives of $\ln \mathsf A^1$ and $\ln \mathsf A^2$ are transformed using integration by parts. For each one, it can be argued using the non-linear integral equations that the surface terms are zero. This yields J = 8 \_[-]{}\^y \^[-y]{} A\^1(y) + 2 \_[i=1]{}\^[t\^1]{} \_[-]{}\^y + 2\_[j=1]{}\^[t\^2]{} \_[-]{}\^y , which is precisely the combination of integrals needed to compute the finite-size corrections: b(x) - - ( \_[j=1]{}\^[t\^2]{}(k\^j-12- ) + \_[i=1]{}\^[t\^1]{}(\^i-12 + 2- ) - ). \[eq:lnbalmostfinal\] The second way of performing the integrals consists in changing the integration variable from $y$ to $\mathsf a$. For the integral involving $\mathsf a^1(x)$, we obtain $$\begin{aligned}
{2}
\int_{-\infty}^\infty \dd y \Big((&\ln \mathsf a^1)' \ln \mathsf A^1 -\ln |\mathsf a^1| (\ln \mathsf A^1)' \Big) = \int_{-\infty}^\infty \dd y\, \frac{\dd \mathsf a^1}{\dd y} \,\bigg( \frac{\ln(1+ \mathsf a^1)}{\mathsf a^1} - \frac{\ln |\mathsf a^1|}{1+ \mathsf a^1} \bigg)\nonumber\\
& = \bigg[\int_{-\infty}^{z_{t^1}}+\int_{z_{t^1}}^{z_{t^1-1}}+ \dots + \int_{z_2}^{z_1}+ \int_{z_1}^{\infty}\bigg] \dd y\, \frac{\dd \mathsf a^1}{\dd y}\, \bigg( \frac{\ln(1+ \mathsf a^1)}{\mathsf a^1} - \frac{\ln |\mathsf a^1|}{1+ \mathsf a^1} \bigg) \nonumber\\
& = \bigg[\int_{0}^{0}+\int_{0}^{0}+ \dots + \int_{0}^{0}+ \int_{0}^{\mathsf a^1(\infty)}\bigg] \dd \mathsf a \,\Big( \frac{\ln(1+ \mathsf a)}{\mathsf a} - \frac{\ln |\mathsf a|}{1+ \mathsf a} \Big) \nonumber\\
& = 2L_+(\mathsf a^1(\infty)) = 2 L\big(\tfrac {\mathsf a^1(\infty)}{1+\mathsf a^1(\infty)}\big)\end{aligned}$$ where the Rogers dilogarithm functions are given by
$$\begin{aligned}
{2}
L(x)&=-\frac12 \int_{0}^x \dd y \, \Big(\frac{\ln (1-y)}y + \frac{\ln y}{1-y} \Big),\\
L_+(x)&= \frac12 \int_0^x \dd y \, \Big(\frac{\ln (1+y)}y - \frac{\ln y}{1+y}\Big)=L\Big(\frac{x}{1+x}\Big).\end{aligned}$$
Recalling that $\mathsf A^2(x) = (1 + \mathsf a^2(x))^2$, the integral involving $\mathsf a^2(x)$ is computed with the same arguments: $$\begin{aligned}
{2}
\int_{-\infty}^\infty \dd y \Big((&\ln \mathsf a^2)' \ln \mathsf A^2 -\ln |\mathsf a^2| (\ln \mathsf A^2)' \Big) = 2\int_{-\infty}^\infty \dd y\, \frac{\dd \mathsf a^2}{\dd y} \,\bigg( \frac{\ln(1+ \mathsf a^2)}{\mathsf a^2} - \frac{\ln |\mathsf a^2|}{1+ \mathsf a^2} \bigg)\nonumber\\
& = 2\bigg[\int_{-\infty}^{y_{t^2}}+\int_{y_{t^2}}^{y_{t^2-1}}+ \dots + \int_{y_2}^{y_1}+ \int_{y_1}^{\infty}\bigg] \dd y\, \frac{\dd \mathsf a^2}{\dd y}\, \bigg( \frac{\ln(1+ \mathsf a^2)}{\mathsf a^2} - \frac{\ln |\mathsf a^2|}{1+ \mathsf a^2} \bigg) \nonumber\\
& = 2\bigg[\int_{-1}^{0}+\int_{0}^{0}+ \dots + \int_{0}^{0}+ \int_{0}^{\mathsf a^2(\infty)}\bigg] \dd \mathsf a \,\Big( \frac{\ln(1+ \mathsf a)}{\mathsf a} - \frac{\ln |\mathsf a|}{1+ \mathsf a} \Big) \nonumber\\
& = 4L_+(\mathsf a^2(\infty)) +4L(1) = 4 L\big(\tfrac {\mathsf a^2(\infty)}{1+\mathsf a^2(\infty)}\big) + 4 L(1).\end{aligned}$$ Putting these results together, we find J = 4 L(1) + 2 L() + 4 L() = {
[ll]{} K\_[-1]{}(3) = 0 &d 0,1 3,\
K\_[-1]{}(0) =3 &d 2 3,
. where the integral $\mathcal K_{-1}(\gamma)$ is computed in \[sec:Rogers\]. From , the final result for the finite-size corrections is \[eq:finalfsc\] b(x) - -N ( \_[j=1]{}\^[t\^2]{} k\^j+\_[i=1]{}\^[t\^1]{} \^i + ), = {
[ll]{} - 12t\^1t\^2 & d 0,13,\
- 12(t\^2+t\^1t\^2) & d 0,13,\
- 12(t\^1+t\^2+t\^1t\^2)-16 & d 23,
. where we used the values of $\phi^1$ and $\phi^2$ given in , and . Comparing with specialised to $c=0$, we find that the conformal dimension of the corresponding conformal state is the content of the parenthesis, namely: \[eq:DelKL\] = \_[j=1]{}\^[t\^2]{}k\^j+\_[i=1]{}\^[t\^1]{}\^i + .
Solution for the ground states {#sec:cfgs}
------------------------------
For the ground state eigenvalue in $\stan_N^d$, our experimentations on small system sizes reveal that the corresponding pattern of zeros is characterised by t\^1 = {
[ll]{} 3 & d 03,\
3 & d 13,\
3 & d 23,
.t\^2 = {
[cl]{} 3 & d 03,\
3 & d 13,\
3 & d 23,
. and by $\frac{N-d}2$ (pairs of) zeros lying on its boundary lines for $D^1(u)$, see \[fig:GSpatterns\]. Each single zero of $D^2(u)$ is joined by a pair of single zeros, sitting at the same height on the edges of the analyticity strip.
For $d \equiv 0 \text{ mod } 3$ and $d \equiv 1 \text{ mod } 3$, we find empirically that the ground state eigenvalue respectively belongs to the subcase A and B. The integers $k^j$ and $\ell^i$ are not fixed by the technique used in \[sec:fsc\]. They can instead be estimated from and using exact diagonalisation on small system sizes. For the ground state of $\stan_N^d$, we find, again empirically, that the $k^j$ and $\ell^i$ are given by k\^j = {
[cl]{} j-1 & d 03,\
j & d 13,\
j & d 23,
. \^i = i. \[fig:GSpatterns\] features the patterns of zeros for the ground states for $N=14,15$.[^2] For $d=0$ and $d=1$, the second analyticity strip is colored in gray, indicating that $D^2(u)=0$ for all $u$ for the Razumov-Stroganov eigenvalue. The integers $k^j$ and $\ell^i$ given there are those obtained using the method described above. To correctly compute the conformal weight from , we recall that $t^1$ counts the zeros in the second analyticity strip, and $t^2$ counts the zeros in the first. This is because, up to prefactors and shifts in the arguments, $\mathfrak a^1(x)$ corresponds to $D^2(u)$ whereas $\mathfrak a^2(x)$ corresponds to $D^1(u)$.
Carrying out the sum in with the data given above, we find in all three cases ($d \equiv 0,1,2 \Mod 3$) that the conformal weight of the ground state of $\stan_N^d$ is = 6 = \_[1,d+1]{}.
Single- and double-column diagrams {#sec:colconfs}
----------------------------------
Our classification of the patterns of zeros for the excited states in \[sec:excitations\] uses the column diagrams introduced in [@PR07]. Here we recall the definitions and some basic results in a self-consistent manner.
#### Single-column diagrams.
A single-column diagram in the set $\Aone{M}{m}$ is a vertical array of $M$ sites of which $m$ are occupied and the other $M-m$ are unoccupied. We draw occupied and unoccupied sites in black and white respectively, as in the example of \[fig:columnconf\]. The sites are assigned the height labels $1, \dots, M$ starting from the bottom. The [*signature*]{} $S = \{S_1, \dots, S_m\}$ of a single-column diagram in $\Aone{M}{m}$ is the list of the heights of its occupied sites in decreasing order. The [*energy*]{} of a single-column diagram is $E = \sum_{j=1}^m S_j$ and its [*weight*]{} is $q^{E}$. An example is given in the left panel of \[fig:columnconf\].
$$\psset{unit=0.5}
\begin{pspicture}[shift=-3.9](0,0)(1,8)
\pspolygon[fillstyle=solid,fillcolor=lightyellow](0,0)(1,0)(1,8)(0,8)
\rput(0.5,0.5){\blackcircle}\rput(-0.5,0.5){$_1$}
\rput(0.5,1.5){\blackcircle}\rput(-0.5,1.5){$_2$}
\rput(0.5,2.5){\whitecircle}\rput(-0.5,2.5){$_3$}
\rput(0.5,3.5){\blackcircle}
\rput(0.5,4.5){\blackcircle}\rput(-0.5,5.2){$\vdots$}
\rput(0.5,5.5){\whitecircle}
\rput(0.5,6.5){\blackcircle}
\rput(0.5,7.5){\whitecircle}\rput(-0.5,7.5){$_M$}
\end{pspicture}
\quad \rightarrow \quad
{\psun
\begin{pspicture}[shift=-1.2](-0.0,-2.8)(1.0,0.5)
\psline{-}(0,-2.8)(0,0)(1,0)(1,-2.8)
\psdots[dotsize=0.09cm](0,-0.3)(0,-0.9)(0,-1.8)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)(0.5,-1.5)(0.5,-2.1)(0.5,-2.4)
\psdots[dotsize=0.09cm](1,-0.3)(1,-0.9)(1,-1.8)
\end{pspicture}
}
\ \hspace{2cm}\
\begin{pspicture}[shift=-2.9](0,0)(2,6)
\pspolygon[fillstyle=solid,fillcolor=lightyellow](0,0)(2,0)(2,6)(0,6)
\rput(0.5,0.5){\blackcircle}\rput(-0.5,0.5){$_1$}
\rput(0.5,1.5){\whitecircle}\rput(-0.5,1.5){$_2$}
\rput(0.5,2.5){\whitecircle}\rput(-0.5,2.5){$_3$}
\rput(0.5,3.5){\blackcircle}\rput(-0.5,4.2){$\vdots$}
\rput(0.5,4.5){\blackcircle}
\rput(0.5,5.5){\whitecircle}\rput(-0.5,5.5){$_M$}
\rput(1.5,0.5){\blackcircle}
\rput(1.5,1.5){\whitecircle}
\rput(1.5,2.5){\blackcircle}
\rput(1.5,3.5){\blackcircle}
\rput(1.5,4.5){\whitecircle}
\rput(1.5,5.5){\blackcircle}
\psline[linewidth=0.5pt,linecolor=gray]{-}(0.5,4.5)(1.5,5.5)
\psline[linewidth=0.5pt,linecolor=gray]{-}(0.5,3.5)(1.5,3.5)
\psline[linewidth=0.5pt,linecolor=gray]{-}(0.5,0.5)(1.5,2.5)
\end{pspicture}
\quad \rightarrow \quad
{\psun
\begin{pspicture}[shift=-1.8](-0.4,-4.0)(1.4,0.5)
\psline{-}(0,-4.0)(0,0)(1,0)(1,-4.0)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-1.8)(0.5,-3.6)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0,-3.0)(1,-3.0)
\psdots[dotsize=0.09cm](0,-0.6)(0,-1.2)(0,-2.4)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)(0.5,-2.4)
\psdots[dotsize=0.09cm](1,-0.6)(1,-1.2)(1,-2.4)
\end{pspicture}
}$$
The generating function $\ang{M}{m}$ is defined as the sum of the weights over the single-column diagrams in $\Aone{M}{m}$. It is a polynomial in $q$ and can be computed using the following recursive argument. By removing the lowest site of a configuration in $\Aone{M}{m}$, we obtain a new configuration containing $M-1$ sites, with either $m$ or $m-1$ occupied sites depending on whether the lowest site of the original configuration was occupied. Moreover, the labels of the new configuration range from $2$ to $M$. The generating function then satisfies the recursion relation = q\^m (+). Here, the factor $q^m$ corrects for the energy difference arising due to the relabelling of the sites. For the second term, it also includes a factor of $q^1$ for the energy contribution of the lowest site of the original single-column diagram. Along with the boundary conditions $\ang{M}{0}=1$ and $\ang{M}{M}=q^{\frac12M(M+1)}$, the recursion relation fixes $\ang{M}{m}$ completely to \[eq:AMm\] = q\^[12m(m+1)]{} where the [*Gaussian polynomial*]{} (or [*$q$-binomial*]{}) is defined as \[eq:Gausspoly\] = , = . The factor $q^{\frac12m(m+1)}$ in is identified as the energy of the single-column diagram in $\Aone{M}{m}$ with all occupied sites at the bottom. We note that if the height labels of the single-column diagram are $a, \dots, M+a-1$ instead of $1, \dots, M$, then the energies of all the eigenvalues are shifted by $m(a-1)$ and the generating function is $q^{m(a-1)} \ang{M}{m} = q^{\frac12m(m+2a-1)}\smallqbinom{M}{m}$.
A single-column diagram is mapped to a pattern of zeros of $D^1(u)$ using the following rule: (i) an occupied site at height $j$ produces a zero of order one in the center of the analyticity strip; (ii) an unoccupied site produces two zeros, each of order one, lying on the edges of the analyticity strip. This map is illustrated in the left panel of \[fig:columnconf\].
#### Double-column diagrams.
We define $\Adom{M}{m}{n}$ to be the set of double-column diagrams that satisfy the condition of [*dominance*]{}. Such diagrams are made of two single-column configurations of $M$ sites drawn side by side, with respectively $m$ and $n$ occupied sites in the left and right columns. An example is given in the right panel of \[fig:columnconf\]. We respectively denote by $L$ and $R$ the signatures of the left and right column. The energy of a double-column diagram is $E=\sum_{i=1}^m L_i + \sum_{j=1}^n R_j$ and its weight is $q^E$.
A double-column configuration in $\Adom{M}{m}{n}$ satisfies the condition of dominance if \[eq:dominance\] L\_i R\_i, j = 1, …, m. This of course presupposes that $0{\leqslant}m{\leqslant}n{\leqslant}M$. The criterion can be translated in terms of a diagrammatic rule for the double-column diagram. One draws $m$ non intersecting lines pairing the top $m$ occupied sites of each column starting from the top. The double-column diagram satisfies dominance if the slope of each line is non-negative.
The generating function for $\Adom{M}{m}{n}$ is denoted $\ang{M}{m,n}$. It is the sum of the weights of the diagrams in $\Adom{M}{m}{n}$. By removing the lowest row of a given configuration in $\Adom{M}{m}{n}$, we obtain a new configuration of height $M-1$ in which the occupation numbers are $(m,n)$, $(m-1,n)$, $(m,n-1)$ or $(m-1,n-1)$. Crucially, the resulting column configuration satisfies dominance in all cases. This is easy to see either from the definition or from the diagrammatic rule. As a result, the generating function satisfies the recursion relation = q\^[m+n]{}(+++), where the prefactor $q^{m+n}$ compensates for the relabelling of the height labels and the energy contribution of the lowest sites. With the conditions = 1, = q\^[12M(M+1)]{},= 0,= q\^[12M(M+1)+12N(N+1)]{}, the recursion relation fixes the generating functions entirely. The result is \[eq:Amn\] = q\^[12m(m+1)+12n(n+1)]{} where $\smallqnarayana{M}{m}{n}$ are the generalised $q$-Narayana numbers =q\^[-M+n]{}(-). \[eq:qnar\] The factor $q^{\frac12m(m+1)+\frac12n(n+1)}$ in is the weight of the double-column diagram in $\Adom{M}{m}{n}$ with minimal energy.
A double-column diagram is mapped to a pattern of zeros of $D^2(u)$ using the following rule: (i) if both sites at height $j$ are occupied, a zero of order two sits in the center of the analyticity strip; (ii) if both sites at height $j$ are unoccupied, a pair of double zeros is inserted on the edges of the analyticity strip; (iii) if one site is occupied and the other is not, this yields three zeros of order one, one of which is inserted in the center of the analyticity strip whereas the two others are inserted on the edges. An example is given in the right panel of \[fig:columnconf\].
Solution for all the eigenvalues {#sec:excitations}
--------------------------------
In this section, we find closed expressions for the finitized spectrum generating functions $\ZdN$, defined by = \_ q\^, where $q$ is the modular nome. We note that the sum is over the finite set of eigenvalues in $\stan_N^d$, characterized by their patterns of zeros, whereas $\Delta$ is the conformal weight of the corresponding pattern of zeros in the scaling limit.
#### Selection rules for the patterns of zeros.
Our computation of $\ZdN$ is based on a conjecture for the selection rules for the eigenvalues in $\stan_N^d$ which we now formulate. These empirical rules give the patterns of zeros and the values taken by the integers $k^j$ and $\ell^i$ for each eigenvalue. Similar selection rules for the model of critical dense polymers on the strip were conjectured in [@PR07] and later proven in [@MD11]. The conjectured selection rules given below are supported by data produced with our computer implementation of the transfer matrices for $N {\leqslant}12$. For a given eigenvalue, our program outputs the corresponding patterns of zeros of $D^1(u)$ and $D^2(u)$. To illustrate, the data corresponding to all the eigenstates in $\stan_8^2$ and $\stan_8^4$ is given in \[fig:pats82,fig:pats84\].
The selection rules for the patterns of zeros are described in terms of the single- and double-column diagrams discussed in \[sec:colconfs\]. Let $(\sigma,\sigma')$ be a pair of column diagrams[^3] with $\sigma \in \Aone{M}{m}$ and $\sigma' \in \Adom{L}{n}{\ell}$. We denote the set of such pairs by $\AA{M}{L}{m}{n}{\ell}$. An example is given in \[fig:A42311\]. For $d \equiv 0,1 \Mod 3$, we indicate whether the patterns of zeros of the corresponding set belong to subcases A or B by writing $\AAAB{M}{L}{m}{n}{\ell}{\tA}$ or $\AAAB{M}{L}{m}{n}{\ell}{\tB}$.
$$\begin{aligned}
{2}
&
\psun
\begin{pspicture}[shift=-0.8](-0.4,-2.2)(1.4,0.1)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](1,-0.3)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.9)(0.5,-1.2)
\psdots[dotsize=0.09cm](0,-0.3)
\end{pspicture}
\begin{pspicture}[shift=-0.8](-0.4,-2.2)(1.4,0.1)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0,-0.6)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](1,-0.6)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-1.2)
\end{pspicture} \quad\
\begin{pspicture}[shift=-0.8](-0.4,-2.2)(1.4,0.1)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](1,-0.6)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.9)(0.5,-1.2)
\psdots[dotsize=0.09cm](0,-0.6)
\end{pspicture}
\begin{pspicture}[shift=-0.8](-0.4,-2.2)(1.4,0.1)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0,-0.6)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](1,-0.6)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-1.2)
\end{pspicture} \quad\
\begin{pspicture}[shift=-0.8](-0.4,-2.2)(1.4,0.1)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](1,-0.9)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-1.2)
\psdots[dotsize=0.09cm](0,-0.9)
\end{pspicture}
\begin{pspicture}[shift=-0.8](-0.4,-2.2)(1.4,0.1)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0,-0.6)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](1,-0.6)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-1.2)
\end{pspicture} \quad\
\begin{pspicture}[shift=-0.8](-0.4,-2.2)(1.4,0.1)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](1,-1.2)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)
\psdots[dotsize=0.09cm](0,-1.2)
\end{pspicture}
\begin{pspicture}[shift=-0.8](-0.4,-2.2)(1.4,0.1)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0,-0.6)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](1,-0.6)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-1.2)
\end{pspicture}
\\[0.2cm] &
\psun
\begin{pspicture}[shift=-0.8](-0.4,-2.2)(1.4,0.1)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](1,-0.3)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.9)(0.5,-1.2)
\psdots[dotsize=0.09cm](0,-0.3)
\end{pspicture}
\begin{pspicture}[shift=-0.8](-0.4,-2.2)(1.4,0.1)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.6)(0,-1.2)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)
\psdots[dotsize=0.09cm](1,-0.6)(1,-1.2)
\end{pspicture} \quad\
\begin{pspicture}[shift=-0.8](-0.4,-2.2)(1.4,0.1)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](1,-0.6)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.9)(0.5,-1.2)
\psdots[dotsize=0.09cm](0,-0.6)
\end{pspicture}
\begin{pspicture}[shift=-0.8](-0.4,-2.2)(1.4,0.1)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.6)(0,-1.2)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)
\psdots[dotsize=0.09cm](1,-0.6)(1,-1.2)
\end{pspicture} \quad \
\begin{pspicture}[shift=-0.8](-0.4,-2.2)(1.4,0.1)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](1,-0.9)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-1.2)
\psdots[dotsize=0.09cm](0,-0.9)
\end{pspicture}
\begin{pspicture}[shift=-0.8](-0.4,-2.2)(1.4,0.1)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.6)(0,-1.2)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)
\psdots[dotsize=0.09cm](1,-0.6)(1,-1.2)
\end{pspicture}\quad \
\begin{pspicture}[shift=-0.8](-0.4,-2.2)(1.4,0.1)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](1,-1.2)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)
\psdots[dotsize=0.09cm](0,-1.2)
\end{pspicture}
\begin{pspicture}[shift=-0.8](-0.4,-2.2)(1.4,0.1)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.6)(0,-1.2)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)
\psdots[dotsize=0.09cm](1,-0.6)(1,-1.2)
\end{pspicture}
\\[0.2cm] &
\psun
\begin{pspicture}[shift=-0.8](-0.4,-1.6)(1.4,0.1)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](1,-0.3)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.9)(0.5,-1.2)
\psdots[dotsize=0.09cm](0,-0.3)
\end{pspicture}
\begin{pspicture}[shift=-0.8](-0.4,-1.6)(1.4,0.1)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0,-1.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](1,-1.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)
\end{pspicture} \quad\
\begin{pspicture}[shift=-0.8](-0.4,-1.6)(1.4,0.1)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](1,-0.6)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.9)(0.5,-1.2)
\psdots[dotsize=0.09cm](0,-0.6)
\end{pspicture}
\begin{pspicture}[shift=-0.8](-0.4,-1.6)(1.4,0.1)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0,-1.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](1,-1.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)
\end{pspicture} \quad\
\begin{pspicture}[shift=-0.8](-0.4,-1.6)(1.4,0.1)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](1,-0.9)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-1.2)
\psdots[dotsize=0.09cm](0,-0.9)
\end{pspicture}
\begin{pspicture}[shift=-0.8](-0.4,-1.6)(1.4,0.1)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0,-1.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](1,-1.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)
\end{pspicture} \quad\
\begin{pspicture}[shift=-0.8](-0.4,-1.6)(1.4,0.1)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](1,-1.2)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)
\psdots[dotsize=0.09cm](0,-1.2)
\end{pspicture}
\begin{pspicture}[shift=-0.8](-0.4,-1.6)(1.4,0.1)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0,-1.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](1,-1.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)
\end{pspicture}\end{aligned}$$
For $\stan_8^2$ and $\stan_8^4$, \[fig:pats82,fig:pats84\] reveal that the patterns of zeros of $D^1(u)$ and $D^2(u)$ are encoded by the following sets:
$$\begin{aligned}
{2}
\stan_8^2&: \quad \AA{4}{0}{1}{0}{0}\,\cup\,\AA{4}{1}{3}{0}{0}\,\cup\,\AA{5}{1}{3}{1}{1}\,\cup\,\AA{5}{2}{5}{1}{1} \,\cup\,\AA{6}{2}{5}{2}{2}\,\cup\,\AA{7}{3}{7}{3}{3},\\[0.2cm]
\stan_8^4&:\quad\AAAB{4}{0}{2}{0}{0}{\tB}\,\cup\,\AAAB{4}{1}{4}{0}{0}{\tB}\,\cup\,\AAAB{5}{1}{4}{1}{1}{\tB}\,\cup\,\AAAB{6}{2}{6}{2}{2}{\tB}\,\cup\,\AAAB{5}{1}{4}{0}{1}{\tA} \,\cup\,\AAAB{6}{2}{6}{1}{2}{\tA}.\end{aligned}$$
In general, for $d>1$, we conjecture that the full set of patterns of zeros in $\stan_N^d$ is given by the following sets:
\[eq:SR\] $$\begin{aligned}
{2}
d=3t&: \hspace{0.1cm}\bigcup_{i=0}^{\frac{N-d}2} \bigcup_{j=0}^{\big\lfloor\frac12\big(\frac{N-d}2-i\big)\big\rfloor} \AAAB{\frac{N+t}2 + i}{i+j+t-1}{2(i+j+t)}{i}{i+t-1}{\tA} \hspace{0.1cm} \cup \hspace{0.0cm} \bigcup_{i=0}^{\frac{N-d-4}2} \bigcup_{j=0}^{\big\lfloor\frac12\big(\frac{N-d-4}2-i\big)\big\rfloor} \AAAB{\frac{N+t}2 + i}{i+j+t}{2(i+j+t)+2}{i}{i+t}{\tB}\ ,\label{eq:SR0t}
\\
d=3t+1&: \hspace{0.1cm}\bigcup_{i=0}^{\frac{N-d}2} \bigcup_{j=0}^{\big\lfloor\frac12\big(\frac{N-d}2-i\big)\big\rfloor} \AAAB{\frac{N+t-1}2 + i}{i+j+t-1}{2(i+j+t)}{i}{i+t-1}{\tB}\hspace{0.1cm} \cup \hspace{0.0cm} \bigcup_{i=0}^{\frac{N-d-2}2} \bigcup_{j=0}^{\big\lfloor\frac12\big(\frac{N-d-2}2-i\big)\big\rfloor} \AAAB{\frac{N+t+1}2 + i}{i+j+t}{2(i+j+t)+2}{i}{i+t}{\tA}\ ,\label{eq:SR1t}
\\
d=3t+2&: \hspace{0.1cm}\bigcup_{i=0}^{\frac{N-d}2} \bigcup_{j=0}^{\big\lfloor\frac12\big(\frac{N-d}2-i\big)\big\rfloor} \AA{\frac{N+t}2 + i}{i+j+t}{2(i+j+t)+1}{i}{i+t}\ .\label{eq:SR2t}\end{aligned}$$ For $d = 0$ and $d=1$, and are ill-defined because some indices are negative. In these cases, we instead have the following selection rules: $$\begin{aligned}
{2}
d=0&: \quad \AA{0}{0}{0}{0}{0} \hspace{0.1cm} \cup \hspace{0.1cm} \bigcup_{i=0}^{\frac{N-4}2} \bigcup_{j=0}^{\big\lfloor\frac12\big(\frac{N-4}2-i\big)\big\rfloor} \AAAB{\frac{N}2 + i}{i+j}{2(i+j)+2}{i}{i}{\tB}\ ,
\\
d=1&: \quad\AA{0}{0}{0}{0}{0}\hspace{0.1cm} \cup \hspace{0.1cm} \bigcup_{i=0}^{\frac{N-3}2} \bigcup_{j=0}^{\big\lfloor\frac12\big(\frac{N-3}2-i\big)\big\rfloor} \AAAB{\frac{N+1}2 + i}{i+j}{2(i+j)+2}{i}{i}{\tA}\ .\end{aligned}$$
where $\AA{0}{0}{0}{0}{0}$ is the set that contains a unique element: the pattern of zeros corresponding to the Razumov-Stroganov eigenvalue.
#### Selection rules for the integers.
As part of the conjectured selection rules, we also provide the prescription for the values taken by the numbers $k^j$ and $\ell^i$ for each eigenvalue. For $N{\leqslant}12$, we have obtained these integers for each pattern of zeros by evaluating and with the finite-size spectra produced by our computer implementation. The precision of the approximate values of the $k^j$ and $\ell^i$ obtained in this way is remarkably good even for small system sizes: The error is less than $0.1$ in almost all the cases. In \[fig:pats82,fig:pats84\], the values of these integers are given alongside the corresponding patterns of zeros. The prescription is as follows: for an element in $\AA{M}{L}{m}{n}{\ell}$,
- the heights of the single-column diagram (corresponding to the $k^j$) are labelled from $0$ to $M-1$ for $d\equiv 0,1 \Mod 3\,$ (A), and from $1$ to $M$ for $d\equiv 0,1 \Mod 3\,$ (B) and $d\equiv 2 \Mod 3$;
- the heights of the double-column diagram (corresponding to the $\ell^i$) are labelled from $1$ to $L$ in all cases.
#### Finitized spectrum generating functions.
The conformal weight $\Delta$ corresponding to a given pattern of zeros is given in . The selection rules and the prescription for the integers allow us to write explicit expressions for the finitized spectrum generating functions. These are obtained as sums of the generating functions of the sets of column diagrams given in . To do so, we compute the minimal conformal weights $\Delta_{\textrm{min}}$ for $\AA{M}{L}{m}{n}{\ell}$, $\AAAB{M}{L}{m}{n}{\ell}{\tA}$ and $\AAAB{M}{L}{m}{n}{\ell}{\tB}$ in terms of the energies $E$, $E^{(\tA)}$ and $E^{(\tB)}$ of the corresponding minimal configurations. Using and the above prescription (i) and (ii), we find:
\[eq:Es\] $$\begin{aligned}
{4}
&d \equiv 0,1 \Mod 3\,(\tA):\qquad&& \Delta_{\textrm{min}} = E^{(\tA)}= \tfrac12(m^2+n^2+\ell^2 - m + n+\ell - mn-m\ell),
\label{eq:EA}\\[0.15cm]
&d \equiv 0,1 \Mod 3\,(\tB):\qquad&& \Delta_{\textrm{min}} = E^{(\tB)}= \tfrac12(m^2+n^2+\ell^2 + n+\ell - mn-m\ell), \label{eq:EB}\\[0.15cm]
&d \equiv 2 \Mod 3:\qquad&& \Delta_{\textrm{min}} = E-\tfrac16 = \tfrac12(m^2+n^2+\ell^2- mn-m\ell)-\tfrac16.
\label{eq:justE}\end{aligned}$$
The spectra generating functions $\ZdN$ are then obtained by summing $q^{\Delta_{\textrm{min}}} \smallqbinom{M}{m} \smallqnarayana{L}{n}{\ell}$ over the sets $\AA{M}{L}{m}{n}{\ell}$ given by . In doing so, we note that the indices $i$ and $j$ run over all possible values for which the sets are well-defined. Equivalently, they run over all values such that the $q$-binomials in the generating functions have positive arguments, with the top one larger or equal to the bottom one. We therefore omit the indices of the sums over $i$ and $j$, understanding these sums as running over $\mathbb Z$, with only finitely many contributions. We obtain:
\[eq:char.identities\] $$\begin{aligned}
{3}
&d = 3t: \quad
&&\ZdN
= q^{\frac{d(d-1)}6} \sum_{i,j} q^{i^2+2j(i+j)+t(2i+3j)}
\qbinom{\frac{N+t}2+i}{2(i+j+t)}\qnarayana{i+j+t-1}{i}{i+t-1}\label{eq:char0}
\\
& &&\hspace{1.25cm} +
q^{\frac{(d+3)(d+4)}6} \sum_{i,j} q^{i(i+3)+2j(i+j+2)+t(2i+3j)}
\qbinom{\frac{N+t}2+i}{2(i+j+t)+2}\qnarayana{i+j+t}{i}{i+t},\nonumber
\\[0.3cm]
&d = 3t+1:\quad
&&\ZdN = q^{\frac{d(d-1)}6} \sum_{i,j} q^{i(i+1)+2j(i+j+\frac12)+t(2i+3j)}
\qbinom{\frac{N+t-1}2+i}{2(i+j+t)}\qnarayana{i+j+t-1}{i}{i+t-1}\label{eq:char1}
\\
& &&\hspace{1.25cm} +
q^{\frac{(d+1)(d+2)}6}\sum_{i,j} q^{i(i+2)+2j(i+j+\frac32)+t(2i+3j)}
\qbinom{\frac{N+t+1}2+i}{2(i+j+t)+2}\qnarayana{i+j+t}{i}{i+t},\nonumber
\\[0.3cm]
&d = 3t+2 : \quad
&&\ZdN = q^{\frac{d(d-1)}6} \sum_{i,j} q^{i(i+1)+2j(i+j+1)+t(2i+3j)}
\qbinom{\frac{N+t}2+i}{2(i+j+t)+1}\qnarayana{i+j+t}{i}{i+t}.\label{eq:char2}\end{aligned}$$
For $d=0$ and $d=1$, and must be modified so that the first line is replaced by $1$.
In \[sec:charids\], we show using $q$-binomial identities that these complicated expressions for $\ZdN$ can be simplified to yield the finitized Kac characters, namely: =q\^[d(d-1)/6]{}(-q\^[d+1]{}) = \_[1,d+1]{}(q). This holds for $d\equiv 0,1,2 \textrm{ mod } 3$.
#### Scaling limit.
The behavior of the finitized character in the scaling limit is easily extracted using 1[(q)\_]{} which holds for fixed $d$. We recall that $(q)_\infty$ is defined in . This yields \_[1,d+1]{}(q).
As a final remark, we note that provides expressions for finitized characters of the irreducible Virasoro representations $\mathsf I_{1,s}$. This is trivially true for $d \equiv 2 \textrm{ mod } 3$ because $\chit_{1,d+1} \simeq \mathsf I_{1,d+1}$ is already irreducible. For $d \equiv 0 \textrm{ mod } 3$, the first and second lines of are respectively finitized characters for $\mathsf I_{1,d+1}$ and $\mathsf I_{1,d+5}$. For $d \equiv 1 \textrm{ mod } 3$, the first and second lines of are respectively finitized characters for $\mathsf I_{1,d+1}$ and $\mathsf I_{1,d+3}$.
Cylinder partition functions {#sec:cylPF}
----------------------------
The partition function $\ZcylMN$ of the loop model on the cylinder of size $2M\times N$ is obtained from the traces of $\Db(u)^M$ over $\stan_N^d$: = \_ U\_[d]{}(2)(u)\^M. This is the so-called Markov trace [@Jones83], used to embed the Temperley-Lieb algebra on a cylinder. This was also used by Jacobsen and Richard [@RJ06] in the context of the $Q$-state Potts model. Here, $\alpha$ is the weight of the non-contractible loops and $U_k(x)$ is the $k$-th Chebyshev polynomial of the second type. The conformal cylinder partition function is obtained by removing the non-universal energy contributions and by taking the scaling limit: \^[2MN f\_(u)+M f\_(u)]{} (q) where the ratio $\frac MN$ is taken to converge to a real number $\delta$, and the modular nome $q$ is given in terms of the lattice data in . The scaling limit is taken separately for the two parities of $N$. We denote by $\Zcylodd(q)$ and $\Zcyleven (q)$ the corresponding partition functions. We then have \[eq:Zcylgen\] (q) = \_[N]{}\_ U\_[d]{}(2) = \_ U\_[d]{}(2)\_[1,d+1]{}(q).
For $\alpha = 1$, simplifies to (q)|\_[= 1]{}=(q)|\_[= 1]{}=1 which is the trivial cylinder partition function of the rational model of percolation. For $\alpha = 2$, $U_{d}(1) = d+1$ and the partition functions can be written in terms of the $u(1)$ characters and their derivatives. Writing $\varkappa_j(q,z) = \varkappa^6_j(q,z)$ and $\varkappa_j(q) = \varkappa^6_j(q)$, we find $$\begin{aligned}
{2}
\Zcyleven(q) &+ \Zcylodd(q)\Big|_{\alpha = 2} = \frac{1}{(q)_\infty}\sum_{d{\geqslant}0}(d+1) (q^{\Delta_{1,d+1}}-q^{\Delta_{1,d+4}})
= \frac{3}{(q)_\infty} \sum_{d {\geqslant}1}q^{\Delta_{1,d+1}}
\nonumber\\& = \frac{3}{2(q)_\infty} \sum_{d \in \mathbb Z}q^{\Delta_{1,d+1}}
= \frac{3}{2(q)_\infty} \sum_{r=0}^5 \sum_{k\in \mathbb Z} q^{\Delta_{1,6k+r+1}}
= 3\big(\varkappa_1(q)+ \varkappa_3(q)+ \varkappa_5(q)\big)\end{aligned}$$ and $$\begin{aligned}
{2}
\Zcyleven(q) &- \Zcylodd(q)\Big|_{\alpha = 2} = \frac{1}{(q)_\infty}\sum_{d{\geqslant}0}(-1)^d(d+1) (q^{\Delta_{1,d+1}}-q^{\Delta_{1,d+4}})
= \frac{1}{(q)_\infty} \sum_{d {\geqslant}1}(-1)^d(2d-1)q^{\Delta_{1,d+1}}
\nonumber\\&
= \frac{1}{(q)_\infty} \sum_{d \in \mathbb Z}(-1)^d d\, q^{\Delta_{1,d+1}}
= \frac{1}{(q)_\infty} \sum_{r=0}^5 (-1)^r\sum_{k\in \mathbb Z}(6k+r) q^{\Delta_{1,6k+r+1}}
\nonumber\\&
= -\varkappa_1(q)+ 3\varkappa_3(q)- 5\varkappa_5(q) -12\frac{\dd}{\dd z} \Big(\varkappa_1(q,z) - 3\varkappa_3(q,z) + \varkappa_5(q,z) \Big)\Big|_{z=1}\ .\end{aligned}$$ The final result is
$$\begin{aligned}
{2}
&\Zcylodd (q)\big|_{\alpha = 2} &&= 2\varkappa_1(q) + 4 \varkappa_5(q) +6\frac{\dd}{\dd z} \Big(\varkappa_1(q,z) - 3\varkappa_3(q,z) + \varkappa_5(q,z) \Big)\Big|_{z=1}\ ,\\[0.15cm]
&\Zcyleven (q)\big|_{\alpha = 2} &&= \varkappa_1(q) + 3\varkappa_3(q) - \varkappa_5(q) -6\frac{\dd}{\dd z} \Big(\varkappa_1(q,z) - 3\varkappa_3(q,z) + \varkappa_5(q,z) \Big)\Big|_{z=1}\ .\end{aligned}$$
Critical percolation with periodic boundary conditions {#sec:periodiccase}
======================================================
The transfer tangle and the enlarged periodic Temperley-Lieb algebra {#sec:PTLandT}
--------------------------------------------------------------------
The transfer tangle with periodic boundary conditions is defined as (u)= \^[N]{}
(-0.2,-0.7)(5.2,1.0) (0,0)[0.16]{}[0]{}[90]{} (1,0)[0.16]{}[0]{}[90]{} (4,0)[0.16]{}[0]{}[90]{} (0,0.5)(-0.2,0.5) (5,0.5)(5.2,0.5) (2.5,0.5)[$\ldots$]{} (3.5,0.5)[$\ldots$]{} (0.5,.5)[$u$]{} (1.5,.5)[$u$]{} (4.5,.5)[$u$]{} (2.5,-0.5)[$\underbrace{\qquad \qquad \qquad \qquad \qquad \qquad}_N$]{}
, \[eq:Tu\] with the elementary face operator given in . Loop models on periodic geometries are usually described by the so-called [*periodic Temperley-Lieb algebra*]{}. This algebra was studied by Levy [@L91] and Martin and Saleur [@MS93] in the context of the Potts model. Its representation theory was also studied by Graham and Lehrer [@GL98], Green [@G98] and Erdmann and Green [@EG99]. The terminology and precise definitions of this algebra tend to vary from author to author, and here we will work with the [*enlarged*]{} periodic Temperley-Lieb algebra $\eptl_N(\alpha, \beta)$, defined [@PRVcyl2010] as \_N(, ) = , \^[-1]{},e\_j;j=1,…,N. It is a unital algebra, with the identity element $I$ obtained from the product of the generators $\Omega$ and $\Omega^{-1}$. The connectivities $I$ and $e_j$ for $1{\leqslant}j{\leqslant}N-1$ are given by the diagrams in , but drawn on a rectangle with periodic boundary conditions in the horizontal direction. The other generators are depicted as e\_N=
(0,-0.55)(2.4,0.35) (0,-0.35)(2.4,-0.35)(2.4,0.35)(0,0.35) (0.2,-0.55)[$_1$]{}(0.6,-0.55)[$_2$]{}(1.0,-0.55)[$_3$]{}(2.2,-0.55)[$_N$]{} (1.4,0.0)[$...$]{} (0.0,0.35)[0.2]{}[-90]{}[0]{} (0.0,-0.35)[0.2]{}[0]{}[90]{} (0.6,0.35)(0.6,-0.35) (1.0,0.35)(1.0,-0.35) (1.8,0.35)(1.8,-0.35) (2.4,-0.35)[0.2]{}[90]{}[180]{} (2.4,0.35)[0.2]{}[180]{}[-90]{} (-0.1,-0.4)(0,0.4) (2.4,-0.4)(2.5,0.4)
,
(-0.7,-0.55)(2.0,0.35) (0.2,-0.55)[$_1$]{}(0.6,-0.55)[$_2$]{}(1.0,-0.55)[$_3$]{}(1.4,-0.55)[$...$]{}(1.8,-0.55)[$_N$]{} (0,-0.35)(2.0,-0.35)(2.0,0.35)(0,0.35) (0,0)(0.4,0)[6]{}[(-0.2,-0.35)(-0.2,-0.0)(0.2,0.0)(0.2,0.35)]{} (-0.3,-0.4)(0,0.4) (2.0,-0.4)(2.4,0.4) (-0.55,0.042)[$\Omega=$]{}
,
(-1.1,-0.55)(2.0,0.35) (0.2,-0.55)[$_1$]{}(0.6,-0.55)[$_2$]{}(1.0,-0.55)[$_3$]{}(1.4,-0.55)[$...$]{}(1.8,-0.55)[$_N$]{} (0,-0.35)(2.0,-0.35)(2.0,0.35)(0,0.35) (0,0)(0.4,0)[6]{}[(-0.2,0.35)(-0.2,-0.0)(0.2,0.0)(0.2,-0.35)]{} (-0.3,-0.4)(0,0.4) (2.0,-0.4)(2.4,0.4) (-0.75,0.07)[$\Omega^{-1}=$]{}
. The defining relations of $\eptl_N(\alpha, \beta)$ include , wherein the indices are in the set $\{1, \dots, N\}$ and are taken modulo $N$, as well as \^[-1]{}= \^[-1]{} = I, e\_i \^[-1]{} = e\_[i-1]{}, \^[N]{} e\_N = e\_N \^[N]{}, (\^[1]{} e\_N)\^[N-1]{} = \^[N]{} (\^[1]{} e\_N). For $N$ even, there are extra relations which remove the non-contractible loops in favor of weights $\alpha$: E \^[1]{} E = E E= e\_2e\_4…e\_[N-2]{}e\_N. We refer to this algebra as $\eptl_N(\alpha, \beta)$ for both parities of $N$, even if the parameter $\alpha$ does not come into play for $N$ odd. The transfer tangle $\Tb(u)$ is an element of $\eptl_N(\alpha, \beta)$. We are interested in the case $\lambda=\frac\pi 3$ and therefore $\beta = 1$ corresponding to the model of critical percolation. The parameter $\alpha$ remains free until \[sec:Pcfgs\], whereafter the case $\alpha=2$ is our main focus.
As an element of $\eptl_N(\alpha,\beta)$, $\Tb(u)$ satisfies a number of properties, namely: (i) the periodicity property $\Tb(u+\pi) = (-1)^N\Tb(u)$, (ii) the commutativity property $[\Tb(u),\Tb(v)] = 0$, and (iii) the specialisations $\Tb(u=0) = \ir^N\Omega$ and $\Tb(u=\lambda) = \ir^N\Omega^{-1}$.
The braid transfer matrices are also elements of $\eptl_N(\alpha,\beta)$. They are defined as \[eq:braidT\] =
(-0.2,-0.7)(5.2,1.0) (-0.2,0.5)(2,0.5) (0.5,0)(0.5,0.35) (0.5,0.65)(0.5,1) (1.5,0)(1.5,0.35) (1.5,0.65)(1.5,1) (2.53,.51)[$\dots$]{} (3.53,.51)[$\dots$]{} (4,0.5)(5.2,0.5) (4.5,0)(4.5,0.35) (4.5,0.65)(4.5,1) (2.5,-0.5)[$\underbrace{\qquad \qquad \qquad \qquad \qquad \qquad}_N$]{}
, =
(-0.2,-0.7)(5.2,1.0) (2.53,.51)[$\dots$]{} (3.53,.51)[$\dots$]{} (0,0.5)(-0.2,0.5) (5,0.5)(5.2,0.5) (0.5,0)(0.5,1) (1.5,0)(1.5,1) (4.5,0)(4.5,1) (0,0.5)(0.35,0.5) (0.65,0.5)(1.35,0.5) (1.65,0.5)(2,0.5) (4,0.5)(4.35,0.5) (4.65,0.5)(5,0.5) (2.5,-0.5)[$\underbrace{\qquad \qquad \qquad \qquad \qquad \qquad}_N$]{}
, with the braid face operators defined in . The two braid transfer matrices are not equal in general. They are obtained as the $u \rightarrow \pm \ir \infty$ limits of $\Tb(u)$ as follows: = \_[u ]{} ()\^N (-)\^N(u), = \_[u -]{} ()\^N (-)\^N(u).
Standard modules {#sec:Pstanmod}
----------------
The standard modules $\stanp_N^d$ over $\eptl_N(\alpha, \beta)$ are defined on the vector space generated by the span of periodic link states on $N$ nodes with $d$ defects, where $N \equiv d \textrm{ mod }2$. The construction is similar to the one for $\tl_N(\beta)$ in \[sec:stanmod\]. One difference is that the link states are drawn on a line segment with the left and right ends identified, and the loop segments can connect via the back (virtual cut) of the cylinder. The dimensions of these vector spaces are \_N\^d = . For example, the link states that generate $\stanp_5^1$ are
[c]{}
(-0.0,0)(2.0,0.5) (0,0)(2.0,0) (0.4,0)[0.2]{}[0]{}[180]{} (1.2,0)[0.2]{}[0]{}[180]{} (1.8,0)(1.8,0.5)
,
(-0.0,0)(2.0,0.5) (0,0)(2.0,0) (0.4,0)[0.2]{}[0]{}[180]{} (1.6,0)[0.2]{}[0]{}[180]{} (1.0,0)(1.0,0.5)
,
(-0.0,0)(2.0,0.5) (0,0)(2.0,0) (0.8,0)[0.2]{}[0]{}[180]{} (1.6,0)[0.2]{}[0]{}[180]{} (0.2,0)(0.2,0.5)
,
(-0.0,0)(2.0,0.5) (0,0)(2.0,0) (0.0,0)[0.2]{}[0]{}[90]{} (2.0,0)[0.2]{}[90]{}[180]{} (1.2,0)[0.2]{}[0]{}[180]{} (0.6,0)(0.6,0.5)
,
(-0.0,0)(2.0,0.5) (0,0)(2.0,0) (0.0,0)[0.2]{}[0]{}[90]{} (2.0,0)[0.2]{}[90]{}[180]{} (0.8,0)[0.2]{}[0]{}[180]{} (1.4,0)(1.4,0.5)
,\
(-0.0,0)(2.0,0.5) (0,0)(2.0,0) (0.2,0)(0.2,0.7)(1.4,0.7)(1.4,0) (0.8,0)[0.2]{}[0]{}[180]{} (1.8,0)(1.8,0.5)
,
(-0.0,0)(2.0,0.5) (0,0)(2.0,0) (0.6,0)(0.6,0.7)(1.8,0.7)(1.8,0) (1.2,0)[0.2]{}[0]{}[180]{} (0.2,0)(0.2,0.5)
,
(-0.0,0)(2.0,0.5) (0,0)(2.0,0) (1.0,0)(1.0,0.7)(2.2,0.7)(2.2,0) (-0.2,0.54)(-0.12,0.43)(0.2,0.4)(0.2,0) (1.6,0)[0.2]{}[0]{}[180]{} (0.6,0)(0.6,0.5) (2.0,-0.1)(2.4,0.9) (0.0,-0.1)(-0.4,0.9)
,
(-0.0,0)(2.0,0.5) (0,0)(2.0,0) (-0.12,0.53)(-0.07,0.545)(0.6,0.545)(0.6,0) (2.12,0.53)(2.07,0.545)(1.4,0.545)(1.4,0) (0.0,0)[0.2]{}[0]{}[90]{} (2.0,0)[0.2]{}[90]{}[180]{} (1.0,0)(1.0,0.5) (2.0,-0.1)(2.4,0.9) (0.0,-0.1)(-0.4,0.9)
,
(-0.0,0)(2.0,0.5) (0,0)(2.0,0) (2.2,0.54)(2.12,0.43)(1.8,0.4)(1.8,0) (-0.2,0)(-0.2,0.7)(1.0,0.7)(1.0,0) (0.4,0)[0.2]{}[0]{}[180]{} (1.4,0)(1.4,0.5) (2.0,-0.1)(2.4,0.9) (0.0,-0.1)(-0.4,0.9)
.
The standard action of $\eptl_N(\alpha, \beta)$ on $\stanp_N^d$ is defect-preserving. For $d=0$, the resulting representations depend on $\beta$ and $\alpha$. For $d>0$, they depend on $\beta$ and a [*twist parameter*]{} $\omega$. To compute the action of a connectivity on a link state, one draws the link state above the connectivity and reads the new link state from the bottom nodes. If some defects have annihilated pairwise, the result is set to zero. Otherwise, a multiplicative factor of $\beta$ is inserted for each contractible loop. For $d=0$, a multiplicative factor of $\alpha$ is also inserted for each non-contractible loop. For $d>0$, a multiplicative factor of $\omega$ or $\omega^{-1}$ is inserted each time a defect crosses the virtual cut of the cylinder: $\omega$ if the defect travels to the left and $\omega^{-1}$ if it travels to the right. Here are examples of the standard action of $\eptl_N(\alpha, \beta)$:
$$\begin{aligned}
{2}
&
\begin{pspicture}[shift=-0.55](0,-0.65)(1.6,1.05)
\pspolygon[fillstyle=solid,fillcolor=lightlightblue,linewidth=0pt](0,-0.5)(1.6,-0.5)(1.6,0.5)(0,0.5)
\psarc[linecolor=blue,linewidth=1.5pt]{-}(1.2,-0.5){0.2}{0}{180}
\psarc[linecolor=blue,linewidth=1.5pt]{-}(0.4,0.5){0.2}{180}{0}
\psbezier[linecolor=blue,linewidth=1.5pt]{-}(0.2,-0.5)(0.2,0)(1.0,0)(1.0,0.5)
\psbezier[linecolor=blue,linewidth=1.5pt]{-}(0.6,-0.5)(0.6,0)(1.4,0)(1.4,0.5)
\psline{-}(0,0.5)(1.6,0.5)
\psarc[linecolor=darkgreen,linewidth=1.5pt]{-}(0,0.5){0.2}{0}{90}
\psbezier[linecolor=darkgreen,linewidth=1.5pt]{-}(0.6,0.5)(0.6,1.2)(1.8,1.2)(1.8,0.5)
\psarc[linecolor=darkgreen,linewidth=1.5pt]{-}(1.2,0.5){0.2}{0}{180}
\psframe[fillstyle=solid,linecolor=white,linewidth=0pt](-0.4,-0.5)(0,1.2)
\psframe[fillstyle=solid,linecolor=white,linewidth=0pt](1.6,-0.5)(2.0,1.2)
\end{pspicture} \ = \alpha \
\begin{pspicture}[shift=0.0](0,0.35)(1.6,0.95)
\psline{-}(0,0.35)(1.6,0.35)
\psarc[linecolor=darkgreen,linewidth=1.5pt]{-}(0.4,0.35){0.2}{0}{180}
\psarc[linecolor=darkgreen,linewidth=1.5pt]{-}(1.2,0.35){0.2}{0}{180}
\end{pspicture}\, ,
\qquad
&&
\begin{pspicture}[shift=-0.55](0,-0.65)(1.6,1.05)
\pspolygon[fillstyle=solid,fillcolor=lightlightblue,linewidth=0pt](0,-0.5)(1.6,-0.5)(1.6,0.5)(0,0.5)
\psarc[linecolor=blue,linewidth=1.5pt]{-}(0.8,0.5){0.2}{180}{0}
\psarc[linecolor=blue,linewidth=1.5pt]{-}(1.6,0.5){0.2}{180}{270}
\psarc[linecolor=blue,linewidth=1.5pt]{-}(0,0.5){0.2}{270}{0}
\psarc[linecolor=blue,linewidth=1.5pt]{-}(0.8,-0.5){0.2}{0}{180}
\psbezier[linecolor=blue,linewidth=1.5pt]{-}(0.2,-0.5)(0.2,0.2)(1.4,0.2)(1.4,-0.5)
\psframe[fillstyle=solid,linecolor=white,linewidth=0pt](-0.4,-0.5)(0,0.5)
\psframe[fillstyle=solid,linecolor=white,linewidth=0pt](1.6,-0.5)(2.0,0.5)
\psline{-}(0,0.5)(1.6,0.5)
\psline[linecolor=darkgreen,linewidth=1.5pt]{-}(0.2,0.5)(0.2,1.0)
\psline[linecolor=darkgreen,linewidth=1.5pt]{-}(0.6,0.5)(0.6,1.0)
\psarc[linecolor=darkgreen,linewidth=1.5pt]{-}(1.2,0.5){0.2}{0}{180}
\end{pspicture} \ = \ 0 \, ,
\\
&\begin{pspicture}[shift=-0.55](0,-0.65)(1.6,1.05)
\pspolygon[fillstyle=solid,fillcolor=lightlightblue,linewidth=0pt](0,-0.5)(1.6,-0.5)(1.6,0.5)(0,0.5)
\psarc[linecolor=blue,linewidth=1.5pt]{-}(0.8,-0.5){0.2}{0}{180}
\psarc[linecolor=blue,linewidth=1.5pt]{-}(0,0.5){0.2}{270}{0}
\psarc[linecolor=blue,linewidth=1.5pt]{-}(1.6,0.5){0.2}{180}{270}
\psbezier[linecolor=blue,linewidth=1.5pt]{-}(0.2,-0.5)(0.2,0)(1.0,0)(1.0,0.5)
\psbezier[linecolor=blue,linewidth=1.5pt]{-}(-0.2,-0.5)(-0.2,0)(0.6,0)(0.6,0.5)
\psbezier[linecolor=blue,linewidth=1.5pt]{-}(1.4,-0.5)(1.4,0)(2.2,0)(2.2,0.5)
\psline{-}(0,0.5)(1.6,0.5)
\psline[linecolor=darkgreen,linewidth=1.5pt]{-}(0.6,0.5)(0.6,1)
\psline[linecolor=darkgreen,linewidth=1.5pt]{-}(1.0,0.5)(1.0,1)
\psarc[linecolor=darkgreen,linewidth=1.5pt]{-}(0,0.5){0.2}{0}{90}
\psarc[linecolor=darkgreen,linewidth=1.5pt]{-}(1.6,0.5){0.2}{90}{180}
\psframe[fillstyle=solid,linecolor=white,linewidth=0pt](-0.4,-0.5)(0,1.2)
\psframe[fillstyle=solid,linecolor=white,linewidth=0pt](1.6,-0.5)(2.25,1.2)
\end{pspicture} \ = \omega \beta \
\begin{pspicture}[shift=0.0](0,0.35)(1.6,0.95)
\psline{-}(0,0.35)(1.6,0.35)
\psarc[linecolor=darkgreen,linewidth=1.5pt]{-}(0.8,0.35){0.2}{0}{180}
\psline[linecolor=darkgreen,linewidth=1.5pt]{-}(0.2,0.35)(0.2,0.85)
\psline[linecolor=darkgreen,linewidth=1.5pt]{-}(1.4,0.35)(1.4,0.85)
\end{pspicture}\, , \qquad
&&\begin{pspicture}[shift=-0.55](0,-0.65)(1.6,1.05)
\pspolygon[fillstyle=solid,fillcolor=lightlightblue,linewidth=0pt](0,-0.5)(1.6,-0.5)(1.6,0.5)(0,0.5)
\psbezier[linecolor=blue,linewidth=1.5pt]{-}(0.6,-0.5)(0.6,0)(-0.2,0)(-0.2,0.5)
\psbezier[linecolor=blue,linewidth=1.5pt]{-}(1.0,-0.5)(1.0,0)(0.2,0)(0.2,0.5)
\psbezier[linecolor=blue,linewidth=1.5pt]{-}(1.4,-0.5)(1.4,0)(0.6,0)(0.6,0.5)
\psbezier[linecolor=blue,linewidth=1.5pt]{-}(1.8,-0.5)(1.8,0)(1,0)(1,0.5)
\psbezier[linecolor=blue,linewidth=1.5pt]{-}(2.2,-0.5)(2.2,0)(1.4,0)(1.4,0.5)
\psbezier[linecolor=blue,linewidth=1.5pt]{-}(0.2,-0.5)(0.2,0)(-0.6,0)(-0.6,0.5)
\psline{-}(0,0.5)(1.6,0.5)
\psline[linecolor=darkgreen,linewidth=1.5pt]{-}(1.4,0.5)(1.4,1)
\psline[linecolor=darkgreen,linewidth=1.5pt]{-}(1.0,0.5)(1.0,1)
\psarc[linecolor=darkgreen,linewidth=1.5pt]{-}(0.4,0.5){0.2}{0}{180}
\psframe[fillstyle=solid,linecolor=white,linewidth=0pt](-0.65,-0.5)(0,1.2)
\psframe[fillstyle=solid,linecolor=white,linewidth=0pt](1.6,-0.5)(2.25,1.2)
\end{pspicture} \ = \omega^{-2} \
\begin{pspicture}[shift=0.0](0,0.35)(1.6,0.95)
\psline{-}(0,0.35)(1.6,0.35)
\psarc[linecolor=darkgreen,linewidth=1.5pt]{-}(1.2,0.35){0.2}{0}{180}
\psline[linecolor=darkgreen,linewidth=1.5pt]{-}(0.6,0.35)(0.6,0.85)
\psline[linecolor=darkgreen,linewidth=1.5pt]{-}(0.2,0.35)(0.2,0.85)
\end{pspicture}\, .\end{aligned}$$
For generic values of $\beta$, $\alpha$ and $\omega$, the modules $\stanp_N^d$ are irreducible modules over $\eptl_N(\alpha,\beta)$. For percolation, the indecomposable structures of $\stanp_N^0$ with $\alpha=1$ and $\stanp_N^{d>0}$ with $\omega^N=1$ are discussed in [@GRSV15].
Fused transfer matrices and the fusion hierarchy {#sec:Pfusion}
------------------------------------------------
Using the Wenzl-Jones projectors for generic values of $\beta$, one can define a sequence of commuting fused transfer tangles $\Tb^n(u)$, starting with $\Tb^1(u) = \Tb(u)$, which satisfy the fusion hierarchy relations \^[n]{}\_0\^[1]{}\_n = h\_n \^[n-1]{}\_0 + h\_[n-1]{} \^[n+1]{}\_0, n 0, \[eq:fushierT\] where \_k\^n = \^n(u + k ), \_0\^0 = h\_[-1]{} , \^[-1]{}\_k = 0, h\_k = s\_[k]{}(u)\^[N]{}. These tangles indeed commute as elements of $\eptl_N(\alpha,\beta)$: $[\Tb^m(u),\Tb^n(v)] = 0$.
In particular, for $\beta\ne 0$, the fused transfer tangle $\Tb^2(u)$ is constructed from the $1\times2$ fused face operator, see , as follows: \^2 (u)= (-1)\^N
(-0.2,-0.7)(5.2,1.0) (0,0)[0.16]{}[0]{}[90]{} (1,0)[0.16]{}[0]{}[90]{} (4,0)[0.16]{}[0]{}[90]{} (0,0.5)(-0.2,0.5) (5,0.5)(5.2,0.5) (2.5,0.5)[$\ldots$]{} (3.5,0.5)[$\ldots$]{} (0.5,.5)[$u$]{}(0.5,0.75) (1.5,.5)[$u$]{}(1.5,0.75) (4.5,.5)[$u$]{}(4.5,0.75) (2.5,-0.5)[$\underbrace{\qquad \qquad \qquad \qquad \qquad \qquad}_N$]{}
.
For the rational values $\lambda = \frac{\pi(p'-p)}{p'}$, see , the hierarchy of fused transfer matrices closes [@MDPR14] at $n=p'$: \^[p’]{}\_0 = \^[p’-2]{}\_1 + 2(-)\^[N(p’-p)]{} h\_[-1]{} , = T\_[p’]{}(12 ), where $T_k(x)$ in the definition of $\Jb$ is the $k$-th Chebyshev polynomial of the first kind. For percolation, this becomes \^[3]{}\_0 = \^[1]{}\_1 + 2(-)\^[N]{} h\_[-1]{} , = T\_3(12 ) = 12 ()\^3-32. We note that the closure relation could equivalently be written in terms of the other braid transfer matrix $\Tb_{\!-\infty}$.
The $\boldsymbol T$-system and the $\boldsymbol Y$-system {#sec:Pfunctional}
---------------------------------------------------------
The $T$-system for the periodic transfer matrices is found using the fusion hierarchy relations and a recursive argument. It takes the form \^[n]{}\_0 \^[n]{}\_1 = h\_[-1]{} h\_n +\^[n+1]{}\_0 \^[n-1]{}\_1, n0. By defining \^n(u) =, n 0, and $\tb^n_k = \tb^n(u+k \lambda)$, we find that the tangles $\tb^n(u)$ satisfy precisely the same $Y$-system as on the strip, namely: \[eq:ysysP\] \^[n]{}\_0 \^[n]{}\_1=(+ \^[n-1]{}\_1)(+ \^[n+1]{}\_0), n 1. For rational values of $\lambda$, this $Y$-system can be written in terms of finitely many objects. Indeed, by defining $\Kbt(u)$ and $\Lambdab$ as follows: (u) = \_1\^[p’-2]{}, = , \[eq:KtangT\] we find the closure relations: + \^[p’-1]{}\_0 = (+ \^\_0)(+ \^[-]{}\_0), \_0\_1 = 1+\_1\^[p’-2]{}. \[eq:Pyclo\] The finite $Y$-system consists of the relations for $n = 1, \dots, p'-2$ and . It is described by the Dynkin diagram $D_{p'-2}$ in \[fig:Dynkin\], with the new feature that the endpoint nodes to the right are distinguished by the factors $ \eE^{\pm\ir \boldsymbol{\Lambda}}$.
For percolation, the $Y$-system involves only the tangles $\tb^1(u)$, $\Kb(u)$, $\Ib$ and $\Lambdab$ and is written as \[eq:ysysT\] \_0\^1\_1\^1 = (+ \^\_0)(+ \^[-]{}\_0), \_0\_1 = + \_1\^1. By defining
\[eq:aUT\] $$\begin{aligned}
{3}
&\ab^1(x) = \tb_0^1(\tfrac {\ir x} 3), \qquad &&\ab^2(x) = (-1)^N\Kb_0(\tfrac {\ir x} 3- \tfrac \pi 6), \label{eq:aUT1}\\
&\Ab^1(x) = \Ib+\ab^1(x), \qquad &&\Ab^2(x) = \big(\Ib+(-1)^N\eE^{\ir \boldsymbol{\Lambda}}\ab^2(x)\big)\big(\Ib+(-1)^N\eE^{-\ir \boldsymbol{\Lambda}}\ab^2(x)\big),\end{aligned}$$
the $Y$-system takes a symmetric form: $$\ab^1(x-\tfrac{\ir \pi}2)\ab^1(x+\tfrac{\ir \pi}2) = \Ab^2(x),\qquad
\ab^2(x-\tfrac{\ir \pi}2)\ab^2(x+\tfrac{\ir \pi}2) = \Ab^1(x).
\label{eq:finalyT}$$
Properties of the eigenvalues {#sec:Pprops}
-----------------------------
#### $\boldsymbol{Y}$-system for the eigenvalues.
The tangles $\Tb_{\!\pm\infty}$ are not proportional to the identity tangle, but nevertheless act as multiples of the identity on the standard modules. For $\alpha=2$ and $\omega =1$, the (unique) eigenvalues of $\Tb_{\!+\infty}$ and $\Tb_{\!-\infty}$ on $\stanp_N^d$ are identical and were computed in [@MDSA13no2] for generic values of $\lambda$. The generalisation to arbitrary $\alpha$ and $\omega$ is straightforward. For $\lambda=\frac \pi 3$, we obtain \[eq:Jeigs\] T\_ = 2 (- ),J = (-1)\^d (3 ), \^ = (-1)\^d \^[3 ]{}, where we parameterise $\alpha$ and $\omega$ in terms of a single parameter $\gamma$ as follows: = 2 , = \^. The $Y$-system satisfied by the eigenvalues in $\stanp_N^d$ is then given by with A\^2(x) = (1+\^[3 ]{} a\^2(x))(1+\^[-3 ]{} a\^2(x)). The cancellation of the factors of $(-1)^d$ and $(-1)^N$ here explains the choice made in to include an extra $(-1)^N$ in the definition of $\ab^2(x)$.
#### Razumov-Stroganov eigenvalues.
For periodic boundary conditions, the Razumov-Stroganov eigenstate appears in two situations: (i) for $d=0$ with $\gamma = \pm \frac\pi3,\pm \frac{2\pi}3$, and (ii) for $d=1$ with $\gamma = 0$, $\pi$. This is consistent with the spin-chain results of [@RS01; @RS01no2; @S01]. Indeed, a simple solution to the $Y$-system in these cases is $K_0 = -1$ and $t_0^1 = 0$, which corresponds to T\^1(u) = - = -\^[N]{} ()\^[N]{}, T\^2(u) = 0. \[eq:RSeigT\] In both cases (i) and (ii), the Razumov-Stroganov eigenvalue is non-degenerate and acts as the ground state of the corresponding standard module.
#### Patterns of zeros and analyticity strips.
The eigenvalues $T^1(u)$ and $T^2(u)$ are centered Laurent polynomials in the variable $z=\eE^{\ir u}$ with minimal and maximal powers $-N$ and $N$. Due to the periodicity properties of $\Tb^n(u)$, the patterns of zeros are periodic with period $\pi$ in the complex $u$ plane and each vertical strip of width $\pi$ has at most $N$ zeros. Examples of patterns of zeros are given in \[fig:eigpatternsp,fig:eigpatternsp2\]. Empirically, we find that for the leading eigenvalues, only a finite number of these zeros are in the following analyticity strips: T\^1(u): -6< (u) < 2, T\^2(u): -3< (u) < 3.
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
$T^1(u):$ $T^2(u):$
\[0.1cm\] ![The location of the zeros for the ground state in $\stanp_{10}^2$ for $\omega = 1$. The analyticity strips are shaded in gray. There is one pair of complex zeros in the analyticity strip for $T^1(u)$ and none for $T^2(u)$.[]{data-label="fig:eigpatternsp"}](eig1d1p.eps "fig:"){width=".45\textwidth"} ![The location of the zeros for the ground state in $\stanp_{10}^2$ for $\omega = 1$. The analyticity strips are shaded in gray. There is one pair of complex zeros in the analyticity strip for $T^1(u)$ and none for $T^2(u)$.[]{data-label="fig:eigpatternsp"}](eig1d2p.eps "fig:"){width=".45\textwidth"}
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
$T^1(u):$ $T^2(u):$
\[0.1cm\] ![The location of the zeros of an excited state in $\stanp_{10}^2$ for $\omega = 1$. There are four complex zeros in the analyticity strip for $T^1(u)$ and they are not related by complex conjugation. For $T^2(u)$, there is a single double-zero on the central line of the analyticity strip.[]{data-label="fig:eigpatternsp2"}](eig2d1p.eps "fig:"){width=".45\textwidth"} ![The location of the zeros of an excited state in $\stanp_{10}^2$ for $\omega = 1$. There are four complex zeros in the analyticity strip for $T^1(u)$ and they are not related by complex conjugation. For $T^2(u)$, there is a single double-zero on the central line of the analyticity strip.[]{data-label="fig:eigpatternsp2"}](eig2d2p.eps "fig:"){width=".45\textwidth"}
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
As opposed to the strip boundary conditions, none of the zeros are common to all the eigenvalues. This is due to the absence of surface terms in the fusion hierarchy relations . In general, the zeros have three possible locations: (i) in the center of the analyticity strip, (ii) on one of the two edges of the analyticity strip, or (iii) outside the analyticity strip. We will be interested in zeros of type (i) and (ii). The zeros that lie in the center of the analyticity strip of $T^1(u)$ are always of order one. For $\gamma = 0$, the zeros in the center of the analyticity strip of $T^2(u)$ can be of order one or two. This degeneracy is lifted away from $\gamma = 0$, but may occur again for special values of $\gamma$. For generic values of $\gamma$, all zeros of $T^2(u)$ are of order one.
We observe that all the patterns of zeros are symmetric under a reflection with respect to the central line of the analyticity strips. This implies that the corresponding eigenvalues are real on this central line. In contrast, some patterns of zeros, for example the one in \[fig:eigpatternsp2\], are not symmetric with respect to a reflection about the real $u$ axis. In general, these eigenvalues are therefore complex for Im$(u)=0$.
We shall see in \[sec:Pfsc\] that, for the finite-size corrections, the zeros in the upper and lower parts of the $u$ plane are tied to contributions to $\Delta$ and $\bar\Delta$ respectively. For small values of $N$, we find that some zeros are very close to the real axis, and even coincide with this axis in certain instances. For the periodic boundary conditions, the line dividing the plane into two chiral halves need not necessarily coincide with the real axis. In fact, it can be chosen arbitrarily, as long as the convention does not impact the finite-size corrections for the leading eigenvalues. This is discussed further in \[sec:fsgf\].
We encode the content of zeros in both the upper- and lower-half planes together in a pattern diagram. For the eigenvalues corresponding to \[fig:eigpatternsp,fig:eigpatternsp2\], the pattern diagrams are
(-0.4,-1.4)(1.4,1.4) (0,-1.4)(0,1.4)(1,-1.4)(1,1.4)(0,0)(1,0) (0,-0.3)(0,-0.6)(0,0.3)(0,0.6) (0.5,-0.9)(0.5,0.9) (1,-0.3)(1,-0.6)(1,0.3)(1,0.6)
(-0.4,-1.4)(1.4,1.4) (0,-1.4)(0,1.4)(1,-1.4)(1,1.4)(0,0)(1,0)
(-0.4,-1.4)(1.4,1.4) (0,-1.4)(0,1.4)(1,-1.4)(1,1.4)(0,0)(1,0) (0,0)(0,-0.3)(0,0.6) (0.5,-0.6)(0.5,-0.9)(0.5,0.3)(0.5,0.9) (1,0)(1,-0.3)(1,0.6)
(-0.4,-1.4)(1.4,1.4) (0,-1.4)(0,1.4)(1,-1.4)(1,1.4)(0,0)(1,0) (0.5,0)
. We work with the following convention. For a given eigenvalue, if the first analyticity strip has a site, occupied or unoccupied, lying on the dashed separation line, it is considered to belong to the upper half-plane. The same convention was used in [@PRVcyl2010] for critical dense polymers. The pattern of zeros in the second strip is described below as a pair of single-column configurations. If the second analyticity strip has a site lying on the dashed separation line, then the corresponding site in the first column is considered to be in the upper-half of the plane and the site in the second column is considered to be in the lower-half. We will come back to this in \[sec:fsgf\] and \[sec:Pcpf\].
The analyticity properties of $K(u)$ and $t^1(u)$ are deduced from those of $T^1(u)$ and $T^2(u)$. From , we see that $K(u)$ has a pole of order $N$ at $u = \frac \pi3$. Likewise, $t^1(u)$ has poles of order $N$ at $u = -\frac\pi3, \frac\pi 3$ and a zero of order $N$ at $u=0$. The analyticity strips have width $2\lambda$: K(u): -2< (u) < 6, t\^1(u): -3< (u) < 3.
A similar analysis on small system sizes reveals that the functions $1+\eE^{\pm \ir \Lambda}K(u)$ and $1+t^1(u)$ are analytic and non-zero in the following strips: 1+\^K(u): -3< (u) < 0, 1+t\^1(u): -6< (u) < 6. We observe that these analyticity strips have width $\lambda$ and are entirely free of zeros and poles.
Finally, in terms of the variables defined in , the analyticity strips take the following elegant forms:
$$\begin{aligned}
{4}
&\mathfrak a^1(x): &&\quad - \pi < \text{Im}(x) < \pi , \qquad
&&\mathfrak a^2(x): &&\quad -\pi < \text{Im}(x) < \pi, \\[0.15cm]
&\mathfrak A^1(x): &&\quad -\frac \pi 2< \text{Im}(x) < \frac \pi 2, \qquad
&&\mathfrak A^2(x): &&\quad -\frac \pi 2< \text{Im}(x) < \frac {\pi} 2.\end{aligned}$$
#### Braid limit.
The braid transfer tangles $\Tb^n_{\!\pm\infty}$ are obtained as the $u\rightarrow \pm\ir \infty$ limit of the transfer tangle $\Tb^n(u)$. For $n=1$, $\Tb^1_{\!\pm\infty} \equiv \Tb_{\!\pm\infty}$ are defined in . For $n=2$, we have
$$\begin{aligned}
{2}
\Tb^2_{\!+\infty} &= \lim_{u \rightarrow \ir \infty} \bigg(\frac{-\eE^{\ir(\pi-\lambda)}}{s_{1}(u)}\bigg)^N \Tb^2(u) = (\Fbtp)^2 - \Ib, \\
\Tb^2_{\!-\infty} &= \lim_{u \rightarrow - \ir \infty} \bigg(\frac{\eE^{\ir(\pi-\lambda)}}{s_{1}(-u)}\bigg)^N \Tb^2(u) = (\Fbtm)^2 - \Ib. $$
The last equalities are obtained by taking the proper $u \rightarrow \pm\ir \infty$ limits of for $n=1$. The braid limits of $\Kb_{\!\pm\infty}$ and $\tb^1_{\pm\infty}$ are likewise given by \_ = (-1)\^N\^2\_= \_[u ]{} (u) = \^1\_, \^1\_ = \^1\_= \_[u ]{} \^1(u) = \^2\_= (\^1\_)\^2-.
The matrix realisations on $\stanp_N^d$ of all these braid tangles are scalar multiples of the identity matrix. For $\beta = 1$, the result for $T^1_{\!\pm\infty}$ is given in and yields \[eq:braida1a2\] a\^1\_ = 4 \^2(+3)-1, a\^2\_ = 2 (+3). The eigenvalues thus have different braid behaviors in the lower- and upper-half planes for generic $\gamma$.
For $\gamma = 0$ and $d \equiv 0,1 \Mod 3$, we have $\mathfrak a^1_{\pm\infty} = 0$ and $\mathfrak a^2_{\pm \infty} = -1$. As for the strip case, the convergence of $1+\mathfrak a^2(x) $ to zero can follow different subcases. In the periodic case however, the convergence can be different for $x \rightarrow \infty$ and $x \rightarrow -\infty$. From our numerical data, we find the following four subcases: $$\begin{aligned}
{3}
\label{eq:subcasesAABB}
&\text{Subcase } \bar \tA: \quad\lim_{x \rightarrow -\infty} \eE^{-\frac{2x}3} \big(1+ \mathfrak a^2(x)\big) = \kappa_-, \qquad &&\text{Subcase } \bar \tB: \quad \lim_{x \rightarrow -\infty} \eE^{-\frac{4x}3} \big(1+ \mathfrak a^2(x)\big) = \kappa'_-,\\[0.2cm]
&\text{Subcase A}: \quad\lim_{x \rightarrow +\infty} \eE^{\frac{2x}3} \big(1+ \mathfrak a^2(x)\big) = \kappa_+, \qquad &&\text{Subcase B}: \quad \lim_{x \rightarrow +\infty} \eE^{\frac{4x}3} \big(1+ \mathfrak a^2(x)\big) = \kappa'_+.\end{aligned}$$ The constants $\kappa_\pm$ and $\kappa'_\pm$ are real and strictly positive, for all eigenvalues except the Razumov-Stroganov eigenvalues. The argument leading to this conclusion is identical to the one given below . From and , the rate of convergence of $\mathfrak a^1(x)$ to zero as $x \rightarrow \pm \infty$ is $$\begin{aligned}
{2}
&\text{Subcase } \bar \tA: \quad \lim_{x \rightarrow -\infty} \eE^{-\frac{2x}3} \mathfrak a^1(x) = -\kappa_-, \qquad &&\text{Subcase } \bar \tB: \quad \lim_{x \rightarrow -\infty} \eE^{\frac{-4x}3} \mathfrak a^1(x) = \kappa'_-,\\
&\text{Subcase } \tA: \quad \lim_{x \rightarrow +\infty} \eE^{\frac{2x}3} \mathfrak a^1(x) = -\kappa_+, \qquad &&\text{Subcase } \tB: \quad \lim_{x \rightarrow +\infty} \eE^{\frac{4x}3} \mathfrak a^1(x) = \kappa'_+.\end{aligned}$$ An eigenstate is thus described by one of these four pairs: $(\bar\tA,\tA)$, $(\bar\tA,\tB)$, $(\bar\tB,\tA)$ or $(\bar\tB,\tB)$. In the pattern diagrams, this is indicated by small letters at the bottom and top of the analyticity strips, as in \[fig:GSpatternsp,fig:patsP62\].
#### Bulk behavior.
The eigenvalues of $\tb^1(u)$ and $\Kb(u)$ have the following bulk behaviors: t\^1\_(u) = 0, K\_(u) = , {-1,1}. \[eq:bulkbP\] Indeed, the behavior of $t^1(u)$ near the origin is governed by the zero of order $N$ at $u = 0$, so the function is approximately zero in this neighborhood. From this remark, it follows that is simply $K_{\text{bulk}}(u)^2 = 1$ in this neighborhood.
For a given pattern of zeros, one can deduce the values of $\sigma$ from the braid value $K_{+\infty}$, and the number of zeros on the central line of the analyticity strip. For instance, let us consider the ground state of $\stanp_{10}^2$ for $\omega =1$, whose pattern is displayed in \[fig:eigpatternsp\]. The braid behavior is $K_{+\infty} = -1$. In the upper half-plane, there is a single zero on the central line of the analyticity strip, so $K(-\frac\pi6)$ is positive. We deduce that $\sigma=1$. In contrast, applying the same reasoning to the eigenvalue corresponding to \[fig:eigpatternsp\], we find $\sigma=-1$. For periodic boundary conditions, both values of $\sigma$ are possible.
Finite-size corrections {#sec:Pfsc}
-----------------------
The eigenvalues $T(u)$ can be factored as \[eq:Tseparation\] T(u) = T\_(u) T\_(u). Compared to , the surface term is absent because of the periodic boundary conditions. The bulk and finite contributions satisfy T\_(u)T\_(u+) = h\_[-1]{}h\_1,T\_(u)T\_(u+) = 1+t\^1(u). \[eq:inv.identities.P\] The solution to the first relation in gives a bulk free energy identical to the one for the strip. Below, we compute the finite-size correction to $T_{\text{f}}(u)$ for any eigenvalue in $\stanp_N^d$. We use the functions $\mathfrak a^1(x)$, $\mathfrak a^2(x)$, $\mathfrak A^1(x)$ and $\mathfrak A^2(x)$ for the computation, as well as b(x) = T\_( 3+ 6). The relation for $T_{\text{f}}(u)$ in then becomes b(x - 2) b(x + 2) = A\^1(x). \[eq:bbUT\] Since defines $\mathfrak b(x)$, only up to a sign, we choose the convention that b\_ = T\^1\_. The rest of this section computes the finite-size corrections to $\mathfrak b(x)$ for any eigenvalue. The results could be obtained for the full interval $\gamma \in [-\pi,\pi]$, but for simplicity we consider $\gamma$ in a neighborhood of $\gamma = 0$, namely $\gamma \in (-\frac\pi3,\frac\pi3)$ for $d \equiv 0 \Mod 3$ and $\gamma \in (-\frac\pi6,\frac\pi6)$ for $d \equiv 1,2 \Mod 3$.
#### TBA equations.
For a given eigenstate of $\mathfrak a^1(x)$ and $\mathfrak a^2(x)$, we respectively denote by $t^1_+$ and $t^1_-$ the numbers of zeros $w_\pm^i$ of $\mathfrak a^1(x)$ in the upper and lower part of the plane. These zeros are ordered such that $w_\pm^i{\geqslant}w_\pm^{i+1}$. Likewise $t^2_+$ and $t^2_-$ count the zeros $x_\pm^j$ of $\mathfrak a^2(x)$ in the upper and lower half-planes, which are ordered as $x_\pm^j{\geqslant}x_\pm^{j+1}$. Depending on $\gamma$, some of the zeros may be degenerate.
The finite-size correction functions $\ell^1(x)$ and $\ell^2(x)$ are defined as
$$\begin{aligned}
{2}
\mathfrak a^1(x) &= \tanh^{N}(\tfrac x 2)\cdot \bigg(\prod_{i = 1}^{t_+^1}\tanh\big(\tfrac{x-w^i_+}2\big) \bigg)\bigg(\prod_{i = 1}^{t_-^1} \tanh\big(\tfrac{x-w^i_-}2\big)\bigg)\cdot \ell^1(x),\label{eq:nlieper1}\\
\mathfrak a^2(x) &= \bigg(\prod_{j = 1}^{t^2_+}\tanh\big(\tfrac{x-x^j_+}2\big) \bigg)\bigg(\prod_{j = 1}^{t^2_-} \tanh\big(\tfrac{x-x^j_-}2\big)\bigg)\cdot \ell^2(x).\end{aligned}$$
The functions $\ell^1(x)$, $\ell^2(x)$, $\mathfrak A^1(x)$ and $\mathfrak A^2(x)$ are analytic and free of zeros in their corresponding analyticity strips, and satisfy \^1(x-2)\^1(x+2) = A\^2(x), \^2(x-2)\^2(x+2) = A\^1(x). For $\gamma$ generic, the four functions have constant asymptotics, so we can define the Fourier transforms of their logarithmic derivatives:
$$\begin{aligned}
{2}
&\displaystyle L^n(k) = \frac{1}{2 \pi} \int_{-\infty}^\infty \dd x\, \big(\ln \ell^n(x)\big)'\eE^{-\ir k x}, \qquad
&& \displaystyle A^n(k) = \frac{1}{2 \pi} \int_{-\infty}^\infty \dd x\, \big(\ln \mathfrak A^n(x)\big)'\eE^{-\ir k x},\\[0.25cm]
&\displaystyle \big(\ln \ell^n(x)\big)' = \int_{-\infty}^\infty \dd k\, L^n(k)\, \eE^{\ir k x}, \qquad
&& \displaystyle \big(\ln \mathfrak A^n(x)\big)' = \int_{-\infty}^\infty \dd k\, A^n(k)\, \eE^{\ir k x}.\end{aligned}$$
Applying the Fourier transform and its inverse to , and subsequently integrating with respect to $x$, we find
\[eq:PTBA\] $$\begin{aligned}
{2}
\ln \mathfrak a^1(x) &= \ln \tanh^{N}(\tfrac x 2) + \sum_{i = 1}^{t^1_+} \ln \Big(\tanh\big(\tfrac{x-w^i_+}2\big) \Big) + \sum_{i = 1}^{t^1_-} \ln \Big(\tanh\big(\tfrac{x-w^i_-}2\big)\Big) + K \ast \ln \mathfrak A^2 + \phi^1,\\
\ln \mathfrak a^2(x) &= \sum_{j = 1}^{t^2_+} \ln \Big(\tanh\big(\tfrac{x-x^j_+}2\big) \Big) +\sum_{j = 1}^{t^2_-} \ln \Big(\tanh\big(\tfrac{x-x^j_-}2\big) \Big) + K \ast \ln \mathfrak A^1 + \phi^2,\end{aligned}$$
where $\phi^1$ and $\phi^2$ are the integration constants. The kernel $K(x)$ and the convolution are defined as for the strip, see and .
We note that for special values of $\gamma$, the asymptotics of the functions $\ell^1(x)$, $\ell^2(x)$, $\mathfrak A^1(x)$ and $\mathfrak A^2(x)$ can be zero. In this case, one should instead define the Fourier transform of the second logarithmic derivative, which results in extra linear terms of the form $\phi\, x$ in . However, this care turns out to be superfluous because the constants $\phi$ are evaluated to zero using the braid limit.
#### Scaling TBA equations.
In the scaling regime, the function $\tanh^{N}(\frac x 2)$ has the following behavior: \_[N ]{} \^[N]{}(12(x + N)) = (-1)\^N(-2 \^[-x]{}). We assume that the following scaling limits also exist: a\^n\_(x) = \_[N ]{} a\^n( (x + N)), A\^n\_(x) = \_[N ]{} A\^n( (x + N)), n = 1,2. Because the functions $\mathfrak a^n(x)$ are not symmetric under the transformation $x \rightarrow -x$, the functions $\mathsf a^n_+(x)$ and $\mathsf a^n_-(x)$ are not equal. We denote the zeros of $\mathsf a^1_\pm(x)$ and $\mathsf a^2_\pm(x)$ by $z_\pm^i$ and $y_\pm^j$, namely z\_\^i = w\_\^i - N, y\_\^j = x\_\^j - N. In the scaling regime, the non linear integral equations become
\[eq:scalNLIET\] $$\begin{aligned}
{2}
\ln \mathsf a_\pm^1(x) &= -2 \,\eE^{-x} + \sum_{i=1}^{t^1_\pm} \ln\big(\!-\!\tanh (\tfrac{x-z_\pm^i}2) \big) + K \ast \ln \mathsf A_\pm^2 +\phi^1_\pm,\\
\ln \mathsf a_\pm^2(x) &= \sum_{j=1}^{t^2_\pm} \ln\big(\!-\!\tanh (\tfrac{x-y^j_\pm}2) \big)+ K \ast \ln \mathsf A_\pm^1 + \phi^2_\pm.\end{aligned}$$
Here, we consider the non linear integral equations to be independent in the upper and lower part of the planes, thus allowing two different sets of constants, $(\phi^1_-,\phi^2_-)$ and $(\phi^1_+,\phi^2_+)$.
#### Evaluation of the constants.
For $d \equiv 0 \Mod 3$, we consider $\gamma \in (-\frac \pi3,\frac \pi3)$. On this interval, the braid limits are both positive. We fix the branches of the logarithms by imposing (-(2)) and \[eq:logbranches\] a\_\^1(x) (4 \^2 -1), a\_\^2(x) (2 ). Applying the braid limit to , we find that d 0 3: \_\^1 = -t\_\^1, \_\^2 = -t\_\^2.
For $d \equiv 1,2 \Mod 3$, the interval $\gamma \in (-\frac \pi 6, \frac \pi 6)$ splits into three cases: $\gamma \in (-\frac {\pi}6,0)$, $\gamma \in (0,\frac \pi6)$ and $\gamma = 0$. Here, $\mathsf a_\pm^2(\infty)$ is negative in all three cases. Fixing the branches of the logarithms fixes the constants $\phi^2_\pm$: a\_\^2(x) (-2 (+3)) + , \^2\_= -(t\^1\_-1). In stark contrast, $\mathsf a_\pm^1(\infty)$ takes opposite signs in the first and second interval and is zero for $\gamma = 0$. For $d \equiv 1 \Mod 3$, $\gamma \in (-\tfrac\pi6,0)$ as well as for $d \equiv 2 \Mod 3$, $\gamma \in (0,\tfrac\pi6)$, we have
[ll]{} a\^1\_-(x) (4 \^2 (-+3)-1), & \^1\_- = -t\^1\_-,\
a\^1\_+(x) (1-4 \^2 (+3)) + , & \^1\_+ = -(t\^1\_+-1).
Likewise, for $d \equiv 1 \Mod 3$, $\gamma \in (0,\tfrac\pi6)$ and $d \equiv 2 \Mod 3$, $\gamma \in (-\tfrac\pi6,0)$, we have
[ll]{} a\^1\_-(x) (1-4 \^2 (-+3)) + , &\^1\_- = -(t\^1\_–1),\
a\^1\_+(x) (4 \^2 (+3)-1), &\^1\_+ = -t\^1\_+.
For $\gamma = 0$, the result depends on the subcases $\bar \tA$ or $\bar \tB$ and $\tA$ or $\tB$:
$$\begin{aligned}
{3}
&\textrm{Subcase } \bar \tA: \quad &&\ln \mathsf a^1_-(x) \xrightarrow{x \rightarrow -\infty} -\tfrac{2x}3 + \ln \kappa_- + \ir \pi && \qquad \phi^1_- = -\ir \pi (t^1_--1), \\[0.15cm]
&\textrm{Subcase } \bar \tB: \quad &&\ln \mathsf a^1_-(x) \xrightarrow{x \rightarrow -\infty} -\tfrac{4x}3 + \ln \kappa'_-&& \qquad \phi^1_- = -\ir \pi t^1_-, \\[0.15cm]
&\textrm{Subcase } \tA: \quad &&\ln \mathsf a^1_+(x) \xrightarrow{x \rightarrow \infty} -\tfrac{2x}3 + \ln \kappa_+ + \ir \pi && \qquad \phi^1_+ = -\ir \pi (t^1_+-1), \\[0.15cm]
&\textrm{Subcase } \tB: \quad &&\ln \mathsf a^1_-(x) \xrightarrow{x \rightarrow \infty} -\tfrac{4x}3 + \ln \kappa'_+&& \qquad \phi^1_+ = -\ir \pi t^1_+.\end{aligned}$$
#### Finite-size corrections.
We apply the Fourier transform to by first removing the zeros of $\mathfrak b(x)$ on the real axis. This yields b(x) = \_[j = 1]{}\^[t\_+\^2]{} (-(2) ) + \_[j = 1]{}\^[t\_-\^2]{} ((2)) + K A\^1 + . The constant $\psi$ can be evaluated from the braid limits to be a multiple of $\pi$. It does not contribute to the finite-size corrections. These are written in terms of the scaling functions and their zeros as follows: $$\begin{aligned}
{2}
\ln \mathfrak b(x) - \psi &= \sum_{j = 1}^{t^2_-} \ln \Big(\!\tanh\big(\tfrac{x-x_-^j}2\big) \Big)+\sum_{j = 1}^{t^2_+} \ln \Big(\!-\!\tanh\big(\tfrac{x-x^j_+}2\big) \Big)\nonumber\\&\hspace{0.8cm}+ \frac{1}{2\pi} \int_{-\ln N}^\infty \dd y \Big( \frac{\ln\mathfrak A^1(y+\ln N)}{\cosh(x-y-\ln N)}+\frac{\ln\mathfrak A^1(-y-\ln N)}{\cosh(x+y+\ln N)}\Big) \label{eq:FSC1T}\\
&\hspace{-1cm} \simeq - \frac1N \bigg(2 \sum_{j=1}^{t_-^2}\eE^{-x-y_-^j} + 2 \sum_{j=1}^{t_+^2}\eE^{x-y_+^j}-\frac{\eE^{-x}}\pi \int_{-\infty}^\infty \dd y\, \eE^{-y} \ln \mathsf A_-^1(y)- \frac{\eE^{x}}\pi \int_{-\infty}^\infty \dd y\, \eE^{-y} \ln \mathsf A_+^1(y)\bigg).\nonumber\end{aligned}$$
#### Zeros of $\boldsymbol{\mathfrak a^1(x)}$ and $\boldsymbol{\mathfrak a^2(x)}$.
To rewrite $\eE^{-y_\pm^j}$ in terms of integrals involving the scaling functions, we use and find a\^2(x\^j\_) = 0 a\^1(x\^j\_-2) = -1 a\^1\_(y\_\^j-2) = -1. Using , we find $$\begin{aligned}
{2}
(2 k_\pm^j-1)\pi &= \ir \ln \mathsf a^1_\pm (y^j_\pm - \tfrac{\ir \pi }2) \nonumber \\&= 2\, \eE^{-y_\pm^j} + \ir \sum_{i=1}^{t_\pm^1} \ln\big(\!-\!\tanh \tfrac12(y_\pm^j-z_\pm^i- \tfrac{\ir\pi}2 )\big) + \frac1{2\pi} \int_{-\infty}^\infty \dd y\, \frac{\ln \mathsf A_\pm^2(y)}{\sinh(y-y_\pm^j)} + \ir \phi^1_\pm\end{aligned}$$ where the $k_\pm^j$ are integers. The terms $\ln\big(\!-\!\tanh \tfrac12(y_\pm^j-z_\pm^i- \tfrac{\ir\pi}2 )\big)$ in this last expression are rewritten using : \[eq:integerKP\] a\^1(w\^i\_) = 0 a\^2(w\^i\_-2) = -1 a\^2\_( z\^j\_-2) = -1.Similarly,$$\begin{aligned}
{2}
(2 \ell_\pm^i-1)\pi &= \ir \ln \mathsf a^2_\pm (z_\pm^i - \tfrac{\ir \pi}2) \nonumber\\
& = \ir \sum_{j=1}^{t_\pm^2} \ln \big(\!-\!\tanh \tfrac12(z_\pm^i-y_\pm^j -\tfrac{\ir \pi}2)\big) + \frac{1}{2 \pi} \int_{-\infty}^\infty \dd y \frac{\ln \mathsf A^1_\pm(y)}{\sinh(y-z_\pm^i)}+ \ir \phi_\pm^2
\label{eq:integerLP}\end{aligned}$$ where the $\ell_\pm^i$ are integers. We can then apply to each of the logarithms in . By combining , and , we obtain $$\begin{aligned}
{2}
\ln \mathfrak b(x) - \psi & \simeq -\frac{2 \pi}N \bigg(\eE^x \bigg[ \sum_{j=1}^{t_+^2}\big( k_+^j -\tfrac12 - \tfrac{\ir \phi^1_+}{2\pi}\big) + \sum_{i=1}^{t^1_+} \big(\ell_+^i-\tfrac12+\tfrac{t^2_+}2 - \tfrac{\ir \phi^2_+}{2\pi}\big) - \frac 1 {2\pi^2} \int_{-\infty}^\infty \dd y \,\eE^{-y} \ln \mathsf A_+^1(y) \nonumber\\
& \hspace{1.5cm}- \frac{1}{4\pi^2} \sum_{i=1}^{t_+^1}\int_{-\infty}^{\infty} \dd y\, \frac{\ln \mathsf A_+^1(y)}{\sinh(y-z_+^i)}- \frac{1}{4\pi^2} \sum_{j=1}^{t_+^2}\int_{-\infty}^{\infty} \dd y\, \frac{\ln \mathsf A_+^2(y)}{\sinh(y-y_+^j)} \bigg] \nonumber\\
&\hspace{1cm}+\eE^{-x} \bigg[ \sum_{j=1}^{t_-^2}\big( k_-^j -\tfrac12 - \tfrac{\ir \phi^1_-}{2\pi}\big) + \sum_{i=1}^{t^1_-} \big(\ell_-^i-\tfrac12+\tfrac{t^2_-}2 - \tfrac{\ir \phi^2_-}{2\pi}\big)- \frac 1 {2\pi^2} \int_{-\infty}^\infty \dd y \,\eE^{-y} \ln \mathsf A_-^1(y) \nonumber\\
&\hspace{1.5cm}- \frac{1}{4\pi^2} \sum_{i=1}^{t_-^1}\int_{-\infty}^{\infty} \dd y\, \frac{\ln \mathsf A_-^1(y)}{\sinh(y-z_-^i)} - \frac{1}{4\pi^2} \sum_{j=1}^{t_-^2}\int_{-\infty}^{\infty} \dd y\, \frac{\ln \mathsf A_-^2(y)}{\sinh(y-y_-^j)} \bigg]
\bigg). \label{eq:FSC2}\end{aligned}$$
#### Dilogarithm technique.
To evaluate , we consider the integrals \[eq:Jints\] J\_= \_[-]{}\^y ((a\_\^1)’ A\_\^1 -|a\_\^1| (A\_\^1)’ ) + \_[-]{}\^y ((a\_\^2)’ A\_\^2 -|a\_\^2| (A\_\^2)’ ) where $\ln |\mathsf a_\pm^1|$ and $\ln |\mathsf a_\pm^2|$ are real for all $x$ and given by |a\_\^n|(x) = a\^n\_(x) + \_\^n(x), n = 1,2. The functions $\theta_+^n(x)$ and $\theta_-^n(x)$ are step functions defined for $x \in \mathbb R$ with the following behavior: starting at $x = \infty$ and progressing to the left on the real $x$ axis, they decrease by $\ir\pi$ each time a zero of the corresponding type is crossed ($z_\pm^i$ for $\theta^1_\pm(x)$ and $y_\pm^j$ for $\theta^2_\pm(x)$). The values at the endpoints are \_\^1(x) = 0 x > z\_\^1, \_\^2(x) = 0 x > y\_\^1.
The integrals $\mathcal J_\pm$ can be evaluated in two ways. For the first, one uses the non-linear integral equations and the symmetries of $K(x)$ and obtains $$\begin{aligned}
{2}
\mathcal J_\pm &= 2\int_{-\infty}^\infty \dd y \, \eE^{-y} \big(\ln \mathsf A_\pm^1 + (\ln \mathsf A_\pm^1)'\big) + \int_{-\infty}^\infty \dd y \sum_{i=1}^{t_\pm^1} \Big[\ln\big(\!-\! \tanh (\tfrac{y-z_\pm^i}2)\big) \Big]' \ln \mathsf A_\pm^1 \nonumber\\
& - \int_{-\infty}^\infty \dd y \Big(\sum_{i=1}^{t_\pm^1} \ln \big(\!-\!\tanh (\tfrac{y-z_\pm^i}2)\big) - \theta_\pm^1(y) - \ir \pi t_\pm^1\Big) \big(\ln \mathsf A_\pm^1\big)' + \int_{-\infty}^\infty \dd y \sum_{j=1}^{t_\pm^2} \Big[\ln \big(\!-\!\tanh (\tfrac{y-y_\pm^j}2)\big) \Big]' \ln \mathsf A_\pm^2 \nonumber\\
& - \int_{-\infty}^\infty \dd y \Big(\sum_{j=1}^{t_\pm^2} \ln \big(\!-\!\tanh (\tfrac{y-y_\pm^j}2)\big) - \theta_\pm^2(y) - \ir \pi t_\pm^2\Big) \big(\ln \mathsf A_\pm^2\big)'.\end{aligned}$$ The integrals involving derivatives of $\ln \mathsf A_\pm^1$ and $\ln \mathsf A_\pm^2$ are transformed using integration by parts. For each one, it can be argued using the non-linear integral equations that the surface terms are zero. This yields J\_= 4 \_[-]{}\^y \^[-y]{} A\_\^1(y) + 2 \_[i=1]{}\^[t\_\^1]{} \_[-]{}\^y + 2\_[j=1]{}\^[t\_\^2]{} \_[-]{}\^y , which is precisely the combination of integrals needed to compute the finite-size corrections: $$\begin{aligned}
{2}\label{eq:finalfscP}
\ln \mathfrak b(x) -\psi \simeq -\frac{2\pi}N \bigg(&\eE^x \Big[\sum_{j=1}^{t_+^2}\big( k_+^j -\tfrac12 - \tfrac{\ir \phi^1_+}{2\pi}\big) + \sum_{i=1}^{t^1_+} \big(\ell_+^i-\tfrac12+\tfrac{t^2_+}2 - \tfrac{\ir \phi^2_+}{2\pi}\big) - \frac{\mathcal J_+}{8 \pi^2}\Big] \nonumber\\
&+ \eE^{-x} \Big[ \sum_{j=1}^{t_-^2}\big( k_-^j -\tfrac12 - \tfrac{\ir \phi^1_-}{2\pi}\big) + \sum_{i=1}^{t^1_-} \big(\ell_-^i-\tfrac12+\tfrac{t^2_-}2 - \tfrac{\ir \phi^2_-}{2\pi}\big) - \frac{\mathcal J_-}{8 \pi^2}\Big]\bigg).\end{aligned}$$ The second way of performing the integrals consists in changing the integration variable from $y$ to $\mathsf a_\pm$. The asymptotic behavior of the scaling functions is a\^1\_(-) = 0, a\^2\_(-) = , a\^1\_() = 4 \^2(+3)-1, a\^2\_() = 2 (+3) with $\sigma \in \{-1,+1\}$. For the integrals involving $\mathsf a^1_\pm$, we find $$\begin{aligned}
{2}
\int_{-\infty}^\infty \dd y \Big((&\ln \mathsf a_\pm^1)' \ln \mathsf A_\pm^1 -\ln |\mathsf a_\pm^1| (\ln \mathsf A_\pm^1)' \Big) = \int_{-\infty}^\infty \dd y\, \frac{\dd \mathsf a_\pm^1}{\dd y} \,\bigg( \frac{\ln(1+ \mathsf a_\pm^1)}{\mathsf a_\pm^1} - \frac{\ln |\mathsf a_\pm^1|}{1+ \mathsf a_\pm^1} \bigg)\nonumber\\
& = \bigg[\int_{-\infty}^{z^\pm_{t_\pm^1}}+\int_{z^\pm_{t_\pm^1}}^{z^\pm_{t_\pm^1-1}}+ \dots + \int_{z^\pm_2}^{z^\pm_1}+ \int_{z^\pm_1}^{\infty}\bigg] \dd y\, \frac{\dd \mathsf a_\pm^1}{\dd y}\, \bigg( \frac{\ln(1+ \mathsf a_\pm^1)}{\mathsf a_\pm^1} - \frac{\ln |\mathsf a_\pm^1|}{1+ \mathsf a_\pm^1} \bigg) \nonumber\\
& = \bigg[\int_{0}^{0}+\int_{0}^{0}+ \dots + \int_{0}^{0}+ \int_{0}^{\mathsf a^1_\pm(\infty)}\bigg] \dd \mathsf a \,\Big( \frac{\ln(1+ \mathsf a)}{\mathsf a} - \frac{\ln |\mathsf a|}{1+ \mathsf a} \Big) \nonumber\\
& =\int_{0}^{\mathsf a^1_\pm(\infty)} \dd \mathsf a \,\Big( \frac{\ln(1+ \mathsf a)}{\mathsf a} - \frac{\ln |\mathsf a|}{1+ \mathsf a} \Big). \label{eq:J1finalper}\end{aligned}$$ Recalling that $\mathsf A^2_\pm = (1 + \eE^{3 \ir \gamma}\mathsf a^2_\pm)(1 + \eE^{-3 \ir \gamma}\mathsf a^2_\pm)$, the integrals involving $\mathsf a^2_\pm$ are obtained with the same arguments. We find $$\begin{aligned}
{2}
\int_{-\infty}^\infty \dd y \Big((\ln \mathsf a_\pm^2)' \ln \mathsf A_\pm^2 -\ln |\mathsf a_\pm^2| (\ln \mathsf A_\pm^2)' \Big) &= \int_{\sigma}^{\mathsf a^2_\pm(\infty)} \dd \mathsf a \,\Big( \frac{\ln(1+ \eE^{3 \ir \gamma}\mathsf a)}{\mathsf a} - \frac{\eE^{3 \ir \gamma}\ln |\mathsf a|}{1+ \eE^{3 \ir \gamma} \mathsf a} \Big) \nonumber\\ &+ \int_{\sigma}^{\mathsf a^2_\pm(\infty)} \dd \mathsf a \,\Big( \frac{\ln(1+ \eE^{-3 \ir \gamma}\mathsf a)}{\mathsf a} - \frac{\eE^{-3 \ir \gamma}\ln |\mathsf a|}{1+ \eE^{-3 \ir \gamma} \mathsf a} \Big). \label{eq:J2finalper}\end{aligned}$$ The sum of the integrals and is expressed as J\_= {
[ll]{} K\_() & d 0 3,\
K\_(3) & d 1 3,\
K\_(3) & d 2 3,\
. \[eq:JpmT\] where $\mathcal K_\sigma (\gamma)$ is defined and evaluated in \[sec:Rogers\]. Comparing with specialised to $c=0$, we find that the first and second bracket in are identified with $\Delta$ and $\bar\Delta$:
$$\begin{aligned}
{2}
\Delta &= \sum_{j=1}^{t_+^2}\big( k_+^j -\tfrac12 - \tfrac{\ir \phi^1_+}{2\pi}\big) + \sum_{i=1}^{t^1_+} \big(\ell_+^i-\tfrac12+\tfrac{t^2_+}2 - \tfrac{\ir \phi^2_+}{2\pi}\big) - \frac{\mathcal J_+}{8 \pi^2}, \\
\bar \Delta &= \sum_{j=1}^{t_-^2}\big( k_-^j -\tfrac12 - \tfrac{\ir \phi^1_-}{2\pi}\big) + \sum_{i=1}^{t^1_-} \big(\ell_-^i-\tfrac12+\tfrac{t^2_-}2 - \tfrac{\ir \phi^2_-}{2\pi}\big) - \frac{\mathcal J_-}{8 \pi^2},\end{aligned}$$
where the integers $k_\pm^j$ and $\ell_\pm^i$ are given in terms of the eigenvalue and its zeros by and .
#### Specialisation to $\boldsymbol{\gamma = 0}$.
For $\gamma=0$, we have $\mathcal J_+ = \mathcal J_-$ and
\[eq:Deltagamma0\] = \_[j=1]{}\^[t\^2\_+]{}k\_+\^j+\_[i=1]{}\^[t\^1\_+]{}\_+\^i + \_+ + (), |= \_[j=1]{}\^[t\^2\_-]{}k\_-\^j+\_[i=1]{}\^[t\^1\_-]{}\_-\^i + \_- + () with $$\begin{aligned}
{2}
\tau_\pm &=
\left\{\begin{array}{cl}
-\frac12(t^1_\pm + t^2_\pm + t^1_\pm t^2_\pm) & d\equiv 0 \Mod 3,\\[0.1cm]
-\frac12t^1_\pm t^2_\pm & d \equiv 1,2 \Mod 3\, (\bar \tA \textrm{ or } \tA),\\[0.1cm]
-\frac12(t^2_\pm + t^1_\pm t^2_\pm) & d \equiv 1,2 \Mod 3\, (\bar \tB \textrm{ or } \tB)
\end{array}\right. \\[0.2cm]
\iota(\sigma) &= -\frac{\mathcal J_\pm}{8\pi^2} =
\left\{\begin{array}{cll}
-\frac1{24} &d\equiv 0 \Mod 3,& \sigma = 1,\\[0.15cm]
-\frac1{6} &d\equiv 0 \Mod 3,& \sigma = -1,\\[0.15cm]
\frac1{8} & d\equiv 1,2 \Mod 3,& \sigma = 1,\\[0.15cm]
0 & d\equiv 1,2 \Mod 3,& \sigma = -1.\\[0.15cm]
\end{array}\right.\end{aligned}$$
Solution for the ground states {#sec:Pcfgs}
------------------------------
In this section, we study the conformal weights of the ground states of each $\stanp_N^d$ for $\gamma = 0$. For the ground state of $\stanp_N^d$, we observe that the pattern of zeros is symmetric with respect to the horizontal axis, implying that $\Delta = \bar
\Delta$. The analysis of the bulk behavior of these eigenvalues (using the method discussed below ) reveals that $\sigma = 1$ for $N$ even and $\sigma = -1$ for $N$ odd. Moreover, for $d \equiv 1,2 \text{ mod } 3$, we observe that the ground state eigenvalue respectively belongs to the subcases $(\bar\tB,\tB)$ and $(\bar\tA,\tA)$. For $N$ even, the zero patterns are characterised by
\[eq:datat1t2\] \[eq:datat1t2even\] t\^1\_= {
[cl]{} 6 & d 06,\
6 & d 26,\
6 & d 46,
.t\^2\_= {
[cl]{} 3 & d 06,\
3 & d 26,\
3 & d 46.
. For $N$ odd, we work with the convention described in \[sec:Pprops\] that the single zeros on the axis are in the upper-half plane. We stress however that the opposite convention produces the same conformal weights. With this convention, we have \[eq:datat1t2odd\] t\^1\_+ = {
[cl]{} 6 & d 16,\
6 & d 36,\
6 & d 56,
.t\^1\_- = {
[cl]{} 6 & d 16,\
6 & d 36,\
6 & d 56,
.t\^2\_= {
[cl]{} 3 & d 16,\
3 & d 36,\
3 & d 56.
.
In both cases, there are also $\frac{N-d}2$ pairs of zeros on the boundary edges of the analyticity strip of $T^1(u)$, split between the upper and lower halves, see the examples in \[fig:GSpatternsp\]. Each zero in the center of the analyticity strip of $T^2(u)$ is of order one and is joined, at the same height, by a pair of single zeros sitting on the edges of the analyticity strip. The integers $k^j_\pm$ and $\ell^i_\pm$ are not fixed by the technique of \[sec:Pfsc\]. We estimate them using our computer program by evaluating and on small system sizes. For the ground state of $\stanp_N^d$, we find: \[eq:kdata\] k\^j\_= {
[cl]{} j & d 03,\
j & d 13,\
j-1 & d 23,
. \^i\_= i.
The patterns of zeros for the ground states for $N=12,13$ are given in \[fig:GSpatternsp\]. In this figure, for $d=1$, the analyticity strip of $T^2(u)$ for the Razumov-Stroganov eigenstate is colored in gray, indicating that $T^2(u)=0$. The zeros on the separation lines are divided between the upper and lower half-planes using the prescription discussed in \[sec:Pprops\].
Inserting the data and into , we find that in all cases, the ground state conformal weights are given by = |= = \_[0,d 2]{}.
Solution for all the eigenvalues {#sec:fsgf}
--------------------------------
In this section, we describe the full spectrum in the standard modules $\stanp_N^d$ for $\gamma = 0$. We compute the finitized spectrum generating functions \[eq:ZdNP\] = \_ q\^|q\^[|]{}, where $\Delta$ and $\bar \Delta$ are the conformal weights of the limiting conformal states, $\varepsilon$ is the overall sign of the eigenvalue and the sum is over eigenstates characterised by the patterns of zeros. Indeed, for periodic boundary conditions, the eigenvalues can have either a positive or a negative overall sign. This sign is crucial in \[sec:mipf\] for the partition function on the $M\times N$ torus with $M$ odd. Concretely, $\varepsilon$ is obtained from the bulk limit $K_{\textrm{bulk}} = \sigma$ and the vertical lattice width $M$ as $\varepsilon = \sigma^M$.
As for the strip, our derivation is built on a set of conjectured selection rules for the patterns of zeros in $\stanp_N^d$. This conjecture is supported by data produced with our computer implementation of the transfer matrices and the computation of the eigenvalues and patterns of zeros, for $1 {\leqslant}N {\leqslant}12$. We present data for $\stanp_6^0$ and $\stanp_6^2$ in \[fig:patsP60,fig:patsP62\].
#### Selection rules for the patterns of zeros.
The selection rules are expressed in terms of triples $(\sigma, \sigma',\sigma'')$ of single-column diagrams, with $\sigma \in \Aone{M}{m}$, $\sigma' \in \Aone{L}{n}$ and $\sigma'' \in \Aone{L}{\ell}$. We denote by $\BB{M}{L}{m}{n}{\ell}$ the set of such triples. In the selection rule, the configuration $\sigma$ describes the content in zeros of the first strip, whereas the pair $(\sigma', \sigma'')$ describes the content of the second strip. Crucially, in contrast with the boundary case, the constraint of dominance is [*not*]{} imposed on the pair $(\sigma', \sigma'')$. The cardinality of $\BB{M}{L}{m}{n}{\ell}$ is therefore simply given by the product of the cardinalities of $\Aone{M}{m}$, $\Aone{L}{n}$ and $\Aone{L}{\ell}$: \[eq:cardB\] || = . For instance, $\BB{4}{2}{3}{1}{1}$ contains 16 configurations. These are precisely the configurations appearing in $\AA{4}{2}{3}{1}{1}$ given in \[fig:A42311\], except that those on the second row are twice degenerate. For $d \equiv 1,2 \Mod 3$, we include $(\bar \tA, \tA)$, $(\bar \tA, \tB)$, $(\bar \tB, \tA)$ or $(\bar \tB, \tB)$ in the upper labels of $\BB{M}{L}{m}{n}{\ell}$ to specify which subcase the corresponding eigenvalues belong to.
We give a conjecture for the full content of zeros in the analyticity strips, with the separation between the upper and lower planes discussed below. For example, the patterns for $\stanp_6^0$ and $\stanp_6^2$ given in \[fig:patsP60,fig:patsP62\] are encoded by the following sets of column configurations:
$$\begin{aligned}
{2}
\stanp_6^0&: \quad \BB{3}{0}{0}{0}{0}\,\cup\,\BB{3}{1}{2}{0}{0}\,\cup\,\BB{4}{1}{2}{1}{1}\,\cup\,\BB{4}{2}{4}{1}{1} \,\cup\,\BB{5}{2}{4}{2}{2}\,\cup\,\BB{6}{3}{6}{3}{3},\\[0.2cm]
\stanp_6^2&:\quad\BBAB{4}{0}{2}{0}{0}{\bar \tA,\tA}\,\cup\,\BBAB{4}{1}{4}{0}{0}{\bar \tA,\tA}\,\cup\,\BBAB{5}{1}{4}{1}{1}{\bar\tA,\tA}\,\cup\,\BBAB{6}{2}{6}{2}{2}{\bar\tA,\tA}\,\cup\,\BBAB{4}{1}{4}{0}{1}{\bar\tA,\tB} \,\cup\,\BBAB{4}{1}{4}{0}{1}{\bar\tB,\tA}.\end{aligned}$$
More generally, for $d\neq 1$, we conjecture that the full set of patterns of zeros in $\stanp_N^d$ is described by the following sets:
\[eq:SRp\] $$\begin{aligned}
{2}
d=3t&: \hspace{0.2cm}\bigcup_{i=0}^{\frac{N-d}2}\,\, \bigcup_{j=0}^{\big\lfloor\frac12\big(\frac{N-d}2-i\big)\big\rfloor} \BB{\frac{N+t}2 + i}{i+j+t}{2(i+j+t)}{i}{i+t} \ ,\label{eq:SRp0t}
\\[0.2cm]
d=3t+1&: \hspace{0.2cm}\bigcup_{i=0}^{\frac{N-d}2}\,\, \bigcup_{j=0}^{\big\lfloor\frac12\big(\frac{N-d}2-i\big)\big\rfloor} \BBAB{\frac{N+t-1}2 + i}{i+j+t-1}{2(i+j+t)}{i}{i+t-1}{\bar\tB,\tB}
\cup\bigcup_{i=0}^{\frac{N-d-2}2}\,\, \bigcup_{j=0}^{\big\lfloor\frac12\big(\frac{N-d-2}2-i\big)\big\rfloor} \BBAB{\frac{N+t+1}2 + i}{i+j+t}{2(i+j+t+1)}{i}{i+t}{\bar\tA,\tB} \nonumber\\
&\cup\bigcup_{i=0}^{\frac{N-d-2}2}\,\, \bigcup_{j=0}^{\big\lfloor\frac12\big(\frac{N-d-2}2-i\big)\big\rfloor} \BBAB{\frac{N+t+1}2 + i}{i+j+t}{2(i+j+t+1)}{i}{i+t}{\bar\tB,\tA} \cup
\bigcup_{i=0}^{\frac{N-d-4}2}\,\, \bigcup_{j=0}^{\big\lfloor\frac12\big(\frac{N-d-4}2-i\big)\big\rfloor} \BBAB{\frac{N+t+3}2 + i}{i+j+t+1}{2(i+j+t+2)}{i}{i+t+1}{\bar\tA,\tA}\ ,
\label{eq:SRp1t}
\\[0.2cm]
d=3t+2&: \hspace{0.2cm}\bigcup_{i=0}^{\frac{N-d}2}\,\, \bigcup_{j=0}^{\big\lfloor\frac12\big(\frac{N-d}2-i\big)\big\rfloor} \BBAB{\frac{N+t+2}2 + i}{i+j+t}{2(i+j+t+1)}{i}{i+t}{\bar\tA,\tA}
\cup \bigcup_{i=0}^{\frac{N-d-4}2}\,\, \bigcup_{j=0}^{\big\lfloor\frac12\big(\frac{N-d-4}2-i\big)\big\rfloor} \BBAB{\frac{N+t+2}2 + i}{i+j+t+1}{2(i+j+t+2)}{i}{i+t+1}{\bar\tA,\tB} \nonumber\\
&\cup \bigcup_{i=0}^{\frac{N-d-4}2}\,\, \bigcup_{j=0}^{\big\lfloor\frac12\big(\frac{N-d-4}2-i\big)\big\rfloor} \BBAB{\frac{N+t+2}2 + i}{i+j+t+1}{2(i+j+t+2)}{i}{i+t+1}{\bar\tB,\tA}
\cup \bigcup_{i=0}^{\frac{N-d-8}2}\,\, \bigcup_{j=0}^{\big\lfloor\frac12\big(\frac{N-d-8}2-i\big)\big\rfloor} \BBAB{\frac{N+t+2}2 + i}{i+j+t+2}{2(i+j+t+3)}{i}{i+t+2}{\bar\tB,\tB}\ .
\label{eq:SRp2t}\end{aligned}$$ For $d=1$, we instead have $$\begin{aligned}
{2}
d=1: \quad&\BB{0}{0}{0}{0}{0} \hspace{0.1cm}\cup\hspace{0.1cm} \bigcup_{i=1}^{\frac{N-1}2}\,\, \bigcup_{j=1}^{\big\lfloor\frac12\big(\frac{N-1}2-i\big)\big\rfloor} \BBAB{\frac{N-1}2 + i}{i+j-1}{2(i+j)}{i}{i-1}{\bar\tB,\tB}
\hspace{0.1cm}\cup\hspace{0.1cm} \bigcup_{i=0}^{\frac{N-3}2}\,\, \bigcup_{j=0}^{\big\lfloor\frac12\big(\frac{N-3}2-i\big)\big\rfloor} \BBAB{\frac{N+1}2 + i}{i+j}{2(i+j+1)}{i}{i}{\bar\tA,\tB} \nonumber\\
&
\cup\hspace{0.1cm} \bigcup_{i=0}^{\frac{N-3}2}\,\, \bigcup_{j=0}^{\big\lfloor\frac12\big(\frac{N-3}2-i\big)\big\rfloor} \BBAB{\frac{N+1}2 + i}{i+j}{2(i+j+1)}{i}{i}{\bar\tB,\tA}
\hspace{0.1cm}\cup\hspace{0.1cm} \bigcup_{i=0}^{\frac{N-5}2}\,\, \bigcup_{j=0}^{\big\lfloor\frac12\big(\frac{N-5}2-i\big)\big\rfloor} \BBAB{\frac{N+3}2 + i}{i+j+1}{2(i+j+2)}{i}{i+1}{\bar\tA,\tA}\ .
\label{eq:SRp1}\end{aligned}$$
Using , one can use binomial identities to show that the sum of the cardinalities of the above sets equals $\dim \stanp_N^d$ in all cases.
#### Separation between upper and lower half-planes.
Writing down explicit expressions for the spectrum generating functions requires understanding how the zeros are split between the upper and lower halves of the plane. Let us denote by $(\Abarone{M_1}{m_1} \smallsep\Aone{M_2}{m_2})$ the set of single-column diagrams in $\Aone{M_1+M_2}{m_1+m_2}$ for which the subconfigurations of the lower and upper half planes respectively belong to $\Aone{M_1}{m_1}$ and $\Aone{M_2}{m_2}$. The bar in $\Abarone{M_1}{m_1}$ is a reminder that this factor contributes powers of $\bar q$ in $\ZdNP$.
We first describe the splitting of the heights in the first strip. A configuration of zeros of the first strip is described by a single-column diagram of $\Aone{M}{m}$. From the selection rules, we see that for the leading eigenvalues, $M$ is a number that grows linearly with the system size $N$, whereas $m$ remains small.[^4] The zeros of these leading eigenvalues lie at a distance $\sim \ln N$ from the real axis and the separating line. There is thus some arbitrariness in the choice of the position of the separation. One can choose \_[k\_1 = 0]{}\^m ( ,) \[eq:k1splitting\] where $\epsilon$ is an arbitrary fixed number that is much smaller than $M$. The union over $k_1$ allows for the zeros to be split between the upper and lower half-planes in any possible way. Depending on $\epsilon$, some eigenvalues are treated differently in terms of the contribution of their zeros to $\Delta$ and $\bar \Delta$. Indeed for finite $N$, the numbers $t^2_-$ and $t^2_+$ of zeros in each half-plane depends on $\epsilon$, and likewise for the resulting conformal weights . However, this does not occur for the leading eigenstates, namely those that correspond to the conformal states in the scaling limit. The finitized partition functions given below depend on $\epsilon$, but by varying $\epsilon$ (while keeping $\epsilon \ll M$), the resulting expressions only change by powers of $q$ and/or $\bar q$ that are linear in $N$. For $\epsilon \ll M$, the scaling limits of these partition functions are therefore independent of $\epsilon$. We note that the data presented in \[fig:patsP60,fig:patsP62\] corresponds to $\epsilon = 0$.
Crucially, the bulk behavior $K_{\text{bulk}}=\sigma$ is a function of the variable $k_1$, which controls the separation between upper and lower halves in : \[eq:sigmak1\] = {
[ll]{} (-1)\^[k\_1]{}& d 0 3,\
(-1)\^[k\_1+1]{} & d 1,2 3.\
. The value of $k_1$ therefore dictates both the selection of $\iota(\sigma)$ in and the sign $\varepsilon$ in .
The second strip does not allow as much arbitrariness in the choice of the separation line. This is because for the leading eigenvalues, the number of sites in the second strip does not scale with the system size $N$. Let us denote by $[(\Abarone{M_1}{m_1} \smallsep\Aone{M_2}{m_2})(\Abarone{L_1}{n_1} \smallsep\Aone{L_2}{m_2})(\Abarone{P_1}{\ell_1} \smallsep\Aone{P_2}{\ell_2})]$, with $L_1+L_2 = P_1 + P_2$, the set of configurations $(\sigma, \sigma', \sigma '')$ in $\BB{M_1+M_2}{L_1+L_2}{m_1+m_2}{n_1+n_2}{\ell_1+\ell_2}$ for which $\sigma \in (\Abarone{L_1}{n_1} \smallsep\Aone{L_2}{m_2})$, $\sigma' \in (\Abarone{L_1}{n_1} \smallsep\Aone{L_2}{m_2})$ and $\sigma'' \in (\Abarone{P_1}{\ell_1} \smallsep\Aone{P_2}{\ell_2})$. We discuss the cases $d \equiv 0 \Mod 3$ and $d \equiv 1,2 \Mod 3$ separately.
For $d \equiv 0 \Mod 3$, we note that all sets in the selection rule are of the form $\BB{M}{L}{2L}{n}{\ell}$. The number of occupied sites in the first strip is twice the number of sites in the second strip. From our numerical data, we find that the splitting of the zeros in the second strip follows the same rule: If $\sigma$ has $k_1$ zeros in the lower half of the first strip, then the total number of sites in the lower half-plane of the second strip (namely those of $\sigma'$ plus those of $\sigma''$) is also equal to $k_1$. The splitting is therefore as follows: \[eq:sep0\] \_[k\_1 = 0]{}\^[2L]{}\_[k\_2 = 0]{}\^[n]{}\_[k\_3 = 0]{}\^ . Thus, if $k_1$ is even, the sites are split evenly between $\sigma'$ and $\sigma''$. If $k_1$ is odd, we choose the lower half-plane of $\sigma''$ to have an extra site compared to the lower half-plane of $\sigma'$. Had we chosen the opposite convention, the finitized characters given below would be slightly different, but their scaling limits would be identical.
For $d \equiv 1,2 \Mod 3$, the sets in the selection rules and are of the form $\BBAB{M}{L}{2(L+1)}{n}{\ell}{\tX,\tY}$ with $\tX, \tY \in \{\tA, \tB\}$. In this case, the splitting is such that the total number of sites in the second strip equals the number of occupied sites in the lower half of the first strip, minus one: \[eq:sep12\] \_[k\_1 = 0]{}\^[2(L+1)]{}\_[k\_2 = 0]{}\^[n]{}\_[k\_3 = 0]{}\^ .
In computing $\ZdNP$, each $[(\Abarone{M_1}{m_1} \smallsep\Aone{M_2}{m_2})(\Abarone{L_1}{n_1} \smallsep\Aone{L_2}{m_2})(\Abarone{P_1}{\ell_1} \smallsep\Aone{P_2}{\ell_2})]$ will contribute \[eq:Bgeneratingfunction\] |q\^[|\_]{}q\^[\_]{} where $\Delta_{\textrm{min}}$ and $\bar\Delta_{\textrm{min}}$ are the conformal weights of the minimal configurations.
#### Selection rules for the integers and spectrum generating functions.
Remarkably, the prescription for the integers $k^j_\pm$ and $\ell^i_\pm$ is very similar to the one found for the strip. We formulate it based on data collected from our computer implementation, namely approximations to $k^j_\pm$ and $\ell^i_\pm$ obtained using and , for all the eigenvalues in $\stanp_N^d$ for $N{\leqslant}12$. In \[fig:patsP60,fig:patsP62\], the values of these integers are displayed with each pattern of zeros for $\stanp_6^0$ and $\stanp_6^2$. The prescription is as follows. For any configuration in $\BB{M}{L}{m}{n}{\ell}$,
- the heights of the single-column diagram (corresponding to the $k^j_\pm$) are labelled from $0$ to $M-1$ for $d\equiv 1,2 \Mod 3\, (\tA)$ or $(\bar \tA)$. They are labeled from $1$ to $M$ for $d\equiv 0 \Mod 3$, and likewise for $d\equiv 1,2 \Mod 3\, (\tB)$ or $(\bar \tB)$;
- the heights of the double-column diagram (corresponding to the $\ell^i_\pm$) are labelled from $1$ to $L$ in all cases.
Using the selection rules and the formula for the conformal weights, we can formulate expressions for $\ZdNP$ by writing down the generating function for each set in and taking their sum over $i$ and $j$. Because of the similarities between the current prescription and the one in \[sec:excitations\] for the strip, the energies of the minimal configuration of each set in is obtained from under the suitable specifications of $m, n$ and $\ell$. The resulting expressions are complicated power-law series in $q$ and $\bar q$, involving products of three $q$-binomials and three $\bar q$-binomials, with sums over five integers: $i$, $j$, $k_1$, $k_2$ and $k_3$. Moreover for $d \equiv 1,2 \Mod 3$, there are four contributions to the generating functions depending on the subcases $\bar \tA$ or $\bar \tB$ and $\tA$ or $\tB$. The resulting expressions are collected in \[sec:Pcpf\].
#### Scaling behavior.
Unlike in the boundary case, for the periodic case we are unable to directly simplify the expressions for the finitized conformal partition functions. We are nevertheless able to perform such simplifications in the scaling limit: Z\_d(q,|q). Indeed, in \[sec:Pcpf\], we use the $q$-binomial identities of \[sec:qids\] to extract the following expressions for $Z_d(q,\bar q)$:
\[eq:Zdresults\] $$\begin{aligned}
{2}
&d \equiv 0 \Mod 3 : \quad
&&Z_d(q,\bar q) =
\frac1{(q)_\infty(\bar q)_\infty} \sum_{\ell\in \mathbb Z} \Big(q^{\Delta_{0,3\ell -d }} \bar q^{\Delta_{0,3 \ell}} + (-1)^M q^{\Delta_{1,3\ell - d}} \bar q^{\Delta_{1,3 \ell}} \Big),\label{eq:Z0results}\\[0.2cm]
&d \equiv 1 \Mod 3 : \quad
&&Z_d(q,\bar q) =
\frac1{(q)_\infty(\bar q)_\infty} \sum_{\ell\in \mathbb Z} \Big( q^{\Delta_{0,3\ell -d +2}} \bar q^{\Delta_{0,3 \ell+2}} + (-1)^Mq^{\Delta_{1,3\ell - d+2}} \bar q^{\Delta_{1,3 \ell+2}} \Big),\label{eq:Z1results}\\[0.2cm]
&d \equiv 2 \Mod 3 : \quad
&&Z_d(q,\bar q) =
\frac1{(q)_\infty(\bar q)_\infty} \sum_{\ell\in \mathbb Z} \Big(q^{\Delta_{0,3\ell -d +1}} \bar q^{\Delta_{0,3 \ell+1}} + (-1)^Mq^{\Delta_{1,3\ell - d+1}} \bar q^{\Delta_{1,3 \ell+1}} \Big).\label{eq:Z2results}\end{aligned}$$
Torus partition functions {#sec:mipf}
-------------------------
Computing the partition function of a model described by the periodic Temperley-Lieb algebra on the $M\times N$ torus requires a proper understanding of the representation theory of this algebra. This was achieved for the $Q$-Potts model by Jacobsen and Richard [@RJ07] and by Aufgebauer and Klümper for quantum spin chains [@AK10]. For the loop model, we set the weight of a non-contractible loop to $\alpha$, independent of its winding numbers around the torus. For $d >0$, let us denote by $\Tr_{d,j}$ the coefficients of the trace of $\Tb(u)^M$ on $\stanp_N^d$ in an expansion in $\omega$: \[eq:traceofT\] (u)\^M|\_[\_N\^d]{} = \_[j Z]{} \_[d,j]{} \^j. For finite $M$, the trace of $\Tb(u)^M$ is a Laurent polynomial in $\omega$, so there are finitely many contributions in the sum on the right-hand side of . The partition function of the loop model for arbitrary $\alpha$ and $\beta$ then reads [@MDPR13] = \_[N0 2]{} (u)\^M|\_[\_N\^0]{} +2 \_ \_[jZ]{} T\_[j d]{}(2) \_[d,j]{} , where $T_k(x)$ is the $k$-th Chebyshev polynomial of the first kind and $j \wedge d$ is the greatest common divisor of $j$ and $d$. Even if our current derivation does not reveal this, we know that $\ZtorMN\big|_{\alpha=1}=1$ for all parities of $M$ and $N$. For $\alpha = 2$, a remarkable simplification occurs: $T_{j\wedge d}(1) = 1$ and |\_[= 2]{} = \_ (2-\_[d,0]{}) (u)\^M|\_[\_N\^d, = 1]{} . The torus conformal partition function is then obtained from the scaling limit of the finite-size partition function: \[eq:Zmodifiedtrace\] \^[MN f\_(u)]{} |\_[= 2]{} (q,|q) = \_(2-\_[d,0]{}) with $q$ given in terms of the lattice data in .
From here, the computation splits between the odd and even parity of $N$. From , we extend the definition of $Z_d(q,\bar q)$ to $d<0$. Using the relation $Z_{d}(q,\bar q) = Z_{-d}(q,\bar q)$, we find = \_[d 2 Z]{} Z\_d(q,|q), = \_[d 2 Z + 1]{} Z\_d(q,|q). These can in turn be written in terms of the $u(1)$ characters $\varkappa_j(q) = \varkappa^6_j(q)$ given in . Defining $y = \frac1{(q)_\infty(\bar q)_\infty}$, we find the following identities for $N$ odd:
$$\begin{aligned}
{2}
&y\sum_{d \in 6 \mathbb Z+3} \sum_{\ell \in \mathbb Z} q^{\Delta_{0,3\ell -d }} \bar q^{\Delta_{0,3 \ell}}= \varkappa_0(q)\varkappa_6(\bar q) + \varkappa_6(q)\varkappa_0(\bar q),
\\
&y\sum_{d \in 6 \mathbb Z+3} \sum_{\ell \in \mathbb Z} q^{\Delta_{1,3\ell -d }} \bar q^{\Delta_{1,3 \ell}}= 2 |\varkappa_3(q)|^2,
\\
&y\Big(\sum_{d \in 6 \mathbb Z+1} \sum_{\ell \in \mathbb Z} q^{\Delta_{0,3\ell -d+2}} \bar q^{\Delta_{0,3 \ell+2}} + \sum_{d \in 6 \mathbb Z+5} \sum_{\ell \in \mathbb Z} q^{\Delta_{0,3\ell -d+1 }} \bar q^{\Delta_{0,3 \ell+1}} \Big)=
2\varkappa_2(q)\varkappa_4(\bar q) + 2\varkappa_4(q)\varkappa_2(\bar q),
\\
&y\Big(\sum_{d \in 6 \mathbb Z+1} \sum_{\ell \in \mathbb Z} q^{\Delta_{1,3\ell -d+2}} \bar q^{\Delta_{1,3 \ell+2}} + \sum_{d \in 6 \mathbb Z+5} \sum_{\ell \in \mathbb Z} q^{\Delta_{1,3\ell -d+1 }} \bar q^{\Delta_{1,3 \ell+1}} \Big)=
2|\varkappa_1(q)|^2 + 2|\varkappa_5(q)|^2.\label{papK1K5}\end{aligned}$$
For $N$ even, we find
$$\begin{aligned}
{2}
&y\sum_{d \in 6 \mathbb Z} \sum_{\ell \in \mathbb Z} q^{\Delta_{0,3\ell -d }} \bar q^{\Delta_{0,3 \ell}}= |\varkappa_0(q)|^2 + |\varkappa_6(q)|^2, \label{eq:forexample}\\
&y\sum_{d \in 6 \mathbb Z} \sum_{\ell \in \mathbb Z} q^{\Delta_{1,3\ell -d }} \bar q^{\Delta_{1,3 \ell}}= 2 |\varkappa_3(q)|^2, \\
&y\sum_{d \in 6 \mathbb Z+2} \sum_{\ell \in \mathbb Z} q^{\Delta_{0,3\ell -d+1 }} \bar q^{\Delta_{0,3 \ell+1}} + y \sum_{d \in 6 \mathbb Z+4} \sum_{\ell \in \mathbb Z} q^{\Delta_{0,3\ell -d+2}} \bar q^{\Delta_{0,3 \ell+2}} = 2|\varkappa_2(q)|^2 + 2|\varkappa_4(q)|^2, \\
&y\sum_{d \in 6 \mathbb Z+2} \sum_{\ell \in \mathbb Z} q^{\Delta_{1,3\ell -d+1 }} \bar q^{\Delta_{1,3 \ell+1}} + y\sum_{d \in 6 \mathbb Z+4} \sum_{\ell \in \mathbb Z} q^{\Delta_{1,3\ell -d+2}} \bar q^{\Delta_{1,3 \ell+2}} = 2\varkappa_1(q)\varkappa_5(\bar q) + 2\varkappa_5(q)\varkappa_1(\bar q).\label{eq:K1K5}\end{aligned}$$
These relations are obtained using simple algebraic manipulations. To illustrate, is proved as follows: $$\begin{aligned}
{2}
y\sum_{d \in 6 \mathbb Z} \sum_{\ell \in \mathbb Z} q^{\Delta_{0,3\ell -d }} \bar q^{\Delta_{0,3 \ell}}
&= y \sum_{k,\ell \in \mathbb Z} q^{\Delta_{0,3\ell -6k}} \bar q^{\Delta_{0,3 \ell}}
= y \sum_{r=0,1}\sum_{j,k \in \mathbb Z} q^{\Delta_{0,6(j-k)+3r}} \bar q^{\Delta_{0,6 j + 3r}} \nonumber\\
& = y \sum_{r=0,1}\sum_{i,j\in \mathbb Z} q^{\Delta_{0,6i+3r}} \bar q^{\Delta_{0,6 j + 3r}} = y \Big| \sum_{i\in \mathbb Z} q^{\Delta_{0,6i}} \Big|^2 + y \Big| \sum_{i\in \mathbb Z} q^{\Delta_{0,6i+3}} \Big|^2 \nonumber\\
& = |\varkappa_0(q)|^2 + |\varkappa_6(q)|^2.\label{eq:K1K1K5K5}\end{aligned}$$ For $N$ even, is the only seemingly anti-diagonal contribution. It can be changed to a diagonal contribution using the relation \[eq:K1K5relation\] \_1(q)\_5(|q) + \_5(q)\_1(|q) = |\_1(q)|\^2 + |\_5(q)|\^2-1, which itself follows from $\varkappa_1(q)-\varkappa_5(q) = 1$. Using , one can also write as an anti-diagonal contribution. Our final expressions for the conformal torus partition functions are
\[eq:subfinalZ\] $$\begin{aligned}
{2}
\Ztorodd &=\varkappa_0(q)\varkappa_6(\bar q) + 2(-1)^M \varkappa_1(q)\varkappa_5(\bar q) + 2 \varkappa_2(q)\varkappa_4(\bar q) + 2(-1)^M |\varkappa_3(q)|^2 \nonumber\\[0.1cm]
&\hspace{1cm}+ 2 \varkappa_4(q)\varkappa_2(\bar q) + 2(-1)^M \varkappa_5(q)\varkappa_1(\bar q) + \varkappa_6(q)\varkappa_0(\bar q) + 2(-1)^M, \\[0.2cm]
\Ztoreven &= |\varkappa_0(q)|^2+ 2(-1)^M |\varkappa_1(q)|^2 + 2 |\varkappa_2(q)|^2+2(-1)^M |\varkappa_3(q)|^2 \nonumber\\[0.1cm]
&\hspace{1cm}+2 |\varkappa_4(q)|^2+ 2(-1)^M |\varkappa_5(q)|^2 + |\varkappa_6(q)|^2 -2(-1)^M.\end{aligned}$$
These can be written in compact form: \[eq:finalZ\] = \_[j=0]{}\^6 ((-1)\^[M j]{} d\_j\^6 \_j(q)\_[6-j]{}(|q)) + 2(-1)\^M, = \_[j=0]{}\^6 ((-1)\^[M j]{} d\_j\^6 |\_j(q)|\^2) - 2(-1)\^M, with $d_j^n$ defined in . In terms of the $u(1)$ characters, the odd and even parities of $N$ thus correspond to different mixtures of diagonal and anti-diagonal sectors. For $M$ and $N$ even, the result agrees with the conjecture of Pearce and Rasmussen[@PRcosetGraphs11] with $n_{2,3} = -1$ so that $Z_{p,p'}(q)=Z^{\text{Circ}}_{p,p'}(q)$. It should also be noted that the partition functions for $M$ odd are not genuine conformal partition functions, since some coefficients in the $q,\bar q$ expansions are negative.
#### Modular invariance and covariance.
The $u(1)$ characters behave as follows under the $T$ and $S$ transformations of the modular group:
\[eq:transforms\] $$\begin{aligned}
\label{eq:Ttransform}
T:\qquad\varkappa^{n}_j(\eE^{2 \pi \ir(\tau+1)}) &= \exp\big(2 \pi \ir (\tfrac{j^2}{4n}-\tfrac1{24})\big) \varkappa^{n}_j(\eE^{2 \pi \ir\tau}),
\\[0.15cm]
\label{eq:Stransform}
S:\qquad\ \ \varkappa^{n}_j(\eE^{-2 \pi \ir/\tau}) &= \sum_{k=0}^{2n-1} S_{jk}\, \varkappa^{n}_k(\eE^{2 \pi \ir\tau}) = \frac{1}{\sqrt{2n}} \sum_{k=0}^{2n-1} \eE^{-\pi \ir j k/n} \varkappa^{n}_k(\eE^{2 \pi \ir\tau}).$$
It follows that diagonal and anti-diagonal terms in transform trivially under $T$:
$$\begin{aligned}
{2}
|\varkappa^{n}_j(\eE^{2 \pi \ir(\tau+1)})|^2 &= |\varkappa^{n}_j(\eE^{2 \pi \ir(\tau)})|^2, \\[0.15cm]
\varkappa^{n}_j(\eE^{2 \pi \ir(\tau+1)})\varkappa^{n}_{n-j}(\eE^{-2 \pi \ir(\tau+1)}) &= (-1)^n \varkappa^{n}_j(\eE^{2 \pi \ir\tau})\varkappa^{n}_{n-j}(\eE^{-2 \pi \ir\tau}),\end{aligned}$$
where we recall that $n = p p' = 6$ for percolation. The partition functions $\Ztorodd$ and $\Ztoreven$ are thus invariant under the action of $T$ for all parities of $M$ and $N$.
The $S$ matrix describing the transformations of the $u(1)$ characters is obtained from and $\varkappa^{n}_j(q) =\varkappa^{n}_{2n-j}(q)$. For $n= pp' = 6$, in the basis $\{\varkappa_j(q),\, j = 0, \dots, 6\}$, we have S = 1[2 3]{} (
1 & 2 & 2 & 2 & 2 & 2 & 1\
1 & & 1 & 0 & -1 & - & -1\
1 & 1 & -1 & -2 & -1 & 1 & 1\
1 & 0 & -2 & 0 & 2 & 0 & -1\
1 & -1 & -1 & 2 & -1 & -1 & 1\
1 & - & 1 & 0 & -1 & & -1\
1 & -2 & 2 & -2 & 2 & -2 & 1
). Applying the $S$ transform to , we find that the torus partition functions transform as follows:
$$\begin{aligned}
{2}\label{eq:covariance}
\Ztoroddodd (\eE^{-2 \pi \ir/\tau}) &= \Ztoroddodd (\eE^{2 \pi \ir\tau}),\\[0.15cm]
\Ztoroddeven (\eE^{-2 \pi \ir/\tau}) &= \Ztorevenodd (\eE^{2 \pi \ir\tau}),\\[0.15cm]
\Ztoreveneven (\eE^{-2 \pi \ir/\tau}) &= \Ztoreveneven (\eE^{2 \pi \ir\tau}).\end{aligned}$$
Therefore the partition functions for $M,N$ both odd or both even are modular invariant, whereas the mixed cases are modular covariant, as they map to one another under the $S$ transformation.
Conclusion {#sec:conclusion}
==========
In this paper, we derive and analyse truncated $T$- and $Y$-systems of functional relations satisfied by the transfer matrix eigenvalues of critical bond percolation considered as the loop model ${\cal LM}(2,3)$. Using analyticity properties of the eigenvalues and empirically based selections rules, we solve non-linear integral equations in the form of $D_3$ TBA equations to obtain exact expressions for the finite-size corrections and conformal data. These calculations are carried out for both the geometry of the strip with the boundaries consisting of simple half-arcs, and for periodic boundary conditions. Formulating selection rules encoding the patterns of zeros for all finite excitations, we give explicit expressions for finite-size spectrum generating functions in the various sectors. On the strip, our refined finitized characters reproduce the finitized Kac characters of [@PRZ06]. Additionally, fixing the weight of the non-contractible loops to $\alpha=2$, we obtain in the continuum scaling limit the conformal cylinder and torus partition functions.
Our expressions for the conformal cylinder partition functions are to be compared to those found using Coulomb gas arguments in [@SB89; @C06]. In particular, the partition function given in [@C06 equation 1] with $g = \frac23$ almost coincides with our result , with the difference that considers separately the partition functions for the $N$ odd and $N$ even cases, whereas the expression given in [@C06] is the sum of the two. The Coulomb gas argument is easily fixed by noting that for the dense loop model, the number of contractible loops has the same parity as $N$, and this produces the correct partition functions.
Our expressions for the conformal torus partition functions depend on the parities of the lattice dimensions $M$ and $N$. In each of the four cases, $\Ztor(q,\bar q)$ is a simple sesquilinear form in the $u(1)$ characters $\varkappa_j^6(q)$. Somewhat surprisingly, for $M,N$ both even, the MIPF Z\_[2,3]{}(q)=Z\^\_[2,3]{}(q)= |\_0(q)|\^2+ 2 |\_2(q)|\^2+2 |\_3(q)|\^2 +2 |\_4(q)|\^2+ 2\[\_1(q)\_5(|q) + \_5(q)\_1(|q)\] + |\_6(q)|\^2 is non-diagonal and not of diagonal $A$-type. Consequently, it differs from the diagonal A-type MIPF of the triplet model [@GRW10]. Nevertheless, it is of the form conjectured by Pearce and Rasmussen [@PRcosetGraphs11] as in : Z\_[p,p’]{}(q)=Z\^\_[p,p’]{}(q)+n\_[p,p’]{} Z\^\_[p,p’]{}(q),n\_[p,p’]{} \[Z=ZnZ\] since our lattice derivation shows that $n_{2,3} = -1$. Based on the assumption of a diagonal A-type MIPF, Pearce and Rasmussen instead conjectured $n_{p,p'}=2$ for all $p,p'$. Knowing what we now know about the action of the modified trace, the result $n_{2,3} = -1$ can be understood by observing that, if $Z_{2,3}(q)$ has the form in , its $q,\bar q$ expansion starts out as \[eq:Z23expansion\] Z\_[2,3]{}(q)=(q|q)\^[-1/24]{} + (1+n\_[2,3]{}) + 2 (q|q)\^[1/8]{} + 2 (q|q)\^[1/3]{} + The coefficient $n_{2,3}$ only appears in the constant term, with $(1+n_{2,3})$ counting the number of states with conformal weights $\Delta = \bar \Delta = 0$, namely the Razumov-Stroganov eigenstates.
The MIPF found from the loop model with $\alpha = 2$ and $M, N$ both even is in fact equal to the partition function of the six-vertex model on a torus with no twist. Indeed, the spin-chain representation of $\eptl_n(\alpha,\beta)$ with magnetisation $m$ has the same spectra [@MDSA13no1], including degeneracies, as the standard module $\stanp_n^d$ with $d=\frac m 2$. Because the representations with magnetisation $\pm m$ are isomorphic, taking the normal trace of $\Tb(u)$ in the full spin-chain representation precisely produces : The sector with $m=0$ is singly degenerate, whereas all others are doubly degenerate. Since there are no Razumov-Stroganov eigenstates for $N$ even in the untwisted case [@RS01; @RS01no2; @S01], the expansion implies that $n_{2,3} = -1$. Our result for $Z_d(q,\bar q)$ should then be compared with the XXZ partition function [@FSZ87; @ABGR88; @AGR89]. An explicit comparison with [@AGR89 equation 1.8] specialised to $h=\frac38$, $Q = \frac d2$, $l = 0$ and $z=q$ reveals that the partition functions coincide for $M$ even. For $M$ odd, however, incorporates factors of $(-1)^M$ to give the correct conformal partition functions for $M$ odd.
Ultimately, one of our goals is to derive the conformal torus partition functions for all the logarithmic minimal models $\mathcal{LM}(p,p')$, for all parities of $M$ and $N$. It is then natural to compare the results found here for percolation with those for critical dense polymers, corresponding to $\mathcal{LM}(1,2)$. For $M$ even, the conformal torus partition functions of this model were found in [@PRVcyl2010; @MDPR13] and expressed in terms of $\mathcal W$-irreducible characters. In \[sec:CDPpartitionfunctions\], we present the conformal partition functions for all parities of $M$ and $N$ in terms of $u(1)$ characters. For $M$ and $N$ both even, we note that the torus partition function is just $Z^{\textrm{Circ}}_{1,2}(q)$, hinting at the possibility that for $\mathcal{LM}(p,p')$, the torus partition function for $M$ and $N$ even may equal $Z^{\textrm{Circ}}_{p,p'}(q)$. For the other parities, the results for the two models are quite different and it is not possible to identify a general pattern. Clearly, more data is required before the full picture can emerge describing the conformal partition function in this case, and the structure of the indecomposable representations in general. From the lattice perspective, the study of the latter was indeed initiated in [@PRZ06] and [@RS07]. This has become the subject of an extensive investigation in the following years, and yet much is still to be understood about the indecomposable structures [@Gurarie; @GR08; @GRW10; @RGW14; @GRSV15] of bulk logarithmic conformal field theories including percolation.
This work leaves open a plethora of avenues for future research on an analytic approach to the logarithmic minimal models ${\cal LM}(p,p')$. In particular, in this paper, we only consider boundary conditions on the strip corresponding to conformal Kac modules with highest weight $\Delta_{1,s}$. However, the methods of this paper should extend to the more general Kac modules with highest weight $\Delta_{r,s}$, which are realised on the lattice by including a seam on the boundary [@PRZ06; @PTC2015; @MDRD2015]. These methods should also extend to boundary conditions described by the one- and two-boundary Temperley-Lieb algebra [@JS08; @DJS09; @PRT14; @BPT2016]. In all these cases, it is expected that the transfer matrix eigenvalues will satisfy the same universal $Y$-system, encoded by the Dynkin diagram $D_{p'}$. However, the analyticity properties will differ in the various cases, leading to the different Kac conformal weights describing the finite-size corrections. Ultimately, it would also be of interest to obtain the modular invariants of the dilute loop models. Needless to say, the analytic and combinatorial classification problems in all these challenges promise to be formidable.
Acknowledgments {#acknowledgments .unnumbered}
---------------
AMD was supported by the Belgian Interuniversity Attraction Poles Program P7/18 through the network DYGEST (Dynamics, Geometry and Statistical Physics) and by the FNRS fellowship CR28075116. AMD and AK acknowledge the support of the ERC grant [*Loop models, integrability and combinatorics*]{} and are grateful for the kind hospitality of Paul Zinn-Justin and the LPTHE where early stages of this work were done. AMD and PAP acknowledge the hospitality of the University of Wuppertal and are grateful for the kind hospitality of Holger Frahm at the ITP in Hannover where later stages of this work were done. All authors were supported by DFG through the program FOG 2316. PAP thanks the APCTP, Pohang for hospitality during the writing of this paper. The authors thank Paul Zinn-Justin, Jørgen Rasmussen and Yacine Ikhlef for useful discussions.
Integrals involving Rogers dilogarithms {#sec:Rogers}
=======================================
In this section, we study the following integrals:
\[defintegrals\] $$\begin{aligned}
{2}
\mathcal I_1 & = \int_{0}^{\mathsf a^1(\infty)} \dd \mathsf a \bigg(\frac{\ln \left(1+ \mathsf a\right)} {\mathsf a} - \frac{\ln |\mathsf a|}{1+\mathsf a}\bigg),\\
\mathcal I_2 & = \int_{\mathsf a^2(-\infty)}^{\mathsf a^2(\infty)} \dd \mathsf a \bigg(\frac{\ln \left(1+ \eE^{3\ir\gamma}\mathsf a\right)} {\mathsf a} - \frac{\eE^{3\ir\gamma}\ln |\mathsf a|}{1+\eE^{3\ir\gamma}\mathsf a}\bigg),\\
\mathcal I_3 & = \int_{\mathsf a^2(-\infty)}^{\mathsf a^2(\infty)} \dd \mathsf a \bigg(\frac{\ln \left(1+ \eE^{-3\ir\gamma}\mathsf a\right)} {\mathsf a} - \frac{\eE^{-3\ir\gamma}\ln |\mathsf a|}{1+\eE^{-3\ir\gamma}\mathsf a}\bigg),\end{aligned}$$
with a\^1() = 4 \^2-1, a\^2() = 2 , and consider two cases, namely $\mathsf a^2(-\infty) = \sigma=\pm 1$. All the integrals coming from the dilogarithm technique in \[sec:fsc,sec:Pfsc\] can be written in terms of $\mathcal K_\sigma(\gamma)$: $$\mathcal K_\sigma(\gamma)=\mathcal I_1+\mathcal I_2+\mathcal I_3.$$ Because $\mathcal K_\sigma(\gamma) = \mathcal K_\sigma(-\gamma) = \mathcal K_\sigma(\gamma+ 2\pi)$, we restrict $\gamma$ to the interval $[0,\pi]$.
We perform the calculations by taking derivatives of $\mathcal K_\sigma=\mathcal K_\sigma(\gamma)$ with respect to $\gamma$. This removes all integrals and allows for calculations with explicit rational functions and logarithms. In some instances, we need to distinguish between the two cases $\sigma=\pm 1$.
First we perform integration by parts on the second integrand in each integral and obtain
$$\begin{aligned}
{2}
\mathcal I_1 & = 2\int_{0}^{\mathsf a^1(\infty)} \dd \mathsf a \frac{\ln
\left(1+ \mathsf a\right)}{\mathsf a} -\ln(1+\mathsf a)\ln |\mathsf a|\Big|_0^{\mathsf a^1(\infty)}\ ,\\
\mathcal I_2 & = 2\int_{\mathsf a^2(-\infty)}^{\mathsf a^2(\infty)} \dd \mathsf a \frac{\ln \left(1+ \eE^{3\ir\gamma}\mathsf a\right)} {\mathsf a} - \ln\big(1+\eE^{3\ir\gamma}\mathsf a\big)\ln |\mathsf a|\Big|_{\mathsf a^2(-\infty)}^{\mathsf a^2(\infty)}\ ,\\
\mathcal I_3 & = 2\int_{\mathsf a^2(-\infty)}^{\mathsf a^2(\infty)} \dd \mathsf a \frac{\ln \left(1+ \eE^{-3\ir\gamma}\mathsf a\right)} {\mathsf a} - \ln\big(1+\eE^{-3\ir\gamma}\mathsf a\big)\ln |\mathsf a|\Big|_{\mathsf a^2(-\infty)}^{\mathsf a^2(\infty)}\ .\end{aligned}$$
Next we use a substitution of the variable of integration $\mathsf a \mapsto
\eE^{\pm 3\ir\gamma}\mathsf a$ which moves the $\gamma$ dependence from the integrand to the terminals, and use explicit expressions for the terminals in terms of $\qo=\eE^{\ir\gamma}$:
$$\begin{aligned}
{2}
\mathcal I_1 & = 2\int_{0}^{\qo^2+1+\qo^{-2}} \dd \mathsf a \frac{\ln
\left(1+ \mathsf a\right)}{\mathsf a} -\ln(1+\mathsf a)\ln |\mathsf a|\Big|_{0}^{\qo^2+1+\qo^{-2}}\ ,\\
\mathcal I_2 & = 2\int_{\sigma \qo^3}^{\qo^2+\qo^4} \dd \mathsf a \frac{\ln \left(1+\mathsf a\right)} {\mathsf a} - \ln\big(1+\qo^3\mathsf a\big)\ln |\mathsf a|\Big|_{\sigma}^{\qo+\qo^{-1}}\ ,\\
\mathcal I_3 & = 2\int_{\sigma \qo^{-3}}^{\qo^{-2}+\qo^{-4}} \dd \mathsf a \frac{\ln \left(1+\mathsf a\right)} {\mathsf a} - \ln\big(1+\qo^{-3}\mathsf a\big)\ln |\mathsf a|\Big|_{\sigma}^{\qo+\qo^{-1}}\ .\end{aligned}$$
Note that the explicit log-log terms vanish at the lower terminals. The derivatives with respect to $\qo$ read $$\begin{aligned}
{2}
\qo\frac{\dd}{\dd \qo}\mathcal K_\sigma=&
+2\frac{2 \qo^2-2 \qo^{-2}}{\qo^2+1+\qo^{-2}}\ln\big(\qo^2+2+\qo^{-2}\big)\nonumber\\
&+2\frac{2\qo^2+4\qo^{4}}{\qo^2+\qo^{4}}\ln\big(1+\qo^2+\qo^{4}\big)
-6\ln\big(1+\sigma \qo^3\big)\nonumber\\
&+2\frac{-2\qo^{-2}-4\qo^{-4}}{\qo^{-2}+\qo^{-4}}\ln\big(1+\qo^{-2}+\qo^{-4}\big)
+6\ln\big(1+\sigma \qo^{-3}\big)\nonumber\\
&-\frac{2 \qo^2-2 \qo^{-2}}{\qo^2+2+\qo^{-2}}\ln\big|\qo^2+1+\qo^{-2}\big|
-\frac{2 \qo^2-2 \qo^{-2}}{\qo^2+1+\qo^{-2}}\ln\big(\qo^2+2+\qo^{-2}\big)\nonumber\\
&-\frac{2 \qo^2+4 \qo^{4}}{1+\qo^2+\qo^{4}}\ln\big|\qo+\qo^{-1}\big|
-\frac{\qo- \qo^{-1}}{\qo+\qo^{-1}}\ln\big(1+\qo^2+\qo^{4}\big)\nonumber\\
&-\frac{-2 \qo^{-2}-4 \qo^{-4}}{1+\qo^{-2}+\qo^{-4}}\ln\big|\qo+\qo^{-1}\big|
-\frac{\qo-\qo^{-1}}{\qo+\qo^{-1}}\ln\big(1+\qo^{-2}+\qo^{-4}\big).
\label{ddoK}\end{aligned}$$ Grouping terms and using $$\ln\big(\qo^2+2+\qo^{-2}\big)=\ln\big(\qo+\qo^{-1}\big)^2=2\ln\big|\qo+\qo^{-1}\big|,$$ we find $$\begin{aligned}
{2}
\qo\frac{\dd}{\dd \qo}\mathcal K_\sigma=&
\frac{7\qo+5\qo^{-1}}{\qo+\qo^{-1}}\ln\big(1+\qo^2+\qo^{4}\big)
-\frac{5\qo+7\qo^{-1}}{\qo+\qo^{-1}}\ln\big(1+\qo^{-2}+\qo^{-4}\big)\nonumber\\
&-2\frac{\qo-\qo^{-1}}{\qo+\qo^{-1}}\ln\big|\qo^2+1+\qo^{-2}\big|
+6\ln\big(1+\sigma \qo^{-3}\big)-6\ln\big(1+\sigma \qo^3\big).\label{ddoKv2}\end{aligned}$$ Next we perform simplifications where care is taken when using the functional relations for the logarithm:
\[logids\] $$\begin{aligned}
{2}
\ln\big(1+\qo^2+\qo^{4}\big)&=\ln\Big[\big(\qo^2+1+\qo^{-2}\big)\qo^2\Big]
=\ln\big|\qo^2+1+\qo^{-2}\big| + 2\ln \qo - n \pi\ir,\label{logid1}\\[0.1cm]
\ln\big(1+\qo^{-2}+\qo^{-4}\big)&=\ln\Big[\big(\qo^2+1+\qo^{-2}\big)\qo^{-2}\Big]
=\ln\big|\qo^2+1+\qo^{-2}\big| - 2\ln \qo + n \pi\ir,\label{logid2}\\[0.1cm]
\ln\big(1+\sigma \qo^{-3}\big)-\ln\big(1+\sigma \qo^3\big)&=-3\ln \qo+m \pi\ir,\label{logid3}\end{aligned}$$
where the integers $m$ and $n$ are given by n = {
[ll]{} 0 & (0,/3),\
1 & (/3,2/3),\
2 & (2/3,),
. m = {
[lll]{} 0 & (0,/3), &= +1,\
2 & (/3,), &= +1,\
1 & (0,2/3), &= -1,\
3 & (2/3,), &= -1.
.
These integers are determined in the following way. The logarithms on the left-sides of - are the same as in and have imaginary parts in $(-\pi,\pi]$ because of our choice of the branches. On the right-sides of -, a real logarithm of a positive real number and $\ln\qo=\ir\gamma$ appear. The argument of the logarithms on the left-side of and is zero for $\gamma = \frac \pi 3, \frac {2\pi} 3$. The range $(0, \pi)$ therefore splits into the three subintervals on which $n$ takes constant integer values: $(0,\frac \pi 3)$, $(\frac \pi 3,\frac {2\pi} 3)$ and $(\frac {2\pi} 3, \pi)$. It is easy to check that the integer $n$ has to be chosen in the above specified manner in order to have the imaginary part of the right-sides of and in $(-\pi,\pi]$.
Exponentiating both sides of , we find that the integer $m$ is even for $\sigma = 1$ and odd for $\sigma = -1$. The arguments of the logarithms on the left-side of assumes the value $0$ for $\gamma=\pi/3\ (2\pi/3)$ for $\sigma=+1\ (-1)$. We treat $\sigma=+1$ first. The integer valued $m$ is constant in $[0,\pi/3)$ and in $(\pi/3,\pi)$. Inserting $\gamma=0$ and $\gamma=2\pi/3$ into we find $m=0$ and $2$ respectively in these intervals. Next we treat $\sigma=-1$. The integer valued $m$ is constant in $(0,2\pi/3)$ and in $(2\pi/3,\pi]$. Inserting $\gamma=\pi/3$ and $\gamma=\pi$ into we find $m=1$ and $3$ in these intervals.
Simplifying by use of yields $$\qo\frac{\dd}{\dd \qo}\mathcal K_\sigma=6\ln \qo+(6m-12 n)\pi\ir.$$ For the two cases, we find explicitly
$$\begin{aligned}
{3}
\qo\frac{\dd}{\dd \qo}\mathcal K_{+} &=
\left\{\begin{array}{c}
6\ln \qo \\[0.2cm]
-12\pi\ir+6\ln \qo
\end{array}\right. \quad &&
\begin{array}{l}
0 {\leqslant}\gamma {\leqslant}\frac {2 \pi} 3, \\[0.2cm]
\frac {2 \pi} 3 {\leqslant}\gamma {\leqslant}\pi,
\end{array}
\\[0.3cm]
\qo\frac{\dd}{\dd \qo}\mathcal K_{-} &=
\left\{\begin{array}{c}
6\pi\ir+ 6\ln \qo\\[0.2cm]
-6\pi\ir+ 6\ln \qo
\end{array}\right.
\quad &&
\begin{array}{l}
0 {\leqslant}\gamma {\leqslant}\frac {\pi} 3, \\[0.2cm]
\frac {\pi} 3 {\leqslant}\gamma {\leqslant}\pi.
\end{array}\end{aligned}$$
Note that $\ln
\qo=\ir\gamma$ and $\frac{\dd}{\dd \gamma}=\ir \qo\frac{\dd}{\dd \qo}$. We next integrate with respect to $\gamma$:
$$\begin{aligned}
{3}
\mathcal K_{+} &=
\left\{\begin{array}{c}
C_+-3\gamma^2 \\[0.2cm]
C_+-8\pi^2+12\pi\gamma-3\gamma^2
\end{array}\right. \quad &&
\begin{array}{l}
0 {\leqslant}\gamma {\leqslant}\frac {2 \pi} 3, \\[0.2cm]
\frac {2 \pi} 3 {\leqslant}\gamma {\leqslant}\pi,
\end{array}
\\[0.3cm]
\mathcal K_{-} &=
\left\{\begin{array}{c}
C_--6\pi\gamma-3\gamma^2\\[0.2cm]
C_--4\pi^2+6\pi\gamma-3\gamma^2
\end{array}\right.
\quad &&
\begin{array}{l}
0 {\leqslant}\gamma {\leqslant}\frac {\pi} 3, \\[0.2cm]
\frac {\pi} 3 {\leqslant}\gamma {\leqslant}\pi,
\end{array}\end{aligned}$$
where $C_\pm$ are integration constants and the required continuity at $\gamma=\pi/3$ and $\gamma=2\pi/3$ has been imposed. The constants are determined quite easily by noting that $\mathcal K_{1}(\frac \pi 3) = \mathcal K_{-1}(\frac {2\pi} 3) = 0$, as for these values of $\gamma$ the upper and lower terminals of all integrals coincide. This yields $C_+=\pi^2/3$ and $C_-=4\pi^2/3$. We finally find
$$\begin{aligned}
{3}
\frac 1{8\pi^2}\mathcal K_+(\gamma) &=
\left\{\begin{array}{c}
\frac1{24} - \frac38 \left(\frac \gamma \pi\right)^2 \\[0.2cm]
-\frac{23}{24} + \frac32 \left(\frac \gamma \pi\right)- \frac38 \left(\frac \gamma \pi\right)^2
\end{array}\right. \quad &&
\begin{array}{l}
0 {\leqslant}\gamma {\leqslant}\frac {2 \pi} 3, \\[0.2cm]
\frac {2 \pi} 3 {\leqslant}\gamma {\leqslant}\pi,
\end{array}
\\[0.3cm]
\frac 1{8\pi^2}\mathcal K_-(\gamma) &=
\left\{\begin{array}{c}
\frac1{6} - \frac34 \left(\frac \gamma \pi\right) - \frac38 \left(\frac \gamma \pi\right)^2\\[0.2cm]
-\frac{1}{3} + \frac34 \left(\frac \gamma \pi\right)- \frac38 \left(\frac \gamma \pi\right)^2
\end{array}\right.
\quad &&
\begin{array}{l}
0 {\leqslant}\gamma {\leqslant}\frac {\pi} 3, \\[0.2cm]
\frac {\pi} 3 {\leqslant}\gamma {\leqslant}\pi.
\end{array}\end{aligned}$$
Spectrum generating functions and characters {#sec:qstuff}
============================================
Identities for $\boldsymbol{q}$-binomials {#sec:qids}
-----------------------------------------
The goal of this section is to derive the relation for the product of two Gaussian polynomials, which is useful in \[sec:charids\] and \[sec:Pcpf\]. The $q=1$ specialisations of the identities derived below are proven in [@R79 Section 1.4], and here we follow the same ideas. From the definition of the $q$-binomial, $\smallqbinom{n}{m}$ is non zero for integers $m,n$ with $0 {\leqslant}m {\leqslant}n$. The definition of the $q$-binomial is extended to $n<0$ by using the definition . This yields = (-1)\^b q\^[-12b(b-1) -ab]{},aZ\_[> 0]{}.\[eq:minusqbin\] It follows that $\smallqbinom{n}{m}$ is non zero for $n<0$, $m\in \mathbb Z_{{\geqslant}0}$. The $q$-binomials satisfy the relations
$$\begin{aligned}
{2}
\qbinom{a}{b} &= \qbinom{a}{a-b}\label{eq:id1}\\
\qbinom{a}{b}\qbinom{b}{c} &= \qbinom{a}{c}\qbinom{a-c}{b-c}\label{eq:id2}\\
\qbinom{a}{b} &= \qbinom{a-1}{b}+q^{a-b}\qbinom{a-1}{b-1}=q^b\qbinom{a-1}{b}+\qbinom{a-1}{b-1}
\label{eq:qbinnm}\end{aligned}$$
as well as the $q$-Vandermonde identity: = \_j q\^[j(a-c+j)]{}\[eq:qVan\]. Here and below, sums without bounds indicate that the variable is summed from $-\infty$ to $\infty$, with the summand being non-zero on a finite range only.
We first derive sum formulas for single $q$-binomials. A first relation is obtained as follows: \_k q\^[k(n-m+k)]{} \_[k]{}(-1)\^k q\^[k(n-m+k)-12k(k-1)-pk]{}. \[eq:id3\] A second relation is given by $$\begin{aligned}
{2}
\qbinom{n-p}{m} &\overset{\textrm{\tiny\eqref{eq:minusqbin}}}{=} (-1)^m q^{-\frac12m(m-1)-m(p-n)}\qbinom{p-n+m-1}{m} \nonumber\\
& \overset{\textrm{\tiny\eqref{eq:qVan}}}{=} (-1)^m q^{-\frac12m(m-1)-m(p-n)} \sum_k \qbinom{p}{m-k}\qbinom{-n+m-1}{k}q^{k(p-m+k)} \nonumber\\
&\overset{\textrm{\tiny\eqref{eq:minusqbin}}}{=}q^{-\frac12m(m-1)-m(p-n)} \sum_k (-1)^{m+k} \qbinom{p}{m-k}\qbinom{n-m+k}{k}q^{k(p-m+k)-\frac12k(k-1)-k(n-m+1)}\nonumber\\
& \ = \sum_{j}(-1)^j \qbinom{n-j}{m-j}\qbinom{p}{j} q^{\frac12j(j+1)+j(n-m-p)},\end{aligned}$$ where we substituted $k = m-j$ at the last step. By replacing $n-p$ with $n$, this identity can be rewritten as $$\begin{aligned}
{2}
\qbinom{n}{m}&=\sum_k(-1)^k \qbinom{n+p-k}{m-k}\qbinom{p}{k}q^{\frac12k(k+1)+k(n-m)} = \sum_k (-1)^k \qbinom{n+p-k}{n-m-k}\qbinom{p}{k}q^{\frac12k(k+1)+km} \nonumber\\
&\hspace{-0.2cm} \overset{\textrm{\tiny\eqref{eq:id1}}}{=} \sum_k (-1)^k \qbinom{n+p-k}{m+p}\qbinom{p}{k}q^{\frac12k(k+1)+km} \label{eq:id4}\end{aligned}$$ where we used $\smallqbinom{n}{m} = \smallqbinom{n}{n-m}$ at the second equality to replace $m$ by $n-m$.
Using these relations, we now derive an identity for the product of two Gaussian polynomials: $$\begin{aligned}
{2}
\qbinom{m}{p}\qbinom{n}{r} & = \qbinom{m-n+r+n-r}{p}\qbinom{n}{r} \overset{\textrm{\tiny\eqref{eq:qVan}}}{=} \sum_k \qbinom{m-n+r}{p-k}\qbinom{n-r}{k}\qbinom{n}{r} q^{k(m-n+r-p+k)}
\nonumber\\&
\hspace{-1.71cm}\overset{\textrm{\tiny\eqref{eq:id2}}}{=} \sum_k \qbinom{m-n+r}{p-k}\qbinom{r+k}{r}\qbinom{n}{r+k} q^{k(m-n+r-p+k)}
\nonumber\\&
\hspace{-1.65cm}\overset{\textrm{\tiny\eqref{eq:id4}}}{=} \sum_k \qbinom{m-n+r}{p-k}\qbinom{r+k}{r}q^{k(m-n+r-p+k)}\sum_j(-1)^{j}\qbinom{n+p-k-j}{p+r}\qbinom{p-k}{j}q^{\frac12j(j+1)+j(r+k)}
\nonumber\\&
\hspace{-1.5cm}= \sum_k \qbinom{m-n+r}{p-k}\qbinom{r+k}{r}q^{k(m-n+r-p+k)}\sum_i(-1)^{p-k-i}\qbinom{n+i}{p+r}\qbinom{p-k}{i}q^{\frac12(p-k-i)(p+k-i+1+2r)}
\nonumber\\&
\hspace{-1.71cm}\overset{\textrm{\tiny\eqref{eq:id1}}}{=} \sum_i \qbinom{n+i}{p+r}\sum_k(-1)^{p-k-i}\qbinom{r+k}{k}\qbinom{m-n+r}{p-k}\qbinom{p-k}{i}q^{k(m-n+r-p+k)+\frac12(p-k-i)(p+k-i+1+2r)}
\nonumber\\&
\hspace{-1.71cm}\overset{\textrm{\tiny\eqref{eq:id2}}}{=} \sum_i \qbinom{n+i}{p+r}\qbinom{m-n+r}{i}\sum_k(-1)^{p-k-i}\qbinom{r+k}{k}\qbinom{m-n+r-i}{p-i-k}q^{k(m-n+r-p+k)+\frac12(p-k-i)(p+k-i+1+2r)}
\nonumber\\&
\hspace{-1.65cm}\overset{\textrm{\tiny\eqref{eq:id3}}}{=} \sum_i \qbinom{n+i}{p+r}\qbinom{m-n+r}{i}(-1)^{p-i}\qbinom{m-n-i-1}{p-i}q^{\frac12i(i-1)+\frac12p(p+1)+pr-pi-ri}
\nonumber\\&
\hspace{-1.65cm}\overset{\textrm{\tiny\eqref{eq:minusqbin}}}{=}\sum_i \qbinom{n+i}{p+r}\qbinom{m-n+r}{i}\qbinom{n-m+p}{p-i}q^{\frac12i(i-1)+\frac12p(p+1)+pr-pi-ri-\frac12(p-i)(p-i-1)+(p-i)(m-n-i-1)}
\nonumber\\&
\hspace{-1.71cm}\overset{\textrm{\tiny\eqref{eq:id1}}}{=} \sum_i \qbinom{n+i}{p+r}\qbinom{m-n+r}{i}\qbinom{n-m+p}{n-m+i} q^{i^2+i(n-m-p-r)+mp-np+pr}.\label{eq:finaldoublebinomial}\end{aligned}$$ The last line is an identity that is used multiple times in \[sec:charids,sec:Pcpf\].
Character identities for the boundary case {#sec:charids}
------------------------------------------
In this section, we show that the finitized partition functions $\ZdN$ given in are equal to the finitized Kac characters $\chit_{1,d+1}^{\textrm{\tiny$(N)$}}(q)$. Let $t \in \mathbb Z$ with $t \equiv N \Mod 2$. We define
\[eq:X1X2X3\] $$\begin{aligned}
{2}
&X^1_{t} = \sum_{k{\geqslant}0}\sum_{i=0}^{k} q^{i^2 + 2k (k-i+\frac12)+t(-t+i-k-1)}\qbinom{\frac{N+t}2+i}{2k+1}\qbinom{k}{i}\qbinom{k+1}{i+t+1},\label{eq:X1}\\
&X^2_{t} = \sum_{k{\geqslant}1}\sum_{i=0}^{k-1} q^{i(i+1) + 2k (k-i-\frac12)+t(-t+i-k+1)}\qbinom{\frac{N+t}2+i}{2k}\qbinom{k-1}{i}\qbinom{k+1}{i+t+1},\\
&X^3_{t} = \sum_{k{\geqslant}0}\sum_{i=0}^{k} q^{-1+i(i+1) + 2k (k-i-\frac12)+t(-t+i-k-2)}\qbinom{\frac{N+t}2+i}{2k}\qbinom{k}{i}\qbinom{k}{i+t+1}.\end{aligned}$$
These quantities are useful because of the following identities:
\[eq:propsyo\] $$\begin{aligned}
{2}
&d=3t: && \ZdN = q^{d(d-1)/6}(X^2_{t} - q^{d+1} X^3_{t} + \delta_{d,0}),\label{eq:prop0}\\[0.05cm]
&d=3t+1: && \ZdN = q^{d(d-1)/6}(X^3_{t-1} - q^{d+1} X^2_{-t-1}),\label{eq:prop1}\\[0.05cm]
&d = 3t+2: \qquad && \ZdN = q^{d(d-1)/6}(X^1_{t} - q^{d+1} X^1_{-t-2}).\label{eq:prop2}\end{aligned}$$
In proving , we start with the case $d \equiv 2 \Mod 3$ which is easiest. By using and substituting $j$ by $k=i+j+t$ in , we find $$\begin{aligned}
{2}
\ZdN = q^{d(d-1)/6} &\sum_{i,k} q^{i^2 + 2k (k-i+\frac12)+t(-t+i-k-1)}\qbinom{\frac{N+t}2+i}{2k+1}\bigg(\qbinom{k}{i}\qbinom{k+1}{i+t+1} - \qbinom{k+1}{i}\qbinom{k}{i+t+1} \bigg).\end{aligned}$$ The positive part is readily identified as $X^1_{t}$. After substituting $i$ for $j=i+t+1$ and setting $t =-s-2$ in the negative part, it is found to equal $-q^{-3(1+s)}X^1_{s} = -q^{d+1}X^1_{-t-2}$, ending the proof of .
For $d = 3t$ and $t>0$, applying to yields four double-sums: $$\begin{aligned}
{2}
\ZdN &= q^{d(d-1)/6}\bigg( \sum_{i,k} q^{i(i+1)+2k(k-i-\frac12)+t(-t+i-k+1)}\qbinom{\frac{N+t}2}{2k}\Big(\qbinom{k-1}{i}\qbinom{k}{i+t}-\qbinom{k}{i}\qbinom{k-1}{i+t}\Big)\nonumber\\
&\hspace{-0.5cm} + \sum_{i,\ell} q^{1+i(i+2)+2\ell(\ell-i-\frac12)+t(-t+i-\ell+2)}\qbinom{\frac{N+t}2}{2\ell}\Big(\qbinom{\ell-1}{i}\qbinom{\ell}{i+t+1}-\qbinom{\ell}{i}\qbinom{\ell-1}{i+t+1}\Big)\bigg)\label{eq:noname},\end{aligned}$$ where the sum on the second line was obtained by substituting $\ell = k+1$. Using the identity , the first terms and second terms of each line of respectively combine, resulting in $$\begin{aligned}
{2}
\ZdN &=q^{d(d-1)/6} \sum_{i,k} q^{i(i+1)+2k(k-i-\frac12)+t(-t+i-k+1)}\qbinom{\frac{N+t}2}{2k}\bigg(\qbinom{k-1}{i}\qbinom{k+1}{i+t+1}-\qbinom{k}{i}\qbinom{k}{i+t+1}\bigg) \nonumber\\
& = q^{d(d-1)/6}(X^2_{t}-q^{d+1} X^3_{t}),\end{aligned}$$ ending the proof of . The case $d=0$ is computed separately and the proof of in this case is a simple exercise in $q$-binomials, with the Razumov-Stroganov eigenvalue responsible for the extra factor $\delta_{d,0}$. The proof of uses the ideas of the previous two cases and is straightforward.
The identity derived in \[sec:qids\] allows us to simplify $X^1_t$, $X^2_t$ and $X^3_t$. The parameters $m$, $n$, $p$ and $q$ are specialised as follows:
$$\begin{aligned}
{5}
&\textrm{for } X^1_{t}:\qquad&&m=\tfrac{N-t-2}2,\quad &&n=\tfrac{N+t}2,\quad &&p=k-t,\quad&&r=k+t+1,\\
&\textrm{for } X^2_{t}:\qquad&&m=\tfrac{N-t-2}2,\quad &&n=\tfrac{N+t}2,\quad &&p=k-t,\quad&&r=k+t,\\
&\textrm{for } X^3_{t}:\qquad&&m=\tfrac{N-t-2}2,\quad&&n=\tfrac{N+t}2,\quad&&p=k-t-1,\quad&&r=k+t+1.\end{aligned}$$
Starting from , for $X^1_t$ we find $$\begin{aligned}
{2}
X^1_t &\overset{\textrm{\tiny\eqref{eq:finaldoublebinomial}}}{=} \sum_{k{\geqslant}0} q^{k(k+1)-t(t+1)} \qbinom{\frac{N+t}2}{k+t+1}\qbinom{\frac{N-t-2}2}{k-t} = \sum_j q^{j(2t+1+j)}\qbinom{\frac{N+t}2}{j+2t+1}\qbinom{\frac{N-t-2}2}{j} \nonumber\\
& \hspace{-0.068cm}
\overset{\textrm{\tiny\eqref{eq:id1}}}{=} \sum_j q^{j(2t+1+j)}\qbinom{\frac{N+t}2}{\frac{N-3t-2}2-j}\qbinom{\frac{N-t-2}2}{j} \overset{\textrm{\tiny\eqref{eq:qVan}}}{=} \qbinom{N-1}{\frac{N-3t-2}2}.
\label{eq:X1result}\end{aligned}$$ Likewise for $X^2_t$: $$\begin{aligned}
{2}
X^2_t &\overset{\textrm{\tiny\eqref{eq:finaldoublebinomial}}}{=} \sum_{k{\geqslant}1} q^{k^2-t^2} \qbinom{\frac{N+t}2}{k+t}\qbinom{\frac{N-t-2}2}{k-t} = \bigg(\sum_{k{\geqslant}0} q^{k^2-t^2} \qbinom{\frac{N+t}2}{k+t}\qbinom{\frac{N-t-2}2}{k-t}\bigg) - \delta_{t,0} \nonumber\\
& \overset{\phantom{\textrm{\tiny\eqref{eq:finaldoublebinomial}}}}{=} \bigg(\sum_j q^{j(2t+j)}\qbinom{\frac{N+t}2}{j+2t}\qbinom{\frac{N-t-2}2}{j}\bigg) - \delta_{t,0}
\nonumber\\
&\hspace{-0.068cm} \overset{\textrm{\tiny\eqref{eq:id1}}}{=} \bigg( \sum_j q^{j(2t+1+j)}\qbinom{\frac{N+t}2}{\frac{N-3t-2}2-j}\qbinom{\frac{N-t-2}2}{j}\bigg) - \delta_{t,0} \overset{\textrm{\tiny\eqref{eq:qVan}}}{=} \qbinom{N-1}{\frac{N-3t-2}2}- \delta_{t,0}
\label{eq:X2result}\end{aligned}$$ where we used $q^{-t^2}\smallqbinom{\frac {N+t}2}{t}\smallqbinom{\frac {N+t}2}{-t}= \delta_{t,0}$ at the second equality. The steps are identical for $X^3_t$: X\^3\_t \_[k]{} q\^[k\^2-(t+1)\^2]{} = \_j q\^[j(j+2t+2)]{} . \[eq:X3result\] Combining , and with , we find = q\^[d(d-1)/6]{}(-q\^[d+1]{})= q\^[d(d-1)/6]{}(-q\^[d+1]{}) = \_[1,d+1]{}(q), which holds for $d \equiv 0,1,2 \Mod 3$. The second equality is obtained (from right to left) using .
Partition functions for the periodic case {#sec:Pcpf}
-----------------------------------------
#### Explicit expressions.
The finitized partition functions $\ZdNP$ defined in are obtained by writing down the generating functions for each set in and summing over the corresponding values of $i$ and $j$. We note that the indices $i$ and $j$ in the selection rules always run over all possible values for which the corresponding sets $\BB{M}{L}{m}{n}{\ell}$ are well defined. This is also true for $k_1, k_2$ and $k_3$ in . For ease of notation, we omit to write the bounds of the sums over $i,j,k_2$ and $k_3$ and interpret the corresponding indices as running over $\mathbb Z$, understanding that only finitely many terms are non-zero. The same applies to $k_1$, except that the sum is split between the odd and even values, as they correspond to different values of $\sigma$, see .
For $d \equiv 0 \Mod 3$, we use the prescription for the separation between upper and lower halves and write down the generating function for each set in . Summing over these sets, we find $$\begin{aligned}
{2}
Z_{d=3t}^{\textrm{\tiny$(N)$}}(q,\bar q) &= \sum_{i,j,k_2,k_3} \!\Big(\sum_{k_1\,\textrm{even}} (q \bar q)^{-\frac1{24}}+ \sum_{k_1\,\textrm{odd}}(-1)^M (q \bar q)^{-\frac16}\Big) (q\bar q)^{E} q^{E_0}
\qqbinom{\lfloor \frac12(\frac{N+t}{2}+i)\rfloor - \epsilon}{k_1} \qqbinom{\lfloor{\frac {k_1}2}\rfloor}{k_2} \qqbinom{\lfloor{\frac {k_1+1}2}\rfloor}{k_3}
\nonumber\\& \hspace{2cm}\times \qbinom{\lfloor \frac12(\frac{N+t}{2}+i+1)\rfloor + \epsilon}{2(i+j+t)-k_1} \qbinom{i+j+t-\lfloor{\frac {k_1}2}\rfloor}{i-k_2}\qbinom{i+j+t-\lfloor{\frac {k_1+1}2}\rfloor}{i+t-k_3}
\label{eq:Z0}\end{aligned}$$ where $E$ and $E_0$ are obtained from the proper specialisations of :
$$\begin{aligned}
{2}
E &= \tfrac12(k_1^2+k_2^2+k_3^2-k_1k_2 - k_1 k_3),\label{eq:EE}\\[0.15cm]
E_0 &= i^2 - i k_1+ 2j(i+j-k_1+\tfrac12 k_2+\tfrac12k_3) + t(2i+3j - \tfrac32 k_1 + k_2 + \tfrac32 t).\label{E0}\end{aligned}$$
For $d \equiv 1,2 \Mod 3$, from , and , $\ZdNP$ is split between contributions coming from the different subcases:
\[eq:Z12separation\] $$\begin{aligned}
{2}
Z_{d=3t+1}^{\textrm{\tiny$(N)$}}(q,\bar q) &= Z_1^{\bar \tB, \tB} + Z_1^{\bar \tA, \tB} + Z_1^{\bar \tB, \tA} + Z_1^{\bar \tA, \tA} + (-1)^M \delta_{d,1},\label{eq:Z1separation} \\[0.15cm]
Z_{d=3t+2}^{\textrm{\tiny$(N)$}}(q,\bar q) &= Z_2^{\bar\tA, \tA} + Z_2^{\bar\tA, \tB} + Z_2^{\bar\tB, \tA} + Z_2^{\bar\tB, \tB}.
\label{eq:Z2separation}\end{aligned}$$
Each $Z_k^{\tX, \tY}$, with $\tX \in \{\bar \tA, \bar \tB\}$, $\tY \in \{\tA, \tB\}$ and $k = 1,2$, is obtained by using the prescription for the separation between upper and lower halves, writing down the corresponding generating function , and summing over the corresponding sets given by the selection rules: $$\begin{aligned}
{2}
Z_k^{\tX, \tY} &= \sum_{i,j,k_2,k_3} \!\Big(\sum_{k_1\,\textrm{even}} (-1)^M+ \sum_{k_1\,\textrm{odd}} (q \bar q)^{\frac18}\Big) (q\bar q)^{E(\tX)} q^{E_k(\tX,\tY)} F_k(\tX, \tY)\end{aligned}$$ with
$$\begin{aligned}
{2}
F_1(\bar \tB, \tB) &=
\qqbinom{\lfloor \frac12(\frac{N+t-1}{2}+i)\rfloor - \epsilon}{k_1} \qqbinom{\lfloor{\frac {k_1-1}2}\rfloor}{k_2} \qqbinom{\lfloor{\frac {k_1}2}\rfloor}{k_3} \nonumber\\& \hspace{0.8cm} \times
\qbinom{\lfloor \frac12(\frac{N+t-1}{2}+i+1)\rfloor + \epsilon}{2(i+j+t)-k_1} \qbinom{i+j+t-1-\lfloor{\frac {k_1-1}2}\rfloor}{i-k_2}\qbinom{i+j+t-1-\lfloor{\frac {k_1}2}\rfloor}{i+t-1-k_3},\\[0.2cm]
F_1(\bar \tA, \tB) &= F_2(\bar \tB, \tA) =
\qqbinom{\lfloor \frac12(\frac{N+t+1}{2}+i)\rfloor - \epsilon}{k_1} \qqbinom{\lfloor{\frac {k_1-1}2}\rfloor}{k_2} \qqbinom{\lfloor{\frac {k_1}2}\rfloor}{k_3} \nonumber\\& \hspace{2.25cm} \times
\qbinom{\lfloor \frac12(\frac{N+t+1}{2}+i+1)\rfloor + \epsilon}{2(i+j+t+1)-k_1} \qbinom{i+j+t-\lfloor{\frac {k_1-1}2}\rfloor}{i-k_2}\qbinom{i+j+t-\lfloor{\frac {k_1}2}\rfloor}{i+t-k_3},\\[0.2cm]
F_1(\bar \tA, \tA) &=
\qqbinom{\lfloor \frac12(\frac{N+t+3}{2}+i)\rfloor - \epsilon}{k_1} \qqbinom{\lfloor{\frac {k_1-1}2}\rfloor}{k_2} \qqbinom{\lfloor{\frac {k_1}2}\rfloor}{k_3} \nonumber\\& \hspace{0.8cm} \times
\qbinom{\lfloor \frac12(\frac{N+t+3}{2}+i+1)\rfloor + \epsilon}{2(i+j+t+2)-k_1} \qbinom{i+j+t+1-\lfloor{\frac {k_1-1}2}\rfloor}{i-k_2}\qbinom{i+j+t+1-\lfloor{\frac {k_1}2}\rfloor}{i+t+1-k_3},\end{aligned}$$
and
$$\begin{aligned}
{2}
F_2(\bar \tA, \tA) &=
\qqbinom{\lfloor \frac12(\frac{N+t+2}{2}+i)\rfloor - \epsilon}{k_1} \qqbinom{\lfloor{\frac {k_1-1}2}\rfloor}{k_2} \qqbinom{\lfloor{\frac {k_1}2}\rfloor}{k_3} \nonumber\\& \hspace{0.8cm} \times
\qbinom{\lfloor \frac12(\frac{N+t+2}{2}+i+1)\rfloor + \epsilon}{2(i+j+t+1)-k_1} \qbinom{i+j+t-\lfloor{\frac {k_1-1}2}\rfloor}{i-k_2}\qbinom{i+j+t-\lfloor{\frac {k_1}2}\rfloor}{i+t-k_3},\\[0.2cm]
F_2(\bar \tA, \tB) &= F_2(\bar \tB, \tA) =
\qqbinom{\lfloor \frac12(\frac{N+t+2}{2}+i)\rfloor - \epsilon}{k_1} \qqbinom{\lfloor{\frac {k_1-1}2}\rfloor}{k_2} \qqbinom{\lfloor{\frac {k_1}2}\rfloor}{k_3} \nonumber\\& \hspace{0.8cm} \times
\qbinom{\lfloor \frac12(\frac{N+t+2}{2}+i+1)\rfloor + \epsilon}{2(i+j+t+2)-k_1} \qbinom{i+j+t+1-\lfloor{\frac {k_1-1}2}\rfloor}{i-k_2}\qbinom{i+j+t+1-\lfloor{\frac {k_1}2}\rfloor}{i+t+1-k_3},
\\[0.2cm]
F_2(\bar \tB, \tB) &=
\qqbinom{\lfloor \frac12(\frac{N+t+2}{2}+i)\rfloor - \epsilon}{k_1} \qqbinom{\lfloor{\frac {k_1-1}2}\rfloor}{k_2} \qqbinom{\lfloor{\frac {k_1}2}\rfloor}{k_3} \nonumber\\& \hspace{0.8cm} \times
\qbinom{\lfloor \frac12(\frac{N+t+2}{2}+i+1)\rfloor + \epsilon}{2(i+j+t+3)-k_1} \qbinom{i+j+t+2-\lfloor{\frac {k_1-1}2}\rfloor}{i-k_2}\qbinom{i+j+t+2-\lfloor{\frac {k_1}2}\rfloor}{i+t+2-k_3}.\end{aligned}$$
The energies for the $\tA$ and $\bar \tA$ subcases are obtained from under the proper specialisations of $m$, $n$ and $\ell$. The minimal energies for the $\tB$ and $\bar \tB$ subcases are likewise obtained from . We find
$$\begin{aligned}
{2}
E(\bar \tA) &= \tfrac12 (k_1^2+k_2^2+k_3^2-k_1+k_2+k_3-k_1 k_2-k_1 k_3),\\[0.15cm]
E(\bar \tB) &= \tfrac12 (k_1^2+k_2^2+k_3^2+k_2+k_3-k_1 k_2-k_1 k_3),\end{aligned}$$
and
$$\begin{aligned}
{2}
q^{E_1(\bar \tB, \tB)} & =q^{1 + i(i+1) - i k_1-\frac12k_1 - k_2 + 2j(i+j-k_1+\frac12 k_2+\frac12k_3+\frac 12) + t(2i+3j - \frac32 k_1 + k_2 + \frac32 t+\frac12)}, \\
q^{E_1(\bar \tA, \tB)} & =q^{2 + i(i+3) - i k_1- \frac32k_1 + 2j(i+j-k_1+\frac12 k_2+\frac12k_3+2) + t(2i+3j - \frac32 k_1 + k_2 + \frac32 t+\frac{7}2)}, \\
q^{E_1(\bar \tB, \tA)} & =q^{1 + i(i+2) - i k_1- \frac32k_1 + 2j(i+j-k_1+\frac12 k_2+\frac12k_3+\frac32) + t(2i+3j - \frac32 k_1 + k_2 + \frac32 t+\frac{5}2)}, \\
q^{E_1(\bar \tA, \tA)} & =q^{5 + i(i+4) - i k_1-\frac52k_1+k_2 + 2j(i+j-k_1+\frac12 k_2+\frac12k_3+3) + t(2i+3j - \frac32 k_1 + k_2 + \frac32 t+\frac{11}2)}, \\
q^{E_2(\bar \tA, \tA)} & =q^{1 + i(i+2) - i k_1-k_1 + 2j(i+j-k_1+\frac12 k_2+\frac12k_3+\frac 32) + t(2i+3j - \frac32 k_1 + k_2 + \frac32 t+\frac52)}, \\
q^{E_2(\bar \tA, \tB)} & =q^{7 + i(i+5) - i k_1- 3k_1 +k_2+ 2j(i+j-k_1+\frac12 k_2+\frac12k_3+\frac72) + t(2i+3j - \frac32 k_1 + k_2 + \frac32 t+\frac{13}2)}, \\
q^{E_2(\bar \tB, \tA)} & =q^{5 + i(i+4) - i k_1-3k_1 +k_2+ 2j(i+j-k_1+\frac12 k_2+\frac12k_3+3) + t(2i+3j - \frac32 k_1 + k_2 + \frac32 t+\frac{11}2)}, \\
q^{E_2(\bar \tB, \tB)} & =q^{15 + i(i+7) - i k_1-5k_1+2k_2 + 2j(i+j-k_1+\frac12 k_2+\frac12k_3+5) + t(2i+3j - \frac32 k_1 + k_2 + \frac32 t+\frac{19}2)}.\end{aligned}$$
#### Scaling behavior for $\boldsymbol{d \equiv 0} \Mod \boldsymbol{3}$.
We explicitly derive the formula for the scaling limit of $Z_{d=3t}^{\textrm{\tiny$(N)$}}(q,\bar q)$. We perform the calculation separately for the odd and even $k_1$ contributions in , namely we write Z\_[d=3t]{}\^(q,|q) = Z\_[0,]{} + (-1)\^M Z\_[0,]{}.
We start with the even case. As discussed in \[sec:fsgf\], for $\epsilon \ll N$, the scaling behavior of is independent of $\epsilon$. In fact, $\epsilon$ can be chosen to depend on $i$, whose values are indeed much smaller than $N$ for the leading eigenvalues. For the same reason, one could also choose $\epsilon$ to depend on $j$, $k_1$, $k_2$ and $k_3$. Here we make a special choice of $\epsilon$ that allows the computation to go forward, namely we choose $\epsilon$ such that the first $\bar q$-binomial in does not depend on $i$. This allows us to perform the sums over $i$ and $j$ first, noting that their summands involve only powers of $q$ and not of $\bar q$. Defining $N_t = N+t$, we get Z\_[0,]{}=(q |q)\^[-1[24]{}]{} \_[k\_2,k\_3]{} \_[k\_1]{} (q |q)\^[E]{} S\_[0,]{} where $$\begin{aligned}
{2}
S_{0,\textrm{even}} &= \sum_{i,j} q^{E_0} \qbinom{\lfloor \frac{N_t}4+\frac12\rfloor+i}{2(i+j+t)-k_1}\qbinom{i+j+t-\frac{k_1}2}{i-k_2}\qbinom{i+j+t-\frac{k_1}2}{i+t-k_3} \nonumber\\
& = \sum_{k,\ell} q^{E_0} \qbinom{\lfloor \frac{N_t}4+\frac12\rfloor+k_2+k}{2\ell-k_1}\qbinom{\ell-\frac{k_1}2}{k}\qbinom{\ell-\frac{k_1}2}{k+t+k_2-k_3}.\end{aligned}$$ At the last equality, we substituted first $\ell = i+j+t$ and then $k = i-k_2$. In the last expression, the sums over $k$ and $\ell$ still run over $\mathbb Z$. Although the notation does not make it explicit, the expression for $E_0$ is understood as changing with every change of summation variables. It remains quadratic in the various parameters. The next step consists in using with m = 4 + 12+ k\_3 - t, n = 4 + 12+ k\_2, p = - 2 - k\_2 + k\_3 - t, r = - 2 + k\_2 - k\_3 + t. This yields S\_[0,]{} = \_q\^[E\_0’]{} where E\_0’ = -14 k\_1\^2 + 12 k\_1 k\_2 + 12 k\_1 k\_3 - k\_2 k\_3 + \^2 - k\_1 + 12 t\^2 + t k\_2 - t k\_3. This remaining sum can be evaluated using : S\_[0,]{} = \_j q\^[E\_0’]{} = q\^[E\_0”]{} . Remarkably, $E_0''$ satisfies E\_0” + E = 32(k\_3 -k\_2-t)\^2 = \_[0,3(k\_3-k\_2-t)]{}+ 1[24]{} with $E$ given in . The right-side depends only on the difference between $k_3$ and $k_2$. Changing the summation variables to $k_3 = \ell + k_2$ and $k_1 = 2i$, we obtain Z\_[0,]{}=|q\^[-1[24]{}]{} \_q\^[\_[0,3(-t)]{}]{} \_[i,k\_2]{} |q\^[E]{} . The expression is thus reduced to sums of products of four binomials, with only one depending on $q$. While one may wish to reduce this expression further to a single sum with one binomial of each kind, this appears not to be feasible because the arguments of the remaining $q$-binomial involve both $\ell$ and $k_2$. However, both entries of this $q$-binomial scale linearly with $N$. We consider standard modules where the number $d$ of defects remains small as $N \rightarrow \infty$, namely values of $d$ such that $t \ll N$. Recalling that $k_2, \ell \ll N_t$ for large $N$, in the scaling limit we have and therefore Z\_[0,]{} \_q\^[\_[0,3(-t)]{}]{} \_[i,k\_2]{} |q\^[E]{} . Here, $X \simeq Y$ means that $X$ and $Y$ are equal up to terms which go to zero in the scaling limit. The sums over $i$ and $k_2$ involve only powers of $\bar q$. The expression is not suitable for us to use to remove the sum over $i$, nor for the sum over $k_2$ because $E$ is not of the correct form. We however note that and \_[i,k\_2]{} |q\^[E]{} = \_i [|q]{}\^[E\_0”’]{} = |q\^[32 \^2]{} where we use and at the first and second equalities. Using $\Delta_{0,3 \ell} = \frac32 \ell^2 - \frac1{24}$, we find Z\_[0,]{} \_q\^[\_[0,3 (-t)]{}]{}|q\^[\_[0,3 ]{}]{}. This is the first term in .
The derivation of the scaling behavior of $Z_{0,\textrm{odd}}$ follows the same steps with only few modifications: Z\_[0,]{}=(q |q)\^[-1[6]{}]{} \_[k\_2,k\_3]{} \_[k\_1]{} (q |q)\^[E]{} S\_[0,]{} with $$\begin{aligned}
{2}
S_{0,\textrm{odd}} &= \sum_{i,j} q^{E_0} \qbinom{\lfloor \frac{N_t}4+\frac12\rfloor+i}{2(i+j+t)-k_1}\qbinom{i+j+t-\frac{k_1-1}2}{i-k_2}\qbinom{i+j+t-\frac{k_1+1}2}{i+t-k_3} \nonumber\\
& = \sum_{k,\ell} q^{E_0} \qbinom{\lfloor \frac{N_t}4+\frac12\rfloor+k_2+k}{2\ell-k_1}\qbinom{\ell-\frac{k_1-1}2}{k}\qbinom{\ell-\frac{k_1+1}2}{k+t+k_2-k_3}\nonumber\\
& = \sum_\ell q^{E_0'} \specialqbinom{\lfloor\tfrac{N_t}4 + \tfrac12\rfloor + k_3 - t}{\ell - \frac{k_1+1}2 - k_2 + k_3 - t}\specialqbinom{ \lfloor\tfrac{N_t}4 + \tfrac12\rfloor + k_2}{\ell - \frac{k_1-1}2 + k_2 - k_3 + t}\nonumber\\
& = q^{E_0''} \specialqbinom{2\lfloor\tfrac{N_t}4 + \tfrac12\rfloor + k_2+k_3 - t}{\lfloor\tfrac{N_t}4 + \tfrac12\rfloor + 2k_2 - k_3 + t + 1} \simeq \frac{q^{E_0''}}{(q)_\infty}\end{aligned}$$ with $E_0'$ and $E_0''$ adapted accordingly. In particular, $E_0''$ satisfies $E_0'' + E = \Delta_{1,3(k_3-k_2-t)}+ \tfrac1{6}$. Changing the summation variables to $k = 2i+1$ and $k_3 = k_2 +\ell$, we find $$\begin{aligned}
{2}
Z_{0,\textrm{odd}} &\simeq \frac{\bar q^{\,-\frac1{6}}}{(q)_\infty} \sum_\ell q^{\Delta_{1,3(\ell-t)}} \sum_{i,k_2} \bar q^{E} \qqbinom{\lfloor \frac{N_t}4\rfloor}{2i+1}\qqbinom{i}{k_2}\qqbinom{i+1}{k_2 + \ell}
\nonumber\\&
\simeq \frac{\bar q^{\,-\frac1{6}}}{(q)_\infty} \sum_\ell q^{\Delta_{1,3(\ell-t)}} \sum_{i,k_2} \bar q^{E} \qqbinom{\lfloor \frac{N_t}4\rfloor+k}{2i+1}\qqbinom{i}{k_2}\qqbinom{i+1}{k_2 + \ell}\nonumber\\
& = \frac{\bar q^{\,-\frac1{6}}}{(q)_\infty} \sum_\ell q^{\Delta_{1,3(\ell-t)}} \sum_{i} \bar q^{E_0'''} \qqbinom{\lfloor \frac{N_t}4\rfloor-\ell}{i+1-\ell} \qqbinom{\lfloor \frac{N_t}4\rfloor}{i+\ell}\nonumber \\
& = \frac1{(q)_\infty} \sum_\ell q^{\Delta_{1,3(\ell-t)}} \bar q^{\Delta_{1,3\ell}} \specialqqbinom{2\lfloor \frac{N_t}4\rfloor - \ell}{\lfloor \frac{N_t}4\rfloor+\ell-1} \simeq \frac{1}{(q)_\infty (\bar q)_\infty} \sum_\ell q^{\Delta_{1,3 (\ell-t)}}\bar q^{\Delta_{1,3 \ell}}.\label{eq:forlaststeps}\end{aligned}$$ This ends the proof of .
#### Scaling behavior for $\boldsymbol{d \equiv 1,2} \Mod \boldsymbol{3}$.
These cases are more complicated because of the various contributions to $\ZdNP$ in . For $d \equiv 1 \Mod 3$, we do the calculations only for $d>1$ for simplicity. The case $d=1$ uses the same arguments, with extra care given to the bounds of the sums and the contribution from the Razumov-Stroganov eigenvalue. Writing $d = 3t+1$ with $t>0$, each contribution to splits into an even and an odd part as \[eq:noinspiration\] Z\_[1]{}\^[,]{} = (-1)\^M Z\_[1,]{}\^[,]{} + Z\_[1,]{}\^[,]{} , {|, |}, {, }. Starting with the even contributions, as before, we choose $\epsilon$ such that the $\bar q$-binomials do not depend on $i$. Defining $N_t = N+t+1$, we obtain Z\^[,]{}\_[1,]{}= \_[k\_2,k\_3]{} \_[k\_1]{} (q |q)\^[E()]{} S\^[,]{}\_[1,]{} with
$$\begin{aligned}
{2}
S^{\bar \tB, \tB}_{1,\textrm{even}} &= \sum_{i,j} q^{E_1(\bar \tB, \tB)} \qbinom{\lfloor \frac{N_t}4+\frac12\rfloor+i-1}{2(i+j+t)-k_1}\qbinom{i+j+t-\frac{k_1}2}{i-k_2}\qbinom{i+j+t-\frac{k_1}2-1}{i+t-k_3-1} \nonumber\\
& = \sum_{k,\ell} q^{E_1(\bar \tB, \tB)} \qbinom{\lfloor \frac{N_t}4+\frac12\rfloor+k_2+k-1}{2\ell-k_1}\qbinom{\ell-\frac{k_1}2}{k}\qbinom{\ell-\frac{k_1}2-1}{k+t+k_2-k_3-1},
\\[0.2cm]
S^{\bar \tB, \tA}_{1,\textrm{even}} &= \sum_{i,j} q^{E_1(\bar \tB, \tA)} \qbinom{\lfloor \frac{N_t}4+\frac12\rfloor+i}{2(i+j+t+1)-k_1}\qbinom{i+j+t+1-\frac{k_1}2}{i-k_2}\qbinom{i+j+t-\frac{k_1}2}{i+t-k_3} \nonumber\\
& = \sum_{k,\ell} q^{E_1(\bar \tB, \tA)} \qbinom{\lfloor \frac{N_t}4+\frac12\rfloor+k_2+k-1}{2\ell-k_1}\qbinom{\ell-\frac{k_1}2}{k-1}\qbinom{\ell-\frac{k_1}2-1}{k+t+k_2-k_3-1},
\\[0.2cm]
S^{\bar \tA, \tB}_{1,\textrm{even}} &= \sum_{i,j} q^{E_1(\bar \tA, \tB)} \qbinom{\lfloor \frac{N_t}4+\frac12\rfloor+i}{2(i+j+t+1)-k_1}\qbinom{i+j+t+1-\frac{k_1}2}{i-k_2}\qbinom{i+j+t-\frac{k_1}2}{i+t-k_3} \nonumber\\
& = \sum_{k,\ell} q^{E_1(\bar \tA, \tB)} \qbinom{\lfloor \frac{N_t}4+\frac12\rfloor+k_2+k}{2\ell-k_1}\qbinom{\ell-\frac{k_1}2}{k}\qbinom{\ell-\frac{k_1}2-1}{k+t+k_2-k_3},
\\[0.2cm]
S^{\bar \tA, \tA}_{1,\textrm{even}} &= \sum_{i,j} q^{E_1(\bar \tA, \tA)} \qbinom{\lfloor \frac{N_t}4+\frac12\rfloor+i+1}{2(i+j+t+2)-k_1}\qbinom{i+j+t+2-\frac{k_1}2}{i-k_2}\qbinom{i+j+t-\frac{k_1}2+1}{i+t-k_3+1} \nonumber\\
& = \sum_{k,\ell} q^{E_1(\bar \tA, \tA)} \qbinom{\lfloor \frac{N_t}4+\frac12\rfloor+k_2+k}{2\ell-k_1}\qbinom{\ell-\frac{k_1}2}{k-1}\qbinom{\ell-\frac{k_1}2-1}{k+t+k_2-k_3}.\end{aligned}$$
It is not possible to simplify each $S^{\tX, \tY}_{1,\textrm{even}}$ individually using . One instead combines $S^{\bar \tB, \tB}$ and $S^{\bar \tB, \tA}$ using $E_1(\bar \tB, \tB) - E_1(\bar \tB, \tA) = j$ and : $$\begin{aligned}
{2}
S^{\bar \tB, \tB}_{1,\textrm{even}} + S^{\bar \tB, \tA}_{1,\textrm{even}} &= \sum_{k,\ell} q^{E_1(\bar \tB, \tA)} \qbinom{\lfloor \frac{N_t}4+\frac12\rfloor+k_2+k-1}{2\ell-k_1}\qbinom{\ell-\frac{k_1}2+1}{k}\qbinom{\ell-\frac{k_1}2-1}{k+t+k_2-k_3-1},\nonumber\\
& = \sum_\ell q^{E_1'(\bar \tB, \tA)} \qbinom{\lfloor\tfrac{N_t}4 + \tfrac12\rfloor + k_3 - t}{\ell - \frac{k_1}2 - k_2 + k_3 - t}\qbinom{ \lfloor\tfrac{N_t}4 + \tfrac12\rfloor + k_2-1}{\ell - \frac{k_1}2 + k_2 - k_3 + t}\nonumber\\
& = q^{E_1''(\bar \tB, \tA)}\specialqbinom{2\lfloor\tfrac{N_t}4 + \tfrac12\rfloor + k_2+k_3 - t-1}{\lfloor\tfrac{N_t}4 + \tfrac12\rfloor + 2k_2 - k_3 + t} \simeq \frac{q^{E_1''(\bar \tB, \tA)}}{(q)_\infty}\end{aligned}$$ with $E_1''(\bar \tB, \tA)$ satisfying $E_1''(\bar \tB, \tA)+E(\tB) = \Delta_{1,3(k_3-k_2-t)+1}$. The relations and were used for the second and third equality. Changing the summation variables in to $k_1 = 2i$ and $k_3 = k_2 + \ell$, we obtain Z\_[1, ]{}\^[|,]{} + Z\_[1, ]{}\^[|,]{} 1[(q)\_]{} \_q\^[\_[1,3(-t)+1]{}]{} \_[i,k\_2]{} |q\^[E()]{} . Likewise for $S^{\bar \tA, \tB}$ and $S^{\bar \tA, \tA}$, we use $E_1(\bar \tA, \tB) - E_1(\bar \tA, \tA) =j$ and find after simplification: S\^[|, ]{}\_[1,]{} + S\^[|, ]{}\_[1,]{} = q\^[E\_1”(|, )]{} with $E_1''(\bar \tA, \tA)+E(\tB) = \Delta_{1,3(k_3-k_2-t-1)+1}$. With $k = 2i$, $k_2 = k_2'-1$ and $k_3 = k_2' + \ell$, we get Z\_[1, ]{}\^[|,]{} + Z\_[1, ]{}\^[|,]{} 1[(q)\_]{} \_q\^[\_[1,3(-t)+1]{}]{} \_[i,k’\_2]{} |q\^[E()]{} . Putting these results together, we find $$\begin{aligned}
{2}
Z_{1, \textrm{even}}^{\bar\tB,\tB}& + Z_{1, \textrm{even}}^{\bar\tB,\tA} + Z_{1, \textrm{even}}^{\bar\tA,\tB} + Z_{1, \textrm{even}}^{\bar\tA,\tA}
\simeq \frac1{(q)_\infty} \sum_\ell q^{\Delta_{1,3(\ell-t)+1}} \sum_{i,k_2} \bar q^{E_1(\tA)} \qqbinom{\lfloor \frac{N_t}4\rfloor}{2i}\qqbinom{i}{k_2}\qqbinom{i}{k_2 + \ell} \nonumber \\
&\simeq \frac1{(q)_\infty} \sum_\ell q^{\Delta_{1,3(\ell-t)+1}} \sum_{i,k_2} \bar q^{E_1(\tA)} \qqbinom{\lfloor \frac{N_t}4\rfloor+k_2}{2i}\qqbinom{i}{k_2}\qqbinom{i}{k_2 + \ell} \nonumber\\
& = \dots = \frac1{(q)_\infty} \sum_\ell q^{\Delta_{1,3(\ell-t)+1}} \bar q^{\Delta_{1,3\ell+2}} \specialqqbinom{2\lfloor \frac{N_t}4\rfloor - \ell}{\lfloor \frac{N_t}4\rfloor+\ell} \simeq \frac{1}{(q)_\infty (\bar q)_\infty} \sum_\ell q^{\Delta_{1,3 (\ell-t)+1}}\bar q^{\Delta_{1,3 \ell+2}}\end{aligned}$$ where the last steps follow those in . The result produces the second term in .
The odd contributions are treated similarly. We thus only write down the intermediate results. Each contribution is written as Z\^[,]{}\_[1,]{}= (q |q)\^[18]{}\_[k\_2,k\_3]{} \_[k\_1]{} (q |q)\^[E()]{} S\^[,]{}\_[1,]{} with
$$\begin{aligned}
{2}
S^{\bar \tB, \tB}_{1,\textrm{odd}} & = \sum_{k,\ell} q^{E_1(\bar \tB, \tB)} \qbinom{\lfloor \frac{N_t}4+\frac12\rfloor+k_2+k-1}{2\ell-k_1}\qbinom{\ell-\frac{k_1+1}2}{k}\qbinom{\ell-\frac{k_1+1}2}{k+t+k_2-k_3-1},
\\[0.2cm]
S^{\bar \tB, \tA}_{1,\textrm{odd}} & = \sum_{k,\ell} q^{E_1(\bar \tB, \tA)} \qbinom{\lfloor \frac{N_t}4+\frac12\rfloor+k_2+k-1}{2\ell-k_1}\qbinom{\ell-\frac{k_1+1}2}{k-1}\qbinom{\ell-\frac{k_1+1}2}{k+t+k_2-k_3-1},
\\[0.2cm]
S^{\bar \tA, \tB}_{1,\textrm{odd}} & = \sum_{k,\ell} q^{E_1(\bar \tA, \tB)} \qbinom{\lfloor \frac{N_t}4+\frac12\rfloor+k_2+k}{2\ell-k_1}\qbinom{\ell-\frac{k_1+1}2}{k}\qbinom{\ell-\frac{k_1+1}2}{k+t+k_2-k_3},
\\[0.2cm]
S^{\bar \tA, \tA}_{1,\textrm{odd}} & = \sum_{k,\ell} q^{E_1(\bar \tA, \tA)} \qbinom{\lfloor \frac{N_t}4+\frac12\rfloor+k_2+k}{2\ell-k_1}\qbinom{\ell-\frac{k_1+1}2}{k-1}\qbinom{\ell-\frac{k_1+1}2}{k+t+k_2-k_3}.\end{aligned}$$
These combine pairwise:
S\^[|, ]{}\_[1,]{} + S\^[|, ]{}\_[1,]{} , S\^[|, ]{}\_[1,]{} + S\^[|, ]{}\_[1,]{} . The partial partition functions then satisfy $$\begin{aligned}
{2}
Z_{1, \textrm{odd}}^{\bar\tB,\tB} + Z_{1, \textrm{odd}}^{\bar\tB,\tA} &\simeq \frac{\bar q^{\frac18}}{(q)_\infty} \sum_\ell q^{\Delta_{0,3(\ell-t)+1}} \sum_{i,k_2} \bar q^{E(\tB)} \qqbinom{\lfloor \frac{N_t}4\rfloor}{2i+1}\qqbinom{i}{k_2}\qqbinom{i}{k_2+\ell},
\\
Z_{1, \textrm{odd}}^{\bar\tA,\tB} + Z_{1, \textrm{odd}}^{\bar\tA,\tA} &\simeq \frac{\bar q^{\frac18}}{(q)_\infty} \sum_\ell q^{\Delta_{0,3(\ell-t)+1}} \sum_{i,k_2} \bar q^{E(\tA)} \qqbinom{\lfloor \frac{N_t}4\rfloor}{2i+1}\qqbinom{i}{k_2-1}\qqbinom{i}{k_2+\ell},\end{aligned}$$ and the final result is Z\_[1, ]{}\^[|,]{} + Z\_[1, ]{}\^[|,]{} + Z\_[1, ]{}\^[|,]{} + Z\_[1, ]{}\^[|,]{} \_q\^[\_[0,3 (-t)+1]{}]{}|q\^[\_[0,3 +2]{}]{}, as announced in .
For $d \equiv 2 \Mod 3$, the derivation is done using the same ideas. One considers separately the odd and even $k_1$ contributions. For the even case (and likewise for the odd case), one chooses $\epsilon$ so only the $q$-dependent part depends on the sum label $i$. One writes down the $S^{\tX, \tY}_{2,\textrm{even}}$ corresponding to each $Z^{\tX,\tY}_{2,\textrm{even}}$, and then combines $S^{\bar \tA, \tA}_{2,\textrm{even}}$ with $S^{\bar \tA, \tB}_{2,\textrm{even}}$ and $S^{\bar \tB, \tA}_{2,\textrm{even}}$ with $S^{\bar \tB, \tB}_{2,\textrm{even}}$. Their scaling limits are evaluated using and as the ratio of a power of $q$ with $(q)_\infty$. Combining the four contributions and using and one last time, the final result is
$$\begin{aligned}
{2}
Z_{2, \textrm{even}}^{\bar\tB,\tB} + Z_{2, \textrm{even}}^{\bar\tB,\tA} + Z_{2, \textrm{even}}^{\bar\tA,\tB} + Z_{2, \textrm{even}}^{\bar\tA,\tA}
&\simeq \frac{1}{(q)_\infty (\bar q)_\infty} \sum_\ell q^{\Delta_{1,3 (\ell-t)-1}}\bar q^{\Delta_{1,3 \ell+1}},
\\
Z_{2, \textrm{odd}}^{\bar\tB,\tB} + Z_{2, \textrm{odd}}^{\bar\tB,\tA} + Z_{2, \textrm{odd}}^{\bar\tA,\tB} + Z_{2, \textrm{odd}}^{\bar\tA,\tA}
&\simeq \frac{1}{(q)_\infty (\bar q)_\infty} \sum_\ell q^{\Delta_{0,3 (\ell-t)-1}}\bar q^{\Delta_{0,3 \ell+1}},\end{aligned}$$
consistent with .
Torus partition functions of critical dense polymers {#sec:CDPpartitionfunctions}
====================================================
For critical dense polymers, namely $\mathcal{LM}(1,2)$, the torus conformal partition functions for even $M$ are written in terms of $\mathcal W$-irreducible characters in [@PRVcyl2010; @MDPR13]. Here we express these results in terms of the $u(1)$ characters and present the results for $M$ odd. In this case, one must keep track of the overall sign $\varepsilon$ of each eigenvalue. In the scaling limit, the partition function in each standard module $\stanp_n^d$ then reads Z\_[d]{}(q,|q) = \_[Z]{} (-1)\^[M ]{} q\^[\_[2+d/2]{}]{} [|q]{}\^[\_[2-d/2]{}]{}. This holds for the odd and even parities of $N$ and for all values of $d$. The torus partition of the loop model with $\alpha = 2$ is computed from the modified trace . After simplification, we find that it can be written in terms of the $u(1)$ characters $\varkappa^8_{j}(q)$:
$$\begin{aligned}
{2}
\Zcdpodd &= \sum_{j=1,3,5,7} d_j^8 \Big( |\varkappa^8_j(q)|^2 + (-1)^M \varkappa^8_j(q)\varkappa^8_{8-j}(\bar q)\Big),\\
\Zcdpeven &= \sum_{j=0,2,4,6,8} d_j^8 \Big( |\varkappa^8_j(q)|^2 + (-1)^M \varkappa^8_j(q)\varkappa^8_{8-j}(\bar q)\Big).\end{aligned}$$
From the transformation laws , the cases with $M$ and $N$ of the same parity are found to be fully modular invariant, whereas the cases with opposite parities are invariant under the action of $T$ and covariant under the action of $S$, as is the case for percolation in . Using \_[2j]{}\^[4n]{}(q) \_[4n-2j]{}\^[4n]{}(q) = \_j\^n(q,1), the torus partition function for critical dense polymers can also be written in terms of the $u(1)$-characters $\varkappa^2_j(q,\pm1)$, with $j$ integer and half-integer for $N$ even and odd respectively:
$$\begin{aligned}
{2}
\label{eq:Zcdptake 2}
\Zcdpodd &= 2 \varkappa^2_{1/2}(q,1)\varkappa^2_{1/2}(\bar q,(-1)^M) + 2 \varkappa^2_{3/2}(q,1)\varkappa^2_{3/2}(\bar q,(-1)^M),\\[0.15cm]
\Zcdpeven &= \varkappa^2_{0}(q,1)\varkappa^2_{0}\big(\bar q,(-1)^M\big) +2 \varkappa^2_{1}(q,1)\varkappa^2_{1}\big(\bar q,(-1)^M\big) + \varkappa^2_{2}(q,1)\varkappa^2_{2}\big(\bar q,(-1)^M\big).\end{aligned}$$
For $M$ and $N$ both even, the resulting conformal torus partition function is $Z^{\text{Circ}}_{1,2}(q)$.
Examples of patterns of zeros
=============================
Strip boundary conditions {#sec:pats}
-------------------------
$$\begin{array}{l}
\begin{array}{rcccccccc}
& d=0 &d=2 & d=4 & d=6 & d=8 & d=10 & d=12 & d=14 \\
&\psun
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\end{pspicture}
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\pspolygon[fillstyle=solid,fillcolor=lightgray,linecolor=lightgray](0,-3)(0,0)(1,0)(1,-3)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\psdots[dotsize=0.09cm](0,-0.3)(0,-0.6)(0,-0.9)(0,-1.2)(0,-1.5)(0,-1.8)
\psdots[dotsize=0.09cm](1,-0.3)(1,-0.6)(1,-0.9)(1,-1.2)(1,-1.5)(1,-1.8)
\psdots[dotsize=0.09cm](0.5,-2.1)
\end{pspicture}
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\psdots[dotsize=0.09cm](0,-0.3)(0,-0.6)(0,-0.9)(0,-1.2)(0,-1.5)
\psdots[dotsize=0.09cm](1,-0.3)(1,-0.6)(1,-0.9)(1,-1.2)(1,-1.5)
\psdots[dotsize=0.09cm](0.5,-1.8)(0.5,-2.1)
\end{pspicture}
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\rput(-0.4,-3.3){\tiny{B}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\psdots[dotsize=0.09cm](0,-0.3)(0,-0.6)(0,-0.9)(0,-1.2)
\psdots[dotsize=0.09cm](1,-0.3)(1,-0.6)(1,-0.9)(1,-1.2)
\psdots[dotsize=0.09cm](0.5,-1.5)(0.5,-1.8)(0.5,-2.1)(0.5,-2.4)
\end{pspicture}
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\psdots[dotsize=0.09cm](0,-0.6)(0.5,-0.6)(1,-0.6)
\rput(-0.4,-3.3){\tiny{A}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\psdots[dotsize=0.09cm](0,-0.3)(0,-0.6)(0,-0.9)
\psdots[dotsize=0.09cm](1,-0.3)(1,-0.6)(1,-0.9)
\psdots[dotsize=0.09cm](0.5,-1.2)(0.5,-1.5)(0.5,-1.8)(0.5,-2.1)(0.5,-2.4)
\end{pspicture}
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\psdots[dotsize=0.09cm](0,-0.6)(0,-1.2)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)
\psdots[dotsize=0.09cm](1,-0.6)(1,-1.2)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\psdots[dotsize=0.09cm](0,-0.3)(0,-0.6)
\psdots[dotsize=0.09cm](1,-0.3)(1,-0.6)
\psdots[dotsize=0.09cm](0.5,-0.9)(0.5,-1.2)(0.5,-1.5)(0.5,-1.8)(0.5,-2.1)(0.5,-2.4)
\end{pspicture}
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\psdots[dotsize=0.09cm](0,-0.6)(0,-1.2)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)
\psdots[dotsize=0.09cm](1,-0.6)(1,-1.2)
\rput(-0.4,-3.3){\tiny{B}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\psdots[dotsize=0.09cm](0,-0.3)
\psdots[dotsize=0.09cm](1,-0.3)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.9)(0.5,-1.2)(0.5,-1.5)(0.5,-1.8)(0.5,-2.1)(0.5,-2.4)(0.5,-2.7)
\end{pspicture}
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\psdots[dotsize=0.09cm](0,-0.6)(0,-1.2)(0,-1.8)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)(0.5,-1.8)
\psdots[dotsize=0.09cm](1,-0.6)(1,-1.2)(1,-1.8)
\rput(-0.4,-3.3){\tiny{A}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)(0.5,-1.2)(0.5,-1.5)(0.5,-1.8)(0.5,-2.1)(0.5,-2.4)(0.5,-2.7)
\end{pspicture}
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\psdots[dotsize=0.09cm](0,-0.6)(0,-1.2)(0,-1.8)(0,-2.4)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)(0.5,-1.8)(0.5,-2.4)
\psdots[dotsize=0.09cm](1,-0.6)(1,-1.2)(1,-1.8)(1,-2.4)
\end{pspicture}
\\[0.4cm]
k^j: & * & 1 & 1,2 & 0,1,2,3 & 1, \dots, 5 & 1, \dots, 6 & 0, \dots, 7 & 1, \dots, 9
\\[0.1cm]
\ell^i: & * & * & * & 1 & 1,2 & 1,2&1,2,3 & 1,2,3,4
\\[0.1cm]
q^{\Delta}: & 1 &q^{1/3} & q^{2} & q^{5} & q^{28/3} & q^{15} & q^{22} & q^{91/3}
\end{array}\\
\\[-0.2cm]
\begin{array}{rccccccccc}
&d=1 &d=3 & d=5 & d=7 & d=9 & d=11 & d=13 & d=15 \\
&\psun
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\end{pspicture}
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\pspolygon[fillstyle=solid,fillcolor=lightgray,linecolor=lightgray](0,-3)(0,0)(1,0)(1,-3)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\psdots[dotsize=0.09cm](0,-0.3)(0,-0.6)(0,-0.9)(0,-1.2)(0,-1.5)(0,-1.8)
\psdots[dotsize=0.09cm](1,-0.3)(1,-0.6)(1,-0.9)(1,-1.2)(1,-1.5)(1,-1.8)
\psdots[dotsize=0.09cm](0.5,-2.1)(0.5,-2.4)
\end{pspicture}
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\rput(-0.4,-3.3){\tiny{A}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\psdots[dotsize=0.09cm](0,-0.3)(0,-0.6)(0,-0.9)(0,-1.2)(0,-1.5)
\psdots[dotsize=0.09cm](1,-0.3)(1,-0.6)(1,-0.9)(1,-1.2)(1,-1.5)
\psdots[dotsize=0.09cm](0.5,-1.8)(0.5,-2.1)(0.5,-2.4)
\end{pspicture}
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\psdots[dotsize=0.09cm](0,-0.6)
\psdots[dotsize=0.09cm](0.5,-0.6)
\psdots[dotsize=0.09cm](1,-0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\psdots[dotsize=0.09cm](0,-0.3)(0,-0.6)(0,-0.9)(0,-1.2)
\psdots[dotsize=0.09cm](1,-0.3)(1,-0.6)(1,-0.9)(1,-1.2)
\psdots[dotsize=0.09cm](0.5,-1.5)(0.5,-1.8)(0.5,-2.1)(0.5,-2.4)
\end{pspicture}
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\psdots[dotsize=0.09cm](0,-0.6)
\psdots[dotsize=0.09cm](0.5,-0.6)
\psdots[dotsize=0.09cm](1,-0.6)
\rput(-0.4,-3.3){\tiny{B}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\psdots[dotsize=0.09cm](0,-0.3)(0,-0.6)(0,-0.9)
\psdots[dotsize=0.09cm](1,-0.3)(1,-0.6)(1,-0.9)
\psdots[dotsize=0.09cm](0.5,-1.2)(0.5,-1.5)(0.5,-1.8)(0.5,-2.1)(0.5,-2.4)(0.5,-2.7)
\end{pspicture}
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\psdots[dotsize=0.09cm](0,-0.6)(0,-1.2)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)
\psdots[dotsize=0.09cm](1,-0.6)(1,-1.2)
\rput(-0.4,-3.3){\tiny{A}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\psdots[dotsize=0.09cm](0,-0.3)(0,-0.6)
\psdots[dotsize=0.09cm](1,-0.3)(1,-0.6)
\psdots[dotsize=0.09cm](0.5,-0.9)(0.5,-1.2)(0.5,-1.5)(0.5,-1.8)(0.5,-2.1)(0.5,-2.4)(0.5,-2.7)
\end{pspicture}
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\psdots[dotsize=0.09cm](0,-0.6)(0,-1.2)(0,-1.8)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)(0.5,-1.8)
\psdots[dotsize=0.09cm](1,-0.6)(1,-1.2)(1,-1.8)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\psdots[dotsize=0.09cm](0,-0.3)
\psdots[dotsize=0.09cm](1,-0.3)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.9)(0.5,-1.2)(0.5,-1.5)(0.5,-1.8)(0.5,-2.1)(0.5,-2.4)(0.5,-2.7)
\end{pspicture}
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\psdots[dotsize=0.09cm](0,-0.6)(0,-1.2)(0,-1.8)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)(0.5,-1.8)
\psdots[dotsize=0.09cm](1,-0.6)(1,-1.2)(1,-1.8)
\rput(-0.4,-3.3){\tiny{B}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)(0.5,-1.2)(0.5,-1.5)(0.5,-1.8)(0.5,-2.1)(0.5,-2.4)(0.5,-2.7)(0.5,-3.0)
\end{pspicture}
\begin{pspicture}[shift=-1.0](-0.4,0)(1.4,-3.7)
\psline{-}(0,-3)(0,0)(1,0)(1,-3)
\psdots[dotsize=0.09cm](0,-0.6)(0,-1.2)(0,-1.8)(0,-2.4)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)(0.5,-1.8)(0.5,-2.4)
\psdots[dotsize=0.09cm](1,-0.6)(1,-1.2)(1,-1.8)(1,-2.4)
\rput(-0.4,-3.3){\tiny{A}}
\end{pspicture}
\\[0.4cm]
k^j: & * & 0,1 & 1,2,3 & 1,2,3,4 & 0,\dots,5 & 1, \dots, 7 & 1, \dots, 8 & 0, \dots, 9
\\[0.1cm]
\ell^i: & * & * & 1 & 1 & 1,2 & 1,2,3 &1,2,3 & 1,2,3,4
\\[0.1cm]
q^\Delta: & 1 &q & q^{10/3} & q^{7} & q^{12} & q^{55/3} & q^{26} & q^{35}\\[-0.3cm]
\end{array}
\end{array}$$
\[fig:GSpatterns\]
$$\begin{aligned}
{2}
&\begin{array}{rcccccccc}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.3)(0,-0.6)(0,-0.9)
\psdots[dotsize=0.09cm](0.5,-1.2)
\psdots[dotsize=0.09cm](1,-0.3)(1,-0.6)(1,-0.9)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.3)(0,-0.6)(0,-1.2)
\psdots[dotsize=0.09cm](0.5,-0.9)
\psdots[dotsize=0.09cm](1,-0.3)(1,-0.6)(1,-1.2)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.3)(0,-0.6)
\psdots[dotsize=0.09cm](0.5,-0.9)(0.5,-1.2)(0.5,-1.5)
\psdots[dotsize=0.09cm](1,-0.3)(1,-0.6)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.3)(0,-0.9)(0,-1.2)
\psdots[dotsize=0.09cm](0.5,-0.6)
\psdots[dotsize=0.09cm](1,-0.3)(1,-0.9)(1,-1.2)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.3)(0,-0.9)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)(0.5,-1.5)
\psdots[dotsize=0.09cm](1,-0.3)(1,-0.9)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.6)(0,-0.9)(0,-1.2)
\psdots[dotsize=0.09cm](0.5,-0.3)
\psdots[dotsize=0.09cm](1,-0.6)(1,-0.9)(1,-1.2)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.3)(0,-1.2)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.9)(0.5,-1.5)
\psdots[dotsize=0.09cm](1,-0.3)(1,-1.2)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.3)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.9)(0.5,-1.2)
\psdots[dotsize=0.09cm](1,-0.3)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0,-0.6)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](1,-0.6)
\end{pspicture}
\\[-0.2cm]
k^j: & 1 & 2 & 1,2,3 & 3 & 1, 2,4 & 4 & 1,3,4 & 1,2,3
\\[0.1cm]
\ell^i: & * & * & 1,1 & * & 1,1 & * & 1,1 & *
\\[0.1cm]
q^{\Delta}: & q^{1/3} & q^{4/3} & q^{7/3} & q^{7/3} & q^{10/3} & q^{10/3} & q^{13/3} & q^{13/3}
\end{array}\\
\\[-0.5cm]&
\begin{array}{rcccccccc}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.6)(0,-0.9)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-1.2)(0.5,-1.5)
\psdots[dotsize=0.09cm](1,-0.6)(1,-0.9)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.3)(0,-1.5)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.9)(0.5,-1.2)
\psdots[dotsize=0.09cm](1,-0.3)(1,-1.5)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.6)(0,-1.2)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.9)(0.5,-1.5)
\psdots[dotsize=0.09cm](1,-0.6)(1,-1.2)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.6)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.9)(0.5,-1.2)
\psdots[dotsize=0.09cm](1,-0.6)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0,-0.6)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](1,-0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.3)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.9)(0.5,-1.2)(0.5,-1.5)(0.5,-1.8)
\psdots[dotsize=0.09cm](1,-0.3)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.6)(0,-1.5)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.9)(0.5,-1.2)
\psdots[dotsize=0.09cm](1,-0.6)(1,-1.5)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.9)(0,-1.2)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-1.5)
\psdots[dotsize=0.09cm](1,-0.9)(1,-1.2)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.9)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-1.2)
\psdots[dotsize=0.09cm](1,-0.9)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0,-0.6)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](1,-0.6)
\end{pspicture}
\\[-0.2cm]
k^j: &1,2,5 & 2,3,4 & 1,3,5 & 1,2,4 & 1,2,3,4,5 & 2,3,5 & 1,4,5 & 1,3,4
\\[0.1cm]
\ell^i: & 1,1 & 1,1 & 1,1 & * & 1,1,2,2 & 1,1 & 1,1 & *
\\[0.1cm]
q^{\Delta}: & q^{13/3} & q^{16/3} & q^{16/3} & q^{16/3} & q^{19/3} & q^{19/3} & q^{19/3} & q^{19/3}
\end{array}\\
\\[-0.5cm]&
\begin{array}{rcccccccc}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.6)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.9)(0.5,-1.2)(0.5,-1.5)(0.5,-1.8)
\psdots[dotsize=0.09cm](1,-0.6)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.9)(0,-1.5)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-1.2)
\psdots[dotsize=0.09cm](1,-0.9)(1,-1.5)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-1.2)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)
\psdots[dotsize=0.09cm](1,-1.2)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0,-0.6)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](1,-0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.9)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-1.2)(0.5,-1.5)(0.5,-1.8)
\psdots[dotsize=0.09cm](1,-0.9)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-1.2)(0,-1.5)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)
\psdots[dotsize=0.09cm](1,-1.2)(1,-1.5)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)(0.5,-1.2)(0.5,-1.5)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0,-0.6)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-1.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](1,-0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-1.2)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)(0.5,-1.5)(0.5,-1.8)
\psdots[dotsize=0.09cm](1,-1.2)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)
\end{pspicture}
\\[-0.2cm]
k^j: & 1,2,3,4,6 & 2,4,5 & 2,3,4 & 1,2,3,5,6 & 3,4,5 & 1,2,3,4,5 & 1,2,4,5,6
\\[0.1cm]
\ell^i: & 1,1,2,2 & 1,1 & * & 1,1,2,2 & 1,1 & 1,1 & 1,1,2,2
\\[0.1cm]
q^{\Delta}: & q^{22/3} & q^{22/3} & q^{22/3} & q^{25/3} & q^{25/3} & q^{25/3} & q^{28/3} \end{array}\\
\\[-0.5cm]&
\begin{array}{rccccccccc}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)(0.5,-1.2)(0.5,-1.5)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.6)(0,-1.2)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)
\psdots[dotsize=0.09cm](1,-0.6)(1,-1.2)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-1.5)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)(0.5,-1.2)(0.5,-1.8)
\psdots[dotsize=0.09cm](1,-1.5)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)(0.5,-1.2)(0.5,-1.5)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0,-1.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](1,-1.2)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-1.8)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)(0.5,-1.2)(0.5,-1.5)
\psdots[dotsize=0.09cm](1,-1.8)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)(0.5,-1.2)(0.5,-1.5)(0.5,-1.8)(0.5,-2.1)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)(0.5,-1.8)
\end{pspicture}
\\[-0.2cm]
k^j: & 1,2,3,4,5 & 1,3,4,5,6 & 1,2,3,4,5 & 2,3,4,5,6 & 1,2,3,4,5,6,7 \\[0.1cm]
\ell^i: & 1,2 & 1,1,2,2 & 2,2 & 1,1,2,2 & 1,1,2,2,3,3\\[0.1cm]
q^{\Delta}: & q^{28/3} & q^{31/3} & q^{31/3} & q^{34/3} & q^{37/3}\\[-0.3cm]
\end{array}\end{aligned}$$
$$\begin{aligned}
{2}
&\begin{array}{rcccccccccc}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.3)(0,-0.6)
\psdots[dotsize=0.09cm](0.5,-0.9)(0.5,-1.2)
\psdots[dotsize=0.09cm](1,-0.3)(1,-0.6)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\rput(-0.4,-2.5){\tiny{B}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.3)(0,-0.9)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)
\psdots[dotsize=0.09cm](1,-0.3)(1,-0.9)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\rput(-0.4,-2.5){\tiny{B}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.3)(0,-1.2)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.9)
\psdots[dotsize=0.09cm](1,-0.3)(1,-1.2)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\rput(-0.4,-2.5){\tiny{B}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](1,-0.6)(1,-0.9)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-1.2)
\psdots[dotsize=0.09cm](0,-0.6)(0,-0.9)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\rput(-0.4,-2.5){\tiny{B}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](1,-0.3)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.9)(0.5,-1.2)(0.5,-1.5)
\psdots[dotsize=0.09cm](0,-0.3)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.6)
\psdots[dotsize=0.09cm](0.5,-0.6)
\psdots[dotsize=0.09cm](1,-0.6)
\rput(-0.4,-2.5){\tiny{A}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.6)(0,-1.2)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.9)
\psdots[dotsize=0.09cm](1,-0.6)(1,-1.2)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\rput(-0.4,-2.5){\tiny{B}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.3)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.9)(0.5,-1.2)(0.5,-1.5)
\psdots[dotsize=0.09cm](1,-0.3)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdot[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)
\rput(-0.4,-2.5){\tiny{B}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](1,-0.6)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.9)(0.5,-1.2)(0.5,-1.5)
\psdots[dotsize=0.09cm](0,-0.6)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.6)
\psdots[dotsize=0.09cm](0.5,-0.6)
\psdots[dotsize=0.09cm](1,-0.6)
\rput(-0.4,-2.5){\tiny{A}}
\end{pspicture}
\\[0.5cm]
k^j: & 1,2 & 1,3 & 2,3 & 1,4 & 0,1,2,3 & 2,4 & 1,2,3,4& 0,1,2,4
\\[0.1cm]
\ell^i: & * & * & * & * & 1& * & 1,1 & 1
\\[0.1cm]
q^{\Delta}: & q^2 & q^3 & q^4 & q^4 & q^5 & q^5 & q^6 & q^6
\\[-0.1cm]
\end{array}
\\
&
\begin{array}{rccccccccccccc}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.9)(0,-1.2)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)
\psdots[dotsize=0.09cm](1,-0.9)(1,-1.2)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\rput(-0.4,-2.5){\tiny{B}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.6)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.9)(0.5,-1.2)(0.5,-1.5)
\psdots[dotsize=0.09cm](1,-0.6)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdot[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)
\rput(-0.4,-2.5){\tiny{B}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.9)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-1.2)(0.5,-1.5)
\psdots[dotsize=0.09cm](1,-0.9)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.6)
\psdots[dotsize=0.09cm](0.5,-0.6)
\psdots[dotsize=0.09cm](1,-0.6)
\rput(-0.4,-2.5){\tiny{A}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.9)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-1.2)(0.5,-1.5)
\psdots[dotsize=0.09cm](1,-0.9)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdot[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)
\rput(-0.4,-2.5){\tiny{B}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-1.2)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)(0.5,-1.5)
\psdots[dotsize=0.09cm](1,-1.2)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.6)
\psdots[dotsize=0.09cm](0.5,-0.6)
\psdots[dotsize=0.09cm](1,-0.6)
\rput(-0.4,-2.5){\tiny{A}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)(0.5,-1.2)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdot[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0,-0.6)
\psdot[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](1,-0.6)
\rput(-0.4,-2.5){\tiny{B}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-1.2)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)(0.5,-1.5)
\psdots[dotsize=0.09cm](1,-1.2)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdot[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)
\rput(-0.4,-2.5){\tiny{B}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-1.5)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)(0.5,-1.2)
\psdots[dotsize=0.09cm](1,-1.5)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.6)
\psdots[dotsize=0.09cm](0.5,-0.6)
\psdots[dotsize=0.09cm](1,-0.6)
\rput(-0.4,-2.5){\tiny{A}}
\end{pspicture}
\\[0.5cm]
k^j: & 3,4 & 1,2,3,5 & 0,1,3,4& 1,2,4,5&0,2,3,4& 1,2,3,4 & 1,3,4,5&1,2,3,4
\\[0.1cm]
\ell^i: & * & 1,1 & 1 & 1,1 & 1 & * & 1,1 &1
\\[0.1cm]
q^{\Delta}: & q^6 & q^7 & q^7 & q^8 & q^8 & q^8 & q^9 & q^9
\\[-0.1cm]
\end{array}\\
&
\begin{array}{rccccccccccccc}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-1.5)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)(0.5,-1.2)
\psdots[dotsize=0.09cm](1,-1.5)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdot[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)
\rput(-0.4,-2.5){\tiny{B}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)(0.5,-1.2)(0.5,-1.5)(0.5,-1.8)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-0.6)
\psdots[dotsize=0.09cm](0.5,-0.6)
\psdots[dotsize=0.09cm](1,-0.6)
\psdot[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-1.2)
\rput(-0.4,-2.5){\tiny{A}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)(0.5,-1.2)(0.5,-1.5)(0.5,-1.8)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0,-1.2)
\psdots[dotsize=0.09cm](0.5,-1.2)
\psdots[dotsize=0.09cm](1,-1.2)
\psdot[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)
\rput(-0.4,-2.5){\tiny{A}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)(0.5,-1.2)(0.5,-1.5)(0.5,-1.8)
\end{pspicture}
\begin{pspicture}[shift=-1.5](-0.4,-3.2)(1.4,0.5)
\psline{-}(0,-2.2)(0,0)(1,0)(1,-2.2)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)
\rput(-0.4,-2.5){\tiny{B}}
\end{pspicture}
\\[0.5cm]
k^j: & 2,3,4,5 & 0,1,2,3,4,5 & 0,1,2,3,4,5 & 1,2,3,4,5,6\\[0.1cm]
\ell^i: & 1,1 & 1,1,2 & 1,2,2 & 1,1,2,2\\[0.1cm]
q^{\Delta}: & q^{10} & q^{10} & q^{11} & q^{12}
\\[-0.3cm]
\end{array}\end{aligned}$$
\[fig:pats84\]
Periodic boundary conditions {#sec:patsperiodic}
----------------------------
$$\begin{array}{l}
\hspace{-1cm}
\begin{array}{rccccccc}
& d=0 &d=2 & d=4 & d=6 & d=8 & d=10 & d=12 \\
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0.3)(0,0.6)(0,0.9)
\psdots[dotsize=0.09cm](0,-0.3)(0,-0.6)(0,-0.9)
\psdots[dotsize=0.09cm](1,0.3)(1,0.6)(1,0.9)
\psdots[dotsize=0.09cm](1,-0.3)(1,-0.6)(1,-0.9)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0)(1,0)
\psdots[dotsize=0.09cm](0,0.3)(0,0.6)
\psdots[dotsize=0.09cm](0,-0.3)(0,-0.6)
\psdots[dotsize=0.09cm](1,0.3)(1,0.6)
\psdots[dotsize=0.09cm](1,-0.3)(1,-0.6)
\psdots[dotsize=0.09cm](0.5,0.9)(0.5,-0.9)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\rput(-0.4,-2.0){\tiny{$\bar\tA$}}
\rput(-0.4,2.0){\tiny{A}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0.3)(0,0.6)
\psdots[dotsize=0.09cm](0,-0.3)(0,-0.6)
\psdots[dotsize=0.09cm](1,0.3)(1,0.6)
\psdots[dotsize=0.09cm](1,-0.3)(1,-0.6)
\psdots[dotsize=0.09cm](0.5,0.9)(0.5,-0.9)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\rput(-0.4,-2.0){\tiny{$\bar\tB$}}
\rput(-0.4,2.0){\tiny{B}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0)(1,0)
\psdots[dotsize=0.09cm](0,0.3)
\psdots[dotsize=0.09cm](0,-0.3)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.9)
\psdots[dotsize=0.09cm](0.5,0.6)(0.5,0.9)
\psdots[dotsize=0.09cm](1,0.3)
\psdots[dotsize=0.09cm](1,-0.3)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,-0.6)
\psdots[dotsize=0.09cm](0,0.6)
\psdots[dotsize=0.09cm](0.5,-0.6)
\psdots[dotsize=0.09cm](0.5,0.6)
\psdots[dotsize=0.09cm](1,-0.6)
\psdots[dotsize=0.09cm](1,0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0.3)
\psdots[dotsize=0.09cm](0,-0.3)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.9)(0.5,-1.2)
\psdots[dotsize=0.09cm](0.5,0.6)(0.5,0.9)(0.5,1.2)
\psdots[dotsize=0.09cm](1,0.3)
\psdots[dotsize=0.09cm](1,-0.3)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,-0.6)
\psdots[dotsize=0.09cm](0,0.6)
\psdots[dotsize=0.09cm](0.5,-0.6)
\psdots[dotsize=0.09cm](0.5,0.6)
\psdots[dotsize=0.09cm](1,-0.6)
\psdots[dotsize=0.09cm](1,0.6)
\rput(-0.4,-2.0){\tiny{$\bar\tA$}}
\rput(-0.4,2.0){\tiny{A}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0)(1,0)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)
\psdots[dotsize=0.09cm](0.5,0.3)(0.5,0.6)(0.5,0.9)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,-0.6)
\psdots[dotsize=0.09cm](0,0.6)
\psdots[dotsize=0.09cm](0.5,-0.6)
\psdots[dotsize=0.09cm](0.5,0.6)
\psdots[dotsize=0.09cm](1,-0.6)
\psdots[dotsize=0.09cm](1,0.6)
\rput(-0.4,-2.0){\tiny{$\bar\tB$}}
\rput(-0.4,2.0){\tiny{B}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)(0.5,-1.2)
\psdots[dotsize=0.09cm](0.5,0.3)(0.5,0.6)(0.5,0.9)(0.5,1.2)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,-0.6)(0,-1.2)
\psdots[dotsize=0.09cm](0,0.6)(0,1.2)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-1.2)
\psdots[dotsize=0.09cm](0.5,0.6)(0.5,1.2)
\psdots[dotsize=0.09cm](1,-0.6)(1,-1.2)
\psdots[dotsize=0.09cm](1,0.6)(1,1.2)
\end{pspicture}
\\[0.4cm]
\sigma: & 1 & 1 & 1& 1& 1& 1& 1
\\[0.1cm]
k_-^j\sep k_+^j: & *\sep* & 0\sep0 & 1\sep1 & 1,2\sep1,2 & 0,1,2\sep0,1,2 & 1,2,3\sep1,2,3 & 1,2,3,4\sep1,2,3,4
\\[0.2cm]
\ell_-^j\sep\ell_+^j: & *\sep* & *\sep* & *\sep* & 1\sep1 & 1\sep1 & 1\sep1&1,2\sep1,2
\\[0.1cm]
q^\Delta\bar q^{\bar \Delta}: & (q\bar q)^{-\frac1{24}} &(q\bar q)^{\frac1{8}} & (q\bar q)^{\frac5{8}} & (q\bar q)^{\frac{35}{24}} & (q\bar q)^{\frac{21}{8}} & (q\bar q)^{\frac{33}{8}} & (q\bar q)^{\frac{143}{24}}
\end{array}\\
\\\hspace{-1cm}
\begin{array}{rccccccc}
& d=1 &d=3 & d=5 & d=7 & d=9 & d=11 & d=13 \\
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\pspolygon[fillstyle=solid,fillcolor=lightgray,linecolor=lightgray](0,-1.7)(0,1.7)(1,1.7)(1,-1.7)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0)(1,0)
\psdots[dotsize=0.09cm](0,0.3)(0,0.6)
\psdots[dotsize=0.09cm](0,-0.3)(0,-0.6)
\psdots[dotsize=0.09cm](1,0.3)(1,0.6)
\psdots[dotsize=0.09cm](1,-0.3)(1,-0.6)
\psdots[dotsize=0.09cm](0.5,0.9)(0.5,-0.9)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0)(0.5,0)(1,0)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0.3)(0,0.6)
\psdots[dotsize=0.09cm](0,-0.3)(0,-0.6)
\psdots[dotsize=0.09cm](1,0.3)(1,0.6)
\psdots[dotsize=0.09cm](1,-0.3)(1,-0.6)
\psdots[dotsize=0.09cm](0.5,0.9)(0.5,1.2)(0.5,-0.9)(0.5,-1.2)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psdots[dotsize=0.09cm](0,0)(0.5,0)(1,0)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\rput(-0.4,-2.0){\tiny{$\bar\tA$}}
\rput(-0.4,2.0){\tiny{A}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0)(1,0)
\psdots[dotsize=0.09cm](0,0.3)
\psdots[dotsize=0.09cm](0,-0.3)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.9)
\psdots[dotsize=0.09cm](0.5,0.6)(0.5,0.9)
\psdots[dotsize=0.09cm](1,0.3)
\psdots[dotsize=0.09cm](1,-0.3)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0)(0.5,0)(1,0)
\rput(-0.4,-2.0){\tiny{$\bar\tB$}}
\rput(-0.4,2.0){\tiny{B}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0.3)
\psdots[dotsize=0.09cm](0,-0.3)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.9)(0.5,-1.2)
\psdots[dotsize=0.09cm](0.5,0.6)(0.5,0.9)(0.5,1.2)
\psdots[dotsize=0.09cm](1,0.3)
\psdots[dotsize=0.09cm](1,-0.3)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0)(0.5,0)(1,0)
\psdots[dotsize=0.09cm](0,-0.6)
\psdots[dotsize=0.09cm](0,0.6)
\psdots[dotsize=0.09cm](0.5,-0.6)
\psdots[dotsize=0.09cm](0.5,0.6)
\psdots[dotsize=0.09cm](1,-0.6)
\psdots[dotsize=0.09cm](1,0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0)(1,0)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)(0.5,-1.2)
\psdots[dotsize=0.09cm](0.5,0.3)(0.5,0.6)(0.5,0.9)(0.5,1.2)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0)(0.5,0)(1,0)
\psdots[dotsize=0.09cm](0,-0.6)
\psdots[dotsize=0.09cm](0,0.6)
\psdots[dotsize=0.09cm](0.5,-0.6)
\psdots[dotsize=0.09cm](0.5,0.6)
\psdots[dotsize=0.09cm](1,-0.6)
\psdots[dotsize=0.09cm](1,0.6)
\rput(-0.4,-2.0){\tiny{$\bar\tA$}}
\rput(-0.4,2.0){\tiny{A}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)(0.5,-0.9)(0.5,-1.2)
\psdots[dotsize=0.09cm](0.5,0.3)(0.5,0.6)(0.5,0.9)(0.5,1.2)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-2.5)(1.4,2.5)
\psline{-}(0,-1.7)(0,1.7)\psline{-}(1,-1.7)(1,1.7)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0)(0.5,0)(1,0)
\psdots[dotsize=0.09cm](0,-0.6)
\psdots[dotsize=0.09cm](0,0.6)
\psdots[dotsize=0.09cm](0.5,-0.6)
\psdots[dotsize=0.09cm](0.5,0.6)
\psdots[dotsize=0.09cm](1,-0.6)
\psdots[dotsize=0.09cm](1,0.6)
\rput(-0.4,-2.0){\tiny{$\bar\tB$}}
\rput(-0.4,2.0){\tiny{B}}
\end{pspicture}
\\[0.4cm]
\sigma: & -1 & -1 & -1& -1& -1& -1& -1
\\[0.1cm]
k_-^j\sep k_+^j: & *\sep* & 1\sep1 & 0,1\sep0,1 & 1,2\sep1,2 & 1,2,3\sep1,2,3 & 0,1,2,3\sep0,1,2,3 & 1,2,3,4\sep 1,2,3,4
\\[0.2cm]
\ell_-^j\sep\ell_+^j: & *\sep* & *\sep1 & *\sep1 & *\sep1 & 1\sep1,2 & 1\sep1,2&1\sep1,2
\\[0.1cm]
q^\Delta\bar q^{\bar \Delta}: & 1 &(q\bar q)^{\frac1{3}} & (q\bar q)^{1} & (q\bar q)^{2} & (q\bar q)^{\frac{10}{3}} & (q\bar q)^{5} & (q\bar q)^{7}
\end{array}
\end{array}$$
\[fig:GSpatternsp\]
$$\begin{aligned}
{2}
&\begin{array}{rcccccccc}
&\psun
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,-0.3)(0,0)(0,0.3)
\psdots[dotsize=0.09cm](1,-0.3)(1,0)(1,0.3)
\end{pspicture}
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,-0.3)(0,0.3)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,0.6)
\psdots[dotsize=0.09cm](1,-0.3)(1,0.3)
\end{pspicture}
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,0)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,0.3)
\psdots[dotsize=0.09cm](1,0)
\end{pspicture}
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0,0)(1,0)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,-0.3)(0,-0.6)
\psdots[dotsize=0.09cm](0.5,0.3)(0.5,0.6)
\psdots[dotsize=0.09cm](1,-0.3)(1,-0.6)
\end{pspicture}
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0.3)(0,0.6)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.3)
\psdots[dotsize=0.09cm](1,0.3)(1,0.6)
\end{pspicture}
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.3)(0.5,0.3)(0.5,0.6)
\psdots[dotsize=0.09cm](1,0)
\end{pspicture}
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)(0.5,0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,-0.3)(0,0.6)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,0.3)
\psdots[dotsize=0.09cm](1,-0.3)(1,0.6)
\end{pspicture}
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,0)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,-0.6)(0,0.3)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,0.6)
\psdots[dotsize=0.09cm](1,-0.6)(1,0.3)
\end{pspicture}
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,0)
\end{pspicture}
\\[0.2cm]
k^j_-\sep k^j_+: & * \sep * & 1 \sep 1 & 1 \sep 1 & * \sep 1,2 & 1,2 \sep * & 1,2 \sep 1,2 & 1 \sep 2 & 2 \sep 1
\\[0.2cm]
\ell^i_-\sep \ell^i_+: & * \sep *& 1 \sep 1 & * \sep *& * \sep 1,1& 1,1 \sep *& 1,1 \sep 1,1 & 1 \sep 1& 1 \sep 1
\\[0.1cm]
q^{\Delta}\bar q^{\bar \Delta}: & q^{-\frac1{24}} \bar q^{-\frac1{24}} & q^{\frac13}\bar q^{\frac13} & q^{\frac13}\bar q^{\frac13} & q^{\frac{23}{24}}\bar q^{-\frac1{24}} & q^{-\frac1{24}}\bar q^{\frac{23}{24}} & q^{\frac{23}{24}}\bar q^{\frac{23}{24}} & q^{\frac43}\bar q^{\frac13} & q^{\frac13}\bar q^{\frac43}
\\[0.1cm]
\sigma: & 1 & -1 & -1 & 1 & 1 & 1 & -1 & -1
\\[-0.1cm]
\end{array}
\\[0.4cm]
&
\begin{array}{rcccccccc}
&\psun
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,-0.6)(0,0.6)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,0.3)
\psdots[dotsize=0.09cm](1,-0.6)(1,0.6)
\end{pspicture}
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,0)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0.3)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,0)
\psdots[dotsize=0.09cm](1,0.3)
\end{pspicture}
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0,0)(1,0)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,-0.3)
\psdots[dotsize=0.09cm](0.5,0.3)(0.5,-0)
\psdots[dotsize=0.09cm](1,-0.3)
\end{pspicture}
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0,0.6)(1,0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.3)(0.5,0.3)(0.5,0.6)
\end{pspicture}
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,-0.6)(0.5,-0.6)(1,-0.6)(0,0.6)(0.5,0.6)(1,0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.3)(0.5,0.3)(0.5,0.6)
\end{pspicture}
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,-0.6)(0.5,-0.6)(1,-0.6)(0,0.6)(0.5,0.6)(1,0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0.3)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.3)(0.5,-0)(0.5,0.6)
\psdots[dotsize=0.09cm](1,0.3)
\end{pspicture}
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)(0.5,0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,-0.3)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,0.0)(0.5,0.3)(0.5,0.6)
\psdots[dotsize=0.09cm](1,-0.3)
\end{pspicture}
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,0.6)(0.5,0)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.3)(0.5,0.3)(0.5,0.6)
\end{pspicture}
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0,-0.6)(0.5,0.6)(1,-0.6)
\end{pspicture}
\\[0.2cm]
k^j_-\sep k^j_+: &2 \sep 2 & 1\sep 2 & * \sep 1,2 & 1,2 \sep 1,2 & 1,2 \sep 1,2 & 1,2 \sep 1,3 & 1 \sep 1,2,3 & 1,2 \sep 1,2
\\[0.2cm]
\ell^i_-\sep \ell^i_+: & 1 \sep 1 & * \sep * & * \sep * & 1 \sep 1 & 1 \sep 1 & 1,1 \sep 1,1 & 1 \sep 1,1,2 & * \sep 1,1
\\[0.1cm]
q^{\Delta}\bar q^{\bar \Delta}: & q^{\frac43}\bar q^{\frac43} & q^{\frac{4}{3}}\bar q^{\frac{1}{3}} & q^{\frac{47}{24}}\bar q^{-\frac{1}{24}} & q^{\frac{35}{24}}\bar q^{\frac{35}{24}} & q^{\frac{35}{24}}\bar q^{\frac{35}{24}} & q^{\frac{47}{24}}\bar q^{\frac{23}{24}} & q^{\frac{7}{3}}\bar q^{\frac{1}{3}} & q^{\frac{23}{24}}\bar q^{\frac{47}{24}}
\\[0.1cm]
\sigma: & -1 & -1 & 1 & 1 & 1 & 1 & -1 & 1
\\[-0.1cm]
\end{array}
\\[0.4cm]
&
\begin{array}{rccccccccccccc}
&\psun
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.3)(0.5,0.3)(0.5,0.6)
\end{pspicture}
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0,0.6)(0.5,-0.6)(1,0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0.6)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.3)(0.5,0)(0.5,0.3)
\psdots[dotsize=0.09cm](1,0.6)
\end{pspicture}
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)(0.5,0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,-0.6)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,0.0)(0.5,0.3)(0.5,0.6)
\psdots[dotsize=0.09cm](1,-0.6)
\end{pspicture}
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,0)(0.5,0.6)
\end{pspicture}
&\psun
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0.5,-0.9)(0.5,-0.6)(0.5,-0.3)(0.5,0.3)(0.5,0.6)(0.5,0.9)
\end{pspicture}
\begin{pspicture}[shift=-0.4](-0.4,-1.0)(1.4,1.0)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)(0.5,0)(0.5,0.6)
\end{pspicture}
\\[0.2cm]
k^j_-\sep k^j_+: & 1,2\sep 1,2 & 1,2\sep 2,3 & 2\sep 1,2,3 & 1,2,3\sep 1,2,3
\\[0.2cm]
\ell^i_-\sep \ell^i_+: & 1,1\sep * & 1,1\sep 1,1 & 1\sep 1,1,2 & 1,1,2\sep 1,1,2
\\[0.1cm]
q^{\Delta}\bar q^{\bar \Delta}: & q^{\frac{47}{24}}\bar q^{\frac{23}{24}} & q^{\frac{71}{24}}\bar q^{\frac{23}{24}} & q^{\frac{7}{3}}\bar q^{\frac{4}{3}} & q^{\frac{7}{3}}\bar q^{\frac{7}{3}}
\\[0.1cm]
\sigma: & 1 & 1 & -1 & -1
\\[-0.1cm]
\end{array}\end{aligned}$$
\[fig:patsP60\]
$$\begin{aligned}
{2}
&\begin{array}{rcccccccccc}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,-0.3)(0,0.3)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,0.6)
\psdots[dotsize=0.09cm](1,-0.3)(1,0.3)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\rput(-0.4,1.3){\tiny{A}}
\rput(-0.4,-1.3){\tiny{$\bar \tA$}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,-0.3)(0,-0.6)
\psdots[dotsize=0.09cm](0.5,0.3)(0.5,0.6)
\psdots[dotsize=0.09cm](1,-0.3)(1,-0.6)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\rput(-0.4,1.3){\tiny{A}}
\rput(-0.4,-1.3){\tiny{$\bar \tA$}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0.3)(0,0.6)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,-0.6)
\psdots[dotsize=0.09cm](1,0.3)(1,0.6)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\rput(-0.4,1.3){\tiny{A}}
\rput(-0.4,-1.3){\tiny{$\bar \tA$}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,-0.6)(0,0.3)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,0.6)
\psdots[dotsize=0.09cm](1,-0.6)(1,0.3)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\rput(-0.4,1.3){\tiny{A}}
\rput(-0.4,-1.3){\tiny{$\bar \tA$}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0.6)(0,-0.3)
\psdots[dotsize=0.09cm](0.5,0.3)(0.5,-0.6)
\psdots[dotsize=0.09cm](1,0.6)(1,-0.3)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\rput(-0.4,1.3){\tiny{A}}
\rput(-0.4,-1.3){\tiny{$\bar \tA$}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.3)(0.5,0.3)(0.5,0.6)
\psdots[dotsize=0.09cm](1,0)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,0)
\rput(-0.4,1.3){\tiny{A}}
\rput(-0.4,-1.3){\tiny{$\bar \tA$}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.3)(0.5,0.3)(0.5,0.6)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0,0)(1,0)
\rput(-0.4,1.3){\tiny{A}}
\rput(-0.4,-1.3){\tiny{$\bar \tA$}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,-0.6)(0,0.6)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,0.3)
\psdots[dotsize=0.09cm](1,-0.6)(1,0.6)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\rput(-0.4,1.3){\tiny{A}}
\rput(-0.4,-1.3){\tiny{$\bar \tA$}}
\end{pspicture}
\\[0.5cm]
\sigma: & 1 & -1 & -1 & 1 &1& -1 & -1 & 1 \\[0.1cm]
k^j_-\sep k^j_+: & 0\sep 0 & *\sep 0,1 & 0,1\sep * & 1\sep 0 & 0\sep 1 & 0,1\sep 0,1 & 0,1\sep 0,1 & 1\sep 1
\\[0.2cm]
\ell^i_-\sep \ell^i_+: & *\sep * & *\sep * & *\sep * & *\sep * & *\sep * & 1\sep 1 & *\sep * & *\sep *
\\[0.1cm]
q^{\Delta}\bar q^{\bar \Delta}: & q^{\frac18}\bar q^{\frac18} & q & \bar q & q^{\frac18}\bar q^{\frac98} & q^{\frac98}\bar q^{\frac18} & q\bar q & q\bar q & q^{\frac98}\bar q^{\frac98}
\end{array}
\\[0.3cm]
&
\begin{array}{rccccccccccccc}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0.3)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.3)(0.5,0)(0.5,0.6)
\psdots[dotsize=0.09cm](1,0.3)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,0)
\rput(-0.4,1.3){\tiny{A}}
\rput(-0.4,-1.3){\tiny{$\bar \tA$}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,-0.3)
\psdots[dotsize=0.09cm](0.5,0.6)(0.5,0.3)(0.5,0)(0.5,-0.6)
\psdots[dotsize=0.09cm](1,-0.3)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,0.6)
\rput(-0.4,1.3){\tiny{A}}
\rput(-0.4,-1.3){\tiny{$\bar \tA$}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.3)(0.5,0.3)(0.5,0.6)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0)(0.5,0)(1,0)
\rput(-0.4,1.3){\tiny{B}}
\rput(-0.4,-1.3){\tiny{$\bar \tA$}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.3)(0.5,0.3)(0.5,0.6)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0)(0.5,0)(1,0)
\rput(-0.4,1.3){\tiny{A}}
\rput(-0.4,-1.3){\tiny{$\bar \tB$}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,0.6)
\psdots[dotsize=0.09cm](0.5,-0.6)(0.5,-0.3)(0.5,0)(0.5,0.3)
\psdots[dotsize=0.09cm](1,0.6)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,0)
\rput(-0.4,1.3){\tiny{A}}
\rput(-0.4,-1.3){\tiny{$\bar \tA$}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0,-0.6)
\psdots[dotsize=0.09cm](0.5,-0.3)(0.5,0)(0.5,0.3)(0.5,0.6)
\psdots[dotsize=0.09cm](1,-0.6)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,0.6)
\rput(-0.4,1.3){\tiny{A}}
\rput(-0.4,-1.3){\tiny{$\bar \tA$}}
\end{pspicture}
&\psun
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[dotsize=0.09cm](0.5,-0.9)(0.5,-0.6)(0.5,-0.3)(0.5,0.3)(0.5,0.6)(0.5,0.9)
\end{pspicture}
\begin{pspicture}[shift=-1.2](-0.4,-1.6)(1.4,1.6)
\psline{-}(0,-1)(0,1)\psline{-}(1,-1)(1,1)\psline[linestyle=dashed,dash=1pt 1pt]{-}(0,0)(1,0)
\psdots[linecolor=black,fillcolor=lightgray,dotstyle=o,dotsize=0.09cm](0.5,-0.6)(0.5,0.6)
\rput(-0.4,1.3){\tiny{A}}
\rput(-0.4,-1.3){\tiny{$\bar \tA$}}
\end{pspicture}
\\[0.5cm]
\sigma: & -1 & 1 & -1 & -1 & -1 & 1 & 1\\[0.1cm]
k^j_-\sep k^j_+: & 0,1\sep 0,2 & 0\sep 0,1,2 & 0,1\sep 1,2 & 1,2\sep 0,1 & 0,1\sep 1,2 & 1\sep 0,1,2 & 0,1,2\sep 0,1,2
\\[0.2cm]
\ell^i_-\sep \ell^i_+: & 1\sep 1 & *\sep 1,1 & *\sep 1 & *\sep 1 & 1\sep 1 & *\sep 1,1 & 1,1\sep 1,1
\\[0.1cm]
q^{\Delta}\bar q^{\bar \Delta}: & q^{2} \bar q^{1} & q^{\frac{17}8} \bar q^{\frac18} & q^2 \bar q & q \bar q^{2} & q^{3} \bar q & q^{\frac{17}8} \bar q^{\frac{9}8} & q^{\frac{17}8} \bar q^{\frac{17}8}
\end{array}\end{aligned}$$
\[fig:patsP62\]
[99]{} S. Broadbent, J. Hammersley, [*Percolation processes I. Crystals and mazes*]{}, Proc. Camb. Phil. Soc. [**53**]{} (1957) 629.
H. Kesten, [*Percolation theory for mathematicians*]{}, Springer Science (1982).
D. Stauffer, A. Aharony, [*Introduction to Percolation Theory*]{}, Taylor and Francis (1992).
G. Grimmett, [*Percolation and disordered systems*]{}, Lectures on Probability Theory and Statistics, Springer (1997).
A.A. Saberi, [*Recent advances in percolation theory and its applications*]{}, Phys. Rep. [**578**]{} (2015) 1–32, .
R.J. Baxter, [*Exactly Solved Models in Statistical Mechanics*]{}, Academic Press (1982).
P. Di Francesco, P. Mathieu, D. Sénéchal, [*Conformal Field Theory*]{}, Springer (1997).
V. Gurarie, [*Logarithmic operators in conformal field theory*]{}, Nucl. Phys. [**B410**]{} (1993) 535–549, .
A. Gainutdinov, D. Ridout, I. Runkel (Guest Editors), [*Special issue on logarithmic conformal field theory*]{}, J. Phys. A: Math. Theor. [**46(49)**]{} (2013).
T.E. Harris, [*A lower bound for the critical probability in a certain percolation process*]{}, Proc. Cambridge Philos. Soc. [**56**]{} (1960) 13–20.
H. Kesten, [*The critical probability of bond percolation on the square lattice equals $1/2$*]{}, Commun. Math. Phys. [**74**]{} (1980) 41–59.
R. Vasseur, J.L. Jacobsen, H. Saleur, [*Logarithmic observables in critical percolation*]{}, J. Stat. Mech. (2012) L07001, .
G. Delfino, M. Picco, R. Santachiara, J. Viti, [*Spin clusters and conformal field theory*]{}, J. Stat. Mech. (2013) P11011, .
H. Temperley, E. Lieb, [*Relations between the “percolation” and “colouring” problem and other graph-theoretical problems associated with regular planar lattices: Some exact results for the “percolation” problem*]{}, Proc. Roy. Soc. London Ser. [**A322**]{} (1971) 251–280.
V.F.R. Jones, [*Planar algebras I*]{}, .
S. Roux, E. Guyon, D. Sornette, [*Hull percolation*]{}, J. Phys. A [**21**]{} (1988) L475–L482.
B. Duplantier, [*Hull percolation and standard percolation*]{}, J. Phys. A [**21**]{} (1988) 3969–3973.
J. Cardy, [*Critical percolation in finite geometries*]{}, J. Phys. A [**25**]{} (1992) L201–L206, ; [*Conformal invariance and percolation*]{}, .
R. Langlands, P. Pouliot, Y. Saint-Aubin, [*Conformal invariance in two-dimensional percolation*]{}, Bull. Am. Math. Soc. [**30**]{} (1994) 1–61, .
J. Cardy, [*Scaling and Renormalization in Statistical Physics*]{}, Cambridge Lecture Notes in Physics, Cambridge University Press (1996).
M.P.M. den Nijs, [*A relation between the temperature exponents of the eight-vertex model and $q$-state Potts model*]{}, J. Phys. A [**12**]{} (1979) 1857–68.
B. Nienhuis, E.K. Riedel, M. Schick, [*Magnetic exponents of the two-dimensional $q$-state Potts model*]{}, J. Phys. A [**13**]{} (1980) L189–92.
B. Nienhuis, [*Exact critical point and critical exponents of $O(n)$ models in two dimensions*]{}, Phys. Rev. Lett. [**49**]{} (1982) 1062–1065.
B. Nienhuis, [*Critical behaviour of two-dimensional spin models and charge asymmetry in the Coulomb gas*]{}, J. Stat. Phys. [**34**]{} (1984) 731–761.
P. di Francesco, H. Saleur, J.-B. Zuber, [*Relations between the Coulomb gas picture and conformal invariance of two-dimensional critical models*]{}, J. Stat. Phys. [**49**]{} (1987) 57–79; [*Generalized Coulomb-gas formalism for two-dimensional critical models based on $SU(2)$ coset construction*]{}, Nucl. Phys. [**B300**]{} (1988) 393–432.
G. Delfino, J. Cardy, [*Universal amplitude ratios in the two-dimensional $q$-state Potts model and percolation from quantum field theory*]{}, Nucl. Phys. [**B519**]{} (1998) 551–578, .
G. Delfino, J. Viti, J. Cardy, [*Universal amplitude ratios of two-dimensional percolation from field theory*]{}, J. Phys. A [**43**]{} (2010) 152001, .
H. Saleur, [*New exact exponents for the two-dimensional self-avoiding walks*]{}, J. Phys. A: Math. Gen. [**19**]{} (1986) L807–L810.
H. Saleur, [*Conformal invariance for polymers and percolation*]{}, J. Phys. A: Math. Gen. [**20**]{} (1987) 455–470.
H. Saleur, [*Magnetic properties of the two-dimensional $n = 0$ vector model*]{}, Phys. Rev. [**B35**]{} (1987) 3657–3660.
B. Duplantier, [*Exact critical exponents for two-dimensional dense polymers*]{}, J. Phys. A: Math. Gen. [**19**]{} (1986) L1009–L1014.
B. Duplantier, H. Saleur, [*Exact surface and wedge exponents for polymers in two dimensions*]{}, Phys. Rev. Lett. [**57**]{} (1986) 3179–3182.
H. Saleur, B. Duplantier, [*Exact determination of the percolation hull exponent in two dimensions*]{}, Phys. Rev. Lett. [**58**]{} (1987) 2325–2328.
H. Saleur, [*Polymers and percolation in two dimensions and twisted $N=2$ supersymmetry*]{}, Nucl. Phys. [**B382**]{} (1992) 486–531, .
T. Deguchi, K. Fabricius, B.M. McCoy, [*The $sl_2$ loop algebra symmetry of the six-vertex model at roots of unity*]{}, J. Stat. Phys. [**102**]{} (2001) 701–736, .
T. Deguchi, [*The $sl_2$ loop algebra symmetry of the twisted transfer matrix of the six-vertex model at roots of unity*]{}, J. Phys. A [**37**]{} (2004) 347–358, ; [*Regular XXZ Bethe states as highest weight vectors of the $sl_2$ loop algebra at roots of unity*]{}, J. Phys. A [**40**]{} (2007) 7473–7508, .
P.A. Pearce, J. Rasmussen, J.-B. Zuber, [*Logarithmic minimal models*]{}, J. Stat. Mech. (2006) P11017, .
P.A. Pearce, J. Rasmussen, I.Y. Tipunin, [*Critical dense polymers with Robin boundary conditions, half-integer Kac labels and ${\mathbb Z}_4$ fermions*]{}, Nucl. Phys. [**B889**]{} (2014) 580–636, .
J.-E. Bourgine, P.A. Pearce, E. Tartaglia, [*Logarithmic minimal models with Robin boundary conditions*]{}, J. Stat. Mech. (2016) 063104, .
R. Couvreur, J.L. Jacobsen, R. Vasseur, [Non-scalar operators for the Potts model in arbitrary dimension]{}, .
P.A. Pearce, J. Rasmussen, [*Coset construction of logarithmic minimal models: branching rules and branching functions*]{}, J. Phys. A [**46**]{} (2013) 355402, .
P. Mathieu, D. Ridout, [*From percolation to logarithmic conformal field theory*]{}, Phys. Lett. [**B29**]{} (2007) 120–129, .
G. Moore, N. Seiberg, [*Classical and quantum conformal field theory*]{}, Commun. Math. Phys. [**123**]{} (1989) 177–254.
A. Morin-Duchesne, Y. Saint-Aubin, [*The Jordan structure of two-dimensional loop models*]{}, J. Stat. Mech. (2011) P04007, .
A. Morin-Duchesne, Y. Saint-Aubin, [*Jordan cells of periodic loop models*]{}, J. Phys. A: Math. Theor. [**46**]{} (2013) 494013, .
J.L. Cardy, [*Operator content of two-dimensional conformally invariant theories*]{}, Nucl. Phys. [**B270**]{} (1986) 186–204.
A. Cappelli, C. Itzykson, J.-B. Zuber, [*Modular Invariant Partition Functions in Two Dimensions*]{}, Nucl. Phys. [**B280**]{} (1987) 445–465.
S. Smirnov, [*Critical percolation in the plane: conformal invariance, Cardy’s formula, scaling limits*]{}, Comptes Rendus de l’Académie des Sciences, Series I - Mathematics, [**333**]{} (2001) 239–244.
J. Asikainen, A. Aharony, B.B. Mandelbrot, E.M. Rauch, J.-P. Hovi, [*Fractal geometry of critical Potts clusters*]{}, Euro. Phys. J. [**B34**]{} (2003) 479–487, .
W. Janke, A.M.J. Schakel, [*Fractal structure of spin clusters and domain walls in the two-dimensional Ising model*]{}, Phys. Rev. [**E71**]{} (2005) 036703, .
W. Janke, A.M.J. Schakel, [*Two-dimensional critical Potts and its tricritical shadow*]{}, Braz. J. Phys. [**36**]{} (2006) 708–716, .
Y. Saint-Aubin, P.A. Pearce, J. Rasmussen, [*Geometric exponents, SLE and logarithmic minimal models*]{}, J. Stat. Mech. (2009) P02028, .
B.B. Mandelbrot, [*The Fractal Geometry of Nature*]{}, W.H. Freeman and Co. (1982).
G.F. Lawler, O. Schramm, W. Werner, [*The dimension of the Brownian frontier is $4/3$*]{}, Math. Rev. Lett. [**8**]{} (2001) 13–24, .
V. Beffara, [*Hausdorff dimensions for $\mbox{SLE}_6$*]{}, Ann. Prob. [**32**]{} (2004) 2606–2629, .
P.A. Pearce, J. Rasmussen, [*Solvable critical dense polymers*]{}, J. Stat. Mech. (2007) P02015, .
P.A. Pearce, J. Rasmussen, S.P. Villani, [*Solvable dense polymers on the cylinder*]{}, J. Stat. Mech. (2010) P02010, .
A. Morin-Duchesne, [*A proof of selection rules for critical dense polymers*]{}, J. Phys. A: Math. Theor. [**44**]{} (2011) 495003, .
P.A. Pearce, J. Rasmussen, S.P. Villani, [*Infinitely extended Kac table of solvable critical dense polymers*]{}, J. Phys. A: Math. Theor. [**46**]{} (2013) 175202, .
A. Morin-Duchesne, P.A. Pearce, J. Rasmussen, [*Modular invariant partition function of critical dense polymers*]{}, Nucl. Phys. [**B874**]{} (2013) 312–357, .
P.A. Pearce, J. Rasmussen, [*Coset graphs in bulk and boundary logarithmic minimal models*]{}, Nucl. Phys. [**B846**]{} (2011) 616–649, .
A. Morin-Duchesne, P.A. Pearce, J. Rasmussen, [*Fusion hierarchies, T-systems and Y-systems of logarithmic minimal models*]{}, J. Stat. Mech. (2014) P05012, .
P.A. Pearce, E. Tartaglia, R. Couvreur, [*Kac boundary conditions of the logarithmic minimal models*]{}, J. Stat. Mech. (2015) P01018, .
A. Morin-Duchesne, J. Rasmussen, D. Ridout, [*Boundary algebras and Kac modules for logarithmic minimal models*]{}, Nucl. Phys. [**B899**]{} (2015) 677–769, .
V.V. Bazhanov, N. Reshetikhin, [*Critical RSOS models and conformal field theory*]{}, Int. J. Mod. Phys. [**A4**]{} (1989) 115–142.
A.B. Zamolodchikov, [*On the thermodynamic Bethe ansatz equations for reflectionless ADE scattering theories*]{}, Phys. Lett. [**B253**]{} (1991) 391–394.
A.B. Zamolodchikov, [*Thermodynamic Bethe ansatz for RSOS scattering theories*]{}, Nucl. Phys. [**B358**]{} (1991) 497–523.
P.A. Pearce, A. Klümper, [*Finite-size corrections and scaling dimensions of solvable lattice models: an analytic method*]{}, Phys. Rev. Lett. [**66**]{} (1991) 974–977.
A. Klümper, P.A. Pearce, [*Conformal weights of RSOS lattice models and their fusion hierarchies*]{}, Physica [**A183**]{} (1992) 304–350.
A. Kuniba, T. Nakanishi, J. Suzuki, [*Functional relations in solvable lattice models I: functional relations and representation theory*]{}, Int. J. Mod. Phys. [**A9**]{} (1994) 5215–5266, .
A. Kuniba, T. Nakanishi, J. Suzuki, [*Functional relations in solvable lattice models II*]{}, Int. J. Mod. Phys. [**A9**]{} (1994) 5267–5312, .
A. Kuniba, T. Nakanishi, J. Suzuki, [*$T$-systems and $Y$-systems in integrable systems*]{}, J. Phys. A: Math. Theor. [**44**]{} (2011) 103001, .
I. Krichever, O. Lipan, P. Wiegmann, A. Zabrodin, [*Quantum integrable models and discrete classical Hirota equations*]{}, Commun. Math. Phys. [**188**]{} (1997) 267–304, .
C.N. Yang, C.P. Yang, [*Thermodynamics of a one-dimensional system of bosons with repulsive delta-function interaction*]{}, J. Math. Phys. [**10**]{} (1969) 1115.
M. Takahashi, [*One-dimensional Heisenberg model at finite temperature*]{}, Prog. Theor. Phys. [**46**]{} (1971) 401.
M. Gaudin, [*Thermodynamics of the Heisenberg-Ising ring for $\Delta{\geqslant}1$*]{}, Phys. Rev. Lett. [**26**]{} (1971) 1301.
A.B. Zamolodchikov, [*Thermodynamic Bethe ansatz in relativistic models: Scaling $3$-state Potts and Lee-Yang models*]{}, Nucl. Phys. [**B342**]{} (1990) 695–720.
A.B. Zamolodchikov, [*On the thermodynamic Bethe ansatz equations for ADE reflectionless scattering theories*]{}, Phys. Lett. [**B253**]{} 391–394; [*Thermodynamic Bethe ansatz for RSOS scattering theories*]{}, Nucl. Phys. **B358** (1991) 497–523; [*From tricritical Ising to critical Ising by thermodynamic Bethe ansatz*]{}, Nucl. Phys. **B358** (1991) 524–546; [*TBA equations for integrable perturbed $SU(2)_k\times SU(2)_1/SU(2)_{k+1}$ coset models*]{}, Nucl. Phys. **B366** (1991) 122–132.
C. Chui, C. Mercat, P.A. Pearce, [*Integrable boundaries and universal TBA functional equations*]{}, Prog. Math. Phys. [**23**]{} (2002) 391–413, .
E.K. Sklyanin, [*Boundary conditions for integrable quantum systems*]{}, J. Phys. [**A21**]{} (1988) 2375–2389.
R.E. Behrend, P.A. Pearce, D.L. O’Brien, [*Interaction-round-a-face models with fixed boundary conditions: The ABF fusion hierarchy*]{}, J. Stat. Phys. [**84**]{} (1996) 1–48, .
A. Klümper, P.A. Pearce, [*Analytic calculation of scaling dimensions: Tricritical hard squares and critical hard hexagons*]{}, J. Stat. Phys. [**64**]{} (1991) 13–76.
D.L. O’Brien, P.A. Pearce, S.O. Warnaar, [*Analytic calculation of conformal partition functions: Tricritical hard squares with fixed boundaries*]{}, Nucl. Phys. [**B501**]{} (1997) 773–799.
Z. Bajnok, O. el Deeb, P.A. Pearce,Ê [*Finite-volume spectra of the Lee-Yang model*]{},Ê JHEP [**04**]{} (2015) 73.
A. Kuniba, K. Sakai, J. Suzuki, [*Continued fraction TBA and functional relations in XXZ model at root of unity*]{}, Nucl. Phys. [**B525**]{} (1998) 597–626, .
D.L. O’Brien, P.A. Pearce, S.O. Warnaar, [*Finitized conformal spectrum of the Ising model on the cylinder and torus*]{}, Physica A [**228**]{} (1996) 63–77.
M.T. Batchelor, J. de Gier, B. Nienhuis, [*The quantum symmetric XXZ chain at $\Delta=-1/2$, alternating sign matrices and plane partitions*]{}, J. Phys. A: Math. Gen. [**34**]{} (2001) L265, .
S. Mitra, B. Nienhuis, J. de Gier, M.T. Batchelor, [*Exact expressions for correlations in the ground state of the dense $O(1)$ loop model*]{}, J. Stat. Mech. (2004) P09010, .
V. Jones, [*Index for subfactors*]{}, Invent. Math. [**72**]{} (1983) 1–25.
P. Martin, [*Potts models and related problems in statistical mechanics*]{}, World Scientific, Singapore (1991).
F. Goodman, H. Wenzl, [*The Temperley-Lieb algebra at roots of unity*]{}, Pacific J. Math. [**161**]{} (1993) 307–334.
B. Westbury, [*The representation theory of the Temperley-Lieb algebras*]{}, Math. Zeit. [**219**]{} (1995) 539–565.
D. Ridout, Y. Saint-Aubin, [*Standard modules, induction and the structure of the Temperley-Lieb algebra*]{}, Adv. Theo. Math. Phys. [**18**]{} (2014) 957–1041, .
J.F. Richard, J.L. Jacobsen, [*Character decomposition of Potts model partition functions. I. Cyclic geometry*]{}, Nucl. Phys. [**B750**]{} (2006) 250–264, .
D. Levy, [*Algebraic structure of translation-invariant spin-$\frac12$ XXZ and $q$-Potts quantum chains*]{}, Phys. Rev. Lett. [**67**]{} (1991) 1971–1974.
P. Martin, H. Saleur, [*On an algebraic approach to higher dimensional statistical mechanics*]{}, Commun. Math. Phys. [**158**]{} (1993) 155–190, .
J.J. Graham, G.I. Lehrer, [*The representation theory of affine Temperley-Lieb algebras*]{}, Enseign. Math. [**44**]{} (1998) 173–218.
R.M. Green, [*On representations of affine Temperley-Lieb algebras*]{}, CMS Conf. Proc. [**24**]{} (1998) 245–261.
K. Erdmann, R.M. Green, [*On representations of affine Temperley-Lieb algebras, II*]{}, Pac. J. Math. [**191**]{} (1999) 243–274, .
A.V. Razumov, Y.G. Stroganov, [*Spin chains and combinatorics*]{}, J. Phys. A: Math. Theor. [**34**]{} (2001) 3185–3190, .
A.V. Razumov, Y.G. Stroganov, [*Spin chains and combinatorics: twisted boundary conditions*]{}, J. Phys. A: Math. Theor. [**34**]{} (2001) 5335–5340, .
Y.G. Stroganov, [*The importance of being odd*]{}, J. Phys. A [**34**]{} (2001) L179–L185, .
J.F. Richard, J.L. Jacobsen, [*Eigenvalue amplitudes of the Potts model on a torus*]{}, Nucl. Phys. [**B769**]{} (2007) 256–274, .
B. Aufgebauer, A. Klümper, [*Quantum spin chains of Temperley-Lieb type: periodic boundary conditions, spectral multiplicities and finite temperature*]{}, J. Stat. Mech. (2010) P05018, .
H. Saleur, M. Bauer, [*On some relations between local height probabilities and conformal invariance*]{}, Nucl. Phys. [**B320**]{} (1989) 591–624.
J. Cardy, [*The $O(n)$ model on the annulus*]{}, J. Stat. Phys. [**125**]{} (2006) .
M.R. Gaberdiel. I. Runkel, S. Wood, [*A modular invariant bulk theory for the $c=0$ triplet model*]{}, J. Phys. A [**44**]{} (2010) .
A. Morin-Duchesne, Y. Saint-Aubin, [*A homomorphism between link and XXZ modules over the periodic Temperley-Lieb algebra*]{}, J. Phys. A: Math. Theor. [**46**]{} (2013) 285207, .
F.C. Alcaraz, M. Baake, U. Grimm, V. Rittenberg, [*Operator content of the XXZ chain*]{}, J. Phys. A: Math. Gen. [**21**]{} (1988) L117–120.
F.C. Alcaraz, U. Grimm, V. Rittenberg,Ê [*The XXZ Heisenberg chain, conformal invariance and the operator content of $c<1$ systems*]{},Ê Nucl. Phys. [**B316**]{} (1989) 735–768.
N. Read, H. Saleur, [*Associative-algebraic approach to logarithmic conformal field theories*]{}, Nucl. Phys. [**B777**]{} (2007) 316–351, .
M.R. Gaberdiel. I. Runkel, [*From boundary to bulk in logarithmic CFT*]{}, J. Phys. A: Math. Theor. [**41**]{} (2008) 075402, .
I. Runkel, M.R. Gaberdiel, S. Wood, [*Logarithmic bulk and boundary conformal field theory and the full centre construction*]{}, in Bai [*et al.*]{} (eds) Conformal Field Theories and Tensor Categories, Math. Lect. from Peking Univ., Springer (2014), .
A.M. Gainutdinov, N. Read, H. Saleur, R. Vasseur, [*The periodic $s\ell(2|1)$ alternating spin chain and its continuum limit as a bulk logarithmic conformal field theory at $c=0$*]{}, J. High Energ. Phys. [**114**]{} (2015), .
J.L. Jacobsen, H. Saleur, [*Conformal boundary loop models*]{}, Nucl. Phys. [B788]{} (2008) 137–166, .
J. Dubail, J.L. Jacobsen, H. Saleur, [*Conformal two-boundary loop model on the annulus*]{}, Nucl. Phys. [**B813**]{} (2009) 430–459, .
J. Riordan, [*Combinatorial identities*]{}, Robert E. Krieger Pub. Co (1979).
[^1]: We note that the Razumov-Stroganov eigenvalue does not fit in any of the two subcases. The analysis presented here holds for all the other eigenvalues.
[^2]: Note that for $N=14,15$, our program can only produce the pattern diagrams for $d {\geqslant}8$. For $d<8$, the corresponding pattern diagrams given in \[fig:GSpatterns\] were inferred from our understanding of the patterns for smaller system sizes.
[^3]: We use the letter $M$ for both the number of sites of a column diagram and the vertical width of the lattice. It should be clear from the context which one is referred to, and likewise in \[sec:fsgf\].
[^4]: Here $M$ is the number of sites of the column diagram in $\Aone{M}{m}$, not the vertical width of the $M\times N$ lattice.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study the notions of weak rational ergodicity and rational weak mixing as defined by Jon Aaronson. We prove that various families of infinite measure-preserving rank-one transformations possess (or do not posses) these properties, and consider their relation to other notions of mixing in infinite measure.'
address:
- |
Harvard College\
University Hall\
Cambridge, MA 02138, USA
- |
University of Minnesota\
Minneapolis, MN 55455-0213, USA
- 'University of California, Los Angeles, CA 90095-1555, US '
- |
Department of Mathematics\
Williams College\
Williamstown, MA 01267, USA
author:
- Irving Dai
- Xavier Garcia
- Tudor Pădurariu
- 'Cesar E. Silva'
bibliography:
- 'References.bib'
title: 'On Rationally Ergodic and Rationally Weakly Mixing Rank-One Transformations'
---
Definitions and Preliminaries
=============================
Let $(X,\mathcal{B},\mu)$ be a standard Borel measure space with a $\sigma$-finite nonatomic measure $\mu$. In most cases, we will assume that $\mu$ is infinite. A transformation $T: X \rightarrow X$ is [**measurable**]{} if $T^{-1}A\in\mathcal B$ for all $A\in\mathcal B$. A measurable transformation $T$ is [**measure-preserving**]{} if $\mu(A)=\mu(T^{-1}A)$ for all $A\in\mathcal B$. We say that $T$ is [**ergodic**]{} if every $T$-invariant set (i.e, $T^{-1}A=A $ mod $\mu$) is null ($\mu(A)=0$) or full $(\mu(X\setminus A)=0$). We say that $T$ is [**conservative**]{} if for every measurable set $A$ of positive measure, there exists a positive integer $n$ such that $\mu(A \cap T^{-n}A)>0$. It follows that $T$ is conservative and ergodic if and only if for every set $A$ of positive measure, $\bigcup_{n=0}^\infty T^{-n}A=X$ mod $\mu$. An [**invertible measurable transformation**]{} is a measurable transformation whose inverse is also measurable. Throughout this paper, we will assume that $T$ is an invertible, conservative ergodic, measure-preserving transformation on $(X,\mathcal{B},\mu)$, and we will typically use the forward images $T^nA$ instead of $T^{-n}A$.\
\
When $T$ is a measure-preserving transformation on a probability space $X$, the Birkhoff ergodic theorem states that ergodicity is equivalent to having the convergence $$\begin{aligned}
\label{birk}
\dfrac{1}{n} \displaystyle \sum_{k = 0}^{n-1} \mu(A \cap T^k B) \rightarrow \mu(A)\mu(B)\end{aligned}$$ for all measurable $A, B \subset X$. This gives a quantitative estimate for the average number of visits of one set to another. When $X$ has infinite measure, however, the Birkhoff ergodic theorem implies that the Cesaro averages of converge to $0$ for all pairs $A, B$ of finite measure. Moreover, in [@Aa77] Aaronson proved that there exists no sequence of normalizing constants for which the averages of converge to $\mu(A)\mu(B)$, and he proposed in turn the definitions of rational ergodicity and weak rational ergodicity.\
For any measurable set $F \subset X$ of finite positive measure, define the $\mathbf{intrinsic}$ $\mathbf{weight}$ $\mathbf{sequence}$ of $F$ to be
$$u_n(F) = \dfrac{\mu(F \cap T^nF)}{\mu(F)^2}$$ and write
$$a_n(F) = \displaystyle \sum_{k = 0}^{n-1} u_k(F).$$
A transformation $T$ is said to be [**weakly rationally ergodic**]{} (see [@Aa77]) if there exists a measurable set $F \subset X$ of positive finite measure such that for all measurable $A$, $B$ $\subset F$, we have
$$\label{e:raterg}
\dfrac{1}{a_n(F)} \displaystyle \sum_{k = 0}^{n-1} \mu(A \cap T^kB) \rightarrow \mu(A)\mu(B)$$
as $n \rightarrow \infty$. If this convergence happens only along a subsequence $\{n_i\}$ of $\mathbb{N}$, we say that $T$ is $\mathbf{subsequence}$ $\mathbf{weakly}$ $\mathbf{rationally}$ $\mathbf{ergodic}$. To emphasize the set $F$, we will sometimes say $T$ is $\mathbf{weakly}$ $\mathbf{rationally}$ $\mathbf{ergodic}$ $\mathbf{on \ F}$. Note that any measure-preserving ergodic transformation on a probability space is trivially weakly rationally ergodic, by taking $F$ to be the whole space itself. Then $a_n(F) = n$, so reduces to the Cesaro sum definition of ergodicity.\
A transformation $T$ is said to be $\mathbf{(spectrally)}$ $\mathbf{weakly}$ $\mathbf{mixing}$ if whenever $f\in L^\infty(X,\mu)$ and $f\circ T= z f $ for some $z\in \mathbb{C}$, then $f$ is constant a.e. When $X$ is a probability space, this is equivalent to ergodicity of the Cartesian square and also to the strong Cesaro convergence $$\dfrac{1}{n} \displaystyle \sum_{k = 0}^{n-1} |\mu(A \cap T^k B) - \mu(A)\mu(B)| \rightarrow 0$$ for all measurable $A, B \subset X$. In [@AaLiWe79], it was shown that for infinite measure-preserving transformations, (spectral) weak mixing is strictly weaker than ergodicity of the Cartesian square.\
\
Another property we consider that is equivalent to weak mixing in the finite measure-preserving case is double ergodicity. This property was introduced by Furstenberg in [@Fu81] and was shown to be equivalent to weak mixing for probability-preserving transformations, but was not given a specific name. A transformation $T$ is said to be [**doubly ergodic**]{} if for every pair of sets $A$ and $B$ with positive measure, there exists a positive integer $n$ for which $\mu(A \cap T^nA)$ and $\mu(B \cap T^nA)$ are simultaneously nonzero. In the infinite measure-preserving case, double ergodicity is strictly stronger than spectral weak mixing and is properly implied by ergodic Cartesian square [@BoFiMaSi01].\
\
More recently, Aaronson introduced another notion of weak mixing for infinite measure that generalizes rational ergodicity. A transformation $T$ is said be [**rationally weakly mixing**]{} (see [@Aa12]) if there exists a measurable set $F \subset X$ of positive finite measure such that for all measurable $A$, $B$ $\subset F$, we have $$\label{e:rwm}
\dfrac{1}{a_n(F)} \displaystyle \sum_{k = 0}^{n-1} |\mu(A \cap T^kB) - \mu(A)\mu(B)u_k(F)| \rightarrow 0$$
as $n \rightarrow \infty$. Again, it is clear that rational weak mixing reduces to the usual definition of weak mixing in the finite measure-preserving case.\
\
We now describe our main results. In Section \[S:ratergodicity\] we prove that a large class of rank-one transformations are weakly rationally ergodic and discuss the notions of rational ergodicity and bounded rational ergodicity in this context. In Section \[S:ratweakmix\] we construct a class of rank-one transformations that are not rationally weakly mixing; in particular, we obtain a transformation which is rationally ergodic and spectrally weakly mixing but not rationally weakly mixing. This negatively answers a question of Aaronson’s. (After this work was completed, we learned that Aaronson had also independently answered this question [@Aa12b].) Section \[S:doubleerg\] shows that rational weak mixing implies double ergodicity and constructs a transformation that is not rationally weakly mixing and which we conjecture to be doubly ergodic. Section \[S:zerotype\] proves that the notion of zero-type for infinite measure-preserving transformations (whose spectral definition is similar to the mixing condition in the case of probability-preserving transformations) is independent of rational weak mixing. Finally, in Section \[S:ratweakmixex\] we present a class of rank-one transformations that are rationally weakly mixing. As remarked in [@Aa12], all the examples of rationally weakly mixing transformations constructed in [@Aa12] are of the type $T\times S$, where $T$ is an infinite measure-preserving $K$-automorphism and $S$ is a mildly mixing probability-preserving transformation. These examples have countable Lebesgue spectrum and are of a different nature than our rank-one constructions.
Acknowledgements
----------------
This paper was based on research done by the Ergodic Theory group of the 2012 SMALL Undergraduate Research Project at Williams College. Support for this project was provided by the National Science Foundation REU Grant DMS - 0353634 and the Bronfman Science Center of Williams College. We are indebted to Jon Aaronson for conversations and suggestions during discussions of our work at the 2012 Williams Ergodic Theory Conference. We would also like to acknowledge the other members of the 2012 Ergodic Theory group: Shelby Heinecke, Emily Wickstrom, and Evangelie Zachos. We would like to thank the referee for comments that improved the paper.
Rank-One Transformations (Basics)
---------------------------------
We briefly review (rank-one) cutting-and-stacking transformations (see e.g. [@Si08]). A $\mathbf{column}$ is an ordered collection of pairwise disjoint intervals (called $\mathbf{levels}$) in $\mathbb{R}$, each of the same measure. We think of the levels in a column as being stacked on top of each other, so that the $(j+1)$-st level is directly above the $j$-th level. Every column $C = \{J_j\}$ is associated with a natural column map $T_C$ sending each point in $J_j$ to the point directly above it in $J_{j+1}$. (Note that $T_C$ is undefined on the top level of $C$.) A $(\mathbf{rank}$-$\mathbf{one})$ $\mathbf{cutting}$-$\mathbf{and}$-$\mathbf{stacking}$ construction for $T$ consists of a sequence of columns $C_n$ such that:
(a) The first column $C_0$ is the unit interval.
(b) Each column $C_{n+1}$ is obtained from $C_n$ by cutting $C_n$ into $r_n\geq 2$ subcolumns of equal width, adding any number of new levels (called $\mathbf{spacers}$) above each subcolumn, and stacking every subcolumn under the subcolumn to its right. In this way, $C_{n+1}$ consists of $r_n$ copies of $C_n$, possibly separated by spacers.
(c) The collection of levels $\displaystyle \bigcup_n C_n$ forms a generating semiring for $\mathcal{B}$.
Observing that $T_{C_{n+1}}$ agrees with $T_{C_n}$ everywhere where $T_{C_n}$ is defined, we then take $T$ to be the limit of $T_{C_n}$ as $n \rightarrow \infty$.
Rank-One Transformations (Notation)
-----------------------------------
Let $T$ be a rank-one transformation, and fix any column $C_n$ of $T$. We denote the number of levels in $C_n$ by $h_n$ and write $w_n$ for the width of each level. We denote the height of any level $J$ in $C_n$ by $h(J)$, with the convention that $0 \leq h(J) < h_n$. For each $0 \leq k < r_n$, let $s_{n,k}$ be the number of spacers added above the $k$-th subcolumn of $C_n$, and denote the number of levels in the $k$-th subcolumn (after adding spacers) by $h_{n, k} = h_n + s_{n, k}$.\
\
Define $T$ to be $\mathbf{normal}$ if $s_{n, r_n-1} > 0$ for infinitely many values of $n$. (This means that at least one spacer is added above the rightmost subcolumn infinitely many times.) In addition, we say that $T$ has a $\mathbf{bounded \ number \ of}$ $\mathbf{cuts}$ if $\sup \{r_n\} < \infty$; this implies that $T$ is partially rigid and of infinite conservative index [@AdFrSi97].\
\
Given any level $J$ from $C_n$ and any column $C_m$ of $T$ with $m \geq n$, we define the $\mathbf{descendants}$ of $J$ in $C_m$ to be the collection of levels in $C_m$ whose disjoint union is $J$. We denote this set by $D(J, m)$. Occasionally, we will also use $D(J, m)$ to refer to the heights of the descendants of $J$ in $C_m$. In the case when $J$ is the unit interval $I$, for each $m \in \mathbb{N}$ it will be convenient to define $M_m = \max(D(I, m))$. (That is, $M_m$ is the height of the uppermost descendant of $I$ in $C_m$.)\
\
We say that $T$ $\mathbf{grows \ exponentially}$ if $2s_{n,r_n-1} \geq h_{n+1}$ for every $n$. Intuitively, this means that the upper half of every column $C_n$ consists of spacers added during the $(n-1)$-st stage of construction. In particular, the descendants of any level $J$ from an earlier column must lie in the lower half of $C_n$. Note that any $T$ which grows exponentially is clearly normal.
Rational Ergodicity {#S:ratergodicity}
===================
In this section, we establish some introductory ideas and prove that a large class of rank-one transformations are rationally ergodic.\
\
We begin with a computational lemma. Suppose that $T$ is a normal rank-one transformation. Then we claim that the partial sums $a_n(J)$ for any level $J$ can be computed from the descendant heights $D(J,N)$ for $N$ sufficiently large. More precisely,
\[L:lem2.1\] Let $T$ be a normal rank-one transformation. Fix any level $J$ and $n \in \mathbb{N}$. Then for every $N$ sufficiently large, we have $$\mu(J \cap T^kJ) = w_N\cdot |D(J, N) \cap (k + D(J, N))|$$ for all $0 \leq k < n$. Consequently, $$\displaystyle \sum_{k = 0}^{n-1} \mu(J \cap T^kJ) = w_N\cdot \sum_{k = 0}^{n-1} |D(J, N) \cap (k + D(J, N))|.$$
Fix any level $J$, and let $n \in \mathbb{N}$ be arbitrary. Since $T$ is normal, we can find some column $C_N$ in which all the heights $D(J, N)$ are at most $h_N - n$. For any $0 \leq k < n$ and level $J_i \in D(J, N)$, the image $T^k(J_i)$ is then the level in $C_N$ of height $h(J_i) + k$. The conclusion follows immediately.
We will sometimes need to compute $\mu(J \cap T^kJ)$ for $k < 0$. For this, simply observe that $$\mu(J \cap T^kJ) = \mu(T^{-k}J \cap J)$$ and $$|D(J, N) \cap (k + D(J, N))| = |(-k + D(J, N)) \cap D(J, N)|,$$ so in fact Lemma \[L:lem2.1\] holds for all $-n < k < n$.\
\
We thus calculate $D(J, N)$. Suppose that $J$ is a level in $C_j$ of height $h(J)$. Then $J$ splits into $r_j$ levels in $C_{j+1}$ of heights $$\{h(J)\} \cup \{h(J) + \displaystyle \sum_{k = 0}^i h_{j, k}: 0 \leq i < r_j - 1\}.$$ Letting $$H_j = \{0\} \cup \left\{\sum_{k = 0}^i h_{j, k} : 0 \leq i < r_j - 1\right\},$$ it follows inductively that $$D(J, N) = h(J) + H_j \oplus H_{j+1} \oplus \cdots \oplus H_{N-1}.$$
\
We now show that every normal rank-one transformation satisfies condition for $A$, $B$ finite unions of levels and $F$ the unit interval. In this context, we note that Aaronson [@Aa77 Theorem 6.1] has shown every set of finite measure $F$ contains a dense algebra of sets satisfying , but at the same time it is never true that is satisfied for all measurable sets in every set $F$ of finite positive measure [@Aa77 Theorem 6.2].
\[T:normal\] Let $T$ be a normal rank-one transformation. Then $T$ satisfies condition for $A$, $B$ finite unions of levels and $F$ the unit interval.
Let $F = I$ denote the unit interval. We begin by proving for $A = B = J$, where $J$ is the bottom level of any column $C_j$. We need to show that $$\dfrac{1}{a_n(I)} \displaystyle \sum_{k = 0}^{n-1} \mu(J \cap T^{k}J) \rightarrow \mu(J)^2$$ as $n \rightarrow \infty$. For $N$ sufficiently large (as a function of $n$), we have $$\displaystyle \sum_{k = 0}^{n-1} \mu(J \cap T^{k}J) = w_N \left( \sum\limits_{k = 0}^{n-1}|D(J,N)\cap (k + D(J,N))| \right)$$ by Lemma \[L:lem2.1\]. Now, writing $$D(I, N) = H_0 \oplus H_1 \oplus \cdots \oplus H_{N-1}$$ and $$D(J, N) = H_j \oplus H_{j+1} \oplus \cdots \oplus H_{N-1},$$ we may express $D(I, N) = A \oplus B$ and $D(J, N) = B$ with $A = H_0 \oplus H_1 \oplus \cdots \oplus H_{j-1}$. Noting that $\mu(J) = 1/|D(I, j)| = 1/|A|$, we thus wish to show $$\displaystyle \dfrac{w_N}{a_n(I)} \left(|A|^2\sum\limits_{k = 0}^{n-1}|B \cap (k + B)|\right) \rightarrow 1.$$ We give the term inside the parentheses a combinatorial interpretation. Let $P(n)$ denote the number of ordered quadruplets $(a, a', b, b')$ with $a, a' \in A$ and $b, b' \in B$ for which $0 \leq b - b' < n$. Then the above quotient is precisely $w_NP(n) / a_n(I)$, since $|B \cap (k + B)|$ counts the number of pairs $b, b' \in B$ with $k = b - b'$.\
\
Now let $M$ be the maximum value of $A - A$. We claim that the following inequality holds: $$\displaystyle \sum_{k = M}^{n - 1 - M} |(A \oplus B) \cap (k + A \oplus B)| \leq P(n) \leq \sum_{k = -M}^{n - 1 + M} |(A \oplus B) \cap (k + A \oplus B)|.$$ Indeed, the sum on the left counts the number of quadruplets $(a, a', b, b')$ with $M \leq a - a' + b - b' < n - M$; the sum on the right counts the number of quadruplets with $-M \leq a - a' + b - b' < n + M$. Clearly, any quadruplet with $M \leq a - a' + b - b' < n - M$ has $0 \leq b - b' < n$. Similarly, any quadruplet with $0 \leq b - b' < n$ has $-M \leq a - a' + b - b' < n + M$. Recalling that $A \oplus B = D(I, N)$, it thus follows that $$\dfrac{1}{a_n(I)} \sum_{k = M}^{n - 1 - M} \mu(I \cap T^kI) \leq \dfrac{w_N}{a_n(I)} (P(n)) \leq \dfrac{1}{a_n(I)} \sum_{k = -M}^{n - 1 + M} \mu(I \cap T^kI).$$ Now, $M$ is a fixed constant, independent of $n$. Furthermore, the sequence $\mu(I \cap T^k(I))$ is bounded above by 1 but has divergent sum. Hence both sides of the above inequality tend to 1 as $n \rightarrow \infty$, showing that $w_NP(n) / a_n(I) \rightarrow 1$, as desired. This proves for $A = B = J$ , where $J$ is the bottom level of any column.\
\
We now prove for $J$ and $J'$ any two levels in the same column. By applying $T^{-1}$ and using the fact that $T$ is measure-preservin, we may assume that one of the two levels (say $J$) is actually the bottom level of the column. Letting $J' = T^d(J)$ for some $d$, we wish to show $$\dfrac{1}{a_n(I)} \sum_{k = 0}^{n-1} \mu(J \cap T^{k + d}J) \rightarrow \mu(J)^2.$$ Now, we have from before that $$\dfrac{1}{a_n(I)} \sum_{k = 0}^{n-1} \mu(J \cap T^{k}J) \rightarrow \mu(J)^2.$$ Since $\mu(J \cap T^{k}J)$ is bounded and $a_n(I) \rightarrow \infty$, the conclusion follows immediately.\
\
Finally, we extend to finite unions of levels. Without loss of generality, we may assume that $J$ and $J'$ are both disjoint unions of images of the same level $K$. The desired statement then follows from summing together the limits for each pair of images.
We now show that under certain conditions, we can extend the results of Theorem \[T:normal\] to all sets $A$ and $B$ (thus proving weak rational ergodicity).
\[T:wraterg\] Let $T$ be an exponentially growing rank-one transformation with a bounded number of cuts. Then $T$ is weakly rationally ergodic.
We show that for $T$ satisfying the above hypotheses, it suffices to prove for finite unions of levels (as in Theorem \[T:normal\]). Indeed, given arbitrary measurable sets $A, B \subset I$, choose $D \subset I$ a finite union of levels for which $\mu(D \triangle B) < \varepsilon$. We claim that there is some constant $c$ such that $$\left| \dfrac{1}{a_n(I)} \sum\limits_{k = 0}^{n-1}\mu(A\cap T^kB) - \dfrac{1}{a_n(I)}\sum_{k = 0}^{n-1}\mu(A\cap T^kD)\right| \leq c\varepsilon$$\
for every $n$. Indeed, let $B_0 = B \cap D$, and write $B = B_0 \cup B_1$ and $D = B_0 \cup D_1$. Then the above difference reduces to $$\left| \dfrac{1}{a_n(I)} \sum\limits_{k = 0}^{n-1}\mu(A\cap T^kB_1) - \dfrac{1}{a_n(I)}\sum_{k = 0}^{n-1}\mu(A\cap T^kD_1)\right|.$$ Now, we claim that we can bound $$\displaystyle \dfrac{1}{a_n(I)} \sum\limits_{k = 0}^{n-1}\mu(A\cap T^kB_1) \leq c \mu(B_1)$$ for some $c$ independent of $n, A,$ and $B_1$. Applying this bound with $D_1$ in place of $B_1$ will bound (4) by $c(\mu(B_1)+\mu(D_1)) \leq 2c\varepsilon$, as desired.\
\
Recall that $M_m = \max(D(I, m))$, for $m\in\mathbb{N}$. Clearly, $\{M_m\}$ is an increasing sequence. For any fixed $n$, if we choose $m$ such that $M_{m-1} \leq n-1 < M_m$, we have $$\displaystyle \sum_{k = 0}^{n-1} \mu(A \cap T^kB_1) \leq \sum_{k = 0}^{n-1} \mu(I \cap T^kB_1) \leq \sum_{k = 0}^{M_m} \mu(I \cap T^kB_1)$$ and $$\displaystyle \sum_{k = 0}^{M_{m-1}} \mu(I \cap T^kI) \leq \sum_{k = 0}^{n-1} \mu(I \cap T^kI) = a_n(I).$$ To prove (5), it thus suffices to find some $c$ such that $$\sum_{k = 0}^{M_m} \mu(I \cap T^kB_1) \leq c\mu(B_1) \sum_{k = 0}^{M_{m-1}} \mu(I \cap T^kI)$$ for every $m$.\
\
Now observe that the sets $T^kI$ with $-M_m \leq k \leq M_m$ cover each point of $I$ exactly $|D(I, m)|$ times. Indeed, consider the column $C_m$ and fix any $x \in I$. Let $x$ be contained in $J$, where $J$ is some level from $D(I, m)$. For any level $J'$ in $D(I, m)$, we claim that there is exactly one value of $k$ between $-M_m$ and $M_m$ for which $T^kJ' \cap J \neq \emptyset$. Indeed, suppose $0 \leq k \leq M_m$ and $T^kJ' \cap J \neq \emptyset$. Any forward image $T^kJ'$ with $0 \leq k \leq M_m$ is just a translation upwards by $k$ levels, since $h_m \geq 2M_m$. (This is implied by our hypothesis that $T$ is exponentially growing.) Hence in this case $k$ must equal $h(J) - h(J')$. On the other hand, suppose $-M_m \leq k < 0$ and $T^kJ' \cap J \neq \emptyset$. Then $J' \cap T^{-k}J \neq \emptyset$, and exactly the same argument shows that $-k = h(J') - h(J)$ (i.e., $k = h(J) - h(J')$). The claim is then immediate.\
\
We thus have $$\sum_{k = -M_m}^{M_m} \mu(I \cap T^kB_1) = \sum_{k = -M_m}^{M_m} \mu(T^kI \cap B_1) = |D(I, m)| \mu(B_1)$$ and $$\sum_{k = -M_{m-1}}^{M_{m-1}} \mu(I \cap T^kI) = |D(I, m-1)|.$$ Hence $$\sum_{k = -M_m}^{M_m} \mu(I \cap T^kB_1) = \left(\dfrac{|D(I,m)|}{|D(I,m-1)|}\right)\mu(B_1)\left(\sum_{k = -M_{m-1}}^{M_{m-1}}\mu(I \cap T^kI)\right)$$ and so $$\sum_{k = 0}^{M_m} \mu(I \cap T^kB_1) \leq \left(\dfrac{|D(I,m)|}{|D(I,m-1)|}\right)\mu(B_1)\left(2\left(\sum_{k = 0}^{M_{m-1}}\mu(I \cap T^kI)\right) - 1\right).$$\
But $|D(I,m)|/|D(I, m-1)| = r_{m-1}$, and $T$ has a bounded number of cuts. We thus easily obtain (6). Hence (4) holds, and we can approximate $B$ with $D$ a finite union of levels. Applying a similar argument to $A$ shows that it suffices to prove for all $A$, $B$ finite unions of levels, which is the content of Theorem \[T:normal\].
We now consider some alternate notions of rational ergodicity, also due to Aaronson [@Aa77]. For any measurable function $f$, recall the notation $$S_n(f) = \displaystyle \sum_{k=0}^{n-1} f \circ T^k.$$ We say that $T$ is $\mathbf{rationally \ ergodic}$ if there exists a set $F$ of positive finite measure which satisfies a $\mathbf{Renyi}$ $\mathbf{inequality}$; i.e., there is some constant $M$ such that $$\label{e:renyi}
\displaystyle \int_F (S_n(1_F))^2 dm \leq M \left( \int_F S_n(1_F) dm \right)^2$$ for every $n \in \mathbb{N}$. If this inequality holds only on a subset $\{n_i\} \subset \mathbb{N}$, we say that $T$ is $\mathbf{subsequence}$ $\mathbf{rationally}$ $\mathbf{ergodic}$. Some authors adopt this as the definition of rational ergodicity instead (see e.g. [@Ci11]). It was shown in [@Aa77] that rational ergodicity implies weak rational ergodicity. It is not currently known whether these notions are equivalent.\
\
We say that $T$ is $\mathbf{boundedly \ rationally \ ergodic}$ (see [@Aa79]) if there exists a set $F$ of positive finite measure such that $$\displaystyle \sup_{n \geq 1} \left\| \dfrac{1}{a_n(F)} S_n(1_F) \right\|_{\infty} < \infty.$$ In [@Aa79], it was shown that bounded rational ergodicity is a strictly stronger property than rational ergodicity. It is not difficult to see that the proof of Theorem \[T:wraterg\] (in particular, the establishment of (5) for all $B_1$) yields bounded rational ergodicity for the transformations in question. Indeed, set $A = I$ in (5). Then there is a constant $c$ such that for any $n$ and $B_1 \subset I$, $$\int_{B_1} \dfrac{1}{a_n(I)} S_n(1_I) dm = \dfrac{1}{a_n(I)} \sum_{k = 0}^{n-1} \mu(I \cap T^kB_1) \leq c \mu(B_1).$$ This means that the average value of $S_n(1_I)/a_n(I)$ on $B_1$ is bounded above by $c$. Since this holds for every $B_1$, the essential supremum of $S_n(1_I)/a_n(I)$ must also be bounded above by $c$. Hence $T$ is boundedly rationally ergodic.\
\
Aaronson proved in [@Aa79] that every dyadic tower over the adding machine is boundedly rationally ergodic; Theorem \[T:wraterg\] extends this result to a larger class of transformations and uses a different approach. Some interesting examples of exponentially growing rank-one transformations with a bounded number of cuts include:
(a) Hajian-Kakutani skyscraper-type constructions [@HaKa70]: $r_n = 2$, $\{s_{n, 0} = 0, s_{n, 1} \geq 2h_n\}.$
(When $s_{n, 1} = 2h_n+1$ the transformation is spectral weakly mixing, see [@AdFrSi97]).
(b) Chacón-like constructions: $r_n = 3$, $\{s_{n, 0} = 0, s_{n, 1} = 1, s_{n, 2} \geq 3h_n + 1\}$.
(When $s_{n, 2} = 3h_n + 1$ the transformation has infinite ergodic index, see [@AdFrSi97], but is not power weakly mixing, see [@GHPSW03].)
\
We now prove a slightly different version of Theorem \[T:wraterg\] without the hypothesis of a bounded number of cuts but obtain the conclusion only a a subsequence.
\[T:renyi\] Let $T$ be an exponentially growing rank-one transformation. Then $T$ is subsequence rationally ergodic on $F = I = (0, 1)$ along the sequence $\{n_m = M_m + 1\}$.
We verify the Renyi inequality with $n = M_m+1$ and $M = 2$. Let $D(I, m) = \{I_j\}$ be the descendants of $I$ in column $C_m$, and set $N = |D(I, m)|$. Order $\{I_j\}$ by height of appearance in $C_m$ so that $I_1$ is the lowermost level of $\{I_j\}$ in $C_m$ and $I_N$ is the uppermost. Denote the heights of $\{I_j\}$ in $C_m$ by $\{h(I_j)\}$.\
\
Now, $S_n(1_I)(x)$ is equal to the number of $k$ with $0 \leq k \leq n-1$ such that $T^k(x) \in I$. Since $T$ is exponentially growing, this implies that $S_n(1_I)$ is constant on each $I_j$ and that the value of $S_n(1_I)$ on any fixed $I_l$ is the cardinality of the intersection $(h(I_l) + \{0, 1, \cdots, n-1\}) \cap \{h(I_j)\}$. (The relevant forward images $T^kI_l$ are simply upward translations.) On the other hand, it is obvious that $h(I_j) \in (h(I_l) + \{0, 1, \cdots, n-1\})$ exactly when $j \geq l$. Hence $S_n(1_I)$ takes the value $N + 1 - l$ on $I_l$. Restricting the domain of $S_n(1_I)$ to $I$, we thus have $$S_n(1_I) = \displaystyle \sum_{l=1}^{N} (N + 1 - l) 1_{I_l}.$$ Proving (7) is thus equivalent to showing $$\displaystyle \sum_{l = 1}^N (N+1-l)^2w_m \leq 2 \left( \sum_{l=1}^N (N + 1 -l)w_m \right)^2.$$ Now, $w_m = 1/|D(I, m)| = 1/N$. Multiplying through by $N^2$ and reindexing yields the equivalent inequality $$\displaystyle N\left( \sum_{l = 1}^N l^2 \right) \leq 2 \left( \sum_{l=1}^Nl \right)^2.$$ The result then follows from the formulas for power sums.
Rational Weak Mixing {#S:ratweakmix}
====================
In this section, we present a large class of transformations that are $not$ rationally weakly mixing. We obtain as a corollary the existence of transformations which are rationally ergodic and spectrally weakly mixing, but not rationally weakly mixing.\
\
We begin with an example of a rank-one transformation which is subsequence rationally weakly mixing.\
\
First, consider the Chacón rank-one transformation $T$ constructed by starting with the unit interval, cutting each column in half, and adding a single spacer on top of the right subcolumn at every step [@Ch69]. This transformation is finite measure-preserving and weakly mixing; thus, it is rationally weakly mixing. We claim that (in particular) $T$ is rationally weakly mixing on the unit interval $I = (0, 1)$.\
\
It is clear from the definition of weak rational ergodicity that if $T$ is weakly rationally ergodic on $F$, then $T$ is weakly rationally ergodic on any subset of $F$. Moreover, it was shown in [@Aa12] that for $T$ rationally weakly mixing, the class of sets $F$ satisfying is the same as the class of sets $F$ satisfying . This establishes the claim.\
\
Now let $$\phi_n(A, B) = \dfrac{1}{a_n(I)} \displaystyle \sum_{k = 0}^{n-1} |\mu(A \cap T^{k}B) - \mu(A)\mu(B)u_k(I)|$$ be the quotient from , and let $D_m$ denote the collection of dyadic intervals of the form $(i/2^m, (i+1)/2^m)$ for $0 \leq i < 2^m$. Since $D_1$ is a finite collection and $T$ is rationally weakly mixing, there exists some natural number $m_1$ such that for all $A, B \in D_1$ we have $\phi_{m_1}(A ,B) < 1/2$. We claim that in fact this inequality is true for every rank-one transformation $\tilde{T}$ which shares its first $m_1$ stages of construction with $T$ (i.e., $\tilde{C}_n = C_n$ for all $n < m_1$). Indeed, for $A, B \subset I$, the value of $\phi_{m_1}(A ,B)$ depends only on the first $m_1$ stages of the construction of $T$, since the heights $D(I, m_1)$ are all less than $h_{m_1} - m_1$.\
\
We now define our desired transformation. We begin by following the construction of the transformation $T$ as described above, until we reach $C_{m_1}$. Then, at the $m_1$-th iteration, we add $2h_{m_1}$ spacers above the right subcolumn. Now, adding one spacer at each subsequent iteration gives another finite measure-preserving transformation, which is also weakly mixing. Hence, there is some $m_2 > m_1$ such that $\phi_{m_2}(A, B) < 1/4$ for all $A,B \in D_{1} \cup D_{2}$.\
\
We thus continue adding a single spacer at each step until we reach $C_{m_2}$, at which point we add $2h_{m_2}$ spacers. Proceeding inductively in this manner, we obtain a cutting-and-stacking transformation $T$ and a sequence $\{m_i\}$ such that for each $i$, $\phi_{m_i}(A, B) < 1/2^i$ for all $A, B \in D_1 \cup D_2 \cup \cdots \cup D_i$. The result is an invertible, infinite measure-preserving transformation which is rationally weakly mixing along $\{m_i\}$ for dyadic intervals.\
\
In order to extend to all subsets of $I$, we use the following result due to Aaronson [@Aa12]:
Let $T$ be an invertible measure-preserving transformation on a Polish space $X$, and assume that $T$ is rationally ergodic on some open set $F$. Suppose there is a countable base $\mathcal{C}$ for the topology of $F$ such that for every finite subcollection $\{C_i\} \subset \mathcal{C}$, there exists a finite subcollection $\{D_i\} \subset \mathcal{C}$ which is disjoint and has the same union. Then to establish rational weak mixing, it suffices to prove condition for elements of $\mathcal{C}$.
Lemma 3.1 also holds for establishing subsequence rational weak mixing, so long as rational ergodicity is known along the same subsequence. Since the transformation $T$ above may expressed as a dyadic tower over the adding machine, $T$ is rationally ergodic [@Aa79]. It follows from Lemma 3.1 that $T$ is subsequence rationally weakly mixing.\
\
We now present a large class of examples that are not rationally weakly mixing. It will be convenient to write $$u_k(A, B) = \dfrac{\mu(A \cap T^kB)}{\mu(A)\mu(B)}$$ so that for $A$, $B$ of positive measure, we can divide by $\mu(A)\mu(B)$ to obtain $$\dfrac{1}{a_n(F)} \displaystyle \sum_{k = 0}^{n-1} |u_k(A,B) - u_k(F)| \rightarrow 0.$$ It is not difficult to see that in this case we must have $a_n(F)/a_n(A,B) \rightarrow 1$ [@Aa12]. This yields the following theorem:
\[T:notrwm\] Let $T$ be a rank-one transformation constructed by cutting $C_n$ in half and adding at least $c_n \geq 2h_n$ spacers on top of the right subcolumn at every step. Then $T$ is not rationally weakly mixing.
We prove by contradiction. Suppose that $T$ is rationally weakly mixing on some set $F$. Choose a level $J$ which is at least $(3/4)$-full of $F$, and let $J_1$ and $J_2$ be the left and right halves of $J$. By applying $T^{-1}$ to $F$, we may assume that $J$ is the bottom level of some column $C_j$. Now, both $J_1$ and $J_2$ intersect $F$ in positive measure, so $$\displaystyle \dfrac{1}{a_n(F)} \displaystyle \sum_{k = 0}^{n-1} \left| u_k(J_1 \cap F) - u_k(F) \right| \rightarrow 0$$ and $$\displaystyle \dfrac{1}{a_n(F)} \displaystyle \sum_{k = 0}^{n-1} \left| u_k(J_1 \cap F, J_2 \cap F) - u_k(F) \right| \rightarrow 0.$$ Moreover, $a_n(F)/a_n(J_1 \cap F) \rightarrow 1$. Multiplying through by this limit and using the triangle inequality, we obtain $$\displaystyle \dfrac{1}{a_n(J_1 \cap F)} \displaystyle \sum_{k = 0}^{n-1} \left| u_k(J_1 \cap F, J_2 \cap F) - u_k(J_1 \cap F) \right| \rightarrow 0.$$ Now, fix $k$ and suppose that $u_k(J_1 \cap F) > 0$. Then $\mu(J_1 \cap T^kJ_1) > 0$, so for sufficiently large $N$ we have $k \in D(J_1, N) - D(J_1, N)$. Similarly, if $u_k(J_1 \cap F, J_2 \cap F) > 0$, then $\mu(J_1 \cap T^{k + h_j}J_1) = \mu(J_1 \cap T^kJ_2) > 0$, which implies that $k + h_j \in D(J_1, N) - D(J_1, N)$. Hence we cannot have both $u_k(J_1 \cap F)$ and $u_k(J_1 \cap F, J_2 \cap F)$ nonzero, since then we would have $h_j \in (D(J_1, N) - D(J_1, N)) - (D(J_1, N) - D(J_1, N))$. As $D(J_1, N) = \{0, h_{j+1}\} \oplus \{0, h_{j+2}\} \oplus \cdots \oplus \{0, h_{N-1}\}$, this is easily seen to be impossible (given the fact that $c_n \geq 2h_n$ for all $n$).\
\
It is then immediate that $$\left| u_k(J_1 \cap F, J_2 \cap F) - u_k(J_1 \cap F) \right| \geq u_k(J_1 \cap F)$$ for every $k$. Indeed, if $u_k(J_1 \cap F) = 0$ then we are done; otherwise, $u_k(J_1 \cap F, J_2 \cap F)$ is $0$. It follows that the quotient (8) is bounded below by 1, which is a contradiction. This shows that $T$ is not rationally weakly mixing.
In particular, we obtain the following:
Let $T$ be the transformation constructed by cutting each column $C_n$ in half and adding $2h_n + 1$ spacers on top of the right subcolumn. Then $T$ is rationally ergodic and spectrally weakly mixing but not rationally weakly mixing.
This transformation is rationally ergodic by Theorem \[T:wraterg\] and the discussion following (and also by [@Aa79 Theorem 1]), and is spectrally weakly mixing by [@AdFrSi97 Proposition 1.1]. By Theorem \[T:notrwm\], however, $T$ is not rationally weakly mixing.
This negatively answers a question of Aaronson’s. (As noted in the introduction, this result was obtained independently by Aaronson in [@Aa12b 1.1].)\
\
We now extend Theorem \[T:notrwm\] to other rank-one transformations. Define $$H = \displaystyle \bigcup_{j = 0}^\infty H_j \setminus \{0\}$$ and observe that the elements of $H$ are increasing when listed in the obvious order. (Begin with successive elements of $H_0 \setminus \{0\}$, followed by successive elements of $H_1 \setminus \{0\}$, and so on.) We say that a rank-one transformation is $\mathbf{steep}$ if $t_{i + 1} \geq 4 t_i$ for every pair of successive $t_i, t_{i+1} \in H$. Clearly, the transformations of Theorem \[T:notrwm\] are steep. In general, such transformations can be constructed by adding an exponentially increasing number of spacers above successive subcolumns.\
\
Steep transformations satisfy several nice properties, chief among which is a linear independence condition that allows us to extend Theorem 3.2. Suppose we have a linear combination $$\displaystyle \sum_{t \in H} c_t t = 0$$ with the coefficients $c_t \in \{-2, -1, 0, 1, 2\}$. Then it is easily seen that all the $c_t$ must be 0. Similarly, we also obtain a uniqueness condition that will be useful in Section 5: every integer $k$ has at most one representation $$k = \sum_{t \in H} c_t t$$ with the $c_t \in \{-1, 0, 1\}$. Altering the definition of steepness slightly yields stronger forms of these properties; for example, requiring $t_{i + 1} \geq 5 t_i$ results in uniqueness of representation with $c_t \in \{-2, -1, 0, 1, 2\}$.
Let $T$ be a normal, steep rank-one transformation. Then $T$ is not rationally weakly mixing.
We sketch the proof and leave the details to the reader. As before, we proceed by contradiction. Suppose $T$ is rationally weakly mixing on $F$, and let $J$ be a level $(3/4)$-full of $F$. Without loss of generality, we may assume that $J$ is the bottom level of some column $C_j$. Now, there must exist at least two descendants $J_1$ and $J_2$ of $J$ in $C_{j+1}$ that have positive intersection with $F$. For these levels, we have $J_2 = T^dJ_1$ for some $d \in H_j - H_j$. It then suffices to show that $d$ cannot be contained in $(D(J_1, N) - D(J_1, N)) - (D(J_1, N) - D(J_1, N))$, which follows from the linear independence (9).
Relation to Double Ergodicity {#S:doubleerg}
=============================
In this section we show that rational weak mixing implies double ergodicity and present an example suggesting the converse implication is false.\
\
We begin by proving that rational weak mixing on $F$ implies double ergodicity for subsets of $F$.
Suppose that $T$ is rationally weakly mixing on $F$. Then $T$ is doubly ergodic for all $A,B \subset F$.
Let $A, B \subset F$, and fix $\delta > 0$ such that $$\delta < \frac{1}{2}\min(\mu(A)^2, \mu(A)\mu(B)).$$
Since $T$ is rationally weakly mixing on $F$, $$\dfrac{1}{a_n(F)} \displaystyle \sum_{k = 0}^{n-1} |\mu(A \cap T^kA) - \mu(A)^2u_k(F)| \rightarrow 0$$ and $$\dfrac{1}{a_n(F)} \displaystyle \sum_{k = 0}^{n-1} |\mu(A \cap T^kB) - \mu(A)\mu(B)u_k(F)| \rightarrow 0.$$ Summing these together, we obtain (by contradiction) that there exists a positive integer $k$ for which $u_k(F)>0$ and $$|\mu(A\cap T^kA)-\mu(A)^2u_k(F)|+|\mu(A\cap T^kB)-\mu(A)\mu(B)u_k(F)|<\delta u_k(F).$$ We thus have $$|\mu(A\cap T^kA)-\mu(A)^2u_k(F)|<\delta u_k(F)$$ and so $$\mu(A\cap T^kA)>u_k(F) (\mu(A)^2-\delta)>0$$ for this $k$. Similarly, $$\mu(A \cap T^kB)>u_k(F) (\mu(A)\mu(B)-\delta)>0.$$ By construction of $\delta$, this shows that $T$ is doubly ergodic on $F$.
We now extend this result to all of $X$. It was shown in [@Aa77] that if $T$ is weakly rationally ergodic on $F$, it is weakly rationally ergodic on any finite union $F_N = F \cup T(F) \cup \cdots \cup T^{N-1}(F)$. It follows that the analogous statement holds for rational weak mixing, giving the following theorem:
Suppose that $T$ is rationally weakly mixing. Then $T$ is doubly ergodic.
Let $T$ be rationally weakly mixing on $F$, and suppose that $T$ is not doubly ergodic. Fix $A, B \subset X$ for which the double ergodicity condition fails; i.e., choose $A$ and $B$ such that for every $n$, either $\mu(A \cap T^nA) = 0$ or $\mu(A \cap T^nB) = 0$. Since $F$ sweeps out $X$, there is some $N$ for which $F_N$ intersects both $A$ and $B$ in positive measure. Then $\tilde{A} = F_N \cap A$ and $\tilde{B} = F_N \cap B$ are sets of positive measure which fail the double ergodicity condition. But $T$ is doubly ergodic on $F_N$, a contradiction.
It is worth noting that (in general) the class of sets on which $T$ is doubly ergodic is $not$ a hereditary ring. For example, let $T$ be any doubly ergodic transformation on $X$, and define $S$ on $X \times \{0, 1\}$ by $S(x, 0) = (T(x), 1)$ and $S(x, 1) = (x, 0)$. Then $S$ is doubly ergodic on both $X \times \{0\}$ and $X \times \{1\}$, but not doubly ergodic on all of $X \times \{0, 1\}$. (Let $A = X \times \{0\}$ and $B = X \times \{1\}$.)\
\
We now investigate whether rational weak mixing is strictly stronger than double ergodicity. It will be useful for us consider transformations that are “almost" steep. Recall that $T$ is steep if for any pair of successive elements $t_i, t_{i + 1}$ in $H = (H_0 \cup H_1 \cup \cdots) \setminus \{0\}$, we have $t_{i+1} \geq 4 t_i$. Now, suppose $T$ is constructed so that:
(a) Each column $C_n$ is cut into at least three subcolumns ($r_n \geq 3$).
(b) We add zero spacers above the first subcolumn and one spacer above the second ($s_{n, 0} = 0$ and $s_{n, 1} = 1$).
(c) We add a sufficient number of spacers above each subsequent subcolumn so that $$\displaystyle \sum_{k = 0}^i h_{n, k} \geq 4 \left( \sum_{k = 0}^{i-1} h_{n, k} \right)$$ for every $2 \leq i \leq r_n -1$.
Then $T$ is “almost" steep, in the sense that $t_{i+1} < 4t_i$ only when $t_i$ and $t_{i+1}$ are the first two nonzero elements of some $H_n$. For such $T$, we can still extract a (slightly technical) algebraic uniqueness condition in the spirit of (9). Indeed, let $$B_n = \{h_{n, 0}, h_{n, 0} + h_{n,1}\} \times \{h_{n, 0}, h_{n, 0} + h_{n,1}\}$$ and define $$A_n = (H_n \times H_n) \setminus (\Delta H_n \cup B_n).$$ (Here, $\Delta H_n = \{(x, x) : x \in H_n\}$.) Then for any $(a, b), (a', b') \in A_n$ and $-M_n \leq k, k' \leq M_n$, the equality $$k + a - b = k' + a' - b'$$ implies $$a = a', b = b', k = k'.$$ (The proof of this is not difficult and is left to the reader.) Before we proceed, it will be useful to establish following lemma:
Let $J$ be any level, and fix $N$ sufficiently large. Suppose $(a, b) \in A_N$ and $-M_N \leq k \leq M_N$. Then $$\mu(J\cap T^{k + a - b}J) = \dfrac{1}{r_N} \mu(J\cap T^{k}J).$$
By Lemma \[L:lem2.1\], we have $$\mu(J\cap T^{k}J) = w_N\cdot |D(J, N) \cap (k + D(J, N))|$$ and $$\mu(J\cap T^{k + a - b}J) = w_{N+1}|D(J, N+1) \cap (k + a - b + D(J, N+1))|.$$ By uniqueness of (10), every representation of $k + a - b$ as an element of $D(J, N+1) - D(J, N+1)$ corresponds to exactly one representation of $k$ as an element of $D(J, N) - D(J, N)$, and vice-versa. Hence $$\mu(J\cap T^{k + a - b}J) = \dfrac{w_{N+1}}{w_N} \mu(J_1\cap T^{k}J) = \dfrac{1}{r_N} \mu(J\cap T^{k}J),$$ as desired.
We now show that if $T$ is almost steep and $\{r_n\}$ is sufficiently large, $T$ cannot be rationally weakly mixing.
\[T:almoststeep\] Let $T$ be a rank-one transformation. Suppose that $T$ is almost steep (as described above), and that $$\displaystyle \sum_{n = 0}^{\infty} \dfrac{1}{r_n} < \infty.$$ Then $T$ is not rationally weakly mixing.
We begin by proving that $T$ is not rationally weakly mixing on levels. Let $J$ be the bottom level of any column $C_j$, and let $J_1$ and $J_2$ be any two descendants of $J$ in $C_{j+1}$. Then $J_1 = T^dJ_2$ for some $d \in H_j - H_j$. As in Theorem \[T:notrwm\], it suffices to disprove the convergence $$\displaystyle \dfrac{1}{a_n(J_1)} \displaystyle \sum_{k = 0}^{n-1} \left|u_k(J_1) - u_k(J_1, J_2) \right| \rightarrow 0.$$ To do this, define $$P_m = \sum\limits_{k = -M_m}^{M_m}|\mu(J_1\cap T^kJ_1)-\mu(J_1\cap T^{k+d}J_1)|$$ and $$Q_m = \sum\limits_{k = -M_m}^{M_m}\mu(J_1\cap T^kJ_1).$$ For $m$ sufficiently large, $R_m = P_m/ Q_m$ approximates the quotient (11), so it is enough to show that $R_m$ is bounded below by some positive constant.\
\
Any choice of $(a, b) \in A_m$ and $- M_m \leq k \leq M_m$ yields a unique number $k + a - b$ between $-M_{m+1}$ and $M_{m+1}$. Hence $$\begin{aligned}
P_{m+1} &= \sum\limits_{k = -M_{m+1}}^{M_{m+1}}|\mu(J_1\cap T^kJ_1)-\mu(J_1\cap T^{k+d}J_1)| \\
&\geq \sum_{(a,b) \in A_m} \sum\limits_{k = -M_m}^{M_m}|\mu(J_1\cap T^{k + a - b}J_1)-\mu(J_1\cap T^{k+a-b+d}J_1)| \\
&= \dfrac{1}{r_m} \left(\sum_{(a,b) \in A_m}\sum\limits_{k = -M_m}^{M_m}|\mu(J_1\cap T^{k}J_1)-\mu(J_1\cap T^{k+d}J_1)|\right) \\
&= \dfrac{|A_m|}{r_m} P_m.\end{aligned}$$ Moreover, the same argument as in Theorem \[T:wraterg\] shows $$Q_m = \sum\limits_{k = -M_m}^{M_m}\mu(J_1\cap T^kJ_1) = |D(J_1, m)|\mu(J_1)$$ from which it follows that $$Q_{m+1} = r_mQ_m.$$ We thus obtain $$R_{m+1} \geq \dfrac{|A_m|}{r_m^2} R_m.$$ Now, $|A_m| = r_m^2 - r_m - 2$, so $R_m$ is bounded below by $$\displaystyle \prod_{k=0}^{\infty} \left(1 - \dfrac{1}{r_k} - \dfrac{2}{r_k^2}\right)R_0$$ which is a positive constant by the hypotheses of the theorem. This bounds (10) from below along the sequence $\{M_m + 1\}$. Since $T$ is rationally ergodic along the same sequence by Theorem \[T:renyi\], it follows that $T$ is not rationally weakly mixing.
We now show that $T$ is doubly ergodic for levels, suggesting that rational weak mixing is strictly stronger than double ergodicity.
The transformation $T$ above is doubly ergodic for levels.
We check that for any pair of levels $A$ and $B$, there exists an integer $n$ such that both $\mu(A\cap T^nA)>0$ and $\mu(B\cap T^nA)>0$. Without loss of generality, we may assume that $A$ is the bottom level of some column $C_j$ and that $B = T^dA$. It then suffices to prove there exists an $n$ such that both $n$ and $n+d$ are in $D(A,N)-D(A,N)$ (for $N$ sufficiently large). This is easy; simply choose $$n = h_{j+1,0}+\cdots+h_{j+d,0}.$$ Then $$n + d = ((2h_{j+1, 0} + 1) +\cdots + (2h_{j+d,0} + 1)) - (h_{j+1,0}+\cdots+h_{j+d,0}),$$ as desired.
Independence from Zero-type {#S:zerotype}
===========================
We now show that (subsequence) rational weak mixing and zero-type are independent (i.e., do not imply each other). We say that $T$ is $\mathbf{zero}$-$\mathbf{type}$ if $\mu(A \cap T^nA) \rightarrow 0$ for all sets $A$ of finite measure [@HaKa64]. It is well-known that in order to show a conservative ergodic transformation is zero-type, it suffices to check this convergence for a single set $A$ of positive finite measure [@HaKa64]. We show that every steep transformation with an increasing number of cuts is zero-type.
Let $T$ be a normal, steep rank-one transformation, and suppose that $\{r_n\}$ is nondecreasing with $\sup\{r_n\} = \infty$. Then $T$ is zero-type.
Consider $I = (0, 1)$. For $N$ sufficiently large, we have $$\mu(I \cap T^k I) = \dfrac{|D(I, N) \cap (k + D(I, N))|}{|D(I, N)|}.$$ Now, $|D(I, N) \cap (k + D(I, N))|$ counts the number of representations $$k = \displaystyle \sum_{i = 0}^{N-1} (d_i - d_i')$$ with $d_i, d'_i \in H_i$. (Recall that $D(I, N) = H_0 \oplus H_1 \oplus \cdots \oplus H_{N-1}$.) If $k \notin D(I, N) - D(I, N)$, then $\mu(I \cap T^k I) = 0$, so suppose that $k \in D(I, N) - D(I, N)$. Then there is at least one representation $$k = \displaystyle \sum_{i = 0}^{N-1} (x_i - x_i')$$ with $x_i, x'_i \in H_i$. Now, fix $n$ and suppose $x_n - x'_n \neq 0$. By uniqueness of representation, any other representation (12) of $k$ must have $d_n = x_n$ and $d'_n = x'_n$. In particular, the only indices $i$ at which (12) can differ from (13) are those for which $x_i - x'_i = 0$. In these cases we must have $d_i = d'_i$, but otherwise there are no restrictions (i.e., $d_i = d'_i$ can be any element of $H_i$). Hence $$\displaystyle |D(I, N) \cap (k + D(I, N))| = \prod \limits_{x_i - x'_i = 0} |H_i|$$ with the product being taken over all $i$ for which $x_i - x'_i = 0$. Since $$\displaystyle |D(I, N)| = \prod \limits_{i = 0}^{N-1} |H_i|$$ it follows that $$\mu(I \cap T^k I) = \left( \prod \limits_{x_i - x'_i \neq 0} |H_i| \right)^{-1}.$$ Now, if $k > M_n$, then the representation (13) of $k$ must have $x_m - x'_m \neq 0$ for some $m \geq n$. This implies that $$\mu(I \cap T^k I) \leq \dfrac{1}{|H_m|} = \dfrac{1}{r_m} \leq \dfrac{1}{r_n},$$ which shows $\mu(I \cap T^k I) \rightarrow 0$ as $k \rightarrow \infty$. Hence $T$ is zero-type, as desired.
We thus have:
There exist rank-one transformations that are zero-type but not rationally weakly mixing.
In [@Aa12b], Aaronson recently constructed a zero-type transformation of the form $T\times S$, where $S$ is a Markov shift, such that $T\times S$ is not subsequence rationally weakly mixing. Our examples, however, are rank-one, so of a different nature, and were constructed independently.\
\
We note that it follows from Theorem F in Aaronson [@Aa12] that there exist subsequence rationally weakly mixing transformation of positive type; a rank-one example is given by the subsequence rationally weakly mixing transformation of Section 3. (Indeed, this is partially rigid since $\mu(I \cap T^{h_i}I) \geq 1/2$ for every $i$.) Aaronson [@Aa12b] also constructed positive-type, rank-one, transformations that are not subsequence rationally ergodic.
Examples of Rational Weak Mixing {#S:ratweakmixex}
================================
We end with a construction of a positive-type rank-one transformation which is rationally weakly mixing. Let $T$ be a Chacón-like transformation ($r_n = 3$, $\{s_{n, 0} = 0, s_{n, 1} = 1, s_{n, 2} \geq 3h_n + 1\}$) with enough spacers added above every third subcolumn so as to have $h_{n+1} = 3^c h_n$ for some fixed integer $c \geq 2$. Then $h_n = 3^{cn}$ and $D(I, n) = \displaystyle H_0 \oplus H_1 \oplus \cdots \oplus H_{n-1}$, where $H_i = \{0,h_i,2h_i +1\}$.
\[T:ratwm\] The above transformation $T$ is rationally weakly mixing.
We prove rational weak mixing for levels. Let $j\in \mathbb N_0$, $J = J_1$ be the bottom level of $C_j$, and let $J_2 = T^dJ_1$. As in the proof of Theorem \[T:almoststeep\], it suffices to show the convergence (11). Now, for any $n$, we may choose $m$ such that $M_{m-1} \leq n-1 < M_m$. Then the quotient (11) is asymptotically bounded above by $P_m / Q_{m-1} = 3 P_m / Q_m$, so it suffices to prove $P_m / Q_m \rightarrow 0$ as $m \rightarrow \infty$.\
\
By the triangle inequality, $$\begin{aligned}
P_m &= \displaystyle\sum_{k = -M_m}^{M_m}|\mu(J \cap T^k J) - m (J \cap T^{k+d}J)| \\
&\leq \displaystyle \sum_{\ell=0}^{d-1}\displaystyle\sum_{k = -M_m}^{M_m} |\mu(J \cap T^{k+\ell} J) - m (J \cap T^{k+1+\ell}J)|.\end{aligned}$$ Since each of the $d$ outer sums on the right differs from the $\ell=0$ sum by a finite number of terms, it suffices to prove the convergence with only the $\ell= 0$ sum. That is, we wish to show $$\frac{\sum\limits_{k = -M_m}^{M_m}|\mu(J\cap T^kJ)-\mu(J\cap T^{k+1}J)|}{\sum\limits_{k = -M_m}^{M_m}\mu(J\cap T^kJ)} \rightarrow 0.$$ To do this, it will be useful to introduce some auxiliary functions. Given $(d, d') \in D(J, m) \times D(J, m)$, write $$d - d' = \displaystyle \left(\sum_{i=j}^{m - 1} d_i\right) - \left(\sum_{i=j}^{m - 1} d_i'\right)$$ with each $d_i, d_i' \in \{0, h_i, 2h_i + 1\}$. Replacing each instance of $2h_i + 1$ with $2h_i$ in (14) yields a sum of the form $$\displaystyle \sum_{i = j}^{m-1} \varepsilon_i3^{ci}$$ with each $\varepsilon_i \in \{-2, -1, 0, 1, 2\}$. This defines a function $$g: D(J, m) \times D(J, m) \rightarrow \{-2, -1, 0, 1, 2\}^{m - j}$$ taking the pair $(d, d')$ to the vector $\varepsilon = (\varepsilon_i)_{i = j}^{m-1}$, with the $\varepsilon_i$ as in (15).\
\
For each $\varepsilon \in \{-2, -1, 0, 1, 2\}^{m - j}$, define the “multiplicity function" $\tilde{\varepsilon}$ on $D(J, m) - D(J, m)$ by $$\tilde{\varepsilon}(k) = |g^{-1} (\varepsilon) \cap \{(d,d') | d,d' \in D(J,m) \text{ and } d-d'=k \}|.$$ That is, $\tilde{\varepsilon}(k)$ counts the number of pairs $(d,d')$ in $g^{-1}(\varepsilon)$ with $d-d'=k$. Then
The following properties hold:
(a) Fix $k \in D(J, m) - D(J, m)$. Then $$\displaystyle \sum_{\varepsilon} \tilde{\varepsilon}(k) = |D(J,m)\cap (k + D(J, m))|.$$ where the sum on the left is taken over all $\varepsilon \in \{-2, -1, 0, 1, 2\}^{m - j}$.
(b) Fix $\varepsilon = (\varepsilon_i)_{i = j}^{m-1} \in \{-2, -1, 0, 1, 2\}^{m - j}$. For each $p \in \{-2, -1, 0, 1, 2\}$, let $a_p$ be the number of $\varepsilon_i$ equal to $p$. Then $$\displaystyle \sum_k \tilde{\varepsilon}(k)=3^{a_0}2^{a_1+a_{-1}}.$$ where the sum on the left is taken over all $k \in D(J, m) - D(J, m)$.
For (a), simply observe that both the left and right-hand expressions count the number of pairs $(d, d') \in D(J, m) \times D(J, m)$ for which $d-d'=k$. For (b), observe that the sum on the left counts the number of pairs $(d,d')$ whose associated vector is $\varepsilon$. Now, if $\varepsilon_i = 0$ in (15), then we must have $d_i = d_i'$ in (14), and there are three ways that this can happen. Similarly, if $\varepsilon_i = 1$, then either $d_i = h_i$ and $d_i' = 0$, or $d_i = 2h_i + 1$ and $d_i' = h_i$. Proceeding in this manner, a counting argument yields the desired equality.
($\textit{Proof of Theorem~\ref{T:ratwm}, continued.}$)\
\
Applying Lemma 6.2 (a) and Lemma 2.1, we thus need to show $$\displaystyle \frac{\sum\limits_{k = -M_m}^{M_m}\left|\sum\limits_{\varepsilon}\tilde{\varepsilon}(k)-\sum\limits_{\varepsilon}\tilde{\varepsilon}(k+1)\right|}{\sum\limits_{k = -M_m}^{M_m}\sum\limits_{\varepsilon}\tilde{\varepsilon}(k)} \rightarrow 0.$$ By the triangle inequality, it suffices to prove this convergence after exchanging the order of summation in both the numerator and denominator. To this end, we exhibit a nonincreasing function $c(t)$ which converges to $0$ such that $$\displaystyle R(\varepsilon):=\frac{\sum\limits_{k}|\tilde{\varepsilon}(k)-\tilde{\varepsilon}(k+1)|}{\sum\limits_{k} \tilde{\varepsilon}(k)}\leq c(a_1+a_{-1})$$ for each $\varepsilon$. Once we have such a function, we obtain the following bound for large enough $N$:
$$\begin{aligned}
\displaystyle \sum_{\varepsilon}\sum\limits_{k = -M_m}^{M_m}|\tilde{\varepsilon}(k)-\tilde{\varepsilon}(k+1)| &\leq \sum_{\varepsilon}\sum\limits_{k = -M_m}^{M_m} \tilde{\varepsilon}(k) c(a_1 + a_{-1}) \\
&\leq c(N) \sum_{\substack{{\varepsilon} \text{ with} \\ {a_1 + a_{-1} \geq N}}}\sum\limits_{k = -M_m}^{M_m} \tilde{\varepsilon}(k) \\
&+ c(0)\sum_{\substack{{\varepsilon} \text{ with} \\ {a_1 + a_{-1} < N}}}\sum\limits_{k = -M_m}^{M_m} \tilde{\varepsilon}(k).\end{aligned}$$
Dividing this by the denominator of (16) yields $$\displaystyle \frac{\sum\limits_{\varepsilon}\sum\limits_{k = -M_m}^{M_m}|\tilde{\varepsilon}(k)-\tilde{\varepsilon}(k+1)|}{\sum\limits_{\varepsilon}\sum\limits_{k = -M_m}^{M_m}\tilde{\varepsilon}(k)} \leq c(N) + c(0) d(N,m)$$ where $$\displaystyle d(N, m) = \dfrac{\sum\limits_{\substack{{\varepsilon} \text{ with} \\ {a_1 + a_{-1} < N}}} \sum\limits_{k = -M_m}^{M_m} \tilde{\varepsilon}(k) }{\sum\limits_{\varepsilon}\sum\limits_{k = -M_m}^{M_m}\tilde{\varepsilon}(k)}.$$ We claim that $d(N, m) \rightarrow 0$ as $m \rightarrow \infty$. Combined with the fact (still to be proven) that $c(N) \rightarrow 0$ as $N \rightarrow \infty$, this will imply the convergence of (16) to zero.\
\
By Lemma 6.2 (b), $$\begin{aligned}
d(N, m) &= \left( \sum\limits_{\substack{{\varepsilon} \text{ with} \\ {a_1 + a_{-1} < N}}} 3^{a_0}2^{a_1+a_{-1}} \right) / \left( \sum\limits_{\varepsilon}3^{a_0}2^{a_1+a_{-1}} \right) \\
&\leq 2^N \left( \sum\limits_{\varepsilon} 3^{a_0} \right) / \left( \sum\limits_{\varepsilon}3^{a_0}2^{a_1+a_{-1}} \right)\end{aligned}$$ with the sums taken over all $\varepsilon \in \{-2, -1, 0, 1, 2\}^{m - j}$. Now, $\sum_{p=-2}^{2} a_p = m - j$ for each such $\varepsilon$, so we can view the sums over $\varepsilon$ as sums over 5-tuples $(a_0,a_{-1},a_{1},a_{-2},a_{2})$ of non-negative integers summing to $m-j$. That is, the above expression is equal to\
$$2^N \left(\sum \binom{m-j}{a_0, a_{-1}, a_{1}, a_{-2}, a_{2}} 3^{a_0} \right) / \left( \sum \binom{m-j}{a_0, a_{-1}, a_{1}, a_{-2}, a_{2}} 3^{a_0}2^{a_1}2^{a_{-1}} \right)$$\
with the summation as described above and $\binom{m-j}{a_0, a_{-1}, a_{1}, a_{-2}, a_{2}}$ the multinomial coefficient “$m-j$ choose $a_0, \cdots, a_2$". By the identity $$(x_1 + x_2 + x_3 + x_4 + x_5)^n = \sum_{e_1 + e_2 + e_3 + e_4 + e_5 = n} \binom{n}{e_1, e_2, e_3, e_4, e_5} x_1^{e_1} \cdots x_5^{e_5},$$ this is equal to $$2^N (3 + 1 + 1 + 1 + 1)^{m-j}/(3 + 2 + 2 + 1 + 1)^{m-j}.$$ Hence $D(N, m) \leq 2^N (7/9)^{m-j}$, which clearly goes to zero as $m \rightarrow \infty$.\
\
Next, we will show that $$s(n):=\sup_{\varepsilon} \left\lbrace R(\varepsilon): a_1+a_{-1}=n\right\rbrace$$ converges to $0$. (See (17) for a definition of $R(\varepsilon)$.) Setting $$c(t):=\sup\left\lbrace s(n): n\geq t\right\rbrace$$ then produces a nonincreasing function with the desired properties and completes the proof.\
\
Fix $\varepsilon$, and let $a$ be the minimum element of $D(J, m) - D(J, m)$ for which $\tilde{\varepsilon}(a) > 0$. Any $k \in D(J,m) - D(J,m)$ is expressible as $k = \sum \varepsilon_i 3^{ci} + \sum (+1) + \sum (-1)$, with the $+1$’s and $-1$’s coming from choosing $2h_i + 1$ for $d_i$ and $d_i'$ in (14). We now ask: how many $+1$’s and $-1$’s do we have for $k = a$? We have only one way of obtaining $\varepsilon_i = 2$ in (14): namely, $(2h_i + 1) - 0$. Similarly, we only have one way of obtaining $\varepsilon_i = -2$: namely, $0 - (2h_i +1)$. This introduces $a_2$ number of $+1$’s and $a_{-2}$ number of $-1$’s. We have three ways of obtaining $\varepsilon_i = 0$, none of which introduce a net number of $+1$’s or $-1$’s. For $\varepsilon_i = 1$, we have two possibilities: either $h_i - 0$ or $(2h_i + 1) - h_i$. Since we want to minimize $a$, we choose the former. Similarly, for $\varepsilon_i = -1$, we must have either $0 - h_i$ or $h_i - (2h_i + 1)$, and to minimize $a$ we choose the latter. It thus follows that $a$ has $a_2$ number of $+1$’s and $a_{-2} + a_{-1}$ number of $-1$’s; moreover, $\tilde{\varepsilon}(a) = 3^{a_0}$. It is then not difficult to see that $$\tilde{\varepsilon}(a + k) = 3^{a_0} \binom{a_1 + a_{-1}}{k}$$ for all $0 \leq k \leq a_1 + a_{-1}$, and is $0$ otherwise.\
\
Letting $n = a_1 + a_{-1}$, we thus have $$R(\varepsilon)=\dfrac{1}{2^n} \left(\sum_{k = 0}^{n-1} \left|\binom{n}{k}-\binom{n}{k+1}\right| + 2\right).$$ Suppose $n = 2l -1$. (The case when $n$ is even is dealt with similarly.) Since $$\left|\binom{n}{k}-\binom{n}{k-1}\right|=\binom{n + 1}{k}\left|\frac{(n + 1)-2k}{n+1}\right|,$$ the above expression yields $$\begin{aligned}
R(\varepsilon) &= \dfrac{1}{2^n} \left( \sum_{k=1}^{n}\binom{n + 1}{k}\left|\frac{(n + 1)-2k}{n+1}\right| + 2\right) \\
&= \dfrac{1}{2^n} \left(\sum_{k=1}^{2l - 1} \binom{2l}{k} \left| \dfrac{l - k}{l} \right| + 2\right) \\
&\leq \dfrac{1}{2^n} \left(2 \sum_{k=0}^{l} \binom{2l}{k} \left(\dfrac{l - k}{l} \right)\right).\end{aligned}$$ Using the combinatorial identity $$\sum_{k=0}^{l} \binom{2l}{k} \left( \frac{l-k}{l} \right) = \frac{l+1}{2l}\binom{2l}{l+1},$$ we obtain $$R(\varepsilon) \leq \dfrac{1}{2^{2l}} \binom{2l}{l+1}.$$ It is not difficult to see that this goes to $0$ as a function of $l$, thus proving that $T$ is rationally weakly mixing for levels. By Theorem \[T:wraterg\] and Lemma 3.1, it follows that $T$ is rationally weakly mixing.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We describe the morphological composition of a sample of 518 galaxies in the field of CL1358+62 at z=0.33, drawn from a large [*HST*]{} mosaic covering 53 sq. arcmin. The sample is complete to $I$=22, corresponding to $M_V$=-18.5 in the rest frame. The galaxy morphologies have been independently classified by the authors of this paper and by Alan Dressler. Dressler’s classifications place our work in context with the previous MORPHS study, and allow us to estimate the scatter between different sets of visual classifications.
We restrict most of our analysis to the brighter part of the sample, $I<21$ ($M_V<-19.5$), where the scatter between the two sets of classifications is $\sim$1 in morphological type. The scatter doubles at $I=22$, presumably due to the lower signal-to-noise and poorer sampling of faint, small galaxy images. To $I$=21 the two sets of classifiers agree on the fraction of early type galaxies (elliptical+S0): 72%. We conclude that CL1358+62 does not contain the large population of spiral galaxies found in other studies of clusters at $z\sim0.3$, and that there is probably a significant spread in the degree of cluster evolution at intermediate redshift.
The two groups of classifiers differ on the relative fraction of S0 and elliptical galaxies. We show that the distributions of ellipticities and bulge/total light cannot resolve this discrepancy. Nonetheless, we can derive significant constraints on physical models for the evolution of the galaxy population in CL1358+62. The higher ratio of S0 to elliptical galaxies (1.6) found by DF/MF/PvD requires that the evolution preserve the relative fraction of elliptical, S0 and spiral galaxies. Alternately, the lower ratio (1.1) found by AD requires that the evolution preserve the early-type to spiral ratio while increasing the S0 to elliptical ratio. In the latter case, a possible evolutionary mechanism is accretion of galaxies that predominantly evolve to S0’s between $z$=0.33 and the present.
We use our large body of spectra to make the correspondence between spectral and morphological type. Our data follow the pattern seen in the field at low redshift: emission line spectra are more prevalent among the later morphological types. The 11 identified k+a galaxies (absorption line spectra with strong Balmer lines) have S0–Sb morphologies.
author:
- Daniel Fabricant
- 'Marijn Franx and Pieter van Dokkum[^1]'
title: 'Galaxy Morphologies in the Cluster CL1358+62 at z=0.33[^2]'
---
*Accepted for publication in the Astrophysical Journal*
Introduction
============
The WFPC2 camera on the Hubble Space Telescope ([*HST*]{}) has made it possible to determine the morphologies of galaxies at intermediate redshift and beyond. It has been known for some time that the photometric and spectral properties of galaxies in intermediate redshift clusters differ from galaxies in nearby clusters; the population of blue, star-forming galaxies and post-starburst galaxies is larger at intermediate redshift, see e.g., [@bo84], [@dr87], [@gd88], [@cs87], and [@dr92]. Given the correlation between the spectral and morphological properties of galaxies ([@mm57], [@mo69], [@ke92]), we might be able to detect a corresponding evolution of galaxy morphology in intermediate redshift clusters. However, the evolution of morphology is likely to be more subtle than spectral evolution, since galaxies of the same morphological type can have significantly different star formation rates (e.g. [@ja99]).
In the most ambitious study of this sort with WFPC2 to date, the MORPHS group have classified over 1200 galaxies in 10 clusters at 0.37$<$z$<$0.56 ([@sm97]). They find that S0 galaxies are less common than in low redshift clusters and that the ratio of S0’s to E’s within a radius of $\sim$600 kpc (for H$_0$=50, q$_0$=0.5) decreases with redshift, falling from 2 in low redshift clusters to less than 0.5 at $z=0.5$.
[@an97] (see also [@an98]) have independently classified galaxies in a WFPC2 image of Cl0939+4713, one of the less concentrated clusters in the MORPHS sample. They find a ratio of S0’s to E’s of $\sim$2, quite comparable to a low-$z$ reference sample in the Coma Cluster. However, they classify 40-50% of the galaxies in Cl0939+4713 as spirals (S), in contrast with 20-30% S in a comparable region of the Coma Cluster. The sample of galaxies in Cl0939+4713 is relatively small ($\sim$70), and redshifts are available for less than one third of these.
[@co98] present a study of the morphological types in three clusters at $z$=0.31, also using WFPC2 images. There is substantial overlap between the authors of this paper and the MORPHS group, and the two groups have attempted to adopt a consistent morphological system. At $z$=0.31, Couch et al. find an excess of S’s, with an abundance at small radii ($\sim$400 kpc for $H_0$=50, $q_0$=0.5) approximately twice that in low-$z$ reference clusters. However, averaged over the three clusters, within 400 kpc, they find a ratio of S0’s to E’s at most slightly depressed relative to regions of comparable galaxy surface density in low-$z$ clusters. [^3]. Note, however, that the average $z$ of the MORPHS clusters is larger (0.46).
[@lu98] report the morphological types in two more distant clusters at $z\sim$0.9. One cluster, CL0023+04, appears to be composed of two low velocity dispersion groups, and contains predominantly S’s. The other, CL1604+43, with a velocity dispersion of $\sim$1200 km s$^{-1}$, contains $\sim$76% early-types. In the latter case, the S0/E ratio is found to be 1.7$\pm$0.9. This result is sensitive to the assumed morphological composition of the foreground/background population, but is evidence that the S0/E ratio does not decline smoothly with $z$.
Our approach is complementary to the [@dr97], [@an97], and [@co98] studies which predominantly describe the galaxy morphologies in cluster cores. We use mosaics of WFPC2 fields to study the galaxy population in a larger region (allowing larger galaxy samples per cluster), and we have acquired large numbers of spectra of cluster galaxies. The spectra remove ambiguity about cluster membership and allow us to directly connect the morphological and spectral properties of the galaxies. Our sample of clusters is x-ray selected, with x-ray luminosities exceeding 4$\times$10$^{44}$ erg s$^{-1}$ in the 0.2–4.5 keV band.
In CL1358+62 at $z$=0.33, we have drawn a complete sample of 518 galaxies to a magnitude limit $I$=22 from a WFPC2 mosaic image of CL1358+62 covering 53 square arcminutes. Spectra for 276 of the 518 galaxies in the morphological sample were previously obtained at the Multiple Mirror and William Herschel Telescopes. The color-magnitude relation of the 194 spectroscopically confirmed cluster members in the [*HST*]{} mosaic (3 are fainter than $I$=22) has been previously described in [@vd98]. The spectroscopic properties of 232 cluster members (some outside the [*HST*]{} mosaic), as well as the cluster dynamics have been described in [@fi98].
Our objectives in this paper are fourfold. (1) We introduce the morphological classification techniques that we will apply to our entire sample of clusters. (2) We classify the galaxies in CL1358+62 at z=0.33, comparing our classifications with those of an experienced external researcher, Alan Dressler. Our deep sample with two independent classifications provides a useful assessment of the scatter between WFPC2 visual morphological classifications at intermediate redshifts. (3) We describe the robust, classifier-independent conclusions and explore the physical implications of the differences between the two sets of classifications. (4) We connect the spectral and morphological types of the cluster galaxies.
The paper is organized in the following fashion. In $\S$ 2, we describe the photometric catalog from which the morphological sample was drawn. The two set of morphological classifications are discussed in $\S$ 3. The morphological composition of the cluster and evidence for morphological evolution are presented in $\S$ 4. The connection between the morphological and spectral properties of the galaxies is made in $\S$ 5. $\S$ 6 contains a brief discussion and conclusions.
Photometric Catalog
===================
Our photometric catalog, from which we draw galaxies for morphological classification, is derived from the [*HST*]{} F814W mosaic image. The techniques used to construct the mosaic are described in [@vd98]. Our use of the mosaic image, instead of individual WFPC2 CCD frames, slightly compromises the accuracy of the photometry, but considerably simplifies the source detection problem by eliminating most of the boundaries. Because our goal is to select a sample of galaxies for morphological classification to a consistent magnitude limit, rather than to perform precision photometry, this is a beneficial tradeoff. We used the SExtractor package, described by [@se96], to detect the galaxies and perform the photometry. We use a detection and analysis threshold of 24.5 mag arcsec$^{-2}$, and a zeropoint of 30.546 (3600 s exposures, May 1997 WFPC2 SYNPHOT update) to convert from instrumental magnitudes to a Cousins $I$ magnitude[^4]. The adopted zeropoint is the average of the value for the 3 WFPC2 CCDs, which vary by $\sim$$\pm$0.01 magnitudes. We use the SExtractor total magnitude estimator (see [@se96]), which is insensitive to the analysis threshold.
In order to allow convenient comparisons with the results of the MORPHS group ([@dr97]), we calculate a conversion of observed $I$ magnitudes (for $z<$0.6) to a $V$ magnitude in the rest frame. Some of the MORPHS classifications used WFPC2 F702W filter data, so we also derive a consistent conversion from F702W magnitudes to a rest frame $V$. The intent in [@dr97] was to work to a consistent limit of $M_V$=-20, but a transcription error from Table 9 of [@ho95] led to the adoption of a deeper limit of $M_V$$\sim$-19. (Here and throughout we use $M_V$ to refer to a rest frame $V$ absolute magnitude.) In contrast with [@dr97], we also apply an evolutionary correction to allow a closer comparison with low $z$ clusters. We adopt the [@dr97] cosmological model (H$_0$=50 km s$^{-1}$ Mpc$^{-1}$, q$_0$=0.5)
We use four numbers in addition to the distance modulus to convert from the WFPC2 magnitudes to $M_V$: (1) the conversion from the Vega-referenced WFPC2 natural filter system to Cousins filter bands, (2) the $z$-dependent $K$ correction to transform the observed Cousins $R$ and $I$ magnitudes to the rest frame, (3) the estimated rest frame galaxy colors, and (4) a $z$-dependent evolutionary correction. Each of these numbers depends on the spectral energy distribution of the galaxies, so the accuracy of this procedure is limited. We use these numbers, however, only to choose sample limiting magnitudes appropriately scaled with $z$. We adopt the approximate expressions described below for these conversions.
We first fit a linear relation to the synthetic transformation of F702W to $R$ as a function of $V-R$ (valid for $V-R$$<$1.5), using results from Fig. 10 of [@ho95]:
$R-F702W=0.31(V-R)$ (1a)
We take the $V-R$ colors of galaxies as a function of $z$ from [@fr94], with a morphological mix of 70% E and 30% Sbc to convert (1a) to:
$R-F702W=0.31(0.5373 - 0.9132z + 10.17z^2 - 11.21z^3)$ (1b)
To a good approximation (0.1 mag), Fig. 9 of [@ho95] shows:
$I-F814W=0$ (1c)
We fit polynomials to the average $K$ corrections of [@fr94] and [@pog97], using 70% E and 30% Sbc or Sc contributions:
$K_R=0.4293z + 3.807z^2 - 2.903z^3$ (2a)
$K_I=0.4910z + 0.4836z^2$ (2b)
From [@fr94] we take:
$V_{rest}-R_{rest}=0.53$ (3a)
$V_{rest}-I_{rest}=1.10$ (3b)
From the the early-type galaxy fundamental plane study of [@ke97] we derive an evolutionary correction:
$EC_{V_{rest}}=-0.77z$ (4)
The corrections are applied in the following fashion:
$I = M_V + DM + EC_{V_{rest}} - (V_{rest}-I_{rest}) + K_I$
Here, $DM$ is the distance modulus. Applying these expressions to CL1358+62, with a distance modulus at $z=0.3283$ of 41.62, we find that the MORPHS limit of $M_V$=-20 corresponds to an observed $I$=20.5.
Our photometric catalog completeness extends below $I=23$, but we have limited our morphological sample to $I=22$ to increase the reliability of the morphological classifications. At $I=22$, we attain a signal-to-noise ratio (S/N) that is very similar to the S/N attained at the $I=23$ classification limit adopted by the MORPHS group for their deeper images ([@sm97]). The standard deviation of the sky noise in our (3600 s) CL1358+62 image has an equivalent surface brightness of 24.4 mag arcsec$^{-2}$, as compared with 25 mag arcsec$^{-2}$ typical for the ($\sim$12600 s) MORPHS images. A crude scaling relation can be derived by assuming that all galaxies have the same surface brightness, and that our criterion is that classifications of equal reliability require the same number of pixels at the same S/N. For a shallower image, we can effectively bin the image of a brighter galaxy that is $n$ times larger $n\times n$ pixels to achieve a S/N that is $n$ times better. To recover a sky S/N deficit of 0.6 mag, we need to set a limit $\sim$1.2 mag brighter.
Morphological Classification
============================
There are 518 galaxies in the morphological catalog after removing stars and image artifacts. In the rest frame, the filter central wavelength corresponds to $\sim$6100 [Å]{}. We use the F814W images for morphological classification because they are deeper than the F606W images. [@dr97] have used either F702W and F814W images for classification at comparable redshifts (between 0.37 and 0.41).
Below, we describe two sets of morphological classifications. The first set was carried out by the authors of the paper using the techniques described in $\S$3.1. We realized that our classifications would be of greater value if we could compare them with morphological classifications from an external expert. Alan Dressler (AD) kindly agreed to independently classify the entire sample. Dressler is a member of the MORPHS group, and most importantly, has classified a large reference sample of low-$z$ cluster galaxies. The strength of the AD classifications is that the same experienced classifier has classified the galaxies at low and intermediate redshifts. Consistency has obvious benefits when searching for evolution.
However, it is important to keep in mind the difficulty of classifying these relatively faint and small galaxies from WFPC2 images. The brighter cluster ellipticals and S0’s in CL1358+62 ($I\sim19$) have effective radii (half-light radii), r$_e$, of 0.5$^{\prime\prime}$ to 0.6$^{\prime\prime}$. The faintest galaxies in the sample ($I\sim$22) have r$_e$$\sim$0.3$^{\prime\prime}$. The 50% encircled energy radius of WFPC2 stellar images measured from our frames is about 0.17$^{\prime\prime}$, so the numbers of meaningful information elements are small: typically 20 to 40. For this reason, we cannot be sure that agreement on a common morphological system is the most significant issue. Even the most experienced classifier may not account for all the systematic differences between the low $z$ (photographic) and the intermediate $z$ WFPC2 images. By comparing the two sets of morphological classifications we will be able to discern which aspects of the classification are most robust.
DF/MF/PvD Classifications
-------------------------
The 518 sample galaxies were independently classified along the revised Hubble sequence by DF, MF and PvD from 96$\times$96 pixel “postage-stamps” drawn from both the mosaic and original individual CCD images. We referred frequently to [@sa61] and [@sa87], as well as to artificially redshifted digital images drawn from the Nearby Field Galaxy Survey ([@ja99]). We assigned numerical types as follows: -5 (elliptical), -4 (elliptical or S0), -2 (S0), 0 (S0 or Sa), 1 (Sa), 3 (Sb), 5 (Sc), 7 (Sd), 9 (Sm), 10 (Im) and 99 (peculiar or merger). Intermediate types (2, 4, 6, and 8) were also assigned.
For 80% of the galaxies, the three independent classifications span a range of three or fewer numerical types (we consider -4, -2 and 0 to be adjacent numerical types), and agree exactly for 34% of the sample. For 6% of the sample, the object is considered unclassifiable by one or more of the authors, or else the classifications disagree wildly. In these cases, we assign a numerical type of 999. The remaining 14% are classified type 1 or later by all the authors, but with a broader range of classifications. Table 1 lists the combination rules used to assign types where the agreement is not exact, including type 15 for indeterminate late types.
AD Classifications
------------------
Following the completion of the DF/MF/PvD classifications, AD independently classified the CL1358+62 galaxies according to techniques described in [@sm97]. AD does not assign galaxies to the DF/MF/PvD intermediate types -4 (E or S0) or 0 (S0 or Sa) in the same fashion, preferring to subdivide these into E/S0 (-4), S0/E (-3), S0/Sa (-1) and Sa/S0 (0). When discussing the cluster population in broad terms, AD’s types -5 and -4 are combined into E, types -3, -2 and -1 into S0, and etc. DF/MF/PvD split the contents of the -4 bin equally into E and S0, and the 0 bin equally into S0 and Sa. When converting AD’s descriptive types into numerical types, we have placed a few galaxies into the merger (99) bin where an individual catalog object corresponds to an interacting or merging pair of objects.
Augustus Oemler, another member of the MORPHS group, independently classified 307 galaxies from the CL1358+62 sample ($I$=22 limit) to check if AD’s classifications adhere to the MORPHS system. Oemler’s classifications agree very well with AD’s overall, with $\sim$10% more S0’s by number, and a corresponding decrease in the numbers of E’s. The number of spirals is identical in both classifications.
Quantitative Comparison of the Classifications
----------------------------------------------
Like any measurement, classifications will suffer from random and systematic errors. We can estimate these errors by comparing multiple sets of classifications. Ideally, such a comparison should be based on classifications using independent imaging data. Here, both sets of classifiers (DF/MF/PvD and AD) worked from the same data set, so we may underestimate the errors. Nonetheless, this comparison is quite interesting. Figure 1 shows the difference between DF/MF/PvD and AD morphological types for individual galaxies as a function of magnitude. As expected, the differences increase at fainter magnitudes. To avoid creating artifical gaps in Figure 1, we condensed and shifted the numerical types of the early type galaxies for this presentation: -3 for E’s, -2 for E/S0’s and S0/E’s, -1 for S0’s, 0 for S0/Sa and Sa/S0. For types Sa and later, the “normal” types were used. Galaxies which were classified as merger, peculiar, or unclassified by one group were ignored.
Figure 2 is a scatter diagram comparing the two sets of classifications. The scatter is dominated by random differences, but there is some evidence for systematic differences for the early type galaxies. We return to this point below. The scatter between the classifications has been measured by calculating the mean absolute deviation, and normalizing it to the mean absolute deviation of a gaussian with an rms of 1. This measurement of the scatter is much more robust than the RMS of the differences. The result is shown in Figure 3. The scatter is $\sim$1 for bright galaxies, but increases strongly faintwards of $I=21$. The appendix describes how classification errors might systematically bias our population estimates; unfortunately we cannot simply calculate correction factors. In what follows, we restrict most of our analysis to the subset of galaxies with $I<21$.
Classification of Ellipticals and S0s
-------------------------------------
Figure 2 shows a systematic difference in the DF/MF/PvD and AD classifications of early type galaxies. This is not surprising as the division between E and SO is a difficult problem in visual classifications. We might hope that looking at two objectively determined structural parameters for the galaxies, ellipticity and the bulge/total light ratio, might be helpful in resolving this issue. For example, [@sm97] compare the ellipticity distributions of the MORPHS E and S0 intermediate-$z$ galaxies with those from an [@an96] study of Coma Cluster galaxies as a consistency check of the MORPHS classifications.
We have therefore examined the structural properties of the CL1358+62 galaxies measured by the Medium Deep Survey (MDS) group[^5], [@ra99]. The MDS group has published structural properties for 70% of the CL1358+62 galaxies in our morphological catalog. The missing 30% are located near frame edges or in one mosaic frame that is not yet included in the MDS database. The MDS group has chosen the best fitting of four structural models for the objects in their catalogs: star, disk, bulge, or disk + bulge, accounting for the point spread function of HST. If a galaxy model is chosen, the MDS group fit the ellipticity of the disk, bulge or disk and bulge separately as appropriate. We focus on two structural parameters derived from these fits: the ratio of bulge to total light, and the weighted ellipticity. The ratio of bulge to total light follows trivially from the best fit model, or from the fits to the disk + bulge model. We calculate the weighted ellipticity from the MDS disk and bulge ellipticities, weighting by the fractions of light in the disk and bulge.
In Figures 4 and 5 we plot the distributions of the bulge to total light ratio for the E and S0 galaxies with $I<21$ in the DF/MF/PvD and AD samples, respectively. Excepting the differences in numbers, the distributions for both E’s and S0’s look remarkably similar for the two sets of classifications.
In Figures 6 and 7 we plot the distribution of weighted ellipticities for the same E and S0 galaxies ($I<21$), again showing the DF/MF/PvD and AD samples independently. Caution must be exercised when comparing to these to other measures of ellipticity since the MDS numbers are corrected for the HST PSF. A larger fraction of the AD E’s have ellipticities exceeding 0.2, but the mean AD E ellipticity is 0.22, only slightly larger than the DF/MF/PvD mean of 0.17. The ellipticities of the two samples of S0’s are similar: AD finds a mean ellipticity of 0.46 and DF/MF/PvD find a mean ellipticity of 0.40. The sizable difference in the numbers of E and S0 galaxies found by the two sets of classifiers is not reflected in a large difference between the structural parameters for the two samples of E’s or S0’s. We conclude that although the total number of early-type galaxies is well established, the SO to E ratio in CL1358+62 is uncertain.
Morphological Composition of CL1358+62 and Evidence for Evolution
=================================================================
We have learned that the differences between the two sets of classifications rise steeply below $I=21$, suggesting this as a practical limit for our morphological study. This is also conveniently close to the effective limit of our spectroscopic completeness. We have redshifts for 277[^6] of the 518 galaxies in the morphological sample, and 191 of these are cluster members by the criteria given in [@fi98]: 0.31461 $<$ z $<$ 0.34201. The spectroscopic completeness is 89% for the galaxies brighter than $I=20.5$ (corresponding to $M_V=-20)$, falling to 59% for $20.5<I<21$, 33% for $21<I<21.5$, and 9% for galaxies with $21.5<I<22$.
Table 2 lists positions, $I$ magnitudes, both sets of morphological classifications, and radial velocities for galaxies brighter than $I$=21.
Morphological Composition
-------------------------
We may study the morphological composition of the subsample of known members, which is nearly complete to the MORPHS’s depth of $M_V=-20$, without concern about background galaxy subtraction. The subsample of known cluster members to $I=20.5$ (or $M_V=-20$, see $\S$ 2) contains 138 galaxies. DF/MF/PvD classify 27$\pm$4% of these as E, 44$\pm$6% as S0, 29$\pm$5% as S, with 1 unclassified galaxy. (The errors here and in the following discussions account only for the Poisson statistics of the number of galaxies per classification bin.) AD classifies 35$\pm$5% as E, 38$\pm$5% as S0, and 27$\pm$4% as S. In both cases, the total early type population is $\sim$72%. To $M_V$=-20, then, the only difference between the two groups of classifiers is the relative numbers of E’s and S0’s.
We consider also the complete photometric sample to a depth of $I=21$ (or $\sim$$M_V=-19.5$), where a small background correction is necessary. To $I=21$ our sample contains 298 galaxies, for which we have 236 redshifts (79.2% completeness). Of the 236 galaxies with redshifts, 65$\pm$8 are nonmembers, yielding a foreground/background count of 82$\pm$10 after correcting for the spectroscopic completeness. We determine the morphological composition of the foreground/background galaxies directly from our spectroscopic sample, which contains 86 foreground/background galaxies. DF/MF/PvD classify 3% E, 13% S0, 77% S, and 6% mergers. AD classifies 11% E, 9% S0, 71% S, and 9% mergers. We average these classifications, and adopt a background composition of 7% E, 11% S0, 74% S and 8% mergers. For comparison, [@dr97] adopt a morphological composition of 10% E, 10% S0, and 80% S.
After correcting for background, to $I=21$, DF/MF/PvD find a population of 25$\pm$4% E’s, 46$\pm$5% S0’s, 29$\pm$6% S’s, and 0$\pm$1% mergers. AD finds 36$\pm$4% E’s, 36$\pm$5% S0’s, 28$\pm$6% S’s, and 1$\pm$1% mergers. These results are indistinguishable from those for the brighter spectrosopic sample, with $\sim$72% early-types in both sets of classifications.
Morphological Evolution
-----------------------
We search for morphological evolution in CL1358+62 by comparing its population with that of equivalent low-$z$ clusters. Judging which low-$z$ clusters are equivalent is somewhat uncertain, but we take as an approximation low-$z$ clusters with a similar number of galaxies within a fixed metric aperture, allowing us to correct for the effects of the morphology-density relation. We use the nearby cluster catalog of [@dr80], reanalyzed and summarized in [@dr97] as a benchmark. We use the subset of high concentration clusters (10 of 55) in the low-$z$ sample for comparison, because CL 1358+62 was selected for its high x-ray luminosity and has a concentration index C$\sim$0.49 ([@fa91]).
The data for the 10 high-concentration clusters are plotted in Fig. 12 of [@dr97]. There are an average of $\sim$63 cluster galaxies within a radius of 1450 kpc in these clusters to $M_V$=-20.4. In CL1358+62, there are $\sim$114 cluster members within this radius to $M_V$=-20.4, so CL1358+62 is richer than the average cluster in the low-$z$ sample. However, the low-$z$ sample does contain a high-concentration cluster as rich as CL1358+62: the Coma Cluster. Furthermore, the density difference between CL1358+62 and the average low-$z$ reference cluster, 0.3 dex, is comparable to the bin size in the morphology-density relation plots in [@dr97].
The comparison between the morphological composition of the CL1358+62 sample to $M_V$=-20 and the low-$z$ reference sample is made in Table 3 and Figure 8. The DF/MF/PVD S0/E ratio, 1.6$\pm$0.3, differs at only 1.4$\sigma$ confidence from the low-$z$ reference sample ratio of 2.1$\pm$0.2. AD’s ratio, 1.1$\pm$0.2, differs from the the low-$z$ reference sample ratio at 3.5$\sigma$ confidence.
Despite this difference in the S0/E ratio, we stress that both classifications for CL1358+62 yield fractions of early types (E+S0), $\sim$72%, and late types (S), $\sim$28%, that are identical within the errors to the low-$z$ reference sample. We can therefore draw a robust conclusion from the two sets of morphological classifications: CL1358+62 does not contain an elevated population of spiral galaxies compared with low $z$ reference clusters. This contrasts with the results of [@an97] and and [@co98] for other intermediate $z$ clusters. Our work suggests that the early type/spiral classifications are likely to be secure, implying that the populations of intermediate $z$ clusters vary significantly, even after accounting for the effects of the low $z$ morphology-density relation.
This conclusion about the spiral population in CL1358+62 limits the range of physical models for evolution in CL1358+62, even allowing for our uncertainty about the E/S0 classifications. As we mentioned earlier, the AD classifications have the strong advantage of the same classifier at low and intermediate $z$. However, given the different character of the low-$z$ photographic images and the intermediate-$z$ WFPC2 images, we must acknowledge the possibility of systematic, redshift-dependent classification uncertainties. Figure 1 provides some reason to be cautious about this issue. Because we do not understand in detail the reasons for the differences between the two sets of visual classifications, we cannot be positive that the two sets of classifications bound our uncertainties. However, the best we can do at present is to leave the issue of E/S0 classifications open, and to explore the consequences of both sets of classifications below.
The DF/MF/PvD classifications would imply that cluster evolution from $z=0.33$ to the present does not affect the cluster morphological composition or its morphology-density relation within the observed 1.4 Mpc radius aperture. Figure 9 shows this relation for CL1358+68, binning the background subtracted data to I=22 ($M_V$=-18.5) radially about the dominant central galaxy. Since we are looking only for radial trends, using the deeper sample is appropriate here. The average galaxy density in each of four radial bins (0–1, 1–2, 2–3, and 3–5 arcmin) is the abscissa for this histogram. The galaxy densities in Figure 9 have been normalized to a $M_V$=-20 limit to allow comparison with the low-$z$ reference sample. We find that the morphology-density relation for CL1358+62 is indistinguishable within the errors from the low-$z$ relation (to $\sim$$M_V$=-20) shown in Figure 3 of [@dr97].
The AD classifications suggest an evolutionary mechanism that decreases the fraction of E’s, increases the fraction of S0’s, while leaving the fraction of spirals unchanged. Assuming that there is no plausible mechanism for directly converting E’s into S0’s, we can exclude models that transform the observed $z$=0.33 cluster spirals into S0’s without accretion of additional galaxies. In order to transform the $z$=0.33 AD morphological mix into the low-$z$ reference sample population while accreting the smallest number of galaxies, the cluster population would increase by 50%. Approximately 70% of the accreted galaxies would become S0’s by the present day, and 30% spirals.
The Morphological-Spectral Connection
=====================================
While a great deal has been learned about the galaxy population in intermediate redshift clusters from relatively small samples of spectra, we must remember that even present-day clusters of galaxies are a heterogeneous group, differing widely in their degree of virialization. Because cluster relaxation may drive galaxy evolution, the range in cluster galaxy populations may be large at any redshift. For this reason, it is desirable to connect the morphological and spectral properties of a large sample of galaxies in each of a number of clusters directly.
We have spectral classifications from [@fi98] sorting each of the galaxies with spectra into one of the four categories: (1) absorption lines only, (2) emission lines present, (3) emission lines plus strong Balmer absorption lines, and (4) k+a (also called E+A). Category (4) contains galaxies with the normal absorption lines of E/S0 galaxies plus strong Balmer absorption lines. Emission line galaxies have \[OII\] 3727 [Å]{} emission with equivalent width (EW) $>$5 [Å]{}. If H$_{\delta}$ absorption of $>$4 [Å]{} EW is detected for an emission line galaxy, the galaxy is classified as emission plus Balmer lines. Galaxies with \[(H$_{\delta}$ + H$_{\gamma}$ + H$_{\beta}$)/3\] EW greater than 4 [Å]{}, but \[OII\] emission with $<$5 [Å]{} EW, are classified as k+a. We refer the reader to [@fi98] for a more complete summary of the spectral properties of these galaxies and a comparison with spectra of low-$z$ galaxies. Table 4 and Figure 10 summarize the comparison between the spectral and morphological properties for the 191 cluster members in common with [@fi98], using the DF/MF/PvD classifications.
In rough terms, the bulk of the CL1358+62 galaxies follow the morphological-spectral correlation expected for bright field galaxies at low redshift: the preponderance of the E and S0 galaxies have pure absorption line spectra, while the fraction of galaxies with emission lines rises for the late type spirals. [@dr99] found a similar behavior for galaxies in the MORPHS sample; see also [@pog99]. We do know, however, that the percentage of galaxies with emission lines and strong Balmer absorption lines, $\sim$19%, is higher than the $\sim$6% content of these galaxies in comparable low $z$ clusters ([@dr87], see also [@fi98]).
If we use the AD classifications for the CL1358+62 galaxies, these conclusions do not change significantly. The most interesting difference between the two sets of morphological classifications is that DF/MF/PvD classify all the E+A galaxies as types S0 to Sb, while AD classifies these galaxies as having a wider range of morphologies from E to Sbc.
Discussion and Conclusions
==========================
For CL1358+62, we have acquired a unique data set including a large mosaic of HST fields and extensive spectroscopy that allows us to unambiguously determine cluster membership for galaxies with $M_V <
-20$. We have directly compared the morphological classifications of two sets of classifiers for the galaxies in CL1358+62. The two sets of classifiers agree that (to a limit of $M_V$=-20) the fraction of early type galaxies (and therefore spirals) in this cluster at $z$=0.33 is indistinguishable from the fraction in comparable low-$z$ clusters. In contrast, previous workers, [@an97] and [@co98], who also studied WFPC2 images of clusters at z$\sim$0.3, found an elevated population of spirals compared with low-$z$ reference samples. Because our work confirms the reliability of early-type/spiral classifications from intermediate $z$ WFPC2 observations, we conclude that this is evidence for a dispersion in the evolution of intermediate-$z$ clusters.
The two groups of classifiers differ on ratio of E to S0 galaxies in CL1358+62. DF/MF/PvD find a population of S0 galaxies (S0/E=1.6$\pm$0.3) that is within 1.4$\sigma$ of the low-$z$ reference sample, while AD finds a significantly smaller ratio (1.1$\pm$0.2). This systematic difference is most likely related to the fact that the transition between S0’s and intermediate luminosity E’s is rather gradual. Many of the intermediate luminosity E’s are thought to have disks, e.g. [@sc98], [@ri90], and [@jo94]. It may only be possible to resolve this issue by direct model fitting to images at low and intermediate $z$.
Even though we conclude that we have not reliably determined the ratio of S0’s to E’s among the early types, our work significantly restricts possible evolutionary models. If we accept the MF/DF/PvD classifications, evolution must preserve the fraction of E’s, S0’s and S’s as well as the morphology-density relation. If the AD classifications are correct, evolution must decrease the fraction of E’s and increase the fraction of S0’s while maintaining the fraction of S’s. A possible mechanism for driving the evolution of the morphological mix in this latter fashion is accretion of additional galaxies from the spiral-rich infall region that become predominantly S0’s. The cluster population within an $\sim$1 Mpc radius must increase by a minimum of 50% from $z$=0.5 to the present day in order convert an intermediate $z$ population rich in E’s to a low-$z$ population rich in S0’s. It will be interesting to see whether such accretion can be produced in simulations of cluster formation. In most cluster formation scenarios, massive clusters form by the merging of pre-existing massive clusters with (presumably) similar populations of early type galaxies. It may therefore be difficult to double the ratio of S0 to E galaxies.
We compare our morphological classifications to our previous spectral classifications and conclude that the morphologies of the spectrally “active” galaxies are as might be expected from the low-$z$ field population: the galaxies with emission lines are predominently spirals and the k+a (or E+A) post-starburst galaxies are typically early type disk galaxies (S0–Sb).
We wish to acknowledge the generous contributions of Alan Dressler to this paper, including his independent classifications and insightful comments. We thank Gus Oemler for checking the classifications and for thoughtful comments on the manuscript, which led to several improvements. We thank Margaret Geller for a critical reading of an earlier version of the manuscript and helpful comments. Our referee, Ian Smail, and our editor, Greg Bothun, both made insightful comments that helped us clarify the paper.
APPENDIX
SYSTEMATIC EFFECTS OF MEASUREMENT ERRORS
The effects of classification errors on the distribution of types can be important. Quantifying these effects is difficult because morphological classification is a subjective procedure, but we can gain some insight by considering simple models for the errors. We begin by assuming that the numerical type is based on a one dimensional measurement with a simple, constant error. This simple model would imply that any peaks in the distribution of types would be softened. If we adopt nominal intrinsic fractions of S:S0:E of 0.20:0.53:0.27, as is approximately correct for the inner 600 kpc of low redshift clusters ([@dr97]), the errors will automatically decrease the fraction of S0’s, and enhance the fraction of E’s and S’s.
We can calculate an upper limit to the loss of S0’s for our nominal 20:53:27 (S:S0:E) population with this model for the errors. We assume that the types are scattered with a mean absolute deviation (MAD) of 1, and all intermediate types are divided equally between adjacent types. The outcome of this experiment is a distribution of 0.27:0.38:0.34 (S:S0:E). In this case, all galaxies scattered beyond the normal type boundaries were assigned to the boundary type (i.e. E). A MAD of 1 may be an overestimate as the difference in the types assigned by the two groups of classifiers is of this order. The intrinsic errors are $\sqrt{2}$ smaller if the errors are independent. For these smaller errors the resulting distribution would be 0.24:0.43:0.33 (S:S0:E). In either case, the results serve to illustrate that the systematic effects can be very significant.
Let us consider a second, more physical model for the visual classification errors. Here, we assume that morphological type is based on two independent variables with continuous distributions: bulge-to-total light fraction ($f_b$), and asymmetric features due to spiral arms ($A$). This is very similar to the quantitative classification devised by [@ab96]. Galaxies with low $A$ will be classified as early type ($t < 0$) and then divided into E’s or S0’s based on whether $f_b$ is above or below a critical value. It has been argued that most $L*$ ellipticals have faint disks, e.g. [@ri90] and [@jo94]. Similarly, the spiral classification will be based on a combination of $f_b$ and $A$.
If the intrinsic distribution of $f_b$ is flat, then errors in $f_b$ will not change the ratio of S0’s to E’s. However, if the intrinsic distribution of $f_b$ is peaked, the errors will have a systematic effect. The sign depends on the details of the intrinsic $f_b$ distributions and the size of the errors. Contributing to this uncertainty, the errors can also be asymmetric, if for example, faint extended disks are missed in noisy data. The situation is similar for errors in $A$, where E’s and Sa’s now share a boundary. Extensive simulations are required to estimate the systematic effects of limited S/N on classification errors, taking into account the the point spread function of WFPC. We will undertake this effort in a future paper.
[DUM]{} Abraham, R. et al. 1996, , 107, 1. Andreon, S., Davoust, E. & Heim, T. 1997, , 323, 337. Andreon, S. 1998, , 501, 533. Andreon, S. et al. 1996, , 116, 429. Bertin, E. & Arnouts, S. 1996, , 117, 393. Butcher, H. &, Oemler, A.. 1984, , 285, 426. Couch, W. & Sharples, R. 1987, , 229, 423. Couch, W. et al. 1998, , 497, 188. Dressler, A. 1980, , 236, 351. Dressler, A. 1987, Nearly Normal Galaxies, ed. Faber, S., (New York:Springer-Verlag), 276. Dressler, A. &Gunn, J.E. 1992, , 78, 1. Dressler, A. et al. 1997, , 490, 577. Dressler, A. et al. 1999, , 122, 51. Fabricant, D., McClintock, J. & Bautz, M. 1991, , 381, 33. Fisher, D., Fabricant, D., Franx, M., & Van Dokkum, P. 1998, , 498, 195. Frei, Z. & Gunn, J. 1994, , 108, 1476. Geller, M. et al. 1997, , 114, 2205. Griffiths, R. et al. 1994, , 435, L19. Gunn, J. & Dressler, A. 1988, Towards Understanding Galaxies at High Redshift, ed. Kron, R. & Renzini, A., (Dordrect:Kluwer), 227. Holtzmann, J. et al. 1995, , 107, 156. Jansen, R. et al. 1999, submitted to . Jorgensen, I. & Franx, M. 1994, , 433, 553. Kelson, D. et al. 1997, , 478, 13. Kennicutt, R. C., Jr. 1992, , 388, 310. Lubin, L. et al. 1998, , 116, 584. McLeod, B. & Rieke, M. 1995, , 454, 611. Morgan, W. W. & Mayall, N. U. 1957, , 69, 291. Morgan, W. W. & Osterbrock, D. E. 1969, , 74, 515. Poggianti, B. M. 1997, , 122, 399. Poggianti, B. M. 1999, , 518, 576. Postman, M. et al. 1998, , 506, 33. Ratnatunga, K., Griffiths, R., & Ostrander, E. 1999, , submitted. Rix, H.-W. & White, S. 1990, , 362, 52. Sandage, A. 1961, The Hubble Atlas of Galaxies (Washingon, DC: Carnegie Institution of Washington) Sandage, A. & Tammann, G. A. 1987, A Revised Shapley-Ames Catalog of Bright Galaxies (Washingon, DC: Carnegie Institution of Washington) Scorza, C. et al. 1998, , 131, 165. Shectman, S. et al. 1996, , 470, 172. Smail, I. et al. 1997, , 110, 213. van Dokkum, P. et al. 1998, , 500, 714.
[cccc]{} -5 & -5 & -4 & -5\
-5 & -5 & -2 & -4\
-5 & -4 & -4 & -4\
-5 & -4 & -2 & -4\
-4 & -4 & -2 & -4\
-2 & -2 & -5 & -4\
-2 & -2 & -4 & -2\
-2 & -2 & 0 & -2\
-2 & -2 & 1 & 0\
-2 & -2 & 2 & 0\
-2 & 0 & 1 & 0\
-2 & 1 & 1 & 0\
-2 & 1 & 2 & 0\
[ccccccc]{} 1 & 2185.23 & 198.20 & 20.85 & 99246 & -4 & -4\
12 & 4165.23 & 344.75 & 19.79 & 97702 & -2 & -4\
15 & 3424.60 & 366.51 & 19.24 & 95807 & -2 & 1\
16 & 2155.44 & 387.90 & 19.36 & 97243 & 1 & 1\
25 & 968.84 & 444.59 & 20.54 & 129864 & 6 & 5\
27 & 4576.15 & 456.72 & 19.00 & - & 5 & 5\
30 & 1479.89 & 477.46 & 19.38 & 98463 & -4 & -5\
33 & 4231.61 & 489.34 & 19.13 & 143869 & 0 & 0\
35 & 4095.85 & 492.36 & 20.47 & 98424 & -4 & -3\
37 & 3526.59 & 495.73 & 20.42 & 96743 & -2 & -2\
38 & 3377.31 & 502.24 & 18.81 & 99803 & 0 & -2\
48 & 3302.50 & 558.38 & 20.98 & 164142 & 15 & 99\
51 & 4062.30 & 568.56 & 20.77 & - & 8 & 4\
52 & 3024.67 & 571.39 & 16.96 & 48524 & 4 & 3\
53 & 3081.65 & 574.07 & 20.08 & - & 4 & 5\
54 & 1525.88 & 580.51 & 15.69 & 25570 & 4 & 5\
61 & 2357.29 & 602.25 & 20.50 & 144702 & 15 & 7\
66 & 4121.06 & 625.81 & 20.99 & - & -2 & -2\
68 & 3470.13 & 635.49 & 18.95 & 96710 & -2 & -2\
70 & 3598.20 & 642.90 & 20.04 & 80512 & 5 & 5\
79 & 4072.28 & 697.05 & 18.92 & 98334 & 0 & -4\
94 & 2580.14 & 725.66 & 19.64 & 97702 & -2 & -2\
101 & 2602.51 & 739.65 & 20.44 & 151970 & 0 & -5\
103 & 3652.82 & 740.90 & 20.43 & 95699 & 3 & 5\
107 & 3499.88 & 757.63 & 20.99 & - & 4 & 5\
114 & 2736.96 & 794.55 & 18.89 & 31012 & 5 & 6\
118 & 2286.07 & 807.47 & 19.08 & 97741 & 0 & -1\
120 & 1458.64 & 807.67 & 20.02 & - & 6 & 5\
125 & 4464.63 & 814.64 & 20.65 & - & 5 & 5\
127 & 3372.60 & 818.52 & 19.95 & 99845 & -4 & -5\
132 & 907.72 & 840.09 & 20.65 & - & -5 & -5\
140 & 1319.17 & 854.62 & 19.83 & 98991 & -4 & -5\
156 & 1940.02 & 919.07 & 19.74 & 149319 & 1 & 99\
159 & 3641.99 & 927.73 & 20.65 & - & 5 & 4\
162 & 2605.02 & 933.83 & 19.94 & - & 99 & -3\
178 & 558.31 & 1001.72 & 20.38 & - & 999 & 99\
180 & 1745.41 & 1004.04 & 18.59 & 96677 & 0 & 1\
186 & 2657.55 & 1009.86 & 20.83 & - & -5 & -5\
192 & 3689.81 & 1034.60 & 20.64 & 100272 & -2 & -2\
197 & 2634.39 & 1041.72 & 19.94 & 144632 & 2 & 2\
205 & 1501.76 & 1068.61 & 19.61 & 120912 & 5 & 1\
206 & 3733.06 & 1075.41 & 20.86 & 98397 & 0 & 0\
222 & 873.00 & 1136.15 & 20.86 & 99237 & -4 & -5\
237 & 2777.48 & 1171.76 & 19.64 & 97525 & 5 & 3\
239 & 1674.39 & 1188.96 & 20.92 & 207514 & 7 & 10\
244 & 3455.89 & 1200.72 & 20.08 & - & 999 & 999\
248 & 1019.98 & 1207.10 & 20.02 & 99291 & -2 & -2\
269 & 698.78 & 1258.32 & 20.54 & - & 4 & 6\
277 & 1244.07 & 1269.37 & 20.59 & - & 2 & 2\
279 & 999.08 & 1272.41 & 20.44 & 162101 & 4 & 3\
283 & 1400.74 & 1280.19 & 20.93 & 98430 & -2 & -2\
285 & 3410.44 & 1288.24 & 20.66 & 100257 & 4 & 4\
286 & 2933.14 & 1291.02 & 20.34 & 98343 & 6 & 4\
295 & 1818.09 & 1320.64 & 17.85 & 25606 & 4 & 3\
299 & 2669.82 & 1334.75 & 18.88 & 100970 & 0 & -2\
303 & 3046.19 & 1343.52 & 19.71 & 97417 & -2 & -2\
304 & 4330.18 & 1348.34 & 17.29 & 23213 & 1 & 1\
314 & 1730.45 & 1364.59 & 20.88 & 97564 & -2 & -2\
316 & 1147.10 & 1373.25 & 20.47 & 99893 & 0 & 3\
325 & 946.39 & 1406.99 & 19.25 & 96686 & -5 & -5\
326 & 3531.82 & 1407.04 & 20.77 & 100281 & -2 & 0\
327 & 3089.78 & 1412.64 & 19.03 & 99330 & 1 & 1\
328 & 1340.92 & 1417.67 & 20.94 & 202745 & 0 & 3\
331 & 3058.10 & 1427.59 & 19.08 & 98481 & -2 & -2\
333 & 4153.37 & 1437.70 & 18.87 & 98104 & -5 & -5\
336 & 3083.01 & 1446.02 & 19.79 & 97627 & -2 & -2\
338 & 2994.05 & 1447.54 & 20.62 & - & 2 & 1\
352 & 869.48 & 1493.94 & 20.70 & 45442 & 3 & 3\
362 & 715.38 & 1525.08 & 20.21 & 99597 & 1 & 1\
368 & 2960.35 & 1547.05 & 19.92 & 99513 & 0 & 0\
382 & 697.67 & 1576.68 & 20.71 & 149212 & 6 & 5\
385 & 1150.11 & 1586.52 & 18.25 & 98976 & -4 & -5\
392 & 3917.25 & 1608.25 & 18.60 & 98541 & 3 & 3\
393 & 2331.21 & 1609.32 & 20.61 & 99237 & -2 & -2\
397 & 2423.80 & 1617.24 & 19.02 & 100430 & -2 & -2\
399 & 765.00 & 1624.56 & 20.76 & 159692 & 0 & -2\
415 & 2527.55 & 1680.94 & 19.65 & 98682 & -4 & -5\
433 & 2519.00 & 1709.59 & 21.00 & - & 6 & 4\
434 & 2712.22 & 1710.19 & 18.44 & 97642 & -5 & -5\
435 & 1048.97 & 1710.26 & 20.02 & 190882 & 3 & 2\
442 & 2657.45 & 1722.45 & 20.88 & 98607 & 1 & -2\
452 & 3575.83 & 1753.56 & 19.99 & 98086 & 1 & 0\
457 & 2202.26 & 1758.15 & 19.29 & 98688 & -2 & -3\
468 & 2524.04 & 1780.65 & 20.30 & 170632 & 2 & 1\
476 & 2164.48 & 1808.28 & 18.87 & 96773 & -4 & -5\
484 & 2281.67 & 1835.66 & 17.57 & 97168 & -5 & -5\
496 & 4657.77 & 1869.51 & 20.93 & - & -4 & -5\
499 & 926.31 & 1877.15 & 19.85 & 160364 & -5 & -5\
513 & 3385.76 & 1937.13 & 19.49 & 99860 & 0 & -2\
519 & 2633.79 & 1953.48 & 17.76 & 98691 & -4 & -5\
520 & 2267.84 & 1953.77 & 19.98 & 98961 & -5 & -5\
521 & 2598.87 & 1954.62 & 20.06 & 99504 & -2 & -2\
526 & 1819.56 & 1976.82 & 20.91 & - & -4 & -5\
529 & 2019.97 & 1981.74 & 20.50 & 118028 & 15 & -3\
530 & 2604.02 & 1985.11 & 19.63 & 99198 & -2 & -2\
533 & 3492.41 & 1992.67 & 19.56 & 97120 & -5 & -5\
541 & 2558.48 & 2016.50 & 20.55 & - & 0 & 1\
545 & 1341.00 & 2022.51 & 20.10 & 149243 & 0 & -4\
546 & 4213.68 & 2023.37 & 17.26 & 23768 & 5 & 4\
550 & 2819.78 & 2034.77 & 20.70 & 100970 & 1 & 0\
558 & 2046.76 & 2042.20 & 19.76 & 95753 & -2 & -2\
559 & 4563.94 & 2045.51 & 20.58 & - & 4 & 4\
561 & 4130.90 & 2050.21 & 19.46 & 207072 & 15 & 99\
562 & 2886.03 & 2061.80 & 18.43 & 69161 & 4 & 3\
563 & 2957.63 & 2065.25 & 20.21 & 100760 & 6 & 0\
564 & 3187.62 & 2070.07 & 19.06 & 97222 & 0 & -2\
570 & 3067.16 & 2074.98 & 20.66 & 99522 & -2 & -5\
581 & 2750.58 & 2107.99 & 19.12 & 98421 & -2 & -2\
591 & 2604.10 & 2133.40 & 18.36 & 97762 & -2 & -2\
602 & 2350.33 & 2175.38 & 20.38 & 94464 & -5 & -5\
606 & 2651.60 & 2182.44 & 19.05 & 96413 & -2 & -4\
611 & 1909.12 & 2186.08 & 18.89 & 97612 & -5 & -5\
623 & 4308.50 & 2227.50 & 19.93 & 99776 & 3 & 2\
624 & 2123.91 & 2231.85 & 19.46 & 52634 & 0 & -3\
629 & 5040.18 & 2242.21 & 17.88 & 76299 & 4 & 1\
636 & 2538.86 & 2258.28 & 18.75 & 97342 & -4 & -5\
639 & 2777.26 & 2264.35 & 20.40 & 98032 & -2 & -2\
641 & 4305.25 & 2279.18 & 20.95 & 23920 & 6 & 4\
643 & 2833.92 & 2289.38 & 20.50 & 148539 & -2 & -2\
647 & 2897.12 & 2311.32 & 20.21 & - & 6 & 5\
651 & 3624.13 & 2320.14 & 21.00 & 100652 & -2 & -1\
653 & 3096.27 & 2325.88 & 19.87 & 100251 & -2 & -4\
655 & 4925.76 & 2334.26 & 20.04 & 98116 & -2 & -2\
656 & 3032.45 & 2336.55 & 20.17 & 99708 & -2 & -2\
662 & 2762.21 & 2350.96 & 19.15 & 95205 & 1 & 1\
663 & 2002.63 & 2352.95 & 20.24 & 83425 & 0 & -5\
670 & 4792.86 & 2365.01 & 19.75 & 98125 & -2 & -2\
679 & 2891.70 & 2395.08 & 19.51 & 96653 & 0 & -5\
680 & 4456.98 & 2396.53 & 20.87 & 124924 & 2 & 3\
685 & 4144.73 & 2406.59 & 20.20 & - & -4 & -5\
698 & 2673.91 & 2444.21 & 20.95 & - & 3 & 4\
707 & 3017.95 & 2463.86 & 20.54 & - & 0 & -2\
711 & 2158.68 & 2471.78 & 19.83 & 96536 & -2 & -5\
712 & 2420.36 & 2476.45 & 19.69 & 99920 & 0 & 1\
713 & 2213.24 & 2477.51 & 20.55 & - & 5 & 4\
717 & 2719.31 & 2486.68 & 18.78 & 96452 & -2 & -2\
719 & 2172.06 & 2489.81 & 20.78 & 99339 & -5 & -5\
724 & 2281.36 & 2494.96 & 18.85 & 98302 & -5 & -5\
730 & 1712.08 & 2511.63 & 19.03 & 99830 & 3 & 1\
732 & 2705.21 & 2518.39 & 20.23 & 96773 & -2 & -2\
736 & 3434.65 & 2523.63 & 20.51 & 98451 & 0 & 0\
743 & 2741.21 & 2541.98 & 20.06 & 98272 & -5 & -5\
745 & 2878.57 & 2542.91 & 17.95 & 97033 & -4 & -5\
748 & 2835.90 & 2554.58 & 20.25 & 97048 & -5 & -5\
767 & 2784.63 & 2608.14 & 19.91 & 98302 & -5 & -5\
768 & 2566.24 & 2608.94 & 18.44 & 101929 & 1 & 1\
771 & 2851.49 & 2618.35 & 19.79 & 97462 & -5 & -2\
774 & 3465.80 & 2625.73 & 18.81 & 97342 & -2 & -5\
775 & 2818.67 & 2625.80 & 19.50 & 96824 & -5 & -5\
778 & 3368.31 & 2635.15 & 20.05 & 99351 & 1 & 1\
793 & 4179.71 & 2668.46 & 19.19 & 97036 & -4 & -2\
794 & 2810.14 & 2668.81 & 19.23 & 98514 & 2 & 1\
799 & 4028.08 & 2680.53 & 19.29 & 96161 & 0 & -4\
801 & 1766.29 & 2685.75 & 18.63 & 96038 & 1 & 2\
803 & 3671.86 & 2687.29 & 18.34 & 99702 & 1 & 0\
810 & 2801.93 & 2719.27 & 17.11 & 98182 & -5 & -5\
813 & 3341.70 & 2731.51 & 19.51 & 96383 & -2 & -2\
822 & 4639.03 & 2748.37 & 19.05 & 96997 & 5 & 6\
828 & 2521.07 & 2770.37 & 19.91 & 97912 & 6 & 6\
831 & 2801.07 & 2781.44 & 18.74 & 97672 & -4 & -5\
838 & 2248.16 & 2789.85 & 20.14 & - & 999 & -5\
842 & 2906.30 & 2794.44 & 20.39 & 100670 & 999 & 0\
843 & 4713.68 & 2796.95 & 20.55 & 48385 & 7 & 7\
846 & 2774.56 & 2803.87 & 19.72 & 96563 & -5 & -5\
867 & 2612.37 & 2853.73 & 18.37 & 97042 & -4 & -2\
872 & 3640.91 & 2868.48 & 20.90 & - & -2 & -1\
877 & 1235.86 & 2875.59 & 20.67 & 121526 & 1 & 0\
886 & 2138.47 & 2909.08 & 20.20 & 95723 & 8 & 9\
887 & 1185.86 & 2912.30 & 20.47 & - & 999 & 999\
888 & 1978.03 & 2915.65 & 19.48 & 96563 & 0 & -2\
889 & 3904.51 & 2916.67 & 20.36 & 97801 & 0 & 1\
890 & 3514.41 & 2918.22 & 20.68 & 97564 & -4 & -2\
898 & 3185.08 & 2942.27 & 20.62 & - & -5 & -1\
900 & 3332.43 & 2943.17 & 20.63 & 97732 & -2 & -2\
905 & 1591.54 & 2966.15 & 20.91 & 98451 & -5 & -5\
909 & 3911.90 & 2973.01 & 20.31 & 187822 & 1 & 1\
912 & 1895.44 & 2979.65 & 20.09 & 98751 & -5 & -3\
913 & 2748.21 & 2980.39 & 19.26 & 95816 & -2 & -2\
922 & 3294.59 & 3021.22 & 19.65 & 98931 & -2 & -2\
927 & 2019.29 & 3026.93 & 19.25 & 95274 & -5 & -5\
929 & 3951.51 & 3033.18 & 20.10 & 96263 & 0 & -1\
933 & 4515.25 & 3035.84 & 19.87 & - & -4 & -5\
938 & 1691.42 & 3049.42 & 19.25 & 71440 & -4 & 0\
942 & 2829.86 & 3058.61 & 20.59 & 96428 & -2 & -5\
943 & 3816.51 & 3058.75 & 19.94 & - & -2 & -2\
944 & 2420.93 & 3065.24 & 20.94 & - & -5 & -5\
948 & 4108.78 & 3075.68 & 20.49 & - & -5 & -4\
961 & 2478.92 & 3095.09 & 19.79 & 99381 & -5 & -5\
967 & 3606.49 & 3104.33 & 20.10 & - & 4 & 1\
971 & 3131.00 & 3106.51 & 19.94 & 47565 & 5 & 3\
975 & 3024.81 & 3120.83 & 20.95 & - & 99 & 99\
978 & 2832.26 & 3123.45 & 20.24 & - & -4 & -5\
993 & 3323.02 & 3154.25 & 19.05 & 97735 & 0 & -1\
1003 & 2815.02 & 3176.78 & 20.76 & 100763 & -5 & -5\
1012 & 2910.90 & 3204.76 & 20.77 & - & -2 & -4\
1019 & 1598.82 & 3220.61 & 20.02 & 98718 & 0 & 1\
1020 & 2661.84 & 3222.75 & 18.46 & 83578 & 3 & 1\
1027 & 2954.26 & 3244.74 & 20.28 & 96686 & 0 & -2\
1036 & 4346.84 & 3277.98 & 19.22 & 99165 & -2 & -2\
1041 & 4453.92 & 3289.61 & 19.78 & 98649 & -2 & -2\
1042 & 3771.63 & 3293.33 & 20.93 & - & -2 & 1\
1045 & 4485.09 & 3295.86 & 19.95 & 44686 & 99 & 99\
1049 & 2469.20 & 3302.45 & 20.58 & - & 0 & 0\
1062 & 3268.23 & 3338.69 & 18.84 & 98182 & 0 & -1\
1064 & 4749.89 & 3342.41 & 19.45 & 96182 & 5 & 2\
1069 & 1366.71 & 3359.61 & 19.95 & 99060 & 0 & -2\
1075 & 4869.37 & 3367.38 & 19.90 & 62244 & 1 & -2\
1080 & 2041.26 & 3397.59 & 20.86 & 98421 & -5 & -5\
1091 & 3315.98 & 3426.51 & 20.51 & 99579 & -2 & 0\
1100 & 4416.75 & 3462.54 & 20.73 & 159095 & 15 & 99\
1105 & 2881.63 & 3475.46 & 18.79 & 98323 & -2 & -2\
1108 & 2836.73 & 3485.88 & 19.42 & 99081 & 1 & -1\
1121 & 1960.44 & 3512.66 & 20.68 & - & -4 & -5\
1127 & 2942.66 & 3523.15 & 20.76 & 159362 & 5 & 3\
1141 & 4517.75 & 3545.75 & 19.93 & 99657 & -2 & -2\
1146 & 4117.26 & 3561.40 & 18.27 & 100080 & -5 & -5\
1153 & 2566.37 & 3583.80 & 20.14 & 61935 & 9 & 99\
1157 & 2487.08 & 3591.79 & 18.64 & 62207 & 9 & 5\
1158 & 3279.02 & 3592.19 & 20.16 & 99935 & -2 & -2\
1161 & 2923.10 & 3597.44 & 20.01 & - & -5 & -5\
1163 & 1348.04 & 3601.03 & 20.12 & 144364 & 1 & 3\
1177 & 2091.53 & 3633.12 & 20.39 & 216269 & 15 & 1\
1178 & 2361.82 & 3635.18 & 20.30 & 99234 & -2 & -1\
1192 & 3212.40 & 3665.38 & 20.39 & 97375 & -5 & -5\
1200 & 2024.55 & 3679.35 & 20.59 & - & 5 & 7\
1207 & 3978.09 & 3693.67 & 20.68 & 100221 & -4 & -2\
1209 & 3395.58 & 3703.43 & 19.10 & 83882 & 1 & 1\
1211 & 2923.22 & 3706.89 & 20.79 & 99177 & -4 & -5\
1214 & 3909.03 & 3714.53 & 19.51 & 100323 & 1 & 1\
1223 & 2310.92 & 3752.63 & 20.49 & - & -5 & -5\
1231 & 3481.26 & 3785.53 & 20.32 & 99018 & -2 & 0\
1235 & 2884.78 & 3791.48 & 20.67 & - & -4 & -2\
1239 & 2759.52 & 3812.84 & 20.00 & 62264 & 3 & 2\
1240 & 2878.69 & 3813.83 & 20.24 & 122164 & 4 & 7\
1253 & 995.43 & 3851.25 & 19.37 & 97093 & 2 & 4\
1255 & 2908.32 & 3863.10 & 20.54 & - & 0 & 3\
1258 & 2180.86 & 3869.70 & 20.60 & - & -5 & -5\
1265 & 3358.24 & 3888.19 & 19.60 & 99896 & -2 & -2\
1272 & 3738.88 & 3907.25 & 20.85 & 99501 & -2 & -1\
1273 & 2182.60 & 3907.71 & 20.83 & - & -5 & 0\
1275 & 2223.31 & 3912.98 & 20.08 & 99141 & -2 & -1\
1285 & 1476.03 & 3937.66 & 19.84 & 188687 & 1 & 1\
1287 & 2727.83 & 3947.24 & 20.20 & - & -2 & -2\
1297 & 2476.43 & 3965.12 & 20.88 & 97102 & -2 & -2\
1298 & 2291.46 & 3965.96 & 20.79 & - & -2 & -2\
1299 & 1553.40 & 3967.72 & 20.25 & 98964 & 3 & 3\
1301 & 1428.04 & 3978.07 & 19.40 & 99342 & 0 & -2\
1303 & 3180.35 & 3982.62 & 18.98 & 96901 & 0 & -2\
1308 & 1799.52 & 3992.55 & 20.30 & - & 5 & 3\
1322 & 1469.20 & 4032.65 & 20.90 & - & -2 & -2\
1330 & 2626.63 & 4050.56 & 19.91 & 97051 & 15 & 10\
1335 & 3007.32 & 4063.99 & 20.39 & - & -4 & -4\
1343 & 2119.25 & 4082.59 & 17.77 & 99156 & -5 & -5\
1357 & 1560.05 & 4129.23 & 19.78 & 97000 & -5 & -2\
1367 & 2972.68 & 4167.09 & 18.02 & 98299 & -5 & -5\
1370 & 3118.58 & 4169.24 & 19.71 & 97738 & -2 & -5\
1372 & 3911.67 & 4178.76 & 20.50 & 98472 & -5 & -5\
1374 & 3888.45 & 4183.67 & 20.52 & 97699 & -2 & -2\
1376 & 3489.91 & 4187.61 & 19.91 & 98275 & -2 & -2\
1377 & 2964.53 & 4192.76 & 19.23 & 97975 & -5 & -5\
1379 & 3412.68 & 4193.30 & 20.93 & - & 0 & 2\
1382 & 2988.76 & 4209.19 & 20.23 & - & 3 & 4\
1387 & 2761.87 & 4222.14 & 19.18 & 99417 & -5 & -5\
1395 & 848.83 & 4239.90 & 19.39 & 99771 & -2 & 1\
1405 & 3807.69 & 4288.35 & 20.12 & 99285 & -4 & -5\
1409 & 3951.25 & 4295.39 & 19.69 & 54964 & -2 & -2\
1411 & 2500.75 & 4322.96 & 19.74 & - & 999 & -2\
1413 & 3281.08 & 4328.24 & 19.37 & 99774 & 2 & 2\
1414 & 1303.46 & 4329.27 & 20.65 & 95615 & 5 & 3\
1415 & 288.53 & 4331.63 & 18.70 & 98844 & -5 & -5\
1423 & 3633.52 & 4361.01 & 20.56 & 98799 & -2 & -2\
1430 & 4193.88 & 4380.15 & 20.61 & 99222 & 999 & 3\
1431 & 3125.60 & 4381.56 & 20.86 & - & -2 & -4\
1438 & 574.34 & 4410.17 & 18.72 & 81105 & 2 & 1\
1439 & 3020.74 & 4422.23 & 19.27 & 99339 & 0 & -5\
1447 & 755.69 & 4452.72 & 20.97 & 171424 & 3 & 4\
1454 & 3375.34 & 4469.90 & 20.28 & 100859 & -2 & -3\
1461 & 4207.84 & 4489.60 & 20.98 & 97279 & -2 & -5\
1466 & 3188.31 & 4501.44 & 20.62 & - & 5 & 5\
1481 & 810.82 & 4540.51 & 20.61 & 99791 & -2 & 2\
1483 & 825.24 & 4542.01 & 19.36 & 99791 & -5 & -5\
1488 & 661.75 & 4553.49 & 19.55 & 80904 & -2 & -4\
1524 & 943.52 & 4664.34 & 20.89 & 190098 & 6 & 5\
1529 & 3450.04 & 4681.31 & 20.81 & 114632 & 15 & 10\
1540 & 3674.58 & 4731.74 & 18.01 & 21317 & 5 & 6\
1541 & 2265.25 & 4732.11 & 20.43 & 98778 & -5 & -5\
1559 & 3242.72 & 4783.73 & 20.72 & 100380 & 1 & 1\
1563 & 4731.77 & 4795.22 & 20.88 & - & -2 & -1\
1567 & 768.30 & 4815.43 & 20.88 & 100284 & 99 & 10\
1568 & 4014.76 & 4819.25 & 19.46 & 36222 & 99 & -5\
1571 & 2338.11 & 4825.66 & 19.97 & 83803 & 4 & 5\
1588 & 4185.04 & 4870.66 & 20.95 & 126167 & 0 & -4\
1594 & 1852.07 & 4887.76 & 19.87 & - & 999 & 999\
1621 & 1467.74 & 4980.41 & 20.03 & - & -5 & 999\
1623 & 1937.90 & 4990.12 & 20.86 & 187086 & -2 & -2\
1634 & 3158.57 & 5037.47 & 19.28 & 31985 & 9 & 5\
1664 & 2256.63 & 5240.59 & 20.61 & 248908 & 5 & 2\
1666 & 2243.04 & 5270.84 & 19.15 & 99552 & 1 & 2\
[cccccc]{} E & 27$\pm$4 & 35$\pm$5 & 23$\pm$2\
S0 & 44$\pm$6 & 38$\pm$5 & 49$\pm$3\
S & 29$\pm$5 & 27$\pm$4 & 27$\pm$2\
[ccccc]{} E & 46 & 1.5 & 0.5 & 0\
S0 & 76.5 & 2.5 & 1.5 & 7\
Sa-Sb & 26.5 & 5 & 2 & 4\
Sbc-Irr& 3 & 4 & 6 & 0\
Merger & 0 & 1 & 1 & 0\
? & 3 & 0 & 0 & 0\
[^1]: Present address: California Institute of Technology, MS 105-24, Pasadena, CA 91125
[^2]: Based on observations with the NASA/ESA Hubble Space Telescope and the W. M. Keck Observatory
[^3]: It is important to account for the morphology-density relation when we consider the morphological content of clusters. At high galaxy densities, typically found in the cores of clusters, the low-$z$ reference population becomes increasingly dominated by E’s, [@dr97]
[^4]: Throughout the paper we consistently refer to a Cousins $I$ magnitude, which is very close to the natural F814W system referred to a Vega zeropoint ([@ho95])
[^5]: The MDS catalog is based on observations with the NASA/ESA HST, obtained at the STSCI, operated by AURA.
[^6]: In one case, a spectrum was obtained of a pair of galaxies separated by 0.7$^{\prime\prime}$, one (\#1295) at $I=21.17$ and the other (\#1297) at $I=20.92$. We have assigned the measured velocity (cz=99657) to both galaxies. Galaxies \#1481 and \#1483 may have both fallen within the spectrograph slit. We assign the measured velocity (cz=99792) to \#1483, which is 1.2 mag brighter.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Ab-initio fully differential cross-sections for electrons scattering in liquid xenon are developed from a solution of the Dirac-Fock scattering equations, using a recently developed framework [@Boyletal15] which considers multipole polarizabilities, a non-local treatment of exchange, and screening and coherent scattering effects. A multi-term solution of Boltzmann’s equation accounting for the full anisotropic nature of the differential cross-section is used to calculate transport properties of excess electrons in liquid xenon. The results were found to agree to within $25$% of the measured mobilities and characteristic energies over the reduced field range of $10^{-4}$–$1$ Td. The accuracies are comparable to those achieved in the gas phase. A simple model, informed by highly accurate gas-phase cross-sections, is presented to improve the liquid cross-sections, which was found to enhance the accuracy of the transport coefficient calculations.'
author:
- 'G. J. Boyle'
- 'R. P. McEachran'
- 'D. G. Cocks'
- 'M. J. Brunger'
- 'S. J. Buckman'
- 'S. Dujko'
- 'R. D. White'
bibliography:
- 'library\_RW.bib'
- 'library.bib'
- 'library\_extra.bib'
- 'ThesisFinal.bib'
title: 'Ab-initio electron scattering cross-sections and transport in liquid xenon'
---
\#1
Introduction
============
An understanding of the behavior of free electrons in liquids and dense systems is of interest to both fundamental physics research and to technological applications [@Boyletal15]. In particular, liquid-phase noble gases are used in high-energy particle detectors and at present several LAr (liquid argon) and LXe (liquid xenon) TPCs (time projection chambers) have been built for dark matter searches [@Regenfus2010; @Aprile2010; @Aprile2011g; @Alner2007g; @baller2014liquid], neutrino detection [@ICARUS2004; @ICARUS2011; @baller2014liquid; @Aprile2010], and have also been used in high-energy beam-line experiments [@ATLAS2014; @baller2014liquid]. Optimizing the performance of these liquid TPC particle detectors requires an accurate understanding of electron drift and diffusion in noble liquids subject to electric fields. In a previous paper, Boyle et al. [@Boyletal15], we investigated the elastic scattering of electrons from gas-phase and liquid-phase argon. In this paper, we extend the previous discussion to consider elastic scattering of electrons in gas-phase and liquid-phase xenon, using the same techniques previously outlined [@Boyletal15].
The study of excess electrons in dense gases and liquids involves many effects that are not significant in dilute gaseous systems. When the de Broglie wavelength of the electrons (near thermal energies) is comparable to the interatomic spacing of the medium, scattering occurs off multiple scattering centres simultaneously, rather than through binary scattering. Furthermore, these scattering centres are highly correlated in space and time. Historically, transport simulations neglected these correlations and simply scaled calculations in the dilute gas phase to higher densities. It has been shown [@Boyletal15] that this simplistic approach cannot explain the non-linearities seen in the experiments . As we have described in Boyle et al. [@Boyletal15], there are other theoretical approaches to exploring the effect of liquid correlations on the transport of light particles [@GeeFree86; @Braglia82; @Borghesani2006; @Sakai2007; @White2009; @Kunhardt1991], however these either require empirical inputs, are applicable only close to equilibrium, or have heuristically combined the liquid effects identified above to obtain an effective cross-section.
As also discussed in Boyle et al. [@Boyletal15], we employ the *ab initio* procedure of Cohen and Lekner [@Cohen1967] using updated scattering theory to address transport in dense systems under a kinetic theory framework. Atrazhev et al. [@Atrazhev1981] have used a similar procedure based on a simplification of the Cohen and Lekner theory where they argued that, for small energies, the effective cross-section becomes dependent on the density only and they obtained good agreement with experiment. However, the distance at which to enforce this new behaviour of the effective cross-section remains a free parameter in the theory and this constant effective cross-section must be found empirically. By performing a detailed analysis of the partial phase shifts, Atrazhev and co-workers [@Atrazhev1996] were able to isolate the important properties of the potential which are required for accurate determination of the transport properties. Our calculations instead avoid these difficulties by using accurate forms for the electron-atom interaction. The transport theory employed in this manuscript also represents an improvement over previous calculations as we use a full multi-term treatment of the velocity distribution function [@White2009; @White2011] which utilizes all the available anisotropic details of the scattering cross-sections. It is well known that the often-used two-term approximation for transport calculations can be in serious error [@White2003a], and we compare our solutions for transport in the xenon system using two-term and multi-term treatments.
In the following sections we consider the calculation of the macroscopic swarm transport properties, in the gaseous and liquid xenon environments, from the microscopic cross-sections, modified by the screening and coherent scattering effects discussed above. We first discuss a multi-term solution of Boltzmann’s equation in Section \[TransportTheory-1\], using the calculation of the elastic scattering cross-section for electrons in dilute gaseous xenon via the Dirac-Fock scattering equations in Section \[ScatteringGas\]. We determine the xenon pair-correlator at the xenon triple point in Section \[PairCorrelation\] using a Monte-Carlo simulation of a Lennard-Jones with parameters fitted from experimental data in the gas phase and compare the results with previous calculations. The pair-correlator allows us to determine an effective liquid scattering cross-section for xenon (see Section \[sub:Liquid-xenon-cross-section\]), in the manner outlined in Boyle et al. [@Boyletal15]. The application of these cross-sections to determine macroscopic transport properties via kinetic theory is also outlined in Section \[TransportTheory-1\]. We present the results of our transport calculations in Section \[sec:Results\]. Initially in Section \[sub:Electrons-in-gaseous\] we consider only electrons in gas-phase xenon, and we compare the reduced mobility and characteristic energies with swarm experiment measurements for a range of fields. Electrons in liquid-phase xenon are then considered in Section \[sub:Electrons-in-liquid\]. The impact of anisotropic scattering and validity of the two-term approximation is also investigated, and we highlight the importance of using a multi-term framework when solving Boltzmann’s equation accurately. Finally, we discuss a scaling procedure to adapt other gas-phase cross-section sets, such as those compiled from various theoretical and experimental sources, to the liquid phase in Section \[sub:Electrons-in-liquid-2\], and compare the transport properties. Throughout this paper we will make use of atomic units ($m_{e}=e=a_{0}=\hbar=1$) unless otherwise specified.
Multi-term solution of Boltzmann’s equation\[TransportTheory-1\]
================================================================
A dilute swarm of electrons moving through gaseous or liquid xenon, subject to an external electric field $E$, can be described by the solution of the Boltzmann’s equation for the phase-space distribution function $f(\vc r,\vc v,t)$ [@Boltzmann1872]: $$\frac{\partial f}{\partial t}+\vc v\cdot\nabla f+\frac{e\vc E}{m_{e}}\cdot\frac{\partial f}{\partial\vc v}=-J(f),\label{eq:BE}$$ where $\vc r,$ $\vc v$ and $e$ denote the position, velocity and charge of the electron respectively. The collision operator $J(f)$ accounts for all the necessary collision types and interactions between the electrons of mass $m_{e}$ and the background medium. In this paper we will only consider reduced electric field strengths $E/N$ (where $N$ is the number density of the background material) such that there is no significant contribution from excitation collisions in xenon.
In swarm experiments only a few macroscopic variables can be controlled and/or measured [@HuxlCrom74]. The most commonly reported transport quantities for electrons in liquids are the mobility $\mu$ ($=W/E$, where $W$ is drift velocity), and the transverse and longitudinal characteristic energies, $D_{T}/\mu$ and $D_{L}/\mu$ respectively. As shown in [@Boyletal15], we can calculate these coefficients through multi-term spherical harmonic representations of the necessary velocity distribution functions:
$$F(\vc v)=\sum_{l=0}^{\infty}\, F_{l}(v)P_{l}^{0}(\cos\theta)\;,\label{sphericalharmonic-1-1}$$
$$F^{(L)}(\vc v)=\sum_{l=0}^{\infty}\, F_{l}^{(L)}(v)P_{l}(\cos\theta)\;,\label{sphericalharmonic-1-1-1}$$
and
$$F^{(T)}(\vc v)=\sum_{l=0}^{\infty}\, F_{l}^{(T)}(v)P_{l}^{1}(\cos\theta)\;.\label{sphericalharmonic-1-1-1-1}$$
Note that $\theta$ denotes the angle relative to the electric field direction (taken to be the $z$-axis) and $P_{l}^{m}(\cos\theta)$ are associated Legendre polynomials. The Boltzmann equation can be re-written as the following hierarchy of equations for these expansion coefficients: $$J^{l}F_{l}+\frac{l+1}{2l+3}a\left(\frac{\partial}{\partial v}+\frac{l+2}{v}\right)F_{l+1}+\frac{l}{2l-1}a\left(\frac{\partial}{\partial v}-\frac{l-1}{v}\right)F_{l-1}=0,\label{eq:homogenous}$$
$$J^{l}F_{l}^{(L)}+\frac{l+1}{2l+3}a\left(\frac{\partial}{\partial v}+\frac{l+2}{v}\right)F_{l+1}^{(L)}+\frac{l}{2l-1}a\left(\frac{\partial}{\partial v}-\frac{l-1}{v}\right)F_{l-1}^{(L)}=v\left(\frac{l+1}{2l+3}F_{l+1}+\frac{l}{2l-1}F_{l-1}\right),\label{eq:longitudinal}$$
and
$$J^{l}F_{l}^{(T)}+\frac{l+2}{2l+3}a\left(\frac{\partial}{\partial v}+\frac{l+2}{v}\right)F_{l+1}^{(T)}+\frac{l-1}{2l-1}a\left(\frac{\partial}{\partial v}-\frac{l-1}{v}\right)F_{l-1}^{(T)}=v\left(\frac{F_{l-1}}{2l-1}-\frac{F_{l+1}}{2l+3}\right),\label{eq:Transverse}$$
where the $J^{l}$ represent the Legendre projections of the collision operator detailed below and $a=eE/m_{e}$. We enforce the normalisation condition: $$4\pi\int_{0}^{\infty}F_{0}(v)v^{2}dv=1\;.$$ The solution of this hierarchy of equations then yields the mobility: $$\mu=\frac{1}{E}\frac{4\pi}{3}\intop_{0}^{\infty}v^{3}F_{1}dv,$$ and the characteristic energy:
$$D_{\left(L,T\right)}=\frac{4\pi}{3}\intop_{0}^{\infty}v^{3}F_{1}^{(L,T)}dv\;.$$
Note, this theory avoids the traditional two-term approximation used in electron transport in liquids [@Cohen1967; @Borghesani2006; @Sakai2007a], and is a true multi-term solution of Boltzmann’s equation, whereby the upper bound in each of the $l$-summations is truncated at a value $l_{\mathrm{max}}$, and this value is incremented until some convergence criteria is met on the distribution function or its velocity moments.
We only consider low-energy elastic scattering in this study. Hence, the collision operator appearing in (\[eq:BE\]), which describes the rate of change of the distribution function due to interactions with the background material, will include elastic collisions only. For the liquid systems considered here, the de Broglie wavelength of the electron is often of the order of the average inter-particle spacing $\sim N^{-1/3}$. In this energy regime, the electron is best viewed as a wave that simultaneously interacts with multiple scattering centres that comprise the medium. For liquid xenon, the average interparticle spacing is approximately $2.6\,{\ifmmode\begingroup\def\b@ld{bold}
\text{\ifx\math@version\b@ld\bfseries\fi\AA}\endgroup\else\AA\fi}$, implying that “low” energies are those less than $\sim0.5$ eV, which is several orders of magnitude larger than the thermal energy of $\sim0.014$ eV. The interaction potential is then modified from the dilute gas phase as discussed in Section \[sub:Liquid-xenon-cross-section\].
The Legendre projections of the elastic collision operator, in the small mass ratio limit, accounting for coherent scattering, are given by:
$$J^{0}\left({\Phi_{l}}\right)=\frac{m_{e}}{Mv^{2}}\frac{d}{dv}\left\{ v\nu_{1}(v)\left[v{\Phi_{l}}+\frac{kT}{m_{e}}\frac{d}{dv}{\Phi_{l}}\right]\right\} \label{eq:Davydov}$$
$$J^{l}{\Phi_{l}}=\tilde{\nu}_{l}(v){\Phi_{l}}\qquad\text{for \ensuremath{l}\ensuremath{\ge}1},\label{eq:structurecollisionfreq}$$
where $M$ is the mass of a xenon atom, $\Phi_{l}=\left\{ F_{l},F^{(L)},F^{(T)}\right\} $ and
$$\nu_{l}(v)=Nv2\pi\int_{0}^{\pi}\sigma(v,\chi)\left[1-P_{l}(\cos\chi)\right]\sin\chi d\chi,\label{eq:collisionfreq_gas}$$
is the binary transfer collision frequency in the absence of coherent scattering effects with $\sigma(v,\chi)$ the differential scattering cross-section. In addition,
$$\tilde{\nu}_{l}(v)=Nv\left(2\pi\int_{0}^{\pi}\Sigma(v,\chi)\left[1-P_{l}(\cos\chi)\right]\sin\chi d\chi\right).\label{eq:collisionfreqs-1}$$
are the structure-modified higher-order collision frequencies that account for coherent scattering through:
$$\Sigma(v,\chi)=\sigma(v,\chi)\; S\left(\frac{2m_{e}v}{\hbar}\sin\frac{\chi}{2}\right),$$
which represents an effective differential cross-section. $S$ is the static structure factor, which can be determined from the pair-correlator as discussed in Section \[sub:Liquid-xenon-cross-section\]. In what follows we also define the momentum transfer cross-sections without and with coherent scattering via $\nu_{1}(v)=Nv\sigma_{m}(v)$ and $\tilde{\nu}_{1}(v)=Nv\Sigma_{m}(v)$, respectively.
Scattering of electrons by xenon gas\[ScatteringGas\]
=====================================================
The theoretical procedures used in this paper, to describe the elastic scattering of electrons from xenon atoms at low energies, are essentially the same as those used in Boyle et al. [@Boyletal15] for electron scattering from argon. We thus present only a short summary here and refer the reader to reference [@Boyletal15] for more details.
In the pure elastic scattering energy region, only two interactions need to be considered, namely polarization and exchange. The polarization interaction is accounted for by means of a long-range multipole polarization potential, while the exchange interaction is represented by a short-range non-local potential formed by antisymmetrizing the total scattering wavefunction. The scattering of the incident electrons, with wavenumber $k$, by xenon atoms can then be described in the gaseous phase by the integral equation formulation of the partial wave Dirac-Fock scattering equations (see Chen et al. [@Chen08] for details).
In matrix form, these equations can be written as:
$$\left(\begin{array}{c}
f_{\kappa}(r)\\
g_{\kappa}(r)
\end{array}\right)=\begin{pmatrix}\begin{array}{c}
v_{1}(kr)\\
v_{2}(kr)
\end{array}\end{pmatrix}+\frac{1}{k}\int_{0}^{r}\mathrm{d}x\, G(r,x)\,\biggl[U(x)\begin{pmatrix}\begin{array}{c}
f_{\kappa}(x)\\
g_{\kappa}(x)
\end{array}\end{pmatrix}-\begin{pmatrix}\begin{array}{c}
W_{Q}(\kappa;x)\\
W_{P}(\kappa;x)
\end{array}\end{pmatrix}\biggr],\label{eq:DiracFock}$$
where the local potential $U(r)$ is given by the sum of the static and local polarization potentials i.e., $$U(r)=U_{\mathrm{s}}(r)+U_{\mathrm{p}}(r)\label{eq:potential}$$ and $W_{P}(\kappa;r)$ and $W_{Q}(\kappa;r)$ represent the large and small components of the exchange interaction.
The precise form of these exchange terms is given in equation (5) of Boyle et al. [@Boyletal15]. In particular, the polarization potential $U_{\mathrm{p}}(r)$ in equation (\[eq:potential\]) was determined using the polarized orbital method [@mceachran77] and contained several static multipole terms as well as the corresponding dynamic polarization term [@McEachran90; @Mimnagh93]. In total, the potential $U(r)$ contained all terms up to and including those that behave as $r^{-16}$ asymptotically.
In equation (\[eq:DiracFock\]), $f_{\kappa}(r)$ and $g_{\kappa}(r)$ are the large and small components of the scattering wavefunction. Here the quantum number $\kappa$ is related to the total and orbital angular momentum quantum numbers $j$ and $l$ according to $\kappa=-l-1$ when $j=l+1/2$ (spin-up) and $\kappa=l$ when $j=l-1/2$ (spin down). Furthermore, $G(r,x)$ is the free particle Green’s function given in equations (23) and (24a,b) of Chen et al. [@Chen08]. The kinetic energy $\epsilon$ of the incident electron and its wavenumber $k$ are related by: $$k^{2}=\frac{1}{\hbar^{2}c^{2}}\,\epsilon\,\bigl(\epsilon+2mc^{2}\bigr),\label{eq:kineticenergy}$$ where $c$ is the velocity of light which, in atomic units, is given by $c=1/\alpha$ where $\alpha$ is the fine-structure constant.
In the integral equation formulation, the scattering phase shifts can be determined from the asymptotic form of the large component of the scattering wavefunction i.e., $$f_{\kappa}(r)_{r\rightarrow\infty}\kern-15pt \longrightarrow A_{\kappa}\,\hat{j}_{l}(kr)-B_{\kappa}\,\hat{n}_{l}(kr),\label{eq:wavefunction}$$ where $\hat{j}_{l}(kr)$ and $\hat{n}_{l}(kr)$ are the Riccati-Bessel and Riccati-Neumann functions while the constants $A_{\kappa}$ and $B_{\kappa}$ are defined in equations (7) and (8) of Boyle et al. [@Boyletal15] . The partial wave phase shifts are then given by: $$\tan\delta^{\pm}(k)=\frac{B_{\kappa}}{A_{\kappa}},\label{eq:partials}$$ where the $\delta^{\pm}$ are the spin-up $(+)$ and spin-down $(-)$ phase shifts. The total elastic and momentum transfer cross-sections are given, in terms of these phase shifts, by: $$\sigma_{{\rm el}}(k^{2})=\frac{4\pi}{k^{2}}\sum_{l=0}^{\infty}\Bigl\{(l+1)\,\sin^{2}\delta_{l}^{+}(k)+l\,\sin^{2}\delta_{l}^{-}(k)\Bigr\}\label{eq:elastic}$$ and $$\begin{gathered}
\sigma_{{\rm mt}}(k^{2})=\frac{4\pi}{k^{2}}\sum_{l=0}^{\infty}\Bigl\{\frac{(l+1)(l+2)}{2l+3}\,\sin^{2}\bigl(\delta_{l}^{+}(k)-\delta_{l+1}^{+}(k)\bigr)+\frac{l(l+1)}{2l+1}\,\sin^{2}\bigl(\delta_{l}^{-}(k)-\delta_{l+1}^{-}(k)\bigr)\\
+\frac{(l+1)}{(2l+1)(2l+3)}\,\sin^{2}\bigl(\delta_{l}^{+}((k)-\delta_{l+1}^{-}(k)\bigr)\Bigr\}.\label{eq:MT}\end{gathered}$$ Analytic fits of the above momentum-transfer cross-section (and for other noble gases) are given in McEachran and Stauffer 2014 [@McEcStau14] to aid in plasma modelling calculations.
In Figure \[fig:Liquid-Xe-cross-sections-1\], the momentum transfer cross-section in the gas-phase is compared to the cross-section recommended by Biagi [@lxcatbiagi14a] from the “Magboltz” Boltzmann equation solver. The Biagi elastic momentum transfer cross-section has been constructed from the unpublished analysis of Elford, fitting to the available drift velocity and diffusion coefficients [@Bordetal13]. It is often considered the reference cross-section for electron-xenon interactions in the gas phase. Our momentum transfer cross-section gives good qualitative agreement with the Biagi reference cross-section, but generally somewhat over-estimates the value. The location and depth of the Ramsauer minima, however, agrees closely.
Scattering of electrons by xenon liquid
=======================================
Xenon pair correlation\[PairCorrelation\]
-----------------------------------------
The only measurements of the liquid phase xenon structure factor that are known to us, are by Becchi and Magli [@Becchi1997] for $T=274.7$ K and $N=8.86\times10^{21}\,\mathrm{cm}^{-3}$, near to the critical point of $T=289.72$ K. In order to obtain a structure factor at the lower temperature of $T=165$ K and $N=14.2\times10^{22}\,\mathrm{cm}^{-3}$, where the transport measurements have been performed, we have used a Lennard-Jones model, calibrated to the high temperature structure factor, to construct a low temperature structure factor by performing Monte-Carlo simulations of the fluid.
The Lennard-Jones fluid has a pair-wise potential given by: $$V(r)=4\epsilon_{LJ}\left[\left(\frac{r}{\sigma_{LJ}}\right)^{12}-\left(\frac{r}{\sigma_{LJ}}\right)^{6}\right],$$ with the characteristic energy $\epsilon_{LJ}$ and length scale $\sigma_{LJ}$. Often, this potential is truncated to a range of $r_{\mathrm{trunc}}=2.5\,\sigma_{LJ}$ but we work with an effectively untruncated potential by extending $r_{\mathrm{trunc}}$ to the system size $r_{\mathrm{trunc}}=L/2$. For the truncated potential, Atrazhev et al [@Atrazhevetal05] have used values of $\epsilon_{LJ}/k_{B}=299\,\mathrm{K}$ and $\sigma_{LJ}=4.05\,\AA$ but, by comparing to the experimental data of reference [@Becchi1997], we find a better fit for our untruncated model by matching the critical temperature of the Lennard-Jones model ($k_{B}T_{\mathrm{crit}}=1.312\,\epsilon_{LJ}$ [@Perez-Pellitero2006]) to the measured value of $289.72$ K leaving us instead with $\epsilon_{LJ}/k_{B}=220.83$ K. Given this value, the best fit to the data of reference [@Becchi1997] is $\sigma_{LJ}=3.86\,\AA$. We then use these parameters to obtain the pair-correlator at the desired temperature of $T=165$ K. A plot of the various structure factors is shown in figure \[fig:structure\_factor\], where we compare with the structure factor calculated from the truncated potential in the paper of Atrazhev et al. [@Atrazhevetal05].
Liquid xenon cross-section\[sub:Liquid-xenon-cross-section\]
------------------------------------------------------------
In order to capture the major effects of increasing the number density of the fluid system, we include two modifications to the gas scattering potentials due to: a) the screening of the long-range polarization potential, and b) the influence of the particles in the medium bulk. The procedure outlined in this section closely follows that of references [@Cohen1967; @Boyletal15]. We summarize the steps here.
Firstly, the effective charge-multipole polarization potential acting between the electron and the induced multipole of an individual atom is reduced by the presence of the induced multipoles in the surrounding atoms, which produces a screening effect. Using the (isotropic) pair-correlator for xenon, $g(r)$, determined in Section \[PairCorrelation\], we have self-consistently calculated the screening function $f(r)$ for an electron located at position $r$ via:
$$f(r)=1-\pi N\int_{0}^{\infty}ds\ \frac{g(s)}{s^{2}}\int_{\left|r-s\right|}^{r+s}dt\ \Theta\left(r,s,t\right)\frac{\alpha(t)f(t)}{t^{2}},$$
which has been represented in bipolar coordinates, $s$ and $t$, and where $$\Theta\left(r,s,t\right)=\frac{3}{2}\frac{\left(s^{2}+t^{2}-r^{2}\right)\left(r^{2}+t^{2}-s^{2}\right)}{s^{2}}+\left(r^{2}+t^{2}-s^{2}\right),$$ arises due to the form of the electric field of a dipole. The screening function is used to determine the screened polarization potential, $\tilde{U}_{p}(r),$ of an electron with one atom in a dense fluid,
$$\tilde{U}_{p}(r)=f(r)U_{p}(r).$$
Secondly, the direct interaction of the electrons with other atoms in the bulk is significant in a dense system, even when the electron is very close to the focus atom. By following the procedure of Lekner [@Lekner1967], we construct an effective potential, $U_{\mathrm{eff}}=U_{1}+U_{2}$, where $U_{1}$ describes the direct (screened) interaction of the electrons and the target atom, and where $U_{2}$ describes the collective interaction of the electron with the rest of the bulk atoms. We approximate $U_{2}$ by an ensemble average of the bulk, i.e., $$U_{2}(r)=\frac{2\pi N}{r}\int_{0}^{\infty}dt\ U_{1}(t)\int_{\left|r-t\right|}^{r+t}ds\ sg(s).$$ We note that taking the ensemble average has the advantage of enforcing spherical symmetry of the total effective potential $U_{\mathrm{eff}}$.
Since, in a dense system, the electron is never in effectively free space, a different measure of the volume “owned” by the focus atom is required. It is natural to define the distance of first turning point of the potential $U_{\mathrm{eff}}$, which is denoted by $r_{m}$, as the spherical distance under the influence of the focus atom. Hence we can say that a single collision event takes place when an electron enters and leaves the radius $r_{m}$ of a single atom. In order to calculate the phase shift at the distance $r_{m}$ rather than infinity, we have set the upper limits of equations (7) and (8) of Boyle et al. [@Boyletal15] for $A_{\kappa}$ and $B_{\kappa}$ to be $r_{m}$.
The full differential elastic scattering cross-section for the gas and liquid phases (with and without coherent scattering) are displayed in Figure \[fig:DifferentialCrossSections\]. For the dilute gas phase, we observe the presence of a forward-peaked minimum in the range $0.2$–$1$ eV, below which the differential cross-section is essentially isotropic. This minimimun is the well-known Ramsauer minimimun which occurs in a number of electron-atom gas-phase cross-sections. At energies above the minimun, the differential cross-section demonstrates increased magnitude and also enhanced anisotropy, with peaks in the forward- and back-scattering directions. When the modification due to screening and interactions from the liquid bulk are included, we observe the suppression of the Ramsauer minimum and a removal of the forward peak for low and moderate energies. At higher energies, the liquid differential cross-section becomes similar both qualitatively and quantitatively to the gas phase cross-section. When the liquid phase differential cross-section is combined with the static structure factor accounting for coherent scattering effects, the resulting differential cross-section $\Sigma(\epsilon,\chi)$ takes on a completely different qualitative structure. At low energies, the differential scattering cross-section has been further reduced in magnitude at all scattering angles. At higher energies the forward-scattering has been reduced, while the back-scattering peak remains essentially unaffected.
The momentum transfer cross-sections corresponding to the differential scattering cross-sections in Figure \[fig:DifferentialCrossSections\] are displayed in Figure \[fig:Liquid-Xe-cross-sections\], along with the cross-sections from Atrazhev et al. [@Atrazhevetal05]. The comments made regarding the differential scattering cross-sections are once again reflected here. The Ramsauer minimum observed in the gas-phase is completely suppressed in the liquid-phase, and there is a large reduction in the magnitude of the cross-sections when the screening and liquid effects are included, and again when coherent scattering effects are included. At higher energies, the liquid-phase cross-sections approach the gas-phase values, with some additional oscillatory structure evident when coherent scattering is included. The cross-sections of Atrazhev et al. [@Atrazhevetal05] have been calculated using a similar formalism, but with a pseudo-potential that replaces the short-range part of the interaction by a boundary condition that reproduces the expected scattering length in the gas-phase. Their cross-sections are qualitatively similar to ours, but are consistently smaller in magnitude over the range of energies considered. We attribute these differences to our treatment of the static and exchange parts of the potential, for both the focus and surrounding atoms, as well as the inclusion of a full multipole polarization potential.
Results\[sec:Results\]
======================
One of the key functions of modern-day swarm experiments is to assess the completeness and accuracy of cross-section [@Boyle2014; @Boyle2014a; @DeUrquijo2014; @White2014] sets. Swarm experiments are many-scattering experiments, where there is a balance established between the number of particles and the momentum and energy transfers occurring. In the following sections we consider the calculation of the macroscopic swarm transport properties in the gaseous and liquid environments from the microscopic cross-sections. Particular attention is given to investigating the validity of the two-term approximation in our calculations, and we compare these with the full multi-term results. In Section \[sub:Electrons-in-liquid-2\], we introduce a simple scaling algorithm to adapt any gas-phase cross-section to the liquid-phase based on the our *ab initio* scattering calculations.
Electrons in gaseous xenon\[sub:Electrons-in-gaseous\]
------------------------------------------------------
The calculated reduced mobility and characteristic energies using the gas-phase cross-sections detailed in Section \[ScatteringGas\] and the reference cross-sections of Biagi [@Bordetal13] are presented in Figure \[fig:Drift-Gas\]. They are compared against various experimental data for xenon gas [@koizumi86; @pack92a]. We restrict ourselves to the reduced electric fields of less than $1$ Td, to ensure we are in the energy regime where only elastic scattering is operative. For the gas calculations, using the present cross-sections, we observe agreement to within $30$% or better for both the reduced mobility and the transverse characteristic energy over the range of the reduced fields considered. These errors decrease to $5$% or better for the transverse characteristic energy, above the field region where the transport properties rapidly increase. This can be compared with the Biagi [@Bordetal13] cross-section calculations, which demonstrate agreement to within $10$% or better (generally less than $5$%) for reduced mobility and transverse characteristic energy. Although the experimental longitudinal characteristic energies exhibit larger variation and error than the transverse counterpart, our calculated energies using the present cross-sections and those of Biagi [@Bordetal13] appear to give good agreement. The major difference between the transport calculations using the two different cross-sections is in the turning point of the longitudinal characteristic energy profile: our ab initio cross-sections cause a turning point at slightly higher reduced electric field strengths than the experiment, whereas the Biagi cross-sections is consistent with the experimental measurements.
In a direct reflection of our increased momentum transfer cross-section, as compared to the reference, the transverse characteristic energy and reduced mobility generally underestimate the experimental measurements [@koizumi86; @pack92a]. The increased momentum transfer tends to increase the randomization of the electron’s direction during a collision, which reduces the fields ability to efficiently pump energy into the system. It is no surprise that the Biagi cross-section gives closer agreement between calculation and experiment, since it is informed by experimental swarm measurements and was not from an *ab initio* **theory*.*
Electrons in liquid xenon\[sub:Electrons-in-liquid\]
----------------------------------------------------
In Figure \[fig:Liquid-Ar\] the reduced mobility and characteristic energies are now compared in both the gaseous and liquid phases. The transport coefficients are presented against reduced electric fields so that any linear dependence on density (as occurs in the dilute-gas limit) has been removed, and so we have a true comparison of the gaseous and liquid phases. Qualitatively, for a given reduced field in the low-energy regime, we observe that the reduced mobility and transverse characteristic energy in the liquid phase are both significantly larger, often by several orders of magnitude over the gaseous phase. In our investigations of electrons in liquid argon [@Boyletal15], the transverse characteristic energy in the liquid phase was instead smaller than in the gaseous phase. The longitudinal characteristic energy is generally larger for reduced electric field strengths of less than $\sim0.2$ Td, but smaller for the higher electric field strength considered. Our calculations using the ab-initio theory are accurate to within $25$% for the mobility (generally less than $15\%$) and $25$% for the transverse characteristic energy. These are of similar magnitude to those for the gas phase.
Although the differential scattering cross-sections for Xe displayed an enhanced anisotropic nature in the liquid phase over that in the gas phase, the impact of this anisotropy on the velocity distribution function was not particularly significant here. The errors associated with the two-term approximation ($l_{\mathrm{max}}=1)$ to the velocity distribution function are displayed in Figure \[fig:Two-term approximation\]. In the gaseous and liquid phases there are differences as large as $10$% and $40$% respectively, in the transverse characteristic energies. The errors associated with the longitudinal characteristic energies was less than $0.1$% for both the gas and liquid phases, and hence were omitted from Figure \[fig:Two-term approximation\].
Rescaling gas phase cross-sections \[sub:Electrons-in-liquid-2\]
----------------------------------------------------------------
It was noted in Section \[sub:Electrons-in-gaseous\] that our gas phase momentum-transfer cross-section overestimates the reference cross-section of Biagi [@lxcatbiagi14a; @Bordetal13], the latter of which is shown to produce transport data accurate to within $10$% of the experimental data, compared with $30$% for the current *ab-initio* theory. In order to utilize the apparent enhanced accuracy associated with the experimental cross-section, we postulate a method of extracting the explicit liquid-based effects from the theory and applying it to the experimentally measured cross-section. The ratio of the liquid to gas phase cross-section using our formalism gives an energy dependent scaling for the importance of screening and coherent scattering. Let us define a scaling factor, $\xi_{l}(v)$, such that: $$\xi_{l}(v)=\frac{\tilde{\nu_{l}}(v)}{\nu_{l}(v)},$$ where these $\nu$ correspond to our scattering calculations using the Dirac-Fock equations described in Section \[ScatteringGas\] and \[sub:Liquid-xenon-cross-section\] respectively. We can then build an effective liquid-phase momentum-transfer cross-section based on the Biagi [@Bordetal13] reference data, $\Sigma_{m}^{Biagi}(v)$ , via: $$\Sigma_{m}^{Biagi}(v)=\xi(v)\sigma_{m}^{Biagi}(v),$$ where $\sigma_{m}^{Biagi}(v)$ is the Biagi recommended gas-phase cross-section [@Bordetal13]. It should be noted that, due to the lack of differential cross-section information for the Biagi set, only a two-term approximation can be used here, i.e. only the momentum transfer cross-section is known. The effective Biagi liquid momentum transfer cross-section is shown in Figure \[fig:Liquid-Xe-cross-sections-2\]. The method is similar in spirit to the procedure proposed in reference [@Boyle2012], though scaling was done with respect to the transport coefficients rather than the cross-sections.
The characteristic energies and reduced mobility calculated using the effective Biagi liquid momentum transfer cross-section is shown in Figure \[fig:Liquid-Xe-1\], along with the previous calculations considered in this manuscript. It appears that this result gives a slightly better agreement with experiment for the reduced mobilities at low field strengths i.e., errors within $20\%$, while it is no worse for the transverse characteristic energiy. The biggest improvements occur at the low electric field strengths, where the errors are effectively half that of the *ab initio* approach.
Conclusions
===========
We have presented an *ab-initio* treatment of electron scattering and transport in liquid xenon. There are no free parameters in this calculation, and hence the agreement to within $25$% in the mobility and the transverse characteristic energy in the reduced field range $10^{-4}$–$1$ Td is considered satisfactory.Given that, in the dilute gas phase, agreement to within $30$% for the mobility and transverse characteristic energy transport coefficients was achieved, with the current scattering calculations, this gives confidence that the majority of the essential physics for considering high mobility noble liquids is present in the theory. A scaling factor formed from the ratio of the calculated liquid to gas phase cross-sections, was postulated to encompass the “liquid”-based effects, enabling more accurate gas-phase cross-sections derived from experiment to be translated to liquid-phase cross-sections. Subsequent enhancements in the accuracy of the transport coefficient calculations were achieved, with the mobility errors reduced to $20$% and and even bigger improvement displayed for the lowest fields considered.
This work was supported in part by the Australian Research Council through its Discovery Program scheme. M. J. B. and S. J. B. acknowledge University Malaysia for their “Icon Professor” appointments, while R. D. White thanks Prof. K Ratnavelu for some financial support through the University Malaysia grant UMRP11A-13AFR. S. D. acknowledges support from MPNTRRS Projects OI171037 and III41011.
![\[fig:Liquid-Xe-cross-sections-1\] The current momentum transfer cross-sections in the gas-phase (Gas), and the reference momentum transfer cross-section of Biagi [@Bordetal13] (Biagi gas).](cs_comp){width="0.9\columnwidth"}
![\[fig:structure\_factor\]Xenon structure factors. The crosses represent the Becchi measurements [@Becchi1997] and the solid blue lines are our Monte-Carlo simulations of the Lennard-Jones fluid at a high temperature near the critical point of $274.7$ K. The solid green line represents the structure factor at $165$ K that we use in this paper and the dashed green line shows the rescaled measurements from argon experimental data [@Yarnell73] used by reference [@Atrazhevetal05].](SF_comparison){width="70.00000%"}
![\[fig:DifferentialCrossSections\] Current differential cross-sections, $\Sigma(\epsilon,\chi)$., in square angstroms for electrons in Xe for a) dilute gas phase, b) effective liquid phase including screening effects, and c) liquid phase including coherent scattering effects.](DCS_comp_image){width="0.9\columnwidth"}
![\[fig:Liquid-Xe-cross-sections\] The current momentum transfer cross-sections in the gas-phase (Gas; Biagi gas), liquid-phase (Scr liq, Atrazhev scr liq) and including their modifications when coherent scattering effects are included (Scr + coh liq, Atrazhev scr + coh liq). A detailed description of the Atrazhev et al. cross-section calculations is given in reference [@Atrazhevetal05].](cs_comp_all){width="0.9\columnwidth"}
![\[fig:Drift-Gas\]The transverse ($D_{T}/\mu$) and longitudinal ($D_{L}/\mu$) characteristic energies (top) and reduced mobility ($n_{0}\mu$) (bottom) of electrons in gaseous xenon, calculated using the current potentials and associated cross-sections detailed in Section \[ScatteringGas\] (Gas), and the recommended cross-section of Biagi [@Bordetal13] (Biagi gas), and compared with available experimental data (Koizumi et al. [@koizumi86] at 300 K; Pack et. al [@pack92a] at 300 K;). The background xenon gas for the calculations was fixed at 300 K.](ChEn_gas){width="0.9\columnwidth"}
![\[fig:Drift-Gas\]The transverse ($D_{T}/\mu$) and longitudinal ($D_{L}/\mu$) characteristic energies (top) and reduced mobility ($n_{0}\mu$) (bottom) of electrons in gaseous xenon, calculated using the current potentials and associated cross-sections detailed in Section \[ScatteringGas\] (Gas), and the recommended cross-section of Biagi [@Bordetal13] (Biagi gas), and compared with available experimental data (Koizumi et al. [@koizumi86] at 300 K; Pack et. al [@pack92a] at 300 K;). The background xenon gas for the calculations was fixed at 300 K.](mob_gas){width="0.9\columnwidth"}
![\[fig:Liquid-Ar\]Comparison of the transverse ($D_{T}/\mu$) and longitudinal ($D_{L}/\mu$) characteristic energies (top) and reduced mobilities ($n_{0}\mu$) (bottom) in gaseous and liquid xenon, with those calculated from the various approximations to the cross-sections. Experimental data ( Liquid phase: Shibamaru et al. [@Shibetal1984] at 165K; Huang and Freeman [@HuangFreem78] at 163 K). The current liquid-phase cross-sections (Liq) have been calculated including coherent scattering effects. All other profiles are the same as given in Figure \[fig:Drift-Gas\].](ChEn_liq){width="0.9\columnwidth"}
![\[fig:Liquid-Ar\]Comparison of the transverse ($D_{T}/\mu$) and longitudinal ($D_{L}/\mu$) characteristic energies (top) and reduced mobilities ($n_{0}\mu$) (bottom) in gaseous and liquid xenon, with those calculated from the various approximations to the cross-sections. Experimental data ( Liquid phase: Shibamaru et al. [@Shibetal1984] at 165K; Huang and Freeman [@HuangFreem78] at 163 K). The current liquid-phase cross-sections (Liq) have been calculated including coherent scattering effects. All other profiles are the same as given in Figure \[fig:Drift-Gas\].](mob_liq){width="0.9\columnwidth"}
![\[fig:Two-term approximation\] Percentage differences between the two-term and multi-term values of the characteristic energy for the gas (Gas) and liquid (Liq) phases. All percentages are relative to the converged multi-term result using the full differential cross-section. ](multiterm){width="0.9\columnwidth"}
![\[fig:Liquid-Xe-cross-sections-2\] The momentum transfer cross-sections in the gas-phase (Gas; Biagi), liquid-phase (Scr liq, Scaled scr liq) and including their modifications when coherent scattering effects are included ( Scr+coh liq, Scaled scr+coh liq). ](cs_comp_scaled){width="0.9\columnwidth"}
![\[fig:Liquid-Xe-1\]Comparison of the transverse ($D_{T}/\mu$) and longitudinal ($D_{L}/\mu$) characteristic energies (top) and reduced mobilities ($n_{0}\mu$) (bottom) in gaseous and liquid xenon, with those calculated from the various approximations to the cross-sections. The current liquid-phase cross-sections (Liq) have been calculated including coherent scattering effects. All other profiles are the same as in Figure \[fig:Drift-Gas\]. The calculations using the re-scaled Biagi cross-sections are given by the red lines (Scaled liq). All other profiles are the same as given in Figures \[fig:Drift-Gas\] and \[fig:Liquid-Ar\]. ](ChEn_scaled){width="0.9\columnwidth"}
![\[fig:Liquid-Xe-1\]Comparison of the transverse ($D_{T}/\mu$) and longitudinal ($D_{L}/\mu$) characteristic energies (top) and reduced mobilities ($n_{0}\mu$) (bottom) in gaseous and liquid xenon, with those calculated from the various approximations to the cross-sections. The current liquid-phase cross-sections (Liq) have been calculated including coherent scattering effects. All other profiles are the same as in Figure \[fig:Drift-Gas\]. The calculations using the re-scaled Biagi cross-sections are given by the red lines (Scaled liq). All other profiles are the same as given in Figures \[fig:Drift-Gas\] and \[fig:Liquid-Ar\]. ](mob_scaled){width="0.9\columnwidth"}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this paper a stochastic equation on compact groups in discrete negative time is studied. This is closely related to Tsirelson’s stochastic differential equation, of which any solution is non-strong. How the group action reflects on the set of solutions is investigated. It is applied to generalize Yor’s result and give a necessary and sufficient condition for existence of a strong solution and for uniqueness in law.'
address:
- 'Jirô Akahori, Department of Mathematical Sciences, Ritsumeikan University, 1-1-1 Noji-higashi, Kusatsu, Shiga, 525-8577, Japan'
- 'Chihiro Uenishi, Cadem Corporation, Ltd., Shin-Yokohama Office, Houei-Shin-Yokohama Bldg., 2-14-9, Shin-Yokohama, Kouhoku-Ku, Yokohama, 222-0033, Japan'
- 'Kouji Yano, Research Institute for Mathematical Sciences, Kyoto University, Kyoto, 606-8502, Japan'
author:
- Jirô Akahori
- Chihiro Uenishi
- Kouji Yano
title: Stochastic equation on compact groups in discrete negative time
---
March 3, 2006
Introduction
============
In contrast with that of ordinary ones, the theory of stochastic differential equations has the distinguished notions of a [*strong solution*]{} and two uniqueness properties: [*pathwise uniqueness*]{} and [*uniqueness in law*]{}. Therefore we have the following four cases: $$\text{
\begin{tabular}{c||c|c}
& unique in law & non-unique in law \\
\hline
\hline
$ \exists $ strong solution & {$ ({\bf C0}) $}& {$ ({\bf C2}) $}\\
\hline
$ \not \exists $ strong solution & {$ ({\bf C1}) $}& {$ ({\bf C3}) $}\end{tabular}
}
\label{table}$$ The celebrated theorem of T. Yamada and S. Watanabe ([@YW]) is stated as follows: [*Pathwise uniqueness implies uniqueness in law and then any solution is strong*]{}. Except for the trivial cases where there is no solution, we may say that pathwise uniqueness implies that the case [[$ ({\bf C0}) $]{}]{} occurs. In most cases the converse is also true (See A. K. Zvonkin and N. V. Krylov [@ZK]).
B. Tsirelson ([@Tsi]) has presented a remarkable example of a stochastic differential equation stated below, which enjoys uniqueness in law property but has no strong solution: In short, the case [[$ ({\bf C1}) $]{}]{} occurs. His equation has deeply been investigated from various viewpoints by many researchers: See, for example, [@ES], [@Kall] and [@LGY].
To prove the non-existence of strong solutions, Tsirelson introduced a stochastic equation on the torus $ \T = \R / \Z $ indexed by discrete negative time: $$\eta_k = \eta_{k-1} + \xi_k ,
\quad k \in -\N,
\label{Tsirel eq}$$ where $ \xi = ( \xi_k , k \in -\N ) $ is a Gaussian driving noise and $ \eta = ( \eta_k , k \in -\N ) $ is an unknown process. We can find its simpler proof, mainly due to N. V. Krylov, in the literature: For example, see [@LS pp.150–151], [@SY pp.149–151] and [@IW pp.195–197]. In these contexts, the authors discussed, instead of , a modified equation on the real line $ \R $ as follows: $$\begin{aligned}
\eta_k = \alpha (\eta_{k-1}) + \xi_k ,
\quad k \in -\N.
\label{Yor eq}\end{aligned}$$ Here $ \alpha (x) $ denotes the fractional part of $ x \in \R $.
M. Yor ([@Yor]) has studied the stochastic equation for an [*arbitrarily*]{} given noise law, which is not necessarily Gaussian. In this case we always have a non-strong solution, so that the case [[$ ({\bf C0}) $]{}]{} never occurs. He successfully obtained striking results to give a necessary and sufficient condition for the trichotomy [[$ ({\bf C1}) $]{}]{}-[[$ ({\bf C3}) $]{}]{} in terms of the noise law. One of his results may roughly be stated as follows. To any given noise law $ \mu $ there corresponds a subgroup $ \Z_\mu $ of $ \Z $ such that the following holds.
\[Intro YorTh\] [(i)]{} The case [[$ ({\bf C1}) $]{}]{} occurs iff $ \Z_\mu = {\left\{ 0 \right\}} $.\
[(ii)]{} The case [[$ ({\bf C2}) $]{}]{} occurs iff $ \Z_\mu = \Z $. If it is true, then, among the solutions, all extremal points are strong and the others non-strong.\
[(iii)]{} The case [[$ ({\bf C3}) $]{}]{} occurs iff $ {\left\{ 0 \right\}} \subsetneq \Z_\mu \subsetneq \Z $.
This result sounds somehow paradoxical; at least, they are mysterious: [*Uniqueness in law implies that the solution is non-strong, and existence of strong solution(s) implies that it is non-unique but extremal*]{}.
The purpose of the present paper is to generalize Yor’s results. We study the following stochastic equation on a compact group $ G $ in discrete negative time: $$\eta_k = \eta_{k-1} \cdot \xi_k ,
\quad k \in -\N,
\label{STYE}$$ which we call the [*simple Tsirelson–Yor equation on $ G $*]{}. We shall introduce the notions of a solution, a strong solution and uniqueness in law following the theory of stochastic differential equations. Then we obtain the following result, which generalizes Theorems 2 of [@Yor].
\[Intro exist\] For any given noise law $ \mu $, there exists a solution $ \bP^*_\mu $ of the equation such that all marginal distributions of $ \bP^*_\mu $ are uniform. If $ G \neq {\left\{ {e}\right\}} $, then the solution $ \bP^*_\mu $ is non-strong.
This result says that, if $ G \neq {\left\{ {e}\right\}} $, then the case [[$ ({\bf C0}) $]{}]{} always fails. Theorem \[Intro exist\] will be restated in Theorem \[YorTh\]
We shall naturally define a group action of $ G $ on the set of solutions. Our key result is roughly stated as follows.
\[Intro action\] The action restricted on the set of extremal points is transitive.
This implies that the set of extremal points is exhausted by the $ G $-orbit of an arbitrary extremal point $ \bP^o $ and is homeomorphic to the homogeneous space $ G/H_\mu(\bP^o) $ where $ H_\mu(\bP^o) $ is the isotropic subgroup at the point $ \bP^o $. The precise statement of Theorem \[Intro action\] will be given in Theorem \[Gal1\].
Based on Theorem \[Intro action\] and fully employing the representation theory of compact groups, we generalize Theorem \[Intro YorTh\] to give a necessary and sufficient condition for the trichotomy [[$ ({\bf C1}) $]{}]{}-[[$ ({\bf C3}) $]{}]{} in terms of two subgroups $ H_\mu(\bP^o) $ and $ H_\mu^s $. Here $ H_\mu^s $ is a closed normal subgroup of $ G $ which we will define in .
\[Intro trichotomy\] [(i)]{} The case [[$ ({\bf C1}) $]{}]{} occurs iff $ H_\mu(\bP^o) = G $ for some (and hence any) extremal point $ \bP^o $.\
[(ii)]{} The case [[$ ({\bf C2}) $]{}]{} occurs iff $ H_\mu^s = {\left\{ 0 \right\}} $. If it is true, then, among the solutions, all extremal points are strong and the others non-strong.\
[(iii)]{} The case [[$ ({\bf C3}) $]{}]{} occurs iff $ H_\mu(\bP^o) \subsetneq G $ and $ H_\mu^s \supsetneq {\left\{ 0 \right\}} $.
Theorem \[Intro trichotomy\] will be proved in section \[sec: trichotomy\].
The subgroups $ H_\mu^s $ and $ H_\mu(\bP^o) $ for extremal points $ \bP^o $ are related as follows.
\[Intro isot group\] For any extremal point $ \bP^o $, the inclusion $$H_\mu(\bP^o) \subset H_\mu^s$$ holds. If $ G $ is abelian, then the equality holds for any extremal point $ \bP^o $.
We will restate Theorem \[Intro isot group\] as Theorem \[thm: isot group\] including further information. In the case where $ G = \T = \R/\Z $, note that the abelian group $ \Z $ is the dual group of $ G $ in the sense of the [*Pontryagin duality*]{}. In this terminology, the subgroup $ \Z_\mu $ of $ \Z $ in Theoren \[Intro YorTh\] is exactly the dual group of $ H_\mu^s $. In the non-abelian cases, however, we have a typical example given in Example \[ex: sym group\] where $ H_\mu(\bP^o) $ is strictly included in $ H_\mu^s $ for any extremal point $ \bP^o $.
Now we recall Tsirelson’s stochastic differential equation (SDE in short). Let $ (t_k : \, k \in -\N) $ be a decreasing sequence of the interval $ (0,1] $ such that $ t_k \to 0 $ as $ k \to -\infty $. Define $ A(t,X) $, which is called [*Tsirelson’s drift*]{}, as $$A(t,X) = \sum_{k \in -\N} \left\{ \frac{X_{t_{k-1}} - X_{t_{k-2}}}{t_{k-1} - t_{k-2}} \right\}
1_{ [t_{k-1},t_{k}) }(t) , \quad 0 < t < 1 .$$ Then Tsirelson’s SDE is given by $$X_t = B_t + \int_0^t A(s,X) ds , \quad 0 \le t \le 1
\label{Tsirel SDE}$$ where $ (B_t) $ is a one-dimensional Brownian motion. If $ ((X_t),(B_t)) $ is a solution of the equation , then the sequences $ (\eta_k) $ and $ (\xi_k) $ defined by $$\eta_k = \frac{X_{t_{k}} - X_{t_{k-1}}}{t_{k} - t_{k-1}} , \quad
\xi_k = \frac{B_{t_{k}} - B_{t_{k-1}}}{t_k - t_{k-1}} ,
\quad k \in -\N$$ satisfy the equation . Conversely, we can reconstruct the process $ (X_t) $ from the processes $ (\eta_k) $ and $ (B_t) $. Hence we may say that all properties of the SDE can be deduced from those of the equation .
To generalize the stochastic equation , we introduce a class of stochastic equations in discrete negative time, which we call [*Tsirelson–Yor equations*]{}. Let $ S $ be a Polish space and $ G $ a compact group. We introduce an operation of $ G $ on the state space $ S $ through a measurable map $ \psi:G \times S \to S $ and consider a measurable map $ \theta:S \to G $. Then our Tsirelson–Yor equation is of the form $$\eta_k = \psi {\left( \theta(\eta_{k-1}),\xi_k \right)} ,
\quad k \in -\N.
\label{TYE}$$ The key to the generalization is the following [*commutation*]{} condition[^1]: $$\theta {\left( \psi( g,s ) \right)} = g \cdot \theta(s), \quad g \in G, \ s \in S.
\label{UeniC}$$ This condition allows us to reduce the equation to the equation $$\theta(\eta_k) = \theta (\eta_{k-1}) \cdot \theta(\xi_k) ,
\quad k \in -\N.
\label{T-Yor eq 2}$$ If we write $ {\widehat}{\eta}_k=\theta(\eta_k) $ and $ {\widehat}{\xi}_k=\theta(\xi_k) $, then the equation is exactly the simple Tsirelson–Yor equation on $ G $.
We will prove in Proposition \[reduction\] that in order to study the Tsirelson–Yor equation with given laws it is sufficient to investigate the simple Tsirelson–Yor equation on $ G $.
We will show in Example \[ex: Yor eq\] that Yor’s equation is an example of our Tsirelson–Yor equation. Proposition \[reduction\] says that the equation is essentially ‘equivalent’ to the equation $$\alpha (\eta_k) = \alpha (\eta_{k-1}) + \alpha (\xi_k) \quad \mbox{modulo 1},
\quad k \in -\N ,
\label{STYE torus}$$ which is nothing but the equation .
In addition, we will show in Example \[ex: Tanaka eq\] that the following equation is also an example: $$\eta_k = \sgn {\left( \eta_{k-1} \right)} \cdot \xi_k ,
\quad k \in -\N
\label{Tanaka}$$ provided that the noise law has no point mass at $ x=0 $. Here $ \sgn (x) = 1 $ if $ x \ge 0 $ and $ \sgn (x) = -1 $ if $ x<0 $. This equation has been dealt with in [@R-Y Chapter IX, Exercise 3.18] and [@LGY pp. 87–88]. Proposition \[reduction\] says that the equation is equivalent to the simple Tsirelson–Yor equation on $ \Z/2\Z $: $$\sgn {\left( \eta_k \right)} = \sgn {\left( \eta_{k-1} \right)} \cdot \sgn {\left( \xi_k \right)} ,
\quad k \in -\N .
\label{STYE Z/2Z}$$ This equation will be dealt with in Example \[ex: Z/2Z\].
The present paper is organized as follows. In section \[sec: pre\], we give the precise definition of the simple Tsirelson–Yor equation on a compact group and introduce the notions of a solution, a strong solution and uniqueness in law. We also prepare some preliminary facts about the set of solutions. Section \[sec: exist\] is devoted to the proof of Theorem \[Intro exist\]. In section \[sec: hom sp\], we give a precise statement and the proof of Theorem \[Intro action\]. In section \[sec: H mu s\], we introduce a subgroup $ H_\mu^s $ which appears in Theorem \[Intro trichotomy\] and restate Theorem \[Intro isot group\] as Theorem \[thm: isot group\]. In section \[sec: example\], we investigate the equation in detail and consider another typical example of the simple Tsirelson–Yor equation. Section \[sec: trichotomy\] is devoted to the proofs of Theorem \[Intro trichotomy\] and the propositions which are given in section \[sec: H mu s\]. The Tsirelson–Yor equations are defined and discussed in section \[sec: TYE\].
[**Acknowledgments**]{}: The authors wish to express sincere thanks to Professors Marc Yor, Freddy Delbaen, Shinzo Watanabe, and Yoichiro Takahashi for stimulating discussions and valuable comments. They also thank Gen Mano and Hidehisa Alikawa, who kindly informed them of basic theorems of the representation theory. The second author expresses his hearty thanks to Professors Hiroki Aoki and Yoshiaki Kobayashi for their kind tutorial lectures.
Definitions and preliminary facts {#sec: pre}
=================================
Let $ G $ be a compact group. We consider the following stochastic equation on $ G $ in discrete negative time: $$\eta_k = \eta_{k-1} \cdot \xi_k ,
\quad k \in -\N.
\tag{\ref{STYE}}$$
The stochastic equation is called the [*simple Tsirelson-Yor equation*]{} on $ G $, which will be abbreviated by “STYE".
For an arbitrary sequence $ \mu = (\mu_k , k \in -\N) $ of Borel probability measures $ \mu_k $ on $ G $, we consider the equation with a general noise law $ \mu $.
Let us give the precise definition of a solution of . We denote by $ \eta = ( \eta_k , k \in -\N ) $ the coordinate mapping process: $ \eta_k(\omega)=\omega(k) $ for $ \omega \in G^{-\N} $ and $ k \in -\N $. Set $$\xi_k = (\eta_{k-1})^{-1} \cdot \eta_k ,
\quad k \in -\N.
\label{STYE2}$$ Let $ \F^{\eta}_k $ and $ \F^{\xi}_k $ for $ k \in -\N $ denote $$\F^{\eta}_k := \sigma (\eta_k, \eta_{k-1}, \ldots) ,
\quad k \in -\N$$ and $$\F^{\xi}_k := \sigma (\xi_k, \xi_{k-1}, \ldots) ,
\quad k \in -\N$$ respectively. It is obvious that $$\F^{\xi}_k \subset \F^{\eta}_k ,
\quad k \in -\N.$$
Since the law on $ G^{-\N} $ of the process $ \eta $ determines that of the noise $ \xi $, the following definition is reasonable.
\[Def sol\] Let $ \mu = (\mu_k , k \in -\N) $ be a sequence of Borel probability measures $ \mu_k $ on $ G $. A [*solution*]{} of the STYE on $ G $ with the noise law $ \mu $ is a probability measure $ \bP $ on $ G^{-\N} $ such that the following two statements hold:\
(i) $ \xi_k $ is independent of $ \F^{\eta}_{k-1} $ under $ \bP $, for any $ k \in -\N $.\
(ii) $ \xi_k $ is distributed as $ \mu_k $ under $ \bP $, for any $ k \in -\N $.
The totality of solutions of the STYE with the noise law $ \mu $ will be denoted by $ \sP_\mu $.
\[SE remark\] For a given $ \mu $, a probability measure $ \bP $ on $ G^{-\N} $ belongs to $ \sP_\mu $ if and only if the following (inhomogeneous) Markov property holds: $$\bE {\left[ \phi(\eta_k) \mid \F_{k-1} \right]}
= \int_{G} \phi( \eta_{k-1} \cdot g ) \mu_k(dg) , \quad k \in -\N
\label{Markov}$$ for any bounded measurable function $ \phi $ on $ G $.
In order to give a precise meaning to the table , we need to introduce the notions of strong solution and uniqueness in law. We follow the usual terminology in the theory of stochastic differential equations.
\[Def s sol\] Let $ \mu $ be given. A solution $ \bP \in \sP_\mu $ is called [*strong*]{} if $$\F^{\eta}_k \subset \F^{\xi}_k \ \mbox{up to $ \bP $-null sets}, \quad k \in -\N .$$ If $ \bP \in \sP_\mu $ is not strong, then it is called [*non-strong*]{}[^2].
\[Def uni\] Let $ \mu $ be given. It is said that [*uniqueness in law*]{} holds if the set $ \sP_\mu $ consists of at least one element.
Let $ \mu = (\mu_k , k \in -\N) $ be given and consider the set $ \sP_\mu $ of solutions of the STYE on $ G $ with the noise law $ \mu $. Since the condition $ \bP \in \sP_\mu $ is equivalent to the Markov property , the following is obvious.
\[lem: closed\] The set $ \sP_\mu $ is a closed convex subset of the linear topological space of signed measures on $ G^{-\N} $ equipped with the weak-star topology.
We denote the totality of extremal points of $ \sP_\mu $ by $ \ext $. Then Lemma \[lem: closed\] implies that $ \ext \subset \sP_\mu $ and any solution $ \bQ \in \sP_\mu $ has an integral representation $$\bQ(\cdot) = \int_{\ext} \bP(\cdot) \, \gamma(d\bP)$$ for some Borel probability measure $ \gamma $ on $ \ext $.
The following is also true.
\[lem: tail\] For $ \bP \in \sP_\mu $, the solution $ \bP $ is extremal if and only if $ \F^{\eta}_{-\infty } $ is $ \bP $-trivial.
We omit the proof, since the proof of [@Yor Theorem 1, 2)] still survives for the STYE’s.
Existence of a non-strong solution {#sec: exist}
==================================
Since the group $ G $ is compact, we have the normalized Haar measure $ \nu $: That is, there exists a unique positive Borel measure $ \nu $ on $ G $ such that $ \nu(G)=1 $ and such that $$\int_G \phi(g h) \nu(dh) =
\int_G \phi(h g) \nu(dh) =
\int_G \phi(h) \nu(dh)$$ for any $ g \in G $ and any bounded measurable function $ \phi $ on $ G $.
A random variable $ U $ is said to be [*uniformly distributed on $ G $*]{} if the law of $ U $ is equal to the normalized Haar measure $ \nu $.
Now we restate Theorem \[Intro exist\].
\[YorTh\] For an arbitrary noise law $ \mu = {\left( \mu_k \right)} $, there exists a unique element $ \bP^*_{\mu} \in \sP_\mu $ whose marginal distributions are uniform: $$\lambda_k = \nu ,
\quad k \in -\N.
\label{YorTh marginal}$$ Moreover, under $ \bP^*_{\mu} $, $$\mbox{each $ \eta_k $ is independent of the noise $ \F^{\xi}=\sigma(\xi_k : \, k \in -\N) $} .
\label{YorTh indep}$$
In particular, the solution $ \bP^*_{\mu} $ is [*non-strong*]{} unless $ G $ is trivial: i.e. $ G = \{ {e}\} $.
Theorem \[YorTh\] immediately implies the following.
Assume that $ G $ is not trivial. Then, for any noise law $ \mu $, the case [[$ ({\bf C0}) $]{}]{} in the table never occurs.
In the sequel, we always assume that $ G $ is not trivial. Now the problem is how to characterize the trichotomy [[$ ({\bf C1}) $]{}]{}-[[$ ({\bf C3}) $]{}]{}.
The key fact to the proof of Theorem \[YorTh\] is the one-to-one and onto correspondence of each solution to what we call an entrance law.
Let $ \mu = (\mu_k , k \in -\N) $ be given.
A family $ \lambda = (\lambda_k , k \in -\N) $ of probability laws $ \lambda_k $ on $ G $ is called an [*entrance law*]{} for the noise law $ \mu $ if the following recurrence relation holds: $$\lambda_k = \lambda_{k-1} * \mu_k ,
\quad k \in -\N.
\label{consistency}$$ Here $ \mu * \lambda $ for two measures $ \mu $ and $ \lambda $ on $ G $ stands for the convolution of $ \mu $ and $ \lambda $: $$\int_G \phi(g) \mu * \lambda (dg) = \int_G \int_G \phi(gh) \mu(dg) \lambda(dh)$$ for any bounded measurable function $ \phi $ on $ G $.
For any $ \bP \in \sP_\mu $, let $ \lambda = (\lambda_k , k \in -\N) $ denote the marginal distributions of $ \eta_k $ on $ G $ , i.e., $$\lambda_k(\cdot) = \bP (\eta_k \in \cdot) ,
\qquad k \in -\N.
\label{marginal}$$ Then the equation implies that holds. Conversely, the following holds.
\[ent law\] Let $ \mu $ be a given noise law. Let $ \lambda = (\lambda_k , k \in -\N) $ be an entrance law for the noise law $ \mu $: The family $ (\lambda_k , k \in -\N) $ satisfies the consistency condition . Then there exists a unique element $ \bP \in \sP_\mu $ such that holds.
This is obvious by Kolmogorov’s extention theorem, so we omit the proof.
By Lemma \[ent law\], we see that there exists a solution $ \bP^*_\mu $ with marginal distributions given by . Let $ \bE^*_\mu $ denote the expectation with respect to $ \bP^*_\mu $. Let $ \phi $ be a bounded measurable function on $ G \times G^{{\left\{ k-n+1,\ldots,0 \right\}}} $. Noting that the independence of $ \eta_{k-n} $ and $ \sigma (\xi_{k-n+1},\ldots,\xi_0) $, we have $$\begin{aligned}
&& \bE^*_\mu {\left[ \phi {\left( \eta_k; \xi_{k-n+1}, \ldots, \xi_0 \right)} \right]}
\\
&& = \bE^*_\mu {\left[ \phi {\left( \eta_{k-n} \cdot \xi_{k-n+1} \cdots \xi_k; \xi_{k-n+1}, \ldots, \xi_0 \right)} \right]}
\\
&& = \bE^*_\mu {\left[ \phi {\left( \eta_{k-n}; \xi_{k-n+1}, \ldots, \xi_0 \right)} \right]} . \end{aligned}$$ This implies .
Homogeneous space structure of the set of extremal points {#sec: hom sp}
=========================================================
Let $ \mu = (\mu_k , k \in -\N) $ be given. For each $ g \in G $, we define a continuous map $ T_g $ on $ \sP_\mu $ by $$T_g(\bP)(\cdot) = \bP {\left( g \cdot \eta \in \cdot \right)} ,
\quad \bP \in \sP_\mu$$ where we write $$g \cdot \eta = {\left( g \cdot \eta_k , k \in -\N \right)} ,
\quad g \in G , \, \eta \in G^{-\N}.$$ Then the following is obvious.
\[Tg action\] The family of maps $ (T_g : \, g \in G ) $ defines a group action of $ G $ on $ \sP_\mu $ with $ \ext $ an invariant subset, i.e., $$T_g(\sP_\mu) \subset \sP_\mu
\quad \mbox{and} \quad
T_g(\ext) \subset \ext
\quad
\mbox{for} \ g \in G
\label{G action 1}$$ and $$T_g T_{h^{-1}} = T_{gh^{-1}} \quad \mbox{for $ g,h \in G $}.
\label{G action 2}$$
The following proposition shows that the point $ \bP^*_\mu $, which is defined in Theorem \[YorTh\], can be considered to be the [*center*]{} of the set of solutions $ \sP_\mu $.
\[center point\] Let $ \mu $ be given. Then, for any $ \bP \in \sP_\mu $, it holds that $$\bP^*_\mu = \int_G T_g(\bP) \nu(dg) .
\label{center point 1}$$
For any bounded measurable function $ \phi $ on $ G $, we have $$\begin{aligned}
&& \int_G \phi(h) \int_G T_g(\bP)(\eta_k \in dh) \nu(dg)
\\
&& = \int_G \bE {\left[ \phi(g \cdot \eta_k) \right]} \nu(dg) = \int_G \phi(g) \nu(dg) \end{aligned}$$ for any $ k \in -\N $. This implies that all marginal distributions of the RHS of are uniform on $ G $.
Now we are in a position to give the precise statement of Theorem \[Intro action\].
\[Gal1\] Let $ \mu $ be given. Then the action of $ (T_g: \, g \in G) $ restricted on $ \ext $ is transitive: That is, if $ \bP^1 $ and $ \bP^2 $ are two solutions in $ \ext $, then there exists an element $ g \in G $ such that $ \bP^1 = T_g (\bP^2) $.
Let an extremal point $ \bP^o \in \ext $ be fixed. We denote by $ H_\mu(\bP^o) $ the isotropic subgroup at $ \bP^o $: $$H_\mu(\bP^o) = \{ g \in G : \, T_g(\bP^o) = \bP^o \} .$$ It is easy to see that $ H_\mu(\bP^o) $ is a closed subgroup of $ G $, and hence the quotient set $ G/H_\mu(\bP^o) $ is a compact Hausdorff set. Thus Theorem \[Gal1\] implies the following corollary, which reveals the homogeneous space structure of the set $ \ext $.
The action of $ (T_g : \, g \in G) $ restricted on $ \ext $ induces a homeomorphism: $$G/H_\mu(\bP^o) \stackrel{\sim}{\longrightarrow} \ext.
\label{homeo}$$
The key to the proof of Theorem \[Gal1\] is the following lemma, which we follow the proof of Yamada–Watanabe’s theorem [@YW Proposition 1]. See also [@IW pp. 163–166].
We consider a product space $ G^{-\N} \times G^{-\N} \times G^{-\N} $ with its coordinate written as $ (\eta^1,\eta^2,\xi) $. Define $$\F^{\eta^1,\eta^2}_k
= \sigma {\left( \eta^1_j , \eta^2_j : \, j \le k \right)} ,
\quad k \in -\N .$$
\[coupling\] Let $ \bP^1 $ and $ \bP^2 $ be two solutions in $ \sP_\mu $. Then there exists a probability measure $ \bQ $ on $ G^{-\N} \times G^{-\N} \times G^{-\N} $ such that the following statements hold:\
(i) For $ i=1,2 $, the law on $ G^{-\N} \times G^{-\N} $ of $ (\eta^i,\xi) $ under $ \bQ $ coincides with that of $ (\eta,\xi) $ under $ \bP^i $ where $ \xi $ is defined in .\
(ii) For any $ k \in -\N $, $$\xi_k \mbox{ is independent of } \F^{\eta^1,\eta^2}_{k-1} \mbox{ under $ \bQ $} .$$
Let $ \bP^1_{\xi}(\cdot) $ and $ \bP^2_{\xi}(\cdot) $ denote the regular conditional probability given $ \F^{\xi}=\sigma(\xi_k: \, k \in -\N) $ such that $$\bP^i(\eta \in A , \xi \in B) = \int_B \bP^i_{\xi}(A) \mu(d\xi)$$ for arbitrary measurable sets $ A $ and $ B $ of $ G^{-\N} $, $ i=1,2 $. Then the desired probability measure $ \bQ $ is obtained as $$\bQ(d\eta^1 \, d\eta^2 \, d\xi) = \bP^1_{\xi}(d\eta^1) \, \bP^2_{\xi}(d\eta^2) \, \mu(d\xi) .$$ We can prove the claims (i) and (ii) in the same way as in the proof of Yamada–Watanabe’s theorem, so we omit the proofs of Lemma \[coupling\].
Let $ \bQ $ be the probability measure given in Lemma \[coupling\]. As far as the end of this paragraph, we omit to write “$ \bQ $-a.s." By Lemma \[coupling\] (i), we have $$\eta^i_k = \eta^i_{k-1} \cdot \xi_k ,
\quad k \in -\N , \ i=1,2.$$ Let $ k \in -\N $ and $ n \in \N $ be arbitrary numbers. Then $$\eta^i_k = \eta^i_{k-n} \cdot \xi_{k-n+1} \cdots \xi_k ,
\quad i=1,2.$$ Thus we have $${\left( \eta^1_k \right)} \cdot {\left( \eta^2_k \right)}^{-1}
= {\left( \eta^1_{k-n} \right)} \cdot {\left( \eta^2_{k-n} \right)}^{-1} ,
\quad k \in -\N , \ n \in \N .
\label{G action3}$$ Since the LHS is irrelevant to $ n \in \N $, there exists a random variable $ \eps $ which is $ \F^{\eta^1,\eta^2}_{-\infty } $-measurable such that $$\eps = {\left( \eta^1_k \right)} \cdot {\left( \eta^2_k \right)}^{-1} , \quad k \in -\N .
\label{G action4}$$
If we denote by $ \bQ {\left( \cdot \mid \eps = g \right)} $ the regular conditional probability given $ \eps=g $, then we have the following disintegration: $$\bQ(\cdot) = \int_G \bQ {\left( \cdot \mid \eps = g \right)} \bQ( \eps \in dg ) .
\label{decomp Q}$$ Hence we obtain an integral expression of $ \bP^1 $: $$\bP^1(\cdot) = \int_G \bQ {\left( \eta^1 \in \cdot \mid \eps = g \right)} \bQ( \eps \in dg ) .$$ By definition, we see that the law $$\bQ {\left( \eta^1 \in \cdot \mid \eps = g \right)}$$ belongs to $ \sP_\mu $ for $ \bQ( \eps \in dg ) $-almost every $ g \in G $. By the assumption that $ \bP^1 $ is extremal, we obtain $$\bP^1(\cdot) = \bQ {\left( \eta^1 \in \cdot \mid \eps = g \right)}
\label{decomp Q2}$$ for $ \bQ( \eps \in dg ) $-almost every $ g \in G $. We obtain the similar identity for $ \bP^2 $. Note that implies $$\bQ {\left( \eta^1 = g \cdot \eta^2 \mid \eps = g \right)} = 1
\quad \mbox{for any $ g \in G $} .
\label{decomp Q3}$$ Therefore we conclude that $$\bP^1 = T_g {\left( \bP^2 \right)}
\quad \mbox{for $ \bQ( \eps \in dg ) $-a.e. $ g \in G $}.$$ This completes the proof.
In the above proof of Theorem \[Gal1\], take $ g_0 \in G $ such that $ \bP^1 = T_{g_0} {\left( \bP^2 \right)} $. Then it is obvious that the measure $ \bQ( g_0^{-1} \eps \in dg ) $ is the normalized Haar measure on $ H_\mu(\bP^2) $. This fact leads to the following paradox: [*The tail $ \sigma $-field $ \F^{\eta^1,\eta^2}_{-\infty } $ is non-trivial under $ \bQ $, whereas both $ \F^{\eta^1}_{-\infty } $ and $ \F^{\eta^2}_{-\infty } $ are trivial.*]{}
Consider the STYE on the direct product group $ G \times G $ $$(\eta^1_k,\eta^2_k) = (\eta^1_{k-1},\eta^2_{k-1}) \cdot (\xi^1_k,\xi^2_k) ,
\quad k \in -\N$$ with the noise law $ {\widetilde}{\mu}=({\widetilde}{\mu}_k : k \in -\N) $ given by $$\int_{G \times G} \phi(g,h) {\widetilde}{\mu}_k(dg \times dh) = \int_G \phi(g,g) \mu_k(dg)$$ for any bounded measurable function $ \phi $ on $ G \times G $ and for $ k \in -\N $. Denote the set of solutions by $ {\widetilde}{\sP}_{{\widetilde}{\mu}} $. We denote the marginal laws of $ (\eta^1,\eta^2) $ under the measure $ \bQ $ by $ {\widetilde}{\bP} $. Then $ {\widetilde}{\bP} $ and the regular conditional probabilities $ {\widetilde}{\bP}(\cdot \mid \eps=g) $ belong to $ {\widetilde}{\sP}_{{\widetilde}{\mu}} $. Note that implies that $${\widetilde}{\bP}(\cdot) = \int_G {\widetilde}{\bP}(\cdot \mid \eps=g) \bQ( \eps \in dg ) .$$ In this integral expression of the solution $ {\widetilde}{\bP} $, the integrand $ {\widetilde}{\bP}(\cdot \mid \eps=g) $ belongs to $ {\rm ex}({\widetilde}{\sP}_{{\widetilde}{\mu}}) $ for $ \bQ( \eps \in dg ) $-almost every $ g \in G $. In fact, by and , we see that $ \F^{\eta^1,\eta^2}_{-\infty } $ is $ {\widetilde}{\bP}(\cdot \mid \eps=g) $-trivial for $ \bQ( \eps \in dg ) $-almost every $ g \in G $.
The subgroup $ H_\mu^s $ {#sec: H mu s}
========================
To begin with, we recall the well-known [*Peter–Weyl theorem*]{} for compact groups (see, e.g., [@Sugiura Chapter 1] and [@JS Corollary 13]).
Let $ G $ be a compact group and let $ \nu $ denote the normalized Haar measure on $ G $. Let $ \sG $ denote the totality of irreducible unitary representations $ \rho $ of $ G $ on a finite-dimensional linear space $ V^{\rho} $. Then the following holds: [ *The family $$( \rho_{i,j} : \, 1 \le i,j \le \dim \rho , \ \rho \in \sG )$$ forms a total family in the space of continuous functions on $ G $*]{}. Here $ ( \rho_{i,j} ) $ denotes the matrix element of a representation $ \rho \in \sG $.
In what follows we consider the STYE on a compact group $ G $ with a fixed noise law $ \mu $.
To characterize the trichotomy, we introduce a subset $ H^s_\mu $ of $ G $ as follows. For an extremal point $ \bP^o \in \ext $ we define $$\sH^s_\mu(\bP^o)
= \Bigl\{ \rho \in \sG : \,
\text{$ \rho(\eta_k) $ is $ \F^{\xi}_k $-m'ble $ \bP^o $-a.s. for $ k \in -\N $}
\Bigr\}.$$ Here the word “m’ble" is abbreviated from “measurable". It is clear from Theorem \[Gal1\] that the set $ \sH^s_\mu (\bP^o) $ is independent of the choice of $ \bP^o \in \ext $. So we simply write $ \sH^s_\mu (\bP^o) $ as $ \sH^s_\mu $, and define $$H^s_\mu
= \{ g \in G : \, \mbox{$ \rho (g) = {\rm id} $ for every $ \rho \in \sH^s_\mu $} \} .
\label{H mu s}$$
Note that we need to know at least one extremal point $ \bP^o $ in order to compute $ H_\mu(\bP^o) $ and $ H^s_\mu $. Let us introduce two subsets $ H^1_\mu $ and $ H^2_\mu $ of $ G $ which can directly be computed from the noise law $ \mu $ as follows. For $ \rho \in \sG $, we set[^3] $$R_k = \int_G \rho(g) \mu_k(dg)
, \quad k \in -\N .
\label{Rk}$$ Then the following two limits exist for any $ k \in -\N $: The first one is $$r^1_k[\rho] =
\lim_{n \to \infty } \| R_{k-n} R_{k-n+1} \cdots R_k \|$$ where $ \| \cdot \| $ denotes the operator norm of linear operators on the representation space $ V^{\rho} $. The second one is $$r^2_k[\rho] =
\lim_{n \to \infty } \left| \det {\left( R_{k-n} R_{k-n+1} \cdots R_k \right)} \right| .$$ The convergence of the first limit is obvious by $ \| R_j \| \le 1 $ for any $ j \in -\N $. That of the second is ensured by $ |\det R_j| \le 1 $ for any $ j \in -\N $, which will be assured by Lemma \[lem: det ineq\]. Now we set $$\sH_\mu^i :=
\{ \rho \in \sG : \, r^i_k[\rho] > 0 \ \text{for some $ k \in -\N $} \}
, \quad i=1,2$$ and define $$H^i_\mu :=
\{ g \in G : \, \mbox{$ \rho (g) = {\rm id} $ for every $ \rho \in \sH^i_\mu $} \}
, \quad i=1,2 ,
\label{H mu}$$ where the symbol ‘$ {\rm id} $’ stands for the identity on the representation space $ V^\rho $.
The following hierarchy is fundamental to our analysis.
\[thm: isot group\] [(i)]{} The three subclasses $ \sH^1_\mu $, $ \sH^2_\mu $ and $ \sH^s_\mu $ satisfy the following inclusions: $$\sH^1_\mu \supset \sH^s_\mu \supset \sH^2_\mu .
\label{subclass inclusion}$$ [(ii)]{} The three subsets $ H_\mu^1 $, $ H_\mu^2 $ and $ H^s_\mu $ are closed normal subgroups of $ G $ such that $$H_\mu^1 \subset H_\mu (\bP^o) \subset H^s_\mu \subset H_\mu^2
\label{isot group inclusion}$$ for any $ \bP^o \in \ext $. If $ G $ is abelian, then the equalities hold: $$H_\mu^1 = H_\mu (\bP^o) = H^s_\mu = H_\mu^2 .
\label{isot group equality}$$
We remark that this result includes the whole statement of Theorem \[Intro isot group\]. The proof of Theorem \[thm: isot group\] will be given in section \[sec: trichotomy\].
\[isot group another pt\] The isotropic subgroup $ H_\mu(\bP) $ at another extremal point $ \bP=T_g(\bP^o) \in \ext $ is related to $ H_\mu(\bP^o) $ by $$H_\mu(\bP) = g H_\mu(\bP^o) g^{-1} .$$ Hence the isotropic subgroup $ H_\mu(\bP^o) $ is not necessarily normal, while the subgroup $ H^s_\mu $ is always normal.
Examples {#sec: example}
========
\[ex: Z/2Z\] Consider the STYE on the group $ \Z/2\Z \simeq {\left\{ 1,-1 \right\}} $. Since the group $ \Z/2\Z $ is abelian, we have the equalities . Note that the class $ \sG $ consists of only one element $ \rho $ such that $$\rho(1) = 1 , \quad \rho(-1) = -1 .$$ For a noise law $ \mu=(\mu_k: k \in -\N) $, we set $ p_k = \mu_k {\left( {\left\{ 1 \right\}} \right)} $. Now set $$r_k = \lim_{n \to \infty } \prod_{j=k-n}^k |2p_j-1| ,
\quad k \in -\N .$$ Then Theorem \[Intro trichotomy\] leads to the following.
The case [[$ ({\bf C1}) $]{}]{} or [[$ ({\bf C2}) $]{}]{} occurs according to whether the infinite product $ r_k $ vanishes for any $ k \in -\N $ or not.
This is obvious, so we omit the proof.
We give a typical example of the STYE on a non-abelian group where $ H_\mu(\bP^o) $ is non-normal and hence strictly included in $ H^s_\mu $.
\[ex: sym group\] Consider the symmetric group of degree 3: $$\fS_3 = \{ {e},(12),(23),(13),(123),(132) \} .$$ Set $$\begin{aligned}
&& H^o = {\left\{ {e},(12) \right\}} , \\
&& H^1 = (13) H^o = {\left\{ (13),(23) \right\}} , \\
&& H^2 = (123) H^o = {\left\{ (123),(132) \right\}} . \end{aligned}$$ Then $ H^o $ is a non-normal subgroup of $ \fS_3 $ such that $$\fS_3 / H^o = {\left\{ H^o,H^1,H^2 \right\}} .$$ Let $ \mu = (\mu_k : k \in -\N) $ be the sequence of the uniform laws on $ H^o $: $ \mu_k = \nu^o $ for any $ k \in -\N $ where $$\nu^o({\left\{ {e}\right\}}) = \nu^o({\left\{ (12) \right\}}) = 1/2 .
\label{iid law}$$
\[prop: sym group\] Consider the STYE on $ \fS_3 $ with the noise law $ \mu $ given above. Then there exists a solution $ \bP^o \in \sP_\mu $ such that the following hold:\
[(i)]{} Each $ \eta_k $ under $ \bP^o $ is uniformly distributed on $ H^o $.\
[(ii)]{} The extremal points $ \ext = {\left\{ \bP^o,\bP^1,\bP^2 \right\}} $, where $$\bP^1:=T_{(13)}(\bP^o) , \quad \bP^2:=T_{(123)}(\bP^o) .
\label{sym group 1}$$ [(iii)]{} The isotropic subgroup $ H_\mu(\bP^o) = H^o $. Hence $$\{ {e}\} = H_\mu^1 \subsetneq H_\mu(\bP^o) = H^o \subsetneq H_\mu^s = H_\mu^2 = \fS_3 .$$
Note that the family $ \mu = (\mu_k : k \in -\N) $ itself forms an entrance law: $ \nu^o = \nu^o * \nu^o $. Thus there exists a solution $ \bP^o $ such that each marginal distribution $ \bP^o(\eta_k \in \cdot) $ for any $ k \in -\N $ coincides with $ \nu^o $. Thus we obtain (i).
Let $ \bP $ be a solution and let $ \lambda = (\lambda_k : k \in -\N) $ be the corresponding entrance law. Then we have $$\lambda_k({\left\{ g \right\}})
= \frac{1}{2} \lambda_{k-1}({\left\{ g \right\}}) + \frac{1}{2} \lambda_{k-1}({\left\{ g (12) \right\}}) ,
\quad g \in \fS_3.$$ This implies that there exist $ p_0,p_1,p_2 \ge 0 $ with $ p_0+p_1+p_2=1 $ such that $$\begin{aligned}
&& \lambda_k({\left\{ {e}\right\}}) =
\lambda_k({\left\{ (12) \right\}}) = p_0/2 , \\
&& \lambda_k({\left\{ (13) \right\}}) =
\lambda_k({\left\{ (23) \right\}}) = p_1/2 , \\
&& \lambda_k({\left\{ (123) \right\}}) =
\lambda_k({\left\{ (132) \right\}}) = p_2/2 \end{aligned}$$ for any $ k \in -\N $. Therefore we obtain $$\bP = p_0 \bP^o + p_1 \bP^1 + p_2 \bP^2 ,$$ where $ \bP^1 $ and $ \bP^2 $ are defined in . Since the measures $ \bP^o $, $ \bP^1 $ and $ \bP^2 $ are mutually singular, we obtain (ii). Hence we obtain (iii). This completes the proof.
Proof of the characterization theorem of the trichotomy {#sec: trichotomy}
=======================================================
First, we prove Theorem \[Intro trichotomy\]. Before proving it, we need the following.
\[van Kampen1\] The set $ \sH = \sH^s_\mu $ is a [*submodule*]{} of $ \sG $, i.e., the following statements hold:\
(0) If $ \rho_1 \in \sH $ and if $ \rho_2 $ is equivalent to $ \rho_1 $, then $ \rho_2 \in \sH $.\
(i) If $ \rho_1,\rho_2 \in \sH $, then $ \rho_1 \otimes \rho_2 \in \sH $.\
(ii) If $ \rho_1,\rho_2 \in \sH $, then $ \rho_1 \oplus \rho_2 \in \sH $.\
(iii) If $ \rho \in \sH $, then $ \bar{\rho} \in \sH $. Here $ \bar{\rho} $ denotes the complex conjugate representation.
This is obvious, so we omit the proof.
We utilize the following fact.
\[Kampen\] Let $ \sH $ be a submodule of $ \sG $. Suppose that $$\mbox{$ \rho(g) = {\rm id} $ for every $ \rho \in \sH $ \ $ \Longrightarrow $ \ $ g=e $ }.$$ Then $ \sH = \sG $.
This fact plays a key role in the proof of the [*Tannaka duality*]{} in the representation theory of compact groups. For the proof of Lemma \[Kampen\], see, e.g., [@Tann Theorem 13.1] and [@JS Lemma 17].
Now we proceed to prove Theorem \[Intro trichotomy\].
The claim (i) is obvious by definition of the isotropic subgroup $ H_\mu(\bP^o) $. The claim (iii) follows immediately from (i) and (ii). Thus we need only to prove the claim (ii).
$ 1^{\circ}). $ Suppose that the case [[$ ({\bf C2}) $]{}]{} occurs, i.e., that there exists a strong solution $ \bP \in \sP_\mu $. Then it holds that $ \F^{\eta}_{-\infty } \subset \F^{\xi}_{-\infty } $ under $ \bP $. Since $ (\xi_k : \, k \in -\N) $ is an independent sequence, Kolmogorov’s 0-1 law holds so that $ \F^{\eta}_{-\infty } $ is $ \bP $-trivial. Then Lemma \[lem: tail\] says that the solution $ \bP $ must be an extremal point: $ \bP \in \ext $. Since $ \sH^s_\mu(\bP) = \sG $, we obtain $ H^s_\mu = {\left\{ {e}\right\}} $.
Theorem \[Gal1\] says that all the solutions of $ \bP' \in \ext $ are obtained by $ \bP' = T_g(\bP) $ for some $ g \in G $. Then it is clear that the solution $ \bP' $ is also strong. Therefore we obtain the last claim: All extremal solutions are strong and the others non-strong.
$ 2^{\circ}). $ Suppose that $ H^s_\mu = \{ {e}\} $. Then we see that $$\mbox{$ \rho(g) = {\rm id} $ for every $ \rho \in \sH^s_\mu $
\ $ \Longrightarrow $ \ $ g = {e}$}.$$ Applying this fact and Lemma \[van Kampen1\] to Theorem \[Kampen\], we conclude that $ \sH^s_\mu $ coincides with the whole $ \sG $. This shows that all the extremal points are strong and that the case [[$ ({\bf C2}) $]{}]{} occurs.
Second, we prove the inclusions in Theorem \[thm: isot group\]. This is an immediate consequence of the following proposition, which generalizes Proposition 2 of [@Yor].
\[Gal3\] Let $ \mu $ be given and let $ \bP \in \sP_\mu $.\
[(i)]{} If $ \rho \in \sH^2_\mu $, then $$\bE {\left[ \rho(\eta_k) \mid \F^{\eta}_{-\infty } \vee \F^{\xi}_k \right]}
= \rho(\eta_k)
, \quad k \in -\N .
\label{eq:3}$$ If, moreover, $ \bP \in \ext $, then $ \rho(\eta_k) $ is $ \F^{\xi}_k $-measurable $ \bP $-a.s. for any $ k \in -\N $.\
[(ii)]{} If $ \rho \notin \sH^1_\mu $, then $$\bE {\left[ \rho(\eta_k) \mid \F^{\eta}_{-\infty } \vee \F^{\xi}_k \right]}
= O
, \quad k \in -\N .
\label{eq:5}$$ If, moreover, $ \bP \in \ext $, then $ \rho(\eta_k) $ is [*never*]{} $ \F^{\xi}_k $-measurable $ \bP $-a.s. for any $ k \in -\N $.
\(i) To prove the claim, it suffices to show that for arbitrary small $ k \in -\N $.
Since $ \rho \in \sH^2_\mu $, it holds that $$\lim_{k \to -\infty } \prod_{j = k }^{k_0} | \det R_j | > 0$$ for arbitrary small $ k_0 \in -\N $. Iterating the equation , we have $$\rho(\eta_{k_0})
= \rho(\eta_{k_0-n}) \Xi_n , \quad n \in \N$$ where $$\Xi_n = \rho(\xi_{k_0-n+1}) \rho(\xi_{k_0-n+2})
\cdots \rho(\xi_{k_0}) , \quad n \in \N .$$ Since $ \det \bE {\left[ \Xi_n \right]} \neq 0 $ for $ n \in \N $, we can define $$\Phi_n = {\left( \bE {\left[ \Xi_n \right]} \right)}^{-1} \Xi_n , \quad n \in \N .$$ Then the sequence $ (\Phi_n : \, n \in \N) $ constitutes a matrix-valued bounded $ (\E^{k_0}_{n}) $-martingale[^4] where $$\E^{k_0}_{n} = \sigma ( \xi_{k_0}, \xi_{k_0-1}, \ldots, \xi_{k_0-n+1} ) ,
\quad n \in \N .$$ Therefore $ \Phi_n $ converges to an $ \F^{\xi}_{k_0} $-measurable $V^{\rho} \otimes V^{\rho} $-valued random element $ \Phi_{\infty } $ almost surely. Since $$| \det \Phi_n | = {\left( \det \bE {\left[ \Xi_n \right]} \right)}^{-1} , \quad n \in \N ,$$ we obtain $ \det \Phi_{\infty } \neq 0 $ almost surely. Taking subsequence if necessary, we see that $$\begin{aligned}
\Psi_n = \rho(\eta_{k_0-n}) \bE {\left[ \Xi_n \right]}
= \rho(\eta_{k_0}) (\Phi_n)^{-1} . \end{aligned}$$ converges to an $\F^{\eta}_{-\infty }$-measurable $V^{\rho} \otimes V^{\rho} $-valued random element $ \Psi_{\infty } $ almost surely. Therefore we conclude that $ \rho(\eta_{k_0}) = \Psi_{\infty } \Phi_{\infty }$ is $ \F^{\eta}_{-\infty } \vee \F^{\xi}_{k_0} $-measurable.
\(ii) We can easily prove the claim by imitating the proof of Proposition 2 of [@Yor]. So we omit the proof.
Third, we prove the rest of Theorem \[thm: isot group\].
$ 1^{\circ}). $ It is obvious by definition that $ H^1_\mu $, $ H^2_\mu $ and $ H^s_\mu $ are closed normal subgroups of $ G $.
$ 2^{\circ}). $ Let $ \bP^o \in \ext $ be fixed. We seek an equivalent expression of the condition that $ g \in H_\mu(\bP^o) $. Note that $ T_g(\bP^o) = \bP^o $ if and only if $$\bP^o (g \cdot \eta_k \in \cdot) = \bP^o (\eta_k \in \cdot)
, \quad k \in -\N ,$$ which is equivalent to $$\rho(g) \bE^o [ \rho(\eta_k) ] = \bE^o [ \rho(\eta_k) ]
, \quad k \in -\N , \ \rho \in \sG
\label{Gal2 eq1}$$ by the Peter–Weyl theorem.
$ 3^{\circ}). $ Suppose that $ g \in H^1_\mu $. Let $ \rho \notin \sH^1_\mu $ and $ \bP^o \in \ext $. Noting that $$\bE^o [ \rho(\eta_k) \mid \F^{\xi}_{k-n} ]
= \rho(\eta_{k-n})
R_{k-n+1} R_{k-n+1} \cdots R_k ,$$ we have $$\left\| \bE^o [ \rho(\eta_k) \mid \F^{\xi}_{k-n} ] \right\|
\le \| R_{k-n+1} R_{k-n+1} \cdots R_k \| .$$ Letting $ n \to \infty $, we have $$\bE^o [\rho(\eta_k) \mid \F^{\xi}_{-\infty } ] = O .$$ Since $ \F^{\xi}_{-\infty } $ is $ \bP^o $-trivial, we obtain $$\bE^o [ \rho(\eta_k) ] = O ,$$ and then we see that holds, which proves that $ H^1_\mu \subset H_\mu (\bP^o) $ for any $ \bP^o \in \ext $.
$ 4^{\circ}). $ Suppose that $ g \in H_\mu(\bP^o) $. Let $ \rho \in \sH^s_\mu $. Since $ T_g(\bP^o)=\bP^o $, we have the following identity between two joint laws on $ G^{-\N} \times G^{-\N} $: $$\bP^o {\left( (g \cdot \eta, \xi) \in \cdot \right)} = \bP^o {\left( (\eta,\xi) \in \cdot \right)} .$$ Thus we have the following identity between two regular conditional distributions on $ G^{-\N} $ given $ \F^{\xi} $ (cf. Proof of Lemma \[coupling\]): $$\bP^o_{\xi} {\left( g \cdot \eta \in \cdot \right)}
= \bP^o_{\xi} {\left( \eta \in \cdot \right)}
\quad \mu-\mbox{a.s.}$$ Hence we have $$\bE^o_{\xi} {\left[ \rho(g) \rho(\eta) \right]}
= \bE^o_{\xi} {\left[ \rho(\eta) \right]}
\quad \mu-\mbox{a.s.}$$ Since $ \rho(\eta) $ is $ \F^{\xi} $-measurable, we obtain $ \rho(g) \rho(\eta) = \rho(\eta) $, $ \mu $-a.s., which implies that $ \rho(g)={\rm id} $. Therefore we obtain the inclusion $ H_\mu(\bP^o) \subset H^s_\mu $.
$ 5^{\circ}). $ The inclusion $ H^s_\mu \subset H^2_\mu $ follows from the inclusion $ \sH^s_\mu \supset \sH^2_\mu $, which is assured by Proposition \[Gal3\].
$ 6^{\circ}). $ If $ G $ is abelian, then all irreducible representations of $ G $ are one-dimensional, so we obtain $ \sH^1_\mu = \sH^2_\mu $, which implies .
Finally, we prove the following lemma, which assures the convergence of $ r^2_k[\rho] $.
\[lem: det ineq\] Let $ \mu_1, \ldots, \mu_n $ be probability measures on $ \{ x \in \C^n : \, |x| := (|x_1|^2 + \cdots + |x_n|^2)^{1/2} \le 1 \} $ and set $$u_i = \int x \mu_i(dx), \quad i = 1, 2, \ldots, n .$$ Then $$| \det (u_1 \cdots u_n) | \le 1 .$$
[*Proof.*]{}[^5] By Hadamard’s inequality, we have $$| \det ( u_1,\ldots,u_n ) | \le \prod_i |u_i| .$$ By Jensen’s inequality, we have $$|u_i| \le \int |x| \mu_i(dx) \le 1 , \quad i = 1, 2, \ldots, n .$$ This completes the proof.
Tsirelson–Yor equations {#sec: TYE}
=======================
Let $ S $ be a Polish space and $ G $ a compact group. Let $ \theta : S \to G $ be a measurable map and let $ \psi : G \times S \to S $ and $\psi^{-1}:G \times S \to S $ be two measurable maps such that $$\label{TsiYor1}
\psi^{-1}( g, \psi(g,s) ) = \psi( g, \psi^{-1}(g,s) ) = s ,
\quad g \in G , \ s \in S .$$ We consider the following stochastic equation in discrete negative time: $$\eta_k = \psi {\left( \big. \theta (\eta_{k-1}) , \xi_k \right)}, \quad k \in -\N .
\tag{\ref{TYE}}$$ Moreover, we assume the following.
\[UC\] The mappings $ \psi $ and $ \theta $ commute in the sense that $$\theta ( \psi(g,x) ) = g \cdot \theta (x) , \quad g \in G .$$
Let $ S,G,\psi,\psi^{-1} $ and $ \theta $ as above and assume that Assumption \[UC\] is satisfied. Then the stochastic equation is called a [*Tsirelson–Yor equation*]{}, which will be abbreviated by “TYE".
Following the case of STYE’s, we introduce the notion of a solution as follows. Let $ \eta = ( \eta_k , k \in -\N ) $ denote the coordinate mapping process on $ S^{-\N} $ and set $$\xi_k = \psi^{-1} {\left( \big. \theta {\left( \eta_{k-1} \right)} , \eta_k \right)},
\quad k \in -\N.
\label{TYE xi}$$ The filtrations $ (\F^{\eta}_k) $ and $ (\F^{\xi}_k) $ are defined in the same way.
\[Def sol2\] Let $ \mu = (\mu_k , k \in -\N) $ be a sequence of Borel probability measures $ \mu_k $ on $ S $. A [*solution*]{} of the TYE with the noise law $ \mu $ is a probability measure $ \bP $ on $ S^{-\N} $ such that the following two statements hold:\
(i) $ \xi_k $ is independent of $ \F^{\eta}_{k-1} $ under $ \bP $, for any $ k \in -\N $.\
(ii) $ \xi_k $ is distributed as $ \mu_k $ under $ \bP $, for any $ k \in -\N $.
The totality of solutions of the TYE with the noise law $ \mu $ will be denoted by $ \sP_\mu $.
We adopt the same notions of strong solutions and uniqueness in law as are defined in Definitions \[Def s sol\] and \[Def uni\].
If $ S=G $, $ \psi(g,s) $ is the product and $ \theta $ is the identity mapping, then the TYE is exactly the STYE on $ G $ and all the notions of a solution, a strong solution and uniqueness in law coincide.
Consider a TYE with a given noise law $ \mu $. Denote $ {\widehat}{\eta}_k = \theta(\eta_k) $, $ {\widehat}{\mu} = \mu \circ \theta^{-\N} $ and so on. Then Assumption \[UC\] implies that $${\widehat}{\eta}_k = {\widehat}{\eta}_{k-1} \cdot {\widehat}{\xi}_k ,
\quad k \in -\N.
\label{hat eta}$$ This is nothing but the STYE on $ G $ with the noise law $ {\widehat}{\mu} $. We define $${\widehat}{\bP} = \text{the law of $ {\widehat}{\eta} $ on $ G^{-\N} $ under $ \bP $}.$$ Then we obtain a mapping $$\sP_\mu \ni \bP \mapsto {\widehat}{\bP} \in {\widehat}{\sP}_{{\widehat}{\mu}}
\label{hat}$$ Here $ {\widehat}{\sP}_{{\widehat}{\mu}} $ denotes the set of solutions of the STYE with the noise law $ {\widehat}{\mu} $.
Let us give two examples.
\[ex: Yor eq\] Let $ S=\R $ and $ G=\T = \R/\Z \simeq [0,1) $. For $ g \in [0,1) $ and $ s \in \R $, we set $ \psi(g,s) = s+g $ and $ \psi^{-1}(g,s) = s-g $. Set $ \theta(s) = \alpha (s) $ for $ s \in \R $. Then the TYE coincides with the equation $$\eta_k = \alpha (\eta_{k-1}) + \xi_k ,
\quad k \in -\N .
\tag{\ref{Yor eq}}$$ Then the equation for $ {\widehat}{\eta_k} = \alpha (\eta_k) $ is the STYE on $ \T $, which is actually .
\[ex: Tanaka eq\] Let $ S=\R \setminus {\left\{ 0 \right\}} $ and $ G = \Z/2\Z \simeq {\left\{ 1,-1 \right\}} $. For $ g=\pm 1 $ and $ s \in \R \setminus {\left\{ 0 \right\}} $, we set $ \psi(\pm 1,s) = \pm s $ and $ \theta(s) = \sgn (s) $. Then the TYE coincides with the equation $$\eta_k = \sgn {\left( \eta_{k-1} \right)} \cdot \xi_k ,
\quad k \in -\N .
\tag{\ref{Tanaka}}$$ Then the equation for $ {\widehat}{\eta_k} = \sgn {\left( \eta_k \right)} $ is the STYE on $ \Z/2\Z $, which is actually .
Let $ g \in G $. For the coordinate process $ (\eta_k : \, k \in -\N) $ on $ S^{-\N} $ and the process $ (\xi_k : \, \in -\N ) $ defined by , we set $$\eta'_k := \psi ( g \cdot \theta (\eta_{k-1}) , \xi_k ) , \quad k \in -\N .
\label{G action1}$$ For $ \bP \in \sP_\mu $ for some $ \mu $, we define $ T_g(\bP) $ by the law of the process $ (\eta'_k : \, k \in -\N) $ under $ \bP $.
Now we have the following.
\[reduction\] (i) The mapping is bijective.\
(ii) $ \bP \in \sP_\mu $ is strong iff so is $ {\widehat}{\bP} \in {\widehat}{\sP}_{{\widehat}{\mu}} $.\
(iii) $ \bP \in \sP_\mu $ is extremal iff so is $ {\widehat}{\bP} \in {\widehat}{\sP}_{{\widehat}{\mu}} $.\
(iv) The family of mappings $ (T_g: \, g \in G) $ defines a group action on $ \sP_\mu $ and its restriction on $ \ext \cap \sP_\mu $ is transitive.\
(v) The case [[$ ({\bf C1}) $]{}]{}, [[$ ({\bf C2}) $]{}]{} or [[$ ({\bf C3}) $]{}]{} occurs for the TYE with the noise law $ \mu $ iff so does for the STYE on $ G $ with the noise law $ {\widehat}{\mu} $, accordingly.
Let $ \bP' \in {\widehat}{\sP}_{{\widehat}{\mu}} $ be given. For any $ k \in -\N $, we define a probability measure $ \pi_k $ on $ (G \times S)^{ {\left\{ k, \ldots, 0 \right\}} } $ in the following way: Let $ (U_k , \xi_k , \xi_{k+1} , \ldots, \xi_0) $ be a family of independent random variables such that $ U_k $ is a $ G $-valued random variable distributed as $ \bP'(\eta_k \in \cdot) $ and $ \xi_j $ is an $ S $-valued random variable distributed as $ \mu_j $ for $ j=k,\ldots,0 $. Set $ \eta_k = \psi ( U_k , \xi_k ) $, $$\eta_j = \psi ( \theta (\eta_{j-1}), \xi_j ) , \quad j = k+1,\ldots,0$$ and $$U_j = \theta (\eta_j) , \quad j = k+1,\ldots,0 .$$ Then we define $$\mbox{$ \pi_k = $ the law of $ ( (U_j,\eta_{j}) : \, j = k, k+1, \ldots, 0 ) $}.$$
Thanks to the consistency assumption , we see that the family $ \{ \pi_k : \, k \in -\N \} $ satisfies Kolmogorov’s consistency condition. Therefore Kolmogorov’s extension theorem ensures the existence of a probability measure $ \bQ $ on $ (G \times S)^{- \N} $ whose projection on $ (G \times S)^{ \{ k, \ldots, 0 \} } $ coincides with $\pi_k$ for $ k \in -\N $. If we define $ \bP $ by the projection of $ \bQ $ on $ S^{-\N} $, then we obtain $ {\widehat}{\bP} = \bP' $. Therefore we conclude that the mapping is surjective.
The rest of the claims are obvious, so we omit their proofs.
[1]{} Émery, M. and Schachermayer, W., [*A remark on Tsirelson’s stochastic differential equation*]{}, Sém. Prob., XXXIII, 291–303, Lecture Notes in Math., [**1709**]{}, Springer, Berlin, 1999.
Ikeda, N. and Watanabe, S., [*Stochastic Differential Equations and Diffusion Processes*]{}, North Holland-Kodansha, Amsterdam and Tokyo, 1981.
Joyal, A. and Street, R., [*An introduction to Tannaka duality and quantum groups*]{}, Category theory (Como, 1990), Lecture Notes in Math., [**1488**]{} (1991), 411–492.
Kallsen, J., [*A stochastic differential equation with a unique (up to indistinguishability) but not strong solution*]{}, Sém. Prob., XXXIII, 315–326, Lecture Notes in Math., [**1709**]{}, Springer, Berlin, 1999.
van Kampen, E. R., [*Almost periodic functions and compact groups*]{}, Ann. of Math., [**37**]{} (1936), No. 1, 78–91.
Le Gall, J. F. and Yor, M., [*Sur l’équation stochastique de Tsirelson*]{}, Sem. Prob. XVII, Lecture Notes in Math., [**986**]{} (1983), 81–88.
Liptser, R. S. and Shiryayev, A. N., [*Statistics of random processes [I]{}*]{}, General theory; translated by A. B. Aries, Applications of Mathematics, Vol. 5, Springer-Verlag, New York, 1977.
Revuz, D. and Yor, M., [*Continuous martingales and Brownian motion*]{}, third edition, Springer-Verlag, Berlin, 1999.
Stroock, D. W. and Yor, M., [*On extremal solutions of martingale problems*]{}, Ann. Sci. Ecole Norm. Sup. IV, Ser. [**13**]{} (1980), no. 1, 95–164.
Sugiura, M., [*Unitary representations and harmonic analysis, An introduction*]{}, Kodansha Ltd., Tokyo, Halstead Press \[John Wiley & Sons\], New York-London-Sydney, 1975.
Tannaka, T., [*Sôtsui genri (Duality principle)*]{} (Japanese), Iwanami shoten, 1966.
Tsirelson, B., [*An example of a stochastic differential equation having no strong solution*]{}, Theory Probab. Appl. [**20**]{} (1975), 416–418; translated from Teor. Veroyatn. Primen. [**20**]{} (1975), 427–430.
Uenishi, C., [*On weak solutions of stochastic differential equations*]{} (Japanese), Master Thesis, March 2004, Ritsumeikan University.
Yamada, T. and Watanabe, S., [*On the uniqueness of solutions of stochastic differential equations*]{}, J. Math. Kyoto Univ., [**11**]{} (1971), 155–167.
Yor, M., [*Tsirel’son’s equation in discrete time*]{}, Probab. Theory Related Fields, [**91**]{} (1992), 135–152.
Zvonkin, A. K. and Krylov, N. V., [*Strong solutions of stochastic differential equations*]{}, Selecta Math. Sovietica, [**1**]{} (1981), no. 1, 19–61; translated by A. B. Aries, Proceedings of the School and Seminar on the Theory of Random Processes (Druskininkai, 1974), Part II (Russian) (1975), 9–88.
[^1]: The condition was discovered in [@Uen].
[^2]: It is sometimes called [*weak*]{}.
[^3]: Here the integral in RHS of is interpreted as the componentwise integral in a fixed matrix representation of $ \rho $.
[^4]: We mean that $ \bE {\left[ \Phi_{n+1} \mid \E^{k_0}_{n} \right]} = \Phi_{n} $.
[^5]: The authors are informed of this simple proof by Y. Takahashi.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We investigate the effect of charge carrier interaction with surface optical phonons on the band properties of monolayer black phosphorus induced by polar substrates. We develop an analytical method based on the Lee-Low-Pines theory to calculate the spectrum of Fr[ö]{}hlich type continuum Hamiltonian in the long-wavelength limit. We examine the modification of a band gap and renormalization of effective masses due to the substrate-related polaronic effect. Our results show that an energy gap in supported monolayer black phosphorus is enlarged depending on a particular substrate and the interlayer distance, $z$. Among the substrate considered, the largest gap broadening at $ z=2.5$ Å is observed for the Al$ _{2} $O$ _{3} $ substrate, which is found to be $ \sim 50$ meV. Carrier-phonon coupling also renormalizes the effective masses which is more pronounced along the zigzag direction. Anisotropy of the effective masses becomes stronger by the influence of the polaronic effect corresponding to direction-dependent carrier-phonon coupling. We conclude that substrate phonons have a non-negligible effect on the static band properties of monolayer black phosphorus, which may be further exploited in its experimental and theoretical studies.'
author:
- 'A. Mogulkoc'
- 'Y. Mogulkoc'
- 'A. N. Rudenko'
- 'M. I. Katsnelson'
bibliography:
- 'referans.bib'
title: Polaronic effects in monolayer black phosphorus on polar substrates
---
Introduction
============
With the discovery of graphene [@novoselov2004electric], two-dimensional (2D) materials have become the focus of many researchers. Recently, a few-layer black phosphorus (BP) has emerged as one of the promising 2D material owing to its unique physical properties. Its peculiar structure leads significantly anisotropic electronic and optical properties [@Li:2014aa; @doi:10.1021/nn501226z; @Xia:2014aa; @Reich2014; @koenig2014electric; @castellanos2014isolation; @Low:2014aa]. Similar to graphene, BP can be mechanically exfoliated [@Li:2014aa; @doi:10.1021/nn501226z] down to a few layers sample. Beside this, few-layer BP has been obtained using liquid-phase exfoliation [@brent2014production; @yasaei2015high] and plasma-assisted fabrication method [@lu2014plasma]. Chemically, BP is less stable than graphene and it degrades quickly in the atmospheric environment [@island2015environmental]. To protect BP samples from the degradation, their encapsulation in other materials is used to achieve better performance of BP-based devices. For example, high carrier mobility of $ \sim $1.350 $ cm^{2}V^{-1}s^{-1} $ at room temperature and high on-off ratios exceeding $ \sim10^{5} $ in a few layer BP encapsulated by atomically thin hexagonal boron nitride (h-BN), which forms a BN-BP-BN heterostructures is demonstrated by Chen *et al.* [@Chen:2015aa]. Interaction with surfaces is, therefore, an important aspect affecting the properties of BP in its practical applications.
Lattice vibrations play an important role in the dynamics of charge carriers in 2D materials [@Samsonidze:2007aa; @Stauber:2007aa; @Hague:2011aa; @Hague:2012aa; @Hague:2014aa; @Kandemir:2012aa; @kandemir2014zone; @wang2015energy; @kandemir2015chiral; @Mogulkoc:2015aa; @mogulkoc2015ab]. Particularly, optical phonons of polar substrates localized around sample-substrate interface affect the behavior of charge carrier, which depends strongly on the phonon frequencies and polarizability of a substrate [@Chen:2008aa; @Fratini:2008aa; @konar2010effect; @Hague:2011aa; @Zhang:2013aa; @Hwang:2013aa; @Tan2013aa; @Scharf:2013aa; @wang2015energy]. Similar to the effect of strain [@Peng:2014aa; @doi:10.1021/nl500935z], interaction of electrons with optical phonons can open a band gap or allow us to tune the existing band gap of the system. Band gap engineering is an important field aiming at tuning on energy gap of material for potential applications in nanoelectronics. For instance, the electron-phonon interaction can induce a small gap in the band structure of graphene [@Samsonidze:2007aa; @Hague:2011aa; @Hague:2012aa; @Hague:2014aa; @kandemir2014zone; @wang2015energy; @kandemir2015chiral]. Moreover, Wang *et al.* [@wang2015energy] found an energy gap in the zeroth Landau level due to the electron-surface optical phonon interaction arising from the polar substrate, this gap can be tuned by choosing the polarization of substrates and changing the distance between the substrate and graphene. Based on density functional theory calculations, Hu and Hong [@hu2015anisotropic] have recently found that the encapsulation of BP layers into h-BN changes the band structure affecting the band gap and effective masses without changing the anisotropic electrical and optical properties of BP layers. Such modification is a result of a single-particle approximation, while the role of the many-body effects remains unclear.
Here, we investigate another mechanism responsible for the formation of band properties of encapsulated or supported monolayer black phosphorus (MBP), namely the surface polaronic effect. We consider an analytical model based on the Lee-Low-Pines (LLP) theory [@lee1953motion] which is based on a $ \boldsymbol{k}\cdot \boldsymbol{p} $-type Hamiltonian for MBP. We apply a series of unitary transformations to diagonalize the Fr[ö]{}hlich type many-body Hamiltonian and use the variational method to find the eigenvalues of the total Hamiltonian. The band structure of MBP is then investigated as a function of model parameters, such as phonon energy, $ \hbar \omega $, effective screening constant, $ \xi $, and the distance between substrate and MBP, $ z $. Finally, the gap modifications and renormalization of the effective masses are examined by considering a number of typical polar substrates (h-BN, HfO$ _{2} $, SiC, SiO$ _{2} $, Al$ _{2} $O$ _{3} $).
Theory
======
The energy spectrum of MBP can be described by a tight-binding model [@Rudenko:2014aa; @Rudenko:2015aa] and effective $ \boldsymbol{k}\cdot \boldsymbol{p} $ model [@Rodin:2014aa; @1504.02452v1; @Zhou:2015aa; @ezawa2014topological] as a continuum approach, which shows good agreement with the tight-binding results in the range of $ -2.0$ to $1.5 $ eV [@1504.02452v1; @Zhou:2015aa]. The band structure of MBP exhibits a direct band gap of $ 1.5 \sim 2.0 $ eV at the $ \Gamma $ point of the Brillouin zone [@doi:10.1021/nn501226z; @Rudenko:2014aa; @Rudenko:2015aa]. In the long-wavelength regime, the continuum Hamiltonian of MBP [@1504.02452v1] with bare phonon field and electron-surface optical phonon Hamiltonian can be written as, $$\mathcal{H}=\mathcal{H}_{0}+\sum\limits_{\mu }\sum\limits_{\boldsymbol{q}%
}\hbar \omega _{\mu }b_{\boldsymbol{q}}^{\dag }b_{%
\boldsymbol{q}}+\mathcal{H}_{e-p} \label{1}$$with $$\mathcal{H}_{0}=\left[
\begin{array}{cc}
u_{0}+\bar{\eta}_{x}p_{x}^{2}+\bar{\eta}_{y}p_{y}^{2} & \delta+\bar{\gamma}_{x}p_{x}^{2}+\bar{\gamma}_{y}p_{y}^{2}+i\bar{\chi}p_{y} \\
\delta+\bar{\gamma}_{x}p_{x}^{2}+\bar{\gamma}_{y}p_{y}^{2}-i\bar{\chi}p_{y} & u_{0}+\bar{\eta}_{x}p_{x}^{2}+\bar{\eta}_{y}p_{y}^{2}
\end{array}%
\right] \label{2}$$and $$\mathcal{H}_{e-p}=\sum\limits_{\mu}\sum\limits_{\boldsymbol{q}}\left[ \mathcal{M}_{\mu \boldsymbol{q}}b_{\boldsymbol{q}}e^{i\boldsymbol{q}\cdot\boldsymbol{r}}+h.c.\right], \label{3}$$where $\mathcal{M}_{\mu\boldsymbol{q}}=\sqrt{e^{2}\xi\hbar\omega_{\mu}/2A\epsilon_{0}q}e^{-qz}$ is the amplitude of carrier-surface optical phonon interaction [@Wang:1972aa; @Mori:1989aa; @Sak:1972aa], $ b_{\boldsymbol{k}}^{\dag }(b_{\boldsymbol{k}}) $ is the phonon creation (annihilation) operator, $ e $ is the elementary electric charge, $ \xi =\left( \kappa_{0}-\kappa_{\infty}\right)/\left[\left(\kappa_{0}+1 \right)\left(\kappa_{\infty}+1 \right) \right] $ is the effective screening constant related to the dielectric constants of a substrate, $ \omega_{\mu} $ is the longitudinal surface optical phonon frequency of the $ \mu^{th} $ branch ($ \mu=1,2 $), $ z $ is the distance between the substrate and MBP \[see Fig.(\[FIGURE1\])\], $ \kappa_{\infty} $ ($ \kappa_{0} $) is the high (low) frequency dielectric constant of the substrate and $ \epsilon_{0} $ is the vacuum permittivity. Here, $ \bar{\eta}_i=\eta_{i}/\hbar^{2} $ ($ \eta_{x}=0.58 $ eVÅ$ ^{2} $ and $ \eta_{y}=1.01 $ eVÅ$ ^{2} $), $ \bar{\gamma}_i=\gamma_{i}/\hbar^{2} $ ($ \gamma_{x}=3.93 $ eVÅ$ ^{2} $ and $ \gamma_{y}=3.83 $ eVÅ$ ^{2} $), $ \bar{\chi}=\chi/\hbar $ ($ \chi=5.25 $ eVÅ), $ u_{0}=-0.42 $ eV and $ \delta=0.76 $ eV. The total Hamiltonian given by Eq.(\[1\]) can be rewritten in a more compact form as,
$$\begin{aligned}
\mathcal{H}&=&\left( u_{0}+\sum\limits_{i=x,y}\bar{\eta}_{i}p_{i}^{2}\right)\sigma_{0}+\left(\delta +\sum\limits_{i=x,y}\bar{\gamma}_{i}p_{i}^{2}\right)\sigma_{1}-\left( \bar{\chi} p_{y} \right)\sigma_{2} \notag \\
&+& \left(\ \sum\limits_{\boldsymbol{q} \mu}\mathcal{M}_{\mu \boldsymbol{q}}b_{\boldsymbol{q}}e^{i\boldsymbol{q}\cdot\boldsymbol{r}}+h.c.\right)\sigma_{0}+\left( \sum\limits_{\mu }\sum\limits_{\boldsymbol{q}%
}\hbar \omega _{\mu }b_{\boldsymbol{q}}^{\dag }b_{%
\boldsymbol{q}}\right) \sigma_{0} \label{4}\end{aligned}$$
Here, $ \sigma_{i} $ $ \left(i=1,2,3 \right) $ are the Pauli matrices, and $ \sigma_{0} $ is the identity matrix. Note that, electron-phonon interaction and bare phonon field are involved in the low-energy Hamiltonian with identity matrix which equally affects both sublattices. On the other hand, we assume that the substrate is isotropic and the interlayer interaction is weak which preserves C$_{2h}$ group invariance of the supported MBP lattice. To diagonalize the Hamiltonian in Eq.(\[4\]), we introduce two unitary transformation given by $$\begin{aligned}
U_{1}&=&\exp \left[ -i\boldsymbol{r}\mathbf{\cdot }\sum\limits_{%
\boldsymbol{q} \mu}\boldsymbol{q} b_{\boldsymbol{q}%
}^{\dag }b_{\boldsymbol{q}}\right] \notag \\
U_{2}&=&\exp \left[\sum\limits_{\boldsymbol{q}%
}\left(\textit{f}_{\boldsymbol{q}} b_{\boldsymbol{q}}^{\dag}-\textit{f}_{\boldsymbol{q}}^{*} b_{\boldsymbol{q}}\right) \right].\notag\end{aligned}$$ The first transformation eliminates the electron coordinates from Eq.(\[4\]) and transforms the phononic and momentum operators by $ b_{\boldsymbol{q}}\rightarrow b_{\boldsymbol{q}}\exp\left(i \boldsymbol{q} \cdot \boldsymbol{r} \right) $ and $ p_{i}\rightarrow p_{i}-\sum_{\boldsymbol{q}} \hbar \boldsymbol{q}b_{\boldsymbol{q}}^{\dag }b_{%
\boldsymbol{q}}$, respectively. The second transformation is the well-known LLP transformation [@lee1953motion], where $ \textit{f}_{\boldsymbol{q}} $ is considered to be a variational function. This transformation enables us to consider the dressed electron states due to the phonon field induced by the substrate which yields the polarization in the lattice. Moreover, it generates the coherent states for the phonon subsystem such that optical phonon operators transform according to the rule $ b_{\boldsymbol{q}}\rightarrow b_{\boldsymbol{q}}+\textit{f}_{q} $. Under the two successive unitary transformation $ \widetilde{\mathcal{H}}=U_{2}^{-1}U_{1}^{-1}\mathcal{H}U_{1}U_{2} $, the transformed Hamiltonian takes the form $$\begin{aligned}
\widetilde{\mathcal{H}}&=&\left[u_{0}+\sum\limits_{i=x,y}\bar{\eta}_{i}p_{i}^{2} +\sum\limits_{i=x,y} \bar{\eta}_{i} \left( \sum\limits_{\boldsymbol{q}}\hbar q_{i} \left|\textit{f}_{\boldsymbol{q}} \right|^{2} \right)^{2}+\sum\limits_{\boldsymbol{q} \mu} \left( \mathcal{M}_{\mu \boldsymbol{q}} \textit{f}_{\boldsymbol{q}}+\mathcal{M}_{\mu \boldsymbol{q}}^{*} \textit{f}_{\boldsymbol{q}}^{*}\right) \right]\sigma_{0} \notag \\
&+& \left[ \sum\limits_{\boldsymbol{q}}\left|\textit{f}_{\boldsymbol{q}} \right|^{2} \left( \hbar \omega_{\mu}-2\hbar\sum\limits_{i=x,y}\bar{\eta}_{i} p_{i}q_{i}+\hbar^{2}\sum\limits_{i=x,y}\bar{\eta}_{i}q_{i}^{2}\right) \right] \sigma_{0}+\left[\delta+\sum\limits_{i=x,y}\bar{\gamma}_{i}p_{i}^{2} +\sum\limits_{i=x,y} \bar{\gamma}_{i} \left( \sum\limits_{\boldsymbol{q}}\hbar q_{i} \left|\textit{f}_{\boldsymbol{q}} \right|^{2} \right)^{2} \right] \sigma_{1} \notag \\
&+&\left[ \sum\limits_{\boldsymbol{q}}\left|\textit{f}_{\boldsymbol{q} } \right|^{2} \left(-2\hbar\sum\limits_{i=x,y}\bar{\gamma}_{i} p_{i}q_{i}+\hbar^{2}\sum\limits_{i=x,y}\bar{\gamma}_{i}q_{i}^{2}\right) \right] \sigma_{1}-\bar{\chi}\left(p_{y}- \sum\limits_{\boldsymbol{q}}\hbar q_{y} \left|\textit{f}_{\boldsymbol{q}} \right|^{2}\right)\sigma_{2}+\left( \sum\limits_{\boldsymbol{q} \mu%
}\hbar \omega _{\mu }b_{\boldsymbol{q}}^{\dag }b_{%
\boldsymbol{q}}\right) \sigma_{0} \notag \\
&+&\sum\limits_{\boldsymbol{q}%
} \left[ \left( \sum\limits_{i=x,y} 2\hbar^{2} \bar{\eta}_{i}q_{i}\sum\limits_{\boldsymbol{q^{'}}} q_{i}^{'} \left|\textit{f}_{\boldsymbol{q}^{'}} \right|^{2}\right) \sigma_{0}+ \left( \sum\limits_{i=x,y} 2\hbar^{2} \bar{\gamma}_{i}q_{i}\sum\limits_{\boldsymbol{q^{'}}} q_{i}^{'} \left|\textit{f}_{\boldsymbol{q}^{'}} \right|^{2}\right) \sigma_{1} \right]b_{\boldsymbol{q}}^{\dag }b_{%
\boldsymbol{q}}+\mathcal{\widetilde{H}}_{ND}, \label{5}\end{aligned}$$ where $ \mathcal{\widetilde{H}}_{ND} $ includes nondiagonal elements in terms of phonon creation and annihilation operators. It is also convenient to express the kinetic energy and linear momentum operators in the second quantization form as, $$\begin{aligned}
\bar{\eta}_{i}p_{i}^{2}&=&\hbar\sum\limits_{\boldsymbol{k}}\eta_{i} k_{i}^{2}a_{\boldsymbol{k}}^{\dag }a_{%
\boldsymbol{k}} \notag \\
\bar{\gamma}_{i}p_{i}^{2}&=&\hbar\sum\limits_{\boldsymbol{k}}\gamma_{i} k_{i}^{2}a_{\boldsymbol{k}}^{\dag }a_{%
\boldsymbol{k}}\notag \\
p_{i}&=&\hbar\sum\limits_{\boldsymbol{k}}k_{i}a_{\boldsymbol{k}}^{\dag }a_{%
\boldsymbol{k}} \notag\end{aligned}$$ with the following vacuum state vector, $$\begin{aligned}
\left| k_{i}, 0\right\rangle &=&a_{\boldsymbol{k}}^{\dag }\left| 0,0\right\rangle \notag \\
\left|0,0 \right\rangle &=& \left|0 \right\rangle_{e}\left| 0\right\rangle_{ph}, \notag \end{aligned}$$ where $ \left|0 \right\rangle_{e} $ ($ \left| 0\right\rangle_{ph} $) is the vacuum state vector for electron (phonon), and $ a_{\boldsymbol{k}}^{\dag }(a_{\boldsymbol{k}}) $ is the electron creation (annihilation) operator. Using the vacuum state vectors, the expectation values of the transformed Hamiltonian ($ \widetilde{H}^{'}=\left\langle 0,0\right| \widetilde{H}\left| 0,0 \right\rangle $) can be written as $$\begin{aligned}
\widetilde{H}^{'}&=&\left[u_{0}+\sum\limits_{i=x,y}\bar{\eta}_{i}k_{i}^{2} +\sum\limits_{i=x,y} \bar{\eta}_{i} \left( \sum\limits_{\boldsymbol{q}}\hbar q_{i} \left|\textit{f}_{\boldsymbol{q}} \right|^{2} \right)^{2}+\sum\limits_{\boldsymbol{q} \mu} \left( \mathcal{M}_{\mu \boldsymbol{q}} \textit{f}_{\boldsymbol{q}}+\mathcal{M}_{\mu \boldsymbol{q}}^{*} \textit{f}_{\boldsymbol{q}}^{*}\right) \right]\sigma_{0} \notag \\
&+&\left[ \sum\limits_{\boldsymbol{q}}\left|\textit{f}_{\boldsymbol{q}} \right|^{2} \left( \hbar \omega_{\mu}-2\hbar\sum\limits_{i=x,y}\bar{\eta}_{i} k_{i}q_{i}+\hbar^{2}\sum\limits_{i=x,y}\bar{\eta}_{i}q_{i}^{2}\right) \right] \sigma_{0}+\left[\delta+\sum\limits_{i=x,y}\bar{\gamma}_{i}k_{i}^{2} +\sum\limits_{i=x,y} \bar{\gamma}_{i} \left( \sum\limits_{\boldsymbol{q}}\hbar q_{i} \left|\textit{f}_{\boldsymbol{q}} \right|^{2} \right)^{2} \right] \sigma_{1} \notag \\
&+& \left[ \sum\limits_{\boldsymbol{q}}\left|\textit{f}_{\boldsymbol{q}} \right|^{2} \left(-2\hbar\sum\limits_{i=x,y}\bar{\gamma}_{i} k_{i}q_{i}+\hbar^{2}\sum\limits_{i=x,y}\bar{\gamma}_{i}q_{i}^{2}\right) \right] \sigma_{1}-\bar{\chi}\left(k_{y}- \sum\limits_{\boldsymbol{q}}\hbar q_{y} \left|\textit{f}_{\boldsymbol{q}} \right|^{2}\right)\sigma_{2} \label{6}\end{aligned}$$
Eq.(\[6\]) can be minimized by employing the following conditions, $$\dfrac{\delta\widetilde{H}^{'}}{\delta \textit{f}_{\boldsymbol{q}}}=0, \quad \dfrac{\delta\widetilde{H}^{'}}{\delta \textit{f}_{\boldsymbol{q}}}^{*}=0. \notag$$ yielding the equation on $ \textit{f}_{\boldsymbol{q}} $,
$$\begin{aligned}
&&\left[ \mathcal{M}_{\mu \boldsymbol{q}}+\textit{f}_{\boldsymbol{q}}^{*} \left(\hbar \omega_{\mu}-2\sum\limits_{i=x,y}{\eta}_{i} k_{i}q_{i}+\sum\limits_{i=x,y}{\eta}_{i}q_{i}^{2}+2\sum\limits_{\boldsymbol{q^{'}}}{\eta}_{i} \left( q_{i}^{'} \left|\textit{f}_{\boldsymbol{q}^{'}} \right|^{2} \right) ^{2}\right) \right] \sigma_{0} \notag \\
&+&\left[\textit{f}_{\boldsymbol{q}}^{*} \left(-2\sum\limits_{i=x,y}{\gamma}_{i} k_{i}q_{i}+\sum\limits_{i=x,y}{\gamma}_{i}q_{i}^{2}+2\sum\limits_{\boldsymbol{q^{'}}}{\gamma}_{i} \left( q_{i}^{'} \left|\textit{f}_{\boldsymbol{q}^{'}} \right|^{2} \right) ^{2}\right) \right] \sigma_{1}+\left( \textit{f}_{\boldsymbol{q}}^{*} \chi q_{y}\right) \sigma_{2}=0. \label{7} \end{aligned}$$
The only preferred direction in this problem is chosen that of $ \boldsymbol{p} $ following the original work of LLP [@lee1953motion]. Thus, the new variation parameters can be introduced as, $$\alpha_{i}k_{i}=\sum\limits_{\boldsymbol{q}}q_{i} \left|\textit{f}_{\boldsymbol{q}} \right|^{2} \label{8}$$ Here, $ \alpha_{x} $ and $ \alpha_{y} $ are the new variation parameters along the $ x $ and $ y $ directions, respectively. $ \alpha_{i} $ is a fraction ($ \alpha_{i}<1 $) which is valid for weak and intermediate coupling regime. Note that in Ref., Eq.(\[8\]) was evaluated analytically by expanding the expression to order of $ ~\boldsymbol{p} $. It is also obvious that, $ \alpha_{i} $ can be regarded as variation parameters, which minimize the energy eigenvalues related with the coupling strength of interaction (for further discussion of the LLP method, see Ref.). The minimum of the quadratic expression in Eq.(\[7\]) can be found by using the variation parameters, $ \textit{f}_{\boldsymbol{q}}^{*} $ as, $$\textit{f}_{\boldsymbol{q}}^{*}=\dfrac{-\mathcal{M}_{\mu \boldsymbol{q}}}{\mathbb{I}_{1}\left(\boldsymbol{k},\boldsymbol{q},\boldsymbol{\alpha},\mu \right) +\left(\mathbb{I}_{2}^{2}\left(\boldsymbol{k},\boldsymbol{q},\boldsymbol{\alpha}\right)+\left(\chi q_{y} \right)^{2} \right)^{1/2} } \label{9}$$ where $$\begin{aligned}
\mathbb{I}_{1}\left(\boldsymbol{k},\boldsymbol{q},\boldsymbol{\alpha},\mu \right)&=&\hbar \omega_{\mu}-2\sum\limits_{i=x,y}\left(1-\alpha_{i} \right) {\eta}_{i} k_{i}q_{i}+\sum\limits_{i=x,y}{\eta}_{i}q_{i}^{2} \notag \\
\mathbb{I}_{2}\left(\boldsymbol{k},\boldsymbol{q},\boldsymbol{\alpha} \right)&=&-2\sum\limits_{i=x,y}\left(1-\alpha_{i} \right) {\gamma}_{i} k_{i}q_{i}+\sum\limits_{i=x,y}{\gamma}_{i}q_{i}^{2} \notag\end{aligned}$$ Using Eqs.(\[8\]) and (\[9\]), the eigenvalues of the Hamiltonian given by Eq.(\[6\]) can be found as, $$\begin{aligned}
E_{\lambda}&=&\left(u_{0}+\sum\limits_{i=x,y}\left( \alpha_{i}^{2}+1\right) \eta k_{i}^{2} \right) \left\langle \sigma_{0}\right\rangle+\sum\limits_{\boldsymbol{q} \mu} \left( \frac{-2\left|\mathcal{M}_{\mu \boldsymbol{q}} \right|^{2} }{\mathbb{I}_{1}\left(\boldsymbol{k},\boldsymbol{q},\boldsymbol{\alpha},\mu \right) +\left(\mathbb{I}_{2}^{2}\left(\boldsymbol{k},\boldsymbol{q},\boldsymbol{\alpha} \right)+\left(\chi q_{y} \right)^{2} \right)^{1/2} }\right) \left\langle \sigma_{0}\right\rangle \notag \\
&+& \sum\limits_{\boldsymbol{q} \mu}\left( \frac{\left|\mathcal{M}_{\mu \boldsymbol{q}} \right|^{2} \left( \mathbb{I}_{1}\left(\boldsymbol{k},\boldsymbol{q},\boldsymbol{\alpha},\mu \right)-2\sum\limits_{i=x,y}\alpha_{i}\eta_{i}k_{i}^{2}\right) }{\left( \mathbb{I}_{1}\left(\boldsymbol{k},\boldsymbol{q},\boldsymbol{\alpha},\mu \right) +\left(\mathbb{I}_{2}^{2}\left(\boldsymbol{k},\boldsymbol{q},\boldsymbol{\alpha} \right)+\left(\chi q_{y} \right)^{2} \right)^{1/2} \right)^{2} }\right) \left\langle \sigma_{0}\right\rangle +\left(\delta+\sum\limits_{i=x,y}\left( \alpha_{i}^{2}+1\right) \gamma k_{i}^{2} \right) \left\langle \sigma_{1}\right\rangle \notag \\
&+& \sum\limits_{\boldsymbol{q} \mu}\left( \frac{\left|\mathcal{M}_{\mu \boldsymbol{q}} \right|^{2} \left( \mathbb{I}_{2}\left(\boldsymbol{k},\boldsymbol{q},\boldsymbol{\alpha} \right)-2\sum\limits_{i=x,y}\alpha_{i}\gamma_{i}k_{i}^{2}\right) }{\left( \mathbb{I}_{1}\left(\boldsymbol{k},\boldsymbol{q},\boldsymbol{\alpha},\mu \right) +\left(\mathbb{I}_{2}^{2}\left(\boldsymbol{k},\boldsymbol{q},\boldsymbol{\alpha} \right)+\left(\chi q_{y} \right)^{2} \right)^{1/2} \right)^{2} }\right) \left\langle \sigma_{1}\right\rangle -\left(\chi k_{y} \right) \left\langle \sigma_{2}\right\rangle \notag \\
&+& \sum\limits_{\boldsymbol{q} \mu}\left( \frac{\left|\mathcal{M}_{\mu \boldsymbol{q}} \right|^{2} \left(\chi k_{y}\right) }{\left( \mathbb{I}_{1}\left(\boldsymbol{k},\boldsymbol{q},\boldsymbol{\alpha},\mu \right) +\left(\mathbb{I}_{2}^{2}\left(\boldsymbol{k},\boldsymbol{q},\boldsymbol{\alpha}\right)+\left(\chi q_{y} \right)^{2} \right)^{1/2} \right)^{2} }\right) \left\langle \sigma_{2}\right\rangle, \qquad \left\langle \sigma_{i} \right\rangle=\left\langle \Psi_{\lambda} \right| \sigma_{i} \left| \Psi_{\lambda}\right\rangle . \label{10}\end{aligned}$$
Here, $ \left| \Psi_{\lambda}\right\rangle $ are the eigenvectors of the unperturbed $ \boldsymbol{k}\cdot\boldsymbol{p} $ Hamiltonian \[Eq.(\[2\])\], which can be expressed as,
$$\begin{aligned}
\left| \Psi_{\lambda}\right\rangle &=&\frac{1}{\sqrt{2}}\left[
\begin{array}{c}
N_{\lambda} \\ \\
1
\end{array}%
\right], \qquad N_{\lambda}=\dfrac{E_{\lambda}^{0}-\left(u_{0}+\sum\limits_{i=x,y}\eta_{i}k_{i}^{2} \right) }{\left(\delta+\sum\limits_{i=x,y}\gamma_{i}k_{i}^{2} \right)}, \notag\end{aligned}$$
which corresponds to the following eigenvalues for electron ($ \lambda=1 $) and holes ($ \lambda=-1 $) $$E_{\lambda}^{0}=u_{0}+\sum\limits_{i=x,y}\eta_{i}k_{i}^{2}+\lambda\left[\left( \delta+\sum\limits_{i=x,y}\gamma_{i}k_{i}^{2}\right)^{2} +\left( \chi k_{y}\right)^{2} \right]^{1/2} \label{11}$$
It can be seen that, the energy expression of pristine MBP in Eq.(\[11\]) leads to a band gap, i.e., $ E_{g}=1.52 $ eV. Finally, the eigenvalues given by Eq.(\[10\]) can be represented as, $$\begin{aligned}
E_{\lambda}&=&E_{\lambda}^{0}+\sum\limits_{i=x,y}\left( \alpha_{i}^{2}\eta_{i}k_{i}^{2}+\alpha_{i}^{2}\gamma_{i}k_{i}^{2}\right)+\mathbf{\Sigma}_{1}\left(\boldsymbol{k},\boldsymbol{\alpha} \right) \notag \\
&+&\dfrac{\lambda}{\left[ \left(\delta+ \sum\limits_{i=x,y}\gamma_{i}k_{i}^{2}\right)^{2}+\left(\chi k_{y} \right)^{2} \right]^{1/2} }\left[\sum\limits_{i=x,y}\left(\delta+ \alpha_{i}^{2}\gamma_{i}k_{i}^{2}\right) \mathbf{\Sigma}_{2}\left(\boldsymbol{k},\boldsymbol{\alpha} \right)-\left( \chi k_{y}\right) \mathbf{\Sigma}_{3}\left(\boldsymbol{k},\boldsymbol{\alpha} \right) \right], \label{12} \end{aligned}$$ with $$\begin{aligned}
\mathbf{\Sigma}_{1}\left(\boldsymbol{k},\boldsymbol{\alpha} \right)&=&\sum\limits_{\boldsymbol{q} \mu} \left( \frac{-2\left|\mathcal{M}_{\mu \boldsymbol{q}} \right|^{2} }{\mathbb{I}_{1}\left(\boldsymbol{k},\boldsymbol{q},\boldsymbol{\alpha},\mu \right) +\left(\mathbb{I}_{2}^{2}\left(\boldsymbol{k},\boldsymbol{q},\boldsymbol{\alpha}\right)+\left(\chi q_{y} \right)^{2} \right)^{1/2} }\right) \notag \\
&+&\sum\limits_{\boldsymbol{q} \mu}\left( \frac{\left|\mathcal{M}_{\mu \boldsymbol{q}} \right|^{2} \left( \mathbb{I}_{1}\left(\boldsymbol{k},\boldsymbol{q},\boldsymbol{\alpha},\mu \right)-2\sum\limits_{i=x,y}\alpha_{i}\eta_{i}k_{i}^{2}\right) }{\left( \mathbb{I}_{1}\left(\boldsymbol{k},\boldsymbol{q},\boldsymbol{\alpha},\mu \right) +\left(\mathbb{I}_{2}^{2}\left(\boldsymbol{k},\boldsymbol{q},\boldsymbol{\alpha} \right)+\left(\chi q_{y} \right)^{2} \right)^{1/2} \right)^{2} }\right) \notag \\
\mathbf{\Sigma}_{2}\left(\boldsymbol{k},\boldsymbol{\alpha} \right)&=& \sum\limits_{\boldsymbol{q} \mu}\left( \frac{\left|\mathcal{M}_{\mu \boldsymbol{q}} \right|^{2} \left( \mathbb{I}_{2}\left(\boldsymbol{k},\boldsymbol{q},\boldsymbol{\alpha}\right)-2\sum\limits_{i=x,y}\alpha_{i}\gamma_{i}k_{i}^{2}\right) }{\left( \mathbb{I}_{1}\left(\boldsymbol{k},\boldsymbol{q},\boldsymbol{\alpha},\mu \right) +\left(\mathbb{I}_{2}^{2}\left(\boldsymbol{k},\boldsymbol{q},\boldsymbol{\alpha}\right)+\left(\chi q_{y} \right)^{2} \right)^{1/2} \right)^{2} }\right) \notag \\
\mathbf{\Sigma}_{3}\left(\boldsymbol{k},\boldsymbol{\alpha} \right)&=&\sum\limits_{\boldsymbol{q} \mu}\left( \frac{\left|\mathcal{M}_{\mu \boldsymbol{q}} \right|^{2} \left(\chi k_{y}\right) }{\left( \mathbb{I}_{1}\left(\boldsymbol{k},\boldsymbol{q},\boldsymbol{\alpha},\mu \right) +\left(\mathbb{I}_{2}^{2}\left(\boldsymbol{k},\boldsymbol{q},\boldsymbol{\alpha} \right)+\left(\chi q_{y} \right)^{2} \right)^{1/2} \right)^{2} }\right). \label{13}\end{aligned}$$
Summations in Eq.(\[13\]) can be replaced with integral over $ \boldsymbol{q} $, i.e., $ \sum_{\boldsymbol{q}}\rightarrow \left( A/4\pi^{2}\right) \int \int dq_{x} dq_{y}$, where $ A $ is the surface area of MBP. While taking the integrals we introduce a cut-off phonon wavevector, $ q_{cut} $ . The eigenvalues are calculated by numeric minimization of Eq.(\[12\]) with respect to the two different variation parameters, $ \alpha_{x} $ and $ \alpha_{y} $, which is necessary to take into account the anisotropic nature of MBP along the $x $ and $y$ direction, respectively.
Results and Discussion
======================
Due to the polaronic effect caused by a polar substrate, the energy band gap of MBP is increased depending on the interlayer distance, dielectric constants, and characteristic phonon frequencies of a substrate. Indeed, from Eq.(\[12\]) one can see that, while the second and third terms cause a uniform shift of the valance and conduction bands and do not produce any gap change, the fourth term changes the positions of the bands for $\lambda=-1$ and $\lambda=+1$ individually thus inducing a gap broadening ($ \Delta E_{g}$). Although electron-phonon coupling in MBP due to the surface optical phonon modes of a substrate can be regarded as weak, the preferred direction choice made in Eq.(\[8\]) allows us to expect more accurate eigenvalues of the total Hamiltonian \[Eq.(\[4\])\] by taking into account correlations introduced by electron recoil [@lee1953motion].
We first analyze general dependence of the energy gap and effective masses in MBP on the characteristic substrate frequency, $\omega $ and on its effective screening parameter, $\xi $. For simplicity, we take $ \omega=\omega_{1}=\omega_{2} $. The corresponding dependencies are shown in Fig.(\[FIGURE2\]). As can be seen from Fig.(\[FIGURE2\]) (a), the gap increases with increasing the values of both $ \hbar \omega $ and $\xi $. From the Figs.(\[FIGURE2\]) (b) and (c), it can be also seen that for higher $ \hbar \omega $, effective masses are decreasing along the $ x $ direction, but increasing along the $ y $ direction. The qualitative difference of the phonon frequency dependence of the masses along the $ x $ and $ y $ directions arises from the linear $ \chi k_{y}$ term which appears in Eq.(\[11\]). Note that, the anisotropy of MBP not only originates from the coefficients in $ k_{x}^{2} $ and $ k_{y}^{2} $ terms, i.e., $ \eta $ and $ \gamma $, but also from the linear $ k_{y} $ term in the Hamiltonian Eq.(\[2\]). In other words, charge carriers of MBP show both Schr[ö]{}dinger-like and Dirac-like character in addition to the anisotropy of the MBP spectrum [@1504.02452v1]. Therefore, when the electron-phonon interaction is switched on, the dynamics of charge carriers also changes significantly along both directions in addition to the existing anisotropy of MPB.
In practical calculations, surface optical phonon modes of polar substrate emerge from its finite size and they can be extracted from its bulk transverse optical phonon modes ($\omega_{TO_{\mu}}$) using the formula, $ \hbar \omega_{\mu}=\hbar \omega_{TO_{\mu}} F_{\mu} $, $ F_{1}=\left[\left(\kappa_{0}+1 \right) /\left(\kappa_{int}+1 \right) \right]^{1/2} $ and $ F_{2}=\left[\left(\kappa_{int}+1 \right) /\left(\kappa_{\infty}+1 \right) \right]^{1/2} $ [@Wang:1972aa; @fischetti2001effective; @Tan2013aa]. Here, $ \kappa_{int} $ is the intermediate frequency dielectric constant of the substrate. Surface optical phonon modes and dielectric constants of polar substrates are given in Table \[tab1\], which have been taken from Refs. . Among these substrates, h-BN has the highest surface optical phonon frequencies, whereas Al$ _{2} $O$ _{3} $ has the highest effective screening constant. In terms of practical applications, polar substrates might give rise to considerable modifications of the electronic band structure of supported samples, which are expected to be especially important for scattering processes. To examine the effect of specific substrates on $ \Delta E_{g}$ in MBP, we have considered a series of typical polar substrates: h-BN, SiC, SiO$_{2}$, HfO$_{2}$, Al$_{2}$O$_{3}$. One can see from Fig.(\[FIGURE3\])(b), $ \Delta E_{g}$ decreases with $ z $ and converged to the band gap of pristine MBP after some values of $ z $. It is also clear from the Fig.(\[FIGURE3\]) (b) that, while the influence of the effective screening constant, $ \xi $, is more pronounced for higher $ z $, the phonon frequencies become more effective at lower $ z $. Since $ \xi $ values of SiO$ _{2} $ is lower than HfO$ _{2}$, $ \Delta E_{g}$ of SiO$ _{2} $ higher than HfO$ _{2} $ below $ z\approx5 $ Å, which is related to the relatively higher phonon frequencies of SiO$ _{2} $. $ \Delta E_{g}$ is also dependent on the cut-off wavevector $ q_{cut}$, which converges at around $ 0.5 \sim 0.6 $ Å$ ^{-1} $. The effect of $ q_{cut} $ on $ \Delta E_{g}$ becomes insignificant when $ z $ increases, and means that only the phonons with low $ q $, are strongly coupled with the charge carriers in MBP. We considered $ q_{cut}=0.5 $ Å$ ^{-1} $ for all numerical calculations presented in this paper.
\[h\]
[l\*[6]{}[c]{}]{} Substrate &$ \hbar \omega_{1} $ (meV) & $ \hbar \omega_{2} $ (meV)& $ \kappa_{0} $ &$ \kappa_{\infty} $ & $ \xi$ &\
h-BN & 101 & 195 & 5.1 & 4.1 & 0.03 &\
SiC & 167 & 116 & 9.7 & 6.5 & 0.04 &\
SiO$ _{2} $ & 146 & 60 & 3.9 & 2.5 & 0.08 &\
HfO$ _{2} $ & 53 & 19 & 22.0 & 5.0 & 0.12 &\
Al$ _{2} $O$ _{3} $ & 94 & 55 & 12.5 & 3.2 & 0.16 &\
\[h\] ![Illustration of a MBP/substrate system. $ z $ is the vertical distance between MBP and a polar substrate.[]{data-label="FIGURE1"}](ILS.jpg "fig:"){width="45.00000%"}
\[h!\] ![(a) Gap broadening ($ \Delta E_{g}$) and (b),(c) renormalization of direction-dependent effective masses in MBP, as a function of phonon frequency, $ \hbar \omega $, for different effective screening constants, $ \xi$. Each case corresponds to a fixed MBP-substrate separation, $z=2 $ Å.[]{data-label="FIGURE2"}](FIG2.eps "fig:"){width="95.00000%"}
\[h!\]
![Evolution of a band gap in MBP with the cutoff wavevector $q_{cut} $ (a) and with distance to the substrate (b) shown for $ z=2 $ Å. Different curves correspond to different substrates. Dashed line corresponds to $q_{cut} $ used in this paper for other calculations.[]{data-label="FIGURE3"}](FIG3.eps){width="70.00000%"}
As a next step, we analyze the effective masses. The effective masses can be calculated as, $ m_{\lambda,i}^{*}=\hbar^{2}/\left(\partial ^{2}E_{\lambda}/\partial k_{i}^{2} \right) $, where $ i=x,y $ and $ \lambda=1 $($ -1 $) corresponds to electron (hole) states. In pristine MBP, charge carriers have different effective masses due to the anisotropy of the system along the $ x $ and $ y $ direction. While the $ x $ component of effective masses is, $ m_{\lambda,x}^{*}=\hbar^2/2\left(\eta_{x}+\lambda\gamma_{x} \right) $ ($ m_{e,x}^{*}=0.846 $ $ m_{0} $ and $ m_{h,x}^{*}=1.14 $$ m_{0}) $, the $ y $ component can be found by expanding the expression around $ k_{y}=0 $ point as, $ m_{\lambda,2}^{*}=\hbar^2/2\left(\eta_{x}+\lambda\gamma_{x}+\lambda \left( \chi^{2}/2\delta \right) \right) $ ($ m_{e,y}^{*}=0.166 $ $ m_{0} $ and $ m_{h,y}^{*}=0.182 $ $ m_{0} $, where $ m_{0} $ is the free electron mass). The dressed effective masses can be found by following the same procedure. In this case, not only $ y $ components but also $ x $ components of the effective masses can be obtained by expanding around $ k_{x}=0 $ point. The many-body correction to effective masses ($ \Delta m_{\lambda,i}^{*} $=$ \bar{m}_{\lambda,i}^{*}-m_{\lambda,i}^{*} $) of the charge carriers in MBP is presented in Fig.(\[FIGURE4\]) for different $ z $ values, where $ \bar{m}_{\lambda,i}^{*} $ are the dressed effective masses under the influence of electron-phonon interaction. The effective masses of the charge carriers of MBP increase by the influence of the polaronic effect. In Fig.(\[FIGURE4\])(a),and (b), the surface polaronic effect on the electron and hole effective masses along the $ x $ direction is presented. One can see that along the $ x$-direction hole-phonon has stronger coupling than the electron-phonon one. Since the effective masses along the $ x $ direction decrease with $ \hbar \omega $ as it shows in Fig.\[FIGURE2\](b), HfO$ _{2} $-induced effective mass enhancement is higher compared to other substrates at high $ z $ values. However, the effect of $ \xi $ becomes more dominant at low $ z $. In the vicinity of $ z=2 $ Å and $ z=6 $ Å, it is clear from Fig.(\[FIGURE4\])(a),and (b) Al$ _{2} $O$ _{3} $ has the largest effect for both electrons and holes due to the higher effective screening constant. In Fig.(\[FIGURE4\])(c), and (d), we show the renormalization of the electron and hole masses along the $ y $ direction which is substantially smaller compared to the $ x $ components. This can be related with a weaker electron-phonon coupling along the $ y $ direction. In contrast to the effective masses along the $ x $ direction, the $ y $ component of effective masses increases with $ \hbar \omega $ as it is seen from Fig.\[FIGURE2\](c). Therefore, the $\Delta m_{\lambda,y}^{*} $ show similar behavior to $ \Delta E_{g}$ given in Fig.(\[FIGURE3\]). That is, the effective screening constant, $ \xi $ basically determines the behavior of effective masses along the $ y $ direction at all distances, whereas the effects of phonon frequencies on effective masses become more pronounced at low $ z $.
\[h!\] ![Renormalization of effective masses in MBP due to the polaronic effect induced by different substrates shown as a function of the MBP-substrate separation. The values are given in units of free electron mass, $ m_{0}$.[]{data-label="FIGURE4"}](FIG4.eps "fig:"){width="\textwidth"}
Conclusion
==========
In summary, we have shown that electron-phonon interaction in MBP on polar substrates gives rise to a polaronic effect, which is an efficient many-body mechanism affecting the band properties of the system. We have used the LLP method to capture the effects of electron-phonon coupling in the energy spectrum of MBP within the long-wavelength approximation. In contrast to standard perturbation theory which is valid only in the weak coupling regime, this method is also applicable to the intermediate interactions. In comparison to the original formulation of the LLP method, we have considered two different variation parameters, i.e., $ \alpha_{x} $ and $ \alpha_{y} $, corresponding to the $ x $ and $ y $ direction, respectively, which is governed by the anisotropy of effective masses in MBP. Following this method, we have calculated the band gap $ \Delta E_{g}$ and effective mass variations in MBP arising from the coupling of carriers with surface optical phonon modes of different substrates. We have found that, the highest contribution to $ \Delta E_{g}$ comes from the Al$ _{2} $O$ _{3}$ substrate. We have also shown that, while the effective screening constant is the main parameter that determines the gap variations at all interlayer separations, phonon energies of a substrate are important only at low $ z $. At $ z=2.5$ Å, which is a typical interlayer distance between graphene and SiO$_{2}$ [@rudenko2011interfacial], the largest $ \Delta E_{g}$ is the order of $ 50$ meV. Our analysis of the effective mass renormalization shows that the interaction of charge-carrier coupling with phonons is significantly stronger along the $ x $ direction than the $ y $ direction. Due to its large effective screening constant, Al$ _{2} $O$ _{3}$-induced mass enhancements are higher with respect to the other substrates at low $ z $. For higher $ z $, one should also involve the energies of surface optical phonons to understand the behavior of effective mass renormalization. We have also shown that the dependence of effective masses on $ \hbar \omega$ is qualitatively different for armchair and zigzag directions. This is due to the highly anisotropic nature of the continuum Hamiltonian of MBP. Moreover, it should be noted that the carrier-surface optical phonon coupling increases the effective mass anisotropy of MBP. Our results provide a starting point for the experimental verification of the surface polaronic effect in supported or encapsulated MBP. We also believe that our results will lead to deeper understanding of the charge-carrier dynamics in MBP from both theoretical and experimental perspectives.
A.M. and Y.M. acknowledge “The Scientific and Technological Research Council of Turkey (TUBITAK)” through “BIDEB-2219 Postdoctoral Research Fellowship.”. Funding from the European Union Seventh Framework Programme under Grant Agreement No. 604391 Graphene Flagship is also acknowledged.
References {#references .unnumbered}
==========
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study analytically the non-Markovianity of a spin ensemble, with arbitrary number of spins and spin quantum number, undergoing a pure dephasing dynamics. The system is considered as a part of a larger spin ensemble of any geometry with pairwise interactions. We derive exact formulas for the reduced dynamics of the system and for its non-Markovianity as assessed by the witness of Lorenzo *et al.* \[Phys. Rev. A **88**, 020102(R) (2013)\]. The non-Markovianity is further investigated in the thermodynamic limit when the environment’s size goes to infinity. In this limit and for finite-size systems, we find that the Markovian’s character of the system’s dynamics crucially depends on the range of the interactions. We also show that, when the system and its environment are initially in a product state, the appearance of non-Markovianity is independent of the entanglement generation between the system and its environment.'
author:
- Rémy Dubertrand
- Alexandre Cesa
- John Martin
title: 'Analytical results for the quantum non-Markovianity of spin ensembles undergoing pure dephasing dynamics'
---
Introduction
============
Open quantum systems can display a large variety of dynamical behaviors, including decoherence [@joos_decoherence_2003; @zurek_decoherence_1991; @haroche_entanglement_1998], thermalization and memory effects. The notion of non-Markovianity, accounting for memory effects, has found applications in many different fields ranging from quantum optics [@de_vega_matter-wave_2008], quantum thermodynamics [@bylicka_thermodynamic_2016; @pezzutto_implications_2016], quantum information theory [@bylicka_non-markovianity_2015; @bellomo_non-markovian_2007] to quantum foundations [@matsuzaki_magnetic_2011; @chin_quantum_2012; @dhar_characterizing_2015; @glick_markovian_2017]. Non-Markovianity has also been identified as a key ingredient to achieve specific tasks in the context of quantum heat machines and quantum information processing [@gonzalez-tudela_dissipative_2010; @malekakhlagh_non-markovian_2016; @fang_non-markovian_2017]. While Markovian dynamics for discrete variable systems is always governed by a master equation of the Gorini-Kossakowski-Sudarshan-Lindblad (GKSL) type, the methods for treating non-Markovian quantum dynamics and their physical interpretation are generally much more complicated, see e.g. [@rivas_quantum_2014; @breuer_colloquium_2016; @de_vega_dynamics_2017; @shibata_generalized_1977; @wilkie_positivity_2000; @Laine2012]. A direct consequence is that non-Markovian master equations are only rarely analytically solvable [@Briegel; @Maniscalco]. The departure from Markovian dynamics can be quantified through measures of non-Markovianity (see Sec. \[definitions\]). Even when the dynamics of the system and its environment is known, evaluating analytically measures of non-Markovianity is often a difficult task, so that up to now only a limited number of analytical results have been obtained [@She17; @bhattacharya_exact_2017; @fischer_correlated_2007]. The aim of this work is to contribute to the analytical treatment of non-Markovianity in the case of spin ensembles undergoing pure dephasing dynamics, with a particular emphasis on the thermodynamic limit of infinitely many spins in the system and/or the environment. Note that non-Markovianity in spin chains has already been studied analytically in [@prosen_exact_2010; @mahmoudi_non-markovian_2016] and numerically in [@buca_note_2012; @cormick_dissipative_2013; @smirne_interaction-induced_2013; @ribeiro_non-markovian_2015].
The paper is organized as follows. In Section \[definitions\], we present three different measures of non-Markovianity and introduce our main model of a spin ensemble with arbitrary pairwise interaction range and longitudinal external field. In Section \[results\], the reduced dynamics and a non-Markovianity witness are evaluated analytically for such an ensemble. The cases of nearest-neighbor and infinite range interactions are discussed in detail, in particular in the limit of an infinite number of spins. A comparison with other measures of non-Markovianity is presented. We also discuss how non-Markovianity is independent of the generation of entanglement between the system and its environment. In Section \[conclusion\], we summarize our results and formulate some perspectives. Some more technical material is presented in the Appendix.
Definitions and system {#definitions}
======================
Measures of non-Markovianity
----------------------------
Different measures of non-Markovianity have been proposed in the literature, relying on different notions of non-Markovianity. Although these notions are not equivalent, they coincide in many instances [@rivas_quantum_2014; @breuer_colloquium_2016; @Zheng2011]. In all cases, non-Markovianity appears as a property of the dynamics, i.e. it does not depend on a particular choice of the initial state(s). The Rivas-Huelga-Plenio (RHP) measure is based on the divisibility of the dynamical map for the reduced system [@rivas_entanglement_2010], the Breuer-Laine-Piilo (BLP) measure is devised from information-theoretical considerations [@laine_measure_2010] and the measure introduced in [@lorenzo_geometrical_2013] relies on a geometrical characterization of the dynamics. The RHP measure quantifies the divisibility of the super-operator describing the time evolution of the reduced density matrix. It can be reformulated as a positivity constraint on the rates of the dynamical equation for the density matrix, when this equation can be cast into GKSL form [@rivas_quantum_2014], see Sec. \[compmeas\]. The BLP measure is probably the most intuitive: it consists of tracking the time evolution of the trace distance between two initially distinct states of the system. When the trace distance is growing, that may be interpreted as back-flow of information to the system [@laine_measure_2010], hence a signature of non-Markovianity (see Sec. \[compmeas\]), despite some recent qualification of this interpretation [@wudarski_petruccione_2016; @budini_2018]. A practical limitation of this measure is that it requires an optimization over the two initial states, which becomes prohibitive when studying large systems.
The measure of non-Markovianity introduced in [@lorenzo_geometrical_2013] relies on the parametrisation of the system’s density matrix by a Bloch vector, see e.g. [@mahler_quantum_1998]. The time evolution is then described by a matrix. The derivative of the determinant of this matrix tells us whether the norm of the Bloch vector is expanding or contracting. Any expansion, i.e. when the derivative of the determinant is positive, is defined as a non-Markovian episode in the time evolution. This corresponds to an increase with time of the volume of accessible states. In contrast, for a Markovian dynamics, the volume of accessible states can only decrease with time. This measure is especially well suited for analytical results and will be mainly considered in this work. It will be compared to the two previously introduced measures only in the simplest cases.
Spin ensemble with pairwise interaction and local longitudinal field
--------------------------------------------------------------------
We are interested in estimating how the time dynamics of a subset of a system of spins can show non-Markovian features. As our formalism allows us to address a quite general problem, we will first express it in a most general framework. Then our results will be applied to the particular case of a spin-$1/2$ chain. From now on, we set $\hbar=1$. We consider a set of $N$ spins with spin quantum number $S$ interacting with each other only through pairwise interaction. Moreover each spin is subject to a local longitudinal field. The Hamiltonian describing such a spin ensemble reads $$\label{Hgeneral}
H={-}\sum_{i=1}^N\sum_{j=1}^N \Cij S_i^z S_{j}^z +\sum_{i=1}^N \Gi S_i^z\ ,$$ where $S_i^z$ stands for the spin operator in the $z$ direction associated to spin $i$ ($i=1,\ldots,N$), and $\Gi$ is the magnitude of the external field applied on spin $i$. The pairwise correlation matrix $(\Cij)$ is only assumed to be real symmetric and accounts for the geometrical arrangement of the $N$ spins and the range of interaction. Note that at this stage, we do not impose any specific geometry nor boundary conditions. For the sake of simplicity, we restrict ourselves to an external longitudinal field, i.e. in the same direction as the interaction, which allows a fully analytical description of the dynamics. The whole set of spins is divided into a subset $\mathcal{S}$ of $p$ spins (labeled hereafter $i=1,\dots, p$ without loss of generality), which defines our system of interest, and the remaining $N-p$ spins ($i=p+1,\dots, N$), which form the environment $\mathcal{E}$. The global system $\mathcal{S}+\mathcal{E}$ is assumed to be isolated, so that it evolves unitarily. If $\rho_{\mathcal{S}+\mathcal{E}}$ denotes its density matrix, it obeys Liouville equation $${\mathrm{i}}\frac{{\mathrm{d}}}{{\mathrm{d}}t}\rho_{\mathcal{S}+\mathcal{E}}=\left[H, \rho_{\mathcal{S}+\mathcal{E}}\right]\ .$$ The global Hamiltonian (\[Hgeneral\]) can be written $$H=H_\mathcal{S}+H_\mathcal{E}+H_\mathcal{SE}\ , \label{Hsum}$$ with $$\begin{aligned}
H_{\cal S}&={-}\sum_{i=1}^p\sum_{j=1}^p \Cij S_i^z S_j^z +\sum_{i=1}^p \Gi S_i^z\ , \label{HqS} \\
H_{\cal E}&= {-}\sum_{i=p+1}^N\sum_{j=p+1}^N \Cij S_i^z S_j^z+\sum_{i=p+1}^N \Gi S_i^z\ ,\label{HqE}\\
H_{\cal SE}&={-}2\sum_{i=1}^p\sum_{j=p+1}^N \Cij S_i^zS_j^z,\label{HqSE}\end{aligned}$$ where $H_{\cal S}$ is the Hamiltonian of the system $\mathcal{S}$ of interest, $H_{\cal E}$ is the Hamiltonian of its environment, and $H_{\cal SE}$ is the interaction Hamiltonian between the system and the environment. The computational basis states are defined as the common eigenstates of all $S_i^z$ operators ($i=1,\ldots,N$). For convenience, we write these states as $$\left| \bm{s} \bm{\sigma}\right> \equiv \left| \bm{s}\rangle\otimes |\bm{\sigma}\right>=\left| s_1 s_2 \dots s_p\right>\otimes \left| \sigma_{p+1}\sigma_{p+2}\dots \sigma_N\right>\ , \label{comp_basis}$$ where $|s_k\rangle$ (resp. $|\sigma_k\rangle$) are the eigenstates of $S_k^z$ for $k=1,\ldots,p$ (resp. $k=p+1,\ldots,N$) of eigenvalue $s_k\,(\sigma_k) \in\{-S,-S+1,\ldots,S\} $. In particular we use different notation to emphasize the distinction between the system and its environment. Note that all three Hamiltonians (\[HqS\]), (\[HqE\]) and (\[HqSE\]) are diagonal in the basis (\[comp\_basis\]), and thus pairwise commute.
Non-Markovianity in a spin ensemble with pairwise interaction {#results}
=============================================================
Derivation of the main result {#mainderiv}
-----------------------------
In this Section, we calculate the reduced density matrix of the system $\mathcal{S}$ at any time $t$ and deduce from it the witness of non-Markovianity following [@lorenzo_geometrical_2013]. The time evolution operator of the global system associated to (\[Hgeneral\]), $U(t)=e^{-{\mathrm{i}}(H_\mathcal{S}+H_\mathcal{E}+H_\mathcal{SE}) t}$, acts on the computational basis states as $$U(t)\left| \bm{s} \bm{\sigma}\right> = e^{-{\mathrm{i}}[H_\mathcal{S}(\bm{s})+H_\mathcal{SE}(\bm{s},\bm{\sigma})+H_\mathcal{E}(\bm{\sigma})] t}\left| \bm{s} \bm{\sigma}\right>\ ,
\label{diag}$$ where $$\begin{aligned}
H_{\cal S}(\bm{s})&={-}\sum_{i=1}^p\sum_{j=1}^p \Cij s_i s_j +\sum_{i=1}^p \Gi s_i\ , \label{HclasS} \\
H_{\cal E}(\bm{\sigma})&={-}\sum_{i=p+1}^N\sum_{j=p+1}^N \Cij \sigma_i \sigma_j+\sum_{i=p+1}^N \Gi \sigma_i\ ,\label{HclasE}\\
H_{\cal SE}(\bm{s},\bm{\sigma})&= {-}2\sum_{i=1}^p\sum_{j=p+1}^N \Cij s_i\sigma_j\label{HclasSE}\end{aligned}$$ are the corresponding scalar Hamiltonians introduced in correspondence to (\[HqS\])-(\[HqSE\]) and contain all the physical description of the dynamics.
We consider a density matrix of the global system that is initially a product state with respect to the bi-partition $\mathcal{S}+\mathcal{E}$, $$\rho_{\mathcal{S}+\mathcal{E}}(0)=\rho_\mathcal{S}(0)\otimes\rho_\mathcal{E}(0)\label{rho_init_sep}\ ,$$ In particular, if the initial state of the whole chain is separable it may *not* stay so during the dynamics. It will stay separable only for some prescribed choices of the initial density matrix of both the system and its environment. This important point about possible creation of entanglement during the time evolution, already present within our simple model, will be discussed in more details in Sect. \[correlations\] below. The reduced density matrix of $\mathcal{S}$ at any time $t$ is given by $$\label{defrhoS}
\rho_\mathcal{S}(t)= {\mathrm{ tr }\;}_\mathcal{E} (\rho_{\mathcal{S}+\mathcal{E}}(t)),\, \mathrm{with}\, \rho_{\mathcal{S}+\mathcal{E}}(t)=e^{-{\mathrm{i}}H t} \rho_\mathcal{S+E}(0) e^{{\mathrm{i}}H t} ,$$ where ${\mathrm{ tr }\;}_\mathcal{E}$ denotes a partial trace over the environment degrees of freedom. This expression can be explicited as $$\begin{aligned}
\rho_\mathcal{S}(t)=&\sum_{\bm{\sigma}} \left< \bm{\sigma}\right| \rho_{\mathcal{S}+\mathcal{E}}(t) \left| \bm{\sigma}\right>\\
= & \sum_{\sigma_{p+1}=-S}^S \sum_{\sigma_{p+2}=-S}^S \dots \sum_{\sigma_{N}=-S}^S \left< \bm{\sigma}\right|\rho_{\mathcal{S}+\mathcal{E}}(t) \left| \bm{\sigma}\right> .
\end{aligned}
\label{deftrE}$$ Expanding the initial state of the environment in the computational basis as $$\rho_\mathcal{E}(0)=\sum_{ \bm{\sigma}', \bm{\sigma}''} a_{ \bm{\sigma}', \bm{\sigma }''} \left| \bm{\sigma}'\right>\left< \bm{\sigma}''\right|,\label{rhoE_gen}$$ the evolved reduced density matrix follows from (\[diag\]), (\[defrhoS\]) and (\[deftrE\]) $$\left< \bm{s}\right| \rho_\mathcal{S}(t)\left| \bm{s'}\right>={} e^{{\mathrm{i}}t [H_\mathcal{S}(\bm{s'})-H_\mathcal{S}(\bm{s})]} \left< \bm{s}\right| \rho_\mathcal{S}(0)\left| \bm{s'}\right>\A \label{rhoS_t_exact}$$ with $$\A = \sum_{\bm{\sigma}} a_{ \bm{\sigma}, \bm{\sigma}} \,e^{{\mathrm{i}}t [H_\mathcal{SE}(\bm{s'},\bm{\sigma})-H_\mathcal{SE}(\bm{s},\bm{\sigma})]}.\label{defA}$$ In Eq. (\[defA\]), the sum runs over the diagonal elements of the expansion (\[rhoE\_gen\]), which comes from the fact that the Hamiltonian of the environment is diagonal in the computational basis. Therefore the reduced density matrix of the system $\mathcal{S}$ only depends on the initial populations of the environment in the computational basis. Equations (\[rhoS\_t\_exact\]) and (\[defA\]) show that the populations of the system are conserved during the dynamics as, for $ \bm{s}= \bm{s'}$, we have $A_{\bm{s},\bm{s}}(t)=1$ for all $t$. This means that the dynamics of the system is purely dephasing. Using the definition (\[HclasSE\]) of the interaction Hamiltonian, Eq. (\[defA\]) becomes $$\A = \sum_{\bm{\sigma}} a_{ \bm{\sigma}, \bm{\sigma}} \exp\left[2{\mathrm{i}}t \left(
\displaystyle\sum_{j=p+1}^N \sigma_j\sum_{i=1}^p \Cij (s_i-s'_i) \right) \right].\label{A}$$ The next step consists of writing the Bloch vector parametrising the density matrix (\[rhoS\_t\_exact\]) in order to compute the determinant of the time evolution operator for the reduced density matrix. This operator is represented by a matrix $M_{\mathcal{S}}(t)$ acting on the Bloch vector, and its determinant is the volume of accessible states. Its exact expression and the calculation of its determinant is a bit lengthy and can be found in Appendix \[Appendix\_A\]. One eventually gets the closed formula $$\label{general_detM}
\det M_{\mathcal{S}}(t)= \prod_{\bm{s},\bm{s'}} \A\ ,$$ where the product over $\bm{s}$ is meant to browse all the eigenstates of the Hamiltonian (\[HclasS\]), i.e. all the $(2S+1)^p$ values of the coordinates of $\bm{s}$ with $s_i=-S,\ldots, S$, and the same for $\bm{s'}$. We find that there is no dependence on the external field. Equation (\[general\_detM\]) is one of the main results of our paper. Following [@lorenzo_geometrical_2013], the dynamics of $\mathcal{S}$ defined by (\[defrhoS\]) will be non-Markovian whenever $$\frac{{\mathrm{d}}}{{\mathrm{d}}t}\det M_{\mathcal{S}}(t)> 0.
\label{NMwitness}$$ This result leads to several remarks. First, Eqs. (\[A\]) and (\[general\_detM\]) show that the couplings between any two spins within the system $\mathcal{S}$ (or the environment $\mathcal{E}$) do not influence the non-Markovianity of $\mathcal{S}$. Instead, non-Markovianity is a feature that only stems from the couplings between $\mathcal{S}$ and $\mathcal{E}$. Second, when the environment is in a computational basis state $\rho_{\mathcal{E}}=|\boldsymbol{\sigma}'\rangle\langle \boldsymbol{\sigma}'|$, the determinant simplifies to $\det M_{\mathcal{S}}(t)=1$ for all times, and the dynamics is Markovian. Last, let us emphasize that the result (\[general\_detM\]) is very general as it is valid for any pairwise interaction strengths $\Cij$ and in particular, for random interactions or for spin glasses [@sherrington_solvable_1975].
Application to spin-$1/2$ chains
--------------------------------
Let us exemplify Eq. (\[general\_detM\]) in the case of $N$ spin-$1/2$. For the sake of simplicity, we consider the environment initially in the maximally mixed state $$\rho_\mathcal{E}(0)=\frac{\mathbb{1}_\mathcal{E}}{2^{N-p}}.
\label{rhoE_mixed}$$ Inserting (\[rhoE\_mixed\]) into (\[A\]), and performing the sum over the environment states by descending recursion, we obtain $$\A =\displaystyle\prod_{j=p+1}^N
\cos\left[\left(\sum_{i=1}^p \Cij(s_i-s'_i) \right)t \right]\label{general_rhoS_t_exact}$$ with $s_i,s'_i\in\{-1/2,1/2\}$. We will use this result to determine when one-dimensional spin chains with periodic boundary conditions display non-Markovianity. We are more particularly interested in studying how the range of the interaction can affect the Markovian character of the dynamics of the system $\mathcal{S}$. We will start by investigating the most common case of nearest neighbor interaction. Then we will study the formal case of infinite range where all the spins of the chain interact with each other. Last, we consider a model with power law range, which interpolates between those two situations.
### Ising model with nearest neighbor interaction
We consider now a spin chain where each spin interacts only with its two nearest neighbors (nn). When comparing with the general form (\[Hgeneral\]), this amounts to take $\Cij=0$ for $i=j$ and $(|i-j|~\mathrm{mod}~N)>2$, and $\Cij=J$ ($J>0$) for $(|i-j|~\mathrm{mod}~N)=1$. In this case, Eq. (\[general\_rhoS\_t\_exact\]) yields $$\A=\cos\left[J t\left(s_p-s'_p \right) \right]
\cos\left[J t\left( s_1-s'_1 \right) \right]\ ,\label{QIshrlongh_rhoS_t_exact}$$ where it was assumed that the environment contains more than one spin ($N>p+1$). This explicit expression allows us to evaluate the determinant of the time evolution operator $M_{\mathcal{S}}(t)$ of the Bloch vector given by Eq. (\[general\_detM\]), $$\det M_{\mathcal{S},\mathrm{nn}}(t) =\cos^{2^{2p}}\left( Jt\right),\, N-p\ge 1. \label{detM_QIshrlongh}$$ This result indicates that the dynamics of the system is always non-Markovian following the criterion (\[NMwitness\]), as the derivative of this expression always reaches positive values. Interestingly Eq. (\[detM\_QIshrlongh\]) does depend neither on the sign of the interaction, nor on the size of the bath. Therefore the system remains non-Markovian in the thermodynamic limit of infinitely large environment ($N\to\infty$). Another choice of the thermodynamic limit can be taken by choosing a system size which is a finite fraction of the whole chain: $p=rN$. From (\[detM\_QIshrlongh\]) it can be immediately seen that the determinant is zero almost everywhere [^1] so that the dynamics becomes Markovian in this limit.
### Infinite range Ising model
In this Section, all spins are assumed to be coupled with each other with the same interaction strength, i.e. $\Cij=J/N$ ($J>0$) for $i\ne j$ and zero otherwise. In particular, we recover for $p=1$ the case of a single spin coupled uniformly to an environment of spins: this is the celebrated central spin model, which has been extensively studied before, see e.g. [@coish_hyperfine_2004; @fischer_correlated_2007; @bhattacharya_exact_2017]. Note that the Hamiltonian (\[HqS\]) of the system $\mathcal{S}$ depends on the size of the environment through the interaction constant $J_{ij}=J/N$. This convention is particularly relevant in order to consider the thermodynamic limit as in this case the interaction part of the Hamiltonian follows the same scaling when $N\to\infty$ as the external field part. Evaluating Eq. (\[general\_rhoS\_t\_exact\]) and inserting the result into Eq. (\[general\_detM\]) yields $$\det M_{\mathcal{S},\mathrm{\infty}}(t) =\prod_{\bm{s},\bm{s'}} \cos^{N-p}\left[\frac{J t}{N} \sum_{i=1}^p (s_i-s_i')\right].
\label{QIinfrlongh_detM_anyp_1}$$ This expression can be further simplified using a simple combinatorial argument. When varying the spin variables $s_i$’s, each of them being $\pm 1/2$, the sum of them is $$\label{sum_si}
\sum_{i=1}^p s_i=\frac{p-2k}{2}, \quad \binom{p}{k} \textrm{ times}\ , 0\le k \le p.$$ The determinant allowing us to estimate the non-Markovianity of the dynamics is then given by $$\det M_{\mathcal{S},\mathrm{\infty}}(t) =\prod_{j=0}^p \prod_{k=0}^p \left[\cos\left(\frac{J t}{N} (j-k)\right)\right]^{(N-p)\binom{p}{k}\binom{p}{j}}. \label{QIinfrlongh_detM_anyp}$$ In this case again, the witness of non-Markovianity does not depend on the sign of the interaction. We shall now consider two thermodynamic limits: when the system size is fixed and the environment size goes to infinity, and when the system $\mathcal{S}$ consists of a finite fraction of the whole system $\mathcal{S+E}$, i.e. $p=rN$, and $N$ goes to infinity.
The first thermodynamic limit is almost trivial. The product (\[QIinfrlongh\_detM\_anyp\]) contains a finite number of factors. One can use for each factor the Taylor expansion $$\cos\left(\frac{J t}{N} (j-k)\right)^{N-p}\simeq \left(1-\frac{(J t)^2(j-k)^2}{2N^2} \right)^{N-p}\ ,$$ to see that each of them will go to $1$ in the limit $N\to\infty$. Eventually one gets $$\det M_{\mathcal{S},\mathrm{\infty}}(t)=1\label{detM_QIinfrlongh}\ .$$ Following the criterion (\[NMwitness\]) this means that the system’s dynamics is Markovian in this thermodynamic limit. Another way to understand this result is that, in this limit, all the coefficients defined in (\[A\]) become $\A=1$ so that the system’s dynamics (\[rhoS\_t\_exact\]) is the same as if it was isolated hence becomes Markovian.
The second thermodynamic limit, which consists of $p=rN$, i.e. both the system and its environment have a infinitely growing size, requires a bit more care. First, counting each index pair once and doing [the]{} change of variable $q\equiv k-j$, [Eq. ]{}(\[QIinfrlongh\_detM\_anyp\]) can be rewritten $$\det M_{\mathcal{S},\mathrm{\infty}}(t)=\left[\prod_{q=1}^{rN} \cos^2\left(\frac{J t q}{N}\right)^{\sum_{k=q}^{rN}\binom{rN}{k}\binom{rN}{k-q}}\right]^{N(1-r)}\ .\label{detM_QIinfr_thermolim_v0}$$ This expression is convenient to see that $\det M_{\mathcal{S},\mathrm{\infty}}(t)$ is a periodic function of $t$ of period $2\pi N/J$. It reaches the value $1$ when $t$ is an integer multiple of that period. It is enough to restrict ourselves to the behavior during one period. For $0 < t < 2\pi N/J$ at least one factor is smaller than one. As it is raised to a power growing with $N$, it is enough to make the whole product vanish to $0$. This can be more precisely written when $t$ is such that $q Jt/N$ is not a multiple of $\pi$ for any $q$ between $1$ and $rN$. The exponent of each factor can be simplified by using Chu-Vandermonde identity $$\sum_{k=q}^{rN}\binom{rN}{k}\binom{rN}{k-q}=\sum_{k=0}^{rN-q}\binom{rN}{k}\binom{rN}{k+q}=\binom{2 rN}{rN-q}\ .$$ Each factor of the product (\[detM\_QIinfr\_thermolim\_v0\]) is Taylor expanded so that the whole product becomes $$\det M_{\mathcal{S},\mathrm{\infty}}(t)\simeq\left[\prod_{q=1}^{rN} \left(1-\frac{1}{2}\left(\frac{J t q}{N}\right)^2\right)^{\binom{2 rN}{rN-q}}\right]^{2N(1-r)}\ ,$$ which can be rewritten $$\label{detM_QIinfr_thermolim_pN}
\det M_{\mathcal{S},\mathrm{\infty}}(t)\simeq \exp\left[ -\frac{1-r}{N}\left({J t}\right)^2 \sum_{q=1}^{rN} \binom{2rN}{rN-q} q^2 \right] \ .$$ As the sum grows at least exponentially when increasing $N$, the determinant converges to $0$ [for all times]{}, which means that the dynamics is Markovian in this limit.
### Power law range Ising model
Here a slightly more general model of spin system is investigated, which includes as limiting cases both the previous examples. Consider a one-dimensional chain, where the interaction between any two spins depends on the distance between those spins through a power law (PL). More specifically, the pairwise correlation matrix is chosen as $\Cij=J_N(\alpha)/r_{ij}^\alpha$ ($J_N(\alpha)>0$) for $i\ne j$ and zero otherwise, where $\alpha$ is the parameter ruling the range of the interaction, and $r_{ij}$ denotes the distance between the $i$th and $j$th sites. The interaction strength $J_N(\alpha)$ depends both on $N$ and $\alpha$. This model is convenient to interpolate between the more common nearest-neighbor interaction ($\alpha\to\infty$) and the infinite range interaction ($\alpha\to 0$). Note that this model for $\alpha=3$ is similar to the RKKY model [@ruderman_indirect_1954; @kasuya_electrical_1956; @yosida_magnetic_1957], and has been previously intensively studied in a spin glass perspective, see e.g. [@walker_computer_1980; @haussler_interrelations_1992]. Using Eqs. (\[general\_rhoS\_t\_exact\]) and (\[general\_detM\]), the witness for non-Markovianity for the dynamics of $\mathcal{S}$ is obtained by checking the variations of $$\label{QIpowrlongh_detM}
\det M_{\mathcal{S},\mathrm{PL}}(t)=\prod_{\bm{s},\bm{s'}}
\prod_{j=p+1}^N
\cos\left[J_N(\alpha) t\left(\sum_{i=1}^p \frac{s_i-s'_i}{r_{ij}^\alpha} \right) \right].$$
Again it is customary to ask whether non-Markovianity survives at the thermodynamic limit of large size. There can be two options for the choice the interaction constant $J_N(\alpha)$: it can be independent of $N$ as for optical atom systems [@beguin_direct_2013; @barredo_demonstration_2014; @saffman_quantum_2010], or it can scale with $N$ to have a unit mean field temperature, see e.g. [@katzgraber_monte_2003]. In both cases we can argue qualitatively the same behavior for the non-Markovian character dynamics of the system. Similarly to the previous case of infinite range system, the non-Markovianity witness is a product of periodic functions. The crucial difference is that all the factors display now incommensurable frequencies. Therefore we predict that, in the large $N$ limit, the whole product should vanish, which is supported by our numerics. In other words the product (32), which is a special case of (22), contains an infinite number of factors. Each of them are raised to a power growing with $N$ so that they become non-zero only for a discrete set of times in the thermodynamic limit. This set is different for each factor so that the whole product vanishes for all time. The situation is different as soon as the support of the interaction is finite. This means that only a finite number of $J_{ij}$ in (22) are non zero. The product now contains a finite number of oscillating factor, hence can generically have piecewise a positive derivative. This is the reason why we conjecture that the dynamics is Markovian at all times whenever the support of the interaction between the system and its environment is infinite, and can become non-Markovian in the case of finitely supported interaction.
Influence of the dimension and of the temperature
-------------------------------------------------
### Higher dimensional spin lattice
It is worth emphasizing that our results (\[rhoS\_t\_exact\]),(\[A\]) and (\[general\_detM\]) can be applied to other partitions. This is particularly relevant for higher dimensional model. For the sake of illustration we will investigate the case of spins-$1/2$ located on a two-dimensional square lattice interacting via a nearest neighbor interaction, and with periodic boundary conditions. In order to use our general results, we consider a lattice $\mathcal{S}+\mathcal{E}$ made of $N=M^2$ sites. The systems $\mathcal{S}$ here consists of the $p=q^2$ spins in the square sub-lattice in the upper left corner. Each spin $s_{ix,iy}^z$ is now labeled with two spatial indices $(ix,iy)$, which locates its position along both directions of the lattice. These two indices can be combined in a single index $i$ ranging from $1$ to $N=M^2$ using $i=(ix-1) {M}+(iy-1)+1.$ In order to facilitate the physical interpretation, we will use the 2d indices $(ix,iy)$ in the following discussion. The nearest-neighbor interaction corresponds to the pairwise correlation matrix given by $J_{(ix,iy),(jx,jy)} = J$ ($J>0$) for $(jx,jy)=(ix,iy+1)$, $(ix,iy-1)$, $(ix+1,iy)$ and $(ix-1,iy)$ and $J_{(ix,iy),(jx,jy)}= 0$ otherwise. In order to ensure periodic boundary conditions, an index taking the value $0$ (resp. $M+1$) corresponds to $M$ (resp. $1$). The initial state of the environment is, in analogy with the one dimensional case (\[rhoE\_mixed\]), $$\rho_\mathcal{E}(0)=\frac{\mathbb{1}_\mathcal{E}}{2^{M^2-q^2}}\ .
\label{rhoE_mixed2d}$$ Inserting (\[rhoE\_mixed2d\]) into (\[A\]) and using the definition of the pairwise correlation matrix given previously leads to $$\begin{aligned}
\A &=\prod_{i=1}^q \cos[Jt(s_{1,i}-s_{1,i}')]\prod_{i=1}^q \cos[Jt(s_{q,i}-s_{q,i}')]
\nonumber\\
&\prod_{i=1}^q \cos[Jt(s_{i,1}-s_{i,1}')]\prod_{i=1}^q \cos[Jt(s_{i,q}-s_{i,q}')]\ .\label{A_shr2d}\end{aligned}$$ The first two products in (\[A\_shr2d\]) correspond respectively to the coupling of the first and last rows of spins in $\mathcal{S}$ with the environment. Similarly, the last two products in (\[A\_shr2d\]) correspond respectively to the coupling of the first and last columns of spins in $\mathcal{S}$ with the environment. Therefore, this shows that only the coupling at the boundary between [the system]{} and the environment contributes to non-Markovianity. This result is similar to the case of the one-dimensional chains with nearest neighbor interaction previously discussed, see Eq. (\[QIshrlongh\_rhoS\_t\_exact\]). The last step consists of computing the non-Markovianity witness using (\[general\_detM\]). The determinant consists of $2^{2q^2}$ factors in two dimensions. There are exactly $2^{2(q^2-1)}$ factors for which $s_{ix,iy}^z$ and $s_{ix,iy}^{\prime\;z}$ are fixed for a given location $(ix,iy)$. One needs to distinguish between $4(q-2)$ edge sites located at $(ix,iy)\in\{(1,i), (i,q), (q,i),(i,1)\}$ for $2\le i\le q-1$ and $4$ corner sites located at $(1,1)$, $(1,q)$, $(q,q)$ and $(q,1)$. Following (\[A\_shr2d\]) the contribution of a given edge site is $$\left[\prod_{s_{ix,iy}^z=\pm 1/2}\prod_{s_{ix,iy}^{\prime\;z}=\pm 1/2} \cos[Jt(s_{ix,iy}^z-s_{ix,iy}^{\prime\;z})]\right]^{2^{2q^2-2}} \ ,$$ whereas the contribution of any of the four corner sites is $$\left[\prod_{s_{ix,iy}^z=\pm 1/2}\prod_{s_{ix,iy}^{\prime\;z}=\pm 1/2} \cos^2[Jt(s_{ix,iy}^z-s_{ix,iy}^{\prime\;z})]\right]^{2^{2q^2-2}} \ ,$$ Multiplying all those contributions leads to the exact formula for the non-Markovianity witness for a two-dimensional [square]{} lattice $$\det M_{\mathcal{S},\mathrm{nn}}(t) =\left[\cos\left( Jt\right)\right]^{q 2^{2q^2+1}},\, M-q\ge 1 \label{detM_QIshrlongh2d} \ .$$ Again it is worth stressing that this formula proves that the dynamics of the sub-lattice will remain non-Markovian for an arbitrary size of the surrounding environment. Conversely, when the size of the system is taken as a finite fraction size of its environment $(q=rM)$, its dynamics becomes Markovian.
### Finite temperature state for the environment
It is worth noticing that our results can be generalized to account for the effect of the temperature. We will illustrate this for the case of the one-dimensional spin$-1/2$ chain with nearest neighbor interaction, and a homogeneous external field.
Start from an initial density matrix for the environment at a given finite temperature $T$: $$\rho_\mathcal{E}(0)=\sum_{\bm{\sigma}} \frac{e^{-\beta H_\mathcal{E}(\bm{\sigma})}}{Z}\left| \bm{\sigma}\right>\left<\bm{\sigma}\right|\;,
\label{rhoE_finiteT}$$ where $\beta=1/k_B T$ is the inverse temperature. The Hamiltonian of the environment is, see Eq. (\[HclasE\]): $$H_\mathcal{E}(\bm{\sigma})=-J\sum_{i=p+1}^{N-1} \sigma_i\sigma_{i+1} + h \sum_{i=p+1}^N \sigma_i\ .
\label{HqE_open}$$ Note that this subchain, defining the environment, obeys open boundary conditions. Last the partition function $Z$ in (\[rhoE\_finiteT\]) is given by: $$Z\equiv Z(T,h)=\sum_{\bm{\sigma}} e^{-\beta H_\mathcal{E}(\bm{\sigma})}$$ As detailed in \[mainderiv\] the way to assess the non-Markovian character of the dynamics will be achieved in two steps. First the coefficients $\A$ as defined in Eq. (\[defA\]) are computed. Then the determinant (\[general\_detM\]) and its first derivative are evaluated numerically. This is illustrated in Fig. \[NM\_vs\_T\], which shows the non-Markovianity witness for different temperatures of the environment. Notice that, similarly to (\[detM\_QIshrlongh\]), the determinant is a periodic function of the time $t$ with period $2\pi/J$. Hence it is plotted only over one period. It can be seen that the dynamics is non Markovian for any non vanishing temperature. Note that, for $T=0$, the initial density matrix of the environment is $|\boldsymbol{\sigma_0}\rangle\langle \boldsymbol{\sigma_0}|$ in the computational basis, where $\left|\boldsymbol{\sigma_0}\right>$ is the ground state of the Hamiltonian. As mentioned earlier after Eq. (\[NMwitness\]), this leads trivially to a Markovian dynamics for the system.
![Non-Markovianity witness (\[NMwitness\]) of a system of two spin-1/2 with an environment made of $8$ spin-1/2 as a function of time. Here, $\mathcal{S}+\mathcal{E}$ is a chain of $N=10$ spins-1/2 with nearest-neighbor interactions, periodic boundary conditions and $h=J$. The environment is initially in a thermal state (\[rhoE\_finiteT\]) with $\beta=0$ (black), $\beta =1/J$ (blue dashed), $\beta=3/J$ (green dotted) and $\beta\rightarrow\infty$ (red dot-dashed).[]{data-label="NM_vs_T"}](fig1.pdf)
Comparison with other non-Markovianity measures for systems of $p=1$ spin {#compmeas}
-------------------------------------------------------------------------
Our results can be used in order to compare different measures of non-Markovianity [@neto_inequivalence_2016]. For the sake of illustration, let us consider here the special case of a system consisting of one spin-$1/2$ ($p=1$). Using (\[general\_detM\]) and (\[general\_rhoS\_t\_exact\]) valid for the environment initially in the maximally mixed state, the determinant of the evolution operator is $$\det M_{\mathcal{S}}(t)=A(t)^2\ ,$$ with $$\label{def_f0}
A(t)=\prod_{j=2}^N \cos\left(J_{1j} t \right).$$ such that, following the criterion (\[NMwitness\]), the dynamics is non-Markovian whenever $$\label{NM_p1}
A(t)A'(t)>0\ .$$ In order to evaluate other witnesses of non-Markovianity, we write explicitly the reduced density matrix of $\mathcal{S}$ at any time $t>0$. Eq. (\[rhoS\_t\_exact\]) yields $$\label{general_rhoS_t_exact_p1}
\rho_\mathcal{S}(t)=\left(
\begin{array}{cc}
\rho_{11} & \rho_{12} A(t)\, e^{-{\mathrm{i}}h_1 t} \\
\rho_{21} A(t)\, e^{{\mathrm{i}}h_1 t} & \rho_{22}
\end{array}\right),$$ where $\rho_{ij}$ ($i,j=1,2$) are the coefficients of the initial density matrix of $\mathcal{S}$ at $t=0$. One can get the corresponding Kraus representation (see e.g. [@andersson_finding_2007]) and deduce from it the master equation for the reduced density matrix [@hall_canonical_2014; @bhattacharya_exact_2017] $$\frac{{\mathrm{d}}}{{\mathrm{d}}t} \rho_\mathcal{S}(t)=-{\mathrm{i}}\left[H^{\rm eff}_\mathcal{S},\rho_\mathcal{S}(t)\right]
+\Gamma_z(t) \Big( \sigma^z \rho_\mathcal{S}(t)\sigma^z-\rho_\mathcal{S}(t)\Big)
\label{master_eq_p1},$$ with the effective Hamiltonian $H^{\rm eff}_\mathcal{S}=h_1 \sigma^z/2$ where $\sigma^z$ stands for the usual Pauli matrix. This master equation models a pure dephasing channel with a time-dependent rate $$\Gamma_z(t)=-\frac{A'(t)}{2A(t)}. \label{Gammaz_p1}$$ The master equation (\[master\_eq\_p1\]), of the GSKL form, can be used to evaluate divisibility criterion, as it can be expressed as a sign constraint on the rate in the master equation. The RHP measure detects a non-Markovian behavior when the rate in the master equation becomes negative [@rivas_quantum_2014]. Due to the explicit expression (\[Gammaz\_p1\]) the dynamics will be non-Markovian if $-A'(t)/A(t)<0$, which trivially agrees with our witness (\[NM\_p1\]). Knowing the exact expression (\[general\_rhoS\_t\_exact\_p1\]) of $\rho_\mathcal{S}(t)$ enables one also to compute BLP distance measure of non-Markovianity [@laine_measure_2010]. The trace distance between two arbitrary states $\rho^a_0$ and $\rho^b_{0}$ is given by $$\begin{aligned}
D(\rho_a(t),\rho_b(t))&={\mathrm{ tr }\;}\left(\sqrt{(\rho_a(t)-\rho_b(t))(\rho_a(t)-\rho_b(t))^\dagger}\right)\nonumber\\
&=\sqrt{(\rho_{11}^a-\rho_{11}^b)^2+A(t)^2|\rho_{12}^a-\rho_{12}^b|^2}.
\label{BLP_p1}\end{aligned}$$ The system is said to be non-Markovian according to the BLP measure whenever $$\frac{{\mathrm{d}}}{{\mathrm{d}}t}D(\rho_a(t),\rho_b(t))=\frac{|\rho_{12}^a-\rho_{12}^b|^2 A(t) A'(t)}{\sqrt{(\rho_{11}^a-\rho_{11}^b)^2+A(t)^2|\rho_{12}^a-\rho_{12}^b|^2}}$$ is strictly positive. As here, $0\leq A(t)^2\leq 1 $, and, for any density operator of a two-level system, we have $|\rho_{12}|\leq \rho_{11}\leq 1$ and $|\rho_{12}|\leq 1/2$, see e.g. [@HornJohnson], the maximum of this expression is reached for $\rho_{11}^a=\rho_{11}^b$ and $\rho_{12}^a=-\rho_{12}^b=1/2$. This condition for non-Markovianity is satisfied whenever $A(t) A'(t)>0$ which agrees again with (\[NM\_p1\]).
Entanglement and non-Markovianity {#correlations}
=================================
The aim of this Section is to investigate the relation between the non-Markovianity of the system $\mathcal{S}$ and the generation of entanglement with the environment $\mathcal{E}$. Let us remind that we consider an initial state without system-environment entanglement of the form (\[rho\_init\_sep\]).
First, let us show that the dynamics of $\mathcal{S}$ can display non-Markovianity, according to the witness , without generating any entanglement with the environment. For this purpose, we consider an initial separable state of the system and the environment as in (\[rho\_init\_sep\]), the initial density matrix of the latter being a classical mixture of computational basis state $$\rho_\mathcal{E}(0)=\sum_{ \bm{\sigma}'} a_{ \bm{\sigma}', \bm{\sigma}'} \left| \bm{\sigma}'\right>\left< \bm{\sigma}'\right|.
\label{rhoE_mixture}$$ According to our previous analysis, the system’s non-Markovianity is given in this case by Eqs. (\[general\_detM\])-(\[NMwitness\]). Writing the initial state of $\mathcal{S}$ as $$\rho_\mathcal{S}(0)=\sum_{ \bm{s},\bm{s}'} r_{ \bm{s}, \bm{s}'} \left| \bm{s}\right>\left< \bm{s}'\right|$$ and using Eqs. (\[diag\]), (\[HclasS\]), (\[HclasE\]), and (\[HclasSE\]), we obtain $$\rho_{\mathcal{S}+\mathcal{E}}(t) = \sum_{\bm{\sigma}} a_{ \bm{\sigma}, \bm{\sigma}}\left(\rho_{\mathcal{S}|\bm{\sigma}}(t)\otimes\left| \bm{\sigma}\right>\left< \bm{\sigma}\right|\right)
\label{rhoSE_mixture}$$ with the conditional state of the system $$\begin{aligned}
\rho_{\mathcal{S}|\bm{\sigma}}(t)=\sum_{\bm{s},\bm{s}'} &e^{{\mathrm{i}}t [H_\mathcal{S}(\bm{s}')-H_\mathcal{S}(\bm{s})+H_\mathcal{SE}(\bm{s}',\bm{\sigma})-H_\mathcal{SE}(\bm{s},\bm{\sigma})]}\\
&\times r_{ \bm{s}, \bm{s}'} \left| \bm{s}\right>\left< \bm{s}'\right|.
\end{aligned}$$ Therefore, we see that the global system $\mathcal{S}+\mathcal{E}$ stays in a separable state at all times as shown by Eq. (\[rhoSE\_mixture\]), independently of the non-Markovianity of $\mathcal{S}$. Moreover, the state has, by definition, zero discord with respect to the environment [@Olliver2001; @Henderson2001]. Note that, similarly, if the system $\mathcal{S}$ starts in a classical mixture of computational basis states, the global system $\mathcal{S}+\mathcal{E}$ stays in a separable state at all times independently of the initial state of the environment. This result is in agreement with previous works on qubit-environment entanglement generation during pure dephasing dynamics [@Roszak2015; @Roszak2018].
Let us now show that the system and its environment can get entangled during the dynamics, when the initial state of the environment $\rho_\mathcal{E}(0)$ has non-vanishing coherences $a_{ \bm{\sigma}', \bm{\sigma}''}$ in the computational basis. As an illustration, we consider a chain of $N=10$ spin-$1/2$ with infinite range or nearest neighbors interaction [and various sizes of the system $\mathcal{S}$]{}. The presence of entanglement between $\mathcal{S}$ and $\mathcal{E}$ is assessed using the negativity $$\mathcal{N}\left(\rho_{\mathcal{S}+\mathcal{E}}(t)\right)=\frac{||\rho_{\mathcal{S}+\mathcal{E}}^{T_\mathcal{S}}(t)||_1-1}{2}$$ where $||\rho||_1=\mathrm{Tr}(\sqrt{\rho\rho^\dagger})$ and $\rho_{\mathcal{S}+\mathcal{E}}^{T_\mathcal{S}}(t)$ is the partial transpose of $\rho_{\mathcal{S}+\mathcal{E}}(t)$ with respect to $\mathcal{S}$. The Peres-Horodecki negativity criterion [@Peres1996; @Horodecki1996] states that whenever the negativity is non-zero, the bipartite system $\mathcal{S}+\mathcal{E}$ is entangled. This criterion is necessary and sufficient in the case of two spin-$1/2$ and two spin-$1$. For higher dimensional systems, all separable states have zero negativity but there also exist entangled states with zero negativity. Figure \[fig\_neg\] illustrates that $\mathcal{N}(\rho_{\mathcal{S}+\mathcal{E}}(t))$ oscillates as a function of time for a system $\mathcal{S}$ made of $p=3$ spins and a given initial state (\[rho\_init\_sep\]). Numerical simulations showed that whenever the coherences of the initial density matrix of the environment are non-vanishing, the dynamics typically generates entanglement between the system and its environment.
![Negativity $\mathcal{N}$ between the system $\mathcal{S}$ made of $p=3$ spin-$1/2$ and the environment $\mathcal{E}$ made of $7$ spin-$1/2$, as a function of time for (a) infinite range interactions, and (b) nearest-neighbor interactions with periodic boundary conditions and $h_i=0$ for all $i=1,\dots,10$. The system $\mathcal{S}$ is initially in the pure state $|\psi_\mathcal{S}(0)\rangle=\sum_{\bm{s}}|\bm{s}\rangle/{2^{p/2}}$. The black solid curves correspond to the environment initially in the state $|\psi_\mathcal{E}(0)\rangle=\sum_{\bm{\sigma}}|\bm{\sigma}\rangle/{2^{7/2}}$, and the red dashed curves to the environment initially in a classical mixture of computational basis state .[]{data-label="fig_neg"}](fig2.pdf)
We have shown in Sec. \[mainderiv\] that, for any given separable global state of the form (\[rho\_init\_sep\]), the non-Markovianity of the system is independent of the coherences of the initial density matrix of the environment. The reason is that the reduced dynamics of $\mathcal{S}$ given by Eqs. and is independent of the off-diagonal elements of $\rho_\mathcal{E}(0)$. Yet, having non-zero coherences will lead to the generation of entanglement between the system and its environment, see Fig. \[fig\_neg\], whereas the initial density matrix of $\mathcal{E}$ with the same populations and no coherence will lead to a separable dynamics, see (\[rhoSE\_mixture\]). As a consequence, we claim that, for our model, the non-Markovianity is independent of the generation of entanglement between the system and its environment.
Last, although our model is sufficiently simple to allow for analytical calculations, it is interesting to note that spins within the system $\mathcal{S}$ undergoing non-Markovian dynamics can display non-trivial entanglement dynamics as illustrated in Fig. \[fig\_neg\_S\]. In particular, when $\rho_{\mathcal{S}+\mathcal{E}}(0)$ is a fully separable $N$-spin state, we observe that the system can display sudden-death and revival of entanglement [@bellomo_non-markovian_2007], whereas the environment stays at all times in a separable state.
![Negativity $\mathcal{N}$ between two spin-$1/2$ that define the system $\mathcal{S}$ as a function of time when $\mathcal{S}+\mathcal{E}$ is a chain of $N$ spin-$1/2$ with nearest-neighbor interactions and periodic boundary conditions and $h_i=0$ for all $i=1,\dots,N$. This results is valid for any size of the environment as soon as $N\geq 4$. The environment is initially in the maximally mixed state (\[rhoE\_mixed\]). The black curve corresponds to the system $\mathcal{S}$ initially in the separable pure state $\left(|-\tfrac{1}{2}\rangle +|\tfrac{1}{2}\rangle\right)\otimes\left(|-\tfrac{1}{2}\rangle +|\tfrac{1}{2}\rangle\right)/2$ while the red dashed curve corresponds to $\mathcal{S}$ initially in the entangled pure state $\left(|-\tfrac{1}{2}\rangle \otimes |-\tfrac{1}{2}\rangle +|\tfrac{1}{2}\rangle \otimes |\tfrac{1}{2}\rangle\right)/\sqrt{2}$. []{data-label="fig_neg_S"}](fig3.pdf)
Conclusion
==========
In this paper, we investigated analytically the quantum non-Markovianity of a spin ensemble ($\mathcal{S}$) undergoing a pure dephasing dynamics arising from the unitary evolution of a larger spin ensemble ($\mathcal{S}+\mathcal{E}$) governed by the Hamiltonian (\[Hgeneral\]). One of our main results is given by Eqs. (\[A\])–(\[general\_detM\]) that apply to spin ensembles of arbitrary size and spin quantum number and allows us to determine analytically whether the dynamics is Markovian or not. For a spin-$1/2$ ensemble $\mathcal{S}+\mathcal{E}$ of finite size, we found out that, when the environment $\mathcal{E}$ is initially in the maximally mixed state, the dynamics of $\mathcal{S}$ is always non-Markovian. We also obtained analytical results in the thermodynamic limit for one-dimensional spin chains. In the limit of infinite size of the environment with fixed size of the system, the quantum dynamics of the system stays non-Markovian for nearest-neighbor interactions whereas it becomes Markovian for infinite range interactions, see Eq. (\[detM\_QIshrlongh\]) vs Eq. (\[detM\_QIinfrlongh\]). In the limit of infinite size of the environment with the size of the system being a fixed fraction of the ensemble $\mathcal{S}+\mathcal{E}$, the quantum dynamics of the system becomes Markovian both for nearest-neighbor and infinite range interactions. In these limits, we found out that Markovianity can appear when (i) the non-Markovian episodes are separated by a period whose value goes to infinity (cases studied with infinite range interaction), and (ii) the non-Markovian episodes occur for a duration shrinking to zero (cases studied with infinite range or nearest-neighbor interaction with a system size being a fixed fraction of $N$). We gave another application of our results to two-dimensional square spin lattice. We also showed that, for our system, non-Markovianity does not stem from the generation of entanglement with the environment. Although these observations are specific to our system, they raise the more general question of the relationship between non-Markovianity and system-environment correlations. Natural extensions of this work include the study of dynamics [more general than purely dephasing]{} or non-integrable dynamics [@ignacio_classical_chaos; @davalos_quantum_2017], e.g. in the presence of transverse field. Experimental realizations of the system studied in this work could be realized with cold atoms in optical lattices, see e.g. [@navarrete-benlloch_simulating_2011].
This work was supported by the ARC grant QUANDROPS 12/17-02.
Determinant of the time evolution operator for the reduced dynamics {#Appendix_A}
===================================================================
We start by explaining how to write the time evolution operator of the Bloch vector when the coefficients of the density matrix are explicitly known. It will be illustrated for the system considered in the main part of the paper: $p$ spin-$S$ interacting via a pairwise interaction, see e.g. (\[HqS\]). In particular the dimension of the Hilbert space of the system under consideration is $D=(2S+1)^p$. The Bloch parametrisation for a density matrix $\rho$ of size $D\times D$ (see e.g. [@mahler_quantum_1998]) consists of re-arranging the $D^2$ entries of the density matrix into a vector, called the Bloch vector. The coordinates $r_j$ of the Bloch vector are called the Bloch parameters. They are divided into two sets: one set containing $D(D-1)$ real Bloch coordinates to parametrise the off diagonal elements $\rho_{ij}$ $(i\ne j)$ of the density matrix. They can be grouped in pairs, for the real and the imaginary part respectively. More precisely one can define $$\begin{aligned}
\hspace{-1cm}r_1={\mathrm{ Re }\;}(\rho_{12}),\;& r_2={\mathrm{ Im }\;}(\rho_{12})\nonumber\\
\hspace{-1cm}r_3={\mathrm{ Re }\;}(\rho_{13}),\;& r_4={\mathrm{ Im }\;}(\rho_{13})\nonumber\\
\vdots & \vdots\nonumber\\
r_{2(D-1)+1}={\mathrm{ Re }\;}(\rho_{21}),\;&\ r_{2(D-1)+2}={\mathrm{ Im }\;}(\rho_{21})\nonumber\\
r_{2(D-1)+3}={\mathrm{ Re }\;}(\rho_{23}),\;&\ r_{2(D-1)+4}={\mathrm{ Im }\;}(\rho_{23})\nonumber\\
\vdots & \vdots\nonumber\\
\hspace{-1cm}r_{D(D-1)-1}={\mathrm{ Re }\;}(\rho_{D-1\; D}),\; & r_{D(D-1)}={\mathrm{ Im }\;}(\rho_{D-1\; D})\nonumber\end{aligned}$$ The second set of the $D^2$ Bloch coordinates are formed by $D$ linear combinations of the diagonal elements of the matrix. $$r_{D(D-1)+l}=\sqrt\frac{2}{l(l+1)}\left(\sum_{k=1}^{l} \rho_{kk}-l\rho_{l+1\; l+1}\right)\ ,$$ for $1\le l \le D-1$. The last remaining coefficient is chosen by convention to be $$r_{D^2}=\sum_{k=1}^{D} \rho_{kk}\ ,$$ so that it is unity for a density matrix. If the $D^2-$dimensional Bloch vector corresponding to the matrix at time $t$ is denoted by $\mathbf{r}(t)$, one can define its time evolution operator $M_{\mathcal{S}}(t)$ through $$\mathbf{r}(t)=M_{\mathcal{S}}(t)\,\mathbf{r}(0). \label{prop_vBloch}$$ It can be shown that the operator $M_{\mathcal{S}}(t)$ is linear, hence can be represented by a $D^2\times D^2$ matrix.
The explicit expression (\[rhoS\_t\_exact\]) allows for a direct evaluation of the coefficients of the matrix representing $M_{\mathcal{S}}(t)$. As the diagonal elements of the density matrix are unchanged, the evolution operator boils down to the identity in the subspace spanned by the second set of Bloch coordinates, as defined above. For the first set, it can be seen directly from (\[rhoS\_t\_exact\]) that each pair of Bloch coordinates $(r_{2j-1},r_{2j})$ for $1\le j \le D(D-1)/2$ follow a rotation, expressed by the time dependent phase, and a dilatation expressed by the factor $\A$ $$\left(
\begin{array}{c}
r_{2j-1}(t) \\
r_{2j}(t)
\end{array}\right)= \mathcal{O}_j
\left(
\begin{array}{c}
r_{2j-1}(0) \\
r_{2j}(0)
\end{array}\right)$$ with $$\mathcal{O}_j=\left(
\begin{array}{cc}
\A \cos \theta_{\bm{s},\bm{s'}}t & \A\sin \theta_{\bm{s},\bm{s'}}t \\
-\A\sin \theta_{\bm{s},\bm{s'}}t & \A \cos \theta_{\bm{s},\bm{s'}}t
\end{array}\right)\ ,$$ where the notation $\theta_{{\bf s},{\bf s'}}=H_\mathcal{S}(\bm{s'})-H_\mathcal{S}(\bm{s})$ was introduced for the sake of brevity. In other words the matrix $M_{\mathcal{S}}(t)$ in (\[prop\_vBloch\]) can be written in a block structure for the first set of Bloch coordinates $$M_{\mathcal{S}}(t)=\left(
\begin{array}{cccc}
\mathcal{O}_1 & 0 & \dots & 0\\
0 & \mathcal{O}_2 & \dots & 0\\
\vdots & \vdots & \ddots & \vdots \\
0& 0 & \dots & \mathcal{O}_{D(D-1)/2}
\end{array}\right)\ ,$$ and its determinant is directly given by $$\det M_{\mathcal{S}}(t)=\prod_{j=1}^{D(D-1)/2} \det \mathcal{O}_j=\prod_{\bm{s},\bm{s'}} \A \ ,$$ which is exactly (\[general\_detM\]).
[99]{}
E. Joos, H. D. Zeh, C. Kiefer, D. Giulini, J. Kupsch, and I.-O. Stamatescu, [*Decoherence and the [Appearance]{} of a [Classical]{} [World]{} in [Quantum]{} [Theory]{}*]{}. Berlin, Heidelberg: Springer Berlin Heidelberg, 2003. DOI: 10.1007/978-3-662-05328-7.
W. H. Zurek, Physics Today [**44**]{}, 36 (1991)
S. Haroche, Physics Today [**51**]{}, 36 (1998)
I. de Vega, D. Porras, and J. Ignacio Cirac, Phys. Rev. Lett. [**101**]{}, 260404 (2008)
B. Bylicka, M. Tukiainen, D. Chruściński, J. Piilo, and S. Maniscalco, Sci. Rep. [**6**]{}, 27989 (2016)
M. Pezzutto, M. Paternostro, and Y. Omar, New J. Phys. [**18**]{}, 123018 (2016)
B. Bylicka, D. Chruściński, and S. Maniscalco, Sci. Rep. [**4**]{}, 5720 (2014)
B. Bellomo, R. Lo Franco, and G. Compagno, Phys. Rev. Lett. [**99**]{}, 160502 (2007)
Y. Matsuzaki, S. C. Benjamin, and J. Fitzsimons, Phys. Rev. A [**84**]{}, 012103 (2011)
A. W. Chin, S. F. Huelga, and M. B. Plenio, Phys. Rev. Lett. [**109**]{}, 233601 (2012)
H. S. Dhar, M. N. Bera, and G. Adesso, Phys. Rev. A [**91**]{}, 032115 (2015)
J. R. Glick and C. Adami, arXiv preprint arXiv:1701.05636 (2017)
A. Gonzalez-Tudela, F. J. Rodríguez, L. Quiroga, and C. Tejedor, Phys. Rev. B [**82**]{}, 115334 (2010)
M. Malekakhlagh, A. Petrescu, and H. E. Türeci, Phys. Rev. A [**94**]{}, 063848 (2016)
Y.-L. L. Fang, F. Ciccarello, and H. U. Baranger, New J. Phys. [**20**]{}, 043035 (2018)
Á. Rivas, S. F. Huelga, and M. B. Plenio, Rep. Prog. Phys. [**77**]{}, 094001 (2014)
H.-P. Breuer, E.-M. Laine, J. Piilo, and B. Vacchini, Rev. Mod. Phys [**88**]{}, 021002 (2016)
I. de Vega and D. Alonso, Rev. Mod. Phys. [**89**]{}, 015001 (2017)
F. Shibata, Y. Takahashi, and N. Hashitsume, J. Stat. Phys. [**17**]{}, 171 (1977)
J. Wilkie, Phys. Rev. E [**62**]{}, 8808 (2000)
E.-M. Laine, K. Luoma, and J. Piilo, J. Phys. B [**45**]{}, 154004 (2012)
H.-J. Briegel, and B.-G. Englert, Phys. Rev. A [**47**]{}, 3311 (1993)
S. Maniscalco, Phys. Rev. A [**75**]{}, 62103 (2007)
H. Z. Shen, D. X. Li, Shi-Lei Su, Y. H. Zhou, and X. X. Yi, Phys. Rev. A [**96**]{}, 033805 (2017)
J. Fischer and H.-P. Breuer, Phys. Rev. A [**76**]{}, 052119 (2007)
S. Bhattacharya, A. Misra, C. Mukhopadhyay, and A. K. Pati, Phys. Rev. A [**95**]{}, 012122 (2017)
T. Prosen and B. Žunkovič, New J. Phys. [**12**]{}, 025016 (2010)
M. Mahmoudi, S. Mahdavifar, T. Mohammad Ali Zadeh, and M. R. Soltani, Phys. Rev. A [**95**]{}, 012336 (2017)
B. Buča and T. Prosen, New J. Phys. [**14**]{}, 073007 (2012)
C. Cormick, A. Bermudez, S. F. Huelga, and M. B. Plenio, New J. Phys. [**15**]{}, 073027 (2013)
A. Smirne, L. Mazzola, M. Paternostro, and B. Vacchini, Phys. Rev. A [**87**]{}, 052129 (2013)
P. Ribeiro and V. R. Vieira, Phys. Rev. B [**92**]{}, 100302(R) (2015)
H.-S. Zeng, N. Tang, Y.-P. Zheng, and G.-Y. Wang, Phys. Rev. A [**84**]{}, 032118 (2011)
Á. Rivas, S. F. Huelga, and M. B. Plenio, Phys. Rev. Lett. [**105**]{} 050403 (2010)
E.-M. Laine, J. Piilo, and H.-P. Breuer, Phys. Rev. A [**81**]{}, 062115 (2010)
S. Lorenzo, F. Plastina, and M. Paternostro, Phys. Rev. A [**88**]{}, 020102(R) (2013)
F. A. Wudarski, F. Petruccione, Eur. Phys. Lett. [**113**]{}, 50001 (2016)
A. Budini, arXiv preprint arXiv:1804.03305 (2018)
G. Mahler and V. A. Weberruß, [*Quantum [Networks]{}*]{}. Berlin, Heidelberg: Springer Berlin Heidelberg, 1998. DOI: 10.1007/978-3-662-03669-3.
D. Sherrington and S. Kirkpatrick, Phys. Rev. Lett. [**35**]{}, 1792 (1975)
Except for the set of point where $\protect \qopname \relax o{cos}\left (
Jt\right )=\pm 1$, which is of measure zero. So we neglect it for the discussion of non-Markovianity.
W. A. Coish and D. Loss, Phys. Rev. B [**70**]{}, 195340 (2004)
M. A. Ruderman and C. Kittel, Phys. Rev. [**96**]{}, 99 (1954)
T. Kasuya, Prog. Theor. Phys. [**16**]{}, 58 (1956)
K. Yosida, Phys. Rev. [**106**]{}, 893 (1957)
L. R. Walker and R. E. Walstedt, Phys. Rev. B [**22**]{}, 3816 (1980)
P. Häussler, Phys. Rep [**222**]{}, 65 (1992)
L. Béguin, A. Vernier, R. Chicireanu, T. Lahaye, and A. Browaeys, Phys. Rev. Lett. [**110**]{} 263201 (2013)
D. Barredo, S. Ravets, H. Labuhn, L. Béguin, A. Vernier, F. Nogrette, T. Lahaye, and A. Browaeys, Phys. Rev. Lett. [**112**]{}, 183002 (2014)
M. Saffman, T. G. Walker, and K. Mølmer, Rev. Mod. Phys. [**82**]{}, 2313 (2010)
H. G. Katzgraber and A. P. Young, Phys. Rev. B [**67**]{}, 134410 (2003)
A. C. Neto, G. Karpat, and F. F. Fanchini, Phys. Rev. A [**94**]{}, 032105 (2016)
E. Andersson, J. D. Cresser, and M. J. W. Hall, J. Mod. Opt [**54**]{}, 1695 (2007)
M. J. W. Hall, J. D. Cresser, L. Li, and E. Andersson, Phys. Rev. A [**89**]{}, 042120(2014)
R. A. Horn, C. R. Johnson, [*Matrix Analysis*]{}, Cambridge Univ. Press, 1990
H. Olliver and W. H. Zurek, Phys. Rev. Lett. [**88**]{}, 017901 (2001)
L. Henderson and V. Vedral, J. Phys. A [**34**]{}, 6899 (2001)
K. Roszak and Ł. Cywi' nski, Phys. Rev. A [**92**]{}, 032310 (2015)
K. Roszak and Ł. Cywi' nski, Phys. Rev. A [**97**]{}, 012306 (2018)
A. Peres, Phys. Rev. Lett. [**77**]{}, 1413 (1996)
M. Horodecki, P. Horodecki, and R. Horodecki, Phys. Lett. A [**223**]{}, 1 (1996)
I. García-Mata, C. Pineda, D. Wisniacki, Phys. Rev. A [**86**]{}, 022114 (2012)
D. Davalos and C. Pineda, Phys. Rev. A [**96**]{}, 062127 (2017)
C. Navarrete-Benlloch, I. de Vega, D. Porras, and J. Ignacio Cirac, New J. Phys. [**13**]{}, 023024 (2011)
[^1]: Except for the set of points where $\cos\left( Jt\right)=\pm 1$, which is of measure zero. So we neglect it for the discussion of non-Markovianity.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Research in dolphin communication and cognition requires detailed inspection of audible dolphin signals. The manual analysis of these signals is cumbersome and time-consuming. We seek to automate parts of the analysis using modern deep learning methods. We propose to learn an autoencoder constructed from convolutional and recurrent layers trained in an unsupervised fashion. The resulting model embeds patterns in audible dolphin communication. In several experiments, we show that the embeddings can be used for clustering as well as signal detection and signal type classification.'
author:
-
-
-
title: An Auto Encoder For Audio Dolphin Communication
---
=1
Bio Acoustics, Dolphin Communication, Neural Network, Deep Learning
Introduction {#introduction .unnumbered}
============
Audible dolphin signals provide insight into dolphin cognition and social structure. Marine mammalogists collect large datasets of underwater recordings when encountering dolphins in the wild. For example, for 29 summers, each with 100 field days, researchers of the Wild Dolphin Project have collected audio and video data while observing wild Atlantic spotted dolphins (*Stenella frontalis*) underwater, in the Bahamas. The analysis of this data involves annotating the videos with observed dolphin behavior as well as dolphin names and audible signal type categories. In order to understand dolphin communication, researchers desire to correlate patterns in the audio with observed behavior. However, finding patterns in audible communication manually involves intensive measurements and comparisons across multiple spectrograms of the field recordings. Every hour of field recordings requires ten hours of manual analysis. We seek to automate several parts of the analysis: signal detection, sound type classification, and pattern identification. In the signal detection step, we locate dolphin signals temporally in the field recordings. The sound type classification step is needed to automatically determine the type of dolphin signal. There are several sound types in dolphin communication. Three prominent types are whistles, burst pulses and echolocation (see Fig. \[fig:examples\]) [@b13]. These types are often indicative of dolphin behavior. For example, dolphins use signature whistles to name each other while echolocation is often used during foraging.
![Example spectrogram of several dolphin signals. Top: a whistle, middle: a series of echolocation clicks, bottom: a burst pulse.[]{data-label="fig:examples"}](images/examples_labeled.jpg){width="50.00000%"}
Signal detection and signal type classification can help researchers to browse the data more efficiently. However, the ultimate goal is to automatically identify patterns in audible dolphin communication. A model of the patterns in dolphin signals needs to be robust to frequency shifting and time warping. One source of these variations is that dolphins are known to shift their communication into higher frequency bands depending on the noise floor. Other variations occur due to the angle of recording and the dolphins’ sound production itself, among others. Another challenge is that there is no fully annotated large dataset that could enable us to train a deep neural network.
We developed a deep autoencoder constructed from convolutional and recurrent layers. Through several experiments we show that we can indeed train a model in an unsupervised manner that enables us to find patterns in dolphin communication and can be used in a transfer learning setup.
Our contributions are
- A deep neural network for audible dolphin signals trained in an unsupervised manner
- A clustering experiment using embeddings from our architecture
- A dolphin signal detection experiment
- A dolphin type classification experiment
- Clustering all non-silent regions of a complete year of audio recordings.
Related Work {#related-work .unnumbered}
============
Marine mammalogists use interactive tools for the manual analysis of animal communication recordings. For example, Cornell’s RAVEN [@b22] allows researchers to annotate animal audio recordings in spectrogram form. Furthermore, Raven includes algorithms for animal signal detection in larger recordings. Noldus Observer enables users to annotate audio and video data together. Observer is primarily used to annotate annotate animal behavior data manually and does not offer automatic analysis capabilities. The automatic analysis of dolphin communication can increase the research speed of marine mammalogists. Previous models required several separate machine learning algorithms. For example, one approach involves learning a feature space that is invariant to signal shifts in frequency using convolutional k-means [@b3]. The next step is to cluster all examples using the dynamic time warping distance in order to account for temporal warpings of the signal. The last step is to learn a probabilistic model using a mixture of hidden Markov models. In order to build a classifier on top of the unsupervised results, the algorithm uses a bag-of-words approach on the cluster IDs [@b1; @b2; @b3]. One drawback is that these methods have to be tuned independently, which requires significant manual labor.
Previously, researchers have used the above methods in isolation. For example, Lampert and O‘Keefe detect dolphin whistles in spectrograms under various distortions using hidden Markov models [@b8]. The dynamic time warping distance is also used to measure similarity between manually extracted contours from spectrograms of dolphin whistles [@b9]. Other models for dolphin communication classification and clustering use neural networks [@b10].
There are also several specialized models for dolphin whistles. Two closely related methods are a Viterbi algorithm based pitch tracker [@b18] and a Kalman filtering approach [@b20]. Other approaches include a frame-based Bayesian approach [@b19] and a pitch detection algorithm designed for human telephone speech that is also capable of extracting whale vocalizations [@b21]. All of these approaches seek to extract the contour of a dolphin whistle. While these methods are very effective for whistle based communication, we wish to model a wider variety of dolphin signals and underwater artifacts.
In the proposed architecture we want to enable an end-to-end deep learning approach to the feature extraction step that handles frequency shifts and temporal warps in one model. Furthermore, each pattern should be encoded by the model into a single vector. Our model is inspired by the machine translation community. Specifically, we are inspired by the encoder-decoder model [@b6].
In a survey [@b24] of neural networks as models for animal communication, the authors suggest to use a sequence to sequence model on top of convolutional neural networks and show how these models can be used for visualization. We will use a similar model and study its performance on our dolphin communication dataset.
The encoder in these models encodes a sequence of words into a single vector using a many-to-one recurrent neural network. The decoder creates the translation from the embedding vector using a one-to-many recurrent neural network. Instead of words, we pass the encoder the output of a convolutional layer, followed by max-pooling. The decoder’s output is not the result itself, but the resulting sequence is passed to a deconvolution layer, similar to a convolutional autoencoder [@b7]. Instead of training the model for translation, we train the encoder and decoder to reconstruct a spectrogram window of dolphin communication. In the following section, we describe our model in more detail.
Proposed Architecture {#proposed-architecture .unnumbered}
=====================
Our goal is to learn a feature space that maps small spectrogram windows ($128$ spectrogram frames or $3/4$ seconds) into a single embedding vector. We propose an autoencoder to account for signal variations in time and frequency. The first layers of our encoder seek to achieve invariance to signal shifts in frequency by using a convolutional layer followed by a pooling operation. The recurrent layers compensate for time warping effects in the dolphin signals. After encoding a window of dolphin communication into a single vector, the decoder reconstructs the input window. Therefore, the encoding process is reversed. In other words, we reconstruct a sequence with the length of the input sequence from the embedding vector using several recurrent layers, and the signal is reconstructed by a series of deconvolutions.
Our architecture is shown in Fig. \[fig:architecure\], and our Keras implementation can be found at https://github.com/dkohlsdorf/wdp-ds/tree/v4.0/ml\_pipeline
Architecture Details {#architecture-details .unnumbered}
--------------------
We extract sliding windows
$$x=\{x_1 ... x_T\}, x_i \in \mathbb{R}^{F}$$
from a spectrogram with $F$ frequency bins. Each spectrogram window is convolved by $256$ filters resulting in a novel sequence:
$$\hat{x} = \{\hat{x_1} ... \hat{x_T}\}, \hat{x_i} \in \mathbb{R}^{F x 256}$$.
Each sample in the new sequence $\hat{x}$ has $256$ channels per frequency bin. Each filter spans $0.02$ seconds and $680$ Hz. In the next step, we apply max pooling. For every channel in a sample, we pool across all frequency bins. In other words, for each spectrogram frame and filter, we select the maximum response across all frequencies. The result is a new sequence:
$$pooled(\hat{x}) = \{pooled(\hat{x_1}) ... pooled(\hat{x_T})\}, pooled(\hat{x_i}) \in \mathbb{R}^{256}$$
with the same length as the spectrogram and one dimension per filter. In this way, we account for frequency shifts since we discard the frequency of the maximum response of a filter.
![The autoencoding architecture. Left: the encoder constructed from convolutional and recurrent layers. Right: The decoder constructed from deconvolutional and recurrent layers. []{data-label="fig:architecure"}](images/architecture.jpg){width="50.00000%"}
In order to account for temporal warping effects, we use the filter response sequence as input to several layers of recurrent units. The first layer is a many-to-many bidirectional long short term memory (LSTM) cell [@b12; @b15]. The second layer is a many-to-one LSTM. The resulting vector is our embedding:
$$e = LSTM(pooled(\hat{x})), e \in R^{128}$$
Dolphin signals can develop in complex ways over time. Therefore, we decided to use LSTM cells in our recurrent neural network for their ability to hold information longer. The first recurrent layer returns a sequence of the same length as the original sequence (many-to-many) serving as the input to the final embedding layer. Furthermore, the first recurrent layer is a bidirectional LSTM enabling the outputs to depend on information from the future and the past. The embedding layer is a simple LSTM with one output (many-to-one).
The decoder aims to reconstruct the input spectrogram from the encoding by reversing the encoding process. First, we build a new sequence from the embedding vector using a one-to-many LSTM which is followed by a bi-directional LSTM. The output is a sequence of the same length as the spectrogram:
$$x'= \{x'_1 ... x'_T\}, x'_i \in R^{F}$$.
The hope is that the embedding vector holds enough information to construct a whole sequence resembling the spectrogram input from it. We then apply a convolution with 256 filters:
$$\hat{x'}= \{\hat{x'_1} ... \hat{x'_T}\}, \hat{x'_i} \in R^{F x 256}$$.
followed by a 1x1 deconvolution which creates the final spectrogram.
Using the convolutions in the decoder we upsample the number of channels to model the decoder as a reverse of the encoder. In order to reshape the final sequence to the same shape as the input spectrogram, we apply a 1x1 deconvolution layer with one kernel. This technique reduces the shape of each sample from $R^{F x 256}$ to $R^{F}$.
Preprocessing And Training {#preprocessing-and-training .unnumbered}
--------------------------
Before the training, we compute the spectrogram using a $0.01$ second window with a $0.005$ second skip. Then, we apply a Hanning window before computing the discrete Fourier transform. We normalize each of the spectrogram windows ($3/4$ seconds) to its standard score. Therefore, we compute the mean and standard deviation for each spectrogram frame and then subtract the mean of the frame and divide it by its standard deviation.
During training, we use a batch size of $50$ spectrogram windows and train for $128$ epochs. We optimize the model by minimizing the mean square error between the input spectrogram and the decoder’s reconstruction using the ADAM optimizer [@b16].
Transfer Learning {#transfer-learning .unnumbered}
-----------------
The autoencoder can be trained in an unsupervised manner. Collecting a large amount of unlabeled data is easy and can be used to learn a feature space embedding appropriate for dolphin communication. Once the model is trained, we can use a smaller labeled dataset to construct a classifier on top of the embedding.
In our experiments, we will show how to use the embedder to distinguish between spectrogram windows with dolphin communication and water noise and how to use the embedder to distinguish between different dolphin communication types.
When building a classifier we simply add three dense layers on top of the encoder’s output and freeze all other layers’ weights except for the last LSTM. The first two dense layers’ activations are rectified linear units and the last layer’s activation is either a sigmoid for binary classification problems or a softmax activation for multiclass problems. By retraining on a labeled dataset we can easily construct a classifier.
Experiments {#experiments .unnumbered}
===========
With the following experiments, we will highlight several aspects of our model. We run a clustering experiment to show that similar patterns are embedded close to each other. In two experiments using the same model, we show that the encoder can be adopted to tasks such as dolphin signal detection and dolphin type classification.
Datasets {#datasets .unnumbered}
--------
In our experiments we use data from in-water field recordings collected by the Wild Dolphin Project. The audio data is extracted from video files from underwater cameras filming Atlantic spotted dolphins.
For our experiments we collect four datasets:
- An unlabeled dataset of 24 minutes collected form several field recordings from 2008, 2010 and 2012
- A small signal detection dataset containing 33 seconds of signal and 83 seconds of noise from field recordings in 2011
- A full year of field recordings consisting of ̃16 hours collected in 2011.
- A small labeled signal-type classification dataset
The classification dataset is labeled for dolphin signals containing:
- Noise: basic water noise, 82 seconds
- Echolocation: echolocation of dolphins, 118 seconds
- Burst Pulses: dense click packages, 71 seconds
- Whistles: the dolphins’ whistles, 64 seconds
The last dataset is a full year of field recordings. It consists of $\approx 16$ hours collected in 2011. All audio is recorded with a sample rate of 44100Hz.
Autoencoding {#autoencoding .unnumbered}
------------
In the first experiment, we train our autoencoder on the $22$ minute long unsupervised dataset. In order to inspect the learned model, we visualize the first layer’s convolutional kernels. As one can see in Fig. \[fig:embedder\_filters\] the result is filters representing up and down sweep in whistles (single line) as well as several representing burst pulses (multiple lines).
![A subset of the first layer convolutional kernels. Kernels that show multiple lines will be able to detect burst pulse signals while kernels with a single line will detect dolphin whistles. Vertical lines in a kernel indicate the presence of echolocation clicks. []{data-label="fig:embedder_filters"}](images/filters_zoomed.jpg){width="50.00000%"}
In another test, we visually inspect the reconstructions of several windows using the autoencoder. As one can see in Fig. \[fig:ae\_predict\] the reconstructions are indeed recognizable as dolphin signals. Clicks and whistles are especially visible, while burst pulses are often smeared.
![Some reconstructions of dolphin signals. One can clearly see whistles and echolocation sounds. The reconstructed whistles include up and down sweeps as well as turning points. The echolocation reconstructions are visible as one or more vertical lines. Most of the burst pulse sounds are not reconstructed sharply since the stereotypical harmonics are not present. []{data-label="fig:ae_predict"}](images/reconstructions.jpg){width="50.00000%"}
We then embed all of the training data and cluster the resulting vectors using k-means. We use 100 clusters and initialize the clustering using k-means++ [@b17]. We also restrict k-means to 1024 iterations. In order to visualize the clustering, we project the embeddings into 2D using T-SNE [@b11]. The results are plotted in Figure \[fig:train\_embedding\]. A magnified version is shown in Figure \[fig:train\_embedding\_zoom\].
![Embedding of the spectrogram windows from the train set. The position is determined by the T-SNE projection of the embeddings. With colors indicating the cluster, in the background we plotted the complete T-SNE map and in the foreground, we zoomed into the four corners. Best viewed in color. For details please enlarge.[]{data-label="fig:train_embedding"}](images/embedding_train.jpg){width="50.00000%"}
![Elarged version of the training set embedding. Signals densely cluster by type and also shape[]{data-label="fig:train_embedding_zoom"}](images/embedding_train_zoom.jpg){width="50.00000%"}
Signal Detection {#signal-detection .unnumbered}
----------------
In this experiment, we desire to show that the basic model can be adapted to the signal detection task. Therefore, we split the signal detection dataset into 60% training data and 40% testing data with each set containing data from all years. We employ a standard neural network for binary classification. In this experiment, we add three dense layers on top of the encoder’s output. First, we batch normalize the output of the encoder and then add two dense layers (64 neurons and 32 neurons). Before adding the classification layer we add one dropout layer with a dropout rate of 0.5. The last layer has one output neuron with a sigmoid activation. During training, we use a batch size of 10 instances and train for 25 epochs using an ADAM optimizer. The results are shown in Table I. As one can see, we achieve an accuracy of 96% on the test set. In the experiment, we do not fix the layers of the encoder and initialize the weights with the encoder from the unsupervised experiment.
truth / prediction dolphin noise
-------------------- --------- -------
dolphin 83 10
noise 5 347
: The confusion matrix for the signal detection experiment.
\[tab:signaldetect\]
Signal Type Classification {#signal-type-classification .unnumbered}
--------------------------
In another experiment, we adapt the model to perform dolphin signal type classification. We classify the dolphin signals into the four categories contained in the dataset. Again we use a 60 / 40 split from all years, and we use the same architecture and training as in the signal detection experiment, except we adjust the last layer to have four neurons with a softmax activation. As can be seen in Table II, we achieve an 85% accuracy in this experiment.
truth / prediction noise echo burst whistle
-------------------- ------- ------ ------- ---------
noise 291 27 15 1
echo 35 434 9 7
burst 38 30 207 8
whistle 8 12 5 181
: The confusion matrix for the signal classification experiment.[]{data-label="tab:signaltype"}
Mining a year of data {#mining-a-year-of-data .unnumbered}
---------------------
In our final experiment, we use the trained encoder and the signal detector to extract and cluster patterns from the whole year of data from 2011. We extract windows classified as dolphin communication using the signal detector. We then embed the windows using the encoder and cluster the resulting embeddings into $100$ clusters using the same method as in the previous clustering experiment. We visualize the result in the same way. After clustering, we estimate the silhouette coefficient for each window. The silhouette coefficient measures how windows are clustered with similar samples [@b23]. We then filter all samples with a coefficient lower than the medium coefficient across all samples. The results are shown in Fig. \[fig:2011embedding\]. In this case, we only plot 25% of the instances for better visibility. A zoomed in version is shown in Figure \[fig:2011embedding\_zoom\].
![Embedding of the spectrogram windows from the test set. The position is determined by the T-SNE projection of the embeddings. The colors indicate the cluster, in the background we plotted the complete T-SNE map and in the foreground we zoomed into the four corners. Best viewed in color. For details please enlarge.[]{data-label="fig:2011embedding"}](images/embedding_test.jpg){width="50.00000%"}
![Elarged version of the training set embedding. Signals densely cluster by type and also shape.[]{data-label="fig:2011embedding_zoom"}](images/embedding_test_zoom.jpg){width="50.00000%"}
We inspect all clusters visually as well. Therefore, we export each cluster into a single wav file. The audio in each window is concatenated with a short gap of silence (see Fig. \[fig:warping\]). We then inspect each spectrogram and note how many clusters have the same type and how many silence clusters still made it through the silence detector. We found that 14 clusters are still mostly noise resulting in an $86\%$ accuracy for the silence detector. We also found four clusters of mixed type (mix of whistles, burst sounds and echolocation).
![ A cluster showing spectrograms of a whistle warped in time and shifted in frequency. Especially in the middle row, we see several distortions of the signal including overlapping signals. []{data-label="fig:warping"}](images/warped.jpg){width="50.00000%"}
In the example in Figure \[fig:warping\] we see three clusters with slight time warps and shifts in frequency.
Discussion {#discussion .unnumbered}
==========
In our experiments, we showed that our model is indeed able to learn an embedding for audible dolphin signals in an unsupervised manner. We showed that the embedding can be trained using an autoencoder that reconstructs dolphin signals. Fig. \[fig:embedder\_filters\] shows the first layer’s convolutional kernels. Nearly all of these kernels look like spectrogram patches from dolphin communication which can be interpreted as evidence for successful training. In the same way, Fig. \[fig:ae\_predict\] presents successful reconstructions of the autoencoder as further evidence. While we are able to successfully reconstruct whistles and echolocation sounds, the burst pulses blur more. Furthermore, by visual inspection, the clusters in the 2D projection seem tight, meaning that the embedding is able to model patterns in dolphin communication. We also showed that the embedder can be used for transfer learning. The signal detector and the type classifier achieve high accuracy on their test sets despite the significantly smaller training data. Because labeling of dolphin signals is cumbersome for marine mammalogists it is expected that more directed efforts towards dolphin communication in specific behavioral contexts will produce smaller datasets. However, our transfer learning results indicate that the model can be adjusted with little data to other tasks. In the final experiment, we ran the signal detector and the embedder in a more realistic scenario. We showed that the signal detector performs well across data recorded throughout a whole year. Furthermore, we showed that the embedding of the detected regions produces clean clusters. In total, we think that we trained a successful feature extractor which will be the basis of our future research.
Future Work {#future-work .unnumbered}
===========
Our experiments show that the model is indeed an effective model of short term fixed-size windows of dolphin communication. In the future, we plan to investigate several methods for sequence analysis, in contrast to the fixed sized windows utilized in our approach. Sliding the encoder across the spectrogram together with the silence detector will create variable length embedding sequences of dolphin communication bounded by silence. We aim to replace the k-means clustering of single embedding vectors with agglomerative clustering of sequences of embedding vectors using the dynamic time warping distance. Furthermore, we aim to create multiple sequence alignments of dolphin communication. We hope that the visualization of aligned spectrograms will ease the comparison of longer sequences of dolphin communication. Finally, we seek to answer questions about the structure of dolphin communication. One question marine mammalogists have is whether dolphin communication displays a similar structure to human communication. One idea is to use recursive neural networks [@b5] on top of sequences of embedding vectors in order to find structural patterns.
Conclusion {#conclusion .unnumbered}
==========
We proposed an autoencoder constructed from convolutional and recurrent layers in order to construct an embedding of short windows of audible dolphin communication. Our architecture is inspired by encoder-decoder models in natural language processing. The encoder transforms short spectrogram windows into a single encoding vector. The decoder reconstructs the complete spectrogram from just the embedding vector. In a series of experiments, we showed the effectiveness of the embedding. First, we visualized the first layer’s convolutions and found that the convolutional kernels seem to pick up on dolphin signals. When plotting the decoder’s reconstructions we clearly saw that we are able to encode enough information in the embedding vector to reconstruct the signals. When clustering the embedded spectrograms we saw that the clusters seem distinct as well. We also showed the model’s performance on two transfer learning tasks. In a signal detection experiment and a type classification experiment, the retrained model showed high accuracies. When running the trained models on a whole year of data we found good clustering and signal detection performance.
[00]{} Kohlsdorf, “Data Mining In Large Audio Collections Of Dolphin Signals,” PhD Thesis, Georgia Institute of Technology, 2015. Kohlsdorf, Herzing and Starner, “Methods for Discovering Models of Behavior: A Case Study with Wild Atlantic Spotted Dolphins,” Animal Behavior and Cognition, 2016. Kohlsdorf, Herzing and Starner, “Feature Learning and Automatic Segmentation for Dolphin Communication Analysis,” Interspeech 16, 2016. Coates, Lee and Ng, “An Analysis of Single-Layer Networks in Unsupervised Feature Learning,” International Conference on Artificial Intelligence and Statistics, 2011. Socher, Lin, Ng and Manning, “Parsing Natural Scenes and Natural Language with Recursive Neural Networks”, ICML, 2011. Cho, van Merrienboer, Gulcehre, Bahdanau, Bougares, Schwenk and Bengio, “Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation,” CoRR, 2014 Ranzato, Huang, Boureau and LeCun, “Unsupervised Learning of Invariant Feature Hierarchies with Applications to Object Recognition,” CVPR, 2007 Lampert and O’Keefe, “A survey of spectrogram track detection algorithms,” Applied Acoustics, vol. 71, no. 2, pp. 87 – 100, 2010. Kershenbaum, Sayigh and Janik, “The encoding of individual identity in dolphin signature whistles: How much information is needed?,” PLoS ONE, vol. 8, no. 10, 2013. Deecke, Ford, and Spong,“Quantifying complex patterns of bioacoustic variation: Use of a neural network to compare killer whale (Orcinus orca) dialects,” Journal of the Acoustical Society of America, 1999. van der Maaten and Hinton, “Visualizing High-Dimensional Data Using t-SNE,” Journal of Machine Learning Research, 2008. Hochreiter and Schmidhuber, “Long Short-Term Memory,” Neural Computation 9.8, 1997. Herzing, “Clicks, whistles and pulses: Passive and active signal use in dolphin communication,” Acta Astronautica, 2014. Schuster and Paliwal, “Bidirectional recurrent neural networks,” IEEE Transactions on Signal Processing, 1997. Kingma and Lei Ba, “ADAM: A method for stochastic optimization,” ICLR, 2015. Arthur and Vassilvitskii, “k-means++: The Advantages of Careful Seeding,” ACM-SIAM symposium on Discrete algorithms, 2007. Kohlsdorf, Mason, Herzing and Starner, “Probabilistic extraction and discovery of fundamental units in dolphin whistles,” ICASSP, 2014. Halkias and Ellis, “Call detection and extraction using Bayesian inference,” Applied Acoustics, 2006. Lampert and O’Keefe, “An active contour algorithm for spectrogram track detection,” Pattern Recognition Letters, 2010. Shapiro and Wang, “A versatile pitch tracking algorithm: From human speech to killer whale vocalizations,” The Journal of the Acoustical Society of America, 2009 Cornell, “Lab of Ornithology’s Raven: Interactive sound analysis software,” Bioacoustics Research Program, 2014 Rousseeuw, “Silhouettes: a Graphical Aid to the Interpretation and Validation of Cluster Analysis,” Computational and Applied Mathematics, 1987 Sainburg, Thielk and Gentner: “Latent Space Visualization, Characterization, and Generation of Diverse Vocal Communication Signals”, Preprint, available at https://www.biorxiv.org/content/10.1101/870311v1
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We report the use of time- and angle-resolved two-photon photoemission to map the bound, unoccupied electronic structure of the weakly coupled graphene/Ir(111) system. The energy, dispersion, and lifetime of the lowest three image-potential states are measured. In addition, the weak interaction between Ir and graphene permits observation of resonant transitions from an unquenched Shockley-type surface state of the Ir substrate to graphene/Ir image-potential states. The image-potential-state lifetimes are comparable to those of mid-gap clean metal surfaces. Evidence of localization of the excited electrons on single-atom-layer graphene islands is provided by coverage-dependent measurements.'
author:
- 'D. Niesner'
- 'Th. Fauster'
- 'J. I. Dadap'
- 'N. Zaki'
- 'K. R. Knox'
- 'P.-C. Yeh'
- 'R. Bhandari'
- 'R. M. Osgood'
- 'M. Petrović'
- 'M. Kralj'
title: Trapping Surface Electrons on Graphene Layers and Islands
---
Graphene on metal surfaces are a materials system of enormous fundamental and applied interest. The graphene/metal interface is encountered in the rapidly expanding technological system of CVD graphene on Cu foil, in the structurally precise monolayer epitaxial systems of graphene on single-crystal Ru, Ir, or Ni, and finally in the metal contacts of graphene field-effect transistors or other devices. Questions then arise on the electronic structure of graphene on metal surfaces and in fact several recent studies have addressed questions such as the role of lattice mismatch on band structures. Most studies of the electronic structure of graphene have focused on the band structure in the vicinity of its K point, near the Fermi edge. Further there has been a paucity of measurements about its unoccupied electronic structure and the dynamics of strongly excited electrons. Image-potential states offer one important approach to probe the excited state manifold and are known to vary with interfacial quality, dielectric properties, and electronic structure. In graphene the large band gap at the $\Gamma$ point results in Bragg reflection from the crystal within a certain range of energy and momentum. In fact a recent theoretical study has shown the existence of a dual Rydberg-like series of even and odd symmetry image-potential states in a single free-standing sheet of graphene [@SIL2010]. Image-potential states on graphene may experience different dynamic constraints. For example, the different phase space for decay in two dimensions compared to three dimensions may affect the lifetimes for electrons trapped in image-potential states on graphene. In fact, more generally the response of the image electron to the composite dielectric/metal systems is itself of basic physics interest.
![(Color online). Arrows indicate 2PPE transitions between surface and image-potential states. The experimental results (dots) are compared to calculations (lines). The projected bulk-band structure of Ir(111) along the $\Gamma$K direction is shaded according to the total and $sp$-density of states (DOS) at the right and left, respectively.[]{data-label="bs"}](irkpxf.ps){width=".67\columnwidth"}
In this Letter, we investigate the uncharted region of the bound, unoccupied electronic structure of epitaxial graphene grown on Ir(111) in the vicinity of the graphene $\Gamma$ point; our measurements are made via the image-potential states using angle- and time-resolved two-photon photoemission (2PPE) as indicated by arrows in Fig. \[bs\]. This system was chosen for several reasons: First, because of the weak coupling in the graphene/Ir(111) system, the electronic structure of the graphene overlayer is nearly intact, with sharp Dirac dispersion characteristics [@PLE2009]. In addition, the moiré corrugation of the epitaxial graphene on Ir(111) has been found to be only $0.35\pm0.10$ [Å]{} based on atomic force microscopy measurements [@SUN2011] indicating a smooth epitaxial graphene surface. Second, the molecular-based growth is well characterized and saturates at precisely one monolayer (ML) of epitaxial graphene [@GAS2009]. Our results show that image-potential states may be excited from the Ir/graphene interfacial region and have binding energies and lifetimes comparable to those of mid-gap clean metal surfaces. In addition, spectral measurements of binding energy versus coverage show clearly that at low graphene coverage, image-potential electrons are trapped on graphene islands by surface work function differences between the metal and graphene regions, an observation of high importance for understanding of transport at graphene-metal interfaces [@XIA2011].
Our choice of two-photon photoemission is the result of its high temporal and energy resolution. Other experimental observations of image-potential states have used scanning tunneling spectroscopy, i. e., graphene on SiC [@BOS2010] and on Ru(0001) [@BOR2010]. This technique, however, measures the image-potential series in the presence of strongly distorting electric field between tip and sample and without time-resolved possibilities.
The experiments were conducted using monochromatic and bichromatic 2PPE, and angle-resolved photoemission (ARPES). Details of the monochromatic 2PPE setup at Columbia which was used in the photon energy range of $3.8<h\nu<4.9$ eV are given in Ref. [@HAO2010]. Bichromatic and time-resolved 2PPE measurements were performed in Erlangen using pump-probe methods with the third harmonic (UV) and the fundamental (IR, $1.51<h\nu<1.62$ eV) as described in Ref. [@Boger05njp]. Additionally, occupied-state ARPES measurements were performed at APE (ELETTRA) using a photon energy of 55 eV with an energy resolution of 20 meV. The resolution of the 2PPE experiments was 40 meV. The base pressure in all three ultrahigh vacuum (UHV) systems was better than $1\times10^{-8}$ Pa. All measurements used $p$-polarized beams.
Graphene was prepared by cycles of temperature programmed growth, TPG (room temperature ethene exposure $6\times10^{-6}$ Pa for 60 s and flashed to $\approx1450$ K), followed by a chemical vapor deposition run ($6\times10^{-6}$ Pa of ethene for 300 s at 1150 K), to form exactly one graphene monolayer [@GAS2009]. Growth was monitored by low-energy electron diffraction (LEED) after each cycle, which showed the development of the characteristic moiré pattern of uniformly oriented graphene [@HAT2011], as graphene coverage varied from 0 to 1 ML; LEED patterns (not shown) revealed these patterns clearly.
![(Color online). a) Intensity map of the 2PPE signal recorded with photon energy $h\nu=1.59$ eV for 1 ML graphene on Ir(111). Points represent the intensity of the lowest $n=1$ band. b) ARPES map showing initial states for $h\nu=55$ eV.[]{data-label="maps"}](maps.eps){width="\columnwidth"}
Figure \[maps\]a shows the measured 2PPE intensity obtained at 1 ML and for $h\nu=1.59$ eV along the $\Gamma$K direction. Three unoccupied bands are observed. The pumping process could be deduced from its photon-energy dependence, thus in bichromatic case, all peak positions shifted linearly with IR photon energy, indicating that the process involves pumping by a UV photon and photoemission by an IR photon [@HOF1997]. All 2PPE features vanished when the IR beam was switched to $s$-polarization, indicating the expected symmetry for image-potential states. The effective masses of all three states are $0.9\pm0.1\,m_{e}$. The binding energies of the three states with respect to the vacuum level are given in Table \[tab1\]. The measured energies and effective masses are close to the free electron mass and fit well to a Rydberg-like-series of image-potential states with a nonvanishing quantum defect [@EP1978].
Figure \[maps\]a shows that the $n=1$ band is most intense for parallel momenta $k_\|$ between 0.08 and 0.17 Å^-1^ [\[]{}cf. points in Fig. \[maps\]a[\]]{}, with the intensity typically decreasing monotonically with increasing $k_\|$ [@GUD2007]. Direct transitions from initial surface bands can lead to intensity resonances [@HAO2010]. In order to identify possible initial states for 2PPE, we have performed ARPES measurements of graphene on Ir(111). The ARPES data in Fig. \[maps\]b show two parabolic-like dispersions with a downward curvature. The two branches are shifted from $k_\|=0$ by $\pm0.033\pm0.001$ Å^-1^ and have a maximum energy of $-0.19\pm0.01$ eV. Similar results were also obtained with the fourth harmonic (6.2 eV) in the 2PPE setup. Rashba-type splittings of similar magnitude are found in other systems, e.g., a Bi/Ag(111) surface alloy [@AST2007]. These bands are also observed on clean Ir(111) [@VEEN1980; @PLE2010] indicating that this surface feature is inherent to the clean metal surface. The surface-state energy reported for clean and graphene-covered surfaces differs by about 0.2 eV, an effect, which is consistent with a charge transfer between substrate and overlayer and which can shift graphene [@PLE2010] or iridium states [@KRA2011; @SUB2011].
The initial band dispersion is plotted together with the measured dispersion of the image-potential states in Fig. \[bs\] (blue dots). The arrows connecting initial states to the $n=1$ image-potential band are at slightly larger $k_\|$ values than the enhanced intensity in Fig. \[maps\]a. In the absence of resonances the 2PPE intensity along image-potential bands decreases continuously with increasing parallel momentum [@GUD2007]. In the present case due to the finite energy and angle resolution the intensity maxima are shifted to lower $k_\|$ values compared to the position found in the dispersion analysis. The additional resonance into the $n=2$ band (see Fig. \[bs\]) can be inferred from the similar intensity as for the $n=1$ state at $k_\|=0$ in Fig. \[maps\]a and is confirmed by photon-energy-dependent data presented in the supplemental material [@suppmat].
$n$ $E^\mathrm{exp}_n$ (eV) $E^\mathrm{calc}_n$ (eV) $\tau$ (fs)
----- ------------------------- -------------------------- -------------
1 $0.83\pm0.02$ 0.59 $35\pm3$
2 $0.19\pm0.02$ 0.18 $114\pm6$
3 $0.09\pm0.02$ 0.08 $270\pm12$
: Experimental and calculated binding energies and lifetimes for image-potential states on graphene/Ir(111).[]{data-label="tab1"}
In order to understand the character of the initial state, we calculated the projected bulk-band structure of Ir(111) using a non-relativistic parameterized tight-binding scheme [@PAP1986]. Figure \[bs\] shows this projected structure along the $\Gamma$K direction, at the right. The shading represents the one-dimensional density of states (1D-DOS). The left part of Fig. \[bs\] shows the 1D-DOS of bands according to their $sp$-character. For $k_\|=0$, the lower edge of the $sp$-band gap is at $-0.7$ eV, which corresponds to the $L_{2'}$ point. The band edge of the total projected bulk-band structure disperses upward from the $L_{3}$ point around $+0.8$ eV and picks up $sp$-contributions. On the other hand the lower $sp$-band edge shows a downward dispersion. The energy of the Shockley surface state was calculated using the $sp$-band edges within a scattering model [@FAU1994]. The calculations used the experimental work function of 4.65 eV for graphene on Ir(111). The calculated bands were shifted by $\pm0.033$ Å^-1^ to account for the experimentally observed Rashba splitting and are drawn as green lines in Fig. \[bs\] in the region below the Fermi energy. The experimentally extracted dispersion shown by dots agrees well with the calculation. The Shockley-type surface state is apparently not quenched by the graphene layer at a distance of 3.4 Å [@BUS2011], because its probability density is concentrated at the Ir(111) surface.
The scattering model was also used to calculate the energies of the image-potential-band series [@FAU1994]. The calculated binding energies, given in Table \[tab1\], are approximately those expected for states located near midgap (see Fig. \[bs\]). However, the calculated $n=1$ binding energy is significantly smaller than the experimental value. This discrepancy is due to the fact that the scattering model calculation neglects the round-trip phase shift $2\phi_{gr}$ of the graphene layer. Using the expressions for the phase shift at the substrate and the image-potential barrier [@SMI1985] we obtain $\phi_{C}=0.63\pi$ and $\phi_{B}=1.02\pi$, respectively. The total phase shift for the $n=1$ state is $2\pi$, from which we obtain $\phi_{gr}=0.18\pi$. Note that such a small phase shift leads to a significant change in binding energy from 0.59 eV to 0.83 eV (see Table \[tab1\]).
![a) Time-resolved measurements of the image-potential-state series at $k_\|=0$. b) Lifetimes as function of binding energy compared to various power-law dependencies (see text).[]{data-label="tr"}](TR_spectran.eps){width="\columnwidth"}
The time-resolved spectra of the image-potential states were also measured and are shown in Fig. \[tr\]a. As summarized in Table \[tab1\], lifetimes of tens to hundreds of femtoseconds are obtained. These are comparable to values obtained for Cu(100) with a similar midgap image-potential state position and hence bulk evanescent decay length in the metal crystal [@HOF1997]. Note, as an aside, that the curve measured at the energy of the $n=4$ image-potential state in Fig. \[tr\]a shows weak quantum beats [@HOF1997] for delay times $<300$ fs. The data in Table \[tab1\] show that lifetimes vary with binding energy approximately $\propto
E^{-1}$ (solid line in Fig. \[tr\]b). The asymptotic, classical $\tau\propto E^{-3/2}$ behavior [@FAU2000] (dashed line in Fig. \[tr\]b) is not reached for $n<4$. Similar behavior has been found on copper surfaces [@ssr].
An important issue for carrier movement at graphene/metal interfaces is the degree of lateral confinement. This confinement can be examined at low graphene coverage, obtained via a small number of sequential TPG cycles. From previous studies, it is known that one TPG cycle covers a fraction of about 20% of uncovered Ir surface [@KRA2011]. After one TPG cycle the typical island size is $(35~\mathrm{nm})^{2}$ and after the second cycle of the order of $(100~\mathrm{nm})^{2}$ [@KRA2011].
As has been shown in earlier work [@FAU1993a], the average and local work function play an important role in interfacial electron localization. Thus the average work function $\Phi$ was measured via monochromatic 2PPE and the expression $\Phi=2h\nu-\Delta E$, i.e., where $h\nu$ is the photon energy and $\Delta E$ is the difference between the Fermi level cutoff and the low-energy cutoff. Figure \[cov\] displays the work function (open symbols) as a function of graphene coverage. The work function decreases approximately linearly from a value $5.79\pm0.10$ eV to $4.65\pm0.10$ eV from Ir(111) to 1 ML graphene. Reported values of the work function for Ir(111) are 5.76 and 5.79 eV [@Ir111WF]. The work function of the graphene-covered surface on Ir(111) is between the values for Pt(111) of 4.87 eV and free-standing graphene of 4.48 eV [@GIO2008], which is consistent with the weak bonding between the Ir(111) and the graphene overlayer and a $p$-doping of the graphene [@PLE2009]. The linear decrease of the work function is known for other systems and is due to the averaging over substrate and overlayer islands [@FAU1993a].
![Sample work function (open symbols) and image-potential states $n=1$, 2, and 3 binding energies (filled symbols) as a function of graphene coverage. Dashed line represents a linear fit for the work-function change.[]{data-label="cov"}](cov.eps){width=".7\columnwidth"}
Image-potential states were observed at all coverages reported here using 2PPE. However, for the clean surface or uncovered substrate areas the available photon energies were not sufficient to populate image-potential states due to the large work function of Ir(111). The image-potential-state energies, measured relative to the Fermi level, are shown in Fig. \[cov\]. The energies are generally constant over the coverage range from 0.2 to 1 ML with the intensity increasing monotonically with coverage. Note that the graphene Dirac cone at the K point has been clearly observed for more than three TPG cycles or 0.5 ML graphene coverage [@KRA2011]. The constant energy of the image-potential series as a function of coverage in Fig. \[cov\] is a direct result of the localization of the electrons on the graphene islands [@FAU1993a]. The electrons respond to the local work function if the average island dimensions are larger than the typical distance of the probability density maximum which is of the order of nanometers for the lowest $n$ image-potential states. Note that the localization on the graphene islands is facilitated by the large work function difference between the graphene layers and the Ir(111) substrate. For small graphene islands, an energy shift proportional to $d^{-2}$, where $d$ denotes the characteristic island size, is expected due to the lateral localization of the electron in a two-dimensional quantum well [@FAU1993b]. However, these shifts would be $<1$ meV for the island sizes expected for the current preparation conditions [@KRA2011].
In summary, we have observed and measured the properties of image-potential states on a graphene monolayer on Ir(111). The binding energy of the $n=1$ image-potential state is 40% larger than expected from the position of the graphene vacuum level relative to the Ir(111) band gap. There is no prominent indication of a second main series of image-potential states as predicted for free-standing graphene [@SIL2010]. Apparently, the underlying metal substrate breaks the mirror symmetry of the graphene layer and the state of odd symmetry shifts up in energy as has been calculated for graphene on Ru(0001) [@BOR2010]. In addition, the image-potential states can be excited efficiently from a downward dispersing Shockley surface state in the $sp$-band gap of the Ir(111) band structure indicating a sizable overlap of the wave functions of these states located at the substrate interface and graphene surface, respectively. The measured lifetimes of the image-potential states are comparable to similar clean metal surfaces. Recently, similar results have also been obtained for graphene on Pt(111) [@Pt111]. Apparently, the evanescent coupling of the image-potential-state wavefunctions to the underlying electronic states of the Ir(111) bulk and surface states is not altered by the graphene layer. Three-dimensional localization of electrons on graphene islands has been observed for submonolayer coverages obtained by individual TPG cycles. However, even for the smallest island size, no energy shift due to localization was observed within the experimental uncertainty. Further development is needed to prepare well-ordered graphene islands with controlled lateral extension. A different approach would be to exploit the moiré pattern on more corrugated graphene layers [@BOR2010].
The work at Columbia University was supported under the US Department of Energy Contract No. DE-FG 02-04-ER-46157. The Zagreb group acknowledges support by the MZOS through project No. 035-0352828-2840 and MZOS-NSF support through contract No. 1/2009. M. K. thanks the Alexander von Humboldt Foundation for a research fellowship. We acknowledge helpful discussions with Branko Gumhalter.
[10]{}
V. M. Silkin [*et al.*]{}, Phys. Rev. B [**80**]{}, 121408 (2009).
I. Pletikosić [*et al.*]{}, Phys. Rev. Lett. [**102**]{}, 056808 (2009).
Z. Sun [*et al.*]{}, Phys. Rev. B [**83**]{}, 081415 (2011).
R. van Gastel [*et al.*]{}, Appl. Phys. Lett. [**95**]{}, 121901 (2009).
F. Xia [*et al.*]{}, Nature Nanotechnol. [**6**]{}, 179 (2011).
S. Bose [*et al.*]{}, New J. Phys. [**12**]{}, 023028 (2010).
B. Borca [*et al.*]{}, Phys. Rev. Lett. [**105**]{}, 036804 (2010).
Z. Hao [*et al.*]{}, Phys. Rev. Lett. [**105**]{}, 017602 (2010).
K. Boger, [Th. Fauster]{}, and M. Weinelt, New J. Phys. [**7**]{}, 110 (2005).
H. Hattab [*et al.*]{}, Appl. Phys. Lett. [**98**]{}, 141903 (2011).
U. H[ö]{}fer [*et al.*]{}, Science [**277**]{}, 1480 (1997).
P. M. Echenique and J. B. Pendry, J. Phys. C [**11**]{}, 2065 (1978).
J. [G[ü]{}dde]{} [*et al.*]{}, Science [**318**]{}, 1287 (2007).
C. R. Ast [*et al.*]{}, Phys. Rev. Lett. [**98**]{}, 186807 (2007).
J. F. van der Veen, F. J. Himpsel, and D. E. Eastman, Phys. Rev. B [**22**]{}, 4226 (1980).
I. Pletikosić [*et al.*]{}, J. Phys.: Condens. Matter [**22**]{}, 135006 (2010).
M. Kralj [*et al.*]{}, Phys. Rev. B [**84**]{}, 075427 (2011).
D. [Subramaniam]{} [*et al.*]{}, ArXiv e-prints (2011).
See Supplemental Material at URL for data showing the resonant transition into th $n=2$ state.
D. A. Papaconstantopoulos, [*Handbook of the Band Structure of Elemental Solids*]{} (Plenum Press, New York, 1986).
, Appl. Phys. A [**59**]{}, 639 (1994).
C. Busse [*et al.*]{}, Phys. Rev. Lett. [**107**]{}, 036101 (2011).
N. V. Smith, Phys. Rev. B [**32**]{}, 3549 (1985).
, [Ch. Reu[ß]{}]{}, I. L. Shumay, and M. Weinelt, Chem. Phys. [ **251**]{}, 111 (2000).
P. M. Echenique [*et al.*]{}, Surf. Sci. Rep. [**52**]{}, 219 (2004).
R. Fischer [*et al.*]{}, Phys. Rev. Lett. [**70**]{}, 654 (1993), . Fischer, Ph. D. thesis, Universität München, 1993.
R. W. Strayer, W. Mackie, and L. W. Swanson, Surf. Sci. [**34**]{}, 225 (1973), . E. Nieuwenhuys, R. Bouwman, and W. M. H. Sachtler, Thin Solid Films [**21**]{}, 51 (1974).
G. Giovannetti [*et al.*]{}, Phys. Rev. Lett. [**101**]{}, 026803 (2008).
R. Fischer, [Th. Fauster]{}, and W. Steinmann, Phys. Rev. B [**48**]{}, 15496 (1993).
D. Nobis, D. Niesner, and [Th. Fauster]{} (unpublished).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We address the question: Why may reaction-diffusion equations with hysteretic nonlinearities become ill-posed and how to amend this? To do so, we discretize the spatial variable and obtain a lattice dynamical system with a hysteretic nonlinearity. We analyze a new mechanism that leads to appearance of a spatio-temporal pattern called [*rattling*]{}: the solution exhibits a propagation phenomenon different from the classical traveling wave, while the hysteretic nonlinearity, loosely speaking, takes a different value at every second spatial point, independently of the grid size. Such a dynamics indicates how one should redefine hysteresis to make the continuous problem well-posed and how the solution will then behave. In the present paper, we develop main tools for the analysis of the spatially discrete model and apply them to a prototype case. In particular, we prove that the propagation velocity is of order $a t^{-1/2}$ as $t\to\infty$ and explicitly find the rate $a$.'
author:
- 'Pavel Gurevich[^1], Sergey Tikhomirov[^2]'
title: 'Spatially discrete reaction-diffusion equations with discontinuous hysteresis'
---
intro.tex
sect2.tex
sect3.tex
sectAux.tex
sect5.tex
sect6.tex
sect7.tex
sectMainTheorem.tex
appendixC.tex
appendixD\_new.tex
Acknowledgement {#acknowledgement .unnumbered}
===============
The authors are grateful to Daria Neverova for her help in preparing the figures. The work of the first author was supported by the DFG Heisenberg Programme, DFG project SFB 910, and the Ministry of Education and Science of Russian Federation (agreement 02.a03.21.0008). The second author would like to thank JSC “Gazprom neft” and Contest “Young Russian Mathematics” for their attention to this work. bibl.tex
[^1]: Free University of Berlin, RUDN University, email: [email protected]
[^2]: Saint-Petersburg State Univeristy; email: [email protected]
| {
"pile_set_name": "ArXiv"
} |
-1.cm -1.5cm =16.cm =24.cm \#1 \#1[\#1]{} \#1 5774[7]{}
**THE “PISCO” SPECKLE CAMERA AT PIC DU MIDI OBSERVATORY[^1]**
**J.-L. Prieur$^*$, L. Koechlin$^*$, C. André$^*$, G. Gallou$^*$, C. Lucuix$^*$**
*$^*$Observatoire Midi-Pyrénées, 14, Avenue E. Belin, F-31400 Toulouse, France.*
[**Abstract:**]{}
We present a new speckle camera designed and built at Observatoire Midi-Pyrénées. This focal instrument has been used for two years with the 2-meter Bernard Lyot Telescope of Pic du Midi observatory. It can be set in various operating modes: full pupil imaging, masked-pupil imaging, spectroscopy, wave-front sensor and stellar coronagraphy, hence its name “PISCO” (“Pupil Interferometry Speckle COronagraph"). Restored images of double and triple stars have demonstrated its capabilities in providing close to diffraction limited images (0.06$\,$ in V). PISCO has been fully tested and is now ready to be used by the whole astronomical community. [**1. Introduction**]{}
10000 This speckle camera was designed and built between 1991 and 1994 by the “Aperture synthesis team” of Observatoire Midi Pyrénées (OMP), as a new focal instrument for the 2 meter Télescope Bernard Lyot (TBL) at Pic du Midi. The aim was to take advantage of the good seeing quality of that site.
The optical design was chosen such that a pupil plane would be accessible and pupil masks could be put into it, to allow for aperture synthesis experiments and simulate telescope arrays such as the ESO VLTI (European Southern Observatory Very Large Telescope in the Interferometric mode) or others. This speckle camera thus provides appropriate experimental tools for the investigation of image restoration techniques with the optical telescope interferometric networks that are currently being built and operated around the world (CHARA, COAST, GI3T, IOTA, NPOI, PTI, SUSI, etc, see [*f.i.*]{} [*Astronomical Interferometry,*]{} Proceedings of SPIE, Vol. 3350, Kona, 20-28/03/98)
In the following we present an overview of PISCO and its optical concept (§2). Then we introduce the various operative modes and illustrate them with observational results (§3). The detectors that have been used with PISCO are presented in §4. In §5, we discuss the current performances of the instrument in relation with the physical limitations of the speckle techniques and the image restoration methods. In §6 we briefly describe the on-going scientific programs which use PISCO, and conclude about the use of a speckle camera in the current context of new high angular resolution techniques.
[**2. The PISCO speckle camera**]{}
10000 Compared to other speckle cameras (Blazit [*et al.*]{}, 1977, Breckinridge [*et al.*]{} 1979, Strittmater, 1980, Beckers [*et al.*]{}, 1983, Foy, 1988a) our instrument presents the advantage of a versatility and a full remote control of all its operating modes. PISCO offers thus a wide range of possibilities with fast switching between modes (less than one minute), allowing an optimal use of the seeing conditions.
[*2.1. Description*]{}
100000 The general layout is shown in Fig. . The external mechanical structure is a rectangular box of 100$\times$40$\times$36 cm$^3$ which is mounted at the Cassegrain focus of the telescope (Fig. ).
The input image plane (I1) of the telescope is located 200 mm downstream from the front flange of PISCO (this value is easily adaptable for the 3.6 m CFH or ESO telescopes). The converging input beam is transformed into a parallel beam by the collimating lens (L2), then focused into the image plane (I2) by the lens (L3).
Lens (L4) magnifies this image and projects it to the detector faceplate. The focal length of (L4) can be selected with the wheel GR that bears a series of eyepieces and microscope objectives. A magnification of at least 20 mas/pixel is needed to obtain a good sampling for speckle observations at the TBL while a lower magnification is used for field acquisition.
Figure 1: Optical diagram of PISCO.
Filters (wheels FA and FB) allow selecting the desired wavelength range while neutral densities ( wheels DA and DB) are used to adjust the light level to the (generally poor) dynamic range of the photon-counting detectors (see §4). As will be explained in § 2.2, a set of Risley prisms correct for the atmospheric chromatic dispersion.
Figure 2: PISCO and photo-counting detector CP40 at the TBL.
When a field lens is selected (wheel CH), the pupil plane is located in the plane of the wheel MA, where pupil masks are available for coronagraphy (§3.3) or multi-aperture interferometry (§3.2). If low dispersion spectroscopy (§3.4) is wanted, a grism can be put into the parallel beam (wheel FA). Wavefront analysis is also possible with a Hartman sensor by selecting a microlens array in the wheel MA, and the pupil imaging mode in wheel GR.
PISCO can then be seen as an optical bench on which mounts and wheels can freely move and rotate. This concept allows a great flexibility for future instrumental developments. All instrumental functions, including wheel positioning and control of the Risley prisms are monitored by a microprocessor and remotely accessible via a RS232 link.
One of us (J.-L. Prieur) has developed a program in the PC/Window environment to facilitate the remote control of PISCO. All the basic functions are available (Fig. ) with mouse-driven menus. The program controls in real time the atmospheric dispersion correction according to the telescope position, the filter and the atmosphere parameters (§2.2). A log file is also produced at the end of each night with the PISCO setup parameters of all the exposures taken during that night.
PISCO was primarily designed to be used at the Cassegrain focus of TBL, but it has also been made mechanically and optically compatible with the Cassegrain foci of the Canada-France-Hawaii (CFH) and European Southern Observatory (ESO) 3.6-m telescopes.
[*2.2. Atmospheric dispersion correction with Risley prisms*]{}
10000 For an astronomical object observed from the ground at an elevation different from zenith, the atmosphere behaves as a dispersive prism (see f.i., Simon, 1966). Polychromatic images are spread into a small vertical spectrum. For instance, for a 250 nm bandwidth centered at 500 nm, the typical atmospheric dispersion is 1 for an elevation $h = 60^\circ$ and 2 for $h =30^\circ$.
In PISCO, the atmospheric dispersion is corrected with “Risley prisms,” which consist in two identical sets of prisms (Breckinridge [*et al.*]{} 1979, Walner, 1990) that can be rotated to produce a tunable chromatic dispersion both in amplitude and direction (Fig. 4a).
Each set is made of two prisms of different dispersion law and roof angles, placed in an upside-down position. These prisms have been designed to have a null mean deviation, and a dispersion allowing atmospheric correction from the zenith down to an elevation of 30$^\circ$ for the blue domain which is the most defavorable (B filter, centered at 450 nm with a 70 nm bandwidth). We used the same combination of Shott glasses (F4, SK10) as the one used for the Kitt Peak speckle camera (Breckinridge [*et al.*]{}, 1979). Wallner (1990) found other combinations which are closer to the atmospheric dispersion curve but we chose the Kitt Peak combination because of its low cost and sufficient efficiency for our purpose. Our Risley prisms reduce the residual dispersion down to a level smaller than 0.01“ for every location of the object in the sky above an elevation of 30$^\circ$, with the 70 nm bandpass B filter which is small compared to the diffraction limit of 0.05” in B at the TBL.
During the observations, a specially designed program (already mentionned in §2.1) computes the elevation of the star and the corresponding atmospheric dispersion using J.C. Owens’ model of atmosphere (Owens, 1967, formulae 29–31). The Risley prisms are then dynamically rotated during data acquisition to compensate for the atmospheric dispersion.
[**3. The observation modes**]{}
10000 PISCO can be used in various modes which are selected during the observations. Switching from one mode to another takes a few mouse clicks and less than one minute for the motors to set the wheels.
[*3.1. Full pupil speckle imaging*]{}
10000 In this mode, no pupil masks are used and a high magnification is selected with wheel GR (Fig. ). It corresponds to the “conventional” way of observing in speckle interferometry. Most speckle cameras (Breckinridge [*et al.*]{} 1979, Strittmater, 1980, Foy, 1988a). offer only this possibility – or would require many optical changes with a full re-calibration of the instrument for other operating modes –. By applying bispectral techniques (§5.2), we have obtained images with an angular resolution close to the diffraction limit of the telescope (Fig. 4b).
to to The full pupil is used and the optical transfer function (OTF) corresponds to that of the telescope, and we shall see in the following that another OTF may be preferred. For instance, the diffraction pattern of the telescope spider may pollute the final image and hinder the detection of faint objects in the vicinity of a bright one.
Another drawback of this OTF is that low spatial frequencies dominate the transfer function. In photon counting mode mode, the few available photons are then mainly spread in the (less useful) low frequencies.
As the limited dynamic range of detectors such as the CP40 imposes the use of neutral densities to reduce the photon flux for bright objects to only a few hundreds per frame(§4), they always work in photon counting mode. This limits the performances of the restoration methods even for bright objects. This is the reason why the next mode could be preferred in that case.
[*3.2. Masked pupil speckle imaging and aperture synthesis*]{}
10000 By inserting masks into the pupil plane (P1) (cf. Fig. ), the pupil function (and thus the OTF) can be modified as desired. For instance the spider diffraction pattern can be removed by placing a Lyot’s mask or a four-hole mask which carefully avoids the shadow of the spider (see §3.3), and telescope arrays can be simulated by placing a mask with appropriately located small holes.
Pupil masks allow to select a sub-sample of spatial frequencies and more accurately measure the corresponding complex visibilities since they will be less attenuated (as the overlap of the fringes is lower); hence, a better use of the maximum number of photons allowed by the detector. The price to pay is to perform an interpolation in the Fourier space (aperture synthesis, see methods in §5.1) and complementary observations to make the process more robust in the case of complex objects.
We made 3 pupil masks by drilling 0.7 mm holes into a 5-cm metallic disk, according to some of the complementary non-redundant networks from Golay (1971). They are displayed in Fig. with the corresponding $(u,v)$ coverage. A successful image of HR 8652 was restored using these masks and the aperture synthesis method from Lannes (1989, 1991) and Anterrieu (1992). These method would work with any other configuration – there is no need for complementary networks –. We chose these masks because the corresponding $(u,v)$ coverage (Fig. ) was rather compact with only small gaps, which makes image restoration more robust.
The main drawback is a lower limiting magnitude in the case of small holes, and this method can only be used for objects with $V < 8$ at the TBL.
to
[*3.3. Coronagraphic mode*]{}
10000 PISCO can be used as a Lyot’s coronagraph by putting adequate masks $m_1$ in the entrance image plane (I1) and $m_2$ in the pupil plane (P1) (wheels EN and MA of Fig. ). This mode was successfully tested in 1994 with long integrations on a conventional CCD detector. Speckle imaging from short-exposured frames has little interest in this mode since the obturation of the mask $m_1$ in (I1) needs to be quite large (a few times the FWHM seeing) to hide most of the brightness of the central target. This mode would take its real advantage with adaptive optics and a small obturation of $m_1$, to investigate closer to the target (see f.i., some recent developments with a phase mask in the image plane, Roddier & Roddier, 1997).
A four-hole pupil mask can also be used to suppress the diffraction image of the spider of the telescope. This reduces the diffusion of a bright object and allows the detection of a possible faint close companion or stellar envelopes, as seen in Fig. . The stellar profile is more concentrated when putting this mask. If we normalize the profiles with the central value, the level of the wings has been reduced by a factor larger than 3.
to
[*3.4. High angular resolution spectroscopy*]{}
10000 Some authors have already shown the feasibility of speckle spectroscopy (Weigelt [*et al.*]{}, 1991, Kuwamura [*et al.*]{}, 1992) which has a great interest for the individual study of binary stars or for determining the physical nature of fine details found in speckle imaging. Two possibilities have been used: spectroscopy with or without a slit in the entrance image plane.
– Both the Hokudai speckle camera at the Okayama 188 cm telescope and the Steward Observatory speckle camera with the spectroscopy module at KPNO used by Kuwamura [*et al.*]{}, 1992, worked in [*objective prism spectroscopy mode*]{}, i.e., slitless spectroscopy. In this mode, the resolution is not fixed and changes as the seeing varies. The spectral calibration is rather difficult to perform since it depends upon the position of the object in the field. But the main advantage is that all the incoming light is used, without any loss.
– Weigelt [*et al.*]{}, 1991, proposed a slit spectroscopy setup which allows a high spectral resolution and a fixed spectral calibration. This is the option we chose because we wanted to be able to do stellar classification of the components of binary stars and work with a good spectral calibration. The main drawback is a loss of sensitivity due to a rejection of the light by the entrance slit. PISCO can be easily converted to a spectrograph by selecting the grism in the wheel FA and a slit in the wheel EN in the entrance image plane (I1) (Fig. ). It then provides a low dispersion spectrographic mode with a spectral range of 350–500 nm, and a spectral resolution of $\sim$300 with a slit of 0.7. This range was chosen to allow stellar classification of close binaries with the hydrogen Balmer series. Unfortunately, the atmospheric turbulence is stronger in the blue domain, which make images more difficult to restore. The wavelength calibration can be performed with calibration spectral lamps in wheel AS (Halogen, Argon, Neon and Xenon lamps).
Quick switching between imaging and spectroscopy is possible since this mode can be remotely selected by rotating the wheels. The observing procedure is the following:
– obtain the autocorrelation of the binary star in the full pupil mode (cf., §3.1) and measure the rotation angle to align the slit on the direction of the two components.
– rotate the telescope flange supporting PISCO and center the object on the slit.
– switch to the spectroscopic mode and record the data.
Restoration of high angular resolution information in the direction of the entrance slit is then done by applying one-dimensional speckle imaging techniques to each monochromatic image of the slit (see §5.2).
The first spectroscopic observations were made in 1995. Unfortunately the poor seeing conditions and the low dynamic range of the detector did not allow us to restore high resolution images. We simply obtained long integration spectra (Fig 5774) and calibrated the whole instrument with known stars.
=8.8cm
Figure 7: First spectrum obtained with PISCO (HD 5774).
[*3.5. Wavefront analyzer*]{}
The atmospheric wavefront can be sensed with the Shack-Hartman method (Roggeman [*et al.*]{}, 1997), by putting a microlens array into the pupil plane (P1) and a specific imaging lens in the GR wheel, (cf. Fig. ). Each microlens has a diameter of 0.7 mm which corresponds to 20 cm on the pupil at the TBL and 10 cm at the 3.6 m CFH or ESO telescopes.
The SCIDAR (SCintillation Detection And Ranging) technique (Vernin & Roddier, 1973) which consists in analysing the images of the pupil brightness lit by two stars to measure the wind speed and the altitude of the turbulence layers, could also be applied with PISCO. In that case, fast detectors operating at frequencies larger than 200 Hz (such as the PAPA or the RANICON cameras, cf. §4) are needed to “freeze” the turbulence.
[**4. The detectors**]{}
10000 PISCO has been used with a wide range of detectors. Actually, the qualities of the detector mainly condition the performances of the image restoration process. A good knowledge of the limitations of the detector is essential to elaborate a strategy of observation and obtain valid measurements. Here we describe the detectors that we have already tested on PISCO. For a wider information about the detectors used in the field of optical speckle interferometry, see for instance a review in Cuby (1988), Richard [*et al.*]{}, (1990) or Cuby [*et al.*]{}, (1990).
[*4.1. The CP40-INSU detector*]{}
10000 The first CP40 detector was designed at CERGA for interferometry observations by A. Blazit (Blazit, 1976, 1987). It is a two-stage intensified CCD camera followed by a photon analyzer which computes the coordinates of the photo-events (Foy, 1988b). The field is covered by a mosaic of 4 CCD’s, 288$\times$384 pixels each. We actually used the duplication of this prototype, financed by INSU to make it available to the French astronomical community and in particular to the instruments of the TBL.
The exposure time is set to 20 msec, which may be too large for speckle applications when the coherence time is smaller than this value. To circumvent this difficulty a rotating shutter was implemented which reduces the exposure time to 5 or 10 msec. This shutter interrupts the light beam in the speckle camera with a rotating opaque sector synchronized with the frame signal of the CP40 (phase-locked motor).
Because of a dead zone between the 4 image quadrants of the CP40, we decided to use only one quadrant and made a special “off-axis” mechanical interface to align the center of the selected quadrant with the optical axis of PISCO.
The geometrical distortion caused by the two-stage amplifier is rather large, of the order of 20% in the edges (Thiebault, 1994).
Another problem is a strongly non-uniform sensitivity of the photo-cathode within a single quadrant (down to nearly zero in one edge), which can hardly be corrected by a flat field map and causes a big non-uniformity of the signal to noise ratio within the elementary frames. The photometry of the image restoration process is also badly affected for intrinsically big objects that spread on the whole image.
10000 The electronic device which computes the coordinates of the photo-events produces an artifact which affects the photometry of the images.
When two photo-events are very close in the image, they merge into a single spot. The photon centering device is unable to identify it properly and discards such an event. This causes a depletion of high spatial frequencies in the power spectrum. A “hole” can be seen in the center of the mean auto-correlation, which becomes larger when the photon flux increases. This problem also affects the photometry since many photons are not recorded in the high intensity regions of the image.
To reduce this effect during our observations, the photon flux had to be limited to around 10,000 photons/sec and a high magnification was used to over-sample by a factor of 3.
[*4.2. The Ranicon*]{}
10000 The Ranicon (“Resistive Anode camera”, described in Clamping and Paresce (1989)) has been built by the Space Telescope Science Institute (Baltimore). The model we used was lent by the Observatoire de la Côte d’Azur (OCA) for some observing runs between 1993 and 1996.
This detector has a S20 photo-cathode and a saturated mode single microchannel amplifier (Gen II). The position analysis of the detected photo-events is made with a resistive anode. Each cloud of amplified electrons, resulting from the impact of a photon on the photo-cathode, produces a charge drift towards the four electrodes which surround the resistive anode. The location of the impact is deduced from the voltage variations et the electrodes and can be measured accurately to about 10 kHz.
Compared to the CP40, the quantum efficiency of the Ranicon is smaller by a factor of $\sim$3. This is due to the lower efficiency of the GEN II compared to the GEN I intensifiers.
Although the micro-channel amplification does not introduce geometric distortion, we have noticed a small distortion with the X and Y axes which are not perfectly perpendicular. A small variation of the geometric scale was also noticed and calibration was needed during the night.
Another unexpected defect was the presence of a small “hole” at the center of the autocorrelation function, similar in some way to that of the CP40 (cf, §4.2), but with a smaller amplitude. This is caused by the depletion of electrons of a micro-channel after a photon-detection: a delay of a few tenths of milliseconds is needed to recover its full charge and efficiency. To reduce this defect, the photon flux had to be lowered to about 8000 photons/sec.
[*4.3. Other detectors*]{}
10000 Two other detectors have been used with PISCO: the ICCD (Intensified Charge Coupled Device) belonging to C. Aime and E. Aristidi’s from Nice University, and P. Nisenson’s PAPA camera from Harvard Smithsonian Center for Astrophysics (CfA):
– The ICCD has a single stage intensifier. It cannot operate in true photon counting mode and is thus limited to objects brighter than V$_{lim} \sim$10. This detector has no significant geometric distortion nor non-linearity problems which would affect the photometry measurements. The exposure time can be set between 64 $\mu$sec and 16 msec and the gain of the micro-channel amplifier can be tuned, thus allowing a wide range of input luminosities. The output is an analog video signal, recorded on SVHS video cassettes. Bispectral image restoration with this detector has been very promising, and the first attempts lead to the restoration of a triple star (Aristidi [*et al.*]{}, 1997).
– The principle of the PAPA camera was described in Papaliolios and Mertz (1982) and Papaliolios [*et al.*]{}, (1985). It features a two-stage electrostatic amplifier, and a fast (P46) phosphor. Amplified photon impacts are analyzed by a set of binary masks which act as an optical computer to instantly digitize the position of photons in the field. The version we used was new and not fully operational, with a new binary mask setup and a refurbished image intensifier jointly made by P. Nisenson, D. Gezari (CfA) and L. Koechlin (OMP). The first observations in June 1997 have shown that the quantum efficiency was very good, slightly larger than that of the CP40. The maximum photon rate per second was as high as 100,000 but a small “hole” at the center of the autocorrelation function was also noticed. The geometric distortion caused by the image intensifier was large and an overall scale variation during the night imposed quasi permanent scale calibrations. To solve this problem, the image intensifier was changed after these observations.
[*4.4. Comparison of the detectors*]{}
10000 Here is a summary of the characteristics of the detectors used with PISCO.
10000
1. The CP40 has a good quantum efficiency but a non-uniform sensitivity and a strong geometric distortion, with a fixed integration time of 20 msec. The “photon-counting hole” affects the photometry and limits the photon flux to around 10000 ph/sec (limiting magnitude at the TBL: V$_{lim}\sim$12).
2. The Ranicon has a very low geometric distortion, but a poor quantum efficiency and a limitation of the usable photon flux to around 8000 ph/sec (V$_{lim}\sim$11). It generates a chronologically ordered list of photon coordinates.
3. The PAPA exhibits a flat-field pattern and geometric distortion. The photon flux is limited to around 100,000 ph/sec, and V$_{lim}\sim$12. It also a chronologically ordered list of photon coordinates.
4. The ICCD of Nice Univ. has a lower gain than the previous detectors, no geometric distortion, virtually no limitation to the photon flux for normal astrophysical use, and V$_{lim}\sim$10. The image rate is 50 Hz, with an electronic shutter, able to reduce the integration time to 0.06 msec
10000 Hence the detectors should be chosen according to the observing program, since some defects may be incompatible with the observation requirements. A good detector for high resolution imaging is still to be desired. Some technical developments are under way in our team to contribute to this problem. A prototype of a new photon-counting camera that would allow a high photon rate and a direct numerisation of photon coordinates is beeing tested (DELTA camera, Koechlin & Morel, 1998).
[**5. Performances and limitations**]{}
10000 [*5.1. Physical limitations*]{}
10000 The effects of seeing on speckle observations is a strong reduction of the limiting magnitude for bad seeing conditions, whereas the angular resolution attainable in the image restoration process degrades more slowly from the theoretical limit of $\lambda/D$, where $\lambda$ is the wavelength and $D$ is the telescope diameter (Roddier, 1981).
As solar observations have demonstrated, the Pic du Midi site sometimes features slow seeing variations and extended isoplanetic patch, which indicates that the TBL is potentially well suited to speckle and adaptive optics observations at short wavelengths, despite its modest size. With a diffraction limit at 0.06$\,$ in V, the TBL can provide high quality data from which many astrophysical programs could benefit.
Due to the necessary short exposure times photon noise is the most severe limiting factor in speckle imaging (Dainty and Greenaway, 1979, Beletic and Goody, 1992). The limiting magnitude depends on the atmospheric seeing, the spectral bandpass, the angular resolution to be achieved, and the quantum efficiency of the detector (Dainty and Greenaway, 1979). We reach with the TBL and detectors such as the CP40 and the PAPA with a bandpass of 70 nm. Without filters, the expected limiting magnitude would increase to about . In that case, the wavelength range is determined by the product of the sensitivity response of the detector with that of the photo-cathode, and the resulting bandwidth is a few hundreds of nanometers. It was shown both experimentally (Hege [*et al.*]{}, 1981) and with numerical simulations (Ziad [*et al.*]{}, 1994) that such extreme observing conditions can be used for faint detection of object duplicity.
[*5.2. Data processing and image restoration*]{}
10000 Whereas data quality is of paramount importance and obviously limits the angular resolution that can be ultimately obtained in the reconstructed images, the nature of the data reduction methods subsequently employed to extract the scientific information contained in these images plays a key role in ensuring the overall success of the scientific programs.
We have written the software to process data from the various detectors used (cf. §4), and the different observing modes: speckle imaging, aperture synthesis with pupil masks, speckle spectroscopy, and coronagraphy. The analysis of the wavefront is not yet implemented.
For speckle imaging, a few observers have independently reached the conclusion that “the bispectrum combined with a constrained iterative deconvolution of amplitudes produces the highest quality imagery” (Beletic and Goody, 1992). Nevertheless, we have used various programs ranging from Knox-Tompson (1974) to full bispectrum methods (Weigelt, 1977, Roddier, 1986, Lannes, 1989) (and even partial bispectrum methods, i.e. using only a subset of all possible closure relations) and found little differences on the restored phasor image of double stars. The pre-processing of the original data (correction of geometric distortion, of flat-field, and various calibrations) is for us the crucial step in the whole image restoration process.
The “Aperture Synthesis” team at OMP has been mainly involved during the last few years in the theoretical aspects of aperture synthesis and related problems such as deconvolution, wavelets and multi-resolution methods, with applications on single aperture interferometry and multi-aperture devices (Lannes [*et al.*]{} 1987; Lannes 1988, 1989, 1991). The approach to these problems is deterministic and based on a least-squares scheme that allows error analysis, hence a good understanding of the stability of the image restoration process. Note that the deconvolution method (Lannes [*et al.*]{} 1987) respects the photometry of the target, which is necessary for many applications, such as the determination of color indices of binary stars for example (cf §6).
[**6. Astrophysical programs**]{}
10000 In this section, we describe are some of the scientific subjects which are being studied with PISCO at Pic du Midi. Other programs which aimed at imaging complex objects (asteroids and stellar enveloppes) have been impossible to do because of bad weather conditions.
10000 [*– Orbits of binary stars*]{}
10000 The study of binaries is a well suited program for speckle cameras as the bibliography of the last two decades can easily show (CHARA project, McAlister and Hartkopf, 1984, 1988, Hartkopf [*et al.*]{}, 1996).
A long term program aims at measuring the position of close binaries to determine the orbits and derive the masses of the components using the parallaxes measured by Hipparcos (Carbillet [*et al.*]{}, 1996, Aristidi [*et al.*]{}, 1997, Aristidi [*et al.*]{}, 1998). New orbital elements have already been recalculated for 8 double stars from these observations (Aristidi [*et al.*]{}, 1998).
We noticed that PISCO was very efficient for binary study, even when the atmospheric conditions were poor and did not allow any other imaging program. Hence binary measurements have been used as a backup program of all our high angular resolution observations.
[*– Stellar classification of components of binary stars*]{}
10000 When images have been restored in B, V, R, with a good photometry (cf. §5.2), color indices can be measured which then allow stellar classification of each of the two stars. This may reveal essential for the stars for which a derivation of mass has been made. Acurate orbit determination (and hence masses) are easier to perform for short-period binaries, which are generally very close and for which only global color indices or spectra are available. The individual stellar classification is then poorly known. The paradox is then that acurate masses are affected to stars with big uncertainties in the stellar classification, or less acurate masses to well identified stars, in the case of binaries with a big angular separation. Hence we see that color indices, or even spectra, of individual stars are crucial for stellar studies.
A study of composite spectrum stars (coll. J.-M. Carquillat and N. Ginestet, OMP) associate the imaging and spectroscopic modes of PISCO. Some stars exhibit the signature of a composite spectrum which could be interpreted as the sum of (at least) two spectra of different type. (Ginestet [*et al.*]{}, 1994) The aim of this program is to detect the possible presence of a companion and then to identify the spectral type of both components either with color indices or (when possible) with a spectrum with high angular resolution which would separate the spectra of the individual stars.
[*– Search for binarity and statistical studies*]{}
10000 The influence of the presence of a companion for star formation (accretion of a disk) and stellar evolution (stellar winds, f.i.) is not yet fully understood. Hence some surveys have been undergone to determine the frequency of binarity among pre-main sequence or post-AGB stars to constrain the theoretical models with these statistical results. Once the binarity has been established, the next step is to identify the nature of each companion, either with its photometry or directly by spectroscopy.
A statistical study of pre-main sequence stars has been started in 1996 with complementary high angular observations made at ESO and CFHT with Adaptive Optics (AO) in the infra-red (Bouvier [*et al.*]{}, 1996).
Another program directed by E. Aristidi and B. Lopez (Nice Univ., France) aims at searching for binaries among Mira-type stars (for which binarity has been suspected by Hipparcos), and studying the interaction between the envelope of the Mira and the atmosphere of the companion (Lopez [*et al.*]{}, 1998).
[**7. Conclusion**]{}
10000 The speckle camera of Observatoire Midi-Pyrénées has been tested in all its operating modes and is now qualified for routine scientific exploitation. Its versatility with multi-mode observational possibilities makes it particularly well suited to testing the new methods of image restoration and aperture synthesis. The experience gained with pupil masks may have direct applications for reducing data from optical interferometric arrays.
The good performances of speckle methods for binary star observations have lead to consequent orbit measurements during the last twenty years all around the world and PISCO has started to bring its contribution to this effort (Carbillet [*et al.*]{}, 1996, Aristidi [*et al.*]{}, 1997, Aristidi [*et al.*]{}, 1998). This high efficiency makes speckle observations a “privileged” tool for binary studies. A new series of speckle programs have been impulsed by the discovery by Hipparcos of thousands of binary candidates (confirmation of binarity, orbits, variability of companions, etc). New space projects such as the space interferometers dedicated to parallax measurements (ESA GAIA) will also need follow-up based-ground observational programs in the future for which PISCO and speckle techniques in general may significantly contribute.
[**Acknowledgements:**]{}
We are indebted to A. Blazit, D. Mourard, E. Aristidi, D. Gezari and P. Nisenson for lending us their detectors, and to A. Lannes, M. Festou, J.-M. Carquillat, N. Ginestet and M. Scardia for the fruitful collaboration for the scientific exploitation of PISCO.
We thank the Observatoire Midi-Pyrénées technical staff, and especially the workshop of Toulouse, Bagnères de Bigorre and Pic du Midi and the night assistants and operators of the TBL, for their participation to this project. We acknowledge the assistance of J. Cadaugade and S. Chastanet for the preparation of the photographs.
This instrument was financed by a grant from the [*Institut National des Sciences de l’Univers*]{} of the [*Centre National de la Recherche Scientifique (CNRS)*]{} to the TBL, with additional support from the [*Unités de Recherche Associées n$^{\circ}$1281*]{} and [*n$^{\circ}$285*]{} (now [*Unité Mixte de Recherche n$^{\circ}$5572*]{}) of CNRS.
**Bibliography**
[^1]: Based on observations made at Télescope Bernard Lyot, Pic du Midi, France
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The break-up of a two-dimensional circular disc by normal and oblique impact on a hard frictionless plate is investigated by molecular dynamics simulations. The disc is composed of numerous unbreakable randomly shaped convex polygons connected together by simple elastic beams that break when bent or stretched beyond a certain limit. It is found that for both normal and oblique impacts the crack patterns are the same and depend solely on the normal component of the impact velocity. Analysing the pattern of breakage, amount of damage, fragment masses and velocities, we show the existence of a critical velocity which separates two regimes of the impact process: below the critical point only a damage cone is formed at the impact site [*(damage)*]{}, cleaving of the particle occurs at the critical point, while above the critical velocity the disc breaks into several pieces [*(fragmentation)*]{}. In the limit of very high impact velocities the disc suffers complete disintegration [*(shattering)*]{} into many small fragments. In agreement with experimental results, fragment masses are found to follow the Gates-Gaudin-Schuhmann distribution (power law) with an exponent independent of the velocity and angle of impact. The velocity distribution of fragments exhibit an interesting anomalous scaling behavior when changing the impact velocity and the size of the disc.'
author:
- 'Bhupalendra Behera$^{1,3}$, Ferenc Kun${}^2$, Sean McNamara${}^1$, and Hans J. Herrmann${}^1$'
title: Fragmentation of a circular disc by Impact on a Frictionless plate
---
Introduction {#intro}
============
The strength and break-up of agglomerates composed of smaller sized primary particles is of particular importance for the storage and handling of materials in process industries such as pharmaceuticals, chemicals, fertilizers, detergent, and food industries. In industrial processes agglomerates often collide with each other and with the hard walls of the equipment resulting in a size reduction, which is desired or not depending on the type of the process. The strength of agglomerates has to be characterized for the design of operating conditions in industrial processes such as milling, tabletting, mixing, and transport in pneumatic conveying. Another important class of agglomerates are the so-called particle compounds, which are the combination of various sized particles embedded in a cementous matrix. The different types of engineering agglomerates and building materials like concretes are some examples of particle compounds. It is of high industrial importance to recycle these particle compounds in order to use the valuable aggregates. The design and optimization of the liberation process of aggregates from the matrix material requires a detailed knowledge of the strength and break-up of compounds.
For the understanding of the strength and break-up process, the study of simple systems like spherical particles is essential. During the last decades several experimental and theoretical studies have been performed to understand the break-up of spherical bodies arising due to impact. The crack pattern of sand-cement spheres by a free fall impact was studied in Ref. [@fra_pattern], which reports observations of meridian cracks, that divide the sphere into two nearly equal parts, and oblique cracks, which are straight like median cracks, but cut the sphere into two unequal pieces. The fracture of glass and plaster spheres by free fall impact and double impact (dynamic loading between two hard plates) have been carried out recently [@powder; @chau]. It was found that at the lowest impact velocities hertzian cone cracks (formed from a surface ring crack) are developed, whereas, at high velocities, oblique cracks propagate before meridian cracks form [@fra_glass]. This finding differs from the experimental results of Ref. [@fra_pattern], where it was found that with increasing impact energy, the number of meridian planes increases and oblique cracks start to develop.
Due to the high speed and violent nature of the break-up process, observations are usually restricted to the final state of impact experiments, where information has to be extracted from the remaining pieces of the body. Hence, computer simulation of models of agglomerate break-up is an indispensable tool in this field. Simulations of realistic models provide a deeper inside into the break-up process and can even complement the experimental findings directly supporting the design of industrial processing of these materials. Analytic approaches have limited capabilities in this field since they cannot capture the disordered microstructure of the material.
The finite element approach and the discrete element modeling have been successfully applied to describe the stress field, crack propagation, and fragment formation in impacting spherical particles [@agglomerate; @bk; @imp_agglo; @poto; @simu; @tsoungui; @imp_ang; @poschel1; @poschel2]. Recent simulations of ball impact revealed two types of crack patterns: oblique cracks radiating from impact point, and secondary cracks perpendicular to the oblique ones. In the framework of the discrete element method it was clarified that depending on the impact velocity the result of the break-up process can be localized damage around the contact zone, fragmentation, or shattering. The evolution of several characteristic quantities of the break-up process when increasing the impact velocity were monitored and analyzed in normal and oblique impact [@agglomerate; @bk; @imp_agglo; @poto; @imp_ang].
From a more general point of view, the break-up of agglomerates presents an important class of fragmentation phenomena which is ubiquitous in everyday life and concerns a wide range of phenomena in science and technology. In general, when an object is subjected to shock or stress it will break up into smaller pieces. The length scales involved in this process range from the collisional evolution of asteroids [@asteroids] to the degradation of materials comprised of small agglomerates [@agglomerate; @bk] employed in the process industries as summarized above. There are also many geological examples associated with the use of explosives for mining and oil shale industry, coal heaps, etc. A wide variety of experiments [@self; @composite; @instable; @ice; @lab; @glassplate] and simulations [@agglomerate; @bk; @discrete; @transition; @twodisc; @granulate; @diehl; @univer; @branching; @aspect; @droplet; @britt; @poto; @simu] revealed that the fragment mass distribution is a power law except for very large fragment sizes. The exponents in the power law region were found experimentally to be between 1.35 and 2.6 depending on the effective dimensionality of the system [@asteroids; @fra_pattern; @lab; @breakup]. Recent studies revealed that power law distributions arise in fragmentation phenomena due to an underlying phase transition [@instable; @transition; @univer]. However, most of the data reported in the literature is concerned with the general behavior of fragmentation processes. There is much less literature where the propagation and orientation of cracks are discussed.
In the present paper we study the normal and oblique impact of a circular brittle particle on a hard frictionless plate, varying the impact velocity and impact angle in a broad range. The particle is composed of numerous unbreakable, undeformable, randomly shaped polygons which are bonded together by elastic beams. The bonds between the polygons can be broken according to a physical breaking rule, which takes into account the stretching and bending of the connections. Based on simulations of the model, we performed a detailed study of the failure evolution at different impact velocities and of the nature of the crack propagation during the fragmentation process, and compared the results with experiments [@fra_pattern; @powder; @lab; @ice; @fra_glass]. In the analysis of the simulation data, we profit from recent theoretical results of general studies of fragmentation processes. We observed that for both normal and oblique impacts, the crack patterns are the same and depend solely on the normal component of the impact velocity. Studying the crack patterns, amount of damage, fragment masses, and velocities, we provide a quantitative foundation of the concept of damage, fragmentation, and shattering in ball impact, which was introduced recently on a more qualitative basis [@agglomerate]. We show the existence of a critical impact velocity $v_c$ which distinguishes two regimes of the impact process, [*i.e.*]{} below the critical velocity damage mainly occurs in a conical region around the impact site with a large residue, however, above $v_c$ an ensemble of oblique cracks develop and the disc breaks up into pieces. In agreement with experimental results, fragment masses are found to follow the Gates-Gaudin-Schuhmann distribution (power law) [@kelly] with an exponent independent of the velocity and angle of impact. The velocity distribution of fragments exhibit an interesting anomalous scaling behavior when changing the impact velocity and the size of the disc.
An important application of our results, besides the ones mentioned at the beginning, is to the optimization and control of tumbling mill performance. These questions are of utmost practical importance as they have a tremendous influence on power draft, wear of the balls and liners and breakage characteristics of the grinding materials. During the cataracting motion where the charge material inside a mill follows a parabolic path [@data], most of the materials are ground as hard balls fall back onto them. There is particular interest in the net energy required to achieve a certain size reduction and the the energy distribution of the fragments during the grinding process. The efficiency of the mills could be controlled if the breakage characteristics of the grinding materials were better understood. Our current work can provide some valuable information for the modernization of the mill design.
Model
=====
In order to study fragmentation of granular solids, we performed molecular dynamic (MD) simulations in two dimensions. To better capture the complex structure of a real solid, we used randomly generated convex polygons that interact with each other elastically. The model consists of three major parts, namely, the construction of a Voronoi cellular structure, the introduction of the elastic behavior, and finally the breaking of the solid. This section gives a detailed overview of these three steps.
In order to take into account the complex structure of the granular solid, we use randomly generated convex polygons, [*i.e.*]{} we divide the solid into grains by a Voronoi cellular structure. The Voronoi construction is a random tessellation of the plane into convex polygons. This is obtained by putting a random set of points onto the plane and then assigning to each point that part of the plane which is nearer to it than to any other point. One advantage of the Voronoi tessellation is that the number of neighbors of each polygon is limited which makes the computer code faster and allows us to simulate larger systems. In our case, the initial configuration of the polygons was constructed using a vectorizable random lattice, which is Voronoi construction with slightly reduced disorder [@rand_lattic]. First, the Voronoi tessellation of a square is performed, and then a circular disc with smooth surface is cut out.
In the model the polygons are rigid bodies. They are neither breakable nor deformable, but they can overlap when pressed against each other. This overlap represents local deformations of the grains. Usually the overlapping polygons have two intersection points which define the contact line. In order to simulate the elastic contact force, we introduce a repulsive force between touching polygons. This force is proportional to the overlapping area $A$ divided by a characteristic length $L_c$ $(\frac{1}{L_c}= \frac{1}{2}[\frac{1}{r_i}+\frac{1}{r_j}]$, where $r_i$, $r_j$ are the radii of circles of the same area as the polygons). The direction of the elastic or normal force is perpendicular to the contact line of the polygons. The complete form of the normal force contains an elastic and damping contribution, whereas the tangential component is responsible for the friction.
Again, to bond the particles together it is necessary to introduce a cohesive force between neighboring polygons. For this purpose we introduce beams. The centers of mass of neighboring polygons are joined together with elastic beams that exert an attractive, restoring force but can break in order to model the fragmentation of the solid. Because of the randomness contained in the Voronoi tessellation, the lattice of beams is also random. The length, the cross-section and the moment of inertia of each beam are determined by the initial configuration of the polygons. The Young’s modulus of the beams and of the particles are considered to be independent of each other. The beams break according to a physical breaking rule, which takes into account the stretching and bending of the connection. The surface of the grains where beams are broken represent cracks. The energy stored in the broken beams represents the energy needed to create these new crack surfaces inside the solid.
In order to simulate the break-up of the disc due to impact with a hard plate, a repulsive force is introduced between the plate and those polygons of the disc which have overlap with the plate. This repulsive force is proportional to the overlap area, similarly to the polygon-polygon contacts but with a higher stiffness value. The contact force of the disc and the plate has vertical direction, tangential component like friction is excluded in the present study.
The time evolution of the system is obtained by numerically solving Newton’s equations of motion of the individual polygons (Molecular Dynamics). For the solution of the equations we use a Gear Predictor-Corrector scheme of fifth order, which means that we have to keep track of the coordinates and all their derivatives up to fifth order. The breaking criterion of beams is evaluated in each iteration time step and those beams which fulfill the condition are removed from the calculations. The simulation stops after no beams break during a certain number of time steps. Previously this model has been applied to study fragmentation of solids in various experimental situations [@discrete; @transition; @twodisc; @granulate; @proj]. For more details of the simulation technique see Ref. [@discrete].
Crack pattern {#crack}
=============
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![\[snapshots\] Snapshots of the fragmentation process in normal impact of a circular disc of radius 30 cm at $250$ cm/sec. (a) $t=0.00150$ sec, a high compressive wave generated at the impact point causes the primary breakage. (b) $t=0.00300$ sec, cracks start at the periphery of the conical region whose base is the contact line between the disc and the plate. (c) $t=0.00600$ sec, oblique cracks move outwards. (d) $t=0.00996$ sec, the oblique cracks reach at the outer surface. (e) The disc is reassembled at the end of the simulation to observe the final crack pattern. ](crack/90_250_0.00150.ps "fig:"){width="30.00000%"} ![\[snapshots\] Snapshots of the fragmentation process in normal impact of a circular disc of radius 30 cm at $250$ cm/sec. (a) $t=0.00150$ sec, a high compressive wave generated at the impact point causes the primary breakage. (b) $t=0.00300$ sec, cracks start at the periphery of the conical region whose base is the contact line between the disc and the plate. (c) $t=0.00600$ sec, oblique cracks move outwards. (d) $t=0.00996$ sec, the oblique cracks reach at the outer surface. (e) The disc is reassembled at the end of the simulation to observe the final crack pattern. ](crack/90_250_0.00300.ps "fig:"){width="30.00000%"} ![\[snapshots\] Snapshots of the fragmentation process in normal impact of a circular disc of radius 30 cm at $250$ cm/sec. (a) $t=0.00150$ sec, a high compressive wave generated at the impact point causes the primary breakage. (b) $t=0.00300$ sec, cracks start at the periphery of the conical region whose base is the contact line between the disc and the plate. (c) $t=0.00600$ sec, oblique cracks move outwards. (d) $t=0.00996$ sec, the oblique cracks reach at the outer surface. (e) The disc is reassembled at the end of the simulation to observe the final crack pattern. ](crack/90_250_0.00600.ps "fig:"){width="30.00000%"}
(a) (b) (c)
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![\[snapshots\] Snapshots of the fragmentation process in normal impact of a circular disc of radius 30 cm at $250$ cm/sec. (a) $t=0.00150$ sec, a high compressive wave generated at the impact point causes the primary breakage. (b) $t=0.00300$ sec, cracks start at the periphery of the conical region whose base is the contact line between the disc and the plate. (c) $t=0.00600$ sec, oblique cracks move outwards. (d) $t=0.00996$ sec, the oblique cracks reach at the outer surface. (e) The disc is reassembled at the end of the simulation to observe the final crack pattern. ](crack/90_250_0.00996.ps "fig:"){width="40.00000%"} ![\[snapshots\] Snapshots of the fragmentation process in normal impact of a circular disc of radius 30 cm at $250$ cm/sec. (a) $t=0.00150$ sec, a high compressive wave generated at the impact point causes the primary breakage. (b) $t=0.00300$ sec, cracks start at the periphery of the conical region whose base is the contact line between the disc and the plate. (c) $t=0.00600$ sec, oblique cracks move outwards. (d) $t=0.00996$ sec, the oblique cracks reach at the outer surface. (e) The disc is reassembled at the end of the simulation to observe the final crack pattern. ](crack/90_250_0.01000.ps "fig:"){width="40.00000%"}
(d) (e)
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
In the present work we apply our model to explore the properties of the fragmentation process of a circular disc when dropped on a frictionless hard plate at different angles. The particle moves with a constant speed $v_0$ without the influence of gravity, which is obtained by supplying a constant velocity to all the polygons constituting the circular particle just before it touches the hard surface. The impact angle $\theta$ defined as the angle of the vector of the impact velocity to the horizontal, was varied between $90^{\circ}$ (normal impact) and $45^{\rm o}$ (oblique impact).
In order to understand the break-up process of discs, we investigated the crack pattern arising both in normal and oblique impacts. Fig. \[snapshots\] presents the time evolution of the crack pattern of normal impact obtained by the simulation of a circular disc of radius $30$cm at $250$cm/sec. When the disc strikes against the hard plate, a high compressive wave is generated at the impact point. Fracture starts from the region of contact point and propagates through the disc. As the time passes, more and more bonds break at the impact region and the area of contact increases progressively. As a result of this primary breakage a cone shaped (triangle shaped in two dimensions) damage area is created whose base corresponds approximately to the area of contact of the specimen and the target (see Fig. \[snapshots\]b) and it is more distinct at the end of the fragmentation process (see Fig. \[snapshots\]d). When the cone is driven into the specimen a large number of cracks are generated starting from the region around the cone (see Fig. \[snapshots\]b). This indicates that a high stress concentration has developed around the conical damage region. Later on these cracks run together to form few oblique cracks (see Fig. \[snapshots\]c) directing radially outward. As crack propagation is very energy dissipative, when these oblique cracks move outwards, the intensity of the compressive wave gradually decreases and hence larger fragments appear opposite the impact point.
To demonstrate the effect of the impact velocity on the break-up process, in Fig. \[snapshot\] final states of the process are shown obtained at different impact velocities, with the fragments reassembled into the initial disc. At low velocities the compressive wave intensity generated at the impact point is low, and hence, the cone could not develop fully. Moreover, only a few oblique cracks are obtained, and they do not reach the opposite surface of the disc (Fig. \[snapshot\]a). As the velocity increases, more oblique cracks develop and cover a greater distance (Fig. \[snapshot\]b) and a considerable part of the initial kinetic energy goes into the motion of the residue resulting in rebound. At the impact velocity where the oblique cracks reach the outer surface of the disc opposite to the impact point, the break-up process drastically changes: below this velocity mostly contact damage occurs in the form of the damage cone and a relatively big residue remains back. Above this velocity, however, the cracks spanning the entire disc result in the break-up of the residue into smaller pieces, see Fig. \[snapshots\]d. Later on it will be shown that the behavior of the system quantitatively changes at this velocity, which we call critical velocity. At impact velocities larger than the critical value, secondary cracks are generated roughly perpendicular to the oblique cracks. Secondary cracks from neighboring oblique cracks may merge with each other as can be seen in Fig. \[snapshot\]c. Also at higher impact velocities, vertical cracks with a direction nearly perpendicular to the target plate are more prominent as the intensity of stress concentration near the tip region of the cone is high as compared to other parts.
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![\[snapshot\] Final reassembled states of the break-up process of a disc of radius $30$ cm at different impact velocities dropped on a hard frictionless plate. (a) $v_0 = 100$ cm/sec. The cone is not fully developed and only few oblique cracks are present. (b) $v_0 = 200$ cm/sec. More oblique cracks develop and travel a greater distance. (c) $v_0 = 600$ cm/sec. Both oblique cracks and secondary cracks are present. ](crack/90_100_0.01000.ps "fig:"){width="30.00000%"} ![\[snapshot\] Final reassembled states of the break-up process of a disc of radius $30$ cm at different impact velocities dropped on a hard frictionless plate. (a) $v_0 = 100$ cm/sec. The cone is not fully developed and only few oblique cracks are present. (b) $v_0 = 200$ cm/sec. More oblique cracks develop and travel a greater distance. (c) $v_0 = 600$ cm/sec. Both oblique cracks and secondary cracks are present. ](crack/90_200_0.01000.ps "fig:"){width="30.00000%"} ![\[snapshot\] Final reassembled states of the break-up process of a disc of radius $30$ cm at different impact velocities dropped on a hard frictionless plate. (a) $v_0 = 100$ cm/sec. The cone is not fully developed and only few oblique cracks are present. (b) $v_0 = 200$ cm/sec. More oblique cracks develop and travel a greater distance. (c) $v_0 = 600$ cm/sec. Both oblique cracks and secondary cracks are present. ](crack/90_600_0.01000.ps "fig:"){width="30.00000%"}
(a) (b) (c)
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Crack patterns obtained in the final state of oblique impacts at impact angles $75^{\circ}, 60^{\circ}$ and $45^{\circ}$ are compared in Fig. \[snap2\]. It is important to emphasize that in our calculations the friction between the target plate and the circular disc is completely excluded. Under this condition, varying the impact velocity while keeping its normal component constant, practically the same crack pattern is obtained (see Fig. \[snap2\]). Thus, the crack propagation and orientation during the fragmentation process solely depends on the normal component of the impact velocity.
Comparing the crack pattern obtained in the simulations to the experimental results [@fra_pattern], we did not find any meridian cracks as they are difficult to detect in two dimensions. The pattern of oblique cracks and secondary cracks has a satisfactory agreement with the experimental results. The simulations confirm that oblique cracks which were observed in experimental investigations [@fra_pattern] develop along the trajectories of maximum compression planes.
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![\[snap2\] The reassembled snapshots of normal and oblique impacts at different velocities keeping the normal component constant as $500$cm/sec. The crack patterns are almost same in all cases. (a) $\theta = 90^{\circ}$. (b) $\theta = 75^{\circ}$. (c) $\theta = 60^{\circ}$. (d) $\theta = 45^{\circ}$. ](crack/crack_90_500.ps "fig:"){width="40.00000%"} ![\[snap2\] The reassembled snapshots of normal and oblique impacts at different velocities keeping the normal component constant as $500$cm/sec. The crack patterns are almost same in all cases. (a) $\theta = 90^{\circ}$. (b) $\theta = 75^{\circ}$. (c) $\theta = 60^{\circ}$. (d) $\theta = 45^{\circ}$. ](crack/crack_75_518.ps "fig:"){width="40.00000%"}
(a) (b)
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![\[snap2\] The reassembled snapshots of normal and oblique impacts at different velocities keeping the normal component constant as $500$cm/sec. The crack patterns are almost same in all cases. (a) $\theta = 90^{\circ}$. (b) $\theta = 75^{\circ}$. (c) $\theta = 60^{\circ}$. (d) $\theta = 45^{\circ}$. ](crack/crack_60_578.ps "fig:"){width="40.00000%"} ![\[snap2\] The reassembled snapshots of normal and oblique impacts at different velocities keeping the normal component constant as $500$cm/sec. The crack patterns are almost same in all cases. (a) $\theta = 90^{\circ}$. (b) $\theta = 75^{\circ}$. (c) $\theta = 60^{\circ}$. (d) $\theta = 45^{\circ}$. ](crack/crack_45_708.ps "fig:"){width="40.00000%"}
(c) (d)
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Results
=======
Studying the evolution of the final crack patterns when changing the impact velocity, we have identified a critical velocity $v_c$ which separates the regimes of different break-up mechanisms. In the following, we analyze characteristic quantities of the break-up process, and show that there are substantial differences between the two regimes.
Size Distribution of Fragments {#size}
------------------------------
![\[maxmas\] The mass of the first and second largest fragment as a function of the normal component of the impact velocity $v_0\sin{\theta}$ at different impact angles $\theta$ and velocities $v_0$ for a system size $30$ cm. The vertical dashed line indicates the critical point, furthermore, the damaged and fragmented regimes are also shown. ](figures/new_maxmass.eps){width="50.00000%"}
Recently, it has been shown that the final outcome of a fragmentation process can be classified into two states depending on the amount of the imparted energy: damaged and fragmented states with a sharp transition in between. Detailed analysis revealed that the transition between the two states occurs as a continuous phase transition which also provides a possible explanation of the power law mass distribution of fragments observed.
To explore the nature of the critical velocity $v_c$ identified in the previous section we investigated the evolution of the mass of the two largest fragments when varying the angle $\theta$ and the velocity $v_0$ of impact. Plotting the largest fragment mass as a function of the normal component of the impact velocity for both normal and oblique impacts in Fig. \[maxmas\], all curves fall over one another. This implies that in the absence of friction between the plate and the disc, the size reduction achieved depends both on $v_0$ and $\theta$ but in such a way that it depends on the combination of the two variables $v_n = v_0 \sin{\theta}$. The curves of the second largest mass exhibit the same data collapse when plotting them as a function of $v_n$ further supporting the above arguments.
The functional form of the two largest fragment masses in Fig. \[maxmas\] shows the existence of two distinct regions. At low impact velocity, breakage takes place only at the impact point and the largest fragment is nearly equal to the original mass. As the velocity of impact increases, more small fragments are chipped off from the impact point and cracks start around the damaged conical region and move towards the outer surface of the disc. At the critical velocity $v_c$ these propagating cracks reach the outer surface opposite to the impact point and the largest fragment break into several big pieces. The impact velocity where the second largest mass attains its maximum value or where there is an inflexion point in the largest fragment mass curve coincides with the critical velocity defined by analyzing the cracking pattern in Figs.\[snapshots\],\[snapshot\]. In our case for normal impact of a system of $30$ cm radius the critical velocity turned out to be $250$ cm/sec.
The quality of the data collapse of the curves in Fig. \[maxmas\] obtained at different impact velocities $v_0$ and angles $\theta$ is excellent for the largest fragment, however, there are large fluctuations of the value of the second largest mass at impact velocities just below the critical point. Above the critical point all curves merge nicely together.
![\[avgfra\] The average fragment mass $\overline M$ as a function of the impact velocity $v_0$. The inset shows the same curves plotted as a function of the normal component of $v_0$. For oblique fragmentation when the velocity approaches the critical point always fluctuations arise, however beyond the critical velocity all curves merge nicely. ](figures/velnor_avgfra_all.eps){width="50.00000%"}
More information about the evolution of fragment sizes with varying impact angle $\theta$ and velocity $v_0$ can be obtained by studying the moments of fragment masses [@instable; @transition; @univer; @proj]. The $k$th moment $M_k$ of fragment masses is defined as $$\label{eq:m_k}
M_k=\sum_i^{N_f} m_i^k-M_\mathrm{max}^k,$$ where $N$ denotes the total number of fragments, $m_i$ is the mass of fragment $i$ and $M_\mathrm{max}$ is the largest fragment mass. The definition Eq. (\[eq:m\_k\]) means that the $k$th power of the largest mass is extracted from the sum of the $k$th power of the fragment masses. The average mass of fragments $\overline M$ can be defined as the ratio of the second and first moments $\overline M \equiv
M_2/M_1$. In order to demonstrate the effect of rescaling the impact velocity, in the main panel of Fig.\[avgfra\] the average fragment mass $\overline M$ is presented as a function of the impact velocity $v_0$ for the system size $R=30$cm obtained at different impact angles $\theta$, and in the inset the same curves are shown as a function of the normal component of $v_0$. It can be seen that for each impact angle $\overline{M}$ has a peak which broadens and gets shifted towards larger velocity values with decreasing impact angle. However, when plotting the same quantity as a function of the normal components of the impact velocity $v_0\sin{\theta}$ all the curves fall on top of each other. Larger fluctuations arise below the critical point which are more dominant for lower impact angles. Note that the position of the maximum in the inset coincides with the transition point determined in Fig. \[maxmas\].
![\[specific\] Average fragment mass $\overline M$ as a function of specific energy $E_0/m_{tot}$. The critical point, [ *i.e.*]{} the position of the maximum, depends on system size. ](figures/speng_avgmas_90_all.eps){width="50.00000%"}
Studies of fragmentation of various types of brittle solids have revealed that a larger amount of energy is required to achieve the same size reduction on systems of larger size. However, in terms of the specific energy, [*i.e.*]{} the energy imparted to the fragmenting system divided by the total mass $E_0/m_{tot}$, all the characteristic quantities show a universal behavior, especially the critical value of the specific energy is independent of the system size [@transition; @proj; @twodisc; @granulate; @poto; @simu]. For impacting discs, however, the critical value of the specific energy shows a clear dependence on the size of disc $R$ as it is illustrated in Fig. \[specific\]. The larger the disc, the higher the energy density is required to break it into pieces. A possible explanation is that for larger discs, a larger part of the imparted energy goes into the motion of the fragments, lowering the efficiency of break-up.
The total number of fragments $N_f$ is also an important measure of the degree of break-up in the impact process. Fig \[nofra\] shows that the number of fragments $N_f$ is uniquely determined by the normal component of the impact velocity $v_n$, [ *i.e.*]{} the curves obtained at different impact angles present a perfect collapse when plotting them as a function of $v_n$. It can be seen in the figure that the number of fragments is a monotonically increasing function of the velocity, however, the functional form of $N_f$ seems to be different on the two sides of the critical point, [*i.e.*]{} up to the critical point the curves show clearly a straight line, whereas, above the critical point all curves are slightly bent towards down as the efficiency of the fragmentation process decreases. Replotting the results using logarithmic scale on the horizontal axis, however, a straight line is obtained above the critical point (see the inset of Fig. \[nofra\]), which implies that $N_f$ has the form $$\begin{aligned}
N_f = a\cdot \ln{\frac{v_n}{v_{nc}}} + N_c, \ \ \mbox{for} \ \ v_n > v_{nc},\end{aligned}$$ where $N_c$ denotes the number of fragments at the critical point and $a$ is the slope of the straight line in the inset of Fig.\[nofra\].
![\[nofra\] Number of fragments $N_f$ as a function of the normal component of the impact velocity.](figures/velnor_nofra_all.eps){width="50.00000%"}
The amount of damage occurring during the break-up process can be quantified by the so-called damage ratio $d$ proposed by Thornton [*et. al*]{} [@agglomerate]. $d$ is defined as the ratio of the number of broken contacts $N_b$ to the total number of contacts $N_c$ existing initially inside the disc. The damage ratio $d$ depends both on the impact angle $\theta$ and the impact velocity $v_0$, [*i.e.*]{}, increasing $v_0$ at a fixed value of $\theta$ results in an increase of $d$, furthermore, increasing the impact angle $\theta$ at a given value of $v_0$ the damage ratio also increases. However, when plotting $d$ as a function of the normal velocity $v_n$ in Fig. \[damage\_ratio\] the curves obtained at different impact angles collapse on top of each other, which implies that $d$ solely depends on $v_n$. Similarly to the number of fragments, $d$ is also a monotonically increasing function of $v_n$, however, its functional form changes at the critical point. It is observed that below the critical point $d$ is a linear function of $v_n$, while above the critical point the curve is non-linear, slightly bending down. On a semilogarithmic plot again a straight line arises which implies the functional form $$d = b\cdot \ln{\left(\frac{v_n}{v_{cn}}\right)}+d_c, \ \ \mbox{for} \
\ v_n > v_c$$ where $d_c$ is the value of $d$ at critical point and $b$ is the slope of the fitted straight line in Fig. \[damage\_ratio\]. A somewhat similar functional form of $d$ has also been pointed out by Thornton [@agglomerate; @imp_agglo] in impact of discs and spherical objects with a hard plate.
![\[damage\_ratio\] Dependency of damage ratio $d$, the number of broken beams $N_b$ to the total number of beams $N_o$, on the normal component of the impact velocity.](figures/impvelnor_beambrk_all.eps){width="50.00000%"}
Plate Force
-----------
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -----
![\[wforce\] (a) Keeping the normal component of impact velocity constant at different impact angle, the force exerted by the plate is almost the same. (b) Force exerted by the plate at various velocities during normal impact. (c) The kinetic energy normalized by the total kinetic energy as a function of time. (d) The state of the disc of size $30$cm after $t=0.00210$sec which corresponds to the maximum plate force and minimum kinetic energy at the impact velocity of $250$cm/sec. ](figures/wallf_200_all.eps "fig:"){width="49.00000%"} ![\[wforce\] (a) Keeping the normal component of impact velocity constant at different impact angle, the force exerted by the plate is almost the same. (b) Force exerted by the plate at various velocities during normal impact. (c) The kinetic energy normalized by the total kinetic energy as a function of time. (d) The state of the disc of size $30$cm after $t=0.00210$sec which corresponds to the maximum plate force and minimum kinetic energy at the impact velocity of $250$cm/sec. ](figures/wforce_90.eps "fig:"){width="49.00000%"}
(a) (b)
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -----
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![\[wforce\] (a) Keeping the normal component of impact velocity constant at different impact angle, the force exerted by the plate is almost the same. (b) Force exerted by the plate at various velocities during normal impact. (c) The kinetic energy normalized by the total kinetic energy as a function of time. (d) The state of the disc of size $30$cm after $t=0.00210$sec which corresponds to the maximum plate force and minimum kinetic energy at the impact velocity of $250$cm/sec. ](figures/energy_90_all.eps "fig:"){width="49.00000%"} ![\[wforce\] (a) Keeping the normal component of impact velocity constant at different impact angle, the force exerted by the plate is almost the same. (b) Force exerted by the plate at various velocities during normal impact. (c) The kinetic energy normalized by the total kinetic energy as a function of time. (d) The state of the disc of size $30$cm after $t=0.00210$sec which corresponds to the maximum plate force and minimum kinetic energy at the impact velocity of $250$cm/sec. ](crack/crack_250_90_0.00210t.ps "fig:"){width="40.00000%"}
(c) (d)
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
An alternative way of showing the effect of the impact angle is by analyzing the force exerted by the plate on the disc. In Fig. \[wforce\](a) and (b) typical time series of the force between the target plate and the disc are presented. If the normal component of the impact velocity $v_n =
v_0\sin{\theta}$ is kept constant while changing $\theta$ and $v_0$, the force practically remains constant except for fluctuations, see Fig. \[wforce\](a). It shows clearly that the force exerted by the target plate only depends on the normal component of impact velocity providing further support for the above findings in consistency with refs. [@agglomerate; @bk; @imp_agglo; @imp_ang].
To take a clear view of the nature of the plate force, we have plotted the plate force as a function of time at various velocities (see Fig. \[wforce\](b)). In general, as the impact velocity is increased, the maximum plate force increases, the duration of the impact decreases. The maximum force exerted by the plate occurs when the kinetic energy has a minimum (see Fig. \[wforce\](c)). The maximum plate force or minimum kinetic energy corresponds to the state of the fragmented disc where most of the bonds break near the contact region and cracks start to propagate radially outwards from the conical damage region (see Fig. \[wforce\](d). Since damage, [*i.e.*]{} bond breaking dissipates energy, the final kinetic energy is significantly less than the initial kinetic energy. Increasing the impact velocity gives rise to an increase in the final kinetic energy and decrease of the duration of plate-disc contact. Moreover, the initial kinetic energy remaining at the end of the impact first decreases with impact velocity until the velocity is sufficient to produce multiple fracture and then increases due to increase in kinetic energy of the broken fragments. Clearly, there are two stages to the bond breaking process during impact. Initially bonds are broken primarily as a result of the high compressive shock wave adjacent to the impact site, which occurs during the period when the plate force is increasing. This is followed by further bond breakage due to crack propagation radially outwards, starting around the conical damage region where high stress concentration occurs, while the plate force decreases.
Mass Distribution of Fragments
------------------------------
![\[mass\] The fragment mass histograms of normal impact for the system size $R = 30$cm with varying initial impact velocity. The straight line shows the power law fitted to the curve at the critical velocity $v_0 = 250$cm/sec with an exponent $\tau = 1.835$.](figures/massdist_90.eps){width="45.00000%"}
The mass (or size) distribution of fragments is one of the most important characteristic quantities of the disc impact which has also a high practical relevance. The fragment mass histograms $F(m)$ of normal impact are presented in the Fig \[mass\], for the system size $30$cm at varying impact velocity $v_0$. In order to resolve the shape of the distribution over a wide range of mass values, logarithmic binning was used [ *i.e.,*]{} the binning is equidistant on logarithmic scale.
It can be observed that the histograms have a maximum at small fragment sizes due to the existence of single unbreakable polygons. The shape of the peak of the small fragments is determined by the mass distribution of single Voronoi polygons obtained by the tessellation procedure. At low velocities, much below the critical point, the distributions are discontinuous: for small fragment masses the distributions are smoothly decreasing functions, while for large fragments $F(m)$ has a peak indicating the presence of large unbroken pieces. In between however, the medium sized fragments are missing. As the impact velocity increases, the large pieces break up into smaller ones, the peak of the large fragments on the right hand side gradually disappears and the entire mass distribution becomes continuous. It is interesting to note that the peak of large fragments disappears completely at the critical point where the cracks starting from the damaged conical region reach at the outer surface of the disc, breaking the disc into several smaller pieces. As a result, $F(m)$ takes a power law form at the critical point $$\begin{aligned}
F(m) \sim m^{-\tau}. \end{aligned}$$ The exponent of the power law $\tau$ fitted to the curve at the critical velocity $v_0 = 250$cm/sec is $\tau =
1.835$. For oblique impact the value of the exponent is nearly the same as in normal impact within a precision of $\pm 0.05$. Simulations with different system sizes such as $R=25$cm, $R=20$cm and $R=15$cm proved that the exponent $\tau$ is also independent of $R$.
Increasing the impact velocity above the critical point, the power law regime of the mass distribution remains unchanged, however, the largest fragment size decreased and the shape of the curve attains an exponential form for large fragments. In the limiting case of very high impact velocities the disc suffers complete disintegration into small pieces. In this shattered phase $F(m)$ gradually transforms to an exponential form. However, the shattered phase is slowly approached when increasing the impact velocity $v_0$ since the damage ratio and the number of fragments have a logarithmic dependence on $v_0$. The results are in good quantitative agreement with the experimental findings on the fragmentation of plate-like objects [@glassplate; @platelike; @composite; @ice].
Scaling of the Velocity Distribution
------------------------------------
In applications like mineral processing, a fragment, formed with a certain velocity, can undergo secondary break-up due to collisions with other fragments or with the walls of the container. To get an estimate about the importance of this secondary fragmentation, it is essential to study the fragment velocities. We investigated the velocity distribution of fragments and its dependence on the macroscopic variables of the system like the impact velocity $v_0$ and radius of the disc $R$.
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![\[veldistv\] The distribution of the $x$ and $y$ components of the velocity of fragments with fixed system size R=$30$ cm varying the initial impact velocity $v_0$.](figures/veldistx_30_90.eps "fig:"){width="49.00000%"} ![\[veldistv\] The distribution of the $x$ and $y$ components of the velocity of fragments with fixed system size R=$30$ cm varying the initial impact velocity $v_0$.](figures/veldisty_30_90.eps "fig:"){width="49.00000%"}
(a) (b)
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![\[veldistr\] The distribution of the $x$ and $y$ components of the velocity of fragments with fixed impact velocity $v_0 = 400$ cm/sec varying the radius $R$ of the particle.](figures/veldistx_radi_all.eps "fig:"){width="49.00000%"} ![\[veldistr\] The distribution of the $x$ and $y$ components of the velocity of fragments with fixed impact velocity $v_0 = 400$ cm/sec varying the radius $R$ of the particle.](figures/veldisty_radi_all.eps "fig:"){width="49.00000%"}
(a) (b)
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
To determine the velocity distribution of fragments and explore its dependence on $v_0$ and $R$, we analyzed the data in two ways. First we fixed the disc radius $R$ and varied the impact velocity $v_0$, then fixed $v_0$ while varying $R$. In both cases the calculations are restricted to normal impact ($\theta = 90^{\circ}$) and the distributions of the velocity components $n(v_x)$, $n(v_y)$ of the center of mass of the fragments are evaluated. In Figs. \[veldistv\],\[veldistr\] we present the results for fixed radius $R = 30$ varying the initial velocity and for fixed $v_0 = 400$ cm/sec varying the radius of the particles, respectively. In Fig. \[veldistv\](a) one can observe that the distribution of the $x$ component of the fragment velocities $n(v_x)$ is symmetric about $v_x=0$ as expected from the symmetry of the initial conditions. The zero mean value is a consequence of momentum conservation. As the impact velocity increases, the distribution broadens. The distribution $n(v_y)$ of the $y$ components also broadens with increasing impact velocity but also shifts towards the negative $y$ direction. This is obvious as the direction of the impact velocity is in the negative direction of $y$ axis and the total linear momentum increases with the impact velocity. However, in the $y$ direction fragments are slower, [*i.e.,*]{} the values of $v_y$ are much smaller than those of $v_x$. Note that there is a small fraction of the debris which has velocity larger than $v_0$, in agreement with experimental findings [@breakup]. When the impact velocity $v_0$ is fixed and the size of the disc $R$ is varied, however, the distribution of the $x$ components $n(v_x)$ remains the same (see Fig. \[veldistr\](a)). For $n(v_y)$ a similar trend is observed as in the case of changing $v_0$, except that the distribution is less dispersed with varying $R$ (Fig. \[veldistr\](b)).
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![\[scalev\] Rescaled plots of the velocity distributions for a fixed system size $30$ cm varying the impact velocity $v_0$ above the critical point.](figures/veldistx_abvcrt.eps "fig:"){width="49.00000%"} ![\[scalev\] Rescaled plots of the velocity distributions for a fixed system size $30$ cm varying the impact velocity $v_0$ above the critical point.](figures/veldisty_scl_90_abvcrt.eps "fig:"){width="49.00000%"}
(a) (b)
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![ \[scaler\] Rescaled plot of the distributions of the $y$ component of fragment velocities $n(v_y)$ for a fixed impact velocity $400$ cm/sec varying the size $R$ of the particle. ](figures/veldisty_sc_allradi.eps){width="49.00000%"}
To reveal the functional form of the dependence of $n(v_x)$ and $n(v_y)$ on the macroscopic variables $v_0$ and $R$ we performed a scaling analysis of the distributions. Figs. \[scalev\], \[scaler\] demonstrate that by appropriate rescaling the axes one can merge the curves obtained at different values of the macroscopic variables onto a single curve. In the case of the $x$ components, the transformation is a stretching and shrinking by a power of $v_0$ on the vertical and horizontal axis, respectively. However, for the $y$ components, a combination of a linear shift and a shrinking by a power of $v_0$ is required. The good quality of the data collapse implies that the $v_0$ and $R$ dependence of the distributions can be cast in the form $$\begin{aligned}
\label{eqx}
n(v_x,v_0) &\sim& v_0^{-\alpha} \phi\left(v_x
v_0^{-\alpha}\right), \\
\label{eqy}
n(v_y,v_0) &\sim& v_0^{-\beta} \psi\left((v_y+\lambda_1 v_0)
v_0^{-\beta}\right), \end{aligned}$$ where the parameter values $\alpha = 0.92$, $\beta = 0.58$, and $\lambda_1 = 0.38$ were obtained by varying them until the best data collapse is obtained in Fig. \[scalev\],\[scaler\]. Similarly at constant velocity while varying the system size $R$ the functional form of the distribution $n(v_y)$ reads as $$\label{eqyr}
n(v_y,R) \sim R^{-\gamma} \xi\left((v_y+\lambda_2 R)R^{-\gamma}\right),$$ where $\gamma = 0.12$ and $\lambda_2 = 6.2$ provide the best collapse in Fig. \[scaler\]. $\phi$, $\psi$, and $\xi$ are scaling functions which seem to have a Gaussian-like shape in Figs.\[scalev\],\[scaler\]. The scaling forms Eqs. (\[eqx\],\[eqy\],\[eqyr\]) show that the width of the scaling functions have a power law dependence on the impact velocity $v_0$ and on the radius of the disc $R$. It has to be emphasized that the above scaling behavior is valid only above the critical velocity $v_c$; below $v_c$ no scaling was found.
The scaling form of the distribution of the velocity components have also consequences for the spatial distribution of the flying pieces after impact. The increase of the width of the distributions with increasing impact velocity and disc radius implies that the flying fragments are more dispersed in space.
Conclusion
==========
We studied the normal and oblique impact of a circular brittle particle on a hard frictionless plate using a cell model of cohesive granular materials. We carried out a detailed analyses of the evolution of the crack pattern arising in the disc during the impact process, and of the mass and velocity distributions of fragments in the final state.
For both normal and oblique impact, a cone shaped damage region is formed at the impact point whose base area increases gradually as the velocity of impact increases. Cracks start to develop from the the conical damaged region where the maximum stress concentration exists. The oblique crack patterns obtained resemble those of the experimental findings [@fra_pattern], where oblique cracks moving along the plane of maximum compression were found. In agreement with the experimental observations, the oblique cracks in our simulation follow the trajectory of the maximum compression plane. Varying the impact velocities while keeping its normal component constant, we observed that the crack pattern remains the same in agreement with recent experimental and theoretical findings [@fra_pattern; @breakup; @powder; @fra_glass; @ball_impact].
Our analyses showed the existence of a critical value of the impact velocity, at which the oblique cracks reach the outer surface of the disc opposite to the impact point. The critical velocity separates two regimes of the break-up process, [*i.e.*]{} below the critical point only a damage cone is formed at the impact site [*(damage)*]{}, cleaving of the particle occurs at the critical point, while above the critical velocity the disc breaks into several pieces [*(fragmentation)*]{}. In the limit of very high impact velocities, the disc suffers complete disintegration into many small fragments. However, this shattered phase is slowly approached since the damage ratio and the number of fragments increase logarithmically with the impact speed. The critical behavior proved to be independent of the impact angle, it solely depends on the normal component of the impact velocity. Studying the average fragment size revealed that the critical value of the specific energy increases with the size of the disc. This implies in practical cases that a higher energy density is required to break a particle of larger size.
Above the critical point, the mass distribution $F(m)$ of fragments was found to obey the Gates-Gaudin-Schuhmann distribution (power law) with an exponent close to $2$. The power law functional form occurs at the critical point and remains unchanged in a broad interval of the impact velocity independent of the system size and of the impact angle. However, in the shattered phase attained in the limit of very high impact velocities, the fragment mass distribution tends to an exponential form. The results are in good quantitative agreement with the experimental findings on the fragmentation of plate-like objects [@glassplate; @platelike; @composite; @ice].
In applications like mineral processing, a fragment, formed with a certain velocity, can undergo secondary break-up due to collisions with other fragments or with the walls of the container. To get an estimate about the importance of this secondary fragmentation, the study of fragment velocities is essential. We determined the distribution of the velocity components of fragments and analysed the scaling behaviour of the distributions when changing the macroscopic variables of the system, [*i.e.*]{} impact velocity $v_0$ amd system size $R$. A very interesting anomalous scaling of the distribution functions were revealed with a power law dependence on $v_0$ and $R$.
In a variety of size reduction operations practiced by a wide range of industries, there is a particular interest in the net energy required to achieve a certain size reduction and the energy distribution of the fragments during the grinding process. To maximize efficiency of such processes, it is important to know the breakage characteristics of the grinding materials. Our current work can provide some of this valuable information.
[99]{}
N. Arbiter, C. C. Harris, and G. A. Stamboltzis, Transactions of the Society of Mining Engs. of AIME [**119**]{}, 244 (1969).
A. D. Salman, C. A. Biggs, J. Fu, I. Angyal, M. Szabo, M. Z. Hounslow, Powder Technology [**128**]{}, 36 (2002).
K. T. Chau, S. Z. Wu, W. C. Zhu, C. A. Tang, and T. X. Yu, in Proceedings of 16[*th*]{} ASCE Engineering Mechanics Conference, July 16-18, 2003, University of Washington, Seattle.
A. D. Salman, and D. A. Gorham, Powder Technology [**107**]{}, 179 (1999).
S. Data, and B. K. Mishra, M. Tech. Thesis, FFT Analysis of Charge Dynamics in Tumbling Mill (1998).
D. L. Turcotte, Jour. Geophys. Res. [**91 B2**]{}, 1921 (1986).
C. Thornton, K. K. Yin, and M. J. Adams, J. Phys. D [**29**]{}, 424 (1996).
B. K. Mishra, C. Thornton, Int. J. Mineral Processing [**61**]{}, 225 (2001).
F. Kun and H. J. Herrmann, Comput. Meth. Appl.Mech. Eng. [**138**]{}, 3 (1996).
F. Kun and H. J. Herrmann, Phys. Rev. E [**59**]{}, 2623 (1999).
F. Kun and H. J. Herrmann, Int. Jour. Mod. Phys. C [**7**]{}, 837 (1996).
G. A. D’Addetta, F. Kun, E. Ramm, and H. J. Herrmann, [*From solids to granulates - Discrete element simulations of fracture and fragmentation processes in geomaterials*]{}, Continuous and discontinuous modelling of cohesive-frictional materials, pp. 231-258 (2001), Lecture Notes in Physics (LNP) 568,Springer Verlag, Berlin.
L. Oddershede, P. Dimon, and J. Bohr, Phys. Rev. Lett., 3107 (1993).
A. Meibom and I. Balslev, Phys. Rev. Lett., 2492 (1996).
E. S. C. Ching, [*et al*]{}, Physica A [**265**]{}, 119 (1999).
A. Diehl [*et al*]{}, Phys. Rev. E [**62**]{}, 4742 (2000).
J. Åström, M. Kellomäki and J. Timonen, Phys. Rev. E [**55**]{}, 4757 (1997).
W. T. Ashurst and B. L. Holian, Phys. Rev. E [**59**]{}, 6742 (1999).
H. Inaoka, E. Toyosawa, and H. Takayasu, Phys. Rev. Lett. [**78**]{}, 3455 (1997).
J. Åström and J. Timonen, Phys. Rev. Lett. [**78**]{}, 3677(1997).
J. A. Aström, B. L. Holian, and J. Timonen, Phys. Rev. Lett. [**84**]{}, 3061 (2000).
M. Arakawa, Icarus [**142**]{}, 34 (1999).
F. Wittel, F. Kun, H. J. Herrmann, and B. H. Kröplin, Phys. Rev. Lett. [**93**]{}, 035504 (2004).
T. Matsui, T. Waza, K. Kani and S. Suzuki, J. of Geophysical Research [**87 B13**]{}, 10968 (1982).
A. Fujiwara and A. Tsukamoto, Icarus [**142**]{}, 44 (1980).
V. Latora, M. Belkacem, and A. Bonasera, Phys. Rev. Lett. [**73**]{}, 1765 (1994).
A. V. Potapov, C. S. Campbell, and M. A. Hopkins, Int. J. Mod. Phys. C [**6**]{}, 399 (1995).
H. Katsuragi, D. Sugino, and H. Honjo Phys. Rev. E [**68**]{}, 046105 (2003).
T. Kadono, Phys. Rev. Lett. [**78**]{}, 1444 (1997).
T. Kadono and M. Arakawa, Phys Rev. E [**65**]{}, 035107(R) (2002).
K. D. Kafui and C. Thornton, Powder Technology [**109**]{}, 113 (1999).
O. Tsoungui, D. Vallet, J. -C. Charmet, and S. Roux, Granular Matter [**2**]{}, 19 (1999).
M. Khanal, W. Schubert, and J. Tomas, Granular Matter [**5**]{}, 177 (2004).
R. Moreno, M. Ghadiri, and S. J. Antony, Powder Technology [**130**]{}, 132 (2003).
E. G. Kelly and D. J. Spottiswood, [ *Introduction to Mineral processing*]{}, Johns Wiley & Sons Inc. New York, 1982.
C. Moukarzel and H. J. Herrmann, Jour. Stat. Phys. [**68**]{}, 911 (1992).
H. J. Tillemans and H. J. Herrmann, Physica A [**217**]{}, 261 (1995).
A. V. Potapov and C. S. Campbell [*Fourth year progress report*]{}, International Fine Particle Research Institute , Dept. of Mech. Eng., University of southern California, USA.
B. Behera, F. Kun, S. MaNamara and H. J. Herrmann, cond-mat/0404057.
V. Buchholtz, J. A. Freund, T. Poschel, Eur. Phys. Jour. B [**16**]{}, 169 (2000). Thorsten Pöschel, [*Dynamik Granularer Systeme: Theorie, Experimente und numerische Experimente*]{}, Logos Verlag Berlin (2000).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present a comparative analysis of multifractal properties of financial time series built on stock indices from developing (WIG) and developed (S&P500) financial markets. It is shown how the multifractal image of the market is altered with the change of the length of time series and with the economic situation on the market. We emphasize that the proper adjustment of scaling range for multiscaling power laws is essential to obtain the multifractal image of time series. We analyze in this paper multifractal properties of real financial time series using Hölder $f(\alpha)$ representation and multifractal-DFA method. It is also investigated how multifractal properties of stocks change with variety of “surgeries” done on the initial real financial time series. This way we reveal main phenomena on the market influencing its multifractal dynamics. In particular, we focus on examining how multifractal picture of real time series changes when one cuts off extreme events like crashes or rupture points, and how fluctuations around the main trend in time series influence the multifractal behavior of financial series in the long-time horizon for both developed and developing markets.'
author:
- |
Łukasz Czarnecki, Dariusz Grech[^1] \
Institute of Theoretical Physics\
University of Wroc[ł]{}aw, PL-50-204 Wroc[ł]{}aw, Poland
title: Multifractal dynamics of stock markets
---
$$$$ **Keywords**: multifractality, econophysics, time series analysis, Hurst exponent\
**PACS:** 89.65.Gh, 89.75.Da, 89.20.-a
Introduction
============
Multifractality \[1-4\] is rather a new concept when applied to financial time series. It may be considered as the higher order extension of the monofractal analysis used successfully e.g. in analysis of persistency level in data – particularly in econophysics. So far it is not very clear what practical applications of multifractality in financial time series might be. However, some preliminary directions in this topic have already been shown (see e.g. \[5\]). In monofractal approach, one usually looks for the probabilistic (stochastic) behavior of some geometrical or topological properties of signal fluctuations around the local trend in given time series. One expects power-law relation between the quantity describing such fluctuations (e.g. variance) and the length of time window $s$ along which the fluctuation is being measured. A good example of such relation is the one provided in Detrended Fluctuation Analysis (DFA) \[6-11\] where the power law relation between the variance of detrended signal $F^2(s)$ and the width of time window $s$ reads $$F^2(s) \sim s^{2H}$$ with the Hurst scaling exponent $H$ ($0<H<1$) \[12,13\]. If $0<H<1/2$ the signal is said to be antipersistent, if $1/2<H<1$ the signal is persistent, while the case $H=1/2$ corresponds to no memory present in signal increments (Brownian motion).\
The relation given by Eq.(1) is usually understood as independence of scaling properties of a system from the scale – thus $H$ is constant. However, in many cases it might not be so. There might be a finite or infinite number of crossover points $s_X$ such that for time scales $s<s_X$ the fractal properties differ from these for $s>s_X$. In this case, the fractal structure will be described by the whole set of $H$ exponents instead of just one, thus revealing the multiple scaling rules. Using the standard DFA procedure one calculates only the leading scaling rule linked to major or more frequent fluctuations. Moreover, distortion of the signal from the leading pattern before or below the crossover point $s_X$ may be very weak, so in order to extract them, one is forced to use more sophisticated technique based on the artificial amplification of weak (with respect to the average ones) fluctuations. Such philosophy makes the background of the modified DFA proposed by Kantelhardt et. al. and called multifractal DFA (MF-DFA) \[14\]. One replaces in this method the “ordinary” fluctuation function $F^2(s)$ from Eq.(1) by its q-th moment $F^2_q(s)$ defined as follows $$F(s,q)=\{1/{2N}\sum^{2N}_k [F^2(s,k)]^{q/2}\}^{1/q}$$ for $q\neq 0$, and $$F(s,0)=\exp\{1/{4N}\sum^{2N}_k \ln[F^2(s,k)]\}$$ for $q=0$, where $N$ counts the number of non-overlapping boxes of size $s$ each for which detrendization procedure is performed.\
One obtains in this manner the whole continuous set of Hurst exponents $h(q)$ labeled by different $q\epsilon \bf{R}$. They are related to different level of amplification of small fluctuation in data. The $h(q)$ dependence is decreasing monotonic function of $q$ for stationary signal what basically reflects the fact that relatively small fluctuations happen more often in this signal than relatively big ones do \[14\]. If so, the so called Hölder spectrum or singularity spectrum $f(\alpha)$ \[4\] of the Hölder exponent $\alpha$ ca be used where $$\alpha = \frac{ds(q)}{dq}$$ and $$s(q)=qh(q)-1$$ while $f(\alpha)$ is determined by Legendre transform \[16,17\] $$f(\alpha)=q(\alpha)\alpha-s(q(\alpha))$$ The latter approach makes a useful representation of multifractality. The singularity spectrum in this case should have rather regular form of inverse parabolic shape as shown in Fig.2. The width of $f(\alpha)$ spectrum measures the multifractality level of the signal. For the pure monofractal signal this width converges to one point $\alpha=H=h(2)$ and $f(\alpha)=1$, while if the bigger amount of multifractality is present in the signal, the width of $f(\alpha)$ spectrum becomes wider.\
We will analyze in this paper multifractal properties of real financial time series using $f(\alpha)$ representation and MF-DFA method argued as working better than other approaches \[15\]. Our main task is to investigate how multifractal properties of stocks change with variety of “surgeries” done on the initial real financial time series. We shall do this in following chapters. Our aim is to see what phenomena on the market influence its multifractal dynamics in the first place. In particular, we will focus on examining how multifractal picture of real time series is changed when one cuts off extreme events like crashes or rupture points, and how established trends or their absence influence the multifractal behavior of developed and developing markets.
Multifractal noise in monofractal signal of finite length
=========================================================
All statistical methods determining the fractal properties of time series give an exact result only for series of infinite length. Since this applies also to the MF-DFA scheme, one should know how the length of series is an important factor in discovering its multifractal structure. In fact, finite signals assumed by their construction to be monofractals, reveal some artificial multifractal structure. This is because big fluctuations in finite time series, measured within MF-DFA, seem to be more rare than for longer or infinite time series. It leads to smaller $h(q)$ value for $q\rightarrow\infty$ in finite series than for infinite ones, and finally, to some artificial multifractal structure of finite time series. This kind of multifractal noise should be subtracted first as a background influencing any real property of monofractal or multifractal time series of finite length.\
To explore this issue we must have a ”test series“ with known fractal characteristic. We will use monofractal series generated by the random midpoint displacement (RMD) algorithm \[22\]. The resulting h(q) functions calculated within MF-DFA for three series of length $L=2^{14}$ generated with input Hurst exponent values $H=0.3, 0.5, 0.7$ respectively are presented in Fig. 1.\
One may notice a small degree of multifractality (short span of $h(q)$) for all three series. The series with higher $H$ show a bigger deviation from monofractality. The non-monotonic behaviour of $h(q)$ is reflected by the ”twist“ on top of the singularity spectrum (see Fig. 2). This effect will be also observed in following chapters where real financial data are analyzed.\
One can also see that the maxima positions of the spectra reflect the assumed Hurst exponents values.\
It is essential to find the influence of series length on the width of multifractal spectrum. To observe this, one has to calculate the highest and lowest value of $\alpha$, as well as the position of the $f(\alpha)$ peak. This was done on a sample of 20 series each for four lengths $L=2^8, 2^{10}, 2^{12}, 2^{14}$ and for different $H$ values. The width of $f(\alpha)$ spectrum was rapidly changing with $L$ and seemed to be weakly dependent on $H$. Table 1 presents results found for $H=0.5$.
[|C|C|C|]{} Series length $L$ & Spectrum width $\Delta\alpha$ & Peak position\
$2^{8}$ & 0.73 & 0.56\
$2^{10}$ & 0.28 & 0.50\
$2^{12}$ & 0.21 & 0.50\
$2^{14}$ & 0.22 & 0.51\
\
They confirm the significant narrowing of $\Delta\alpha$ spectrum width when the length $L$ of time series is increasing. Since we will deal later on with series lengths around $L\sim 2^{14}$, we may assume that the finite size effects should be around $\delta_{finite}(\Delta\alpha)\sim 0.20$.\
Before analyzing real data we also need to observe the scaling range of the $F(s,q) \sim s^{h(q)}$ relation. For the artificial data, this range spans almost the entire available time window lengths (see Fig. 3). However, for the entire set of real data (later sections will also discuss parts of these data), the scaling range becomes shorter and decreases with increasing $\left|q\right|$ values.\
The corresponding effect for the S&P500 is visible in Fig. 4. It is even more significant for positive and larger $q$ what is shown in Fig. 5 (compare with Fig. 4). To solve this problem a careful separate study of all $F(s,q) \sim s^{h(q)}$ relations and manual selection of the scaling range is needed. It was done in analysis of all financial data presented below.
Long-term data analysis
=======================
The first and most obvious approach is to make the long-term data analysis, i.e., to calculate the multifractal structure of the stock market indices in the whole available time horizon. Our analysis focuses on two indices:
- WIG (Warsaw Stock Exchange)) - from April 16$^{th}$ 1991 till October 10$^{th}$ 2008
- S&P500 (NYSE & Nasdaq) - from December 30$^{th}$ 1927 till September 3$^{rd}$ 2008
They are examples of developing (Poland) and developed (USA) markets respectively.\
All time series were created from daily closing values of each index. The S&P 500 data were downloaded from the financial web-site of Yahoo.com \[18\] and WIG data from a stock market web-site of Wirtualna Polska \[19\].\
The singularity spectra $f(\alpha)$ and $h(q)$ plots of all discussed time series (as well as their shuffles) are presented in Figs 6, 7. Surprisingly, all markets show unexpected non-monotonic $h(q)$ behavior. This phenomenon, in addition to the problems with short scaling ranges (mentioned in the previous section), strongly suggests a disturbance in the multifractal structure of the series. Spectra of all three shuffled data sets are positioned around the expected location $\alpha = 0.5$. These sets of data also show a monotonic behavior of the $h(q)$ function, contrary to originally ordered data. The spectra width for original as well for shuffled data are much wider than those of the corresponding RMD spectra and hence indicate the multifractal content in dynamics of long-term market data. Simultaneously, the non-monotonic $h(q)$ behavior seems to indicate non-stationary effects in these data. The latter case needs more detailed study partly provided below.
Multifractal analysis of financial data in established trends
=============================================================
It has been shown by a number of authors (see e.g.\[5,20,21\]) that positive and negative fluctuations in time series have different fractal properties. Oświȩcimka et. al. has shown in Ref.\[5\] that DAX data in increasing and decreasing trends has different multifractal properties for particular choice of periods Dec.1997-Dec.1999 and May 2004-May 2006. It is quite natural to investigate whether these properties are general and if so, in what extent they are general for other markets. In particular, it is interesting to know how multifractal properties of established trends depend on the internal structure of these trends. We were interested in the beginning in positive (negative) long-lasting trends with no sub-trends in opposite direction. We tried to exactly repeat this way the analysis made for DAX in Ref.\[5\]. Such trends turned out to be very rare for S&P 500 and WIG indices. The selected fragments of both indices with required properties are shown in Fig. 8. They have both long-lasting positive trend (bullish phase) followed by a long negative trend (bearish phase) and no astonishing internal events.\
The resulting singularity spectra (see Fig.9) show the expected shift of the bearish phase spectra towards higher values of $\alpha$. Since these results were obtained from data with a much wider scaling range, they should be considered as much more reliable then previous ones presented for entire indices. It is also worth noting that the spectra are now much wider, which suggest a more complex multifractal structure. In the case of WIG this effect might be amplified by the short data set, nevertheless the spectra width of a RMD series with similar length is still narrower.\
This improvement of singularity spectra for well determined trends in both indices (i.e. quasi quadratic dependence $f(\alpha)$ from $\alpha$) with respect to spectra obtained for the entire indices might suggest that the multifractality is strongly affected by non-stationarity of financial time series. To verify such hypothesis we checked what multifractal properties are hidden in subparts of entire index. The S&P500 index is particularly good for this search because it is long enough comparing with WIG, thus containing much more amount of data. If one divides the entire S&P500 history into two regimes - one with small fluctuations (small volatility) and the second one with high fluctuations (high volatility) as shown in Fig. 10, a very intriguing result for singularity spectra is found in both regimes. Results of this search are plotted in Figs.11, 12, where a comparison of singularity spectra and $h(q)$ dependence for both regimes with $f(\alpha)$ and $h(q)$ behavior for the entire index is provided. It is clear that both regimes show separately a monotonic $h(q)$ dependence leading to reversed parabolic shape of $f(\alpha)$ . It is not the case of the entire index.\
However, both spectra for two regimes are slightly different – the width of multifractal spectrum (and the width of $h(q)$ dependence) is larger for the first regime characterized by lower volatility and no opposite sub-trends inside. We checked the same happened for other, even shorter trends, whenever they are affected by well formed sub-trends of opposite direction or by extreme events like crashes, etc. This phenomena is shown on example of S&P500 time series without and with internal structure (see Figs.13, 14). Fig.13 presents an example of increasing trend with no visible distortion inside. We called it “good” data. Its dynamics is associated with nice multifractal properties - well established quadratic $f(\alpha)$ dependence and the monotonic $h(q)$ behaviour. Simultaneously, time series of the same length, but collecting data from the very next period, has some extreme events inside and it contains opposite sub-trends connected with these events (see Fig.14). We call them “bad” data. The scaling $F_q(s)\sim s^{h(q)}$ is worse here and the presence of “twist” in singularity spectrum $f(\alpha)$ confirms the non-monotonic character of $h(q)$ function and leads also to non-quadratic $f(\alpha)$ dependence. The multifractal character of time series is here very much affected by the presence of abrupt events, i.e. crashes or data with respectively large fluctuations.
Influence of extreme events on multifractal dynamics of stocks
==============================================================
The sub-trends containing abrupt events can be artificially removed to see if such removal would improve the multiscaling properties of the new trend constructed after such “surgery”. The applied procedure is simple. First, the time series with abrupt events is differentiated. Then, at the level of returns, these events are removed and the time series of remaining returns is integrated to get the new artificial evolution of financial index. Examples of such modification are visible for variety of events for both S&P 500 and WIG indices in Figs.15, 16 and 17 respectively. It is remarkable, how much such surgery significantly improves multi-scaling properties of financial time series and restores the monotonic character of decreasing $h(q)$ function. The width of singularity spectrum has also noticeably increased when abrupt events had been removed.
Conclusions
===========
We presented a comparative analysis of multifractal properties of financial time series built from stock indices of developing (WIG) and emergent (S&P500) markets using MF-DFA technique. We found that financial time series in a very long time horizon show in both cases very short multiscaling ranges and in turn, their multifractal image is somehow obscured. Therefore we analyzed in shorter time horizon the selected parts of financial data chosen due to their specific properties.\
The most important result is that division of time series into regimes of distinctly different (large or small) fluctuations around the main trend, as well as removal of abrupt events like crashes or significant opposite sub-trends, radically improves the multiscaling ranges and restores the monotonic behavior of $h(q)$ dependence as well as the reversed parabolic shape of multifractal singularity spectrum $f(\alpha)$. The width of singularity spectrum also increases noticeably in these cases. This leads us to conclusion that non-stationarity of financial time series significantly influences their multifractal character leading to behavior beyond our expectation for multifractal spectrum shape known from studies with artificially created mono- and multifractal signals.\
It is clear that further research of length effects as well as the effects of non-stationarity in real time series influencing multiscaling and multifractality is needed with hope for further practical applications.\
[99]{}
J.W. Kantelhardt, arXiv: 0804.0747v1 \[physics.data-an\] Z. Eisler, J. Kertész, Physica A 343 (2004) 603 H.G.E. Hentschel, I. Procaccia, Physica D 8 (1983) 435 T.C Halsey, M.H. Jensen, L.P. Kadanoff, I. Procaccia, B.I. Shraiman, Phys. Rev. A33 (1983) 1141 P. Oświȩcimka, J. Kwapień, S. Drożdż, A.Z. Górski, R. Rak, Acta Phys. Pol. A 114 (2008) 547 J.W. Kantelhardt, E. Koscielny-Bunde, H.H.A. Rego, S. Havlin, A. Bunde, Physica A 295 (2001) 441 C.-K. Peng, S.V. Buldyrev, S. Havlin, M. Simons, H.E. Stanley, A.L. Golberger, Phys. Rev. E 49 (1994) 1685 N. Vandewalle, M. Ausloos, Physica A 246 (1997) 454 M. Ausloos, N. Vandewalle, Ph. Boveroux, A. Minguet, K. Ivanova, Physica A 274 (1999) 229 G. M. Viswanathan, C.-K. Peng, H.E. Stanley, A.L. Goldberger, Phys. Rev. E 55 (1997) 845 K. Hu, P. Ch. Ivanov, Z. Chen, P. Carpena, H. E. Stanley, Phys. Rev. E 64 (2001) 011114 H.E. Hurst, Trans. Am. Soc. Civ. Eng. 116 (1951) 770 B.B. Mandelbrot, J.R. Wallis, Water Resour. Res. 5(2) (1969) 321 J.W. Kantelhardt, S.A. Zschiegner, E. Koscielny-Bunde, S. Havlin, A. Bunde, H.E. Stanley, Physica A 316 (2002) 87 P. Oświȩcimka, J. Kwapień, S. Drożdż, A.Z. Górski, R. Rak, Acta Phys. Pol. B 36 (2005) 2447 F. Feder, Fractals (Plenum Press, New York 1988) H.-O. Peitgen, H. Jürgens, D. Saupe, Chaos and fractal (Springer, Berlin, 2004) http://finance.yahoo.com/ http://gielda.wp.pl/ K. Ohashi, L.A.N Amaral, B.H. Natelson, Y. Yamamoto, Phys. Rev. E 68 (2003) 065204(R) N.B. Ferreira, R. Menezes, D.A. Mendes, Physica A 382 (2007) 73 R.H. Voss, Random fractal forgeries, in: R.A. Earnshaw (ed.), Fundamental Algorithm for Computer Graphics, NATO ASI Series, Vol. F17, Springer, Berlin, Heidelberg, 1985
{width="\columnwidth"}
{width="\columnwidth"}
{width="\columnwidth"}
{width="\columnwidth"}
{width="\columnwidth"}
{width="\columnwidth"} {width="\columnwidth"}
{width="\columnwidth"} {width="\columnwidth"}
{width="\columnwidth"}
{width="\columnwidth"} {width="\columnwidth"}
{width="\columnwidth"}
{width="\columnwidth"}
{width="\columnwidth"}
{width="\columnwidth"}
{width="\columnwidth"}
{width="\columnwidth"}
{width="\columnwidth"}
{width="\columnwidth"}
[^1]: [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We have derived oxygen and nitrogen abundances of a sample of late-type, low surface brightness (LSB) galaxies found in the Sloan Digital Sky Survey (SDSS). Furthermore, we have computed a large grid (5000 models) of chemical evolution models (CEMs) testing various time-scales for infall, baryon densities and several power-law initial mass functions (IMFs) as well. Because of the rather stable N/O-trends found both in CEMs (for a given IMF) and in observations, we find that the hypotheses that LSB galaxies have stellar populations dominated by low-mass stars, i.e., very bottom-heavy IMFs (see Lee et al. 2004), can be ruled out. Such models predict much too high N/O-ratios and generally too low O/H-ratios. We also conclude that LSB galaxies probably have the same ages as their high surface brightness counterparts, although the global rate of star formation must be considerably lower in these galaxies.'
author:
- 'Lars Mattsson, Brady Caldwell, and Nils Bergvall'
title: The Chemical Evolution of LSB Galaxies
---
Late-type LSBGs in the SDSS
===========================
We use the data compilation by Caldwell & Bergvall (2006) consisting of a sample of 1199 close to edge-on ($b/a\leq 0.25$) low surface brightness (LSB) galaxies in the fourth data release of the Sloan Digital Sky Survey (SDSS-DR4). All galaxies in the sample were selected in order to obtain a sample of isolated, bulge-less systems, which is ideal for comparison with simple CEMs since the merger histories of such galaxies have less influence on the evolution of elemental ratios. This total sample is divided by $g-r$ colour limits into a blue Sample A (377 galaxies), a green (intermediate) Sample B (436 galaxies) and a red Sample C (386 galaxies). From these 1199 galaxies we pick out the ones with a relative error in the H$\alpha$-flux less than $25\%$.
Numerical Model Grid
====================
We have computed a large grid (5000 models) of one-zone CEMs, testing time-scales for infall from $\tau_{\rm inf} = 1.0$ Gyr to $\tau_{\rm inf} = 7.9$ Gyr, baryon densities ranging between $\Sigma_{\rm final} = 1.0 M_\odot$ pc$^{-2}$ and $\Sigma_{\rm final} = 6000.0 M_\odot$ pc$^{-2}$, which should cover most of the possible variations within dwarf and late-type galaxies. Star formation is prescribed by a simple Schmidt-law, $\dot{\Sigma}_\star = \eta\,\Sigma_{\rm gas}^{1.5}$, where the constant $\eta$ is varied between $\eta = 0.001$ to $\eta = 0.2$. Furthermore, we have tried several power-law initial IMFs $\phi(m)\sim m^{-(1+x)}$ in order to test the bottom-heavy IMF hypothesis ($x=2.85$) suggested by Lee et al. (2004). The nucleosynthesis prescriptions are taken from Chieffi & Limongi (2004) for high mass stars and van den Hoek & Groenewegen (1997) for low and intermediate stars and all model tracks where evolved over 13.7 Gyr (one Hubble-age).
![ \[models\] Observed abundances and model results (grey dots) for different IMFs. Slowly evolving tracks populate the $\log(\rm{N/O})$-plateau, while rapidly evolving tracks populate the “secondary branch”. The abundances of the late-type LSBGs are shown as blue diamonds (Sample A), green triangles (Sample B) and red squares (Sample C). Black dots and circles show data for dwarf LSBGs from Rönnback & Bergvall (1995), van Zee et al. (1996) and van Zee & Haynes (2006). ](mattsson_fig1a "fig:"){width="4.33cm"} ![ \[models\] Observed abundances and model results (grey dots) for different IMFs. Slowly evolving tracks populate the $\log(\rm{N/O})$-plateau, while rapidly evolving tracks populate the “secondary branch”. The abundances of the late-type LSBGs are shown as blue diamonds (Sample A), green triangles (Sample B) and red squares (Sample C). Black dots and circles show data for dwarf LSBGs from Rönnback & Bergvall (1995), van Zee et al. (1996) and van Zee & Haynes (2006). ](mattsson_fig1b "fig:"){width="4.33cm"} ![ \[models\] Observed abundances and model results (grey dots) for different IMFs. Slowly evolving tracks populate the $\log(\rm{N/O})$-plateau, while rapidly evolving tracks populate the “secondary branch”. The abundances of the late-type LSBGs are shown as blue diamonds (Sample A), green triangles (Sample B) and red squares (Sample C). Black dots and circles show data for dwarf LSBGs from Rönnback & Bergvall (1995), van Zee et al. (1996) and van Zee & Haynes (2006). ](mattsson_fig1c "fig:"){width="4.33cm"}
Results and Conclusions
=======================
In Fig. 1 we show a comparison between our abundances derived from spectral data from SDSS-DR4 using the so-called P-method (see, e.g., Pilyugin 2003) and our model results. Clearly, the late-type LSBGs do not deviate from the general, rising trend found in HSB galaxies. Based on the results from our model grid we draw the following conclusions:\
[ - The $\log({\rm N/O})$-ratios in dwarf irregular and late-type LSBGs follow the same basic pattern as their HSB counterparts.\
- The chemical evolution, as revealed by the HII-regions, is not compatible with an extremely bottom-heavy IMF as suggested by Lee et al (2004), nor is it compatible with a top-heavy IMF.\
- The LSB property of these objects is most likely due to a low global star formation density, although the efficiency of star formation in star-forming regions must be as high as in HSBGs in order to explain the high $\log(\rm{N/O})$-ratios.\
- Given these results, it is unlikely (although not impossible) that the ages of LSBGs are significantly different from the ages of corresponding HSBGs. ]{}
Caldwell B. & Bergvall N., 2006, in the proceedings of the IAU Symposium no. 235 Chieffi A. & Limongi M., 2004, , 608, 405 van den Hoek L.B. & Groenewegen M.A.T., 1997, , 123, 305 Lee H-C., Gibson B.K., Flynn C., Kawata D. & Beasley M.A., 2004, , 353, 113 Pilyugin L.S., Thuan T.X. & Vilchez J.M., , 2003, 397, 487 Rönnback J. & Bergvall N., 1995, , 302, 353 van Zee L., Haynes M. & Salzer J.J., 1997, , 114, 2479 van Zee L. & Haynes M., 2006, , 636, 214
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The CMS experiment uses missing $E_\text{T}$ to both measure processes in the Standard Model and test models of physics beyond the Standard Model. These proceedings show the performance of the missing $E_\text{T}$ reconstruction evaluated by using 4.6 fb${}^{-1}$ of proton-proton collision data at the center-of-mass energy $\sqrt{s}$ = 7 TeV collected in 2011 with the CMS detector at the Large Hadron Collider. Missing $E_\text{T}$ was reconstructed based on a particle-flow technique. Jet energy corrections were propagated to missing $E_\text{T}$. After anomalous signals and events were addressed, the missing $E_\text{T}$ spectrum was well reproduced by MC simulation. The multiple proton-proton interactions in a single bunch crossing, pile-up events, degraded the performance of the missing $E_\text{T}$ reconstruction. Mitigations of this degradation have been developed.'
address: |
Mitchell Institute for Fundamental Physics and Astronomy\
Department of Physics & Astronomy, Texas A&M University, College Station, TX 77843, USA
author:
- Tai Sakuma for the CMS collaboration
bibliography:
- 'proceedings.bib'
title: 'Missing $E_\text{T}$ Reconstruction with the CMS Detector'
---
=1
110
= 10000 = 10000
Introduction
============
In the CMS experiment, *missing* $E_\text{T}$ (MET) in proton-proton collisions is reconstructed and used in a wide range of physics analyses. MET is the imbalance in the transverse momentum of all *visible particles*, particles which interact with the *electromagnetic* or *strong forces*, in the final state of proton-proton collisions. Because momentum is conserved in each direction, MET is the transverse momentum that must have been carried by something *invisible*. *Neutrinos*, for example, are invisible particles; therefore, MET is an estimate of transverse momentum of neutrinos. We use MET in measurements of *W bosons*, *top quarks*, and *tau leptons* as these particles can decay into neutrinos. Further, many models of *physics beyond the Standard Model* predict the existence of particles or something else which are invisible and can carry momentum; e.g., *Dark Matter* models, *supersymmetric* models, *unparticle* models, and models with *large extra dimensions*. For this reason, we use MET to test such models.
Accurate reconstruction of MET is demanding because it entails reconstruction of all visible particles in an event with precision. This requires a *hermetic* detector which can detect all particles which electromagnetically or strongly interact with matter. The CMS detector, located at one of two *high luminosity interaction points* of the *Large Hadron Collider* (LHC), meets this requirement. The subsystems of the CMS detector include *highly granular electromagnetic calorimeters*, *hermetic hadronic calorimeters*, *redundant muon systems*, and *all silicon trackers* in a *strong magnetic field*. A thorough description of the CMS detector can be found in Ref. [@Chatrchyan:2008aa].
Based on 36 pb${}^{-1}$ of data collected at the center-of-mass energy $\sqrt{s}$ = 7 TeV in 2010, the CMS collaboration published the results of comprehensive studies of the MET reconstruction performance [@Chatrchyan:2011tn]. In 2011, the CMS detector collected a considerably larger amount of data. These proceedings summarize preliminary results of the MET reconstruction performance study based on the data collected in 2011 [@CMS-DP-2011-010; @CMS-DP-2012-003].
With the increase in the LHC luminosity, the number of multiple proton-proton interactions in the same *bunch crossing* (*in-time pile-up*) and overlapping detector signal from previous or following bunch crossings (*out-of-time pile-up*) significantly increased. The pile-up events worsen the performance of the MET reconstruction. We have developed several techniques to mitigate the effect of pile-up events.
The 7 TeV proton-proton collision run in 2011
=============================================
These proceedings use 4.6 fb${}^{-1}$ of proton-proton collision data collected with the CMS detector at the center-of-mass energy $\sqrt{s}$ = 7 TeV in 2011. We divided the data collection period into two *eras*: Run 2011A and Run 2011B. For proton-proton collisions, Run 2011A started in March 2011 and ended in August 2011; Run 2011B started in September 2011 and ended in October 2011. Of the 4.6 fb${}^{-1}$ of data, 2.0 fb${}^{-1}$ was collected in Run 2011A and 2.6 fb${}^{-1}$ was collected in Run 2011B. Since the LHC luminosity rapidly increased over the year, Run 2011B was shorter than Run 2011A but more data were collected. In addition, the number of pile-up events also increased. The average number of the reconstructed *vertices* in a single bunch crossing, which indicates the average number of in-time pile-up events, increased from 5.5 in Run 2011A to 9.2 in Run 2011B. The distributions of the number of the reconstructed vertices in a single bunch crossing are shown in Figure \[143737\_22Jul12\].
MET reconstruction algorithms and corrections
=============================================
MET used in these proceedings is based on a *particle-flow algorithm* and includes the *Type-I* MET *correction*, a propagation of *jet energy corrections*. This MET, both with and without the correction, is called *particle-flow* MET (pfMET) in Ref. [@Chatrchyan:2011tn]. Ref. [@Chatrchyan:2011tn] contains more detailed description of pfMET and the Type-I MET correction as well as other definitions of MET and corrections to MET.
First, *raw* MET was defined as the imbalance in the *transverse energy*[^1] of all particles in the event reconstructed by a particle-flow algorithm. The detail of the particle-flow algorithm at CMS can be found in Ref.[@CMS-PAS-PFT-09-001]. In short, the particle-flow algorithm uses information from all CMS detector subsystems, i.e., trackers, calorimeters, and muon systems; then, it reconstructs four momenta of all visible particles, each of which is identified as one of five particle types, i.e., *electrons*, *photons*, *muons*, *charged hadrons*, and *neutral hadrons*.
The raw MET is systematically different from *true* MET, i.e., the transverse momentum carried by invisible particles, for many reasons including the non-compensating nature of the calorimeters. To make MET a better estimate of true MET, the Type-I MET correction was applied. The Type-I MET correction replaces the transverse energy of particles which can be clustered as jets with the transverse momentum of the jets to which jet energy corrections are applied. The detail of jet energy corrections at CMS can be found in Ref. [@Chatrchyan:2011ds].
False MET and event cleaning
============================
Large MET is caused not only by interesting physics processes in proton-proton collisions such as production of invisible particles. In fact, large MET has more often uninteresting causes such as detector noise, *cosmic rays*, and *beam-halo* particles. MET with uninteresting causes is called *false* MET or *anomalous* MET. For an accurate reconstruction of MET, it is, therefore, not sufficient to reconstruct all visible particles produced in proton-proton collisions.
We developed several algorithms to identify false MET. These algorithms, for example, use timing, pulse shape, and topology of signal. After the identified false MET was removed, the agreement of the MET spectrum with simulation, in which causes of false MET were not explicitly simulated, significantly improved (Figure \[135414\_22Jul12\]).
Performance of MET reconstruction
==================================
MET *response* and *resolution* are measures of MET reconstruction performance. We evaluated them by artificially inducing MET in events very likely to have only little or no true MET. We used events with *vector bosons*, either *Z bosons* decaying into *dimuons* or photons. These vector bosons are predominantly produced in interactions with no true MET, such as $qg \rightarrow
q\gamma$, $q\bar{q} \rightarrow Z$, $qg \rightarrow qZ$, $q\bar{q}
\rightarrow gZ$. Therefore, an event is primarily composed of a vector boson and its *hadronic recoil*. The dimuons and photons were measured with precision by, respectively, the trackers and muon systems, and the electromagnetic calorimeters. To induce MET for the performance evaluation, we excluded the dimuons or photons from the MET reconstruction; the dimuons or photons played the role of invisible particles. Then, the MET performance can be indicated by how close reconstructed MET is to the transverse momentum of the excluded dimuons or photons.
[r]{}[7.0cm]{} 
The transverse momentum of the vector boson is $\qT$. The vector sum of the transverse energy of all reconstructed particles except the vector boson, i.e. the hadronic recoil, is $\uT$. Thus, the induced MET is $\vecMET=-\uT$. The projection of $\uT$ on $\qT$ is $\upar$, which is typically negative. The component of $\uT$ perpendicular to $\qT$ is $\uper$. These kinematic variables are illustrated in Figure \[104058\_27Jul12\].
MET response is defined as $-\upar/|\qT|$, the ratio of the MET component parallel to the transverse momentum of the vector boson and the magnitude of the transverse momentum of the vector boson. Figure \[105714\_27Jul12\] shows the MET response in events with Z bosons decaying into dimuons. The response is close to unity at above certain $|\qT|$ and has little dependence on the number of pile-up events.
MET resolutions are defined as the RMS of $\upar$ and $\uper$ about their mean values, denoted, respectively, as $\textrm{RMS}(\upar)$ and $\textrm{RMS}(\uper)$. Figure \[110832\_27Jul12\] shows the MET resolutions in Z events as a function of the number of reconstructed vertices for data collected in Run2011A and Run2011B. Figure \[110911\_27Jul12\] shows the MET resolutions as functions of $|\qT|$ in photon events for four different numbers of reconstructed vertices for data collected in early Run2011A. The resolutions degrade as the number of reconstructed vertices (in-time pile-up events) increases. The resolutions are worse in events for Run2011B than events with the same number of in-time pile-up events for Run2011A because the out-of-time pile-up increased.
 
Mitigation of MET degradation in high pile-up events
====================================================
This section briefly introduces one of the techniques that CMS used to mitigate the degradation of the MET performance caused by the pile-up events.
As shown in the previous section, the MET performance becomes worse as the number of pile-up events increases. Then, when MET is used in event selections in physics analyses, the efficiency of selecting signal events decreases. In order to prevent the efficiency from decreasing, an analysis of the search for the *Higgs boson* decaying into a pair of W bosons [@Chatrchyan:2012ty] defined another kinematic variable similar to MET, which is called *TrackMET* in Ref.[@CMS-DP-2012-003]. TrackMET is the imbalance in the transverse momentum of charged particles originating from the primary vertex of the high-$\pT$ event. By its construction, it depends little on the number of pile-up events. It was shown that using a combination of TrackMET and MET improved the signal efficiency in this analysis [@Chatrchyan:2012ty].
Summary
=======
MET is an estimate of the transverse momentum of neutrinos and other invisible particles or some stuff with momentum. In the CMS experiment, MET plays a central role in both precision measurements of Standard Model physics and searches for physics beyond the Standard Model.
Using 4.6 fb${}^{-1}$ of proton-proton collision data at the center-of-mass energy $\sqrt{s}$ = 7 TeV collected with the CMS detector in 2011, we evaluated the performance of the reconstruction of MET, which was based on a particle-flow algorithm, and in particular how it was affected by pile-up events. After false MET was removed, the MET spectrum was well described by MC simulation. While the MET response exhibited little dependence on the number of pile-up events, the MET resolutions became worse as the number of pile-up events increased. We have developed several techniques to mitigate the effect of the pile-up events. As the LHC luminosity is expected to keep rapidly increasing, it will be important to continue to develop and refine techniques to handle a large number of pile-up events.
In 2012, the CMS detector is collecting proton-proton collision data at the center-of-mass energy $\sqrt{s}$ = 8 TeV at the increasing luminosity. MET will remain an important object to be reconstructed for a variety of physics analyses at the CMS experiment.
References {#references .unnumbered}
==========
[^1]: We will use the transverse momentum instead of the transverse energy in the near future.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We establish the existence of multi-bump solutions for the following class of quasilinear problems $$- \Delta_{ p(x) } u + \big( \lambda V(x) + Z(x) \big) u ^{ p(x)-1 } = f(x,u) \text{ in } \mathbb R^N, \, u \ge 0 \text{ in } \mathbb R^N,$$ where the nonlinearity $ f \colon \mathbb R^N \times \mathbb R \to \mathbb R $ is a continuous function having a subcritical growth and potentials $ V, Z \colon \mathbb R^N \to \mathbb R $ are continuous functions verifying some hypotheses. The main tool used is the variational method.'
author:
- |
Claudianor O. Alves[^1] and Marcelo C. Ferreira [^2]\
Universidade Federal de Campina Grande\
Unidade Acadêmica de Matemática\
CEP:58429-900, Campina Grande - PB, Brazil.
title: ' Multi-bump solutions for a class of quasilinear problems involving variable exponents [^3]'
---
[ 35A15, 35B09, 35B40, 35H30.]{}
[ Variational Methods, Positive solutions, Asymptotic behavior of solutions, $p(x)$-Laplacian ]{}
Introduction
============
In this paper, we considered the existence and multiplicity of solutions for the following class of problems $$\big( P_\lambda \big) \;
\begin{cases}
- \Delta_{ p(x) } u + \big( \lambda V(x) + Z(x) \big) u^{ p(x)-1 }= f(x,u), \text{ in } \mathbb R^N, \\
u \ge 0, \text{ in } \mathbb R^N, \\
u \in W^{ 1,p(x) } \big( \mathbb R^N \big),
\end{cases}$$ where $ \Delta_{ p(x) } $ is the $ p(x) $-Laplacian operator given by $$\Delta_{ p(x) } u = \text{div} \left( \big| \nabla u \big|^{p(x)-2} \nabla u \right).$$ Here, $ \lambda > 0 $ is a parameter, $ p \colon \mathbb R^N \to \mathbb R $ is a Lipschitz function, $ V,Z \colon \mathbb R^N \to \mathbb R $ are continuous functions with $ V \ge 0 $, and $ f \colon \mathbb R^N \times \mathbb R \to \mathbb R $ is continuous having a subcritical growth. Furthermore, we take into account the following set of hypotheses:
1. $ 1 < p_- \le p_+ < N $.
2. $ \Omega = \text{int } V^{ -1 } (0) \ne \emptyset $ and bounded, $ \overline{\Omega} = V^{ -1 }(0) $ and $ \Omega $ can be decomposed in $ k $ connected components $ \Omega_1, \ldots, \Omega_k $ with $ \text{dist} \big( \Omega_i, \Omega_j \big) > 0, \, i \ne j $.
3. There exists $ M > 0 $ such that $$\lambda V(x) + Z(x) \ge M, \, \forall x \in \mathbb R^N, \lambda \ge 1.$$
4. There exists $ K > 0 $ such that $$\big| Z(x) \big| \le K, \, \forall x \in \mathbb R^N.$$
5. $$\limsup_{ |t| \to \infty } \frac{|f(x,t)|}{|t|^{ q(x)-1 }} < \infty, \text{ uniformly in } x \in \mathbb R^N,$$ where $ q \colon \mathbb R^N \to \mathbb R $ is continuous with $ p_+ < q_- $ and $ q \ll p^* $.
6. $ f(x,t) = o \big( |t|^{ p_+ - 1} \big), \, t \to 0, \text{ uniformly in } x \in \mathbb R^N $.
7. There exists $ \theta > p_+ $ such that $$0 < \theta F(x,t) \le f(x,t) t, \, \forall x \in \mathbb R^N, t > 0,$$ where $ F(x,t) = \int_0^t f(x,s) \, ds $.
8. $ \dfrac{f(x,t)}{t^{p_+ - 1}} $ is strictly increasing in $ (0,\infty) $, for each $ x \in \mathbb R^N $.
9. $ \forall a, b \in \mathbb R, \, a < b, \, \displaystyle \sup_{ x \in \mathbb R^N \atop t \in [a,b] } |f(x,t)| < \infty $.
A typical example of nonlinearity verifying $ (f_1)-(f_5) $ is $$f(x,t) = |t|^{ q(x)-2 }t, \, \forall \, x \in \mathbb R^N \, \mbox{ and } \, \forall t \in \mathbb R,$$ where $ p_+ < q_- $ and $ q \ll p^* $.
Partial differential equations involving the $ p(x) $-Laplacian arise, for instance, as a mathematical model for problems involving electrorheological fluids and image restorations, see [@Acerbi1; @Acerbi2; @Antontsev; @CLions; @Chen; @Ru]. This explains the intense research on this subject in the last decades. A lot of works, mainly treating nonlinearities with subcritical growth, are available (see [@Alves2; @Alves3; @AlvesBarreiro; @AlvesSouto; @AlvesFerreira; @AlvesFerreira1; @Fan1; @Fan; @FanZhao0; @FZ1; @FanShenZhao; @BSS; @FuZa; @MR] for interesting works). Nevertheless, to the best of the author’s knowledge, this is the first work dealing with multi-bump solutions for this class of problems.
The motivation to investigate problem $ \big( P_\lambda \big) $ in the setting of variable exponents has been the papers [@Alves] and [@DingTanaka]. In [@DingTanaka], inspired by [@DelPinoFelmer] and [@Sere] the authors considered $ \big( P_\lambda \big) $ for $ p = 2 $ and $ f(u) = u^q $, $ q \in \big(1, \frac{N+2}{N-2} \big) $ if $ N \ge 3 $; $ q \in (1, \infty) $ if $ N = 1, 2 $. The authors showed that $ \big( P_\lambda \big) $ has at least $ 2^k-1 $ solutions $u_\lambda$ for large values of $ \lambda $. More precisely, one solution for each non-empty subset $ \Upsilon $ of $ \{ 1,\ldots,k \} $. Moreover, fixed $ \Upsilon \subset \{ 1,\ldots,k \}$, it was proved that, for any sequence $ \lambda_n \to \infty $ we can extract a subsequence $( \lambda_{n_i}) $ such that $( u_{ \lambda_{n_i} } )$ converges strongly in $ W^{ 1,p(x) } \big( \mathbb R^N \big) $ to a function $ u $, which satisfies $ u = 0 $ outside $ \Omega_\Upsilon = \bigcup_{ j \in \Upsilon } \Omega_j $ and $ u_{|_{\Omega_j}}, \, j \in \Upsilon $, is a least energy solution for $$\begin{cases}
- \Delta u + Z(x) u = u^q, \text{ in } \Omega_j, \\
u \in H^1_0 \big( \Omega_j \big), \, u > 0, \text{ in } \Omega_j.
\end{cases}$$
In [@Alves], employing some different arguments than those used in [@DingTanaka], Alves extended the results described above to the $ p$-Laplacian operator, assuming that in $ \big( P_\lambda \big) $ the nonlinearity $ f $ possesses a subcritical growth and $ 2 \le p < N $. In particular, fixed $ \Upsilon \subset \{ 1,\ldots,k \}$, for any sequence $ \lambda_n \to \infty $ we can extract a subsequence $ (\lambda_{n_i}) $ such that $ (u_{ \lambda_{n_i} }) $ converges strongly in $ W^{ 1,p(x) } \big( \mathbb R^N \big) $ to a function $ u $, which satisfies $ u = 0 $ outside $ \Omega_\Upsilon $ and $ u_{|_{\Omega_j}}, \, j \in \Upsilon $, is a least energy solution for $$\begin{cases}
- \Delta_p u + Z(x) u = f(u), \text{ in } \Omega_j, \\
u \in W^{ 1,p }_0 \big( \Omega_j \big), \, u > 0, \text{ in } \Omega_j.
\end{cases}$$
In the present paper, we extend the results found in [@Alves] to the $ p(x)$-Laplacian operator. However, we would like emphasize that in a lot of estimates, we have used different arguments from that found in [@Alves]. The main difference is related to the fact that for equations involving the $p(x)$-Laplacian operator it is not clear that Moser’s iteration method is a good tool to get the estimates for the $L^{\infty}$-norm. Here, we adapt some ideas explored in [@FanZhao0] and [@FuscoSbordone] to get these estimates. For more details see Section 5.
Since we intend to find nonnegative solutions, throughout this paper, we replace $ f $ by $ f^+ \colon \Bbb R^N \times \Bbb R \to \Bbb R $ given by $$f^+(x,t) =
\begin{cases}
f(x,t), \text { if } t > 0 \\
\qquad \, 0, \text{ if } t \le 0.
\end{cases}$$ Nevertheless, for the sake of simplicity, we still write $ f $ instead of $ f^+ $.
The main theorem in this paper is the following:
\[main\] Assume that $ (H_1)-(H_4) $ and $ (f_1)-(f_5) $ hold. Then, there exist $ \lambda_0 > 0 $ with the following property: for any non-empty subset $ \Upsilon $ of $ \{1, 2, . . . , k \} $ and $ \lambda \ge \lambda_0 $, problem $ \big( P_\lambda \big) $ has a solution $u_\lambda$. Moreover, if we fix the subset $ \Upsilon $, then for any sequence $ \lambda_n \to \infty $ we can extract a subsequence $ (\lambda_{n_i}) $ such that $ (u_{ \lambda_{n_i} }) $ converges strongly in $ W^{ 1,p(x) } \big( \mathbb R^N \big) $ to a function $ u $, which satisfies $ u = 0 $ outside $ \Omega_\Upsilon = \bigcup_{ j \in \Upsilon } \Omega_j $ and $ u_{|_{\Omega_j}}, \, j \in \Upsilon $, is a least energy solution for $$\begin{cases}
- \Delta_{ p(x) } u + Z(x) u = f(x,u), \text{ in } \Omega_j, \\
u \in W^{ 1,p(x) }_0 \big( \Omega_j \big), \, u \ge 0, \text{ in } \Omega_j.
\end{cases}$$
[**Notations:**]{} The following notations will be used in the present work:\
$\bullet$ $C$ and $C_i$ will denote generic positive constant, which may vary from line to line;\
$\bullet$ In all the integrals we omit the symbol $ dx $.\
$\bullet$ If $u$ is a mensurable function, we denote $u^+$ and $u^-$ its positive and negative part, i.e., $u^+(x) = \max\{ u(x), 0 \}$ and $ u^-(x) = \min\{ u(x), 0 \}. $\
$\bullet$ For $ u,v \in C \big( \mathbb R^N \big) $, the notation $ u \ll v $ means that $ \displaystyle \inf_{ x \in \mathbb R^N } \big( v(x)-u(x) \big) > 0 $, $ u_- = \displaystyle \inf_{ x \in \mathbb R^N } u(x) $. Moreover, we will denote by $u^*$ the function $$u^*(x) =
\begin{cases}
\frac{Nu(x)}{N-u(x)}, \text{ if } u(x) < N, \\
\ \ \quad \infty, \text{ if } u(x) \ge N.
\end{cases}$$
Preliminaries on variable exponents Lebesgue and Sobolev spaces
===============================================================
In this section, we recall some results on variable exponents Lebesgue and Sobolev spaces found in [@AlvesFerreira1; @FZ; @FanShenZhao] and their references.
Let $ h \in L^\infty \big( \mathbb R^N \big) $ with $ h_- = \text{ess} \displaystyle \inf_{ \! \! \! \! \! \mathbb R^N} h \ge 1$. The *variable exponent Lebesgue space* $ L^{ h(x) } \big( \mathbb R^N \big) $ is defined by $$L^{ h(x) } \big( \mathbb R^N \big) = \left\{ u \colon \mathbb R^N \to \mathbb R \, ; \, u \text{ is measurable and } \int_{ \mathbb R^N} \left\vert u \right\vert^{ h(x) } < \infty \right\},$$ endowed with the norm $$\left\vert u \right\vert _{ h(x) } = \inf \left\{ \lambda > 0 \, ; \, \int_{ \mathbb R^N } \left\vert \frac{u}{\lambda} \right\vert^{ h(x) } \le 1 \right\}.$$ The *variable exponent Sobolev space* is defined by $$W^{ 1,h(x) } \big( \mathbb R^N \big) = \left\{ u \in L^{ h(x) } \big( \mathbb R^N \big) \, ; \, \big\vert \nabla u \big\vert \in L^{ h(x) } \big( \mathbb R^N \big) \right\},$$ with the norm $$\left\Vert u \right\Vert _{ 1,h(x) } = \inf \left\{ \lambda > 0 \, ; \, \int_{ \mathbb R^N } \left( \left\vert \frac{\nabla u}{\lambda} \right\vert^{ h(x) } + \left\vert \frac{u}{\lambda} \right\vert^{ h(x) } \right) \le 1 \right\}.$$ If $ h_- > 1 $, the spaces $ L^{ h(x) } \big( \mathbb R^N \big) $ and $ W^{ 1,h(x) } \big( \mathbb R^N \big) $ are separable and reflexive with these norms.
We are mainly interested in subspaces of $ W^{ 1,h(x) } \big( \mathbb R^N \big) $ given by $$E_W = \left\{ u \in W^{ 1,h(x) } \big( \mathbb R^N \big) \, ; \, \int_{ \mathbb R^N } W(x) |u|^{ h(x) } < \infty \right\},$$ where $ W \in C \big( \mathbb R^N \big) $ such that $ W_- > 0 $. Endowing $ E_W $ with the norm $$\left\Vert u \right\Vert _W = \inf \left\{ \lambda > 0 \, ; \, \int_{ \mathbb R^N } \left( \left\vert \frac{\nabla u}{\lambda} \right\vert^{ h(x) } + W(x)\left\vert \frac{u}{\lambda} \right\vert^{ h(x) } \right) \le 1 \right\},$$ $ E_W $ is a Banach space. Moreover, it is easy to see that $ E_W \hookrightarrow W^{ 1,h(x) } \big( \mathbb R^N \big) $ continuously. In addition, we can show that $ E_W $ is reflexive. For the reader’s convenience, we recall some basic results.
\[normxmodular\] The functional $ \varrho \colon E_W \to \mathbb R $ defined by $$\varrho(u) = \int_{ \mathbb R^N } \left( \big\vert \nabla u \big\vert^{ h(x) } + W(x) \left\vert u \right\vert^{ h(x) } \right),$$ has the following properties:
1. If $ \left\Vert u \right\Vert_W \ge 1 $, then $ \left\Vert u \right\Vert_W^{ h_- } \le \varrho(u) \le \left\Vert u\right\Vert_W^{ h_+ } $.
2. If $ \left\Vert u \right\Vert_W \le 1$, then $ \left\Vert u \right\Vert_W^{ h_+ }\le \varrho(u) \le \left\Vert u \right\Vert_W^{h_- }$.
In particular, for a sequence $ (u_n) $ in $ E_W $, $$\begin{gathered}
\left\Vert u_n \right\Vert_W \rightarrow 0 \iff \varrho(u_n) \rightarrow 0, \ \text{and}, \\
(u_n) \ \text{is bounded in} \ E_W \iff \varrho(u_n) \ \text{is bounded in} \ \mathbb R.\end{gathered}$$
For the functional $ \varrho_{ h(x) } \colon L^{ h(x) } \big( \mathbb R^N \big) \rightarrow \mathbb R $ given by $$\varrho_{ h(x) }(u) = \int_{ \mathbb R^N } \left\vert u \right\vert^{ h(x) },$$ the same conclusion of Proposition \[normxmodular\] also holds.
\[estimate0\] Let $ m \in L^\infty \big( \mathbb R^N \big) $ with $ 0 < m_- \le m(x) \le h(x) \text{ for a.e. } x \in \mathbb R^N $. If $ u \in L^{ h(x) } \big( \mathbb R^N \big) $, then $ |u|^{ m(x) } \in L^{ \frac{h(x)}{m(x)} } \big( \mathbb R^N \big) $ and $$\left| |u|^{ m(x) } \right|_{ \frac{h(x)}{m(x)} } \le \max \left\{ |u|_{ h(x) }^{ m_- }, |u|_{ h(x) }^{ m_+ } \right\} \le |u|_{ h(x) }^{ m_- } + |u|_{ h(x) }^{ m_+ }.$$
Related to the Lebesgue space $ L^{ h(x) } \big( \mathbb R^N \big) $, we have the following generalized Hölder’s inequality.
\[Hölder’s inequality\] \[Holder’s inequality\] If $ h_- > 1 $, let $ h' \colon \mathbb R^N \to \mathbb R $ such that $$\frac{1}{h( x)} + \frac{1}{h'(x)} = 1 \text{ for a.e. } x \in \mathbb R^N.$$ Then, for any $ u\in L^{ h( x) } \big( \mathbb R^N \big) $ and $ v \in L^{ h'(x) } \big( \mathbb R^N \big) $, $$\int_{\mathbb R^N} | uv | \, dx \le \left( \frac{1}{h_-} + \frac{1}{h'_-} \right) |u| _{ h( x) } |v|_{ h'(x) }.$$
We can define *variable exponent Lebesgue spaces with vector values*. We say $ u = ( u_1, \ldots, u_L ) \colon \mathbb R^N \to \mathbb R^L \in L^{ h(x) } \big( \mathbb R^N, \mathbb R^L \big) $ if, and only if, $ u_i \in L^{ h(x) } \big( \mathbb R^N \big) $, for $ i = 1, \ldots, L $. On $ L^{ h(x) } \big( \mathbb R^N, \mathbb R^L \big) $, we consider the norm $ \displaystyle | u |_{ L^{ h(x) }( \mathbb R^N, \mathbb R^L) } = \sum_{i=1}^L | u_i |_{ h(x) } $.
We state below lemmas of Brezis-Lieb type. The proof of the two first results follows the same arguments explored at [@Kavian], while the proof of the latter can be found at [@AlvesFerreira1].
\[Brezis-Lieb lemma, first version\] \[first Brezis-Lieb\] Let $ \left( u_n \right) $ be a bounded sequence in $ L^{ h(x) } \big( \mathbb R^N, \mathbb R^L \big) $ such that $ u_n(x) \to u(x) \text{ for a.e. } x \in \mathbb R^N $. Then, $ u \in L^{ h(x) } \big( \mathbb R^N, \mathbb R^L \big) $ and $$\int_{ \mathbb R^N } \left| \left| u_n \right|^{ h(x) } - \left| u_n - u \right|^{ h(x) } - \left| u \right|^{ h(x) } \right| \, dx = o_n(1).$$
\[Brezis-Lieb lemma, second version\] \[second Brezis-Lieb\] Let $ \left( u_n \right) $ be a bounded sequence in $ L^{ h(x) } \big( \mathbb R^N, \mathbb R^L \big) $ with $ h_- > 1 $ and $ u_n(x) \to u(x) \text{ for a.e. } x \in \mathbb R^N
$. Then $$u_n \rightharpoonup u \text{ in } L^{ h(x) } \big( \mathbb R^N, \mathbb R^L \big).$$
\[Brezis-Lieb lemma, third version\] \[third Brezis-Lieb\] Let $ (u_n) $ be a bounded sequence in $ L^{ h(x) } \big( \mathbb R^N, \mathbb R^L \big) $ with $ h_- > 1 $ and $ u_n(x) \to u(x) $ for a.e. $ x \in \mathbb R^N $. Then $$\int_{ \mathbb R^N } \left| \left| u_n \right|^{ h(x)-2 } u_n - \left| u_n - u \right|^{ h(x)-2 } \left( u_n - u \right) - | u |^{ h(x)-2 } u \right|^{ h'(x) } \, dx = o_n(1),$$
To finish this section, we notice that for any open subset $ \Omega \subset \mathbb R^N $, we can define of the same way the spaces $ L^{ h(x) } \big( \Omega \big) $ and $ W^{ 1,h(x) } \big( \Omega \big) $. Moreover, all the above propositions hold for these spaces and, besides, we have the following embedding Theorem of Sobolev’s type.
\[[[@FanShenZhao Theorems 1.1, 1.3]]{}\] \[embedding theorem\] Let $ \Omega \subset \mathbb R^N $ an open domain with the cone property, $ h \colon \overline{\Omega} \to \mathbb R $ satisfying $ 1 < h_- \le h_+ < N $ and $ m \in L^{\infty}_+ \big( \Omega \big) $.
1. If $ h $ is Lipschitz continuous and $ h \le m \le h^{\ast} $, the embedding $ W^{ 1,h(x) } \big( \Omega \big) \hookrightarrow L^{ m(x) } \big( \Omega \big) $ is continuous;
2. If $ \Omega $ is bounded, $ h $ is continuous and $ m \ll h^{\ast} $, the embedding $ W^{ 1,h(x) } \big( \Omega \big) \hookrightarrow L^{ m(x) } \big( \Omega \big) $ is compact.
An auxiliary problem
====================
In this section, we work with an auxiliary problem adapting the ideas explored in del Pino & Felmer [@DelPinoFelmer] (see also [@Alves]).
We start noting that the energy functional $ I_\lambda \colon E_\lambda \to \mathbb R $ associated with $ \big( P_\lambda \big) $ is given by $$I_\lambda (u) = \int_{ \mathbb R^N } \frac{1}{p(x)} \left( \big| \nabla u \big|^{ p(x) } + \big( \lambda V(x) + Z(x) \big) | u |^{ p(x) } \right) - \int_{ \mathbb R^N } F(x,u),$$ where $ E_\lambda = \big( E, \| \cdot \|_\lambda \big) $ with $$E = \left\{ u \in W^{ 1,p(x) } \big( \mathbb R^N \big) \, ; \, \int_{ \mathbb R^N } V(x) |u|^{ p(x) } < \infty \right\},$$ and $$\| u \|_\lambda = \inf \left\{ \sigma > 0 \, ; \, \varrho_{ \lambda } \left( \frac{u}{\sigma} \right) \le 1 \right\},$$ being $$\varrho_{ \lambda }(u) = \int_{ \mathbb R^N } \left( \big| \nabla u \big|^{ p(x) } + \big( \lambda V(x) + Z(x) \big) | u |^{ p(x) } \right).$$
Thus $ E_\lambda \hookrightarrow W^{ 1,p(x) } \big( \mathbb R^N \big) $ continuously for $ \lambda \ge 1 $ and $ E_\lambda $ is compactly embedded in $ L_{ loc }^{ h(x) } \big( \mathbb R^N \big) $, for all $ 1 \le h \ll p^* $. In addition, we can show that $ E_\lambda $ is a reflexive space. Also, being $ {\cal O} \subset \mathbb R^N $ an open set, from the relation $$\label{modular relation 1}
\varrho_{ \lambda, {\cal O} }(u) = \int_{ \cal O } \left( \big| \nabla u \big|^{ p(x) } + \big( \lambda V(x) + Z(x) \big) | u |^{ p(x) } \right) \ge M \int_{ \cal O } |u|^{ p(x) } = M \varrho_{ p(x), \cal{O} }(u),$$ for all $ u \in E_\lambda $ with $ \lambda \ge 1 $, writing $ M = ( 1-\delta )^{ -1 } \nu $, for some $ 0 < \delta < 1 $ and $ \nu > 0 $, we derive $$\label{modular relation 2}
\varrho_{ \lambda, \cal{O} }(u) - \nu \varrho_{ p(x), \cal{O} }(u) \ge \delta \varrho_{ \lambda,\cal{O} }(u), \, \forall u \in E_\lambda, \, \lambda \ge 1.$$
From the above commentaries, in this work the parameter $\lambda$ will be always bigger than or equal to 1.
We recall that for any $ \epsilon > 0 $, the hypotheses $ (f_1) $, $ (f_2) $ and $ (f_5) $ yield $$\label{f estimate}
f(x,t) \le \epsilon | t |^{ p(x)-1 } + C_\epsilon | t |^{ q(x)-1 }, \, \forall x \in \mathbb R^N, \, t \in \mathbb R,$$ and, consequently, $$\label{F estimate}
F(x,t) \le \epsilon | t |^{ p(x) } + C_\epsilon | t |^{ q(x) }, \, \forall x \in \mathbb R^N, \, t \in \mathbb R,$$ where $ C_\epsilon $ depends on $ \epsilon $. Moreover, for each $\nu >0$ fixed, the assumptions $ (f_2) $ and $ (f_3) $ allow us considering the function $ a \colon \mathbb R^N \to \mathbb R$ given by $$\label{fc a}
a(x) = \min \left\{ a > 0 \, ; \, \frac{f(x,a)}{a^{ p(x)-1 }} = \nu \right\}.$$ From $(f_2)$, it follows that $$\label{a_}
0 < a_- = \inf_{ x \in \mathbb R^N } a(x).$$
Using the function $a(x)$, we set the function $ \tilde{f} \colon \mathbb R^N \times \mathbb R \to \mathbb R $ given by $$\tilde{f}(x,t) =
\begin{cases}
\ \, f(x,t), \ t \le a(x) \\
\nu t^{ p(x)-1 }, \ t \ge a(x)
\end{cases},$$ which fulfills the inequality $$\label{til f estimate}
\tilde{f}(x,t) \le \nu | t |^{ p(x)-1 }, \, \forall x \in \mathbb R^N, t \in \mathbb R.$$ Thus $$\label{t til f estimate}
\tilde{f}(x,t) t \le \nu | t |^{ p(x) }, \, \forall x \in \mathbb R^N, t \in \mathbb R,$$ and $$\label{til F estimate}
\tilde{F}(x,t) \le \frac{\nu}{p(x)} | t |^{ p(x) }, \, \forall x \in \mathbb R^N, t \in \mathbb R,$$ where $ \tilde{F}(x,t) = \int_0^t \tilde{f}(x,s) \, ds $.
Now, once that $ \Omega=\text{int } V^{ -1 } (0) $ is formed by $ k $ connected components $ \Omega_1, \ldots, \Omega_k $ with $ \text{dist} \big( \Omega_i, \Omega_j \big) > 0, \, i \ne j $, then for each $ j \in \{ 1, \ldots, k \} $, we are able to fix a smooth bounded domain $ \Omega'_j $ such that $$\label{omega}
\overline{\Omega_j} \subset \Omega'_j \, \text{ and } \, \overline{\Omega'_i} \cap \overline{\Omega'_j} = \emptyset, \text{ for } i \ne j.$$
From now on, we fix a non-empty subset $ \Upsilon \subset \left\{ 1, \ldots, k \right\} $ and $$\Omega_\Upsilon = \bigcup_{ j \in \Upsilon } \Omega_j, \, \Omega'_\Upsilon = \bigcup_{ j \in \Upsilon } \Omega'_j, \,
\chi_\Upsilon =
\begin{cases}
1, \text{ if } x \in \Omega'_\Upsilon \\
0, \text{ if } x \notin \Omega'_\Upsilon .
\end{cases}$$ Using the above notations, we set the functions $$g(x,t) = \chi_\Upsilon(x) f(x,t) + \big( 1-\chi_\Upsilon(x) \big) \tilde{f}(x,t), \, (x,t) \in \mathbb R^N \times \mathbb R$$ and $$G(x,t) = \int_0^t g(x,s) \, ds, \, (x,t) \in \mathbb R^N \times \mathbb R,$$ and the auxiliary problem $$\big( A_\lambda \big) \,
\begin{cases}
- \Delta_{ p(x) } u + \big( \lambda V(x) + Z(x) \big) | u |^{ p(x)-2 } u = g(x,u), \text{ in } \mathbb R^N, \\
u \in W^{ 1,p(x) } \big( \mathbb R^N \big).
\end{cases}$$
The problem $ \big( A_\lambda \big) $ is related to $ \big( P_\lambda \big) $, in the sense that, if $ u_\lambda $ is a solution for $ \big( A_\lambda \big) $ verifying $$u_\lambda (x) \le a(x), \, \forall x \in \mathbb R^N \setminus \Omega'_\Upsilon,$$ then it is a solution for $ \big( P_\lambda \big) $.
In comparison to $ \big( P_\lambda \big) $, problem $ \big( A_\lambda \big) $ has the advantage that the energy functional associated with $ \big( A_\lambda \big) $, namely, $ \phi_\lambda \colon E_\lambda \to \mathbb R $ given by $$\phi_\lambda(u) = \int_{ \mathbb R^N } \frac{1}{p(x)} \left( \left| \nabla u \right|^{ p(x) } + \big( \lambda V(x) + Z(x) \big) | u |^{ p(x) } \right) - \int_{ \mathbb R^N } G(x,u),$$ satisfies the $ (PS) $ condition, whereas $ I_\lambda $ does not necessarily satisfy this condition. This way, the mountain pass level (see Theorem \[TAlambda\]) is a critical value for $ \phi_\lambda $.
\[mpg\] $ \phi_\lambda $ satisfies the mountain pass geometry.
From (\[F estimate\]) and (\[til F estimate\]), $$\phi_\lambda(u) \ge \frac{1}{p_+} \varrho_\lambda (u) - \epsilon \int_{ \mathbb R^N } |u|^{ p(x) } - C_\epsilon \int_{ \mathbb R^N } |u|^{ q(x) } - \frac{\nu}{p_-} \int_{\mathbb R^N} |u|^{ p(x) },$$ for $ \epsilon > 0 $ and $ C_\epsilon > 0 $ be a constant depending on $ \epsilon $. By , fixing $ \epsilon < \frac{M}{p_+} $ and $ \nu < p_- M \left( \frac{1}{p_+}-\frac{\epsilon}{M} \right) $ and assuming $ \| u \|_\lambda < \min \left\{ 1, 1/C_q \right\} $, where $ | v |_{ q(x) } \le C_q \| v \|_\lambda, \, \forall v \in E_\lambda $, we derive from Proposition \[normxmodular\] $$\phi_\lambda (u) \ge \alpha \| u \|^{ p_+ }_\lambda - C \| u \|^{q_-}_\lambda,$$ where $ \alpha = \left( \frac{1}{p_+} - \frac{\epsilon}{M} \right) - \frac{\nu}{p_-M} > 0 $. Once $ p_+ < q_- $, the first part of the mountain pass geometry is satisfied. Now, fixing $v \in C^{\infty}_{0}(\Omega_\Upsilon)$, we have for $t \geq 0$ $$\phi_{\lambda}(tv)= \int_{ \mathbb R^N } \frac{t^{p(x)}}{p(x)} \left( \left| \nabla v \right|^{ p(x) } + Z(x) \big) | v |^{ p(x) } \right) - \int_{ \mathbb R^N } F(x,tv).$$ If $t>1$, by $(f_3)$, $$\phi_{\lambda}(tv) \leq \frac{t^{p^{+}}}{p_{-}} \int_{ \mathbb R^N }\left( \left| \nabla v \right|^{ p(x) } + Z(x) \big) | v |^{ p(x) } \right) -C_1t^{\theta}\int_{\mathbb{R}^{N}}|v|^{\theta}-C_2,$$ and so, $$\phi_\lambda(tv) \to -\infty \quad \mbox{as} \quad t \to +\infty.$$ The last limit implies that $\phi_{\lambda}$ verifies the second geometry of the mountain pass.
\[boundedness\] All $ (PS)_d $ sequences for $ \phi_\lambda $ are bounded in $ E_\lambda $.
Let $ (u_n) $ be a $ (PS)_d $ sequence for $ \phi_\lambda $. So, there is $ n_0 \in \mathbb N $ such that $$\phi_\lambda (u_n) - \frac{1}{\theta} \phi_\lambda'(u_n) u_n \le d+1 + \| u_n \|_\lambda, \text { for } n \ge n_0.$$ On the other hand, by (\[t til f estimate\]) and (\[til F estimate\]) $$\tilde{F}(x,t) - \frac{1}{\theta} \tilde{f}(x,t)t \le \left( \frac{1}{p(x)} - \frac{1}{\theta} \right) \nu | t |^{ p(x) }, \, \forall x \in \mathbb R^N, t \in \mathbb R,$$ which together with (\[modular relation 2\]) gives $$\phi_\lambda (u_n) - \frac{1}{\theta} \phi_\lambda'(u_n) u_n \ge \left( \frac{1}{p_+} - \frac{1}{\theta} \right) \delta \varrho_\lambda (u_n), \, \forall n \in \mathbb N.$$ Hence $$d+1 + \max \left\{ {\varrho_\lambda (u_n)}^{1/p_-}, {\varrho_\lambda (u_n)}^{1/p_+} \right\} \ge \left( \frac{1}{p_+} - \frac{1}{\theta} \right) \delta \varrho_\lambda (u_n), \, \forall n \ge n_0,$$ from where it follows that $ (u_n) $ is bounded in $ E_\lambda $.
\[Estimativa no infinito\] If $(u_n)$ is a $(PS)_d$ sequence for $\phi_{\lambda}$, then given $\epsilon>0$, there is $R>0$ such that $$\label{Estimativa}
\limsup_n \int_{ \mathbb R^N \setminus B_R (0) } \left( \big| \nabla u_n \big|^{ p(x) } + \big( \lambda V(x) + Z(x) \big) | u_n |^{ p(x) } \right) < \epsilon.$$ Hence, once that $g$ has a subcritical growth, if $ u \in E_\lambda $ is the weak limit of $ (u_n) $, then $$\int_{\mathbb R^N}g(x,u_n)u_n\,dx \to \int_{\mathbb R^N} g(x,u)u \, dx \, \text{ and } \, \int_{\mathbb R^N} g(x,u_n)v \, dx \to \int_{\mathbb R^N} g(x,u)v \, dx, \, \forall v \in E_\lambda.$$
Let $ (u_n) $ be a $ (PS)_d $ sequence for $ \phi_\lambda $, $ R > 0 $ large such that $ \Omega'_\Upsilon \subset B_{ \frac{R}{2} }(0) $ and $ \eta_R \in C^\infty \big( \Bbb R^N \big) $ satisfying $$\eta_R (x) =
\begin{cases}
0, \, x \in B_{ \frac{R}{2} }(0) \\
1, \, x \in \Bbb R^N \setminus B_R (0)
\end{cases},$$ $ 0 \le \eta_R \le 1 $ and $ \big| \nabla \eta_R \big| \le \dfrac{C}{R} $, where $ C > 0 $ does not depend on $ R $. This way, $$\begin{aligned}
\mbox{} & \int_{ \Bbb R^N } \left( \big| \nabla u_n \big|^{ p(x) } + \big( \lambda V(x) + Z(x) \big) | u_n |^{ p(x) } \right) \eta_R \\
= & \phi_\lambda'(u_n) \left( u_n \eta_R \right) - \int_{ \Bbb R^N } u_n \big| \nabla u_n \big|^{ p(x)-2 } \nabla u_n \cdot \nabla \eta_R + \int_{ \Bbb R^N \setminus \Omega'_\Upsilon } \tilde{f}(x,u_n) u_n \eta_R.\end{aligned}$$ Denoting $$I =\int_{ \Bbb R^N } \left( \big| \nabla u_n \big|^{ p(x) } + \big( \lambda V(x) + Z(x) \big) | u_n |^{ p(x) } \right) \eta_R,$$ it follows from (\[t til f estimate\]), $$I \le \phi_\lambda'(u_n) \left( u_n \eta_R \right) + \frac{C}{R} \int_{ \Bbb R^N } | u_n | \big| \nabla u_n \big|^{ p(x)-1 } + \nu \int_{ \Bbb R^N } | u_n |^{ p(x) } \eta_R.$$ Using Hölder’s inequality \[Holder’s inequality\] and Proposition \[estimate0\], we derive $$I \le \phi_\lambda'(u_n) \left( u_n \eta_R \right) + \frac{C}{R} | u_n |_{ p(x) } \max \left\{ \big| \nabla u_n \big|^{ p_- -1 }_{ p(x) }, \big| \nabla u_n \big|^{ p_+ -1 }_{ p(x) } \right\} + \frac{\nu}{M}I.$$ Since $ (u_n) $ and $ \Big( \big| \nabla u_n \big| \Big) $ are bounded in $ L^{ p(x) } \big( \Bbb R^N \big) $ and $\frac{\nu}{M}=1-\delta$, we obtain $$\int_{ \Bbb R^N \setminus B_R(0) } \left( \big| \nabla u_n \big|^{ p(x) } + \big( \lambda V(x) + Z(x) \big) | u_n |^{ p(x) } \right) \le o_n(1) + \frac{C}{R}.$$ Therefore $$\limsup_n \int_{ \Bbb R^N \setminus B_R(0) } \left( \big| \nabla u_n \big|^{ p(x) } + \big( \lambda V(x) + Z(x) \big) | u_n |^{ p(x) } \right) \le \frac{C}{R}.$$ So, given $ \epsilon > 0 $, choosing a $ R > 0 $ possibly still bigger, we have that $ \dfrac{C}{R} < \epsilon $, which proves (\[Estimativa\]). Now, we will show that $$\int_{\Bbb R^N}g(x,u_n)u_n \to \int_{\Bbb R^N}g(x,u)u.$$ Using the fact that $g(x,u)u \in L^{1}(\Bbb R^N)$ together with (\[Estimativa\]) and Sobolev embeddings, given $\epsilon >0$, we can choose $R>0$ such that $$\limsup_{n \to +\infty}\int_{\Bbb R^N \setminus B_R(0)}|g(x,u_n)u_n| \leq \frac{\epsilon}{4} \quad \mbox{and} \quad \int_{\Bbb R^N \setminus B_R(0)}|g(x,u)u| \leq \frac{\epsilon}{4}.$$ On the other hand, since $g$ has a subcritical growth, we have by compact embeddings $$\int_{B_{R}(0)}g(x,u_n)u_n \to \int_{B_{R}(0)}g(x,u)u.$$ Combining the above informations, we conclude that $$\int_{\Bbb R^N}g(x,u_n)u_n \to \int_{\Bbb R^N}g(x,u)u.$$ The same type of arguments works to prove that $$\int_{\Bbb R^N}g(x,u_n)v \to \int_{\Bbb R^N}g(x,u)v \quad \forall v \in E_{\lambda}.$$
\[(PS) condition\] $ \phi_\lambda $ verifies the $ (PS) $ condition.
Let $ (u_n) $ be a $ (PS)_d $ sequence for $ \phi_\lambda $ and $ u \in E_\lambda $ such that $u_n \rightharpoonup u$ in $E_{\lambda}$. Thereby, by Proposition \[Estimativa no infinito\] $$\int_{ \mathbb R^N } g(x,u_n)u_n \to \int_{ \mathbb R^N } g(x,u)u \, \text{ and } \, \int_{ \mathbb R^N } g(x,u_n)v \to \int_{ \mathbb R^N } g(x,u)v, \, \forall v \in E_\lambda.$$ Moreover, the weak limit also give $$\int_{ \mathbb R^N } \big| \nabla u \big|^{ p(x)-2 } \nabla u \cdot \nabla ( u_n-u ) \to 0$$ and $$\int_{ \mathbb R^N } \big( \lambda V(x) + Z(x) \big) | u |^{ p(x)-2 }u ( u_n-u ) \to 0.$$ Now, if $$P_n^1 (x) = \left( \big| \nabla u_n \big|^{ p(x)-2 } \nabla u_n - \big| \nabla u \big|^{ p(x)-2 } \nabla u \right) \cdot \left( \nabla u_n - \nabla u \right)$$ and $$P_n^2 (x) = \left( | u_n|^{ p(x)-2 } u_n - | u |^{ p(x)-2 } u \right) ( u_n - u ),$$ we derive $$\begin{gathered}
\int_{ \mathbb R^N } \! \! \Big( P_n^1(x) + \big( \lambda V(x) + Z(x) \big) P_n^2(x) \Big) = \phi_\lambda'(u_n)u_n + \int_{ \mathbb R^N } \! g(x,u_n)u_n - \phi_\lambda'(u_n)u - \int_{ \mathbb R^N } \! g(x,u_n)u \\
- \int_{ \mathbb R^N } \left( \big| \nabla u \big|^{ p(x)-2 } \nabla u \cdot \nabla ( u_n-u ) + \big( \lambda V(x) + Z(x) \big) | u |^{ p(x)-2 }u ( u_n-u ) \right).\end{gathered}$$ Recalling that $\phi_\lambda'(u_n)u_n=o_n(1)$ and $\phi_\lambda'(u_n)u=o_n(1)$, the above limits lead to $$\int_{ \mathbb R^N } \Big( P_n^1(x) + \big( \lambda V(x) + Z(x) \big) P_n^2(x) \Big) \to 0.$$ Now, the conclusion follows as in [@AlvesFerreira1].
\[TAlambda\] The problem $ \big( A_\lambda \big) $ has a (nonnegative) solution, for all $ \lambda \ge 1 $.
The proof is an immediate consequence of the Mountain Pass Theorem due to Ambrosetti & Rabinowitz [@AmRb].
The $ (PS)_\infty $ condition
=============================
A sequence $ (u_n) \subset W^{ 1,p(x) } \big( \mathbb R^N \big) $ is called a $ (PS)_\infty $ *sequence for the family* $ \left( \phi_\lambda \right)_{\lambda \ge 1} $, if there is a sequence $ ( \lambda_n ) \subset [1, \infty) $ with $ \lambda_n \to \infty $, as $ n \to \infty $, verifying $$\phi_{ \lambda_n }(u_n) \to c \text{ and } \left\| \phi'_{ \lambda_n }(u_n) \right\| \to 0, \text{ as } n \to \infty.$$
\[(PS) infty condition\] Let $ (u_n) \subset W^{ 1,p(x) } \big( \mathbb R^N \big) $ be a $ (PS)_\infty $ sequence for $ \left( \phi_\lambda \right)_{\lambda \ge 1} $. Then, up to a subsequence, there exists $ u \in W^{ 1,p(x) } \big( \mathbb R^N \big) $ such that $ u_n \rightharpoonup u $ in $ W^{ 1,p(x) } \big( \mathbb R^N \big) $. Furthermore,
1. $ \varrho_{ \lambda_n } (u_n-u) \to 0 $ and, consequently, $ u_n \to u $ in $ W^{ 1,p(x) } \big( \mathbb R^N \big) $;
2. $ u = 0 $ in $ \mathbb R^N \setminus \Omega_\Upsilon $, $ u \ge 0 $ and $ u_{|_{\Omega_j}}, \, j \in \Upsilon $, is a solution for $$(P_j) \;
\begin{cases}
- \Delta_{ p(x) } u + Z(x) | u |^{ p(x)-2 } u = f(x,u), \text{ in } \Omega_j, \\
u \in W^{ 1,p(x) }_0 \big( \Omega_j \big);
\end{cases}$$
3. $ \displaystyle \int_{\mathbb R^N} \lambda_n V(x) | u_n |^{ p(x) } \to 0 $;
4. $ \varrho_{ \lambda_n, \Omega'_j } (u_n) \to \displaystyle \int_{ \Omega_j } \left( \big| \nabla u \big|^{ p(x) } + Z(x) | u |^{ p(x) } \right), \text{ for } j \in \Upsilon $;
5. $ \varrho_{ \lambda_n, \mathbb R^N \setminus \Omega_\Upsilon } (u_n) \to 0 $;
6. $ \phi_{ \lambda_n } (u_n) \to \displaystyle \int_{ \Omega_\Upsilon } \frac{1}{p(x)} \left( \big| \nabla u \big|^{ p(x) } + Z(x) | u |^{ p(x) } \right) - \int_{ \Omega_\Upsilon } F(x,u) $.
Using the same reasoning as in the proof of Proposition \[boundedness\], we obtain that $ \big( \varrho_{ \lambda_n }(u_n) \big) $ is bounded in $ \mathbb R $. Then $ \big( \| u_n \|_{ \lambda_n } \big) $ is bounded in $ \mathbb R $ and $ (u_n) $ is bounded in $ W^{ 1,p(x) } \big( \mathbb R^N \big) $. So, up to a subsequence, there exists $ u \in W^{ 1,p(x) } \big( \mathbb R^N \big) $ such that $$u_n \rightharpoonup u \text{ in } W^{ 1,p(x) } \big( \mathbb R^N \big) \, \text{ and } \, u_n(x) \to u(x) \text{ for a.e. } x \in \mathbb R^N.$$ Now, for each $ m \in \mathbb N $, we define $ C_m = \left\{ x \in \mathbb R^N \, ; \, V(x) \ge \dfrac{1}{m} \right\} $. Without loss of generality, we can assume $ \lambda_n < 2 ( \lambda_n-1 ), \, \forall n \in \mathbb N $. Thus $$\int_{ C_m } | u_n |^{ p(x) } \le \frac{2m}{\lambda_n} \int_{ C_m } \big( \lambda_n V(x)+Z(x) \big) | u_n |^{ p(x) } \le \frac{2m}{\lambda_n} \varrho_{ \lambda_n }(u_n) \le \frac{C}{\lambda_n}.$$ By Fatou’s lemma, we derive $$\int_{ C_m } | u |^{ p(x) } = 0,$$ which implies that $ u = 0 $ in $ C_m $ and, consequently, $ u = 0 $ in $ \mathbb R^N \setminus \overline{\Omega} $. From this, we are able to prove $(i)-(vi)$.
1. Since $ u = 0 $ in $ \mathbb R^N \setminus \overline{\Omega} $, repeating the argument explored in Proposition \[(PS) condition\] we get $$\int_{ \mathbb R^N } \Big( P_n^1(x) + \big( \lambda_n V(x) + Z(x) \big) P_n^2(x) \Big) \to 0,$$ where $$P_n^1 (x) = \left( \big| \nabla u_n \big|^{ p(x)-2 } \nabla u_n - \big| \nabla u \big|^{ p(x)-2 } \nabla u \right) \cdot \left( \nabla u_n - \nabla u \right) \\$$ and $$P_n^2 (x) = \left( | u_n|^{ p(x)-2 } u_n - | u |^{ p(x)-2 } u \right) ( u_n - u ).$$ Therefore, $ \varrho_{ \lambda_n } ( u_n-u ) \to 0 $, which implies $ u_n \to u $ in $ W^{ 1,p(x) } \big( \mathbb R^N \big) $.
2. Since $ u \in W^{ 1,p(x) } \big( \mathbb R^N \big) $ and $ u = 0 $ in $ \mathbb R^N \setminus \overline{\Omega} $, we have $ u \in W^{ 1,p(x) }_0 \big( \Omega \big) $ or, equivalently, $ u_{ |_{\Omega_j} } \in W^{ 1,p(x) }_0 \big( \Omega_j \big) $, for $ j = 1, \ldots, k $. Moreover, the limit $u_n \to u$ in $W^{1,p(x)}(\mathbb R^N)$ combined with $\phi'_{\lambda_n}(u_n)\varphi \to 0$ for $\varphi \in C^{\infty}_0 \big( \Omega_j \big)$ implies that $$\label{u is solution}
\int_{\Omega_j} \left( \big| \nabla u \big|^{ p(x)-2 } \nabla u \cdot \nabla \varphi + Z(x) | u |^{ p(x)-2 } u \varphi \right) - \int_{\Omega_j} g(x,u) \varphi = 0,$$ showing that $ u_{ |_{\Omega_j} } $ is a solution for $$\begin{cases}
- \Delta_{ p(x) } u + Z(x) | u |^{ p(x)-2 } u = g(x,u), \text{ in } \Omega_j, \\
u \in W^{ 1,p(x) }_0 \big( \Omega_j \big).
\end{cases}$$ This way, if $ j \in \Upsilon $, then $ u_{ |_{\Omega_j} } $ satisfies $ (P_j) $. On the other hand, if $ j \notin \Upsilon $, we must have $$\int_{ \Omega_j } \left( \big| \nabla u \big|^{ p(x) } + Z(x) | u |^{ p(x) } \right) - \int_{ \Omega_j } \tilde{f}(x,u)u = 0.$$ The above equality combined with (\[t til f estimate\]) and (\[modular relation 2\]) gives $$0 \ge \varrho_{ \lambda, \Omega_j }(u) - \nu \varrho_{ p(x), \Omega_j }(u) \ge \delta \varrho_{ \lambda, \Omega_j }(u) \ge 0,$$ from where it follows $ u_{|_{\Omega_j}} = 0 $. This proves $ u = 0 $ outside $ \Omega_\Upsilon $ and $ u \ge 0 $ in $ \Bbb R^N $.
3. It follows from (i), since $$\int_{ \mathbb R^N } \lambda_n V(x) | u_n |^{ p(x) } = \int_{ \mathbb R^N } \lambda_n V(x) | u_n-u |^{ p(x) } \le 2 \varrho_{ \lambda_n }(u_n-u).$$
4. Let $ j \in \Upsilon $. From (i), $$\varrho_{ p(x), \Omega'_j }( u_n-u ), \varrho_{ p(x), \Omega'_j } \big( \nabla u_n - \nabla u \big) \to 0.$$ Then by Proposition \[first Brezis-Lieb\], $$\int_{ \Omega'_j } \big( \big| \nabla u_n \big|^{ p(x) } - \big| \nabla u \big|^{ p(x) } \big) \to 0 \quad \mbox{and} \quad \int_{ \Omega'_j } Z(x) \big( | u_n |^{ p(x) } - | u |^{ p(x) } \big) \to 0.$$ From (iii), $$\int_{ \Omega'_j } \lambda_n V(x) \big( | u_n |^{ p(x) } - | u |^{ p(x) } \big) = \int_{ \Omega'_j \setminus \overline{\Omega_j} } \lambda_n V(x) | u_n |^{ p(x) } \to 0.$$ This way $$\varrho_{ \lambda_n, \Omega'_j } (u_n) - \varrho_{ \lambda_n, \Omega'_j } (u) \to 0.$$ Once $ u = 0 \text{ in } \Omega'_j \setminus \Omega_j $, we get $$\varrho_{ \lambda_n, \Omega'_j } (u_n) \to \int_{ \Omega_j } \left( | \nabla u |^{ p(x) } + Z(x) | u |^{ p(x) } \right).$$
5. By (i), $ \varrho_{ \lambda_n }( u_n-u ) \to 0 $, and so, $$\varrho_{ \lambda_n, \mathbb R^N \setminus \Omega_\Upsilon }(u_n) \to 0.$$
6. We can write the functional $\phi_{\lambda_n}$ in the following way $$\begin{gathered}
\phi_{ \lambda_n } (u_n) = \sum_{ j \in \Upsilon } \int_{ \Omega'_j } \frac{1}{p(x)} \left( \big| \nabla u_n \big|^{ p(x) } + \big( \lambda_n V(x) + Z(x) \big) | u_n |^{ p(x) } \right) \\
+ \int_{ \mathbb R^N \setminus \Omega'_\Upsilon } \frac{1}{p(x)} \left( \big| \nabla u_n \big|^{ p(x) } + \big( \lambda_n V(x) + Z(x) \big) | u_n |^{ p(x) } \right) - \int_{ \mathbb R^N } G(x,u_n).
\end{gathered}$$ From $(i)-(v)$, $$\int_{ \Omega'_j } \frac{1}{p(x)} \left( \big| \nabla u_n \big|^{ p(x) } + \big( \lambda_n V(x) + Z(x) \big) | u_n |^{ p(x) } \right) \to \int_{ \Omega_j } \frac{1}{p(x)} \left( \big| \nabla u \big|^{ p(x) } + Z(x) | u |^{ p(x) } \right),$$ $$\int_{ \mathbb R^N \setminus \Omega'_\Upsilon } \frac{1}{p(x)} \left( \big| \nabla u_n \big|^{ p(x) } + \big( \lambda_n V(x) + Z(x) \big) | u_n |^{ p(x) } \right) \to 0.$$ and $$\int_{ \mathbb R^N } G(x,u_n) \to \int_{ \Omega_\Upsilon } F(x,u).$$ Therefore $$\phi_{ \lambda_n } (u_n) \to \int_{ \Omega_\Upsilon } \frac{1}{p(x)} \left( | \nabla u |^{ p(x) } + Z(x) | u |^{ p(x) } \right) - \int_{ \Omega_\Upsilon } F(x,u).$$
The boundedness of the $ \big( A_\lambda \big) $ solutions
==========================================================
In this section, we study the boundedness outside $ \Omega'_\Upsilon $ for some solutions of $ \big( A_\lambda \big ) $. To this end, we adapt for our problem arguments found in [@FanZhao0] and [@FuscoSbordone].
\[P:boundedness of the solutions\] Let $ \big( u_\lambda \big) $ be a family of solutions for $ \big( A_\lambda \big) $ such that $ u_\lambda \to 0 $ in $ W^{ 1,p(x) } \big( \mathbb R^N \setminus \Omega_\Upsilon \big) $, as $ \lambda \to \infty $. Then, there exists $ \lambda^* > 0 $ with the following property: $$\left| u_\lambda \right|_{ \infty, \mathbb R^N \setminus \Omega'_\Upsilon } \le a_-, \, \forall \lambda \ge \lambda^*.$$ Hence, $u_{\lambda}$ is a solution for $(P_\lambda)$ for $\lambda \geq \lambda^*$.
Before to prove the above proposition, we need to show some technical lemmas.
\[boundary’s cover\] There exist $ x_1, \ldots, x_l \in \partial \Omega'_\Upsilon $ and corresponding $ \delta_{x_1}, \ldots, \delta_{x_l} > 0 $ such that $$\partial \Omega'_\Upsilon \subset {\cal N} \left( \partial \Omega'_\Upsilon \right) : = \bigcup_{ i=1 }^l B_{ \frac{\delta_{x_i}}{2} } (x_i).$$ Moreover, $$\label{qplus x pminus}
q^{x_i}_+ \le \big( p^{x_i}_- \big)^*,$$ where $$q^{x_i}_+ = \sup_{ B_{ \delta_{x_i} }({x_i}) } q, \ p^{x_i}_- = \inf_{ B_{\delta_{x_i}}(x_i) } p \text{ \, and \, } \big( p^{x_i}_- \big)^* = \frac{N p^{x_i}_-}{N-p^{x_i}_-}.$$
From , $ \overline{\Omega_\Upsilon} \subset \Omega'_\Upsilon $. So, there is $ \delta > 0 $ such that $$\overline{B_{\delta}(x)} \subset \mathbb R^N \setminus \overline{\Omega_\Upsilon}, \, \forall x \in \partial \Omega'_\Upsilon.$$ Once $ q \ll p^* $, there exists $ \epsilon > 0 $ such that $ \epsilon \le p^*(y) - q(y) $, for all $ y \in \mathbb R^N $. Then, by continuity, for each $ x \in \partial \Omega'_\Upsilon $ we can choose a sufficiently small $ 0 < \delta_x \le \delta $ such that $$q^x_+ \le \big( p^x_- \big)^*,$$ where $$q^x_+ = \sup_{ B_{ \delta_x }(x) } q, \ p^x_- = \inf_{ B_{ \delta_x }(x) } p \text{ \, and \, } \big( p^x_- \big)^* = \frac{N p^x_-}{N-p^x_-}.$$ Covering $ \partial \Omega'_\Upsilon $ by the balls $ B_{ \frac{\delta_x}{2} }(x), \, x \in \partial \Omega'_\Upsilon $, and using its compactness, there are $ x_1, \ldots, x_l \in \partial \Omega'_\Upsilon $ such that $$\partial \Omega'_\Upsilon \subset \bigcup_{ i=1 }^l B_{ \frac{\delta_{x_i}}{2} }(x_i).$$
\[good estimate\] If $ u_\lambda $ is a solution for $ \big( A_\lambda \big) $, in each $ B_{ \delta_{x_i} }(x_i), \, i = 1, \ldots, l $, given by Lemma \[boundary’s cover\], it is fulfilled $$\int_{ A_{k,\overline{\delta},x_i} } \big| \nabla u_\lambda \big|^{ p^{x_i}_- } \le C \Bigg( \big( k^{ q_+ } + 2 \big) \big| { A_{k,\widetilde{\delta},x_i} } \big| + \left(\widetilde{\delta}-\overline{\delta} \right)^{ -\big( p^{x_i}_- \big)^* } \int_{ A_{k,\widetilde{\delta},x_i} } \left( u_\lambda-k \right)^{ \big( p^{x_i}_- \big)^* } \Bigg),$$ where $ 0 < \overline{\delta} < \widetilde{\delta} < \delta_{ x_i } $, $ k \ge \dfrac{a_-}{4} $, $ C = C \big( p_-, p_+, q_-, q_+, \nu, \delta_{ x_i } \big) > 0 $ is a constant independent of $ k $, and for any $ R > 0 $, we denote by $A_ { k,R,x_i }$ the set $$A_ { k,R,x_i } = B_R(x_i) \cap \left\{ x \in \mathbb R^N \, ; \, u_\lambda(x) > k \right\}.$$
We choose arbitrarily $ 0 < \overline{\delta} < \widetilde{\delta} < \delta_{x_i} $ and $ \xi \in C^{ \infty } \big( \mathbb R^N \big) $ with $$0 \le \xi \le 1, \, \text{ supp } \xi \subset B_{ \widetilde{\delta} }(x_i), \, \xi = 1 \text{ in } B_{ \overline{\delta} }(x_i) \, \text{ and } \, \big| \nabla \xi \big| \le \frac{2}{\widetilde{\delta}-\overline{\delta}}.$$ For $ k \ge \dfrac{a_-}{4} $, we define $ \eta = \xi^{ p_+ } ( u_\lambda-k )^+ $. We notice that $$\nabla \eta = p_+ \xi^{ p_+-1 } (u_\lambda-k) \nabla \xi + \xi^{ p_+ } \nabla u_\lambda$$ on the set $ \left\{ u_\lambda > k \right\} $. Then, writing $ u_\lambda = u $ and taking $ \eta $ as a test function, we obtain $$\begin{gathered}
p_+ \int_{ A_{k,\widetilde{\delta},x_i} } \xi^{ p_+-1 } (u-k) \big| \nabla u \big|^{ p(x)-2 } \nabla u \cdot \nabla \xi + \int_{ A_{k,\widetilde{\delta},x_i} } \xi^{ p_+ } \big| \nabla u \big|^{ p(x) } \\
+ \int_{ A_{k,\widetilde{\delta},x_i} } \big( \lambda V(x) + Z(x) \big) u^{ p(x)-1 } \xi^{ p_+ } (u-k) = \int_{ A_{k,\widetilde{\delta},x_i} } g(x,u) \xi^{ p_+ } (u-k).\end{gathered}$$ If we set $$J = \int_{ A_{k,\widetilde{\delta},x_i} } \xi^{ p_+ } \big| \nabla u \big|^{ p(x) },$$ using that $ \nu \le \lambda V(x) + Z(x), \, \forall x \in \mathbb R^N $, we get $$\begin{gathered}
\label{J estimate}
J \le p_+ \int_{ A_{k,\widetilde{\delta},x_i} } \xi^{ p_+-1 } (u-k) \big| \nabla u \big|^{ p(x)-1 } \big| \nabla \xi \big| \\
- \int_{ A_{k,\widetilde{\delta},x_i} } \nu u^{ p(x)-1 } \xi^{ p_+ } (u-k) + \int_{ A_{k,\widetilde{\delta},x_i} } g(x,u) \xi^{ p_+ } (u-k).\end{gathered}$$ From , and , $$\begin{gathered}
J \le p_+ \int_{ A_{k,\widetilde{\delta},x_i} } \xi^{ p_+-1 } (u-k) \big| \nabla u \big|^{ p(x)-1 } \big| \nabla \xi \big| - \int_{ A_{k,\widetilde{\delta},x_i} } \nu u^{ p(x)-1 } \xi^{ p_+ } (u-k) \\
+ \int_{ A_{k,\widetilde{\delta},x_i} } \big( \nu u^{ p(x)-1 } + C_\nu u^{ q(x)-1 } \big) \xi^{ p_+ } (u-k),\end{gathered}$$ from where it follows $$J \le p_+ \int_{ A_{k,\widetilde{\delta},x_i} } \xi^{ p_+-1 } (u-k) \big| \nabla u \big|^{ p(x)-1 } \big| \nabla \xi \big| + C_\nu \int_{ A_{k,\widetilde{\delta},x_i} } u^{ q(x)-1 } (u-k).$$ Using Young’s inequality, we obtain, for $ \chi \in (0,1) $, $$\begin{gathered}
J \le \frac{p_+ (p_+-1)}{p_-} \chi^{ \frac{p_-}{p_+-1}} J + \frac{2^{ p_+ } p_+}{p_-} \chi^{ -p_+ } \int_{ A_{k,\widetilde{\delta},x_i} } \left( \frac{u-k}{\widetilde{\delta}-\overline{\delta}} \right)^{ p(x) } \\
+ \frac{C_\nu (q_+-1)}{q_-} \int_{ A_{k,\widetilde{\delta},x_i} } u^{ q(x) } + \frac{C_\nu \left( 1 + \delta_{ x_i }^{ q_+ } \right)}{q_-} \int_{ A_{k,\widetilde{\delta},x_i} } \left( \frac{u-k}{\widetilde{\delta}-\overline{\delta}}\right)^{ q(x) }.\end{gathered}$$ Writing $$Q = \int_{ A_{k,\widetilde{\delta},x_i} } \left( \frac{u-k}{\widetilde{\delta}-\overline{\delta}} \right)^{ \left( p^{x_i}_- \right)^* },$$ for $ \chi \approx 0^+ $ fixed, due to , we deduce $$\begin{gathered}
J \le \frac{1}{2} J + \frac{2^{ p_+ } p_+}{p_-} \chi^{ -p_+ } \Big( \big| { A_{k,\widetilde{\delta},x_i} } \big| + Q \Big) + \frac{C_\nu 2^{ q_+ } (q_+-1) \left( 1 + \delta_{ x_i }^{ q_+ } \right)}{q_-} \Big( \big| { A_{k,\widetilde{\delta},x_i} } \big| + Q \Big) \\
+ \frac{C_\nu 2^{ q_+ } (q_+-1) \left( 1+k^{ q_+ } \right)}{q_-} \big| { A_{k,\widetilde{\delta},x_i} } \big| + \frac{C_\nu \left( 1 + \delta_{ x_i }^{ q_+ } \right)}{q_-} \Big( \big| { A_{k,\widetilde{\delta},x_i} } \big| + Q \Big).\end{gathered}$$ Therefore $$\int_{ A_{k,\overline{\delta},x_i} } \big| \nabla u \big|^{ p(x) } \le J \le C \left[ \big( k^{ q_+ } + 1 \big) \big| A_{k,\widetilde{\delta},x_i} \big| + Q \right],$$ for a positive constant $ C = C \big( p_-, p_+, q_-, q_+, \nu, \delta_{ x_i } \big) $ which does not depend on $ k $. Since $$\big| \nabla u \big|^{ p^{x_i}_- } - 1 \le \big| \nabla u \big|^{ p(x) }, \, \forall x \in B_{\delta_{x_i}}(x_i),$$ we obtain $$\begin{aligned}
\int_{ A_{k,\overline{\delta},x_i} } \big| \nabla u \big|^{ p^{x_i}_- } & \le C \left[ \big( k^{ q_+ } + 1 \big) \big| A_{k,\widetilde{\delta},x_i} \big| + Q \right] + \big| A_{k,\widetilde{\delta},x_i} \big| \\
& \le C \left( \big( k^{ q_+ } + 2 \big) \big| A_{k,\widetilde{\delta},x_i} \big| + \left(\widetilde{\delta}-\overline{\delta} \right)^{ -\big( p^{x_i}_- \big)^* } \int_{ A_{k,\widetilde{\delta},x_i} } \left( u-k \right)^{ \big( p^{x_i}_- \big)^* } \right),\end{aligned}$$ for a positive constant $ C = C \big( p_-, p_+, q_-, q_+, \nu, \delta_{ x_i } \big) $ which does not depend on $ k $.
The next lemma can be found at ([@LadyUral Lemma 4.7]).
\[Ladyzhenskaya\] Let $ (J_n) $ be a sequence of nonnegative numbers satisfying $$J_{ n+1 } \le C B^n J_n^{ 1+\eta }, \, n=0,1,2,\ldots,$$ where $ C, \eta > 0 $ and $ B > 1 $. If $$J_0 \le C^{ - \frac{1}{\eta} } B^{ - \frac{1}{{\eta}^2} },$$ then $ J_n \to 0 $, as $ n \to \infty $.
\[boundedness of the solutions\] Let $ \big( u_\lambda \big) $ be a family of solutions for $ \big( A_\lambda \big) $ such that $ u_\lambda \to 0 $ in $ W^{ 1,p(x) } \big( \mathbb R^N \setminus \Omega_\Upsilon \big) $, as $ \lambda \to \infty $. Then, there exists $ \lambda^* > 0 $ with the following property: $$\left| u_\lambda \right|_{ \infty, {\cal N} \left( \partial \Omega'_\Upsilon \right) } \le a_-, \, \forall \lambda \ge \lambda^*.$$
It is enough to prove the inequality in each ball $ B_{\frac{\delta_{x_i}}{2}} (x_i), \, i = 1, \ldots, l $, given by Lemma \[boundary’s cover\]. We set $$\widetilde{\delta}_n = \frac{\delta_{x_i}}{2} + \frac{\delta_{x_i}}{2^{n+1}}, \ \overline{\delta}_n = \frac{\widetilde{\delta}_n + \widetilde{\delta}_{n+1}}{2}, \ k_n = \frac{a_-}{2} \left( 1-\frac{1}{2^{n+1}} \right), \, \forall n = 0, 1, 2, \ldots .$$ Then $$\widetilde{\delta}_n \downarrow \frac{\delta_{x_i}}{2}, \quad \widetilde{\delta}_{ n+1 } < \overline{\delta}_n < \widetilde{\delta}_n, \quad k_n \uparrow \frac{a_-}{2}.$$ From now on, we fix $$J_n(\lambda) = \int_{ A_{k_n,\widetilde{\delta}_n,x_i} } \big( u_\lambda(x) - k_n \big)^{ \left( p^{x_i}_- \right)^* }, \, n = 0, 1, 2, \ldots.$$ and $ \xi \in C^1 \big( \mathbb R \big) $ such that $$0 \le \xi \le 1, \ \xi(t) = 1, \text{ for } t \le \frac{1}{2}, \text{ and } \xi(t) = 0, \text{ for } t \ge \frac{3}{4}.$$ Setting $$\xi_n(x) = \xi \Bigg( \frac{2^{ n+1 }}{\delta_{x_i}} \bigg( \big| x-x_i \big|-\frac{\delta_{x_i}}{2} \bigg) \Bigg), \, x \in \mathbb R^N, \, n = 0, 1, 2, \ldots,$$ we have $ \xi_n = 1 $ in $ B_{ \widetilde{\delta}_{n+1} }(x_i) $ and $ \xi_n = 0 $ outside $ B_{ \overline{\delta}_n }(x_i) $. Writing $ u_\lambda = u $, we get $$\begin{aligned}
J_{ n+1 } & \le \int_{ A_{k_{n+1},\overline{\delta}_n,x_i} } \big( (u(x) - k_{ n+1 } ) \xi_n(x) \big)^{ \left( p^{x_i}_- \right)^* } \\
& = \int_{ B_{ \delta_{ x_i } }(x_i) } \big( ( u-k_{ n+1 } )^+(x) \xi_n(x) \big)^{ \left( p^{x_i}_- \right)^* } \\
& \le C \big( N, p^{x_i}_- \big) \left( \int_{ B_{ \delta_{ x_i } }(x_i) } \big| \nabla \big( ( u-k_{ n+1 } )^+ \xi_n \big)(x) \big|^{ p^{x_i}_- } \right)^{ \frac{\left( p^{x_i}_- \right)^*}{ p^{x_i}_-} } \\
& \le C \big( N, p^{x_i}_- \big) \left( \int_{ A_{k_{n+1},\overline{\delta}_n,x_i} } \big| \nabla u \big|^{ p^{x_i}_- } + \int_{ A_{k_{n+1},\overline{\delta}_n,x_i} } ( u-k_{ n+1 } )^{ p^{x_i}_- } \big| \nabla \xi_n \big|^{ p^{x_i}_- } \right)^{ \frac{\left( p^{x_i}_- \right)^*}{ p^{x_i}_-} }.\end{aligned}$$ Since $$\big| \nabla \xi_n(x) \big| \le C \big( \delta_{ x_i } \big) 2^{ n+1 }, \, \forall x \in \mathbb R^N,$$ writing $ J_{ n+1 }^{ \frac{p^{x_i}_-}{\left( p^{x_i}_- \right)^*} } = \widetilde{J}_{ n+1 } $, we obtain $$\widetilde{J}_{ n+1 } \le C \Big( N, p^{x_i}_-, \delta_{ x_i } \Big) \left( \int_{ A_{k_{n+1},\overline{\delta}_n,x_i} } \big| \nabla u \big|^{ p^{x_i}_- } + 2^{ n p^{x_i}_- } \int_{ A_{k_{n+1},\overline{\delta}_n,x_i} } ( u-k_{ n+1 } )^{ p^{x_i}_- } \right).$$ Using Lemma \[good estimate\], $$\begin{aligned}
\widetilde{J}_{ n+1 } & \le C \Big( N, p^{x_i}_-, \delta_{ x_i } \Big) \bigg( \left( k_{ n+1 }^{ q_+ }+2 \right) \big| A_{ k_{ n+1 },\widetilde{\delta}_n,x_i } \big| \\
& \phantom{ \le } + \left( \frac{2^{ n+3 }}{\delta_{ x_i }} \right)^{ \left( p^{x_i}_- \right)^* } \int_{ A_{ k_{ n+1 },\widetilde{\delta}_n,x_i } } ( u-k_{ n+1 } )^{ \left( p^{x_i}_- \right)^* } + 2^{ n p^{x_i}_- } \int_{ A_{ k_{ n+1 },\widetilde{\delta}_n,x_i } } ( u-k_{ n+1 } )^{ p^{x_i}_- } \bigg) \\
& \le C \Big( N, p^{x_i}_-, \delta_{ x_i } \Big) \bigg( \left( k_{ n+1 }^{ q_+ }+2 \right) \big| A_{ k_{ n+1 },\widetilde{\delta}_n,x_i } \big| \\
& \phantom{ \le } + 2^{ n \left( p^{x_i}_- \right)^* } \int_{ A_{ k_{ n+1 },\widetilde{\delta}_n,x_i } } ( u-k_{ n+1 } )^{ \left( p^{x_i}_- \right)^* } + 2^{ n p^{x_i}_- } \int_{ A_{ k_{ n+1 },\widetilde{\delta}_n,x_i } } ( u-k_{ n+1 } )^{ p^{x_i}_- } \bigg).\end{aligned}$$ From Young’s inequality $$\int_{ A_{ k_{ n+1 },\widetilde{\delta}_n,x_i } } ( u-k_{ n+1 } )^{ p^{x_i}_- } \le C \Big( p^{x_i}_- \Big) \left( \big| A_{ k_{ n+1 },\widetilde{\delta}_n,x_i } \big| + \int_{ A_{ k_{ n+1 },\widetilde{\delta}_n,x_i } } ( u-k_{ n+1 } )^{ \left( p^{x_i}_- \right)^* } \right).$$ Thus $$\widetilde{J}_{ n+1 } \le C \Big( N, p^{x_i}_-, \delta_{ x_i } \Big) \Bigg( \bigg( \left( \frac{a_-}{2} \right)^{ q_+ }+2+2^{ n p^{x_i}_- } \bigg) \big| A_{ k_{ n+1 },\widetilde{\delta}_n,x_i } \big| + 2^{ n \left( p^{x_i}_- \right)^* } J_n + 2^{ n p^{ x_i }_- } J_n \Bigg).$$ Now, since $$J_n \ge \int_{ A_{ k_{ n+1 },\widetilde{\delta}_n,x_i } } ( u-k_n)^{ \left( p^{x_i}_- \right)^* } \ge ( k_{ n+1 }-k_n )^{ \left( p^{x_i}_- \right)^* } \big| A_{ k_{ n+1 },\widetilde{\delta}_n,x_i } \big|$$ it follows that $$\big| A_{ k_{ n+1 },\widetilde{\delta}_n,x_i } \big| \le \left( \frac{2^{ n+3 }}{a_-} \right)^{ \left( p^{x_i}_- \right)^* } J_n,$$ and so, $$\begin{aligned}
\widetilde{J}_{ n+1 } & \le C \Big( N, p^{x_i}_-, \delta_{ x_i }, a_-, q_+ \Big) \left( 2^{ n \left( p^{x_i}_- \right)^* } J_n + 2^{ n \big( p^{ x_i }_- + \left( p^{x_i}_- \right)^* \big) } J_n + 2^{ n \left( p^{x_i}_- \right)^* } J_n + 2^{ n p^{ x_i }_- } J_n \right).\end{aligned}$$ Fixing $ \alpha = \big( p^{ x_i }_- + \left( p^{x_i}_- \right)^* \big) $, it follows that $$J_{ n+1 } \le C \Big( N, p^{x_i}_-, \delta_{ x_i }, a_-, q_+ \Big) \left( 2^{ \alpha \frac{\left( p^{x_i}_- \right)^*}{p^{ x_i }_-} } \right)^n { J_n }^{ \frac{\left( p^{x_i}_- \right)^*}{p^{ x_i }_-} },$$ and consequently $$J_{ n+1 } \le C B^n J_n^{ 1+\eta },$$ where $ C = C \Big( N, p^{x_i}_-, \delta_{ x_i }, a_-, q_+ \Big) $, $ B = 2^{ \alpha \frac{\left( p^{x_i}_- \right)^*}{p^{ x_i }_-} } $ and $ \eta = \frac{\left( p^{x_i}_- \right)^*}{p^{ x_i }_-} -1 $. Now, once that $ u_\lambda \to 0 $ in $ W^{ 1,p(x) } \big( \mathbb R^N \setminus \Omega_\Upsilon \big) $, as $ \lambda \to \infty $, there exists $ \lambda_i > 0 $ such that $$\int_{ A_{ \frac{a_-}{4}, \delta_{ x_i }, x_i } } \left( u_\lambda-\frac{a_-}{4} \right)^{ \left( p^{x_i}_- \right)^* } = J_0(\lambda) \le C^{ - \frac{1}{\eta} } B^{ - \frac{1}{{\eta}^2} }, \quad \lambda \geq \lambda_i.$$ From Lemma \[Ladyzhenskaya\], $ J_n(\lambda) \to 0 $, $ n \to \infty $, for all $ \lambda \geq \lambda_i$, and so, $$u_\lambda \le \frac{a_-}{2} < a_-, \text{ in } B_{\frac{\delta_{x_i}}{2}}, \text{ for all } \lambda \ge \lambda_i.$$ Now, taking $ \lambda^* = \max \{ \lambda_1, \ldots, \lambda_l \} $, we conclude that $$\left| u_\lambda \right|_{ \infty, {\cal N} \left( \partial \Omega'_\Upsilon \right) } < a_-, \, \forall \lambda \ge \lambda^*.$$
\[Proof of Proposition \[P:boundedness of the solutions\]\] Fix $ \lambda \ge \lambda^* $, where $ \lambda^* $ is given at Lemma \[boundedness of the solutions\], and define $ \widetilde{u}_\lambda \colon \mathbb R^N \setminus \Omega'_\Upsilon \to \mathbb R $ given by $$\widetilde{u}_\lambda(x) = \left( u_\lambda-a_- \right)^+ (x).$$ From Lemma \[boundedness of the solutions\], $ \widetilde{u}_\lambda \in W^{ 1,p(x) }_0 \big( \mathbb R^N \setminus \Omega'_\Upsilon \big) $. Our goal is showing that $\widetilde{u}_\lambda = 0 $ in $ \mathbb R^N \setminus \Omega'_\Upsilon $. This implies $$\left| u_\lambda \right|_{ \infty, \mathbb R^N \setminus \Omega'_\Upsilon } \le a_-.$$ In fact, extending $ \widetilde{u}_\lambda = 0 $ in $ \Omega'_\Upsilon $ and taking $ \widetilde{u}_\lambda $ as a test function, we obtain $$\int_{ \mathbb R^N \setminus \Omega'_\Upsilon } \! \! \big| \nabla u_\lambda \big|^{ p(x)-2 } \nabla u_\lambda \cdot \nabla \widetilde{u}_\lambda + \int_{ \mathbb R^N \setminus \Omega'_\Upsilon } \! \! \! \! \big( \lambda V(x) + Z(x) \big) u_{\lambda}^{ p(x)-2 } u_\lambda \widetilde{u}_\lambda = \! \! \int_{ \mathbb R^N \setminus \Omega'_\Upsilon } g \left( x, u_\lambda \right) \widetilde{u}_\lambda.$$ Since $$\begin{gathered}
\int_{ \mathbb R^N \setminus \Omega'_\Upsilon } \big| \nabla u_\lambda \big|^{ p(x)-2 } \nabla u_\lambda \cdot \nabla \widetilde{u}_\lambda = \int_{ \mathbb R^N \setminus \Omega'_\Upsilon } \big| \nabla \widetilde{u}_\lambda \big|^{ p(x) }, \\
\int_{ \mathbb R^N \setminus \Omega'_\Upsilon } \! \! \! \!\big( \lambda V(x) + Z(x) \big) u_{\lambda}^{ p(x)-2 } u_\lambda \widetilde{u}_\lambda = \int_{ \left( \mathbb R^N \setminus \Omega'_\Upsilon \right)_+ } \! \! \! \! \big( \lambda V(x) + Z(x) \big) u_{\lambda}^{ p(x)-2 } \left( \widetilde{u}_\lambda+a_- \right) \widetilde{u}_\lambda\end{gathered}$$ and $$\int_{ \mathbb R^N \setminus \Omega'_\Upsilon } g \left( x, u_\lambda \right) \widetilde{u}_\lambda = \int_{ \left( \mathbb R^N \setminus \Omega'_\Upsilon \right)_+ } \frac{g \left( x, u_\lambda \right)}{u_\lambda} \left( \widetilde{u}_\lambda+a_- \right) \widetilde{u}_\lambda,$$ where $$\left( \mathbb R^N \setminus \Omega'_\Upsilon \right)_+ = \left\{ x \in \mathbb R^N \setminus \Omega'_\Upsilon \, ; \, u_\lambda(x) > 0 \right\},$$ we derive $$\int_{ \mathbb R^N \setminus \Omega'_\Upsilon } \! \! \big| \nabla \widetilde{u}_\lambda \big|^{ p(x) } + \int_{ \left( \mathbb R^N \setminus \Omega'_\Upsilon \right)_+ } \! \! \! \! \left( \big( \lambda V(x) + Z(x) \big) u_{\lambda}^{ p(x)-2 } -\frac{g \left( x, u_\lambda \right)}{u_\lambda} \right) \left( \widetilde{u}_\lambda+a_- \right) \widetilde{u}_\lambda = 0,$$ Now, by , $$\big( \lambda V(x) + Z(x) \big) u_{\lambda}^{ p(x)-2 } - \frac{g \left( x, u_\lambda \right)}{u_\lambda} > \nu u_{\lambda}^{ p(x)-2 } - \frac{\tilde{f} \left( x, u_\lambda \right)}{u_\lambda} \ge 0 \quad \mbox{in} \quad \left( \mathbb R^N \setminus \Omega'_\Upsilon \right)_+ .$$ This form, $ \widetilde{u}_\lambda = 0 $ in $ \left( \mathbb R^N \setminus \Omega'_\Upsilon \right)_+ $. Obviously, $ \widetilde{u}_\lambda = 0 $ at the points where $ u_\lambda = 0 $, consequently, $ \widetilde{u}_\lambda = 0 $ in $ \mathbb R^N \setminus \Omega'_\Upsilon $.
A special critical value for $ \phi_\lambda $
=============================================
For each $ j = 1, \ldots, k $, consider $$I_j(u) = \int_{ \Omega_j } \frac{1}{p(x)} \left( \big| \nabla u \big|^{ p(x) } + Z(x) | u |^{ p(x) } \right) - \int_{ \Omega_j } F(x,u), \ u \in W^{ 1,p(x) }_0 \big( \Omega_j \big),$$ the energy functional associated to $ (P_j) $, and $$\phi_{ \lambda,j }(u) = \int_{ \Omega'_j } \frac{1}{p(x)} \left( \big| \nabla u \big|^{ p(x) } + \big( \lambda V(x) + Z(x) \big) | u |^{ p(x) } \right) - \int_{ \Omega'_j } F(x,u), \ u \in W^{ 1,p(x) } \big( \Omega'_j \big),$$ the energy functional associated to $$\begin{cases}
- \Delta_{ p(x) } u + \big( \lambda V(x) + Z(x) \big) | u |^{ p(x)-2 } u = f(x,u), \text{ in } \Omega'_j, \\
\phantom{ - \Delta_{ p(x) } u + \big( \lambda V(x) + Z(x) \big) | u |^{ p(x)-} } \frac{\partial u}{\partial \eta} = 0, \text{ on } \partial \Omega'_j.
\end{cases}$$ It is fulfilled that $ I_j $ and $ \phi_{ \lambda,j } $ satisfy the mountain pass geometry and let $$c_j = \inf_{ \gamma \in \Gamma_j } \max_{ t \in [0,1] } I_j \big( \gamma(t) \big) \, \text{ and } \, c_{ \lambda,j } = \inf_{ \gamma \in \Gamma_{ \lambda,j } } \max_{ t \in [0,1] } \phi_{ \lambda,j } \big( \gamma(t) \big),$$ their respective mountain pass levels, where $$\Gamma_j = \left\{ \gamma \in C \Big( [0,1], W^{ 1,p(x) }_0 \big( \Omega_j \big) \Big) \, ; \, \gamma(0) = 0 \text{ and } I_j \big( \gamma(1) \big) < 0 \right\}$$ and $$\Gamma_{ \lambda,j } = \left\{ \gamma \in C \Big( [0,1], W^{ 1,p(x) } \big( \Omega'_j \big) \Big) \, ; \, \gamma(0) = 0 \text{ and } \phi_{ \lambda,j } \big( \gamma(1) \big) < 0 \right\}.$$ Invoking the $ (PS) $ condition on $ I_j $ and $ \phi_{ \lambda,j } $, we ensure that there exist $ w_j \in W^{ 1,p(x) }_0 \big( \Omega_j \big) $ and $ w_{ \lambda,j } \in W^{ 1,p(x) } \big( \Omega'_j \big) $ such that $$I_j \big( w_j \big) = c_j \, \text{ and } \, I'_j \big( w_j \big) = 0$$ and $$\phi_{ \lambda,j } \big( w_{ \lambda,j } \big) = c_{ \lambda,j } \, \text{ and } \, \phi'_{ \lambda,j } \big( w_{ \lambda,j } \big) = 0.$$
There holds that
1. $ 0 < c_{ \lambda,j } \le c_j, \, \forall \lambda \ge 1, \, \forall j \in \left\{ 1, \ldots, k \right\} $;
2. $ c_{ \lambda,j } \to c_j, \text{ as } \lambda \to \infty, \, \forall j \in \left\{ 1, \ldots, k \right\} $.
<!-- -->
1. Once $ W^{ 1,p(x) }_0 \big( \Omega_j \big) \subset W^{ 1,p(x) } \big( \Omega'_j \big) $ and $ \phi_{ \lambda,j } \big( \gamma(1) \big) = I_j \big( \gamma(1) \big) $ for $ \gamma \in \Gamma_j $, we have $ \Gamma_j \subset \Gamma_{ \lambda,j } $. This way $$c_{ \lambda,j } = \inf_{ \gamma \in \Gamma_{ \lambda,j } } \max_{ t \in [0,1] } \phi_{ \lambda,j } \big( \gamma(t) \big) \le \inf_{ \gamma \in \Gamma_j } \max_{ t \in [0,1] } \phi_{ \lambda,j } \big( \gamma(t) \big) = \inf_{ \gamma \in \Gamma_j } \max_{ t \in [0,1] } I_j \big( \gamma(t) \big) = c_j.$$
2. It suffices to show that $ c_{ \lambda_n,j } \to c_j, \text{ as } n \to \infty $, for all sequences $ ( \lambda_n ) $ in $ [1,\infty) $ with $ \lambda_n \to \infty, \text{ as } n \to \infty $. Let $ \left( \lambda_n \right) $ be such a sequence and consider an arbitrary subsequence of $ \left( c_{ \lambda_n,j } \right) $ (not relabelled) . Let $ w_n \in W^{ 1,p(x) } \big( \Omega'_j \big) $ with $$\phi_{ \lambda_n,j } \big( w_n \big) = c_{ \lambda_n,j } \, \text{ and } \, \phi'_{ \lambda_n,j } \big( w_n \big) = 0.$$ By the previous item, $ \big( c_{ \lambda_n,j } \big) $ is bounded. Then, there exists $ \big( w_{ n_k } \big) $ subsequence of $ \big( w_n \big) $ such that $ \phi_{ \lambda_{ n_k },j } \big( w_{ n_k } \big) $ converges and $ \phi'_{ \lambda_{ n_k },j } \big( w_{ n_k } \big) = 0 $. Now, repeating the same type of arguments explored in the proof of Proposition \[(PS) infty condition\], there is $ w \in W^{ 1,p(x) }_0 \big( \Omega_j \big) \setminus \{0\} \subset W^{ 1,p(x) } \big( \Omega'_j \big) $ such that $$w_{ n_k } \to w \text{ in } W^{ 1,p(x) } \big( \Omega'_j \big), \text{ as } k \to \infty.$$ Furthermore, we also can prove that $$c_{ \lambda_{ n_k },j } = \phi_{ \lambda_{ n_k },j } \big( w_{ n_k } \big) \to I_j(w)$$ and $$0 = \phi'_{ \lambda_{ n_k },j } \big( w_{ n_k } \big) \to I'_j(w).$$ Then, by $ (f_4) $, $$\lim_k c_{ \lambda_{ n_k },j } \ge c_j.$$ The last inequality together with item (i) implies $$c_{ \lambda_{ n_k },j } \to c_j, \text{ as } k \to \infty.$$ This establishes the asserted result.
In the sequel, let $ R > 1 $ verifying $$\label{R}
0< I_j \left( \frac{1}{R} w_j \right), I_j(R w_j)< c_j, \text{ for } j = 1, \ldots, k.$$ There holds that $$c_j = \max_{ t \in [1/R^2,1] } I_j (t R w_j ), \text{ for } j = 1, \ldots, k.$$ Moreover, to simplify the notation, we rename the components $ \Omega_j $ of $ \Omega $ in way such that $ \Upsilon = \{ 1, 2, \ldots, l \} $ for some $ 1 \le l \le k $. Then, we define: $$\begin{gathered}
\gamma_0 ( t_1, \ldots, t_l )(x) = \sum_{j=1}^l t_j R w_j(x), \, \forall ( t_1, \ldots, t_l )\in [1/R^2,1]^l, \\
\Gamma_\ast = \Big\{ \gamma \in C \big( [1/R^2,1]^l, E_\lambda \setminus \{ 0 \} \big) \, ; \, \gamma = \gamma_0 \text{ on } \partial [1/R^2,1]^l \Big\}\end{gathered}$$ and $$b_{ \lambda, \Upsilon } = \inf_{ \gamma \in \Gamma_\ast } \max_{ ( t_1, \ldots, t_l )\in [1/R^2,1]^l } \phi_\lambda \big( \gamma ( t_1, \ldots, t_l ) \big).$$
Next, our intention is proving that $ b_{ \lambda, \Upsilon } $ is a critical value for $ \phi_\lambda $. However, to do this, we need to some technical lemmas. The arguments used are the same found in [@Alves], however for reader’s convenience we will repeat their proofs
\[solution’s existence\] For all $ \gamma \in \Gamma_\ast $, there exists $ (s_1, \ldots, s_l ) \in [1/R^2,1]^l $ such that $$\phi'_{ \lambda,j } \big( \gamma ( s_1, \ldots, s_l ) \big) \big( \gamma ( s_1, \ldots, s_l ) \big) = 0, \, \forall j \in \Upsilon.$$
Given $ \gamma \in \Gamma_\ast $, consider $ \widetilde{\gamma} \colon [1/R^2,1]^l \to \mathbb R^l $ such that $$\widetilde{\gamma} ( \textbf{t} ) = \Big( \phi'_{ \lambda,1 } \big( \gamma ( \textbf{t} ) \big) \gamma ( \textbf{t} ), \ldots, \phi'_{ \lambda,l } \big( \gamma ( \textbf{t} ) \big) \gamma ( \textbf{t} ) \Big), \text{ where } \textbf{t} = ( t_1, \ldots, t_l ).$$ For $ \textbf{t} \in \partial [1/R^2,1]^l $, it holds $ \widetilde{\gamma} ( \textbf{t} ) = \widetilde{\gamma_0} ( \textbf{t} ) $. From this, we observe that there is no $ \textbf{t} \in \partial [1/R^2,1]^l $ with $ \widetilde{\gamma} ( \textbf{t} ) = 0 $. Indeed, for any $ j \in \Upsilon $, $$\phi'_{ \lambda,j } \big( \gamma_0 ( {\bf{t}} ) \big) \gamma_0 ( {\bf{t}} ) = I'_j ( t_j R w_j ) ( t_j R w_j ).$$ This form, if $ {\bf{t}} \in \partial [1/R^2,1]^l $, then $ t_{j_0} =1 $ or $ t_{j_0} = \frac{1}{R^2} $, for some $ j_0 \in \Upsilon $. Consequently, $$\phi'_{ \lambda,j_0 } \big( \gamma_0 ( {\bf{t}} ) \big) \gamma_0 ( {\bf{t}} ) = I'_{j_0} ( R w_{j_0} ) ( R w_{j_0} ) \, \text{ or } \, \phi'_{ \lambda,j_0 } \big( \gamma_0 ( {\bf{t}} ) \big) \gamma_0 ( {\bf{t}} ) = I'_{j_0} \left( \frac{1}{R} w_{j_0} \right) \left( \frac{1}{R} w_{j_0} \right).$$ Therefore, if $ \phi'_{ \lambda,j_0 } \big( \gamma_0 ( {\bf{t}} ) \big) \gamma_0 ( {\bf{t}} ) = 0 $, we get $ I_{j_0} ( R w_{j_0} ) \ge c_{j_0} $ or $ I_{j_0} \left( \frac{1}{R} w_{j_0} \right) \ge c_{j_0} $, which is a contradiction with .
Now, we compute the degree $ \deg \big( \widetilde{\gamma}, (1/R^2,1)^l, (0, \ldots, 0 ) \big) $. Since $$\deg \big( \widetilde{\gamma}, (1/R^2,1)^l, (0, \ldots, 0 ) \big) = \deg \big( \widetilde{\gamma_0}, (1/R^2,1)^l, (0, \ldots, 0 ) \big),$$ and, for $ \textbf{t} \in (1/R^2,1)^l $, $$\widetilde{\gamma_0} ( \textbf{t} ) = 0 \iff {\bf{t}} = \left( \frac{1}{R}, \ldots, \frac{1}{R} \right),$$ we derive $$\deg \big( \widetilde{\gamma}, (1/R^2,1)^l, (0, \ldots, 0 ) \big) \ne 0.$$ This shows what was stated.
\[blambdagamma\] If $ c_{ \lambda,\Upsilon } = \displaystyle \sum_{ j=1 }^l c_{ \lambda,j } \, \text{ and } \, c_\Upsilon = \sum_{ j=1 }^l c_j $, then
1. $ c_{ \lambda,\Upsilon } \le b_{ \lambda,\Upsilon } \le c_\Upsilon, \, \forall \lambda \ge 1 $;
2. $ b_{ \lambda,\Upsilon } \to c_\Upsilon, \text{ as } \lambda \to \infty $;
3. $ \phi_\lambda \big( \gamma({\bf{t}}) \big) < c_\Upsilon, \, \forall \lambda \ge 1, \gamma \in \Gamma_\ast \text{ and } {\bf{t}} = (t_1, \ldots, t_l ) \in \partial [1/R^2,1]^l $.
<!-- -->
1. Once $ \gamma_0 \in \Gamma_\ast $, $$b_{ \lambda,\Upsilon } \le \max_{ ( t_1, \ldots, t_l ) \in [1/R^2,1]^l } \phi_\lambda \big( \gamma_0 ( t_1, \ldots, t_l ) \big) = \max_{ ( t_1, \ldots, t_l )\in [1/R^2,1]^l } \sum_{ j=1 }^l I_j ( t_j R w_j ) = c_\Upsilon.$$ Now, fixing $ {\bf s} = (s_1, \ldots, s_l) \in [1/R^2,1]^l $ given in Lemma \[solution’s existence\] and recalling that $$c_{ \lambda,j } = \inf \left\{ \phi_{ \lambda,j } (u) \, ; \, u \in W^{ 1,p(x) } \big( \Omega'_j \big) \setminus \{ 0 \} \text{ and } \phi'_{ \lambda,j }(u)u = 0 \right\},$$ it follows that $$\phi_{ \lambda,j } \big( \gamma( {\bf s } ) \big) \ge c_{ \lambda,j }, \, \forall j \in \Upsilon.$$ From , $$\phi_{ \lambda, \mathbb R^N \setminus \Omega'_\Upsilon } (u) \ge 0, \, \forall u \in W^{ 1,p(x) } \big( \mathbb R^N \setminus \Omega'_\Upsilon \big),$$ which leads to $$\phi_\lambda \big( \gamma( {\bf t} ) \big) \ge \sum_{ j=1 }^l \phi_{ \lambda,j } \big( \gamma( {\bf t} ) \big), \, \forall \textbf{t} = (t_1, \ldots, t_l) \in [1/R^2,1]^l.$$ Thus $$\max_{ ( t_1, \ldots, t_l )\in [1/R^2,1]^l } \phi_\lambda \big( \gamma ( t_1, \ldots, t_l ) \big) \ge \phi_\lambda \big( \gamma( \textbf{s} ) \big) \ge c_{ \lambda,\Upsilon },$$ showing that $$b_{ \lambda,\Upsilon } \ge c_{ \lambda,\Upsilon };$$
2. This limit is clear by the previous item, since we already know $ c_{ \lambda,j } \to c_j $, as $ \lambda \to \infty $;
3. For $ \textbf{t} = ( t_1, \ldots, t_l ) \in \partial [1/R^2,1]^l $, it holds $ \gamma ( \textbf{t} ) = \gamma_0 ( \textbf{t} ) $. From this, $$\phi_\lambda \big( \gamma ( \textbf{t} ) \big) = \sum_{j=1}^l I_j ( t_j R w_j ).$$ Writing $$\phi_\lambda \big( \gamma ( \textbf{t} ) \big) = \sum_{j=1 \atop j \ne j_0 }^l I_j ( t_j R w_j ) + I_{j_0} ( t_{j_0} R w_{j_0} ),$$ where $ t_{j_0} \in \left\{ \frac{1}{R^2}, 1 \right\} $, from we derive $$\phi_\lambda \big( \gamma ( \textbf{t} ) \big) \le c_\Upsilon - \epsilon,$$ for some $ \epsilon > 0 $, so (iii).
$ b_{ \lambda,\Upsilon } $ is a critical value of $ \phi_\lambda $, for $ \lambda $ sufficiently large.
Assume $ b_{ \widetilde{\lambda},\Upsilon } $ is not a critical value of $ \phi_{\widetilde{\lambda}} $ for some $ \widetilde{\lambda}$. We will prove that exists $ \lambda_1 $ such that $ \widetilde{\lambda} < \lambda_1 $. Indeed, by item (iii) of Proposition \[blambdagamma\], we have seen that $$\phi_\lambda \big( \gamma_0 ( \textbf{t} ) \big) < c_\Upsilon , \, \forall \lambda \ge 1, \, \textbf{t} \in \partial [1/R^2,1]^l.$$ This way $${\cal M} = \max_{ \textbf{t} \in \partial [1/R^2,1]^l } \phi_{ \widetilde{\lambda} } \big( \gamma_0 ( \textbf{t} ) \big) < c_\Upsilon.$$ Since $ b_{ \lambda,\Upsilon } \to c_\Upsilon $ (item (ii) of Proposition \[blambdagamma\]), there exists $ \lambda_1 > 1 $ such that if $ \lambda \ge \lambda_1 $, then $${\cal M} < b_{ \lambda,\Upsilon }.$$ So, if $ \widetilde{\lambda} \ge \lambda_1 $, we can find $ \tau = \tau( \widetilde{\lambda} ) > 0 $ small enough, with the ensuing property $$\label{M estimate}
{\cal M} < b_{ \widetilde{\lambda},\Upsilon } - 2\tau.$$ From the deformation’s lemma [@W Page 38], there is $ \eta \colon E_\lambda \to E_\lambda $ such that $$\eta \left( \phi_{\widetilde{\lambda}}^{ b_{ \widetilde{\lambda},\Upsilon } +\tau } \right) \subset \phi_{\widetilde{\lambda}}^{ b_{ \widetilde{\lambda},\Upsilon} -\tau } \, \text{ and } \, \eta(u) = u, \text{ for } u \notin \phi_{\widetilde{\lambda}}^{-1} \big( [b_{ \widetilde{\lambda},\Upsilon }-2 \tau, b_{ \widetilde{\lambda},\Upsilon }+2 \tau] \big).$$ Then, by , $$\eta \big( \gamma_0 ( \textbf{t} )\big) = \gamma_0 ( \textbf{t} ), \, \forall \textbf{t} \in \partial [1/R^2,1]^l.$$ Now, using the definition of $ b_{ \widetilde{\lambda},\Upsilon } $, there exists $ \gamma_\ast \in \Gamma_\ast $ satisfying $$\label{estimate}
\max_{ \textbf{t} \in [1/R^2,1]^l } \phi_{ \widetilde{\lambda} } \big( \gamma_\ast ( \textbf{t} ) \big) < b_{ \widetilde{\lambda},\Upsilon }+\tau.$$ Defining $$\widetilde{\gamma} ( \textbf{t} ) = \eta \big( \gamma_\ast ( \textbf{t} ) \big), \, \textbf{t} \in [1/R^2,1]^l,$$ due to , we obtain $$\phi_{ \widetilde{\lambda} } \big( \widetilde{\gamma}( \textbf{t} ) \big) \le b_{ \widetilde{\lambda},\Upsilon } - \tau, \, \forall \textbf{t} \in [1/R^2,1]^l.$$ But since $ \widetilde{\gamma} \in \Gamma_\ast $, we deduce $$b_{ \widetilde{\lambda},\Upsilon } \le \max_{ \textbf{t} \in [1/R^2,1]^l } \phi_{\widetilde{\lambda}} \big( \widetilde{\gamma} ( \textbf{t} ) \big) \le b_{ \widetilde{\lambda},\Upsilon }-\tau,$$ a contradiction. So, $ \widetilde{\lambda} < \lambda_1 $.
The proof of the main theorem
=============================
To prove Theorem \[main\], we need to find nonnegative solutions $ u_\lambda $ for large values of $ \lambda $, which converges to a least energy solution in each $ \Omega_j $ $ (j \in \Upsilon) $ and to $ 0 $ in $\Omega_\Upsilon^{c}$ as $ \lambda \to \infty $. To this end, we will show two propositions which together with the Propositions \[(PS) infty condition\] and \[P:boundedness of the solutions\] will imply that Theorem \[main\] holds.
Henceforth, we denote by $$r = R^{ p_+ } \sum_{ j=1 }^l \left( \frac{1}{p_+}-\frac{1}{\theta} \right)^{-1} c_j, \quad {\cal B}_r^\lambda = \big\{ u \in E_\lambda \, ; \, \varrho_\lambda (u) \le r \big\}$$ and $$\phi_\lambda^{ c_\Upsilon } = \big\{ u \in E_\lambda \, ; \, \phi_\lambda(u) \le c_{ \Upsilon } \big\}.$$
Moreover, for small values of $ \mu $, $${\cal A}_\mu^\lambda = \left\{ u \in {\cal B}_r^{\lambda }\, ; \, \varrho_{ \lambda, \mathbb R^N \setminus \Omega_\Upsilon } (u) \le \mu, \, \left| \phi_{ \lambda,j }(u)-c_j \right| \le \mu, \, \forall j \in \Upsilon \right\}.$$ We observe that $$w = \sum_{ j=1 }^l w_j \in {\cal A}_\mu^\lambda \cap \phi_\lambda^{ c_\Upsilon },$$ showing that $ {\cal A}_\mu^\lambda \cap \phi_\lambda^{ c_\Upsilon } \ne \emptyset $. Fixing $$\label{mu estimate}
0 < \mu < \frac{1}{4} \min_{ j \in \Gamma } c_j,$$ we have the following uniform estimate of $ \big\| \phi'_{ \lambda }(u) \big\| $ on the region $ \left( {\cal A}_{ 2 \mu }^\lambda \setminus {\cal A}_\mu^\lambda \right) \cap \phi_\lambda^{ c_\Upsilon } $.
\[derivative estimate\] Let $ \mu > 0 $ satisfying . Then, there exist $ \Lambda_\ast \ge 1 $ and $ \sigma_0 >0 $ independent of $ \lambda $ such that $$\big\| \phi'_{ \lambda }(u) \big\| \ge \sigma_0, \text{ for } \lambda \ge \Lambda_\ast \text{ and all } u \in \left( {\cal A}_{ 2 \mu }^\lambda \setminus {\cal A}_\mu^\lambda \right) \cap \phi_\lambda^{ c_\Upsilon }.$$
We assume that there exist $ \lambda_n \to \infty $ and $ u_n \in \left( {\cal A}_{ 2 \mu }^{\lambda_n} \setminus {\cal A}_\mu^{\lambda_n} \right) \cap \phi_{\lambda_n}^{ c_\Upsilon } $ such that $$\big\| \phi'_{ \lambda_n }(u_n) \big\| \to 0.$$ Since $ u_n \in {\cal A}_{ 2 \mu }^{ \lambda_n } $, this implies $ \big( \varrho_{ \lambda_n } (u_n) \big) $ is a bounded sequence and, consequently, it follows that $ \big( \phi_{ \lambda_n }(u_n) \big) $ is also bounded. Thus, passing a subsequence if necessary, we can assume $\phi_{ \lambda_n }(u_n) $ converges. Thus, from Proposition \[(PS) infty condition\], there exists $ 0 \le u \in W^{ 1,p(x) }_0 \big( \Omega_\Upsilon \big) $ such that $ u_{ |_{ \Omega_j } }, \, j \in \Upsilon $, is a solution for $ (P_j) $, $$\varrho_{ \lambda_n, \mathbb R^N \setminus \Omega_\Upsilon } (u_n) \to 0 \, \text{ and } \, \phi_{ \lambda_n,j } (u_n) \to I_j(u).$$ We know that $ c_j $ is the least energy level for $ I_j $. So, if $ u_{ |_{ \Omega_j } } \ne 0 $, then $ I_j(u) \ge c_j $. But since $ \phi_{ \lambda_n } (u_n) \le c_\Upsilon $, we must analyze the following possibilities:
1. $ I_j(u) = c_j, \, \forall j \in \Upsilon $;
2. $ I_{ j_0 }(u) = 0 $, for some $ j_o \in \Upsilon $.
If (i) occurs, then for $ n $ large, it holds $$\varrho_{ \lambda_n, \mathbb R^N \setminus \Omega_\Upsilon } (u_n) \le \mu \, \text{ and } \, \left| \phi_{ \lambda_n,j }(u_n)-c_j \right| \le \mu, \, \forall j \in \Upsilon.$$ So $ u_n \in {\cal A}_\mu^{\lambda_n} $, a contradiction.
If (ii) occurs, then $$\left| \phi_{ \lambda_n,j_0 }(u_n)-c_{ j_0 } \right| \to c_{ j_0 } > 4 \mu,$$ which is a contradiction with the fact that $ u_n \in {\cal A}_{ 2 \mu }^{\lambda_n} $. Thus, we have completed the proof.
\[P\] Let $ \mu > 0 $ satisfying and $ \Lambda_\ast \ge 1 $ given in the previous proposition. Then, for $ \lambda \ge \Lambda_\ast $, there exists a solution $ u_\lambda $ of $ (A_\lambda) $ such that $ u_\lambda \in {\cal A}_\mu^\lambda \cap \phi_\lambda^{ c_\Upsilon } $.
Let $ \lambda \ge \Lambda_\ast $. Assume that there are no critical points of $ \phi_\lambda $ in $ {\cal A}_\mu^\lambda \cap \phi_\lambda^{ c_\Upsilon } $. Since $ \phi_\lambda $ is a $ (PS) $ functional, there exists a constant $ d_\lambda > 0 $ such that $$\big\| \phi'_\lambda(u) \big\| \ge d_\lambda, \text{ for all } u \in {\cal A}_\mu^\lambda \cap \phi_\lambda^{ c_\Upsilon }.$$ From Proposition \[derivative estimate\] we have $$\big\| \phi'_\lambda(u) \big\| \ge \sigma_0, \text{ for all } u \in \left( {\cal A}_{ 2 \mu }^{\lambda} \setminus {\cal A}_\mu^{\lambda} \right) \cap \phi_{\lambda}^{ c_\Upsilon },$$ where $ \sigma_0 > 0 $ does not depend on $ \lambda $. In what follows, $ \Psi \colon E_\lambda \to \mathbb R $ is a continuous functional verifying $$\Psi(u) = 1, \text{ for } u \in {\cal A}_{\frac{3}{2} \mu}^\lambda, \ \Psi(u) = 0, \text{ for } u \notin {\cal A}_{2 \mu}^\lambda
\, \text{ and } \, 0 \le \Psi(u) \le 1, \, \forall u \in E_\lambda.$$ We also consider $ H \colon \phi_\lambda^{ c_\Upsilon } \to E_\lambda $ given by $$H(u) =
\begin{cases}
- \Psi(u) \big\| Y(u) \big\|^{ -1 } Y(u), \text{ for } u \in {\cal A}_{2 \mu}^\lambda, \\
\phantom{- \Psi(u) \big\| Y(u) \big\|^{ -1 } Y()} 0, \text{ for } u \notin {\cal A}_{2 \mu}^\lambda, \\
\end{cases}$$ where $ Y $ is a pseudo-gradient vector field for $ \Phi_\lambda $ on $ {\cal K} = \left\{ u \in E_\lambda \, ; \, \phi'_\lambda(u) \ne 0 \right\} $. Observe that $ H $ is well defined, once $ \phi'_\lambda(u) \ne 0 $, for $ u \in {\cal A}_{2 \mu}^\lambda \cap \phi_\lambda^{ c_\Upsilon } $. The inequality $$\big\| H(u) \big\| \le 1, \, \forall \lambda \ge \Lambda_* \text{ and } u \in \phi_\lambda^{ c_\Upsilon },$$ guarantees that the deformation flow $ \eta \colon [0, \infty) \times \phi_\lambda^{ c_\Upsilon } \to \phi_\lambda^{ c_\Upsilon } $ defined by $$\frac{d \eta}{dt} = H(\eta), \ \eta(0,u) = u \in \phi_\lambda^{ c_\Upsilon }$$ verifies $$\begin{gathered}
\frac{d}{dt} \phi_\lambda \big( \eta(t,u) \big) \le - \frac{1}{2} \Psi \big( \eta(t,u) \big) \big\| \phi'_\lambda \big( \eta(t,u) \big) \big\| \le 0, \label{eta derivative}\\
\left\| \frac{d \eta}{dt} \right\|_\lambda = \big\| H(\eta) \big\|_\lambda \le 1\end{gathered}$$ and $$\label{eta}
\eta(t,u) = u \text{ for all } t \ge 0 \text{ and } u \in \phi_\lambda^{ c_\Upsilon } \setminus {\cal A}_{2 \mu}^\lambda.$$ We study now two paths, which are relevant for what follows:\
$ \noindent \bullet $ The path $ {\bf t} \mapsto \eta \big( t, \gamma_0( {\bf t} ) \big), \text{ where } \textbf{t} = (t_1,\ldots,t_l) \in [1/R^2, 1]^l $.
The definition of $ \gamma_0 $ combined with the condition on $ \mu $ gives $$\gamma_0( {\bf t } ) \notin {\cal A}_{2 \mu}^\lambda, \, \forall {\bf t } \in \partial [1/R^2, 1]^l.$$ Since $$\phi_\lambda \big( \gamma_0( {\bf t } ) \big) < c_\Upsilon, \, \forall {\bf t } \in \partial [1/R^2, 1]^l,$$ from (\[eta\]), it follows that $$\eta \big( t, \gamma_0( {\bf t} ) \big) = \gamma_0( {\bf t} ), \, \forall {\bf t} \in \partial [1/R^2, 1]^l.$$ So, $ \eta \big( t, \gamma_0( {\bf t} ) \big) \in \Gamma_\ast $, for each $ t \ge 0 $.
$ \noindent \bullet $ The path $ {\bf t} \mapsto \gamma_0( {\bf t} ), \text{ where } \textbf{t} = (t_1,\ldots,t_l) \in [1/R^2, 1]^l $.
We observe that $$\text{supp} \big( \gamma_0 ( {\bf t} ) \big)\subset \overline{\Omega_\Upsilon}$$ and $$\phi_\lambda \big( \gamma_0 ( {\bf t} ) \big) \text{ does not depend on } \lambda \ge 1,$$ forall $ {\bf t} \in [1/R^2, 1]^l $. Moreover, $$\phi_\lambda \big( \gamma_0 ( {\bf t} ) \big) \le c_\Upsilon, \, \forall {\bf t} \in [1/R^2, 1]^l$$ and $$\phi_\lambda \big( \gamma_0 ( {\bf t} ) \big) = c_\Upsilon \text{ if, and only if, } t_j = \frac{1}{R}, \, \forall j \in \Upsilon.$$ Therefore $$m_0 = \sup \left\{ \phi_\lambda(u) \, ; \, u \in \gamma_0 \big( [1/R^2,1]^l \big) \setminus A_\mu^\lambda \right\}$$ is independent of $ \lambda $ and $ m_0 < c_\Upsilon $. Now, observing that there exists $ K_\ast > 0 $ such that $$\big| \phi_{ \lambda,j }(u) - \phi_{ \lambda,j }(v) \big| \le K_* \| u-v \|_{ \lambda, \Omega'_j }, \, \forall u,v \in {\cal B}_r^\lambda \text{ and } \forall j \in \Upsilon,$$ we derive $$\label{max estimate}
\max_{ {\bf t } \in [1/R^2,1]^l } \phi_\lambda \Big( \eta \big( T, \gamma_0 ( {\bf t} ) \big) \Big) \le \max \left\{ m_0, c_\Upsilon-\frac{1}{2 K_\ast} \sigma_0 \mu \right\},$$ for $ T > 0 $ large.
In fact, writing $ u = \gamma_0( {\bf t} ) $, $ {\bf t } \in [1/R^2,1]^l $, if $ u \notin A_\mu^\lambda $, from (\[eta derivative\]), $$\phi_\lambda \big( \eta( t, u ) \big) \le \phi_\lambda (u) \le m_0, \, \forall t \ge 0,$$ and we have nothing more to do. We assume then $ u \in A_\mu^\lambda $ and set $$\widetilde{\eta}(t) = \eta (t,u), \ \widetilde{d_\lambda} = \min \left\{ d_\lambda, \sigma_0 \right\} \text{ and } T = \frac{\sigma_0 \mu}{K_\ast \widetilde{d_\lambda}}.$$ Now, we will analyze the ensuing cases:\
[**Case 1:**]{} $ \widetilde{\eta}(t) \in {\cal A}_{\frac{3}{2} \mu}^\lambda, \, \forall t \in [0,T] $.
[**Case 2:**]{} $ \widetilde{\eta}(t_0) \in \partial {\cal A}_{\frac{3}{2} \mu}^\lambda, \text{ for some } t_0 \in [0,T] $.\
[**Analysis of Case 1**]{}
In this case, we have $ \Psi \big( \widetilde{\eta}(t) \big) = 1 $ and $ \big\| \phi'_\lambda \big( \widetilde{\eta}(t) \big) \big\| \ge \widetilde{d_\lambda} $ for all $ t \in [0,T] $. Hence, from (\[eta derivative\]), $$\phi_\lambda \big( \widetilde{\eta}(T) \big) = \phi_\lambda (u) + \int_0^T \frac{d}{ds} \phi_\lambda \big( \widetilde{\eta}(s) \big) \, ds \le c_\Upsilon - \frac{1}{2} \int_0^T \widetilde{d_\lambda} \, ds,$$ that is, $$\phi_\lambda \big( \widetilde{\eta}(T) \big) \le c_\Upsilon - \frac{1}{2} \widetilde{d_\lambda} T = c_\Upsilon - \frac{1}{2 K_\ast} \sigma_0 \mu,$$ showing (\[max estimate\]).\
[**Analysis of Case 2**]{}
In this case, there exist $ 0 \le t_1 \le t_2 \le T $ satisfying $$\begin{gathered}
\widetilde{\eta}(t_1) \in \partial {\cal A}_\mu^\lambda, \\
\widetilde{\eta}(t_2) \in \partial {\cal A}_{\frac{3}{2} \mu}^\lambda,\end{gathered}$$ and $$\widetilde{\eta}(t) \in {\cal A}_{\frac{3}{2} \mu}^\lambda \setminus {\cal A}_\mu^\lambda, \, \forall t \in (t_1,t_2].$$ We claim that $$\big\| \widetilde{\eta}(t_2)-\widetilde{\eta}(t_1) \big\| \ge \frac{1}{2 K_\ast} \mu.$$ Setting $ w_1 = \widetilde{\eta}(t_1) $ and $ w_2 = \widetilde{\eta}(t_2) $, we get $$\varrho_{ \lambda, \mathbb R^N \setminus \Omega_\Upsilon } (w_2) = \frac{3}{2} \mu \ \text{ or } \, \big| \phi_{ \lambda, j_0 } (w_2) - c_{j_0} \big| = \frac{3}{2} \mu,$$ for some $ j_0 \in \Upsilon $. We analyse the latter situation, once that the other one follows the same reasoning. From the definition of $ {\cal A}_\mu^\lambda $, $$\big| \phi_{ \lambda, j_0 } (w_1) - c_{j_0} \big| \le \mu,$$ consequently, $$\| w_2-w_1 \| \ge \frac{1}{K_\ast} \big| \phi_{ \lambda, j_0 } (w_2) - \phi_{ \lambda, j_0 } (w_1) \big| \ge \frac{1}{2 K_\ast} \mu.$$ Then, by mean value theorem, $ t_2-t_1 \ge \frac{1}{2 K_\ast} \mu $ and, this form, $$\phi_\lambda \big( \widetilde{\eta}(T) \big) \le \phi_\lambda(u) - \int_0^T \Psi \big( \widetilde{\eta}(s) \big) \big\| \phi'_\lambda \big( \widetilde{\eta}(s) \big) \big\| \, ds$$ implying $$\phi_\lambda \big( \widetilde{\eta}(T) \big) \le c_\Upsilon - \int_{t_1}^{t_2} \sigma_0 \, ds = c_\Upsilon - \sigma_0 (t_2-t_1) \le c_\Upsilon - \frac{1}{2 K_\ast} \sigma_0 \mu,$$ which proves \[max estimate\]. Fixing $ \widehat{\eta} (t_1, \ldots, t_l) = \eta \big( T, \gamma_0 (t_1,\ldots,t_l) \big) $, we have that $ \widehat{\eta} \in \Gamma_\ast $ and, hence, $$b_{ \lambda, \Gamma } \le \max_{ (t_1,\ldots,t_l) \in [1/R^2, 1] } \phi_\lambda \big( \widehat{\eta} (t_1,\ldots,t_l) \big) \le \max \left\{ m_0, c_\Upsilon - \frac{1}{2 K_\ast} \sigma_0 \mu \right\} < c_\Upsilon,$$ which contradicts the fact that $ b_{ \lambda, \Upsilon } \to c_\Upsilon $.
\[Proof of Theorem \[main\]\] According Proposition \[P\], for $\mu$ satisfying and $ \Lambda_\ast \ge 1 $, there exists a solution $ u_\lambda $ for $ (A_\lambda) $ such that $ u_\lambda \in {\cal A}_\mu^\lambda \cap \phi_\lambda^{ c_\Upsilon } $, for all $\lambda \geq \Lambda_*$.\
[**Claim:**]{} There are $\lambda_0 \geq \Lambda_*$ and $\mu_0>0$ small enough, such that $u_\lambda$ is a solution for $ \big( P_\lambda \big)$ for $\lambda \geq \Lambda_0$ and $\mu \in (0, \mu_0)$.
Indeed, assume by contradiction that there are $ \lambda_n \to \infty $ and $ \mu_n \to 0 $, such that $(u_{\lambda_n})$ is not a solution for $(P_{\lambda_n})$. From Proposition \[P\], the sequence $ (u_{\lambda_n}) $ verifies:
1. $ \phi'_{ \lambda_n }(u_{\lambda_n}) = 0, \, \forall n \in \Bbb N $;
2. $ \varrho_{ \lambda_n, \Bbb R^N \setminus \Omega_\Upsilon }(u_{\lambda_n}) \to 0$;
3. $ \phi_{ \lambda_n,j } (u_{\lambda_n}) \to c_j, \, \forall j \in \Upsilon. $
The item (b) ensures we can use Proposition \[P:boundedness of the solutions\] to deduce $ u_{\lambda_n} $ is a solution for $ \big( P_{\lambda_n} \big) $, for large values of $ n $, which is a contradiction, showing this way the claim.\
Now, our goal is to prove the second part of the theorem. To this end, let $(u_{\lambda_n})$ be a sequence verifying the above limits. Since $ \phi_{ \lambda_n }(u_{ \lambda_n } ) $ is bounded, passing a subsequence, we obtain that $ \phi_{ \lambda_n }(u_{ \lambda_n } ) \to c $. This way, using Proposition \[(PS) infty condition\] combined with item (c), we derive $ u_{ \lambda_n } $ converges in $ W^{ 1,p(x) } \big( \Bbb R^N \big ) $ to a function $ u \in W^{ 1,p(x) } \big( \Bbb R^N \big ) $, which satisfies $ u = 0 $ outside $ \Omega_\Upsilon $ and $ u_{|_{\Omega_j}}, \, j \in \Upsilon $, is a least energy solution for $$\begin{cases}
- \Delta_{ p(x) } u + Z(x) u = f(u), \text{ in } \Omega_j, \\
u \in W^{ 1,p(x) }_0 \big( \Omega_j \big), \, u \ge 0, \text{ in } \Omega_j.
\end{cases}$$
[99]{} E. Acerbi & G. Mingione, Regularity results for stationary electrorheological fluids, [*Arch. Rational Mech. Anal.*]{} [**164**]{} (2002), 213-259. E. Acerbi & G. Mingione, Regularity results for electrorheological fluids: stationary case, [*C.R. Math. Acad. Sci. Paris*]{} [**334**]{} (2002), 817-822. C.O. Alves, Existence of multi-bump solutions for a class of quasilinear problems, [*Adv. Nonlinear Stud.*]{} [**6**]{} (2006), 491-509. C.O. Alves, Existence of solutions for a degenerate $ p(x) $-Laplacian equation in $ \mathbb R^N $, [*J. Math. Anal. Appl.*]{} [**345**]{} (2008) 731-742. C.O. Alves, Existence of radial solutions for a class of $ p(x) $-Laplacian equations with critical growth, [*Differential and Integral Equations*]{} [**23**]{} (2010), 113-123. C.O. Alves & J.L.P. Barreiro, Existence and multiplicity of solutions for a $ p(x) $-Laplacian equation with critical growth, [*J. Math. Anal. Appl.*]{} [**403**]{} (2013) 143-154. C.O. Alves & M.C. Ferreira, Nonlinear perturbations of a $p(x)$-Laplacian equation with critical growth in $ \mathbb R^N $, to appear in [*Math. Nach.*]{} (2013). C.O. Alves & M.C. Ferreira, Existence of solutions for a class of $ p(x)$-Laplacian equations involving a concave-convex nonlinearity with critical growth in $ \mathbb R^N $, to appear in [*Topol. Methods Nonlinear Anal.*]{} (2013). C.O. Alves & M.A.S. Souto, Existence of solutions for a class of problems in $ \mathbb R^N $ involving $ p(x) $-Laplacian, [*Prog. Nonlinear Differential Equations and their Appl.*]{} [**66** ]{} (2005), 17-32. A. Ambrosetti & P.H. Rabinowitz, Dual variational methods in critical point theory and applications, J. Funct. Anal. 14 (1973) 349-381. S.N. Antontsev & J.F. Rodrigues, On stationary thermo-rheological viscous flows, [*Ann. Univ. Ferrara Sez. VII Sci. Mat.*]{} [**52**]{} (2006), 19-36. A. Chambolle & P.L. Lions, Image recovery via total variation minimization and related problems, [*Numer. Math.*]{} [**76**]{} (1997), 167-188. Y. Chen, S. Levine & M. Rao, Variable exponent, linear growth functionals in image restoration, [*SIAM J. Appl. Math.*]{} [ **66**]{} (2006), 1383-1406. M. del Pino & P.L. Felmer, Local mountain passes for semilinear elliptic problems in unbounded domains, [ *Calc. Var. PDE*]{} [**4**]{} (1996), 121-137. Y.H. Ding & K. Tanaka, Multiplicity of positive solutions of a nonlinear Schrödinger equation, [*Manuscripta Math.*]{} [**112(1)**]{} (2003) 109-135 X.L. Fan, On the sub-supersolution method for $ p(x) $-Laplacian equations, [*J. Math. Anal. Appl.*]{} [**330**]{} (2007), 665-682. X.L. Fan, $ p(x) $-Laplacian equations in $ \mathbb R^N $ with periodic data and nonperiodic perturbations, [*J. Math. Anal. Appl.*]{} [**341**]{} (2008), 103-119. X. Fan & D. Zhao, A class of De Giorgi type and Hölder continuity, [*Nonlinear Anal.*]{} [**36**]{} (1999), 295-318. X.L. Fan & D. Zhao, On the Spaces $ L^{ p(x) } \big( \Omega \big) $ and $ W^{ 1,p(x) } \big( \Omega \big) $, [*J. Math. Anal. Appl.*]{} [**263**]{} (2001), 424-446. X.L. Fan & D. Zhao, Nodal solutions of $ p(x) $-Laplacian equations, [*Nonlinear Anal.*]{} [**67**]{} (2007), 2859-2868. X.L. Fan, J.S. Shen & D. Zhao, Sobolev embedding theorems for spaces $ W^{ k,p(x) } \big( \Omega \big) $, [*J. Math. Anal. Appl.*]{} [**262**]{} (2001) 749-760. J. Fernández Bonder, N. Saintier & A. Silva. On the Sobolev embedding theorem for variable exponent spaces in the critical range, [*J. Differential Equations*]{} [**253**]{} (2012), 1604-1620 Y. Fu & X. Zhang, Multiple solutions for a class of $ p(x) $-Laplacian equations in involving the critical exponent, [*Proceedings Roy. Soc. of Edinburgh Sect A*]{} [**466**]{} (2010) 1667-1686. N. Fusco & C. Sbordone, Some remarks on the regularity of minima of anisotropic integrals, [*Comm. Partial Differential Equations* ]{} [**18(1-2)**]{} (1993), 153-167. O. Kavian, Introduction à la théorie de points critiques et applications aux problèmes elliptiques, Springer-Verlag France, Paris, 1993. O. A. Ladyzhenskaya & N. N. Ural’tseva, Linear and quasilinear elliptic equations, Acad. Press, 1968. M. Mih$\breve{a}$ilescu & V. R$\breve{a}$dulescu, On a nonhomogeneous quasilinear eigenvalue problem in Sobolev spaces with variable exponent, [*Proc. Amer. Math. Soc.*]{} [**135(9)**]{} (2007) 2929-2937 (electronic). M. Ruzicka, Electrorheological fluids: Modeling and mathematical theory. Lecture Notes in Mathematics, vol. 1748, Springer-Verlag, Berlin, 2000. E. Séré, Existence of infinitely many homoclinic orbits in Halmitonian systems, [*Math Z*]{} [**209**]{} (1992), 27-42. M. Willem, Minimax Theorems, Birkhäuser Boston, MA, 1996.
[^1]: C.O. Alves was partially supported by CNPq/Brazil 303080/2009-4, e-mail:[email protected]
[^2]: e-mail:[email protected]
[^3]: Partially supported by INCT-MAT and PROCAD
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We rewrite the standard 4-dimensional Dirac equation in terms of quaternionic spinors, leading to a formalism which treats both massive and massless particles on an equal footing. The resulting unified description has the correct particle spectrum to be a generation of leptons, with the correct number of spin/helicity states. Furthermore, precisely three such generations naturally combine into an octonionic description of the 10-dimensional massless Dirac equation, as previously discussed in [@Dim].'
author:
- |
Tevian Dray\
[*Department of Mathematics, Oregon State University, Corvallis, OR 97331*]{}\
[tevian[@]{}math.orst.edu]{}\
- |
Corinne A. Manogue\
[*Department of Physics, Oregon State University, Corvallis, OR 97331*]{}\
[corinne[@]{}physics.orst.edu]{}
date: '1 December 1998 (revised 30 September 1999)'
title: '**QUATERNIONIC SPIN [^1]** '
---
**INTRODUCTION**
================
We recently outlined a new dimensional reduction scheme [@Dim]. We show here in detail that applying this mechanism to the 10-dimensional massless Dirac equation on Majorana-Weyl spinors leads to a quaternionic description of the full 4-dimensional (free) Dirac equation which treats both massive and massless particles on an equal footing. Furthermore, there are naturally 3 such descriptions, each of which corresponds to a generation of leptons with the correct number of spin/helicity states.
The massive Dirac equation is usually formulated in the context of 4-component [*Dirac spinors*]{}. The 4 degrees of freedom correspond to the choice of spin (up or down) and the choice of particle or antiparticle. Similarly, 2-component [*Penrose spinors*]{}, which can be thought of as the square roots of null vectors, correspond to massless objects, such as photons. In Section \[COMPLEX\] we set the stage by reviewing these standard properties of the chiral description of the momentum-space Dirac equation.
Penrose spinors are usually thought of as Weyl projections of Dirac spinors; a Dirac spinor contains twice the information of a single Penrose spinor. As an alternative to doubling the number of (complex) components, however, we double the dimension of the underlying division algebra, from the complex numbers ${{\mathbb C}}$ to the quaternions ${{\mathbb H}}$. The anticommutativity of the quaternions then enables us to package two complex representations of opposite chirality into the (now quaternionic) 2-component formalism. In Section \[QUATERNIONIC\] we show how to replace the usual 4-component complex Dirac description with an equivalent 2-component quaternionic Penrose description, and further discuss how this puts the massive and massless Dirac equations on an equal footing.
We then consider in Section \[OCTONIONIC\] the massless Dirac equation on Majorana-Weyl spinors (in momentum space) in 10 dimensions, which can be nicely described in terms of 2-component spinors over the octonions ${{\mathbb O}}$, the only other normed division algebra besides ${{\mathbb R}}$, ${{\mathbb C}}$, and ${{\mathbb H}}$. Solutions of this equation are automatically quaternionic, and thus lend themselves to the preceding quaternionic description.
The final, and most important, ingredient in our approach is the the dimensional reduction scheme introduced in [@Dim]. In Sections \[CHOOSING\] and \[SPIN\] we describe how the choice of a preferred octonionic unit, or equivalently of a preferred complex subalgebra ${{\mathbb C}}\subset{{\mathbb O}}$, naturally reduces 10 spacetime dimensions to 4, and further allows us to use the standard representation of the Lorentz group $SO(3,1)$ as $SL(2,{{\mathbb C}})\subset SL(2,{{\mathbb O}})$. Putting this all together, we show in Section \[PARTICLES\] that the quaternionic spin/helicity eigenstates correspond precisely to the particle spectrum of a generation of leptons, consisting of 1 massive and 1 massless particle and their antiparticles.
In Section \[SPINOP\], we discuss the remarkable fact that the quaternionic spin eigenstates are in fact simultaneous eigenstates of all 3 spin operators, although the other two eigenvalues are not real. Finally, in Section \[DISCUSSION\] we discuss our results, in particular that, in a natural sense, there are precisely 3 such quaternionic subalgebras of the octonions, which we interpret as generations.
There is a long history of trying to use the quaternions in 4-dimensional quantum mechanics; see the comprehensive treatment in [@Adler] and references therein. Our approach is different in that we use the additional degrees of freedom to repackage existing information, rather than increasing the size of the underlying space of scalars. Ultimately, this leads us to work in more than 4 spacetime dimensions.
We also note a relatively unknown paper by Dirac [@Dirac] which, much to our surprise, contains the precursors of several of the key ideas presented here.
The octonions were first introduced into quantum mechanics by Jordan [@Jordan; @JNW]. There has in fact been much recent interest in using the octonions in (higher-dimensional) field theory; excellent modern treatments can be found in [@GT; @Okubo; @Dixon].
After much of this work was completed, we became aware of the recent work of Schücking [*et al.*]{} [@Schucking; @SubStandard], who also use a quaternionic formalism to describe a single generation of leptons. They further speculate that extending the formalism to the octonions would yield a description of a single generation of quarks as well. Although the language is strikingly similar, our approach differs fundamentally from theirs in its description of momentum. Ultimately, this hinges on our interpretation of the obvious $SU(2)$ as spin, whereas Schücking and coworkers interpret it as isospin.
**COMPLEX FORMALISM** {#COMPLEX}
=====================
The standard Weyl representation of the gamma matrices in signature is $$\label{CGamma}
\gamma_t = \pmatrix{0& I\cr \noalign{\smallskip} I& 0\cr}
\qquad\qquad
\gamma_a
= \pmatrix{0& \sigma_a\cr \noalign{\smallskip} -\sigma_a& 0\cr}$$ where $\sigma_a$ for $a=x,y,z$ denote the usual Pauli matrices [^2] $$\sigma_x = \pmatrix{0& 1\cr \noalign{\smallskip} 1& 0\cr} \qquad
\sigma_y = \pmatrix{0& -\ell\cr \noalign{\smallskip} \ell& 0\cr} \qquad
\sigma_z = \pmatrix{1& 0\cr \noalign{\smallskip} 0& -1\cr}$$ and $I$ is the $2\times2$ identity matrix.
The original formulation of the Dirac equation involves the even part of the Clifford algebra, historically written in terms of the matrices $\alpha_a=\gamma_t\gamma_a$ and $\beta=\gamma_t$. Explicitly, the momentum-space Dirac equation in this signature can be written as $$\label{DiracI}
(\gamma_t\gamma_\alpha \, p^\alpha - m \, \gamma_t) \, \Psi = 0$$ where $\alpha=t,x,y,z$ and $\Psi$ is a 4-component complex (Dirac) spinor.
Writing $\Psi$ in terms of two 2-component complex Weyl (or Penrose) spinors $\theta$ and $\eta$ as $$\Psi = \pmatrix{\theta\cr \noalign{\smallskip} \eta}$$ and expanding (\[DiracI\]) leads to $$\label{DiracII}
\pmatrix{p^tI-p^a\sigma_a& -m\cr \noalign{\smallskip} -m& p^tI+p^a\sigma_a}
\pmatrix{\theta\cr \noalign{\smallskip} \eta} = 0$$ This leads us to identify the momentum 4-vector with the $2\times2$ Hermitian matrix $${{\bf p}}= p^\alpha \sigma_\alpha
= \pmatrix{p^t+p^z& p^x-\ell p^y\cr \noalign{\smallskip} p^x+\ell p^y& p^t-p^z}$$ where we have set $\sigma_t=I$, which reduces (\[DiracII\]) to the two equations $$\begin{aligned}
-{\widetilde}{{\bf p}}\,\theta - m\eta &=& 0 \label{DiracIIIa}\\
-m\theta + {{\bf p}}\,\eta &=& 0 \label{DiracIIIb}\end{aligned}$$ where the tilde denotes trace-reversal. Explicitly, $${\widetilde}{{\bf p}}= {{\bf p}}- {{\rm tr\,}}({{\bf p}}) I$$ which reverses the sign of $p^t$, so that $-{\widetilde}{{\bf p}}$ can be identified with the 1-form dual to ${{\bf p}}$. This interpretation is strengthened by noting that $$-{\widetilde}{{\bf p}}{{\bf p}}= \det({{\bf p}}) I = p_\alpha p^\alpha I = m^2 I$$ where the identification of the norm of $p^\alpha$ with $m$ is just the compatibility condition between (\[DiracIIIa\]) and (\[DiracIIIb\]).
**QUATERNIONIC FORMALISM** {#QUATERNIONIC}
==========================
The [*quaternions*]{} ${{\mathbb H}}$ are the associative, noncommutative, normed division algebra over the reals. The quaternions are spanned by the identity element $1$ and three imaginary units, usually denoted $i$, $j$, $k:=ij$. Quaternionic conjugation, denoted with a bar, is given by reversing the sign of each imaginary unit. Each imaginary unit squares to $-1$, and they anticommute with each other; the full multiplication table then follows using associativity.[^3] However, in order to avoid conflict with our subsequent conventions for the octonions, we will instead label our quaternionic basis $\ell$, $k$, $\ell k$. The imaginary unit $\ell$ will play the role of the complex unit $i$, and, as we will see later, $k$ will label this particular quaternionic subalgebra of the octonions. In terms of the Cayley-Dickson process [@Dickson; @Schafer], we have $$\label{Cayley}
{{\mathbb H}}= {{\mathbb C}}+ {{\mathbb C}}k = ({{\mathbb R}}+ {{\mathbb R}}\ell) + ({{\mathbb R}}+ {{\mathbb R}}\ell)k$$
As vector spaces, ${{\mathbb H}}= {{\mathbb C}}^2$, which allows us to identify ${{\mathbb H}}^2$ with ${{\mathbb C}}^4$ in several different ways. We choose the identification $$\label{IdentifyC}
\pmatrix{A\cr B\cr C\cr D\cr}
\longleftrightarrow
\pmatrix{C - k B\cr \noalign{\smallskip} D + k A\cr}$$ with $A,B,C,D\in{{\mathbb C}}$. Equivalently, we can write this identification in terms of the Weyl (Penrose) spinors $\theta$ and $\eta$ as $$\label{IdentifyK}
\Psi =
\pmatrix{\theta\cr \noalign{\smallskip} \eta\cr}
\longleftrightarrow
\eta + \sigma_k \theta$$ where we have introduced the generalized Pauli matrix $$\sigma_k = \pmatrix{0& -k\cr \noalign{\smallskip} k& 0\cr}$$
Since (\[IdentifyK\]) is clearly a vector space isomorphism, there is also an isomorphism relating the linear maps on these spaces. We can use the induced isomorphism to rewrite the Dirac equation (\[DiracI\]) in 2-component quaternionic language. Direct computation yields the correspondences $$\gamma_t \gamma_a
= \pmatrix{-\sigma_a& 0\cr \noalign{\smallskip} 0& \sigma_a\cr}
\longleftrightarrow
\sigma_a$$ and $$\gamma_t \longleftrightarrow \sigma_k$$ and of course also $$\gamma_t\gamma_t \longleftrightarrow \sigma_t$$ since the left-hand-side is the $4\times4$ identity matrix and the right-hand-side is the $2\times2$ identity matrix. Direct translation of (\[DiracI\]) now leads to the quaternionic Dirac equation $$\label{DiracIV}
({{\bf p}}- m\sigma_k) (\eta + \sigma_k\theta) = 0$$ Working backwards, we can separate this into an equation not involving $k$, which is precisely (\[DiracIIIb\]), and an equation involving $k$, which is $$\label{DiracV}
{{\bf p}}\, \sigma_k \theta - m \sigma_k \eta = 0$$ Multiplying this equation on the left by $\sigma_k$, and using the remarkable identity $$\label{Remarkable}
\sigma_k \, {{\bf p}}\, \sigma_k = - {\widetilde}{{\bf p}}$$ reduces (\[DiracV\]) to (\[DiracIIIa\]), as expected.
So far, we have done nothing more or less than rewrite the usual momentum-space Dirac equation in 2-component quaternionic language. However, the appearance of the term $m\sigma_k$ suggests a way to put in the mass term on the same footing as the other terms, which we now exploit. Multiplying (\[DiracIV\]) on the left by $-\sigma_k$ and using (\[Remarkable\]) brings the Dirac equation to the form $$\label{DiracVI}
({\widetilde}{{\bf p}}+m\sigma_k) \, \psi = 0$$ where we have introduced the 2-component quaternionic spinor $$\psi = \sigma_k (\eta + \sigma_k\theta) = \theta + \sigma_k\eta$$ When written out in full, (\[DiracVI\]) takes the form $$\pmatrix{-p^t+p^z& p^x-\ell p^y-km\cr
\noalign{\smallskip}
p^x+\ell p^y+km& -p^t-p^z}
\psi = 0$$ This clearly suggests viewing the mass as an additional spacelike component of a higher-dimensional vector. Furthermore, since the matrix multiplying $\psi$ has determinant zero, this higher-dimensional vector is null. We thus appear to have reduced the massive Dirac equation in 4 dimensions to the massless Dirac, or Weyl, equation in higher dimensions, thus putting the massive and massless cases on an equal footing. This expectation is indeed correct, as we show in the next section in the more general octonionic setting.
**OCTONIONIC FORMALISM** {#OCTONIONIC}
========================
Octonionic Penrose Spinors
--------------------------
[=3.5in]{}
The [*octonions*]{} ${{\mathbb O}}$ are the nonassociative, noncommutative, normed division algebra over the reals. The octonions are spanned by the identity element $1$ and seven imaginary units, which we label as $\{i,j,k,k\ell,j\ell,i\ell,\ell\}$. Each imaginary unit squares to $-1$ $$i^2 = j^2 = k^2 = ... = \ell^2 = -1$$ and the full multiplication table can be conveniently encoded in the 7-point projective plane, as shown in Figure \[Omult\]; each line is to be thought of as a circle. The octonionic units can be grouped into (the imaginary parts of) quaternionic subalgebras in 7 different ways, corresponding to the 7 lines in the figure; these will be referred to as quaternionic triples. Within each triple, the arrows give the orientation, so that e.g. $$ij = k = -ji$$ Any three imaginary basis units which do not lie in a such a triple anti-associate. Note that any two octonions automatically lie in (at least one) quaternionic triple, so that expressions containing only two independent imaginary octonionic directions do associate. [*Octonionic conjugation*]{} is given by reversing the sign of the imaginary basis units, and the norm is just $$|p| = \sqrt{p {\overline}{p}}$$ which satisfies the defining property of a normed division algebra, namely $$|pq| = |p| |q|$$
We follow [@SUDBERY; @Chung] in representing real $(9+1)$-dimensional Minkowski space in terms of $2\times2$ Hermitian octonionic matrices. [^4] In analogy with the complex case, a vector field $q^\mu$ with $\mu=0,...,9$ can be thought of under this representation as a matrix $$Q = \pmatrix{~~q^+& {\overline}{q}\cr \noalign{\smallskip} q& ~~q^-}$$ where $q^\pm=q^0\pm q^9 \in{{\mathbb R}}$ are the components of $q^\mu$ in 2 null directions and $q=q^1 +q^2 i + ... + q^8 \ell \in{{\mathbb O}}$ is an octonion representing the transverse spatial coordinates. Following [@Sudbery], we define $${\widetilde}{Q} = Q - {{\rm tr\,}}(Q) I$$ Furthermore, since $Q$ satisfies its characteristic polynomial, we have $$-Q{\widetilde}{Q} = -{\widetilde}{Q}Q = -Q^2 + {{\rm tr\,}}(Q) Q = \det(Q) I
= g_{\mu\nu} q^\mu q^\nu I$$ where $g_{\mu\nu}$ is the Minkowski metric (with signature ($+$ $-$ ... $-$)). We can therefore identify the tilde operation with the metric dual, so that $-{\widetilde}{Q}$ represents the covariant vector field $q_\mu$.
Just as in the complex case (compare (\[CGamma\])), this can be thought of (up to associativity issues) as a Weyl representation of the underlying Clifford algebra ${{\cal C \it l}}(9,1)$ in terms of $4\times4$ gamma matrices of the form $$\label{MatDef}
q^\mu \gamma_\mu
= \pmatrix{0&Q\cr \noalign{\smallskip} -{\widetilde}{Q}&0\cr}$$ which are now octonionic. It is readily checked that $$\gamma_\mu\gamma_\nu + \gamma_\nu\gamma_\mu = 2\,g_{\mu\nu}$$ as desired.
In this language, a Majorana spinor $\Psi=\pmatrix{\psi\cr\noalign{\smallskip}\chi}$ is a 4-component octonionic column, whose chiral projections are the Majorana-Weyl spinors $\pmatrix{\psi\cr\noalign{\smallskip}0}$ and $\pmatrix{0\cr\noalign{\smallskip}\chi}$, which can be identified with the 2-component octonionic columns $\psi$ and $\chi$, which in turn can be thought of as generalized Penrose spinors. Writing $$\gamma_\mu
= \pmatrix{0& \sigma_\mu\cr \noalign{\smallskip}-{\widetilde}\sigma_\mu& 0\cr}$$ or equivalently $$\label{Qdef}
Q = q^\mu \sigma_\mu = q_\mu \sigma^\mu$$ defines the [*octonionic Pauli matrices*]{} $\sigma_\mu$. The matrices $\sigma_a$, with $a=1,...,9$, are the natural generalization of the ordinary Pauli matrices to the octonions, and $\sigma_0=I$. In analogy with our treatment of the complex case, we have $$\label{OctoI}
\gamma_0 \gamma_\mu
= \pmatrix{-{\widetilde}\sigma_\mu& 0\cr \noalign{\smallskip} 0& \sigma_\mu\cr}$$
For completeness, we record some useful relationships. The adjoint ${\overline}\Psi$ of the Majorana spinor $\Psi$ is given as usual by $${\overline}\Psi = \Psi^\dagger \gamma_0$$ since $$\gamma_\mu^\dagger \gamma_0^\dagger = \gamma_0 \gamma_\mu$$ Given a Majorana spinor $\Psi=\pmatrix{\psi\cr\noalign{\smallskip}\chi}$, we can construct a real vector [^5] $$q^\mu[\Psi] = {{\rm Re}}(\Psi^\dagger\gamma^0\gamma^\mu\Psi)$$ corresponding in traditional language to ${\overline}\Psi\gamma^\mu\Psi$. We can further identify this with a $2\times2$ matrix $Q[\Psi]$ as in (\[MatDef\]) above. Direct computation using the cyclic property of the trace, e.g. for octonionic columns $\Psi_1$, $\Psi_2$, and octonionic matrices $\gamma$ $${{\rm Re}}(\Psi_1^\dagger \gamma \Psi_2^{\phantom\dagger})
= {{\rm Re}}\left({{\rm tr\,}}(\Psi_1^\dagger \gamma \Psi_2^{\phantom\dagger})\right)
= {{\rm Re}}\left({{\rm tr\,}}(\Psi_2^{\phantom\dagger} \Psi_1^\dagger \gamma)\right)$$ shows that [@FairlieI; @FairlieII] $$\label{MSq}
Q[\Psi] = 2 \, \psi\psi^\dagger - 2 \, {\widetilde}{\chi\chi^\dagger}$$
Octonionic Dirac Equation
-------------------------
The momentum-space massless Dirac equation (Weyl equation) in 10 dimensions can be written in the form $$\label{Weyl}
\gamma_0\gamma_\mu \, p^\mu \, \Psi = 0$$ Choosing $\Psi=\pmatrix{\psi\cr\noalign{\smallskip}0}$ to be a Majorana-Weyl spinor, and using (\[OctoI\]) and (\[Qdef\]), (\[Weyl\]) finally takes the form $$\label{StringII}
{\widetilde}{P}\psi = 0$$ which is the octonionic Weyl equation. In matrix notation, it is straightforward to show that the momentum $p^\mu$ of a solution of the Weyl equation must be null: (\[StringII\]) says that the $2\times2$ Hermitian matrix $P$ has $0$ as one of its eigenvalues, [^6] which forces $\det(P)=0$, which is $$\label{StringI}
{\widetilde}{P}P = 0$$ which in turn is precisely the condition that $p^\mu$ be null.
Equations (\[StringI\]) and (\[StringII\]) are algebraically the same as the octonionic versions of two of the superstring equations of motion, as discussed in [@FairlieI; @FairlieII; @Sudbery], and are also the octonionic superparticle equations [@Schray]. As implied by those references, (\[StringI\]) implies the existence of a 2-component spinor $\theta$ such that $$\label{SolI}
P = \pm\theta\theta^\dagger$$ where the sign corresponds to the time orientation of $P$, and the general solution of (\[StringII\]) is $$\label{SolII}
\psi = \theta\xi$$ where $\xi\in{{\mathbb O}}$ is arbitrary. The components of $\theta$ lie in the complex subalgebra of ${{\mathbb O}}$ determined by $P$, so that (the components of) $\theta$ and $\xi$ (and hence also $P$) belong to a quaternionic subalgebra of ${{\mathbb O}}$. Thus, for solutions (\[SolII\]), the Weyl equation (\[Weyl\]) itself becomes quaternionic.
Furthermore, it follows immediately from (\[SolII\]) that $$\label{Proportional}
\psi\psi^\dagger = \pm |\xi|^2 P$$ Comparing this with (\[MSq\]), we see that the vector constructed from $\psi$ is proportional to $P$, or in more traditional language $${\overline}\Psi\gamma^\mu\Psi \sim p^\mu$$ which can be interpreted as the requirement that the Pauli-Lubanski spin vector be proportional to the momentum for a massless particle.
**DIMENSIONAL REDUCTION AND SPIN**
==================================
Choosing a Preferred Complex Subalgebra {#CHOOSING}
---------------------------------------
The description in the preceding section of 10-dimensional Minkowski space in terms of Hermitian octonionic matrices is a direct generalization of the usual description of ordinary (4-dimensional) Minkowski space in terms of complex Hermitian matrices. If we fix a complex subalgebra ${{\mathbb C}}\subset{{\mathbb O}}$, then we single out a 4-dimensional Minkowski subspace of 10-dimensional Minkowski space. The projection of a 10-dimensional null vector onto this subspace is a causal 4-dimensional vector, which is null if and only if the original vector was already contained in the subspace, and timelike otherwise. The time orientation of the projected vector is the same as that of the original, and the induced mass is given by the norm of the remaining 6 components. Furthermore, the ordinary Lorentz group $SO(3,1)$ clearly sits inside the Lorentz group $SO(9,1)$ via the identification of their double-covers, the spin groups $\hbox{Spin}(d,1)$, namely [^7] $$\hbox{Spin}(3,1) = SL(2,{{\mathbb C}}) \subset SL(2,{{\mathbb O}}) = \hbox{Spin}(9,1)$$
Therefore, all it takes to break 10 spacetime dimensions to 4 is to choose a preferred octonionic unit to play the role of the complex unit. We choose $\ell$ rather than $i$ to fill this role, preferring to save $i$, $j$, $k$ for a (distinguished) quaternionic triple. The projection $\pi$ from ${{\mathbb O}}$ to ${{\mathbb C}}$ is given by $$\pi(q) = {1\over2} (q + \ell q {\overline}\ell)$$ and we thus obtain a preferred $SL(2,{{\mathbb C}})$ subgroup of $SL(2,{{\mathbb O}})$, corresponding to the “physical” Lorentz group.
Spin {#SPIN}
----
Since we now have a preferred 4-d Lorentz group, we can use its rotation subgroup $SU(2)\subset SL(2,{{\mathbb C}})$ to define spin. However, care must be taken when constructing the Lie algebra $su(2)$, due to the lack of commutativity.
Under the usual action of $M\in SU(2)$ on a Hermitian matrix $Q$ (thought of as a spacetime vector via (\[Qdef\])), namely $$Q \mapsto MQM^\dagger$$ we can identify the basis rotations as usual as $$R_z = \pmatrix{e^{\ell{\phi\over2}}& 0\cr
\noalign{\smallskip} 0&
e^{-\ell{\phi\over2}}\cr}
\qquad
R_y = \pmatrix{~~\cos{\phi\over2}& \sin{\phi\over2}\cr
\noalign{\smallskip}
-\sin{\phi\over2}& \cos{\phi\over2}\cr}
\qquad
R_x = \pmatrix{\cos{\phi\over2}& \ell\sin{\phi\over2}\cr
\noalign{\smallskip}
\ell\sin{\phi\over2}& \cos{\phi\over2}\cr}$$ corresponding to rotations by the angle $\phi$ about the $z$, $y$, and $x$ axes, respectively.
The infinitesimal generators of the Lie algebra $su(2)$ are obtained by differentiating these group elements, via $$L_a = {d R_a\over d\phi} \Bigg|_{\phi=0}$$ where as before $a=x,y,z$. For reasons which will become apparent, we have [*not*]{} multiplied these generators by $-\ell$ to obtain Hermitian matrices. We have instead $$2 L_z = \pmatrix{\ell& 0\cr \noalign{\smallskip} 0& -\ell\cr}
\qquad
2 L_y = \pmatrix{0& 1\cr \noalign{\smallskip} -1& 0\cr}
\qquad
2 L_x = \pmatrix{0& \ell\cr \noalign{\smallskip} \ell& 0\cr}$$ which satisfy the commutation relations $$\left[ L_a , L_b \right] = \epsilon_{abc} L_c$$ where $\epsilon$ is completely antisymmetric and $$\epsilon_{xyz} = 1$$
Spin eigenstates are usually obtained as eigenvectors of the Hermitian matrix $-\ell L_z$, with real eigenvalues. Here we must be careful to multiply by $\ell$ in the correct place. We define $${\widehat}{L}_z \psi := - L_z \psi \ell$$ which is well-defined by alternativity, so that $${\widehat}{L}_z = - \ell_R \circ L_z$$ where the operator $\ell_R$ denotes right multiplication by $\ell$ and where $\circ$ denotes composition. The operators ${\widehat}{L}_a$ are self-adjoint with respect to the inner product $$\label{SelfAdj}
\langle \psi,\chi \rangle = \pi \!\left( \psi^\dagger\chi \right)$$ We therefore consider the eigenvalue problem $$\label{HEigen}
{\widehat}{L}_z \psi = \psi \lambda$$ with $\lambda\in{{\mathbb R}}$. It is straightforward to show that the real eigenvalues are $$\lambda_\pm = \pm {1\over2}$$ as expected. However, the form of the eigenvectors is a bit more surprising: $$\psi_+ = \pmatrix{A\cr kD\cr} \qquad\qquad
\psi_- = \pmatrix{kB\cr C\cr}$$ where $A,B,C,D\in{{\mathbb C}}$ are any elements of the preferred complex subalgebra, and $k$ is any imaginary octonionic unit orthogonal to $\ell$, so that $k$ and $\ell$ anticommute. Thus, the components of spin eigenstates are contained in the quaternionic subalgebra ${{\mathbb H}}\subset{{\mathbb O}}$ which is generated by $\ell$ and $k$.
Therefore, if we wish to consider spin eigenstates, $\ell$ must be in the quaternionic subalgebra ${{\mathbb H}}$ defined by the solution. We can further assume without loss of generality that ${{\mathbb H}}$ takes the form given in (\[Cayley\]). Thus, the only possible nonzero components of $p_\mu$ are $p_t=p_0$, $p_x=p_1$, $p_k=p_4$, $p_{k\ell}=p_5$, $p_y=p_8$, and $p_z=p_9$, corresponding to the gamma matrices with components in ${{\mathbb H}}$. We can further assume (via a rotation in the ($k,k\ell$)-plane if necessary) that $p_5=0$, so that $$\label{PComplex}
P = \pi(P) + m \, \sigma^k$$ where $$\pi(P) = p_\alpha \sigma^\alpha \equiv {{\bf p}}$$ with $\alpha = 0,1,8,9$ (or equivalently $\alpha=t,x,y,z$) is complex, and corresponds to the 4-dimensional momentum of the particle, with squared mass $$m^2 = p_\alpha p^\alpha = -\det(\pi(P))$$ Inserting (\[PComplex\]) into (\[StringII\]), we recover precisely (\[DiracVI\]), and we see that we have come full circle: Solutions of the [*octonionic*]{} Weyl equation (\[Weyl\]) are described precisely by the [*quaternionic*]{} formalism of Section \[QUATERNIONIC\], and the dimensional reduction scheme determines the mass term.
Particles {#PARTICLES}
---------
For each solution $\psi$ of (\[SolII\]), the momentum is proportional to $\psi\psi^\dagger$ by (\[Proportional\]). Up to an overall factor, we can therefore read off the components of the 4-dimensional momentum $p_\alpha$ directly from $\pi(\psi\psi^\dagger)$. We can use a Lorentz transformation to bring a massive particle to rest, or to orient the momentum of a massless particle to be in the $z$-direction.
If $m\ne0$, we can distinguish particles from antiparticles by the sign of the term involving $m$, which is the coefficient of $\sigma_k$ in $P$. Equivalently, we have the particle/antiparticle projections (at rest) $$\Pi_\pm
= {1\over2} \left( \sigma_t \pm \sigma_k \right)$$ If $m=0$, however, we can only distinguish particles from antiparticles in momentum space by the sign of $p^0$, as usual; this is the same as the sign in (\[Proportional\]). Similarly, in this language, the chiral projection operator is constructed from $$\Upsilon^5
= \sigma^t \sigma^x \sigma^y \sigma^z
= - \pmatrix{\ell& 0\cr 0& \ell\cr}$$ However, as with spin, we must multiply by $\ell$ in the correct place, obtaining $${\widehat}\Upsilon^5 = \ell_R \circ \Upsilon^5$$ As a result, even though $\Upsilon^5$ is a multiple of the identity, ${\widehat}\Upsilon^5$ is not, and the operators ${1\over2}(\sigma_t\pm{\widehat}\Upsilon^5)$ project ${{\mathbb H}}^2$ into the Weyl subspaces ${{\mathbb C}}^2\oplus{{\mathbb C}}^2 k$ as desired.
Combining the spin and particle information, over the quaternionic subalgebra ${{\mathbb H}}\subset{{\mathbb O}}$ determined by $k$ and $\ell$, we thus find 1 massive [spin-${1\over2}$]{} particle at rest, with 2 spin states, namely $${e_{\scriptscriptstyle\uparrow}}= \pmatrix{1\cr k\cr}
\qquad
{e_{\scriptscriptstyle\downarrow}}= \pmatrix{-k\cr ~~1\cr}$$ whose antiparticle is obtained by replacing $k$ by $-k$ (and changing the sign in (\[Proportional\])). We also find 1 massless [spin-${1\over2}$]{} particle involving $k$ moving in the $z$-direction, with a single helicity state, $${\nu_z}= \pmatrix{0\cr k\cr}$$ which corresponds, as usual, to both a particle and its antiparticle. It is important to note that $${\nu_{{\scriptscriptstyle-}z}}= \pmatrix{k\cr 0\cr}$$ corresponds to a massless particle with the same helicity moving in the opposite direction, not to a different particle with the opposite helicity. Each of the above states may be multiplied (on the [*right*]{}) by an arbitrary complex number.
There is also a single [*complex*]{} massless [spin-${1\over2}$]{} particle, with the opposite helicity, which is given in momentum space by $$\label{Oz}
\hbox{\O}_z=\pmatrix{0\cr 1\cr}$$ As with the other massless momentum space states, this describes both a particle and an antiparticle. Alone among the particles, this one does not contain $k$, and hence does not depend on the choice of identification of a particular quaternionic subalgebra ${{\mathbb H}}$ satisfying ${{\mathbb C}}\subset{{\mathbb H}}\subset{{\mathbb O}}$.
Spin Operators {#SPINOP}
--------------
We saw in the previous section that the spin up particle state ${e_{\scriptscriptstyle\uparrow}}$ is a simultaneous eigenvector of the spin operator and particle projections, that is $$2{\widehat}{L}_z {e_{\scriptscriptstyle\uparrow}}= {e_{\scriptscriptstyle\uparrow}}= \Pi_+ {e_{\scriptscriptstyle\uparrow}}$$ Remarkably, ${e_{\scriptscriptstyle\uparrow}}$ is also an eigenvector of the remaining spin operators, namely $$2{\widehat}{L}_x {e_{\scriptscriptstyle\uparrow}}= -{e_{\scriptscriptstyle\uparrow}}\,k \qquad
2{\widehat}{L}_y {e_{\scriptscriptstyle\uparrow}}= -{e_{\scriptscriptstyle\uparrow}}\,k\ell$$ although the eigenvalues are not real. Similar statements hold for the corresponding spin down and antiparticle states, although with different eigenvalues.
We find it illuminating to consider the equivalent right eigenvalue problem $$L_z \psi = \psi \lambda$$ for the non-Hermitian operator $L_z$. The operator $L_z$ admits imaginary eigenvalues $\pm \ell/2$, which correspond to the usual spin eigenstates. But $L_z$ also admits other imaginary eigenvalues! These correspond precisely to the eigenvalues of ${\widehat}{L}_z$ which are not real, and in fact not in ${{\mathbb C}}$. We emphasize that the spin operators [*are*]{} self-adjoint (with respect to (\[SelfAdj\])). However, over the octonions it is not true that all the eigenvalues of Hermitian matrices, are real [@Eigen; @NonReal]; the case of self-adjoint operators is similar.
How does it affect the traditional interpretation of quantum mechanics to have simultaneous eigenstates of all 3 spin operators? The essential feature which permits this is that only one of the eigenvalues is real, and only real eigenvalues correspond to observables. Thus, from this point of view, the reason that the spin operators fail to commute is not that they do not admit simultaneous eigenstates, but rather that their [*eigenvalues*]{} fail to commute!
Furthermore, while (right) multiplication by a (complex) phase does not change any of the real eigenvalues, the nonreal eigenvalues do depend on the phase, since the phase doesn’t commute with the eigenvalue! Does this allow one [*in principle*]{} to determine the exact spin orientation, even if no corresponding measurement exists?
**DISCUSSION** {#DISCUSSION}
==============
We have shown how the massless Dirac equation in 10 dimensions reduces to the (massive and massless) Dirac equation in 4 dimensions when a preferred octonionic unit is chosen.
The quaternionic Dirac equation discussed in Section \[QUATERNIONIC\] describes 1 massive particle with 2 spin states, 1 massless particle with only 1 helicity, and their antiparticles. We identify this set of particles with a generation of leptons.
Furthermore, as can be seen from Figure \[Omult\], there is room in the octonions for exactly 3 such quaternionic descriptions which have only their complex part in common, corresponding to replacing $k$ in turn by $i$ and $j$. We identify these 3 quaternionic spaces as describing 3 generations of leptons.
There is, however, one additional massless particle/antiparticle pair, given by (\[Oz\]). Being purely complex, it does not belong to any generation, and it has the opposite helicity from the other massless particles. We do not currently have a physical interpretation for this additional particle; if this theory is to correspond to nature, then this additional particle must for some reason not interact much with anything else.
Note that the mass appears in this theory as an overall scale, which can be thought of as the length scale associated with the corresponding quaternionic direction. In particular, antiparticles must have the same mass as the corresponding particles. This suggests that the only free parameters in this theory are 3 length scales, corresponding to the masses in each generation.
The theory presented here can be elegantly rewritten in terms of [*Jordan matrices*]{}, i.e. $3\times3$ octonionic Hermitian matrices, along the lines of the approach to the superparticle presented in [@Schray]. This approach, which is briefly described in [@Other], demonstrates that the theory is invariant under a much bigger group than the Lorentz group, namely the exceptional group $E_6$ (actually, $E_7$, since only conformal transformations are involved). We therefore believe it may be possible to extend the theory so as to include quarks and color.
Finally, as noted in [@Dim], we have worked only in momentum space, and have discussed only free particles. Perhaps our most intriguing result is the observation that the introduction of position space would require a preferred complex unit in the Fourier transform. Similarly, a description of interactions based on minimal coupling would again involve a preferred complex unit. Therefore, it does not appear to be [*possible*]{} to use the formalism presented here to give a full, interacting, 10-dimensional theory in which all 10 spacetime dimensions are on an equal footing. We view this as a tantalizing hint that not only interactions, but even 4-dimensional spacetime itself, may arise as a consequence of the symmetry breaking from 10 dimensions to 4!
It is a pleasure to thank David Griffiths, Phil Siemens, Tony Sudbery, and Pat Welch for comments on earlier versions of this work, Paul Davies for moral support, and Reed College for hospitality.
[99]{}
Corinne A. Manogue and Tevian Dray, [*Dimensional Reduction*]{}, Mod. Phys. Lett. [**A14**]{}, 93–97 (1999).
Stephen L. Adler, [**Quaternionic Quantum Mechanics and Quantum Fields**]{}, Oxford University Press, New York, 1995.
P. A. M. Dirac, Proc. Roy. Irish Acad., Sect. A, Vol. L, 261 (1945).
P. Jordan, [*Über die Multiplikation quantenmechanischer Größen*]{}, Z. Phys. [**80**]{}, 285–291 (1933).
P. Jordan, J. von Neumann, and E. Wigner, [*On an Algebraic Generalization of the Quantum Mechanical Formalism*]{}, Ann. Math. [**35**]{}, 29–64 (1934).
Feza Gürsey and Chia-Hsiung Tze, [ **On the Role of Division, Jordan, and Related Algebras in Particle Physics**]{}, World Scientific, Singapore, 1996.
S. Okubo, [**Introduction to Octonion and Other Non-Associative Algebras in Physics**]{}, Cambridge University Press, Cambridge, 1995.
Geoffrey M. Dixon, [**Division Algebras: Octonions, Quaternions, Complex Numbers and the Algebraic Design of Physics**]{}, Kluwer Academic Publishers, Boston, 1994.
E. L. Schücking, Jerome Epstein, William P. Kowalski, and Salvatore Lauro, [*What is Space-Time Made Of?*]{}, In: [**Quantum Gravity**]{}, Proceedings of the XIV Course of the International School of Cosmology and Gravitation (Erice 1995), eds. P. G. Bergmann, V. de Sabbata, and H.-J. Treder, World Scientific, Singapore, 1996, pp. 342–365.
Engelbert Schücking and Jerome Epstein, [**The (Sub)Standard Theory**]{}, in preparation.
Michael J. Crowe, [**A History of Vector Analysis**]{}, Dover, Mineola, NY, 1984 (originally published 1967).
L. E. Dickson, Ann. Math. [**20**]{}, 155 (1919).
Richard D. Schafer, [**An Introduction to Nonassociative Algebras**]{}, Academic Press, New York, 1966 & Dover, Mineola NY, 1995.
A. Sudbery, J. Phys. [**A17**]{}, 939 (1987).
K. W. Chung and A. Sudbery, [*Octonions and the Lorentz and Conformal Groups of Ten-Dimensional Space-Time*]{}, Phys. Lett. [**B198**]{}, 161 (1987).
T. Kugo and P. Townsend, [*Supersymmetry and the division algebras*]{}, Nucl. Phys. [**B221**]{}, 357 (1983).
David B. Fairlie and Corinne A. Manogue, [*Lorentz Invariance and the Composite String*]{}, Phys. Rev. [**D34**]{}, 1832-1834 (1986).
David B. Fairlie and Corinne A. Manogue, [*A Parameterization of the Covariant Superstring*]{}, Phys. Rev. [**D36**]{}, 475-479 (1987).
Corinne A. Manogue and Anthony Sudbery, [*General Solutions of Covariant Superstring Equations of Motion*]{}, Phys. Rev. [**D40**]{}, 4073-4077 (1989).
Jörg Schray, [*The General Classical Solution of the Superparticle*]{}, Class. Quant. Grav. [**13**]{}, 27 (1996).
Jörg Schray, [**Octonions and Supersymmetry**]{}, Ph.D. thesis, Department of Physics, Oregon State University, 1994.
Tevian Dray and Corinne A. Manogue, [*The Octonionic Eigenvalue Problem*]{}, Adv. Appl. Clifford Algebras [**8**]{}, 341–364 (1998).
Tevian Dray and Corinne A. Manogue, [*Finding Octonionic Eigenvectors Using [*Mathematica*]{}*]{}, Comput. Phys. Comm. [**115**]{}, 536–547 (1998).
Tevian Dray and Corinne A Manogue [*The Exceptional Jordan Eigenvalue Problem*]{}, Internat. J. Theoret. Phys. (1999; to appear).
Corinne A. Manogue and Jörg Schray, [*Finite Lorentz transformations, automorphisms, and division algebras*]{}, J. Math. Phys. [**34**]{}, 3746–3767 (1993).
Corinne A. Manogue and Tevian Dray, [*Octonionic Möbius Transformations*]{}, Mod. Phys. Lett. [**A14**]{}, 1243–1255 (1999).
Tevian Dray, Jason Janesky, and Corinne A. Manogue, [*Octonionic Hermitian Matrices with Non-Real Eigenvalues*]{} (in preparation).
[^1]: Much of this material was presented in an invited talk entitled [*Choosing a Preferred Complex Subalgebra of the Octonions*]{} given at the [**5th International Conference on Clifford Algebras and their Applications in Mathematical Physics**]{} in Ixtapa, MÉXICO, in June 1999.
[^2]: For later compatibility with our octonion conventions we use $\ell$ rather than $i$ to denote the complex unit.
[^3]: The use of ${{\hbox{\boldmath$\vec{\imath}$}}}$, ${{\hbox{\boldmath$\vec{\jmath}$}}}$, ${{\hbox{\boldmath$\vec{k}$}}}$ for Cartesian basis vectors originates with the quaternions, which were introduced by Hamilton as an early step towards vectors [@Crowe]. Making the obvious identification of vectors ${{\hbox{\boldmath$\vec{v}$}}}$, ${{\hbox{\boldmath$\vec{w}$}}}$ with imaginary quaternions $v$, $w$, then the real part of the quaternionic product $vw$ is (minus) the dot product ${{\hbox{\boldmath$\vec{v}$}}}\cdot{{\hbox{\boldmath$\vec{w}$}}}$, while the imaginary part is the cross product ${{\hbox{\boldmath$\vec{v}$}}}\times{{\hbox{\boldmath$\vec{w}$}}}$.
[^4]: A number of authors, such as [@KT], have used this approach to describe supersymmetric theories in 10 dimensions. Fairlie & Manogue [@FairlieI; @FairlieII] and Manogue & Sudbery [@Sudbery] described solutions of the superstring equations of motion using octonionic parameters, and Schray [@Schray; @Thesis] described the superparticle. A more extensive bibliography appears in [@Schray].
[^5]: We assume here that the components of our spinors are [*commuting*]{}, as we believe that the anticommuting nature of fermions may be carried by the octonionic units themselves. An analogous result for anticommuting spinors was obtained in both of [@FairlieII; @Schray].
[^6]: It is [*not*]{} true in general [@Eigen; @Find] that the determinant of an [$n \times n$]{} Hermitian octonionic matrix is the product of its (real) eigenvalues, unless $n=2$; however, see also [@Other].
[^7]: The last equality is more usually discussed at the Lie algebra level. Manogue & Schray [@Lorentz] gave an explicit representation using this language of the [*finite*]{} Lorentz transformations in 10 spacetime dimensions. For further discussion of the notation $SL(2,{{\mathbb O}})$, see also [@Mobius].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'This paper deals with a precise description of the region of [*zitterbewegung*]{} below the Compton scale and the stochastic nature associated with it. We endeavour to delineate this particular region by dint of [*Ito’s calculus*]{} and instigate certain features that are in sharp contrast with conventional physics.'
author:
- |
B.G. Sidharth[^1], B.M. Birla Science Centre,\
Adarsh Nagar, Hyderabad - 500 063, India\
\
Abhishek Das[^2], B.M. Birla Science Centre,\
Adarsh Nagar, Hyderabad - 500 063, India\
title: The Zitterbewegung Region
---
Introduction
============
The phenomenon of [*zitterbewegung*]{} has been a subject of widespread interest since Schrodinger [@Erwin] proposed the said concept. Many authors [@huang; @David; @Barut] including the author Sidharth [@bgs1; @bgs2] have studied such a phenomenon extensively. Particularly, the affinity of the [*zitterbewegung*]{} and the Compton scale has been studied by Sidharth, where the Compton scale has been considered to be elementary rather than the Planck scale.\
The approach presented in this paper is distinguished in the sense that we identify the entire space to be comprised of an interior and an exterior region separated by a Jordan curve, which has been discussed in the second section. Our endeavours are related to the interior region which is must be treated in a meticulous and precise manner. The methodology used is that which was introduced by Kiyosi Ito [@Ito1]. Ito’s calculus has been implemented to describe the stochastic nature of the interior region in the section 3.\
Section 4 deals with the various transforms (Laplace, Fourier) that are duly modified in the interior region, on account of Ito’s stochastic nature. The fifth section brings out the junction of classical and quantum physics that arises in this interior region and some other related aspects.\
Finally, we discuss the various aspects of the interior region and how it can be looked upon as the region of juncture between classical and quantum mechanics. The quantum mechanical spin has been shown to arise from our considerations in the Compton scale, which therefore establishes the Compton scale in a fundamental level.
The Jordan Curve
================
Apart from the previous papers by Sidharth, in some other papers [@bgs3; @bgs4; @bgs5] it was found that there is indeed a transition from the [*Planck scale*]{} to the [*Compton scale*]{}, from whence we have the [*coherence length*]{} or [*coherence parameter*]{} [@bgs6; @bgs7] as\
$$\xi = \frac{h}{mc} = \frac{2\pi\hbar}{mc}$$\
Or,\
$$\xi = 2\pi l_{c}$$\
where, $l_{c} = \frac{\hbar}{mc}$ is the Compton length of a particle. Now, this can be looked upon as the circumference of a circle of radius $l_{c}$ or merely, a nonintersecting closed curve whose length is $2\pi l_{c}$. In a generalized manner, we can say that this is a [*Jordan curve*]{} ($J$) that has two connected components: one being the [*interior region*]{} ($I$) bounded by $J$ and the other being the [*exterior region*]{} ($E$).\
Now, since the sets $E$ and $I$ are disjoint and $J$ is the boundary separating them, we can infer that together they form a disconnected set $S$ which is the set of the whole space. Hereby, we begin our theory by making the presumption that the interior region ($I$) can be visualized as being below the Compton scale and the exterior region ($E$) being above the Compton scale, separated by the [*Jordan curve*]{} ($J$) whose length is given by the [*coherence parameter*]{} as above. This [*coherence parameter*]{} ($\xi$) being related to the Compton length is the measure of a [*fundamental minimum length*]{}.\
Suppose, the state describing the exterior ($E$) is $\psi_{E}$ and that describing the interior ($I$) is $\psi_{I}$. Therefore, if $\psi_{S}$ is the state describing the entire space then we can write\
$$\psi_{S} = \psi_{E} + \psi_{I}$$ Now, if $\psi_{S1}$ and $\psi_{S2}$ be represent two different states of the set $S$ with probability $P_{S1}$ and $P_{S2}$. Thus, the density operator would be given as\
$$\rho = P_{S1} |\psi_{S1}><\psi_{S1}| + P_{S2} |\psi_{S2}><\psi_{S2}|$$ Using equation (1) we have\
$$\rho = P_{S1} |\psi_{E1} + \psi_{I1}><\psi_{E1} + \psi_{I1}| + P_{S2} |\psi_{E2} + \psi_{I2}><\psi_{E2} + \psi_{I2}|$$ Here, we would like to mention that according to the Feshbach-Villars approach [@Feshbach] Sidharth had shown [@bgs8; @bgs9] that if we invoke a two component wavefunction\
$$\psi = \left(\matrix {\phi\cr \chi}\right)$$\
where $\phi$ and $\chi$ represent the positive and negative energy solutions respectively, then above the Compton scale the $\phi$ component is predominant and below the Compton scale the $\chi$ component is predominant. In our approach, we infer that the $\phi$ component dominates in the $E$-region and the $\chi$ component dominates in the $I$-region, although both components are present in the said regions.\
Now as we can see from equation (3), the state describing the interior region ($I$) is present in the density matrix. So, if the space $S$ is studied on the whole then all we get is a mixed description of the two regions. Now, this is not feasible as we shall argue later, because the interior region is in it’s entirety a completely different picture. Due to [*zitterbewegung*]{} effects the processes going on in the region $I$ is stochastic and in this sense it is distinguished from the region $E$. The [*Jordan curve*]{} ($J$) not only separates the $I$-region from the $E$-region, but also poses some restrictions on the region $I$. In a nutshell, this implies that the methodology used to describe the $E$-region cannot be implemented in case of the $I$-region. The [*Jordan curve*]{} ($J$) whose magnitude is related to the Compton length binds the stochastic effects below the Compton scale. Consequently, equation (2) will not be able to provide a feasible description of the state of the whole space $S$. Also, the processes going on in the $E$-region will not be affected by those of the $I$-region.\
All of this suggests that the regions $E$ and $I$ must be studied separately in a meticulous manner, in order to get a better conception of the space $S$ on the whole. Again, it is interesting to point out the fact that if the space $S$ was connected, i.e. the curve $J$ didn’t exist or more precisely the [*coherence parameter*]{} wasn’t epitomized as a [*fundamental minimum length*]{} then we would simply have the density operator as\
$$\rho = P_{S1} |\psi_{S1}><\psi_{S1}| + P_{S2} |\psi_{S2}><\psi_{S2}|$$ where, $\psi$ simply equals to $\psi_{S}$. The entire space $S$ would be described merely by $\psi_{S}$. But, as we know matters aren’t so simple and thus we would like to study elaborately the stochastic processes going on inside the region $I$.\
The Interior Region
===================
As we mentioned in the preceding section, the interior region ($I$) is a bounded region. All the processes going on in it are stochastic due to [*zitterbewegung*]{} effects. The processes could be classical or quantum in nature but they are always stochastic and epitomize the uniqueness of the $I$-region. Also, these can be looked upon as [*Ito processes*]{} ($p$) [@Ito1; @Ito2] culminating from Brownian-type motion [@Wim] of the particles. Suppose, $\phi_{t}$ describes the processes $p_{t}$ at the instant ’$t$’ such that\
$$\phi_{t} = \exp(\pm p_{t})$$\
Or, $$p(\phi_{t}) = p_{t} = \mp \ln\phi_{t}$$ where, $\phi_{t}$ is itself a stochastic process due to random fluctuations going on in every aspect. Also, we consider a [*Wiener process*]{} $W_{t}$ [@Wim] which itself is a stochastic process such that\
$$W_{t} = \int_{0}^{t} f(p_{t}){\rm d}\phi_{t}$$ This is [*Ito’s integral*]{} [@Ito1; @Anders] which defines [*Ito’s calculus*]{} and $f(p_{t})$ is another process which is adapted to the filtration generated by $\phi_{t}$. Here ’filtration’ means the set of information available up to the time ’$t$’. This means that a stochastic process depends on the information available till the instant the said process occurs. More precisely, a stochastic process depends on the previous stochastic processes. We shall endeavour to prove this more rigorously.\
Let us consider [*Ito’s lemma*]{} for the stochastic [*drift diffusion*]{} processes where the process $p(\phi_{t})$ do not explicitly depend on time ($\frac{\partial p(\phi_{t})}{\partial t} = 0$), such that\
$${\rm d}p(\phi_{t}) = p^{\prime}(\phi_{t}){\rm d}\phi_{t} + \frac{1}{2} p^{\prime\prime}(\phi_{t})\sigma_{t}^{2}\phi_{t}^{2}{\rm d}t$$ where, the primes ($\prime$) denote partial derivatives with respect to $\phi_{t}$ and the expectation\
$$E[{\rm d}\phi_{t}^{2}] = \phi_{t}^{2}{\rm d}t$$\
In parallel to this, as an analogue to methods used in stochastic mathematical finance [@Mall] we consider the processes ($\phi_{t}$) to be characterized by [*geometric Brownian motion*]{}, such that\
$${\rm d}\phi_{t} = \phi_{t}[\mu{\rm d}t + \sigma_{t}{\rm d}W_{t}]$$ where, $\mu$ is the mean or [*drift*]{} and $\sigma_{t}$ is an infinitesimal standard deviation. Now, in the $E$-region where stochastic processes have no relevance ($W_{t} = 0$), this equation will be simply\
$${\rm d}\phi = \phi\mu{\rm d}t$$\
and with ordinary calculus have the solution\
$$\phi_{t} = \phi_{0}\exp[\pm\mu t]$$ But, in the $I$-region when we resort to Ito’s stochastic calculus the solution differs a lot. Putting equation (8) in (7) we have\
$${\rm d}p_{t} = p^{\prime}\phi_{t}[\mu{\rm d}t + \sigma_{t}{\rm d}W_{t}] + \frac{1}{2} p^{\prime\prime}\sigma_{t}^{2}\phi_{t}^{2}{\rm d}t$$ Using relation (5) we have\
$$\mp{\rm d}(\ln\phi_{t}) = \mu{\rm d}t + \sigma_{t}{\rm d}W_{t} \pm \frac{1}{2}\sigma_{t}^{2}{\rm d}t$$ Thus, we obtain finally the solution\
$$\phi_{t} = \phi_{0}\exp[\pm\{\sigma_{t}W_{t} + (\mu \pm \frac{\sigma_{t}^{2}}{2})t\}]$$ It is obvious that this solution is entirely different from that of the solution give in equation (9). This is a stringent reason as to why the $I$-region must be looked upon differently from than the $E$-region. Due to the stochastic nature of the processes going on in the $I$-region the use of Ito’s calculus yields different results [@Hui].\
Now, in terms of partial derivatives we can write equation (8) as\
$$\frac{\partial W_{t}}{\partial \phi_{t}} = \frac{1}{\sigma_{t}}[\frac{1}{\phi_{t}} - \frac{\mu}{\frac{\partial \phi_{t}}{\partial t}}]$$\
Again, using the solution (12) this gives us\
$$\frac{\partial W_{t}}{\partial \phi_{t}} = \frac{1}{\sigma_{t}}[\frac{1}{\phi_{t}} - \frac{\mu}{\phi_{0}(\mu - \frac{\sigma_{t}^{2}}{2})}]$$ This can also be written as\
$$\frac{\partial W_{t}}{\partial \phi_{t}} = f(\phi_{t})$$\
where, $f(\phi_{t}) = \frac{1}{\sigma_{t}}[\frac{1}{\phi_{t}} - \frac{\mu}{\phi_{0}(\mu - \frac{\sigma_{t}^{2}}{2})}]$. Therefore, we have\
$$W_{t} = \int_{0}^{t} f(\phi_{t}) {\rm d}\phi_{t}$$ where, $f(\phi_{t})$ is another stochastic Ito process. This methodology can be continued to find a succession of such stochastic processes. Also, the process $f(p_{t})$ in equation (6) can be traced back to another stochastic Ito process. Therefore, we can conclude that in the $I$-region there are plethora of stochastic processes going on and all of them are connected to each other. The chain of such processes carries on indefinitely as one process triggers the start of another process and so on. This is the underlying feature of the $I$-region that distinguishes it from the $E$-region and all of this is caused due to the [*Jordan curve*]{} ($J$) that acts as the barrier between the aforesaid regions.\
Now, let us consider any two arbitrary processes $p(\phi_{t1})$ and $p(\phi_{t2})$ occurring at two instants $t_{1}$ and $t_{2}$. If $Pr[p(\phi_{t1})]$ and $Pr[p(\phi_{t2})]$ are the individual probabilities of the processes then in terms of conditional probability it is obvious from our results that\
$$Pr[p(\phi_{t2})] \neq Pr[p(\phi_{t2})|p(\phi_{t1}))]$$ since the processes are connected. Thus, we can conclude that all the processes in the $I$-region are dependent of one another. Again, the processes in the $E$-region are not connected to those in the $I$-region. Therefore, for a process $X$ in the $E$-region and a process $p(\phi_{t})$ in the $I$-region we would have\
$$Pr[X] = Pr[X|p(\phi_{t})]$$ Consequently, as we mentioned earlier, when one studies the space $S$ the regions $E$ and $I$ must be studied separately and with relevant methods: ordinary calculus for the $E$-region and Ito’s calculus for the $I$-region. Otherwise, there would be discrepancies and one would not be able to get the true picture of the whole space. The reason is obviously the stochastic nature of the $I$-region which emanates from the different nature of the physics inside the $I$-region.\
The Laplace transform in the [*I*]{}-region
===========================================
Now, let us consider that the interior ($I$) region is a subset of [**C**]{}, the complex plane. If ’$s$’ and ’$t$’ denote the frequency and time respectively then the Laplace transform of a function $f(t)$ is be given as\
$$F(s) = L[f(t)] = \int_{0}^{\infty} e^{-st}f(t) {\rm d}t$$ Now, suppose that in the $I$-region $f(t)$ and $F(s)$ represent stochastic processes as functions of time and frequency respectively. Then, we would like to see the nature of this transform when [*Ito’s calculus*]{} is taken into consideration. One may write equation (17) as\
$${\rm d}F(s) = e^{-st}f(t){\rm d}t$$\
But, for the $I$-region if we consider [*Ito’s calculus*]{} then we may write\
$${\rm d}F(s) = \dot{F}(s){\rm d}t + \frac{1}{2}F^{\prime\prime}(s)\sigma_{t}^{2}s^{2}{\rm d}t$$ where the prime ($\prime$) denotes derivative with respect to the frequency ($s$) and the [*dot*]{} denotes partial derivative with respect to time ($t$), assuming that the $F(s)$ depends explicitly on ’$t$’ only ($\frac{\partial F(s)}{\partial s} = 0$). Here\
$$\dot{F}(s) = e^{-st}f(t)$$\
Now, in the region outside $I$, the second term on the right hand side of equation (18) does not exist because there we have the conventional calculus. But, in the $I$-region considering [*Ito’s calculus*]{} the second term is of non-trivial importance. Integrating equation (18) we have\
$$F(s) = \int_{0}^{\infty} F^{\prime}(s){\rm d}t + \frac{\sigma_{t}^{2}}{2}\int_{0}^{\infty} s^{2}\frac{\rm d}{{\rm d}s}[e^{-st}f(t)]{\rm d}t$$\
Assuming, the frequency is independent of the integration with respect to time, this gives\
$$F(s) = L[f(t)] - \frac{\sigma_{t}^{2}}{2}s^{3} \int_{0}^{\infty} e^{-st}f(t) {\rm d}t + \frac{\sigma_{t}^{2}}{2}s^{2} \int_{0}^{\infty} e^{-st}h(t) {\rm d}t$$\
where, $h(t) = f^{\prime}(t)$ is another stochastic processes. Therefore, we have\
$$F(s) = L[f(t)] + \frac{\sigma_{t}^{2}}{2} (s^{2}L[h(t)] - s^{3}L[f(t)])$$ More rigorously, this can be written as\
$$F(s) = L[f(t)] + \frac{\sigma_{t}^{2}}{2} G(s, t)$$ where, $G(s, t) = s^{2}L[h(t)] - s^{3}L[f(t)]$ is another stochastic process. This result substantiates that in the $I$-region the expression for the Laplace transform is modified due to the host of stochastic processes going on. Ostensibly, outside the $I$-region we would have $\sigma_{t} = 0$, and consequently then we would get back the original Laplace transform. Now, from equation (19) we can also infer that the Laplace transform of a stochastic process in the complex plane is dependent of other stochastic processes going on in the $I$-region. This is exactly in favour of what we derived in the previous section.\
Another interesting conclusion that we can draw from this result is that for other transforms also such stochastic effects would be present. Since, the Fourier, Mellin transforms are related to the bilateral or two-sided Laplace transform one can easily conclude a stochastic nature of the Fourier and Mellin transforms. The same can be deduced for the Borel transform and the z-transform.\
The juncture of Classical and Quantum mechanics
===============================================
Although, the present approach is purely classical it is interesting to point out the fact that the author Sidharth had used a slightly similar technique distinguishing the interior region from the exterior region. Averaging over the interior region one arrives at the domain of conventional physics. In fact, the author was also able to derive the Compton length from the classical point of view, averaging over the physically inaccessible interior region. This coincides with our approach, since we have considered the magnitude of the boundary curve $J$ to be related to the Compton length itself, via the [*coherence parameter*]{} ($\xi$).\
In accordance with a previous work of the author Sidharth [@bgs10] and Moller [@Moller] we would like to stress another key feature of the interior region. It can be shown that in the relativistic case, if we consider the $I$-region as the rest system then we have a host of mass centres ($C_{i}$’s) that form a two dimensional circular disk perpendicular to the direction of the angular momentum ($L = l \times p$) of the system with centre at the proper centre of mass. The radius of such a disk is given as\
$$r = \frac{L}{mc} = \frac{l \times p}{mc}$$\
Now, if the momentum corresponds to the De Broglie wavelength of the system ($I$-region) such that\
$$p = \frac{h}{l}$$\
then, we have\
$$r = 2\pi \frac{\hbar}{mc} = 2\pi l_{c}$$ which is exactly the [*coherence parameter*]{} ($\xi$). Since, the Compton length ($l_{c}$) is a purely quantum mechanical we find that due to relativistic considerations classical and quantum physics have an affine connection. More interestingly, it was also shown by the author Sidharth [@bgs10; @bgs11] that we can indeed obtain the quantum mechanical spin starting from classical physics, in the Compton region which is basically the $I$-region in this paper. As stated by Sidharth in light of Dirac’s approach [@Dirac], the interior region represents itself as a doubly connected space which has a nodal singularity or quantized vortex which gives rise to the quantum mechanical spin. The $I$-region is inherently the region of [*zitterbewegung*]{} and inside this region we realize the juncture of classical and quantum mechanics.\
We would like to obtain further insight regarding this subject. Now, it is known that for two different solutions $\psi_{1}(x)$ and $\psi_{2}(x)$ of the Dirac equation\
$$(\gamma^{\mu}p^{\mu} - mc)\psi(x) = 0$$\
we have due to linearity [@Bjorken]\
$$c\bar{\psi}_{2}\gamma^{\mu}\psi_{1} = \frac{1}{2m}[\bar{\psi}_{2}p^{\mu}\psi_{1} - (p^{\mu}\bar{\psi}_{2})\psi_{1}] - \frac{i}{2m}p_{\nu}(\bar{\psi}_{2}\sigma^{\mu\nu}\psi_{1})$$ where, $\gamma^{\mu}$’s are the gamma matrices and $\sigma^{\mu\nu} = \frac{i}{2}[\gamma^{\mu}, \gamma^{\nu}]$. It is also known that this equation leads to the considerations of both positive and negative energy solutions. Now, in equation (22) we identify the term consisting $\sigma^{\mu\nu}$ as the term arising due to the motion about the proper centre of mass (as discussed above). Consequently, the momentum associated with it can be identified as the spin angular momentum due to the presence of $\sigma^{\mu\nu}$ which consists of the gamma matrices. Interestingly, calculating the average current which is given by the expectation value of the velocity operator $c\alpha$ \[where, $\alpha_{i} = \left(\matrix {0 & \sigma_{i}\cr \sigma_{i} & 0}\right)$\] yields cross terms between both the positive and negative energy solutions that fluctuate rapidly leading to the phenomenon of [*zitterbewegung*]{}.\
These negative energy solutions are localized within the $I$-region (of radius or extension $l_{c}$) and bounded by the [*Jordan curve*]{} (of length = $2\pi l_{c}$). Therefore, we see that the quantum mechanical spin arises quite elegantly from our considerations in the $I$-region, when we take into account the motion about the centre of proper mass of the system. Now, it must be stringently borne in mind that outside the $I$-region the negative energy solutions are dominated by the positive energy solutions and they are no longer localized. Consequently, the phenomenon of [*zitterbewegung*]{} has no effects in the $E$-region.
As we discussed previously, it was shown by the author Sidharth [@bgs8; @bgs9] that the in the $I$-region the negative energy solutions dominate whereas in case of the $E$-region the positive energy solutions dominate. This feature of the $I$-region was also discussed recently [@bgs7] from the context of an [*Ising-like model*]{}.
Discussions
===========
1) Continuing in the vein of Section 5, we observe that as John Wheeler put it, it is the spin half that separates Quantum Mechanics from Classical Physics [@wheeler]. The spin half is also connected with the doubly connected space feature of Quantum Theory as brought out by Fermionic four spinor wave functions. More explicitly [@schweber], a scalar is a one component object of the rotation rule, under which $$\xi \to \xi' = \xi$$ On the other hand a spinor (of rank 1) is a two component object which transforms according to $$\xi \to \xi' = (1 + \frac{1}{2} \imath \epsilon \sigma_{i})
\xi\label{110}$$
under the rotation rule, i.e. a rank 1 spinor gets transformed by a $2 \times 2$ unitary matrix (determinant = 1). In the same vein a vector is a three component object. In the case of the Quantum Mechanical four spinor, $$\left(\begin{array}{ll}
\phi\\
\chi\end{array}\right)$$ alluded to above, the two spinor $\chi$ transforms according to $$\chi \to - \chi$$ under reflections while $$\phi \to \phi$$ under the same reflection. Here, four dimensional reflection of the Dirac matrices is essential because the usual laws of electromagnetic theory are invariant under reflections [@heine].\
The Quantum Mechanical wave function defines a doubly connected space. In this sense, rather than ordinary physics simply connected with space, herein comes the difference between Classical Physics and Quantum Physics.\
2) Also, an interesting fact can be mentioned in accordance with our work. The author Sidharth in some of his papers [@bgs8; @bgs9] had considered a double Wiener process (random forward-backward motion of time) that lead to a complex velocity $V - \imath U$, which means that the conventional space coordinate $x$ is replaced as\
$$x \rightarrow x + \imath x^{\prime}$$\
where, $x^{\prime}$ is an arbitrary function of time, i.e. we have a new imaginary coordinate (in 1D). Let us discuss this a little elaborately. It has been show by the author Sidharth that within the Compton scale, the [*zitterbewegung*]{} or self interaction effects gives rise to the inertial mass. In such a fundamental domain the fractal nature of space-time and a stochastic underpinning are found to be correlated. Particularly, in our work the $I$-region is such a domain and there we have time as a back and forth motion leading to a double Wiener process. In the vein of such stochastic nature of time the forward and backward velocities can be defined as\
$$\frac{{\rm d}_{+}}{{\rm d}t}x(t) = b_{+}$$\
$$\frac{{\rm d}_{-}}{{\rm d}t}x(t) = b_{-}$$\
This leads to the well known Fokker-Planck equations \[\] as below\
$$\frac{\partial \rho}{\partial t} + \nabla(\rho b_{+}) = V\Delta\rho$$ and\
$$\frac{\partial \rho}{\partial t} + \nabla(\rho b_{-}) = -U\Delta\rho$$ where, we have defined\
$$V = \frac{b_{+} + b_{-}}{2}; U = \frac{b_{+} - b_{-}}{2}$$\
Using equations (1) and (2) we have\
$$\frac{\partial \rho}{\partial t} + \nabla(\rho V) = 0$$ and $$U = \nu\nabla\ln\rho$$ where, $\nu = lv$ is the diffusion constant, $l$ is the mean free path and $v$ is the velocity. Here, $V$ and $U$ are the statistical averages of the corresponding velocities. Introducing the following identifications\
$$V = 2\nu\nabla S$$ and $$V - \imath U = -2\imath\nu\nabla(\ln \psi)$$ we get the complex wavefunction $$\psi = \sqrt{\rho}e^{\frac{\imath S}{\hbar}}$$ which is the solution to the Schrodinger wave equation of quantum mechanics. Now, as can be seen we have introduced the complex velocity $V -\imath U$ which insinuates that the ordinary coordinates will get complexified as below:\
$$x \rightarrow x + \imath x^{\prime}$$\
where, the real part of the velocity would be\
$$\frac{{\rm d}X_{r}}{{\rm d}t} = V$$\
and the imaginary part would be given as\
$$\frac{{\rm d}X_{i}}{{\rm d}t} = U$$\
Therefore, we have the total Wiener-type velocity as\
$$W = \frac{{\rm d}}{{\rm d}t}(X_{r} - \imath X_{i})$$ Taking, $x^{\prime} = ct$ we finally find that the Minkowski coordinate system ($x, \imath ct$) emerges very elegantly and so does special relativity. One may apply the methodology of [*quaternions*]{} to generalize this concept to 3D. As it was shown by the author Sidharth that this leads to four dimensions automatically, considering the Pauli matrices.\
Thus, due to the random nature of a double Wiener process, the coordinates get complexified inside the $I$-region and as a consequence we obtain the roots of inertial mass, special relativity and quantum mechanics. This alone emphasizes the fundamental nature of the $I$-region and the Jordan curve (boundary of a fundamental minimum length).\
It is worth noting that equation (7) resembles a covariant derivative that epitomizes the geometry of the $I$-region itself. From this covariant derivative one can understand the underlying fundamental structure of space-time.\
Also, many years ago it was shown by Ezra Newman [@Ezra] that if one considers the extension of Maxwell’s equations into a complex Minkowski space and their invariance under the complex Poincare group then new solutions can be found. The precise value of the gyromagnetic ratio ($g = 2$) according to Dirac’s theory can also be explained elegantly. Actually, from the derivations of our current paper we can indeed infer that such fundamental features of quantum mechanics begin in the $I$-region and are extended to the $E$-region, but with stochastic effects distinguishing the two different regions of physics.
3) To see all this in the context of Dirac’s nodal singularity let us start with $$\psi = \psi_1 e^{\imath S},\label{5xe2}$$ where $\psi_1$ is the usual wave function with what may be called an integrable phase. Further the phase $S$ does not have a definite value at each point. The four gradient $$K^\mu = \partial^\mu S\label{5xe3}$$ is well defined. We use temporarily natural units, $\hbar = c = 1$. Dirac then goes on to identify $K$ in (\[5xe3\]) (except for the numerical factor $hc/e$) with the electromagnetic field potential, as in the Weyl gauge invariant theory.\
Next Dirac considered the case of a nodal singularity, which is closely related to what was later called a quantized vortex a term we have already noted (Cf. for example ref.[@vasu]). In this case a circuit integral of a vector as in (\[5xe3\]) gives, in addition to the electromagnetic term, a term like $2 \pi n$, so that we have for a change in phase for a small closed curve around this nodal singularity, $$2 \pi n + e \int \vec B \cdot d \vec S\label{5xe4}$$ In (\[5xe4\]) $\vec B$ is the magnetic flux across a surface element $d \vec S$ and $n$ is the number of nodes within the circuit. The expression (\[5xe4\]) directly lead to the Monopole in Dirac’s formulation.\
Let us now reconsider the above arguments in terms of our earlier developments. As we saw the Dirac equation for a spin half particle throws up a complex or non Hermitian position coordinate [@bgsfpl172004; @bgsfpl162003]. Dirac as noted identified the imaginary part with zitterbewegung effects and argued that this would be eliminated when averages over intervals of the order of the Compton scale are taken to recover meaningful physics [@diracpqm]. Over the decades the significance of such cut off space time intervals has been stressed by T.D. Lee and several other scholars [@bgs10; @kard; @bom; @leepl]. Indeed we saw that with a minimum cut off length $l$, it was shown by Snyder that there would be a non commutative but Lorentz invariant spacetime structure. At the Compton scale we would have [@uof], $$[x,y] = 0(l^2)\label{5xe5}$$ and similar relations.\
In fact starting from the Dirac equation itself, we deduced directly the non commutativity (\[5xe5\]) (Cf.refs.[@bgsfpl172004; @bgsfpl162003]).\
Let us now return to Dirac’s formulation of the monopole in the light of the above comments. As noted above, the non integrability of the phase $S$ in (\[5xe2\]) gives rise to the electromagnetic field, while the nodal singularity gives rise to a term which is an integral multiple of $2 \pi$. As is well known [@heap] we have $$\vec \nabla S = \vec p\label{5xe6}$$ where $\vec p$ is the momentum vector. When there is a nodal singularity, as noted above, the integral over a closed circuit of $\vec p$ does not vanish. In fact in this case we have a circulation given by $$\Gamma = \oint \vec \nabla S \cdot d \vec r = \hbar \oint dS = 2 \pi
n\label{5xe7}$$ It is because of the nodal singularity that though the $\vec p$ field is irrotational, there is a vortex - the singularity at the central point associated with the vortex makes the region multiply connected, or alternatively, in this region we cannot shrink a closed smooth curve about the point to that point. In fact if we use the fact as seen above that the Compton wavelength is a minimum cut off, then we get from (\[5xe7\]) using (\[5xe6\]), and on taking $n = 1$, $$\oint \vec \nabla S \cdot d\vec r = \int \vec p \cdot d \vec r =
2\pi mc \frac{1}{2mc} = \frac{h}{2}\label{5xe8}$$ $l = \frac{\hbar}{2mc}$ is the radius of the circuit and $\hbar = 1$ in the above in natural units. In other words the nodal singularity or quantized vortex gives us the mysterious Quantum Mechanical spin half (and other higher spins for other values of $n$). In the case of the Quantum Mechanical spin, there are $2 \times n/2 + 1 = n + 1$ multiply connected regions, exactly as in the case of nodal singularities.
[99]{} E. Schrodinger, Sitzungber. Preuss. Akad. Wiss. Phys.-Math. Kl. 24, 418 (1930). K. Huang, *On the zitterbewegung of the electron*, Am. J. Phys. 47, 797 (1949). David Hestenes, *The Zitterbewegung Interpretation of Quantum Mechanics*, Found. Physics., Vol. 20, No. 10, (1990) 1213-1232. A. O. Barut and A. J. Bracken, *Zitterbewegung and the internal geometry of the electron*, Phys. Rev. D 23, 2454 (1981). B.G. Sidharth, *ZPF, Zitterbewegung and Inertial Mass* Int. J. Mod. Phys. E, 18, 1863 (2009). B.G. Sidharth, *Revisiting Zitterbewegung*, IJTP, February 2009, Volume 48, Issue 2, pp 497-506. Kiyosi Ito, *Stochastic Integral*, Proc. Imperial Acad. Tokyo 20, 519-524, 1944. B.G. Sidharth, *Gravitation and Electromagnetism*, Nuovo. Cim. B116, 735 (2001). B.G. Sidharth, *The New Cosmos*, Chaos, Solitons and Fractals, vol. 18, no. 1, pp. 197-201, 2003. B.G. Sidharth, *An Underpinning for Spacetime*, Chaos, Solitons and Fractals 25, pp.965-968, 2005. B.G. Sidharth, *Gravitation and Electromagnetism*, arXiv:physics/0106051v2 \[physics.gen-ph\] 24 Dec 2015. B.G. Sidharth and Abhishek Das, *2 State Asymmetries: Matter, Antimatter and Other Effects*, IJMPA, Vol. 31, No. 32 (2016) 1650167.
H. Feshbach and F. Villars, *Elementary Relativistic Wave Mechanics of Spin 0 and Spin 1/2 Particles* Rev.Mod.Phys. Vol.30, No.1, January 1958, pp.24-45. B.G. Sidharth, *A Note on a Particle-Antiparticle Interaction*, IJMPE, Vol. 20, No. 9, 2011. B.G. Sidharth, *Negative Energy Solutions and Symmetries*, IJMPE, Vol. 20, No. 10, 2011. Kiyosi Ito, *On stochastic differential equations*, Memoirs, American Mathematical Society 4, 1-51, 1951. Wim Vervaat, *A Relation between Brownian Bridge and Brownian Excursion*, The Annals of Probability, Vol. 7, No. 1 (Feb., 1979), pp. 143-149. Anders Bolund and Klaus Molmer, *How many atoms get excited when they decay?*, Phys. Rev. A 89, 023827 (2014). A. G. Malliaris, *Ito’s Calculus in Financial Decision Making*, SIAM Review, Vol. 25, No. 4 (Oct., 1983), pp. 481-496. Hui-Hsiung Kuo, *Introduction to Stochastic Integration*, Springer, 2006. B.G. Sidharth, *The Chaotic Universe: From the Planck to the Hubble Scale*, pp. 21-22, Nova Science, New York (2001). C. Moller, *The Theory of Relativity*, pp. 170-173, Oxford University Press, 1952. B.G. Sidharth, *The Thermodynamic Universe* (World Scientific, Singapore, 2008), pp. 176-177. P.A.M. Dirac, *An Extensible model of the electron*, Proc.Roy.Soc.Lond. A268, pp. 57-67, 1962. J.D. Bjorken and S.D. Drell, [*Relativistic Quantum Mechanics*]{}, Mc-Graw Hill, New York, 1964, pp. 35-40. J.A. Wheeler, *Superspace and the Nature of Quantum Geometrodynamics*, *Battelles Rencontres, Lectures*, De Witt, B.S. and Wheeler, J.A. (eds.) (Benjamin, New York, 1968). S.S. Schweber, *An Introduction to Relativistic Quantum Field Theory* (Harper and Row, New York, 1961). V. Heine, *Group Theory in Quantum Mechanics* (Pergamon Press, Oxford, 1960), pp.364. Ezra T. Newman, *Maxwell’s Equations and Complex Minkowski Space*, Journal of Mathematical Physics, Vol. 14, No. 1, 1973. R. Vasudevan, *Hydrodynamical Formulation of Quantum Mechanics*, in *Perspectives in Theoretical Nuclear Physics*, Srinivas Rao, K., and Satpathy L. (eds.) (Wiley Eastern, New Delhi, 1994), pp.216ff. B.G. Sidharth, *Found.Phys.Lett.* 17, (5), 2004, pp.503–506. B.G. Sidharth, *Found.Phys.Lett.* 16, (1), 2003, pp.91–97. P.A.M. Dirac, *The Principles of Quantum Mechanics* (Clarendon Press, Oxford, 1958), pp.4ff, pp.253ff. V.G. Kadyshevskii, *Translated from Doklady Akademii Nauk SSSR* Vol.147, No.6 December 1962, pp.1336–1339. L. Bombelli, J. Lee, D. Meyer, D. and R.D. Sorkin, *Physical Review Letters* Vol.59, No.5, August 1987, pp.521–524. T.D. Lee, *Physics Letters* Vol.122B, No.3,4, 10 March 1983, pp.217–220. B.G. Sidharth, *The Universe of Fluctuations* (Springer, Netherlands, 2005). B.G. Sidharth, *Chaos, Solitons and Fractals* (12), (1), 2000, pp.173–178.
[^1]: [email protected]
[^2]: [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We offer a brief response to the criticisms put forward by Cusin et al in Ref. [@Cusin:2018ump] about our work Refs. [@Jenkins:2018kxc; @Jenkins:2018uac], emphasising that none of these criticisms are relevant to our main results.'
author:
- 'Alexander C. Jenkins'
- Mairi Sakellariadou
- Tania Regimbau
- Eric Slezak
- 'Richard O‘Shaughnessy'
- Daniel Wysocki
bibliography:
- 'response.bib'
title: 'Response to Cusin et al’s comment on `arXiv:1810.13435`'
---
All of the criticisms raised by Cusin et al in Ref. [@Cusin:2018ump] are about the analytical approach that we introduced in Ref. [@Jenkins:2018uac]. Indeed, we stressed explicitly ourselves throughout Ref. [@Jenkins:2018uac] that this analytical approach is inaccurate. In particular, after introducing the analytical expression, we wrote:
> “We emphasize that Eq. (69) \[the analytical expression\] is only a simple approximation of the true angular spectrum of the anisotropies… We also note that we have extended Eq. (64) \[the power-law approximation of the galaxy-galaxy two-point correlation function\] beyond its realm of validity by assuming that it holds for all distances $d$… We therefore turn to a more detailed and accurate approach in the following section, to address the deficiencies of this simple model.”
The main results of Refs. [@Jenkins:2018kxc; @Jenkins:2018uac] are *not* based on this analytical approximation; they come from a thorough and careful analysis of a large simulated galaxy catalogue, based on the Millennium simulation [@Springel:2005nw; @Blaizot:2003av; @Lemson:2006ee; @DeLucia:2006szx]. The analytical approach was only ever intended as a simplistic first pass at the problem, before doing the full catalogue analysis. While the analytical approach can provide valuable insights (particularly for rapid investigations of e.g. tens of thousands of different astrophysical models [@Jenkins:2018kxc]), it should not be regarded as a confident prediction, whereas the catalogue results should be. It is unambiguously clear that none of the criticisms in Ref. [@Cusin:2018ump] are relevant to our catalogue approach.
Thus, it remains to understand the difference between the results of our catalogue approach and the results of Cusin et al in Ref. [@Cusin:2018rsq]. In order to clarify the difference between the approaches, we list the steps that must go into a first-principles calculation of the $C_\ell$’s of the astrophysical stochastic background, and described how each of these is achieved for us and for Cusin et al:
1. The cosmological perturbations must be evolved from some initial power spectrum at early times, giving the dark matter overdensity field down to redshift zero.
Jenkins et al
: This is done for us by the Millennium simulation, where the N-body gravitational clustering dynamics is simulated numerically, in a way that automatically accounts for all non-linear effects.
Cusin et al
: This is done using the Boltzmann code <span style="font-variant:small-caps;">CMBQuick</span>, which treats the overdensities as small, linearised perturbations around a homogeneous background, and solves the linear evolution equations for these overdensities. The linear approximation is fine for CMB calculations at $z>10^3$ (which is what these types of codes were originally designed for), but breaks down at $z\approx0$, particularly on small scales. Cusin et al attempt to account for non-linear effects at $z\approx0$ by using the HALOFIT algorithm [@Smith:2002dz]; this is essentially an ad-hoc fitting function, calibrated to N-body simulations. However, HALOFIT is known to under-estimate the matter power spectrum in $\Lambda$CDM.[^1] What’s more, the simulations that HALOFIT is based on are older than Millennium, and smaller by a factor of $(2160/256)^3\approx600$.
2. The dark matter haloes must be populated with galaxies, accounting for the fact that the galaxies are more tightly clustered than the haloes themselves.
Jenkins et al
: This is done for us by the simulated galaxy catalogue, which is based on a sophisticated semi-analytical model [@DeLucia:2006szx] that accounts for a whole host of messy astrophysical feedback processes to model the galaxy population and distribution within each dark matter halo.
Cusin et al
: This is done by writing the galaxy power spectrum as a biased form of the matter power spectrum, $P_\mathrm{gal}=b^2P_\mathrm{mat}$. The bias function $b$ is itself based on a linearised approximation, and is assumed to be scale-invariant (which is known to be false, particularly on small scales).
3. The gravitational-wave (GW) emission of each galaxy must be calculated, assuming a particular model for the compact binary populations.
Jenkins et al
: This is done using detailed information from the Millennium simulation about the star formation rate in each galaxy as a function of time, which allows us to calculate the compact binary merger rate at the time of GW emission. We also use information about the metallicity and peculiar velocity of each galaxy, as these influence the observed GW flux.
Cusin et al
: This is done using analytical formulae, treating the emitted GW flux of each galaxy as a function of the mass of the host halo only.
Following steps 1–3, one can then superimpose the GW flux from all the galaxies to calculate the $C_\ell$’s.
In Ref. [@Jenkins:2018kxc] we sample several thousand possible binary black hole (BBH) populations supported by the LIGO/Virgo O1 detections, and find that these only affect the $C_\ell$’s at a level of $\approx3\%$. So it seems that step 3 is not the cause of the discrepancy between us and Cusin et al (they state that they agree with this in Ref. [@Cusin:2018rsq]). One must therefore look at the different approaches to steps 1 and 2. For both steps, we have in the discussion above defended the accuracy of our catalogue approach relative to that of Cusin et al. In particular, their use of linear perturbation theory and a linear, scale-invariant galaxy bias can be expected to lead to a loss of clustering on small scales. This is important because, although we are interested in large angular scales (small $\ell$), most of the astrophysical background comes from low redshifts ($z\lesssim1$), and the anisotropies are dominated by the very lowest redshifts (this is because changing the number of galaxies in a given direction at low redshifts causes a much more significant fluctuation than at higher redshifts, as the GW flux from each individual galaxy is much greater—see Fig. \[fig:C\_ell\_diff\]). All of the GW-brightest sources included in our catalogue are at distances less than 10 Mpc, where non-linear effects will be important (see Fig. \[fig:brightest-galaxies\]). So even on large angular scales, the $C_\ell$’s are sensitive to GW sources at small distances from us, and therefore to small scales in the galaxy power spectrum.
{width="\textwidth"}
{width="\textwidth"}
To confirm that the linearised approximation (augmented with HALOFIT) adopted by Cusin et al in Ref. [@Cusin:2018rsq] lead to smaller anisotropies, we compared in Ref.[@Jenkins:2018kxc] the $C_\ell$’s obtained using the Millenium catalogue and that of Ref. [@Cusin:2018rsq] for the same (maximum-likelihood) BBH distribution and fixing all other details of the astrophysical model to be the same. We refer the reader to Fig. 2 of our paper [@Jenkins:2018kxc]
In summary, we have pointed out that none of Cusin et al’s criticisms in Ref. [@Cusin:2018rsq] are at all relevant to our catalogue approach, which is the basis for all of our main results. We have defended the accuracy of this approach compared to the linearised approach used by Cusin et al in Ref. [@Cusin:2018rsq].
[^1]: From John Peacock’s website where the code is hosted, [www.roe.ac.uk/\~jap/haloes/](www.roe.ac.uk/~jap/haloes/) — “HALOFIT has received a lot of use, and has been incorporated into CMB packages such as CMBFAST and CAMB. Nevertheless, it is not perfect: it reflected accurately the state of the art of simulations as of 2003, but subsequent work has pushed measurements to smaller scales and higher degrees of nonlinearity. This has revealed that HALOFIT tends to underpredict the power on the smallest scales in standard LCDM universes (although HALOFIT was designed to work for a much wider range of power spectra).”
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'This paper focuses on data structures for multi-core reachability, which is a key component in model checking algorithms and other verification methods. A cornerstone of an efficient solution is the storage of visited states. In related work, static partitioning of the state space was combined with thread-local storage and resulted in reasonable speedups, but left open whether improvements are possible. In this paper, we present a scaling solution for shared state storage which is based on a lockless hash table implementation. The solution is specifically designed for the cache architecture of modern CPUs. Because model checking algorithms impose loose requirements on the hash table operations, their design can be streamlined substantially compared to related work on lockless hash tables. Still, an implementation of the hash table presented here has dozens of sensitive performance parameters (bucket size, cache line size, data layout, probing sequence, etc.). We analyzed their impact and compared the resulting speedups with related tools. Our implementation outperforms two state-of-the-art multi-core model checkers (SPIN and DiVinE) by a substantial margin, while placing fewer constraints on the load balancing and search algorithms.'
author:
- |
[Alfons Laarman, Jaco van de Pol, Michael Weber]{}\
[{a.w.laarman,vdpol,michaelw}@cs.utwente.nl]{}\
[Formal Methods and Tools, University of Twente, The Netherlands]{}
bibliography:
- '9\_LLDBS.bib'
title: 'Boosting Multi-Core Reachability Performance with Shared Hash Tables'
---
Acknowledgements
================
We thank the chair of Computational Materials Science at UTwente for making their cluster available for our experiments. In particular, we thank Anton Starikov for aiding our experiments and at the same time configuring the cluster to facilitate them. We thank Petr Ročkai and Jiř[í]{} Barnat for the help and support they provided on the DiVinE toolkits. We also thank the Linux Kernel developers for their help in tracing down the cause of the performance regression on newer kernels. This resulted in 10% improvements of the results.
APPENDIX A – SPEEDUPS
=====================
This appendix contains detailed figures about per-model speedups with the different model checkers.
![Speedup of BEEM models with LTSmin[]{data-label="dve"}](dve_speedups-nr.pdf){width="90.00000%"}
![Speedup of BEEM models with DiVinE 2.2[]{data-label="div2"}](div2_speedups-nr.pdf){width="90.00000%"}
![Speedup of BEEM models with SPIN[]{data-label="div2-spin"}](spin_speedups.pdf){width="90.00000%"}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this paper we give a complete classification of totally-reflective, primitive genera in dimension 3 and 4. Our method breaks up into two parts. The first part consists of classifying the square free, totally-reflective, primitive genera by calculating strong bounds on the prime factors of the determinant of positive definite quadratic forms (lattices) with this property. We achieve these bounds by combining the Minkowski-Siegel mass formula with the combinatorial classification of reflective lattices accomplished by Scharlau & Blaschke. In a second part, we use a lattice transformation that goes back to Watson, to generate all totally-reflective, primitive genera when starting with the square free case.'
author:
- Ivica Turkalj
bibliography:
- 'References.bib'
title: 'Totally-Reflective Genera of Integral Lattices'
---
Introduction
============
The investigation of totally-reflective genera came up with the problem of classifying the cofinite, arithmetic reflection groups on the hyperbolic space. These groups occur as Weyl groups of Lorentzian lattices with a fundamental polyhedron of finite volume. Vinberg showed in [@vin72] that a necessary condition for (certain) Lorentzian lattices to induce a cofinite, arithmetic reflection group, is that a naturally associated, positive definite genus is totally-reflective. In the subsequent work [@vin81], he proved that totally-reflective genera only appear in dimension $< 30$. It was unknown whether this bound is sharp since the largest known dimension of such genera was $20$. An example was found by Borcherds in [@bor87], §$8$, Example $5$. By a more detailed investigation of the existence of non reflective lattices in high dimensional genera, Esselmann proved in [@ess96] that $20$ is actually the largest dimension in which a totally-reflective genus can exists. Furthermore, it is well known that there are only finitely many totally-reflective, primitive genera in any fixed dimension (cf. [@sw92], Theorem 1.4). An explicit classification has been carried out only in dimenson 3 for the special case of square free genera, cf. [@wal93] (which we reproduce here). In this work, all totally-reflective, primitive genera in dimension $3$ and $4$ have been found. Also, the related problem of classifying the hyperbolic reflection groups is outlined. We give a brief overview of our methods.
Our goal of classifying all totally-reflective genera in dimension $3$ and $4$ is achieved as follows:
- Let $L$ be a strongly square free, totally-reflective lattice with $\dim L = 4$ (resp. $\dim = 3$). Hence $\det L$ is of the form $\det L = p_1^2 \cdots p_r^2 \cdot q_1 \cdots q_s$ (resp. $r=0$). Using the mass formula, we prove that $r \leq 9$ and $s \leq 8-r$ (resp. $s \leq 10)$ (section 3.1).
- By applying the combinatorial classification of Scharlau & Blaschke, we prove that there are bounds $\overline{p_i}$ and $\overline{q_j}$ (one for every prime factor) depending only on the number of prime factors, such that $p_i \leq \overline{p_i}$ and $q_j \leq \overline{q_j}$. Thus the number of local invariants that need to be taken into account is effectively bounded and the enumeration is computationally feasible (section 3.2)
- After finishing the strongly square free classification, we obtain all square free, totally-reflective, primitive genera by partial dualization (section 4.1).
- The last step consists in dropping the assumption “square free” by determining the pre-images of square free genera under the Watson transformation (section 4.2).
Background
==========
Integral lattices
-----------------
Let $R$ be a prinicipal ideal domain and $K:=\operatorname{Quot}(R)$ its quotient field. A *lattice* over $R$ is a pair $(L,b)$, where $L$ is a free $R$-module of finite rank and $b:L \times L \longrightarrow K$ a symmetric bilinear form. As usual, $V:=L \otimes_R K$ means the enveloping $K$-space of $L$. By $O(L)$ we denote the isometry group of $(L,b)$. The *determinant* of $L$ is the determinant of any Gram matrix of $(L,b)$, which is well defined modulo squares of units in $R$.
We say that $L$ is *integral*, if $b(L,L) \subseteq R$. An $R$-lattice is called *even* if $b(x,x) \in 2R$ for all $x \in L$, and *odd* otherwise. By ${}^{\alpha}\!{L}$ we mean the lattice $(L,\alpha b)$ obtained by scaling the bilinear form, where $\alpha \in K$. An integral lattice is said to be *primitive* if it is not the scaled version of another intergral lattice. We denote by $L^{\#}$ the *dual lattice* of $L$ which is defined as $$\begin{aligned}
L^{\#} := \left\{ v \in V \| \forall x \in L : b(x,v) \in R \right\}.\end{aligned}$$ Clearly, $L$ is integral iff $L \subseteq L^{\#}$. For an integral lattice, the group $L^{\#}/L$ has order $\det L$ and is called the *discriminant group* of $L$. A lattice is *unimodular* if $L = L^{\#}$. More generally, for $\alpha \in K$, we say $L$ is *$\alpha$-modular* if $L = \alpha L^{\#}$. It is easy to see that any $\alpha$-modular lattice $K$ can be written as $K={}^{\alpha}\!{L}$, where $L$ is unimodular.
Let $\PP$ be the set of all rational primes. Two $\ZZ$-lattices $L_1, L_2$ are in the same *genus* if they become isometric over all completions $\ZZ_p$: $$\begin{aligned}
L_1 \otimes_{\ZZ} \ZZ_p \cong L_2 \otimes_{\ZZ} \ZZ_p, \ \text{for all $p \in \PP \cup \{ \infty \}$}.\end{aligned}$$ It is well known that any genus consists of finitely many isometry classes. We write $\mathcal{G}(L)$ for the set of all isometry classes in the genus of $L$ and define $h(L):=\# \mathcal{G}(L)$ as the *class number* of $L$.
Recall that every $\ZZ$-lattice $L$ locally posseses a *Jordan decomposition*, $$\begin{aligned}
L \otimes_{\ZZ} \ZZ_p = {}^{p^{-r}}\!{L_{-r}} \perp \dots \perp {}^{p^{-1}}\!{L_{-1}} \perp L_0 \perp {}^{p}\!{L_1} \perp \dots \perp {}^{p^r}\!{L_r},\end{aligned}$$ where all $L_i$ are unimodular (possibly zero-dimensional). For $p \neq 2$ the Jordan decomposition is unique up to isometry. Unfortunately this is not true for $p = 2$. At least the following data remains invariant: $\dim L_i$, $2^i$ and the property even/odd for every $L_i$. We refer to $L$ as *square free* if the Jordan decomposition of $L$ is of the form $L_0 \perp {}^{p}\!{L_1}$ at every prime $p$, and as *strongly square free* if additionally $\dim L_0 \geq \dim L_1$ holds.
We remind the reader of the *genus symbol* as introduced in [@cs99], Chapter $15$. This symbol is a list of local symbols for each prime $p$ dividing $2\det L$. Assuming a Jordan decomposition as above, the local symbol at the prime $p \neq 2$ is the formal product $$\begin{aligned}
\text{$\prod_{i=-r}^{s} (p^{i})^{\varepsilon_i,n_i}$, where $\varepsilon_i := \left(\frac{\det L_i}{p}\right)$ and $n_i:=\dim L_i$.}\end{aligned}$$ Due to the lack of uniqueness, the $2$-adic symbol is more complicated and two other invariants have to be taken into account (oddity and parity). We refer to [@cs99] for a more detailed investigation.
The mass formula
----------------
From now on all $\ZZ$-lattices $L$ are assumed to be positive definite (except the Lorentzian lattices in section $5$). Hence $O(L)$ is a finite group and the following definition makes sense.
Let $L$ be a $\ZZ$-lattice. The *mass* of $L$ is defined as $$\begin{aligned}
\operatorname{m}(L):= \sum_{M \in \mathcal{G}(L)}\frac{1}{|O(M)|}.
\end{aligned}$$
An indispensable tool for our investigation is the Minkowski-Siegel mass formula which relates the mass of a lattice to local quantities, which can be derived from the genus symbol. Originally the mass formula is stated in terms of $p$-adic densities, cf. [@sie35] for further details. However, for computational reasons, we found the approach of Conway and Sloane more suitable, cf. [@cs88]. Below we outline the fundamental aspects. Let $n:=\dim L, s:=\lceil \frac{n}{2} \rceil$ and $D:=(-1)^s\det L$. A helpful notion is the so-called *standard mass*, $$\begin{aligned}
\operatorname{std}(n,D):= 2\pi^{-n(n+1)/4} \cdot \prod_{j=1}^{n}\Gamma(\tfrac{j}{2}) \cdot \zeta(2)\zeta(4) \cdots \zeta(2s-2)\zeta_D(s),\end{aligned}$$ where $\Gamma$ denotes the Gamma function, $\zeta$ the Riemann zeta function and $\zeta_D$ refers to the $L$-function $$\begin{aligned}
\zeta_D(s)=
\begin{cases}
\prod_{p \in \PP}\left(1-\left(\tfrac{D}{p}\right)\frac{1}{p^s}\right)^{-1}, & n\ \text{even}, \\
1, & n\ \text{odd}.
\end{cases}\end{aligned}$$ The actual mass of $L$ is gained from the standard mass by multiplying with certain correction factors, one for every prime $p$ dividing $2\det L$. Unlike the standard mass, these correction factors depend on the local structure of $L$.
Let $L$ be a $\ZZ$-lattice with Jordan decompositions as in $(1)$. Then $$\begin{aligned}
\operatorname{m}(L) = \operatorname{std}(n,D) \cdot \prod_{p \mid 2\det L} \left( \operatorname{m_{p}}(L) \cdot 2 \prod_{j=2}^{s}\left(1-p^{2-2j}\right) \right),
\end{aligned}$$ where for $p \neq 2$ $$\begin{aligned}
\operatorname{m_{p}}(L) =\prod_{\substack{i \in \ZZ,\\ \dim L_i \neq 0}}\left(\tfrac{1}{2}\left(1+\varepsilon p^{-s_i}\right)^{-1} \cdot \prod_{i=2}^{s_i}\left(1-p^{2-2i}\right)^{-1}\right)
\cdot \prod_{\substack{k,l \in \ZZ,\\ k<l}} p^{\tfrac{1}{2}(l-k)n_ln_k}.
\end{aligned}$$
In the above proposition, $\varepsilon = 0$ if $\dim L_i$ is odd, and $\varepsilon \in \left\{1,-1 \right\}$ if $\dim L_i$ is even. The exact value of $\varepsilon$ depends on the species of the orthogonal group $O_{n_i}(p)$ over $\FF_p$, which can be read off the genus symbol. Again, the case $p=2$ is more complicated, for which we refer to [@cs88].
Reflective lattices and totally-reflective genera
-------------------------------------------------
For a nonzero vector $v \in L$, the reflection $$\begin{aligned}
s_v: x \longmapsto x - \frac{2b(x,v)}{b(v,v)}v\end{aligned}$$ is an isometry of $V$. We call $v$ a *root* of $L$ if $v$ is primitive (that is $v/m \notin L$ for all integers $m > 0$) and $s_v(L) = L$. The set $R(L)$ of all roots of $L$ is a root system in the usual sense of Lie theory. The subgroup $W(L) \leqslant O(L)$ generated by all reflections $s_v$, with $v \in R(L)$, is called *Weyl group* of $L$. Besides the combinatorial structure induced by the Dynkin diagram, the root system of a lattice inherits the quadratic form, thus $R(L)$ decomposes into scaled irreducible components ${\hspace{0.5mm}^{\alpha}\hspace*{-0.5mm}\mathsf A}_n, {\hspace{0.5mm}^{\alpha}\hspace*{-0.2mm}\mathsf B}_n, {\hspace{0.5mm}^{\alpha}\hspace*{-0.2mm}\mathsf C}_n, {\hspace{0.5mm}^{\alpha}\hspace*{-0.2mm}\mathsf D}_n,
{\hspace{0.5mm}^{\alpha}\hspace*{-0.2mm}\mathsf E}_6, {\hspace{0.5mm}^{\alpha}\hspace*{-0.2mm}\mathsf E}_7, {\hspace{0.5mm}^{\alpha}\hspace*{-0.2mm}\mathsf E}_8, {\hspace{0.5mm}^{\alpha}\hspace*{-0.2mm}\mathsf F}_4, {\hspace{0.5mm}^{\alpha}\hspace*{-0.2mm}\mathsf G}_2$. We mention [@bou08] as a general reference to the theory of root systems.
A positive definite lattice is called reflective if, roughly spoken, it is “almost” a root lattice (it is a root lattice up to finte index). More precisely:
The positive definite, integral lattice $L$ is called *reflective* if its root system $R(L)$ generates a sublattice of the same rank.
Certainly, $L$ is reflective iff $W(L)$ has no nonzero fixed vectors (while acting on $V$).
Crucial to our investigations is the work of Scharlau & Blaschke. They classified all indecomposable, reflective lattices in low dimensions by pairs $({\mathsf R}, \CL)$, where ${\mathsf R}$ is a scaled root system and $\CL$ the so-called glue code (a subgroup of the discriminant group of $\left\langle {\mathsf R}\right\rangle$) (cf. [@bs96], $4.4$, $4.5$, $4.7$). Given a pair $({\mathsf R},\CL)$, the associated lattice $L$ is constructed by $L = \left\langle {\mathsf R}\right\rangle + \left\langle x \| \overline{x} \in \CL \right\rangle$. We will refer to this result in Lemma $3.4$, Lemma $3.5$ and Lemma $3.9$.
Let $L$ be an integral lattice and $\CG$ its genus.
- We call $\CG$ *totally-reflective* if each lattice in $\CG$ is reflective.
- The integral lattice $L$ is called *totally-reflective* if its genus $\CG$ is totally-reflective.
One can deduce from the work of Biermann [@bie81], that there are only finitely many totally-reflective genera in any fixed dimension (cf. [@sw92], Theorem 1.4). Furthermore, Esselmann proved in [@ess96] that $20$ is the largest dimension of totally-reflective genera, thus a classification is possible (at least) in priniciple. With the present paper we contribute to this problem by classifying the dimensions $3$ and $4$.
Bounds for strongly square free, totally-reflective genera
==========================================================
In this section we prove the results concerning step 1 and 2 of the general strategy. The basic idea is to compare the whole mass of a lattice $L$ with the part coming from the reflective lattices within $\CG(L)$. Since the latter quantity is crucial, we make the following
Let $L$ be an integral lattice. We refer to $$\begin{aligned}
\operatorname{m_{ref}}(L) := \sum_{\substack{M \in \CG(L),\\\text{$M$ is reflective}}} \frac{1}{|O(M)|}
\end{aligned}$$ as the *reflective part* of the mass.
An important (though trivial) observation is that $\operatorname{m_{ref}}(L) \leq \operatorname{m}(L)$ and $L$ is totally-reflective iff $\operatorname{m_{ref}}(L) = \operatorname{m}(L)$. We will obtain our bounds by showing that the reflective part of the mass grows more slowly than the whole mass. We begin with
Let $D \in \NN$. Then
- $\zeta_D(2) \geq \frac{\zeta(4)}{\zeta(2)} = \frac{\pi^2}{15}$.
- $\zeta_{-D}(1) \leq 1+\frac{1}{2} \ln(D)$.
Part (a) follows from an elementary calculation which can be found in [@kl13], $5.1$ and (b) is proved in [@wat79], $5.10$.
With part (a) of the previous lemma and the mass formula as stated in section 2.2, one can control the growth of $\operatorname{m}$ in the following sense.
Let $L$ be a strongly square free lattice with determinant $d$.
- For $\dim L = 3$ set $$\begin{aligned}
\operatorname{M}(L):=\frac{1}{6} \cdot \frac{1}{8} \cdot \prod_{\substack{p \mid d,\\p \neq 2}} \frac{p-1}{2}.
\end{aligned}$$
- For $\dim L = 4$ set $$\begin{aligned}
\operatorname{M}(L):=\frac{1}{90}\cdot \frac{1}{24} \cdot \prod_{\substack{p \mid d,\\v_p(d) = 2 \\ p \neq 2}} \frac{1}{2}p^2\cdot\frac{p-1}{p+1}
\cdot \prod_{\substack{p \mid d,\\v_p(d)=1 \\ p \neq 2}}\frac{1}{2}p^{\tfrac{3}{2}}.
\end{aligned}$$
Then, in both cases, we have $\operatorname{m}(L) \geq \operatorname{M}(L)$.
This follows directly from the mass formula and Lemma $3.2$.
It is somewhat more difficult to control the growth of $\operatorname{m_{ref}}$. For this, the following observation is helpful (recall Remark $2.4$).
Let $L$ be an indecomposable reflective lattice with $\dim L \in \{2,3,4\}$. Then $O(L)$ only depends on the combinatorial class of $R(L)$. In particular, $O(L)$ does not depend on the glue-code nor the scaling. Referring to [@bs96], $4.4$, $4.5$, $4.7$, we have
- in dimension $2:$
$
\begin{array}{c|c|c|c|c}
& (a) & (b) & (c) & (d) \\
\hline
|O(L)| & 4 & 12 & 4 & 8
\end{array}
$
- in dimension $3:$
$
\begin{array}{c|c|c|c|c|c}
& (a) & (b) & (c) & (d) & (e)\\
\hline
|O(L)| & 8 & 8 & 16 & 48 & 48
\end{array}
$
- in dimension $4:$
$
\begin{array}{c|c|c|c|c|c|c}
& (a),(b),(c) & (d),(e) & (f),(g),(i) & (h) & (j),(k) & (l)\\
\hline
|O(L)| & 16 & 32 & 96 & 72 & 240 & 1152
\end{array}
$
Let $R(L)$ be decomposed as $R(L) = {\hspace{0.5mm}^{\alpha_1}\hspace*{-0.2mm}\mathsf R}_1 \cdots {\hspace{0.5mm}^{\alpha_k}\hspace*{-0.2mm}\mathsf R}_k$, with ${\mathsf R}_i$ irreducible and let $O(R(L))$ be the stabilizer of $R(L)$ in $O(V)$. It is well known that $O(R(L)) \cong W(R(L)) \rtimes A(L)$, where $A(L)$ can be identified with a subgroup of the outer automorphism group of $R(L)$. It follows from [@bs96], $4.4$, $4.5$, $4.7$, and the relation $W(R(L)) \subseteq O(L) \subseteq O(R(L))$ that $O(L) = W(R(L))$ holds in dimension $2$ and $3$ since $A(L)$ is trivial, except for ${\mathsf A}_2, {\mathsf A}_3$ and ${\mathsf D}_3$. Furthermore, $W(R(L))$ does not depend on the scaling since $W({\hspace{0.5mm}^{\alpha_1}\hspace*{-0.2mm}\mathsf R}_1 \dots {\hspace{0.5mm}^{\alpha_k}\hspace*{-0.2mm}\mathsf R}_k) \cong W({\hspace{0.5mm}^{\alpha_1}\hspace*{-0.2mm}\mathsf R}_1) \times \cdots \times W({\hspace{0.5mm}^{\alpha_k}\hspace*{-0.2mm}\mathsf R}_k) \cong W({\mathsf R}_1) \times \cdots \times W({\mathsf R}_k).$ In dimension 4, one can use the same arguments, except in the cases $(h),(j)$ and $(k)$, where $A(L) \neq 1$. But there we have $$\begin{aligned}
A(L) =
\begin{cases}
\ZZ/2\ZZ, & \text{in $(j)$ and $(k)$}, \\
\ZZ/2\ZZ \times \ZZ/2\ZZ, & \text{in $(h),$} \\
\end{cases}
\end{aligned}$$ which obviously do not depend on scaling nor the glue-code. Weyl groups of unscaled, irreducible root systems, particularly their orders, are well known and can be found, for instance, in [@bou08].
Bounds on the number of prime factors
-------------------------------------
Regarding the statement on the number of prime factors outlined in step 1, the following way of controlling $\operatorname{m_{ref}}$ seems appropriate (we will introduce an alternative way in the next subsection). We recall that $\omega(d)$ (resp. $\Omega(d)$) refers to the number of (not necessarily) different prime factors of $d$.
Let $L$ be a strongly square free lattice with determinant $d$.
- For $\dim L = 3$ we set $$\begin{aligned}
\operatorname{M_{ref}}(L):=\sum_{\substack{x \mid d}}2^1 \cdot \frac{1}{2} \cdot \frac{2}{\pi} \cdot \prod_{p \mid x}\frac{1}{2}\sqrt{p} \cdot \left(1+\frac{1}{2}\ln(x)\right).
\end{aligned}$$
- For $\dim L = 4$ we set $$\begin{aligned}
\operatorname{M_{ref}}(L) &:=3\frac{4^{\Omega(d)}}{16} + 2\frac{3^{\Omega(d)}}{32} + \frac{2^{\Omega(d)}}{72} + 3\frac{2^{\Omega(d)}}{96} + \frac{53}{5760} \\
&+ \sum_{\substack{x \mid d}} 2^{\omega(x)+1} \cdot \frac{1}{4} \cdot \frac{2}{\pi}
\cdot \left(1+\frac{1}{2}\ln(x)\right)
\cdot \prod_{\substack{p \mid x,\\v_p(d)=2}}\frac{2p}{2p-1} \cdot \prod_{\substack{p \mid x,\\v_p(d)=1}}\frac{1}{2}\sqrt{p},
\end{aligned}$$
Then, in both cases, we have $\operatorname{m_{ref}}(L) \leq \operatorname{M_{ref}}(L)$.
$(b)$ We take a closer look on how the quantity $\operatorname{m_{ref}}$ is composed, more precisely we write $$\begin{aligned}
\operatorname{m_{ref}}(L) = \operatorname{m}^{(4)}_{\mathrm{ref}}(L) + \operatorname{m}^{(3)}_{\mathrm{ref}}(L) + \operatorname{m}^{(2)}_{\mathrm{ref}}(L),
\end{aligned}$$ where $\operatorname{m}^{(4)}_{\mathrm{ref}}(L)$ refers to the part of the mass coming from the indecomposable, reflective lattices within the genus, $\operatorname{m}^{(3)}_{\mathrm{ref}}(L)$ refers to the part which comes from the reflective lattices with a $3$-dimensional, indecomposable component and $\operatorname{m}^{(2)}_{\mathrm{ref}}(L)$ means the contribution of the reflective lattices which have a $2$-dimensional, indecomposable compontent or which are diagonalisable. An estimate of $\operatorname{m}^{(4)}_{\mathrm{ref}}(L)$ is easily obtained by combining the classification theorem from [@bs96] and Lemma $3.4$. Considering that a bound for the number of possible isometry classes for a given pair $({\mathsf R},\CL)$ of determinant $d$ is $a^{\Omega(d)}$, where $a$ is the number of occurring scaling factors, we get $$\begin{aligned}
\operatorname{m}^{(4)}_{\mathrm{ref}}(L) \leq 3\frac{4^{\Omega(d)}}{16} + 2\frac{3^{\Omega(d)}}{32} + \frac{2^{\Omega(d)}}{72} + 3\frac{2^{\Omega(d)}}{96} + \frac{53}{5760}.
\end{aligned}$$ As Berger showed in [@ber93], a $4$-dimensional, strongly square free, reflective lattice can not have a $3$-dimensional, indecomposable component, so $\operatorname{m}^{(3)}_{\mathrm{ref}}(L) = 0$. For the estimate on $\operatorname{m}^{(2)}_{\mathrm{ref}}(L)$ we use the mass formula: $$\begin{aligned}
\operatorname{m}^{(2)}_{\mathrm{ref}}(L) &= \sum_{\substack{\text{$M \in \CG(L)$ refl,} \\M = M_1 \perp M_2,\\ \dim M_1=2}} \frac{1}{|O(M)|}
\leq \frac{1}{4} \sum_{\substack{\text{$M \in \CG(L)$ refl,} \\M = M_1 \perp M_2,\\ \dim M_1=2}} \frac{1}{|O(M_1)|} \\
&\leq \sum_{\substack{x \mid d}} 2^{\omega(x)+1} \cdot \frac{1}{4} \cdot \frac{2}{\pi}
\cdot \left(1+\frac{1}{2}\ln(x)\right)
\cdot \prod_{\substack{p \mid d,\\v_p(d)=2}}\frac{2p}{2p-1} \cdot \prod_{\substack{p \mid d,\\v_p(d)=1}}\frac{1}{2}\sqrt{p},
\end{aligned}$$ where $2^{\omega(x)+1}$ is an estimate for the number of $2$-dimensional, square free genera of determinant $x$. The estimate on $\operatorname{m}(M_1)$ follows from Proposition $2.2$ and Lemma $3.2. (b)$.
$(a)$ Here we consider $$\begin{aligned}
\operatorname{m_{ref}}(L) = \operatorname{m}^{(3)}_{\mathrm{ref}}(L) + \operatorname{m}^{(2)}_{\mathrm{ref}}(L).
\end{aligned}$$ Again, $\operatorname{m}^{(3)}_{\mathrm{ref}}(L) = 0$, because $4$ divides the determinant of a $3$-dimensional, indecomposable, reflective lattice which is never the case for strongly square free lattices. For $\operatorname{m}^{(2)}_{\mathrm{ref}}(L)$ we get: $$\begin{aligned}
\operatorname{m}^{(2)}_{\mathrm{ref}}(L) &= \sum_{\substack{\text{$M \in \CG(L)$ refl,} \\M = M_1 \perp M_2,\\ \dim M_1=2}} \frac{1}{|O(M)|}
\leq \frac{1}{2} \sum_{\substack{\text{$M \in \CG(L)$ refl,} \\M = M_1 \perp M_2,\\ \dim M_1=2}} \frac{1}{|O(M_1)|} \\
&\leq \sum_{\substack{x \mid d}} 2^1 \cdot \frac{1}{2} \cdot \frac{2}{\pi} \cdot \prod_{p \mid x}\frac{1}{2}\sqrt{p} \cdot \left(1+\frac{1}{2}\ln(x)\right).
\end{aligned}$$ Notice that, unlike in the $4$-dimensional case, an estimate for the number of $2$-dimensional, square free genera of determinant $x$ is $2$. In the situation $\CG(M_1 \perp N_1) = \CG(M_2 \perp N_2)$ with $\dim M_i =2, \dim N_i=1$ and $\det N_i = x$ it follows from Witt’s cancellation theorem for nondyadic local rings that $M_1 \otimes_{\ZZ} \ZZ_p \cong M_2 \otimes_{\ZZ} \ZZ_p$ for $p \neq 2$, whereas $M_1 \otimes_{\ZZ} \ZZ_2$ and $M_2 \otimes_{\ZZ} \ZZ_2$ can be distinguished only through the parity, cf. [@oma00], $92:3, 93:16$.
By combining Lemma $3.3$ and Lemma $3.5$, we see that a strongly square free, totally-reflective lattice satisfies the condition $\operatorname{M_{ref}}(L) / \operatorname{M}(L) \geq 1$. Actually, the estimates on $\operatorname{m_{ref}}(L)$ and $\operatorname{m}(L)$ depend only on the determinant of $L$, so we may write $\operatorname{M}(L) = \operatorname{M}(\det L)$ and $\operatorname{M_{ref}}(L) = \operatorname{M_{ref}}(\det L)$.
To show that the number of prime factors can not be arbitrarily large, it is important to investigate how the ratio of $\operatorname{M_{ref}}$ and $\operatorname{M}$ behaves when prime factors are appended to the determinant. Notice that both $\operatorname{M_{ref}}$ and $\operatorname{M}$ tend to $\infty$ when the number of prime factors increases.
Let $d \in \mathbb{N}$ and $q \in \PP$.
- In dimension $3$ we have $$\begin{aligned}
\frac{\operatorname{M_{ref}}(d)}{\operatorname{M}(d)} \geq \frac{\operatorname{M_{ref}}(dq)}{\operatorname{M}(dq)}
\end{aligned}$$ if $q \geq 17$ and $\Omega(d) \geq 2$.
- In dimension $4$ we have $$\begin{aligned}
\frac{\operatorname{M_{ref}}(d)}{\operatorname{M}(d)} \geq \frac{\operatorname{M_{ref}}(dq^2)}{\operatorname{M}(dq^2)}
\end{aligned}$$ if $q \geq 7$ and $\Omega(d) \geq 2$, as well as $$\begin{aligned}
\frac{\operatorname{M_{ref}}(d)}{\operatorname{M}(d)} \geq \frac{\operatorname{M_{ref}}(dq)}{\operatorname{M}(dq)}
\end{aligned}$$ if $q \geq 5$ and $\Omega(d) \geq 2$.
We proof part (b) in detail so that the general idea will be clear. Part (a) is proven analogously. It is helpful to consider the quantity $\operatorname{M_{ref}}$ piecewise. We write $\operatorname{M_{ref}}(d) = \operatorname{M}^{(4)}_{\mathrm{ref}}(d) + \operatorname{M}^{(2)}_{\mathrm{ref}}(d)$ with $$\begin{aligned}
\operatorname{M}^{(4)}_{\mathrm{ref}}(d):=3\frac{4^{\Omega(d)}}{16} + 2\frac{3^{\Omega(d)}}{32} + \frac{2^{\Omega(d)}}{72} + 3\frac{2^{\Omega(d)}}{96} + \frac{53}{5760}
\end{aligned}$$ and $$\begin{aligned}
\operatorname{M}^{(2)}_{\mathrm{ref}}(d):=\sum_{\substack{x \mid d}} 2^{\omega(x)+1} \cdot \frac{1}{4} \cdot \frac{2}{\pi}
\cdot \left(1+\frac{1}{2}\ln(x)\right)
\cdot \prod_{\substack{p \mid d,\\v_p(d)=2}}\frac{2p}{2p-1} \cdot \prod_{\substack{p \mid d,\\v_p(d)=1}}\frac{1}{2}\sqrt{p}.
\end{aligned}$$ The case $\operatorname{M}^{(4)}_{\mathrm{ref}}$ is easily done. From the mass formula we get the equivalence $$\begin{aligned}
\frac{\operatorname{M}^{(4)}_{\mathrm{ref}}(d)}{\operatorname{M}(d)} \geq \frac{\operatorname{M}^{(4)}_{\mathrm{ref}}(dq^2)}{\operatorname{M}(dq^2)}
\Longleftrightarrow \operatorname{M}^{(4)}_{\mathrm{ref}}(d) \cdot \frac{q^2(q-1)}{2 (q+1)} \geq \operatorname{M}^{(4)}_{\mathrm{ref}}(dq^2)
\end{aligned}$$ and because of $$\begin{aligned}
\operatorname{M}^{(4)}_{\mathrm{ref}}(dq^2) \geq 4^2 \cdot \operatorname{M}^{(4)}_{\mathrm{ref}}(d)
\end{aligned}$$ the inequality on the right hand side is satisfied when $\tfrac{q^2(q-1)}{2 (q+1)} \geq 4^2$, that is when $q \geq 7$.
The case $\operatorname{M}^{(2)}_{\mathrm{ref}}$ is more difficult. Let $p_1$ be an arbitrary prime dividing $d$. Define $D(d):=\{ x \in \NN \| \text{$d$ is divisible by $x$}\}$. For the first part of (b) the starting point is the decomposition $$\begin{aligned}
D(dq^2) & = D(d)\setminus \{p_1\} \\
& \cupdot q \cdot \left(D(d)\setminus \{p_1,1\} \right) \\
& \cupdot q^2 \cdot \left(D(d)\setminus \{p_1,1\} \right) \\
& \cupdot \left\{q, q^2, p_1, q p_1, q^2 p_1 \right\}.
\end{aligned}$$ The idea is to compare the summands of $\operatorname{M}^{(2)}_{\mathrm{ref}}(dq^2)$ corresponding to the right hand side of $(2), (3)$ and $(4)$ (resp. $(5)$) with the summands of $\operatorname{M}^{(2)}_{\mathrm{ref}}(d)$ corresponding to $D(d)\setminus \{p_1\}$ (resp. $p_1$). For all $x \in D(d) \setminus \{p_1\}$ the mass formula implies $$\begin{aligned}
& \left( 2^{\omega(x)+1} \cdot \frac{1}{4} \cdot \frac{2}{\pi}
\cdot \left(1+\frac{1}{2}\ln(x)\right)
\cdot \prod_{\substack{p \mid d,\\v_p(d)=2}}\frac{2p}{2p-1} \cdot \prod_{\substack{p \mid d,\\v_p(d)=1}}\frac{1}{2}\sqrt{p}\right)
\cdot \frac{q^2(q-1)}{2 (q+1)} \\
& \geq 2^{\omega(x)+1} \cdot \frac{1}{4} \cdot \frac{2}{\pi}
\cdot \left(1+\frac{1}{2}\ln(x)\right)
\cdot \prod_{\substack{p \mid d,\\v_p(d)=2}}\frac{2p}{2p-1} \cdot \prod_{\substack{p \mid d,\\v_p(d)=1}}\frac{1}{2}\sqrt{p} \\
& + 2^{\omega(qx)+1} \cdot \frac{1}{4} \cdot \frac{2}{\pi}
\cdot \left(1+\frac{1}{2}\ln(qx)\right)
\cdot \prod_{\substack{p \mid d,\\v_p(d)=2}}\frac{2p}{2p-1} \cdot \prod_{\substack{p \mid d,\\v_p(d)=1}}\frac{1}{2}\sqrt{p} \cdot \frac{1}{2}\sqrt{q}\\
& + 2^{\omega(q^2x)+1} \cdot \frac{1}{4} \cdot \frac{2}{\pi}
\cdot \left(1+\frac{1}{2}\ln(q^2x)\right)
\cdot \prod_{\substack{p \mid d,\\v_p(d)=2}}\frac{2p}{2p-1} \cdot \frac{2q}{2q-1} \cdot \prod_{\substack{p \mid d,\\v_p(d)=1}}\frac{1}{2}\sqrt{p},
\end{aligned}$$ which is equivalent to $$\begin{aligned}
& \left(\frac{q^2(q-1)}{2 (q+1)}-1\right) \cdot \left(1+\frac{1}{2}\ln(x)\right) \\
& \geq \left( \sqrt{q}+\frac{\sqrt{q}}{2} \ln(q) + \frac{4q}{2q-1}+\frac{2q}{2q-1} \ln(q^2) \right) + \left( \frac{\sqrt{q}}{2} + \frac{2q}{2q-1} \right) \ln(x).
\end{aligned}$$ Since $x \geq 1$, and thus $\ln(x) = 0$ or $\ln(x) \geq 1$, the latter inequality is true for $q \ge 7$. For $x = p_1$ we obtain $$\begin{aligned}
& \left( 2^{\omega(p_1)+1} \cdot \frac{1}{4} \cdot \frac{2}{\pi}
\cdot \left(1+\frac{1}{2}\ln(p_1)\right)
\cdot \frac{1}{2}\sqrt{p_1} \right) \cdot \frac{q^2(q-1)}{2 (q+1)} \\
& + 2^{\omega(p_1)+1} \cdot \frac{1}{4} \cdot \frac{2}{\pi}
\cdot \left(1+\frac{1}{2}\ln(p_1)\right)
\cdot \frac{1}{2}\sqrt{p_1} \\
& + 2^{\omega(q)+1} \cdot \frac{1}{4} \cdot \frac{2}{\pi}
\cdot \left(1+\frac{1}{2}\ln(q)\right)
\cdot \frac{1}{2}\sqrt{q} \\
& + 2^{\omega(q^2)+1} \cdot \frac{1}{4} \cdot \frac{2}{\pi}
\cdot \left(1+\frac{1}{2}\ln(q^2)\right)
\cdot \frac{2q}{2q-1} \\
& + 2^{\omega(qp_1)+1} \cdot \frac{1}{4} \cdot \frac{2}{\pi}
\cdot \left(1+\frac{1}{2}\ln(qp_1)\right)
\cdot \frac{1}{2}\sqrt{q} \cdot \frac{1}{2} \sqrt{p_1} \\
& + 2^{\omega(q^2p_1)+1} \cdot \frac{1}{4} \cdot \frac{2}{\pi}
\cdot \left(1+\frac{1}{2}\ln(q^2p_1)\right)
\cdot \frac{2q}{2q-1} \cdot \frac{1}{2}\sqrt{p_1}.
\end{aligned}$$ It is easy to check that this is true for all $p_1 \geq 2$ and $q \geq 7$. The second part of (b) works similarly, but uses the decompostion $$\begin{aligned}
D(dq) & = D(d) \setminus \{p_1\} \\
& \cupdot q \cdot \left(D(d)\setminus \{p_1\}\right) \\
& \cupdot \left\{q, p_1, q p_1 \right\}.
\end{aligned}$$
The second lemma of this section clarifies the behavior of $\operatorname{M_{ref}}/ \operatorname{M}$ if the number of prime factors is fixed while the prime numbers increase. Again, a priori that is not clear since $\operatorname{M_{ref}}$ and $\operatorname{M}$ tend to $\infty$ when the primes increase.
In both dimensions $\frac{\operatorname{M_{ref}}(d)}{\operatorname{M}(d)}$ is monotonically decreasing in each prime factor of $d$.
Let $q$ be a prime with $v_q(d) = 2$ (the easier case $v_p(d) = 1$ is proven analogously). Like in the previous proof we consider $\operatorname{M}^{(4)}_{\mathrm{ref}}(d)$ and $\operatorname{M}^{(2)}_{\mathrm{ref}}(d)$ separated. Since $\operatorname{M}^{(4)}_{\mathrm{ref}}(d)$ only depends on the number of prime factors (and not on the primes itself), it is clear that $\frac{\operatorname{M}^{(4)}_{\mathrm{ref}}(d)}{\operatorname{M}(d)}$ decreases monotonically. In order to deal with $\operatorname{M}^{(2)}_{\mathrm{ref}}(d)$, we have to isolate the effect of $q$. For this the following decomposition is helpful: $$\begin{aligned}
D(d) = D(\tfrac{d}{q^2}) \cupdot q D(\tfrac{d}{q^2}) \cupdot q^2 D(\tfrac{d}{q^2}).
\end{aligned}$$ Thus we can write $$\begin{aligned}
& \operatorname{M}^{(2)}_{\mathrm{ref}}(d) = \sum_{\substack{x \mid d}} 2^{\omega(x)+1} \cdot \frac{1}{4} \cdot \frac{2}{\pi}
\cdot \left(1+\frac{1}{2}\ln(x)\right)
\cdot \prod_{\substack{p \mid d,\\v_p(d)=2}}\frac{2p}{2p-1} \cdot \prod_{\substack{p \mid d,\\v_p(d)=1}}\frac{1}{2}\sqrt{p} \\
& = \underbrace{\sum_{\substack{x \mid d/q}} 2^{\omega(x)+1} \cdot \frac{1}{4} \cdot \frac{2}{\pi}
\cdot \left(1+\frac{1}{2}\ln(x)\right)
\cdot \prod_{\substack{p \mid d/q,\\v_p(d/q)=2}}\frac{2p}{2p-1} \cdot \prod_{\substack{p \mid d/q,\\v_p(d/q)=1}}\frac{1}{2}\sqrt{p}}_{\text{$:=c_1$, does not depend on $q$}} \\
& + \sum_{x \mid d/q} \underbrace{2^{\omega(x)+1} \cdot \frac{1}{4} \cdot \frac{2}{\pi}
\cdot \prod_{\substack{p \mid d/q,\\v_p(d/q)=2}}\frac{2p}{2p-1} \cdot \prod_{\substack{p \mid d/q,\\v_p(d/q)=1}}\frac{1}{2}\sqrt{p} \cdot \frac{1}{2}}_{\text{$:=c_2(x)$, does not depend on $q$}}
\cdot \sqrt{q} \cdot \left(1+\frac{1}{2}\ln(qx)\right)\\
& + \sum_{x \mid d/q} \underbrace{2^{\omega(x)+1} \cdot \frac{1}{4} \cdot \frac{2}{\pi}
\cdot \prod_{\substack{p \mid d/q,\\v_p(d/q)=2}}\frac{2p}{2p-1} \cdot \prod_{\substack{p \mid d/q,\\v_p(d/q)=1}}\frac{1}{2}\sqrt{p}}_{\text{$:=c_3(x)$, does not depend on $q$}}
\cdot \frac{2q}{2q-1} \cdot \left(1+\frac{1}{2}\ln(q^2x)\right).
\end{aligned}$$ Furthermore, by isolating the effect of $q$ in $\operatorname{M}(d)$, $$\begin{aligned}
\operatorname{M}(d) = \underbrace{\frac{1}{90}\cdot \frac{1}{24} \cdot \prod_{\substack{p \mid d/q,\\v_p(d) = 2}} \frac{1}{2}p^2\cdot\frac{p-1}{p+1}
\cdot \prod_{\substack{p \mid d,\\v_p(d)=1}}\frac{1}{2}p^{\tfrac{3}{2}} \cdot \frac{1}{2}}_{\text{$:=c$, does not depend on $q$}}
\cdot q^2\cdot\frac{q-1}{q+1}
\end{aligned}$$ we can treat $$\begin{aligned}
\frac{\operatorname{M}^{(2)}_{\mathrm{ref}}(d)}{\operatorname{M}(d)} = \frac{c_1 + \sum_{x} c_2(x)\cdot \sqrt{q} \cdot \left(1+\tfrac{1}{2}\ln(qx)\right) +
\sum_{x} c_3(x) \cdot \tfrac{2q}{2q-1} \cdot \left(1+\tfrac{1}{2}\ln(q^2x)\right)}{c \cdot q^2\cdot\frac{q-1}{q+1}}
\end{aligned}$$ as a differentiable function of $q$. Then one easily checks that the first derivative after $q$ is $<0$.
The main theorem of this subsection is now a direct consequence of Lemma $3.6$ and Lemma $3.7$.
Let $L$ be a strongly square free, totally-reflective lattice.
- Let $\dim L=3$ and $\det L =q_1 \cdots q_s$. Then $s \leq 10$.
- Let $\dim L=4$ and $\det L =p_1^2 \cdots p_r^2 q_1 \cdots q_s$. Then $r \leq 9$ and $s \leq 8-r$.
We have to decide when the necessary condition $\frac{\operatorname{M_{ref}}}{\operatorname{M}} \geq 1 $ is violated.
\(a) Let $L$ be a strongly square free, totally-reflective lattice with $\det L = q_1 \cdots q_s$ and $s \geq 11$. Assume the prime factors are ordered such that $q_1 < \dots < q_s$. We start with the observation $$\begin{aligned}
\frac{\operatorname{M_{ref}}(2 \cdot 3 \cdot 5 \cdot 7 \cdot 11 \cdot 13 \cdot 17 \cdot 19 \cdot 23 \cdot 29 \cdot 31)}{\operatorname{M}(2 \cdot 3 \cdot 5 \cdot 7 \cdot 11 \cdot 13 \cdot 17 \cdot 19 \cdot 23 \cdot 29 \cdot 31)} < 1.
\end{aligned}$$ Using the monotony statement of Lemma $3.7$, we get $$\begin{aligned}
\frac{\operatorname{M_{ref}}(q_1 \cdot q_2 \cdot q_3 \cdot q_4 \cdot q_5 \cdot q_6 \cdot q_7 \cdot q_8 \cdot q_9 \cdot q_{10} \cdot q_{11})}
{\operatorname{M}(q_1 \cdot q_2 \cdot q_3 \cdot q_4 \cdot q_5 \cdot q_6 \cdot q_7 \cdot q_8 \cdot q_9\cdot q_{10} \cdot q_{11})} < 1,
\end{aligned}$$ which is possible since $s \geq 11$. Now we apply Lemma $3.6$ (a) and see that $$\begin{aligned}
\frac{\operatorname{M_{ref}}(q_1 \cdots q_{11} \cdot q_{12} \cdots q_s)}{\operatorname{M}(q_1 \cdots q_{11} \cdot q_{12} \cdots q_s)} = \frac{\operatorname{M_{ref}}(\det L)}{\operatorname{M}(\det L)} < 1.
\end{aligned}$$ Thus $L$ is not totally-reflective.
\(b) Let $L$ be a strongly square free, totally-reflective lattice of determinant $\det L = p_1^2 \cdots p_r^2 q_1 \cdots q_s$ with $r \geq 10$. First we prove the statement regarding $r$. Assume $p_1 < \dots < p_r$ and $q_1 < \dots < q_s$. We have $$\begin{aligned}
\frac{\operatorname{M_{ref}}(2^2 \cdot 3^2 \cdot 5^2 \cdot 7^2 \cdot 11^2 \cdot 13^2 \cdot 17^2 \cdot 19^2 \cdot 23^2 \cdot 29^2)}
{\operatorname{M}(2^2 \cdot 3^2 \cdot 5^2 \cdot 7^2 \cdot 11^2 \cdot 13^2 \cdot 17^2 \cdot 19^2 \cdot 23^2 \cdot 29^2)} < 1
\end{aligned}$$ and with Lemma $3.7$ $$\begin{aligned}
\frac{\operatorname{M_{ref}}(p_1^2 \cdot p_2^2 \cdot p_3^2 \cdot p_4^2 \cdot p_5^2 \cdot p_6^2 \cdot p_7^2 \cdot p_8^2 \cdot p_9^2 \cdot p_{10}^2)}
{\operatorname{M}(p_1^2 \cdot p_2^2 \cdot p_3^2 \cdot p_4^2 \cdot p_5^2 \cdot p_6^2 \cdot p_7^2 \cdot p_8^2 \cdot p_9^2 \cdot p_{10}^2)} < 1.
\end{aligned}$$ Applying Lemma $3.6$ part (a) and (b), we get $$\begin{aligned}
\frac{\operatorname{M_{ref}}(p_1^2 \cdots p_{10}^2 \cdot p_{11}^2 \cdots p_{r}^2 \cdot q_1 \cdots q_s)}
{\operatorname{M}(p_1^2 \cdots p_{10}^2 \cdot p_{11}^2 \cdots p_{r}^2 \cdot q_1 \cdots q_s)} = \frac{\operatorname{M_{ref}}(\det L)}{\operatorname{M}(\det L)} < 1.
\end{aligned}$$ Thus $L$ is not totally-reflective.
To prove the statement concerning $s$, we fix the number of quadratic prime factors $r \leq 9$. Let $\det L = p_1^2 \cdots p_r^2 q_1 \cdots q_s$ with $s \geq 9-r$. Define $\PP(r+s)$ to be the finite set consisting of the first $r+s$ primes. For each combination $(\tilde{p}_1, \dots, \tilde{p}_r,\tilde{q}_1, \dots, \tilde{q}_s) \in \PP(r+s)^{r+s}$ with $\tilde{p}_1 < \dots < \tilde{p}_r$ and $\tilde{q}_1 < \dots < \tilde{q}_s$ we have $$\begin{aligned}
\frac{\operatorname{M_{ref}}(\tilde{p}_1^2 \cdots \tilde{p}_r^2 \tilde{q}_1 \cdots \tilde{p}_s)}
{\operatorname{M}(\tilde{p}_1^2 \cdots \tilde{p}_r^2 \tilde{q}_1 \cdots \tilde{p}_s)} < 1.
\end{aligned}$$ Then Lemma $3.6$ and Lemma $3.7$ implie $$\begin{aligned}
\frac{\operatorname{M_{ref}}(p_1^2 \cdots p_r^2 q_1 \cdots q_s)}
{\operatorname{M}(p_1^2 \cdots p_r^2 q_1 \cdots q_s)} < 1.
\end{aligned}$$
Bounds on the prime factors
---------------------------
In the previous subsection we used the mass formula to estimate the part of the mass coming from those reflective lattices that decompose into one- and two-dimensional sublattices. For the purpose of this subsection (step $2$ of the general strategy) a different approach seems more appropriate.
Let $L$ be a strongly square free lattice with determinant $d$.
- For $\dim L = 3$ we set $$\begin{aligned}
\operatorname{N_{ref}}(L):=\sum_{x \mid d} \frac{1}{2} \left( 2\frac{2^{\Omega(x)}}{4} + \frac{5}{24} \right).
\end{aligned}$$
- For $\dim L = 4$ we set $$\begin{aligned}
\operatorname{N_{ref}}(L) &:=3\frac{4^{\Omega(d)}}{16} + 2\frac{3^{\Omega(d)}}{32} + \frac{2^{\Omega(d)}}{72} + 3\frac{2^{\Omega(d)}}{96} + \frac{53}{5760} \\
&+ \sum_{x \mid d} \frac{1}{4} \left( 2\frac{2^{\Omega(x)}}{4} + \frac{5}{24} \right)\cdot \left( 2\frac{2^{\Omega(d/x)}}{4} + \frac{5}{24} \right).
\end{aligned}$$
Then, in both cases, we have $\operatorname{m_{ref}}(L) \leq \operatorname{N_{ref}}(L)$.
\(b) Like in the proof of Lemma $3.5$ we start with the equation $$\begin{aligned}
\operatorname{m_{ref}}(L) = \operatorname{m}^{(4)}_{\mathrm{ref}}(L) + \operatorname{m}^{(3)}_{\mathrm{ref}}(L) + \operatorname{m}^{(2)}_{\mathrm{ref}}(L).
\end{aligned}$$ The estimates on $\operatorname{m}^{(3)}_{\mathrm{ref}}(L)$ and $\operatorname{m}^{(4)}_{\mathrm{ref}}(L)$ are the same as in $3.5$. But this time we rather use the classification theorem from [@bs96] than the mass formula to control $\operatorname{m}^{(2)}_{\mathrm{ref}}(L)$. Consider a reflective lattice $M = M_1 \perp M_2$ with $\det M_1 =x$ and the associated pair $({\mathsf R}(M_1),\CL(M_1))$. The classification [@bs96], $4.4$, $4.5$, $4.7$, implies that a bound for the number of possible isometry classes for $M_1$ is $a^{\Omega(x)}$, where $a$ is the number of occurring scaling factors. Furthermore, each of the $a^{\Omega(x)}$ isometry classes can occur $a^{\Omega(d/x)}$ times, thus $$\begin{aligned}
\operatorname{m}^{(2)}_{\mathrm{ref}}(L) & = \sum_{\substack{\text{$M \in \CG(L)$ refl,} \\M = M_1 \perp M_2,\\ \dim M_1=2}} \frac{1}{|O(M)|}
\leq \frac{1}{4} \sum_{\substack{\text{$M \in \CG(L)$ refl,} \\M = M_1 \perp M_2,\\ \dim M_1=2}} \frac{1}{|O(M_1)|} \\
& \leq \sum_{x \mid d} \frac{1}{4} \left( 2\frac{2^{\Omega(x)}}{4} + \frac{5}{24} \right)\cdot \left( 2\frac{2^{\Omega(d/x)}}{4} + \frac{5}{24} \right).
\end{aligned}$$
\(a) When considering a $3$-dimensional lattice $M$ decomposed as $M = M_1 \perp M_2$, such that $\dim M_1 =2, \dim M_2 =1$, the estimate for the number of possible isometry classes for $M_1$ works similarly. Each isometry class, however, occurs only one time which follows from Witt’s cancelation theorem. Thus $$\begin{aligned}
\operatorname{m}^{(2)}_{\mathrm{ref}}(L) &= \sum_{\substack{\text{$M \in \CG(L)$ refl,} \\M = M_1 \perp M_2,\\ \dim M_1=2}} \frac{1}{|O(M)|}
\leq \frac{1}{2} \sum_{\substack{\text{$M \in \CG(L)$ refl,} \\M = M_1 \perp M_2,\\ \dim M_1=2}} \frac{1}{|O(M_1)|} \\
& \leq \sum_{x \mid d} \frac{1}{2} \left( 2\frac{2^{\Omega(x)}}{4} + \frac{5}{24} \right).
\end{aligned}$$
In analogy to $\operatorname{M_{ref}}$, the quantity $\operatorname{N_{ref}}$ only depends on the prime factors of the determinant, thus the formulation in the following lemma makes sense.
In both dimensions $\frac{\operatorname{N_{ref}}(d)}{\operatorname{M}(d)}$ is monotonically decreasing in each prime factor of $d$.
The enumerator $\operatorname{N_{ref}}(d)$ only depends on the number of prime factors of $d$ and, as shown in the proof of Lemma $3.7$, the denominator $\operatorname{M}(d)$ is monotonically increasing in each prime factor.
As a first application we can slightly improve our bounds for the number of prime factors in dimension $3$.
Let $L$ be a strongly square free, totally-reflective lattice with $\dim L=3$ and $\det L =q_1 \cdots q_s$. Then $s \leq 9$.
This follows from $$\begin{aligned}
\frac{\operatorname{N_{ref}}(2 \cdot 3 \cdot 5 \cdot 7 \cdot 11 \cdot 13 \cdot 17 \cdot 19 \cdot 23 \cdot 29)}{\operatorname{M}(2 \cdot 3 \cdot 5 \cdot 7 \cdot 11 \cdot 13 \cdot 17 \cdot 19 \cdot 23 \cdot 29)} < 1
\end{aligned}$$ and Lemma $3.10$.
Keeping in mind that a strongly square free, totally-refelctive lattice $L$ fulfills the condition $\operatorname{N_{ref}}(L) / \operatorname{M}(L) \geq 1$, Lemma $3.10$ provides (from the computational point of view) strong bounds on the prime factors of $\det L$.
Let $L$ be a strongly square free, totally-reflective lattice.
- Let $\dim L=3$ and $\det L =q_1 \cdots q_s$ with $s \leq 9$. Assume $q_1 < \dots < q_s$. Then
$
\begin{array}{c|c|c|c|c|c|c|c|c|c|}
& q_1 & q_2 & q_3 & q_4 & q_5 & q_6 & q_7 & q_8 & q_9\\
\hline
\leq & 89 & 257 & 733 & 1063 & 1033 & 607 & 293 & 113 & 37
\end{array}.
$
- Let $\dim L=4$ and $\det L =p_1^2 \cdots p_r^2 q_1 \cdots q_s$ with $r \leq 9$ and $s \leq 8-r$. Assume $p_1 < \dots < p_r$ and $q_1 < \dots < q_s$. Then
$
\begin{array}{c|c|c|c|c|c|c|c|c|c|}
& p_1 & p_2 & p_3 & p_4 & p_5 & p_6 & p_7 & p_8 & p_9 \\
\hline
\leq & 191 & 661 & 1601 & 2069 & 1831 & 997 & 449 & 157 & 47
\end{array},
$
$
\begin{array}{c|c|c|c|c|c|c|c|c|}
& q_1 & q_2 & q_3 & q_4 & q_5 & q_6 & q_7 & q_8\\
\hline
\leq & 11287 & 6427 & 3613 & 1597 & 653 & 229 & 67 & 19
\end{array}.
$
By using Lemma $3.10$ we can repeatedly increase a prime factor (and thus decrease the function $\operatorname{N_{ref}}/\operatorname{M}$) until the necessary condition $\operatorname{N_{ref}}(\det L) / \operatorname{M}(\det L) \geq 1$ is violated.
- With the help of Theorem $3.8$ and Theorem $3.12$, the enumeration of all strongly square free, totally-reflective genera can be carried out computationally. One produces all genera up to the given bounds and checks whether they are totally-reflective. The number of possible genera is finite since we deal with strongly square free lattices. The list is included in section $6$.
- It turns out in section $6$ that the largest occuring value for the number of prime factors is $$\begin{aligned}
(r,s)=
\begin{cases}
(3,3), & \text{in dimension $4$}, \\
(0,4), & \text{in dimension $3$}.
\end{cases}
\end{aligned}$$ The largest prime factor $p$ occuring in dimension $4$ is $$\begin{aligned}
p=
\begin{cases}
13, & \text{if $v_p(\det) = 2$}, \\
17, & \text{if $v_p(\det) = 1$}.
\end{cases}
\end{aligned}$$ and $p = 23$ in dimension 3.
Completing the classification
=============================
Now, after we found all strongly square free, totally-reflective genera, our next goal is to gradually weaken the restriction “strongly square free”. This is done in two steps. First, we drop the assumption “strong” by applying the partial dualization operator (this is easy). In a second step we drop the assumption “square free” which turnes out to be somewhat more difficult. We need to clarify for which primes we have to consider pre-images under the Watson transformation.
From strongly square free to square free
----------------------------------------
Let $L$ be an integral lattice and $p \in \PP$. The *partial dual* of $L$ at $p$ is defined as $\operatorname{D}_p(L):={}^{p}\!{(\tfrac{1}{p}L \cap L^{\#}})$.
In contrast to the usual dual operator, the partial dual operator only dualizes the lattice at the prime spot $p$. That means $$\begin{aligned}
\operatorname{D}_p(L) \otimes_{\ZZ} \ZZ_q =
\begin{cases}
{}^{p}\!{\left(L_q^{\#}\right)}, & \text{if $q = p$}, \\
{}^{p}\!{L_q}, & \text{if $q \neq p$}.
\end{cases}\end{aligned}$$ This has the following effect on the Jordan decomposition of a square free lattice $L \otimes_{\ZZ} \ZZ_p = L_0 \perp {}^{p}{L_1}$: $$\begin{aligned}
\operatorname{D}_p(L) \otimes_{\ZZ} \ZZ_p = L_1 \perp {}^{p}{L_0}.\end{aligned}$$ Thus, starting with a strongly square free lattice, one can construct a (not necessarily strongly) square free, primitive lattice by applying $\operatorname{D}_p$ for $p \mid \det L$ (and vice versa). For a set of primes $I:=\left\{p_1,\cdots,p_k \right\} \subseteq \PP$ we use the abbreviation $\operatorname{D}_I:=\operatorname{D}_{p_1} \circ \cdots \circ \operatorname{D}_{p_k}$ (where $\operatorname{D}_\emptyset := \mathrm{id}$) which is well-defined since two partial dual operators with respect to different primes commute. Clearly, $\operatorname{D}_p$ extends to a well-defined bijective function $\CG(L) \longrightarrow \operatorname{D}_p(\CG(L))=\CG(\operatorname{D}_p(L))$.
The next lemma shows that the partial dual behaves “well” relative to the property “totally-reflective”.
Let $L$ be an integral lattice. Then $L$ is totally-reflective if and only if $\operatorname{D}_p(L)$ is totally-reflective.
Recall that the property “reflective” can be characterized by the action of $W(L)$ on $V$. Since $W(L)=W(\operatorname{D}_p(L))$, it is clear that $L$ is reflective iff $\operatorname{D}_p(L)$ is reflective. Thus, the assertion follows from the bijectivity of $\operatorname{D}_p: \CG(L) \longrightarrow \operatorname{D}_p(\CG(L))$.
Let $\CT_{n}$ be the set of all strongly square free, totally-reflective genera in dimension $n \in \{3,4\}$. Let $\CP(d)$ be the power set of the set of all prime factors of $d:=\det \CG$. Then $$\begin{aligned}
\bigcup_{\CG \in \CT_n} \bigcup_{I \in \CP(d)} \left\{\operatorname{D}_I(\CG)\right\}
\end{aligned}$$ is the set of all square free, totally-reflective, primitve genera in dimension $n$.
This is a consequence of Lemma $4.2$, the bijectivity of $\CG(L) \longrightarrow \operatorname{D}_p(\CG(L))$ and the above discussion.
From square free to all
-----------------------
The techniques we will use in this subsection are based on the following definition, going back to Watson, cf. [@wat62], [@wat73].
Let $L$ be an integral lattice and $p \in \PP$. The *Watson transformation* of $L$ at $p$ is defined as $\operatorname{E}_p(L):=L + (\tfrac{1}{p}L \cap pL^{\#})$.
The usefulness of $\operatorname{E}_p$ becomes clear when we consider its effect on the Jordan decomposition. Let $L$ be an integral lattice with $L_p = L_0 \perp {}^{p}\!{L_1} \perp \dots \perp {}^{p^r}\!{L_r}$. Then $$\begin{aligned}
\operatorname{E}_p(L) \otimes_{\ZZ} \ZZ_q =
\begin{cases}
\left(L_0 \perp L_2\right) \perp {}^{p}\!{\left(L_1 \perp L_3\right)} \perp {}^{p^2}\!{L_4} \perp \dots \perp {}^{p^{r-2}}\!{L_r} , & \text{if $q = p$}, \\
L \otimes_{\ZZ} \ZZ_q, & \text{if $q \neq p$}.
\end{cases}\end{aligned}$$ Hence, after repeatedly applying the Watson transformation, a primitive lattice transforms into a square free, primitive lattice. Similar to the partial dual, $\operatorname{E}_p$ extends to a well-defined surjective function $\CG(L) \longrightarrow \operatorname{E}_p(\CG(L)) = \CG(\operatorname{E}_p(L))$.
Let $L$ be a totally-reflective lattice. Then $\operatorname{E}_p(L)$ is totally-reflective.
The assertion implies that $W(L)$ has no nonzero fixed vectors, thus neither has $W(\operatorname{E}_p(L))$ since $W(L) \subseteq W(\operatorname{E}_p(L))$. Hence the assertion follows from the surjectivity of $\operatorname{E}_p$.
It may happen that prime factors disappear from the determinant after applying the Watson transformation. Thus, when calculating pre-images under $\operatorname{E}_p$, one has to decide which primes $p$ to consider (besides the prime factors of the determinant). An answer to this question is given by the follwing two lemmata.
Let $L$ be an integral lattice, $p$ an odd prime with $p \nmid \det L$ and $K \in \operatorname{E}_p(L)^{-1}$.
- If $\dim L = 3$ then, $$\begin{aligned}
\operatorname{m}(K) \geq \left(\left(\frac{1}{1+p^{-1}}\right)^2 p^2 \left(1-p^{-2}\right)\right) \cdot \operatorname{m}(L).
\end{aligned}$$
- If $\dim L = 4$ then, $$\begin{aligned}
\operatorname{m}(K) \geq \left( \frac{\zeta(4)}{2\zeta(2)^2} \left(\frac{1}{1+p^{-1}}\right)^2 p^3 \left(1-p^{-2}\right)\right) \cdot \operatorname{m}(L).
\end{aligned}$$
This follows from the mass formula and Lemma $3.2$.
Since we no longer deal only with strongly square free lattices, the following extended definition of $\operatorname{N_{ref}}(L)$ becomes necessary: $$\begin{aligned}
\operatorname{N_{ref}}(L) := \operatorname{N}^{(4)}_{\mathrm{ref}}(L) + \operatorname{N}^{(3)}_{\mathrm{ref}}(L) + \operatorname{N}^{(2)}_{\mathrm{ref}}(L),\end{aligned}$$ with $$\begin{aligned}
\operatorname{N}^{(4)}_{\mathrm{ref}}(L) &:=3\frac{4^{\Omega(d)}}{16} + 2\frac{3^{\Omega(d)}}{32} + \frac{2^{\Omega(d)}}{72} + 3\frac{2^{\Omega(d)}}{96} + \frac{53}{5760},\\
\operatorname{N}^{(3)}_{\mathrm{ref}}(L) &:=\sum_{x \mid d}2\frac{3^{\Omega(x)}}{8} + \frac{2^{\Omega(x)}}{16}+\frac{1}{24},\\
\operatorname{N}^{(2)}_{\mathrm{ref}}(L) &:=
\begin{cases}
\sum_{x \mid d} \frac{1}{4} \left( 2\frac{2^{\Omega(x)}}{4} + \frac{5}{24} \right)\cdot \left( 2\frac{2^{\Omega(d/x)}}{4} + \frac{5}{24} \right), & \text{if $\dim L =4$},\\
\sum_{x \mid d} \frac{1}{2} \left( 2\frac{2^{\Omega(x)}}{4} + \frac{5}{24} \right), & \text{if $\dim L =3$}.
\end{cases}\end{aligned}$$ It follows from a calculation similar to Lemma $3.9$ that $\operatorname{m_{ref}}(L) \leq \operatorname{N_{ref}}(L)$, for a not necessarily strongly square free lattice $L$.
Let $L$ be an integral lattice, $p$ an odd prime with $p \nmid \det L$ and $K \in \operatorname{E}_p(L)^{-1}$.
- If $\dim L = 3$ then, $$\begin{aligned}
\operatorname{N_{ref}}(K) \leq 81 \cdot \operatorname{N_{ref}}(L).
\end{aligned}$$
- If $\dim L = 4$ then, $$\begin{aligned}
\operatorname{N_{ref}}(K) \leq 5103 \cdot \operatorname{N_{ref}}(L).
\end{aligned}$$
\(b) Let $d:=\det L$. The assumption implies $K \otimes_{\ZZ} \ZZ_p = K_0 \perp {}^{p^{2}}\!{K_{2}}$, in particular $\det K = \det L \cdot p^{2\dim K_2}$, where $n_2:=\dim K_2 \leq 3$. Thus $$\begin{aligned}
\operatorname{N}^{(4)}_{\mathrm{ref}}(K) &=3\frac{4^{\Omega(dp^{2n_2})}}{16} + 2\frac{3^{\Omega(dp^{2n_2})}}{32} + \frac{2^{\Omega(dp^{2n_2})}}{72}
+ 3\frac{2^{\Omega(dp^{2n_2})}}{96} + \frac{53}{5760} \\
& \leq 4^{2n_2} \cdot \operatorname{N}^{(4)}_{\mathrm{ref}}(L).
\end{aligned}$$ For the other two cases we consider the decomposition $$\begin{aligned}
D(\det K) = \bigcupdot_{i=0}^{2n_2}p^i D(d).
\end{aligned}$$ Hence $$\begin{aligned}
\operatorname{N}^{(3)}_{\mathrm{ref}}(K) &=\sum_{x \mid D(d)}2\frac{3^{\Omega(x)}}{8} + \frac{2^{\Omega(x)}}{16}+\frac{1}{24} \\
&+\sum_{x \mid D(d)}2\frac{3^{\Omega(px)}}{8} + \frac{2^{\Omega(px)}}{16}+\frac{1}{24}\\
& \vdots \\
&+\sum_{x \mid D(d)}2\frac{3^{\Omega(p^{2n_2}x)}}{8} + \frac{2^{\Omega(p^{2n_2}x)}}{16}+\frac{1}{24}\\
&\leq 3^{2n_2} (1+2n_2) \cdot \operatorname{N}^{(3)}_{\mathrm{ref}}(L).
\end{aligned}$$ and $$\begin{aligned}
\operatorname{N}^{(2)}_{\mathrm{ref}}(K) &= \sum_{x \mid d} \frac{1}{4} \left( 2\frac{2^{\Omega(x)}}{4} + \frac{5}{24} \right)\cdot \left( 2\frac{2^{\Omega(dp^{2n_2}/x)}}{4} + \frac{5}{24} \right)\\
&= \sum_{x \mid d} \frac{1}{4} \left( 2\frac{2^{\Omega(px)}}{4} + \frac{5}{24} \right)\cdot \left( 2\frac{2^{\Omega(dp^{2n_2-1}/x)}}{4} + \frac{5}{24} \right) \\
& \vdots \\
&= \sum_{x \mid d} \frac{1}{4} \left( 2\frac{2^{\Omega(p^{2n_2}x)}}{4} + \frac{5}{24} \right)\cdot \left( 2\frac{2^{\Omega(d/x)}}{4} + \frac{5}{24} \right)\\
&\leq 2^{2n_2}(1+2n_2) \cdot \operatorname{N}^{(2)}_{\mathrm{ref}}(L).
\end{aligned}$$ Finally note that $\max\{4^{2n_2}, 3^{2n_2} (1+2n_2), 2^{2n_2}(1+2n_2) \} \leq 5103$ for $n_2 \in \{1,2,3\}$.
Part (a) of this lemma follows from a similar calculation.
Let $L$ be an integral lattice, $p$ an odd prime with $p \nmid \det L$ and $K \in \operatorname{E}_p(L)^{-1}$.
- If $K$ is totally-reflective and $\dim L =3$ then $$\begin{aligned}
81 \cdot \frac{\operatorname{N_{ref}}(L)}{\operatorname{m}(L)} \cdot \left(\left(\frac{1}{1+p^{-1}}\right)^2 p^2 \left(1-p^{-2}\right)\right)^{-1} \geq 1.
\end{aligned}$$
- If $K$ is totally-reflective and $\dim L =4$ then $$\begin{aligned}
5103 \cdot \frac{\operatorname{N_{ref}}(L)}{\operatorname{m}(L)} \cdot \left( \frac{\zeta(4)}{2\zeta(2)^2} \left(\frac{1}{1+p^{-1}}\right)^2 p^3 \left(1-p^{-2}\right)\right)\geq 1.
\end{aligned}$$
Combine Lemma $4.6$ and Lemma $4.7$.
Since the $p$-term in the above inequalities depends monotonically decreasingly on $p$ (for $p \nmid 2\det L$) and $\operatorname{N_{ref}}(L)/{\operatorname{m}(L)}$ does not depend on $p$ at all, it is straightforward to decide when the statement of Corollary $4.8$ is satisfied.
Given the set of all square free, totally-reflective, primitive genera, one can produce all totally-reflective, primitive genera by using Corollary $4.8$ and Lemma $4.5$. First, Corollary $4.8$ tells us which (finitely many) primes one has to concider when calculating pre-images under $\operatorname{E}_p$. Then, during the process of repeatedly generating lattices $K \in \operatorname{E}_p(L)^{-1}$, Lemma $4.5$ tells us that we can stop and proceed with the next lattice when a non totally-reflective lattice occurs. Eventually this process will terminate since the number of totally-reflective genera is finite. Furthermore, Lemma $4.5$ implies that every totally-reflective genus will be produced this way.
Reflective Lorentzian lattices and hyperbolic reflection groups
===============================================================
We want to outline some applications to the theory of reflective Lorentzian lattices and hyperbolic reflection groups. A more detailed investigation will be presented in a (near) future work.
A Lorentzian lattice $E$, that is an integral $\ZZ$-lattice of signature $(n,1)$, is called *reflective* if the Weyl group of $E$ has a finite index in $O(E)$. In terms of reflection groups this means that $W(E)$ induces an arithmetic reflection group on the hyperbolic space of dimension $n$ that has a fundamental polyhedron of finite volume. The connection to totally-reflective, postive definite lattices is described in the next lemmata.
Let $E$ be a square free, reflective lattice of signatrue $(n,1)$ with $n \geq 4$. Then $E$ can be written as $E = {}^{\alpha}\!{\mathbb{H}} \perp L$ with $L$ beeing a square free, totally-reflective, positive definite lattice of dimension $n-1$ and $\alpha \in \ZZ$. If $E$ is strongly square free one can choose $ \alpha = 1$ in the cases $n = 6, n \geq 8$ and $\alpha \in \{1,2\}$ in the cases $n=4,5,7$.
For $n = 6$ and $n \geq 8$ this follows immediately from [@vin81], Proposition $20$. In cases where this proposition is not applicable one uses elementary manipulations of the genus symbol of $E$ to see that a (scaled) hyperbolic plane splits $E$. This is possible since the class number of square free Lorentzian lattices is $1$. Vinberg’s Lemma implies that the positive definite part is totally-reflective, cf. [@vin72].
The group theoretical version of this lemma reads as follows:
Let $W$ be a maximal, arithmetic reflection group on the hyperbolic space of dimension $n \geq 4$. Then $W$ can be written as $$\begin{aligned}
W = W^{+}\left(\mathbb{H} \perp L\right) \quad \text{or} \quad W = W^{+}\left({}^{2}\!{\mathbb{H}} \perp L\right)
\end{aligned}$$ with $L$ beeing a square free, totally-reflective, positive definite lattice of dimension $n-1$. The $2$ scaled hyperbolic plane is only necessary in the cases $n \in \{4,5\}$.
Since $W$ is maximal and arithmetic one can find a Lorentzian lattice $E$ with $W = W(E)$. After repeated use of $\operatorname{D}_p$ and $\operatorname{E}_p$ we can assume that $E$ is strongly square free. The assertion then follows from the previous lemma.
The following list is a complete classification of all strongly square free, reflective lattices of signature $(5,1)$. As mentioned before, a more detailed proof will be given in a future work.
The Lorentzian lattices of signature $(5,1)$ in the table below are reflective. Every square free, reflective Lorentzian lattice of signature $(5,1)$ is isometric to one in the table below.
Our classification of totally-refelctive lattices provides a list of possible candidates to which we apply Vinberg’s algorithm, cf. [@vin272], §3. In the cases where the algorithm does not terminate, we prove non-reflectivity with the help of a method introduced by Bugaenko in [@bug92] and by embedding non-reflective lattices of smaller dimension.
The notation ${}^{\alpha}\!{\mathbb{H}} \perp \text{Genus}$ means that $L$ can be choosen arbitrary within the given $4$-dimensional genus. The combinatorial structure of the fundamental polyhedron is given as follows:
- = Number of fundamental roots = Number of $4$-dimensional faces,
- = Number of $3$-dimensional faces,
- = Number of $2$-dimensional faces,
- = Number of edges,
- = Number of vertices,
- = Number of cusps (vertices at infinity).
No. Lattice $r$ $f_3$ $f_2$ $e$ $v$ $c$
----- ----------------- --------------------------------------------------------------------------- ----- ------- ------- ------ ----- -----
1 1 $\mathbb{H} \perp \mathrm{I}(1_{4}^{+4})$ 6 15 20 15 5 1
2 2 $\mathbb{H} \perp \mathrm{I}(2_{5}^{-1})$ 7 20 30 24 8 1
3 3 $\mathbb{H} \perp \mathrm{I}(3^{-1})$ 7 21 33 27 9 1
4 3 $\mathbb{H} \perp \mathrm{I}(3^{+1})$ 8 25 40 34 12 1
5 4 $\mathbb{H} \perp \mathrm{II}(2_{\mathrm{II}}^{-2})$ 6 15 20 15 5 1
6 4 $\mathbb{H} \perp \mathrm{I}(2_{2}^{+2})$ 7 21 33 27 8 2
7 5 $\mathbb{H} \perp \mathrm{II}(5^{+1})$ 6 15 20 15 5 1
8 5 $\mathbb{H} \perp \mathrm{I}(5^{+1})$ 8 25 40 34 11 2
9 5 $\mathbb{H} \perp \mathrm{I}(5^{-1})$ 9 32 57 51 18 1
10 6 $\mathbb{H} \perp \mathrm{I}(2_{7}^{+1}3^{+1})$ 9 32 57 51 18 1
11 6 $\mathbb{H} \perp \mathrm{I}(2_{5}^{-1}3^{-1})$ 9 31 53 45 14 2
12 7 $\mathbb{H} \perp \mathrm{I}(5^{+1})$ 11 42 77 70 24 2
13 8 $\mathbb{H} \perp \mathrm{I}(2_{3}^{+3})$ 8 25 40 33 10 2
14 9 $\mathbb{H} \perp \mathrm{II}(3^{+2})$ 7 21 33 27 9 1
15 9 $\mathbb{H} \perp \mathrm{I}(3^{+2})$ 8 28 50 44 14 2
16 9 $\mathbb{H} \perp \mathrm{I}(3^{-2})$ 9 32 57 51 16 3
17 10 $\mathbb{H} \perp \mathrm{I}(2_{3}^{-1}5^{-1})$ 12 49 94 86 28 3
18 12 $\mathbb{H} \perp \mathrm{II}(2_{2}^{+2}3^{-1})$ 7 21 33 27 9 1
19 12 $\mathbb{H} \perp \mathrm{II}(2_{6}^{+2}3^{+1})$ 8 25 40 34 12 1
20 $2^2 \cdot 12$ ${}^{2}\!{\mathbb{H}} \perp \mathrm{I}(2_{\mathrm{II}}^{+2}3^{-1})$ 8 25 40 34 12 1
21 $2^2 \cdot 12$ ${}^{2}\!{\mathbb{H}} \perp \mathrm{I}(2_{\mathrm{II}}^{-2}3^{+1})$ 7 21 33 27 9 1
22 12 $\mathbb{H} \perp \mathrm{I}(2_{0}^{+2}3^{-1})$ 9 33 59 50 14 3
23 12 $\mathbb{H} \perp \mathrm{I}(2_{0}^{+2}3^{+1})$ 11 42 77 68 20 4
24 14 $\mathbb{H} \perp \mathrm{I}(2_{5}^{-1}7^{+1})$ 15 67 135 127 42 4
25 15 $\mathbb{H} \perp \mathrm{I}(3^{+1}5^{+1})$ 16 74 153 148 52 3
26 15 $\mathbb{H} \perp \mathrm{I}(3^{-1}5^{-1})$ 15 66 131 122 40 4
27 15 $\mathbb{H} \perp \mathrm{I}(3^{+1}5^{-1})$ 12 54 114 113 42 1
28 18 $\mathbb{H} \perp \mathrm{I}(2_{3}^{-1}3^{-2})$ 14 62 125 116 36 5
29 18 $\mathbb{H} \perp \mathrm{I}(2_{1}^{+1}3^{+2})$ 12 51 101 93 30 3
30 20 $\mathbb{H} \perp \mathrm{II}(2_{\mathrm{II}}^{+2}5^{+1})$ 8 25 40 34 11 2
31 20 $\mathbb{H} \perp \mathrm{II}(2_{\mathrm{II}}^{-2}5^{-1})$ 8 25 40 34 12 1
32 20 $\mathbb{H} \perp \mathrm{I}(2_{2}^{+2}5^{-1})$ 16 74 151 140 44 5
33 21 $\mathbb{H} \perp \mathrm{II}(3^{-1}7^{+1})$ 9 32 57 51 18 1
34 24 $\mathbb{H} \perp \mathrm{I}(2_{1}^{-3}3^{+1})$ 11 43 80 71 22 3
35 24 $\mathbb{H} \perp \mathrm{I}(2_{5}^{-3}3^{-1})$ 11 43 80 71 22 3
36 25 $\mathbb{H} \perp \mathrm{II}(5^{-2})$ 9 33 61 57 21 1
37 25 $\mathbb{H} \perp \mathrm{I}(5^{-2})$ 21 120 282 288 102 5
38 25 $\mathbb{H} \perp \mathrm{I}(5^{+2})$ 14 67 144 142 46 7
39 27 $\mathbb{H} \perp \mathrm{I}(3^{+3})$ 9 34 64 58 18 3
40 27 $\mathbb{H} \perp \mathrm{I}(3^{-3})$ 9 34 64 58 18 3
41 28 $\mathbb{H} \perp \mathrm{II}(2_{2}^{+2}7^{+1})$ 11 42 77 70 24 2
42 $2^2 \cdot 28$ ${}^{2}\!{\mathbb{H}} \perp \mathrm{I}(2_{\mathrm{II}}^{+2}7^{+1})$ 11 42 77 70 24 2
43 36 $\mathbb{H} \perp \mathrm{II}(2_{\mathrm{II}}^{-2}3^{-2})$ 7 21 33 27 8 2
44 36 $\mathbb{H} \perp \mathrm{II}(2_{\mathrm{II}}^{+2}3^{+2})$ 8 28 50 44 14 2
45 36 $\mathbb{H} \perp \mathrm{II}(2_{0}^{+2}3^{+2})$ 10 38 69 59 17 3
46 $2^2 \cdot 36$ ${}^{2}\!{\mathbb{H}} \perp \mathrm{I}(2_{\mathrm{II}}^{+2}3^{+2})$ 10 38 69 59 17 3
47 36 $\mathbb{H} \perp \mathrm{I}(2_{2}^{+2}3^{-2})$ 16 83 184 177 54 8
48 36 $\mathbb{H} \perp \mathrm{I}(2_{2}^{+2}3^{+2})$ 11 47 94 87 27 4
49 45 $\mathbb{H} \perp \mathrm{II}(3^{-2}5^{+1})$ 12 50 98 92 30 4
50 49 $\mathbb{H} \perp \mathrm{II}(7^{+2})$ 16 80 176 176 64 2
51 54 $\mathbb{H} \perp \mathrm{I}(2_{3}^{-1}3^{+3})$ 16 75 156 145 42 8
52 54 $\mathbb{H} \perp \mathrm{I}(2_{3}^{-1}3^{-3})$ 18 99 230 231 78 6
53 60 $\mathbb{H} \perp \mathrm{II}(2_{2}^{+2}3^{+1}5^{+1})$ 16 74 153 148 52 3
54 60 $\mathbb{H} \perp \mathrm{II}(2_{2}^{+2}3^{-1}5^{-1})$ 15 66 131 122 40 4
55 60 $\mathbb{H} \perp \mathrm{II}(2_{6}^{+2}3^{+1}5^{-1})$ 12 54 114 113 42 1
56 $2^2 \cdot 60$ ${}^{2}\!{\mathbb{H}} \perp \mathrm{I}(2_{\mathrm{II}}^{+2}3^{+1}5^{+1})$ 15 66 131 122 40 4
57 $2^2 \cdot 60$ ${}^{2}\!{\mathbb{H}} \perp \mathrm{I}(2_{\mathrm{II}}^{+2}3^{-1}5^{-1})$ 16 74 153 148 52 3
58 $2^2 \cdot 60$ ${}^{2}\!{\mathbb{H}} \perp \mathrm{I}(2_{\mathrm{II}}^{+2}3^{-1}5^{+1})$ 12 54 114 113 42 1
59 72 $\mathbb{H} \perp \mathrm{I}(2_{5}^{-3}3^{-2})$ 24 128 284 274 84 12
60 72 $\mathbb{H} \perp \mathrm{I}(2_{7}^{+3}3^{+2})$ 14 66 140 134 44 4
61 75 $\mathbb{H} \perp \mathrm{I}(3^{+1}5^{-2})$ 86 672 1788 1902 660 42
62 84 $\mathbb{H} \perp \mathrm{II}(2_{\mathrm{II}}^{-2}3^{+1}7^{+1})$ 16 74 153 148 52 3
63 100 $\mathbb{H} \perp \mathrm{II}(2_{\mathrm{II}}^{-2}5^{+2})$ 9 33 61 57 19 3
64 100 $\mathbb{H} \perp \mathrm{II}(2_{\mathrm{II}}^{+2}5^{-2})$ 21 120 282 288 102 6
65 108 $\mathbb{H} \perp \mathrm{II}(2_{6}^{+2}3^{-3})$ 9 34 64 58 18 3
66 108 $\mathbb{H} \perp \mathrm{II}(2_{6}^{-2}3^{+3})$ 9 34 64 58 18 3
67 $2^2 \cdot 108$ ${}^{2}\!{\mathbb{H}} \perp \mathrm{I}(2_{\mathrm{II}}^{+2}3^{-3})$ 9 34 64 58 18 3
68 $2^2 \cdot 108$ ${}^{2}\!{\mathbb{H}} \perp \mathrm{I}(2_{\mathrm{II}}^{-2}3^{+3})$ 9 34 64 58 18 3
69 108 $\mathbb{H} \perp \mathrm{I}(2_{6}^{+2}3^{+3})$ 16 85 193 188 56 10
70 108 $\mathbb{H} \perp \mathrm{I}(2_{4}^{-2}3^{-3})$ 16 85 193 188 56 10
71 125 $\mathbb{H} \perp \mathrm{II}(5^{+3})$ 10 40 80 80 30 2
72 125 $\mathbb{H} \perp \mathrm{I}(5^{+3})$ 20 115 280 295 100 12
73 180 $\mathbb{H} \perp \mathrm{II}(2_{\mathrm{II}}^{-2}3^{+2}5^{+1})$ 16 80 177 178 64 3
74 196 $\mathbb{H} \perp \mathrm{II}(2_{\mathrm{II}}^{-2}7^{-2})$ 15 72 156 159 54 8
75 216 $\mathbb{H} \perp \mathrm{I}(2_{7}^{+3}3^{+3})$ 24 138 324 324 104 12
76 216 $\mathbb{H} \perp \mathrm{I}(2_{1}^{-3}3^{-3})$ 24 138 324 324 104 12
77 300 $\mathbb{H} \perp \mathrm{II}(2_{6}^{-2}3^{+1}5^{-2})$ 86 672 1788 1902 660 42
78 $2^2 \cdot 300$ ${}^{2}\!{\mathbb{H}} \perp \mathrm{I}(2_{\mathrm{II}}^{-2}3^{-1}5^{-2})$ 86 672 1788 1902 660 42
79 500 $\mathbb{H} \perp \mathrm{II}(2_{\mathrm{II}}^{+2}5^{+3})$ 20 115 280 295 100 12
80 500 $\mathbb{H} \perp \mathrm{II}(2_{\mathrm{II}}^{-2}5^{-3})$ 12 56 124 126 44 4
: Reflective lattices of siganture $(5,1)$
Every maximal, arithmetic reflection group on the hyperbolic $5$-space is of the form $W^{+}(E)$ where $E$ is a lattice from the table above.
Totally-reflective genera
=========================
All discussed calculations were performed using $\mathrm{MAGMA}$, cf. [@mag]. We used some self-implemented programs to produce all genus symbols of strongly square free genera up to a given bound and determine whether a genus is totally-reflective as well as an algorithm for calculating pre-images under the Watson-transformation. Furthermore, we used a program to generate a representative $\ZZ$-lattice for a given genus symbol implemented by Kirschmer & Lorch, cf. [@kl13]. The (self-implemented) algorithms are available on request.
Dimension 3
-----------
$1234$ genera, of which $289$ are square free and 52 strongly square free.
$\det$ Genus h $\det$ Genus h $\det$ Genus h
-------- ------------------------------------------------------------- ---- -------- ------------------------------------------------------------- --- -------- ------------------------------------------------------------- ----
1 $\mathrm{I}(1_{3}^{+3})$ 1 2 $\mathrm{I}(2_{1}^{+1})$ 1 3 $\mathrm{I}(3^{-1})$ 1
3 $\mathrm{I}(3^{+1})$ 1 4 $\mathrm{I}(2_{2}^{+2})$ 1 4 $\mathrm{II}(4_{3}^{-1})$ 1
4 $\mathrm{I}(4_{1}^{+1})$ 1 5 $\mathrm{I}(5^{-1})$ 1 5 $\mathrm{I}(5^{+1})$ 1
6 $\mathrm{II}(2_{1}^{+1}3^{-1})$ 1 6 $\mathrm{I}(2_{1}^{+1}3^{+1})$ 1 6 $\mathrm{I}(2_{3}^{-1}3^{-1})$ 1
7 $\mathrm{I}(7^{+1})$ 2 8 $\mathrm{I}(2_{1}^{+1}4_{1}^{+1})$ 1 9 $\mathrm{I}(3^{-2})$ 1
9 $\mathrm{I}(9^{-1})$ 1 9 $\mathrm{I}(9^{+1})$ 1 9 $\mathrm{I}(3^{+2})$ 1
10 $\mathrm{I}(2_{1}^{+1}5^{-1})$ 2 10 $\mathrm{I}(2_{1}^{+1}5^{+1})$ 1 11 $\mathrm{I}(11^{-1})$ 1
12 $\mathrm{I}(2_{2}^{+2}3^{+1})$ 1 12 $\mathrm{I}(2_{\mathrm{II}}^{-2}3^{+1})$ 1 12 $\mathrm{I}(2_{4}^{-2}3^{-1})$ 1
12 $\mathrm{II}(4_{1}^{+1}3^{-1})$ 1 12 $\mathrm{I}(4_{7}^{+1}3^{-1})$ 1 12 $\mathrm{I}(4_{1}^{+1}3^{+1})$ 1
12 $\mathrm{II}(4_{5}^{-1}3^{+1})$ 1 12 $\mathrm{I}(4_{3}^{-1}3^{+1})$ 1 14 $\mathrm{II}(2_{1}^{+1}7^{+1})$ 1
14 $\mathrm{I}(2_{7}^{+1}7^{+1})$ 2 15 $\mathrm{I}(3^{+1}5^{+1})$ 2 15 $\mathrm{I}(3^{-1}5^{-1})$ 2
15 $\mathrm{I}(3^{+1}5^{-1})$ 1 15 $\mathrm{I}(3^{-1}5^{+1})$ 1 16 $\mathrm{I}(4_{\mathrm{II}}^{-2})$ 1
16 $\mathrm{I}(4_{2}^{+2})$ 1 16 $\mathrm{I}(16_{5}^{-1})$ 1 16 $\mathrm{I}(16_{1}^{+1})$ 2
17 $\mathrm{I}(17^{+1})$ 2 18 $\mathrm{II}(2_{3}^{-1}3^{+2})$ 1 18 $\mathrm{I}(2_{3}^{-1}3^{-2})$ 1
18 $\mathrm{I}(2_{1}^{+1}9^{-1})$ 2 18 $\mathrm{I}(2_{1}^{+1}3^{+2})$ 1 20 $\mathrm{I}(2_{2}^{-2}5^{+1})$ 2
20 $\mathrm{I}(2_{6}^{-2}5^{-1})$ 1 20 $\mathrm{II}(4_{7}^{+1}5^{+1})$ 1 20 $\mathrm{I}(4_{1}^{+1}5^{-1})$ 1
20 $\mathrm{I}(4_{1}^{+1}5^{+1})$ 2 20 $\mathrm{II}(4_{3}^{-1}5^{-1})$ 1 21 $\mathrm{I}(3^{+1}7^{+1})$ 2
21 $\mathrm{I}(3^{-1}7^{+1})$ 2 21 $\mathrm{I}(3^{+1}7^{-1})$ 1 24 $\mathrm{I}(2_{1}^{+1}4_{1}^{+1}3^{+1})$ 2
24 $\mathrm{I}(2_{3}^{-1}4_{1}^{+1}3^{-1})$ 1 25 $\mathrm{I}(5^{+2})$ 1 25 $\mathrm{I}(5^{-2})$ 1
27 $\mathrm{I}(3^{-1}9^{+1})$ 1 27 $\mathrm{I}(3^{-1}9^{-1})$ 1 27 $\mathrm{I}(3^{+1}9^{-1})$ 1
27 $\mathrm{I}(3^{+1}9^{+1})$ 1 28 $\mathrm{I}(2_{\mathrm{II}}^{+2}7^{+1})$ 1 28 $\mathrm{I}(2_{0}^{+2}7^{+1})$ 2
28 $\mathrm{II}(4_{1}^{+1}7^{+1})$ 2 28 $\mathrm{I}(4_{7}^{+1}7^{+1})$ 2 30 $\mathrm{I}(2_{5}^{-1}3^{+1}5^{-1})$ 1
30 $\mathrm{II}(2_{1}^{+1}3^{-1}5^{-1})$ 2 30 $\mathrm{II}(2_{1}^{+1}3^{+1}5^{+1})$ 1 30 $\mathrm{I}(2_{1}^{+1}3^{-1}5^{-1})$ 2
30 $\mathrm{I}(2_{7}^{+1}3^{+1}5^{+1})$ 2 32 $\mathrm{I}(2_{3}^{-1}16_{7}^{+1})$ 1 32 $\mathrm{I}(2_{1}^{+1}16_{5}^{-1})$ 1
32 $\mathrm{I}(2_{3}^{-1}16_{3}^{-1})$ 1 32 $\mathrm{I}(2_{1}^{+1}16_{1}^{+1})$ 1 33 $\mathrm{I}(3^{+1}11^{-1})$ 2
33 $\mathrm{I}(3^{-1}11^{+1})$ 2 35 $\mathrm{I}(5^{+1}7^{-1})$ 1 35 $\mathrm{I}(5^{-1}7^{+1})$ 2
36 $\mathrm{I}(2_{\mathrm{II}}^{-2}3^{+2})$ 1 36 $\mathrm{I}(2_{4}^{-2}3^{-2})$ 1 36 $\mathrm{I}(2_{2}^{+2}9^{+1})$ 2
36 $\mathrm{I}(2_{6}^{+2}3^{+2})$ 1 36 $\mathrm{II}(4_{3}^{-1}3^{-2})$ 1 36 $\mathrm{II}(4_{3}^{-1}9^{-1})$ 1
36 $\mathrm{II}(4_{3}^{-1}9^{+1})$ 1 36 $\mathrm{I}(4_{3}^{-1}3^{+2})$ 1 36 $\mathrm{II}(4_{7}^{+1}3^{+2})$ 1
36 $\mathrm{I}(4_{1}^{+1}3^{-2})$ 1 36 $\mathrm{I}(4_{1}^{+1}9^{+1})$ 2 36 $\mathrm{I}(4_{1}^{+1}3^{+2})$ 1
39 $\mathrm{I}(3^{+1}13^{-1})$ 1 40 $\mathrm{I}(2_{1}^{+1}4_{1}^{+1}5^{+1})$ 2 42 $\mathrm{II}(2_{3}^{-1}3^{-1}7^{+1})$ 2
42 $\mathrm{I}(2_{3}^{-1}3^{-1}7^{+1})$ 2 42 $\mathrm{I}(2_{5}^{-1}3^{-1}7^{-1})$ 2 44 $\mathrm{II}(4_{1}^{+1}11^{-1})$ 1
45 $\mathrm{I}(3^{+2}5^{+1})$ 2 45 $\mathrm{I}(3^{-2}5^{-1})$ 2 45 $\mathrm{I}(3^{+2}5^{-1})$ 1
45 $\mathrm{I}(3^{-2}5^{+1})$ 1 45 $\mathrm{I}(9^{-1}5^{+1})$ 2 48 $\mathrm{I}(4_{2}^{+2}3^{+1})$ 1
48 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+1})$ 1 48 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1})$ 1 48 $\mathrm{I}(4_{4}^{-2}3^{+1})$ 1
48 $\mathrm{I}(4_{2}^{-2}3^{-1})$ 1 48 $\mathrm{II}(16_{1}^{+1}3^{-1})$ 2 48 $\mathrm{I}(16_{1}^{+1}3^{+1})$ 2
48 $\mathrm{I}(16_{7}^{+1}3^{+1})$ 1 48 $\mathrm{I}(16_{5}^{-1}3^{+1})$ 2 48 $\mathrm{I}(16_{3}^{-1}3^{+1})$ 2
49 $\mathrm{I}(7^{+2})$ 2 50 $\mathrm{I}(2_{5}^{-1}5^{-2})$ 1 50 $\mathrm{I}(2_{1}^{+1}5^{+2})$ 2
51 $\mathrm{I}(3^{-1}17^{+1})$ 2 54 $\mathrm{II}(2_{1}^{+1}3^{-1}9^{-1})$ 2 54 $\mathrm{I}(2_{1}^{+1}3^{+1}9^{+1})$ 1
54 $\mathrm{I}(2_{1}^{+1}3^{+1}9^{-1})$ 1 54 $\mathrm{I}(2_{3}^{-1}3^{-1}9^{-1})$ 1 54 $\mathrm{I}(2_{3}^{-1}3^{-1}9^{+1})$ 1
56 $\mathrm{I}(2_{7}^{+1}4_{1}^{+1}7^{+1})$ 3 57 $\mathrm{I}(3^{+1}19^{-1})$ 2 60 $\mathrm{I}(2_{2}^{-2}3^{+1}5^{+1})$ 2
60 $\mathrm{I}(2_{4}^{-2}3^{-1}5^{+1})$ 1 60 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+1}5^{+1})$ 2 60 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-1}5^{-1})$ 1
60 $\mathrm{I}(2_{2}^{+2}3^{-1}5^{-1})$ 2 60 $\mathrm{II}(4_{5}^{-1}3^{+1}5^{-1})$ 1 60 $\mathrm{II}(4_{5}^{-1}3^{-1}5^{+1})$ 1
60 $\mathrm{I}(4_{3}^{-1}3^{+1}5^{+1})$ 2 60 $\mathrm{I}(4_{3}^{-1}3^{-1}5^{-1})$ 2 60 $\mathrm{II}(4_{1}^{+1}3^{+1}5^{+1})$ 2
60 $\mathrm{II}(4_{1}^{+1}3^{-1}5^{-1})$ 2 60 $\mathrm{I}(4_{7}^{+1}3^{-1}5^{+1})$ 1 63 $\mathrm{I}(3^{+2}7^{-1})$ 2
63 $\mathrm{I}(3^{-2}7^{-1})$ 2 63 $\mathrm{I}(3^{+2}7^{+1})$ 1 64 $\mathrm{I}(4_{1}^{+1}16_{5}^{-1})$ 2
65 $\mathrm{I}(5^{+1}13^{+1})$ 2 66 $\mathrm{I}(2_{1}^{+1}3^{+1}11^{-1})$ 3 68 $\mathrm{II}(4_{3}^{-1}17^{+1})$ 2
69 $\mathrm{I}(3^{-1}23^{+1})$ 3 72 $\mathrm{I}(2_{1}^{+1}4_{3}^{-1}3^{+2})$ 1 72 $\mathrm{I}(2_{3}^{-1}4_{3}^{-1}3^{-2})$ 2
75 $\mathrm{I}(3^{-1}5^{+2})$ 1 75 $\mathrm{I}(3^{+1}5^{-2})$ 1 75 $\mathrm{I}(3^{-1}5^{-2})$ 2
75 $\mathrm{I}(3^{+1}5^{+2})$ 2 77 $\mathrm{I}(7^{+1}11^{-1})$ 3 78 $\mathrm{II}(2_{1}^{+1}3^{-1}13^{-1})$ 3
80 $\mathrm{I}(4_{\mathrm{II}}^{+2}5^{-1})$ 1 80 $\mathrm{I}(4_{2}^{-2}5^{-1})$ 1 80 $\mathrm{I}(4_{6}^{-2}5^{+1})$ 2
80 $\mathrm{I}(4_{\mathrm{II}}^{-2}5^{+1})$ 1 80 $\mathrm{I}(16_{1}^{+1}5^{-1})$ 2 80 $\mathrm{I}(16_{1}^{+1}5^{+1})$ 3
80 $\mathrm{I}(16_{5}^{-1}5^{+1})$ 4 81 $\mathrm{I}(3^{+1}27^{-1})$ 1 81 $\mathrm{I}(9^{-2})$ 1
81 $\mathrm{I}(9^{+2})$ 1 84 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+1}7^{+1})$ 2 84 $\mathrm{I}(2_{2}^{+2}3^{+1}7^{-1})$ 2
84 $\mathrm{I}(2_{6}^{-2}3^{+1}7^{+1})$ 2 84 $\mathrm{II}(4_{7}^{+1}3^{-1}7^{+1})$ 2 84 $\mathrm{II}(4_{7}^{+1}3^{+1}7^{-1})$ 1
84 $\mathrm{I}(4_{1}^{+1}3^{+1}7^{+1})$ 2 84 $\mathrm{I}(4_{1}^{+1}3^{+1}7^{-1})$ 2 84 $\mathrm{II}(4_{3}^{-1}3^{+1}7^{+1})$ 2
90 $\mathrm{II}(2_{7}^{+1}3^{-2}5^{+1})$ 2 90 $\mathrm{I}(2_{7}^{+1}3^{+2}5^{+1})$ 1 90 $\mathrm{II}(2_{3}^{-1}3^{+2}5^{-1})$ 1
90 $\mathrm{I}(2_{3}^{-1}3^{-2}5^{+1})$ 2 90 $\mathrm{I}(2_{3}^{-1}3^{+2}5^{-1})$ 2 96 $\mathrm{I}(2_{3}^{-1}16_{3}^{-1}3^{+1})$ 2
96 $\mathrm{I}(2_{1}^{+1}16_{1}^{+1}3^{+1})$ 2 96 $\mathrm{I}(2_{3}^{-1}16_{5}^{-1}3^{-1})$ 1 96 $\mathrm{I}(2_{1}^{+1}16_{7}^{+1}3^{+1})$ 2
96 $\mathrm{I}(2_{3}^{-1}16_{1}^{+1}3^{-1})$ 1 96 $\mathrm{I}(2_{5}^{-1}16_{7}^{+1}3^{-1})$ 1 96 $\mathrm{I}(2_{1}^{+1}16_{5}^{-1}3^{+1})$ 2
96 $\mathrm{I}(2_{7}^{+1}16_{3}^{-1}3^{-1})$ 1 98 $\mathrm{II}(2_{7}^{+1}7^{+2})$ 1 98 $\mathrm{I}(2_{7}^{+1}7^{+2})$ 2
99 $\mathrm{I}(3^{+2}11^{+1})$ 2 99 $\mathrm{I}(3^{-2}11^{-1})$ 2 100 $\mathrm{I}(2_{6}^{-2}5^{-2})$ 1
100 $\mathrm{I}(2_{2}^{+2}5^{+2})$ 2 100 $\mathrm{II}(4_{3}^{-1}5^{+2})$ 1 100 $\mathrm{I}(4_{5}^{-1}5^{-2})$ 1
100 $\mathrm{I}(4_{5}^{-1}5^{+2})$ 2 100 $\mathrm{II}(4_{7}^{+1}5^{-2})$ 1 105 $\mathrm{I}(3^{+1}5^{-1}7^{+1})$ 2
105 $\mathrm{I}(3^{-1}5^{-1}7^{-1})$ 2 105 $\mathrm{I}(3^{-1}5^{+1}7^{+1})$ 2 105 $\mathrm{I}(3^{-1}5^{-1}7^{+1})$ 3
108 $\mathrm{I}(2_{2}^{+2}3^{+1}9^{-1})$ 1 108 $\mathrm{I}(2_{2}^{+2}3^{+1}9^{+1})$ 1 108 $\mathrm{I}(2_{\mathrm{II}}^{-2}3^{+1}9^{+1})$ 2
108 $\mathrm{I}(2_{4}^{-2}3^{-1}9^{+1})$ 1 108 $\mathrm{I}(2_{4}^{-2}3^{-1}9^{-1})$ 1 108 $\mathrm{II}(4_{1}^{+1}3^{-1}9^{+1})$ 1
108 $\mathrm{II}(4_{1}^{+1}3^{-1}9^{-1})$ 1 108 $\mathrm{I}(4_{1}^{+1}3^{+1}9^{+1})$ 2 108 $\mathrm{II}(4_{5}^{-1}3^{+1}9^{-1})$ 1
108 $\mathrm{II}(4_{5}^{-1}3^{+1}9^{+1})$ 1 108 $\mathrm{I}(4_{3}^{-1}3^{+1}9^{-1})$ 1 108 $\mathrm{I}(4_{3}^{-1}3^{+1}9^{+1})$ 1
108 $\mathrm{I}(4_{3}^{-1}3^{-1}9^{+1})$ 1 108 $\mathrm{I}(4_{3}^{-1}3^{-1}9^{-1})$ 1 112 $\mathrm{I}(4_{\mathrm{II}}^{+2}7^{+1})$ 2
112 $\mathrm{I}(4_{2}^{+2}7^{+1})$ 2 117 $\mathrm{I}(3^{+2}13^{-1})$ 1 120 $\mathrm{I}(2_{5}^{-1}4_{3}^{-1}3^{+1}5^{-1})$ 2
120 $\mathrm{I}(2_{7}^{+1}4_{1}^{+1}3^{+1}5^{+1})$ 3 121 $\mathrm{I}(11^{-2})$ 1 125 $\mathrm{I}(5^{+1}25^{-1})$ 1
125 $\mathrm{I}(5^{+1}25^{+1})$ 1 126 $\mathrm{I}(2_{3}^{-1}3^{+2}7^{+1})$ 2 126 $\mathrm{II}(2_{1}^{+1}3^{+2}7^{-1})$ 2
126 $\mathrm{II}(2_{1}^{+1}9^{-1}7^{+1})$ 3 126 $\mathrm{I}(2_{1}^{+1}3^{+2}7^{-1})$ 2 128 $\mathrm{I}(2_{1}^{+1}64_{1}^{+1})$ 4
132 $\mathrm{I}(2_{2}^{+2}3^{-1}11^{+1})$ 3 132 $\mathrm{II}(4_{3}^{-1}3^{+1}11^{-1})$ 2 132 $\mathrm{II}(4_{3}^{-1}3^{-1}11^{+1})$ 2
132 $\mathrm{I}(4_{1}^{+1}3^{-1}11^{+1})$ 3 135 $\mathrm{I}(3^{+1}9^{+1}5^{+1})$ 2 135 $\mathrm{I}(3^{-1}9^{+1}5^{-1})$ 3
135 $\mathrm{I}(3^{+1}9^{-1}5^{+1})$ 2 135 $\mathrm{I}(3^{-1}9^{-1}5^{+1})$ 2 135 $\mathrm{I}(3^{+1}9^{+1}5^{-1})$ 1
135 $\mathrm{I}(3^{+1}9^{-1}5^{-1})$ 1 140 $\mathrm{II}(4_{1}^{+1}5^{+1}7^{-1})$ 1 140 $\mathrm{II}(4_{1}^{+1}5^{-1}7^{+1})$ 2
144 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-2})$ 1 144 $\mathrm{I}(4_{\mathrm{II}}^{-2}9^{-1})$ 1 144 $\mathrm{I}(4_{\mathrm{II}}^{-2}9^{+1})$ 1
144 $\mathrm{I}(4_{4}^{-2}3^{+2})$ 1 144 $\mathrm{I}(4_{6}^{-2}3^{-2})$ 1 144 $\mathrm{I}(4_{2}^{+2}9^{+1})$ 2
144 $\mathrm{I}(4_{6}^{+2}3^{+2})$ 1 144 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+2})$ 1 144 $\mathrm{II}(16_{3}^{-1}3^{-2})$ 2
144 $\mathrm{I}(16_{5}^{-1}3^{+2})$ 1 144 $\mathrm{I}(16_{3}^{-1}3^{+2})$ 2 144 $\mathrm{I}(16_{5}^{-1}9^{+1})$ 3
144 $\mathrm{I}(16_{7}^{+1}3^{+2})$ 2 144 $\mathrm{I}(16_{1}^{+1}9^{+1})$ 4 144 $\mathrm{I}(16_{1}^{+1}3^{+2})$ 2
147 $\mathrm{I}(3^{-1}7^{-2})$ 2 147 $\mathrm{I}(3^{+1}7^{+2})$ 1 147 $\mathrm{I}(3^{+1}7^{-2})$ 2
150 $\mathrm{II}(2_{1}^{+1}3^{+1}5^{-2})$ 2 150 $\mathrm{I}(2_{7}^{+1}3^{+1}5^{-2})$ 2 150 $\mathrm{I}(2_{1}^{+1}3^{-1}5^{-2})$ 1
150 $\mathrm{II}(2_{5}^{-1}3^{-1}5^{+2})$ 1 150 $\mathrm{I}(2_{5}^{-1}3^{-1}5^{+2})$ 2 153 $\mathrm{I}(3^{+2}17^{-1})$ 2
156 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+1}13^{+1})$ 3 156 $\mathrm{II}(4_{5}^{-1}3^{+1}13^{-1})$ 1 160 $\mathrm{I}(2_{3}^{-1}16_{3}^{-1}5^{+1})$ 2
160 $\mathrm{I}(2_{1}^{+1}16_{1}^{+1}5^{+1})$ 2 160 $\mathrm{I}(2_{1}^{+1}16_{5}^{-1}5^{+1})$ 2 160 $\mathrm{I}(2_{5}^{-1}16_{7}^{+1}5^{+1})$ 2
162 $\mathrm{I}(2_{1}^{+1}9^{+2})$ 2 165 $\mathrm{I}(3^{+1}5^{+1}11^{-1})$ 2 171 $\mathrm{I}(3^{+2}19^{+1})$ 2
175 $\mathrm{I}(5^{+2}7^{-1})$ 2 175 $\mathrm{I}(5^{-2}7^{+1})$ 1 176 $\mathrm{I}(4_{\mathrm{II}}^{-2}11^{-1})$ 1
180 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+2}5^{+1})$ 1 180 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-2}5^{-1})$ 2 180 $\mathrm{I}(2_{0}^{+2}3^{+2}5^{+1})$ 2
180 $\mathrm{I}(2_{0}^{+2}3^{-2}5^{-1})$ 2 180 $\mathrm{I}(2_{4}^{-2}3^{+2}5^{-1})$ 1 180 $\mathrm{II}(4_{7}^{+1}3^{+2}5^{-1})$ 1
180 $\mathrm{II}(4_{7}^{+1}3^{-2}5^{+1})$ 1 180 $\mathrm{II}(4_{7}^{+1}9^{-1}5^{+1})$ 2 180 $\mathrm{I}(4_{1}^{+1}3^{-2}5^{-1})$ 2
180 $\mathrm{I}(4_{1}^{+1}3^{+2}5^{-1})$ 1 180 $\mathrm{II}(4_{3}^{-1}3^{+2}5^{+1})$ 2 180 $\mathrm{II}(4_{3}^{-1}3^{-2}5^{-1})$ 2
180 $\mathrm{I}(4_{5}^{-1}3^{+2}5^{+1})$ 2 189 $\mathrm{I}(3^{-1}9^{+1}7^{+1})$ 2 189 $\mathrm{I}(3^{+1}9^{-1}7^{-1})$ 1
189 $\mathrm{I}(3^{-1}9^{-1}7^{+1})$ 2 189 $\mathrm{I}(3^{+1}9^{+1}7^{-1})$ 1 192 $\mathrm{I}(4_{1}^{+1}16_{7}^{+1}3^{+1})$ 2
192 $\mathrm{I}(4_{3}^{-1}16_{1}^{+1}3^{+1})$ 2 192 $\mathrm{I}(4_{1}^{+1}16_{5}^{-1}3^{+1})$ 2 195 $\mathrm{I}(3^{-1}5^{-1}13^{-1})$ 3
196 $\mathrm{I}(2_{6}^{+2}7^{+2})$ 2 196 $\mathrm{I}(2_{\mathrm{II}}^{+2}7^{+2})$ 1 196 $\mathrm{I}(4_{5}^{-1}7^{+2})$ 2
196 $\mathrm{II}(4_{7}^{+1}7^{+2})$ 2 198 $\mathrm{I}(2_{1}^{+1}3^{+2}11^{-1})$ 3 200 $\mathrm{I}(2_{5}^{-1}4_{1}^{+1}5^{-2})$ 2
204 $\mathrm{II}(4_{1}^{+1}3^{-1}17^{+1})$ 2 207 $\mathrm{I}(3^{+2}23^{+1})$ 3 210 $\mathrm{II}(2_{7}^{+1}3^{+1}5^{+1}7^{+1})$ 2
210 $\mathrm{I}(2_{7}^{+1}3^{+1}5^{+1}7^{+1})$ 3 216 $\mathrm{I}(2_{1}^{+1}4_{1}^{+1}3^{+1}9^{+1})$ 2 216 $\mathrm{I}(2_{1}^{+1}4_{1}^{+1}3^{+1}9^{-1})$ 2
216 $\mathrm{I}(2_{3}^{-1}4_{1}^{+1}3^{-1}9^{+1})$ 2 224 $\mathrm{I}(2_{7}^{+1}16_{5}^{-1}7^{+1})$ 3 224 $\mathrm{I}(2_{1}^{+1}16_{7}^{+1}7^{+1})$ 3
224 $\mathrm{I}(2_{7}^{+1}16_{1}^{+1}7^{+1})$ 3 224 $\mathrm{I}(2_{3}^{-1}16_{3}^{-1}7^{+1})$ 3 225 $\mathrm{I}(3^{-2}5^{+2})$ 1
225 $\mathrm{I}(3^{+2}5^{-2})$ 1 225 $\mathrm{I}(9^{+1}5^{+2})$ 2 225 $\mathrm{I}(3^{-2}5^{-2})$ 2
225 $\mathrm{I}(3^{+2}5^{+2})$ 2 228 $\mathrm{II}(4_{3}^{-1}3^{+1}19^{-1})$ 2 234 $\mathrm{II}(2_{3}^{-1}3^{+2}13^{-1})$ 3
240 $\mathrm{I}(4_{6}^{-2}3^{+1}5^{+1})$ 2 240 $\mathrm{I}(4_{6}^{-2}3^{-1}5^{-1})$ 2 240 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}5^{-1})$ 1
240 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}5^{+1})$ 1 240 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+1}5^{+1})$ 2 240 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{-1}5^{-1})$ 2
240 $\mathrm{I}(4_{6}^{+2}3^{-1}5^{+1})$ 1 240 $\mathrm{II}(16_{5}^{-1}3^{-1}5^{+1})$ 2 245 $\mathrm{I}(5^{-1}7^{+2})$ 1
245 $\mathrm{I}(5^{+1}7^{-2})$ 2 250 $\mathrm{I}(2_{1}^{+1}5^{-1}25^{-1})$ 3 252 $\mathrm{I}(2_{2}^{-2}3^{+2}7^{-1})$ 2
252 $\mathrm{I}(2_{6}^{-2}3^{+2}7^{+1})$ 2 252 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+2}7^{-1})$ 2 252 $\mathrm{I}(2_{\mathrm{II}}^{+2}9^{+1}7^{+1})$ 3
252 $\mathrm{II}(4_{5}^{-1}3^{-2}7^{-1})$ 2 252 $\mathrm{II}(4_{5}^{-1}3^{+2}7^{+1})$ 1 252 $\mathrm{I}(4_{3}^{-1}3^{+2}7^{+1})$ 2
252 $\mathrm{I}(4_{3}^{-1}3^{+2}7^{-1})$ 2 252 $\mathrm{II}(4_{1}^{+1}3^{+2}7^{-1})$ 2 256 $\mathrm{I}(16_{6}^{-2})$ 1
256 $\mathrm{I}(16_{2}^{+2})$ 2 260 $\mathrm{II}(4_{3}^{-1}5^{+1}13^{+1})$ 2 270 $\mathrm{I}(2_{5}^{-1}3^{+1}9^{+1}5^{-1})$ 2
270 $\mathrm{II}(2_{1}^{+1}3^{-1}9^{+1}5^{-1})$ 2 270 $\mathrm{II}(2_{1}^{+1}3^{+1}9^{-1}5^{+1})$ 2 270 $\mathrm{II}(2_{1}^{+1}3^{-1}9^{-1}5^{-1})$ 2
270 $\mathrm{I}(2_{1}^{+1}3^{-1}9^{+1}5^{-1})$ 2 270 $\mathrm{I}(2_{1}^{+1}3^{-1}9^{-1}5^{-1})$ 2 270 $\mathrm{I}(2_{7}^{+1}3^{+1}9^{-1}5^{+1})$ 3
272 $\mathrm{I}(4_{\mathrm{II}}^{-2}17^{+1})$ 2 276 $\mathrm{II}(4_{7}^{+1}3^{-1}23^{+1})$ 3 288 $\mathrm{I}(2_{1}^{+1}16_{3}^{-1}3^{+2})$ 1
288 $\mathrm{I}(2_{7}^{+1}16_{5}^{-1}3^{+2})$ 1 288 $\mathrm{I}(2_{3}^{-1}16_{7}^{+1}3^{-2})$ 2 288 $\mathrm{I}(2_{5}^{-1}16_{1}^{+1}3^{+2})$ 1
288 $\mathrm{I}(2_{1}^{+1}16_{5}^{-1}3^{-2})$ 2 288 $\mathrm{I}(2_{3}^{-1}16_{3}^{-1}3^{-2})$ 2 288 $\mathrm{I}(2_{1}^{+1}16_{1}^{+1}3^{-2})$ 2
288 $\mathrm{I}(2_{1}^{+1}16_{7}^{+1}3^{+2})$ 1 289 $\mathrm{I}(17^{+2})$ 2 294 $\mathrm{I}(2_{7}^{+1}3^{-1}7^{+2})$ 2
294 $\mathrm{II}(2_{5}^{-1}3^{-1}7^{-2})$ 2 294 $\mathrm{I}(2_{5}^{-1}3^{-1}7^{-2})$ 2 297 $\mathrm{I}(3^{-1}9^{-1}11^{+1})$ 3
297 $\mathrm{I}(3^{+1}9^{+1}11^{-1})$ 2 297 $\mathrm{I}(3^{+1}9^{-1}11^{-1})$ 2 300 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+1}5^{+2})$ 1
300 $\mathrm{I}(2_{0}^{+2}3^{-1}5^{-2})$ 2 300 $\mathrm{I}(2_{0}^{+2}3^{+1}5^{+2})$ 2 300 $\mathrm{I}(2_{\mathrm{II}}^{-2}3^{-1}5^{-2})$ 2
300 $\mathrm{I}(2_{4}^{-2}3^{+1}5^{-2})$ 1 300 $\mathrm{II}(4_{1}^{+1}3^{-1}5^{+2})$ 1 300 $\mathrm{II}(4_{1}^{+1}3^{+1}5^{-2})$ 1
300 $\mathrm{I}(4_{7}^{+1}3^{+1}5^{-2})$ 1 300 $\mathrm{I}(4_{7}^{+1}3^{-1}5^{-2})$ 2 300 $\mathrm{I}(4_{7}^{+1}3^{+1}5^{+2})$ 2
300 $\mathrm{II}(4_{5}^{-1}3^{-1}5^{-2})$ 2 300 $\mathrm{II}(4_{5}^{-1}3^{+1}5^{+2})$ 2 308 $\mathrm{II}(4_{7}^{+1}7^{+1}11^{-1})$ 3
315 $\mathrm{I}(3^{+2}5^{+1}7^{+1})$ 2 315 $\mathrm{I}(3^{+2}5^{-1}7^{-1})$ 2 315 $\mathrm{I}(3^{-2}5^{+1}7^{-1})$ 2
315 $\mathrm{I}(3^{+2}5^{+1}7^{-1})$ 3 320 $\mathrm{I}(4_{5}^{-1}16_{1}^{+1}5^{+1})$ 6 324 $\mathrm{I}(2_{2}^{+2}9^{+2})$ 2
324 $\mathrm{II}(4_{3}^{-1}3^{+1}27^{-1})$ 1 324 $\mathrm{II}(4_{3}^{-1}9^{-2})$ 1 324 $\mathrm{II}(4_{3}^{-1}9^{+2})$ 1
324 $\mathrm{I}(4_{1}^{+1}9^{+2})$ 2 325 $\mathrm{I}(5^{-2}13^{-1})$ 2 330 $\mathrm{II}(2_{3}^{-1}3^{-1}5^{-1}11^{-1})$ 4
336 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+1}7^{+1})$ 2 336 $\mathrm{I}(4_{2}^{-2}3^{+1}7^{+1})$ 2 336 $\mathrm{I}(4_{6}^{-2}3^{+1}7^{-1})$ 2
336 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}7^{+1})$ 2 336 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}7^{-1})$ 1 336 $\mathrm{I}(16_{1}^{+1}3^{+1}7^{-1})$ 3
336 $\mathrm{I}(16_{5}^{-1}3^{+1}7^{-1})$ 4 351 $\mathrm{I}(3^{+1}9^{-1}13^{-1})$ 2 360 $\mathrm{I}(2_{3}^{-1}4_{7}^{+1}3^{+2}5^{-1})$ 3
360 $\mathrm{I}(2_{7}^{+1}4_{3}^{-1}3^{+2}5^{+1})$ 2 363 $\mathrm{I}(3^{-1}11^{-2})$ 2 363 $\mathrm{I}(3^{+1}11^{+2})$ 2
375 $\mathrm{I}(3^{-1}5^{-1}25^{+1})$ 3 375 $\mathrm{I}(3^{+1}5^{-1}25^{-1})$ 2 375 $\mathrm{I}(3^{-1}5^{+1}25^{+1})$ 1
375 $\mathrm{I}(3^{-1}5^{+1}25^{-1})$ 1 378 $\mathrm{II}(2_{3}^{-1}3^{-1}9^{-1}7^{+1})$ 4 378 $\mathrm{I}(2_{5}^{-1}3^{-1}9^{-1}7^{-1})$ 3
384 $\mathrm{I}(2_{3}^{-1}64_{1}^{+1}3^{-1})$ 4 392 $\mathrm{I}(2_{7}^{+1}4_{7}^{+1}7^{+2})$ 3 396 $\mathrm{I}(2_{4}^{-2}3^{+2}11^{+1})$ 3
396 $\mathrm{II}(4_{1}^{+1}3^{+2}11^{+1})$ 2 396 $\mathrm{II}(4_{1}^{+1}3^{-2}11^{-1})$ 2 396 $\mathrm{I}(4_{3}^{-1}3^{+2}11^{+1})$ 3
400 $\mathrm{I}(4_{\mathrm{II}}^{-2}5^{+2})$ 1 400 $\mathrm{I}(4_{2}^{-2}5^{-2})$ 1 400 $\mathrm{I}(4_{6}^{-2}5^{+2})$ 2
400 $\mathrm{I}(4_{\mathrm{II}}^{+2}5^{-2})$ 1 400 $\mathrm{I}(16_{5}^{-1}5^{-2})$ 2 400 $\mathrm{I}(16_{5}^{-1}5^{+2})$ 3
400 $\mathrm{I}(16_{1}^{+1}5^{+2})$ 4 405 $\mathrm{I}(9^{+2}5^{+1})$ 2 420 $\mathrm{I}(2_{6}^{+2}3^{-1}5^{-1}7^{+1})$ 3
420 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-1}5^{-1}7^{+1})$ 2 420 $\mathrm{II}(4_{3}^{-1}3^{+1}5^{-1}7^{+1})$ 2 420 $\mathrm{II}(4_{3}^{-1}3^{-1}5^{-1}7^{-1})$ 2
420 $\mathrm{II}(4_{3}^{-1}3^{-1}5^{+1}7^{+1})$ 2 420 $\mathrm{II}(4_{7}^{+1}3^{-1}5^{-1}7^{+1})$ 3 420 $\mathrm{I}(4_{1}^{+1}3^{-1}5^{-1}7^{+1})$ 3
432 $\mathrm{I}(4_{2}^{+2}3^{+1}9^{-1})$ 1 432 $\mathrm{I}(4_{2}^{+2}3^{+1}9^{+1})$ 1 432 $\mathrm{I}(4_{6}^{+2}3^{-1}9^{+1})$ 1
432 $\mathrm{I}(4_{6}^{+2}3^{-1}9^{-1})$ 1 432 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+1}9^{-1})$ 1 432 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+1}9^{+1})$ 1
432 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}9^{+1})$ 1 432 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}9^{-1})$ 1 432 $\mathrm{I}(4_{4}^{-2}3^{+1}9^{+1})$ 2
432 $\mathrm{II}(16_{1}^{+1}3^{-1}9^{+1})$ 2 432 $\mathrm{II}(16_{1}^{+1}3^{-1}9^{-1})$ 2 432 $\mathrm{I}(16_{1}^{+1}3^{+1}9^{+1})$ 4
432 $\mathrm{I}(16_{7}^{+1}3^{+1}9^{+1})$ 2 432 $\mathrm{I}(16_{5}^{-1}3^{+1}9^{+1})$ 4 432 $\mathrm{I}(16_{3}^{-1}3^{+1}9^{-1})$ 2
432 $\mathrm{I}(16_{3}^{-1}3^{+1}9^{+1})$ 2 441 $\mathrm{I}(3^{-2}7^{-2})$ 2 441 $\mathrm{I}(3^{+2}7^{+2})$ 1
441 $\mathrm{I}(3^{+2}7^{-2})$ 2 450 $\mathrm{I}(2_{3}^{-1}3^{+2}5^{-2})$ 1 450 $\mathrm{I}(2_{5}^{-1}3^{-2}5^{-2})$ 2
450 $\mathrm{II}(2_{7}^{+1}3^{+2}5^{+2})$ 1 450 $\mathrm{II}(2_{7}^{+1}3^{-2}5^{-2})$ 2 450 $\mathrm{I}(2_{1}^{+1}3^{+2}5^{+2})$ 2
468 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+2}13^{+1})$ 3 468 $\mathrm{II}(4_{7}^{+1}3^{+2}13^{-1})$ 1 480 $\mathrm{I}(2_{3}^{-1}16_{1}^{+1}3^{+1}5^{-1})$ 2
480 $\mathrm{I}(2_{5}^{-1}16_{7}^{+1}3^{+1}5^{-1})$ 2 480 $\mathrm{I}(2_{7}^{+1}16_{5}^{-1}3^{+1}5^{+1})$ 3 480 $\mathrm{I}(2_{3}^{-1}16_{5}^{-1}3^{+1}5^{-1})$ 2
480 $\mathrm{I}(2_{5}^{-1}16_{3}^{-1}3^{+1}5^{-1})$ 2 480 $\mathrm{I}(2_{1}^{+1}16_{7}^{+1}3^{+1}5^{+1})$ 3 480 $\mathrm{I}(2_{7}^{+1}16_{1}^{+1}3^{+1}5^{+1})$ 3
480 $\mathrm{I}(2_{3}^{-1}16_{3}^{-1}3^{+1}5^{+1})$ 3 484 $\mathrm{II}(4_{3}^{-1}11^{-2})$ 1 495 $\mathrm{I}(3^{+2}5^{-1}11^{-1})$ 2
500 $\mathrm{I}(2_{2}^{-2}5^{+1}25^{+1})$ 3 500 $\mathrm{II}(4_{7}^{+1}5^{+1}25^{-1})$ 1 500 $\mathrm{II}(4_{7}^{+1}5^{+1}25^{+1})$ 1
500 $\mathrm{I}(4_{1}^{+1}5^{+1}25^{+1})$ 3 507 $\mathrm{I}(3^{+1}13^{-2})$ 1 513 $\mathrm{I}(3^{+1}9^{-1}19^{-1})$ 3
525 $\mathrm{I}(3^{+1}5^{-2}7^{-1})$ 3 525 $\mathrm{I}(3^{-1}5^{-2}7^{-1})$ 2 525 $\mathrm{I}(3^{+1}5^{-2}7^{+1})$ 2
525 $\mathrm{I}(3^{+1}5^{+2}7^{-1})$ 2 528 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}11^{-1})$ 2 528 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}11^{+1})$ 2
528 $\mathrm{I}(4_{2}^{+2}3^{-1}11^{+1})$ 3 539 $\mathrm{I}(7^{+2}11^{+1})$ 3 540 $\mathrm{I}(2_{2}^{-2}3^{+1}9^{+1}5^{+1})$ 2
540 $\mathrm{I}(2_{2}^{-2}3^{+1}9^{-1}5^{+1})$ 2 540 $\mathrm{I}(2_{4}^{-2}3^{-1}9^{-1}5^{+1})$ 2 540 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+1}9^{+1}5^{+1})$ 2
540 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-1}9^{+1}5^{-1})$ 2 540 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+1}9^{-1}5^{+1})$ 2 540 $\mathrm{I}(2_{2}^{+2}3^{-1}9^{+1}5^{-1})$ 3
540 $\mathrm{II}(4_{5}^{-1}3^{-1}9^{-1}5^{+1})$ 2 540 $\mathrm{II}(4_{5}^{-1}3^{+1}9^{+1}5^{-1})$ 1 540 $\mathrm{II}(4_{5}^{-1}3^{+1}9^{-1}5^{-1})$ 1
540 $\mathrm{I}(4_{3}^{-1}3^{+1}9^{+1}5^{+1})$ 2 540 $\mathrm{I}(4_{3}^{-1}3^{-1}9^{+1}5^{-1})$ 3 540 $\mathrm{I}(4_{3}^{-1}3^{+1}9^{-1}5^{+1})$ 2
540 $\mathrm{II}(4_{1}^{+1}3^{+1}9^{+1}5^{+1})$ 2 540 $\mathrm{II}(4_{1}^{+1}3^{-1}9^{+1}5^{-1})$ 3 540 $\mathrm{II}(4_{1}^{+1}3^{+1}9^{-1}5^{+1})$ 2
540 $\mathrm{I}(4_{7}^{+1}3^{-1}9^{-1}5^{+1})$ 2 560 $\mathrm{I}(4_{\mathrm{II}}^{-2}5^{+1}7^{-1})$ 1 560 $\mathrm{I}(4_{\mathrm{II}}^{-2}5^{-1}7^{+1})$ 2
576 $\mathrm{I}(4_{3}^{-1}16_{1}^{+1}3^{+2})$ 2 576 $\mathrm{I}(4_{1}^{+1}16_{5}^{-1}9^{+1})$ 6 576 $\mathrm{I}(4_{3}^{-1}16_{3}^{-1}3^{+2})$ 2
576 $\mathrm{I}(4_{1}^{+1}16_{7}^{+1}3^{+2})$ 2 585 $\mathrm{I}(3^{+2}5^{+1}13^{-1})$ 3 588 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+1}7^{-2})$ 2
588 $\mathrm{I}(2_{2}^{-2}3^{+1}7^{-2})$ 2 588 $\mathrm{I}(2_{6}^{-2}3^{+1}7^{+2})$ 2 588 $\mathrm{II}(4_{1}^{+1}3^{-1}7^{-2})$ 2
588 $\mathrm{II}(4_{1}^{+1}3^{+1}7^{+2})$ 1 588 $\mathrm{I}(4_{7}^{+1}3^{+1}7^{+2})$ 2 588 $\mathrm{I}(4_{7}^{+1}3^{+1}7^{-2})$ 2
588 $\mathrm{II}(4_{5}^{-1}3^{+1}7^{-2})$ 2 600 $\mathrm{I}(2_{5}^{-1}4_{7}^{+1}3^{-1}5^{+2})$ 3 600 $\mathrm{I}(2_{1}^{+1}4_{5}^{-1}3^{-1}5^{-2})$ 2
612 $\mathrm{II}(4_{3}^{-1}3^{+2}17^{-1})$ 2 624 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}13^{-1})$ 1 630 $\mathrm{II}(2_{5}^{-1}3^{+2}5^{-1}7^{-1})$ 2
630 $\mathrm{I}(2_{5}^{-1}3^{+2}5^{-1}7^{-1})$ 3 660 $\mathrm{I}(2_{\mathrm{II}}^{-2}3^{+1}5^{+1}11^{+1})$ 4 660 $\mathrm{II}(4_{7}^{+1}3^{+1}5^{+1}11^{-1})$ 2
675 $\mathrm{I}(3^{-1}9^{+1}5^{+2})$ 1 675 $\mathrm{I}(3^{-1}9^{-1}5^{+2})$ 1 675 $\mathrm{I}(3^{+1}9^{+1}5^{-2})$ 2
675 $\mathrm{I}(3^{-1}9^{+1}5^{-2})$ 2 675 $\mathrm{I}(3^{+1}9^{-1}5^{+2})$ 3 675 $\mathrm{I}(3^{-1}9^{-1}5^{-2})$ 2
684 $\mathrm{II}(4_{1}^{+1}3^{+2}19^{+1})$ 2 700 $\mathrm{II}(4_{5}^{-1}5^{+2}7^{-1})$ 2 700 $\mathrm{II}(4_{5}^{-1}5^{-2}7^{+1})$ 1
702 $\mathrm{II}(2_{1}^{+1}3^{-1}9^{-1}13^{-1})$ 6 720 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+2}5^{+1})$ 2 720 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{-2}5^{-1})$ 2
720 $\mathrm{I}(4_{6}^{+2}3^{+2}5^{+1})$ 2 720 $\mathrm{I}(4_{2}^{-2}3^{-2}5^{-1})$ 2 720 $\mathrm{I}(4_{6}^{-2}3^{+2}5^{-1})$ 1
720 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}5^{-1})$ 1 720 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-2}5^{+1})$ 1 720 $\mathrm{I}(4_{\mathrm{II}}^{-2}9^{-1}5^{+1})$ 2
720 $\mathrm{II}(16_{7}^{+1}3^{+2}5^{-1})$ 2 726 $\mathrm{I}(2_{1}^{+1}3^{-1}11^{+2})$ 3 735 $\mathrm{I}(3^{-1}5^{+1}7^{+2})$ 3
735 $\mathrm{I}(3^{+1}5^{+1}7^{+2})$ 2 735 $\mathrm{I}(3^{-1}5^{+1}7^{-2})$ 2 735 $\mathrm{I}(3^{-1}5^{-1}7^{+2})$ 2
750 $\mathrm{II}(2_{1}^{+1}3^{+1}5^{+1}25^{-1})$ 2 750 $\mathrm{I}(2_{7}^{+1}3^{+1}5^{+1}25^{-1})$ 3 756 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+1}9^{+1}7^{+1})$ 4
756 $\mathrm{I}(2_{2}^{+2}3^{+1}9^{+1}7^{-1})$ 3 756 $\mathrm{II}(4_{7}^{+1}3^{-1}9^{+1}7^{+1})$ 2 756 $\mathrm{II}(4_{7}^{+1}3^{+1}9^{-1}7^{-1})$ 1
756 $\mathrm{II}(4_{7}^{+1}3^{-1}9^{-1}7^{+1})$ 2 756 $\mathrm{II}(4_{7}^{+1}3^{+1}9^{+1}7^{-1})$ 1 756 $\mathrm{I}(4_{1}^{+1}3^{+1}9^{+1}7^{-1})$ 3
768 $\mathrm{I}(16_{2}^{+2}3^{+1})$ 2 768 $\mathrm{I}(16_{0}^{+2}3^{+1})$ 2 768 $\mathrm{I}(16_{\mathrm{II}}^{-2}3^{-1})$ 2
768 $\mathrm{I}(16_{4}^{-2}3^{+1})$ 2 768 $\mathrm{I}(16_{6}^{-2}3^{+1})$ 1 780 $\mathrm{II}(4_{1}^{+1}3^{-1}5^{-1}13^{-1})$ 3
784 $\mathrm{I}(4_{6}^{+2}7^{+2})$ 2 784 $\mathrm{I}(4_{\mathrm{II}}^{+2}7^{+2})$ 2 800 $\mathrm{I}(2_{5}^{-1}16_{1}^{+1}5^{-2})$ 2
800 $\mathrm{I}(2_{7}^{+1}16_{3}^{-1}5^{-2})$ 2 800 $\mathrm{I}(2_{5}^{-1}16_{5}^{-1}5^{-2})$ 2 800 $\mathrm{I}(2_{7}^{+1}16_{7}^{+1}5^{-2})$ 2
816 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}17^{+1})$ 2 825 $\mathrm{I}(3^{-1}5^{-2}11^{-1})$ 2 828 $\mathrm{II}(4_{5}^{-1}3^{+2}23^{+1})$ 3
845 $\mathrm{I}(5^{-1}13^{-2})$ 2 847 $\mathrm{I}(7^{+1}11^{+2})$ 3 864 $\mathrm{I}(2_{3}^{-1}16_{3}^{-1}3^{+1}9^{-1})$ 2
864 $\mathrm{I}(2_{1}^{+1}16_{1}^{+1}3^{+1}9^{+1})$ 2 864 $\mathrm{I}(2_{1}^{+1}16_{1}^{+1}3^{+1}9^{-1})$ 2 864 $\mathrm{I}(2_{3}^{-1}16_{5}^{-1}3^{-1}9^{+1})$ 2
864 $\mathrm{I}(2_{5}^{-1}16_{3}^{-1}3^{-1}9^{+1})$ 2 864 $\mathrm{I}(2_{1}^{+1}16_{7}^{+1}3^{+1}9^{+1})$ 2 864 $\mathrm{I}(2_{1}^{+1}16_{7}^{+1}3^{+1}9^{-1})$ 2
864 $\mathrm{I}(2_{3}^{-1}16_{1}^{+1}3^{-1}9^{+1})$ 2 864 $\mathrm{I}(2_{5}^{-1}16_{7}^{+1}3^{-1}9^{+1})$ 2 864 $\mathrm{I}(2_{1}^{+1}16_{3}^{-1}3^{+1}9^{+1})$ 2
864 $\mathrm{I}(2_{7}^{+1}16_{5}^{-1}3^{+1}9^{+1})$ 2 864 $\mathrm{I}(2_{1}^{+1}16_{5}^{-1}3^{+1}9^{-1})$ 2 867 $\mathrm{I}(3^{+1}17^{-2})$ 2
882 $\mathrm{II}(2_{3}^{-1}3^{+2}7^{-2})$ 2 882 $\mathrm{I}(2_{5}^{-1}3^{+2}7^{+2})$ 2 882 $\mathrm{II}(2_{7}^{+1}9^{-1}7^{+2})$ 3
882 $\mathrm{I}(2_{7}^{+1}3^{+2}7^{-2})$ 2 900 $\mathrm{I}(2_{\mathrm{II}}^{-2}3^{-2}5^{-2})$ 2 900 $\mathrm{I}(2_{4}^{-2}3^{+2}5^{-2})$ 1
900 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+2}5^{+2})$ 1 900 $\mathrm{I}(2_{0}^{+2}3^{-2}5^{-2})$ 2 900 $\mathrm{I}(2_{0}^{+2}3^{+2}5^{+2})$ 2
900 $\mathrm{II}(4_{3}^{-1}3^{-2}5^{+2})$ 1 900 $\mathrm{II}(4_{3}^{-1}3^{+2}5^{-2})$ 1 900 $\mathrm{II}(4_{3}^{-1}9^{+1}5^{+2})$ 2
900 $\mathrm{I}(4_{5}^{-1}3^{+2}5^{+2})$ 2 900 $\mathrm{I}(4_{5}^{-1}3^{+2}5^{-2})$ 1 900 $\mathrm{II}(4_{7}^{+1}3^{-2}5^{-2})$ 2
900 $\mathrm{II}(4_{7}^{+1}3^{+2}5^{+2})$ 2 900 $\mathrm{I}(4_{1}^{+1}3^{-2}5^{-2})$ 2 912 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}19^{-1})$ 2
945 $\mathrm{I}(3^{-1}9^{-1}5^{+1}7^{+1})$ 3 945 $\mathrm{I}(3^{+1}9^{+1}5^{-1}7^{+1})$ 2 945 $\mathrm{I}(3^{+1}9^{-1}5^{-1}7^{+1})$ 2
975 $\mathrm{I}(3^{+1}5^{-2}13^{+1})$ 3 980 $\mathrm{II}(4_{7}^{+1}5^{-1}7^{+2})$ 1 980 $\mathrm{II}(4_{7}^{+1}5^{+1}7^{-2})$ 2
990 $\mathrm{II}(2_{1}^{+1}3^{+2}5^{+1}11^{-1})$ 4 1008 $\mathrm{I}(4_{2}^{-2}3^{+2}7^{+1})$ 2 1008 $\mathrm{I}(4_{6}^{-2}3^{+2}7^{-1})$ 2
1008 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-2}7^{-1})$ 2 1008 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}7^{+1})$ 1 1008 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+2}7^{-1})$ 2
1008 $\mathrm{I}(16_{3}^{-1}3^{+2}7^{+1})$ 3 1008 $\mathrm{I}(16_{7}^{+1}3^{+2}7^{+1})$ 4 1014 $\mathrm{II}(2_{5}^{-1}3^{-1}13^{+2})$ 3
1029 $\mathrm{I}(3^{+1}7^{-1}49^{+1})$ 2 1040 $\mathrm{I}(4_{\mathrm{II}}^{-2}5^{+1}13^{+1})$ 2 1050 $\mathrm{II}(2_{3}^{-1}3^{-1}5^{-2}7^{-1})$ 2
1050 $\mathrm{I}(2_{3}^{-1}3^{-1}5^{-2}7^{-1})$ 3 1080 $\mathrm{I}(2_{5}^{-1}4_{1}^{+1}3^{+1}9^{+1}5^{-1})$ 4 1083 $\mathrm{I}(3^{+1}19^{+2})$ 2
1089 $\mathrm{I}(3^{-2}11^{-2})$ 2 1089 $\mathrm{I}(3^{+2}11^{+2})$ 2 1104 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}23^{+1})$ 3
1125 $\mathrm{I}(3^{+2}5^{+1}25^{-1})$ 3 1125 $\mathrm{I}(3^{+2}5^{-1}25^{+1})$ 1 1125 $\mathrm{I}(3^{+2}5^{-1}25^{-1})$ 1
1125 $\mathrm{I}(3^{-2}5^{+1}25^{+1})$ 2 1125 $\mathrm{I}(9^{-1}5^{+1}25^{-1})$ 3 1134 $\mathrm{II}(2_{1}^{+1}9^{+2}7^{+1})$ 3
1152 $\mathrm{I}(2_{7}^{+1}64_{3}^{-1}3^{+2})$ 4 1156 $\mathrm{II}(4_{3}^{-1}17^{+2})$ 2 1188 $\mathrm{II}(4_{3}^{-1}3^{-1}9^{-1}11^{+1})$ 3
1188 $\mathrm{II}(4_{3}^{-1}3^{+1}9^{+1}11^{-1})$ 2 1188 $\mathrm{II}(4_{3}^{-1}3^{+1}9^{-1}11^{-1})$ 2 1200 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{-1}5^{-2})$ 2
1200 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+1}5^{+2})$ 2 1200 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}5^{+2})$ 1 1200 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}5^{-2})$ 1
1200 $\mathrm{I}(4_{2}^{-2}3^{+1}5^{-2})$ 1 1200 $\mathrm{I}(4_{6}^{-2}3^{-1}5^{-2})$ 2 1200 $\mathrm{I}(4_{6}^{-2}3^{+1}5^{+2})$ 2
1200 $\mathrm{II}(16_{1}^{+1}3^{+1}5^{-2})$ 2 1225 $\mathrm{I}(5^{-2}7^{+2})$ 1 1225 $\mathrm{I}(5^{+2}7^{-2})$ 2
1232 $\mathrm{I}(4_{\mathrm{II}}^{-2}7^{+1}11^{-1})$ 3 1260 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+2}5^{+1}7^{-1})$ 2 1260 $\mathrm{I}(2_{0}^{+2}3^{+2}5^{+1}7^{-1})$ 3
1260 $\mathrm{II}(4_{1}^{+1}3^{+2}5^{+1}7^{+1})$ 2 1260 $\mathrm{II}(4_{1}^{+1}3^{+2}5^{-1}7^{-1})$ 2 1260 $\mathrm{II}(4_{1}^{+1}3^{-2}5^{+1}7^{-1})$ 2
1260 $\mathrm{II}(4_{5}^{-1}3^{+2}5^{+1}7^{-1})$ 3 1260 $\mathrm{I}(4_{3}^{-1}3^{+2}5^{+1}7^{-1})$ 3 1280 $\mathrm{I}(16_{2}^{+2}5^{+1})$ 4
1280 $\mathrm{I}(16_{2}^{-2}5^{-1})$ 2 1280 $\mathrm{I}(16_{6}^{-2}5^{+1})$ 3 1296 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}27^{-1})$ 1
1296 $\mathrm{I}(4_{\mathrm{II}}^{-2}9^{-2})$ 1 1296 $\mathrm{I}(4_{\mathrm{II}}^{-2}9^{+2})$ 1 1296 $\mathrm{I}(4_{2}^{+2}9^{+2})$ 2
1296 $\mathrm{I}(16_{5}^{-1}9^{+2})$ 3 1296 $\mathrm{I}(16_{1}^{+1}9^{+2})$ 4 1300 $\mathrm{II}(4_{7}^{+1}5^{-2}13^{-1})$ 2
1323 $\mathrm{I}(3^{-1}9^{+1}7^{-2})$ 2 1323 $\mathrm{I}(3^{+1}9^{-1}7^{+2})$ 1 1323 $\mathrm{I}(3^{-1}9^{-1}7^{-2})$ 2
1323 $\mathrm{I}(3^{+1}9^{+1}7^{+2})$ 1 1344 $\mathrm{I}(4_{1}^{+1}16_{5}^{-1}3^{+1}7^{-1})$ 6 1350 $\mathrm{II}(2_{1}^{+1}3^{+1}9^{+1}5^{-2})$ 2
1350 $\mathrm{II}(2_{1}^{+1}3^{+1}9^{-1}5^{-2})$ 2 1350 $\mathrm{I}(2_{7}^{+1}3^{+1}9^{+1}5^{-2})$ 2 1350 $\mathrm{I}(2_{7}^{+1}3^{+1}9^{-1}5^{-2})$ 2
1350 $\mathrm{I}(2_{1}^{+1}3^{-1}9^{-1}5^{-2})$ 2 1350 $\mathrm{II}(2_{5}^{-1}3^{-1}9^{+1}5^{+2})$ 2 1350 $\mathrm{I}(2_{5}^{-1}3^{-1}9^{+1}5^{+2})$ 3
1404 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+1}9^{+1}13^{+1})$ 6 1404 $\mathrm{II}(4_{5}^{-1}3^{+1}9^{-1}13^{-1})$ 2 1440 $\mathrm{I}(2_{3}^{-1}16_{3}^{-1}3^{+2}5^{-1})$ 3
1440 $\mathrm{I}(2_{1}^{+1}16_{1}^{+1}3^{+2}5^{-1})$ 3 1440 $\mathrm{I}(2_{7}^{+1}16_{7}^{+1}3^{+2}5^{+1})$ 2 1440 $\mathrm{I}(2_{3}^{-1}16_{7}^{+1}3^{+2}5^{-1})$ 3
1440 $\mathrm{I}(2_{5}^{-1}16_{1}^{+1}3^{+2}5^{+1})$ 2 1440 $\mathrm{I}(2_{1}^{+1}16_{5}^{-1}3^{+2}5^{-1})$ 3 1440 $\mathrm{I}(2_{7}^{+1}16_{3}^{-1}3^{+2}5^{+1})$ 2
1440 $\mathrm{I}(2_{7}^{+1}16_{5}^{-1}3^{+2}5^{+1})$ 2 1452 $\mathrm{I}(2_{4}^{-2}3^{+1}11^{+2})$ 3 1452 $\mathrm{II}(4_{1}^{+1}3^{-1}11^{-2})$ 2
1452 $\mathrm{II}(4_{1}^{+1}3^{+1}11^{+2})$ 2 1452 $\mathrm{I}(4_{3}^{-1}3^{+1}11^{+2})$ 3 1470 $\mathrm{II}(2_{1}^{+1}3^{+1}5^{-1}7^{+2})$ 2
1470 $\mathrm{I}(2_{1}^{+1}3^{+1}5^{-1}7^{+2})$ 3 1500 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-1}5^{-1}25^{+1})$ 2 1500 $\mathrm{I}(2_{2}^{+2}3^{-1}5^{-1}25^{+1})$ 3
1500 $\mathrm{II}(4_{5}^{-1}3^{+1}5^{-1}25^{-1})$ 2 1500 $\mathrm{II}(4_{5}^{-1}3^{-1}5^{+1}25^{+1})$ 1 1500 $\mathrm{II}(4_{5}^{-1}3^{-1}5^{+1}25^{-1})$ 1
1500 $\mathrm{II}(4_{1}^{+1}3^{-1}5^{-1}25^{+1})$ 3 1500 $\mathrm{I}(4_{7}^{+1}3^{-1}5^{-1}25^{+1})$ 3 1521 $\mathrm{I}(3^{+2}13^{-2})$ 1
1568 $\mathrm{I}(2_{5}^{-1}16_{1}^{+1}7^{+2})$ 3 1568 $\mathrm{I}(2_{7}^{+1}16_{3}^{-1}7^{+2})$ 3 1568 $\mathrm{I}(2_{5}^{-1}16_{5}^{-1}7^{+2})$ 3
1568 $\mathrm{I}(2_{7}^{+1}16_{7}^{+1}7^{+2})$ 3 1575 $\mathrm{I}(3^{+2}5^{-2}7^{+1})$ 3 1575 $\mathrm{I}(3^{-2}5^{-2}7^{+1})$ 2
1575 $\mathrm{I}(3^{+2}5^{+2}7^{+1})$ 2 1575 $\mathrm{I}(3^{+2}5^{-2}7^{-1})$ 2 1584 $\mathrm{I}(4_{6}^{+2}3^{+2}11^{+1})$ 3
1584 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}11^{+1})$ 2 1584 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-2}11^{-1})$ 2 1587 $\mathrm{I}(3^{+1}23^{+2})$ 3
1600 $\mathrm{I}(4_{1}^{+1}16_{5}^{-1}5^{+2})$ 6 1620 $\mathrm{II}(4_{7}^{+1}9^{+2}5^{+1})$ 2 1650 $\mathrm{II}(2_{7}^{+1}3^{+1}5^{-2}11^{-1})$ 4
1680 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}5^{-1}7^{+1})$ 2 1680 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}5^{-1}7^{-1})$ 2 1680 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}5^{+1}7^{+1})$ 2
1680 $\mathrm{I}(4_{2}^{-2}3^{-1}5^{-1}7^{+1})$ 3 1680 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{-1}5^{-1}7^{+1})$ 3 1728 $\mathrm{I}(4_{3}^{-1}16_{1}^{+1}3^{+1}9^{+1})$ 4
1728 $\mathrm{I}(4_{1}^{+1}16_{3}^{-1}3^{+1}9^{+1})$ 4 1728 $\mathrm{I}(4_{1}^{+1}16_{5}^{-1}3^{+1}9^{+1})$ 4 1764 $\mathrm{I}(2_{2}^{-2}3^{+2}7^{+2})$ 2
1764 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+2}7^{-2})$ 2 1764 $\mathrm{I}(2_{\mathrm{II}}^{+2}9^{+1}7^{+2})$ 3 1764 $\mathrm{I}(2_{0}^{+2}3^{+2}7^{-2})$ 2
1764 $\mathrm{II}(4_{3}^{-1}3^{-2}7^{-2})$ 2 1764 $\mathrm{II}(4_{3}^{-1}3^{+2}7^{+2})$ 1 1764 $\mathrm{I}(4_{5}^{-1}3^{+2}7^{-2})$ 2
1764 $\mathrm{I}(4_{5}^{-1}3^{+2}7^{+2})$ 2 1764 $\mathrm{II}(4_{7}^{+1}3^{+2}7^{-2})$ 2 1800 $\mathrm{I}(2_{3}^{-1}4_{5}^{-1}3^{+2}5^{-2})$ 2
1800 $\mathrm{I}(2_{1}^{+1}4_{7}^{+1}3^{+2}5^{+2})$ 3 1815 $\mathrm{I}(3^{-1}5^{+1}11^{-2})$ 2 1872 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}13^{-1})$ 1
1890 $\mathrm{II}(2_{7}^{+1}3^{+1}9^{-1}5^{+1}7^{+1})$ 4 1936 $\mathrm{I}(4_{\mathrm{II}}^{-2}11^{-2})$ 1 1980 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+2}5^{-1}11^{+1})$ 4
1980 $\mathrm{II}(4_{5}^{-1}3^{+2}5^{-1}11^{-1})$ 2 2000 $\mathrm{I}(4_{6}^{-2}5^{+1}25^{+1})$ 3 2000 $\mathrm{I}(4_{\mathrm{II}}^{-2}5^{+1}25^{-1})$ 1
2000 $\mathrm{I}(4_{\mathrm{II}}^{-2}5^{+1}25^{+1})$ 1 2000 $\mathrm{I}(16_{1}^{+1}5^{+1}25^{+1})$ 6 2000 $\mathrm{I}(16_{5}^{-1}5^{+1}25^{+1})$ 6
2025 $\mathrm{I}(9^{+2}5^{+2})$ 2 2028 $\mathrm{I}(2_{\mathrm{II}}^{-2}3^{+1}13^{+2})$ 3 2028 $\mathrm{II}(4_{1}^{+1}3^{+1}13^{-2})$ 1
2052 $\mathrm{II}(4_{3}^{-1}3^{+1}9^{-1}19^{-1})$ 3 2100 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+1}5^{-2}7^{-1})$ 2 2100 $\mathrm{I}(2_{0}^{+2}3^{+1}5^{-2}7^{-1})$ 3
2100 $\mathrm{II}(4_{7}^{+1}3^{-1}5^{-2}7^{-1})$ 2 2100 $\mathrm{II}(4_{7}^{+1}3^{+1}5^{-2}7^{+1})$ 2 2100 $\mathrm{II}(4_{7}^{+1}3^{+1}5^{+2}7^{-1})$ 2
2100 $\mathrm{II}(4_{3}^{-1}3^{+1}5^{-2}7^{-1})$ 3 2100 $\mathrm{I}(4_{5}^{-1}3^{+1}5^{-2}7^{-1})$ 3 2156 $\mathrm{II}(4_{1}^{+1}7^{+2}11^{+1})$ 3
2160 $\mathrm{I}(4_{6}^{-2}3^{+1}9^{+1}5^{+1})$ 2 2160 $\mathrm{I}(4_{6}^{-2}3^{+1}9^{-1}5^{+1})$ 2 2160 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}9^{-1}5^{+1})$ 2
2160 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{+1}5^{-1})$ 1 2160 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{-1}5^{-1})$ 1 2160 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+1}9^{+1}5^{+1})$ 2
2160 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{-1}9^{+1}5^{-1})$ 3 2160 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+1}9^{-1}5^{+1})$ 2 2160 $\mathrm{I}(4_{2}^{+2}3^{-1}9^{+1}5^{-1})$ 3
2160 $\mathrm{I}(4_{6}^{+2}3^{-1}9^{-1}5^{+1})$ 2 2160 $\mathrm{II}(16_{5}^{-1}3^{-1}9^{-1}5^{+1})$ 4 2178 $\mathrm{I}(2_{1}^{+1}3^{+2}11^{+2})$ 3
2205 $\mathrm{I}(3^{+2}5^{-1}7^{+2})$ 3 2205 $\mathrm{I}(3^{+2}5^{-1}7^{-2})$ 2 2205 $\mathrm{I}(3^{+2}5^{+1}7^{+2})$ 2
2205 $\mathrm{I}(3^{-2}5^{-1}7^{+2})$ 2 2250 $\mathrm{II}(2_{3}^{-1}3^{+2}5^{-1}25^{+1})$ 2 2250 $\mathrm{I}(2_{3}^{-1}3^{+2}5^{-1}25^{+1})$ 3
2268 $\mathrm{I}(2_{\mathrm{II}}^{+2}9^{+2}7^{+1})$ 3 2304 $\mathrm{I}(16_{\mathrm{II}}^{-2}3^{-2})$ 2 2304 $\mathrm{I}(16_{4}^{-2}3^{+2})$ 2
2304 $\mathrm{I}(16_{2}^{-2}3^{+2})$ 1 2304 $\mathrm{I}(16_{6}^{-2}9^{+1})$ 3 2304 $\mathrm{I}(16_{2}^{+2}9^{+1})$ 4
2304 $\mathrm{I}(16_{6}^{+2}3^{+2})$ 2 2304 $\mathrm{I}(16_{0}^{+2}3^{+2})$ 2 2340 $\mathrm{II}(4_{3}^{-1}3^{+2}5^{+1}13^{-1})$ 3
2352 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+1}7^{-2})$ 2 2352 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}7^{-2})$ 2 2352 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}7^{+2})$ 1
2352 $\mathrm{I}(4_{2}^{-2}3^{+1}7^{+2})$ 2 2352 $\mathrm{I}(4_{6}^{-2}3^{+1}7^{-2})$ 2 2352 $\mathrm{I}(16_{7}^{+1}3^{+1}7^{+2})$ 3
2352 $\mathrm{I}(16_{3}^{-1}3^{+1}7^{+2})$ 4 2400 $\mathrm{I}(2_{1}^{+1}16_{1}^{+1}3^{-1}5^{-2})$ 2 2400 $\mathrm{I}(2_{3}^{-1}16_{5}^{-1}3^{-1}5^{+2})$ 3
2400 $\mathrm{I}(2_{5}^{-1}16_{3}^{-1}3^{-1}5^{+2})$ 3 2400 $\mathrm{I}(2_{3}^{-1}16_{1}^{+1}3^{-1}5^{+2})$ 3 2400 $\mathrm{I}(2_{5}^{-1}16_{7}^{+1}3^{-1}5^{+2})$ 3
2400 $\mathrm{I}(2_{1}^{+1}16_{3}^{-1}3^{-1}5^{-2})$ 2 2400 $\mathrm{I}(2_{3}^{-1}16_{7}^{+1}3^{-1}5^{-2})$ 2 2400 $\mathrm{I}(2_{1}^{+1}16_{5}^{-1}3^{-1}5^{-2})$ 2
2448 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}17^{-1})$ 2 2475 $\mathrm{I}(3^{+2}5^{-2}11^{-1})$ 2 2535 $\mathrm{I}(3^{-1}5^{+1}13^{+2})$ 3
2601 $\mathrm{I}(3^{+2}17^{-2})$ 2 2625 $\mathrm{I}(3^{-1}5^{+1}25^{+1}7^{+1})$ 3 2640 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}5^{+1}11^{-1})$ 2
2646 $\mathrm{I}(2_{7}^{+1}3^{-1}9^{-1}7^{+2})$ 3 2646 $\mathrm{II}(2_{5}^{-1}3^{-1}9^{-1}7^{-2})$ 4 2700 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+1}9^{-1}5^{+2})$ 2
2700 $\mathrm{I}(2_{0}^{+2}3^{-1}9^{+1}5^{-2})$ 2 2700 $\mathrm{I}(2_{0}^{+2}3^{+1}9^{-1}5^{+2})$ 3 2700 $\mathrm{I}(2_{0}^{+2}3^{-1}9^{-1}5^{-2})$ 2
2700 $\mathrm{I}(2_{\mathrm{II}}^{-2}3^{-1}9^{+1}5^{-2})$ 2 2700 $\mathrm{I}(2_{\mathrm{II}}^{-2}3^{-1}9^{-1}5^{-2})$ 2 2700 $\mathrm{I}(2_{4}^{-2}3^{+1}9^{+1}5^{-2})$ 2
2700 $\mathrm{II}(4_{1}^{+1}3^{-1}9^{+1}5^{+2})$ 1 2700 $\mathrm{II}(4_{1}^{+1}3^{-1}9^{-1}5^{+2})$ 1 2700 $\mathrm{II}(4_{1}^{+1}3^{+1}9^{+1}5^{-2})$ 2
2700 $\mathrm{I}(4_{7}^{+1}3^{+1}9^{+1}5^{-2})$ 2 2700 $\mathrm{I}(4_{7}^{+1}3^{-1}9^{+1}5^{-2})$ 2 2700 $\mathrm{I}(4_{7}^{+1}3^{+1}9^{-1}5^{+2})$ 3
2700 $\mathrm{I}(4_{7}^{+1}3^{-1}9^{-1}5^{-2})$ 2 2700 $\mathrm{II}(4_{5}^{-1}3^{-1}9^{+1}5^{-2})$ 2 2700 $\mathrm{II}(4_{5}^{-1}3^{+1}9^{-1}5^{+2})$ 3
2700 $\mathrm{II}(4_{5}^{-1}3^{-1}9^{-1}5^{-2})$ 2 2736 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}19^{+1})$ 2 2800 $\mathrm{I}(4_{\mathrm{II}}^{-2}5^{+2}7^{-1})$ 2
2800 $\mathrm{I}(4_{\mathrm{II}}^{-2}5^{-2}7^{+1})$ 1 2925 $\mathrm{I}(3^{+2}5^{-2}13^{+1})$ 3 2940 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-1}5^{+1}7^{+2})$ 2
2940 $\mathrm{I}(2_{0}^{+2}3^{-1}5^{+1}7^{+2})$ 3 2940 $\mathrm{II}(4_{5}^{-1}3^{+1}5^{+1}7^{+2})$ 2 2940 $\mathrm{II}(4_{5}^{-1}3^{-1}5^{+1}7^{-2})$ 2
2940 $\mathrm{II}(4_{5}^{-1}3^{-1}5^{-1}7^{+2})$ 2 2940 $\mathrm{I}(4_{3}^{-1}3^{-1}5^{+1}7^{+2})$ 3 2940 $\mathrm{II}(4_{1}^{+1}3^{-1}5^{+1}7^{+2})$ 3
3024 $\mathrm{I}(4_{6}^{-2}3^{+1}9^{+1}7^{-1})$ 3 3024 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}9^{+1}7^{+1})$ 2 3024 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{-1}7^{-1})$ 1
3024 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}9^{-1}7^{+1})$ 2 3024 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{+1}7^{-1})$ 1 3024 $\mathrm{I}(16_{1}^{+1}3^{+1}9^{+1}7^{-1})$ 6
3024 $\mathrm{I}(16_{5}^{-1}3^{+1}9^{+1}7^{-1})$ 6 3042 $\mathrm{II}(2_{7}^{+1}3^{+2}13^{+2})$ 3 3087 $\mathrm{I}(3^{+2}7^{+1}49^{-1})$ 2
3120 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}5^{-1}13^{-1})$ 3 3150 $\mathrm{II}(2_{1}^{+1}3^{+2}5^{-2}7^{+1})$ 2 3150 $\mathrm{I}(2_{1}^{+1}3^{+2}5^{-2}7^{+1})$ 3
3249 $\mathrm{I}(3^{+2}19^{+2})$ 2 3267 $\mathrm{I}(3^{-1}9^{+1}11^{-2})$ 2 3267 $\mathrm{I}(3^{-1}9^{-1}11^{-2})$ 2
3267 $\mathrm{I}(3^{+1}9^{+1}11^{+2})$ 3 3300 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-1}5^{-2}11^{+1})$ 4 3300 $\mathrm{II}(4_{3}^{-1}3^{-1}5^{-2}11^{-1})$ 2
3312 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}23^{+1})$ 3 3375 $\mathrm{I}(3^{-1}9^{+1}5^{-1}25^{+1})$ 6 3375 $\mathrm{I}(3^{-1}9^{-1}5^{+1}25^{+1})$ 2
3375 $\mathrm{I}(3^{-1}9^{-1}5^{+1}25^{-1})$ 2 3375 $\mathrm{I}(3^{+1}9^{+1}5^{-1}25^{-1})$ 2 3375 $\mathrm{I}(3^{+1}9^{-1}5^{-1}25^{-1})$ 2
3380 $\mathrm{II}(4_{7}^{+1}5^{-1}13^{-2})$ 2 3388 $\mathrm{II}(4_{5}^{-1}7^{+1}11^{+2})$ 3 3456 $\mathrm{I}(2_{3}^{-1}64_{1}^{+1}3^{-1}9^{+1})$ 8
3468 $\mathrm{II}(4_{1}^{+1}3^{+1}17^{-2})$ 2 3600 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-2}5^{+2})$ 1 3600 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}5^{-2})$ 1
3600 $\mathrm{I}(4_{\mathrm{II}}^{-2}9^{+1}5^{+2})$ 2 3600 $\mathrm{I}(4_{2}^{-2}3^{+2}5^{+2})$ 2 3600 $\mathrm{I}(4_{2}^{+2}3^{+2}5^{-2})$ 1
3600 $\mathrm{I}(4_{6}^{+2}3^{-2}5^{-2})$ 2 3600 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{-2}5^{-2})$ 2 3600 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+2}5^{+2})$ 2
3600 $\mathrm{II}(16_{3}^{-1}3^{+2}5^{-2})$ 2 3630 $\mathrm{II}(2_{1}^{+1}3^{+1}5^{-1}11^{+2})$ 4 3675 $\mathrm{I}(3^{-1}5^{-2}7^{+2})$ 2
3675 $\mathrm{I}(3^{+1}5^{+2}7^{+2})$ 2 3675 $\mathrm{I}(3^{+1}5^{-2}7^{-2})$ 2 3675 $\mathrm{I}(3^{+1}5^{-2}7^{+2})$ 3
3780 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-1}9^{+1}5^{-1}7^{+1})$ 4 3780 $\mathrm{II}(4_{3}^{-1}3^{-1}9^{-1}5^{+1}7^{+1})$ 3 3780 $\mathrm{II}(4_{3}^{-1}3^{+1}9^{+1}5^{-1}7^{+1})$ 2
3780 $\mathrm{II}(4_{3}^{-1}3^{+1}9^{-1}5^{-1}7^{+1})$ 2 3840 $\mathrm{I}(16_{\mathrm{II}}^{-2}3^{-1}5^{+1})$ 2 3900 $\mathrm{II}(4_{5}^{-1}3^{+1}5^{-2}13^{+1})$ 3
3920 $\mathrm{I}(4_{\mathrm{II}}^{-2}5^{-1}7^{+2})$ 1 3920 $\mathrm{I}(4_{\mathrm{II}}^{-2}5^{+1}7^{-2})$ 2 4032 $\mathrm{I}(4_{7}^{+1}16_{3}^{-1}3^{+2}7^{+1})$ 6
4116 $\mathrm{II}(4_{7}^{+1}3^{+1}7^{-1}49^{+1})$ 2 4225 $\mathrm{I}(5^{-2}13^{-2})$ 2 4320 $\mathrm{I}(2_{3}^{-1}16_{1}^{+1}3^{+1}9^{+1}5^{-1})$ 4
4320 $\mathrm{I}(2_{5}^{-1}16_{7}^{+1}3^{+1}9^{+1}5^{-1})$ 4 4320 $\mathrm{I}(2_{3}^{-1}16_{5}^{-1}3^{+1}9^{+1}5^{-1})$ 4 4320 $\mathrm{I}(2_{5}^{-1}16_{3}^{-1}3^{+1}9^{+1}5^{-1})$ 4
4332 $\mathrm{II}(4_{1}^{+1}3^{+1}19^{+2})$ 2 4356 $\mathrm{I}(2_{2}^{+2}3^{+2}11^{+2})$ 3 4356 $\mathrm{II}(4_{3}^{-1}3^{-2}11^{-2})$ 2
4356 $\mathrm{II}(4_{3}^{-1}3^{+2}11^{+2})$ 2 4356 $\mathrm{I}(4_{1}^{+1}3^{+2}11^{+2})$ 3 4410 $\mathrm{II}(2_{3}^{-1}3^{+2}5^{+1}7^{+2})$ 2
4410 $\mathrm{I}(2_{3}^{-1}3^{+2}5^{+1}7^{+2})$ 3 4500 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+2}5^{+1}25^{-1})$ 2 4500 $\mathrm{I}(2_{0}^{+2}3^{+2}5^{+1}25^{-1})$ 3
4500 $\mathrm{II}(4_{7}^{+1}3^{+2}5^{-1}25^{+1})$ 1 4500 $\mathrm{II}(4_{7}^{+1}3^{+2}5^{-1}25^{-1})$ 1 4500 $\mathrm{II}(4_{7}^{+1}3^{-2}5^{+1}25^{+1})$ 2
4500 $\mathrm{II}(4_{7}^{+1}9^{-1}5^{+1}25^{-1})$ 3 4500 $\mathrm{II}(4_{3}^{-1}3^{+2}5^{+1}25^{-1})$ 3 4500 $\mathrm{I}(4_{5}^{-1}3^{+2}5^{+1}25^{-1})$ 3
4563 $\mathrm{I}(3^{+1}9^{-1}13^{-2})$ 2 4624 $\mathrm{I}(4_{\mathrm{II}}^{-2}17^{+2})$ 2 4725 $\mathrm{I}(3^{-1}9^{+1}5^{-2}7^{-1})$ 2
4725 $\mathrm{I}(3^{-1}9^{-1}5^{-2}7^{-1})$ 2 4725 $\mathrm{I}(3^{+1}9^{+1}5^{+2}7^{-1})$ 3 4752 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}9^{-1}11^{+1})$ 3
4752 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{+1}11^{-1})$ 2 4752 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{-1}11^{-1})$ 2 4761 $\mathrm{I}(3^{+2}23^{+2})$ 3
4900 $\mathrm{II}(4_{3}^{-1}5^{-2}7^{+2})$ 1 4900 $\mathrm{II}(4_{3}^{-1}5^{+2}7^{-2})$ 2 4950 $\mathrm{II}(2_{5}^{-1}3^{+2}5^{-2}11^{-1})$ 4
5040 $\mathrm{I}(4_{2}^{+2}3^{+2}5^{+1}7^{-1})$ 3 5040 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+2}5^{+1}7^{-1})$ 3 5040 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}5^{+1}7^{+1})$ 2
5040 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}5^{-1}7^{-1})$ 2 5040 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-2}5^{+1}7^{-1})$ 2 5184 $\mathrm{I}(4_{1}^{+1}16_{1}^{+1}9^{+2})$ 6
5200 $\mathrm{I}(4_{\mathrm{II}}^{-2}5^{-2}13^{-1})$ 2 5292 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+1}9^{+1}7^{-2})$ 4 5292 $\mathrm{I}(2_{6}^{-2}3^{+1}9^{+1}7^{+2})$ 3
5292 $\mathrm{II}(4_{1}^{+1}3^{-1}9^{+1}7^{-2})$ 2 5292 $\mathrm{II}(4_{1}^{+1}3^{+1}9^{-1}7^{+2})$ 1 5292 $\mathrm{II}(4_{1}^{+1}3^{-1}9^{-1}7^{-2})$ 2
5292 $\mathrm{II}(4_{1}^{+1}3^{+1}9^{+1}7^{+2})$ 1 5292 $\mathrm{I}(4_{3}^{-1}3^{+1}9^{+1}7^{+2})$ 3 5376 $\mathrm{I}(16_{2}^{+2}3^{+1}7^{-1})$ 4
5376 $\mathrm{I}(16_{6}^{-2}3^{+1}7^{-1})$ 3 5400 $\mathrm{I}(2_{1}^{+1}4_{5}^{-1}3^{-1}9^{-1}5^{-2})$ 4 5445 $\mathrm{I}(3^{+2}5^{-1}11^{-2})$ 2
5616 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{-1}13^{-1})$ 2 5808 $\mathrm{I}(4_{6}^{+2}3^{+1}11^{+2})$ 3 5808 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}11^{-2})$ 2
5808 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}11^{+2})$ 2 5929 $\mathrm{I}(7^{+2}11^{+2})$ 3 6000 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}5^{-1}25^{-1})$ 2
6000 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}5^{+1}25^{+1})$ 1 6000 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}5^{+1}25^{-1})$ 1 6000 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{-1}5^{-1}25^{+1})$ 3
6000 $\mathrm{I}(4_{2}^{+2}3^{-1}5^{-1}25^{+1})$ 3 6084 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+2}13^{+2})$ 3 6084 $\mathrm{II}(4_{3}^{-1}3^{+2}13^{-2})$ 1
6300 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+2}5^{-2}7^{+1})$ 2 6300 $\mathrm{I}(2_{0}^{+2}3^{+2}5^{-2}7^{+1})$ 3 6300 $\mathrm{II}(4_{5}^{-1}3^{-2}5^{-2}7^{+1})$ 2
6300 $\mathrm{II}(4_{5}^{-1}3^{+2}5^{+2}7^{+1})$ 2 6300 $\mathrm{II}(4_{5}^{-1}3^{+2}5^{-2}7^{-1})$ 2 6300 $\mathrm{II}(4_{1}^{+1}3^{+2}5^{-2}7^{+1})$ 3
6300 $\mathrm{I}(4_{7}^{+1}3^{+2}5^{-2}7^{+1})$ 3 6348 $\mathrm{II}(4_{1}^{+1}3^{+1}23^{+2})$ 3 6400 $\mathrm{I}(16_{2}^{-2}5^{-2})$ 2
6400 $\mathrm{I}(16_{6}^{-2}5^{+2})$ 3 6400 $\mathrm{I}(16_{2}^{+2}5^{+2})$ 4 6480 $\mathrm{I}(4_{\mathrm{II}}^{-2}9^{+2}5^{+1})$ 2
6615 $\mathrm{I}(3^{-1}9^{-1}5^{-1}7^{+2})$ 3 6615 $\mathrm{I}(3^{+1}9^{+1}5^{+1}7^{+2})$ 2 6615 $\mathrm{I}(3^{+1}9^{-1}5^{+1}7^{+2})$ 2
6750 $\mathrm{II}(2_{1}^{+1}3^{+1}9^{-1}5^{+1}25^{-1})$ 4 6750 $\mathrm{I}(2_{7}^{+1}3^{+1}9^{-1}5^{+1}25^{-1})$ 6 6912 $\mathrm{I}(16_{2}^{+2}3^{+1}9^{-1})$ 2
6912 $\mathrm{I}(16_{2}^{+2}3^{+1}9^{+1})$ 2 6912 $\mathrm{I}(16_{0}^{+2}3^{+1}9^{+1})$ 4 6912 $\mathrm{I}(16_{\mathrm{II}}^{-2}3^{-1}9^{+1})$ 2
6912 $\mathrm{I}(16_{\mathrm{II}}^{-2}3^{-1}9^{-1})$ 2 6912 $\mathrm{I}(16_{4}^{-2}3^{+1}9^{+1})$ 4 6912 $\mathrm{I}(16_{6}^{-2}3^{+1}9^{+1})$ 2
7056 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-2}7^{-2})$ 2 7056 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}7^{+2})$ 1 7056 $\mathrm{I}(4_{2}^{-2}3^{+2}7^{-2})$ 2
7056 $\mathrm{I}(4_{2}^{+2}3^{+2}7^{+2})$ 2 7056 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+2}7^{-2})$ 2 7056 $\mathrm{I}(16_{5}^{-1}3^{+2}7^{+2})$ 3
7056 $\mathrm{I}(16_{1}^{+1}3^{+2}7^{+2})$ 4 7200 $\mathrm{I}(2_{3}^{-1}16_{1}^{+1}3^{+2}5^{-2})$ 2 7200 $\mathrm{I}(2_{1}^{+1}16_{3}^{-1}3^{+2}5^{+2})$ 3
7200 $\mathrm{I}(2_{3}^{-1}16_{7}^{+1}3^{+2}5^{-2})$ 2 7200 $\mathrm{I}(2_{5}^{-1}16_{5}^{-1}3^{+2}5^{+2})$ 3 7200 $\mathrm{I}(2_{3}^{-1}16_{5}^{-1}3^{+2}5^{-2})$ 2
7200 $\mathrm{I}(2_{5}^{-1}16_{3}^{-1}3^{+2}5^{-2})$ 2 7200 $\mathrm{I}(2_{1}^{+1}16_{7}^{+1}3^{+2}5^{+2})$ 3 7200 $\mathrm{I}(2_{7}^{+1}16_{1}^{+1}3^{+2}5^{+2})$ 3
7260 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-1}5^{+1}11^{+2})$ 4 7260 $\mathrm{II}(4_{5}^{-1}3^{-1}5^{+1}11^{-2})$ 2 7350 $\mathrm{II}(2_{5}^{-1}3^{-1}5^{-2}7^{+2})$ 2
7350 $\mathrm{I}(2_{5}^{-1}3^{-1}5^{-2}7^{+2})$ 3 7605 $\mathrm{I}(3^{+2}5^{-1}13^{+2})$ 3 7875 $\mathrm{I}(3^{+2}5^{-1}25^{-1}7^{-1})$ 3
7920 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}5^{-1}11^{-1})$ 2 7938 $\mathrm{II}(2_{7}^{+1}9^{+2}7^{+2})$ 3 8000 $\mathrm{I}(4_{1}^{+1}16_{1}^{+1}5^{+1}25^{+1})$ 12
8100 $\mathrm{II}(4_{3}^{-1}9^{+2}5^{+2})$ 2 8112 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}13^{-2})$ 1 8208 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{-1}19^{-1})$ 3
8400 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+1}5^{-2}7^{-1})$ 3 8400 $\mathrm{I}(4_{6}^{+2}3^{+1}5^{-2}7^{-1})$ 3 8400 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}5^{-2}7^{-1})$ 2
8400 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}5^{-2}7^{+1})$ 2 8400 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}5^{+2}7^{-1})$ 2 8624 $\mathrm{I}(4_{\mathrm{II}}^{-2}7^{+2}11^{+1})$ 3
8820 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+2}5^{-1}7^{+2})$ 2 8820 $\mathrm{I}(2_{0}^{+2}3^{+2}5^{-1}7^{+2})$ 3 8820 $\mathrm{II}(4_{7}^{+1}3^{+2}5^{-1}7^{-2})$ 2
8820 $\mathrm{II}(4_{7}^{+1}3^{+2}5^{+1}7^{+2})$ 2 8820 $\mathrm{II}(4_{7}^{+1}3^{-2}5^{-1}7^{+2})$ 2 8820 $\mathrm{II}(4_{3}^{-1}3^{+2}5^{-1}7^{+2})$ 3
8820 $\mathrm{I}(4_{5}^{-1}3^{+2}5^{-1}7^{+2})$ 3 9075 $\mathrm{I}(3^{+1}5^{-2}11^{-2})$ 2 9126 $\mathrm{II}(2_{5}^{-1}3^{-1}9^{-1}13^{+2})$ 6
9360 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}5^{+1}13^{-1})$ 3 9408 $\mathrm{I}(4_{3}^{-1}16_{3}^{-1}3^{+1}7^{+2})$ 6 9450 $\mathrm{II}(2_{3}^{-1}3^{-1}9^{+1}5^{-2}7^{-1})$ 4
9747 $\mathrm{I}(3^{+1}9^{-1}19^{+2})$ 3 9900 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+2}5^{-2}11^{+1})$ 4 9900 $\mathrm{II}(4_{1}^{+1}3^{+2}5^{-2}11^{-1})$ 2
10125 $\mathrm{I}(9^{+2}5^{+1}25^{-1})$ 3 10140 $\mathrm{II}(4_{5}^{-1}3^{-1}5^{+1}13^{+2})$ 3 10404 $\mathrm{II}(4_{3}^{-1}3^{+2}17^{-2})$ 2
10500 $\mathrm{II}(4_{3}^{-1}3^{-1}5^{+1}25^{+1}7^{+1})$ 3 10800 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{-1}9^{+1}5^{-2})$ 2 10800 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+1}9^{-1}5^{+2})$ 3
10800 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{-1}9^{-1}5^{-2})$ 2 10800 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}9^{+1}5^{+2})$ 1 10800 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}9^{-1}5^{+2})$ 1
10800 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{+1}5^{-2})$ 2 10800 $\mathrm{I}(4_{2}^{-2}3^{+1}9^{+1}5^{-2})$ 2 10800 $\mathrm{I}(4_{6}^{-2}3^{-1}9^{+1}5^{-2})$ 2
10800 $\mathrm{I}(4_{6}^{-2}3^{+1}9^{-1}5^{+2})$ 3 10800 $\mathrm{I}(4_{6}^{-2}3^{-1}9^{-1}5^{-2})$ 2 10800 $\mathrm{II}(16_{1}^{+1}3^{+1}9^{+1}5^{-2})$ 4
10890 $\mathrm{II}(2_{3}^{-1}3^{+2}5^{+1}11^{+2})$ 4 11025 $\mathrm{I}(3^{-2}5^{-2}7^{+2})$ 2 11025 $\mathrm{I}(3^{+2}5^{-2}7^{-2})$ 2
11025 $\mathrm{I}(3^{+2}5^{+2}7^{+2})$ 2 11025 $\mathrm{I}(3^{+2}5^{-2}7^{+2})$ 3 11520 $\mathrm{I}(16_{\mathrm{II}}^{-2}3^{+2}5^{-1})$ 2
11700 $\mathrm{II}(4_{7}^{+1}3^{+2}5^{-2}13^{+1})$ 3 11760 $\mathrm{I}(4_{6}^{-2}3^{-1}5^{+1}7^{+2})$ 3 11760 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}5^{+1}7^{+2})$ 2
11760 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}5^{+1}7^{-2})$ 2 11760 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}5^{-1}7^{+2})$ 2 11760 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{-1}5^{+1}7^{+2})$ 3
12096 $\mathrm{I}(4_{1}^{+1}16_{1}^{+1}3^{+1}9^{+1}7^{-1})$ 12 12348 $\mathrm{II}(4_{5}^{-1}3^{+2}7^{+1}49^{-1})$ 2 12675 $\mathrm{I}(3^{+1}5^{-2}13^{+2})$ 3
12996 $\mathrm{II}(4_{3}^{-1}3^{+2}19^{+2})$ 2 13068 $\mathrm{II}(4_{1}^{+1}3^{-1}9^{+1}11^{-2})$ 2 13068 $\mathrm{II}(4_{1}^{+1}3^{-1}9^{-1}11^{-2})$ 2
13068 $\mathrm{II}(4_{1}^{+1}3^{+1}9^{+1}11^{+2})$ 3 13200 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}5^{-2}11^{-1})$ 2 13230 $\mathrm{II}(2_{1}^{+1}3^{+1}9^{-1}5^{-1}7^{+2})$ 4
13500 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-1}9^{+1}5^{-1}25^{+1})$ 4 13500 $\mathrm{I}(2_{2}^{+2}3^{-1}9^{+1}5^{-1}25^{+1})$ 6 13500 $\mathrm{II}(4_{5}^{-1}3^{-1}9^{-1}5^{+1}25^{+1})$ 2
13500 $\mathrm{II}(4_{5}^{-1}3^{-1}9^{-1}5^{+1}25^{-1})$ 2 13500 $\mathrm{II}(4_{5}^{-1}3^{+1}9^{+1}5^{-1}25^{-1})$ 2 13500 $\mathrm{II}(4_{5}^{-1}3^{+1}9^{-1}5^{-1}25^{-1})$ 2
13500 $\mathrm{II}(4_{1}^{+1}3^{-1}9^{+1}5^{-1}25^{+1})$ 6 13500 $\mathrm{I}(4_{7}^{+1}3^{-1}9^{+1}5^{-1}25^{+1})$ 6 13520 $\mathrm{I}(4_{\mathrm{II}}^{-2}5^{-1}13^{-2})$ 2
13552 $\mathrm{I}(4_{\mathrm{II}}^{-2}7^{+1}11^{+2})$ 3 13872 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}17^{-2})$ 2 14700 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+1}5^{-2}7^{+2})$ 2
14700 $\mathrm{I}(2_{0}^{+2}3^{+1}5^{-2}7^{+2})$ 3 14700 $\mathrm{II}(4_{1}^{+1}3^{-1}5^{-2}7^{+2})$ 2 14700 $\mathrm{II}(4_{1}^{+1}3^{+1}5^{+2}7^{+2})$ 2
14700 $\mathrm{II}(4_{1}^{+1}3^{+1}5^{-2}7^{-2})$ 2 14700 $\mathrm{II}(4_{5}^{-1}3^{+1}5^{-2}7^{+2})$ 3 14700 $\mathrm{I}(4_{3}^{-1}3^{+1}5^{-2}7^{+2})$ 3
15120 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}9^{-1}5^{+1}7^{+1})$ 3 15120 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{+1}5^{-1}7^{+1})$ 2 15120 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{-1}5^{-1}7^{+1})$ 2
15600 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}5^{-2}13^{+1})$ 3 15876 $\mathrm{I}(2_{\mathrm{II}}^{+2}9^{+2}7^{+2})$ 3 16128 $\mathrm{I}(16_{2}^{-2}3^{+2}7^{+1})$ 3
16128 $\mathrm{I}(16_{6}^{+2}3^{+2}7^{+1})$ 4 16464 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}7^{-1}49^{+1})$ 2 16900 $\mathrm{II}(4_{3}^{-1}5^{-2}13^{-2})$ 2
17328 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}19^{+2})$ 2 17424 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-2}11^{-2})$ 2 17424 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}11^{+2})$ 2
17424 $\mathrm{I}(4_{2}^{+2}3^{+2}11^{+2})$ 3 18000 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+2}5^{+1}25^{-1})$ 3 18000 $\mathrm{I}(4_{6}^{+2}3^{+2}5^{+1}25^{-1})$ 3
18000 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}5^{-1}25^{+1})$ 1 18000 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}5^{-1}25^{-1})$ 1 18000 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-2}5^{+1}25^{+1})$ 2
18000 $\mathrm{I}(4_{\mathrm{II}}^{-2}9^{-1}5^{+1}25^{-1})$ 3 18150 $\mathrm{II}(2_{5}^{-1}3^{-1}5^{-2}11^{+2})$ 4 18252 $\mathrm{I}(2_{\mathrm{II}}^{-2}3^{+1}9^{+1}13^{+2})$ 6
18252 $\mathrm{II}(4_{1}^{+1}3^{+1}9^{-1}13^{-2})$ 2 18375 $\mathrm{I}(3^{-1}5^{-1}25^{-1}7^{+2})$ 3 18900 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+1}9^{-1}5^{-2}7^{-1})$ 4
18900 $\mathrm{II}(4_{7}^{+1}3^{-1}9^{+1}5^{-2}7^{-1})$ 2 18900 $\mathrm{II}(4_{7}^{+1}3^{-1}9^{-1}5^{-2}7^{-1})$ 2 18900 $\mathrm{II}(4_{7}^{+1}3^{+1}9^{+1}5^{+2}7^{-1})$ 3
19044 $\mathrm{II}(4_{3}^{-1}3^{+2}23^{+2})$ 3 19200 $\mathrm{I}(16_{\mathrm{II}}^{-2}3^{+1}5^{-2})$ 2 19600 $\mathrm{I}(4_{\mathrm{II}}^{-2}5^{-2}7^{+2})$ 1
19600 $\mathrm{I}(4_{\mathrm{II}}^{-2}5^{+2}7^{-2})$ 2 20736 $\mathrm{I}(16_{6}^{-2}9^{+2})$ 3 20736 $\mathrm{I}(16_{2}^{+2}9^{+2})$ 4
21168 $\mathrm{I}(4_{6}^{+2}3^{+1}9^{+1}7^{+2})$ 3 21168 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}9^{+1}7^{-2})$ 2 21168 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{-1}7^{+2})$ 1
21168 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}9^{-1}7^{-2})$ 2 21168 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{+1}7^{+2})$ 1 21168 $\mathrm{I}(16_{7}^{+1}3^{+1}9^{+1}7^{+2})$ 6
21168 $\mathrm{I}(16_{3}^{-1}3^{+1}9^{+1}7^{+2})$ 6 21600 $\mathrm{I}(2_{1}^{+1}16_{1}^{+1}3^{-1}9^{-1}5^{-2})$ 4 21600 $\mathrm{I}(2_{1}^{+1}16_{3}^{-1}3^{-1}9^{-1}5^{-2})$ 4
21600 $\mathrm{I}(2_{3}^{-1}16_{7}^{+1}3^{-1}9^{-1}5^{-2})$ 4 21600 $\mathrm{I}(2_{1}^{+1}16_{5}^{-1}3^{-1}9^{-1}5^{-2})$ 4 21780 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+2}5^{-1}11^{+2})$ 4
21780 $\mathrm{II}(4_{7}^{+1}3^{+2}5^{-1}11^{-2})$ 2 22050 $\mathrm{II}(2_{7}^{+1}3^{+2}5^{-2}7^{+2})$ 2 22050 $\mathrm{I}(2_{7}^{+1}3^{+2}5^{-2}7^{+2})$ 3
23716 $\mathrm{II}(4_{3}^{-1}7^{+2}11^{+2})$ 3 24336 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}13^{-2})$ 1 25200 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-2}5^{-2}7^{+1})$ 2
25200 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}5^{+2}7^{+1})$ 2 25200 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}5^{-2}7^{-1})$ 2 25200 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+2}5^{-2}7^{+1})$ 3
25200 $\mathrm{I}(4_{2}^{+2}3^{+2}5^{-2}7^{+1})$ 3 25392 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}23^{+2})$ 3 26460 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-1}9^{+1}5^{+1}7^{+2})$ 4
26460 $\mathrm{II}(4_{5}^{-1}3^{-1}9^{-1}5^{-1}7^{+2})$ 3 26460 $\mathrm{II}(4_{5}^{-1}3^{+1}9^{+1}5^{+1}7^{+2})$ 2 26460 $\mathrm{II}(4_{5}^{-1}3^{+1}9^{-1}5^{+1}7^{+2})$ 2
27225 $\mathrm{I}(3^{+2}5^{-2}11^{-2})$ 2 28224 $\mathrm{I}(4_{5}^{-1}16_{1}^{+1}3^{+2}7^{+2})$ 6 29040 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}5^{+1}11^{-2})$ 2
30420 $\mathrm{II}(4_{7}^{+1}3^{+2}5^{-1}13^{+2})$ 3 31500 $\mathrm{II}(4_{1}^{+1}3^{+2}5^{-1}25^{-1}7^{-1})$ 3 32000 $\mathrm{I}(16_{2}^{+2}5^{+1}25^{+1})$ 6
32000 $\mathrm{I}(16_{6}^{-2}5^{+1}25^{+1})$ 6 32400 $\mathrm{I}(4_{\mathrm{II}}^{-2}9^{+2}5^{+2})$ 2 33075 $\mathrm{I}(3^{-1}9^{+1}5^{-2}7^{+2})$ 2
33075 $\mathrm{I}(3^{-1}9^{-1}5^{-2}7^{+2})$ 2 33075 $\mathrm{I}(3^{+1}9^{+1}5^{+2}7^{+2})$ 3 34560 $\mathrm{I}(16_{\mathrm{II}}^{-2}3^{-1}9^{-1}5^{+1})$ 4
35280 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+2}5^{-1}7^{+2})$ 3 35280 $\mathrm{I}(4_{6}^{+2}3^{+2}5^{-1}7^{+2})$ 3 35280 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}5^{-1}7^{-2})$ 2
35280 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}5^{+1}7^{+2})$ 2 35280 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-2}5^{-1}7^{+2})$ 2 36300 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+1}5^{-2}11^{+2})$ 4
36300 $\mathrm{II}(4_{1}^{+1}3^{+1}5^{-2}11^{-2})$ 2 37632 $\mathrm{I}(16_{6}^{+2}3^{+1}7^{+2})$ 4 37632 $\mathrm{I}(16_{2}^{-2}3^{+1}7^{+2})$ 3
38025 $\mathrm{I}(3^{+2}5^{-2}13^{+2})$ 3 38988 $\mathrm{II}(4_{1}^{+1}3^{+1}9^{-1}19^{+2})$ 3 39600 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}5^{-2}11^{-1})$ 2
40500 $\mathrm{II}(4_{7}^{+1}9^{+2}5^{+1}25^{-1})$ 3 40560 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}5^{+1}13^{+2})$ 3 41616 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}17^{-2})$ 2
42000 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}5^{+1}25^{+1}7^{+1})$ 3 44100 $\mathrm{I}(2_{6}^{+2}3^{+2}5^{-2}7^{+2})$ 3 44100 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+2}5^{-2}7^{+2})$ 2
44100 $\mathrm{II}(4_{3}^{-1}3^{-2}5^{-2}7^{+2})$ 2 44100 $\mathrm{II}(4_{3}^{-1}3^{+2}5^{-2}7^{-2})$ 2 44100 $\mathrm{II}(4_{3}^{-1}3^{+2}5^{+2}7^{+2})$ 2
44100 $\mathrm{II}(4_{7}^{+1}3^{+2}5^{-2}7^{+2})$ 3 44100 $\mathrm{I}(4_{1}^{+1}3^{+2}5^{-2}7^{+2})$ 3 46800 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}5^{-2}13^{+1})$ 3
48384 $\mathrm{I}(16_{2}^{+2}3^{+1}9^{+1}7^{-1})$ 6 48384 $\mathrm{I}(16_{6}^{-2}3^{+1}9^{+1}7^{-1})$ 6 49392 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}7^{+1}49^{-1})$ 2
50700 $\mathrm{II}(4_{1}^{+1}3^{+1}5^{-2}13^{+2})$ 3 51984 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}19^{+2})$ 2 52272 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}9^{+1}11^{-2})$ 2
52272 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}9^{-1}11^{-2})$ 2 52272 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{+1}11^{+2})$ 3 54000 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}9^{-1}5^{+1}25^{+1})$ 2
54000 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}9^{-1}5^{+1}25^{-1})$ 2 54000 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{+1}5^{-1}25^{-1})$ 2 54000 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{-1}5^{-1}25^{-1})$ 2
54000 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{-1}9^{+1}5^{-1}25^{+1})$ 6 54000 $\mathrm{I}(4_{2}^{+2}3^{-1}9^{+1}5^{-1}25^{+1})$ 6 54450 $\mathrm{II}(2_{3}^{-1}3^{+2}5^{-2}11^{+2})$ 4
55125 $\mathrm{I}(3^{+2}5^{+1}25^{+1}7^{+2})$ 3 57600 $\mathrm{I}(16_{\mathrm{II}}^{-2}3^{+2}5^{-2})$ 2 58800 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+1}5^{-2}7^{+2})$ 3
58800 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}5^{-2}7^{+2})$ 2 58800 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}5^{+2}7^{+2})$ 2 58800 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}5^{-2}7^{-2})$ 2
58800 $\mathrm{I}(4_{6}^{-2}3^{+1}5^{-2}7^{+2})$ 3 66150 $\mathrm{II}(2_{5}^{-1}3^{-1}9^{+1}5^{-2}7^{+2})$ 4 67600 $\mathrm{I}(4_{\mathrm{II}}^{-2}5^{-2}13^{-2})$ 2
73008 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{-1}13^{-2})$ 2 73500 $\mathrm{II}(4_{5}^{-1}3^{-1}5^{-1}25^{-1}7^{+2})$ 3 75600 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}9^{+1}5^{-2}7^{-1})$ 2
75600 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}9^{-1}5^{-2}7^{-1})$ 2 75600 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{+1}5^{+2}7^{-1})$ 3 76176 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}23^{+2})$ 3
84672 $\mathrm{I}(4_{3}^{-1}16_{3}^{-1}3^{+1}9^{+1}7^{+2})$ 12 87120 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}5^{-1}11^{-2})$ 2 94864 $\mathrm{I}(4_{\mathrm{II}}^{-2}7^{+2}11^{+2})$ 3
105840 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}9^{-1}5^{-1}7^{+2})$ 3 105840 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{+1}5^{+1}7^{+2})$ 2 105840 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{-1}5^{+1}7^{+2})$ 2
108900 $\mathrm{I}(2_{\mathrm{II}}^{-2}3^{+2}5^{-2}11^{+2})$ 4 108900 $\mathrm{II}(4_{3}^{-1}3^{+2}5^{-2}11^{-2})$ 2 112896 $\mathrm{I}(16_{6}^{-2}3^{+2}7^{+2})$ 3
112896 $\mathrm{I}(16_{2}^{+2}3^{+2}7^{+2})$ 4 121680 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}5^{-1}13^{+2})$ 3 126000 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}5^{-1}25^{-1}7^{-1})$ 3
132300 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+1}9^{-1}5^{-2}7^{+2})$ 4 132300 $\mathrm{II}(4_{1}^{+1}3^{-1}9^{+1}5^{-2}7^{+2})$ 2 132300 $\mathrm{II}(4_{1}^{+1}3^{-1}9^{-1}5^{-2}7^{+2})$ 2
132300 $\mathrm{II}(4_{1}^{+1}3^{+1}9^{+1}5^{+2}7^{+2})$ 3 145200 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}5^{-2}11^{-2})$ 2 152100 $\mathrm{II}(4_{3}^{-1}3^{+2}5^{-2}13^{+2})$ 3
155952 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{-1}19^{+2})$ 3 162000 $\mathrm{I}(4_{\mathrm{II}}^{-2}9^{+2}5^{+1}25^{-1})$ 3 172800 $\mathrm{I}(16_{\mathrm{II}}^{-2}3^{+1}9^{+1}5^{-2})$ 4
176400 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-2}5^{-2}7^{+2})$ 2 176400 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}5^{-2}7^{-2})$ 2 176400 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}5^{+2}7^{+2})$ 2
176400 $\mathrm{I}(4_{6}^{+2}3^{+2}5^{-2}7^{+2})$ 3 176400 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+2}5^{-2}7^{+2})$ 3 202800 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}5^{-2}13^{+2})$ 3
220500 $\mathrm{II}(4_{7}^{+1}3^{+2}5^{+1}25^{+1}7^{+2})$ 3 294000 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}5^{-1}25^{-1}7^{+2})$ 3 338688 $\mathrm{I}(16_{6}^{+2}3^{+1}9^{+1}7^{+2})$ 6
338688 $\mathrm{I}(16_{2}^{-2}3^{+1}9^{+1}7^{+2})$ 6 435600 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}5^{-2}11^{-2})$ 2 529200 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}9^{+1}5^{-2}7^{+2})$ 2
529200 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}9^{-1}5^{-2}7^{+2})$ 2 529200 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{+1}5^{+2}7^{+2})$ 3 608400 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}5^{-2}13^{+2})$ 3
882000 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}5^{+1}25^{+1}7^{+2})$ 3
: Totally-Reflective Genera in dimension $3$
Dimension 4
-----------
$930$ genera, of which $230$ are square free and 88 strongly square free.
$\det$ Genus h $\det$ Genus h $\det$ Genus h
-------- ----------------------------------------------------------- ---- --------- ----------------------------------------------------------- ---- --------- ---------------------------------------------------------------- ----
1 $\mathrm{I}(1_{4}^{+4})$ 1 2 $\mathrm{I}(2_{1}^{+1})$ 1 3 $\mathrm{I}(3^{-1})$ 1
3 $\mathrm{I}(3^{+1})$ 1 4 $\mathrm{II}(2_{\mathrm{II}}^{-2})$ 1 4 $\mathrm{I}(2_{2}^{+2})$ 1
4 $\mathrm{I}(4_{3}^{-1})$ 1 4 $\mathrm{I}(4_{1}^{+1})$ 1 5 $\mathrm{II}(5^{+1})$ 1
5 $\mathrm{I}(5^{+1})$ 1 5 $\mathrm{I}(5^{-1})$ 1 6 $\mathrm{I}(2_{1}^{+1}3^{-1})$ 2
6 $\mathrm{I}(2_{1}^{+1}3^{+1})$ 1 7 $\mathrm{I}(7^{+1})$ 2 8 $\mathrm{I}(2_{3}^{+3})$ 1
8 $\mathrm{II}(2_{1}^{+1}4_{3}^{-1})$ 1 8 $\mathrm{I}(2_{1}^{+1}4_{1}^{+1})$ 1 9 $\mathrm{II}(3^{+2})$ 1
9 $\mathrm{I}(3^{+2})$ 1 9 $\mathrm{I}(3^{-2})$ 1 9 $\mathrm{I}(9^{-1})$ 1
10 $\mathrm{I}(2_{5}^{-1}5^{-1})$ 2 12 $\mathrm{II}(2_{2}^{+2}3^{-1})$ 1 12 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-1})$ 1
12 $\mathrm{I}(2_{2}^{+2}3^{+1})$ 1 12 $\mathrm{II}(2_{2}^{-2}3^{+1})$ 1 12 $\mathrm{I}(2_{\mathrm{II}}^{-2}3^{+1})$ 1
12 $\mathrm{I}(2_{4}^{-2}3^{-1})$ 1 12 $\mathrm{I}(4_{1}^{+1}3^{-1})$ 1 12 $\mathrm{I}(4_{7}^{+1}3^{-1})$ 1
12 $\mathrm{I}(4_{1}^{+1}3^{+1})$ 2 12 $\mathrm{I}(4_{3}^{-1}3^{+1})$ 2 14 $\mathrm{I}(2_{7}^{+1}7^{+1})$ 3
15 $\mathrm{I}(3^{+1}5^{+1})$ 2 15 $\mathrm{I}(3^{-1}5^{-1})$ 3 15 $\mathrm{I}(3^{+1}5^{-1})$ 1
16 $\mathrm{I}(2_{2}^{+2}4_{1}^{+1})$ 1 16 $\mathrm{II}(4_{4}^{-2})$ 1 16 $\mathrm{I}(4_{2}^{-2})$ 1
16 $\mathrm{I}(4_{\mathrm{II}}^{-2})$ 1 16 $\mathrm{I}(4_{2}^{+2})$ 1 16 $\mathrm{I}(16_{5}^{-1})$ 2
16 $\mathrm{I}(16_{1}^{+1})$ 2 18 $\mathrm{I}(2_{1}^{+1}3^{-2})$ 2 18 $\mathrm{I}(2_{1}^{+1}3^{+2})$ 2
20 $\mathrm{II}(2_{\mathrm{II}}^{+2}5^{+1})$ 1 20 $\mathrm{I}(2_{2}^{+2}5^{-1})$ 2 20 $\mathrm{I}(2_{2}^{+2}5^{+1})$ 2
20 $\mathrm{II}(2_{\mathrm{II}}^{-2}5^{-1})$ 1 20 $\mathrm{I}(4_{7}^{+1}5^{+1})$ 2 20 $\mathrm{I}(4_{1}^{+1}5^{+1})$ 2
21 $\mathrm{II}(3^{-1}7^{+1})$ 1 21 $\mathrm{I}(3^{-1}7^{+1})$ 2 24 $\mathrm{I}(2_{1}^{-3}3^{+1})$ 2
24 $\mathrm{I}(2_{5}^{-3}3^{-1})$ 1 24 $\mathrm{II}(2_{1}^{+1}4_{1}^{+1}3^{-1})$ 2 24 $\mathrm{I}(2_{1}^{+1}4_{7}^{+1}3^{-1})$ 2
24 $\mathrm{II}(2_{1}^{+1}4_{5}^{-1}3^{+1})$ 1 24 $\mathrm{I}(2_{1}^{+1}4_{3}^{-1}3^{+1})$ 2 25 $\mathrm{II}(5^{-2})$ 1
25 $\mathrm{I}(5^{-2})$ 1 25 $\mathrm{I}(5^{+2})$ 2 27 $\mathrm{I}(3^{+3})$ 1
27 $\mathrm{I}(3^{-1}9^{-1})$ 2 27 $\mathrm{I}(3^{-3})$ 1 27 $\mathrm{I}(3^{+1}9^{-1})$ 1
27 $\mathrm{I}(3^{+1}9^{+1})$ 1 28 $\mathrm{II}(2_{2}^{+2}7^{+1})$ 2 28 $\mathrm{I}(2_{\mathrm{II}}^{+2}7^{+1})$ 2
28 $\mathrm{I}(4_{7}^{+1}7^{+1})$ 3 30 $\mathrm{I}(2_{1}^{+1}3^{-1}5^{-1})$ 4 32 $\mathrm{I}(2_{1}^{+1}4_{\mathrm{II}}^{-2})$ 1
32 $\mathrm{I}(2_{1}^{+1}4_{2}^{+2})$ 1 32 $\mathrm{I}(2_{3}^{-1}16_{7}^{+1})$ 1 32 $\mathrm{I}(2_{1}^{+1}16_{5}^{-1})$ 1
32 $\mathrm{I}(2_{3}^{-1}16_{3}^{-1})$ 2 32 $\mathrm{I}(2_{1}^{+1}16_{1}^{+1})$ 2 33 $\mathrm{I}(3^{+1}11^{-1})$ 2
36 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{-2})$ 1 36 $\mathrm{II}(2_{\mathrm{II}}^{-2}9^{-1})$ 1 36 $\mathrm{II}(2_{\mathrm{II}}^{-2}9^{+1})$ 1
36 $\mathrm{II}(2_{4}^{-2}3^{+2})$ 1 36 $\mathrm{I}(2_{\mathrm{II}}^{-2}3^{+2})$ 1 36 $\mathrm{I}(2_{4}^{-2}3^{-2})$ 2
36 $\mathrm{II}(2_{\mathrm{II}}^{+2}3^{+2})$ 1 36 $\mathrm{I}(2_{2}^{+2}3^{+2})$ 2 36 $\mathrm{I}(4_{5}^{-1}3^{-2})$ 2
36 $\mathrm{I}(4_{7}^{+1}3^{+2})$ 2 36 $\mathrm{I}(4_{7}^{+1}3^{-2})$ 2 36 $\mathrm{I}(4_{1}^{+1}3^{+2})$ 2
40 $\mathrm{I}(2_{3}^{-3}5^{+1})$ 2 40 $\mathrm{II}(2_{5}^{-1}4_{3}^{-1}5^{-1})$ 2 45 $\mathrm{II}(3^{+2}5^{-1})$ 1
45 $\mathrm{II}(3^{-2}5^{+1})$ 1 45 $\mathrm{I}(3^{+2}5^{-1})$ 2 45 $\mathrm{I}(3^{-2}5^{+1})$ 2
48 $\mathrm{I}(2_{\mathrm{II}}^{-2}4_{1}^{+1}3^{+1})$ 1 48 $\mathrm{I}(2_{4}^{-2}4_{1}^{+1}3^{-1})$ 1 48 $\mathrm{I}(2_{2}^{+2}4_{3}^{-1}3^{+1})$ 2
48 $\mathrm{II}(4_{2}^{+2}3^{-1})$ 1 48 $\mathrm{II}(4_{6}^{+2}3^{+1})$ 1 48 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{-1})$ 1
48 $\mathrm{I}(4_{2}^{+2}3^{+1})$ 2 48 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+1})$ 1 48 $\mathrm{II}(4_{2}^{-2}3^{-1})$ 1
48 $\mathrm{II}(4_{6}^{-2}3^{+1})$ 1 48 $\mathrm{I}(4_{2}^{-2}3^{-1})$ 1 48 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1})$ 1
48 $\mathrm{I}(4_{4}^{-2}3^{+1})$ 2 48 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1})$ 1 48 $\mathrm{I}(4_{4}^{-2}3^{-1})$ 1
48 $\mathrm{I}(16_{1}^{+1}3^{-1})$ 2 48 $\mathrm{I}(16_{5}^{-1}3^{-1})$ 2 49 $\mathrm{II}(7^{+2})$ 1
54 $\mathrm{I}(2_{1}^{+1}3^{+1}9^{-1})$ 2 54 $\mathrm{I}(2_{3}^{-1}3^{+3})$ 2 54 $\mathrm{I}(2_{3}^{-1}3^{-3})$ 1
56 $\mathrm{I}(2_{1}^{+3}7^{+1})$ 3 56 $\mathrm{II}(2_{7}^{+1}4_{3}^{-1}7^{+1})$ 3 60 $\mathrm{II}(2_{6}^{-2}3^{+1}5^{+1})$ 2
60 $\mathrm{II}(2_{2}^{+2}3^{-1}5^{-1})$ 3 60 $\mathrm{II}(2_{6}^{+2}3^{+1}5^{-1})$ 1 60 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+1}5^{+1})$ 3
60 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-1}5^{-1})$ 2 60 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-1}5^{+1})$ 1 60 $\mathrm{I}(4_{3}^{-1}3^{+1}5^{+1})$ 3
60 $\mathrm{I}(4_{1}^{+1}3^{+1}5^{-1})$ 2 63 $\mathrm{I}(3^{+2}7^{-1})$ 3 63 $\mathrm{I}(3^{+2}7^{+1})$ 3
64 $\mathrm{I}(4_{1}^{-3})$ 1 64 $\mathrm{I}(2_{2}^{-2}16_{1}^{+1})$ 2 64 $\mathrm{I}(2_{4}^{-2}16_{7}^{+1})$ 1
64 $\mathrm{I}(2_{2}^{-2}16_{5}^{-1})$ 2 64 $\mathrm{I}(2_{4}^{-2}16_{3}^{-1})$ 1 64 $\mathrm{I}(4_{3}^{+3})$ 1
64 $\mathrm{II}(4_{3}^{-1}16_{1}^{+1})$ 2 64 $\mathrm{I}(4_{1}^{+1}16_{1}^{+1})$ 3 68 $\mathrm{II}(2_{\mathrm{II}}^{-2}17^{+1})$ 2
72 $\mathrm{I}(2_{1}^{-3}3^{+2})$ 2 72 $\mathrm{I}(2_{3}^{+3}3^{-2})$ 2 72 $\mathrm{II}(2_{1}^{+1}4_{3}^{-1}3^{-2})$ 2
72 $\mathrm{I}(2_{1}^{+1}4_{3}^{-1}3^{+2})$ 3 72 $\mathrm{I}(2_{1}^{+1}4_{5}^{-1}3^{-2})$ 3 72 $\mathrm{II}(2_{1}^{+1}4_{7}^{+1}3^{+2})$ 2
75 $\mathrm{I}(3^{-1}5^{+2})$ 3 75 $\mathrm{I}(3^{+1}5^{-2})$ 3 80 $\mathrm{I}(2_{2}^{+2}4_{5}^{-1}5^{+1})$ 3
80 $\mathrm{II}(4_{0}^{+2}5^{+1})$ 2 80 $\mathrm{I}(4_{2}^{+2}5^{-1})$ 2 80 $\mathrm{I}(4_{\mathrm{II}}^{-2}5^{+1})$ 2
80 $\mathrm{I}(4_{2}^{-2}5^{-1})$ 2 80 $\mathrm{I}(4_{6}^{-2}5^{+1})$ 2 80 $\mathrm{I}(4_{2}^{-2}5^{+1})$ 2
80 $\mathrm{I}(16_{1}^{+1}5^{+1})$ 5 80 $\mathrm{I}(16_{5}^{-1}5^{+1})$ 5 81 $\mathrm{II}(3^{+2}9^{-1})$ 1
81 $\mathrm{II}(3^{+2}9^{+1})$ 1 81 $\mathrm{I}(3^{+2}9^{-1})$ 1 81 $\mathrm{I}(3^{+2}9^{+1})$ 1
81 $\mathrm{I}(3^{-2}9^{+1})$ 1 81 $\mathrm{I}(3^{-2}9^{-1})$ 1 81 $\mathrm{I}(9^{+2})$ 2
84 $\mathrm{II}(2_{\mathrm{II}}^{+2}3^{-1}7^{+1})$ 2 84 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+1}7^{+1})$ 2 90 $\mathrm{I}(2_{7}^{+1}3^{+2}5^{+1})$ 4
90 $\mathrm{I}(2_{1}^{+1}3^{+2}5^{-1})$ 4 96 $\mathrm{I}(2_{1}^{+1}4_{\mathrm{II}}^{+2}3^{+1})$ 1 96 $\mathrm{I}(2_{1}^{+1}4_{\mathrm{II}}^{-2}3^{-1})$ 2
96 $\mathrm{I}(2_{1}^{+1}4_{4}^{-2}3^{+1})$ 2 96 $\mathrm{I}(2_{1}^{+1}4_{2}^{-2}3^{-1})$ 2 96 $\mathrm{I}(2_{1}^{+1}16_{7}^{+1}3^{+1})$ 3
96 $\mathrm{I}(2_{7}^{+1}16_{1}^{+1}3^{+1})$ 3 96 $\mathrm{I}(2_{1}^{+1}16_{3}^{-1}3^{+1})$ 4 96 $\mathrm{I}(2_{7}^{+1}16_{5}^{-1}3^{+1})$ 4
98 $\mathrm{I}(2_{7}^{+1}7^{+2})$ 4 100 $\mathrm{II}(2_{\mathrm{II}}^{-2}5^{+2})$ 1 100 $\mathrm{I}(2_{2}^{-2}5^{-2})$ 2
100 $\mathrm{II}(2_{\mathrm{II}}^{+2}5^{-2})$ 1 108 $\mathrm{II}(2_{2}^{+2}3^{-1}9^{-1})$ 2 108 $\mathrm{II}(2_{6}^{+2}3^{-3})$ 1
108 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-1}9^{+1})$ 1 108 $\mathrm{I}(2_{2}^{+2}3^{+1}9^{+1})$ 2 108 $\mathrm{I}(2_{6}^{+2}3^{+3})$ 1
108 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-3})$ 1 108 $\mathrm{II}(2_{2}^{-2}3^{+1}9^{-1})$ 1 108 $\mathrm{II}(2_{2}^{-2}3^{+1}9^{+1})$ 1
108 $\mathrm{II}(2_{6}^{-2}3^{+3})$ 1 108 $\mathrm{I}(2_{\mathrm{II}}^{-2}3^{+1}9^{+1})$ 2 108 $\mathrm{I}(2_{4}^{-2}3^{-1}9^{-1})$ 2
108 $\mathrm{I}(2_{\mathrm{II}}^{-2}3^{+3})$ 1 108 $\mathrm{I}(2_{\mathrm{II}}^{-2}3^{-1}9^{-1})$ 1 108 $\mathrm{I}(2_{4}^{-2}3^{-3})$ 1
108 $\mathrm{I}(4_{1}^{+1}3^{+3})$ 2 108 $\mathrm{I}(4_{1}^{+1}3^{+1}9^{+1})$ 4 108 $\mathrm{I}(4_{5}^{-1}3^{-3})$ 1
108 $\mathrm{I}(4_{3}^{-1}3^{-3})$ 1 108 $\mathrm{I}(4_{3}^{-1}3^{+1}9^{-1})$ 2 108 $\mathrm{I}(4_{3}^{-1}3^{+1}9^{+1})$ 2
108 $\mathrm{I}(4_{3}^{-1}3^{+3})$ 2 112 $\mathrm{II}(4_{2}^{-2}7^{+1})$ 2 112 $\mathrm{I}(4_{\mathrm{II}}^{-2}7^{+1})$ 2
117 $\mathrm{II}(3^{+2}13^{-1})$ 2 120 $\mathrm{I}(2_{1}^{+3}3^{+1}5^{+1})$ 4 120 $\mathrm{II}(2_{1}^{+1}4_{1}^{+1}3^{-1}5^{-1})$ 4
125 $\mathrm{II}(5^{+3})$ 1 125 $\mathrm{I}(5^{+3})$ 1 125 $\mathrm{I}(5^{-3})$ 1
128 $\mathrm{I}(2_{3}^{-1}4_{1}^{+1}16_{3}^{-1})$ 2 128 $\mathrm{I}(2_{1}^{+1}4_{1}^{+1}16_{1}^{+1})$ 2 132 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+1}11^{-1})$ 2
132 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{-1}11^{+1})$ 2 135 $\mathrm{I}(3^{-3}5^{+1})$ 1 135 $\mathrm{I}(3^{+1}9^{-1}5^{+1})$ 4
135 $\mathrm{I}(3^{+3}5^{+1})$ 3 135 $\mathrm{I}(3^{-3}5^{-1})$ 2 144 $\mathrm{I}(2_{\mathrm{II}}^{-2}4_{1}^{+1}3^{+2})$ 1
144 $\mathrm{I}(2_{2}^{+2}4_{3}^{-1}3^{+2})$ 2 144 $\mathrm{I}(2_{\mathrm{II}}^{-2}4_{3}^{-1}3^{+2})$ 1 144 $\mathrm{II}(4_{\mathrm{II}}^{-2}3^{+2})$ 1
144 $\mathrm{II}(4_{4}^{-2}3^{-2})$ 2 144 $\mathrm{I}(4_{2}^{-2}3^{-2})$ 2 144 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-2})$ 2
144 $\mathrm{I}(4_{4}^{-2}3^{+2})$ 2 144 $\mathrm{I}(4_{2}^{-2}3^{+2})$ 2 144 $\mathrm{I}(4_{6}^{-2}3^{-2})$ 2
144 $\mathrm{II}(4_{\mathrm{II}}^{+2}3^{+2})$ 1 144 $\mathrm{II}(4_{0}^{+2}3^{+2})$ 2 144 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+2})$ 2
144 $\mathrm{I}(4_{2}^{+2}3^{+2})$ 2 160 $\mathrm{I}(2_{5}^{-1}4_{\mathrm{II}}^{-2}5^{-1})$ 2 162 $\mathrm{I}(2_{1}^{+1}3^{-2}9^{-1})$ 3
162 $\mathrm{I}(2_{1}^{+1}3^{+2}9^{+1})$ 2 162 $\mathrm{I}(2_{1}^{+1}3^{+2}9^{-1})$ 2 180 $\mathrm{II}(2_{\mathrm{II}}^{+2}3^{+2}5^{-1})$ 2
180 $\mathrm{II}(2_{\mathrm{II}}^{+2}3^{-2}5^{+1})$ 2 180 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+2}5^{+1})$ 2 180 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{-2}5^{-1})$ 2
189 $\mathrm{II}(3^{-3}7^{-1})$ 1 189 $\mathrm{II}(3^{-1}9^{-1}7^{+1})$ 3 189 $\mathrm{I}(3^{-3}7^{-1})$ 2
192 $\mathrm{I}(4_{3}^{+3}3^{+1})$ 2 192 $\mathrm{I}(2_{4}^{-2}16_{5}^{-1}3^{-1})$ 2 192 $\mathrm{I}(2_{\mathrm{II}}^{+2}16_{1}^{+1}3^{+1})$ 2
192 $\mathrm{I}(2_{0}^{+2}16_{1}^{+1}3^{+1})$ 3 192 $\mathrm{I}(2_{2}^{+2}16_{7}^{+1}3^{+1})$ 3 192 $\mathrm{I}(2_{6}^{+2}16_{7}^{+1}3^{-1})$ 2
192 $\mathrm{I}(2_{4}^{-2}16_{1}^{+1}3^{-1})$ 2 192 $\mathrm{I}(2_{\mathrm{II}}^{+2}16_{5}^{-1}3^{+1})$ 2 192 $\mathrm{I}(2_{0}^{+2}16_{5}^{-1}3^{+1})$ 3
192 $\mathrm{I}(2_{2}^{+2}16_{3}^{-1}3^{+1})$ 3 192 $\mathrm{I}(2_{6}^{+2}16_{3}^{-1}3^{-1})$ 2 192 $\mathrm{I}(4_{1}^{-3}3^{-1})$ 1
192 $\mathrm{I}(4_{5}^{-3}3^{+1})$ 2 192 $\mathrm{I}(4_{3}^{-3}3^{-1})$ 1 192 $\mathrm{II}(4_{1}^{+1}16_{1}^{+1}3^{-1})$ 2
196 $\mathrm{II}(2_{\mathrm{II}}^{-2}7^{-2})$ 2 216 $\mathrm{I}(2_{3}^{+3}3^{-3})$ 2 216 $\mathrm{I}(2_{7}^{+3}3^{+3})$ 1
216 $\mathrm{I}(2_{5}^{-3}3^{-1}9^{+1})$ 2 216 $\mathrm{II}(2_{3}^{-1}4_{3}^{-1}3^{+3})$ 2 216 $\mathrm{I}(2_{3}^{-1}4_{3}^{-1}3^{-3})$ 2
216 $\mathrm{II}(2_{3}^{-1}4_{7}^{+1}3^{-3})$ 1 216 $\mathrm{II}(2_{1}^{+1}4_{5}^{-1}3^{+1}9^{-1})$ 2 216 $\mathrm{I}(2_{3}^{-1}4_{1}^{+1}3^{+3})$ 2
224 $\mathrm{I}(2_{7}^{+1}4_{\mathrm{II}}^{-2}7^{+1})$ 3 225 $\mathrm{II}(3^{+2}5^{+2})$ 2 225 $\mathrm{I}(3^{+2}5^{-2})$ 2
240 $\mathrm{II}(4_{2}^{-2}3^{+1}5^{+1})$ 2 240 $\mathrm{II}(4_{2}^{-2}3^{-1}5^{-1})$ 3 240 $\mathrm{II}(4_{6}^{-2}3^{+1}5^{-1})$ 1
240 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}5^{+1})$ 2 240 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}5^{-1})$ 3 240 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}5^{-1})$ 1
243 $\mathrm{I}(3^{-1}9^{+2})$ 2 243 $\mathrm{I}(3^{+1}9^{-2})$ 1 243 $\mathrm{I}(3^{+1}9^{+2})$ 1
250 $\mathrm{I}(2_{5}^{-1}5^{+3})$ 2 252 $\mathrm{I}(2_{\mathrm{II}}^{-2}3^{+2}7^{+1})$ 3 252 $\mathrm{II}(2_{2}^{+2}3^{+2}7^{-1})$ 3
252 $\mathrm{II}(2_{6}^{+2}3^{+2}7^{+1})$ 3 252 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+2}7^{-1})$ 3 256 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{1}^{+1})$ 1
256 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{7}^{+1})$ 1 256 $\mathrm{I}(4_{2}^{+2}16_{5}^{-1})$ 2 256 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{3}^{-1})$ 1
256 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{5}^{-1})$ 1 256 $\mathrm{I}(4_{6}^{+2}16_{7}^{+1})$ 1 256 $\mathrm{I}(16_{6}^{-2})$ 2
256 $\mathrm{I}(16_{2}^{+2})$ 5 270 $\mathrm{I}(2_{1}^{+1}3^{-3}5^{+1})$ 4 288 $\mathrm{I}(2_{1}^{+1}4_{\mathrm{II}}^{-2}3^{-2})$ 2
288 $\mathrm{I}(2_{1}^{+1}4_{6}^{-2}3^{-2})$ 3 288 $\mathrm{I}(2_{1}^{+1}4_{6}^{+2}3^{+2})$ 3 288 $\mathrm{I}(2_{1}^{+1}4_{\mathrm{II}}^{+2}3^{+2})$ 2
297 $\mathrm{I}(3^{-3}11^{-1})$ 2 300 $\mathrm{II}(2_{2}^{+2}3^{-1}5^{+2})$ 3 300 $\mathrm{II}(2_{2}^{+2}3^{+1}5^{-2})$ 3
300 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-1}5^{-2})$ 3 300 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+1}5^{+2})$ 3 320 $\mathrm{I}(2_{2}^{+2}16_{1}^{+1}5^{+1})$ 6
320 $\mathrm{I}(2_{2}^{+2}16_{5}^{-1}5^{+1})$ 6 320 $\mathrm{I}(4_{7}^{-3}5^{+1})$ 2 320 $\mathrm{I}(4_{1}^{-3}5^{+1})$ 2
320 $\mathrm{II}(4_{7}^{+1}16_{1}^{+1}5^{+1})$ 5 320 $\mathrm{I}(4_{1}^{+1}16_{1}^{+1}5^{+1})$ 10 324 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{-2}9^{+1})$ 1
324 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{-2}9^{-1})$ 1 324 $\mathrm{II}(2_{\mathrm{II}}^{-2}9^{-2})$ 1 324 $\mathrm{II}(2_{\mathrm{II}}^{-2}9^{+2})$ 1
324 $\mathrm{II}(2_{4}^{-2}3^{+2}9^{-1})$ 2 324 $\mathrm{I}(2_{\mathrm{II}}^{-2}3^{+2}9^{+1})$ 2 324 $\mathrm{I}(2_{4}^{-2}3^{-2}9^{+1})$ 2
324 $\mathrm{I}(2_{4}^{-2}3^{-2}9^{-1})$ 2 324 $\mathrm{II}(2_{\mathrm{II}}^{+2}3^{+2}9^{-1})$ 1 324 $\mathrm{II}(2_{\mathrm{II}}^{+2}3^{+2}9^{+1})$ 1
324 $\mathrm{I}(2_{2}^{+2}3^{+2}9^{-1})$ 2 324 $\mathrm{I}(2_{2}^{+2}3^{+2}9^{+1})$ 2 324 $\mathrm{I}(4_{3}^{-1}3^{-2}9^{+1})$ 2
324 $\mathrm{I}(4_{3}^{-1}3^{-2}9^{-1})$ 2 324 $\mathrm{I}(4_{5}^{-1}3^{+2}9^{+1})$ 3 324 $\mathrm{I}(4_{7}^{+1}3^{+2}9^{+1})$ 3
324 $\mathrm{I}(4_{1}^{+1}3^{-2}9^{+1})$ 2 324 $\mathrm{I}(4_{1}^{+1}3^{-2}9^{-1})$ 2 343 $\mathrm{I}(7^{+3})$ 2
360 $\mathrm{I}(2_{7}^{-3}3^{+2}5^{+1})$ 4 360 $\mathrm{I}(2_{5}^{-3}3^{+2}5^{-1})$ 4 360 $\mathrm{II}(2_{1}^{+1}4_{7}^{+1}3^{+2}5^{-1})$ 4
360 $\mathrm{II}(2_{7}^{+1}4_{5}^{-1}3^{+2}5^{+1})$ 4 375 $\mathrm{I}(3^{+1}5^{+3})$ 3 375 $\mathrm{I}(3^{-1}5^{-3})$ 2
375 $\mathrm{I}(3^{-1}5^{+3})$ 1 384 $\mathrm{I}(2_{1}^{+1}4_{1}^{+1}16_{7}^{+1}3^{+1})$ 6 384 $\mathrm{I}(2_{7}^{+1}4_{1}^{+1}16_{5}^{-1}3^{+1})$ 6
392 $\mathrm{I}(2_{7}^{+3}7^{+2})$ 4 392 $\mathrm{II}(2_{7}^{+1}4_{1}^{+1}7^{+2})$ 4 400 $\mathrm{II}(4_{\mathrm{II}}^{-2}5^{-2})$ 2
400 $\mathrm{I}(4_{6}^{-2}5^{-2})$ 2 400 $\mathrm{I}(4_{2}^{-2}5^{-2})$ 2 405 $\mathrm{II}(3^{-2}9^{-1}5^{+1})$ 2
405 $\mathrm{II}(3^{+2}9^{+1}5^{-1})$ 1 405 $\mathrm{II}(3^{+2}9^{-1}5^{-1})$ 1 405 $\mathrm{I}(3^{-2}9^{-1}5^{+1})$ 3
405 $\mathrm{I}(3^{+2}9^{+1}5^{-1})$ 2 405 $\mathrm{I}(3^{+2}9^{-1}5^{-1})$ 2 420 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+1}5^{-1}7^{+1})$ 3
420 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{-1}5^{+1}7^{+1})$ 3 432 $\mathrm{I}(2_{\mathrm{II}}^{-2}4_{3}^{-1}3^{+3})$ 1 432 $\mathrm{I}(2_{4}^{-2}4_{3}^{-1}3^{-3})$ 1
432 $\mathrm{I}(2_{\mathrm{II}}^{-2}4_{1}^{+1}3^{+1}9^{+1})$ 3 432 $\mathrm{I}(2_{2}^{+2}4_{3}^{-1}3^{+1}9^{+1})$ 4 432 $\mathrm{I}(2_{6}^{+2}4_{3}^{-1}3^{+3})$ 2
432 $\mathrm{II}(4_{2}^{+2}3^{+3})$ 1 432 $\mathrm{II}(4_{2}^{+2}3^{-1}9^{-1})$ 2 432 $\mathrm{II}(4_{6}^{+2}3^{-3})$ 1
432 $\mathrm{II}(4_{6}^{+2}3^{+1}9^{-1})$ 1 432 $\mathrm{II}(4_{6}^{+2}3^{+1}9^{+1})$ 1 432 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+3})$ 1
432 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{-1}9^{-1})$ 2 432 $\mathrm{I}(4_{2}^{+2}3^{+1}9^{+1})$ 4 432 $\mathrm{I}(4_{6}^{+2}3^{+3})$ 2
432 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{-3})$ 1 432 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+1}9^{+1})$ 2 432 $\mathrm{I}(4_{0}^{+2}3^{-3})$ 1
432 $\mathrm{II}(4_{2}^{-2}3^{+3})$ 1 432 $\mathrm{II}(4_{2}^{-2}3^{-1}9^{-1})$ 2 432 $\mathrm{II}(4_{6}^{-2}3^{-3})$ 1
432 $\mathrm{II}(4_{6}^{-2}3^{+1}9^{+1})$ 2 432 $\mathrm{I}(4_{6}^{-2}3^{-3})$ 1 432 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+3})$ 1
432 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}9^{-1})$ 2 432 $\mathrm{I}(4_{4}^{-2}3^{+1}9^{+1})$ 4 432 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-3})$ 1
432 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{-1})$ 1 432 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{+1})$ 1 432 $\mathrm{I}(4_{4}^{-2}3^{+3})$ 2
432 $\mathrm{I}(16_{7}^{+1}3^{-3})$ 2 432 $\mathrm{I}(16_{3}^{-1}3^{-3})$ 2 441 $\mathrm{I}(3^{+2}7^{+2})$ 4
448 $\mathrm{I}(4_{3}^{+3}7^{+1})$ 3 480 $\mathrm{I}(2_{1}^{+1}4_{\mathrm{II}}^{+2}3^{-1}5^{-1})$ 4 484 $\mathrm{II}(2_{\mathrm{II}}^{-2}11^{-2})$ 1
486 $\mathrm{I}(2_{1}^{+1}3^{+1}9^{+2})$ 2 500 $\mathrm{II}(2_{\mathrm{II}}^{+2}5^{+3})$ 1 500 $\mathrm{I}(2_{2}^{+2}5^{-3})$ 2
500 $\mathrm{II}(2_{\mathrm{II}}^{-2}5^{-3})$ 1 500 $\mathrm{I}(2_{2}^{-2}5^{+3})$ 2 500 $\mathrm{I}(4_{7}^{+1}5^{+3})$ 2
500 $\mathrm{I}(4_{5}^{-1}5^{+3})$ 2 512 $\mathrm{I}(2_{3}^{-1}16_{0}^{+2})$ 2 512 $\mathrm{I}(2_{1}^{+1}16_{6}^{-2})$ 1
512 $\mathrm{I}(2_{5}^{-1}16_{2}^{-2})$ 2 512 $\mathrm{I}(2_{1}^{+1}16_{2}^{+2})$ 2 512 $\mathrm{I}(2_{5}^{-1}16_{4}^{-2})$ 2
512 $\mathrm{I}(2_{7}^{+1}16_{6}^{+2})$ 1 540 $\mathrm{II}(2_{6}^{-2}3^{-3}5^{+1})$ 1 540 $\mathrm{I}(2_{\mathrm{II}}^{-2}3^{-3}5^{-1})$ 3
540 $\mathrm{II}(2_{2}^{+2}3^{+1}9^{-1}5^{+1})$ 4 540 $\mathrm{II}(2_{6}^{+2}3^{+3}5^{+1})$ 3 540 $\mathrm{II}(2_{6}^{+2}3^{-3}5^{-1})$ 2
540 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+3}5^{-1})$ 1 540 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-1}9^{+1}5^{-1})$ 4 540 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+3}5^{+1})$ 2
540 $\mathrm{I}(4_{3}^{-1}3^{-3}5^{+1})$ 2 540 $\mathrm{I}(4_{1}^{+1}3^{-3}5^{-1})$ 3 567 $\mathrm{I}(3^{+2}9^{-1}7^{-1})$ 4
576 $\mathrm{I}(2_{\mathrm{II}}^{-2}16_{1}^{+1}3^{+2})$ 2 576 $\mathrm{I}(2_{2}^{-2}16_{7}^{+1}3^{+2})$ 3 576 $\mathrm{I}(2_{0}^{+2}16_{5}^{-1}3^{+2})$ 3
576 $\mathrm{I}(2_{6}^{-2}16_{1}^{+1}3^{+2})$ 3 576 $\mathrm{I}(2_{\mathrm{II}}^{+2}16_{3}^{-1}3^{+2})$ 2 576 $\mathrm{I}(2_{0}^{+2}16_{3}^{-1}3^{+2})$ 3
576 $\mathrm{I}(2_{\mathrm{II}}^{+2}16_{7}^{+1}3^{+2})$ 2 576 $\mathrm{I}(4_{3}^{+3}3^{-2})$ 2 576 $\mathrm{I}(4_{7}^{+3}3^{+2})$ 2
576 $\mathrm{I}(2_{\mathrm{II}}^{-2}16_{5}^{-1}3^{+2})$ 2 576 $\mathrm{I}(4_{1}^{+3}3^{+2})$ 2 576 $\mathrm{I}(4_{5}^{+3}3^{-2})$ 2
625 $\mathrm{I}(5^{+2}25^{-1})$ 2 648 $\mathrm{I}(2_{1}^{-3}3^{+2}9^{+1})$ 2 648 $\mathrm{I}(2_{1}^{-3}3^{+2}9^{-1})$ 2
648 $\mathrm{I}(2_{3}^{+3}3^{-2}9^{+1})$ 3 648 $\mathrm{II}(2_{1}^{+1}4_{3}^{-1}3^{-2}9^{-1})$ 3 648 $\mathrm{I}(2_{1}^{+1}4_{3}^{-1}3^{+2}9^{+1})$ 3
648 $\mathrm{I}(2_{1}^{+1}4_{3}^{-1}3^{+2}9^{-1})$ 3 648 $\mathrm{II}(2_{1}^{+1}4_{7}^{+1}3^{+2}9^{+1})$ 2 648 $\mathrm{II}(2_{1}^{+1}4_{7}^{+1}3^{+2}9^{-1})$ 2
675 $\mathrm{I}(3^{-3}5^{+2})$ 3 675 $\mathrm{I}(3^{+3}5^{-2})$ 3 686 $\mathrm{I}(2_{7}^{+1}7^{+3})$ 3
729 $\mathrm{II}(3^{-1}9^{-1}27^{-1})$ 3 729 $\mathrm{I}(3^{+1}9^{+1}27^{-1})$ 1 729 $\mathrm{I}(9^{-3})$ 1
750 $\mathrm{I}(2_{5}^{-1}3^{+1}5^{-3})$ 4 756 $\mathrm{II}(2_{\mathrm{II}}^{+2}3^{-3}7^{-1})$ 2 756 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+3}7^{-1})$ 2
768 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{3}^{-1}3^{+1})$ 2 768 $\mathrm{I}(4_{\mathrm{II}}^{+2}16_{7}^{+1}3^{-1})$ 1 768 $\mathrm{I}(4_{2}^{+2}16_{1}^{+1}3^{+1})$ 3
768 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{5}^{-1}3^{-1})$ 1 768 $\mathrm{I}(4_{\mathrm{II}}^{+2}16_{1}^{+1}3^{+1})$ 2 768 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{1}^{+1}3^{-1})$ 1
768 $\mathrm{I}(4_{2}^{-2}16_{7}^{+1}3^{-1})$ 2 768 $\mathrm{I}(4_{\mathrm{II}}^{+2}16_{5}^{-1}3^{+1})$ 2 768 $\mathrm{I}(4_{2}^{+2}16_{3}^{-1}3^{+1})$ 3
768 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{7}^{+1}3^{+1})$ 2 768 $\mathrm{I}(4_{2}^{-2}16_{1}^{+1}3^{-1})$ 2 768 $\mathrm{I}(4_{\mathrm{II}}^{+2}16_{3}^{-1}3^{-1})$ 1
784 $\mathrm{II}(4_{\mathrm{II}}^{-2}7^{+2})$ 1 810 $\mathrm{I}(2_{7}^{+1}3^{+2}9^{-1}5^{+1})$ 7 864 $\mathrm{I}(2_{3}^{-1}4_{\mathrm{II}}^{-2}3^{+3})$ 2
864 $\mathrm{I}(2_{3}^{-1}4_{4}^{-2}3^{-3})$ 2 864 $\mathrm{I}(2_{3}^{-1}4_{6}^{-2}3^{+3})$ 2 864 $\mathrm{I}(2_{1}^{+1}4_{\mathrm{II}}^{+2}3^{+1}9^{-1})$ 2
864 $\mathrm{I}(2_{3}^{-1}4_{\mathrm{II}}^{+2}3^{-3})$ 1 864 $\mathrm{I}(2_{3}^{-1}16_{3}^{-1}3^{-3})$ 3 864 $\mathrm{I}(2_{1}^{+1}16_{1}^{+1}3^{-3})$ 4
864 $\mathrm{I}(2_{3}^{-1}16_{7}^{+1}3^{-3})$ 4 864 $\mathrm{I}(2_{1}^{+1}16_{5}^{-1}3^{-3})$ 3 900 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{-2}5^{+2})$ 2
900 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+2}5^{-2})$ 2 900 $\mathrm{I}(2_{4}^{-2}3^{+2}5^{-2})$ 4 900 $\mathrm{I}(4_{3}^{-1}3^{+2}5^{-2})$ 4
900 $\mathrm{I}(4_{5}^{-1}3^{+2}5^{-2})$ 4 960 $\mathrm{I}(4_{7}^{-3}3^{+1}5^{+1})$ 3 960 $\mathrm{I}(4_{5}^{+3}3^{+1}5^{-1})$ 2
972 $\mathrm{II}(2_{2}^{+2}3^{-1}9^{+2})$ 2 972 $\mathrm{II}(2_{6}^{+2}3^{+1}9^{+2})$ 1 972 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-1}9^{+2})$ 1
972 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-1}9^{-2})$ 1 972 $\mathrm{I}(2_{2}^{+2}3^{+1}9^{+2})$ 2 972 $\mathrm{II}(2_{2}^{-2}3^{+1}9^{-2})$ 1
972 $\mathrm{I}(2_{\mathrm{II}}^{-2}3^{+1}9^{+2})$ 2 972 $\mathrm{I}(2_{4}^{-2}3^{-1}9^{+2})$ 2 972 $\mathrm{I}(4_{5}^{-1}3^{+1}9^{+2})$ 4
972 $\mathrm{I}(4_{3}^{-1}3^{+1}9^{-2})$ 2 972 $\mathrm{I}(4_{3}^{-1}3^{+1}9^{+2})$ 2 1000 $\mathrm{I}(2_{7}^{+3}5^{-3})$ 2
1000 $\mathrm{II}(2_{5}^{-1}4_{7}^{+1}5^{+3})$ 2 1008 $\mathrm{II}(4_{2}^{-2}3^{+2}7^{-1})$ 3 1008 $\mathrm{II}(4_{6}^{-2}3^{+2}7^{+1})$ 3
1008 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}7^{-1})$ 3 1008 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}7^{+1})$ 3 1024 $\mathrm{I}(4_{\mathrm{II}}^{-2}64_{1}^{+1})$ 3
1024 $\mathrm{I}(4_{3}^{-1}16_{\mathrm{II}}^{-2})$ 2 1024 $\mathrm{I}(4_{1}^{+1}16_{2}^{+2})$ 3 1029 $\mathrm{II}(3^{-1}7^{-3})$ 1
1029 $\mathrm{I}(3^{-1}7^{-3})$ 2 1053 $\mathrm{II}(3^{+2}9^{-1}13^{-1})$ 4 1080 $\mathrm{I}(2_{1}^{+3}3^{+3}5^{-1})$ 4
1080 $\mathrm{II}(2_{1}^{+1}4_{1}^{+1}3^{-3}5^{+1})$ 4 1125 $\mathrm{II}(3^{+2}5^{-3})$ 1 1125 $\mathrm{II}(3^{-2}5^{+3})$ 1
1125 $\mathrm{I}(3^{+2}5^{-3})$ 2 1125 $\mathrm{I}(3^{-2}5^{+3})$ 2 1188 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+3}11^{+1})$ 2
1188 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{-3}11^{-1})$ 2 1200 $\mathrm{II}(4_{2}^{+2}3^{-1}5^{+2})$ 3 1200 $\mathrm{II}(4_{2}^{+2}3^{+1}5^{-2})$ 3
1200 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}5^{+2})$ 3 1200 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}5^{-2})$ 3 1215 $\mathrm{I}(3^{+1}9^{+2}5^{+1})$ 4
1280 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{5}^{-1}5^{+1})$ 2 1280 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{3}^{-1}5^{+1})$ 2 1280 $\mathrm{I}(4_{6}^{-2}16_{5}^{-1}5^{+1})$ 6
1280 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{7}^{+1}5^{+1})$ 2 1280 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{1}^{+1}5^{+1})$ 2 1280 $\mathrm{I}(16_{2}^{+2}5^{+1})$ 10
1280 $\mathrm{I}(16_{6}^{-2}5^{+1})$ 10 1296 $\mathrm{I}(2_{\mathrm{II}}^{-2}4_{1}^{+1}3^{+2}9^{+1})$ 2 1296 $\mathrm{I}(2_{2}^{+2}4_{3}^{-1}3^{+2}9^{+1})$ 3
1296 $\mathrm{I}(2_{\mathrm{II}}^{-2}4_{3}^{-1}3^{+2}9^{+1})$ 2 1296 $\mathrm{II}(4_{\mathrm{II}}^{-2}3^{+2}9^{-1})$ 2 1296 $\mathrm{II}(4_{\mathrm{II}}^{-2}3^{+2}9^{+1})$ 1
1296 $\mathrm{II}(4_{4}^{-2}3^{-2}9^{+1})$ 2 1296 $\mathrm{II}(4_{4}^{-2}3^{-2}9^{-1})$ 2 1296 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-2}9^{+1})$ 2
1296 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-2}9^{-1})$ 2 1296 $\mathrm{I}(4_{4}^{-2}3^{+2}9^{+1})$ 4 1296 $\mathrm{I}(4_{2}^{-2}3^{+2}9^{-1})$ 2
1296 $\mathrm{I}(4_{2}^{-2}3^{+2}9^{+1})$ 2 1296 $\mathrm{I}(4_{6}^{-2}3^{-2}9^{+1})$ 2 1296 $\mathrm{I}(4_{6}^{-2}3^{-2}9^{-1})$ 2
1296 $\mathrm{II}(4_{\mathrm{II}}^{+2}3^{+2}9^{-1})$ 2 1296 $\mathrm{II}(4_{0}^{+2}3^{+2}9^{+1})$ 3 1296 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+2}9^{+1})$ 3
1296 $\mathrm{I}(4_{2}^{+2}3^{+2}9^{-1})$ 2 1296 $\mathrm{I}(4_{2}^{+2}3^{+2}9^{+1})$ 2 1296 $\mathrm{I}(4_{6}^{+2}3^{-2}9^{+1})$ 2
1296 $\mathrm{I}(4_{6}^{+2}3^{-2}9^{-1})$ 2 1372 $\mathrm{II}(2_{6}^{+2}7^{+3})$ 2 1372 $\mathrm{I}(2_{\mathrm{II}}^{+2}7^{+3})$ 2
1372 $\mathrm{I}(4_{5}^{-1}7^{+3})$ 3 1440 $\mathrm{I}(2_{7}^{+1}4_{\mathrm{II}}^{+2}3^{+2}5^{+1})$ 4 1440 $\mathrm{I}(2_{1}^{+1}4_{\mathrm{II}}^{-2}3^{+2}5^{-1})$ 4
1500 $\mathrm{I}(2_{\mathrm{II}}^{-2}3^{+1}5^{-3})$ 1 1500 $\mathrm{I}(2_{\mathrm{II}}^{-2}3^{-1}5^{-3})$ 3 1500 $\mathrm{II}(2_{2}^{+2}3^{+1}5^{+3})$ 3
1500 $\mathrm{II}(2_{2}^{+2}3^{-1}5^{-3})$ 2 1500 $\mathrm{II}(2_{6}^{+2}3^{-1}5^{+3})$ 1 1500 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+1}5^{+3})$ 2
1500 $\mathrm{I}(4_{7}^{+1}3^{-1}5^{-3})$ 3 1500 $\mathrm{I}(4_{1}^{+1}3^{-1}5^{+3})$ 2 1536 $\mathrm{I}(2_{3}^{-1}16_{2}^{-2}3^{+1})$ 4
1536 $\mathrm{I}(2_{1}^{+1}16_{0}^{+2}3^{+1})$ 6 1536 $\mathrm{I}(2_{7}^{+1}16_{2}^{+2}3^{+1})$ 4 1536 $\mathrm{I}(2_{1}^{+1}16_{4}^{-2}3^{+1})$ 6
1536 $\mathrm{I}(2_{7}^{+1}16_{6}^{-2}3^{+1})$ 3 1536 $\mathrm{I}(2_{5}^{-1}16_{6}^{+2}3^{+1})$ 3 1568 $\mathrm{I}(2_{7}^{+1}4_{\mathrm{II}}^{+2}7^{+2})$ 4
1620 $\mathrm{II}(2_{\mathrm{II}}^{+2}3^{-2}9^{-1}5^{+1})$ 3 1620 $\mathrm{II}(2_{\mathrm{II}}^{+2}3^{+2}9^{+1}5^{-1})$ 2 1620 $\mathrm{II}(2_{\mathrm{II}}^{+2}3^{+2}9^{-1}5^{-1})$ 2
1620 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+2}9^{+1}5^{+1})$ 2 1620 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{-2}9^{+1}5^{-1})$ 3 1620 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+2}9^{-1}5^{+1})$ 2
1701 $\mathrm{II}(3^{-1}9^{+2}7^{+1})$ 3 1728 $\mathrm{I}(2_{2}^{-2}16_{5}^{-1}3^{-3})$ 2 1728 $\mathrm{I}(2_{4}^{-2}16_{3}^{-1}3^{-3})$ 2
1728 $\mathrm{I}(2_{\mathrm{II}}^{+2}16_{7}^{+1}3^{+3})$ 2 1728 $\mathrm{I}(2_{2}^{+2}16_{1}^{+1}3^{-3})$ 2 1728 $\mathrm{I}(4_{3}^{+3}3^{-3})$ 1
1728 $\mathrm{I}(4_{3}^{+3}3^{+1}9^{-1})$ 2 1728 $\mathrm{I}(4_{3}^{+3}3^{+1}9^{+1})$ 2 1728 $\mathrm{I}(4_{7}^{+3}3^{+3})$ 2
1728 $\mathrm{I}(2_{4}^{-2}16_{5}^{-1}3^{+3})$ 3 1728 $\mathrm{I}(2_{\mathrm{II}}^{+2}16_{1}^{+1}3^{+1}9^{+1})$ 6 1728 $\mathrm{I}(2_{6}^{+2}16_{7}^{+1}3^{+3})$ 3
1728 $\mathrm{I}(4_{1}^{+3}3^{-3})$ 1 1728 $\mathrm{I}(2_{4}^{-2}16_{1}^{+1}3^{+3})$ 3 1728 $\mathrm{I}(2_{\mathrm{II}}^{+2}16_{5}^{-1}3^{+1}9^{+1})$ 6
1728 $\mathrm{I}(2_{6}^{+2}16_{3}^{-1}3^{+3})$ 3 1728 $\mathrm{I}(4_{1}^{-3}3^{+3})$ 2 1728 $\mathrm{I}(4_{5}^{-3}3^{+1}9^{+1})$ 4
1728 $\mathrm{I}(2_{4}^{-2}16_{7}^{+1}3^{-3})$ 2 1728 $\mathrm{I}(2_{\mathrm{II}}^{+2}16_{3}^{-1}3^{+3})$ 2 1728 $\mathrm{II}(4_{3}^{-1}16_{3}^{-1}3^{-3})$ 2
1764 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+2}7^{+2})$ 2 1792 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{7}^{+1}7^{+1})$ 3 1792 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{3}^{-1}7^{+1})$ 3
1875 $\mathrm{I}(3^{+1}5^{-2}25^{-1})$ 5 1944 $\mathrm{I}(2_{5}^{-3}3^{-1}9^{+2})$ 2 1944 $\mathrm{II}(2_{1}^{+1}4_{5}^{-1}3^{+1}9^{+2})$ 2
2000 $\mathrm{I}(2_{2}^{-2}4_{5}^{-1}5^{+3})$ 3 2000 $\mathrm{II}(4_{0}^{+2}5^{+3})$ 2 2000 $\mathrm{I}(4_{2}^{+2}5^{-3})$ 2
2000 $\mathrm{I}(4_{2}^{+2}5^{+3})$ 2 2000 $\mathrm{I}(4_{6}^{+2}5^{-3})$ 2 2000 $\mathrm{I}(4_{\mathrm{II}}^{-2}5^{+3})$ 2
2000 $\mathrm{I}(4_{2}^{-2}5^{+3})$ 2 2000 $\mathrm{I}(16_{1}^{+1}5^{+3})$ 5 2000 $\mathrm{I}(16_{5}^{-1}5^{+3})$ 5
2160 $\mathrm{II}(4_{2}^{-2}3^{-3}5^{+1})$ 1 2160 $\mathrm{II}(4_{2}^{-2}3^{+1}9^{-1}5^{+1})$ 4 2160 $\mathrm{II}(4_{6}^{-2}3^{+3}5^{+1})$ 3
2160 $\mathrm{II}(4_{6}^{-2}3^{-3}5^{-1})$ 2 2160 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-3}5^{+1})$ 1 2160 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{-1}5^{+1})$ 4
2160 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+3}5^{+1})$ 3 2160 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-3}5^{-1})$ 2 2250 $\mathrm{I}(2_{7}^{+1}3^{+2}5^{+3})$ 4
2250 $\mathrm{I}(2_{5}^{-1}3^{+2}5^{-3})$ 4 2268 $\mathrm{II}(2_{2}^{+2}3^{+2}9^{-1}7^{-1})$ 4 2268 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+2}9^{+1}7^{-1})$ 4
2304 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{1}^{+1}3^{-2})$ 2 2304 $\mathrm{I}(4_{\mathrm{II}}^{+2}16_{5}^{-1}3^{+2})$ 2 2304 $\mathrm{I}(4_{2}^{+2}16_{3}^{-1}3^{+2})$ 3
2304 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{7}^{+1}3^{-2})$ 2 2304 $\mathrm{I}(4_{\mathrm{II}}^{+2}16_{3}^{-1}3^{+2})$ 2 2304 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{3}^{-1}3^{-2})$ 2
2304 $\mathrm{I}(4_{\mathrm{II}}^{+2}16_{7}^{+1}3^{+2})$ 2 2304 $\mathrm{I}(4_{6}^{+2}16_{1}^{+1}3^{+2})$ 3 2304 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{5}^{-1}3^{-2})$ 2
2304 $\mathrm{I}(4_{\mathrm{II}}^{+2}16_{1}^{+1}3^{+2})$ 2 2500 $\mathrm{II}(2_{\mathrm{II}}^{-2}5^{+2}25^{-1})$ 1 2500 $\mathrm{II}(2_{\mathrm{II}}^{-2}5^{+2}25^{+1})$ 1
2592 $\mathrm{I}(2_{1}^{+1}4_{\mathrm{II}}^{-2}3^{-2}9^{-1})$ 3 2592 $\mathrm{I}(2_{1}^{+1}4_{4}^{-2}3^{+2}9^{+1})$ 3 2592 $\mathrm{I}(2_{1}^{+1}4_{6}^{+2}3^{+2}9^{-1})$ 3
2592 $\mathrm{I}(2_{1}^{+1}4_{\mathrm{II}}^{+2}3^{+2}9^{+1})$ 2 2592 $\mathrm{I}(2_{1}^{+1}4_{\mathrm{II}}^{+2}3^{+2}9^{-1})$ 2 2700 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-3}5^{-2})$ 3
2700 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+3}5^{+2})$ 3 2700 $\mathrm{II}(2_{2}^{-2}3^{-3}5^{+2})$ 3 2700 $\mathrm{II}(2_{2}^{-2}3^{+3}5^{-2})$ 3
2744 $\mathrm{I}(2_{7}^{+3}7^{+3})$ 3 2744 $\mathrm{II}(2_{7}^{+1}4_{7}^{+1}7^{+3})$ 3 2916 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{-1}9^{+1}27^{+1})$ 2
2916 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+1}9^{+1}27^{-1})$ 1 2916 $\mathrm{II}(2_{\mathrm{II}}^{-2}9^{-3})$ 1 2916 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{-1}9^{-1}27^{+1})$ 1
2916 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+1}9^{-1}27^{-1})$ 2 2916 $\mathrm{II}(2_{\mathrm{II}}^{-2}9^{+3})$ 1 3000 $\mathrm{I}(2_{5}^{-3}3^{-1}5^{+3})$ 4
3000 $\mathrm{II}(2_{5}^{-1}4_{5}^{-1}3^{+1}5^{-3})$ 4 3072 $\mathrm{I}(4_{1}^{+1}16_{\mathrm{II}}^{-2}3^{-1})$ 2 3072 $\mathrm{I}(4_{\mathrm{II}}^{-2}64_{1}^{+1}3^{-1})$ 4
3087 $\mathrm{I}(3^{+2}7^{+3})$ 3 3087 $\mathrm{I}(3^{+2}7^{-3})$ 3 3240 $\mathrm{I}(2_{5}^{-3}3^{+2}9^{+1}5^{-1})$ 7
3240 $\mathrm{II}(2_{7}^{+1}4_{5}^{-1}3^{+2}9^{-1}5^{+1})$ 7 3375 $\mathrm{I}(3^{+3}5^{-3})$ 1 3375 $\mathrm{I}(3^{-1}9^{+1}5^{-3})$ 4
3375 $\mathrm{I}(3^{+3}5^{+3})$ 2 3375 $\mathrm{I}(3^{-3}5^{-3})$ 3 3456 $\mathrm{I}(2_{3}^{-1}4_{3}^{-1}16_{7}^{+1}3^{-3})$ 6
3456 $\mathrm{I}(2_{1}^{+1}4_{3}^{-1}16_{1}^{+1}3^{-3})$ 6 3600 $\mathrm{II}(4_{\mathrm{II}}^{-2}3^{+2}5^{+2})$ 2 3600 $\mathrm{II}(4_{4}^{-2}3^{+2}5^{-2})$ 4
3600 $\mathrm{I}(4_{2}^{-2}3^{+2}5^{-2})$ 4 3600 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}5^{-2})$ 4 3600 $\mathrm{I}(4_{6}^{-2}3^{+2}5^{-2})$ 4
3780 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+3}5^{-1}7^{-1})$ 3 3780 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{-3}5^{+1}7^{-1})$ 3 3840 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{7}^{+1}3^{+1}5^{+1})$ 3
3840 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{1}^{+1}3^{+1}5^{-1})$ 2 3840 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{5}^{-1}3^{+1}5^{-1})$ 2 3840 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{3}^{-1}3^{+1}5^{+1})$ 3
3888 $\mathrm{I}(2_{\mathrm{II}}^{-2}4_{1}^{+1}3^{+1}9^{+2})$ 3 3888 $\mathrm{I}(2_{2}^{+2}4_{3}^{-1}3^{+1}9^{+2})$ 4 3888 $\mathrm{II}(4_{2}^{+2}3^{-1}9^{+2})$ 2
3888 $\mathrm{II}(4_{6}^{+2}3^{+1}9^{-2})$ 1 3888 $\mathrm{II}(4_{6}^{+2}3^{+1}9^{+2})$ 1 3888 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{-1}9^{+2})$ 2
3888 $\mathrm{I}(4_{2}^{+2}3^{+1}9^{+2})$ 4 3888 $\mathrm{I}(4_{\mathrm{II}}^{+2}3^{+1}9^{+2})$ 2 3888 $\mathrm{II}(4_{2}^{-2}3^{-1}9^{+2})$ 2
3888 $\mathrm{II}(4_{6}^{-2}3^{+1}9^{+2})$ 2 3888 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}9^{+2})$ 2 3888 $\mathrm{I}(4_{4}^{-2}3^{+1}9^{+2})$ 4
3888 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{-2})$ 1 3888 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{+2})$ 1 3969 $\mathrm{I}(3^{+2}9^{-1}7^{+2})$ 4
3993 $\mathrm{I}(3^{-1}11^{-3})$ 2 4000 $\mathrm{I}(2_{5}^{-1}4_{\mathrm{II}}^{+2}5^{+3})$ 2 4096 $\mathrm{I}(16_{7}^{-3})$ 2
4096 $\mathrm{I}(16_{3}^{+3})$ 2 4116 $\mathrm{II}(2_{\mathrm{II}}^{+2}3^{-1}7^{-3})$ 2 4116 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+1}7^{-3})$ 2
4320 $\mathrm{I}(2_{1}^{+1}4_{\mathrm{II}}^{+2}3^{-3}5^{+1})$ 4 4500 $\mathrm{II}(2_{\mathrm{II}}^{+2}3^{+2}5^{-3})$ 2 4500 $\mathrm{II}(2_{\mathrm{II}}^{+2}3^{-2}5^{+3})$ 2
4500 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+2}5^{+3})$ 2 4500 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{-2}5^{-3})$ 2 4860 $\mathrm{II}(2_{2}^{+2}3^{+1}9^{+2}5^{+1})$ 4
4860 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-1}9^{+2}5^{-1})$ 4 4900 $\mathrm{II}(2_{\mathrm{II}}^{-2}5^{-2}7^{+2})$ 2 5120 $\mathrm{I}(4_{1}^{+1}16_{2}^{+2}5^{+1})$ 10
5120 $\mathrm{I}(4_{7}^{+1}16_{\mathrm{II}}^{-2}5^{+1})$ 5 5184 $\mathrm{I}(2_{2}^{-2}16_{7}^{+1}3^{+2}9^{+1})$ 4 5184 $\mathrm{I}(2_{2}^{+2}16_{3}^{-1}3^{+2}9^{+1})$ 4
5184 $\mathrm{I}(4_{1}^{-3}3^{-2}9^{+1})$ 2 5184 $\mathrm{I}(4_{1}^{-3}3^{-2}9^{-1})$ 2 5184 $\mathrm{I}(2_{\mathrm{II}}^{+2}16_{3}^{-1}3^{+2}9^{+1})$ 4
5184 $\mathrm{I}(2_{\mathrm{II}}^{+2}16_{7}^{+1}3^{+2}9^{+1})$ 4 5184 $\mathrm{I}(4_{3}^{+3}3^{-2}9^{+1})$ 2 5184 $\mathrm{I}(4_{3}^{+3}3^{-2}9^{-1})$ 2
5184 $\mathrm{I}(4_{7}^{+3}3^{+2}9^{+1})$ 3 5184 $\mathrm{I}(2_{\mathrm{II}}^{-2}16_{5}^{-1}3^{+2}9^{+1})$ 4 5184 $\mathrm{I}(2_{\mathrm{II}}^{+2}16_{1}^{+1}3^{+2}9^{+1})$ 4
5184 $\mathrm{I}(4_{1}^{+3}3^{+2}9^{+1})$ 3 5488 $\mathrm{II}(4_{6}^{-2}7^{+3})$ 2 5488 $\mathrm{I}(4_{\mathrm{II}}^{-2}7^{+3})$ 2
6000 $\mathrm{II}(4_{2}^{-2}3^{+1}5^{+3})$ 3 6000 $\mathrm{II}(4_{2}^{-2}3^{-1}5^{-3})$ 2 6000 $\mathrm{II}(4_{6}^{-2}3^{-1}5^{+3})$ 1
6000 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}5^{+3})$ 3 6000 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}5^{-3})$ 2 6000 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}5^{+3})$ 1
6084 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+2}13^{-2})$ 2 6750 $\mathrm{I}(2_{7}^{+1}3^{+3}5^{+3})$ 4 6912 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{3}^{-1}3^{-3})$ 1
6912 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{3}^{-1}3^{+1}9^{-1})$ 2 6912 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{3}^{-1}3^{+1}9^{+1})$ 2 6912 $\mathrm{I}(4_{2}^{-2}16_{5}^{-1}3^{+3})$ 3
6912 $\mathrm{I}(4_{\mathrm{II}}^{+2}16_{7}^{+1}3^{+3})$ 2 6912 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{5}^{-1}3^{+3})$ 2 6912 $\mathrm{I}(4_{\mathrm{II}}^{+2}16_{1}^{+1}3^{-3})$ 1
6912 $\mathrm{I}(4_{\mathrm{II}}^{+2}16_{1}^{+1}3^{+1}9^{+1})$ 4 6912 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{1}^{+1}3^{+3})$ 2 6912 $\mathrm{I}(4_{\mathrm{II}}^{+2}16_{5}^{-1}3^{-3})$ 1
6912 $\mathrm{I}(4_{\mathrm{II}}^{+2}16_{5}^{-1}3^{+1}9^{+1})$ 4 6912 $\mathrm{I}(4_{2}^{+2}16_{3}^{-1}3^{-3})$ 2 6912 $\mathrm{I}(4_{6}^{+2}16_{3}^{-1}3^{+3})$ 3
6912 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{7}^{+1}3^{-3})$ 1 6912 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{7}^{+1}3^{+1}9^{-1})$ 2 6912 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{7}^{+1}3^{+1}9^{+1})$ 2
6912 $\mathrm{I}(4_{\mathrm{II}}^{+2}16_{3}^{-1}3^{+3})$ 2 6912 $\mathrm{I}(4_{2}^{+2}16_{5}^{-1}3^{-3})$ 2 7500 $\mathrm{II}(2_{2}^{+2}3^{+1}5^{-2}25^{-1})$ 5
7500 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-1}5^{-2}25^{+1})$ 5 7776 $\mathrm{I}(2_{1}^{+1}4_{\mathrm{II}}^{+2}3^{+1}9^{+2})$ 2 8000 $\mathrm{I}(4_{5}^{+3}5^{+3})$ 2
8000 $\mathrm{I}(2_{2}^{-2}16_{5}^{-1}5^{+3})$ 6 8000 $\mathrm{I}(2_{2}^{+2}16_{1}^{+1}5^{+3})$ 6 8000 $\mathrm{I}(4_{7}^{-3}5^{+3})$ 2
8000 $\mathrm{II}(4_{7}^{+1}16_{1}^{+1}5^{+3})$ 5 8000 $\mathrm{I}(4_{5}^{-1}16_{5}^{-1}5^{+3})$ 10 8100 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{-2}9^{+1}5^{+2})$ 2
8100 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+2}9^{-1}5^{-2})$ 3 8100 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{-2}9^{-1}5^{+2})$ 2 8100 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+2}9^{+1}5^{-2})$ 3
8640 $\mathrm{I}(4_{1}^{-3}3^{-3}5^{-1})$ 3 8640 $\mathrm{I}(4_{3}^{+3}3^{-3}5^{+1})$ 2 9000 $\mathrm{I}(2_{7}^{+3}3^{+2}5^{+3})$ 4
9000 $\mathrm{I}(2_{5}^{-3}3^{+2}5^{-3})$ 4 9000 $\mathrm{II}(2_{5}^{-1}4_{7}^{+1}3^{+2}5^{-3})$ 4 9000 $\mathrm{II}(2_{7}^{+1}4_{5}^{-1}3^{+2}5^{+3})$ 4
9072 $\mathrm{II}(4_{2}^{-2}3^{+2}9^{-1}7^{-1})$ 4 9072 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}9^{-1}7^{-1})$ 4 9261 $\mathrm{II}(3^{-3}7^{+3})$ 1
9261 $\mathrm{II}(3^{-1}9^{-1}7^{-3})$ 3 9261 $\mathrm{I}(3^{-3}7^{+3})$ 2 10125 $\mathrm{II}(3^{+2}9^{+1}5^{-3})$ 1
10125 $\mathrm{II}(3^{-2}9^{+1}5^{+3})$ 2 10125 $\mathrm{II}(3^{+2}9^{-1}5^{-3})$ 1 10125 $\mathrm{I}(3^{+2}9^{+1}5^{-3})$ 2
10125 $\mathrm{I}(3^{-2}9^{+1}5^{+3})$ 3 10125 $\mathrm{I}(3^{+2}9^{-1}5^{-3})$ 2 10500 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+1}5^{+3}7^{-1})$ 3
10500 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{-1}5^{-3}7^{-1})$ 3 10800 $\mathrm{II}(4_{6}^{+2}3^{-3}5^{+2})$ 3 10800 $\mathrm{II}(4_{6}^{+2}3^{+3}5^{-2})$ 3
10800 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-3}5^{+2})$ 3 10800 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+3}5^{-2})$ 3 10976 $\mathrm{I}(2_{7}^{+1}4_{\mathrm{II}}^{+2}7^{+3})$ 3
11664 $\mathrm{II}(4_{\mathrm{II}}^{-2}3^{-1}9^{-1}27^{-1})$ 6 12000 $\mathrm{I}(2_{5}^{-1}4_{\mathrm{II}}^{+2}3^{+1}5^{-3})$ 4 12288 $\mathrm{I}(16_{5}^{+3}3^{-1})$ 2
12288 $\mathrm{I}(16_{1}^{-3}3^{-1})$ 2 12348 $\mathrm{II}(2_{6}^{-2}3^{+2}7^{+3})$ 3 12348 $\mathrm{I}(2_{\mathrm{II}}^{-2}3^{+2}7^{-3})$ 3
12348 $\mathrm{I}(2_{\mathrm{II}}^{-2}3^{+2}7^{+3})$ 3 12348 $\mathrm{II}(2_{6}^{+2}3^{+2}7^{-3})$ 3 12960 $\mathrm{I}(2_{7}^{+1}4_{\mathrm{II}}^{+2}3^{+2}9^{-1}5^{+1})$ 7
13500 $\mathrm{II}(2_{2}^{-2}3^{+3}5^{+3})$ 2 13500 $\mathrm{II}(2_{2}^{+2}3^{+3}5^{-3})$ 1 13500 $\mathrm{II}(2_{2}^{+2}3^{-1}9^{+1}5^{-3})$ 4
13500 $\mathrm{II}(2_{6}^{+2}3^{-3}5^{-3})$ 3 13500 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-3}5^{+3})$ 1 13500 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+1}9^{-1}5^{+3})$ 4
13500 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+3}5^{+3})$ 3 13500 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-3}5^{-3})$ 2 13500 $\mathrm{I}(4_{5}^{-1}3^{+3}5^{+3})$ 3
13500 $\mathrm{I}(4_{7}^{+1}3^{+3}5^{-3})$ 2 13824 $\mathrm{I}(2_{3}^{-1}16_{4}^{-2}3^{-3})$ 6 13824 $\mathrm{I}(2_{5}^{-1}16_{2}^{-2}3^{-3})$ 3
13824 $\mathrm{I}(2_{1}^{+1}16_{2}^{+2}3^{-3})$ 3 13824 $\mathrm{I}(2_{3}^{-1}16_{0}^{+2}3^{-3})$ 6 13824 $\mathrm{I}(2_{5}^{-1}16_{6}^{+2}3^{-3})$ 4
13824 $\mathrm{I}(2_{1}^{+1}16_{6}^{-2}3^{-3})$ 4 14400 $\mathrm{I}(4_{3}^{+3}3^{+2}5^{-2})$ 4 14400 $\mathrm{I}(4_{5}^{+3}3^{+2}5^{-2})$ 4
15552 $\mathrm{I}(4_{3}^{+3}3^{+1}9^{-2})$ 2 15552 $\mathrm{I}(4_{3}^{+3}3^{+1}9^{+2})$ 2 15552 $\mathrm{I}(2_{\mathrm{II}}^{-2}16_{1}^{+1}3^{+1}9^{+2})$ 6
15552 $\mathrm{I}(2_{\mathrm{II}}^{+2}16_{5}^{-1}3^{+1}9^{+2})$ 6 15552 $\mathrm{I}(4_{5}^{-3}3^{+1}9^{+2})$ 4 15876 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+2}9^{-1}7^{+2})$ 2
15876 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+2}9^{+1}7^{+2})$ 2 15972 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+1}11^{+3})$ 2 15972 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{-1}11^{-3})$ 2
16384 $\mathrm{I}(16_{\mathrm{II}}^{-2}64_{3}^{-1})$ 3 16875 $\mathrm{I}(3^{+3}5^{-2}25^{+1})$ 5 19440 $\mathrm{II}(4_{2}^{-2}3^{+1}9^{+2}5^{+1})$ 4
19440 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}9^{+2}5^{+1})$ 4 19652 $\mathrm{II}(2_{\mathrm{II}}^{-2}17^{+3})$ 2 19773 $\mathrm{II}(3^{+2}13^{-3})$ 2
20250 $\mathrm{I}(2_{5}^{-1}3^{+2}9^{+1}5^{-3})$ 7 20480 $\mathrm{I}(16_{3}^{+3}5^{+1})$ 5 20480 $\mathrm{I}(16_{7}^{-3}5^{+1})$ 5
20580 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+1}5^{+1}7^{+3})$ 3 20580 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{-1}5^{-1}7^{+3})$ 3 20736 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{1}^{+1}3^{-2}9^{+1})$ 2
20736 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{1}^{+1}3^{-2}9^{-1})$ 2 20736 $\mathrm{I}(4_{\mathrm{II}}^{+2}16_{5}^{-1}3^{+2}9^{+1})$ 3 20736 $\mathrm{I}(4_{2}^{+2}16_{3}^{-1}3^{+2}9^{+1})$ 4
20736 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{7}^{+1}3^{-2}9^{+1})$ 2 20736 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{7}^{+1}3^{-2}9^{-1})$ 2 20736 $\mathrm{I}(4_{\mathrm{II}}^{+2}16_{3}^{-1}3^{+2}9^{+1})$ 3
20736 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{3}^{-1}3^{-2}9^{+1})$ 2 20736 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{3}^{-1}3^{-2}9^{-1})$ 2 20736 $\mathrm{I}(4_{\mathrm{II}}^{+2}16_{7}^{+1}3^{+2}9^{+1})$ 3
20736 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{5}^{-1}3^{-2}9^{+1})$ 2 20736 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{5}^{-1}3^{-2}9^{-1})$ 2 20736 $\mathrm{I}(4_{\mathrm{II}}^{+2}16_{1}^{+1}3^{+2}9^{+1})$ 3
21952 $\mathrm{I}(4_{5}^{+3}7^{+3})$ 3 22500 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+2}5^{-2}25^{+1})$ 2 22500 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+2}5^{-2}25^{-1})$ 2
24000 $\mathrm{I}(4_{7}^{-3}3^{-1}5^{-3})$ 3 24000 $\mathrm{I}(4_{5}^{+3}3^{-1}5^{+3})$ 2 27000 $\mathrm{I}(2_{5}^{-3}3^{-3}5^{-3})$ 4
27000 $\mathrm{II}(2_{7}^{+1}4_{7}^{+1}3^{+3}5^{+3})$ 4 27648 $\mathrm{I}(4_{3}^{-1}16_{\mathrm{II}}^{-2}3^{-3})$ 2 27648 $\mathrm{I}(4_{\mathrm{II}}^{-2}64_{3}^{-1}3^{-3})$ 4
27783 $\mathrm{I}(3^{+2}9^{-1}7^{-3})$ 4 30000 $\mathrm{II}(4_{2}^{+2}3^{+1}5^{-2}25^{-1})$ 5 30000 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+1}5^{-2}25^{-1})$ 5
30375 $\mathrm{I}(3^{-1}9^{+2}5^{-3})$ 4 32000 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{5}^{-1}5^{+3})$ 2 32000 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{3}^{-1}5^{+3})$ 2
32000 $\mathrm{I}(4_{2}^{+2}16_{1}^{+1}5^{+3})$ 6 32000 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{7}^{+1}5^{+3})$ 2 32000 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{1}^{+1}5^{+3})$ 2
32000 $\mathrm{I}(16_{2}^{+2}5^{+3})$ 10 32000 $\mathrm{I}(16_{6}^{-2}5^{+3})$ 10 34560 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{7}^{+1}3^{-3}5^{+1})$ 2
34560 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{1}^{+1}3^{-3}5^{-1})$ 3 34560 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{5}^{-1}3^{-3}5^{-1})$ 3 34560 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{3}^{-1}3^{-3}5^{+1})$ 2
35937 $\mathrm{I}(3^{+3}11^{-3})$ 2 36000 $\mathrm{I}(2_{7}^{+1}4_{\mathrm{II}}^{+2}3^{+2}5^{+3})$ 4 36000 $\mathrm{I}(2_{5}^{-1}4_{\mathrm{II}}^{+2}3^{+2}5^{-3})$ 4
37044 $\mathrm{II}(2_{\mathrm{II}}^{+2}3^{-3}7^{+3})$ 2 37044 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+3}7^{+3})$ 2 40500 $\mathrm{II}(2_{\mathrm{II}}^{+2}3^{+2}9^{+1}5^{-3})$ 2
40500 $\mathrm{II}(2_{\mathrm{II}}^{+2}3^{-2}9^{+1}5^{+3})$ 3 40500 $\mathrm{II}(2_{\mathrm{II}}^{+2}3^{+2}9^{-1}5^{-3})$ 2 40500 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{-2}9^{-1}5^{-3})$ 3
40500 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+2}9^{+1}5^{+3})$ 2 40500 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+2}9^{-1}5^{+3})$ 2 49152 $\mathrm{I}(16_{\mathrm{II}}^{-2}64_{1}^{+1}3^{-1})$ 4
49392 $\mathrm{II}(4_{2}^{-2}3^{+2}7^{+3})$ 3 49392 $\mathrm{II}(4_{6}^{-2}3^{+2}7^{-3})$ 3 49392 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}7^{+3})$ 3
49392 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}7^{-3})$ 3 54000 $\mathrm{II}(4_{2}^{-2}3^{+3}5^{-3})$ 1 54000 $\mathrm{II}(4_{2}^{-2}3^{-1}9^{+1}5^{-3})$ 4
54000 $\mathrm{II}(4_{6}^{-2}3^{+3}5^{+3})$ 2 54000 $\mathrm{II}(4_{6}^{-2}3^{-3}5^{-3})$ 3 54000 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+3}5^{-3})$ 1
54000 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}9^{+1}5^{-3})$ 4 54000 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+3}5^{+3})$ 2 54000 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-3}5^{-3})$ 3
57600 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{1}^{+1}3^{+2}5^{-2})$ 4 57600 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{7}^{+1}3^{+2}5^{-2})$ 4 57600 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{3}^{-1}3^{+2}5^{-2})$ 4
57600 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{5}^{-1}3^{+2}5^{-2})$ 4 62208 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{3}^{-1}3^{+1}9^{-2})$ 2 62208 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{3}^{-1}3^{+1}9^{+2})$ 2
62208 $\mathrm{I}(4_{\mathrm{II}}^{+2}16_{1}^{+1}3^{+1}9^{+2})$ 4 62208 $\mathrm{I}(4_{\mathrm{II}}^{+2}16_{5}^{-1}3^{+1}9^{+2})$ 4 62208 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{7}^{+1}3^{+1}9^{-2})$ 2
62208 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{7}^{+1}3^{+1}9^{+2})$ 2 67500 $\mathrm{II}(2_{6}^{+2}3^{+3}5^{-2}25^{+1})$ 5 67500 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{-3}5^{-2}25^{-1})$ 5
81000 $\mathrm{I}(2_{7}^{+3}3^{+2}9^{-1}5^{+3})$ 7 81000 $\mathrm{II}(2_{5}^{-1}4_{7}^{+1}3^{+2}9^{+1}5^{-3})$ 7 83349 $\mathrm{II}(3^{-1}9^{+2}7^{-3})$ 3
87808 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{1}^{+1}7^{+3})$ 3 87808 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{5}^{-1}7^{+3})$ 3 94500 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+3}5^{+3}7^{+1})$ 3
94500 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{-3}5^{-3}7^{+1})$ 3 96000 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{7}^{+1}3^{-1}5^{-3})$ 3 96000 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{1}^{+1}3^{-1}5^{+3})$ 2
96000 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{5}^{-1}3^{-1}5^{+3})$ 2 96000 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{3}^{-1}3^{-1}5^{-3})$ 3 108000 $\mathrm{I}(2_{7}^{+1}4_{\mathrm{II}}^{+2}3^{+3}5^{+3})$ 4
110592 $\mathrm{I}(16_{3}^{+3}3^{-3})$ 2 110592 $\mathrm{I}(16_{7}^{-3}3^{-3})$ 2 111132 $\mathrm{I}(2_{\mathrm{II}}^{-2}3^{+2}9^{+1}7^{-3})$ 4
111132 $\mathrm{II}(2_{6}^{+2}3^{+2}9^{-1}7^{-3})$ 4 121500 $\mathrm{II}(2_{2}^{+2}3^{-1}9^{+2}5^{-3})$ 4 121500 $\mathrm{I}(2_{\mathrm{II}}^{+2}3^{+1}9^{+2}5^{+3})$ 4
128000 $\mathrm{I}(4_{5}^{-1}16_{2}^{+2}5^{+3})$ 10 128000 $\mathrm{I}(4_{7}^{+1}16_{\mathrm{II}}^{-2}5^{+3})$ 5 143748 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+3}11^{-3})$ 2
143748 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{-3}11^{+3})$ 2 177957 $\mathrm{II}(3^{+2}9^{-1}13^{-3})$ 4 185220 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+3}5^{+1}7^{-3})$ 3
185220 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{-3}5^{-1}7^{-3})$ 3 216000 $\mathrm{I}(4_{7}^{-3}3^{+3}5^{-3})$ 2 216000 $\mathrm{I}(4_{1}^{-3}3^{+3}5^{+3})$ 3
270000 $\mathrm{II}(4_{6}^{+2}3^{+3}5^{-2}25^{+1})$ 5 270000 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+3}5^{-2}25^{+1})$ 5 324000 $\mathrm{I}(2_{5}^{-1}4_{\mathrm{II}}^{+2}3^{+2}9^{+1}5^{-3})$ 7
442368 $\mathrm{I}(16_{\mathrm{II}}^{-2}64_{3}^{-1}3^{-3})$ 4 444528 $\mathrm{II}(4_{6}^{-2}3^{+2}9^{-1}7^{-3})$ 4 444528 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{+2}9^{-1}7^{-3})$ 4
486000 $\mathrm{II}(4_{2}^{-2}3^{-1}9^{+2}5^{-3})$ 4 486000 $\mathrm{I}(4_{\mathrm{II}}^{-2}3^{-1}9^{+2}5^{-3})$ 4 512000 $\mathrm{I}(16_{3}^{+3}5^{+3})$ 5
512000 $\mathrm{I}(16_{7}^{-3}5^{+3})$ 5 514500 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+1}5^{-3}7^{-3})$ 3 514500 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{-1}5^{+3}7^{-3})$ 3
864000 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{7}^{+1}3^{+3}5^{-3})$ 2 864000 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{1}^{+1}3^{+3}5^{+3})$ 3 864000 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{5}^{-1}3^{+3}5^{+3})$ 3
864000 $\mathrm{I}(4_{\mathrm{II}}^{-2}16_{3}^{-1}3^{+3}5^{-3})$ 2 4630500 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{+3}5^{-3}7^{+3})$ 3 4630500 $\mathrm{II}(2_{\mathrm{II}}^{-2}3^{-3}5^{+3}7^{+3})$ 3
: Totally-Reflective Genera in dimension $4$
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A rigorous theorem due to Aizenman and Wehr asserts that there can be no latent heat in a two-dimensional system with quenched random impurities. We examine this result, and its possible extensions to higher dimensions, in the context of several models. For systems whose pure versions undergo a strong first-order transition, we show that there is an asymptotically exact mapping to the random field Ising model, at the level of the interface between the ordered and disordered phases. This provides a physical explanation for the above result and also implies a correspondence between the problems in higher dimensions, including scaling relations between their exponents. The particular example of the $q$-state Potts model in two dimensions has been considered in detail by various authors and we review the numerical results obtained for this case. Turning to weak, fluctuation-driven first-order transitions, we describe analytic renormalisation group calculations which show how the continuous nature of the transition is restored by randomness in two dimensions.'
---
[cond-mat/9806355\
JLC-98-3]{}
[$^{\dag}$ Invited talk to be presented at STATPHYS 20, Paris, July 1998.]{}
Introduction
============
The effect of quenched random impurities which couple to the local energy density on the critical behaviour of a system whose pure version undergoes a continuous transition is well understood in terms of the Harris criterion [@Harris]: if the specific heat exponent $\alpha$ of the pure system is negative, such impurities do not change the qualitative nature of the transition or its universality class, while in the opposite case there is expected to be different behaviour described by a new random fixed point of the renormalisation group (RG). Given the volume of literature devoted to this subject, it is rather surprising to find relatively few studies of the effects of such impurities on systems whose pure versions undergo a first-order transition, despite the ubiquity of such systems in nature. A naive extension of the Harris criterion, based on the observation that the effective value of $\alpha$ is unity at a first-order transition, might suggest that randomness is always strongly relevant. On the other hand, first-order behaviour is accompanied by a finite correlation length and is commonly assumed to be stable under small perturbations. Following early work by Imry and Wortis [@IW], Aizenman and Wehr [@AW] and Hui and Berker [@HB] showed that the role of dimensionality $d$ is crucial. In fact the first authors proved a rigorous theorem which states that for $d\leq 2$ the Gibbs state is always unique, for arbitrarily small but non-zero concentration of impurities. This means that there can be no phase coexistence at the transition and hence no latent heat.
This result raises a number of important questions: (a) what is its physical mechanism; (b) is the continuous transition accompanied by a divergent correlation length; (c) if so, what are the (presumably) universal critical exponents characterising the transition; and (d) what happens for $d>2$? Initial Monte Carlo studies of (c) [@CFL; @Dom] suggested that the critical behaviour of all such systems might be strongly universal, exhibiting Ising-like exponents independent of the underlying symmetry of the model. This received some theoretical support [@Kardar]. However, more recent numerical studies (to be described below) support earlier theoretical results [@Ludwig] in suggesting that the critical behaviour of, for example, the random bond $q$-state Potts model depends continuously on $q$, both in the region where the pure model has a continuous transition, and where it is first-order. The aim of this talk is to summarise some of these recent developments, and to provide at least partial answers to the above questions.
Mapping to the random field Ising model
=======================================
First, I want to describe a mapping [@CJ] of this problem to the *random field *Ising model (RFIM), which is asymptotically valid for very strong first-order transitions, and which provides a physical explanation of the Aizenman-Wehr result [@AW], as well as a prediction for what happens when $d>2$.**
Consider a pure system at a thermal first-order transition point. There will be coexistence between a (generally unique) disordered phase and the (generally non-unique) ordered phases. The internal energies $U_1$ and $U_2$ of these two kinds of phase will differ by the latent heat, $L$. Now consider a interface between the disordered phase and one of the ordered phases, with surface tension $\sigma$ (measured in units of $kT_c$). If $\sigma$ is large, there will be very few isolated bubbles of the opposite phase above or below the main interface. Its equilibrium statistics may therefore be described by a free energy functional equal to its area multiplied by $\sigma$. Let us compare this with the interface between the two *ordered *phases of an Ising model, with spins taking the values $\pm1$, at low temperatures. Once again, there will be few bubbles of the opposite phase, and the bare interfacial tension will be $\sim 2J$, where $J$ is the reduced exchange coupling. In the limit when $\sigma\sim 2J$ is large, these interfacial models are therefore identical[^1]**
Now consider the effects of randomness in these two models, in the first case random bonds, coupling to the local energy density, and in the second random fields, coupling to the local magnetisation. In the RFIM, these are accounted for by adding a term $\Sigma_{r>}h(r)-
\Sigma_{r<}h(r)$, where $h(r)$ is the local random field, and in the two terms $r$ is summed respectively above and below the instantaneous position of the interface. For random impurities of local concentration $\delta x(r)$ we have, similarly, $U_1\Sigma_{r>}\delta x(r)+U_2\Sigma_{r<}\delta x(r)$. Apart from a constant independent of the position of the interface, this may be written as $L$ times $\Sigma_{r>}\delta x(r)-\Sigma_{r<}\delta x(r)$, exactly the same form as for the RFIM.
We may therefore set up a dictionary between these two cases, in which the thermal variables of the random bond system are related to the magnetic variables of the RFIM: $$\begin{aligned}
\sigma/kT_c&\longleftrightarrow&J/kT\\
(L/kT_c)\,x&\longleftrightarrow& h_{RF}/kT\\
(T-T_c)\,L&\longleftrightarrow&H\cdot M,\end{aligned}$$ where the last relation is between the fields $(T-T_c)$ and a *uniform *magnetic field $H$ which respectively distinguish between the two phases. Although the above mapping may seem ill-defined in its use of the local energy density as a kind of order parameter, it may be made completely explicit, for example, for the $q$-state Potts model through the mapping to the random cluster model, where $\sigma\sim
L\sim\ln q$ for large $q$ [@CJ].**
The interfacial model of the RFIM has been studied extensively [@RFIMinterface] and, in particular, RG equations have been derived which are asymptotically exact in the limit of low temperatures and weak randomness, just where our mapping is valid. For $d=2$ the variable $h_{RF}/J\sim xL/\sigma$ is (marginally) relevant, and, just as this destroys the spontaneous magnetisation for the RFIM, the latent heat vanishes for the random bond case, in accord with the Aizenman-Wehr theorem [@AW]. For the RFIM, the RG trajectories flow towards a paramagnetic fixed point at which the correlation length $\xi\sim\exp({\rm const}(J/h_{RF})^3)$ is finite, but note that this is outside the region where the mapping to the random bond problem is valid. We cannot conclude, therefore, anything about the nature of the latter’s true critical behaviour from this argument. In fact, numerical and other analytic studies indicate that the actual correlation length is divergent, and hence $\xi$ is merely a crossover length in this case.
The predictions are more interesting for $d>2$, when the RFIM exhibits a critical fixed point at $kT/J=0$ and a finite value of $h_{RF}/J$.
As shown in Fig. \[fig1\], there is now a region in which the spontaneous magnetisation (latent heat) is non-vanishing as the sign of the uniform field ($T-T_c$) is changed. This region is bounded, in the case of the RFIM by the critical curve, close to which the critical behaviour is determined by the zero-temperature fixed point. Similarly, the first-order region of the random bond system will be bounded by a line of *tricritical *points, above which (presumably) the transition becomes continuous. Since the fixed point occurs in the region where the mapping between the two systems is valid, we may infer some of the tricritical exponents from those of the RFIM [@CJ]. For example, the latent heat should vanish as $(x_c-x)^\beta$, where $\beta$ is the usual magnetisation exponent of the RFIM. Similarly, the correlation length on the critical surface should diverge as $x\to x_c$ with the usual exponent $\nu$ of the RFIM. But the behaviour for $T\not=T_c$ is related to the *magnetic *properties of the RFIM, and is complicated by the fact that the temperature at the RFIM fixed point is dangerously irrelevant, with an RG eigenvalue $-\theta$ which is responsible for the violation of hyperscaling $\alpha=2-(d-\theta)\nu$ in that model. As a result, for example, the correlation length in the random bond model at $x=x_c$ diverges as $(T-T_c)^{1/y}$, with $y=d-\theta-\beta/\nu$.****
Results for the Potts model.
============================
As is well-known, the pure $q$-state Potts model undergoes a first-order transition for $q>4$ in $d=2$, and it is a relatively simple system to study numerically in the random bond case. By choosing a suitable distribution of randomness, one can fix the model to be self-dual so that the critical point is determined exactly. One method, which is very effective for pure two-dimensional critical system, is to study the finite-size scaling behaviour of the eigenvalues $\Lambda_i$ of the transfer matrix in a strip of width $N$. By conformal invariance [@JCconf], these are related to the scaling dimensions $x_i$ by $2\pi x_i/N\sim\ln(\Lambda_0/\Lambda_i)$ as $N\to\infty$. For the Potts model, it is possible [@JC] to write the transfer matrix in the so-called connectivity basis [@BN], allowing $q$ to appear as an easily tunable variable. In the random case, however, the transfer matrices for different rows do not commute, and the role of the $\Lambda_i$ is taken by the Lyapunov exponents which govern the average growth rate of the norm of vectors under the action of the transfer matrices. The asymptotic behaviour of the correlation functions is related to these exponents as for the pure case. In order to use conformal invariance, it is necessary to assume translational invariance which is only recovered after quenched averaging. However, the Lyapunov exponents themselves are not self-averaging, only their logarithms. This is related to the phenomenon of multi-scaling, whereby the average of the $p$th power of the correlation function is governed by an exponent $x^{(p)}$ which is not linear in $p$. Since this occurs in most random systems, however, I shall not treat the problem in detail here. It still proves possible, effectively by measuring the whole distribution of the $\ln \Lambda_i$, to extract the decay of the average correlation function, and hence, for example, the magnetic exponent $x_1\equiv\beta/\nu$. The results from our study [@JC] are shown in Fig. \[fig2\]. We see that the value of $x_1$ appears to increase
steadily with $q$. For $q<3$ it agrees with the analytic $(q-2)$-expansion [@Ludwig; @Dot]. There appears to be no break at $q=4$ where the pure transition becomes first-order. Our results disagree with earlier Monte Carlo work [@CFL] for $q=8$ where a number close to the Ising value of $\frac18$ was reported. However, more recent Monte Carlo results by Picco [@Picco] find $x_1=0.150-0.155$ for $q=8$ (and $0.185\pm0.005$ for $q=64$, while Chatelain and Berche [@CB] report $x_1=0.153\pm0.003$ for $q=8$. The small discrepancy with our results may be explained by a careful study of the crossover behaviour which shows that stronger randomness (beyond the reach of our methods) must be considered as $q$ increases [@Picco].
All studies report a thermal exponent $\nu\approx1$. This approximately saturates the lower bound proved by Chayes et al [@Chayes]. It has been suggested [@Davis] that, in general, this might be a result of the averaging procedure and that in some random systems the ‘true’ value of $\nu$ might be less than unity. However, it can shown that this is not the case here.
Analytic results for weak first-order transitions.
==================================================
Although the $q$-state Potts model is relatively simple to analyse numerically, like most other first-order transitions it is difficult to study in any kind of perturbative RG approach, since the randomness is strongly relevant. However, this is not always the case for systems which exhibit weak ‘fluctuation-driven’ first-order transitions. An example is afforded by $N$ Ising models with spins $s_i(r)$, coupled through their energy densities, with a hamiltonian $${\cal H}=-\sum_{r,r'}J(r,r')\sum_is_i(r)s_i(r')+
g\sum_{r,r'}\sum_{i\not=j}s_i(r)s_i(r')s_j(r)s_j(r')$$ For uniform couplings $J$, this exhibits a first-order transition when $N>2$, and in fact, on the critical surface is equivalent to the Gross-Neveu model, which may be analysed nonperturbatively to show that the correlation length is $\xi\sim\exp({\rm const}/(N-2)g)$ [@GN]. Even with random bonds $J(r,r')$ of strength $\Delta$ the one-loop RG equations may be derived By very simple combinatorial methods [@JCbook], to give [@RFD] $$\begin{aligned}
dg/d\ell&=&4(N-2)g^2-8g\Delta+\cdots\label{flows1}\\
d\Delta/d\ell&=& -8\Delta^2+8(N-1)g\Delta+\cdots\label{flows2}\end{aligned}$$ These flows are illustrated in Fig. \[fig3\].
When $\Delta=0$, $g$ runs away to infinity, in a manner consistent with the non-perturbative result for $\xi$. When randomness is added, however, it is marginally relevant, and in fact the flows are quite similar to those found for the RFIM quoted earlier if we identify the interfacial tension $\sim\xi$. However, in this case, we also see where the flows end: in this case at the critical fixed point corresponding to $N$ decoupled pure Ising models. (This is the only example I know of where the infrared and ultraviolet fixed points of a set of RG flows are the same.) There have been various generalisations of this calculation to coupled Potts models [@gen]. In most cases the addition of randomness induces flow towards a critical fixed point which is perturbatively accessible. In higher dimensions, however, different outcomes are possible. Adding randomness to models in $4-\epsilon$ dimensions with cubic anisotropy appears not to change their fluctuation-driven first-order character [@RFD]. However, for impure $n$-component superconductors in $4-\epsilon$ dimensions there is a critical concentration above which the transition becomes continuous [@CBoy].
Outlook
=======
There are still many open questions in this relatively little explored area. A lot more needs to be learned about the nature of the universal critical behaviour (if indeed it is universal), both in $d=2$ and for $d>2$ when the randomness is sufficiently strong. In the former case, it may be possible to solve the problem by conformal field theory methods. Even the limit of large $q$ appears non-trivial, however. Similar ideas apply to quantum phase transitions and may have relevance for the quantum Hall effect. Most importantly, it should be possible to find experimental systems which realise some of the predictions discussed above. There are, after all, many three-dimensional examples of first-order transitions. However, it should be stressed that the randomness should couple only to the local energy density, not to the order parameter, otherwise this becomes the random field problem. It is also important to ensure that the tricritical behaviour is driven by the effects discussed and not by some simple mean-field mechanism (which would give rise to mean-field tricritical exponents in $d=3$, part from logs.) Finally it is important to understand the dynamics of these systems: it may be that, like the random field problem, they are plagued by logarithmically slow time scales close to the critical point [@RFdyn].
I am grateful to Jesper Jacobsen for his continuing collaboration on this problem. This work was supported in part by EPSRC Grant GR/J78327.
[9]{} A. B. Harris, J. Phys. C [**7**]{}, 1671 (1974). Y. Imry and M. Wortis, Phys. Rev. B [**19**]{}, 3581 (1979). M. Aizenman and J. Wehr, Phys. Rev. Lett. [**62**]{}, 2503 (1989). K. Hui and A. N. Berker, Phys. Rev. Lett. [**62**]{}, 2507 (1989); Phys. Rev. Lett. [**63**]{}, 2433 (1989) (erratum). S. Chen, A. M. Ferrenberg and D. P. Landau, Phys. Rev. E [**52**]{}, 1377 (1995). E. Domany and S. Wiseman, Phys. Rev. E [**51**]{}, 3074, 1995. M. Kardar, A. L. Stella, G. Sartoni and B. Derrida, Phys. Rev. E [**52**]{}, R1269 (1995). A. W. W. Ludwig and J. L. Cardy, Nucl. Phys. B [**285**]{} \[FS19\], 687 (1987); A. W. W. Ludwig, Nucl. Phys. B [**285**]{}, 97 (1987). J. Cardy and J. L. Jacobsen, Phys. Rev. Lett. [**79**]{}, 4063 (1997). See, for example, A. J. Bray and M. A. Moore, J. Phys. C [**18**]{}, L927 (1985). J. L. Cardy, J. Phys. A [**17**]{}, L385 (1984). J. L. Jacobsen and J. L. Cardy, Nucl. Phys. B [**515**]{}, 701 (1998). H. W. J. Blöte and M. P. Nightingale, Physica A [**112**]{}, 405 (1982). V. Dotsenko, M. Picco and P. Pujol, Phys. Lett. B [**347**]{}, 113 (1995); Nucl. Phys. B [**455**]{}, 701 (1995). M. Picco, cond-mat/9802092. C. Chatelain and B. Berche, Phys. Rev. Lett. [**80**]{}, 1670 (1998). J. T. Chayes, L. Chayes, D. S. Fisher and T. Spencer, Phys. Rev. Lett. [**57**]{}, 2999 (1986); J. T. Chayes, L. Chayes, D. S. Fisher and T. Spencer, Commun. Math. Phys. [**120**]{}, 501 (1989). F. Pázmándi, R. T. Scalettar and G. T.. Zimányi, Phys. Rev. Lett. [**79**]{}, 5130 (1997). D. Gross and A. Neveu, Phys. Rev. D [**10**]{}, 3235 (1974); E. Fradkin, Phys. Rev. Lett. [**54**]{}, 1967 (1984). J. L. Cardy, in [*Scaling and Renormalization in Statistical Physics*]{}, ch. 5 (Cambridge University Press, 1996). J. L. Cardy, J. Phys. A [**29**]{}, 1897 (1996). P. Pujol, Europhys. Lett. [**35**]{}, 283 (1996); P. Simon, Phys. Lett. B [**408**]{}, 293 (1997), Europhys. Lett. [**41**]{}, 605 (1998). D. Boyanovsky and J. L. Cardy, Phys. Rev. B [**25**]{}, 7058 (1982). D. S. Fisher, Phys. Rev. Lett. [**56**]{}, 416 (1986).
[^1]: This is not the case when bubbles of the opposite phase are included. For example, regions of ordered phase appearing in the disordered phase are counted with the degeneracy factor of the ordered phases, as compared with a factor of unity for the Ising case.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study the commutation relations and normal ordering between families of operators on symmetric functions. These operators can be naturally defined by the operations of multiplication, Kronecker product, and their adjoints. As applications we give a new proof of the skew Littlewood–Richardson rule and prove an identity about the Kronecker product with a skew Schur function.'
address:
- 'Departamento de Matemática Aplicada I, Escuela Técnica Superior de Ingeniería Informática, Avda. Reina Mercedes, S/N, 41012 Sevilla, España'
- 'Department of Mathematics, Bucknell University, Lewisburg, PA 17837, USA'
- 'Dartmouth College, Mathematics Department, 6188 Kemeny Hall, Hanover, NH 03755, USA'
- 'Departamento de Álgebra, Facultad de Matemáticas, Universidad de Sevilla, Avda. Reina Mercedes, Sevilla, España'
author:
- Emmanuel Briand
- 'Peter R. W. McNamara'
- Rosa Orellana
- Mercedes Rosas
bibliography:
- 'commutators.bib'
title: Commutation and normal ordering for operators on symmetric functions
---
Introduction
============
Due to the their connection to representation theory, Schubert calculus, and their beautiful combinatorial description, Schur functions are ubiquitous in algebraic combinatorics. For this reason, it not surprising that identities involving Schur functions greatly improve the understanding of these subjects
There are two important products naturally defined on symmetric functions: the ordinary and the Kronecker product. The well-known Littlewood-Richardson coefficients are the structure coefficients for the ordinary product, while the elusive Kronecker coefficients are the structure coefficients for the the Kronecker product. They naturally define linear operators on symmetric functions.
Let $f$ be a symmetric function. Then, define $U_f$ to be the operator “multiplication by $f$”, and $K_f$ be “Kronecker multiplication by $f$.” Explicitly, the operators act on a symmetric function $g$ by $${U_{f}}(g)=f g, \qquad {K_{f}}(g)=f \ast g.$$ The Hall inner product allow us to define the adjoint of $U_f$, which is denoted by $D_f$, and sometimes called the skewing operator. With respect to this inner product $K_f$ is self-adjoint. We will also consider another intriguing operator related to the Kronecker product, ${\overline{K}_{\lambda}}$, defined on the Schur basis as follows: Let $\lambda$ be a partition and $g$ be any homogeneous symmetric function of degree $n$, then $${\overline{K}_{\lambda}}(g)=s_{(n-|\lambda|,\lambda)}\ast g\,,$$ where $(n-|\lambda|,\lambda)=(n-|\lambda|,\lambda_1,\lambda_2,\ldots)$ is a sequence of integers. This sequence is not always decreasing. To deal with this issue, we define the Schur function $s_{(n-|\lambda|,\lambda)}$ by means of the Jacobi–Trudi formula: $
s_{(\alpha_1,\alpha_2,\ldots,\alpha_N)}=\det(h_{\alpha_{i}+j-i})_{i,j=1\ldots N}.$ This determinant coincides with the Schur function $s_{\alpha}$ when $\alpha$ is weakly decreasing (*i.e.,* is a partition) but makes sense even when $\alpha$ is not. Once the operators ${\overline{K}_{\lambda}}$ are defined, the definition is extended by linearity to ${\overline{K}_{f}}$ for any symmetric function $f$.
In this paper we study identities involving four families of operators on the ring of symmetric functions, ${\textit{Sym}}$: ${U_{\lambda}}$, ${D_{\lambda}}$, ${K_{\lambda}}$, and ${\overline{K}_{\lambda}}$. Here the index $\lambda$ is a partition and indicates that the operator is product by the Schur function $s_\lambda$. We consider the following questions: Given any pair of them. Can we establish the commutator relations between them? Given a word involving them, how can we put it in a normal form? Is this expression unique? Is it possible to express some of them in terms of the other ones?
For a motivating example, let us look at the operators ${U_{(1)}}$ and ${D_{(1)}}$. They are well-known to satisfy the commutator relation $
{D_{(1)}}{U_{(1)}} = {U_{(1)}}{D_{(1)}} +1.$ That is Leibniz’s rule for multiplication, when ${U_{(1)}}$ is multiplication by $x$, and ${D_{(1)}}=\frac{\partial}{\partial x}$. This identity is the defining relation for the algebra of Weyl, and the building identity for Stanley’s theory of differential posets, [@Stanley-diffposet].
Our main result is the following theorem that gives beautiful commutation relations four each pair of operators.
\[thm:main\] For any partitions $\alpha$ and $\beta$ we have the following identities (where $\lambda$, $\tau$ and $\nu$ each run over the set of all partitions). $$\begin{aligned}
{D_{\beta}}{U_{\alpha}}&=
\sum_{\lambda} {U_{{\alpha/\lambda}}} {D_{{\beta/\lambda}}} \label{eqDbetaUalpha}\\
{U_{\alpha}}{D_{\beta}}&=\sum_{\lambda} (-1)^{|\lambda|} {D_{{\beta/\lambda'}}} {U_{{\alpha/\lambda}}}\label{eqUalphaDbeta}\\
{K_{\beta}} {U_{\alpha}} &=\sum_{\lambda} {U_{s_{\beta/\lambda} \ast s_{\alpha}}} {K_{\lambda}} \label{KU2}\\
{D_{\alpha}}{K_{\beta}} &=\sum_{\lambda} {K_{\lambda}} {D_{s_{\beta/\lambda} \ast s_{\alpha}}} \label{DK2}\\
{\overline{K}_{\beta}} {U_{\alpha}}&=
\sum_{\tau,\nu} {U_{(s_{\beta/\nu} \ast s_{\tau}) s_{\alpha/\tau}}} {\overline{K}_{\nu}} \label{KKKU2}\\
{D_{\alpha}} {\overline{K}_{\beta}} &=
\sum_{\tau,\nu} {\overline{K}_{\nu}} {D_{(s_{\beta/\nu} \ast s_{\tau}) s_{\alpha/\tau}}}\label{DKKK2}
$$
To answer our set of questions, and in particular, to prove these identities, we associate a generating series to each of our operators. Then using known operations of formal series we are able to obtain a uniform and elegant method to tackle them. More precisely, let $P$ be any of operators ${U_{}}$, ${D_{}}$, ${K_{}}$ and ${\overline{K}_{}}$. We associate to $P$ a formal series of operators $
\sum_{\lambda} s_{\lambda}[A] P_{\lambda}.
$ We call this series the Schur generating series of $P$. The Schur generating series of $P$ defines a linear map that sends any symmetric function $g \in {\textit{Sym}}(X)$ to the expression $
\sum_{\lambda} s_{\lambda}[A] P_{\lambda}(g).
$ The operator $P_\lambda$ can be recovered from this series using the scalar product on ${\textit{Sym}}$.
The first part of our paper concludes with the expansion of the identities in Theorem 1.1 in the Schur basis, and are summarized in Theorem \[thm:main-cor\]. These expansions give normal ordering relations for the operators $U, D, K,$ and $ {\overline{K}_{}}$.
An important question remains. Are such expressions unique? We consider this question in Proposition \[finite expansions\], where we show that finite expansions with respect to the pairs $(U,D)$, $(D,U)$, $(U,K)$, $(K,D)$, $(U,{\overline{K}_{}})$ and $({\overline{K}_{}},D)$ are unique. In contrast, observe that expansions with respect to $(K,U)$ or to $(D,K)$ are not unique. For instance we have the relation ${K_{p_2}} {U_{p_1}} =0$, that is straightforwardly equivalent to the relation ${K_{2}}{U_{1}} = {K_{1,1}}{U_{1}}$. Taking adjoints, we have the relation ${D_{1}}{K_{2}} = {D_{1}} {K_{1,1}}$.
The following charming and well-known identity $
{\overline{K}_{(1)}}={U_{(1)}}{D_{(1)}}-1$ describes a relation between the operators ${\overline{K}_{}}, U,$ and $D$. It translate to one of the few known cases where it is possible to give a combinatorial description for a Kronecker product. We extend this identity to similar expression for the Kronecker product of an arbitrary Skew Schur function with $s_{(n-1,1)}$, and for other simple families of Schur functions (hooks and two–row shapes).
Remarkably, the previous identity shows that ${\overline{K}_{(1)}}$ can be rewritten it in terms of the much simpler operators ${U_{(1)}}$ and ${D_{(1)}}$. In Proposition \[KBf in U D\] we vastly generalize this observation, and show that for any symmetric function $f$ the operator ${\overline{K}_{f}}$ lies in the subalgebra of $\operatorname{End}({\textit{Sym}})$ generated by the operators $U_g$ and $D_g$.
Dealing with naturally defined objects, like our families of operators, you are bound to recover some classical results. A testimony of the elegance of this approach is that both Foulkes’s and Littlewood’s identities can be easily derived from Theorem \[thm:main-cor\]. However, a new identity of the same nature is also obtained, as described in the following table. This is discussed in Section \[sec:commutations\].
--------------------------- ---------------------------------- -----------------------
Product Coproduct equivalent identities
\[2mm\] Ordinary, $\cdot$ Adjoint of the ordinary product and
Kronecker, $\ast$ Adjoint of the ordinary product and
Ordinary, $\cdot $ Adjoint of the Kronecker product and
--------------------------- ---------------------------------- -----------------------
: The three bialgebra structures on ${\textit{Sym}}$ and the corresponding identities $\partial_x \circ \mu = \mu \circ \partial_{\Delta(x)}$.[]{data-label="table pro copro"}
We finish our work with two combinatorial applications of our identities. The first one is a proof of the *skew Littlewood–Richardson Rule*, a combinatorial rule that gives the product of two skew Schur functions as a linear combination of skew Schur functions, based on counting Young tableaux (Theorem \[eqskewLR\]). This rule was conjectured in [@AssafMcNamara], and proved in [@LamLauveSottile]. Our proof relies on the normal ordering relation that decomposes the products ${U_{\alpha}} {D_{\beta}}$ as linear combinations of products of the form ${D_{\beta/\lambda'}} {U_{\alpha/\lambda}}$. It generalizes the algebraic proof given by Thomas Lam of the skew Pieri Rule (a particular case of the skew Littlewood–Richardson Rule) in the appendix of [@AssafMcNamara]. Indeed, Lam’s proof relies on the same normal ordering relation, in the special case of $\beta$ having only one part.
The second application exploits our normal ordering relation for the products ${\overline{K}_{1}} D_{\lambda}$. We extend the combinatorial rule for the expansion in the Schur basis of the Kronecker product of $s_{(n-1,1)}$ with a Schur function, to the Kronecker product of $s_{(n-1,1)}$ with any skew Schur function (Theorem \[thm:skewcorners\]). Additionally, we give a different, combinatorial, proof of this result.
Statement of the results and Historical context {#sec:commutations}
===============================================
The main result of the paper is a list of six normal ordering relations, stated in two equivalent ways in Theorems \[thm:main\] and \[thm:main-cor\]. These relations will be proved in Section \[sec:proofs\].
\[thm:main-cor\] For any partitions $\alpha$ and $\beta$ we have the following identities. $$\begin{aligned}
{D_{\beta}}{U_{\alpha}}&=
\sum_{\mu,\nu}
\left(\sum_{\lambda} c^{\alpha}_{\lambda,\mu} \;c^{\beta}_{\lambda,\nu}\right)
{U_{\mu}}{D_{\nu}}\\
{U_{\alpha}}{D_{\beta}}
&=
\sum_{\mu,\nu}
\left(\sum_{\lambda} (-1)^{|\lambda|} c^{\alpha}_{\lambda,\mu} \; c^{\beta}_{\lambda',\nu}\right)
{D_{\nu}}{U_{\mu}}\\
{K_{\beta}} {U_{\alpha}} &=
\sum_{\mu,\nu}
\left(\sum_{\lambda} g_{\alpha,\lambda,\mu} \; c^{\beta}_{\lambda,\nu} \right)
{U_{\mu}}{K_{\nu}}\\
{D_{\alpha}}{K_{\beta}} &=
\sum_{\mu,\nu}
\left(\sum_{\lambda} g_{\alpha,\lambda,\mu} \; c^{\beta}_{\lambda,\nu} \right)
{K_{\nu}}{D_{\mu}}\\
{\overline{K}_{\beta}} {U_{\alpha}}&=
\sum_{\mu,\nu}
\left( \sum_{\lambda, \sigma, \tau, \theta}
g_{\lambda,\tau,\theta} \; c^{\beta}_{\lambda,\nu} \; c^{\alpha}_{\tau,\sigma} \; c_{\theta,\sigma}^{\mu} \right) {U_{\mu}} {\overline{K}_{\nu}}\\
{D_{\alpha}} {\overline{K}_{\beta}} &=
\sum_{\mu,\nu}
\left( \sum_{\lambda, \sigma, \tau, \theta}
g_{\lambda,\tau,\theta} \; c^{\beta}_{\lambda,\nu} \; c^{\alpha}_{\tau,\sigma} \; c_{\theta,\sigma}^{\mu} \right) {\overline{K}_{\nu}} {D_{\mu}}\end{aligned}$$
Identities and are avatars of well-known identities of Foulkes and Littlewood. Indeed, if we apply the operators in and to the Schur function $s_{\gamma}$ we get $$\begin{aligned}
D_{\beta}(s_{\alpha}s_{\gamma})&=\sum_{\lambda} s_{\alpha/\lambda} {D_{\beta/\lambda}}(s_{\gamma}),\label{Foulkes}\\
s_{\beta}\ast (s_{\alpha} s_{\gamma}) &=\sum_{\lambda} (s_{\beta/\lambda} \ast s_{\alpha}) (s_{\lambda} \ast s_{\gamma}).\label{equ:littlewood}\end{aligned}$$ By linearity, we can replace $s_{\alpha}$ and $s_{\gamma}$ with arbitrary symmetric functions $f$ and $g$. Also if we expand $s_{\beta/\lambda}=\sum_{\mu} c^{\beta}_{\lambda, \mu} s_{\mu}$, we obtain $$\begin{aligned}
{D_{\beta}}(fg)&=\sum_{\lambda,\mu} c^{\beta}_{\lambda,\mu} {D_{\lambda}}(f) {D_{\mu}}(g), \label{TrueFoulkes}\\
s_{\beta}\ast (fg) &=\sum_{\lambda, \mu} c^{\beta}_{\lambda, \mu} (s_{\mu} \ast f)(s_{\lambda} \ast g).
\label{TrueLittlewood}\end{aligned}$$ Formula was obtained by Foulkes ([@FoulkesDifferentialOperators §3.b] , also mentioned in [@MacdonaldBook I.§5 Ex. 25.(d)], while is due to Littlewood ([@Littlewood1956 Theorem III], see also [@MacdonaldBook I.§7 Ex. 23.(c)]). An expression similar to and can be derived the same way from . This is: $$\label{Similar}
D_{\beta}(f \ast g) = \sum_{\lambda} g_{\beta, \lambda, \mu} D_{\lambda}(f) \ast D_{\mu}(g).$$ The similarity between , and has a nice explanation. It is provided by J.–Y. Thibon for and , see [@ThibonCoproductCR p.554], [@ThibonHopf Proposition 6.4], and [@ThibonSlides], but applies as well for . The explanation is as follows: let $B$ be a bialgebra with product $\odot$ and coproduct $\Delta$. The coproduct $\Delta$ induces a product $\mu$ on the dual space $B^*$. For any $x \in B$ or in $B \otimes B$, let $\partial_x$ be the adjoint of the $\odot$–product by $x$. Then, for any $x \in B$: $$\label{Thibon}
\partial_x \circ \mu = \mu \circ \partial_{\Delta(x)}$$ See the aforementionned references by J.–Y. Thibon for a proof.
Consider ${\textit{Sym}}$ with a product that is either the ordinary product or the Kronecker product, and a coproduct that is either the adjoint of the ordinary product, or the adjoint of the ordinary coproduct. This gives four possibilities, but only three of them make ${\textit{Sym}}$ a bialgebra. The three identities , and are obtained by applying to these three bialgebra structures, with $x=s_{\beta}$. See table \[table pro copro\]. Note that for any symmetric function $f$, the operator $K_f$ is its own adjoint.
Thibon also relates in [@ThibonSlides] Identity to Mackey’s formula in group theory (see [@Hazewinkel 18.15]). Let us mention that Cummins [@Cummins identity (13)] also derives an identity very close to .
Formula can be stated as $$s_{\alpha} s_{\gamma/\beta}=\sum_{\lambda} (-1)^{|\lambda|} {D_{\beta/\lambda'}}(s_{\alpha/\lambda}s_{\gamma}). \label{ReverseFoulkes}$$ Formula happens to be closely related to the *Skew Littlewood–Richardson Rule*, see section \[SkewLRrule\]. In [@LamLauveSottile Lemma 1.1] Lam, Lauve, and Sottile obtain a more general version of Formula valid for arbitrary pairs of dual Hopf algebras.
As mentioned in the introduction, Ira Gessel, [@Gessel], established special cases of and when the Schur functions are indexed by one-row or one-column shapes. He showed that $$\begin{aligned}
&&D_n U_m = \sum_i U_{m-i} D_{n-i}\,, \\
&&U_m D_n= D_n U_m - D_{n-1} U_{m-1}\,,\\
&&D_{(1^n)} U_m= U_m D_{(1^n)} + U_{m-1} D_{(1^{n-1})}.\end{aligned}$$
Since ${K_{\beta}}$ and ${\overline{K}_{\beta}}$ are self-adjoint, and $U_{\alpha}$ and $D_{\alpha}$ are adjoint of each other, and are obtained from and respectively by taking adjoints.
An interesting and elegant way of stating some of the results of Theorem \[thm:main\] is in terms of commutators.
\[commutators\] For any two partitions $\alpha$ and $\beta$, we have $$\begin{aligned}
[{D_{\beta}}, {U_{\alpha}}]
&= \sum_{\lambda \neq (0)} {U_{\alpha/\lambda}} {D_{\beta/\lambda}} = \sum_{\lambda \neq (0)} (-1)^{|\lambda|-1} {D_{\beta/\lambda'}} {U_{\alpha/{\lambda}}}\,,\\
[{\overline{K}_{\beta}}, {U_{\alpha}}]&=
\sum_{(\tau,\nu) \neq ((0),\beta)} {U_{(s_{\beta/\nu} \ast s_{\tau}) s_{\alpha/\tau}}} {\overline{K}_{\nu}}\,, \\
[{D_{\alpha}}, {\overline{K}_{\beta}}] &=
\sum_{(\tau,\nu) \neq ((0),\beta)} {\overline{K}_{\nu}} {D_{(s_{\beta/\nu} \ast s_{\tau}) s_{\alpha/\tau}}}\,,
$$ where $(0)$ denotes the empty partition.
Proof of the identities {#sec:proofs}
=======================
In this section, we prove all six identities of Theorem \[thm:main\]. We begin with an overview of the method of proof. Let $P$ be any of the families of operators ${U_{}}$, ${D_{}}$, ${K_{}}$ and ${\overline{K}_{}}$. These operators act on ${\textit{Sym}}={\textit{Sym}}(X)$, the symmetric functions in some alphabet $X$. Introduce an auxiliary alphabet $A$. The Schur generating series of $P$ is defined as $$\sum_{\lambda} s_{\lambda}[A] P_{\lambda} .$$ This can be interpreted as the linear map that sends any symmetric function $g \in {\textit{Sym}}(X)$ to the expression $$\sum_{\lambda} s_{\lambda}[A] P_{\lambda}(g).$$ For each of the four operators under consideration, the effect of the Schur generating function operator is described nicely by means of operations on alphabets (Lemma \[lemma:effects\]). The identities of Theorem \[thm:main\] are derived at the level of generating series. The result is then recovered by extracting coefficients by means of the appropriate scalar products.
Preliminary: operations on alphabets
------------------------------------
Let $X=\{x_1, x_2, \ldots\}$ be the underlying alphabet for the symmetric functions in ${\textit{Sym}}$. Any infinite alphabet $A$ gives rise to a copy ${\textit{Sym}}(A)$ of ${\textit{Sym}}$. In this copy, the corresponding scalar product will be denoted by ${\left\langle\, \, \middle| \, \, \right\rangle}_A$ and the element corresponding to $f\in{\textit{Sym}}$, by $f[A]$. Accordingly, the scalar product ${\left\langle\, \, \middle| \, \, \right\rangle}$ of ${\textit{Sym}}$ and elements $f\in{\textit{Sym}}$ will be denoted sometimes by ${\left\langle\, \, \middle| \, \, \right\rangle}_X$ and $f[X]$.
If $A$ and $B$ are two alphabets, the tensor product ${\textit{Sym}}(A) \otimes {\textit{Sym}}(B)$ is endowed with the induced scalar product ${\left\langle\, \, \middle| \, \, \right\rangle}_{A,B}$.
Both the Kronecker product $\ast$ and the adjoint $D_f$ of the operator of multiplication by a symmetric function $f$ will be only considered with respect to ${\textit{Sym}}={\textit{Sym}}(X)$.
Given a morphism of algebras $A$ from ${\textit{Sym}}$ to some commutative algebra $\mathcal{R}$, it will be convenient to write it as $f \mapsto f[A]$ (rather than $f \mapsto A(f)$) for any *morphism of algebras* and consider it as a “specialization at the virtual alphabet $A$.”
Since the power sum symmetric functions $p_k$ ($k \geq 1$) generate ${\textit{Sym}}$ and are algebraically independent, the map $$\label{bijection}
A \mapsto (p_1[A], p_2[A], \ldots)$$ is a bijection from the set of all morphisms of algebras from ${\textit{Sym}}$ to $\mathcal{R}$ to the set of infinite sequences of elements from $\mathcal{R}$. This set of sequences is endowed with its operations of component-wise sum and product, and multiplication by a scalar. The bijection is used to lift these operations to the set of morphisms from ${\textit{Sym}}$ to $\mathcal{R}$. This defines expressions like $f[A+B]$ and $f[AB]$, where $f$ is a symmetric function and $A$ and $B$ are two “virtual alphabets,” and more general expressions $f[P(A,B,\ldots)]$ where $P(A,B,\ldots)$ is a polynomial in several virtual alphabets $A$, $B$ …with coefficients in the base field. Note that, by definition, for any power sum $p_k$ ($k \geq 1$), virtual alphabets $A$ and $B$, and scalar $z$, $$p_k[A+B]=p_k[A]+p_k[B], \qquad p_k[AB]=p_k[A] \cdot p_k[B], \qquad p_k[zA]=z \; p_k[A].$$ In our calculations below, the morphism $f \mapsto f[1]$ will appear: it is the specialization at $x_1=1$, $x_2=0$, $x_3=0$ …, sending each $p_k$ to $1$. The morphism $f \mapsto f[X]$ is just the identity of ${\textit{Sym}}$. The morphism $f \mapsto f[X^{\perp}] = f^{\perp}=D_f$ associates to $f$ the adjoint of the operator “multiplication by $f$”.
Let ${\sigma}$ be the generating series for the complete homogeneous symmetric functions $h_n$, meaning ${\sigma}=\sum_{n=0}^{\infty} h_n$, where $h_0=1$. Recall from [@MacdonaldBook I.§2] that we also have $${\sigma}=\exp\left(\sum_{k=1}^{\infty} \frac{p_k}{k} \right)=\sum_{\lambda} \frac{p_{\lambda}}{z_{\lambda}}$$ where the last sum is carried over all partitions $\lambda$.
We will make use of the following known identities, whose proofs we include for the sake of completeness.
\[properties:ss\] Let $A$ and $B$ be any two alphabets, and $f$ and $g$ be any two symmetric functions. Then we have the following identities. $$\begin{aligned}
&{\sigma}[A+B]={\sigma}[A] {\sigma}[B]&& \label{factorization}\\
&{\sigma}[AB]=\sum_{\lambda} s_{\lambda}[A] s_{\lambda}[B] && \text{ (Cauchy Identity)} \label{Cauchy}\\
&{\sigma}[-AB]=\sum_{\lambda} (-1)^{|\lambda|} s_{\lambda'}[A] s_{\lambda}[B] && \label{dual Cauchy}\\
&{D_{{\sigma}[AX]}}(f[X])=f[X+A] && \label{sum}\\
&{\sigma}[AX] \ast f[X] =f[AX] && \label{product}\\
&{\left\langle\, f[AX] \, \middle| \, g[X] \, \right\rangle}_X=(f\ast g)[A] && \label{scalar}\\
&{\left\langle\, {\sigma}[AB] \, \middle| \, g[B] \, \right\rangle}_B=g[A] && \text{ (Reproducing Kernel)}\label{reproducing}\end{aligned}$$
Generating series
-----------------
Let $P$ be any of the families of operators $U$, $D$, $K$ and ${\overline{K}_{}}$. Introduce an auxiliary alphabet $A$ and the following generating series for $P$: $$\sum_{\lambda} s_{\lambda}[A] P_{\lambda}.$$ We use the linearity of $f \mapsto P_f$ to simplify this expression as follows: $$\begin{aligned}
\sum_{\lambda} s_{\lambda}[A] P_{\lambda}
= \sum_{\lambda} s_{\lambda}[A] P_{s_\lambda[X]}
= P_{\sum_{\lambda} s_{\lambda}[A] s_\lambda[X] }
= P_{{\sigma}[AX]}\end{aligned}$$ by the Cauchy Identity .
Note that any operator $P_{\lambda}$ can be recovered from the generating series with a coefficient extraction by means of a scalar product: $$P_{\lambda}(f) = {\left\langle\, P_{{\sigma}[AX]}(f[X]) \, \middle| \, s_{\lambda}[A] \, \right\rangle}_A\,.$$
The generating series $P_{{\sigma}[AX]}$ also acts linearly on symmetric functions. The following lemma describes the effect of all four generating series ${U_{{\sigma}[AX]}}$, ${D_{{\sigma}[AX]}}$, ${K_{{\sigma}[AX]}}$ and ${\overline{K}_{{\sigma}[AX]}}$.
\[lemma:effects\] Let $f[X]$ be any symmetric function. $$\begin{aligned}
{U_{{\sigma}[AX]}}(f[X]) &= {\sigma}[AX] \cdot f[X] \label{effectU}\\
{D_{{\sigma}[AX]}}(f[X]) &= f[X+A] \label{effectD} \\
{K_{{\sigma}[AX]}}(f[X]) &= f[AX] \label{effectK} \\
{\overline{K}_{{\sigma}[AX]}}(f[X]) &= {\sigma}[-A] \cdot f[X(A+1)]\label{effectKbar}\end{aligned}$$
Equation is straightforward. Equation is . Equation is . Let us prove . For any symmetric functions $f$ and $g$, $
{\overline{K}_{f}}(g)=\Gamma_1 f \ast g
$, where $\Gamma_1$ is the vertex operator: $$\label{Gamma-VO}
\Gamma_1=\left( \sum_{i=0}^{\infty} U_{(i)}\right) \left( \sum_{j=0}^{\infty} (-1)^j {D_{(1^j)}}\right)$$
We will make use of the following identity (see [@ScharfThibonWybourne §3]. For a combinatorial approach to this identity see [@RosasGamma] ): $$\label{vertex}
\Gamma_1 f = {\sigma}[X] f[X-1].$$ Therefore, we have [$$\begin{aligned}
{\overline{K}_{{\sigma}[AX]}}(f)
&=\left(\Gamma_1 \; {\sigma}[AX]\right) \ast f[X] && \\
&=\left({\sigma}[X] \; {\sigma}[A(X-1)]\right) \ast f[X] && \\
&={\sigma}[X+A(X-1)] \ast f[X] && \text{by \eqref{factorization},}\\
&={\sigma}[X(A+1)-A] \ast f[X] && \\
&={\sigma}[-A] \; \left({\sigma}[X(A+1)] \ast f[X] \right)&& \text{by \eqref{factorization} again,}\\
&={\sigma}[-A] \cdot f[X(A+1)] && \text{by \eqref{product}.}\\\end{aligned}$$ ]{}
Operators $U$ and $D$
---------------------
Let us prove the first two identities in Theorem \[thm:main\], which are , $$\begin{aligned}
{D_{\beta}} \; {U_{\alpha}} &=
\sum_{\lambda} {U_{{\alpha/\lambda}}} {D_{{\beta/\lambda}}}\,, $$ and , $$\begin{aligned}
{U_{\alpha}} \; {D_{\beta}}&=
\sum_{\lambda} (-1)^{|\lambda|} {D_{{\beta/\lambda'}}} {U_{{\alpha/\lambda}}}\,. $$
To this aim, we first establish commutation relations between the generating series of the operators $U$ and $D$.
Let $A$ and $B$ be two alphabets. We have [$$\begin{aligned}
{{D_{{\sigma}[B X]}}}{{U_{{\sigma}[A X]}}}&={\sigma}[AB] {{U_{{\sigma}[A X]}}}{{D_{{\sigma}[B X]}}}, \label{eqDU}\\
{{U_{{\sigma}[A X]}}}{{D_{{\sigma}[B X]}}}&={\sigma}[-AB] {{D_{{\sigma}[B X]}}}{{U_{{\sigma}[A X]}}}.\label{eqUD}\end{aligned}$$ ]{}
The second of these identities is obtained straightforwardly from the first after noting that ${\sigma}[-AB]$ is the inverse of ${\sigma}[AB]$ (see ).
Let us prove using Lemmas \[properties:ss\] and \[lemma:effects\]. We have, for any symmetric function $f[X]$, [$$\begin{aligned}
{{D_{{\sigma}[B X]}}} {{U_{{\sigma}[A X]}}} (f[X]) &= {{D_{{\sigma}[B X]}}} ({\sigma}[AX] f[X]) && \\
&= {\sigma}[A(X+B)] f[X+B] && \text{by \eqref{effectD},}\\
&= {\sigma}[AB]{\sigma}[AX]f[X+B] && \text{by \eqref{factorization},}\\
&= {\sigma}[AB]{\sigma}[AX]{D_{{\sigma}[BX]}}(f[X]) && \text{by \eqref{effectD},}\\
&= {\sigma}[AB]{{U_{{\sigma}[A X]}}}{{D_{{\sigma}[B X]}}}(f[X]).&&\end{aligned}$$ ]{}
We will use that, since ${U_{}}$ and ${D_{}}: {\textit{Sym}}\rightarrow \operatorname{End}({\textit{Sym}})$ are morphisms of algebras, we can write, for any symmetric function $f$, that ${U_{f}}=f[U]$ and ${D_{f}}=f[D]$. In particular, for the generating series, we have ${U_{{\sigma}[AX]}}={\sigma}[AU]$ and ${D_{{\sigma}[BX]}}={\sigma}[BD]$.
In , the operator ${D_{\beta}}{U_{\alpha}}$ is the coefficient of $s_{\alpha}[A]s_{\beta}[B]$ in the expansion in the Schur basis of ${\sigma}[AB] {{U_{{\sigma}[A X]}}}{{D_{{\sigma}[B X]}}}$, which is extracted by performing the scalar product with $s_{\alpha}[A]s_{\beta}[B]$. Thus [$$\begin{aligned}
{D_{\beta}}{U_{\alpha}}
&={\left\langle\, {\sigma}[AB] {{U_{{\sigma}[A X]}}} {{D_{{\sigma}[B X]}}} \, \middle| \, s_{\alpha}[A] s_{\beta}[B] \, \right\rangle}_{A,B} && \\
&=\sum_{\lambda} {\left\langle\, s_{\lambda}[A] s_{\lambda}[B] {{U_{{\sigma}[A X]}}} {{D_{{\sigma}[B X]}}} \, \middle| \, s_{\alpha}[A] s_{\beta}[B] \, \right\rangle}_{A,B} && \text{by \eqref{Cauchy},}\\
&=\sum_{\lambda} {\left\langle\, s_{\lambda}[A] {{U_{{\sigma}[A X]}}} \, \middle| \, s_{\alpha}[A] \, \right\rangle}_A {\left\langle\, s_{\lambda}[B] {{D_{{\sigma}[B X]}}} \, \middle| \, s_{\beta}[B] \, \right\rangle}_{B} && \\
&=\sum_{\lambda} {\left\langle\, {{U_{{\sigma}[A X]}}} \, \middle| \, s_{\alpha/\lambda}[A] \, \right\rangle}_A {\left\langle\, {{D_{{\sigma}[B X]}}} \, \middle| \, s_{\beta/\lambda}[B] \, \right\rangle}_{B} && \\
&=\sum_{\lambda} {\left\langle\, {\sigma}[AU] \, \middle| \, s_{\alpha/\lambda}[A] \, \right\rangle}_A {\left\langle\, {\sigma}[BD] \, \middle| \, s_{\beta/\lambda}[B] \, \right\rangle}_{B} && \\
&=\sum_{\lambda} s_{\alpha/\lambda}[U] s_{\beta/\lambda}[D] && \text{ by \eqref{reproducing},}\\
&=\sum_{\lambda} {U_{\alpha/\lambda}} {D_{\beta/\lambda}}. && $$ ]{}
Identity is derived from analogously.
Operators $U$ and $K$
---------------------
We now turn to the proof of : $${K_{\beta}} {U_{\alpha}} =\sum_{\lambda} {U_{s_{\beta/\lambda} \ast s_{\alpha}}} {K_{\lambda}} $$ (The identity follows straightforwardly from by taking adjoints).
Again we consider first commutation relations for the generating series of the operators $K$ and $U$.
Let $A$ and $B$ be two alphabets. We have $$\label{KKUU}
{{K_{{\sigma}[B X]}}} {{U_{{\sigma}[A X]}}}= {{U_{{\sigma}[AB X]}}} {{K_{{\sigma}[B X]}}}.$$
We have, for any symmetric function $f$, [$$\begin{aligned}
{{K_{{\sigma}[B X]}}} {{U_{{\sigma}[A X]}}} (f[X]) &= {{K_{{\sigma}[B X]}}}({\sigma}[AX] f[X])&&\\
&= {\sigma}[ABX] f[BX] && \text{by \eqref{effectK},}\\
&= {{U_{{\sigma}[AB X]}}}(f[BX]) && \\
&= {{U_{{\sigma}[AB X]}}}{{K_{{\sigma}[B X]}}}(f[X])&& \text{by \eqref{effectK}.}\end{aligned}$$ ]{}
We now get ${K_{\beta}}{U_{\alpha}}$ from by extracting the coefficient of $s_{\beta}[B] s_{\alpha}[A]$ in its expansion in terms of Schur functions: [$$\begin{aligned}
{K_{\beta}} {U_{\alpha}}
&= {\left\langle\, {{U_{{\sigma}[AB X]}}} {{K_{{\sigma}[B X]}}} \, \middle| \, s_{\beta}[B] s_{\alpha}[A] \, \right\rangle}_{A,B} && \\
&= \sum_{\lambda} {\left\langle\, {{U_{{\sigma}[AB X]}}} s_{\lambda}[B] \, \middle| \, s_{\beta}[B] s_{\alpha}[A] \, \right\rangle}_{A,B} {K_{\lambda}}&& \\
&= \sum_{\lambda} {\left\langle\, {{U_{{\sigma}[AB X]}}} \, \middle| \, s_{\beta/\lambda}[B] s_{\alpha}[A] \, \right\rangle}_{A,B} {K_{\lambda}}&& \\
&= \sum_{\lambda} {\left\langle\, {\left\langle\, {{U_{{\sigma}[AB X]}}} \, \middle| \, s_{\alpha}[A] \, \right\rangle}_A \, \middle| \, s_{\beta/\lambda}[B] \, \right\rangle}_{B} {K_{\lambda}}&& \\
&= \sum_{\lambda} {\left\langle\, {\left\langle\, {\sigma}[ABU] \, \middle| \, s_{\alpha}[A] \, \right\rangle}_A \, \middle| \, s_{\beta/\lambda}[B] \, \right\rangle}_{B} {K_{\lambda}}&& \\
&= \sum_{\lambda} {\left\langle\, s_{\alpha}[BU] \, \middle| \, s_{\beta/\lambda}[B] \, \right\rangle}_{B} {K_{\lambda}}&& \text{by \eqref{reproducing},}\\
&= \sum_{\lambda} (s_{\alpha}\ast s_{\beta/\lambda})[U] {K_{\lambda}}&& \text{by \eqref{scalar},}\\
&= \sum_{\lambda} {U_{s_{\alpha}\ast s_{\beta/\lambda}}} {K_{\lambda}}.&& $$ ]{}
Operators $U$ and ${\overline{K}_{}}$
-------------------------------------
We now proceed to proving : $${\overline{K}_{\beta}} {U_{\alpha}}=
\sum_{\tau,\nu} {U_{(s_{\beta/\nu} \ast s_{\tau}) s_{\alpha/\tau}}} {\overline{K}_{\nu}}\,. $$ (The identity is deduced by taking adjoints). Again, we first consider commutation relations for the generating series of the families of operators involved.
Let $A$ and $B$ be two alphabets. We have $$\label{GKbarU}
{\overline{K}_{{\sigma}[BX]}} {{U_{{\sigma}[A X]}}} = {{U_{{\sigma}[A(B+1) X]}}} {\overline{K}_{{\sigma}[BX]}}.$$
We have, for any symmetric function $f$, [$$\begin{aligned}
{\overline{K}_{{\sigma}[BX]}}{{U_{{\sigma}[A X]}}}(f)
&={\overline{K}_{{\sigma}[BX]}}({\sigma}[AX] f[X])&&\\
&={\sigma}[-B]{\sigma}[AX(B+1)] f[X(B+1)]&& \text{by \eqref{effectKbar}, }\\
&={\sigma}[AX(B+1)] {\sigma}[-B] f[X(B+1)]&&\\
&={\sigma}[AX(B+1)] {\overline{K}_{{\sigma}[BX]}}(f)&& \text{by \eqref{effectKbar} again, }\\
&={{U_{{\sigma}[A(B+1) X]}}} {\overline{K}_{{\sigma}[BX]}}(f).&&\end{aligned}$$ ]{}
From we extract the term ${\overline{K}_{\beta}}{U_{\alpha}}$ by taking scalar product with $s_{\alpha}[A]\;s_{\beta}[B]$. This yields $${\overline{K}_{\beta}}{U_{\alpha}}
={\left\langle\, {{U_{{\sigma}[A(B+1) X]}}} {\overline{K}_{{\sigma}[BX]}} \, \middle| \, s_{\alpha}[A] s_{\beta}[B] \, \right\rangle}_{A,B}.$$ Expanding in the scalar product the generating function ${\overline{K}_{{\sigma}[BX]}}$, we get $${\overline{K}_{\beta}}{U_{\alpha}}
=\sum_{\nu}{\left\langle\, {{U_{{\sigma}[A(B+1) X]}}} s_{\nu}[B] \, \middle| \, s_{\alpha}[A] s_{\beta}[B] \, \right\rangle}_{A,B} {\overline{K}_{\nu}}.$$ This simplifies as follows: [$$\begin{aligned}
{\overline{K}_{\beta}}{U_{\alpha}}
&=\sum_{\nu}{\left\langle\, {\left\langle\, {{U_{{\sigma}[A(B+1) X]}}} \, \middle| \, s_{\alpha[A]} \, \right\rangle}_A s_{\nu}[B] \, \middle| \, s_{\beta}[B] \, \right\rangle}_{B} {\overline{K}_{\nu}}&& \\
&=\sum_{\nu}{\left\langle\, {\left\langle\, {\sigma}[A(B+1)U] \, \middle| \, s_{\alpha[A]} \, \right\rangle}_A s_{\nu}[B] \, \middle| \, s_{\beta}[B] \, \right\rangle}_{B} {\overline{K}_{\nu}}&& \\
&=\sum_{\nu}{\left\langle\, s_{\alpha[(B+1)U]} s_{\nu}[B] \, \middle| \, s_{\beta}[B] \, \right\rangle}_{B} {\overline{K}_{\nu}}&& \text{by \eqref{reproducing}, } \\
&=\sum_{\nu}{\left\langle\, s_{\alpha[BU+U]} s_{\nu}[B] \, \middle| \, s_{\beta}[B] \, \right\rangle}_{B} {\overline{K}_{\nu}}&& \\
&=\sum_{\nu}{\left\langle\, \sum_{\tau} s_{\tau}[BU] s_{\alpha/\tau}[U] \, \middle| \, s_{\beta/\nu}[B] \, \right\rangle}_{B} {\overline{K}_{\nu}}&& \\
&=\sum_{\nu,\tau}{\left\langle\, s_{\tau}[BU] \, \middle| \, s_{\beta/\nu}[B] \, \right\rangle}_{B} s_{\alpha/\tau}[U]{\overline{K}_{\nu}}&& \\
&=\sum_{\nu,\tau} (s_{\tau}\ast s_{\beta/\nu})[U] s_{\alpha/\tau}[U]{\overline{K}_{\nu}}&& \text{by \eqref{scalar},}\\
&=\sum_{\nu,\tau} {U_{(s_{\tau} \ast s_{\beta/\nu})s_{\alpha/\tau}}}{\overline{K}_{\nu}}\,.&& $$ ]{}
In Section \[SkewKronecker\] we present as an application a combinatorial rule for the Kronecker product of any skew Schur function by $s_{(n-1,1)}$. Other interesting particular cases of correspond to the cases when $\lambda=(k)$ (Kronecker product with a two-row shape) and $\lambda=(1^k)$ (Kronecker product with a hook), where we get: $${\overline{K}_{(k)}} {U_{\alpha}} =
\sum_{j=0}^k \left(\sum_{\rho \vdash k-j} {U_{\alpha/\rho}} {U_{\rho}}\right) {\overline{K}_{(j)}},\qquad
{\overline{K}_{(1^k)}} {U_{\alpha}} =
\sum_{j=0}^k \left(\sum_{\rho \vdash k-j} {U_{\alpha/\rho}} {U_{\rho'}}\right) {\overline{K}_{(1^j)}}\,.$$ Setting $n=|\alpha|$ and $m=|\gamma|$, the same identities can be stated as: $$\begin{aligned}
s_{(n+m-k,k)}\ast(s_{\alpha}s_{\gamma}) =
\sum_{j=0}^k \left(\sum_{\rho \vdash k-j} s_{\alpha/\rho} s_{\rho} \right)( s_{\gamma}\ast s_{(m-j,j)}),\\
s_{(n+m-k,1^k)} \ast (s_{\alpha}s_{\gamma}) =
\sum_{j=0}^k \left(\sum_{\rho \vdash k-j} s_{\alpha/\rho} s_{\rho'}\right) (s_{\gamma} \ast s_{(m-j,1^j)}).\end{aligned}$$ The terms of the form $\sum_{\rho \vdash q} s_{\alpha/\rho} s_{\rho}$ and $\sum_{\rho \vdash q} s_{\alpha/\rho} s_{\rho'}$ that appear in these identities can take the following alternative forms, as a direct consequence of Littlewood’s Identity : $$\sum_{\rho \vdash q} s_{\alpha/\rho} s_{\rho} =s_{\alpha} \ast h_{(n-q,q)}, \quad \text{ and likewise }
\sum_{\rho \vdash q} s_{\alpha/\rho} s_{\rho'} =s_{\alpha} \ast (h_{n-q}e_q).$$
Uniqueness of expansions {#uniqueness}
========================
The identities in Theorem \[thm:main\] and Corollary \[thm:main-cor\] express some operators as linear combinations of operators ${U_{\mu}} {D_{\nu}}$, ${D_{\nu}} {U_{\mu}}$, ${U_{\mu}} {K_{\nu}}$, etc. Are such expressions unique?
To answer this question, we associate to any pair $P$, $Q$ of linear maps from ${\textit{Sym}}$ to $\operatorname{End}({\textit{Sym}})$ a generating series depending on four independent alphabets $X$, $A$, $B$, $T$. This generating series is $$\sum_{\alpha, \beta,\lambda} P_{\alpha}(Q_{\beta}(s_{\lambda}[X])) s_{\alpha}[A] s_{\beta}[B] s_{\lambda}[T].$$ It can be written more concisely as $$P_{{\sigma}[AX]} Q_{{\sigma}[BX]} ({\sigma}[XT]).$$ We also associate to the pair $P$, $Q$ the linear map $\Phi_{P,Q}$ from ${\textit{Sym}}(A) \otimes_Q {\textit{Sym}}(B)$ to the set of formal series of the form $\sum_{\alpha, \beta} a_{\alpha,\beta} s_{\alpha}[X] s_{\beta}[T]$, defined on simple tensors by: $$\Phi_{P,Q}( f[A] g[B]) = {\left\langle\, P_{{\sigma}[AX]} Q_{{\sigma}[BX]} ({\sigma}[XT]) \, \middle| \, f[A] g[B] \, \right\rangle}_{A,B}$$
Let us say that *finite expansions with respect to $(P,Q)$ are unique* if, for any $M \in \operatorname{End}({\textit{Sym}})$, there is at most one expansion $$M = \sum_{\alpha, \beta} a_{\alpha, \beta} P_{\alpha} Q_{\beta}.$$ That is, finite expansions with respect to $(P,Q)$ are unique when the operators $P_{\alpha} Q_{\beta}$, for $\alpha$ and $\beta$ partitions, are linearly independent.
\[finite expansions\] Finite expansions with respect to the pairs $(U,D)$, $(D,U)$, $(U,K)$, $(K,D)$, $(U,{\overline{K}_{}})$ and $({\overline{K}_{}},D)$ are unique.
In contrast, observe that expansions with respect to $(K,U)$ or to $(D,K)$ are not unique. For instance we have the relation ${K_{p_2}} {U_{p_1}} =0$, that is straightforwardly equivalent to the relation ${K_{2}}{U_{1}} = {K_{1,1}}{U_{1}}$. Taking adjoints, we have the relation ${D_{1}}{K_{2}} = {D_{1}} {K_{1,1}}$.
Proposition \[finite expansions\] will be proved using the following lemma.
\[coordinate\_independent\] Let $P$ and $Q$ of linear maps from ${\textit{Sym}}$ to $\operatorname{End}({\textit{Sym}})$.
1. Finite expansions with respect to $(P,Q)$ are unique if and only if $\Phi_{P,Q}$ is injective.
2. Let $M \in \operatorname{End}({\textit{Sym}})$.
- The operator $M$ is in the linear span of the operators $P_f Q_g$, for $f$, $g \in {\textit{Sym}}$, if and only if $M({\sigma}[XT])$ lies in the image of $\Phi_{P,Q}$.
- If $M({\sigma}[XT]) = \Phi_{P,Q}(F)$ then $M$ is the image of $F$ under the linear map defined on the Schur basis by $s_{\alpha}[A] s_{\beta}[B] \mapsto P_{\alpha} Q_{\beta}$.
We start with a computation: let $M \in \operatorname{End}({\textit{Sym}})$. We have: $$\begin{aligned}
M = \sum_{\alpha, \beta} a_{\alpha, \beta} P_{\alpha} Q_{\beta}
&\Leftrightarrow \text{ for any partition $\lambda$, } M(s_{\lambda}[X]) = \sum_{\alpha, \beta} a_{\alpha, \beta} P_{\alpha} Q_{\beta} (s_{\lambda}[X]) \\
&\Leftrightarrow \sum_{\lambda} M(s_{\lambda}[X]) s_{\lambda}[T] = \sum_{\alpha, \beta, \lambda} a_{\alpha, \beta} P_{\alpha} Q_{\beta} (s_{\lambda}[X]) s_{\lambda}[T] \\
&\Leftrightarrow M({\sigma}[XT]) = \sum_{\alpha, \beta} a_{\alpha, \beta} P_{\alpha} Q_{\beta} ({\sigma}[XT]) \\
&\Leftrightarrow M({\sigma}[XT]) = \Phi_{P, Q}\left(\sum_{\alpha,\beta} a_{\alpha, \beta} s_{\alpha}[A] s_{\beta}[B]\right).\end{aligned}$$ This proves (2), since the linear span of the operators $P_f Q_g$, for $f$, $g \in {\textit{Sym}}$, is also the linear span of the operators $P_{\alpha} Q_{\beta}$, for $\alpha$ and $\beta$ partitions.
To obtain (1), take $M=0$ in the above equivalence.
For each of the pairs $(U,D)$, $(D,U)$, $(U,K)$ and $(U,{\overline{K}_{}})$, we compute the corresponding generating series by means of Lemma \[lemma:effects\]. Next we deduce a description of the corresponding map $\Phi$ to show it is injective. The uniqueness of finite expansions follows then from (1) in Lemma \[coordinate\_independent\].
For $(U,D)$, the detail of the calculation is as follows. The generating series is $$\begin{aligned}
U_{{\sigma}[AX]} D_{{\sigma}[BX]} ({\sigma}[XT])
&=U_{{\sigma}[AX]} ({\sigma}[(X+B)T], \\
&={\sigma}[AX] {\sigma}[(X+B)T], \\
&={\sigma}[AX+XT+BT].\end{aligned}$$
We deduce from this a formula for $\Phi_{U,D}$. Let $f$ and $g$ be any symmetric functions. We have $$\begin{aligned}
\Phi_{U,D}(f[A] g[B])
&= {\left\langle\, {\sigma}[AX+XT+BT] \, \middle| \, f[A] g[B] \, \right\rangle}_{A,B}&\\
&={\left\langle\, {\sigma}[AX]{\sigma}[XT]{\sigma}[BT] \, \middle| \, f[A] g[B] \, \right\rangle}_{A,B}&\\
&={\sigma}[XT ]{\left\langle\, {\sigma}[AX]{\sigma}[BT] \, \middle| \, f[A] g[B] \, \right\rangle}_{A,B}&\\
&={\sigma}[XT ]{\left\langle\, {\sigma}[AX] \, \middle| \, f[A] \, \right\rangle}_A {\left\langle\, {\sigma}[BT] \, \middle| \, g[B] \, \right\rangle}_{B}&\\
&={\sigma}[XT ] f[X] g[T] & \text{ by \eqref{reproducing}.}\end{aligned}$$ This shows that $\Phi_{U,D}$ is injective, since the series ${\sigma}[XT]$ is invertible.
For the other three pairs, we skip the details of the calculations.
The generating series for $(D,U)$ is: $$D_{{\sigma}[BX]} U_{{\sigma}[AX]} ({\sigma}[XT]) = {\sigma}[(A+T)(B+X)].$$ The map $\Phi_{D,U}$ is $$f[A] g[B] \mapsto {\sigma}[XT] {\sigma}[X^{\perp} T^{\perp}] (f[X] gT] )$$ The map $\Phi_{D,U}$ is injective, since the series ${\sigma}[XT]$ is invertible, and the operator ${\sigma}[X^{\perp} T^{\perp}]$ is invertible as well (its inverse is ${\sigma}[- X^{\perp} T^{\perp}]$).
The generating series for $(U,K)$ is $$U_{{\sigma}[AX]} K_{{\sigma}[BX]} ({\sigma}[XT]) = {\sigma}[B(X+A)T].$$ The map $\Phi_{U,K}$ is $$f[A] g[B] \mapsto f[X] g[XT]$$ To check that $\Phi_{U,K}$ is injective, post–compose it with the specialization of $T$ at $T/X$: the map obtained is $f[A] g[B] \mapsto f[X] g[T]$, which is injective. Thus $\Phi_{U,K}$ is injective.
The generating series for $(U,{\overline{K}_{}})$ is $$U_{{\sigma}[AX]} {\overline{K}_{{\sigma}[BX]}} ({\sigma}[XT]) = {\sigma}[AX+B(XT-1)+XT].$$ The map $\Phi_{U,{\overline{K}_{}}}$ is $$f[A] g[B] \mapsto {\sigma}[XT] f[X] g[XT-1]$$ This map $\Phi_{U,{\overline{K}_{}}}$ is injective. Indeed, post–composing first with the product with the inverse of ${\sigma}[XT]$, and next by the specialization of $T$ at $(T+1)/X$, yields the map $f[A] g[B] \mapsto f[X] g[T]$, which is injective.
This proves the uniqueness of finite expansions with respect to $(U,D)$, $(D,U)$, $(U,K)$ and $(U,{\overline{K}_{}})$. The uniqueness of finite expansions with respect to $(K,D)$ and $({\overline{K}_{}},D)$ is obtained by taking adjoints.
The uniqueness of finite expansions with respect to $(U, D)$ and $(D,U)$ can alternatively be proved by switching to the basis of power sums. The algebra generated by the operators $U_g$ and $D_g$ is also generated by $1$, the $U_{p_k}$ and the $D_{p_k}$ ($k>0$). The following maps define an isomorphism between this algebra and, firstly, the bosonic creation and annihilation operator algebra (this appears for instance in [@JimboMiwa]) and, secondly, the Weyl algebra in infinitely many generators. $$\begin{aligned}
{U_{p_k}} &\mapsto a_k^{\dagger} \mapsto \widehat{x_k}\,,\\
{D_{p_k}} &\mapsto k\; a_k \mapsto k\;\frac{\partial \phantom{F}}{\partial x_k}\,,\end{aligned}$$ where the $a_k$ are the creation operators, and the $a_k^{\dagger}$ are the annihilation operator. It is well-known that in the bosonic creation and annihilation operator algebra, the monomials in normal order $$(a_1^{\dagger})^{m_1} (a_2^{\dagger})^{m_2} \cdots a_1^{n_1} a_2^{n_2} \cdots$$ as well as the monomials in antinormal order $$a_1^{n_1} a_2^{n_2} \cdots (a_1^{\dagger})^{m_1} (a_2^{\dagger})^{m_2} \cdots$$ are linearly independent. This shows that the operators ${U_{p_\lambda}} {D_{p_{\mu}}}$ are linearly independent, and so are the operators $ {D_{p_{\mu}}} {U_{p_\lambda}}$. From this one deduces that finite expansions with respect to $(U,D)$ and with respect to $(D,U)$ are unique.
We finish this section with an expansion of the operators ${\overline{K}_{f}}$ in terms of operators $U_g$ and $D_g$.
From (2) in Lemma \[coordinate\_independent\], we get the following result.
\[KBf in U D\] Let $f$ be a symmetric function. The operator ${\overline{K}_{f}}$ lies in the subalgebra of $\operatorname{End}({\textit{Sym}})$ generated by the operators $U_g$ and $D_g$ (for $g \in {\textit{Sym}}$).
More precisely, $${\overline{K}_{f}} = \sum_{\lambda} {U_{f[X-1] \ast s_{\lambda}}} {D_{\lambda}}.$$
The subalgebra of $\operatorname{End}({\textit{Sym}})$ generated by the operators $U_g$ and $D_g$, for $g \in {\textit{Sym}}$, is the linear span of the operators $U_{\alpha} D_{\beta}$, for $\alpha$ and $\beta$ partitions. After (2) in Lemma \[coordinate\_independent\], and the calculations in the proof of Proposition \[finite expansions\], it is the set of operators $M$ such that $M({\sigma}[XT])$ is ${\sigma}[XT]$ times an element of ${\textit{Sym}}(X) \otimes_{{\mathbb{Q}}} {\textit{Sym}}(T)$.
After ,${\overline{K}_{f}}({\sigma}[XT])={\sigma}[X] f[X-1] * {\sigma}[XT]$, which is equal, after , to ${\sigma}[XT] f[XT-1]$. This proves Proposition \[KBf in U D\].
To get an explicit decomposition of ${\overline{K}_{f}}$, we decompose $f[XT-1]$ as an element of ${\textit{Sym}}(X) \otimes_{{\mathbb{Q}}} {\textit{Sym}}(T)$. We start with $$f[XT-1] = f[X-1] \ast \sigma[XT].$$ (The Kronecker product $\ast$ is relative to the symmetric functions in $X$). Thus, $$f[XT-1]
= f[X-1] \ast \sum_{\lambda} s_{\lambda}[X] s_{\lambda}[T]
= \sum_{\lambda} f[X-1]\ast s_{\lambda}[X] s_{\lambda}[T].$$ Therefore, $${\overline{K}_{f}}({\sigma}[XT]) = \Phi_{U,D}\left( \sum_{\lambda} f[A-1] \ast s_{\lambda}[A] s_{\lambda}[B]\right)$$ The conclusion comes by applying the second part of (2) in Lemma \[coordinate\_independent\].
For instance, for $f=h_{k}$ we have: $
{\overline{K}_{(k)}} = \sum_{\lambda \vdash k} {U_{\lambda}} {D_{\lambda}} - \sum_{\lambda \vdash k-1} {U_{\lambda}} {D_{\lambda}}
,
$ since $h_k[X-1]=h_k-h_{k-1}$. The simplest case is $k=1$. Here we obtain $${\overline{K}_{(1)}}={U_{(1)}}{D_{(1)}}-1\label{KB1}\,.$$ By the Pieri rule, for any partition $\alpha$, $${U_{(1)}}{D_{(1)}} s_{\alpha} = \sum_{\beta} s_{\beta},$$ where each term in the sum corresponds to a choice of a corner in the diagram of $\alpha$, that is removed, and then a choice of a box that is added, to give the diagram of a partition $\beta$. There are two cases: the box can be added where the corner was removed, or not. Accordingly the sum splits: $${U_{(1)}}{D_{(1)}} s_{\alpha} = {\#\mathrm{corners}(\alpha)} s_{\alpha} + \sum_{\beta \in {{\alpha^{\mp}}}} s_{\beta}.$$ Therefore $${\overline{K}_{(1)}} s_{\alpha}={U_{(1)}}{D_{(1)}} s_{\alpha}-s_{\alpha} = ({\#\mathrm{corners}(\alpha)}-1) s_{\alpha} + \sum_{\beta \in {{\alpha^{\mp}}}} s_{\beta}\,$$ which is .
Application to the skew Littlewood–Richardson rule {#SkewLRrule}
==================================================
In this section we present our first application of Theorem \[thm:main\]: a new proof of the skew Littlewood–Richardson rule as conjectured by Assaf and the second author [@AssafMcNamara] and proved by Lam, Lauve and Sottile [@LamLauveSottile]. As in [@LamLauveSottile], our starting point is . In [@LamLauveSottile], first an “algebraic skew Littlewood–Richardson rule” is derived, involving sums of products of Littlewood–Richardson coefficients. Then, the combinatorial skew Littlewood–Richardson rule is obtained by interpreting these Littlewood–Richardson coefficients as counting semistandard Young tableaux with given rectification. Our proof fits more closely to the statement of the skew Littlewood–Richardson rule: we avoid going through the algebraic skew Littlewood–Richardson rule and use the interpretation of the Littlewood–Richardson coefficients as counting semistandard Young tableaux with content and Yamanouchi constraints. Our proof appears in Subsections \[sub:combskewLRrule\] and \[sub:combskewLRrule2\] and is largely combinatorial.
For a positive integer $k$ and a partition $\gamma$, the classical Pieri rule [@Pieri] gives a simple and beautiful expression for the product $s_{(k)} s_\gamma$ as a sum of Schur functions. A *$k$-horizontal* (resp. *$k$-vertical*) *strip* is a skew shape with $k$ boxes that has at most one box in each column (resp. row). The Pieri rule states that $$s_{(k)} s_\gamma = \sum_\lambda s_{{\widehat{\gamma}}}\,,$$ where the sum is over all partitions ${\widehat{\gamma}}$ such that ${\widehat{\gamma}}/\gamma$ is a $k$-horizontal strip. In [@AssafMcNamara], Assaf and the second author generalized the Pieri rule to the setting of skew shapes as follows: $$\label{equ:skewpieri}
s_{(k)} s_{\gamma/\beta} = \sum_{i=0}^k (-1)^i \sum_{{\widehat{\gamma}},{\widecheck{\beta}}} s_{{\widehat{\gamma}}/{\widecheck{\beta}}}\ ,$$ where the sum is over all partitions ${\widehat{\gamma}}$ and ${\widecheck{\beta}}$ such that ${\widehat{\gamma}}/\gamma$ is a $(k-i)$-horizontal strip, and $\beta/{\widecheck{\beta}}$ is an $i$-vertical strip. We will use the skew Pieri rule with $k=1$ in Section \[SkewKronecker\].
For the next level of generality, it is natural to ask for a similarly combinatorial expression for $s_\alpha s_{\beta/\gamma}$ for any partition $\alpha$. Equation gives one expression, but it does not mimic in the sense that it does not give the answer as a signed sum of skew Schur functions. Instead, the skew Littlewood–Richardson rule [@LamLauveSottile] gives an expression for the even more general product $s_{\alpha/\delta} s_{\beta/\gamma}$ as a signed sum of skew Schur functions. In this section we will derive the skew Littlewood–Richardson rule from in the following way. In Subsection \[sub:combskewLRrule\], we will use a combinatorial approach to obtain from the skew Littlewood–Richardson rule in the case when $\delta$ is empty, and then we will use a linearity argument to derive the result for general $\delta$ in Subsection \[sub:combskewLRrule2\].
The combinatorial skew Littlewood–Richardson rule {#sub:skewLRrule}
-------------------------------------------------
In order to state the skew Littlewood–Richardson rule, we first need some terminology. As usual, a sequence of positive integers $\omega$ is said to be a *lattice permutation* if any prefix of $\omega$ contains at least as many appearances of $i$ as $i+1$, for all $i\geq1$. For a partition $\delta$, we will say that $\omega$ is a $\delta$-lattice permutation if the word obtained by prefixing $\omega$ with $\delta_1$ copies of 1 followed by $\delta_2$ copies of 2, etc., is a lattice permutation.
We will draw our Young tableaux in French notation, implying that the entries of an SSYT weakly increase along the rows and strictly increase up the columns. An *anti-semistandard Young tableau* (ASSYT) $T_1$ of shape $\alpha/\beta$ is a filling of the boxes of $\alpha/\beta$ so that the entries strictly decrease along the rows and weakly decrease up the columns. Equivalently, $T_1$ is an ASSYT if the tableau $(T_1')^r$ obtained by transposing $T_1$ and then rotating it 180$^\circ$ is an SSYT. The *reverse reading word of an SSYT* $T_2$ is defined as usual as the word obtained by reading right-to-left along the rows of $T_2$, taking the rows from bottom to top. In contrast, the *reverse reading word of an ASSYT* $T_1$ is the word obtained by reading up the columns of $T$, taking the columns from right-to-left. Equivalently, we can take the usual reverse reading work of the SSYT $(T_1')^r$. Given a pair of tableaux $(T_1, T_2)$, where $T_1$ is an ASSYT and $T_2$ is an SSYT, we define the *reverse reading word of the pair* as the concatenation of the reading word of $T_1$ with that of $T_2$. We will encounter such pairs as in the the figure below, where the entries in the bottom left form an ASSYT, and the entries above or to the right of the outlined skew shape form an SSYT. $$\label{equ:ASSYTSSYT}
\begin{tikzpicture}[scale=0.5]
\draw[thick] (3,0) -- (3,2) -- (0,2) -- (0,4) -- (1,4) -- (1,3) -- (4,3) -- (4,2) -- (5,2) -- (5,1) -- (7,1) -- (7,0) --cycle;
\draw (2.5, 0.5) node {2};
\draw (2.5, 1.5) node {1};
\draw (1.5, 0.5) node {3};
\draw (1.5, 1.5) node {3};
\draw (0.5, 1.5) node {5};
\draw (7.5, 0.5) node {2};
\draw (8.5, 0.5) node {4};
\draw (5.5, 1.5) node {1};
\draw (6.5, 1.5) node {4};
\draw (7.5, 1.5) node {4};
\draw (8.5, 1.5) node {5};
\draw (4.5, 2.5) node {3};
\draw (1.5, 3.5) node {5};
\draw (2.5, 3.5) node {6};
\end{tikzpicture}$$ The reverse reading word of $(T_1, T_2)$ shown in is 21335425441365, which is certainly not a lattice permutation but is a 5321-lattice permutation.
We are now ready to state the skew Littlewood–Richardson rule.
\[thm:skewLR\] For skew shapes $\alpha/\delta$ and $\gamma/\beta$, $$\label{eqskewLR}
s_{\alpha/\delta} s_{\gamma/\beta} = \sum_{\genfrac{}{}{0pt}{}{T_1 \in \mathrm{ASSYT}(\beta/{\widecheck{\beta}})}{T_2 \in \mathrm{SSYT}({\widehat{\gamma}}/\gamma)}} (-1)^{|\beta/{\widecheck{\beta}}|} s_{{\widehat{\gamma}}/{\widecheck{\beta}}}\ ,$$ where the sum is over all ASSYT $T_1$ of shape $\beta/{\widecheck{\beta}}$ for some ${\widecheck{\beta}} \subseteq \beta$, and SSYT $T_2$ of shape ${\widehat{\gamma}}/\gamma$ for some ${\widehat{\gamma}} \supseteq \gamma$, with the following properties:
1. the combined content of $T_1$ and $T_2$ is the component-wise difference $\alpha-\delta$, and
2. the reverse reading word of $(T_1,T_2)$ is a $\delta$-lattice permutation.
For example, the ASSYT and SSYT pair of contribute $-s_{9953/1}$ to the product $s_{755431/5321} s_{7541/33}$. Note that when $\beta$ and $\delta$ are empty, we recover the classical Littlewood–Richardson rule.
Recovering a special case of the combinatorial skew Littlewood–Richardson rule {#sub:combskewLRrule}
------------------------------------------------------------------------------
Our first step to reproving Theorem \[thm:skewLR\] is to start with and show it implies Theorem \[thm:skewLR\] in the case when $\delta=(0)$, the empty partition. Instead of , we work with the equivalent : $$s_{\alpha} s_{\gamma/\beta}=\sum_{\lambda} (-1)^{|\lambda|} {D_{\beta/\lambda'}} (s_{\alpha/\lambda}s_{\gamma}).$$ First, let us examine the product $s_{\alpha/\lambda}s_{\gamma}$ from the right–hand side, and expand it in terms of Schur functions. Note that only those $s_\nu$ with $\nu \supseteq \gamma$ will appear in the Schur expansion with nonzero coefficient. Thus we can write $$s_{\alpha/\lambda}s_{\gamma} = \sum_{{\widehat{\gamma}} \supseteq \gamma} a_{{\widehat{\gamma}}} s_{{\widehat{\gamma}}}\,.$$ We have $$a_{{\widehat{\gamma}}} = {\left\langle\, s_{{\widehat{\gamma}}} \, \middle| \, s_{\alpha/\lambda}s_{\gamma} \, \right\rangle} = {\left\langle\, s_{{\widehat{\gamma}}/\gamma} \, \middle| \, s_{\alpha/\lambda} \, \right\rangle} = {\left\langle\, s_{{\widehat{\gamma}}/\gamma} s_\lambda \, \middle| \, s_\alpha \, \right\rangle}.$$ The product $s_{{\widehat{\gamma}}/\gamma} s_\lambda$ is equal to the skew Schur function of the shape $({\widehat{\gamma}}/\gamma) \oplus \lambda$. (The notation $\oplus$ denotes that ${\widehat{\gamma}}/\gamma$ is positioned so that its bottom-right corner box is immediately northwest of the top-left corner box of $\lambda$.) Therefore, the coefficient $a_{{\widehat{\gamma}}}$ is equal to the number of Littlewood–Richardson fillings (LR-fillings) of that skew shape that have content $\alpha$. Any LR-filling of that shape must just fill the $i$th row of $\lambda$ with the number $i$, for all $i$. Thus $a_{{\widehat{\gamma}}}$ equals the number of SSYT of shape ${\widehat{\gamma}}/\gamma$ whose reverse reading word is a $\lambda$-lattice permutation and whose content is the component-wise difference $\alpha-\lambda$. Hence is equivalent to $$\label{equ:gammayamanouchi}
s_{\alpha} s_{\gamma/\beta} = \sum_{\lambda} (-1)^{|\lambda|} {D_{\beta/\lambda'}} \sum_{T_2} s_{{\widehat{\gamma}}}$$ where second sum is over all SSYT $T_2$ of shape ${\widehat{\gamma}}/\gamma$ for some ${\widehat{\gamma}}\supseteq \gamma$, and content $\alpha-\lambda$, whose reverse reading word is a $\lambda$-lattice permutation.
Next, we will examine the term $s_{\beta/\lambda'}$. The coefficient of $s_\nu$ in this term is exactly the Littlewood–Richardson coefficient $c^\beta_{\lambda'\nu}$, which is nonzero only if $\nu \subseteq \beta$. Thus we wish to determine the coefficient of $s_{{\widecheck{\beta}}}$ in $s_{\beta/\lambda'}$ when ${\widecheck{\beta}} \subseteq \beta$, which equals the number of LR-fillings of $\beta/{\widecheck{\beta}}$ of content $\lambda'$. We claim that such fillings $T$ are in a shape-preserving bijection with ASSYT that are lattice permutations of content $\lambda$ (as opposed to content $\lambda'$ previously). Indeed the bijection $\psi$ is defined as mapping the $i$th appearance (in the reverse reading word of the SSYT $T$) of the number $j$ to the number $i$, for all $i$ and $j$. For example, $$\Yvcentermath1
\young(34,233,:122,::1111)\ \ \longrightarrow\ \ \young(31,321,:521,::4321).$$ Then, one can check that $\psi$ has the following necessary properties.
- The inverse of $\psi$ is given by the ASSYT analogue of $\psi$: map the $j$th appearance (in the reverse reading word, now in the ASSYT sense) of the number $i$ to the number $j$, for all $i$ and $j$.
- The image $\psi(T)$ of an LR-filling $T$ is indeed an ASSYT whose reverse reading word is a lattice permutation.
- Such a $\psi(T)$ maps to an LR-filling under the inverse map.
- Both $\psi$ and its inverse transpose the content partition.
Thus is equivalent to $$s_{\alpha} s_{\gamma/\beta} = \sum_{\lambda} (-1)^{|\lambda|} \sum_{T_1} {D_{{\widecheck{\beta}}}}
\sum_{T_2} s_{{\widehat{\gamma}}} = \sum_{\lambda} (-1)^{|\lambda|} \sum_{T_1, T_2} s_{{\widehat{\gamma}}/{\widecheck{\beta}}}\,,$$ where the relevant sums are over all $T_1$ and $T_2$ such that
- $T_1$ is an ASSYT having content $\lambda$, whose reverse reading word is a lattice permutation, and with shape $\beta/{\widecheck{\beta}}$ for some ${\widecheck{\beta}} \subseteq \beta$, and
- $T_2$ is an SSYT having content $\alpha-\lambda$, a $\lambda$-lattice permutation as reverse reading word, and shape ${\widehat{\gamma}}/\gamma$ for some ${\widehat{\gamma}}\supseteq\gamma$.
Note that $T_1$ tells us that $|\lambda| = |\beta/{\widecheck{\beta}}|$, and we have arrived at Theorem \[thm:skewLR\] in the case when $\delta=(0)$.
Recovering the full combinatorial skew Littlewood–Richardson rule {#sub:combskewLRrule2}
-----------------------------------------------------------------
Our second step is to use a linearity argument to derive Theorem \[thm:skewLR\] for general $\delta$. For this, observe that the coefficient of $(-1)^{|\beta/{\widecheck{\beta}}|} s_{{\widehat{\gamma}}/{\widecheck{\beta}}}$ on the right-hand side of is the number of pairs of tableaux $(T_1, T_2)$ with $T_1$ an ASSYT and $T_2$ a SSYT, fulfilling conditions (a) and (b) in Theorem \[thm:skewLR\]. But $T_1$ being an ASSYT is equivalent to $(T_1')^r$ being an SSYT, and the reverse reading word of the ASSYT $T_1$ is defined so that $(T_1')^r$ has the same reverse reading word as an SSYT. Therefore, the coefficient of $(-1)^{|\beta/{\widecheck{\beta}}|} s_{{\widehat{\gamma}}/{\widecheck{\beta}}}$ on the right-hand side of equals the number of SSYT of shape $({\widehat{\gamma}}/\gamma) \oplus (\beta'/{\widecheck{\beta'}})^r$ and content $\alpha$ whose reverse reading word is a $\delta$–lattice permutation. This is the number of SSYT of shape $({\widehat{\gamma}}/\gamma) \oplus (\beta'/{\widecheck{\beta'}})^r \oplus \delta$ and content $\alpha$ whose reverse reading word is a lattice permutation. By the Littlewood–Richardson rule, this quantity equals the coefficient of $s_\alpha$ in the Schur expansion of $s_{({\widehat{\gamma}}/\gamma) \oplus (\beta'/{\widecheck{\beta'}})^r \oplus \delta}$. This skew Schur function being equal to the product $s_{{\widehat{\gamma}}/\gamma} s_{(\beta'/{\widecheck{\beta'}})^r}s_{\delta}$, this coefficient is equal to $${\left\langle\, s_{{\widehat{\gamma}}/\gamma} s_{(\beta'/{\widecheck{\beta'}})^r}s_\delta \, \middle| \, s_\alpha \, \right\rangle}\,,$$ which is equal to $${\left\langle\, s_{{\widehat{\gamma}}/\gamma} s_{(\beta'/{\widecheck{\beta'}})^r} \, \middle| \, s_{\alpha/\delta} \, \right\rangle}\,.$$ Therefore, is equivalent to $$\label{equ:altskewLR}
s_{\alpha/\delta} s_{\gamma/\beta} = \sum_{{\widehat{\gamma}}, {\widecheck{\beta'}}} (-1)^{|\beta/{\widecheck{\beta}}|} {\left\langle\, s_{{\widehat{\gamma}}/\gamma} s_{(\beta'/{\widecheck{\beta'}})^r} \, \middle| \, s_{\alpha/\delta} \, \right\rangle} s_{{\widehat{\gamma}}/{\widecheck{\beta}}}\,,$$ where the sums are over all partitions ${\widehat{\gamma}}$ and ${\widecheck{\beta'}}$ such that ${\widehat{\gamma}} \supseteq \gamma$ and ${\widecheck{\beta}} \subseteq \beta$.
The key observation is that is linear in $s_{\alpha/\delta}$. Given any partitions $\beta$ and $\gamma$, will be true for all partitions $\alpha$ and $\delta$ when $$f \cdot s_{\gamma/\beta} = \sum_{{\widehat{\gamma}}, {\widecheck{\beta'}}} (-1)^{|\beta/{\widecheck{\beta}}|} {\left\langle\, s_{{\widehat{\gamma}}/\gamma} s_{(\beta'/{\widecheck{\beta'}})^r} \, \middle| \, f \, \right\rangle} s_{{\widehat{\gamma}}/{\widecheck{\beta}}}\ ,$$ holds for any symmetric function $f$. Since the Schur functions form a basis for the space of symmetric functions, it is enough to check it for $f=s_\alpha$, all partitions $\alpha$. This is what was done in Subsection \[sub:combskewLRrule\].
A combinatorial interpretation for the Kronecker product of a skew Schur function by $s_{(n-1,1)}$. {#SkewKronecker}
===================================================================================================
As another application of the identities of Section \[sec:commutations\], our goal for this section is to derive a combinatorial formula for Kronecker products involving skew Schur functions. Let $\alpha$ be a partition of $n$, and let us speak of partitions and their Young diagrams interchangeably.
A well-known case of ${\overline{K}_{\lambda}}$ is when $\lambda=(1)$. We pause to describe an identity that we will generalize in Section \[SkewKronecker\] using the normal ordering relations. A *corner* of $\alpha$ is a box of $\alpha$ whose removal results in another partition, and we denote by ${\#\mathrm{corners}(\alpha)}$ the number of corners of $\alpha$. Denote by ${\alpha^-}$ the set of partitions that result from removing a corner of $\alpha$. Similarly, ${\alpha^+}$ will denote the set of those partitions $\beta$ such that $\alpha \in {\beta^-}$. We use ${{\alpha^{\mp}}}$ to denote the set of partitions not equal to $\alpha$ that can be obtained by removing a corner of $\alpha$ and then adding a box to the result. Equivalently, ${{\alpha^{\mp}}}$ is the set of partitions that can be obtained from $\alpha$ by first adding a box and then removing a different box. For example, $31$ has two corners, and ${{(31)^{\mp}}} = \{(4), (22), (211)\}$. We finish Section \[uniqueness\] by relating the combinatorial identity $$\label{equ:straightcorners}
{\overline{K}_{(1)}}s_{\alpha} = ({\#\mathrm{corners}(\alpha)} - 1) s_\alpha + \sum_{\beta \in {{\alpha^{\mp}}}} s_\beta\,.$$ to the decomposition of the operators ${\overline{K}_{f}}$ in terms of operator $U_g$ and $D_g$.
We aim to generalize to skew Schur functions. This leads to our next use of the identities of Section 2. Corollary \[commutators\] implies the relation $[{D_{\theta}}, {\overline{K}_{(1)}}]={D_{s_{\theta/(1)} s_1}}$, which gives $${\overline{K}_{(1)}} {D_{\theta}} (s_\alpha) = {D_{\theta}} {\overline{K}_{(1)}} (s_\alpha) - {D_{s_{\theta/(1)} s_1}} (s_\alpha).$$ Applying and the fact that $$s_{\theta/(1)} s_1 = {\#\mathrm{corners}(\theta)}s_\theta + \sum_{\phi \in {{\theta^{\mp}}}} s_\phi,$$ we get $$s_{\alpha/\theta} * s_{(n-|\theta|-1,1)} = ({\#\mathrm{corners}(\alpha)} - {\#\mathrm{corners}(\theta)} - 1)s_{\alpha/\theta} + \sum_{\beta \in {{\alpha^{\mp}}}} s_{\beta/\theta} - \sum_{\phi \in {{\theta^{\mp}}}} s_{\alpha/\phi}\ ,$$
Thus we have an algebraic proof of the following.
\[thm:skewcorners\] Suppose $\alpha \vdash n$ and $\theta \vdash k$ with $\theta \subseteq \alpha$. Then $$s_{\alpha/\theta} * s_{(n-k-1,1)} = ({\#\mathrm{corners}(\alpha)} - {\#\mathrm{corners}(\theta)} - 1)s_{\alpha/\theta} + \sum_{\beta \in {{\alpha^{\mp}}}} s_{\beta/\theta} - \sum_{\phi \in {{\theta^{\mp}}}} s_{\alpha/\phi}\, .$$
This results begs for a a combinatorial proof. We offer a proof which is “two-thirds” combinatorial. The part which is non-combinatorial makes use of , which is turn is proved using Littlewood’s Identity .
The proof will work in three stages. In the short first stage, which is the non-combinatorial one, we will apply to express $s_{\alpha/\theta} * s_{(n-k-1,1)}$ in a form not involving any Kronecker products. Then, using the skew Pieri rule , we will reduce the problem to showing an identity that is effectively purely about SSYT. This identity will be proved in the third stage using jeu de taquin.
First, we will need to determine the result of applying ${D_{(1)}}$ to the skew Schur function $s_{\alpha/\theta}$. We have $${\left\langle\, s_\beta \, \middle| \, {D_{(1)}} s_{\alpha/\theta} \, \right\rangle}
= {\left\langle\, s_\beta s_{(1)} \, \middle| \, s_{\alpha/\theta} \, \right\rangle}
= \sum_{\delta\in{\theta^+}} {\left\langle\, s_\beta s_\delta \, \middle| \, s_\alpha \, \right\rangle}
= {\left\langle\, s_\beta \, \middle| \, \sum_{\delta\in{\theta^+}} s_{\alpha/\delta} \, \right\rangle},$$ and so ${D_{(1)}} s_{\alpha/\theta} = \sum_{\delta\in{\theta^+}} s_{\alpha/\delta}$. Applying to $s_{\alpha/\theta}$, we immediately deduce $$\label{equ:nokronecker}
s_{\alpha/\theta} * s_{(n-k-1,1)} = s_{(1)} \sum_{\delta \in {\theta^+}} s_{\alpha/\delta} - s_{\alpha/\theta}\,.$$
We next wish to apply the skew Pieri rule to $s_{(1)} \sum_{\delta \in {\theta^+}} s_{\alpha/\delta}$, but it will prove worthwhile to perform a preliminary step. By definition, $s_{\alpha/\delta} = 0$ unless $\delta \subseteq \alpha$, so it suffices to sum over those $\delta$ such that $\delta \in {\theta^+}$ and $\delta \subseteq \alpha$. We will denote that $\delta$ satisfies both conditions by writing $\delta \in{\theta^{+\alpha}}$. So we now apply the skew Pieri rule to $$s_{\alpha/\theta} * s_{(n-k-1,1)} = s_{(1)} \sum_{\delta \in {\theta^{+\alpha}}} s_{\alpha/\delta} - s_{\alpha/\theta}$$ to yield $$s_{\alpha/\theta} * s_{(n-k-1,1)}
= \sum_{\genfrac{}{}{0pt}{}{\gamma \in {\alpha^+}}{\delta \in {\theta^{+\alpha}}}} s_{\gamma/\delta}
- \sum_{\genfrac{}{}{0pt}{}{\delta \in {\theta^{+\alpha}}}{\phi\in {\delta^-}}} s_{\alpha/\phi}
- s_{\alpha/\theta}\ .$$
Let us examine the second sum. For any $\delta$ in ${\theta^{+\alpha}}$, we can choose $\phi = \theta$. We see that the other $\phi$ that arise will be exactly those elements of ${{\theta^{\mp}}}$ that are contained in $\alpha$. Therefore, $$s_{\alpha/\theta} * s_{(n-k-1,1)} = \sum_{\genfrac{}{}{0pt}{}{\gamma \in {\alpha^+}}{\delta \in {\theta^{+\alpha}}}} s_{\gamma/\delta} - |{\theta^{+\alpha}}| s_{\alpha/\theta} - \sum_{\phi \in {{\theta^{\mp}}}} s_{\alpha/\phi} - s_{\alpha/\theta}\ .$$ Thus to prove Theorem \[thm:skewcorners\], it remains to show that $$\label{equ:tableauxmanipulation}
\sum_{\genfrac{}{}{0pt}{}{\gamma \in {\alpha^+}}{\delta \in {\theta^{+\alpha}}}} s_{\gamma/\delta} - |{\theta^{+\alpha}}| s_{\alpha/\theta}
= ({\#\mathrm{corners}(\alpha)} - {\#\mathrm{corners}(\theta)})s_{\alpha/\theta} + \sum_{\beta \in {{\alpha^{\mp}}}} s_{\beta/\theta}\ .$$
Our main tool for proving the above identity will be jeu de taquin but, like with our application of the skew Pieri rule, it will be worthwhile to rewrite in a slightly different form. Observe that for any partition $\alpha$, we have ${\#\mathrm{corners}(\alpha)} = |{\alpha^+}|-1$. For $\theta \subseteq \alpha$, denote those elements of ${\theta^+}$ that are not contained in $\alpha$ by ${\theta^{+\alpha^c}}$, which we can check can only be non-empty if $\alpha/\theta$ has some empty rows or columns. We can now rewrite as $$\label{equ:tableauxmanipulation2}
\sum_{\genfrac{}{}{0pt}{}{\gamma \in {\alpha^+}}{\delta \in {\theta^{+\alpha}}}} s_{\gamma/\delta}
= (|{\alpha^+}| - |{\theta^{+\alpha^c}}|)s_{\alpha/\theta} + \sum_{\beta \in {{\alpha^{\mp}}}} s_{\beta/\theta}\ .$$ For intuition, we can call the positions of the form $\lambda/\alpha$ for some $\lambda \in {\alpha^+}$ the *outside corners* of $\alpha$. Then the term $|{\alpha^+}| - |{\theta^{+\alpha^c}}|$ is the number of outside corners of $\alpha$, excluding those that are also outside corners of $\theta$. See Example \[exa:jdt\] below for a fully worked example of the remainder of the proof.
To prove using jeu de taquin (jdt), consider an SSYT $T$ that contributes to the left-hand side, meaning $T$ has shape $\gamma/\delta$, where $\gamma \in {\alpha^+}$ and $\delta \in {\theta^{+\alpha}}$. Notice that the unique box $b$ of $\delta/\theta$ is not an element of $T$, and that the unique box $c$ of $\gamma/\alpha$ is in $T$. Perform a jdt slide of $T$ into $b$, and let $T'$ denote the resulting SSYT. There are three possibilities that can arise.
1. $T'=T$, meaning there is no way to fill $b$ under a jdt slide of $T$.
2. $T'$ contains $b$, and the single vacated box under the jdt slide is not $c$.
3. $T'$ contains $b$, and the single vacated box under the jdt slide is $c$.
By definition of jdt, Case (a) can happen if and only if $b$ is a corner of $\gamma$. Since $b \in \delta \subseteq \alpha \subset \gamma$, it must also be the case that $b$ is a corner of $\alpha$. Therefore, since $b$ is the unique box of $\delta/\theta$ and is not an element of $T'$ while $c$ is the unique box of $\gamma/\alpha$ and is in $T'$, the shape $\gamma/\delta$ of $T'$ can be written in the equivalent form $\beta/\theta$, where $\beta$ is obtained from $\alpha$ by removing $b$ and adding $c$. In particular, $\beta \in {{\alpha^{\mp}}}$, and so such $T'$ contribute part of the sum on the right of .
We claim that Case (b) contributes the rest of the sum on the right of . We see that the shape of $T'$ is obtained from the shape of $T$ by making exactly two changes: $T'$ contains $b$, and a box $c'$ different from $c$ has been vacated. As a result, $T'$ has shape $\beta/\theta$ for some $\beta \in {{\alpha^{\mp}}}$.
To prove our claim from the start of the previous paragraph, let $T'$ be an SSYT of shape $\beta/\theta$, where $\beta \in {{\alpha^{\mp}}}$. We wish to show that $T'$ arises as the image under jdt of exactly one $T$ from Cases (a) and (b). In short, the reason is that jdt slides are reversible, but let us be more precise. Suppose $\beta$ is obtained from $\alpha$ by removing a box $d$ and adding a different box $e$. Perform a jdt slide of $T'$ into $d$. (Some references would call this a *reverse* jdt slide, since $d$ is outside $\beta$.) There are two possibilities that can arise.
1. There is no way to fill $d$ under a jdt slide of $T'$. This can happen if and only if $d$ is an outside corner of $\theta$. Thus $\beta/\theta$ can be written in the equivalent form $\gamma/\delta$, where $\gamma$ equals $\alpha$ with $e$ added, and $\delta$ equals $\theta$ with $d$ added. Since $d$ is a corner of $\alpha$, we also have that $d$ is a corner of $\gamma$. These are exactly the conditions for $T'$ to arise as an image under jdt of a $T$ from Case (a) (where in fact $T=T'$ and our $d$ here corresponds to $b$ in Case (a)).
2. Performing a jdt slide of $T'$ into $d$ fills $d$ and vacates an outside corner of $\theta$. This is exactly the reverse of the jdt slide from Case (b) (where $d$ is playing the role of $c'$).
Thus $T'$ arises as the image of a single $T$ from Cases (a) and (b).
It remains to consider Case (c). Note that all $T'$ in Case (c) are of shape $\alpha/\theta$. We would like to show that each $T'$ is the image under jdt of $k$ distinct $T$, where $k=(|{\alpha^+}| - |{\theta^{+\alpha^c}}|)$. This would show that the $T'$ from Case (c) together contribute the term $(|{\alpha^+}| - |{\theta^{+\alpha^c}}|)s_{\alpha/\theta}$ from the right-hand side of , and would be proved.
So pick a $T'$ from Case (c). Pick an outside corner $c$ of $\alpha$, and perform a (reverse) jdt slide of $T'$ into $c$. If $c$ is not an outside corner of $\theta$, then this jdt slide will fill $c$, and the result will be an SSYT $T$ of shape $\gamma/\delta$ with $\gamma \in {\alpha^+}$ and $\delta \in {\theta^{+\alpha}}$. These are exactly the conditions for $T'$ to arise in Case (c).
On the other hand, if $c$ is an outside corner of $\alpha$ and also of $\theta$, then $T'$ will remain fixed under the (reverse) jdt slide. Thus any $T$ that maps to such a $T'$ under a jdt slide must also have shape $\alpha/\theta$. Such a $T$ from the left-hand side of does not exist, since $T$ would contain the single box $\gamma/\alpha$ whereas $T'$ does not. We conclude that each $T'$ in Case (c) is the image of exactly $k$ distinct $T$, as required.
\[exa:jdt\] Suppose $\alpha=(4,1,1)$ and $\theta=(2,1)$. $$\begin{tikzpicture}[scale=0.4]
\tikzstyle{every node}=[inner sep=3pt];
\draw (-1.5,1.5) node {$\alpha/\theta=$};
\begin{scope}[xshift=2mm]
\draw[thick] (0,2) rectangle (1,3);
\draw[thick] (2,0) rectangle (3,1);
\draw[thick] (3,0) rectangle (4,1);
\draw[dashed] (0,0) rectangle (1,1);
\draw[dashed] (0,1) rectangle (1,2);
\draw[dashed] (1,0) rectangle (2,1);
\end{scope}
\end{tikzpicture}$$ The complete set of shapes $\gamma/\delta$, and SSYT $T$ and $T'$ from the proof of are show in Table \[tab:jdt\]. Deliberately omitted from the table is the scenario from the last paragraph of the proof, where $c$ is an outside corner of both $\alpha$ and $\theta$, which does not contribute to either side of . In this example, there is $1=|{\theta^{+\alpha^c}}|$ such situation, shown below. $$\begin{tikzpicture}[scale=0.4]
\tikzstyle{every node}=[inner sep=3pt];
\draw (-1.5,1.5) node {$\alpha/\theta=$};
\begin{scope}[xshift=2mm]
\draw[thick] (0,2) rectangle (1,3);
\draw[thick] (2,0) rectangle (3,1);
\draw[thick] (3,0) rectangle (4,1);
\draw[dashed] (0,0) rectangle (1,1);
\draw[dashed] (0,1) rectangle (1,2);
\draw[dashed] (1,0) rectangle (2,1);
\draw (1.5,1.5) node {$c$};
\end{scope}
\end{tikzpicture}$$
[c||c|c||c|c||c|c||]{} & & &\
(0,1) node [$\gamma/\delta$]{}; (0,0) node ; (0,4) node ;
&
(1,1) rectangle (2,2); (2,0) rectangle (3,1); (3,0) rectangle (4,1); (0,0) rectangle (1,1); (0,1) rectangle (1,2); (1,0) rectangle (2,1); (0,2) rectangle (1,3); (0.5,2.5) node [$b$]{}; (1.5,1.5) node [$c$]{};
&
(4,0) rectangle (5,1); (2,0) rectangle (3,1); (3,0) rectangle (4,1); (0,0) rectangle (1,1); (0,1) rectangle (1,2); (1,0) rectangle (2,1); (0,2) rectangle (1,3); (0.5,2.5) node [$b$]{}; (4.5,0.5) node [$c$]{};
&
(0,3) rectangle (1,4); (0,2) rectangle (1,3); (3,0) rectangle (4,1); (0,0) rectangle (1,1); (0,1) rectangle (1,2); (1,0) rectangle (2,1); (2,0) rectangle (3,1); (2.5,0.5) node [$b$]{}; (0.5,3.5) node [$c$]{}; (3.5,0.5) node [$c'$]{};
&
(3,0) rectangle (4,1); (0,2) rectangle (1,3); (1,1) rectangle (2,2); (0,0) rectangle (1,1); (0,1) rectangle (1,2); (1,0) rectangle (2,1); (2,0) rectangle (3,1); (2.5,0.5) node [$b$]{}; (1.5,1.5) node [$c$]{}; (3.5,0.5) node [$c'$]{};
&
(4,0) rectangle (5,1); (0,2) rectangle (1,3); (3,0) rectangle (4,1); (0,0) rectangle (1,1); (0,1) rectangle (1,2); (1,0) rectangle (2,1); (2,0) rectangle (3,1); (2.5,0.5) node [$b$]{}; (4.5,0.5) node [$c$]{};
&
(0,3) rectangle (1,4); (2,0) rectangle (3,1); (3,0) rectangle (4,1); (0,0) rectangle (1,1); (0,1) rectangle (1,2); (1,0) rectangle (2,1); (0,2) rectangle (1,3); (0.5,2.5) node [$b$]{}; (0.5,3.5) node [$c$]{};
\
\[2ex\]
(0,1) node [$T$]{}; (0,0) node ;
&
(1,1) rectangle (2,2); (2,0) rectangle (3,1); (3,0) rectangle (4,1); (0,0) rectangle (1,1); (0,1) rectangle (1,2); (1,0) rectangle (2,1); (0,2) rectangle (1,3); (0.5,2.5) node [$b$]{}; (1.5,1.5) node [$A$]{}; (2.5,0.5) node [$B$]{}; (3.5,0.5) node [$C$]{};
&
(4,0) rectangle (5,1); (2,0) rectangle (3,1); (3,0) rectangle (4,1); (0,0) rectangle (1,1); (0,1) rectangle (1,2); (1,0) rectangle (2,1); (0,2) rectangle (1,3); (0.5,2.5) node [$b$]{}; (2.5,0.5) node [$A$]{}; (3.5,0.5) node [$B$]{}; (4.5,0.5) node [$C$]{};
&
(0,3) rectangle (1,4); (0,2) rectangle (1,3); (3,0) rectangle (4,1); (0,0) rectangle (1,1); (0,1) rectangle (1,2); (1,0) rectangle (2,1); (2,0) rectangle (3,1); (2.5,0.5) node [$b$]{}; (0.5,2.5) node [$A$]{}; (0.5,3.5) node [$B$]{}; (3.5,0.5) node [$C$]{};
&
(3,0) rectangle (4,1); (0,2) rectangle (1,3); (1,1) rectangle (2,2); (0,0) rectangle (1,1); (0,1) rectangle (1,2); (1,0) rectangle (2,1); (2,0) rectangle (3,1); (2.5,0.5) node [$b$]{}; (0.5,2.5) node [$A$]{}; (1.5,1.5) node [$B$]{}; (3.5,0.5) node [$C$]{};
&
(4,0) rectangle (5,1); (0,2) rectangle (1,3); (3,0) rectangle (4,1); (0,0) rectangle (1,1); (0,1) rectangle (1,2); (1,0) rectangle (2,1); (2,0) rectangle (3,1); (2.5,0.5) node [$b$]{}; (0.5,2.5) node [$A$]{}; (3.5,0.5) node [$B$]{}; (4.5,0.5) node [$C$]{};
&
(0,3) rectangle (1,4); (2,0) rectangle (3,1); (3,0) rectangle (4,1); (0,0) rectangle (1,1); (0,1) rectangle (1,2); (1,0) rectangle (2,1); (0,2) rectangle (1,3); (0.5,2.5) node [$b$]{}; (0.5,3.5) node [$A$]{}; (2.5,0.5) node [$B$]{}; (3.5,0.5) node [$C$]{};
\
\[2ex\]
(0,1) node [$T'$]{}; (0,0) node ;
&
(1,1) rectangle (2,2); (2,0) rectangle (3,1); (3,0) rectangle (4,1); (0,0) rectangle (1,1); (0,1) rectangle (1,2); (1,0) rectangle (2,1); (0.5,2.5) node [$d$]{}; (1.5,1.5) node [$A$]{}; (2.5,0.5) node [$B$]{}; (3.5,0.5) node [$C$]{};
&
(4,0) rectangle (5,1); (2,0) rectangle (3,1); (3,0) rectangle (4,1); (0,0) rectangle (1,1); (0,1) rectangle (1,2); (1,0) rectangle (2,1); (0.5,2.5) node [$d$]{}; (2.5,0.5) node [$A$]{}; (3.5,0.5) node [$B$]{}; (4.5,0.5) node [$C$]{};
&
(0,3) rectangle (1,4); (0,2) rectangle (1,3); (0,0) rectangle (1,1); (0,1) rectangle (1,2); (1,0) rectangle (2,1); (2,0) rectangle (3,1); (2.5,0.5) node [$C$]{}; (0.5,2.5) node [$A$]{}; (0.5,3.5) node [$B$]{}; (3.5,0.5) node [$d$]{};
&
(0,2) rectangle (1,3); (1,1) rectangle (2,2); (0,0) rectangle (1,1); (0,1) rectangle (1,2); (1,0) rectangle (2,1); (2,0) rectangle (3,1); (2.5,0.5) node [$C$]{}; (0.5,2.5) node [$A$]{}; (1.5,1.5) node [$B$]{}; (3.5,0.5) node [$d$]{};
&
(3,0) rectangle (4,1); (0,2) rectangle (1,3); (0,0) rectangle (1,1); (0,1) rectangle (1,2); (1,0) rectangle (2,1); (2,0) rectangle (3,1); (2.5,0.5) node [$B$]{}; (0.5,2.5) node [$A$]{}; (3.5,0.5) node [$C$]{}; (4.5,0.5) node [$c$]{};
&
(2,0) rectangle (3,1); (3,0) rectangle (4,1); (0,0) rectangle (1,1); (0,1) rectangle (1,2); (1,0) rectangle (2,1); (0,2) rectangle (1,3); (0.5,2.5) node [$A$]{}; (0.5,3.5) node [$c$]{}; (2.5,0.5) node [$B$]{}; (3.5,0.5) node [$C$]{};
\
Acknowledgments {#acknowledgments .unnumbered}
===============
We thank Aaron Lauve for helpful comments. R. Orellana is grateful for the hospitality of the University of Sevilla and IMUS. E. Briand, P. McNamara and M. Rosas are grateful for the hospitality of the University of Rennes 1 and IRMAR in summer 2014.
E. Briand and M. Rosas have been partially supported by projects MTM2010–19336, MTM2013-40455-P, FQM–333, P12–FQM–2696 and FEDER. P. McNamara was partially supported by a grant from the Simons Foundation (\#245597). R. Orellana was partially supported by NSF Grant DMS-130512.
| {
"pile_set_name": "ArXiv"
} |
---
address: |
Instituto de Física, Universidad Nacional Autónoma de México,\
Apartado Postal 20-364, México D.F. 01000, México
author:
- 'Cecilia Noguez and C. E. Román-Velázquez'
title: 'Casimir force between a sphere and a plane: spectral representation formalism'
---
Introduction
============
Recent experiments$^{1-4}$ to measure the Casimir force have been done choosing a spherical surface and a plate, instead of the plate-plate configuration originally proposed by Casimir[@casimir]. The interpretation of the Casimir force in these precision experiments has relied in the Proximity Theorem, developed by Derjaguin and collaborators[@proximidad]. The Proximity Theorem estimates the Casimir force per unit area between two curved surfaces of radii $R_1$ and $R_2$, in terms of the Casimir energy per unit area between parallel planes[@raul], ${\cal V}(z)$, where $z$ is the distance between plates. Assuming that the Casimir force on a small area of one curved surface is due to locally “flat” portions on the other curved surface, Derjaguin found that, $$F_{\rm PT}(z) = 2 \pi \left(\frac{R_1R_2}{R_1 + R_2}\right) {\cal V}(z).$$ In the limit, when $R_1 =R$, and $R_2 \to \infty$, the problem reduces to the case of a sphere of radius $R$ and a flat plane, that yields to $F_{\rm PT}(z) = 2 \pi R {\cal V}(z)$. The force obtained is a power-law function of the separation, $F_{\rm PT}(z) \propto z^{-\beta}$, where at “large” distances $\beta = 3$ while at short distances $\beta = 2$. This theorem is assumed to hold only when $z \ll R_1, R_2$, however, it is not clear up to what limit this approach is valid.
The origin of dispersive forces between atoms and macroscopic bodies may be attributed to interactions between the induced charge distributions on them by quantum vacuum fluctuations[@casimir]. The charge distribution can be represented, in the simplest approximation, by electric dipoles. This dipole approximation was employed by London [@london] to calculate the van der Waals energy $V_{\rm vW} (z)$ between two identical polarizable molecules by using perturbation theory in quantum mechanics. Considering that the molecules interact through a quasi-static potential (or non-retarded limit), London found that $V_{\rm vW}(z) \propto - 1/z^\beta$, where $\beta = 6$ and $z$ is the magnitude of the distance between molecules. In 1948, Casimir and Polder [@casypol] studied the interaction of a neutral atom with a perfectly conducting plane, such that, the force is proportional to $ -1/z^\beta$ with $\beta = 4$ in the non-retarded limit [@casypol]. Furthermore, they found that the vW interaction could be attributed to the change of the electromagnetic zero-point energy of the classical proper electromagnetic modes of the system. The same behavior can be also obtained calculating the change of the zero-point energy between a induced dipole moment on a sphere of polarizability $\alpha$ with its own image dipole in a plane. Contrary to what one may expect, the result for perfect conductors differs from the one obtained using the Proximity Theorem by a factor of $1/z^2$. Therefore, an exact calculation of the Casimir force between a sphere and a planar surface becomes essential.
In this paper, we study the Casimir force in the non-retarded or quasi-static limit between a sphere and a plane, both with arbitrary dielectric properties. We calculate the difference of the zero-point energy when the two bodies are at a distance $z$, and when they are at infinite, as $${\cal E}(z) = \frac{\hbar}{2}\sum_{s} \left[\omega^{s}(z) - \omega_\infty^s \right], \label{ener}$$ where $\omega^s(z)$ and $\omega_\infty^s$, are the proper electromagnetic modes at $z$ and $z \to \infty$, respectively. By including multipolar interactions to all orders, we calculate exactly the proper electromagnetic modes of the system using a Spectral Representation formalism[@ceci2]. We show that this spectral representation formalism has the advantage that separates the contribution of the dielectric properties of the sphere from the contribution of its geometrical properties, allowing to perform a systematic study of the system as a function of the radius of the sphere and its dielectric properties.
Dipole approximation within the non-retarded limit
==================================================
{width="3in"}
Let us consider a sphere of radius $a$ and dielectric function $\epsilon_s(\omega)$ which is at a minimum distance $z$ from a semi-infinite plate of dielectric function $\epsilon_p(\omega)$, as shown in Fig. 1. The sphere and plate are electrically neutral. The quantum vacuum fluctuations will induce a charge distribution on the sphere which also induces a charge distribution in the plate, such that, the $lm$-th multipolar moment on the sphere is given by $$Q_{lm} = \alpha_{lm} \left[ V_{lm}^{\rm vac} + V_{lm}^{\rm sub} \right], \label{qlm}$$ where $V_{lm}^{\rm vac}$ is the field associated to the quantum vacuum fluctuations at the zero-point energy, $V_{lm}^{\rm sub}$ is the induced field due to the presence of the plate, and $\alpha_{lm}$ is the $lm$-th polarizability of the sphere.
In the dipolar approximation, one can find using the image method, that the charge distribution on the plate is described by a dipole moment ${\vec p}_{\rm p}(\omega)$ which is related with the induced dipole moment on the sphere by $${\vec p}_{\rm p}(\omega) = \frac{1 - \epsilon_p(\omega)} { 1 + \epsilon_p (\omega)} {\mathbb M} \cdot {\vec p}_{\rm s}(\omega), \label{psub}$$ where ${\mathbb M}= (1, 1, -1)$ is a diagonal matrix, and ${\vec p}_{\rm s}$ is the dipole moment on the sphere which, according with Eq. (\[qlm\]), is given by $${\vec p}_{\rm s}(\omega) = \alpha(\omega) \left[ {\vec E}^{\rm vac} + {\mathbb T} \cdot {\vec p}_{\rm p}(\omega) \right]. \label{psph}$$ Here, $\mathbb T= (3 {\vec r} {\vec r} - r^2{\mathbb I})/r^5$ is a tensor that couples the interaction between the induced dipoles in the sphere and plate, ${\mathbb I}$ is the identity matrix, and ${\vec r}=(0,0,2(z+a))$ is the vector from the center of the sphere to the center of the image charge on the plane, such that, ${\mathbb M}\cdot {\mathbb T}= (-1/r^3, -1/r^3, -2/r^3)$. Substituting Eq. (\[psub\]) in Eq. (\[psph\]), we find $$\left[\frac{1}{\alpha} + f_c \frac{b}{r^3} \right] \mathbb I \cdot {\vec p}_{\rm s} = {\vec E}^{\rm vac}. \label{q2}$$ with $f_c = [ 1 - \epsilon_{\rm p}(\omega)]/[ 1 + \epsilon_{\rm p}(\omega)]$, and $b= 1$, $1$ or $2$. The eigenfrequencies of the sphere-substrate system are found from the above equation and must be independent of ${\vec E}^{\rm vac}$. The normal frequencies can be obtained when the determinant of the matrix in the left-side of Eq. (\[q2\]) is equal to zero. Then, there are three modes (two are degenerated), with frequencies given by the solution of $$\left[\frac{1}{{\alpha}(\omega)} +f_c \frac{1}{[2(z+a)]^{3}}\right]^2
\left[ \frac{1}{{\alpha}(\omega)} +f_c \frac{2}{[2(z+a)]^{3}}\right] = 0. \label{ceros}$$ One should notice that the left-side of each term of the above equation is related only with the dielectric properties of the sphere, while the right-side is related with its geometrical properties only.
Spectral Representation of the sphere-plane configuration {#formalism}
=========================================================
The polarizability of a sphere in the non-retarded limit is $\alpha/a^3 \equiv \tilde{\alpha}=[\epsilon_{\rm s} -1]/[\epsilon_{\rm s} + 2]$. Now, let us define the variable $u = [1 - \epsilon_{\rm s}]^{-1}$, then $\tilde{\alpha}(\omega)^{-1} = -3 u(\omega) + 1 $, and substituting in Eq. (\[q2\]), one obtains $$\left[ - u(\omega) {\mathbb I} + {\mathbb H} \right] \cdot {\bf p}_{\rm s}(\omega) = {\mathbf V}^{\rm vac}, \label{matriz}$$ with ${\mathbf V}^{\rm vac} = \frac{1}{3}a^3{\mathbf E}^{\rm vac}$, and ${\mathbb H} = \frac{1}{3} [{\mathbb I} - f_c a^3 {\mathbb T}\cdot{\mathbb M}]$. The later is a dimensionless matrix that only depends on the geometry of the system through the ratio $z/a$.
Let consider when $f_c$ is real[@fc], then ${\mathbb H}$ is real and symmetric, in such case, we can always find a unitary transformation ${\mathbb U}$ that diagonalizes it, ${\mathbb U}^{-1}{\mathbb H}{\mathbb U} = n_s$, being $n_s$ the eigenvalues of ${\mathbb H}$. Therefore, the proper modes depend on the separation $z$, and can be obtained from $$\det{[- u(\omega) {\mathbb I} + \mathbb H]} = \prod_s [-u(\omega) + n_s(z)] =0.$$ Alternatively, the eigenfrequencies can be found from the poles of the Green’s function of the system defined as $\mathbb G(u) = [ - u(\omega) {\mathbf 1} + {\mathbb H}]^{-1}$, whose $ij$-th element is given by $$G_{ij}(u) = \sum_s \frac{U_{is}(U_{js})^{-1}}{u-n_s(z)}. \label{green}$$ Once the proper modes are obtained, the interaction energy is calculated using Eq. (1), as the difference between the zero-point energy when the sphere is at a distance $z$ from the substrate and when it is at $z \to \infty$, then the Casimir force is calculated as $F=-d{\cal E}(z)/dz$.
One should notice that in the spectral representation formalism, the material properties of the sphere are contained in the spectral variable $u$, while the geometrical properties, like the radius of the sphere and the separation of the sphere to the substrate are in ${\mathbb H}$. Furthermore, ${\mathbb H}$ is dimensionless and depends on the ratio $z/R$, its eigenvalues are independent of ${\mathbf V}^{\rm vac}$. On the other hand, the dielectric properties of the substrate are in $f_c$ which can be a real function even for dispersive materials. Therefore, with the spectral representation one can study separately the contribution of the dielectric properties of the sphere from its geometrical properties, allowing for a systematic investigation of the system.
Exact calculation within the non-retarded limit
===============================================
The $lm$-th multipolar moment induced in the sphere on Eq. (2) can be written using a multipolar expansion[@claro], that yields to $$Q_{lm} = \frac{-(2l+1)}{4 \pi} \alpha_{lm} [ V_{lm}^{\rm vac} + \sum_{l', m'} (-1)^{m^{^{\prime }}+l^{^{\prime }}} A_{lm}^{l'm'} {\hat{Q}}_{l'm'} ], \label{q}$$ where $\hat{Q}_{l'm'}$ is the $l'm'$-th induced multipolar moment in the substrate located at ${\vec r}= (2z+2a,\pi,\varphi)$ from the center of the sphere, and $A_{lm}^{l'm'}$ is the matrix that couples the interaction between sphere and substrate[@claro]. The induced $l'm'$-th multipolar moment in the substrate is related with the multipolar charge distribution on the sphere like $\hat{Q}_{l'm'} = (-1)^{l'+m'} f_c Q_{l'm'}$. Substituting this in Eq. (\[q\]), one finds $$-\sum_{l'm'} \left[\frac{4 \pi \delta_{ll'} \delta_{mm'} }{(2l+1)\alpha_{l'm'}} + f_c A_{lm}^{l'm'} \right] {Q}_{l'm'} = V_{lm}^{\rm vac}. \label{q3}$$ If the sphere is homogeneous its polarizabilities are independent of $m$, and are given by[@claro] $$\alpha_l (\omega) = \frac{l[\epsilon_{\rm s}(\omega) - 1]} {l [ \epsilon_{\rm s}(\omega) + 1] + 1} a^{2l+1} = \frac{n_{l0}}{n_{l0} - u(\omega)} a^{2l+1}, \label{alfa}$$ where $n_{l0} = l/(2l+1)$. The poles of the right-side of Eq. (\[alfa\]), $u(\omega_{l0}) = n_{l0}$, yield the frequencies of the proper modes of the isolated sphere. Again, we have separated the material and geometrical properties of the sphere in Eq. (\[alfa\]).
Using Eq. (\[alfa\]), we can rewrite Eq. (\[q2\]) as $$\sum_{\mu'} [ - u(\omega) \delta_{\mu \mu'} + {H}^{\mu '}_{\mu} ] {x}_{\mu'} = b_{\mu},$$ -.2in $${\rm where} \quad \mu \equiv (l,m), \quad x_{\mu} = \frac{Q_{lm}} { (l a^{2l+1})^{1/2}} , \quad b_{\mu} = - \frac{(l a^{2l+1})^{1/2}} {4 \pi} V_{lm}^{\rm vac}, \quad {\rm with}$$ $$H_{\mu}^{\mu'} = n_{l'0} \delta_{\mu \mu'} + f_c \frac{(a^{l+l'+1})} {4 \pi} (ll')^{1/2} A_{\mu}^{\mu'}. \label{h}$$ It was shown in Ref. 12 that, $A_{\mu}^{\mu'}$ is a symmetric matrix that depends on the distance between the center of the “image-charge” distribution and the center of the sphere as, $1/[2(z+a)]^{l+l'+1}$. Then, $\mathbb H$ is dimensionless, symmetric, and depends only on the geometry of the system. As we showed, it is possible to find the Green’s function, $\mathbb G$, of the system, in terms of the unitary matrix that diagonalizes $\mathbb H$. The poles of $\mathbb G$, yield the proper frequencies of the system.
Results and discussion
======================
To illustrate the procedure, we use the Drude model for the dielectric function of the sphere, $\epsilon_s(\omega) = 1 - \omega_p^2/[\omega(\omega + i/\tau)]$, with $\omega_p $ the plasma frequency and $\tau$ the relaxation time. In this case, $u(\omega) \omega_p^2 = \omega(\omega + i/\tau )$, such that, the proper modes are given by $$\omega_s(z) = -i/2\tau + \sqrt{(i/2\tau)^2 + \omega_p^2 n_s(z)} \approx \omega_p \sqrt{n_s(z)},$$ with $s = (l,m)$. In the left-side we have considered that $ 1 \gg (\tau\omega_p)^{-1}$. In this case, the fluctuations of the zero-point energy, according with Eq. (1), are $${\cal E}(z) = \frac{\hbar \omega_p}{2} \sum_{l,m}\left[ \sqrt{n_{lm}} - \sqrt{n_{l0}} \right].$$ One should notice that it is possible to find the behavior of the energy as a function of $z/a$ for any Drude’s sphere independently of its plasma frequency, like $\tilde{{\cal E}} \equiv {\cal E}/\hbar \omega_p $. Furthermore, the Casimir force can be also studied independently of the plasma frequency and radius of the sphere, since ${\cal E}(z/a)$ one can find the force like $$F=-\frac{d{\cal E}(z/a)}{dz} = -\frac{\partial{\cal E}(z/a)}{\partial(z/a)} \frac{\partial(z/a)}{\partial z},$$ and then define the dimensionless variable $\tilde{F}a \equiv F a/ \hbar \omega_p$. In summary, we have shown, for a Drude sphere with $ 1 \gg (\tau\omega_p)^{-1}$, that given the dielectric properties of the substrate the behavior of the energy and force is quite general, showing the potentiality of the Spectral Representation formalism.
{width="4.8in"}
In Fig. 2, we show the energy $\tilde{{\cal E}}$ and force $\tilde{F}a$ for a Drude sphere, of arbitrary plasma frequency and radius over sapphire[@sapphire], as a function of $z/a$. We calculate the energy and force when multipolar interactions up to $l$ and $l' = 2000$ are taken into account, as well as up to dipolar, and up to quadrupolar interactions are considered. When we consider multipolar interactions greater than dipolar ones ($l, l' > 1$), it is observed that $\tilde{{\cal E}}$ and $\tilde{F}a$ do not follow a simple power-law as a function of the separation. We find that $\tilde{{\cal E}}$ and $\tilde{F}a$ are proportional to $1/(z/a)^{\beta}$, where $\beta (z) $ is a positive integer which is also a function of the separation. As the separation decreases ($z \to 0$) the power increases ($\beta \to \infty$). Then, the force can suddenly increases more than four orders of magnitude as compare with the dipolar approximation, when the sphere approaches the substrate. While at large distances ($ 7a > z > 2a$) the force can be obtained exactly if up to quadrupolar interactions are considered. And for $z > 7a$, the interaction between the sphere and the substrate can be modeled using the dipolar approximation only, like in the Casimir and Polder model. Finally, the increment of the force at small separations could explain the physical origin of the large deviations observed in the deflection of atomic beams by metallic surfaces, as well as some instabilities detected in micro and nano devices. However, specific experiments and calculations have to be performed to prove the latter.
Acknowledgments {#acknowledgments .unnumbered}
===============
This work has been partly financed by CONACyT grant No. 36651-E and by DGAPA-UNAM grant No. IN104201.
[99]{} S. K. Lamoreaux, .
U. Mohideen and A. Roy, .
H. B. Chan, [*et al.*]{}, .
R.S. Decca, [*et al.*]{},
H. B. G. Casimir, .
B.V. Derjaguin, I.I. Abrikosova, .
The energy per unit area between parallel plates can be calculated for real materials using the Lifshitz formula \[see, E. M. Lifshitz, Sov. Phys. JETP [**2**]{}, 73 (1956)\] and some of its extensions. See for example: R. Esquivel-Sirvent and C. Villarreal and G. H. Cocoletzi, ; W.L. Mochán, C. Villarreal, R. Esquivel-Sirvent, ; R. Esquivel-Sirvent, C. Villarreal, W. L. Mochán, G.H. Cocoletzi, ; C. Villarreal, R. Esquivel-Sirvent, G. H. Cocoletzi, ; R. Esquivel-Sirvent, C. Villarreal, ; R. Esquivel-Sirvent, C. Villarreal, W. L, Mochán, .
F. London, .
H. B. G. Casimir, D. Polder, .
C.E. Román-Velázquez, C. Noguez, C. Villarreal, and R. Esquivel-Sirvent, quant-ph/0303172; C. Noguez, C.E. Román-Velázquez, R. Esquivel-Sirvent, and C. Villarreal, quant-ph/0310068.\[ceci2\]
$a$ and $z$ must be smaller than the characteristic length $l_c$ of the system. For metals, $l_c=c/\omega_p$, where $\omega_p$ is the plasma frequency.
F. Claro [*Phys. Rev.*]{} B [**30**]{}, 4989 (1984).
For purely imaginary frequencies $\xi = i \omega$, it is known that $\epsilon(\xi)$ is real, therefore, $f_c$ is also real.
The dielectric function of sapphire is real and constant in a wide range of the electromagnetic spectrum, with $\epsilon_s = 3.13$ and $f_c= -0.516$.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We prove that if $G$ is a finite simple group of Lie type and $S$ a subset of $G$ of size at least two then $G$ is a product of at most $c\log|G|/\log|S|$ conjugates of $S$, where $c$ depends only on the Lie rank of $G$. This confirms a conjecture of Liebeck, Nikolov and Shalev in the case of families of simple groups of bounded rank. We also obtain various related results about products of conjugates of a set within a group.'
address:
- 'László Pyber and Endre SzabóA. Rényi Institute of MathematicsHungarian Academy of SciencesP.O. Box 127H-1364 Budapest'
- 'Nick Gill and Ian ShortDepartment of Mathematics and StatisticsThe Open UniversityMilton Keynes, MK7 6AAUnited Kingdom'
author:
- 'N. Gill, L. Pyber, I. Short, E. Szabó'
bibliography:
- 'paper6.bib'
title: On the product decomposition conjecture for finite simple groups
---
[^1] [^2] [^3]
Introduction
============
Our starting point is the following conjecture of Liebeck, Nikolov and Shalev [@lns2].
\[c: 2\] There exists an absolute constant $c$ such that if $G$ is a finite simple group and $S$ is a subset of $G$ of size at least two, then $G$ is a product of $N$ conjugates of $S$ for some $N\leq c \log|G|/ \log |S|$.
Note that we must have $N\ge\log|G|\big/\log|S|$ by order considerations, and so the bound above is best possible up to the value of the constant $c$.
The conjecture is an extension of a deep (and widely applied) theorem of Liebeck and Shalev. Indeed, the main result of [@lieshal] states that the above conjecture holds when $S$ is a conjugacy class or, more generally, a normal subset (that is, a union of conjugacy classes) of $G$. In [@lns2] Conjecture \[c: 2\] is also proved for sets of bounded size.
Somewhat earlier Liebeck, Nikolov and Shalev [@lns] posed the following (still unproved) weaker conjecture.
\[c: 1\] There exists an absolute constant $c$ such that if $G$ is a finite simple group and $H$ is any nontrivial subgroup of $G$, then $G$ is a product of $N$ conjugates of $H$ for some $N\leq c \log|G|/ \log |H|$.
Conjecture \[c: 1\] itself represents a dramatic generalization of a host of earlier work on product decompositions of finite simple groups, most of which prove Conjecture \[c: 1\] for particular subgroups $H$. For instance, in [@liepy] it is proved that a finite simple group of Lie type in characteristic $p$ is a product of 25 Sylow $p$-subgroups (see also [@bnp] for a recent improvement from 25 to 5).
Further positive evidence for Conjecture \[c: 1\] is provided by [@lns3], [@lub] and [@nikolov] (when $H$ is of type $SL_n$). Certain results of this type are essential to prove that finite simple groups can be made into expanders (see the announcement [@kln]).
The main purpose of this note is to prove Conjecture \[c: 2\] for finite simple groups of Lie type of bounded rank. Put another way, we prove a version of Conjecture \[c: 2\] in which the constant $c$ depends on the rank of the group $G$. Our main result follows.
\[t: gps\] Fix a positive integer $r$. There exists a constant $c=c(r)$ such that if $G$ is a finite simple group of Lie type of rank $r$ and $S$ is a subset of $G$ of size at least two then $G$ is a product of $N$ conjugates of $S$ for some $N\leq c \log|G|/ \log |S|$.
In [@lns2] a weaker bound of the form $N\leq\big(\log|G|/ \log |S|\big)^{c(r)}$ is obtained. Also, in [@lns], Theorem \[t: gps\] is proved when $S$ is a maximal subgroup of $G$.
As a byproduct of our proof we obtain two results of independent interest. In these results, and throughout the paper, we denote by $S^g$ the conjugate $g^{-1}Sg$ of a subset $S$ of a group $G$ by an element $g$ of $G$, and, given a positive integer $m$, we denote by $S^m$ the product $SS\dotsb S$ of $m$ copies of $S$. There should be no confusion between these two similar notations because the type of the exponent will always be given.
\[t: 2\] Fix a positive integer $r$. There exists a positive constant $\varepsilon=\varepsilon(r)$ such that if $G$ is a finite simple group of Lie type of rank $r$ and $S$ is a subset of $G$ then for some $g$ in $G$ we have $\left|S S^g\right|\ge|S|^{1+\varepsilon}$ or $S^3=G$.
The next theorem is similar, but concerns only normal subsets, in which case we obtain absolute constants.
\[t: normal subsets\] There exists $\varepsilon>0$ and a positive integer $b$ such that if $G$ is a finite simple group and $S$ is a normal subset of $G$ then $\left|S^2\right|\ge|S|^{1+\varepsilon}$ or $S^b=G$.
Theorem \[t: normal subsets\] relates to a result of Shalev [@shalev Theorem 7.4], which we strengthen in Section \[s: shalev\].
Note that the theorem would not be true were we to consider sets that are not normal. For instance, take $S$ to be a maximal parabolic subgroup in $G=PSL_n(q)$ with index $\frac{q^n-1}{q-1}$. Clearly $S^b=S$ for all positive integers $b$; on the other hand, for any positive number $\varepsilon$, and any $g$ in $G$, we have $|S S^g|\leq |G|\leq |S|^{1+\varepsilon}$ once $n$ is large enough. We conclude that neither of the given options can hold in this more general situation.
Theorems \[t: 2\] and \[t: normal subsets\], and the remarks of the previous paragraph, lead us to make the following conjecture.
\[c: 3\] There exists $\varepsilon>0$ and a positive integer $b$ such that if $S$ is a subset of a finite simple group $G$ then for some $g$ in $G$ we have $\left|S S^g\right|\ge|S|^{1+\varepsilon}$ or $G$ is the product of $b$ conjugates of $S$.
Note that, by Theorems \[t: gps\] and \[t: 2\], Conjectures \[c: 2\], \[c: 1\] and \[c: 3\] hold for all exceptional simple groups. Note too that all three conjectures could be phrased in terms of *translates* of the set $S$, rather than conjugates. This follows from the simple fact that a product of translates of $S$ is equal to a translate of a product of conjugates of $S$. Similarly a product of conjugates of a translate of $S$ is equal to a translate of a product of conjugates of $S$, a fact which will be useful in its own right.
It is possible that Conjecture \[c: 3\] actually holds with $b=3$. When $b=2$ counterexamples are given by large non-real conjugacy classes (see the final section of [@shalev] for some related issues). Further counterexamples are given by certain families of maximal subgroups (see for example [@lps Corollary 2], which states that large enough simple unitary groups of odd dimension cannot be decomposed into the product of two proper subgroups).
We derive Theorems \[t: gps\] and \[t: 2\] as consequences of the recent Product theorem for finite simple groups, proved independently by Breuillard, Green and Tao [@bgt2], and Pyber and Szabó [@ps2] (see Section \[s: proof2\]). Theorem \[t: normal subsets\] follows from a version of Conjecture \[c: 2\] for normal subsets due to Liebeck and Shalev [@lieshal] and an extension of Plünnecke’s theorem [@taovu Theorem 6.27] to normal subsets of nonabelian groups (see Section \[s: doubling\]).
In the final section we use a result of Petridis [@petridis] to derive an analogue of the classical Doubling lemma, a special case of Plünnecke’s theorem. We refer to the new result as the Skew doubling lemma; it can be thought of as a nonabelian version of the classical Doubling lemma. The Skew doubling lemma is applied to prove that Conjecture \[c: 2\] implies Conjecture \[c: 3\]. In the other direction, a standard argument (similar to the proof of Corollary \[c: polynomial\]) shows that Conjecture \[c: 3\] implies that a simple group $G$ is a product of $\left(\log|G|/\log|S|\right)^c$ conjugates of $S$, a weaker version of Conjecture \[c: 2\].
Proof of Theorem \[t: 2\] {#s: proof2}
=========================
We begin with a result of Petridis [@petridis Theorem 4.4], which extends work of Helfgott, Ruzsa and Tao [@helfgott3; @Ru2009; @Ru2010; @tao]. It relates to the Doubling lemma for abelian groups, which we return to in Section \[s: doubling\].
\[l: petridis\] Let $S$ be a finite subset of a group $G$. Suppose that there exist positive numbers $J$ and $K$ such that $|S^2|\leq J|S|$ and $|SgS|\leq K|S|$ for each $g$ in $S$. Then $|S^3| \leq J^7K|S|$.
Suppose now that $G$ is a finite group, and let ${\mathrm{minclass}}(G)$ denote the size of the smallest nontrivial conjugacy class in $G$. Let ${\mathrm{minclass}}(S,G)$ denote the size of the smallest nontrivial conjugacy class in $G$ that intersects $S$, and let ${\mathrm{mindeg}}(G)$ denote the dimension of the smallest nontrivial complex irreducible representation of $G$.
As observed in [@npy], a result of Gowers [@gowers] implies the following.
\[p: gowers\] Let $G$ be a finite group and let $k={\mathrm{mindeg}}(G)$. Take $S\subseteq G$ such that $|S| \geq \frac{|G|}{\sqrt[3]k}.$ Then $G=S^3$.
Now let $G=G_r(q)$ be a simple group of Lie type of rank $r$ over ${\mathbb{F}_q}$, the finite field of order $q$. We need some facts about $G$. The first result can be deduced, for example, from [@kl Tables 5.1 and Theorem 5.2.2].
\[p: conjugacy size\] We have $q^r\leq {\mathrm{minclass}}(G)<|G|\leq q^{8r^2}$.
\[p: landseitz\] Let $k={\mathrm{mindeg}}(G)$. Then $|G|<k^{8r^2}$.
We use the lower bounds on projective representations given by Landazuri and Seitz [@landseitz], allowing for the slight errors corrected in [@kl Table 5.3.A]. For $G\neq PSL_2(q)$, we see that $k\geq q$, and so the result follows from Proposition \[p: conjugacy size\].
Now suppose that $G=PSL_2(q)$; then $|G|<q^3$ and $r=1$. For $q\geq 5$ and $q\neq 9$, $k=\frac1{(2,q-1)}(q-1)$ and it is clear that $k^8>q^3$. When $q=4$ we have $k=2$ and the result follows; likewise when $q=9$ we have $k=3$ and the result follows.
The next result was obtained independently in [@guralnick-kantor] and [@stein].
\[p: stein\] Each finite simple group $G$ is $\frac32$-generated; that is, for any nontrivial element $g$ of $G$ there exists $h$ in $G$ such that $\langle g, h\rangle=G$.
\[c: 32\] Let $G$ be a finite simple group and let $S$ be a subset of $G$ of size at least two. Then some translate of $S$ generates $G$.
Let $u$ and $v$ be distinct elements of $S$. Since $G$ is $\frac32$-generated, there exists $x$ in $G$ such that $\langle vu^{-1}, x\rangle=G$. Therefore the translate $Su^{-1}x$, which contains $x$ and $vu^{-1}x$, generates $G$.
The next result, the Product theorem, is our primary tool for proving Theorems \[t: gps\] and \[t: 2\]. Versions of this result can be found in [@bgt2; @ps2]. It was first proved by Helfgott for the groups $PSL_2(p)$ and $PSL_3(p)$ in [@helfgott2; @helfgott3].
\[t: generating sets\] Fix a positive integer $r$. There exists a positive constant $\eta=\eta(r)$ such that, for $G$ a finite simple group of Lie type of rank $r$ and $S$ a generating set of $G$, either $S^3=G$ or $|S^3|\geq |S|^{1+\eta}.$
We can now prove Theorem \[t: 2\].
Given a positive integer $r$, let $\eta$ be the constant from Theorem \[t: generating sets\]. It suffices to prove Theorem \[t: 2\] for sets $S$ of size larger than some constant $L>1$ that depends only on $\eta$, because if $|S|<L$, and $S^3\neq G$, then, by the simplicity of $G$, there is an element $g$ of $G$ such that $|SS^g|\geq |S|+1$, and $|S|+1\geq |S|^{1+\delta}$, where $\delta=\log(L+1)/\log L - 1$. In particular, we assume that $|S|\geq 8^{\frac{2}{\eta}}$, and we define $\varepsilon = \tfrac{1}{16}\min\left\{\eta,\tfrac{1}{24r^2}\right\}$.
Since $G$ is $\frac32$-generated, there exists an element $g$ of $G$ such that the set $T=S\cup \{g\}$ generates $G$. We can apply Theorem \[t: generating sets\] to $T$ to conclude that either $T^3=G$ or $|T^3|\geq |S|^{1+\eta}$.
Now, $T^3$ is the union of the eight sets $SSS$, $SSg$, $SgS$, $gSS$, $Sgg$, $gSg$, $ggS$ and $\{ggg\}$. Suppose that $|T^3|\geq |S|^{1+\eta}$. By the pigeon-hole principle at least one of the eight sets is larger than $\frac18 |S|^{1+\eta}$. We assumed earlier that $|S|\geq 8^{\frac{2}{\eta}}$, from which it follows that $\frac18 |S|^{1+\eta} > |S|^{1+\frac{\eta}{2}}$. Therefore one of the first seven of the eight sets is larger than $|S|^{1+\frac{\eta}{2}}$. All of these seven sets except $SSS$ are equal to a translate of the product of one or two conjugates of $S$, so if any of these have size at least $|S|^{1+\frac{\eta}{2}}$ then $|SS^h|\geq |S|^{1+\frac{\eta}{2}}$ for some element $h$ of $G$. If, on the other hand, $|SSS|\geq|S|^{1+\frac{\eta}{2}}$, then Lemma \[l: petridis\] (with $J=K=|S|^{\frac{\eta}{16}}$) implies that there is an element $h$ of $S\cup \{1\}$ with $|SS^h|\geq |S|^{1+\frac{\eta}{16}}$. Thus in both cases there is an element $h$ with $|SS^h|\geq |S|^{1+\varepsilon}$.
The remaining possibility is that $T^3=G$. If $S^3\neq G$ then Proposition \[p: gowers\] implies that $|S|\leq |G|/\sqrt[3]{k}$ where $k={\mathrm{mindeg}}(G)$. But Proposition \[p: landseitz\] gives that $|S|\leq |G|^{1-\frac{1}{24r^2}}$, and this implies, in particular, that $|T^3|=|G|\geq |S|^{1+\frac{1}{24r^2}}$. The argument of the previous paragraph applies again, to give $|SS^h|\geq |S|^{1+\varepsilon}$ for some element $h$.
Note that we can immediately deduce the following result of [@lns] (which we will use later).
\[c: polynomial\] Fix a positive integer $r$. There exists a constant $d$ such that if $G$ is a finite simple group of Lie type of rank $r$ and $S$ is a subset of $G$ of size at least two then $G$ is a product of $N$ conjugates of $S$ for some $N\leq 3 (\log|G|/ \log |S|)^d$.
Let $\varepsilon$ be the constant from Theorem \[t: 2\], and define $d=\log_{1+\varepsilon}2$. Let $M$ be the integer part of $\log_{1+\varepsilon} \frac{\log|G|}{\log|S|}$. Theorem \[t: 2\] implies that $G$ is the product of $3\cdot 2^M$ conjugates of $S$, and $$3\cdot 2^M \leq 3\left(\frac{\log|G|}{\log|S|}\right)^{d}.$$
The results in this section motivate a common generalisation of the Product theorem, and Conjecture \[c: 3\], for groups of Lie type.
\[c: 4\] There exists $\varepsilon>0$ and a positive integer $b$ such that the following statement holds. For each integer $r$ there is a positive integer $c(r)$ such that if $G$ is a finite simple group of Lie type of rank $r$ and $S$ a generating set of $G$, then either $|SS^g|\ge|S|^{1+\varepsilon}$ for some $g\in S^{c(r)}$, or else $G$ is the product of $b$ conjugates $S^{g_1},\dots,S^{g_b}$, where $g_1,\dots,g_b\in S^{c(r)}$.
It would be interesting to prove Conjecture \[c: 3\] in the case when $S$ is a subgroup of $G$. A rather general qualitative result in this direction was obtained by Bergman and Lenstra [@bergman-lenstra]. They show that if $H$ is a subgroup of a group $G$ satisfying $\big|H H^g\big|\leq K|H|$ for all $g$ in $G$, then $H$ is “close to” some normal subgroup $N$ of $G$, in the sense that $\big|H:H\cap N\big|$ and $\big|N:H\cap N\big|$ are both bounded in terms of $K$.
Proof of Theorem \[t: gps\] {#s: proof1}
===========================
Given an element $g$ of a group $G$ we define $$g^G=\{g^h\,:\, h\in G\},$$ and, for a subset $Z$ of $G$, $$Z^G = \{Z^h\,:\, h\in G\}.$$ We begin the proof of Theorem \[t: gps\] with a simple combinatorial lemma, which enables us to deal with “small” sets.
\[l: get a big set\] Let $S$ be a subset of a finite group $G$. There exist a positive integer $m$ and $m$ conjugates of $S$ such that their product $X$ satisfies $$|X|= |S|^m\geq \frac{\sqrt{{\mathrm{minclass}}(SS^{-1},G)}}{|S|}\geq
\frac{\sqrt{{\mathrm{minclass}}(G)}}{|S|}.$$
Define $X_1=S$ and, if possible, choose an element $g$ of $G$ such that $X_1^{-1}X_1\cap gSS^{-1}g^{-1}=\{1\}$. Define $X_2=X_1gSg^{-1}$. Notice that if $x_L, x_R\in X_1, s_L, s_R\in S$, and $x_Lgs_Lg^{-1}=x_Rgs_Rg^{-1}$, then $x_R^{-1}x_L=gs_Rs_L^{-1}g^{-1}$. Hence $x_R^{-1}x_L\in X_1^{-1}X_1\cap gSS^{-1}g^{-1}$, and so $x_L=x_R$ and $s_L=s_R$. It follows that $|X_2|=|X_1||S|$. Now repeat this process with $X_2$ replacing $X_1$, and so on.
The process terminates with a set $X$ of size $|S|^m$, which is a product of $m$ conjugates of $S$, and such that $|X^{-1}X\cap gSS^{-1}g^{-1}|\geq 2$ for all $g$ in $G$.
Let $T$ be a set of smallest possible size that intersects every conjugate of $Z=SS^{-1}$ nontrivially, and write $t=|T|$. Let $n=|G:N_G(Z)|$, the number of $G$-conjugates of $Z$. By the pigeonhole principle there exists an element $g$ of $Z$ that lies in at least $\frac{n}{t}$ different conjugates of $Z$. Let us count the set $$\Omega=
\big\{(g',Z') \in g^G\times Z^G \, \big| \, g'\in Z'\big\}$$ in two different ways.
First, since every conjugate of $g$ lies in the same number of conjugates of $Z$, we know that $|g^G| \frac{n}{t}\leq |\Omega|.$ On the other hand it is clear that $|\Omega|\leq n|Z|$. Putting these together we obtain that $|g^G|\frac{n}{t}\leq n|Z|$. Therefore $$t\geq \frac{|g^G|}{|Z|}\ge\frac{{\mathrm{minclass}}(SS^{-1},G)}{|S|^2}$$ and using $|X|^2\geq |X^{-1}X|\geq t$ our statement follows.
Lemma \[l: get a big set\] and Proposition \[p: conjugacy size\] imply that if $G$ is a simple group of Lie type of rank $r$ and $S$ a subset of size less that $q^{r/4}$ then we have $\big|SS^g\big|=|S|^2$ for some $g$ in $G$.
We are now ready to prove Theorem \[t: gps\].
As observed above, a product of conjugates of a translate of $S$ is equal to the translate of a product of conjugates of $S$. By Corollary \[c: 32\], a translate of $S$ generates $G$. Therefore we assume that $S$ generates $G$.
Suppose that $|S| \geq \big|{\mathrm{minclass}}(G)\big|^{1/4}$; then $|G|<|S|^{32r}$ by Proposition \[p: conjugacy size\]. Now Corollary \[c: polynomial\] implies that $G$ is a product of fewer than $3(32r)^d$ conjugates of $S$. The theorem holds in this case with $c=3(32r)^d$.
Suppose instead that $|S| < |{\mathrm{minclass}}(G)|^{1/4}$. By Lemma \[l: get a big set\] we can choose conjugates $S_1,\dots,S_m$ of $S$ such that the set $X=S_1\dotsb S_m$ satisfies $|X|=|S|^m$ and $$|X|\geq \frac{\sqrt{|{\mathrm{minclass}}(G)|}}{|S|} \ge
\big|{\mathrm{minclass}}(G)\big|^{1/4} \,.$$ It follows from the first part of the proof that $G$ is a product of fewer than $c\log|G|/\log|X|$ conjugates of $X$. Therefore $G$ is a product of fewer than $mc\log|G|/\log|X|$ conjugates of $S$ and, since $\log|X|=m\log|S|$, the result follows.
Plünnecke-Ruzsa estimates for nonabelian groups {#s: doubling}
===============================================
The following basic result in additive combinatorics is due to Plünnecke [@plun; @plun2] (see also [@taovu Section 6.5]).
\[t: doubling\] Let $A$ and $B$ be finite sets in an abelian group $G$ and suppose that $|A B|\leq K|A|$ where $K$ is a positive number. Then for any positive integer $m$ there exists a nonempty subset $X$ of $A$ such that $$|X B^m|\leq K^m|X|.$$ In particular, $|B^2|\leq K|B|$ implies that $|B^m|\leq K^m|B|$ for $m=1,2,\dotsc$.
The last statement (“In particular…”) is called the Doubling lemma; it does not hold for nonabelian groups, however, as we saw in Lemma \[l: petridis\], there are useful analogues in this context due to Helfgott, Petridis, Ruzsa and Tao [@helfgott3; @petridis; @Ru2009; @Ru2010; @tao]. Petridis also proved the following lemma [@petridis Proposition 2.1].
\[l: petridis 2\] Let $X$ and $B$ be finite sets in a group. Suppose that $$\frac{|XB|}{|X|} \leq \frac{|ZB|}{|Z|}$$ for all $Z\subseteq X$. Then, for all finite sets $C$, $$|CXB|\leq \frac{|CX| |XB|}{|X|}.$$
Using this lemma we can extend Plünnecke’s theorem to normal subsets of nonabelian groups. The statement and proof mimic [@petridis Theorem 3.1], which is a stronger version of Theorem \[t: doubling\].
\[t: normal doubling\] Let $A$ and $B$ be finite sets in a group $G$ with $B$ normal in $G$. Suppose that $|A B|\leq K|A|$ for some positive number $K$. Then there exists a nonempty subset $X$ of $A$ such that $$|X B^m|\leq K^m|X|$$ for $m=1,2,\dotsc$. In particular, $|B^2|\leq K|B|$ implies that $|B^m|\leq K^m|B|$ for $m=1,2,\dotsc$
We proceed by induction on $m$. First choose $X\subseteq A$ such that $$\frac{|X B|}{|X|} \leq \frac{|Z B|}{|Z|}$$ for all $Z\subseteq A$. Then $$|X B|\leq |X| \frac{|A B|}{|A|}\leq K|X|,$$ so the result is true for $m=1$.
Now suppose that $|X B^{m}|\leq K^{m}|X|$ for some positive integer $m$. Normality of $B$ implies that $|X B^{m+1}|=|B^{m} X B|$, and then Lemma \[l: petridis 2\] gives $$|X B^{m+1}| = |B^{m} X B|\leq \frac{|B^mX||XB|}{|X|} \leq K^{m+1}|X|.$$ This verifies the inductive step, and completes the proof of the theorem.
Following an argument of Petridis (see the proof of [@petridis Theorem 1.2]) we observe that the Plünnecke-Ruzsa estimates [@taovu Corollary 6.29] can also be generalised using Theorem \[t: normal doubling\].
Suppose that $A$ and $B$ are subsets of a group $G$, with $B$ normal in $G$, and $|AB|\leq K|A|$. Then $$|B^mB^{-n}|\leq K^{m+n}|A|$$ for all positive integers $m$ and $n$.
Theorem \[t: normal doubling\] suggests that certain techniques in additive combinatorics concerning subsets of abelian groups can be applied to normal subsets of nonabelian groups. The next example – which is a consequence of Plünnecke’s theorem, and generalises [@Ru2009 Corollary 2.4] – supports this suggestion.
\[t: plun\] Let $A$ and $B$ be subsets of a group $G$ with $B$ normal in $G$, and suppose that $|AB^j|\leq K|A|$ for some positive integer $j$. If $m\geq j$ then $$|B^m| \leq K^{\frac{m}{j}}|A|.$$
We use the notation of [@taovu Section 6.5]. Construct the $m$-tuple of directed bipartite graphs $$(G_{A,B}, G_{A B, B}, \dots, G_{A B^{m-1}, B}).$$ This $m$-tuple is a Plünnecke graph. Now Plünnecke’s theorem [@taovu Theorem 6.27] yields the result immediately.
Proof of Theorem \[t: normal subsets\] {#s: shalev}
======================================
In this section we prove Theorem \[t: normal subsets\] and generalise some related results of Shalev. We will need the following theorem of Liebeck and Shalev [@lieshal].
\[t: normal\] There exists an absolute positive constant $a$ such that, if $G$ is a finite simple group and $S$ is a nontrivial normal subset of $G$, then $G=S^m$, where $m\leq a\frac{\log|G|}{\log|S|}$.
Let $a$ be the absolute constant from Theorem \[t: normal\]. Choose a positive integer $b$ larger than $2a$. Suppose first that $|S|\geq \sqrt{|G|}$. Then Theorem \[t: normal\] implies that $G=S^m$ where $$m\leq \frac{a\log|G|}{\log|S|} \leq 2a \leq b,$$ and hence $S^b=G$.
Now suppose that $|S|\leq \sqrt{|G|}$. Then $$\frac{\log|S|}{a\log|G|} \geq \frac{\log|S|}{2a(\log|G|-\log|S|)}=\frac{\log|S|}{2a(\log(|G|/|S|)}.$$ Theorem \[t: normal\] implies, once again, that for some $m\leq\frac{a\log|G|}{\log|S|}$ we have $G=S^m$. Hence, applying Theorem \[t: normal doubling\] to the normal subset $S$, we see that $$\frac{|S^2|}{|S|} \geq \left(\frac{|S^m|}{|S|}\right)^{\frac{1}{m}}
\geq \left(\frac{|G|}{|S|}\right)^{\frac{\log|S|}{a\log|G|}}
\geq \left(\frac{|G|}{|S|}\right)^{\frac{\log|S|}{2a(\log(|G|/|S|)}}
=|S|^{\frac{1}{2a}}\geq |S|^\frac{1}{b},$$ and this completes the proof.
The next result is a strengthening of [@shalev Theorem 7.4].
\[p: shalev\] For every $\delta>0$ there exists $\varepsilon>0$ such that for any finite simple group $G$ and subsets $A$ and $B$ of $G$ with $B$ normal in $G$ and $|A|\leq |G|^{1-\delta}$ we have $$|AB|\geq |A||B|^{\varepsilon}.$$
We assume that $A$ is nonempty and $B$ is nontrivial, otherwise the result is immediate.
By Theorem \[t: normal\], $G=B^m$, where $m\leq a\frac{\log|G|}{\log|B|}$. Let $K=|AB|/|A|$. Then, by Theorem \[t: normal doubling\], there is a nonempty subset $X$ of $A$ such that $|XB^m|\leq K^m|X|$. It follows that $$|G|=|B^m|=|XB^m| \leq K^m|X|\leq K^m|A|.$$ Since $|A|\leq |G|^{1-\delta}$ and $m\leq a\frac{\log|G|}{\log|B|}$ we can rearrange this inequality to give $$|G|^\delta \leq K^{a\frac{\log|G|}{\log|B|}}.$$ This is equivalent to $|B|^\frac{\delta}{a}\leq K$, which, with $\varepsilon=\frac{\delta}{a}$, is the required result.
Proposition \[p: shalev\] constitutes the expansion result for $B^2$ that was partially proven in [@shalev Proposition 10.4]. Furthermore it goes some way towards a proof of [@shalev Conjecture 10.3] although what remains is the more difficult part of the conjecture.
We can strengthen [@shalev Proposition 10.4] in a different direction as follows.
\[p: shalev 2\] For every $\delta>0$ and positive integer $r$ there exists $\varepsilon>0$ such that for any finite simple group $G$ of Lie type of rank $r$ and any set $S\subseteq G$ such that $|S|\leq |G|^{1-\delta}$, there exists $g$ in $G$ such that $$|SS^g|\geq |S|^{1+\varepsilon}.$$
Given $\delta>0$ and a positive integer $r$, let $\varepsilon$ be the positive constant from Theorem \[t: 2\]. Now choose any subset $S$ of $G$ such that $|S|\leq |G|^{1-\delta}$. According to Theorem \[t: 2\], either $|S S^g|\geq |S|^{1+\varepsilon}$ or else $S^3=G$. In the former case the result is proven. In the latter case we apply Lemma \[l: petridis\] with $J=K=(|S^3|/|S|)^{1/10}$ to deduce the existence of an element $g$ of $G$ with $|SgS|> K|S|$. Then, using $S^3=G$ and $|G|\geq |S|^{1+\delta}$, it follows that $$|SgS| > \left(\frac{|S^3|}{|S|}\right)^{\frac{1}{10}}|S|\geq |S|^{1+\frac{\delta}{10}}.$$ Provided that $\varepsilon$ is chosen to be smaller than $\tfrac{\delta}{10}$, the inequality $|S S^g|\geq |S|^{1+\varepsilon}$ is again satisfied.
The Skew doubling lemma {#s: skew doubling}
=======================
The next result is another analogue of the Doubling lemma for nonabelian groups, which we call the [*Skew doubling lemma*]{}.
\[l: skew doubling\] If $S$ is a finite subset of a group $G$ such that, for some positive number $K$, $|S S^g|\leq K|S|$ for every conjugate $S^g$ of $S$, then $$|S_1\dotsb S_m|\leq K^{14(m-1)}|S|$$ for $m=1,2,\dotsc$, where each of $S_1,\dots,S_m$ is any conjugate of either $S$ or $S^{-1}$.
To prove Lemma \[l: skew doubling\] we will use Lemma \[l: petridis\] and the following result, Ruzsa’s triangle inequality [@ruzsa] (see also [@taovu Section 2.3]).
\[l: ruzsa\] Let $U$, $V$ and $W$ be finite subsets of a group $G$. Then $$\frac{|VW^{-1}|}{|U|} \leq \frac{|UV^{-1}|}{|U|} \frac{|UW^{-1}|}{|U|}.$$
First we prove a special case of Lemma \[l: skew doubling\].
\[l: tao\] Let $S$ be a finite subset of a group $G$. Suppose that $K$ is a positive number such that $|SS^g| \leq K|S|$ for each $g$ in $G$. Then $|S_1S_2S_3| \leq K^{14}|S|$, where each of $S_1$, $S_2$ and $S_3$ is any conjugate of either $S$ or $S^{-1}$.
Choose elements $a$ and $b$ of $G$. We can apply Lemma \[l: petridis\] with $J=K$ to obtain $$|S^3|\leq K^8|S|.$$ Using this inequality and Lemma \[l: ruzsa\] (with $U=S^{-1}$, $V=SS$ and $W=S$) we obtain $$\frac{|SSS^{-1}|}{|S|}\leq
\frac{|S^{-1}S^{-1}S^{-1}|}{|S|}\frac{|S^{-1}S^{-1}|}{|S|}=\frac{|S^3|}{|S|}\frac{|S^2|}{|S|}\leq
K^{9}.$$ Using this inequality and Lemma \[l: ruzsa\] (with $U=S$, $V=S^{-1}$ and $W=SS^{-1}$) we obtain $$\frac{|S^{-1}SS^{-1}|}{|S|}\leq
\frac{|SS|}{|S|}\frac{|SSS^{-1}|}{|S|}\leq K^{10}.$$ Using this inequality and Lemma \[l: ruzsa\] (with $U=S^{-1}$, $V=SS^{-1}$ and $W=Sa$) we obtain $$\frac{|SS^{-1}a^{-1}S^{-1}|}{|S|}\leq
\frac{|S^{-1}SS^{-1}|}{|S|}\frac{|S^{-1}a^{-1}S^{-1}|}{|S|}\leq K^{11}.$$ Using this inequality and Lemma \[l: ruzsa\] (with $U=S$, $V=SaS$ and $W=S^{-1}b^{-1}$) we obtain $$\label{e1}
\frac{|SaSbS|}{|S|}\leq
\frac{|SS^{-1}a^{-1}S^{-1}|}{|S|}\frac{|SbS|}{|S|}\leq K^{12}.$$ Using this inequality and Lemma \[l: ruzsa\] (with $U=S$, $V=S^{-1}$ and $W=S^{-1}b^{-1}S^{-1}a^{-1}$) we obtain $$\label{e2}
\frac{|S^{-1}aSbS|}{|S|}\leq \frac{|SS|}{|S|}\frac{|SaSbS|}{|S|}\leq K^{13}.$$ Finally, using this inequality and Lemma \[l: ruzsa\] (with $U=S^{-1}$, $V=S^{-1}aSb$ and $W=S$) we obtain $$\label{e3}
\frac{|S^{-1}aSbS^{-1}|}{|S|}\leq
\frac{|S^{-1}b^{-1}S^{-1}a^{-1}S|}{|S^{-1}|}\frac{|S^{-1}S^{-1}|}{|S^{-1}|} =
\frac{|S^{-1}aSbS|}{|S|}\frac{|SS|}{|S|} \leq
K^{14}.$$ Equations , and imply that, given any conjugates $S_1$, $S_2$ and $S_3$ of either $S$ or $S^{-1}$, we have $|S_1S_2S_3|/|S| \leq K^{14}$, as required.
We need the following proposition.
\[p: top\] If $A$ and $B$ are finite subsets of a group $G$ such that, for some positive number $K$, $|BB^g|\leq K|B|$ for every conjugate $B^g$ of $B$, then $$|AB_1B_2|\leq K^{14}|AB_3|,$$ where each of $B_1$, $B_2$ and $B_3$ is any conjugate of $B$ or $B^{-1}$.
By Lemma \[l: tao\] we have $$\frac{|B_3^{-1}B_1B_2|}{|B_3|} \leq K^{14},$$ where each of $B_1$, $B_2$ and $B_3$ is any conjugate of $B$ or $B^{-1}$. Applying Lemma \[l: ruzsa\] with $U=B_3^{-1}$, $V=A$ and $W=B_2^{-1}B_1^{-1}$ we obtain $$\frac{|AB_1B_2|}{|AB_3|}=\frac{|AB_1B_2|}{|B_3^{-1}A^{-1}|}\leq \frac{|B_3^{-1}B_1B_2|}{|B_3|}\leq K^{14},$$ as required.
We can finally prove Lemma \[l: skew doubling\].
The result holds trivially when $m=1$ and $m=2$. Suppose that $m\geq 3$. Apply Proposition \[p: top\] with $B=S$, $A=S_1\dotsb S_{n-2}$, $B_1=B_3=S_{n-1}$ and $B_2=S_{n}$ to see that $$\frac{|S_1\dotsb S_{n}|}{|S_1\dotsb S_{n-1}|} \leq K^{14}$$ for $n=3,4,\dots,m$. It follows that $$\begin{aligned}
\frac{|S_1\dotsb S_{m}|}{|S|} &= \left(\frac{|S_1\dotsb S_{m}|}{|S_1\dotsb S_{m-1}|}\right)\left(\frac{|S_1\dotsb S_{m-1}|}{|S_1\dotsb S_{m-2}|}\right)\dotsb \left(\frac{|S_1S_2S_3|}{|S_1S_2|}\right)\left(\frac{|S_1S_2|}{|S_1|}\right) \\
&\leq (K^{14})^{m-2}K \\
& \leq K^{14(m-1)},\end{aligned}$$ as required.
Using the Skew doubling lemma we can derive Conjecture \[c: 3\] from Conjecture \[c: 2\]. The proof is similar to the proof of Theorem \[t: normal subsets\].
Let $c$ be the absolute constant from Conjecture \[c: 2\]. We define $b$ to be a positive integer greater than $2c$, and $\varepsilon=1/(28c)$. Suppose first that $|S|\geq \sqrt{|G|}$. Then Conjecture \[c: 2\] implies that $G=S_1\dotsb S_N$, for conjugates $S_1,\dots,S_N$ of $S$, where $$N\leq \frac{c\log|G|}{\log|S|} \leq 2c < b,$$ and hence $G$ is certainly the product of $b$ conjugates of $S$.
Now suppose that $|S|\leq \sqrt{|G|}$. Then $$\frac{\log|G|-\log|S|}{c\log|G|-\log|S|} \geq
\frac{\log|G|-\log|S|}{c\log|G|} \geq \frac{1}{2c}.$$ In particular observe that $$c\log|G|-\log|S| \leq 2c(\log|G|-\log|S|) = 2c\log(|G|/|S|).$$ Conjecture \[c: 2\] implies, once again, that for some $N\leq\frac{c\log|G|}{\log|S|}$ we have $G=S_1\dotsb S_N$, for conjugates $S_1,\dots,S_N$ of $S$. Using the Skew doubling lemma, Lemma \[l: skew doubling\], we see that there is an element $g$ of $G$ for which $$\frac{|SS^g|}{|S|} \geq \left(\frac{|S_1\dots
S_N|}{|S|}\right)^{\frac{1}{14(N-1)}}
\geq \left(\frac{|G|}{|S|}\right)^{\frac{\log|S|}{14(c\log|G|-\log|S|)}}
\geq \left(\frac{|G|}{|S|}\right)^{\frac{\log|S|}{28c(\log(|G|/|S|))}}
\geq |S|^{\frac{1}{28c}},$$ and this completes the proof.
[^1]: L.P. is supported in part by OTKA NK78439 and K84233
[^2]: E.Sz. is supported in part by OTKA NK81203 and K84233
[^3]: N.G. would like to thank Harald Helfgott for allowing the use of his research funds to facilitate a visit to the Rényi Institute during which work on this paper was initiated. He would also like to thank the University of Bristol to which he has been a regular visitor over the course of writing this paper.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Scalarizing functions have been widely used to convert a multiobjective optimization problem into a single objective optimization problem. However, their use in solving (computationally) expensive multi- and many-objective optimization problems in Bayesian multiobjective optimization is scarce. Scalarizing functions can play a crucial role on the quality and number of evaluations required when doing the optimization. In this article, we study and review 15 different scalarizing functions in the framework of Bayesian multiobjective optimization and build Gaussian process models (as surrogates, metamodels or emulators) on them. We use expected improvement as infill criterion (or acquisition function) to update the models. In particular, we compare different scalarizing functions and analyze their performance on several benchmark problems with different number of objectives to be optimized. The review and experiments on different functions provide useful insights when using and selecting a scalarizing function when using a Bayesian multiobjective optimization method.
**Keywords:** metamodelling, machine learning, multiple criteria decision making, Pareto optimality, computational cost, Bayesian optimization
author:
- Tinkle Chugh
bibliography:
- 'Master\_File.bib'
title: Scalarizing Functions in Bayesian Multiobjective Optimization
---
INTRODUCTION
============
Bayesian multiobjective optimization methods have been widely used for solving (computationally) expensive multiobjective optimization problems (MOPs) and in the last few years, many articles have been published. A survey of many Bayesian and non-Bayesian methods with their advantages and limitations is provided in [@Chugh2018; @Jin2005; @Jin2011; @Richard2016; @Deb2018; @Horn2015].
A generic framework of a Bayesian optimization method is shown in Figure \[fig:surrogate\], where the steps are mentioned inside squares.
In the first step, a set of samples is generated e.g. using a design of experiment technique (DOE) like Latin hypercube sampling [@Mckay2000]. These samples are then evaluated with expensive objective functions in the second step. A termination criterion (typically maximum number of expensive function evaluations) is checked in the third step. If the termination criterion is met, nondominated solutions from all evaluated solutions (i.e. from expensive evaluations) are used as the final solutions. Otherwise, surrogate models are built using evaluated solutions in the fourth step. An optimizer e.g. an evolutionary algorithm is then used with the models in the fifth step to find promising samples by using an appropriate infilling criterion (or updating criterion or acquisition function). In the sixth step, a fixed number of samples generated with the optimizer is selected which are then evaluated with expensive objective functions.
All six steps mentioned above are important in the performance of a Bayesian optimization method. In this article, our focus is on the fourth step i.e. building or training surrogates. Once, the samples are evaluated with expensive objective functions in the second step, there can be different ways to build surrogates to be used in a Bayesian optimization method. The most common way is to build surrogate for each objective function [@Chugh2016a; @Ponweiser2008a; @Zhang2010; @Lim2010]. For instance, in [@Chugh2016a; @Zhang2010; @Ponweiser2008a], Gaussian process models are used for different objective functions. Another way is to build a surrogate for a scalarizing function after converting multiobjective optimization problem into a single objective optimization problem [@Knowles2006; @Alma2017], which is also the focus of this article. There are other approaches for building surrogates like classification of solutions into different ranks or classes [@CSEA2018; @Loshchilov2009; @Loshchilov2010; @Pilat2011b; @Bandaru2014]. For details on other ways of building surrogates, see [@Chugh2018; @Richard2016]. Note that scalarizing function can also be used after building models on each objective function e.g. as in [@Zhang2010; @Yew_Soon2012] to do a local search. In this article, we focus on building surrogate after converting a multiobjective optimization problem into a single objective by using a scalarizing function.
There are two main advantages to make a single surrogate in solving an expensive MOP. The first one is that only one surrogate is used in the solution process instead of multiple surrogates, which reduces the computational burden e.g. the training time especially in many-objective (usually more than three objectives) optimization problems. The second advantage is that one can use an infill criterion proposed for single-objective optimization problems which also reduces the computational complexity. In the literature, a little attention has been paid in using scalarizing functions for building surrogates on them and only a few studies exist in the literature. For instance, in [@Knowles2006], an augmented achievement scalarizing function (AASF) was used and in [@Alma2017], hypervolume improvement, dominance rank and minimum signed distance were used (more details are provided in the Section \[sec:scf\_functions\]). It is worth important to be pointed out that several studies exist [@Coello2017; @Ishibuchi2016; @Coello2017b; @Miettinen2002; @Miettinen1999] on using scalarizing functions without using them in a Bayesian optimization method. In this article, we study 15 different scalarizing functions to build surrogates on them in solving (computationally) expensive MOPs and particularly, focus on answering the following research questions:
1. are different scalarizing functions perform different to each other under the same framework?
2. are surrogate models built on a given scalarizing function sensitive to the number of objectives ?
To answer the questions above, we embed different scalarizing functions into the framework of efficient global optimization (EGO) [@Jones1998]. The algorithm uses the expected improvement criterion for updating the surrogates build on the scalarizing function. We then test the method with different scalarizing functions on several benchmark problems with different numbers of objectives. We analyze the accuracy and uncertainty provided by Gaussian process models on different scalarizing functions with the different number of objectives.
The rest of the article is organized as follows. In the next section, we provide a literature survey on using scalarizing functions with and without building surrogates on them. In Section 3, we give a brief introduction to different functions with their mathematical formulations and merits and demerits. In Section 4, numerical experiments are conducted to answer the research questions mentioned above. We finally conclude and mention the future research directions in Section 5.
Related work
============
Use of scalarizing functions has been a topic of interest in different communities. For instance, in Multiple Criteria Decision Making (MCDM) problems, several scalarizing functions have been used to incorporate a decision maker’s preferences [@Miettinen1999; @Hwang1979; @Steuer1986; @Miettinen2002]. In evolutionary multiobjective optimization (EMO) algorithms especially decomposition based, utilization of scalarizing functions became popular in the last few years. Many studies and algorithms exist in the literature utilizing different functions for solving multi- and many-objectives optimization problems. Some of the well-known decomposition based EMO algorithms utilizing different scalarizing functions are nondominated-sorting genetic algorithm III (NSGA-III) [@Deb2014; @Deb2014b], multiobjective optimization based on decomposition (MOEA/D) [@Zhang2007] and its numerous versions [@Survey_MOEA/D] and reference vector guided evolutionary algorithm (RVEA) [@Ran2016]. For more details on the working principle of decomposition based algorithms, see surveys on many-objective optimization evolutionary algorithms [@Ishibuchi2008b; @Li2015].
In [@Hisao2010], two different scalarizing functions weighted sum and augmented Chebyshev was used adaptively in the solution process in the framework of MOEA/D by using a multi-grid scheme. The proposed idea was tested on a knapsack problem with four and six objectives and performed better than the original version of MOEA/D. The authors extended their work in [@Hisao2016], and modified the penalty boundary intersection (PBI) function in MOEA/D to handle different kinds of Pareto front. Two modified PBI functions called two-level PBI and quadratic PBI were tested in the MOEA/D framework and the algorithm with two new scalarizing functions performed better than the original version.
A detailed study on different scalarizing functions and their corresponding parameters was conducted in [@Derbel2014]. Instead of proposing a new algorithm, the authors showed the performance of the different scalarizing function in a simple (1+$\lambda$)-evolutionary algorithm [@Derbel2014] on bi-objective optimization problems. In [@Coello2017], the performance of 15 different scalarizing functions was tested in MOEA/D [@Zhang2007] and MOMBI-II [@Coello2015] algorithms. The authors used a tool called EVOCA [@Riff2013] to tune the parameter values in different functions. Two algorithms MOEA/D and MOMBI-II [@Coello2015] with different scalarizing functions were tested on Lame Supersphere test problems [@Emmerich2007]. The authors found out that the performance of two algorithms depends on the choice of scalarizing functions.
In [@Coello2017b], a hyper-heuristic was used to rank different scalarizing functions with a measure called s-energy [@Hardin2004]. The proposed algorithm was tested on ZDT, DTLZ and WFG benchmark problems with 2-10 objectives and compared with MOMBI-II, MOEA/D and NSGA-III and found better results in most of the instances. In [@Jiang2018], two new scalarizing functions called the multiplicative scalarizing function (MSF) and penalty-based scalarizing function (PSF) was used in MOEA/D-DE [@Zhang2017]. The proposed scalarizing functions performed better than penalty boundary intersection, Chebyshev and weighted sum scalarizing functions.
The first algorithm in Bayesian multiobjective optimization method using a scalarizing function and building a surrogate on it was proposed in [@Knowles2006] and known as ParEGO (for Pareto based efficient global optimization). A Gaussian process model [@Rasmussen2006] was used as a surrogate of the scalarizing function. In the algorithm, a set of reference vectors was uniformly generated in the objective space using simplex lattice-design method [@Cornell2011]. In each iteration, the algorithm randomly selected a reference vector among a set of vectors, which was then used in the Chebyshev function to build the surrogate. The algorithm was compared with NSGA-II [@Deb2002] and performed significantly better. In a recent study in [@Alma2017], three scalarizing functions, hypervolume improvement, dominance ranking, and minimum signed distance were proposed in solving expensive MOP. These scalarizing functions do not use reference vectors and preserve the dominance relationship. The authors also used the framework of EGO and compared with SMS-EGO [@Ponweiser2008a] and ParEGO (i.e. with Chebyshev scalarizing function). The results outperformed the ParEGO algorithm and performed similarly to SMS-EGO in some cases. However, the reason for the better performance of their approach comparison to ParEGO was not mentioned.
Scalarizing functions {#sec:scf_functions}
=====================
In this section, we summarize 15 different scalarizing functions studies in this work with their merits and demerits. All these functions have already been explained in details in the literature [@Coello2007; @Miettinen1999]. Therefore, we provide a brief summary of these functions. Before going into description of scalarizing function, we define a multiobjective optimization problem (MOP) as:
$$\begin{gathered}
\mbox{minimize } {\left\{f_1(x),\ldots,f_k(x)\right\}} \\
\mbox{subject to} \ x \in S
\end{gathered}
\label{MOOP}$$
with $k (\geq2)$ objective functions $f_i(x)$: *S*$\rightarrow\Re^n$. The vector of objective function values is denoted by $f(x)=(f_1(x),\ldots,f_k(x))^T$. The (nonempty) feasible space $S$ is a subset of the decision space $\Re^n$ and consists of decision vectors $x=(x_1,\ldots,x_n)^T$ that satisfy all the constraints. Different scalarizing functions are defined as follows:
1. **Weighted sum (WS)**: The weighted sum combines different objectives linearly and has been widely used [@Miettinen1999]. It converts the a MOP into a single-objective optimization problems as: $$ g = \sum_{i=1}^{k}w_if_i.$$ One major limitation in using WS is that it cannot find solutions in non-convex parts of the Pareto front.
2. **Exponential weighted criterion (EWC)**: EWC was first used in [@Athan1996] to overcome the limitations of WS: $$g = \sum_{i=1}^{k}\exp(pw_i-1)\exp(pf_i).$$ As mentioned in [@Coello2007], the performance of the function depends on the value of $p$ and usually a large value of $p$ is needed.
3. **Weighted power (WPO)**: The weighted power defined below can find solutions when the Pareto front is not convex. However, it also depends on the parameter $p$ as in EWC. $$g = \sum_{i=1}^{k}w_i(f_i)^p.$$
4. **Weighted norm (WN)**: The weighted norm (or $L_p$ metric) is a generalized form of the weighted sum: $$g = (\sum_{i=1}^{k}w_i|f_i|^p)^{1/p}.$$ The weighted norm has been used in MOEA/D [@Zhang2007]. Likewise EWC and WPO, its performance also depends on the choice of the $p$ parameter.
5. **Weighted product (WPR)**: The weighted power is defined as: $$g = \prod_{i=1}^{k}(f_i)^{w_i}.$$ It is also called as product of powers [@Coello2007] and can find solutions in non-convex parts of the Pareto front. However, like EWC, WPO and WN, its performance depends on the value of $p$.
6. **Chebyshev function (TCH)**: The Chebyshev function can be derived from the WN function with $p=\infty$. This function has been used in many EMO algorithms such as MOEA/D [@Zhang2007] and its versions [@Survey_MOEA/D]: $$g = \max_i[w_i|f_i - z_i^*|],$$ where $z_i^*$ is the ideal or utopian objective vector. In this work, we normalize the expensive objective function values in the range \[0,1\] before building the model. Therefore, $z_i^*$ is a vector of zeros.
7. **Augmented Chebyshev (ATCH)**: In [@Steuer1986], it was suggested that weakly Pareto optimal solutions can be avoided by adding an augmented term to TCH: $$g = \max_i[w_i|f_i - z_i^*|] + \alpha\sum_{i=1}^{k}|f_i - z_i^*|$$ Moreover, this function was the first to be used as a surrogate in [@Knowles2006].
8. **Modified Chebyshev (MTCH)**: A slightly modified form of MTCH was used in [@Kaliszewski1987]: $$g = \max_i\big[ w_i(|f_i - z_i^*| + \alpha\sum_{i=1}^{k}|f_i - z_i^*|)\big]$$ As mentioned in [@Miettinen1999] that main different between ATCH and MTCH is the slope to avoid weakly Pareto optimal solutions.
9. **Penalty boundary intersection (PBI)**: The PBI function was first used in MOEA/D and used as the selection criterion for balancing convergence and diversity: $$g = d_1 + \theta d_2,$$where $d_1 = |\mathbf{f} \cdot \frac{\mathbf{w}}{\| \mathbf{w} \|}|$ and $d_2 = \|\mathbf{f} - d_1 \frac{\mathbf{w}}{\| \mathbf{w} \|}\|$. The PBI function has been widely used in EMO algorithms [@Survey_MOEA/D]. However, as shown in [@Ishibuchi2010] its performance is effected by the $\theta$ parameter.
10. **Inverted penalty boundary intersection (IPBI)**: To enhance the diversity of solutions, IPBI was proposed in [@Sato2014]: $$g = \theta d_2 - d_1,$$ where $d_1 = |\mathbf{f}^{'} \cdot \frac{\mathbf{w}}{\| \mathbf{w} \|}|$ and $d_2 = \|\mathbf{f}^{'} - d_1 \frac{\mathbf{w}}{\| \mathbf{w} \|}\|$ and $\mathbf{f}^{'} = \mathbf{z}^{nadir} - \mathbf{f}$. In this work, we considered the vector of worst expensive objective function values as $\mathbf{z^nadir}$.
11. **Quadratic PBI (QPBI)**: Recently, an enhanced version of PBI function was proposed in [@Hisao2016]: $$g = d_1 + \theta d_2 \frac{d_2}{d^*},$$ where $d_1$ and $d_2$ are same as in PBI and $d^*$ is an adaptive parameter and defined as: $$d^* = \alpha \frac{1}{H} \frac{1}{k} \sum (z^{nadir} - z^{ideal})$$ where $\alpha$ is a pre-defined parameter and $H$ is a parameter used in generating the reference (or weight vectors) in decomposition based EMO algorithms. For more details about these parameters, see [@Zhang2007; @Ran2016].
12. **Angle penalized distance (APD)**: The APD function is a recently proposed scalarizing function and used as the selection criterion in reference vector guided many-objective evolutionary algorithm (RVEA) [@Ran2016]: $$g = 1 + P(\theta) \cdot \|\mathbf{f}\|,$$ where $P(\theta) = k (\frac{FE}{FE^{max}})^{\alpha} \frac{\theta}{\gamma}$, $FE$ is the number of expensive function evaluations at the current iteration and $FE^{max}$ is the maximum number of expensive function evaluations. The angle between an objective vector $\mathbf{f}$ and the reference vector to which it is assigned is represented by $\theta$ and the minimum of all angles between a reference vector selected and other reference vectors is represented by $\gamma$. This function adaptively balances the convergence and diversity based on the maximum number of function evaluations. The performance of the function depends on the value of $\alpha$ and $FE^{max}$.
13. **Hypervolume improvement (HypI)**: The HypI is a recently proposed scalarizing function in a surrogate-based algorithm [@Alma2017]. In contrast to its name, this scalarizing function is designed in such a way that it uses the Pareto dominance. Given a set of solutions $X$, a nondominated sorting is performed to find fronts of different ranks as in [@Deb2002]. Let the different fronts of ranks $1,2, \ldots$ are denoted by $P_1, P_2, \ldots$ and the hypervolume of a front $P_k$ given a reference point $r$ is denoted by $H(P_k)$. Then hypervolume improvement a solution $x \in X$ belongs to the front $P_k$ is then given by: $$g = H(x \cup P_{k+1}).$$ In this way, the Pareto dominance is preserved when calculating the hypervolume contributions of solutions. For problems with many-objectives, where all solutions are non-dominated i.e. belong to only one front, this scalarizing function might not be suitable. In other words, if all solutions are nondominated, their HypI values will be similar and the algorithm embedding the model will not be able to solve problems with many-objectives. Moreover, the performance of the function can be sensitive to the reference point in calculating the hypervolume.
14. **Dominance ranking (DomRank)**: This function is also recently proposed in [@Alma2017] and assigns fitness values based on the ranks of different solutions as done in the MOGA algorithm [@Fonseca1993]. In [@Fonseca1993], a solution is assigned a rank as: rank(x)= 1+$p$, where $p$ is number of solutions dominating x. Similarly, in [@Alma2017], given a set of solutions (with expensive evaluations) $X$ the fitness of a solution $x$ is: $$g = 1 - \frac{rank(x)-1}{|X|-1}.$$ For instance, the rank of a solution $x'$ dominated by all other solutions would be $rank(x') = 1 + |X| -1$, and therefore, the fitness of solution would be $g(x') = 1 - \frac{1 + |X| - 1 -1}{|X|-1} = 0$. Similarly, the rank of solutions belong the first front will be 1. This function is maximized to find samples for training the surrogates. Similar to HypI, this function also uses Pareto dominance and might not be suitable for problems with many-objectives.
15. **Minimum Signed Distance (MSD) [@Alma2017]**: The MSD function is proposed in [@Alma2017] and defined as: $$g = \min d(x', x)$$ where $x'$ are the solutions belong the first front (or rank one solutions) and $d(x',x) = \sum_{i=1}^k f_i(x') - f_i(x)$. For instance, if a solution a dominates another solution b, then $g(a,x') > g(b,x')$.
Numerical experiments
=====================
This section provides a comparison of different functions and analysis when building surrogates on them. As mentioned we used the framework of EGO [@Jones1998] and built Gaussian process as surrogates on the scalarizing functions. The expected improvement criterion in EGO is maximized with a genetic algorithm for selecting samples when updating the surrogates.
Performance of different scalarizing functions
----------------------------------------------
To compare the performance of different scalarizing functions, we used DTLZ [@Deb2005] and WFG [@Huband2005] problems with 2, 3, 5 and 10 number of objectives. In DTLZ suite, the number of variables is kept to $k$+5-1, where $k$ is the number of objectives. For the WFG suite, the number of variables is defined by position ($d$) and distance ($l$) parameters. We used the number of parameters as suggested in [@Huband2005]. For two objectives, $d$ is set to four and $2\times(k-1)$ for the rest of the objectives. The distance parameter is set to four for all objectives. To be summarized, the number of variables ($n$) for DTLZ and WFG suites is given in Table \[tab:no\_variables\_WFG\]:
DTLZ
----- ----- ------ ----- -----
$k$ $d$ $l$ $n$ $n$
2 4 4 8 6
3 4 4 8 7
5 8 4 12 9
10 18 4 22 14
: \[tab:no\_variables\_WFG\]Number of variables ($n$) for WFG and DTLZ suites
There are several parameters in different scalarizing functions and we used the recommended values from the respective articles. Different parameter values used are provided in Table \[tab:parameters\]:
Scalarizing function Parameter value
---------------------- ------------------------------------------------------
EWC $p$ = 100
WPO $p$ = 3
WN $p$ = 0.5
ATCH $\alpha$ = 0.0001
MTCH $\alpha$ = 0.0001
PBI $\theta$ = 5
IPBI $\theta$ = 5
QPBI $\theta$ = 1
APD $\alpha$ = 2
HypI $r$ (In table \[tab:ref\_points\] for DTLZ problems)
: \[tab:parameters\] Parameters values used in different scalarizing functions
Problem reference point
--------- --------------------
DTLZ1 400 $\times$ (1,k)
DTLZ2 1.5 $\times$ (1,k)
DTLZ3 900 $\times$ (1,k)
DTLZ4 2 $\times$(1,k)
DTLZ5 1.5 $\times$ (1,k)
DTLZ6 6 $\times$ (1,k)
DTLZ7 5 $\times$ (1,k)
: \[tab:ref\_points\]Reference point in calculating hypevolumes in HypI and as performance measure:
For WFG problems, the reference point $r$ in HyPI is used as $2 \times z^{nadir}$. The vector $z^{nadir}$ contains the maximum objective function values in the Pareto front of the given problem. We ran 21 independent runs for each scalarizing function with 300 maximum number of expensive function evaluations. We show the performance of different functions with inverted generational distance (IGD) and hypervolume. The mean IGD values with standard deviation (in parentheses) for DTLZ problems are provided in Table \[tab:IGD\_DTLZ\] and other results including IGD and hypervolume on WFG are in the supplementary material[^1]. To compare the results of different functions, we used the Wilcoxon rank sum test and used the Gaussian (or RBF) kernel when using the Gaussian process model.
In tables, the values statistically similar to the best value are in bold, the best value is in the circle and the worst value is underlined. As can be seen, the performance of different functions varies with different problems. Moreover, we provide in how many instances, a particular scalarizing function has the best IGD value in Figure \[fig:bar\_graph\]. The total number of instances tested were (7 + 9) $\times$ 4 = 64 i.e. (number of DTLZ problems + number of WFG problems) $\times$ number of cases with respect to objectives.
![\[fig:bar\_graph\]Number of wins of different scalarizing functions](DTLZ_WFG_bar_graph_IGD.pdf)
As can be seen from tables and the figure, three functions PBI, HypI, and APD outperformed other functions. Moreover, WS, ATCH, and DomRank never had the best IGD value in any of 64 instances. Note that, all these three functions have statistically similar values to the best IGD value in many cases as can be seen from tables. Another interesting observation from the results is that three different scalarizing functions ATCH, DomRank and MSD used in the literature for building surrogates on them were not in the top list. On the other hand, scalarizing functions used as the selection criterion in EMO algorithms e.g. APD and PBI performed significantly better. These results indicate that it is not straightforward to select a particular scalarizing function. However, one can analyze the fitness landscape of different functions after building surrogates. Moreover, this landscape can be sensitive to the number of objectives and the details are provided in the next subsection.
Sensitivity towards number of objectives
----------------------------------------
When using scalarizing functions for building surrogates on them, the number of objectives can play a crucial role in their performance. Before using the surrogate built on the scalarizing function, we can analyze and visualize the fitness landscape with respect to the accuracy and uncertainty of approximations (or predictions). We start by showing the bar graph of the number of wins with respect to 2, 3, 5 and 10 objectives in Figure \[fig:bar\_graph\_objectives\].
\
As can be seen in the figure that the performance of different functions varies with the number of objectives. The reason is that different formulations of scalarizing functions provide a different landscape. For a better illustration, an example of fitness landscapes of DTLZ2 with two and 10 objectives for different functions are shown in Figures \[fig:fitness\_landscape\_DTLZ2\_2\] and \[fig:fitness\_landscape\_DTLZ2\_10\], respectively. In figures, we used a training data set (shown as circles in figures) and built the surrogates on scalarizing functions. Then we generated a testing data set and used the model to get the predicted values (solid lines in figures) and uncertainties of the predicted values (shaded region in figures). The dotted line represents the actual scalarizing function values (denoted by g in figures) on the testing data set. For visualization purposes, we could show only one of the variables values.
              
              
As can be seen, for the same training data set (and therefore same objective function values), different scalarizing functions give different values. Therefore, the surrogate model to be built strongly depends on the g function values. For instance, in many-objective case, most of the solutions were nondominated and functions which use the property of Pareto dominance produce similar g values, as can be seen in the plots of HypI, DomRank, and MSD in 10 objective cases. As the g function values are the same for most of the solutions, the algorithm (EGO in this work) for finding a sample to update the surrogate by optimizing an infill criterion (EI in this work) will not be able to enhance the accuracy of the surrogates This is also the reason that HypI performed very well in a low number of objectives and its performance deteriorated with an increase in the number of objectives. On the other hand, functions like APD and PBI are not very sensitive to the number of objectives as they do not use the property of Pareto dominance and suitable for a large number of objectives when building surrogates on them. In addition to the sensitivity towards many objectives, some functions, EWC and WS did not perform well in most of the problems. In EWC, the g function landscape is very rugged [@Malan2009] as can be seen in figures and WS has the problem of finding solutions in the non-convex parts of the Pareto front. These results show that one needs to see or analyze the landscape of scalarizing functions for the given training data set before using them in the optimization. We observed the similar behavior on other problems and landscapes on other problems with different objectives are provided in the supplementary material.
Conclusions and future research directions
==========================================
This work focused on reviewing, analyzing and comparing different scalarizing functions in Bayesian multiobjective optimization for solving (computationally) expensive multi- and many-objective optimization problems. We provided an overview of different functions with their merits and demerits. We built the surrogates on the different functions in the framework of Bayesian multiobjective optimization. We compared different functions with the different number of objectives by using IGD and hypervolume as the performance metrics. The results clearly showed that some functions outperformed others in many cases and some did not work in most of the cases. We then analyzed the fitness landscape of different functions with respect to the number of objectives. We found out that some of the functions are very sensitive to the number of objectives and an analysis of the landscape of different functions might be helpful in selecting functions when building surrogates on them and using them in the optimization for solving expensive MOPs.
In this work, we did not compare the scalarizing functions with other non-Bayesian multiobjective optimization algorithms. Therefore, comparison with other algorithms are topics for future research directions. Moreover, working on an adaptive strategy in using and selecting scalarizing functions and their corresponding parameters are also in the list of future research.
Acknowledgement {#acknowledgement .unnumbered}
===============
This research is supported by the Natural Environment Research Council \[NE/P017436/1\]
[^1]: available from https://github.com/tichugh/SCF
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Discriminative training has turned out to be effective for robust tracking. However, online learning could be simply applied for classification branch, while still remains challenging to adapt to regression branch due to its complex design. In this paper, we present the first fully convolutional online tracking framework (FCOT), with a focus on enabling online learning for both classification and regression branches. Our key contribution is to introduce an anchor-free box regression branch, which unifies the whole tracking pipeline into a simpler fully convolutional network. This unified framework is beneficial to greatly decrease the complexity of tracking system and allows for more efficient training and inference. In addition, thanks to its simplicity, we are able to design a regression model generator (RMG) to perform online optimization of regression branch, making the whole tracking pipeline more effective in handling target deformation during tracking procedure. The proposed FCOT sets a new state-of-the-art results on five benchmarks including GOT-10k, LaSOT, TrackingNet, UAV123 and NFS, and performs on par with the state-of-the-art trackers on OTB100, with a high running speed of 53 FPS. The code and models will be made available at <https://github.com/MCG-NJU/FCOT>.'
author:
- Yutao Cui
- Cheng Jiang
- 'Limin Wang[^1]'
- Gangshan Wu
bibliography:
- 'egbib.bib'
title: Fully Convolutional Online Tracking
---
Introduction
============
Visual object tracking is a fundamental computer vision task, which aims at estimating the state of an arbitrary target in every frame of a video, given its bounding box in the first frame. It has a variety of applications such as human-computer [@introduction1] and visual surveillance [@introduction2]. However, tracking still remains as a highly challenging task due to several factors such as illumination changes, occlusion, and background clutter. In addition, target appearance variation along temporal dimension will further add difficulty for robust tracking.
In general, object tracking comprises a classification branch, to locate the target coarsely by discriminating the target from the background, and an regression branch [@atom], to generate an accurate bounding box of the target. For classification task, the current approaches could be roughly divided into generative trackers (e.g. SiamFC [@siamfc]) and discriminative tracker (e.g., DiMP [@dimp]). The generative trackers typically employs a fixed target template without modeling background clutter, while the discriminative trackers learns an adaptive filter by maximizing the response gap between target and background. It is well established that this discriminative training would increase the robustness of tracking [@dimp]. For regression task, the existing methods usually depend on hand-crafted designs, such as anchor box placement [@siamrpn; @siamrpnPlus; @dasiamrpn], or box sampling and selection [@dimp]. Due to this complex design, this regression branch is cannot be easily optimized with online learning for each tracked target just as the updating classification branch in a discriminative tracker. Therefore, a natural question arises [*whether we could also design a simple box regression branch, analogue to classification branch, that could be easily updated with online learning and efficiently deployed in practice*]{}.
{width="12cm"}
\[fig:1\]
Based on the above analysis, we present a Fully Convolutional Online Tracker, termed as FCOT, to yield a conceptually simple, relatively efficient, and more precise tracking framework. Our FCOT allows the whole method to run in a principled fully convolutional manner without any hand-crafted design, but also enable to both classification and regression branches to be online optimized for more robust and precise tracking. The key part of our FCOT framework is to design an online anchor-free box regression branch, that direct regresses the bounding box size of target in each frame. This anchor-free box branch could be deployed in a fully convolutional manner during tracking process. Thus it enables the whole tracking system to be efficiently optimized during training and easily deployed for inference. In addition, due to its simplicity in design, an online optimization algorithm is proposed to adaptively tune the parameters of box regression branch, making it to effectively deal with object deformation along time and thus yield more precise tracked results, as shown in Figure \[fig:1\].
Specifically, our FCOT framework starts with an new encoder-decoder architecture for high resolution feature extraction. Introducing upsampling layers into tracking system turns to be crucial for improving the accuracy of tracked results. Then, FCOT is composed of a classification branch for roughly localizing object center and a regression branch for regressing bounding box size. Both classification and regression branches are implemented with deformable convolutions due to its good performance handling deformation. To discriminate the tracked target with background and tackle the deformation issue of object shape during training, we propose to adpatively tune the parameters of both branches by online learning. Inspired by DiMP, we design a novel online regression model generator (RMG), composed of a model initializer and an online model optimizer. The effectiveness of our FCOT framework is demonstrated on the common tracking datasets, and it demonstrates our online box regressor is able to consistently improve tracking performance, in particular for higher IoU criteria. In summary, our main contributions are three-fold:
- We propose a Fully Convolutional Online Tracker (FCOT) with a simple architecture to implement target classification and regression directly, which can improve the tracking accuracy yet guaranteeing efficiency.
- We design a Regression Model Generator (RMG) to online optimize the regression model, which can estimate a precise target box in face of target appearance variations and deformations.
- The proposed FCOT outperforms all the popular state-of-the-art trackers on five benchmark datasets including GOT-10k [@got10k], LaSOT [@lasot], TrackingNet [@trackingnet], UAV123 [@uav123] and NFS [@nfs], and performs on par with the state-of-the-art trackers on OTB100 [@otb], while running at a high speed of 53 fps.
Related Work
============
Generally, visual objective tracking can be divided into target classification and regression subtasks [@siamDW; @CRPN; @Learning_the_Model_Update; @Real_Time_Object_Tracking]. In this section, we briefly review recent trackers from the two aspects.
Target Classification
---------------------
Modern tracking methods can be categorised as generative trackers and discriminative trackers. The former one is based on template matching, typically using Siamese networks [@siamfc; @siamrpnPlus; @Siamese_Cascaded_Region_Proposal_Networks_for_Real_Time_Visual_Tracking; @StructSiam; @Learning_dynamic] to perform similarity learning. Bertinetto *et al.* [@siamfc] first employ Siemese network to measure the similarity between the target and the search area with a tracking speeds of over 100 fps. SiamRPN [@siamrpn] formulates visual tracking as a local one-shot detection task in inference by introducing a Region Proposal Network to Siamese network. SiamRPN++ [@siamrpnPlus] improves SiamRPN by substituting the modified AlexNet [@alexnet] with Resnet-50 [@resnet50], which enables the backbone to extract abundant features.
Discriminative trackers aims at learning an adaptive filter by maximizing the response gap between target and background. Paticularly, the correlation-filter-based [@kcf; @eco] trackers and classifier-based trackers [@Learning_Multi_domain] are typical methods to online update the classification model so as to distinguish the target from background. However, these approaches rely on complicated online learning procedures that cannot be easily formulated in an end-to-end learning architecture. Bhat *et al.* [@dimp] and Park *et al.* [@meta-tracker] further learns to learn during tracking based on the meta-learning framework. DiMP [@dimp] introduces a target model predictor to online optimizing the target model instructed by the discriminative loss, which achieves leading performace in various benchmarks. In our work, we employ the target model predictor to perform online classification.
Target Regression
-----------------
Previous trackers can be divided into three categories based on the task of target regression. DCF [@dcf_] and SiamFC [@siamfc] employ brutal multi-scale test to estimate the target scale roughly. RPN-based trackers [@siamrpn; @siamrpnPlus] regress the location shift and size difference between pre-defined anchor boxes and target location. ATOM [@atom] and DiMP [@dimp] employ IoUNet to iteratively refine the inital multiple boxes. In this work, we take inspiration from FCOS [@fcos] to regress the distance from estimated target center to the sides of the bounding box, which is similar with Siamfc++ [@siamfc++]. However, our FCOT is different on several important aspects. First, our FCOT is essentially an discrminative tracker with a focus on enabling online optimization for both classification and regression branches, while Siamfc++ is a generative tracker with fixed kernels for both branches. In addition, to fully unleash the power of FCOT, we resort to higher resolution of feature map produce classification and regression results.
{width="12cm"}
\[fig:architecture\]
Our Method
==========
In order to obtain a simple, efficient, and precise tracking method, we design a fully convolutional online tracker (FCOT), guided by the following principles.
- **A simple and unified architecture.** We hope the components of feature extraction, classification branch, and regression branch could be implemented in a single and unified network architecture. A fully convolutional network is employed to locate the target center and regress the offsets from the target sides to the center directly, which can avoid designing hand-crafted box size estimation head such as hyper-parameters and IoU prediction in DiMP [@dimp] or anchor box placement and design in SiamRPN [@siamrpnPlus]. The unified fully convolutional scheme also enables the FCOT to be efficient for both training and inference.
- **Accurate regression and classification.** First, compared with previous trackers, FCOT generates larger score map and box offset maps, ensuring more precise target center location and target bounding box regression. In addition, due its simplicity of FCOT, it’s for the first time to online optimize the regression model implemented by our proposed Regression Model Generator with steepest descent methodology. In this way, FOCT can update the regression model online thereby regressing the bounding box accurately facing the issue of target appearance changing in the subsequent frames. For classification, we utilize an online target model generator introduced by DiMP [@dimp] to distinguish the target from the background.
As shown in Fig \[fig:architecture\], our FCOT comprises a Resnet-50 [@resnet50] backbone to extract general features, classification and regression heads to generate task specific features, online model generators for the two tasks to predict online models, a classification convolutional layer locating the target center and a regression convolutional layer estimating the offsets from the four sides to the target center. Thus, our discriminative tracker can integrate online updating target-specific information into classification and regression so as to predict bounding boxes accurately with such a simple FCN-based architecture.
Classification and Regression via Fully Convolutional Network
-------------------------------------------------------------
In general, current trackers can be divided into two categories, generative and discriminative trackers. With taking the background information into consideration, discriminative trackers such as ATOM [@atom] and DiMP [@dimp] achieve leading performance on various benchmarks. However, these approaches are two-stage and perform a complicated target regression procedure. Therefore, we introduce a simple fully convolutinal network for both classification and regression to overcome these issues.
We denote the training set as [$(I_{(i)},B_{i})_{i=1}^N$ ]{}, which is composed of a set of training frames $(I_{i})_{i=1}^N$ of length $N$ with its annotated bounding box $(B_{i})_{i=1}^N$. Specifically, the training images are selected from the tracked frames with the predicted bounding boxes in the online tracking phase.
Similarly, the test frame can be represented as $I_{t}$, whose expected bounding box $B_{t}=(x_{t}^{0}, x_{t}^{1}, y_{t}^{0}, y_{t}^{1})$. The goals of the network are locating the target center of the test frame $I_{t}$ and estimating the distance from the predicted center $\tilde{c_t}$ to the four sides $(x_{t}^{0}, x_{t}^{1}, y_{t}^{0}, y_{t}^{1})$ of the bounding box $B_{t}$.
![The structure of the Classification Heads and Regression Heads, where GN represents the Group Normalization. The classification heads in training branch and test branch share the same structure and weights. The regression heads in the two branches are different. The Regression Head-1 outputs 1024 feature maps to be used for generating four regression filters while the Regression Head-2 outputs 256 feature maps to be performed a regression convolution with the four filters.[]{data-label="fig:architecture_heads"}](architecture_heads.pdf){width="10cm"}
### Feature Extraction.
With the aforementioned inputs, the general features are extracted by an encoder-decoder backbone. The encoder covers from the Layer-1 to Layer-4 of Resnet-50. And the decoder contains a $1\times1$ convolution and 2 simple up-sample layers. The spatial down-sampling ratio of the general feature maps is 4. Then the Classification heads and the regression heads extract task-specific features to cope with classification and regression tasks separately. The structure of the classification heads and the regression heads can be seen in Fig \[fig:architecture\_heads\]. Specifically, the classification heads employed in training and test branch share the same network with common weights while the regression heads in the two branches are different, which performs well as the experiments discover.
The Regression Head-1 outputs 1024 feature maps to generate four regression filters while the Regression Head-2 outputs 256 feature maps to be performed a regression convolution with the four filters.
### Classification and Regression. {#cla_and_reg_chap}
In our FCOT, we formulate the tracking as a per-pixel prediction problem. We predict a target center confidence map $M_{cls}$ and four offsets maps $M_{reg}$ via classification and regression branch, which is defined as: $$\begin{split}
M_{cls} = \phi(I_{t}) * f_{cls}, \\
M_{reg} = \theta(I_{t}) * f_{reg}.
\end{split}$$ The parameter $\phi$ is the feature extractor of classification branch, $\theta$ is the feature extractor of regression branch, $f_{cls}$ and $f_{reg}$ represent the filters generated by the corresponding model generators and $*$ denotes a convolution operation.
For each location $(x,y)$ on the final feature maps, we can map it back onto the input image as $(\lfloor \frac{s}{2} + xs \rfloor, \lfloor \frac{s}{2} + ys \rfloor)$, where $s$ is the stride of the feature extractor(In this work, $s=4$). For classification, $M_{cls}(x,y)$ denotes the confidence score of the pixel being a target center. During training, the classification target is a Gaussian function map centered at $c_{t}$. For regression, $M_{reg}(x,y)$ is expected to a 4D vector $(l^{*},r^{*},t^{*},b^{*})$ representing the distance from $(x,y)$ to the sides of the bounding box in the final feature maps. Hence, the regression targets of position $(x,y)$ can be formulated as follows: $$\begin{split}
l^{*} = \lfloor \frac{s}{2} + xs \rfloor - x_{0}, \ \ \ r^{*} = x_{1} - \lfloor \frac{s}{2} + xs \rfloor,\\
t^{*} = \lfloor \frac{s}{2} + ys \rfloor - y_{0}, \ \ \ b^{*} = y_{1} - \lfloor \frac{s}{2} + ys \rfloor.
\end{split}$$ We regress for positions in the vicinity of the target center $c_t$ (the area with a radius of 2 in this work) rather than for the only pixel $c_t$. Experimental results demonstrate its effectiveness in section \[training\_ablation\].
{width="9cm"}
\[fig:estimation\_subnet\]
Regression Model Generator
--------------------------
Inspired by DiMP [@dimp], we present a regression model generator to online optimize the target regression model for the first time, which can alleviate the impact of target appearance changing on box regression. As shown in Fig \[fig:estimation\_subnet\], the regression model contains a model initializer and a model optimizer. The model initializer takes the regression features and bounding box of the first frame as input and generate the initial model which is a regression convolution filter. The features of the training set and their corresponding bounding boxes are then input into the model optimizer to update the model iteratively. According to the design, our tracker can not only update the filter online to fit with the changing target appearance by the optimizer but reduce optimization steps to improve the tracking speed.
The structure of the model initializer is a single ROI-pooling layer with the size of $3\times3$. For improving the efficiency, the initializer performs ROI pooling only on the features of the first frame thus generating a rough model. And the model optimizer is derived from online regression training loss: $$L(f) = \frac{1}{N} \sum\limits_{(X,c) \in S_{train}}\left\| M_{reg}^{(c)}- X^{(c)} * f \right\|_2 + \left\| \lambda f \right\|_2.$$ The parameter $N$ is the length of the online training set $S_{train}$ which is composed of the tracked frames with high classification scores, $X$ is the features extracted by the Regression Head-1, $M_{reg}^{(c)}$ denotes the 4D distance vector of center position $c$ in the regression map $M_{reg}$ as described in \[cla\_and\_reg\_chap\], $X^{(c)}$ is a portion of $X$ with an area of $3\times3$(the same with the regression model size) centered at $c$, $f$ is the regression convolution filter, $*$ denotes convolution and $\lambda$ is a regularization factor. The objective is to optimize the regression convolution filter $f$. Since the gradient descent is slow, we solve the issue with the steepest descent methodology, which compute a step length $\alpha$ to update the model as follows: $$f^{(i+1)} = f^{i} - \alpha \nabla{L(f^{(i)})}.$$ The parameter $i$ denotes the number of iterations of optimizing. We compute $\alpha$ and $\nabla{L(f^{(i)})}$ according to the similar expression with DiMP [@dimp].
Offiline Training
-----------------
The parts to be trained offline include the encoder and decoder backbone, classification and regression heads, the parameter $\lambda$ in regression model generator and some parameters in classification model generator as in DiMP [@dimp]. Our offline training are performed at two stages, which it first trains the entire network except for the regression optimizer in the model generator, and then updates the regression optimizer with the rest of the network freezed. In this way, the training time is reduced largely since online optimizing the model is time-consuming and there is just one parameter $\lambda$ in the regression model generator to be trained. The total loss for offline training can be formulated as $L_{tot}=\alpha L_{cls} + \beta L_{reg}$, which $\alpha$ is 100 and $\beta$ is 0.1 in this work. For classification branch, we use the same loss $L_{cls}$ and training strategies as DiMP [@dimp]. We use an IoU loss [@iounet] for $L_{reg}$ to train the regression branch and perform regression for the points in the vicinity of the target center with a radius of 2. This strategy can improve the accuracy of the target box regression during tracking especially when the detected target center is deviated from the groundtruth center.
Online Tracking {#online_tracking}
---------------
Given the first frame with groundtruth bounding box, we construct a training set of size 15 with performing augmentation on the frame. The initial regression and classification model (convolutional filter) is generated by the model initializers. Then the initial models are optimized using the augmented training set. We present two simple strategies to online update classification and regression models. First, we add the frames with the highest classification score every 25 frames to the online training set so as to guarantee the quality of the training samples. Second, we merge the latest model $f_{lat}$ with the model $f_{1}$ optimized with the augmented training set with the first frame, so the current model $f_{cur}$ can be formulated as: $$f^{(cur)} = \lambda f_{1} + (1-\lambda) f_{lat}.$$ It turns out to boost the performance of the tracker.
Experiments
===========
Implement Details
-----------------
Our tracker FCOT is implemented with Pytorch based on the project Pytracking [@pytracking]. We use ADAM [@adam] with learning rate decay of 0.2 every 25 epochs. We train FCOT for 100 epochs and 5 epochs in the first and the second stage respectively. We spend 50 hours training the whole model offline by 8 RTX 2080ti GPUs. While for inference, the average tracking speed is 53 fps on a single RTX 2080ti GPU. The encoder in backbone is initializied with ImageNet weights. The training set we used to train FCOT including TrackingNet [@trackingnet], LaSOT [@lasot], GOT-10k [@got10k] and COCO [@coco] training dataset. The regression model is a convolution filter with size of $((4*256)\times3\times3)$ and the classification model of $(256\times4\times4)$.
Ablation Study
--------------
### Decoder Feature Layers.
[c|cc|cc|ccc]{}
------------------------------------------------------------------------
\*[No.]{} & & & \*[SR$_{0.5}$($\%$)]{} & \*[SR$_{0.75}$($\%$)]{} & \*[AO($\%$)]{}\
------------------------------------------------------------------------
& $U1$ & $U2$ & $U1$ & $U2$\
1 & & & & & 74.0 & 48.7 & 61.1\
2 & & & & & 74.6 & 50.4 & 61.9\
3 & & & & & 73.2 & 48.1 & 60.4\
4 & & & & & **75.3 & **51.7 & **62.7\
******
We evaluate the impact of using different feature blocks from the Decoder(in Table \[table:1\]). For regression, we can compare the feature layers settings of No.1, No.2 and No.4. Using features from only $U2$ performs better than only $U1$, which demonstrates that using features with higher resolution leading to more accurate tracking results. Fusing features from both the blocks leading to significant improvement, giving scores of 75.3%, 51.7% and 62.7% for the three metrics respectively. For Classification, we can compare the feature layers settings of No.3 and No.4. The setting of using both $U1$ and $U2$ is difficult to train and cannot boost the performance than using only the $U2$ so we overlook it in this paper. It can be derived that using features from $U2$ performs better than $U1$, with gains of 2.1%, 3.6% and 2.3% respectively in $SR_{0.5}$, $SR_{0.75}$ and AO.
### Online Model Generators.
[c|cc|cc|ccc]{}
------------------------------------------------------------------------
\*[No.]{} & & & \*[SR$_{0.5}$($\%$)]{} & \*[SR$_{0.75}$($\%$)]{} & \*[AO($\%$)]{}\
------------------------------------------------------------------------
& Cls & Reg & Cls & Reg\
1 & & & & & 70.9 & 48.7 & 59.3\
2 & & & & & 72.7 & 49.9 & 60.9\
3 & & & & & 71.6 & 49.2 & 59.8\
4 & & & & & 73.2 & 50.1 & 61.2\
5 & & & & & 74.1 & 50.7 & 61.8\
6 & & & & & 73.7 & 50.4 & 61.6\
7 & & & & & **75.3 & **51.7 & **62.7\
******
Here, we analyse the effect of our online model generators (Table \[table:2\]). **Optimization using the augmented training set with the first frame:** It can be derived from the optimization settings of No.1, No.2, NO.3 and No.4 that performing optimization of the classification and regression model in the first frame can both improve the tracking accuracy and success rate. **Optimization online:** Comparing the optimization settings of No.4, No.5, NO.6 and No.7, we find that online optimization of the two models can both boost the performance largely. Compared with the setups of No.1, the setups of No. 7 achieves large gains of 4.4%, 3.0% and 3.4% in terms of the three metrics respectively, which demonstrates the effectiveness of the regression model generator and classification model generator.
### Offline Training. {#training_ablation}
-------- --------------------------- ------------------- ----------
radius SR$_{0.5}$($\%$) SR$_{0.75}$($\%$) AO($\%$)
0 72.6 43.8 59.3
1 74.3 50.2 61.7
2 **75.3 & **51.7 & **62.7\
3 & 74.6 & 51.1 & 62.4\
******
-------- --------------------------- ------------------- ----------
: Ablation study of the size of the area where the points are performed regression during offline training phase. Radius 0 means the only target center point, while radius $x$ means a square of side length $2x+1$ around the target center point.[]{data-label="table:3"}
------------------- ----------------------------- --------------------------- ------------------- ----------
if merge filters? if select the best samples? SR$_{0.5}$($\%$) SR$_{0.75}$($\%$) AO($\%$)
73.8 51.0 61.7
74.5 50.9 62.1
74.1 51.0 61.9
**75.3 & **51.7 & **62.7\
******
------------------- ----------------------------- --------------------------- ------------------- ----------
: Ablation study of two online training strategies referred in section \[online\_tracking\]. **column 1:** whether to merge the optimized filter in the first frame with the online optimized filter. **column 2:** whether to select the tracked frames with highest classification confidence score every 25 frames.[]{data-label="table:4"}
Here, we analyze the impact of the regression area around the target center point, where the points are performed regression during offline training. As shown in Table \[table:3\], we find that it’s useful for imporving the robustness of the FCOT to regress for the points in the vicinity of the target center. And the best size of the area is $5\times5$ (the radius is 2).
### Online Training Strategy.
Here, we analyze the two training strategies referred in section \[online\_tracking\]. It can be seen from Table \[table:4\] that each of the strategy is effective for improving the tracking accuracy. And It’s obvious that FCOT acquire excellent results when the two strategies being jointly employed.
Comparison with the state of the art
------------------------------------
we test the proposed FCOT on six tracking benchmarks and compare our results with the state-of-the-art trackers.
### GOT10k
GOT10k [@got10k] is a large-scale dataset with over 10000 video segments and has 180 segments for the test set. Apart from generic classes of moving objects and motion patterns, the object classes in the train and test set are zero-overlapped. This feature makes the dataset focus one-shot tracking capability of the generic tracker.
------------------- ---------- -------- ----------- --------- --------------------------------------------------------------------------------- ------------- ------
MDNet ECO SiamFC ATOM DiMP SiamFC++ FCOT
[@mdnet] [@eco] [@siamfc] [@atom] [@dimp] [@siamfc++]
SR$_{0.5}$ ($\%$) 30.3 30.9 35.3 63.4 **[ 71.7]{} & 69.5 & **[ 75.3]{}\
SR$_{0.75}$ ($\%$) & 9.9 & 11.1 &9.8 & 40.2 & **[ 49.2]{} & 47.9 & **[ 51.7]{}\
AO ($\%$) & 29.9 & 31.6 & 34.8 & 55.6 & **[ 61.1]{} & 59.5 & **[ 62.7]{}\
************
------------------- ---------- -------- ----------- --------- --------------------------------------------------------------------------------- ------------- ------
: State-of-the-art comparison on the GOT-10k test set in terms of average overlap (AO), and success rates (SR) at overlap thresholds 0.5 and 0.75. []{data-label="table:5"}
------------ -------- ---------- ------------ ---------------- --------- ------------------------------------------------------------------------------- ------
ECO MDNet SiamRPN SiamRPN++ ATOM DiMP FCOT
[@eco] [@mdnet] [@siamrpn] [@siamrpnPlus] [@atom] [@dimp]
AUC ($\%$) 52.5 52.8 52.7 61.3 63.2 **[ 64.3]{} & **[ 65.7]{}\
Precision ($\%$) & 74.1 & - & 74.8 & 80.7 & 84.4 & **[ 84.9]{} & **[ 87.6]{}\
********
------------ -------- ---------- ------------ ---------------- --------- ------------------------------------------------------------------------------- ------
: state-of-the-art comparison on the UAV123 dataset in terms of success rate (AUC) and precision.[]{data-label="table:6"}
We show state-of-the-art comparison on Table \[table:5\]. DiMP achieves an average overlap(AO) score of 61.1%. Compared with DiMP, Our FCOT improves 1.6% in AO. Impressively, Our FCOT achieves gains of 3.6% and 2.5% in success rate of threshold 0.5 and 0.75 respectively, which demonstrates that FCOT has the ability to generate accurate bounding boxes.
### UAV123
UAV123 [@uav123] is a large dataset captured from low-altitude UAVs. Thus compared with other benchmarks, the targets tend to be farther from the camera in UAV123. This dataset has a total of over 110K frames and 123 video sequences. We compare our FCOT with previous approaches on this dataset. As shown in Table \[table:6\], DiMP achieves an AUC score of 64.3% and a precision score of 84.9%. Our FCOT outperforms the previous approaches reaching 65.7% in AUC score and 87.6% in precision score, respectively. This demonstrates the powerful capabilities of our method on tracking far and tiny objects.
### Need For Speed
Need For Speed [@nfs] contains a total of 380K frames in 100 videos captured with high frame rate cameras from real world scenarios. We evaluate our FCOT on the 30 FPS version of this dataset and compare with the recent approaches. The results are shown on Table \[table:7\]. Specifically, FCOT obtains an AUC score of 62.2% and precision score of 74.5%.
### TrackingNet
TrackingNet [@trackingnet] provides over 30K videos with more than 14 million dense bounding box annotations. We validate FCOT on its test set, which consists of 511 videos, with an average of 441 frames per sequence. The results generated by our tracker are submitted to the evaluation server. Then three metrics success, precision and normalized precision are caculated and feedbacked. As shown in Table \[table:8\], our FCOT reaches a precision score of 71.4% and a normalized score of 81.7% surpassing DiMP and SiamFC++. We also achieve a success score of 74.5%, which is competitive with the state-of-the-art method.
------------ --------- -------- ---------- --------- --------- -------------------------------------------------------------------- ------
CCOT ECO MDNet UPDT ATOM DiMP FCOT
[@ccot] [@eco] [@mdnet] [@updt] [@atom] [@dimp]
AUC ($\%$) 48.8 46.6 42.2 53.7 58.0 **[61.5]{} & **[62.2]{}\
Precision ($\%$) & - & - & - & - & 70.0 & **[74.1]{} & **[74.5]{}\
********
------------ --------- -------- ---------- --------- --------- -------------------------------------------------------------------- ------
: State-of-the-art comparison on the NFS [@nfs] dataset in terms of success rate (AUC) and precision.[]{data-label="table:7"}
------------------ -------- ---------- --------- ---------------- --------- --------------------------------------------------------------------------------- ------
ECO MDNet ATOM SiamRPN++ DiMP SiamFC++ FCOT
[@eco] [@mdnet] [@atom] [@siamrpnPlus] [@dimp] [@siamfc++]
Precision ($\%$) 49.2 56.5 64.8 69.4 68.7 **[70.5]{} & **[ 71.4]{}\
Norm. Prec. ($\%$)&61.8 & 70.5& 77.1 & 80.0 & **[ 80.1]{} & 80.0 & **[ 81.7]{}\
Success ($\%$)&55.4 & 60.6 & 70.3 & 73.3 & 74.0 & **[ 75.4]{} & **[ 74.5]{}\
************
------------------ -------- ---------- --------- ---------------- --------- --------------------------------------------------------------------------------- ------
: State-of-the-art comparison on the TrackingNet [@trackingnet] dataset in terms of success rate, precision and normalized precision.[]{data-label="table:8"}
### LaSOT
LaSOT [@lasot] has 280 videos in its test set. With an average of 2500 frames, sequences of LaSOT are longer than other dataset, which poses great challenges to trackers. We evaluate our FCOT on the test set to validate its long-term capability. The results are shown on Fig \[fig:lasot\]. FCOT achieves a success score of 55.4%, which is better than other methods except for DiMP. In normalized precision, we reach the best performance of 65.7%. Our method outperforms DiMP when overlap threshold is above 0.5, which means our online regression branch predicts more accurate bounding boxes. The overall gap between DiMP and FCOT comes from coarse-grained tracking. As also confirmed in normalized precision plots, our FCOT does better in accurate localization , while DiMP finds more coarse-grained predictions.
### OTB-100
![State-of-the-art comparison on the LaSOT dataset in terms of success rate (right) and precision (left).[]{data-label="fig:lasot"}](lasot.pdf){width="10cm"}
![State-of-the-art comparison on the OTB-100 dataset in terms of success rate (left) and precision (right).[]{data-label="fig:otb100"}](otb100.pdf){width="10cm"}
We evaluate our FCOT on the 100 videos of OTB-100 [@otb] dataset. Fig \[fig:otb100\] reports our overall performance of FCOT on OTB-100. We achieve 69.2% and 90.6% in success score and precision score, which outperforms DiMP with relative gains of 0.8% and 1.2% in two metrics.
Conclusions
===========
In this paper, we have presented a fully convolutional online tracker (FCOT), by unifying the components of feature extraction, classification head, and regression head into a encoder-decoder architecture. Our key contribution is to design an online anchor-free box size regression branch to directly estimate bbox sizes. This new design enables the whole tracking framework to be performed in a simple fully convolutional manner, and also allows for online optimization regression branch to well handle target deformation. Extensive experiments on several benchmarks demonstrate the high precision of our proposed anchor-free and online regression branch. Our FCOT outperforms the state-of-the-art trackers on most benchmarks at a high speed of 53 FPS.
Appendix A: Visualization on OTB100 {#section_vis .unnumbered}
===================================
We provide visualization examples generated by FCOT and DiMP [@dimp] on OTB100 dataset [@otb] in the attached folder named *visualization* which is available at <https://github.com/MCG-NJU/FCOT/tree/master/visualization>. These sequences suffer from the limitations including occlusion, scale variation, deformation, motion blur and so on. It can be seen that our tracker performs well on these sequences. Particularly, the bounding boxes are more precise than DiMP [@dimp] once the objects have been roughly located by the classification branch.
Furthermore, we find that the labels of OTB100 dataset are not accurate enough. It can be seen from Fig. \[fig:otb\_anno\] that the ground-truth bounding box(colored in green) are not accurate and the definitions of objects are ambiguous leading to the performance degradation of our FCOT.
![Visualization of the ground-truth bounding boxes and the tracking results of FCOT on OTB100 dataset [@otb]. The labels of OTB100 dataset are not accurate enough, which may degrade the performance of FCOT.[]{data-label="fig:otb_anno"}](otb_wrong_anno.pdf){width="12cm"}
Appendix B: Score Map Comparison {#section_hm .unnumbered}
================================
In this section, we compare the classification score maps generated by DiMP [@dimp] and FCOT. We can derive from Fig. \[fig:p2\] that the score map of FCOT is more precise than DiMP, which ensures the tracking accuracy of our tracker. It demonstrates the effectiveness of our designed encoder-decoder architecture.
![Visualization of score maps generated by DiMP [@dimp] and FCOT. The initial size of score maps of DiMP and FCOT are 18 and 72 respectively. We zoom the score maps to the same size in this visualization result.[]{data-label="fig:p2"}](p2.pdf){width="12cm"}
[^1]: Corresponding Author.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We propose an efficient scheme for generating fake network traffic to disguise the real event notification in the presence of a global eavesdropper, which is especially relevant for the quality of service in delay-intolerant applications monitoring rare and spatially sparse events, and deployed as large wireless sensor networks with single data collector. The efficiency of the scheme that provides statistical source anonymity is achieved by partitioning network nodes randomly into several node groups. Members of the same group collectively emulate both temporal and spatial distribution of the event. Under such dummy-traffic framework of the source anonymity protection, we aim to better model the global eavesdropper, especially her way of using statistical tests to detect the real event, and to present the quality of the location protection as relative to the adversary’s strength. In addition, our approach aims to reduce the per-event work spent to generate the fake traffic while, most importantly, providing a guaranteed latency in reporting the event. The latency is controlled by decoupling the routing from the fake-traffic schedule. We believe that the proposed source anonymity protection strategy, and the quality evaluation framework, are well justified by the abundance of the applications that monitor a rare event with known temporal statistics, and uniform spatial distribution.'
author:
- |
Silvija Kokalj-Filipović, Fabrice Le Fessant, and Predrag Spasojević,\
INRIA Saclay, WINLAB Rutgers University,\
*{silvija.kokalj-filipovic,fabrice.le\_fessant}@inria.fr, [email protected]*
bibliography:
- 'EveBib.bib'
title: ' **The Quality of Source Location Protection in Globally Attacked Sensor Networks**'
---
Introduction {#sec:intro}
============
Privacy issues are an important aspect of monitoring applications in wireless sensor networks [*(WSNs)*]{}. A recent survey of state-of-the-art research on privacy protection in WSNs [@LiSurvey], among other problems, reviews strategies to protect the object observed by a WSN node, referred to as [*source*]{}, from the global eavesdropper [*(Eve)*]{} [@Mehta07], which can infer the location of the object based on the established location of the source.
Problem Description {#subsec:problem}
-------------------
The observed object may be a smuggler crossing the border, an important person entering a classified area, or endangered animals monitored in their habitats. Messages from the source are propagated in a traditional hop-by-hop manner, and directed to a fixed data collector, referred to as a base station, or a [*sink*]{}. Sink protection is not an issue as the adversary usually knows the sink, in fact she may even know the whole topology of the sensor network. In addition, Eve detects the timing and location of all transmissions in the network (hence [*global*]{}); she can hear and capture any packet sent in the network (either with a very powerful antenna or she has her own sensor network deployed in the area). Eve is powerful: she can employ complex statistical algorithms for detection, and arbitrary localization techniques. However, the message itself is encrypted and Eve does not know the encryption, hence she cannot capture the message and infer the object’s position from the content. Moreover, Eve, despite being so powerful and omnipresent, needs to stay invisible. We define [*outage*]{} as the event when, following an eavesdropped transmission, Eve reveals itself by taking actions based on the false suspicion that an event occurred. The actions may involve physical presence of the attacker or her faculties, in order to capture or destroy the object. Hence, a false-alarm presents a risk of personal exposure and liability.
The adversary gathers the source of the transmission based on the change in the traffic pattern; a conspicuous case would be when a node starts transmitting after a prolonged period of inactivity in the WSN. For many event-reporting applications, despite the fact that the attacker cannot learn the details from the message content, inferring the [*contextual*]{} information, i.e. whether, when and where a concerned event has happened, may be enough to jeopardize monitored resources. Intuitively, the persistent [*dummy (fake) traffic*]{} is the only way to obfuscate the events, and the formal proof for it is available in [@Mehta07]. Dummy packets follow a predefined schedule, aligned with the expected timeline of real packets, so that Eve cannot observe the change. To better explain the intricacies of this approach, especially in light of the existing research, we next introduce two models of monitored phenomena. Let us first define the application delay as the delay in event reporting. In both scenarios, the duration of time is relative to the application delay constraint, which is a known value $\Delta$. We assume that the WSN is divided into cells, such that each sensor node monitors a unique cell and that the events are occurring in a uniform manner over time and space. An active event is any event that is not reported to the sink yet.
### Scenario $\mathcal{A}$: Frequent and Dense Events
This model describes monitoring of a physical phenomena that creates on average one event per cell over one [*cycle*]{}, a relatively short period of time, whose duration is larger but of the same order of magnitude as the application latency constraint. In other words, in any cycle, there are many active events in the network.
### Scenario $\mathcal{B}$: Rare and Spatially Sparse Events
In this model, the events are rare and isolated. For example, if the allowed application delay is in minutes, the expected interval between events is measured in hours or days. They are spatially sparse: we assume that there is at most [*one active event*]{} at a time. In fact, the examples of monitoring applications at the beginning of \[subsec:problem\] all represent the scenario $\mathcal{B}.$ No single node can statistically emulate the spatial and temporal characteristics of the events in this model. In addition, by observing a single node for the duration several orders of magnitude larger than $\Delta,$ Eve can not reliably deduce deviation from the expected behavior. Consequently, Eve attempts to observe abnormalities in the network-wide traffic pattern. The anonymity protection scheme described here implements the traffic pattern in a decentralized manner, so that the occurrence of real events does not cause observable abnormalities.
Solution Outline
----------------
The uniform spatial/time distribution of events guides naturally the [*baseline scenario*]{} for the dummy traffic: all cells in the network send dummy messages at a constant rate regardless of whether a real event has occurred or not. That means that an event would have to wait to be reported on average for half of the inter-transmission interval. However, since the traffic in the network always keeps the same pattern, it effectively defeats any traffic analysis techniques. The main problem with dummy traffic is immediately obvious from the basic scenario: limiting the reporting delay calls for a high-rate fake traffic, which is not only expensive but may quickly burn out the network.
Our approach stipulates the importance of knowing the event’s temporal dynamics, but only in terms of the two introduced scenarios; it allows us to design energy-efficient protection strategies. It is natural to assume that the expected frequency of events will be known to both the attacker and the network architect, given that we design the monitoring application for a particular physical phenomenon (an important person does not enter a classified area every single minute, or even hour), and that the easiest characterization of a random process is through its first moment, or the moment’s estimate. Next, among the all-positive probability distributions with a given expected value, exponential distribution has the highest entropy. Hence, assuming that inter-event times follow an exponential distribution of the estimated mean leads to a good and justifiable model of the event’s randomness. In terms of traffic overhead/ energy-consumption and interference, the optimal design would force each node to transmit as rarely as possible, and that would be in exponentially distributed intervals of the expected duration [*exactly*]{} equal to the expected time between real events; smaller intervals generate more traffic and therefore cost more, while larger ones create too few opportunities for embedding the real-traffic, especially under delay constraints.
The second major underpinning of our approach is the network-centric view of the problem. The following aspects of the problem are looked at from both the event’s and the network’s perspective:
- Event is characterized as a spatio-temporal process over the whole network area,
- Event-reporting delay includes routing latency,
- Fake-traffic shaping is a decentralized process, collaboratively maintained by all nodes
- Energy consumption per-event of the protection strategy is equally split among network nodes, and substantially decreased with respect to the baseline strategy due to nodes’ collaboration.
Finally, our source-anonymity protection scheme aims to achieve statistical event unobservability. The absolute protection under baseline strategy is not applicable to delay-sensitive applications, such is the majority of event monitoring in WSNs. Secondly, we do not adjust the timing of real events as in [@ShaoINFOCOM08], to make the event pass the statistical test under the test parameters assumed to be used by Eve. Instead, we make the event pass the test with the same probability as the dummy transmissions, making it statistically indistinguishable.
Existing Research
-----------------
There are several papers that study the WSN source anonymity but, in our opinion, propose solutions that are efficient under scenario $\mathcal{A}$ only[@Mehta07; @YangWiSec08; @ShaoINFOCOM08]. At the expense of substantial traffic overhead, a practical tradeoff between security and latency is proposed in [@ShaoINFOCOM08]. This approach, as ours, assumes that the attacker knows the defense strategy of the WSN, and that she will use a state-of-the-art statistical test to distinguish the real event from the fake once. The rationale is that fake traffic, even if random, is designed to follow a distribution, while the real events may not. According to [@ShaoINFOCOM08], the event sources should run the same test, and adjust the time of the real event to pass the test. The event sources test and correct the intervals between [*their own*]{} transmissions of fake and real events, striving to maintain their exponentiality. The paper compares two varieties of this approach: one in which the real event embeds itself by waiting until next scheduled transmission (ProbRate), and another one (FitProbRate), when it waits as little as needed to pass the goodness-of-fit test for the exponential distribution, inferred from the previous transmissions. The latter approach results in smaller delay, but requires the correction of the schedule, quite likely for all the post-event transmissions.
We further observe that the scheme proposed in [@ShaoINFOCOM08] defines delay as the time between the event occurrence and the source’s transmission, which [*holds only*]{} for WSN applications in which the sink is one hop away from any source. If the packet is delivered to the sink in a hop-by-hop manner, the latency includes another random part due to summation of the exponentially distributed delays associated with such transmission schedule of each relay. We refer to this additional delay as the [*publishing route (PR) latency*]{}. When the expected value $\lambda$ of the inter-transmission times is the same at each node, as in [@ShaoINFOCOM08], and designed to imitate a relatively rare event pattern, for the source-sink route of $h$ hops, the PR latency becomes an Erlang-distributed random variable with mean $h\lambda.$ From this point of view too, Scenario $\mathcal{B}$ requires a modified approach to source anonymity. For the cases simulated in [@ShaoINFOCOM08] , the mean of dummy message intervals is $20s,$ and real events arrive according to a Poisson process with the rate changing from $1/20$ to $1/100.$ Their protection scheme achieves the average latency of less than 1s. If we replace seconds with hours, having in mind events that happen once a day, or once a couple of days, the delay of one hour does not seem to be acceptable. Additionally, the PR latency for rare events is prohibitive, even for applications that are not delay-sensitive. To decrease the overhead of the dummy protection scheme in [@ShaoINFOCOM08], in [@YangWiSec08] the same authors propose a WSN with several proxy nodes, which pick up transmissions from surrounding nodes, and filter out the dummy packets. Apart from requiring mitigating solutions, frequent dummy traffic inevitably leads to interference, which additionally increases the PR latency.
The next section briefly introduces our solution to this problem. Section \[sec:Decentral\] describes and analyzes two decentralized algorithms that implement the proposed solution: one is suitable for uniform spatio-temporal distribution of events, while another is more resilient to the distribution outliers (spatial clusters, and temporal bursts of events), however, more expensive in terms of overhead. Section \[sec:BurstOutage\] motivates and presents some of the simulations, while some are left out due to space constraints. Finally, in \[sec:Conclude\], we conclude.
Our Approach {#sec:Appr}
============
System Model {#subsec:Model}
------------
We have a static WSN of $n$ nodes. There is one static sink collecting event notifications from all nodes. The monitoring application is delay sensitive: the time between the event occurrence and the sink’s notification must be smaller than $\Delta$. We assume that monitored events have Poisson temporal distribution of a known rate $\lambda=1/\mu$, and uniform spatial distribution over the area of network deployment. Hence, the time between the events is distributed according to a exponential distribution $\zeta_{\mu}$ (of expected value $\mu>>\Delta$). The source is assumed to transmit a burst of packets, all describing the event. Due to space limitations, we here analyze the burst of unit length (one packet), although our second algorithm is designed for exponential distributions perturbed by both random outliers and event bursts.
Problem Formulation {#subsec:Formul}
-------------------
We established that the only way to confuse Eve is through persistent network-wide transmissions. Simultaneously, as hop-by-hop is prevalent data transfer model in WSNs, and distance to a sink may be considerable, to satisfy the application latency constraints, we need to decouple routing from the fake traffic schedule by allowing immediate relaying of event notifications as opposed to piggybacking them on the existing fake transmissions. However, as such a route may be backtraced to the source, a similar routing path should be emulated from each fake source (see Figure \[fig:spatial\]). Given the mentioned constraints, our goal is to achieve statistically strong source anonymity through methods that optimize energy and delay [@Mehta07]. We hereby propose a pattern of fake traffic that scales well with network size, and satisfies application-latency constraints, while protecting the source location up to a given significance level, defined under the strong statistical tests available to Eve [@AndersonDarling; @WaldSeqAnal]. To confuse Eve, we propose to replicate the spatio-temporal process through the following mechanisms:
- [**Source Emulation:**]{} a subset of $d$ nodes regularly wakes up to act as dummy sources. As a result, any real event is [*covered*]{} by $d$ dummy sources, which we refer to as [*dummy population*]{}. To explain what it means for an event to be covered, we introduce a time interval, dubbed [*round*]{}, whose duration is equal to the expected inter-event time. Hence, the length of a round is $\mu.$ Covering an event implies the expected existence of fake transmissions in the same round in which the event occurs. To engage all nodes equally, we may divide the network in $d$ groups and assign one representative of the group to a distinct round. Each group will maintain a schedule that emulates the event distribution $\zeta_{\mu}.$ Due to the size of the dummy population, the probability distribution of the intervals between any two consecutive dummy events is $\zeta_{\mu/d}$ (exponential of expected value $\mu/d$). Such cooperative shaping of the fake traffic in order to emulate a sufficiently dense Poisson distribution is amenable to distributed implementation, which is thoroughly explained in Section \[sec:Decentral\]. However, as the attacker overhears every transmission that occurs in the network, and may integrate all recorded temporal data into one global network activity timeline, it is judicious to ignore for a moment the decentralized implementation. Instead, we look at the global timeline as if it was produced by a single source sampling the values of event inter-arrivals from the distribution $\zeta_{\mu/d}.$ The joint empirical distribution of inter-transmission times, created by extending the fake schedule with the immediate (undelayed) transmissions of real events, is, based on the transmissions’ independence, $\zeta_{\mu/(d+1)},$ i.e. exponential with the expected value $\mu/(d+1).$ For sufficiently large $d,$ $\zeta_{\mu/(d+1)}$ does not diverge perceptibly from the distribution on the global timeline of fake events $\zeta_{\mu/d)}$.
- [**Route Emulation:**]{} each source (dummy or real) forwards the packet along a predetermined route towards the sink (see Figure \[fig:spatial\]). The inter-transmission time between relays is constant and significantly shorter than $\Delta$ (and, consequently, orders of magnitude smaller than $\mu,$ as opposed to [@ShaoINFOCOM08] where it is tied to the inter-transmission time of dummy sources). As the real source starts transmitting without delay, the application latency is equal to the routing delay, which is now decoupled from the time dynamics of the fake traffic, and can be further optimized by minimizing the number of hops.
- [**Knowing the Attacker’s Detection Methods:**]{} Eve is assumed to be able to estimate the distribution of recorded transmission times. The estimated distribution is used as a reference point in the real-event detection strategy that involves a statistical test. We base our analysis on the assumption of a single statistical test, namely [*Anderson-Darling (A-D)*]{} test for exponentiality [@AndersonDarling]. Apart from the motives of simplicity, and the existence of a reference that employs the same test [@ShaoINFOCOM08], the additional arguments can be found in [@StephensEDF]. The A-D test belongs to the class of [*goodness-of-fit*]{} tests that evaluate the distance between the distribution of the sample data and a specified probability distribution. Alternatively, the test evaluates if the current sample comes from the same distribution as the previously evaluated ones. If the distance is statistically significant, where the significance level is derived from a parameter of the test, which also defines the percentage FA of [*false alarms*]{}, it is decided that data do not follow this distribution.
As the latency issue is decoupled from the fake traffic design, we seek to determine the minimal dummy population size needed to secure a given statistical anonymity, hence minimizing the overhead. For a WSN of size $n,$ we define $W_n$ as per-event and per-round energy consumption of the source anonymity mechanism. We express $W_n$ in terms of the number of packet-forwarding hops, where we upper-bound the length of the publishing route by $h=O{\left(\sqrt{\frac{n}{\log{n}}}\right)}$ [@GuptaKumarPower]. Hence, the cost of each fake source transmission will be of the same order. With $d$ fake sources covering each real event, $W_n=O{\left((d+1)\sqrt{\frac{n}{\log{n}}}\right)},$ which demonstrates the importance of optimizing $d$, as the source anonymity calls for a sufficiently large $d.$ However, note that $d$ would have to be on the order of $\sqrt{n\log{n}}$ to exert the same overhead as the method in [@ShaoINFOCOM08].
{width="5.0in"}
\[fig:spatial\]
Decentralized Generation of Fake Traffic {#sec:Decentral}
=========================================
Baseline Decentralized Algorithm
---------------------------------
In Section \[sec:Appr\] we highlighted the importance of cooperative and distributed shaping of the fake traffic that should result in sufficiently dense Poisson distribution of dummy transmissions. Let us now propose the realization of such a decentralized system. We first establish the baseline for the decentralized implementation. As the dummy population covering each event needs to include on average $d$ fake sources, we define en epoch of duration $T=\mu\frac{n}{d}$ in which each network node will get to be a dummy once. Then, we let each node draw a time instant to transmit in this epoch by sampling uniform distribution $U(0, T).$ The causality of transmissions will arrange all node samples in increasing order, resulting in exponential distribution $\zeta_{\mu/d}$ of inter-transmission times. The procedure can be extended to the consecutive epochs, so that in the $i$th epoch nodes draw their transmission times from $U((i-1)T, iT).$ Note that the particular uniform distribution range does not overlap with the ranges of distributions pertaining to other epochs. The collective empirical distribution of transmission times is the distribution of [*almost*]{} independent disjoint events, and therefore it approximates the Poissonian distribution. The independence of transmission events is broken only on the boundary of the epochs, as with each new epoch the nodes sample from a uniform distribution of different disjoint range. Hence, the distribution of the interval $Z$ between the first event in the new epoch and the last event in the previous epoch is not exponential. $$\begin{aligned}
\nonumber Z &= T+U-V\\
&= T + \min{\left\{X_1,\cdots,X_n\right\}}-\max{\left\{X_1,\cdots,X_n\right\}}.{\label{eqn:firstz}}\end{aligned}$$ $$\begin{aligned}
\nonumber F_U(u) &= {\mathcal{P}\left(U\leq u\right)}= 1-{\mathcal{P}\left(U>u\right)}\\
\nonumber &= 1-\prod^n_{i=1}{{\mathcal{P}\left(X_i>u\right)}}\\
\nonumber &= 1-\prod^n_{i=1}{{\left(1-{\mathcal{P}\left(X_i\leq u\right)}\right)}}\\
&= 1-\prod^n_{i=1}{{\left(1-F_X(u)\right)}}=1-{\left(1-F_X(u)\right)}^n.\end{aligned}$$ The probability distribution for $U$ is $$\begin{aligned}
\nonumber f_U(u) &=& n{\left(1-F_X(u)\right)}^{n-1}f_X(u)\\
&=&\frac{d}{\mu}{\left(1-\frac{u}{T}\right)}^{n-1},\ \ \mbox{for $T \geq u\geq 0,$}\end{aligned}$$ and o.w. $0,$ which is for sufficiently large $n$ clearly exponential distribution of expected value $\frac{\mu}{d}$ $$\begin{aligned}
f_U(u) &\approx& \frac{1}{\frac{\mu}{d}}{{e}^{-\frac{u}{\frac{\mu}{d}}}}.\end{aligned}$$ Further, $$\begin{aligned}
\nonumber F_V(v) &=& {\mathcal{P}\left(V\leq v\right)}= \prod^n_{i=1}{{\mathcal{P}\left(X_i\leq v\right)}}\\
&=& \prod^n_{i=1}{F_X(v)}=F^n_X(v).{\label{eqn:cumV}}\end{aligned}$$ $$\begin{aligned}
f_V(v) &=& nF^{n-1}_X(v)f_X(v)=\frac{d}{\mu}{\left(\frac{v}{T}\right)}^{n-1},\end{aligned}$$ for $T \geq v\geq 0.$ For sufficiently large $n$ $$\begin{aligned}
f_V(v) &=\frac{d}{\mu}{{e}^{-\frac{T-v}{\frac{\mu}{d}}}}.\end{aligned}$$ For $W=T-V,$ $f_W(w) =\frac{d}{\mu}{{e}^{-\frac{w}{\frac{\mu}{d}}}},$ where now $$\begin{aligned}
Z&=U+W. {\label{eqn:secondz}}\end{aligned}$$ The random variable $Z$ has a range ${\left[0,2T\right]},$ and, from [(\[eqn:secondz\])]{}, its probability distribution is Erlang, with the shape parameter of $2,$ denoted $\epsilon_{(2,\frac{\mu}{d})}$: $$\begin{aligned}
\nonumber f_Z(z) &= \int^{T}_{0}{f_U(\tau)f_W(z-\tau)d\tau}\\
&= \int^{z}_{0}{\frac{d}{\mu}{{e}^{-\frac{\tau}{\frac{\mu}{d}}}}\frac{d}{\mu}{{e}^{-\frac{z-\tau}{\frac{\mu}{d}}}}d\tau}= z\frac{d^2}{\mu^2}{{e}^{-\frac{z}{\frac{\mu}{d}}}}.\end{aligned}$$ As the probability of an inter-epoch sample in the collection of test samples of size $t,$ where $t<n,$ is $\frac{1}{t}$ or $0,$ the average test-failure probability will be at most $\frac{1}{t}FA_{E}+\frac{t-1}{t}FA,$ where $FA_{E}$ denotes failure probability when Erlang-distributed samples are tested on exponentiality. Even though the distribution of $Z$ differs from $\zeta_{\mu/d},$ we observe that the baseline decentralized implementation will pass the exponentiality test for any reasonable size of $t$ (no estimate is made based on 3 samples). Such a distribution satisfies our needs for real-event obfuscation. However, for other reasons, related to cases when the outliers of the real-event temporal distribution coincide with the spatial correlation of events, we propose the following realization of the dummy traffic, dubbed [*group implementation*]{}.
Group Algorithm
----------------
In the initialization, the WSN nodes are divided into $d$ groups of size $\left\lfloor n/d \right\rfloor$, and every node in the group is assigned an index $i$, denoting the round in which the node will cover the source. Hence, the $k$th round, where $k\equiv i\ \mbox{\cal{mod}}\left\lfloor n/d \right\rfloor,\ i\in{\left\{1,\left\lfloor n/d \right\rfloor\right\}},$ will have a dummy population of $d$ nodes belonging to different groups. A group schedule is created by letting the $i$th member select the specific time instant to transmit a fake message, sampled from the uniform distribution $U((i-1)\mu, i\mu)$. Such an algorithm is amenable to distributed implementation, since each node can independently measure time and keep count of the current round. Once the round index corresponds to node’s index modulo group size, the node draws a sample from the pertaining distribution, and determines its transmission time. As the independence of transmission events is broken only on the boundary of the rounds, the distribution of the interval $Z$ between the first event in the new round and the last event in the previous round is not exponential. Once again, $Z = U+W,$ where $W=\mu -V,$ $V=\max{\left\{X_1,\cdots,X_d\right\}},$ and $U=\min{\left\{X_1,\cdots,X_d\right\}}.$ $$\begin{aligned}
F_U(u) &=1-{\left(1-F_X(u)\right)}^d,\ \ \mbox{and}\end{aligned}$$ $$\begin{aligned}
f_U(u) &=\frac{d}{\mu}{\left(1-\frac{u}{\mu}\right)}^{d-1},\ \ \mbox{for $\mu \geq u\geq 0,$}\end{aligned}$$ and o.w. $0,$ which is, for large enough $d,$ exponential distribution of expected value $\frac{\mu}{d}.$ Now, following a derivation similar to [(\[eqn:cumV\])]{}, we obtain $$\begin{aligned}
f_V(v)&=\frac{d}{\mu}{\left(\frac{v}{\mu}\right)}^{d-1},\end{aligned}$$ for $\mu \geq v\geq 0.$ For large $d,$ the distribution of $W$ is $\zeta_{\mu/d}.$ The random variable $Z$ has a range ${\left[0,2\mu\right]},$ and the probability distribution is $$\begin{aligned}
f_Z(z) &= z\frac{d^2}{\mu^2}{{e}^{-\frac{z}{\frac{\mu}{d}}}}=\epsilon_{(2,\frac{\mu}{d})},\end{aligned}$$ for sufficiently large $d.$ Hence, with group implementation $Z$ follows the same Erlang distribution $\epsilon_{(2,\frac{\mu}{d})}$ as in the baseline decentralized implementation. However, to be able to state that the test sample of size $t<n,$ which includes inter-round intervals with probability $1/d,$ will be statistically indistinguishable from the sample of exponentially distributed intervals, we need to impose a stricter requirement for the value of $d.$ The average test-failure probability will be $\frac{1}{d}FA_{E}+\frac{d-1}{d}FA.$ Upper-bounding the failure probability $FA_{E}$ (when Erlang samples are tested on exponentiality) with one, we obtain that $d$ should be at least $\frac{1}{FA}$. If the spatially-uniform events do follow the distribution $\zeta_{\mu}$ in time, this may unnecessarily increase the per-event overhead with respect to the baseline. However, if we have a more complex event distribution, by selecting the group algorithm we are able to not only render the event’s temporal characteristics indistinguishable, but also to obfuscate spatial correlation.
![[**(A)** ]{}Illustration of our group algorithm for fake-traffic generation, where we randomly form $d$ groups of nodes to cooperatively create the schedule by sampling a series of uniform distributions. [**(B)** ]{}Large enough $d$ renders small divergence from the Poisson distribution statistically irrelevant. Bottom axis: for sufficiently large $d,$ mixing several “group” schedules and the timeline of the real events (red stars) into a global transmission schedule observed by Eve, statistically indistinguishable from Poisson schedule.](unifpoisson1.eps){width="3.5in"}
\[fig:fitting\]
Simulations {#sec:BurstOutage}
===========
When designing the simulations, we dismiss the possibility that Eve would test the schedule of any single node, since, in our scenario, inter-transmission times per-node are large with respect to $\mu$ (inter-event times), and, hence, it takes a lot of time to record a reasonable test sample. Our primary goal was to demonstrate the influence of the dummy population size $d$ to the statistical properties of the network-wide transmission schedule, both in the absence of real events, and under different stochastic models for real events. For sufficiently large $d,$ which is still much smaller than $n,$ our simulations show that the insertion of events does not statistically change the time axis. Therefore, by running the statistical tests, Eve does not obtain any additional information that would help her capture the monitored object, even if the time of transmission of a real source is not delayed. With the existing work [@ShaoINFOCOM08], an adjustment delay is added, and another mechanism may be needed to fix the sample mean affected by the adjustments, to delude Eve’s sequential analysis tests, such as SPRT [@WaldSeqAnal].
Performing the A-D test, which is a powerful test for exponentiality [@ADPoisson; @GFitExpStephens], on the samples drawn from an [*exponential distribution*]{} results in a percentage of failures, which represent false alarms. The percentage of false alarms is a random variable whose mean corresponds to the false-alarm parameter of the test (also referred to as the [*significance level*]{}), denoted by $\alpha$. Due to randomness, over certain sets of test samples this percentage will fluctuate around the value of the parameter provided by the test. Our testing strategy monitors the rate of test failures to evaluate if it behaves as expected for the exponential distribution.
[c]{}\
[**(A) - FA has constant mean**]{}\
\
[**(B) - FA has constant mean**]{}\
\
[**(C) - constant mean only for d=100**]{}\
\
[**(D) - constant mean only for d=100**]{}
\[fig:samplesize\]
If $d$ is not sufficiently large, goodness-of-fit tests of exponentiality [@AndersonDarling] will fail in a much larger percentage of cases than the value of the test’s FA parameter. For small $d,$ the test results are also sensitive to the sample size as shown in Figure \[fig:samplesize\] which compares the pure exponential samples whose false alarm statistics are independent of the sample size (pane [**(A)**]{}), with the samples obtained using our group algorithm for $d=10$ and $d=100.$ Samples that span just one round will most likely pass the test since ordered uniform samples in any range ${\left[(i-1)\mu, i\mu\right]}$ produce intervals described by the exponential distribution $\zeta_{\mu/d}$ (Figure \[fig:samplesize\] [**(B)**]{}). Large number of samples would include many rounds, with $1/d$ portion of samples not belonging to the exponential distribution. Note that for sample size $d=100$ our algorithm achieves close approximation of the exponential distribution as it exhibits the constant FA mean across the rounds, as opposed to $d=10$ (panes [**(C)**]{} and [**(D)**]{}). In pane [**(D)**]{}, at each new round we test 200 preceding samples. The same sample size was used to test the influence the inserted real events had on the percentage of test failures. As expected, our simulations confirm that the real events are statistically imperceptible for $d>10.$
Conclusion {#sec:Conclude}
==========
The proposed decentralized implementations of the fake traffic provide desired statistical source anonymity with minimal overhead and a delay that depends only on the efficiency of packet routing. Simultaneously, they utilize the network resources in a balanced and fair way, and provide flexibility necessary to handle different temporal and spatial profiles of the event process(omitted here due to space constraints). By designing only one parameter, the size of the dummy population $d,$ according to the known statistical characteristic of the observed process we achieve such flexibility. The minimal value of $d$ depends on the implementation, as the deployment of group implementation requires $d$ to be at least $\frac{1}{FA}.$ Uniform spatial distribution of events does not call for the group implementation, and this constraint on $d$ does not hold. However, $d$ needs to be large enough to render $\zeta_{\mu/d}$ and $\zeta_{\mu/(d+1)}$ statistically indistinguishable.
Our future work is to formalize a metric for the quality of a WSN source anonymity scheme that includes the Eve’s outage probability and her work needed to collect statistically relevant samples. By including the adversary’s work and vulnerabilities, we aim to better model a global eavesdropper, and to present the quality of the source anonymity protection as relative to the adversary’s strength. In addition, the quality metric should include the statistically guaranteed anonymity level, the work spent to obfuscate the events, and the latency guarantees by the proposed algorithm.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'anonymity, random graph, big component, adversary'
author:
-
-
-
bibliography:
- 'bibliography.bib'
title: |
How to Cooperate Locally to Improve Global Privacy in Social Networks?\
On Amplification of Privacy Preserving Data Aggregation
---
[^1]
[^1]: Supported by Polish National Science Center - NCN, decision number DEC-2013/ 08/M/ST6/00928 (Harmonia)
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'New H$\alpha$ Fabry-Perot interferometer observations of the Virgo cluster spiral galaxy NGC 4522 are presented. A velocity field up to a radius of 60$''''$ was obtained. The observed rotation curve is symmetric. It is not perturbed in the inner disk. For radii $>50''''$ it shows a solid body rotation. We compare these data with a numerical model which includes the effects of ram pressure stripping. The model can reproduce the main characteristics of the H$\alpha$ emission distribution and velocity field. We therefore conclude that the stripping event which caused the H[i]{} deficiency and the distorted H$\alpha$ distribution and velocity field has been due to ram pressure. The closest passage of the galaxy to the cluster center is estimated to be $\sim$6.510$^{8}$ yr ago. The observed recent star formation is due to the re-accretion of gas clouds which were accelerated to velocities below the escape velocity during the stripping event. Furthermore, the gas which is located at radii $>60''''$ is part of an asymmetric expanding ring of gas clouds. A substantial part of these clouds is located above the disk plane. We predict the H[i]{} gas distribution and velocity field as it would be observed by deep 21 cm observations.'
author:
- 'B. Vollmer, M. Marcelin, P. Amram, C. Balkowski, V. Cayatte'
- 'O. Garrido'
date: 'Received / Accepted'
title: The consequences of ram pressure stripping on the Virgo cluster spiral galaxy NGC 4522
---
Introduction
============
It is a well established fact that spiral galaxies in clusters have less atomic gas than isolated spirals of the same morphological type and same optical diameter, i.e. they are H[i]{} deficient (Chamaraux et al. 1980, Bothun et al. 1982, Giovanelli & Haynes 1985, Gavazzi 1987, 1989). There are mainly two kinds of mechanisms which are able to cause the removal of the atomic gas: (i) tidal interactions or (ii) the interaction of the interstellar medium (ISM) with the hot intracluster medium (ICM). The mapping of the gas content of spiral galaxies in the Virgo (Cayatte et al. 1990, 1994) and Coma cluster (Bravo-Alfaro et al. 2000) showed that the H[i]{} disk sizes of cluster spirals are considerably reduced. In addition, galaxies with a symmetric optical disk that have an asymmetric H[i]{} gas distribution are quite frequent in the cluster core. These observational results indicate that the gas removal due to the rapid motion of the galaxy within the ICM (ram pressure stripping; Gunn & Gott 1972) is responsible for the H[i]{} deficiency and the distorted gas disks of the cluster spirals. Nevertheless, it has not yet been unambiguously shown that ram pressure stripping is responsible for the H[i]{} distribution and kinematics of one given spiral galaxy. The possibility of a tidal interaction producing the distortion has never been completely ruled out.
Very few simulations have been done to quantify ram pressure stripping (Gaetz et al. 1987, Balsara et al. 1994, Tosa 1994, Abadi et al. 1999). All of them have considered that the Interstellar medium (ISM) is continuous. In order to take into account the clumpiness of the ISM, we use a sticky particle model in which each particle represents a cloud complex with an assigned mass dependent radius. The viscosity of the clumpy ISM is due to inelastic collisions between the particles. The effect of ram pressure is modeled as an additional acceleration applied on the particles located at the front side of the galaxy motion. The gas distribution and the velocity field of a given simulation snapshot can be directly compared with observations.
In the Virgo cluster only few spiral galaxies with peculiar gas distributions were studied in detail:
- NGC 4419: Kenney et al. (1990) observed an asymmetric CO distribution in this cluster spiral. They suggest that its ISM is undergoing a strong interaction with the ICM.
- NGC 4254: Phookun & Mundy (1993) observed a counter-rotating, very low surface density H[i]{} plume.
- NGC 4654: Phookun & Mundy (1995) found a very asymmetric H[i]{} distribution, with a compressed edge on one side and a long tenuous tail on the other. They conclude that this is a result of the combination of rotation and ram pressure.
- NGC 4388: Veilleux et al. (1999) made H$\alpha$ and \[O[iii]{}\] $\lambda$5007 observations with the Hawaii Imaging Fabry-Perot Interferometer (HIFI). They found a large complex of highly ionized gas that extends well above the disk. A ram pressure model with a Mach cone of opening angle $\sim$80$^{\rm o}$ is proposed to explain the observed distribution and kinematics of the ionized gas.
The H[i]{} deficient spiral galaxy NGC 4522 was observed in the optical and H$\alpha$ with the WIYN telescope. Kenney & Koopmann (1999) showed that its old stellar disk is relatively undisturbed. However, the H$\alpha$ distribution is very peculiar. Filaments emerge from the outer edge of the disk which is abruptly truncated beyond 0.35$R_{25}$. They suggest that this features are due to a bow shock caused by ram pressure which is still acting on the galaxy. Nevertheless, they made their conclusion only on morphological grounds.
With the imaging of the gas content of galaxies only two ($\alpha$, $\delta$) out of six components ($x$, $y$, $z$, $v_{\rm x}$, $v_{\rm y}$, $v_{\rm z}$) of phase space are accessible. The knowledge of the velocity field represents a considerable increase of information. But still, the derived three-dimensional distribution and velocity field depends strongly on the applied model. The number of possible model solutions decreases strongly with the number of observed quantities. Thus, the knowledge of both the gas distribution and velocity field gives very strong constraints on dynamical models.
We therefore observed NGC 4522 with the Fabry-Perot Interferometer at the Observatoire de Haute Provence (OHP). We were able to recover its velocity field up to a radius of $\sim$60$''$. We first present the observations (Section 2) and the numerical model (Section 3). The observational results are compared to the model in Section 4. We discuss the outcomes of the comparison and draw our conclusions in Section 5.
We adopt a distance of 17 Mpc for the Virgo cluster.
Observations and data reductions
================================
The GHASP Survey
----------------
The Gassendi HAlpha survey of SPiral galaxies [^1] consists of mapping the H$\alpha$ distribution and kinematics of a sample of about 200 nearby spirals using a scanning Fabry Perot interferometer (FP). The aim of the survey begun in October 1998 is to obtain the kinematics with both high spatial and spectral resolutions in the optical disk for a homogeneous sample of galaxies. This should allow us to tune the parameters of the mass distribution accurately and to study the internal dynamics of disks. This project is complementary to the Westerbork Survey of H[i]{} in Spiral Galaxies[^2].
The FP is placed in a focal reducer bringing the original f/15 focal ratio of the Cassegrain focus to f/3.8. The focal reducer is attached at the Cassegrain focus of the 1.93 m OHP telescope (Observatoire de Haute Provence, France). Narrow band and high transmission interference filters are used to isolate the studied emission line. The detector used is an image photon counting system (IPCS). The IPCS, with a time sampling of 1/50 s and zero readout noise makes it possible to scan the interferometer rapidly (typically 10 s per channel) avoiding sky transparency, airmass and seeing variation problems during the exposures. The field of view is 4.1 arcmin x 4.1 arcmin and the pixel size 0.96 arcsec. A full description of the GHASP survey and equipment will be given in a forthcoming paper but the basic principles of a similar instrument can be found in Amram et al. (1991).
Observations
------------
The observation of NGC 4522 have been made with the GHASP equipment in April 1999. Using a scanning FP of interference order 796 at H$\alpha$ the free spectral range of 378 kms$^{-1}$ was scanned through 24 channels (the finesse of the interferometer being 12) providing a spectral sampling of 17 kms$^{-1}$. The journal of the observations is given in Table \[tab:journal\].
$$\begin{array}{ll}
\hline
\noalign{\smallskip}
{\rm Observations} & \\
\hline
\noalign{\smallskip}
{\rm Telescope} & {\rm OHP\ 1.93\ m} \\
{\rm Equipment} & {\rm GHASP\ @\ Cassegrain\ focus} \\
{\rm Date} & {\rm April,\ 7\ and\ 13\ 1999} \\
{\rm Seeing} & $$\sim$$ 2" \\
\noalign{\smallskip}
\hline
\noalign{\smallskip}
{\rm Interference\ Filter} & \\
\hline
\noalign{\smallskip}
{\rm Central\ Wavelength} & 6615 {\rm \AA} \\
{\rm FWHM} & 11 {\rm \AA} \\
{\rm Transmission} & 0.6 \\
\noalign{\smallskip}
\hline
\noalign{\smallskip}
{\rm Calibration} & \\
\hline
\noalign{\smallskip}
{\rm Neon\ Comparison\ light} & $$\lambda$$ 6598.95 {\rm \AA} \\
\noalign{\smallskip}
\hline
\noalign{\smallskip}
{\rm Fabry-Perot} & \\
\hline
\noalign{\smallskip}
{\rm Interference\ Order} & 796 @ 6562.78 {\rm \AA} \\
{\rm Free\ Spectral\ Range\ at\ H}$$\alpha$$ & 380\ {\rm km\,s}$$^{-1}$$ \\
{\rm Finesse\ at\ H}$$\alpha$$ & 12 \\
{\rm Spectral\ resolution\ at\ H}$$\alpha$$ & 18750\ {\rm at\ the\ sample\ step} \\
\noalign{\smallskip}
\hline
\noalign{\smallskip}
{\rm Sampling} & \\
\hline
\noalign{\smallskip}
{\rm Number\ of\ Scanning\ Steps} & 24 \\
{\rm Sampling\ Step} & 0.35 {\rm \AA}\ (16\ {\rm km\,s}$$^{-1}$$) \\
{\rm Total\ Field} & 4.1'$$\times $$4.1' (256$$\times $$256 px$$^{2}$$) \\
{\rm Pixel\ Size} & 0.96'' \\
\noalign{\smallskip}
\hline
\noalign{\smallskip}
{\rm Exposures\ times} & \\
\hline
\noalign{\smallskip}
{\rm Total\ exposure} & 3.5\ {\rm hours} \\
{\rm Elementary\ scanning} & \\
{\rm exposure\ time} & {\rm 10\ s\ per\ channel} \\
{\rm Total\ exposure\ time} & \\
{\rm per\ channel} & 525\ {\rm s}\\
\noalign{\smallskip}
\hline
\end{array}$$
$$\begin{array}{ll}
\hline
\noalign{\smallskip}
{\rm Other\ names} & {\rm UGC~7711} \\
& {\rm VCC~1516} \\
& {\rm CGCG~070-168} \\
$$\alpha$$\ (1950)$$^{\rm a}$$ & 12$$^{\rm h}31^{\rm m}7.62^{\rm s}$$\\
$$\delta$$\ (1950)$$^{\rm a}$$ & 9$$^{\rm o}27'3.1''$$\\
{\rm Morphological\ type}$$^{\rm a,\ b}$$ & {\rm SBcd,\ Sc/Sb} \\
{\rm Optical\ diameter\ D}_{25}$$^{\rm a}$$\ ($$'$$) & 3.7\\
{\rm B}$$_{T}^{0}$$$$^{\rm a}$$ & 12.02\\
{\rm Systemic\ heliocentric\ velocity}$$^{\rm c}$$\ {\rm (km\,s}$$^{-1}$$) & 2337$$\pm$$5\\
{\rm (optical)} & \\
{\rm HI\ heliocentric\ velocity}$$^{\rm d}$$\ {\rm (km\,s}$$^{-1}$$) & 2330$$\pm$$5 \\
{\rm Distance\ D\ (Mpc)} & 17 \\
{\rm Log\ L(H}$$\alpha$$)$$^{\rm e}$$\ ({\rm erg\,s}$$ ^{-1}$$) & 40.11\\
{\rm Vrot}$$_{\rm max}^{\rm f}$$\ {\rm (km\,s}$$^{-1}$$) & 110$$\pm$$10 \\
{\rm R}$$_{\rm max}^{\rm f}$$\ {\rm (kpc)} & 4.5$$\pm$$0,5 \\
{\rm M}$$_{\rm tot}^{\rm g}$$\ ($$\times$$ 10$$^{10}$$\ {\rm M}$$_{\odot}$$) & 2 \\
{\rm PA\ (gas\ kinematics)}$$^{\rm f}$$ & 33$$^{\rm o}$$ \\
{\rm PA\ (optical\ image)}$$^{\rm a}$$ & 33$$^{\rm o}$$ \\
{\rm Inclination\ of\ the\ gaseous\ disk}$$^{\rm e}$$ & 75$$^{\rm o} \pm 5^{\rm o}$$ \\
{\rm FWHM\ of\ central\ profiles}$$^{\rm f}$$\ {\rm (km\,s}$$^{-1}$$) &80$$\pm$$10 \\
{\rm HI\ deficiency}$$^{\rm h}$$ & 0.51 \\
\noalign{\smallskip}
\hline
\end{array}$$
RC3
Binggeli et al. (1985)
Rubin et al. (1998)
Helou et al. (1984)
Kenney & Koopmann (1999)
this paper
inside D$_{25}$ assuming a flat rotation curve of $V_{\rm max}$=100 kms$^{-1}$
Giovanelli & Haynes (1985)
Reductions
----------
The data were reduced using the ADHOCw software package [^3]. The data reduction procedure has been extensively described in Amram et al. (1996) and references therein.
Wavelength calibrations were obtained by scanning the narrow Ne 6599 Å line under the same conditions as the observations. The relative velocities with respect to the systemic velocity are very accurate, with an error of a fraction of a channel width ($<$3 kms$^{-1}$) over the whole field.
The signal measured along the scanning sequence was separated into two parts: (1) an almost constant level produced by the continuum light in a 10 Å passband around H$\alpha $ (continuum map), and (2) a varying part produced by the H$\alpha $ line (monochromatic map). The continuum level was taken to be the mean of the three faintest channels, to avoid channel noise effects. The monochromatic map was obtained by integrating the monochromatic profile in each pixel. The velocity sampling was 16 kms$^{-1}$. The monochromatic maps had one-pixel resolution in the center of the galaxy. Spectral profiles were binned in the outer parts (to 5 $\times$ 5 pixels) in order to increase the signal-to-noise ratio. When multiple components were visually present, the lines were decomposed into multiple Gaussian components. OH night sky lines passing through the filters were subtracted by determining the shapes and intensities of the lines away from the galaxies (Laval et al. 1987).
The numerical model
===================
We used the three-dimensional N-body code described in detail in Vollmer et al. (2000). The particles represent gas cloud complexes which are evolving in an analytically given gravitational potential of the galaxy. This potential consists of two spherical parts: the dark matter halo and the stellar bulge. The outcoming velocity field has a constant rotation curve of $v_{\rm rot} \sim$130 kms$^{-1}$ (Fig. \[fig:initrotationvel\]).
10000 particles of different masses are rotating within this gravitational potential. A radius, which depends on the mass, is attributed to each particle. During the disk evolution they can have inelastic collisions. The outcome of these collisions is simplified following Wiegel (1994):
- for $r_{1}-r_{2} < b < r_{1}+r_{2}$:\
fragmentation
- for $b \le r_{1}-r_{2}$ and $v_{\rm esc} > v_{\rm f}$:\
mass exchange
- for $b \le r_{1}-r_{2}$ and $v_{\rm esc} \le v_{\rm f}$:\
coalescence,
where $b$ is the impact parameter, $r_{1}$ and $r_{2}$ the cloud radii, $v_{\rm f}$ the final velocity difference, and $v_{\rm esc}$ is the escape velocity. This results in an effective gas viscosity in the disk.
As the galaxy moves through the ICM its clouds are accelerated by ram pressure. Within the galaxy’s inertial system the clouds are exposed to a wind coming from the front side of its motion. The effect of ram pressure on the clouds is simulated by an additional force on the clouds in the wind direction. Only clouds which are not protected by other clouds against the wind are affected.
As the galaxy approaches the cluster center its velocity increases. At the same time the surrounding ICM density increases. This leads to an increase of the ram pressure on the ISM clouds $p_{\rm ram}=
\rho_{\rm ICM} v_{\rm gal}^{2}$, where $\rho_{\rm ICM}$ is the ICM density and $v_{\rm gal}$ is the velocity of the galaxy. We take this evolution of $p_{\rm ram}$ into account in adopting the following profile $p_{\rm ram}=28\,\rho_{0} v_{0}^{2} {\rm exp} \big(
-(t/8\,10^{7})^{2} \big)$, where $\rho_{0}=10^{-4}$ cm$^{-3}$ and $v_{0}$=1000 kms$^{-1}$. This corresponds to a galaxy’s closest approach of $\sim 10^{5}$ pc to the cluster center.
The choice of the maximal ram pressure and the inclination angle $i$ between the disk and the orbital plane depends on the resulting H[i]{} deficiency and the fraction of re-accreted gas (see Section \[sec:results\]). Furthermore, the observed gas distribution gives an indication for the right range of the inclination angle $i$. The chosen parameters $i$=20$^{\rm o}$ and $p_{\rm ram}^{\rm max}=28\,\rho_{0} v_{0}^{2}$ lead to a model H[i]{} deficiency of $DEF$=0.4 compared to the observed deficiency of $DEF$=0.5. The correspondence is good enough in the way that the simulation shows all characteristics of the distribution and kinematics necessary to identify the stripping process unambiguously. We have let the galaxy evolve during 10$^{9}$ yr before beginning the ram pressure simulations.
Results\[sec:results\]
======================
A sequence of the galaxy evolution is shown in Fig. \[fig:film\].
At $t_{0}=6\,10^{8}$ yr ram pressure is at maximum. An overdensity builds up within the disk in the wind direction (south-west) together with an asymmetric ring structure. About 10$^{8}$ yr later, the material in the ring moves to the north-east leaving the disk. At this stage of the evolution ram pressure has already ceased and the subsequent evolution of the ISM is mainly governed by rotation and re-accretion. As for the case of edge-on stripping, there is an accelerated arm where the wind pushes the clouds in the sense of rotation and a decelerated arm where the clouds are decelerated by the wind (see Vollmer et al. 2000). The accelerated arm extends to the north, the decelerated arm is horizontal. After $t-t_{0} \sim 2\,10^{8}$ yr the clouds, which where not accelerated to the escape velocity, begin to fall back onto the galaxy. This happens within two distinct arms. A small, counter-rotating arm in the north-west and a more massive arm coming from the north-east where the clouds fall back to the galaxy in the sense of rotation. It hits the outer galactic disk in the south-west carrying a part of the disk gas clouds with it to larger galactic radii. This leads to the formation of an expanding asymmetric ring at $t-t_{0} \sim 3\,10^{8}$ yr. During the expansion the re-accreting gas clouds of the counter-rotating arm collide with the clouds located in the ring. We suggest that these collisions lead to star formation activity within the ring structure.
Vollmer et al. (2000) have shown that in general a substantial part of the stripped ICM re-accretes between 2 and 710$^{8}$ yr after the closest passage of the galaxy to the cluster center. Furthermore, the effects of ram pressure within a Virgo spiral galaxy can only be observed after its passage in the cluster core. This is also the case for the simulation shown in Fig. \[fig:film\]. The large projected distance of NGC 4522 with respect to the Virgo cluster center (M87) of 3.3$^{\rm o}$ ($d \sim$1 Mpc) makes it thus very unlikely that the galaxy is just entering the cluster. The ICM density at this distance is only $\rho_{\rm ICM}\sim 3\,10^{-5}$ cm$^{-3}$ (Schindler et al. 1999). Even if the galaxy has a velocity of 1800 kms$^{-1}$ the resulting ram pressure does not exceed $\rho_{0} v_{0}^{2}$. For these low values the influence of ram pressure on the neutral component of the ISM is not observable.
We therefore come to the conclusion that NGC 4522 has already passed the cluster center. Since the galaxy is leaving the cluster now, we see the consequences of a past strong stripping event. Assuming a mean galaxy velocity of $v_{\rm gal}$=1500 kms$^{-1}$ we can calculate the time elapsed since its closest passage to the cluster center: $t\simeq d/v_{\rm gal}\simeq$ 6.510$^{8}$ yr. We therefore compare a snapshot of our simulation at this timestep with our observations. In this stage most of the re-accretion has already taken place. Since our model does not explicitly include star formation we can only speculate that this re-accretion has lead to an increase of the star formation rate making the galaxy bluer. At the present day, accretion is still happening and the collisions between the infalling clouds and the ISM probably still induces star formation. For direct comparison we use the position angle, inclination and systemic velocity of NGC 4522 (Table \[tab:parameters\]).
First, we show in Fig. \[fig:point\_dist\] the three-dimensional particles distribution and velocity field of the snapshot of our simulation that we want to compare with the observations. It shows the gas cloud distribution of the galaxy in a face-on and edge-on view. It corresponds to the last snapshot of Fig. \[fig:film\].
The H[i]{} disk has a diameter of $\sim$14 kpc. The asymmetric ring extends up to radii of $\sim$20 kpc. A part of the ring is located beyond the disk plane (Fig. \[fig:point\_dist\] lower panel). At this stage of evolution, a tenuous gas component forms complex three-dimensional structure around the galaxy disk.
The three-dimensional velocity field can be seen in Fig. \[fig:point\_vel\].
The galaxy rotates counter-clockwise. The gas which is located in the north is accreting back to the galaxy. We observe a lot of counter-rotating clouds in the expanding ring. In particular, there is a streamer which runs almost perpendicular to the disk (Fig. \[fig:point\_vel\] lower panel). It comes from below and crosses the disk at its eastern edge. We want to stress here again that there are a lot of clouds whose trajectories will lead to collisions.
Since we have used a galaxy with an optical diameter of $D$=30 kpc in our simulations (see Vollmer et al. 2000) whereas NGC 4522 has an optical diameter of $D_{25}$=18 kpc, we adapted the length scale of our simulations to that of NGC 4522 (roughly a factor 2). This three-dimensional cloud distribution and velocity field can now be projected using the observed position angle and inclination. We will directly compare the outcomes of this projection with the observational results. The model images have almost the resolution of the observations.
The distribution of the ionized gas
-----------------------------------
In order to model the H$\alpha$ distribution, we assume that star formation is induced by the compression of the clouds during cloud–cloud collisions. Consequently, we have to search for clouds with crossing trajectories in order to detect gas streams which lead to frequent inelastic collisions. We thus project the velocity vectors (${\bf v_{1}}$, ${\bf v_{2}}$) of each pair of clouds on their relative distance vector ${\bf r_{12}}$: $v_{1}^{\rm proj}={\bf v_{1}}\cdot
{\bf r_{12}}\ ;\ v_{2}^{\rm proj}={\bf v_{2}}\cdot {\bf r_{12}}\ .$ We apply a weighting factor to each cloud which is equal to the number of possible collisions fulfilling the criterion: $r_{12} \leq 1$ kpc, $v_{1}^{\rm proj}$ has the opposite sign of $v_{2}^{\rm proj}$, and $R > 10$ kpc, where $R$ is the distance of the cloud to the galaxy center. At distances $R \leq 10$ kpc, i.e. within the disk, the mean free path of a cloud is smaller than outside the disk. The criterion $r_{12} \leq 1$ kpc might thus not be valid at small distances. Moreover, in the inner disk star formation is more likely due to density waves within the stellar and gaseous disk. Since this aspect is not included in the model we prefer to give the clouds at $R \leq 10$ kpc a uniform weight. The model emission inside the galaxy disk is thus density weighted, the emission outside the disk is collision weighted. This treatment ensures in our view the simplest approach in order to model the observed H$\alpha$ emission. Fig. \[fig:distributions\] shows the observed together with the model of the H$\alpha$ emission distribution.
The observed H$\alpha$ emission distribution is already discussed in detail in Kenney & Koopmann (1999). We therefore just summarize the main features.
- The H$\alpha$ disk is strongly truncated.
- 10 % of the total H$\alpha$ emission arises from a one-sided, extra planar distribution.
- The extra planar ionized gas is organized into filaments.
- The extra planar H$\alpha$ emission has also a diffuse component.
- The region of highest H$\alpha$ emission in the core of the disk is more extended to the south-west.
- The region of intermediate H$\alpha$ emission in the galaxy disk is more extended to the north-east.
Almost all characteristics of the observed H$\alpha$ emission distribution can also be found in the model distribution (Fig. \[fig:distributions\] lower panel):
- The disk is strongly truncated. This is due to the truncation of the H[i]{} gas disk at a radius $R \sim 7$ kpc.
- A fraction of several percent of the total emission can be found in the north-west, far away from the disk. This emission is due to colliding clouds which form stars ionizing their environment. This happens predominantly in the expanding ring.
- Our model image does not have enough resolution to show filaments. A careful inspection of the observed H$\alpha$ image of Kenney & Koopmann (1999) shows that the filaments are mainly located near the north-eastern and south-western end of the galaxy’s disk. These are the places where the expanding ring joins the disk. Moreover, the filamentary structure can be due to magneto–hydrodynamic phenomena which are not included in our code.
- We have not yet included the mechanism of star formation in the model. We therefore can not make the difference between clumped and diffuse emission directly. Nevertheless, since the expanding ring has a considerable extension in the disk plane with a relatively low cloud density, we expect that there is a part of UV emission which escapes out of the cloud complex, where the stars were built. This environment can be other clouds in the expanding ring or the tenuous material which is distributed around the galaxy disk (see Fig. \[fig:point\_dist\]). The diffuse component of the observed H$\alpha$ emission might therefore be the trace of this tenuous gas component.
- The maximum emission of the disk shows the same asymmetry as the observed H$\alpha$ emission distribution.
The velocity field of the ionized gas
-------------------------------------
We will first discuss the overall behavior of the observed H$\alpha$ rotation curve averaged for both sides. For the comparison with our model, the approaching and receding sides are separated. The H$\alpha$ rotation curve can be seen in Fig. \[fig:mdm\_model\].
It is made with a constant position angle and a constant inclination angle. The error bars correspond to the difference between the approaching and the receding side. The rotation curve rises rapidly up to a distance of 20$''$ and more slowly for distances between 20$''$ and 50$''$. Between 50$''$ and 60$''$ the velocities of the approaching and receding side differ considerably ($\sim$50%). A drop of the rotation velocity from 90 kms$^{-1}$ to 70 kms$^{-1}$ can be observed at 65$''$. The rotation curve rises then linearly up to a distance of 90$''$. This rise is due to the H$\alpha$ blobs located in the north west direction above the disk’s major axis. The difference between both sides is quite small ($\sim$20 kms$^{-1}$) for this outer part of the rotation curve.
The mass distribution has been analyzed using a mass model and a code written by Carignan (1985). The solid line in Fig. \[fig:mdm\_model\] corresponding to this model rotation curve is based on the R luminosity profile (Koopmann & Kenney 2000), an intrinsic ratio Qo=0.15, a constant M/L=5, and a maximum disk assumption. The mass surface density of the disk is derived from the surface brightness profile at each radius. This explains the dip in the model rotation curve at 15$''$ caused by a dip in the R luminosity profile, which is due to the galaxy’s internal structure, i.e. the bar. The model curve reproduces accurately the H$\alpha$ rotation curve up to 60$''$. Outside, it was impossible to reproduce the rising solid body rotation curve using a maximum disk mass model or a best fit model with any known shape of dark halo. We therefore conclude here that the H$\alpha$ emission blobs which are located in the north west must be located beyond the disk plane and/or must be accelerated.
In the following paragraph the observed and the model rotation curves will be directly compared. Fig. \[fig:vfields\] shows the velocity field of the H$\alpha$ observations together with that of the model for the approaching and receding side separately.
The overall structure of both velocity fields is in good agreement. The difference in the inner disk is due to the initial model rotation curve, which rises more rapidly than the observed one.
The derived rotation curves within a sector of 5$^{\rm o}$ around the major axis is plotted in Fig. \[fig:compare5\].
Our observations confirm the non-perturbed rising rotation curve measured by Rubin et al. (1999).
The observed rotation curve continues to rise at both sides up to a distance of $|R|\sim 50''$. The receding side shows a little bump at $R \sim 20''$. The model rotation curve does not differ from the initial model rotation curve for radii $R < 40''$, i.e. it stays constant. The outer end at positive distances begins to rise, whereas the outer end at the opposite side shows the inverse trend.
As mentioned in the previous section, the initial model rotation curve is not rising but flat. This changes obviously the velocity field of the inner disk but not that for larger distances to the galaxy center. The main result of the simulation at this point is that the rotation curve of the inner disk is not altered by the ram pressure stripping event, i.e. it stays flat.
The most important test for the model velocity field is the velocity structure of the emission at the north-eastern side of the galaxy disk. These regions are included in the rotation curve derived within a sector of 40$^{\rm o}$ around the major axis (Fig. \[fig:compare40\]).
The observed rotation curve is symmetric. It rises steeply up to a distance of $|R| \sim 20''$ from the galaxy center. Then, the velocity increases more slowly up to a distance of $|R|\sim 50''$. After a sudden drop it rises linearly (solid body rotation) and reaches $\sim 200$ kms$^{-1}$ at a distance of $|R| \sim 100''$.
The model rotation curve shows the same overall behavior. It rises linearly at both sides and covers the same velocity range than the observed rotation curve. The main difference between the model and the observed rotation curve is the asymmetry of the model rotation curve. While the receding side has a very similar behavior as the observed rotation curve, the linear rise of the rotation curve at the approaching side begins at a smaller distance ($R \sim -30''$). For $R < -50''$ the rotation curve drops abruptly. The observed rotation curve shows also a decrease but at larger radius ($R \sim -90''$).
Off-plane material and predicted H[i]{} gas distribution and velocity field
---------------------------------------------------------------------------
We can investigate if the material in the expanding ring is located beyond the galaxy’s disk plane as suggested by Kenney & Koopmann (1999). In order to study the emission distribution in the $z$ direction we removed the disk particles from the model and plotted the emission distribution for the galaxy plane seen edge-on (Fig. \[fig:edge-on\]).
A substantial part of the gas clouds in the expanding ring located at radii $R > 10$ kpc is thus well situated up to 2 kpc above the disk plane.
With the help of the three-dimensional model cloud distribution and velocity field we can attempt to predict the overall gas distribution and velocity field as it would be detected by very deep H[i]{} 21 cm observations (Fig. \[fig:HI\_all\]).
In doing so, we have to be very cautious. The major caveat for this prediction is the unknown mass loss rate of the neutral gas clouds which are located in the hot ICM due to evaporation. Since the stripping event has already happened a long time ago, it is possible that a significant fraction of the extra planar neutral gas clouds has already evaporated and is no longer observable in H[i]{}. Very deep H[i]{} observations (VLA D array) will therefore not only be a test for the chosen model but they will also give some clues on the evaporation timescale of the stripped neutral gas clouds.
Discussion and conclusions
==========================
We have shown new H$\alpha$ Fabry-Perot interferometer observations of the Virgo cluster spiral galaxy NGC 4522. The velocity field can be traced up to radius of $\sim 60''$. We compare the results of these observations with a numerical model which includes the effects of ram pressure acting on the galaxy’s gas content.
We derived rotation curves within a sector of 5$^{\rm o}$ and 40$^{\rm o}$ around the major axis. Thus, we have been able to compare the emission distribution, the velocity field, and the rotation curves of the observations with those of the model.
For the inner disk of the galaxy, the main result is that the ram pressure stripping event which happened 6.510$^{8}$ yr ago did not alter the initial rotation curve of the galaxy. For the outer disk, the stripped material which has not been accelerated above the escape velocity is re-accreting on the galaxy. The majority of this material forms an expanding, asymmetric ring structure. A small fraction of the stripped gas is falling back onto the galaxy within a counter-rotating arm. Inelastic collisions between the gas clouds located in the ring on the one hand and between clouds of the counter-rotating arm and clouds located in the ring on the other hand happen frequently. This leads to an enhanced star formation activity within the ring structure. Since we have traced the total number of collisions during the simulation, we can estimate the star formation rate due to cloud–cloud collisions: during the last $\Delta t=$510$^{7}$ yr the number of collisions is $n_{\rm coll}$=31. With a cloud mean mass $\overline{m}_{\rm cl}$=810$^{5}$ M$_{\odot}$ and assuming that 10 % of the gas is turned into stars this gives $$SFR = 0.1\,n_{\rm coll}\,2\,\overline{m}_{\rm cl}/\Delta t \simeq 0.1
\ {\rm M}_{\odot}\,{\rm yr}^{-1}$$ This is in excellent agreement with the measured star formation rate of $SFR = 0.11\ {\rm M}_{\odot}\,{\rm yr}^{-1}$ (Kenney & Koopmann 1999). The newly formed massive and hot stars ionize their surroundings giving rise to the H$\alpha$ emission. These detached emission blobs lead generally to a rise of the rotation curve, and sometimes to a decreasing rotation curve. This corresponds to the behavior of the observed velocity field.
It is important to mention that the numerical simulation does not pretend to reproduce the H$\alpha$ distribution and velocity field in each detail. The aim of the comparison is to find the same general characteristics which allow us to discriminate whether the model describes the observations well. In our case, the comparison shows the following similarities:
- The H$\alpha$ disk is strongly truncated.
- A considerable amount of the total H$\alpha$ emission arises from a one-sided, extra planar distribution.
- The region of highest H$\alpha$ emission in the core of the disk is more extended to the south-west.
- The region of intermediate H$\alpha$ emission in the galaxy disk is more extended to the north-east.
- The rotation curves of the inner disk are those of an undisturbed galaxy.
- The rotation curves of the outer disk shows a linear rise (solid body rotation).
- The detached emission blobs lead generally to a rise of the rotation curve, and sometimes to a decreasing rotation curve.
The dissimilar aspects are by far less numerous:
- The observed H$\alpha$ emission off-plane comes mainly from three distinct regions, whereas the model emission off-plane covers 180$^{\rm o}$.
- The model rotation curve is not symmetric.
- The model emission distribution does not show a filamentary structure.
- The model emission distribution does not show a diffuse component.
The last two points are due to numerics (discrete model, missing resolution, no magnetic fields, no star formation mechanism). The asymmetry of the model rotation curve is most likely due to the initial conditions of the simulation. We have run several simulations with different initial conditions, but since the parameter space for these initial conditions is very large, we were not able to find a model snapshot which fits our observations more tidily.
We therefore conclude that the model describes very well the overall aspects of the H$\alpha$ observations.
This leads to the final conclusion that the galaxy’s closest passage to the cluster center is $\sim$6.510$^{8}$ yr ago and the galaxy is coming out of the cluster core. Its positive radial velocity with respect to the cluster mean velocity places it behind the cluster center (M87).
The authors wish to thank Jacques Boulesteix for his help on the data acquisition system and data reduction software for GHASP. BV is supported by a TMR Programme of the European Community (Marie Curie Research Training Grant).
Abadi, M.G., Moore, B., Bower, R.G., 1999, MNRAS, 308, 947 Amram, P., Boulesteix, J., Georgelin, Y.M., et al., 1991, The Messanger, 64, 44 Amram, P., Balkowski, C., Boulesteix, J., Cayatte, V., Marcelin, M., Sullivan, W.T.III, 1996, A&A, 310, 737 Balsara, D., Livio, M., O’Dea, C.P., 1994, ApJ, 437, 83 Binggeli, B., Sandage, A., Tammann, G.A., 1985, AJ, 90, 1681 Bothun, G., Schommer, R.A., Sullivan, W.T.III, 1982, AJ, 87, 731 Boulesteix, J. 1999, ADHOCw Users Manual (Pub. Obs. Marseille) Bravo-Alfaro, H., Cayatte, V., van Gorkom, J.H., Balkowski, C., 2000, AJ, 119, 580 Carignan, C., 1985, ApJ, 299, 59 Cayatte, V., van Gorkom, J.H., Balkowski, C., Kotanyi, C., 1990, AJ, 100, 604 Cayatte, V., Kotanyi, C., Balkowski, C., van Gorkom, J.H., 1994, AJ, 107, 1003 Chamaraux, P., Balkowski, C., Gérard, E., 1980, A&A, 83, 38 de Vaucouleurs, G., de Vaucouleurs, A., Corwin, H.G., Buta, R.J., Paturel, G., Fouqué, P., 1991, Third Reference Catalogue of Bright Galaxies,(New York:Springer)(RC3) Gaetz, T.J., Salpeter, E.E., Shaviv, G., 1987, ApJ, 316, 530 Gavazzi, G., 1987, ApJ, 320, 96 Gavazzi, G., 1989, ApJ, 346, 59 Giovanelli, R., Haynes, M.P., 1985, ApJ, 292, 404 Gunn, J.E., Gott, J.R., 1972, ApJ, 176, 1 Helou, G., Hoffman, G.L., Salpeter, E.E., 1984, ApJS, 55, 433 Kenney, J.P.D., Young, J.S., Hasegawa, T., Nakai, N., 1990, ApJ, 353, 460 Kenney, J.P.D., Koopmann, R.A., 1999, AJ, 117, 181 Koopmann, R.A., Kenney, J.P.D., 2000, AJ, submitted Laval, A., Boulesteix, J., Georgelin, Y.P., Georgelin, Y.M., Marcelin, M., 1987, A&A, 175, 199 Phookun, B., Mundy, L.G., 1993, ApJ, 418, 113 Phookun, B., Mundy, L.G., 1995, ApJ, 453, 154 Rubin, V., Waterman, A.H., Kenney, J.P.D., 1999, AJ, 118, 236 Schindler, S., Binggeli, B., Böhringer, H., 1999, A&A, 343, 420 Tosa, M., 1994, ApJ, 426, L81 Veilleux, S., Bland-Hawthorn, J., Cecil, G., Tully, R.B., Miller, S.T., 1999, ApJ, 520, 111 Vollmer, B., Cayatte, V., Balkowski, C., Duschl, W.J., 2000, ApJ, submitted Wiegel, W., 1994, Diploma thesis, University of Heidelberg
[^1]: http://www-obs.cnrs-mrs.fr/interferometrie/ghasp.html
[^2]: http://thales.astro.rug.nl/ whisp/
[^3]: ftp availaible http://www-obs.cnrs-mrs.fr/adhoc/adhoc.html, Boulesteix et al. (1999)
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The dispersion relations for nucleon-nucleon ($NN$) $T$-matrix in the framework of the Bethe-Salpeter equation for a two spin-one-half particle system and with a separable kernel of interaction are considered. The developed expressions are applied for construction of the separable kernel of interaction for $S$ partial-waves in singlet and triplet channels. We calculate the low energy scattering parameters, the phase shifts, and the deuteron binding energy with the separable interaction. The approach can be easily extended to higher partial-waves for $NN$-scattering and other reactions ($\bar N N$-, $\pi N$- scattering).'
author:
- 'Serge G. Bondarenko'
- 'Valery V. Burov'
- Naohide Hamamoto
- Hiroshi Toki
title: 'Separable Kernel of Nucleon-Nucleon Interaction in the Bethe-Salpeter Approach'
---
Introduction
============
Recent and planned experiments on $NN$-scattering [@lehar] as well as reactions with the deuteron and light nuclei reach large momentum transfer from the probe particle to the investigated system. In this context, relativistic effects play an essential role in understanding reaction mechanisms. Studies of elastic electron-deuteron scattering (see, for example, [@GH; @pair]) in the nonrelativistic framework require to take into account mesonic exchange currents (a part of them has a pure relativistic origin, i.e. the so-called pair currents [@BB97]). Theoretical calculations should therefore follow relativistic approaches.
The approach based on the Bethe-Salpeter equation (BS approach) [@BS51] gives a powerful method to investigate various nuclear reactions both for electromagnetic and strong interactions. To find an analytic solution for the BS equation for $T$-matrix of $NN$ scattering describing bound and scattering states, the separable kernel of interaction is very useful [@yam]. The separable kernel is well known from nonrelativistic consideration to solve the Lippman-Schwinger equation analytically [@pless]. The separable form of interaction is good not only for the technical reason but, in our opinion, also for the reflection of the nonlocal type of elementary nucleon-nucleon interaction (it was stressed first in papers [@yam]). This allows us not only to avoid the divergence peculiar of local theories but also to take into account the internal structure of the nucleon.
In the paper, we consider the BS equation for $T$-matrix and solve it using the separable kernel of interaction for a spin-one-half particle system (for a spin-zero case see, for example, [@zingl] and for trinucleon systems [@rupp-tjon]). This allows us to solve the BS equation without referring to the ladder approximation. We also find dispersion relations for the obtained solution which allow us to perform explicit analytic calculations and connect internal parameters of the kernel with experimental data: low-energy parameters, bound state energy, and phase shifts.
The paper is organized as follows: after the partial-wave decomposition of the BS equation (section \[sec:decomp\]), the one-rank separable kernel of interaction is introduced and used for solving the BS equation (section \[sec:separable\]). Dispersion relations for the $T$-matrix are discussed in section \[sec:disper\]. In section \[sec:ladder\], it is shown that the separable kernel of interaction includes the kernel in the ladder approximation, and the connection between parameters of two kernels is found. Section \[sec:results\] is devoted to results of calculation of the low-energy parameters, bound state energy, and phase shifts of elastic $NN$-scattering and to the discussion for the future works.
Partial-wave decomposition of the BS equation {#sec:decomp}
=============================================
The BS equation for the two-nucleon $T$-matrix with momenta $p_1 (p_1^{\prime}),
p_2 (p_2^{\prime})$ could be represented in the following form: $$\begin{aligned}
T(p^{\prime}, p; P) = V(p^{\prime}, p; P) +
\frac{i}{(2\pi)^4}\int dk\, V(p^{\prime}, k; P) S(k; P) T(k, p; P),
\label{t00}\end{aligned}$$ where $P=p_1+p_2$, $p=(p_1-p_2)/2$ \[$p^{\prime}=(p_1^{\prime}-p_2^{\prime})/2$\] are total and relative momenta, and $V(p^{\prime}, p; P)$ is the $NN$ kernel of interaction. The two-body propagator in terms of the total and relative momenta has the form: $$\begin{aligned}
S(k; P)=1/\left[({\hat P}/2+{\hat k}-m+i\epsilon)
({\hat P}/2-{\hat k}-m+i\epsilon)\right],\end{aligned}$$ with ${\hat a} = a_{\mu} \cdot \gamma^{\mu}$, and $\gamma^{\mu}$ are Dirac matrices.
After the partial wave decomposition in the rest frame of the two-nucleon system, the equation (\[t00\]) becomes ($s=P^2$), $$\begin{aligned}
T_{\alpha \beta}(\pzp, \pmp, p_0, p; s) &=&
V_{\alpha \beta}(\pzp, \pmp, p_0, p; s)
\label{t01}\\
&+& \frac{i}{4\pi^3} \sum_{\gamma \delta} \int\, dk_0\,\int\,
k^2\, dk\, V_{\alpha \gamma}(\pzp, \pmp, k_0, k; s) S_{\gamma
\delta}(k_0, k; s) T_{\alpha \beta}(k_0, k, p_0, p; s).
\nonumber\end{aligned}$$ (Here and below $p \equiv |\pmb|$, $\pmp
\equiv |\pmpb|$, $k \equiv |\kmb|$). Greek letters denote partial waves in the channels under consideration. In $LSJ\rho$-basis [@KU72; @bb:cov2], the singlet $^1S_0$ channel consists of four partial waves, and the triplet $^3S_1$ channel, of eight partial-waves.
Below we omit all states which have at least one negative-energy nucleon and take only $\Sp$-wave in the singlet channel and neglect also $\Dpp$-wave and take only $\Spp$-wave in the triplet channel. Therefore eq. (\[t01\]) could be rewritten as: $$\begin{aligned}
t(\pzp, \pmp, p_0, p; s) = v(\pzp, \pmp, p_0, p; s)
\hskip 70mm
\label{t02}\\
+ \frac{i}{4\pi^3}
\int\, dk_0\,\int\, k^2\, dk\, \frac{v(\pzp, \pmp, k_0, k;s)\,
t(k_0, k, p_0, p; s)}{(\sqrt{s}/2-e_k+i\epsilon)^2-k_0^2},
\nonumber\end{aligned}$$ where $t \equiv T_{{\tilde \alpha}\, {\tilde \alpha}},
v \equiv V_{{\tilde \alpha}\, {\tilde \alpha}}$, with ${\tilde \alpha} = \Sp\mbox{ or }\Spp$, and $e_k = \sqrt{k^2+m^2}$ and $m$ is the nucleon mass.
The $T$-matrix normalization condition could be written in the on-mass-shell form. Introducing phase shifts $\delta(s)$, one could write $$\begin{aligned}
t(s) \equiv t(0,\pp,0,\pp,s) = - \frac{16 \pi}{\sqrt{s}\sqrt{s-4m^2}}\,
e^{i\delta(s)}\, \sin{\delta(s)},
\label{t03}\end{aligned}$$ with $\pp = \sqrt{s/4-m^2} = \sqrt{2mT_{lab}}$.
To introduce also the low-energy parameters of $NN$-scattering, it is suitable to expand the above expression into a series of $\pp$-terms with the following expression [@brown]: $$\begin{aligned}
\pp\, {\cot}\, \delta(s) = - a_0^{-1} + \frac{r_0}{2}\pp^2 + {\cal O}(\pp^3).
\label{t03a}\end{aligned}$$ Taking into consideration only the first two terms of the decomposition (\[t03a\]), we define the scattering length $a_0$ and the effective radius $r_0$.
As for the mass of the bound state ($M_b$), the squared $T$-matrix has a simple pole in the total momentum squared ($s$), and the bound state conditions could be written in the following form: $$\begin{aligned}
t(\pzp, \pmp, p_0, p; s) = \frac{B(\pzp, \pmp, p_0, p; s=M_b^2)}{s-M_b^2}
+ R(\pzp, \pmp, p_0, p; s),
\label{t03b}\end{aligned}$$ where functions $B$ and $R$ are regular at the point $s=M_b^2$. The bound state energy $E_b$ is connected with $M_b$: $M_b = 2m-E_b$.
Rank I separable kernel of interaction {#sec:separable}
======================================
The [*ansatz*]{} for the separable kernel of interaction with rank I has the form: $$\begin{aligned}
v(\pzp, \pmp, p_0, p; s) = \lambda g(\pzp, \pmp) g(p_0, p).
\label{t04}\end{aligned}$$ Here introduced is the g-function for $NN$-vertices, and $\lambda$ is the coupling parameter.
To solve (\[t02\]) we could assume the following separable form for $T$-matrix: $$\begin{aligned}
t(\pzp, \pmp, p_0, p; s) = \tau(s) g(\pzp, \pmp) g(p_0, p)
\label{t05}\end{aligned}$$ and substituting (\[t04\]) and (\[t05\]) into (\[t02\]), we obtain the function $\tau(s)$: $$\begin{aligned}
\tau(s) = 1/(\lambda^{-1} + h(s)),
\label{t06}\end{aligned}$$ where the function $h(s)$ is: $$\begin{aligned}
h(s) = -\frac{i}{4\pi^3}\, \int\, dk_0\,\int\, k^2\, dk\,
\frac{[g(k_0,k)]^2}{(\sqrt{s}/2-e_k+i\epsilon)^2-k_0^2}.
\label{t07}\end{aligned}$$
Therefore, the $T$-matrix could be rewritten in the following form: $$\begin{aligned}
t(\pzp, \pmp, p_0, p; s) = \frac{g(\pzp, \pmp) g(p_0, p)}
{\lambda^{-1} + h(s)},
\label{t08}\end{aligned}$$ and the on-mass-shell expression is: $$\begin{aligned}
t(s) = \frac{n(s)}{d(s)} = \frac{[g(0, \pp)]^2}{\lambda^{-1} +
h(s)}. \label{t09}\end{aligned}$$ It should be noted that the expression (\[t09\]) has the so-called $N/D$-form widely used in the nonrelativistic $T$-matrix theory [@brown] and in some relativistic methods.
Using (\[t09\]), it is easy to connect the solution of $T$-matrix and the phase shifts $\delta(s)$. To achieve this, we assume that the imaginary part of the function $n(s)$ satisfies the following condition: $$\begin{aligned}
\im{} n(s) = 0. \label{t010}\end{aligned}$$ This condition is connected with the specific choice of g-functions for $NN$-vertices to be discussed in the next section. By using (\[t09\]) and (\[t010\]), the phase shifts $\delta(s)$ could be expressed by the formula: $$\begin{aligned}
{\cot}\, \delta(s) = \frac{\re t(s)}{\im t(s)} = -\,
\frac{\lambda^{-1} + \re h(s)}{\im h(s)}
\label{t011}\end{aligned}$$
To express the low-energy parameters in terms of the $T$-matrix solution, it is suitable to expand the function $h(s)$ in a series of $\pp$ terms: $$\begin{aligned}
&& h(s) = h_0 + i\pp h_1 + \pp^2 h_2 + i \pp^3 h_3 + {\cal O}(\pp^4),
\label{t012a}\\
&& \re h(s) = h_0 + \pp^2 h_2 + {\cal O}(\pp^4),
\label{t012b}\\
&& \im h(s) = \pp (h_1 + \pp^2 h_3 + {\cal O}(\pp^3)).
\label{t012}\end{aligned}$$ Using now the definition (\[t03a\]) and (\[t011\]-\[t012\]), we obtain the $a_0$ and $r_0$ parameters: $$\begin{aligned}
a_0 = \frac{h_1}{\lambda^{-1}+h_0},\qquad r_0 =
\frac{2}{h_1}\left[(\lambda^{-1}+h_0)\frac{h_3}{h_1}-h_2\right].
\label{t013}\end{aligned}$$ The bound state condition (\[t03b\]) with the help of (\[t08\]) is now expressed in the form: $$\begin{aligned}
\lambda^{-1} = -h(s=M_b^2).
\label{t014}\end{aligned}$$
Dispersion relations {#sec:disper}
====================
Let us consider now the analytic properties of the BS solution (namely, the function $h$). The simplest choice of the function $g(p_0,p)$ is the [*Yamaguchi*]{} type [@yam]: $$\begin{aligned}
g(p_0,p) = (p_0^2-p^2-\beta^2+i\epsilon)^{-1}.
\label{t21a}\end{aligned}$$
In this case, the function $h$ could be rewritten as $$\begin{aligned}
h(s,\beta) = -\,\frac{i}{4\pi^3}\, \db\, \int dp_0\,\int\, p^2\, dp\
\frac{1}{(\sqrt{s}/2-e_p+i\epsilon)^2-p_0^2}\ \frac{1}{p_0^2-\eb^2+i\epsilon}
\label{t21}\end{aligned}$$ with $\db \equiv \partial / \partial \beta^2$ and $\eb=\sqrt{p^2+\beta^2}$. We have introduced the second argument $\beta$ to extract the explicit dependence of the function $h$ on this parameter.
Analyzing equation (\[t21\]), we could conclude that there are four poles in the complex plane $p_0$, namely: $$\begin{aligned}
&& p^{(1)}_0(s) = \frac{\sqrt{s}}{2} - e_p + i\epsilon \qquad\qquad
p^{(2)}_0(s) = - \frac{\sqrt{s}}{2} + e_p - i\epsilon
\label{t22}\\
&& p^{(3)}_0(s) = - \eb + i\epsilon \qquad\qquad\quad\quad
p^{(4)}_0(s) = \eb - i\epsilon
\nonumber\end{aligned}$$
With the change of the variable $s$, the poles $p^{(1)}_0$ and $p^{(3)}_0$ move, and there could be a situation when two poles “pinch” real $p_0$ axes. It means, the function $h(s)$ has at this $s$-point the leap and imaginary part. First points at which this condition is satisfied (branch points) could be found from the following equations: $$\begin{aligned}
&& p^{(1)}_0(s) = - p^{(1)}_0(s) \qquad \Rightarrow \qquad s_0 = 4m^2,
\label{t23}\\
&& p^{(1)}_0(s) = - p^{(2)}_0(s) \qquad \Rightarrow \qquad s_1 = 4(m+\beta)^2.
\nonumber\end{aligned}$$
Summarizing the situation, one could say that the function $h(s)$ has two cuts starting at points $s_0$ and $s_1$, respectively, and therefore could be written in a dispersion form: $$\begin{aligned}
&& h(s,\beta) = \int\limits_{4m^2}^{+\infty}
\frac{\rho(s^{\prime},\beta)\,ds^{\prime}}{s^{\prime}-s-i\epsilon},
\label{t24}\\
&& \rho(s^{\prime},\beta) = \theta (s^{\prime}-4m^2)
\rho_{el}(s^{\prime},\beta) + \theta
(s^{\prime}-4(m+\beta)^2)\rho_{in}(s^{\prime},\beta)
\nonumber\end{aligned}$$ with two spectral functions $\rho_{el;in}$ ([*el*]{} stands for [*elastic*]{} and [*in*]{} for [*inelastic*]{}) which are connected with the imaginary parts as $$\begin{aligned}
\rho(s^{\prime},\beta) = \frac{1}{\pi} \im h(s^{\prime},\beta) =
\frac{1}{2\pi i} (h-h^{*}).
\label{t25}\end{aligned}$$
To find spectral functions, one should perform $p_0$-integration in (\[t21\]) which gives the following result: $$\begin{aligned}
h(s,\beta) = -\frac{1}{2\pi^2}\,\db\, \int p^2\, dp\,
\frac{1}{s/4-\sqrt{s}e_p+m^2-\beta^2+i\epsilon}
\left[\frac{1}{\sqrt{s}-2e_p+i\epsilon}+\frac{1}{2\eb}\right].
\label{t26}\end{aligned}$$ Taking into account the following symbolic equation: $$\begin{aligned}
\frac{1}{x-x_0 \pm i\epsilon} = \frac{{\cal P}}{x-x_0} \mp i\pi\delta(x-x_0),
\label{t27}\end{aligned}$$ it is easy to find the spectral functions: $$\begin{aligned}
&& s^{\prime} \ge 4m^2
\label{d28a}\\
&& \hskip 10mm
\rho_{el}(s^{\prime},\beta) =
\sqrt{\spr}\sqrt{\spr-4m^2}/(\pi^2(\spr-4m^2+4\beta^2)^2)
\\
&& \spr \ge 4(m+\beta)^2
\label{d28b}\\
&& \hskip 10mm
\rho_{in}(\spr,\beta) =
-(64\beta^6+16\beta^4\spr-192\beta^4m^2-20\beta^2\spr^2
\nonumber\\
&& \hskip 10mm
+192\beta^2m^4-32\beta^2\spr m^2+16m^4\spr+3\spr^3-64m^6-12m^2\spr^2)/
\nonumber\\
&& \hskip 10mm
(2\pi^2\spr(\spr-4m^2+4\beta^2)^2\sqrt{\spr-4(m+\beta)^2}
\sqrt{\spr-4(m-\beta)^2}).
\nonumber\end{aligned}$$
To perform integration (\[t21\]), it is suitable to introduce new variables: $$\begin{aligned}
&& \lmb = \frac{\beta}{m},\quad\quad\qquad\qquad\qquad t=\frac{s}{4m^2}
\label{t211a}\\
&& \rho^2 = 1 - \frac{4m^2}{s} = 1 - \frac{1}{t},\qquad
w^2 = \frac{m^2}{m^2-\beta^2} = \frac{\lmb^2}{1-\lmb^2},
\label{t211b}\\
&& v^2=\frac{m+\beta}{m-\beta}= \frac{1+\lmb}{1-\lmb},
\qquad
\sigma^2 = - \frac{s - 4(m+\beta)^2}{s - 4(m-\beta)^2}=
-\frac{t-(1+\lmb)^2}{t-(1-\lmb)^2}.
\label{t211c}\end{aligned}$$
In case when the following conditions are valid $$\begin{aligned}
&& m > \beta > 0 \qquad \Rightarrow \qquad 1> \lmb > 0, \\
\label{t212a}
&& 4(m+\beta)^2 > s >4m^2 \qquad \Rightarrow \qquad (1+\lmb)^2 > t > 1,
\label{t212b}\end{aligned}$$ the introduced parameters are real and positive: $$\begin{aligned}
1 > \rho^2 > 0,
\qquad\qquad
w^2 > 0,
\\
v^2 > 1 > 0,
\qquad\qquad
\sigma^2 > 0.
\nonumber\end{aligned}$$ The condition (\[t212b\]) means that the second (inelastic) imaginary part does not contribute to the function $\im h(s)$ when the phase shifts are calculated in the region $4(m+\beta)^2 > s >4m^2$, and therefore: $$\begin{aligned}
\im h(s,\beta) = \im h_{el}(s,\beta) =
\frac{\rho(\rho^2-1)(1+w^2)^2}{4m^2\pi(\rho^2+w^2)^2},\quad\quad
\mbox{ if }\quad 4(m+\beta)^2 > s >4m^2.
\label{t213}\end{aligned}$$ Performing integration (\[t24\]), we obtain for the real part of function $h(s)$ (the imaginary part is given by (\[t213\])): $$\begin{aligned}
\re h(s,\beta) &=& h_{el}(s,\beta) + h_{in}(s,\beta),
\label{t215a}
\\
h_{el}(s,\beta) &=&
\label{t215b}
\frac{\rho(\rho^2-1)(1+w^2)^2}
{4m^2\pi^2(\rho^2+w^2)^2}
\ln{\left|\frac{\rho+1}{\rho-1}\right|}
\nonumber\\
&&
-\frac{(\rho^2-1)(1+w^2)^2(w^2-\rho^2)}{4m^2\pi^2(\rho^2+w^2)^2w}\arctan{1/w}
\nonumber\\
&&
-\frac{(\rho^2-1)(1+w^2)}{4m^2\pi^2(\rho^2+w^2)},
\nonumber\\
h_{in}(s) &=&
\label{t215c}
-\frac{(1+\sigma^2)(1+v^2)^2}{32m^2\pi^2(v^4+\sigma^2)}
\ln{\left|\frac{v^2+1}{v^2-1}\right|}
\nonumber\\
&&
-\frac{(1+\sigma^2)(1+v^2)^2}{16m^2\pi^2(v^2-1)^2v(v^2-\sigma^2)^2}
\nonumber\\
&&
\times
(v^6+v^4\sigma^2-4v^4+4v^2\sigma^2-v^2-\sigma^2)\arctan{1/v}
\nonumber\\
&&
+
\frac{(1+\sigma^2)(1+v^2)^3}{16m^2\pi^2\sigma(v^4+\sigma^2)(v^2-\sigma^2)^2\pi^2(v^2-1)^2}
\nonumber\\
&&
\times
(-v^6+2v^6\sigma^2-3v^4\sigma^2+3\sigma^4v^2-2\sigma^4+\sigma^6)\arctan{1/\sigma}
\nonumber\\
&&
-\frac{(1+\sigma^2)(1+v^2)^2}{16m^2\pi^2(v^2-1)(v^2-\sigma^2)}
\nonumber\end{aligned}$$
To find also expressions for low-energy parameters, we should return to (\[t012\])-(\[t013\]) and expand the function $h(s)$ in a series of $\pp$ terms: $$\begin{aligned}
h_0(\beta) &=&
\frac{(1+w^2)}{4m^2\pi^2w^3}(w+(1+w^2)\arctan{1/w})
\label{t216a}\\
&&
+
\frac{5+20v^2+5v^8+14v^4+20v^6}{4m^2\pi^2(v^2-1)^3(3v^2+1)}
\sqrt{\frac{3v^2+1}{v^2+3}}\arctan{1/\sqrt{\frac{3v^2+1}{v^2+3}}}
\nonumber\\
&&
-\frac{(v^2+1)^4}{4m^2\pi^3v(v^2-1)^3}\arctan{1/v}
-\frac{1}{8m^2\pi^2}\ln{\left|\frac{v^2+1}{v^2-1}\right|}
-\frac{(v^2+1)^2}{4m^2\pi^2(v^2-1)^2}
\nonumber\\
h_2(\beta) &=&
-
\frac{(1+w^2)^2}{4m^4\pi^2w^5}(3w+(3+w^2)\arctan{1/w})
\label{t216b}\\
&&
-\frac{1}{4m^4\pi^2(v^2-1)^5(v^2+3)(3v^2+1)^2}
(304v^2+1212v^{12}+3790v^8
\nonumber\\
&&
+304v^{14}+2704v^{10}+29v^{16}+1212v^4+
2704v^6+29)
\nonumber\\
&&
\times
\sqrt{\frac{3v^2+1}{v^2+3}}
\arctan{1/\sqrt{\frac{3v^2+1}{v^2+3}}}
\nonumber\\
&&
+
\frac{(v^4+10v^2+1)(v^2+1)^4}{4m^4\pi^2v(v^2-1)^5}\arctan{1/v}
\nonumber\\
&&
+
\frac{1}{8m^4\pi^2}\ln{\left|\frac{v^2+1}{v^2-1}\right|}
+\frac{11v^8+52v^6+66v^4+52v^2+11)(v^2+1)^2}{8m^4\pi^2(v^2-1)^4(3v^2+1)(v^2+3)}
\nonumber\\
h_1(\beta) &=& \frac{(1+w^2)^2}{4m^3\pi w^4}
\label{t216c}\\
h_3(\beta) &=& -\frac{(1+w^2)^2(4+3w^2)}{8m^5\pi w^6}
\label{t216d}\end{aligned}$$
At this point we have an[*explicit analytic*]{} expressions which connect the parameters of the separable kernel of interaction ($\lambda$ and $\beta$ \[$\lmb$\]) with observables: phase shifts $\delta(s)$ (eq. (\[t011\])), low-energy parameters $a_0$ and $r_0$ (eq. (\[t013\])), and bound state energy (eq. (\[t014\])).
Separable and one-meson exchange kernels of interaction {#sec:ladder}
=======================================================
In this section, we show that the so-called realistic kernel of interaction (meson-nucleon) in the ladder approximation is included in the separable kernel of the Yamaguchi-type g-functions. To achieve this, we introduce a simple approximation for the ladder kernel: $$\begin{aligned}
V(p_0,p;k_0,k) \to {\tilde V}(p_0,p;k_0,k) =
\frac{V(p_0,p;0,0)V(0,0;k_0,k)}{V(0,0;0,0)}.
\label{ns01}\end{aligned}$$ Using the expressions of the kernel for scalar-meson (sc) exchange [@KU72],[@bonn]: $$\begin{aligned}
V_{sc}(p_0,p;k_0,k) =
-\frac{g_{sc}^2}{4\pi}\frac{1}{\pi^2}\frac{1}{4pke_pe_k}
\left[(e_pe_k+m^2)Q_0(z_{\mu})-pkQ_1(z_{\mu})\right],
\label{ns02}\end{aligned}$$ where $z_{\mu}=(p^2+k^2-(p_0-k_0)^2+\mu^2)/(2pk)$, $\mu$ is the exchange meson mass, and $Q_{i}(z)$ are the Legendre functions of second kind, $Q_0(z) = 1/2\ln{(z+1)/(z-1)}$, $Q_1(z) = zQ_0(z) - 1$, it is clear that $$\begin{aligned}
V_{sc}(p_0,p;0,0) = \lim_{k_0\to 0, k\to 0} V_{sc}(p_0,p;k_0,k)
= a_p g_{\mu}(p_0,p),
\label{ns03}\end{aligned}$$ with $$\begin{aligned}
a_p = \frac{g_{sc}^2}{4\pi}\frac{1}{\pi^2}\frac{e_p+m}{2e_p},
\nonumber\\
g_{\mu}(p_0,p) = 1/(p_0^2-p^2-\mu^2).
\label{ns05}\end{aligned}$$ Expressions for $V_{sc}(0,0;k_0,k)$ could be obtained from eqs. (\[ns03\]-\[ns05\]) by the following substitutions: $p_0 \to k_0$ and $p \to k$. To make connection between parameters more clear, we perform the $p/m$-decomposition in function $a_p$ up to ${\cal O}(p^2/m^2)$ term: $$\begin{aligned}
a_p = a_k = \frac{g_{sc}^2}{4\pi}\frac{1}{\pi^2},
\nonumber\\
V_{sc}(0,0;0,0) = -\frac{g_{sc}^2}{4\pi}\frac{1}{\pi^2}\frac{1}{\mu^2}.
\label{ns06}\end{aligned}$$
Using now expression (\[ns01\]), we could write: $$\begin{aligned}
{\tilde V}_{sc}(p_0,p;k_0,k) = -\frac{g_{sc}^2}{4\pi}(\frac{\mu}{\pi})^2
g_{\mu}(p_0,p) g_{\mu}(k_0,k).
\label{ns07}\end{aligned}$$
Comparing the above expressions with the separable form of the kernel introduced by (\[t04\]), we could write the following connection between parameters: $$\begin{aligned}
\beta = \mu,\qquad\qquad
\lambda_{sc} = -\frac{g_{sc}^2}{4\pi}(\frac{\mu}{\pi})^2.
\label{ns08}\end{aligned}$$
Equations (\[ns08\]) are valid also for the vector-meson (vc) exchange kernel with the substitution of $g_{sc} \to g_{vc}$ and $\mu$ by the vector-meson mass and could be derived from the following expression: $$\begin{aligned}
V_{vc}(p_0,p;k_0,k) =
-\frac{g_{vc}^2}{4\pi}\frac{1}{\pi^2}\frac{1}{4pke_pe_k}
\left[-2(2e_pe_k+m^2)Q_0(z_{\mu})\right].
\label{ns09}\end{aligned}$$
The pseudoscalar-meson (ps) exchange kernel of interaction has the form: $$\begin{aligned}
V_{ps}(p_0,p;k_0,k) =
-\frac{g_{sc}^2}{4\pi}\frac{1}{\pi^2}\frac{1}{4pke_pe_k}
\left[-(e_pe_k-m^2)Q_0(z_{\mu})+pkQ_1(z_{\mu})\right],
\label{ns10}\end{aligned}$$ and for $p,k \to 0$ we could write: $$\begin{aligned}
{\tilde V}_{ps}(p_0,p;k_0,k) \sim p^2 k^2 g_{\mu}(p_0,p) g_{\mu}(k_0,k).
\label{ns11}\end{aligned}$$ In this case, the expression for ${\tilde V}_{ps}$ goes to zero for $p,k \to 0$, and it is impossible to find a connection between parameters similar to (\[ns08\]).
We have obtained connections between parameters for various cases of the scalar and vector exchange, which are given in Ref. [@bonn]. Results are given in Table \[ns12\].
$J^P$ $\mu$ (GeV) $g^2/4\pi$ $\lambda$ (GeV$^{2}$)
------- --------------------- ------------- ------------ -----------------------
$0^+$ $NN\delta$ 0.983 0.64 -.06265954891
$0^+$ $NN\sigma^{\prime}$ 0.550 7.07 -.2166930823
$1^-$ $NN\rho$ 0.769 0.43 -.02576448048
$1^-$ $NN\omega$ 0.7826 10.6 -.6577877886
: \[ns12\] Connection between parameters of two kernels.
Results and discussion {#sec:results}
======================
At this moment, we could find internal parameters of the kernel of interaction ($\lambda$, $\beta$) to reproduce experimental values for low-energy parameters $a_{0s}^{exp}, r_{0s}^{exp}$ for the singlet channel ($^1S_0$) and $a_{0t}^{exp}$ and bound state (deuteron) energy $E_d^{exp}$ for the triplet channel ($^3S_1$).
[**In the case of $^1S_0$-channel**]{}, we use equation (\[t013\]) with $a_{0} \equiv a_{0s}^{exp}$ to find $\lambda$: $$\begin{aligned}
{\lambda}^{-1} = (a_{0s}^{exp})^{-1} h_1(\beta) - h_0(\beta).
\label{dis01}\end{aligned}$$ Inserting the obtained expression into equation (\[t014\]) with $r_0
\equiv r_{0s}^{exp}$, we find: $$\begin{aligned}
r_{0s}^{exp} = \frac{2}{h_1(\beta)} \left[ (a_{0s}^{exp})^{-1}
h_3(\beta) - h_2(\beta)\right]. \label{dis02}\end{aligned}$$ Solving the nonlinear equation (\[dis02\]), we could find the value of $\beta$ and, then, using expression (\[dis01\]), the value of $\lambda$.
[**In the case of $^3S_1$-channel**]{}, we obtain $\lambda$ from the bound state condition (\[t014\]) with $E_b \equiv
E_d^{exp}$: $$\begin{aligned}
\lambda^{-1} = - h(s=(M_d^{exp})^2,\beta), \label{dis03}\end{aligned}$$ where $M_d = 2m-E_d$. Inserting the obtained expression into (\[t013\]) with $a_0 \equiv a_{0t}^{exp}$, we find: $$\begin{aligned}
a_{0t}^{exp} = \frac{h_1(\beta)}{h_0(\beta) - h(s=M_d^2,\beta)}.
\label{dis04}\end{aligned}$$ Solving the nonlinear eq. (\[dis04\]), we could find the value of $\beta$ and, then, using expression (\[dis03\]) - the value of $\lambda$.
To solve equations (\[dis02\]) and (\[dis04\]), we use FORTRAN code with the DZEROX subroutine (as a part of CERNLIB package). As a result, we find the following values for $\lambda$ and $\beta$: $$\begin{aligned}
\begin{array}{ll}
\mbox{for $^1S_0$ channel}: & \\
& \lambda = -0.29425404 \mbox{ GeV}^{-2},
\qquad \beta = 0.22412880 \mbox{ GeV},\\
&\\
\mbox{for $^3S_1$ channel}: & \\
& \lambda = -0.79271213 \mbox{ GeV}^{-2},
\qquad \beta = 0.27160579 \mbox{ GeV},\\
\end{array}\end{aligned}$$
The results on the binding energy of the deuteron and the low-energy parameters are shown in Table \[table3\]. The experimental data are taken from [@Kroll].
$1S_0$ $a_{0s}$ (Fm) $r_{0s}$ (Fm)
------------ --------------------- ------------------- -------------------------
Calculated -23.748 2.75
Experiment -23.748 $\pm$ 0.010 2.75 $\pm$ 0.05
$3S_1$ $a_{0t}$ (Fm) $r_{0t}$ (Fm) $E_d$ (MeV)
Calculated 5.424 1.774 2.224644
Experiment 5.424 $\pm$ 0.004 1.759 $\pm$ 0.005 2.224644 $\pm$ 0.000046
: \[table3\] The binding energy and low-energy parameters for singlet and triplet channels.
The phase shifts $\delta_s(s)$ and $\delta_t(s)$ calculated with the above parameters are presented in Fig. \[fig1\]. The experimental data are taken from [@arndt]. As it is seen from the figure, even with the simplest choice of the separable kernel of interaction - rank I with only two parameters $\lambda$ and $\beta$, we find the low-energy parameters of elastic $NN$ scattering $a_s$ and $r_s$ in the singlet channel and $a_t$ and bound state (deuteron) energy $E_d$ in the triplet channel with required accuracy and reproduce phase shifts in a region till $T_{lab} =
100-150$ MeV.
![\[fig1\] The singlet $\delta_s$ and triplet $\delta_t$ phase shifts.](phases.eps){width="110mm"}
To summarize, we recall once again the basic points of our lines of thought and the approximations made in the analysis. We start with the Bethe-Salpeter equation for $T$-matrix and perform the partial-wave decomposition. At this point, we omit all states which have at least one negative-energy nucleon and consider only $\Sp$-wave in the singlet channel and neglect also $\Dpp$-wave and left only $\Spp$-wave in the triplet channel. In order to find the explicit analytic solution for $T$-matrix, we introduce separable ansatz for the kernel of interaction. As a result, we find the $N/D$-form for on-shell $T$-matrix. Taking the simplest choice of g-function in a [*Yamaguchi*]{} form for $NN$-vertex, we analyze the analytic structure of the obtained solution for $T$-matrix and write the dispersion representation of the $T$-matrix. Performing the integration in the dispersion formula, we obtain [*explicit analytic*]{} expressions which connect the parameters of the separable kernel of interaction ($\lambda$ and $\beta$ \[$\lmb$\]) with observables: phase shifts, low-energy parameters and bound state energy. This connection is used to calculate intrinsic parameters of the kernel of interaction and then to find phase shifts.
We find also that the kernel of interaction in the ladder approximation is included into the separable kernel. To obtain this result, we use a simple approximation for the expressions in the one-meson exchange model.
The main result of the paper is the dispersion form of $T$-matrix for elastic $NN$ scattering obtained for the separable kernel of interaction which allows us to perform analytic calculations and to connect explicitly parameters of the kernel and observables. The discussed method could be technically expanded to the multi-rank separable kernels and more complex forms of g-function for $NN$-vertex to achieve required accuracy for bound state energy, low-energy parameters and phase shifts in a wide region of cm laboratory energy. The approach can be easily extended to higher partial-waves for $NN$-scattering and other reactions ($\bar N N$-, $\pi N$- scattering).
[**Acknowledgments.**]{} We would like to thank Prof. Y. Yamaguchi and Prof. Y. Nambu for useful discussions. One of us (V.V.B.) expresses his deep gratitude to the directorate of RCNP for hospitality. The work was supported in part by the RFBR grant [*N*]{} 00-15-96737.
[100]{}
C. Lechanoine-LeLuc, F. Lehar, Rev. of Modern Phys., [**65**]{} (1993), p. 47.
Gari M., Hyuga H., Z.Phys. A [**58**]{} (1976), p.291.
V.V. Burov, V.N. Dostovalov, and S.Eh.Sus’kov, Czech. J. of Phys. [**41**]{} (1991) p. 1139; Particles and Nuclei [**23**]{} (1992), p.721.
S.G. Bondarenko, V.V. Burov, M. Beyer, S.M. Dorkin, Phys. Rev. C [**58**]{} (1998) 3143-3152.
E.E. Salpeter and H.A. Bethe, Phys. Rev. C [**84**]{} (1951) p.1232.
Y. Yamaguchi, Phys. Rev. [**95**]{}, (1954) p.1628; Y. Yamaguchi, Y. Yamaguchi, Phys. Rev. [**95**]{}, (1954) p.1635.
L. Mathelitsch, W. Plessas, W. Schweiger, Phys. Rev. C [**26**]{} (1982) p.65; W. Schweiger, J. Haidenbauer, W. Plessas, Phys. Rev. C [**32**]{} (1985), p.1261.
G. Rupp and J.A. Tjon, Phys. Rev. C37, 1729 (1988); G. Rupp and J.A. Tjon, Phys. Rev. C41, 472 (1990); G. Rupp, Nucl. Phys. A508, 131c (1990); G. Rupp and J.A. Tjon, Phys. Rev. C45, 2133 (1992); G. Rupp, Proc. of PAN XIII, 13th Int. Conf. on Particles and Nuclei, Perugia, June-July 1993, World Scientific, Ed. A. Pascolini, p. 673-675 (1994).
K. Schwarz [*et al.*]{}, Acta Physica Austriaca, [**53**]{} (1981), p.191; J. Fr" ohlich [*et al.*]{}, Phys. Rev. C [**25**]{} (1982), p.2591; J. Fr" ohlich [*et al.*]{}, Phys. Rev. C [**27**]{} (1983), p.265; J. Haidenbauer, and J. Froehlich, Phys. Rev. C[**33**]{}, 456 (1986).
J.J. Kubis, Phys. Rev. D[**6**]{} (1972) p.547.
S.G. Bondarenko, V.V. Burov, M. Beyer, S.M. Dorkin, Few Body Syst. [**26**]{} (2-4) (1999) p.185.
G.E. Brown, A.D. Jackson, “The Nucleon-nucleon interaction”, North-Holland Publishing Conpany, 1976.
R. Machleidt [*et al.*]{}, Phys. Rep. [**149**]{} (1987) p.1.
O. Dumbrajs [*et al.*]{}, Nucl. Phys. B [**216**]{} (1983) p.277.
R. A. Arndt, L. D. Roper, R. A. Bryan, R. B. Clark, B. J. VerWest, and P. Signell, Phys. Rev. D [**28**]{} (1983) p.97; R.A. Arndt, J.S. Hyslop and L.D. Roper, Phys.Rev. D [**35**]{} (1987) p.128.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
In this paper we define a new type of continued fraction expansion for a real number $x \in I_m:=[0,m-1], m\in N_+, m\geq 2$: $$x = \frac{m^{-b_1(x)}}{\displaystyle 1+\frac{m^{-b_2(x)}}{1+\ddots}}:=[b_1(x), b_2(x), \ldots]_m.$$ Then, we derive the basic properties of this continued fraction expansion, following the same steps as in the case of the regular continued fraction expansion. The main purpose of the paper is to prove the convergence of this type of expansion, i.e. we must show that $$x= \lim_{n\rightarrow\infty}[b_1(x), b_2(x), \ldots, b_n(x)]_m.$$
[**Keywords:**]{} [*continued fractions, incomplete quotients*]{}
author:
- |
Ion COLTESCU, Dan LASCU\
“Mircea cel Batran” Naval Academy, 1 Fulgerului,\
900218 Constanta, Romania\
: [email protected], [email protected]
title: A NEW TYPE OF CONTINUED FRACTION EXPANSION
---
INTRODUCTION
============
In this section we make a brief presentation of the theory of regular continued fraction expansions.
It is well-known that the regular continued fraction expansion of a real number looks as follows: $$\frac{1}{\displaystyle a_1+\frac{1}{\displaystyle a_2+\ddots + \frac{1}{a_n + \ddots}}}$$ where $a_n \in N$, $\forall n \in N_+$. We can write this expression more compactly as $$[0;a_1, a_2, \ldots, a_n, \ldots].
\label{eq11}$$ The terms $a_1, a_2, \ldots$ are called the incomplete quotients of the continued fraction. Continued fractions theory starts with the procedure known as Euclid’s algorithm for finding the greatest common divisor. To generalize Euclid’s algorithm to irrational numbers from the unit interval $I$, consider the continued fraction transformation $\tau:I \rightarrow I$ defined by $$\tau(x):=\frac{1}{x} - \left[\frac{1}{x}\right], x \neq 0, \tau(0):=0,
\label{eq12}$$ where $[\cdot]$ denotes the floor (entire) function. Thus, we define $a_1=a_1(x)=\left[\frac{1}{x}\right]$ and $a_n = a_1 (\tau ^{n-1}(x))$, $\forall n \in N$, where $\tau ^0(x)=x$, and $\tau^n(x)=\tau(\tau^{n-1}(x))$. Then, from relation (\[eq12\]), we have: $$x= \frac{1}{\displaystyle a_1+\tau(x)} =
\frac{1}{\displaystyle a_1+\frac{1}{\displaystyle a_2+\tau^2(x)}} = \ldots = [0;a_1,a_2,...,a_n+\tau^n(x)].$$
The metrical theory of continued fractions expansions is about the sequence $(a_n)_{n\in N}$ of its incomplete quotients, and related sequences. This theory started with Gauss’ problem. In modern notation, Gauss wrote that $$\lim_{n\rightarrow\infty} \lambda\left(\left\{x\in[0,1); \tau^n(x)\leq z \right\}\right) = \frac{\log(z+1)}{\log 2}, \ 0\leq z\leq 1,
\label{eq13}$$ where $\lambda$ is the Lebesgue measure. Gauss asked Laplace to prove (\[eq13\]) and to estimate the error-term $r_n(z)$, defined by $r_n(z) := \lambda(\tau^{-n}([0,z])) -
\frac{\log(z+1)}{\log 2}$, $n\geq 1$. (Note that, when we omit the logarithm base, we will consider the natural logarithm.) The first one who proved (\[eq13\]) and at the same time answered Gauss’ question was Kuzmin (1928), followed by L' evy. From that time on, a great number of such Gauss-Kuzmin theorems followed. To mention a few: F. Schweiger (1968), P. Wirsing (1974), K.I. Babenko (1978), and more recently by M. Iosifescu (1992).
Apart from regular continued fractions, there are many other continued fractions expansions: Engel continued fractions, Rosen expansions, the nearest integer continued fraction, the grotesque continued fractions, etc.
ANOTHER CONTINUED FRACTION EXPANSION
====================================
We start this section by showing that any $x \in I_m :=[0, m-1]$, $m\in N_+$, $m\geq 2$, can be written in the form $$\frac{m^{-b_1(x)}}{\displaystyle 1+\frac{m^{-b_2(x)}}{1+\ddots}}:=[b_1(x), b_2(x), \ldots]_m
\label{eq21}$$ where $b_n = b_n(x)$ are integer values, belonging to the set $Z_{\geq-1}:=\{-1,0,1,2,\ldots\}$, for any $n\in N_+$.
[**Proposition 2.1**]{} For any $x \in I_m:=[0,m-1]$, there exist integers numbers $b_n(x) \in \{-1,0,1,2,\ldots\}$ such that $$x = \frac{m^{-b_1(x)}}{\displaystyle 1+\frac{m^{-b_2(x)}}{1+\ddots}}
\label{eq22}$$ [**Proof**]{}. If $x\in [0, m-1]$, then we can find an integer $b_1(x)\in Z_{\geq-1}$ such that $$\frac{1}{m^{b_1(x)+1}}<x<\frac{1}{m^{b_1(x)}}.
\label{eq23}$$ Thus, there is a $p\in[0,1]$ such that $$x = (1-p)\frac{1}{m^{b_1(x)}} + p\frac{1}{m^{b_1(x)+1}} = \frac{m-mp+p}{m} m^{-b_1(x)}.$$ If we set $x_1 = \frac{mp-m}{m-mp+p}$, then we can write $x$ as $$x = \frac{m^{-b_1(x)}}{1+x_1}.$$ Since $x_1\in[0,m-1]$, we can repeat the same iteration and obtain $$x = \frac{m^{-b_1(x)}}{\displaystyle 1+\frac{m^{-b_2(x)}}{1+\ddots}}$$ which completes the proof.
Next, we define on $I_m:=[0,m-1]$, $m\in N_+$, $m\geq 2$, the transformation $\tau_m$ by $$\tau_m: I_m \rightarrow I_m,$$ $$\tau_m(x):=m^{\frac{\log x^{-1}}{\log m}-\left[\frac{\log x^{-1}}{\log m}\right]}-1, x\neq 0, \tau(0):=0,
\label{eq24}$$ where $[\cdot]$ denotes the floor (entire) function.
For any $x \in I_m$, put $$b_n(x)=b_1\left(\tau^{n-1}_m(x)\right), n \in N_+,$$ $$b_1(x) = \left[\frac{\log x^{-1}}{\log m}\right], x \neq 0, b_1(0)=\infty.$$ Let $\Omega_m$ be the set of all irrational numbers from $I_m$. In the case when $x\in I_m\backslash \Omega_m$, we have $$b_n(x) = \infty, \forall n\geq k(x) \geq m,\mbox{ and } b_n(x) \in Z_{\geq -1}, \forall n < k(x).$$ Therefore, in the rational case, the continued fraction expansion (\[eq21\]) is finite, unlike the irrational case, when we have an infinite number of incomplete quotients from the set $\{-1,0,1,2,\ldots\}$.
Let $\omega \in \Omega_m$. We have $$\omega = m^{\log_m \omega} = m ^{-\frac{\log \omega^{-1}}{\log m}} = \frac{m^{-\left[\frac{\log \omega^{-1}}{\log m}\right]}}{m^{\frac{\log \omega^{-1}}{\log m}-\left[\frac{\log \omega^{-1}}{\log m}\right]}} = \frac {m^{-b_1(\omega)}}{1+\tau_m(\omega)}.$$ Since, $$\begin{aligned}
\tau_m(\omega) &=& m^{\log_m\tau_m(\omega)} = m^{-\frac{\log \tau_m^{-1}(\omega)}{\log m}} = \frac{m^{-\left[\frac{\log \tau_m^{-1}(\omega)}{\log m}\right]}}{m^{\frac{\log \tau_m^{-1}(\omega)}{\log m} - \left[\frac{\log \tau_m^{-1}(\omega)}{\log m}\right]}} \nonumber \\
&=&\frac{m^{-b_1(\tau_m(\omega))}}{1+\tau_m(\tau_m(\omega))} = \frac{m^{-b_2(\omega)}}{1+\tau^2_m(\omega)} \nonumber\end{aligned}$$ then, we have $$\omega = \frac{m^{-b_1(\omega)}}{\displaystyle 1+ \frac{m^{-b_2(\omega)}}{1+\tau^2_m(\omega)}} = \ldots = \frac{m^{-b_1(\omega)}}{\displaystyle 1+ \frac{m^{-b_2(\omega)}}{\displaystyle 1+ \ddots + \frac{m^{-b_n(\omega)}}{1+ \tau^n_m(\omega)}}}
\label{eq25}$$ If $[b_1(\omega)] = m^{-b_1(\omega)}$, and $[b_1(\omega), b_2(\omega), \ldots, b_n(\omega)] = \frac{m^{-b_1(\omega)}}{1+[b_2(\omega),b_3(\omega),\ldots,b_n(\omega)]}$, $\forall n\geq 2$, then (\[eq25\]) can be written as $$\begin{aligned}
\omega &=& \left[b_1(\omega) + \frac{\log(1+\tau_m(\omega))}{\log m}\right] = \left[b_1(\omega), b_2(\omega) + \frac{\log(1+\tau^2_m(\omega))}{\log m}\right] = \ldots = \nonumber\\
&=& \left[b_1(\omega), b_2(\omega), \ldots, b_{n-1}(\omega), b_n(\omega) + \frac{\log(1+\tau^n_m(\omega))}{\log m}\right].\nonumber\end{aligned}$$
It is obvious that we have the relations $$\tau_m(\omega) = \frac{m^{-b_2(\omega)}}{1+\tau^2_m(\omega)}, \ldots, \tau^{n-1}_m(\omega) = \frac{m^{-b_n(\omega)}}{1+\tau^n_m(\omega)}, \forall n\in N_+, \forall \omega \in \Omega_m,
\label{eq26}$$
CONVERGENTS. BASIC PROPERTIES
=============================
In this section we define and give the basic properties of the convergents of this continued fraction expansion.
[**Definition 3.1**]{} A finite truncation in (\[eq21\]), i.e. $$\omega_n(\omega):=\frac{p_n(\omega)}{q_n(\omega)} = [b_1(\omega), b_2(\omega), \ldots, b_n(\omega)]_m, n\in N_+
\label{eq31}$$ is called the $n$-th convergent of $\omega$.
The integer valued functions sequences $(p_n)_{n\in N}$ and $(q_n)_{n\in N}$ can be recursively defined by the formulae: $$\begin{aligned}
p_n(\omega) &=& m^{b_n(\omega)}p_{n-1}(\omega) + m^{b_{n-1}(\omega)}p_{n-2}, \forall n\geq 2, \nonumber \\
q_n(\omega) &=& m^{b_n(\omega)}q_{n-1}(\omega) + m^{b_{n-1}(\omega)}q_{n-2}, \forall n\geq 2,
\label{eq32}\end{aligned}$$ with $p_0(\omega)=0$, $q_0(\omega) = 1$, $p_1(\omega)=1$ and $q_1(\omega)=m^{b_1(\omega)}$.
By induction, it is easy to prove that $$p_n(\omega)q_{n+1}(\omega) - p_{n+1}(\omega)q_n(\omega) = (-1)^{n+1}m^{b_1(\omega) + \ldots + b_n(\omega)}, \forall n \in N_+,
\label{eq33}$$ and that $$\frac{m^{-b_1(\omega)}}{\displaystyle 1+ \frac{m^{-b_2(\omega)}}{\displaystyle 1+ \ddots + \frac{m^{-b_n(\omega)}}{1+ t}}} = \frac{p_n(\omega) + tm^{b_n(\omega)}p_{n-1}(\omega)}{q_n(\omega) + tm^{b_n(\omega)}q_{n-1}(\omega)}, \forall n \in N_+, t\geq 0.
\label{eq34}$$
Now, combining the relations (\[eq25\]) and (\[eq34\]), it can be shown that, for any $\omega \in \Omega_m$, we have $$\omega = \frac{p_n(\omega) + \tau_m^n(\omega)m^{b_n(\omega)}p_{n-1}(\omega)}{q_n(\omega) + \tau_m^n(\omega)m^{b_n(\omega)}q_{n-1}(\omega)}, \forall n \in N_+.
\label{eq35}$$
MAIN RESULT
===========
At this moment, we are able to present the main result of the paper, which is the convergence of this new continued fraction expansion, i.e. we must show that $$\omega = \lim_{n\rightarrow\infty}[b_1(\omega), b_2(\omega), \ldots, b_n(\omega)]_m,$$ for any $\omega \in \Omega_m$.
[**Theorem**]{} For any $\omega \in \Omega_m:=I_m \backslash Q$, we have $$\omega - \omega_n(\omega) = \frac{(-1)^n\tau^n_m(\omega)m^{b_1(\omega)+\ldots+ b_n(\omega)}}{q_n(\omega)\left(q_n(\omega)+\tau^n_m(\omega)m^{b_n(\omega)}q_{n-1}(\omega)\right)}, \forall n \in N_+.
\label{eq41}$$ For any $\omega \in \Omega_m$, we have $$\frac{m^{b_1(\omega)+\ldots+b_n}(\omega)}{q_n(\omega)\left(q_{n+1}(\omega)+(m-1)^{n+1}m^{b_{n+1}(\omega)}q_{n}(\omega)\right)} <
|\omega - \omega_n(\omega)|<$$ $$< \frac{1}{\max(F_n, m^{b_1(\omega)+\ldots+b_n(\omega)})}, \forall n\in N_+,
\label{eq42}$$ and $$\lim_{n\rightarrow\infty}\omega_n(\omega) = \omega
\label{eq43}$$ Here $F_n$ denotes the $n$-th Fibonacci number.\
[**Proof.**]{} Using relations (\[eq33\]) and (\[eq35\]), we obtain: $$\begin{aligned}
\omega - \omega_n(\omega) &=& \frac{p_n(\omega) + \tau^n_m(\omega) m^{b_n(\omega)}p_{n-1}(\omega)}{q_n(\omega) + \tau^n_m(\omega)m^{b_n(\omega)}q_{n-1}(\omega)} - \frac{p_n(\omega)}{q_n(\omega)} \nonumber \\
&=& \frac{(-1)^n\tau^n_m(\omega)m^{b_1(\omega)+\ldots+ b_n(\omega)}}{q_n(\omega)\left(q_n(\omega)+\tau^n_m(\omega)m^{b_n(\omega)}q_{n-1}(\omega)\right)}. \nonumber\end{aligned}$$
Next, by (\[eq26\]) and (\[eq41\]), it follows: $$\begin{aligned}
|\omega - \omega_n(\omega)| & =& \frac{\tau^n_m(\omega)m^{b_1(\omega)+\ldots+ b_n(\omega)}}{q_n(\omega)\left(q_n(\omega)+ \tau^n_m(\omega)m^{b_n(\omega)}q_{n-1}(\omega)\right)} \nonumber \\
&=& \frac{m^{-b_{n+1}(\omega)}}{1+\tau^{n+1}_m(\omega)} \cdot \frac{m^{b_1(\omega)+\ldots+ b_n(\omega)}} {q_n(\omega)\left(q_n(\omega)+ \frac{m^{-b_{n+1}(\omega)}}{1+\tau^{n+1}_m(\omega)} m^{b_n(\omega)}q_{n-1}(\omega)\right)} \nonumber \\
&=& \frac{m^{b_1(\omega)+\ldots+ b_n(\omega)}} {m^{b_{n+1}(\omega)}q_n(\omega)\left(q_n(\omega) + \tau^{n+1}_m(\omega)q_n(\omega) + m^{-b_{n+1}(\omega)}m^{b_n(\omega)}q_{n+1}(\omega)\right)} \nonumber \\
&=& \frac{m^{b_1(\omega)+\ldots+ b_n(\omega)}} {q_n(\omega)\left(q_{n+1}(\omega) + \tau^{n+1}_m(\omega)m^{b_{n+1}(\omega)}q_n(\omega)\right)}
\label{eq44}\end{aligned}$$ Now, we know that the Fibonacci numbers are defined by recurrence $$F_{n+1} = F_{n} + F_{n-1}, \forall n \in N_+, \mbox{ and } F_0 = F_1=1.$$ Also, from the recurrence relation (\[eq32\]), we infer that $$p_{n+1} \geq F_{n+1} \mbox{ and } q_n \geq F_n, \forall n \in N_+, n \geq 2.
\label{eq45}$$ Also, we have that $$\begin{aligned}
q_n(\omega) &=& m^{b_n(\omega)} q_{n-1}(\omega) + m^{b_{n-1}(\omega)}q_{n-2}(\omega) \geq m^{b_n(\omega)}q_{n-1}(\omega) \geq \nonumber\\
& \geq& m^{b_n(\omega)} m^{b_{n-1}(\omega)} q_{n-2}(\omega) \geq \ldots \geq m^{b_1(\omega)+\ldots+b_n(\omega)}q_0(\omega). \nonumber\end{aligned}$$ i.e. $$q_n(\omega) \geq m^{b_1(\omega)+\ldots+b_n(\omega)}, \forall n \in N_+.
\label{eq46}$$ Thus, from relations (\[eq45\]) and (\[eq46\]), we have that $$q_n(\omega) \geq \max(F_n, m^{b_1(\omega)+\ldots+b_n(\omega)}), \forall n \in N_+.$$ Now, since the transformation $\tau_m$ belonging to $(0, m-1)$ and from the last two relations, we can show that $$\frac{m^{b_1(\omega)+\ldots+ b_n(\omega)}}{q_n(\omega)\left(q_{n+1}(\omega) + \tau^{n+1}_m(\omega) m^{b_{n+1}(\omega)}
q_n(\omega)\right)} \leq \frac{m^{b_1(\omega)+\ldots+ b_n(\omega)}}{q_n(\omega)q_{n+1}(\omega)}\leq$$ $$\leq \frac{1}{q_n(\omega)} \leq \frac{1}{\max(F_n, m^{b_1(\omega)+\ldots+b_n(\omega)})}.$$ It is obvious that the left inequality is true. Since $\max(F_n,
m^{b_1(\omega)+\ldots+b_n(\omega)})$ is an increasing function, we have $$\lim_{n \rightarrow \infty}\omega_n(\omega) = \omega.$$ The proof is complete.
REMARK
======
This paper is the first one which addresses this type of continued fraction expansion, and will be followed by other papers which will present the metrical theory of this expansion, the principal aim being solving Gauss’ problem.
[\[01\]]{} K. Dajani, C. Kraaikamp, [*Ergodic theory of numbers*]{}, Cambridge University Press, 2002. A.I. Hincin, [*Fractii continue*]{}, Editura Tehnica, Bucuresti, 1960. M. Iosifescu, [*A very simple proof of a generalization of the Gauss-Kuzmin-Lévy theorem on continued fractions, and related questions*]{}, Rev. Roumaine Math. Pures Appl. 37 (1992), 901-914. M. Iosifescu, C. Kraaikamp, [*Metrical theory of continued fractions*]{}, Kluwer Academic, 2002. M. Iosifescu, G.I. Sebe, [*An exact convergence rate in a Gauss-Kuzmin-Lévy problem for some continued fraction expansion*]{}, in vol. Mathematical Analysis and Applications, 90-109. AIP Conf. Proc. 835 (2006), Amer. Inst. Physics, Melville, NY. A.M. Rockett, P. Sz" usz, [*Continued fractions*]{}, World Scientific, Singapore, 1992. P. Sz" usz, [*" Uber einen Kusminschen Satz*]{}, Acta Math. Acad. Sci. Hungar, 12 (1961), 447-453. G.I. Sebe, [*A Wirsing-type approach to some continued fraction expansion*]{}, Int. J. Math. Math. Sci., 12 (2005), 1943-1950. E. Wirsing, [*On the theorem of Gauss-Kuzmin-Lévy and Frobenius-type theorem for function space*]{}, Acta Arithmetica 24 (1974), 507-528.
| {
"pile_set_name": "ArXiv"
} |
1.4cm 1.4cm = 15truecm = 21truecm = -1.3truecm = -2truecm 0.3cm
[**INTERACTING FAMILIES OF CALOGERO-TYPE PARTICLES AND SU(1,1) ALGEBRA**]{}
S. MELJANAC$^{a}$ [[^1]]{}, M. MILEKOVIĆ $^{b}$ [[^2]]{}, A. SAMSAROV$^{a}$[ [^3]]{} and M. STOJIĆ$^{a}$[ [^4]]{}\
$^{a}$ Rudjer Bošković Institute, Bijenička c.54, HR-10002 Zagreb, Croatia\
$^{b}$ Theoretical Physics Department, Faculty of Science, P.O.B. 331, Bijenička c.32,\
HR-10002 Zagreb, Croatia\
We study a one-dimensional model with F interacting families of Calogero-type particles. The model includes harmonic, two-body and three-body interactions. We emphasize the universal SU(1,1) structure of the model. We show how SU(1,1) generators for the whole system are composed of SU(1,1) generators of arbitrary subsystems. We find the exact eigenenergies corresponding to a class of the exact eigenstates of the F-family model. By imposing the conditions for the absence of the three-body interaction, we find certain relations between the coupling constants. Finally, we establish some relations of equivalence between two systems containing F families of Calogero-type particles.\
Keywords: Multispecies Calogero model, SU(1,1) symmetry, Fock space .\
PACS number(s): 03.65.Fd, 03.65.Sq, 05.30.Pr\
Introduction
============
Since its inception, the ordinary Calogero model $^1$ continues to be of interest for both physics and mathematics community$^2$. The model describes $N$ identical (single-species) particles on the line which interact through an inverse-square two-body interaction and are subjected to a common confining harmonic force. The inverse-square potential can be regarded as a pure statistical interaction $^3$ and the model maps to an ideal gas of particles obeying fractional Haldane statistics $^4$. The role of Haldane statistical parameters is played by a universal coupling constant in the two-body interaction. However, in Haldane’s formulation of statistics there is a possibility of having particles of different species with a mutual statistical coupling parameter depending on the species coupled. This suggests the generalization of the single-species Calogero model to the multispecies Calogero model. Distinguishabillity of the species can be introduced by allowing particles to have different masses and different couplings to each other. While the single-species Calogero model is completely solvable $^{1,5}$, very little is known about spectra and wave functions of the multispecies Calogero model $^6$.\
Recently, we used an operator method to analyse a one-dimensional multispecies Calogero model with two- and three-body interactions $^7$. We succeeded in finding a class of, but not all, exact eigenstates and eigenenergies of the model Hamiltonian. The analysis relied heavily on the SU(1,1) algebraic structure of the Hamiltonian and once more stressed the importance of the conformal symmetry of the quantum singular oscillator $^8$. We were also able to generalize the model of Ref.7 to arbitrary dimensions $^9$.\
In the present Letter, which is in a sense a continuation of our investigation of the ordinary Calogero model $^{10}$ and the multispecies Calogero model $^{7,9}$, we turn to the important problem of interacting families of Calogero-type particles in one dimension, a theme which is already announced in $^7$. We consider a model with a potential that generally includes harmonic, two-body and three-body interactions acting between particles belonging to different families, as well as the interaction between particles belonging to the same family with the coupling constant that may be different for different families. In Section 2 we prepare all necessary tools for handling the problem of interacting families. We collect, without rederiving, the main results of the analysis of the one-dimensional multispecies Calogero model $^7$. In Section 3 we apply these results to the case of two interacting families of Calogero particles. We display the model Hamiltonian and find the ground state energy. We construct generators of SU(1,1) algebra for interacting families and underline the importance of the dilatation part of the algebra, i.e. generator $T_0$. Furthermore, by imposing the conditions for the absence of the three-body interaction in the initial Hamiltonian, we find certain relations between the coupling constants. In Section 4 we extend these results to three and more interacting families of Calogero particles. We show that the underlying SU(1,1) structure is universal, i.e. holds for an arbitrary number of families, arbitrary masses of Calogero particles and arbitrary coupling constants. We particularly show how to obtain SU(1,1) generators of the whole system from SU(1,1) generators of arbitrary subsystem, i.e. we establish composition rules for SU(1,1) generators. We also find the ${\em {exact}}$ eigenenergies corresponding to a class of the ${\em {exact}}$ eigenstates of the model with F interacting families. We discuss the relations between the coupling constants in the case when a three-body interaction vanishes. Finally, we establish some relations of equivalence between two systems containing F families. Section 5 is a short conclusion.
A multispecies Calogero model: main results
===========================================
The model of Ref.7 is specified by masses of particles, $ m_{i}$, and the coupling constants $\omega$ and $\nu_{ij},\; i,j = 1,2,...,N$. The Hamiltonian is $$H(\omega) = -\frac{1}{2}\sum_{i=1}^{N}\frac{1}{m_{i}}\frac{{\partial}^{2}}
{\partial x_{i}^{2}} + \frac{{\omega}^{2}}{2}\sum_{i=1}^{N }m_{i} x_{i}^{2}
+ \frac{1}{4} \sum_{i \neq j }\frac{{\nu}_{ij}({\nu}_{ij}-1)}
{{(x_{i}-x_{j})}^{2}}(\frac{1}{m_{i}} + \frac{1}{m_{j}}) +$$ $$+\frac{1}{2}\sum_{i \neq j, i \neq k, j \neq k
}\frac{{\nu}_{ij}{\nu}_{jk}}
{m_{j}(x_{j}-x_{i}) (x_{j}-x_{k})}.$$ The ground state wave function is of the Calogero type: $$\Psi_{0}(x_{1},...,x_{n})= \prod_{i<j}|x_{i}-x_{j}|^{\nu_{ij}}
e^{-\frac{\omega}{2}\sum_{i=1}^{N}
m_{i}x_{i}^{2}}\equiv
\Delta e^{-\frac{\omega}{2}\sum_{i=1}^{N}
m_{i}x_{i}^{2}}$$ and the corresponding ground state energy is $$E_0=\omega \epsilon_{0} = \omega ( \frac{N}{2} + \sum_{i < j}\nu_{ij} ).$$ When all couplings $\nu_{ij}$ are equal, Eq.(3) reduces to the well-known Calogero result $\epsilon_{0}=\frac{N}{2}+ \nu \frac{N (N-1)}{2}$.\
After performing a similarity transformation $$\begin{array}{l}
\tilde{H}(\omega) = {\Delta}^{-1} H(\omega) \Delta ,\\
\tilde{\Psi} = {\Delta}^{-1} \Psi ,
\end{array}$$ one obtains a non-Hermitean Hamiltonian $\tilde{H}(\omega)$ with a hidden three-body interaction: $$\tilde{H}(\omega) = -\frac{1}{2}\sum_{i=1}^{N}\frac{1}{m_{i}}\frac{{\partial}^{2}}
{\partial x_{i}^{2}} + \frac{{\omega}^{2}}{2}\sum_{i=1}^{N }m_{i} x_{i}^{2}
- \frac{1}{2} \sum_{i \neq j }\frac{{\nu}_{ij}}
{(x_{i}-x_{j})}(\frac{1}{m_{i}} \frac{\partial}{{\partial} x_{i}}
- \frac{1}{m_{j}} \frac{\partial}{{\partial} x_{j}})=$$ $$={\omega}^{2}T_{+}-T_{-},$$ where $$T_{-} = -\tilde{H}(\omega = 0) ,\qquad T_{+} = \frac{1}{2}\sum_{i=1}^{N }m_{i} x_{i}^{2} ,$$ $$T_{0} = \frac{1}{2} (\sum_{i=1}^{N} x_{i}\frac{\partial}{{\partial} x_{i}}
+ \epsilon_{0}).$$ The set of operators $\{T_{\pm},T_0\}$ satisfy the SU(1,1) algebra: $$[T_{-},T_{+}] = 2T_{0}, \;\;\;\; [T_{0},T_{\pm}] = \pm T_{\pm}.$$ Note that $T_0$ serves as a dilatation operator. One can deduce that $$T_{0}\Delta = ( \frac {1}{2}\sum_{i<j} \nu_{ij} + \frac{\epsilon_{0}}{2} )\Delta$$ $$T_{-}\Delta = 0.$$ It is convenient to introduce the centre-of-mass coordinate $ X = \frac{1}{M} \sum_{i=1}^{N}m_{i} x_{i} $ ( where $ M = \sum_{i=1}^{N}m_{i}$ ) and relative coordinates $ \xi_{i} = x_{i}-X $. In terms of these coordinates, the Hamiltonian $\tilde{H}(\omega)$, Eq.(4), separates into parts which describe its centre-of-mass motion (CM) and its relative motion (R), namely $\tilde{H}(\omega)=\tilde{H}(\omega)_{CM} + \tilde{H}(\omega)_{R}$. In the same way one can split the generators $ T_{\pm}$ and $ T_{0}$ into the centre-of-mass and relative parts, i.e. $T_{\pm,0}=
T_{\pm,0_{(CM)}} + T_{\pm,0_{(R)}}$.\
In the next section we apply these results to the case of two interacting families.
Two interacting families
==========================
Let us consider two families, ${\cal {F}}_1$ and ${\cal {F}}_2$, of Calogero particles. The first one, denoted by $ {\cal {F}}_1= \{m_1, \nu_1, N_1\}$, is described by $ N_{1} $ particles of mass $m_{1}$, the coupling constant $\nu_{1}$ and the coordinates of the particles are $\{x_i\}=\{x_{1},x_{2},...,x_{N_{1}}\}$. Similarly, the second one, denoted by ${\cal {F}}_2 =\{m_2, \nu_2, N_2\}$, is described by $ N_{2} $ particles of mass $m_{2}$, the coupling constant $\nu_{2} $ and the coordinates of the particles are $\{ z_{\alpha}\}=\{z_{1},z_{2},...,z_{N_{2}}\}$. The interaction strength between the first and the second family is $\nu_{12} = \kappa$.\
The full Hamiltonian now reads $$H(\omega) = H_{1}(\omega) + H_{2}(\omega) + H_{int},$$ where $H_{int}$ is given by $$H_{int} = \frac{1}{4} \sum_{i}^{N_{1}}\sum_{\alpha }^{N_{2}}
\frac{\kappa(\kappa -1)}{(x_{i}-z_{\alpha})^{2}}\left(
\frac{1}{ m_1} + \frac{1} { m_2 } \right) +$$
$$+ \frac{1}{4}\sum_{i}^{N_{1}} \sum_{\alpha \neq \beta}^{N_{2}}
\left( \frac{{\kappa}^{2}}
{m_{1}(x_{i}-z_{\alpha}) (x_{i}-z_{\beta})}\right) +
\frac{1}{2}\sum_{i}^{N_{1}} \sum_{\alpha \neq \beta}^{N_{2}} \left(\frac{\nu_2 \kappa}
{m_{2} (z_{\alpha}-x_{i}) (z_{\alpha}-z_{\beta})} \right) +$$
$$+ \frac{1}{4}\sum_{i \neq j}^{N_{1}} \sum_{\alpha}^{N_{2}}
\left( \frac{{\kappa}^{2}}
{m_{2} (z_{\alpha} - x_{i}) (z_{\alpha} - x_{j})} \right )+
\frac{1}{2}\sum_{i \neq j}^{N_{1}} \sum_{\alpha}^{N_{2}} \left( \frac{\nu_1 \kappa}
{m_{1} (x_{i}-z_{\alpha}) (x_{i}-x_{j})} \right ),$$
and $H_1(\omega)$ ( $H_2(\omega)$ ) are Calogero Hamiltonians, Eq.(1), for the first and the second family, respectively.\
The corresponding ground state wave function of the Hamiltonian (8) is $$\Psi_{0}(x_{1},...,x_{N_{1}},z_{1},...,z_{N_{2}}) = \prod_{i,\alpha}(x_{i}-z_{\alpha})^{\kappa}\Psi_{0,1}(x_{1},...,x_{N_{1}})
\Psi_{0,2}(z_{1},...,z_{N_{2}})$$ $$\equiv\Delta_{12}\Psi_{0,1}(x_{1},...,x_{N_{1}})
\Psi_{0,2}(z_{1},...,z_{N_{2}}),$$ where $\Psi_{0,1}$ and $\Psi_{0,2}$ are the Calogero ground states (when $\kappa =0$), Eq.(2), for the families ${\cal {F}}_1$ and ${\cal {F}}_2$, respectively.\
We can perform a similarity transformation with a $\Delta_{1}\Delta_{2}$ part of the full Jastrow prefactor $\Delta=\Delta_{1}\Delta_{2}\Delta_{12}$ in (8,10), to obtain $${\Delta_{1}}^{-1}{\Delta_{2}}^{-1} H(\omega) \Delta_{1}\Delta_{2} =
\tilde{H}_{1}(\omega) + \tilde{H}_{2}(\omega) + H_{int},$$ $${\Delta_{1}}^{-1}{\Delta_{2}}^{-1}\Psi_{0} =
\prod_{i,\alpha}(x_{i}-z_{\alpha})^{\kappa}\tilde{\Psi}_{0,1}
\tilde{\Psi}_{0,2}.$$ The ground state energy of the Hamiltonian (8) can be split into three terms: $$\epsilon_{0} = \epsilon_{0,1} + \epsilon_{0,2} + \kappa N_{1}N_{2},$$ describing the ground state energies of each family and the interaction between them, respectively.\
For each family, one can define $SU(1,1)$ generators $T_{\pm}^{(I)}, T_{0}^{(I)}$, $I=1,2$. These two sets of generators, i.e. the corresponding $SU(1,1)$ algebras, mutually commute. From the following two relations: $$T_{0} = T_{0}^{(1)} + T_{0}^{(2)} + \frac{1}{2} \kappa N_{1}N_{2},$$ $$T_{0}\Delta_{12} = \frac{1}{2}( \kappa N_{1}N_{2} + \epsilon_{0})\Delta_{12},$$ we find $$(T_{0}^{(1)} + T_{0}^{(2)}) \Delta_{12} = \frac{\epsilon_{0}}{2}\Delta_{12}.$$ Furthermore, from Eq. (7) and after multiplication by ${\Delta_{1}}^{-1}{\Delta_{2}}^{-1}$ from the left, it follows that $$T_{-}\Delta_{12} \equiv (T_{-}^{(1)} + T_{-}^{(2)} - H_{int})\Delta_{12} = 0.$$ Note that $ T_{+} = T_{+}^{(1)} + T_{+}^{(2)}$.
As we have already shown in Ref.7, for the general $\nu_{ij}$ and $m_j$ the three-body interactions in the initial Hamiltonian (1) vanish identically if the following conditions are satisfied for all triples of indices $i,j,k$: $$\frac{\nu_{ij} \nu_{jk}}{m_{j}} = \frac{\nu_{ji} \nu_{ik}}{m_{i}} =
\frac{\nu_{ik} \nu_{kj}}{m_{k}}.$$ In this case, the Hamiltonian contains the two-body interactions (i.e. inverse-square interactions) only. The unique solution of these conditions is $\nu_{ij}=\lambda \, m_i\, m_j$, $\lambda $ being some universal constant.
In our two-family system this corresponds to the condition $$\nu_{ij} = \lambda m_{i} m_{j} = \kappa, \qquad \forall \, i,j$$ or explicitly $$\nu_{1} = \lambda m_{1}^{2}, \qquad \nu_{2} = \lambda m_{2}^{2},\qquad
\nu_{12}= \kappa = \lambda m_{1}m_{2},$$ from which it follows $$\nu_{1} \nu_{2} = {\kappa}^{2} ,$$ $$\nu_{2} = {(\frac{m_{2}}{m_{1}})}^{2}\nu_{1}.$$ Note that Eqs.(16-18) imply that the couplings $\nu_{1}, \nu_{2}$ and $\kappa $ have to be simultaneously positive, negative or zero.\
The connection between the coupling constants $\{\nu_{1},\nu_{2},\kappa\}$, Eqs.(18), is ascribed to the weak-strong coupling duality in Ref.11, but it is [*[de facto]{}*]{} a simple consequence of the absence of the three-body interaction in the starting Hamiltonian (1). We also point out that all the above relations for $T_{0}^{(1)}+T_{0}^{(2)}$ and $T_{-}^{(1)} + T_{-}^{(2)}$ (Eqs.(13,14)) hold generally for arbitrary masses $m_{1},m_{2}$ and arbitrary coupling constants $ \nu_{1}, \nu_{2}, \kappa$, i.e. irrespectively of the presence/absence of the three-body interaction.\
Three and more interacting families, exact eigenstates and equivalences between models
======================================================================================
We extend the above analysis to the case of three families, ${\cal {F}}_1$, ${\cal {F}}_2$ and ${\cal {F}}_3$, of Calogero particles. The families are characterized by mutually distinct numbers of particles, masses of particles and different coupling constants. One can immediatelly generalize the results (13,14): $$\Delta = \Delta_{1}\Delta_{2}\Delta_{3}\Delta_{12}\Delta_{13}\Delta_{23},$$ $$T_{0} = T_{0}^{(1)} + T_{0}^{(2)} + T_{0}^{(3)} + \frac{1}{2}(\epsilon_{0}
- \epsilon_{0,1} - \epsilon_{0,2} - \epsilon_{0,3}),$$ $$T_{+} = T_{+}^{(1)} + T_{+}^{(2)} + T_{+}^{(3)},$$ $$T_{-} = T_{-}^{(1)} + T_{-}^{(2)} + T_{-}^{(3)} - H_{int},$$ $$(T_{0}^{(1)} + T_{0}^{(2)} + T_{0}^{(3)})\Delta_{12}\Delta_{13}\Delta_{23}
= \frac{\epsilon_{0}}{2}\Delta_{12}\Delta_{13}\Delta_{23},$$ $$(T_{-}^{(1)} + T_{-}^{(2)} + T_{-}^{(3)} - H_{int})
\Delta_{12}\Delta_{13}\Delta_{23} = 0.$$ For the initial $ N $ -body multispecies Calogero model we can write composition laws for the SU(1,1) generators: $$T_{0} = \sum_{i = 1}^{N} T_{0}^{(i)} + \frac{1}{2}\sum_{i<j}\nu_{ij},$$ $$T_{+} = \sum_{i = 1}^{N}T_{+}^{(i)}, \qquad T_{-} = \sum_{i = 1}^{N}T_{-}^{(i)}- H_{int},$$ from which it follows $$T_{0}\Delta = \frac{1}{2}(\sum_{i<j}\nu_{ij} + \epsilon_{0})\Delta,$$ $$(\sum_{i = 1}^{N} T_{0}^{(i)})\Delta = \frac{\epsilon_{0}}{2} \Delta,$$ $$(\sum_{i = 1}^{N}T_{-}^{(i)} - H_{int})\Delta = 0.$$ These relations are general, valid for an arbitrary number of families F (i.e. for an arbitrary partition of a multispecies Calogero model), and for an arbitrary choice of masses $ m_{i}$ and coupling constants $ \nu_{ij}$.
The infinite set of exact eigenstates of the Hamiltonian (1) can be constructed by applying ladder operators $$A_{1}^{\pm} = \frac{1}{\sqrt{2}}(\sqrt{M\omega} X \mp
\frac{1}{\sqrt{M\omega}} \frac{\partial}{\partial X})$$ and $$B_{2}^{\pm}= \frac{1}{2}(\omega T_{+} + \frac{T_{-}}{\omega}) \mp T_{0}- \frac{1}{2}{A_{1}^{\pm}}^{2}$$ to the vacuum $$\tilde{\Psi}_{0}(x_1,x_2,\cdots x_N)
=\tilde{\Psi}_{0}(X)\tilde{\Psi}_{0}(\xi_1, \xi_2 \cdots \xi_N)=
e^{-\frac{M\omega}{2}X^2} e^{-\frac{\omega}{2}\sum^{N}_{i=1} m_i \xi_i^2} .$$ The exact eigenstates (corresponding to the center-of-mass states and global dilatation states) are $$\tilde{\Psi}_{n_{1}n_{2}} = {A_{1}^{+}}^{n_{1}}{B_{2}^{+}}^{n_{2}},
\tilde{\Psi}_{0}. \qquad n_{1}, n_{2} = 0,1,2,,,$$ The exact eigenenergies corresponding to these states are ($I,J=1,2,...F$) $$E_{n_{1}n_{2}}= \omega(n_{1} + 2n_{2} + \epsilon_{0}),$$ $$\epsilon_0=\sum_{I=1}^{F} \epsilon_{0,I} + \sum_{I<J; I,J=1 }^{F} \epsilon_{0,IJ},$$ $$\epsilon_{0,I}= \frac{N_I}{2} + \nu_I \frac{N_I (N_I-1)}{2} ,$$ $$\epsilon_{0,IJ}=\nu_{IJ}\, N_I \, N_J .$$ In the special case, when there is no three-body interaction (i.e. relations (15) are satisfied), we can identify $$\nu_I = \lambda m_I^2, \qquad \forall I=1,2...F,$$ $$\nu_{IJ}=\lambda m_I m_J , \qquad \forall I,J=1,2...F.$$ Since the masses are positive, the couplings $\nu_I$ and $\nu_{IJ}$ have the same sign, depending on the sign of the free parameter $\lambda$.
Now we establish some relations of equivalence between the two systems containing F families of Calogero particles.
[**[Case 1.]{}**]{} [*[Complete equivalence of the two systems.]{}*]{}\
Let $ {\em {S}}=\{\omega, m_I, \nu_I, \nu_{IJ}, N_I\}$ and ${\em {S'}}=\{\omega, m'_I, \nu'_I, \nu'_{IJ}, N'_I\}$ be two Calogero systems with F families. We call them completely equivalent if $$\epsilon_{0,I}=\epsilon'_{0,I}, \qquad \epsilon_{0,IJ}=\epsilon'_{0,IJ}, \qquad N_I=N'_I.$$ These conditions imply $$\nu_I=\nu'_I, \qquad \nu_{IJ}=\nu'_{IJ}.$$
[**[Case 2.]{}**]{} [*[Partial equivalence of the two systems.]{}*]{}\
We call the two systems $ {\em {S}}$ and ${\em {S'}}$ partially equivalent if $$\epsilon_{0}=\epsilon'_{0} ,$$ while the number of particles, $N$ and $N'$, may be the same or different. For example, in the case of one-family systems ($F=1$) and $N\neq N'$, the above condition implies that $$N + \nu N (N-1) =N' + \nu' N' (N'-1).$$
[**[Case 3.]{}**]{} [*[Special case: the single system.]{}*]{}\
Consider a single system with F families of Calogero particles. We can demand that $$\epsilon_{0,1}=\epsilon_{0,I}, \qquad \epsilon_{0,12}=\epsilon_{0,IJ}, \qquad \forall I,J=1,2...F.$$ In the case of the two-family system ($F=2$), we have $$N_1 + \nu_1 N_1 (N_1-1) =N_2 + \nu_2 N_2 (N_2-1).$$ We obtain very interesting relations between the couplings $\nu_1$ and $ \nu_2$ if we impose the strong-weak duality condition on the couplings, namely $\nu_1 \nu_2=1$. (We fix $\kappa^2 =1$ in Eq.(18)).\
The quadratic equation (28) then has two solutions: $$(i) \qquad \nu_1=\frac{ N_2 - 1 }{N_1 - 1 } > 0,$$ $$(ii) \qquad \nu_1= - ( \frac{N_2}{N_1} ) < 0.$$ Their physical implications are summarized in Table 1.
Table 1. $$
[ccccc]{} $\nu_1$ & $\kappa$ & $\lambda$ & $\epsilon_0$ & Comments\
\
$\frac{ N_2 -1 }{ N_1-1 }$ & +1 & $\lambda >0 $ & $2 N_1 N_2 > 0$ & Physical solution, no three-body interaction.\
\
$- ( \frac{N_2}{N_1} ) $ & +1 & $ -$ & $N_1 + N_2 > 0$ & Physical solution, with a three-body interaction.\
\
$\frac{ N_2 -1 }{ N_1-1 }$ & - 1 & $ - $ & $ 0 $ & Unphysical solution, with a three-body interaction.\
\
$- ( \frac{N_2}{N_1} ) $ & -1 & $\lambda < 0 $& $ N_1 + N_2 - 2 N_1 N_2 < 0$ & Unphysical solution, no three-body interaction.\
\
$$ A few remarks are in order.\
[*[Remark 1]{}*]{}. Notice that solution $(i)$ requires $N_{1,2} \geq 2$.\
[*[Remark 2]{}*]{}. If the generalized strong-weak duality condition is imposed, i.e. $\nu_I \nu_J =1$, ($I \neq J=1,2, ...F $), then it follows that it can be satisfied for $F=2$ only.\
[*[Remark 3]{}*]{}. In Refs.7 and 10, we showed that there existed a critical point $\epsilon_{0R}=0$ at which the system described by $\tilde{H}(\omega)_R$ collapsed completely, i.e. the relative momenta, the relative energy and the relative coordinates were all zero at this critical point. The ground state was a square-integrable function only for $\epsilon_{0R}>0$. This is the reason why we ascribe the term ’unphysical’ to the last solutions in Table 1.
Conclusion
==========
In this Letter we have studied the most general Calogero model on the line with a three-body interaction possessing an arbitrary number of mutually interacting families of Calogero particles. We have found the exact eigenenergies corresponding to a class of the exact eigenstates of the model. We have established relations of equivalence between two systems with F families which imply a certain connection between the coupling constants. Particularly interesting appear the relations between the coupling constants in the single system with F families of Calogero particles when a strong-weak duality condition is imposed. We have paid special attention to the SU(1,1) structure of the model. We have found certain relations between SU(1,1) generators that are universal for all choices of masses and coupling constants. We particularly show how to obtain SU(1,1) generators of the whole system from SU(1,1) generators of arbitrary subsystem. Moreover, the same relations are valid for an arbitrary number of dimensions and for all potentials that behave as a kinetic energy term under the dilatation represented by the generator $T_0$. There is only one difference between one and higher dimensions. In the case of one dimension, one can exclude the three-body interaction between particles from the beginning, while there is no known way how to do this in dimensions higher than one. Our results can also be extended to other systems with the underlying conformal or superconformal symmetry $^{12}$.
[**Acknowledgment**]{}\
This work was supported by the Ministry of Science and Technology of the Republic of Croatia under contracts No. 0098003 and No. 0119261.
[99]{}
F. Calogero, J. Math. Phys. [**10**]{}, 2191, 2197 (1969) ;\
F. Calogero, J. Math. Phys. [**12**]{}, 419 (1971);\
B. Sutherland, J. Math. Phys. [**12**]{}, 246 (1971). M. Lassigg, Phys. Rev. Lett. [**77**]{}, 526 (1996);\
N.F.Johnson, L.Quiroga, Phys. Rev. Lett. [**74**]{}, 4277 (1995) ;\
H. Azuma, S. Iso, Phys. Lett. B [**331**]{}, 107 (1994) ;\
N. Kawakami,S.K.Yang, Phys. Rev. Lett. [**67**]{}, 2493 (1991);\
G.W. Gibbons, P.K. Townsend, Phys. Lett. B [**454**]{}, 187 (1999);\
D.Birmingham, K.S. Gupta, S. Sen, Phys. Lett. B [**505**]{}, 191 (2001);\
H. Ujino, M. Wadati, J. Phys. Soc. Jpn. [**65**]{}, 653 (1996);\
L. Lapointe, L. Vinet, Comm. Math. Phys. [**178**]{}, 425 (1996). D. Bernard, Y.-S.Wu, ArXiv cond-mat/9404025;\
M.V.N.Murthy, R. Shankar, Phys. Rev. Lett. [**73**]{}, 1574 (1994). F.D.M. Haldane, Phys. Rev. Lett. [**67**]{}, 937 (1991);\
Y.S.Wu, [*ibid.*]{} [**73**]{}, 922 (1994). A.P. Polychronakos, Phys. Rev. Lett. [**69**]{}, 703 (1992);\
L. Brink, T.H. Hansson, M. Vasiliev, Phys. Lett. B [**286**]{}, 109 (1992);\
L. Brink, T.H. Hansson, S.E. Konstein, M. Vasiliev, Nucl. Phys. B [**384**]{}, 591 (1993). P. Forrester, J. Phys. A: Math. Gen. [**25**]{}, L607 (1992) ;\
C. Furtlehner, S. Ouvry, Mod.Phys.Lett. B[**9**]{}, 503 (1995);\
D. Sen, Nucl. Phys. B [**479**]{}, 554 (1996);\
A. Dasniers de Veigy, Nucl. Phys. B [**483**]{}, 580 (1997);\
S. Maskhevich, Phys. Lett. A [**233**]{}, 30 (1997). S. Meljanac, M. Mileković , A.Samsarov, Phys. Lett. B[**573**]{}, 202 (2003). V. de Alfaro, S. Fubini, G. Furlan, Nuovo Cimento A [**34**]{}, 569 (1976);\
A.M. Perelomov, [*[Generalized Coherent States and Their Applications]{}*]{} (Springer, 1986). S. Meljanac, M. Mileković , A.Samsarov, “A multispecies model in arbitrary dimensions” ( in preparation ). S. Meljanac, M. Mileković , M. Stojić , Eur. Phys. J. C [**60**]{}, 331 (2002);\
V. Bardek, L. Jonke, S. Meljanac, M. Milekovi, Phys.Lett. B [**531**]{}, 311 (2002). I. Andrić , D. Jurman, Phys.Lett. A [**313**]{}, 252 (2003). P.K.Ghosh, J.Phys.A [**34**]{}, 5583 (2001).
[^1]: e-mail: [email protected]
[^2]: e-mail: [email protected]
[^3]: e-mail:[email protected]
[^4]: e-mail:[email protected]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We provide complete characterizations, on Banach spaces with cotype 2, of those linear operators which happen to be weakly mixing or strongly mixing transformations with respect to some nondegenerate Gaussian measure. These characterizations involve two families of small subsets of the circle: the countable sets, and the so-called *sets of uniqueness* for Fourier-Stieltjes series. The most interesting part, the sufficient conditions for weak and strong mixing, is valid on an arbitrary (complex, separable) Fréchet space.'
address:
- 'Clermont Université, Université Blaise Pascal, Laboratoire de Mathématiques, BP 10448, F-63000 Clermont-Ferrand - CNRS, UMR 6620, Laboratoire de Mathématiques, F-63177 Aubière.'
- 'Laboratoire de Mathématiques de Lens, Université d’Artois, Rue Jean Souvraz S. P. 18, 62307 Lens.'
author:
- Frédéric Bayart
- Étienne Matheron
title: |
Mixing operators\
and small subsets of the circle
---
Introduction
============
A basic problem in topological dynamics is to determine whether a given continuous map $T:X\to X$ acting on a topological space $X$ admits an ergodic probability measure. One may also ask for stronger ergodicity properties such as weak mixing or strong mixing, and put additional constraints on the measure $\mu$, for example that $\mu$ should have no discrete part, or that it should belong to some natural class of measures related to the structure of the underlying space $X$. Especially significant is the requirement that $\mu$ should have *full support* ( $\mu (V)>0$ for every open set $V\neq\emptyset$) since in this case any ergodicity property implies its topological counterpart. There is, of course, a huge literature on these matters since the classical work of Oxtoby and Ulam ([@OU]).
In recent years, the above problem has received a lot of attention in the specific setting of *linear* dynamics, when the transformation $T$ is a continuous linear operator acting on a topological vector space $X$ ([@F], [@BG2], [@BG3], [@BoGE], [@Sophie]). The main reason is that people working in linear dynamics are mostly interested in studying *hypercyclic* operators, operators having dense orbits. When the space $X$ is second-countable, it is very easy to see that if a continuous map $T:X\to X$ happens to be ergodic with respect to some Borel probability measure $\mu$ with full support, then almost every $x\in X$ (relative to $\mu$) has a dense $T$-orbit. (In fact, one can say more: it follows from Birkhoff’s ergodic theorem that almost all $T$-orbits visit every non-empty open set along a set of integers having positive lower density. In the linear setting, an operator having at least one orbit with that property is said to be *frequently hypercyclic*. This notion was introduced in [@BG3] and extensively studied since then; see the books [@BM] and [@GP] for more information). Hence, to find an ergodic measure with full support is an efficient way of showing that a given operator is hypercyclic, which comes as a measure-theoretic counterpart to the more traditional Baire category approach.
Throughout the paper, we shall restrict ourselves to the best understood infinite-dimensional measures, the so-called *Gaussian* measures. Moreover, the underlying topological vector space $X$ will always be a complex separable Fréchet space. (The reason for considering *complex* spaces only will become clear in the next few lines). In this setting, a Borel probability measure on $X$ is Gaussian if and only if it is the distribution of an almost surely convergent random series of the form $\xi=\sum_0^\infty g_n x_n$, where $(x_n)\subset X$ and $(g_n)$ is a sequence of independent, standard complex Gaussian variables. Given any property (P) relative to measure-preserving transformations, we shall say that an operator $T\in\mathfrak L(X)$ has property (P) *in the Gaussian sense* if there exists some Gaussian probability measure $\mu$ on $X$ with full support with respect to which $T$ has (P). The problem of determining which operators are ergodic in the Gaussian sense was investigated by E. Flytzanis ([@F]), in a Hilbert space setting. The fundamental idea of [@F] is that one has to look at the *$\TT$-eigenvectors* of the operator, the eigenvectors associated with eigenvalues of modulus $1$: roughly speaking, ergodicity is equivalent to the existence of “sufficiently many $\TT$-eigenvectors and eigenvalues". This is of course to be compared with the now classical eigenvalue criterion for hypercyclicity found by G. Godefroy and J. Shapiro ([@GS]), which says in essence that an operator having enough eigenvalues inside and outside the unit circle must be hypercyclic.
The importance of the $\TT$-eigenvectors is easy to explain. Indeed, it is almost trivial that if $T\in\mathfrak L(X)$ is an operator whose $\TT$-eigenvectors span a dense subspace of $X$, then $T$ admits an invariant Gaussian measure with full support: choose a sequence of $\TT$-eigenvectors $(x_n)_{n\geq 0}$ (say $T(x_n)=\lambda_n x_n$) with dense linear span such that $\sum_0^\infty\Vert x_n\Vert<\infty$ for every continuous semi-norm $\Vert\,\cdot\,\Vert$ on $X$, and let $\mu$ be the distribution of the random variable $\xi=\sum_0^\infty g_n x_n$. That $\mu$ is $T$-invariant follows from the linearity of $T$ and the rotational invariance of the Gaussian variables $g_n$ ($\mu\circ T^{-1}\sim\sum_0^\infty g_n T(x_n)=\sum_0^\infty (\lambda_ng_n)\, x_n\sim\sum_0^\infty g_n x_n=\mu$). However, this particular measure $\mu$ cannot be ergodic ([@Sophie]).
Building on Flytzanis’ ideas, the first named author and S. Grivaux came rather close to characterizing the weak and strong mixing properties for Banach space operators in terms of the $\TT$-eigenvectors ([@BG3], [@BG2]). However, this was not quite the end of the story because the sufficient conditions for weak or strong mixing found in [@BG3] and [@BG2] depend on some geometrical property of the underlying Banach space, or on some “regularity" property of the $\TT$-eigenvectors (see the remark just after Corollary \[eigvectfield\]).
In the present paper, our aim is to show that in fact, these assumptions can be completely removed. Thus, we intend to establish “optimal" sufficient conditions for weak and strong mixing in terms of the $\TT$-eigenvectors which are valid on an arbitrary Fréchet space. These conditions turn out to be also necessary when the underlying space $X$ is a Banach space with *cotype 2*, and hence we get complete characterizations of weak and strong mixing in this case. We shall in fact consider some more general notions of “mixing", but our main concerns are really the weak and strong mixing properties.
At this point, we should recall the definitions. A measure-preserving transformation $T:(X,\mathfrak B,\mu)\to(X,\mathfrak B,\mu)$ is *weakly mixing* (with respect to $\mu$) if $$\frac{1}{N}\sum_{n=0}^{N-1} \vert \mu(A\cap T^{-n}(B))-\mu (A)\mu(B)\vert\xrightarrow{N\to\infty} 0$$ for any measurable sets $A,B\subset X$; and $T$ is *strongly mixing* if $$\mu(A\cap T^{-n}(B))\xrightarrow{n\to\infty} \mu(A)\mu(B)$$ for any $A,B\in\mathfrak B$. (Ergodicity can be defined exactly as weak mixing, but removing the absolute value in the Cesàro mean).
According to the “spectral viewpoint" on ergodic theory, weakly mixing transformations are closely related to *continuous* measures on the circle $\TT$, and strongly mixing transformations are related to *Rajchman* measures, i.e. measures whose Fourier coefficients vanish at infinity. Without going into any detail at this point, we just recall that, by a classical result of Wiener (see [@Ktz]), continuous measures on $\TT$ are characterized by the behaviour of their Fourier coefficients: a measure $\sigma$ is continuous if and only if $$\frac{1}{N}\sum_{n=0}^{N-1} \vert \widehat\sigma (n)\vert\xrightarrow{N\to\infty} 0\, .$$ Wiener’s lemma is usually stated with symmetric Cesàro means, but this turns out to be equivalent. Likewise, by the so-called *Rajchman’s lemma*, a measure $\sigma$ is Rajchman if and only if $\widehat\sigma (n)\to 0$ as $n\to +\infty$ (that is, a one-sided limit is enough). Especially important for us will be the corresponding families of “small" sets of the circle; that is, the sets which are annihilated by every positive measure in the family under consideration (continuous measures, or Rajchman measures). Obviously, a Borel set $D\subset\TT$ is small for continuous measures if and only if it is countable. The small sets for Rajchman measures are the so-called *sets of extended uniqueness* or *sets of uniqueness for Fourier-Stieltjes series*, which have been extensively studied since the beginning of the 20th century (see [@KL]). The family of all sets of extended uniqueness is usually denoted by $\mathcal U_0$.
Our main results can now be summarized as follows.
\[WS\] Let $X$ be a complex separable Fréchet space, and let $T\in\mathfrak L(X)$.
1. Assume that the $\TT$-eigenvectors are *perfectly spanning*, in the following sense: for any countable set $D\subset \TT$, the linear span of $\bigcup_{\lambda\in\TT\setminus D}\ker (T-\lambda)$ is dense in $X$. Then $T$ is weakly mixing in the Gaussian sense.
2. Assume that the $\TT$-eigenvectors are *$\mathcal U_0$-perfectly spanning*, in the following sense: for any Borel set of extended uniqueness $D\subset \TT$, the linear span of $\bigcup_{\lambda\in\TT\setminus D}\ker (T-\lambda)$ is dense in $X$. Then $T$ is strongly mixing in the Gaussian sense.
3. In [(1)]{} and [(2)]{}, the converse implications are true if $X$ is a Banach space with [cotype 2]{}.
Some remarks are in order regarding the scope and the “history" of these results.
1. When $X$ is a Hilbert space, (1) is stated in [@F] (with some additional assumptions on the operator $T$) and a detailed proof is given in [@BG3] (without these additional assumptions). The definition of “perfectly spanning" used in [@BG3] is formally stronger than the above one, but the two notions are in fact equivalent ([@Sophie]).
2. It is shown in [@Sophie] that under the assumption of (1), the operator $T$ is frequently hypercyclic. The proof is rather complicated, and it is not clear that it could be modified to get weak mixing in the Gaussian sense. However, some of the ideas of [@Sophie] will be crucial for us. In particular, sub-section \[Sophiesection\] owes a lot to [@Sophie].
3. In the weak mixing case, (3) is proved in [@BG2 Theorem 4.1].
4. It seems unnecessary to recall here the definition of cotype (see any book on Banach space theory, [@AK]). Suffices it to say that this is a geometrical property of the space, and that Hilbert space has cotype $2$ as well as $L^p$ spaces for $p\in [1,2]$ (but *not* for $p>2$).
5. As observed in [@BG2 Example 4.2], (3) does not hold on an arbitrary Banach space $X$. Indeed, let $X:=\mathcal C_0([0,2\pi])=\{ f\in\mathcal C([0,2\pi]);\; f(0)=0\}$ and let $V:L^2(0,2\pi)\to X$ be the Volterra operator, $Vf(t)=\int_0^t f(s)\, ds$. There is a unique operator $T:X\to X$ such that $TV=VM_\phi$, where $M_\phi:L^2(0,2\pi)\to L^2(0,2\pi)$ is the multiplication operator associated with the function $\phi (t)=e^{it}$. The operator $T$ is given by the formula $$\label{Kal}
Tf(t)=\phi(t)f(t)-\int_0^t \phi'(s)f(s)\, ds\, .$$ It is easy to check that $T$ has no eigenvalues. On the other hand, $T$ is strongly mixing with respect to the Wiener measure on $\mathcal C_0([0,2\pi])$.
As it turns out, ergodicity and weak mixing in the Gaussian sense are in fact equivalent (see e.g. [@G], or [@BG2 Theorem 4.1]). Hence, from Theorem \[WS\] we immediately get the following result. (A Gaussian measure $\mu$ is *nontrivial* if $\mu\neq\delta_0$).
\[characexistergod\] For a linear operator $T$ acting on a Banach space $X$ with cotype 2, the following are equivalent:
- $T$ admits a nontrivial ergodic Gaussian measure;
- there exists a closed, $T$-invariant subspace $Z\neq \{ 0\}$ such that $$\overline{\rm span}\, \bigcup_{\lambda\in\TT\setminus D}\ker (T_{\vert Z}-\lambda)=Z$$ for every countable set $D\subset\TT$.
In this case, $T$ admits an ergodic Gaussian measure with support $Z$, for any such subspace $Z$.
If $T$ admits an ergodic Gaussian measure $\mu\neq\delta_0$, then $Z:={\rm supp}(\mu)$ is a non-zero $T$-invariant subspace, and $Z$ satisfies (b) by Theorem \[WS\] (3). The converse follows from Theorem \[WS\] (1).
For concrete applications, it is useful to formulate Theorem \[WS\] in terms of *$\TT$-eigenvector fields* for the operator $T$. A $\TT$-eigenvector field for $T$ is a map $E:\Lambda\to X$ defined on some set $\Lambda\subset\TT$, such that $$TE(\lambda)=\lambda E(\lambda)$$ for every $\lambda\in\Lambda$. (The terminology is not perfectly accurate: strictly speaking, $E(\lambda)$ is perhaps not a $\TT$-eigenvector because it is allowed to be $0$). Recall also that a closed set $\Lambda\subset\TT$ is *perfect* if it has no isolated points or, equivalently, if $V\cap\Lambda$ is uncountable for any open set $V\subset\TT$ such that $V\cap\Lambda\neq\emptyset$. Analogously, a closed set $\Lambda\subset \TT$ is said to be *$\mathcal U_0$-perfect* if $V\cap \Lambda$ is not a set of extended uniqueness for any open set $V$ such that $V\cap\Lambda\neq\emptyset$. (For example, any nontrivial closed arc is $\mathcal U_0$-perfect).
\[eigvectfield\] Let $X$ be a separable complex Fréchet space, and let $T\in\mathfrak L(X)$. Assume that one has at hand a family of *continuous* $\TT$-eigenvector fields $(E_i)_{i\in I}$ for $T$, where $E_i:\Lambda_i\to X$ is defined on some closed set $\Lambda_i\subset\TT$, such that ${\rm span}\left(\bigcup_{i\in I} E_i(\Lambda_i)\right)$ is dense in $X$.
1. If each $\Lambda_i$ is a perfect set, then $T$ is weakly mixing in the Gaussian sense.
2. If each $\Lambda_i$ is $\mathcal U_0$-perfect, then $T$ is strongly mixing in the Gaussian sense.
This follows immediately from Theorem \[WS\]. Indeed, if $\Lambda\subset\TT$ is a perfect set then $\Lambda\setminus D$ is dense in $\Lambda$ for any countable set $D$, whereas if $\Lambda$ is $\mathcal U_0$-perfect then $\Lambda\setminus D$ is dense in $\Lambda$ for any $\mathcal U_0$-set $D$. Since the $\TT$-eigenvector fields $E_i$ are assumed to be continuous, it follows that the $\TT$-eigenvectors of $T$ are perfectly spanning in case (i), and $\mathcal U_0$-perfectly spanning in case (ii).
Several results of this kind are proved in [@BG2] and in [@BM Chapter 5], all of them being based on an interplay between the geometry of the (Banach) space $X$ and the regularity of the $\TT$-eigenvector fields $E_i$. For example, it is shown that if $X$ has *type 2*, then continuity of the $E_i$ is enough, whereas if the $E_i$ are Lipschitz and defined on (nontrivial) closed arcs then no assumption on $X$ is needed. “Intermediate" cases involve the [type]{} of the Banach space $X$ and H" older conditions on the $E_i$. What Corollary \[eigvectfield\] says is that continuity of the $E_i$ is *always* enough, regardless of the underlying space $X$.
We also point out that the assumption in (i), i.e. the existence of $\TT$-eigenvector fields with the required spanning property defined on perfect sets, is in fact equivalent to the perfect spanning property ([@Sophie]). Likewise, the assumption in (ii) is equivalent to the $\mathcal U_0$-perfect spanning property (see Proposition \[perfect\]).
In order to illustrate our results, two examples are worth presenting immediately. Other examples will be reviewed in section \[final\].
Let $\mathbf w=(w_n)_{n\geq 1}$ be a bounded sequence of nonzero complex numbers, and let $B_{\bf w}$ be the associated *weighted backward shift* acting on $X_p=\ell^p(\NN)$, $1\leq p<\infty$ or $X_\infty=c_0(\NN)$; that is, $B_{\bf w}(x_0,x_1,x_2,\dots )=(w_1x_1,w_2x_2,\dots ).$ Solving the equation $B_{\mathbf w}(x)=\lambda x$, it is easy to check that $B_{\mathbf w}$ has eigenvalues of modulus 1 if and only if $$\label{shift}
\hbox{the sequence $\displaystyle\left(\frac{1}{w_0\cdots w_n}\right)_{n\geq 0}$ is in $X_p$}$$ (we have put $w_0:=1$). In this case the formula $$E(\lambda):=\sum_{n=0}^\infty \frac{\lambda^n}{w_0\cdots w_n}\, e_n$$ defines a continuous $\TT$-eigenvector field $E:\TT\to X_p$ such that $\overline{\rm span}\, E(\TT)=X_p$. Hence $B_{\mathbf w}$ is strongly mixing in the Gaussian sense. This is known since [@BG2] if $p<\infty$, but it appears to be new for weighted shifts on $c_0(\NN)$.
The converse is true if $p\leq 2$ ( (\[shift\]) is satisfied if $B_{\mathbf w}$ is strongly mixing in the Gaussian sense) since in this case $X_p$ has cotype 2, but the case $p>2$ is not covered by Theorem \[WS\]. However, it turns out that the converse does hold true for any $p<\infty$. In fact, (\[shift\]) is satisfied as soon as the weighted shift $B_{\mathbf w}$ is frequently hypercyclic ([@BR]). As shown in [@BG2], this breaks down completely when $p=\infty$: there is a frequently hypercyclic weighted shift $B_{\mathbf w}$ on $c_0(\NN)$ whose weight sequence satisfies $w_1\cdots w_n=1$ for infinitely many $n$. Such a weighted shift does not admit any (nontrivial) invariant Gaussian measure.
Let us also recall that, in contrast with the ergodic properties, the hypercyclicity of $B_{\bf w}$ does not depend on $p$: by a well known result of H. Salas ([@Sal]), $B_{\mathbf w}$ is hypercyclic on $X_p$ for any $p$ if and only if $\sup_{n\geq 1} \vert w_1\cdots w_n\vert=\infty$. Likewise, $B_{\mathbf w}$ is strongly mixing in the topological sense (on any $X_p$) iff $\vert w_1\cdots w_n\vert\to\infty$. Hence, we see that strong mixing in the topological sense turns out to be equivalent to strong mixing in the Gaussian sense for weighted shifts on $c_0(\NN)$.
Let $T$ be the operator defined by formula (\[Kal\]), but acting on $L^2(0,2\pi)$. It is straightforward to check that for any $t\in (0,2\pi)$, the function $f_t=\mathbf 1_{(0,t)}$ is an eigenvector for $T$ with associated eigenvalue $\lambda =e^{it}$. Moreover the map $E:\TT\setminus\{ \mathbf 1\}\to L^2(0,2\pi)$ defined by $E(e^{it})=f_t$ is clearly continuous. Now, let $\Lambda$ be an arbitrary compact subset of $\TT\setminus\{ \mathbf 1\}$. Let us denote by $H_\Lambda$ the closed linear span of $E(\Lambda)$, and let $T_\Lambda$ be the restriction of $T$ to $H_\Lambda$. By a result of G. Kalisch ([@Kal], see also [@BG1 Lemma 2.12]), the point spectrum of $T_\Lambda$ is exactly equal to $\Lambda$. By Corollary \[eigvectfield\] and Theorem \[WS\] (3), it follows that the operator $T_\Lambda$ is weakly mixing in the Gaussian sense if and only if $\Lambda$ is a perfect set, and strongly mixing iff $\Lambda$ is $\mathcal U_0$-perfect. Hence, any perfect $\mathcal U_0$-set $\Lambda$ gives rise to a very simple example of a weakly mixing transformation which is not strongly mixing. This could be of some interest since the classical concrete examples are arguably more complicated (see the one given in [@P section 4.5]).
Regarding the difference between weak and strong mixing, it is also worth pointing out that there exist Hilbert space operators which are weakly mixing in the Gaussian sense but not even strongly mixing in the [topological]{} sense. Indeed, in the beautiful paper [@BadG], C. Badea and S. Grivaux are able to construct a weakly mixing operator (in the Gaussian sense) which is *partially power-bounded*, $\sup_{n\in I} \Vert T^n\Vert<\infty$ for some infinite set $I\subset\NN$. This line of investigations was pursued even much further in [@BadG2] and [@EG].
We have deliberately stressed the formal analogy between weak and strong mixing in the statement of Theorem \[WS\]. In view of this analogy, it should not come as a surprise that Theorem \[WS\] can be deduced from some more general results dealing with abstract notions of “mixing". (In order not to make this introduction exceedingly long, these results will be described in the next section). In particular, (1) and (2) are formal consequences of Theorem \[abstract\] below. However, even though the proof of Theorem \[abstract\] is “conceptually" simple, the technical details make it rather long. This would be exactly the same for the strong mixing case (i.e. part (2) of Theorem \[WS\]), but in the weak mixing case it is possible to give a technically much simpler and hence much shorter proof. For the sake of readability, it seems desirable to present this proof separately. But since there is no point in repeating identical arguments, we shall follow the abstract approach as long as this does not appear to be artificial.
The paper is organized as follows. In section 2, we present our abstract results. In section 3, we review some basic facts concerning Gaussian measures and we outline the strategy for proving the abstract results and hence Theorem \[WS\]. Apart from some details in the presentation and the level of generality, this follows the scheme described in [@BG3], [@BG2] and [@BM]. In section 4, we prove part (1) of Theorem \[WS\] (the sufficient condition for weak mixing). The abstract results are proved in sections \[proofabstract1\] and \[proofabstract2\]. Section \[final\] contains some additional examples and miscellaneous remarks. In particular, we briefly discuss the “continous" analogues of our results (i.e. the case of $1$-parameter semigroups), and we show that for a large class of strongly mixing weighted shifts, the set of hypercyclic vectors turns out to be rather small, namely Haar-null in the sense of Christensen. We conclude the paper with some possibly interesting questions.
[**Notation and conventions.**]{} The set of natural numbers is denoted either by $\NN$ or by $\ZZ_+$. We denote by $\mathcal M(\TT)$ the space of all complex measures on $\TT$, endowed with total variation norm. The Fourier transform of a measure $\sigma\in\mathcal M(\TT)$ is denoted either by $\widehat\sigma$ or by $\mathcal F(\sigma)$. As a rule, all measurable spaces $(\Omega,\mathfrak A)$ are standard Borel, and all measure spaces $(\Omega,\mathfrak A, m)$ are sigma-finite. All Hilbert spaces $\mathcal H$ are (complex) separable and infinite-dimensional. The scalar product $\langle u,v\rangle_{\mathcal H}$ is linear with respect to $u$ and conjugate-linear with respect to $v$.
Abstract results
================
$\mathbf S$-mixing
------------------
It is well known (and easy to check) that the definitions of ergodicity, weak and strong mixing can be reformulated as follows. Let $(X,\mathfrak B,\mu)$ be a probability space, and set $$L^2_0(\mu):=\left\{ f\in L^2(\mu);\; \int_X f\, d\mu=0\right\} .$$ Then, a measure-preserving transformation $T:(X,\mathfrak B,\mu)\to(X,\mathfrak B,\mu)$ is ergodic with respect to $\mu$ if and only if $$\frac{1}{N}\sum_{n=0}^{N-1} \langle f\circ T^n,g\rangle_{L^2(\mu)}\xrightarrow{N\to\infty} 0$$ for any $f,g\in L^2_0(\mu)$. The transformation $T$ is weakly mixing iff $$\frac{1}{N}\sum_{n=0}^{N-1} \left\vert \langle f\circ T^n,g\rangle_{L^2(\mu)}\right\vert\xrightarrow{N\to\infty} 0$$ for any $f,g\in L^2_0(\mu)$, and $T$ is strongly mixing iff $$\langle f\circ T^n,g\rangle_{L^2(\mu)}\xrightarrow{n\to\infty} 0\, .$$
Now, let us denote by $V_T:L^2(\mu)\to L^2(\mu)$ the Koopman operator associated with a measure-preserving transformation $T:(X,\mathfrak B,\mu)\to(X,\mathfrak B,\mu)$, the isometry defined by $$V_Tf=f\circ T\, .$$
For any $f,g\in L^2(\mu)$, there is a uniquely defined complex measure $\sigma_{f,g}=\sigma_{f,g}^T$ on $\TT$ such that $$\widehat\sigma_{f,g} (n)=\left\{
\begin{matrix} \langle V_T^n f,g\rangle_{L^2(\mu)}&{\rm if}&n\geq 0\\
\langle V_T^{*\vert n\vert} f,g\rangle_{L^2(\mu)}&{\rm if}&n< 0
\end{matrix}
\right.$$ (When $f=g$, this follows from Bochner’s theorem because in this case the right-hand side defines a positive-definite function on $\ZZ$; and then one can use a “polarization" argument). We denote by $\Sigma(T,\mu)$ the collection of all measures $\sigma_{f,g}$, $f,g\in L^2_0(\mu)$, and forgetting the measure $\mu$ we refer to $\Sigma(T,\mu)$ as “the spectral measure of $T$".
With these notations, we see that $T$ is weakly mixing with respect to $\mu$ iff all measures $\sigma\in \Sigma(T,\mu)$ are continuous (by Wiener’s lemma), and that $T$ is strongly mixing iff all measures $\sigma\in \Sigma(T,\mu)$ are Rajchman (by Rajchman’s lemma). Likewise, $T$ is ergodic iff $\sigma (\{ \mathbf 1\})=0$ for every $\sigma\in \Sigma(T,\mu)$.
More generally, given any family of measures $\mathcal B\subset\mathcal M(\TT)$, one may say that $T$ is *$\mathcal B$-mixing* with respect to $\mu$ if the spectral measure of $T$ lies in $\mathcal B$, all measures $\sigma\in\Sigma(T,\mu)$ are in $\mathcal B$. We shall in fact consider a more specific case which seems to be the most natural one for our concerns. Let us denote by $\mathcal F_+:\mathcal M(\TT)\to \ell^\infty(\ZZ_+)$ the positive part of the Fourier transformation, i.e. $\mathcal F_+(\sigma)= \widehat\sigma_{\vert \ZZ_+}$.
Given any family $\mathbf S\subset\ell^\infty(\ZZ_+)$, we say that a measure $\sigma\in\mathcal M(\TT)$ is *$\mathbf S$-continuous* if $\mathcal F_+(\sigma)\in\mathbf S$. A measure-preserving transformation $T:(X,\mu)\to (X,\mu)$ is *$\mathbf S$-mixing* with respect to $\mu$ if every measure $\sigma \in\Sigma (T,\mu)$ is $\mathbf S$-continuous.
Thus, strong mixing is just $\mathbf S$-mixing for the family $\mathbf S=c_0(\ZZ_+)$, weak mixing is $\mathbf S$-mixing for the the family $\mathbf S$ of all sequences $(a_n)\in\ell^\infty(\ZZ_+)$ such that $\vert a_n\vert\to 0$ in the Cesàro sense, and ergodicity corresponds to the family $\mathbf S$ of all $a\in\ell^\infty (\ZZ_+)$ tending to $0$ in the Cesàro sense. In what follows, these families will be denoted by $\mathbf S_{\rm strong}$, $\mathbf S_{\rm weak}$ and $\mathbf S_{\rm erg}$, respectively.
Small subsets of the circle
---------------------------
Given a family of measures $\mathcal B\subset\mathcal M(\TT)$, it is quite natural in harmonic analysis to try to say something about the *$\mathcal B$-small* subsets of $\TT$, i.e. the sets $D\subset\TT$ that are annihilated by all positive measures $\sigma\in\mathcal B$. By this we mean that for any such measure $\sigma$, one can find a Borel set $\widetilde D$ (possibly depending on $\sigma$) such that $D\subset\widetilde D$ and $\sigma (\widetilde D)=0$. When the family $\mathcal B$ has the form $\mathcal B=\mathcal F_+^{-1}(\mathbf S)$ for some $\mathbf S\subset\ell^\infty (\ZZ_+)$, we call these sets *$\mathbf S$-small*.
To avoid trivialities concerning $\mathcal B$-small sets, the family $\mathcal B$ under consideration should contain nonzero [positive]{} measures, and in fact it is desirable that it should be *hereditary* with respect to absolute continuity; that is, any measure absolutely continuous with respect to some $\sigma\in\mathcal B$ is again in $\mathcal B$. The following simple lemma shows how to achieve this for families of the form $\mathcal F_{+}^{-1}(\mathbf S)$. Let us say that a family $\mathbf S\subset\ell^\infty(\ZZ_+)$ is *translation-invariant* if it is invariant under both the forward and the backward shift on $\ell^\infty(\ZZ_+)$.
\[hereditary\] If $\mathbf S$ is a translation-invariant linear subspace of $\ell^\infty(\ZZ_+)$ such that $\mathcal F_{+}^{-1}(\mathbf S)$ is norm-closed in $\mathcal M(\TT)$, then $\mathcal F_{+}^{-1}(\mathbf S)$ is hereditary with respect to absolute continuity.
If $\sigma\in\mathcal F_+^{-1}(\mathbf S)$ then $P\sigma$ is in $\mathcal F_+^{-1}(\mathbf S)$ for any trigonometric polynomial $P$, by translation-invariance. So the result follows by approximation.
We shall also make use of the following well known result concerning *$\mathcal B$-perfect* sets. By definition, a set $\Lambda\subset \TT$ is $\mathcal B$-perfect if $V\cap\Lambda$ is not $\mathcal B$-small for any open set $V\subset\TT$ such that $V\cap\Lambda\neq\emptyset$.
\[Bperfect\] Let $\mathcal B$ be a norm-closed linear subspace of $\mathcal M(\TT)$, and assume that $\mathcal B$ is hereditary with respect to absolute continuity. For a closed set $\Lambda\subset\TT$, the following are equivalent:
- $\Lambda$ is $\mathcal B$-perfect;
- $\Lambda$ is the support of some probability measure $\sigma\in\mathcal B$.
That (b) implies (a) is clear (without any assumption on $\mathcal B$). Conversely, assume that $\Lambda$ is $\mathcal B$-perfect. Let $(W_n)_{n\geq 1}$ be a countable basis of open sets for $\Lambda$, with $W_n\neq \emptyset$. Since $\mathcal B$ is hereditary, one can find for each $n$ a probability measure $\sigma_n\in\mathcal B$ such that ${\rm supp}(\sigma_n)\subset\Lambda$ and $\sigma_n (W_n)>0$. Then the probability measure $\sigma=\sum_1^\infty 2^{-n}\sigma_n$ is in $\mathcal B$ and ${\rm supp}(\sigma)=\Lambda$.
The results
-----------
To state our results we need two more definitions.
\[defBperfspan\] Let $T$ be an operator acting on a complex separable Fréchet space $X$, and let $\mathcal B\subset\mathcal M(\TT)$. We say that the $\TT$-eigenvectors of $T$ are *$\mathcal B$-perfectly spanning* if, for any Borel $\mathcal B$-small set $D\subset \TT$, the linear span of $\bigcup_{\lambda\in\TT\setminus D}\ker (T-\lambda)$ is dense in $X$. When $\mathcal B$ has the form $\mathcal F_+^{-1}(\mathbf S)$, the terminology *$\mathbf S$-perfectly spanning* is used.
Thus, perfect spanning is $\mathcal B$-perfect spanning for the family of continuous measures, and $\mathcal U_0$-perfect spanning is $\mathcal B$-perfect spanning for the family of Rajchman measures.
At some places, we will encounter sets which are *analytic* but perhaps non Borel. Recall that a set $D$ in some Polish space $Z$ is [analytic]{} if one can find a Borel relation $B\subset Z\times E$ (where $E$ is Polish) such that $z\in D\Leftrightarrow\exists u\;:\; B(z,u)$. If the spanning property of the above definition holds for every analytic $\mathcal B$-small set $D$, we say that the $\TT$-eigenvectors of $T$ are [$\mathcal B$-perfectly spanning]{} *for analytic sets*.
We shall say that a family $\mathbf S\subset\ell^\infty(\ZZ_+)$ is *$c_0$-like* if it has the form $$\mathbf S=\{ a\in\ell^\infty(\ZZ_+);\; \lim_{n\to\infty} \Phi_n(a)=0\}\,$$ where $(\Phi_n)_{n\geq 0}$ is a uniformly bounded sequence of $w^*$-$\,$continuous semi-norms on $\ell^\infty(\ZZ_+)$. (By “uniformly bounded", we mean that $\Phi_n(a)\leq C\, \Vert a\Vert_\infty$ for all $n$ and some finite constant $C$).
For example, the families $\mathbf S_{\rm strong}$, $\mathbf S_{\rm weak}$ and $\mathbf S_{\rm erg}$ are $c_0$-like: just put $\Phi_n(a)=\vert a_n\vert$ in the strong mixing case, $\Phi_n(a)=\frac{1}{n}\sum_{k=0}^{n-1}\vert a_k\vert$ in the weak mixing case, and $\Phi_n(a)=\left\vert \frac{1}{n}\sum_{k=0}^{n-1} a_k\right\vert$ in the ergodic case.
Our main result is the following theorem, from which (1) and (2) in Theorem \[WS\] follow immediately. Recall that a family $\mathbf S\subset \ell^\infty(\ZZ_+)$ is an *ideal* if it is a linear subspace and $ua\in \mathbf S$ for any $(u,a)\in\ell^\infty(\ZZ_+)\times\mathbf S$.
\[abstract\] Let $X$ be a separable complex Fréchet space, and let $T\in\mathfrak L(X)$. Let also $\mathbf S\subset \ell^\infty (\ZZ_+)$. Assume that $\mathbf S$ is a translation-invariant and $c_0$-like [ideal]{}, and that any $\mathbf S$-continuous measure is continuous. If the $\TT$-eigenvectors of $T$ are $\mathbf S$-perfectly spanning, then $T$ is $\mathbf S$-mixing in the Gaussian sense.
This theorem cannot be applied to the ergodic case, for two reasons: the family $\mathbf S_{\rm erg}$ is not an ideal of $\ell^\infty (\ZZ_+)$, and $\mathbf S_{\rm erg}$-continuous measures need not be continuous (a measure $\sigma$ is $\mathbf S_{\rm erg}$-continuous iff $\sigma(\{ \mathbf 1\})=0$).
Theorem \[abstract\] will be proved in section \[proofabstract1\]. The following much simpler converse result (which corresponds to (3) in Theorem \[WS\]) will be proved in section \[proofabstract2\].
\[converse\] Let $X$ be a separable complex Banach space, and assume that $X$ has cotype 2. Let also $\mathbf S$ be an arbitrary subset of $\ell^\infty(\ZZ_+)$. If $T\in\mathfrak L(X)$ is $\mathbf S$-mixing in the Gaussian sense, then the $\TT$-eigenvectors of $T$ are $\mathbf S$-perfectly spanning for analytic sets.
Two “trivial" cases are worth mentioning. If we take $\mathbf S=\mathbf S_{\rm erg}$, then $\mathcal F_+^{-1}(\mathbf S)=\{ \sigma\in\mathcal M(\TT);\; \sigma (\{\mathbf 1\})=0\}$ and hence a set $D\subset\TT$ is $\mathbf S$-small if and only if $D\subset\{\mathbf 1\}$. If we take $\mathbf S=\ell^\infty (\ZZ_+)$, then $\mathcal F_+^{-1}(\mathbf S)=\mathcal M(\TT)$ and hence the only $\mathbf S$-small set is $D=\emptyset$. So, assuming that $X$ has cotype $2$, we get the following (known) results: if $T$ admits an invariant Gaussian measure with full support, then the $\TT$-eigenvectors of $T$ span a dense subspace of $X$; and if $T$ is ergodic in the Gaussian sense, then ${\overline{\rm span}}\,\left(\bigcup_{\lambda\in\TT\setminus\{ \mathbf 1\}}\ker(T-\lambda)\right)=X$. As already pointed out, much more is true in the ergodic case: it is proved in [@BG2 Theorem 4.1] that if $T$ is ergodic in the Gaussian sense, then in fact the $\TT$-eigenvectors are perfectly spanning (provided that $X$ has cotype 2).
Our last result (to be proved also in section \[proofabstract2\]) says that when the space $X$ is well-behaved, the assumptions in Theorem \[abstract\] can be relaxed: it is no longer necessary to assume that the family $\mathbf S$ is $c_0$-like, nor that $\mathbf S$-continuous measures are continuous. However, the price to pay is that one has to use the strengthened version of $\mathbf S$-perfect spanning.
\[abstracteasy\] Let $X$ be a separable complex Banach space, and assume that $X$ has *type 2*. Let also $\mathbf S$ be a norm-closed, translation-invariant ideal of $\ell^\infty(\ZZ_+)$, and let $T\in\mathfrak L(X)$. If the $\TT$-eigenvectors of $T$ are $\mathbf S$-perfectly spanning for analytic sets, then $T$ is $\mathbf S$-mixing in the Gaussian sense.
Since Hilbert space has both type 2 and cotype 2, we immediately deduce
\[Smix=span\] Let $\mathbf S$ be a norm-closed, translation-invariant ideal of $\ell^\infty(\ZZ_+)$. For Hilbert space operators, $\mathbf S$-mixing in the Gaussian sense is equivalent to the $\mathbf S$-spanning property of $\TT$-eigenvectors (for analytic sets).
The reader should wonder why analytic sets appear in Theorem \[abstracteasy\], whereas only Borel sets are needed in Theorem \[abstract\]. The reason is that the family of $\mathbf S$-small sets has a quite special structural property (the so-called *covering property*) when the family $\mathbf S$ is $c_0$-like: any analytic $\mathbf S$-small set can be covered by a sequence of *closed* $\mathbf S$-small sets (see [@MZ]). In particular, any analytic $\mathbf S$-small set is contained in a Borel $\mathbf S$-small set and hence the two notions of $\mathbf S$-perfect spanning are equivalent.
The covering property is of course trivially satisfied in the weak mixing case, for the family of countable sets. In the strong mixing case, for the sets of extended uniqueness, this is a remarkable result due to G. Debs and J. Saint Raymond ([@DSR]).
The proof of Theorem \[abstracteasy\] turns out to be considerably simpler than that of Theorem \[abstract\]. Roughly speaking, the reason is the following. Without any assumption on $X$, in the case of Theorem \[abstract\], we will have to be extremely careful to ensure that some “integral" operators of a certain kind are *gamma-radonifying* (see sub-section \[gamrad\]). On the other hand, such integral operators are [automatically]{} gamma-radonifying when $X$ is a Banach space with type 2. So the main technical difficulty just disappears, and the proof becomes rather easy.
background
==========
Throughout this section, $X$ is a separable complex Fréchet space.
Gaussian measures and gamma-radonifying operators {#gamrad}
-------------------------------------------------
This sub-section is devoted to a brief review of the basic facts that we shall need concerning Gaussian measures. For a reasonably self-contained exposition specifically tailored to linear dynamics, we refer to [@BM]; and for in-depth studies of Gaussian measures, to the books [@Bo] and [@CTV].
By a *Gaussian measure* on $X$, we mean a Borel probability measure $\mu$ on $X$ which is the distribution of a random variable of the form $$\sum_{n=0}^\infty g_n x_n\,$$ where $(g_n)$ is a standard complex Gaussian sequence defined on some probability space $(\Omega,\mathfrak A,\mathbb P)$ and $(x_n)$ is a sequence in $X$ such that the series $\sum g_n x_n$ is almost surely convergent.
We recall that $(g_n)$ is a standard complex Gaussian sequence if the $g_n$ are independent complex-valued random variables with distribution $\gamma_\sigma\otimes\gamma_\sigma$, where $\gamma_\sigma=\frac{1}{\sqrt{2\pi}\sigma}e^{-t^2/2\sigma^2}\, dt$ is the usual Gaussian distribution with mean $0$ and variance $\sigma^2=1/2$. This implies in particular that $\mathbb Eg_n=0$ and $\mathbb E\vert g_n\vert^2=1$.
“Our" definition of a Gaussian measure is not the usual one. However, it is indeed equivalent to the classical definition, each continuous linear functional $x^*$ has a complex symmetric Gaussian distribution when considered as a random variable on the probability space $(X,\mathfrak B(X),\mu)$ (where $\mathfrak B(X)$ is the Borel sigma-algebra of $X$).
It is well known that for a Fréchet space valued Gaussian series $\sum g_n x_n$, almost sure convergence is equivalent to convergence in the $L^p$ sense, for any $p\in [1,\infty[$. It follows at once that if $\mu\sim \sum_{n\geq 0} g_n x_n$ is a Gaussian measure on $X$, then $\int_X \Vert x\Vert^2\, d\mu(x)<\infty$ for every continuous semi-norm $\Vert\,\cdot\,\Vert$ on $X$. In particular, any continuous linear functional $x^*\in X^*$ is in $L^2(\mu)$ when considered as a random variable on $(X,\mu)$. We also note that we consider only *centred* Gaussian measures $\mu$, $\int_X x\, d\mu(x)=0$.
Gaussian measures correspond in a canonical way to *gamma-radonifying operators*. Let $\mathcal H$ be a Hilbert space. An operator $K:\mathcal H\to X$ is said to be gamma-radonifying if for some (equivalently, for any) orthonormal basis $(e_n)_{n\geq 0}$ of $\mathcal H$, the Gaussian series $\sum g_n K(e_n)$ is almost surely convergent. By rotational invariance of the Gaussian variables $g_n$, the distribution of the random variable $\sum_{n\geq 0} g_n K(e_n)$ does not depend on the orthonormal basis $(e_n)$, so the operator $K$ gives rise to a Gaussian measure which may be denoted by $\mu_K$: $$\mu_K\sim \sum g_n K(e_n)\, .$$ Conversely, it is not hard to show that any Gaussian measure $\mu\sim\sum g_n x_n$ is induced by some gamma-radonifying operator.
The following examples are worth keeping in mind: when $X$ is a Hilbert space, an operator $K:\mathcal H\to X$ is gamma-radonifying if and only if it is a *Hilbert-Schmidt* operator; and for an arbitrary $X$, the operator $K$ is gamma-radonifying as soon as $\sum_0^\infty \Vert K(e_n)\Vert<\infty$, for some orthonormal basis of $\mathcal H$ and every continuous semi-norm $\Vert\,\cdot\,\Vert$ on $X$. This follows at once from the equivalence between almost sure convergence and $L^2$ or $L^1$ convergence for Gaussian series.
We note that the support of a Gaussian measure $\mu\sim\sum_{n\geq 0} g_n x_n$ is the closed linear span of the $x_n$. Therefore, a Gaussian measure $\mu=\mu_K$ has full support if and only if the gamma-radonifying operator $K:\mathcal H\to X$ has dense range.
If $K:\mathcal H\to X$ is a continuous linear operator from some Hilbert space $\mathcal H$ into $X$, we consider its adjoint $K^*$ as an operator from $X^*$ into $\mathcal H$. Hence, $K^*:X^*\to \mathcal H$ is a *conjugate-linear* operator. If $K$ is gamma-radonifying and $\mu=\mu_K$ is the associated Gaussian measure, then we have the following fundamental identity: $$\langle x^*, y^*\rangle_{L^2(\mu)}=\langle K^*y^*, K^*x^*\rangle_{\mathcal H}$$ for any $x^*,y^*\in X^*$. The proof is a simple computation using the orthogonality of the Gaussian variables $g_n$.
Let us also recall the definition of the *Fourier transform* of a Borel probability measure $\mu$ on $X$: this is the complex-valued function defined on the dual space $X^*$ by $$\widehat\mu (x^*)=\int_X e^{-i{\rm Re}\, \langle x^*, x\rangle}d\mu (x)\, .$$
One very important property of the Fourier transform is that it uniquely determines the measure: if $\mu_1$ and $\mu_2$ have the same Fourier transform, then $\mu_1=\mu_2$.
If $\mu\sim \sum_{n\geq 0} g_n x_n$ is a Gaussian measure, a simple computation shows that $\widehat\mu (x^*)=\exp\left(-\frac{1}{4} \sum_{0}^\infty \vert\langle x^*, x_n\rangle\vert^2\right)$ for any $x^*\in X^*$. It follows that if $K:\mathcal H\to X$ is a gamma-radonifying operator then $$\widehat{\mu_K}(x^*)=e^{-\frac14{\Vert K^*x^*\Vert^2}}\, .$$ In particular, two gamma-radonifying operators $K_1$ and $K_2$ define the same Gaussian measure if and only if $$\Vert K_1^*x^*\Vert= \Vert K^*_2x^*\Vert$$ for every $x^*\in X^*$.
Finally, let us say a few words about type 2 and cotype 2 Banach spaces. A Banach space $X$ has *type 2* if $$\mathbb E\left\Vert \sum_ng_n x_n\right\Vert\leq C\left( \sum_n \Vert x_n\Vert^2\right)^{1/2}$$ for some finite constant $C$ and any finite sequence $(x_n)$ in $X$; and $X$ has *cotype 2* if the reverse inequality holds. Thus, type 2 makes the convergence of Gaussian series easier, whereas on a cotype 2 space the convergence of such a series has strong implications. This is apparent in the following proposition, which contains all the results that we shall need concerning these notions. (See (\[defke\]) below for the definition of the operators $K_E$).
\[typecotype\] Let $X$ be a separable Banach space.
1. If $X$ has type 2 then any operator $K_E:L^2(\Omega,m)\to X$ is gamma-radonifying.
2. If $X$ has cotype 2 then any gamma-radonifying operator $K:L^2(\Omega, m)\to X$ has the form $K_E$, for some vector field $E:\Omega\to X$.
These results are nontrivial (see [@BM]), but quite easy to use.
The general procedure
---------------------
Most of what is needed for the the proof of Theorem \[abstract\] is summarized in Proposition \[background\] below, whose statement requires to introduce some terminology. Recall that all measure spaces under consideration are *sigma-finite*, and that all $L^2$ spaces are separable.
Let $(\Omega ,\mathfrak A,m)$ be a measure space. By a ($X$-valued) *vector field* on $(\Omega, \mathfrak A,m)$ we mean any measurable map $E:\Omega\to X$ which is in $L^2(\Omega,m, X)$. Such a vector field gives rise to an operator $K_E:L^2(\Omega,m)\to X$ defined as follows:
$$\label{defke} K_E f=\int_\Omega f(\omega) E(\omega)\, dm(\omega)\, .$$
It is easy to check that the operator $K_E$ has dense range if and only if the vector field $E$ is *$m$-spanning* in the following sense: for any measurable set $\Delta\subset \Omega$ with $m(\Delta)=0$, the linear span of $\{ E(\omega);\; \omega\in\Omega\setminus\Delta\}$ is dense in $X$. This happens in particular if $\Omega$ is a topological space, $m$ is a Borel measure with full support, and the vector field $E$ is continuous with ${\overline{\rm span}}\, E(\Omega)=X$.
We also note that the operator $K_E$ is always compact, and that it is a Hilbert-Schmidt operator if $X$ is Hilbert space (because $E$ is in $L^2(\Omega,m,X)$). Moreover, the adjoint operator $K_E^*:X^*\to L^2(\Omega, m)$ is given by the formula $$K_E^*x^*=\overline{\langle x^*, E(\,\cdot\,)\rangle}\, .$$
The next definition will be crucial for our purpose.
Let $T\in\mathfrak L(X)$. By a *$\TT$-eigenfield* for $T$ on $(\Omega ,\mathfrak A, m)$ we mean a pair of maps $(E,\phi)$ where
- $E:\Omega \to X$ is a vector field;
- $\phi:\Omega\to \TT$ is measurable;
- $TE(\omega)=\phi (\omega) E(\omega)$ for every $\omega\in \Omega$.
For example, if $E:\Lambda\to X$ is a continuous $\TT$-eigenvector field for $T$ defined on some compact set $\Lambda\subset\TT$ and if we put $\phi (\lambda)=\lambda$, then $(E,\phi)$ is a $\TT$-eigenfield for $T$ on $(\Lambda, \mathfrak B(\Lambda), m)$ for any Borel probability measure $m$ on $\Lambda$.
The key fact about $\TT$-eigenfields is the following: if $(E,\phi)$ is a $\TT$-eigenfield for $T$ on $(\Omega,m)$ and if $M_\phi=L^2(\Omega, m)\to L^2(\Omega, m)$ is the (unitary) multiplication operator associated with $\phi$, then the *intertwining equation* $$TK_E=K_E M_\phi$$ holds. The proof is immediate.
\[background\] Let $\mathbf S$ be a norm-closed ideal of $\ell^\infty (\ZZ_+)$, and let $T\in\mathcal L(X)$. Assume that one can find a $\TT$-eigenfield $(E,\phi)$ for $T$ defined on some measure space $(\Omega, m)$, such that
- the operator $K_E : L^2(\Omega, m)\to X$ is gamma-radonifying;
- the vector field $E$ is $m$-spanning;
- for any $f\in L^1(\Omega, m)$, the image measure $\sigma_f=(f\, m)\circ\phi^{-1}$ is $\mathbf S$-continuous.
Then $T$ is $\mathbf S$-mixing in the Gaussian sense. More precisely, $T$ is $\mathbf S$-mixing with respect to $\mu_{K_E}$.
As mentioned above, this proposition is essentially all what is needed to understand the proof of Theorem \[abstract\]. It will follow at once from the next two lemmas, that will also be used in the proof of Proposition \[converse\].
\[back1\] Let $T\in\mathfrak L(X)$, and let $K:\mathcal H\to X$ be gamma-radonifying. Let also $\mathbf S$ be a norm-closed ideal of $\ell^\infty (\ZZ_+)$.
- The measure $\mu=\mu_{K}$ is $T$-invariant if and only if one can find an operator $M:\mathcal H\to\mathcal H$ such that $\mathcal H_{K}:=\mathcal H\ominus \ker(K)$ is $M^*$-invariant, $M^{*}$ is an isometry on $\mathcal H_K$ and $TK=KM$.
- The operator $T$ is $\mathbf S$-mixing with respect to $\mu$ if and only if $M^{*}$ is *$\mathbf S$-mixing on $\mathcal H_{K}$*, the sequence $\left(\langle M^{*n}u,v\rangle_{\mathcal H})\right)_{n\geq 0}$ is in $\mathbf S$ for any $u,v\in \mathcal H_{K}$.
Since $T$ is a linear operator, the image measure $\mu_{K}\circ T^{-1}$ is equal to $\mu_{TK}$. Taking the Fourier transforms, it follows that $\mu_K$ is $T$-invariant if and only if $$\Vert K^*(x^*)\Vert=\Vert K^*(T^*x^*)\Vert$$ for every $x^*\in X^*$. This means exactly that one can find an isometry $V:{\rm ran}(K^*)\to{\rm ran}(K^*)$ such that $K^*T^*=VK^*$. Since ${\overline{{\rm ran}(K^*)}}=\mathcal H\ominus \ker(K)$, this proves (1).
As for (2), the basic idea is very simple. Recall that $T$ is $\mathbf S$-mixing with respect to $\mu$ if and only if $$\label{L_0} \mathcal F_+({\sigma_{f,g}})\in\mathbf S$$ for any $f,g\in L^2_0(\mu)$. When $f=x^*$ and $g=y^*$ are continuous linear functionals on $X$ (recall that $\mu$ is centred, so that $X^*\subset L^2_0(\mu)$) we have $\widehat\sigma_{f,g}(n)=\langle f\circ T^n, g\rangle_{L^2(\mu)}=\langle K^*T^{*n}(x^*), K^*y^*\rangle_{\mathcal H}=\langle M^{*n}(K^*x^*), K^*y^*\rangle_{\mathcal H}$ for every $n\geq 0$. Since $M^*$ is power-bounded on $\mathcal H_K$ and $\mathcal H_K={\overline{{\rm ran}(K^*)}}$, it follows that the $\mathbf S$-mixing property of $M^*$ on $\mathcal H_K$ is equivalent to the validity of (\[L\_0\]) for all continuous linear functionals $f,g$. So the point is to pass from linear functionals to arbitrary $f,g\in L^2_0(\mu)$.
In the “classical" cases (weak and strong mixing), this can be done in at least two ways: either by reproducing a rather elementary argument of R. Rudnicki ([@R pp 108–109], see also [@GE] or [@BM Theorem 5.24]), or by the more abstract approach of [@BG3], which relies on the theory of *Fock spaces*. In the more general case we are considering, one possible proof would consist in merely copying out pp 5105–5108 of [@BG3]. The fact that $\mathbf S$ is norm-closed would be needed for the approximation argument on p. 5105, and the ideal property would be used on p. 5108 in the following way: if $(a_n)\in\mathbf S$ and if a sequence $(b_n)$ satisfies $\vert b_n\vert\leq C\, \vert a_n\vert$ for all $n\in\ZZ_+$ (and some finite constant $C$), then $(b_n)\in\mathbf S$.
We outline a more direct approach, which is in fact exactly the same as in [@BG3] but without any algebraic apparatus. In what follows, we denote by ${\rm Re}(X^*)$ the set of all continuous, real-valued, real-linear functionals on $X$. For any $u\in{\rm Re}(X^*)$, we denote by $u^*$ the unique complex-linear functional with real part $u$, which is given by the formula $\langle u^*,x\rangle=u(x)-iu(ix)$.
First, we note that if (\[L\_0\]) holds for all continuous linear functionals, then it holds for all $f,g\in{\rm Re}(X^*)$. Indeed, using the invariance of $\mu$ under the transformation $x\mapsto ix$, it is easily checked that $\langle u,v\rangle_{L^2(\mu)}=\frac{1}{2}\,{\rm Re}\left(\langle u^*,v^*\rangle_{L^2(\mu)}\right)$ for any $u,v\in{\rm Re}(X^*)$. Applying this with $u:=f\circ T^n$, $n\geq0$ and $v:=g$ and since $\mathbf S$, being an ideal, is closed under taking real parts, it follows that if $f,g\in{\rm Re}(X^*)$ and $\mathcal F_+({\sigma_{f^*,g^*}})\in\mathbf S$ then $\mathcal F_+({\sigma_{f,g}})\in\mathbf S$.
Now, let us denote by $H_k$, $k\geq 0$ the classical real *Hermite polynomials*, i.e. the orthogonal polynomials associated with the standard Gaussian distribution $\gamma=\frac{1}{\sqrt{2\pi}}\, e^{-t^2/2}dt$ on $\RR$. For every $k\geq 0$, set $$\mathcal H_k:=\overline{\rm span}^{L^2(\mu)}\{ H_k(f);\; f\in \mathcal S\}\, ,$$ where $\mathcal S$ is the set of all $f\in{\rm Re}(X^*)$ such that $\Vert f\Vert_{L^2(\mu)}=1$. It is well known that $L^2(\mu)$ is the orthogonal direct sum of the subspaces $\mathcal H_k$, $k\geq 0$ (this is the so-called *Wiener chaos decomposition*) and hence that $L^2_0(\mu)=\bigoplus_{k\geq 1} \mathcal H_k$. Moreover, it is also well known that $$\langle H_k(u),H_k(v)\rangle_{L^2(\mu)}=\langle u,v\rangle_{L^2(\mu)}^k$$ for any $u,v\in\mathcal S$ and every $k\geq 0$ (see [@D Chapter 9], where the proofs are given in a Hilbert space setting but work equally well on a Fréchet space). Taking $u:=f\circ T^n$, $n\geq 0$ and $v:=g$, it follows that $\mathcal F_+(\sigma_{H_k(f),H_k(g)})=\left[ \mathcal F_+(\sigma_{f,g})\right]^k$ for any $f,g\in\mathcal S$ and all $k\geq 0$. Since $\mathbf S$ is a closed ideal in $\ell^\infty(\ZZ_+)$ and since the map $(f,g)\mapsto \mathcal F_+({\sigma_{f,g}})$ is continuous from $L^2(\mu)\times L^2(\mu)$ into $\ell^\infty(\ZZ_+)$ we conclude that (\[L\_0\]) does hold true for any $f,g\in L^2_0(\mu)$ as soon as it holds for linear functionals.
It is apparent from the above proof that the implication “$T$ is $\mathbf S$-mixing implies $M^*$ is $\mathbf S$-mixing on $\mathcal H_K$" requires no assumption on the family $\mathbf S$. This will be used in the proof of Proposition \[converse\].
\[back2\] Let $M=M_{\phi}$ be a unitary multiplication operator on $\mathcal H=L^{2}(\Omega,m)$ associated with a measurable function $\phi:\Omega\to\TT$. Let also $\mathcal H_1\subset \mathcal H$ be a closed $M^*$-$\,$invariant subspace, and let us denote by $\mathcal H_1\cdot \mathcal H$ the set of all $f\in L^1(m)$ that can be written as $f=h_1h$, where $h_1\in \mathcal H_1$ and $h\in\mathcal H$. Finally, let $\mathbf S$ be an arbitrary subset of $\ell^\infty (\ZZ_+)$. Consider the following assertions:
- $M^{*}$ is $\mathbf S$-mixing on $\mathcal H_1$;
- for any $f\in\mathcal H_1\cdot\mathcal H$, the image measure $\sigma_f =(fm)\circ\phi^{-1}$ is $\mathbf S$-continuous;
- $\mathbf 1_{\{ \phi \in D\}}h\perp \mathcal H_1$, for any $\mathbf S$-small analytic set $D\subset \TT$ and every $h\in\mathcal H$.
Then [(i)]{} and [(ii)]{} are equivalent and imply [(iii)]{}.
We note that (iii) makes sense because $\phi^{-1}(D)$ is $m$-measurable for any analytic $D\subset\TT$ (see [@K]).
A straightworward computation shows that for any $u,v\in L^2(\Omega, m)$, the Fourier coefficients of $\sigma_{u\bar v}$ are given by $$\widehat\sigma_{u\bar v} (n)=\langle M^{*n} u,v\rangle_{\mathcal H}\, .$$ Moreover, since $\mathcal H_1$ is $M^*$-invariant we have $$\langle M^{*n} h_1,h\rangle_{\mathcal H}=\langle M^{*n} h_1,\pi_{1}h\rangle_{\mathcal H}\hskip 1cm (n\geq 0)$$ for any $h_1\in\mathcal H_1$ and every $h\in \mathcal H$, where $\pi_1$ is the orthogonal projection onto $\mathcal H_1$. It follows that $M^*$ is $\mathbf S$-mixing on $\mathcal H_1$ if and only if $\mathcal F_+({\sigma_f})\in\mathbf S$ for any $f=h_1\bar h\in\mathcal H_1\cdot\mathcal H$. In other words, (i) and (ii) are equivalent.
For any analytic set $D\subset\TT$ and $h,h_1\in\mathcal H$, we have $$\langle h_1,\mathbf 1_{\{\phi\in D\}} h\rangle_{\mathcal H}=\sigma_{h_1\bar h} (D)\, .$$ Hence, (iii) says exactly that $\sigma_f(D)=0$ for any $f\in\mathcal H_1\cdot\mathcal H$ and every $\mathbf S$-small analytic set $D\subset \TT$. From this, it is clear that (ii) implies (iii).
Properties (i), (ii), (iii) are in fact equivalent in the strong mixing case, i.e. $\mathbf S=c_0(\ZZ_+)$. Indeed, by a famous theorem of R. Lyons ([@L], see also [@KL]), a positive measure $\sigma$ is Rajchman *if and only if* $\sigma (D)=0$ for every Borel set of extended uniqueness $D\subset\TT$. From this, it follows that if (iii) holds then $\sigma_f$ is Rajchman for every nonnegative $f\in\mathcal H\cdot \mathcal H_1$, and it is easy to check that this implies (ii).
Let $M_\phi :L^2(\Omega ,m)\to L^2(\Omega, m)$ be the unitary multiplication operator associated with $\phi$. Since $TM_\phi =M_\phi K_E\, ,$ the Gaussian measure $\mu=\mu_{K_E}$ is $T$-invariant by Lemma \[back1\]. Moreover, $\mu$ has full support since $K_E$ has dense range (by (b)), and $T$ is $\mathbf S$-mixing with respect to $\mu$ by Lemmas \[back1\] and \[back2\].
An examination of the proof reveals that assumption (c) in Proposition \[background\] is a little bit stronger than what is actually needed: since $K_E^*:X^*\to L^2(\Omega,m)$ is given by $K_E^*x^*=\overline{\langle x^*, E(\,\cdot\,)\rangle}$, it is enough to assume that the measure $(fm)\circ\phi^{-1}$ is $\mathbf S$-continuous for every $f\in L^1(\Omega,m)$ of the form $f=\langle x^*, E(\,\cdot\,)\rangle\,\overline{\langle y^*, E(\,\cdot\, )\rangle}$, where $x^*,y^*\in X^*$. However, the proposition is easier to remember as stated.
Somewhat ironically, it will follow from Theorem \[abstract\] that the seemingly crucial gamma-radonifying assumption (a) in Proposition \[background\] is in fact not necessary to ensure $\mathbf S$-mixing in the Gaussian sense. Indeed, it is easily checked that if one can find a $\TT$-eigenfield for $T$ satisfying (b) and (c), then the $\TT$-eigenvectors of $T$ are $\mathbf S$-perfectly spanning. In fact, the proof of Theorem \[abstract\] essentially consists in showing that if there exists a $\TT$-eigenfield satisfying (b) and (c), then it is possible to construct one satisfying (b), (c) *and* (a).
An “exhaustion" lemma {#Sophiesection}
---------------------
Besides Proposition \[background\], the following lemma will also be needed in the proof of Theorems \[abstract\] and \[abstracteasy\].
\[exhaust\] Let $\mathcal B\subset\mathcal M(\TT)$, and let $T\in\mathcal L(X)$. Assume that the $\TT$-eigenvectors of $T$ are $\mathcal B$-perfectly spanning for analytic sets. Finally, put $$\mathbf V:=\{ (x,\lambda)\in X\times \TT;\; x\neq 0\;{\rm and}\; T(x)=\lambda x\}\, .$$ Then, one can find a closed subset $\mathbf Z$ of $\mathbf V$ with the following properties:
- for any (relatively) open set $O\subset\mathbf V$ such that $O\cap\mathbf Z\neq\emptyset$, the set $\pi_2(O\cap\mathbf Z):=\{ \lambda\in\TT;\;\exists x\in O\cap \mathbf Z\;:\; (x,\lambda)\in \mathbf Z\}$ is not $\mathcal B$-small;
- the linear span of $\pi_1(\mathbf Z):=\{ x\in X;\; \exists\lambda\in\TT\;:\; (x,\lambda)\in\mathbf Z\}$ is dense in $X$.
Let us denote by $\mathbf O$ the union of all relatively open sets $O\subset \mathbf V$ such that $\pi_2(O)$ is $\mathcal B$-small, and set $\mathbf Z:=\mathbf V\setminus \mathbf O$. Then $\mathbf Z$ is closed in $\mathbf V$ and satisfies the first required property (by its very definition). Moreover, by the Lindelöf property and since any countable union of $\mathcal B$-small sets is $\mathcal B$-small, the set $D:=\pi_2(\mathbf O)$ is $\mathcal B$-small; and $D$ is an analytic set because $\mathbf O$ is Borel in $X\times \TT$. By assumption on $T$, the linear span of $\bigcup_{\lambda\in\TT\setminus D}\ker (T-\lambda)$ is dense in $X$. Now, any $\TT$-eigenvector $x$ for $T$ is in $\pi_1(\mathbf V)$, and if the associated eigenvalue $\lambda$ is not in $D$ then $(x,\lambda)\in\mathbf Z$ by the definition of $D$. It follows that $\ker (T-\lambda)\setminus\{ 0\}$ is contained in $\pi_1(\mathbf Z)$ for any $\lambda\in\TT\setminus D$, and hence that $\overline{\rm span}\,( \pi_1(\mathbf Z))=X$.
Despite its simplicity, Lemma \[exhaust\] will be quite useful for us. We illustrate it by proving the following result.
\[perfect\] Let $\mathcal B\subset \mathcal M(\TT)$ be hereditary with respect to absolute continuity, and let $T\in\mathfrak L(X)$. The following are equivalent:
1. the $\TT$-eigenvectors of $T$ are $\mathcal B$-perfectly spanning for analytic sets;
2. one can find a countable family of [continuous]{} $\TT$-eigenvector fields $(E_i)_{i\in I}$ for $T$, where $E_i:\Lambda_i\to X$ is defined on some $\mathcal B$-perfect set $\Lambda_i\subset\TT$, such that ${\rm span}\left(\bigcup_{i\in I} E_i(\Lambda_i)\right)$ is dense in $X$.
Since the $E_i$ are assumed to be continuous in (ii), it is plain that (ii) implies (i). Conversely, assume that (i) holds true, and let $\mathbf Z$ be the set given by Lemma \[exhaust\].
Choose a countable dense set $\{ (x_i,\lambda_i);\; i\in\NN\}\subset\mathbf Z$, and let $({\varepsilon}_i)_{i\in\NN}$ be any sequence of positive numbers tending to $0$. Let also $d$ be a compatible metric $d$ on $X$ and put $B_i:=\{ x\in X;\; d(x,x_i)<{\varepsilon}_i\}$. By the definition of $\mathbf Z$ and since $\mathcal B$ is hereditary with respect to absolute continuity, one can find for each $i\in\NN$, a probability measure $\sigma_i\in\mathcal B$ such that $${\rm supp}(\sigma_i)\subset\{\lambda\in\TT;\; \exists x\in B_i\; :\; T(x)=\lambda x\}\, .$$
Put $\Lambda_i:={\rm supp}(\sigma_i)$, so that $\Lambda_i$ is $\mathcal B$-perfect. The set $$A_i=\{ (x,\lambda)\in B_i\times\Lambda_i;\; T(x)=\lambda x\}$$ is closed in the Polish space $B_i\times\Lambda_i$ and projects onto $\Lambda_i$. By the Jankov–von Neumann uniformization theorem (see [@K 18.A]), one can find a universally measurable map $E_i:\Lambda_i\to B_i$ such that $(E_i(\lambda), \lambda)\in A_i$ for every $\lambda\in \Lambda_i$. In other words we have a universally measurable $\TT$-eigenvector field $E_i:\Lambda_i\to X$ such that $d(E_i(\lambda),x_i)< {\varepsilon}_i$ for every $\lambda\in\Lambda_i$. By Lusin’s theorem on measurable functions (see [@K 17.D]), one can find a closed set $\Lambda'_i$ of positive $\sigma_i$-measure such that $(E_i)_{\vert \Lambda'_i}$ is continuous. So, upon replacing the measure $\sigma_i$ with its restriction $\sigma'_i$ to $\Lambda'_i$ (which is still in $\mathcal B$ since it is absolutely continuous with respect to $\sigma_i$) and $\Lambda_i$ with ${\rm supp}(\sigma_i')$, we may assume that in fact each $E_i$ is *continuous*. Since ${\rm span}\left(\bigcup_{i\in \NN} E_i(\Lambda_i)\right)$ is dense in $X$, the proof is complete.
When $\mathcal B$ is the family of continuous measures, the equivalence of (i) and (ii) is proved in [@Sophie Proposition 4.2].
The weak mixing case {#WMcase}
====================
In this section, we concentrate on part (1) of Theorem \[WS\]. So we assume that the $\TT$-eigenvectors of our operator $T\in\mathfrak L(X)$ are perfectly spanning, and we want to show that $T$ is weakly mixing in the Gaussian sense. For simplicity, we assume that $X$ is a *Banach* space in order to avoid working with sequences of continuous semi-norms.
Why things may not be obvious
-----------------------------
Before embarking on the proof, let us briefly explain why it could go wrong. In view of Proposition \[background\], we need to construct a $\TT$-eigenfield $(E,\phi)$ for $T$ on some measure space $(\Omega ,m)$ such that
- the operator $K_E : L^2(\Omega, m)\to X$ is gamma-radonifying;
- the vector field $E$ is $m$-spanning;
- for any $f\in L^1(\Omega, m)$, the image measure $\sigma_f=(f\, m)\circ\phi^{-1}$ is continuous.
To achieve (a) and (b), the most brutal way is to choose some total sequence $(x_n)_{n\in\NN}\subset B_X$ consisting of $\TT$-eigenvectors for $T$, say $T(x_n)=\lambda_n x_n$, to take $\Omega=\NN$ with the counting measure $m$, and to define $(E,\phi)$ by $E(n)=2^{-n}x_n$ and $\phi (n)=\lambda_n$. Then the vector field $E$ is obviously $m$-spanning, and the operator $K_E$ is gamma-radonifying since $K_E(e_n)=2^{-n}x_n$ and hence $\sum_0^\infty\Vert K_E(e_n)\Vert<\infty$. (Here, of course, $(e_n)$ is the canonical basis of $L^2(\Omega,m)=\ell^2(\NN)$). However, the image measure $\sigma =m\circ\phi^{-1}$ is purely atomic, so (c) is certainly not satisfied. (In fact, by [@Sophie Theorem 5.1], $T$ will never be ergodic with respect to any measure induced by a random variable of the form $\xi=\sum_0^\infty \chi_n x_n$, where $(x_n)$ is a sequence of $\TT$-eigenvectors for $T$ and $(\chi_n)$ is a sequence of independent, rotation-invariant centred random variables). To get (c) we must take a more complicated measure space $(\Omega, m)$ and avoid atoms; but then it will be harder to show that an operator defined on $L^2(\Omega, m)$ is gamma-radonifying. Thus, conditions (a) and (c) go in somewhat opposite directions, and we have to find a kind of balance between them.
In [@BG2] and [@BM], the measure space $(\Omega,m)$ was an open arc of $\mathbb T$ (with the Lebesgue measure), and the difficulty was partially settled by requiring some regularity on the $\TT$-eigenvector fields and combining this with the geometrical properties of the underlying Banach space. In the present paper, we will allow more freedom on the measure space, and this will enable us to get rid of any assumption on $X$.
The main part of the proof is divided into two steps. We shall first explain how to produce gamma-radonifying operators of the form $K_E$ in a reasonably flexible way, and then we show how this can be used to construct suitable $\TT$-eigenfields for $T$ under the perfectly spanning assumption. Once this has been done, the conclusion follows easily.
How to construct gamma-radonifying operators {#bordel01}
--------------------------------------------
The first part of our program relies essentially on the following observation. Let $G$ be a compact metrizable abelian group with normalized Haar measure $m_G$ and dual group $\Gamma$, and let $E:G\to X$ be a vector field on $(G, m_G)$. Then the operator $K_E:L^2(G,m_G)\to X$ is gamma-radonifying as soon as $$\sum_{\gamma\in \Gamma} \Vert\widehat E(\gamma )\Vert <\infty\, ,$$ where $\widehat E$ is the Fourier transform of $E$. This is indeed obvious since the characters of $G$ form an orthonormal basis $(e_\gamma)_{\gamma\in \Gamma}$ of $L^2(G,m_G)$ and $K_E(e_\gamma)=\widehat E(\gamma )$ for every $\gamma\in\Gamma$.
The compact group we shall use is the usual *Cantor group* $G:=\{ 0,1\}^\NN$, where addition is performed modulo 2. The dual group $\Gamma$ is identified with the set all of finite subsets of $\NN$, which we denote by $\rm FIN$. A set $I\in {\rm FIN}$ corresponds to the character $\gamma_I\in\Gamma$ defined by the formula $$\gamma_I(\omega)=\prod_{i\in I} {\varepsilon}_i(\omega)\, ,$$ where ${\varepsilon}_i(\omega_0, \omega_1, \dots )=(-1)^{\omega_i}$. An empty product is declared to be equal to $1$, so that $\gamma_\emptyset=\mathbf 1$.
It is common knowledge that any “sufficiently regular" function has an absolutely convergent Fourier series. In our setting, regularity will be quantified as follows. Let us denote by $d$ the usual ultrametric distance on $G$, $$d(\omega, \omega')=2^{-n(\omega,\omega')}\, ,$$ where $n(\omega,\omega')$ is the first $i$ such that $\omega_i\neq \omega'_i$. We shall say that a map $E:G\to X$ is *super-Lipschitz* if it is $\alpha$-Hölderian for some $\alpha >1$, $$\Vert E(\omega)-E(\omega')\Vert \leq C\, d(\omega,\omega')^\alpha$$ for any $\omega, \omega'\in G$ and some finite constant $C$. (Of course, there are no nonconstant super-Lipschitz maps on $\TT$; but life is different on the Cantor group).
The following lemma is the kind of result that we need.
\[halfkey0\] If $E: G\to X$ is super-Lipschitz, then $E$ has an absolutely convergent Fourier series.
Assume that $E$ is $\alpha$-Hölderian, $\alpha >1$, with constant $C$. The key point is the following
Let $n\in\NN$. If $I\in{\rm FIN}$ satisfies $I\ni n$ then $\Vert\widehat E(\gamma_I)\Vert\leq (C/2)\times 2^{-\alpha n}\, .$
Indeed, setting $\mathcal F_n:=\{ I\in{\rm FIN};\; I\neq\emptyset\;{\rm and}\;\max I=n\}$ (which has cardinality $2^n$), this yields that $$\sum_{I\in\mathcal F_n} \Vert \widehat E(\gamma_I)\Vert\leq 2^n\times (C/2)\times 2^{-\alpha n}=(C/2)\times 2^{-\beta n}$$ for all $n\geq 0$ (where $\beta=\alpha-1>0$), and the result follows.
Write $I=J\cup\{ n\}$, where $J\in {\rm FIN}$ and $n\not\in J$. Then $$\langle \gamma_I,\omega\rangle = {\varepsilon}_n(\omega) \langle \gamma_J,\omega\rangle\, ,$$ and hence $$\begin{aligned}
\widehat E(\gamma_I)&=&\int_{\{\omega;\; {\varepsilon}_n(\omega)=1\}}\langle \gamma_J,\omega\rangle\, E(\omega)\, dm_G(\omega) - \int_{\{\omega;\; {\varepsilon}_n(\omega)=-1\}}\langle \gamma_J,\omega\rangle\, E(\omega)\, dm_G(\omega)\\
&=&\int_{\{\omega;\; {\varepsilon}_n(\omega)=1\}} \langle \gamma_J,\omega\rangle\, \left[ E(\omega)-E(\omega+\delta^n)\right] dm_G(\omega) ,\end{aligned}$$ where $\delta^n\in G$ is the sequence which is $0$ everywhere except at the $n$-th coordinate. By assumption on $E$, we know that $\Vert E(\omega)-E(\omega+\delta^n)\Vert\leq C\times 2^{-\alpha n}$; and since the set $\{ {\varepsilon}_n=1\}$ has measure $1/2$, this concludes the proof.
Construction of suitable $\TT$-eigenfields {#bordel02}
------------------------------------------
The second part of our program is achieved by the following lemma.
\[superkey0\] Put $\mathbf V :=\{ (x,\lambda)\in X\times \TT;\; x\neq 0\;{\rm and}\; T(x)=\lambda x\}$, and let $\bf Z\subset \mathbf V$. Assume that for any (relatively) open set $O\subset\mathbf V$ such that $O\cap\mathbf Z\neq\emptyset$, the set $\{ \lambda\in \TT;\;\exists x\, :\, (x,\lambda)\in O\}$ is uncountable. [Then,]{} [for any $(x_0,\lambda_0)$ in $\bf Z$ and ${\varepsilon}>0$,]{} [one can construct a $\TT$-eigenfield $(E,\phi)$ for $T$ on the Cantor group $(G, m_G)$]{} [such that]{}
$\Vert E(\omega )-x_0\Vert<{\varepsilon}$ for all $\omega\in G$;
$E:G\to X$ is super-Lipschitz;
$\phi:G\to\TT$ is a homeomorphic embedding.
This is a standard Cantor-like construction. Let us denote by $\mathcal S$ the set of all finite $0$-$1$ sequences (including the empty sequence $\emptyset$). We write $\vert s\vert$ for the length of a sequence $s\in\mathcal S$, and if $i\in\{ 0,1\}$ we denote by $si$ the sequence “$s$ followed by $i$".
Put $V_\emptyset:=\TT$ and choose an open set $U_\emptyset\subset X$ with $x_0\in U_\emptyset$ and $\diam (U_\emptyset)<{\varepsilon}$. Let us also fix an arbitrary Hölder exponent $\alpha >1$. We construct inductively two sequences of open sets $(U_s)_{s\in\mathcal S}$ in $X$ and $(V_s)_{s\in\mathcal S}$ in $\TT$, such that the following properties hold for all $s\in\mathcal S$ and $i\in\{ 0,1\}$.
1. $\overline{U_{si}}\subset U_s$ and $\overline{V_{si}}\subset V_s$;
2. $\overline{U_{s0}}\cap\overline{U_{s1}}=\emptyset$ and $\overline{V_{s0}}\cap\overline{V_{s1}}=\emptyset$;
3. $\diam(U_s)\leq 2^{-\alpha\vert s\vert}$ and $\diam (V_{si})\leq\frac 12\, \diam (V_s)$;
4. $(U_s\times V_s)\cap\mathbf Z\neq \emptyset$.
The inductive step is easy. Assume that $U_s$ and $V_s$ have been constructed for some $s$. Since $(U_s\times V_s)\cap\mathbf Z\neq \emptyset$, we know that the set $\{ \lambda\in \TT;\;\exists x\, :\, (x,\lambda)\in (U_s\times V_s)\cap \mathbf Z\}$ is uncountable, and hence contains at least 2 distinct points $\lambda^0, \lambda^1$. Pick $x^i\in U_{s}$ such that $(x^i,\lambda^i)\in\mathbf Z$. Then $x^0\neq x^1$ because $T(x^i)=\lambda^i x^i$ and $x^i\neq 0$. Thus, one may choose small enough open sets $U_{si}\ni x^i$ and $V_{si}\ni \lambda^i$ to ensure (i),$\,\dots $, (iv) at steps $s0$ and $s1$.
For any $\omega\in G$ and $n\in\NN$, let us denote by $\omega_{\vert n}$ the finite sequence $(\omega_0,\dots ,\omega_n)\in\mathcal S$. By (i), (ii), (iii), the intersection $\bigcap_{n\geq 0} U_{\omega_{\vert n}}$ is a single point $\{ x_\omega\}$, and the intersection $\bigcap_{n\geq 0} V_{\omega_{\vert n}}$ is a single point $\{ \lambda_\omega\}$. Moreover, the map $\omega\mapsto \lambda_{\omega}:=\phi(\omega)$ is a homeomorphic embedding, and the map $\omega\mapsto x_\omega:=E(\omega)$ is $\alpha$-Hölderian and hence super-Lipschitz (by (iii)). Finally, it follows easily from the continuity of $T$ that $T(x_\omega)=\lambda_\omega x_\omega$ for every $\omega\in G$. In other words, $(E,\phi)$ is a $\TT$-eigenfield for $T$. Since $\Vert E(\omega)-x_0\Vert\leq \diam(U_\emptyset)<{\varepsilon}$ for all $\omega\in G$, this concludes the proof.
The proof {#proofWM}
---------
We now just have to put the pieces together. Let us recall what we are trying to do: we are given an operator $T\in\mathfrak L(X)$ whose $\TT$-eigenvectors are perfectly spanning, and we want to show that $T$ is weakly mixing in the Gaussian sense.
Let $\mathbf V :=\{ (x,\lambda)\in X\times \TT;\; x\neq 0\;{\rm and}\; T(x)=\lambda x\}$. Applying Lemma \[exhaust\] to the family of continuous measures, we get a closed set $\mathbf Z\subset \mathbf V$ with the following properties:
- $\mathbf Z$ satisfies the assumption of Lemma \[superkey0\], for any $O\subset \mathbf V$ open such that $O\cap \mathbf Z\neq \emptyset$, the set $\{ \lambda\in \TT;\;\exists x\, :\, (x,\lambda)\in O\}$ is uncountable;
- ${\rm span}\left( \pi_1(\mathbf Z)\right)$ is dense in $X$.
Let us fix a countable dense set $\{ (x_i,\lambda_i);\; i\in\NN\}\subset\mathbf Z$, and let us apply Lemma \[superkey0\] to each point $(x_i,\lambda_i)$, with ${\varepsilon}_i=2^{-i}$. Taking Lemma \[halfkey0\] into account and since ${\varepsilon}_i\to 0$, this give a sequence of $\TT$-eigenfields $(E_i,\phi_i)$ defined on $(\Omega_i, m_i):=(G,m_G)$, such that
- each operator $K_{E_i} :L^2(\Omega_i, m_i)\to X$ is gamma-radonifying;
- each $E_i$ is continuous and $\overline{\rm span}\left(\bigcup_{i\in\NN} E_i(\Omega_i)\right)=X$;
- each $\phi_i$ is one-to-one.
Now, let $(\Omega, m)$ be the “disjoint union" of the measure spaces $(\Omega_i,m_i)$ (so that $L^2(\Omega, m)$ is the $\ell^2$-direct sum $\oplus_i L^2(\Omega_i, m_i)$). Choose a sequence of small positive numbers $(\alpha_i)_{i\in\NN}$, and define a $\TT$-eigenfield $(E,\phi)$ on $(\Omega, m)$ as follows: $E(\omega_i)=\alpha_i E_i(\omega_i)$ and $\phi(\omega_i)=\phi_i(\omega_i)$ for each $i$ and every $\omega_i\in\Omega_i$. If the $\alpha_i$ are small enough, then $E$ is indeed in $L^2(\Omega, m)$ and (using property (1)) the operator $K_E:L^2(\Omega, m)\to X$ is gamma-radonifying. By (2) and since $m$ has full support, the vector field $E$ is $m$-spanning and hence the operator $K_E$ has dense range. Moreover, since $TK_{E_i}=K_{E_i}M_{\phi_i}$ for all $i$, the intertwining equation $TK_E=K_EM_\phi$ holds. Finally, since $(\Omega_i,m_i)$ is atomless, it follows from (3) that each measure $m_{i}\circ\phi_i^{-1}$ is continuous, and hence that $\sigma=m\circ\phi^{-1}$ is continuous as well. By Proposition \[background\], we conclude that $T$ is weakly mixing in the Gaussian sense.
Proof of the abstract results (1) {#proofabstract1}
=================================
In this section, we prove Theorem \[abstract\]. The proof runs along exactly the same lines as that of the weak mixing case: we first explain how to produce gamma-radonifying operators of the form $K_E$ in a flexible way, then we show how this can be used to construct suitable $\TT$-eigenfields for $T$ under the $\mathbf S$-perfectly spanning assumption, and the conclusion follows easily. However, the first two steps are technically more involved than the corresponding ones from the weak mixing case, the difficulty being not less important when dealing with the strong mixing case only. Roughly speaking, the main reason is that it is much harder for a measure to be Rajchman, or more generally to be $\mathbf S$-continuous, than to be merely continuous.
To avoid artificial complications, we shall first assume that the underlying space $X$ is a Banach space (which will dispense us from using sequences of semi-norms), and we will indicate only at the very end of the proof how it can be adapted in the Fréchet space case (this is really a matter of very minor changes). Thus, in most of this section, $T$ is a linear operator acting on a complex separable Banach space $X$. We are also given a $c_0$-like translation ideal $\mathbf S\subset\ell^\infty (\ZZ_+)$ such that every $\mathbf S$-continuous measure is continuous. We are assuming that the $\TT$-eigenvectors of $T$ are $\mathbf S$-perfectly spanning, and our aim is to show that $T$ is $\mathbf S$-mixing in the Gaussian sense.
How to construct gamma-radonifying operators {#bordel1}
--------------------------------------------
Just as in the weak mixing case, the guiding idea of the first part of our program is the observation that we made in sub-section \[bordel01\], namely that if $E:\Omega\to X$ is a vector field defined on a compact abelian group $\Omega$ (with Haar measure $m$ and dual group $\Gamma$), then the operator $K_E:L^2(\Omega,m)\to X$ is gamma-radonifying as soon as $$\sum_{\gamma\in \Gamma} \Vert\widehat E(\gamma )\Vert <\infty\, .$$
Unfortunately, we are not able to use this observation as stated. Instead of compact groups, we will be forced to use some slightly more complicated objects $(\Omega ,m)$, due to the structure of the inductive construction that will be performed in the *second part* of our program.
Let us denote by $\mathfrak Q$ the set of all finite sequences of integers $\bar q=(q_1,\dots ,q_l)$ with $q_s\geq 2$ for all $s$. For any $\bar q=(q_1,\dots ,q_l)\in\mathfrak Q$, we put $$\Omega(\bar q):=\bigsqcup_{s=1}^l\Omega(q_s)\, ,$$ where $\Omega (q)=\{ \xi\in\TT;\; \xi^{q}=\bf 1\}$ is the group of all $q$-roots of $\bf 1$, and the symbol $\sqcup$ stands for a “disjoint union".
The following notation will be useful: for any $\bar q=(q_1,\dots ,q_l)\in\mathfrak Q$, we set $$l(\bar q)=l\;\;{\;\rm and}\;\;\;w(\bar q) =q_1+\dots +q_l .$$ In particular, $\Omega(\bar q)$ has cardinality $w( \bar q)$.
We endow each finite group $\Omega (q)$ with its normalized Haar measure $m_q$ ( the normalized counting measure), and each $\Omega(\bar q)=\Omega (q_1,\dots ,q_l)$ with the probability measure $$m_{\bar q}=\frac{1}{l(\bar q)}\sum_{s=1}^{l(\bar q)} m_{q_s}\, .$$ Here, of course, $\Omega (q_s)$ is considered as a subset of $\Omega(\bar q)$, so that the measures $m_{q_s}$ are disjointly supported.
For any infinite sequence $\mathbf q=(\bar q_n)_{n\geq 1}\in\mathfrak Q^\NN$, we put $$\Omega (\mathbf q):=\prod_{n=1}^\infty \Omega (\bar q_n)\, ,$$ and we endow $\Omega (\mathbf q)$ with the product measure $$m_{\mathbf q}=\bigotimes_{n=1}^\infty m_{\bar q_n}\, .$$
Given $\mathbf q=(\bar q_n)_{n\geq 1}\in \mathfrak Q^\NN$, it is not difficult to describe a “canonical" orthonormal basis for $L^2(\Omega (\mathbf q),m_{\mathbf q})$. However, we have to be careful with the notation.
For each sequence $\bar q=(q_1,\dots ,q_l)\in\mathfrak Q$, let us denote by $\Gamma (\bar q)$ the disjoint union $\bigsqcup_{s=1}^l \Gamma(q_s)$, where $\Gamma (q)$ is the dual group of $\Omega (q)$, $\Gamma(q)=\ZZ_q$. For any $\gamma_s\in \Gamma (q_s)\subset\Gamma (\bar q)$, define a function $e_{\gamma_s}:\Omega (\bar q)\to \CC$ by $$e_{\gamma_s} (\xi)=\left\{
\begin{array}{cl}
\sqrt{l(\bar q)}\, \langle \gamma_s, \xi\rangle & {\rm if\;\;}\xi\in\Omega (q_s),\\
0&{\rm otherwise.}
\end{array}
\right.$$
Since the characters of $\Omega (q)$ form an orthonormal basis of $L^2(\Omega (q), m_q)$ for every positive integer $q$, it is clear that the $e_{\gamma_s}$, $\gamma_{s}\in\Gamma (\bar q)$ form an orthonormal basis of $L^2(\Omega (\bar q), m_{\bar q})$ for every $\bar q\in \mathfrak Q$.
Now, write each $\bar q_n$ as $\bar q_n=(q_{n,1}, \dots ,q_{n,l_n})$, and let us denote by $\Gamma (\mathbf q)$ the set of all finite sequences of the form $\gamma=(\gamma_{s_1},\dots ,\gamma_{s_N})$ with $\gamma_{s_n}\in\Gamma (q_{n,s_n})\subset \Gamma (\bar q_n)$ for all $n$ and $\gamma_{s_N}\neq 0$. For any $\gamma=(\gamma_{s_1},\dots ,\gamma_{s_N})\in\Gamma (\mathbf q)$, we define $e_\gamma :\Omega (\mathbf q)\to \CC$ as expected:
$$e_\gamma (\omega)=\prod_{n=1}^N e_{\gamma_{s_n}}(\omega _n)\, .$$ In other words, $e_\gamma=e_{\gamma_{s_1}}\otimes\cdots\otimes e_{\gamma_{s_N}}$. We also include the empty sequence $\emptyset$ in $\Gamma (\mathbf q)$ and we put $e_\emptyset=\mathbf 1$. The following lemma is essentially obvious:
The family $(e_\gamma)_{\gamma\in\Gamma (\mathbf q)}$ is an orthonormal basis of $L^2(\Omega (\mathbf q), m_{\bf q})$, for any $\mathbf q\in\mathfrak Q^\NN$.
We note that if $\mathbf q=(\bar q_n)_{n\geq 1}\in\mathfrak Q^\NN$ and each sequence $\bar q_n$ has length $1$, $\bar q_n=(q_n)$ for some $q_n\geq 2$, then $(\Omega (\mathbf q), m_{\mathbf q})$ is just the compact group $\prod_{n\geq 1} \Omega (q_n)$ with its normalized Haar measure, and $\Gamma (\mathbf q)$ “is" the character group of $\Omega (\mathbf q)$. Moreover, if $E:\Omega(\mathbf q)\to X$ is a vector field on $(\Omega (\mathbf q), m_{\mathbf q})$ then $K_E(e_\gamma)$ is the Fourier coefficient $\widehat E(\gamma )$, for any $\gamma\in\Gamma (\mathbf q)$. Accordingly, we shall use the following notation for an *arbitrary* $\mathbf q\in\mathfrak Q^\NN$: given a vector field $E:\Omega(\mathbf q)\to X$ on $(\Omega (\mathbf q), m_{\mathbf q})$, we set $$\widehat E(\gamma )=K_E(e_\gamma)$$ for every $\gamma\in\Gamma (\mathbf q)$.
Next, we introduce a partially defined “metric" on every $\Omega (\mathbf q)$, as follows. Write $\mathbf q=(\bar q_n)_{n\geq 1}$ and $\bar q_n=(q_{n,1},\dots , q_{n,l_n})$. If $\omega=(\omega_n)$ and $\omega'=(\omega'_n)$ are distinct elements of $\Omega (\bf q)$, let us denote by $n(\omega, \omega')$ the smallest $n$ such that $\omega_n\neq \omega'_n$. If $\omega_n$ and $\omega'_n$ are in the same $\Omega (q_{n(\omega,\omega'), s})$, we denote this $s$ by $s(\omega,\omega')$ and we put $$d_{\mathbf q}(\omega,\omega'):=\frac{1}{w(\bar q_1) \cdots w(\bar q_{n(\omega,\omega')-1})}\times \frac{1}{l_{n(\omega,\omega')}^{1/4}\, q_{n(\omega,\omega'), s(\omega,\omega')}}\, \cdot$$ (The value of an empty product is declared to be $1$). Otherwise, $d_{\mathbf q} (\omega,\omega')$ is not defined.
Let $\mathbf q\in\mathfrak Q^\NN$. We shall say that a map $E:\Omega(\mathbf q)\to X$ is *super-Lipschitz* if $$\Vert E(\omega)-E(\omega')\Vert\leq C\, d_{\mathbf q} (\omega,\omega')^2$$ for some finite constant $C$, whenever $d_{\mathbf q}(\omega,\omega')$ is defined.
The terminology is arguably not very convincing, since super-Lipschitz maps need not be continuous. Moreover, the definition of $d_{\bf q}$ may look rather special, mainly because of the strange term $l_{n(\omega,\omega')}^{1/4}$. We note, however, that when all sequences $\bar q_n$ have length $1$, say $\bar q_n=(q_n)$, then $d_{\mathbf q}$ is a quite natural true (ultrametric) distance on the Cantor-like group $\Omega(\mathbf q)=\prod_{n\geq 1} \Omega (q_n)$: $$d_{\mathbf q}(\omega,\omega')=\frac{1}{q_1\cdots q_{n(\omega, \omega')}}\, \cdot$$ In this case, the terminology “super-Lipschitz" seems adequate (forgetting that we allowed arbitrary Hölder exponents $\alpha >1$ in the weak mixing case). More importantly, the following lemma is exactly what is needed to carry out the first part of our program.
\[halfkey\] Let $\mathbf q\in\mathfrak Q^\NN$. If $E:\Omega(\mathbf q)\to X$ is a super-Lipschitz vector field on $(\Omega (\mathbf q), m_{\mathbf q})$, then $$\displaystyle\sum_{\gamma\in\Gamma(\mathbf q)} \Vert \widehat E(\gamma)\Vert<\infty\, .$$
Throughout the proof we put $\mathbf q=(\bar q_n)_{n\geq 1}$, and each $\bar q_n$ is written as $\bar q_n=(q_{n,1},\dots ,q_{n, l_n})$.
For notational simplicity, we write $\Gamma$ instead of $\Gamma (\mathbf q)$. We denote by $\vert\gamma\vert$ the length of a sequence $\gamma\in\Gamma$. That is, $\vert\emptyset\vert=0$ and $\vert \gamma\vert =N$ if $\gamma=(\gamma_{s_1}, \dots ,\gamma_{s_N})$ with $\gamma_{s_n}\in \Gamma (q_{n,s_n})$ for all $n$ and $\gamma_{N, s_N}\neq 0$. Then $\Gamma$ is partitioned as $$\Gamma=\bigcup_{N=0}^\infty \Gamma_N\,$$ where $\Gamma_N=\{\gamma\in\Gamma;\; \vert\gamma\vert=N\}$.
For any $\gamma=(\gamma_{s_1},\dots ,\gamma_{s_N})\in\Gamma_N$, we put $s(\gamma)=s_N$. In other words, $s(\gamma)$ is the unique $s\in\{ 1,\dots ,l_N\}$ such that the $N$-th coordinate of $\gamma$ belongs to $\Omega(q_{N, s})$.
If $E:\Omega(\mathbf q)\to X$ is super-Lipschitz with constant $C$, then $$\Vert\widehat E(\gamma)\Vert\leq \frac{C}{l_N}\,\sqrt{l_1\cdots l_{N-1}} \prod_{n\leq N-1} w(\bar q_n)^{-2}\times \frac{1}{q_{N,s(\gamma)}^2}$$ for every $\gamma\in \Gamma_N$, $N\geq 1$.
Let us fix $\gamma=(\gamma_{s_1},\dots ,\gamma_{s_N})\in\Gamma_N$, so that $s(\gamma)=s_N$. For notational simplicity (again), we put $q_\gamma=q_{N, s_N}=q_{N,s(\gamma)}$. For any $\xi\in\Omega (q_{\gamma})$ and $\omega\in\Omega(\mathbf q)$ with $\omega_N\in\Omega (q_{\gamma})$, let us denote by $\xi\omega$ the element $\omega'$ of $\Omega(\mathbf q)$ defined by $\omega'_n=\omega_n$ if $n\neq N$ and $\omega'_N=\xi\omega_N$. Then $$\begin{aligned}
\widehat E(\gamma)&=&\int_{\{\omega;\;\omega_N\in\Omega(q_{\gamma})\}} e_\gamma (\omega)\, E(\omega)\, dm_{\bf q}(\omega)\\
&=&\sum_{\xi\in\Omega(q_{\gamma})}\int_{\{\omega;\; \omega_N=\xi\}}e_\gamma (\omega)\, E(\omega)\, dm_{\bf q}(\omega)\\
&=&\sum_{\xi\in\Omega(q_{\gamma})} \int_{\{\omega;\; \omega_N=\mathbf 1_{\Omega (q_\gamma)}\}} e_\gamma( \xi\omega)\, E(\xi\omega)\, dm_{\bf q}(\omega)\\
&=&\int_{\{\omega;\;\omega_N=\mathbf 1_{\Omega (q_\gamma)}\}} e_\gamma (\omega)\times\left(\sum_{\xi\in\Omega(q_{\gamma})} {\langle\gamma_{s_N}, \xi\rangle}E(\xi\omega)\right) dm_{\bf q}(\omega).\end{aligned}$$
Now, $\gamma_{s_N}$ is assumed to be a non-trivial character of $\Omega(q_{\gamma})$. So we have $$\sum_{\xi\in\Omega (q_{\gamma})} {\langle\gamma_{s_N}, \xi\rangle}=0\, ,$$ and it follows that $$\widehat E(\gamma)=\int_{\{\omega;\;\omega_N=\mathbf 1_{\Omega (q_\gamma)}\}} e_\gamma (\omega)\times\left[\sum_{\xi\in\Omega(q_{\gamma})} {\langle\gamma_{s_N}, \xi\rangle}\, \Bigl(E(\xi\omega)-E(\omega)\Bigr)\right] dm_{\mathbf q}(\omega)\, .$$
By the definition of a super-Lipschitz map, since $\vert e_\gamma (\omega)\vert\leq \sqrt{l_1\cdots l_N}$ and since the set $\{\omega;\; \omega_N=\mathbf 1_{\Omega(q_\gamma)}\}$ has $m_{\bf q}$-measure $1/l_Nq_{\gamma}$, we conclude that $$\begin{aligned}
\Vert \widehat E(\gamma)\Vert&\leq &\sqrt{l_1\cdots l_N}\times \frac{1}{l_N}\times C\prod_{n\leq N-1} w( \bar q_n)^{-2}\times \frac{1}{l_N^{1/2}\, q_{N,s(\gamma)}^2}\, , \end{aligned}$$ which is the required estimate.
It is now easy to conclude the proof of the lemma. For each $N\geq 1$ and every $s\in\{ 1,\dots ,l_N\}$, the set $\Gamma_{N,s}=\{ \gamma\in\Gamma_N;\; s(\gamma)=s\}$ has cardinality $$\vert\Gamma_{N,s}\vert=\prod_{n\leq N-1} w( \bar q_n)\times q_{N,s}\, .$$ By the above fact and since $w(\bar q_n)=q_{n,1}+\dots +q_{n,l_n}\geq 2l_n$ for all $n$, it follows that
$$\begin{aligned}
\sum_{\gamma\in\Gamma_{N, s}} \Vert \widehat E(\gamma)\Vert&\leq&\frac{C}{l_N}\,\sqrt{l_1\cdots l_{N-1}}\,\prod_{n\leq N-1} w(\bar q_n)^{-1}\times \frac{1}{q_{N,s}}\\
&\leq& \frac{C}{l_N}\, 2^{-N}\end{aligned}$$
for each $s\in\{ 1,\dots ,l_N\}$. Hence, we get $\sum\limits_{\gamma\in\Gamma_{N}} \Vert \widehat E(\gamma)\Vert\leq C\, 2^{-N}$ for every $N\geq 1$, and the result follows.
Construction of suitable $\TT$-eigenfields {#bordel2}
------------------------------------------
We now turn to the second part of our program, which is the most technical one. Our aim is to prove the following lemma.
\[superkey\] Put $\mathbf V :=\{ (x,\lambda)\in X\times \TT;\; x\neq 0\;{\rm and}\; T(x)=\lambda x\}$, and let $\bf Z$ be a closed subset of $\mathbf V$. Assume that for any (relatively) open set $O\subset\mathbf V$ such that $O\cap\mathbf Z\neq\emptyset$, the set $\{ \lambda\in \TT;\;\exists x\, :\, (x,\lambda)\in O\}$ is not $\mathbf S$-small. [Then,]{} [for any $(x_0,\lambda_0)$ in $\bf Z$ and ${\varepsilon}>0$,]{} [one can construct a $\TT$-eigenfield $(E,\phi)$ for $T$ on some $(\Omega(\mathbf q), m_{\mathbf q})$]{} [such that]{}
$\Vert E(\omega )-x_0\Vert<{\varepsilon}$ for all $\omega\in \Omega(\mathbf q)$;
$E:\Omega(\mathbf q)\to X$ is super-Lipschitz;
$\phi:\Omega(\mathbf q)\to\TT$ is a homeomorphic embedding, and the image measure $\sigma=m_{\bf q}\circ\phi^{-1}$ is $\mathbf S$-continuous.
Throughout the proof, we denote (as usual) by $\mathcal M(\TT)$ the space of all complex measures on $\TT$ endowed with the total variation norm $\Vert\hskip 0.6mm\cdot\hskip 0.6mm\Vert_{}$. We recall that $\mathcal M(\TT)$ is the dual space of $\mathcal C(\TT)$ (the space of all continuous complex-valued functions on $\TT$), so we can use the $w^*$ topology on $\mathcal M(\TT)$. Since our family $\mathbf S$ is $c_0$-like, we may fix a uniformly bounded sequence of $w^*$-$\,$continuous semi-norms $(\Phi_k)_{k\geq 0}$ on $\ell^\infty (\ZZ_+)$ such that $$\mathbf S=\left\{ a\in\ell^\infty (\ZZ_+);\; \Phi_k(a)\xrightarrow{k\to\infty} 0\right\}\, .$$ Without loss of generality, we may assume that $\Phi_k(a)\leq \Vert a\Vert_\infty$ for all $k$ and every $a\in\ell^\infty (\ZZ_+)$. Moreover, upon replacing $\Phi_k(a)$ by $\max(\vert a_k\vert, \Phi_k(a))$, we may also assume that $\Phi_k(a)\geq \vert a_k\vert$ for every $a\in\ell^\infty(\ZZ_+)$.
Finally, to avoid notational heaviness, we denote by $\widehat\sigma$ the *positive part* of the Fourier transform of a measure $\sigma\in\mathcal M(\TT)$, we write $\widehat\sigma$ instead of $\widehat\sigma_{\vert \ZZ_+}$ or $\mathcal F_+(\sigma)$. We then have $$\vert\widehat\sigma(k)\vert\leq \Phi_k(\widehat\sigma)\leq \Vert\widehat\sigma\Vert_\infty\leq \Vert \sigma\Vert$$ for all $k$ and every $\sigma\in\mathcal M(\TT)$. In particular, if a bounded sequence $(\sigma_n)\subset\mathcal M(\TT)$ is such that the sequence $(\widehat\sigma_n)$ is Cauchy with respect to each semi-norm $\Phi_k$, then $(\sigma_n)$ is $w^*$ convergent in $\mathcal M(\TT)$.
Let us introduce some terminology. By an *admissible sequence of open sets in $\TT$*, we shall mean a finite sequence of open sets $(V_i)_{i\in I}\subset\TT$ such that the $V_i$ have pairwise disjoint closures. An admissible sequence $(W_j)_{j\in J}$ is *finer* than an admissible sequence $(V_i)_{i\in I}$ if the following hold:
- for every $j\in J$, one can find $i\in I$ such that ${\overline{W_j}}\subset V_i$; then $j$ is called a *successor* of $i$;
- all $i\in I$ have the same number of successors $j\in J$.
In this situation, we write $i\prec j$ when $j\in J$ is a successor of $i\in I$. We define in the same way admissible sequences of open sets $(U_i)_{i\in I}$ in $X$ and the corresponding refinement relation.
An *admissible pair* is a pair $(\sigma, (V_i)_{i\in I})$ where $(V_i)_{i\in I}$ is an admissible sequence of open sets in $\TT$ and $\sigma$ is a positive $\mathbf S$-continuous measure such that
- $\supp(\sigma)\subset\bigcup_{i\in I} V_i$;
- all open sets $V_i$ have the same $\sigma$-measure.
An admissible pair $(\sigma, (V_i)_{i\in I})$ is *normalized* if $\sigma$ is a probability measure.
Let $(\sigma, (V_i)_{i\in I})$ be a normalized admissible pair. Given $N\in\NN$ and $\eta >0$, one can find two admissible sequences of open sets $(V_j')_{j\in J}$ and $(W_j)_{j\in J}$ (with the same index set $J$) and an $\mathbf S$-continuous probability measure $\nu$ such that
- $(V_j')_{j\in J}$ is finer than $(V_i)_{i\in I}$ and $\diam(V_j')<\eta$ for all $j\in J$;
- $(\nu, (W_j)_{j\in J})$ is an admissible pair finer than $(\sigma, (V_i)_{i\in I})$;
- $\supp(\nu)\subset\supp(\sigma)$;
- $\supp(\sigma)\cap V_j'\neq\emptyset$ for all $j\in J$;
- ${\bigcup_j {\overline{V_j'}}}\cap{\bigcup_j{\overline{W_j}}}=\emptyset$;
- $\Vert\nu-\sigma\Vert<\eta$;
- whenever $\sigma'$ is a probability measure such that $(\sigma', (V'_j)_{j\in J})$ is an admissible pair, it follows that $\Phi_k(\widehat{\sigma}'-\widehat\nu)<\eta$ for all $k\leq N$.
Let $\eta'$ and $\eta''$ be small positive numbers to be chosen later.
First, we note that since the measure $\sigma$ is *continuous* by assumption on $\mathbf S$, one can partition $\supp(\sigma)\cap V_i$, $i\in I$ into Borel sets $A_{i,1},\dots ,A_{i, K_i}$ with $\diam(A_{i,s})<\eta'$ and $\sigma(A_{i,1})=\dots =\sigma(A_{i,K_i})$. This can be done as follows. Split $\supp(\sigma)\cap V_i$ into finitely many Borel sets $B_1,\dots, B_N$ with diameters less than $\eta'/2$ and positive $\sigma$-measure. Using the continuity of $\sigma$, choose a Borel set $B'_1\subset B_1$ such that $\sigma (B'_1)>0$ and $\sigma (B'_1)/\sigma (V_i)$ is a *rational* number. Then choose a Borel set $B'_2$ such that $B_1\setminus B'_1\subset B'_2\subset (B_1\setminus B'_1)\cup B_2$ and $\sigma (B'_2)/\sigma (V_i)$ is a positive rational number, and so on. This gives a partition of $\supp(\sigma)\cap V_i$ into pairwise disjoint Borel sets $B'_1,\dots ,B'_N$ with diameter less than $\eta'$ such that $\sigma (B'_k)/\sigma (V_i)$ is a positive rational number for each $k$, say $\sigma (B'_k)=\frac{p_k}{q}\, \sigma (V_i)$ where $p_k, q\in\NN$ (the same $q$ for all $k$). Finally, use again the continuity of $\sigma$ to split each $B'_k$ into $p_k$ Borel sets $B''_{k,1},\dots ,B''_{k,p_k}$ with $\sigma (B''_{k,s})=\frac{\sigma(V_i)}{q}$, and relabel the collection of all these sets $B''_{k,s}$ as $A_{i,1},\dots ,A_{i, K_i}$. That all the splittings can indeed be made follows from the ($1$-dimensional case of the) classical *Liapounoff convexity theorem*; see [@Rudin Theorem 5.5].
Next, given any positive number $m_i$, one can partition further each $A_{i,s}$ into $m_i$ Borel sets $B$ with the same $\sigma$-measure. Taking $m_i:=\prod_{j\neq i} K_j$, we then have the same number of Borel sets $B$ inside each open set $V_i$. Thus, we may in fact assume from the beginning that we have the same number of sets $A_{i,s}$ inside each $V_i$. We denote this number by $K$, and we put $J:=I\times\{ 1,\dots ,K\}$. We note that $\sigma(A_{i,s})=\frac{1}{\vert J\vert}$ for all $(i,s)\in J$.
Since the measure $\sigma$ is regular and continuous, one can pick, for each $(i,s)\in J$, a compact set $C_{i,s}\subset A_{i,s}$ such that $0<\sigma(A_{i,s}\setminus C_{i,s})<\eta''$, and then a point $a_{i,s}\in (A_{i,s}\setminus C_{i,s})\cap\supp(\sigma)$. Then we may choose open sets $W_{i,s}\supset C_{i,s}$ with pairwise disjoint closures contained in $V_i$, and open sets $V_{i,s}'\ni a_{i,s}$ with pairwise disjoint closures contained in $V_i$ and $\diam(V'_{i,s})<\eta'$, in such a way that ${\bigcup_{j\in J} {\overline{V_j'}}}\cap{\bigcup_{j\in J}{\overline{W_j}}}=\emptyset$.
If $\eta''$ is small enough, then the probability measure $$\nu =\frac 1{\vert J\vert}\, \sum_{(i,s)\in J}\frac{\sigma_{\vert C_{i,s}}}{\sigma (C_{i,s})}$$ satisfies $\Vert\sigma-\nu\Vert <\eta$. Moreover, $(\nu ,(W_j)_{j\in J})$ is an admissible pair finer than $(\sigma, (V_i)_{i\in I})$.
Let us denote by $\omega_f$ the modulus of continuity of a function $f\in\mathcal C(\TT)$: $$\omega_f(\delta)=\sup\{ \vert f(u)-f(v)\vert;\; \vert u-v\vert<\delta\}\, .$$ Since the sets $C_{i,s}$, $(i,s)\in J$ form a partition of $\supp(\nu)$ with $\nu(C_{i,s})=\frac{1}{\vert J\vert}$, and since $\vert z-a_{i,s}\vert\leq \diam(A_i{,s})<\eta'$ for all $z\in C_{i,s}$, we have $$\left\vert\int_\TT f\, d\nu-\frac{1}{\vert J\vert}\sum_{(i,s)\in J} f(a_{i,s})\right\vert\leq \omega_f(\eta')$$ for any $f\in\mathcal C(\TT)$. Similarly, if $\sigma'$ is any probability measure with $\supp(\sigma')\subset\bigcup_{j\in J} V'_j$ and $\sigma'(V_j')=\frac{1}{\vert J\vert}$ for all $j\in J$, then $$\left\vert\int_\TT f\, d\sigma'-\frac{1}{\vert J\vert}\sum_{(i,s)\in J} f(a_{i,s})\right\vert\leq \omega_f(\eta')\, .$$ Hence, we see that $\sigma'$ is close to $\nu$ in the $w^*$ topology of $\mathcal M(\TT)$ if $\eta'$ is sufficiently small. Since each semi-norm $\Phi_k$ is $w^*$-$\,$continuous and since the Fourier transformation $\mathcal F_+:\mathcal M(\TT)\to\ell^\infty(\ZZ_+)$ is $(w^*,w^*)\,$-$\,$continuous on bounded sets, it follows that if $\eta'$ is small enough then, for any $\sigma'$ such that $(\sigma', (V_j')_{j\in J})$ is an admissible pair, we do have $\Phi_k(\widehat{\sigma}'-\widehat\nu)<\eta$ for all $k\leq N$.
When considering two sequences of open sets $(V_j')_{j\in J}$ and $(W_j)_{j\in J}$ both finer than a given sequence $(V_i)_{i\in I}$ and with the same index set $J$, we will always assume that the corresponding “extension" relations $\prec$ on $I\times J$ are in fact the same.
At this point, we need to introduce some more terminology. By a *convenient triple*, we mean a triple $(\sigma, (U_i)_{i\in I}, (V_i)_{i\in I})$, where $(U_i)\subset X$ and $(V_i)\subset\TT$ are admissible sequences of open sets (with the same index set $I$), and $\sigma$ is a positive $\mathbf S$-continuous measure such that the pair $(\sigma , (V_i)_{i\in I})$ is admissible and the following holds: for each $i\in I$ one can find a closed set $F_i\subset X$ such that $$F_i\subset U_i\setminus\{ 0\}\;\;{\rm and}\;\;\supp (\sigma)\subset\bigcup_{i\in I}\{ \lambda\in V_i;\; \exists x\in F_i\, :\, (x,\lambda)\in \mathbf Z\}\, .$$
There is a natural notion of refinement for convenient triples, which is defined exactly as for admissible pairs. Moreover, we shall say that two convenient triples $(\sigma, (U_i), (V_i))$ and $(\sigma', (U_{i'}), (V_{i'}))$ are *disjoint* if $\bigcup_i {\overline{U_i}}\cap \bigcup_{i'} {\overline{U_{i'}}}=\emptyset$ and $\bigcup_i {\overline{V_i}}\cap \bigcup_{i'} {\overline{V_{i'}}}=\emptyset$.
The following fact is the key point in order to prove Lemma \[superkey\]: it is the basic inductive step towards the construction of the measure space $(\Omega(\mathbf q),m_{\mathbf q})$ and the map $\phi$. The technical difficulty comes essentially from condition (c), which is here to ensure (3) in Lemma \[superkey\].
Let $(\sigma, (U_i)_{i\in I}, (V_i)_{i\in I})$ be a normalized convenient triple, and let $\eta >0$. Then one can find a positive integer $l=l(\eta)$ and a finite sequence of pairwise disjoint normalized convenient triples $(\sigma'_1, (U_j')_{j\in J_1}, (V_j')_{j\in J_1}),
\dots,(\sigma'_l, (U_j')_{j\in J_l}, (V_j')_{j\in J_l})$ finer than $(\sigma, (U_i),(V_i))$ such that
- $\diam(V_j')<\eta$ for all $s\in\{ 1,\dots ,l\}$ and every $j\in J_s$;
- $\diam(U_j')<\frac{\eta}{l(\eta)^{1/2}\vert J_s\vert^2}$ for all $s$ and every $j\in J_s$;
- If we put $$\sigma':=\frac{1}{l}\sum_{s=1}^l \sigma'_s$$ then $\sup_{k\geq 0}\,\Phi_k(\widehat{\sigma}'-\widehat\sigma)\leq\eta$.
Put $\nu_0=\sigma=\sigma'_0$, $J_0=I$ and $W_j=V_j$, $j\in J_0$. Let also $\eta^*$ be a positive number to be chosen later but depending only on $\eta$, and let $l$ be the smallest integer such that $l>1/\eta^*$.
Since $\nu_0$ is $\mathbf S$-continuous, one can find $N_0\in\NN$ such that $\Phi_k(\widehat{\nu}_0)<\eta^*$ for all $k>N_0$.
Applying Fact 1 to the pair $(\sigma, (V_i)_{i\in I})=(\nu_0, (V_j)_{j\in J_0})$, we find an admissible sequence of open sets $(V'_j)_{j\in J_1}$ and a normalized admissible pair $(\nu_1, (W_j)_{j\in J_1})$ such that
- $(V_j')_{j\in J_1}$ is finer than $(W_j)_{j\in J_{0}}$ and $\diam(V_j')<\eta^*$ for all $j\in J_1$;
- $(\nu_1, (W_{j})_{j\in J_1})$ is finer than $(\nu_0, (W_{j})_{j\in J_0})$;
- $\supp(\nu_1)\subset\supp(\nu_0)$;
- $\supp(\nu_0)\cap V_{j}'\neq\emptyset$ for all $j\in J_1$;
- ${\bigcup_{j\in J_1} {\overline{V'_{j}}}}\cap{\bigcup_{j\in J_1}{\overline{W_{j}}}}=\emptyset$;
- $\Vert\nu_1-\nu_{0}\Vert<\eta^*/l$;
- whenever $\sigma'$ is a probability measure such that $(\sigma', (V'_j)_{j\in J_1})$ is an admissible pair, it follows that $\Phi_k(\widehat{\sigma}'-\widehat{\nu}_1)<\eta^*$ for all $k\leq N_{0}$.
Each $j\in J_1$ has a unique “predecessor" $i\in J_0$. In what follows, we denote this predecessor by $j^-$.
Since $V'_j\cap\supp(\nu_0)\neq\emptyset$ and since $(\nu_0, (U_i)_{i\in J_0}, (V_i)_{i\in J_0})$ is a convenient triple, one can pick, for each $j\in J_1$, a point $\lambda_j\in V'_j\cap\supp(\nu_0)$ and a point $x_j\in U'_{j^-}$ such that $(x_j,\lambda_j)\in \mathbf Z$. The points $x_j$ are pairwise distinct because $x_j\neq 0$ and $T(x_j)=\lambda_j x_j$.
Using again the fact that $(\nu_0, (U_i)_{i\in J_0}, (V_i)_{i\in J_0})$ is a convenient triple, one can find closed sets $F_{i,0}\subset X$ with $F_{i,0}\subset U_i\setminus\{ 0\}$ such that $$\supp (\nu_0)\subset\bigcup_{i\in I}\{ \lambda\in V_i;\; \exists x\in F_{i,0}\, :\, (x,\lambda)\in \mathbf Z\}\, .$$
Since $\mathbf Z$ is closed in $(X\setminus\{ 0\})\times \TT$, the set $$F_{j}=\left\{ x\in F_{j^-, 0};\,\exists \lambda\in \supp(\nu_1)\cap {\overline{W_j}}\, :\, (x,\lambda)\in \mathbf Z\right\}$$ is closed in $X$, for each $j\in J_1$. Moreover, the sets $F_{j}$ are pairwise disjoint and we have $\bigcup_{j\in J_1}F_{j}\cap\{ x_j;\; j\in J_1\}=\emptyset$, because $\bigcup_{j\in J_1}{\overline{V_j'}}\cap\bigcup_{j\in J_1} {\overline{W_j}}=\emptyset$. It follows that one can find open sets $U_j\subset X$, $j\in J_1$ with pairwise disjoint closures and open sets $U_{j}'\subset X$ with pairwise disjoint closures, such that
- $\diam(U_j')<\frac{\eta^*}{l^{1/2}\vert J_1\vert^2}$;
- ${\overline{U_{j}}}\subset U_{j^-}$ and ${\overline{U'_j}}\subset U_{j^-}$;
- $x_j\in U'_j$ and $F_{j}\subset U_{j}$;
- $\bigcup_{j\in J_1} {\overline{U_{j}}}\cap\bigcup_{j\in J_1}{\overline{U'_{j}}}=\emptyset$.
We note that $(\nu_1, (U_{j})_{j\in J_1}, (W_j)_{j\in J_1})$ is a convenient triple, by the very definition of the closed sets $F_{j}$.
Now, we use the assumption on $\mathbf Z$: since $\mathbf Z\cap (U_j'\times V'_j)\neq\emptyset$, the set $$A_j=\{ \lambda\in V'_j;\;\exists x\in U_j'\,:\, (x,\lambda)\in \mathbf Z\}$$ is not $\mathbf S$-small, for any $j\in J_1$. Moreover, $A_j$ is an [analytic set]{} because $Z$ is a Borel subset of $X\times\TT$; in particular, $A_j$ is universally measurable (see [@K]), and hence it contains a *compact* set which is not $\mathbf S$-small. Since the family of $\mathbf S$-continuous measures is hereditary with respect to absolute continuity, it follows that one can find, for each $j\in J_1$, an $\mathbf S$-continuous probability measure $\widetilde\sigma_j$ such that $$\supp(\widetilde\sigma_j)\subset \{ \lambda\in V'_j;\;\exists x\in U_j'\,:\, (x,\lambda)\in \mathbf Z\}\, .$$ Moreover, since $U'_j\setminus\{ 0\}$ is an $F_\sigma$ set in $X$, we may in fact assume that there is a closed set $F_j'\subset X$ with $F_j'\subset U_j'\setminus\{ 0\}$ such that $$\supp(\widetilde\sigma_j)\subset \{ \lambda\in V'_j;\;\exists x\in F_j'\,:\, (x,\lambda)\in \mathbf Z\}\, .$$
If we put $$\sigma'_1:=\frac{1}{\vert J_1\vert}\,\sum_{j\in J_1} \widetilde\sigma_j\, ,$$ it follows that $(\sigma'_1, (U_j')_{j\in J_1}, (V_j')_{j\in J_1})$ is a convenient triple. In particular, the pair $(\sigma'_1, (V_j)_{j\in J_1})$ is admissible, so that $\Phi_k(\widehat\sigma'_1-\widehat{\nu}_1)<\eta^*$ for all $k\leq N_0$.
Let us summarize what we have done up to now. Starting with the convenient triple $(\sigma, (U_i)_{i\in I}, (V_i)_{i\in I})=(\nu_0, (U_j)_{j\in J_0}, (W_j)_{j\in J_0})$ and a positive integer $N_0$ such that $\Phi_k(\widehat\nu_0)<\eta^*$ for all $k>N_0$, we have found two convenient triples $(\nu_1, (U_{j})_{j\in J_1}, (W_j)_{j\in J_1})$ and $(\sigma_1', (U'_j)_{j\in J_1}, (V'_j)_{j\in J_1})$ both finer than $(\sigma, (U_i), (V_i))=(\nu_0, (U_j)_{j\in J_0}, (W_j)_{j\in J_0})$, such that
- $\bigcup_j {\overline{V'_j}}\cap\bigcup_j {\overline{W_j}}=\emptyset$;
- $\Vert\nu_1-\nu_0\Vert<\eta^*/l$;
- $\Phi_k(\widehat\sigma'_1-\widehat\nu_1)<\eta^*$ for all $k\leq N_0$.
Now, since $\nu_1$ and $\sigma'_1$ are $\mathbf S$-continuous, we can choose $N_1>N_0$ such that $\Phi_k(\widehat{\nu}_1 )<\eta^*$ and $\Phi_k(\widehat{\sigma}'_1)<\eta^*$ for all $k>N_1$, and we repeat the whole procedure with $(\nu_1, (U_{j})_{j\in J_1}, (W_j)_{j\in J_1})$ in place of $(\nu_0, (U_j)_{j\in J_0}, (W_j)_{j\in J_0})$. This produces two new normalized convenient triples $(\nu_2, (U_{j})_{j\in J_2}, (W_j)_{j\in J_2})$ and $(\sigma_2', (U'_j)_{j\in J_2}, (V'_j)_{j\in J_2})$. Then we start again with $\nu_2$ and a positive integer $N_2>N_1$ witnessing that $\nu_2$ and $\sigma'_2$ are $\mathbf S$-continuous, and so on.
After $l$ steps, we will have constructed positive integers $N_0<N_1<\dots <N_l$ and, for each $s\in\{1,\dots ,l\}$, two normalized convenient triples $(\nu_2, (U_{j})_{j\in J_s}, (W_j)_{j\in J_s})$ and $(\sigma_s', (U'_j)_{j\in J_s}, (V'_j)_{j\in J_s})$, such that the following properties hold:
- both triples $(\sigma'_s, (U'_j)_{j\in J_s}, (V_j')_{j\in J_s})$ and $(\nu_s, (U_j)_{j\in J_s}, (W_j)_{i\in J_{s}})$ are finer than $(\nu_{s-1}, (U_j)_{j\in J_{s-1}}, (W_j)_{i\in J_{s-1}})$;
- $\diam (U'_j)<\frac{\eta^*}{l^{1/2}\vert J_s\vert^2}$ and $\diam(V_j')<\eta^*$;
- ${\bigcup_{j\in J_s} {\overline{V'_{j}}}}\cap{\bigcup_{j\in J_s}{\overline{W_{j}}}}=\emptyset$;
- $\Vert\nu_l-\nu_{l-1}\Vert<\eta^*/l$;
- $\Phi_k(\widehat\sigma'_s-\widehat\nu_{l})<\eta^*$ for all $k\leq N_{s-1}$;
- $\Phi_k(\widehat\nu_s)<\eta^*$ and $\Phi_k(\widehat\sigma'_s)<\eta^*$ for all $k>N_s$.
Then (a) and (b) hold, and we have to check (c). That is, we have to show that the measure $$\sigma'=\frac{1}{l}\sum_{s=1}^l \sigma'_s\, ,$$ satisfies $\Phi_k( \widehat{\sigma}'-\widehat\sigma)\leq\eta$ for all $k\geq 0$ if $\eta^*$ is small enough (depending on $\eta$ only) and $l>1/\eta^*$.
First, we note that $\Vert{\nu_s}-\sigma\Vert=\Vert\nu_s-\nu_0\Vert<s\eta^*/l\leq \eta^*$ for every $s\in\{ 1,\dots ,l\}$. Since $\Phi_k(\widehat{\nu}_s-\widehat\sigma)\leq \Vert{\nu_s}-\sigma\Vert$, it follows that $$\Phi_k(\widehat{\sigma}'-\widehat\sigma)\leq \eta^*+\frac{1}{l}\,\sum_{s=1}^l\Phi_k(\widehat{\sigma}'_s-\widehat{\nu}_s)$$ for all $k\geq 0$.
If $k\leq N_0$ then $\Phi_k(\widehat\sigma'_s-\widehat\nu_s)<{\varepsilon}$ for every $s\in\{ 1,\dots, l\}$, and hence $$\Phi_k(\widehat\sigma'-\widehat\sigma)\leq 2\eta^*\, .$$
If $k>N_0$, let us denote by $s(k)$ the largest $s\in\{ 0,\dots ,l\}$ such that $k>N_s$. Then $\Phi_k(\widehat\sigma'_s-\widehat\nu_s)<\eta^*$ if $s>s(k)+1$ because $k\leq N_{s(k)+1}\leq N_{s-1}$; and $\Phi_k(\widehat\sigma'_s-\widehat\nu_s)\leq\Phi_k(\widehat\sigma'_s) +\Phi_k(\widehat\nu_s)<2\eta^*$ if $s\leq s(k)$, because $k >N_s$. So we get $$\begin{aligned}
\Phi_k(\widehat\sigma'_s-\widehat\sigma)&\leq&\eta^*+ \frac{1}{l} \left(\sum_{s\leq s(k)} 2\eta^*+\Vert \widehat\sigma'_{s(k)+1}-\widehat\nu_{s(k)+1}\Vert_\infty+\sum_{s>s(k)+1}\eta^*\right)\\
&\leq&4\eta^*+\frac{2}{l}\, \cdot\end{aligned}$$
Taking $\eta^*=\eta/6$, this concludes the proof of Fact 2.
We can now start the actual proof of Lemma \[superkey\].
Recall that $\mathfrak Q$ is the set of all finite sequences of integers $\bar q=(q_1,\dots ,q_l)$ with $q_s\geq 2$ for all $s$. We denote by $\mathfrak Q^{<\NN}$ the set of all finite sequences $\mathbf s=(\bar q_1,\dots ,\bar q_n)$ with $\bar q_j\in\mathfrak Q$, plus the empty sequence $\emptyset$. We denote by $\vert \mathbf s\vert$ the length of a sequence $\mathbf s\in \mathfrak Q^{<\omega}$ (the empty sequence $\emptyset$ has length $0$), and by $\prec$ the natural extension ordering on $\mathfrak Q^{<\omega}$.
If $\mathbf s=(\bar q_1,\dots ,\bar q_n)\in\mathfrak Q^{<\NN}$ we put $$\Omega (\mathbf s):=\prod_{j=1}^n \Omega( \bar q_j)\, ,$$ and we denote by $m_{\mathbf s}$ the product measure $\otimes_{j=1}^n m_{\bar q_j}$. We also put $\Omega(\emptyset)=\{\emptyset\}$, and $m_\emptyset =\delta_{\emptyset}$.
Let us fix $(x_0,\lambda_0)\in \mathbf Z$ and ${\varepsilon}>0$. Put $V_\emptyset:=\TT$, and choose an open set $U_\emptyset\subset X\setminus\{ 0\}$ such that $\diam(U_\emptyset)<{\varepsilon}$ and $(x_0,\lambda_0)\in U_\emptyset\times V_\emptyset$. Let also pick an open set $U'_\emptyset\subset X$ such that $x_\emptyset\in U'_\emptyset$ and ${\overline{U'_\emptyset}}\subset U_\emptyset$. By assumption, one can find an $\mathbf S$-continuous probability measure $\sigma_0$ such that $\supp(\sigma_0)\subset\{ \lambda\in V_\emptyset;\;\exists x\in U'_\emptyset\, :\, (x,\lambda)\in Z\}\, .$ Then $(\sigma_\emptyset, U_\emptyset, V_\emptyset)$ is a convenient triple.
We construct by induction a sequence $(\mathbf s_n)_{n\geq 0}\subset \mathfrak Q^{<\omega}$, a sequence of probability $\mathbf S$-continuous measures $(\sigma_n)_{n\geq 0}$ and, for each $n\geq 0$, two sequences of open sets $(U_\xi)_{\xi\in\Omega (\mathbf s_n)}\subset X$ and $(V_\xi)_{\xi\in\Omega(\mathbf s_n)}\subset\TT$. If $n\geq 1$, we write $\mathbf s_n=(\bar q_1,\dots ,\bar q_n)$ and $\bar q_j=(q_{j,1},\dots ,q_{j, l_j}\}$. If $\xi=(\xi_1,\dots ,\xi_{n-1})\in\Omega(\mathbf s_{n-1})$ and $\tau\in\Omega (\bar q_n)$, we denote by $\xi\tau$ the sequence $(\xi_1,\dots ,\xi_{n-1},\tau)\in\Omega (\mathbf s_n)$ (if $n=1$ then $\xi\tau=\emptyset\tau=\tau$). Finally, we put ${\varepsilon}_0=1$ and $${\varepsilon}_n:=\frac{1}{w(\bar q_1)\cdots w(\bar q_n)}$$ if $n\geq 1$. The following requirements have to be fulfilled for all $n\geq 0$.
(i) $\vert \mathbf s_n\vert=n$, and $\mathbf s_{n-1}\prec \mathbf s_n$ if $n\geq 1$.
(ii) If $n\geq 1$ and $\xi\in\Omega(\mathbf s_{n-1})$, then
- ${\overline{U_{\xi\tau}}}\subset U_\xi$ and ${\overline{V_{\xi\tau}}}\subset V_\xi$ for every $\tau\in\Omega (\bar q_n)$;
- ${\overline{U_{\xi\tau}}}\cap {\overline{U_{\xi\tau'}}}=\emptyset$ and ${\overline{V_{\xi\tau}}}\cap{\overline{V_{\xi\tau}}}=\emptyset$ if $\tau\neq \tau'$;
- $\diam(U_{\xi\tau})\leq\frac 12\diam (U_\xi)$ and $\diam(V_{\xi\tau})\leq\frac 12\diam (V_\xi)$.
(iii) If $n\geq 1$ and $\xi\in\Omega(\mathbf s_{n-1})$, then $$\diam (U_{\xi\tau})<\frac{{\varepsilon}_{n-1}}{l_n^{1/2} \, q_{n,s}^2}$$ for all $s\in\{ 1,\dots ,l_{n}\}$ and every $\tau\in\Omega(q_{n,s})$.
(iv) One can find closed sets $F_\xi\subset X$, $\xi\in\Omega (\mathbf s_n)$ with $F_\xi\subset U_\xi\setminus\{ 0\}$ and such that $\supp(\sigma_n)\subset \bigcup_{\xi\in\Omega (\mathbf s_n)} \{\lambda\in V_\xi;\;\exists x\in F_\xi\, :\, (x,\lambda)\in \mathbf Z\}$. In particular: $$\supp(\sigma_n)\subset \bigcup_{\xi\in\Omega (\mathbf s_n)} \{\lambda\in V_\xi;\;\exists x\in U_\xi\, :\, (x,\lambda)\in \mathbf Z\}\, .$$
(v) If $i\leq n$ and $\xi\in\Omega (\mathbf s_i)$, then $\sigma_n(V_\xi)=m_{\mathbf s_i}(\{ \xi\}).$
(vi) If $n\geq 1$ then $\sup_{k\geq 0}\,\Phi_k(\widehat\sigma_{n}-\widehat\sigma_{n-1})<2^{-n}$.
Since $(\sigma_0, U_\emptyset, V_\emptyset)$ is a convenient triple, condition (iv) is satisfied for $n=0$; and the other conditions are (trivially) satisfied as well.
Applying Fact 2 with the convenient triple $(\sigma_0, U_\emptyset, V_\emptyset)$ and a small enough $\eta>0$, we get a positive integer $l$ and a finite sequence of pairwise disjoint convenient triples $(\sigma'_1, (U_j')_{j\in J_1}, (V_j')_{j\in J_1}),
\dots,$ $(\sigma'_l, (U_j')_{j\in J_l}, (V_j')_{j\in J_l})$ finer than $(\sigma_0, U_\emptyset, V_\emptyset)$ such that
- $\diam(U_j')<\frac 12\diam(U_\emptyset)$ and $\diam(V'_j)<\frac 12\diam(V_\emptyset)$ for all $s\in\{ 1,\dots ,l\}$ and every $j\in J_s$;
- $\diam(U_j')<\frac{{\varepsilon}_0}{l^{1/2}\vert J_s\vert^2}$ for all $s$ and every $j\in J_s$;
- $\sup_{k\geq 0}\,\Phi_k(\widehat{\sigma}'-\widehat\sigma_0)\leq 2^{-1}$, where $\sigma'=\frac{1}{l}\sum_{s=1}^l \sigma'_s\, . $
If we put $q_{1,s}:=\vert J_s\vert$, $s\in\{ 1,\dots ,l\}$ and $\bar q_1=(q_{1,1},\dots ,q_{1,l})$, we may enumerate in the obvious way the open sets $U'_j$, $V'_j$ as $U_{\xi}$, $V_\xi$, $\xi\in\Omega (\bar q_1)$. Then (i),$\,\dots$,(vi) are clearly satisfied for $n=1$ with $\sigma_1=\sigma'$.
The general inductive step is very much the same. Assume that everything has been constructed up to some stage $n\geq 1$. Then, for every $\tilde\xi\in\Omega (\mathbf s_{n-1})$, the triple $\mathcal T_{\tilde\xi}=((\sigma_n)_{\vert V_{\tilde\xi}}, (U_{\tilde\xi\tau})_{\tau\in\Omega (\bar q_n)}, (V_{\tilde\xi\tau})_{\tau\in\Omega (\bar q_n)})$ is a (non-normalized) convenient triple. Given $\eta>0$, it is not hard to see (by examining the proof) that we can apply Fact 2 simultaneously to all triples $\mathcal T_{\tilde\xi}$, $\tilde\xi\in\Omega (\mathbf s_{n-1})$, with the same $l=l(\eta)$ and the same index set $J$. As above, we may take $J=\Omega(\bar q_n)\times \Omega (\bar q)$, for some $\bar q=(q_1,\dots ,q_l)$. This gives positive $\mathbf S$-continuous measures $\sigma_{\tilde\xi}$ and open sets $U_{\tilde\xi\tau'}$, $V_{\tilde\xi\tau'}$, $\tau'\in\Omega (\bar q_n)\times\Omega(\bar q)$. Then, if $\eta$ is small enough, conditions (i),$\,\dots$,(vi) will be met at stage $n+1$ with $\mathbf s_{n-1}=\mathbf s_n\bar q$, $\sigma_{n+1}=\sum_{\tilde\xi}\sigma_{\tilde\xi}$ and the open sets $U_\xi,V_\xi$ for $\xi\in\Omega(\mathbf s_{n+1})=\Omega(\mathbf s_{n-1})\times\Omega(\bar q_n)\times \Omega (\bar q)$.
Let us denote by $\mathbf q$ the “limit" of the increasing sequence $(\mathbf s_n)$, the infinite sequence $(\bar q_n)_{n\geq 1}\in\mathfrak Q^\NN$.
It follows from (ii) that for any $\omega=(\omega_n)_{n\geq 1}\in\Omega (\mathbf q)$, the intersection $\bigcap_{n\geq 1} U_{\omega_{\vert n}}$ is a single point $\{ E(\omega )\}$, the intersection $\bigcap_{n\geq 1} V_{\omega_{\vert n}}$ is a single point $\{ \phi(\omega)\} $, and the maps $\phi :\Omega (\mathbf q)\to\TT$ and $E:\Omega (\mathbf q)\to X$ are homeomorphic embeddings. Moreover, condition (iii) says exactly that $E$ is super-Lipschitz. And since $x_0\in U_\emptyset$ and $\diam (U_\emptyset)<{\varepsilon}$, we have $\Vert E(\omega)-x_0\Vert<{\varepsilon}$ for every $\omega\in \Omega(\mathbf q)$.
For each $\xi\in\bigcup_{n\geq 0}\Omega (\mathbf s_n)$, let us pick a point $\lambda_\xi\in V_\xi\cap\supp(\sigma_n)$, where $n$ is the length of $\xi$, $\xi\in\Omega (\mathbf s_n)$. By (iv), one can find $x_\xi\in U_\xi$ such that $(x_\xi,\lambda_\xi)\in \mathbf Z$. In particular, we have $T(x_\xi)=\lambda_\xi x_\xi$ for every $\xi\in\bigcup_{n\geq 0}\Omega (\mathbf s_n)$. Since $x_{\omega_{\vert n}}\to E(\omega)$ and $\lambda_{\omega_{\vert n}}\to\phi(\omega)$ as $n\to\infty$, it follows that $TE(\omega )=\phi(\omega)\, E(\omega)$ for every $\omega\in\Omega (\mathbf q)$. In other words, $(E,\phi)$ is a $\TT$-eigenfield for $T$.
By (vi), the sequence $(\sigma_n)$ converges $w^*$ to a probability $\mathbf S$-continuous measure $\sigma$. To conclude the proof, it remains to show that $\sigma$ is the image measure of the measure $m_{\mathbf q}$ on $\Omega(\mathbf q)$ under the embedding $\phi :\Omega (\mathbf q)\to\TT$. That is, we want to check that
$$\int_{\Omega (\mathbf q)} f\circ \phi (\omega)\, dm_{\mathbf q}(\omega)=\int_\TT f\, d\sigma$$ for any $f\in\mathcal C(\TT)$. Let us fix $f$.
For each $\xi\in\bigcup_{n\geq 0}\Omega (\mathbf s_n)$, let us (again) pick a point $\lambda_\xi\in V_\xi$. Then $\lambda_{\omega_{\vert n}}\to \phi(\omega)$ as $n\to\infty$, for every $\omega\in\Omega (\mathbf q)$. Setting $\Omega_\xi:=\{ \omega\in\Omega (\mathbf q);\; \xi\subset \omega\}$ and using Lebesgue’s theorem, it follows that
$$\begin{aligned}
\int_{\Omega(\mathbf q)} f\circ\phi (\omega)\, dm_{\mathbf q}(\omega)&=&\lim_{n\to\infty} \sum_{\xi\in\Omega (\mathbf s_n)} f(\lambda_\xi)\times m_{\mathbf q} (\Omega_\xi)\\
&=&\lim_{n\to\infty}\,\sum_{\xi\in\Omega (\mathbf s_n)} f(\lambda_\xi)\times m_{\mathbf s_n} (\{\xi\})\, .\end{aligned}$$
On the other hand, by the definition of $\sigma$ and since $\diam (V_\xi)\to 0$ as $\vert \xi\vert\to\infty$ we also have $$\begin{aligned}
\int_\TT f\, d\sigma&=&\lim_{n\to\infty} \int_\TT f\, d\sigma_n\\
&=&\lim_{n\to\infty}\sum_{\xi\in\Omega (\mathbf s_n)} f(\lambda_\xi)\times \sigma_n(V_\xi)\, .\end{aligned}$$
By condition (v), this concludes the proof.
The proof {#realproof}
---------
Assume that the $\TT$-eigenvectors are $\mathbf S$-perfectly spanning, and let us show that $T$ is $\mathbf S$-mixing in the Gaussian sense. We recall that, since $\mathbf S$ is $c_0$-like, the $\mathbb T$-eigenvectors are in fact $\mathbf S$-perfectly spanning for analytic sets (see Remark 1 just after Corollary \[Smix=span\]).
Using Lemma \[halfkey\], Lemma \[superkey\] and proceeding exactly as in sub-section \[proofWM\], we find a sequence of $\TT$-eigenfields $(E_i,\phi_i)$ defined on some $\Omega (\mathbf{q}_i)$, such that
- each operator $K_{E_i} :L^2(\Omega ({\mathbf q_i}), m_{{\mathbf q_i}})\to X$ is gamma-radonifying;
- each $E_i$ is continuous and $\overline{\rm span}\left(\bigcup_{i\in\NN} {\rm ran}\, (E_i)\right)=X$;
- each $\phi_i$ is a homeomorphic embedding, and $\sigma_i=m_{{\mathbf q_i}}\circ\phi_i^{-1}$ is $\mathbf S$-continuous.
Put $(\Omega_i, m_i):=(\Omega(\mathbf q_i), m_{\mathbf q_i})$ and let $(\Omega, m)$ be the disjoint union of the measure spaces $(\Omega_i,m_i)$. Choose a sequence of small positive numbers $(\alpha_i)_{i\in\NN}$, and define a $\TT$-eigenfield $(E,\phi)$ on $(\Omega, m)$ as expected: $E(\omega_i)=\alpha_i E_i(\omega_i)$ and $\phi(\omega_i)=\phi_i(\omega_i)$ for each $i$ and every $\omega_i\in\Omega_i$. By (1), the operator $K_E:L^2(\Omega, m)\to X$ is gamma-radonifying if the $\alpha_i$ are small enough. By (2) and since $m$ has full support, the vector field $E$ is $m$-spanning and hence the operator $K_E$ has dense range. The intertwining equation $TK_E=K_EM_\phi$ holds by the definition of $E$. Finally, it follows from (3) that the measure $(f_im_{i})\circ\phi_i^{-1}$ is $\mathbf S$-continuous for each $i$ and every $f_i\in L^1(\Omega_i, m_i)$. Hence, the measure $\sigma_f=(fm)\circ\phi^{-1}$ is $\mathbf S$-continuous for every $f\in L^1(\Omega, m)$. By Proposition \[background\], this shows that $T$ is $\mathbf S$-mixing in the Gaussian sense.
Fréchet spaces
--------------
The above proof can be reproduced almost word for word in the Fréchet space setting. More precisely, the following changes should be made.
- Modify the definition of “super-Lipschitz": a map $E:\Omega(\mathbf q)\to X$ is super-Lipschitz if $E:\Omega(\mathbf q)\to (X,\Vert\hskip 0.6mm\cdot\hskip 0.6mm\Vert_{})$ is super-Lipschitz for any continuous semi-norm $\Vert\hskip 0.6mm\cdot\hskip 0.6mm\Vert$ on $X$.
- In Lemma \[halfkey\], add “for any continuous semi-norm $\Vert\hskip 0.6mm\cdot\hskip 0.6mm\Vert$ on $X$".
- In Lemma \[superkey\], fix a nondecreasing sequence of semi-norms $(\Vert\hskip 0.6mm\cdot\hskip 0.6mm\Vert_i)_{i\in\NN}$ generating the topology of $X$, and put $$\Vert x\Vert=\sum_{i\in\NN}2^{-i} \min(\Vert x\Vert_i, 1)\, .$$ (Of course this is not even a semi-norm, but the notation is convenient anyway). Then perform exactly the same construction.
- Do the same when starting the proof of Theorem \[abstract\].
proof of the abstract results (2) {#proofabstract2}
=================================
Proof of theorem \[abstracteasy\]
---------------------------------
Let the Banach space $X$ have type 2. Let $T\in\mathfrak L(X)$ and assume that the $\TT$-eigenvectors of $T$ are $\mathbf S$-perfectly spanning for analytic sets. By Lemma \[hereditary\] and Proposition \[perfect\], one can find a countable family of [continuous]{} $\TT$-eigenvector fields $(E_i)_{i\in I}$ for $T$, where $E_i:\Lambda_i\to X$ is defined on some $\mathbf S$-perfect set $\Lambda_i\subset\TT$, such that ${\rm span}\left(\bigcup_{i\in I} E_i(\Lambda_i)\right)$ is dense in $X$. By Lemma \[Bperfect\], each $\Lambda_i$ is the support of some $\mathbf S$-continuous probability measure $\sigma_i$. If we put $\phi_i(\lambda)=\lambda$, this gives a continuous $\TT$-eigenfield $(E_i,\phi_i)$ for $T$ on the measure space $(\Lambda_i,\sigma_i)$ such that the image measure $\sigma_i\circ \phi_i^{-1}=\sigma_i$ is $\mathbf S$-continuous; and by Proposition \[typecotype\], the operator $K_{E_i}$ is gamma-radonifying because $X$ has type 2. So the proof can be completed exactly as in sub-section \[realproof\] above.
Proof of Proposition \[converse\]
---------------------------------
Let the Banach space $X$ have cotype 2, assume that $T\in\mathfrak L(X)$ is $\mathbf S$-mixing with respect to some Gaussian measure $\mu$ with full support, and let us show that the $\TT$-eigenvectors of $T$ are perfectly spanning for analytic sets.
We start with the following fact, which follows rather easily from Lemma \[back1\] (1) (see the beginning of the proof of Theorem 5.46 in [@BM]).
One can find a gamma-radonifying operator $K:\mathcal H\to X$ such that $\mu=\mu_K$ and a *unitary* operator $M=\mathcal H\to\mathcal H$ such that $TK=KM$.
By the spectral theorem, we may assume that $\mathcal H=L^2(\Omega ,m)$ for some measure space $(\Omega, m)$ and that $M$ is a multiplication operator, $M=M_\phi$ for some measurable map $\phi :\Omega\to\TT$. Then we use the cotype 2 assumption on $X$: by Proposition \[typecotype\], the gamma-radonifying operator $K:L^2(\Omega, m)\to X$ is in fact given by a vector field; that is, $K=K_E$ for some vector field $E:(\Omega,m)\to X$.
Now, let $D\subset \TT$ be an analytic $\mathbf S$-small set. By the remark just after Lemma \[back1\] and by Lemma \[back2\] (with $\mathcal H_1=\mathcal H\ominus\ker(K_E)$), we know that $\mathbf 1_{\{ \phi\in D\}}h\in\ker (K_E)$ for any $h\in L^2(\Omega, m)$. In other words, we have $$\int_{\{\phi\in D\}} h(\omega) E(\omega)\, dm(\omega)=0$$ for every $h\in L^2(\Omega, m)$. It follows at once that $E$ is almost everywhere $0$ on the set $\{ \phi\in D\}$; and since $E$ is $m$-spanning (because $\mu=\mu_{K_E}$ has full support), this implies that the linear span of $\{ E(\omega);\; \omega\in \Omega\setminus\phi^{-1}(D)\}$ is dense in $X$. By the very definition of a $\TT$-eigenfield, it follows that the linear span of $\bigcup_{\lambda\in\TT\setminus D} \ker(T-\lambda)$ is dense in $X$, and we conclude that the $\TT$-eigenvectors of $T$ are $\mathbf S$-perfectly spanning.
Miscellaneous remarks {#final}
=====================
Examples
--------
Theorem \[WS\] is a theoretical statement. However, it is extremly useful to give concrete examples of mixing operators, including new ones ( the backward shift operators on $c_0$, or the translation semigroups in sub-section \[Semigroups\] below). In this sub-section, we review several examples that were already known to be mixing in the Gaussian sense, either by an application of the results of [@BG3], [@BG2] or [@BM], or by using an ad-hoc argument. What we want to point out is that Theorem \[WS\] now makes the proofs completely straightforward. In all cases, Theorem \[WS\] is used through the following immediate consequence.
\[mainbis\] Let $T$ be an operator acting on a complex separable Fréchet space $X$. Assume that one can find a $\TT$-eigenfield $(E,\phi)$ for $T$ on some topological measure space $(\Omega, m)$ such that the measure $m$ has full support, $E$ is continuous with $\overline{\rm span}\, E(\Omega)=X$, and $(fm)\circ\phi^{-1}$ is a Rajchman measure for every $f\in L^1(m)$. Then $T$ is strongly mixing in the Gaussian sense.
If $D\subset\TT$ is a Borel set of extended uniqueness, then $m(\phi^{-1}(D))=0$ because $(fm)\circ\phi^{-1}(D)=0$ for every $f\in L^1(m)$. Since $m$ has full support, it follows that $\Omega\setminus\phi^{-1}(D)$ is dense in $\Omega$ and hence that the linear span of $E(\Omega\setminus\phi^{-1}(D))$ is dense in $X$ because $E$ is assumed to be continuous. Since $E(\Omega\setminus \phi^{-1}(D))\subset \bigcup_{\lambda\in\TT\setminus D}\ker(T-\lambda)$, this shows that the $\TT$-eigenvectors of $T$ are $\mathcal U_0$-perfectly spanning.
When the measure $m$ is finite, the third condition in Proposition \[mainbis\] just means that $m\circ\phi^{-1}$ is a Rajchman measure.
### Weighted shifts again
Weighted backward shifts on $X_p=\ell^p(\NN)$, $1\leq p<\infty$ or $X_\infty=c_0(\NN)$ have already been considered in the introduction (Example 1): a weighted shift $B_{\mathbf w}$ is strongly mixing in the Gaussian sense as soon as the sequence $\left(\frac{1}{w_0\cdots w_n}\right)_{n\geq 0}$ is in $X_p$, due to the existence of a a continuous $\TT$-eigenvector field $E:\TT\to X_p$ such that $\overline{\rm span}\, E(\TT)=X_p$, namely $$E(\lambda):=\sum_{n=0}^\infty \frac{\lambda^n}{w_0\cdots w_n}\, e_n\, .$$
Here, we just would like to point out one curious fact. If $p<\infty$, the operator $K_E:L^2(\TT)\to X_p$ turns out to be gamma-radonifying (see [@BM]) and hence there is a very natural explicit ergodic Gaussian measure for $B_{\mathbf w}$, namely the distribution of the random variable $\xi=\sum_0^\infty \frac{g_n}{w_0\cdots w_n}\, e_n$. As shown in [@BG2 Example 3.13], this need not be true when $p=\infty$, $X=c_0$; that is, the weight sequence can be chosen in such a way that $K_E$ is not gamma-radonifying. So there is no “obvious" Gaussian measure in this case.
### Composition operators
Let $\alpha:\DD\to\DD$ be an automorphism of the unit disk $\DD$ without fixed points in $\DD$, and let $C_\alpha$ be the associated composition operator acting on $X=H^p(\DD)$, $1\leq p<\infty$: $$C_\alpha (f)=f\circ\alpha\, .$$ As shown in [@BG3], there is a natural continuous $\TT$-eigenfield $(E,\phi)$ for $C_\alpha$, defined on $\Omega=\RR_+$ in the parabolic case and $\Omega=[0,2\pi)$ in the hyperbolic case (endowed with Lebesgue measure $m$) such that ${\overline{\rm span}}\, E(\Omega)= X$ and the image measure $m\circ\phi^{-1}$ is absolutely continuous with respect to Lebesgue measure on $\TT$. By Proposition \[mainbis\], it follows that $C_\alpha$ is strongly mixing in the Gaussian sense. This was shown in [@BG2], with a longer proof because the authors had to play with the regularity of the vector field $E$ and the geometry of the space $X$ to show that $K_E$ is gamma-radonifying; but the proof of [@BG2] is also more informative since it provides an explicit Gaussian measure for $C_\alpha$.
### Operators with analytic $\TT$-eigenvector fields
The following consequence of Proposition \[mainbis\] is worth stating explicitely.
Let $X$ be a complex separable Fréchet space, and let $T\in\mathfrak L(X)$. Assume that $T$ is not a scalar multiple of the identity and that one can find a map $E:U\to X$ defined on some connected open set $U\subset \CC$ such that $E$ is holomorphic or “anti-holomorphic" ( $E(\bar s)$ is holomorphic), each $E(s)$ is an eigenvector for $T$, the associated eigenvalue has modulus 1 for at least one $s\in U$ and $\overline{\rm span}\, E(U)=X$. Then $T$ is strongly mixing in the Gaussian sense.
Let us denote by $\phi(s)$ the eigenvalue associated with $E(s)$. Using the Hahn-Banach theorem, it is easily seen that $\phi :U\to\CC$ is holomorphic or anti-holomorphic. Moreover, $\phi$ is non-constant since $T$ is not scalar. By the inverse function theorem, it follows that $\phi (U)\cap \TT$ contains a non trivial arc $\Lambda$ such that the restriction of $\phi$ to $\Omega:=\phi^{-1}(\Lambda)$ is a homeomorphism from $\Omega$ onto $\Lambda$. By the Hahn-Banach theorem and the identity principle for holomorphic (or anti-holomorphic) functions, the linear span of $E(\Omega)$ is dense in $X$. Hence, if we denote by $m$ the image of the Lebesgue measure on $\Lambda$ by $\phi^{-1}$, the $\TT$-eigenfield $(E,\phi)$ restricted to $\Omega$ satisfies the assumptions of Proposition \[mainbis\].
One could also have used Theorem \[WS\] directly. Indeed, if $D\subset\TT$ is a Borel $\mathcal U_0$-set, then $\phi^{-1}(\TT\setminus D)$ is an uncountable Borel set (because $\phi(U)$ contains a nontrivial arc by the open mapping theorem), and as such it contains an uncountable compact set $K$. Then $\overline{\rm{span}}\, E(K)=X$ by the identity principle, and since $E(K)\subset\bigcup_{\lambda\in\TT\setminus D}\ker(T-\lambda)$, it follows that the $\TT$-eigenvectors of $T$ are $\mathcal U_0$-perfectly spanning.
This lemma may be applied, for example, in the following two cases.
1. $T=M_\phi^*$, where $M_\phi$ is a (non-scalar) multiplication operator on some reproducing kernel Hilbert space of analytic functions on a connected open set $U\subset\CC$, and $\phi(U)\cap\TT\neq\emptyset$.
2. $T$ is a (non-scalar) operator on the space of entire functions $H(\CC)$ commuting with all translation operators.
In the first case one may take $E(s):=k_s$, the reproducing kernel at $s\in U$ (which depends anti-holomorphically on $s$ and satisfies $M_\phi^*(k_s)=\overline{\phi (s)}\, k_s$). In the second case the assumptions of the lemma are satisfied with $E(s)=e_s$, where $e_s(z)=e^{sz}$ (denoting by $\tau_z$ the operator of translation by $z$, use the relation $\tau_ze_s=e_s(z)e_s$ to show that $Te_s=(Te_s(0))\times e_s$).
Non Gaussian measures
---------------------
It is natural to ask whether one gets a more general notion of mixing for an operator $T\in\mathfrak L(X)$ by requiring only that $T$ should be mixing with respect to *some* probability measure $\mu$ on $X$ with full support. The following remark (essentially contained in [@R], and also in [@BG3]) provides a partial answer.
Let $X$ be a separable Banach space, and assume that $X$ has *type $2$*. Let also $\mu$ be a centred Borel probability measure on $X$ such that $\int_X\Vert x\Vert^2d\mu (x)<\infty$. Then there is a unique Gaussian measure $\nu$ on $X$ such that $$\Vert x^*\Vert_{L^2(\nu)}=\Vert x^*\Vert_{L^2(\mu)}$$ for every $x^*\in X^*$. Moreover,
- if $\mu$ has full support then so does $\nu$;
- if $\mu$ is $T$-invariant for some $T\in\mathfrak L(X)$, then so is $\nu$;
- if $T\in\mathfrak L(X)$ is weakly mixing or strongly mixing with respect to $\mu$, then the same is true with respect to $\nu$.
Put $\mathcal H:=L^2(\mu)$. By assumption on $\mu$, there is a well defined (conjugate-linear) “inclusion" operator $J:X^*\to \mathcal H$, namely $J(x^*)=\overline{x^*}$ considered as an element of $L^2(\mu)$. Moreover, it follows from Lebesgue’s theorem and the $w^*$-$\,$metrizability of $B_{X^*}$ that $J$ is $(w^*,w^*)\,$-$\,$continuous on $B_{X^*}$. Hence, $J$ is the adjoint of a bounded operator $K:\mathcal H\to X$. By definition, this means that $K^*x^*=\overline{x^*}$ considered as an element of $L^2(\mu)$, so we have in particular $$\Vert K^*x^*\Vert_{\mathcal H}=\Vert x^*\Vert_{L^2(\mu)}\, .$$
It is fairly easy to show that the “inclusion" operator $K^*=J:X^*\to L^2(\mu)$ is *absolutely $2$-summing* (see [@AK] for the definition). Since $X$ has type 2, it follows that $K$ is gamma-radonifying (see [@CTV], or [@BM Proposition 5.19]). If we denote by $\nu=\mu_K$ the associated Gaussian measure, then $\Vert x^*\Vert_{L^2(\nu)}=\Vert x^*\Vert_{L^2(\mu)}$ for every $x^*\in X^*$ by the very definition of $\nu$. Moreover, if $\nu'$ is another Gaussian measure with the same properties, then $\nu$ and $\nu'$ have the same Fourier transform and hence $\nu=\nu'$.
To prove (1), assume that $\mu$ has full support. Then the operator $K^*=J:X^*\to \mathcal H$ is one-to-one because a continuous function on $X$ is $0$ in $L^2(\mu)$ if and only if it is identically $0$. It follows that $K$ has dense range, and hence that $\nu=\mu_K$ has full support.
To prove (2), assume that $\mu$ is $T$-invariant. Then $\Vert T^*x^*\Vert_{L^2(\mu)}=\Vert x^*\Vert_{L^2(\mu)}$ and hence $\Vert K^*(T^*x^*)\Vert_{L^2(\nu)}=\Vert K^*(x^*)\Vert_{L^2(\nu)}$ for every $x^*\in X^*$. By the proof of Lemma \[back1\], this shows that $\nu$ is $T$-invariant
Finally, (3) follows from the following observations : (i) weak mixing or strong mixing of $T$ with respect to $\mu$ is characterized by a certain behaviour (B) of the sequence $(\langle f\circ T^n,g\rangle_{L^2(\mu)})_{n\geq 0}$, for any $f,g\in L^2_0(\mu)$; (ii) when specialized to linear functionals, this gives that the sequence $(\langle T^{*n}x^*,y^*\rangle_{L^2(\mu)})$ satisfies (B) for any $x,y^*\in X^*$; (iii) by the definition of $\nu$, this means that $(\langle T^{*n}x^*,y^*\rangle_{L^2(\nu)})$ satisfies (B) for any $x^*,y^*$; (iv) since $\nu$ is a Gaussian measure, this is enough to ensure weak mixing or strong mixing of $T$ with respect to $\nu$, by Rudnicki [@R] or Bayart-Grivaux [@BG3].
When $X$ is a Hilbert space, it follows from this remark (and from Theorem \[WS\]) that an operator $T\in\mathfrak L(X)$ is strongly mixing with respect to some centred probability measure $\mu$ on $X$ with full support such that $\int_X\Vert x\Vert^2d\mu (x)<\infty$ if and only if the $T$-eigenvectors of $T$ are $\mathcal U_0$-perfectly spanning, in which case the measure $\mu$ can be assumed to be Gaussian. It would be interesting to know if this remains true without any a priori assumption on the measure $\mu$.
A characterization of $\mathcal U_0$-sets {#bizarre}
-----------------------------------------
Our results easily yield the following curious characterization of sets of extended uniqueness.
Let $D$ be an analytic subset of $\TT$. Then $D$ is a set of extended uniqueness if an only if the following holds: for every Hilbert space operator $T$ which is strongly mixing in the Gaussian sense, the linear span of $\bigcup_{\lambda\in\TT\setminus D}\ker(T-\lambda)$ is dense in the underlying Hilbert space.
The “only if" part follows immediately from Proposition \[converse\]. Conversely, assume that $D\not\in\mathcal U_0$. Then, since $D$ is universally measurable, it contains a compact set $\Lambda$ which is the support of some Rajchman probability measure, $\Lambda$ is [$\mathcal U_0$-perfect]{}. Let $T_\Lambda:H_\Lambda\to H_\Lambda$ be the Kalisch operator from Example 2 in the introduction. Then $T_\Lambda$ is strongly mixing in the Gaussian sense, but ${\rm span}\left(\bigcup_{\lambda\in\TT\setminus D}\ker(T-\lambda)\right)$ is certainly not dense in $H$ since it is $\{ 0\}$ (recall that $\sigma_p(T)=\Lambda$ and $\Lambda\subset D$).
The non-ergodicity index
------------------------
Loosely speaking, Corollary \[characexistergod\] is a kind of “perfect set theorem" for ergodicity. This can be made precise as follows. Let $T\in\mathfrak L(X)$ (where $X$ is a Banach space with cotype 2), and consider the following “derivation" on closed, $T$-invariant subspaces of $X$: for any such subspace $E$, set $$\mathcal D_T(E):=\bigcap_{D}\overline{\rm span}\, \bigcup_{\lambda\in\TT\setminus D}\ker (T_{\vert E}-\lambda)\, ,$$ where the intersection ranges over all countable sets $D\subset\TT$. By transfinite induction, one defines the iterates $\mathcal D_T^\alpha(X)$ for every ordinal $\alpha$, in the obvious way: $$\displaylines{
\mathcal D_T^{\alpha+1}(X)=\mathcal D_T[\mathcal D_T^\alpha(X)]\, ,\cr
\mathcal D_T^\lambda(X)=\displaystyle\bigcap_{\alpha <\lambda} \mathcal D_T^\alpha(X)\;\;\;\;\hbox{($\lambda$ limit).}
}$$
Since $X$ is a Polish space, the process must stabilize at some countable ordinal $\alpha(T)$ and hence $\mathcal D_T^\infty (X):=\bigcap_{\alpha} \mathcal D_T^\alpha (X)$ is well-defined and is a fixed point of $\mathcal D_T$ (in fact, the largest fixed point). Then, Corollary \[characexistergod\] says that $T$ admits a nontrivial ergodic Gaussian measure iff $\mathcal D_T^\infty(X)\neq\{ 0\}$, in which case one can find an ergodic measure with support $\mathcal D_T^\infty(X)$. The subspace $\mathcal D_T^\infty (X)$ is the “perfect kernel" associated with the derivation $\mathcal D_T$, a canonical witness of the ergodicity of $T$.
Let us say that $T$ is *totally non-ergodic* if it does not admit any nontrivial ergodic Gaussian measure. Then the ordinal $\alpha(T)$ may be called the “non-ergodicity index" of $T$. It is quite natural to wonder whether this index can be arbitrarily large: given any countable ordinal $\alpha$, is it possible to construct a totally non-ergodic operator $T$ with $\alpha (T)>\alpha$?
The scope of the abstract results
---------------------------------
Some comments are in order regarding the assumptions made on the family $\mathbf S$ in Theorems \[abstract\] and \[abstracteasy\].
The main trouble with Theorem \[abstract\] is that we have no idea of how to prove it without assuming that the family $\mathbf S$ is $c_0$-like. Perhaps unexpectedly, what makes the definition of a $c_0$-like family very restrictive is the uniform boundedness assumption of the sequence of semi-norms $(\Phi_n)$. For example, any growth condition of the form $$a_n=o({\varepsilon}_n)\, ,$$ where $\bar{\varepsilon}=({\varepsilon}_n)$ is a sequence of positive numbers tending to $0$ and satisfying ${\varepsilon}_{n\pm k}\leq C_k \, {\varepsilon}_n$, defines a translation-invariant ideal $\mathbf S_{\bar{\varepsilon}}\subset\ell^\infty(\ZZ_+)$ which has the correct form *except* for this uniform boundedness condition (just put $\Phi_n(a)=\frac1{{\varepsilon}_n}\, \vert a_n\vert$). In fact, in this case one cannot hope for positive results of any kind: as observed by V. Devinck ([@Vincent]) it follows from a result of C. Badea and V. Müller ([@BadMull]) that $\mathbf S_{\bar{\varepsilon}}\,$-mixing operators just do not exist at all (at least on a Hilbert space).
In the same spirit, it can be shown that it is impossible to find any $c_0$-like description for $\mathcal B=L^1(m)$, the family of all measures absolutely continuous with respect to Lebesgue measure on $\TT$ (see [@MZ]). One may also note that $\mathcal FL^1(m)$ is not an ideal of $\ell^\infty(\ZZ)$ (see [@JPK Chapter 5, Proposition 6]).
As a more extreme example, the well studied notion of *mild mixing* (see [@Aa]) does not fit at all into the framework. Indeed, in this case the relevant family of measures $\mathcal B$ is the following: a measure $\sigma$ is in $\mathcal B$ iff it annihilates every *weak Dirichlet set* (a Borel set $D\subset \TT$ is weak Dirichlet if, for any measure $\mu$ supported on $D$, one can find an increasing sequence of integer $(n_k)$ such that $z^{n_k}\to\mathbf 1$ in $L^2(\mu)$). By a result of S. Kahane ([@SK]), the family $\mathcal B$ is extremely complicated, namely *non Borel* in $(\mathcal M(\TT),w^*)$. So there is absolutely no hope of finding a $c_0$-like description for $\mathcal B$.
Now, if one is interested in Hilbert spaces only, Theorem \[abstracteasy\] is arguably rather general since the $c_0$-like property is not required. However, that the family $\mathbf S$ should be *norm-closed* in $\ell^\infty(\ZZ_+)$ is already quite a restrictive condition: indeed, this implies that the strongest mixing property that can be reached is, precisely, strong mixing. In particular, Theorem \[abstracteasy\] cannot be applied to any “summability" condition on the Fourier coefficients. On the other hand, the following example can be handled: let $\mathcal F$ be any translation-invariant filter on $\ZZ_+$, and take as $\mathbf S$ the family of all sequences $(a_n)\in\ell^\infty (\ZZ_+)$ tending to $0$ along $\mathcal F$. Perhaps more importantly, there are quite natural (norm-closed) families $\mathbf S$ which are not ideals of $\ell^\infty (\ZZ_+)$. The most irritating example is the ergodic case. As already observed a set $D\subset \TT$ is $\mathbf S_{\rm erg}$-small if and only if $D\subset\{ \mathbf 1\}$. Hence, if Theorem \[abstracteasy\] could be applied in this case, the result would read as follows: *if $T\in\mathfrak L(X)$ and if ${\rm span}\left(\bigcup_{\lambda\in\TT\setminus\{\mathbf 1\}} \ker(T-\lambda)\right)$ is dense in $X$, then $T$ is ergodic in the Gaussian sense.* But this is clearly not true since for example $T=- id$ satisfies the assumption and is not even hypercyclic.
In spite of that, it might happen that an operator $T\in\mathfrak L(X)$ is ergodic in the Gaussian sense as soon as *it is hypercyclic* and the $\TT$-eigenvectors span $X$; but the proof of such a result would require at least one new idea. This is, of course, related to the following well known open problem: is it true that every *chaotic* operator ( a hypercylic operator with a dense set of periodic points) is frequently hypercyclic?
Semigroups {#Semigroups}
----------
The proof of Theorem \[WS\] can be easily adapted to get analogous results in the continuous case, for one-parameter semigroups of operators $(T_t)_{t\geq 0}$. Indeed, the mixing properties make perfect sense in the continuous case, and the sets of extended uniqueness are well defined on $\RR$ just like on any locally compact abelian group. The only “difference" is that, according to the (point-)spectral mapping theorem for $C_0$-semigroups, the unimodular eigenvalues of the single operator $T$ from the discrete case should be replaced with the purely imaginary eigenvalues of the semigroup generator in the continuous case. Hence, the continuous analogue of Theorem \[WS\] reads as follows.
\[THMSG\] Let $X$ be a separable Banach space and let $\mathcal T=(T_t)_{t\geq 0}$ be a $C_0$-semigroup in $\mathfrak L(X)$ with infinitesimal generator $A$.
1. If the linear span of $\bigcup_{\theta\in\RR\setminus D}\ker (A-i\theta)$ is dense in $X$ for any countable set $D\subset\RR$, then $\mathcal T$ is weakly mixing in the Gaussian sense.
2. If the linear span of $\bigcup_{\theta\in\RR\setminus D}\ker (A-i\theta)$ is dense in $X$ for any $\mathcal U_0$-set $D\subset\RR$, then $\mathcal T$ is strongly mixing in the Gaussian sense.
Since the proof would essentially be a matter of changing the notation, we shall not give any detail. We note, however, that the semigroup case is obviously of some interest in view of its connections with partial differential equations. See [@BK], [@Las] or [@R2] for more on these matters.
In another direction, a perhaps ambitious program would be to consider linear representations of more general semigroups; in other words, to establish results like Theorem \[WS\] for semigroups of operators $(T_\gamma)_{\gamma\in\Gamma}$ with $\Gamma$ no longer equal to $\NN$ or $\RR_+$. There is no difficulty in defining ergodicity or strong mixing in this setting, and the spectral approach makes sense if $\Gamma$ is a locally compact abelian group. However, it is not clear what the correct “perfect spanning" property should be.
### Translation semigroups
To illustrate Theorem \[THMSG\], let us give a new and very simple example of a strongly mixing $C_0$-semigroup, that cannot be reached by applying the results of [@BG3], [@BG2] or [@BM].
Let $\rho:\mathbb R_+\to(0,\infty)$ be a locally bounded positive function on $\RR_+$, and let $$\mathcal C_{0}(\RR_+,\rho):=\left\{f\in \mathcal C(\mathbb R_+); \lim_{x\to+\infty} f(x)\rho(x)=0\right\}$$ endowed with its natural norm, $\|f\|=\|f \rho\|_\infty$. Moreover, assume that $\rho$ is an *admissible weight* in the sense of [@DSW], which means that $$C(t):=\sup_{x\in\RR_+}\frac{\rho(x)}{\rho(x+t)}<\infty$$ for any $t\geq 0$ and $C(t)$ is locally bounded on $\RR_+$. Then the *translation semigroup* $\mathcal T=(T_t)_{t\geq 0}$ defined by $$T_tf(x)= f(x+t)$$ is a $C_0$-semigroup on $\mathcal C_{0}(\RR_+,\rho)$.
It is proved in [@BBCP] that $\mathcal T$ is strongly mixing in the topological sense if and only if $\rho(x)\xrightarrow{x\to\infty} 0$ as $x\to\infty$. We now show that this is in fact equivalent to strong mixing in the Gaussian sense.
The translation semigroup $\mathcal T$ is strongly mixing in the Gaussian sense on $\mathcal C_{0}(\RR_+,\rho)$ if and only if $\rho(x)\to 0$ as $x\to+\infty$.
Assume that $\rho(x)\to 0$ as $x\to+\infty$. The infinitesimal generator of $\mathcal T$ is the derivation operator (denoted by $A$), whose domain includes all $\mathcal C^1$ functions on $\RR_+$ with a bounded and uniformly continuous derivative. For any $\theta\in\RR$, the function $e_{\theta}(x)=e^{i\theta x}$ is in $\ker(A-i\theta)$, and the map $\theta\mapsto e_{\theta}$ is clearly continuous from $\RR$ into $\mathcal C_{0}(\RR_+,\rho)$. Moreover, it follows easily from (the Hahn-Banach theorem and) the injectivity of the Fourier transformation that the linear span of the functions $e_{\theta}$ is dense in $\mathcal C_{0}(\RR_+,\rho)$. By Theorem \[THMSG\], we conclude that $\mathcal T$ is strongly-mixing in the Gaussian sense.
One may also consider the weighted $L^p$ spaces $L^p(\RR_+,\rho)$ ($1\leq p<\infty$) defined by the condition $$\int_0^\infty \vert f(x)\vert^p\, \rho(x)\, dx<\infty\, .$$ With exactly the same proof as above, one gets that the translation semigroup is strongly mixing in the Gaussian sense on $L^p(\RR_+,\rho)$ as soon as $$\int_0^\infty \rho (x)\, dx<\infty\, .$$
The size of the set of hypercyclic vectors
------------------------------------------
It is well known that if $T$ is a hypercyclic operator acting on a separable Fréchet space $X$, then $HC(T)$ (the set of hypercyclic vectors for $T$) is a dense $G_\delta$ subset of $X$. Moreover, as observed in the introduction, if $T$ happens to be ergodic with respect to some probability measure $\mu$ with full support, then $\mu$-almost every $x\in X$ is a hypercyclic vector for $T$. Thus, $HC(T)$ is large both in the Baire category sense and in a measure-theoretic sense.
Now, there are many other natural notions of “largeness" in analysis. A quite popular one is that of *prevalence*, which is discussed at length in [@HSY]. In a Polish abelian group $G$, a set is prevalent if its complement $A$ is Haar-null in the sense of Christensen [@Chr], one can find a Borel probability measure $\nu$ on $G$ such that $\nu (A+g)=0$ for every $g\in G$. Some results concerning prevalence and hypercyclicity are proved in [@BMM], and much more spectacular results regarding the size of the set of hypercyclic vectors are to be found in [@GR].
For some reasons, it is not incongruous to expect that if an operator $T$ is ergodic in the Gaussian sense, then $HC(T)$ is *not* prevalent and even Haar-null. We are not able to prove this, but this is indeed true for a large class of ergodic weighted shifts, as shown by the following result.
\[PROPHAARNULL\] Let $X$ be a Banach space, and let $T\in\mathcal L(X)$. Assume that one can find $u\in X$ such that $$\sum_{n=0}^\infty \frac1{\|T^n (u)\|}<\infty\, .$$ Then $HC(T)$ is Haar-null.
Considering only the real-linear structure of $X$, we may assume that $X$ is a real Banach space.
An efficient way of proving that a set $A\subset X$ is Haar-null is to exhibit some finite-dimensional subspace $V$ of $X$ such that $$\forall x\in X\;\forall^{a.e.} v\in V\; :\; x+v\not\in A\, ,$$ where $\forall^{a.e.}$ refers to Lebesgue measure on $V$. In the terminology of [@HSY], such a subspace $V$ is called a *probe* for $A$.
We show that the one-dimensional subspace $V=\RR u$ is a probe for $HC(T)$. So let $x\in X$ be arbitrary, set $\Lambda:=\{\lambda\in\RR;\ x+\lambda u\in HC(T)\}$, and let us prove that $\Lambda$ has Lebesgue measure $0$.
For any $\lambda\in \Lambda$, one can find arbitrary large $n\in\NN$ such that $$\|T^n (x)+\lambda T^n (u)\|\leq 1,$$ and hence such that $$\left\vert \lambda-({\|T^n(x)\|}/{\|T^n (u)\|})\right\vert\leq \frac1{\|T^n (u)\|}\,\cdot$$ Putting $a_n:= {\|T^n(x)\|}/{\|T^n (u)\|}$, it follows that $$|\lambda|\in\bigcap_{N\in\NN}\, \bigcup_{n\geq N}\left[a_n-\frac1{\|T^n (u)\|},a_n+\frac1{\|T^n (u)\|}\right]$$ for every $\lambda\in\Lambda$. In particular, the Lebesgue measure of $\Lambda$ is not greater than $$\inf_{N\in\NN} \; 2 \sum_{n\geq N}\frac1{\|T^n u\|}\, ,$$ which is a complicated way to write $0$.
Let $B_{\mathbf w}$ be a weighted backward shift on $X_p=\ell^p(\NN)$, $1\leq p<\infty$ or $X_\infty =c_0(\NN)$. Assume that the weight sequence $\mathbf w=(w_n)_{n\geq 1}$ satisfies $$\sum_{n=1}^\infty \frac{1}{\vert w_1\cdots w_n\vert^{p/p+1}}<\infty\, .$$ Then $HC(B_{\mathbf w})$ is Haar-null. This holds for example if $\liminf\limits_{n\to\infty} \vert w_n\vert>1$.
Set $w_0:=0$ and denote by $(e_k)_{k\geq 0}$ the canonical basis of $X_p$. If $p<\infty$, consider the vector $$u:=\sum_{k=0}^\infty \frac{1}{\vert w_0\cdots w_k\vert^{1/p+1}}\, e_k\, .$$ If $p=\infty$, put $$u:=\sum_{k=0}^\infty \frac{A_k}{w_0\cdots w_k}\, e_k\, ,$$ where $(A_k)$ is any sequence of positive numbers such that $A_k=o(w_0\cdots w_k)$ and $\sum_0^\infty 1/A_k<\infty$.
Thus, we see that if $B$ is the usual (unweighted) backward shift, then $HC(\lambda B)$ is Haar-null in $X_p$ for any $p\in[1,\infty]$ if $\vert\lambda\vert>1$.
Some questions
--------------
To conclude the paper, we list a few “natural" questions, most of which have already been raised.
1. Is Theorem \[abstract\] true without assuming that the family $\mathbf S$ is $c_0$-like?
2. Is Theorem \[abstracteasy\] true for $L^1(m)$-mixing, or in the mild mixing case?
3. Let $T$ be a hypercyclic operator on $X$ whose $\TT$-eigenvectors span a dense subspace of $X$. Is $T$ ergodic in the Gaussian sense? Is $T$ frequently hypercyclic?
4. \[truc\] Let $T\in\mathcal L(X)$, and assume that for any set $D\subset\TT$ *with empty interior*, the linear span of $\bigcup_{\lambda\in\TT\setminus D}\ker(T-\lambda)$ is dense in $X$. Is it possible to find a countable family of continuous $\TT$-eigenvector fields $(E_i)_{i\in I}$, where each $E_i$ is defined on some nontrivial closed arc $\Lambda_i\subset\TT$, such that ${\rm span}\left(\bigcup_{i\in I} E_i(\Lambda_i)\right)$ is dense in $X$?
5. Let $H$ be a Hilbert space, and let $T\in\mathfrak L(H)$ be mixing with respect to some Borel measure on $H$ with full support. Is $T$ mixing in the Gaussian sense?
6. On which Banach spaces is it possible to find ergodic operators (in the Gaussian sense) with no eigenvalues? The space $X$ should of course not have cotype 2, but this is not enough: for example, if $X$ is hereditarily indecomposable then there are no frequently hypercyclic operators at all on it, and hence no ergodic operators either ([@Shk]). In fact, the question splits into two separate problems: (i) on which Banach spaces is it possible to find ergodic operators? (ii) what can be said if the space does not have cotype 2? We refer to [@GRosetc] for general results regarding the first problem.
7. Is there an ergodic weighted shift on $c_0(\NN)$ with no unimodular eigenvalues?
8. Let $T\in\mathfrak L(X)$ be ergodic in the Gaussian sense. Are the ergodic measures with full support dense in the set of all $T$-invariant measures (endowed with the usual Prokhorov topology)? See [@Sig] for a positive answer in a completely different situation, and [@CS] for more in that direction.
9. Is it true that if $T$ is an ergodic operator, then $HC(T)$ is Haar-null? Is this true at least for weighted shifts?
J. Aaronson, *An introduction to infinite ergodic theory*. Mathematical Surveys and Monographs [**50**]{}, Amer. Math. Soc. (1997). F. Albiac and N. Kalton, *Topics in Banach space theory.* Graduate Texts in Mathematics [**233**]{}, Springer (2006). C. Badea and S. Grivaux, [*Unimodular eigenvalues, uniformly distributed sequences and linear dynamics*]{}. Adv. Math. [**211**]{} (2007), 766–793. C. Badea and S. Grivaux, [*Size of the peripherical point spectrum under power or resolvent growth conditions*]{}. J. Funct. Anal. [**247**]{} (2007), 302–329. C. Badea and V. Müller, [*On weak orbits of operators*]{}. Topology Appl. [**156**]{} (2009), 1381–1385. F. Bayart and S. Grivaux, [*Hypercyclicity and unimodular point spectrum*]{}. J. Funct. Anal. [**226**]{} (2005), 281–300. F. Bayart and S. Grivaux, [*Frequently hypercyclic operators*]{}. Trans. Amer. Math. Soc. [**358**]{} (2006), no. 11, 5083–5117. F. Bayart and S. Grivaux, [*Invariant Gaussian measures for operators on Banach spaces and linear dynamics*]{}. Proc. Lond. Math. Soc. [**94**]{} (2007), 181–210. F. Bayart and É. Matheron, *Dynamics of linear operators*. Cambridge Tracts in Mathematics [**179**]{}, Cambridge University Press (2009). F. Bayart, É. Matheron and P. Moreau, [*Small sets and hypercyclic vectors*]{}. Comment. Math. Univ. Carolinae [**49**]{} (2008), 53–65. F. Bayart and I. Z. Rusza, [*Difference sets and frequently hypercyclic weighted shifts*]{}. Preprint (2011). T. Bermúdez, A. Bonilla, J. Conejero and A. Peris, [*Hypercyclic, topologically mixing and chaotic semigroups on Banach spaces*]{}. Studia Math. [**170**]{} (2005), 57–75. V. I. Bogachev *Gaussian measures.* Mathematical Surveys and Monographs [**62**]{}, Amer. Math. Soc. (1998). A. Bonilla and K. -G. Grosse-Erdmann, [*On a theorem of Godefroy and Shapiro*]{}. Integral Equations operator Theory [**56**]{} (2006), 151–162. P. Brunovský and J. Komornik, [*Ergodicity and exactness of the shift on $C[0,\infty]$ and the semiflow of a first order partial differential equation.*]{} J. Math. Anal. Appl. [**104**]{} (1984), no. 1, 235–245. S. A. Chobanyan, V. I. Tarieladze and N. N. Vakhania, *Probability distributions on Banach spaces.* Mathematics and its Applications (Soviet Series) [**14**]{}, Springer (1987). J. P. R. Christensen, [*On sets of [H]{}aar measure zero in Polish abelian groups*]{}. Israel J. Math. [**13**]{} (1972), 255–260 Y. Coudène and B. Schapira, [*Generic measures for hyperbolic flows on non-compact spaces*]{}. Israel J. Math. [**179**]{} (2010), 157–172. G. Da Prato, *An introduction to infinite-dimensional analysis*. Universitext, Springer (2006). M. De la Rosa, L. Frerick, S. Grivaux and Alfredo Peris, [*Frequent hypercyclicity, chaos, and unconditional Schauder decompositions.*]{} Israel J. Math. (to appear). G. Debs and J. Saint Raymond, [*Ensembles boréliens d’unicité et d’unicité au sens large*]{}. Ann. Inst. Fourier, Grenoble [**37**]{} (1987), no. 3, 217–239. W. Desch, W. Schappacher and G. F. Webb, [*Hypercyclic and chaotic semigroups of linear operators*]{}. Ergodic Theory Dynam. Systems [**17**]{} (1997), 793–819. V. Devinck, [*Strongly mixing operators on Hilbert spaces and speed of mixing*]{}. Preprint (2011). T. Eisner and S. Grivaux, [*Hilbertian Jamison sequences and rigid dynamical systems*]{}. Preprint (2011). E. Flytzanis, [*Unimodular eigenvalues and linear chaos in Hilbert spaces*]{}. Geom. Funct. Anal. [**5**]{} (1995), 1–13. E. Glasner, *Ergodic theory via joinings*. Mathematical Surveys and Monographs [**101**]{}, Amer. Math. Soc. (2003). G. Godefroy and J. Shapiro, [*Operators with dense, invariant, cyclic vector manifolds*]{}. J. Funct. Anal. [**98**]{} (1991), 229–269. S. Grivaux, [*A new class of frequently hypercyclic operators*]{}. Indiana Univ. Math. J. (2011), [to appear]{}. S. Grivaux and M. Roginskaya, [*On Read’s type operators on Hilbert spaces*]{}. Int. Math. Res. Not. IMRN (2008), Art. ID rnn083, 42 pp. K. -G Grosse-Erdmann, [*Dynamics of linear operators.*]{} Topics in operator theory and complex analysis 41–84, University of Malaga (2007). K. -G Grosse-Erdmann and A. Peris, *Linear chaos.* Universitext, Springer (to appear). B. R. Hunt, T. Sauer and J. A. Yorke, [*Prevalence: a translation invariant “almost every" on infinite-dimensional spaces*]{}. Bull. Amer. Math. Soc. [**27**]{} (1992), no. 2, 217–238. J. P. Kahane, *Some random series of functions*. Cambridge Studies in Advanced Mathematics [**5**]{}, Cambridge University Press (1985). S. Kahane, [*On the complexity of sums of Dirichlet measures*]{}. Ann. Inst. Fourier, Grenoble [**43**]{} (1993), no. 1, 111–123. G. Kalisch, [*On operators on separable Banach spaces with arbitrary prescribed point spectrum*]{}. Proc. Amer. Math. Soc. [**34**]{} (1972) 207–208. Y. Katznelson, *An introduction to harmonic analysis.* Dover (1976). A. S. Kechris, *Classical descriptive set theory*. Graduate Texts in Mathematics [**156**]{}, Springer (1995). A. S. Kechris and A. Louveau, *Descriptive set theory and the structure of sets of uniqueness*. London Math. Soc. Lecture Notes Series [**128**]{}, Cambridge University Press (1987). A. Lasota, [*Invariant measures and a linear model of turbulence.*]{} Rend. Sem. Mat. Univ. Padova [**61**]{} (1979), 39–48. R. Lyons, [*Fourier–Stieltjes coefficients and asymptotic distribution modulo $1$.*]{} Ann. Math. [**122**]{} (1985) 155–170. É. Matheron and M. Zelený, [*Rudin-like sets and hereditary families of compact sets*]{}. Fund. Math. [**185**]{} (2005), 97–116. J. C. Oxtoby and S. M. Ulam, [*On the existence of a measure invariant under a transformation*]{}. Ann. Math [**40**]{} (1939), 560–566. K. Petersen, *Ergodic theory*. Cambridge Studies in Advanced Mathematics [**2**]{}, Cambridge University Press (1983). W. Rudin, *Functional analysis*. McGraw-Hill (1973). R. Rudnicki, [*Gaussian measure-preserving linear transformations*]{}. Univ. Iagel Acta Math. [**30**]{} (1993), 105–112. R. Rudnicki, [*Chaos for some infinite-dimensional dynamical systems.*]{} Math. Methods Appl. Sci. [**27**]{} (2004), no. 6, 723–738. H. Salas, [*Hypercyclic weighted shifts*]{}. Trans. Amer. Math. Soc. [**347**]{} (1995), no. 3, 993–1004. S. Shkarin, [*On the spectrum of frequently hypercyclic operators*]{}. Proc. Amer. Math. Soc. [**137**]{} (2009) no. 1, 123–134. K. Sigmund, [*Generic properties of invariant measures for axiom A diffeomorphisms*]{}. Inventiones Math. [**11**]{} (1970), 99–109.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- |
\
Panjab University, Chandigarh, India\
E-mail:
title: |
Dynamical net charge fluctuations at RHIC\
energies in STAR
---
Introduction
============
The STAR experiment at the Relativistic Heavy Ion Collider (RHIC) provides the ability to investigate the behaviour of strongly interacting matter at high density and to study the Quark Gluon Plasma (QGP). In the year 2010 (Run 10), RHIC started its Beam Energy Scan (BES) program and collided Au+Au ions from $\sqrt{s_{NN}}$ = 39 GeV down to 7.7 GeV covering 112 $<$ $\mu_{B}$ $<$ 410 MeV. This allows one to access and probe broad regions of the QCD phase diagram. Event-by-event net-charge fluctuations have been proposed as an indicator of the QGP formation in heavy ion collisions. The fluctuation in net-charge depends on the squares of the charges present in the system, which depends on the state from which it originates. The system passing through a QGP phase which has quarks as charge carriers should result in a significantly different net-charge fluctuation as compared to Hadron Gas (HG). The variance of the event-by-event difference of the numbers of positive and negative particles scaled by the total charged particle multiplicity, a quantity called $\textit{D}$, should be approximately four times smaller in a QGP medium than in a gas of hadrons. The charge fluctuation is measured in terms of $ \textit{D} $ defined as : $$D=4\dfrac{\langle\delta Q^{2}\rangle}{\langle N_{ch}\rangle},$$ where $\langle\delta Q^{2}\rangle$ is the net charge variance, $Q=N_{+}-N_{-}$ and $N_{ch}=N_{+}+N_{-}$. Here $N_{+}$ and $N_{-}$ are the number of negative and positive particles, measured in specific transverse momentum $(p_{T})$ and pseudorapidity $(\eta)$ window.\
The value of $\textit{D}$ is 4 for an uncorrelated pion gas and is reduced by about 30$\%$ in the presence of resonances. For a thermal system of free quarks and gluons, the value is significantly lower and has been calculated to be $\approx$ 1. [@ebye_1; @ebye_2]\
The event-by-event net-charge fluctuations have also been estimated by calculating the quantity $\nu_{+-,dyn}$ defined as: [@b2; @pruneau; @monika; @prc] : $$\nu_{+-,dyn}=\dfrac{\langle N_{+}(N_{+}-1)\rangle}{\langle N_{+}\rangle^{2}}+\dfrac{\langle N_{-}(N_{-}-1)\rangle}{\langle N_{-}\rangle^{2}}
-2\dfrac{\langle N_{-}N_{+}\rangle}{\langle N_{+}\rangle\langle N_{-}\rangle}$$ which is a measure of the relative correlation of $+$$+$, $-$ $-$ and $+$$-$ pairs. The $\nu_{+-,dyn}$ has been found to be robust against experimental inefficiencies [@robust]. The value of *D* is related to $\nu_{+-,dyn}$ as : $$\langle N_{ch}\rangle \nu_{+-,dyn}=D-4 .$$
Analysis Details and Results
============================
The measurements of net-charge fluctuations as a function of centrality in Au+Au collisions at $\sqrt{s_{NN}}$ = 7.7, 11.5, 19.6, 27 and 39 GeV data collected in years 2010 and 2011 by the STAR experiment are reported. For this analysis, we use charged
![ Dynamical net-charge fluctuations, $\nu_{+-,dyn}$, of charged particles as a function of number of participating nucleons, $\langle N_{part}\rangle$.[]{data-label="fig1"}](icpaqgp_fig/sep14_nudyn.pdf){width="80mm"}
particle tracks from the Time Projection Chamber (TPC) for -0.5 $<$ $\eta$ $<$ 0.5 with transverse momenta in the range 0.2 $<$ $p_{T}$ $<$ 5.0 GeV/$c$. For the
![Dynamical net-charge fluctuations, $\nu_{+-,dyn}$, scaled with the average number of participating nucleons as a function of $\langle N_{part}\rangle$.[]{data-label="fig2"}](icpaqgp_fig/sep14_scaling.pdf){width="80mm"}
centrality selection, the uncorrected multiplicity of charged particles within 0.5 $<$ $|\eta|$ $<$ 1.0 is used in order to avoid the autocorrelation between the centrality definition and the $\nu_{+-,dyn}$ observable.\
Figure \[fig1\] shows the measurements of $\nu_{+-,dyn}$ at $\sqrt{s_{NN}}$ = 7.7, 11.5, 19.6, 27 and 39 GeV as a function of the average number of participating nucleons, $\langle N_{part}\rangle$. It also includes the previous results from STAR at $\sqrt{s_{NN}}$ = 200, 130 and 62.4 GeV, and ALICE results for Pb+Pb collisions at $\sqrt{s_{NN}}$ = 2.76 TeV [@monika; @alice]. The uncertainties shown are only statistical. In all cases, the values of $\nu_{+-,dyn}$ are negative indicating the dominance of the correlation between positive and negative charged particles. The strength of the correlation decreases while going from peripheral to central collisions. Also, the magnitude of fluctuations decreases as the beam energy increases.\
Figure \[fig2\] shows $\nu_{+-,dyn}$ scaled with the average number of participating nucleons, $N_{part}\nu_{+-,dyn}$, as a function of $\langle N_{part}\rangle$. The measured data scaled by the number of participating nucleons exhibits either a weak or no centrality dependence at all of the measured energies.\
The magnitude of the net-charge correlations is affected by the global charge
![$\langle N_{ch}\rangle \nu_{+-,dyn}^{corr}$ (left-axis) and $D$ (right-axis) as a function of beam energy for 0-5% collisions. The theoretical predictions for a Pion Gas, HG and a QGP are also indicated. Systematic uncertainties are not yet included.[]{data-label="fig3"}](icpaqgp_fig/sep16_dval.pdf){width="100mm"}
conservation and the finite size of the colliding system. The contribution of the global charge conservation effect is estimated to be -4/$\langle N_{total}\rangle$, [@pruneau] where $\langle N_{total}\rangle$ is the average number of charged particles produced over the full phase space. The corrected value of $\nu_{+-,dyn}$ is calculated as $$\nu_{+-,dyn}^{corr} = \nu_{+-,dyn} + \dfrac{4}{\langle N_\mathrm{{total}}\rangle}$$ Here the $\langle N_{ch}\rangle$ is the efficiency corrected average charged particle multiplicity for Au+Au collisions. Both $\langle N_{ch}\rangle$ and $\langle N_{total}\rangle$ have been estimated from the PHOBOS experiment data [@phobos_ntot]. Figure \[fig3\] shows $\langle N_{ch}\rangle \nu_{+-,dyn}^{corr}$ along left y-axis and $D$ along right y-axis as a function of the colliding energy for 0-5% central collisions within $|\eta|$ $<$ 0.5. The uncertainties shown are statistical only. The theoretical predictions for Pion Gas, Hadron Resonance Gas and QGP have also been displayed in the figure. It can be observed that the net-charge fluctuations corrected for the global charge conservation when scaled with $\langle N_{ch}\rangle$ decrease with increases in the beam energy. The $D$ measure linearly decreases with increasing energy. The net-charge fluctuations observables approach the expectation for a Pion Gas as the beam energy decreases.\
In summary, we report recent results of the net-charge fluctuations for Au+Au collisions at 7.7, 11.5, 19.6, 27 and 39 GeV. There is a monotonic reduction in the magnitude of dynamical net-charge fluctuations with increasing number of participants. Dynamical net-charge fluctuations are observed to follow approximate $N_{part}$ scaling. Top 5% central collisions results show that $\langle N_{ch}\rangle \nu_{+-,dyn}^{corr}$ generally decreases with increasing colliding energy.
[20]{} S. Jeon, V. Koch, Phys. Rev. Lett. 83, 5435 (1999). S. Jeon, V. Koch, Phys. Rev. Lett. 85, 2076 (2000). J. Adams *et al.* \[STAR Collaboration\] Phys. Rev. C 68, 044905 (2003). C. Pruneau, S. Gavin, S. Voloshin, Phys. Rev. C 66, 044904 (2002). B.I. Abelev *et al.* \[STAR Collaboration\] Phys. Rev. C 79,024906 (2009). Bhanu Sharma, Madan M. Aggarwal, Nihar Ranjan Sahoo and Tapan K. Nayak,\
Phys. Rev. C. 91, 024909 (2015) J. Nystrand, E. Stenlund, H. Tydesjo, Phys. Rev. C68, 034902 (2003). B. Abelev *et al.* (ALICE Collaboration), Phys. Rev. Lett. 110, 152301 (2013). B. Adler *et al.* (PHOBOS Collaboration), Phys. Rev. C 83 024913 (2011).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Using data obtained through simultaneous measurements of $(\vec p,\vec
p\,')$ spin-transfer observables and $(\vec p,p' \gamma)$ coincident spin observables, we have made a model-independent determination of the complete scattering amplitude for the 15.11 MeV, 1$^+$, $T$=1 state in $^{12}$C at an incident proton energy of 200 MeV, for four proton scattering angles ranging from $\theta_{\rm c.m.}$ = 5.5$^\circ$ to 16.5$^\circ$. At each angle, 16 different observables were determined, whereas only 11 independent quantities are required to specify the transition amplitude for this state. It had been shown previously that the set of observables measured span the allowed space; hence the system is overdetermined, which allowed us to extract, in a model-independent fashion, each of the individual spin-operator amplitudes that characterize the reaction. Additional insight into the physical mechanisms that drive this transition is obtained by mapping out the momentum-transfer dependence of these amplitudes. We also compare the magnitudes and phases determined for each of the spin-operator amplitudes to the predictions of calculations performed in both relativistic and nonrelativistic frameworks, and discuss the physics content of these comparisons.
address: ' Indiana University Cyclotron Facility, Bloomington, Indiana 47408 '
author:
- 'S. P. Wells[@spw] and S. W. Wissink'
title: ' Model-independent determination of the $^{12}$C[$(p,p'')$]{}$^{12}$C$^{\ast}$ (15.11 MeV, 1$^+$, [$T$]{}=1) transition amplitude at 200 MeV '
---
Introduction
============
At intermediate energies of $\sim$150–500 MeV, hadron-induced nuclear reactions have served as a rich source of information, both to further our understanding of nuclear structure, and to illustrate how the elementary nucleon-nucleon ($NN$) interaction may be modified in the nuclear medium [@Bag92]. This is particularly true of spin observables, which often contain contributions from interferences between different pieces of the full scattering amplitude. These interferences, inaccessible through measurements of differential cross sections alone, can thus provide information about not only the magnitudes, but also the relative phases of individual terms in the transition amplitude. Ideally, one would like to extract precise values for each of these (complex) terms in a manner which does not rely on specific model assumptions. This information, in turn, can then provide the most stringent tests of a given theoretical prediction for the hadronic process of interest.
To obtain deeper insight into the nucleon-nucleus ($NA$) interaction, several inelastic transitions have been identified as being particularly amenable to experimental investigation. The strong, isovector 1$^+$ state in $^{12}$C at an excitation energy of 15.11 MeV has been extensively studied, using a variety of probes operating under a broad range of kinematic conditions. Due to its $1^+$ nature, hadronic excitation of this state represents an “unnatural parity” transition; consequently, this process should be particularly sensitive to the spin-dependent parts of the $NA$ interaction, and in fact has long served as a critical test for our understanding of the $\Delta S = \Delta T = 1$ component of the $NN$ effective interaction [@Lov85]. As an example, the spin observable $P-A_y$, i.e., the difference between the polarization induced in the outgoing proton and the scattering yield asymmetry that results from use of a polarized incident beam, has been investigated both experimentally [@Car82; @Hic88] and theoretically [@Lov84]. This quantity, which in a non-relativistic framework contains only interference terms between competing pieces of the transition amplitude, has been shown to be sensitive to the coupling of the nucleon spin to the bound nucleon current [@Lov84]. In later relativistic treatments of proton-nucleus scattering [@Pik85; @Shep86], it became clear that these same nuclear current terms appear more naturally in a relativistic formalism, and are produced through the linear couplings between the upper and lower components of the bound nucleon wave function [@Pik85; @Shep86]. In either description, $P-A_y$ is dependent on the momentum of the nucleons inside the nucleus, and is therefore sensitive to the off-shell behavior of these nucleons, and the non-local or exchange nature of the interaction.
Thus, detailed comparisons of measured spin observable data to various predictions for specific nuclear transitions has traditionally been our best means of constraining theoretical models. In recent years, though, such comparisons have raised a number of concerns, at both the experimental and theoretical ends. For example, it has been shown [@Mos82; @Ble82] that in a direct-only plane-wave impulse approximation (PWIA), certain combinations of $(\vec p,\vec p\,')$ spin-transfer observables are directly related to individual terms in the effective $NN$ interaction, weighted by corresponding nuclear response functions. When physical processes such as distortion of the projectile waves and knock-on exchange are included, however, it becomes less clear what these combinations of $(\vec p,\vec p\,')$ observables represent physically, thereby blurring any simple interpretation of these quantities. On the experimental side, counting arguments show that even “complete” sets of $(\vec p,\vec p\,')$ spin-transfer measurements provide (at most) eight independent pieces of information, while a transition of the general form $0^+ \to
J^{\pi}$ requires knowledge of $(8J+3)$ independent quantities in order to fully specify the scattering amplitude.
To reduce this ambiguity, it is clearly desirable to establish a more direct connection between theory and experiment, such as would be obtained from a model-independent extraction of individual elements of the scattering amplitude. In so doing, one bypasses the ‘intermediate’ role played by spin observables, which represent non-trivial combinations of many of these elements. Carrying out such a program, though, requires access to information beyond that provided by $(\vec
p,\vec p\,')$ spin-transfer measurements alone. In particular, one must probe the polarization state of the ‘other particle’ involved in the reaction: the recoil nucleus. This can be achieved through study of the angular correlation in the final state between the scattered (out-going) proton and the particle(s) emitted in the decay of the excited nucleus, as in reactions of the type $(\vec p,p' \gamma)$. More specifically, for proton excitation of a $0^+ \to 1^+$ transition, followed by electromagnetic decay back to the ground state, it has recently been proven formally [@Pik90; @Ram94] that certain $(\vec p,p' \gamma )$ measurements, when combined with complete sets of $(\vec p,\vec p\,')$ observables, allow for a complete description of the transition amplitude. Note that in this case a total of $8J+3=11$ independent quantities must be determined.
In this paper, we report the first model-independent determination of the complete transition amplitude for any nuclear final state with $J
\neq 0$. This analysis is based on a detailed set of measurements [@SPW95] carried out at the Indiana University Cyclotron Facility (IUCF), in which both $(\vec p,\vec p\,')$ spin-transfer and $(\vec
p,p' \gamma)$ coincident observables were determined simultaneously for excitation of the isovector $1^+$ state at 15.11 MeV in $^{12}$C. Data were taken at four proton scattering angles ($\theta_{\rm c.m.}$ = 5.5$^\circ$, 8.8$^\circ$, 12.1$^\circ$, and 16.5$^\circ$) at an incident proton beam energy of 200 MeV. As will be shown below, this data set allows for extraction of each individual (complex) spin-dependent term in the transition amplitude in a model-independent manner. Moreover, because this procedure has been applied at a number of scattering angles, the momentum-transfer ($q$) dependence of each spin-operator amplitude has been mapped out, allowing for a clearer interpretation of the physical mechanisms that drive this transition.
This paper will be organized in the following way. In Sec. II we present the formalism adopted (from Ref. [@Pik90]) for analysis of the data. In Sec. III, we present the observables measured in the IUCF experiment [@SPW95], and express each of these in terms of real and imaginary parts of the individual spin-operator amplitudes. We describe in detail the procedure used to extract these amplitudes from the observables in Sec. IV, and present the results of this procedure in Sec. V. Section VI provides a detailed discussion of the momentum-transfer dependence determined for each amplitude (both magnitude and phase), and compares our results to predictions of calculations performed in both relativistic and nonrelativistic formalisms. Where possible, we discuss the specific physics issues to which each amplitude is sensitive. Our most significant results and conclusions are summarized in Sec. VII.
Formalism
=========
The formalism adopted here follows that of Piekarewicz [*et al.*]{} [@Pik90], and has been presented in detail in Ref. [@SPW95]; only the most important results are given here for completeness. We work in the orthogonal coordinate system defined by $${\bf n} \equiv {\bf p} \times {\bf p'}~; \qquad {\bf K} \equiv
{\bf p} + {\bf p'}~; \qquad {\bf q} \equiv {\bf n} \times {\bf K}~,
\label{eq:coord}$$ where [**p**]{} ([**p**]{}$'$) is the incident (outgoing) proton momentum, [**n**]{} is directed normal to the scattering plane, [**K**]{} is along the direction of the average proton momentum, and (neglecting the reaction $Q$-value) [**q**]{} points in the direction of momentum transfer ${\bf p^{\prime}}$–${\bf p}$. In this frame, the most general form of the scattering amplitude allowed by angular momentum and parity conservation for a $0^+ \to 1^+$ transition can be written in the form [@Pik90] $$\begin{aligned}
\hat T ^p ({\bf p},{\bf p'})
&~=~& A_{n0} ({\bf \hat \Sigma} \cdot {\bf \hat n})
~+~ A_{nn} ({\bf \hat \Sigma} \cdot {\bf \hat n})
({\bf \sigma} \cdot {\bf \hat n})
~+~ A_{KK} ({\bf \hat \Sigma} \cdot {\bf \hat K})
({\bf \sigma} \cdot {\bf \hat K}) \nonumber \\
&~+~& A_{Kq} ({\bf \hat \Sigma} \cdot {\bf \hat K})
({\bf \sigma} \cdot {\bf \hat q})
~+~ A_{qK} ({\bf \hat \Sigma} \cdot {\bf \hat q})
({\bf \sigma} \cdot {\bf \hat K})
~+~ A_{qq} ({\bf \hat \Sigma} \cdot {\bf \hat q})
({\bf \sigma} \cdot {\bf \hat q}) ~,
\label{eq:Tpp}\end{aligned}$$ where $${\bf \hat \Sigma} _M ~\equiv~
\vert 1^+ M \rangle \langle 0^+ \vert$$ is the polarization operator for promoting a $0^+$ state to a state with $J^{\pi}$=$1^+$ and magnetic substate $M$, while ${\vec \sigma}$ are the Pauli spin operators for the projectile. In Eq. \[eq:Tpp\], the $A_{i\mu}$’s are scalar functions of energy and momentum transfer, where the subscripts $i (=n,K,q)$ and $\mu (=0,n,K,q)$ indicate the polarization components of the recoil $1^+$ nucleus ($\bf \Sigma$) and the scattered proton ($\vec \sigma$, with $\sigma _0 \equiv 1$), respectively. Because there are six allowed complex amplitudes, 11 pieces of information (after eliminating one overall phase) are required to specify the scattering amplitude for this transition.
If the final polarization of the nuclear state is undetected, one can sum incoherently over the $\bf \Sigma$ index. In this case, the relevant spin observables can be expressed as $$\begin{aligned}
{{d\sigma}\over{d\Omega _p}}
&~=~& \sum _{ij\mu} A_{i\mu}A^{\ast}_{j\mu} \delta _{ij} ~,
\nonumber \\
{{d\sigma}\over{d\Omega _p}}D_{\alpha \beta}
&~=~& \sum _{ij\mu \nu} A_{i\mu}A^{\ast}_{j\nu} \delta _{ij}
{1\over 2}Tr \{ \sigma _{\alpha} \sigma _\mu \sigma _{\beta}
\sigma _{\nu} \} ~,
\label{eq:Dab}\end{aligned}$$ where $\alpha ,\beta = 0,n,K,q.$ Using the scattering amplitude of Eq. \[eq:Tpp\] and carrying through the Pauli algebra, it is easily shown that only eight of the 16 possible singles $(\vec p,\vec p\,')$ spin observables (the $D_{\alpha \beta}$) are nonzero. Because 11 independent quantities appear in $\hat T ^p ({\bf p},{\bf p'})$, one sees that singles measurements alone can not uniquely define this amplitude; or, put another way, there is information contained in $\hat T ^p ({\bf p},{\bf p'})$ which is not accessible via $(\vec
p,\vec p\,')$ observables. In particular, the presence in Eq. \[eq:Dab\] of the Kronecker $\delta _{ij}$ makes it impossible to determine the relative phase between any two spin operator amplitudes ($A_{i\mu}$’s) that correspond to orthogonal orientations of the recoil nuclear polarization.
If one makes the assumption that the $(p,p' \gamma)$ reaction is strictly a two-step process, then the total transition amplitude (excitation plus decay) can be written as the product of the strong and electromagnetic amplitudes. In this case, all of the coincident $(p,p' \gamma)$ observables may be written in terms of just the singles $(p,p')$ spin operator amplitudes, $A_{i\mu}$, and the $\gamma$-ray branching ratio to the ground state $R$ [@Pik90]. In complete analogy with the singles observables (Eq. \[eq:Dab\]), the spin-dependent coincident observables for this transition can be written in the form [@Pik90] $$\begin{aligned}
{{d^2 \sigma}\over{d\Omega_p d\Omega_{\gamma}}}({\bf \hat k})
&~=~& {{3R}\over{8 \pi}} \sum _{ij \mu}
A_{i \mu} A^{\ast}_{j \mu} t_{ij}({\bf \hat k}) ~, \nonumber \\
{{d^2 \sigma}\over{d\Omega_p d\Omega_{\gamma}}}({\bf \hat k})
D_{\alpha \beta} ({\bf \hat k})
&~=~& \sum _{ij\mu\nu} A_{i\mu} A^{\ast}_{j\nu} t_{ij} ({\bf \hat k})
\, {1\over 2}Tr \{ \sigma_{\alpha} \sigma_\mu \sigma_{\beta}
\sigma_{\nu} \} ~,
\label{eq:Dabk}\end{aligned}$$ where ${\bf \hat k}$ is the momentum direction of the emitted photon, and the photon polarization tensor $t_{ij}$ is given by $$t_{ij} ({\bf \hat k} ) ~\equiv~
\delta_{ij} - ({\bf \hat k} \cdot {\bf \hat e}_i)
({\bf \hat k} \cdot {\bf \hat e}_j) ~,
\label{eq:tij}$$ with ${\bf \hat e}_i$ a unit vector lying along one of the $({\bf \hat
n},{\bf \hat K},{\bf \hat q})$ coordinates defined in Eq. \[eq:coord\]. It is important to emphasize that the coincident observables defined in Eq. \[eq:Dabk\] are written in terms of the [*same*]{} $A_{i\mu}$’s that appear in the definitions of the singles observables, Eq. \[eq:Dab\]. The crucial difference between these two sets of equations is the presence of $t_{ij} ({\bf \hat k}
)$ in the definition of the coincident observables, replacing the $\delta _{ij}$ for the singles observables. Thus, certain $(p,p'
\gamma)$ spin observables [*will*]{} be sensitive to the relative phases between amplitudes for different recoil nuclear polarizations, provided that the emitted $\gamma$–ray has momentum components along [*both*]{} ${\bf \hat e}_i$ and ${\bf \hat e}_j$. Thus, the specific information about the transition amplitude which is lost for the singles observables can be accessed through coincident $(\vec p,p'
\gamma)$ measurements.
Summary of Observables
======================
In this paper, we present detailed analysis of the data from a previous experiment [@SPW95]. In that work, some spin observables were measured using an incident proton beam whose polarization vector was normal to the (horizontal) scattering plane, while others required use of a beam in which the polarization vector had been precessed to lie in the scattering plane. Due to technical problems encountered during data acquisition [@SPW95], reliable values of the differential cross section, $d\sigma / d\Omega_p$, could not be extracted from the experimental yields. While this problem had no effect on the determination of spin-dependent observables, it was necessary to use previously measured values of $d\sigma / d\Omega_p$ for this transition [@Com81] to set the scale for the overall magnitudes of the $A_{i\mu}$ amplitudes. In this paper, these amplitudes will be presented in units of $\sqrt{\mu {\rm b} / {\rm
sr}}$.
The singles observables obtained in Ref. [@SPW95] include the three normal-component spin-transfer coefficients $A_y$, $P$, and $D_{N'N}$, and two linear combinations of the in-plane spin transfer coefficients, $$\begin{aligned}
D_{\lambda} & \equiv &
D_{L'L} \sin \alpha + D_{S'L} \cos \alpha ~, \nonumber \\
D_{\sigma} & \equiv &
D_{L'S} \sin \alpha + D_{S'S} \cos \alpha ~,\end{aligned}$$ where here $\alpha$ is the spin precession angle about ${\bf \hat n}$ experienced by the scattered protons in passing through the magnetic spectrometer used in the experiment. Through use of Eq. \[eq:Dab\], all of these measured $(\vec p,\vec p\,')$ observables can be expressed in terms of the $A_{i\mu}$ amplitudes as follows: $$\begin{aligned}
{{d\sigma}\over{d\Omega _p}} &~=~&
\vert A_{n0} \vert^2 + \vert A_{nn} \vert^2 + \vert A_{KK} \vert^2
+ \vert A_{Kq} \vert^2 + \vert A_{qK} \vert^2 + \vert A_{qq} \vert^2
~, \nonumber \\
{{d\sigma}\over{d\Omega _p}} A_y &~=~&
2[\Re (A_{n0}A^{\ast}_{nn}) +
\Im (A_{KK}A^{\ast}_{Kq} + A_{qK}A^{\ast}_{qq})] ~, \nonumber \\
{{d\sigma}\over{d\Omega _p}} P &~=~&
2[\Re (A_{n0}A^{\ast}_{nn}) -
\Im (A_{KK}A^{\ast}_{Kq} + A_{qK}A^{\ast}_{qq})] ~, \nonumber \\
{{d\sigma}\over{d\Omega _p}} D_{N'N} &~=~&
\vert A_{n0} \vert^2 + \vert A_{nn} \vert^2 - \vert A_{KK} \vert^2
- \vert A_{Kq} \vert^2 - \vert A_{qK} \vert^2 - \vert A_{qq} \vert^2 ~;
\label{eq:norm}\end{aligned}$$ and $$\begin{aligned}
{{d\sigma}\over{d\Omega _p}} D_{\lambda} &~=~& \cos \theta_{pK}
[\sin (\alpha - \theta_{\rm c.m.} + \theta_{pK})
{{d\sigma}\over{d\Omega _p}} D_{KK} +
\cos (\alpha - \theta_{\rm c.m.} + \theta_{pK})
{{d\sigma}\over{d\Omega _p}}D_{qK}] \nonumber \\
&~-~ & \sin \theta_{pK}
[\sin (\alpha - \theta_{\rm c.m.} + \theta_{pK})
{{d\sigma}\over{d\Omega _p}} D_{Kq} +
\cos (\alpha - \theta_{\rm c.m.} + \theta_{pK})
{{d\sigma}\over{d\Omega _p}} D_{qq}] ~, \nonumber \\
{{d\sigma}\over{d\Omega _p}} D_{\sigma} &~=~& \sin \theta _{pK}
[\sin (\alpha - \theta_{\rm c.m.} + \theta_{pK})
{{d\sigma}\over{d\Omega_p}} D_{KK} +
\cos (\alpha - \theta_{\rm c.m.} + \theta_{pK})
{{d\sigma}\over{d\Omega _p}} D_{qK}] \nonumber \\
&~+~& \cos \theta_{pK}
[\sin (\alpha - \theta_{\rm c.m.} + \theta_{pK})
{{d\sigma}\over{d\Omega _p}} D_{Kq} +
\cos (\alpha - \theta_{\rm c.m.} + \theta_{pK})
{{d\sigma}\over{d\Omega _p}} D_{qq}] ~,
\label{eq:Dsig}\end{aligned}$$ where $\theta_{\rm c.m.}$ is the proton center-of-mass scattering angle, $\theta_{pK}$ is the angle between the incident beam momentum ${\bf p}$ and the average momentum ${\bf K}$ (see Eq. \[eq:coord\]), and $\alpha$ is the spin precession angle defined above. The $D_{\alpha \beta}$ spin transfer coefficients that appear in the previous equation can also be expressed in terms of the spin operator amplitudes as: $$\begin{aligned}
{{d\sigma}\over{d\Omega_p}} D_{KK} &~=~&
\vert A_{n0} \vert^2 - \vert A_{nn} \vert^2 + \vert A_{KK} \vert^2
- \vert A_{Kq} \vert^2 + \vert A_{qK} \vert^2 - \vert A_{qq} \vert^2
~, \nonumber \\
{{d\sigma}\over{d\Omega_p}} D_{Kq} &~=~&
-2[\Im (A_{n0}A^{\ast}_{nn}) -
\Re (A_{KK}A^{\ast}_{Kq} + A_{qK}A^{\ast}_{qq})] ~, \nonumber \\
{{d\sigma}\over{d\Omega_p}} D_{qK} &~=~&
2[\Im (A_{n0}A^{\ast}_{nn}) +
\Re (A_{KK}A^{\ast}_{Kq} + A_{qK}A^{\ast}_{qq})] ~, \nonumber \\
{{d\sigma}\over{d\Omega_p}} D_{qq} &~=~&
\vert A_{n0} \vert^2 - \vert A_{nn} \vert^2 - \vert A_{KK} \vert^2
+ \vert A_{Kq} \vert^2 - \vert A_{qK} \vert^2 + \vert A_{qq} \vert^2 ~.
\label{eq:Dij}\end{aligned}$$
For the coincident $(\vec p,p' \gamma)$ measurements, the incident beam polarization was either pointed along the normal ${\bf \hat n}$ to the scattering plane, or was rotated to lie [*in*]{} the scattering plane. During vertical polarization running, three photon detectors were positioned in the scattering plane on beam right, and one was placed directly above the target [@SPW95]. If a sum over the two beam spin states was performed (effectively producing the results that would be obtained with an unpolarized beam), the coincident yields from the three in-plane detectors could be used to extract the in-plane double-differential cross section, which takes the form [@Pik90] $${{8\pi}\over{3R}} {{d^2 \sigma}\over{d\Omega_{\gamma}d\Omega_p}}
(\theta_{\gamma}) ~=~ A(\theta_p) + B(\theta_p) \cos 2\theta_{\gamma}
+ C(\theta_p) \sin 2\theta_{\gamma} ~,
\label{eq:d2sig}$$ where $R$ is again the $\gamma$-ray branching ratio to the ground state, $\theta_{\gamma}$ is the photon angle in the scattering plane, measured with respect to the ${\bf \hat q}$ direction, and $A$, $B$, and $C$ are unknown functions of the proton scattering angle. These can be written in terms of amplitudes as: $$\begin{aligned}
A &~=~& \vert A_{n0} \vert^2 + \vert A_{nn} \vert^2 + {1 \over 2}
[\vert A_{KK} \vert^2 + \vert A_{Kq} \vert^2 + \vert A_{qK} \vert^2
+ \vert A_{qq} \vert^2] ~, \nonumber \\
B &~=~& {1 \over 2} [\vert A_{KK} \vert^2 + \vert A_{Kq} \vert^2
- \vert A_{qK} \vert^2 - \vert A_{qq} \vert^2] ~, \nonumber \\
C &~=~& -\Re [A_{KK}A^{\ast}_{qK} + A_{Kq}A^{\ast}_{qq}] ~.
\label{eq:ABC}\end{aligned}$$ We point out that in this work our definitions of $A$, $B$, and $C$ do not include the branching ratio normalization factor $8\pi / (3R)$, and are therefore different from the definitions given in Ref. [@Pik90]. A more important comment is that, in order to eliminate any sensitivity to detection efficiency in the photon detectors, only the [*ratios*]{} of the symmetric and antisymmetric pieces of the in-plane coincident cross section, $B/A$ and $C/A$, were used in our determination of the transition amplitude. These quantities have the obvious advantage of being independent of any overall normalization error.
We now turn to the spin-dependent coincident observables which, at a given proton scattering angle $\theta_p$, will be functions of the angle $\theta_\gamma$ at which the photon is emitted. If the photon is emitted in the scattering plane, then the normal-component coincident analyzing power, scaled by the coincident cross section, can be cast in a form similar to that of Eq. \[eq:d2sig\], i.e., $${{8\pi}\over{3R}} {{d^2 \sigma} \over {d\Omega_{\gamma} d\Omega_p}}
(\theta_{\gamma}) A_y (\theta_{\gamma}) ~=~ \epsilon_A (\theta_p ) +
\epsilon_B (\theta_p ) \cos 2\theta_{\gamma} +
\epsilon_C (\theta_p ) \sin 2\theta_{\gamma} ~,$$ where $$\begin{aligned}
\epsilon_A &~=~& 2\Re (A_{n0}A^{\ast}_{nn}) +
\Im (A_{KK}A^{\ast}_{Kq} + A_{qK}A^{\ast}_{qq}) ~, \nonumber \\
\epsilon_B &~=~& \Im (A_{KK}A^{\ast}_{Kq} - A_{qK}A^{\ast}_{qq}) ~,
\nonumber \\
\epsilon_C &~=~& -\Im (A_{KK}A^{\ast}_{qq} - A_{Kq}A^{\ast}_{qK}) ~,\end{aligned}$$ and $d^2 \sigma / d\Omega_\gamma d\Omega_p$ is given by Eqs. \[eq:d2sig\] and \[eq:ABC\]. Note that with three values measured for the normal-component spin asymmetry (corresponding to the three angles of the photon detectors), we have three expressions from which the coefficients $\epsilon_A$, $\epsilon_B$, and $\epsilon_C$ can be determined. Although this inversion is possible in principle, it is more efficient to use the measured asymmetries directly in our fitting procedure to determine the independent $A_{i\mu}$’s, which we describe in the next section.
With a normally-polarized beam, the photon detector directly above the target could measure another piece of the normal-component coincident analyzing power, given by $${{d^2 \sigma} \over {d\Omega_\gamma d\Omega_p}}
({\bf \hat n}) A_y ({\bf \hat n}) ~=~
2\Im (A_{KK}A^{\ast}_{Kq} + A_{qK}A^{\ast}_{qq}) ~,
\label{eq:dAyn}$$ where $${{d^2 \sigma} \over {d\Omega_\gamma d\Omega_p}}
({\bf \hat n}) ~=~ \vert A_{KK} \vert^2 + \vert A_{Kq} \vert^2
+ \vert A_{qK} \vert^2 + \vert A_{qq} \vert^2 ~.
\label{eq:d2sign}$$
It can be shown algebraically that the ratio of Eqs. \[eq:dAyn\] and \[eq:d2sign\] can be expressed in terms of the normal-component [*singles*]{} observables in the form $$A_y ({\bf \hat n}) ~=~ -{{(P-A_y )} \over {(1-D_{N'N})}} ~.
\label{eq:Ayn}$$ In Ref. [@SPW95] this equation was used to demonstrate that the independently measured $(\vec p,p' \gamma)$ coincident and $(\vec
p,\vec p\,')$ singles observables were internally consistent.
Finally, the four remaining $(\vec p,p' \gamma)$ observables were extracted from the spin asymmetries measured when the incident beam polarization had been oriented to lie in the scattering plane. These are components of the longitudinal and sideways analyzing powers, $D_{0L}({\bf \hat k})$ and $D_{0S}({\bf \hat k})$, respectively. These quantities, which vanish identically for singles measurements, are related to the corresponding center-of-mass asymmetries $D_{0K}({\bf
\hat k})$ and $D_{0q}({\bf \hat k})$ via a rotation in the reaction plane through the angle $\theta_{pK}$: $${{D_{0L}({\bf \hat k})} \choose {D_{0S}({\bf \hat k})}} ~=~
{{\cos \theta_{pK} ~~-\sin \theta_{pK}} \choose
{\sin \theta_{pK} ~~~~~\cos \theta_{pK}}}
{{D_{0K}({\bf \hat k})} \choose {D_{0q}({\bf \hat k})}} ~,
\label{eq:D0L}$$ with $D_{0K}({\bf \hat k})$ and $D_{0q}({\bf \hat k})$ defined by $$\begin{aligned}
{{8\pi}\over{3R}} {{d^2 \sigma} \over {d\Omega_p d\Omega_\gamma}}
({\bf \hat k}) D_{0K}({\bf \hat k})
&~=~& 2[\Re (A_{n0}A^{\ast}_{KK}) - \Im (A_{nn}A^{\ast}_{Kq})]
t_{nK} ({\bf \hat k}) \nonumber \\
&~+~& 2[\Re (A_{n0}A^{\ast}_{qK}) - \Im (A_{nn}A^{\ast}_{qq})]
t_{nq} ({\bf \hat k}) ~, \nonumber \\
{{8\pi}\over{3R}} {{d^2 \sigma} \over {d\Omega_p d\Omega_\gamma}}
({\bf \hat k}) D_{0q}({\bf \hat k})
&~=~& 2[\Re (A_{n0}A^{\ast}_{Kq}) + \Im (A_{nn}A^{\ast}_{KK})]
t_{nK} ({\bf \hat k}) \nonumber \\
&~+~& 2[\Re (A_{n0}A^{\ast}_{qq}) + \Im (A_{nn}A^{\ast}_{qK})]
t_{nq} ({\bf \hat k}) ~.
\label{eq:D0K}\end{aligned}$$
The four coefficients of $t_{nK}({\bf \hat k})$ and $t_{nq}({\bf \hat
k})$ that appear in the above expressions (i.e., these four combinations of the $A_{i\mu}$’s) can be viewed as independent observables, and were determined in Ref. [@SPW95] by a fit to the dual sinusoidal dependence of the measured asymmetries on both the outgoing photon direction ${\bf \hat k}$ and the orientation of the incident proton polarization in the scattering plane. Contained in these four observables are clear sensitivities to the relative phases between terms in the transition amplitude corresponding to proton and recoil nuclear polarization projections that are normal to, and oriented in, the reaction plane; relative phases which, by virtue of Eq. \[eq:Dab\], are inaccessible via singles $(\vec p,\vec p\,')$ measurements.
Minimization Procedures
=======================
In the set of combined $(\vec p,\vec p\,')$ and $(\vec p,p' \gamma)$ observables discussed above, we have a total of 16 measured quantities at each proton scattering angle, which can be used to determine the 11 independent quantities required to specify the complete scattering amplitude. Although it has been shown formally [@Ram94] that certain combinations of these observables can provide an analytic solution of this problem (via matrix inversion), this method of analysis has several disadvantages in practice. Because only particular linear combinations of the measured quantities are used, some statistical information is invariably lost in this method. Of more concern is the loss of any statistical gauge of the quality or internal consistency of the amplitude extraction procedure; because one will always obtain an ‘answer,’ the assignment of errors to the $A_{i\mu}$ amplitudes becomes somewhat ambiguous.
In this work, [*all*]{} of the observables measured at each scattering angle are used as input to a single $\chi^2$ minimization procedure. This method not only makes use of the full statistical information contained in the data set, but also minimizes any bias that could be introduced through data manipulation (e.g., forming various linear combinations of the data) prior to the actual fitting. Thus, our approach can be viewed conceptually as the following: we seek values for the six complex amplitudes $A_{i\mu}$ contained in Eq. \[eq:Tpp\] that, when used to form the 16 combinations provided in Section II, minimize the differences between these combinations and the measured values of the corresponding observables. Details of the minimization procedure we followed will be provided in the next few paragraphs, but we first point out a few subtle features unique to this problem. Because the amplitudes are complex, one can seek values for either their magnitudes and phases, or for their real and imaginary components. In this work, we carried out independent fits to both parameter sets, and obtained (as one would hope) equivalent results. However, it became clear that when attempting to resolve various discrete ambiguities in the fitting results, or invoking arguments of ‘smoothness’ (in momentum transfer), use of magnitudes and phases as the fitting parameters was favored. It was also necessary to hold one phase fixed during the fitting, as the observables are insensitive to any uniform shift in all the phases. Because $\vert A_{nn} \vert$ was consistently one of the largest magnitudes over the entire angular range studied, we chose to define $A_{nn}$ to be real and positive, and thus determined (in effect) the phase of every other $A_{i\mu}$ relative to that of $A_{nn}$.
A final choice which required careful thought was the selection of appropriate starting values for the parameters to be fit. To avoid any bias, we noted that the form of the singles cross section, $d\sigma /
d\Omega_p$ (Eq. \[eq:norm\]), reveals that the maximum allowed value for the magnitude of any of the $A_{i\mu}$’s is constrained by $$\vert A_{i\mu} \vert \le \sqrt{d\sigma \over d\Omega_p} ~.$$ Thus, the allowed parameter space for the magnitudes must be restricted to this range. In our fitting, we therefore assigned starting values for each $\vert A_{i\mu} \vert$ by using a linear random number generator to select a number between zero and this maximum. Similarly, starting values for each phase were chosen at random over the allowed range $$\begin{aligned}
0 \le \phi_{i\mu} \le 2\pi ~.\end{aligned}$$ By holding $\phi_{nn}$ at zero, then searching the 11-dimensional parameter space for minima in $\chi^2$, we were able to determine the values for the $A_{i\mu}$ that were most consistent with our entire 16-observable data set.
We now describe the actual fitting procedure in greater detail. We begin by defining $$\chi ^2 ~\equiv~ \sum ^{16}_{j,k=1} [F_j - f_j (A_{i\mu})] \,
W_{jk} \, [F_k - f_k (A_{i\mu})] ~,
\label{eq:chi2}$$ where $F_j$ are the measured values for the 16 observables, and $f_j$ are the expressions given in the previous section for these same observables in terms of the $A_{i\mu}$’s. In this equation, $W_{jk}$ is the weight matrix, given by $$W_{jk} ~=~ [\epsilon ^{-1}]_{jk} ~.$$ Here $\epsilon$ is the full error matrix associated with the set of observables. In the absence of any correlations among the observables, i.e., if each of the observables had been determined independently, $W_{jk}$ would be diagonal, with each element equal to the inverse of the square of the error assigned to each observable, $$W_{jj} ~=~ {1 \over {\delta F_{j}^2}} ~.$$ In this work, however, $W_{jk}$ has been generalized to include known correlations (off-diagonal elements) between specific observables, given their method of determination [@SPW95]. For example, the values obtained for the in-plane coincident cross-section coefficients, $B/A$ and $C/A$, were deduced from the same data set by fitting sinusoidal functions to the measured photon yields. Thus, the resulting coefficients of the fit are highly correlated.
The minimization procedure we employed uses a combination of algorithms [@Bev69] to locate the minima of an arbitrary, nonlinear function in an arbitrarily large parameter space. The $\chi^2$ function (Eq. \[eq:chi2\]) can be linearized around a minimum value $\chi^2_0$ via $$\chi^2 ~=~ \chi _{0}^2 ~+~ \sum_{\mu =1}^{11}
\left. {\partial \chi^2 \over \partial A_\mu} \right\vert_{A_0}
\delta A_\mu ~.
\label{eq:linchi}$$ In this and following equations, we have simplified our notation by denoting all $A_{i\mu}$ with a single index, $A_\mu$, representing either a magnitude or phase. In the above equation, $A_0$ is the [*set*]{} of $A_\mu$ values which minimize $\chi^2$, i.e., their values at $\chi^2 = \chi^2_0$. Thus, our minimization condition can be defined by $${\partial \chi^2 \over \partial A_\nu} ~=~
\left. {\partial \chi^2 \over \partial A_\nu} \right\vert_{A_0} ~+~
\sum_{\mu =1}^{11} \left. {\partial^2 \chi^2 \over \partial A_\nu
\partial A_\mu} \right\vert_{A_0} \delta A_\mu ~=~ 0 ~.
\label{eq:mincond}$$ Once initial values for the $A_\mu$’s are chosen, the parameter search begins by solving for optimal changes in the parameters, $\delta
A_\mu$, then generating new values for these parameters via $$A_\mu ~=~ A^{\rm init}_\mu ~+~ \delta A_\mu ~,$$ from which point the parameter search can begin again. This procedure is repeated until a minimum in $\chi^2$ is found.
The minimization condition can be written as a matrix equation $$\alpha \, \overrightarrow{\delta A} ~=~ \overrightarrow{\beta}$$ or $$\overrightarrow{\delta A} ~=~ \alpha^{-1} \overrightarrow{\beta}$$ where $$\alpha_{\mu \nu} ~\equiv~ {1 \over 2} \left. {\partial^2 \chi^2
\over \partial A_\mu \partial A_\nu} \right\vert_{A_0}$$ and $$\beta_\mu ~=~ -{1 \over 2}
\left. {\partial \chi^2 \over \partial A_\mu} \right\vert_{A_0} ~.$$
The algorithm [@Bev69] used to find the $\chi^2$ minimum utilizes a gradient search in the early stages of the fitting process, which transforms smoothly into a linearization of the fitting function as the fit converges. This is achieved by introducing a parameter $\lambda$ which sets the scale for the size of steps taken along the gradient. To do so, the diagonal elements of the curvature matrix $\alpha$ defined above are modified according to $$\alpha_{\mu \mu} \to \alpha_{\mu \mu}^{\prime} ~=~
(1 + \lambda)\, \alpha_{\mu \mu} ~.$$ Upon inspection of this equation, we note that if $\lambda \gg 1$, then $$\delta A_\mu ~\approx~
{1 \over {\lambda \, \alpha_{\mu \mu}}}\beta_\mu$$ and the search is approximately a gradient search. If, on the other hand, $\lambda \ll 1$, then $$\alpha_{\mu \nu}^{\prime} ~\approx ~\alpha_{\mu \nu}$$ and so the function $\chi^2$ is (approximately) linearized, as described in Eqs. \[eq:linchi\] and \[eq:mincond\]. The algorithm is designed such that the closer $\chi ^2$ gets to a minimum (determined by the size of the changes in the fitting parameters required to ‘step across’ the minimum), the smaller the value of $\lambda$ chosen, and thus a transition from a gradient search to a linearization of $\chi^2$ is effected.
At each scattering angle, a total of 10,000 randomly chosen sets of the $A_\mu$’s were used as starting points to this algorithm. Approximately 95% of the time, the algorithm was successful and converged on a minimum in the multi-dimensional $\chi^2$ surface. It quickly became clear, however, that the $\chi^2$ function describing these observables contained many local minima, and the fit would often become ‘trapped’ in these shallower regions, rather than converging on the ‘true’ minimum value, $\chi^2_{\rm best}$. (This latter quantity was defined as simply the lowest value of $\chi^2_{\rm min}$ determined in any of the 10,000 fits.) As a result, the algorithm was able to locate minima with $\chi^2_{min} \le 2 \, \chi^2 _{\rm best}$ only about 10% of the time, an indication of the complexity of the space being probed. Nevertheless, this yielded a sample of roughly 1000 sets of amplitudes at each angle which gave reasonably good descriptions of all 16 measured observables.
The procedures used to extract final values for each spin-operator amplitude will be presented in the next section. For now, we note that within these 1000 or so acceptable solutions, the [*magnitudes*]{} of the amplitudes, $\vert A_{i\mu} \vert$, were generally found to be very stable, and largely independent of the choice of starting parameters. The values found for the corresponding [*phases*]{}, $\phi_{i\mu}$, on the other hand, were highly dependent on the starting parameters, and thus exhibited much larger fit-to-fit variations. Based on this, a second round of fitting was undertaken, in which the starting parameters were [*not*]{} chosen at random: in this analysis, the magnitudes were initially set to their best-fit values, as determined from the first round of fitting, while the phases were each stepped through all allowed values in a multi-dimensional grid search. Specifically, each phase was assigned a starting value between 0 and $5\pi/3$ in steps of $\pi/3$, with all possible combinations used as starting sets.
As a final check on the robustness of the entire fitting procedure, we used a set of theoretical amplitudes (see Sect. VI) to generate values for each of the observables that had been measured in Ref. [@SPW95], thereby producing a ‘data set’ comparable to that obtained in the actual measurement. Using this ‘data’ as input to our fitting code, we were indeed able to reproduce the input amplitudes, i.e., the algorithm could always find the correct solution. Once this had been established, we performed a more realistic test in which the calculated value of each observable was randomized with a root-mean-square deviation $\sigma$ equal to a typical experimental error bar for that observable. Using this data set, whose statistical precision matched that of the actual measurements, we were again able to reproduce our input amplitudes, within acceptable errors.
Results of the Fitting
======================
After performing 10,000 fits to the data at each scattering angle, our next step was to try to converge on a unique set of solutions for the $A_{i\mu}$’s. At each angle, we first discarded all fits in which the resulting $\chi^2$ minimum, $\chi^2_{\rm min}$, was more than twice the lowest value found, $\chi^2_{\rm best}$. For the number of degrees of freedom in our fitting function, this provided a confidence level of 85% that the true solution was included among the 1000 or so fits kept at each angle [@Bro85]. Among these 1000 fits, though, the solutions tended to cluster tightly around a very small number of regions in parameter space, yielding a set of roughly 5–10 distinct solutions for each angle. As mentioned previously, the amplitudes extracted from these different solutions were often quite similar in magnitude, but showed discrete (and correlated) variations in phase.
To proceed further, i.e., to select from among these few distinct sets of solutions, it was necessary to introduce additional assumptions concerning the angle dependence of the amplitudes. By imposing the constraint that the amplitudes vary slowly and smoothly as a function of momentum transfer (as do all the observables described by these amplitudes), we were able to eliminate most of the remaining ambiguities in the values determined for the $A_{i\mu}$’s. We will illustrate this procedure with examples below. Before invoking such arguments, it is important to note that as the magnitude of a complex quantity goes through a minimum, both the real [*and*]{} the imaginary components must pass close to zero; hence the phase will typically change by roughly 180$^\circ$ as one passes through this minimum. Conversely, it is difficult to produce such a large phase shift [*unless*]{} a magnitude becomes very small.
We now examine in detail the two amplitudes $A_{nn}$ and $A_{KK}$. Shown in Fig. 1 are the fitted magnitudes and the relative phase of these two amplitudes, each plotted versus the center-of-mass scattering angle. (For plotting purposes, all phases lie in the range $-180^\circ \le \phi_{i\mu} \le 180^\circ$.) Different plotting symbols at each angle correspond to different sets of solutions, and have been displaced slightly in angle. The magnitudes of both $A_{nn}$ and $A_{KK}$ are large over the range $5^\circ \le \theta_{\rm c.m.}
\le 13^\circ$ for all valid solutions, and decrease smoothly with angle. The phase difference $(\phi _{nn}-\phi _{KK})$ is close to zero everywhere [*except*]{} for a single solution at $\theta_{\rm c.m.}$ = 8.8$^\circ$, indicated by a daggered ‘X’, which is near $-180^\circ$. If one assumes that neither $A_{nn}$ nor $A_{KK}$ passes near zero in this region, then this phase, and all other fitted values associated with this solution, are almost certainly unphysical. While one could attempt to make this argument more quantitative, e.g., by fitting the magnitudes of $A_{nn}$ and $A_{KK}$ with simple functions that did or did not pass near zero somewhere in this angle range, we feel that even a cursory examination of Fig. 1 tends to rule out the possibility of a zero-crossing. These arguments will become even stronger (though obviously somewhat more model-dependent) when we compare our results to a wide range of realistic theoretical predictions for these quantities, all of which display very smooth variations over this range of momentum transfer.
It is interesting to note that the three quantities presented in Fig. 1 largely determine the observable $D_{0q}$, which in the direction ${\bf \hat k} \cdot {\bf \hat n}$ = ${\bf \hat k} \cdot {\bf
\hat K}$ = $1/\sqrt{2}$, ${\bf \hat k} \cdot {\bf \hat q}=0$ has the value $$D_{0q} \approx \Im (A_{nn}A^{\ast}_{KK}) =
\vert A_{nn}\vert \vert A_{KK} \vert \sin (\phi_{nn} -\phi_{KK})~.$$ Our measured value for this observable, being close to zero, drives the phase difference to either zero or 180$^\circ$, but supplies little additional information to the fit. Constraints on the magnitudes of $A_{nn}$ and $A_{KK}$, and the resolution of this 180$^\circ$ phase ambiguity, must therefore be supplied by other observables, demonstrating again the advantages of a large, diverse data set.
By applying similar arguments to other $A_{i\mu}$’s, we were able to eliminate most of the remaining solution sets, and arrive at a nearly ambiguity-free determination of the magnitude and phase of each amplitude at all angles. For most quantities, the few distinct solutions left were sufficiently close in value that a simple average could be taken, with errors enlarged slightly to reflect the range of solutions included. Exceptions to this behavior occurred only at $\theta_{\rm c.m.} = 16.5^\circ$, where two of the phase differences exhibited two-fold discrete ambiguities that could not be resolved; an interpretation for this will be given in the following section. In all cases, the final values determined for the $A_{i\mu}$ magnitudes and phases were consistent with those found in the fits which yielded the lowest minimum for $\chi^2$, denoted here by $\chi^2_{\rm best}$.
As the final step in our analysis, the values obtained for $\chi^2_{\rm best}$ were normalized to the number of degrees of freedom in the fit, to yield $\chi^2_{\nu}$, a quantity which statistically should lie close to unity. In this work, the values for $\chi^2_{\nu}$ at $\theta_{\rm c.m.}$ = 5.5$^\circ$, 8.8$^\circ$, 12.1$^\circ$, and 16.5$^\circ$ were 0.75, 2.14, 2.17, and 6.47, respectively. Because the fitting function should provide an accurate description of the data (that is, one is [*not*]{} gauging the appropriateness of the model in this case), a minimum $\chi^2_\nu$ in substantial excess of 1 suggests an underestimate of the input error for at least part of the fitted data set. To compensate for this, the error determined in the fitting process for the magnitudes and phases of each of the $A_{i\mu}$’s was artificially increased by a factor of $\sqrt{\chi^2_{\nu}}$. This ensures that when the extracted amplitudes are used to determine best-fit values (with errors) for the observables, one will reproduce the measured input data within one standard deviation, on average. Thus, we believe that the errors quoted here for the magnitudes and phases of the $A_{i\mu}$’s provide a realistic estimate of the true uncertainties inherent in the fits to these data.
Comparison of Amplitudes with Calculations
==========================================
It has been shown previously [@SPW95] that the large number of spin observables measured for this transition provides a severe test for any theoretical model. None of the calculations presented in Ref. [@SPW95] could describe the momentum transfer dependence of all the observables over the entire range covered by the data. To determine more precisely where weaknesses lie in these models, it is useful to bring experiment ‘closer’ to theory by comparing not the measured observables, as was done in Ref. [@SPW95] but by comparing theoretical predictions for the $A_{i\mu}$ amplitudes with values extracted directly from the data, as described here.
In this work, we will consider five different sets of predictions for the scattering amplitude. To carry out these calculations, many details must be specified, such as: the method used to generate the distorting potential, and how well this describes the elastic scattering data; assumptions made regarding the structure of the excited state; the extent to which medium corrections are incorporated into the effective interaction; the handling of exchange contributions to the interaction; etc. Here, we will examine predictions for the $A_{i\mu}$’s derived from the [*same*]{} five models as those to which the [*observables*]{} were compared in Ref. [@SPW95]. Extensive discussion on the content of each calculation is presented in that paper, and the interested reader is encouraged to refer to Ref. [@SPW95] for more detail; only fairly brief descriptions will be provided here.
Of the five calculations to be shown, all use free $t$-matrices for the effective interaction, and Cohen-Kurath matrix elements [@Coh65] to specify the transition to the excited state. Two are relativistic calculations in which both direct and exchange contributions are included explicitly (DREX) [@DREX]. The distorting potential for the incident and outgoing projectile waves, however, is generated either “self-consistently” (the same interaction that induces the $0^+ \rightarrow 1^+$ transition is also folded with the $^{12}$C ground state transition density to produce the distortions), or from an optical potential, with parameters fit to p+$^{12}$C elastic scattering cross section and analyzing power data. In all of the figures, the predictions of these two calculations will be shown as a solid line (self-consistent) and a long-dashed line (optical potential). Two nonrelativistic calculations also include both direct and exchange contributions explicitly (DW81) [@DW81], and also incorporate the effects of distortion using the same two methods, i.e., either self-consistently (dotted line) or from the same optical potential as was used in the relativistic calculation (dot-dashed line). Finally, we will compare our values of the scattering amplitudes to those predicted using a relativistic calculation in which the full interaction has been parameterized in terms of direct scattering processes [*only*]{}, with distorted waves generated self-consistently (short-dashed line). Although exchange processes are expected to contribute significantly to this reaction, this last calculation appeared to describe the [*observables*]{} measured for this reaction better than any model in which exchange contributions had been explicitly included [@SPW95]. This would suggest that most current methods of accounting for exchange processes are inadequate to describe proton-nucleus scattering at intermediate energies, at least for unnatural parity transitions.
Before making detailed comparisons between data and theory, there is one ambiguity which must be pointed out. Because the individual spin-operator amplitudes have dimensions of $\sqrt{\mu {\rm b} / {\rm
sr}^2}$, cross section data [@Com81] were used to provide an overall normalization factor for all spin observables considered in the fitting procedure. This ensures, for example, that Eq. \[eq:norm\] is obeyed, and the sum of the squared amplitudes equals the measured cross section. However, as can be seen in Fig. 2, the five calculations considered here do not reproduce this cross section. In particular, all of the calculations underpredict $d\sigma
/ d\Omega_p$ at small angles, except for the DW81 calculation using distortions generated in an optical model, which overpredicts the data at small angles, then decreases much more rapidly with angle than the data. The DW81 calculation which uses self-consistent distortions, on the other hand, is a much more shallow function of angle than the data suggest.
In light of these discrepancies, one could consider scaling all of the calculated magnitudes, $\vert A_{i\mu} \vert$, by a single multiplicative factor, in order to more closely reproduce the measured cross section. In this way, one is effectively comparing the [*relative*]{} sizes of the spin-operator amplitudes for each model to the measured values. This also ensures that the dominant amplitudes will be well reproduced. On the other hand, the differences observed between the predicted and measured cross sections might result from particular amplitudes (especially the larger ones) being grossly over- or under-predicted, while the calculated values for others are actually in close agreement with the experimentally determined values. In this case, one is best served by [*not*]{} applying an artificial normalization, and directly comparing data and theory for each amplitude. Because our primary goal is to identify more narrowly the weaknesses in individual calculations, we have adopted the latter approach here. The potential drawback, of course, is that a calculation that is correct in all respects other than reproducing the measured cross section will systematically miss [*each*]{} of the extracted amplitudes by roughly the same factor.
With the above caveat in mind, we now compare the predictions of these five models for the magnitude and phase of each amplitude to the values deduced from our fits. The quantity $A_{n0}$, shown in Fig. 3, is the amplitude for polarizing the recoil $^{12}$C nucleus along the ${\bf \hat n}$ direction (perpendicular to the scattering plane) with an unpolarized incident proton. Its magnitude (upper plot) is predicted by each calculation to be quite weak over the entire angle range studied in this work, a feature also seen in the data. The differences among the various calculations are, in absolute terms, very small. Thus, given the level of precision with which this quantity can be determined experimentally, not much useful information can be obtained from $\vert A_{n0} \vert$, other than confirming the small probability for producing this particular spin configuration. Due to the small size of $\vert A_{n0} \vert$, the phase difference $\phi _{n0} - \phi _{nn}$, shown in the lower half of Fig. 3, exhibits a twofold discrete ambiguity at the largest scattering angle. As discussed earlier, if the magnitude of a complex amplitude passes near zero, its phase can change by nearly 180$^\circ$. Our data suggest that $A_{n0}$ passes near zero (has a local minimum) somewhere around $\theta_{\rm c.m.}$ = 16.5$^\circ$, but our measurements can not establish this unambiguously. We also note that at smaller angles the phase difference is relatively flat, in agreement with all of the calculations, although the values deduced from the data are significantly more negative (by $\sim$ 90$^\circ$) than any of the calculations predict.
In contrast to this weak amplitude, we next examine $A_{nn}$, shown in Fig. 4, which is the amplitude for polarizing the recoil nucleus along the ${\bf \hat n}$ direction when the incident proton is also polarized along ${\bf \hat n}$. The data show that $\vert A_{nn}
\vert$ is large throughout the angular range studied, and decreases smoothly with increasing momentum transfer. All five calculations predict this general behavior, but differ significantly in strength, relative to the precision with which this quantity has been determined. It is useful to note the striking similarities between this figure and Fig. 2, the unpolarized cross section. (For a more quantitative comparison, one would need to square the results shown in Fig. 4.) Just as for $d\sigma / d\Omega_p$, the values of $\vert
A_{nn} \vert$ at small momentum transfer are underpredicted slightly by four of the calculations, and overpredicted by the DW81 calculation which uses optical model distortions. The momentum transfer dependence of the data is reasonably well described by the three relativistic calculations, but is too steep (shallow) for the DW81 calculations that use optical model (self-consistent) distortions. The inabilities of the models to reproduce this particular amplitude are thus directly reflected in the discrepancies found between the predictions and measured values for the scattering cross section.
Similar behavior is seen in Fig. 5 for the other large amplitude, $A_{KK}$, the amplitude for polarizing the recoil nucleus along the average momentum direction using an incident proton polarized along the same direction. The measured magnitude $\vert A_{KK} \vert$ (upper plot), like $\vert A_{nn} \vert$, closely follows the angular dependence of the differential cross section, and decreases monotonically with $\theta_{\rm c.m.}$. In this case, the three relativistic calculations [*quantitatively*]{} describe this behavior, and agree with the data at all values of momentum transfer. The two nonrelativistic calculations, on the other hand, do very poorly, either over- or underpredicting the strength of this amplitude at small angles, and predicting angle dependences which are either too steep or too shallow, depending on the method used to generate distortions. Also shown in Fig. 5 is the phase difference $\phi_{nn} - \phi_{KK}$, which is predicted by all of the calculations to be very close to zero over this entire angular range. The data support this idea, albeit with a fairly large statistical uncertainty, suggesting that the two largest amplitudes contributing to this reaction are closely matched in phase.
The third “diagonal” amplitude, $A_{qq}$, has some intriguing properties. $A_{qq}$ is the amplitude for polarizing the recoil $^{12}$C nucleus along the [**q**]{} (momentum transfer) direction, with the proton also polarized along [**q**]{}, i.e., this amplitude is associated with the spin-operator combination $({\bf \Sigma}
\cdot$[**q**]{})(${\bf \sigma} \cdot$[**q**]{}). In a meson-exchange formulation of $NN$ scattering [@Sto93], the one-pion exchange amplitude is associated with the spin-operator $({\bf \sigma _1}
\cdot$[**q**]{})(${\bf \sigma _2} \cdot$[**q**]{}), which is very similar. Perhaps of more significance, in a relativistic PWIA, $A_{qq}$ is the only amplitude that contains contributions from the pseudovector invariant $NN$ amplitude, which carries the same quantum numbers as the pion. Because the excitation of the 15.11 MeV state is a $\Delta T=1$ transition, one-$\pi$ and one-$\rho$ exchange should dominate the reaction process. The potentials for $\pi$ and $\rho$ exchange interfere constructively at very small momentum transfer, but the one-$\pi$ exchange term changes sign relative to one-$\rho$ exchange at relatively low momentum transfer [@Hor94], which can lead to large cancellations. Thus, a zero crossing in the $A_{qq}$ amplitude may be a direct reflection of $\pi - \rho$ interference in the reaction mechanism for this transition.
Inspection of the extracted values for the magnitude and phase of $A_{qq}$, plotted in Fig. 6, clearly indicates that this amplitude does indeed cross zero somewhere between $\theta_{\rm c.m.}$ = 5.5$^\circ$ and 12.1$^\circ$. The magnitude $\vert A_{qq} \vert$, which exhibits a very different $q$ dependence than either $\vert
A_{nn} \vert$ or $\vert A_{KK} \vert$, is very well described by all of the relativistic calculations, which each predict a zero crossing somewhere near $\theta_{\rm c.m.} \sim 9^\circ$. The data are not described quite as well by the nonrelativistic calculation which uses self-consistently generated distortions, and are in strong disagreement with the other DW81 calculation. The large experimental uncertainty in the phase difference at $\theta_{\rm c.m.}$ = 8.8$^\circ$ supports the idea that a zero crossing occurs near this point. The two relativistic calculations which have distortions generated self-consistently describe the momentum transfer dependence of both the magnitude and phase of $A_{qq}$ quite well, and thus best describe the physics contained in this amplitude.
The last two amplitudes, the off-diagonal terms $A_{Kq}$ and $A_{qK}$, are also intriguing. Physically, they represent the amplitudes for polarizing the recoil $^{12}$C nucleus along either the [**K**]{} or [**q**]{} direction when the proton probe is polarized along either the [**q**]{} or [**K**]{} direction, respectively. In a nonrelativistic PWIA, these two amplitudes should be identically zero for this transition, and only become non-zero if non-local effects, such as knock-on exchange, are explicitly included [@Pik85]. In a relativistic formulation, on the other hand, $A_{Kq}$ and $A_{qK}$ include contributions due to linear couplings between the upper and lower components of the bound nucleon wave function, even in PWIA. Because the lower components are momentum-dependent, the nucleon is manifestly off-shell, and non-local effects arise “naturally.” Formally, $A_{Kq}$ and $A_{qK}$ are proportional to the tensor and axial vector pieces of the invariant $NN$ amplitude, respectively. By carrying out the spin algebra [@Pik85], these can be written in terms of the composite spin-convection current amplitudes $\langle {\bf \sigma}
\times {\bf J} \rangle$ and $\langle {\bf \sigma} \cdot {\bf J}
\rangle$, again respectively. Thus, these two amplitudes should be sensitive to the off-shell behavior of the nucleons inside the $^{12}$C nucleus.
In Fig. 7, we see that the extracted magnitude of $A_{Kq}$ is significantly smaller than all of the predicted values, especially at small angles. This suggests a quenching of the tensor component of the $NN$ amplitude within the nucleus, though one can not tell if this is a nuclear structure or an interaction effect. It is very curious that the only relativistic calculation that does [*not*]{} include exchange (short-dashed line) also predicts the largest magnitude for an amplitude that, at least nonrelativistically, is driven largely by exchange contributions. Unfortunately, little insight is provided by the phase difference, $\phi_{nn} - \phi_{Kq}$. Note, however, that if the phase found at the largest angle is simply shifted by 360$^\circ$, the general shape of the angular dependence of the phase difference is followed reasonably well by three of the calculations, albeit with a $\sim 90^\circ$ offset.
In striking contrast, the other off-diagonal amplitude, $A_{qK}$, is predicted to be somewhat smaller than $A_{Kq}$ over most of this angle range, while the experimentally determined values are seen in Fig. 8 to be quite a bit larger. Only the DWIA calculation using optical model distortions is consistent with the typical strength suggested by the data. Figure 8 also shows, though, that the limited statistical precision with which this amplitude has been determined makes it difficult to draw any conclusions about the momentum transfer dependence of this quantity. This limitation is also evident in the phase difference, in which a tight clustering about zero is broken only at the largest angle, where a two-fold discrete ambiguity is observed at $16.5^\circ$. As was the case for $A_{n0}$, our statistical accuracy is such that we cannot determine experimentally whether this amplitude is crossing zero in this angle regime or not. Two calculations predict amplitudes that, in the complex plane, pass near zero on one side ($\Delta \phi = +180^\circ$), two pass on the other side ($-180^\circ$), and one phase remains constant. All of these possibilities are consistent with one of the allowed solutions extracted from the data.
Summary and Conclusions
=======================
We have made a model-independent determination of the full transition amplitude for the $^{12}$C$(p,p')^{12}$C$^{\ast}$ (15.11 MeV, 1$^+$, $T=1$) reaction at an incident beam energy of 200 MeV at four scattering angles. This represents the first such determination for any hadron-induced nuclear transition other than those with $J_i = J_f
= 0$. By imposing only loose “smoothness” constraints on the momentum transfer dependence of the individual spin-operator amplitudes, we have performed a nearly ambiguity-free extraction of these quantities, which has provided deeper insight into the physics driving this transition.
As expected theoretically, the two diagonal spin-operator amplitudes $A_{nn}$ and $A_{KK}$ were found to be the dominant amplitudes over the entire momentum transfer range studied. Each of these was better described by the three calculations carried out in relativistic frameworks than by the two nonrelativistic calculations we considered. The third diagonal amplitude, $A_{qq}$, which has a spin-operator structure similar to that of one-pion exchange, exhibited behavior characteristic of a “zero crossing.” The value of momentum transfer at which this occurred was again much better matched by the relativistic calculations than the DWIA predictions. This crossing has physical significance in that it may reflect interference between $\pi$ and $\rho$ exchange, and hence may serve as a gauge of the relative strengths of these two contributions within the nuclear medium.
The three off-diagonal amplitudes are all much weaker than the three just discussed, and as such were determined with much larger experimental uncertainties. The magnitude of the $A_{n0}$ amplitude provided little insight, though the phase was consistently (i.e., at all angles) off by about $90^\circ$ relative to most of the predicted values. The amplitudes $A_{Kq}$ and $A_{qK}$ are expected to be sensitive to the off-shell behavior of the nucleons inside the $^{12}$C nucleus, and should therefore probe the non-local or exchange nature of the scattering process. Despite the sizable errors on the experimentally determined values for these two amplitudes, neither are described well by any of the five calculations considered here, which may indicate problems in our present treatment of non-local effects in both relativistic and non-relativistic frameworks.
**ACKNOWLEDGEMENTS**
We thank those who participated in the experiment from which the observables discussed here were obtained. We also thank J. Piekarewicz and J. R. Shepard for providing us with the calculations and for useful discussions.
This work was supported in part by the US National Science Foundation under Grant No. PHY-9602872.
Present address: Center for Applied Physics Studies, Louisiana Tech University, Ruston, Louisiana 71272.
See, for example, H. Baghaei [*et al.*]{}, Phys. Rev. Lett. [**69**]{}, 2054 (1992), and references therein.
W. G. Love, A. Klein, and M. A. Franey, in [*Antinucleon- and Nucleon-Nucleus Interactions*]{}, edited by G. E. Walker, C. D. Goodman, and C. Olmer (Plenum, New York, 1985), p. 1.
T. A. Carey, J. M. Moss, S. J. Seestrom-Morris, A. D. Bacher, D. W. Miller, H. Nann, C. Olmer, P. Schwandt, E. J. Stephenson, and W. G. Love, Phys. Rev. Lett. [**49**]{}, 266 (1982).
K. H. Hicks [*et al.,*]{} Phys. Lett. B [**201**]{}, 29 (1988).
W. G. Love and J. R. Comfort, Phys. Rev. C [**29**]{}, 2135 (1984) and references therein.
J. Piekarewicz, R. D. Amado, and D. A. Sparrow, Phys. Rev. C [**32**]{}, 949 (1985).
J. R. Shepard, E. Rost, and J. A. McNeil, Phys. Rev. C [**33**]{}, 634 (1986).
J. M. Moss, Phys. Rev. C [**26**]{}, 727 (1982).
E. Bleszynski, M. Bleszynski, and C. A. Witten, Phys. Rev. C [**26**]{}, 2063 (1982).
J. Piekarewicz, E. Rost, and J. R. Shepard, Phys. Rev. C [**41**]{}, 2277 (1990).
G. Ramachandran, A. R. Usha Devi, and A. Sudha Rao, Phys. Rev. C [**49**]{}, R623 (1994).
S. P. Wells [*et al.,*]{} Phys. Rev. C [**52**]{}, 2559 (1995).
J. R. Comfort, R. E. Segel, G. L. Moake, D. W. Miller, and W. G. Love, Phys. Rev. C [**23**]{}, 1858 (1981).
Philip R. Bevington, [*Data Reduction and Error Analysis for the Physical Sciences*]{}, McGraw-Hill, New York (1969).
I. N. Bronshtein and K. S. Semendyayev, [*Handbook of Mathematics,*]{} Van Nostrand Reinhold Company, New York (1985).
S. Cohen and D. Kurath, Nucl. Phys. [**73**]{}, 1 (1965); T. H. S. Lee and D. Kurath, Phys. Rev. C [**21**]{}, 293 (1980).
E. Rost and J. R. Shepard, computer code [DREX]{}, unpublished; E. Rost and J. R. Shepard, Phys. Rev. C [**35**]{}, 681 (1987).
R. Schaeffer and J. Raynal, Saclay Report No. CEA-R4000, 1970; modifications by J. R. Comfort.
V. G. J. Stoks, R.A.M. Klomp, M.C.M. Rentmeester, and J.J. de Swart, Phys. Rev. C [**46**]{}, 792 (1993).
C. J. Horowitz and J. Piekarewicz, Phys. Rev. C [**50**]{}, 2540 (1994).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Deep Learning (DL) algorithms have become the [*de facto*]{} Machine Learning (ML) algorithm for large scale data analysis. DL algorithms are computationally expensive – even distributed DL implementations which use MPI require days of training (model learning) time on commonly studied datasets. Long running DL applications become susceptible to faults – requiring development of a fault tolerant system infrastructure, in addition to fault tolerant DL algorithms. This raises an important question: [*What is needed from MPI for designing fault tolerant DL implementations?*]{} In this paper, we address this problem for permanent faults. We motivate the need for a fault tolerant MPI specification by an in-depth consideration of recent innovations in DL algorithms and their properties, which drive the need for specific fault tolerance features. We present an in-depth discussion on the suitability of different parallelism types (model, data and hybrid); a need (or lack thereof) for check-pointing of any critical data structures; and most importantly, consideration for several fault tolerance proposals (user-level fault mitigation (ULFM), Reinit) in MPI and their applicability to fault tolerant DL implementations. We leverage a distributed memory implementation of Caffe, currently available under the Machine Learning Toolkit for Extreme Scale (MaTEx). We implement our approaches by extending MaTEx-Caffe for using ULFM-based implementation. Our evaluation using the ImageNet dataset and AlexNet, and GoogLeNet neural network topologies demonstrates the effectiveness of the proposed fault tolerant DL implementation using OpenMPI based ULFM.'
author:
- Vinay Amatya
- Abhinav Vishnu
- Charles Siegel
- Jeff Daily
bibliography:
- 'bibTexts.bib'
title: 'What does fault tolerant Deep Learning need from MPI?'
---
Introduction {#sec:intro}
============
Deep Learning (DL) algorithms are a class of Machine Learning and Data Mining algorithms, which conduct model learning by emulating the computational structure of a mammalian brain. A deep neural network (DNN) – which stores the model of a DL algorithm – contains several [*layers*]{} of [*neurons*]{} inter-connected with [*synapses*]{}. By using deep layers, DL algorithms are able to conduct transformations on highly non-linear data, which is commonplace in many scientific datasets. DL algorithms have shown amazing results in many Computer Vision tasks [@NIPS2012_4824; @43022] and science domains such as High Energy Physics [@Baldi:2014kfa], Climate Modeling [@liu2016application] and Computational Chemistry [@goh2017deep]. DL implementations such as TensorFlow [@tensorflow2015-whitepaper], Caffe [@jia2014caffe], Theano [@bergstra+al:2010-scipy; @Bastien-Theano-2012], and Torch [@Collobert02torch:a] have become available. These implementations are primarily geared towards compute nodes that may contain a multi-core architecture (such as Intel Xeon/KNC/KNL) and/or many-core architectures (GPUs) as commonplace in Leadership Class Facilities (LCFs).
DL algorithms can be applied to a variety of input representations. The tabular input representations typically leverage Multi-layer Perceptrons (MLPs). The images, videos and speech tend to leverage the Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs). The CNNs and RNNs are computationally expensive and typically require significant time for training even on relatively modest data set sizes with modest number of hidden layers. The problem is further exacerbated by the increasing number of layers (such as recently proposed Residual Networks have up to 1000 layers) and ever-increasing volume of data produced by simulations, experiments and hand-held devices. An important solution to these problems is the design and implementation of DL algorithms that are capable of execution on distributed memory systems. Table \[table:newcomparisons\] shows a table of prominent distributed DL implementations.
Name HPC Ready Fault Tolerance
----------------------------------------- ----------- -----------------
FireCaffe [@firecaffe]
S-Caffe [@scaffe]
MaTEx [@matex]
Poseidon [@poseidon]
Petuum [@petuum]
GeePS [@geeps] ?
ProjectAdam [@projectadam]
TensorFlow [@tensorflow2015-whitepaper]
MXNET [@mxnet]
CaffeonSpark [@caffeonspark]
SparkNet [@sparknet]
DogWild [@sparknet] ?
CNTK [@cntk]
Parle [@parle]
PaddlePaddle [@mxnet]
Caffe2 [@goyal:arxiv17]
Proposed FT-Caffe
: []{data-label="table:newcomparisons"}
\
An important artifact of the large scale systems is the increased frequency of faults, which are commonplace in large scale systems [@Schroeder:4775906]. Distributed DL implementations such as distributed TensorFlow, distributed memory implementations of Caffe and even recently proposed Caffe2 [@Caffe2] are primarily geared towards performance. As shown in Table \[table:newcomparisons\], we observe that state of the art HPC ready DL implementations are not fault tolerant. On the other hand, automatic fault tolerance is provided by MapReduce instantiations such as Hadoop, and Spark. However, the implementations are not HPC ready. At the same time, DL implementations are known to take days even on modest size datasets, significantly increasing the probability of observing a fault during the training phase. This raises two important questions: [*1) What are the elements of fault tolerant DL algorithms? and 2) What is needed from MPI for implementing these fault tolerant DL algorithms?*]{}
Contributions
-------------
In this paper, we address these questions and make the following contributions:
- We present the case for several types of parallelism (model, data and hybrid) as motivated from common use-cases and DNN topologies. We use this discussion to derive the suitability of fault tolerance proposals in MPI.
- We consider several design choices for implementing fault tolerant DL implementations. Specifically, we consider checkpoint-restart, Reinit (when a fault occurs, re-initialize the MPI automatically) and user-level fault mitigation.
- We consider several approaches for recovery from faults. We primarily rely on “continued execution” – where the DL implementation continues to execute by using the remaining set of compute nodes.
- We implement our design using MaTEx-Caffe and leverage the ULFM implementation available with OpenMPI. We provide an evaluation of fault tolerant MaTEx-Caffe using the ImageNet-1K dataset and widely studied neural network topologies such as AlexNet and GoogLeNet.
Our evaluation on a 16 node Intel Haswell system connected with InfiniBand indicates that the proposed fault tolerant MaTEx-Caffe is able to scale well and continues execution in the presence of actual permanent node faults. It incurs no observable overhead in the absence of faults, and provides expected performance after recovering from faults, since the overall number of compute nodes are reduced. We also observe that both Reinit [@Reinit] and ULFM [@Bland2013] proposals are suitable for addressing permanent node faults for DL algorithms. However, ULFM is simple – and versatile enough – since it obviates the need for any checkpoint/restart, re-reading of the entire dataset and allows continued execution in the presence of permanent node faults.
The rest of the paper is organized as follows: In section \[sec:background\], we present the background of the proposed research. We present elements of scalable DL algorithms in section \[sec:motivation\] and a solution space for fault tolerant MaTEx-Caffe in section \[sec:design\]. In section \[sec:existing\_proposals\] we present existing features and proposals for fault tolerance in MPI and provide implementation details in section \[sec:implDetail\]. We discuss experimental results in section \[sec:exp\], related work in section \[sec:related\] followed by conclusions in section \[sec:conclusions\].
Background {#sec:background}
==========
In this section, we provide a brief overview of the proposed research. Specifically, we focus on deep neural networks (DNNs).
Deep Neural Networks
--------------------
Symbol Meaning
---------------- ----------------------------------- -- -- -- -- -- -- -- --
$W^{(\ell)}$ weights of layer $\ell$
$b^{(\ell)}$ biases of layer $\ell$
$z^{(\ell+1)}$ $W^{(\ell)}a^{(\ell)}+b^{(\ell)}$
$a^{(\ell)}$ $\sigma(z^{(\ell)})$
$\sigma$ $\mathrm{ReLU}(x)=\max(0, x)$
$n_\ell$ number of layers
: Symbols for Backpropagation.[]{data-label="table:notation"}
\
A dataset is a collection of samples. Samples are often images, speech, text or raw vectors of numbers. ML algorithms typically split a dataset into a [*training set*]{}, used for learning the details of a model; a [*validation set*]{}, used to prevent overfitting and to tune hyper-parameters; and a [*testing set*]{}, used for the accuracy calculation after the final model is trained. Deep neural networks (DNNs) are a class of ML algorithm that learn nonlinear functions by emulating the computational structure of a mammalian brain. It consists of simple computational units called [*neurons*]{} which are connected with [*synapses*]{}.
The values of the synapses, called [*weights*]{} are learned through the [*back-propagation*]{} algorithm. It iteratively updates the weights of the DNN to find a local minimum of an objective/ cost function. With this algorithm, each sample is an input to the [*feed-forward*]{} step. The output is a [*predicted value*]{} which is compared to a [*label*]{}. The difference between the label and predicted value is used to calculate the [*gradients*]{} which are applied to update the weights. This difference is called the [*cost*]{} and the objective of training is to minimize this value on the training set while ensuring that the value on the validation set decreases as well.
Back-propagation is a special case of [*gradient descent*]{}. Any gradient descent variant uses the update rule $$\begin{aligned}
\label{graddesc}
\mathbf{w}'&=&\mathbf{w}+\lambda \nabla_{\mathbf{w}}C\\
\mathbf{b}'&=&\mathbf{b}+\lambda \nabla_{\mathbf{b}}C.\end{aligned}$$ where $\mathbf{w}$ are the weights, $\mathbf{b}$ the biases, $\lambda$ the learning rate, and $C$ is a cost function to be optimized.
We use the notation of Table \[table:notation\] and describe back-propagation in Algorithm \[alg:bp\].
\[alg:bp\] **input:** Data $X\in {\mathbb{R}}^{n\times p}$ and labels $Y\in {\mathbb{R}}^{n\times \ell}$ Compute all $z^{(\ell)}$ and $a^{(\ell)}$. $\delta^{(n_\ell)} = -(y-a^{n_{\ell}})\odot \sigma(z^{(n_\ell)})$ $\delta^{(\ell)}=W^{\ell} \delta^{(\ell+1)}\odot \sigma'(z^{(\ell)})$ $\nabla_{W^{(\ell)}}C = \delta^{(\ell+1)}{a^{(\ell)}}^T$ $\nabla_{b^{(\ell)}}C = \delta^{(\ell+1)}$
Algorithm \[alg:bp\] is most directly applicable to fully-connected neural networks. For structured data, however, convolutional neural networks (CNNs) are more useful. The fundamental unit of computation in a CNN are convolutions – which are stored as arrays of some dimension – unlike vectors in a fully-connected neural network as described above. Each neuron in a convolution layer considers input from a small window (such as `3x3,5x5`) in an image, applies a convolution and computes a value. The computation can be reduced to a matrix-vector multiplication with redundant weights, allowing the above algorithm to be applied.
Elements of Deep Learning Algorithms for Fault Tolerance Consideration {#sec:motivation}
======================================================================
In this section, we present the motivation of our work. Specifically, we consider the properties of DL algorithms, distinguishing between MLPs, CNNs and RNNs in terms of their expected execution on large scale systems. This distinction provides the necessary guidelines for requirements from MPI in terms of fault tolerance. As pointed out by Gropp and Lusk [@gropp:ijhpca02], “fault tolerance is a property of MPI programs and specification”. Hence, it is critical to consider these in conjunction. Our first element of discussion is the expected type of parallelism for scaling out DL algorithms.
Master-Slave Paradigm
---------------------
Over the last few years, several researchers have considered the possibility of scaling out DL algorithms [@NIPS2012_0598; @geeps; @projectadam; @mxnet; @caffeonspark]. The classical work in scaling out DL algorithms considered a [*master-slave*]{} paradigm, which was proposed under the DistBelief framework [@NIPS2012_0598]. It considered a hierarchical organization of [*parameter servers*]{} which would hold the latest copy of the model. The [*workers*]{} would periodically update the master with their updates and request the latest copy of the model. Several extensions to this fundamental paradigm have been proposed in the literature [@geeps; @mxnet; @firecaffe]. The limitations of the master-slave model have been well-studied in the distributed systems research [@das:arxiv16; @lian:nip15]. In addition to being a single point of failure, and a communication bottleneck, the limitation of this approach is that the convergence of master-slave paradigm worsens at scale-out. For extreme scale systems, this approach is infeasible. Hence, we disregard this approach which would be leveraged in practical deployments especially of HPC systems such as Leadership Class Facilities (LCFs).
It is also worthwhile noting that this approach is amenable to fault tolerance, especially if the reliability of the parameter server is higher than workers. A possible implementation in the master-slave paradigm is either re-spawning of new workers and splitting the original training set among these new workers (by using `MPI_Comm_spawn`) or continue executing in the presence of faults using the remaining set of compute nodes. Other researchers have made similar observations in the context of generic master-slave applications [@gropp:ijhpca02] and they are readily applicable to DL algorithms. However, due to the fundamental scaling issues of the master-slave paradigm for DL implementations, we disregard this approach from implementation.
Model Parallelism
-----------------
Another possibility which has been presented in literature for scaling out DL implementations is [*model parallelism*]{}. In this specific type, individual layers of the overall DNN model are split among different compute nodes. The training set itself is split among the compute nodes as well. Let us consider the example of the AlexNet neural network topology as shown in Figure \[fig:alexnet\_topology\]. In a sample execution of model parallelism, each of the hidden layers is resident on a single compute node.
![A pictorial representation of AlexNet [@imagenet] neural network topology[]{data-label="fig:alexnet_topology"}](./alexnet_topology.pdf){width="0.7\columnwidth"}
During the feedforward step, a batch of samples is executed on the first hidden layer. The output of the first hidden layer – which is typically referred to as activations – is forwarded to the next hidden layer, resulting in point-to-point communication between two compute nodes. This procedure is repeated untill the last layer of the DNN is reached, at which point the [*error*]{} is calculated. During the back-propagation step, the error is used to calculate the updates to the weights ([*gradients*]{}) which are communicated between compute nodes in the reverse order to the feedforward step.
Data Parallelism
----------------
A widely used option in scaling out DL implementations is [*data parallelism*]{} [@zheng:icpads16; @siegel:bigdata16; @siegel2016adaptive; @vishnu:cluster15-a; @vishnu:cluster15-b; @vishnu:ipdps16; @shohdy:icpp16; @shohdy:hipc16; @zheng:icpads16]. Under this type of parallelism, the model is replicated and the data is split among multiple compute nodes. A pictorial representation of the data parallelism is shown in the Figure \[fig:data\_parallelism\].
![A pictorial representation of data parallelism in DL algorithms using AlexNet neural network topology and four compute nodes. The model is synchronized at the end of each batch using `MPI_Allreduce` and other primitives such as NVIDIA Collective Communication Library (NCCL)[]{data-label="fig:data_parallelism"}](./data_parallelism.pdf){width="0.7\columnwidth"}
As shown in the figure, at the end of each batch each compute node (assuming that the implementation uses shared address space programming model such as OpenMP/pthread on a node) executes an `MPI_Allreduce`. By executing the all-to-all reduction primitive, the algorithm ensures that it is equivalent to the default sequential DL algorithm [@Keuper:MHLPC2015]. An important consideration for data parallelism is that strong scaling of work is essential to ensure the equivalence of the implementation to the sequential algorithm. Specifically, let us consider a batch size $b$, and let $n$ be the number of compute nodes. The overall expected complexity of the data parallelism based implementation is $\Theta\left(\frac{b}{n} + \log(n)\right)$. Naturally, the ratio of communication to computation increases with strong scaling – which is a potential downside to data parallelism. Several solutions have been proposed to handle this situation [@Krizhevsky14oneweird]. One possibility is to consider increasing the batch size and increasing the values of other hyper-parameters (such as learning rate) by Krizhevsky [@Krizhevsky14oneweird]. Recently proposed solutions such as S-Caffe [@scaffe] improve the scalability of data parallelism by leveraging the overlap of communication with computation. While a few of these approaches provide strict equivalence to the sequential algorithm, other approaches such as asynchronous variants (also referred to as asynchronous gradient descent (AGD)) are still useful, but do not provide strict equivalence to the default stochastic gradient descent (SGD) algorithm.
Scalability Analysis
--------------------
An advantage of model parallelism is its potential to scale-out the DL algorithms very well. For example, if there are 1,000 layers, then in an ideal situation each compute node may have one layer, resulting in scale-out. However, there are several reasons to not consider pure model parallelism based techniques for scaling out the DL implementations.
Method Symbol
--- ------------------------------------ ----------------- -- --
1 Features in previous layer $n_1$
2 Features in current layer $n_2$
3 Activation shape in previous layer $x_1\times x_2$
4 Window size in current layer $w_1\times w_2$
5 Strides for current layer $s_1\times s_2$
: Symbols Used For Computing Activations and Parameters[]{data-label="table:activations"}
\
For any DNN with several hidden layers, let us consider two consecutive layers: $L_1$ and $L_2$. We compute the number of parameters and activations for $L_2$ as follows: If $L_2$ is a convolution layer, then the array containing the parameters is $w_1\times w_2\times n_1\times n_2$. However, the activations are an array of size $\frac{x_1}{s_1}\times \frac{x_2}{s_2}\times n_2$. As $w_i<\frac{x_i}{s_i}$, the number of parameters is less than the number of activations. Conversely, if $L_2$ is a fully-connected layer, then there are $n_1\times n_2$ parameters and only $n_2$ activations. We note that if $L_1$ is a convolution layer, then $n_1$ must be replaced by the total number of activations of that layer, namely $\frac{x_1}{s_1}\times \frac{x_2}{s_2}\times n_1$. This implies, in general, convolutional networks have lesser communication volume of weights than activations. However, for fully-connected networks the trend is reversed. In the case of AlexNet (shown in Figure \[fig:alexnet\_topology\]), a well-established DNN architecture, this can be seen directly. The first convolutional layer contains 34,848 parameters, but has 301,056 activations, a difference of an order of magnitude. The final fully-connected layer, however, has 4,096,000 parameters and 1,000 activations.
Lessons Learned
---------------
In the previous section, we provided an in-depth discussion on the possibilities of scaling out DL implementations. While model parallelism looks attractive, in practice the ratio of communication of activations to model parameters prohibits effective scaling. This is because CNNs are increasingly becoming commonplace – including the winners of last 5 years of ImageNet classification challenge [@imagenet]. At the same time, data parallelism provides scaling out possibilities, but has limitations regarding the growth of batch size [@Keuper:MLHPC2016]. However, solutions proposed by other researchers [@Krizhevsky14oneweird; @scaffe] still make data parallelism an amenable choice for scaling out DL implementations. It is worth noting that a possibility which combines DL model and data parallelism – hybrid parallelism – has been proposed in literature as well [@das:arxiv16]. However, usually model parallelism is applied on multiple GPUs/multiple sockets on each compute node and data parallelism is applied for multiple compute nodes. In essence, we already consider hybrid parallelism, while implicitly leveraging model parallelism within a node. [**Hence, we consider data parallelism for fault tolerance considerations in rest of the paper.**]{}
Solution Space {#sec:design}
==============
In this section, we present a solution space for designing fault tolerant DL implementations using data parallelism, as discussed in the previous section. An important consideration is exploring the suitability of existing features in MPI for this purpose, with detailed considerations for the primary proposals.
Critical Data Structures in DL Algorithms
-----------------------------------------
The first step is the identification of critical data structures in DL algorithms. Specifically, there are several data structures which are used during the feed-forward and the back-propagation phase of the DL implementations. During the feed-forward step, the input dataset and the model are used – both of which are read-only during the step. However, at the back-propagation step, the model weights are updated while the input dataset remains untouched. Hence, the critical data structure for DL implementations is the model weights which are updated iteratively till convergence. It is worth noting that there are auxiliary data structures which are updated as well. As an example for DL implementations with [*momentum*]{}, data structures such as [*history*]{} are updated, which can be recalculated from the model weights. The gradients – iterative updates to the model weights – are calculated during the back-propagation step. However, they are accumulated iteratively in the model weights, and hence they are not critical.
Process Recovery Model
----------------------
An important design point is the process recovery model. Several scientific applications such as LULESH [@IPDPS13:LULESH] and NAS Parallel Benchmarks [@Bailey:SC91] typically require a fixed topology (such as a square/quadratic) in terms of number of MPI processes. However, there are no such requirements for DL implementations. Hence, it is possible to continue execution with the remaining set of compute nodes, without affecting the correctness of the DL implementation. A natural advantage of this approach is that it requires little support from process managers for practical deployments. Hence, we use this approach for designing fault tolerant DL implementation.
Suitability of Existing MPI Features and Proposals {#sec:existing_proposals}
==================================================
Suitability of Existing Error Handling in MPI
---------------------------------------------
An important design consideration is the suitability of existing error handling in MPI for designing fault tolerant DL implementations. Specifically, by initializing with `MPI_ERRORS_RETURN`, it is possible for a DL implementation to prevent aborting when a fault is detected either during MPI communication or an external system software component (such as SLURM).
By setting an explicit error handler, it is possible for the DL implementation to checkpoint their critical data structures, exit the application and re-start the application from the recent saved checkpoint. In this specific case, the critical data structure is the model parameters of the DNN, since the dataset is read-only, and it can be readily recovered from the disk/file-system. The DL implementation may be re-started using $n$ compute nodes (if spare compute nodes are available) or $n - 1$ compute nodes, since DL implementations do not have specific requirements of a topology. This approach is definitely a suitable possibility. However, it may not be necessary, since this will result in a recovery complexity of $O(n)$, a function of number of compute nodes, instead of the degree of failure. The reasoning behind this time complexity is due to the fact that the $n - 1$ compute nodes would need to read the entire dataset back from the disk (prohibitive data movement), in addition to the checkpointed model files. Hence, this approach may be considered as the baseline approach, but not necessarily as the optimal approach for handling permanent node faults in DL implementations.
Suitability of User-Level Fault Mitigation (ULFM) Proposal
----------------------------------------------------------
One of the fault tolerance proposal which has been considered for inclusion in the MPI specification for the last few years is ULFM. The salient features of ULFM are: 1) ability to provide non-collective global fault notification, 2) ability to recover from faults by fixing the [*broken*]{} communicator on the fly and 3) support for fixed/shrinking process set.
ULFM is particularly suited for applications which have small process-specific state information. Usually, resetting the global state information is non-trivial and requires writing a complex error handler. Naturally, for large-scale applications – which have been developed over decades – writing a correct error handler even for a subset of fault cases is non-trivial.
However, with data parallelism the overall state information that is required for DL implementations is minimal. Since the model is replicated across the compute nodes, the DL implementation requires no checkpointing. Hence, ULFM is potentially the right fit for implementing fault tolerant DL algorithms.
Suitability of Reinit Fault Tolerance Proposal
----------------------------------------------
Recently, Laguna [*et al.*]{} have proposed [*Reinit*]{} proposal for handling faults in MPI. The objective of Reinit is to address the limitations of ULFM, and is particularly suitable to applications where the code complexity of the recovery module is high. The salient features of the Reinit proposal are: 1) automatic re-initialization of MPI after a fault is detected, 2) semi-automatic recovery from the intermediate checkpoints, and 3) ability to handle shrinking/fixed process set.
We consider the suitability of the Reinit proposal to data parallelism based DL implementations. We observe that DL implementations would be required to check-point the model weights periodically, which would be used by the Reinit implementation during recovery. We also observe that the application would be re-started requiring the entire dataset to be read from the disk.
Lessons Learned
---------------
We observe that existing local fault notification in the MPI specification and implementations may be used for developing fault tolerant DL implementations. However, there is a significant amount of work needed within MPI and at the application level (such as intermediate checkpointing) to leverage the existing functionality. The Reinit proposal is suitable as well. However, there are two potential downsides that are readily observed: 1) Reinit would require DL implementations to consider intermediate check-pointing, when the DL algorithm does not mandate it, and 2) Reinit would require application to read the entire dataset from the disk, when reading the data could be fairly localized to the degree of failure.
ULFM has positive attributes which are definitely suitable for designing fault tolerant DL implementations. The primary functionality that is required is automatic fixing of the communicator, and reading the partial dataset from the disk, while continuing to execute with the existing set of compute nodes. The ULFM specification itself has a few implementation caveats, including the cost of fault detection (which is relatively lesser for Reinit), cost of global notification and cost of recovering the communicator. Yet, the overall cost of computation recovery is at most one batch – while in the case of Reinit, it is expected to be much higher depending upon the degree of checkpointing. Hence, we consider ULFM for implementing fault tolerant DL algorithms. In the next section, we present the implementation details for ULFM based DL algorithm.
Design and Implementation Details {#sec:implDetail}
=================================
In this section, we present design and implementation details for fault tolerant DL algorithms. We leverage the ULFM implementation provided by OpenMPI for this purpose and implement our changes in Caffe runtime. Figure \[fig:code\_snippet\] shows the overall interaction between different components.
![Overall Caffe architecture and code flow for the original code. We require changes to dataset reading and the callback for implementing fault tolerance.[]{data-label="fig:code_snippet"}](./code_snippet.pdf){width="0.8\columnwidth"}
As shown in figure, the Caffe architecture has layers such as for MLPs, CNNs and RNNs, which are defined in a prototxt file. Hence, these are already resident on disk. Our extensions of Caffe – also referred to as MaTEx-Caffe for rest of the paper – support parallel NetCDF format which requires changes for fault tolerance. Caffe runtime supports several types of solvers such as SGD, Adam and others. These solvers use a common substrate for data parallelism, where `ForwardBackward()` step computes the forward and back-propagation steps of the overall implementation. The resulting gradients are then synchronized using a callback, which is extended by MaTEx-Caffe for using `MPI_Allreduce`. The resulting synchronized gradients are then applied using the `ApplyGradients` function before the next batch of samples are ingested.
Pseudo-code Walk-through
------------------------
Figure \[fig:ulfm\_code\] shows the difference between the original non-fault-tolerant implementation of the callback `on_gradients_ready` and the fault tolerant version on the right. Figure \[fig:code\_flow\] presents the code changes for data readback in the fault tolerant version and compares it to the original code.
![Code snippet for ULFM based fault tolerant DL implementation[]{data-label="fig:ulfm_code"}](./ulfm_code.pdf){width="\columnwidth"}
### Original Callback
The original call-back receives the gradients from the `Forwardbackward` function and uses an all-to-all reduction to synchronize the gradients across all compute nodes. The resulting gradient is divided by the number of compute nodes and applied to the local model using `ApplyGradients` function.
### Callback with ULFM Changes
As shown in Figure \[fig:ulfm\_code\], the ULFM changes are handful and primarily restricted to a single callback. Specifically, when a fault is detected, we leveraget the `MPIX_Comm_shrink` function is used to shrink the communicator from original to the new communicator. Once the communicator is reset, then all to all reduction is retried till return code is `MPI_SUCCESS`.
### Data Readback for Fault Tolerant Version
Figure \[fig:code\_flow\] shows the original code flow and the code flow for fault tolerant DL. The data is read only when a fault is detected by the `on_gradients_ready`.
![Code flow for original and fault tolerant implementations for data readback[]{data-label="fig:code_flow"}](./code_flow.pdf){width="\columnwidth"}
Performance Evaluation {#sec:exp}
======================
In this section, we present a detailed performance evaluation of the proposed fault tolerant approach using ULFM. Table \[table:arch\] shows the hardware and software details of our systems. Table \[table:heur\_desc\] shows a description of proposed approaches that we have implemented. Table \[table:datasets\] provides an overview of datasets, and the associated neural networks.
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Name CPU (\#cores) GPU Baseline Caffe Network MPI cuDNN CUDA Nodes \#cores
---------------------------------------------------------------------------------------------------------- --------------- ----- ---------------- --------- ----- ------- ------ ------- ---------
**[PUMA]{} & Haswell (20) & N/A & Intel-Caffe [@caffe-intel] & IB & ULFM-OpenMPI & N/A & N/A & 16 & 320\
**
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
\
Name Type Implemented Description of DL Algorithm and Implementation
--------------- -------------------------------------------- ------------- -----------------------------------------------------------------------
SGD (Default) Stochastic Gradient Descent Yes Implements strong scaling by dividing batch and all-to-all reduction.
FT-SGD Fault Tolerant Stochastic Gradient Descent Yes Implements strong scaling by dividing batch and all-to-all reduction.
\
[max width=]{}
Dataset Model Description Training Samples Validation Samples Image Size Classes
------------------------------------------ ------------------------------ -------------------- ------------------ -------------------- ----------------------- --------- --
MNIST [@mnistlecun] LeNet-3 [@lecun1998gradient] Handwritten Digits 60000 10000 $28 \times28$ 10
CIFAR-10 [@Krizhevsky09learningmultiple] Caffe-default Small Images 60000 10000 $32 \times32\times3$ 10
ImageNet [@ILSVRC15] AlexNet [@NIPS2012_4824] Diverse Images 1,281,167 50,000 $256\times256\times3$ 1,000
ImageNet [@ILSVRC15] GoogLeNet Diverse Images 1,281,167 50,000 $256\times256\times3$ 1,000
Objective
---------
The objective of our performance evaluation is to understand the performance overhead of using ULFM based implementation and correctness implications (if any) of the existing ULFM implementations.
Fault Injection Methodology
---------------------------
To emulate the process faults, we insert a fault in a process by using `SIGKILL`. For controlled experiment the fault is injected after 300 batches. This effectively emulates compute node faults, since we use one process on a compute node and multiple threads for each process.
Correctness Analysis
--------------------
For understanding the correctness, we compare the [*loss curves*]{} – a measure of the error as observed during the training phase. The curves are compared for SGD and FT-SGD implementations in Figures \[fig:alexnet\_loss\_n4\], \[fig:alexnet\_loss\_n8\], and \[fig:alexnet\_loss\_n16\] using 4, 8, and 16 compute nodes respectively. The FT-SGD evaluation consists of exactly one process fault – which is usually the case in real scenarios. We observe that the loss curves for both SGD and FT-SGD implementation track each other roughly. However, the FT-SGD is behind since with reduced number of available compute nodes, the overall batch size is reduced as well. The effect is diminished on 16 compute nodes since the overall effect of losing one compute node is reduced. Similar for other datasets as shown in Figures \[fig:googlenet\_loss\_n16\], \[fig:cifar10\_loss\_n16\] and \[fig:mnist\_loss\_n16\], the convergence of FT-SGD and SGD is similar.
Performance Analysis
--------------------
Figures \[fig:alexnet\_run\_time\], \[fig:googlenet\_run\_time\], \[fig:cifar10\_run\_time\] and \[fig:mnist\_run\_time\] shows the performance comparisons of SGD and FT-SGD using AlexNet, GoogLeNet, CifarNet (CIFAR10) and LeNet-3 (MNIST). The overall evaluation uses 1024 batches, which is a relatively small number of batches in comparison to the standard number of batches such as 60K for AlexNet. For the FT-SGD evaluation, exactly one process executes `SIGKILL` after 300 batch updates, resulting in $n - 1$ remaining number of compute nodes. Since the batch on each compute node remains constant, the overall computation time as observed on each compute node is similar for FT-SGD and SGD. Also, since we have one compute node failure, the overall difference in communication time is also negligible.
Figures \[fig:alexnet\_dataload\_time\], \[fig:googlenet\_dataload\_time\], \[fig:cifar10\_dataload\_time\] and \[fig:mnist\_dataload\_time\] shows the comparison of reading the overall dataset and partial dataset when a fault occurs for AlexNet, GoogLeNet, and other datasets. These charts are specifically useful to understand the cost of reading the dataset as a function of number of compute nodes. We observe that reading the partial dataset is significantly faster than reading the entire dataset, which is not surprising. This is especially validated for ImageNet dataset which is much larger than MNIST and CIFAR10 datasets. We observe that for MNIST and CIFAR10, partial reading is actually slower, since these datasets are trivially small.
Figures \[fig:alexnet\_shrink\_time\], \[fig:googlenet\_shrink\_time\], \[fig:cifar10\_shrink\_time\] and \[fig:mnist\_shrink\_time\] shows the overhead of cumulative `MPI_Comm_shrink` as a function of number of compute nodes. These functions are executed only if a fault is detected, otherwise this code is not executed. We observe that the overhead increases with the number of compute nodes, which is expected. However, the overall time is relatively insignificant to the batch update time. Hence, the ULFM specification and the ULFM implementation are sufficient for providing functionality and performance in implementing DL algorithms.
\
\
Related Work {#sec:related}
============
\[checkpoint\_restart\]
The majority of fault tolerant solutions proposed in the literature and practice have focused on checkpoint-restart mechanisms. Under these solutions, the applications periodically save the state of the data and computation either explicitly or implicitly by using OS level approaches such as Berkeley Lab Checkpoint Restart (BLCR) or virtualization based approaches. As an example, [@Hursey:4228333; @Ma:5289172], primarily focus on checkpoint-restart method for fault tolerance, storing the checkpointed data into the filesystem, while [@Gamell:InMemory:7576485] have presented disk-less in-memory checkpointing storage-restart scheme at application level leveraging ULFM. Others [@Ouyang:CRFS:6047205] have focused on alleviating filesystem I/O bottleneck due to checkpointing, using other libraries for checkpointing; [@Kutlu:6507503] explores an algorithm based fault tolerance for data intensive algorithms with data replication techniques. Others [@Walters:4633353] have explored asynchronous decentralized replication with standard checkpoint restart techniques. Wang [*et al.*]{} [@Wang:Hybrid:5695644] discusses hybrid checkpointing, alternating between incremental checkpointing and full checkpointing, resulting in minimized checkpointed data size. All these techniques have checkpoint-restart as the fundamental method for providing fault tolerance. Guo [*et al.*]{} [@Guo:SC2015] have discussed detect-resume model for MapReduce, using MPI, in addition to Reinit model. With Detect-Resume model, the workload from the faulted process is redistributed to the remaining nodes. In their approach, the lost work is recomputed from scratch in the remaining processes, leading to a longer recovery time. However in our approach, for DL algorithms, there is no need to recompute the work from the lost process(es), hence recovery time is greatly reduced. Chakravorty [*et al.*]{} in their work [@Chakravorty:IHPC06] have explored the concept of predictive fault tolerance. While this approach introduces lesser overhead compared to the checkpoint-restart, the suitability of predictive approach is limited.
Conclusions {#sec:conclusions}
===========
In this paper, we have addressed the question of the requirements of MPI for designing fault tolerant Deep Learning (DL) algorithms. We have presented the case for several types of parallelism as motivated from common use-cases, and DNN topologies. We have used the discussion to derive the suitability of fault tolerance proposals in MPI. We have considered several design choices for implementing fault tolerant DL implementations. Specifically, we have considered checkpoint-restart, Reinit (when a fault occurs, re-initialize the MPI automatically) and user-level fault mitigation (ULFM). We have implemented our design using MaTEx-Caffe and leveraged ULFM implementation available from OpenMPI. We have provided an evaluation of fault tolerant MaTEx-Caffe using ImageNet-1K dataset and widely studied neural network topologies and datasets such as AlexNet, GoogLeNet on ImageNet dataset, and MNIST and CIFAR-10 datasets as well. Our evaluation has indicated the effectiveness of ULFM both in terms of its suitability as a specification and readiness for practical deployments.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- |
$^a$, Pedro Bicudo$^b$, Marco Cardoso$^b$, Antje Peters$^a$, Marc Wagner$^a$\
\
$^a$Goethe-Universität Frankfurt am Main, Institut für Theoretische Physik,\
Max-von-Laue-Stra[ß]{}e 1, D-60438 Frankfurt am Main, Germany\
$^b$CeFEMA, Instituto Superior Técnico, Universidade de Lisboa, Av. Rovisco Pais,\
1049-001 Lisboa, Portugal\
\
E-mail: , , , ,
title: '$\bar{b}\bar{b}ud$ tetraquark resonances in the Born-Oppenheimer approximation using lattice QCD potentials'
---
Introduction
============
A challenging and modern problem in particle physics and QCD is to improve our understanding of exotic hadrons. A possible approach to study heavy-heavy-light-light four-quark systems and the existence of tetraquarks is to compute potentials of two static antiquarks $\bar{Q}\bar{Q}$ in the presence of two light quarks $qq$ and to use these potentials in the Schrödinger equation to search for bound states (cf. e.g. [@Detmold:2007wk; @Wagner:2010ad; @Bali:2010xa; @Wagner:2011ev; @Bicudo:2012qt; @Brown:2012tm; @Bicudo:2015vta; @Bicudo:2015kna; @Bicudo:2016ooe]). In this way a stable $\bar{b}\bar{b}ud$ tetraquark with quantum numbers $I(J^P) = 0(1^+)$ has been predicted [@Bicudo:2012qt; @Brown:2012tm]. Recently it has been confirmed by lattice computations using $\bar{b}$ quarks of finite mass treated with Non Relativistic QCD [@Francis:2016hui; @Junnarkar:2018twb]. In this work, we extend our investigation of the $\bar{b}\bar{b}ud$ four-quark system by exploring the existence of tetraquark resonances. To this end we use the emergent wave method from scattering theory [@Bicudo:2015bra].
For a more detailed discussion of this work cf. [@Bicudo:2017szl].
\[sec:latticePotentials\]Lattice QCD potentials of two static antiquarks $\bar{Q}\bar{Q}$ in the presence of two light quarks $qq$
==================================================================================================================================
In previous studies we have computed potentials $V(r)$ of two static antiquarks $\bar{Q} \bar{Q}$ in the presence of two light quarks $q q$ with lattice QCD. Computations have been performed for different light quark flavor combinations $q q$ with $q \in \{ u, d, s, c \}$. Moreover, different parity and total angular momentum sectors have been studied (cf. e.g. [@Bicudo:2015vta; @Bicudo:2015kna]). There are both attractive as well as repulsive potentials. Of particular interest with respect to the existence of tetraquarks are two of the attractive potentials with $q \in \{ u , d \}$. The corresponding quantum numbers are $(I=0,j=0)$ and $(I=1,j=1)$, where $I$ denotes isospin and $j$ the total angular momentum of the light quarks and gluons around the $\bar{b}\bar{b}$ separation axis. The two potentials are shown in Figure \[fig:potentials\] for lattice spacing $a \approx 0.079 \, \textrm{fm}$ and $u$ and $d$ quark masses corresponding to a pion mass $m_\pi \approx 340 \, \textrm{MeV}$.
 $(I=0,j=0)$ potential. (right) $(I=1,j=1)$ potential.](potential_scalar_light0.pdf "fig:"){width="45.00000%"}  $(I=0,j=0)$ potential. (right) $(I=1,j=1)$ potential.](potential_vector_light0.pdf "fig:"){width="45.00000%"}
The existence or non-existence of a stable tetraquark and its binding energy depends on the light quark mass $q$ [@Bicudo:2015vta]. Thus, we have performed computations of $\bar{Q} \bar{Q}$ potentials for three different light $u$ and $d$ quark masses corresponding to $m_\pi \in \{ 340 \, \textrm{MeV} , 480 \, \textrm{MeV} , 650 \, \textrm{MeV} \} $. The results, which can be parameterized by a screened Coulomb potential $$\label{eq:potential} V(r) = -\frac{\alpha}{r} e^{-r^2 / d^2} ,$$ have been extrapolated to $ m_\pi= 140 \, \textrm{MeV}$ [@Bicudo:2015kna]. The parameterization (\[eq:potential\]) is motivated by one-gluon exchange for small $\bar{Q} \bar{Q}$ separations $r$ and the formation of two $B$ mesons at larger $r$ as a consequence of color screening as sketched in Figure \[fig:screening\]. Even though this approach is phenomenologically motivated, it is fully consistent with our lattice QCD results, i.e. the corresponding fits yield small $\chi^2 / \textrm{dof}$. The numerical values of the parameters $\alpha$ and $d$ are collected for both potentials in Table \[tab:parameters\]. Clearly, the $(I = 0,j = 0)$ potential is more attractive than the $(I = 1,j = 1)$ potential.
 At small separations the static antiquarks $\bar{Q} \bar{Q}$ interact by perturbative one-gluon exchange. (b) At large separations the light quarks $q q$ screen the interaction and the four quarks form two rather weakly interacting $B$ mesons.](strings_screened_ab.pdf){width="0.75\columnwidth"}
---------- ---------- ----------------------------- -----------------------------
$I$ $j$ $\alpha$ $d$ in $\textrm{fm}$
$\ 0 \ $ $\ 0 \ $ $\ 0.34^{+0.03}_{-0.03} \ $ $\ 0.45^{+0.12}_{-0.10} \ $
$1$ $1$ $0.29^{+0.05}_{-0.06}$ $0.16^{+0.05}_{-0.02}$
---------- ---------- ----------------------------- -----------------------------
: \[tab:parameters\]Parameters $\alpha$ and $d$ of the parameterization (\[eq:potential\]) of the two attactive $\bar{Q} \bar{Q}$ potentials with $(I = 0,j = 0)$ and $(I = 1,j = 1)$.
The potential parameterization (\[eq:potential\]) with $\alpha$ and $d$ from Table \[tab:parameters\] can be inserted into the Schrödinger equation, i.e. they can be used to explore the existence of stable tetraquarks or tetraquark resonances in the Born-Oppenheimer approximation. A stable $\bar{b} \bar{b} u d$ tetraquark with quantum numbers $I(J^P) = 0(1^+)$ around $60 \, \textrm{MeV}$ below the $B B^\ast$ threshold has been predicted in [@Bicudo:2012qt; @Bicudo:2015vta; @Bicudo:2015kna; @Bicudo:2016ooe]. The search for $\bar{b} \bar{b} u d$ tetraquark resonances is discussed in [@Bicudo:2017szl] and the following sections.
\[sec:emergent\_wave\]The emergent wave method
==============================================
In this section we discuss the emergent wave method, which allows to extract phase shifts and resonance parameters. More details can be found e.g. in [@Bicudo:2015bra].
We start by considering the Schrödinger equation $$\label{eq:schro} \Big(H_{0} + V(r)\Big) \Psi = E \Psi$$ and by splitting the wave function into two parts, $$\label{eq:sep_psi} \Psi = \Psi_{0} + X .$$ $\Psi_{0}$ is the incident wave, which is a solution of the free Schrödinger equation, i.e. $H_{0} \Psi_{0} = E \Psi_{0}$, and $X$ denotes the emergent wave. Inserting eq. (\[eq:sep\_psi\]) into eq. (\[eq:schro\]) and using the free Schrödinger equation we obtain $$\label{eq:schro_scatter} \Big(H_{0} + V(r) - E\Big) X = -V(r) \Psi_{0} .$$ Solving this equation for given energy $E$ provides the emergent wave $X$. The asymptotic behavior of $X$ determines the phase shifts. We find the poles of the $\mbox{S}$ matrix and the $\mbox{T}$ matrix in the complex energy plane and identify them with resonances, when located in the second Riemann sheet at $\mathcal{E} - i \Gamma/2$, where $\mathcal{E}$ is the energy and $\Gamma$ is the decay width of the resonance.
Partial wave decomposition
--------------------------
The Hamiltonian describing the two heavy antiquarks $\bar{b} \bar{b}$ is $$\label{EQN005} H = H_0 + V(r) = -\frac{\hbar^{2}}{2 \mu} \triangle + V(r) ,$$ where $\mu = M/2$ is the reduced mass and $M = 5 \, 280 \, \textrm{MeV}$ is the mass of the $B$ meson from the PDG [@Tanabashi:2018oca]. One can express the incident plane wave $\Psi_{0} = e^{i \mathbf{k} \cdot \mathbf{r}}$ as a sum over spherical waves, $$\label{eq:expansionsphericalbessel} \Psi_{0} = e^{i \mathbf{k} \cdot \mathbf{r}} = \sum_{l} (2l+1) i^{l} j_{l}(k r) P_{l}(\hat{\mathbf{k}} \cdot \hat{\mathbf{r}}) ,$$ where $j_{l}$ are spherical Bessel functions, $P_{l}$ are Legendre polynomials and the relation between energy and momentum is $\hbar k = \sqrt{2 \mu E}$. Since the potential $V(r)$ is spherically symmetric, we can also expand the emergent wave $X$ in terms of Legendre polynomials $P_{l}$, $$\label{eq:001} X = \sum_{l} (2l+1) i^{l} \frac{\chi_l(r)}{k r} P_{l}(\hat{\mathbf{k}} \cdot \hat{\mathbf{r}}) .$$ Inserting eq. (\[eq:expansionsphericalbessel\]) and eq. (\[eq:001\]) into eq. (\[eq:schro\_scatter\]) leads to a set of ordinary differential equations for $\chi_l$, $$\label{eq:1cl0:radial} \bigg(-\frac{\hbar^2}{2 \mu} \frac{d^{2}}{dr^{2}} + \frac{l (l+1)}{2 \mu r^{2}} + V(r) - E\bigg) \chi_l(r) = -V(r) k r j_l(k r) .$$
Solving the differential equations for the emergent wave
--------------------------------------------------------
$V(r)$, eq. (\[eq:potential\]), is exponentially screened, i.e. $V(r) \approx 0$ for $r \geq R$, where $R \gg d$. Consequently, the emergent wave is a superposition of outgoing spherical waves for large separations $r \geq R$ and can be expressed by spherical Hankel functions of the first kind $h_l^{(1)}$, $$\label{eq:002} \frac{\chi_l(r)}{k r} = i t_l h_l^{(1)}(k r) .$$ To compute the complex prefactors $t_l$, which will lead to the phase shifts, we solve the ordinary differential equation (\[eq:1cl0:radial\]) using the following boundary conditions:
- At $r = 0$: $\chi_l(r) \propto r^{l+1}$.
- For $r \geq R$: eq. (\[eq:002\]).
We emphasize that the boundary condition for $r \geq R$ depends on $t_l$. Solving the differential equation for a given value of the energy $E$, this boundary condition is only fulfilled for a specific value of $t_l$. In other words the boundary condition for $r \geq R$ fixes $t_l$ as a function of $E$.
To solve eq. (\[eq:1cl0:radial\]) numerically, we have implemented two different approaches:
- A fine uniform discretization of the interval $[0,R]$ reducing the differential equation to a large set of linear equations, which can be solved rather efficiently, since the corresponding matrix is tridiagonal.
- A standard 4-th order Runge-Kutta shooting method.
Phase shifts, $\mbox{S}$ and $\mbox{T}$ matrix poles
----------------------------------------------------
$t_l$ is an eigenvalue of the $\mbox{T}$ matrix (see standard textbooks on quantum mechanics and scattering, e.g. [@Merzbacher]). From $t_l$ we can determine the phase shift $\delta_l$ and also the corresponding $\mbox{S}$ matrix eigenvalue $$\label{eq:003} s_l \equiv 1 + 2 i t_l = e^{2 i \delta_l}$$ (at large distances $r \geq R$ the radial wave function is $k r ( j_l (kr) + i t_l h_l^{(1)}(k r)) = (k r /2) (h_l ^{(2)}(kr) + e^{2 i \delta_l} h_l^{(1)}(k r))$). Note that both the $\mbox{S}$ matrix and the $\mbox{T}$ matrix are analytical in the complex plane and are also defined for complex energies $E$. Thus, we solve the differential equation (\[eq:1cl0:radial\]) for complex $E$ and find the $\mbox{S}$ and $\mbox{T}$ matrix poles by scanning the complex energy plane $(\textrm{Re}(E) , \textrm{Im}(E))$ and by applying Newton’s method to find the roots of $1 / t_l(E)$. These poles correspond to complex resonance energies $E = \mathcal{E} - i \Gamma/2$ and must be located in the second Riemann sheet with a negative imaginary part of $E$.
\[sec:results\]Results for phase shifts, $\mbox{S}$ and $\mbox{T}$ matrix poles and prediction of resonances
============================================================================================================
Phase shifts $\delta_l$
-----------------------
We consider the more attractive $\bar{b} \bar{b}$ potential with $(I=0,j=0)$ (cf. section \[sec:latticePotentials\]), compute $t_l$ for real energies $E$ and apply eq. (\[eq:003\]) to determine phase shifts $\delta_l$ for orbital angular momenta $l=0,1,2,3,4$. A clear indication for a resonance would be a strongly increasing $\delta_{{\textit{\textrm{l}}}}$ from $0$ to almost $\pi$. Such a behavior is, however, not observed (cf. Figure \[fig:phase\_shifts\] (left)). Thus, we have to check more thoroughly, whether there are resonances or not.
 Phase shift $\delta_l$ as a function of the energy $E$ for orbital angular momenta $l=0,1,2,3,4$ for the $(I=0,j=0)$ potential ($\alpha = 0.34$, $d = 0.45 \, \textrm{fm}$). (right) Phase shift $\delta_1$ as a function of the energy $E$ for different $\alpha$ and fixed $d = 0.45 \, \textrm{fm}$.](phase_shifts.pdf "fig:"){width="48.00000%"}  Phase shift $\delta_l$ as a function of the energy $E$ for orbital angular momenta $l=0,1,2,3,4$ for the $(I=0,j=0)$ potential ($\alpha = 0.34$, $d = 0.45 \, \textrm{fm}$). (right) Phase shift $\delta_1$ as a function of the energy $E$ for different $\alpha$ and fixed $d = 0.45 \, \textrm{fm}$.](delta_wp.pdf "fig:"){width="48.00000%"}\
It is also interesting to consider the $l=1$ channel for even more attractive potentials by increasing the parameter $\alpha$, while $d$ is fixed. We show the resulting phase shifts $\delta_1$ in Figure \[fig:phase\_shifts\] (right). For $\alpha \gtrsim 0.65$ resonances are clearly indicated. For $\alpha \gtrsim 0.72$ there are even bounds states, i.e. the phase shifts start at $\pi$ and decrease monotonically. However, this observation does not allow to make a clear statement, whether there is a resonance for $\alpha = 0.34$.
Resonances as poles of the $\mbox{S}$ and $\mbox{T}$ matrices for complex energies $E$
--------------------------------------------------------------------------------------
Now we search for poles of the $\mbox{T}$ matrix eigenvalue $t_l$ in the complex energy plane, which indicate resonances. For orbital angular momentum $l=1$ and the $(I=0,j=0)$ potential we find a pole, which is shown in Figure \[fig:complexPlane\] (left), where $t_1$ is plotted as a function of the complex energy $E$. For a better understanding of the resonance and its dependence on the potential we determine the pole of $t_1$ for various parameters $\alpha$. In Figure \[fig:complexPlane\] (right) we show the location of the pole for several values of $\alpha$ in the $(\textrm{Re}(E),\textrm{Im}(E))$ plane. Indeed, starting at $\alpha=0.21$ we find poles. Consequently, we can predict a resonance at $\alpha =0.34$. For orbital angular momenta $l > 1$ as well as for the less attractive potential $(I=1,j=1)$ no poles have been found.
: $\mbox{T}$ matrix eigenvalue $t_1$ as a function of the complex energy $E$ for the $(I=0,j=0)$ potential ($\alpha = 0.34$, $d = 0.45 \, \textrm{fm}$). Along the vertical axis we show the norm $|t_1|$, while the phase $\textrm{arg}(t_1)$ is visualized by different colors. (right) Trajectory of the pole of the $\mbox{T}$ matrix eigenvalue $t_1$ in the complex energy plane $(\textrm{Re}(E),\textrm{Im}(E))$ corresponding to a variation of the parameter $\alpha$. The cloud of blue points represents the systematic error of our prediction.](complex_plane.pdf "fig:"){width="48.00000%"} : $\mbox{T}$ matrix eigenvalue $t_1$ as a function of the complex energy $E$ for the $(I=0,j=0)$ potential ($\alpha = 0.34$, $d = 0.45 \, \textrm{fm}$). Along the vertical axis we show the norm $|t_1|$, while the phase $\textrm{arg}(t_1)$ is visualized by different colors. (right) Trajectory of the pole of the $\mbox{T}$ matrix eigenvalue $t_1$ in the complex energy plane $(\textrm{Re}(E),\textrm{Im}(E))$ corresponding to a variation of the parameter $\alpha$. The cloud of blue points represents the systematic error of our prediction.](traj_wp.pdf "fig:"){width="48.00000%"}
Analysis of statistical and systematic errors
---------------------------------------------
We perform a detailed statistical and systematic error analysis for the pole of $t_1$ in the complex energy plane $(\textrm{Re}(E),\textrm{Im}(E))$ using the same method as for our study of bound states [@Bicudo:2015vta]. We parametrize the lattice QCD data for the potential $V^{\textrm{lat}}(r)$ with an uncorrelated $\chi^2$ minimizing fit using the ansatz (\[eq:potential\]), i.e. we minimize the expression $$\label{eq:chisquared} \chi^2 = \sum_{r_\textrm{min} \leq r \leq r_\textrm{max}} \bigg(\frac{V(r)-V^{\textrm{lat}}(r)}{\Delta V^{\textrm{lat}}(r)}\bigg)^2$$ with respect to $\alpha$ and $d$, where $\Delta V^{\textrm{lat}}(r)$ denotes the corresponding statistical errors. To estimate the systematic error, we perform fits for various fit ranges $r_\textrm{min} \leq r \leq r_\textrm{max}$. Additionally, we vary the range of the temporal separation $t_\textrm{min} \leq t \leq t_\textrm{max}$, where $V^{\textrm{lat}}(r)$ is read off. For each fit we determine the pole of $t_1$, i.e. the resonance energy $\mathcal{E}$ and the decay width $\Gamma$. As systematic error we take the spread of these results, while the statistical error is determined via the jackknife method. Applying this combined systematic and statistical error analysis, we find a resonance energy $\mathcal{E} = \textrm{Re}(E) = 17^{+4}_{-4} \, \textrm{MeV}$ above the $B B$ threshold and a decay width $\Gamma = -2 \textrm{Im}(E) = 112^{+90}_{-103} \, \textrm{MeV}$. Studying the symmetries of the quarks with respect to color, flavor, spin and their spatial wave function and considering the Pauli principle we determine the quantum numbers as $I(J^P) = 0(1^-)$. The mass of this $\bar{b}\bar{b}ud$ tetraquark resonance is given by $m = 2 M + \textrm{Re}(E) = 10 \, 576^{+4}_{-4} \, \textrm{MeV}$.
Conclusion
==========
We have explored the existence of $\bar{b}\bar{b}ud$ tetraquark resonances applying lattice QCD potentials for two static antiquarks in the presence of two light quarks, the Born-Oppenheimer approximation and the emergent wave method. We predict a new resonance with quantum numbers $I(J^P)=0(1^-)$, a resonance mass $\textrm{Re}(E) = 17^{+4}_{-4} \, \textrm{MeV}$ and a decay width $\Gamma = 112^{+90}_{-103} \, \textrm{MeV}$.
Acknowledgements {#acknowledgements .unnumbered}
================
We acknowledge useful conversations with K. Cichy.
P.B. acknowledges the support of CeFEMA (grant FCT UID/CTM/04540/2013) and is thankful for hospitality at the Institute of Theoretical Physics of Goethe-University Frankfurt am Main. M.C. acknowledges the support of CeFEMA and the FCT contract SFRH/BPD/73140/2010. M.W. acknowledges support by the Emmy Noether Programme of the DFG (German Research Foundation), grant WA 3000/1-1.
This work was supported in part by the Helmholtz International Center for FAIR within the framework of the LOEWE program launched by the State of Hesse.
Calculations on the LOEWE-CSC and on the on the FUCHS-CSC high-performance computer of the Frankfurt University were conducted for this research. We would like to thank HPC-Hessen, funded by the State Ministry of Higher Education, Research and the Arts, for programming advice.
[99]{}
W. Detmold, K. Orginos and M. J. Savage, “$BB$ Potentials in Quenched Lattice QCD,” Phys. Rev. D [**76**]{}, 114503 (2007) \[hep-lat/0703009 \[hep-lat\]\]. M. Wagner \[ETM Collaboration\], “Forces between static-light mesons,” PoS LATTICE [**2010**]{}, 162 (2010) \[arXiv:1008.1538 \[hep-lat\]\]. G. Bali [*et al.*]{} \[QCDSF Collaboration\], “Static-light meson-meson potentials,” PoS LATTICE [**2010**]{}, 142 (2010) \[arXiv:1011.0571 \[hep-lat\]\]. M. Wagner \[ETM Collaboration\], “Static-static-light-light tetraquarks in lattice QCD,” Acta Phys. Polon. Supp. [**4**]{}, 747 (2011) \[arXiv:1103.5147 \[hep-lat\]\]. P. Bicudo [*et al.*]{} \[European Twisted Mass Collaboration\], “Lattice QCD signal for a bottom-bottom tetraquark,” Phys. Rev. D [**87**]{}, 114511 (2013) \[arXiv:1209.6274 \[hep-ph\]\]. Z. S. Brown and K. Orginos, “Tetraquark bound states in the heavy-light heavy-light system,” Phys. Rev. D [**86**]{}, 114506 (2012) \[arXiv:1210.1953 \[hep-lat\]\]. P. Bicudo, K. Cichy, A. Peters, B. Wagenbach and M. Wagner, “Evidence for the existence of $u d \bar{b} \bar{b}$ and the non-existence of $s s \bar{b} \bar{b}$ and $c c \bar{b} \bar{b}$ tetraquarks from lattice QCD,” Phys. Rev. D [**92**]{}, 014507 (2015) \[arXiv:1505.00613 \[hep-lat\]\]. P. Bicudo, K. Cichy, A. Peters and M. Wagner, “$B B$ interactions with static bottom quarks from Lattice QCD,” Phys. Rev. D [**93**]{}, 034501 (2016) \[arXiv:1510.03441 \[hep-lat\]\]. P. Bicudo, J. Scheunert and M. Wagner, “Including heavy spin effects in the prediction of a $\bar{b} \bar{b} u d$ tetraquark with lattice QCD potentials,” Phys. Rev. D [**95**]{}, 034502 (2017) \[arXiv:1612.02758 \[hep-lat\]\]. A. Francis, R. J. Hudspith, R. Lewis and K. Maltman, “Lattice prediction for deeply bound doubly heavy tetraquarks,” Phys. Rev. Lett. [**118**]{}, 142001 (2017) \[arXiv:1607.05214 \[hep-lat\]\]. P. Junnarkar, N. Mathur and M. Padmanath, “Study of doubly heavy tetraquarks in Lattice QCD,” arXiv:1810.12285 \[hep-lat\]. P. Bicudo and M. Cardoso, “Tetraquark bound states and resonances in the unitary and microscopic triple string flip-flop quark model, the light-light-antiheavy-antiheavy $q q \bar Q\bar Q$ case study,” Phys. Rev. D [**94**]{}, 094032 (2016) \[arXiv:1509.04943 \[hep-ph\]\]. P. Bicudo, M. Cardoso, A. Peters, M. Pflaumer and M. Wagner, “$u d \bar{b} \bar{b}$ tetraquark resonances with lattice QCD potentials and the Born-Oppenheimer approximation,” Phys. Rev. D [**96**]{}, 054510 (2017) \[arXiv:1704.02383 \[hep-lat\]\]. M. Tanabashi [*et al.*]{} \[Particle Data Group\], “Review of particle physics,” Phys. Rev. D [**98**]{}, 030001 (2018). E. Merzbacher, “Quantum mechanics (3rd edition),” Wiley (1998).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We show that the output of a (residual) with an appropriate prior over the weights and biases is a in the limit of infinitely many convolutional filters, extending similar results for dense networks. For a , the equivalent kernel can be computed exactly and, unlike “deep kernels”, has very few parameters: only the hyperparameters of the original . Further, we show that this kernel has two properties that allow it to be computed efficiently; the cost of evaluating the kernel for a pair of images is similar to a single forward pass through the original with only one filter per layer. The kernel equivalent to a 32-layer ResNet obtains 0.84% classification error on MNIST, a new record for with a comparable number of parameters. [^1]'
author:
- |
Adrià Garriga-Alonso\
department of Engineering\
University of Cambridge\
`[email protected]` Carl Edward Rasmussen\
Department of Engineering\
University of Cambridge\
`[email protected]` Laurence Aitchison\
Department of Engineering\
University of Cambridge\
`[email protected]`
bibliography:
- 'already-read.bib'
title: |
Deep Convolutional Networks as\
shallow Gaussian Processes
---
Introduction
============
Convolutional Neural Networks () have powerful pattern-recognition capabilities that have recently given dramatic improvements in important tasks such as image classification [@krizhevsky2012imagenet]. However, as are increasingly being applied in real-world, safety-critical domains, their vulnerability to adversarial examples [@szegedy2013intriguing; @kurakin2016adversarial], and their poor uncertainty estimates are becoming increasingly problematic. Bayesian inference is a theoretically principled and demonstrably successful [@snoek2012practical; @deisenroth2011pilco] framework for learning in the face of uncertainty, which may also help to address the problems of adversarial examples [@gal2018idealised]. Unfortunately, Bayesian inference in is extremely difficult due to the very large number of parameters, requiring highly approximate factorised variational approximations [@blundell2015weight; @gal2015dropout], or requiring the storage [@lakshminarayanan2017simple] of large numbers of posterior samples [@welling2011bayesian; @mandt2017stochastic].
Other methods such as those based on are more amenable to Bayesian inference, allowing us to compute the posterior uncertainty exactly [@rasmussen2006gaussian]. This raises the question of whether it might be possible to combine the pattern-recognition capabilities of with the exact probabilistic computations in . Two such approaches exist in the literature. First, deep convolutional kernels [@wilson2016deep] parameterise a kernel using the weights and biases of a , which is used to embed the input images into some latent space before computing their similarity. The parameters of the resulting kernel then are optimised by gradient descent. However, the large number of kernel parameters in the reintroduces the risk of overconfidence and overfitting. To avoid this risk, we need to infer a posterior over the kernel parameters, which is as difficult as directly inferring a posterior over the parameters of the original . Second, it is possible to define a convolutional [@markvdw2017convolutional] or a deep convolutional [@kumar2018deep] by defining a that takes an image patch as input, and using that as a component in a larger -like system. However, inference in such systems is very computationally expensive, at least without the use of potentially severe variational approximations [@markvdw2017convolutional].
An alternative approach is suggested by the underlying connection between Bayesian Neural Networks () and . In particular, @neal1996bayesian showed that the function defined by a single-layer fully-connected with infinitely many hidden units, and random independent zero-mean weights and biases is equivalent to a , implying that we can do exact Bayesian inference in such a by working with the equivalent . Recently, this result was extended to arbitrarily deep fully-connected with infinitely many hidden units in each layer [@lee2017deep; @gpbehaviour]. However, these fully-connected networks are rarely used in practice, as they are unable to exploit important properties of images such as translational invariance, raising the question of whether state-of-the-art architectures such as [@lecun1990handwritten] and ResNets [@he2016deep] have equivalent representations. Here, we answer in the affirmative, giving the kernel corresponding to arbitrarily deep and to (convolutional) residual neural networks [@he2016deep]. In this case, if each hidden layer has an infinite number of convolutional *filters*, the network prior is equivalent to a .
Furthermore, we show that two properties of the kernel induced by a allow it to be computed very efficiently. First, in previous work it was necessary to compute the covariance matrix for the output of a single convolutional filter applied at all possible locations within a single image [@markvdw2017convolutional], which was prohibitively computationally expensive. In contrast, under our prior, the downstream weights are independent with zero-mean, which decorrelates the contribution from each location, and implies that it is necessary only to track the patch variances, and not their covariances. Second, while it is still necessary to compute the variance of the output of a convolutional filter applied at all locations within the image, the specific structure of the kernel induced by the means that the variance at every location can be computed simultaneously and efficiently as a convolution.
Finally, we empirically demonstrate the performance increase coming from adding translation-invariant structure to the prior. Without computing any gradients, and without augmenting the training set (e.g.using translations), we obtain 0.84% error rate on the MNIST classification benchmark, setting a new record for nonparametric GP-based methods.
behaviour in a
================
For clarity of exposition, we will treat the case of a 2D convolutional . The result applies straightforwardly to $n$D convolutions, dilated convolutions and upconvolutions (“deconvolutions”), since they can be represented as linear transformations with tied coefficients (see Fig. \[fig:conv-as-linear\]).
A 2D convolutional network prior\[sec:define-convnet\]
------------------------------------------------------
The network takes an arbitrary input image ${{\bf {X}}}$ of height $H{^{(0)}}$ and width $D{^{(0)}}$, as a $C{^{(0)}} \times (H{^{(0)}}
D{^{(0)}})$ real matrix. Each row, which we denote ${{\bf {x}}}_1,{{\bf {x}}}_2,\dotsc,{{\bf {x}}}_{C{^{(0)}}}$, corresponds to a channel of the image (e.g. $C{^{(0)}} = 3$ for RGB), flattened to form a vector. The first activations ${{\bf {A}}}{^{(1)}}({{\bf {X}}})$ are a linear transformation of the inputs. For $i \in \{1,\dotsc,C{^{(1)}}\}$: $${{\bf {a}}}_i{^{(1)}}({{\bf {X}}}) {:=}b_i{^{(1)}} {\bf 1} + \sum_{j=1}^{C{^{(0)}}} {{\bf {W}}}{^{(1)}}_{i,j} {{\bf {x}}}_j\;.
\label{eq:network-base}$$
We consider a network with $L$ hidden layers. The other activations of the network, from ${{\bf {A}}}{^{(2)}}({{\bf {X}}})$ up to ${{\bf {A}}}{^{(L+1)}}({{\bf {X}}})$, are defined recursively: $${{\bf {a}}}_i{^{(\ell+1)}}({{\bf {X}}}) := b_i{^{(\ell+1)}} {\bf 1} + \sum_{j=1}^{C{^{(\ell)}}} {{\bf {W}}}_{i,j}{^{(\ell+1)}} \phi\left( {{\bf {a}}}_j{^{(\ell)}}({{\bf {X}}}) \right).
\label{eq:network-recursive}$$
The activations ${{\bf {A}}}{^{(\ell)}}({{\bf {X}}})$ are $C{^{(\ell)}} \times (H{^{(\ell)}}
D{^{(\ell)}})$ matrices. Each row ${{\bf {a}}}_i{^{(\ell+1)}}$ represents the flattened $j$th channel of the image that results from applying a convolutional filter to $\phi({{\bf {A}}}{^{(\ell)}}({{\bf {X}}}))$.
The structure of the pseudo-weight matrices ${{\bf {W}}}{^{(\ell+1)}}_{i,j}$ and biases $b_i{^{(\ell+1)}}$, for $i\in \{1,\dotsc,C{^{(\ell+1)}}\}$ and $j \in \{1,\dots,C{^{(\ell)}}\}$, depends on the architecture. For a convolutional layer, each row of ${{\bf {W}}}{^{(\ell+1)}}_{i,j}$ represents a *position* of the filter, such that the dot product of all the rows with the image vector ${{\bf {x}}}_j$ represents applying the convolutional filter ${{\bf {U}}}{^{(\ell+1)}}_{i,j}$ to the $j$th channel. Thus, the elements of each row of ${{\bf {W}}}{^{(\ell+1)}}_{i,j}$ are: 0 where the filter does not apply, and the corresponding element of ${{\bf {U}}}_{i,j}{^{(\ell+1)}}$ where it does, as illustrated in Fig. \[fig:conv-as-linear\]. The outputs of the network are the last activations, ${{\bf {A}}}{^{(L+1)}}({{\bf {X}}})$. In the classification or regression setting, the outputs are not spatially extended, so we have $H{^{(L+1)}} = D{^{(L+1)}} = 1$, which is equivalent to a fully-connected output layer. In this case, the pseudo-weights ${{\bf {W}}}{^{(L+1)}}_{i,j}$ only have one row, and the activations ${{\bf {a}}}_i{^{(L+1)}}$ are single-element vectors.
Finally, we define the prior distribution over functions by making the filters ${{\bf {U}}}_{i,j}{^{(\ell)}}$ and biases $b_i{^{(\ell)}}$ be independent Gaussian . For each layer $\ell$, channels $i,j$ and locations within the filter ${x},{y}$: $$\begin{aligned}
\label{eq:prior}
U{^{(\ell)}}_{i,j,{x},{y}} &\sim {\mathcal{N}\left(0, \sigma_\text{w}^2/C{^{(\ell)}}\right)}, & b{^{(\ell)}}_i &\sim {\mathcal{N}\left(0, \sigma_\text{b}^2\right)}.\end{aligned}$$ Note that, to keep the activation variance constant, the weight variance is divided by the number of input channels. The weight variance can also be divided by the number of elements of the filter, which makes it equivalent to the weight initialisation scheme introduced by @he2016deep.
Argument for behaviour\[sec:argument-gp\]
-----------------------------------------
We follow the proofs by @lee2017deep and @gpbehaviour to show that the output of the described in the previous section, ${{\bf {A}}}{^{(L+1)}}$, defines a indexed by the inputs, ${{\bf {X}}}$. Their proof [@lee2017deep] proceeds by applying the multivariate to each layer in sequence, i.e. taking the limit as $N{^{(1)}} \rightarrow \infty$, then $N{^{(2)}} \rightarrow \infty$ etc, where $N{^{(\ell)}}$ is the number of hidden units in layer $\ell$. By analogy, we sequentially apply the multivariate CLT by taking the limit as the number of channels goes to infinity, i.e. $C{^{(1)}} \rightarrow \infty$, then $C{^{(2)}} \rightarrow \infty$ etc. While this is the simplest approach to taking the limits, other potentially more realistic approaches also exist [@gpbehaviour]. In Appendix \[sec:actual-proof\] we take the latter approach.
The fundamental quantity we consider is a vector formed by concatenating the feature maps (or equivalently channels), ${{\bf {a}}}_j{^{(\ell)}}({{\bf {X}}})$ and ${{\bf {a}}}_j{^{(\ell)}}({{\bf {X}}}')$ from data points ${{\bf {X}}}$ and ${{\bf {X}}}'$, $$\label{eq:vapair}
{{\bf {a}}}_i{^{(\ell)}}({{\bf {X}}}, {{\bf {X}}}') =
\begin{pmatrix} {{\bf {a}}}_i{^{(\ell)}}({{\bf {X}}})\phantom{'} \\ {{\bf {a}}}_i{^{(\ell)}}({{\bf {X}}}') \end{pmatrix}.$$ This quantity (and the following arguments) can all be extended to the case of finitely many input points.
#### Induction base case.
For any pair of data points, ${{\bf {X}}}$ and ${{\bf {X}}}'$ the feature-maps corresponding to the $j$th channel, ${{\bf {a}}}_j{^{(1)}}({{\bf {X}}}, {{\bf {X}}}')$ have a multivariate Gaussian joint distribution. This is because each element is a linear combination of shared Gaussian random variables: the biases, ${{\bf {b}}}_j{^{(0)}}$ and the filters, ${{\bf {U}}}{^{(0)}}_{j,:}$. Following Eq. , $$\begin{aligned}
{{\bf {a}}}_i^{(1)}({{\bf {X}}}, {{\bf {X}}}')
&= b_i^{(1)} {\bf 1}
+ \sum_{i=1}^{C{^{(0)}}}
\begin{pmatrix} {{\bf {W}}}_{i,j}{^{(1)}} & {{\bf {0}}}\\ {{\bf {0}}}& {{\bf {W}}}_{i,j}{^{(1)}} \end{pmatrix}
\begin{pmatrix} {{\bf {x}}}_i \\ {{\bf {x}}}_i' \end{pmatrix},\end{aligned}$$ where ${\bf 1}$ is a vector of all-ones. While the elements within a feature map display strong correlations, different feature maps are conditioned on the data (i.e. ${{\bf {a}}}_i{^{(1)}}({{\bf {X}}},{{\bf {X}}}')$ and ${{\bf {a}}}_{i'}{^{(1)}}({{\bf {X}}},{{\bf {X}}}')$ are for $i\neq i'$), because the parameters for different feature-maps (i.e. the biases, $b_i{^{(1)}}$ and the filters, ${{\bf {W}}}{^{(1)}}_{i,:}$) are themselves .
#### Induction step.
Consider the feature maps at the $\ell$th layer, ${{\bf {a}}}_j{^{(\ell)}}({{\bf {X}}}, {{\bf {X}}}')$, to be multivariate Gaussian (i.e. for $j\neq j'$, ${{\bf {a}}}_{j}{^{(\ell)}}({{\bf {X}}}, {{\bf {X}}}')$ and ${{\bf {a}}}_{j'}{^{(\ell)}}({{\bf {X}}}, {{\bf {X}}}')$ are ). Our goal is to show that, taking the number of channels at layer $\ell$ to infinity (i.e. $C{^{(\ell)}} \rightarrow \infty$), the same properties hold at the next layer (i.e. all feature maps, ${{\bf {a}}}_i{^{(\ell+1)}}({{\bf {X}}},{{\bf {X}}}')$, are multivariate Gaussian ). Writing Eq. for two training examples, ${{\bf {X}}}$ and ${{\bf {X}}}'$, we obtain, $$\begin{aligned}
\label{eq:network-recursive-pair}
{{\bf {a}}}_i{^{(\ell+1)}}({{\bf {X}}},{{\bf {X}}}')
&= b_i{^{(\ell+1)}} {\bf 1}
+ \sum_{j=1}^{C{^{(\ell)}}} \begin{pmatrix} {{\bf {W}}}_{i,j}{^{(\ell+1)}} & {{\bf {0}}}\\ {{\bf {0}}}& {{\bf {W}}}_{i,j}{^{(\ell+1)}} \end{pmatrix}
\phi({{\bf {a}}}_j{^{(\ell)}}({{\bf {X}}},{{\bf {X}}}'))\end{aligned}$$ We begin by showing that ${{\bf {a}}}_i{^{(\ell+1)}}({{\bf {X}}},{{\bf {X}}}')$ is a multivariate Gaussian . The first term is multivariate Gaussian, as it is a linear function of $b_i{^{(\ell+1)}}$, which is itself Gaussian. We can apply the multivariate to show that the second term is also Gaussian, because, in the limit as $C{^{(\ell)}} \rightarrow \infty$, it is the sum of infinitely many terms: ${{\bf {a}}}_j^{(\ell)}({{\bf {X}}},{{\bf {X}}}')$ are by assumption, and ${{\bf {W}}}_{i,j}{^{(\ell+1)}}$ are by definition. Note that the same argument applies to all feature maps jointly, so all elements of ${{\bf {A}}}{^{(\ell+1)}}({{\bf {X}}},{{\bf {X}}}')$ (defined by analogy with Eq. \[eq:vapair\]) are jointly multivariate Gaussian.
Following @lee2017deep, to complete the argument, we need to show that the output feature maps are , i.e. ${{\bf {a}}}_i{^{(\ell+1)}}({{\bf {X}}},{{\bf {X}}}')$ and ${{\bf {a}}}_{i'}{^{(\ell+1)}}({{\bf {X}}},{{\bf {X}}}')$ are for $i\neq i'$. They are identically distributed, as $b_i{^{(\ell+1)}}$ and ${{\bf {W}}}_{i, j}{^{(\ell+1)}}$ are and $\phi({{\bf {a}}}_j{^{(\ell)}}({{\bf {X}}},{{\bf {X}}}'))$ is shared. To show that they are independent, remember that ${{\bf {a}}}_i{^{(\ell+1)}}({{\bf {X}}},{{\bf {X}}}')$ and ${{\bf {a}}}_{i'}{^{(\ell+1)}}({{\bf {X}}},{{\bf {X}}}')$ are jointly Gaussian, so it is sufficient to show that they are uncorrelated. We can show that they are uncorrelated noting that the weights ${{\bf {W}}}_{i, j}{^{(\ell+1)}}$ are independent with zero-mean, eliminating any correlations that might arise through the shared random vector, $\phi({{\bf {a}}}_j{^{(\ell)}}({{\bf {X}}},{{\bf {X}}}'))$.
The ConvNet and ResNet kernels
==============================
Here we derive a computationally efficient kernel corresponding to the described in the previous section. It is surprising that we can compute the kernel efficiently because the feature maps, ${{\bf {a}}}_i{^{(\ell)}}({{\bf {X}}})$, display rich covariance structure due to the shared convolutional filter. Computing and representing these covariances would be prohibitively computationally expensive. However, in many cases we only need the variance of the output, e.g. in the case of classification or regression with a final dense layer. It turns out that this propagates backwards through the convolutional network, implying that for every layer, we only need the “diagonal covariance” of the activations: the covariance between the corresponding elements of ${{\bf {a}}}_j{^{(\ell)}}({{\bf {X}}})$ and ${{\bf {a}}}_j{^{(\ell)}}({{\bf {X}}}')$, that is, $\operatorname{diag}{{\left( } \right)}{\operatorname*{\mathbb{C}}{{\left[ } \right]}{{{\bf {a}}}_j{^{(\ell)}}({{\bf {X}}}), {{\bf {a}}}_j{^{(\ell)}}({{\bf {X}}}')}}$.
mean and covariance\[sec:gp-covariance\]
-----------------------------------------
A is completely specified by its mean and covariance (or kernel) functions. These give the parameters of the joint Gaussian distribution of the indexed by any two inputs, ${{\bf {X}}}$ and ${{\bf {X}}}'$. For the purposes of computing the mean and covariance, it is easiest to consider the network as being written entirely in index notation, $$A_{i,\mu}{^{(\ell+1)}}({{\bf {X}}}) = b_i{^{(\ell+1)}} +
\sum_{j=1}^{C{^{(\ell)}}} \ \sum_{\nu=1}^{H{^{(\ell)}}\!D{^{(\ell)}}} W{^{(\ell+1)}}_{i,j,\mu,\nu}\, \phi(A{^{(\ell)}}_{j,\nu}({{\bf {X}}})).$$ where $\ell$ and $\ell+1$ denote the input and output layers respectively, $j$ and $i \in \{1,\dotsc,C{^{(\ell+1)}}\}$ denote the input and output channels, and $\nu$ and $\mu \in \{1,\dotsc,H{^{(\ell+1)}} D{^{(\ell+1)}}\}$ denote the location within the input and output channel or feature-maps.
The mean function is thus easy to compute, $$\operatorname*{\mathbb{E}}{{\left[ } \right]}{A_{i,\mu}{^{(\ell+1)}}({{\bf {X}}})} = \operatorname*{\mathbb{E}}{{\left[ } \right]}{b_i{^{(\ell+1)}}} +
\sum_{j=1}^{C{^{(\ell)}}} \ \sum_{\nu=1}^{H{^{(\ell)}}\!D{^{(\ell)}}} \operatorname*{\mathbb{E}}{{\left[ } \right]}{W{^{(\ell+1)}}_{i,j,\mu,\nu} \, \phi(A{^{(\ell)}}_{j,\nu}({{\bf {X}}}))} = 0,$$ as $b_i{^{(\ell+1)}}$ and $W{^{(\ell+1)}}_{i,j,\mu,\nu}$ have zero mean, and $W{^{(\ell+1)}}_{i,j,\nu,\mu}$ are independent of the activations at the previous layer, $\phi(A{^{(\ell)}}_{j,\nu}({{\bf {X}}}))$.
Now we show that it is possible to efficiently compute the covariance function. This is surprising because for many networks, we need to compute the covariance of activations between all pairs of locations in the feature map (i.e. $\operatorname*{\mathbb{C}}{{\left[ } \right]}{A_{i,\mu}{^{(\ell+1)}}({{\bf {X}}}), A_{i,\mu'}{^{(\ell+1)}}({{\bf {X}}}')}$ for $\mu,\mu'\in\{1,\dotsc,H{^{(\ell+1)}}D{^{(\ell+1)}}\}$) and this object is extremely high-dimensional, $N^2 (H{^{(\ell+1)}} D{^{(\ell+1)}})^2$. However, it turns out that we only need to consider the “diagonal” covariance, (i.e. we only need $\operatorname*{\mathbb{C}}{{\left[ } \right]}{A_{i,\mu}{^{(\ell+1)}}({{\bf {X}}}), A_{i,\mu}{^{(\ell+1)}}({{\bf {X}}}')}$ for $\mu\in\{1,\dotsc,H{^{(\ell+1)}}D{^{(\ell+1)}}\}$), which is a more manageable quantity of size $N^2 (H{^{(\ell+1)}} D{^{(\ell+1)}})$.
This is true at the output layer $(L+1)$: in order to achieve an output suitable for classification or regression, we use only a single output location $H{^{(L+1)}} = D{^{(L+1)}} =1$, with a number of “channels” equal to the number of of outputs/classes, so it is only possible to compute the covariance at that single location. We now show that, if we only need the covariance at corresponding locations in the outputs, we only need the covariance at corresponding locations in the inputs, and this requirement propagates backwards through the network.
Formally, as the activations are composed of a sum of terms, their covariance is the sum of the covariances of all those underlying terms, $$\begin{aligned}
&\operatorname*{\mathbb{C}}{{\left[ } \right]}{A_{i, \mu}{^{(\ell+1)}}({{\bf {X}}}), A_{i, \mu}{^{(\ell+1)}}({{\bf {X}}}')} =
\operatorname*{\mathbb{V}}{{\left[ } \right]}{b_i{^{(\ell)}}} + \\
&\hspace{3em}
\sum_{j=1}^{C{^{(\ell)}}} \sum_{j'=1}^{C{^{(\ell)}}} \ \sum_{\nu=1}^{H{^{(\ell)}}\!D{^{(\ell)}}} \ \sum_{\nu'=1}^{H{^{(\ell)}}\!D{^{(\ell)}}} \operatorname*{\mathbb{C}}{{\left[ } \right]}{W{^{(\ell+1)}}_{i,j,\mu,\nu} \phi(A{^{(\ell)}}_{j,\nu}({{\bf {X}}})), W{^{(\ell+1)}}_{i,j',\mu,\nu'} \phi(A{^{(\ell)}}_{j',\nu'}({{\bf {X}}}'))}.
\end{aligned}$$ As the terms in the covariance have mean zero, and as the weights and activations from the previous layer are independent, $$\begin{aligned}
&\operatorname*{\mathbb{C}}{{\left[ } \right]}{A_{i, \mu}{^{(\ell+1)}}({{\bf {X}}}), A_{i, \mu}{^{(\ell+1)}}({{\bf {X}}}')} =
\sigma_\text{b}^2 + \\
&\hspace{3em} \sum_{j=1}^{C{^{(\ell)}}} \sum_{j'=1}^{C{^{(\ell)}}} \ \sum_{\nu=1}^{H{^{(\ell)}}\!D{^{(\ell)}}} \ \sum_{\nu'=1}^{H{^{(\ell)}}\!D{^{(\ell)}}}
\operatorname*{\mathbb{E}}{{\left[ } \right]}{W{^{(\ell+1)}}_{i,j,\mu,\nu} W{^{(\ell+1)}}_{i,j',\mu,\nu'}}
\operatorname*{\mathbb{E}}{{\left[ } \right]}{\phi(A{^{(\ell)}}_{j,\nu}({{\bf {X}}})) \phi(A{^{(\ell)}}_{j',\nu'}({{\bf {X}}}'))}.
\end{aligned}$$ The weights are independent for different channels: ${{\bf {W}}}{^{(\ell+1)}}_{i,j}$ and ${{\bf {W}}}{^{(\ell+1)}}_{i,j'}$ are for $j\neq j'$, so $\operatorname*{\mathbb{E}}{{\left[ } \right]}{W{^{(\ell+1)}}_{i,j,\mu,\nu} W{^{(\ell+1)}}_{i,j',\mu,\nu'}} = 0$ for $j\neq j'$. Further, each row $\mu$ of the weight matrices ${{\bf {W}}}{^{(\ell+1)}}_{i,j}$ only contains independent variables or zeros (Fig. \[fig:conv-as-linear\]), so $\operatorname*{\mathbb{E}}{{\left[ } \right]}{W{^{(\ell+1)}}_{i,j,\mu,\nu} W{^{(\ell+1)}}_{i,j',\mu,\nu'}} = 0$ for $\nu \neq
\nu'$. Thus, we can eliminate the sums over $j'$ and $\nu'$: $$\operatorname*{\mathbb{C}}{{\left[ } \right]}{A_{i, \mu}{^{(\ell+1)}}({{\bf {X}}}), A_{i, \mu}{^{(\ell+1)}}({{\bf {X}}}')} =
\sigma_\text{b}^2 +
\sum_{j=1}^{C{^{(\ell)}}} \ \sum_{\nu=1}^{H{^{(\ell)}}\!D{^{(\ell)}}}
\operatorname*{\mathbb{E}}{{\left[ } \right]}{W{^{(\ell+1)}}_{i,j,\mu,\nu} W{^{(\ell+1)}}_{i,j,\mu,\nu}}
\operatorname*{\mathbb{E}}{{\left[ } \right]}{\phi(A{^{(\ell)}}_{j,\nu}({{\bf {X}}})) \phi(A{^{(\ell)}}_{j,\nu}({{\bf {X}}}'))}.$$ The $\mu$th row of ${{\bf {W}}}{^{(\ell+1)}}_{i,j}$ is zero for indices $\nu$ that do not belong to its convolutional patch, so we can restrict the sum over $\nu$ to that region. We also define $v_g{^{(1)}}({{\bf {X}}}, {{\bf {X}}}')$, to emphasise that the covariances are independent of the output channel, $j$. The variance of the first layer is $$\begin{aligned}
\label{eq:kernel-base}
K_\mu{^{(1)}}({{\bf {X}}}, {{\bf {X}}}') =
\operatorname*{\mathbb{C}}{{\left[ } \right]}{A_{i, \mu}{^{(1)}}({{\bf {X}}}), A_{i, \mu}{^{(1)}}({{\bf {X}}}')} &=
\sigma_\text{b}^2 +
\frac{\sigma_\text{w}^2}{C{^{(0)}}} \sum_{i=1}^{C{^{(0)}}} \;\sum_{\nu\in \mu\text{th patch}}
X_{i,\nu} X'_{i,\nu}.
\intertext{And we do the same for the other layers,}
\label{eq:kernel-recursive}
K_\mu{^{(\ell+1)}}({{\bf {X}}}, {{\bf {X}}}') =
\operatorname*{\mathbb{C}}{{\left[ } \right]}{A_{i, \mu}{^{(\ell+1)}}({{\bf {X}}}), A_{i, \mu}{^{(\ell+1)}}({{\bf {X}}}')} &=
\sigma_\text{b}^2 +
\sigma_\text{w}^2 \sum_{\nu\in \mu\text{th patch}} V_\nu{^{(\ell)}}({{\bf {X}}}, {{\bf {X}}}'),
$$ where $$\label{eq:phi-cov}
V_\nu{^{(\ell)}}({{\bf {X}}}, {{\bf {X}}}') = \operatorname*{\mathbb{E}}{{\left[ } \right]}{\phi(A{^{(\ell)}}_{j,\nu}({{\bf {X}}})) \phi(A{^{(\ell)}}_{j,\nu}({{\bf {X}}}'))}$$ is the covariance of the activations, which is again independent of the channel.
Covariance of the activities
----------------------------
The elementwise covariance in the right-hand side of Eq. can be computed in closed form for many choices of $\phi$ if the activations are Gaussian. For each element of the activations, one needs to keep track of the 3 distinct entries of the bivariate covariance matrix between the inputs, $K_\mu{^{(\ell+1)}}({{\bf {X}}}, {{\bf {X}}})$, $K_\mu{^{(\ell+1)}}({{\bf {X}}}, {{\bf {X}}}')$ and $K_\mu{^{(\ell+1)}}({{\bf {X}}}', {{\bf {X}}}')$.
For example, for the ReLU nonlinearity ($\phi(x) = \max(0, x)$), one can adapt @cho2009kernel in the same way as @gpbehaviour [section 3] to obtain $$ V_\nu{^{(\ell)}}({{\bf {X}}}, {{\bf {X}}}') = \frac{\sqrt{K_\nu{^{(\ell)}}({{\bf {X}}},{{\bf {X}}})K_\nu{^{(\ell)}}({{\bf {X}}}',{{\bf {X}}}')}}{\pi} {{\left( } \right)}{\sin \theta_\nu{^{(\ell)}} + (\pi - \theta_\nu{^{(\ell)}}) \cos \theta_\nu{^{(\ell)}}}
\label{eq:nlin-relu}$$ where $\theta_\nu{^{(\ell)}} = \cos^{-1}\left( K_\nu{^{(\ell)}}({{\bf {X}}},{{\bf {X}}}') /
\sqrt{K_\nu{^{(\ell)}}({{\bf {X}}},{{\bf {X}}})K_\nu{^{(\ell)}}({{\bf {X}}}',{{\bf {X}}}')}\right)$. Another example is the error function (erf) nonlinearity, similar to the hyperbolic tangent (tanh). The form of its relevant expectation [@williams1997computing] is in Appendix \[sec:erf\].
*Input*: two images, ${{\bf {X}}},{{\bf {X}}}' \in \mathbb{R}^{C^{(0)} \times (H^{(0)}W^{(0)})}$. Compute $K_\mu{^{(1)}}({{\bf {X}}}, {{\bf {X}}})$, $K_\mu{^{(1)}}({{\bf {X}}}, {{\bf {X}}}')$, and $K_\mu{^{(1)}}({{\bf {X}}}', {{\bf {X}}}')$\
for $\mu \in \{1,\dotsc,H{^{(1)}}D{^{(1)}}\}$; using Eq. . Compute $V_\mu{^{(\ell)}}({{\bf {X}}}, {{\bf {X}}}')$, $V_\mu{^{(\ell)}}({{\bf {X}}}, {{\bf {X}}}')$ and $V_\mu{^{(\ell)}}({{\bf {X}}},{{\bf {X}}}')$\
for $\mu \in \{1,\dotsc,H{^{(\ell)}}D{^{(\ell)}}\}$; using Eq. , or some other nonlinearity. Compute $K_\mu{^{(\ell+1)}}({{\bf {X}}},{{\bf {X}}})$, $K_\mu{^{(\ell+1)}}({{\bf {X}}},{{\bf {X}}}')$, and $K_\mu{^{(\ell+1)}}({{\bf {X}}}',{{\bf {X}}}')$\
for $\mu \in \{1,\dotsc,H{^{(\ell+1)}}D{^{(\ell+1)}}\}$; using Eq. . Output the scalar $K_1{^{(L+1)}}({{\bf {X}}},{{\bf {X}}}')$.
Efficiency of the ConvNet kernel\[sec:efficiency\]
--------------------------------------------------
We now have all the pieces for computing the kernel, as written in Algorithm \[alg:kernel\].
Putting together Eq. and Eq. gives us the surprising result that the diagonal covariances of the activations at layer $\ell+1$ only depend on the diagonal covariances of the activations at layer $\ell$. This is very important, because it makes the computational cost of the kernel be within a constant factor of the cost of a forward pass for the equivalent with 1 filter per layer.
Thus, the algorithm is more efficient that one would naively think. A priori, one needs to compute the covariance between all the elements of ${{\bf {a}}}_{j}{^{(\ell)}}({{\bf {X}}})$ and ${{\bf {a}}}_{j}{^{(\ell)}}({{\bf {X}}}')$ combined, yielding a $2H{^{(\ell)}}D{^{(\ell)}} \times
2H{^{(\ell)}}D{^{(\ell)}}$ covariance matrix for every pair of points. Instead, we only need to keep track of a $H{^{(\ell)}}D{^{(\ell)}}$-dimensional vector per layer and pair of points. Furthermore, the particular form for the kernel (Eq. \[eq:network-base\] and Eq. \[eq:network-recursive\]) implies that the required variances and covariances at all required locations can be computed efficiently as a convolution.
Kernel for a residual
----------------------
The induction step in the argument for behaviour from Sec. \[sec:argument-gp\] depends only on the previous activations being Gaussian. Since all the activations are Gaussian, we can add skip connections between the activations of different layers while preserving behaviour, e.g. ${{\bf {A}}}^{(\ell+1)}$ and ${{\bf {A}}}^{(\ell-s)}$ where $s$ is the number of layers that the skip connection spans. If we change the recursion (Eq. \[eq:network-recursive\]) to $${{\bf {a}}}_i^{(\ell+1)}({{\bf {X}}}) := {{\bf {a}}}_i^{(\ell-s)}({{\bf {X}}}) + {{\bf {b}}}_i^{(\ell+1)} + \sum_{j=1}^{C^{(\ell)}} {{\bf {W}}}_{i,j}^{(\ell)} \phi\left( {{\bf {a}}}_j^{(\ell)}({{\bf {X}}}) \right),
\label{eq:resnet-recursive}$$ then the kernel recursion (Eq. \[eq:kernel-recursive\]) becomes $$\label{eq:rekernel-recursive}
K_\mu{^{(\ell+1)}}({{\bf {X}}}, {{\bf {X}}}') = K_\mu{^{(\ell-s)}}({{\bf {X}}}, {{\bf {X}}}')
+ \sigma_\text{b}^2 + \sigma_\text{w}^2
\sum_{\nu \in \text{$\mu$th patch}} V_\nu{^{(\ell)}}({{\bf {X}}}, {{\bf {X}}}').$$ This way of adding skip connections is equivalent to the “pre-activation” shortcuts described by @he2016identity. Remarkably, the natural way of adding residual connections to is the one that performed best in their empirical evaluations.
Experiments
===========
We evaluate our kernel on the MNIST handwritten digit classification task. Classification likelihoods are not conjugate for , so we must make an approximation, and we follow @lee2017deep, in re-framing classification as multi-output regression.
The training set is split into $N=50000$ training and $10000$ validation examples. The regression targets ${{\bf {Y}}}\in \{-1, 1\}^{N \times 10}$ are a one-hot encoding of the example’s class: $y_{n,c} = 1$ if the $n$th example belongs to class $c$, and $-1$ otherwise.
Training is exact conjugate likelihood regression with noiseless targets ${{\bf {Y}}}$ [@rasmussen2006gaussian]. First we compute the $N\times N$ kernel matrix ${{\bf {K}}}_{xx}$, which contains the kernel between every pair of examples. Then we compute ${{\bf {K}}}_{xx}^{-1}{{\bf {Y}}}$ using a linear system solver.
The test set has $N_T=10000$ examples. We compute the $N_T \times N$ matrix ${{\bf {K}}}_{x^*x}$, the kernel between each test example and all the training examples. The predictions are given by the row-wise maximum of ${{\bf {K}}}_{x^*x}{{\bf {K}}}_{xx}^{-1}{{\bf {Y}}}$.
For the “ConvNet ” and “Residual ”, (Table \[mnist-results\]) we optimise the kernel hyperparameters by random search. We draw $M$ random hyperparameter samples, compute the resulting kernel’s performance in the validation set, and pick the highest performing run. The kernel hyperparameters are: $\sigma_\text{b}^2$, $\sigma_\text{w}^2$; the number of layers; the convolution stride, filter sizes and edge behaviour; the nonlinearity (we consider the error function and ReLU); and the frequency of residual skip connections (for Residual CNN GPs). We do not retrain the model on the validation set after choosing hyperparameters.
The “ResNet ” (Table \[mnist-results\]) is the kernel equivalent to a 32-layer version of the basic residual architecture by @he2016deep. The differences are: an initial $3
\times 3$ convolutional layer and a final dense layer instead of average pooling. We chose to remove the pooling because computing its output variance requires the off-diagonal elements of the filter covariance, in which case we could not exploit the efficiency gains described in Sec. \[sec:efficiency\]. We found that the despite it not being optimised, the 32-layer ResNet outperformed all other comparable architectures (Table \[mnist-results\]), including the NNGP in @lee2017deep, which is state-of-the-art for non-convolutional networks, and convolutional [@markvdw2017convolutional; @kumar2018deep]. That said, our results have not reached state-of-the-art for methods that incorporate a parametric neural network, such as a standard ResNet [@chen2018neural] and a Gaussian process with a deep neural network kernel [@bradshaw2017adversarial].
[**Method**]{} [**\#samples**]{} [**Validation error**]{} [**Test error**]{}
----------------------------------------------------- ------------------- -------------------------- -------------------- --
NNGP [@lee2017deep] $\approx 250$ – 1.21%
Convolutional [@markvdw2017convolutional] SGD – 1.17%
Deep Conv. [@kumar2018deep] SGD – 1.34%
ConvNet 27 0.71% 1.03%
Residual CNN 27 0.71% 0.93%
ResNet – [0.68%]{} [**0.84%**]{}
+ parametric deep kernel [@bradshaw2017adversarial] SGD – 0.60%
ResNet [@chen2018neural] – – [**0.41%**]{}
: MNIST classification results. \#samples gives the number of kernels that were randomly sampled for the hyperparameter search. “ConvNet GP” and “Residual CNN GP” are random CNN architectures with a fixed filter size, whereas “ResNet GP” is a slight modification of the architecture by @he2016identity. Entries labelled “SGD” used stochastic gradient descent for tuning hyperparameters, by maximising the likelihood of the training set. The last two methods use parametric neural networks. The hyperparameters of the ResNet GP were not optimised [they were fixed based on the architecture from @he2016identity]. See Table \[table:hyperparameters\] (appendix) for optimised hyperparameter values. \[mnist-results\]
To check whether the GP limit is applicable to relatively small networks used practically (with of the order of $100$ channels in the first layers), we randomly sampled $10,000$ 32-layer ResNets, with 3, 10, 30 and 100 channels in the first layers. Following the usual practice for ResNets we increase the number the number of hidden units when we downsample the feature maps. Then, we compare the sampled and limiting theoretical distributions of ${{\bf {A}}}{^{(32)}}({{\bf {X}}})$ for a given input ${{\bf {X}}}$.
The probability density plots show a good match around 100 channels (Fig. \[fig:random\]A), which matches a more sensitive graphical procedure based on quantile-quantile plots (Fig. \[fig:random\]B). Notably, even for only 30 channels, the empirical moments (computed over many input images) match closely the limiting ones (Fig. \[fig:random\]C). For comparison, typical ResNets use from 64 [@he2016deep] to 192 [@zagoruyko2016wide] channels in their first layers. We believe that this is because the moment propagation equations only require the Gaussianity assumption for propagation through the ReLU, and presumably this is robust to non-Gaussian input activations.
![ Distribution of the output ${{\bf {A}}}{^{(32)}}({{\bf {X}}})$: limiting density and samples of finite 32-layer ResNets [@he2016identity] with $C=3,10,30,100$ channels. **A)** Empirical histogram and limiting density function for one input image. **B)** A more sensitive test of Gaussianity is a quantile-quantile plot, which plots in $x$ the value of a quantile in the limiting density and in $y$ the corresponding quantile in the empirical one, for one input image. **C)** The empirical moments (variances and covariances) over 100 training images show a good match for all numbers of channels. \[fig:random\] ](img/random.pdf)
#### Computational efficiency.
Asymptotically, computing the kernel matrix takes $O(N^2 LD)$ time, where $L$ is the number of layers in the network and $D$ is the dimensionality of the input, and inverting the kernel matrix takes $O(N^3)$. As such, we expect that for very large datasets, inverting the kernel matrix will dominate the computation time. However, on MNIST, $N^3$ is only around a factor of $10$ larger than $N^2 LD$. In practice, we found that it was more expensive to *compute* the kernel matrix than to invert it. For the ResNet kernel, the most expensive, computing ${{\bf {K}}}_{xx}$, and ${{\bf {K}}}_{xx*}$ for validation and test took $3$h $40$min on two Tesla P100 GPUs. In contrast, inverting ${{\bf {K}}}_{xx}$ and computing validation and test performance took $43.25\pm 8.8$ seconds on a single Tesla P100 GPU.
Related work
============
Van der Wilk et al. [@markvdw2017convolutional] also adapted to image classification. They defined a prior on functions $f$ that takes an image and outputs a scalar. First, draw a function $g \sim \mathcal{GP}(0, k_p({{\bf {X}}}, {{\bf {X}}}'))$. Then, $f$ is the sum of the output of $g$ applied to each of the convolutional patches. Their approach is also inspired by convolutional , but their kernel $k_p$ is applied to all pairs of patches of ${{\bf {X}}}$ and ${{\bf {X}}}'$. This makes their convolutional kernel expensive to evaluate, requiring inter-domain inducing point approximations to remain tractable. The kernels in this work, directly motivated by the infinite-filter limit of a , only apply something like $k_p$ to the *corresponding* pairs of patches within ${{\bf {X}}}$ and ${{\bf {X}}}'$ (Eq. \[eq:kernel-base\]). As such, the kernels are cheaper to compute and exhibit superior performance (Table \[mnist-results\]), despite the use of an approximate likelihood function.
@kumar2018deep define a prior over functions by stacking several with van der Wilk’s convolutional kernel, forming a “Deep ” [@damianou13a]. In contrast, the kernel in this paper confines all hierarchy to the definition of the kernel, and the resulting is shallow.
@wilson2016deep introduced and @bradshaw2017adversarial improved deep kernel learning. The inputs to a classic kernel $k$ (e.g. RBF) are preprocessed by applying a feature extractor $g$ (a deep ) prior to computing the kernel: $k_\text{deep}({{\bf {X}}}, {{\bf {X}}}') {:=}k(g({{\bf {X}}}; \theta), g({{\bf {X}}}', \theta))$. The parameters are optimised by gradient ascent using the likelihood as the objective, as in standard kernel learning [@rasmussen2006gaussian Chapter 5]. Since deep kernel learning incorporates a state-of-the-art with over $10^6$ parameters, we expect it to perform similarly to a applied directly to the task of image classification. At present both and deep kernel learning display superior performance to the kernels in this work. However, the kernels defined here have far fewer parameters (around $10$, compared to their $10^6$). @borovykh also suggests that a exhibits behaviour. However, they take the infinite limit with respect to the *filter size*, not the number of filters. Thus, their infinite network is inapplicable to real data which is always of finite dimension.
Finally, there is a series of papers analysing the mean-field behaviour of deep and which aims to find good random initializations, i.e. those that do not exhibit vanishing or exploding gradients or activations [@schoenholz2016deep; @yang2017mean]. Apart from their very different focus, the key difference to our work is that they compute the variance for a single training-example, whereas to obtain the kernel, we additionally need to compute the output covariances for different training/test examples [@xiao2018dynamical].
Conclusions and future work
===========================
We have shown that deep Bayesian with infinitely many filters are equivalent to a with a recursive kernel. We also derived the kernel for the equivalent to a , and showed that, in handwritten digit classification, it outperforms all previous approaches that do not incorporate a parametric into the kernel. Given that most state-of-the-art neural networks incorporate structure (convolutional or otherwise) into their architecture, the equivalence between and is potentially of considerable practical relevance. In particular, we hope to apply in domains as widespread as adversarial examples, lifelong learning and k-shot learning, and we hope to improve them by developing efficient multi-layered inducing point approximation schemes.
Appendix
========
Technical notes on limits\[sec:notes-limits\]
---------------------------------------------
The key technical issues in the proof (and the key differences between @lee2017deep and @matthews2018gaussian) arise from exactly how and where we take limits. In particular, consider the activations as being functions of the activities at the previous layer, $$\begin{aligned}
{{\bf {A}}}{^{(4)}} &= {{\bf {A}}}{^{(4)}}({{\bf {A}}}{^{(3)}}({{\bf {A}}}{^{(2)}}({{\bf {A}}}{^{(1)}}({{\bf {X}}}))))\end{aligned}$$ Now, there are two approaches to taking limits. First, both our argument in the main text, and the argument in @lee2017deep is valid if we are able to take limits “inside” the network, $$\begin{aligned}
{{\bf {A}}}{^{(4)}}_\text{L} &= \lim_{C{^{(3)}}\rightarrow \infty} {{\bf {A}}}{^{(4)}} {{\left( } \right)}{\lim_{C{^{(2)}}\rightarrow \infty}{{\bf {A}}}{^{(3)}}{{\left( } \right)}{\lim_{C{^{(1)}}\rightarrow \infty} {{\bf {A}}}{^{(2)}}{{\left( } \right)}{{{\bf {A}}}{^{(1)}}({{\bf {X}}})}}}.\end{aligned}$$ However, @gpbehaviour [@matthews2018gaussian] argue that is preferable to take limits “outside” the network. In particular, @matthews2018gaussian take the limit with all layers simultaneously, $$\begin{aligned}
{{\bf {A}}}{^{(4)}}_\text{M} &= \lim_{n\rightarrow \infty} {{\bf {A}}}{^{(4)}} {{\left( } \right)}{{{\bf {A}}}{^{(3)}}{{\left( } \right)}{ {{\bf {A}}}{^{(2)}}{{\left( } \right)}{{{\bf {A}}}{^{(1)}}({{\bf {X}}})}}},\end{aligned}$$ where $C{^{(\ell)}} = C{^{(\ell)}}(n)$ goes to infinity as $n \rightarrow \infty$. That said, similar technical issues arise if we take limits in sequence, but outside the network.
Extending the derivations of @matthews2018gaussian to the convolutional case\[sec:actual-proof\]
------------------------------------------------------------------------------------------------
In the main text, we follow @lee2017deep in sequentially taking the limit of each layer to infinity (i.e. $C{^{(1)}}\rightarrow \infty$, then $C{^{(2)}}\rightarrow \infty$ etc.). This dramatically simplified the argument, because taking the number of units in the previous layer to infinity means that the inputs from that layer are exactly Gaussian distributed. However, @matthews2018gaussian argue that the more practically relevant limit is where we take all layers to infinity simultaneously. This raises considerable additional difficulties, because we must reason about convergence in the case where the previous layer is finite. Note that this section is not intended to stand independently: it is intended to be read alongside @matthews2018gaussian, and we use several of their results without proof.
Mirroring Definition 3 in @matthews2018gaussian, we begin by choosing a set of “width” functions, $C{^{(\ell)}}(n)$, for $\ell \in \{1,\dotsc,L\}$ which all approach infinity as $n \rightarrow \infty$. In @matthews2018gaussian, these functions described the number of hidden units in each layer, whereas here they describe the number of channels. Our goal is then to extend the proofs in @matthews2018gaussian (in particular, of theorem 4), to show that the output of our convolutional networks converge in distribution to a Gaussian process as $n \rightarrow \infty$, with mean zero and covariance given by the recursion in Eqs. (\[eq:kernel-base\] – \[eq:phi-cov\]).
The proof in @matthews2018gaussian has three main steps. First, they use the Cramér-Wold device, to reduce the full problem to that of proving convergence of scalar random variables to a Gaussian with specified variance. Second, if the previous layers have finite numbers of channels, then the channels ${{\bf {a}}}_j{^{(\ell)}}({{\bf {X}}})$ and ${{\bf {a}}}_j{^{(\ell)}}({{\bf {X}}}')$ are uncorrelated but no longer independent, so we cannot apply the CLT directly, as we did in the main text. Instead, they write the activations as a sum of exchangeable random variables, and derive an adapted CLT for exchangeable (rather than independent) random variables [@blum1958central]. Third, they show that moment conditions required by their exchangeable CLT are satisfied.
To extend their proofs to the convolutional case, we begin by defining our networks in a form that is easier to manipulate and as close as possible to Eq. (21-23) in @matthews2018gaussian, $$\begin{aligned}
A_{i, \mu}{^{(1)}} = f_{i, \mu}{^{(1)}}(x) &= \frac{\sigma_\text{w}}{\sqrt{C{^{(0)}}}} \sum_{j=1}^{C{^{(0)}}} \sum_{\nu \in \text{$\mu$th patch}} \epsilon_{i,j,\mu,\nu}{^{(1)}} x_{j, \nu} + b_i{^{(1)}}, \quad i \in \mathbb{N}\\
g_{i, \mu}{^{(\ell)}}(x) &= \phi{{\left( } \right)}{f_{i, \mu}{^{(\ell)}}(x)}\\
\label{eq:def:f:rec}
A_{i, \mu}{^{(\ell+1)}} = f_{i, \mu}{^{(\ell+1)}}(x) &= \frac{\sigma_\text{w}}{\sqrt{C{^{(\ell)}}(n)}} \sum_{j=1}^{C{^{(\ell)}}(n)} \sum_{\nu \in \text{$\mu$th patch}} \epsilon_{i,j,\mu,\nu} {^{(\ell+1)}} g_{j, \nu}{^{(\ell)}}(x) + b_i{^{(\ell+1)}}, \quad i \in \mathbb{N}
\intertext{where,}
\epsilon_{i,j,\mu,\nu} &\sim \mathcal{N}(0, 1).\end{aligned}$$
The first step is to use the Cramér-Wold device [Lemma 6 in @matthews2018gaussian], which indicates that convergence in distribution of a sequence of finite-dimensional vectors is equivalent to convergence on all possible linear projections to the corresponding real-valued random variable. Mirroring Eq. 25 in @matthews2018gaussian, we consider convergence of random vectors, $f{^{(\ell)}}_{i,\mu}(x)[n] - b_{i}{^{(\ell)}}$, projected onto $\alpha{^{(x, i, \mu)}}$, $$\begin{aligned}
\mathcal{T}{^{(\ell)}}{{\left( } \right)}{\mathcal{L}, \alpha}[n] = \sum_{(x, i, \mu) \in \mathcal{L}} \alpha{^{(x, i, \mu)}} {{\left[ } \right]}{f{^{(\ell)}}_{i,\mu}(x)[n] - b_{i}{^{(\ell)}}}.\end{aligned}$$ where $\mathcal{L} \subset \mathcal{X} \times \mathbb{N} \times \{1,\dotsc,H{^{(\ell)}}D{^{(\ell)}}\}$ is a finite set of tuples of data points and channel indicies, $i$, and indicies of elements within channels/feature maps, $\mu$. The suffix $[n]$ indicates width functions that are instantiated with input, $n$.
Now, we must prove that these projections converge in distribution a Gaussian. We begin by defining summands, as in Eq. 26 in @matthews2018gaussian, $$\begin{aligned}
\gamma_j{^{(\ell)}}{{\left( } \right)}{\mathcal{L}, \alpha}[n] := \sigma_\text{w} \sum_{(x, i, \mu) \in \mathcal{L}} \alpha{^{(x, i, \mu)}} \sum_{\nu \in \text{$\mu$th patch}} \epsilon_{i,j,\mu,\nu}{^{(\ell)}} g_{j, \nu}{^{(\ell-1)}}(x)[n],\end{aligned}$$ such that the projections can be written as a sum of the summands, exactly as in Eq. 27 in @matthews2018gaussian, $$\begin{aligned}
\mathcal{T}{^{(\ell)}}{{\left( } \right)}{\mathcal{L}, \alpha}[n] = \frac{1}{\sqrt{C{^{(\ell-1)}}(n)}} \sum_{j=1}^{C{^{(\ell-1)}}(n)} \gamma_j{^{(\ell)}}{{\left( } \right)}{\mathcal{L}, \alpha}[n].\end{aligned}$$ Now we can apply the exchangeable CLT to prove that $\mathcal{T}{^{(\ell)}}{{\left( } \right)}{\mathcal{L}, \alpha}[n]$ converges to the limiting Gaussian implied by the recursions in the main text. To apply the exchangeable CLT, the first step is to mirror Lemma 8 in @matthews2018gaussian, in showing that for each fixed $n$ and $\ell \in \{2,\dotsc,L+1\}$, the summands, $\gamma_j{^{(\ell)}}{{\left( } \right)}{\mathcal{L}, \alpha}[n]$ are exchangeable with respect to the index $j$. In particular, we apply de Finetti’s theorem, which states that a sequence of random variables is exchangeable if and only if they are i.i.d. conditional on some set of random variables, so it is sufficient to exhibit such a set of random variables. Mirroring Eq. 29 in @matthews2018gaussian, we apply the recursion, $$\begin{aligned}
\gamma_j{^{(\ell)}}{{\left( } \right)}{\mathcal{L}, \alpha}[n] := &\sigma_\text{w} \sum_{(x, i, \mu) \in \mathcal{L}} \alpha{^{(x, i, \mu)}} \sum_{\nu \in \text{$\mu$th patch}} \epsilon_{i,j,\mu,\nu}{^{(\ell)}}\\
&\hspace{1em}\phi{{\left( } \right)}{\frac{\sigma_\text{w}}{\sqrt{C{^{(\ell-2)}}(n)}} \sum_{k=1}^{C{^{(\ell-2)}}(n)} \sum_{\xi \in \text{$\nu$th patch}} \epsilon_{j,k,\nu,\xi}{^{(\ell-1)}} g_{k, \xi}{^{(\ell-2)}}(x)[n] + b_j{^{(\ell+1)}}}
\end{aligned}$$ As such, the summands are iid conditional on the finite set of random variables $\left\{ g_{k, \xi}{^{(\ell-2)}}(x)[n]: k\in\{1,\dotsc,C{^{(\ell-2)}}\},\xi\in\{1,\dotsc,H{^{(\ell-2)}}D{^{(\ell-2)}}\}, x\in\mathcal{L}_\mathcal{X}\right\}$, where $\mathcal{L}_\mathcal{X}$ is the set of input points in $\mathcal{L}$.
The exchangeable CLT in Lemma 10 in @matthews2018gaussian indicates that $\mathcal{T}{^{(\ell)}}{{\left( } \right)}{\mathcal{L}, \alpha}[n]$ converges in distribution to $\mathcal{N}{{\left( } \right)}{0, \sigma_*^2}$ if the summands are exchangeable (which we showed above), and if three conditions hold,
1. $\operatorname*{\mathbb{E}}_n{{\left[ } \right]}{\gamma_j{^{(\ell)}} \gamma_{j'}{^{(\ell)}}} = 0$
2. $\lim_{n\rightarrow \infty} \operatorname*{\mathbb{E}}_n{{\left[ } \right]}{{{\left( } \right)}{\gamma_j{^{(\ell)}}}^2 {{\left( } \right)}{\gamma_{j'}{^{(\ell)}}}^2} = \sigma_*^4$
3. $\operatorname*{\mathbb{E}}_n{{\left[ } \right]}{|\gamma_j{^{(\ell)}}|^3} = o{{\left( } \right)}{\sqrt{C{^{(\ell)}}(n)}}$
Condition a) follows immediately as the summands are uncorrelated and zero-mean. Conditions b) and c) are more involved as convergence in distribution in the previous layers does not imply convergence in moments for our activation functions.
We begin by considering the extension of Lemma 20 in @matthews2018gaussian, which allow us to show conditions b) and c) above, even in the case of unbounded but linearly enveloped nonlinearities [Definition 1 in @matthews2018gaussian]. Lemma 20 states that the eighth moments of $f_{i, \mu}{^{(t)}}(x)[n]$ are bounded by a finite constant independent of $n \in \mathbb{N}$. We prove this by induction. The base case is trivial, as $f_{j, \mu}{^{(1)}}(x)[n]$ is Gaussian. Following @matthews2018gaussian, assume the condition holds up to $\ell-1$, and show that the condition holds for layer $\ell$. Using Eq. , we can bound the activations at layer $\ell$, $$\begin{aligned}
\operatorname*{\mathbb{E}}{{\left[ } \right]}{\lvert f_{i, \mu}{^{(\ell)}}(x)[n] \rvert^8} \leq 2^{8-1} \operatorname*{\mathbb{E}}{{\left[ } \right]}{\lvert b_i{^{(\ell)}}\rvert^8 + \left\lvert \frac{\sigma_\text{w}}{\sqrt{C{^{(\ell-1)}}}}\sum_{j=1}^{C{^{(\ell-1)}}(n)} \sum_{\nu \in \text{$\mu$th patch}} \epsilon_{i,j,\mu,\nu}{^{(\ell)}} g_{j, \nu}{^{(\ell-1)}}(x)[n] \right\rvert^8}\end{aligned}$$ Following Eq. 48 in @matthews2018gaussian, which uses Lemma 19 in @matthews2018gaussian, we have, $$\begin{gathered}
\label{eq:bound}
\operatorname*{\mathbb{E}}{{\left[ } \right]}{\left\lvert \frac{\sigma_\text{w}}{\sqrt{C{^{(\ell-1)}}}}\sum_{j=1}^{C{^{(\ell-1)}}(n)} \sum_{\nu \in \text{$\mu$th patch}} \epsilon_{i,j,\mu,\nu}{^{(\ell)}} g_{j, \nu}{^{(\ell-1)}}(x)[n] \right\rvert^8} \\
= \frac{2^4 \Gamma(4 + 1/2)}{\Gamma(1/2)} \operatorname*{\mathbb{E}}{{\left[ } \right]}{\left\lvert \frac{\sigma_\text{w}^2}{C{^{(\ell-1)}}(n)} \lVert g_{j\in\{1,\dotsc,C{^{(\ell-1)}}(n)\},\nu\in\text{$\mu$th patch}}{^{(\ell-1)}}(x)[n] \rVert_2^2 \right\rvert^4}.\end{gathered}$$ where $g_{j\in\{1,\dotsc,C{^{(\ell-1)}}(n)\},\nu\in\text{$\mu$th patch}}{^{(\ell-1)}}(x)[n]$ is the set of post-nonlinearities corresponding to $j\in\{1,\dotsc,C{^{(\ell-1)}}(n)\}$ and $\nu\in\text{$\mu$th patch}$. Following @matthews2018gaussian, observe that, $$\begin{aligned}
\frac{1}{C{^{(\ell-1)}}(n)} \lVert g_{j\in\{1,\dotsc,C{^{(\ell-1)}}(n)\},\nu\in\text{$\mu$th patch}}{^{(\ell-1)}}(x)[n] \rVert_2^2
&= \frac{1}{C{^{(\ell-1)}}(n)} \sum_{j=1}^{C{^{(\ell-1)}}(n)} \sum_{\nu \in \text{$\mu$th patch}} {{\left( } \right)}{g_{j, \nu}{^{(\ell-1)}}(x)[n]}^2\\
&\hspace{-3em}\leq \frac{1}{C{^{(\ell-1)}}(n)} \sum_{j=1}^{C{^{(\ell-1)}}(n)} \sum_{\nu \in \text{$\mu$th patch}} {{\left( } \right)}{c + m \lvert f_{j, \nu}{^{(\ell-1)}}(x)[n] \rvert}^2\end{aligned}$$ by the linear envelope property, $\lvert\phi(u)\rvert \leq c + m \lvert u\rvert$. Following @matthews2018gaussian, we substitute this bound back into Eq. and suppress a multiplicative constant independent of $x$ and $n$, $$\begin{gathered}
\operatorname*{\mathbb{E}}{{\left[ } \right]}{\left\lvert \frac{\sigma_\text{w}}{\sqrt{C{^{(\ell-1)}}(n)}}\sum_{j=1}^{C{^{(\ell-1)}}(n)} \sum_{\nu \in \text{$\mu$th patch}} \epsilon_{i,j,\mu,\nu}{^{(\ell)}} g_{j, \nu}{^{(\ell-1)}}(x)[n] \right\rvert^8} \\
\leq \frac{1}{{{\left( } \right)}{C{^{(\ell-1)}}(n)}^4} \operatorname*{\mathbb{E}}{{\left[ } \right]}{\left\lvert \sum_{j=1}^{C{^{(\ell-1)}}(n)} \sum_{\nu \in \text{$\mu$th patch}} c^2 + 2cm \lvert f_{j, \mu}{^{(\ell-1)}}(x)[n] \rvert + m^2 \lvert f_{j, \mu}{^{(\ell-1)}}(x)[n]\rvert^2 \right\rvert^4}\end{gathered}$$ This can be multiplied out, yielding a weighted sum of expectations of the form, $$\begin{aligned}
\operatorname*{\mathbb{E}}{{\left[ } \right]}{
\lvert f_{k,\nu}{^{(\ell-1)}}(x)[n]\rvert^{p_1}
\lvert f_{l,\xi}{^{(\ell-1)}}(x)[n]\rvert^{p_2}
\lvert f_{r,\pi}{^{(\ell-1)}}(x)[n]\rvert^{p_3}
\lvert f_{q,\rho}{^{(\ell-1)}}(x)[n]\rvert^{p_4}
}\end{aligned}$$ with $p_i \in \{0, 1, 2\}$ for $i=1,2,3,4$, and $k,l,r,q\in \{1,\dotsc,C{^{(\ell-1)}}(n)\}$, and $\nu,\xi,\pi,\rho\in\text{$\mu$th patch}$ where the weights of these terms are independent of $n$. Using Lemma 18 in @matthews2018gaussian, each of these terms is bounded if the eighth moments of $f_{k,\mu}{^{(\ell-1)}}(x)[n]$ are bounded, which is our inductive hypothesis. The number of terms in the expanded sum is upper bounded by ${{\left( } \right)}{2 C{^{(\ell-1)}}(n) \lvert \text{$\mu$th patch} \rvert}^4$, where $\lvert \text{$\mu$th patch} \rvert$ is the number of elements in a convolutional patch. Thus, we can use the same constant for any $n$ due to the $1/{{\left( } \right)}{C{^{(\ell-1)}}(n)}^4$ scaling. As in @matthews2018gaussian, noting that $f_{j,\mu}{^{(\ell-1)}}(x)[n]$ are exchangeable over $j$ for any $x$ and $n$ concludes the proof.
Using this result, we can obtain a straightforward adaptation of Lemmas 15, 16 and 21 in @matthews2018gaussian. Lemma 15 gives condition b), Lemma 16 gives condition c); Lemma 15 requires uniform integrability, which is established by Lemma 21.
Calibration of Gaussian process uncertainty
-------------------------------------------
It is important to check that the estimates of uncertainty produced by our Gaussian process are reasonable. However, to make this assessment, we needed to use a proper likelihood, and not the squared-error loss in the main text. We therefore used our kernel to perform the full, multi-class classification problem in GPflow [@matthews2017gpflow], with a RobustMax likelihood [@robustmax]. The more difficult non-conjugate inference problem forced us to use 1000 inducing points, randomly chosen from the training inputs. Both our kernel and an RBF kernel have similar calibration curves, that closely track the diagonal, indicating accurate uncertainty estimation. However, even in the inducing point setting, our convolutional kernel gave considerably better performance than the RBF kernel (2.4% error vs 3.4% error). See Fig. \[fig:calibration\].
Closed-form expectation for the error function nonlinearity\[sec:erf\]
----------------------------------------------------------------------
The error function (erf) is given by the integral $\phi(x) = \frac{2}{\sqrt{\pi}}\int_0^xe^{-t^2}dt$, and is related to the Gaussian CDF. It is very similar to the hyperbolic tangent (tanh), with the difference that erf’s tails approach 0 and 1 faster than the tanh’s tails.
@williams1997computing gives us the relevant closed-form integral: $$V_\nu{^{(\ell)}}({{\bf {X}}}, {{\bf {X}}}')
= \frac{2}{\pi} \sin^{-1}\left( \frac{2\,K_\nu{^{(\ell)}}({{\bf {X}}},{{\bf {X}}}')}{\sqrt{(1 + 2\,K_\nu{^{(\ell)}}({{\bf {X}}},{{\bf {X}}})(1 + 2\,K_\nu{^{(\ell)}}({{\bf {X}}}',{{\bf {X}}}'))}}\right).
\label{eq:nlin-erf}$$
![ Calibration plots for an RBF kernel (left) and the ResNet kernel (right). The x-axis gives GP prediction for the label probability. The points give corresponding proportion of test points with that label, and the bars give the proportion of training examples in each bin. \[fig:calibration\] ](img/reliability.pdf){width="\textwidth"}
[**Hyperparameters**]{} [**ConvNet GP**]{} [**Residual CNN GP**]{} [**ResNet GP**]{}
------------------------- -------------------- ------------------------- ------------------- --
$\sigma^2_\text{b}$ 7.86 4.69 0.0
$\sigma^2_\text{w}$ 2.79 7.27 1.0
\#layers 7 9 32
Stride 1 1 mostly 1, some 2
Filter sizes 7 4 3
Padding SAME SAME SAME
Nonlinearity ReLU ReLU ReLU
Skip connections – every 1 layer every 2 layers
Test error 1.03% 0.93% 0.84%
: Optimised hyperparameter values. The ResNet has $\sigma^2_\text{b}=0$ because there are no biases in the architecture of @he2016deep. Because of this, and the fact that the nonlinearity is a ReLU, the value of $\sigma^2_\text{w}$ does not matter except for numerical stability: the $\sigma^2_\text{w}$ for every layer can be taken out of the nonlinearity and multiplied together, and kernel functions that are equal up to multiplication give the same classification results in this Gaussian likelihood setting. \[table:hyperparameters\]
[^1]: Code to replicate this paper is available at <https://github.com/convnets-as-gps/convnets-as-gps>
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
ASAS is a long term project to monitor bright variable stars over the whole sky. It has discovered 50,122 variables brighter than ${\rm V < 14 }$ mag south of declination ${\rm + 28^{\circ}}$, and among them 11,099 eclipsing binaries. We present a preliminary analysis of 5,384 contact, 2,957 semi-detached, and 2,758 detached systems. The statistics of the distribution provides a qualitative confirmation of decades old idea of Flannery and Lucy that W UMa type binaries evolve through a series of relaxation oscillations: ASAS finds comparable number of contact and semidetached systems.
The most surprising result is a very small number of detached eclipsing binaries with periods ${\rm P < 1 }$ day, the systems believed to be the progenitors of W UMa stars. As many (perhaps all) contact binaries have companions, there is a possibility that some were formed in a Kozai cycle, as suggested by Eggleton and his associates.
author:
- |
B. Paczy[ń]{}ski$^{1}$[^1] and D. Szczygiel$^{2}$, B. Pilecki$^{2}$, G. Pojmański$^{2}$ [^2]\
$^{1}$Princeton University Observatory, Peyton Hall, Princeton, NJ 08544, USA\
$^{2}$Warsaw University Observatory, Al. Ujazdowskie 4, PL-00-478, Poland
date: 'Accepted –. Received – ; in original form –'
title: Eclipsing binaries in ASAS catalog
---
\[firstpage\]
stars: eclipsing – stars: binary – stars: evolution
Introduction to ASAS {#sect:intro1}
====================
ASAS - All Sky Automated Survey, is a long term project dedicated to detection and monitoring variability of bright stars. This paper presents the results of several years of observations done at the Las Campanas Observatory with a single instrument: a telescope with the aperture of 7 cm, the focal length of 20 cm, done through a standard V-band filter and a ${\rm 2K \times 2K }$ CCD camera with ${\rm 15 \mu }$ pixels from Apogee (Pojmański 1997, 1998, 2000, 2002, 2003, Pojmański and Maciejewski 2004, 2005, Pojmański, Pilecki and Szczygiel 2005). More information about ASAS is provided on the WWW:
http://www.astrouw.edu.pl/\~gp/asas/asas.html
http://archive.princeton.edu/\~asas/
{width="\linewidth"}
The variable stars were discovered quasi - uniformly for declination ${\rm < + 28^{\circ} }$, covering almost 3/4 of the full sky. Fig. 1 shows the distribution of ASAS variables in the sky in the Galactic coordinates. The Milky Way is clearly visible, together with the dust lanes. The distribution of ASAS stars as a function of V-mag is shown on Fig. 2 for all stars (17,000,000), all variable stars (50,122), and all eclipsing binaries, the latter divided into Eclipsing Contact binaries (EC, 5,384), Eclipsing Semi Detached binaries (ESD, 2,957), and Eclipsing Detached binaries (ED, 2758). The statistics for stars with ${\rm 8 < V < 12 }$ mag appears to be approximately complete, but the efficiency falls rapidly within the range of ${\rm 12 < V < 14 }$ mag as the detection limit is approached. Also, the statistics deteriorates for ${\rm 8 < V }$ because of saturation effects.
{width="\linewidth"}
All stars were observed for about 5 years, with a small subset for 8 years. Typical number of photometric measurements was several hundred. The distribution of this number is shown in Fig. 3. The total number of photometric measurements was 2,916,000. As ASAS continues its operation the number of measurements will increase, approximately 100 V-band photometric measurements per year per variable. We intend to continue the project indefinitely, with some upgrades. While only V-band results are reported here, the I-band photometry was accumulated, and several years of data are already stored on a RAID-5 disk system. However, it will take another year to process I-band data, well over a Tera-byte. There are more stars detectable in the I-band, so we expect that the number of variables will more than double. We are also planning an expansion of ASAS to the northern hemisphere, to fully cover the whole sky.
{width="\linewidth"}
The results presented in this paper are not final in any sense. Please note, that all data are public domain. A different classification of binaries can be readily done by whoever feels like verifying and/or correcting our presentation.
This is an observational paper, but section 2 gives a short introduction to the main ideas about the structure and evolution of contact binaries. Section 3 gives the information about ASAS data, the classification scheme, and some examples of a diversity of light curves. Section 4 provides simple statistics of ASAS binaries. Finally, in section 5 we make a somewhat speculative discussion based on this observational paper.
Introduction to contact binaries {#sect:intro2}
================================
Contact binaries, also known as W UMa stars, are in a physical contact, with continuously changing brightness because of large tidal distortion of the two components.
The first theoretical milestone in the understanding of contact binaries was due to Lucy (1968a,b), who proposed that the two components share common envelope with the same entropy, thereby making the effective temperature almost constant over the surface of the two stars. As contact binaries have the mass ratio distinctly different from one, most nuclear energy is generated in the more massive component and it is redistributed around the whole surface through a moderately thick convective envelope.
The second theoretical milestone was the recognition of the consequences of the fact that the mass - radius relation for Zero Age Main Sequence (ZAMS) stars is much steeper than for the two Roche lobes. There can be no stable equilibrium between the two stars with a common envelope. The system evolves through a sequence of relaxation oscillations, with the mass flowing from star A to B, next from B to A, etc. (Flannery 1976, Lucy 1976, Robertson and Eggleton 1977). The cycle repeats on a thermal (Kelvin-Helmholtz) time scale. According to thermal relaxation model the binary oscillates between thermal contact, with the two eclipses of almost equal depth, and a semidetached phase in which one eclipse is much deeper than another.
Hazlehurst (1970) suggested that nuclear evolution of the primary component of a contact binary affects its structure. Stepień (2003, 2005) suggested that the currently more massive primary was originally the less massive of the two. The nuclear burning formed a small helium core, the star expanded and transferred mass to the original secondary. In analogy with Algol systems the currently more massive component is the less evolved, while the present secondary has a small helium core, and it is more advanced in its nuclear evolution.
It is interesting that there is a controversy about thermal relaxation oscillations in W UMa systems. Some authors claim there are no such oscillations (Webbink 2003), while others claim that such oscillations exist (Quin 2003, Yakut and Eggleton 2005, and references therein). ASAS statistics resolves this controversy on purely observational grounds.
ASAS data {#sect:statistics}
=========
Close binaries with a deep common envelope are in thermal contact and they have eclipses of almost equal depth. If the contact is shallow, or if there is no physical contact, then the effective temperatures of the two stars are different, and the two eclipses have different depth.
Theoretical models of relaxation oscillations indicate that the radii of the two components change relatively little throughout the cycle (Flannery 1976, Lucy 1976, Robertson and Eggleton 1977, Yakut and Eggleton 2005, and references therein). With the geometry of the two stars almost unchanged, tidal distortions due to geometry remain almost the same, and the most profound difference in the light curve is the relative depth of the two eclipses.
{width="\linewidth"}
For the purpose of this paper the classification of eclipsing binaries was done by decomposing their light curves into Fourier coefficients: ${\rm a_1, a_2, a_3, a_4 }$, following Ruciński (1997a,b, 1998, 2002) and Pojmański (2002). The classification is shown in Fig. 4 in the ${\rm a_2 - a_4 }$ plane, with the EC, ESD, and ED binaries shown with different symbols: ESD stars have smaller marks than EC and ED. Some examples of bright binaries of EC, ESD, and ED type are shown in Fig. 5. The parameters of these 15 binaries are listed in Table 1. The type of an eclipsing binary, EC, ESD, or ED, is indicated at the upper right corner of the light curve. The complete version of Table 1, for all eclipsing binaries used in this paper, and with all eight Fourier parameters, is in the file [Fourier.E]{} on the web page:
http://www.astrouw.edu.pl/\~gp/asas/asas\_paper\_data.html
http://archive.princeton.edu/\~asas/asas\_paper\_data.html
where all the data on which this paper is based are available electronically. These include all photomertic measurements in compressed files: ec.tgz (49MB), esd.tgz (27MB), ed.tgz (27MB).
[ccc]{} & &\
& &\
& &\
& &\
& &\
Some investigators (e.g. P. Eggleton) strongly prefer classification of close binaries into EW, EB, and EA types, on the grounds that photometry alone cannot provide unique geometry. Our classification into EC, ESD, and ED types should be considered preliminary, as we have only single band photometry and no spectroscopic information for thousands of our binaries.
ASAS ID P$[days]$ $a_{1}$ $a_{2}$ $a_{3}$ $a_{4}$ $b_{1}$
--------------- ----------- --------- --------- --------- --------- ---------
125340-5010.6 0.4047 -0.003 -0.158 -0.003 -0.028 -0.003
111621-5427.2 0.4971 -0.026 -0.123 -0.012 -0.020 -0.000
220425-0603.4 0.7273 -0.062 -0.139 -0.034 -0.027 0.021
181437-1324.0 1.3992 -0.001 -0.110 -0.004 -0.054 0.004
125028-6039.8 2.0439 -0.030 -0.085 -0.010 -0.031 -0.001
162959-4515.5 1.1376 -0.068 -0.155 -0.032 -0.049 0.001
072345-6500.6 3.1240 -0.006 -0.066 -0.003 -0.060 -0.001
185503-3359.5 1.1678 -0.019 -0.070 -0.015 -0.060 0.002
175924-0954.7 2.3438 -0.115 -0.119 -0.090 -0.088 0.006
123140-6237.3 1.9307 -0.009 -0.184 -0.004 -0.051 0.001
172116-3759.2 5.2056 -0.021 -0.135 -0.005 -0.033 0.007
012917-7243.4 180.60 -0.050 -0.190 0.016 -0.052 0.041
051700-5555.4 0.7904 -0.059 -0.126 -0.028 -0.021 0.020
105554-4349.6 1.1541 -0.034 -0.185 -0.019 -0.030 0.037
132831-4107.9 16.515 -0.010 -0.069 -0.011 0.003 -0.015
: The most important Fourier parameters for light curves from Fig. 5 are listed along with orbital periods. Each row is separated with a horizontal line. The $a_{2}$ parameter can be easily translated to the amplitude, $a_{1}$ and $a_{3}$ determine the difference between the two eclipses, and $b_{1}$ is related to the difference between the two maxima. Full Table 1 is given electronically as file [Fourier.E]{} on our web page. \[tabelka\]
A sample of the catalog of ASAS eclipsing binaries is shown in Table 2. The full table has the information for all 11,099 binaries, and can be found on our web page, in the file [Vars.E]{}.
----------------- ---------- ------------- ---- ---------- ----------- --------- ------- -------- ----------- -------
ID RA DEC T0 V Amp Class Other Other
- (2000) (2000) 2450000+ $[mag]$ $[mag]$ - ID Class
065819+1028.4 06:58:19 +10:28:24 0 447763 2387.63 11.04 0.28 EC
070017+0202.2 07:00:17 +02:02:12 1 397550 1873.07 10.53 0.57 EC V0460 Mon EB/KE
211201$-$2003.2 21:12:01 $-$20:03:12 0 352202 1872.84 13.22 0.34 EC/ESD
025535$-$0219.9 02:55:35 $-$02:19:54 0 79274 1920.00 11.59 0.54 ESD
095930$-$6440.1 09:59:30 $-$64:40:06 2 2428 1869.42 11.48 1.12 ESD
202110$-$4333.9 20:21:10 $-$43:33:54 1 00218 1876.053 12.08 0.71 ESD/ED V2265 Sgr EA
001855$-$7954.9 00:18:55 $-$79:54:54 0 90310 1870.45 11.28 0.77 ED
114500$-$7247.6 11:45:00 $-$72:47:36 22 23037 1895.60 11.15 0.09 ED
235052$-$2316.7 23:50:52 $-$23:16:42 1 4023 1871.06 9.42 0.29 ED
----------------- ---------- ------------- ---- ---------- ----------- --------- ------- -------- ----------- -------
A subset of all ASAS results related to binary stars was ‘frozen’ for the epoch of this paper. ASAS is an on-going project, with more data added every clear night. Therefore, as time goes on, the volume of data related to eclipsing binaries will increase, and the quality will improve. To make the results presented in this paper reproducible we decided to provide ‘frozen’ data.
Note: all light curves presented in Fig. 5 are for bright variables discovered with ASAS survey. They were not known before. Just as expected, the majority of ASAS variables are new discoveries (Paczyński 1997).
Statistics of contact binaries {#sect:statistics}
==============================
All the statistics of this section is based on data provided in the file [Vars.E]{}, which can be found on our web page. Table 2 is a sample of this file. In particular, the classification of ASAS binaries is given there.
The Galactic distribution of ASAS eclipsing binaries shows a significant difference, with detached binaries most strongly concentrated to the galactic plane, while the short period contact binaries have almost isotropic distribution. It is well known that there is a period - luminosity relation for W UMa stars (Rucinski 1996, Klagyivik and Csizmadia 2004), and the Galactic distribution of those stars is not surprising. The strong concentration of detached binaries to the Galactic plane implies these are even more luminous and massive, on average.
The binary period distribution is shown in Fig. 6 for stars with the Galactic latitude ${\rm |b| > 30^{\circ}}$ to reduce the luminosity bias. The distribution of contact systems (EC) peaks near 0.37 days, it has a sharp cut-off at 0.2 days, and a long tail extending beyond 1 day. In fact the tail extends to more than one hundred days, as it is apparent in Fig. 8 and Fig. 10. Also shown is the distribution of orbital periods of semi - detached (ESD) and detached (ED) binaries.
{width="\linewidth"}
The important topic ot this paper is the presentation of eclipse depth of EC binaries, as this is directly related to the issue of relaxation oscillations. For contact binaries the depth of both eclipses is directly related to the Fourier coefficients of the light curve. The following analytical formulae approximate these relations with better then $5\%$ accuracy for the majority of stars: $${\rm D_{p} = -2 \left( 1.2a_{2} - 2a_{2}^2 + (a_{1}+a_{3}) \right) \hskip 1.0cm [mag] } ,
\eqno(1)$$ $${\rm D_{s} = -2 \left( 1.2a_{2} - 2a_{2}^2 - (a_{1}+a_{3}) \right) \hskip 1.0cm [mag] } ,
\eqno(2)$$ where ${\rm D_p }$ and ${\rm D_s }$ correspond to the depth of the primary and secondary minimum, respectively. Three examples are presented in Fig. 7.
[c]{}\
\
\
{width="\linewidth"}
The distribution of primary eclipse depth with binary period is shown in Fig. 8. A histogram of primary eclipse depth integrated over binary period is shown in Fig. 9. Note that while the vast majority of W UMa binaries have periods in the range ${\rm 0.2 < P < 1.2 }$ days, there are contact binaries with periods over 100 days, as it is apparent in Fig. 5 and Fig. 8.
{width="\linewidth"}
The eclipse depth ratio is shown in Fig. 10 for EC binaries, and in Fig. 11 for EC and ESD binaries. These are important figures. They demonstrate that some EC binaries are in good thermal contact, as the depth of secondary eclipse is almost the same as the primary. But many other EC binaries have the two eclipses which are very unequal, implying that they are not in thermal contact, yet their geometry is close to that expected for a contact system. Finally, there are ESD binaries, in which the geometry is very different, and of course the two eclipses have usually different depth. Qualitatively, this is just what was expected in the models with relaxation oscillations (cf. Lucy 1976, Flannery 1976, Robertson and Eggleton 1977, Yakut and Eggleton 2005). There is no shortage of stars without thermal contact. Combining ASAS data with models of relaxation oscillations should provide quantitative verification of the theory, but this task is beyond the scope of our paper.
{width="\linewidth"}
{width="\linewidth"}
Contact binaries often have maxima of different height, with the maximum following the primary (i.e. deeper) eclipse being either higher or lower than the maximum following the secondary eclipse. The distribution of the difference is shown in Fig. 12. The positive value of the ${\rm b1/|a2| } $ parameter indicates that the maximum following the primary eclipse is brighter of the two. The asymmetry in distribution, i.e. the excess of positive values of ${\rm b1/|a2| }$, is known as the O’Connell (1951) effect (cf. also Ruciński 1997b): the maximum following the primary eclipse is on average brighter, presumably due to stream of gas flowing between the two stars. An extreme case of this phenomenon is shown in V361 Lyr (Kaluzny 1991).
{width="\linewidth"}
Discussion {#sect:discuss}
==========
Our conclusion, based on the distribution of eclipse depths (Fig. 10, 11), is that relaxation oscillations, first proposed by Lucy and Flannery, are real. There is no shortage of binaries corresponding to no thermal contact, with a very different depth of the two eclipses. Model calculations of the type recently done by Yakut and Eggleton (2005), when combined with ASAS data, should allow a quantitative verification of the theory.
A very large number of contact or near contact systems reveal features never noticed before: the distribution of eclipse depth ratios as shown in Fig. 10 and Fig. 11 has a distinct break around ${\rm log P \approx -0.4 }$ and ${\rm D_s/D_p \approx 0.7 }$. We do not speculate on the origin of this feature, but we bring this break to the attention of our readers. This break is best seen in systems with the most robust classification.
The most surprising result of this paper is presented in Fig. 6. A traditional view for the origin of W UMa contact binaries is to assume that they come from detached binaries of comparable periods. For the first time we have approximately complete statistics of binaries of all types with orbital periods shorter than one day, and there are very few detached binaries. Obviously, they are more difficult to find than either contact or semi - detached systems. As times goes on the statistics of ASAS detached binaries will improve, and a statistical analysis will tell us if there is a problem with the origin of W UMa stars. At this time the contact systems seem to appear “out of nowhere”.
In about a year we shall have I-band data for our eclipsing binaries. This will make it possible to be more quantitative about the distribution properties, including the distribution in our Galaxy. Also, it will be much easier to quantify the impression about a shortage of detached binaries and the space density of systems of different types: EC, ESD, and ED. At this time it is premature for us to speculate about the outcome of binary statistics while we wait for the I-band data.
While this is an observational paper, written to promote the usefulness of the ASAS catalog of variable stars, we are tempted to speculate about possible interpretation of Fig. 6, which was so surprising to us. We are not consistent with the previous paragraph, but the temptation is hard to resist. The following is a speculative hypothesis.
There has been a gradual emergence of the notion that contact systems have companions (Ruciński and Kaluzny 1982, Chambliss 1992, Hendry and Mochnacki 1998, also Tokovinin 2004). Recently. Pribulla and Ruciński (2005) found that up to 50% of W UMa binaries have companions. This opens up a possibility that Kozai (1962) cycle operates in some such triples, as suggested by Kiseleva, Eggleton, and Mikkola (1998), and by Eggleton and Kiseleva-Eggleton (2001). The mechanical three-body orbital evolution, with a large range of inclinations of the two orbits, inner and outer, may induce large variations in the eccentricity of the inner binary. In time the eccentricity may become large enough to make a contact system out of the inner binary.
In this scenario the inner binary has the initial period that is relatively long, and in most cases no eclipses would be detectable. During a Kozai cycle the inner binary occasionally has a chance to reach a physical contact: it may either merge forming a single star, or it may become a W UMa star. This process is somewhat similar to the model of formation of close binaries in globular clusters (Pooley et al. 2003, and references therein).
The surprisingly small number of short period detached eclipsing binaries as seen in Fig. 6 may indicate that the Kozai cycle is not just a curiosity, but it may be an important channel for forming W UMa stars. This possibility cannot be rejected without careful analysis. After all W UMa stars are rare, with the local space density of just 0.2% of the main sequence stars (Rucinski 2002, and references therein).
Acknowledgements {#acknowledgements .unnumbered}
================
We are very grateful to P. P. Eggleton, D. Fabrycky, S. Ruciński and K. Stepień for many helpful discussions.
The ASAS project was supported to a large extent by the generous donation of William Golden. This work was partially supported by the Polish MNiI grant no. 1P03D-008-27.
Chambliss, C. R. 1992, PASP, 104, 663
Eggleton, P. P., Kiseleva, L. 2001, ApJ, 562, 1012
Flannery, B. P. 1976, ApJ, 205, 217 Hazlehurst, J. 1970, MNRAS, 149, 129
Hendry, P. D., Mochnacki, S. W. 1998, ApJ, 504, 978
Kaluzny, J. 1991, AcA, 41, 17 Klagyivik, P., and Csizmadia, Sz. 2004, PADEU, 14, 303
Kiseleva, L. G., Eggleton, P. P., Mikkola, 1998, MNRAS, 300, 292
Kozai, Y. 1962, AJ, 67, 579
Lucy, L. B. 1968a, ApJ, 151, 1123 Lucy, L. B. 1968b, ApJ, 153, 877 Lucy, L. B. 1976, ApJ, 205, 208 O’Connell, D. J. K. 1951, Riverview Pub. 2, 85
Paczyński, B. 1997, vsar.conf, 357
Pojmański, G. 1997, AcA, 47, 467
Pojmański, G. 1998, AcA, 48, 35
Pojmański, G. 2000, AcA, 50, 177
Pojmański, G. 2002, AcA, 52, 397
Pojmański, G. 2003, AcA, 53, 341
Pojmański, G., and Maciejewski, G. 2004, AcA, 54, 153
Pojmański, G., and Maciejewski, G. 2005, AcA, 55, 97
Pojmański, G., Pilecki, B., Szczygiel, D. 2005, astro-ph/050817
Pooley, D. et al. 2003, ApJL, 591, L131
Pribulla, T., Rucinski, S. M. 2005, private communication
Quin, S. 2003, MNRAS, 342, 1260
Robertson, J. A., and Eggleton, P. P. 1977, MNRAS, 179, 359
Rucinski, S. M. 1996, ASPC, 90, 270
Rucinski, S. M. 1997a, AJ, 113, 407
Rucinski, S. M. 1997b, AJ, 113, 1112
Rucinski, S. M. 1998, AJ, 115, 1135
Rucinski, S. 2002, PASP, 114, 1124
Rucinski, S. M., Kaluzny, J. 1982, Ap&SS, 88, 433
Stepień, K. 2003, IAUS 219, 134 Stepień, K. 2005, astro-ph/0510464, submitted to A&A Tokovinin, A. 2004, RMxAC, 21, 7
Webbink, R. F. 2003, ASPC, 293, 76
Yakut, K., Eggleton, P. P. 2005, ApJ, 629, 1055
\[lastpage\]
[^1]: e-mail: [email protected]
[^2]: e-mail: dszczyg, pilecki, [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
author:
- |
G. Maciejewski[^1]\
Centrum Astronomii UMK, ul. Gagarina 11, 87-100 Toruń\
Astrophysikalisches Institut und Universitäts-Sternwarte, Schillergässchen 2-3, D-07745 Jena, Germany\
B. Mihov, Ts. Georgiev\
Institute of Astronomy, Bulgarian Academy of Sciences, 72 Tsarigradsko Chausse Blvd., 1784 Sofia, Bulgaria
title: 'The open cluster Berkeley 53[^2]'
---
*Abstract*: We present a photometric study of the neglected open cluster Berkeley 53. We derived its fundamental parameters, such as the age, the interstellar reddening, and the distance from the Sun, based on $BV$ photometry combined with near-infrared $JHK_{\rm{S}}$ data. The structure and the mass function of the cluster were also studied and the total number of members and the total mass were estimated. The cluster was found to be a rich and massive stellar system, located in the Perseus Arm of the Milky Way, $3.1\pm0.1$ kpc from the Sun. Its age exceeds 1 Gy but it seems to be very young in the context of its dynamical evolution. The analysis of the two-color diagrams and color-magnitude diagrams indicates that the cluster is significantly reddened. However, both methods resulted in different values of $E(B-V)$, i.e. $1.21\pm0.04$ and $1.52\pm0.01$, respectively. This discrepancy suggests the presence of an abnormal interstellar extinction law toward the cluster.
*Keywords*: open clusters and associations: individual: Berkeley 53
Introduction
============
The open cluster Berkeley 53 (C 2055+508) was discovered by Setteducati & Weaver (1960). Ruprecht (1966) classified it as a poor, concentrated open cluster of Trumpler type II3p. According to the *New catalogue of optically visible open clusters and candidates* by Dias et al. (2002), the cluster’s apparent diameter is $12'$ and Trumpler type is III2m. No dedicated studies of this object have been performed to date.
Berkeley 53 consists of stars fainter than $V=18$ mag and is located in the vicinity ($3.\!'1$ from the cluster center) of the bright ($V=6.6$ mag) foreground star HD 199578. This makes photometric observations of the cluster difficult.
In this paper we present a photometric study of Berkeley 53 resulting in the determination of its basic parameters, such as the age, the interstellar reddening, and the distance from the Sun. The structure and the mass function of the cluster are also studied and the total number of members and the total mass are estimated.
Observation and data reduction
==============================
Observations were performed with the 2/16 m Ritchey-Chrétien telescope of the Rozhen National Astronomical Observatory (NAO, Bulgaria), operated by the Institute of Astronomy, Bulgarian Academy of Sciences. The instrument was used in a direct imaging mode and was equipped with a Princeton Instruments VersArray:1300B CCD camera mounted in the Ritchey-Chrétien focus. The field of view was $5.\!'8 \times 5.\!'6$ with a scale of $0.26$ arcsec per pixel. Observations were carried out on August 19, 2007. Two exposures in the $B$ filter and three frames in the $V$ band were acquired. The exposure time was 600 s and 300 s, respectively. Four exposures in the $U$ filter with 900 s exposure time were also acquired but they occurred to be not deep enough to detect the cluster stars.
The field around Berkeley 53 is presented in Fig. \[rys1\] where the fragment of the sky covered by our observations is also sketched. The telescope was not pointed on the cluster center due to the nearby bright star HD 199578. The $B$ band exposures were affected by a narrow strip of its reflected light passing horizontally through the center of frames.
CCD frames were processed using a standard procedure that included subtraction of bias frame, flat-fielding with twilight flats, aperture and point spread function (PSF) photometry, transformation to the standard system, and astrometric calibration. The aperture photometry and astrometric calibration were performed with the software pipeline developed for the Semi-Automatic Variability Search sky survey (Niedzielski et al. 2003). Exposures in a given band were averaged. The PSF magnitudes were obtained with IRAF[^3] package DAOPHOT. A second-order variable PSF was used to compensate for variability of stellar profiles across frames. Aperture corrections were determined using aperture photometry of 13 isolated stars that were used while building the PSF profile.
![A $ 9'$ $\times$ $9'$ field around Berkeley 53. The $5.\!'8$ $\times$ $5.\!'6$ field of view of the telescope is marked with a square frame. The redetermined center of the cluster is marked with a cross. The core radius boundary is also sketched. The bright star at the left edge of the image is HD 199578. The image was taken from the Digitized Sky Surveys (DSS). North is up, East to the left.[]{data-label="rys1"}](gm-fig-field.eps){width="8.3cm"}
The raw instrumental magnitudes were corrected for the scattered light effect (Markov 2008). The following correction formulae were obtained by us: $$\Delta b = (0.122\pm0.006)\,d^2 \, , \;$$ $$\Delta v = (0.106\pm0.004)\,d^2 \, , \;$$ where $\Delta b$, $\Delta v$ are the additive magnitude corrections in the corresponding filters and $d$ is the distance of a star from the CCD matrix center normalized by the detector half-size.
The calibration coefficients that transform instrumental magnitudes into standard ones were determined using 16 stars from the Messier 92 field (Majewski et al. 1994, see their Table 1) and 40 stars from the field of NGC 7790 (Odewahn et al. 1992). The magnitude range was between $13.144$ and $18.315$ mag in $V$ and the $(B-V)$ color index coverage was in the range between $-0.111$ and $1.918$ mag. The following equations were derived: $$b - B = (0.58\pm0.03) X - (0.17\pm0.01)(B-V)-23.06 \, , \;$$ $$v - V = (0.33\pm0.02) X - (0.16\pm0.01)(B-V)-23.67\, , \;$$ where $b$, $v$ are the corrected instrumental magnitudes, $B$, $V$ are the magnitudes in the standard system, and $X$ is the airmass.
The final list of stars contains equatorial coordinates, $V$ magnitudes, and $(B-V)$ color indices. It is available in electronic form at the survey web site[^4] and the WEBDA[^5] database (Mermilliod 1996).
The optical data were complemented with near-infrared $JHK_{\mathrm{S}}$ photometry extracted from the 2-Micron All Sky Survey (2MASS, Strutskie et al. 2006). The extraction radius was set to $30'$ around the cluster center (see Sect. 3.1). Optical photometry was combined with near-infrared data to perform a comprehensive photometric study of the cluster (e.g. Maciejewski 2008).
Data analysis and results
=========================
Cluster structure
-----------------
![The radial density profile. The solid line denotes the fitted density distribution while the dashed one marks the level of the stellar background density.[]{data-label="rys2"}](gm-fig-rdp.eps){width="7cm"}
The cluster structure was investigated with the radial density profile (RDP). For this purpose only the 2MASS data set was used for constructing RDP due to unlimited field of view. We started with redetermining the cluster center. The algorithm was adopted from Maciejewski & Niedzielski (2007) and started from a tentative position determined by eye. Two perpendicular stripes were cut along declination and right ascension starting from the approximate cluster center and the histogram of star counts was built along each stripe. The bin with the maximum value in both coordinates was taken as a new cluster center what resulted in $\mathrm{RA}=20^{\rm{h}}55^{\rm m}56^{\rm s}$ and $\mathrm{DEC}=+51^{\circ}02.\!'8$ for epoch J2000.0 ($l=90.\!\!^{\circ}29$, $b=3.\!\!^{\circ}75$). As one can see in Fig. \[rys1\], these values do not point to the center of the field of view. Next, the profiles were constructed by counting stars inside concentric rings of width $1'$, centered at the redetermined cluster center. The density uncertainty in each ring was estimated assuming Poisson statistics. The RDP is plotted in Fig. \[rys2\].
To parametrize the density distribution, a two-parameter King (1966) density profile was fitted with the least-squares method in which the uncertainties were used as weights. We derived the core radius (the distance where the stellar density drops to half its maximum value) $r_{\mathrm{c}}=2.5\pm0.1$ arcmin, the central density $f_{0}=18.0\pm0.5$ stars/arcmin$^{2}$, and the background density level $f_{\mathrm{bg}}=6.9\pm0.1$ stars/arcmin$^{2}$. The fitted profile is sketched with a solid line in Fig. 2 while the dashed line marks the level of $f_{\mathrm{bg}}$. The cluster limiting radius $r_{\mathrm{lim}}$ was roughly estimated by eye-inspection in the RDP. The stellar density excess is visible up to at least $11'$ – a value almost two times greater than the literature one.
The RDP allowed us to estimate number of observed stars belonging to the cluster to be $\sim$900. That suggests that Berkeley 53 is a very rich and massive stellar system.
Two-color diagrams
------------------
![$(H-K_{\mathrm{S}})$ $vs.$ $(B-V)$ and $(J-H)$ $vs.$ $(H-K_{\mathrm{S}})$ diagrams constructed for the stars located within the core radius from the cluster center. The stars that form the main sequence of the cluster are marked with filled circles. The main sequence, fitted by shifting along the reddening vector, is sketched with a continuous line in each diagram.[]{data-label="rys3"}](gm-fig-tcd.eps){width="10.3cm"}
$(H-K_{\mathrm{S}})$ $vs.$ $(B-V)$ and $(J-H)$ $vs.$ $(H-K_{\mathrm{S}})$ two color diagrams (TCDs) were constructed to estimate the interstellar reddening toward the cluster. Only stars located within the core radius from the cluster center were considered to minimize the influence of the background star contamination. Among them only stars forming the main sequence of the cluster were taken for the further analysis. They are marked with filled circles in Fig. \[rys3\] where the TCDs are plotted.
The theoretical main sequence was extracted from the Padova isochrones for solar metallicity $Z=0.019$ (Giraldi at al. 2002). The reddenings $E(B-V)=1.18$ and $E(H-K_{\mathrm{S}})=0.26$ were obtained by shifting the main sequence (a continuous line in Fig. 3) along the reddening vectors whose normal slopes were calculated assuming the universal interstellar extinction law by Schlegel et al. (1998). To obtain an independent determination of the color excess in $(B-V)$ color index, the value of $E(H-K_{\mathrm{S}})$ was transformed into $E(B-V)$ applying the relation $\frac{E(H-K_{\mathrm{S}})}{E(B-V)} = 0.209$ taken from Schlegel et al. (1998). We derived $E'(B-V)=1.24$. The results seem to be consistent with a mean value of $\left\langle {E}(B-V) \right\rangle =1.21\pm0.04$.
Color-magnitude diagrams
------------------------
A preliminary $J$ $vs.$ $(J-K_{\mathrm{S}})$ color-magnitude diagram (CMD) for overall cluster region ($r<r_{\mathrm{lim}}$) is presented in Fig. \[rys4\]. We applied a decontamination procedure to remove background-star contamination. The details of this procedure can be found in Maciejewski & Niedzielski (2007). The CMDs were built for the cluster region and for an offset field. A concentric offset field of width $10'$ and starting at $r=r_{\rm{lim}}+2'$ from the cluster center was used. Then the CMDs were divided into two-dimensional bins and the number of stars within each box was counted. The cleaned (decontaminated) cluster CMD was built by subtracting the number of stars in the offset box from the number of stars in the corresponding cluster box. The latter number was weighted by the cluster to offset field area ratio. Knowing the number of cluster stars occupying each box, the algorithm randomly chose the required number of stars located in the cluster area and with the adequate magnitude and color index.
As one can see in Fig. \[rys4\], a rich main sequence is clearly visible, as well as a red giant clump. The estimated number of observed cluster stars is $\sim$1000 after rejecting stars with outstanding magnitudes and colors. This value is comparable with the one obtained from the RDP analysis. We found $\sim$200 evolved stars constituting the red clump.
![The near-IR color-magnitude diagram constructed for the overall ($r<r_{\rm{lim}}$) area of Berkeley 53, after running the decontamination procedure. A morphology typical for a rich old stellar cluster is clearly visible.[]{data-label="rys4"}](gm-fig-morph.eps){width="7cm"}
To determine fundamental astrophysical parameters of the cluster, we built 8 CMDs combining $V$ and $J$ magnitudes with $(B-V)$, $(V-J)$, $(V-H)$, and $(V-K_{\mathrm{S}})$ color indices. The broad color baselines were expected to minimize photometric errors and to determine precisely the reddening. Only stars located within the core radius around the redetermined cluster center were considered to minimize the influence of the background star contamination on further analysis. The diagrams are plotted in Fig. \[rys5\].
{width="14cm"}
Initially, we tried to derive the distance modulus and the age of the cluster via fitting a set of theoretical Padova isochrones (Giraldi at al. 2002). Since there is no information about the cluster metallicity, the solar metallicity of $Z=0.019$ was assumed. The fitting algorithm is based on the least-squares method and uses stellar magnitudes as weights. Stars with extremal color indices or magnitudes were rejected manually before running the isochrone-fitting procedure. The fitting and decontamination procedures were run independently for each diagram with a fixed value of reddening. The latter quantity was calculated for a given color using $\left\langle {E}(B-V) \right\rangle$ from the two-color diagram analysis and assuming the universal interstellar extinction law by Schlegel et al. (1998). However, we were unable to find a convergent and satisfactory final solution and the reddening was found to be underestimated.
To explain this discrepancy, we ran the isochrone-fitting algorithm with reddening as a third free parameter. The results of individual fits are collected in Table 1 and the respective isochrones are drawn with solid lines in Fig. 5. The dereddened distance modulus $(m-M)_{0}$ was calculated assuming the universal interstellar extinction law by Schlegel et al. (1998) with a total-to-selective absorption ratio of $R=3.315$. The value of $E(B-V)$ was found to be about 0.3 mag greater than the two-color diagram analysis suggested. One can note that $E(B-V)$ tends to be slightly smaller for colors including more reddish bands. We obtained $\frac{E(V-J)}{E(B-V)}=2.414\pm0.025$, $\frac{E(V-H)}{E(B-V)}=2.707\pm0.017$, and $\frac{E(V-K_{\rm{S}})}{E(B-V)}=2.893\pm0.023$. Comparing to the values given by Schlegel et al. (1998), i.e. 2.413, 2.739, and 2.948, respectively, a discrepancy can be noted in bands $H$ and $K_{\rm{S}}$ and increasing with wavelength. This effect explains the lower value of the interstellar reddening obtained from the TCDs analysis. Adopting redetermined reddening vectors, we obtained consistent results of TCDs and CMDs analysis. In the absence of photometric observations in the $U$ band, it is impossible to answer the question about the nature of the observed effect. It may be caused by an abnormal interstellar extinction law toward Berkeley 53 or by systematics in 2MASS photometry.
------------------------ ------------- -------- --------- ---------- -----------
Diagram $\log(age)$ $E$ $(m-M)$ $E(B-V)$ $(m-M)_0$
(mag) (mag) (mag) (mag)
$V~vs.~(B-V)$ $9.05$ $1.54$ $17.40$ $1.54$ $12.35$
$V~vs.~(V-J)$ $9.10$ $3.72$ $17.53$ $1.54$ $12.48$
$V~vs.~(V-H)$ $9.10$ $4.15$ $17.49$ $1.52$ $12.44$
$V~vs.~(V-K_{\rm{S}})$ $9.10$ $4.45$ $17.52$ $1.51$ $12.47$
$J~vs.~(B-V)$ $9.10$ $1.53$ $13.76$ $1.53$ $12.39$
$J~vs.~(V-J)$ $9.10$ $3.69$ $13.84$ $1.53$ $12.47$
$J~vs.~(V-H)$ $9.10$ $4.16$ $13.88$ $1.52$ $12.51$
$J~vs.~(V-K_{\rm{S}})$ $9.10$ $4.43$ $13.87$ $1.50$ $12.50$
------------------------ ------------- -------- --------- ---------- -----------
Finally, the following mean results were obtained from the CDMs analysis: $\log(age)=9.09\pm0.02$, $E(B-V)=1.52\pm0.01$, $(m-M)_0=12.45\pm0.05$, and a distance of $3.1\pm0.1$ kpc. The linear diameter was found to be $19.8\pm0.5$ pc. The main possible cause of systematic error in the values given above is the unknown metallicity of the cluster. To estimate its influence on the results, we repeated the isochrone-fitting procedure for the super- and sub-solar metallicities of $Z=0.030$, $Z=0.008$, and $Z=0.004$. We noticed the increase of the reddening with the decrease of the metallicity ($\Delta E(B-V) \approx 0.07$ for each metallicity step, including $Z=0.019$) while the age and the apparent distance modulus remained stable. It is also worth noting that the discrepancy between the TCDs and CMDs values of the reddening cannot be justified by non-solar metallicity of the cluster.
Total mass and number of members
--------------------------------
Studies of the mass function (MF) were carried out to estimate the total mass of the cluster and the number of its members. The analysis is based on the 2MASS photometry due to the wide field of view available and uses the algorithm adopted from Maciejewski & Niedzielski (2007). The first step was to build a luminosity function (LF) for overall ($r<r_{\rm{lim}}$) cluster region. The bright end of the LF was determined by the main-sequence turn-off point while the faint end was set for $J=15.8$ mag – the value of the 99.9% Point Source Catalogue Completeness Limit[^6]. We used bins as small as 0.1 mag due to cluster richness. Another LF was built for a concentric offset field of width $10'$ and starting at $r=r_{\rm{lim}}+2'$. The LF of the offset field was subtracted, bin by bin, from the cluster LF, taking the area proportion into account. The resulting LF was converted into an MF using the respective isochrone. It is woth noting that the cluster area, especially the core ($r<r_{\rm{c}}$), may suffer significant incompletness due to stellar crowding. Therefore, our further results must be treated as a lower limit.
![The mass function derived for the main-sequence stars of Berkeley 53. []{data-label="rys6"}](gm-fig-mf.eps){width="9.0cm"}
The mass function $\phi(m)$ was approximated by a standard relation of the form $\log\phi(m)=-(1+\chi)\log{m}+b_{0}$ where $m$ is the stellar mass, $\chi$ is the mass-function slope, and $b_{0}$ is a constant. The derived parameters allowed us to estimate the total mass $M_{\rm{tot}}$ and the total number of stars $N_{\rm{tot}}$. These quantities were calculated extrapolating the MF from the main-sequence turn-off point down to the H-burning mass limit of 0.08 $M_{\odot}$ (see Maciejewski & Niedzielski 2007 and references therein for details). The mass function is plotted in Fig. \[rys6\]. We obtained $\chi=1.7\pm1.6$, a value comparable within error bars to the universal initial mass function (IMF) $\chi_{\rm{IMF}}=1.3 \pm 0.3$ given by Kroupa (2001). The high uncertainty is caused by a small range of covered stellar masses. The cluster was found to be in fact very rich and massive with $N_{\rm{tot}}=31000$ and $M_{\rm{tot}}=12000$ $\rm{M}_{\odot}$.
To describe a state of cluster dynamic evolution, the dynamical-evolution parameter $\tau$ was calculated in the form $\tau=\frac{age}{t_{\rm{relax}}}$ where $t_{\rm{relax}}$ is the relaxation time (see Maciejewski & Niedzielski 2007 for details). We derived $\log\tau=-0.3$ which suggests that the cluster is dynamically younger than its relaxation time.
Summary
=======
Berkeley 53 was found to be a rich and massive open cluster belonging to the Perseus Arm. Its age exceeds 1 Gy, but it seems to be very young in the context of its dynamical evolution. A subtle discrepancy in the interstellar extinction law toward the cluster in the $H$ and $K_{\rm{S}}$ bands was detected. That suggests the presence of an abnormal interstellar extinction law toward the cluster.
*Acknowledgements*: We thank the referee for remarks that improved our paper. This research has made use of the WEBDA and SIMBAD data bases and is supported by UMK grant 411-A and the grant VU-NZ-01/06 of the Ministry of Education and Science of Bulgaria. GM acknowledge support from the EU in the FP6MC ToK project MTKD-CT-2006-042514. 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.
Dias, W.S., Alessi, B.S., Moitinho, A., Lepine, J.R.D.: 2002, A&A 389, 871 Giraldi, L., Bertelli, G., Bressan, A., et al.: 2002, A&A 391, 195 King, I.: 1966, AJ 71, 64 Kroupa, P.: 2001, MNRAS 322, 231 Maciejewski, G.: 2008, Acta Astron. 58, 389 Maciejewski, G., Niedzielski, A.: 2007, A&A 467, 1065 Majewski, S.R., Kron, R.G., Koo, D.C., Bershady, M.A.: 1994, PASP 106, 1258 Markov, H.: 2008, Bulgarian Astronomical Journal 10, 85 Mermilliod, J.C.: 1996, in The Origins, Evolution and Destinies of Binary Stars in Cluster, ed. E. F. Milone & J.-C. Mermilliod, ASP Conf. Ser. 90, 475 Niedzielski, A., Maciejewski, G., Czart, K.: 2003, Acta Astron. 53, 281 Odewahn, S.C., Bryja, C., Humphreys, R.M.: 1992, PASP 104, 553 Ruprecht, J.: 1966, BAICz 17, 33 Schlegel, D.J., Finkbeiner, D.P., Davis, M.: 1998, AJ 500, 525 Setteducati, A.F., Weaver, H.F.: 1960, in Newly Found Star Clusters, Radio Astronomy Laboratory, Berkeley Strutskie, M.F., Cutri, M.F., Stiening, R., et al.: 2006, AJ 131, 1163
[^1]: [email protected]
[^2]: This work is based on observations made with the 2-m telescope of the Rozhen National Astronomical Observatory, which is operated by the Institute of Astronomy, Bulgarian Academy of Sciences.
[^3]: 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.
[^4]: http://www.astri.uni.torun.pl/\~gm/OCS
[^5]: http://www.univie.ac.at/webda/
[^6]: Following the Level 1 Requirement, according to *Explanatory Supplement to the 2MASS All Sky Data Release and Extended Mission Products* (http://www.ipac.caltech.edu/2mass/releases/allsky/doc)
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'It is known that the fluctuations of suitable linear statistics of Haar distributed elements of the compact classical groups satisfy a central limit theorem. We show that if the corresponding test functions are sufficiently smooth, a rate of convergence of order almost $1/n$ can be obtained using a quantitative multivariate CLT for traces of powers that was recently proven using Stein’s method of exchangeable pairs.'
author:
- Christian Döbler
- Michael Stolz
title: A quantitative central limit theorem for linear statistics of random matrix eigenvalues
---
[^1]
Introduction
============
For $n \in \nn$ let $M_n$ denote a random $n \times n$ matrix, distributed according to Haar measure on one of the compact classical groups, i.e. the unitary, orthogonal, and (if $n$ is even) unitary symplectic group. Its eigenvalues $\lambda_{n1}, \ldots, \lambda_{nn}$ lie on the unit circle line $\torus$ of the complex plane $\cc$. Write $L_n = \ll_n (M_n)$ for the empirical measure $\frac{1}{n} \sum_{j=1}^n \delta_{\lambda_{nj}}$ of the $\lambda_{nj}$. It is well known (see [@MeckesMeckes12+] for a recent quantitative version) that, as $n \to \infty$, $L_n$ tends a.s. weakly to the uniform distribution on $\torus$. Furthermore, for suitable test functions $f: \torus \to \rr$, the fluctuations $n(L_n(f) - \erw L_n(f))$ have been proven to tend to a Gaussian limit as $n \to \infty$ (see, e.g., [@DiaconisEvansTAMS; @SoshnikovLocal; @Wieand02]). In fact, these results are part of a broader interest in fluctuations of linear spectral statistics of various random matrix ensembles, both from the physics (see the seminal paper [@CostinLebowitz]) and mathematics points of view (see, e.g., [@ChattFluct] and the references therein).\
For Haar distributed matrices from the compact classical groups, and in the special case that $f$ is a trigonometric polynomial, Johansson has obtained in [@JohannssonAnnMath97] an exponential rate of convergence to the Gaussian limit, using sophisticated analytic tools related to Szegö’s strong limit theorem for Toeplitz determinants. In this note we are concerned with the speed of convergence in the case of test functions whose Fourier expansion does not necessarily terminate. In this more general setting we have been unable to find quantitative convergence results in the literature. We point out that for sufficiently smooth test functions a rate close to $\on{O}(1/n)$ is an easy consequence of a recent quantitative multivariate CLT for a vector of traces of powers of a Haar distributed element of the compact classical groups, where, crucially, the length of the vector may grow with the matrix size $n$. This CLT was proven by the present authors in [@DoeSt11], using the “exchangeable pairs” version of Stein’s method (see [@CGSbook] for background) and building upon a construction of an exchangeable pair that had already been successfully used by Fulman [@Ful10] in the univariate case. Note that a different way of using Stein’s method to study the linear eigenvalue statistics of different random matrix ensembles was devised by Chatterjee in [@ChattFluct].\
In what follows we concentrate on the case of random unitary matrices. The orthogonal and symplectic cases can be treated along the same lines, leading to the same rates of convergence towards a (this time not necessarily centered) Gaussian limit. In Section \[setup\] we recall some Fourier analysis as well as the crucial quantitative multivariate CLT for traces of powers. In Section \[ratessmoothcase\] we state and prove our main result on linear statistics.
Set-up and background {#setup}
=====================
Let $f \in \on{L}^1(\torus)$ be real valued. We will view $f$ as a $2\pi$-periodic function on $\rr$ by tacitly identifying $f(e^{ix})$ with $f(x)$. For $j \in \zz$ the $j$-th Fourier coefficient of $f$ is defined as $$\hat{f}_j = \hat{f}(j) = \frac{1}{2\pi} \int_0^{2\pi} f(x) e^{-ijx}dx.$$ Note that since $f$ is real valued, $\hat{f}(-j) = \overline{\hat{f}(j)}.$ It is a well known fact about Fourier coefficients that smoothness of the function implies quantitive information on the decay of the coefficients:
\[coeffbound\] If $f \in \on{C}^k(\torus)$, then for all $0 \neq j \in \zz$ there holds $$|\hat{f}(j)| \le \frac{\|f^{(k)}\|_1}{ |j|^k}.$$
See, e.g., [@Katznelson I.4.4].
Slightly generalizing this, we will consider functions $f \in \on{L}^1(\torus)$ with the property that there exist $\kappa > 1$ and $C_{\kappa} > 0$ such that for all $0 \neq j \in \zz$ there holds $$\label{kappa} |\hat{f}(j)| \le \frac{C_{\kappa}}{ |j|^{\kappa}}.$$ Proposition \[coeffbound\] then says that for $f \in \on{C}^k(\torus)$ $(k \ge 2)$ condition holds with $\kappa = k$, and it follows from the results in [@GraClassical Sec. 3.2.2] that holds with $\kappa = 2$ if $f$ has a Lipschitz first derivative. Note that since $|e^{ijx}| = 1$ for all $j$ and $x$, our assumption $\kappa > 1$ implies that the Fourier expansion of $f$ will converge normally, hence (by compactness of the torus) uniformly to $f$. So we have that
$$\sum_{j \in \zz} \hat{f}(j) e^{ijx} = \hat{f}(0) + \sum_{j=1}^{\infty}\left( \hat{f}(j) e^{ijx}
+ \overline{\hat{f}(j)\ e^{ijx}}\right).$$
Let $M_n$ be a Haar distributed element of $\on{U}_n$ and write $$L_n = \ll_n(M_n) = \frac{1}{n} \sum_{j = 1}^n \delta_{\lambda_{nj}},$$ where $\lambda_{n1}, \ldots, \lambda_{nn} \in \torus$ are the eigenvalues (with multiplicities) of $M_n$. Observe that by Fubini’s theorem and the left and right invariance of Haar measure one has $$\begin{aligned}
\erw L_n(f) &= \frac{1}{n} \erw[f(\lambda_{n1}) + \ldots + f(\lambda_{nn})]\\
&= \frac{1}{n} \int_{\torus} \erw[f(t\lambda_{n1}) + \ldots + f(t\lambda_{nn})] dt \\
&= \frac{1}{n} \erw \int_{\torus} [f(t\lambda_{n1}) + \ldots + f(t\lambda_{nn})] dt \\
&= \frac{1}{n} \erw\left[ n \int_{\torus} f(t) dt \right] = \int_{\torus} f(t) dt = \hat{f}(0).\end{aligned}$$ In view of this, it follows from the above that the $n$-scaled fluctuation of $L_n(f)$ has a pointwise expression $$\begin{aligned}
\label{mainexpansion} n(L_n(f) - \erw(L_n(f))) &= f(\lambda_{n1}) + \ldots + f(\lambda_{nn}) - n \hat{f}(0)\nonumber \\
&= \sum_{j=1}^{\infty} \hat{f}(j) \on{Tr}(M_n^j) + \sum_{j=1}^{\infty} \overline{\hat{f}(j) \on{Tr}(M_n^j)}.\end{aligned}$$
The following quantitative CLT for vectors of traces of powers of Haar unitaries, proven in [@DoeSt11], will make it possible to control finite sections of $n$-dependent length of the expansion : Let $M = M_n$ be distributed according to Haar measure on $\on{U}_n$. For $d \in \nn,\
r = 1, \ldots, d$, consider the $r$-dimensional complex random vector $$W := W(d, r, n) := (\on{Tr}(M^{d-r+1}), \on{Tr}(M^{d-r+2}), \ldots, \on{Tr}(M^{d})).$$ Let $Z := (Z_{d-r+1} \ldots, Z_d)$ denote an $r$-dimensional complex standard normal random vector, i.e., there are iid real random variables $X_{d-r+1}, \ldots, X_d,$ $Y_{d-r+1}, \ldots, Y_d$ with distribution $\on{N}(0, 1/2)$ such that $Z_j = X_j + i Y_j$ for $j = d-r+1, \ldots, d$. We take $\Sigma$ to denote the diagonal matrix $\on{diag}(d-r+1, d-r+2, \ldots, d)$ and write $Z_{\Sigma} := \Sigma^{1/2} Z.$\
Recall that the Wasserstein distance for probability distributions $\mu$, $\nu$ on $(\cc^d,\calb(\cc^d))$ is defined by
$$d_\calw(\mu,\nu):=\sup\left\{|\int h d\mu-\int h d\nu|\,:\, h:\cc^d\rightarrow\rr\text{ and } \|h\|_{\on{Lip}}\leq 1\right\},$$ $\|h\|_{\on{Lip}}$ denoting the minimum Lipschitz constant $$\|h\|_{\on{Lip}} :=\sup_{x\not=y}\frac{|h(x)-h(y)|}{\|x-y\|_2}\in[0,\infty]$$ of $h$. If $\mu$ is the law of $X$ and $\nu$ is the law of $Y$, we write $d_{\calw}(X, Y) = d_{\calw}(\mu, \nu).$\
We are now in a position to state the crucial quantitative CLT for vectors of traces of powers [@DoeSt11 Thm. 1.1].
\[trofpow\] If $n \ge 2d$, the Wasserstein distance between $W$ and $Z_{\Sigma}$ is $$\label{ordnungsformel}
d_\calw(W,Z_\Sigma)=\on{O}\left(\frac{\max\left\{\frac{r^{7/2}}{(d-r+1)^{3/2}},\,(d-r)^{3/2}\sqrt{r}\right\}}{n}\right)\,.$$ In particular, for $r = d$ we have $$\label{trofpow-speziell}
d_{\calw}(W, Z_{\Sigma}) = \on{O}\left(\frac{d^{7/2}}{n}\right).$$
We will also need the following orthogonality relations for traces of powers:
\[orth\] Let $n, i, j \in \nn$. Then $$\erw(\on{Tr}(M_n^i) \on{Tr}(M_n^j)) = 0$$ and $$\erw(\on{Tr}(M_n^i) \overline{\on{Tr}(M_n^j)}) = \delta_{ij} (j \wedge n).$$
The first equality follows from the unitary invariance of Haar measure, see also [@StolzDSM Thm. 3.29] For the second equality see [@DiaconisEvansTAMS Thm.2.1(b)].
Rates of convergence for linear statistics {#ratessmoothcase}
==========================================
The following theorem is the main result of this note.
Let $M_n$ be a Haar distributed unitary $n \times n$ matrix, and let $f \in \on{L}^1(\torus)$ be real valued and satisfy condition above with $\kappa > 1$. Then, as $n \to \infty$, the fluctuation $n(L_n(f) - \erw L_n(f))$ converges in distribution to a centered Gaussian random variable with variance $$\sigma^2 := \sum_{j = -\infty}^{\infty} |\hat{f}_j|^2 |j| = 2 \sum_{j=1}^{\infty} |\hat{f}_j|^2 j < \infty.$$ If $\call_n$ denotes the law of $n(L_n(f) - \erw L_n(f))$, then there exists $C = C(f, \kappa) > 0$ such that for any $n \in \nn$ the Wasserstein distance between $\call_n$ and $\on{N}(0, \sigma^2)$ can be bounded as follows:
$$d_\calw(\call_n, \on{N}(0, \sigma^2)) \le C\ n^{-\left(\frac{k-1}{k + 5/2}\right)}.$$
In particular, if $f$ is in $\on{C}^{\infty}(\torus)$, then $d_\calw(\call_n, \on{N}(0, \sigma^2))$ is $\on{O}\left( \frac{1}{n^{1-\epsilon}}\right)$ for all $\epsilon > 0$.
In [@DiaconisEvansTAMS] the corresponding weak convergence result without a bound on the speed of convergence is proven under the weaker assumption that $\sigma^2$ be finite, which amounts to requiring that $f$ be an element of the Sobolev space $H^{1/2}(\torus)$. Since on the one hand, elements of $H^{1/2}(\torus)$ need not be continuous [@Katznelson I.8.11], and on the other hand, our method of proof requires the assumption that $\kappa > 1$, which implies that $f$ must be the uniform limit of a sequence of continuous functions, the present result applies to a strictly smaller class of test functions.
It follows from that $\sigma^2 < \infty.$ Set $$S_n := \sum_{j=1}^{\infty} \hat{f}(j) \on{Tr}(M_n^j) + \sum_{j=1}^{\infty} \overline{\hat{f}(j) \on{Tr}(M_n^j)},$$ and write, for $d = d(n) \le \frac{n}{2}$ to be chosen later, $$S^{(1)}_n := \sum_{j=1}^{d} \hat{f}(j) \on{Tr}(M_n^j) + \sum_{j=1}^{d} \overline{\hat{f}(j) \on{Tr}(M_n^j)},$$ and $$\label{S2} S_n^{(2)} := S_n - S_n^{(1)}.$$ Then, by Proposition \[coeffbound\] and dominated convergence, $$\begin{aligned}
\erw|S_n^{(2)}|^2 &= \sum_{i, j = d+1}^{\infty} \erw\left( \hat{f}_i \on{Tr}(M_n^i) + \overline{\hat{f}_i}\ \overline{\on{Tr}(M_n^i)}\right)
\overline{\left( \hat{f}_j \on{Tr}(M_n^j) + \overline{\hat{f}_j}\ \overline{\on{Tr}(M_n^j)}\right)}\\
&= \sum_{i, j = d+1}^{\infty} \hat{f}_i \overline{\hat{f}_j}\ \erw\left( \on{Tr}(M_n^i) \overline{\on{Tr}(M_n^j)}\right) + \hat{f}_i \hat{f}_j\ \erw\left( \on{Tr}(M_n^i) \on{Tr}(M_n^j)\right) \\
&\quad\quad + \overline{\hat{f}_i \hat{f}_j}\ \overline{ \erw\left( \on{Tr}(M_n^i) \on{Tr}(M_n^j)\right)} + \overline{\hat{f}_i} \hat{f}_j\ \erw\left( \overline{\on{Tr}(M_n^i)} \on{Tr}(M_n^j)\right)\\
&= 2 \sum_{j=d+1}^{\infty} |\hat{f}_j|^2 (j \wedge n),\end{aligned}$$ where we have used Proposition \[orth\] in the last step. Consequently, Proposition \[coeffbound\] implies that $$\begin{aligned}
\erw|S_n^{(2)}|^2 &\le 2 \sum_{j = d+1}^{\infty} j |\hat{f}_j|^2 \le 2 \| f^{(k)}\|_1^2 \sum_{j = d+1}^{\infty} \frac{1}{ j^{2k - 1}}\nonumber\\
&\le 2 \| f^{(k)}\|_1^2 \int_d^{\infty} \frac{1}{x^{2k - 1}} dx = 2 \| f^{(k)}\|_1^2 \frac{1}{(2k - 2) d^{(2k-2)}}.\label{boundS2}\end{aligned}$$
For $z \in \cc^d$ define $$\phi(z) = \phi_n(z) = \sum_{j=1}^d \hat{f}_j z_j + \overline{ \hat{f}_j} \overline{z_j}.$$ We write $\underline{f} = (\hat{f}_1, \ldots, \hat{f}_d)$. Then, for any $z, w \in \cc^d$, Cauchy-Schwarz implies $$|\phi(z) - \phi(w)| = \left| \sum_{j=1}^d \hat{f}_j (z_j - w_j) + \overline{\hat{f}_j}(\overline{z_j} - \overline{w_j})
\right| \le 2 \| \underline{f}\|_2 \| z-w\|_2.$$ We have thus obtained that $$\|\phi\|_{\on{Lip}} \le
2 \| \underline{f}\|_2.$$ Note that, by Proposition \[coeffbound\], $$\| \underline{f}\|^2_2\ = \sum_{j=1}^{d(n)} |\hat{f}_j|^2 \le \|f^{(k)}\|_1^2 \sum_{j = 1}^{\infty} \frac{1}{j^{2k}} < \infty$$ is bounded by a constant independent of $n$ and of the specific form of $f$.\
Now let $\zeta$ be a real standard normal random variable, and $Z, \Sigma, Z_{\Sigma}$ as above in Section \[setup\]. Let $g: \rr \to \rr$ be Lipschitz with $\|g\|_{\on{Lip}} \le 1$. Then $$\begin{aligned}
&|\erw(g(S_n)) - \erw(g(\sigma\zeta))| \nonumber\\
&\le\ |\erw(g(S_n)) - \erw(g(S_n^{(1)}))| + |\erw(g(S_n^{(1)})) - \erw(g(\phi(Z_{\Sigma})))| + |\erw(g(\phi(Z_{\Sigma}))) - \erw(g(\sigma \zeta)))|
\label{EinsZweiDrei}\end{aligned}$$ In view of and , the first summand in can be bounded by $$\|g\|_{\on{Lip}}\ \erw|S_n^{(2)}| \le \|g\|_{\on{Lip}} \sqrt{ \erw|S_n^{(2)}|^2 }
\le \|g\|_{\on{Lip}}\ \|f^{(k)}\|_1 \sqrt{ \frac{1}{(2k - 2)\ d^{2k-2}}}.$$ Writing $W_n := (\on{Tr}(M_n), \ldots, \on{Tr}(M_n^d))$, by Proposition \[trofpow\] the second summand in has an upper bound $$\begin{aligned}
|\erw(g(S_n^{(1)})) - \erw(g(\phi(Z_{\Sigma})))| &= |\erw(g(\phi(W_n))) - \erw(g(\phi(Z_{\Sigma})))|\\
&= \|\phi\|_{\on{Lip}} \left| \erw \left( \frac{1}{\|\phi\|_{\on{Lip}}} (g \circ \phi)(W_n)\right) -
\erw \left( \frac{1}{\|\phi\|_{\on{Lip}}} (g \circ \phi)(Z_{\Sigma})\right)\right|\\
&\le \|\phi\|_{\on{Lip}}\ d_{\calw}(W_n, Z_{\Sigma}) \le 2 \| \underline{f}\|_2\ C\ \frac{d^{7/2}}{n},\end{aligned}$$ where $C$ is the constant from Proposition \[trofpow\].\
To bound the third summand in , we will first study the distribution of $\phi(Z_{\Sigma})$. Recall that $Z_{\Sigma}$ is the random vector $(Z_1, \sqrt{2} Z_2, \ldots, \sqrt{d}Z_d)$, where the $Z_j\ (j = 1, \ldots,
d)$ are iid standard complex normal random variables, i.e., there are iid real random variables $X_{1}, \ldots, X_d,$ $Y_{1}, \ldots, Y_d$ with distribution $\on{N}(0, 1/2)$ such that $Z_j = X_j + i Y_j$ for $j = 1, \ldots, d$. Then $$\begin{aligned}
\phi(Z_{\Sigma}) &= \sum_{j=1}^d \hat{f}_j \sqrt{j} Z_j + \sum_{j=1}^d \overline{\hat{f}_j} \sqrt{j}\ \overline{Z_j}\\
&= 2 \sum_{j=1}^d \sqrt{j}\ \on{Re}( \hat{f}_j Z_j) \\
&= 2 \sum_{j=1}^d \sqrt{j}\ \left( \on{Re}( \hat{f}_j) X_j - \on{Im}( \hat{f}_j) Y_j\right) \end{aligned}$$ $\phi(Z_{\Sigma})$ is thus a centered real normal random variable with variance $$4 \sum_{j=1}^d \frac{j}{2} (\on{Re}(\hat{f}_j)^2 + \on{Im}(\hat{f}_j)^2) = 2 \sum_{j=1}^{d(n)} j |\hat{f}_j|^2
=: \sigma_n^2.$$ So we obtain $$\begin{aligned}
|\erw (g(\phi(Z_{\Sigma}))) - \erw(g(\sigma \zeta))| &= |\erw (g(\sigma_n \zeta)) - \erw(g(\sigma \zeta))|\\
&\le\ \|g\|_{\on{Lip}}\ \erw|\sigma_n \zeta - \sigma \zeta|\\ &\le \|g\|_{\on{Lip}}\ |\sigma_n - \sigma|\ \erw|\zeta|\\ &\le \|g\|_{\on{Lip}}\ |\sigma_n - \sigma|.\end{aligned}$$ Now, $$\begin{aligned}
|\sigma_n^2 - \sigma^2| &= 2 \sum_{j = d+1}^{\infty} j |\hat{f}_j|^2 \le 2 \|f^{(k)}\|_1^2\ \sum_{j = d+1}^{\infty} \frac{1}{j^{2k - 1}}\\ &\le
2 \|f^{(k)}\|_1^2\ \int_d^{\infty} \frac{1}{x^{2k-1}} dx\\
&= 2 \|f^{(k)}\|_1^2\ \frac{1}{(2k - 2)\ d^{2k - 2}}.\end{aligned}$$ Since $|\sigma_n^2 - \sigma^2| = |(\sigma_n - \sigma) (\sigma_n + \sigma)|$, we have that $$|\sigma_n - \sigma| = \frac{|\sigma_n^2 - \sigma^2|}{ \sigma_n + \sigma} \le \frac{| \sigma_n^2 - \sigma^2|}{\sigma},$$ and the third term in may thus be bounded by $$\label{BoundDrei}
2 \|g\|_{\on{Lip}}\ \|f^{(k)}\|_1^2\ \frac{1}{ \sigma (2k-2)\ d^{2k-2}}.$$
Comparing the bounds that we have obtained for the individual summands in , we obtain that $$\begin{aligned}
\label{ZusammenschauSchranken}
|\erw(g(S)) - \erw(g(\sigma \zeta))| &\le \|g\|_{\on{Lip}}\ C_{f, k}\ \max\left\{ \frac{1}{d^{k-1}},\ \frac{d^{7/2}}{n}\right\}\nonumber\\
&\le \|g\|_{\on{Lip}}\ C_{f, k}\ \left( \frac{1}{d^{k-1}} + \frac{d^{7/2}}{n}\right),\end{aligned}$$ where $C_{f, k}$ depends only on $f$ and $k$.\
One verifies that, on $]0, \infty[$, the function $f(x) := \frac{1}{x^{k-1}} + \frac{x^{7/2}}{n}$ has a unique global minimum at $$x_0 := \left(\frac{2(k-1)}{7}\right)^{\frac{2}{5+2k}} n^{\frac{2}{5+2k}}.$$ Furthermore, $f$ is strictly falling on $]0, x_0[$ and strictly growing on $]x_0, \infty[$. Setting $d_1 := \lfloor x_0 \rfloor$ and $d_2 := d_1 + 1$, we see that the minimizer of $f$ on the positive integers is in $\{ d_1, d_2\}$.\
There are $0 < \alpha < \beta$, depending only on $k$, such that for all $n \in \nn$ one has $$\alpha n^{\frac{2}{5+2k}} \le d_1 < d_2 \le \beta n^{\frac{2}{5+2k}}.$$ Hence, $$\begin{aligned}
f(x_0) &\le \frac{1}{d_1^{k-1}} + \frac{d_2^{7/2}}{n} \le \frac{1}{\alpha^{k-1} n^{\frac{2k - 2}{2k + 5}}} + \beta^{7/2} n^{\left( \frac{7}{5+2k} - 1\right)}\\
&= \frac{1}{\alpha^{k-1} n^{\frac{2k - 2}{2k + 5}}} + \beta^{7/2} n^{- \frac{2k - 2}{2k + 5}} = \gamma n^{- \frac{2k - 2}{2k + 5}},\end{aligned}$$ where $\gamma := \frac{1}{\alpha^{k-1}} + \beta^{7/2}.$ Comparing with yields the theorem.
[CGS11]{}
Louis H. Y. Chen, Larry Goldstein, and Qi-Man Shao. . Probability and its Applications (New York). Springer, Heidelberg, 2011.
Sourav Chatterjee. Fluctuations of eigenvalues and second order [P]{}oincaré inequalities. , 143(1-2):1–40, 2009.
Ovidiu Costin and Joel L. Lebowitz. Gaussian fluctuation in random matrices. , 75:69–72, Jul 1995.
Persi Diaconis and Steven N. Evans. Linear functionals of eigenvalues of random matrices. , 353(7):2615–2633, 2001.
Christian D[ö]{}bler and Michael Stolz. Stein’s method and the multivariate [CLT]{} for traces of powers on the classical compact groups. , 16:2375–2405, 2011.
Jason Fulman. Stein’s method, heat kernel, and traces of powers of elements of compact [L]{}ie groups. , 2010.
Loukas Grafakos. . Graduate Texts in Mathematics 249. Springer, second edition, 2008.
Kurt Johansson. On random matrices from the compact classical groups. , 145(3):519–545, 1997.
Yitzhak Katznelson. . Cambridge Mathematical Library. Cambridge University Press, Cambridge, third edition, 2004.
Elizabeth Meckes and Mark Meckes. Concentration and convergence rates for spectral measures of random matrices. To appear in [*Probab. Theory Related Fields*]{}.
Alexander Soshnikov. The central limit theorem for local linear statistics in classical compact groups and related combinatorial identities. , 28(3):1353–1370, 2000.
Michael Stolz. On the [D]{}iaconis-[S]{}hahshahani method in random matrix theory. , 22(4):471–491, 2005.
Kelly Wieand. Eigenvalue distributions of random unitary matrices. 123(2):202–224, 2002.
[^1]: Ruhr-Universität Bochum, Fakultät für Mathematik, D-44780 Bochum, Germany.\
email: [email protected] and [email protected].\
Both authors have been supported by Deutsche Forschungsgemeinschaft via SFB-TR 12.\
Keywords: Random matrices, Haar measure, unitary group, speed of convergence, central limit theorem, traces of powers.\
MSC 2010: 60F05, 60B15, 60B20
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
The generalization error bound of support vector machine (SVM) depends on the ratio of radius and margin, while standard SVM only considers the maximization of the margin but ignores the minimization of the radius. Several approaches have been proposed to integrate radius and margin for joint learning of feature transformation and SVM classifier. However, most of them either require the form of the transformation matrix to be diagonal, or are non-convex and computationally expensive.
In this paper, we suggest a novel approximation for the radius of minimum enclosing ball (MEB) in feature space, and then propose a convex radius-margin based SVM model for joint learning of feature transformation and SVM classifier, i.e., F-SVM. An alternating minimization method is adopted to solve the F-SVM model, where the feature transformation is updated via gradient descent and the classifier is updated by employing the existing SVM solver. By incorporating with kernel principal component analysis, F-SVM is further extended for joint learning of nonlinear transformation and classifier. Experimental results on the UCI machine learning datasets and the LFW face datasets show that F-SVM outperforms the standard SVM and the existing radius-margin based SVMs, e.g., RMM, R-SVM$^{+}$ and R-SVM$_{\mu}^{+}$.
author:
- 'Xiaohe Wu, Wangmeng Zuo, Yuanyuan Zhu, Liang Lin, [^1][^2]'
bibliography:
- 'IEEEabrv.bib'
- 'fsvm.bib'
title: 'F-SVM: Combination of Feature Transformation and SVM Learning via Convex Relaxation'
---
[Shell : F-SVM: Combination of Feature Transformation and SVM Learning via Convex Relaxation]{}
Support vector machine, radius margin bound, convex relaxation, max-margin.
Introduction
============
vector machine (SVM) and its extensions have been one of the most successful machine learning methods [@vapnik1998statistical; @cristianini2000introduction], and have been adopted in various fields, e.g., computer vision [@osuna1997training; @heisele2001face; @guo2001support; @wu2015linear], signal processing [@samanta2003artificial; @chen2001support], natural language processing [@collobert2011natural; @tsochantaridis2004support] and bioinformatics [@do2009feature; @weston2000feature; @guyon2002gene; @rakotomamonjy2003variable]. Despite its popularity, SVM aims to seek the optimal hyperplane with the maximum margin principle, but the generalization error of SVM actually is a function of the ratio of radius and margin [@vapnik2000bounds]. Given feature space, the radius is fixed and can be ignored, thus SVM can minimize the generalization error by maximizing the margin. However, for joint learning of feature transformation and classifier, the radius information will be valuable and cannot be ignored.
By minimizing the radius-margin ratio, the generalization error of SVM can be optimized for joint learning of feature transformation and classifier. Since the radius-margin error bound is non-convex, relaxation and approximation of radius is generally adopted in the existing models [@shivaswamy2010maximum; @zhu2012learning]. Several approaches have been proposed from the perspective of radius-margin error [@do2009feature; @shivaswamy2010maximum; @zhu2012learning; @do2013convex], but most ones suffer from the limitations of computational burden and simplified forms of transformation. RMM [@shivaswamy2010maximum] only considers the spread of the data along the direction perpendicular to the classification hyperplane. Radius-margin based SVMs, e.g., MR-SVM [@do2009feature], R-SVM$^{+}$ [@do2013convex] and RSVM$_{\mu}^{+}$ [@do2013convex], are based on the constraint that the linear transformation matrix should be diagonal.
Another strategy for joint feature transformation and classifier learning is to incorporate metric learning with SVM, where metric learning can be adopted to learn a better linear transformation matrix [@xing2002distance; @zhu2012learning; @globerson2005metric; @schultz2004learning; @wang2013kernel; @shen2014efficient; @guillaumin2009you; @wang2014shrinkage; @tran2008human]. One simple approach to combine metric learning and SVM is to directly deploy the transformation obtained using metric learning into SVM. This approach, however, usually cannot lead to satisfactory performance improvement [@xu2012distance]. Therefore, other approaches have been proposed to integrate metric learning to SVM, e.g., support vector metric learning (SVML) [@xu2012distance] and metric learning with SVM (MSVM) [@zhu2012learning]. But SVML [@xu2012distance] was designed for RBF-SVM and ignored the radius information, while MSVM [@zhu2012learning] is non-convex.
In this paper, we propose a novel radius-margin based SVM model for joint learning of feature transformation and SVM classifier, i.e., F-SVM. Compared with the existing radius-margin based SVM methods, we derive novel lower and upper bounds for the relaxation of the radius-margin ratio. Unlike MR-SVM [@do2009feature], R-SVM$^{+}$ [@do2013convex] and RSVM$_{\mu}^{+}$ [@do2013convex] which are suggested for joint feature weighting and SVM learning, F-SVM can simultaneously learn feature transformation ${\bf L}$ and classifier $( {\bf w}, b )$. Compared with the existing metric learning for SVM methods, our F-SVM model considers both the radius and the margin information, and is convex. Then, an alternating minimization algorithm is proposed to solve our F-SVM model, which iterates by updating feature transformation and classifier alternatively. Note that kernel SVM is equivalent to perform linear SVM in the kernel PCA space. We further suggest to conduct linear FSVM in the kernel PCA space for joint learning of nonlinear transformation and classifier. The contribution of this paper is of three-fold:
- A novel convex formulation of radius-margin based SVM model, i.e., F-SVM, is proposed. Unlike MR-SVM [@do2009feature], R-SVM$^{+}$ [@do2013convex] and RSVM$_{\mu}^{+}$ [@do2013convex], our F-SVM is capable of joint learning feature transformation and classifier, and is robust against outliers. Experimental results show that F-SVM outperforms SVM and the existing radius-margin based SVMs.
- In F-SVM, we derive the lower and upper bounds for the radius of minimum enclosing ball (MEB) in feature space, resulting in a novel approximation of the radius. Compared with the approximations proposed in [@do2013convex], ours is much simple and can be easily adopted in developing radius-margin based SVM models.
- An alternating minimization algorithm is developed for solving F-SVM via iterating between gradient descent and SVM learning. Therefore, the off-the-shelf SVM solvers can be employed to improve the computational efficiency. Moreover, a semi-whitened PCA method is developed for the initialization of $ {\bf M} = {\bf L}^T {\bf L} $.
The remainder of the paper is organized as follows: Section 2 reviews the related work on the radius-margin ratio based bounds and their applications. Section 3 describes the model and algorithm of the proposed F-SVM method. Section 4 extends F-SVM to the kernelized version for nonlinear classification. Section 5 provides the experimental results on the UCI machine learning datasets and the LFW dataset. Finally, we conclude the paper in Section 6.
Related work
============
The radius-margin bound not only provides theoretical explanation on the generalization performance of SVM [@vapnik1998statistical], but also has been extensively adopted for improving kernel classification methods, e.g., model selection [@chapelle2002choosing; @sydorov2014deep], multiple kernel learning (MKL) [@gai2010learning; @do2009margin; @liu2013efficient; @liu2014efficient], and mapping of nominal attributes [@peng2014improved]. Denote a training set by $\mathcal{S}=\left\{ \left( {{\mathbf{x}}_{1}},{{y}_{1}} \right),...,\left( {{\mathbf{x}}_{n}},{{y}_{n}} \right) \right\}$ and a feature space by $\mathcal{H} : \Phi \left( \mathbf{x} \right)$. In [@gai2010learning; @do2009margin], the radius ${R}$ of minimum enclosing ball (MEB) in feature space is computed as: $$\begin{aligned}
\label{eq:class_R}
\min_{ {R}, {\mathbf{x}}_{0} } {R} ^{2}, s.t. \| \Phi ({{\mathbf{x}}_{i}})-\Phi ({{\mathbf{x}}_{0}}) \|_{2}^{2} \le {{R}^{2}},i=1,2,\cdots ,n.\end{aligned}$$ Assuming that the training set is separable, given the optimal hyperplane $( {\bf w}, b )$, Vapnik [@vapnik1998statistical] suggested a radius-margin error bound which showed that the expectation of the misclassification probability depends on ${R}^{2} \| {\mathbf{w}} \|_{2}^{2}$.
The standard SVM is known as a max-margin method which only considers the margin $1/\| {\mathbf{w}} \|_{2}^{2}$ in the algorithm. When the feature space is fixed, the radius is a constant and can thus be ignored. But in many classification tasks, the model parameters [@chapelle2002choosing], combination of basis kernels [@do2009margin], feature reweighting or transformation [@weston2000feature] usually should be learned or tuned based on the training data, where integration of radius has been demonstrated to be very effective in improving the classification performance. In model selection, radius-margin bound has been applied for choosing tradeoff parameter and scaling factors of SVM and $L_{1}$-SVM [@chapelle2002choosing]. In multiple kernel learning (MKL) [@gai2010learning; @do2009margin; @liu2013efficient] and feature reweighting [@weston2000feature], several variants of radius had been developed.
This paper aims to jointly learn SVM together with feature transformation by minimizing the radius-margin ratio, i.e., radius-margin based SVM, and more detailed review is given on this topic. Except [@zhu2012learning], most existing approaches [@do2009feature; @do2013convex; @schultz2004learning] require the transformation matrix to be diagonal, i.e., feature reweighting and selection. Direct use of radius-margin ratio ${R}^{2} \| {\mathbf{w}} \|_{2}^{2}$ in SVM results in a non-convex optimization problem, which makes the learning algorithm computationally expensive and unstable. By restricting the feature transformation to be diagonal ${{\mathbf{D}}_{\bf{\mu }}}\!=\!Diag\left( \bf{\mu } \right)$ with ${{\mu }_{k}} \!\ge\! 0$, Do et al. [@do2009feature] suggested that the radius is bounded with ${{\max }_{k}}\ {{\mu }_{k}}R_{k}^{2}\le R_{\mathbf{\mu }}^{2}\le \sum\nolimits_{k}{{{\mu }_{k}}R_{k}^{2}}$, where ${{R}_{k}}$ is the radius on dimension ${k}$. By approximating $R_{\mathbf{\mu }}^{2}$ with its upper bound $\sum\nolimits_{k}{{{\mu }_{k}}R_{k}^{2}}$, MR-SVM in [@do2009feature] solved the following convex relaxation problem: $$\begin{aligned}
\label{eq:class_MR-SVM}
\begin{aligned}
\min \limits_{ {\bf w},{b},{\mathbf \xi},{\mathbf \mu} } \ \ \ \ & {\frac{1}{2}}\sum\nolimits_{k}\frac{w_{k}^{2}}{{\mu}_{k}}+{\frac{C}{\sum\nolimits_{k}{{{\mu }_{k}}R_{k}^{2}}}}\sum\nolimits_{i}{{{\xi }_{i}}}, \\
s.t. \ \ \ \ & {{y}_{i}}({{{\mathbf{w}}^{T}}{{\mathbf{x}}_{i}}}+b)\ge 1-{{\xi }_{i}}, \forall i, \\
\ \ \ \ & {{\xi }_{i}} \ge 0, i=1,2,\cdots ,n,\\
\ \ \ \ & \sum\nolimits_{k}{{{\mu }_{k}}}=1,{{\mu }_{k}}\ge 0,\forall k.
\end{aligned}\end{aligned}$$ Denote ${{R}_{O}}$ by the half value of the maximum pairwise distances. Do et al. in [@do2013convex] introduced a tighter bound of the radius ${{R}_{O}}\le {{R}_{\mathbf{\mu }}}\le \frac{1+{\sqrt{3}}}{2}{{R}_{O}}$ and proposed another convex model R-SVM$_{\mu}^{+}$: $$\begin{aligned}
\label{eq:class_R-SVM-mu+}
\begin{aligned}
\min \limits_{ {\bf w},{b},{\mathbf \xi},{\mathbf \mu},{r}} \ \ \ \ & {\frac{1}{2}}\sum\nolimits_{k}\frac{w_{k}^{2}}{{\mu}_{k}}+{\lambda}{r}+ C\sum\nolimits_{i}{{{\xi }_{i}}}, \\
s.t. \ \ \ \ & {{y}_{i}}({{\mathbf{w}}^{T}}{{\mathbf{x}}_{i}}+b)\ge 1-{{\xi }_{i}}, \forall i, \\
\ \ \ \ & {{\xi }_{i}} \ge 0, i=1,2,\cdots ,n, \\
\ \ \ \ & \sum\nolimits_{k}{{{\mu }_{k}}}=1,{{\mu }_{k}}\ge 0,\forall k, \\
\ \ \ \ & \frac{1}{2}{{\left\| {{D}_{\sqrt{\mathbf{\mu }}}}{{\mathbf{x}}_{i}}-{{D}_{\sqrt{\mathbf{\mu }}}}{{\mathbf{x}}_{j}} \right\|}^{2}}\le r,\forall i,j.
\end{aligned}\end{aligned}$$ Furthermore, R-SVM$^{+}$ was developed in [@do2013convex] by controlling both the radius and margin with ${\bf w}$.
Rather than feature reweighting and selection, Zhu et al. [@zhu2012learning] proposed a metric learning with SVM (MSVM) method for joint learning of the linear transformation and SVM classifier. In [@zhu2012learning], given the transformation matrix ${\mathbf A}$, an alternative $\bar{R}={{\max }_{i}}\left\| \mathbf{A}{{\mathbf{x}}_{i}}-\mathbf{A\bar{x}} \right\|_{2}^{2}$ of the radius $R$ was adopted, where $\mathbf{\bar{x}}$ is the mean of the training samples. Although Zhu et al. [@zhu2012learning] claimed that $R=\bar{R}$, as demonstrated in [**Theorem 1**]{} of this work, $\bar{R}$ is an upper bound of $R$. The MSVM model in [@zhu2012learning] was formulated as: $$\begin{aligned}
\label{eq:class_MSVM}
\begin{aligned}
\min \limits_{\mathbf{w},b,\mathbf{A}} \ \ \ \ & \frac{1}{2}\left\| \mathbf{w} \right\|_{2}^{2}+C\sum\nolimits_{i}{{{\xi }_{i}}}, \\
s.t. \ \ \ \ & {{y}_{i}}({{\mathbf{w}}^{T}}\mathbf{A}{{\mathbf{x}}_{i}}+b)\ge 1-{{\xi }_{i}}, \forall i, \\
\ \ \ \ & {{\xi }_{i}} \ge 0, i=1,2,\cdots ,n, \\
\ \ \ \ & {{\left\| \mathbf{A}{{\mathbf{x}}_{i}}-\mathbf{A\bar{x}} \right\|}^{2}}\le 1,\forall i.
\end{aligned}\end{aligned}$$ Note that MSVM is non-convex and solved using gradient projection.
In this paper, we propose a novel relaxed convex model of radius-margin based SVM, i.e., F-SVM, for joint learning of feature transformation and SVM classifier. Compared with the existing radius-margin based SVM methods, F-SVM has some distinguishing features. Our F-SVM model is convex, while MSVM [@zhu2012learning] is non-convex. Unlike RMM [@shivaswamy2010maximum], the transformation in F-SVM is learned to minimize the radius of the enclosing ball of all samples rather than to only shrink the sample span along the direction perpendicular to the hyperplane. Moreover, F-SVM is also different with MR-SVM [@do2009feature], R-SVM$^{+}$ [@do2013convex] and R-SVM$_{\mu}^{+}$ [@do2013convex] from three aspects: $(i)$ Instead of feature reweighting and selection, F-SVM can learn feature transformation and classifier simultaneously; $(ii)$ F-SVM adopts a new approximation for the radius of MEB in feature space; $(iii)$ In F-SVM, individual inequality constraints are combined into one holistic inequality constraint to improve the robustness against outliers. All these make F-SVM very promising for joint learning of feature transformation and SVM classifier, and the results also validate the effectiveness of F-SVM.
Radius-margin based Support Vector Machine
==========================================
Problem Formulation
-------------------
Denote $\mathcal{S}=\left\{ \left( {{\mathbf{x}}_{1}},{{y}_{1}} \right),...,\left( {{\mathbf{x}}_{n}},{{y}_{n}} \right) \right\}$ by a training set, where ${{\mathbf{x}}_{i}} \in {{\mathbf{R}}^{d}}$ and ${{y}_{i}}\in \left\{ -1,+1 \right\}$ denote the [*i*]{}th training sample and the corresponding class label, respectively. By introducing the slack variables ${{\xi }_{i}}\left( i=1,2,...,n \right)$, SVM aims to find the optimal separating hyperplane by solving the following optimization problem: $$\begin{aligned}
\label{eq:class_SVM}
\begin{aligned}
\min \limits_{\mathbf{u},b,{\xi}} \ \ \ \ & \frac{1}{2}\left\| \mathbf{u} \right\|_{2}^{2}+C\sum\nolimits_{i}{{{\xi }_{i}}}, \\
s.t. \ \ \ \ & {{y}_{i}}({{\mathbf{u}}^{T}}{{\mathbf{x}}_{i}}+b)\ge 1-{{\xi }_{i}}, \forall i, \\
\ \ \ \ & {{\xi }_{i}} \ge 0, i=1,2,\cdots ,n.
\end{aligned}\end{aligned}$$ where $({\bf u},b)$ are the parameters to describe the learned hyperplane ${{\mathbf{u}}^{T}}\mathbf{x}+b=0$, ${\xi}_{i}$ denotes the [*i*]{}th slack variable, and $C$ stands for the tradeoff parameter. The objective function in Eq. (\[eq:class\_SVM\]) aims to maximize the margin $\gamma ={1}/{{{\left\| \mathbf{u} \right\|}^{2}}}$ while minimizing the empirical risk $\sum\nolimits_{i=1}^{n}{{{\xi }_{i}}}$. For joint learning, we introduce a linear transformation matrix ${\bf A}$ and integrate the radius information, resulting in the following radius-margin based SVM model: $$\begin{aligned}
\label{eq:class_radiu-margin-based-SVM}
\begin{aligned}
\min \limits_{\mathbf{u},b,{\xi},{\mathbf{A}},R} \ \ \ \ & \frac{1}{2}\left\| \mathbf{u} \right\|_{2}^{2}{R}^2+C\sum\nolimits_{i}{{{\xi }_{i}}}, \\
s.t. \ \ \ \ & {{y}_{i}}({{\mathbf{u}}^{T}}{\bf A}{{\mathbf{x}}_{i}}+b)\ge 1-{{\xi }_{i}}, \forall i, \\
\ \ \ \ & {{\xi }_{i}} \ge 0, i=1,2,\cdots ,n.
\end{aligned}\end{aligned}$$ where the radius $R$ is defined as: $$\begin{aligned}
\label{eq:class_F-SVM-R}
\min_{ {R}, {\mathbf{x}}_{0} } {R} ^{2}, s.t. \| {{\bf A}{\mathbf{x}}_{i}}- {{\bf A}{\mathbf{x}}_{0}} \|_{2}^{2} \le {{R}^{2}},i=1,2,\cdots ,n.\end{aligned}$$ Note that $R^2$ depends on matrix $\bf A$ and the problem in Eq. (\[eq:class\_radiu-margin-based-SVM\]) is non-convex [@do2013convex]. Denote ${\bf {x}_{0}}$ by the center of all instances, and $\bar R$ by the largest squared distance from the center in transformed feature space. Let ${{\mathbf{x}}_{0}}=\mathbf{\bar{x}}=\sum\nolimits_{i=1}^{n}{{{\mathbf{x}}_{i}}}$ and $\bar{R}= \max_{i} \left\| {\bf A}{{\bf x}_{i}}-{\bf A}{\bf \bar{x}} \right\|_{2}^{2}$. We prove that the radius $R$ is bounded by $\bar R$.
[**Theorem 1**]{}. The radius $R$ is bounded by $\bar R$ by: $$\begin{aligned}
\label{eq:class_R-bound}
\frac{1}{2}\overline{R}\le R\le \overline{R}.\end{aligned}$$ Please refer to [**Appendix A**]{} for the proof of [**Theorem 1**]{}. In [@zhu2012learning], Zhu et al. claimed that $R=\bar{R}$ . From [**Theorem 1**]{}, $\bar R$ is only an approximation of $R$, and counter examples can be easily found to illustrate $R\ne \bar{R}$. Let $\mathbf{w}={{\mathbf{A}}^{T}}\mathbf{u}$ and $\mathbf{M}={{\mathbf{A}}^{T}}\mathbf{A}$. Since the radius $R$ is upper bounded by $\bar R$, we can approximate $R$ with $\bar R$. With simple algebra, the radius-margin SVM model in Eq. (\[eq:class\_radiu-margin-based-SVM\]) is relaxed into the following formulation: $$\begin{aligned}
\label{eq:class_radiu-margin-based-SVM-relaxation}
\begin{aligned}
\min_{\mathbf{w}\!,b\!,\mathbf{\xi }\!,\mathbf{M}\!,\bar{R}\!} \ \ \ \ & F(\mathbf{w},b,\mathbf{\xi },\mathbf{M},\bar{R}) \!\!=\!\! \frac{1}{2}\left( {{\mathbf{w}}^{T}}\!{{\mathbf{M}}^{-1}}\!\mathbf{w}\! \right)\bar{R}_{{}}^{2}\!+\!C\!\sum\limits_{i=1}^{n}{{{\xi }_{i}}}, \\
s.t. \ \ \ \ & {{y}_{i}}({{\mathbf{w}}^{T}}{{\mathbf{x}}_{i}}+b)\ge 1-{{\xi }_{i}}, \forall i,\\
\ \ \ \ & {{\xi }_{i}}\ge 0,i=1,2,\cdots ,n, \\
\ \ \ \ & {{({{\mathbf{x}}_{i}}-\mathbf{\bar{x}})}^{T}}\mathbf{M}({{\mathbf{x}}_{i}}-\mathbf{\bar{x}})\le {{{\bar{R}}}^{2}}.
\end{aligned}\end{aligned}$$
[**Theorem 2**]{}. The problem in Eq. (\[eq:class\_radiu-margin-based-SVM-relaxation\]) is equivalent with the following problem: $$\begin{aligned}
\label{eq:class_radiu-margin-based-SVM-relaxation-equivalentaion}
\begin{aligned}
\min_{\mathbf{w},b,\mathbf{\xi },\mathbf{M}} \ \ \ \ & L(\mathbf{w},b,\mathbf{\xi },\mathbf{M})\!\!=\!\!
\left\{ \frac{1}{2}\left( {{\mathbf{w}}^{T}}{{\mathbf{M}}^{-1}}\mathbf{w} \right)\!\!+\!\!C\sum\limits_{i=1}^{n}{{{\xi }_{i}}} \right\}, \\
s.t. \ \ \ \ & {{y}_{i}}({{\mathbf{w}}^{T}}{{\mathbf{x}}_{i}}+b)\ge 1-{{\xi }_{i}}, \forall i,\\
\ \ \ \ & {{\xi }_{i}}\ge 0, i=1,2,\cdots ,n, \\
\ \ \ \ & {{({{\mathbf{x}}_{i}}-\mathbf{\bar{x}})}^{T}}\mathbf{M}({{\mathbf{x}}_{i}}-\mathbf{\bar{x}})\le 1, \forall i,\\
\ \ \ \ & \mathbf{M}\succ 0.
\end{aligned}\end{aligned}$$
Denote $(\mathbf{\hat{w}},\hat{b},\mathbf{\hat{\xi }},\mathbf{\hat{M}},\hat{R})$ by the optimal solution to the problem in Eq. (\[eq:class\_radiu-margin-based-SVM-relaxation\]). Let $\mathbf{\tilde{M}}=\mathbf{\hat{M}}/{{\hat{R}}^{2}}$ and $\tilde{R}=1$. It is obvious to see that $(\mathbf{\hat{w}},\hat{b},\mathbf{\hat{\xi }},\mathbf{\tilde{M}},\tilde{R})$ is also the optimal solution to the problem in Eq. (\[eq:class\_radiu-margin-based-SVM-relaxation\]) because $F(\mathbf{\hat{w}},\hat{b},\mathbf{\hat{\xi }},\mathbf{\hat{M}},\hat{R})=F(\mathbf{\hat{w}},\hat{b},\mathbf{\hat{\xi }},\mathbf{\tilde{M}},\tilde{R})$.
Next we will show that $(\mathbf{\hat{w}},\hat{b},\mathbf{\hat{\xi }},\mathbf{\tilde{M}})$ is the optimal solution to the problem in Eq. (\[eq:class\_radiu-margin-based-SVM-relaxation-equivalentaion\]). If $(\mathbf{\hat{w}},\hat{b},\mathbf{\hat{\xi }},\mathbf{\tilde{M}})$ is not the optimal solution to Eq. (\[eq:class\_radiu-margin-based-SVM-relaxation-equivalentaion\]), there must exist some $({{\mathbf{w}}^{*}},{{b}^{*}},{{\mathbf{\xi }}^{*}},{{\mathbf{M}}^{*}})$ that satisfies all inequality constraints and $L({{\mathbf{w}}^{*}},{{b}^{*}},{{\mathbf{\xi }}^{*}},{{\mathbf{M}}^{*}})<L(\mathbf{\hat{w}},\hat{b},\mathbf{\hat{\xi }},\mathbf{\tilde{M}})$. Then we can define $\tilde{R}=1$ and have $F({{\mathbf{w}}^{*}},{{b}^{*}},{{\mathbf{\xi }}^{*}},{{\mathbf{M}}^{*}},\tilde{R})<F(\mathbf{\hat{w}},\hat{b},\mathbf{\hat{\xi }},\mathbf{\tilde{M}},\tilde{R})$, which is contradictory with the assumption that $(\mathbf{\hat{w}},\hat{b},\mathbf{\hat{\xi }},\mathbf{\tilde{M}},\tilde{R})$ is the optimal solution to Eq. (\[eq:class\_radiu-margin-based-SVM-relaxation\]). Thus, we can solve the problem in Eq. (\[eq:class\_radiu-margin-based-SVM-relaxation-equivalentaion\]) with the optimal solution $(\mathbf{\hat{w}},\hat{b},\mathbf{\hat{\xi }},\mathbf{\tilde{M}})$, and then obtain the optimal solution $(\mathbf{\hat{w}},\hat{b},\mathbf{\hat{\xi }},\mathbf{\tilde{M}},\tilde{R})$ to Eq. (\[eq:class\_radiu-margin-based-SVM-relaxation\]).
Without loss of generality, we assume $\tilde{R}=1$ and seek the corresponding optimal ${\bf w}$ and ${\bf M}$ by solving Eq. (\[eq:class\_radiu-margin-based-SVM-relaxation-equivalentaion\]). Moreover, to make the model robust against outliers and noisy samples, we combine the individual inequality constraints ${{({{\mathbf{x}}_{i}}-\mathbf{\bar{x}})}^{T}}\mathbf{M}({{\mathbf{x}}_{i}}-\mathbf{\bar{x}})\le 1,i=1,2,\cdots ,n$ into one integrated inequality constraint [@liu2013efficient], resulting in the following radius-margin based SVM model: $$\begin{aligned}
\label{eq:class_F-SVM-1}
\begin{aligned}
\min \limits_{\mathbf{w},b,\mathbf{\xi },\mathbf{M}} \ \ \ \ & \frac{1}{2}\left( {{\mathbf{w}}^{T}}{{\mathbf{M}}^{-1}}\mathbf{w} \right)+C\sum\limits_{i=1}^{n}{{{\xi }_{i}}}, \\
s.t. \ \ \ \ & {{y}_{i}}({{\mathbf{w}}^{T}}{{\mathbf{x}}_{i}}+b)\ge 1-{{\xi }_{i}}, \forall i,\\
\ \ \ \ & {{\xi }_{i}}\ge 0, i=1,2,\cdots ,n, \\
\ \ \ \ & \sum\limits_{i=1}^{n}{{{({{\mathbf{x}}_{i}}-\mathbf{\bar{x}})}^{T}}\mathbf{M}({{\mathbf{x}}_{i}}-\mathbf{\bar{x}})}\le \kappa, \\
\ \ \ \ & \mathbf{M}\succ 0.
\end{aligned}\end{aligned}$$ By defining the scattering matrix of the training set $\mathbf{S}=\sum\nolimits_{i=1}^{n}{({{\mathbf{x}}_{i}}-\mathbf{\bar{x}}){{({{\mathbf{x}}_{i}}-\mathbf{\bar{x}})}^{T}}}$, based on the Lagrangian multiplier method [@boyd2004convex], the problem expressed in Eq. (\[eq:class\_F-SVM-1\]) can be equivalently reformulated as the following F-SVM model: $$\begin{aligned}
\label{eq:class_F-SVM-2}
\begin{aligned}
\min \limits_ {\mathbf{w},b,\mathbf{\xi },\mathbf{M}} \ \ \ \ & \frac{1}{2}\left( {{\mathbf{w}}^{T}}{{\mathbf{M}}^{-1}}\mathbf{w} \right)+C\sum\limits_{i=1}^{n}{{{\xi }_{i}}}+\rho tr\left( \mathbf{MS} \right), \\
s.t. \ \ \ \ & {{y}_{i}}({{\mathbf{w}}^{T}}{{\mathbf{x}}_{i}}+b)\ge 1-{{\xi }_{i}}, \forall i,\\
\ \ \ \ & {{\xi }_{i}}\ge 0,i=1,2,\cdots ,n, \\
\ \ \ \ & \mathbf{M}\succ 0.
\end{aligned}\end{aligned}$$ where ${\rho}$ is the regularization parameter determined by ${\kappa}$. In the following, we prove that our F-SVM model is convex.
[**Theorem 3**]{}. The F-SVM model is a convex optimization problem. The proof can be found in [**Appendix B**]{}.
{width="6.0in"}
Alternating minimization
------------------------
In this section, we propose an efficient alternating minimization algorithm to solve the proposed F-SVM model, as illustrated in Fig. \[fig:framework\]. First, a semi-whitened PCA based initialization scheme is proposed to initialize ${\bf M}$. Then, the alternating minimization algorithm is adopted by updating $( {\bf w}, b)$ and ${\bf M}$ alternatively. Fixing ${\bf M}$, the model can be reformulated as the SVM in the transformed feature space, and be solved using the off-the-shelf SVM solvers to update $( {\bf w}, b)$. Fixing $( {\bf w}, b)$, the gradient descent method [@boyd2004convex] is adopted to update the matrix ${\bf M}$. When the algorithm converges, as shown in Fig. \[fig:framework\], we can learn both a better matrix M to reduce the radius of MEB in feature space and a max-margin classifier $( {\bf w}, b)$. The alternating optimization procedure is summarized in [**Algorithm 1**]{}.
Training set $\left\{ \left( {{\mathbf{x}}_{i}},{{y}_{i}} \right)\left| \forall i \right. \right\}$. Optimal ${\bf M}$ and $( {\bf w}, b)$. $k$ = 1, Initialize\
${{\mathbf{M}}_{k}}=\sqrt{{{\tau }'}}\mathbf{U\Xi }{{\mathbf{U}}^{T}}$,\
$\mathbf{\Xi }={diag}\{{{\left( {{\lambda }_{1}} \right)}^{-\frac{1}{2}}},{{\left( {{\lambda }_{2}} \right)}^{-\frac{1}{2}}},\cdots ,{{\left( {{\lambda }_{d
}} \right)}^{-\frac{1}{2}}}\}$, // Lines 7-9: updating $({\bf w}, b)$. Do Eigenvalue decomposition on ${{\mathbf{M}}_{k}}$: ${{\mathbf{M}}_{k}}=\mathbf{V\Sigma }{{\mathbf{V}}^{T}}$, Perform linear transformation on ${{\mathbf{x}}_{i}}$: ${{\mathbf{z}}_{i}}\!\leftarrow\!{{\mathbf{\Sigma }}^{{1}/{2}}}\!{{\mathbf{V}}^{T}}\!{{\mathbf{x}}_{i}}$, Update the SVM classifier $({\bf w}, b)$ based on ${\bf Z}$, // Lines 11-18: updating ${\bf M}$. $\mathbf{M}={{\mathbf{M}}_{k}}$, ${\it t}=1$, Compute the gradient of ${\bf M}$:\
$\nabla f(\mathbf{M})=-\frac{1}{2}\mathbf{M}_{{}}^{-1}\mathbf{w}{{\mathbf{w}}^{T}}\mathbf{M}_{{}}^{-1}+\rho \mathbf{S}$, Update ${\bf M}$: $\mathbf{M}={{\mathcal{P}}_{{{\mathcal{S}}_{+}}}}\left( \mathbf{M}-t\nabla f(\mathbf{M}) \right)$, Update the stepsize ${\it t}$: $t\leftarrow \beta *t$, ${{\mathbf{M}}_{k+1}}=\mathbf{M}$, $k\leftarrow k+1$,
### Initialization of M
Because the proposed F-SVM model is convex, alternating minimization can converge to global optimum for any initialization of ${\bf M}$ and $({{\bf w}, b})$, but proper initialization is helpful in improving the computational efficiency. Thus, by further relaxing the F-SVM model in Eq. (\[eq:class\_F-SVM-2\]), we propose a semi-whitened PCA based initialization method on ${\bf M}$.
Note that ${{\mathbf{w}}^{T}}{{\mathbf{M}}^{-1}}\mathbf{w}$ is upper bounded by [@hom1991topics]: $$\begin{aligned}
\label{eq:class_Initial_M}
\begin{aligned}
{\bf w}^{T}{{\bf M}^{-1}}{\bf w} & =tr\left( {\bf w}{\bf w}^{T}{{\bf M}^{-1}} \right)\\
& \le \left\| {\bf w} \right\|_{2}^{2}{{\left\|{{\bf M}^{-1}} \right\|}_{2}}\\
& \le \left\| {\bf w} \right\|_{2}^{2}{{\left\| {{{\bf M}}^{-1}} \right\|}_{*}}.
\end{aligned}\end{aligned}$$ where ${{\left\| \mathbf{A} \right\|}_{2}}$ and ${{\left\| \mathbf{A} \right\|}_{*}}$ denote the L$_2$-norm and the nuclear norm of a positive semi-definite matrix ${\bf A}$, respectively. Based on Eq. (\[eq:class\_F-SVM-2\]) and Eq. (\[eq:class\_Initial\_M\]), by setting $\mathbf{B}={{\mathbf{M}}^{-1}}$, the subproblem on ${\bf M}$ can be rewritten as the problem on ${\bf B}$ formulated as: $$\begin{aligned}
\label{eq:class_B}
\begin{aligned}
\min \limits_{\mathbf{B}} \ \ \ \ & L(\mathbf{B}) = {{\left\| \mathbf{B} \right\|}_{*}}+{\tau }'tr\left( {{\mathbf{B}}^{-1}}\mathbf{S} \right), \\
s.t. \ \ \ \ & \mathbf{B}\succ 0.
\end{aligned}\end{aligned}$$ where ${\tau }'={\rho }/{{{\left\| \mathbf{w} \right\|}^{2}}}$. The eigenvalue decomposition of ${\bf S}$ is $\mathbf{S}=\mathbf{U\Lambda }{{\mathbf{U}}^{T}}$, where $\mathbf{\Lambda }=diag({{\lambda }_{1}},{{\lambda }_{2}},\cdots ,{{\lambda }_{d}})$ (${{\lambda }_{1}}\ge {{\lambda }_{2}}\ge \cdots \ge {{\lambda }_{d}}\ge 0$), ${{\lambda }_{i}}$ and the [*i*]{}th column of ${\bf U}$ denote the [*i*]{}th eigenvalue and eigenvector, rsespectively. With ${\bf U}$ and $\mathbf{\Lambda }$, we define $\mathbf{\hat{B}}$ as: $$\begin{aligned}
\label{eq:class_15}
& \mathbf{\hat{B}}\!=\!\mathbf{U\Sigma}{{\mathbf{U}}^{T}}\!, \mathbf{\Sigma }\!=\!{diag}\{{{({\tau }'\!{{\lambda }_{1}})}^{\!1/\!2}},\cdots ,{{({\tau }'\!{{\lambda }_{d}})}^{\!1/\!2}}\}.\end{aligned}$$
[**Theorem 4**]{}. Given a SPD matrix ${\bf S}$ and ${\tau }'>0$, $\mathbf{\hat{B}}$ defined in Eq. (\[eq:class\_15\]) is the optimal solution to the problem: $$\begin{aligned}
\label{eq:class_16}
\mathbf{\hat{B}}=\arg \min \limits_{\mathbf{B}} \left\{ L(\mathbf{B},{\tau }') = {{\left\| \mathbf{B} \right\|}_{*}}+{\tau }'\left( tr\left( {{\mathbf{B}}^{-1}}\mathbf{S} \right) \right) \right\}.\end{aligned}$$ The proof can be found in [**Appendix C**]{}. With $\mathbf{\hat{B}}$, the initialization of ${\bf M}$ in Eq. (\[eq:class\_F-SVM-2\]) is then defined as: $$\begin{aligned}
\label{eq:class_17}
\begin{aligned}
{{\bf M}_{0}} \! = \! \sqrt{{\tau }'}\mathbf{U\Xi }{{\mathbf{U}}^{T}}, \mathbf{\Xi } \!=\! {diag}\{{\left( {{\lambda }_{1}} \right)}^{\!-1\!/\!2},\cdots ,{{\left( {{\lambda }_{d}} \right)}^{\!-\!1/\!2}}\}.
\end{aligned}\end{aligned}$$ Noted that we assume that ${{\left\| \mathbf{w} \right\|}^{2}}$ is known for the initialization of ${\bf M}$. From Eq. (\[eq:class\_17\]), ${{\left\| \mathbf{w} \right\|}^{2}}$ only affects the scale factor $\sqrt{{{\tau }'}}$ to the linear transformation. Thus, we simply let ${{\left\| \mathbf{w} \right\|}^{2}}=1$ in our implementation.
It is interesting to point out that ${{\bf M}_{0}}$ in Eq. (\[eq:class\_17\]) implies a semi-whitening PCA transformation because ${{\mathbf{M}}_{0}}=\mathbf{U}{{\mathbf{\Lambda }}^{-1/2}}{{\mathbf{U}}^{T}}$, where the linear transformation can then be defined as ${{\mathbf{\Lambda }}^{-1/4}}{{\mathbf{U}}^{T}}{{\mathbf{x}}_{i}}$. In literature [@hyvarinen2000independent], ${{\mathbf{\Lambda }}^{-1/2}}{{\mathbf{U}}^{T}}{{\mathbf{x}}_{i}}$ was called the PCA whitening transformation, and data whitening has been widely exploited in many applications, e.g., face recognition, object detection, and image classification [@yang2005ica; @krizhevsky2009learning; @hariharan2012discriminative; @girshick2013training]. Considering its connection with data whitening, semi-whitening is also expected to be effective in other tasks and applications.
### The subproblem on $( {\bf w}, b)$
Given ${\bf M}$, the F-SVM model can be formulated as: $$\begin{aligned}
\label{eq:class_18}
\begin{aligned}
\min \limits_{\mathbf{w},b,\mathbf{\xi }} \ \ \ \ & \frac{1}{2}{{\mathbf{w}}^{T}}\mathbf{Bw}+C\sum\limits_{i=1}^{n}{{{\xi }_{i}}}, \\
s.t. \ \ \ \ & {{y}_{i}}({{\mathbf{w}}^{T}}{{\mathbf{x}}_{i}}+b)\ge 1-{{\xi }_{i}}, \forall i,\\
\ \ \ \ & {{\xi }_{i}}\ge 0,i=1,2,\cdots ,n.
\end{aligned}\end{aligned}$$ where $\mathbf{B}={{\mathbf{M}}^{-1}}$. The eigenvalue decomposition of ${\bf B}$ is $\mathbf{B}=\mathbf{V\Sigma }{{\mathbf{V}}^{T}}$. By introducing $\mathbf{B}={{\mathbf{L}}^{T}}\mathbf{L}$, the transformation matrix ${\bf L}$ can be rewritten as to $\mathbf{L}={{\mathbf{\Sigma }}^{\frac{1}{2}}}{{\mathbf{V}}^{T}}$. Let ${{\mathbf{z}}_{i}}={{\mathbf{\Sigma }}^{-\frac{1}{2}}}{{\mathbf{V}}^{T}}{{\mathbf{x}}_{i}}$ and $\mathbf{v}=\mathbf{Lw}$. With simple algebra, the problem in Eq. (\[eq:class\_18\]) can be reformulated as: $$\begin{aligned}
\label{eq:class_19}
\begin{aligned}
\min \limits_{\mathbf{v},b,\mathbf{\xi }} \ \ \ \ & \frac{1}{2}{{\mathbf{v}}^{T}}\mathbf{v}+C\sum\limits_{i=1}^{n}{{{\xi }_{i}}}, \\
s.t. \ \ \ \ & {{y}_{i}}({{\mathbf{v}}^{T}}{{\mathbf{z}}_{i}}+b)\ge 1-{{\xi }_{i}}, \forall i,\\
\ \ \ \ & {{\xi }_{i}}\ge 0,i=1,2,\cdots ,n.
\end{aligned}\end{aligned}$$ which can be solved using the off-the-shelf SVM solvers. Given the solution ${\bf v}$, $\mathbf{w}=\mathbf{V}{{\mathbf{\Sigma }}^{-\frac{1}{2}}}\mathbf{v}$ can then be obtained.
### The subproblem on M
Given $( {\bf w}, b)$, the sub-problem on ${\bf M}$ can be reformulated as: $$\begin{aligned}
\label{eq:class_20}
\begin{aligned}
\min \limits_{\mathbf{M}} \ \ \ \ & f(\mathbf{M})=\frac{1}{2}\left( {{\mathbf{w}}^{T}}{{\mathbf{M}}^{-1}}\mathbf{w} \right)+\rho tr\left( \mathbf{MS} \right), \\
s.t. \ \ \ \ & \mathbf{M}\succ 0.
\end{aligned}\end{aligned}$$ Since the objective function in Eq. (\[eq:class\_20\]) is convex and differentiable with respect to ${\bf M}$, the gradient projection method [@boyd2004convex] is adopted to update ${\bf M}$. According to [@petersen2008matrix], the gradient of $f({\bf M})$ can be obtained by: $$\begin{aligned}
\label{eq:class_21}
\nabla f(\mathbf{M})=-\frac{1}{2}{{\mathbf{M}}^{-1}}\mathbf{w}{{\mathbf{w}}^{T}}{{\mathbf{M}}^{-1}}+\rho \mathbf{S}.\end{aligned}$$ As presented in [**Algorithm 1**]{}, we use gradient projection $$\begin{aligned}
\label{eq:class_22}
\mathbf{M}={{\mathcal{P}}_{{{\mathcal{S}}_{+}}}}\left( \mathbf{M}-t\nabla f(\mathbf{M}) \right)\end{aligned}$$ to update ${\bf M}$ by choosing proper stepsize ${\it t}$ and gradually decreasing it along with iterations, where ${{\mathcal{P}}_{{{\mathcal{S}}_{+}}}}\left( \cdot \right)$ projects a matrix onto the cone of positive semidefinite matrices.
Discussion
----------
The proposed F-SVM method has several interesting advantages while compared with the other radius-margin based SVMs, e.g., RMM [@shivaswamy2010maximum], MR-SVM [@do2009feature], R-SVM$^{+}$ [@do2013convex], R-SVM$_{\mu}^{+}$ [@do2013convex] and M-SVM [@zhu2012learning]. RMM [@shivaswamy2010maximum] is suggested to maximize the margin while restricting the spread of the data along the direction perpendicular to the separating hyperplane, while our F-SVM is proposed to minimize the convex relaxation of the radius-margin ratio. The generalization error is bounded by the radius and margin ratio, and the radius is determined by the spread along all possible directions rather than only the direction perpendicular to the separating hyperplane, making F-SVM theoretically more promising.
MR-SVM [@do2009feature], R-SVM$^{+}$ [@do2013convex] and R-SVM$_{\mu}^{+}$ [@do2013convex] aim to learn the diagonal feature transformation ${{\mathbf{D}}_{\mathbf{\mu }}}=Diag\left( \mathbf{\mu } \right)$ with ${{\mu }_{k}}\ge 0$, while F-SVM is developed for joint learning of feature transformation and SVM classifier. Both R-SVM$^{+}$ and R-SVM$_{\mu}^{+}$ need to solve a Quadratically Constrained Quadratic Programming (QCQP) optimization problem, which is computationally expensive than the alternating minimization method used in our F-SVM. Moreover, R-SVM$^{+}$ and R-SVM$_{\mu}^{+}$ adopted a tighter approximation ${{R}_{O}}$ of the radius. In F-SVM, a new approximation $\bar{R}$ of the radius is proposed, which is also tighter than that used in MR-SVM [@do2009feature]. Moreover, the individual inequality constraints on $\bar{R}$ are combined to improve the robustness against outliers. It is interesting to note that we have: $$\begin{aligned}
\label{eq:class_23}
\sum\limits_{i,j}{{{({{\mathbf{x}}_{i}}-{{\mathbf{x}}_{j}})}^{T}}\!\mathbf{M}\!({{\mathbf{x}}_{i}}-{{\mathbf{x}}_{j}})}\!=\!tr\left( \mathbf{M}{{\mathbf{S}}_{t}} \right)\!=\!4n\left( tr\left( \mathbf{MS} \right) \right),\end{aligned}$$ where ${{\mathbf{S}}_{t}}=\sum\nolimits_{i,j}{({{\mathbf{x}}_{i}}-{{\mathbf{x}}_{j}}){{({{\mathbf{x}}_{i}}-{{\mathbf{x}}_{j}})}^{T}}}$. Eq. (\[eq:class\_23\]) indicates that, if all the inequality constraints on ${{R}_{O}}$, i.e., ${{\left\| {{D}_{\sqrt{\mathbf{\mu }}}}{{\mathbf{x}}_{i}}-{{D}_{\sqrt{\mathbf{\mu }}}}{{\mathbf{x}}_{j}} \right\|}^{2}}\le r$, are combined into one integrated inequality constraint: $$\begin{aligned}
\label{eq:class_24}
\sum\limits_{i,j}\!\!{{{({{\mathbf{x}}_{i}}-{{\mathbf{x}}_{j}})}^{T}}\!D_{\sqrt{\mathbf{\mu }}}^{T}{{D}_{\sqrt{\mathbf{\mu }}}}\!({{\mathbf{x}}_{i}}-{{\mathbf{x}}_{j}})}\!=\!tr\left( D_{\sqrt{\mathbf{\mu }}}^{T}{{D}_{\sqrt{\mathbf{\mu }}}}{{\mathbf{S}}_{t}} \right)\!\le\!{\kappa }'.\end{aligned}$$ Let ${\kappa }'=4n\kappa$ and $\mathbf{M}=D_{\sqrt{\mathbf{\mu }}}^{T}{{D}_{\sqrt{\mathbf{\mu }}}}$. One can see that the integrated inequality constraint will be equivalent with that adopted in Eq. (\[eq:class\_F-SVM-1\]).
MSVM [@zhu2012learning] was developed for simultaneous learning of the linear transformation and SVM classifier, but the MSVM model is non-convex and solved using gradient projection. Moreover, although Zhu et al. [@zhu2012learning] claimed that $R=\bar{R}$, as discussed in Section 3.1, $\bar{R}$ is only a lower bound of $R$ and counter examples can be easily found to illustrate $R\ne \bar{R}$. Compared with MSVM [@zhu2012learning], the F-SVM model is convex and robust against noise and outliers, and can be efficient solved using the optimization method introduced in Section 3.2.
Kernelization of F-SVM
======================
With the incorporation of kernel principal component analysis, linear F-SVM can be extended to kernel version for nonlinear classification. First, we show that kernel SVM is equivalent to perform linear SVM in the kernel PCA space. Then, kernel F-SVM is introduced by conducting linear F-SVM in the kernel PCA space.
Let the kernel function be $K({{\mathbf{x}}_{i}},{{\mathbf{x}}_{j}})=\varphi {{({{\mathbf{x}}_{i}})}^{T}}\varphi ({{\mathbf{x}}_{j}})$, where $\varphi \left( \mathbf{x} \right)$ defines an implicit mapping from the data space to high or infinite dimensional feature space. For the training set $\mathcal{S}=\left\{ \left( {{\mathbf{x}}_{1}},{{y}_{1}} \right),...,\left( {{\mathbf{x}}_{n}},{{y}_{n}} \right) \right\}$, we use $\mathbf{W}=\left[ {{\mathbf{w}}_{1}},{{\mathbf{w}}_{2}},...,{{\mathbf{w}}_{D}} \right]$ to denote all the PCA eigenvectors corresponding to positive eigenvalues. Let $\mathbf{\bar{W}}$ be a set of basis vectors in the complementary space of $\mathbf{W}$. Assuming the training set is centered, for any ${{\bf x}_{i}}$, we have ${{\mathbf{\bar{W}}}^{T}}\varphi ({{\mathbf{x}}_{i}})=0$, and thus can get: $$\begin{aligned}
\label{eq:class_25}
\begin{aligned}
K({{\mathbf{x}}_{i}},{{\mathbf{x}}_{j}})&\!=\!\varphi {{({{\mathbf{x}}_{i}})}^{T}}\mathbf{W}{{\mathbf{W}}^{T}}\varphi ({{\mathbf{x}}_{j}})\!+\!\varphi {{({{\mathbf{x}}_{i}})}^{T}}\mathbf{\bar{W}}{{\mathbf{\bar{W}}}^{T}}\varphi ({{\mathbf{x}}_{j}})\\
&=\varphi {{({{\mathbf{x}}_{i}})}^{T}}\mathbf{W}{{\mathbf{W}}^{T}}\varphi ({{\mathbf{x}}_{j}}).
\end{aligned}\end{aligned}$$ Let ${{\mathbf{f}}_{i}}={{\mathbf{W}}^{T}}\varphi ({{\mathbf{x}}_{i}})$. The dual problem of SVM in the kernel PCA space can be formulated as: $$\begin{aligned}
\label{eq:class_26}
\begin{aligned}
\max \limits_{\mathbf{\alpha }} \ \ \ \ & Q\left( \mathbf{\alpha } \right)=\sum\nolimits_{i=1}^{n}{{{\alpha}_{i}}}-\frac{1}{2}\sum\nolimits_{i,j=1}^{n}{{{\alpha}_{i}}{{\alpha}_{j}}{{y}_{i}}{{y}_{j}}\left\langle {{\mathbf{f}}_{i}},{{\mathbf{f}}_{j}} \right\rangle }, \\
s.t. \ \ \ \ & \sum\nolimits_{i=1}^{n}{{{\alpha }_{i}}{{y}_{i}}} = 0, \\
\ \ \ \ & 0\le {{\alpha }_{i}}\le C,\ \ i=1,...,n.
\end{aligned}\end{aligned}$$ where $\left\langle {{\mathbf{f}}_{i}},{{\mathbf{f}}_{j}} \right\rangle =K\left( {{\mathbf{x}}_{i}},{{\mathbf{x}}_{j}} \right)$. Therefore, kernel SVM is equivalent to performing linear SVM in the kernel PCA space. To extend F-SVM to its kernelized version, we first project each training sample ${{\bf x}_i}$ to the kernel PCA space ${{\mathbf{f}}_{i}}={{\mathbf{W}}^{T}}\varphi ({{\mathbf{x}}_{i}})$, and then solve the following F-SVM model: $$\begin{aligned}
\label{eq:class_27}
\begin{aligned}
\min \limits_{\mathbf{w},b,\mathbf{\xi },\mathbf{M}} \ \ \ \ & \frac{1}{2}\left( {{\mathbf{w}}^{T}}{{\mathbf{M}}^{-1}}\mathbf{w} \right)+C\sum\limits_{i=1}^{n}{{{\xi}_{i}}}+\rho tr\left( \mathbf{M}{{\mathbf{S}}_{f}} \right), \\
s.t. \ \ \ \ & {{y}_{i}}({{\mathbf{w}}^{T}}{{\mathbf{f}}_{i}}+b)\ge 1-{{\xi }_{i}}, \forall i,\\
\ \ \ \ & {{\xi }_{i}}\ge 0,i=1,2,\cdots ,n, \\
\ \ \ \ & \mathbf{M}\succ 0.
\end{aligned}\end{aligned}$$ where ${{\mathbf{S}}_{f}}=\sum\nolimits_{i=1}^{n}{{{\mathbf{f}}_{i}}\mathbf{f}_{i}^{T}}$. [**Algorithm 1**]{} can be adopted to solve the model in Eq. (27). In our implementation, instead of using all the eigenvectors, we also consider to employ the PCA eigenvectors corresponding to first [*d*]{} largest eigenvalues, i.e., $\mathbf{W}=\left[ {{\mathbf{w}}_{1}},{{\mathbf{w}}_{2}},...,{{\mathbf{w}}_{D}} \right]$, and Section 5.2 reports the empirical result on the influence of [*d*]{} on classification accuracy.
Experiments
===========
In this section, we use both the UCI machine learning datasets and the Labeled Faces in the Wild (LFW) database to evaluate the proposed F-SVM method, and compare our F-SVM with the competing methods, including SVM and several representative radius-margin based SVM methods, i.e., RMM [@shivaswamy2010maximum], R-SVM$^{+}$ [@do2013convex] and R-SVM$_{\mu}^{+}$ [@do2013convex]. MR-SVM [@do2009feature] and MSVM [@zhu2012learning] are not considered in our experiments because their source codes are not publicly available. The 10-fold cross validation (CV) is adopted to determine the optimal values of hyper-parameters for each method. The mean classification accuracy is adopted by averaging the 100 runs of the 10-fold CV. The methods are evaluated by two performance indicators: accuracy and training time (seconds, [*s*]{}).
Dataset \# of samples \# of classes \# of attributes
--------------- --------------- --------------- ------------------
Breast cancer 286 2 9
Diabetes 768 2 8
Solar Flare 144 3 9
German 1000 2 20
Heart 270 2 13
Image 2310 7 19
Ringnorm 7400 2 20
Splice 3190 3 60
Thyroid 215 3 5
Twonorm 7400 2 20
Waveform 5000 3 21
: Summary of the UCI datasets used in the experiments.[]{data-label="table1"}
Evaluation on linear F-SVM using the UCI datasets
-------------------------------------------------
We evaluate the performance of linear F-SVM on the 11 datasets from the UCI machine learning repository, where the reason to choose them is that they had been widely adopted for evaluating SVM and kernel methods [@ratsch2001soft; @scholkopft1999fisher; @cawley2003efficient]. Table \[table1\] provides a brief summary of these UCI datasets, which includes 6 2-class problems and 5 multi-class problems. Tables \[table2\] and \[table3\] list the mean classification accuracy and training time of five linear classifiers, i.e., linear SVM, linear RMM [@shivaswamy2010maximum], linear R-SVM$^{+}$ [@do2013convex], linear R-SVM$_{\mu}^{+}$ [@do2013convex], and linear F-SVM. RMM [@shivaswamy2010maximum], R-SVM$^{+}$ [@do2013convex], R-SVM$_{\mu}^{+}$ [@do2013convex], and F-SVM consider both margin and radius information, while SVM only considers margin. As shown in Table \[table2\], the radius-margin based SVM methods generally outperform SVM in terms of classification accuracy, which indicates that the incorporation of radius can improve the classification performance. As listed in Table \[table3\], the training time of SVM is much less than the other four methods, indicating that the introduction of radius makes the model more complex to train.
Dataset SVM RMM R-SVM$^{+}$ R-SVM$_{\mu}^{+}$ F-SVM
--------------- ------- ----------- ------------- ------------------- -----------
Breast cancer 71.40 70.19 71.54 71.14 **71.68**
Diabetes 76.57 76.29 76.67 76.42 **77.00**
Solar Flare 67.66 67.38 67.66 67.66 **67.69**
German 75.58 75.99 76.01 75.87 **76.04**
Heart 83.61 83.64 83.83 83.96 **84.02**
Image 83.77 84.12 **84.39** 83.97 84.32
Ringnorm 75.41 75.78 75.63 75.43 **77.05**
Splice 84.54 **85.05** 84.67 84.74 84.81
Thyroid 89.76 91.19 **91.23** 90.09 86.81
Twonorm 96.92 **97.79** 97.41 97.39 97.08
Waveform 86.95 **88.54** 88.51 86.88 86.76
: Comparison of the average classification accuracy (%) of linear SVM, linear RMM [@shivaswamy2010maximum], linear R-SVM$^{+}$ [@do2013convex], linear R-SVM$_{\mu}^{+}$ [@do2013convex], and linear F-SVM.[]{data-label="table2"}
Dataset SVM RMM R-SVM$^{+}$ R-SVM$_{\mu}^{+}$ F-SVM
--------------- ------------------------------------- ------------------------------------- ------------------------------------- ------------------------------------- -------------------------------------
Breast cancer $\!\!6.20\!\!\times\!\!10^{\!-\!3}$ $\!\!3.70\!\!\times\!\!10^{\!+\!1}$ $\!\!2.29\!\!\times\!\!10^{\!+\!1}$ $\!\!0.20\!\!\times\!\!10^{\!+\!0}$ $\!\!6.60\!\!\times\!\!10^{\!-\!3}$
Diabetes $\!\!1.12\!\!\times\!\!10^{\!-\!2}$ $\!\!2.63\!\!\times\!\!10^{\!+\!2}$ $\!\!2.80\!\!\times\!\!10^{\!+\!2}$ $\!\!1.20\!\!\times\!\!10^{\!+\!3}$ $\!\!1.38\!\!\times\!\!10^{\!-\!2}$
Solar Flare $\!\!7.79\!\!\times\!\!10^{\!-\!4}$ $\!\!1.38\!\!\times\!\!10^{\!+\!1}$ $\!\!3.41\!\!\times\!\!10^{\!+\!1}$ $\!\!4.61\!\!\times\!\!10^{\!+\!8}$ $\!\!6.60\!\!\times\!\!10^{\!-\!4}$
German $\!\!4.83\!\!\times\!\!10^{\!-\!2}$ $\!\!3.77\!\!\times\!\!10^{\!+\!2}$ $\!\!7.38\!\!\times\!\!10^{\!+\!2}$ $\!\!1.55\!\!\times\!\!10^{\!+\!2}$ $\!\!5.95\!\!\times\!\!10^{\!-\!2}$
Heart $\!\!1.50\!\!\times\!\!10^{\!-\!3}$ $\!\!3.06\!\!\times\!\!10^{\!+\!1}$ $\!\!5.09\!\!\times\!\!10^{\!+\!1}$ $\!\!0.29\!\!\times\!\!10^{\!+\!0}$ $\!\!1.80\!\!\times\!\!10^{\!-\!3}$
Image $\!\!2.29\!\!\times\!\!10^{\!+\!0}$ $\!\!2.21\!\!\times\!\!10^{\!+\!3}$ $\!\!5.79\!\!\times\!\!10^{\!+\!3}$ $\!\!1.48\!\!\times\!\!10^{\!+\!3}$ $\!\!2.40\!\!\times\!\!10^{\!+\!0}$
Ringnorm $\!\!1.42\!\!\times\!\!10^{\!+\!1}$ $\!\!7.46\!\!\times\!\!10^{\!+\!3}$ $\!\!2.63\!\!\times\!\!10^{\!+\!4}$ $\!\!1.13\!\!\times\!\!10^{\!+\!3}$ $\!\!1.42\!\!\times\!\!10^{\!+\!1}$
Splice $\!\!4.02\!\!\times\!\!10^{\!+\!1}$ $\!\!2.77\!\!\times\!\!10^{\!+\!3}$ $\!\!1.09\!\!\times\!\!10^{\!+\!4}$ $\!\!5.69\!\!\times\!\!10^{\!+\!3}$ $\!\!1.12\!\!\times\!\!10^{\!+\!0}$
Thyroid $\!\!8.70\!\!\times\!\!10^{\!-\!3}$ $\!\!2.06\!\!\times\!\!10^{\!+\!1}$ $\!\!2.79\!\!\times\!\!10^{\!+\!1}$ $\!\!0.21\!\!\times\!\!10^{\!+\!0}$ $\!\!1.59\!\!\times\!\!10^{\!-\!1}$
Twonorm $\!\!5.99\!\!\times\!\!10^{\!+\!0}$ $\!\!3.14\!\!\times\!\!10^{\!+\!1}$ $\!\!9.91\!\!\times\!\!10^{\!+\!3}$ $\!\!1.16\!\!\times\!\!10^{\!+\!2}$ $\!\!3.16\!\!\times\!\!10^{\!-\!1}$
Waveform $\!\!1.30\!\!\times\!\!10^{\!-\!1}$ $\!\!6.54\!\!\times\!\!10^{\!+\!4}$ $\!\!5.20\!\!\times\!\!10^{\!+\!3}$ $\!\!1.19\!\!\times\!\!10^{\!+\!1}$ $\!\!6.71\!\!\times\!\!10^{\!-\!1}$
: Comparison of the training time (${\it s}$) of linear SVM, linear RMM [@shivaswamy2010maximum], linear R-SVM$^{+}$ [@do2013convex], linear R-SVM$_{\mu}^{+}$ [@do2013convex], and linear F-SVM.[]{data-label="table3"}
We further compare linear F-SVM with the competing methods. From Table \[table2\], F-SVM achieves higher classification accuracy than SVM on 9, RMM [@shivaswamy2010maximum] on 7, R-SVM$^{+}$ [@do2013convex] on 7, and R-SVM$_{\mu}^{+}$ [@do2013convex] on 8 of the 11 datasets. The better classification accuracy of our F-SVM should be attributed to that: $(i)$ compared with SVM, F-SVM incorporates radius in the convex model; $(ii)$ unlike RMM [@shivaswamy2010maximum], our F-SVM considers the spread along all directions rather than only the direction perpendicular to the separating hyperplane. $(iii)$ instead of feature reweighting and selection in R-SVM$^{+}$ [@do2013convex] and R-SVM$_{\mu}^{+}$ [@do2013convex], general linear transformation is learned in F-SVM. To improve the efficiency of F-SVM in training, we adopt the warm-start strategy, where the solution $( {\bf w}, b)$ of the previous iteration is used as the initialization of the next iteration. From Table \[table3\], one can see that our F-SVM is only a little slower than SVM, but is much more efficient than the other competing methods in training. F-SVM is about $10^3\!\!\sim\!\!10^4$ times faster than RMM [@shivaswamy2010maximum], R-SVM$^{+}$ and R-SVM$_{\mu}^{+}$ in [@do2013convex]. In summary, F-SVM obtains the best classification accuracy among all competing methods, and is more efficient in training than the other radius-margin based SVM methods.
Dataset SVM RMM R-SVM$^{+}$ R-SVM$_{\mu}^{+}$ F-SVM
--------------- ------- ----------- ------------- ------------------- -----------
Breast cancer 73.74 **74.15** 71.54 71.14 73.95
Diabetes 76.83 74.97 77.05 76.80 **78.84**
Solar Flare 67.64 66.33 **67.66** 67.54 **67.66**
German 76.36 76.58 76.01 75.93 **76.90**
Heart 83.43 82.19 83.83 83.96 **84.25**
Image 97.14 96.41 84.39 83.97 **96.93**
Ringnorm 98.41 86.16 75.63 75.43 **98.58**
Splice 90.16 89.34 84.85 84.97 **90.55**
Thyroid 95.91 95.86 91.23 91.31 **96.13**
Twonorm 97.59 95.43 97.41 97.39 **97.79**
Waveform 89.75 **92.60** 88.51 89.38 90.95
: Comparison of the average classification accuracy (%) of kernel SVM, kernel RMM [@shivaswamy2010maximum], kernel R-SVM$^{+}$ [@do2013convex], kernel R-SVM$_{\mu}^{+}$ [@do2013convex], and kernel F-SVM.[]{data-label="table4"}
Dataset SVM RMM R-SVM$^{+}$ R-SVM$_{\mu}^{+}$ F-SVM
--------------- ------------------------------------- ------------------------------------- ------------------------------------- ------------------------------------- -------------------------------------
Breast cancer $\!\!1.90\!\!\times\!\!10^{\!-\!3}$ $\!\!1.00\!\!\times\!\!10^{\!+\!3}$ $\!\!2.24\!\!\times\!\!10^{\!+\!1}$ $\!\!1.83\!\!\times\!\!10^{\!+\!1}$ $\!\!6.60\!\!\times\!\!10^{\!-\!3}$
Diabetes $\!\!1.11\!\!\times\!\!10^{\!-\!2}$ $\!\!5.70\!\!\times\!\!10^{\!+\!2}$ $\!\!3.47\!\!\times\!\!10^{\!+\!1}$ $\!\!5.02\!\!\times\!\!10^{\!+\!1}$ $\!\!2.57\!\!\times\!\!10^{\!-\!2}$
Solar Flare $\!\!1.10\!\!\times\!\!10^{\!-\!3}$ $\!\!9.01\!\!\times\!\!10^{\!+\!1}$ $\!\!3.52\!\!\times\!\!10^{\!+\!1}$ $\!\!4.58\!\!\times\!\!10^{\!+\!3}$ $\!\!1.80\!\!\times\!\!10^{\!-\!3}$
German $\!\!4.70\!\!\times\!\!10^{\!-\!2}$ $\!\!1.17\!\!\times\!\!10^{\!+\!3}$ $\!\!5.30\!\!\times\!\!10^{\!+\!2}$ $\!\!1.83\!\!\times\!\!10^{\!+\!3}$ $\!\!6.08\!\!\times\!\!10^{\!-\!2}$
Heart $\!\!1.00\!\!\times\!\!10^{\!-\!3}$ $\!\!3.46\!\!\times\!\!10^{\!+\!1}$ $\!\!5.08\!\!\times\!\!10^{\!+1\!}$ $\!\!1.15\!\!\times\!\!10^{\!+\!3}$ $\!\!1.20\!\!\times\!\!10^{\!-\!3}$
Image $\!\!5.30\!\!\times\!\!10^{\!-\!2}$ $\!\!4.31\!\!\times\!\!10^{\!+\!3}$ $\!\!3.44\!\!\times\!\!10^{\!+\!3}$ $\!\!1.59\!\!\times\!\!10^{\!+\!3}$ $\!\!5.98\!\!\times\!\!10^{\!-\!2}$
Ringnorm $\!\!3.08\!\!\times\!\!10^{\!-\!1}$ $\!\!2.93\!\!\times\!\!10^{\!+\!4}$ $\!\!3.59\!\!\times\!\!10^{\!+\!4}$ $\!\!6.25\!\!\times\!\!10^{\!+\!1}$ $\!\!3.75\!\!\times\!\!10^{\!-\!1}$
Splice $\!\!6.28\!\!\times\!\!10^{\!-\!1}$ $\!\!6.49\!\!\times\!\!10^{\!+\!3}$ $\!\!3.13\!\!\times\!\!10^{\!+\!4}$ $\!\!6.69\!\!\times\!\!10^{\!+\!3}$ $\!\!6.07\!\!\times\!\!10^{\!-\!1}$
Thyroid $\!\!4.22\!\!\times\!\!10^{\!-\!4}$ $\!\!4.51\!\!\times\!\!10^{\!+\!1}$ $\!\!4.13\!\!\times\!\!10^{\!+\!1}$ $\!\!7.13\!\!\times\!\!10^{\!+\!0}$ $\!\!5.95\!\!\times\!\!10^{\!-\!4}$
Twonorm $\!\!2.81\!\!\times\!\!10^{\!-\!1}$ $\!\!3.26\!\!\times\!\!10^{\!+\!4}$ $\!\!1.97\!\!\times\!\!10^{\!+\!4}$ $\!\!2.62\!\!\times\!\!10^{\!+\!1}$ $\!\!3.83\!\!\times\!\!10^{\!-\!1}$
Waveform $\!\!5.79\!\!\times\!\!10^{\!-\!1}$ $\!\!4.24\!\!\times\!\!10^{\!+\!4}$ $\!\!1.04\!\!\times\!\!10^{\!+\!4}$ $\!\!3.18\!\!\times\!\!10^{\!+\!1}$ $\!\!6.19\!\!\times\!\!10^{\!-\!1}$
: Comparison of the training time (${\it s}$) of kernel SVM, kernel RMM [@shivaswamy2010maximum], kernel R-SVM$^{+}$ [@do2013convex], kernel R-SVM$_{\mu}^{+}$ [@do2013convex], and kernel F-SVM.[]{data-label="table5"}
Evaluation on kernel F-SVM using the UCI datasets
-------------------------------------------------
In this subsection, we evaluate the performance of kernel F-SVM on the 11 UCI datasets. The Gaussian RBF kernel is adopted in our experiments, which includes an extra kernel parameter ${\sigma}$. As discussed in Section 4, we also consider the number of kernel PCA components in kernel F-SVM. Using four datasets, i.e., [*Breast cancer*]{}, [*Thyroid*]{}, [*Heart*]{}, and [*German*]{}, Fig. \[fig:fig2\] illustrates the classification accuracy of kernel SVM and kernel F-SVM under different kernel PCA dimensions. It is interesting to note that, the proper decreasing of kernel PCA dimension can consistently improve the classification accuracy both for kernel SVM and kernel F-SVM. Also from Fig. \[fig:fig2\] one can see that the kernel F-SVM is superior to kernel SVM under different dimensions. One possible explanation may be that the decreasing of kernel PCA dimension would make the learned transformation more stable.
Tables \[table4\] and \[table5\] list the mean classification accuracy and training time of five linear classifiers, i.e., kernel SVM, kernel RMM [@shivaswamy2010maximum], kernel R-SVM$^{+}$ [@do2013convex], kernel R-SVM$_{\mu}^{+}$ [@do2013convex], and kernel F-SVM. For kernel methods, the superiority of F-SVM against the competing methods is more significant. The Kernel F-SVM outperforms kernel SVM on 10, kernel RMM [@shivaswamy2010maximum] on 9, kernel R-SVM$^{+}$ [@do2013convex] on 11, and kernel R-SVM$_{\mu}^{+}$ [@do2013convex] on 11 of all the 11 datasets in terms of classification accuracy. By training time, the kernel F-SVM is a little slower than SVM, but is about $10^3\!\!\sim\!\!10^4$ times faster than the other competing methods.
Results on the LFW Database
---------------------------
In this subsection, the LFW database is used to evaluate F-SVM for face verification. The database consists of more than 13,233 face images from 5,749 persons. The face images in the LFW database were collected from the Internet, and vary in pose, illumination, expression, and age, making LFW very suitable for studying unconstrained face verification. The face recognition method can be evaluated with two test protocols for LFW: the restricted and the unrestricted settings. Under the restricted setting, the only available information is whether each pair of training images is matched or not, and the performance of the face verification method is evaluated by 10-fold cross validation on a set of 3000 positive and 3000 negative image pairs.
In our experiment, we adopt the restricted setting with the face images aligned by the funneling method [@huang2007unsupervised]. Fig. \[fig:fig3\] shows some examples of similar and dissimilar pairs. We extract two kinds of features for each face image: SIFT feature and attribute feature, and compare F-SVM with SVM, RMM [@shivaswamy2010maximum], R-SVM$^{+}$ [@do2013convex], and R-SVM$_{\mu}^{+}$ [@do2013convex], and several representative face verification methods, including LDML [@guillaumin2009you], Nowak [@nowak2007learning], V1-like/MKL [@pinto2009far], and MERL+Nowak [@huang2008lfw].
Fig. \[fig:fig4\] shows the verification accuracies of SVM and F-SVM under different PCA dimensions by using the attribute feature and the combined features of SIFT and attributes, respectively. F-SVM using the combined features of SIFT and attributes (F-SVM-combined) achieves its best performance of 83.25% when the dimension d = 300, and 82.58% when the dimension d = 73 using the attribute features (F-SVM-attribute). SVM using the combined features of SIFT and attributes (SVM-combined) achieves its best performance of 81.90% when the dimension d = 400, and 80.12% when the dimension d = 73 using the attribute features (SVM-attribute). Thus, F-SVM can get better accuracy than SVM on the LFW database.
We further compare F-SVM with several other face verification methods. Table 6 lists the accuracy of F-SVM, SVM, RMM [@shivaswamy2010maximum], R-SVM$^{+}$ [@do2013convex], R-SVM$_{\mu}^{+}$ [@do2013convex], LDML [@guillaumin2009you],Nowak [@nowak2007learning], V1-like/MKL [@pinto2009far], and MERL+Nowak [@huang2008lfw]. We report the accuracy of F-SVM, SVM, RMM [@shivaswamy2010maximum], R-SVM$^{+}$ [@do2013convex], R-SVM$_{\mu}^{+}$ [@do2013convex], LDML [@guillaumin2009you] using the attribute and the SIFT features, report the accuracy of Nowak [@nowak2007learning] and MERL+Nowak [@huang2008lfw] using SIFT and geometry feature, and report the accuracy of V1-like/MKL [@pinto2009far] using the V1-like features. For either the combined features of SIFT and attributes or the attribute features, F-SVM achieves higher accuracy than SVM, RMM [@shivaswamy2010maximum],R-SVM$^{+}$ [@do2013convex], R-SVM$_{\mu}^{+}$ [@do2013convex], LDML [@guillaumin2009you] separately from Table \[table6\]. Fig. \[fig:fig5\] shows the ROC curves of the competing methods. Also one can see that F-SVM-combined gets better performance than other face verification methods.
Method, restricted Accuracy
--------------------------------------------- ----------- -- -- -- --
F-SVM-combined
F-SVM-attribute
R-SVM$_{\mu}^{+}$-combined [@do2013convex] $ 81.90 $
R-SVM$_{\mu}^{+}$-attribute [@do2013convex] $ 81.50 $
R-SVM$^{+}$-combined [@do2013convex] $81.67$
R-SVM$^{+}$-attribute [@do2013convex] $81.46$
RMM-combined [@shivaswamy2010maximum] $81.30$
RMM-attribute [@shivaswamy2010maximum] $81.27$
SVM-combined $81.90$
SVM-attribute $80.12$
Nowak [@nowak2007learning] $73.93$
LDML [@guillaumin2009you] $79.27$
V1-like/MKL [@pinto2009far] $79.35$
MERL+Nowak [@huang2008lfw] $76.18$
: Comparison of accuracy obtained using different face verification methods. The top two results are shown in red and blue fonts, respectively.[]{data-label="table6"}
Conclusion
==========
In this paper, we proposed a convex radius-margin based SVM model (F-SVM) for joint learning of feature transformation and SVM classifier. For the formulation of F-SVM, lower and upper bounds of the radius of MEB are introduced to derive a novel approximation of radius-margin ratio, and all the individual inequality constraints are combined into one integrated inequality constraint, resulting in a convex relaxation of the radius-margin based SVM model. For model optimization, a semi-whitened PCA based method is proposed for the initialization of the learned transformation, and an alternating minimization algorithm is adopted to learn the feature transformation and SVM classifier. Further, F-SVM is kernelized by using kernel PCA. Our experimental results show that, F-SVM obtains higher classification accuracy than SVM and the state-of-the-art radius-margin based SVM methods, and is more efficient in training than the other radius-margin based SVM methods, e.g., RMM [@shivaswamy2010maximum], R-SVM$^{+}$ [@do2013convex] and R-SVM$_{\mu}^{+}$ [@do2013convex]. In our future work, we will extend the proposed relaxed radius-margin based error bound to other classification methods and extend the proposed model for learning other forms of feature transformation tailored for specific applications.
[**Lemma A.1**]{}. $\bar{R}\ge R$.
Based on the definition of the radius, we get $$\begin{aligned}
\label{eq:class_A.1.1}
\begin{aligned}
\nonumber {{R}^{2}} & =\min \limits_{{{\mathbf{x}}_{0}}} \max \limits_{i}\,{\left\| \mathbf{A}{{\mathbf{x}}_{i}}-\mathbf{A}{{\mathbf{x}}_{0}} \right\|_{2}^{2}} \\
& \le \max \limits_{i}{\left\| \mathbf{A}{{\mathbf{x}}_{i}}-\mathbf{A\bar{x}} \right\|_{2}^{2}} \\
& ={{{\bar{R}}}^{2}}.
\end{aligned}\end{aligned}$$
Denote ${{R}_{p}}$ by the maximum pairwise distance. We have $$\begin{aligned}
\label{eq:class_A.1.2}
\nonumber {{R}_{p}}=\max \limits_{i,j}\{\left\| \mathbf{A}{{\mathbf{x}}_{i}}-\mathbf{A}{{\mathbf{x}}_{j}} \right\|_{2}^{2}\}.\end{aligned}$$
[**Lemma A.2**]{} [@do2013convex]. $R\ge {{R}_{p}}/2$.
[**Lemma A.3**]{}. $\bar{R}\le {{R}_{p}}$.
Let ${{\mathbf{{x}'}}_{i}}=\mathbf{A}{{\mathbf{x}}_{i}}-\mathbf{A\bar{x}}$. We have $\mathbf{A}{{\mathbf{x}}_{i}}-\mathbf{A}{{\mathbf{x}}_{j}}={{\mathbf{{x}'}}_{i}}-{{\mathbf{{x}'}}_{j}}$. Based on the definition of $\bar{R}$ $$\begin{aligned}
\label{eq:class_A.3.1}
\nonumber {{\bar{R}}^{2}}=\max \limits_{i} \left\{ {{\left\| {{{\mathbf{{x}'}}}_{i}} \right\|}^{2}} \right\}={{\left\| {{{\mathbf{{x}'}}}_{{{i}^{*}}}} \right\|}^{2}},\end{aligned}$$ we will prove that there exist some ${{j}^{*}}$ which makes ${{\left\| {{{\mathbf{{x}'}}}_{{{i}^{*}}}}-{{{\mathbf{{x}'}}}_{{{j}^{*}}}} \right\|}^{2}}\ge {{\left\| {{{\mathbf{{x}'}}}_{{{i}^{*}}}} \right\|}^{2}}$. Based on the definition of $\mathbf{\bar{x}}$, we have $\sum\nolimits_{j}{{{{\mathbf{{x}'}}}_{j}}}=0$. Then, we derive $$\begin{aligned}
\label{eq:class_A.3.2}
\nonumber \sum\nolimits_{j}{{{{\mathbf{{x}'}}}_{j}}{{{\mathbf{{x}'}}}_{{{i}^{*}}}}}=0\Rightarrow \min \limits_{j}\left\{ {{{\mathbf{{x}'}}}_{j}}{{{\mathbf{{x}'}}}_{{{i}^{*}}}} \right\}={{\mathbf{{x}'}}_{{{j}^{*}}}}{{\mathbf{{x}'}}_{{{i}^{*}}}}\le 0.\end{aligned}$$ Since ${{\left\| {{{\mathbf{{x}'}}}_{{{j}^{*}}}} \right\|}^{2}}\ge 0$ and $-2{{\mathbf{{x}'}}_{{{j}^{*}}}}{{\mathbf{{x}'}}_{{{i}^{*}}}}\ge 0$, one can easily see that $$\begin{aligned}
\label{eq:class_A.3.3}
\nonumber {{\left\| {{{\mathbf{{x}'}}}_{{{i}^{*}}}}-{{{\mathbf{{x}'}}}_{{{j}^{*}}}} \right\|}^{2}}\ge {{\left\| {{{\mathbf{{x}'}}}_{{{i}^{*}}}} \right\|}^{2}}.\end{aligned}$$ Based on the definition of ${{R}_{p}}$: $$\begin{aligned}
\label{eq:class_A.3.4}
\nonumber R_{p}^{2}\ge {{\left\| {{{\mathbf{{x}'}}}_{{{i}^{*}}}}-{{{\mathbf{{x}'}}}_{{{j}^{*}}}} \right\|}^{2}}.\end{aligned}$$
Combining the two inequalities above, we can prove $\bar{R}\le {{R}_{p}}$.
Finally, by combining [**Lemmas**]{} 1$\sim$3, we obtain the following theorem:
[**Theorem A.1**]{}. The margin R is bounded by $\bar{R}$ by: $$\begin{aligned}
\label{eq:class_Theorem A.1}
\nonumber \frac{1}{2}\overline{R}\le R\le \overline{R}.\end{aligned}$$
[**Lemma B.1**]{} [@petersen2008matrix]. Given two symmetric positive definite (SPD) matrices ${\bf A}$ and ${\bf B}$, we have $$\begin{aligned}
\label{eq:class_28}
{{\left( \mathbf{A}+\mathbf{B} \right)}^{-1}}={{\mathbf{A}}^{-1}}-{{\mathbf{A}}^{-1}}{{({{\mathbf{A}}^{-1}}+{{\mathbf{B}}^{-1}})}^{-1}}{{\mathbf{A}}^{-1}},\end{aligned}$$ $$\begin{aligned}
\label{eq:class_29}
\begin{aligned}
{{\left( \mathbf{A}+\mathbf{B}\right)}^{-1}} & = {{\mathbf{A}}^{-1}}{{({{\mathbf{A}}^{-1}}+{{\mathbf{B}}^{-1}})}^{-1}}{{\mathbf{B}}^{-1}} \\
& = {{\mathbf{B}}^{-1}}{{({{\mathbf{A}}^{-1}}+{{\mathbf{B}}^{-1}})}^{-1}}{{\mathbf{A}}^{-1}}.
\end{aligned}\end{aligned}$$
[**Theorem B.2**]{}. The problem Eq. (\[eq:class\_F-SVM-2\]) is a convex optimization problem.
Note that all the constraints define a convex set, and $\sum\nolimits_{i}{{{\xi }_{i}}}$ and $tr\left( \mathbf{MS} \right)$ are linear to $\mathbf{\xi }$ and ${\bf M}$, respectively. Then the key step is to prove that the function ${{\mathbf{w}}^{T}}{{\mathbf{M}}^{-1}}\mathbf{w}$ is convex for $\mathbf{M}\succ 0$, i.e., for any $1\ge \theta \ge 0$, $$\begin{aligned}
\label{eq:class_B.2.1}
\begin{aligned}
\nonumber & \theta \mathbf{w}_{1}^{T}\mathbf{M}_{1}^{-1}{{\mathbf{w}}_{1}}+(1-\theta )\mathbf{w}_{2}^{T}\mathbf{M}_{2}^{-1}{{\mathbf{w}}_{2}} \ge \\
& {{\left( \theta {{\mathbf{w}}_{1}}+(1-\theta ){{\mathbf{w}}_{2}} \right)}^{T}}{{\left( \theta {{\mathbf{M}}_{1}}+(1-\theta ){{\mathbf{M}}_{2}} \right)}^{-1}}\left( \theta {{\mathbf{w}}_{1}}+(1-\theta ){{\mathbf{w}}_{2}} \right).
\end{aligned}\end{aligned}$$ Note that ${{\left( \theta {{\mathbf{w}}_{1}}\!\!+\!\!(1\!\!-\!\!\theta ){{\mathbf{w}}_{2}} \right)}^{T}}{{\left( \theta {{\mathbf{M}}_{1}}\!\!+\!\!(1\!\!-\!\!\theta ){{\mathbf{M}}_{2}} \right)}^{-1}}\left( \theta {{\mathbf{w}}_{1}}\!\!+\!\!(1\!\!-\!\!\theta ){{\mathbf{w}}_{2}} \right)$ contains three terms: $$\begin{aligned}
\label{eq:class_B.2.2}
\nonumber {{\theta }^{2}}\mathbf{w}_{1}^{T}{{\left( \theta {{\mathbf{M}}_{1}}+(1-\theta ){{\mathbf{M}}_{2}} \right)}^{-1}}{{\mathbf{w}}_{1}} \\
\nonumber {{(1-\theta )}^{2}}\mathbf{w}_{2}^{T}{{\left( \theta {{\mathbf{M}}_{1}}+(1-\theta ){{\mathbf{M}}_{2}} \right)}^{-1}}{{\mathbf{w}}_{2}} \\
\nonumber \theta (1-\theta )\mathbf{w}_{1}^{T}{{\left( \theta {{\mathbf{M}}_{1}}+(1-\theta ){{\mathbf{M}}_{2}} \right)}^{-1}}{{\mathbf{w}}_{2}}.\end{aligned}$$ First, we have $$\begin{aligned}
\label{eq:class_30}
\begin{aligned}
& {{\theta }^{2}}{{\left( \theta {{\mathbf{M}}_{1}}+(1-\theta ){{\mathbf{M}}_{2}} \right)}^{-1}}=\theta {{\left( {{\mathbf{M}}_{1}}+\frac{(1-\theta )}{\theta }{{\mathbf{M}}_{2}} \right)}^{-1}} \\
&=\theta \!\! \left( \!\! \mathbf{M}_{1}^{-1}\!\!-\!\!\mathbf{M}_{1}^{-1}{{\left( \mathbf{M}_{1}^{-1}\!\!+\!\!{{\left( \frac{(1\!\!-\!\!\theta )}{\theta }{{\mathbf{M}}_{2}} \right)}^{-1}} \right)}^{-1}}\mathbf{M}_{1}^{-1} \!\! \right) \\
&=\theta \mathbf{M}_{1}^{-1}-\mathbf{M}_{1}^{-1}{{\left( {{\left( \theta {{\mathbf{M}}_{1}} \right)}^{-1}}+{{\left( (1-\theta ){{\mathbf{M}}_{2}} \right)}^{-1}} \right)}^{-1}}\mathbf{M}_{1}^{-1},
\end{aligned}\end{aligned}$$ and then we get $$\begin{aligned}
\label{eq:class_31}
\begin{aligned}
& \theta \mathbf{w}_{1}^{T}\mathbf{M}_{1}^{-1}{{\mathbf{w}}_{1}}-{{\theta }^{2}}\mathbf{w}_{1}^{T}{{\left( \theta {{\mathbf{M}}_{1}}+(1-\theta ){{\mathbf{M}}_{2}} \right)}^{-1}}{{\mathbf{w}}_{1}} \\
& =\mathbf{w}_{1}^{T}\mathbf{M}_{1}^{-1}{{\left( {{\left( \theta {{\mathbf{M}}_{1}} \right)}^{-1}}+{{\left( (1-\theta ){{\mathbf{M}}_{2}} \right)}^{-1}} \right)}^{-1}}\mathbf{M}_{1}^{-1}{{\mathbf{w}}_{1}}.
\end{aligned}\end{aligned}$$ Analogously, we get $$\begin{aligned}
\label{eq:class_32}
\begin{aligned}
& (1\!-\! \theta )\mathbf{w}_{2}^{T}\mathbf{M}_{2}^{-1}{{\mathbf{w}}_{2}}\!-\!{{(1\!-\!\theta )}^{2}}\mathbf{w}_{2}^{T}{{\left( \theta {{\mathbf{M}}_{1}}\!+\!(1\!-\!\theta ){{\mathbf{M}}_{2}} \right)}^{-1}}{{\mathbf{w}}_{2}} \\
& =\mathbf{w}_{2}^{T}\mathbf{M}_{2}^{-1}{{\left( {{\left( \theta {{\mathbf{M}}_{1}} \right)}^{-1}}+{{\left( (1\!-\!\theta ){{\mathbf{M}}_{2}} \right)}^{-1}} \right)}^{-1}}\mathbf{M}_{2}^{-1}{{\mathbf{w}}_{2}}.
\end{aligned}\end{aligned}$$ With Eq. (\[eq:class\_30\]), we have $$\begin{aligned}
\label{eq:class_33}
\begin{aligned}
& \mathbf{w}_{1}^{T}{{\left( \theta {{\mathbf{M}}_{1}}+(1-\theta ){{\mathbf{M}}_{2}} \right)}^{-1}}{{\mathbf{w}}_{2}} \\
& =\mathbf{w}_{1}^{T}\mathbf{M}_{1}^{-1}{{\left( {{\left( \theta {{\mathbf{M}}_{1}} \right)}^{-1}}+{{\left( (1-\theta ){{\mathbf{M}}_{2}} \right)}^{-1}} \right)}^{-1}}\mathbf{M}_{2}^{-1}{{\mathbf{w}}_{2}}.
\end{aligned}\end{aligned}$$ Combining Eqns. (29)-(32), we get $$\begin{aligned}
\label{eq:34}
\begin{aligned}
\nonumber & \theta \mathbf{w}_{1}^{T}\mathbf{M}_{1}^{-1}{{\mathbf{w}}_{1}}\!+\!(1\!-\!\theta )\mathbf{w}_{2}^{T}\mathbf{M}_{2}^{-1}{{\mathbf{w}}_{2}} \\
& \!-{{\left( \theta {{\mathbf{w}}_{1}}\!\!+\!\!(1\!-\!\theta ){{\mathbf{w}}_{2}} \right)}^{T}}{{\left( \theta {{\mathbf{M}}_{1}}\!\!+\!\!(1\!-\!\theta ){{\mathbf{M}}_{2}} \right)}^{\!-\!1}}\left( \theta {{\mathbf{w}}_{1}}\!\!+\!\!(1\!-\!\theta ){{\mathbf{w}}_{2}} \right) \\
& \!= \mathbf{w}_{1}^{T}\mathbf{M}_{1}^{\!-\!1}{{\left( {{\left( \theta {{\mathbf{M}}_{1}} \right)}^{\!-\!1}}\!\!+\!\!{{\left( (1\!-\!\theta ){{\mathbf{M}}_{2}} \right)}^{\!-\!1}} \right)}^{\!-1\!}}\mathbf{M}_{1}^{\!-\!1}{{\mathbf{w}}_{1}}\\
& \!+ \mathbf{w}_{2}^{T}\mathbf{M}_{2}^{\!-\!1}{{\left( {{\left( \theta {{\mathbf{M}}_{1}} \right)}^{\!-\!1}}\!\!+\!\!{{\left( (1\!-\!\theta ){{\mathbf{M}}_{2}} \right)}^{\!-\!1}} \right)}^{\!-1\!}}\mathbf{M}_{2}^{\!-\!1}{{\mathbf{w}}_{2}} \\
& \!- \mathbf{w}_{1}^{T}\mathbf{M}_{1}^{\!-\!1}{{\left( {{\left( \theta {{\mathbf{M}}_{1}} \right)}^{\!-\!1}}\!\!+\!\!{{\left( (1\!-\!\theta ){{\mathbf{M}}_{2}} \right)}^{\!-\!1}} \right)}^{\!-1\!}}\mathbf{M}_{2}^{\!-\!1}{{\mathbf{w}}_{2}} \\
& \!= {{\left\| {{\left( {{\left( \theta {{\mathbf{M}}_{1}} \right)}^{\!-\!1}}\!+\!{{\left( (1\!-\!\theta ){{\mathbf{M}}_{2}} \right)}^{\!-\!1}} \right)}^{-\frac{1}{2}}}\left( \mathbf{M}_{1}^{\!-\!1}{{\mathbf{w}}_{1}}\!-\!\mathbf{M}_{2}^{\!-\!1}{{\mathbf{w}}_{2}} \right) \right\|}^{2}} \\
& \ge 0.
\end{aligned}\end{aligned}$$ Thus, the F-SVM model is convex.
[**Theorem 4**]{}. Given a SPD matrix ${\bf S}$ and ${\tau }'>0$, $\mathbf{\hat{B}}$ defined in Eq. (\[eq:class\_17\]) is the optimal solution to the problem: $$\begin{aligned}
\label{eq:class_C.1}
\nonumber \mathbf{\hat{B}}=\arg \min \limits_{\mathbf{B}}\left\{ L(\mathbf{B},{\tau }')={{\left\| \mathbf{B} \right\|}_{*}}+{\tau }'\left( tr\left( {{\mathbf{B}}^{-1}}\mathbf{S} \right) \right) \right\}.\end{aligned}$$
$L(\mathbf{B},{\tau }')$ is strictly convex with respect to ${\bf B}$ \[28\]. Given $g(\mathbf{B})={\tau }'tr\left( {{\mathbf{B}}^{-1}}\mathbf{S} \right)$, we have: $$\begin{aligned}
\label{eq:class_C.2}
\nonumber \frac{\partial g}{\partial \mathbf{B}}=-{\tau }'{{\left( {{\mathbf{B}}^{-1}}\mathbf{S}{{\mathbf{B}}^{-1}} \right)}^{T}}\end{aligned}$$
From [@cai2010singular; @watson1992characterization], the set of sub-gradients of the nuclear norm $\partial {{\left\| \mathbf{B} \right\|}_{*}}$ can be represented as $$\begin{aligned}
\label{eq:class_C.3}
\nonumber \partial {{\left\| \mathbf{B} \right\|}_{*}} \!\!=\!\! \{\mathbf{\bar{U}\bar{\Sigma }}{{\mathbf{\bar{U}}}^{T}}\!\!+\!\!\mathbf{W}|\mathbf{W} \!\! \in \!\! {{\mathbf{R}}^{d \!\times\! d}},{{\mathbf{\bar{U}}}^{T}}\mathbf{W}\!\!=\!\!0,\mathbf{W\bar{U}}\!\!=\!\!0,{{\left\| \mathbf{W} \right\|}_{2}}\!\!\le\!\! 1\},\end{aligned}$$ where $\mathbf{\bar{U}\bar{\Sigma }}{{\mathbf{\bar{U}}}^{T}}$ is the eigenvalue decomposition of ${\bf B}$, each column of $\mathbf{\bar{U}}$ is a eigenvector, $\mathbf{\bar{\Sigma }}$ is a diagonal matrix with $\mathbf{\bar{\Sigma }}=diag({{\sigma }_{1}},{{\sigma }_{2}},\cdots ,{{\sigma }_{d}})$ ($0\le {{\sigma }_{1}}\le {{\sigma }_{2}}\le \cdots \le {{\sigma }_{d}}$).
To prove that $\mathbf{\hat{B}}$ is the optimal solution, we will show that $$\begin{aligned}
\label{eq:class_C.4}
\nonumber 0\in -{\tau }'{{\left( {{{\mathbf{\hat{B}}}}^{-1}}\mathbf{S}{{{\mathbf{\hat{B}}}}^{-1}} \right)}^{T}}+\partial {{\left\| {\mathbf{\hat{B}}} \right\|}_{*}}.\end{aligned}$$
With the matrix $\mathbf{\hat{B}}$ in Eq. (\[eq:class\_B\]), we have $$\begin{aligned}
\label{eq:class_C.4}
\nonumber {\tau }'{{\left( {{{\mathbf{\hat{B}}}}^{-1}}\mathbf{S}{{{\mathbf{\hat{B}}}}^{-1}} \right)}^{T}}=\mathbf{U}{{\mathbf{U}}^{T}}\end{aligned}$$
Let ${\bf W}=0$. We have ${{\mathbf{U}}^{T}}\mathbf{W}=0$, $\mathbf{WU}=0$, and ${{\left\| \mathbf{W} \right\|}_{2}}\le 1$. Thus $\mathbf{U}{{\mathbf{U}}^{T}}\in \partial {{\left\| {\mathbf{\hat{B}}} \right\|}_{*}}$, and $\mathbf{\hat{B}}$ is the optimal minimizer of $L(\mathbf{B},{\tau }')$.
Acknowledgment {#acknowledgment .unnumbered}
==============
The work is partially supported by the NSFC fund of China under contract NO. 61271093, the program of ministry of education for New Century Excellent Talents in University under Grant NCET-12-0150.
[Author A]{} Biography text here.
[Author B]{} Biography text here.
[Author C]{} Biography text here.
[^1]: X. Wu, W. Zuo and Y. Zhu are with School of Computer Science and Technology, Harbin Institute of Technology, Harbin, China.
[^2]: L. Lin is with School of Advanced Computing, Sun Yat-sen University, Guangzhou, China.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Let $f:\ov\C\to\ov\C$ be a rational map on the Riemann sphere , such that for every $f$-critical point $c\in J$ which forward trajectory does not contain any other critical point, $|(f^n)'(f(c))|$ grows exponentially fast (Collet–Eckmann condition), there are no parabolic periodic points, and else such that Julia set is not the whole sphere. Then smooth (Riemann) measure of the Julia set is 0.
For $f$ satisfying additionally Masato Tsujii’s condition that the average distance of $f^n(c)$ from the set of critical points is not too small, we prove that Hausdorff dimension of Julia set is less than 2. This is the case for $f(z)=z^2+c$ with $c$ real, $0\in J$, for a positive line measure set of parameters $c$.
address: |
Institute of Mathematics\
Polish Academy of Sciences\
ul. Śniadeckich 8\
00 950 Warszawa, Poland
author:
- Feliks Przytycki
date: '(version of July 26, 1995)'
title: 'On measure and Hausdorff dimension of Julia sets for holomorphic Collet–Eckmann maps'
---
\#1
§
Ł[[L]{}]{}
=18.88pt =18.88pt =14.21pt
[^1]
Introduction {#introduction .unnumbered}
============
It is well-known that if $f:\ov\C\to\ov\C$ a rational map of the Riemann sphere is hyperbolic, i.e. expanding on its Julia set $J=J(f)$ namely $|(f^n)'|>1$ for an integer $n>0$, then Hausdorff dimension $\HD(J)<2$.
The same holds for a more general class of subexpanding maps, namely such maps that all critical points in $J(f)$ are non-reccurrent, supposed $J(f)\not=\ov\C$, see \[U\].
On the other hand there is an abundance of rational maps with $J\not=\ov\C$ and $\HD(J)=2$, \[Shi\].
Recently Chris Bishop and Peter Jones proved that for every finitely generated not geometrically finite Kleinian groups for the Poincaré limit set $\Lambda$ one has $\HD(\Lambda)=2$. As geometrically finite exhibits some analogy to subexpanding in the Kleinian Groups – Rational Maps dictionary, the question arised, expressed by Ch. Bishop and M. Lyubich at MSRI Berkeley conference in January 1995, isn’t it true for every non-subexpanding rational map with connected Julia set, that $HD(J)=2$ ?
Here we give a negative answer. For a large class of “non-uniformly" hyperbolic so called Collet–Eckmann maps, studied in \[P1\], satisfying an additional Tsujii condition, $ \HD(J)<2$.
[**Notation.**]{} For a rational map $f:\ov\C\to\ov\C$ denote by $\Crit(f)$ the set of all critical points of $f$, i.e.points where $f'=0$. Let $\nu:=\sup\{\hbox{multiplicity of}f^n \hbox{at} c:
c\in\Crit(f)\cap J\}$. Finally denote by $\Crit'(f)$ the set of all critical points of $f$ in $J(f)$ which forward trajectories do not contain other critical points.
We prove in this paper the following results:
[**Theorem A.**]{} Let $f$ be a rational map on the Riemann sphere $f:\ov\C\to\ov\C$, and there exist $\lambda>1, C>0$ such that for every $f$-critical point $c\in \Crit'(f)$ $$|(f^n)'(f(c))|\ge C\lambda^n, \eqno (0.1)$$ there are no parabolic periodic points, and $J(f)\not=\ov\C$. Then $\Vol(J(f)=0$, where Vol denotes Riemann measure on $\ov\C$.
[**Theorem B.**]{} In the conditions of Theorem A assume additionally that $$\lim_{t\to\infty}\limsup_{n\to\infty}{1\over n}\sum_{j=1}^n
\max(0,-\log\bigl(\dist(f^j(c),\Crit(f))\bigr)-t) = 0. \eqno (0.2)$$ Then $\HD(J(f))<2$.
For $f(z)=z^2+c, c\in [-2,0]$ real, it is proved in \[T\] that (0.1) and (0.2) are satisfied for a positive measure set of parameters $c$ for which there is no sink in the interval $[c,c^2+c]$. Tsujii’s condition in \[T\] called there [*weak regularity*]{} is in fact apparently stronger than (0.2). The set of subexpanding maps satisfying (0.1) and weak regularity has measure 0, \[T\]. Thus Theorem B answers Bishop–Lyubich’s question.
[**Remark.**]{} In \[DPU\] it is proved that for every rational map $f\ov\C\to\ov\C, c\in\Crit'$ $$\limsup_{n\to\infty} {1\over
n}\sum_{j=1}^n-\log\dist(f^j(c),\Crit(f))\le C_f$$ where $C_f$ depends only on $f$. Here in the condition (0.2) it is sufficient, for Theorem B to hold, to have a positive constant instead of 0 on the right hand side, unfortunately apparently much smaller than $C_f$.
Crucial in proving Theorems A and B is the following intermediate result:
[**Theorem 0.1 (on the existence of pacim)**]{}, see \[P1\]. Let $f:\ov\C\to\ov\C$ satisfies the assumptions of Theorem A. Let $\mu$ be an $\a$-conformal measure on the Julia set $J=J(f)$ for an arbitrary $\a>0$. Assume there exists $0<\lambda'<\lambda$ such that for every $n\ge 1$ and every $c\in \Crit'(f)$ $$\int{d\mu \over \dist (x,f^n(c))^{(1-1/\nu)\a} }< C^{-1}(\lambda')^{\a
n/\nu}. \eqno (0.3)$$ Then there exists an $f$-invariant probability measure $m$ on $J$ absolutely continuous with respect to $\mu$ (pacim).
Recall that a probability measure $\mu$ on $J$ is called $\a$-conformal if for every Borel $B\subset J$ on which $f$ is injective $\mu(f(B)=\int_B |f'|^\a d\mu$. In particular $|f'|^\a$ is Jacobian for $f$ and $\mu$. The number $\a$ is called the exponent of the conformal measure.
If $\Vol(J)>0$ then the restriction of Vol to $J$, normalized, is 2-conformal and obviously satisfies (0.3). If $\HD(J)=2$ then by \[P1\] we know there exists a 2-conformal measure $\mu$ on $J$ but we do not know whether it is not too singular, namely whether it satisfies (0.3). Under the additional assumption (0.2) we shall prove that it is so for every $\a$-conformal measure.
[**Notation.**]{} Const will denote various positive constants which may change from one formula to another, even in one string of estimates.
More on pacim. Proof of Theorem A.
==================================
[**Proposition 1.1.**]{} In the situation of Theorem 0.1 there exists $K>0$ such that $\mu$-a.e. ${dm \over d\mu} \ge K$.
[**Proof.**]{} In Proof of Theorem 0.1 \[P1\] one obtains $m$ as a weak\* limit of a subsequence of the sequence of measures ${1\over n}\sum_{j=0}^{n-1}f_*^j(\mu)$.
It is sufficient to prove that there exists $K>0$ and $n_0>0$ such that for $\mu$-a.e. $y\in J(f)$ $${df^n_*(\mu) \over d\mu}(y)={\cal L}^n(\1) \ge K. \eqno (1.1)$$ Here $\cal L$ denotes the transfer operator, which can be defined for example by ${\cal L}(\f)(y)=\sum_{f(z)=y}|f'(z)|^{-\a}\f(z)$. $\1$ is the constant function of value 1. We can assume $y\notin\bigcup_{n>0}
f^n(\Crit(f))$ because $$\mu(\bigcup_{n>0} f^n(\Crit(f)))=0 \eqno (1.2)$$ If a critical value for $f^n$ were an atom then a critical point would have $\mu$ measure equal to $\infty$.
It is sufficient to prove (1.1) for $y\in B(x,\d)\cap J(f)$ for an [*a priori*]{} chosen $x$ and an arbitrarily small $\d$ and next to use the fact that there exists $m\ge 0$ such that $f^m(B(x,\d))\supset
J(f)$ (called [*topological exactness*]{}). Indeed $$\L^n(\1)(w)=\sum_{f^m(y)=w}\L^{n-m}(\1)(y)|(f^m)'|^{-\a}\ge\
(\sup|(f^m)'|)^{-\a}\L^{n-m}(y_0)$$ where $y_0\in f^{-m}(\{w\})\cap B(x,\d)$.
Recall the estimate from \[P1\]. For an arbitrary $\g>1$ there exists $C>0$ such that for every $x\in J(f)$ $${\cal L}^n(\1)(x)\le C+C \sum_{c\in\Crit(f)\cap J} \sum_{j=0}^{\infty}
{\g^j\la^{-j\a/\nu}\over \dist (x,f^j(f(c)))^{(1-1/\nu)\a} }. \eqno
(1.3)$$ By the assumptions (0.1) and (0.3) the above function is $\mu$-integrable if $\g$ is small enough.
Pay attention to the assumption (0.3). It concerns only $c\in\Crit'$. Fortunately there is only a finite number of summands in (1.3) for which $f^{j_0}(c)\in\Crit, j_0\ge j$ . Each summand is integrable because up to a constant it is bounded by $\L^j(\1)$.
So $$\sum_{c\in\Crit(f)\cap J}
\sum_{j=s}^{\infty} {\g^j\la^{-j\a/\nu}\over \dist
(x,f^j(f(c)))^{(1-1/\nu)\a} }\to 0 \ \ \mu-{\rm a.e.} \ \hbox{as}
s\to\infty . \eqno (1.4)$$
Fix from now on an arbitrary $x\in J(f)$ for which (1.4) holds, $(dm/d\mu) (x)\ge 1$ and $x\notin\bigcup_{n>0}\f^n(\Crit(f))$ (possible by (1.2) and by $\int(dm/d\mu)d\mu=1$).
We need now to repeat from \[P1\] a part of Proof of Theorem 0.1:
For every $y\in B(x,\d)$ and $n>0$ $$\eqalign{ \L^n(\1)(y)&=\sum_{y'\in f^{-n}(y), {\rm
regular}}|(f^n)'(y')|^{-\a} + \sum_{(y',s) {\rm
singular}}\L^{n-s}(\1)(y')|(f^s)'(y')|^{-\a}\cr &=\sum_{{\rm
reg},y}+\sum_{{\rm sing},y} .} \eqno (1.5)$$
We shall recall the definitions of [*regular*]{} and [*singular*]{}: Take an arbitrary subexponentially decreasing sequence of positive numbers $b_j, j=1,2\dots$ with $\sum b_j=1/100$. Denote by $B_{[k}$ the disc $B(x,(\prod_{j=1}^k (1-b_j))2\d)$. We call $s$ the [*essentially critical time*]{} for a sequence of compatible components $W_j=\Comp f^{-j}(B_{[j})$, compatible means $f(W_j)\subset W_{j-1}$, if there exists a critical point $c\in W_s$ such that $f^s(c)\in
B_{[s}$.
We call $y'$ [*regular*]{} in (1.5) if for the sequence of compatible components $W_s, s=0,1,\dots,n, W_n\ni y'$ no $s<n$ is essentially critical.
We call a pair $(y',s)$ [*singular*]{} if $f^s(y')=y$ and for the sequence of compatible components $W_j$, for $j=0,1,\dots,s$, with $W_s\ni y'$, the integer $s$ is the first (i.e., the only) essentially critical time.
If $\d$ is small enough then all $s$ in $\sum_{{\rm sing},x}$ are sufficiently large that $\sum_{{\rm sing},x}\le 1/2$. This follows from the estimates in \[P1\]; here is the idea of the proof: Transforming $\sum_{{\rm sing},x}$ in (1.5) using the induction hypothesis (1.3) we obtain the summands $$C{\g^j\la^{-j\a/\nu}\over \dist (x,f^{s+j-1}(f(c)))^{(1-1/\nu)\a} },\
j=0,\dots,n-s$$ multiplied by $$\Const |(f^{s-1})'(x')|^{-\a/\nu} a_s< \g^{s-1}\la^{-(s-1)\a/\nu}.$$ The numbers $a_s$ are constants arising from distortion estimates, related to $b_s$. The numbers $\g^s$ swallow them and other constants.
(There is a minor inaccuracy here: $(s,x')$ is a singular pair where the summand appears, provided $x'$ is not in the forward trajectory of another critical point, otherwise one moves back to it, see \[P1\] for details.)
Now $\sum_{{\rm sing},x}\le 1/2$ follows from (1.4).
The result is that $\sum_{{\rm reg},x}\ge 1/2$. So by the uniformly bounded distortion along regular branches of $f^{-n}$ on $B(x,\d)$ we obtain $$\L^n(\1)(y)\ge \sum_{{\rm reg},y}\ge\Const \sum_{{\rm
reg},x}\ge\Const>0$$ The name [*regular*]{} concerned formally $y'\in f^{-n}(y)$ but in fact it concerns the branch of $f^{-n}$ mapping $y$ to $y'$ not depending on $y\in B(x,\d)$.
By distortion of any branch $g$ of $f^{-n}$ on a set $U$ we mean $$\sup_{z\in B}|g'(z)|/\inf_{z\in B}|g'(z)|.$$ This proves Proposition 1.1.$\square$
[**Corollary 1.2**]{} In the situation of Theorem 0.1 for measure-theoretic entropy $\h_m(f)>0$.
[**Proof.**]{} Denote $dm/d\mu$ by $u$.
Consider an open set $U\subset \ov\C$ intersecting $J(f)$ such that there exist two branches $g_1$ and $g_2$ of $f^{-1}$ on it. Then by the $f$-invariance of $m$ we have $\Jac_m(g_1)+\Jac_m(g_2) \le 1$ ($=1$ if we considered all branches of $f^{-1}$). $\Jac_m(g_i)$ means Jacobian with respect to $m$ for $g_i$.
We have $m(U)>0$ because $\mu$ does not vanish on open sets in $J$ (by the topological exactness of $f$ on $J$) and by Proposition 1.2. At $m$-a.e. $x\in U$ $$\Jac_m(g_i)(x)=u(g_i(x))|g_i'(x)|u(x)^{-1}>0,$$ (we used here also (1.4)).
Hence $\Jac_m(g_i)<1$, so $\Jac_m(f)>1$ on the set $g_i(U), i=1,2$ of positive measure $m$. Now we use Rokchlin’s formula and obtain $$\h_m(f)=\int\log \Jac_m(f) dm >0 .\eqno\square$$
Let $\chi_m=\int\log|f'|dm$ denote characteristic Lyapunov exponent.
[**Corollary 1.3**]{} In the situation of Theorem 0.1, $\chi_m>0$.
[**Proof.**]{} This Corollary follows from Ruelle’s inequality $\h_m(f)<2\chi_m$, see \[R\].
[**Proof of Theorem A.**]{} Suppose $\Vol(J(f))>0$. After normalization we obtain a 2-conformal measure $\mu$ on $J(f)$ and by Theorem 0.1 and Corollary 1.3 a pacim $m$ with $\chi_m>0$. By Pesin’s Theory \[Pesin\] in the iteration in dimension 1 case \[Le\] (\[Le\] is on the real case, but the complex one is similar), for $m$-a.e. $x$, there exists a sequence of integers $n_j\to\infty$ and $r>0$ such that for every $j$ there exists a univalent branch $g_j$ of $f^{-n_j}$ on $B_j:=B(f^{n_j}(x),r)$ mapping $f^{n_j}(x)$ to $x$ and $g_j$ has distortion bounded by a constant. By $\chi_m>0 \ \diam
g_j(B(f^{n_j}(x),r)\to 0$. (This follows also automatically from the previous assertions by the definition of Julia set \[GPS\].) Now we can forget about the invariant measure $m$ and go back to $\Vol$. Because $J(f)$ is nowhere dense in $\ov\C$, there exists $\e>0$ such that for every $z\in J(f)$ $${\Vol (B(z,r)\setminus J(f)) \over \Vol (B(z,r))}>\e.$$
Bounded distortion for $g_j$ on $B(z,r)$, $z=f^{n_j}(x)$ allows to deduce that the same part of each small disc$\approx g_j(B_j)$ around $x$ is outside $J(f)$ , up to multiplication by a constant. This is so because we can write for every $X\subset B(z,r), y\in B(z,r)$ $$\Vol g_j(X)\approx |g_j'(y)|^2 \Vol(X) \eqno (1.6)$$ where $\approx$ means up to the multiplication by a uniformly bounded factor. So $x$ is not a density point of $J(f)$. On the other hand a.e. point is a density point. So $\Vol J(f)=0$ and we arrived at a contradiction.$\square$
Proof of Theorem B.
===================
As mentioned in the Introduction, the following result is crucial:
[**Lemma 2.1.**]{} Under the conditions of Theorem B, i.e. in the situation of Theorem A and assuming (0.2), the condition (0.3) holds for every $\a$-conformal measure, $\a>0$.
[**Proof. Step 1.**]{} Denote the expression from (0.2) $$\max\bigl(0,-\log\inf_{c\in\Crit'(f)}\dist(f^n(c),\Crit(f))-t\bigr)$$ by $\f_t(n)$. Consider the following union of open-closed intervals $$A_t':=\bigcup_n (n,n+\f_t(n)\cdot K_f] \ \hbox{and write} \
A_t:=\Z_+\setminus A_t',$$ for an arbitrary constant $K_f> \nu/\log\la$.
By (0.2) for every $a>0$ there exist $t>0$ and $n(a,t)$ such that for every $n\ge n(a,t)$ $$A_t\cap [n,n(1+a)]\not=\emptyset \eqno (2.1)$$
Moreover, fixed an arbitrary integer $M>0$, we can guarantee for every $n'\ge n(1+a),n\ge n(a,t)$ $$\sharp (A_t \cap \{j\in [n,n']: j \hbox{divisable by}M\}) \ge {1\over
2M} (n'-n). \eqno (2.2)$$
Observe that for every $n_0,n$, (2.1) transforms into $$[n_0+n,n_0+n+a(n_0+n)]=[n_0+n,n_0+n+a({n_0 \over n}+1)n].$$ The result is that if $n \ge bn_0$ for an arbitrary $b>0$ then $$A_t\cap [n_0+n,n_0+n+a(b^{-1}+1)n]\not=\emptyset. \eqno (2.3)$$ Denote in the sequel $a(b^{-1}+1)$ by $a'$.
[**Step 2.**]{} Observe now that if $n\in A_t$ then for every $c\in\Crit'(f)$ there exist branches $g_s, s=1,2,\dots,n-1$ of $f^{-s}$ on $B_n:=B(f^n(c),\d)$ with uniformly bounded distortions, where $\d=\e\exp-t\nu$ for a constant $\e$ small enough. Sometimes to exhibit the dependence on $n$ we shall write $g_{s,n}$.
Indeed, define $g_s$ on $B_{[s}=B(f^n(c),\prod_{j=1}^s (1-b_j)2\d)$ for $s=1,2,\dots,n-1$ according to the procedure described in Proof of Theorem A. If there is an obstruction, namely $s$ an essential critical time, then for every $z\in B_{[s}$ $$|g_{s-1}'(z)| \le \la^{-s}\th^s \le \exp (-s\nu/K_f) \eqno (2.4)$$ for $\th>1$ arbitrarily close to 1 (in particular such that $K_f>{\nu\over \log\la-\log\th}$) and for $s$ large enough. The constant $\th$ takes care of distortion. (2.4) holds for $z=f^s(q)$, where $q$ is the critical point making $s$ critical time, without $\th$ by (0.1) (with the constant $C$ instead). The small number $\e$ takes care of $s$ small, which cannot be then essential critical.
The inequality (2.4) and rooting ($1/\nu$ to pass from $s-1$ to $s$) imply $\f_t(f^{n-s}(c))\ge s/K_f$, so $n\notin A_t$, a contradiction.
[**Step 3.**]{} By uniformly bounded distortion for the maps $g_{j,n}$, $n\in A_t$ we obtain (compare (1.6)) for every $n_0>0$ large enough, $ c\in\Crit'$ $$\mu B(f^{n_0}(c), r_j)\approx r_j^\a \eqno (2.5)$$ for a sequence $r_j, j=1,2,\dots$ such that $$r_1>\exp -Lbn_0 \eqno (2.6)$$ $$r_{j+1} > r_{j}^{1+ \s} \eqno (2.7)$$ $$\and \ r_{j+1}<r_{j}/2. \eqno (2.8)$$ Here $L:=2\sup |f'|$ and $b, \s$ are arbitrarily close to 0.
Indeed, we can find $r_j$ satisfying the conditions above by taking $$r_j:=\diam g_{n_j, n_0+n_j}(B(f^{n_0+n_j}(c),\d))$$ where $n_j\in A_t$ are taken consecutively so that $$n_{j+1}\in [n_j+(1+\th)n_j, n_j+(1+\th)(1+a')n_j \ \for \ j\ge 2 \
\and$$ $$n_1\in [n_0+bn_0, n_0+bn_0+a'b n_0],$$ where $\th>0$ is an arbitrary constant close enough to 0.
This gives $$r_{j+1}/r_j \ge \exp (-2(\log L)a'n_j) \eqno (2.9)$$ To conclude we need to know that $r_j$ shrink exponentially fast with $n_j\to\infty$. For that we need the following fact (see for example \[GPS\], find the analogous fact in Proof of Theorem A):
[**(\*)**]{} For every $r>0$ small enough and $\xi,
C>0$ there exists $m_0$ such that for every $m\ge m_0, x\in J(f)$ and branch $g$ of $f^{-m}$ on $B(x,r)$ having distortion less than $C$ we have $\diam g(B(x,r))<\xi r$.
Apply now (2.2) to $n=n_0, n'=n_j+n_0$. We obtain a “telescope": For all consecutive $\tau_1,\tau_2,\dots,\tau_{k(j)}\in
A_t\cap [n_0, n_j+n_0]$ divisible by $M$ $$g_{\tau_{i+1}-\tau_i, \tau_{i+1}} (B(f^{\tau_{i+1}}(c),\d))\subset
B(f^{\tau_i}(c),\d/2)$$ for $M\ge m_0$ from (\*).
Hence using (2.2) $$r_j\le 2^{-n_j/2M} . \eqno (2.10)$$ Denote $2a'\log L$ by $\g$ and $(\log 2)/2M$ by $\g'$. (2.9) and (2.10) give $$r_{j+1} \ge r_j\exp -\g n_j \ge r_j (\exp -\g'n_j)^{\g/\g'} \ge
r_j^{1+\g/\g'}.$$ As $\g'$ is a constant and $\g$ can be done arbitrarily small if $a$ is small enough, we obtain (2.7).
The condition (2.8) follows from the fact that for $n_0$ large enough all $n_{j+1}-n_j$ are large enough to apply (\*).
[**Conclusion.**]{} We obtain $$\int{d\mu\over\dist(x,f^{n_0}(c))^{(1-1/\nu)\a}} \le$$ $$\mu(\ov\C \setminus B(f^{n_0}(c), r_1)) {1\over r_1^{(1-1/\nu)\a}} +
\sum_{j\ge 2} \mu(B(f^{n_0}(c), r_{j-1})\setminus B(f^{n_0}(c), r_j))
{1\over r_j^{(1-1/\nu)\a}} \le$$ $$\exp (Lbn_0(1-1/\nu)\a) +\Const\sum_{j\ge 2} {r_{j-1}^\a \over
r_j^{(1-1/\nu)\a}} \le$$ $$(\exp (Lb(1-1/\nu)\a))^{n_0} + \Const\sum_{j\ge 2} r_{j-1}^\a
r_{j-1}^{-(1-1/\nu)\a (1+\s)}.$$
The latter series has summands decreasing exponentially fast for $\s$ small enough so it sums up to a constant, hence the first summand dominates. We obtain the bound by $(\la')^{n_0}$ with $\la'>1$ arbitrarily close to 1. Thus (0.3) has been proved.$\square$
[**Remark 2.2.**]{} The only result in our disposal on the abundance of non-subexpanding maps satisfying (0.1) and (0.2) is Tsujii’s one concerning $z^2+c, c$ real (see Introduction). For this class however the exponential convergence of $$\diam\Comp
f^{-n_j}(B(f^{n_j+n_0}(0),\d)$$ to 0 follows from \[N\] (the component containing $f^{n_0}(c)$). So restricting our interests to this class we could skip (2.2) and considerations leading to (2.10) above.
By \[N\] $\diam\bigl(\Comp( f^{-n}(B(x,\d)))\cap \R\bigr) <
C\tilde\la^{-n}$ for some constants $C>0, \tilde\la>1, \d$ small enough and every component Comp. Just the uniform convergence of the diameters to 0 as $n\to\infty$ follows from \[P1\], but I do not know how fast is it.
[**Proof of Theorem B.**]{} Suppose that $\HD(J)=2$. Then there exist a 2-conformal measure $\mu$ on $J$. This follows from the existence of an $\a$-conformal measure for $\a=\HDess(J)$, where $\HDess$ is the essential Hausdorff dimension which can be defined for example as supremum of Hausdorff dimensions of expanding Cantor sets in $J$, see \[DU\] \[P2\] and \[PUbook\], and from $\HDess(J)=\HD(J)$, see \[P1\]. The former holds for every rational map, the latter was proved in \[P1\] only for Collet–Eckmann maps.
By Lemma 2.1 the condition (0.3) holds, hence there exists a pacim $m\ll\mu$. Moreover $\chi_m>0$ by Corollary 1.3. As in Proof of Theorem A by Pesin Theory there exists $X\subset J$, $m(X)=\mu(X)=1$, such that for every $x\in X$ there exists a sequence of integers $n_j(x)\to\infty$, $r>0$ and univalent branches $g_j$ of $f^{-n_j}$ on $B(f^{n_j}(x),r)$ mapping $f^{n_j}$ to $x$ with uniformly bounded distortion. Write $B_{x,j}:=g_j (B(f^{n_j}(x),r))$.
We obtain for every $x\in X$ applying (1.6) to Vol and $\mu$ (similarly as in Proof of Theorem A) $$\mu(B_{x,j})\le\Const\Vol(B_{x,j})\le\Const(\Vol B(x,\diam B_{x,j})) .$$ If $\Vol X=0$ then there exists a covering of $X$ by discs $B(x_t,\diam B_{x_t,j_t}), t=1,2,\dots$ which union has $\Vol<\e$ for $\e$ arbitrarily close to 0, of multiplicity less than a universal constant (Besicovitch’s theorem). Hence $$\e\ge\Const\sum_t\Vol B(x_t,\diam B_{x_t,j_t})\ge\Const \mu \sum_t
B_{x_t,j_t}\ge 1,$$ a contradiction. Hence $\Vol J\ge \Vol X>0$.
This contradicts Theorem A that $\Vol J=0$ and the proof is over.
Remark that we could end the proof directly: As in Proof of Theorem A we show that no point of $X$ is a point of density of the $\Vol$ measure. Hence $\Vol X=0$. (I owe this remark to M. Urbański.)$\square$
Ch. Bishop, P. Jones: Hausdorff dimension and Kleinian groups. Preprint SUNY at Stony Brook, IMS 1994/5.
M. Denker, F. Przytycki, M. Urbański: On the transfer operator for rational functions on the Riemann sphere. Preprint SFB 170 Göttingen, 4 (1994). To appear in Ergodic Th. and Dyn. Sys..
M. Denker, M. Urbański: On Sullivan’s conformal measures for rational maps of the Riemann sphere. Nonlinearity 4 (1991), 365-384.
P. Grzegorczyk, F. Przytycki, W. Szlenk: On iterations of Misiurewicz’s rational maps on the Riemann sphere. Ann. Inst. H. Poincaré, Phys. Théor. 53 (1990), 431-444.
F. Ledrappier: Some properties of absolutely continuous invariant measures on an interval. Ergod. Th. & Dynam. Sys. 1 (1981), 77-93.
T. Nowicki: A positive Liapunov exponent for the critical value of an $S$-unimodal mapping implies niform hyperbolicity. Ergodic Th. & Dynamic. Sys. 8 (1988), 425-435.
Ya. B. Pesin: Characteristic Lyapunov exponents and smooth ergodic theory. Russ. Math. Surv. 32 (1977), 45-114.
F. Przytycki: Iterations of holomorphic Collet–Eckmann maps: conformal and invariant measures. Preprint n$^o$ 57, Lab. Top. Université de Bourgogne, Février 1995.
F. Przytycki: Lyapunov characteristic exponents are non-negative. Proc. Amer. Math. Soc. 119(1) (1993), 309-317.
F. Przytycki, M. Urbański: To appear.
D. Ruelle: An inequality for the entropy of differentiable maps. Bol. Soc. Bras. Mat. 9 (1978), 83-87.
M. Shishikura: The Hausdorff dimension of the boundary of the Mandelbrot set and Julia set. Preprint SUNY at Stony Brook, IMS 1991/7.
M. Tsujii: Positive Lyapunov exponents in families of one dimensional dynamical systems. Invent. Math. 111 (1993), 113-137.
M. Urbański: Rational functions with no recurrent critical points. Ergodic Th. and Dyn. Sys. 14.2 (1994), 391-414.
[^1]: The author acknowledges support by Polish KBN Grant 2 P301 01307 “Iteracje i Fraktale, II”. He expresses also his gratitude to the MSRI at Berkeley (partial support by NSF grant DMS-9022140) and ICTP at Trieste, where parts of this paper were written.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
In the present work we show that the local generalized monotonicity of a lower semicontinuous set-valued operator on some certain type of dense sets ensures the global generalized monotonicity of that operator. We achieve this goal gradually by showing at first that the lower semicontinuous set-valued functions of one real variable, which are locally generalized monotone on a dense subsets of their domain are globally generalized monotone. Then, these results are extended to the case of set-valued operators on arbitrary Banach spaces. We close this work with a section on the global generalized convexity of a real valued function, which is obtained out of its local counterpart on some dense sets.
*Keywords:*[ locally generalized monotone functions; generalized monotone operators; self segment-dense sets; generalized convex functions]{}
*MSC:* 47H04; 47H05; 26B25; 26E25
address:
- 'S. L'' aszl'' o, Department of Mathematics, Technical University of Cluj Napoca, Str. Memorandumului nr. 28, 400114 Cluj-Napoca, Romania.'
- 'A. Viorel,Department of Mathematics, Technical University of Cluj Napoca, Str. Memorandumului nr. 28, 400114 Cluj-Napoca, Romania.'
author:
- 'Szilárd L'' aszl'' o$^*$ Adrian Viorel$^\dagger$'
title: Generalized monotone operators on dense sets
---
[^1] [^2]
Introduction
============
In this paper we provide sufficient conditions which ensure that a lower semicontinuous set-valued operator satisfying a generalized monotonicity property locally is actually globally generalized monotone. We are concerned on strictly quasimonotone, pseudomonotone as well as strictly pseudomonotone set-valued operators, and these results extend and improve the results obtained in the single-valued case in [@La] and [@KaPiLa].
We study generalized monotonicity properties on certain dense subsets that we call self segment-dense. By an example we show that this new concept differs from that of a segment-dense set introduced by Dinh The Luc [@DTL] in the context of densely quasimonotone, respectively densely pseudomonotone operators.
Two counterexamples circumscribe the area of our research. We show that the local (generalized) monotonicity of an upper semicontinuous set-valued operator does not imply its global counterpart (Example \[ex2\]). Then we also provide an example of lower semicontinuous set-valued operator, which is locally quasimonotone on its domain but is not globally quasimonotone (Example \[ex1\]). Hence, we deal with locally strictly quasimonotone, locally pseudomonotone, respectively locally strictly pseudomonotone lower semicontinuous set-valued operators.
The outline of the paper is the following. In the next section we introduce the necessary apparatus that we need to obtain our results. We also introduce the notion of a self segment-dense set and we show by an example that this notion differs to the notion of segment-dense set introduced in [@DTL]. Further, we emphasize that the local generalized monotonicity on a self segment-dense set contains as particular cases the local monotonicity conditions considered in [@KaPiLa], [@KaPiLa1] and [@La]. Section 3 contains the main results. Our analysis starts with the one dimensional case ($X=\mathbb{R}$) in which we show that a lower semicontinuous set-valued operator that is locally generalized monotone on a dense subset of its domain must be globally generalized monotone. These results are then extended to the case of arbitrary Banach spaces using the concept of a self segment-dense subset. In Section 4 the previous results are applied to obtain the global generalized convexity of locally generalized convex functions under mild assumptions. Also an example of a locally quasiconvex continuously differentiable function which is not globally quasiconvex is given.
Preliminaries
=============
In what follows $X$ denotes a real Banach space and $X^{\ast }$ denotes the topological dual of $X.$ For $x\in X$ and $x^{\ast }\in X^{\ast }$ we denote by $\<x^{\ast },x\>$ the scalar $x^{\ast }(x)$.
We will also often use the following notations for the open, respectively closed, line segments in $X$ with the endpoints $x$ and $y$ $$\begin{aligned}
(x,y) &=&\big\{z\in X:z=x+t(y-x),\,t\in (0,1)\big\}, \\
\lbrack x,y] &=&\big\{z\in X:z=x+t(y-x),\,t\in \lbrack 0,1]\big\}.\end{aligned}$$
For a non-empty set $D\subseteq X$, we denote by $\operatorname*{int}(D)$ its interior and by $\operatorname*{cl}(D)$ its closure. We say that $P\subseteq D$ is dense in $D$ if $D\subseteq \operatorname*{cl}(P)$, and that $P\subseteq X$ is closed regarding $D$ if $\operatorname*{cl}(P)\cap D=P\cap D.$
Let $T:X\rightrightarrows X^{\ast }$ be a set-valued operator. We denote by $D(T)=\{x\in X:T(x)\neq \emptyset \}$ its domain and by $R(T)=\displaystyle\bigcup_{x\in D(T)}{T(x)}$ its range. The graph of the operator $T$ is the set $G(T)=\{(x,x^{\ast })\in X\times X^{\ast }:\,x^{\ast }\in T(x)\}.$
Recall that $T$ is said to be *upper semicontinuous* at $x\in D(T)$ if for every open set $N\subseteq X^{\ast }$ containing $T(x)$, there exists a neighborhood $M\subseteq X$ of $x$ such that $T(M)\subseteq N.$ $T$ is said to be *lower semicontinuous* at $x\in D(T)$ if for every open set $N\subseteq X^{\ast }$ satisfying $T(x)\cap N\neq \emptyset $, there exists a neighborhood $M\subseteq X$ of $x$ such that for every $y\in M\cap D(T)$ one has $T(y)\cap N\neq \emptyset .$ $T$ is upper semicontinuous (lower semicontinuous) on $D(T)$ if it is upper semicontinuous (lower semicontinuous) at every $x\in D(T).$
It can easily be observed that the definition of lower semicontinuity is equivalent to the following (see [@AF] Def. 1.4.2).
\[r21\] $T$ is lower semicontinuous at $x\in D(T)$, if and only if, for every sequence $({x_n})\subseteq D(T)$ such that $x_n\longrightarrow x$ and for every $x^*\in T(x)$ there exists a sequence of elements $x^*_n \in T(x_n)
$ such that $x^*_{n}\longrightarrow x^*.$
Obviously, when $T$ is single-valued then upper semicontinuity and also lower semicontinuity become the usual notion of continuity.
\[d3.1\] Let $T:X\longrightarrow X^*$ be a single valued operator. We say that $T$ is hemicontinuous at $x\in X$, if for all $(t_n)_{n\in{{\mathbb N}}}\subset {{\mathbb R}},\,t_n{\longrightarrow} 0,\, {n\longrightarrow\infty}$ and $y\in X,$ we have $T(x+t_n y)\rightharpoonup^* T(x),\, n\longrightarrow \infty,$ where $"\rightharpoonup^*"$ denotes the convergence with respect to the $weak^*$ topology of $X^*$.
Local generalized monotonicities
--------------------------------
In what follows we recall the definitions of several monotonicity concepts (see, for instance [@HKS; @L-S; @PQ]). The operator $T:X\rightrightarrows X^*$ is called:
1. * monotone* if for all $(x,x^*),(y,y^*)\in G(T)$ one has the following $$\left\langle x^*-y^*,x-y\right\> \geq0;$$
2. * pseudomonotone* if for all $(x,x^*),(y,y^*)\in G(T),$ the following implication holds $$\<x^*,y-x\> \ge 0\Longrightarrow \<y^*,y-x\>\ge 0,$$ or equivalently, for all $(x,x^*),(y,y^*)\in G(T),$ one has $$\<x^*,y-x\> > 0\Longrightarrow \<y^*,y-x\>>0;$$
3. *strictly pseudomonotone* if for all $(x,x^*),(y,y^*)\in G(T),\, x\neq y,$ the following implication holds $$\<x^*,y-x\> \ge 0\Longrightarrow \<y^*,y-x\>> 0;$$
4. *quasimonotone* if for every $(x,x^*),(y,y^*)\in G(T)$ the following implication holds $$\<x^*,y-x\> > 0 \Longrightarrow \<y^*,y-x\>\ge 0;$$
5. *strictly quasimonotone*, if $T$ is quasimonotone, and for all $x,y\in D(T),\,x\neq y$ there exist $z\in(x,y)$ and $z^*\in T(z)$ such that$\<z^*,y-x\> \neq 0.$
The relation among these concepts is shown bellow.
-------------------------------------------------------- -------------------------------
$$ $T \mbox{ is monotone} $
$$ $\Downarrow $
$ T \mbox{ is strictly pseudomonotone}\Longrightarrow$ $T \mbox{ is pseudomonotone}$
$\Downarrow$ $\Downarrow$
$T \mbox{ is strictly quasimonotone}\Longrightarrow$ $T \mbox{ is quasimonotone}.$
-------------------------------------------------------- -------------------------------
The concepts of local monotonicity can be defined as follows.
We say that the operator $T$ is locally monotone, (respectively, locally pseudomonotone, locally strict pseudomonotone, locally quasimonotone, locally strict quasimonotone), on its domain $D(T)$, if every $x\in D(T)$ admits an open neighborhood $U_x\subseteq X$ such that the restriction of the operator $T$ on $U_x\cap D(T),\, T\big|_{U_x\cap D(T)}$ is monotone, (respectively, pseudomonotone, strictly pseudomonotone, quasimonotone, strictly quasimonotone).
According to [@KaPiLa1], respectively [@La], local monotonicity, respectively local generalized monotonicity of a real valued function of one real variable in general implies its global counterpart. More precisely the following hold.
\[p21\] Let $I\subseteq\mathbb{R}$ be an open interval and let $f:I\longrightarrow\mathbb{R}$ be a function.
- [(see Lemma 2.1, [@KaPiLa1])]{} If $f$ is locally increasing on $I$, then $f$ is globally increasing on $I$.
- [(see Theorem 2.2, [@La])]{} If $f$ is locally strictly quasimonotone on $I$, then $f$ is globally strictly quasimonotone on $I$.
- [(see Theorem 2.3, [@La])]{} If $f$ is locally pseudomonotone on $I$, then $f$ is globally pseudomonotone on $I$.
- [(see Theorem 2.4, [@La])]{} If $f$ is locally strictly pseudomonotone on $I$, then $f$ is globally strictly pseudomonotone on $I$.
However, local quasimonotonicity does not imply global quasimonotonicity as the next example shows.
(Example 2.6, [@La]) Let $f:{{\mathbb R}}\longrightarrow{{\mathbb R}},\,$ $$f(x)=\left\{
\begin{array}{lll}
-x-1,\,\mbox{if\,}\, x<-1\\
0,\,\mbox{if\,}\, x\in[-1,1]\\
-x+1,\,\mbox{if\,}\,x>1.\\
\end{array}
\right.$$ We can see that $f$ is continuous and locally quasimonotone on ${{\mathbb R}}$ but $f$ is not globally quasimonotone.
Next we extend the statements $(a)-(d)$ in Proposition \[p21\] to the case of set-valued operators. We need the following notion.
Let $Z$ and $Y$ be two arbitrary sets. A single valued selection of the set-valued map $F:Z\rightrightarrows Y$ is the single valued map $f:Z\longrightarrow Y$ satisfying $f(z)\in F(z)$ for all $z\in Z.$
\[p31\] Let $I\subseteq {\mathbb{R}}$ be an interval and let $F:I\rightrightarrows {\mathbb{R}}$ be a set-valued function. Assume that $F$ is locally strictly quasimonotone, (respectively locally monotone, locally pseudomonotone, locally strictly pseudomonotone) on $I$. Then $F$ is strictly quasimonotone, (respectively monotone, pseudomonotone, strictly pseudomonotone) on $I$.
We treat the case when $F$ is strictly quasimonotone, the other cases can be treated similarly. Observe first that due to the local strict quasimonotonicity property of $F$, for every $x,y\in I,\,x\neq y,$ there exists $z\in (x,y)$ and $z^*\in F(z)$ such that $\<z^*,y-x\>\neq 0.$
Let $x,y\in \operatorname*{int}(I).$ Suppose the contrary, that is, $\<x^*,y-x\>>0$ for some $x^*\in F(x)$ and $\<y^*,y-x\><0,$ for some $y^*\in F(y).$
Let $f$ be a single valued selection of $F$, which is also locally strictly quasimonotone, and assume that $f(x)=x^*$ and $f(y)=y^*.$ According to Proposition \[p31\] $(b)$ we get that $f$ is globally strictly quasimonotone on $I$, hence $\<f(y),y-x\>=\<y^*,y-x\>\ge 0,$ contradiction. Thus, $F$ is globally strictly quasimonotone on $\operatorname*{int}(I).$
If $I$ is closed from left assume that $x$ is the left endpoint of $I$ and $\<x^*,y-x\>> 0$ for some $x^*\in F(x)$ and $y\in I.$ Then obviously $x^*>0.$ Assume further that there exists $y^*\in F(y)$ such that $\<y^*,y-x\><0,$ i.e. $y^*<0.$ Since $F$ is locally strictly quasimonotone on $I$, there exists an open neighborhood $U_x$ of $x$, such that $F|_{U_x\cap I}$ is strictly quasimonotone, which in particular shows that for some $u\in \operatorname*{int}(I)\cap U_x$ there exists $u^*\in
F(u),\, u^*>0.$ On the other hand, it can be deduced in the same manner, that there exists an open neighborhood $U_y$ of $y$, such that $F|_{U_y\cap I}
$ is strictly quasimonotone, which in particular shows that for some $v\in \operatorname*{int}(I)\cap U_y$ there exists $v^*\in F(v),\, v^*<0.$ Obviously we can assume $u<v.$ But then, $\<u^*,v-u\>>0$ and $\<v^*,v-u\><0$ which is in contradiction with the fact that $F$ is strictly quasimonotone on $\operatorname*{int}(I).$
The case when $I$ is closed from right can be treated similarly and we omit it.
\[r31\] Note that for the case when $F$ is locally monotone, locally pseudomonotone, respectively locally strictly pseudomonotone, one may use Proposition \[p31\] $(a),\,(c)$ and $(d)$ respectively.
The next example shows, that locally quasimonotone set-valued operators are not globally quasimonotone in general. Even more, our operator is also lower semicontinuous, hence searching for conditions that ensure global quasimonotonicity of a lower semicontinuous set-valued operator, based on its local quasimonotonicity, is meaningless.
\[ex1\] Consider the operator $$F:\mathbb{R} \rightrightarrows\mathbb{R} ,\, F(x):=\left\{
\begin{array}{lll}
[0,-x-1],\mbox{ if } x<-1, & & \\
\, 0,\mbox{ if } x\in[-1,1], & & \\
\, [-x+1,0],\mbox{ if } x>1.
\end{array}
\right.$$ Then $F$ is lower semicontinuous and locally [quasimonotone]{} on ${\mathbb{R}}$, but is not globally [quasimonotone]{}.
It is easy to check that $F$ is locally [quasimonotone]{} and lower semicontinuous on ${\mathbb{R}}.$ On the other hand, for $x=-2,\,y=2$ and $x^{\ast }=1\in F(x)$ we have $\<x^{\ast },y-x\>=4>0,$ but for $y^{\ast }=-1\in F(y)$ we have $\<y^{\ast },y-x\>=-4<0,$ which shows that $F$ is not [quasimonotone]{}.
Next we define the local generalized monotonicity of an operator on a subset of its domain.
\[d2.1\] Let $T:X\rightrightarrows X^*$ be a set-valued operator and let $D\subseteq D(T).$ We say that $T$ is locally quasimonotone, (respectively, locally monotone, locally pseudomonotone, locally strictly pseudomonotone, locally strictly quasimonotone), on $D$, if every $x\in D$ admits an open neighborhood $U_x$, such that the restriction $T|_{U_x\cap D}$ is [quasimonotone]{}, (respectively, [monotone]{}, [pseudomonotone]{}, [strictly pseudomonotone]{}, [strictly quasimonotone]{}).
\[r2.2\]
- -
Some global generalized monotonicity results for single-valued operators
------------------------------------------------------------------------
Having Definition \[d2.1\] in mind, an important question is related to which properties must the subset $D\subseteq D(T)$ have such that the local generalized monotonicity on $D$ implies the global monotonicity of the same kind. The following results concerning single-valued operators, established in [@KaPiLa1], respectively [@La], will serve as a starting point for our investigations in Section 3.
\[p\]*(Theorem 3.4, [@KaPiLa1])* Let $X$ be a real Banach space and let $D\subseteq X$ be open and convex. Let $C\subseteq D$ be a set closed regarding $D,$ with empty interior, satisfying the following condition $$\forall\,x,y\in D\setminus C \mbox{ the set } [x,y]\cap C \mbox{ is countable, possibly empty.}$$ Assume that $T:D\longrightarrow X^*$ is a single-valued operator. If $T$ is continuous and the restriction $T|_{D\setminus C}$ is locally monotone, then $T$ is monotone on $D$
\[pp\] Let $X$ be a real Banach space and let $D\subseteq X$ be open and convex. Let $C\subseteq D$ be a set closed regarding $D,$ with empty interior, satisfying the following condition $$\forall\,x,y\in D\setminus C \mbox{ the set } [x,y]\cap C \mbox{ is countable, possibly empty.}$$ Assume that $T:D{\longrightarrow}X^*$ is a hemicontinuous operator with the property that $\<T(z),y-x\>\neq 0$ for all $z\in [x,y]\cap C,\,x,y\in D,\, x\neq y.$ Then the following hold.
- *(Theorem 3.6, [@La])* If $T|_{D\setminus C}$ is locally strictly quasimonotone, then $T$ is strictly quasimonotone on $D.$
- *(Theorem 3.7, [@La])* If $T|_{D\setminus C}$ is locally pseudomonotone, then $T$ is pseudomonotone on $D.$
- *(Theorem 3.8, [@La])* If $T|_{D\setminus C}$ is locally strictly pseudomonotone, then $T$ is strictly pseudomonotone on $D.$
Recall that a set $V\subseteq X$ is of first category in the sense of Baire, if $V=\bigcup_{n=1}^{\infty }V_{n}$, where $V_{n}\subseteq X,\,n\in{{\mathbb N}}$ are nowhere dense sets i. e. ${\operatorname*{int}}\left( {\operatorname*{cl}}(V_{n})\right) =\emptyset $.
Note that condition $$\forall\,x,y\in D\setminus C \mbox{ the set } [x,y]\cap C \mbox{ is countable, possibly empty}$$ imposed on the set $C$ in Proposition \[pp\], has been weakened in [@KaPiLa]. More precisely, in [@KaPiLa] it was established the following result.
\[ppp\]*(Theorem 3, [@KaPiLa])* Let $X$ be a real Banach space and let $V\subseteq X$ be a set of first category with the following property $$\forall x,y\in D\setminus V,\mbox{ the set }[x,y]\cap V\mbox{ is countable, possibly empty.}$$ Assume that $T:D{\longrightarrow}X^*$ is hemicontinuous and locally monotone on $D\setminus V$. Then $T$ is monotone on $D$.
Note that the conclusion of Proposition \[ppp\] remains valid if the condition $ \forall x,y\in D\setminus V,\mbox{ the set }[x,y]\cap V\mbox{ is countable, possibly empty}$ is replaced by the condition $$\forall x,y\in D\setminus V,\mbox{ the set }[x,y]\cap V\mbox{ contains no nonempty perfect subsets.}$$
\[r22\] It is worth mentioning that Proposition \[ppp\] cannot be extended to the set-valued case since, according to a theorem of Kenderov (see [@Ke], Proposition 2.6), a monotone operator that is lower semicontinuous at a point of its domain must be single-valued at that point.
Furthermore, Proposition \[ppp\] does not remain valid for set-valued upper semicontinuous operators (see [@KaPiLa], Example 2). Next we also provide an example of an upper semicontinuous set-valued operator which is strictly pseudomonotone on a dense subset of its open and convex domain, but which is not even quasimonotone on its domain.
\[ex2\] Consider the operator $$F:\mathbb{R} \rightrightarrows\mathbb{R} ,\, F(x):=\left\{
\begin{array}{ll}
x+2,\mbox{ if } x\in{\mathbb{R}}\setminus\{0\}, \\
\, [-1,3],\mbox{ if } x=0.
\end{array}
\right.$$
Let $V:=\{0\}.$ Then, $F$ is upper semicontinuous on $D(F)={\mathbb{R}}$ and strictly pseudomonotone on $D(F) \setminus V,$ but is not even quasimonotone on $D(F).$\
Obviously $D(F)\setminus V={\mathbb{R}}\setminus\{0\}$ is dense in $D(F)={\mathbb{R}}.$ Since $F(x)=x+2$ on $D(F)\setminus V,$ it is obvious that there is strictly pseudomonotone. It can easily be verified that $F$ is upper semicontinuous on ${\mathbb{R}}.$ But, for $(x,x^*)=(-1,1)\in G(F)$ and $(y,y^*)=(0,-1)\in G(F),$ we get $\langle x^*,y-x\rangle=1>0$ and $\langle y^*,y-x\rangle=-1<0,$ which shows that $F$ is not quasimonotone on ${\mathbb{R}}.$
According to Example \[ex1\], Example \[ex2\] and Remark \[r22\] we can obtain global generalized monotonicity results for a set-valued operator, based on its local appropriate generalized monotonicity property, only in the case when the operator is lower semicontinuous and the mentioned monotonicity property is one of the following: strict quasimonotonicity, pseudomonotonicity or strict pseudomonotonicity, respectively.
Self segment-dense sets
-----------------------
In [@DTL], Definition 3.4, The Luc has introduced the notion of a so-called *segment-dense* set. Let $V\subseteq X$ be a convex set. One says that the set $U\subseteq V$ is segment-dense in $V$ if for each $x\in V$ there can be found $y\in U$ such that $x$ is a cluster point of the set $[x,y]\cap U.$
In what follows we introduce a denseness notion, slightly different from the concept of The Luc presented above, but which is sufficient for the conditions in Proposition \[p\], Proposition \[pp\], respectively Proposition \[ppp\].
\[dd\] Consider the sets $U\subseteq V\subseteq X$ and assume that $V$ is convex.
We say that $U$ is self segment-dense in $V$ if $U$ is dense in $V$ and $$\forall x,y\in U,\mbox{ the set }[x,y]\cap U\mbox{ is dense in }[x,y].$$
Obviously in one dimension the concepts of a segment-dense set respectively a self segment-dense set are equivalent to the concept of a dense set.
Assume that $C, D,$ respectively $V$ are sets as in Proposition \[p\], Proposition \[pp\], respectively Proposition \[ppp\]. Then the set $D\setminus C$, respectively the set $D\setminus V$ is self segment-dense in $D.$
In what follows we provide some examples of self segment-dense sets.
(see also Example 3.1, [@LYAK])\[EEE\] Let $V$ be the two dimensional Euclidean space ${{\mathbb R}}^2$ and define $U$ to be the set $$U :=\{(p,q) \in{{\mathbb R}}^2 : p\in \mathbb{Q},\, q\in\mathbb{Q}\},$$ where $\mathbb{Q}$ denotes the set of all rational numbers. Then, it is clear that $U$ is dense in ${{\mathbb R}}^2.$ On the other hand $U$ is not segment-dense in ${{\mathbb R}}^2,$ since for $x=(0, \sqrt{2})\in {{\mathbb R}}^2 $ and for every $y=(p,q)\in U$, one has $[x, y] \cap U = \{y\}.$
It can easily be observed that $U$ is also self segment-dense in ${{\mathbb R}}^2$, since for every $x,y\in U$ $x=(p,q),\,y=(r,s)$ we have $[x,y]\cap U=\{(p+t(r-p),q+t(s-q)): t\in[0,1]\cap\mathbb{Q}\},$ which is obviously dense in $[x,y].$
Let $V^{0}\left( \left[ 0,1\right] ,\mathbb{R}\right) $ be the space of piecewise constant $\mathbb{R}$-valued functions.
Obviously $V^{0}\left( \left[ 0,1\right] ,\mathbb{Q}\right) $ is dense in $V^{0}\left( \left[ 0,1\right] ,\mathbb{R}\right) $. So, it follows from the density of $V^{0}\left( \left[ 0,1\right] ,\mathbb{R}\right) $ in $L^{2}\left( \left[ 0,1\right] ,\mathbb{R}\right) $ that $$V^{0}\left( \left[ 0,1\right] ,\mathbb{Q}\right) \text{ is dense in }
L^{2}\left( \left[ 0,1\right] ,\mathbb{R}\right) .$$ Moreover, $V^{0}\left( \left[ 0,1\right] ,\mathbb{Q}\right) $ is self segment-dense in $L^{2}\left( \left[ 0,1\right] ,\mathbb{R}\right) $, and for any $u,v\in V^{0}\left( \left[ 0,1\right] ,\mathbb{Q}\right) $ the segment $\left[ u,v\right] $ is not contained in $V^{0}\left( \left[ 0,1\right] ,\mathbb{Q}\right) $. However, $V^{0}\left( \left[ 0,1\right] ,\mathbb{Q}\right) $ is not segment-dense (in the sense of The Luc) in $L^{2}\left( \left[ 0,1\right] ,\mathbb{R}\right) $, as can be seen by taking $u\left( x\right) =\sqrt{2}$ for all $x\in \left[ 0,1\right] $ and for any $v\in V^{0}\left( \left[ 0,1\right] ,\mathbb{Q}\right) $ we have that $\left[
u,v\right] \cap V^{0}\left( \left[ 0,1\right] ,\mathbb{Q}\right) =\{v\}$.
The proof of the fact that $V^{0}\left( \left[ 0,1\right] ,\mathbb{R}\right)
$ is dense in $L^{2}\left( \left[ 0,1\right] ,\mathbb{R}\right) $ relies on the following classical arguments.
Let $\varepsilon >0$ be given. We have that $C_{c}^{\infty }\left( \left[ 0,1\right] ,\mathbb{R}\right) $ is dense in $L^{2}\left( \left[ 0,1\right] ,\mathbb{R}\right) $ (see for example [@B]), so for a given $u\in
L^{2}\left( \left[ 0,1\right] ,\mathbb{R}\right) $ the exists $u^{\varepsilon }\in C_{c}^{\infty }\left( \left[ 0,1\right] ,\mathbb{R}\right) $ such that$$\left\Vert u-u^{\varepsilon }\right\Vert _{2}<\varepsilon /2.$$Now since $u^{\varepsilon }$ is continuous on the compact set $[0,1]$, uniform continuity guarantees that there exists $\delta >0$ such that for any $x,y\in \left[ 0,1\right] $ with $\left\vert x-y\right\vert <\delta $ we have$$\left\vert u^{\varepsilon }\left( x\right) -u^{\varepsilon }\left( y\right)
\right\vert <\varepsilon /2.$$If we take a regular grid (partition) with step $h$ small enough ($h<\delta $) and choose $u^{h}\in V^{0}\left( \left[ 0,1\right] ,\mathbb{R}\right) $ such that $u^{h}\left( \left( i-\frac{1}{2}\right) h\right) =u^{\varepsilon
}\left( \left( i-\frac{1}{2}\right) h\right) $ for $i:=0,1,\ldots \frac{1}{h}
$ then$$\left\Vert u^{h}-u^{\varepsilon }\right\Vert _{\infty }<\varepsilon /2.$$Hence$$\begin{aligned}
\left\Vert u-u^{h}\right\Vert _{2} &\leq &\left\Vert u-u^{\varepsilon
}\right\Vert _{2}+\left\Vert u^{\varepsilon }-u^{h}\right\Vert _{2} \\
&\leq &\varepsilon /2+\varepsilon /2.\end{aligned}$$
In the next section we show that a lower semicontinuous set-valued operator that possesses a local generalized monotonicity property on a self segment-dense subset of its domain is actually globally generalized monotone. According to the previous remark, this result not only extends but also improves the results presented in Proposition \[pp\].
On the local generalized monotonicity of set-valued maps on dense subsets
=========================================================================
In this section we show that the local strict quasimonotonicity, local pseudomonotonicity, respectively local strict pseudomonotonicity property of a lower semicontinuous set-valued operator on a self segment-dense subset of its domain provides the global appropriate monotonicity of that operator under some mild conditions. We attain this goal gradually, by showing first that the previous statement holds for real set-valued lower semicontinuous functions of one real variable. We also show that our conditions cannot be weakened. As consequences we obtain some results that improve the results stated in [@La].
Locally quasiomonotone and locally strictly quasimonotone operators
-------------------------------------------------------------------
In this paragraph we study the local quasimonotonicity, respectively the local strict quasimonotonicity of set-valued operators on dense sets. As it was expected some additional conditions are needed in order to assure that local generalized monotonicity on a dense subset of a lower semicontinuous set-valued operator implies its global counterpart. Such conditions will be given and by an example we show that our conditions are essential in obtaining these results.
Let $F:I\subseteq{\mathbb{R}}\rightrightarrows {\mathbb{R}}$ be a set-valued function. Let us introduce the following notations. $$F(x)\ge 0\Leftrightarrow x^*\ge 0,\,\forall x^*\in F(x),\mbox{ and } F(x)>0, \mbox{ if } F(x)\ge0\mbox{ and }0\not\in F(x).$$ Obviously the inequalities $F(x)\le 0$, respectively $F(x)<0$ can be introduced analogously.
\[l31\] Let $I\subseteq {\mathbb{R}}$ be an interval and consider the lower semicontinuous set-valued function $F:I\rightrightarrows {\mathbb{R}}.$ Let $J\subseteq I$ dense in $I$ and assume that $F$ is locally quasimonotone on $J.$ Then, for any fixed $x\in I$ one has $F(x)\ge0,$ or $F(x)\le 0.$
Let $x\in I$, and suppose, that there exist $x_1^*,x_2^*\in F(x),\, x_1^*>0,\,x_2^*<0.$ Consider the intervals $V_1=(x_1^*-\epsilon,x_1^*+ \epsilon),$ respectively $V_2=(x_2^*-\epsilon,x_2^*+\epsilon),$ where $\epsilon>0$ such that $x_1^*-\epsilon>0$ and $x_2^*+\epsilon<0.$ By the lower semicontinuity of $F$ there exists an open neighborhood $U_x$ of $x$, such that for every $y\in U_x\cap I$ one has $F(y)\cap V_1\neq\emptyset$ and $F(y)\cap V_2\neq\emptyset.$ Let $y\in U_x\cap J.$ Then, according to the hypothesis of the lemma, there exists an open neighborhood $U_y$ of $y$ such that $F|_{U_y\cap J}$ is quasimonotone.
Let $U=U_x\cap U_y\cap J$ and let $u,v\in U,\, u\neq v.$ Then, there exist $u_1^*,u_2^*\in F(u),\, v_1^*,v_2^*\in F(v)$ such that $u_1^*>0,v_1^*>0,\mbox{ and }u_2^*<0,v_2^*<0.$
Hence, $$\<u_1^*,v-u\>>0,\mbox{ or } \<u_2^*,v-u\>>0,$$ on the other hand $$\<v_1^*,v-u\><0,\mbox{ or } \<v_2^*,v-u\><0,$$ which contradicts the fact that $F$ is quasimonotone in $U.$
Consequently, for any $x\in I$ one has $F(x)\ge 0,$ or $F(x)\le 0.$
Further, we need the following result which proof is straightforward.
\[p32\] Let $F:I\subseteq{\mathbb{R}}\rightrightarrows {\mathbb{R}}$ a set-valued function.
- If $F(x)\ge 0$ for every $x\in I$, (respectively $F(x)\le0$ for every $x\in I$) , then $F$ is quasimonotone on $I$.
- If $F(x)=\{0\}$ for some $x\in I$, and $F(z)\le 0$ for every $z<x,\,z\in I,$ respectively $F(y)\ge 0$ for every $x< y,\,y\in I,$ then $F$ is quasimonotone.
Now we are able to state and prove the following result concerning on the strict quasimonotonicity of a set-valued function.
\[t31\] Let $I\subseteq {\mathbb{R}}$ be an interval and consider the lower semicontinuous set-valued function $F:I\rightrightarrows {\mathbb{R}}.$ Let $J\subseteq I$ dense in $I$ and assume that $F$ is locally [strictly quasimonotone]{} on $J.$ Further, assume that $0\not\in F(x),$ for all $x\in I\setminus J.$ Then $F$ is [strictly quasimonotone]{} on $I.$
According to Lemma \[l31\], for any $x\in I$ one has $F(x)\ge 0,$ or $F(x)\le 0.$ Note that according to the hypothesis of the theorem, for any $x\in I\setminus J$ one has $F(x)>0,$ or $F(x)<0.$
We show next that $F$ is locally strictly quasimonotone on $I$, that is, every $x\in I$ admits an open neighborhood $U$, such that $F|_{U\cap I}$ is strictly quasimonotone. Let $x\in I$ and assume that $F(x)\neq\{0\}.$ Then we have $F(x)\ge 0,$ or $F(x)\le 0.$ Assume that $F(x)\ge 0$ the other case can be treated analogously. Obviously there exists $x^*\in F(x), x^*>0.$ Let $V=(x^*-\epsilon,x_1^*+\epsilon),$ an open interval, where $\epsilon>0$ such that $x^*-\epsilon>0.$ Since $F$ is lower semicontinuous, there exists an open neighborhood $U_x$ of $x$, such that for every $y\in U_x\cap I$ we have $F(y)\cap V\neq\emptyset.$ But then, according to Lemma \[l31\] $F(y)\ge 0,$ for all $y\in U_x\cap I,$ hence from Proposition \[p32\] we get that $F$ is quasimonotone on $U_x\cap I.$ Let $u,v\in U_x\cap I,\, u\neq v.$ Then, by the denseness of $J$ in $I$, we obtain that there exists $w\in(u,v)\cap J,$ and according to the hypothesis of the theorem $w$ admits an open neighborhood $U_w$ such that $F|_{U_w\cap J}$ is strictly quasimonotone. This in particular shows, that there exists $z\in(u,v)$ and $z^*\in F(z)$, such that $\<z^*,v-u\>\neq 0.$ Hence, $F$ is strictly quasimonotone on $U_x\cap I.$
If $F(x)=\{0\}$, then $x\in J$, hence there exists an open neighborhood $U_x$ of $x$ such that $F|_{U_x\cap J}$ is strictly quasimonotone. Note that according to Proposition \[p32\], in case that there exists an open neighborhood $U$ of $x$ such that one of the following conditions is fulfilled
- $F(y)\ge0$ for every $y\in U\cap I,$ or,
- $F(y)\le 0$ for every $y\in U\cap I,$ or,
- $F(z)\le 0$ and $F(y)\ge 0$ or every $z\le x\le y,\,y,z\in U\cap
I,$
then $F|_{U\cap I}$ is strictly quasimonotone.
Assume now, that none of the conditions $(a)-(c)$ is fulfilled. Then, for every neighborhood $U$ of $x$, there exists $y,z\in U\cap I,\, z>x,\,y>x$ (or $z<x,\,y<x,$) such that $F(y)\ge 0,\, F(y)\neq\{0\}$ and $F(z)\le 0,\,F(z)\neq\{0\}.$ Let $y,z\in U_x\cap I,\, z>x,\,y>x,$ (the case $y,z\in U\cap I,\, z<x,\,y<x$ can be treated similarly). Obviously, we can assume that $z>y.$ According to the previous part of the proof there exist an open neighborhood $U_y$ of $y,$ respectively $U_z$ of $z$ such that $F(u)\ge 0$ for all $u\in{U_y\cap I,}$ respectively $F(v)\le0$ for all $v\in{U_z\cap I}.$
Let $u\in U_y\cap U_x\cap J,$ respectively $v\in U_z\cap U_x\cap J,$ such that $u<v,$ and with the property that there exist $u^*\in F(u),\, u^*>0$, respectively $v^*\in F(v),\,v^*<0.$ Such $u^*$ and $v^*$ must exist since $F$ is strictly quasimonotone on $U_x\cap J.$ Then, $\<u^*,v-u\>>0$ and $\<v^*,v-u\><0,$ contradiction.
We have shown, that $F$ is locally strictly quasimonotone on $I$. According to Proposition \[p31\], $F$ is globally strictly quasimonotone on $I$ and the proof is completely done.
In what follows we extend Theorem \[t31\] for lower semicontinuous operators that are locally strictly quasimonotone on a self segment-dense subset of their convex domain.
\[t32\] Let $X$ be a real Banach space and let $X^*$ be its topological dual. Let $T:X\rightrightarrows X^*$ be a lower semicontinuous operator with convex domain $D(T).$ Let $D\subseteq D(T)$ self segment-dense in $D(T)$ and assume that $T$ is locally strictly quasimonotone on $D.$ Assume further, that the following condition holds: $$\label{e1}
\forall x,y\in D(T),\, z\in[x,y]\cap(D(T)\setminus D)\mbox{ and }z^*\in T(z) \mbox{ one has }\<z^*,y-x\>\neq0.$$ Then $T$ is [strictly quasimonotone]{}.
First of all observe, that for every $x,y\in D(T),\,x\neq y$ there exists $z\in(x,y)$ and $z^*\in T(z)$ such that $\<z^*,y-x\>\neq0.$ Indeed, if $[x,y]\cap D$ is dense in $[x,y]$ then the statement follows from the local strict quasimonotonicity property of $T$ on $D.$ Otherwise, there exists $z\in(x,y),\,z\not\in D$. But then the statement follows from condition (\[e1\]). Hence, it is enough to show, that for all $(x,x^*),(y,y^*)\in G(T),$ such that $\<x^*,y-x\>>0$ we have $\<y^*,y-x\>\ge0.$
We divide the proof into two cases.
If $x,y\in D$ then consider the real set-valued function $$F_{x,y}:[0,1]\rightrightarrows{\mathbb{R}},\,F_{x,y}(t)=\{\<z^*,y-x\>:z^*\in T(x+t(y-x))\}.$$
Obviously $T$ is lower semicontinuous on $[x,y].$ Since $T$ is locally strict quasimonotone on $D$, obviously $T$ is locally strict quasimonotone on $[x,y]\cap D.$ On the other hand $D$ is self segment-dense in $D(T)$, hence $[x,y]\cap D$ is dense in $[x,y].$ Let $$J=\{t\in[0,1]:x+t(y-x)\in [x,y]\cap D\}.$$ Obviously $J$ is dense in $[0,1]$ and $F_{x,y}$ is locally strict quasimonotone on $J$. On the other hand, from \[e1\] we obtain, that $0\not\in F_{x,y}(s)$ for every $s\in[0,1]\setminus J.$ Hence, according to Theorem \[t31\], $F_{x,y} $ is strictly quasimonotone on $[0,1].$ In particular, the latter shows, that for $t^*\in
F_{x,y}(0)$ one has $t^*>0$ then $s^*\ge0$ for all $s^*\in F_{x,y}(1)$, or equivalently if $\<x^*,y-x\>>0$ for some $x^*\in T(x)$ then $\<y^*,y-x\>\ge0$ for all $y^*\in T(y).$
Let now $x,y\in D(T)$ arbitrary and assume that $\<x^*,y-x\>>0.$ We show that $\<y^*,y-x\>\ge0,$ for all $y^*\in T(y).$ Since $T$ is lower semicontinuous, according to Remark \[r21\], for every sequence $({x_n})\subseteq D(T)$ such that $x_n\longrightarrow x$ there exists a sequence of elements $x^*_n \in T(x_n)$ such that $x^*_{n}\longrightarrow x^*,$ respectively for every sequence $({y_n})\subseteq D(T)$ such that $y_n\longrightarrow y$ and for every $y^*\in T(y)$ there exists a sequence of elements $y^*_n \in T(y_n)$ such that $y^*_{n}\longrightarrow y^*.$ Let us fix $y^*\in T(y).$ Since $D$ is dense in $D(T)$, one can consider $(x_n),(y_n)\subseteq D.$ Obviously $\<x_n^*,y_n-x_n\>\longrightarrow \<x^*,y-x\>>0,$ hence $\<x_n^*,y_n-x_n\>>0$ if $n$ is big enough. According to the first part of the proof $\<y_n^*,y_n-x_n\>\ge0.$ By taking the limit $n\longrightarrow\infty$ one obtains $\<y^*,y-x\>\ge 0.$
According to the next corollary the previous result improves Theorem 3.6 from [@La].
Let $T:D(T){\longrightarrow}X^*$ be a continuous operator with convex domain $D(T).$ Let $D\subseteq D(T)$ self segment-dense in $D(T)$ and assume that $T$ is locally strictly quasimonotone on $D.$ Assume further, that the following condition holds: $$\forall x,y\in D(T),\, z\in[x,y]\cap(D(T)\setminus D) \mbox{ one has }\<T(z),y-x\>\neq0.$$ Then $T$ is strictly quasimonotone.
Locally pseudomonotone and locally strictly pseudomonotone operators
--------------------------------------------------------------------
In what follows we study the local pseudomonotonicity, respectively the local strict pseudomonotonicity of set-valued operators on self segment-dense sets. We need the following result which proof is straightforward.
\[p33\] Let $F:I\subseteq{\mathbb{R}}\rightrightarrows {\mathbb{R}}$ a set-valued function.
- If $F(x)> 0$, for every $x\in I$, (respectively $F(x)<0$ for every $x\in I$), then $F$ is strictly pseudomonotone on $I$.
- If $F(x)=\{0\}$ for some $x\in I$, and $F(z)< 0$ for every $z<x,\,z\in I,$ respectively $F(y)> 0$ for every $x< y,\,y\in I,$ then $F$ is strictly pseudomonotone on $I$.
- If $I$ is closed from left, $x$ is the left endpoint of $I$ such that $F(x)\ge 0$ and $F(y)>0$ for all $y>x,\, y\in I$, then $F$ is strictly pseudomonotone on $I$.
- If $I$ is closed from right, $x$ is the right endpoint of $I$ such that $F(x)\le 0$ and $F(y)<0$ for all $y<x,\, y\in I$, then $F$ is strictly pseudomonotone on $I$.
The following result ensures the pseudomonotonicity of a locally pseudomonotone set-valued function on a dense subset.
\[t33\] Let $I\subseteq {\mathbb{R}}$ be an interval and consider the lower semicontinuous set-valued function $F:I\rightrightarrows {\mathbb{R}}.$ Let $J\subseteq I$ dense in $I$ and assume that $F$ is locally [pseudomonotone]{} on $J.$ Further, assume that $0\not\in F(x),$ for all $x\in I\setminus J.$ Then $F$ is [pseudomonotone]{} on $I.$
We show that $F$ is locally pseudomonotone on $I$. According to Lemma \[l31\], for any $x\in I$ one has $F(x)\ge 0,$ or $F(x)\le 0.$ Note that according to the hypothesis of the theorem, for any $x\in I\setminus J$ one has $F(x)>0,$ or $F(x)<0.$
Let $x\in I$ and assume that $F(x)\neq\{0\}.$ We have $F(x)\ge 0,$ or $F(x)\le 0.$ Assume that $F(x)\ge 0$ the other case can be treated analogously. Obviously there exists $x^*\in F(x), x^*>0.$ Let $V=(x^*-\epsilon,x_1^*+\epsilon),$ an open interval, where $\epsilon>0$ such that $x^*-\epsilon>0.$ Since $F$ is lower semicontinuous, there exists an open neighborhood $U_x$ of $x$, such that for every $y\in U_x\cap I$ we have $F(y)\cap V\neq\emptyset.$ But then, according to Lemma \[l31\] $F(y)\ge 0,$ for all $y\in U_x\cap I.$ One can observe as well that $F(y)\neq \{0\},$ for all $y\in U_x\cap I.$
We show next, that $F(y)>0,$ for all $y\in U_x\cap I,\,y\neq x.$ Indeed, if $0\in F(y)$ for some $y\in U_x\cap I,$ then according to the hypothesis we have $y\in J.$ Hence, there exists an open neighborhood $U_y\subseteq U_x\cap I$ of $y$ such that $F|_{U_y\cap J}$ is pseudomonotone. But then for $y_1<y,\,y_1\in U_y\cap J$ we have an $y_1^*\in F(y_1),\, y_1^*>0$, hence $\<0,y_1-y\>=0$ and $\<y_1^*,y_1-y\><0,$ contradiction.
Note that we can have $0\in F(x)$ if, and only if, $I$ is closed from left and $x$ is the left endpoint of $I.$ Otherwise, according to the previous part of the proof for $x_1<x,\,x_1\in U_x\cap J$ we have an $x_1^*\in F(x_1),\, x_1^*>0$, hence $\<0,x_1-x\>=0$ and $\<x_1^*,x_1-x\><0,$ contradiction.
Hence, according to Proposition \[p33\], in this case $F$ is strictly pseudomonotone on $U_x\cap I.$
If $F(x)=\{0\}$, then $x\in J$, hence there exists an open neighborhood $U_x$ of $x$ such that $F|_{U_x\cap J}$ is pseudomonotone. We show that $F$ is pseudomonotone on $U_x.$ Suppose the contrary, that is, there exist $u,v\in U_x$ and $u^*\in T(u),\,v^*\in T(v)$ such that $\<u^*,v-u\>\ge 0$ and $\<v^*,v-u\><0.$
Assume that $v>u$, the other case can be treated analogously.
We have $u^*\ge 0$, respectively $v^*<0$, hence according to Lemma \[l31\] $F(v)\le 0.$ Hence, by the lower semicontinuity of $F$ and by the denseness of $J$ in $I$, there exists $w\in(u,v),\,w\in J$ such that $w^*< 0$ for some $w^*\in F(w).$
If $u\in J$ then $\<u^*,w-u\>\ge 0$ and $\<w^*,w-u\><0,$ which contradicts the fact that $F$ is locally pseudomonotone on $U_x\cap J.$
If $u\in I\setminus J$ then from $\<u^*,v-u\>\ge 0$ we have $u^*>0$, hence, by the lower semicontinuity of $F$ and by the denseness of $J$ in $I$, there exists $z\in(u,w),\,z\in J$ such that $z^*> 0$ for some $z^*\in F(z).$ But then $\<z^*,w-z\>\ge 0$ and $\<w^*,w-z\><0,$ contradicts the fact that $F$ is locally pseudomonotone on $U_x\cap J.$
We have shown, that $F$ is locally pseudomonotone on $I$. According to Proposition \[p31\], $F$ is globally pseudomonotone on $I$ and the proof is completely done.
Now we are ready to state and prove one of the main results of this section concerning on pseudomonotonicity of a lower semicontinuous set-valued operator, locally pseudomonotone on a self segment-dense subset of its domain.
\[t34\] Let $X$ be a real Banach space and let $X^*$ be its topological dual. Let $T:X\rightrightarrows X^*$ be a lower semicontinuous operator with convex domain $D(T).$ Let $D\subseteq D(T)$ be self segment-dense in $D(T)$ and assume that $T$ is locally [pseudomonotone]{} on $D.$ Assume further, that the following condition holds: $$\label{e2}
\forall x,y\in D(T),\, z\in[x,y]\cap(D(T)\setminus D)\mbox{ and }z^*\in T(z) \mbox{ one has }\<z^*,y-x\>\neq0.$$ Then $T$ is [pseudomonotone]{}.
For $x,y\in D$ consider the real set-valued function $$F_{x,y}:[0,1]\rightrightarrows{\mathbb{R}},\,F_{x,y}(t)=\{\<z^*,y-x\>:z^*\in
T(x+t(y-x))\}.$$
Obviously $T$ is lower semicontinuous on $[x,y].$ Since $T$ is locally pseudomonotone on $D$, obviously $T$ is locally pseudomonotone on $[x,y]\cap D.$ On the other hand $[x,y]\cap D$ is dense in $[x,y].$ Let $$J=\{t\in[0,1]:x+t(y-x)\in [x,y]\cap D\}.$$ Obviously $J$ is dense in $[0,1]$ and $F_{x,y}$ is locally pseudomonotone on $J$. On the other hand, from (\[e2\]) we obtain, that $0\not\in F_{x,y}(s)$ for every $s\in[0,1]\setminus J.$ Hence, according to Theorem \[t32\], $F_{x,y} $ is pseudomonotone on $[0,1].$ In particular, the latter shows, that for $t^*\in
F_{x,y}(0)$ one has $t^*\ge0$ then $s^*\ge0$ for all $s^*\in F_{x,y}(1)$, or equivalently if $\<x^*,y-x\>\ge0$ for some $x^*\in T(x)$ then $\<y^*,y-x\>\ge0$ for all $y^*\in T(y).$
Let $x\in D(T),\,y\in D(T)\setminus D$ and assume that $\<x^*,y-x\>>0.$ We show that $\<y^*,y-x\>>0,$ for all $y^*\in T(y).$ Since $T$ is lower semicontinuous, according to Remark \[r21\], for every sequence $({x_n})\subseteq D(T)$ such that $x_n\longrightarrow x$ there exists a sequence of elements $x^*_n \in T(x_n)$ such that $x^*_{n}\longrightarrow x^*,$ respectively for every sequence $({y_n})\subseteq D(T)$ such that $y_n\longrightarrow y$ and for every $y^*\in T(y)$ there exists a sequence of elements $y^*_n \in T(y_n)$ such that $y^*_{n}\longrightarrow y^*.$ Let us fix $y^*\in T(y).$ Since $D$ is dense in $D(T)$, one can consider $(x_n),(y_n)\subseteq D.$ Obviously $\<x_n^*,y_n-x_n\>\longrightarrow \<x^*,y-x\>>0,$ hence $\<x_n^*,y_n-x_n\>>0$ if $n$ is big enough. According to the first part of the proof $\<y_n^*,y_n-x_n\>>0.$ By taking the limit $n\longrightarrow\infty$ one obtains $\<y^*,y-x\>\ge 0.$ Since $y\in D(T)\setminus D$ then by the assumption of the theorem $\<y^*,y-x\>\neq 0,$ hence $\<y^*,y-x\>> 0.$
In single-valued case we have the following corollary which improves Theorem 3.7 from [@La].
Let $T:D(T){\longrightarrow}X^*$ be a continuous operator with convex domain $D(T).$ Let $D\subseteq D(T)$ self segment-dense in $D(T)$ and assume that $T$ is locally pseudomonotone on $D.$ Assume further, that the following condition holds: $$\forall x,y\in D(T),\, z\in[x,y]\cap(D(T)\setminus D) \mbox{ one has }\<T(z),y-x\>\neq0.$$ Then $T$ is pseudomonotone.
We have some similar result to those established in Theorem \[t33\] for locally strict pseudomonotone set-valued functions.
\[t35\] Let $I\subseteq {\mathbb{R}}$ be an interval and consider the lower semicontinuous set-valued function $F:I\rightrightarrows {\mathbb{R}}.$ Let $J\subseteq I$ dense in $I$ and assume that $F$ is locally [strictly pseudomonotone]{} on $J.$ Further, assume that $0\not\in F(x),$ for all $x\in I\setminus J.$ Then $F$ is [strictly pseudomonotone]{} on $I.$
We prove that $F$ is locally strictly pseudomonotone on $I.$ Let $x\in I$ and $U_x$ a neighborhood of $x$, such that $F$ is locally strict pseudomonotone on $U_x\cap J.$ Then, according to Theorem \[t32\] $F$ is pseudomonotone on $U_x\cap I.$ Assume that there exist $u,v\in U_x\cap I$ and $u^*\in F(u),\, v^*\in F(v)$ such that $\<u^*,v-u\>\ge 0$ and $\<v^*,v-u\>\le 0.$ Since $F$ is pseudomonotone on $U_x\cap I$, we must have $\<v^*,v-u\>=0.$ Hence $v^*=0$ and due to the hypothesis of the theorem we obtain $v\in J.$ But $\<v^*,u-v\>=0$ and by the pseudomonotonicity of $F$ on $U_x\cap I$ we get $\<u^*,u-v\>\ge 0.$ Hence $u^*=0$ and arguing as before $u\in J.$
But $F$ is strictly pseudomonotone on $U_x\cap J$ hence $$\<u^*,v-u\>= 0\Rightarrow \<v^*,v-u\>> 0,$$ contradiction.
Since $F$ is locally strict pseudomonotone on $I$ according to Proposition \[p31\] $F$ is strict pseudomonotone on $I.$
In infinite dimension we have the following result concerning on strict pseudomonotonicity of a lower semicontinuous set-valued operator, locally strictly pseudomonotone on a self segment-dense subset of its domain..
\[t36\] Let $X$ be a real Banach space and let $X^*$ be its topological dual. Let $T:X\rightrightarrows X^*$ be a lower semicontinuous operator with convex domain $D(T).$ Let $D\subseteq D(T)$ be self segment-dense in $D(T)$ and assume that $T$ is locally [strictly pseudomonotone]{} on $D.$ Assume further, that the following condition holds: $$\label{e3}
\forall x,y\in D(T),\, z\in[x,y]\cap(D(T)\setminus D)\mbox{ and }z^*\in T(z) \mbox{ one has }\<z^*,y-x\>\neq0.$$ Then $T$ is [strictly pseudomonotone]{}.
For $x,y\in D$ the proof is similar to the proof of Theorem \[t34\]. If $x\in D(T),\,y\in D(T)\setminus D$ then by (\[e3\]) one has $\<y^*,y-x\>\neq 0,$ for all $ y^*\in T(y)$, hence the implication $$\<x^*,y-x\>\ge0\Rightarrow \<y^*,y-x\>>0$$ can be obtained as in the proof of Theorem \[t34\].
In single-valued case we have the following corollary which improves Theorem 3.8 from [@La].
Let $T:D(T){\longrightarrow}X^*$ be a continuous operator with convex domain $D(T).$ Let $D\subseteq D(T)$ self segment-dense in $D(T)$ and assume that $T$ is locally strictly pseudomonotone on $D.$ Assume further, that the following condition holds: $$\forall x,y\in D(T),\, z\in[x,y]\cap(D(T)\setminus D) \mbox{ one has }\<T(z),y-x\>\neq0.$$ Then $T$ is strictly pseudomonotone.
The condition $$\forall x,y\in D(T),\, z\in[x,y]\cap(D(T)\setminus D)\mbox{ and }z^*\in T(z) \mbox{ one has }\<z^*,y-x\>\neq0$$ in the hypothesis of Theorem \[t32\], Theorem \[t34\], respectively Theorem \[t36\], in particular the condition $0\not\in F(x),$ for all $x\in I\setminus J$ in the hypothesis of Theorem \[t31\], Theorem \[t33\], respectively Theorem \[t35\], is essential as the next example shows.
Let $$F:\mathbb{R} \rightrightarrows\mathbb{R} ,\, F(x):=\left\{
\begin{array}{lll}
(0,-x],\mbox{ if } x<0, & & \\
\, 0,\mbox{ if } x=0, & & \\
\, [-x,0),\mbox{ if } x>0.
\end{array}
\right.$$ Then $F$ is lower semicontinuous on ${\mathbb{R}}$ and locally [strictly pseudomonotone]{} on ${\mathbb{R}}\setminus\{0\}$, but $F$ is not even [quasimonotone]{} on ${\mathbb{R}}$.
It is easy to check that $F$ is locally [strictly pseudomonotone]{} on $D={\mathbb{R}} \setminus\{0\}$ and lower semicontinuous on $D(F)={\mathbb{R}}.$ Obviously ${
\mathbb{R}}\setminus\{0\}$ is self segment-dense in ${\mathbb{R}},$ hence all the assumptions in the hypothesis of Theorem \[t32\], Theorem \[t34\], respectively Theorem \[t36\] are satisfied excepting the one that $$\forall x,y\in D(F),\, z\in[x,y]\cap(D(F)\setminus D)\mbox{ and }z^*\in F(z) \mbox{ one has }\<z^*,y-x\>\neq0.$$ Consequently, their conclusion fail.
Indeed, for $(x,x^*)=(-1,1)\in G(F)$ and $(y,y^*)=(1,-1)\in G(F)$ we have $\<x^*,y-x\>=2>0,$ and $\<y^*,y-x\>=-2<0,$ which shows that $F$ is not [quasimonotone]{} on ${\mathbb{R}}.$
Generalized convex functions on dense subsets
=============================================
In this section we apply the results obtained in Section 3 to prove the generalized convexity of a locally generalized convex function on a self segment-dense subset of its domain. In the sequel $X$ denotes a real Banach space and $X^*$ denotes its topological dual.
In order to continue our analysis we need the following concepts (see, for instance [@C; @R]). Let $f:X\longrightarrow\overline{{\mathbb{R}}}={\mathbb{R}}\cup\{-\infty,+\infty\}$ be a given function, and let $x\in X$ such that $f(x)\in{\mathbb{R}}.$ Recall that the Clarke-Rockafellar generalized derivative of $f$ at $x$ in direction $v$ is defined by $$f^{\uparrow}(x,v)=\sup_{\epsilon>0}\limsup_{(y,\alpha)\downarrow_f x\, t\downarrow0}\inf_{u\in B(v,\epsilon)}\frac{f(y+tu)-\alpha}{t},$$ where $(y,\alpha)\downarrow_f x$ means $y\longrightarrow x,\,\alpha\longrightarrow f(x),\,\alpha\ge f(y)$ and $B(v,\e)$ denotes the open ball with center $v$ and radius $\e.$
If $f$ is lower semicontinuous at $x$ the Clarke-Rockafellar generalized derivative of $f$ at $x$ in direction $v$ (see [@R]) can be expressed as $$f^{\uparrow}(x,v)=\sup_{\epsilon>0}\limsup_{y\downarrow_f x\, t\downarrow0}\inf_{u\in B(v,\epsilon)}\frac{f(y+tu)-f(y)}{t},$$ where $y\downarrow_f x$ means $y\longrightarrow x,\,f(y)\longrightarrow f(x).$
When $f$ is locally Lipschitz $f^{\uparrow}$ coincides with the Clarke directional derivative (see [@C]), i.e. $$f^{C}(x,v)=\limsup_{\overset{y\to x\, u\to v}{t\downarrow0}}\frac{ f(y+tu)-f(y)}{t}.$$
The Clarke-Rockafellar subdifferential of $f$ at $x$ is given by $$\partial^\uparrow f(x)=\{x^*\in X^*: \<x^*,v\>\le f^\uparrow (x,v),\,\forall v\in X\}.$$
In what follows we present some convexity notions of a real valued function $f:X\longrightarrow{{\mathbb{R}\cup\{\infty\}}}$ (see, for instance [@A; @ACL; @HKS; @P1; @PQ]). Recall that the domain of $f$ is the set $\operatorname*{dom}f=\{x\in X: f(x)\in {{\mathbb R}}\}.$ In the sequel we assume that $\operatorname*{dom}f$ is a convex subset of $X.$ Recall that the function $f$ is called:
1. * convex*, if for all $x,y\in X,\,t\in [0,1]$ one has the following $$f(x+t(y-x))\le f(x)+t(f(y)-f(x))\},$$
2. *pseudoconvex*, if for all $x,y\in \operatorname*{dom}f,$ the following implication holds $$f(y)<f(x)\Rightarrow \forall x^*\in \partial^\uparrow f (x):\<x^*,y-x\>< 0,$$
3. *strictly pseudoconvex*, if for all $x,y\in \operatorname*{dom}f,$ the following implication holds $$f(y)\le f(x)\Rightarrow \forall x^*\in \partial^\uparrow f (x):\<x^*,y-x\><0,$$
4. *quasiconvex*, if for all $x,y\in \operatorname*{dom}f,\,t\in[0,1]$ one has the following $$f(x+t(y-x))\le \max\{f(x),f(y)\},$$
5. *strictly quasiconvex*, if for all $x,y\in \operatorname*{dom}f,\,x\neq y,\,t\in(0,1)$ one has the following $$f(x+t(y-x))< \max\{f(x),f(y)\}.$$
The study of connection between the (generalized) convexity property of a real valued function and appropriate monotonicity of its Clarke-Rockafellar subdifferential ha a rich literature (see for instance [@A; @ACL; @DH; @DH1; @Ko; @P; @P1] and the references therein). In what follows we assume that $f:X\longrightarrow{\mathbb{R}}\cup\{\infty\}$ is locally Lipschitz. The relation among the previously presented convexity concepts is shown bellow.
------------------------------------------------------ -----------------------------
$$ $f \mbox{ is convex} $
$$ $\Downarrow $
$ f \mbox{ is strictly pseudoconvex}\Longrightarrow$ $f \mbox{ is pseudoconvex}$
$\Downarrow$ $\Downarrow$
$f \mbox{ is strictly quasiconvex}\Longrightarrow$ $f \mbox{ is quasiconvex}.$
------------------------------------------------------ -----------------------------
\[r4\] Observe that de definition of pseudoconvexity, respectively strict pseudoconvexity is equivalent to the following: for all $x,y\in \operatorname*{dom}f,$ $$\exists x^*\in \partial^\uparrow f (x):\<x^*,y-x\>\ge0\Rightarrow f(y)\ge f(x),$$ respectively $$\exists x^*\in \partial^\uparrow f (x):\<x^*,y-x\>\ge0\Rightarrow f(y)> f(x),$$
The following statement relates the quasiconvexity of a function to the quasimonotonicity of its Clarke-Rockafellar subdifferential (see [@ACL; @P; @P1]):
\[p41\] A lower semicontinuous function $f:X\longrightarrow{\mathbb{R}}\cup\{\infty\}$ is quasiconvex, if and only if, $\partial^\uparrow f$ is quasimonotone.
The following statement holds (see [@DH]):
\[p42\] A locally Lipschitz function $f:X\longrightarrow{\mathbb{R}}\cup\{\infty\}$ is strictly quasiconvex, if and only if, $\partial^\uparrow f$ is strictly quasimonotone.
The next result is well-known, see for instance [@PQ].
\[p43\] A lower semicontinuous, radially continuous function $f:X\longrightarrow{\mathbb{R}}\cup\{\infty\}$ is pseudoconvex, if and only if, $\partial^\uparrow f$ is pseudomonotone.
A similar result concerning on the strict pseudomonotonicity of the subdifferential of a strictly pseudoconvex function was established in [@PQ].
\[p44\] A locally Lipschitz function $f:X\longrightarrow{\mathbb{R}}\cup\{\infty\}$ is strictly pseudoconvex, if and only if, $\partial^\uparrow f$ is strictly pseudomonotone.
\[d41\] A real valued function $f:X\longrightarrow{{\mathbb{R}\cup\{\infty\}}}$ is said to be locally convex, (respectively, locally quasiconvex, locally strictly quasiconvex, locally pseudoconvex, locally strictly pseudoconvex) if every $x\in X$ possesses an open and convex neighborhood $U_x$ such that the restriction of $f$ on $U_x$, $f|_{U_x}$ is convex, (respectively, quasiconvex, strictly quasiconvex, pseudoconvex, strictly pseudoconvex).
It is known, that local convexity and also local generalized convexity of a differentiable function in general implies its global counterpart (see [@KaPiLa; @La]). However, according to the next example local quasiconvexity does not implies global quasiconvexity.
\[Example 4.9 [@La]\] Consider the function $F:{{\mathbb R}}\longrightarrow{{\mathbb R}},$ $$F(x)={\displaystyle}\left\{\begin{array}{lll}
{\displaystyle}-\frac{x^2}{2}-x,\,\mbox{if\,}\, x<-1,\\
{\displaystyle}\frac12,\,\mbox{if\,}\, x\in[-1,1],\\
{\displaystyle}-\frac{x^2}{2}+x,\,\mbox{if\,}\, x>1.\\
\end{array}\right.$$ Then it can easily be verified that $F$ is continuously differentiable and locally quasiconvex but is not quasiconvex globally.
Assume now, that the set $D\subseteq \operatorname*{dom}f$ is self segment dense in $\operatorname*{dom}f.$ We define the local generalized convexity of $f$ on $D$ as follows.
We say that the function $f$ is:
1. *locally pseudoconvex* on $D$, if every $z\in D$ admits an open neighborhood $U_z$ such that for all $x,y\in U_z\cap D$ the following implication holds $$f(y)<f(x)\Rightarrow \forall x^*\in \partial^\uparrow f (x):\<x^*,y-x\>< 0,$$
2. *locally strictly pseudoconvex* on $D$, if every $z\in D$ admits an open neighborhood $U_z$ such that for all $x,y\in U_z\cap D$ the following implication holds $$f(y)\le f(x)\Rightarrow \forall x^*\in \partial^\uparrow f (x):\<x^*,y-x\>< 0,$$
3. *locally quasiconvex* on $D$, if every $z\in D$ admits an open neighborhood $U_z$ such that for all $x,y\in U_z\cap D$ and $t\in[0,1]$ such that $x+t(y-x)\in D$ one has the following $$f(x+t(y-x))\le \max\{f(x),f(y)\},$$
4. *locally strictly quasiconvex* on $D$, if every $z\in D$ admits an open neighborhood $U_z$ such that for all $x,y\in U_z\cap D$ and $t\in(0,1)$ such that $x+t(y-x)\in D$ one has the following $$f(x+t(y-x))< \max\{f(x),f(y)\}.$$
The following lemma will be useful in the sequel.
\[l41\] Let $f:X\longrightarrow{\mathbb{R}}\cup\{\infty\}$ be lower semicontinuous with convex domain, and let $D\subseteq\operatorname*{dom}f$ self segment-dense in $\operatorname*{dom}f.$ Assume that $f$ is locally strictly quasiconvex on $D$ and that the Clarke-Rockafellar subdifferential of $f,\, \partial^\uparrow f$ is lower semicontinuous on $\operatorname*{dom}f.$ Then $\partial^\uparrow f$ is locally quasimonotone on $D$. If $f$ is also locally Lipschitz, then $\partial^\uparrow f$ is locally strictly quasimonotone on $D$.
Let $z\in D$ and consider $U_z$ an open neighborhood of $z$ such that for all $x,y\in U_z\cap D$ and $t\in(0,1)$ with $x+t(y-x)\in D$ we have $f(x+t(y-x))< \max\{f(x),f(y)\}.$ We show that $\partial^\uparrow f$ is strictly quasimonotone on $U_z\cap D.$
Indeed, let $x,y\in U_z\cap D$ and assume that $\<x^*,y-x\>>0$ for some $x^*\in\partial^\uparrow f(x).$ Then $f^\uparrow (x,y-x)>0$, hence there exists $\e>0$ and the sequences $x_n{\longrightarrow}x,\, t_n\searrow 0$ such that $$\inf_{v\in B(y-x,\e)}\frac{f(x_n+t_n v)-f(x_n)}{t_n}>0.$$ Since $D$ is self segment-dense in $\operatorname*{dom}f$ we can assume that $(x_n),(x_n+t_n(y-x_n))\subseteq D.$ If $n$ is big enough then $\|x_n-x\|<\e$, hence $y-x_n\in B(y-x,\e).$ Thus, $f(x_n+t_n(y-x_n)>f(x_n)$ and in virtue of locally strict quasiconvexity of $f$ on $D$ we get $f(x_n)<f(y).$ Using the lower semicontinuity of $f$ we have $$f(x)\le\liminf_{n{\longrightarrow}\infty}f(x_n),$$ thus $f(x)\le f(y).$
Suppose that there exists $y^*\in\partial^\uparrow f(y)$ such that $\<y^*,y-x\><0.$ Then $\<y^*,x-y\>>0$ which leads to $f^\uparrow (y,x-y)>0$ and using the same arguments as before, we conclude that $f(y)\le f(x).$ Hence, we have $f(x)=f(y).$
By the continuity property of the duality pairing we obtain that there exists an open neighborhood $V$ of $y^*$ such that $\<v^*,x-y\>>0$ for all $v^*\in V.$ Since $\partial^\uparrow f$ is lower semicontinuous, there exists an open neighborhood $U$ of $y$ such that $\partial^\uparrow f(u)\cap V\neq\emptyset$ for all $u\in U\cap U_z.$ Let $u\in (x,y)\cap U\cap D.$ Then there exists $u^*\in \partial^\uparrow f(u)\cap V$ such that $\<u^*,x-y\>>0.$ Since $u=x+t(y-x)$ for some $t\in(0,1)$ we have $\<x^*,u-x\>=t\<x^*,y-x\>>0$ and $\<u^*,x-u\>=t\<u^*,x-y\>>0$, hence by using the same arguments as in firs part of the proof, we conclude that $f(u)= f(x).$ But then $f(u)=\max\{f(x),f(y)\}$ which contradicts the fact that $f$ is locally strictly quasiconvex on $D.$
We proved that $\partial^\uparrow f$ is locally quasimonotne on $D.$ It remained to show, in the case when $f$ is locally Lipschitz, that for all $x,y\in U_z\cap D$ there exists $u\in (x,y)\cap D$ such that $\<u^*,y-x\>\neq 0$ for some $u^*\in \partial^\uparrow f(u).$ Let $x,y\in U_z\cap D.$ Assume that $f(x)\neq f(y)$. This can be assumed since, otherwise, in virtue of locally strict quasiconvexity of $f$ on $D$ one can take $y'\in(x,y)\cap D$ such that $f(y') <\max \{f(x), f(y)\}\Rightarrow f(y')\neq f(x).$ Assume that $f(y)-f(x)>0$, the case $f(y)-f(x)<0$ can be treated similarly.
According to Lebourg mean value theorem (see [@C]), there exists $u\in(x,y)$ and $u^*\in\partial^\uparrow f(u)$ such that $\<u^*,y-x\>= f(y)-f(x)>0.$
In what follows we provide, in a Banach space context, sufficient conditions for strict quasiconvexity of a locally strictly quasiconvex functions.
\[t41\] Let $f:X\longrightarrow{\mathbb{R}}\cup \{\infty\}$ be a locally Lipschitz function, locally strictly quasiconvex on $D,$ where $D\subseteq \operatorname*{dom}f$ is self segment-dense in $\operatorname*{dom}f.$ If $\partial^\uparrow f$ is lower semicontinuous and has the property, that $\<z^*,x-y\>\neq 0$ for all $z\in[x,y]\cap \operatorname*{dom}f\setminus D,\, x,y\in \operatorname*{dom}f,\, x\neq y,\, z^*\in \partial^\uparrow f(z)$ then $f$ is globally quasiconvex on $X.$
According to Lemma \[l41\], $\partial^\uparrow f$ is locally strictly quasimonotone on $D.$ According to Theorem \[t32\], $\partial^\uparrow f$ is strictly quasimonotone on $\operatorname*{dom}f.$ The conclusion follows from Proposition \[p42\].
Similar results to Theorem \[t41\] hold for locally pseudoconvex, respectively locally strict pseudoconvex functions.
\[t42\] Let $f:X\longrightarrow{\mathbb{R}}\cup \{\infty\}$ be a locally Lipschitz function, locally pseudoconvex on $D,$ where $D\subseteq X$ is self segment-dense in $\operatorname*{dom}f.$ If $\partial^\uparrow f$ is lower semicontinuous and has the property, that $\<z^*,x-y\>\neq 0$ for all $z\in[x,y]\cap X\setminus D,\, x,y\in \operatorname*{dom}f,\, x\neq y,\, z^*\in \partial^\uparrow f(z)$ then $f$ is globally pseudoconvex on $X.$
Let $z\in D$ and consider $U_z$ an open neighborhood of $z$ such that for all $x,y\in U_z\cap D$ we have $f(y)< f(x)\Rightarrow \forall x^*\in \partial^\uparrow f (x):\<x^*,y-x\>< 0.$ We show that $\partial^\uparrow f$ is pseudomonotone on $U_z\cap D.$
First of all, observe that $f$ is quasiconvex on $U_z\cap D.$ Indeed, let $x,y\in U_z\cap D$ and assume that $f(u)>\max\{f(x),f(y)\}$ for some $u\in(x,y)\cap D.$ But then $u=x+t(y-x)$ for some $t\in(0,1)$ and by the pseudoconvexity of $f$ on $U_z\cap D$ we obtain that $\forall u^*\in \partial^\uparrow f (u)$ we have $\<u^*,x-u\>< 0$ and $\<u^*,y-u\>< 0$ or, equivalently $\<u^*,-t(y-x)\>< 0$ and $\<u^*,(1-t)(y-x)\>< 0$, impossible.
Assume $\partial^\uparrow f$ is not pseudomonotone on $U_z\cap D$ that is, there exists $x,y\in U_z\cap D$ such that $\<x^*,y-x\>>0$ for some $x^*\in\partial^\uparrow f(x)$ and $\<y^*,y-x\>\le0$ for some $y^*\in\partial^\uparrow f(y).$ According to Remark \[r4\] $\<x^*,y-x\>>0\Rightarrow f(y)\ge f(x),$ and $\<y^*,y-x\>\le0\Rightarrow f(y)\le f(x),$ hence $f(x)=f(y).$
Since $\<x^*,y-x\>>0$ we have $f^\uparrow (x,y-x)>0$, hence there exists $\e>0$ and the sequences $x_n{\longrightarrow}x,\, t_n\searrow 0$ such that $$\inf_{v\in B(y-x,\e)}\frac{f(x_n+t_n v)-f(x_n)}{t_n}>0.$$ Since $D$ is self segment-dense in $\operatorname*{dom}f$ we can assume that $(x_n),(x_n+t_n(y-x_n))\subseteq D.$ If $n$ is big enough then $\|x_n-x\|<\frac{\e}{2}$, hence $y_1-x_n\in B(y-x,\e)$ if $y_1\in B\left(y,\frac{\e}{2}\right).$ Thus, $f(x_n+t_n(y_1-x_n)>f(x_n)$ and in virtue of local quasiconvexity of $f$ on $D$ we get $f(x_n)< f(y_1)$ for all $y_1\in B\left(y,\frac{\e}{2}\right)\cap D.$ The latter relation combined with the pseudoconvexity of $f$ on $U_z\cap D$ in particular shows that $0\not\in \partial^\uparrow f(y).$
Using the continuity of $f$ we have $$f(y)=f(x)=\lim_{n{\longrightarrow}\infty}f(x_n)\le f(y_1),$$ Hence, $f(y)\le f(y_1)$ for all $y_1\in B\left(y,\frac{\e}{2}\right)\cap D$ which shows that $y$ is a local minimum on $U_z\cap D.$ We show that $y$ is a minimum on $B\left(y,\frac{\e}{2}\right)\cap \operatorname*{dom}f.$ Indeed, let $u\in B\left(y,\frac{\e}{2}\right)\cap \operatorname*{dom}f.$ Since $D$ is dense in $\operatorname*{dom}f,$ there exists a sequence $u_n\in B\left(y,\frac{\e}{2}\right)\cap D,\, u_n{\longrightarrow}u.$ Obviously $f(u_n)\ge f(y)$, and since $f$ is continuous we have $f(u)=\lim_{n{\longrightarrow}\infty}f(u_n)\ge f(y).$ Hence $y$ is a local minimum on $U_z\cap \operatorname*{dom}f$, which implies $0\in \partial^\uparrow f(y),$ contradiction.
Thus, $\partial^\uparrow f$ is locally pseudomonotone on $ D.$ According to Theorem \[t33\], $\partial^\uparrow f$ is pseudomonotone on $\operatorname*{dom}f.$ In virtue of Proposition \[p43\], $f$ is pseudoconvex.
Next we establish some conditions that ensure that a locally strictly pseudoconvex function on a self segment-dense subset in its domain is strictly pseudoconvex.
\[t43\] Let $f:X\longrightarrow{\mathbb{R}}\cup \{\infty\}$ be a locally Lipschitz function, locally strictly pseudoconvex on $D,$ where $D\subseteq X$ is self segment-dense in $\operatorname*{dom}f.$ If $\partial^\uparrow f$ is lower semicontinuous and has the property, that $\<z^*,x-y\>\neq 0$ for all $z\in[x,y]\cap X\setminus D,\, x,y\in \operatorname*{dom}f,\, x\neq y,\, z^*\in \partial^\uparrow f(z)$ then $f$ is globally strictly pseudoconvex on $X.$
Let $z\in D$ and consider $U_z$ an open neighborhood of $z$ such that for all $x,y\in U_z\cap D$ we have $f(y)\le f(x)\Rightarrow \forall x^*\in \partial^\uparrow f (x):\<x^*,y-x\>< 0.$ We show that $\partial^\uparrow f$ is strictly pseudomonotone on $U_z\cap D.$
Suppose the contrary, that is there exist $x,y\in U_z\cap D$ such that $\<x^*,y-x\>\ge0$ for some $x^*\in \partial^\uparrow f(x)$ and $\<y^*,y-x\>\le0$ for some $y^*\in \partial^\uparrow f(y).$
According to Remark \[r4\] $$\<x^*,y-x\>\ge0\Rightarrow f(y)>f(x),$$ while $$\<y^*,y-x\>\le0\Rightarrow f(x)>f(y),$$ impossible.
Since $\partial^\uparrow f$ is strictly pseudomonotone on $D,$ according to Theorem \[t34\], $\partial^\uparrow f$ is strictly pseudomonotone on $\operatorname*{dom}f.$ In virtue of Proposition \[p44\], $f$ is strictly pseudoconvex.
The assumptions imposed on $\partial^\uparrow f$ in the hypothesis of the previous theorems cannot be dropped as the next example shows.
Let $f:{{\mathbb R}}{\longrightarrow}{{\mathbb R}},\, f(x)=-|x|.$ Then $f$ is locally strictly pseudoconvex on ${{\mathbb R}}\setminus\{0\}$. We show that $\partial^\uparrow f$ is not lower semicontinuous in $x=0$ and and that $f$ is not even quasiconvex on ${{\mathbb R}}.$
Indeed, it can easily be verified that $$\partial^\uparrow f(x)={\displaystyle}\left\{\begin{array}{lll}
1,\,\mbox{if\,}\, x<0,\\
\, [-1,1],\,\mbox{if\,}\, x=0,\\
-1,\,\mbox{if\,}\, x>0.\\
\end{array}\right.$$ Obviously $\partial^\uparrow f$ is not lower semicontinuous. Let $x=-1,\,y=1.$ Then for all $z\in (x,y)$ one has $f(z)>-1=\max\{f(x),f(y)\}$, which shows that $f$ is not quasiconvex.
[99]{} D. Aussel, *Subdifferential properties of quasiconvex and pseudoconvex functions: Unified approach*, J. Optimiz Theory App. **97**, 29-45 (1998)
D. Aussel, J.N. Corvellec, M. Lassonde, *Subdifferential characterization of quasiconvexity and convexity*, J. Convex Anal. [**1**]{}, 195-201 (1994)
J.P. Aubin, H. Frankowska, Set-valued analysis, Birkhäuser, Boston (1990)
H. Brézis, Functional Analysis, Sobolev Spaces and Partial Differential Equations, Springer (2010)
F.H. Clarke, [Optimization and nonsmoth analysis]{}, Wiley, New York (1983)
A. Daniilidis, N. Hadjisavvas, *Characterization of nonsmooth semistrictly quasiconvex and strictly quasiconvex functions*, J. Optimiz Theory App. **102**, 525-536 (1999)
A. Daniilidis, N. Hadjisavvas,*On the subdifferentials of quasiconvex and pseudoconvex functions and cyclic monotonicity*, J. Math Anal Appl. **237**, 30-42 (1999)
N. Hadjisavvas, *Generalized convexity, generalized monotonicity and nonsmooth analysis*, in Handbook of Generalized Convexity and Generalized Monotonicity, N. Hadjisavas, S. Komlósi, and S. Schaible, eds., Springer Series Nonconvex Optimization and its Applications, Springer, New York (2005)
G. Kassay, C. Pintea, S. László, *Monotone operators and first category sets*, Positivity **16**, 565-577 (2012)
G. Kassay, C. Pintea, S. László, *Monotone operators and closed countable sets*, Optimization **60**, 1059-1069 (2011)
P. Kenderov, *Semi-continuity of set-valued mappings*, Fund. Math. **88**, 61-69 (1975)
S. Komlósi, *Generalized Convexity and Generalized Derivatives*, Handbook of Generalized Convexity and Generalized Monotonicity, N. Hadjisavas, S. Komlósi, and S. Schaible, eds., Springer Series Nonconvex Optimization and its Applications, Springer, New York (2005)
S. László, * Generalized monotone operators, generalized convex functions and closed countable sets*, J. Convex Anal. **18**, 1075-1091 (2011)
L.J. Lin, M.F.Yang, Q.H. Ansari, G. Kassay, *Existence results for Stampacchia and Minty type implicit variational inequalities with multivalued maps*, Nonlinear Anal-Theor. **61**, 1-19 (2005)
D.T. Luc, *Existence Results for Densely Pseudomonotone Variational Inequalities*, J. Math Anal Appl. **254**, 291-308 (2001)
D.T. Luc, S. Schaible, *Generalized Monotone Nonsmooth Maps*, J. Convex Anal. **3**, 195-205 (1996)
J.P. PENOT, *Generalized convex functions in the light of non smooth analysis*, Lecture notes in Economics and Math. Systems **429**, Springer Verlag, 269–291 (1995)
J.P. PENOT, *Are generalized derivatives useful for generalized convex functions?*, in Generalized Convexity, Generalized Monotonicity: Recent results, Crouzeix (Eds.), 3–59 (1998)
J.P. Penot, P. H. Quang, *Generalized convexity of functions and generalized monotonicity of set-valued maps*, J. Optimiz Theory App. **92**, 343-356 (1997)
R.T. Rockafellar, *Generalized directional derivatives and subgradients of nonconvex functions*, Canadian Journal of Mathematics **32**, 257-280 (1980)
[^1]: $^*$This work was supported by a grant of the Romanian Ministry of Education, CNCS - UEFISCDI, project number PN-II-RU-PD-2012-3 -0166.
[^2]: $^\dagger$This research was supported by a grant of the Romanian National Authority for Scientific Research CNCS - UEFISCDI, project number PN-II-PCE-2011-3-0094.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Density-wave instabilities have been observed and studied in a multitude of materials. Most recently, in the context of unconventional superconductors like the iron-based superconductors, they have excited considerable interest. We analyze the fluctuation corrections to the equation of state of the density-wave order parameter for commensurate charge-density waves and spin-density waves due to perfect nesting. For XY magnets, we find that contributions due to longitudinal and transverse fluctuations cancel each other, making the mean-field analysis of the problem controlled. This is consistent with the analysis of fluctuation corrections to the BCS theory of superconductivity \[Š. Kos, A. J. Millis, and A. I. Larkin, Phys. Rev. B **70**, 214531 (2004)\]. However, this cancellation does not occur in density-wave systems when the order parameter is a real $N$-component object with $N\neq2$. Then, the number of transverse fluctuating modes differs from the number of longitudinal fluctuating modes. Indeed, in the case of charge-density waves as well as spin-density waves with Heisenberg symmetry, we find that fluctuation corrections are not negligible, and hence mean-field theories are not justified. These singular fluctuations originate from the intermediate length-scale regime, with wavelengths between the lattice constant and the $T=0$ correlation length. The resulting logarithmic fluctuation contributions to the gap equation originate from the derivative of the anomalous polarization function, and the crucial process is an interaction of quasiparticles through the exchange of fluctuations.'
author:
- Mareike Hoyer
- Jörg Schmalian
title: 'Role of fluctuations for density-wave instabilities: Failure of the mean-field description'
---
Introduction
============
In condensed matter systems, the interaction between electrons often gives rise to an instability of the Fermi-liquid ground state, resulting in the formation of a low-temperature ordered phase. Examples of such ground states include superconductivity, charge-density wave order and magnetically ordered states, and are expected in all materials at least in the clean limit. Often, the low-temperature ordered phase can be described in terms of an effective mean field that characterizes the novel ground state and serves as an order parameter (Fig. \[fig:visu-fluctuations\]). The poster child of such a mean-field theory is the celebrated BCS theory [@BardeenCooperSchrieffer-PR106.1957; @BardeenCooperSchrieffer-PR108.1957] of superconductivity, which offers arguably the single most successful mean-field description of an interacting many-body system. The principal reasons for the success of this theory are the following:
1. the instability towards the BCS state occurs for arbitrarily small interactions,
2. fluctuations beyond mean-field theory can be neglected for small interaction strength since the coherence length is large compared to the inverse Fermi momentum in conventional superconductors, $\xi\gg k_\mathrm{F}^{-1}$, resulting in a narrow Ginzburg regime.
![Visualization of the order parameter and fluctuations around its mean-field value (red dot) for three examples: (a) commensurate charge-density wave order, (b) superconductivity, and (c) commensurate spin-density wave order of Heisenberg spins. The longitudinal mode is indicated in green while the $N-1$ transverse modes are shown in blue. In case of (a) and (b), the manifold of degenerate ground states (red) is shown within the corresponding energy landscape (gray) below the phase transition, whereas for (c) only the former.[]{data-label="fig:visu-fluctuations"}](visu-fluct-OP){width="\columnwidth"}
There are instabilities in the particle-hole channel with perfect nesting [@Peierls1955; @Overhauser-PR1962] which share the first characteristic with the BCS theory [@FeddersMartin-PR1966; @ChanHeine-JPhysF1973; @Gorkov-JETPLett1973; @Gorkov-JETP1974; @RiceScott-PRL1975; @HirschScalapino-PRL1986], and which are governed by the same self-consistent mean-field equation, $$\Delta_0={\rho_{\mathrm{F}}}V\int_{-\omega_0}^{\omega_0}\mathrm{d}\varepsilon_{{\boldsymbol{k}}}\,\frac{[1-2n_\mathrm{F}(E_{{\boldsymbol{k}}})]\Delta_0}{2E_{{\boldsymbol{k}}}}{~ ,}\label{eq:gap-equation}$$ which determines the order parameter $\Delta_0$. Here, we abbreviated $E_{{\boldsymbol{k}}}=\sqrt{\varepsilon_{{\boldsymbol{k}}}^2+\Delta_0^2}$, with the energy dispersion $\varepsilon_{{\boldsymbol{k}}}$. Furthermore, ${\rho_{\mathrm{F}}}V$ is the dimensionless interaction leading to long-range order, $n_\mathrm{F}(E_{{\boldsymbol{k}}})$ is the Fermi distribution, and $\omega_0$ refers to the energetic cut-off of the theory. The physical interpretation of $\Delta_0$ is the pairing gap in the case of superconductivity. In the case of charge-density wave (CDW) or spin-density wave (SDW) ordering, it corresponds to the amplitude of the modulation of the commensurate charge density or spin density, respectively. At zero temperature, the gap equation can be integrated straightforwardly, resulting for ${\rho_{\mathrm{F}}}V\ll1$ in the well-known expression for the zero-temperature energy gap $$\Delta_0=2\omega_0\mathrm{e}^{-\frac{1}{{\rho_{\mathrm{F}}}V}}{~ .}\label{eq:gap}$$
Experimentally, deviations from mean-field behavior have been observed for density-wave systems in the quantum regime [@JaramilloEtAl-Nature2009; @SokolovEtAl-PRB2014; @FreitasEtAl-PRB2015; @FengEtAl-NPhys2015]. The aim of this paper is to theoretically clarify to what extent CDW and SDW systems fulfill the second characteristic of the BCS theory, namely a lack of fluctuation corrections as discovered in the case of superconductivity [@GeorgesYedidia-PRB1991; @vanDongen-PRL1991; @Martin-RoderoFlores-PRB1992; @KosMillisLarkin-PRB2004; @EberleinMetzner-PRB2013; @Eberlein-PRB2014; @FischerEtAl2017]. We find that the result depends on the order-parameter manifold (in particular, the number of components), and on whether longitudinal or transverse fluctuations, i.e., fluctuations of the amplitude or of the phase of the order parameter, dominate. In Fig. \[fig:visu-fluctuations\], the different types of fluctuating modes are visualized for three examples. In all cases, we show that the characteristic length scales of these fluctuations are within the coherence volume $\xi^d$, with coherence length $\xi\sim v_\mathrm{F}/\Delta_0$.
Our analysis focuses on the stability of the ordered state at $T=0$. A complementary view is the investigation of the instability of the disordered state. For the density-wave systems discussed here, the latter can be performed using a renormalization group (RG) treatment [@Shankar-RevModPhys1994]. Consistent with our results, channel interference of the RG analysis suggests corrections to mean-field behavior. This is discussed briefly in Sec. \[sec:rg\] and Appendix \[app:two-band\]. The appeal of our approach is, however, that it offers a simple physical picture for the role of order-parameter fluctuations: amplitude variations suppress the order parameter compared to its mean-field value, whereas phase fluctuations enhance it.
In this paper, we address the role of fluctuations for mean-field theories of commensurate density-wave order in systems with perfect nesting: We provide a self-consistent calculation of Gaussian fluctuation corrections to the zero-temperature gap equation. We come to the conclusion that, in contrast to BCS theory, these mean-field theories are not justified since sizable fluctuation corrections are inherent to these theories. The mean-field approach is then only valid if the number of amplitude and phase fluctuations is the same. Thus the mean-field approach is neither valid for Heisenberg SDWs nor for CDWs. The only exception is spin-density wave order of XY spins, where longitudinal and transverse fluctuation contributions cancel exactly – in accordance with the cancellation of amplitude and phase fluctuations in superconductors [@KosMillisLarkin-PRB2004; @FischerEtAl2017].
Mean-field theory for density-wave instabilities
================================================
Density-wave order arises naturally for systems in which the Fermi surface is perfectly nested, i.e., different parts of the Fermi surface are connected by the vector ${\boldsymbol{Q}}$ such that $\varepsilon_{{\boldsymbol{k}}+{\boldsymbol{Q}}}=-\varepsilon_{{\boldsymbol{k}}}$ holds for the energy dispersion. This nesting condition suggests we formulate mean-field theories for density-wave order in complete structural analogy to the BCS theory of superconductivity. The noninteracting part of the Hamiltonian $\mathcal{H}=\mathcal{H}_0+\mathcal{H}_\mathrm{int}$ is of the usual form $$\label{eq:H0}
\mathcal{H}_0= \sum_{{\boldsymbol{k}},\sigma}\varepsilon_{{\boldsymbol{k}}}\psi^\dagger_{{\boldsymbol{k}},\sigma}\psi^{}_{{\boldsymbol{k}},\sigma}{~ ,}$$ where $\psi_{{\boldsymbol{k}},\sigma}^{(\dagger)}$ annihilates (creates) a fermionic state with momentum ${\boldsymbol{k}}$ and spin $\sigma$, where the nesting condition for the dispersion will be implied in the remainder. The most general $\mathrm{SU}(2)$-invariant interaction can be written as $$\begin{aligned}
\mathcal{H}_\mathrm{int}&= \frac{1}{2}\sum\psi^\dagger_{{\boldsymbol{k}}_1,\sigma_1}\psi^\dagger_{{\boldsymbol{k}}_2,\sigma_2} U_{\sigma_1\sigma_2;\sigma_3\sigma_4}({\boldsymbol{k}}_1,{\boldsymbol{k}}_2;{\boldsymbol{k}}_3,{\boldsymbol{k}}_4) \nonumber \\ &\qquad \times \psi^{}_{{\boldsymbol{k}}_3,\sigma_3}\psi^{}_{{\boldsymbol{k}}_4,\sigma_4}\delta({\boldsymbol{k}}_1+{\boldsymbol{k}}_2-{\boldsymbol{k}}_3-{\boldsymbol{k}}_4){~ ,}\label{eq:H_interaction}\end{aligned}$$ where the summation is over momenta ${\boldsymbol{k}}_i$ and spins $\sigma_i$. The spin sector of the interaction can be decomposed into charge (ch) and spin (sp) channel according to $$U_{\sigma_1\sigma_2;\sigma_3\sigma_4} = U_\mathrm{ch}\,\delta_{\sigma_1\sigma_4}\delta_{\sigma_2\sigma_3} +U_\mathrm{sp}\,{\boldsymbol{\sigma}}_{\sigma_1\sigma_4}\cdot{\boldsymbol{\sigma}}_{\sigma_2\sigma_3}{~ ,}\label{eq:decomposition}$$ where we introduced ${\boldsymbol{\sigma}}=(\sigma_1,\sigma_2,\sigma_3)$ as the vector of Pauli matrices in spin space. Furthermore, we could allow for spin-orbit interaction and consider Ising spins or XY spins instead of the Heisenberg spins introduced in Eq. by restricting ourselves to ${\boldsymbol{\sigma}}=(\sigma_1)$ or ${\boldsymbol{\sigma}}=(\sigma_1,\sigma_2)$, respectively.
When formulating the mean-field theories for charge-density wave (CDW) order and spin-density wave (SDW) order, we consider these two channels separately. The interaction projected onto the respective channels takes the form
\[eq:interaction\] $$\begin{aligned}
\mathcal{H}_\mathrm{ch}&=-\frac{V_\mathrm{ch}}{8}\sum_{\sigma\sigma^\prime}\sum_{{\boldsymbol{k}}{\boldsymbol{k}}^\prime{\boldsymbol{q}}} (\psi^\dagger_{{\boldsymbol{k}},\sigma}\psi^{}_{{\boldsymbol{k}}+{\boldsymbol{q}},\sigma})(\psi^\dagger_{{\boldsymbol{k}}^\prime,\sigma^\prime}\psi^{}_{{\boldsymbol{k}}^\prime-{\boldsymbol{q}},\sigma^\prime}) {~ ,}\\
\mathcal{H}_\mathrm{sp}&=-\frac{V_\mathrm{sp}}{8}\sum_{\sigma_1^{}\sigma_2^{}}\sum_{\sigma_1^\prime\sigma_2^\prime}\sum_{{\boldsymbol{k}}{\boldsymbol{k}}^\prime{\boldsymbol{q}}} (\psi^\dagger_{{\boldsymbol{k}},\sigma_1}{\boldsymbol{\sigma}}_{\sigma_1\sigma_2}\psi^{}_{{\boldsymbol{k}}+{\boldsymbol{q}},\sigma_2}) \nonumber \\
&\qquad \qquad \cdot (\psi^\dagger_{{\boldsymbol{k}}^\prime,\sigma^\prime_1}{\boldsymbol{\sigma}}_{\sigma_1^\prime\sigma_2^\prime}\psi^{}_{{\boldsymbol{k}}^\prime-{\boldsymbol{q}},\sigma^\prime_2}){~ ,}\end{aligned}$$
where we assumed the couplings $U_{\mathrm{ch},\mathrm{sp}}$ to be independent of momenta and thus introduced $U_{\mathrm{ch},\mathrm{sp}}({\boldsymbol{k}}_1,{\boldsymbol{k}}_2;{\boldsymbol{k}}_3,{\boldsymbol{k}}_4)\equiv -V_{\mathrm{ch},\mathrm{sp}}/4$.
The corresponding mean-field theories can be derived from the microscopic Hamiltonian by the usual procedure of introducing an effective bosonic field ${\boldsymbol{\Phi}}$ via a Hubbard-Stratonovich decoupling $$\begin{aligned}
&\mathrm{e}^{\frac{V}{2}\sum_{{\boldsymbol{q}}}{\boldsymbol{b}}_{{\boldsymbol{q}}}\cdot{\boldsymbol{b}}_{-{\boldsymbol{q}}}}= \nonumber \\
&\quad \int\mathcal{D}{\boldsymbol{\Phi}}\,\mathrm{e}^{-\frac{1}{2V}\sum_{{\boldsymbol{q}}}{\boldsymbol{\Phi}}_{{\boldsymbol{q}}}\cdot{\boldsymbol{\Phi}}_{-{\boldsymbol{q}}}+\frac{1}{2}\sum_{{\boldsymbol{q}}}{\boldsymbol{\Phi}}_{{\boldsymbol{q}}}\cdot{\boldsymbol{b}}_{-{\boldsymbol{q}}}+\frac{1}{2}\sum_{{\boldsymbol{q}}}{\boldsymbol{b}}_{{\boldsymbol{q}}}\cdot{\boldsymbol{\Phi}}_{-{\boldsymbol{q}}}}\label{eq:decoupling}\end{aligned}$$ of the interaction in the channel of interest and subsequently integrating out the fermions. This requires $V>0$ for the interaction in the respective channel, where we skip the subindex ch or sp in what follows. The effective field ${\boldsymbol{\Phi}}\in\mathds{R}^N$ plays the role of an order parameter and its dimensionality $N$ depends on the channel in which the decoupling of the interaction is performed. Formulated in the language of field integrals, the corresponding mean-field theory follows immediately at the level of the saddle-point approximation, i.e., assuming that the order parameter be temporally homogeneous and $\delta$-distributed in momentum space, which we denote ${\boldsymbol{\Phi}}_0$.
The mean-field theories for density-wave order resulting from nesting bear the same structure as the BCS theory of superconductivity. The ground-state energy $$E_\mathrm{MF}(\Phi_0)=E_0-2{\rho_{\mathrm{F}}}L^d\Phi_0^2\bigg[\ln\bigg(\frac{2E_\mathrm{F}}{\Phi_0}\bigg)+\frac{1}{2}-\frac{1}{{\rho_{\mathrm{F}}}V}\bigg]$$ deep inside the ordered phase (where $\Phi_0\equiv|{\boldsymbol{\Phi}}_0|$ is the magnitude of the order parameter) is reduced as compared to the high-temperature ground-state energy $E_0$. For the instantaneous electronic interaction considered here, the energy window of the attractive pairing goes up to the Fermi energy. That means, in contrast to the BCS theory where the energetic cut-off $\omega_0$ of the theory is given by the Debye energy, we consider $\omega_0\simeq E_\mathrm{F}$ in the remainder. Furthermore, ${\rho_{\mathrm{F}}}$ refers to the density of states at the Fermi level and $L^d$ denotes the system’s volume. This structure results in the characteristic logarithm appearing in the zero-temperature mean-field gap equation $$0=\frac{1}{{\rho_{\mathrm{F}}}L^d}\frac{{\mathrm{d}}E_\mathrm{MF}(\Phi_0)}{{\mathrm{d}}\Phi_0^2}
\approx \frac{1}{{\rho_{\mathrm{F}}}V}-\ln\bigg(\frac{2E_\mathrm{F}}{\Phi_0}\bigg){~ ,}\label{eq:gap-equation2}$$ which self-consistently determines the magnitude of the mean-field value of the order parameter. Details on the derivation can be found in Appendix \[app:calculation\]. Note that Eq. is equivalent to Eq. at $T=0$. The logarithmic contribution arises from a relative sign provided by the nesting condition $\varepsilon_{{\boldsymbol{k}}+{\boldsymbol{Q}}}=-\varepsilon_{{\boldsymbol{k}}}$ in the case of density-wave order, whereas for superconductivity it results from the pairing of time-reversed states. Hence, the density-wave analog for the Nambu spinors is given by $$\Psi_{{\boldsymbol{k}}}=\begin{pmatrix} \psi^\dagger_{{\boldsymbol{k}},\uparrow} & \psi^\dagger_{{\boldsymbol{k}},\downarrow} & \psi^\dagger_{{\boldsymbol{k}}+{\boldsymbol{Q}},\uparrow} &\psi^\dagger_{{\boldsymbol{k}}+{\boldsymbol{Q}},\downarrow}\end{pmatrix}^T {~ ,}\label{eq:spinor}$$ allowing us to represent the mean-field ground-state energy as $E_\mathrm{MF}(\Phi_0)=2\Phi_0^2/V-\int_k\operatorname{tr}\ln[-(\mathcal{G}_k^\mathrm{MF})^{-1}]$. The corresponding matrix Green’s function in Matsubara representation is given by $$\begin{aligned}
&\big(\mathcal{G}_k^{\mathrm{MF}}\big)^{-1}=\frac{1}{2}\left(\mathrm{i}\nu_n\tau_0\sigma_0-\varepsilon_{{\boldsymbol{k}}}\tau_3\sigma_0\right) \nonumber \\
&\qquad +\frac{1}{2}\tau_1\left\{\begin{matrix}\Phi_0\sigma_0 & \text{for CDW order} {~ ,}\\ {\boldsymbol{\Phi}}_0\cdot {\boldsymbol{\sigma}}& \text{for Heisenberg SDW order}{~ ,}\end{matrix}\right.\end{aligned}$$ where we abbreviated $k\equiv({\boldsymbol{k}},\nu_n)$ with the fermionic Matsubara frequency $\nu_n=(2n+1)\pi T$. The Pauli matrices $\sigma_i$ refer to the spin sector whereas the $\tau_i$ denote Pauli matrices in the band space that emerges due to a doubling of the unit cell. We assume $2{\boldsymbol{Q}}$ to be a reciprocal lattice vector.
Fluctuation corrections to the gap equation
===========================================
The validity of the mean-field approximation as discussed in the previous section relies on the assumption that deviations of the order parameter from its mean-field value lead to negligible corrections to the mean-field theory in the sense that fluctuation corrections to physical observables are small compared to their mean-field value.
Of course, there are regimes in which the role of long-wavelength ($|{\boldsymbol{q}}|\ll2\xi^{-1}$) fluctuations is well-understood. Firstly, thermal fluctuations drive the phase transition. The Ginzburg regime in which these thermal fluctuations lead to sizable corrections to mean-field theory is restricted to the vicinity of the critical point. Secondly, in low-dimensional systems fluctuations are important down to lower temperatures, where they can lead to the breakdown of true long-range order [@MerminWagner-PRL1966; @Hohenberg-PR1967]. At zero temperature, deep inside the ordered phase, (quantum) fluctuations are expected to have severe consequences only for one-dimensional systems, while they are small in spatial dimensions $d\geq2$. Here we consider different fluctuations with characteristic length scales shorter than $\xi$.
To self-consistently check the validity of a mean-field theory, we determine fluctuation corrections to the zero-temperature gap equation and compare the resulting contributions due to fluctuations to the mean-field terms. This approach was put forward in the context of superconductivity by Kos, Millis, and Larkin [@KosMillisLarkin-PRB2004]; and their self-consistent calculation of corrections to the BCS gap equation indeed showed that the BCS mean-field theory is justified. This analysis for $s$-wave superconductors can be extended to charged superconductors [@FischerEtAl2017] as well as anisotropic superconductors [@ParamekantiEtAl-PRB2000; @BarlasVarma-PRB2013; @Hoyer-Thesis2017], resulting in the conclusion that fluctuation corrections to the zero-temperature gap equation are generally negligible for superconductors. Remarkably, this is due to an exact cancellation of individually large contributions that can be assigned to fluctuations of the amplitude and the phase of the order parameter. Therefore, a natural question in the context of density-wave instabilities is whether an analogous mechanism of cancellation of longitudinal and transverse fluctuation corrections ensures the validity of the respective mean-field theories – or whether quantum fluctuations become sizable such that the mean-field description is not justified. This question will be addressed in the remainder of this paper.
For density-wave instabilities, the order parameter governing the effective action is a real $N$-component vector, $\boldsymbol{\Phi}(\boldsymbol{r},\tau)\in\mathds{R}^N$, where $\tau$ refers to the imaginary time in Matsubara formalism. This order parameter can be split into the static and homogeneous mean-field value and spatial and temporal fluctuations around this mean-field value as $$\boldsymbol{\Phi}(\boldsymbol{r},\tau)=\Phi_0\hat{\boldsymbol{e}}+\delta\boldsymbol{\Phi}(\boldsymbol{r},\tau){~ .}$$ Here, we introduced $\hat{\boldsymbol{e}}$ as the unit vector along the direction of the mean-field order parameter and the magnitude of the mean-field order parameter $\Phi_0$ can be determined self-consistently from the gap equation. The fluctuations $\delta\boldsymbol{\Phi}$ of the order parameter around the mean-field configuration can be further split into one longitudinal mode $\parallel \hat{\boldsymbol{e}}$ and $N-1$ transverse modes $\perp \hat{\boldsymbol{e}}$. In the remainder of this paper, we restrict ourselves to the leading contribution and evaluate the fluctuation corrections for Gaussian fluctuations, i.e., taking into account contributions up to $\mathcal{O}[(\delta{\boldsymbol{\Phi}})^2]$.
![Diagrammatic representation of contributions to the gap equation. Straight lines represent the fermionic propagators while wiggly lines stand for the fluctuation propagator. The order parameter (indicated by a dashed line) is added for the sake of clarity here, however, it does not contribute to the derivative ${\mathrm{d}}E/{\mathrm{d}}\Phi_0^2$. The logarithmic contribution to the mean-field gap equation is shown in (a), while the structure of Gaussian fluctuation corrections to the gap equation is presented in (b).[]{data-label="fig:gap-equation"}](gap-equation){width="\columnwidth"}
Including fluctuations around the saddle-point configuration results in additional contributions to the ground-state energy, $E(\Phi_0)=E_\mathrm{MF}(\Phi_0)+E_\mathrm{fluct}(\Phi_0)$, which are also reflected in the gap equation as $$\frac{1}{{\rho_{\mathrm{F}}}L^d}\frac{{\mathrm{d}}E(\Phi_0)}{{\mathrm{d}}\Phi_0^2}=\frac{1}{\lambda}-\ln\Big(\frac{2E_\mathrm{F}}{\Phi_0}\Big)+\frac{1}{{\rho_{\mathrm{F}}}L^d}\frac{{\mathrm{d}}E_\mathrm{fluct}(\Phi_0)}{{\mathrm{d}}\Phi_0^2}=0{~ ,}\label{eq:gap-equation-fluct}$$ where we introduced the dimensionless interaction $\lambda={\rho_{\mathrm{F}}}V$. Unless the additional contribution stemming from fluctuations is smaller than the first two terms already appearing at mean-field level \[cf. Eq. \], the mean-field theory is not justified. The logarithmic contribution already appearing at mean-field level corresponds to the fermionic loop diagram depicted in Fig. \[fig:gap-equation\](a). The Gaussian fluctuation corrections studied here take the form presented in Fig. \[fig:gap-equation\](b). In the remainder, we show that such terms indeed give rise to a logarithmic contribution to the gap equation. The usual contributions due to critical fluctuations, important in the long-wavelength limit and for small frequencies, correspond to the limit where the fermionic triangle becomes structureless, cf. Fig. \[fig:long-wavelength\], but our analysis shows that the internal structure of the triangle diagram is indeed important.
\] becomes structureless. This limit corresponds to the usual contributions known from the consideration of critical fluctuations. Note that in the opposite regime of short-wavelength fluctuations ($|{\boldsymbol{q}}|\gg2\xi^{-1}$), the inner structure of the fermionic triangle becomes crucial, cf. Fig. \[fig:triangles\].[]{data-label="fig:long-wavelength"}](long-wavelength){width="\columnwidth"}
If the Gaussian fluctuation corrections are of the same order as the mean-field contribution to the zero-temperature gap equation, they can be effectively restated as a modification of the prefactor of the logarithmic contribution. Then the solution of Eq. can be written in the familiar form by introducing the effective interaction $$\lambda_\mathrm{eff}=\lambda\bigg[1-\frac{1}{{\rho_{\mathrm{F}}}L^{d}\ln\big(\frac{2 E_\mathrm{F}}{\Phi_0}\big)}\frac{{\mathrm{d}}E_\mathrm{fluct}(\Phi_0)}{{\mathrm{d}}\Phi_0^2}\bigg]{~ .}$$ Hence the sign of the fluctuation contribution plays an interesting role as negative fluctuation corrections further enhance the gap as compared to its mean-field value whereas positive fluctuation contributions result in a reduction, i.e., weaken the ordered state.
Let us now analyze the structure of fluctuation corrections in the case of density-wave instabilities due to nesting following the logic of Ref. , also building on the calculations presented in Ref. . More details specific to our derivation for both charge-density wave order and spin-density wave order are presented in Appendix \[app:calculation\]. The Gaussian fluctuation corrections can be expressed in terms of the fluctuation propagator $\boldsymbol{\mathcal{D}}_q$ as $E_\mathrm{fluct}(\Phi_0)=\frac{L^d}{2}\int_q\ln\det({\boldsymbol{\mathcal{D}}}^{-1}_q)$, where the dimensionality of the order parameter translates to the dimensionality of the fluctuation propagator. Accordingly, the Gaussian fluctuation corrections to the zero-temperature gap equation take the form $$\frac{1}{{\rho_{\mathrm{F}}}L^d}\frac{{\mathrm{d}}E_\mathrm{fluct}(\Phi_0)}{{\mathrm{d}}\Phi_0^2}=\frac{1}{2{\rho_{\mathrm{F}}}}\int_q\,\frac{\frac{{\mathrm{d}}}{{\mathrm{d}}\Phi_0^2}\det(\boldsymbol{\mathcal{D}}^{-1}_q)}{\det(\boldsymbol{\mathcal{D}}^{-1}_q)}{~ ,}\label{eq:fluctuation-corrections}$$ where we introduced $q\equiv({\boldsymbol{q}},\omega)$ and the integration $\int_q\ldots=\int\frac{{\mathrm{d}}\omega}{2\pi}\int\frac{{\mathrm{d}}^d{\boldsymbol{q}}}{(2\pi)^d}\ldots$ runs over all external frequencies and momenta up to the cutoff. Consequently, an expansion of the fluctuation spectrum in small momenta and frequencies as often discussed in the context of collective modes is not sufficient here. Instead, the fluctuation propagator has to be determined for all frequencies and momenta, and in fact, the short-wavelength fluctuations with momenta $2\xi^{-1}=2\Delta_0/v_\mathrm{F}\ll|{\boldsymbol{q}}|\ll k_\mathrm{F}$ turn out to be crucial, as they give rise to an additional logarithmic contribution to the zero-temperature gap equation as discussed in Sec. \[sec:results\].
The inverse matrix of the fluctuation propagator for density-wave order can be stated in terms of the polarization matrix $\boldsymbol{\Pi}_q$ as $$\boldsymbol{\mathcal{D}}^{-1}_q=\frac{1}{V}\boldsymbol{\mathds{1}}-\boldsymbol{\Pi}_q{~ .}$$ For density-wave order, transverse and longitudinal fluctuations are not coupled and hence the polarization matrix is diagonal. The Gaussian fluctuation corrections can thus be expressed in terms of longitudinal and transverse contributions using $$E_\mathrm{fluct}(\Phi_0)=\frac{L^d}{2}\int_q\ln\Big[4^N\Big(\frac{1}{V}-\Pi^\perp_q\Big)^{N-1}\Big(\frac{1}{V}-\Pi^\parallel_q\Big)\Big]{~ .}$$ Both the longitudinal and the transverse contribution consist of a normal part $\Pi^\mathrm{n}_q=\frac{1}{2}\int_k(G_{k+q}G_{-k}+G_{-k}G_{k-q})$ which survives the limit $\Phi_0\rightarrow0$, and an anomalous contribution $\Pi^\mathrm{a}_q=-\int_kF_kF_{k+q}$ which vanishes in the high-temperature normal state, $$\begin{aligned}
\Pi^\perp_q&=\Pi^\mathrm{n}_q-\Pi^\mathrm{a}_q{~ ,}\label{eq:Pi-perp} \\
\Pi^\parallel_q&=\Pi^\mathrm{n}_q+\Pi^\mathrm{a}_q{~ .}\label{eq:Pi-parallel}\end{aligned}$$ These integrals (see Appendix \[app:polarization-function\] for details) are of the same structure as those arising in the context of superconductivity, and hence we can build on the results obtained by previous studies [@KosMillisLarkin-PRB2004; @FischerEtAl2017] in the remainder.
Results and Discussion {#sec:results}
======================
![Diagrammatic representation of the terms contributing to the derivative of the polarization function. For the definition of the diagrammatic elements see Fig. \[fig:diagrammatic-elements\]. (a) and (b) Derivatives of the anomalous part, while (c) represents the derivative of the normal contribution. While in the long-wavelength regime, the fermionic triangle part depicted here becomes structureless (cf. Fig. \[fig:long-wavelength\]), it is of particular importance for the contributions stemming from the regime of short wavelengths. The diagram which yields the crucial contribution to the gap equation discussed in this paper is highlighted in gray.[]{data-label="fig:triangles"}](triangle-diagram-DW){width="\columnwidth"}
We can straightforwardly adopt the formalism developed in the context of superconductivity [@KosMillisLarkin-PRB2004; @FischerEtAl2017] to calculate the leading-order corrections to the zero-temperature mean-field gap equation due to Gaussian fluctuations of the order parameter for density-wave order. We consider the regimes of small and large momenta/frequencies separately, which is possible [@VaksGalitskiiLarkin-JETP1962] since the integrals only depend on the combination $r=\sqrt{\omega^2+(v_\mathrm{F}|{\boldsymbol{q}}|\cos\theta)^2}/(2\Phi_0)$, where $\theta$ denotes the angle between fermionic momentum ${\boldsymbol{k}}$ and bosonic momentum ${\boldsymbol{q}}$. The crucial contribution to the fluctuation corrections, as given by Eq. and diagrammatically represented in Fig. \[fig:gap-equation\](b), stems from the regime where $r\gg1$ and $v_\mathrm{F}|{\boldsymbol{q}}|>\omega$, whereas long-wavelength fluctuations lead to corrections that are negligible compared to the mean-field terms. This is due to the fact that the fermionic triangle (cf. Fig. \[fig:triangles\]) associated with the derivative of the polarization function becomes structureless in the limit $r\ll1$ and hence fluctuation corrections reduce to the simpler form shown in Fig. \[fig:long-wavelength\].
In the regime of interest, the fluctuation propagator is dominated by the normal part of the polarization function, while its derivative is largely determined by the anomalous part. Therefore, the leading contribution to the gap equation evaluates to
\[eq:result\]$$\begin{aligned}
&\frac{1}{{\rho_{\mathrm{F}}}L^d}\frac{{\mathrm{d}}E_\mathrm{fluct}(\Phi_0)}{{\mathrm{d}}\Phi_0^2}\Big|_{r\gg1} \approx \frac{1}{2{\rho_{\mathrm{F}}}}\int_q\frac{[(N-1)-1]\frac{{\mathrm{d}}\Pi^\mathrm{a}_q}{{\mathrm{d}}\Phi_0^2}}{\frac{1}{V}-\Pi^\mathrm{n}_q} \\
&\qquad \qquad \approx -\frac{1}{2}\frac{(N-1)-1}{8\pi (d-1)}\ln\Big(\frac{E_\mathrm{F}}{\Phi_0}\Big){~ .}\end{aligned}$$
This reveals that Gaussian fluctuations indeed quite generically entail an additional logarithmic contribution to the gap equation for density-wave instabilities due to nesting. This logarithmic divergence can be traced back to the fermionic triangle depicted in Fig. \[fig:triangles\](a), which results from the derivative of the anomalous part of the polarization function that enters the numerator of ${\mathrm{d}}E_\mathrm{fluct}(\Phi_0)/{\mathrm{d}}\Phi_0^2$, while the two other contributions \[Figs. \[fig:triangles\](b) and \[fig:triangles\](c)\] do not give rise to additional logarithms. Thus, the precarious process that invalidates the mean-field approach in the case of density-wave order is the interaction of quasiparticles through fluctuations.
fluctuations
------------------ -------------- ----------------
Heisenberg SDW ($N=3$) increase gap
XY SDW ($N=2$) are negligible
Ising SDW or CDW ($N=1$) decrease gap
: Gaussian fluctuation corrections in $\mathrm{O}(N)$ models. Depending on the dimensionality of the order parameter, fluctuation corrections can increase or decrease the gap value compared to its mean-field value.[]{data-label="tab:summary-sdw-fluctuations"}
The above result is in stark contrast to the insignificant role that fluctuations play in the context of superconductivity: Analogous contributions to the BCS gap equation are not only suppressed by the smallness of the Debye energy as compared to the Fermi energy, but the corresponding contributions stemming from fluctuations of phase and amplitude of the complex order parameter even cancel exactly. Nonetheless, the corrections due to the longitudinal mode and the $N-1$ transverse modes enter the result with opposite signs, and hence the prefactor depends on the number of transverse modes. Only for XY spins ($N=2$), the large fluctuation corrections stemming from the regime $r\gg1$ cancel – which is consistent with previous results in the context of superconductivity [@KosMillisLarkin-PRB2004; @FischerEtAl2017]. In conclusion, the analysis of the role of fluctuation corrections in the case of density-wave order reveals that mean-field approaches are generally not justified in this situation, the only exception being spin-density wave order of XY spins.
Furthermore, the sign of the fluctuation corrections allows us to judge whether the presence of fluctuations is advantageous or detrimental to the formation of density-wave order: The effective interaction $$\begin{aligned}
\lambda_\mathrm{eff} &= \lambda\bigg[1+\frac{N-2}{16\pi(d-1)}\bigg]{~ ,}\label{eq:effective-interaction}
$$ which governs the ordering in the presence of fluctuations, either decreases (${\mathrm{d}}E_\mathrm{fluct}/{\mathrm{d}}\Phi_0^2>0$) or increases (${\mathrm{d}}E_\mathrm{fluct}/{\mathrm{d}}\Phi_0^2<0$), and the same is true for the solution of the zero-temperature gap equation in the presence of fluctuations. For $N=1$ or $N=3$ and $d=3$, the relative change in $\lambda$ is $1/(32\pi)\simeq0.01$, i.e., rather small. More important than the numerical value of this correction is the fact that there is no guarantee that even higher-order processes give rise to equally non-negligible corrections. Our analysis of fluctuation corrections shows that contributions stemming from the longitudinal mode lead to a decrease of the gap compared to its mean-field value, whereas transverse fluctuations increase the gap. Hence, if the latter dominate, fluctuations are favorable to the formation of density-wave order, as it is the case for spin-density wave order of Heisenberg spins. If, on the other hand, transverse fluctuations cannot compensate for the effect of the longitudinal mode, the ordered state is weakened. This applies to mean-field theories for charge-density wave order as well as for spin-density wave order of Ising spins, see also Table \[tab:summary-sdw-fluctuations\] for an overview of our results.
Channel interference in the renormalization group approach {#sec:rg}
==========================================================
{width="\textwidth"}
Our analysis of the role of fluctuations for density-wave instabilities has been performed at zero temperature, i.e., deep inside the ordered phase. In doing so, we concentrated on a single channel and neglected the potential presence of competing instabilities. Another, complementary perspective is provided by a renormalization group (RG) analysis [@Shankar-RevModPhys1994], in which competing instabilities of the system can be treated on equal footing and thereby channel interference can be studied within this framework. However, within the RG scheme, the phase transition is approached coming from high energies, allowing us to determine the leading instability of the system, but this approach is less suited to explore the ordered state further. For example, the model defined by the Hamiltonian $\mathcal{H}=\mathcal{H}_0+\mathcal{H}_\mathrm{int}$ as stated in Eqs. and is in principle also unstable towards the formation of superconductivity, which is expected to interfere with SDW ordering.
In fact, the complex phase diagrams of materials of current interest such as the iron-based superconductors can be understood as a result of the interplay of different competing instabilities: Already the simple two-band model allows for superconductivity as well as SDW order and CDW order [@ChubukovEfremovEremin-PRB2008; @PodolskyKeeKim-EPL2009; @KangTesanovic-PRB2011], owing to the nested nature of the Fermi surface. This model is essentially a band-basis translation of the model investigated in this paper. Hence the calculation of fluctuation corrections to the corresponding zero-temperature gap equations is straightforward, see Appendix \[app:calculation\]. What is more, we can use the RG equations derived for the same model [@ChubukovEfremovEremin-PRB2008] to analyze the effect of channel interference on the transition temperature towards a given ordered state. The resulting flow of the couplings in the density-wave channels as a function of $t=\log\frac{W}{E}$, where $W$ is the bandwidth and $E$ the running energy scale, takes the form
\[eq:channel-interference-couplings\] $$\begin{aligned}
\dot{\Gamma}_\mathrm{SDW}&= (\Gamma_\mathrm{SDW})^2\pm 2u_3(u_1-u_2-u_4){~ ,}\\
\dot{\Gamma}_\mathrm{CDW}&= (\Gamma_\mathrm{CDW})^2\mp 2u_3(u_1+u_2-u_4) {~ .}\end{aligned}$$
The couplings $u_i$ refer to different intraband and interband processes that are connected to the couplings in spin and charge channel according to Eq. . If the second term were zero, this would result in the usual logarithmic divergence $\Gamma=\Gamma_0/(1-\Gamma_0\log\frac{W}{E})$ of the coupling in the respective channel. Hence the presence of the second term implies corrections due to channel interference that are intrinsic to the model as long as the interband pair-hopping process which is associated with $u_3$, and involves a momentum transfer of $2{\boldsymbol{Q}}$, is effective. Motivated by the structure of Eq. , we used $u_3^{(0)}$ as a measure of channel interference in our brief analysis, where we tuned the bare couplings such that the instability in the channel under consideration is favored within the mean-field description, while the bare couplings in the competing channels were tuned to zero. We then kept the bare value of the coupling in a given channel fixed while increasing the channel interference strength. Details on our derivation can be found in Appendix \[app:two-band\]. We find that the energy scale at which the coupling diverges is affected by channel interference. Furthermore, the effect of channel interference on charge-density wave order and spin-density wave order is different: While increasing the channel-interference strength via $u_3^{(0)}$ results in higher transition temperatures for SDW order (since the divergence is pushed to lower energies), the opposite is true for CDW order. In Fig. \[fig:channel-interference\], we present our numerical solution of the RG equations for SDW order characterized by a real order parameter and CDW order associated with a purely imaginary order parameter, since these are the two instabilities arising from the parameter range usually discussed in the context of iron-based superconductors.
While our analysis and the perturbative RG investigation address rather different phenomena, we note that the trends we find from the RG flow are consistent with our analysis of fluctuation corrections: Channel interference is favorable for the formation of spin-density wave order as it leads to an increase of the corresponding transition temperature, whereas the transition towards charge-density wave order is hindered by channel interference in the sense that the corresponding transition temperature decreases.
Summary
=======
Mean-field theories for density-wave order resulting from a nesting of the Fermi surface can be derived from a microscopic model of interacting fermions in full analogy to the formulation of BCS theory in the language of field integrals. The BCS theory of superconductivity is the poster child of mean-field theories since neither thermal nor quantum fluctuations lead to sizable effects in conventional superconductors and hence can be neglected. In this paper, we showed that, in contrast to superconductivity (where the order parameter is a complex scalar), the impact of fluctuations is crucial in the case of commensurate density-wave order (characterized by a real $N$-component order parameter) as long as $N\neq2$.
To be specific, we have investigated the role of fluctuations for charge-density wave order and spin-density wave order due to nesting of the Fermi surface. Our main finding is that, generally, fluctuation corrections to the zero-temperature gap equations for such density-wave instabilities are of the same order $\mathcal{O}[\ln(E_\mathrm{F}/\Phi_0)]$ as the terms already appearing at mean-field level. In conclusion, the mean-field theories for density-wave instabilities are not justified since the large fluctuation corrections imply that the respective mean-field theory cannot capture all relevant contributions. Of course, our analysis does not imply that the effect of fluctuations is merely to replace $\lambda$ by $\lambda_\mathrm{eff}$ of Eq. , as there is no guarantee that even higher-order processes will not give rise to corrections of the same order. Moreover, we find that the additional logarithmic contribution to the gap equation stemming from fluctuations originates from the derivative of the anomalous polarization function, and the crucial process is the interaction of quasiparticles through the exchange of fluctuations depicted in Fig. \[fig:triangles\](a).
Interestingly, we find that the impact of longitudinal and transverse modes is quite different: Longitudinal fluctuations always yield $E_\mathrm{fluct}>0$ and are thus detrimental to the formation of long-range order. Transverse fluctuations, on the other hand, only yield $E_\mathrm{fluct}>0$ in the long-wavelength regime. In the opposite regime of transverse fluctuations on lengthscales smaller than the coherence length, we find that the respective fluctuation correction surprisingly lowers the energy. Furthermore, this contribution is the dominant one since it yields the additional logarithm to the gap equation in the case of perfect nesting that ultimately leads to an increase of the gap as compared to its mean-field value. It is due to the twist of the “phase” induced by the excitation of quasiparticles inside the coherence volume which enhances the kinetic energy of quasiparticle excitations. In contrast, longitudinal fluctuations can only lead to an increase of the energy since the potential energy is already minimized by the mean-field configuration.
Because of the different nature of longitudinal and transverse fluctuations, the case $N=2$ is an interesting exception: The logarithmic contributions to the gap equation stemming from the longitudinal mode and the single transverse mode cancel exactly which renders the overall fluctuation corrections negligible. This cancellation legitimates the mean-field approach to density-wave order of XY spins. This is in accordance with the analysis of fluctuations in the context of superconductivity [@KosMillisLarkin-PRB2004; @FischerEtAl2017], where fluctuation corrections from phase and amplitude mode cancel analogously, providing the justification of the BCS mean-field theory.
Acknowledgments
===============
We thank M. Bard, A. V. Chubukov, S. Fischer, M. Hecker, N. Kainaris, and M. S. Scheurer for helpful discussions.
Calculation of fluctuation corrections {#app:calculation}
======================================
This Appendix provides technical details on our calculation of fluctuation corrections, closely following the logic and notation introduced in the context of superconductivity in Ref. and extended by Ref. .
In what follows, we consider the partition function $\mathcal{Z}=\int\mathcal{D}[\bar{\psi},\psi]\,\exp(-\mathcal{S}[\bar{\psi},\psi])$ with the appropriate action $\mathcal{S}[\bar{\psi},\psi]=\int_0^\beta{\mathrm{d}}\tau\,(\sum_\sigma\int{\mathrm{d}}^d{\boldsymbol{x}}\,\bar{\psi}_\sigma({\boldsymbol{x}})\partial_\tau\psi_\sigma({\boldsymbol{x}})+\mathcal{H})$ corresponding to the Hamiltonian as stated in Eqs. and . The effective theory in terms of the order parameter ${\boldsymbol{\Phi}}$ follows from the Hubbard-Stratonovich transformation and successively integrating out the fermions, resulting in $\mathcal{Z}=\int\mathcal{D}{\boldsymbol{\Phi}}\,\exp(-\mathcal{S}_\mathrm{eff}[{\boldsymbol{\Phi}}])$.
Since the derivation of fluctuation corrections to the zero-temperature mean-field gap equation for charge-density wave (CDW) order and spin-density wave (SDW) order follow the same logic and primarily differ in the dimensionality of the respective order parameter ${\boldsymbol{\Phi}}\in\mathds{R}^N$, we treat them simultaneously here. For CDW order, the order parameter (associated with the charge density $\rho$) introduced by the Hubbard-Stratonovich transformation is a scalar, $$\rho_{{\boldsymbol{q}}}=\sum_{{\boldsymbol{k}},\sigma,\sigma^\prime}\big<\psi_{{\boldsymbol{k}},\sigma}^\dagger\delta_{\sigma\sigma^\prime}\psi_{{\boldsymbol{k}}+{\boldsymbol{Q}}+{\boldsymbol{q}},\sigma^\prime}^{}\pm\psi_{{\boldsymbol{k}}+{\boldsymbol{Q}},\sigma}^\dagger\delta_{\sigma\sigma^\prime}\psi_{{\boldsymbol{k}}+{\boldsymbol{q}},\sigma^\prime}^{}\big>{~ ,}$$ corresponding to $N=1$. The upper sign refers to CDW order characterized by a real (r) order parameter, whereas the lower sign refers to CDW order with an imaginary (i) order parameter. The latter follows from assuming $V<0$ in the respective channel. All derivations for iCDW order can be performed in complete analogy to those for rCDW order and since we come to the same conclusions in both cases, we concentrate on rCDW order in the following. For SDW order of Heisenberg spins, the order parameter (associated with the magnetization ${\boldsymbol{M}}$) is a three-component vectorial object, $${\boldsymbol{M}}_{{\boldsymbol{q}}}=\sum_{{\boldsymbol{k}},\sigma,\sigma^\prime}\big<\psi_{{\boldsymbol{k}},\sigma}^\dagger{\boldsymbol{\sigma}}_{\sigma\sigma^\prime}\psi_{{\boldsymbol{k}}+{\boldsymbol{Q}}+{\boldsymbol{q}},\sigma^\prime}^{}\pm\psi_{{\boldsymbol{k}}+{\boldsymbol{Q}},\sigma}^\dagger{\boldsymbol{\sigma}}_{\sigma\sigma^\prime}\psi_{{\boldsymbol{k}}+{\boldsymbol{q}},\sigma^\prime}^{}\big>{~ ,}$$ which corresponds to $N=3$. Again, for the sake of clarity, we only discuss rSDW order (upper sign) since, mutatis mutandis, the same results can be obtained for iSDW order (lower sign).
For both types of density-wave order, the order parameter (which we denote ${\boldsymbol{\Phi}}$ henceforth) can be split into the static and homogeneous mean-field value $\Phi_0$ and fluctuations $\delta{\boldsymbol{\Phi}}$ around this mean-field value as $${\boldsymbol{\Phi}}_q=\Phi_0\hat{{\boldsymbol{e}}}\delta_{q,0}+\delta{\boldsymbol{\Phi}}_q{~ ,}$$ where we introduced $\hat{{\boldsymbol{e}}}$ as the unit vector along the direction of the mean-field order parameter. For CDW order, $\hat{{\boldsymbol{e}}}=1$, while for SDW order, we assume w.l.o.g. $\hat{{\boldsymbol{e}}}=\hat{{\boldsymbol{e}}}_3$ in the remainder.
The usual procedure of integrating out the fermions after the decoupling then leads to the effective action in terms of the fermionic Green’s function $$\mathcal{S}_\mathrm{eff}(\Phi_0)=\int_q\frac{2|{\boldsymbol{\Phi}}_q|^2}{V}-\int_{k,k^\prime}\operatorname{tr}\ln(\mathcal{G}_{kk^\prime}^{-1}) {~ ,}$$ which can be split into a mean-field part and fluctuations as well using $$\begin{aligned}
&\operatorname{tr}\ln\big(-\mathcal{G}^{-1}_{kk^\prime}\big)=\operatorname{tr}\ln\big[-\big((\mathcal{G}^\mathrm{MF}_k)^{-1}\delta_{kk^\prime}+\eta_{kk^\prime}\big)\big]\\
&\quad =\operatorname{tr}\ln\big[-\big(\mathcal{G}^{\mathrm{MF}}_k)^{-1}\delta_{kk^\prime}\big)\big]-\frac{1}{2}\operatorname{tr}\big(\mathcal{G}^\mathrm{MF}_k\eta_{kk^\prime}\mathcal{G}^\mathrm{MF}_{k^\prime}\eta_{k^\prime k}\big) \nonumber \\
&\quad \qquad +\mathcal{O}\big[(\delta{\boldsymbol{\Phi}})^3\big] \end{aligned}$$ by expanding the fluctuations up to Gaussian order. Here, the mean-field part of the inverse matrix Green’s function in Matsubara representation is given by $$\begin{aligned}
&\big(\mathcal{G}_k^{\mathrm{MF}})^{-1}=\tfrac{1}{2}\left(\mathrm{i}\nu_n\tau_0\sigma_0-\varepsilon_{{\boldsymbol{k}}}\tau_3\sigma_0\right) \nonumber \\
&\qquad +\tfrac{1}{2}\tau_1\left\{\begin{matrix}\Phi_0 \sigma_0 & \text{for CDW order} {~ ,}\\ {\boldsymbol{\Phi}}_0\cdot {\boldsymbol{\sigma}}& \text{for SDW order}{~ ,}\end{matrix}\right.\end{aligned}$$ while the fluctuation matrix is given by $$\eta_{kk^\prime}=\tfrac{1}{2}\tau_1\left\{\begin{matrix} \delta\Phi_{k-k^\prime}\sigma_0 & \text{for CDW order} {~ ,}\\ \delta{\boldsymbol{\Phi}}_{k-k^\prime}\cdot{\boldsymbol{\sigma}} & \text{for SDW order}{~ .}\end{matrix} \right.$$ Note that when considering density-wave order with an imaginary order parameter, the order parameter and its fluctuations are associated with $\tau_2$ rather than $\tau_1$. One easily finds that this change will not affect the conclusions of our analysis. Then the partition function can be expressed as $$\mathcal{Z}=\int\mathcal{D}[\bar{\Psi},\Psi]\, \mathrm{e}^{-(\mathcal{S}_0+\mathcal{S}_\mathrm{int})}\approx\mathrm{e}^{-(\mathcal{S}_\mathrm{MF}+\mathcal{S}_\mathrm{fluct})} {~ ,}$$ where the mean-field action takes the form $$\mathcal{S}_\mathrm{MF}(\Phi_0)=\frac{2\Phi_0^2}{V}-\int_k\operatorname{tr}\ln\Big[-\big(\mathcal{G}_k^\mathrm{MF}\big)^{-1}\Big]{~ ,}$$ resulting in the famous form of the gap equation $$\begin{aligned}
0&=\frac{{\mathrm{d}}\mathcal{S}_\mathrm{MF}(\Phi_0)}{{\mathrm{d}}\Phi_0^2}=\frac{2}{V}-\int\frac{{\mathrm{d}}^d{\boldsymbol{k}}}{(2\pi)^d}\int_{-\infty}^\infty\frac{{\mathrm{d}}\nu}{2\pi}\,\frac{2}{\varepsilon_{{\boldsymbol{k}}}^2+\nu^2+\Phi_0^2} \nonumber \\
&\approx\frac{2}{V}-2{\rho_{\mathrm{F}}}\ln\Big(\frac{2 E_\mathrm{F}}{\Phi_0}\Big) \end{aligned}$$ at zero temperature. The Gaussian fluctuation part can be integrated exactly, $$\begin{aligned}
\begin{split}
\mathrm{e}^{-\mathcal{S}_\mathrm{fluct}}&=\int\mathcal{D}[\delta{\boldsymbol{\Phi}}]\,\mathrm{e}^{-\frac{1}{2}\int_q\delta{\boldsymbol{\Phi}}_q\cdot{\boldsymbol{\mathcal{D}}}_q^{-1}\cdot\delta{\boldsymbol{\Phi}}_{-q}} \\
&= \mathrm{e}^{-\frac{1}{2}\int_q\ln\det[{\boldsymbol{\mathcal{D}}}^{-1}_q]}{~ .}\end{split}\end{aligned}$$ Alternatively, the inverse fluctuation propagator matrix ${\boldsymbol{\mathcal{D}}}^{-1}_q$, and consequently the Gaussian fluctuation corrections to the action, can be expressed in terms of the polarization matrix ${\boldsymbol{\Pi}}_q$ as $${\boldsymbol{\mathcal{D}}}_q^{-1}= \frac{4}{V}{\boldsymbol{\mathds{1}}}-{\boldsymbol{\Pi}}_q$$ where ${\boldsymbol{\Pi}}_q$ is either given by $$\Pi^\mathrm{CDW}_q=-\frac{1}{4}\int_k\operatorname{tr}\big[\mathcal{G}^\mathrm{CDW}_{k+\frac{q}{2}}(\tau_1\sigma_0)\mathcal{G}^\mathrm{CDW}_{k-\frac{q}{2}}(\tau_1\sigma_0)\big] {~ ,}$$ or by $$\begin{aligned}
({\boldsymbol{\Pi}}^\mathrm{SDW}_q)_{ij}&=-\frac{1}{4}\int_k\operatorname{tr}\big[\mathcal{G}^\mathrm{SDW}_{k+\frac{q}{2}}(\tau_1\sigma_i)\mathcal{G}_{k-\frac{q}{2}}^\mathrm{SDW}(\tau_1\sigma_j)\big] \\
&=\begin{pmatrix} 4\Pi^\perp_q & 0 &0 \\ 0 & 4\Pi^\perp_q & 0\\ 0 & 0 & 4 \Pi^\parallel_q\end{pmatrix}\label{eq:Pi-SDW-matrix}{~ .}\end{aligned}$$ In the last line, we introduced longitudinal and transverse contributions $$\begin{aligned}
\Pi^\perp_q&=\Pi^\mathrm{n}_q-\Pi^\mathrm{a}_q \\
\Pi^\parallel_q&=\Pi^\mathrm{n}_q+\Pi^\mathrm{a}_q \end{aligned}$$ in terms of the normal ($\Pi_q^\mathrm{n}$) and anomalous ($\Pi_q^\mathrm{a}$) part of the polarization function, which are discussed in more detail in Appendix \[app:polarization-function\]. The generalization to spin dimensionality $N$ is straightforward, and the resulting polarization matrix differs from Eq. only in the number of transverse modes. The fluctuation corrections to the action then take the form $$\mathcal{S}_\mathrm{fluct}(\Phi_0)=\frac{1}{2}\int_q\ln\Big[4^N\Big(\frac{1}{V}-\Pi_q^\perp\Big)^{N-1}\Big(\frac{1}{V}-\Pi_q^\parallel\Big)\Big]$$ for both SDW order and CDW order, where the latter corresponds to $N=1$.
Owing to the diagonal structure of ${\boldsymbol{\mathcal{D}}}^{-1}_q$, the Gaussian fluctuation corrections to the gap equation, $$\begin{aligned}
\frac{{\mathrm{d}}\mathcal{S}_\mathrm{fluct}(\Phi_0)}{{\mathrm{d}}\Phi_0^2}&=\frac{1}{2}\int_q\frac{\frac{{\mathrm{d}}}{{\mathrm{d}}\Phi_0^2}\det({\boldsymbol{\mathcal{D}}}_q^{-1})}{\det({\boldsymbol{\mathcal{D}}}_q^{-1})} \\
&= -\frac{1}{2}\int_q\Bigg[\frac{(N-1)\frac{{\mathrm{d}}\Pi^\perp_q}{{\mathrm{d}}\Phi_0^2}}{\frac{1}{V}-\Pi^\perp_q}+\frac{\frac{{\mathrm{d}}\Pi^\parallel_q}{{\mathrm{d}}\Phi_0^2}}{\frac{1}{V}-\Pi^\parallel_q}\Bigg]{~ ,}\nonumber\end{aligned}$$ can alternatively be written as $$\frac{{\mathrm{d}}\mathcal{S}_\mathrm{fluct}(\Phi_0)}{{\mathrm{d}}\Phi_0^2}=-\frac{1}{2}\int_q
\Big({\boldsymbol{\mathcal{D}}}_q\cdot\frac{{\mathrm{d}}{\boldsymbol{\Pi}}_q}{{\mathrm{d}}\Phi_0^2}\Big){~ ,}\label{eq:fluctuation-integral}$$ which is represented by the diagram in Fig. \[fig:gap-equation\](b), where the fermionic triangle (also see Fig. \[fig:triangles\]) represents the derivative of the polarization function, which is also evaluated in Appendix \[app:polarization-function\]. As shown in Refs. and , the integral is dominated by contributions from the regime where $r=\sqrt{\omega^2+(v_\mathrm{F}|{\boldsymbol{q}}|\cos\theta)^2}/(2\Phi_0)\gg1$ as well as $v_\mathrm{F}|{\boldsymbol{q}}|>\omega$, whereas corrections stemming from long-wavelength fluctuations are negligible. In that regime, the fluctuation propagator is dominated by the normal part of the polarization function, $|1/V-\Pi^\mathrm{n}_q|\gg|\Pi^\mathrm{a}_q|$, while the leading contribution to the derivative of the polarization function stems from the anomalous part, $|{\mathrm{d}}\Pi^\mathrm{a}_q/{\mathrm{d}}\Phi_0^2|\gg|{\mathrm{d}}\Pi^\mathrm{n}_q/{\mathrm{d}}\Phi_0^2|$. Consequently, the crucial correction terms to the gap equation due to fluctuations are given by $$\begin{aligned}
\left.\frac{{\mathrm{d}}\mathcal{S}_\mathrm{fluct}(\Phi_0)}{{\mathrm{d}}\Phi_0^2}\right|_{r\gg1}\approx \frac{1}{2}\int_q\frac{[(N-1)-1]\frac{{\mathrm{d}}\Pi_q^\mathrm{a}}{{\mathrm{d}}\Phi_0^2}}{\frac{1}{V}-\Pi_q^\mathrm{n}}{~ .}\end{aligned}$$ Then, depending on the number of transverse modes, i.e., the dimensionality of the order parameter, the fluctuation corrections are either positive ($N<2$), negative ($N>2$), or negligible ($N=2$).
The polarization function and its derivatives {#app:polarization-function}
=============================================
For the sake of completeness, we summarize the results for the polarization function and its derivatives obtained by previous studies [@KosMillisLarkin-PRB2004; @FischerEtAl2017]. To pinpoint the physical meaning of the individual terms eventually contributing to the gap equation, we introduce normal and anomalous fermionic Green’s functions as $$\begin{aligned}
G_k&\equiv G({\boldsymbol{k}},\nu_n)= -\frac{\mathrm{i}\nu_n+\varepsilon_{{\boldsymbol{k}}}}{\varepsilon_{{\boldsymbol{k}}}^2+\nu_n^2+\Phi_0^2} \\
\text{and} \quad F_k&\equiv F({\boldsymbol{k}},\nu_n)= \frac{\Phi_0}{\varepsilon_{{\boldsymbol{k}}}^2+\nu_n^2+\Phi_0^2}{~ ,}\end{aligned}$$ respectively. It holds that $F_k=F_{-k}$, and furthermore, nesting implies that $G({\boldsymbol{k}}+{\boldsymbol{Q}},\nu_n)=-G(-{\boldsymbol{k}},-\nu_n)\equiv -G_{-k}$ as well as $F({\boldsymbol{k}}+{\boldsymbol{Q}},\nu_n)=F({\boldsymbol{k}},\nu_n)\equiv F_k$. Using these definitions, the matrix Green’s functions can be rewritten as $$\begin{aligned}
\mathcal{G}_k^\mathrm{CDW}&= 2\begin{pmatrix} G_k\sigma_0 & F_k\sigma_0 \\ F_k\sigma_0 & -G_{-k}\sigma_0 \end{pmatrix} \\
\text{and} \quad \mathcal{G}_k^\mathrm{SDW} &= 2\begin{pmatrix} G_k \sigma_0 & F_k \sigma_3 \\ F_k\sigma_3 & -G_{-k}\sigma_0\end{pmatrix} {~ ,}\end{aligned}$$ which makes the structural congruence with the BCS theory of superconductivity obvious. The normal part of the polarization function is then solely determined by normal Green’s functions. It corresponds to $\Pi^\mathrm{n}_q=\frac{1}{2}(\Pi_{11,q}+\Pi_{22,q})$ in the notation of Ref. and reads
$$\begin{aligned}
\Pi^\mathrm{n}_q&= \frac{1}{2}\int_k\big[G_{k+\frac{q}{2}}G_{-(k-\frac{q}{2})}+G_{-(k+\frac{q}{2})}G_{k-\frac{q}{2}}\big] \\
&=\int_k\frac{\nu_+\nu_-+\varepsilon_+\varepsilon_-}{(\varepsilon_+^2+\nu_+^2+\Phi_0^2)(\varepsilon_-^2+\nu_-^2+\Phi_0^2)} \\
&={\rho_{\mathrm{F}}}\ln\bigg(\frac{2 E_\mathrm{F}}{\Phi_0}\bigg)-{\rho_{\mathrm{F}}}\int_\Omega\frac{(2r^2+1)\operatorname{arsinh}(r)}{2r\sqrt{r^2+1}}\label{eq:normal-Pi} {~ ,}\end{aligned}$$
where we abbreviated $\int_k\ldots={\rho_{\mathrm{F}}}\int{\mathrm{d}}\varepsilon\int\frac{{\mathrm{d}}\nu}{2\pi}\int_\Omega\ldots$ and the integration $\int_\Omega\dots =\frac{1}{\Omega_d}\int{\mathrm{d}}\Omega\dots$ refers to the integration over the direction of the fermionic momentum with $\Omega_d$ being the volume of a $d$-dimensional sphere. Since we consider the spectrum to be isotropic, the integrand only depends on the momentum direction via $\theta=\sphericalangle({\boldsymbol{k}},{\boldsymbol{q}})$. Furthermore, we introduced $\nu_\pm=\nu\pm\frac{\omega}{2}$ as well as ${\boldsymbol{k}}_\pm={\boldsymbol{k}}\pm\frac{1}{2}{\boldsymbol{q}}$, and linearized the dispersion $\varepsilon_\pm=\varepsilon\pm \frac{1}{2}v_\mathrm{F}|{\boldsymbol{q}}|\cos\theta$. For the evaluation of the polarization function, it is useful that the integrals discussed here depend on external momenta and frequency only via the combination $$r=\frac{\sqrt{\omega^2+(v_\mathrm{F} |{\boldsymbol{q}}|\cos\theta)^2}}{2\Phi_0}{~ ,}$$ see Refs. and for details. Evaluating the above expression at $|{\boldsymbol{q}}|=0$ and $\omega=2\Phi_0r$ then results in the last line, where the angular integration still remains to be done. Unfortunately, this cannot be performed for arbitrary values of $r$, and we resort to approximations in the regimes $r\ll1$ and $r\gg1$, cf. Refs. and for details.
Analogously, the anomalous part of the polarization function can be expressed as
$$\begin{aligned}
\Pi^\mathrm{a}_q&= -\int_k F_{k+\frac{q}{2}}F_{k-\frac{q}{2}} \\
&= -\int_k\frac{\Phi_0^2}{(\varepsilon_+^2+\nu_+^2+\Phi_0^2)(\varepsilon_-^2+\nu_-^2+\Phi_0^2)} \\
&= -{\rho_{\mathrm{F}}}\int_\Omega\frac{\operatorname{arsinh}(r)}{2r\sqrt{r^2+1}} \label{eq:anomalous-Pi} {~ .}\end{aligned}$$
In the notation of Ref. , this corresponds to $\Pi^\mathrm{a}_q=\frac{1}{2}(\Pi_{11,q}-\Pi_{22,q})$. This part vanishes in the limit $\Phi_0\rightarrow0$, i.e., in the disordered high-temperature phase. Furthermore, it is obvious from the expressions and that $$\left|\tfrac{1}{V}-\Pi_q^\mathrm{n}\right|>\left|\Pi_q^\mathrm{a}\right|$$ holds for arbitrary $r$.
The corresponding derivatives can also be expressed in terms of normal and anomalous Green’s function, and the individual terms are represented by the fermionic triangle diagrams shown in Fig. \[fig:triangles\]. All contributions to the derivative of the normal part, $\frac{{\mathrm{d}}\Pi_q^\mathrm{n}}{{\mathrm{d}}\Phi_0^2}=-\frac{1}{2}\Big(\frac{{\mathrm{d}}\mathcal{D}^{-1}_{11,q}}{{\mathrm{d}}\Phi_0^2}+\frac{{\mathrm{d}}\mathcal{D}^{-1}_{22,q}}{{\mathrm{d}}\Phi_0^2}\Big)$, have the structure presented in Fig. \[fig:triangles\](c), and the analytic expression is given by
$$\begin{aligned}
\frac{{\mathrm{d}}\Pi_q^\mathrm{n}}{{\mathrm{d}}\Phi_0^2}&=-\frac{1}{2\Phi_0}\int_k\big[G_{k+\frac{q}{2}}F_{k+\frac{q}{2}}G_{-(k-\frac{q}{2})}+G_{k+\frac{q}{2}}F_{-(k-\frac{q}{2})}G_{-(k-\frac{q}{2})}+G_{k-\frac{q}{2}}F_{k-\frac{q}{2}}G_{-(k+\frac{q}{2})}+G_{k-\frac{q}{2}}F_{-(k+\frac{q}{2})}G_{-(k+\frac{q}{2})}\big] \\
&=-\frac{{\rho_{\mathrm{F}}}}{4\Phi_0^2}\int_\Omega\bigg[\frac{\operatorname{arsinh}(r)}{r(r^2+1)^{3/2}}+\frac{1}{r^2+1}\bigg]{~ .}\end{aligned}$$
The derivation of $\Pi_q^\mathrm{a}$ w.r.t. $\Phi_0^2$ generates two different types of contributions – the first \[cf. Fig. \[fig:triangles\](a)\] comes from the exchange of fluctuations between quasiparticles while the second \[cf. Fig. \[fig:triangles\](b)\] is solely determined by anomalous propagators,
$$\begin{aligned}
\frac{{\mathrm{d}}\Pi_q^\mathrm{a}}{{\mathrm{d}}\Phi_0^2}&=-\frac{1}{2\Phi_0}\int_k\big[G_{k+\frac{q}{2}}G_{-(k+\frac{q}{2})}F_{k-\frac{q}{2}}+G_{k-\frac{q}{2}}G_{-(k-\frac{q}{2})}F_{k+\frac{q}{2}}-F_{k+\frac{q}{2}}F_{k+\frac{q}{2}}F_{k-\frac{q}{2}}-F_{k+\frac{q}{2}}F_{k-\frac{q}{2}}F_{k-\frac{q}{2}}\big]\\
&=-\frac{{\rho_{\mathrm{F}}}}{4\Phi_0^2}\int_\Omega\bigg[\frac{(2r^2+1)\operatorname{arsinh}(r)}{r(r^2+1)^{3/2}}-\frac{1}{r^2+1}\bigg]{~ .}\end{aligned}$$
The crucial contribution to $\frac{{\mathrm{d}}\Pi_q^\mathrm{a}}{{\mathrm{d}}\Phi_0^2}=-\frac{1}{2}\Big(\frac{{\mathrm{d}}\mathcal{D}^{-1}_{11,q}}{{\mathrm{d}}\Phi_0^2}-\frac{{\mathrm{d}}\mathcal{D}^{-1}_{22,q}}{{\mathrm{d}}\Phi_0^2}\Big)$ that ultimately generates the additional logarithm contributing to the gap equation stems from the term $\propto\int_\Omega \ln(2r)/r^2$ in ${\mathrm{d}}\Pi_q^\mathrm{a}/{\mathrm{d}}\Phi_0^2$ in the regime $r\gg1$. It can be related to the first term, i.e., the contribution depicted in Fig. \[fig:triangles\](a). The diagrammatic key elements used throughout the paper are introduced in Fig. \[fig:diagrammatic-elements\].
![Diagrammatic elements. The normal and anomalous propagators are represented by straight lines, and the numbers refer to the matrix structure in the band space emerging as a consequence of the doubling of the unit cell, as introduced in Eq. . Furthermore, two types of vertices appear in the diagrammatic representation of the gap equation: the coupling to the order parameter as well as to the respective fluctuations.[]{data-label="fig:diagrammatic-elements"}](triangle-diagram-vertices){width="\columnwidth"}
Effect of channel interference on density-wave instabilities {#app:two-band}
============================================================
In this appendix, we demonstrate how the presence of competing instabilities can affect the transition towards a new low-temperature ordered phase. We use the two-band model of iron-based superconductors as an example, which has been studied in great detail in the recent past [@FernandesChubukov-RepProgPhys2017]. In our brief analysis, we greatly build on the RG analysis of this model as presented in Ref. . We start with a brief summary of their results before using them to analyze the effect of channel interference on density-wave instabilities.
The model analyzed in this appendix consists of two nested Fermi pockets: one with a hole-like dispersion centered around ${\boldsymbol{0}}$, and another one with an electron-like dispersion centered around ${\boldsymbol{Q}}$, which we assume to be perfectly nested. The notation used in this appendix then merely differs from the notation used in the main text in that we introduce the band index $j\in\{1,2\}$ and measure momenta ${\boldsymbol{k}}$ as deviations from ${\boldsymbol{0}}$ and ${\boldsymbol{Q}}$, respectively. The noninteracting part \[cf. Eq. \] of the Hamiltonian $\mathcal{H}=\mathcal{H}_0+\mathcal{H}_\mathrm{int}$ then reads $$\mathcal{H}_0= \sum_{j,{\boldsymbol{k}},\sigma}\varepsilon_{j,{\boldsymbol{k}}}\psi^\dagger_{j,{\boldsymbol{k}},\sigma}\psi_{j,{\boldsymbol{k}},\sigma}{~ ,}$$ and the nesting condition takes the form $\varepsilon_{1,{\boldsymbol{k}}}=-\varepsilon_{2,{\boldsymbol{k}}}$. Furthermore, the interaction translated to band notation is given by $$\begin{aligned}
\mathcal{H}_\mathrm{int}&= \frac{1}{2}\sum\psi^\dagger_{j_1,{\boldsymbol{k}}_1,\sigma_1}\psi^\dagger_{j_2,{\boldsymbol{k}}_2,\sigma_2} U^{\sigma_1\sigma_2,\sigma_3\sigma_4}_{j_1j_2,j_3j_4}({\boldsymbol{k}}_1,{\boldsymbol{k}}_2;{\boldsymbol{k}}_3,{\boldsymbol{k}}_4) \nonumber \\ &\qquad \times \psi^{}_{j_3,{\boldsymbol{k}}_3,\sigma_3}\psi^{}_{j_4,{\boldsymbol{k}}_4,\sigma_4}\delta({\boldsymbol{k}}_1+{\boldsymbol{k}}_2-{\boldsymbol{k}}_3-{\boldsymbol{k}}_4){~ ,}\label{eq:H_int}\end{aligned}$$ where the summation is over band indices $j_i$, momenta ${\boldsymbol{k}}_i$, and spins $\sigma_i$. In contrast to the discussion in the main text, we allow for a weak momentum dependence of the couplings in the sense that they still depend on band indices, i.e., on whether the momenta are close to ${\boldsymbol{0}}$ or close to ${\boldsymbol{Q}}$. This results in several coupling constants associated with the different intraband and interband processes. After decomposing the interaction into charge (ch) and spin (sp) channel according to $$\begin{aligned}
U^{\sigma_1\sigma_2,\sigma_3\sigma_4}_{j_1j_2,j_3j_4}({\boldsymbol{k}}_1{\boldsymbol{k}}_2;{\boldsymbol{k}}_3,{\boldsymbol{k}}_4)= U_{j_1j_2;j_3j_4}^\mathrm{ch}\delta_{\sigma_1\sigma_4}\delta_{\sigma_2\sigma_3} \nonumber \\ + U_{j_1j_2;j_3j_4}^\mathrm{sp}{\boldsymbol{\sigma}}_{\sigma_1\sigma_4}\cdot{\boldsymbol{\sigma}}_{\sigma_2\sigma_3}{~ ,}\end{aligned}$$ the spin sums can be partially evaluated and the resulting expression for the interaction term contains five independent interaction constants $U_i^{(0)}$, $$\begin{aligned}
\mathcal{H}^\prime_\mathrm{int}&= U_1^{(0)}\sum_{{\boldsymbol{k}}{\boldsymbol{k}}^\prime{\boldsymbol{q}}}\psi^\dagger_{1,{\boldsymbol{k}},\sigma}\psi^\dagger_{2,{\boldsymbol{k}}^\prime,\sigma^\prime}\psi^{}_{2,{\boldsymbol{k}}^\prime-{\boldsymbol{q}},\sigma^\prime}\psi^{}_{1,{\boldsymbol{k}}+{\boldsymbol{q}},\sigma} \nonumber \\
&\quad + U_2^{(0)}\sum_{{\boldsymbol{k}}{\boldsymbol{k}}^\prime{\boldsymbol{q}}}\psi^\dagger_{2,{\boldsymbol{k}},\sigma}\psi^\dagger_{1,{\boldsymbol{k}}^\prime,\sigma^\prime}\psi^{}_{2,{\boldsymbol{k}}^\prime-{\boldsymbol{q}},\sigma^\prime}\psi^{}_{1,{\boldsymbol{k}}+{\boldsymbol{q}},\sigma} \nonumber \\
&\quad + \frac{U_3^{(0)}}{2}\sum_{{\boldsymbol{k}}{\boldsymbol{k}}^\prime{\boldsymbol{q}}}\big[\psi^\dagger_{2,{\boldsymbol{k}},\sigma}\psi^\dagger_{2,{\boldsymbol{k}}^\prime,\sigma^\prime}\psi_{1,{\boldsymbol{k}}^\prime-{\boldsymbol{q}},\sigma^\prime}\psi_{1,{\boldsymbol{k}}+{\boldsymbol{q}},\sigma}+\mathrm{H.\,c.}\big] \nonumber \\
&\quad + \frac{U_4^{(0)}}{2}\sum_{{\boldsymbol{k}}{\boldsymbol{k}}^\prime{\boldsymbol{q}}}\psi^\dagger_{2,{\boldsymbol{k}},\sigma}\psi^\dagger_{2,{\boldsymbol{k}}^\prime,\sigma^\prime}\psi^{}_{2,{\boldsymbol{k}}^\prime-{\boldsymbol{q}},\sigma^\prime}\psi^{}_{2,{\boldsymbol{k}}+{\boldsymbol{q}},\sigma} \nonumber \\
&\quad + \frac{U_5^{(0)}}{2}\sum_{{\boldsymbol{k}}{\boldsymbol{k}}^\prime{\boldsymbol{q}}}\psi^\dagger_{1,{\boldsymbol{k}},\sigma}\psi^\dagger_{1,{\boldsymbol{k}}^\prime,\sigma^\prime}\psi^{}_{1,{\boldsymbol{k}}^\prime-{\boldsymbol{q}},\sigma^\prime}\psi^{}_{1,{\boldsymbol{k}}+{\boldsymbol{q}},\sigma}{~ ,}\label{eq:H_int2}\end{aligned}$$ which we labeled in accordance with Ref. . Here the index $0$ indicates that these are the bare couplings. Note that the interband pair-hopping process associated with $U_3^{(0)}$ is only allowed if $2{\boldsymbol{Q}}$ is from the reciprocal lattice, as it is the case for iron-based superconductors. $U_4^{(0)}$ and $U_5^{(0)}$ are the intraband couplings in the two bands, and as we consider the system at perfect nesting, i.e., particle-hole symmetry, it holds that $U_4^{(0)}=U_5^{(0)}$. $U_1^{(0)}$ and $U_2^{(0)}$ refer to interband processes with a momentum transfer of ${\boldsymbol{0}}$ and ${\boldsymbol{Q}}$, respectively. The couplings $U_i^{(0)}$ are connected to the couplings in spin and charge channel by
\[eq:couplings-connection\] $$\begin{aligned}
U_{11,11}^\mathrm{ch}&= \frac{U_5^{(0)}}{4} {~ ,}& U_{11,11}^\mathrm{sp}&=-\frac{U_5^{(0)}}{4} {~ ,}\\
U_{22,22}^\mathrm{ch}&= \frac{U_4^{(0)}}{4} {~ ,}& U_{22,22}^\mathrm{sp}&=-\frac{U_4^{(0)}}{4} {~ ,}\\
U_{11,22}^\mathrm{ch}&= \frac{U_3^{(0)}}{4} {~ ,}& U_{11,22}^\mathrm{sp}&=-\frac{U_3^{(0)}}{4} {~ ,}\\
U_{12,12}^\mathrm{ch}&= -\frac{U_1^{(0)}}{4}+\frac{U_2^{(0)}}{2} {~ ,}& U_{12,12}^\mathrm{sp}&=-\frac{U_1^{(0)}}{4} {~ ,}\\
U_{12,21}^\mathrm{ch}&= \frac{U_1^{(0)}}{2}-\frac{U_2^{(0)}}{4} {~ ,}& U_{12,21}^\mathrm{sp}&=-\frac{U_2^{(0)}}{4} {~ .}\end{aligned}$$
In the remainder, we will work with dimensionless quantities, and to this end, we introduce dimensionless couplings via $u_i={\rho_{\mathrm{F}}}U_i$, where ${\rho_{\mathrm{F}}}$ is the density of states at the Fermi level.
The presence of the interaction implies three different types of instabilities: towards the formation of charge-density wave (CDW) order and spin-density wave (SDW) order, both with momentum ${\boldsymbol{Q}}$, as well as a superconducting (SC) instability resulting from Cooper pairing either in the conventional $s^{++}$-wave channel or in the sign-changing $s^{+-}$ channel. The couplings $\Gamma$ in the respective channels \[cf. the couplings $V_\mathrm{ch}$ and $V_\mathrm{sp}$ as introduced in Eq. in the main text\] are given by the combinations
\[eq:couplings\] $$\begin{aligned}
\Gamma_\mathrm{rSDW}&= u_1+ u_3 {~ ,}& \Gamma_\mathrm{iSDW}&= u_1-u_3 {~ ,}\\
\Gamma_\mathrm{iCDW}&= u_1+u_3-2u_2 {~ ,}& \Gamma_\mathrm{rCDW}&= u_1-u_3-2u_2 {~ ,}\\
\Gamma_{s^{+-}}&=u_4-u_3 {~ ,}& \Gamma_{s^{++}}&= u_4+u_3 {~ .}\end{aligned}$$
Here, the labels r and i for density-wave instabilities refer to density-wave order characterized by a purely real and a purely imaginary order parameter, respectively. What is more, the interaction possesses an $\mathrm{SO}(6)$ symmetry provided that $u_2^{(0)}=0$ and $u_4^{(0)}=-u_1^{(0)}$, as discussed by Ref. . Then, three out of the six states of emerging order are degenerate in energy – namely rSDW, iCDW, and $s^{+-}$ SC for repulsive interband pair hopping ($u_3^{(0)}>0$), whereas attractive interband pair hopping ($u_3^{(0)}<0$) may result in iSDW, rCDW, and $s^{++}$ SC. Real materials only approximately exhibit this enhanced symmetry, meaning that one of the instabilities wins. In the parent compounds of iron-based superconductors, for instance, spin-density wave order is realized and hence the coupling in the rSDW channel is the one which diverges first upon successively integrating out high-energy modes in a renormalization group (RG) analysis of the model. However, since other candidates for low-energy ordered phases are close in energy, they are competing for the same electrons, implying that phase competition is important in such systems. In the remainder, we analyze the effect of channel interference on the energy scale at which the instability towards density-wave order occurs.
The RG flow of the couplings $u_i$ as functions of $t=\log\tfrac{W}{E}$, where $W$ is the bandwidth and $E$ the running energy scale, is governed by the coupled differential equations
\[eq:RG-equations\] $$\begin{aligned}
\dot{u}_1&= u_1^2+u_3^2 {~ ,}\\
\dot{u}_2&= 2u_2(u_1-u_2) {~ ,}\\
\dot{u}_3&= 2u_3(2u_1-u_2-u_4) {~ ,}\\
\dot{u}_4&= -u_3^2-u_4^2 {~ .}\end{aligned}$$
For the derivation and a detailed discussion, we refer to Ref. . Analogous results have been obtained by Refs. for a related model without the pair-hopping process. Consequently, the flow of the couplings in the density-wave channels takes the form
\[eq:channel-interference\] $$\begin{aligned}
\dot{\Gamma}_\mathrm{SDW}&= (\Gamma_\mathrm{SDW})^2\pm 2u_3(u_1-u_2-u_4){~ ,}\\
\dot{\Gamma}_\mathrm{CDW}&= (\Gamma_\mathrm{CDW})^2\mp 2u_3(u_1+u_2-u_4) {~ ,}\end{aligned}$$
where the upper sign refers to rSDW and rCDW, while the lower sign refers to iSDW and iCDW. If the second term were zero, this would result in the usual logarithmic divergence $\Gamma=\Gamma_0/(1-\Gamma_0\log\frac{W}{E})$ of the coupling in the respective channel. Hence the presence of the second term implies corrections due to channel interference, meaning that these effects are intrinsic to the model. They vanish only for $u_3^{(0)}=0$, which we cannot generically assume for the iron-based superconductors as $2{\boldsymbol{Q}}$ is a reciprocal lattice vector here. Let us further note here that the second term combines the effect of competing density-wave and superconducting instabilities, even though we cannot differentiate between the effect of different channels on a given instability within this approach.
Motivated by Eq. , we use $u_3^{(0)}$ as a measure of channel-interference strength. We may then analyze whether channel interference is beneficial or detrimental to the formation of a certain type of order by numerically solving the RG equations for different values of $u_3^{(0)}$. For notational convenience, let us concentrate on the parameter range appropriate to describe the physics of iron-based superconductors here, i.e., repulsive interband pair hopping $u_3^{(0)}>0$ leading to the competition of rSDW order with iCDW order and $s^{+-}$ SC. We note here that the same trends are found mutatis mutandis for attractive $u_3^{(0)}<0$, i.e., for competing iSDW, rCDW, and $s^{++}$ SC instabilities. In order to analyze the effect of channel interference, we compare the flow of the coupling in a given channel for fixed $\Gamma^{(0)}$ upon varying $u_3^{(0)}$, which constitutes a measure of channel interference strength. Here, the bare parameters $u_i^{(0)}$ are chosen such that the ordering in the channel under consideration is favorable within the mean-field description, that is, $\Gamma^{(0)}\neq0$, while the bare couplings in the competing channels are tuned to zero. Although the couplings in the competing channels grow with the flow and are relevant as well, the leading instability remains the same as long as the channel interference does not change which of the couplings diverges first. However, the energy scale at which the couplings diverge turns out to be affected by the bare value of the interband pair hopping $u_3^{(0)}$, which can be used as a sign indicating whether channel interference promotes or hinders the formation of order in a given channel. In Fig. \[fig:channel-interference\](a), we exemplarily show the flow of $\Gamma_\mathrm{rSDW}$ obtained from solving Eqs. for fixed bare interaction $\Gamma_\mathrm{rSDW}^{(0)}=0.5$. Upon increasing the channel interference strength via increasing $u_3^{(0)}$, the energy scale at which the interaction in the rSDW channel diverges is pushed to lower energies, that is, happens at higher transition temperatures. As a result, we find that channel interference is beneficial for the formation of SDW order. On the other hand, the energy scale at which the interaction $\Gamma_\mathrm{iCDW}$ in the iCDW channel diverges is pushed to higher energies upon increasing the channel interference strength $u_3^{(0)}$ while keeping $\Gamma_\mathrm{iCDW}^{(0)}=0.5$ fixed. Therefore, channel interference is detrimental to the formation of CDW order as the phase transition now happens at lower temperatures, as illustrated in Fig. \[fig:channel-interference\](b).
[34]{}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 [****, ()](\doibase 10.1103/PhysRev.106.162) [****, ()](\doibase 10.1103/PhysRev.108.1175) @noop [**]{} (, ) [****, ()](\doibase
10.1103/PhysRev.128.1437) [****, ()](\doibase 10.1103/PhysRev.143.245) [****, ()](http://stacks.iop.org/0305-4608/3/i=4/a=022) [****, ()](http://www.jetpletters.ac.ru/ps/1543/article_23614.shtml) [****, ()](http://jetp.ac.ru/cgi-bin/e/index/e/38/4/p830?a=list) [****, ()](\doibase
10.1103/PhysRevLett.35.120) [****, ()](\doibase
10.1103/PhysRevLett.56.2732) [****, ()](https://doi.org/10.1038/nature08008) [****, ()](\doibase 10.1103/PhysRevB.90.035139) [****, ()](\doibase 10.1103/PhysRevB.92.205123) [****, ()](\doibase
http://dx.doi.org/10.1038/nphys3416 10.1038/nphys3416) [****, ()](\doibase 10.1103/PhysRevB.43.3475) [****, ()](\doibase 10.1103/PhysRevLett.67.757) [****, ()](\doibase 10.1103/PhysRevB.45.13008) [****, ()](\doibase 10.1103/PhysRevB.70.214531) [****, ()](\doibase 10.1103/PhysRevB.87.174523) [****, ()](\doibase
10.1103/PhysRevB.90.115125) [****, ()](\doibase 10.1103/PhysRevB.97.054510) [****, ()](\doibase
10.1103/RevModPhys.66.129) [****, ()](\doibase 10.1103/PhysRevLett.17.1133) [****, ()](\doibase
10.1103/PhysRev.158.383) [****, ()](\doibase 10.1103/PhysRevB.62.6786) [****, ()](\doibase
10.1103/PhysRevB.87.054503) [ (), 10.5445/IR/1000074135](\doibase 10.5445/IR/1000074135) @noop [****, ()]{} [****, ()](\doibase 10.1103/PhysRevB.78.134512) [****, ()](\doibase 10.1209/0295-5075/88/17004) [****, ()](\doibase 10.1103/PhysRevB.83.020505) [****, ()](http://stacks.iop.org/0034-4885/80/i=1/a=014503) [****, ()](\doibase 10.1103/PhysRevB.59.12379) [****, ()](\doibase 10.1103/PhysRevB.68.125105)
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Two integrable $U(1)$-invariant peakon equations are derived from the NLS hierarchy through the tri-Hamiltonian splitting method. A Lax pair, a recursion operator, a bi-Hamiltonian formulation, and a hierarchy of symmetries and conservation laws are obtained for both peakon equations. These equations are also shown to arise as potential flows in the NLS hierarchy by applying the NLS recursion operator to flows generated by space translations and $U(1)$-phase rotations on a potential variable. Solutions for both equations are derived using a peakon ansatz combined with an oscillatory temporal phase. This yields the first known example of a peakon breather. Spatially periodic counterparts of these solutions are also obtained.'
author:
- |
Stephen C. Anco${}^1$\
\
Fatane Mobasheramini${}^{1,2}$\
\
\
\
title: |
Integrable U(1)-invariant peakon equations\
from the NLS hierarchy
---
[^1]
Introduction {#intro}
============
There has been considerable recent interest in equations that possess peaked solitary wave solutions, known as peakons. One of the first well-studied peakon equations is the Camassa-Holm (CH) equation [@CamHol; @CamHolHym] $u_t -u_{txx} + 3u u_x - 2 u_x u_{xx} - u u_{xxx}=0$, which arises from the theory of shallow water waves. This equation possesses peakon solutions, $u=c \exp(-|x-ct|)$, and multi-peakon solutions which are linear superpositions of single peakons with time-dependent amplitudes and speeds. The CH equation also possesses a large class of solutions in which smooth initial data evolves to form a cusped wave in a finite time (i.e., a gradient blow up) while the wave amplitude remains bounded. These features are shared by many other equations, all of which belong to a general family of peakon equations [@AncRec] $u_t -u_{txx} + f(u,u_x)(u-u_{xx}) + (g(u,u_x)(u-u_{xx}))_x=0$ where $f(u,u_x)$ and $g(u,u_x)$ are arbitrary non-singular functions.
The CH equation is also an integrable system [@CamHol; @FisSch; @FucFok; @Qia03]. In particular, it has a Lax pair, a bi-Hamiltonian formulation, and an infinite hierarchy of symmetries and conservation laws generated by a recursion operator. Moreover, the CH equation is related to the integrable hierarchy that contains the Korteweg-de Vries (KdV) equation $v_t + vv_x +v_{xxx}=0$, which itself is an integrable system arising from the theory of shallow water waves. Firstly, the CH equation can be obtained from a negative flow in the KdV hierarchy by a hodograph transformation [@Fuc]. Secondly, the CH equation also can be derived as a potential flow by applying the KdV recursion operator to the flow generated by $u_x$ where $v=u-u_{xx}$ relates the variables in the two equations [@ChoQu]. Thirdly, the Hamiltonian structures of the CH equation can be derived from those of the KdV equation by a tri-Hamiltonian splitting method [@OlvRos].
The KdV equation is well-known to be related to the modified KdV (mKdV) equation $v_t + v^2 v_x +v_{xxx}=0$ by a Miura transformation. This equation is also a well-known integrable system, and it is related to a peakon equation $u_t -u_{txx} + ((u^2- u_x^2)(u -u_{xx}))_x =0$ called the modified CH (mCH) equation (also known as the Fokas-Olver-Rosenau-Qiao (FORQ) equation), which is an integrable system. In particular, the mCH equation arises from the theory of surface water waves [@Fok95b] and its integrability was derived by applying the tri-Hamiltonian splitting method to the Hamiltonian structures of the mKdV equation [@OlvRos; @Fok95a; @FokOlvRos]. It can also be obtained as a potential flow by using the mKdV recursion operator. A derivation based on spectral methods is given in where the Lax pair and single peakon solutions of the mCH equation were first obtained. Other recent work on the mCH equation appears in .
The main purpose of the present paper is to derive complex, $U(1)$-invariant peakon equations from the integrable hierarchy that contains the nonlinear Schrodinger (NLS) equation. Two different peakon equations will be obtained. One is a complex analog of the mCH equation. The other is a peakon analog of the NLS equation itself. Both of these peakon equations are integrable systems, and the main aspects of their integrability will be presented: a Lax pair, a bi-Hamiltonian formulation, a recursion operator, and an infinite hierarchy of symmetries and conservation laws.
These two peakon equations were first derived as 2-component coupled systems in by Lax pair methods, without consideration of the NLS hierarchy. The derivation in the present paper shows how both of these two equations describe negative flows in the NLS hierarchy and provides a common origin for their Lax pairs. In addition, their integrability structure is obtained in a simple, systematic way by combining the tri-Hamiltonian splitting method and the AKNS zero-curvature method [@AblKauNewSeg].
In section \[method\], the derivation of the mCH equation from the mKdV hierarchy will be briefly reviewed, using the tri-Hamiltonian splitting method, which utilizes the bi-Hamiltonian structure of the mKdV hierarchy, and using the recursion operator method, which is closely connected to a zero-curvature matrix formulation of the mKdV hierarchy. Compared to the original presentations of these two methods [@Fuc; @OlvRos; @FokOlvRos] and [@ChoQu], some new aspects will be developed, including an explicit recursion formula for all of the Hamiltonians in the mCH hierarchy, and a derivation of the mCH recursion operator directly from the AKNS zero-curvature equation [@AblKauNewSeg].
In section \[splitting\], the tri-Hamiltonian splitting method will be applied to the Hamiltonian structures in the NLS hierarchy. Rather than use the standard bi-Hamiltonian structure of the NLS equation itself, whose tri-Hamiltonian splitting is known to lead to a somewhat trivial equation [@OlvRos], a third Hamiltonian structure of the NLS equation [@MarSanWan; @AncMyr] will be used instead. This Hamiltonian structure is connected with a higher flow in the NLS hierarchy, given by the Hirota equation [@Hir1973], which is a complex, $U(1)$-invariant generalization of the mKdV equation.
In section \[derivation\], the Hamiltonian operators derived from the tri-Hamiltonian splitting method will be used to construct the two $U(1)$-invariant peakon equations along with their bi-Hamiltonian formulation. These two peakon equations will also be shown to arise as potential flows in the NLS hierarchy by applying the NLS recursion operator to the flows generated by $x$-translations and $U(1)$-phase rotations on a potential variable. One of the peakon equations has an NLS-type form which does not admit a real reduction, while the other peakon equation has a complex mCH-type form whose real reduction is given by the mCH equation.
In section \[laxpair\], for each of the two $U(1)$-invariant peakon equations, their bi-Hamiltonian structure will be used to derive a hierarchy of symmetries and conservation laws, and a Lax pair will be obtained by modifying the zero-curvature equation of the mCH equation.
In section \[peakons\], solutions for the two $U(1)$-invariant peakon equations will be derived by using a peaked travelling wave expression modified by a temporal phase oscillation, $u=a \exp(\inum(\phi+\omega t)-|x-ct|)$. Specifically, the complex mCH-type equation will be shown to possess only a standard peakon with a constant phase, while in contrast the NLS-type peakon equation will be shown to possess a stationary peakon with a temporal phase oscillation given by $w=\tfrac{1}{3}a^3$. This provides the first ever example of a peakon breather. Moreover, spatially periodic counterparts of these peakons will be derived.
Finally, some concluding remarks will be made in section \[remarks\].
Derivation of the mCH peakon equation {#method}
=====================================
We start by considering the mKdV equation $$\label{mkdv-eqn}
v_t + \tfrac{3}{2}v^2 v_x +v_{xxx} =0$$ where we have chosen the scaling factor in the nonlinear term to simplify subsequent expressions. This equation has the bi-Hamiltonian structure $$\label{mkdv-biHam-eqn}
v_t=\H(\delta H/\delta v)=\E(\delta E/\delta v)$$ where $$\begin{aligned}
\H & =-D_x ,
\label{mkdv-Hop}
\\
\E & =-D_x^3-D_xvD\inv_x vD_x
\label{mkdv-Eop}\end{aligned}$$ are compatible Hamiltonian operators, and where $$\begin{aligned}
H & =\int_{-\infty}^{\infty} \tfrac{1}{8}v^4 -\tfrac{1}{2}v_x^2\,dx ,
\\
E & =\int_{-\infty}^{\infty} \tfrac{1}{2}v^2\,dx\end{aligned}$$ are the corresponding Hamiltonians.
Recall [@Olv-book], a linear operator $\D$ is a Hamiltonian operator iff its associated bracket $$\label{bracket}
\{H,E\}_\D = \int_{-\infty}^{\infty} (\delta H/\delta v)\D(\delta E/\delta v)\, dx$$ is skew and satisfies the Jacobi identity, for all functionals $H$ and $E$. Two Hamiltonian operators are called compatible if every linear combination of them is a Hamiltonian operator. Also recall, the composition of a Hamiltonian operator and the inverse of a compatible Hamiltonian operator is a recursion operator that obeys a hereditary property [@FucFok]. In particular, a linear operator $\R$ is hereditary if it satisfies $$\label{hereditaryprop}
\lieder_{\X_{\R\eta}}\R = \R(\lieder_{\X_{\eta}}\R)$$ holding for all differential functions $\eta(x,v,v_x,\ldots)$, where $\lieder_{\X_f}$ denotes the Lie derivative with respect to a vector field $\X_f= f(x,v,v_x,\ldots)\partial_v$. An equivalent formulation of the hereditary property is given by [@Olv-book] $$\label{hereditaryop}
\pr\X_{\R\eta}\R +[\R,(\R\eta){}^\prime]= \R(\pr\X_{\eta}\R+[\R,\eta{}^\prime])$$ where $f^\prime$ denotes the Frechet derivative of a differential function $f(x,v,v_x,\ldots)$ and $\pr\X_f$ denotes the prolongation of $\X_f$ to the coordinate space $J^\infty=(x,v,v_x,\ldots)$.
The compatible Hamiltonian operators – yield the hereditary recursion operator $$\label{mkdv-Rop}
\R=\E\H\inv=D_x^2+v^2 +v_xD\inv_x v .$$ This operator and both of the Hamiltonian operators are invariant under $x$-translations applied to $v$, which corresponds to the invariance of the mKdV equation under the symmetry operator $\X=-v_x\partial/\partial_v$ representing infinitesimal $x$-translations. The recursion operator can be combined with this symmetry to express the mKdV equation as a flow $$v_t=\R({-v_x})=-\tfrac{3}{2}v^2 v_x -v_{xxx} .$$ Higher flows are generated by $$\label{mkdv-hierarchy}
P^{(n)}=\R^n({-v_x}),
\quad
n=1,2,\ldots$$ corresponding to an integrable hierarchy of equations $v_t= P^{(n)}$, where the $n=1$ flow is the mKdV equation and each successive flow $n\geq 2$ inherits a bi-Hamiltonian structure $P^{(n)} = \H(\delta H^{(n)}/\delta v)=\E(\delta H^{(n-1)}/\delta v)$ which comes from Magri’s theorem, with $H^{(0)}=E$ and $H^{(1)}=H$. The gradients of these Hamiltonians are generated by $\delta H^{(n)}/\delta v = {\R^*}^n(v) = Q^{(n)}$, $n=1,2,\ldots$, in terms of the adjoint recursion operator $$\label{mkdv-adRop}
\R^*=\H\inv\E=D_x^2+v^2 -vD\inv_x v_x$$ where $Q^{(0)}=\delta H^{(0)}/\delta v=\delta E/\delta v = v$.
It is useful to observe that $P^{(0)}=-v_x$ also can be expressed as a bi-Hamiltonian flow, in a certain formal sense. Its first Hamiltonian structure is simply $P^{(0)} = \H(\delta H^{(0)}/\delta v)$, while its second Hamiltonian structure arises from the relation $\E(0) = -v_xD\inv_x(0)$ if $D\inv_x(0)$ is redefined by the addition of some non-zero constant, so that $D\inv_x(0)=c\neq 0$. This yields $P^{(0)} = c^{-1}\E(0)$, where the corresponding Hamiltonian is trivial. Then this flow has the bi-Hamiltonian structure $$\label{mkdv-rootflow}
P^{(0)} = \H(\delta H^{(0)}/\delta v) =c^{-1}\E(0) .$$
Each higher flow in the mKdV hierarchy corresponds to a higher symmetry operator $\X^{(n)}=P^{(n)}\partial/\partial_v$, $n=1,2,\ldots$, all of which represent infinitesimal symmetries that are admitted by every equation $v_t= P^{(n)}$, $n=1,2,\ldots$, in the hierarchy. Each Hamiltonian $H^{(n)}$, $n=0,1,2,\ldots$, of the flows in the hierarchy corresponds to a conservation law $\frac{d}{dt} H^{(n)}=0$, all of which hold for every equation $v_t= P^{(n)}$, $n=1,2,\ldots$, in the hierarchy.
The mKdV equation has the scaling symmetry $v\to\lambda v$, $x\to\lambda^{-1}x$, $t\to\lambda^{-3}t$. This symmetry can be used to derive a simple scaling formula that yields the Hamiltonians $H^{(n)}$ in the mKdV hierarchy, by the general scaling method [@Anc03] shown in the appendix. The formula is given by $$\label{mkdv-H-Q-expr}
H^{(n)}=\int_{-\infty}^{\infty}h^{(n)}\,dx,
\quad
h^{(n)} = \tfrac{1}{2n+1} D\inv_x(vD_xQ^{(n)})$$ where the Hamiltonian density $h^{(n)}$ can be freely changed by the addition of a total $x$-derivative. Furthermore, the gradient recursion formula $Q^{(n)}= {\R^*}^n(v)$ can be used to convert the Hamiltonian expression into a recursion formula for $h^{(n)}$ itself, as follows. First, note expression can be inverted to give $Q^{(n)}= (2n+1) D\inv_xv^{-1} D_x h^{(n)}$. Next, replacing $Q^{(n)}= {\R^*}Q^{(n-1)} =(2n-1) (D_xv^{-1} D_x h^{(n-1)}+v h^{(n-1)})$ in expression yields $(2n+1)h^{(n)} = (2n-1)D\inv_x\big(vD_x(v+D_x v^{-1}D_x) h^{(n-1)}\big)$ modulo a total $x$-derivative, and thus $$\label{mkdv-H-expr}
h^{(n)} = \tfrac{2n-1}{2n+1}D\inv_x\big(vD_xv(1+(v^{-1}D_x)^2)h^{(n-1)}\big) .$$ This provides an explicit recursion formula for all of the Hamiltonians in the mKdV hierarchy, starting from the Hamiltonian density $h^{(0)}= \tfrac{1}{2}v^2$.
Tri-Hamiltonian splitting method
--------------------------------
The basis for the method of tri-Hamiltonian splitting [@OlvRos] is the observation that if a linear operator $\D$ is Hamiltonian then so is the scaled operator $\D_{(\lambda)}$ defined by scaling $v\to \lambda v$ (and similarly scaling all $x$-derivatives of $v$) in $\D$. When $\D_{(\lambda)}$ consists of two terms with different powers of $\lambda$, each term will define a Hamiltonian operator, and the resulting two operators can be shown to be a compatible Hamiltonian pair.
Under scaling, the first mKdV Hamiltonian operator is invariant, $\H_{(\lambda)} = -D_x = \H$, while the second mKdV Hamiltonian operator becomes $\E_{(\lambda)} = -D_x^3 -\lambda^2 D_xvD\inv_x vD_x$, which yields two operators $$\label{mkdv-splitEops}
\E_1=-D_x^3,
\quad
\E_2=-D_xvD\inv_x vD_x .$$ These operators are a compatible Hamiltonian pair. Furthermore, since the first operator is invariant, all three operators and can be shown to be mutually compatible. Hence, $\H$, $\E_1$, $\E_2$ constitute a compatible Hamiltonian triple. These operators can be combined to obtain a new pair of compatible Hamiltonian operators: $$\label{mkdv-peakon-ops}
\wtil\H = \H - \E_1=-D_x + D_x^3,
\quad
\wtil\E= \E_2=-D_xvD\inv_x vD_x .$$
The following steps are used to derive a peakon equation from this Hamiltonian pair . First, observe that the operator $\wtil\H$ has the factorization $$\label{mkdv-Delaops}
\wtil\H = -D_x\Delta = -\Delta D_x,
\quad
\Delta = 1- D_x^2$$ where $\Delta=\Delta^*$ is a symmetric operator. Second, introduce a potential variable $u$ in terms of $v$, $$\label{mkdv-potential}
v = \Delta u$$ where these variables satisfy the variational relation $$\label{mkdv-varders}
\Delta\frac{\delta}{\delta v} =\frac{\delta}{\delta u} .$$ Next, consider the flow defined by $x$-translations on $v$, $P^{(0)} = -v_x$. This root flow inherits a natural bi-Hamiltonian structure with respect to the two Hamiltonian operators . The first Hamiltonian structure arises from expressing $-v_x = \wtil\H(\Delta\inv v)$ with $\Delta\inv v= \delta \wtil H/\delta v$, so then $$-v_x = \wtil\H(\delta \wtil H/\delta v)$$ with $v= \Delta(\delta \wtil H/\delta v) = \delta \wtil H/\delta u$ through the variational relation , where the Hamiltonian is given by $$\label{mkdv-peakon-H}
\wtil H = \int_{-\infty}^{\infty} \tfrac{1}{2} uv\,dx
= \int_{-\infty}^{\infty} \tfrac{1}{2}(u^2 +u_x^2)\,dx$$ using expression and integrating by parts. The second Hamiltonian structure comes directly from splitting the operator $\E=\E_1+\E_2$ in the bi-Hamiltonian equation for $P^{(0)} = -v_x$. This yields $c^{-1}\E(0)= c^{-1}\E_2(0) = -v_x$, since $\E_1(0) =0$. Hence, $$-v_x = c^{-1}\E(0)$$ holds with a trivial Hamiltonian.
Then, Magri’s theorem [@Mag] implies that the bi-Hamiltonian flow $$P^{(0)} = -v_x = \wtil\H(\delta \wtil H/\delta v) = c^{-1}\wtil\E(0)$$ defined by $x$-translations is a root flow for an integrable hierarchy of higher flows that are generated by the hereditary recursion operator $$\label{mkdv-peakon-Rop}
\wtil\R=\wtil\E\wtil\H\inv= D_x v D\inv_x v\Delta\inv .$$ Each flow in the resulting hierarchy $\wtil P^{(n)} = \wtil\R^n({-v_x})$, $n=1,2,\ldots$, yields an equation $v_t = \wtil P^{(n)}$ that has a bi-Hamiltonian structure $\wtil P^{(n)} = \wtil\H(\delta \wtil H^{(n)}/\delta v)=\wtil\E(\delta \wtil H^{(n-1)}/\delta v)$ with $\wtil H^{(0)}=\wtil H$. The gradients of the Hamiltonians $\wtil H^{(n)}$, $n=0,1,2,\ldots$, are generated by $\delta \wtil H^{(n)}/\delta v = \wtil\R^*{}^n(u) =\wtil Q^{(n)}$, $n=1,2,\ldots$, in terms of the adjoint recursion operator $$\label{mkdv-peakon-adRop}
\wtil\R^*=\wtil\H\inv\wtil\E=\Delta\inv v D\inv_x vD_x$$ where $\delta \wtil H^{(0)}/\delta v=\delta \wtil H/\delta v = u$. These Hamiltonian gradients are equivalently given by $$\label{mkdv-peakon-Q}
\delta \wtil H^{(n)}/\delta u = \wtil\K^n(v) = \Delta \wtil Q^{(n)}$$ in terms of the operator $$\label{mkdv-peakon-Kop}
\wtil\K=\Delta\wtil\R^*\Delta\inv=v D\inv_x vD_x\Delta\inv .$$ Moreover, as $\wtil\K$ is scaling homogeneous in terms of $u$, the Hamiltonians can be obtained by the general scaling method [@Anc03] shown in the appendix. This yields the formula $$\label{mkdv-peakon-H-Q-expr}
\wtil H^{(n)} = \int_{-\infty}^{\infty} \tilde h^{(n)}\,dx,
\quad
\tilde h^{(n)}= \tfrac{1}{2(n+1)} u \wtil Q^{(n)}$$ modulo boundary terms. Note this formula reproduces the Hamiltonian for $n=0$.
Finally, the $n=1$ flow $$\label{mch-flow}
\wtil P^{(1)} = \wtil\R({-v_x}) = - \tfrac{1}{2} ((u^2-u_x^2)v)_x$$ yields the mCH equation $$\label{mch-eqn}
v_t = - \tfrac{1}{2} ((u^2-u_x^2)v)_x
= \wtil\H(\delta \wtil E/\delta v)=\wtil\E(\delta \wtil H/\delta v)$$ along with its bi-Hamiltonian structure, where $\wtil H=\wtil H^{(0)}$ is the Hamiltonian and $\wtil E=\wtil H^{(1)}$ is the Hamiltonian given by $$\label{mkdv-peakon-E}
\wtil E = \int_{-\infty}^{\infty} \tfrac{1}{8}( u(u^2-u_x^2)v ) \,dx
= \int_{-\infty}^{\infty} \tfrac{1}{24}( 3u^4 +6 u^2 u_x^2 -u_x^4) \,dx$$ from formula with $n=1$, after integration by parts. Thus, we will refer to the integrable hierarchy of higher flows $$\label{mch-hierarchy}
\wtil P^{(n)} = \wtil\R^n({-v_x}),
\quad
n=1,2,\ldots$$ as the mCH hierarchy.
The flows in the mCH hierarchy correspond to symmetry operators $\X^{(n)}=\wtil P^{(n)}\partial/\partial_v$, $n=1,2,\ldots$, while the Hamiltonians $\wtil H^{(n)}$ for these flows correspond to conservation laws $\frac{d}{dt}\wtil H^{(n)}=0$, $n=0,1,2,\ldots$, all of which are admitted by each equation $v_t= \wtil P^{(n)}$, $n=1,2,\ldots$, in the hierarchy.
A useful remark is that the recursion formula for Hamiltonian gradients can be used to convert the Hamiltonian expression into a recursion formula for the Hamiltonian densities, similarly to the mKdV case. Specifically, substitution of $\wtil Q^{(n)}= \K \wtil Q^{(n-1)} =2n u^{-1} \tilde h^{(n-1)}$ into expression gives $$\label{mch-H-expr}
\tilde h^{(n)} =\tfrac{n}{n+1} uvD\inv_x\big(vD_x\Delta\inv(u^{-1}\tilde h^{(n-1)})\big)$$ modulo a total $x$-derivative. This provides an explicit recursion formula for all of the Hamiltonians in the mCH hierarchy, starting from the Hamiltonian density $\tilde h^{(0)}= \tfrac{1}{2}vu$.
All of the higher symmetries and higher Hamiltonian densities are found to be nonlocal expressions in terms of $u$, $u_x$, $v$, and $x$-derivatives of $v$. In particular, the first higher symmetry $\X^{(2)}=\wtil P^{(2)}\partial/\partial_v$ admitted by the mCH equation is given by $$\wtil P^{(2)} = \wtil\R\wtil P^{(1)}
= \tfrac{1}{2} \big(v(uw-u_xw_x - \tfrac{1}{4}(u^2-u_x^2)^2)\big)_x$$ where $w$ is another potential which is defined by $$\Delta w = v(u^2-u_x^2) .$$ Likewise, the first higher Hamiltonian density $\tilde h^{(2)}$ admitted by the mCH equation is given by $$\tilde h^{(2)} =\tfrac{2}{3} uvD\inv_x\big(vD_x\Delta\inv(u^{-1}\tilde h^{(1)})\big)
= \tfrac{1}{12}vu(uw -u_xw_x-\tfrac{1}{4}(u^2-u_x^2)^2)$$ which yields the Hamiltonian $$\begin{aligned}
\tilde H^{(2)} & = \int_{-\infty}^{\infty} \tfrac{1}{12}vu(uw -u_xw_x-\tfrac{1}{4}(u^2-u_x^2)^2)\, dx
\\
& = \int_{-\infty}^{\infty} \tfrac{1}{12}( u^3w +\tfrac{3}{2} uu_x^2w - \tfrac{1}{2} u_x^3 w_x +\tfrac{1}{4}u^2u_x^2(u^2+u_x^2) -\tfrac{1}{4} u^6 -\tfrac{1}{10}u_x^6 )\, dx
\end{aligned}$$ modulo boundary terms.
Recursion operator method
-------------------------
From the splitting of the mKdV Hamiltonian operators, the recursion operator in the mCH hierarchy has a simple relationship to the mKdV recursion operator . First, the split operators can be expressed in terms of the mCH operators by $$\H=\Delta\inv\wtil\H,
\quad
\E_1 = (1-\Delta) \H,
\quad
\E_2 =\wtil\E .$$ Next, substitution of these operator expressions into the mKdV recursion operator yields $$\label{mkdv-Rrel1}
\R=\E\H\inv= \E_1\H\inv +\E_2\H\inv$$ where $$\label{mkdv-Rrel2}
\E_1\H\inv = 1-\Delta,
\quad
\wtil\E_2\H\inv = \wtil\E_2\wtil\H\inv\Delta = \wtil\R\Delta .$$ Then, when these two equations – are combined, this gives the relation $$\label{mkdv-recursion-rel}
\R-1 =(\wtil\R-1)\Delta .$$
Since $\wtil\R$ is a hereditary recursion operator, so is $\wtil\R-1$, and therefore $$\label{mch-shifted-Rop}
\what\R = (\R-1)\Delta\inv$$ defines a hereditary recursion operator. The hereditary property of this operator can be verified to hold directly, using the hereditary property of the mKdV recursion operator $\R$, which does not require splitting of the mKdV Hamiltonian operators.
Furthermore, since $\R$ is invariant under $x$-translations, so is $\what\R$. Therefore, by a standard result in the theory of recursion operators [@Olv77; @Olv-book], $\what\R$ can be used to generate an integrable hierarchy of flows $\what P^{(n)} = \what\R^n(-v_x)$, $n=1,2,\ldots$, starting from the root flow $P^{(0)}=-v_x$ for the mKdV and mCH hierarchies. Note the integrability structure provided by the recursion operator $\what\R$ for these flows consists of a hierarchy of higher symmetries $\X^{(n)}=\what P^{(n)}\partial_v$, $n=1,2,\ldots$, that are admitted by every equation $v_t= \what P^{(n)}$, $n=1,2,\ldots$, in the hierarchy.
The first flow in the resulting hierarchy is given by a linear combination of the root flow and the first flow in the mCH hierarchy : $\hat P^{(1)} = \wtil P^{(1)} - P^{(0)} = \what\R(-v_x) = (\R-1)(-u_x)$ where $$\R(-u_x) = -u_{xxx} - \tfrac{1}{2}((u^2-u_x^2)v)_x .$$ This yields $$\label{mch-variant-flow}
\hat P^{(1)} = \wtil P^{(1)} - P^{(0)} = (\R-1)(-u_x) = v_x - \tfrac{1}{2}((u^2-u_x^2)v)_x .$$ Since all flows in that hierarchy have a bi-Hamiltonian structure given by the pair of mCH Hamiltonian operators , the flow is also bi-Hamiltonian. The corresponding bi-Hamiltonian equation $v_t=\hat P^{(1)}$ is a slight generalization of the mCH equation : $$\label{mch-variant-eqn}
v_t - v_x + \tfrac{1}{2} ((u^2-u_x^2)v)_x =0 .$$ In particular, a Galilean transformation $x\rightarrow \tilde x=x+t$, $t\rightarrow\tilde t=t$ applied to the mCH equation yields the generalized equation .
Lax pair
--------
The shifted recursion operator , which generates linear combinations of flows in the mCH hierarchy , is closely connected to a Lax pair for the mCH equation .
To explain the connection, we first consider the mKdV shifted recursion operator $\R-1$. This operator arises in the following way from the AKNS scheme [@AblKauNewSeg] using a matrix zero-curvature equation $$\label{zero-curv}
U_t-V_x +[U,V]=0$$ given by the matrices $$\begin{aligned}
U & =
\begin{pmatrix}
\lambda & \tfrac{1}{2} v
\\
-\tfrac{1}{2} v & -\lambda
\end{pmatrix},
\label{real-akns-U}
\\
V & =
\begin{pmatrix}
\lambda K & \tfrac{1}{2} \omega + \lambda P
\\
-\tfrac{1}{2} \omega + \lambda P & -\lambda K
\end{pmatrix}
\label{real-akns-V}\end{aligned}$$ where $\lambda$ is the spectral parameter. These matrices belong to the Lie algebra $sl(2,\Rnum)$, with $U$ being a reduction of the standard AKNS matrix to the case of a real spectral parameter $\lambda$ and a real variable $v$, and with $V$ being parameterized such that $\begin{pmatrix} K & 0 \\ 0 & -K \end{pmatrix}$, $\begin{pmatrix} 0 & \omega \\ -\omega & 0 \end{pmatrix}$, $\begin{pmatrix} 0 & P \\ P & 0 \end{pmatrix}$ are mutually orthogonal in the Cartan-Killing inner product (as defined by the trace of a matrix product) in $sl(2,\Rnum)$. The components of the zero-curvature equation yield
$$\begin{gathered}
v_t = D_x \omega -4\lambda^2 P ,
\label{real-akns-flow-eqn}
\\
\omega = v K + D_x P ,
\label{real-akns-grad-eqn}
\\
D_x K = v P .
\label{real-akns-aux-eqn}\end{gathered}$$
The second and third equations can be used to express $K=D\inv_x(vP)$ and $\omega = v D\inv_x(vP) + D_x P$, and then the first equation becomes $v_t = D_x^2P +D_x(vD\inv_x(vP)) -4\lambda^2 P$. If we now put $\lambda =\tfrac{1}{2}$ then this yields $$\label{mkdv-lax-pair-flow}
v_t = (\R-1)P$$ holding for an arbitrary differential function $P(v,v_x,v_{xx},\ldots)$, where $\R$ is the mKdV recursion operator . When $P=-v_x$ is given by the $x$-translation symmetry $\X=P\partial/\partial_v$ of this recursion operator, the zero-curvature equations – yield $K= -\tfrac{1}{2} v^2 + c$ and $\omega = -v_{xx} -\tfrac{1}{2}v^3 +c v$, where $c$ is an arbitrary constant given by the freedom in $D\inv_x$. The flow equation then gives $$v_t = (c+1)v_x - \tfrac{3}{2}v^2 v_x -v_{xxx}$$ which is the mKdV equation up to a convective term $(c+1)v_x$. This term vanishes if $c=-1$, whereby $K= -\tfrac{1}{2} v^2 - 1$ and $\omega = -v_{xx} -\tfrac{1}{2}v^3 -v$, so thus the matrices – become $$U = \tfrac{1}{2}
\begin{pmatrix}
1 & v \\ -v & -1
\end{pmatrix},
\quad
V = \tfrac{1}{2}
\begin{pmatrix}
-\tfrac{1}{2} v^2 - 1 & -v_{xx} -\tfrac{1}{2}v^3 -v -v_x
\\
v_{xx} +\tfrac{1}{2}v^3 +v -v_x & \tfrac{1}{2} v^2 + 1
\end{pmatrix} .$$ These matrices satisfy the zero-curvature equation whenever $v$ is a solution of the mKdV equation . We can obtain a matrix Lax pair from them by applying a gauge transformation that consists of the mKdV scaling symmetry $v\to\lambda v$, $x\to\lambda^{-1}x$, $t\to\lambda^{-3}t$, combined with the scaling $U\to\lambda U$, $V\to\lambda^3 V$. This is easily seen to produce the standard mKdV matrix Lax pair $$\label{mkdv-laxpair}
U = \tfrac{1}{2}
\begin{pmatrix}
\lambda & v \\ -v & -\lambda
\end{pmatrix},
\quad
V = \tfrac{1}{2}
\begin{pmatrix}
-\tfrac{1}{2}\lambda v^2 - \lambda^3 & -v_{xx} -\tfrac{1}{2}v^3 -\lambda^2 v -\lambda v_x
\\
v_{xx} +\tfrac{1}{2}v^3 +\lambda^2 v -\lambda v_x & \tfrac{1}{2}\lambda v^2 + \lambda^3
\end{pmatrix}$$ up to a rescaling of the spectral parameter $\lambda$.
We now use the relation between the mKdV recursion operator and the recursion operator of the mCH equation to re-write the flow equation coming from the zero-curvature equation. This gives $$\label{mkdv-lax-pair-peakon-flow}
v_t = (\wtil\R-1)\Delta P = (\wtil\R-1)\wtil P$$ with $$P = \Delta\inv\wtil P .$$ Since $\wtil\R(-v_x) = \wtil P^{(1)}$ produces the mCH flow when $D\inv_x(0)=0$, we consider $\wtil P = -v_x$ and, correspondingly, $P=-u_x$. Then the zero-curvature equations – give $K=-\tfrac{1}{2}(u^2-u_x^2)+ c$ and $\omega = -\tfrac{1}{2}(u^2-u_x^2)v -u_{xx}+ cv$, where $c=D\inv_x(0)$ is an arbitrary constant. Hence, the flow equation becomes $$v_t = (\wtil\R-1)(-v_x) = (c+1)v_x -\tfrac{1}{2}((u^2-u_x^2)v)_x$$ which is the mCH equation up to a convective term $(c+1)v_x$. We put $c=-1$ to make this term vanish, which yields $$K= -\tfrac{1}{2}(u^2-u_x^2)-1,
\quad
\omega = -\tfrac{1}{2}(u^2-u_x^2)v -u .$$ The matrices – thereby satisfy the zero-curvature equation whenever $v$ is a solution of the mCH equation . In particular, they are given by $$U = \tfrac{1}{2}
\begin{pmatrix}
1 & v \\ -v & -1
\end{pmatrix},
\quad
V = \tfrac{1}{4}
\begin{pmatrix}
u_x^2-u^2 -2 & (u_x^2-u^2)v -2(u +u_x)
\\
(u^2-u_x^2)v +2(u -u_x) & u^2-u_x^2 +2
\end{pmatrix} .$$ A Lax pair is obtained from these matrices by applying a gauge transformation defined by the mCH scaling symmetry $u\to\lambda^{-1} u$, $x\to x$, $t\to\lambda^{2}t$, combined with the scaling $U\to U$, $V\to\lambda^{-2} V$. This yields $$\begin{aligned}
U & = \tfrac{1}{2}
\begin{pmatrix}
1 & \lambda v \\ -\lambda v & -1
\end{pmatrix},
\label{mch-U}
\\
V & = \tfrac{1}{4}
\begin{pmatrix}
u_x^2 - u^2 -2\lambda^{-2}
& \lambda(u_x^2 -u^2)v -2\lambda^{-1} (u +u_x)
\\
\lambda(u^2-u_x^2)v +2\lambda^{-1} (u -u_x)
& u^2-u_x^2 + 2\lambda^{-2}
\end{pmatrix},
\label{mch-V}\end{aligned}$$ which is the standard mCH matrix Lax pair (up to rescaling of the spectral parameter $\lambda$).
Thus, we have shown how the AKNS zero-curvature equation can be used to derive the matrix Lax pair for the mCH equation. This derivation has not appeared previously in the literature.
An important final remark arises when we compare this Lax pair – to the mKdV Lax pair . In both of these Lax pairs, $U$ depends linearly on $\lambda$, while $V$ contains negative powers of $\lambda$ in the case of the mCH Lax pair but only positive powers of $\lambda$ in the case of the mKdV Lax pair. This indicates that the mCH equation can be viewed as a negative flow in the mKdV hierarchy.
Tri-Hamiltonian splitting in the NLS hierarchy {#splitting}
==============================================
The tri-Hamiltonian splitting method was originally applied to the NLS hierarchy in by considering the standard bi-Hamiltonian structure of the NLS equation. However, this did not lead to a peakon equation, because the operator $\Delta=1-D_x^2$ did not appear when the split Hamiltonian operators were recombined. We will show how to overcome this obstacle by using a third Hamiltonian structure of the NLS equation, which is connected with the first higher flow in the NLS hierarchy.
NLS hierarchy
-------------
We begin from the NLS equation $$\label{nls-eqn}
\s\inum v_t + \tfrac{1}{2}|v|^2 v +v_{xx}=0$$ which has the tri-Hamiltonian structure $$\label{nls-triHam-eqn}
v_t=\s\inum(\tfrac{1}{2}|v|^2 v +v_{xx})
= \I(\delta I/\delta\bar v)
= \H(\delta H/\delta\bar v)
=\D(\delta D/\delta\bar v)$$ where $$\begin{aligned}
\I & =\s \inum ,
\label{nls-Iop}
\\
\H & =-D_x -\inum vD\inv_x\Im\bar v ,
\label{nls-Hop}
\\
\D & =\s\inum( D_x^2 +vD\inv_x\Re\bar v D_x +\inum D_xv D\inv_x\Im\bar v)
\label{nls-Dop}\end{aligned}$$ are mutually compatible Hamiltonian operators, with $\Re$ and $\Im$ viewed as algebraic operators, and where $$\begin{aligned}
D & =\int_{-\infty}^{\infty} |v|^2\,dx ,
\\
H & =\int_{-\infty}^{\infty} \Im(\bar v v_x)\,dx ,
\\
I & =\int_{-\infty}^{\infty} \tfrac{1}{4} |v|^4 - |v_x|^2 \,dx\end{aligned}$$ are the corresponding Hamiltonians.
The two lowest order Hamiltonian operators – yield the hereditary recursion operator $$\label{nls-Rop}
\R=\H\I\inv=\s\inum(D_x+v D\inv_x \Re\bar v) .$$ This operator and all three of the Hamiltonian operators are invariant under $x$-translations applied to $v$, as well as phase rotations applied to $v$, corresponding to the invariance of the NLS equation under infinitesimal $x$-translations and infinitesimal phase rotations, which are represented by the symmetry operators $\X=-v_x\partial_v$ and $\X=\s\inum v\partial_v$. These operators yield the respective flows $$\label{nls-root-flows}
P^{(1)} = -v_x,
\quad
P^{(0)} = \s\inum v$$ which are related by the NLS recursion operator : $$\label{nls-root-flow-rel}
P^{(1)} = \R P^{(0)} .$$ The NLS equation corresponds to a higher flow $$v_t=-\R({-v_x})=-\R^2(\s\inum v)= \s\inum(\tfrac{1}{2}|v|^2 v +v_{xx})$$ where the square of the recursion operator is given by $$\label{nls-Rsqop}
\R^2=-\D\I\inv=-(D_x^2+D_xv D\inv_x \Re\bar v +\inum v D\inv_x \Im\bar v D_x) .$$ Note that the reduction of this operator under the reality condition $\bar v=v$ is given by the negative of the mKdV recursion operator . Consequently, for comparison with the mKdV hierarchy, the most natural way to generate higher flows using the NLS recursion operator is by dividing the NLS hierarchy into even and odd flows $$\begin{aligned}
P^{(2n)} &=(-\R^2)^n(\s\inum v),
\quad
n=1,2,\ldots
\label{nls-even-flows}
\\
P^{(2n+1)} &=(-\R^2)^n({-v_x}),
\quad
n=1,2,\ldots
\label{nls-odd-flows}\end{aligned}$$ which are related by $$P^{(2n+1)} = \R P^{(2n)} .$$ Then all of the odd flows will admit a real reduction that yields a corresponding flow in the mKdV hierarchy, whereas all of the even flows do not possess a real reduction.
Every flow in the NLS hierarchy – inherits a tri-Hamiltonian structure, due to Magri’s theorem. For the even flows, this structure is given by $$\label{nls-even-triHam}
P^{(2n)} = \I(\delta E^{(n)}/\delta\bar v)
= \H(\delta H^{(n-1)}/\delta\bar v)
=\D(\delta E^{(n-1)}/\delta\bar v),
\quad
n=1,2,\ldots$$ where the gradients of the Hamiltonians are generated by $$\begin{aligned}
\delta E^{(n)}/\delta\bar v & = (-\R^*{}^2)^n(v) = Q^{(2n)},
\quad
n=0,1,2,\ldots
\\
\delta H^{(n)}/\delta\bar v & = -(-\R^*{}^2)^n(\s\inum v_x) = Q^{(2n+1)}= -\R^*Q^{(2n)},
\quad
n=0,1,2,\ldots\end{aligned}$$ and where $\R^*$ is the adjoint recursion operator $$\label{nls-adRop}
\R^*=\I\inv\H=\s(\inum D_x -vD\inv_x \Im\bar v) .$$ The odd flows have a similar but enlarged Hamiltonian structure $$\label{nls-odd-triHam}
P^{(2n+1)} = -\I(\delta H^{(n)}/\delta\bar v)
= \H(\delta E^{(n)}/\delta\bar v)
=-\D(\delta H^{(n-1)}/\delta\bar v)
= \E(\delta E^{(n-1)}/\delta\bar v),
\quad
n=1,2,\ldots$$ where $$\label{nls-Eop}
\begin{aligned}
\E=-\R^2\H &
=-\big( D_x^3 +D_xvD\inv_x\Re\bar vD_x
+ \inum D_x^2vD\inv_x\Im\bar v + \inum vD\inv_x\Im\bar v D_x^2
\\&\qquad\qquad
+\tfrac{1}{2}\inum v(|v|^2 D\inv_x\Im\bar v + D\inv_x |v|^2 \Im\bar v) \big)
\end{aligned}$$ is a fourth Hamiltonian operator which is mutually compatible with $\I$, $\H$, $\D$. In particular, the first higher flow in this hierarchy is an integrable $U(1)$-invariant version of the mKdV equation given by $$\label{hirota-eqn}
v_t = \R(\s\inum(\tfrac{1}{2}|v|^2 v +v_{xx})) = -\R^2(-v_x)
= -v_{xxx}-\tfrac{3}{2}|v|^2v_x$$ which is the Hirota equation [@Hir1973]. It possesses the quad-Hamiltonian structure $$\label{hirota-quadHamil}
v_t = -v_{xxx}-\tfrac{3}{2}|v|^2v_x
= -\I(\delta H^{(1)}/\delta\bar v)
= \H(\delta E^{(1)}/\delta\bar v)
= -\D(\delta H^{(0)}/\delta\bar v)
= \E(\delta E^{(0)}/\delta\bar v) .$$ Its recursion operator is given by the NLS squared recursion operator .
In the NLS hierarchy, the real reductions of $\H$ and $\E$ match the mKdV Hamiltonian operators –, while the Hamiltonians $E^{(n)}$, $n=0,1,2,\ldots$ match the mKdV Hamiltonians , , multiplied by a factor of $2$, with the real reduction of the variational derivative $\delta/\delta\bar v = \tfrac{1}{2}(\delta/\delta\Re v +\inum \delta/\delta\Im v)$ having a compensating factor of $1/2$.
The root flow in the NLS hierarchy of even flows has the Hamiltonian structure $$\label{nls-evenroot-flow}
P^{(0)} = \s\inum v
=\I(\delta E^{(0)}/\delta\bar v)$$ while the root flow in the NLS hierarchy of odd flows has the bi-Hamiltonian structure $$\label{nls-oddroot-flow}
P^{(1)} = -v_x
=-\I(\delta H^{(0)}/\delta\bar v)
= \H(\delta E^{(0)}/\delta\bar v) .$$ Both of these root flows possess another Hamiltonian structure if, similarly to the mKdV case, $D\inv_x(0)$ is redefined by the addition of some non-zero constant, so that $D\inv_x(0)=c\neq 0$. Then the relations $\H(0)=-\inum v D\inv_x(0)$ and $\D(0)=\s(\inum v -v_x)D\inv_x(0)$ yield the Hamiltonian structures $$\label{nls-rootflows-othstruct}
-P^{(0)} = \s c^{-1}\H(0),
\quad
P^{(1)} = c^{-1}(\H(0) \s \D(0))$$ where the corresponding Hamiltonians are trivial.
Each higher flow in the combined NLS hierarchy corresponds to a higher symmetry operator $\X^{(n)}=P^{(n)}\partial_v$, $n=1,2,\ldots$, all of which represent infinitesimal symmetries that are admitted by every equation $v_t= P^{(n)}$, $n=1,2,\ldots$, in the combined hierarchy. The Hamiltonians $E^{(n)}$, $H^{(n)}$, $n=0,1,2,\ldots$, of the even and odd flows in the combined hierarchy correspond to conservation laws $\frac{d}{dt} H^{(n)}=\frac{d}{dt} E^{(n)}=0$, all of which hold for every equation $v_t= P^{(n)}$, $n=1,2,\ldots$, in the combined hierarchy. All of the higher symmetries and higher Hamiltonian densities are local expressions in terms of $v$ and its $x$-derivatives.
Similarly to the mKdV case, the general scaling method [@Anc03] shown in the appendix can be applied to derive the simple scaling formulas $$\begin{aligned}
& E^{(n)} =\int_{-\infty}^{\infty} e^{(n)}\,dx,
\quad
e^{(n)} = \tfrac{2}{2n+1}D\inv_x\Re\big(\bar vD_x(-\R^2)^nv\big),
\quad
n=0,1,2,\ldots ,
\label{nls-H-expr}
\\
& H^{(n)}=\int_{-\infty}^{\infty}h^{(n)}\,dx,
\quad
h^{(n)} = \tfrac{1}{n+1}D\inv_x\Im\big(\s\bar vD_x(-\R^2)^nv_x\big),
\quad
n=0,1,2,\ldots .
\label{nls-E-expr}\end{aligned}$$
Hamiltonian triples
-------------------
Using the preceding preliminaries, we will now proceed with splitting the second and third NLS Hamiltonian operators and . Under scaling, the second Hamiltonian operator splits into two operators $$\label{nls-splitHops}
\H_1=-D_x,
\quad
\H_2 = -\inum vD\inv_x\Im\bar v$$ which are a compatible Hamiltonian pair. Likewise, the third Hamiltonian operator splits into two operators $$\label{nls-splitDops}
\D_1=\s\inum D_x^2 ,
\quad
\D_2 = \s\inum(D\inv_x\Re\bar v D_x +\inum D_xv D\inv_x\Im\bar v)$$ which are a compatible Hamiltonian pair. The first Hamiltonian operator , which obviously cannot be split, is compatible with each of the two pairs and . Hence, this yields two different Hamiltonian triples: $$\label{nls-triple1}
\I,
\quad
\H_1,
\quad
\H_2$$ and $$\label{nls-triple2}
\I,
\quad
\D_1,
\quad
\D_2 .$$
In , the first triple was used to obtain the compatible Hamiltonian pair $\wtil\H = \I \pm \H_1=\s\inum \mp D_x$ and $\wtil\D = \H_2 = -\inum vD\inv_x\Im\bar v$, where $\wtil\H = (1 \pm \I D_x) \I$ factorizes to yield a symmetric operator $\Upsilon=1 \pm \I D_x$ which is the counterpart of the operator $\Delta=1-D_x^2$ in the mKdV case. The NLS root even flow can be expressed as a bi-Hamiltonian flow with respect to the Hamiltonian operators $\wtil\H$ and $\wtil\D$ through the introduction of a potential $u$ given by $v= \Upsilon u = u \pm \s\inum u_x$: $$P^{(0)} = \s\inum v = \wtil\H(u) = -(\s c^{-1})\wtil\D(0)$$ where $u= \wtil\H(\delta\wtil H^{(0)}/\delta\bar v)$ holds for the Hamiltonian $$\label{nls-peakon-H0}
\wtil H^{(0)} = \int_{-\infty}^{\infty} \Re(\bar v u)\, dx
= \int_{-\infty}^{\infty} (|u|^2 \pm \s\Im(\bar u_x u))\, dx .$$ Then, from Magri’s theorem, the hereditary recursion operator $\wtil\R = \wtil\D\wtil\H\inv$ produces the integrable equation $$\label{OlvRos-eqn}
v_t = \wtil\R(\s\inum v) = \pm (\s\inum \tfrac{1}{2}|u|^2 v) ,$$ which has a bi-Hamiltonian structure $v_t = \wtil\D(\delta\wtil H^{(0)}/\delta\bar v) = \wtil\H(\delta\wtil H^{(1)}/\delta\bar v)$ coming from $\wtil\R(\s\inum v) = \wtil\D(u) = \wtil\H(\Upsilon\inv(\pm\tfrac{1}{2}|u|^2 v))$ with $u= \delta\wtil H^{(0)}/\delta\bar v$ and $\Upsilon\inv(\pm\tfrac{1}{2}|u|^2 v) = \delta\wtil H^{(1)}/\delta\bar v$. The first Hamiltonian $\wtil H^{(0)}$ is given by expression , while the second Hamiltonian $\wtil H^{(1)}$ can be obtained by applying the variational relation $$\Upsilon\frac{\delta}{\delta\bar v} = \frac{\delta}{\delta\bar u}$$ to get $\delta\wtil H^{(1)}/\delta\bar u = \pm\tfrac{1}{2}|u|^2 v$, which yields $$\label{nls-peakon-H1}
\wtil H^{(1)}= \int_{-\infty}^{\infty} \pm\tfrac{1}{2}|u|^2\Re(\bar u v)\, dx
= \int_{-\infty}^{\infty} \tfrac{1}{2}|u|^2 (|u|^2+ \s\Im(\bar u_x u))\, dx .$$ However, as discussed in , this bi-Hamiltonian equation is a peakon equation, as it does not contain the operator $\Delta$. We add the remark that the NLS root odd flow also does not lead to a peakon equation. In particular, the first Hamiltonian operator $\wtil\H$ yields $$P^{(1)} = -v_x = \wtil\H(-\s\inum u_x)$$ where $-\s\inum u_x= \delta\wtil E^{(0)}/\delta\bar v$ holds for the Hamiltonian $$\label{nls-peakon-E0}
\wtil E^{(0)}= \int_{-\infty}^{\infty} \s\Im(\bar v u_x)\, dx
= \int_{-\infty}^{\infty} \s\Im(\bar u u_x)\, dx .$$ However, another Hamiltonian structure cannot be found for this flow by using the second Hamiltonian operator $\wtil\D$.
We will instead make use of the other compatible Hamiltonian pair $$\label{hirota-peakon-ops}
\wtil\H = \I -\D_1 =\s\inum(1-D_x^2),
\quad
\wtil\D = \D_2 = \s\inum(vD\inv_x\Re\bar v D_x +\inum D_xv D\inv_x\Im\bar v)$$ where the first operator has the factorization $$\wtil\H =\I\Delta = \Delta\I,
\quad
\Delta=1-D_x^2 .$$ The corresponding hereditary recursion operator is given by $$\label{hirota-peakon-Rop}
\wtil\R =\wtil\D\wtil\H\inv
= D_xv D\inv_x\Re\bar v \Delta\inv + \inum v D\inv_x\Im\bar v D_x\Delta\inv .$$ Both of the NLS root flows have a Hamiltonian structure with respect to the first Hamiltonian operator in the pair : $$\begin{aligned}
& P^{(0)} = \s\inum v = \wtil\H(\delta\wtil H^{(0)}/\delta\bar v) ,
\label{hirota-rootflow-even}
\\
& P^{(1)} = -v_x = \wtil\H(\delta\wtil E^{(0)}/\delta\bar v) ,
\label{hirota-rootflow-odd}\end{aligned}$$ with $\delta \wtil H^{(0)}/\delta\bar v = \Delta\inv v = u$ and $\delta \wtil E^{(0)}/\delta\bar v = -\s\inum\Delta\inv v_x = -\s\inum u_x$, where the potential $u$ is now given by $$\label{peakon-u}
v= \Delta u = u- u_{xx}$$ and $v$ is referred to as the momentum variable. By using the variational relation , we can formulate the previous Hamiltonian gradients as $\delta \wtil H^{(0)}/\delta\bar u =v$ and $\delta \wtil E^{(0)}/\delta\bar u = -\s\inum v_x$, from which we obtain the Hamiltonians $$\begin{aligned}
& \wtil H^{(0)} = \int_{-\infty}^{\infty} \Re(\bar u v)\, dx
= \int_{-\infty}^{\infty} (|u|^2 +|u_x|^2)\,dx ,
\\
& \wtil E^{(0)}= \int_{-\infty}^{\infty} \s\Im(\bar u v_x)\, dx
= \int_{-\infty}^{\infty} \s\Im(\bar u_x (u_{xx}-u))\,dx\end{aligned}$$ after integration by parts. The second Hamiltonian operator in the pair gives the Hamiltonian structure $$\wtil\D(0) = \s (c_1\inum v + c_2v_x) .$$ if we redefine $D\inv_x(0)$ by the addition of an arbitrary non-zero constant, similarly to the mKdV case, so that $D\inv_x(0)=c\neq 0$. Note that we have introduced two separate constants $c_1$ and $c_2$ due to the two separate $D\inv_x$ terms that occur in $\wtil\D$. In this sense, we can view each of the NLS root flows – as having a second Hamiltonian structure, $P^{(0)} = \s\inum v = c_1^{-1}\wtil\E(0)$ and $P^{(1)} = -v_x = -\s c_2^{-1}\wtil\E(0)$, with a trivial Hamiltonian.
$U(1)$-invariant peakon equations {#derivation}
=================================
We will now explicitly show that the recursion operator obtained from splitting the third Hamiltonian structure of the NLS equation generates a bi-Hamiltonian equation when it is applied to each of the root flows – in the NLS hierarchy, with $u$ defined to be the peakon potential . This will yield two $U(1)$-invariant peakon equations.
From the even root flow , we obtain the first higher flow $$\label{hirota-evenflow}
\wtil P^{(0,1)} = \wtil\R(\s\inum v)
= \s\inum\big( \tfrac{1}{2}(|u|^2-|u_x^2|)v +\inum(\Im(\bar u u_x)v)_x \big)$$ where $\wtil\R=\wtil\D\wtil\H\inv $ is the recursion operator , and where $\wtil\D$, $\wtil\H$ are the recombined Hamiltonian operators . The bi-Hamiltonian structure of this flow arises from $$\wtil\R(\s\inum v)
= \wtil\D(\Delta\inv v)
= \wtil\H(\Delta\inv( \tfrac{1}{2}(|u|^2-|u_x^2|)v +\inum(\Im(\bar u u_x)v)_x ))$$ in accordance with Magri’s theorem, by expressing $$\label{hirota-H0-gradient}
\Delta\inv v = \delta \wtil H^{(0)}/\delta\bar v$$ and $$\label{hirota-H1-gradient}
\Delta\inv( \tfrac{1}{2}(|u|^2-|u_x^2|)v +\inum(\Im(\bar u u_x)v)_x )
= \delta \wtil H^{(1)}/\delta\bar v$$ with $v=\Delta u$. These two gradients have an alternative formulation $$\begin{aligned}
\delta\wtil H^{(0)}/\delta\bar u & = v ,
\label{hirota-H0-u-gradient}
\\
\delta\wtil H^{(1)}/\delta\bar u & = \tfrac{1}{2}(|u|^2-|u_x^2|)v +\inum(\Im(\bar u u_x)v)_x
\label{hirota-H1-u-gradient} \end{aligned}$$ through the variational identity . We remark that existence of the Hamiltonians $\wtil H^{(0)}$ and $\wtil H^{(1)}$ requires that the right-hand side of each gradient expression – satisfies the Helmholtz conditions [@Olv-book], as explained in the appendix. The first gradient yields the same Hamiltonian appearing in the Hamiltonian structure of the even root flow: $$\label{hirota-peakon-H0}
\wtil H^{(0)} = \int_{-\infty}^{\infty} \Re(\bar u v)\, dx
= \int_{-\infty}^{\infty} |u|^2 +|u_x|^2\,dx .$$ For the second gradient , we can straightforwardly verify that the Helmholtz conditions hold by a direct computation, with $v=\Delta u$. Then, since this gradient is given by a homogeneous expression under scaling of $u$, we can obtain the Hamiltonian by using the scaling formula shown in the appendix. This yields $$\label{hirota-peakon-H1}
\begin{aligned}
\wtil H^{(1)} & = \int_{-\infty}^{\infty} \tfrac{1}{2}\Re\big(\bar u (\tfrac{1}{2}(|u|^2-|u_x^2|)v +\inum(\Im(\bar u u_x)v)_x)\big)\,dx
\\
& = \int_{-\infty}^{\infty} \tfrac{1}{4}(|u|^4-|u_x|^4) + \tfrac{1}{2}\Re(\bar u_x^2 uv)\,dx
\end{aligned}$$ modulo boundary terms. Hence, the flow has the bi-Hamiltonian structure $$\wtil P^{(0,1)}
= \wtil\D(\delta\wtil H^{(0)}/\delta\bar v)
= \wtil\H(\delta\wtil H^{(1)}/\delta\bar v) .$$ The corresponding flow equation $v_t = \wtil P^{(0,1)}$ is given by $$\label{hirota-even-peakon-eqn}
\s \inum v_t +\inum(\Im(\bar u u_x)v)_x +\tfrac{1}{2} (|u|^2-|u_x^2|)v =0$$ which is a NLS-type peakon equation, with the bi-Hamiltonian formulation $$\s v_t = \tfrac{1}{2} \inum(|u|^2-|u_x^2|)v -(\Im(\bar u u_x)v)_x
= \wtil\D(\delta\wtil H^{(0)}/\delta\bar v)
= \wtil\H(\delta\wtil H^{(1)}/\delta\bar v) .$$
Similarly, from the odd root flow , we obtain the first higher flow $$\label{hirota-oddflow}
\wtil P^{(1,1)} = \wtil\R(-v_x)
= \wtil\D(\s\inum u_x) = -\tfrac{1}{2}((|u|^2-|u_x^2|)v)_x -\inum\Im(\bar u u_x)v .$$ The bi-Hamiltonian structure of this flow arises from $$\label{hirota-E0-E1-gradient}
\wtil\R(-v_x)
= \wtil\D(\s\inum\Delta\inv v_x)
= \wtil\H(\Delta\inv( \s\inum\tfrac{1}{2}((|u|^2-|u_x^2|)v)_x-\Im(\bar u u_x)v ))$$ by expressing $\s\inum\Delta\inv v_x$ and $\s\inum\tfrac{1}{2}((|u|^2-|u_x^2|)v)_x-\Im(\bar u u_x)v$ in the gradient form $$\begin{aligned}
\delta\wtil E^{(0)}/\delta\bar u & = \s\inum v_x ,
\label{hirota-E0-u-gradient}
\\
\delta\wtil E^{(1)}/\delta\bar u & = \s\inum\tfrac{1}{2}((|u|^2 -|u_x^2|)v)_x -\Im(\bar u u_x)v
\label{hirota-E1-u-gradient}\end{aligned}$$ through the variational identity . As in the case of the previous flow, existence of the Hamiltonians $\wtil E^{(0)}$ and $\wtil E^{(1)}$ requires that the right-hand side of each gradient expression – satisfies the Helmholtz conditions. The first gradient yields the Hamiltonian that appears in the Hamiltonian structure of the odd root flow: $$\label{hirota-peakon-E0}
\wtil E^{(0)}= \int_{-\infty}^{\infty} \s\Im(\bar u_x v)\, dx
= \int_{-\infty}^{\infty} \s\Im(\bar u_x (u_{xx}-u))\,dx .$$ For the second gradient , we can straightforwardly verify that the Helmholtz conditions hold by a direct computation, with $v=\Delta u$. Then, since this gradient is given by a homogeneous expressions under scaling of $u$, we can obtain the Hamiltonian by using the scaling formula shown in the appendix. This yields $$\label{hirota-peakon-E1}
\begin{aligned}
\wtil E^{(1)} & = \int_{-\infty}^{\infty} \tfrac{1}{2}\Re\big(\bar u( \s\inum\tfrac{1}{2}((|u|^2 -|u_x^2|)v)_x -\Im(\bar u u_x)v )\big)\,dx
\\
& = -\int_{-\infty}^{\infty} \tfrac{3}{4}|u|^2\Im(\bar u u_x) +\tfrac{1}{4}(2|u|^2-|u_x|^2)\Im(\bar u_x u_{xx}) \,dx
\end{aligned}$$ modulo boundary terms. Hence, the flow has the bi-Hamiltonian structure $$\wtil P^{(1,1)}
= \wtil\D(\delta\wtil E^{(0)}/\delta\bar v)
= \wtil\H(\delta\wtil E^{(1)}/\delta\bar v) .$$ The corresponding flow equation $v_t = \wtil P^{(1,1)}$ is given by $$\label{hirota-odd-peakon-eqn}
v_t +\tfrac{1}{2}((|u|^2-|u_x^2|)v)_x +\inum\Im(\bar u u_x)v
=0$$ which is a Hirota-type peakon equation, with the bi-Hamiltonian formulation $$v_t= -\tfrac{1}{2}((|u|^2-|u_x^2|)v)_x -\inum\Im(\bar u u_x)v
= \wtil\D(\delta\wtil E^{(0)}/\delta\bar v)
= \wtil\H(\delta\wtil E^{(1)}/\delta\bar v) .$$
Both of these peakon equations and are invariant under phase rotations $$\label{phaserot-inv}
v\rightarrow e^{\s\inum\phi} v,
\quad
\phi=\const$$ generated by $\X=\s\inum v\partial_v$, where $u=\Delta\inv v$ transforms in the same way as $v$.
Under the reality condition $\bar v=v$, the first peakon equation becomes trivial, while the second peakon equation reduces to the mCH peakon equation .
These two peakon equations and can be formulated as $2$-component integrable systems by decomposing $u$ and $v$ into their real and imaginary parts: $$u= u_1 + \inum u_2,
\quad
v= v_1 + \inum v_2$$ with $$u_1 = \Re u,
\quad
u_2 = \Im u,
\quad
v_1 = \Re v,
\quad
v_2 = \Im v .$$ Then, the NLS-type peakon equation is equivalent to the integrable system $$\s v_1{}_t + Av_2 +(Bv_1)_x =0,
\quad
\s v_2{}_t -Av_1 +(B v_2)_x =0$$ and the Hirota-type peakon equation is equivalent to the integrable system $$v_1{}_t + (Av_1)_x -Bv_2 =0,
\quad
v_2{}_t +(Av_2)_x +B v_1 =0$$ where $$A = \tfrac{1}{2}( u_1^2+u_2^2 -(u_1{}_x)^2 -(u_2{}_x)^2 ),
\quad
B= u_1 u_2{}_x-u_2 u_1{}_x .$$ Each of these integrable systems is invariant under $SO(2)$ rotations on $(v_1,v_2)$, with $(u_1,u_2)$ transforming the same way.
Any linear combination of the $U(1)$-invariant bi-Hamiltonian peakon equations and is again a $U(1)$-invariant bi-Hamiltonian peakon equation $$\label{hirota-combined-peakon-eqn}
\begin{aligned}
v_t & = -c_1( (Av)_x +\inum Bv ) + \s \inum c_2( Av+\inum (Bv)_x )
\\
& = \wtil\D(\delta(c_1\wtil E^{(0)}+c_2\wtil H^{(0)})/\delta\bar v)
= \wtil\H(\delta(c_1\wtil E^{(1)}+c_2\wtil H^{(1)})/\delta\bar v)
\end{aligned}$$ with $$A = \tfrac{1}{2}(|u|^2 -|u_x|^2),
\quad
B= \Im(\bar u u_x)$$ where $c_1,c_2$ are arbitrary real constants. An equivalent form of this peakon equation is given by $$\begin{aligned}
v_t & = \tilde c_1((A +\s B)v)_x + \tilde c_2 \inum(A+\s B)v
\\
& = -\wtil\D\big( \delta\big( \tfrac{1}{2}(\tilde c_1+\tilde c_2)\wtil E^{(0)} + \tfrac{1}{2}(\tilde c_1-\tilde c_2)\wtil H^{(0)} \big)/\delta\bar v\big)
\\
& = -\wtil\H\big( \delta\big( \tfrac{1}{2}(\tilde c_1+\tilde c_2)\wtil E^{(1)} + \tfrac{1}{2}(\tilde c_1-\tilde c_2)\wtil H^{(1)} \big)/\delta\bar v\big)
\end{aligned}$$ where $\tilde c_1,\tilde c_2$ are arbitrary real constants.
We will conclude this derivation by pointing out a simple relationship between the recursion operator used in obtaining the $U(1)$-invariant peakon equations and , and the NLS recursion operator . First, we express the split Hamiltonian operators in terms of the recombined Hamiltonian operators by $$\D_1 = (1-\Delta) \I,
\quad
\D_2 =\wtil\D .$$ Next, we substitute these operator expressions into the NLS squared recursion operator , which is also the recursion operator of the Hirota equation . This yields $$\label{nls-Rrel1}
-\R^2=\D\I\inv=\D_1\I\inv +\D_2\I\inv$$ where $$\label{nls-Rrel2}
\D_1\I\inv = 1-\Delta,
\quad
\wtil\D_2\I\inv = \wtil\D\wtil\H\inv\Delta = \wtil\R\Delta .$$ Then, we combine these two equations – to get the relation $$\label{nls-recursion-rel}
-(\R^2+1)=(\wtil\R-1)\Delta .$$ Since $\wtil\R$ is a hereditary recursion operator, so is $\wtil\R-1$, and therefore $$\label{hirota-nls-Rop}
\what\R = -(\R^2+1)\Delta\inv$$ defines a hereditary recursion operator, where the hereditary property can be checked to hold directly, without the need for splitting the NLS Hamiltonian operators.
The NLS-type peakon equation arises from applying this recursion operator to the even root flow , giving $$\hat P^{(0,1)} = \what\R(\s\inum v) = -(\R^2+1)(\s \inum u)$$ where $$\R^2(\s\inum u) = \s\inum( - u_{xx} - \tfrac{1}{2}(u^2-u_x^2)v +\inum (\Im(\bar u_x u)v)_x ) .$$ Hence, we have $$\label{hirota-variant-evenflow}
\hat P^{(0,1)} = -(\R^2+1)(\s \inum u)
= -\s\inum v+ \s\tfrac{1}{2} \inum(|u|^2-|u_x^2|)v - \s (\Im(\bar u u_x)v)_x
= \wtil P^{(0,1)} - P^{(0)}$$ which is a linear combination of the root flow and the NLS-type peakon flow . Since both of those flows possess a bi-Hamiltonian structure given by the pair of Hamiltonian operators , the resulting flow is also bi-Hamiltonian. The corresponding bi-Hamiltonian equation $v_t=\hat P^{(0,1)}$ is a slight generalization of the NLS-type peakon equation : $$\label{hirota-variant-even-peakon-eqn}
\s \inum v_t -v +\inum (\Im(\bar u u_x)v)_x + \tfrac{1}{2} (|u|^2-|u_x^2|)v =0 .$$ In particular, a phase transformation $v\rightarrow \tilde v=e^{-\s\inum t}v$ applied to the NLS-type peakon equation yields the generalized equation .
In a similar way, the Hirota-type peakon equation arises from applying the recursion operator to the odd root flow , giving $$\hat P^{(1,1)} = \what\R(-v_x) = -(\R^2+1)(-u_x)$$ where $$\R^2(-u_x) = u_{xxx} + \tfrac{1}{2}((u^2-u_x^2)v)_x +\inum \Im(\bar u_x u)v .$$ This yields a linear combination of the root flow and the Hirota-type peakon flow , $$\label{hirota-variant-oddflow}
\hat P^{(1,1)} = -(\R^2+1)(-u_x)
= v_x -\tfrac{1}{2} ((|u|^2-|u_x^2|)v)_x +\inum\Im(\bar u u_x)v
= \wtil P^{(1,1)} - P^{(1)}$$ which is a bi-Hamiltonian flow with respect to the pair of Hamiltonian operators . The corresponding bi-Hamiltonian equation $v_t=\hat P^{(1,1)}$ is a slight generalization of the Hirota-type peakon equation : $$\label{hirota-variant-odd-peakon-eqn}
v_t -v_x + \tfrac{1}{2} ((|u|^2-|u_x^2|)v)_x -\inum\Im(\bar u u_x)v =0 .$$ In particular, a Galilean transformation $x\rightarrow \tilde x=x+t$, $t\rightarrow\tilde t=t$ applied to the Hirota-type peakon equation yields the generalized equation .
Finally, although the two root flows and are related by the NLS recursion operator , the same relation does not hold for the two $U(1)$-invariant peakon flows and , since $-u_x - \R(\s\inum u) = -\inum\Im(\bar u_x u)v\neq0$ shows that $P^{(1,1)}\neq \R\hat P^{(0,1)}$.
Integrability properties {#laxpair}
========================
We will now derive some further aspects of the integrability structure of the two $U(1)$-invariant peakon equations and , specifically, their hierarchies of symmetries and conservation laws, and their Lax pairs.
Symmetries
----------
The recursion operator which generates the $U(1)$-invariant peakon equations and by applying it to the even and odd root flows in the NLS hierarchy is invariant under $x$-translations and $U(1)$-phase rotations. Consequently, from a standard result in the theory of recursion operators [@Olv77; @Olv-book], this operator $\wtil\R$ gives rise to two hierarchies of flows $$\begin{aligned}
\wtil P^{(0,n)} & = \wtil\R^n(\s\inum v),
\quad
n=1,2,\ldots ,
\label{peakon-even-flows}
\\
\wtil P^{(1,n)} & = \wtil\R^n(-v_x),
\quad
n=1,2,\ldots ,
\label{peakon-odd-flows}\end{aligned}$$ starting from the two NLS root flows $P^{(0)}=\s\inum v$ and $P^{(1)}=-v_x$, where the $n=1$ flows correspond to the respective $U(1)$-invariant peakon equations and , as given by $$\begin{aligned}
v_t & = \wtil P^{(0,1)}
= \s\tfrac{1}{2} \inum(|u|^2-|u_x^2|)v -\s(\Im(\bar u u_x)v)_x ,
\label{peakon-even-eqn}
\\
v_t & = \wtil P^{(1,1)}
= -\tfrac{1}{2}((|u|^2-|u_x^2|)v)_x -\inum\Im(\bar u u_x)v .
\label{peakon-odd-eqn}\end{aligned}$$ Every flow in these two hierarchies is invariant under $x$-translations and $U(1)$-phase rotations. These two symmetries are respectively generated by $$\X_{\rm trans.} = -v_x\partial/\partial_v,
\quad
\X_{\rm phas.} = \inum v\partial/\partial_v .$$
In each hierarchy and , every higher flow corresponds to a higher symmetry operator $$\begin{aligned}
\X^{(0,n)}& =\wtil P^{(0,n)}\partial/\partial_v,
\quad
n=2,3,\ldots
\label{peakon-even-symms}
\\
\X^{(1,n)}& =\wtil P^{(1,n)}\partial/\partial_v,
\quad
n=2,3,\ldots
\label{peakon-odd-symms}\end{aligned}$$ representing an infinitesimal symmetry that is admitted by the respective $U(1)$-invariant peakon equations and . All of these higher symmetries are nonlocal. In particular, the first higher symmetry in each hierarchy is given by $$\begin{aligned}
\wtil P^{(0,2)} & =
\s\inum v\big( \Re(\bar u w-\bar u_x w_x) -\Im(\bar u W_x -\bar u_x W) + \tfrac{1}{2}(\Im(\bar u u_x))^2 -\tfrac{1}{4}(|u|^2-|u_x|^2) \big)
\\&\qquad
+\big(v\big( \Re(\bar u W-\bar u_x W_x) +\Im(\bar u w_x -\bar u_x w) - \Im(\bar u u_x)(|u|^2-|u_x|^2) \big)\big)_x
\end{aligned}$$ and $$\begin{aligned}
\wtil P^{(1,2)} & =
-\big(v\big( \Re(\bar u w-\bar u_x w_x) -\Im(\bar u W_x -\bar u_x W) + \tfrac{1}{2}(\Im(\bar u u_x))^2 -\tfrac{1}{4}(|u|^2-|u_x|^2) \big)\big)_x
\\&\qquad
+ \inum v\big( \Re(\bar u W-\bar u_x W_x) +\Im(\bar u w_x -\bar u_x w) - \Im(\bar u u_x)(|u|^2-|u_x|^2) \big)
\end{aligned}$$ where $w$ and $W$ are potentials defined in terms of $\Delta$ by the relations $$\label{hirota-1stord-potentials}
\Delta w = (|u|^2-|u_x|^2)v,
\quad
\Delta W = \Im(\bar u u_x)v .$$
Conservation laws
-----------------
Since the recursion operator is hereditary, Magri’s theorem indicates that both hierarchies of flows and have a bi-Hamiltonian structure $$\begin{aligned}
\wtil P^{(0,n)} &
= \wtil\D(\delta\wtil H^{(n-1)}/\delta\bar v)
= \wtil\H(\delta\wtil H^{(n)}/\delta\bar v),
\quad
n=1,2,\ldots,
\label{peakon-even-flows-biHamil}
\\
\wtil P^{(1,n)} &
= \wtil\D(\delta\wtil E^{(n-1)}/\delta\bar v)
= \wtil\H(\delta\wtil E^{(n)}/\delta\bar v),
\quad
n=1,2,\ldots .
\label{peakon-odd-flows-biHamil}\end{aligned}$$ The Hamiltonian gradients are given by $$\begin{aligned}
\delta H^{(n)}/\delta\bar v & = \R^*{}^n(u) = Q^{(0,n)},
\quad
n=0,1,2,\ldots ,
\\
\delta E^{(n)}/\delta v & = \R^*{}^n(\s\inum u_x) = Q^{(1,n)},
\quad
n=0,1,2,\ldots , \end{aligned}$$ starting from the respective gradients and , by using the adjoint recursion operator $$\label{hirota-peakon-adRop}
\wtil\R^* =\wtil\H\inv \wtil\D
= \Delta\inv(v D\inv_x\Re\bar v D_x + \inum D_x v D\inv_x\Im\bar v) .$$ These gradients also have the formulation $$\begin{aligned}
\delta H^{(n)}/\delta\bar u & = \Delta\wtil\R^*{}^n(u) = \I\inv \R^n \I (v) = \I\inv P^{(0,n)},
\quad
n=0,1,2,\ldots ,
\label{hirota-H-u-gradient}
\\
\delta E^{(n)}/\delta v & = \Delta\wtil\R^*{}^n(\s\inum u_x) = \I\inv \R^n\I(\s\inum v_x) = \I\inv P^{(1,n)},
\quad
n=0,1,2,\ldots ,
\label{hirota-E-u-gradient}\end{aligned}$$ which is obtained through the variational identity combined with the recursion operator relation $$\I\Delta\wtil\R^* = \wtil\R\I\Delta .$$ We can get a formula for the Hamiltonians by using the general scaling method [@Anc03] shown in the appendix, which relies on the gradients and being scaling-homogeneous expressions of $u$ and $x$-derivatives of $u$. This yields $$\begin{aligned}
\wtil H^{(n)} & = \int_{-\infty}^{\infty} \tilde h^{(n)}\,dx,
\quad
\tilde h^{(n)}= \tfrac{1}{n+1} \Im(\bar u P^{(0,n)}) ,
\label{hirota-peakon-H-Q-expr}
\\
\wtil E^{(n)} & = \int_{-\infty}^{\infty} \tilde e^{(n)}\,dx,
\quad
\tilde e^{(n)}= \tfrac{1}{n+1} \Im(\bar u P^{(1,n)})
\label{hirota-peakon-E-Q-expr}\end{aligned}$$ modulo boundary terms. All of the higher Hamiltonian densities are nonlocal. In particular, the first higher density in each hierarchy is given by $$\begin{aligned}
\tilde h^{(2)} & = -\tfrac{1}{3}\s \Big(
\Re(\bar u v)\big( \Re(\bar u w-\bar u_x w_x) -\Im(\bar u W_x -\bar u_x W) + \tfrac{1}{2}(\Im(\bar u u_x))^2 -\tfrac{1}{4}(|u|^2-|u_x|^2) \big)
\\&\qquad
+\Im(\bar u_x v)\big( \Re(\bar u W-\bar u_x W_x) +\Im(\bar u w_x -\bar u_x w) - \Im(\bar u u_x)(|u|^2-|u_x|^2) \big)
\Big)
\end{aligned}$$ and $$\begin{aligned}
\tilde e^{(2)} & = \tfrac{1}{3}\s \Big(
\Im(\bar u_x v)\big( \Re(\bar u w-\bar u_x w_x) -\Im(\bar u W_x -\bar u_x W) + \tfrac{1}{2}(\Im(\bar u u_x))^2 -\tfrac{1}{4}(|u|^2-|u_x|^2) \big)
\\&\qquad
-\Re(\bar u v)\big( \Re(\bar u W-\bar u_x W_x) +\Im(\bar u w_x -\bar u_x w) - \Im(\bar u u_x)(|u|^2-|u_x|^2) \big)
\Big)
\end{aligned}$$ where $w$ and $W$ are the potentials .
Lax pair
--------
The recursion operator generating the two hierarchies of $U(1)$-invariant peakon flows – can be derived similarly to the mCH recursion operator by using a matrix zero-curvature equation based on the AKNS scheme. Here the zero-curvature matrices are taken to have the form $$\begin{aligned}
U & =
\begin{pmatrix}
\lambda & \tfrac{1}{2} v
\\
-\tfrac{1}{2}\bar v & -\lambda
\end{pmatrix},
\label{akns-U}
\\
V & =
\begin{pmatrix}
\lambda K +\tfrac{1}{2}\inum J & \tfrac{1}{2} \omega + \lambda P
\\
-\tfrac{1}{2}\bar \omega + \lambda\bar P & -\lambda K -\tfrac{1}{2}\inum J
\end{pmatrix} ,
\label{akns-V}\end{aligned}$$ belonging to the Lie algebra $sl(2,\Cnum)$, where $K$ and $J$ are real functions, and where $P$ and $\omega$ are complex functions. Note that $U$ is the standard AKNS matrix, but with the spectral parameter $\lambda$ being real, and $V$ is parameterized such that $\begin{pmatrix} K & 0 \\ 0 & -K \end{pmatrix}$, $\begin{pmatrix} \inum J & 0 \\ 0 & -\inum J \end{pmatrix}$, $\begin{pmatrix} 0 & \omega \\ -\bar\omega & 0 \end{pmatrix}$, $\begin{pmatrix} 0 & P \\ \bar P & 0 \end{pmatrix}$ are mutually orthogonal in the Cartan-Killing inner product in $sl(2,\Cnum)$. (Recall, this inner product is defined by the trace of the product of a $sl(2,\Cnum)$ matrix and a hermitian conjugated $sl(2,\Cnum)$ matrix.) In this representation, the components of the zero-curvature equation yield
$$\begin{aligned}
v_t & = D_x \omega + \inum J v -4\lambda^2 P,
\label{akns-flow-eqn}
\\
\omega & = K v + D_x P ,
\label{akns-grad-eqn}
\\
D_x K & = \Re(\bar v P) ,
\label{akns-aux-eqn1}
\\
D_x J & = \Im(\bar v\omega) .
\label{akns-aux-eqn2}\end{aligned}$$
Equations – can be used to express $K=D\inv_x\Re(\bar v P)$ and $\omega = v D\inv_x\Re(\bar v P) + D_x P$, and equation then gives $J = D\inv_x\Im(\bar v\omega) = D\inv_x\Im(\bar v D_x P)$, which yields $v_t = D_x^2P +D_x(vD\inv_x\Re(\bar vP)) + \inum v D\inv_x\Im(\bar v D_x P) -4\lambda^2 P$ from equation . If we now put $\lambda =\tfrac{1}{2}$ then we obtain $$\label{hirota-lax-pair-flow}
v_t = -(\R^2+1)P$$ holding for an arbitrary differential function $P(v,v_x,v_{xx},\ldots)$, where $\R$ is the NLS recursion operator and $-\R^2$ is the recursion operator that generates the even flows and the odd flows in the NLS hierarchy. This shows that, in the matrices –, $P$ can be identified with the generator of any flow in the NLS hierarchy.
As a first step to obtain a Lax pair in a systematic way, we will show how to use these matrices – to derive a Lax pair for the root flows in each of the two hierarchies and .
In particular, for $P=\inum v$, which corresponds to the even root flow in the NLS hierarchy, the zero-curvature equations – yield $K= c_1$, $\omega = \inum v_x +c_1 v$, $J= \tfrac{1}{2}|v|^2 +c_2$, where $c_1,c_2$ are arbitrary constants given by the freedom in $D\inv_x$. The flow equation then produces the NLS equation if we choose $c_1=0$ and $c_2=-1$. Consequently, substitution of $$K=0,
\quad
\omega = \inum v_x ,
\quad
J= \tfrac{1}{2}|v|^2 -1$$ into these matrices –, followed by a gauge transformation consisting of the NLS scaling symmetry $v\to\lambda v$, $x\to\lambda^{-1}x$, $t\to\lambda^{-2}t$ combined with the scaling $U\to\lambda U$, $V\to\lambda^2 V$, can be seen to give the standard NLS matrix Lax pair $$\label{nls-laxpair}
U = \tfrac{1}{2}
\begin{pmatrix}
\lambda & v \\ -\bar v & -\lambda
\end{pmatrix},
\quad
V = \tfrac{1}{2} \inum
\begin{pmatrix}
\tfrac{1}{2}|v|^2 +\lambda^2 & v_x +\lambda v
\\
\bar v_x -\lambda \bar v & - \tfrac{1}{2}|v|^2 -\lambda^2
\end{pmatrix}$$ up to rescaling the spectral parameter. Likewise, for $P=-v_x$, which corresponds to the odd root flow in the NLS hierarchy, the zero-curvature equations – yield $K= -\tfrac{1}{2} |v|^2 + c_1$, $\omega = -v_{xx} -\tfrac{1}{2}|v|^2v +c_1 v$, $J=\Im(\bar v_x v) +c_2$. If we now choose $c_1=-1$ and $c_2=0$, then the flow equation produces the Hirota equation . Substitution of $$K= -\tfrac{1}{2} |v|^2 - 1,
\quad
\omega = -v_{xx} -\tfrac{1}{2}|v|^2v -v,
\quad
J = \Im(\bar v_x v)$$ into the matrices –, followed by a gauge transformation $v\to\lambda v$, $x\to\lambda^{-1}x$, $t\to\lambda^{-3}t$, $U\to\lambda U$, $V\to\lambda^3 V$, thereby gives a matrix Lax pair for the Hirota equation $$\label{hirota-laxpair}
U = \tfrac{1}{2}
\begin{pmatrix}
\lambda & v \\ -\bar v & -\lambda
\end{pmatrix},
\quad
V = \tfrac{1}{2}
\begin{pmatrix}
\inum\Im(\bar v_x v) -\tfrac{1}{2}\lambda |v|^2 - \lambda^3
& -v_{xx} -\tfrac{1}{2}|v|^2v -\lambda^2 v -\lambda v_x
\\
\bar v_{xx} +\tfrac{1}{2}|v|^2\bar v +\lambda^2\bar v -\lambda\bar v_x
& -\inum\Im(\bar v_x v) +\tfrac{1}{2}\lambda |v|^2 +\lambda^3
\end{pmatrix}$$ up to rescaling the spectral parameter.
We will next adapt the previous steps to both the NLS-type peakon equation and the Hirota-type peakon equation . The main idea is to use the NLS recursion operator relation to express the zero-curvature flow equation in terms of the recursion operator $\wtil\R$ that generates the two hierarchies of $U(1)$-invariant peakon flows –. This yields $$\label{hirota-lax-pair-peakon-flow}
v_t = (\wtil\R-1)\Delta P = (\wtil\R-1)\wtil P$$ with $$P = \Delta\inv\wtil P .$$ As a result, $P$ can be chosen to be the generator of any flow in the hierarchies –. We proceed by separately considering the two flows $$\wtil P^{(0,0)} = \inum v,
\quad
\wtil P^{(1,0)} = -v_x$$ which are the respective ($n=0$) root flows in these two hierarchies. The corresponding flows on the potential $u$ are given by $$P^{(0,0)}= \inum u,
\quad
P^{(1,0)}=-u_x .$$
For the first flow $P=P^{(0,0)} = \inum u$, the zero-curvature equations – give $K=\Im(\bar u_x u)+ c_1$, $\omega = \Im(\bar u_x u)v+ c_1v +\inum u_x$, $J= \tfrac{1}{2}(|u|^2-|u_x|^2)+ c_2$. Hence, the flow equation becomes $$v_t = (\wtil\R-1)(\inum v) = \inum(c_2-1)v +c_1v_x +\tfrac{1}{2}\inum (|u|^2-|u_x|^2)v + (\Im(\bar u_x u)v)_x$$ which is the NLS-type peakon equation if we put $c_2=1$ and $c_1=0$. Then, by substituting $$K=\Im(\bar u_x u),
\quad
\omega = \Im(\bar u_x u)v +\inum u_x,
\quad
J= \tfrac{1}{2}(|u|^2-|u_x|^2)+ 1$$ into the matrices –, and applying a gauge transformation given by $$\label{gaugetrans}
u\to\lambda^{-1} u,
\quad
x\to x,
\quad
t\to\lambda^2 t,
\quad
U\to U,
\quad
V\to\lambda^{-2} V,$$ we obtain the Lax pair $$\begin{aligned}
U & = \tfrac{1}{2}
\begin{pmatrix}
1 & \lambda v \\ -\lambda\bar v & -1
\end{pmatrix},
\label{nls-peakon-U}
\\
V & = \tfrac{1}{4}
\begin{pmatrix}
\inum(|u|^2 -|u_x|^2) +2\Im(\bar u_x u) +2\inum\lambda^{-2}\
& 2\lambda\Im(\bar u_x u)v +2\inum\lambda^{-1}(u +u_x)
\\
-2\lambda\Im(\bar u_x u) \bar v -2\inum\lambda^{-1}(\bar u -\bar u_x)\
& \inum(|u_x|^2 -|u|^2) -2\Im(\bar u_x u) -2\inum\lambda^{-2}
\end{pmatrix} .
\label{nls-peakon-V}\end{aligned}$$
For the second flow $P= P^{(1,0)} = -u_x$, the zero-curvature equations – give $K=-\tfrac{1}{2}(|u|^2-|u_x|^2)+ c_1$, $\omega = -\tfrac{1}{2}(|u|^2-|u_x|^2)v+ c_1v -u_{xx}$, $J= \Im(\bar u_x u)+ c_2$. Hence, the flow equation becomes $$v_t = (\wtil\R-1)(\inum v) = \inum c_2v +(c_1+1)v_x -\tfrac{1}{2}((|u|^2-|u_x|^2)v)_x +\inum \Im(\bar u_x u)v$$ which is the Hirota-type peakon equation if we put $c_1=-1$ and $c_2=0$. Then, by substituting $$K=-\tfrac{1}{2}(|u|^2-|u_x|^2) -1,
\quad
\omega = -\tfrac{1}{2}(|u|^2-|u_x|^2)v - u,
\quad
J= \Im(\bar u_x u)$$ into the matrices –, and applying the gauge transformation , we obtain the Lax pair $$\begin{aligned}
U & = \tfrac{1}{2}
\begin{pmatrix}
1 & \lambda v \\ -\lambda\bar v & -1
\end{pmatrix} ,
\label{hirota-peakon-U}
\\
V & = \tfrac{1}{4}
\begin{pmatrix}
|u_x|^2 -|u|^2 +2\inum \Im(\bar u_x u) -2\lambda^{-2}\
& \lambda(|u_x|^2-|u|^2)v -2\lambda^{-1}(u +u_x)
\\
\lambda(|u|^2 -|u_x|^2)\bar v +2\lambda^{-1}(\bar u -\bar u_x)
& |u|^2 -|u_x|^2 -2\inum \Im(\bar u_x u) +2\lambda^{-2}\
\end{pmatrix} .
\label{hirota-peakon-V}\end{aligned}$$
Comparing the NLS Lax pair to these two Lax pairs – and –, we see that $U$ depends linearly on $\lambda$ while $V$ contains negative powers of $\lambda$ in the case of the peakon equations but only positive powers of $\lambda$ in the case of the NLS equation. This indicates that the two peakon equations can be viewed as negative flows in the NLS hierarchy.
Peakon solutions {#peakons}
================
A peakon $u(t,x)$ is a peaked travelling wave $$\label{peakon}
u = a \exp(-|x-c t|),
\quad
a,c=\const$$ where the amplitude $a$ and the speed $c$ are related by some algebraic equation. This expression is motivated by the form of the kernel of the operator $\Delta = 1- \partial_x^2$. The corresponding momentum variable is a distribution $$v = u-u_{xx} = 2a \delta(x-ct) .$$
Peakons are not classical solutions and instead are commonly understood as weak solutions [@ConStr; @ConMol2000; @ConMol2001; @GuiLiuTian; @Hak; @GuiLiuOlvQu] in the setting of an integral (weak) formulation of a given peakon equation. A weak formulation is defined by multiplying the peakon equation by test function $\psi(t,x)$ and integrating by parts to remove all terms involving $v$ and derivatives of $v$, leaving at most $u$, $u_x$, and $u_t$ in the integral. The weak formulation of the mCH equation is given by $$\label{weak-CH}
0=\iint_{\Rnum^2} \Big(
(\psi-\psi_{xx})u_t + (3\psi -\psi_{xx})u u_x
-\tfrac{1}{2} \psi_x u_x^2
\Big) dx\; dt .$$ Its well-known peakon solution [@Qia06; @GuiLiuOlvQu] has the amplitude-speed relation $a=c$: $$\label{mCH-1peakon}
u = c \exp(-|x-c t|),
\quad
c=\const$$
For both the NLS-type peakon equation and the Hirota-type peakon equation , their $U(1)$-invariance allows for the possibility of oscillatory peakon solutions whose form is given by a peaked travelling wave $a\exp(-|x-c t|)$ modulated by an oscillatory plane-wave phase $\exp(\inum (\phi+\w t+\k x))$.
We will begingby considering smooth plane-wave solutions $$\label{planewave}
u = a\exp(\inum(\k x+\w t)),
\quad
a,\k,\w=\const$$ Substitution of this expression into the Hirota-type peakon equation yields $$\w= \tfrac{1}{2}a^2 \k(\k^2-3)$$ which represents a nonlinear dispersion relation for the plane-wave. Similarly, the NLS-type peakon equation yields $$\w= \tfrac{1}{2}a^2 (1-3\k^2)$$ which is a different nonlinear dispersion relation. Note the speeds $c=-\w/\k$ of the resulting plane-waves are respectively given by $$\label{hirota-planewave-rel}
c= \tfrac{1}{2}a^2 (3-\k^2)$$ and $$\label{nls-planewave-rel}
c= \tfrac{1}{2}a^2 (3\k^2-1)/\k .$$
In the Hirota case , the plane-wave has the amplitude-speed form $$\label{hirota-planewave}
u = a\exp\big( \pm\inum \sqrt{3-2c/a^2}(x-ct) \big)$$ where $c \leq \tfrac{3}{2}a^2$. This wave can move in either direction but has a maximum speed of $c_{\max}= \tfrac{3}{2}a^2$ in the positive $x$ direction. In the NLS case , the amplitude-speed form of the plane-wave is given by $$\label{nls-planewave}
u = a\exp\big( \inum \tfrac{1}{3}(\pm\sqrt{3+c^2/a^4}+c/a^2)(x-ct) \big) .$$ This wave can move in either direction, with no restriction on its speed.
These features of the plane-wave solutions are analogous to the features of the oscillatory solitons [@AncMiaWil] of the Hirota equation and the usual solitons [@AblKauNewSeg] of the NLS equation .
We will now seek oscillatory peakon solutions $$\label{oscil-peakon}
u = a \exp(\inum (\phi+\w t+\k x))\exp(-|\xi|),
\quad
\xi= x-c t,
\quad
a,c,\phi,\w,\k=\const$$ where $\w/(2\pi)$ is a temporal oscillation frequency, $2\pi/k$ is a spatial modulation length, and $\phi$ is a phase angle. The momentum variable corresponding to expression is given by the distribution $$v = u-u_{xx}
= 2 a \exp(\inum (\phi+\w t)) \delta(\xi) + a \exp(\inum (\phi+\w t+\k x) -|\xi|) \big( \inum 2\k\, \sgn(\xi) +\k^2 \big) .$$
To proceed, we first observe that neither of the $U(1)$-invariant peakon equations and has a weak formulation that involves only $u$, $\bar u$, and their first derivatives. In particular, after multiplication by a complex-valued test function $\psi(t,x)$, the NLS-type equation contains problematic terms $\psi |u_x|^2u_{xx} = \tfrac{1}{2}\psi \bar u_x (u_x^2)_x$ and $\psi (\bar u_x uu_{xx})_x \equiv -\psi_x \bar u_x uu_{xx}$, which cannot be expressed as total $x$-derivatives (where “$\equiv$” denotes equality modulo a total $x$-derivative). Likewise the Hirota-type equation contains problematic terms $\psi (|u_x|^2u_{xx})_x \equiv -\psi_x \bar u_x u_x u_{xx}$ and $\psi \bar u_x uu_{xx}$.
But since the oscillatory peakon expression factorizes into a standard peakon amplitude expression $a \exp(-|x-c t|)$ and an oscillatory phase expression $\exp(i(\phi+\w t+\k \xi))$, we can consider a weak formulation in which a polar form $$\label{polarform}
u=A\exp(\inum\Phi),
\quad
\bar u=A\exp(-\inum\Phi)$$ is used such that any derivatives of $A$ of second and higher orders are removed. Specifically, first we substitute this polar form into the peakon equations and multiplied by the test function $\psi(t,x)$; next we use integration by parts to remove all terms involving $A_{xx}$, $A_{tx}$, and their derivatives; then we split the resulting integral equation into its real and imaginary parts. This yields what we will call a *weak-amplitude* formulation.
\[lem:weakform\] (i) The NLS-type peakon equation in polar form has the weak-amplitude formulation $$\begin{aligned}
& \begin{aligned}
0=&\iint_{\Rnum^2} \Big(
\psi_{1xx} \big( A_t+\Phi_x A^2 A_x \big)
+\psi_{1x} \big( 2 \Phi_x AA_x^2+\Phi_{xx}A^2A_x \big)
-\psi_{1} \big( A_t +\tfrac{5}{6}\Phi_x A_x^3
\\&\qquad
+\tfrac{1}{2} \Phi_{xx} AA_x^2
+\tfrac{1}{2}(5+7\Phi_x^2)\Phi_xA^2 A_x
+\tfrac{1}{2}(1+7\Phi_x^2)\Phi_{xx} A^3 \big)
\Big) dx\; dt,
\end{aligned}
\label{weak-nls-peakon-eqn-Re}
\\
& \begin{aligned}
0=&\iint_{\Rnum^2} \Big(
\psi_{2xx} A \Phi_t
-\psi_{2x}\big( \tfrac{1}{6}A_x^3 +\tfrac{1}{2}(3\Phi_x^2 -1)A^2A_x +\Phi_x\Phi_{xx}A^3 \big)
-\psi_{2}\big( \tfrac{1}{2}(3\Phi_x^2 -1)AA_x^2
\\&\qquad
+\Phi_x\Phi_{xx}A^2A_x -\tfrac{1}{2}(1+\Phi_x^2)(1-3\Phi_x^2)A^3
+\Phi_t A \big)
\Big) dx\; dt,
\end{aligned}
\label{weak-nls-peakon-eqn-Im}\end{aligned}$$ where $\psi_{1}(t,x),\psi_{2}(t,x)$ are real test functions. This pair of integral equations – is satisfied by all classical solutions of the NLS-type peakon equation . (ii) The Hirota-type equation in polar form has the weak-amplitude formulation $$\begin{aligned}
& \begin{aligned}
0=&\iint_{\Rnum^2} \Big(
\psi_{1xx}\big( \tfrac{1}{6}A_x^3 + \tfrac{1}{2}(\Phi_x^2-1)A^2A_x-A_t \big)
+\psi_{1x} \big( \tfrac{1}{2}(3\Phi_x^2-1) AA_x^2 +\Phi_x\Phi_{xx}A^2A_x \big)
\\&\qquad
-\psi_{1} \big( \tfrac{1}{2}(3\Phi_x^4-2\Phi_x^2-3)A^2A_x -A_t
+(2\Phi_x^2-1)\Phi_x\Phi_{xx} A^3 \big)
\Big) dx\; dt ,
\end{aligned}
\label{weak-hirota-peakon-eqn-Re}
\\
& \begin{aligned}
0=&\iint_{\Rnum^2} \Big(
\psi_{2xx} A \Phi_t
+\psi_{2x}\big( \tfrac{5}{6}\Phi_xA_x^3 +\tfrac{1}{2}\Phi_{xx}A A_x^2
+\tfrac{1}{2}(\Phi_x^2 -3)\Phi_xA^2A_x +\tfrac{1}{2}(\Phi_x^2-1)\Phi_{xx} A^3 \big)
\\&\qquad
+\psi_{2}\big( 2\Phi_x AA_x^2 +\Phi_{xx}A^2A_x
+(\Phi_x^2+1)(3-\Phi_x^2)\Phi_x A^3
+\Phi_t A \big)
\Big) dx\; dt .
\end{aligned}
\label{weak-hirota-peakon-eqn-Im}\end{aligned}$$ This pair of integral equations – is satisfied by all classical solutions of the Hirota-type peakon equation .
All non-smooth solutions of these integral equations will be *weak-amplitude solutions* of the corresponding peakon equations on the real line, $x\in\Rnum$.
To derive the 1-peakon solutions of the NLS-type peakon equation and the Hirota-type peakon equation , we begin by substituting expression in polar form $A=a e^{-|\xi|}$, $\Phi=\phi+\w t+\k x$ into the corresponding weak-amplitude integral equations. Next, we change the spatial integration variable from $x$ to $\xi$ and split its integration domain into $(-\infty,0)$ and $(0,\infty)$. Finally, we use integration by parts to evaluate the integrals.
Hirota peakons
--------------
The Hirota peakon weak-amplitude integrals – are respectively given by $$\begin{aligned}
0 = & \iint_{\Rnum^2} \Big(
\big( c a e^{-|\xi|} +\tfrac{1}{2}(3\k^4 -2\k^2 -3) a^3 e^{-3|\xi|} \big)\sgn(\xi)\psi_{1}(\xi+ct,t)
\\&\qquad\qquad
+\big( \tfrac{1}{2} (3\k^2-1) a^3 e^{-3|\xi|} \big)\psi_{1x}(\xi+ct,t)
\\&\qquad\qquad
+\big( \tfrac{1}{6} (2-3\k^2) a^3 e^{-3|\xi|} - c a e^{-|\xi|} \big)\sgn(\xi)\psi_{1xx}(\xi+ct,t)
\Big) d\xi\; dt
\\
= &
\iint_{\Rnum^2} \big( \tfrac{1}{2}\k^2(3\k^2 -2) a^3 e^{-3|\xi|} \big)\sgn(\xi)\psi_{1}(\xi+ct,t)\, d\xi\; dt
\\&\qquad
+a(2c+a^2 \k^2 -\tfrac{2}{3}a^2)\int_{\Rnum} \psi_{1x}(ct,t) \;dt
\end{aligned}$$ and $$\begin{aligned}
0 = & \iint_{\Rnum^2} \Big(
\big( \w a e^{-|\xi|} -\tfrac{1}{2}\k(\k^4 -2\k^2 -7) a^3 e^{-3|\xi|} \big)\psi_{2}(\xi+ct,t)
\\&\qquad\qquad
+\big( \tfrac{1}{6}\k (3\k^2 -4) a^3 e^{-3|\xi|} \big)\sgn(\xi)\psi_{2x}(\xi+ct,t)
%\\&\qquad\qquad
-\big( \w a e^{-|\xi|} \big) \psi_{2xx}(\xi+ct,t)
\Big) d\xi\; dt
\\
= &
\iint_{\Rnum^2}
\big( \tfrac{1}{2}\k(3 +5\k^2 -\k^4)a^3 e^{-3|\xi|} \big)\psi_{2}(\xi+ct,t)
d\xi\; dt
\\&\qquad
+a(2\w-3a^2\k^3 +\tfrac{4}{3}a^2\k)\int_{\Rnum} \psi_{1}(ct,t)\; dt .
\end{aligned}$$ This pair of equations must hold for all test functions $\psi_{1}$ and $\psi_{2}$, respectively. Hence, we obtain the conditions $$\begin{gathered}
a\k(3\k^2 -2) =0,
\quad
a(2c+a^2\k^2 -\tfrac{2}{3}a^2) =0,
\\
a\k(\k^4 -5\k^2 -3) =0,
\quad
a(2\w-3a^2\k^3 +\tfrac{4}{3}a^2\k) =0 . \end{gathered}$$ We want $a\neq 0$, and so then these four conditions yield $$\label{hirota-oscil-peakon-vals}
\k=0,
\quad
\w=0,
\quad
c = \tfrac{1}{3} a^2 .$$ Thus, the 1-peakon solution of the Hirota-type peakon equation is given by $$\label{hirota-oscil-peakon-soln}
u = a \exp(\inum \phi -|x-\tfrac{1}{3}a^2t|),
\quad
a,\phi=\const$$ which represents a complex-valued peaked travelling wave with a constant phase. In fact, this solution is simply the mCH peakon multiplied by the phase $e^{\inum \phi}$, and its existence is a direct consequence of the Hirota-type peakon equation being $U(1)$-invariant and reducing to the mCH equation under $u=\bar u$.
In contrast, the NLS-type peakon equation becomes trivial under $u=\bar u$, and so we expect that its peakon solution will be qualitatively different than the mCH peakon and the Hirota peakon .
NLS peakon breathers
--------------------
The NLS peakon weak-amplitude integrals – are respectively given by $$\begin{aligned}
0 = & \iint_{\Rnum^2} \Big(
\big( c a e^{-|\xi|} -\tfrac{1}{6}(21\k^2 +20) a^3 e^{-3|\xi|} \big)\sgn(\xi)\psi_{1}(\xi+ct,t)
\\&\qquad\qquad
-\big( 2\k a^3 e^{-3|\xi|} \big) \psi_{1x}(\xi+ct,t)
%\\&\qquad\qquad
+\big( \k a^3 e^{-3|\xi|} - c a e^{-|\xi|} \big)\sgn(\xi)\psi_{1xx}(\xi+ct,t)
\Big) d\xi\; dt
\\
= &
-\iint_{\Rnum^2} \big( \tfrac{1}{6}\k(21\k^2+2) a^3 e^{-3|\xi|} \big)\sgn(\xi)\psi_{1}(\xi+ct,t)\, d\xi\; dt
\\&\qquad
+2a(c-a^2\k)\int_{\Rnum} \psi_{1x}(ct,t) \;dt
\end{aligned}$$ and $$\begin{aligned}
0 = & \iint_{\Rnum^2} \Big(
\big( \w a e^{-|\xi|} +\tfrac{1}{2}(\k^2+2)(3\k^2 -1) a^3 e^{-3|\xi|} \big)\psi_{2}(\xi+ct,t)
\\&\qquad\qquad
+\big( \tfrac{1}{6}(2-9\k^2) a^3 e^{-3|\xi|} \big)\sgn(\xi)\psi_{2x}(\xi+ct,t)
%\\&\qquad\qquad
-\big( w a e^{-|\xi|} \big) \psi_{2xx}(\xi+ct,t)
\Big) d\xi\; dt
\\
= &
\iint_{\Rnum^2} \big( \tfrac{1}{2}\k^2(3\k^2-4)a^3 e^{-3|\xi|} \big)\psi_{2}(\xi+ct,t)
d\xi\; dt
\\&\qquad
+a(2\w +3a^2\k^2 -\tfrac{2}{3}a^2)\int_{\Rnum} \psi_{2}(ct,t) \;dt .
\end{aligned}$$ Since this pair of equations must hold for all test functions $\psi_{1}$ and $\psi_{2}$, we obtain the respective conditions $$\begin{gathered}
a\k(21\k^2 +2) =0,
\quad
a(c -a^2\k) =0,
\\
a \k(3\k^2-4) =0,
\quad
a(2\w +3a^2 \k^2 -\tfrac{2}{3}a^2) =0 , \end{gathered}$$ where we want $a\neq 0$. These four conditions yield $$\label{nls-oscil-peakon-vals}
c=0,
\quad
\k=0,
\quad
\w = \tfrac{1}{3} a^2 ,$$ and thus the 1-peakon solution of the NLS-type peakon equation is given by $$\label{nls-oscil-peakon-soln}
u = a \exp(\inum (\phi+\tfrac{1}{3}a^2 t) -|x|),
\quad
a,\phi=\const.$$ This solution represents a : it has a stationary peakon profile $|u|= a e^{-|x|}$ that is temporally modulated by $e^{\inum \w t}$ where the oscillation frequency is given by $\nu = \w/(2\pi) = \tfrac{1}{6\pi}a^2$ in terms of the amplitude $a$.
The existence of this novel type of peakon solution indicates that the NLS-type peakon equation can be expected to exhibit qualitatively new phenomena compared to other peakon equations.
Periodic peakons and peakon-breathers
-------------------------------------
Both of the peakon solutions and are defined on the real line, $x\in \Rnum$. We will next consider spatially periodic counterparts of these solutions, which are defined on the circle, $x\in S^1$.
We use the same weak-amplitude formulation stated in Lemma \[lem:weakform\] except that the integration domain $\Rnum^2$ for $(t,x)$ is now replaced by $\Rnum\times S^1$ and the smooth test functions $\psi_1(t,x),\psi_2(t,x)$ are now periodic in $x$. All non-smooth solutions of the resulting integral equations will be periodic weak-amplitude solutions on the circle, $x\in S^1$. Hereafter we will use the notation $\{\xi\}=\xi-\lfloor\xi\rfloor\in [0,1)$ to denote the fractional part of a variable $\xi$, where $\lfloor\xi\rfloor$ is the floor function.
Real periodic peakons are peaked continuous travelling waves on $x\in [0,1]=\Rnum/\Znum$, where the endpoints $x=0$ and $x=1$ are identified with each other. The general form for a real periodic peakon can be expressed as $$\label{periodic-peakon}
u = a (\exp(-\{|\xi|\}) + \exp(-\{|\xi-1|\}))
= a (\exp(-\{\xi\}) + \exp(\{\xi\}-1))$$ with $\xi= x-c t$, where $a$ is the amplitude and $c$ is the wave speed. An equivalent way of writing this expression is given by $$\label{periodic-peakon-cosh}
u = \tilde a \cosh(\{\xi\} -\tfrac{1}{2})$$ where $\tilde a = 2a e^{-\frac{1}{2}}$.
For both the NLS-type peakon equation and the Hirota-type peakon equation , we will seek oscillatory periodic peakon solutions $$\label{oscil-periodic-peakon}
u = \tilde a \exp(\inum (\phi+\w t+\k \xi)) \cosh(\{\xi\} -\tfrac{1}{2}),
\quad
\xi= x-c t,
\quad
a,c,\phi,\w,\k=\const$$ whose form consists of a real periodic peakon modulated by an oscillatory plane-wave phase $\exp(\inum (\phi+\w t+\k x)$. We use the same steps as in the derivation of the oscillatory peakon solutions and .
The Hirota peakon weak-amplitude integrals – formulated on $\Rnum\times S^1$ are respectively given by $$\begin{aligned}
0 = & \iint_{\Rnum\times S^1} \Big(
\big( \tfrac{1}{2}(3+2\k^2 -3\k^4) a^3 \cosh(\xi-\tfrac{1}{2})^2 -ca \big) \sinh(\xi-\tfrac{1}{2}) \psi_{1}(\xi+ct,t)
\\&\qquad\qquad
+\big( \tfrac{1}{2} (3\k^2-1) a^3 \cosh(\xi-\tfrac{1}{2}) \big) \sinh(\xi-\tfrac{1}{2})^2
\psi_{1x}(\xi+ct,t)
\\&\qquad\qquad
+\big( \tfrac{1}{6} (3\k^2-2) a^3 \cosh(\xi-\tfrac{1}{2})^2
+\tfrac{1}{6} (6c-a^2) a \big)\sinh(\xi-\tfrac{1}{2}) \psi_{1xx}(\xi+ct,t)
\Big) d\xi\; dt
\end{aligned}$$ and $$\begin{aligned}
0 = & \iint_{\Rnum\times S^1} \Big(
\big( \w a -2\k a^3 -\tfrac{1}{2}\k(\k^4 -2\k^2 -7) a^3 \cosh(\xi-\tfrac{1}{2})^2 \big)\cosh(\xi-\tfrac{1}{2}) \psi_{2}(\xi+ct,t)
\\&\qquad\qquad
+\big( \tfrac{5}{6}\k a^3 -\tfrac{1}{6}\k (3\k^2 -4) a^3 \cosh(\xi-\tfrac{1}{2})^2 \big)\sinh(\xi-\tfrac{1}{2}) \psi_{2x}(\xi+ct,t)
\\&\qquad\qquad
-\big( \w a \cosh(\xi-\tfrac{1}{2}) \big) \psi_{2xx}(\xi+ct,t)
\Big) d\xi\; dt .
\end{aligned}$$ Using integration by parts and periodicity of the test functions, we find that these two integrals yield $$\begin{aligned}
0 = & \iint_{\Rnum\times S^1}
\big( \tfrac{1}{2}\k^2 a^3 -\tfrac{1}{2}\k^2 (3\k^2 -2) a^3 \cosh(\xi-\tfrac{1}{2})^2 \big)\sinh(\xi-\tfrac{1}{2}) \psi_{1}(\xi+ct,t)\; d\xi\; dt
\\&\qquad
+a\sinh(\tfrac{1}{2})\big( \cosh(\tfrac{1}{2})^2 a^2(\k^2 -\tfrac{2}{3}) -\tfrac{1}{3}a^2 +2c \big) \int_{\Rnum} \psi_{1x}(ct,t) \;dt
\end{aligned}$$ and $$\begin{aligned}
0 = & \iint_{\Rnum\times S^1}
\big( \tfrac{1}{2}\k(3+5\k^2 -\k^4) a^3 \cosh(\xi-\tfrac{1}{2})^2 - \tfrac{1}{2}\k(2\k^2+3) a^3 \big)\cosh(\xi-\tfrac{1}{2}) \psi_{2}(\xi+ct,t) \; d\xi\; dt
\\&\qquad
-a\sinh(\tfrac{1}{2})\big( \cosh(\tfrac{1}{2})^2 a^2\k(\k^2 -\tfrac{4}{3}) -\tfrac{5}{3}a^2\k -2w\big) \int_{\Rnum} \psi_{2}(ct,t) \;dt .
\end{aligned}$$ This pair of equations must hold for all test functions $\psi_{1}$ and $\psi_{2}$, and hence we obtain the conditions $$\begin{gathered}
\k a=0 ,
\quad
\k (3\k^2 -4) a=0 ,
\quad
a\big( \cosh(\tfrac{1}{2})^2 a^2(\k^2 -\tfrac{2}{3}) -\tfrac{1}{3}a^2 +2c \big) =0,
\\
\k(\k^4 -5\k^2 -3) a =0,
\quad
\k(2\k^2+3) a =0,
\quad
a\big( \cosh(\tfrac{1}{2})^2 a^2(\k^2 -\tfrac{4}{3}) -\tfrac{5}{3}a^2\k -2w\big) =0.
\end{gathered}$$ Since we want $a\neq 0$, these six conditions yield $$\k=0,
\quad
\w=0,
\quad
c = \tfrac{1}{6}(2\cosh(\tfrac{1}{2})^2+1) a^2= \tfrac{1}{6}(2+\cosh(1)) a^2 .$$ Thus, the periodic 1-peakon solution of the Hirota-type peakon equation is given by $$\label{hirota-oscil-periodic-peakon-soln}
u = \tilde a \exp(\inum \phi) \cosh(\{ x-ct\} -\tfrac{1}{2}),
\quad
c = \tfrac{1}{6}(2+\cosh(1)) \tilde a^2 ,$$ which has the equivalent form $$u = a \exp(\inum \phi)\big(
\exp( -\{|x-\tfrac{1}{12}a^2(e^{-2}+4e^{-1}+1)t|\})
+ \exp( -\{|x-\tfrac{1}{12}a^2(e^{-2}+4e^{-1}+1)t -1|\}) \big) .$$ This solution represents a complex-valued periodic peaked travelling wave with a constant phase. In fact, the amplitude of the solution is the same as the real periodic peakon [@QuLiuLiu] of the mCH equation , which is a direct consequence of the Hirota-type peakon equation reducing to the mCH equation under $u=\bar u$.
The NLS peakon weak-amplitude integrals – formulated on $\Rnum\times S^1$ are respectively given by $$\begin{aligned}
0 = & \iint_{\Rnum\times S^1} \Big(
\big( \tfrac{1}{6}(21\k+20) a^3 \cosh(\xi-\tfrac{1}{2})^2 -\tfrac{1}{6}(5\k a^2 +6c)a \big) \sinh(\xi-\tfrac{1}{2}) \psi_{1}(\xi+ct,t)
\\&\qquad\qquad
-\big( 2\k a^3 \cosh(\xi-\tfrac{1}{2}) \sinh(\xi-\tfrac{1}{2})^2 \big)
\psi_{1x}(\xi+ct,t)
\\&\qquad\qquad
+\big( ca - \k a^3 \cosh(\xi-\tfrac{1}{2})^2 \big)\sinh(\xi-\tfrac{1}{2}) \psi_{1xx}(\xi+ct,t)
\Big) d\xi\; dt
\end{aligned}$$ and $$\begin{aligned}
0 = & \iint_{\Rnum\times S^1} \Big(
\big( \w a +\tfrac{1}{2}(2-3\k^2) a^3 +\tfrac{1}{2}(\k^2+2)(3\k^2-1) a^3 \cosh(\xi-\tfrac{1}{2})^2 \big)\cosh(\xi-\tfrac{1}{2}) \psi_{2}(\xi+ct,t)
\\&\qquad\qquad
-\big( \tfrac{1}{6} (9\k^2 -2) a^3 \cosh(\xi-\tfrac{1}{2})^2 -\tfrac{1}{6} a^3 \big)\sinh(\xi-\tfrac{1}{2}) \psi_{2x}(\xi+ct,t)
\\&\qquad\qquad
-\big( \w a \cosh(\xi-\tfrac{1}{2}) \big) \psi_{2xx}(\xi+ct,t)
\Big) d\xi\; dt .
\end{aligned}$$ Integrating by parts and using periodicity of the test functions, we find that these two integrals yield $$\begin{aligned}
0 = & \iint_{\Rnum\times S^1}
\big( \tfrac{1}{6}\k (21\k^2 +2) a^3 \cosh(\xi-\tfrac{1}{2})^2 -\tfrac{5}{6}\k a^3 \big)\sinh(\xi-\tfrac{1}{2}) \psi_{1}(\xi+ct,t)\; d\xi\; dt
\\&\qquad
+2a\sinh(\tfrac{1}{2})\big( \cosh(\tfrac{1}{2})^2 a^2\k +c \big) \int_{\Rnum} \psi_{1x}(ct,t) \;dt
\end{aligned}$$ and $$\begin{aligned}
0 = & \iint_{\Rnum\times S^1}
\big( \tfrac{1}{2}\k^2(3\k^2 -4) a^3 \cosh(\xi-\tfrac{1}{2})^2 + \tfrac{3}{2}\k^2 a^3 \big)\cosh(\xi-\tfrac{1}{2}) \psi_{2}(\xi+ct,t) \; d\xi\; dt
\\&\qquad
+a\sinh(\tfrac{1}{2})\big( \cosh(\tfrac{1}{2})^2 a^2(3\k^2 -\tfrac{2}{3}) -\tfrac{1}{3}a^2 +2w\big) \int_{\Rnum} \psi_{2}(ct,t) \;dt .
\end{aligned}$$ Since this pair of equations must hold for all test functions $\psi_{1}$ and $\psi_{2}$, we obtain the respective conditions $$\begin{gathered}
\k (21\k^2 +2) a=0,
\quad
\k a=0,
\quad
a\big( \cosh(\tfrac{1}{2})^2 a^2\k +c \big)=0,
\\
\k(3\k^2 -4) a=0,
\quad
\k a =0,
\quad
a\big( \cosh(\tfrac{1}{2})^2 a^2(3\k^2 -\tfrac{2}{3}) -\tfrac{1}{3}a^2 +2w\big) =0
\end{gathered}$$ where we want $a\neq 0$. These six conditions yield $$c=0,
\quad
\k=0,
\quad
\w = \tfrac{1}{6}(2\cosh(\tfrac{1}{2})^2+1) a^2= \tfrac{1}{6}(2+\cosh(1)) a^2$$ and thus the periodic 1-peakon solution of the NLS-type peakon equation is given by $$\label{nls-oscil-periodic-peakon-soln}
u = \tilde a \exp(\inum (\phi+\w t)) \cosh(\{x\} -\tfrac{1}{2}),
\quad
\w = \tfrac{1}{6}(2+\cosh(1)) \tilde a^2 .$$ An equivalent form is $$u = a \exp\big(\inum (\phi+\tfrac{1}{12} a^2(e^{-2}+4e^{-1}+1) t)\big)
\big( \exp(-\{x\}) + \exp(\{x\}-1) \big) .$$ This solution represents a : it has a stationary periodic peakon profile $|u|= \tilde a\cosh(\{x\}-\tfrac{1}{2})$ that is temporally modulated by $e^{\inum \w t}$.
Concluding remarks {#remarks}
==================
In this paper we have derived two integrable $U(1)$-invariant peakon equations from the NLS hierarchy. These integrable equations are associated with the first two flows in this hierarchy, which consist of the NLS equation and the Hirota equation (a $U(1)$-invariant version of mKdV equation), so consequently one equation can be viewed as an NLS-type peakon equation and the other can be viewed as a Hirota-type peakon equation (a complex analog of the mCH/FORQ equation).
For both peakon equations, we have obtained a Lax pair, a recursion operator, a bi-Hamiltonian formulation, and a hierarchy of symmetries and conservation laws. These two peakon equations have been derived previously as real 2-component coupled systems [@XiaQia; @XiaQiaZho] by Lax pair methods, without consideration of the NLS hierarchy.
We have also investigated oscillatory peakon solutions for these $U(1)$-invariant peakon equations. The Hirota peakon equation possesses only a moving non-oscillatory peakon with a constant phase. In contrast, the NLS peakon equation possesses a stationary oscillatory peakon representing a peaked breather. Breathers are familiar solutions for soliton equations but they had not been found previously for peakon equations. We have also obtained spatially periodic counterparts of these peakon solutions.
An important goal will be to find multi-peakon solutions for both the NLS-type peakon equation and the Hirota-type peakon equation . These solutions should be given by a superposition of oscillatory 1-peakons $$\label{multi-modulated-peakon}
u = \sum_{j=1}^{N} A_j \exp(\inum\Phi_j),
\quad
A_j = \alpha_j(t) \exp(-|x-\beta_j(t)|),
\quad
\Phi_j = \phi_j+\omega_j(t)$$ with time-dependent positions $\beta_j(t)$, amplitudes $\alpha_j(t)$, and phases $\omega_j(t)$. However, the weak-amplitude formulation in Lemma \[lem:weakform\] is not general enough to allow solutions of this form to be found, and the lack of a more general weak formulation implies that some type of explicit regularization of products of distributions involving $\sgn(x-\beta_j(t))$ and $\delta(x-\beta_k(t))$, $j,k=1,\ldots,N$, will have to be considered. This problem will be addressed elsewhere.
There are several directions that can be pursued for future work: (1) studying solutions with non-zero asymptotic boundary conditions and smooth soliton solutions; (2) establishing existence of weak solutions with general initial data and finding conditions for wave breaking; (3) finding a geometric realization of both peakon equations based on the well-known equivalence of the NLS equation to the vortex filament equation.
Appendix {#appendix .unnumbered}
========
Scaling formula
---------------
When a Hamiltonian structure possesses a scaling symmetry, an explicit formula can be derived to express the Hamiltonian in terms of its gradient, by using the scaling method in .
Consider a Hamiltonian evolution equation $$\label{H-eqn}
v_t = \H(\delta H/\delta v)$$ where $$\label{H-dens}
H=\int_{-\infty}^{\infty} h(x,v,v_x,\ldots)\,dx$$ is a Hamiltonian functional, and $\H$ is a Hamiltonian operator. Note the Hamiltonian gradient is given by $$\label{H-grad}
Q = \delta H/\delta v = E_v(h)$$ where $$\label{eulerop}
E_v(f) = \partial_{v} f -D_x(\partial_{v_x} f) +D_x^2(\partial_{v_{xx}} f) + \cdots$$ denotes the Euler operator [@Olv-book] with respect to $v$.
Suppose the evolution equation possesses a scaling symmetry $$\label{scaling}
x\to \lambda^a x,
\quad
t\to \lambda^b t,
\quad
v\to \lambda^c v$$ where $a,b,c=\const$ are the scaling weights. Also suppose the Hamiltonian structure is scaling homogeneous $$\label{scaling-h}
h\to \lambda^p h,
\quad
\H\to \lambda^q \H$$ where the scaling weights $p,q=\const$ are related to $a,b,c$ by $$p+q+b-2c=0$$ due to homogeneity. Then the Hamiltonian gradient has the scaling $$Q\to \lambda^{p-c} Q .$$ Note the generator of the scaling is given by the vector field $$\label{scal-generator}
\X_{\scal}= (cv -axv_x-btv_t)\partial_v$$ in evolutionary form.
The scaling homogeneity of the Hamiltonian density $h$ implies $$\pr\X_{\scal} h = p h -axD_x h - btD_t h$$ under the prolongation of the scaling generator. The explicit action of this generator is given by $$\pr\X_{\scal} h \equiv (cv -axv_x-btv_t)E_v(h) = (cv -axv_x-btv_t)Q$$ where $\equiv$ denotes equality modulo total $x$-derivatives. This yields $$\label{H-scal}
p h \equiv (cv -axv_x -btv_t)Q +axD_x h +btD_t h$$ which can be simplified by the following steps.
First, the Hamiltonian structure implies [@Olv-book] that the Hamiltonian is conserved, $\frac{d}{dt}H=0$, on the space of solutions $v(t,x)$ of the evolution equation . This conservation equation can be expressed in the form $D_t h = D_x\Phi$ for some flux $\Phi$. Consequently, in the scaling equation , the term $btD_t h= D_x(bt\Phi)$ is a total $x$-derivative.
Next, the property that a Hamiltonian operator is skew [@Olv-book] implies that $Q\H(Q) = D_x\Theta$ holds for some expression $\Theta$. Hence, on the space of solutions $v(t,x)$ of the evolution equation , the term $btv_tQ = btQ\H(Q)= bt D_x\Theta = D_x(bt \Theta)$ in the scaling equation is a total $x$-derivative.
Finally, the identity $xf = D_x(xD\inv_x f) - D\inv_x f$ can be used to express the terms $axD_x h-axv_xQ$ in the scaling equation as $ax(D_x h -v_xQ) \equiv aD\inv_x(v_xQ) -ah$.
Hence, the scaling equation simplifies to give $$(p+a) h \equiv cvQ +aD\inv_x(v_xQ)
= (c+a)vQ -aD\inv_x(vD_xQ)$$ which provides an explicit formula for obtaining the Hamiltonian density in terms of its gradient whenever $p+a\neq 0$. This condition is equivalent to $\pr\X_{\scal} H \neq 0$.
For any scaling invariant Hamiltonian evolution equation , the Hamiltonian can be recovered from its gradient by the scaling formula for the Hamiltonian density (modulo a total $x$-derivative) $$\label{H-Q-expr}
h \equiv \tfrac{1}{p+a}\big( (c+a)vQ -aD\inv_x(vD_xQ) \big)$$ provided that the Hamiltonian is not scaling invariant, $p+a\neq 0$, where $a,c,p$ are the scaling weights –.
Helmholtz conditions
--------------------
An expression $Q(x,v,v_x,\ldots)$ is a Hamiltonian gradient iff it satisfies the Helmholtz conditions [@Olv-book]. These conditions state that the Frechet derivative of $Q$ with respect to $v$ is required to be a self-adjoint operator.
Let $\delta v$ be an infinitesimal variation of $v$, namely $\X v = \delta v$ where $\X= \delta v\partial_v$ is the corresponding vector field. Then the Frechet derivative of $Q$ is given by $$\delta Q =
\delta v\partial_v Q +(D_x\delta v)\partial_{v_x} Q +(D_x^2\delta v) \partial_{v_{xx}} Q + \cdots .$$ Its adjoint is defined by integration by parts $$\delta^* Q=
\delta v \partial_v Q -D_x(\delta v\partial_{v_x}Q)) +D_x^2(\delta v\partial_{v_{xx}}Q) + \cdots .$$ Hence, the Frechet derivative is a self-adjoint operator iff $\delta Q = \delta^* Q$ holds for an arbitrary $\delta v$. Self-adjointness can be expressed in an equivalent form [@Olv-book] by using the higher Euler operators: $$\label{highereulerop}
E_v^{(l)}(Q) =
\partial_{\partial_x^l v} Q
-\txtbinom{l+1}{l}D_x\big(\partial_{\partial_x^{l+1} v} Q\big)
+ \txtbinom{l+2}{l}D_x^2\big(\partial_{\partial_x^{l+2} v} Q\big)
+ \cdots ,
\quad
l=1,2,\ldots$$ where $l=0$ corresponds to the ordinary Euler operator . In terms of the Euler operators , the Frechet derivative and its adjoint are given by $$\begin{aligned}
\delta Q& =
\delta v E_v(Q) + D_x(\delta v E_v^{(1)}(Q)) + D_x^2(\delta v E_v^{(2)}(Q)) + \cdots ,
\\
\delta^* Q & =
\delta v E_v(Q) -(D_x\delta v) E_v^{(1)}(Q) + (D_x^2\delta v) E_v^{(2)}(Q) + \cdots . \end{aligned}$$ Hence, $\delta Q = \delta^* Q$ is equivalent to the following explicit conditions.
An expression $Q(x,v,v_x,\ldots)$ is a Hamiltonian gradient iff it satisfies
$$\begin{aligned}
\partial_v Q & = E_v(Q)
\\
\partial_{v_x}Q & = -E_v^{(1)}(Q)
\\
\partial_{v_{xx}}Q & = E_v^{(2)}(Q)
\\
& \vdots
\nonumber\end{aligned}$$
which comprise the Helmholtz conditions.
Acknowledgements {#acknowledgements .unnumbered}
================
S.C.A. is supported by an NSERC research grant. The reviewers are thanked for helpful comments which have improved this paper.
[99]{}
R. Camassa, D.D. Holm, An integrable shallow water equation with peaked solitons, [*Phys. Rev. Lett.*]{} 71 (1993), 1661–1664.
R. Camassa, D.D. Holm, J.M. Hyman, A new integrable shallow water equation, [*Adv. Appl. Mech.*]{} 31 (1994), 1–33.
S.C. Anco and E. Recio, A general family of multi-peakon equations, preprint (2016).\
arXiv:1609.04354 math-ph
M. Fisher, J. Schiff, The Camassa Holm equation: conserved quantities and the initial value problem, [*Phys. Lett. A*]{} 259 (1999), 371–376.
B. Fuchssteiner, A.S. Fokas, Symplectic structures, their Bäcklund transformations and hereditary symmetries, [*Phys. D*]{} 4 (1981/1982), 47–66.
Z. Qiao, The Camassa-Holm hierarchy, N-dimensional integrable systems, and algebro-geometric solution on a symplectic submanifold, [*Commun. Math. Phys.*]{} 239 (2003), 309–341.
B. Fuchssteiner, Some tricks from the symmetry-toolbox for nonlinear equations: generalizations of the Camassa-Holm equation, [*Phys. D*]{} 95 (1996), 229-–243.
K.S. Chou and C. Qu, Integrable equations arising from motions of plane curves I., [*Phys. D*]{} 162 (2002), 9–33.
P.J. Olver and P. Rosenau, Tri-Hamiltonian duality between solitons and solitary-wave solutions having compact support, [*Phys. Rev. E*]{} 53 (1996) 1900–1906.
A.S. Fokas, On a class of physically important integrable equations, [*Phys. D*]{} 87 (1995), 145–150.
A. Fokas, The Korteweg-de Vries equation and beyond, [*Acta Appl. Math.*]{} 39 (1995), 295–305.
A.S. Fokas, P.J. Olver, P. Rosenau, A plethora of integrable bi-Hamiltonian equations. In: [*Algebraic Aspects of Integrable Systems*]{}, 93–101. Progr. Nonlinear Differential Equations Appl., vol. 26, Brikhauser Boston, 1997.
Z. Qiao, A new integrable equation with cuspons and W/M-shape-peaks solitons, [*J. Math. Phys.*]{} 47 (2006), 112701.
Z. Qiao, New integrable hierarchy, parametric solutions, cuspons, one-peak solitons, and M/W-shape peak solutions, [*J. Math. Phys.*]{} 48 (2007), 082701 (20pp).
Z. Qiao, X.Q. Li, A integrable equation with nonsmooth solitons, [*Theor. Math. Phys.*]{} 267 (2011), 584–589.
G. Gui, Y. Liu, P.J. Olver, C. Qu, Wave-breaking and peakons for a modified Camassa-Holm equation, [*Commun. Math. Phys.*]{} 319 (2013), 731–759.
J. Kang, X. Liu, P.J. Olver, C. Qu, Liouville correspondence between the modified KdV hierarchy and its dual integrable hierarchy, [*J. Nonlin. Sci.*]{} 26 (2016), 141–170.
B. Xia and Z. Qiao, A new two-component integrable system with peakon solutions, [Proc. Roy. Soc. A]{} 471 (2015), 20140750 (20pp).
B. Xia, Z. Qiao, R. Zhou, A synthetical two-component model with peakon solutions, [*Stud. Appl. Math.*]{} 135 (2015) 248–276.
M.J. Ablowitz, D.J. Kaup, A.C. Newell, H. Segur, The inverse scattering transform — Fourier analysis for nonlinear problems, [*Stud. Appl. Math.*]{} 53 (1974), 249–315.
G. Mari Beffa, J. Sanders, J.-P. Wang, Integrable systems in three-dimensional Riemannian geometry, [*J. Nonlinear Sci.*]{} 12 (2002), 143–167.
S.C. Anco and R. Myrzakulov, Integrable generalizations of Schrodinger maps and Heisenberg spin models from Hamiltonian flows of curves and surfaces, [*J. Geom. Phys.*]{} 60 (2010), 1576–1603.
R. Hirota, Exact envelope‐soliton solutions of a nonlinear wave equation, [*J. Math. Phys.*]{} 14 (1973) 805–809.
P.J. Olver, [*Applications of the groups to differential equations*]{} (2nd ed.) Springer-Verlag, 1993.
F. Magri, A simple model of the integrable Hamiltonian equation, [*J. Math. Phys.*]{} 19 (1978), 1156–1162.
P.J. Olver, Evolution equations possessing infinitely many symmetries, [*J. Math. Phys.*]{} 18 (1977), 1212–1215.
S.C. Anco, Conservation laws of scaling-invariant field equations, [*J. Phys. A: Math. and Gen.*]{} 36 (2003), 8623–8638.
A. Constantin, W.A. Strauss, Stability of peakons, [*Comm. Pure Appl. Math.*]{} 53 (2000), 603–610.
A. Constantin, L. Molinet, Global weak solutions for a shallow water wave equation, [*Comm. Math. Phys.*]{} 211 (2000), 45–61.
A. Constantin, L. Molinet, Orbital stability of solitary waves for a shallow water wave equation, [*Physica D*]{} 157 (2001), 75–89.
G. Gui, Y. Liu, L. Tian, Global existence and blow-up phenomena for the peakon $b$-family of equations, [*Indiana Univ. Math. J.*]{} 57 (2008), 1209–1233.
S. Hakkaev, Stability of peakons for an integrable shallow water equation, [*Phys. Lett. A*]{} 354 (2006), 137-–144.
S.C. Anco, A.S. Mia, M.R. Willoughby, Oscillatory solitons of U(1)-invariant mKdV equations II: Asymptotic behavior and constants of motion, J. Math. Phys. 56 (2015) 121504 (21 pages)
C. Qu, X. Liu, Y. Liu, Stability of peakons for an integrable modified Camassa-Holm equation with cubic nonlinearity, [*Commun. Math. Phys.*]{} 322 (2013), 967–997.
[^1]: [email protected], [email protected].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We give a simple and a more explicit proof of a mod $4$ congruence for a series involving the little $q$-Jacobi polynomials which arose in a recent study of a certain restricted overpartition function.'
address: 'Department of Mathematics, Indian Institute of Technology Gandhinagar, Palaj, Gandhinagar, Gujarat 382355, India'
author:
- Atul Dixit
title: 'A simple proof of a congruence for a series involving the little $q$-Jacobi polynomials'
---
Introduction {#intro}
============
In [@adsy1], Andrews, Schultz, Yee and the author studied the overpartition function $\overline{p}_{\omega}(n)$, namely, the number of overpartitions of $n$ such that all odd parts are less than twice the smallest part, and in which the smallest part is always overlined. In the same paper, they obtained a representation for the generating function of $\overline{p}_{\omega}(n)$ in terms of a ${}_3\phi_{2}$ basic hypergeometric series and an infinite series involving the little $q$-Jacobi polynomials. The latter are given by [@aa Equation (3.1)] $$\label{little}
p_n(x;{\alpha},{\beta}:q):={}_{2}\phi_{1}\bigg(\begin{matrix}q^{-n},& {\alpha}{\beta}q^{n+1}\\
&{\alpha}q\end{matrix}\, ;qx\bigg),$$ where the basic hypergeometric series ${}_{r+1}\phi_{r}$ is defined by $$\label{bhs}
{}_{r+1}\phi_{r}\left(\begin{matrix} a_1, a_2, \ldots, a_{r+1}\\
b_1, b_2, \ldots, b_{r} \end{matrix}\,; q,
z \right) :=\sum_{n=0}^{\infty} \frac{(a_1;q)_n (a_2;q)_n \cdots (a_{r+1};q)_n}{(q;q)_n (b_1;q)_n \cdots (b_{r};q)_n} z^n,$$ and where we use the notation $$\begin{aligned}
&(A;q)_0 =1;\quad (A;q)_n = (1-A)(1-Aq)\cdots(1-Aq^{n-1}),\quad n \geq 1,\\
&(A;q)_{{\infty}} = \lim_{n\to{\infty}}(A;q)_n\hspace{3mm} (|q|<1)\nonumber.\end{aligned}$$ The precise representation for the generating function of $\overline{p}_{\omega}(n)$ obtained in [@adsy1] is as follows.
\[fgfthm\] The following identity holds for $|q|<1$: $$\begin{aligned}
\label{fgf}
\overline{P}_{\omega}(q):=\sum_{n=1}^{\infty}\overline{p}_{\omega}(n)q^n&=-\frac{1}{2}\frac{(q;q)_{\infty}(q;q^2)_{\infty}}{(-q;q)_{\infty}(-q;q^2)_{\infty}}{}_{3}\phi_{2}\left(\begin{matrix}&-1, &iq^{1/2},&-iq^{1/2}\\
& q^{1/2},& -q^{1/2}
\end{matrix}\, ;q, q \right)\nonumber\\
&\quad+\frac{(-q;q)_{\infty}}{(q;q)_{\infty}}\sum_{n=0}^{\infty}\frac{(q;q^2)_n(-q)^n}{(-q;q^2)_{n}(1+q^{2n})}p_{2n}(-1;q^{-2n-1},-1:q).\end{aligned}$$
Later, Bringmann, Jennings-Shaffer and Mahlburg [@bjm Theorem 1.1] showed that $\overline{P}_{\omega}(q)+\frac{1}{4}-\frac{\eta(4\tau)}{2\eta(2\tau)^2}$, where $q=e^{2\pi i\tau}$ and $\eta(\tau)$ is the Dedekind eta function, can be completed to a function $\hat{P}_{\omega}(\tau)$, which transforms like a weight $1$ modular form. They called the function $\overline{P}_{\omega}(q)+\frac{1}{4}-\frac{\eta(4\tau)}{2\eta(2\tau)^2}$ a *higher depth mock modular form*.
While the series involving the little $q$-Jacobi polynomials in Theorem \[fgfthm\] itself looks formidable, it was shown in [@adsy1 Theorem 1.3] that modulo $4$, it is a simple $q$-product. The mod $4$ congruence proved in there is given below.
\[congmod4\] The following congruence holds: $$\label{mod4}
\sum_{n=0}^{\infty}\frac{(q;q^2)_n(-q)^n}{(-q;q^2)_{n}(1+q^{2n})}p_{2n}(-1;q^{-2n-1},-1:q)\equiv\frac{1}{2} \frac{(q;q^2)_{\infty}}{(-q;q^2)_{\infty}} \pmod{4}.$$
The proof of this congruence in [@adsy1] is beautiful but somewhat involved. The objective of this short note is to give a very simple proof of it. In fact, we derive it as a trivial corollary of the following result.
\[congmod41\] For $|q|<1$, we have $$\begin{aligned}
&\sum_{n=0}^{\infty}\frac{(q;q^2)_n(-q)^n}{(-q;q^2)_{n}(1+q^{2n})}p_{2n}(-1;q^{-2n-1},-1:q)\nonumber\\
&=\frac{1}{2} \frac{(q;q^2)_{\infty}}{(-q;q^2)_{\infty}}+\frac{4q^2}{(1+q)}\sum_{n=0}^{\infty}\frac{(q^3;q^2)_n(-q)^n}{(-q^3;q^2)_{n}(1+q^{2n+2})}\sum_{j=0}^{n}\frac{(-q;q)_{2j}q^{2j}}{(q^2;q)_{2j}}.\end{aligned}$$
The presence of $4$ in front of the series on the right-hand side in the above equation immediately implies that Theorem \[congmod4\] holds.
Proof of Theorem \[congmod41\]
==============================
Observe that from , $$\begin{aligned}
\label{diff1}
\sum_{n=0}^{\infty}\frac{(q;q^2)_n(-q)^n}{(-q;q^2)_{n}(1+q^{2n})}p_{2n}(-1;q^{-2n-1},-1:q)=
\sum_{n=0}^{\infty}\frac{(q;q^2)_n(-q)^n}{(-q;q^2)_{n}(1+q^{2n})}\sum_{j=0}^{2n}\frac{(-1;q)_j}{(q;q)_j}(-q)^j.\end{aligned}$$ However, let us first consider $$\label{diff2}
A(q):=\sum_{n=0}^{\infty}\frac{(q;q^2)_n(-q)^n}{(-q;q^2)_{n}(1+q^{2n})}\sum_{j=0}^{2n}\frac{(-1;q)_j}{(q;q)_j}q^j.$$ The only difference in the series on the right-hand side of and the series in is the presence of $(-1)^{j}$ inside the finite sum in the former.
To simplify $A(q)$, we start with a result of Alladi [@alladi p. 215, Equation (2.6)]: $$\label{alla}
\frac{(abq;q)_n}{(bq;q)_n}=1+b(1-a)\sum_{j=1}^{n}\frac{(abq;q)_{j-1}q^j}{(bq;q)_j}.$$ Let $a=-1, b=1$ and replace $n$ by $2n$ so that $$\label{aftall}
\sum_{j=0}^{2n}\frac{(-1;q)_jq^j}{(q;q)_j}=\frac{(-q;q)_{2n}}{(q;q)_{2n}}.$$ Substitute in to see that $$\begin{aligned}
\label{diff3}
A(q)&=\sum_{n=0}^{\infty}\frac{(q;q^2)_n(-q)^n}{(-q;q^2)_{n}(1+q^{2n})}\frac{(-q;q)_{2n}}{(q;q)_{2n}}\nonumber\\
&=\frac{1}{2}+\sum_{n=1}^{\infty}\frac{(-q^2;q^2)_{n-1}}{(q^2;q^2)_n}(-q)^n\nonumber\\
&=\frac{1}{2}\frac{(q;q^2)_{\infty}}{(-q;q^2)_{\infty}},\end{aligned}$$ where in the last step we used the $q$-binomial theorem $\sum_{n=0}^{\infty}\frac{(a;q)_n}{(q;q)_n}z^n=\frac{(az;q)_{\infty}}{(z;q)_{\infty}}$, valid for $|z|<1$ and $|q|<1$.
From and , [$$\begin{aligned}
\label{diff4}
&\sum_{n=0}^{\infty}\frac{(q;q^2)_n(-q)^n}{(-q;q^2)_{n}(1+q^{2n})}p_{2n}(-1;q^{-2n-1},-1:q)-A(q)\nonumber\\
&=\sum_{n=0}^{\infty}\frac{(q;q^2)_n(-q)^n}{(-q;q^2)_{n}(1+q^{2n})}\sum_{j=0}^{2n}((-1)^j-1)\frac{(-1;q)_jq^j}{(q;q)_j}\nonumber\\
&=-2\sum_{n=0}^{\infty}\frac{(q;q^2)_n(-q)^n}{(-q;q^2)_{n}(1+q^{2n})}\sum_{j=1}^{n}\frac{(-1;q)_{2j-1}q^{2j-1}}{(q;q)_{2j-1}}\nonumber\\
&=\frac{4q^2}{(1+q)}\sum_{n=0}^{\infty}\frac{(q^3;q^2)_n(-q)^n}{(-q^3;q^2)_{n}(1+q^{2n+2})}\sum_{j=0}^{n}\frac{(-q;q)_{2j}q^{2j}}{(q^2;q)_{2j}}.\end{aligned}$$]{} Invoking , we see that the proof of Theorem \[congmod41\] is complete.
[00]{}
K. Alladi, *Variants of classical q-hypergeometric identities and partition implications*, Ramanujan J. **31** Issues 1-2 (2013), 213–238.
G. E. Andrews and R. A. Askey, *Enumeration of partitions: The role of Eulerian series and $q$-orthogonal polynomials*, Higher Combinatorics, M. Aigner, ed., Reidell Publ. Co., Dordrecht, Holland, pp. 3-26 (1977).
G. E. Andrews, A. Dixit, D. Schultz and A. J. Yee, *Overpartitions related to the mock theta function $\omega(q)$*, Acta Arith. **181** No. 3 (2017), 253–286.
K. Bringmann, C. Jennings-Shaffer and K. Mahlburg, *On a modularity conjecture of Andrews, Dixit, Schultz and Yee for a variation of Ramanujan’s $\omega(q)$*, Adv. Math. **325** (2018), 505–532.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'For every integer $n\ge 1$ let $a_n$ be the smallest positive integer such that $n+a_n$ is prime. We investigate the behavior of the sequence $(a_n)_{n\ge 1}$, and prove asymptotic results for the sums $\sum_{n\le x} a_n$, $\sum_{n\le x} 1/a_n$ and $\sum_{n\le x} \log a_n$.'
author:
- 'Brăduţ Apostol, , Lucian Petrescu, and László Tóth'
title: '[**Some properties of a sequence defined with the aid of prime numbers**]{}'
---
Journal of Integer Sequences [**18**]{} (2015), Article 15.5.5
: 11A41, 11N05
[*Key Words and Phrases*]{}: prime numbers, difference of consecutive primes, asymptotic behavior
Introduction
============
For every integer $n\ge 1$ let $a_n$ be the smallest positive integer such that $n+a_n$ is prime. Here $a_1=1$, $a_2=1$, $a_3=2$, $a_4=1$, $a_5=2$, $a_6=1$, $a_7=4$, etc. This is sequence A013632 in Sloane’s Online Encyclopedia of Integer Sequences [@OEIS]. For $n\ge 2$, $a_n$ is the smallest positive integer such that $\gcd(n!,n+a_n)=1$. In this paper we study the behavior of the sequence $(a_n)_{n\ge 1}$, and prove asymptotic results for the sums $\sum_{n\le x} a_n$, $\sum_{n\le x} 1/a_n$ and $\sum_{n\le x} \log a_n$.
We are going to use the following standard notation:
$\pi(x)$ is the number of primes $\le x$,
$\pi_2(x)$ is the number of twin primes $p,p+2$ such that $p\le x$,
$p_n$ is the $n$-th prime,
$d_n=p_{n+1}-p_n$,
$f(x) \ll g(x)$ means that $|f(x)|\le C g(x)$, where $C$ is an absolute constant,
$g(x) \gg f(x)$ means that $f(x) \ll g(x)$,
$f(x) = F(x)+ O(g(x))$ means that $f(x) - F(x) \ll g(x)$,
$f(x) \asymp g(x)$ means that $c f(x) \le g(x) \le Cf(x)$ for some positive absolute constants $c$ and $C$,
$f(x) \sim g(x)$ means that $\lim_{x\to \infty} f(x)/g(x)=1$.
We will apply the following known asymptotic results concerning the distribution of the primes: $$\pi(x) \sim \frac{x}{\log x}, \quad p_n\sim n\log n \quad {\text{\rm (Prime number theorem)}},$$ $$\label{H}
\sum_{p_n\le x} d_n^2\ll x^{23/18+\varepsilon} \quad {\text{\rm for
every $\varepsilon >0$ (unconditional result of Heath-Brown
\cite{H})}},$$ $$\label{S}
\sum_{p_n\le x} d_n^2\ll x (\log x)^3 \quad {\text{\rm (assuming
the Riemann hypothesis, result of Selberg \cite{S})}},$$ $$\label{P}
\left(\frac{d_2d_3\cdots d_n}{(\log 2)(\log 3)\cdots (\log
n)}\right)^{1/n}\asymp 1 \quad {\text{\rm (due to Panaitopol
\cite[Prop.\ 3]{P})}}.$$
This research was initiated by Laurenţiu Panaitopol (1940–2008), former professor at the Faculty of Mathematics, University of Bucharest, Romania. The present paper is dedicated to his memory.
Equations and identities
========================
By the definition of $a_n$, for every $n\ge 1$ we have $n+a_n=p_{\pi(n)+1}$, that is $$\label{exp_form}
a_n= p_{\pi(n)+1}-n.$$
From we deduce that for every $k\ge 1$, $$\label{values}
a_{p_k}= p_{k+1}-p_k, a_{p_k+1}= p_{k+1}-p_k-1,\ldots, a_{p_{k+1}-1}=1.$$
For every integer $a\ge 1$ the equation $a_n=a$ has infinitely many solutions.
Let $A_k=\{1,2,\ldots, p_{k+1}-p_k\}$. Since $\limsup_{k\to \infty}
(p_{k+1}-p_k)=\infty$, it follows from that for every integer $a\ge 1$ there exist infinitely many integers $k\ge 1$ such that $a\in A_k$, whence the equation $a_n=a$ has infinitely many solutions.
Now we compute the sum $\displaystyle S_n=\sum_{i=1}^n a_i$.
For every prime $n\ge 3$ we have $$\label{sum_prime}
S_n= \frac1{2}\left(2p_{\pi(n)+1}-p_{\pi(n)}+\sum_{k=1}^{\pi(n)-1} d_k^2 \right),$$ and for every composite number $n\ge 4$, $$\label{sum_composite}
S_n= \frac1{2}\left(p_{\pi(n)}^2+ 2(n+1-p_{\pi(n)}) p_{\pi(n)+1}
+\sum_{k=1}^{\pi(n)-1} d_k^2 -n^2-n \right).$$
If $n\ge 3$ is a prime, then $n=p_m$ for some $m\ge 2$. By using , $$\begin{aligned}
S_n & = \sum_{i=1}^n \left(p_{\pi(i)+1}-i\right) \\
& = 2+3+(5+5)+\cdots+(p_m-p_{m-1})p_m+p_{m+1}-\frac{n(n+1)}{2} \\
& = 2+ \sum_{k=2}^m p_k(p_k-p_{k-1})+p_{m+1}- \frac{n(n+1)}{2}\\
& = \frac1{2}\left(p_1^2+ 2 \sum_{k=2}^m p_k^2 - 2\sum_{k=2}^m p_k
p_{k-1} +2p_{m+1}- n^2-n\right)\\
& = \frac1{2}\left(2p_{m+1}-n+ \sum_{k=1}^{m-1} (p_{k+1}-p_k)^2
\right)\end{aligned}$$ and follows by using that $m=\pi(n)$.
Now let $t\ge 4$ be composite. Let $m\ge 2$ be such that $p_m<t<p_{m+1}$. By applying for $n=p_m$, where $m=\pi(n)=\pi(t)$, we deduce $$\begin{aligned}
S_t & = S_n+ \sum_{i=n+1}^t a_i = S_n + \sum_{i=n+1}^t \left(p_{\pi(i)+1}-i \right)\\
& = \frac1{2}\left(2p_{\pi(t)+1}-p_{\pi(t)}+\sum_{k=1}^{\pi(t)-1}
(p_{k+1}-p_k)^2 \right)+ \frac{(2p_{\pi(t)+1}-n-t-1)(t-n)}{2}\\
& = \frac1{2}\left(2p_{\pi(t)+1}-p_{\pi(t)}+\sum_{k=1}^{\pi(t)-1}
(p_{k+1}-p_k)^2 + 2p_{\pi(t)+1}(t-n)-t^2-t+n^2+n\right)\\
& = \frac1{2}\left(p_{\pi(t)}^2+ 2(t+1-p_{\pi(t)}) p_{\pi(t)+1}
+\sum_{k=1}^{\pi(t)-1} (p_{k+1}-p_k)^2-t^2-t \right)\end{aligned}$$ and is proved.
If $n$ is prime, then reduces to . Therefore, the identity holds for every integer $n\ge 3$.
Next we compute the product $\displaystyle P_n=\prod_{i=1}^n a_i$.
For every prime $n\ge 3$ we have $$\label{prod_prime}
P_{n-1} = \prod_{k=1}^{\pi(n)-1} d_k!,$$ and for every composite number $n\ge 4$, $$\label{prod_composite} P_{n-1} = \prod_{k=1}^{\pi(n)-1} d_k!
\prod_{k=1}^{n-p_{\pi(n)}} (p_{\pi(n)+1}-p_{\pi(n)}-k+1).$$
Let $n=p_m\ge 3$ be a prime. By using , $$\begin{aligned}
P_{n-1} & = \prod_{i=2}^m (p_i-p_{i-1})!= \prod_{i=1}^{m-1} (p_{i+1}-p_i)!,\end{aligned}$$ which proves .
Now let $t\ge 4$ be composite such that $p_m<t<p_{m+1}$. By applying for $n=p_m$, where $m=\pi(n)=\pi(t)$, we deduce $$\begin{aligned}
P_{t-1} & = P_{n-1} \prod_{i=n}^{t-1} a_i = P_{n-1} \prod_{i=n}^{t-1} \left(p_{\pi(i)+1}-i \right)\\
& = \prod_{k=1}^{\pi(t)-1} d_k! \prod_{j=1}^{t-p_m} \left(p_{m+1}-p_m-j+1 \right) \\
& = \prod_{k=1}^{\pi(t)-1} d_k! \prod_{k=1}^{t-p_{\pi(t)}} \left(p_{\pi(t)+1}-p_{\pi(t)}-k+1 \right)\end{aligned}$$ and is proved.
If $n$ is prime, then the second product in is empty and reduces to . Hence the identity holds for every integer $n\ge 3$.
Asymptotic results
==================
For every $\varepsilon >0$, $$\label{asymp_sum}
x\log x \ll \sum_{n\le x} a_n \ll x^{23/18+\varepsilon},$$ where $23/18\doteq 1.277$. If the Riemann hypothesis is true, then the upper bound in is $x(\log x)^3$.
Let $x\ge 2$ and let $p_k\le x< p_{k+1}$. By using for $n=p_{k+1}$, $$\begin{aligned}
\sum_{n\le x} a_n & \le \sum_{i=1}^{p_{k+1}} a_i =
\frac1{2}\left(2p_{k+2}-p_{k+1} +\sum_{i=1}^k d_i^2
\right)\\
& \ll p_{k+2} + \sum_{p_i\le x} d_i^2.\end{aligned}$$
Taking into account the estimate due to Heath-Brown, and the fact that $p_{k+2}\sim p_k \le x$ we get the unconditional upper bound in . If the Riemann hypothesis is true, then by using Selberg’s result we obtain the upper bound $x(\log x)^3$.
Now, for the lower bound we use the trivial estimate $$\sum_{p_n\le x} d_n^2 \gg x\log x,$$ which follows from the inequality between the arithmetic and quadratic means. We deduce that $$\begin{aligned}
\sum_{n\le x} a_n & \ge \sum_{i=1}^{p_k} a_i =
\frac1{2}\left(2p_{k+1}-p_k +\sum_{i=1}^{k-1} d_i^2
\right)\\
& \gg \sum_{p_i\le p_{k-1}} d_i^2 \gg p_{k-1} \log
p_{k-1} \sim x\log x,\end{aligned}$$ since $p_{k-1}\sim k\log k$ and $k=\pi(x)\sim x/\log x$, $\log k
\sim \log x$.
To prove our next result we need the following
We have $$\label{asympt_lemma}
\sum_{2\le n\le x} \log d_n = x\log \log x + O(x).$$
The inequalities can be written as $$c n < \sum_{i=2}^n \log d_i -\sum_{i=2}^n \log \log i < C n$$ for some positive absolute constants $c$ and $C$. Now emerges by applying the well known asymptotic formula $$\sum_{2\le n\le x} \log \log n = x\log \log x + O(x).$$
We have $$\label{asymp_harm}
\sum_{n\le x} \frac1{a_n} = \frac{x\log \log x}{\log x} +
O\left(\frac{x}{\log x}\right).$$
For $x=p_m-1$ ($m\ge 2$) we have by , $$\sum_{n\le p_m-1} \frac1{a_n} = 1+ \sum_{i=2}^m
\left(1+\frac1{2}+\cdots+\frac1{p_i-p_{i-1}}\right).$$
For an arbitrary $x\ge 3$ let $p_k$ ($k\ge 2$) be the prime such that $p_k\le x< p_{k+1}$. Using the familiar inequalities $$\log m < 1+\frac1{2}+\cdots+\frac1{m}\le 1+\log m \quad (m\ge 1)$$ we deduce $$\log (p_i-p_{i-1})< 1+\frac1{2}+\cdots+\frac1{p_i-p_{i-1}} \le
1+\log (p_i-p_{i-1}) \quad (i\ge 2)$$ and $$1+\sum_{i=2}^k \log (p_i-p_{i-1}) +\frac1{d_k} < \sum_{n\le p_k-1} \frac1{a_n}+ \frac1{a_{p_k}}$$ $$\le \sum_{n\le x} \frac1{a_n} \le \sum_{n\le p_{k+1}-1} \frac1{a_n}
\le 1+ k + \sum_{i=2}^{k+1} \log (p_i-p_{i-1}).$$
By we obtain $$\sum_{n\le x} \frac1{a_n} = k\log \log k + O(k),$$
Here $k=\pi(x)\sim x/\log x$, $\log k
\sim \log x$ and we deduce .
One has $$x \ll \sum_{n\le x} \log a_n \ll x\log x.$$
For an arbitrary $x\ge 3$ let $p_k$ ($k\ge 2$) be the prime such that $p_k\le x< p_{k+1}$. Using the elementary inequalities $$m \log m - m + 1 \le \log m! \le m\log m \quad (m\ge 1)$$ we deduce by applying that $$\sum_{n\le x} \log a_n \le \sum_{n\le p_{k+1}-1} \log a_n =
\sum_{i=1}^k \log d_i! \le \sum_{i=1}^k d_i \log d_i$$ $$< \sum_{i=1}^k d_i \log p_i < (\log p_k) \sum_{i=1}^k d_i < (\log
p_k)p_{k+1},$$ where we also used that $d_i=p_{i+1}-p_i<p_i$ by Chebyshev’s theorem. Here $$\label{k_x}
p_k\sim k\log k, \quad k=\pi(x)\sim x/\log x, \quad \log k \sim \log
x,$$ and we obtain the upper bound $x\log x$.
On the other hand, $$\sum_{n\le x} \log a_n > \sum_{n\le p_k-1} \log a_n =
\sum_{i=1}^{k-1} \log d_i!$$ $$> \sum_{i=1}^{k-1} (d_i \log d_i-d_i+1 ) = \sum_{i=2}^{k-1} d_i\log d_i - p_k+
k+1.$$
Here $$\begin{aligned}
\sum_{i=2}^{k-1} d_i \log d_i & = \sum_{\substack{i=2\\ d_i\ge
3}}^{k-1} d_i\log d_i + 2\log 2 \sum_{\substack{i=2\\ d_i=2}}^{k-1}
1 \\
& \ge (\log 3) \sum_{\substack{i=2\\ d_i\ge 3}}^{k-1} d_i + (2\log
2)\pi_2(k-1)\\
& = (\log 3)\left( \sum_{i=2}^{k-1} d_i - \sum_{\substack{i=2\\
d_i=2}}^{k-1} d_i \right) + (2\log 2)\pi_2(k-1) \\
& = (\log 3)\left(p_k-p_2 - 2\pi_2(k-1) \right) + (2\log
2)\pi_2(k-1)\\
& = (\log 3)p_k -2\log(3/2)\pi_2(k-1)-3\log 3 \\
& > (\log 3)p_k -2\log(3/2)k- 3\log 3,\end{aligned}$$ where it is sufficient to use the obvious estimate $\pi_2(k-1)< k$. Note that $\log 3\doteq 1.09$, $2\log(3/2)\doteq 0.81$, $3\log
3\doteq 3.29$.
We deduce that $$\sum_{n\le x} \log a_n > 0.09 p_k - 3.$$
Now, gives the lower bound $x$.
[99]{}
D. R. Heath–Brown, The differences between consecutive primes, III, [*J. London Math. Soc.*]{} (2) [**20**]{} (1979), 177–178.
L. Panaitopol, Properties of the series of differences of prime numbers, [*Publ. Centre de Rech. Math. Pures Neuchâtel (Serie 1)*]{} [**31**]{} (2000), 21–28.
A. Selberg, On the normal density of primes in small intervals, and the difference between consecutive primes, [*Arch. Math. Naturvid.*]{} [**47**]{} (1943), No. 6, 87–105.
N. J. A. Sloane, The On-Line Encyclopedia of Integer Sequences. <http://oeis.org>
B. Apostol\
Pedagogic High School “Spiru Haret”, Str. Timotei Cipariu 5, RO–620004 Focşani, Romania\
E-mail: `[email protected]`
L. Petrescu\
Technical College “Henri Coandă”, Str. Tineretului 2, RO–820235 Tilcea, Romania\
E-mail: `[email protected]`
L. Tóth\
Department of Mathematics, University of Pécs\
Ifjúság útja 6, H–7624 Pécs, Hungary\
E-mail: `[email protected]`
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We propose a modified color-magnitude diagram for novae in outburst, i.e., $(B-V)_0$ versus $(M_V-2.5 \log f_{\rm s})$, where $f_{\rm s}$ is the timescaling factor of a (target) nova against a comparison (template) nova, $(B-V)_0$ is the intrinsic $B-V$ color, and $M_V$ is the absolute $V$ magnitude. We dub it the time-stretched color-magnitude diagram. We carefully reanalyzed 20 novae based on the time-stretching method and revised their extinctions $E(B-V)$, distance moduli in the $V$ band $(m-M)_V$, distances $d$, and timescaling factors $f_{\rm s}$ against the template nova LV Vul. We have found that these 20 nova outburst tracks broadly follow one of the two template tracks, LV Vul/V1668 Cyg or V1500 Cyg/V1974 Cyg group, in the time-stretched color-magnitude diagram. In addition, we estimate the white dwarf masses and $(m-M)_V$ of the novae by directly fitting the absolute $V$ model light curves ($M_V$) with observational apparent $V$ magnitudes ($m_V$). A good agreement in the two estimates of $(m-M)_V$ confirms the consistency of the time-stretched color-magnitude diagram. Our distance estimates are in good agreement with the results of Gaia Data Release 2.'
author:
- Izumi Hachisu
- Mariko Kato
title: 'The $UBV$ Color Evolution of Classical Novae. III. Time-Stretched Color-Magnitude Diagram of Novae in Outburst'
---
Introduction
============
A nova is a thermonuclear explosion event on a mass-accreting white dwarf (WD) in a close binary. When the mass of the hydrogen-rich envelope on the WD reaches a critical value, hydrogen ignites at the bottom of the hydrogen-rich envelope to trigger a nova explosion. Then, the envelope expands to a red-giant size and emits strong winds [@kat94h]. After the maximum expansion of the photosphere (when the maximum wind mass-loss rate is attained), the photospheric radius begins to shrink and the photospheric temperature rises with time. The main emitting wavelength region thus shifts from optical to UV and supersoft X-ray [see, e.g., Figure 23 of @hac06kb for V1974 Cyg]. A large part of the hydrogen-rich envelope is blown in the wind. The wind stops and the nova enters a supersoft X-ray source (SSS) phase. The envelope mass decreases further by nuclear burning. When it decreases to below the critical mass (minimum mass for steady hydrogen shell-burning), nuclear burning extinguishes. The WD cools down and the nova outburst ends. These evolutions of novae are briefly summarized in @hac17k.
Evolutions of nova outbursts can be followed by the optically thick wind theory. @kat94h calculated nova evolution for various WD masses (from $0.5~M_\sun$ to $1.38~M_\sun$) and showed that the main parameter which determines the speed class of novae is the WD mass. A nova on a more massive WD evolves faster because the ignition mass (envelope mass at the epoch when hydrogen ignites) is much smaller and is blown off in the wind in a much shorter time.
The acceleration mechanism of the wind is the so-called continuum radiation-driven [@fri66]. The matter in the envelope is accelerated deep inside the photosphere, in other words, at/near the iron peak of the radiative opacity ($\log T$ (K) $\sim 5.2$). So, it is called the optically thick winds. This mechanism is common among all the WD masses ranging from $\sim0.5~M_\sun$ to $1.38~M_\sun$ [@kat94h].
Nova spectra in the decay phase are dominated by free-free emission [@gal76; @enn77; @kra84; @nai09] until the nova entered the nebular phase where strong emission lines such as \[\] are dominant. Free-free emission of novae is originating from the expanding hot and optically thin gaseous plasma, that is, outside the photosphere of the optically thick winds. Free-free emission light curves of novae are theoretically calculated from the wind-mass loss rate $\dot M_{\rm wind}$, photospheric velocity $v_{\rm ph}$, and photospheric radius $R_{\rm ph}$, that is, $F_\nu \propto \dot M^2_{\rm wind} / v^2_{\rm ph} R_{\rm ph}$ [@hac06kb], where $F_\nu$ is the flux of free-free emission at the frequency $\nu$. It should be noted that the flux of free-free emission $F_\nu$ is almost independent of the frequency $\nu$ in the optical and near-infrared (NIR) region.
@hac06kb calculated free-free emission light curves for various WD masses (from $0.5~M_\sun$ to $1.38~M_\sun$) and envelope chemical compositions. Since the flux of free-free emission $F_\nu$ is almost independent of the frequency $\nu$, the light curve shapes are very similar among the optical and NIR bands until the nebular phase or dust formation started. For example, the $V$, $J$, $H$, $2.3\mu$m, and $3.6 \mu$m band light curves of V1668 Cyg show an almost identical light curve shape, which are well reproduced by a theoretical light curve until the optically-thin dust shell formation makes a deviation from the theoretical one [see, e.g., Figure 18 of @hac06kb].
@hac06kb further found that the theoretical free-free emission light curves and UV (1455 Å) light curves show a scaling law among various WD masses and chemical compositions. The UV 1455Å band is a narrow (1445–1465Å) and emission line-free band invented by @cas02 and represents continuum flux of UV. All the UV 1455Å model light curves also follow the same timescaling law as that of free-free emission light curves. In other words, if we properly squeeze the timescales of novae, these theoretical light curves converge into one [see Figure 18 of @hac06kb]. The main part of theoretical free-free emission light curve decays as $F_\nu \propto t^{-1.75}$. @hac06kb called this property the universal decline law of classical novae.
Because the decline rate of optical light curve is a good indicator of the WD mass, we can estimate the WD mass from comparison of the theoretical light curve with the observed optical one. If multiwavelength light curves are available, we are able to determine the WD mass more accurately. Using multiwavelength light curves in optical, UV 1455Å, and supersoft X-ray turn-on/turnoff times, @hac06kb [@hac10k; @hac16k] determined the WD masses of V1974 Cyg and V1668 Cyg. The UV 1455Å band flux represents the photospheric emission in the wind phase while the supersoft X-ray flux does in the post-wind phase. These multiwavelength data are so useful because their dependences on the WD mass and chemical composition are different. It should be noted that the soft X-ray light curves do not obey the timescaling law [see, e.g., Figure 6 of @hac10k]. We are further able to discriminate these two parameters from multiwavelength light curve fitting. Thus, @hac15k [@hac16k] determined the WD masses within error of $\pm0.01~M_\sun$ by simultaneously reproducing the observed multiwavelength light curves for a number of novae.
@hac10k introduced the timescaling factor $f_{\rm s}$, which is the timescale ratio of a target nova against a template nova. Using this $f_{\rm s}$, we can compare various nova light curves in a quantitative way. They showed that, if a large part of the nova light/color curves overlap each other by properly squeezing/stretching their timescales, the absolute brightness ($M_V$) of the target nova is related to the absolute brightness ($M'_V$) of the other (template) nova by $M_V = M'_V + 2.5 \log f_{\rm s}$. This means that the time-stretched absolute light curves of the two novae overlap each other in the $(t/f_{\rm s})$-$(M_V-2.5 \log f_{\rm s})$ plane. They adopted V1668 Cyg and V1974 Cyg as template novae, of which the distances and extinctions are known. Using this property, @hac10k [@hac15k; @hac16k] determined the absolute magnitude $M_V$ for each free-free emission light curve of different WD mass and chemical composition. They called this method the time-stretching method. By directly fitting the theoretical light curves of $M_V$ with the observed $m_V$, we can also estimate the WD mass $M_{\rm WD}$ and distance modulus in the $V$ band of $\mu_V\equiv (m-M)_V$ for a target nova. @hac18kb confirmed that the time-stretching method is applicable to fast and very fast novae in our Galaxy, LMC, and M31.
There are also common properties in the color evolution in nova outbursts. @hac14k [hereafter Paper I] found that many novae in outburst evolve on a common path in the $(B-V)_0$-$(U-B)_0$ color-color diagram. Here, $(B-V)_0$ and $(U-B)_0$ are the intrinsic colors of $B-V$ and $U-B$, respectively. The evolution path starts from the nova-giant sequence and then moves to the free-free emission phase followed by the emission line dominant nebular phase. (See, e.g., their Figure 28.)
@hac16kb [Paper II] found that many novae follow similar evolution path in the $(B-V)_0$-$M_V$ color-magnitude diagram. Here, $M_V$ is the absolute $V$ magnitude in the $V$ band. They extensively examined the color-magnitude tracks of 40 novae in outburst and found several distinct “template tracks” in the $(B-V)_0$-$M_V$ diagram. They proposed six template tracks of novae and dubbed them the V1500 Cyg, V1668 Cyg, V1974 Cyg, LV Vul, FH Ser, and PU Vul type tracks. The difference among the groups may be related to the envelope mass, which is closely related to the nova evolution speed, but was not fully clarified. They categorized 40 novae into one of these six subgroups, depending on their similarity of the tracks in the $(B-V)_0$-$M_V$ diagram. They are not fully successful, however, to explain why each nova has a similar $M_V$ in the $(B-V)_0$-$M_V$ diagram.
Nova light curves show a similar property in the $(t/f_{\rm s})$-$(M_V-2.5 \log f_{\rm s})$ plane. The color-magnitude evolution should also be treated in the same way, that is, based on the time-stretching method. In the present work, we propose the $(B-V)_0$-$(M_V-2.5\log f_{\rm s})$ color magnitude diagram. We call this “time-stretched color-magnitude diagram” in order to distinguish it from the usual $(B-V)_0$-$M_V$ diagram. The time-stretching method is useful because many nova light curves follow a universal decline law. Once their absolute $V$ magnitude is time-stretched, nova light curves overlap each other in the $(t/f_{\rm s})$-$(M_V-2.5\log f_{\rm s})$ light curve [@hac10k]. Here, we stretch/squeeze the timescale of a target nova as $t'=t/f_{\rm s}$ and overlap its light curve to the comparison (template) nova. Then, we have the relation of $$\begin{aligned}
\left( M_V[t] \right)_{\rm template}
&=& \left( M'_V[t'] \right)_{\rm target} \cr
&=& \left( M_V[t/f_{\rm s}]-2.5\log f_{\rm s} \right)_{\rm target}
\label{time-stretching_general}\end{aligned}$$ between $(t', M'_V)$ and $(t, M_V)$ coordinates systems [see Appendix A of @hac16k], where $M_V[t]$ means that $M_V$ is a function of time $t$. The color $(B-V)_0$ is not affected by this time-stretch because the both fluxes of $B$ and $V$ are shifted by the same quantities, $-2.5\log f_{\rm s}$ [@hac18k]. We thus examine each nova evolution in the $(B-V)_0$-$(M_V-2.5\log f_{\rm s})$ diagram. The above two reasons, Equation (\[time-stretching\_general\]) and that the color $(B-V)_0$ is not affected by time-stretch, are the theoretical background for the property that each nova track overlaps in the time-stretched color-magnitude diagram.
To summarize the theoretical background, both the colors $(B-V)_0$ and $(U-B)_0$ are not affected by time-stretch. Therefore, the $(t/f_{\rm s})$-$(B-V)_0$ and $(t/f_{\rm s})$-$(U-B)_0$ color curves overlap each other after time-stretch. As a result, many novae in outburst evolve on a common path in the usual $(B-V)_0$-$(U-B)_0$ color-color diagram. This is the reason why @hac14k found a common track in the $(B-V)_0$-$(U-B)_0$ diagram. On the other hand, the $(t/f_{\rm s})$-$M_V$ light curves do not overlap because the absolute $M_V$ magnitudes are different among novae. This is the reason why @hac16kb did not find common tracks in the $(B-V)_0$-$M_V$ diagram. If we use the $(B-V)_0$-$(M_V-2.5\log f_{\rm s})$ diagram, we could find common tracks because the $(t/f_{\rm s})$-$(M_V-2.5\log f_{\rm s})$ light curves overlap each other after time-stretch.
In this work, we examine nova evolution in the time-stretched color-magnitude diagram. We adopt total 20 novae, including 12 novae that were already analyzed in Paper II, but reanalyzed here in a new light of the time-stretched color-magnitude diagram. Our aim is to provide a uniform set of 20 novae analyzed by the same time-stretched color-magnitude diagram method.
Our paper is organized as follows. First we describe our method for four well-studied novae in Section \[method\_example\]. Then, we apply this method to four recent novae V574 Pup, V679 Car, V1369 Cen, and V5666 Sgr to determine their various physical parameters in Sections \[v574\_pup\_cmd\], \[v679\_car\_cmd\], \[v1369\_cen\_cmd\], and \[v5666\_sgr\_cmd\], respectively. In Section \[twelve\_novae\], we reanalyze twelve novae, which were studied in Paper II on the usual color-magnitude diagram. We confirm that all the novae follow one of the two template tracks, V1500 Cyg or LV Vul, in the time-stretched color-magnitude diagram. In Section \[discussion\], we compared our distance estimates with the Gaia Data Release 2. Conclusions follows in Section \[conclusions\].
[llllrrll]{} CI Aql & 2000 & 1.0 & 15.7 & 3.3 & $-0.22$ & 15.15 & 3\
V1419 Aql & 1993 & 0.52 & 15.0 & 4.7 & $+0.15$ & 15.35 & 4\
V679 Car & 2008 & 0.69 & 16.1 & 6.2 & $+0.0$ & 16.1 & 4\
V705 Cas & 1993 & 0.45 & 13.45 & 2.6 & $+0.45$ & 14.55 & 4\
V1065 Cen & 2007 & 0.45 & 15.0 & 5.3 & $+0.0$ & 15.0 & 2\
V1369 Cen & 2013 & 0.11 & 10.25 & 0.96 & $+0.17$ & 10.65 & 4\
IV Cep & 1971 & 0.65 & 14.5 & 3.1 & $+0.0$ & 14.5 & 2\
T CrB & 1946 & 0.056 & 10.1 & 0.96 & $-1.32$ & 13.4 & 3\
V407 Cyg & 2010 & 1.0 & 16.1 & 3.9 & $-0.37$ & 15.2 & 3\
V1500 Cyg & 1975 & 0.45 & 12.3 & 1.5 & $-0.22$ & 11.75 & 1\
V1668 Cyg & 1978 & 0.30 & 14.6 & 5.4 & $+0.0$ & 14.6 & 1\
V1974 Cyg & 1992 & 0.30 & 12.2 & 1.8 & $+0.03$ & 12.3 & 1\
V2362 Cyg & 2006 & 0.60 & 15.4 & 5.1 & $+0.25$ & 16.05 & 4\
V2468 Cyg & 2008 & 0.65 & 16.2 & 6.9 & $+0.38$ & 17.15 & 4\
V2491 Cyg & 2008 & 0.45 & 17.4 & 15.9 & $-0.34$ & 16.55 & 4\
YY Dor & 2010 & 0.12 & 18.9 & 50.0 & $-0.72$ & 17.1 & 3\
V446 Her & 1960 & 0.40 & 11.95 & 1.38 & $+0.0$ & 11.95 & 4\
V533 Her & 1963 & 0.038 & 10.65 & 1.28 & $+0.08$ & 10.85 & 4\
V838 Her & 1991 & 0.53 & 13.7 & 2.6 & $-1.22$ & 10.65 & 3\
V959 Mon & 2012 & 0.38 & 13.15 & 2.5 & $+0.14$ & 13.5 & 2\
RS Oph & 2006 & 0.65 & 12.8 & 1.4 & $-1.02$ & 10.25 & 3\
V2615 Oph & 2007 & 0.90 & 15.95 & 4.3 & $+0.20$ & 16.45 & 4\
V574 Pup & 2004 & 0.45 & 15.0 & 5.3 & $+0.10$ & 15.25 & 4\
U Sco & 2010 & 0.26 & 16.3 & 12.6 & $-1.32$ & 13.0 & 3\
V745 Sco & 2014 & 0.70 & 16.6 & 7.8 & $-1.32$ & 13.3 & 3\
V1534 Sco & 2014 & 0.93 & 17.6 & 8.8 & $-1.22$ & 14.55 & 3\
V496 Sct & 2009 & 0.45 & 13.7 & 2.9 & $+0.30$ & 14.45 & 4\
V5114 Sgr & 2004 & 0.47 & 16.65 & 10.9 & $-0.12$ & 16.35 & 4\
V5666 Sgr & 2014 & 0.50 & 15.4 & 5.8 & $+0.25$ & 16.0 & 4\
V382 Vel & 1999 & 0.25 & 11.5 & 1.4 & $-0.29$ & 10.75 & 4\
LV Vul & 1968 & 0.60 & 11.85 & 1.0 & $+0.0$ & 11.85 & 1,2,4\
PW Vul & 1984 & 0.57 & 13.0 & 1.8 & $+0.35$ & 13.85 & 4\
LMCN 2009a & 2014 & 0.12 & 18.9 & 50.0 & $-0.52$ & 17.6 & 3\
LMCN 2012a & 2012 & 0.12 & 18.9 & 50.0 & $-1.22$ & 15.85 & 3\
LMCN 2013 & 2013 & 0.12 & 18.9 & 50.0 & $-0.42$ & 17.85 & 3\
SMCN 2016 & 2016 & 0.08 & 16.8 & 20.4 & $-0.72$ & 15.0 & 3\
M31N 2008-12a & 2015 & 0.30 & 24.8 & 780 & $-1.32$ & 21.5 & 3\
[lrlllll]{} CI Aql & $-0.22$ & 1.18 & — & — & — & interp.\
V1419 Aql & $+0.15$ & 0.90 & — & — & — & CO3\
V679 Car & $+0.0$ & 0.98 & — & — & — & CO3\
V705 Cas & $+0.45$ & 0.78 & 0.78 & — & — & CO4\
V1065 Cen & $+0.0$ & 0.98 & — & — & — & CO3\
V1369 Cen & $+0.17$ & 0.90 & — & — & — & CO3\
IV Cep & $+0.0$ & 0.98 & — & — & — & CO3\
T CrB & $-1.32$ & 1.38 & — & — & — & interp.\
V407 Cyg & $-0.37$ & 1.22 & — & — & — & interp.\
V1500 Cyg & $-0.22$ & 1.20 & — & — & — & Ne2\
V1668 Cyg & $+0.0$ & 0.98 & 0.98 & — & — & CO3\
V1974 Cyg & $+0.03$ & 0.98 & 0.98 & 0.98 & 0.98 & CO3\
V2362 Cyg & $+0.25$ & 0.85 & — & — & — & interp.\
V2468 Cyg & $+0.38$ & 0.85 & — & — & — & CO4\
V2491 Cyg & $-0.34$ & 1.35 & — & 1.35 & 1.35 & Ne2\
YY Dor & $-0.72$ & 1.29 & — & — & — & interp.\
V446 Her & $+0.0$ & 0.98 & — & — & — & CO3\
V533 Her & $+0.08$ & 1.03 & — & — & — & Ne2\
V838 Her & $-1.22$ & 1.35 & 1.35 & — & — & Ne2\
V838 Her & $-1.22$ & 1.37 & 1.37 & — & — & Ne3\
V959 Mon & $+0.14$ & 0.95 & — & 0.95 & — & CO3\
V959 Mon & $+0.14$ & 1.05 & — & 1.05 & — & Ne2\
V959 Mon & $+0.14$ & 1.1 & — & 1.10 & — & Ne3\
RS Oph & $-1.02$ & 1.35 & 1.35 & 1.35 & 1.35 & evol.\
V2615 Oph & $+0.20$ & 0.90 & — & — & — & CO3\
V574 Pup & $+0.10$ & 1.05 & — & 1.05 & 1.05 & Ne2\
U Sco & $-1.32$ & 1.37 & — & 1.37 & 1.37 & evol.\
V745 Sco & $-1.32$ & 1.38 & — & 1.385 & 1.385 & evol.\
V1534 Sco & $-1.22$ & 1.37 & — & — & — & interp.\
V496 Sct & $+0.30$ & 0.85 & — & — & —0 & CO3\
V5114 Sgr & $-0.12$ & 1.15 & — & — & — & Ne2\
V5666 Sgr & $+0.25$ & 0.85 & — & — & — & CO3\
V382 Vel & $-0.29$ & 1.23 & — & — & 1.23 & Ne2\
LV Vul & $+0.0$ & 0.98 & — & — & — & CO3\
PW Vul & $+0.35$ & 0.83 & 0.83 & — & — & CO4\
LMC N 2009a & $-0.52$ & 1.25 & — & 1.25 & 1.25 & Ne3\
LMC N 2012a & $-1.22$ & 1.37 & — & — & — & interp.\
LMC N 2013 & $-0.42$ & 1.23 & — & — & — & interp.\
SMC N 2016 & $-0.72$ & 1.29 & — & — & 1.3 & Ne3\
SMC N 2016 & $-0.72$ & 1.29 & — & — & 1.25 & Ne2\
M31N 2008-12a & $-1.32$ & 1.38 & — & 1.38 & 1.38 & evol.\
Method and Template Novae {#method_example}
=========================
Method {#reddening_distance_method}
------
First, we need to know the intrinsic color and time-stretching factor $f_{\rm s}$ in order to analyze a particular nova light curve. The procedure is summarized as follows:\
[**1.**]{} Calculate the intrinsic colors of $(B-V)_0$ and $(U-B)_0$ as $$(B-V)_0 = (B-V) - E(B-V),
\label{dereddening_eq_bv}$$ and $$(U-B)_0 = (U-B) - 0.64 E(B-V),
\label{dereddening_eq_ub}$$ where the factor of $0.64$ is taken from @rie85. The color excess $E(B-V)$ is taken from literature, if available.\
[**2.**]{} Determine the time-stretching factor $f_{\rm s}$ of a target nova from a comparison with a template nova in the light/color curves (see, e.g., Figure \[v574\_pup\_v1974\_cyg\_v\_bv\_logscale\]). We adopt the template nova from one of LV Vul, V1500 Cyg, V1668 Cyg, and V1974 Cyg, of which the distance modulus $\mu_V\equiv (m-M)_V$ and extinction $E(B-V)$ are well determined. We use LV Vul as the template nova unless otherwise specified. The timescaling factor $f_{\rm s}$ is measured against the timescale of LV Vul. In the light/color curves, we stretch/squeeze the timescale of the target nova (horizontal shift in a logarithmic timescale) and also shift the magnitude in the vertical direction. The light curve of the target nova overlaps the template nova by stretching horizontally (in the time-direction) by a factor of $f_{\rm s}$ and by shifting vertically by $\Delta V$, we estimate the distance modulus of the target nova, $$(m-M)_{V,\rm target}
= \left( (m-M)_V + \Delta V \right)_{\rm template} - 2.5 \log f_{\rm s}.
\label{distance_modulus_general_temp}$$ @hac10k called this method the time-stretching method. We usually increase/decrease the horizontal shift by a step of $\delta (\Delta \log t) = \delta (\log f_{\rm s}) = 0.01$ and the vertical shift by a step of $\delta (\Delta V)= 0.1$ mag, and determine the best one by eye. The model light curves obey the universal decline law. This phase corresponds to the optically thick wind phase before the nebular phase started. In fact, many novae show a good agreement with the model light curve in the optically thick phase. Moreover, we find that the $V$ light curve of a target nova often show a good agreement with that of a template nova even in the nebular (optically thin) phase. In such a case, we use all the phases to determine the timescaling factor (see, e.g., Figure \[lv\_vul\_yy\_dor\_lmcn\_2009a\_v\_b\_logscale\_2fig\]).\
[**3.**]{} Determine the WD mass and absolute brightness from the comparison with model light curves calculated by @hac10k [@hac15k; @hac16k]. These light curves are obtained for various WD masses and chemical compositions. The distance modulus in the $V$ band, $(m-M)_V$, is determined independently of procedure No.2.\
[**4.**]{} The above procedures in No.2 and No.3 are independent methods. Thus, good agreement of the two $(m-M)_V$ confirms our method. We obtain the distance $d$ to the target nova from the relation of $$\begin{aligned}
(m-M)_V = 3.1 E(B-V) + 5 \log (d/10~{\rm pc}),
\label{distance_modulus_rv}\end{aligned}$$ where the factor $R_V=A_V/E(B-V)=3.1$ is the ratio of total to selective extinction [e.g., @rie85].\
[**5.**]{} We apply the same time-stretching method to the $U$, $B$, $I_{\rm C}$, and $K_{\rm s}$ bands, if they are available, that is, $$(m-M)_{U,\rm target} = ((m-M)_U + \Delta U
- 2.5 \log f_{\rm s})_{\rm template},
\label{distance_modulus_general_temp_u}$$ $$(m-M)_{B,\rm target} = ((m-M)_B + \Delta B
- 2.5 \log f_{\rm s})_{\rm template},
\label{distance_modulus_general_temp_b}$$ $$(m-M)_{I,\rm target} = ((m-M)_I + \Delta I_{\rm C}
- 2.5 \log f_{\rm s})_{\rm template},
\label{distance_modulus_general_temp_i}$$ $$(m-M)_{K,\rm target} = ((m-M)_K + \Delta K_{\rm s}
- 2.5 \log f_{\rm s})_{\rm template},
\label{distance_modulus_general_temp_k}$$ where the timescaling factor $f_{\rm s}$ is unique and the same as that in Equation (\[distance\_modulus\_general\_temp\]). Thus, we independently obtain the five distance moduli in the $U$, $B$, $V$, $I_{\rm C}$, and $K_{\rm s}$ bands (for example, Figure \[v574\_pup\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_k\_logscale\_4fig\]). Then, we obtain the relation between the distance $d$ and color excess $E(B-V)$ to the target nova where we use the relation of $$\begin{aligned}
(m-M)_U = 4.75 E(B-V) + 5 \log (d/10~{\rm pc}),
\label{distance_modulus_ru}\end{aligned}$$ $$\begin{aligned}
(m-M)_B = 4.1 E(B-V) + 5 \log (d/10~{\rm pc}),
\label{distance_modulus_rb}\end{aligned}$$ $$\begin{aligned}
(m-M)_I = 1.5 E(B-V) + 5 \log (d/10~{\rm pc}),
\label{distance_modulus_ri}\end{aligned}$$ $$\begin{aligned}
(m-M)_K = 0.35 E(B-V) + 5 \log (d/10~{\rm pc}),
\label{distance_modulus_rk}\end{aligned}$$ and $A_U/ E(B-V)=4.75$, $A_B/E(B-V)=4.1$, $A_I/E(B-V)= 1.5$, and $A_K/E(B-V)= 0.35$ are taken from @rie85. We plot these relations of Equations (\[distance\_modulus\_ru\]), (\[distance\_modulus\_rb\]), (\[distance\_modulus\_rv\]), (\[distance\_modulus\_ri\]), and (\[distance\_modulus\_rk\]) in the reddening-distance plane. If these lines cross at the same point, this point gives correct values of reddening $E(B-V)$ and distance $d$ (for example, Figure \[distance\_reddening\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\](a)).\
[**6.**]{} Using $E(B-V)$, $(m-M)_V$, and $f_{\rm s}$ obtained in the above procedures, we plot the $(B-V)_0$-$M_V$ color-magnitude diagram and the $(B-V)_0$-$(M_V - 2.5\log f_{\rm s})$ time-stretched color-magnitude diagram of the target nova (for example, Figures \[hr\_diagram\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\_outburst\_mv\] and \[hr\_diagram\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\_outburst\], respectively). We derive the relation of $$\begin{aligned}
(m-M')_{V,\rm target} & \equiv & \left( m_V-(M_V -
2.5\log f_{\rm s})\right)_{\rm target} \cr
&=& ((m-M)_V + \Delta V)_{\rm template},
\label{distance_modulus_general_dot}\end{aligned}$$ from Equations (\[time-stretching\_general\]) and (\[distance\_modulus\_general\_temp\]).\
[**7.**]{} If the track of the target nova overlaps to the LV Vul (or V1500 Cyg, V1668 Cyg, V1974 Cyg) template track in the time-stretched color-magnitude diagram, we regard that our set of $f_{\rm s}$, $E(B-V)$, $(m-M)_V$, and distance $d$ are reasonable.
Reddening and distance {#distance_comparison}
----------------------
In what follows, we use LV Vul, V1500 Cyg, V1668 Cyg, and V1974 Cyg as template novae, because their distances and extinctions have been well determined. Figure \[distance\_reddening\_lv\_vul\_v1500\_cyg\_v1668\_cyg\_v1974\_cyg\] shows the distance-reddening relations toward the four template novae, (a) LV Vul, (b) V1500 Cyg, (c) V1668 Cyg, and (d) V1974 Cyg. We have already analyzed them in our previous papers [e.g., @hac16k; @hac16kb], but here we reanalyze them because new distance-reddening relations are recently available [e.g., @gre18; @chen18]. To summarize, we use five results: (1) @mar06 published a three-dimensional (3D) extinction map of our Galaxy. The range is $-100\fdg0 \le l \le 100\fdg0$ and $-10\fdg0 \le b \le +10\fdg0$ and the resolution of grids is $\Delta l=0\fdg25$ and $\Delta b=0\fdg25$, where $(l, b)$ is the galactic coordinates. (2) @sal14 presented a 3D reddening map with the region of $30\arcdeg \le l < 215\arcdeg$ and $|b|<5\arcdeg$. Their data are based on the INT Photometric H-Alpha Survey (IPHAS) photometry. (3) @gre15 published a 3D reddening map with a wider range of three quarters of the sky and with much finer grids of 34 to 137 whose distance resolution is 25%. The data of distance-reddening relation were recently revised by @gre18. (4) @ozd16 obtained distance-reddening relations toward 46 novae based on the unique position of the red clump giants in the color-magnitude diagram. Recently, @ozd18 added the data of recent novae. (5) @chen18 presented 3D interstellar dust reddening maps of the galactic plane in three colors, $E(G - K_{\rm s})$, $E(G_{\rm BP}-G_{\rm RP})$, and $E(H - K_{\rm s})$. We use the conversion of $E(B-V)= 0.75 E(G_{\rm BP}-G_{\rm RP})$ from @chen18. The maps have a spatial angular resolution of 6 arcmin and covers the galactic longitude $0 < l < 360^\circ$ and latitude $|b| < 10^\circ$. The maps are constructed from parallax estimates from the Gaia Data Release 2 (Gaia DR2) combined with the photometry from the Gaia DR2 and the infrared photometry from the 2MASS and WISE surveys.
These five 3D dust maps often show large discrepancies (see, e.g., Figure \[distance\_reddening\_lv\_vul\_v1500\_cyg\_v1668\_cyg\_v1974\_cyg\](a)). The 3D dust maps essentially give an averaged value of a relatively broad region and thus, the pinpoint reddening could be different from it, because the resolutions of these dust maps are considerably larger than molecular cloud structures observed in the interstellar medium. For example, Figure \[distance\_reddening\_lv\_vul\_v1500\_cyg\_v1668\_cyg\_v1974\_cyg\](a) shows a large discrepancy among the various relations. The orange line [@gre18] is quite consistent with our values, i.e., the crossing point of $(m-M)_V=11.85$ (blue line) and $E(B-V)=0.60$ (vertical solid red line). This will be discussed in detail in Section \[lv\_vul\].
Template nova light curves {#template_novae}
--------------------------
Here, we reexamine the four template novae LV Vul, V1500 Cyg, V1668 Cyg, and V1974 Cyg, with the new distance-reddening relations [@gre18; @chen18] and fix the distances and extinctions for later use.
### LV Vul 1968\#1 {#lv_vul}
The reddening toward LV Vul was estimated by @fer69 to be $E(B-V)=0.6\pm 0.2$ from the color excesses of 14 B stars near the line of sight. @tem72 obtained $E(B-V)=0.55$ from the color at optical maximum; i.e., $E(B-V)=(B-V)_{\rm max} - (B-V)_{0, \rm max}=
0.9 - 0.35 = 0.55$. He adopted $(B-V)_{0, \rm max}= + 0.35$ [@sch57] instead of $(B-V)_{0, \rm max}=+0.23$ [@van87]. @hac14k obtained $E(B-V)=0.60\pm0.05$ by fitting with the typical color-color evolution track of nova outbursts. @sla95 obtained the distance toward LV Vul to be $d=0.92\pm 0.08$ kpc by a nebular expansion parallax method.
Figure \[lv\_vul\_yy\_dor\_lmcn\_2009a\_v\_b\_logscale\_2fig\] shows the (a) $V$ and (b) $B$ light curves of LV Vul as well as YY Dor (2004) and LMC N 2009a. The LV Vul data are the same as those in Figure 4 of Paper II. The light curves of YY Dor and LMC N 2009a are time-stretched against LV Vul. These two novae are analyzed by @hac18kb and the various properties are summarized in their Tables 1, 2, and 3. They determined the timescaling factors of YY Dor and LMC N 2009a against LV Vul, which are plotted in Figure \[lv\_vul\_yy\_dor\_lmcn\_2009a\_v\_b\_logscale\_2fig\]. In this case, we shift up/down the YY Dor and LMC N 2009a light curves in a step of $\delta(\Delta V)= 0.1$ mag (or $\delta(\Delta B)= 0.1$ mag) and find the best fit one by eye. We use all the part of the light curve if they overlap each other. If not, we use only the part of the optically thick wind phase before the nebular (or dust blackout) phase started. Fortunately, the three light curves of LV Vul, YY Dor, and LMC N 2009a broadly overlap each other even during the nebular phase as shown in Figure \[lv\_vul\_yy\_dor\_lmcn\_2009a\_v\_b\_logscale\_2fig\].
YY Dor and LMC N 2009a belong to the Large Magellanic Cloud (LMC) and their distances are well constrained. We adopt $\mu_{0,\rm LMC}=18.493\pm0.048$ [@pie13]. The reddenings toward YY Dor is assumed to be $E(B-V)=0.12$, which is the typical reddening toward the LMC [@imara07]. The distance modulus in the $V$ band is $(m-M)_V= \mu_0 + A_V= 18.49 + 3.1\times 0.12 = 18.86$ toward the LMC novae.
Applying the obtained $f_{\rm s}$ and $\Delta V$ to Equation (\[distance\_modulus\_general\_temp\]), we have the relation of $$\begin{aligned}
(m&-&M)_{V, \rm LV~Vul} \cr
&=& \left( (m-M)_V + \Delta V\right)_{\rm YY~Dor} - 2.5 \log 3.16 \cr
&=& 18.86 - 5.75\pm0.2 - 1.25 = 11.86\pm0.2 \cr
&=& \left( (m-M)_V + \Delta V\right)_{\rm LMCN~2009a} - 2.5 \log 2.0 \cr
&=& 18.86 - 6.25\pm0.2 - 0.75 = 11.86\pm0.2.
\label{distance_modulus_lv_vul_yy_dor_lmcn_2009a_v}\end{aligned}$$ Thus, we obtain $(m-M)_V=11.86\pm0.1$ for LV Vul, being consistent with our previous results of $(m-M)_V=11.85\pm0.1$ [@hac18kb].
In the same way, we obtain $\Delta B$ from Figure \[lv\_vul\_yy\_dor\_lmcn\_2009a\_v\_b\_logscale\_2fig\](b). Using this $\Delta B$ and $f_{\rm s}$, we obtain $$\begin{aligned}
(m&-&M)_{B, \rm LV~Vul} \cr
&=& \left( (m-M)_B + \Delta B\right)_{\rm YY~Dor} - 2.5 \log 3.16 \cr
&=& 18.98 - 5.3\pm0.2 - 1.25 = 12.43\pm0.2 \cr
&=& \left( (m-M)_B + \Delta B\right)_{\rm LMCN~2009a} - 2.5 \log 2.0 \cr
&=& 18.98 - 5.8\pm0.2 - 0.75 = 12.43\pm0.2,
\label{distance_modulus_lv_vul_yy_dor_lmcn_2009a_b}\end{aligned}$$ from Equation (\[distance\_modulus\_general\_temp\_b\]). Here, we adopt $(m-M)_B = 18.49 + 4.1\times 0.12= 18.98$ for the LMC novae. Thus, we obtain $(m-M)_B=12.43\pm0.1$ for LV Vul.
We determine the stretching factor $f_{\rm s}$ by horizontally shifting the light and color curves. Its step is $\delta \log f_{\rm s} = 0.01$ as explained in Section \[reddening\_distance\_method\]. Usually we use three $BVI_{\rm C}$ light curves and $(B-V)_0$ color curve to overlap them with those of the template novae against a unique $f_{\rm s}$. The error of $\log f_{\rm s}$ depends on how well these light/color curves overlap. Typically we have $0.03-0.05$ for the error of $\log f_{\rm s}$. This error is not independent but propagates to the error of vertical fit of $\Delta V$. We estimate the typical total error of $\epsilon (\Delta V - 2.5 \log f_{\rm s})= (\pm 0.2) + (\pm 0.1)$. In the case of LV Vul, $\log f_{\rm s}$ is rather well determined against YY Dor and LMC N 2009a, that is, the error of $\epsilon( \log f_{\rm s}) = \pm0.02$ [see Section 5.1 of @hac18kb]. Therefore, the total error is about $\epsilon (\Delta V - 2.5 \log f_{\rm s})= (\pm 0.2) + (\pm 0.05)$. In what follows, we include the error of the $f_{\rm s}$ determination in the error of vertical fit $\Delta V$, because the vertical fit always reflect the error of $f_{\rm s}$ determination.
We plot these two distance-reddening relations of Equations (\[distance\_modulus\_rv\]) and (\[distance\_modulus\_rb\]) in Figure \[distance\_reddening\_lv\_vul\_v1500\_cyg\_v1668\_cyg\_v1974\_cyg\](a) by the thin solid blue and magenta lines, respectively. The crossing point of the blue, magenta, orange, and vertical red lines, i.e., $d=1.0$ kpc and $E(B-V)=0.60$, is in reasonable agreement with $d=0.92\pm 0.08$ kpc [@sla95] and $E(B-V)=0.60\pm0.05$ [@hac14k; @hac16kb]. Thus, we confirm that the distance modulus in the $V$ band is $(m-M)_V=11.85\pm0.1$, the reddening is $E(B-V)=0.60\pm0.05$, and the distance is $d=1.0\pm0.2$ kpc. These values are listed in Table \[extinction\_various\_novae\].
### V1500 Cyg 1975 {#v1500_cyg}
The distance to V1500 Cyg was discussed by many authors. @you76 estimated the distance to be $1.4 \pm 0.1$ kpc for $E(B-V)= 0.45$ [@tom76] from the distance-reddening law toward the nova (depicted by the filled red circles in Figure \[distance\_reddening\_lv\_vul\_v1500\_cyg\_v1668\_cyg\_v1974\_cyg\](b)). We add Marshall et al.’s (2006) relations for four directions close to V1500 Cyg, that is, $(l, b)=(89\fdg75, 0\fdg00)$ (red open squares), $(90\fdg00, 0\fdg00)$ (green filled squares), $(89\fdg75, -0\fdg25)$ (blue asterisks), and $(90\fdg00, -0\fdg25)$ (magenta open circles). We further add the relations of @gre15 [@gre18], @sal14, @ozd16, and @chen18. Green et al.’s and Sale et al.’s relations (black and green lines) give $d=1.5$ kpc for $E(B-V)=0.45$.
A firm upper limit to the apparent distance modulus was obtained as $(m-M)_V \le 12.5$ by @and76 from the galactic rotational velocities of interstellar H and K absorption lines. The nebular expansion parallax method also gives an estimate of the distance. @bec80 first imaged an expanding nebular ($0\farcs 25$ yr$^{-1}$) of V1500 Cyg and estimated a distance of 1350 pc from the expansion velocity of $v_{\rm exp} = 1600$ km s$^{-1}$. However, @wad91 resolved an expanding nebula and obtained a much lower expansion rate of $0\farcs 16$ yr$^{-1}$; they estimated the distance to be 1.56 kpc, with a much smaller expansion velocity of $v_{\rm exp} = 1180$ km s$^{-1}$ [@coh85]. @sla95 obtained a similar expanding angular velocity of the nebula ($0\farcs16$ yr$^{-1}$) and obtained a distance of 1550 pc from $v_{\rm exp} = 1180$ km s$^{-1}$. Therefore, the distance is reasonably determined to be $d=1.5\pm0.1$ kpc from the nebular expansion parallax method.
@hac14k obtained $(m-M)_V=12.3$ for V1500 Cyg by the time-stretching method together with the light curves of GK Per and V603 Aql (see Figure 40 and Equation (A1) of Paper I). We plot $(m-M)_V=12.3$ (blue solid line) and $E(B-V)=0.45$ (vertical red solid line) in Figure \[distance\_reddening\_lv\_vul\_v1500\_cyg\_v1668\_cyg\_v1974\_cyg\](b). The crossing point of these two lines is consistent with the distance-reddening relations of @gre15 [@gre18], @sal14, @ozd16, and @chen18 except @mar06. Thus, we confirm $d=1.5\pm0.1$ kpc, $E(B-V)=0.45\pm0.05$, and $(m-M)_V=12.3\pm0.1$. The timescaling factor of V1500 Cyg is determined to be $f_{\rm s}=0.60$ against LV Vul. We list these values in Table \[extinction\_various\_novae\].
We check the distance moduli of V1500 Cyg with the LMC novae, YY Dor and LMC N 2009a, based on the time-stretching method. Figure \[v1500\_cyg\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\] shows the $BVI$ light curves of V1500 Cyg as well as YY Dor and LMC N 2009a. Here we use the $I$ band data of V1500 Cyg because no $I_{\rm C}$ data are available. We apply Equation (\[distance\_modulus\_general\_temp\_b\]) for the $B$ band to Figure \[v1500\_cyg\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](a) and obtain $$\begin{aligned}
(m&-&M)_{B, \rm V1500~Cyg} \cr
&=& ((m - M)_B + \Delta B)_{\rm YY~Dor} - 2.5 \log 3.2 \cr
&=& 18.98 - 5.0\pm0.2 - 1.25 = 12.73\pm0.2 \cr
&=& ((m - M)_B + \Delta B)_{\rm LMC~N~2009a} - 2.5 \log 2.0 \cr
&=& 18.98 - 5.5\pm0.2 - 0.75 = 12.73\pm0.2,
\label{distance_modulus_b_v1500_cyg_yy_dor_lmcn2009a}\end{aligned}$$ where we adopt $(m-M)_B = 18.49 + 4.1\times 0.12=18.98$ for the LMC novae. Thus, we obtain $(m-M)_{B, \rm V1500~Cyg}= 12.73\pm0.1$.
For the $V$ light curves in Figure \[v1500\_cyg\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](b), we similarly obtain $$\begin{aligned}
(m&-&M)_{V, \rm V1500~Cyg} \cr
&=& ((m - M)_V + \Delta V)_{\rm YY~Dor} - 2.5 \log 3.2 \cr
&=& 18.86 - 5.3\pm0.2 - 1.25 = 12.31\pm0.2 \cr
&=& ((m - M)_V + \Delta V)_{\rm LMC~N~2009a} - 2.5 \log 2.0 \cr
&=& 18.86 - 5.8\pm0.2 - 0.75 = 12.31\pm0.2,
\label{distance_modulus_v_v1500_cyg_yy_dor_lmcn2009a}\end{aligned}$$ where we adopt $(m-M)_V= 18.49 + 3.1\times 0.12 = 18.86$ for the LMC novae. Thus, we obtain $(m-M)_{V, \rm V1500~Cyg}= 12.31\pm0.1$.
We apply Equation (\[distance\_modulus\_general\_temp\_i\]) to Figure \[v1500\_cyg\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](c) and obtain $$\begin{aligned}
(m&-&M)_{I, \rm V1500~Cyg} \cr
&=& ((m - M)_I + \Delta I_C)_{\rm YY~Dor} - 2.5 \log 3.2 \cr
&=& 18.66 - 5.85\pm0.3 - 1.25 = 11.56\pm 0.3 \cr
&=& ((m - M)_I + \Delta I_C)_{\rm LMC~N~2009a} - 2.5 \log 2.0 \cr
&=& 18.66 - 6.35\pm0.3 - 0.75 = 11.56\pm 0.3,
\label{distance_modulus_i_v1500_cyg_yy_dor_lmcn2009a}\end{aligned}$$ where we adopt $(m-M)_I = 18.49 + 1.5\times 0.12=18.66$ for the LMC novae. Thus, we obtain $(m-M)_{I, \rm V1500~Cyg}= 11.56\pm0.2$.
We plot three distance moduli in the $B$, $V$, and $I_{\rm C}$ bands in Figure \[distance\_reddening\_lv\_vul\_v1500\_cyg\_v1668\_cyg\_v1974\_cyg\](b) by the magenta, blue, and cyan lines, that is, $(m-M)_B= 12.73$, $(m-M)_V= 12.31$, and $(m-M)_I= 11.56$, together with Equations (\[distance\_modulus\_rb\]), (\[distance\_modulus\_rv\]), and (\[distance\_modulus\_ri\]), respectively. These three lines cross at $d=1.5$ kpc and $E(B-V)=0.45$. This crossing point is an independent confirmation of the distance and reddening.
### V1668 Cyg 1978 {#v1668_cyg}
Figure \[v1668\_cyg\_lv\_vul\_u\_b\_v\_logscale\_3fig\] shows the $UBV$ light curves of V1668 Cyg and LV Vul. The timescale of V1668 Cyg is the same as LV Vul, that is, $f_{\rm s}=1.0$ against LV Vul. If the chemical composition of V1668 Cyg is similar to that of LV Vul, both the WD masses should be similar. The peak of V1668 Cyg is slightly brighter than that of LV Vul. In general, the peak brightness depends mainly on the WD mass and initial envelope mass. The peak magnitude is brighter for the more massive ignition mass even if the WD masses are the same. Such a tendency was discussed in detail in @hac10k (see their Figure 2 for the physical explanation). The initial envelope mass, that is, the ignition mass, is closely related to the mass accretion rate on the WD. The larger the mass accretion rate, the smaller the ignition mass [see, e.g., Figure 3 of @kat14shn]. We suppose that the mass accretion rate is smaller in V1668 Cyg than in LV Vul even if the two WD masses are similar.
These two nova light curves well overlap except for the peak brightnesses. The distance moduli of LV Vul are well determined in Section \[lv\_vul\], so the distance moduli of V1668 Cyg are calculated to be $(m-M)_U = 12.84 + 2.25\pm0.1 = 15.1\pm0.1$, $(m-M)_B = 12.45 + 2.45\pm0.1 = 14.9\pm0.1$, and $(m-M)_V = 11.85 + 2.75\pm0.1 = 14.6\pm0.1$ from the time-stretching method. These three distance moduli cross at $d= 5.4$ kpc and $E(B-V)= 0.30$ in the distance-reddening plane of Figure \[distance\_reddening\_lv\_vul\_v1500\_cyg\_v1668\_cyg\_v1974\_cyg\](c).
@hac06kb [@hac16k] calculated free-free emission model light curves based on the optically thick wind model [@kat94h] for various WD masses and chemical compositions and fitted their free-free emission model light curves with the $V$ and $y$ light curves of V1668 Cyg. They further fitted their blackbody emission UV 1455Å model light curves with the [*International Ultraviolet Explore (IUE)*]{} UV 1455Å observation, and estimated the WD mass. This is a narrow-band (1445–1465 Å) flux that represents well the continuum UV fluxes of novae [@cas02].
Assuming the chemical composition of CO nova 3 for V1668 Cyg, @hac16k obtained the best fit model of $0.98~M_\sun$ WD with both the $V$ and UV 1455 Å light curves. (Such model light curve fittings are also plotted in Figure \[v679\_car\_lv\_vul\_v1668\_cyg\_v\_bv\_ub\_logscale\] in Appendix \[v679\_car\].) We plot the distance-reddening relation of $(m-M)_V=14.6\pm0.1$ (thick solid blue line flanked by two thin blue lines in Figure \[distance\_reddening\_lv\_vul\_v1500\_cyg\_v1668\_cyg\_v1974\_cyg\](c)) and another distance-reddening relation of the UV 1455Å fit (thick solid magenta line flanked by two thin solid magenta lines) of $$\begin{aligned}
& & 2.5 \log F_{1455}^{\rm mod}
- 2.5 \log F_{1455}^{\rm obs} \cr
&=& 5 \log \left({d \over {10\mbox{~kpc}}} \right) + 8.3 \times E(B-V),
\label{qu_vul_uv1455_fit_eq2}\end{aligned}$$ where $F_{1455}^{\rm mod}$ is the model flux at the distance of 10 kpc and $F_{1455}^{\rm obs}$ the observed flux. Here we assume an absorption of $A_\lambda= 8.3 \times E(B-V)$ at $\lambda= 1455$ Å [@sea79].
We further plot other distance-reddening relations toward V1668 Cyg in Figure \[distance\_reddening\_lv\_vul\_v1500\_cyg\_v1668\_cyg\_v1974\_cyg\](c): given by @slo79 (filled red circles), by @mar06, by @gre15 [@gre18], and by @chen18. Marshall et al.’s are for four nearby directions: $(l, b)= (90\fdg75,-6\fdg75)$ (red open squares), $(91\fdg00,-6\fdg75)$ (green filled squares), $(90\fdg75,-7\fdg00)$ (blue asterisks), and $(91\fdg00,-7\fdg00)$ (magenta open circles). These trends/lines broadly cross at $d=5.4\pm0.5$ kpc and $E(B-V)= 0.30\pm0.05$ except that of @chen18. The extinction almost saturates at the distance of $d\gtrsim 3$ kpc as shown in Figure \[distance\_reddening\_lv\_vul\_v1500\_cyg\_v1668\_cyg\_v1974\_cyg\](c). This is consistent with the galactic 2D dust absorption map of $E(B-V)=0.29 \pm 0.02$ in the direction toward V1668 Cyg at the NASA/IPAC Infrared Science Archive[^1], which is calculated on the basis of data from @schl11. Thus, we confirm $E(B-V)= 0.30\pm0.05$.
### V1974 Cyg 1992 {#v1974_cyg}
We obtain the reddening and distance from the time-stretching method. We plot the $UBV$ light curves of V1974 Cyg together with those of LV Vul and V1668 Cyg in Figure \[v1974\_cyg\_v1668\_cyg\_lv\_vul\_u\_b\_v\_logscale\_3fig\]. The $UBV$ light curves well overlap until the nebular phase started on Day $\sim 60$. We use only the part of optically thick wind (or optically thick ejecta) phase, which are represented by the decline trend of $F_\nu \propto t^{-1.75}$ [@hac06kb]. The timescaling factor of $f_{\rm s}= 1.07$ against LV Vul is well determined from the fitting of UV 1455Å light curve between V1668 Cyg and V1974 Cyg (see, e.g., Figure \[v382\_vel\_lv\_vul\_v1668\_cyg\_v1974\_cyg\_v\_bv\_ub\_color\_logscale\](a) in Appendix \[v382\_vel\]).
We apply Equation (\[distance\_modulus\_general\_temp\_u\]) for the $U$ band to Figure \[v1974\_cyg\_v1668\_cyg\_lv\_vul\_u\_b\_v\_logscale\_3fig\](a) and obtain $$\begin{aligned}
(m&-&M)_{U, \rm V1974~Cyg} \cr
&=& ((m - M)_U + \Delta U)_{\rm LV~Vul} - 2.5 \log 1.07 \cr
&=& 12.85 - 0.1\pm0.2 - 0.08 = 12.67\pm0.2 \cr
&=& ((m - M)_U + \Delta U)_{\rm V1668~Cyg} - 2.5 \log 1.07 \cr
&=& 15.1 - 2.3\pm0.2 - 0.08 = 12.72\pm0.2,
\label{distance_modulus_u_v1974_cyg_lv_vul_v1668_cyg}\end{aligned}$$ where we adopt $(m-M)_{U, \rm LV~Vul}=11.85 + (4.75-3.1) \times 0.60
=12.85$, and $(m-M)_{U, \rm V1668~Cyg}=14.6 + (4.75-3.1) \times 0.30
=15.10$. Thus, we obtain $(m-M)_{U, \rm V1974~Cyg}= 12.7\pm0.1$. For the $B$ light curves in Figure \[v1974\_cyg\_v1668\_cyg\_lv\_vul\_u\_b\_v\_logscale\_3fig\](b), we similarly obtain $$\begin{aligned}
(m&-&M)_{B, \rm V1974~Cyg} \cr
&=& ((m - M)_B + \Delta B)_{\rm LV~Vul} - 2.5 \log 1.07 \cr
&=& 12.45 + 0.1\pm0.2 - 0.08 = 12.47\pm0.2 \cr
&=& ((m - M)_B + \Delta B)_{\rm V1668~Cyg} - 2.5 \log 1.07 \cr
&=& 14.9 - 2.3\pm0.2 - 0.08 = 12.52\pm0.2,
\label{distance_modulus_b_v1974_cyg_lv_vul_v1668_cyg}\end{aligned}$$ where we adopt $(m-M)_{B, \rm LV~Vul}= 11.85 + 1.0\times 0.6 =12.45$ and $(m-M)_{B, \rm V1668~Cyg}= 14.6 + 1.0\times 0.3 =14.9$. Thus, we obtain $(m-M)_{B, \rm V1974~Cyg}= 12.5\pm0.1$. We apply Equation (\[distance\_modulus\_general\_temp\]) to Figure \[v1974\_cyg\_v1668\_cyg\_lv\_vul\_u\_b\_v\_logscale\_3fig\](c) and obtain $$\begin{aligned}
(m&-&M)_{V, \rm V1974~Cyg} \cr
&=& ((m - M)_V + \Delta V)_{\rm LV~Vul} - 2.5 \log 1.07 \cr
&=& 11.85 + 0.4\pm0.2 - 0.08 = 12.17\pm0.2 \cr
&=& ((m - M)_V + \Delta V)_{\rm V1668~Cyg} - 2.5 \log 1.07 \cr
&=& 14.6 - 2.3\pm0.2 - 0.08 = 12.22\pm0.2.
\label{distance_modulus_v_v1974_cyg_lv_vul_v1668_cyg}\end{aligned}$$ Thus, we obtain $(m-M)_{V, \rm V1974~Cyg}= 12.2\pm0.1$. These three distance moduli, that is, $(m-M)_U= 12.7$, $(m-M)_B= 12.5$, and $(m-M)_V= 12.2$ together with Equations (\[distance\_modulus\_ru\]), (\[distance\_modulus\_rb\]), and (\[distance\_modulus\_rv\]), cross at $d=1.8$ kpc and $E(B-V)=0.30$ as shown in Figure \[distance\_reddening\_lv\_vul\_v1500\_cyg\_v1668\_cyg\_v1974\_cyg\](d).
@hac16k modeled the $V$, UV 1455Å, and supersoft X-ray light curves and fitted a $0.98~M_\sun$ WD model with the observed $V$, UV 1455Å, and X-ray fluxes assuming the chemical composition of CO nova 3 (see, e.g., Figure \[v533\_her\_lv\_vul\_v1974\_cyg\_v\_bv\_ub\_logscale\](a) for such a model light curve fit). The supersoft X-ray flux was calculated assuming that the photospheric emission is approximated by blackbody emission in the supersoft X-ray band. The $V$ model light curve gives a distance modulus in the $V$ band of $(m-M)_V=12.2$ (blue line) as shown in Figure \[distance\_reddening\_lv\_vul\_v1500\_cyg\_v1668\_cyg\_v1974\_cyg\](d), while the UV 1455 Å light curve fit gives a relation of magenta line.
These two lines (blue and magenta lines) cross at $d\approx 1.8$ kpc and $E(B-V)\approx 0.29$ as shown in Figure \[distance\_reddening\_lv\_vul\_v1500\_cyg\_v1668\_cyg\_v1974\_cyg\](d). The figure also depicts other distance-reddening relations toward V1974 Cyg, whose galactic coordinates are $(l, b)=(89\fdg1338, 7\fdg8193)$. The four sets of data points with error bars correspond to the distance-reddening relations in four directions close to V1974 Cyg: $(l, b)=(89\fdg00,7\fdg75)$ (red open squares), $(89\fdg25, 7\fdg75)$ (green filled squares), $(89\fdg00, 8\fdg00)$ (blue asterisks), and $(89\fdg25, 8\fdg00)$ (magenta open circles), the data for which are taken from @mar06. We also add distance-reddening relations given by @gre15 [@gre18] (black/orange lines), @ozd16 (open cyan-blue diamonds with error bars), and @chen18 (cyan-blue line). The orange line is consistent with the crossing point at $d\approx 1.8$ kpc and $E(B-V)\approx 0.30$.
@cho97a estimated the distance to V1974 Cyg as $d=1.77\pm 0.11$ kpc by a nebular expansion parallax method. This distance is consistent with our value of $d=1.8\pm 0.1$ kpc. The reddening was also estimated by many researchers. For example, @aus96 obtained the reddening toward V1974 Cyg mainly on the basis of the UV and optical line ratios for days 200 through 500, i.e., $E(B-V)=0.3 \pm 0.1$. The NASA/IPAC galactic 2D dust absorption map gives $E(B-V)=0.35 \pm 0.01$ in the direction toward V1974 Cyg. This slightly larger value of reddening indicates that the reddening toward V1974 Cyg does not saturate yet at the distance of 1.8 kpc, as shown in Figure \[distance\_reddening\_lv\_vul\_v1500\_cyg\_v1668\_cyg\_v1974\_cyg\](d). These are all consistent with our estimates, that is, $(m-M)_V=12.2\pm0.1$, $d=1.8\pm 0.1$ kpc, and $E(B-V)=0.3 \pm 0.05$. These values are listed in Table \[extinction\_various\_novae\].
Template nova tracks in the time-stretched color-magnitude diagram {#template_tracks}
------------------------------------------------------------------
@hac16kb proposed six template tracks of novae with well-determined distance modulus in the $V$ band $(m-M)_V$ and color excess $E(B-V)$. They dubbed them the V1500 Cyg, V1668 Cyg, V1974 Cyg, LV Vul, FH Ser, and PU Vul tracks as shown in Figure \[hr\_diagram\_6types\_novae\_stretch.epsi\](a). Here, we adopt $E(B-V)=0.60$ and $(m-M)_V=11.85$ for LV Vul from Section \[lv\_vul\], $E(B-V)=0.45$ and $(m-M)_V=12.3$ for V1500 Cyg from Section \[v1500\_cyg\], $E(B-V)=0.30$ and $(m-M)_V=14.6$ for V1668 Cyg from Section \[v1668\_cyg\], $E(B-V)=0.30$ and $(m-M)_V=12.2$ for V1974 Cyg from Section \[v1974\_cyg\], $E(B-V)=0.60$ and $(m-M)_V=11.7$ for FH Ser and $E(B-V)=0.30$ and $(m-M)_V=14.3$ for PU Vul both from @hac16kb. They categorized 40 novae into one of these six subgroups, depending on their similarity in the $(B-V)_0$-$M_V$ diagram. The LV Vul and V1974 Cyg tracks split into two branches just after the nebular phase started. As already discussed in our previous papers [see, e.g., Figure 10 and Section 3.3 of @hac14k], strong emission lines such as \[\] contribute to the blue edge of $V$ filter. A small difference in the $V$ filter response function makes a large difference in the $V$ magnitude and results in a large difference in the $B-V$ color. This is the reason why the track splits into two (or three or even four) branches among various observatories.
In the present work, we propose the $(B-V)_0$-$(M_V-2.5\log f_{\rm s})$ diagram, that is, Figure \[hr\_diagram\_6types\_novae\_stretch.epsi\](b). In this new color-magnitude diagram, the above six nova tracks are shifted up or down. Here, we show the above six tracks of V1500 Cyg, V1974 Cyg, V1668 Cyg, LV Vul, FH Ser, and PU Vul, adopting $(m-M')_V=11.75$ for V1500 Cyg, $(m-M')_V=12.3$ for V1974 Cyg, $(m-M')_V=11.85$ for LV Vul. $(m-M')_V=14.6$ for V1668 Cyg, $(m-M')_V=12.45$ for FH Ser, and $(m-M')_V=16.4$ for PU Vul.
The tracks of V1500 Cyg and V1974 Cyg are different from each other in the early phase but almost overlap in the middle part of the tracks. They turn from toward the blue to toward the red at the similar place, that is, the turning corner. Therefore, we regard that V1974 Cyg belongs to the V1500 Cyg type in the time-stretched color-magnitude diagram.
Also, the V1668 Cyg track almost follows the template track of LV Vul. Thus, these two template tracks can be merged into the same group. The track of FH Ser is close to that of V1668 Cyg and LV Vul. The track of PU Vul is close to that of V1500 Cyg. Thus, the six tracks seem to closely converge into one of the two groups, LV Vul/V1668 Cyg or V1500 Cyg/V1974 Cyg, in the time-stretched color-magnitude diagram.
The main difference between them is that the V1668 Cyg/LV Vul group novae evolve almost straight down along $(B-V)_0\sim -0.03$ during $(M_V-2.5\log f_{\rm s})= -7$ and $-4$, while the V1500 Cyg/V1974 Cyg group novae evolve blueward up to $(B-V)_0\sim -0.5$ during $(M_V-2.5\log f_{\rm s})= -5$ and $-3$. This tendency is clear, e.g., in the case of V2362 Cyg (Figure \[hr\_diagram\_v705\_cas\_v382\_vel\_v5114\_sgr\_v2362\_cyg\_outburst\](d)).
V574 Pup 2004 {#v574_pup_cmd}
=============
@hac10k analyzed the optical and X-ray light curves of V574 Pup. They assumed $A_V=2.2$ after @burl08, which will be revised later. @hac15k determined the absolute magnitude of theoretical $V$ light curve for various WD masses and chemical compositions. Using their theoretical light curves, we reexamine the light/color curves of V574 Pup,
@nes07a obtained $E(B-V)=0.50$ during the SSS phase from the relation of $E(B-V)=N_{\rm H}/4.8\times 10^{21}$ cm$^{-2}$ [@boh78] together with their estimate of $N_{\rm H}= (2.5\pm0.6)
\times 10^{21}$ cm$^{-2}$ calculated with an X-ray model spectrum fitting. Here, $N_{\rm H}$ is the neutral hydrogen column density. @nai10 derived the distance of $d=5.5$ kpc from the maximum magnitude versus rate of decline (MMRD) relation of @del95 and obtained the reddening of $E(B-V)=A_V/3.1= 1.95/3.1= 0.63$ from the empirical relation of $(B-V)_{0, \rm max}=+0.23\pm0.06$ at optical $V$ maximum [@van87].
Distance-reddening relations toward V574 Pup {#distance_reddening_v574_pup}
--------------------------------------------
We obtain the distance and reddening toward V574 Pup based on the time-stretching method. Figure \[distance\_reddening\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\](a) shows four distance moduli in the $B$, $V$, $I_{\rm C}$, and $K_{\rm s}$ bands by the thin solid magenta, blue, cyan, cyan-blue lines, that is, $(m-M)_B= 15.43$, $(m-M)_V= 15.0$, $(m-M)_I= 14.22$, and $(m-M)_K= 13.68$, which are obtained in Appendix \[v574\_pup\], together with Equations (\[distance\_modulus\_rb\]), (\[distance\_modulus\_rv\]), (\[distance\_modulus\_ri\]), and (\[distance\_modulus\_rk\]). These four lines cross at $d=5.3$ kpc and $E(B-V)=0.45$. Thus, we obtain the reddening toward V574 Pup, $E(B-V)=0.45\pm0.05$, the distance of V574 Pup, $d=5.3\pm0.5$ kpc, and the timescaling factor, $f_{\rm s}=1.26$ against LV Vul. The distance modulus in the $V$ band is $(m-M)_V=15.0\pm0.1$. These values are listed in Table \[extinction\_various\_novae\].
We further check our estimates of the distance modulus in the $V$ band $(m-M)_V=15.0\pm0.1$ and extinction $E(B-V)=0.45\pm0.05$. Figure \[distance\_reddening\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\](a) shows the distance-reddening relations toward V574 Pup, whose galactic coordinates are $(l,b)=(242\fdg5695, -1\fdg9933)$. The thick solid black/orange lines denote the distance-reddening relations given by @gre15 [@gre18], respectively. The open cyan-blue diamonds with error bars are taken from @ozd16. The thick solid cyan-blue line denotes the result of @chen18. These four distance-reddening relations based on the dust map are consistent with each other until $d \lesssim 5$ kpc.
Green et al.’s black/orange lines are roughly consistent with our results of $d=5.3\pm0.5$ kpc and $E(B-V)=0.45\pm0.05$. The NASA/IPAC galactic 2D dust absorption map gives $E(B-V)=0.60\pm0.02$ toward V574 Pup, which is slightly larger than our value of $E(B-V)=0.45\pm0.05$. This is because the reddening increases further beyond the position of V574 Pup as shown in Figure \[distance\_reddening\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\](a).
Color-magnitude diagram {#cmd_v574_pup}
-----------------------
Using $E(B-V)=0.45$ and $(m-M)_V=15.0$, we plot the color-magnitude diagram of V574 Pup in Figure \[hr\_diagram\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\_outburst\_mv\](a). We add tracks of V1500 Cyg (solid green line), LV Vul (solid orange line), and V1974 Cyg (solid magenta line) for comparison. The data of V574 Pup are scattered in the early phase owing to its oscillatory behavior but close to that of V1974 Cyg (solid magenta line). Therefore, V574 Pup belongs to the V1974 Cyg type in the $(B-V)_0$-$M_V$ color-magnitude diagram. The track in the early oscillatory phase shows a circular movement similar to that of V705 Cas as shown in Figure 37(b) of Paper II.
In Figure \[hr\_diagram\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\_outburst\](a), adopting $E(B-V)=0.45$ and $(m-M')_V=15.25$ in Equation (\[absolute\_mag\_lv\_vul\_v574\_pup\_only\]), we obtain the time-stretched color-magnitude diagram of V574 Pup. The text “$(m-M')_V=15.25(-0.25)$” in the figure means that $(m-M')_V=15.25$ and $(m-M)_V=15.25 - 0.25 = 15.0$.
We note the starting epoch of the nebular phase in Figures \[hr\_diagram\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\_outburst\_mv\](a) and \[hr\_diagram\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\_outburst\](a). A nebular phase could be identified by the first clear appearance of the nebular emission lines \[\] or \[\] which are stronger than permitted lines. The nebular phase of V574 Pup started at $m_V\sim12$ from the spectra of the Small Medium Aperture Telescope System (SMARTS) database[^2] [@wal12], which is denoted by the large open cyan square in Figures \[hr\_diagram\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\_outburst\_mv\](a) and \[hr\_diagram\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\_outburst\](a). After the onset of the nebular phase, the track turns to the right (red). This is because the strong emission lines of \[\] contribute to the blue edge of the $V$ filter and its large contribution makes the $B-V$ color redder.
@nes07a discussed that the SSS phase started between May and July in 2005. This epoch corresponds to $m_V \sim 13.2$ and denoted by “SSS on” in Figures \[hr\_diagram\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\_outburst\_mv\](a) and \[hr\_diagram\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\_outburst\](a). The $(B-V)_0$ color seems to keep a constant value of $\sim 0.0$ during the SSS phase, as shown in Figure \[v574\_pup\_v1974\_cyg\_v\_bv\_logscale\](b).
In Figure \[hr\_diagram\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\_outburst\](a), the track of V574 Pup almost follows those of V1500 Cyg and V1974 Cyg except for the early oscillation phase. Thus, we conclude that V574 Pup belongs to the V1500 Cyg/V1974 Cyg type in the time-stretched color-magnitude diagram.
Model light curve fitting {#model_light_curve}
-------------------------
@kat94h calculated evolutions of nova outbursts based on the optically thick wind theory. Their numerical models provide various physical quantities such as the photospheric temperature, radius, velocity, and wind mass-loss rate of a nova hydrogen-rich envelope (mass of $M_{\rm env}$) for a specified WD mass ($M_{\rm WD}$) and chemical composition of the envelope. @hac15k calculated the absolute $V$ magnitude light curves based on Kato & Hachisu’s envelope models. Their model $V$ light curve is composed of photospheric emission and free-free emission. The photospheric emission is approximated by blackbody emission at the photosphere, while the free-free emission is calculated from the photospheric radius, velocity, and wind mass-loss rate [@hac15k]. Fitting the absolute $V$ magnitudes ($M_V$) of model light curve with the observed apparent $V$ magnitudes ($m_V$), we are able to specify the $(m-M)_V$ (and even $M_{\rm WD}$) for the nova. Such results are presented in @hac15k [@hac16k; @hac18k; @hac18kb].
@hac10k estimated the WD mass of V574 Pup to be $1.05~M_\sun$. In Figure \[v574\_pup\_v1974\_cyg\_v\_bv\_logscale\](a), we plot the $1.05~M_\sun$ WD model (solid black line for $V$ magnitudes and solid red line for soft X-ray fluxes) with the chemical composition of Ne nova 3 [@hac16k]. We use the absolute $V$ magnitudes of the $1.05~M_\sun$ WD model calculated in @hac16k. We add a $0.98~M_\sun$ WD model (solid green lines both for $V$ and soft X-ray) with the chemical composition of CO nova 3 [@hac16k] for comparison, the timescale of which is stretched by a factor of $f_{\rm s}= 1.17$. The absolute $V$ light curve of the $1.05~M_\sun$ WD model reasonably follows the observed apparent $V$ light curves of V574 Pup for $(m-M)_V=15.0$. In these models, the optically thick winds stop at the open black circle (at the right edge of the solid black line).
The model supersoft X-ray light curve also shows good fit with the X-ray observation, i.e., turn-on and turnoff times. The X-ray data are the same as those of @hac10k. Thus, the $1.05~M_\sun$ WD model is consistent with the X-ray data. We list the WD mass of V574 Pup in Table \[wd\_mass\_novae\].
V679 Car 2008 {#v679_car_cmd}
=============
V679 Car was discovered by Katarzyna Malek at mag 7.9 on UT 2008 November 26.26 (JD 2454795.76). It reached mag 7.55 on UT 2008 November 27.27 (JD 2454796.77) [@waa08]. The spectrum showed that the object is a classical nova of the type Recently, @fra18 suggested that V679 Car is a candidate for GeV gamma-ray detected nova with [*Fermi*]{}/Large Area Telescope (LAT).
Distance-reddening relations toward V679 Car {#distance_reddening_v679_car}
--------------------------------------------
We plot three distance moduli in the $B$, $V$, and $I_{\rm C}$ bands in Figure \[distance\_reddening\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\](b) by the magenta, blue, and cyan lines, that is, $(m-M)_B= 16.8$, $(m-M)_V= 16.1$, and $(m-M)_I= 15.0$ in Appendix \[v679\_car\] together with Equations (\[distance\_modulus\_rb\]), (\[distance\_modulus\_rv\]), and (\[distance\_modulus\_ri\]), respectively. These three lines cross at $d=6.2$ kpc and $E(B-V)=0.69$. Thus, we obtain the reddening toward V679 Car, $E(B-V)=0.69\pm0.05$, the distance of V574 Pup, $d=6.2\pm0.7$ kpc, and the timescaling factor, $f_{\rm s}=1.0$ against LV Vul. The distance modulus in the $V$ band is $(m-M)_V=16.1\pm0.1$. These values are listed in Table \[extinction\_various\_novae\].
We check the distance and reddening toward V679 Car, whose galactic coordinates are $(l,b)=(291\fdg4697, -0\fdg5479)$, on the basis of dust extinction maps calculated by @mar06 and @chen18. Figure \[distance\_reddening\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\](b) shows the relations given by @mar06: $(l, b)=(291\fdg25, -0\fdg50)$ by open red squares, $(291\fdg50, -0\fdg50)$ by filled green squares, $(291\fdg25, -0\fdg75)$ by blue asterisks, and $(291\fdg50, -0\fdg75)$ by open magenta circles. The filled green squares of @mar06 are the closest direction toward V679 Car among the four nearby directions. The solid cyan-blue line denotes the relation given by @chen18. Neither Green et al.’s (2015, 2018) nor Özdörmez et al.’s (2016) data are available toward this direction. The three relations of $(m-M)_B= 16.8$ (magenta line), $(m-M)_V= 16.1$ (blue line), and $(m-M)_I= 15.0$ (cyan line) cross at $d=6.2$ kpc and $E(B-V)=0.69$. This crossing point is significantly different from the closest relation (filled green squares) given by @mar06. However, Chen et al.’s relation (solid cyan-blue line) seems to show possible agreement with our crossing point, although the data end at $d=4.5$ kpc.
This large discrepancy can be understood as follows: Marshall et al.’s 3D dust map gives an averaged value of a relatively broad region, and thus a pinpoint reddening toward the nova could be different from the values of the 3D dust maps, because the resolution of the dust map is considerably larger than molecular cloud structures observed in the interstellar medium. For example, Marshall et al.’s 3D dust map gave a significantly different value for the reddening of LV Vul as shown in Figure \[distance\_reddening\_lv\_vul\_v1500\_cyg\_v1668\_cyg\_v1974\_cyg\](a).
Color-magnitude diagram {#cmd_v679_car}
-----------------------
The $V$ light and $(B-V)_0$ color curve shapes of V679 Car are similar to those of LV Vul and V1668 Cyg as shown in Figures \[v679\_car\_lv\_vul\_v1668\_cyg\_b\_v\_logscale\_2fig\] and \[v679\_car\_lv\_vul\_v1668\_cyg\_v\_bv\_ub\_logscale\]. This suggests that the color-magnitude diagram of V679 Car is also similar to them. Adopting $E(B-V)=0.69$ and $(m-M)_V=16.1$, we plot the color-magnitude track in Figure \[hr\_diagram\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\_outburst\_mv\](b) together with V1500 Cyg (solid green line) and LV Vul (solid orange line). The SMARTS spectra of V679 Car show that the nebular phase started between JD 2454845.71 (UT 2009 January 14.21) and JD 2454860.67 (UT 2009 January 29.17), corresponding to $m_V=12.45$ and $B-V=0.40$, which is denoted by the large open red square in Figure \[hr\_diagram\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\_outburst\_mv\](b). The start of the nebular phase usually corresponds to the turning point from toward blue to toward red (from toward left to toward right) as already discussed in Section \[v574\_pup\_cmd\]. The track of V679 Car broadly locates on the track of LV Vul (orange line), although the $(B-V)_0$ colors of the early phase are rather scattered and slightly bluer than that of LV Vul. After this turning point, the track of V679 Car follows that of LV Vul. We conclude that V679 Car belongs to the LV Vul type in the color-magnitude diagram (Paper II).
Adopting $E(B-V)=0.69$ and $(m-M')_V=16.1$ from Equation (\[absolute\_mag\_lv\_vul\_v679\_car\_only\]), we plot the time-stretched color-magnitude diagram of V679 Car in Figure \[hr\_diagram\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\_outburst\](b). The $(B-V)_0$ intrinsic color is not affected by time-stretch [see @hac18k]. Because $f_{\rm s}=1.0$, the time-stretched color-magnitude track is the same as in Figure \[hr\_diagram\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\_outburst\_mv\](b), but the track of V1500 Cyg is shifted down. The overlap of the track to the LV Vul track supports our values of $E(B-V)=0.69$ and $(m-M')_V=16.1$, that is, $f_{\rm s}=1.0$, $E(B-V)=0.69\pm0.05$, $(m-M)_V=16.1\pm0.1$, and $d=6.2\pm0.7$ kpc. We list our results in Table \[extinction\_various\_novae\].
Model light curve fitting with V679 Car {#model_lc_v679_car}
---------------------------------------
Figure \[v679\_car\_lv\_vul\_v1668\_cyg\_v\_bv\_ub\_logscale\](a) shows the model light curve of a $0.98~M_\sun$ WD (solid red lines) with the envelope chemical composition of CO Nova 3 [@hac16k]. Taking the distance modulus in the $V$ band of $(m-M)_V=16.1$, we reasonably fit the model absolute $V$ light curve with the observed apparent $V$ magnitudes of V679 Car. Thus, our obtained value of $(m-M)_V=16.1$ is consistent with the model light curve. We also add the $0.98~M_\sun$ WD model (solid green lines) with the same envelope chemical composition of CO nova 3 but a slightly larger initial envelope mass, assuming $(m-M)_V=14.6$ for V1668 Cyg. This model fits with the $V$ and UV 1455Å light curves of V1668 Cyg as already discussed in our previous paper [@hac16k].
Our model light curve of the $0.98~M_\sun$ WD predicts the SSS phase between day 250 and day 600 as shown in Figure \[v679\_car\_lv\_vul\_v1668\_cyg\_v\_bv\_ub\_logscale\](a). V679 Car was observed with [*Swift*]{} eight times, but no SSS phase was detected [@schw11]. The epochs of the [*Swift*]{} observations are indicated by the downward magenta arrows. The second and third arrows almost overlap. We expect detection of supersoft X-rays at the last one or two. Non-detection may be owing to a large extinction toward V679 Car, as large as $E(B-V)=0.69$, corresponding to $N_{\rm H}=
8.3\times 10^{21} E(B-V) = 0.6 \times 10^{22}$ cm$^{-2}$ [@lis14], or see also Figure 3 in @schw11,
V1369 Cen 2013 {#v1369_cen_cmd}
==============
V1369 Cen was discovered by J. Seach at mag 5.5 on UT 2013 December 2.692 (JD 2456629.192) [@gui13] and reached $m_{V,\rm max}\approx3.0$ on JD 2456639.18 (VSOLJ data by S. Kiyota, however, we omit his $m_V=2.8$ on JD 2456637.21 because it is too bright compared with other data). @mas18 obtained the reddening to be $E(B-V)=0.15$ from the D2 $\lambda 5890$ interstellar absorption line and the best fit column density of $N_{\rm H}=7.2\times 10^{21}$ cm$^{-2}$ derived from the model of the interstellar Ly$\alpha$ absorption. This value of $N_{\rm H}$ would be a typo of $N_{\rm H}=7.2\times 10^{20}$ cm$^{-2}$, because of $E(B-V)=N_{\rm H}/4.8\times 10^{21}$ cm$^{-2}$ [@boh78].
This nova is characterized by pre-nova X-ray observation and GeV gamma-ray detection. @kuu13 reported a pre-nova X-ray detection between UT 2002 February 20 16:57 and 21:08 with an absorbed 0.3-10 keV flux of $(1.0\pm0.1)\times 10^{-12}$ erg s$^{-1}$cm$^2$ and $N_{\rm H}=(2\pm1)
\times 10^{21}$ cm$^{-2}$ for a non-isothermal collisional plasma. @che13 reported a gamma-ray detection during UT 2013 December 7-10 with an average flux of $F(E>100$ MeV$)\sim
(2.1\pm0.6) \times 10^{-7}$ ph cm$^{-2}$ s$^{-1}$.
Distance and reddening toward V1369 Cen {#distance-reddening_relation_v1369_cen}
---------------------------------------
We obtain the distance and reddening toward the nova based on the time-stretching method. We plot four distance moduli in the $B$, $V$, $I_{\rm C}$, and $K_{\rm s}$ bands in Figure \[distance\_reddening\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\](c), which are obtained in Appendix \[v1369\_cen\]. These four (magenta, blue, cyan, cyan-blue) lines cross at the distance of $0.96$ kpc and the reddening of $E(B-V)=0.11$. Thus, we obtain the distance $d= 0.96\pm0.1$ kpc, the reddening $E(B-V)= 0.11\pm0.03$, and the timescaling factor $f_{\rm s}= 1.48$ against LV Vul.
@izz13 obtained an extinction of $E(B-V) = 0.11\pm0.08$ from the relation given by @poz12 or $E(B-V) = 0.14$ from the relation given by @mun97, both between the widths of D doublet and the extinction. @sho14 obtained the extinction of $E(B-V)\sim0.1$, using UV interstellar spectra. They also estimated the distance of $d\sim 2.4$ kpc comparing the UV fluxes with those of V339 Del. As already mentioned, @mas18 obtained the reddening of $E(B-V)=0.15$ from the D2 $\lambda 5890$ interstellar absorption line and the best fit neutral hydrogen column density derived from the model of the interstellar Ly$\alpha$ absorption.
We further examine the reddening and distance based on various distance-reddening relations. Figure \[distance\_reddening\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\](c) shows the distance-reddening relations toward V1369 Cen, whose galactic coordinates are $(l,b)=(310\fdg9816, +2\fdg7274)$. We plot Marshall et al.’s (2006) relations: $(l, b)=(310\fdg75, +2\fdg50)$ by open red squares, $(311\fdg00, +2\fdg50)$ by filled green squares, $(310\fdg75, +2\fdg75)$ by blue asterisks, and $(311\fdg00, +2\fdg75)$ by open magenta circles. The open magenta circles are the closest direction among the four nearby directions. Green et al.’s or Özdörmez et al.’s data are not available. We add the relation (cyan-blue line) given by @chen18, which is consistent with our crossing point of $E(B-V)=0.11$ and $d=0.96$ kpc. Although Marshall et al.’s relations do not reach $E(B-V)\lesssim 0.2$, it seems that its trend of linear extension toward $E(B-V)= 0.11$ are roughly consistent with our crossing point at $d=0.96\pm0.1$ kpc and $E(B-V)=0.11\pm0.03$.
Color-magnitude diagram {#color-magnitude-diagram}
-----------------------
Adopting $E(B-V)=0.11$ and $(m-M)_V=10.25$, we plot the color-magnitude diagram of V1369 Cen in Figure \[hr\_diagram\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\_outburst\_mv\](c). We also plot the tracks of V1500 Cyg (solid green line), LV Vul (solid orange line), and V496 Sct (filled cyan triangles). The data of V496 Sct are the same as those in Figures 72(b) and 73 of Paper II. The track of V1369 Cen and V496 Sct almost follow that of LV Vul in the early phase until the dust-shell formation, which is denoted by the large open blue square. After that, the track of V1369 Cen underlies the track of LV Vul.
Figure \[hr\_diagram\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\_outburst\](c) shows the time-stretched color-magnitude diagram of V1369 Cen. We adopt $E(B-V)=0.11$ and $(m-M')_V=10.65$ from Equation (\[absolute\_mag\_v1369\_cen\_lv\_vul\]). Now, the track of V1369 Cen and V496 Sct almost follow that of LV Vul. Thus, we conclude that V1369 Cen belongs to the LV Vul type in the time-stretched color-magnitude diagram. This overlapping of V1369 Cen track to that of LV Vul suggests that our adopted values of $E(B-V)=0.11$, $f_{\rm s}=1.48$, and $(m-M')_V=10.65$ are reasonable.
Model light curve fitting {#model-light-curve-fitting}
-------------------------
Figure \[all\_mass\_v1369\_cen\_v1668\_cyg\_x45z02c15o20\](a) shows the model absolute $V$ light curve of a $0.90~M_\sun$ WD (thick solid red line) with the envelope chemical composition of CO Nova 3 [@hac16k]. Here, we assume $(m-M)_V=10.25$. The chemical composition of V1369 Cen ejecta is not known, so we assume CO Nova 3 because V1369 Cen showed a optically-thin dust blackout similar to V1668 Cyg, of which the chemical composition is represented roughly by CO Nova 3 [@hac06kb; @hac16k].
In free-free emission dominant spectra, optical and near-infrared (NIR), i.e., $B$, $V$, and $I_{\rm C}$ light curves should have the same shape [see, e.g., @hac06kb]. The V1369 Cen data show such a property in the middle phase of light curves and are in good agreement with the model light curve. The observed $B$ and $V$ magnitudes departed from the theoretical light curve at $t\gtrsim 100$ day because of strong contributions of emission lines to the $B$ and $V$ bands in the nebular phase (see Paper I). This model light curve again confirms our result of $(m-M)_V=10.25$. The WD mass is around $0.9~M_\sun$.
Our $0.90~M_\sun$ WD model also predicts a supersoft X-ray source phase from Day 340 until Day 960 (thin solid red line in Figure \[all\_mass\_v1369\_cen\_v1668\_cyg\_x45z02c15o20\](a)). @pag14 reported an X-ray spectrum on UT 2014 March 8 (JD 2456724.5). This epoch (Day $\sim100$) corresponds to the deepest dust blackout (see Figure \[v1369\_cen\_v\_bv\_ub\_color\_curve\]), and is much earlier than the X-ray turn-on time in our model. @mas18 presented the X-ray count rates of V1369 Cen obtained with [*Swift*]{}. We plot the X-ray data taken from the [*Swift*]{} Web site[^3] [@eva09] in Figure \[all\_mass\_v1369\_cen\_v1668\_cyg\_x45z02c15o20\](a). The blue pluses represent the hard X-ray (1.5–10 kev) and the magenta crosses denote the soft X-ray (0.3–1.5 keV) components. No clear supersoft X-rays were detected until Day 370. We may conclude that the X-rays detected on Day $\sim 100$ are not from the WD surface but shock-origin.
V5666 Sgr 2014 {#v5666_sgr_cmd}
==============
V5666 Sgr was discovered by S. Furuyama at mag 8.7 on UT 2014 January 26.857 (JD 2456684.357) [@fur14]. It reached $m_V=10.0$ on JD 2456697.3 (estimated from S. Kiyota’s $m_V=9.8$ of VSOLJ). The nova was identified by A. Arai as an type [@fur14].
We estimate the distance moduli of V5666 Sgr in the four bands, i.e., $(m-M)_B=15.87$, $(m-M)_V=15.38$, $(m-M)_I=14.59$, and $(m-M)_K=14.07$ in Appendix \[v5666\_sgr\]. We plot these four relations by the magenta, blue, cyan, and cyan-blue lines in Figure \[distance\_reddening\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\](d). These four lines cross at $d=5.8$ kpc and $E(B-V)=0.50$. Thus, we adopt $d=5.8\pm0.6$, $E(B-V)=0.50\pm0.05$, and the timescaling factor $f_{\rm s}=1.78$ against LV Vul. Figure \[distance\_reddening\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\](d) also shows various distance-reddening relations toward V5666 Sgr, whose galactic coordinates are $(l,b)=(9\fdg8835, -4\fdg6567)$. We add Marshall et al.’s (2006) relations: $(l, b)=(9\fdg75, -4\fdg50)$ by open red squares, $(10\fdg00, -4\fdg50)$ by filled green squares, $(9\fdg75, -4\fdg75)$ by blue asterisks, and $(10\fdg00, -4\fdg75)$ by open magenta circles, each with error bars. The closest direction among the four nearby directions is that of open magenta circles. We also add the relations given by @gre15 [@gre18] (solid black and orange lines, respectively) and the relation by @chen18 (solid cyan-blue line). Özdörmez et al.’s data are not available. The orange line is consistent with our results of $d=5.8\pm0.6$ kpc and $E(B-V)=0.50\pm0.05$.
@schu14 determined interstellar extinction as a function of distance in the galactic bulge covering $-10\fdg0 < l < 10\fdg0$ and $-10\fdg0 <b< 5\fdg0$, using data from the VISTA Variables in the Via Lactea (VVV) survey together with the Besançon stellar population synthesis model of the Galaxy. The resolution is $0\fdg1\times0\fdg1$ and the distance is extended up to 10 kpc in a 0.5 kpc step. We plot four Schuletheis et al.’s distance-reddening relations toward near the direction of V5666 Sgr, i.e., $(l,b)=(9\fdg8, -4\fdg6)$, $(9\fdg8, -4\fdg7)$, $(9\fdg9, -4\fdg6)$, and $(9\fdg9, -4\fdg7)$ by very thin solid cyan lines in Figure \[distance\_reddening\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\](d), where we adopt the relations of $A_{K_s}=0.364~E(B-V)$ [@sai13] and $A_{K_s}=0.528~E(J-K_s)$ [@nis09shogo] in the conversion of $E(B-V)$ from their $E(J-K_s)$ in their Table 1. The four lines show zigzag patterns, although the reddening should increase monotonically with the distance. In this sense, Schuletheis et al.’s relation may not be appropriate in the middle distance. Our crossing point, $d=5.8$ kpc and $E(B-V)=0.50$, however, is consistent with theirs.
Adopting $E(B-V)=0.50$ and $(m-M)_V=15.4$, we plot the color-magnitude diagram of V5666 Sgr in Figure \[hr\_diagram\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\_outburst\_mv\](d). In the same figure, we add the tracks of V1500 Cyg (solid green line) and LV Vul (solid orange line). Although the track of V5666 Sgr almost follows that of LV Vul, its position is slightly lower than that of LV Vul.
Figure \[hr\_diagram\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\_outburst\](d) shows the time-stretched color-magnitude diagram of V5666 Sgr. We adopt $E(B-V)=0.50$ and $(m-M')_V=16.05$ in Equation (\[absolute\_mag\_v5666\_sgr\_lv\_vul\_stretch\]). Now, the track of V5666 Sgr follows well that of LV Vul. Thus, we conclude that V5666 Sgr belongs to the LV Vul type in the time-stretched color-magnitude diagram. This overlap supports our results of $E(B-V)=0.50$ and $(m-M')_V=16.05$, that is, $E(B-V)=0.50\pm0.05$, $(m-M)_V=15.4\pm0.1$, $f_{\rm s}=1.78$, and $d=5.8\pm0.6$ kpc. We list our results in Table \[extinction\_various\_novae\].
We further confirm $(m-M)_V=15.4$ from our model light curve fitting. Taking $(m-M)_V=15.4$, we plot a $0.85~M_\sun$ WD model with the chemical composition of CO nova 3 [@hac16k] in Figure \[all\_mass\_v5666\_sgr\_v1668\_cyg\_x45z02c15o20\](a). The $0.85~M_\sun$ WD (solid red lines) model does not match with the early phase of oscillatory flat peak, but reasonably fit with the mid and later phases, where free-free emission dominates in the $B$, $V$, and $I_{\rm C}$ magnitudes. Thus, we again confirm $(m-M)_{V, \rm V5666~Sgr}=15.4\pm0.2$. The estimated WD mass is about $M_{\rm WD}=0.85~M_\sun$ for the chemical composition of CO Nova 3.
Revisiting 12 Novae in the Time-stretched Color-Magnitude Diagram {#twelve_novae}
=================================================================
In what follows, we reexamine 12 novae studied in Paper II with a new light of time-stretched color-magnitude diagram.
V446 Her 1960 {#v446_her_cmd}
-------------
For the reddening toward V446 Her, @hac14k obtained the arithmetic average of the values in literature to be $E(B-V)=0.41\pm0.15$. Recently, @ozd16 proposed another average of $E(B-V)=0.37\pm0.04$ including the result of @sel13, $E(B-V)=0.38\pm0.04$, from the archival 2200Å feature. See Paper II and @ozd16 for a summary of other estimates on the distance and reddening. In the present paper, we adopt $E(B-V)=0.40\pm0.05$ after @hac14k. We obtain the distance of $d=1.38\pm0.2$ kpc from Equation (\[distance\_modulus\_rv\]), $E(B-V)=0.40\pm0.05$, $f_{\rm s}= 1.0$ against LV Vul, and $(m-M)_V=11.95\pm0.1$ in Appendix \[v446\_her\].
Figure \[distance\_reddening\_v446\_her\_v533\_her\_pw\_vul\_v1419\_aql\](a) shows various distance-reddening relations toward V446 Her, whose galactic coordinates are $(l, b)= (45\fdg4092, +4\fdg7075)$. The vertical solid red line denotes the reddening of $E(B-V)=0.40$. The solid blue line denotes the distance modulus in the $V$ band, i.e., $(m-M)_V=11.95$ and Equation (\[distance\_modulus\_rv\]). These two lines cross at $E(B-V)=0.40$ and $d=1.38$ kpc. We further plot Marshall et al.’s (2006) relations in four directions close to the direction of V446 Her; toward $(l, b)=(45\fdg25, 4\fdg50)$ by open red squares, $(45\fdg50, 4\fdg50)$ by filled green squares, $(45\fdg25, 4\fdg75)$ by blue asterisks, and $(45\fdg50, 4\fdg75)$ by open magenta circles. The closest direction in the galactic coordinates is that of open magenta circles among the four nearby directions. The solid black/orange lines denote the distance-reddening relations given by @gre15 [@gre18], respectively. The solid green line represents the relation given by @sal14. The open cyan diamonds depict the relation of @ozd16. The cyan-blue line represents the relation given by @chen18. Our crossing point at $E(B-V)=0.40$ and $d=1.38$ kpc is consistent with Marshall et al.’s (open magenta circles) and Green et al.’s (solid black/orange lines) relations. Our crossing point is also consistent with the distance-reddening relation given by @ozd16. @coh85 obtained the distance of V446 Her by the expansion parallax method to be $d=(v_{\rm exp} \times t)/r_{\rm shell} =
(1235~{\rm km~s}^{-1} \times 24~{\rm yr})/4\farcs5 =1.35$ kpc. Our value of $d=1.38\pm0.2$ kpc is consistent with Cohen’s value. Thus, we confirm again that our obtained values of $(m-M)_V=11.95\pm0.1$, $E(B-V)=0.40\pm0.05$, and $d=1.38\pm0.2$ kpc are reasonable.
Adopting $E(B-V)=0.40$ and $(m-M')_V=11.95$ in Equation (\[absolute\_mag\_lv\_vul\_v446\_her\]), we obtain the time-stretched color-magnitude diagram of V446 Her in Figure \[hr\_diagram\_v446\_her\_v533\_her\_pw\_vul\_v1419\_aql\_outburst\](a). The track of V446 Her is just on the LV Vul track (solid orange line). Therefore, V446 Her belongs to the LV Vul type. This overlapping to the LV Vul track supports our values of $E(B-V)=0.40$ and $(m-M')_V=11.95$, that is, $f_{\rm s}=1.0$, $E(B-V)=0.40\pm0.05$, $(m-M)_V=11.95\pm0.1$, and $d=1.38\pm0.2$ kpc. We list our results in Table \[extinction\_various\_novae\].
Taking $(m-M)_V=11.95$ for V446 Her, we add a $0.98~M_\sun$ WD model (solid red line) with the chemical composition of CO nova 3 [@hac16k] in Figure \[v446\_her\_v1668\_cyg\_lv\_vul\_v\_color\_logscale\](a). The model absolute $V$ light curve reasonably fits with the observed apparent $V$ light curve of V446 Her. This confirms that the distance modulus of $(m-M)_V=11.95$ is reasonable. In the same figure, we depict another $0.98~M_\sun$ WD model with $(m-M)_V=14.6$ for V1668 Cyg (solid black lines). The difference between these two models represents the difference in the initial hydrogen-rich envelope mass; the initial envelope mass $M_{\rm env,0}$ of the solid red line is slightly smaller than that of the solid black line.
V533 Her 1963 {#v533_her_cmd}
-------------
V533 Her was studied in Paper II on the $(B-V)_0$-$M_V$ diagram but we here examine it on the time-stretched color-magnitude diagram. @hac16kb took $E(B-V)=0.038\pm0.002$ from the NASA/IPAC galactic absorption map. We adopt their value of $E(B-V)=0.038$. We determine the distance modulus in the $V$ band to be $(m-M)_V=10.65\pm0.1$ together with $f_{\rm s}= 1.20$ against LV Vul in Appendix \[v533\_her\], and plot it by the solid blue line in Figure \[distance\_reddening\_v446\_her\_v533\_her\_pw\_vul\_v1419\_aql\](b). The vertical solid red line is the color excess of $E(B-V)=0.038$. These two lines cross at $d=1.28$ kpc (and $E(B-V)=0.038$). Thus, we obtain the distance of $d=1.28\pm0.2$ kpc.
Figure \[distance\_reddening\_v446\_her\_v533\_her\_pw\_vul\_v1419\_aql\](b) also shows a few distance-reddening relations toward V533 Her, whose galactic coordinates are $(l, b)= (69\fdg1887, +24\fdg2733)$. None of Marshall et al.’s, Sale et al.’s, Özdörmez et al.’s, and Chen et al.’s relations is available. The solid black/orange lines denote the distance-reddening relations given by @gre15 [@gre18], respectively. Our crossing point is broadly consistent with Green et al.’s distance-reddening relations.
Adopting $E(B-V)=0.038$ and $(m-M')_V=10.85$ in Equation (\[absolute\_mag\_v533\_her\_lv\_vul\_v\]), we obtain the time-stretched color-magnitude diagram of V533 Her in Figure \[hr\_diagram\_v446\_her\_v533\_her\_pw\_vul\_v1419\_aql\_outburst\](b). We also add the template tracks of LV Vul (thick solid orange lines), V1500 Cyg (thick solid green line), and V1974 Cyg (solid magenta lines). The V533 Her track broadly follows the mid part of V1974 Cyg tracks at least until the onset of the nebular phase (see also Figure \[v533\_her\_lv\_vul\_v1974\_cyg\_v\_bv\_ub\_logscale\](a) and (b)). We regard that V533 Her belongs to the V1500 Cyg type in the time-stretched color-magnitude diagram. This supports our values of $(m-M')_V=10.85$ and $E(B-V)=0.038$, that is, $(m-M)_V=10.65\pm0.1$, $E(B-V)=0.038\pm0.01$, and $f_{\rm s}=1.20$. Our results are listed in Table \[extinction\_various\_novae\].
We again discuss the effect of different responses in the color filters. The color-magnitude track of V533 Her bifurcates at the onset of the nebular phase around UT 1963 April 19 [@chi64r], i.e., at $m_V=7.2$ denoted by the large open blue square in Figure \[hr\_diagram\_v446\_her\_v533\_her\_pw\_vul\_v1419\_aql\_outburst\](b). We plot four color-magnitude data taken from @kre66, @she64, @gen63, and @chi64, from upper to lower (or from right to left), and one of the three major branches immediately turns to the right. After the onset of the nebular phase, strong emission lines \[\] began to contribute to the blue edge of $V$ filter. The response of each $V$ filter is different to each other at the blue edge and this difference makes a significant difference in the $V$ magnitude. This effect can be clearly seen in Figures 3, 4, and 5 of @kre66, in which the $V$ magnitude started to branch off after UT 1963 April 10 while the $B$ magnitude did not among the various observers.
We check our obtained value of $(m-M)_V=10.65$ by fitting with our model $V$ light curve in Figure \[v533\_her\_lv\_vul\_v1974\_cyg\_v\_bv\_ub\_logscale\](a). We add a $1.03~M_\sun$ WD model (solid red line) with the chemical composition of Ne nova 2 [@hac10k]. Taking $(m-M)_V=10.65$ for V533 Her, the absolute model $V$ light curve reproduces the observed apparent $V$ light curve of V533 Her (filled red circles). This confirms that the distance modulus of $(m-M)_V=10.65$ is reasonable.
PW Vul 1984\#1 {#pw_vul_cmd}
--------------
PW Vul was examined in Paper II on the $(B-V)_0$-$M_V$ diagram. Here, we reexamine it on the time-stretched color-magnitude diagram. In Appendix \[pw\_vul\], we obtain $(m-M)_V=13.0\pm0.2$ together with $f_{\rm s}= 2.24$ against LV Vul based on the time-stretching method. Taking $(m-M)_V=13.0$ for PW Vul, we plot a $0.83~M_\sun$ WD model (solid red line) with the chemical composition of CO nova 4 [@hac15k] in Figure \[pw\_vul\_lv\_vul\_v1668\_cyg\_v1500\_cyg\_v\_bv\_ub\_color\_logscale\](a). The model absolute $V$ light curve reasonably fits with the observed apparent $V$ light curve of PW Vul after $t > 60$ day. This confirms that the distance modulus of $(m-M)_V=13.0$ is reasonable.
We also fit our UV 1455Å light curve model of the $0.83~M_\sun$ WD with the observation as shown in Figure \[pw\_vul\_lv\_vul\_v1668\_cyg\_v1500\_cyg\_v\_bv\_ub\_color\_logscale\](a). We obtain $F_{1455}^{\rm mod}= 15$ and $F_{1455}^{\rm obs}= 6$ in units of $10^{-12}$ erg cm$^{-2}$ s$^{-1}$ Å$^{-1}$ at the upper bound of Figure \[pw\_vul\_lv\_vul\_v1668\_cyg\_v1500\_cyg\_v\_bv\_ub\_color\_logscale\](a). With these values, we plot the distance-reddening relation of Equation (\[qu\_vul\_uv1455\_fit\_eq2\]) in Figure \[distance\_reddening\_v446\_her\_v533\_her\_pw\_vul\_v1419\_aql\](c) by the solid magenta line. The solid blue line of $(m-M)_V=13.0$ and this magenta line cross at $d=1.8$ kpc and $E(B-V)= 0.57$. These light curve fittings are essentially the same as in the previous work [@hac15k]. Thus, we obtain the distance of $d=1.8\pm0.2$ kpc from Equation (\[distance\_modulus\_rv\]), $E(B-V)=0.57\pm0.05$, and $(m-M)_V=13.0\pm0.2$. This distance estimate is consistent with the distance estimated by @dow00, $d=1.8\pm0.05$ kpc, with the nebular expansion parallax method. We list the results in Table \[extinction\_various\_novae\].
For comarison, in Figure \[pw\_vul\_lv\_vul\_v1668\_cyg\_v1500\_cyg\_v\_bv\_ub\_color\_logscale\](a), we depict a $0.98~M_\sun$ WD model (solid blue lines) with the chemical composition of CO nova 3 [@hac15k], assuming that $(m-M)_V=14.6$ for V1668 Cyg. With the same stretching factor of $f_{\rm s}=2.24$, both the $V$ and UV 1455Å light curves almost overlap between PW Vul and V1668 Cyg. This confirms that the optical light curves and UV 1455Å light curves follow the same timescaling law as already explained in Section \[introduction\].
We check the distance and reddening based on various distance-reddening relations. @hac15k obtained $E(B-V)=0.55\pm0.05$ from the fitting in the color-color diagram (Paper I), being consistent with the crossing point mentioned above. Figure \[distance\_reddening\_v446\_her\_v533\_her\_pw\_vul\_v1419\_aql\](c) shows various distance-reddening relations toward PW Vul, whose galactic coordinates are $(l, b)=(61\fdg0983, +5\fdg1967)$. We plot the constraint of $d=1.8\pm0.05$ kpc by @dow00 (solid horizontal black lines). We further add other distance-reddening relations toward PW Vul. The vertical solid red line is $E(B-V)=0.57$. We plot the distance-reddening relations (solid black/orange lines) given by @gre15 [@gre18], respectively. We also plot four distance-reddening relations of @mar06; toward $(l, b)=(61\fdg00, +5\fdg00)$ denoted by open red squares, $(61\fdg25, +5\fdg00)$ by filled green squares, $(61\fdg00, +5\fdg25)$ by blue asterisks, and $(61\fdg25, +5\fdg25)$ by open magenta circles, each with error bars. The closest direction in the galactic coordinates is that of blue asterisks. The open cyan-blue diamonds represent the relation given by @ozd16. Our crossing point at $d=1.8$ kpc and $E(B-V)=0.57$ is $\Delta E(B-V)=0.15$ mag larger than the trends of Marshall et al.’s blue asterisks data. To summarize, our set of $d=1.8$ kpc and $E(B-V)=0.57$ is not consistent with all the 3D dust maps given by @mar06, @ozd16, @gre15 [@gre18], and @chen18. This is the first and only the case that our crossing point is largely different from the several 3D dust maps. We discuss the reason of such a large discrepancy below.
The 3D dust maps basically give an averaged value of a relatively broad region, so the pinpoint reddening could be different from the values of the 3D dust maps. The pinpoint estimate of the reddening toward PW Vul indicates $E(B-V)=0.55\pm0.05$ as already discussed in @hac15k. For example, @and91 obtained $E(B-V)=0.58 \pm 0.06$ from $\lambda1640/\lambda4686$ ratio and $E(B-V)=0.55 \pm 0.1$ from the interstellar absorption feature at 2200Å for the reddening toward PW Vul. @sai91 reported $E(B-V)=0.60 \pm 0.06$ from $\lambda1640/\lambda4686$ ratio. @hac14k estimated the reddening to be $E(B-V)=0.55\pm0.05$ from the color-color diagram fit of PW Vul. As for the distance to PW Vul, @dow00 obtained $d=1.8\pm0.05$ kpc with the nebular expansion parallax method. These pinpoint constraints are all consistent with our crossing point of $E(B-V)=0.57\pm0.05$, $(m-M)_V = 13.0\pm0.2$, and $d=1.8\pm0.2$ kpc.
We further obtain the distance moduli of $UBVI$ bands and check their crossing points in the distance-reddening plane. We obtain $(m-M)_U= 13.92$, $(m-M)_B= 13.55$, and $(m-M)_V= 13.0$ (and $(m-M)_I= 12.12$) in Appendix \[pw\_vul\], and plot them in Figure \[distance\_reddening\_v446\_her\_v533\_her\_pw\_vul\_v1419\_aql\](c) by the thin solid green, cyan, thick solid blue (and thin solid blue-magenta) lines, respectively. These lines cross at $E(B-V)=0.57$ and $d=1.8$ kpc, confirming again our results.
Adopting $E(B-V)=0.57$ and $(m-M')_V=13.85$ in Equation (\[absolute\_mag\_pw\_vul\_lv\_vul\]), we obtain the time-stretched color-magnitude diagram of PW Vul in Figure \[hr\_diagram\_v446\_her\_v533\_her\_pw\_vul\_v1419\_aql\_outburst\](c). Here, we plot the data taken from @nos85, @kol86, and @rob95. PW Vul follows the V1500 Cyg track except for the very early phase. Therefore, PW Vul belongs to the V1500 Cyg type in the time-stretched color-magnitude diagram. This overlapping to the V1500 Cyg track supports our estimates of $E(B-V)=0.57$ and $(m-M')_V=13.85$, that is, $f_{\rm s}=2.24$, $E(B-V)=0.57\pm0.05$, $(m-M)_V=13.0\pm0.2$, and $d=1.8\pm0.2$ kpc. We list our results in Table \[extinction\_various\_novae\].
V1419 Aql 1993 {#v1419_aql_cmd}
--------------
V1419 Aql was examined in Paper II on the $(B-V)_0$-$M_V$ diagram. Here, we reexamine it on the time-stretched color-magnitude diagram. We determine the distance modulus in the $V$ band to be $(m-M)_V=15.0\pm0.2$ together with $f_{\rm s}= 1.41$ against LV Vul in Appendix \[v1419\_aql\]. For the reddening toward V1419 Aql, @hac14k [@hac16kb] obtained $E(B-V)=0.50\pm0.05$ from the fitting in the color-color diagram (Paper I). On the other hand, the NASA/IPAC galactic 2D dust absorption map gives $E(B-V)=0.55 \pm 0.01$ in the direction toward V1419 Aql. See @hac14k [@hac16kb] and @ozd16 for a summary of other estimates on the extinction and distance of V1419 Aql. We adopt the arithmetic mean of these two values, i.e., $E(B-V)=0.52\pm0.05$ in the present paper. Then, we obtain the distance of $d=4.7\pm0.5$ kpc from Equation (\[distance\_modulus\_rv\]).
We check the distance and reddening toward V1419 Aql, whose galactic coordinates are $(l, b)=(36\fdg8110, -4\fdg1000)$, based on various distance-reddening relations in Figure \[distance\_reddening\_v446\_her\_v533\_her\_pw\_vul\_v1419\_aql\](d). The solid blue line denotes the distance modulus of $(m-M)_V = 15.0$ and Equation (\[distance\_modulus\_rv\]). The vertical solid red line is $E(B-V)=0.52$. These two lines cross at $d=4.7$ kpc (and $E(B-V)=0.52$). We add the distance-reddening relations (solid black/orange lines) given by @gre15 [@gre18], respectively. We also plot four distance-reddening relations of @mar06; toward $(l, b)=(36\fdg75, -4\fdg00)$ denoted by open red squares, $(37\fdg00, -4\fdg00)$ by filled green squares, $(36\fdg75, -4\fdg25)$ by blue asterisks, and $(37\fdg00, -4\fdg25)$ by open magenta circles, each with error bars. The closest direction is that of open red squares. The solid green line is the relation given by @sal14. The open cyan-blue diamonds with error bars are the relation given by @ozd16. The solid cyan-blue line is the relation given by @chen18. Our set of $E(B-V)=0.52$ and $d=4.7$ kpc is broadly consistent with the distance-reddening relations of @mar06, @sal14, @gre18, and @ozd16. Thus, we confirm that our adopted values of $E(B-V)=0.52\pm0.05$, $(m-M)_V = 15.0\pm0.2$, and $d=4.7\pm0.5$ kpc are reasonable. We list our results in Table \[extinction\_various\_novae\].
Adopting $E(B-V)=0.52$ and $(m-M')_V=15.35$ in Equation (\[absolute\_mag\_v1419\_aql\_lv\_vul\]), we obtain the time-stretched color-magnitude diagram of V1419 Aql in Figure \[hr\_diagram\_v446\_her\_v533\_her\_pw\_vul\_v1419\_aql\_outburst\](d). Here, we plot the data taken from @mun94b and IAU Circular Nos. 5794, 5802, 5807, and 5829. The track of V1419 Aql is considerably affected by dust formation, the start of which is denoted by the large magenta square. We regard that V1419 Aql belongs to the LV Vul type because the V1419 Aql track follows a part of the V1668 Cyg track (blue lines) until the dust blackout started. For comparison, we also plot the V1065 Cen track (filled blue triangles) that shows a similar shallow dust blackout [see Figures 4 and 6 of @hac18k]. The track of V1065 Cen almost follows the lower branch of LV Vul, while the track of V1419 Aql is not clear owing to dust blackout.
Finally, we plot a $0.90~M_\sun$ WD model with the chemical composition of CO nova 3 [@hac16k] by the solid red lines in Figure \[v1419\_aql\_v1668\_cyg\_lv\_vul\_v\_bv\_ub\_color\_logscale\](a). The $V$ model light curve reasonably fits with the $V$ light curve of V1419 Aql during $\log t$ (day)$= 1.0-1.5$. This confirms our value of $(m-M)_V= 15.0$.
V705 Cas 1993 {#v705_cas_cmd}
-------------
V705 Cas was examined in Paper II, but we reexamine it on the time-stretched color-magnitude diagram. For the reddening toward V705 Cas, we adopt $E(B-V)=0.45\pm0.05$ after @hac14k. We do not repeat the discussion. Papers I & II give a summary on the distance and reddening toward V705 Cas, $(l,b)=(113\fdg6595,-4\fdg0959)$. We obtain the timescaling factor of $f_{\rm s}= 2.8$ against LV Vul and the distance of $d=2.6\pm0.3$ kpc from Equation (\[distance\_modulus\_rv\]), $E(B-V)=0.45\pm0.05$, and $(m-M)_V=13.45\pm0.2$ in Appendix \[v705\_cas\]. Using the expansion parallax method, @eyr96 and @dia01 obtained $d=2.5$ kpc and $d=2.9\pm0.4$ kpc, respectively. These distances are all consistent with our value of $d=2.6\pm0.3$ kpc. We list our results in Table \[extinction\_various\_novae\].
Next, we discuss the model light curve fitting. We obtain $(m-M)_V=13.45$ based on the time-stretching method in Appendix \[v705\_cas\]. Assuming that $(m-M)_V=13.45$, we plot a $0.78~M_\sun$ WD model (solid magenta lines) with the chemical composition of CO nova 4 [@hac15k] in Figure \[v705\_cas\_pw\_vul\_lv\_vul\_v\_bv\_ub\_color\_curve\_logscale\](a). The model absolute $V$ and UV 1455Å light curves of the $0.78~M_\sun$ WD reasonably fit with the observed apparent $V$ and UV 1455Å light curves of V705 Cas. This model light curve fitting is essentially the same as Figure 15 of @hac15k and we again confirm that the distance modulus of $(m-M)_V=13.45$ is reasonable for V705 Cas. In the same figure, we add a $0.83~M_\sun$ WD model (solid blue lines) with the same chemical composition of CO nova 4, assuming that $(m-M)_V=13.0$ for PW Vul.
We reanalyze the distance and reddening toward V705 Cas based on various distance-reddening relations in Figure \[distance\_reddening\_v705\_cas\_v382\_vel\_v5114\_sgr\_v2362\_cyg\](a). The solid blue line shows $(m-M)_V = 13.45$ and Equation (\[distance\_modulus\_rv\]). The solid magenta line is the UV 1455Å fit of our $0.78~M_\sun$ WD model. The UV 1455Å light curve fit gives the relation of Equation (\[qu\_vul\_uv1455\_fit\_eq2\]) together with $F_{1455}^{\rm obs}=6.0$ and $F_{1455}^{\rm mod}=12.5$ in units of $10^{-12}$ erg cm$^{-2}$ s$^{-1}$ Å$^{-1}$ at the upper bound of Figure \[v705\_cas\_pw\_vul\_lv\_vul\_v\_bv\_ub\_color\_curve\_logscale\](a). We also plot the relations given by @gre15 [@gre18] by the solid black/orange lines, respectively. No data of @mar06 are available in this direction. The solid green line represent the relation given by @sal14. The open cyan-blue diamonds with error bars denote the relation given by @ozd16. The solid cyan-blue line denotes the relation given by @chen18. The blue, magenta, and red lines cross at $E(B-V)=0.45$ and $d=2.6$ kpc. This crossing point is also consistent with the distance-reddening relation given by @gre15 [black line], @sal14 [green line], and @ozd16 [cyan-blue diamonds].
Adopting $E(B-V)=0.45$ and $(m-M')_V=14.55$ in Equation (\[absolute\_mag\_v705\_cas\_lv\_vul\]), we obtain the time-stretched color-magnitude diagram of V705 Cas in Figure \[hr\_diagram\_v705\_cas\_v382\_vel\_v5114\_sgr\_v2362\_cyg\_outburst\](a). Here, we plot the data taken from @mun94a, @hri98, and IAU Circular Nos. 5905, 5912, 5914, 5920, 5928, 5929, 5945, and 5957. The track of V705 Cas is also affected by dust formation as denoted by the large black square. We also plot the V1065 Cen track [see Figures 4 and 6 of @hac18k] in Figure \[hr\_diagram\_v705\_cas\_v382\_vel\_v5114\_sgr\_v2362\_cyg\_outburst\](a). V1065 Cen shows a similar dust blackout to V705 Cas but its depth of dust blackout is much shallower (see Figure \[v496\_sct\_v1065\_cen\_lv\_vul\_v\_bv\_ub\_color\_logscale\]) than that of V705 Cas. The V1065 Cen track comes back soon after the dust blackout and follows again the LV Vul track, so that V1065 Cen belongs to the LV Vul type in the time-stretched color-magnitude diagram. We regard that V705 Cas also belongs to the LV Vul type because the V705 Cas track almost follows that of LV Vul and V1065 Cen until the dust blackout started. The early pulse (oscillation) in the $V$ light curve makes a loop in the color-magnitude diagram. This kind of loops were also seen in multiple-peak novae like V723 Cas and V5558 Sgr as discussed by @hac16kb. We list our results in Table \[extinction\_various\_novae\].
V382 Vel 1999 {#v382_vel_cmd}
-------------
V382 Vel was also studied in Paper II on the $(B-V)_0$-$M_V$ diagram. In this subsection, we examine it on the $(B-V)_0$-$(M_V-2.5\log f_{\rm s}$) diagram. For the reddening toward V382 Vel, $(l,b)=(284\fdg1674, +5\fdg7715)$, @ste99 obtained $E(B-V)=A_V/3.1=0.8/3.1=0.26$ from an equivalent width of 0.024 nm of the interstellar K line. @ori02 obtained the hydrogen column density toward V382 Vel to be $N_{\rm H}\sim 2 \times 10^{21}$ cm$^{-2}$ from X-ray spectrum fittings. This value can be converted with $E(B-V)= N_{\rm H}/ 4.8 \times 10^{21} \sim 0.4$ [@boh78], $E(B-V)= N_{\rm H}/ 5.8 \times 10^{21} \sim 0.34$ [@gue09], and $E(B-V)= N_{\rm H}/ 6.8 \times 10^{21} \sim 0.3$ [@lis14]. On the other hand, @del02 obtained $E(B-V)=0.05$ from various line ratios and $E(B-V)=0.09$ from D interstellar absorption features. @sho03 obtained $E(B-V)= 0.20$ from the resemblance of [*IUE*]{} spectra to that of V1974 Cyg. @sho03 and @nes05 obtained $N_{\rm H}=1.2\times 10^{21}$ cm$^{-2}$, corresponding to $E(B-V)= N_{\rm H}/ 4.8 \times 10^{21} = 0.25$, $E(B-V)= N_{\rm H}/ 5.8 \times 10^{21} = 0.21$, and $E(B-V)= N_{\rm H}/ 6.8 \times 10^{21} = 0.18$. @hac14k obtained $E(B-V)=0.15\pm0.05$ by fitting the color-color track of V382 Vel with the general track of novae. To summarize, there are two estimates, $E(B-V)=0.1\pm0.05$ and $E(B-V)=0.25\pm0.05$.
We obtain three distance moduli of $B$, $V$, and $I_{\rm C}$ bands in Appendix \[v382\_vel\] based on the time-stretching method. We plot these distance moduli in Figure \[distance\_reddening\_v705\_cas\_v382\_vel\_v5114\_sgr\_v2362\_cyg\](b) by the magenta, blue, and cyan lines, that is, $(m-M)_B= 11.7$, $(m-M)_V= 11.48$, and $(m-M)_I= 11.09$ together with Equations (\[distance\_modulus\_rb\]), (\[distance\_modulus\_rv\]), and (\[distance\_modulus\_ri\]). These three lines broadly cross at $d=1.4$ kpc and $E(B-V)=0.25$. Thus, we obtain the timescaling factor $f_{\rm s}= 0.51$ against LV Vul, the reddening toward V382 Vel, $E(B-V)=0.25\pm0.05$, and the distance to V382 Vel, $d=1.4\pm0.2$ kpc. The distance modulus in the $V$ band is $(m-M)_V=11.5\pm0.2$. These values are listed in Table \[extinction\_various\_novae\].
Adopting $E(B-V)=0.25$ and $(m-M')_V=10.75$ in Equation (\[absolute\_mag\_v382\_vel\_lv\_vul\]), we plot the time-stretched color-magnitude diagram of V382 Vel in Figure \[hr\_diagram\_v705\_cas\_v382\_vel\_v5114\_sgr\_v2362\_cyg\_outburst\](b). The nebular phase had started at least by the end of 1999 June, i.e., $\sim40$ days after the optical maximum [@del02]. We plot this phase ($M_V=m_V-(m-M)_V\approx7.09-11.5=-4.41$) by the large open red square in Figure \[hr\_diagram\_v705\_cas\_v382\_vel\_v5114\_sgr\_v2362\_cyg\_outburst\](b). The track of V382 Vel follows the V1500 Cyg/V1974 Cyg tracks at least until the nebular phase started. Thus, we regard that V382 Vel belongs to the V1500 Cyg type in the time-stretched color-magnitude diagram. This overlapping to the tracks of V1500 Cyg/V1974 Cyg supports our values of $E(B-V)=0.25$ and $(m-M')_V=10.75$, that is, $E(B-V)=0.25\pm0.05$, $(m-M)_V=11.5\pm0.2$, $f_{\rm s}=0.51$, and $d=1.4\pm0.2$ kpc.
Taking $(m-M)_V=11.5$ for V382 Vel, we plot the $V$ model light curve (total flux of free-free plus photospheric emission: solid red line) of a $1.23~M_\sun$ WD with the envelope chemical composition of Ne nova 2 [@hac10k; @hac16k] in Figure \[v382\_vel\_lv\_vul\_v1668\_cyg\_v1974\_cyg\_v\_bv\_ub\_color\_logscale\](a). The model $V$ light curve reasonably reproduces the observation (filled red circles). This model light curve fitting gives the same result as in Figure 27 of @hac16k. Thus, we again confirm that the distance modulus of $(m-M)_V=11.5\pm0.2$ is reasonable.
We further examine the distance and reddening toward V382 Vel, $(l, b)= (284\fdg1674, +5\fdg7715)$, based on various distance-reddening relations in Figure \[distance\_reddening\_v705\_cas\_v382\_vel\_v5114\_sgr\_v2362\_cyg\](b). We plot the four sets of distance-reddening relation calculated by @mar06; toward $(l, b)= (284\fdg00, +5\fdg75)$ denoted by open red squares, $(284\fdg25, +5\fdg75)$ by filled green squares, $(284\fdg00, +6\fdg00)$ by blue asterisks, and $(284\fdg25, +6\fdg00)$ by open magenta circles. The filled green squares are the closest direction among the four nearby directions. The open cyan-blue diamonds with error bars denote the relation given by @ozd16. The solid cyan-blue line denotes the relation given by @chen18. The vertical solid red line is the color excess of $E(B-V)=0.25$. Our crossing point at $d=1.4$ kpc and $E(B-V)=0.25$ is broadly consistent with Marshall et al.’s and Özdörmez et al.’s relations within error bars.
The distance toward V382 Vel was recently estimated by @tom15 to be $d=(v_{\rm exp} \times t)/r_{\rm shell} = (1800\pm100~{\rm km~s}^{-1}
\times 12~{\rm yr})/(6\farcs0 \pm 0\farcs25) \approx 0.8\pm0.1$ kpc with the expansion parallax method. This value is much smaller than our estimate of $d=1.4\pm0.2$ kpc. However, the distance obtained with the expansion parallax method depends largely on the assumed velocity, $v_{\rm exp}$. If we adopt other velocities estimated in the early outburst phase $v_{\rm exp}\sim3500$ km s$^{-1}$ [e.g., @del02] or $v_{\rm exp}\sim2900$ km s$^{-1}$ [e.g., @nes05], we have large distances such as $d\sim1.5$ kpc or $d\sim1.2$ kpc, respectively.
V5114 Sgr 2004 {#v5114_sgr_cmd}
--------------
V5114 Sgr was studied in Paper II on the $(B-V)_0$-$M_V$ diagram. In this subsection, we reexamine it but on the $(B-V)_0$-$(M_V-2.5\log f_{\rm s}$) diagram. We obtain three distance moduli in the $U$, $B$, and $V$ bands in Appendix \[v5114\_sgr\]. We plot these three distance moduli in Figure \[distance\_reddening\_v705\_cas\_v382\_vel\_v5114\_sgr\_v2362\_cyg\](c) by the magenta, cyan, and blue lines, that is, $(m-M)_U= 17.45$, $(m-M)_B= 17.15$, and $(m-M)_V= 16.65$ together with Equations (\[distance\_modulus\_ru\]), (\[distance\_modulus\_rb\]), and (\[distance\_modulus\_rv\]), respectively. These three lines do not exactly but broadly cross at $d=10.9$ kpc and $E(B-V)=0.47$. Thus, we obtain the timescaling factor of $f_{\rm s}= 0.76$ against LV Vul, the reddening of $E(B-V)=0.47\pm0.05$, and the distance of $d=10.9\pm1$ kpc. These values are listed in Table \[extinction\_various\_novae\].
Various reddening estimates were summarized in Paper II and their arithmetic mean was calculated to be $E(B-V)=0.51\pm 0.09$. The NASA/IPAC galactic dust absorption map gives $E(B-V)=0.49 \pm 0.02$ in the direction toward V5114 Sgr. @hac14k derived the reddening of $E(B-V)=0.45\pm 0.05$ on the basis of the general track of color-color evolution of novae. These are all consistent with our new value of $E(B-V)=0.47\pm0.05$.
Next, we reanalyze the distance and reddening toward V5114 Sgr, $(l, b)=(3\fdg9429, -6\fdg3121)$, based on various distance-reddening relations in Figure \[distance\_reddening\_v705\_cas\_v382\_vel\_v5114\_sgr\_v2362\_cyg\](c). The vertical solid red line is $E(B-V)=0.47$. We plot four distance-reddening relations of @mar06; toward $(l, b)=(3\fdg75, -6\fdg50)$ denoted by open red squares, $(4\fdg00, -6\fdg50)$ by filled green squares, $(3\fdg75, -6\fdg25)$ by blue asterisks, and $(4\fdg00, -6\fdg25)$ by open magenta circles, each with error bars. The closest direction in the galactic coordinates is that of open magenta circles. The black/orange lines are the relations given by @gre15 [@gre18], respectively. The cyan-blue line is the relation given by @chen18. We further add the 3D reddening map given by @schu14. We plot four Schuletheis et al.’s distance-reddening relations toward near the direction of V5114 Sgr: $(l,b)=(3\fdg9, -6\fdg4)$, $(3\fdg9, -6\fdg3)$, $(4\fdg0, -6\fdg4)$, and $(4\fdg0, -6\fdg3)$ by the very thin solid cyan lines. The lines show zigzag patterns in this case, although the reddening must increase monotonically with the distance. In this sense, Schuletheis et al.’s distance-reddening relation may not be appropriate in the middle part of the lines. Our set of $E(B-V)=0.47$ and $d=10.9$ kpc is consistent with the trend of Marshall et al.’s data (open magenta circles), although $\Delta E(B-V)\sim0.05-0.1$ mag smaller than Green et al.’s.
Adopting $E(B-V)=0.47$ and $(m-M')_V=16.35$ in Equation (\[absolute\_mag\_v5114\_sgr\_lv\_vul\]), we obtain the time-stretched color-magnitude diagram of V5114 Sgr in Figure \[hr\_diagram\_v705\_cas\_v382\_vel\_v5114\_sgr\_v2362\_cyg\_outburst\](c). Here, we plot the data taken from @ede06, IAU Circular Nos. 8306 and 8310, and SMARTS. V5114 Sgr follows those of V1500 Cyg/V1974 Cyg. Therefore, we regard that V5114 Sgr belongs to the V1500 Cyg type in the time-stretched color-magnitude diagram. This overlapping supports our estimates of $E(B-V)=0.47$ and $(m-M')_V=16.35$, that is, $f_{\rm s}=0.76$, $E(B-V)=0.47\pm0.05$, $(m-M)_V=16.65\pm0.1$, and $d=10.9\pm1$ kpc. We list our results in Table \[extinction\_various\_novae\].
Taking $(m-M)_V=16.65$ for V5114 Sgr, we plot the $V$ model light curve (solid red line) of a $1.15~M_\sun$ WD with the envelope chemical composition of Ne nova 2 [@hac10k; @hac16k] in Figure \[v5114\_sgr\_lv\_vul\_v1668\_cyg\_v\_color\_logscale\](a). The model $V$ light curve reasonably reproduces the observation (filled red circles). This again confirms that the distance modulus of $(m-M)_V=16.65\pm0.1$ is reasonable.
V2362 Cyg 2006 {#v2362_cyg_cmd}
--------------
V2362 Cyg was also studied in Paper II. Here, we reexamine it on the $(B-V)_0$-$(M_V-2.5\log f_{\rm s}$) diagram. V2362 Cyg shows a prominent secondary maximum [e.g., @kim08; @mun08b see Figure \[v2362\_cyg\_lv\_vul\_v1500\_cyg\_v\_bv\_ub\_color\_logscale\]]. The origin of the secondary maximum was discussed by @hac09ka.
The color excess of V2362 Cyg was estimated by various authors (see Section 3.22 of Paper II). We adopt $E(B-V)=0.60\pm0.05$ after Hachisu & Kato (Paper II). Then, the timescaling factor is obtained to be $f_{\rm s}= 1.78$ against LV Vul and the distance is $d=5.1\pm0.5$ kpc from Equation (\[distance\_modulus\_rv\]) together with $(m-M)_V=15.4\pm0.2$ in Appendix \[v2362\_cyg\].
Adopting $E(B-V)=0.60$ and $(m-M')_V=16.05$ from Equation (\[absolute\_mag\_lv\_vul\_v2362\_cyg\_v\]), we plot the time-stretched color-magnitude diagram of V2362 Cyg in Figure \[hr\_diagram\_v705\_cas\_v382\_vel\_v5114\_sgr\_v2362\_cyg\_outburst\](d). We also add the template tracks of LV Vul (thick solid orange line) and V1500 Cyg (thick solid green line). We further add another track of V1500 Cyg (thick solid cyan line) which is shifted toward red by $\Delta (B-V)=0.20$ mag.
V2362 Cyg shows an interesting evolution in the color-magnitude diagram. V2362 Cyg goes down along the LV Vul track (orange line) or the red-shifted V1500 Cyg track (cyan line) just after the optical maximum. In the secondary maximum phase, it goes up along the original track of V1500 Cyg (solid green line) as denoted by open blue circles, and then goes down along the same V1500 Cyg track (solid green line) as denoted by filled magenta circles with black outlines. After the onset of the nebular phase, the track turns to the right and follows again the upper LV Vul track (orange line) or the red-shifted V1500 Cyg track (cyan line).
These distinct two tracks are a hint to resolve the reason why the two types of V1500 Cyg and LV Vul tracks are clearly separated in the time-stretched color-magnitude diagram. @mun08b wrote “the emission spectrum at second maximum was quite different from that at first maximum, reflecting the much higher temperature of the underlying continuum. At second maximum, emission lines were not seen and were replaced by , , , \[\], and .” The strong emission lines (/ as well as Balmer lines such as H$\gamma$, H$\delta$, H$\epsilon$) contribute to the $B$-band and make $B-V$ blue in the secondary maximum phase. Thus, the difference in the $(B-V)_0$ color is real rather than the difference in the $V$ filter response. We can conclude that the two tracks reflect the difference in the ionization state which originates from the temperature difference of the underlying continuum. We call the redder location the LV Vul type and the bluer location the V1500 Cyg type. The separation is $\Delta (B-V)_0\sim 0.2$ mag. Overlapping of these tracks of V2362 Cyg with the two templates novae, LV Vul and V1500 Cyg, supports our values of $(m-M')_V=16.05$ and $E(B-V)=0.60$, that is, $f_{\rm s}=1.78$, $E(B-V)=0.60\pm0.05$, $(m-M)_V=15.4\pm0.2$, and $d=5.1\pm0.5$ kpc. We list our results in Table \[extinction\_various\_novae\].
We reexamine the distance-reddening relation toward V2362 Cyg, $(l, b)= (87\fdg3724,-2\fdg3574)$, based on several distance-reddening relations in Figure \[distance\_reddening\_v705\_cas\_v382\_vel\_v5114\_sgr\_v2362\_cyg\](d). The solid blue line denotes the distance-reddening relation of Equation (\[distance\_modulus\_rv\]) together with $(m-M)_V=15.4$. The vertical solid red line is the color excess of V2362 Cyg estimated in Paper II. These two lines cross at $d=5.1$ kpc (and $E(B-V)=0.60$). We plot the four sets of distance-reddening relation calculated by @mar06; toward $(l, b)= (87\fdg25,-2\fdg25)$ denoted by open red squares, $(87\fdg50,-2\fdg25)$ by filled green squares, $(87\fdg25,-2\fdg50)$ by blue asterisks, and $(87\fdg50,-2\fdg50)$ by open magenta circles. The open red squares are the closest direction toward V2362 Cyg among the four nearby directions. The solid black/orange lines denote the distance-reddening relations given by @gre15 [@gre18], respectively. The solid green line represent the relation given by @sal14 and the open cyan-blue diamonds with error bars indicate the relation given by @ozd16. The solid cyan-blue line is the relation given by @chen18. Our crossing point at $d=5.1$ kpc and $E(B-V)=0.60$ is roughly consistent with Marshall et al.’s (open red squares) and Green et al.’s (solid black/orange lines) relations, although the reddenings given by @sal14 and @ozd16 slightly deviates from the crossing point. Thus, we confirm again that our estimated values of $(m-M)_V=15.4\pm0.2$, $E(B-V)=0.60\pm0.05$, and $d=5.1\pm0.5$ kpc are reasonable.
V2615 Oph 2007 {#v2615_oph_cmd}
--------------
V2615 Oph was also studied in Paper II on the $(B-V)_0$-$M_V$ diagram. In this subsection, we reexamine it on the $(B-V)_0$-$(M_V-2.5\log f_{\rm s}$) diagram.
We obtain three distance moduli in the $B$, $V$, and $I_{\rm C}$ bands in Appendix \[v2615\_oph\]. We plot these three distance moduli in Figure \[distance\_reddening\_v2615\_oph\_v2468\_cyg\_v2491\_cyg\_v496\_sct\](a) by the thin solid magenta, blue, and cyan lines. These three lines cross at $d=4.3$ kpc and $E(B-V)=0.90$. Therefore, we adopt $E(B-V)=0.90\pm0.05$. See Paper II for other various estimates on the reddening and distance. Thus, we obtain the timescaling factor of $f_{\rm s}= 1.58$ against LV Vul and the distance of $d=4.3\pm0.4$ kpc, $E(B-V)=0.90\pm0.05$, and $(m-M)_V=15.95\pm0.2$.
We reanalyze the distance and reddening toward V2615 Oph, $(l, b)=(4\fdg1475, +3\fdg3015)$, based on various distance-reddening relations in Figure \[distance\_reddening\_v2615\_oph\_v2468\_cyg\_v2491\_cyg\_v496\_sct\](a). We plot four distance-reddening relations of @mar06; toward $(l, b)=(4\fdg00, +3\fdg25)$ denoted by open red squares, $(4\fdg25, +3\fdg25)$ by filled green squares, $(4\fdg00, +3\fdg50)$ by blue asterisks, and $(4\fdg25, +3\fdg50)$ by open magenta circles, each with error bars. The direction toward V2615 Oph is between those of open red squares and filled green squares. We add the distance-reddening relations (solid black/orange lines) given by @gre15 [@gre18], respectively. We further add four distance-reddening relations (very thin solid cyan lines) given by @schu14. The open cyan-blue diamonds are the relation given by @ozd16. The cyan-blue line represents the relation given by @chen18; one is toward $(4\fdg15, +3\fdg25)$ and the other is toward $(4\fdg15, +3\fdg35)$. Our crossing point at $E(B-V)=0.90$ and $d=4.3$ kpc is consistent with the both trends of Marshall et al. and Green et al. The NASA/IPAC galactic 2D dust absorption map gives $E(B-V)=0.87\pm0.02$ toward V2615 Oph, being consistent with our value of $E(B-V)=0.90\pm0.05$. We list our results in Table \[extinction\_various\_novae\].
Adopting $E(B-V)=0.90$ and $(m-M')_V=16.45$ in Equation (\[absolute\_mag\_v2615\_oph\_lv\_vul\]), we obtain the time-stretched color-magnitude diagram of V2615 Oph in Figure \[hr\_diagram\_v2615\_oph\_v2468\_cyg\_v2491\_cyg\_v496\_sct\_outburst\](a). Here, we add the time-stretched track of V1065 Cen by the filled cyan triangles. The track of V2615 Oph is similar to that of V1065 Cen until the dust blackout. It broadly follows the LV Vul track. Therefore, V2615 Oph belongs to the LV Vul type in the time-stretched color-magnitude diagram. This overlapping supports our estimates of $E(B-V)=0.90$ and $(m-M')_V=16.45$, that is, $f_{\rm s}=1.58$, $E(B-V)=0.90\pm0.05$, $(m-M)_V=15.95\pm0.2$, and $d=4.3\pm0.4$ kpc.
Taking $(m-M)_V=15.95$ for V2615 Oph, we plot the $V$ model light curve (solid red line) of a $0.90~M_\sun$ WD with the envelope chemical composition of CO nova 3 [@hac16k] in Figure \[v2615\_oph\_lv\_vul\_v1419\_aql\_v\_bv\_ub\_color\_logscale\](a). The model $V$ light curve reasonably reproduces the observation (filled red circles). This confirms that the distance modulus of $(m-M)_V=15.95\pm0.2$ is reasonable.
V2468 Cyg 2008 {#v2468_cyg_cmd}
--------------
V2468 Cyg was studied in Paper II. Here, we reexamine it on the $(B-V)_0$-$(M_V-2.5\log f_{\rm s}$) diagram.
We obtain three distance moduli in the $B$, $V$, and $I_{\rm C}$ bands in Appendix \[v2468\_cyg\]. We plot these three distance moduli in Figure \[distance\_reddening\_v2615\_oph\_v2468\_cyg\_v2491\_cyg\_v496\_sct\](b) by the thin solid magenta, blue, and cyan lines. These three lines cross at $d=6.9$ kpc and $E(B-V)=0.65$. We adopt $f_{\rm s}= 2.4$ against LV Vul, $E(B-V)=0.65\pm0.05$, and $(m-M)_V=16.2\pm0.2$. The previous values in Paper II are $E(B-V)=0.75\pm0.05$, $(m-M)_V=15.6\pm0.2$, and $d=4.5\pm0.5$ kpc, being smaller than our new values. The main difference is the improvement in the timescaling factor. See the discussion in Paper II for a summary of other estimates on the reddening and distance toward V2468 Cyg.
Figure \[distance\_reddening\_v2615\_oph\_v2468\_cyg\_v2491\_cyg\_v496\_sct\](b) also shows several distance-reddening relations toward V2468 Cyg, $(l, b)=(66\fdg8084, +0\fdg2455)$. We plot four distance-reddening relations of @mar06; toward $(l, b)=(66\fdg75, +0\fdg00)$ denoted by open red squares, $(67\fdg00, +0\fdg00)$ by filled green squares, $(66\fdg75, +0\fdg25)$ by blue asterisks, and $(67\fdg00, +0\fdg25)$ by open magenta circles, each with error bars. The closest direction in the galactic coordinates is that of blue asterisks. We add the distance-reddening relations (solid black/orange lines) given by @gre15 [@gre18], respectively. The solid green line represents the relation given by @sal14 and the open cyan-blue diamonds with error bars are given by @ozd16. The cyan-blue line represents the relation given by @chen18. Our crossing point at $E(B-V)=0.65$ and $d=6.9$ kpc is consistent with the distance-reddening relation given by Özdörmez et al. Thus, we confirm that our set of $E(B-V)=0.65\pm0.05$ and $d=6.9\pm0.8$ kpc are reasonable. We list our results in Table \[extinction\_various\_novae\].
Adopting $E(B-V)=0.65$ and $(m-M')_V=17.15$ in Equation (\[absolute\_mag\_v2468\_cyg\_lv\_vul\]), we obtain the time-stretched color-magnitude diagram of V2468 Cyg in Figure \[hr\_diagram\_v2615\_oph\_v2468\_cyg\_v2491\_cyg\_v496\_sct\_outburst\](b). Here, we plot the data of V2468 Cyg taken from AAVSO. Although the color data are rather scattered, V2468 Cyg broadly follows the V1500 Cyg track. Therefore, V2468 Cyg belongs to the V1500 Cyg type. This overlapping supports our estimates of $E(B-V)=0.65$ and $(m-M')_V=17.15$, that is, $f_{\rm s}=2.4$, $E(B-V)=0.65\pm0.05$, $(m-M)_V=16.2\pm0.2$, and $d=6.9\pm0.8$ kpc.
Taking $(m-M)_V=16.2$ for V2468 Cyg, we plot the $V$ model light curve (solid red line) of a $0.85~M_\sun$ WD with the envelope chemical composition of CO nova 4 [@hac15k] in Figure \[v2468\_cyg\_lv\_vul\_v1500\_cyg\_v\_color\_logscale\_no2\](a). The model $V$ light curve reasonably reproduces the upper bound of observation (filled red circles) during $\log t~({\rm day})\sim 1-2$. This suggests that the distance modulus of $(m-M)_V=16.2\pm0.2$ is reasonable.
V2491 Cyg 2008\#2 {#v2491_cyg_cmd}
-----------------
V2491 Cyg was studied in Paper II on the $(B-V)_0$-$M_V$ diagram. In this subsection, we reexamine it on the $(B-V)_0$-$(M_V-2.5\log f_{\rm s}$) diagram. V2491 Cyg showed a small secondary maximum [e.g., @hac09ka]. This nova is also characterized by the detection of pre-outburst X-ray emission [@iba08a; @iba08b]. @hac16kb adopted $E(B-V)=0.23$ after @mun11 while we here adopt a different value of $E(B-V)=0.45$. These two values are significantly different, so we explained the reason below.
Figure \[distance\_reddening\_v2615\_oph\_v2468\_cyg\_v2491\_cyg\_v496\_sct\](c) plots three distance moduli in the $B$, $V$, and $I_{\rm C}$ bands obtained from the time-stretching method in Appendix \[v2491\_cyg\] by the magenta, blue, and cyan lines, that is, $(m-M)_B= 17.83$, $(m-M)_V= 17.41$, and $(m-M)_I= 16.62$ together with Equations (\[distance\_modulus\_rb\]), (\[distance\_modulus\_rv\]), and (\[distance\_modulus\_ri\]). These three lines broadly cross at $d=15.9$ kpc and $E(B-V)=0.45$.
Figure \[distance\_reddening\_v2615\_oph\_v2468\_cyg\_v2491\_cyg\_v496\_sct\](c) also shows several distance-reddening relations toward V2491 Cyg, $(l, b)= (67\fdg2287,+4\fdg3532)$. We plot the distance-reddening relations given by @mar06; toward $(l, b)= (67\fdg00,4\fdg25)$ denoted by open red squares, $(67\fdg25,4\fdg25)$ by filled green squares, $(67\fdg00,4\fdg50)$ by blue asterisks, and $(67\fdg25,4\fdg50)$ by open magenta circles. The closest direction in the galactic coordinates is that denoted by filled green squares. The solid green line represents the relation given by @sal14, black/orange lines by @gre15 [@gre18]. The open cyan-blue diamonds with error bars denote the relation given by @ozd16 and the solid cyan-blue line represents the relation given by @chen18.
Marshall et al.’s distance-reddening relations (green squares and blue asterisks) match the smaller value of $E(B-V)=0.23\pm0.01$. On the other hand, Green et al.’s and Sale et al.’s relations are consistent with the larger value of $E(B-V)=0.45\pm0.05$, which is consistent with our new estimate of $E(B-V)=0.45\pm0.05$.
The reddening toward V2491 Cyg was obtained by @lyn08b to be $E(B-V)=0.3$ from lines, which was revised by @rud08b to be $E(B-V)=0.43$ from the lines at $0.84~\mu$m and $1.13~\mu$m. @mun11 obtained $E(B-V)=0.23\pm0.01$ from an average of $E(B-V)=0.24$ from 5889.953 line profiles, $E(B-V)= (B-V)_{\rm max} - (B-V)_{0, \rm max} =
0.46 - (0.23 \pm 0.06) = 0.23\pm 0.06$, and $E(B-V)= (B-V)_{t2} - (B-V)_{0, t2} = 0.20 - (-0.02\pm 0.04)
= 0.22\pm 0.04$, from the intrinsic colors at maximum and $t_2$ time [@van87]. The distance modulus and distance to V2491 Cyg were estimated by @mun11 as $(m-M)_V= m_{V,\rm max} - M_{V,\rm max}= 7.45 - (-9.06)=16.51$ from the MMRD relation [@coh88] together with $t_2=4.8$ days, and then derived the distance of $d=14$ kpc. On the other hand, the NASA/IPAC Galactic 2D dust absorption map gives $E(B-V)=0.48 \pm 0.03$ in the direction toward V2491 Cyg. To summarize, there are two estimates on the reddening, i.e., $E(B-V)=0.23\pm0.01$ and $E(B-V)=0.43\pm0.05$. Our value is consistent with the larger one. We here adopt $E(B-V)=0.45\pm0.05$ and $d=15.9\pm2$ kpc. The distance modulus in the $V$ band is $(m-M)_V=17.4\pm0.2$. These values are listed in Table \[extinction\_various\_novae\].
Using $E(B-V)=0.45$ and $(m-M')_V=16.55$ from Equation (\[absolute\_mag\_lv\_vul\_v2491\_cyg\]), we plot the time-stretched color-magnitude diagram of V2491 Cyg in Figure \[hr\_diagram\_v2615\_oph\_v2468\_cyg\_v2491\_cyg\_v496\_sct\_outburst\](c). We add the template tracks of LV Vul (thick solid orange line), V1500 Cyg (thick solid green line), and V1974 Cyg (solid magenta line). We also add the V1500 Cyg track shifted toward blue by $\Delta (B-V)=-0.25$ (thick solid cyan line). V2491 Cyg approximately follows the blue-shifted V1500 Cyg track (cyan line) except for during the tiny secondary maximum. The track turns to the right after the onset of the nebular phase (Paper II). This redward excursion is due to large contribution of the strong emission lines \[\] to the $V$ magnitude as already discussed in the previous subsections.
@mun11 suggested the low metallicity of \[Fe/H\]$=-0.25$ for V2491 Cyg. This sub-solar metallicity is roughly consistent with $-0.5<$ \[Fe/H\] $<-0.3$ at the galacto-centric distance ($\sim 14$ kpc) of V2491 Cyg. @hac18kb clarified that the color-magnitude tracks of LMC novae are located at the bluer side by $\Delta (B-V)=-0.3$ than those of galactic novae. For example, the color-magnitude track of YY Dor overlaps with the V1668 Cyg track shifted by $\Delta (B-V)=-0.3$. They supposed that this blue-shift is caused by the lower metallicity of LMC stars, \[Fe/H\]$=-0.55$ [see, e.g., @pia13]. The LMC N 2009a track is in good agreement with the 0.2 mag blue-shifted V1500 Cyg and 0.15 mag blue-shifted U Sco tracks [@hac18kb]. This bluer feature of the LMC N 2009a track is also due to the lower metallicity. The blue-shifted nature of the V2491 Cyg track is consistent with the nature of LMC novae [@hac18kb].
It should be noted that the empirical relations proposed by @van87, i.e., $(B-V)_0= 0.23\pm0.06$ at maximum and $(B-V)_0= -0.02\pm0.04$ at $t_2$, are not applicable to novae of sub-solar metallicity as clearly shown in Figure \[hr\_diagram\_v2615\_oph\_v2468\_cyg\_v2491\_cyg\_v496\_sct\_outburst\](c). This is one of the reason why @mun11 obtain the smaller reddening of $E(B-V)=0.23\pm0.01$.
@mun11 obtained the abundance of ejecta to be $X = 0.573$, $Y = 0.287$, $Z = 0.140$ by mass weight, with those of individual elements being $X_{\rm N} = 0.074$, $X_{\rm O} = 0.049$, and $X_{\rm Ne} = 0.015$ [see also @tar14b for another abundance estimate]. This abundance is close to that of Ne nova 2 [@hac10k]. Taking $(m-M)_V=17.4$ for V2491 Cyg, we plot the $V$ model light curve of $1.35~M_\sun$ (solid red lines) and $1.30~M_\sun$ (solid green lines) WDs with the envelope chemical composition of Ne nova 2 in Figure \[v2491\_cyg\_lv\_vul\_v1500cyg\_cyg\_v\_color\_logscale\](a). The model $V$ light curve reasonably reproduces the observation (filled red circles) except during the secondary maximum. This suggests that the distance modulus of $(m-M)_V=17.4\pm0.2$ is reasonable.
Our model light curve fitting suggests that the WD mass is approximately $1.35~M_\sun$. The WD mass of RS Oph has been estimated to be $\sim 1.35~M_\sun$ by @hac06b [@hac07kl] and @hac18kb. It is interesting to compare the $V$ (or $y$) and supersoft X-ray light curves of these two novae (Figure \[light\_rs\_oph2006\_v2491\_cyg2008\_xray\]). The starting time of the SSS phases are almost the same between the two novae while the duration of the SSS phase is rather different and much longer in the recurrent nova RS Oph than in the classical nova V2491 Cyg.
V496 Sct 2009 {#v496_sct_cmd}
-------------
V496 Sct was studied in Paper II on the $(B-V)_0$-$M_V$ diagram. In this subsection, we reexamine it on the $(B-V)_0$-$(M_V-2.5\log f_{\rm s}$) diagram.
We obtain the distance moduli of $BVI_{\rm C}$ bands in Appendix \[v496\_sct\] and plot them by the magenta, blue, and cyan lines, i.e., $(m-M)_B=14.15$, $(m-M)_V = 13.71$, and $(m-M)_I=13.02$, in Figure \[distance\_reddening\_v2615\_oph\_v2468\_cyg\_v2491\_cyg\_v496\_sct\](d). These three lines broadly cross each other at $d=2.9$ kpc and $E(B-V)=0.45$. Thus, we determine the distance and reddening to be $d=2.9\pm0.3$ kpc and $E(B-V)=0.45\pm0.05$. The distance modulus in the $V$ band is $(m-M)_V = 13.7\pm0.2$.
@hac16kb obtained $E(B-V)=0.50\pm0.05$ by assuming that the intrinsic $B-V$ color curve of V496 Sct is the same as those of FH Ser and NQ Vul. See Paper II for a summary for the other estimates on the reddening. We reanalyze the distance and reddening toward V496 Sct, $(l, b)=(25\fdg2838, -1\fdg7678)$, based on various distance-reddening relations in Figure \[distance\_reddening\_v2615\_oph\_v2468\_cyg\_v2491\_cyg\_v496\_sct\](d). We plot four distance-reddening relations of @mar06; toward $(l, b)=(25\fdg25, -1\fdg75)$ denoted by open red squares, $(25\fdg50, -1\fdg75)$ by filled green squares, $(25\fdg25, -2\fdg00)$ by blue asterisks, and $(25\fdg50, -2\fdg00)$ by open magenta circles, each with error bars. The closest direction in the galactic coordinates is that of open red squares. We add the distance-reddening relations (solid black/orange lines) given by @gre15 [@gre18], respectively. The open cyan-blue diamonds with error bars represent the relation given by @ozd16. The solid cyan-blue line denotes the relation given by @chen18. Our set of $d=2.9$ kpc and $E(B-V)=0.45$ is consistent with the trends of Marshall et al., Green et al., Özdörmez et al, and Chen et al. Our results are listed in Table \[extinction\_various\_novae\].
Adopting $E(B-V)=0.45$ and $(m-M')_V=14.45$ in Equation (\[absolute\_mag\_v496\_sct\_lv\_vul\]), we obtain the time-stretched color-magnitude diagram of V496 Sct in Figure \[hr\_diagram\_v2615\_oph\_v2468\_cyg\_v2491\_cyg\_v496\_sct\_outburst\](d). Here, we plot the data taken from @raj12 and SMARTS [@wal12]. V496 Sct follows the LV Vul track until the dust blackout. Note that, even after the dust blackout, V496 Sct comes back and follows the LV Vul track again (upper branch after the onset of nebular phase). Therefore, V496 Sct belongs to the LV Vul type in the time-stretched color-magnitude diagram. This overlapping supports our estimates of $E(B-V)=0.45$ and $(m-M')_V=14.45$, that is, $f_{\rm s}=2.0$, $E(B-V)=0.45\pm0.05$, $(m-M)_V=13.7\pm0.2$, and $d=2.9\pm0.3$ kpc.
Taking $(m-M)_V=13.7$ for V496 Sct, we plot the $V$ model light curve (solid red line) of a $0.85~M_\sun$ WD with the envelope chemical composition of CO nova 3 [@hac16k] in Figure \[v496\_sct\_lv\_vul\_v1668\_cyg\_v\_bv\_ub\_color\_logscale\_no2\](a). The model $V$ light curve reasonably reproduces the observation (filled red circles). This confirms that the distance modulus of $(m-M)_V=13.7\pm0.2$ is reasonable.
Discussion
==========
Comparison with Gaia DR2 distances {#gaia_distance}
----------------------------------
Recently, @schaefer18 listed distances of 64 novae from Gaia data release 2 (DR2) in his Table 1. Among them, seven novae dubbed the “very well observed light curve ($< 30$% error)” coincide with our analyzed novae in Table \[extinction\_various\_novae\]. These distances are compared with the present results (Gaia DR2) as follows: CI Aql, 3.3 kpc (3.19 kpc); V705 Cas, 2.6 kpc (2.16 kpc); V1974 Cyg, 1.8 kpc (1.63 kpc): V446 Her, 1.38 kpc (1.36 kpc); V533 Her, 1.28 kpc (1.20 kpc); V382 Vel 1.4 kpc (1.80 kpc), PW Vul 1.8 kpc (2.42 kpc). These values are in good agreement within each error box.
Conclusions
===========
We have obtained the following results:\
[**1.**]{} We improved the timescaling factor $f_{\rm s}$ of each nova by overlapping not only $V$ light curves but also $(B-V)_0$ and $(U-B)_0$ color curves. Applying the improved $f_{\rm s}$ to the time-stretching method, we revised the distance modulus in the $V$ band $\mu_V=(m-M)_V$. The results are summarized in Table \[extinction\_various\_novae\].\
[**2.**]{} We also reexamine the color excess $E(B-V)$ from the multi-band time-stretching method. We obtain the distance moduli in the $UBVI_{\rm C}K_{\rm s}$ bands, $(m-M)_U$, $(m-M)_B$, $(m-M)_V$, $(m-M)_I$, and $(m-M)_K$, of each nova. The crossing point of these distance-reddening relations gives a reasonable reddening and distance toward the nova. The results are also summarized in Table \[extinction\_various\_novae\].\
[**3.**]{} With the improved values of $f_{\rm s}$, $(m-M)_V$, and $E(B-V)$, we obtain the $(B-V)_0$-$(M_V-2.5\log f_{\rm s})$ color-magnitude diagram. We call it the time-stretched color-magnitude diagram. In general, each nova evolves from the upper right (red) to the lower left (blue) and then turns back toward the right (red) at the onset of the nebular phase. We found two representative tracks in this diagram, the tracks of LV Vul and V1500 Cyg. The template LV Vul goes down along the line of $(B-V)_0=-0.03$, the color of optically thick free-free emission, in the middle part of the track. The template V1500 Cyg track is almost parallel to, but located at the bluer side of $\Delta (B-V)_0\sim -0.2$ mag than the LV Vul track. This difference is caused by the difference in the ionization state which originates from the temperature difference of the underlying continuum.\
[**4.**]{} Among the eight novae studied in Sections \[method\_example\]–\[v5666\_sgr\_cmd\], V1668 Cyg (see Figure \[hr\_diagram\_v446\_her\_v533\_her\_pw\_vul\_v1419\_aql\_outburst\](d)), V679 Car, V1369 Cen, and V5666 Sgr follow the template track of LV Vul while V1974 Cyg (see Figure \[hr\_diagram\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\_outburst\](a)) and V574 Pup follow the template track of V1500 Cyg.\
[**5.**]{} We reanalyzed additional 12 novae in Section \[twelve\_novae\] on the time-stretched color-magnitude diagram. Among the 12 novae, V446 Her, V1419 Aql, V705 Cas, V2615 Oph, and V496 Sct broadly follow the LV Vul (or V1668 Cyg) template track, while V533 Her, PW Vul, V382 Vel, V5114 Sgr, V2468 Cyg, and V2491 Cyg follow the template track of V1500 Cyg (or V1974 Cyg). Only V2362 Cyg follows the track of LV Vul in the first decline and then follows the V1500 Cyg track during the secondary maximum (a large rebrightening). Thus, we establish the two representative tracks in the $(B-V)_0$-$(M_V-2.5\log f_{\rm s})$ diagram.\
[**6.**]{} The location of V2491 Cyg track is about 0.25 mag bluer than that of the original V1500 Cyg track (Figure \[hr\_diagram\_v2615\_oph\_v2468\_cyg\_v2491\_cyg\_v496\_sct\_outburst\](c)). This bluer location is due to a lower metallicity of the nova ejecta (e.g., subsolar by \[Fe$/$H\]$= -0.25$) as suggested by @mun11. This kind of low metallicity effect is already discussed in @hac18kb for LMC novae.\
[**7.**]{} We estimated the white dwarf masses and $(m-M)_V$ of the novae by directly fitting the absolute $V$ model light curves ($M_V$) with observational apparent $V$ magnitudes ($m_V$). The obtained results of $(m-M)_V$ are in good agreement with the estimates by the time-stretched color-magnitude diagram method.\
[**8.**]{} The white dwarf masses are estimated from the $V$, UV 1455Å, and supersoft X-ray light curve fittings, assuming an appropriate chemical composition of ejecta. They are $0.98~M_\sun$ (LV Vul, CO3), $1.2~M_\sun$ (V1500 Cyg, Ne2), $0.98~M_\sun$ (V1668 Cyg, CO3), $0.98~M_\sun$ (V1974 Cyg, CO3), $1.05~M_\sun$ (V574 Pup, Ne3), $0.98~M_\sun$ (V679 Car, CO3), $0.90~M_\sun$ (V1369 Cen, CO3), $0.85~M_\sun$ (V5666 Sgr, CO3), $0.98~M_\sun$ (V446 Her, CO3), $1.03~M_\sun$ (V533 Her, Ne2), $0.83~M_\sun$ (PW Vul, CO4), $0.90~M_\sun$ (V1419 Aql, CO3), $0.78~M_\sun$ (V705 Cas, CO4), $1.23~M_\sun$ (V382 Vel, Ne2), $1.15~M_\sun$ (V5114 Sgr, Ne2), $0.85~M_\sun$ (V2362 Cyg, interpolation), $0.90~M_\sun$ (V2615 Oph, CO3), $0.85~M_\sun$ (V2468 Cyg, CO3), $1.35~M_\sun$ (V2491 Cyg, Ne2), and $0.85~M_\sun$ (V496 Sct, CO3). These results are summarized in Table \[wd\_mass\_novae\].\
[**9.**]{} Our distance estimates are in good agreement with the results of Gaia Data Release 2.
We express our gratitude to T. Iijima and the Astronomical Observatory of Padova (Asiago) for the warm hospitality during which we initiated the present work. We are grateful to the late A. Cassatella for providing us with UV 1455 Å data for [*IUE*]{} novae. We thank the American Association of Variable Star Observers (AAVSO) and the Variable Star Observers League of Japan (VSOLJ) for the archival data of various novae. We are also grateful to the anonymous referee for useful comments regarding how to improve the manuscript. This research has been supported in part by Grants-in-Aid for Scientific Research (15K05026, 16K05289) from the Japan Society for the Promotion of Science.
Time-stretched Light Curves of V574 Pup, V679 Car, V1369 Cen, and V5666 Sgr {#v574_pup_v679_car_v1369_cen_v5666_sgr}
===========================================================================
V574 Pup 2004 {#v574_pup}
-------------
Figure \[v574\_pup\_v\_bv\_ub\_color\_curve\] shows (a) the $V$ light curve and (b) $(B-V)_0$ color curve of V574 Pup on a linear timescale. We deredden the color with $E(B-V)=0.45$ as obtained in Section \[distance\_reddening\_v574\_pup\]. The $BV$ data are taken from the Small Medium Aperture Telescope System (SMARTS) data base [@wal12], American Association of Variable Star Observers (AAVSO), and the Variable Star Observers League of Japan (VSOLJ). The $B-V$ data of AAVSO (open blue circles) are systematically bluer by 0.2 mag compared with the other two colors of SMARTS and VSOLJ, so we shift them toward red by 0.2 mag in Figure \[v574\_pup\_v\_bv\_ub\_color\_curve\](b). Such differences come from slightly different responses of $V$ filters at different observatories [see, e.g., discussion in @hac06kb]. The horizontal solid red line indicates the $B-V$ color of optically thick free-free emission, i.e., $(B-V)_0=-0.03$ (see Paper I). The typical error of SMARTS $BVRI$ data is about 0.001 mag, being much smaller than the size of each symbol. The error of AAVSO and VSOLJ data are not reported.
Figure \[v574\_pup\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_k\_logscale\_4fig\] shows the (a) $B$, (b) $V$, (c) $I_{\rm C}$, and (d) $K_{\rm s}$ light curves of V574 Pup together with YY Dor and LMC N 2009a. Here, we assume that V574 Pup outbursted on JD 2453326.0 (day 0). @hac18kb have already analyzed YY Dor and LMC N 2009a and their various properties are summarized in their Tables 1, 2, and 3. We increase/decrease the horizontal shift by a step of $\delta (\Delta \log t) = \delta \log f_{\rm s} = 0.01$ and the vertical shift by a step of $\delta (\Delta V)= 0.1$ mag, and determine the best overlapping one by eye. As mentioned in Section \[introduction\], the model light curves obey the universal decline law. This phase corresponds to the optically thick wind phase before the nebular phase started. Therefore, we should use only the optically thick phase. However, we usually use all the phases if they overlap each other even after the nebular phase in order to determine the timescaling factor accurately as much as possible. If not, we use only the part of optically thick wind phase before the nebular (or dust blackout) phase started. It should be noted that the timescaling factor of $f_{\rm s}$ is common among these four band light curves. In the case of Figure \[v574\_pup\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_k\_logscale\_4fig\], the four band light curves well overlap each other even after the nebular phase started. For the $B$ band, we apply Equation (\[distance\_modulus\_general\_temp\_b\]) to Figure \[v574\_pup\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_k\_logscale\_4fig\](a) and obtain $$\begin{aligned}
(m&-&M)_{B, \rm V574~Pup} \cr
&=& ((m - M)_B + \Delta B)_{\rm YY~Dor} - 2.5 \log 6.6 \cr
&=& 18.98 - 1.5\pm0.2 - 2.05 = 15.43\pm0.2 \cr
&=& ((m - M)_B + \Delta B)_{\rm LMC~N~2009a} - 2.5 \log 4.2 \cr
&=& 18.98 - 2.0\pm0.2 - 1.55 = 15.43\pm0.2,
\label{distance_modulus_b_yy_dor_u_sco_lmcn2009a}\end{aligned}$$ where we adopt $(m-M)_B = 18.49 + 4.1\times 0.12=18.98$ for the LMC novae. Thus, we obtain $(m-M)_{B, \rm V574~Pup}= 15.43\pm0.1$.
For the $V$ light curves in Figure \[v574\_pup\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_k\_logscale\_4fig\](b), we similarly obtain $$\begin{aligned}
(m&-&M)_{V, \rm V574~Pup} \cr
&=& ((m - M)_V + \Delta V)_{\rm YY~Dor} - 2.5 \log 6.6 \cr
&=& 18.86 - 1.8\pm0.2 - 2.05 = 15.01\pm0.2 \cr
&=& ((m - M)_V + \Delta V)_{\rm LMC~N~2009a} - 2.5 \log 4.2 \cr
&=& 18.86 - 2.3\pm0.2 -1.55 = 15.01\pm0.2,
\label{distance_modulus_v_yy_dor_u_sco_lmcn2009a}\end{aligned}$$ where we adopt $(m-M)_V = 18.49 + 3.1\times 0.12= 18.86$ for the LMC novae. Thus, we obtain $(m-M)_{V, \rm V574~Pup}= 15.0\pm0.1$.
We apply Equation (\[distance\_modulus\_general\_temp\_i\]) for the $I_{\rm C}$ band to Figure \[v574\_pup\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_k\_logscale\_4fig\](c) and obtain $$\begin{aligned}
(m&-&M)_{I, \rm V574~Pup} \cr
&=& ((m - M)_I + \Delta I_C)_{\rm YY~Dor} - 2.5 \log 6.6 \cr
&=& 18.67 - 2.4\pm0.3 - 2.05 = 14.22\pm0.3 \cr
&=& ((m - M)_I + \Delta I_C)_{\rm LMC~N~2009a} - 2.5 \log 4.2 \cr
&=& 18.67 - 2.9\pm0.3 -1.55 = 14.22\pm0.3,
\label{distance_modulus_i_yy_dor_u_sco_lmcn2009a}\end{aligned}$$ $(m-M)_I = 18.49 + 1.5\times 0.12=18.67$ for the LMC novae. Thus, we obtain $(m-M)_{I, \rm V574~Pup}= 14.22\pm0.2$.
For the $K_s$ light curves in Figure \[v574\_pup\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_k\_logscale\_4fig\](d), we apply Equation (\[distance\_modulus\_general\_temp\_k\]) and obtain $$\begin{aligned}
(m&-&M)_{K, \rm V574~Pup} \cr
&=& ((m - M)_K + \Delta K_s)_{\rm YY~Dor} - 2.5 \log 6.6 \cr
&=& 18.53 - 2.8\pm0.4 - 2.05 = 13.68\pm0.4 \cr
&=& ((m - M)_K + \Delta K_s)_{\rm LMC~N~2009a} - 2.5 \log 4.2 \cr
&=& 18.53 - 3.3\pm0.4 -1.55 = 13.68\pm0.4,
\label{distance_modulus_k_yy_dor_u_sco_lmcn2009a}\end{aligned}$$ $(m-M)_{K, \rm YY~Dor}=18.49 + 0.35\times 0.12= 18.53$ for the LMC novae. Thus, we obtain $(m-M)_{K, \rm V574~Pup}= 13.68\pm0.2$.
Figure \[v574\_pup\_v1974\_cyg\_v\_bv\_logscale\] shows the $V$ light curve and $(B-V)_0$ color curve of V574 Pup on a logarithmic timescale in comparison with V1974 Cyg. The V1974 Cyg data are taken from Figure 38 of Paper II. We add visual magnitudes (small red dots) of V574 Pup, which are taken from AAVSO. The $V$ light curves of these two novae do not exactly but broadly overlap each other. Note that there are two different values of V1974 Cyg $V$ magnitudes in the nebular phase. This is because the response of each $V$ filter is different between these two observations. We set the $V$ light curve of V574 Pup to overlap with the lower (fainter) branch of the V1974 Cyg light curves. We also take into account the $(B-V)_0$ color evolution in Figure \[v574\_pup\_v1974\_cyg\_v\_bv\_logscale\](b) to determine the horizontal shift of $\Delta \log t= \log f_{\rm s}$. We set $\log f_{\rm s}$ to overlap the two $(B-V)_0$ color evolution as much as possible.
Applying the obtained $f_{\rm s}$ and $\Delta V$ to Equation (\[distance\_modulus\_general\_temp\]), we have the relation of $$\begin{aligned}
(m&-&M)_{V, \rm V574~Pup} \cr
&=& \left( (m-M)_V + \Delta V\right)_{\rm V1974~Cyg} - 2.5 \log 1.17 \cr
&=& 12.2 + 3.0\pm0.2 - 0.175 = 15.025\pm0.3,
\label{distance_modulus_v574_pup_v1500_cyg_v1668_cyg_v1974_cyg}\end{aligned}$$ where we adopt $(m-M)_{V, \rm V1974~Cyg}=12.2$ in Section \[v1974\_cyg\]. Thus, we obtain $f_{\rm s}=1.07\times 1.17= 1.26$ against LV Vul (see Table \[extinction\_various\_novae\]) and $(m-M)_V=15.0\pm0.3$ for V574 Pup.
Using $(m-M)_V=15.0\pm0.3$ and $f_{\rm s}= 1.26$, we have the relation of $$\begin{aligned}
(m-M')_{V, \rm V574~Pup}
&\equiv& (m_V - (M_V - 2.5\log f_{\rm s}))_{\rm V574~Pup} \cr
&=& ((m - M)_V + 2.5\log f_{\rm s})_{\rm V574~Pup} \cr
&=& 15.0\pm0.3 + 2.5\times 0.10 = 15.25\pm0.3.
\label{absolute_mag_lv_vul_v574_pup_only}\end{aligned}$$
V679 Car 2008 {#v679_car}
-------------
Figure \[v679\_car\_v\_bv\_ub\_color\_curve\] shows the (a) $V$ and (b) $(B-V)_0$ evolutions of V679 Car on a linear timescale. The $BV$ data are taken from AAVSO (open blue circles), VSOLJ (filled green stars), and SMARTS (filled red circles). Here, $(B-V)_0$ are dereddened with $E(B-V)=0.69$ as obtained in Section \[distance\_reddening\_v679\_car\], but the $B-V$ data of VSOLJ (filled green stars) are systematically shifted toward red by 0.25 mag to overlap them to the other color data. The typical error of SMARTS $BVRI$ data is about 0.003 mag, being much smaller than the size of each symbol. The error of AAVSO and VSOLJ data are not reported.
Figure \[v679\_car\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\] shows the $B$, $V$, and $I_{\rm C}$ light curves of V679 Car together with those of the LMC novae YY Dor and LMC N 2009a. Here, we assume that the V679 Car outburst started on JD 2454794.5 (day 0). As in Figure \[v679\_car\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](c), the $I_{\rm C}$ light curve of V679 Car overlaps with those of YY Dor and LMC N 2009a only in the first 60 days because of the different contributions of line emission in the nebular phase. Thus, we use only the $V$ light curves that overlap longer period to determine the timescaling factor $f_{\rm s}$.
We apply Equation (\[distance\_modulus\_general\_temp\_b\]) for the $B$ band to Figure \[v679\_car\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](a) and obtain $$\begin{aligned}
(m&-&M)_{B, \rm V679~Car} \cr
&=& ((m - M)_B + \Delta B)_{\rm YY~Dor} - 2.5 \log 5.2 \cr
&=& 18.98 - 0.4\pm0.2 - 1.8 = 16.78\pm0.2 \cr
&=& ((m - M)_B + \Delta B)_{\rm LMC~N~2009a} - 2.5 \log 3.3 \cr
&=& 18.98 - 0.9\pm0.2 - 1.3 = 16.78\pm0.2.
\label{distance_modulus_b_v679_car_yy_dor_lmcn2009a}\end{aligned}$$ Thus, we obtain $(m-M)_{B, \rm V679~Car}= 16.78\pm0.1$.
For the $V$ light curves in Figure \[v679\_car\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](b), we similarly obtain $$\begin{aligned}
(m&-&M)_{V, \rm V679~Car} \cr
&=& ((m - M)_V + \Delta V)_{\rm YY~Dor} - 2.5 \log 5.2 \cr
&=& 18.86 - 1.0\pm0.2 - 1.8 = 16.06\pm0.2 \cr
&=& ((m - M)_V + \Delta V)_{\rm LMC~N~2009a} - 2.5 \log 3.3 \cr
&=& 18.86 - 1.5\pm0.2 - 1.3 = 16.06\pm0.2.
\label{distance_modulus_v_v679_car_yy_dor_lmcn2009a}\end{aligned}$$ Thus, we obtain $(m-M)_{V, \rm V679~Car}= 16.06\pm0.1$.
We apply Equation (\[distance\_modulus\_general\_temp\_i\]) to Figure \[v679\_car\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](c) and obtain $$\begin{aligned}
(m&-&M)_{I, \rm V679~Car} \cr
&=& ((m - M)_I + \Delta I_C)_{\rm YY~Dor} - 2.5 \log 5.2 \cr
&=& 18.67 - 1.9\pm0.3 - 1.8 = 14.97\pm 0.3 \cr
&=& ((m - M)_I + \Delta I_C)_{\rm LMC~N~2009a} - 2.5 \log 3.3 \cr
&=& 18.67 - 2.4\pm0.3 -1.3 = 14.97\pm 0.3.
\label{distance_modulus_i_v679_car_yy_dor_lmcn2009a}\end{aligned}$$ Thus, we obtain $(m-M)_{I, \rm V679~Car}= 14.97\pm0.2$.
Figure \[v679\_car\_lv\_vul\_v1668\_cyg\_b\_v\_logscale\_2fig\] show the comparison with the galactic novae LV Vul and V1668 Cyg in the (a) $B$ and (b) $V$ light curves. Note that we do not time-stretch these two novae because their timescales are almost the same as that of V679 Car.
Applying Equation (\[distance\_modulus\_general\_temp\_b\]) for the $B$ band to Figure \[v679\_car\_lv\_vul\_v1668\_cyg\_b\_v\_logscale\_2fig\](a), we have the relation of $$\begin{aligned}
(m&-&M)_{B, \rm V679~Car} \cr
&=& ((m - M)_B + \Delta B)_{\rm LV~Vul} - 2.5 \log 1.0 \cr
&=& 12.45 + 4.3\pm0.2 + 0.0 = 16.75\pm 0.2 \cr
&=& ((m - M)_B + \Delta B)_{\rm V1668~Cyg} - 2.5 \log 1.0 \cr
&=& 14.9 + 1.9\pm0.2 + 0.0 = 16.8\pm 0.2.
\label{distance_modulus_b_v679_car_lv_vul_v1668_cyg}\end{aligned}$$ Thus, we obtain $(m-M)_{B, \rm V679~Car}= 16.78\pm0.2$, being consistent with Equation (\[distance\_modulus\_b\_v679\_car\_yy\_dor\_lmcn2009a\]).
Applying Equation (\[distance\_modulus\_general\_temp\]) to Figure \[v679\_car\_lv\_vul\_v1668\_cyg\_b\_v\_logscale\_2fig\](b), we have the relation of $$\begin{aligned}
(m &-& M)_{V, \rm V679~Car} \cr
&=& (m-M)_{V, \rm LV~Vul} + \Delta V - 2.5 \log 1.0 \cr
&=& 11.85 + 4.2\pm 0.2 + 0.0 = 16.05\pm 0.2 \cr
&=& (m-M)_{V, \rm V1668~Cyg} + \Delta V - 2.5 \log 1.0 \cr
&=& 14.6 + 1.5\pm 0.2 + 0.0 = 16.1\pm 0.2.
\label{distance_modulus_v679_car_lv_vul_v1668_cyg_v}\end{aligned}$$ Thus, we obtain $f_{\rm s}=1.0$ against LV Vul and $(m-M)_V=16.08\pm0.1$ for V679 Car, again being consistent with Equation (\[distance\_modulus\_v\_v679\_car\_yy\_dor\_lmcn2009a\]). Thus, we confirm that both the time-stretching methods for the galactic novae and LMC novae give the same results for the $B$ and $V$ bands of V679 Car.
Figure \[v679\_car\_lv\_vul\_v1668\_cyg\_v\_bv\_ub\_logscale\] shows the same $V$ light curve as in Figure \[v679\_car\_lv\_vul\_v1668\_cyg\_b\_v\_logscale\_2fig\](b), but also shows the color evolution. We confirm that we do not need to stretch the light/color curves because its timescale is almost the same as those of LV Vul and V1668 Cyg. We shift the $V$ light curves of LV Vul by 4.2 mag and V1668 Cyg by 1.5 mag downward in order to overlap these $V$ light curves.
The $(B-V)_0$ color evolution in Figure \[v679\_car\_lv\_vul\_v1668\_cyg\_v\_bv\_ub\_logscale\](b) is also used to constrain the horizontal shift of $\Delta \log t= \log f_{\rm s}= 0.0$. The $(B-V)_0$ color evolution of LV Vul splits into two branches just after the nebular phase started as shown in Figure \[v679\_car\_lv\_vul\_v1668\_cyg\_v\_bv\_ub\_logscale\](b). This is because each response function of $V$ filters is slightly different from each other between these two observations. We set the $(B-V)_0$ color evolution of V679 Car to overlap with the lower branch of LV Vul in Figure \[v679\_car\_lv\_vul\_v1668\_cyg\_v\_bv\_ub\_logscale\](b). The timescaling factor of $f_{\rm s}=1.0$ gives a good match between the two $(B-V)_0$ color curves.
From Equations (\[time-stretching\_general\]), (\[distance\_modulus\_general\_temp\]), and (\[distance\_modulus\_v679\_car\_lv\_vul\_v1668\_cyg\_v\]), we have the relation of $$\begin{aligned}
(m- M')_{V, \rm V679~Car}
&\equiv& (m_V - (M_V - 2.5\log f_{\rm s}))_{\rm V679~Car} \cr
&=& \left( (m-M)_V + \Delta V \right)_{\rm LV~Vul} \cr
&=& 11.85 + 4.2\pm0.2 = 16.05\pm0.2.
\label{absolute_mag_lv_vul_v679_car_only}\end{aligned}$$
V1369 Cen 2013 {#v1369_cen}
--------------
Figure \[v1369\_cen\_v\_bv\_ub\_color\_curve\] shows the (a) $V$ light and (b) $(B-V)_0$ color curves of V1369 Cen, where $(B-V)_0$ are dereddened with $E(B-V)=0.11$ as obtained in Section \[v1369\_cen\_cmd\]. The $V$ data of AAVSO (blue open circles), VSOLJ (filled green stars), and SMARTS (filled red circles) are very similar to each other.
Figure \[v1369\_cen\_v496\_sct\_lv\_vul\_v\_bv\_ub\_color\_logscale\] shows the light/color curves of V1369 Cen as well as LV Vul and V496 Sct on a logarithmic timescale. Here we assume that the nova outbursted on JD 2456627.5 (Day 0). We add a straight solid black line labeled “$t^{-3}$” that indicates the homologously expanding ejecta, i.e., free expansion, after the optically thick winds stop [see, e.g., @woo97; @hac06kb]. Both the V1369 Cen and V496 Sct light curves have wavy structures in the early phase, but we overlap these three novae light/color curves as much as possible. In the middle phase, both V1369 Cen and V496 Sct show a sharp and shallow dip due to dust blackout. In the later nebular phase, the $V$ light curves of V1369 Cen and V496 Sct well overlap the upper branch of LV Vul. The $V$ light curve of LV Vul splits into two branches in the nebular phase due to the different responses of their $V$ filters as discussed in Paper II. This is because strong \[\] lines contributes to the blue edge of $V$ filter in the nebular phase and small differences among the response functions of $V$ filters make large differences in the $V$ magnitudes. The $(B-V)_0$ color curves of LV Vul also splits into two branches for $\log t\gtrsim 2.0$, that is, in the nebular phase. The $V$ light curve of V1369 Cen follows V496 Sct and the upper branch of LV Vul, while the $(B-V)_0$ color curve of V1369 Cen follows V496 Sct and the lower branch of LV Vul.
Applying Equation (\[distance\_modulus\_general\_temp\]) to Figure \[v1369\_cen\_v496\_sct\_lv\_vul\_v\_bv\_ub\_color\_logscale\], we have the relation of $$\begin{aligned}
(m&-&M)_{V, \rm V1369~Cen} \cr
&=& (m-M + \Delta V)_{V, \rm LV~Vul} - 2.5 \log 1.48 \cr
&=& 11.85 - 1.2\pm 0.2 - 0.425 = 10.23\pm 0.2 \cr
&=& (m-M + \Delta V)_{V, \rm V496~Sct} - 2.5 \log 0.74 \cr
&=& 13.7 - 3.8\pm 0.2 + 0.325 = 10.23\pm 0.2.
\label{distance_modulus_v1369_cen_v496_sct}\end{aligned}$$ Thus, we obtain $f_{\rm s}=1.48$ against LV Vul and $(m-M)_V=10.23\pm0.2$ for V1369 Cen. These values are summarized in Table \[extinction\_various\_novae\]
From Equations (\[time-stretching\_general\]), (\[distance\_modulus\_general\_temp\]), and (\[distance\_modulus\_v1369\_cen\_v496\_sct\]), we have the relation of $$\begin{aligned}
(m&-& M')_{V, \rm V1369~Cen} \cr
&\equiv& (m_V - (M_V - 2.5\log f_{\rm s}))_{\rm V1369~Cen} \cr
&=& \left( (m-M)_V + \Delta V \right)_{\rm LV~Vul} \cr
&=& 11.85 - 1.2\pm0.2 = 10.65\pm0.2.
\label{absolute_mag_v1369_cen_lv_vul}\end{aligned}$$
We further compare with other novae having similar light curves to V1369 Cen, i.e., V496 Sct and V5666 Sgr. Figure \[v1369\_cen\_v496\_sct\_v5666\_sgr\_b\_v\_i\_k\_logscale\_4fig\] shows the time-stretched light curves of $B$, $V$, $I_{\rm C}$, and $K_{\rm s}$ bands. The light curves well overlap except for the $K_{\rm s}$ band. Applying Equation (\[distance\_modulus\_general\_temp\_b\]) for the $B$ band to Figure \[v1369\_cen\_v496\_sct\_v5666\_sgr\_b\_v\_i\_k\_logscale\_4fig\](a), we have the relation of $$\begin{aligned}
(m&-&M)_{B, \rm V1369~Cen} \cr
&=& \left( (m-M)_B + \Delta B\right)_{\rm V496~Sct} - 2.5 \log 0.74 \cr
&=& 14.15 - 4.1\pm0.2 + 0.325 = 10.38\pm0.2,
\label{distance_modulus_v1369_cen_v496_sct_v5666_sgr_b}\end{aligned}$$ where we adopt $(m-M)_{B, \rm V496~Sct}=13.7 + 1.0\times 0.45= 14.15$ from Section \[v496\_sct\_cmd\]. Thus, we obtain $(m-M)_B=10.38\pm0.2$ for V1369 Cen. We obtain $d=0.97\pm0.1$ kpc from Equation (\[distance\_modulus\_rb\]) together with $E(B-V)=0.11$ and $(m-M)_B=10.38\pm0.2$. We plot this relation of $(m-M)_B=10.38$ by the solid magenta line in Figure \[distance\_reddening\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\](c).
Applying Equation (\[distance\_modulus\_general\_temp\]) to Figure \[v1369\_cen\_v496\_sct\_v5666\_sgr\_b\_v\_i\_k\_logscale\_4fig\](b), we have the relation of $$\begin{aligned}
(m&-&M)_{V, \rm V1369~Cen} \cr
&=& \left( (m-M)_V + \Delta V\right)_{\rm V496~Sct} - 2.5 \log 0.74 \cr
&=& 13.7 - 3.8\pm0.2 + 0.325 = 10.23\pm0.2,
\label{distance_modulus_v1369_cen_v496_sct_v5666_sgr_v}\end{aligned}$$ where we adopt $(m-M)_{V, \rm V496~Sct}=13.7$ from Section \[v496\_sct\_cmd\]. Thus, we obtain $(m-M)_V=10.23\pm0.2$ for V1369 Cen. We obtain $d=0.95\pm0.1$ kpc from Equation (\[distance\_modulus\_rv\]) together with $E(B-V)=0.11$. We plot this relation of $(m-M)_V=10.23$ by the solid blue line in Figure \[distance\_reddening\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\](c).
From the $I_{\rm C}$ band data in Figure \[v1369\_cen\_v496\_sct\_v5666\_sgr\_b\_v\_i\_k\_logscale\_4fig\](a), we obtain $$\begin{aligned}
(m&-&M)_{I, \rm V1369~Cen} \cr
&=& ((m - M)_I + \Delta I_C)_{\rm V496~Sct} - 2.5 \log 0.74 \cr
&=& 12.98 - 3.2\pm0.2 + 0.325 = 10.11\pm0.2,
\label{distance_modulus_i_v1369_cen_v496_sct_v5666_sgr}\end{aligned}$$ where we adopt $(m-M)_{I, \rm V496~Sct}=13.7 - 1.6\times 0.45= 12.98$. Thus, we obtain $(m-M)_{I, \rm V574~Pup}= 10.11\pm0.2$. We obtain $d=0.97\pm0.1$ kpc from Equation (\[distance\_modulus\_ri\]) together with $E(B-V)=0.11$. We plot this relation of $(m-M)_I=10.11$ by the solid cyan line in Figure \[distance\_reddening\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\](c).
V1369 Cen and V496 Sct showed a shallow dust blackout around 100 days ($\log t \sim 2.0$) after the outbursts (Figure \[v1369\_cen\_v496\_sct\_v5666\_sgr\_b\_v\_i\_k\_logscale\_4fig\]). Correspondingly, the $K_{\rm s}$ light curves have an enhancement above the line of universal decline law ($F_\nu\propto t^{-1.75}$), as shown in Figure \[v1369\_cen\_v496\_sct\_v5666\_sgr\_b\_v\_i\_k\_logscale\_4fig\](d). In such a case, we do not use this part for overlapping. In the later phase, however, the decline trend of V1369 Cen becomes similar to those of V496 Sct and V5666 Sgr. We, thus, try to overlap as much as possible in the later phase. Applying Equation (\[distance\_modulus\_general\_temp\_k\]) for the $K_{\rm s}$ band, we obtain $$\begin{aligned}
(m&-&M)_{K, \rm V1369~Cen} \cr
&=& ((m - M)_K + \Delta K_{\rm s})_{\rm V496~Sct} - 2.5 \log 0.74 \cr
&=& 12.46 - 2.8\pm0.3 + 0.325 = 9.99\pm 0.3,
\label{distance_modulus_k_v1369_cen_v496_sct_v5666_sgr}\end{aligned}$$ where we adopt $(m-M)_{K, \rm V496~Sct}=13.7 - 2.75\times 0.45= 12.46$. Thus, we obtain $(m-M)_{K, \rm V1369~Cen}= 9.99\pm0.3$. The distance is determined to be $d=0.98\pm0.3$ kpc from Equation (\[distance\_modulus\_rk\]) together with $E(B-V)=0.11$. This distance is consistent with those determined from $B$, $V$, and $I_{\rm C}$ bands. We plot this relation of $(m-M)_K=9.99$ by the solid cyan-blue line in Figure \[distance\_reddening\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\](c).
Figure \[all\_mass\_v1369\_cen\_v1668\_cyg\_x45z02c15o20\] shows comparison with V1668 Cyg, one of well studied novae. The timescale of V1668 Cyg is stretched with $f_{\rm s}=1.48$ and the $V$ magnitude difference is $\Delta V=-3.9$. From the time-stretching method, we have the relation of $$\begin{aligned}
(m&-&M)_{V, \rm V1369~Cen} \cr
&=& (m-M + \Delta V)_{V, \rm V1668~Cyg} - 2.5 \log 1.48 \cr
&=& 14.6 - 3.9\pm0.2 - 0.43 = 10.27\pm0.2,
\label{distance_modulus_v1369_cen_v1668_cyg}\end{aligned}$$ where we adopt $(m-M)_{V, \rm V1668~Cyg}=14.6$ from Section \[v1668\_cyg\]. Thus, we again confirm that $f_{\rm s}=1.48$ against LV Vul and $(m-M)_V=10.27\pm0.2$. To summarize, we adopt $(m-M)_V=10.25\pm0.1$ from various estimates based on the time-stretching method.
V5666 Sgr 2014 {#v5666_sgr}
--------------
Figure \[v5666\_sgr\_v\_bv\_ub\_color\_curve\] shows (a) the $V$ and visual light and (b) $(B-V)_0$ color curves of V5666 Sgr. The $V$ and visual data of AAVSO (blue open circles and magenta dots) and $V$ data of VSOLJ (filled green stars) are systematically shifted downward by 0.3 mag and 0.2 mag, respectively, to overlap them to the SMARTS data (filled red circles). Here, $(B-V)_0$ are dereddened with $E(B-V)=0.50$ as obtained in Section \[v5666\_sgr\_cmd\]. The $B-V$ data of AAVSO (blue open circles) and VSOLJ (filled green stars) are systematically shifted upward by 0.03 mag and 0.02 mag, respectively, to overlap them to the SMARTS data (filled red circles).
Figure \[v5666\_sgr\_v1369\_cen\_v496\_sct\_v\_bv\_ub\_color\_logscale\] shows (a) the $V$ light curves of V5666 Sgr, LV Vul, V496 Sct, and V1369 Cen, and (b) their $(B-V)_0$ color curves on logarithmic timescales. Here we assume that the V5666 Sgr outbursted on JD 2456681.0 (Day 0). Figure \[v5666\_sgr\_v1369\_cen\_v496\_sct\_v\_bv\_ub\_color\_logscale\] compares V5666 Sgr with the similar $V$ light curve novae, V496 Sct and V1369 Cen, in addition to LV Vul. We stretch the timescales of LV Vul, V496 Sct, and V1369 Cen by 1.78, 0.89, and 1.20, respectively, against V5666 Sgr, to overlap them with the V5666 Sgr $V$ light and $(B-V)_0$ color curves. From the time-stretching method, we have the relation of $$\begin{aligned}
(m&-&M)_{V, \rm V5666~Sgr} \cr
&=& (m-M + \Delta V)_{V, \rm LV~Vul} - 2.5 \log 1.78 \cr
&=& 11.85 + 4.2\pm0.2 - 0.63 = 15.42\pm0.2 \cr
&=& (m-M + \Delta V)_{V, \rm V496~Sct} - 2.5 \log 0.89 \cr
&=& 13.7 + 1.6\pm0.2 + 0.13 = 15.43\pm0.2 \cr
&=& (m-M + \Delta V)_{V, \rm V1369~Cen} - 2.5 \log 1.20 \cr
&=& 10.25 + 5.3\pm0.2 - 0.20 = 15.35\pm0.2.
\label{distance_modulus_v5666_sgr_v}\end{aligned}$$ Thus, we obtained $(m-M)_{V, \rm V5666~Sgr}=15.4\pm0.1$.
From Equations (\[time-stretching\_general\]), (\[distance\_modulus\_general\_temp\]), and (\[distance\_modulus\_v5666\_sgr\_v\]), we have the relation of $$\begin{aligned}
(m&-& M')_{V, \rm V5666~Sgr} \cr
&\equiv& (m_V - (M_V - 2.5\log f_{\rm s}))_{\rm V5666~Sgr} \cr
&=& \left( (m-M)_V + \Delta V \right)_{\rm LV~Vul} \cr
&=& 11.85 + 4.2\pm0.2 = 16.05\pm0.2.
\label{absolute_mag_v5666_sgr_lv_vul_stretch}\end{aligned}$$
We further compare the V5666 Sgr light curves with other novae having similar light curves, V1369 Cen and V496 Sct. Figure \[v5666\_sgr\_v496\_sct\_v1369\_cen\_b\_v\_i\_k\_logscale\_4fig\] shows the same time-stretched light curves of $B$, $V$, $I_{\rm C}$, and $K_{\rm s}$ bands as Figure \[v1369\_cen\_v496\_sct\_v5666\_sgr\_b\_v\_i\_k\_logscale\_4fig\]. Applying Equation (\[distance\_modulus\_general\_temp\_b\]) for the $B$ band to Figure \[v5666\_sgr\_v496\_sct\_v1369\_cen\_b\_v\_i\_k\_logscale\_4fig\](a), we have the relation of $$\begin{aligned}
(m&-&M)_{B, \rm V5666~Sgr} \cr
&=& \left( (m-M)_B + \Delta B\right)_{\rm V496~Sct} - 2.5 \log 0.89 \cr
&=& 14.15 + 1.6\pm0.2 + 0.125 = 15.88\pm0.2 \cr
&=& \left( (m-M)_B + \Delta B\right)_{\rm V1369~Cen} - 2.5 \log 1.2 \cr
&=& 10.36 + 5.7\pm0.2 - 0.20 = 15.86\pm0.2,
\label{distance_modulus_v5666_sgr_v496_sct_v1369_cen_b}\end{aligned}$$ where we adopt $(m-M)_{B, \rm V496~Sct}=13.7 + 1.0\times 0.45= 14.15$ from Section \[v496\_sct\_cmd\] and $(m-M)_{B, \rm V1369~Cen}=10.25 + 1.0\times 0.11=10.36$ from Section \[v1369\_cen\_cmd\]. Thus, we obtain $(m-M)_B=15.87\pm0.1$ for V5666 Sgr. We obtain $d=5.8\pm0.6$ kpc from Equation (\[distance\_modulus\_rb\]) together with $E(B-V)=0.50$ and $(m-M)_B=15.87\pm0.1$. We plot this relation of $(m-M)_B=15.87$ by the solid magenta line in Figure \[distance\_reddening\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\](d).
Applying Equation (\[distance\_modulus\_general\_temp\]) to Figure \[v5666\_sgr\_v496\_sct\_v1369\_cen\_b\_v\_i\_k\_logscale\_4fig\](b), we have the relation of $$\begin{aligned}
(m&-&M)_{V, \rm V5666~Sgr} \cr
&=& \left( (m-M)_V + \Delta V\right)_{\rm V496~Sct} - 2.5 \log 0.89 \cr
&=& 13.7 + 1.6\pm0.2 + 0.125 = 15.4\pm0.2 \cr
&=& \left( (m-M)_V + \Delta V\right)_{\rm V1369~Cen} - 2.5 \log 1.2 \cr
&=& 10.25 + 5.3\pm0.2 - 0.20 = 15.35\pm0.2,
\label{distance_modulus_v5666_sgr_v496_sct_v1369_cen_v}\end{aligned}$$ where we adopt $(m-M)_{V, \rm V496~Sct}=13.7$ from Section \[v496\_sct\_cmd\] and $(m-M)_{V, \rm V1369~Cen}=10.25$ from Section \[v1369\_cen\_cmd\]. Thus, we obtain $(m-M)_V=15.38\pm0.1$ for V5666 Sgr. We obtain $d=5.8\pm0.6$ kpc from Equation (\[distance\_modulus\_rv\]) together with $E(B-V)=0.50\pm0.05$ and $(m-M)_V=15.38\pm0.1$. We plot this relation of $(m-M)_V=15.38$ by the solid blue line in Figure \[distance\_reddening\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\](d).
From the $I_{\rm C}$ band data in Figure \[v5666\_sgr\_v496\_sct\_v1369\_cen\_b\_v\_i\_k\_logscale\_4fig\](c), we obtain $$\begin{aligned}
(m&-&M)_{I, \rm V5666~Sgr} \cr
&=& ((m - M)_I + \Delta I_C)_{\rm V496~Sct} - 2.5 \log 0.89 \cr
&=& 12.98 + 1.5\pm0.2 + 0.125 = 14.61\pm0.2 \cr
&=& ((m - M)_I + \Delta I_C)_{\rm V1369~Cen} - 2.5 \log 1.2 \cr
&=& 10.07 + 4.7\pm0.2 - 0.20 = 14.57\pm0.2,
\label{distance_modulus_i_v5666_sgr_v496_sct_v1369_cen}\end{aligned}$$ where we adopt $(m-M)_{I, \rm V496~Sct}=13.7 - 1.6\times 0.45= 12.98$ and $(m-M)_{I, \rm V1369~Cen}=10.25 - 1.6\times 0.11= 10.07$. Thus, we obtain $(m-M)_{I, \rm V5666~Sgr}= 14.59\pm0.1$. We obtain $d=5.9\pm0.6$ kpc from Equation (\[distance\_modulus\_ri\]) together with $E(B-V)=0.50\pm0.05$ and $(m-M)_I=14.59\pm0.1$. We plot this relation of $(m-M)_I=14.59$ by the solid cyan line in Figure \[distance\_reddening\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\](d).
We try to overlap the $K_{\rm s}$ light curves as much as possible in the later phase of Figure \[v5666\_sgr\_v496\_sct\_v1369\_cen\_b\_v\_i\_k\_logscale\_4fig\](d). Applying Equation (\[distance\_modulus\_general\_temp\_k\]) for the $K_{\rm s}$ band, we obtain $$\begin{aligned}
(m&-&M)_{K, \rm V5666~Sgr} \cr
&=& ((m - M)_K + \Delta K_{\rm s})_{\rm V496~Sct} - 2.5 \log 0.89 \cr
&=& 12.46 + 1.5\pm0.3 + 0.125 = 14.09\pm 0.3 \cr
&=& ((m - M)_K + \Delta K_{\rm s})_{\rm V1369~Cen} - 2.5 \log 1.2 \cr
&=& 9.95 + 4.3\pm0.3 - 0.20 = 14.05\pm 0.3,
\label{distance_modulus_k_v5666_sgr_v496_sct_v1369_cen}\end{aligned}$$ where we adopt $(m-M)_{K, \rm V496~Sct}=13.7 - 2.75\times 0.45= 12.46$, $(m-M)_{K, \rm V1369~Cen}=10.25 - 2.75\times 0.11= 9.95$. Thus, we obtain $(m-M)_{K, \rm V5666~Sgr}= 14.07\pm0.2$. We obtain $d=6.0\pm0.6$ kpc from Equation (\[distance\_modulus\_rk\]) together with $E(B-V)=0.50\pm0.05$ and $(m-M)_K=14.07\pm0.2$. We plot this relation of $(m-M)_K=14.07$ by the solid cyan-blue line in Figure \[distance\_reddening\_v574\_pup\_v679\_car\_v1369\_cen\_v5666\_sgr\](d).
Figure \[all\_mass\_v5666\_sgr\_v1668\_cyg\_x45z02c15o20\] also shows another comparison with LV Vul and V1668 Cyg. The light/color curves of LV Vul and V1668 Cyg are stretched by a factor of 1.78, so we obtain $$\begin{aligned}
(m&-&M)_{V, \rm V5666~Sgr} \cr
&=& (m-M + \Delta V)_{V, \rm LV~Vul} - 2.5 \log 1.78 \cr
&=& 11.85 + 4.2\pm0.2 - 0.63 = 15.42\pm0.2 \cr
&=& (m-M + \Delta V)_{V, \rm V1668~Cyg} - 2.5 \log 1.78 \cr
&=& 14.6 + 1.4\pm0.2 - 0.63 = 15.37\pm0.2,
\label{distance_modulus_v5666_sgr_v1668_cyg}\end{aligned}$$ from our time-stretching method. Thus, we again confirm $(m-M)_{V, \rm V5666~Sgr}=15.4\pm0.2$.
Reanalyzed Light Curves of 12 Novae {#reanalyse_twelve_novae}
===================================
V446 Her 1960 {#v446_her}
-------------
Figure \[v446\_her\_v1668\_cyg\_lv\_vul\_v\_color\_logscale\] shows the light/color curves of V446 Her on a logarithmic timescale as well as LV Vul and V1668 Cyg. We overlap the light/color curves of these three novae as much as possible. Based on the time-stretching method, we have the relation of $$\begin{aligned}
(m&-&M)_{V, \rm V446~Her} \cr
&=& (m - M + \Delta V)_{V, \rm LV Vul} - 2.5 \log 1.0 \cr
&=& 11.85 + 0.1\pm0.2 + 0.0 = 11.95\pm0.2 \cr
&=& (m - M + \Delta V)_{V, \rm V1668~Cyg} - 2.5 \log 1.0 \cr
&=& 14.6 - 2.6\pm0.2 + 0.0 = 12.0\pm0.2.
\label{distance_modulus_v466_her_lv_vul_v1668_cyg}\end{aligned}$$ Thus, we obtain $f_{\rm s}=1.0$ and $(m-M)_V=11.95\pm0.1$ for V446 Her. The new distance modulus is slightly larger than the previous value of $(m-M)_V=11.7\pm0.1$ (Paper II). This difference comes from the improved values of $f_{\rm s}$ and $\Delta V$. From Equations (\[distance\_modulus\_general\_dot\]) and (\[distance\_modulus\_v466\_her\_lv\_vul\_v1668\_cyg\]), we have the relation of $$\begin{aligned}
(m&-& M')_{V, \rm V446~Her} \cr
&\equiv& (m_V - (M_V - 2.5\log f_{\rm s}))_{\rm V446~Her} \cr
&=& (m-M+ \Delta V)_{V, \rm LV~Vul} \cr
&=& 11.85 + 0.1\pm0.2 = 11.95\pm0.2.
\label{absolute_mag_lv_vul_v446_her}\end{aligned}$$
V533 Her 1963 {#v533_her}
-------------
We obtain $f_{\rm s}$ and $(m-M)_V$ by the time-stretching method as in Figure \[v533\_her\_lv\_vul\_v1974\_cyg\_v\_bv\_ub\_logscale\], which shows the $V$ light curve as well as the dereddened $(B-V)_0$ and $(U-B)_0$ color curves. We add the light/color curves of LV Vul and V1974 Cyg and overlap these light/color curves as much as possible. Applying the time-stretching method of Equation (\[distance\_modulus\_general\_temp\]) to Figure \[v533\_her\_lv\_vul\_v1974\_cyg\_v\_bv\_ub\_logscale\](a), we have the relation of $$\begin{aligned}
(m&-&M)_{V, \rm V533~Her} \cr
&=& ((m - M)_V + \Delta V)_{\rm LV~Vul} - 2.5 \log 1.20 \cr
&=& 11.85 - 1.0\pm0.3 - 0.20 = 10.65\pm0.3 \cr
&=& ((m - M)_V + \Delta V)_{\rm V1974~Cyg} - 2.5 \log 1.15 \cr
&=& 12.2 - 1.4\pm0.3 - 0.15 = 10.65\pm0.3.
\label{distance_modulus_v533_her_v1974_cyg_v1668_cyg_v}\end{aligned}$$ The new result of $(m-M)_V=10.65\pm0.2$ is almost the same as the previous result of $(m-M)_V=10.8\pm0.2$ (Paper II), but slightly improve the timescaling factors of $f_{\rm s}$ and vertical fit of $\Delta V$. We obtain $f_{\rm s}=1.20$ against the template LV Vul. The distance is calculated to be $d= 1.28$ kpc from Equation (\[distance\_modulus\_rv\]) together with $(m-M)_V=10.65$ and $E(B-V)= 0.038$. This value is almost the same as what @hac16kb concluded from various results in the literature.
From Equations (\[distance\_modulus\_general\_dot\]) and (\[distance\_modulus\_v533\_her\_v1974\_cyg\_v1668\_cyg\_v\]), we have the relation of $$\begin{aligned}
(m&-& M')_{V, \rm V533~Her} \cr
&\equiv& (m_V - (M_V - 2.5\log f_{\rm s}))_{\rm V533~Her} \cr
&=& \left( (m-M)_V + \Delta V \right)_{\rm LV~Vul} \cr
&=& 11.85 - 1.0\pm0.3 = 10.85\pm0.3.
\label{absolute_mag_v533_her_lv_vul_v}\end{aligned}$$
PW Vul 1984\#1 {#pw_vul}
--------------
Figure \[pw\_vul\_lv\_vul\_v1668\_cyg\_v1500\_cyg\_v\_bv\_ub\_color\_logscale\] shows the light/color curves of PW Vul on a logarithmic timescale as well as LV Vul, V1500 Cyg, and V1668 Cyg. We deredden the colors of PW Vul with $E(B-V)= 0.57$ as obtained in Section \[pw\_vul\_cmd\]. In this figure, we regard that the $V$ light curve of PW Vul oscillates between the $V$ light curves of V1668 Cyg and V1500 Cyg during $\log t~({\rm day}) \sim1.4$–$2.1$. Based on the time-stretching method, we have the relation of $$\begin{aligned}
(m&-&M)_{V, \rm PW~Vul} \cr
&=& (m - M + \Delta V)_{V, \rm LV~Vul} - 2.5 \log 2.24 \cr
&=& 11.85 + 2.0\pm0.3 - 0.88 = 12.97\pm0.3 \cr
&=& (m - M + \Delta V)_{V, \rm V1668~Cyg} - 2.5 \log 2.24 \cr
&=& 14.6 - 0.7\pm0.3 - 0.88 = 13.02\pm0.3 \cr
&=& (m - M + \Delta V)_{V, \rm V1500~Cyg} - 2.5 \log 3.7 \cr
&=& 12.3 + 2.1\pm0.3 - 1.43 = 12.97\pm0.3.
\label{distance_modulus_pw_vul_t_pyx_lv_vul_v1668_cyg_v1500_cyg}\end{aligned}$$ Thus, we obtain $f_{\rm s}=2.24$ against the template nova LV Vul and confirm the previous result of $(m-M)_V=13.0\pm0.2$ for PW Vul [@hac15k; @hac16kb]. From Equations (\[distance\_modulus\_general\_dot\]) and (\[distance\_modulus\_pw\_vul\_t\_pyx\_lv\_vul\_v1668\_cyg\_v1500\_cyg\]), we have the relation of $$\begin{aligned}
(m&-& M')_{V, \rm PW~Vul} \cr
&\equiv& (m_V - (M_V - 2.5\log f_{\rm s}))_{\rm PW~Vul} \cr
&=& (m-M + \Delta V)_{V,\rm LV~Vul} \cr
&=& 11.85 + 2.0\pm0.3 =13.85\pm0.3.
\label{absolute_mag_pw_vul_lv_vul}\end{aligned}$$
We further obtain the distance moduli of $BVI_{\rm C}$ bands and examine the distance and reddening toward PW Vul. Figure \[pw\_vul\_v679\_car\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\] shows the three band light curves of PW Vul as well as V679 Car, YY Dor, and LMC N 2009a. For the $B$ band, we apply Equation (\[distance\_modulus\_general\_temp\_b\]) to Figure \[pw\_vul\_v679\_car\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](a) and obtain $$\begin{aligned}
(m&-&M)_{B, \rm PW~Vul} \cr
&=& ((m - M)_B + \Delta B)_{\rm YY~Dor} - 2.5 \log 11.7 \cr
&=& 18.98 - 2.8\pm0.2 - 2.67 = 13.51\pm0.2 \cr
&=& ((m - M)_B + \Delta B)_{\rm LMC~N~2009a} - 2.5 \log 7.4 \cr
&=& 18.98 - 3.3\pm0.2 - 2.17 = 13.51\pm0.2 \cr
&=& ((m - M)_B + \Delta B)_{\rm V679~Car} - 2.5 \log 2.24 \cr
&=& 16.79 - 2.4\pm0.2 - 0.87 = 13.52\pm0.2,
\label{distance_modulus_b_pw_vul_v679_car_yy_dor_lmcn2009a}\end{aligned}$$ where we adopt $(m-M)_{B, \rm V679~Car}=16.1 + 1.0\times 0.69= 16.79$. Thus, we obtain $(m-M)_{B, \rm PW~Vul}= 13.52\pm0.1$.
For the $V$ light curves in Figure \[pw\_vul\_v679\_car\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](b), we similarly obtain $$\begin{aligned}
(m&-&M)_{V, \rm PW~Vul} \cr
&=& ((m - M)_V + \Delta V)_{\rm YY~Dor} - 2.5 \log 11.7 \cr
&=& 18.86 - 3.2\pm0.2 - 2.67 = 12.99\pm0.2 \cr
&=& ((m - M)_V + \Delta V)_{\rm LMC~N~2009a} - 2.5 \log 7.4 \cr
&=& 18.86 - 3.7\pm0.2 - 2.17 = 12.99\pm0.2 \cr
&=& ((m - M)_V + \Delta V)_{\rm V679~Car} - 2.5 \log 2.24 \cr
&=& 16.1 - 2.2\pm0.2 - 0.87 = 13.03\pm0.2,
\label{distance_modulus_v_pw_vul_v679_car_yy_dor_lmcn2009a}\end{aligned}$$ where we adopt $(m-M)_{V, \rm V679~Car}=16.1$ in Section \[v679\_car\_cmd\]. Thus, we obtain $(m-M)_{V, \rm PW~Vul}= 13.0\pm0.1$.
Applying Equation (\[distance\_modulus\_general\_temp\_i\]) for the $I_{\rm C}$ band to Figure \[pw\_vul\_v679\_car\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](c), we obtain $$\begin{aligned}
(m&-&M)_{I, \rm PW~Vul} \cr
&=& ((m - M)_I + \Delta I_C)_{\rm YY~Dor} - 2.5 \log 11.7 \cr
&=& 18.67 - 3.9\pm0.3 - 2.67 = 12.1\pm0.3 \cr
&=& ((m - M)_I + \Delta I_C)_{\rm LMC~N~2009a} - 2.5 \log 7.4 \cr
&=& 18.67 - 4.4\pm0.3 - 2.17 = 12.1\pm0.3 \cr
&=& ((m - M)_I + \Delta I_C)_{\rm V679~Car} - 2.5 \log 2.24 \cr
&=& 15.0 - 2.0\pm0.3 - 0.87 = 12.13\pm0.3,
\label{distance_modulus_i_pw_vul_v679_car_yy_dor_lmcn2009a}\end{aligned}$$ where we adopt $(m-M)_{I, \rm V679~Car}=16.1 - 1.6\times 0.69= 15.0$. However, we should not use this result because no $I_{\rm C}$ but only $I$ data of PW Vul are available [@rob95] and the $I$ light curve of PW Vul (filled blue circles) does not accurately follow the other $I_{\rm C}$ data which follow the universal decline law as shown in Figure \[pw\_vul\_v679\_car\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](c).
We also plot the $U$, $B$, and $V$ light curves of PW Vul together with those of LV Vul, V1668 Cyg, and V5114 Sgr in Figure \[pw\_vul\_v5114\_sgr\_lv\_vul\_v1668\_cyg\_u\_b\_v\_logscale\_3fig\]. We apply Equation (\[distance\_modulus\_general\_temp\_u\]) for the $U$ band to Figure \[pw\_vul\_v5114\_sgr\_lv\_vul\_v1668\_cyg\_u\_b\_v\_logscale\_3fig\](a) and obtain $$\begin{aligned}
(m&-&M)_{U, \rm PW~Vul} \cr
&=& ((m - M)_U + \Delta U)_{\rm LV~Vul} - 2.5 \log 2.24 \cr
&=& 12.85 + 1.9\pm0.2 - 0.87 = 13.88\pm0.2 \cr
&=& ((m - M)_U + \Delta U)_{\rm V1668~Cyg} - 2.5 \log 2.24 \cr
&=& 15.1 - 0.3\pm0.2 - 0.87 = 13.93\pm0.2 \cr
&=& ((m - M)_U + \Delta U)_{\rm V5114~Sgr} - 2.5 \log 2.95 \cr
&=& 17.43 - 2.3\pm0.2 - 1.17 = 13.96\pm0.2,
\label{distance_modulus_u_pw_vul_v5114_sgr_lv_vul_v1668_cyg}\end{aligned}$$ where we adopt $(m-M)_{U, \rm LV~Vul}=11.85 + (4.75-3.1) \times 0.60 =12.85$, and $(m-M)_{U, \rm V1668~Cyg}=14.6 + (4.75-3.1) \times 0.30 =15.10$, and $(m-M)_{U, \rm V5114~Sgr}=16.65 + (4.75-3.1) \times 0.47 =17.43$. Thus, we obtain $(m-M)_{U, \rm PW~Vul}= 13.92\pm0.2$. For the $B$ light curves in Figure \[pw\_vul\_v5114\_sgr\_lv\_vul\_v1668\_cyg\_u\_b\_v\_logscale\_3fig\](b), we similarly obtain $$\begin{aligned}
(m&-&M)_{B, \rm PW~Vul} \cr
&=& ((m - M)_B + \Delta B)_{\rm LV~Vul} - 2.5 \log 2.24 \cr
&=& 12.45 + 2.0\pm0.2 - 0.87 = 13.58\pm0.2 \cr
&=& ((m - M)_B + \Delta B)_{\rm V1668~Cyg} - 2.5 \log 2.24 \cr
&=& 14.9 - 0.5\pm0.2 - 0.87 = 13.53\pm0.2 \cr
&=& ((m - M)_B + \Delta B)_{\rm V5114~Sgr} - 2.5 \log 2.95 \cr
&=& 17.12 - 2.4\pm0.2 - 1.17 = 13.55\pm0.2,
\label{distance_modulus_b_pw_vul_v5114_sgr_lv_vul_v1668_cyg}\end{aligned}$$ where we adopt $(m-M)_{B, \rm LV~Vul}= 11.85 + 1.0\times 0.6 =12.45$, $(m-M)_{B, \rm V1668~Cyg}= 14.6 + 1.0\times 0.3 =14.9$, and $(m-M)_{B, \rm V5114~Sgr}=16.65 + 1.0\times 0.47=17.12$. Thus, we obtain $(m-M)_{B, \rm PW~Vul}= 13.55\pm0.2$. We apply Equation (\[distance\_modulus\_general\_temp\]) to Figure \[pw\_vul\_v5114\_sgr\_lv\_vul\_v1668\_cyg\_u\_b\_v\_logscale\_3fig\](c) and obtain $$\begin{aligned}
(m&-&M)_{V, \rm PW~Vul} \cr
&=& ((m - M)_V + \Delta V)_{\rm LV~Vul} - 2.5 \log 2.24 \cr
&=& 11.85 + 2.0\pm0.2 - 0.87 = 12.98\pm0.2 \cr
&=& ((m - M)_V + \Delta V)_{\rm V1668~Cyg} - 2.5 \log 2.24 \cr
&=& 14.6 - 0.7\pm0.2 - 0.87 = 13.03\pm0.2 \cr
&=& ((m - M)_V + \Delta V)_{\rm V5114~Sgr} - 2.5 \log 2.95 \cr
&=& 16.65 - 2.5\pm0.2 - 1.17 = 12.98\pm0.2.
\label{distance_modulus_v_pw_vul_v5114_sgr_lv_vul_v1668_cyg}\end{aligned}$$ Thus, we obtain $(m-M)_{V, \rm PW~Vul}= 13.0\pm0.1$. This result is essentially the same as those in Equations (\[distance\_modulus\_pw\_vul\_t\_pyx\_lv\_vul\_v1668\_cyg\_v1500\_cyg\]) and (\[distance\_modulus\_v\_pw\_vul\_v679\_car\_yy\_dor\_lmcn2009a\]).
We plot these four distance moduli of $U$, $B$, $V$, and $I_{\rm C}$ bands in Figure \[distance\_reddening\_v446\_her\_v533\_her\_pw\_vul\_v1419\_aql\](c) by the thin solid green, cyan, and thick solid blue, and thin solid blue-magenta lines, that is, $(m-M)_U= 13.92$, $(m-M)_B= 13.55$, and $(m-M)_V= 13.0$ and $(m-M)_I= 12.12$, respectively. These four lines cross at $d=1.8$ kpc and $E(B-V)=0.57$.
V1419 Aql 1993 {#v1419_aql}
--------------
Figure \[v1419\_aql\_v1668\_cyg\_lv\_vul\_v\_bv\_ub\_color\_logscale\] shows the light/color curves of V1419 Aql as well as LV Vul and V1668 Cyg. We regard that the $V$ light curve of V1419 Aql follows V1668 Cyg and the upper branch of LV Vul during $\log t~({\rm day})\sim1.0$–$1.4$. Based on the time-stretching method, we have the relation of $$\begin{aligned}
(m&-&M)_{V, \rm V1419~Aql} \cr
&=& (m - M + \Delta V)_{V, \rm LV~Vul} - 2.5 \log 1.41 \cr
&=& 11.85 + 3.5\pm0.3 - 0.38 = 14.97\pm0.3 \cr
&=& (m - M + \Delta V)_{V, \rm V1668~Cyg} - 2.5 \log 1.41 \cr
&=& 14.6 + 0.8\pm0.3 - 0.38 = 15.02\pm0.3.
\label{distance_modulus_v1419_aql_lv_vul_v1668_cyg}\end{aligned}$$ Thus, we obtain $f_{\rm s}=1.41$ against the template nova LV Vul and $(m-M)_V=15.0\pm0.2$. The new value is slightly larger than the previous estimate of $(m-M)_V=14.6\pm0.1$ by @hac16kb, because the timescaling factor of $f_{\rm s}$ and $\Delta V$ are improved. From Equations (\[distance\_modulus\_general\_dot\]) and (\[distance\_modulus\_v1419\_aql\_lv\_vul\_v1668\_cyg\]), we have the relation of $$\begin{aligned}
(m&-& M')_{V, \rm V1419~Aql} \cr
&\equiv& (m_V - (M_V - 2.5\log f_{\rm s}))_{\rm V1419~Aql} \cr
&=& (m-M + \Delta V)_{V,\rm LV~Vul} \cr
&=& 11.85 + 3.5\pm0.3 =15.35\pm0.3.
\label{absolute_mag_v1419_aql_lv_vul}\end{aligned}$$
V705 Cas 1993 {#v705_cas}
-------------
Figure \[v705\_cas\_pw\_vul\_lv\_vul\_v\_bv\_ub\_color\_curve\_logscale\] shows the light/color curves of V705 Cas as well as LV Vul and PW Vul. We deredden the colors of V705 Cas with $E(B-V)=0.45$ as obtained in Section \[v705\_cas\_cmd\]. Based on the time-stretching method, we have the relation of $$\begin{aligned}
(m&-&M)_{V, \rm V705~Cas} \cr
&=& (m - M + \Delta V)_{V, \rm LV~Vul} - 2.5 \log 2.8 \cr
&=& 11.85 + 2.7\pm0.3 - 1.13 = 13.42\pm0.3 \cr
&=& (m - M + \Delta V)_{V, \rm PW~Vul} - 2.5 \log 1.26 \cr
&=& 13.0 + 0.7\pm0.3 - 0.25 = 13.45\pm0.3.
\label{distance_modulus_v705_cas_lv_vul_pw_vul}\end{aligned}$$ Thus, we obtain $f_{\rm s}=2.8$ against the template nova LV Vul and $(m-M)_V=13.45\pm0.2$. This value is consistent with the previous value of $(m-M)_V=13.4\pm0.1$ estimated by @hac15k. From Equations (\[distance\_modulus\_general\_dot\]) and (\[distance\_modulus\_v705\_cas\_lv\_vul\_pw\_vul\]), we have the relation of $$\begin{aligned}
(m&-& M')_{V, \rm V705~Cas} \cr
&\equiv& (m_V - (M_V - 2.5\log f_{\rm s}))_{\rm V705~Cas} \cr
&=& (m-M + \Delta V)_{V,\rm LV~Vul} \cr
&=& 11.85 + 2.7\pm0.3 =14.55\pm0.3.
\label{absolute_mag_v705_cas_lv_vul}\end{aligned}$$
V382 Vel 1999 {#v382_vel}
-------------
Figure \[v382\_vel\_lv\_vul\_v1668\_cyg\_v1974\_cyg\_v\_bv\_ub\_color\_logscale\] shows the light/color curves of V382 Vel as well as LV Vul, V1668 Cyg, and V1974 Cyg. The data of V382 Vel are the same as those in Figures 1 and 27 of Paper II. We add a straight solid black line labeled “$t^{-3}$” that indicates the homologously expanding ejecta, i.e., free expansion after the optically thick winds stop [see, e.g., @woo97; @hac06kb]. The $V$ light curves of these four novae roughly overlap each other except for the very early phase of LV Vul. Based on the time-stretching method, we have the relation of $$\begin{aligned}
(m&-&M)_{V, \rm V382~Vel} \cr
&=& (m - M + \Delta V)_{V, \rm LV~Vul} - 2.5 \log 0.51 \cr
&=& 11.85 - 1.1\pm0.3 + 0.73 = 11.48\pm0.3 \cr
&=& (m - M + \Delta V)_{V, \rm V1668~Cyg} - 2.5 \log 0.51 \cr
&=& 14.6 - 3.8\pm0.3 + 0.73 = 11.53\pm0.3 \cr
&=& (m - M + \Delta V)_{V, \rm V1974~Cyg} - 2.5 \log 0.54 \cr
&=& 12.2 - 1.4\pm0.3 + 0.68 = 11.48\pm0.3.
\label{distance_modulus_v382_vel_lv_vul}\end{aligned}$$ Thus, we obtain $f_{\rm s}=0.51$ against the template nova LV Vul and $(m-M)_V=11.5\pm0.2$. The result of $(m-M)_V=11.5\pm0.2$ is the same as $(m-M)_V=11.5\pm0.2$ in the previous work [@hac16k]. From Equations (\[distance\_modulus\_general\_dot\]) and (\[distance\_modulus\_v382\_vel\_lv\_vul\]), we have the relation of $$\begin{aligned}
(m&-& M')_{V, \rm V382~Vel} \cr
&\equiv& (m_V - (M_V - 2.5\log f_{\rm s}))_{\rm V382~Vel} \cr
&=& \left( (m-M)_V + \Delta V \right)_{\rm LV~Vul} \cr
&=& 11.85 - 1.1\pm0.3 = 10.75\pm0.3.
\label{absolute_mag_v382_vel_lv_vul}\end{aligned}$$
We obtain the reddening and distance from the time-stretching method. We plot the $B$, $V$, and $I_{\rm C}$ light curves of V382 Vel together with those of the LMC novae YY Dor and LMC N 2009a in Figure \[v382\_vel\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\]. We apply Equation (\[distance\_modulus\_general\_temp\_b\]) for the $B$ band to Figure \[v382\_vel\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](a) and obtain $$\begin{aligned}
(m&-&M)_{B, \rm V382~Vel} \cr
&=& ((m - M)_B + \Delta B)_{\rm YY~Dor} - 2.5 \log 2.7 \cr
&=& 18.98 - 6.2\pm0.3 - 1.08 = 11.7\pm0.3 \cr
&=& ((m - M)_B + \Delta B)_{\rm LMC~N~2009a} - 2.5 \log 1.7 \cr
&=& 18.98 - 6.7\pm0.3 - 0.58 = 11.7\pm0.3.
\label{distance_modulus_b_v382_vel_yy_dor_lmcn2009a}\end{aligned}$$ Thus, we obtain $(m-M)_{B, \rm V382~Vel}= 11.7\pm0.2$. For the $V$ light curves in Figure \[v382\_vel\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](b), we similarly obtain $$\begin{aligned}
(m&-&M)_{V, \rm V382~Vel} \cr
&=& ((m - M)_V + \Delta V)_{\rm YY~Dor} - 2.5 \log 2.7 \cr
&=& 18.86 - 6.3\pm0.3 - 1.08 = 11.48\pm0.3 \cr
&=& ((m - M)_V + \Delta V)_{\rm LMC~N~2009a} - 2.5 \log 1.7 \cr
&=& 18.86 - 6.8\pm0.3 -0.58 = 11.48\pm0.3.
\label{distance_modulus_v_v382_vel_yy_dor_lmcn2009a}\end{aligned}$$ Thus, we obtain $(m-M)_{V, \rm V382~Vel}= 11.48\pm0.2$. We apply Equation (\[distance\_modulus\_general\_temp\_i\]) for the $I_{\rm C}$ band to Figure \[v382\_vel\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](c) and obtain $$\begin{aligned}
(m&-&M)_{I, \rm V382~Vel} \cr
&=& ((m - M)_I + \Delta I_C)_{\rm YY~Dor} - 2.5 \log 2.7 \cr
&=& 18.67 - 6.5\pm0.3 - 1.08 = 11.09\pm 0.3 \cr
&=& ((m - M)_I + \Delta I_C)_{\rm LMC~N~2009a} - 2.5 \log 1.7 \cr
&=& 18.67 - 7.0\pm0.3 -0.58 = 11.09\pm 0.3.
\label{distance_modulus_i_v382_vel_yy_dor_lmcn2009a}\end{aligned}$$ Thus, we obtain $(m-M)_{I, \rm V382~Vel}= 11.09\pm0.2$.
V5114 Sgr 2004 {#v5114_sgr}
--------------
Figure \[v5114\_sgr\_lv\_vul\_v1668\_cyg\_v\_color\_logscale\] shows the light/color curves of V5114 Sgr as well as LV Vul and V1668 Cyg. We regard that the $V$ light curve of V5114 Sgr follows V1668 Cyg and LV Vul. Based on the time-stretching method, we have the relation of $$\begin{aligned}
(m&-&M)_{V, \rm V5114~Sgr} \cr
&=& (m - M + \Delta V)_{V, \rm LV~Vul} - 2.5 \log 0.76 \cr
&=& 11.85 + 4.5\pm0.2 + 0.30 = 16.65\pm0.2 \cr
&=& (m - M + \Delta V)_{V, \rm V1668~Cyg} - 2.5 \log 0.76 \cr
&=& 14.6 + 1.75\pm0.2 + 0.30 = 16.65\pm0.2.
\label{distance_modulus_v5114_sgr_lv_vul_iv_cep_v1668_cyg}\end{aligned}$$ Thus, we obtain $f_{\rm s}=0.76$ against the template nova LV Vul and $(m-M)_V=16.65\pm0.1$. This value is slightly larger than the previous value of $(m-M)_V=16.5\pm0.1$ estimated by @hac16kb. From Equations (\[distance\_modulus\_general\_dot\]) and (\[distance\_modulus\_v5114\_sgr\_lv\_vul\_iv\_cep\_v1668\_cyg\]), we have the relation of $$\begin{aligned}
(m&-& M')_{V, \rm V5114~Sgr} \cr
&\equiv& (m_V - (M_V - 2.5\log f_{\rm s}))_{\rm V5114~Sgr} \cr
&=& (m-M + \Delta V)_{V,\rm LV~Vul} \cr
&=& 11.85 + 4.5\pm0.2 =16.35\pm0.2.
\label{absolute_mag_v5114_sgr_lv_vul}\end{aligned}$$
We obtain the reddening and distance from the time-stretching method. We plot the $U$, $B$, and $V$ light curves of V5114 Sgr together with those of LV Vul and V1668 Cyg in Figure \[v5114\_sgr\_lv\_vul\_v1668\_cyg\_u\_b\_v\_logscale\_3fig\]. We apply Equation (\[distance\_modulus\_general\_temp\_u\]) for the $U$ band to Figure \[v5114\_sgr\_lv\_vul\_v1668\_cyg\_u\_b\_v\_logscale\_3fig\](a) and obtain $$\begin{aligned}
(m&-&M)_{U, \rm V5114~Sgr} \cr
&=& ((m - M)_U + \Delta U)_{\rm LV~Vul} - 2.5 \log 0.76 \cr
&=& 12.85 + 4.3\pm0.2 + 0.30 = 17.45\pm0.2 \cr
&=& ((m - M)_U + \Delta U)_{\rm V1668~Cyg} - 2.5 \log 0.76 \cr
&=& 15.1 + 2.05\pm0.2 + 0.30 = 17.45\pm0.2,
\label{distance_modulus_u_v5114_sgr_lv_vul_v1668_cyg}\end{aligned}$$ where we adopt $(m-M)_{U, \rm LV~Vul}=11.85 + (4.75-3.1) \times 0.60
=12.85$, and $(m-M)_{U, \rm V1668~Cyg}=14.6 + (4.75-3.1) \times 0.30
=15.10$. Thus, we obtain $(m-M)_{U, \rm V5114~Sgr}= 17.45\pm0.2$. For the $B$ light curves in Figure \[v5114\_sgr\_lv\_vul\_v1668\_cyg\_u\_b\_v\_logscale\_3fig\](b), we similarly obtain $$\begin{aligned}
(m&-&M)_{B, \rm V5114~Sgr} \cr
&=& ((m - M)_B + \Delta B)_{\rm LV~Vul} - 2.5 \log 0.76 \cr
&=& 12.45 + 4.4\pm0.2 + 0.30 = 17.15\pm0.2 \cr
&=& ((m - M)_B + \Delta B)_{\rm LMC~N~2009a} - 2.5 \log 0.76 \cr
&=& 14.9 + 1.95\pm0.2 + 0.30 = 17.15\pm0.2,
\label{distance_modulus_b_v5114_sgr_yy_dor_lmcn2009a}\end{aligned}$$ where we adopt $(m-M)_{B, \rm LV~Vul}= 11.85 + 1.0\times 0.6 =12.45$ and $(m-M)_{B, \rm V1668~Cyg}= 14.6 + 1.0\times 0.3 =14.9$. Thus, we obtain $(m-M)_{B, \rm V5114~Sgr}= 17.15\pm0.2$. We apply Equation (\[distance\_modulus\_general\_temp\]) to Figure \[v5114\_sgr\_lv\_vul\_v1668\_cyg\_u\_b\_v\_logscale\_3fig\](c) and obtain $$\begin{aligned}
(m&-&M)_{V, \rm V5114~Sgr} \cr
&=& ((m - M)_V + \Delta V)_{\rm LV~Vul} - 2.5 \log 0.76 \cr
&=& 11.85 + 4.5\pm0.2 + 0.30 = 16.65\pm0.2 \cr
&=& ((m - M)_V + \Delta V)_{\rm V1668~Cyg} - 2.5 \log 0.76 \cr
&=& 14.6 + 1.75\pm0.2 + 0.30 = 16.65\pm0.2.
\label{distance_modulus_v_v5114_sgr_yy_dor_lmcn2009a}\end{aligned}$$ Thus, we obtain $(m-M)_{V, \rm V5114~Sgr}= 16.65\pm0.2$. This result is essentially the same as that in Equation (\[distance\_modulus\_v5114\_sgr\_lv\_vul\_iv\_cep\_v1668\_cyg\]). We plot these three distance moduli of $U$, $B$, and $V$ bands in Figure \[distance\_reddening\_v705\_cas\_v382\_vel\_v5114\_sgr\_v2362\_cyg\](c) by the magenta, cyan, and blue lines, that is, $(m-M)_U= 17.45$, $(m-M)_B= 17.15$, and $(m-M)_V= 16.65$ together with Equations (\[distance\_modulus\_ru\]), (\[distance\_modulus\_rb\]), and (\[distance\_modulus\_rv\]), respectively. These three lines cross at $d=10.9$ kpc and $E(B-V)=0.47$.
V2362 Cyg 2006 {#v2362_cyg}
--------------
Figure \[v2362\_cyg\_lv\_vul\_v1500\_cyg\_v\_bv\_ub\_color\_logscale\] shows the light/color curves of V2362 Cyg on a logarithmic timescale as well as LV Vul and V1500 Cyg. We regard that the $V$ light curve of V2362 Cyg follows that of V1500 Cyg. Based on the time-stretching method, we have the relation of $$\begin{aligned}
(m&-&M)_{V, \rm V2362~Cyg} \cr
&=& (m - M + \Delta V)_{V, \rm LV~Vul} - 2.5 \log 1.78 \cr
&=& 11.85 + 4.2\pm0.3 - 0.63 = 15.42\pm0.3 \cr
&=& (m - M + \Delta V)_{V, \rm V1500~Cyg} - 2.5 \log 2.95 \cr
&=& 12.3 + 4.3\pm0.2 - 1.18 = 15.42\pm0.2.
\label{distance_modulus_v2362_cyg_lv_vul_v1500_cyg_v1668_cyg_v}\end{aligned}$$ Thus, we obtain $f_{\rm s}=1.78$ against the template nova LV Vul and $(m-M)_V=15.4\pm0.2$.
This new distance modulus in the $V$ band is smaller than the previous value of $(m-M)_V=15.9\pm0.1$ in Paper II, because we improved both the timescaling factor of $f_{\rm s}$ and the vertical fit of $\Delta V$. The LV Vul upper branch in the time-stretched color-magnitude diagram corresponds to the redder (lower) branch after the onset of the nebular phase in the $(B-V)_0$ color curves of Figure \[v2362\_cyg\_lv\_vul\_v1500\_cyg\_v\_bv\_ub\_color\_logscale\](b). We regard that the V2362 Cyg color curves overlap with this redder branch of LV Vul in the $(B-V)_0$ color curve. Thus, we have more carefully determined the timescaling factor of $f_{\rm s}$ than the previous work (Paper II).
From Equations (\[distance\_modulus\_general\_dot\]) and (\[distance\_modulus\_v2362\_cyg\_lv\_vul\_v1500\_cyg\_v1668\_cyg\_v\]), we obtain the relation of $$\begin{aligned}
(m&-& M')_{V, \rm V2362~Cyg} \cr
&\equiv& (m_V - (M_V - 2.5\log f_{\rm s}))_{\rm V2362~Cyg} \cr
&=& (m-M+ \Delta V)_{V, \rm LV~Vul} \cr
&=& 11.85 +4.2\pm0.3 = 16.05\pm0.3.
\label{absolute_mag_lv_vul_v2362_cyg_v}\end{aligned}$$
V2615 Oph 2007 {#v2615_oph}
--------------
Figure \[v2615\_oph\_lv\_vul\_v1419\_aql\_v\_bv\_ub\_color\_logscale\] shows the light/color curves of V2615 Oph as well as LV Vul and V1419 Aql. The colors of V2615 Oph are dereddened with $E(B-V)=0.90$ as obtained in Section \[v2615\_oph\_cmd\]. The $BV$ light curve data of this nova were obtained by @mun08a. However, their $B-V$ data show systematic differences among three observatories, a, b, and c (see their Figure 1). So, we shift the $B-V$ data of group “a” (filled red circles in Figure \[hr\_diagram\_v2615\_oph\_v2468\_cyg\_v2491\_cyg\_v496\_sct\_outburst\](a)) by 0.15 mag bluer. The $B-V$ data of group “b” and “c” (filled blue circles in Figure \[hr\_diagram\_v2615\_oph\_v2468\_cyg\_v2491\_cyg\_v496\_sct\_outburst\](a)) are not shifted. As a result, the $B-V$ data of these three groups broadly overlap each other in Figures \[v2615\_oph\_lv\_vul\_v1419\_aql\_v\_bv\_ub\_color\_logscale\](b) and \[hr\_diagram\_v2615\_oph\_v2468\_cyg\_v2491\_cyg\_v496\_sct\_outburst\](a). We also add the $V$ data taken from SMARTS [@wal12]. Based on the time-stretching method, we have the relation of $$\begin{aligned}
(m&-&M)_{V, \rm V2615~Oph} \cr
&=& (m - M + \Delta V)_{V, \rm LV~Vul} - 2.5 \log 1.58 \cr
&=& 11.85 + 4.6\pm0.2 - 0.50 = 15.95\pm0.2 \cr
&=& (m - M + \Delta V)_{V, \rm V1419~Aql} - 2.5 \log 1.12 \cr
&=& 15.0 + 1.1\pm0.2 - 0.13 = 15.97\pm0.2.
\label{distance_modulus_v2615_oph_v1419_aql_lv_vul}\end{aligned}$$ Thus, we obtain $f_{\rm s}=1.58$ against the template nova LV Vul and $(m-M)_V=15.95\pm0.2$. This value is smaller than the previous estimate of $(m-M)_V=16.5\pm0.1$ (Paper II). This is due not only to our careful $(B-V)_0$ fitting that improves the timescaling factor of $f_{\rm s}$ but also to the revised vertical $V$ fit to the LV Vul light curve. From Equations (\[distance\_modulus\_general\_dot\]) and (\[distance\_modulus\_v2615\_oph\_v1419\_aql\_lv\_vul\]), we have the relation of $$\begin{aligned}
(m&-& M')_{V, \rm V2615~Oph} \cr
&\equiv& (m_V - (M_V + 2.5\log f_s))_{\rm V2615~Oph} \cr
&=& (m-M + \Delta V)_{V,\rm LV~Vul} \cr
&=& 11.85 + 4.6\pm0.2 =16.45\pm0.2.
\label{absolute_mag_v2615_oph_lv_vul}\end{aligned}$$
We obtain the reddening and distance from the time-stretching method. We plot the $B$, $V$, and $I_{\rm C}$ light curves of V2615 Oph together with those of V1368 Cen, V834 Car, and the LMC novae, YY Dor and LMC N 2009a, in Figure \[v2615\_oph\_v1368\_cen\_v834\_car\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\] on a logarithmic timescale. We apply Equation (\[distance\_modulus\_general\_temp\_b\]) for the $B$ band to Figure \[v2615\_oph\_v1368\_cen\_v834\_car\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](a) and obtain $$\begin{aligned}
(m&-&M)_{B, \rm V2615~Oph} \cr
&=& ((m - M)_B + \Delta B)_{\rm V1368~Cen} - 2.5 \log 0.66 \cr
&=& 18.53 - 2.1\pm0.4 + 0.45 = 16.88\pm0.3 \cr
&=& ((m - M)_B + \Delta B)_{\rm V834~Car} - 2.5 \log 2.5 \cr
&=& 17.75 + 0.1\pm0.4 - 0.97 = 16.88\pm0.3 \cr
&=& ((m - M)_B + \Delta B)_{\rm YY~Dor} - 2.5 \log 8.3 \cr
&=& 18.98 + 0.2\pm0.4 - 2.3 = 16.88\pm0.3 \cr
&=& ((m - M)_B + \Delta B)_{\rm LMC~N~2009a} - 2.5 \log 5.2 \cr
&=& 18.98 - 0.3\pm0.4 - 1.8 = 16.88\pm0.3,
\label{distance_modulus_b_v2615_oph_v1535_sco_v834_car_yy_dor_lmcn2009a}\end{aligned}$$ where we adopt $(m-M)_{B, \rm V1368~Cen}=17.6 + 0.93 = 18.53$ and $(m-M)_{B, \rm V834~Car}= 17.25 + 0.50 = 17.75$ from @hac19k. Thus, we obtain $(m-M)_{B, \rm V2615~Oph}= 16.88\pm0.2$. For the $V$ light curves in Figure \[v2615\_oph\_v1368\_cen\_v834\_car\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](b), we similarly obtain $$\begin{aligned}
(m&-&M)_{V, \rm V2615~Oph} \cr
&=& ((m - M)_V + \Delta V)_{\rm V1368~Cen} - 2.5 \log 0.66 \cr
&=& 17.6 - 2.1\pm0.3 + 0.45 = 15.95\pm0.3 \cr
&=& ((m - M)_V + \Delta V)_{\rm V834~Car} - 2.5 \log 2.5 \cr
&=& 17.25 - 0.3\pm0.3 - 0.97 = 15.98\pm0.3 \cr
&=& ((m - M)_V + \Delta V)_{\rm YY~Dor} - 2.5 \log 8.3 \cr
&=& 18.86 - 0.6\pm0.3 - 2.3 = 15.96\pm0.3 \cr
&=& ((m - M)_V + \Delta V)_{\rm LMC~N~2009a} - 2.5 \log 5.2 \cr
&=& 18.86 - 1.1\pm0.3 - 1.8 = 15.96\pm0.3,
\label{distance_modulus_v_v2615_oph_v1535_sco_v834_car_yy_dor_lmcn2009a}\end{aligned}$$ where we adopt $(m-M)_{V, \rm V1368~Cen}= 17.6$ and $(m-M)_{V, \rm V834~Car}= 17.25$ from @hac19k. Thus, we obtain $(m-M)_{V, \rm V2615~Oph}= 15.96\pm0.2$, being consistent with Equation (\[distance\_modulus\_v2615\_oph\_v1419\_aql\_lv\_vul\]). We apply Equation (\[distance\_modulus\_general\_temp\_i\]) for the $I_{\rm C}$ band to Figure \[v2615\_oph\_v1368\_cen\_v834\_car\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](c) and obtain $$\begin{aligned}
(m&-&M)_{I, \rm V2615~Oph} \cr
&=& ((m - M)_I + \Delta I_C)_{\rm V1368~Cen} - 2.5 \log 0.66 \cr
&=& 16.11 - 2.0\pm0.5 + 0.45 = 14.56\pm 0.3 \cr
&=& ((m - M)_I + \Delta I_C)_{\rm V834~Car} - 2.5 \log 2.5 \cr
&=& 16.45 - 0.9\pm0.5 - 0.97 = 14.58\pm 0.3 \cr
&=& ((m - M)_I + \Delta I_C)_{\rm YY~Dor} - 2.5 \log 8.3 \cr
&=& 18.67 - 1.8\pm0.5 - 2.3 = 14.57\pm 0.3 \cr
&=& ((m - M)_I + \Delta I_C)_{\rm LMC~N~2009a} - 2.5 \log 5.2 \cr
&=& 18.67 - 2.3\pm0.5 - 1.8 = 14.57\pm 0.3,
\label{distance_modulus_i_v2615_oph_v1535_sco_v834_car_yy_dor_lmcn2009a}\end{aligned}$$ where we adopt $(m-M)_{I, \rm V1368~Cen}=17.6 - 1.6\times 0.93 = 16.11$, and $(m-M)_{I, \rm V834~Car}=17.25 - 1.6\times 0.50 = 16.45$ from @hac19k. Thus, we obtain $(m-M)_{I, \rm V2615~Oph}= 14.57\pm0.2$. We plot these three distance moduli in Figure \[distance\_reddening\_v2615\_oph\_v2468\_cyg\_v2491\_cyg\_v496\_sct\](a) by the thin solid magenta, blue, and cyan lines. These three lines cross at $d=4.3$ kpc and $E(B-V)=0.90$.
V2468 Cyg 2008\#1 {#v2468_cyg}
-----------------
Figure \[v2468\_cyg\_lv\_vul\_v1500\_cyg\_v\_color\_logscale\_no2\] shows the light/color curves of V2468 Cyg as well as LV Vul and V1500 Cyg. The $(B-V)_0$ color is dereddened with $E(B-V)=0.65$ as obtained in Section \[v2468\_cyg\_cmd\]. We regard that the early phase of the V2468 Cyg $V$ light curve is superbright like V1500 Cyg and then oscillates between LV Vul and V1500 Cyg in the phase of $\log t~({\rm day})=1.0$–$2.0$. The spectra of V1500 Cyg during the superbright phase (until day $\sim 5$ and above the model $V$ light curve of red line) are approximated with blackbody [@gal76; @enn77]. After the superbright phase ended, the spectrum of V1500 Cyg changed to be of free-free emission, so that $V$ magnitudes are fitted with our model light curve (solid red line). Our time-stretching method is applicable to the part of free-free emission dominated phase of optical light curves.
Based on the time-stretching method, we have the relation of $$\begin{aligned}
(m&-&M)_{V, \rm V2468~Cyg} \cr
&=& (m - M + \Delta V)_{V, \rm LV~Vul} - 2.5 \log 2.4 \cr
&=& 11.85 + 5.3\pm0.3 - 0.95 = 16.2\pm0.3 \cr
&=& (m - M + \Delta V)_{V, \rm V1500~Cyg} - 2.5 \log 4.0 \cr
&=& 12.3 + 5.4\pm0.3 - 1.5 = 16.2\pm0.3.
\label{distance_modulus_v2468_cyg_v1668_cyg_lv_vul_iv_cep}\end{aligned}$$ Thus, we obtain $f_{\rm s}=2.4$ against the template nova LV Vul and $(m-M)_V=16.2\pm0.2$. The newly obtained value is larger than the previous value of $(m-M)_V=15.6\pm0.1$ estimated in Paper II because we improved the timescaling factor $f_{\rm s}$ and the vertical $\Delta V$ fit. From Equations (\[distance\_modulus\_general\_dot\]) and (\[distance\_modulus\_v2468\_cyg\_v1668\_cyg\_lv\_vul\_iv\_cep\]), we have the relation of $$\begin{aligned}
(m&-& M')_{V, \rm V2468~Cyg} \cr
&\equiv& (m_V - (M_V - 2.5\log f_s))_{\rm V2468~Cyg} \cr
&=& (m-M + \Delta V)_{V,\rm LV~Vul} \cr
&=& 11.85 + 5.3\pm0.3 = 17.15\pm0.3.
\label{absolute_mag_v2468_cyg_lv_vul}\end{aligned}$$
We obtain the reddening and distance from the time-stretching method. We plot the $B$, $V$, and $I_{\rm C}$ light curves of V2468 Cyg together with those of V1535 Sco, V834 Car, and the LMC novae, YY Dor and LMC N 2009a, in Figure \[v2468\_cyg\_v1535\_sco\_v834\_car\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\] on a logarithmic timescale. We apply Equation (\[distance\_modulus\_general\_temp\_b\]) for the $B$ band to Figure \[v2468\_cyg\_v1535\_sco\_v834\_car\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](a) and obtain $$\begin{aligned}
(m&-&M)_{B, \rm V2468~Cyg} \cr
&=& ((m - M)_B + \Delta B)_{\rm V1535~Sco} - 2.5 \log 1.0 \cr
&=& 19.08 - 2.2\pm0.4 - 0.0 = 16.88\pm0.3 \cr
&=& ((m - M)_B + \Delta B)_{\rm V834~Car} - 2.5 \log 3.7 \cr
&=& 17.75 + 0.5\pm0.4 - 1.42 = 16.83\pm0.3 \cr
&=& ((m - M)_B + \Delta B)_{\rm YY~Dor} - 2.5 \log 12.6 \cr
&=& 18.98 + 0.6\pm0.4 - 2.75 = 16.83\pm0.3 \cr
&=& ((m - M)_B + \Delta B)_{\rm LMC~N~2009a} - 2.5 \log 7.9 \cr
&=& 18.98 + 0.1\pm0.4 - 2.25 = 16.83\pm0.3,
\label{distance_modulus_b_v2468_cyg_v1535_sco_v834_car_yy_dor_lmcn2009a}\end{aligned}$$ where we adopt $(m-M)_{B, \rm V1535~Sco}=18.3 + 0.78 = 19.08$ and $(m-M)_{B, \rm V834~Car}= 17.25 + 0.50 = 17.75$ from @hac19k. Thus, we obtain $(m-M)_{B, \rm V2468~Cyg}= 16.85\pm0.2$. For the $V$ light curves in Figure \[v2468\_cyg\_v1535\_sco\_v834\_car\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](b), we similarly obtain $$\begin{aligned}
(m&-&M)_{V, \rm V2468~Cyg} \cr
&=& ((m - M)_V + \Delta V)_{\rm V1535~Sco} - 2.5 \log 1.0 \cr
&=& 18.3 - 2.1\pm0.3 - 0.0 = 16.2\pm0.3 \cr
&=& ((m - M)_V + \Delta V)_{\rm V834~Car} - 2.5 \log 3.7 \cr
&=& 17.25 + 0.4\pm0.3 - 1.42 = 16.23\pm0.3 \cr
&=& ((m - M)_V + \Delta V)_{\rm YY~Dor} - 2.5 \log 12.6 \cr
&=& 18.86 + 0.1\pm0.3 - 2.75 = 16.21\pm0.3 \cr
&=& ((m - M)_V + \Delta V)_{\rm LMC~N~2009a} - 2.5 \log 7.9 \cr
&=& 18.86 - 0.4\pm0.3 - 2.25 = 16.21\pm0.3,
\label{distance_modulus_v_v2468_cyg_v1535_sco_v834_car_yy_dor_lmcn2009a}\end{aligned}$$ where we adopt $(m-M)_{V, \rm V1535~Sco}= 18.3$ and $(m-M)_{V, \rm V834~Car}= 17.25$ from @hac19k. Thus, we obtain $(m-M)_{V, \rm V2468~Cyg}= 16.21\pm0.2$, being consistent with Equation (\[distance\_modulus\_v2468\_cyg\_v1668\_cyg\_lv\_vul\_iv\_cep\]). We apply Equation (\[distance\_modulus\_general\_temp\_i\]) for the $I_{\rm C}$ band to Figure \[v2468\_cyg\_v1535\_sco\_v834\_car\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](c) and obtain $$\begin{aligned}
(m&-&M)_{I, \rm V2468~Cyg} \cr
&=& ((m - M)_I + \Delta I_C)_{\rm V1535~Sco} - 2.5 \log 1.0 \cr
&=& 17.05 - 1.9\pm0.5 - 0.0 = 15.15\pm 0.3 \cr
&=& ((m - M)_I + \Delta I_C)_{\rm V834~Car} - 2.5 \log 3.7 \cr
&=& 16.45 + 0.1\pm0.5 - 1.42 = 15.13\pm 0.3 \cr
&=& ((m - M)_I + \Delta I_C)_{\rm YY~Dor} - 2.5 \log 12.6 \cr
&=& 18.67 - 0.8\pm0.5 - 2.75 = 15.12\pm 0.3 \cr
&=& ((m - M)_I + \Delta I_C)_{\rm LMC~N~2009a} - 2.5 \log 7.9 \cr
&=& 18.67 - 1.3\pm0.5 - 2.25 = 15.12\pm 0.3,
\label{distance_modulus_i_v2468_cyg_v1535_sco_v834_car_yy_dor_lmcn2009a}\end{aligned}$$ where we adopt $(m-M)_{I, \rm V1535~Sco}=18.3 - 1.6\times 0.78 = 17.05$, and $(m-M)_{I, \rm V834~Car}=17.25 - 1.6\times 0.50 = 16.45$ from @hac19k. Thus, we obtain $(m-M)_{I, \rm V2468~Cyg}= 15.13\pm0.2$.
V2491 Cyg 2008\#2 {#v2491_cyg}
-----------------
Figure \[v2491\_cyg\_lv\_vul\_v1500cyg\_cyg\_v\_color\_logscale\] shows the light/color curves of V2491 Cyg as well as LV Vul and V1500 Cyg. Based on the time-stretching method, we have the relation of $$\begin{aligned}
(m&-&M)_{V, \rm V2491~Cyg} \cr
&=& (m - M + \Delta V)_{V, \rm LV~Vul} - 2.5 \log 0.46 \cr
&=& 11.85 + 4.7\pm0.4 + 0.85 = 17.4\pm0.4 \cr
&=& (m - M + \Delta V)_{V, \rm V1500~Cyg} - 2.5 \log 0.76 \cr
&=& 12.3 + 4.8\pm0.4 + 0.30 = 17.4\pm0.4.
\label{distance_modulus_v2491_cyg_lv_vul_iv_cep_v1668_cyg}\end{aligned}$$ Thus, we obtain $f_{\rm s}=0.46$ against the template nova LV Vul and $(m-M)_V=17.4\pm0.3$. The new distance modulus in the $V$ band is larger than the previous value of $(m-M)_V=16.5\pm0.1$ (Paper II). This is because the reddening is revised to be $E(B-V)=0.45$ and the timescaling factor is improved to be $f_{\rm s}=0.46$. We carefully redetermine the timescaling factor by fitting the $(B-V)_0$ color curve of V2491 Cyg (filled red circles) with that of LV Vul (filled cyan stars) and V1500 Cyg (open blue circles) as shown in Figure \[v2491\_cyg\_lv\_vul\_v1500cyg\_cyg\_v\_color\_logscale\](b). From Equations (\[distance\_modulus\_general\_dot\]) and (\[distance\_modulus\_v2491\_cyg\_lv\_vul\_iv\_cep\_v1668\_cyg\]), we obtain $$\begin{aligned}
(m&-& M')_{V, \rm V2491~Cyg} \cr
&\equiv& (m_V - (M_V - 2.5\log f_{\rm s}))_{\rm V2491~Cyg} \cr
&=& (m-M+ \Delta V)_{V, \rm LV~Vul} \cr
&=& 11.85 +4.7\pm0.4 = 16.55\pm0.4.
\label{absolute_mag_lv_vul_v2491_cyg}\end{aligned}$$
We obtain the reddening and distance from the time-stretching method. We plot the $B$, $V$, and $I_{\rm C}$ light curves of V2491 Cyg together with those of the LMC novae YY Dor and LMC N 2009a in Figure \[v2491\_cyg\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\] on a logarithmic timescale. We apply Equation (\[distance\_modulus\_general\_temp\_b\]) for the $B$ band to Figure \[v2491\_cyg\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](a) and obtain $$\begin{aligned}
(m&-&M)_{B, \rm V2491~Cyg} \cr
&=& ((m - M)_B + \Delta B)_{\rm YY~Dor} - 2.5 \log 2.4 \cr
&=& 18.98 - 0.2\pm0.4 - 0.95 = 17.83\pm0.4 \cr
&=& ((m - M)_B + \Delta B)_{\rm LMC~N~2009a} - 2.5 \log 1.51 \cr
&=& 18.98 - 0.7\pm0.4 - 0.45 = 17.83\pm0.4.
\label{distance_modulus_b_v2491_cyg_yy_dor_lmcn2009a}\end{aligned}$$ Thus, we obtain $(m-M)_{B, \rm V2491~Cyg}= 17.83\pm0.3$. For the $V$ light curves in Figure \[v2491\_cyg\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](b), we similarly obtain $$\begin{aligned}
(m&-&M)_{V, \rm V2491~Cyg} \cr
&=& ((m - M)_V + \Delta V)_{\rm YY~Dor} - 2.5 \log 2.4 \cr
&=& 18.86 - 0.5\pm0.3 - 0.95 = 17.41\pm0.3 \cr
&=& ((m - M)_V + \Delta V)_{\rm LMC~N~2009a} - 2.5 \log 1.51 \cr
&=& 18.86 - 1.0\pm0.3 -0.45 = 17.41\pm0.3.
\label{distance_modulus_v_v2491_cyg_yy_dor_lmcn2009a}\end{aligned}$$ Thus, we obtain $(m-M)_{V, \rm V2491~Cyg}= 17.41\pm0.2$. We apply Equation (\[distance\_modulus\_general\_temp\_i\]) for the $I_{\rm C}$ band to Figure \[v2491\_cyg\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](c) and obtain $$\begin{aligned}
(m&-&M)_{I, \rm V2491~Cyg} \cr
&=& ((m - M)_I + \Delta I_C)_{\rm YY~Dor} - 2.5 \log 2.4 \cr
&=& 18.67 - 1.1\pm0.5 - 0.95 = 16.62\pm 0.5 \cr
&=& ((m - M)_I + \Delta I_C)_{\rm LMC~N~2009a} - 2.5 \log 1.51 \cr
&=& 18.67 - 1.6\pm0.5 - 0.45 = 16.62\pm 0.5.
\label{distance_modulus_i_v2491_cyg_yy_dor_lmcn2009a}\end{aligned}$$ Thus, we obtain $(m-M)_{I, \rm V2491~Cyg}= 16.62\pm0.3$.
We plot these three distance moduli of $B$, $V$, and $I_{\rm C}$ bands in Figure \[distance\_reddening\_v2615\_oph\_v2468\_cyg\_v2491\_cyg\_v496\_sct\](c) by the magenta, blue, and cyan lines, that is, $(m-M)_B= 17.83$, $(m-M)_V= 17.41$, and $(m-M)_I= 16.62$ together with Equations (\[distance\_modulus\_rb\]), (\[distance\_modulus\_rv\]), and (\[distance\_modulus\_ri\]). These three lines broadly cross at $d=15.9$ kpc and $E(B-V)=0.45$.
V496 Sct 2009 {#v496_sct}
-------------
Figure \[v496\_sct\_lv\_vul\_v1668\_cyg\_v\_bv\_ub\_color\_logscale\_no2\] shows the light/color curves of V496 Sct as well as LV Vul and V1668 Cyg. In the early phase, the $V$ light curve of V496 Sct overlaps that of LV Vul. In the middle phase, V496 Sct show a sharp and shallow dip due to dust blackout. In the later nebular phase, V496 Sct overlaps well with the upper branch of LV Vul, where the $V$ light curve of LV Vul splits into two branches in the nebular phase due to the different responses of their $V$ filters as discussed in Paper II. This is because strong \[\] lines contributes to the blue edge of $V$ band filter in the nebular phase and small differences among the response functions of $V$ filters make large differences in the $V$ magnitudes. The $(B-V)_0$ color curves of LV Vul also splits into two branches for $\log t\gtrsim 2.0$, that is, in the nebular phase. The lower branch of LV Vul in the $(B-V)_0$ color curve is close to that of V496 Sct. We regard that the $V$ light curve of V496 Sct follows V1668 Cyg and the upper branch of LV Vul and the $(B-V)_0$ color curve of V496 Sct follows the lower branch of LV Vul.
Based on the time-stretching method, we have the relation of $$\begin{aligned}
(m&-&M)_{V, \rm V496~Sct} \cr
&=& (m - M + \Delta V)_{V, \rm LV~Vul} - 2.5 \log 2.0 \cr
&=& 11.85 + 2.6\pm0.3 - 0.75 = 13.7\pm0.3 \cr
&=& (m - M + \Delta V)_{V, \rm V1668~Cyg} - 2.5 \log 2.0 \cr
&=& 14.6 - 0.1\pm0.3 - 0.75 = 13.75 \pm0.3 .
\label{distance_modulus_v496_sct_lv_vul_v1668_cyg}\end{aligned}$$ Thus, we obtain $f_{\rm s}=2.0$ against the template nova LV Vul and $(m-M)_V=13.7\pm0.2$. The new value is much smaller than the previous estimate of $(m-M)_V=14.4\pm0.1$ (Paper II). This is because we improved the timescaling factor of $f_{\rm s}$ and the vertical $V$ light curve fitting. From Equations (\[distance\_modulus\_general\_dot\]) and (\[distance\_modulus\_v496\_sct\_lv\_vul\_v1668\_cyg\]), we have the relation of $$\begin{aligned}
(m&-& M')_{V, \rm V496~Sct} \cr
&\equiv& (m_V - (M_V - 2.5\log f_s))_{\rm V496~Sct} \cr
&=& (m-M + \Delta V)_{V,\rm LV~Vul} \cr
&=& 11.85 + 2.6\pm0.3 =14.45\pm0.3.
\label{absolute_mag_v496_sct_lv_vul}\end{aligned}$$
Figure \[v496\_sct\_v1065\_cen\_lv\_vul\_v\_bv\_ub\_color\_logscale\] shows the light/color curves of V496 Sct as well as LV Vul and V1065 Cen. The light/color curves of V1065 Cen show different paths of V496 Sct. We regard that the $V$ light curve of V1065 Cen follows the lower branch of LV Vul and the $(B-V)_0$ color curve evolves along the upper branch of LV Vul in the nebular phase ($t > 100$ day). For V1065 Cen, @hac18k derived $(m-M)_V=15.0\pm0.2$ and $E(B-V)=0.45\pm0.05$ (and $f_{\rm s}=1.0$ against LV Vul) as summarized in Table \[extinction\_various\_novae\]. Applying Equation (\[distance\_modulus\_general\_temp\]) to Figure \[v496\_sct\_v1065\_cen\_lv\_vul\_v\_bv\_ub\_color\_logscale\], we have the relation of $$\begin{aligned}
(m&-&M)_{V, \rm V496~Sct} \cr
&=& (m-M + \Delta V)_{V, \rm LV~Vul} - 2.5 \log 2.0 \cr
&=& 11.85 + 2.6\pm 0.2 - 0.75 = 13.7\pm 0.2 \cr
&=& (m-M + \Delta V)_{V, \rm V1065~Cen} - 2.5 \log 2.0 \cr
&=& 15.0 - 0.5\pm 0.2 - 0.75 = 13.75\pm 0.2.
\label{distance_modulus_v496_sct_lv_vul_v1065_cen_v}\end{aligned}$$ This value of $(m-M)_V=13.7\pm0.2$ is consistent with Equation (\[distance\_modulus\_v496\_sct\_lv\_vul\_v1668\_cyg\]).
Figure \[v496\_sct\_v382\_vel\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\] plots the $B$, $V$, and $I_{\rm C}$ light curves of V496 Sct together with those of V382 Vel, and the LMC novae YY Dor and LMC N 2009a. We apply Equation (\[distance\_modulus\_general\_temp\_b\]) for the $B$ band to Figure \[v496\_sct\_v382\_vel\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](a) and obtain $$\begin{aligned}
(m&-&M)_{B, \rm V496~Sct} \cr
&=& ((m - M)_B + \Delta B)_{\rm YY~Dor} - 2.5 \log 10.5 \cr
&=& 18.98 - 2.3\pm0.2 - 2.55 = 14.13\pm0.2 \cr
&=& ((m - M)_B + \Delta B)_{\rm LMC~N~2009a} - 2.5 \log 6.6 \cr
&=& 18.98 - 2.8\pm0.2 - 2.05 = 14.13\pm0.2 \cr
&=& ((m - M)_B + \Delta B)_{\rm V382~Vel} - 2.5 \log 3.9 \cr
&=& 11.75 + 3.9\pm0.2 - 1.47 = 14.18\pm0.2.
\label{distance_modulus_b_v496_sct_v382_vel_yy_dor_lmcn2009a}\end{aligned}$$ Thus, we obtain $(m-M)_{B, \rm V496~Sct}= 14.15\pm0.1$, For the $V$ light curves in Figure \[v496\_sct\_v382\_vel\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](b), we similarly obtain $$\begin{aligned}
(m&-&M)_{V, \rm V496~Sct} \cr
&=& ((m - M)_V + \Delta V)_{\rm YY~Dor} - 2.5 \log 10.5 \cr
&=& 18.86 - 2.6\pm0.2 - 2.55 = 13.71\pm0.2 \cr
&=& ((m - M)_V + \Delta V)_{\rm LMC~N~2009a} - 2.5 \log 6.6 \cr
&=& 18.86 - 3.1\pm0.2 - 2.05 = 13.71\pm0.2 \cr
&=& ((m - M)_V + \Delta V)_{\rm V382~Vel} - 2.5 \log 3.9 \cr
&=& 11.5 + 3.7\pm0.2 - 1.47 = 13.73\pm0.2.
\label{distance_modulus_v_v496_sct_v382_vel_yy_dor_lmcn2009a}\end{aligned}$$ Thus, we obtain $(m-M)_{V, \rm V496~Sct}= 13.71\pm0.1$. We apply Equation (\[distance\_modulus\_general\_temp\_i\]) for the $I_{\rm C}$ band to Figure \[v496\_sct\_v382\_vel\_yy\_dor\_lmcn\_2009a\_b\_v\_i\_logscale\_3fig\](c) and obtain $$\begin{aligned}
(m&-&M)_{I, \rm V496~Sct} \cr
&=& ((m - M)_I + \Delta I_C)_{\rm YY~Dor} - 2.5 \log 10.5 \cr
&=& 18.67 - 3.1\pm0.2 - 2.55 = 13.02\pm 0.2 \cr
&=& ((m - M)_I + \Delta I_C)_{\rm LMC~N~2009a} - 2.5 \log 6.6 \cr
&=& 18.67 - 3.6\pm0.2 - 2.05 = 13.02\pm 0.2 \cr
&=& ((m - M)_I + \Delta I_C)_{\rm V382~Vel} - 2.5 \log 3.9 \cr
&=& 11.1 + 3.4\pm0.2 - 1.47 = 13.03\pm 0.2.
\label{distance_modulus_i_v496_sct_v382_vel_yy_dor_lmcn2009a}\end{aligned}$$ Thus, we obtain $(m-M)_{I, \rm V496~Sct}= 13.02\pm0.2$.
Ando, H., & Yamashita, Y. 1976, , 28, 171 Andreä, J., Drechsel, H., Snijders, M. A. J., & Cassatella, A. 1991, , 244, 111 Arkhipova, V. P., & Zaitseva, G. V. 1976, Soviet Astron. Lett., 2, 35 Austin, S. J., Wagner, R. M., Starrfield, S. et al. 1996, , 111, 869 Becker, H. J., & Duerbeck, H. W. 1980, , 92, 792 Belokon, E. T., & Larionov, V. M. 1977, Soviet Astronomy, 21, 355 Bode, M. F., Darnley, M. J., Beardmore, A. P., et al. 2016, , 818, 145 Bohlin, R. C., Savage, B. D., & Drake, J. F. 1978, , 224, 132 Burlak, M. A. 2008, Astronomy Letters, 34, 249 Cassatella, A., Altamore, A., & González-Riestra, R. 2002, , 384, 1023 Chincarini, G. 1964, , 76, 289 Chincarini, G., & Rosino, L. 1964, Annales d’Astrophysique, 27, 469 Chochol, D., Grygar, J., Pribulla, T., et al. 1997, , 318, 908 Chochol, D., Hric, L., Urban, Z., et al. 1993, , 277, 103 Chen, B.-Q., Huang, Y., Liu, X.-W., et al. 2018, , 483, 4277 Cheung, C. C., Jean, P., Shore, S. N., et al. 2013, The Astronomer’s Telegram, 5649, 1 Cohen, J. G. 1985, , 292, 90 Cohen, J. G. 1988, The extragalactic distance scale, Proceedings of the ASP 100th Anniversary Symposium, eds. S. van den Bergh & C. J. Prechet (San Francisco, ASP), ASP conference series, 4, 114 della Valle, M., & Livio, M. 1995, , 452, 704 della Valle, M., Pasquini, L., Daou, D., & Williams, R. E. 2002, , 390, 155 Diaz, M. P., Costa, R. D. D., & Jatenco-Pereira, V. 2001, , 113, 1554 Downes, R. A., & Duerbeck, H. W. 2000, , 120, 2007 Ederoclite, A., Mason, E., della Valle, M., et al., , 459, 875 Ennis, D., Becklin, E. E., Beckwith, S., et al. 1977, , 214, 478 Evans, P. A., Beardmore, A. P., Page, K. L., et al. 2009, , 397, 1177 Eyres, S. P. S., Davis, R. J., Bode, M. F., & Lloyd, H. M. 1996, , 6292, 2 Fernie, J. D. 1969, , 81, 374 Franckowiak, A., Jean, P., Wood, M., Cheung, C. C., & Buson, S. 2018, , 609, A120 Friedjung, M. 1966, , 132, 143
Furuyama, S., & Pearce, A. 2014, CBET, 3802, 1 Gallagher, J. S., & Holm, A. V. 1974, , 189, L123 Gallagher, J. S., & Ney, E. P. 1976, , 204, L35 Green, G. M., Schlafly, E. F., Finkbeiner, D. P., et al. 2015, , 810, 25 Green, G. M., Schlafly, E. F., Finkbeiner, D. P., et al. 2018, , 478, 651 Guido, E., Howes, N., Nicolini, M., et al. 2013, , 9265, 1 Güver, T., & Özel, F. 2009, , 400, 2050 Hachisu, I., & Kato, M. 2006, , 167, 59 Hachisu, I., & Kato, M. 2009, , 694, L103 Hachisu, I., & Kato, M. 2010, , 709, 680 Hachisu, I., & Kato, M. 2014, , 785, 97 (Paper I) Hachisu, I., & Kato, M. 2015, , 798, 76 Hachisu, I., & Kato, M. 2016a, , 816, 26 Hachisu, I., & Kato, M. 2016b, , 223, 21 (Paper II) Hachisu, I., & Kato, M. 2017, in Proceedings of the Palermo Workshop 2017 on “The Golden Age of Cataclysmic Variables and Related Objects - IV”, ed. F. Giovannelli et al. (Trieste: SISSA PoS), 315, 47 Hachisu, I., & Kato, M. 2018a, , 858, 108 Hachisu, I., & Kato, M. 2018b, , 237, 4 Hachisu, I., & Kato, M. 2019, , submitted Hachisu, I., Kato, M., Kiyota, S., et al. 2006b, , 651, L141 Hachisu, I., Kato, M., Kiyota, S., et al. 2008b, RS Ophiuchi (2006) and the Recurrent Nova Phenomenon, ASP Conf. Ser. 401, eds. A. Evans et al. (San Francisco, ASP), p.206. Hachisu, I., Kato, M., & Luna, G. J. M. 2007, , 659, L153 Harrison, T. E., Bornak, J., McArthur, B. E., & Benedict, G. F. 2013, , 767, 7 Hric, L., Petrík, K., Urban, Z., Hanžl, D., , 133, 211 Ibarra, A., & Kuulkers, E. 2008, The Astronomer’s Telegram, 1473, 1 Ibarra, A., Kuulkers, E., Beardmore, A., et al. 2008, The Astronomer’s Telegram, 1478, 1 Imara, N., & Blitz, L. 2007, , 662, 969 Izzo, L., Mason, E., Vanzi, L., et al. 2013, The Astronomer’s Telegram, 5639, 1 Kato, M., & Hachisu, I., 1994, , 437, 802 Kato, M., Saio, H., Hachisu, I., & Nomoto, K. 2014, , 793, 136 Kimeswenger, S., Dalnodar, S., Knapp, A., et al. 2008, , 479, L51 Kolotilov, E. A., & Noskova, R. I. 1986, Soviet Astronomy Letters, 12, 370 Krautter, J., Beuermann, K., Leitherer, C., et al. 1984, , 37, 307 Kreiner, J. M., Kurpińska, M., & Winiarski, M. 1966, Acta Astronomica, 16, 137 Kuulkers, E., Ness, J.-U., Ibarra, A., et al. 2013, The Astronomer’s Telegram, 5628, 1 Liszt, H. S. 2014, , 780, 10 Lynch, D. K., Russell, R. W., Rudy, R. J., Woodward, C. E., & Schwarz, G. J. 2008b, , 8935, 1 Marshall, D. J., Robin, A. C., Reylé, C., Schultheis, M., & Picaud, S. 2006, , 453, 635 Mason, E., Shore, S. N., De Gennaro Aquino, I., et al. 2018, , 853, 27 Munari, U., & Zwitter, T. 1997, , 318, 269 Munari, U., Yudin, B. F., Kolotilov, E. A., et al. 1994a, , 284, L9 Munari, U., Tomov, T. V., Hric, L., Hazucha, P. 1994b, Inf. Bull. Variable Stars, 3977, 1 Munari, U., Henden, A., Valentini, M., et al. 2008a, , 387, 344 Munari, U., Siviero, A., Henden, A., et al. 2008b, , 492, 145 Munari, U., Siviero, A., Dallaporta, S., et al. 2011, New Astronomy, 16, 209 Naik, S., Banerjee, D. P. K., & Ashok, N. M. 2009, , 394, 1551 Naik, S., Banerjee, D. P. K., Ashok, N. M., Das, R. K. 2010, , 404, 367 Naito, H., Mizoguchi, S., Arai, A., et al. 2012, , 543, A86 Ness, J.-U., Schwarz, G. J., Retter, A., et al. 2007a, , 663, 505 Ness, J.-U., Starrfield, S., Jordan, C., Krautter, J., & Schmitt, J. H. M. M. 2005, , 364, 1015 Nishiyama, S., Tamura, M., Hatano, H., et al. 2009, , 696, 1407 Noskova, R. I., Zaitseva, G. V., & Kolotilov, E. A. 1985, Soviet Astronomy Letters, 11, 257 Orio, M., Parmar, A. N., Greiner, J., et al. 2002, , 333, L11 Özdörmez, A., Güver, T., Cabrera-Lavers, A., Ak, T. 2016, , 461, 1177 Özdörmez, A., Ege, E., Güver, T., & Ak, T. 2018, , 476, 4162 Page, K. L., Beardmore, A. P., Osborne, J. P., & Schwarz, G. J. 2014, The Astronomer’s Telegram, 5966, 1 Piatti, A., & Geisler, D. 2013 , 145, 17 Pietrzyński, G., Graczyk, D., Gieren, W., et al. 2011, , 495, 76 Pfau, W. 1976, Inf. Bull. Variable Stars, 1106, 1 Poznanski, D., Prochaska, J. X., Bloom, J. S. 2012, , 426, 1465 Raj, A., Ashok, N. M., Banerjee, D. P. K., et al. 2012, , 425, 2576 Rieke, G. H., & Lebofsky, M. J. 1985, , 288, 618 Robb, R. M., & Scarfe, C. D. 1995, , 273, 347 Rudy, R. J., Lynch, D. K., Russell, R. W., Woodward, C. E., & Covey, K. 2008b, , 8938, 2 Saito, R. K., Minniti, D., Angeloni, R., et al. 2013, , 554, A123 Saizar, P., Starrfield, S., Ferland, G. J., et al. 1991, , 367, 310 Sale, S. E., Drew, J. E., Barentsen, G., et al. 2014, , 443, 2907 Schaefer, B. E. 2018, , 481, 3033 Schmeer, P., & Gualdoni, C. 2008, , 8999, 4 Schlafly, E. F., & Finkbeiner, D. P. 2011, , 737, 103 Schmidt, Th. 1957, Z. Astrophys., 41, 181 Schultheis, M., Chen, B. Q., Jiang, B. W., et al. 2014, , 566, A120 Schwarz, G. J., Ness, J.-U., Osborne, J. P., et al. 2011, , 197, 31 Seaton, M. J. 1979, , 187, 73 Selvelli, P., & Gilmozzi, R. 2013, , 560, A49 Shen, L.-Z., et al. 1964, Acta Astronomica Sinica, 12, 83 Shore, S. N., Schwarz, G., Bond, H. E., et al. 2003, , 125, 1507 Shore, S. N., Schwarz, G. J., Walter, F. M., et al. 2014, The Astronomer’s Telegram, 6413, 1 Slavin, A. J., O’Brien, T. J., Dunlop, J. S. 1995, , 276, 353 Slovak, M. H. & Vogt, S. S. 1979, Nature, 277, 114 Steiner, J. E., Campos, R., & Cieslinski, D. 1999, , 7185, 2 Tarasova, T. N. 2014b, Astronomy Reports, 58, 302 Tempesti, P. 1972, , 20, 63 Tempesti, P. 1979, Astronomische Nachrichten, 300, 51 The, P. S., & van der Klis, M. 1976, Inf. Bull. Variable Stars, 1089, 1 Tomkin, J., Lambert, D. L., & Woodman, J. 1976, , 48, 319 Tomov, T., Swierczynski, E., Miko[ł]{}ajewski, M., & Ilkiewicz, K. 2015, , 576, A119 van den Bergh, S., & Younger, P. F. 1987, , 70, 125 van Genderen, A. M. 1963, Bull. of the Astr. Inst. of the Netherlands, 17, 293 Waagen, E. O., Henden, A., Maury, A., et al. 2008, , 8999, 1 Wade, R. A., Ciardullo, R., Jacoby, G. H., & Sharp, N. A. 1991, , 102, 1738 Walter, F. M., Battisti, A., Towers, S. E., Bond, H. E., & Stringfellow, G. S. 2012, , 124, 1057 Williamon, R. M. 1977, , 89, 44 Williams, R. E. 1992, , 104, 725 Woodward, C. E., Gehrz, R. D., Jones, T. J., Lawrence, G. F., & Skrutskie, M. F. 1997, , 477, 817 Wu, C.-C., Holm, A. V., Panek, R. J., et al. 1989, , 339, 443 Young, P. J., Corwin, H. G., Bryan, J., & de Vaucouleurs, G. 1976, , 209, 882
[^1]: http://irsa.ipac.caltech.edu/applications/DUST/
[^2]: http://www.astro.sunysb.edu/fwalter/SMARTS/NovaAtlas/atlas.html
[^3]: http://www.swift.ac.uk/
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
The main goal of this article is to understand the trace properties of nonlocal minimal graphs in ${\ensuremath{\mathbb{R}}}^3$, i.e. nonlocal minimal surfaces with a graphical structure.
We establish that at any boundary points at which the trace from inside happens to coincide with the exterior datum, also the tangent planes of the traces necessarily coincide with those of the exterior datum.
This very rigid geometric constraint is in sharp contrast with the case of the solutions of the linear equations driven by the fractional Laplacian, since we also show that, in this case, the fractional normal derivative can be prescribed arbitrarily, up to a small error.
We remark that, at a formal level, the linearization of the trace of a nonlocal minimal graph is given by the fractional normal derivative of a fractional Laplace problem, therefore the two problems are formally related. Nevertheless, the nonlinear equations of fractional mean curvature type present very specific properties which are strikingly different from those of other problems of fractional type which are apparently similar, but diverse in structure, and the nonlinear case given by the nonlocal minimal graphs turns out to be significantly more rigid than its linear counterpart.
author:
- 'Serena Dipierro${}^{(1)}$'
- 'Ovidiu Savin${}^{(2)}$'
- 'Enrico Valdinoci${}^{(1)}$'
title: |
Boundary properties of fractional objects:\
flexibility of linear equations\
and rigidity of minimal graphs
---
\(1) – Department of Mathematics and Statistics\
University of Western Australia\
35 Stirling Highway, WA6009 Crawley (Australia)\
\(2) – Department of Mathematics\
Columbia University\
2990 Broadway, NY 10027 New York (USA)
E-mail addresses: [[email protected]]{},\
[[email protected]]{},\
[[email protected]]{}
Introduction
============
Boundary behavior of fractional objects
---------------------------------------
This article investigates the geometric properties at the boundary of solutions of fractional problems. Two similar, but structurally significantly different, situations are taken into account. On the one hand, we will consider the solution of the [*linear fractional equation*]{} $$\begin{cases}
(-\Delta)^\sigma u=0 & {\mbox{ in }} B_1\cap \{x_n>0\},\\
u=0 & {\mbox{ in }} \{x_n<0\},\end{cases}$$ where $\sigma\in(0,1)$, and, for $x'=(x_1,\dots,x_{n-1})$ with $|x'|<1$, we consider the “fractional boundary derivative” $$\label{56:62734899fi02}
\displaystyle\lim_{x_n\searrow0} \frac{u(x',x_n)}{x_n^\sigma}
.$$ Interestingly, the function in plays an important role in understanding fractional equations, see [@MR3168912]. In particular, while classical elliptic equations are smooth up to the boundary, the solutions of fractional equations with prescribed exterior datum are in general not better than Hölder continuous with exponent $\sigma$, and therefore the function in is the crucial ingredient to detect the growth of the solution in the vicinity of the boundary.
As a first result, we will show here that, roughly speaking, the function in can be [*arbitrarily prescribed, up to an arbitrarily small error*]{}. That is, one can construct solutions of linear fractional equations whose fractional boundary derivative behaves in an essentially arbitrary way.
Then, we turn our attention to the boundary property of [*nonlocal minimal graphs*]{}, i.e. minimizers of the fractional perimeter functional which possess a graphical structure. In this case, we show that the boundary properties are subject to severe geometric constraints, in sharp contrast with the case of fractional equations.
First of all, the continuity properties of nonlocal minimal graphs are very different from those of the solutions of fractional equations, since we have established in [@MR3516886; @MR3596708] that nonlocal minimal graphs are not necessarily continuous at the boundary. In addition, the boundary discontinuity of nonlocal minimal graphs in the plane happens to be a “generic” situation, as we have recently proved in [@2019arXiv190405393D].
The focus of this article is on the three-dimensional setting, i.e. the case in which the graph is embedded in ${\ensuremath{\mathbb{R}}}^3$. In this situation, the graph can be continuous at a given point, but the discontinuity may occur along the trace, at nearby points. More precisely, we will look at a function $u:{\ensuremath{\mathbb{R}}}^2\to{\ensuremath{\mathbb{R}}}$, which is an $s$-minimal graph in $(-2,2)\times(0,4)$ and is such that $u=0$ in $ (-2,2)\times(-h,0)$, for some $h>0$. In this setting, we will consider its trace along $\{x_2=0\}$, namely we consider the function $$\lim_{x_2\searrow0}u(x_1,x_2).$$ The main question that we address in this article is precisely whether or not [*the trace of a nonlocal minimal graph possesses any distinctive feature or satisfies any particular geometric constraint*]{}.
We will prove that, differently from the case of the linear equations (and also in sharp contrast to the case of classical minimal surfaces), the traces of nonlocal minimal graphs [*cannot have an arbitrary shape*]{}, and, in fact, [*matching points from the two sides must necessarily occur with horizontal tangencies*]{}.
This result relies on a classification theory for homogeneous graphs, since we will show that, in this case, [*the matching at the origin is sufficient to make a nonlocal minimal graph trivial*]{}.
In the rest of this introduction, we will present the precise mathematical framework in which we work and provide the formal statements of our main results.
Boundary flexibility of linear fractional equations
---------------------------------------------------
We discuss now the case of fractional linear equations, showing that the fractional boundary derivative of the solutions can be essentially arbitrarily prescribed, up to a small error. For this, we denote by $B'_r$ the $(n-1)$-dimensional ball of radius $r$ centered at the origin, namely $$B'_r := \{ x'\in{\ensuremath{\mathbb{R}}}^{n-1} {\mbox{ s.t. }} |x'|<r\}.$$ As customary, given $\sigma\in(0,1)$, we define the fractional Laplacian as $$(-\Delta)^\sigma u(x):=\int_{{\ensuremath{\mathbb{R}}}^n}\frac{2u(x)-u(x+y)-u(x-y)}{|y|^{n+\sigma}}
\,dy.$$ Then, we have:
\[FLFL\] Let $n{\geqslant}2$, $\sigma\in(0,1)$, $k\in{\ensuremath{\mathbb{N}}}$ and $f\in C^k(\overline{B_1'})$. Then, for every ${\varepsilon}>0$ there exist $f_{\varepsilon}\in C^k({\ensuremath{\mathbb{R}}}^{n-1})$ and $u_{\varepsilon}\in C({\ensuremath{\mathbb{R}}}^n)$ such that $$\label{56:01} \begin{cases}
(-\Delta)^\sigma u_{\varepsilon}=0 & {\mbox{ in }} B_1\cap \{x_n>0\},\\
u_{\varepsilon}=0 & {\mbox{ in }} \{x_n<0\},\end{cases}$$ $$\label{56:02}
\displaystyle\lim_{x_n\searrow0} \frac{u_{\varepsilon}(x',x_n)}{x_n^\sigma}=f_{\varepsilon}(x')\qquad{\mbox{ for all }}
x'\in B_1',$$ and $$\label{56:03}
\| f_{\varepsilon}-f\|_{C^k(\overline{B_1'})}{\leqslant}{\varepsilon}.$$
On the one hand, Theorem \[FLFL\] falls in the research line opened in [@MR3626547] according to which “all functions are $\sigma$-harmonic up to a small error”, namely it states an interesting flexibility offered by solutions of fractional equations which can adapt themselves in order to capture essentially any prescribed behavior. This flexible feature has been recently studied in several fractional contexts, including time-fractional derivatives, non-elliptic operators, and higher order operators, see [@MR3716924; @MR3935264; @KRYL; @CAR; @CARBOO]. In addition, the flexible properties of fractional equations can be effectively exploited to construct interesting counterexamples, see [@MR3783214], and they have consequences in concrete scenarios involving also mathematical biology and inverse problems, see [@MR3579567; @MR3774704]. Differently from the previous literature, Theorem \[FLFL\] aims at detecting a “boundary” flexibility of fractional equations, rather than an “interior” one.
On the other hand, differently from all the other fractional flexibility results in the literature, which have no counterpart for the case of the classical Laplacian, Theorem \[FLFL\] shares a common treat with the Laplace equation and possesses a full classical analogue (we present its classical counterpart in Appendix \[APPA\]).
Boundary rigidity of fractional minimal surfaces
------------------------------------------------
We now discuss the boundary behavior of $s$-minimal surfaces and we will show its striking differences with respect to the linear fractional equations. To this end, we recall the setting introduced in [@MR2675483]. Given $s\in(0,1)$, we consider the interaction of two disjoint (measurable) sets $F$, $G\subseteq{\ensuremath{\mathbb{R}}}^N$ defined by $$\label{nNdi}
{\mathcal{I}}_s(F,G):=\iint_{F\times G}\frac{dx\,dy}{|x-y|^{N+s}}.$$ Given a bounded reference domain $\Omega$ with Lipschitz boundary, we define the $s$-perimeter of a set $E\subseteq{\ensuremath{\mathbb{R}}}^N$ in $\Omega$ by $${\rm Per}_s(E;\Omega):=
{\mathcal{I}}_s(E\cap\Omega,E^c\cap\Omega)+
{\mathcal{I}}_s(E\cap\Omega,E^c\cap\Omega^c)+
{\mathcal{I}}_s(E\cap\Omega^c,E^c\cap\Omega).$$ As customary, we have used here the complementary set notation $E^c:={\ensuremath{\mathbb{R}}}^N\setminus E$.
Let $\Omega\subset{\ensuremath{\mathbb{R}}}^N$ be bounded and with Lipschitz boundary. Let $E\subseteq{\ensuremath{\mathbb{R}}}^N$. We say that $E$ is $s$-minimal in $\Omega$ if ${\rm Per}_s(E;\Omega)<+\infty$ and $$\label{MINIMIZZ} {\rm Per}_s(E;\Omega){\leqslant}{\rm Per}_s(E';\Omega)$$ for every $E'\subseteq{\ensuremath{\mathbb{R}}}^N$ such that $E'\cap\Omega^c=E\cap\Omega^c$.
Moreover, given $U\subset{\ensuremath{\mathbb{R}}}^N$, we say that $E$ is locally $s$-minimal in $U$ if it $s$-minimal in $
\Omega$, for every $\Omega$ which is bounded, with Lipschitz boundary and strictly contained in $U$.
We remark that one can make sense of the minimization procedure also in unbounded domains by saying that $E$ is locally $s$-minimal in a (possibly unbounded) domain $\Omega$ if $E$ is $s$-minimal in every bounded and Lipschitz domain $\Omega'\Subset\Omega$ (see Section 1.3 in [@MR3827804] for additional details on these minimality notions).
The regularity theory of nonlocal minimal surfaces is a fascinating topic of investigation, still possessing a number of fundamental open problems. We refer to [@MR3090533; @MR3107529; @MR3331523; @BV] for interior regularity results, [@MR3798717; @CCC] for a precise discussion on stable nonlocal cones, and [@MR3588123; @MR3824212] for recent surveys containing the state of the art of this problem.
A particularly important case of locally $s$-minimal sets is given by the ones which have a graph structure. Namely, given $\Omega_0\subseteq{\ensuremath{\mathbb{R}}}^n$ and $u:\Omega_0\to{\ensuremath{\mathbb{R}}}$, we let $$\label{LEYUDEF} E_u:=\{ X=(x,x_{n+1}) {\mbox{ s.t. $x\in\Omega_0$ and $ x_{n+1}<u(x)$}}\}.$$ With respect to the notation in , we are taking here $N=n+1$.
We say that $u$ is an $s$-minimal graph in $\Omega_0$ if $E_u$ is a locally $s$-minimal set in $\Omega_0\times{\ensuremath{\mathbb{R}}}$.
Interestingly, $s$-minimal graphs enjoy suitable Bernstein-type properties, see [@MR3680376; @PISA; @2018arXiv180705774C], and they have a smooth interior regularity theory, as proved in [@MR3934589].
See also [@NOCHETTO] for several very precise simulations on nonlocal minimal graphs and a sharp numerical analysis of their properties.
With this, we are ready to state the main result of this article, which gives some precise geometric conditions on the trace of nonlocal minimal graphs. We establish that the trace graph has necessarily [*zero derivatives when the trace crosses zero*]{}. The precise result that we have is the following one:
\[NLMS\] Let $u$ be an $s$-minimal graph in $(-2,2)\times(0,4)$. Assume that there exists $h>0$ such that $u=0$ in $ (-2,2)\times(-h,0)$, and let $$\label{MR3516886EQ}
f(x_1):=\lim_{x_2\searrow0}u(x_1,x_2).$$ Then, there exist $\delta_0\in\left(0,\frac1{100}\right)$ and $C>0$ such that if $\zeta_0\in\left(-\frac{3}2,\frac32\right)$ is such that $f(\zeta_0)=0$, then $$\label{VANP} |u(x)|{\leqslant}C\,|x-(\zeta_0,0)|^{\frac{3+s}2}$$ for every $x\in B_{\delta_0}(\zeta_0,0)$, and, in particular, $$\label{THAN-2}
{\mbox{$f'(\zeta_0)=0$.}}$$
We remark that the existence of the limit in is warranted by Theorem 1.1 in [@MR3516886].
![*[[A nonlocal minimal graph, in the light of Theorem \[NLMS\]. The red curve represents the graph of the function $f$ defined in .]{}]{}*[]{data-label="DOMA"}](traccia.pdf){width="8.9cm"}
The statement of Theorem \[NLMS\] is described[^1] in Figures \[DOMA\] and \[DOMA-2\]. With respect to this, we stress the remarkable geometric property given by the [*vanishing of the gradient of the trace at the zero crossing points*]{}. We observe that this situation is completely different with respect to the one arising for solutions of linear equations, and one can compare the structurally “rigid” geometry imposed by Theorem \[NLMS\] with the almost completely arbitrariness arising in Theorem \[FLFL\].
We remark that, at a formal level, the settings in Theorems \[FLFL\] and \[NLMS\] are strictly related, since the linearization of the trace of a nonlocal minimal graph is given by the fractional normal derivative of a fractional Laplace problem. More specifically, when one takes into account the improvement of flatness argument for an ${\varepsilon}$-flat nonlocal minimal graph $u$ (see the forthcoming Lemma \[NOANpierjfppp34\]), one sees that $u/{\varepsilon}$ shadows a function $\bar u$, which is a solution of $(-\Delta)^{\sigma}\bar u=0$ in $\{x_2>0\}$, with $\sigma:=\frac{1+s}2$: in this context the first order of $\bar u$ near the origin takes the form $\bar a \,x_2^{\sigma}$, for some $a\in{\ensuremath{\mathbb{R}}}$. Comparing with , one has that $\bar a$ is exactly the fractional normal derivative of the solution of a linear equation, which, in view of Theorem \[FLFL\], can be prescribed in an essentially arbitrary way.
In this spirit, if the linearization procedure produced a “good approximation” of the nonlinear geometric problem, one would expect that the original nonlocal minimal graph $u$ is well approximated near the origin by a term of the form ${\varepsilon}\,\bar a \,x_2^{\sigma}$, with no prescription whatsoever on $\bar a$. Quite surprisingly, formula (applied here with $\zeta_0:=0$) tells us that this is not the case, and the correct behavior of a nonlocal minimal graph at the boundary [*cannot be simply understood “by linearization”*]{}.
![*[[How we expect the graph of the function $f$ defined in when $n=2$ and the exterior datum is $$u(x):=\chi_{(-2,0)\times(-1,-3)}(x)-
\chi_{(0,2)\times(-1,-3)}$$ for all $x\in{\ensuremath{\mathbb{R}}}^2\setminus\big( (-2,2)\times(0,4)\big)$. Notice the horizontal tangency at the origin.]{}]{}*[]{data-label="DOMA-2"}](traccia-2.pdf){width="8.9cm"}
Theorem \[NLMS\] also reveals a [*structural difference*]{} of the boundary regularity theory of fractional minimal surfaces embedded in ${\ensuremath{\mathbb{R}}}^{n+1}$ when $n=2$ with respect to the case in which $n=1$. Indeed, when $n=1$, an $s$-minimal graph in which the exterior datum is attained continuously at a boundary point is necessarily $C^{1,\frac{1+s}2}$ in a neighborhood of such a point (see Theorem 1.2 in [@2019arXiv190405393D]). Instead, when $n=2$, a similar result does not hold: as a matter of fact, when $n=2$,
- Theorem \[NLMS\] guarantees that boundary points which attain the flat exterior datum in a continuous way have necessarily horizontal tangency,
- conversely, boundary points in which the $s$-minimal graph experience a jump have necessarily a vertical tangency (see [@MR3532394]).
Consequently, [*points with vertical tangency accumulate to zero crossing points possessing horizontal tangency*]{}, preventing a differentiable boundary regularity of the surface in a neighborhood of the latter type of points.
The proof of Theorem \[NLMS\] will require a fine understanding of fractional minimal homogeneous graphs. Indeed, as a pivotal step towards the proof of Theorem \[NLMS\], we establish the surprising feature that [*if a homogeneous fractional locally minimal graph vanishes in $\{x_n<0\}$ and it is continuous at the origin, then it necessarily vanishes at all points of $\{x_n=0\}$*]{}. The precise result that we obtain is the following:
\[QC\] Let $u$ be an $s$-minimal graph in ${\ensuremath{\mathbb{R}}}\times(0,+\infty)$. Assume that $$\label{ADX}
{\mbox{$u(x_1,x_2)=0$ if~$x_2<0$.}}$$ Assume also that $u$ is positively homogeneous of degree $1$, i.e. $$\label{ADX6}
u(tx)=tu(x)\qquad{\mbox{ for all $x\in{\ensuremath{\mathbb{R}}}^2$ and~$t>0$.}}$$ Suppose that $$\label{ADX3}
u(0):=\lim_{x\to0} u(x)=0.$$ Then $u(x)=0$ for all $x\in{\ensuremath{\mathbb{R}}}^2$.
We take this opportunity to state and discuss [*some new interesting research lines*]{} opened by the results obtained in the present paper.
\[16\]
In the setting of Theorem \[NLMS\], can one construct examples in which $f'(\zeta)=\pm\infty$ for some $\zeta\in\left(-\frac32,\frac32\right)$?
Namely, is it possible to construct examples of nonlocal minimal graphs embedded in ${\ensuremath{\mathbb{R}}}^3$ which are flat from outside and whose trace develops vertical tangencies?
The trace of such possible pathological examples is depicted in Figure \[DOMA-122\].
![*[[Open problem \[16\]: is it possible to construct nonlocal minimal graphs traces with vertical tangencies?]{}]{}*[]{data-label="DOMA-122"}](traccia-3.pdf){width="8.9cm"}
\[17\]
It would be interesting to determine whether or not a result similar to Theorem \[NLMS\] holds true in higher dimension. Similarly, it would be interesting to determine the possible validity of Theorem \[QC\] in higher dimensions.
[F]{}rom the technical point of view, we observe that some of the auxiliary results exploited towards the proof of Theorem \[QC\] (such as Lemma \[STEP1\] and Corollary \[STEP11\]) are expected to carry over in higher dimension, therefore one can in principle try to argue by induction, supposing that a statement such as the one of Theorem \[QC\] holds true in dimension $n$ with the aim of proving it in dimension $n+1$. The catch in this argument is that one is led to study the points at which the gradient of the trace attains its maximal, and this makes an important connection between this line of research and that of Open Problem \[16\].
It would be interesting to detect the behavior of a nonlocal minimal graph and of its trace at the corners of the domain and in their vicinity, in particular understanding (dis)continuity and tangency properties, possibly also in relation with the convexity or concavity of the corner. This is related to the analysis of nonlocal minimal cones in either convex or concave sectors with zero exterior datum.
As a first step towards it, one can try to understand how to complete Figure \[DOMA-2\] near $x_1=\pm2$.
Organization of the paper
-------------------------
The rest of this manuscript is organized as follows. Theorem \[FLFL\] is proved in Section \[ISsta\]. The arguments used will exploit a method that we have recently introduced in [@MR3626547] to show that “all functions are locally fractional harmonic”, and a careful discussion of the homogeneous solutions of fractional equations on cones, see [@MR2075671; @MR3810469].
Then, in Section \[15kdttd3\] we present the proof of Theorem \[QC\]. The arguments used here exploit and develop a series of fine methods from the theory of nonlocal equations, comprising boundary Lipschitz bounds, blow-up classification results, continuity implies differentiability results, nonlocal geometric equations and nonlocal obstacle-type problems.
In Section \[HDB\] we construct a useful barrier, that we exploit to rule out the case of boundary Lipschitz singularities for nonlocal minimal graphs.
The proof of Theorem \[NLMS\] is contained in Section \[657699767395535\]. Finally, in Appendix \[APPA\], we point out the classical analogue of Theorem \[FLFL\].
Proof of Theorem \[FLFL\] {#ISsta}
=========================
One important ingredient towards the proof of Theorem \[FLFL\] consists in the construction of a homogeneous solution of a linear fractional equation with a suitable growth from the vertex of a cone. This is indeed a classical topic of research, which also bridges mathematical analysis and probability, see [@MR1936081; @MR2075671; @MR3810469] for specific results on fractional harmonic functions on cones. In our setting, we can reduce to the two-dimensional case (though the higher dimensional case can be treated in a similar way), and, for any $\alpha>0$, we let $$\label{89apq0}
{\mathcal{C}}_\alpha:=\left\{ x=(x_1,x_2)\in{\ensuremath{\mathbb{R}}}^{2}
{\mbox{ s.t. $ x_2>0$ and $x_1+\displaystyle\frac{ x_2 }{\alpha }>0
$}}\right\},$$ and the result that we need is the following one:
\[BAN\] For every $\sigma\in(0,1)$ and every $\vartheta\in{\ensuremath{\mathbb{R}}}$ there exist $\alpha>0$, $\beta\in(0,2\sigma)\setminus\{ \vartheta\}$ and $\bar{v}:S^1\to [0,+\infty)$ with $\| \bar{v}\|_{L^\infty(S^{1})}=1$ such that the function $$\label{eca-a} {\ensuremath{\mathbb{R}}}^2\ni x\mapsto v(x):=|x|^\beta\,\bar{v}\left( \frac{x}{|x|}\right)$$ satisfies $$\label{eca} \begin{cases}
(-\Delta)^\sigma v = 0 & {\mbox{ in }}{\mathcal{C}}_\alpha,\\
v=0 & {\mbox{ in }}{\ensuremath{\mathbb{R}}}^2\setminus{\mathcal{C}}_\alpha.\end{cases}$$
Let $\vartheta\in{\ensuremath{\mathbb{R}}}$. By Theorem 3.2 in [@MR2075671], for every $\alpha>0$ there exist $\beta(\alpha)\in(0,2\sigma)$ and $\bar{v}_\alpha:S^1\to [0,+\infty)$ with $\| \bar{v}_\alpha\|_{L^\infty(S^{1})}=1$ such that the function $${\ensuremath{\mathbb{R}}}^n\ni x\mapsto v_\alpha(x):=|x|^\beta\,\bar{v}_\alpha\left( \frac{x}{|x|}\right)$$ satisfies $$\begin{cases}
(-\Delta)^\sigma v_\alpha = 0 & {\mbox{ in }} {\mathcal{C}}_\alpha,\\
v_\alpha=0 & {\mbox{ in }}{\ensuremath{\mathbb{R}}}^2\setminus{\mathcal{C}}_\alpha.\end{cases}$$ Let us focus on the case $\alpha=1$. If $\beta(1)\ne\vartheta$, then the claims of Lemma \[BAN\] are satisfied by choosing $\alpha:=1$, $\beta:=\beta(1)$ and $\bar{v}:=\bar{v}_1$.
If instead $\beta(1)=\vartheta$, we exploit Lemma 3.3 in [@MR2075671]. Namely, since ${\mathcal{C}}_{1/2}\supset {\mathcal{C}}_1$, we deduce from Lemma 3.3 in [@MR2075671] that $\beta(1/2)<
\beta(1)=\vartheta$, and thus the claims of Lemma \[BAN\] are satisfied in this case by choosing $\alpha:=1/2$, $\beta:=\beta(1/2)$ and $\bar{v}:=\bar{v}_{1/2}$.
Exploiting Lemma \[BAN\], we will obtain that the boundary derivatives of $\sigma$-harmonic functions have maximal span. For this, given $k\in{\ensuremath{\mathbb{N}}}$, we define by ${\mathcal{H}}_k$ the set of all functions $u\in C({\ensuremath{\mathbb{R}}}^n)$ for which there exists $r>0$ such that $$\label{KK} \begin{cases}
(-\Delta)^\sigma u=0 & {\mbox{ in }} B_r\cap \{x_n>0\},\\
u=0 & {\mbox{ in }} \{x_n<0\},\end{cases}$$ and for which there exists $\phi\in C^k(\overline{B'_r})$ such that $${\mathcal{T}}u(x'):=
\lim_{x_n\searrow0} \frac{u(x',x_n)}{x_n^\sigma}=\phi(x')\qquad{\mbox{ for all }}
x'\in B_r'.$$ Also, given $r>0$ and $\phi\in C^k(\overline{B'_r})$, we consider the array $$\label{mul}{\mathcal{D}}^k\phi(0):=\big(
D^\gamma \phi(0)
\big)_{{\gamma\in{\ensuremath{\mathbb{N}}}^{n-1}}\atop{|\gamma|{\leqslant}k}}\;.$$ Namely, the array ${\mathcal{D}}^k\phi(0)$ contains all the derivatives of $\phi$ at the origin, up to order $k$. Fixing some order in the components of the multiindex, we can consider ${\mathcal{D}}^k\phi(0)$ as a vector in ${\ensuremath{\mathbb{R}}}^{N_k}$, with $$\label{ennekappa} N_k:=\sum_{j=0}^k (n-1)^j.$$ The following result states that the linear space produced in this way is “as large as possible”:
\[fan\] We have that $$\label{SP} \Big\{ {\mathcal{D}}^k\phi(0), {\mbox{ with }}\phi={\mathcal{T}}u
{\mbox{ and }}u\in {\mathcal{H}}_k\Big\}={\ensuremath{\mathbb{R}}}^{N_k}.$$
Suppose, by contradiction, that the linear space in the left hand side of does not exhaust the whole of ${\ensuremath{\mathbb{R}}}^{N_k}$. Then, there would exist $$\label{0876tg456}
\omega\in {\ensuremath{\mathbb{R}}}^{N_k}\setminus\{0\}$$ such that the linear space in the left hand side of lies in the orthogonal space of $\omega$. Namely, for every $u\in {\mathcal{H}}_k$ with $\phi={\mathcal{T}}u$, $$\label{mul2}
{\mathcal{D}}^k\phi(0)\cdot\omega=0.$$ Recalling the notation in , we can write $\omega=(\omega_\gamma)_{{\gamma\in{\ensuremath{\mathbb{N}}}^{n-1}}\atop{|\gamma|{\leqslant}k}}$, and then takes the form $$\label{mul3}
\sum_{{\gamma\in{\ensuremath{\mathbb{N}}}^{n-1}}\atop{|\gamma|{\leqslant}k}}\omega_\gamma\,
D^\gamma\phi(0)=0,\qquad
{\mbox{ for all~$u\in {\mathcal{H}}_k$ with~$\phi={\mathcal{T}}u$. }}$$ Now, we exploit Lemma \[BAN\] with $\vartheta:=\sigma$. In the notation of Lemma \[BAN\], we take $P=(P_1,P_2)\in{\mathcal{C}}_\alpha\subset{\ensuremath{\mathbb{R}}}^2$ such that $v(P)>0$ and define $$\tilde v(x_1,x_2):=\frac{v(P)\;x_2^\sigma}{P_2^\sigma}.$$ We observe that $\tilde v(P)=v(P)$. Accordingly, by the Boundary Harnack Inequality (see Theorem 1 on page 44 of [@MR1438304]), we have that, for all ${\varepsilon}>0$, $$\label{5302345}
\left[ \frac1C,C\right]\ni \Lambda:=\lim_{x_n\searrow 0} \frac{v({\varepsilon},x_n)}{
\tilde v({\varepsilon},x_n)}=
\lim_{\tau\searrow 0} \frac{P_2^\sigma \;v({\varepsilon},\tau)}{
v(P)\;\tau^\sigma},$$ for some $C{\geqslant}1$.
Now, for all ${\varepsilon}>0$ and $\zeta\in S^{n-2}$ we define $$\label{eca2} w(x):=v(x'\cdot\zeta+{\varepsilon},x_n).$$ Also, if $r:=\frac{{\varepsilon}}{2}$ and $x\in B_r\cap\{x_n>0\}$, it follows that $x'\cdot\zeta+{\varepsilon}+\frac{x_n}{\alpha}{\geqslant}-r+{\varepsilon}>0
$, and thus $(x'\cdot\zeta+{\varepsilon},x_n)\in
{\mathcal{C}}_\alpha$. Consequently, by , we have that $$\label{eca4}
{\mbox{$(-\Delta)^\sigma w=0$ in~$B_r\cap\{x_n>0\}$.}}$$ Moreover, using and , we see that $$\label{eca5}
{\mbox{if~$x_n<0$
then~$w(x)=0$. }}$$ [F]{}rom and (see e.g. Section 1.1 in [@MR3694738], or [@MR3293447; @MR3276603]), it also follows that the function ${\rm dist}_{\{x_n<0\}}^{-\sigma} w$ belongs to $C^\infty(B_r\cap\overline{ \{x_n>0\} })$, hence we can define $$\label{2389IKA}
\psi(x'):=\lim_{x_n\searrow0} \frac{w(x',x_n)}{{\rm dist}_{\{x_n<0\}}^\sigma(x',x_n)}
=
\lim_{x_n\searrow0} \frac{w(x',x_n)}{x_n^\sigma}={\mathcal{T}}w(x').$$ This, and give that $
w\in{\mathcal{H}}_k$. As a consequence, recalling , we have that $$\label{PL} \sum_{{\gamma\in{\ensuremath{\mathbb{N}}}^{n-1}}\atop{|\gamma|{\leqslant}k}}
\omega_\gamma\,
D^\gamma\psi(0)=0.$$ In view of and , we also observe that $$\label{8343546}
\psi(x')=
\lim_{x_n\searrow0} \frac{v(x'\cdot\zeta+{\varepsilon},x_n)}{x_n^\sigma}.$$ Accordingly, recalling , $$\label{8343546BIS}
\psi(0)=
\lim_{x_n\searrow0} \frac{v({\varepsilon},x_n)}{x_n^\sigma}=\frac{\Lambda\,v(P)}{P_2^\sigma}=:\tilde\Lambda\ne0.$$ Also, by , we can write that $$\label{7u12jd2}
\psi(x')=\psi_0(x'\cdot\zeta),$$ for some $\psi_0:{\ensuremath{\mathbb{R}}}\to[0,+\infty)$. Hence, recalling the homogeneity in , for all $t>0$, $$\begin{aligned}
&&\psi_0\big(t (x'\cdot\zeta+{\varepsilon})-{\varepsilon}\big)=
\psi\big(t (x'+{\varepsilon}\zeta)-{\varepsilon}\zeta\big)=
\lim_{x_n\searrow0} \frac{v\big((t (x'+{\varepsilon}\zeta)-{\varepsilon}\zeta)\cdot\zeta+{\varepsilon}, x_n\big)}{x_n^\sigma}\\
&&\qquad=
\lim_{x_n\searrow0} \frac{v(
t x'\cdot\zeta+t{\varepsilon}, x_n)}{x_n^\sigma}=
\lim_{y_n\searrow0}
\frac{v\big(
t (x'\cdot\zeta+{\varepsilon}), ty_n\big)}{t^\sigma y_n^\sigma}
=
\lim_{y_n\searrow0} \frac{t^\beta v(
x'\cdot\zeta+{\varepsilon}, y_n)}{t^\sigma y_n^\sigma}\\&&\qquad=
t^{\beta-\sigma}\psi(x')=t^{\beta-\sigma}\psi_0(x'\cdot\zeta)\end{aligned}$$ Taking $m$ derivatives in $t$ of this identity, we conclude that $$\label{evrt}
\psi_0^{(m)}\big(t (x'\cdot\zeta+{\varepsilon})-{\varepsilon}\big)\,
(x'\cdot\zeta+{\varepsilon})^m
=\prod_{i=0}^{m-1}(\beta-\sigma-i)\;t^{\beta-\sigma-m}\,\psi_0(x'\cdot\zeta).$$ In addition, by , we have that $D^\gamma\psi(x')=
\zeta^\gamma\psi_0^{(|\gamma|)}(x'\cdot\zeta)$. Hence, evaluating at $t:=1$ and $x':=0$, if $m=|\gamma|$ we have that $$\label{5379x34} \begin{split}&{\varepsilon}^m D^\gamma\psi(0)=
\zeta^\gamma\psi_0^{(m)}(0)\,{\varepsilon}^m
=\zeta^\gamma \kappa_m\,\psi_0(0)
=\zeta^\gamma \kappa_m\,\psi(0),\\
{\mbox{where }}\qquad&\kappa_m:=\prod_{i=0}^{m-1}(\beta-\sigma-i)
.\end{split}$$ Plugging this information into , and recalling , we find that $$\label{345732}\begin{split}&
0=\frac1{\tilde\Lambda}\sum_{m=1}^k
\sum_{{\gamma\in{\ensuremath{\mathbb{N}}}^{n-1}}\atop{|\gamma|= m}}
\omega_\gamma\,
D^\gamma\psi(0)=\frac1{\tilde\Lambda}
\sum_{m=1}^k
\sum_{{\gamma\in{\ensuremath{\mathbb{N}}}^{n-1}}\atop{|\gamma|= m}}
\kappa_m\,{\varepsilon}^{-m}
\omega_\gamma \zeta^\gamma\psi(0)\\&\qquad\qquad=\frac1{\tilde\Lambda}
\sum_{{\gamma\in{\ensuremath{\mathbb{N}}}^{n-1}}\atop{|\gamma|{\leqslant}k}}
\kappa_{|\gamma|}\,
\omega_\gamma \left(\frac{\zeta}{{\varepsilon}}\right)^\gamma\psi(0)=
\sum_{{\gamma\in{\ensuremath{\mathbb{N}}}^{n-1}}\atop{|\gamma|{\leqslant}k}}
\kappa_{|\gamma|}\,
\omega_\gamma \left(\frac{\zeta}{{\varepsilon}}\right)^\gamma
.\end{split}$$ Since $\frac{\zeta}{\varepsilon}$ ranges in an open set of ${\ensuremath{\mathbb{R}}}^{n-1}$, we deduce from and the Identity Principle for polynomials that $\kappa_{|\gamma|}\,
\omega_\gamma=0$ for all $\gamma\in{\ensuremath{\mathbb{N}}}^{n-1}$ with $|\gamma|{\leqslant}k$. As a consequence, by , we find that $
\omega_\gamma=0$ for all $\gamma\in{\ensuremath{\mathbb{N}}}^{n-1}$ with $|\gamma|{\leqslant}k$, hence $\omega=0$. This is in contradiction with and thus we have proved the desired result.
With this, we are in the position of completing the proof of Theorem \[FLFL\] by arguing as follows:
\[89-ppA\] Since the claims in Theorem \[FLFL\] have a linear structure in $f$, $f_{\varepsilon}$ and $u_{\varepsilon}$, by the Stone-Weierstra[ß]{} Theorem, it is enough to prove Theorem \[FLFL\] if $f$ is a monomial. Hence, we fix ${\varepsilon}\in(0,1)$, possibly to be taken conveniently small, and we suppose that $$\label{mu1}
f(x')=\frac{(x')^\mu}{\mu!}\qquad{\mbox{for some }}\mu\in{\ensuremath{\mathbb{N}}}^{n-1}.$$ Then, we apply Lemma \[fan\], finding a suitable function $u_\star
\in{\mathcal{H}}_k$ with $\phi_\star:={\mathcal{T}}u_\star$ that satisfies $$\label{mu2} D^\gamma \phi_\star(0)=\begin{cases}
1 & {\mbox{ if }} \gamma=\mu,\\
0 & {\mbox{ if $|\gamma|{\leqslant}|\mu|+k$ and~$ \gamma\ne\mu$.}}
\end{cases}$$ We define $$\begin{aligned}
&&
u_{\varepsilon}(x):= {\varepsilon}^{-\sigma-|\mu|} \,u_\star({\varepsilon}x)\\
{\mbox{and }}&& f_{\varepsilon}(x'):={\varepsilon}^{-|\mu|} \,\phi_\star({\varepsilon}x').\end{aligned}$$ Then, if $x\in B_1\cap\{x_n>0\}$, we have that $(-\Delta)^\sigma u_{\varepsilon}(x)=
{\varepsilon}^{\sigma-|\mu|} \,(-\Delta)^\sigma u_\star({\varepsilon}x)=0$ as long as ${\varepsilon}$ is sufficiently small. In addition, we see that $u_{\varepsilon}<0$ in $\{x_n<0\}$, and that $$\lim_{x_n\searrow0} \frac{u_{\varepsilon}(x',x_n)}{x_n^\sigma}=
\lim_{x_n\searrow0} \frac{{\varepsilon}^{-\sigma-|\mu|} \,u_\star({\varepsilon}x',{\varepsilon}x_n)}{x_n^\sigma}=
\lim_{\tau\searrow0} \frac{{\varepsilon}^{-|\mu|} \,u_\star({\varepsilon}x',\tau)}{\tau^\sigma}={\varepsilon}^{-|\mu|} \,\phi_\star({\varepsilon}x')=
f_{\varepsilon}(x').$$ These observations prove and .
Furthermore, $$D^\gamma f_{\varepsilon}(x')={\varepsilon}^{|\gamma|-|\mu|} \,D^\gamma \phi_\star({\varepsilon}x').$$ Consequently, if we set $g_{\varepsilon}(x'):=f_{\varepsilon}(x')-f(x')$, we deduce from and that $$D^\gamma g_{\varepsilon}(0)=0,$$ for all $\gamma\in{\ensuremath{\mathbb{N}}}^{n-1}$ such that $|\gamma|{\leqslant}|\mu|+k$.
This observation and a Taylor expansion give that, for all $x'\in B_1'$ and all $\zeta
\in{\ensuremath{\mathbb{N}}}^{n-1}$ such that $|\zeta|{\leqslant}k$, $$\begin{aligned}
&&
| D^\zeta f_{\varepsilon}(x') - D^\zeta f(x')|=
|D^\zeta g_{\varepsilon}(x')|{\leqslant}C_k\,\sup_{y'\in B_1'}
\sum_{{\alpha\in{\ensuremath{\mathbb{N}}}^n}\atop{|\alpha| =|\mu|+k+1}}
|D^\alpha g_{\varepsilon}(y')|\\&&\qquad=
C_k\,\sup_{y'\in B_1'}
\sum_{{\alpha\in{\ensuremath{\mathbb{N}}}^n}\atop{|\alpha| =|\mu|+k+1}}
|D^\alpha f_{\varepsilon}(y')|=
C_k\,\sup_{y'\in B_1'}
\sum_{{\alpha\in{\ensuremath{\mathbb{N}}}^n}\atop{|\alpha| =|\mu|+k+1}}
{\varepsilon}^{|\alpha|-|\mu|} \,
|D^\alpha \phi_\star({\varepsilon}x')|\\&&\qquad
= C_k\,{\varepsilon}^{k+1} \,\sup_{y'\in B_1'}
\sum_{{\alpha\in{\ensuremath{\mathbb{N}}}^n}\atop{|\alpha| =|\mu|+k+1}}
|D^\alpha \phi_\star({\varepsilon}x')|{\leqslant}C_k'\,{\varepsilon},\end{aligned}$$ for some $C_k$, $C_k'>0$. This establishes , up to renaming ${\varepsilon}$.
Proof of Theorem \[QC\] {#15kdttd3}
=======================
In this section, for the sake of generality, some results are proved in arbitrary dimension $n{\geqslant}2$, whenever the proof would not experience significant simplifications in the case $n=2$ (then, for the proof of Theorem \[QC\], we restrict ourselves to the case $n=2$, see also Open Problem \[17\]). As customary, given $E\subset{\ensuremath{\mathbb{R}}}^{n+1}$, it is convenient to consider the nonlocal mean curvature at a point $x\in\partial E$, defined by $$\label{jfgnjbj96768769} {\mathcal{H}}_E^s(x):=\int_{{\ensuremath{\mathbb{R}}}^{n+1}}
\frac{\chi_{{\ensuremath{\mathbb{R}}}^{n+1}\setminus E}(y)-\chi_E(y)}{|x-y|^{n+1+s}}\,dy.$$ The first step to prove Theorem \[QC\] is to establish the existence of a small vertical cone not intersecting the boundary of a homogeneous nonlocal minimal surface on a hyperplane with null exterior datum. Letting $e_{n+1}:=(0,\dots,0,1)\in{\ensuremath{\mathbb{R}}}^{n+1}$, the precise result that we have is the following one:
\[3243-237gndlsm\] Let $E\subset{\ensuremath{\mathbb{R}}}^{n+1}$ be a locally $s$-minimal set in ${\ensuremath{\mathbb{R}}}^n\times(0,+\infty)$. Assume that $$\label{ADX-EE}
E\cap\{x_n<0\}=\{x_{n+1}<0\}\cap\{x_n<0\}$$ and that $$\label{ADX6-EE}
{\mbox{$tE=E$ for every~$t>0$.}}$$ Then, $$\label{6BOu-1BIS}
\min\Big\{ {\rm dist}\big(e_{n+1}, (\partial E)\cap\{x_n>0\}\big),\;
{\rm dist}\big(-e_{n+1},(\partial E)\cap\{x_n>0\}\big)\Big\}>0.$$
We claim that $$\label{6BOu-1} e_{n+1}\not\in
\overline{(\partial E)\cap\{x_n>0\}}.$$ Indeed, suppose by contradiction that $e_{n+1}\in\overline{(
\partial E)\cap\{x_n>0\}}$, and thus, by , also $te_{n+1}\in\overline{(\partial E)
\cap\{x_n>0\}}$, for all $t>0$. By , we have that $B_{1/2}(e_{n+1})\cap\{x_n<0\}\subset {\ensuremath{\mathbb{R}}}^{n+1}\setminus E$. Hence (see e.g. Theorem B.9 in [@MR3926519]), we have that ${\mathcal{H}}_{E}^s(te_{n+1})=0$ for all $t\in\left[\frac{9}{10},\frac{11}{10}\right]$. In particular, if $F:=E-\frac{e_{n+1}}{100}$, we see that $F\subset E$, and thus $$\begin{aligned}
&& 0={\mathcal{H}}_{E}^s(e_{n+1})
-{\mathcal{H}}_{F}^s(e_{n+1})
=-2\int_{{\ensuremath{\mathbb{R}}}^{n+1}}
\frac{\chi_{E\setminus F}(y)}{|e_{n+1}-y|^{n+1+s}}\,dy\\&&\qquad{\leqslant}-2
\int_{{\ensuremath{\mathbb{R}}}^{n-1}\times(-\infty,0)\times\left(-\frac1{100},0\right)}
\frac{dy}{|e_{n+1}-y|^{n+1+s}}<0.\end{aligned}$$ This contradiction proves .
Similarly, one proves that $-e_{n+1}\not\in
\overline{(\partial E)\cap\{x_n>0\}}$. Consequently, since $\partial E$ is a closed set, we obtain , as desired.
As a byproduct of Lemma \[3243-237gndlsm\], we obtain that the second blow-up of an $s$-minimal graph which is flat from one side is necessarily a graph as well (see e.g. Lemmata 2.2 and 2.3 in [@2019arXiv190405393D] for the basic properties of the second blow-up). The precise result goes as follows:
\[BOAmdfiUAMP\] Let $u$ be an $s$-minimal graph in $(-2,2)^{n-1}\times(0,4)$. Assume that there exists $h>0$ such that $$\label{568-029-3984112}
{\mbox{$u=0$ in~$ (-2,2)^{n-1}\times(-h,0)$.}}$$ Let $E_{00}$ be the second blow-up of $E_u$, being $E_u$ defined in . Then: $$\begin{aligned}
\label{POS-E001}
&& {\mbox{either $E_{00}\cap\{x_n>0\}=\varnothing$,}}
\\ \label{POS-E002}&& {\mbox{or $E_{00}\cap\{x_n>0\}=\{x_n>0\}$,}}\\&&
{\mbox{or $E_{00}$
has a graphical structure, namely}}\nonumber\\&&\label{POS-E003}
{\mbox{there exists~$u_{00}:{\ensuremath{\mathbb{R}}}^n\to{\ensuremath{\mathbb{R}}}$ such that~$E_{00}=E_{u_{00}}$}.}\end{aligned}$$ Furthermore, if holds true, then $$\label{POS-E004}
\lim_{x\to0}u_{00}(x)=0.$$
We suppose that $$\label{454562382023}
{\mbox{\eqref{POS-E001} and~\eqref{POS-E002} do not hold,}}$$ and we aim at showing that and are satisfied. We start by proving . For this, we first observe that $E_{00}$ has a generalized “hypographical” structure, that is $$\label{657865-349}
{\mbox{if~$(y,y_{n+1})\in({\ensuremath{\mathbb{R}}}^n\times{\ensuremath{\mathbb{R}}})\cap E_{00}$,
then~$(y,\tau)\in E_{00}$ for every~$\tau{\leqslant}y_{n+1}$.}}$$ Indeed, each rescaling of $E_u$ has such property, and since these rescalings approach $E_{00}$ in the Hausdorff distance (see [@2019arXiv190405393D]), the claim in follows.
Moreover, by , $$\label{ADX-EE-568-029-3984112}
E_{00}\cap\{x_n<0\}=\{x_{n+1}<0\}\cap\{x_n<0\},$$ and, by Lemma 2.2 in [@2019arXiv190405393D], we have that $$\label{ADX-EE-568-029-3984112-NIA}
{\mbox{$tE_{00}=E_{00}$ for every~$t>0$.
}}$$ [F]{}rom this and , we are in the position of using Lemma \[3243-237gndlsm\], and thus deduce from that $$\label{6BOu-1BIS-TRIS}
\min\Big\{ {\rm dist}\big(e_{n+1},(\partial E_{00})\cap\{x_n>0\}\big),\;
{\rm dist}\big(-e_{n+1},(\partial E_{00})\cap\{x_n>0\}\big)\Big\}>0.$$ Then, for all $x\in{\ensuremath{\mathbb{R}}}^{n-1}\times(0,+\infty)$, we set $$u_{00}(x):=\sup\big\{
y {\mbox{ s.t. }}(x,y)\in E_{00}
\big\}.$$ By extending $u_{00}$ to vanish in ${\ensuremath{\mathbb{R}}}^{n-1}\times(-\infty,0)$, we find that $E_{00}$ is the subgraph of $u_{00}$, as desired. To this aim, it remains to prove that the image of $u_{00}$ is ${\ensuremath{\mathbb{R}}}$, namely that for every $x\in{\ensuremath{\mathbb{R}}}^{n-1}\times(0,+\infty)$, $$\label{71-02-23884oe}
u_{00}(x)\not\in\{-\infty,+\infty\}.$$ As a matter of fact, in light of , it is sufficient to prove for every $x\in S^{n-1}_+:=\{x=(x_1,\dots,x_n)\in{\ensuremath{\mathbb{R}}}^n
{\mbox{ s.t. $|x|=1$ and~$x_n>0$}}\}$. Hence, we set $$\begin{aligned}
&&
\omega :=\big\{ x\in S^{n-1}_+{\mbox{ s.t. }}u_{00}(x)\not\in\{-\infty,+\infty\}\big\},\\
&&
\omega_+ :=\big\{ x\in S^{n-1}_+{\mbox{ s.t. }}u_{00}(x)=+\infty\big\}\\
{\mbox{and }}&&
\omega_-:=\big\{ x\in S^{n-1}_+{\mbox{ s.t. }}u_{00}(x)=-\infty\big\},\end{aligned}$$ and, to prove , we want to show that $\omega_+=\varnothing=\omega_-$.
For a contradiction, assume that $\omega_-\ne\varnothing$. By , we also know that $\omega_-\ne S^{n-1}_+$. Hence, we can take $x_-\in\omega_-$ and $x_+\in\omega\cup\omega_+$. By construction, we have that $u_{00}(x_-)=-\infty$ and $u_{00}(x_+)\in{\ensuremath{\mathbb{R}}}\cup\{+\infty\}$, and therefore $(x_-,y)\in
{\ensuremath{\mathbb{R}}}^{n+1}\setminus E_{00}$ for all $y\in{\ensuremath{\mathbb{R}}}$, while $(x_+,y_+)\in E_{00}$ for some $y_+\in{\ensuremath{\mathbb{R}}}$.
This and give that $(x_+,y)\in E_{00}$ for all $y{\leqslant}y_+$. In particular, for all $k\in{\ensuremath{\mathbb{N}}}$ sufficiently large, we have that $(x_+,-k)\in E_{00}$, with $(x_-,-k)\in {\ensuremath{\mathbb{R}}}^{n+1}\setminus E_{00}$. Consequently, by , we see that $
\left(\frac{x_+}k,-1\right)\in E_{00}$ and $\left(\frac{x_-}k,-1\right)\in {\ensuremath{\mathbb{R}}}^{n+1}\setminus E_{00}$. As a result, by taking the limit as $k\to+\infty$, we conclude that $-e_{n+1}\in
\overline{(\partial E_{00})\cap\{x_n>0\}}$. But this is in contradiction with , and therefore necessarily $\omega_-=\varnothing$.
Similarly, one proves that $\omega_+=\varnothing$, and this completes the proof of .
The proof of is thus completed, and we now focus on the proof of . For this, assume the converse: then, there exists a sequence $x^{(k)}=(x^{(k)}_1,\dots,x^{(k)}_n)$ such that $x^{(k)}_n>0$, $x^{(k)}\to0$ as $k\to+\infty$ and $|u(x^{(k)})|{\geqslant}a_0$, for some $a_0>0$. Up to a sign change, we can suppose that $
u(x^{(k)}){\geqslant}a_0$. Hence, we have that $( x^{(k)}, u(x^{(k)}))\in
(\partial E_{00})\cap\{x_n>0\}$ and then, recalling , we deduce that $
\left(\frac{ x^{(k)}}{ u(x^{(k)})},1\right)\in
(\partial E_{00})\cap\{x_n>0\}$.
Accordingly, taking the limit as $k\to+\infty$, we find that $e_{n+1}\in\overline{
(\partial E_{00})\cap\{x_n>0\} }$. This is in contradiction with and, as a result, the proof of is complete.
[F]{}rom Lemma \[3243-237gndlsm\] we also deduce a regularity result of the following type:
\[lemma33\] Let $u$ be an $s$-minimal graph in ${\ensuremath{\mathbb{R}}}^{n-1}\times(0,+\infty)$. Assume that $$\label{pppADX}
{\mbox{$u(x_1,\dots,x_n)=0$ if~$x_n<0$.}}$$ Assume also that $u$ is positively homogeneous of degree $1$, i.e. $$\label{pppADX6}
u(tx)=tu(x)\qquad{\mbox{ for all $x\in{\ensuremath{\mathbb{R}}}^n$ and~$t>0$.}}$$ Then, we have that $$\label{ADX0}
u\in
L^\infty_{\rm loc}({\ensuremath{\mathbb{R}}}^n)\cap C([-1,1]^{n-1}\times[0,1])\cap
C^\infty( {\ensuremath{\mathbb{R}}}^{n-1}\times(0,+\infty)).$$
We claim that $$\label{6BOu} u\in
L^\infty_{\rm loc}({\ensuremath{\mathbb{R}}}^n).$$ Suppose not. Then there exist $R>0$ and $x^{(k)}=(x^{(k)}_1,\dots,x^{(k)}_n)\in B_R$ such that $|u(x^{(k)})|{\geqslant}k$. Without loss of generality, we can suppose that $u(x^{(k)}){\geqslant}k$. Then, by , we see that $x^{(k)}_n>0$, and, recalling , we have that $\big( t x^{(k)},t u(x^{(k)})\big)\in
\partial E_u$ for every $t>0$ (being $E_u$ defined in ), and then, in particular, $$\left( \frac{x^{(k)}}{u(x^{(k)})},1\right)\in\partial E_u.$$ Accordingly, taking the limit as $k\to+\infty$, we find that $e_{n+1}\in\overline{\partial E_u\cap\{x_n>0\}}$. This is in contradiction with , whence the proof of is complete.
As a result, by Theorem 1.1 in [@MR3934589] we obtain that $u$ is smooth in $\{x_n>0\}$, and the continuity up to $\{x_n=0\}$ follows from Theorem 1.1 of [@MR3516886]. This proves the claim in .
Now, we show that if the second blow-up is either empty or full in a halfspace, then the original $s$-minimal graph is necessarily boundary discontinuous:
\[554327h:AJSJD2374859ty\] Let $u$ be an $s$-minimal graph in $(-2,2)\times(0,4)$. Assume that there exists $h>0$ such that $$\label{Rg68231opp}
{\mbox{$u=0$ in~$ (-2,2)\times(-h,0)$.}}$$ Let $E_{00}$ be the second blow-up of $E_u$. Then, $$\begin{aligned}
\label{FAG:1}
&& {\mbox{if~$E_{00}={\ensuremath{\mathbb{R}}}\times(-\infty,0)\times(-\infty,0)$, then }}
\lim_{x_2\searrow0} u(0,x_2)<0;
\\
\label{FAG:2}\nonumber
&& {\mbox{if~$E_{00}=\big({\ensuremath{\mathbb{R}}}\times(-\infty,0)\times(-\infty,0)\big)
\cup \big({\ensuremath{\mathbb{R}}}\times(0,+\infty)\times{\ensuremath{\mathbb{R}}}\big)$,}}\\
&&{\mbox{then }}\lim_{x_2\searrow0} u(0,x_2)>0.\end{aligned}$$
We focus on the proof of , since the proof of is similar. We recall (see [@2019arXiv190405393D]) that, as $k\to+\infty$, $$\label{INCLA0-QUI}
\begin{split}&{\mbox{up to a subsequence, $\chi_{k E_u}$ converges to~$\chi_{{E_{00}}}$
in~$L^1_{\rm loc}({\ensuremath{\mathbb{R}}}^3)$,}}\\
&{\mbox{and $kE_u$ converges to~$E_{00}$
locally in the Hausdorff distance.}}\end{split}$$ We claim that for every $M>0$ there exists $k_M\in{\ensuremath{\mathbb{N}}}$ such that if $k{\geqslant}k_M$ then $$\label{INCLA1}(
kE_u)\cap B_M\subseteq\left\{x_2<\frac1M\right\}.$$ To check this, we argue for a contradiction and suppose that, for some $M>1$, there are infinitely many $k$’s for which there exists $p^{(k)}=(p^{(k)}_1,p^{(k)}_2,p^{(k)}_3)\in
(kE_u)\cap B_M$ with $p^{(k)}_2{\geqslant}\frac1M$. We observe that $B_{1/(2M)}(p^{(k)})$ cannot be contained in $kE_u$, otherwise, recalling the structure of $E_{00}$ in , $$\begin{aligned}
&& \int_{ B_{1/(2M)}(p^{(k)}) }|\chi_{k E_u}(x)-\chi_{{{E_{00}}}}(x)|\,dx=
\int_{ B_{1/(2M)}(p^{(k)}) }|\chi_{k E_u}(x)|\,dx\\&&\qquad \qquad = |B_{1/(2M)}(p^{(k)})|=
|B_{1/(2M)}|,\end{aligned}$$ which is in contradiction with .
As a result, there exists $q^{(k)}\in B_{1/(2M)}( p^{(k)} )$ with $q^{(k)}\in\partial (k E_u)$. In particular, we have that $q^{(k)}_2{\geqslant}p^{(k)}_2-\frac1{2M}{\geqslant}\frac1{2M}$, whence, using the clean ball condition in [@MR2675483], there exist $c\in(0,1)$, $r_0\in\left(0,\frac1{4M}\right)$ and $\tilde q^{(k)}\in B_{r_0}( q^{(k)})$ such that $B_{cr_0}( \tilde q^{(k)})\subseteq
(kE_u)\cap B_{r_0}( q^{(k)})$.
We remark that if $x\in B_{cr_0}( \tilde q^{(k)})$ then $$x_2{\geqslant}\tilde q^{(k)}_2-cr_0{\geqslant}q^{(k)}_2 -| q^{(k)}-\tilde q^{(k)}|-r_0
{\geqslant}\frac1{2M}-2r_0>0.$$ Consequently, recalling the structure of $E_{00}$ in , $$\int_{ B_{cr_0}( \tilde q^{(k)}) }
|\chi_{k E_v}(x)-\chi_{{{F}}}(x)|\,dx=
\big| B_{cr_0}( \tilde q^{(k)}) \big|=| B_{cr_0}|.$$ This is in contradiction with and so it proves .
Now, for all $\lambda$, $\tau\in(-2,2)$ and $t>1$ consider the ball $B_1(\lambda,t,\tau)$. By , there exists $k_0\in{\ensuremath{\mathbb{N}}}$ such that if $t{\geqslant}2$ we have that $$\label{980-029-1}
B_1(\lambda,t,\tau)\subseteq {\ensuremath{\mathbb{R}}}^3\setminus (kE_u),$$ for all $\lambda$, $\tau\in(-2,2)$ and $k{\geqslant}k_0$.
Now we claim that the claim in holds true for all $t>1$ (and not just $t{\geqslant}2$) with respect to the same $k_0$: namely, we show that for all $\lambda$, $\tau\in(-2,2)$, $k{\geqslant}k_0$ and $t>1$, we have that $$\label{980-029-2}
B_1(\lambda,t,\tau)\cap\big(\partial (kE_u)\big)=\varnothing.$$ Indeed, if not, there would exist $\lambda$, $\tau\in(-2,2)$ and $k{\geqslant}k_0$, and a suitable $t_\star>1$, for which $B_1(\lambda,t_\star,\tau)\cap\big(\partial (kE_u)\big)\not=\varnothing$. More precisely, if were false, by , we can slide $B_1(\lambda,\cdot,\tau)$ with respect to the parameter $t$ from the right till it touches $\partial (kE_u)$, say at a point $Z=(Z_1,Z_2,Z_3)$. In this way, find that $t_\star\in(1,2)$ and $B_1(\lambda,t,\tau)\subseteq {\ensuremath{\mathbb{R}}}^3\setminus (kE_u)$ for all $t>t_\star$, with $Z\in(\partial (kE_u))\cap B_1(\lambda,t_\star,\tau)$.
Now, we show that, for any fixed $M>1$, $$\label{Diar34t5y6be9er}
(kE_u) \cap\{ x_2-Z_2>2\}\subseteq {\ensuremath{\mathbb{R}}}^3\setminus B_{M/4}(Z),$$ as long as $k$ is sufficiently large. Indeed, if not, take $P=(P_1,P_2,P_3)\in (kE_u)\cap B_{M/4}(Z)$ with $P_2>Z_2+2$. By construction, $$|P_1|{\leqslant}|P_1-Z_1|+|Z_1-\lambda |+|\lambda |{\leqslant}\frac{M}{4}+1+2<\frac{M}{3},$$ as long as $M$ is sufficiently large, and similarly $|P_2|<\frac{M}{3}$ and $|P_3|<\frac{M}{3}$. As a consequence, we have that $P\in B_M$. This and give that $P_2<\frac1M$, and then $$\frac1M>P_2>Z_2+2{\geqslant}t_\star -|t_\star-Z_2|+2{\geqslant}t_\star-1+2{\geqslant}2.$$ This is a contradiction and therefore the proof of is complete.
Furthermore, when $k>2/h$, we observe that $$\label{Diar34t5y6be9er-BIS}\begin{split}&
(kE_u) \cap\{ x_2-Z_2\in(-hk,-2)\}\subseteq\left\{
x_3<ku\left( \frac{x_1}k,\frac{x_2}k\right)\right\}\cap\{ x_2\in(-hk,0)\}\\&\qquad\qquad
\subseteq \{ x_3<0\},\end{split}$$ thanks to .
Then, as a consequence of and , $$\int_{ B_{M/4}(Z)\cap \{2< |x_2-Z_2|<hk\}}\frac{\chi_{{\ensuremath{\mathbb{R}}}^3\setminus(kE_u)}(y)
-\chi_{ kE_u }(y)}{|Z-y|^{3+s}}\,dy{\geqslant}c,$$ for some universal constant $c>0$ depending only on $s$.
On this account, if $M$ and $k$ are sufficiently large, we find that $$\label{Nysjdf33-1}
\int_{ \{ |x_2-Z_2|>2\}}\frac{\chi_{{\ensuremath{\mathbb{R}}}^3\setminus(kE_u)}(y)
-\chi_{ kE_u }(y)}{|Z-y|^{3+s}}\,dy{\geqslant}\frac{c}{2}.$$ Moreover, for a sufficiently small $\rho>0$, using the fact that $
B_1(\lambda,t_\star,\tau)\subseteq {\ensuremath{\mathbb{R}}}^3\setminus (kE_u)$, we see that $$\label{Nysjdf33-2}
\int_{ B_\rho(Z)}\frac{\chi_{{\ensuremath{\mathbb{R}}}^3\setminus(kE_u)}(y)
-\chi_{ kE_u }(y)}{|Z-y|^{3+s}}\,dy{\geqslant}-C\rho^{1-s}{\geqslant}-\frac{c}{4},$$ see e.g. Lemma 3.1 in [@MR3516886] for computational details.
Fixing such a $\rho$ from now on, we deduce from and that $$\label{Nysjdf33-3}
\int_{ B_\rho(Z)\cup \{ |x_2-Z_2|>2\}}\frac{\chi_{{\ensuremath{\mathbb{R}}}^3\setminus(kE_u)}(y)
-\chi_{ kE_u }(y)}{|Z-y|^{3+s}}\,dy{\geqslant}\frac{c}{4}.$$ Also, by and , $$\int_{\{ |x_2-Z_2|{\leqslant}2\}\setminus B_\rho(Z)}\frac{\chi_{{\ensuremath{\mathbb{R}}}^3\setminus(kE_u)}(y)
-\chi_{ kE_u }(y)}{|Z-y|^{3+s}}\,dy{\geqslant}- \frac{c}{8},$$ as long as $M$ is sufficiently large. This and yield that $$\int_{{\ensuremath{\mathbb{R}}}^3}\frac{\chi_{{\ensuremath{\mathbb{R}}}^3\setminus(kE_u)}(y)
-\chi_{ kE_u }(y)}{|Z-y|^{3+s}}\,dy{\geqslant}\frac{c}{8}>0,$$ which is a contradiction with the minimality of $kE_u$. This completes the proof of .
Consequently, by , $$(-2,2)\times(0,2)\times(-2,2)\subseteq {\ensuremath{\mathbb{R}}}^3\setminus (k_0 E_u).$$ Therefore $$\left(-\frac2{k_0},\frac2{k_0}\right)\times
\left(0,\frac2{k_0}\right)
\times\left(-\frac2{k_0},\frac2{k_0}\right)\subseteq {\ensuremath{\mathbb{R}}}^3\setminus E_u,$$ and thus, for all $y=(y_1,y_2,y_3)\in\left(-\frac2{k_0},\frac2{k_0}\right)\times
\left(0,\frac2{k_0}\right)
\times\left(-\frac2{k_0},\frac2{k_0}\right)$, we have that $y_3{\geqslant}u(y_1,y_2)$. In particular, choosing $y_1:=0$ and $y_3:=-\frac1{k_0}$, we find that, for all $y_2\in\left(0,\frac2{k_0}\right)$, $$u(0,y_2){\leqslant}-\frac1{k_0}.$$ Then, we can send $y_2\searrow0$ and obtain $$\limsup_{y_2\searrow0} u(0,y_2)<0.$$ This limit in fact exists, thanks to Theorem 1.1 of [@MR3516886], whence follows, as desired.
Next result discusses how the boundary continuity of a nonlocal minimal graph at some boundary points implies the differentiability up to the boundary:
\[STEP1\] Let the assumptions of Theorem \[QC\] hold. Assume also that $$\label{6tsoe}
\lim_{x\to (1,0)} u(x)=0.$$ Then $u\in C^1({\ensuremath{\mathbb{R}}}^2\setminus\ell_-)$, where $\ell_-:=(-\infty,0]\times \{0\}$.
Similarly, if $$\label{6tsoe-2}
\lim_{x\to (-1,0)} u(x)=0.$$ Then $u\in C^1({\ensuremath{\mathbb{R}}}^2\setminus\ell_+)$, where $\ell_+:=[0,+\infty)\times \{0\}$.
If both and are satisfied, then $u\in C^1({\ensuremath{\mathbb{R}}}^2)$.
We suppose that is satisfied, since the case in which holds true is similar (and so is the casein which both and are fulfilled). Using and , for any $\tau>0$, $$\label{8hsfr}u(\tau,0):=
\lim_{x\to (\tau,0)} u(x)
=0.$$ We define $$\label{sucjp}
v(x_1,x_2):=u(x_1+1,x_2).$$ We consider the homogeneous second blow up ${F}$ of $E_v$ (see Lemmata 2.2 and 2.3 in [@2019arXiv190405393D]), and we have that, as $k\to+\infty$, $$\label{INCLA0}
\begin{split}&{\mbox{up to a subsequence, $\chi_{k E_v}$ converges to~$\chi_{{F}}$
in~$L^1_{\rm loc}({\ensuremath{\mathbb{R}}}^3)$,}}\\
&{\mbox{and $kE_v$ converges to~$F$
locally in the Hausdorff distance.}}\end{split}$$ By , $$\label{Gina}
{{F}}\cap\{x_2<0\}={\ensuremath{\mathbb{R}}}\times(-\infty,0)\times(-\infty,0).$$ In addition, we claim that $$\label{DUIA} {{F}}+(r,0,0)={{F}},$$ for all $r\in{\ensuremath{\mathbb{R}}}$. Indeed, we set $v_k(x):=kv\left(\frac{x}k\right)$, and, by , we see that, for every $x\in{\ensuremath{\mathbb{R}}}^2$, $$\begin{aligned}
&&v_k\left(\frac{k+1}{k}x_1+1,\frac{k+1}{k}x_2
\right)=
k v\left( \frac{k+1}{k^2}x_1+\frac1k,\frac{k+1}{k^2}x_2\right)\\&&\quad=
k u\left( \frac{k+1}{k^2}x_1+\frac1k+1,\frac{k+1}{k^2}x_2\right)=
k u\left( \frac{k+1}{k^2}x_1+\frac{k+1}k,\frac{k+1}{k^2}x_2\right)\\&&\quad=
(k+1)\,
u\left( \frac{x_1}k+1,\frac{x_2}k\right)=(k+1)\,
v\left( \frac{x_1}k,\frac{x_2}k\right)=\frac{k+1}{k}\,v_k(x).\end{aligned}$$ For this reason, taking the limit as $k\to+\infty$ and recalling , we see that $ {{F}}+(1,0,0)={{F}}$. Since $F$ is a cone (see Lemma 2.2 in [@2019arXiv190405393D]), this completes the proof of .
Accordingly, by and the dimensional reduction (see [@MR2675483]), we can write that ${{F}}={\ensuremath{\mathbb{R}}}\times G$, for some cone $G\subseteq{\ensuremath{\mathbb{R}}}^2$ which is locally $s$-minimal in $(0,+\infty)\times{\ensuremath{\mathbb{R}}}$. In view of we have that $G\cap ((-\infty,0)\times{\ensuremath{\mathbb{R}}})=
(-\infty,0)\times(-\infty,0)$. Therefore, by minimality, $$\label{nosn-21}\begin{split}
&{\mbox{either~$G={\ensuremath{\mathbb{R}}}\times(-\infty,0)$,
or~$G=(-\infty,0)\times(-\infty,0)$,}}\\&
{\mbox{or~$G=\big( (-\infty,0)\times(-\infty,0)\big)\cup \big(
(0,+\infty)\times{\ensuremath{\mathbb{R}}}\big)$.}}\end{split}$$ We claim that $$\label{nosn-22}
G={\ensuremath{\mathbb{R}}}\times(-\infty,0).$$ The proof of is by contradiction. Suppose not, then, by , we can suppose that $
G=(-\infty,0)\times(-\infty,0)$ (the case $G=\big( (-\infty,0)\times(-\infty,0)\big)\cup \big(
(0,+\infty)\times{\ensuremath{\mathbb{R}}}\big)$ being similar), and therefore $${{F}}={\ensuremath{\mathbb{R}}}\times(-\infty,0)\times(-\infty,0).$$ This says that we can exploit Lemma \[554327h:AJSJD2374859ty\] here (with $u$ replaced by $v$), and then deduce from that $$\lim_{x_2\searrow0} u(1,x_2)=\lim_{x_2\searrow0} v(0,x_2)<0.$$ This is in contradiction with , whence the proof of is complete.
[F]{}rom , it follows that $${{F}}=\{x_3<0\}.$$ [F]{}rom this and the density estimates in [@MR2675483], we have that, up to a subsequence, for every $\delta>0$ there exists $k(\delta)\in{\ensuremath{\mathbb{N}}}$ such that, if $k{\geqslant}k(\delta)$, $$\label{4568rgu}
\partial (kE_v)\cap\left( [-1,1]\times\left[ \frac14,2\right]\times{\ensuremath{\mathbb{R}}}\right)\subseteq\{|x_3|{\leqslant}\delta\}.$$ Now, to complete the proof of the desired result, we need to show that $u\in C^1({\ensuremath{\mathbb{R}}}^2\setminus\ell_-)$. By we know that $u\in C^1({\ensuremath{\mathbb{R}}}\times(-\infty,0))$, with $\nabla u=0$ in ${\ensuremath{\mathbb{R}}}\times(-\infty,0)$. Moreover, by [@MR3934589], we know that $u\in C^1({\ensuremath{\mathbb{R}}}\times(0,+\infty))$. Hence, to complete the proof of the desired result, it is enough to show that, for all $q>0$, $$\label{3rt7uJS:SP}
\lim_{ {x\to(q,0)}\atop{x_2>0}}|\nabla u(x)|=0.$$ We observe that is proved once we demonstrate that $$\label{3rt7uJS:SP2}
\lim_{ {t\searrow0}}|\nabla u( 1,t)|=0.$$ Indeed, if holds true, using we have that $$\lim_{ {x\to(q,0)}\atop{x_2>0}}|\nabla u(x)|=
\lim_{ {x\to(q,0)}\atop{x_2>0}}\left|
\nabla u\left(1,\frac{x_2}{x_1}\right)\right|=
\lim_{ {t\searrow0}}|\nabla u(1,t)|=0,$$ which gives in this case.
In view of these considerations, we focus on the proof of . For this, we exploit the notation in , and we aim at showing that $$\lim_{ t\searrow0}|\nabla v(0,t)|=0,$$ or, equivalently, that for all ${\varepsilon}>0$ there exists $k_{\varepsilon}\in{\ensuremath{\mathbb{N}}}$ such that if $k{\geqslant}k_{\varepsilon}$ and $t\in
\left( 0,\frac1k\right)$, we have that $|\nabla v(0,t)|{\leqslant}{\varepsilon}$.
For this, it is sufficient to show that if $k{\geqslant}k_{\varepsilon}$ and $t\in
\left[ \frac{1}{k+1},\frac1k\right)$ then $$\label{COnsit949}
|\nabla v(0,t)|{\leqslant}{\varepsilon}.$$ To this end, we suppose, by contradiction, that there exists $a>0$ such that for every $K\in{\ensuremath{\mathbb{N}}}$ with $K{\geqslant}1$ there exist $k\in{\ensuremath{\mathbb{N}}}$ with $k{\geqslant}K$ and $t^{(k)}\in
\left[ \frac{1}{k+1},\frac1k\right)$ such that $$\label{P9iqdwf}|\nabla v(0,t^{(k)})|{\geqslant}a.$$ We let $$\label{7SH37eud} T^{(k)}:=kt^{(k)}\in
\left[ \frac{k}{k+1},1\right)\subseteq
\left[ \frac{1}{2},1\right].$$ By and the improvement of flatness result in [@MR2675483], choosing $\delta$ conveniently small, we know that, for sufficiently large $k$, the set $
(kE_v)\cap\left( \left[-\frac12,\frac12\right]\times\left[ \frac12,1\right]\times{\ensuremath{\mathbb{R}}}\right)$ is the subgraph of a function $w$, with $|\nabla w|{\leqslant}\frac{a}{2}$. By construction, $$w(x)=k v\left(\frac{x}{k}\right),$$ and hence, using , $$\frac{a}2{\geqslant}|\nabla w(0,T^{(k)})|=\left|\nabla
v\left(0,\frac{T^{(k)}}{k}\right)\right|=|\nabla v(0,t^{(k)})|.$$ This is in contradiction with , and the proof of is thereby complete.
It is now convenient to take into account the “Jacobi field” associated to the fractional perimeter (see e.g. formula (1.5) in [@MR3798717], or formula (4.30) in [@SAEZ], or Lemma C.1 in [@MR3824212], or Section 1.3 in [@MR3934589]), namely we define $\sigma:=\frac{1+s}2$, $$\begin{aligned}
&& {\mathcal{L}}^\sigma_E\,\eta(x):=
\int_{\partial E} \frac{\eta(y)-\eta(x)}{|y-x|^{n+2\sigma}}\,d{\mathcal{H}}^n(y),\\
&&A^\sigma_E(x):=\sqrt{
\int_{\partial E} \frac{1-\nu(y)\cdot\nu(x)}{|y-x|^{n+2\sigma}}\,d{\mathcal{H}}^n(y)}\\
{\mbox{and }}&&
{\mathcal{J}}^\sigma_E\,\eta(x):={\mathcal{L}}^\sigma_E\,\eta(x)
+\big( A^\sigma_E(x)\big)^2\,\eta(x),\end{aligned}$$ where $\nu=(\nu_1,\dots,\nu_{n+1})$ is the exterior normal of $E$ (we will often write $\nu(x)$ to denote this normal at the point $(x,u(x))$ if $E=E_u$). It is known (see Theorem 1.3(i) in [@MR3934589]) that if $E$ is $s$-minimal in $B_r(x)$, with $x\in\partial E$, and $(\partial E)\cap B_r(x)$ is of class $C^3$, then $$\label{JAN-AIKS}
{\mathcal{J}}^\sigma_E\,\nu_i(x)=0,\qquad{\mbox{ for every }}i\in\{1,\dots,n+1\}.$$ With this notation, we have the following classification result:
\[nu-co\] Let $u\in C^1({\ensuremath{\mathbb{R}}}^n)$ be an $s$-minimal graph in a domain $\Omega\subseteq{\ensuremath{\mathbb{R}}}^n$ and assume that there exist $i\in\{1,\dots,n+1\}$ and $x^\star\in\Omega$ such that $$\label{8US:1197mi}0{\leqslant}\nu_i(x^\star){\leqslant}\nu_i(x)\qquad{\mbox{
for every~$x\in{\ensuremath{\mathbb{R}}}^n$.}}$$ Then $\nu_i$ is constant in ${\ensuremath{\mathbb{R}}}^n$.
We observe that $u\in C^\infty(\Omega)$, due to [@MR3934589], and therefore we can exploit and obtain that $$\begin{aligned}
&& 0={\mathcal{J}}^\sigma_E\,\nu_i(x^\star)=
\int_{\partial E} \frac{\nu_i(y)-\nu_i(x^\star)}{|y-x^\star|^{n+2\sigma}}
\,d{\mathcal{H}}^n(y)
+\big( A^\sigma_E(x^\star)\big)^2\,\nu_i(x^\star)\\&&\qquad\qquad
{\geqslant}\int_{\partial E} \frac{\nu_i(y)-\nu_i(x^\star)}{|y-x^\star|^{n+2\sigma}}
\,d{\mathcal{H}}^n(y).\end{aligned}$$ This and give that $\nu_i(y)=\nu_i(x^\star)$ for every $y\in{\ensuremath{\mathbb{R}}}^n$.
[F]{}rom this and Lemma \[STEP1\] we deduce that the boundary continuity of homogeneous nonlocal minimal graphs give full rigidity and symmetry results:
\[STEP11\] Let the assumptions of Theorem \[QC\] hold. Assume also that $$\lim_{x\to (1,0)} u(x)=0=\lim_{x\to (-1,0)} u(x).$$ Then $u(x)=0$ for all $x\in{\ensuremath{\mathbb{R}}}^2$.
By Lemma \[STEP1\], we know that $u\in C^1({\ensuremath{\mathbb{R}}}^2)$. Also, recalling , we have that $\nabla u(x)=0$ if $x_2<0$. As a consequence, we see that $\nabla u(x)=0$ for all $x$ with $x_2{\leqslant}0$ and therefore $$\label{nn}
\nu_3(x)=1 \qquad{\mbox{ for all~$x$ with~$x_2{\leqslant}0$.}}$$ Now we take $x^\star=(x^\star_1,x^\star_2)$ such that $$\nu_3(x^\star)=\min_{S^1} \nu_3.$$ By , we know that, for any $x\in{\ensuremath{\mathbb{R}}}^2$, $$\label{7trferf274} \nu_3(x)=\nu_3\left( \frac{x}{|x|}\right){\geqslant}\nu_3(x^\star).$$ We claim that $$\label{nu3}
\nu_3(x)=1\qquad{\mbox{ for all~$x\in{\ensuremath{\mathbb{R}}}^2$.}}$$ To prove this, we distinguish two cases. If $x^\star_2>0$, recalling , we are in the position of using Lemma \[nu-co\]. In this way we obtain that, for every $x\in{\ensuremath{\mathbb{R}}}^2$, $$\nu_3(x)=\nu_3(0)=1,$$ which proves in this case.
If instead $x_2^\star{\leqslant}0$, we deduce from and that, for any $x\in{\ensuremath{\mathbb{R}}}^2$, $$1=\nu_3(x^\star){\leqslant}\nu_3(x){\leqslant}|\nu(x)|=1,$$ thus completing the proof of .
[F]{}rom we deduce that $\nu(x)=(0,0,1)$ and hence $\nabla u(x)=0$ for every $x\in{\ensuremath{\mathbb{R}}}^2$, from which we obtain the desired result.
Another useful ingredient towards the proof of Theorem \[QC\] consists in the following rigidity result:
\[HAusjc293ierf\] Let $u\in L^\infty_{\rm loc}({\ensuremath{\mathbb{R}}}^n)$. Let $\Omega\subseteq{\ensuremath{\mathbb{R}}}^n$ be a smooth and convex domain. Assume that $u\in C^1({\ensuremath{\mathbb{R}}}^n\setminus\overline{\Omega})$ and that there exists $x^\star\in\Omega$ such that $$\label{1g6778}
\partial_1 u(x^\star){\geqslant}\partial_1 u(x)\qquad{\mbox{ for all }}
x\in {\ensuremath{\mathbb{R}}}^n\setminus (\partial\Omega).$$ Suppose also that there exists a ball ${\mathcal{B}}\Subset
{\ensuremath{\mathbb{R}}}^n\setminus (\partial\Omega)$ such that $$\label{1g6778C}
\partial_1 u(x^\star)>\partial_1 u(x)\qquad{\mbox{ for all }}
x\in {\mathcal{B}}.$$ Then, $u$ cannot be an $s$-minimal graph in $\Omega$.
Up to a translation, we suppose that $x^\star=0$. To prove the desired result, we argue for a contradiction, supposing that $u$ is an $s$-minimal graph in $\Omega$. Then, by [@MR3934589], we have that $u$ is smooth inside $\Omega$ and thus, by formula (49) in [@MR3331523], we know that, for every $x\in\Omega$, $$\label{g6778}
\int_{{\ensuremath{\mathbb{R}}}^n} F\left( \frac{u(x+y)-u(x)}{|y|}\right)\,
\frac{dy}{|y|^{n+s}}=0,$$ with $$F(t):=\int_0^t \frac{d\tau}{(1+\tau^2)^{\frac{n+1+s}2}}.$$ Now, the idea that we want to implement is the following: if one formally takes a derivative with respect to $x_1$ of and computes it at the origin, the positivity of $F'$ and leads to the fact that $\partial_1u$ must be constant, in contradiction with . Unfortunately, this approach cannot be implemented directly, since $u$ is not smooth across $\partial\Omega$ and therefore one cannot justify the derivative of under the integral sign.
To circumvent this difficulty, we argue as follows. We let ${\varepsilon}\in(0,1)$ to be taken as small as we wish in what follows. We also take $\delta>0$ such that $B_{3\delta}\subset\Omega$, and we set $$G_{\varepsilon}(y):=F\left( \frac{u({\varepsilon}e_1+y)-u({\varepsilon}e_1)}{|y|}\right),$$ where $e_1=(1,0,\dots,0)\in{\ensuremath{\mathbb{R}}}^n$. We also denote by $G_0$ the function $G_{\varepsilon}$ when ${\varepsilon}=0$.
We observe that $$\label{adesso}
\begin{split}
G_{\varepsilon}(y)-G_0(y)\;=\;&F\left( \frac{u({\varepsilon}e_1+y)-u({\varepsilon}e_1)}{|y|}\right)-
F\left( \frac{u(y)-u(0)}{|y|}\right)\\
=\;& a_{\varepsilon}(y)\;
\,\frac{u({\varepsilon}e_1+y)-u({\varepsilon}e_1)-u(y)+u(0)}{|y|},
\end{split}$$ where $$a_{\varepsilon}(y):=
\int_0^1 F'\left( \frac{t[u({\varepsilon}e_1+y)-u({\varepsilon}e_1)]+(1-t)[u(y)-u(0)]}{|y|}\right)\,dt\in(0,1].$$ We remark that $$\label{VUSJDrg}
\lim_{{\varepsilon}\searrow0}a_{\varepsilon}(y)=a_0(y):=
F'\left( \frac{ u(y)-u(0) }{|y|}\right).$$ Furthermore, we observe that, if ${\varepsilon}\in(0,\delta)$, $$\label{2347gBS12}\Big|
u({\varepsilon}e_1)-u(0)- {\varepsilon}\partial_1 u(0)\Big|{\leqslant}{\varepsilon}^2\| u\|_{C^2(B_{\delta})}.$$ Now, we let ${\mathcal{R}}:={\ensuremath{\mathbb{R}}}^n\setminus (\partial\Omega)$ and we distinguish two regions of space, namely if $y\in {\mathcal{R}}\setminus B_\delta$ and if $y\in B_\delta$. Firstly, if $y\in {\mathcal{R}}\setminus B_\delta$, we consider the segment joining $y$ to ${\varepsilon}e_1+y$, and we observe that it meets $\partial\Omega$ in at most one point, in light of the convexity of the domain. That is, we have that $y+\tau e_1\in {\mathcal{R}}$ for every $\tau\in[0,\tau_{\varepsilon})\cup(\tau_{\varepsilon},{\varepsilon})$ for a suitable $\tau_{\varepsilon}\in(0,{\varepsilon}]$ (with the notation that when $\tau_{\varepsilon}={\varepsilon}$ the set $(\tau_{\varepsilon},{\varepsilon})$ is empty). Then, we have that $$\begin{aligned}
u({\varepsilon}e_1+y)-u(y)&=&u(y+{\varepsilon}e_1)-
u(y+\tau_{\varepsilon}e_1)+u(y+\tau_{\varepsilon}e_1)
-u(y)\\&=&
\int_{\tau_{\varepsilon}}^{\varepsilon}\partial_1 u(y+\tau e_1)\,d\tau+
\int^{\tau_{\varepsilon}}_0 \partial_1 u(y+\tau e_1)\,d\tau\\&{\leqslant}&
\int_{\tau_{\varepsilon}}^{\varepsilon}\partial_1 u(0)\,d\tau+
\int^{\tau_{\varepsilon}}_0 \partial_1 u(0)\,d\tau\\&=&{\varepsilon}\partial_1 u(0),\end{aligned}$$ thanks to . Therefore, for every $y\in {\mathcal{R}}\setminus B_\delta$, recalling and we have that $$\begin{aligned}
G_{\varepsilon}(y)-G_0(y)&{\leqslant}&
a_{\varepsilon}(y)\;
\,\frac{{\varepsilon}\partial_1 u(0)-u({\varepsilon}e_1)+u(0)}{|y|}\\
&{\leqslant}&
\frac{{\varepsilon}^2\| u\|_{C^2(B_{\delta})}}{|y|}.\end{aligned}$$ Consequently, $$\label{JSY:001}
\int_{ {\mathcal{R}}\setminus (B_\delta\cup{\mathcal{B}}) }\frac{
G_{\varepsilon}(y)-G_0(y)}{{\varepsilon}}\,\frac{dy}{|y|^{n+s}}{\leqslant}{\varepsilon}\,\| u\|_{C^2(B_{\delta})}\,
\int_{ {\mathcal{R}}\setminus B_\delta }\frac{dy}{|y|^{n+s+1}}
{\leqslant}\frac{C\,{\varepsilon}\,\| u\|_{C^2(B_{\delta})}}{\delta^{s+1}},$$ for some $C>0$.
Furthermore, if $y\in{\mathcal{B}}\setminus B_\delta$, we recall and we write that $$\begin{aligned}
u({\varepsilon}e_1+y)-u(y)-u({\varepsilon}e_1)+u(0)&{\leqslant}& {\varepsilon}\int_0^1\partial_1u(\tau {\varepsilon}e_1+y)\,d\tau-
{\varepsilon}\partial_1 u(0)+{\varepsilon}^2\| u\|_{C^2(B_{\delta})}\\
&=&{\varepsilon}\int_0^1\big(\partial_1 u(\tau {\varepsilon}e_1+y)-
\partial_1 u(0)\big)\,d\tau+{\varepsilon}^2\| u\|_{C^2(B_{\delta})}.\end{aligned}$$ This gives that $$\label{JSY:002}
\begin{split}&
\int_{ {\mathcal{B}}\setminus B_\delta }\frac{
G_{\varepsilon}(y)-G_0(y)}{{\varepsilon}}\,\frac{dy}{|y|^{n+s}}\\ {\leqslant}\;&\int_{ {\mathcal{B}}\setminus B_\delta }\left[
\int_0^1\big(\partial_1 u(\tau {\varepsilon}e_1+y)-
\partial_1 u(0)\big)\,d\tau\right]\,\frac{a_{\varepsilon}(y)\,dy}{|y|^{n+s+1}}+\frac{C\,
{\varepsilon}\| u\|_{C^2(B_{\delta})}}{ \delta^{s+1}}
,\end{split}$$ up to renaming $C>0$.
Now we focus on the case $y\in B_\delta$. In this case, using that $F'$ is even, we see that $$\begin{aligned}
&& G_{\varepsilon}(y)+G_{\varepsilon}(-y)-G_0(y)-G_0(-y)\\
&=& F\left( \frac{u({\varepsilon}e_1+y)-u({\varepsilon}e_1)}{|y|}\right)+F\left( \frac{u({\varepsilon}e_1-y)-u({\varepsilon}e_1)}{|y|}\right)
\\&&\qquad-F\left( \frac{u(y)-u(0)}{|y|}\right)-F\left( \frac{u(-y)-u(0)}{|y|}\right)
\\ &=&\int_0^{\varepsilon}\Bigg[
F'\left( \frac{u(t e_1+y)-u(t e_1)}{|y|}\right)
\frac{\partial_1 u(t e_1+y)-\partial_1 u(t e_1)}{|y|}\\&&\qquad+
F'\left( \frac{u(t e_1-y)-u(t e_1)}{|y|}\right)
\frac{\partial_1 u(t e_1-y)-\partial_1 u(t e_1)}{|y|}
\Bigg]\,dt\\ &{\leqslant}&\int_0^{\varepsilon}\Bigg[
\Bigg| F'\left( \frac{u(t e_1+y)-u(t e_1)}{|y|}\right)\Bigg|\;
\frac{|\partial_1 u(t e_1+y)
+\partial_1 u(t e_1-y)-2\partial_1 u(t e_1)|
}{|y|}\\&&\qquad+\Bigg|F'\left( \frac{u(t e_1+y)-u(t e_1)}{|y|}\right)-
F'\left( \frac{u(t e_1-y)-u(t e_1)}{|y|}\right)\Bigg|\;
\frac{|\partial_1 u(t e_1-y)-\partial_1 u(t e_1)|}{|y|}
\Bigg]\,dt\\ &{\leqslant}& C\,\int_0^{\varepsilon}\Bigg[
\frac{|\partial_1 u(t e_1+y)
+\partial_1 u(t e_1-y)-2\partial_1 u(t e_1)|
}{|y|}\\&&\qquad+\Bigg|F'\left( \frac{u(t e_1+y)-u(t e_1)}{|y|}\right)-
F'\left( \frac{u(t e_1)-u(t e_1-y)}{|y|}\right)\Bigg|\;
\frac{|\partial_1 u(t e_1-y)-\partial_1 u(t e_1)|}{|y|}
\Bigg]\,dt
\\&{\leqslant}& C\,\int_0^{\varepsilon}\Bigg[
\frac{\|u\|_{C^3(B_{2\delta})}\,|y|^2
}{|y|}+
\frac{|u(t e_1+y)+u(t e_1-y)-2u(t e_1)|}{|y|}
\;\frac{|\partial_1 u(t e_1-y)-\partial_1 u(t e_1)|}{|y|}
\Bigg]\,dt\\&{\leqslant}& C\,\int_0^{\varepsilon}\Bigg[
\|u\|_{C^3(B_{2\delta})}\,|y|+
\frac{\|u\|_{C^2(B_{2\delta})}^2\,|y|^2}{|y|}
\Bigg]\,dt\\
&{\leqslant}& C\,\big(1+\|u\|_{C^3(B_{2\delta})}^2\big)\,{\varepsilon}\,|y|,\end{aligned}$$ up to renaming $C$.
As a consequence, $$\label{JSY:003}\begin{split}&
\int_{ B_\delta }\frac{
G_{\varepsilon}(y)-G_0(y)}{{\varepsilon}}\,\frac{dy}{|y|^{n+s}}
=\frac12
\int_{ B_\delta }\frac{
\big(G_{\varepsilon}(y)-G_0(y)\big)+\big(
G_{\varepsilon}(-y)-G_0(-y)\big)
}{{\varepsilon}}\,\frac{dy}{|y|^{n+s}}\\&\qquad
{\leqslant}C\,\big(1+\|u\|_{C^3(B_{2\delta})}^2\big)\,
\int_{ B_\delta }\frac{dy}{|y|^{n+s-1}}=C\,\big(1+\|u\|_{C^3(B_{2\delta})}^2\big)\,\delta^{1-s},
\end{split}$$ for some $C>0$ independent of ${\varepsilon}$ and $\delta$.
Now we combine , and with . In this way, we find that $$\begin{aligned}
0&=&\frac1{\varepsilon}\left[
\int_{{\ensuremath{\mathbb{R}}}^n} F\left( \frac{u({\varepsilon}e_1+y)-u({\varepsilon}e_1)}{|y|}\right)\,
\frac{dy}{|y|^{n+s}}-
\int_{{\ensuremath{\mathbb{R}}}^n} F\left( \frac{u(y)-u(0)}{|y|}\right)\,
\frac{dy}{|y|^{n+s}}\right]\\&=&
\int_{{\ensuremath{\mathbb{R}}}^n}\frac{
G_{\varepsilon}(y)-G_0(y)}{{\varepsilon}}\,\frac{dy}{|y|^{n+s}}\\
&=&
\int_{{\mathcal{R}}\setminus( B_\delta\cup{\mathcal{B}})}\frac{
G_{\varepsilon}(y)-G_0(y)}{{\varepsilon}}\,\frac{dy}{|y|^{n+s}}+
\int_{B_\delta}\frac{
G_{\varepsilon}(y)-G_0(y)}{{\varepsilon}}\,\frac{dy}{|y|^{n+s}}\\&&\qquad
+
\int_{{\mathcal{B}}\setminus B_\delta}\frac{
G_{\varepsilon}(y)-G_0(y)}{{\varepsilon}}\,\frac{dy}{|y|^{n+s}}\\&{\leqslant}&
\int_{ {\mathcal{B}}\setminus B_\delta }\left[
\int_0^1\big(\partial_1 u(\tau {\varepsilon}e_1+y)-
\partial_1 u(0)\big)\,d\tau\right]\,\frac{a_{\varepsilon}(y)\,dy}{|y|^{n+s+1}}\\
&&\qquad+\frac{2C\,
{\varepsilon}\| u\|_{C^2(B_{\delta})}}{\delta^{s+1}}+C\,\big(1+\|u\|_{C^3(B_{2\delta})}^2\big)\,\delta^{1-s}.\end{aligned}$$ For this reason, and recalling , when we take the limit as ${\varepsilon}\searrow0$ we see that $$0{\leqslant}\int_{ {\mathcal{B}}\setminus B_\delta }\big(\partial_1 u(y)-
\partial_1 u(0)\big)\,\frac{a_0(y)\,dy}{|y|^{n+s+1}}+C\,\big(1+\|u\|_{C^3(B_{2\delta})}^2\big)\,\delta^{1-s}
.$$ Now we can take the limit as $\delta\searrow0$ and conclude that $$0{\leqslant}\int_{ {\mathcal{B}} }\big(\partial_1 u(y)-
\partial_1 u(0)\big)\,\frac{a_0(y)\,dy}{|y|^{n+s+1}}
.$$ [F]{}rom this and a contradiction plainly follows.
The proof of Theorem \[QC\] will also rely on the following simple, but interesting, calculus observation:
\[87923-4656\] Let $\delta>0$ and $g\in C([0,\delta])\cap C^1((0,\delta])$. Assume that $$\label{7US-dc}
\lim_{t\searrow0} g'(t)=+\infty$$ and let $$\label{grty0}
V(t):=g(t)-tg'(t).$$ Suppose also that the following limit exists $$\label{8903456wdf}
{\ensuremath{\mathbb{R}}}\cup\{-\infty\}\ni V(0):=\limsup_{t\searrow0} V(t){\leqslant}g(0).$$ Then there exists $t_0\in(0,\delta)$ such that $$\label{jva7823}V(t_0)> V(0).$$ In particular, there exists $t_\star\in(0,\delta]$ such that $$\label{jva7823BIS} \max_{t\in[0,\delta]} V(t)=V(t_\star).$$
For every $r{\geqslant}0$ we consider the straight line $$R_r(t):= \frac{\big(g(\delta)-g(0)\big)\,t}{\delta}+g(0)+r.$$ We observe that, if $r{\geqslant}4\|g\|_{L^\infty([0,\delta])}+1$, for all $t\in[0,\delta]$, $$R_r(t)-g(t){\geqslant}-
\frac{2\|g\|_{L^\infty([0,\delta])}\,t}{\delta}-\|g\|_{L^\infty([0,\delta])}+r-\|g\|_{L^\infty([0,\delta])}
{\geqslant}r-4\|g\|_{L^\infty([0,\delta])}{\geqslant}1.$$ Hence, we take $r_\star{\geqslant}0$ to be the smallest $r$ for which $R_r(t)-g(t){\geqslant}0$ for all $t\in[0,\delta]$.
We claim that $$\label{rsya}
r_\star>0.$$ Indeed, if not, we have that, for all $r{\geqslant}0$ and all $t\in(0,\delta]$, $$\begin{aligned}
0{\leqslant}\frac{R_r(t)-g(t)}t=\frac{ g(\delta)-g(0) }{\delta}+ \frac{g(0)+r-g(t)}t,\end{aligned}$$ and thus $$0{\leqslant}\frac{ g(\delta)-g(0) }{\delta}+ \frac{g(0)-g(t)}t=
\frac{ g(\delta)-g(0) }{\delta}- \frac{1}t\int_0^t g'(\tau)\,d\tau
.$$ Recalling , we have that for any $M>0$ there exists $t_M>0$ such that, for every $t\in(0,t_M)$, $g'(t){\geqslant}M$ and consequently $$M{\leqslant}\frac{1}{t_M}\int_0^{t_M} g'(\tau)\,d\tau{\leqslant}\frac{ g(\delta)-g(0) }{\delta}.$$ This is a contradiction if $M$ is taken sufficiently large, and hence the proof of is complete.
Then, there exists $t_0\in[0,\delta]$ such that $R_{r_\star}(t_0)=g(t_0)$. In addition, since, by , $$\begin{aligned}
&& R_{r_\star}(0)=g(0)+r_\star>g(0)\\
\quad{\mbox{and }}&&
R_{r_\star}(\delta)= \frac{\big(g(\delta)-g(0)\big)\,\delta}{\delta}+g(0)+r_\star
=g(\delta)+r_\star>g(\delta)
,\end{aligned}$$ we have that $t_0\not\in\{0,\delta\}$.
We show now that $t_0$ satisfies . For this, using that $R_{r_\star}(t_0)-g(t_0)=0{\leqslant}R_{r_\star}(t)-g(t)$ for every $t\in[0,\delta]$, it follows that $$0=
R_{r_\star}'(t_0)-g'(t_0)=
\frac{g(\delta)-g(0)}{\delta}-g'(t_0).$$ This and say that $$\begin{aligned}
&& V(t_0)= g(t_0)-t_0\,g'(t_0)=
g(t_0)-
\frac{\big(g(\delta)-g(0)\big)\,t_0}{\delta}\\&&\qquad
=g(t_0)-R_{r_\star}(t_0)+g(0)+r_\star=g(0)+r_\star>g(0).\end{aligned}$$ This, together with , proves , which in turn implies .
Now, we can complete the proof of Theorem \[QC\] in dimension $n:=2$ (the case $n:=1$ being already covered by Theorem 4.1 in [@2019arXiv190405393D]), by arguing as follows.
By in Lemma \[lemma33\], we can define, for all $x_1\in[-1,1]$, $$\label{uias23} u(x_1,0):=\lim_{{(y_1,y_2)\to(x_1,0)}\atop{y_2>0}}u(y_1,y_2).$$ Notice that if $u(-1,0)=u(1,0)=0$, then the desired result follows from Corollary \[STEP11\]. Hence, without loss of generality, we can assume that $$\label{Qnd}
(0,+\infty)\ni u(1,0){\geqslant}|u(-1,0)|.$$ We claim that such case cannot hold, by reaching a contradiction. To this end, in view of and [@MR3532394], we have that, in a small neighborhood of $P=(P_1,P_2,P_3):=(1,0,u(1,0))$, one can write $$\label{UNAJdf}
{\mbox{$(\partial E_u)\cap\{x_2>0\}$ as the graph of a $C^{1,\frac{1+s}2}$-function in the $e_2$-direction,}}$$ that is, in the vicinity of the point $P$, the set $\{x_{3}=u(x)\}$ in $\{x_2>0\}$ coincides with the set $\{x_2=v(x_1,x_{3})\}$, for a suitable $v\in
C^{1,\frac{1+s}2}({\ensuremath{\mathbb{R}}}^2)$, with $$\label{x2x0}
{\mbox{$v(P_1,x_3)=0$
if~$x_3{\leqslant}P_3$, and $v(P_1,x_3){\geqslant}0$
if~$x_3{\geqslant}P_3$.}}$$ Hence, for $x$ close to $P$ with $x_2>0$, we can write that $$\label{x2x1} x_2=v\big(x_1,u(x_1,x_2)\big).$$
Now, we let $\delta_0\in(0,1]$ to be taken conveniently small in what follows, we define $${\mathcal{Q}}:=\Big\{x=(x_1,x_2)\in{\ensuremath{\mathbb{R}}}^2 {\mbox{ s.t. either $|x_1|=1$
and~$0<|x_2|{\leqslant}\delta_0$,
or $|x_1|{\leqslant}1$
and~$|x_2|=\delta_0$}}\Big\},$$ and we claim that there exists $x^\star=(x^\star_1,x^\star_2)$ such that $x^\star_2>0$ and $$\label{2347we}
\partial_1 u(x^\star)=\max_{{\mathcal{Q}}}\partial_1 u\in(0,+\infty).$$ For this, we define $$\label{la-ggad}
g(t):=u(1,t)\qquad{\mbox{and}}\qquad V(t):=\partial_1 u(1,t).$$ We use and the smoothness of $u$ in $\{x_2>0\}$ (see [@MR3934589]), to see that, if $x_2>0$ is sufficiently small, $$0 =\partial_2\Big( x_2-v\big(1,u(1,x_2)\big)\Big)
= 1-\partial_3 v\big(1,u(1,x_2)\big)\,\partial_2 u(1,x_2).$$ In particular, we have that $$\begin{aligned}
&&
\label{pap3}\partial_3 v\big(1,u(1,x_2)\big)\ne0,\\
\label{pap5a}&&\partial_2 u(1,x_2)\ne 0\\
\label{pap5}
{\mbox{and }}&& \partial_2 u(1,x_2)=
\frac{1}{ \partial_3 v\big(1,u(1,x_2)\big) }\end{aligned}$$ if $x_2>0$ is sufficiently small.
By and , we conclude that, if $x_2>0$ is sufficiently small, $$\partial_3 v\big(1,u(1,x_2)\big)>0.$$ This, and give that $$\label{degeru7443y}
\lim_{x_2\searrow0} \partial_2 u(1,x_2)=\lim_{x_3\searrow P_3}
\frac{1}{ \partial_3 v\big(1,x_3\big) } =+\infty,$$ that is, in the notation of , $$\label{LLxv23e0}
\lim_{t\searrow0}g'(t)=+\infty.$$ Also, by , we have that $ u(x)=\nabla u(x)\cdot x$, and consequently $$\label{alsousing}
u(1,x_2)=\partial_1u(1,x_2)+x_2\partial_2u(1,x_2).$$ Thus, recalling the notation in , if $t>0$ is sufficiently small, $$\label{LLxv23e1}
V(t)-g(t)+tg'(t)=\partial_1 u(1,t)-u(1,t)+t\partial_2 u(1,t)
=0.$$ Using , and the notation of , we see that $$V(t)=\partial_1 u(1,t)= u(1,t)-t\partial_2 u(1,t){\leqslant}u(1,t)=g(t),$$ and consequently $$\label{LLxv23e} {\ensuremath{\mathbb{R}}}\cup\{-\infty\}\ni \limsup_{t\searrow0}V(t){\leqslant}g(0).$$ Now, in light of , and , we see that conditions , and are satisfied in this setting. Consequently, we can exploit Lemma \[87923-4656\] and deduce from that there exists $t_\star\in(0,\delta_0]$ such that, for all $t\in[0,\delta_0]$, $$\label{365}
\partial_1u(1,t)= V(t){\leqslant}V(t_\star)=\partial_1u(1,t_\star)=\max_{{\mathcal{Q}}\cap\{x_2>0\}} \partial_1u.$$ Now we claim that $$\label{o3ra3s2i234}
\max_{{\mathcal{Q}}\cap\{x_2>0\}} \partial_1u>0.$$ Indeed, if not, we have that $\partial_1u(1,t){\leqslant}0$, as long as $t\in(0,\delta_0)$. In particular, if $\tau\in(0,\delta_0^2)$ and $\sigma\in(\sqrt{\tau},1)$, we have that $\frac\tau\sigma{\leqslant}\sqrt\tau<\delta_0$ and, as a result, exploiting , $$\begin{aligned}
&& u(1,\tau)-u(\sqrt{\tau},\tau)=
\int_{ \sqrt{\tau} }^1 \partial_1 u(\sigma,\tau)\,d\sigma=
\int_{ \sqrt{\tau} }^1 \partial_1 u\left(1,\frac\tau\sigma\right)\,d\sigma{\leqslant}0.\end{aligned}$$ Consequently, taking the limit as $\tau\searrow0$ and recalling and , we conclude that $u(1,0){\leqslant}0$. This is in contradiction with and thus this completes the proof of .
Since $\partial_1 u=0$ in $\{x_2<0\}$, from and , we have that $$(0,+\infty)\ni\partial_1u(1,t_\star)=\max_{{\mathcal{Q}}} \partial_1u,$$ and hence follows directly .
This, together with , allows us to exploit Lemma \[HAusjc293ierf\], used here with $\Omega:=\{x\in(0,2)\times(0,+\infty)
{\mbox{ s.t. }}\frac{x_2}{x_1}<\delta_0\}$, ${\mathcal{B}}:=B_1(-2,-2)$ and $x^\star:=
(1,t_\star)$, and this yields that $u$ cannot be $s$-minimal in $\Omega\subset\{x_2>0\}$. This is a contradiction with our assumptions, and hence the setting in cannot occur.
As a result, we have that $u(x_1,0)=0$ for all $x_1\in{\ensuremath{\mathbb{R}}}$. Then, in view of Corollary \[STEP11\], we conclude that $u$ vanishes identically, and this completes the proof of Theorem \[QC\].
It is interesting to point out that, as a byproduct of Theorem \[QC\], one also obtains the following alternatives on the second blow-up:
\[COR:ALTEzz\] Let $u$ be an $s$-minimal graph in $(-2,2)\times(0,4)$. Assume that there exists $h>0$ such that $u=0$ in $ (-2,2)\times(-h,0)$.
Let $E_{00}$ be the second blow-up of $E_u$. Then: $$\begin{aligned}
\label{POS-E001-BI1}
&& {\mbox{either $E_{00}\cap\{x_2>0\}=\varnothing$,}}
\\ \label{POS-E001-BI2}&& {\mbox{or $E_{00}\cap\{x_2>0\}=\{x_2>0\}$,}}\\&&
\label{POS-E001-BI3}{\mbox{or $E_{00}=\{x_{3}<0\}$.}}\end{aligned}$$
We assume that neither nor hold true, and we prove that is satisfied. For this, we first exploit Lemma \[BOAmdfiUAMP\], deducing from and that $E_{00}$ has a graphical structure, with respect to some function $u_{00}$, satisfying $$\label{ADX662738495jdjfgjg-2}
\lim_{x\to0}u_{00}(x)=0.$$ Moreover, we know that $E_{00}$ is a homogeneous set (see e.g. Lemma 2.2 in [@2019arXiv190405393D]), and thus $$\label{ADX662738495jdjfgjg}
u_{00}(tx)=tu_{00}(x)\qquad{\mbox{ for all $x\in{\ensuremath{\mathbb{R}}}^2$ and~$t>0$.}}$$ In view of and , we are in the position of applying Theorem \[QC\] to the function $u_{00}$, and thus we conclude that $u_{00}$ vanishes identically, and this establishes .
We also observe that Corollary \[COR:ALTEzz\] can be further refined in light of Lemma \[554327h:AJSJD2374859ty\].
\[COR:ALTE\] Let $u$ be an $s$-minimal graph in $(-2,2)\times(0,4)$. Assume that there exists $h>0$ such that $u=0$ in $ (-2,2)\times(-h,0)$.
Let $E_{00}$ be the second blow-up of $E_u$. Then: $$\begin{aligned}
\label{POS-E001-BI1b}
&& {\mbox{either $E_{00}\cap\{x_2>0\}=\varnothing$
and }}\lim_{x_2\searrow0} u(0,x_2)<0,
\\ \label{POS-E001-BI2b}&& {\mbox{or $E_{00}\cap\{x_2>0\}=\{x_2>0\}$
and }}
\lim_{x_2\searrow0} u(0,x_2)>0,
\\&&
\label{POS-E001-BI3b}{\mbox{or $E_{00}=\{x_{3}<0\}$.}}\end{aligned}$$
In the setting of Corollary \[COR:ALTEzz\], if holds true, then we exploit and readily follows.
Similarly, combining with , we obtain .
Finally, the situation in coincides with that in , thus exhausting all the available possibilities.
Useful barriers {#HDB}
===============
In this section we construct an auxiliary barrier, that we will exploit in the proof of Theorem \[NLMS\] to rule out the case of boundary Lipschitz singularities. For this, we will rely on a special function introduced in Lemma 7.1 of [@2019arXiv190405393D] and on a codimension-one auxiliary construction (given the possible use of such barriers in other context, we give our construction in a general dimension $n$, but we will then restrict to the case $n=2$ when dealing with the proof of our main theorems, see Open Problem \[17\]).
These barriers rely on a purely nonlocal feature, since they present a corner at the origin, which maintains a significant influence on the nonlocal mean curvature in a full neighborhood (differently from the classical case, in which the mean curvature is a local operator).
To perform our construction, we recall the definition of the nonlocal mean curvature in and we first show that flat higher dimensional extensions preserve the nonlocal mean curvature, up to a multiplicative constant:
\[5444542e62734\] Let $G\subset{\ensuremath{\mathbb{R}}}^2$ and $(a,b)\in\partial G$. Let $$G^\star:=\{(x_1,\dots,x_n,x_{n+1})\in{\ensuremath{\mathbb{R}}}^{n+1}
{\mbox{ s.t. }} (x_n,x_{n+1})\in G\}.$$ Then, for every $p=(p_1,\dots,p_n,p_{n+1})\in{\ensuremath{\mathbb{R}}}^{n+1}$ with $(p_n,p_{n+1})=(a,b)$, $${\mathcal{H}}^s_{G^\star}(p)=C(n,s)\,{\mathcal{H}}^s_{G}(a,b),$$ for a suitable constant $C(n,s)>0$.
Using the notation $y'=(y_1,\dots,y_{n-1})$ and the change of variable $$\zeta':=\frac{y'-p'}{\sqrt{|y_n-p_n|^2+|y_{n+1}-p_{n+1}|^2}},$$ we have that $$\begin{aligned}
{\mathcal{H}}^s_{G^\star}(p)&=&\int_{{\ensuremath{\mathbb{R}}}^{n+1}}\frac{\chi_{{\ensuremath{\mathbb{R}}}^{n+1}\setminus G^\star}(y)-
\chi_{G^\star}(y)
}{|y-p|^{n+1+s}}\,dy\\&=&
\int_{{\ensuremath{\mathbb{R}}}^{n+1}}\frac{\chi_{{\ensuremath{\mathbb{R}}}^2\setminus G}(y_n,y_{n+1})-
\chi_{G}(y_n,y_{n+1})
}{\big(|y'-p'|^2 +|y_n-p_n|^2+|y_{n+1}-p_{n+1}|^2\big)^{\frac{n+1+s}2}}\,dy\\&=&
\int_{{\ensuremath{\mathbb{R}}}^2}\left[
\int_{{\ensuremath{\mathbb{R}}}^{n-1}}\frac{\chi_{{\ensuremath{\mathbb{R}}}^2\setminus G}(y_n,y_{n+1})-
\chi_{G}(y_n,y_{n+1})
}{\big(1+|\zeta'|^2\big)^{\frac{n+1+s}2}\,
\big(|y_n-a|^2+|y_{n+1}-b|^2\big)^{\frac{2+s}2}}\,d\zeta'\right]\,d(y_n,y_{n+1})\\
&=& \int_{{\ensuremath{\mathbb{R}}}^{n-1}}\frac{d\zeta'}{\big(1+|\zeta'|^2\big)^{\frac{n+1+s}2}}\;{\mathcal{H}}^s_{G}(a,b),\end{aligned}$$ which gives the desired result.
While Lemma \[5444542e62734\] deals with a flat higher dimensional extension of a set, we now turn our attention to the case in which a higher dimensional extension is obtained by a given function $\Phi\in C^{1,\vartheta}({\ensuremath{\mathbb{R}}}^{n-1})$, with $\vartheta\in(s,1]$. In this setting, using the notation $x'=(x_1,\dots,x_{n-1})$, given $L>0$ and a function $\beta:(-L,L)\to{\ensuremath{\mathbb{R}}}$, for every $\varpi{\geqslant}0$ we define $$E^{(\beta,\varpi)}:= \big\{
x=(x_1,\dots,x_n,x_{n+1})\in{\ensuremath{\mathbb{R}}}^{n+1}{\mbox{ s.t. }}x_{n+1}<\beta(x_n)-\varpi\,\Phi(x')
\big\}.$$ We also extend $\beta$ to take value equal to $-\infty$ outside $(-L,L)$. Then, recalling the framework in , we can estimate the nonlocal mean curvature of $E^{(\beta,\varpi)}$ with that of $E_\beta$ as follows:
\[VGbjatrtta9\] Let $\gamma>0$. If $p=(p_1,\dots,p_n,p_{n+1})\in\partial E^{(\beta,\varpi)}$ with $p_n\in(-L,L)$, $\beta\in C^{1,1}\big([p_n-\gamma,p_n+\gamma]\big)$, and $C(n,s)>0$ is as in Lemma \[5444542e62734\], we have that $$\big| {\mathcal{H}}^s_{E^{(\beta,\varpi)}}(p)-C(n,s)\,
{\mathcal{H}}^s_{E_\beta}(p_n,p_{n+1})\big|{\leqslant}C\,\varpi\,(1+\varpi),$$ for a suitable constant $C>0$ depending only on $n$, $s$, $L$, $\gamma$, $\Phi$ and $\beta$.
We use the notation $x=(x_1,\dots,x_n)$ and $\beta_\varpi(x):=\beta(x_n)-\varpi\,\Phi(x')$. We recall to write that $$\label{923461d52s99tgv7}
\begin{split}
2{\mathcal{H}}^s_{E^{(\beta,\varpi)}}(p)\,=\,&
2{\mathcal{H}}^s_{E_{\beta_\varpi}}(p)
\\ =\,&
\int_{{\ensuremath{\mathbb{R}}}^n} \left[
F\left( \frac{\beta_\varpi(p+y)-\beta_\varpi(p)}{|y|}\right)+
F\left( \frac{\beta_\varpi(p-y)-\beta_\varpi(p)}{|y|}\right)\right]
\,
\frac{dy}{|y|^{n+s}}\\
=\,&
\int_{{\ensuremath{\mathbb{R}}}^n} \left[
F\left( \frac{\beta(p_n+y_n)-\beta(p_n)}{|y|}-\frac{\varpi \big(\Phi(p'+y')
-\Phi(p')\big)}{|y|}\right)\right.\\
&\qquad+\left.
F\left( \frac{\beta(p_n-y_n)-\beta(p_n)}{|y|}-\frac{\varpi \big(\Phi(p'-y')
-\Phi(p')\big)}{|y|}\right)\right]
\,
\frac{dy}{|y|^{n+s}}\,.
\end{split}$$ Now we define $$\begin{aligned}
\Psi(\varpi)&:=&
F\left( \frac{\beta(p_n+y_n)-\beta(p_n)}{|y|}-\frac{\varpi \big(\Phi(p'+y')
-\Phi(p')\big)}{|y|}\right)
\\&&\qquad+F\left( \frac{\beta(p_n-y_n)-\beta(p_n)}{|y|}-\frac{\varpi
\big(\Phi(p'-y')
-\Phi(p')\big)}{|y|}\right).\end{aligned}$$ We remark that $$\begin{aligned}
|\Psi'(\varpi)|&=&\Bigg|
F'\left( \frac{\beta(p_n+y_n)-\beta(p_n)}{|y|}-\frac{\varpi \big(\Phi(p'+y')
-\Phi(p')\big)}{|y|}\right)
\frac{\Phi(p'+y')
-\Phi(p')}{|y|}
\\&&\qquad+F'\left( \frac{\beta(p_n-y_n)-\beta(p_n)}{|y|}-\frac{\varpi
\big(\Phi(p'-y')
-\Phi(p')\big)}{|y|}\right)
\frac{\Phi(p'-y')
-\Phi(p')}{|y|}\Bigg|
\\&{\leqslant}&
\left|F'\left( \frac{\beta(p_n+y_n)-\beta(p_n)}{|y|}-\frac{\varpi \big(\Phi(p'+y')
-\Phi(p')\big)}{|y|}\right)\right|\\
&&\quad\times\left|
\frac{\Phi(p'+y')
-\Phi(p')}{|y|}+\frac{\Phi(p'-y')
-\Phi(p')}{|y|}\right|\\
&&+
\left|F'\left( \frac{\beta(p_n+y_n)-\beta(p_n)}{|y|}-\frac{\varpi \big(\Phi(p'+y')
-\Phi(p')\big)}{|y|}\right)\right.\\
&&\quad\left.-F'\left( \frac{\beta(p_n+y_n)-\beta(p_n)}{|y|}-\frac{\varpi \big(\Phi(p'-y')
-\Phi(p')\big)}{|y|}\right)\right|\;\left|
\frac{\Phi(p'-y')
-\Phi(p')}{|y|}\right|
.\end{aligned}$$ We also use that $F'$ is even to observe that $$\begin{aligned}
&&
\Bigg|
F'\left( \frac{\beta(p_n+y_n)-\beta(p_n)}{|y|}-\frac{\varpi
\big(\Phi(p'+y')
-\Phi(p')\big)}{|y|}\right)\\
&&\qquad-F'\left( \frac{\beta(p_n-y_n)-\beta(p_n)}{|y|}-\frac{\varpi \big(\Phi(p'-y')
-\Phi(p')\big)}{|y|}\right)
\Bigg|\\
&=&
\Bigg|
F'\left( \frac{\beta(p_n+y_n)-\beta(p_n)}{|y|}-\frac{\varpi
\big(\Phi(p'+y')
-\Phi(p')\big)}{|y|}\right)\\
&&\qquad-F'\left( \frac{\beta(p_n)-\beta(p_n-y_n)}{|y|}+\frac{\varpi
\big(\Phi(p'-y')
-\Phi(p')\big)}{|y|}\right)
\Bigg|\\
&{\leqslant}& \min\Bigg\{2\,\| F'\|_{L^\infty({\ensuremath{\mathbb{R}}})},\\&&\quad
\| F''\|_{L^\infty({\ensuremath{\mathbb{R}}})}\left(
\frac{|\beta(p_n+y_n)+\beta(p_n-y_n)-2\beta(p_n)|}{|y|}+\varpi\,
\frac{\big|\Phi(p'+y')+\Phi(p'-y')
-2\Phi(p')\big|}{|y|}
\right)
\Bigg\}\\
&{\leqslant}& C\,(1+\varpi)\,\min\{ 1,|y|^\vartheta\},\end{aligned}$$ for some $C>0$.
Moreover, since $F'(-\infty)=0$, $$\begin{aligned}
&&
\Bigg|
F'\left( \frac{\beta(p_n+y_n)-\beta(p_n)}{|y|}-\frac{\varpi \big(\Phi(p'+y')
-\Phi(p')\big)}{|y|}\right)
\Bigg|\\
&=&
\Bigg|
F'\left( \frac{\beta(p_n+y_n)-\beta(p_n)}{|y|}-\frac{\varpi \big(\Phi(p'+y')
-\Phi(p')\big)}{|y|}\right)
\Bigg|\;\chi_{(-p_n-L,-p_n+L)}(y_n)
\\&{\leqslant}& C\,\chi_{(-2L,2L)}(y_n),\end{aligned}$$ up to renaming $C>0$.
[F]{}rom these observations, we deduce that $$|\Psi'(\varpi)|{\leqslant}C\,(1+\varpi)\,\left(
\min\{ 1,|y|^\vartheta\}+
\frac{|y'|^{1+\vartheta}\chi_{(-2L,2L)}(y_n)}{|y|}
\right),$$ up to renaming $C>0$.
Consequently, recalling , and writing $E^{(\beta,0)}$ to denote $E^{(\beta,\varpi)}$ with $\varpi:=0$, we find that $$\begin{aligned}
&&
\big| {\mathcal{H}}^s_{E^{(\beta,\varpi)}}(p)-{\mathcal{H}}^s_{E^{(\beta,0)}}(p)\big|\\
&=&\frac12\,\left|\int_{{\ensuremath{\mathbb{R}}}^n}\Big(
\Psi(\varpi)-\Psi(0)\Big)
\,\frac{dy}{|y|^{n+s}}\right|\\
&{\leqslant}& C\,\varpi\,(1+\varpi)\,
\int_{{\ensuremath{\mathbb{R}}}^n}
\left(
\min\{ 1,|y|^\vartheta\}+
\frac{|y'|^{1+\vartheta}\chi_{(-2L,2L)}(y_n)}{|y|}
\right)
\frac{dy}{|y|^{n+s}}\\
&{\leqslant}& C\,\varpi\,(1+\varpi)\,\left(1+
\int_{{\ensuremath{\mathbb{R}}}^n}
|y'|^{1+\vartheta}\chi_{(-2L,2L)}(y_n)\,
\frac{dy}{|y|^{n+1+s}}\right)\\
&{\leqslant}& C\,\varpi\,(1+\varpi)\,\left(1+
\int_{{{\ensuremath{\mathbb{R}}}^n}\atop{\{|y'|>4L\}}}
|y'|^{1+\vartheta}\chi_{(-2L,2L)}(y_n)\,
\frac{dy}{|y|^{n+1+s}}\right)\\
&{\leqslant}& C\,\varpi\,(1+\varpi)\,\left(1+
\int_{{{\ensuremath{\mathbb{R}}}^{n-1}}\atop{\{|y'|>4L\}}}
\frac{dy'}{|y'|^{n-\vartheta+s}}\right)\\&{\leqslant}& C\,\varpi\,(1+\varpi),\end{aligned}$$ up to renaming $C$ line after line.
The desired result now plainly follows from the latter inequality and the fact that, in view of Lemma \[5444542e62734\], we know that ${\mathcal{H}}^s_{E^{(\beta,0)}}(p)=
C(n,s)\,
{\mathcal{H}}^s_{E_\beta}(p_n,p_{n+1})$.
In the light of Lemma \[VGbjatrtta9\], we can now construct the following useful barrier:
\[RCAIcgaue023\] Let $\tilde\ell>0$ and $\bar\ell\in[-\tilde\ell,\tilde\ell]$. Let $\lambda>0$ and $L\in(\lambda,+\infty)$. Let $
a$, $b$, $c>0$, $\alpha\in(0,s)$.
Let ${\varepsilon}\in(0,1)$ and assume that $$\label{43266-78j1}
L{\geqslant}\frac{c}{{\varepsilon}^{1/s}}.$$ Let $$\beta(x_n):=\begin{cases}
\bar\ell x_n & {\mbox{ if }}x_n\in(-L,0),\\
\bar\ell x_n+{\varepsilon}a x_n& {\mbox{ if }}x_n\in[0,\lambda],\\
\bar\ell x_n-{\varepsilon}bx_n^{1+\alpha} & {\mbox{ if }}x_n\in(\lambda,L).
\end{cases}$$ Then there exist $$\label{43266-78j2}
\mu\in \left(0,\frac\lambda8\right)$$ depending only on $n$, $s$, $\tilde\ell$, $\lambda$, $a$, $b$, $c$ and $\alpha$ (but independent of ${\varepsilon}$ and $L$), and $C_\star>0$, depending only on $n$, $s$, $\tilde\ell$, $\lambda$, $a$, $b$, $c$ and $\alpha$, such that, if $$\label{43266-78j3}
\varpi\in \left[0,\min\left\{1,\,\frac{C_\star\,{\varepsilon}}{\mu^s}\right\}\right],$$ then $$\label{LAOS923i828dgtwqr}
{\mathcal{H}}^s_{E^{(\beta,\varpi)}}(p){\leqslant}-\frac{C^\star\,{\varepsilon}}{\mu^s}
<0$$ for every $p=(p_1,\dots,p_n,p_{n+1})\in\partial E^{(\beta,\varpi)}$ with $p_n\in(0,\mu)$, where $C^\star>0$ depends only on $n$, $s$, $\bar\ell$, $\lambda$, $a$, $b$, $c$ and $\alpha$.
Moreover, if $E^{(\beta,\varpi,R)}:=E^{(\beta,\varpi)}\cap \{ |x'|<R\}$ and $$\label{LAOS923i828dgtwqr-2}
R{\geqslant}\frac{C_\sharp\,\mu}{{\varepsilon}^{\frac1s}}
,$$ for a suitable $C_\sharp>0$ depending only on $n$, $s$, $\tilde\ell$, $\lambda$, $a$, $b$, $c$ and $\alpha$, then $$\label{LAOS923i828dgtwqr-3}
{\mathcal{H}}^s_{E^{(\beta,\varpi,R)}}(p)<0$$ for every $p=(p_1,\dots,p_n,p_{n+1})\in\partial E^{(\beta,\varpi,R)}$ with $p_n\in(0,\mu)$ and $|p'|<R/2$.
Firstly, we prove . By Lemma 7.1 in [@2019arXiv190405393D], we know that, under assumptions and , if $\mu$ is sufficiently small we have that $${\mathcal{H}}^s_{E_\beta}(p_n,p_{n+1}){\leqslant}-\frac{C'(\tilde\ell,a)\,{\varepsilon}}{\mu^s},$$ for some $C'(\tilde\ell,a)>0$.
[F]{}rom this and Lemma \[VGbjatrtta9\], we deduce that $$\begin{aligned}
{\mathcal{H}}^s_{E^{(\beta,\varpi)}}(p)&{\leqslant}& C(n,s)\,
{\mathcal{H}}^s_{E_\beta}(p_n,p_{n+1})+ C\,\varpi\,(1+\varpi)\\&
{\leqslant}& -\frac{C(n,s)\,C'(\tilde\ell,a)\,{\varepsilon}}{\mu^s}+ C\,\varpi\,(1+\varpi).\end{aligned}$$ [F]{}rom this and we obtain the desired result in .
Furthermore, the sets $E^{(\beta,\varpi,R)}$ and $E^{(\beta,\varpi)}$ coincide in $\{|x'|<R\}$ and therefore, if additionally $|p'|{\leqslant}R/2$, $$\begin{aligned}
&&{\mathcal{H}}^s_{E^{(\beta,\varpi,R)}}(p)
{\leqslant}{\mathcal{H}}^s_{E^{(\beta,\varpi)}}(p)+\int_{{\ensuremath{\mathbb{R}}}^{n+1}\cap \{|y'|>R\}
}\frac{dy}{|y-p|^{n+1+s}}
\\&&\qquad {\leqslant}{\mathcal{H}}^s_{E^{(\beta,\varpi)}}(p)+\int_{{\ensuremath{\mathbb{R}}}^{n+1}\setminus B_{R/2}
}\frac{dz}{|z|^{n+1+s}}
{\leqslant}-\frac{C^\star\,{\varepsilon}}{\mu^s}+\frac{C}{R^s}{\leqslant}-\frac{C^\star\,{\varepsilon}}{2\mu^s}
,\end{aligned}$$ thanks to and , and this proves .
Proof of Theorem \[NLMS\] {#657699767395535}
=========================
The proof of Theorem \[NLMS\] consists in combining Theorem \[QC\] (or, more specifically, Corollary \[COR:ALTE\]) with the boundary Harnack Inequality and the boundary improvement of flatness methods introduced in [@2019arXiv190405393D]. More specifically, in view of the boundary Harnack Inequality in [@2019arXiv190405393D], one can rephrase Lemma 6.2 of [@2019arXiv190405393D] in our setting and obtain the following convergence result of the vertical rescalings to a linearized equation:
\[NOANpierjfppp34\] Let ${\varepsilon}\in(0,1)$, $\alpha\in(0,s)$ and $u:{\ensuremath{\mathbb{R}}}^2\to{\ensuremath{\mathbb{R}}}$. Set $$u_{\varepsilon}(x):=\frac{u(x)}{\varepsilon}.$$ There exists $c_0\in(0,1)$, depending only on $\alpha$ and $s$, such that the following statement holds true. Assume that $u$ is an $s$-minimal graph in $(-2^{\tilde k_0},2^{\tilde k_0})
\times(0,2^{\tilde k_0})$, with $$\tilde k_0:=\left\lceil\frac{|\log{\varepsilon}|}{c_0}\right\rceil.$$ Suppose also that $$|u(x)|{\leqslant}{\varepsilon}^{\frac1{c_0}}\,|x|^{1+\alpha}
\qquad{\mbox{for all $x=(x_1,x_2)\in B_{ 2^{\tilde k_0} }$
with~$x_2<0$}}$$ and $$|u(x)|{\leqslant}{\varepsilon}\,(2^k)^{1+\alpha}\qquad{\mbox{for all $x=(x_1,x_2)\in B_{ 2^{k} }$
with~$x_2>0$, for all $k\in\{0,\dots,\tilde k_0\}$.}}$$ Then, as ${\varepsilon}\searrow0$, up to a subsequence, $u_{\varepsilon}$ converges locally uniformly in ${\ensuremath{\mathbb{R}}}^2$ to a function $\bar u$ satisfying $$\begin{aligned}
&&
\sup_{x\in{\ensuremath{\mathbb{R}}}^2}\frac{|\bar u(x)|}{1+|x|^{1+\alpha}}<+\infty\\
{\mbox{and }}&&(-\Delta)^{\frac{1+s}2}\bar u=0\;\,
{\mbox{ in }}\;\,{\ensuremath{\mathbb{R}}}\times(0,+\infty).\end{aligned}$$ Furthermore, if $|x|\to0$ with $x_2>0$, we can write that $$\bar u(x)=\bar a \,x_2^{\frac{1+s}2}+O(|x|^{\frac{3+s}2}),$$ for some $\bar a\in{\ensuremath{\mathbb{R}}}$.
With this, one can obtain a suitable improvement of flatness result as follows:
\[ijsd78AIsjjd\] Let ${\varepsilon}$, $h\in(0,1)$, $\alpha\in(0,s)$, $u:{\ensuremath{\mathbb{R}}}^2\to{\ensuremath{\mathbb{R}}}$ be such that $$\label{65676589cs0an3o}
{\mbox{$u(x)=0$ for all~$x\in (-2^{\tilde k_0},2^{\tilde k_0})\times(-h,0)$,}}$$ and assume that $u$ is an $s$-minimal graph in $(-2^{\tilde k_0},2^{\tilde k_0})
\times(0,2^{\tilde k_0})$, with $$\label{LAmbielsz} \tilde k_0:=\left\lceil\frac{|\log{\varepsilon}|}{c_0}\right\rceil,$$ and with $c_0\in(0,1)$ suitably small.
Suppose that $$\label{ESAy} \lim_{{x\to0}\atop{x_2>0}} u(x)=0.$$ Then, there exists $\tilde{\varepsilon}_0\in(0,1)$ such that if ${\varepsilon}\in(0,\tilde{\varepsilon}_0]$ the following statement holds true.
If $$\begin{aligned}
\label{LAPwkeryuifmma}&&
|u(x)|{\leqslant}{\varepsilon}^{\frac1{c_0}}\,|x|^{1+\alpha}\quad{\mbox{for all $x\in B_{ 2^{\tilde k_0} }$
with~$x_2<0$}}\end{aligned}$$ and $$\begin{aligned}
\label{LAPwkeryuifmma2}&&
|u(x)|{\leqslant}{\varepsilon}\,(2^k)^{1+\alpha}\quad{\mbox{for all $x\in B_{ 2^{k} }$
with~$x_2>0$, for all $k\in\{0,\dots,\tilde k_0\}$,}}\end{aligned}$$ then, for all $j\in{\ensuremath{\mathbb{N}}}$, $$|u(x)|{\leqslant}\frac{\varepsilon}{2^{j(1+\alpha)}}\qquad{\mbox{ for all $x\in
B_{1/2^j}$ with~$x_2>0$}}.$$ Moreover, $$|u(x)|{\leqslant}4{\varepsilon}|x|^{1+\alpha}
\qquad{\mbox{ for all $x\in
B_{1/2}$ with~$x_2>0$}}.$$
The proof of Theorem 8.1 of [@2019arXiv190405393D] carries over to this case, with the exception of the proof of (8.9) in [@2019arXiv190405393D] (which in turn uses the one-dimensional barrier built in Lemma 7.1 of [@2019arXiv190405393D], that is not available in the higher dimensional case that we deal with here). More precisely, using Lemma \[NOANpierjfppp34\], we obtain that, given $\delta\in(0,1)$, if ${\varepsilon}$ is sufficiently small, for all $x\in B_3$ with $x_2>0$, we have that $$\label{78hs:93e}
|u(x)-{\varepsilon}\,\bar a\,x_2^{\frac{1+s}2}|{\leqslant}\overline{C}{\varepsilon}\,\big(|x|^{\frac{3+s}2}+\delta\big),$$ for some $\overline{C}>0$.
With this, which replaces formula (8.8) of [@2019arXiv190405393D] in this setting, the proof of Theorem 8.1 of [@2019arXiv190405393D] can be applied to our framework, once we show that $$\label{RUsnsca}
\bar a=0.$$ To check that is satisfied, and thus complete the proof of Theorem \[ijsd78AIsjjd\], we argue for a contradiction and assume, for instance, that $
\bar a>0$ (the case $\bar a<0$ being similar). We take $E^{(\beta,\varpi,R)}$ as in Lemma \[RCAIcgaue023\], with $n:=2$, $$\begin{aligned}
&&
\Phi(x_1):=|x_1|^{\frac{3+s}{2}},\qquad
\bar\ell:={\varepsilon}\delta,\qquad
\lambda:=\min\left\{\frac1{10^{\frac{2}{1-s}}},\frac{\bar a}{4\,\overline{C}}\right\},\\&&
c_0:=\min\left\{ s\log 2,\frac{1-\log2}{2},\frac{s}{2(s+1+\alpha)}\right\},\qquad\varpi:=\frac{C_\star\,{\varepsilon}}{\mu^s},\qquad
L:=\frac{1}{4\,{\varepsilon}^{\frac{\log 2}{c_0}}},
\\&&
a:=\frac{\bar a}{8},\qquad b:=2^{2(2+\alpha)}+\frac{1}{\lambda^\alpha}+
\frac{2^{4(2+\alpha)}}{\lambda^{1+\alpha}},
\qquad c:=\frac{1}{4}\qquad{\mbox{ and }}\qquad
R:=\frac{C_\sharp\,\mu}{{\varepsilon}^{\frac1s}},\end{aligned}$$ and we slide it from below till it touches the graph of $u$ (by choosing conveniently the free parameters such that ${\varepsilon}\ll \delta\ll\mu\ll1$). As a matter of fact, we have that , and are satisfied. Consequently, we are in the position of using Lemma \[RCAIcgaue023\] and deduce from that $$\label{MPLE1}
\begin{split}
{\mbox{if~$E^{(\beta,\varpi,R)}\setminus{\mathcal{D}}$
is below the graph of~$u$,}}\\
{\mbox{then~$E^{(\beta,\varpi,R)}$
is below the graph of~$u$.}}
\end{split}$$ where ${\mathcal{D}}:=\{|x_1|<R/2\}\cap\{x_2\in(0,\mu)\}$.
Now we claim that $$\label{MPLE2}
{\mbox{$E^{(\beta,\varpi,R)}\setminus{\mathcal{D}}$
lies below the graph of~$u$.}}$$ Indeed, if $x_2{\leqslant}-L$, or $x_2{\geqslant}L$, or $|x_1|{\geqslant}R$, then the result is obvious. Hence, to prove , we can focus on the region $\{|x_1|<R\}\times\{|x_2|<L\}$.
If $x_2\in(-L,0)$, we notice that $$\label{Kytilsq} 2^{\tilde k_0}{\geqslant}2^{\frac{|\log{\varepsilon}|}{c_0}-1}=2L,$$ thanks to . As a consequence, $|x|{\leqslant}|x_1|+|x_2|<
R+L<2L{\leqslant}2^{\tilde k_0}$, and thus we can exploit and find that $$|u(x)|{\leqslant}{\varepsilon}^{\frac1{c_0}}|x|^{1+\alpha}.$$ Combining this with , and noticing that $|x_1|<R=\frac{C_\sharp\,\mu}{{\varepsilon}^{\frac1s}}<2^{\tilde k_0}$, we see that $$|u(x)|{\leqslant}{\varepsilon}^{\frac1{c_0}}|x|^{1+\alpha}\chi_{(-\infty,h)}(x_2),$$ and, as a result, $$\label{9423331236d65}
\begin{split}
&\beta(x_2)-\varpi\,\Phi(x_1)-u(x){\leqslant}{\varepsilon}\delta x_2-\varpi\,\Phi(x_1)+{\varepsilon}^{\frac1{c_0}}|x|^{1+\alpha}\chi_{(-\infty,h)}(x_2)\\
&\qquad{\leqslant}- {\varepsilon}\delta |x_2|-\varpi\,\Phi(x_1)+
2^{\frac{1+\alpha}2}{\varepsilon}^{\frac1{c_0}}\big(|x_1|^{1+\alpha}\chi_{(-\infty,h)}(x_2)+|x_2|^{1+\alpha}\big)\\
&\qquad{\leqslant}{\varepsilon}|x_2|\big( 2^{\frac{1+\alpha}2}L^\alpha{\varepsilon}^{\frac{1-c_0}{c_0}}-\delta\big)
-\varpi\,\Phi(x_1)+
2^{\frac{1+\alpha}2}{\varepsilon}^{\frac1{c_0}} |x_1|^{1+\alpha}\chi_{(-\infty,h)}(x_2)\\
&\qquad= {\varepsilon}|x_2|\left( 2^{\frac{1+\alpha}2}\left(
\frac{1}{4{\varepsilon}^{\frac{\log2}{c_0}}}\right)^\alpha{\varepsilon}^{\frac{1-c_0}{c_0}}-\delta\right)
-\varpi\,\Phi(x_1)+
2^{\frac{1+\alpha}2}{\varepsilon}^{\frac1{c_0}} |x_1|^{1+\alpha}\chi_{(-\infty,h)}(x_2)
\\&\qquad{\leqslant}-\frac{\delta{\varepsilon}}2 |x_2|-\varpi\,\Phi(x_1)+
2^{\frac{1+\alpha}2}{\varepsilon}^{\frac1{c_0}} |x_1|^{1+\alpha}\chi_{(-\infty,h)}(x_2).
\end{split}$$ Now we consider two regimes: when $|x_2|{\leqslant}h$, we deduce from that $$\label{9423331236d65:2}
\begin{split}
&\beta(x_2)-\varpi\,\Phi(x_1)-u(x){\leqslant}-\frac{\delta{\varepsilon}}2 |x_2|<0.
\end{split}$$ If instead $|x_2|>h$, using we infer that $$\label{9423331236d65:3}
\begin{split}
&\beta(x_2)-\varpi\,\Phi(x_1)-u(x){\leqslant}-\frac{\delta{\varepsilon}}2 h+
2^{\frac{1+\alpha}2}{\varepsilon}^{\frac1{c_0}} |x_1|^{1+\alpha}\\
&\qquad{\leqslant}- {\varepsilon}\left(
\frac{\delta}2 h-
2^{\frac{1+\alpha}2}{\varepsilon}^{\frac{1-c_0}{c_0}}R^{1+\alpha}\right)=-
{\varepsilon}\left(
\frac{\delta}2 h-
2^{\frac{1+\alpha}2}
C_\sharp^{1+\alpha}\,\mu^{1+\alpha}
{\varepsilon}^{\frac{1-c_0}{c_0}-
\frac{1+\alpha}s
}\right)\\&\qquad{\leqslant}-\frac{\delta{\varepsilon}h}{4}<0.
\end{split}$$ In view of and , we conclude that is satisfied when $x_2\in(-L,0)$ and $|x_1|<R$.
[F]{}rom these considerations, we see that it is enough now to check with $x_2\in(0,\mu)$ and $|x_1|\in(R/2,R)$, and with $x_2\in(\mu,L)$ and $|x_1|<R$.
If $x_2\in(0,\mu)$ and $|x_1|\in(R/2,R)$, in light of we have that $$\begin{aligned}
&&\beta(x_2)-\varpi\,\Phi(x_1)-u(x){\leqslant}{\varepsilon}\delta x_2+{\varepsilon}a x_2 -\varpi\,\Phi(x_1)-
{\varepsilon}\,\bar a\, x_2^{\frac{1+s}2}+ \overline{C}{\varepsilon}\,\big(|x|^{\frac{3+s}2}+\delta\big)
\\&&\qquad
{\leqslant}{\varepsilon}\delta x_2+{\varepsilon}a x_2 -\varpi\,\Phi(x_1)-
{\varepsilon}\,\bar a\,x_2^{\frac{1+s}2}+ \overline{C}{\varepsilon}\delta+
2^{\frac{3+s}4}\,
\overline{C}{\varepsilon}\,\big(x_2^{\frac{3+s}2}+|x_1|^{\frac{3+s}2}\big)\\&&
\qquad
{\leqslant}{\varepsilon}\,x_2^{\frac{1+s}2}\big( (\delta+a)\mu^{\frac{1-s}2}+
2^{\frac{3+s}4}\,\overline{C}\mu
-\bar a\big)
+ \overline{C}{\varepsilon}\delta-\big(\varpi\,\Phi(x_1)-
2^{\frac{3+s}4}\,
\overline{C}{\varepsilon}\, |x_1|^{\frac{3+s}2}\big)\\&&
\qquad
{\leqslant}{\varepsilon}\left(\overline{C}\delta-\left(\frac{C_\star}{\mu^s}\,\Phi(x_1)-
2^{\frac{3+s}4}\,
\overline{C}\, |x_1|^{\frac{3+s}2}\right)\right)\\&&
\qquad
{\leqslant}{\varepsilon}\left(\overline{C}\delta-\left(\frac{C_\star}{\mu^s}-
2^{\frac{3+s}4}\,
\overline{C}\right)|x_1|^{\frac{3+s}2}\right)
\\&&
\qquad
{\leqslant}{\varepsilon}\left(\overline{C}\delta-\frac{C_\star}{2\mu^s}\,\left(\frac{R}2\right)^{\frac{3+s}2}\right)<0.\end{aligned}$$ This checks when $x_2\in(0,\mu)$ and $|x_1|\in(R/2,R)$, and therefore we are only left with the case in which $x_2\in(\mu,L)$ and $|x_1|<R$.
In this setting, we distinguish when $x_2\in(\mu,\lambda)$ and $|x_1|<R$ and when $
x_2\in(\lambda,L)$ and $|x_1|<R$.
Then, when $x_2\in(\mu,\lambda)$ and $|x_1|<R$, we make use of and we see that $$\begin{aligned}
&&\beta(x_2)-\varpi\,\Phi(x_1)-u(x){\leqslant}{\varepsilon}\delta x_2+{\varepsilon}a x_2-\varpi\,\Phi(x_1)
-
{\varepsilon}\,\bar a\, x_2^{\frac{1+s}2}+ \overline{C}{\varepsilon}\,\big(|x|^{\frac{3+s}2}+\delta\big)\\&&
\qquad
{\leqslant}{\varepsilon}(\delta + a) x_2-\varpi\,\Phi(x_1)
-
{\varepsilon}\,\bar a\, x_2^{\frac{1+s}2}+ 2^{\frac{3+s}4}\overline{C}{\varepsilon}\,\big(
|x_1|^{\frac{3+s}2}+x_2^{\frac{3+s}2}\big)+
\overline{C}{\varepsilon}\delta
\\&&
\qquad
={\varepsilon}x_2^{\frac{1+s}2}\left(
(\delta + a) x_2^{\frac{1-s}2}+
2^{\frac{3+s}4}\overline{C}x_2
+\frac{\overline{C}\delta}{
x_2^{\frac{1+s}2}}
-\bar a \right)+ 2^{\frac{3+s}4}\overline{C}{\varepsilon}\,
|x_1|^{\frac{3+s}2}
-\varpi\,\Phi(x_1)\\&&
\qquad
{\leqslant}{\varepsilon}x_2^{\frac{1+s}2}\left(
(\delta + a) \lambda^{\frac{1-s}2}+
2^{\frac{3+s}4}\overline{C}\lambda
+\frac{\overline{C}\delta}{
\mu^{\frac{1+s}2}}
-\bar a \right)+ {\varepsilon}|x_1|^{\frac{3+s}2}\left(2^{\frac{3+s}4}\overline{C}\,
-\frac{C_\star}{\mu^s}\right)\\
&&
\qquad
{\leqslant}-\frac{{\varepsilon}\bar a}{16} x_2^{\frac{1+s}2}<0,\end{aligned}$$ that gives in this case.
If instead $
x_2\in(\lambda,L)$ and $|x_1|<R$, we exploit . In this case, we claim that $$\label{8UAll12-39jjdna}
|u(x)|{\leqslant}2^{1+\alpha}{\varepsilon}\max\{1,|x|^{1+\alpha}\}.$$ Indeed, if $|x|<1$, we use with $k:=0$ and we obtain . If instead $|x|{\geqslant}1$, we remark that $|x|{\leqslant}|x_1|+|x_2|<R+L<2^{\tilde k_0}$, and therefore we are in the position of applying with $k$ such that $2^{k-1}{\leqslant}|x|{\leqslant}2^k$, and conclude that $$|u(x)|{\leqslant}{\varepsilon}\,(2^k)^{1+\alpha}{\leqslant}2^{1+\alpha}{\varepsilon}\,|x|^{1+\alpha}.$$ This completes the proof of .
Consequently, using , we find that $$\label{ABSbaatiina}
\begin{split}
&\beta(x_2)-\varpi\,\Phi(x_1)-u(x){\leqslant}{\varepsilon}\delta x_2-{\varepsilon}b x_2^{1+\alpha}-\varpi\,\Phi(x_1)+
2^{1+\alpha}{\varepsilon}\max\{1,|x|^{1+\alpha}\}\\
&\qquad{\leqslant}{\varepsilon}\delta x_2-{\varepsilon}b x_2^{1+\alpha}-\varpi\,\Phi(x_1)+
2^{2(1+\alpha)}{\varepsilon}\,\big(1+|x_1|^{1+\alpha}+x_2^{{1+\alpha}}\big)\\
&\qquad{\leqslant}{\varepsilon}x_2^{1+\alpha}
\left(\frac\delta{x_2^\alpha}+
2^{2(1+\alpha)}- b
\right)+2^{2(1+\alpha)}{\varepsilon}+{\varepsilon}\left(
2^{2(1+\alpha)} |x_1|^{1+\alpha}-\frac{C_\star}{\mu^s}\Phi(x_1)
\right).
\end{split}$$ Now we consider the function $$[0,+\infty)\ni t\mapsto g(t):= t^{1+\alpha}-K t^{\frac{3+s}2},$$ for a given $K>0$, and we claim that $$\label{y7so239dj}
g(t){\leqslant}\frac{1}{K^{\frac{2(1+\alpha)}{1+s-2\alpha}}}.$$ Indeed, if $t{\geqslant}\frac{1}{K^{\frac{2}{1+s-2\alpha}}}$, we see that $$g(t)=t^{1+\alpha}\left(1-K t^{\frac{1+s-2\alpha}2}\right) {\leqslant}0,$$ and so is satisfied. If instead $t\in\left[0,\frac{1}{K^{\frac{2}{1+s-2\alpha}}}\right)$ we have that $$g(t){\leqslant}t^{1+\alpha}{\leqslant}\frac{1}{K^{\frac{2(1+\alpha)}{1+s-2\alpha}}},$$ proving also in this case.
Now, from , $$\begin{aligned}
&&
2^{2(1+\alpha)} |x_1|^{1+\alpha}-\frac{C_\star}{\mu^s}\Phi(x_1)=
2^{2(1+\alpha)} \left(|x_1|^{1+\alpha}-\frac{C_\star}{2^{2(1+\alpha)}\mu^s}
|x_1|^{\frac{3+s}2}\right)\\
&&\qquad{\leqslant}2^{2(1+\alpha)}\left( \frac{2^{2(1+\alpha)}\mu^s}{C_\star}\right)^{\frac{2(1+\alpha)}{1+s-2\alpha}}
= \tilde C\,\mu^{\frac{2s(1+\alpha)}{1+s-2\alpha}},\end{aligned}$$ for a suitable $\tilde C>0$.
Plugging this information into , we conclude that $$\begin{aligned}
&&
\beta(x_2)-\varpi\,\Phi(x_1)-u(x){\leqslant}{\varepsilon}x_2^{1+\alpha}
\left(\frac\delta{x_2^\alpha}+
2^{2(1+\alpha)}- b
\right)+2^{2(1+\alpha)}{\varepsilon}+\tilde C\,{\varepsilon}\mu^{\frac{2s(1+\alpha)}{1+s-2\alpha}}\\&&\qquad
{\leqslant}{\varepsilon}x_2^{1+\alpha}
\left(\frac\delta{\lambda^\alpha}+
2^{2(1+\alpha)}- b
\right)+2^{2(2+\alpha)}{\varepsilon}{\leqslant}-\frac{2^{4(2+\alpha)}{\varepsilon}x_2^{1+\alpha}}{\lambda^{1+\alpha}}+2^{2(2+\alpha)}{\varepsilon}\\&&\qquad
{\leqslant}-2^{4(2+\alpha)}{\varepsilon}+2^{2(2+\alpha)}{\varepsilon}<0,\end{aligned}$$ and this completes the proof of .
Then, and give that $E^{(\beta,\varpi,R)}$ is below the graph of $u$.
As a result, if $x_2\in(0,\lambda)$ and $|x_1|<R/2$, $${\varepsilon}\delta x_2+\frac{{\varepsilon}\bar a}8 x_2-\varpi\Phi(x_1)
=
\bar\ell x_2+{\varepsilon}a x_2-\varpi\Phi(x_1)=\beta(x)-\varpi\Phi(x_1){\leqslant}u(x).$$ Consequently, if we consider the second blow-up of the graph of $u$, as in Lemma \[BOAmdfiUAMP\], we conclude that $$\left\{x_{3}< \left({\varepsilon}\delta +\frac{{\varepsilon}\bar a}8\right) x_2
\right\}\cap \{x_2>0\} \subseteq E_{00}.$$ [F]{}rom this and Corollary \[COR:ALTE\], we conclude that and cannot hold true, and therefore necessarily must be satisfied, namely $E_{00}\cap\{x_2>0\}=\{x_2>0\}$ and $$\lim_{x_2\searrow0} u(0,x_2)>0.$$ The latter inequality is in contradiction with and therefore this completes the proof of the desired claim in .
Now, we can complete the proof of Theorem \[NLMS\] :
The core of the proof is to establish , since would then easily follow.
To this end, from Theorem \[ijsd78AIsjjd\], arguing as in Theorem 8.2 of [@2019arXiv190405393D], and exploiting Corollary \[COR:ALTE\] here (instead of Theorem 4.1 in [@2019arXiv190405393D]) to obtain (8.19) in [@2019arXiv190405393D], one establishes with $\frac{3+s}2$ replaced by $1+\alpha$, for a given $\alpha\in(0,s)$.
Then, to improve this regularity exponent and complete the proof of , one can proceed as in the proof of Theorem 1.2 of [@2019arXiv190405393D].
A classical counterpart of Theorem \[FLFL\] {#APPA}
===========================================
In this appendix, we show that Theorem \[FLFL\] possesses a classical counterpart for the Laplace equation, namely:
\[FLFLCLA\] Let $n{\geqslant}2$, $k\in{\ensuremath{\mathbb{N}}}$ and $f\in C^k(\overline{B_1'})$. Then, for every ${\varepsilon}>0$ there exist $f_{\varepsilon}\in C^k({\ensuremath{\mathbb{R}}}^{n-1})$ and $u_{\varepsilon}\in C({\ensuremath{\mathbb{R}}}^n)$ such that $$\begin{cases}
\Delta u_{\varepsilon}=0 & {\mbox{ in }} B_1\cap \{x_n>0\},\\
u_{\varepsilon}=0 & {\mbox{ in }} B_1\cap\{x_n=0\},\end{cases}$$ $$\displaystyle\lim_{x_n\searrow0} \frac{u_{\varepsilon}(x',x_n)}{x_n}=f_{\varepsilon}(x')\qquad{\mbox{ for all }}
x'\in B_1',$$ and $$\| f_{\varepsilon}-f\|_{C^k(\overline{B_1'})}{\leqslant}{\varepsilon}.$$
We observe that formally Theorem \[FLFLCLA\] corresponds to Theorem \[FLFL\] with $\sigma:=1$. The arguments that we proposed for $\sigma\in(0,1)$ can be carried out also when $\sigma=1$, and thus prove Theorem \[FLFLCLA\]. Indeed, for the classical Laplace equation, one can obtain Lemma \[BAN\] by taking, for instance, the real part of the holomorphic function ${\ensuremath{\mathbb{C}}}\ni z\mapsto z^k$, with $k\in{\ensuremath{\mathbb{N}}}$, and then use this homogeneous solution in the proof of Lemma \[fan\]. Then, once Lemma \[fan\] is proved with $\sigma:=1$, one can exploit the proof of Theorem \[FLFL\] presented on page with $\sigma:=1$ and obtain Theorem \[FLFLCLA\].
However, in the classical case there is also an explicit polynomial expansion which recovers Lemma \[fan\] with $\sigma:=1$, that is when is replaced by $$\begin{cases}
\Delta u=0 & {\mbox{ in }} B_r\cap \{x_n>0\},\\
u=0 & {\mbox{ in }} B_r\cap\{x_n=0\}.\end{cases}$$
Hence, to establish Theorem \[FLFLCLA\], we focus on the following argument of classical flavor (which is not reproducible for $\sigma\in(0,1)$ and thus requires new strategies in the case of Theorem \[FLFL\]).
Given $\kappa=\big(\kappa_\gamma
\big)_{{\gamma\in{\ensuremath{\mathbb{N}}}^{n-1}}\atop{|\gamma|{\leqslant}k}}$ in ${\ensuremath{\mathbb{R}}}^{N_k}$, where $N_k$ is defined in , we aim at finding a function $u\in C(\overline{B_1})$ such that $$\label{QLA} \begin{cases}
\Delta u=0 & {\mbox{ in }} B_1\cap \{x_n>0\},\\
u=0 & {\mbox{ in }} B_1\cap\{x_n=0\},\end{cases}$$ and for which there exists $\phi\in C^k(\overline{B'_1})$ such that $$\label{uvA4}
\lim_{x_n\searrow0} \frac{u(x',x_n)}{x_n}=\phi(x')\qquad{\mbox{ for all }}
x'\in B_1',$$ with, in the notation of , $$\label{78ssdJA}
{\mathcal{D}}^k\phi(0)=\kappa .$$ To this end, we define $\phi$ to be the polynomial $$\phi(x'):= \sum_{{\gamma\in{\ensuremath{\mathbb{N}}}^{n-1}}\atop{|\gamma|{\leqslant}k}}
\frac{\kappa_\gamma}{\gamma!} (x')^\gamma.$$ In this way, is automatically satisfied. Furthermore, we observe that $\Delta_{x'}^j\phi$ vanishes identically when $2j>k$, and therefore we can set $$u(x',x_n):=\sum_{{j\in{\ensuremath{\mathbb{N}}}}\atop{j{\leqslant}k/2}} \frac{(-1)^j\,
\Delta_{x'}^j \phi(x')}{(2j+1)!}\,x_n^{2j+1}
=
\sum_{j=0}^{+\infty} \frac{(-1)^j\,
\Delta_{x'}^j \phi(x')}{(2j+1)!}\,x_n^{2j+1}.$$ We observe that $$\lim_{x_n\searrow0} \frac{u(x',x_n)}{x_n}=
\lim_{x_n\searrow0}
\sum_{{j\in{\ensuremath{\mathbb{N}}}}\atop{j{\leqslant}k/2}} \frac{(-1)^j\,
\Delta_{x'}^j \phi(x')}{(2j+1)!}\,x_n^{2j}=
\phi(x'),$$ which establishes .
Also, we see that $u(x',0)=0$, and moreover, using the notation $J:=j-1$, $$\begin{aligned}
\Delta u(x)&=&\Delta_{x'} u(x)+\partial^2_{x_n}u(x)\\
&=&
\sum_{j=0}^{+\infty} \frac{(-1)^j\,
\Delta_{x'}^{j+1} \phi(x')}{(2j+1)!}\,x_n^{2j+1}
+
\sum_{j=1}^{+\infty} \frac{(-1)^j\,
\Delta_{x'}^j \phi(x')}{(2j-1)!}\,x_n^{2j-1}\\
&=&
\sum_{j=0}^{+\infty} \frac{(-1)^j\,
\Delta_{x'}^{j+1} \phi(x')}{(2j+1)!}\,x_n^{2j+1}
+
\sum_{J=0}^{+\infty} \frac{(-1)^{J+1}\,
\Delta_{x'}^{J+1} \phi(x')}{(2J+1)!}\,x_n^{2J+1}\\&=&0,\end{aligned}$$ which gives .
Acknowledgments {#acknowledgments .unnumbered}
===============
The first and third authors are member of INdAM and are supported by the Australian Research Council Discovery Project DP170104880 NEW “Nonlocal Equations at Work”. The first author’s visit to Columbia has been partially funded by the Fulbright Foundation and the Australian Research Council DECRA DE180100957 “PDEs, free boundaries and applications”. The second author is supported by the National Science Foundation grant DMS-1500438. The third author’s visit to Columbia has been partially funded by the Italian Piano di Sostegno alla Ricerca “Equazioni nonlocali e applicazioni”.
References {#references .unnumbered}
==========
[^1]: We observe that drawing the trace along $\{x_2=0\}$ in Figure \[DOMA\] in a communicative way is not completely easy, since the vertical tangencies make the nonlocal minimal surface in $\{x_2>0\}$ hide its own trace.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Building an accurate automatic speech recognition (ASR) system requires a large dataset that contains many hours of labeled speech samples produced by a diverse set of speakers. The lack of such open free datasets is one of the main issues preventing advancements in ASR research. To address this problem, we propose to augment a natural speech dataset with synthetic speech. We train very large end-to-end neural speech recognition models using the LibriSpeech dataset augmented with synthetic speech. These new models achieve state of the art Word Error Rate (WER) for character-level based models without an external language model.'
author:
- |
Jason Li, Ravi Gadde[^1] , Boris Ginsburg, Vitaly Lavrukhin\
NVIDIA\
Santa Clara, CA 95051\
`{jasoli,rgadde,bginsburg,vlavrukhin}@nvidia.com`\
bibliography:
- 'nips\_2018.bib'
title: Training Neural Speech Recognition Systems with Synthetic Speech Augmentation
---
Introduction {#intro}
============
There has been a large amount of success in using neural networks (NN) for automatic speech recognition (ASR). Classical ASR systems use complicated pipelines with many heavily engineered processing stages, including specialized input features, acoustic models, and Hidden Markov Models (HMMs). Deep NNs have traditionally been used for acoustic modeling [@Waibel1989; @Hinton2012]. A key breakthrough occurred when a state of the art ASR system, Deep Speech, was built using an end-to-end deep learning approach [@DeepSpeech2014]. This model replaced both acoustic modeling, and HMMs with very deep neural networks and was able to directly translate spectrograms into English text transcriptions. This end-to-end approach was extended in follow-up papers [@Amodei2016DeepS2; @collobert2016wav2letter]. These NNs have been shown to achieve human-level performance for some tasks [@Amodei2016DeepS2]. Recent advances in ASR have further improved upon these models using even more advanced techniques such as replacing n-gram language models with a neural language model, usually in the form of a recurrent neural network (RNN) [@ASRAttention2018; @Povey2018SemiOrthogonalLM; @CAPIO2017].
As opposed to making neural networks more complex, we were interested in an orthogonal direction of study: whether we can improve quality by creating larger models. In order to train such large models, deep NNs require a vast quantity of data to be available. However the lack of a large public speech dataset blocked us from successfully building large NN models for ASR. We were inspired by recent work in improving translation systems using synthetic data [@FacebookGERENGBackTranslation], and as such, we decided to augment speech with synthetic data. [^2] Furthermore, given the recent impressive improvement in neural speech synthesis models [@WaveNet; @shen2018natural], it becomes cheap to generate high quality speech with varying prosody.
We show that by using synthetic speech created from a neural speech synthesis model, we can improve ASR performance compared to models trained using only LibriSpeech data [@panayotov2015librispeech]. By naively increasing the depth of the model, we observe that the synthetic data allows us to achieve state of the art WER using a greedy decoder.
Synthetic Speech Dataset {#creation}
========================
We use the Tacotron-2 like model from the OpenSeq2Seq [^3] toolkit [@OpenSeq2Seq] and add Global Style Tokens (GST) [@GlobalStyleTokens] to learn multiple speaker identities. Tacotron-2 with GST (T2-GST) was trained on the MAILABS English-US dataset [@mailabs] with approximately 100 hours of audio recorded by 3 speakers. T2-GST was able to learn all 3 different speaking styles and different accents that they use to portray different characters.
Using the T2-GST model, we created a fully synthetic version of the LibriSpeech training audio. In order to produce audio, T2-GST requires a spectrogram used for the style and the audio transcription. For the transciption, we took the transcripts from the train-clean-100, train-clean-360, and train-other-500 LibriSpeech splits and randomly paired them with style spectrograms from the MAILABS English-US dataset. At the end, we had a dataset that was the same size as the original training portion of the LibriSpeech dataset but spoken in the tones of the speakers from the MAILABS dataset.
The audio from the T2-GST model could be further controlled by the amount of dropout in the prenet of the decoder. By decreasing the dropout rate, we found that we could slightly distort the audio. The main difference that we noticed was that the lower the dropout rate, the faster the resulting audio would sound. Using this observation, we further increased the size of the synthetic dataset. Thus, we used 46%, 48%, and 50% for the dropout and created a synthetic speech dataset that was 3 times as large as the LibriSpeech training dataset.
Training Speech Recognition with Synthetic Data {#results}
===============================================
Neural Speech Recognition Models
--------------------------------
Our speech recognition model is an end-to-end neural network that takes logarithmic mel-scale spectrograms as inputs and produces characters. We use a deep convolutional NN model, which we will further address as Wave2Letter+ (w2lp) [^4] . It is based on Wav2Letter [@collobert2016wav2letter] except:
- We use ReLU instead of Gated Linear Unit
- We use batch normalization instead of weight normalization
- We add residual connections between convolutional blocks
- We use Connectionist Temporal Classification loss instead of Auto Segmentation Criterion
- We use Layer-wise Adaptive Rate clipping (LARC) for gradient clipping
{width="14cm"}
The base Wave2Letter+ model has 19 convolutional layers:
- 1 pre-processing layer at the beginning of the network
- 15 layers consists of 5 blocks of 3 repeating convolutional layers
- 3 post-processing layers at the end of the network
We made the base model deeper and experimented with 24, 34, 44, and 54 layer networks. The 24 layer networks consists of 5 blocks of 4 repeating convolutional layers. The 34, 44, and 54 layer network consists of 10 blocks of 3, 4, and 5 repeating layers respectively.
Word Error Rate Improvement using Synthetic Augmentation
--------------------------------------------------------
The training dataset was created by combining the synthetic data with the original training data from LibriSpeech. Despite the synthetic data being larger than the natural data, we found it most helpful to sample natural and synthetic data at a 50/50 rate.
Our best performing model, the 54 layer network, currently has a word error rate of 4.32% on test-clean and a word error rate of 14.08% on test-other by greedily choosing the most probable character at each step without any language model rescoring. To the best of our knowledge, the previous best performing model without using any language model achieves 4.87% on test-clean and 15.39% on test-other [@ASRAttention2018]. The complete results can be found in Table \[LibriVsSyn\]. Models trained on the combined dataset outperform those trained on original LibriSpeech. The augmented dataset improves results on test-clean by 0.15, and 0.44 for the 24 and 34 layer models. For test-other, we see an improvement of 0.08, and 0.74 for the 24 and 34 layer models.
By using beam search with beam width 128 and the 4-gram OpenSLR [^5] language model rescoring, we improved our WER on test-other to 12.21% on the 54 layer model which is better than previous public 4-gram language models and comparable to LSTM language models [@ASRAttention2018].
------------------------ -------------- ------- ------- ------- -------
Model Dataset Used
Clean Other Clean Other
attention-Zeyer et al. LibriSpeech 4.87 14.37 4.87 15.39
w2lp-24 LibriSpeech 5.44 16.57 5.31 17.09
w2lp-24 Combined 5.12 16.25 5.16 17.01
w2lp-34 LibriSpeech 5.10 15.49 5.10 16.21
w2lp-34 Combined 4.60 14.98 4.66 15.47
w2lp-44 Combined 4.24 13.87 4.36 14.37
w2lp-54 Combined 4.32 13.74 4.32 14.08
\[1ex\]
------------------------ -------------- ------- ------- ------- -------
: Greedy WER on LibriSpeech for Different Models and Datasets[]{data-label="LibriVsSyn"}
How To Mix Natural and Synthetic Speech
---------------------------------------
We performed a number of additional experiments to find the best sampling ratio between synthetic data and LibriSpeech. We tested training on only LibriSpeech, a 50/50 split, a 33/66 split, and a pure synthetic dataset. All tests were done on the 34 layer model. The results are shown in Table \[SampleProb\].
--------- --------------------- ------- ------- ------- -------
Model Sampling Ratio
(Natural/Synthetic) Clean Other Clean Other
w2lp-34 Natural 5.10 15.49 5.10 16.21
w2lp-34 50/50 4.60 14.98 4.66 15.47
w2lp-34 33/66 4.91 15.18 4.81 15.81
w2lp-34 Synthetic 51.39 80.27 49.80 81.78
\[1ex\]
--------- --------------------- ------- ------- ------- -------
: Greedy WER for Different Ratios Between Natural and Synthetic Datasets[]{data-label="SampleProb"}
Despite the larger amount of synthetic data, the synthetic dataset fails to capture the larger variety of LibriSpeech. We believe that this effect could be moderated if a speech synthesis model with larger speaker variety was used as opposed to the current 3 speaker speech synthesis model. A 50/50 split between the natural and synthetic seems to be a good ratio for our dataset.
Traditional Speech Augmentation vs Synthetic Speech
---------------------------------------------------
Adding synthetic data proved to be better regularization than standard regularization techniques. In addition to dropout which is employed for all models, OpenSeq2Seq supports speech augmentation such as adding noise and time stretching. Using these 3 techniques, we tested 4 additional models. We tested 2 larger dropout factors, and on top on this, we tested whether speech augmentation would improve performance. Since the dropout factor varies by layer, we multiply the local dropout keep probabilities by a global dropout keep factor. All tests were done on the 34 layer model. The tests and results are detailed in Table \[reg\].
A slightly larger dropout resulted in minor improvement in WER. The effects of speech augmentation seem to be negligible or, in the case of large dropout, make WER worse. Adding synthetic data significantly outperforms all other methods of regularization.
--------------- ------------- -------------- -------------- ------- ------- ------- -------
Model Dropout Time Stretch Noise
Keep Factor Factor (dB) Clean Other Clean Other
LibriSpeech None None None 5.10 15.49 5.10 16.21
Dropout 0.9 None None 5.01 15.15 5.15 15.70
Dropout + Aug 0.9 0.05 \[-90, -60\] 5.07 15.00 5.02 15.83
Dropout 0.75 None None 5.46 15.77 5.39 16.62
Dropout + Aug 0.75 0.1 \[-90, -60\] 5.80 16.33 5.72 17.41
Combined None None None 4.60 14.98 4.66 15.47
--------------- ------------- -------------- -------------- ------- ------- ------- -------
: Greedy WER on Using Different Regularization Techniques[]{data-label="reg"}
Conclusion and Future Plans
===========================
Using synthetic data is an effective way to build large neural speech recognition systems. The synthetic data should be combined with the natural data in the correct ratio to obtain best results. With this method, we achieved a WER of 4.32% on test-clean and a WER of 14.08% on test-other using a greedy decoder. This is the current state of the art on character-level based greedy decoding. Furthermore, using a language model and a beam search width of 128, we get 12.21% WER on test-other.
For future studies, we are interested in creating a larger synthetic dataset with noise. For now, we have restricted ourselves to take transcriptions from the training subsets of LibriSpeech, but the speech synthesis models are general enough to accept any transcript. It would be interesting to scrape text from another source and add to the training set additional phrases not found in LibriSpeech.
### Acknowledgments {#acknowledgments .unnumbered}
We would like to thank Rafael Valle, Ryan Leary, Igor Gitman, Oleksii Kuchaiev and Ujval Kapasi for helpful conversation and advice throught this project.
[^1]: work done while the author was an intern at NVIDIA
[^2]: Synthesized speech has previously been used to improve speech recognition for low-resource languages [@Rygaard2015].
[^3]: We used OpenSeq2Seq both to create a synthetic speech dataset and build ASR.
[^4]: More model details: [https://nvidia.github.io/OpenSeq2Seq/html/speech-recognition/wave2letter.html](OS2S)
[^5]: The LM can be found here: www.openslr.org/11
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'It is showed how the Hamiltonian lattice $loop$ $representation$ can be cast straightforwardly in the Lagrangian formalism. The procedure is general and here we present the simplest case: pure compact QED. This connection has been shaded by the non canonical character of the algebra of the fundamental loop operators. The loops represent tubes of electric flux and can be considered the dual objects to the Nielsen-Olesen strings supported by the Higgs broken phase. The lattice loop classical action corresponding to the Villain form is proportional to the quadratic area of the loop world sheets and thus it is similar to the Nambu string action. This loop action is used in a Monte Carlo simulation and its appealing features are discussed.'
---
UAB-FT-341\
June 1994
[**THE LAGRANGIAN LOOP REPRESENTATION .3cm OF LATTICE U(1) GAUGE THEORY**]{}
[ [ **J. M. Aroca** ]{} 0.5cm Departament de Matemàtiques,\
Universitat Politècnica de Catalunya,\
Escola Tècnica Superior d’Enginyers de Telecomunicació, 08034\
Barcelona, Spain \
and \
[**M. Baig and H. Fort**]{} 0.5cm Grup de Física Teòrica\
and\
Institut de Física d’Altes Energies\
Universitat Autònoma de Barcelona\
08193 Bellaterra (Barcelona) Spain]{}
Introduction
============
A unified quantum theory which describes the gauge fields and the gravitation is one of the main goals pursued by the physicists for long time. A good candidate for accomplishing this comprehensive framework is the [*loop representation*]{}. This loop approach was introduced in the early eighties by Gambini and Trias [@gt],[@gt1] as a Hamiltonian representation of gauge theories in terms of their natural physical excitations: the loops. The original aim of this general analytical Hamiltonian approach for gauge theories was to avoid the redundancy introduced by the gauge symmetry working directly in the space of physical states. However, soon it was realized that the loop formalism goes far beyond of a simple gauge invariant description. The introduction by Ashtekar [@a] of a new set of variables that cast general relativity in the same language as gauge theories allowed to apply loop techniques as a natural non-perturbative description of Einstein’s theory. Furthermore, the loop representation appeared as the most appealing application of the loop techniques to this problem [@rs],[@gam].
The Hamiltonian techniques for gauge theories have been developed during the last decade and they provide interesting results for several lattice models [@gt2]-[@af]. On the other hand a Lagrangian approach in terms of loops has been elusive, due mainly to the non-canonical character of the loop algebra. This feature forbids the possibility of performing a Legendre transformation as a straightforward way to obtain the Lagrangian from the Hamiltonian.
In the case of non-Abelian gauge theory a major problem has been whether we can write a reasonably simple Lagrangian in terms of “electric vector potentials” [@Ma]. A Lagrangian loop formulation will give rise to new computation techniques providing a a useful complement to the Hamiltonian loop studies.
Recently, it was proposed a tentative classical action in terms of loop variables for the U(1) gauge theory [@aggs]. Shortly afterwards, we proved that the lattice version of this action is equivalent to Villain form for D=2+1 dimensions but is slightly different for D=3+1 dimensions [@af1]. In fact, this action written in terms of variables directly attached to spatial loops seems to fail in describing all the dynamical degrees of freedom for D=4.
This paper is organized as follows. In section 2 we show how the loops, originally thought up within the Hamiltonian formalism, can be introduced in a natural way in the lattice Lagrangian theory. We follow a different approach to that of reference [@af1]: we show how the electric loops can be traced in the statistical lattice formulation of $4d$ $U(1)$ theory giving rise to an expression of the partition function as a sum of integer closed surfaces. We interpret these surfaces as the world sheets of the electric loops. We discuss the connection of this classical loop action with the Nambu string action. A clear analogy is patent between the gauge theory in the loop representation and the bosonic and fermionic strings, as it was previously suspected [@P], [@gn] but never (as far as we know) demonstrated explicitly. The parallelism of the loop representation with the topological representation of the broken Higgs phase in terms of Nielsen-Olesen strings [@no] is also pointed out. In section 3 we use the loop action equivalent to the Villain form for performing a Monte Carlo simulation. It turns out that this action is the same as the $\gamma \rightarrow \infty$ limit for the non-compact Abelian-Higgs theory [@bd] ($\gamma$ : Higgs coupling constant).
The Lagrangian loop Representation
==================================
The loop based approach of ref.[@gt] describes the quantum electrodynamics in terms of the gauge invariant holonomy (Wilson loop) $$\hat{W} (\gamma ) = \exp [i e \oint_{\gamma} A_a (y) dy^a] ,
\label{eq:Wloop}$$ and the conjugate electric field $\hat{E}^a (x)$ . They obey the commutation relations $$[\hat{E}^a(x), \hat{W} (\gamma)] = e \int_\gamma \delta(x - y) dy^a \hat{W}
(\gamma) .
\label{eq:alg}$$
These operators act on a state space of abelian loops $\psi(\gamma)$ that may be expressed in terms of the transform
$$\psi (\gamma) = \int d_\mu [A] <\gamma \mid A> <A \mid \psi>
= \int d_\mu [A] \psi [A] \exp [- i e \oint_{\gamma} A_a dy^a] .
\label{eq:trans}$$
This loop representation has many appealing features: In first place, it allows to do away with the first class constraints of gauge theories. That is, the Gauss law is automatically satisfied. In second place, the formalism only involves gauge invariant objects. Finally, all the gauge invariant operators have a transparent geometrical meaning when they are realized in the loop space.
When this loop representation is implemented in the lattice it offers a gauge invariant description of physical states in terms of kets $\mid C >=\hat{W}(C) \mid 0>$, where $C$ labels a closed path in the [*spatial*]{} lattice. Eq.(\[eq:alg\]) becomes $$[\hat{E}_l,\hat{W}(C)] = N_l(C)\hat{W}(C),
\label{eq:alg1}$$ where $l$ denotes the links of the lattice, $\hat{E}(l)$ the lattice electric field operator, $\hat{W}(C)=\prod_{l\in C}\hat{U}(l)$ and $N_l(C)$ is the number of times that the link $l$ appears in the closed path $C$.
In this loop representation, the Wilson loop acts as the loop creation operator:
$$\hat{W}(C')\mid C> = \mid C'\cdot C>.
\label{eq:Wloop1}$$
The physical meaning of an abelian loop may be deduced from (\[eq:alg1\]) and (\[eq:Wloop1\]), in fact $$\hat{E}_l \mid C> = N_l(C) \mid C>,
\label{eq:E}$$ which implies that $\mid C>$ is an eigenstate of the electric field. The corresponding eigenvalue is different from zero if the link $l$ belongs to $C$. Thus $C$ represents a confined line of electric flux.
In order to cast the loop representation in Lagrangian form it is convenient to use the language of differential forms on the lattice of ref.[@g]. Besides the great simplifications to which this formalism lead its advantages consists in the general character of the expressions obtained. That is, most of the transformations are independent on the space-time dimension or on the rank of the fields. So let us sumarize the main concepts and some useful results of the formalism of differential forms on the lattice.
A k-form is a function defined on the k-cells of the lattice (k=0 sites, k=1 links, k=2 plaquettes, etc.) over an abelian group which shall be [**R**]{}, [**Z**]{}, or U(1)=[reals module 2$\pi$]{}.
Integer forms can be considered geometrical objects on the lattice. For instance, a 1-form represents a path and the integer value on a link is the number of times that the path traverses this link.
Let us introduce $\nabla$ is the co-border operator which maps k-forms onto (k+1)-forms. It is the gradient operator when acting on scalar functions (0-forms) and it is the rotational on vector functions (1-forms). We shall consider the scalar product of p-forms defined $<\alpha \mid \beta> = \sum_{c_k}\alpha (c)\beta (c)$ where the sum runs over the k-cells of the lattice. Under this product the $\nabla$ operator is adjoint to the border operator $\partial$ which maps k-forms onto (k-1)-forms and which corresponds to minus times the usual divergence operator. That is,
$$\begin{aligned}
<\alpha \mid \nabla \beta> = <\partial \alpha \mid \beta>,\\
<\nabla \alpha \mid \beta> = <\alpha \mid \partial \beta>.
\label{eq:inter}\end{aligned}$$
The co-border $\nabla$ and border $\partial$ operators verify
$${\nabla}^2 = 0,\; \; {\partial}^2 = 0.
\label{eq:0sq}$$
The Laplace-Beltrami operator operator is defined by
$$\Box =\nabla \partial +\partial \nabla .
\label{eq:Box}$$
It is a symmetric linear operator which commutes with $\nabla$ and $\partial$, and differs only by a minus sign of the current Laplacian $ \Delta_\mu \Delta_\mu$.
From Eq.(\[eq:Box\]) is easy to show the Hodge-identity:
$$1=\partial {\Box}^{-1}\nabla + \nabla {\Box}^{-1}\partial .
\label{eq:Hodge}$$
A useful tool to consider is the [*duality transformation*]{} which maps biyectively k-forms over (D-k)-forms. We denote by $*p_{c_{D-k}}$ the dual form of the $p_{c_k}$ form. For example, for $D=2$, to plaquettes there correspond sites of the dual lattice, i.e. those vertices obtained from the original ones by a translation of vector (a/2, a/2).
Under duality the border and co-border operators interchange:
$$\partial = *\nabla * .
\label{eq:dualrel}$$
After this digresion about differential forms on the lattice let us consider the generating functional for the Wilson U(1) lattice action: $$Z_W = \int_{-\pi}^{\pi} (d\theta_l )\exp(-\frac{\beta}{2}\sum_p \cos
{\theta}_p),$$ where the subscripts $l$ and $p$ stand for the lattice links and plaquettes respectively.
Fourier expanding the $\exp [ \cos \theta ]$ we get $$Z_W = \int_{-\pi}^{\pi} (d\theta_l ) \prod_p \sum_{n_p} I_{n_p}(\beta )
e^{in_p {\theta}_p},$$ which can be written, using the language of differential forms as $$Z_W = \sum_{ \left\{ n_p \right\} }
\int_{-\pi}^{\pi} (d\theta_l ) \exp (\sum_p \ln I_{n_p}(\beta ) )
e^{i<n , \nabla \theta_l >}.$$
In the above expression, $\theta_l$ is a real periodic 1-form, that is, a real number $\theta \in [-\pi, \pi ]$ defined on each link of the lattice; $\nabla$ is the co-border operator; $n_p$ are integer 2-forms, defined at the lattice plaquettes.
By eq. (\[eq:inter\]) and integrating over $\theta_l$ we obtain a $\delta (\partial n_p)$. Then, $$Z_W \propto \sum_{ \left\{ n_p; \partial n_p = 0 \right\} }
\exp (\sum_p \ln I_{n_p}(\beta ) ),
\label{eq:Wil}$$ the constraint $\partial n_p = 0$ means that the sum is restricted to $closed$ 2-forms. Thus, the sum runs over collections of plaquettes constituting closed surfaces. This expression was obtained by Savit [@s] as an intermediate step towards the $dual$ representation.
An alternative and more easy to handle lattice action than the Wilson form is the Villain form. The partition function of that form is given by $$Z_V = \int (d\theta ) \sum_{ s }\exp
(-\frac{{\beta}_V}{2}\mid\mid \nabla \theta -2\pi s\mid\mid^2),
\label{eq:Villain}$$ where $\mid \mid \ldots \mid \mid^2 = <\ldots , \ldots>$. If we use the Poisson summation formula $$\sum_s f(s) = \sum_n \int_{-\infty}^{\infty} d\phi
f(\phi) e^{2\pi i\phi n}$$ and we integrate the continuum $\phi$ variables we get $$Z_V = (2\pi {\beta}_V)^{-N_p/2} \int (d\theta ) \sum_{n }\exp
(-\frac{1}{2{\beta}_V}<n,n>+i<n,\nabla \theta >),
\label{}$$ where $N_p$ in the number of plaquettes of the lattice. Again, we can use the equality: $<n,\nabla \theta>=<\partial n,\theta >$ and integrating over $\theta$ we obtain a $\delta (\partial n)$. Then, $$Z_V = (2\pi {\beta}_V)^{-N_p/2} \sum_{\left\{ n; \partial n = 0 \right\}}
\exp(-\frac{1}{2{\beta}_V}<n,n>),
\label{eq:Vill}$$ where $n$ are integer 2-forms. Eq. (\[eq:Vill\]) is obtained from Eq.(\[eq:Wil\]) in the $\beta \rightarrow \infty$ limit.
If we consider the intersection of one of such surfaces with a $t=constant$ plane we get a loop $C_t$. It is easy to show that the creation operator of this loop is just the creation operator of the loop representation, namely the Wilson loop operator. Repeating the steps from Eq.(\[eq:Villain\]) to Eq.(\[eq:Vill\]) we get for $<\hat{W}(C_t)>$
$$<W(C_t)> = \frac{1}{Z}(2\pi {\beta}_V)^{-N_p/2}
\sum_{
\begin{array} {c} n\\
\left( \partial n = C_t \right)
\end{array}
}\exp
(-\frac{1}{2{\beta}_V}<n,n>).
\label{eq:W}$$
This is a sum over all closed world sheets and over all world sheets spanned on the loop $C_t$. In other words, we have arrived to an expression of the partition function of compact electrodynamics in terms of the world sheets of loops: the $loop$ (Lagrangian) representation.
There are other equivalent representations which can be obtained from the Villain form. First, we have the the $dual$ representation [@s] obtained essentially by using the Poisson identity and then performing a duality transformation. Actually, the loop representation for the compact U(1) gauge model is reached following this procedure but stoping before the duality transformation. Second, for any lattice theory with Abelian compact variables, the ‘$topological$’ or $BKT$ (for Berezinskii-Kosterlitz-Thouless) representation [@bwpp] via the ‘$Banks-Kogut-Myerson$’ transformation [@bkm]. The $BKT$ expression for the partition function of a lattice theory with Abelian compact variables is given by $$Z_V \propto \sum_{
\begin{array} {c} *t\\
\left( \partial *t = 0 \right)
\end{array}
}\exp
(-\frac{2\pi^2}{g^2} <*t,\hat{\Delta}*t>),$$ i.e. a sum over closed $(D-k-2)$ topological forms $*t$ attached to the cells $c_{(D-k-2)}$ of the dual lattice and where $\hat{\Delta}$ represents the propagator operator. In the case of compact electrodynamics, $*t\equiv *m$ i.e. the topological objects are monopoles (particles for D=2+1 and loops for D=3+1) and $\hat{\Delta}\equiv \frac{1}{\Box}$.
Returning to the loop representation of the partition function Eq.(\[eq:Vill\]) we can observe that the loop action is proportional to the $quadratic$ $area$ $A_2$: $$S_V = -\frac{1}{{\beta}_V} A_2 =
-\frac{1}{{\beta}_V}\sum_{p \in {\cal S}} n_p^2 = -\frac{1}{{\beta}_V}<n,n>,
\label{eq:A2}$$ i.e. the sum of the squares of the multiplicities $s_p$ of plaquettes which constitute the loop’s world sheet [ S]{}. It is interesting to note the similarity of this action with the continuous Nambu action or its lattice version, the Weingarten action [@we] which are proportional to the area swept out by the bosonic string [^1].
On the other hand, we know that in the continuum the classical action of topological string-like solitons, namely Nielsen-Olesen vortices, reduces to the Nambu action in the strong coupling limit [@no]. The Nielsen-Olesen strings are static solutions of the Higgs Abelian model. A patent analogy is observed when we compare the $loop$-representation of lattice compact pure $QED$ and the $BKT$-representation of lattice $Higgs$ non-compact $QED$. (We compare with the non-compact instead of the compact version because this last, in addition to Nielsen-Olesen strings, also has Dirac monopoles as topological solutions and then we have to consider open as much as closed world sheets. In ref. [@af2] we considered the [*Higgs compact QED*]{} which exhibits duality between both representations).
In the case of $Higgs$ non-compact $QED$ model ( a non compact gauge field $A_\mu$ coupled to a scalar field $\Phi=|\Phi|e^{i\phi}$) we have $t*\equiv \sigma*$, where $\sigma*$ represents a 2-form which corresponds to the world sheet of the topological objects namely the Nielsen-Olesen strings [@pwz] and $\hat{\Delta}\equiv \frac{1}{\Box+M^{2}}$ (M is the the mass acquired by the gauge field due to the Higgs mechanism). Thus, both models consist in a sum over closed surfaces which are the world sheets of closed electric strings (loops) and closed magnetic strings (closed Nielsen-Olesen vortices) respectively. The corresponding lattice actions are essentially the quadratic area of the world sheets in both cases. Moreover, the creation operator of both loops and N.O. strings is essentially the same: the Wilson loop operator [@gt] [@pwz].
It is also possible to regard the connection between this two models from a different point of view : the Villain form of $U(1)$ is the Higgs coupling $\rightarrow \infty$ limit of [*Higgs non-compact QED*]{}. The standard action of [*Higgs non-compact QED*]{} is given by $$S = -\frac{\beta}{2} \sum_p {\theta}_p^2 + \gamma \sum_l
\bar{\phi}_xU(l)\phi_{x+l}.
\label{eq:sncqed}$$
It was pointed out in reference [@bd] that in the limit $\gamma \rightarrow \infty$ $$U(l) = e^{i{\theta}_l} \rightarrow 1,$$ which implies for the angular variables $${\theta}_l=2\pi n_l,$$ and so, the action (\[eq:sncqed\]) becomes $$S_\infty = -\frac{\beta}{2}\sum_{p \in {\cal S}} (2\pi n_p)^2 =
-2{\pi}^2\beta<n,n>,$$ where $n_p=\nabla n_l$. We note that this is just the Villain action of gauge $U(1)$ model, Eq.(\[eq:A2\]) but with $2{\pi}^2\beta$ instead of $\frac{1}{{\beta}_V}$.
Numerical Computations
======================
Here we shall present the results of numerical simulations carried out for the loop action (\[eq:A2\]) corresponding to Villain form of $U(1)$ model. In fact, we have simulated the $dual$ action in terms of the dual integer variables $*n_p$ $$S_d = -\frac{1}{{\beta}_V}\sum_{p} *n_p^2 = -\frac{1}{{\beta}_V}
\sum_{p} (\nabla *n_l)^2
= -\frac{1}{{\beta}_V} \sum_{p} (\sum_{l \in p} *n_l)^2 ,
\label{eq:dual}$$ where $*n_l$ are integer 1-forms attached to links of the dual lattice and the integer 2-forms $*n_p$ correspond to their lattice curl, i.e. $\partial n_p = 0$ means that $n_p=\partial n_c$ where $n_c$ are integer 3-forms attached to the elementary cubes and according to (\[eq:dualrel\]) $*n_p=\nabla *(n_c) = \nabla *n_l$. Note that this action implies the assignment of unbounded integer variables to the links of the lattice and the action is defined through the square of the ordered sum of the integers of an elementary square of the lattice. We have implemented a Metropolis algorithm fixing the acceptance ratio, as it is usual in random surfaces analysis.
We have studied this model for different lattice sizes, imposing periodic boundary conditions. Simple thermal cycles showed the presence of a phase transition in the neighborhood of $\beta_V=0.639$. To study the order of this transitions we have analyzed the energy histogram and we have checked for the presence of tunneling between the phases. We have not applied any reweighting extrapolation technique, only a direct observation of the histograms.
In Fig. 1a we present the histogram of the plaquette energy density including 80.000 measures, after discarding 40.000 thermalizing iterations, for a $12^4$ lattice just at the Villain transition point ${\beta}_V = 0.639$. One can observe clearly a two-peak structure.
In Fig. 1b we present the time evolution of the total internal energy during the simulation. Each point is the average over 100 consecutive iterations. This analysis shows clearly the presence of “tunneling” between the phases.
Our numerical results, using this loop action equivalent to Villain form and imposing periodic boundary conditions ($PBC$), exhibit a first order phase transition signal. This is in agreement with the standard lattice numerical simulations of QED using the Wilson action (again using the standard $PBC$) [@jnz],[@acg]. Nevertheless, some differences between the simulations using Wilson or Villain actions arise. In particular, we have not observed here the strange persistence of the phases -and the absence of tunneling- seen in the simulations performed with the Wilson action. Remember that according [@lne]-[@bf] the first order nature of this phase transition seems to be a spurious effect produced by the non-trivial topology that periodic boundary conditions bear with.
Typical loop configurations, obtained by intersecting the lattice with $x_4$ = constant planes, are showed in Fig.2 for two values of the coupling constant, one to the left (strong coupling) and one to the right (weak coupling) of the transition point. The difference between the two typical configurations is patent. These configurations are obtained by taking the last measure after thermalization with 20.000 iterations and they represent all the plaquettes with integer value $n_p \ne 0$. In order to obtain this plaquettes we proceed as follows: first, we stored the $*n_p \neq 0$ of the dual plaquettes and then we performed a duality transformation to get the corresponding non zero $n_p$.
In TABLE 1 we present the spectra of plaquettes $\left\{ n_p \right\}$ (in all the lattice, not only in a particular cut $x_4 = t$ ) for different values of the ${\beta}_V$ coupling. Comparing the $\left\{ n_p \right\}$ at ${\beta}_V$=0.637 and ${\beta}_V$=0.641 i.e. just before and after the critical Villain coupling $\beta_c$=0.639 one can observe an abrupt increment of non zero plaquettes.
TABLE 1.
[||c|l|l|l|l|l|l|l|l|l||]{}\
${\beta}_V$ &$n=-4$ &$n=-3$ &$n=-2$ &$n=-1$ & $n=0$ & $n=1$ & $n=2$ & $n=3$ & $n=4$\
0.200 & 0 & 0 & 0 & 15 & 124386 & 15 & 0 & 0 & 0\
0.500 & 0 & 0 & 2 & 1657 & 121099 & 1659 & 0 & 0 & 0\
0.555 & 0 & 0 & 2 & 3661 & 117093 & 3655 & 5 & 0 & 0\
0.637 & 0 & 0 & 70 & 12208 & 99850 & 12228 & 60 & 0 & 0\
0.641 & 0 & 0 & 115 & 15104 & 93972 & 15116 & 109 & 0 & 0\
0.714 & 0 & 0 & 299 & 19573 & 84639 & 19639 & 266 & 0 & 0\
1.000 & 1 & 12 & 1292 & 25787 & 70239 & 25776 & 1300 & 9 & 0\
2.000 & 12 & 569 & 6695 & 30141 & 49474 & 30337 & 6630 & 543 & 15\
Incidentally, we want to remark that the apparent first order behaviour of the [*Higgs non-compact QED*]{} in the $\gamma \rightarrow \infty$ limit found in ref. [@bd] can be clarified under this new perspective. In that limit it turns out a sort of “compactification” which transforms the original model in the pure [*compact QED*]{} (Villain form) we have studied here.
Conclusions
===========
As it was mentioned, the loop space provides a common scenario for a non-local description of gauge theories and quantum gravity. Up to now, the loop approach was exclusively a Hamiltonian formalism and no lagrangian counterpart was available. A classical action for the Yang-Mills theory in terms of loop variables would be very valuable in its own right because they are the natural candidates to describe the theory in a confining phase. In addition, it may be useful to obtain semiclassical approximations to gauge theories or to general relativity in terms of Ashtekar’s variables. Here we present some small steps in this direction which continue those ones of ref. [@aggs] and [@af1].
In relation with the considered case of lattice Abelian gauge theory, we observed the known analogy between the confining phase of lattice electrodynamics described in terms of electric loops and the Higgs phase described in terms of Nielsen-Olesen magnetic vortices.
We can also ask whether the loops are no more than a useful representation or if, perhaps, they have a deeper physical meaning. Lattice QED exhibits a confining-deconfining transition, although in principle, ordinary continuum QED has only one non-confining phase. However, there are studies which indicate that also there is a phase transition for QED in the continuum [@mn]. In addition to the usual weak coupling phase, a strong coupling confining phase exists above a critical coupling $\alpha_c$. This new phase could explain a mysterious collection of data from heavy ion collisions [@sch]-[@cg]. The unexpected feature is the observation of positron-electron resonances with narrow peak energy in the range of 1.4-1.8 MeV. This suggest the existence of ’electro-mesons’ in a strongly coupled phase of QED. Moreover, a new two-phases model of continuum QED and a mass formula for taking account of the positronium spectrum in the strong coupling phase has been regarded recently [@az]. Thus, in principle, we can speculate about the existence in nature of abelian electric tubes providing a real support for abelian loops and the possibility of being on equal footing with the observed magnetic vortices.
This work has been supported in part by the CICYT project AEN93-0474, H.F. has been supported by the COMMISSION OF THE EUROPEAN COMMUNITIES.
We wish to thank Rodolfo Gambini for useful discussions and comments.
[99]{}
R.Gambini and A.Trias, Phys.Rev.D [**22**]{} (1980) 1380.
R.Gambini and A.Trias, Phys.Rev.D [**23**]{} (1981) 553.
A. Ashtekar [ Phys. Rev. Lett.]{} [**57**]{}, (1986) 2244; A. Ashtekar, [ Phys. Rev.]{} [**D36**]{}, (1987) 1587.
C. Rovelli and L. Smolin, [ Phys. Rev. Lett.]{} [**61**]{} (1988) 1155; [ Nuc. Phys.]{} [**B331**]{} (1990) 80.
R. Gambini [ Phys. Lett.]{} [**B235**]{} (1991) 180;
R.Gambini and A.Trias, Nuc. Phys. [**B278**]{} (1986) 436.
R.Gambini, L.Leal and A.Trias, Phys.Rev.[**D39**]{} (1989) 3127. C. Di Bartolo, R.Gambini and A.Trias, Phys.Rev.[**D39**]{} (1989) 1756.
J.M. Aroca and H. Fort, Phys.Lett.[**B299**]{} (1993) 305; J.M. Aroca and H. Fort, Phys.Lett.[**B317**]{} (1993) 604.
S. Mandelstam, Phys. Rev. [**D19**]{} (1979) 2391.
D.Armand Ugon, R.Gambini, J.Griego and L.Setaro, Preprint IF/FCN-93 July, 1993.
J.M.Aroca and H.Fort, Phys.Lett.[**B325**]{} (1994) 166.
A. M. Polyakov, [ Nuc. Phys.]{} [**B164**]{}, (1979) 171.
J.L. Gervais and A. Neveu, [ Phys. Lett.]{}[**80B**]{} (1979) 255.
H.B.Nielsen and P.Olesen, Nucl. Phys. [**B61**]{} (1973) 45.
M. Baig and E. Dagotto, Nucl. Phys. [**B17**]{} (Proc. Suppl.) (1990) 671; M. Baig, E. Dagotto and E. Moreo Phys. Lett. [**242**]{} (1990) 444.
A. H. Guth, Phys.Rev.[**D21**]{} (1980) 2291.
R.Savit, Rev.Mod.Phys.[**52**]{} (1980) 453.
A.K. Bukenov, U.J. Wiese, M.I.Polikarpov and A.V. Pochinskii , Phys.At.Nucl.[**56**]{} (1993) 122.
T. Banks, R. Myerson and J.B. Kogut, Nucl. Phys. [**B129**]{} (1977) 493. D. Weingarten, Phys.Lett. [**B90**]{} (1980) 280.
V.A. Kazakov, T.A. Kozhamkulov and A.A. Migdal, Sov.J.Nucl.Phys. [**43**]{} (1986) 301.
J.M.Aroca and H. Fort Preprint UAB-FT-329/94.
M.I.Polikarpov U.J.Wiese and M.A.Zubkov, Phys.Lett. , [**B309**]{} (1993) 133.
J. Jerzak, T. neuhaus and P.M. Zerwas, Phys.Lett. [ **B133**]{} (1983) 103.
V. Azcoiti, G. Di Carlo and A. F. Grillo, Phys. Lett. [**B267**]{} (1991) 101.
C.B. Lang and T. Neuhaus, Nuclear Physics [**B**]{} (Proc. Suppl.) [**34**]{} (1994) 543.
W. Kerler, C. Rebbi and A. Weber, Preprint BU-HEP 94-7.
M. Baig and H. Fort, Preprint UAB-FT 338/94 (Phys. Lett. B, in press)
T. Maskawa and H. Nakajima, Progr.Theor.Phys. [**52**]{} (1974) 1326; T. Maskawa and H. Nakajima, Progr.Theor.Phys.[**54**]{} (1975) 860; R.Fukuda and T.Kugo, Nucl.Phys.[**B117**]{} (1976) 250; V.A.Miransky, Nuovo Cimento [**90A**]{} (1985) 149; V.A.Miransky and P.I.Fomin, Sov.J.Part.Nucl.[**16**]{} (1985) 203;
J. Schweppe et al, Phys.Rev.Lett.[**51**]{} (1983) 2261.
M.Clemente et al, Phys.Lett.[**B137**]{} (1984) 41.
T. Cowan et al, Phys.Rev.Lett.[**54**]{} (1985) 1761; T. Cowan et al, Phys.Rev.Lett.[**56**]{} (1986) 444.
H. Tsertos et al, Phys.Lett.[**B162**]{} (1985) 273; Z.Phys.[**A326**]{} (1987) 235.
T. Cowan and J.Greenberg,in [*Physics of Strong Fields*]{} edited by W.Greiner (Plenum, New York, 1987).
Awada and Zoller, Nucl. Phys. [**B365**]{} 699 (1993).
[**Figure Captions.**]{}
1a
: Histogram of the plaquette energy density corresponding to 80.000 measures on a $12^4$ lattice just at the Villain transition point ${\beta}_V = 0.639$.
1b
: Time evolution of the total internal energy during the simulation. Each point is the average over 100 consecutive iterations.
2a
: Typical loop configurations, obtained by intersecting the lattice with $x_4=6$, at ${\beta}_V =0.5$ (strong coupling phase).
2b
: Typical loop configurations, obtained by intersecting the lattice with $x_4=6$, at ${\beta}_V =1$ (weak coupling phase).
[^1]: The relation between the surfaces of the Wilson action and those of Weingarten action has been analyzed by Kazakov et al in ref.[@kkm].
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'Daniele Barducci,'
- 'Andreas Goudelis,'
- Suchita Kulkarni
- and Dipan Sengupta
bibliography:
- 'paper.bib'
title: 'One jet to rule them all: monojet constraints and invisible decays of a 750 GeV diphoton resonance.'
---
HEPHY-PUB 963/15\
LAPTH-068/15
Introduction {#sec:intro}
============
The ATLAS [@ATLAS-CONF-2015-081] and CMS [@CMS-PAS-EXO-15-004] collaborations recently announced their first results on searches for new resonances decaying into two photons at 13 TeV centre-of-mass (CM) energy $pp$ collisions, with integrated luminosities of 3.2 fb$^{-1}$ and 2.6 fb$^{-1}$ respectively. They both observe an excess of events in the diphoton invariant mass bins around $750$ GeV, with a 3.6$\sigma$ (2.0$\sigma$) and 2.6$\sigma$ (1.2$\sigma$) local (global) significance respectively. A large number of papers have already appeared, studying potential implications of such an observation and numerous ways to interpret it in terms of New Physics (NP) scenarios [@Harigaya:2015ezk; @Mambrini:2015wyu; @Backovic:2015fnp; @Angelescu:2015uiz; @Knapen:2015dap; @Buttazzo:2015txu; @Pilaftsis:2015ycr; @Franceschini:2015kwy; @DiChiara:2015vdm; @Higaki:2015jag; @McDermott:2015sck; @Ellis:2015oso; @Low:2015qep; @Bellazzini:2015nxw; @Gupta:2015zzs; @Petersson:2015mkr; @Molinaro:2015cwg; @Dutta:2015wqh; @Cao:2015pto; @Matsuzaki:2015che; @Kobakhidze:2015ldh; @Martinez:2015kmn; @Cox:2015ckc; @Becirevic:2015fmu; @No:2015bsn; @Demidov:2015zqn; @Chao:2015ttq; @Fichet:2015vvy; @Curtin:2015jcv; @Bian:2015kjt; @Chakrabortty:2015hff; @Ahmed:2015uqt; @Agrawal:2015dbf; @Csaki:2015vek; @Falkowski:2015swt; @Aloni:2015mxa; @Bai:2015nbs; @Gabrielli:2015dhk; @Benbrik:2015fyz; @Kim:2015ron; @Alves:2015jgx; @Megias:2015ory; @Carpenter:2015ucu; @Bernon:2015abk; @Chala:2015cev; @Han:2015qqj; @Murphy:2015kag].
The situation is of course still extremely uncertain, partly because of the low significance of the excess which could be due to a statistical fluctuation. Still, it is interesting to examine various facets of the consequences of such an observation being confirmed in the near future. A first statement that can be made with some certainty is that if indeed a new particle is being observed in the diphoton channel, it should have spin-0 or 2 by virtue of the Landau-Yang theorem [@Landau:1948kw; @Yang:1950rg] (see, however, [@Bernon:2015abk; @Chala:2015cev]). In this work we focus on the spin-0 case. On the slightly more speculative side, the excess appears to be compatible with fairly large cross section values, lying at the limits of (although surviving) the LHC Run-1 constraints. Lastly, it looks compatible with a particle of a fairly large width, with first estimates even pointing to a particle as broad as $45$ GeV [@ATLAS-CONF-2015-081].
It has already been shown (see, , [@Aloni:2015mxa]) that decays into Standard Model (SM) particles alone cannot account for a width as large as $45$ GeV. One interesting way through which a broad resonance can be explained is by invoking decays into some invisible final state. If, moreover, these final state particles are also stable on cosmological timescales, then one could eventually entertain the possibility that they may constitute the dark matter (DM) in the Universe, while the resonance itself could actually play the role of a “portal” between the SM particles and the DM sector [@Patt:2006fw]. Needless to say, this portal scenario could in principle also be viable even if the resonance turns out to be narrow.
This simple picture is, nonetheless, subject to numerous constraints. First, the coupling of the resonance to gluons or quarks is constrained (albeit weakly) by LHC dijet searches at 8 TeV. Then, the decays into invisible states are subject to bounds from the monojet + missing energy ($j + E_{T}^{\rm miss}$) searches, which are the main topic of this paper. Finally, if one wishes to make a connection to DM physics, then one should examine the compatibility of all the LHC constraints with those coming from DM abundance considerations and, eventually, direct/indirect detection.
In this paper we make an effort to put some of these pieces together in a systematic manner. We recast a supersymmetry (SUSY) monojet search to obtain constraints on the parameter space of the considered model and show their interplay with the diphoton resonance production cross section, its decay width into invisible final states, 13 TeV dijet cross section predictions as well as with cosmological considerations on DM. The paper is organised as follows: in Sec. \[sec:LagDM\] we describe our parametrisation for the resonance interactions with SM and invisible particles, summarise the experimental situation on the collider side and comment on DM-related properties. In Sec. \[sec:analysis\] we describe the setup of our analysis, the tools we employ and present our main findings. Finally, in Sec. \[sec:outro\] we summarise our results and conclude.
Working assumptions, collider and DM constraints {#sec:LagDM}
================================================
Effective description of a 750 GeV resonance {#sec:description}
--------------------------------------------
Our working assumption is that the observed excess around $750$ GeV is due to a SM gauge singlet scalar particle $s$ that (effectively) couples to the SM gluons and electroweak (EW) gauge bosons, as well as to a new species of Majorana fermions $\psi$. We neglect all potential couplings of $s$ to SM fermions (which, for a singlet $s$, can also only arise through higher-dimensional operators) as well as to the 125 GeV Higgs boson (which are allowed at tree-level).
Numerous conventions have been adopted by different authors in order to describe such effective interactions. We choose to parametrise our Lagrangian as[^1] $$\begin{aligned}
\label{eq:Lcpe}
{\cal{L}}_{\rm NP, CPE} & = \frac{1}{2} (\partial_\mu s)^2 - \frac{\mu_s^2}{2} s^2 + \frac{1}{2} \bar{\psi} (i \slashed{\partial} - m_\psi) \psi - \frac{y_{\psi}}{2} s \bar{\psi} \psi \\ \nonumber
& - \frac{g_1^2}{4 \pi} \frac{1}{4 \Lambda_1} s ~ B_{\mu\nu} B^{\mu\nu}
- \frac{g_2^2}{4 \pi} \frac{1}{4 \Lambda_2} s ~ W_{\mu\nu} W^{\mu\nu}
- \frac{g_3^2}{4 \pi} \frac{1}{4 \Lambda_3} s ~ G_{\mu\nu} G^{\mu\nu}\end{aligned}$$ where $B_{\mu\nu}$, $W_{\mu\nu}$ and $G_{\mu\nu}$ are the $U(1)_Y$, $SU(2)_L$ and $SU(3)_c$ field strength tensors respectively and $g_{1,2,3}$ are the corresponding SM coupling constants. The Lagrangian actually corresponds to the case where $s$ is even under the charge-parity ($CP$) symmetry. In the case of a pseudoscalar particle, the Lagrangian becomes $$\begin{aligned}
\label{eq:Lcpo}
{\cal{L}}_{\rm NP, CPO} & = \frac{1}{2} (\partial_\mu s)^2 - \frac{\mu_s^2}{2} s^2 + \frac{1}{2} \bar{\psi} (i \slashed{\partial} - m_\psi) \psi - i \frac{y_{\psi}}{2} s \bar{\psi} \gamma^5 \psi \\ \nonumber
& - \frac{g_1^2}{4 \pi} \frac{1}{4 \Lambda_1} s ~ B_{\mu\nu} \tilde{B}^{\mu\nu}
- \frac{g_2^2}{4 \pi} \frac{1}{4 \Lambda_2} s ~ W_{\mu\nu} \tilde{W}^{\mu\nu}
- \frac{g_3^2}{4 \pi} \frac{1}{4 \Lambda_3} s ~ G_{\mu\nu} \tilde{G}^{\mu\nu}\end{aligned}$$ where $\tilde{B}$, $\tilde{W}$ and $\tilde{G}$ are the field strength duals, $\tilde{F}_{\mu\nu} = 1/2 \epsilon_{\mu\nu\rho\sigma} F^{\rho\sigma}$. The collider phenomenology aspects of $s$ we will focus on depend only mildly on its $CP$ nature, unlike the DM properties of $\psi$.
The interpretation of the suppression mass scales in Eqs. and is heavily model-dependent. The most straightforward way of obtaining such interactions is, , by integrating out loops of heavy vector-like fermions. In our analysis the $\Lambda$ couplings will be treated merely as a parametrisation of the underlying physics, without any detailed reference to their potential ultraviolet (UV) origins, and the parameter ranges we will choose to work with are mostly motivated by the requirements of satisfying the various experimental constraints on the resonance $s$ and studying whether they can be reconciled. For the sake of illustration, in App. \[uvcomplete\] we nevertheless comment on the type of physics that could lead to such couplings and point out some of the corresponding model-building challenges.
Collider implications and observational status {#sec:implications}
----------------------------------------------
The Lagrangian gives rise to a variety of collider signatures. The singlet $s$ can be produced through gluon, vector boson fusion (VBF) or photon fusion and can decay into $g$/$\gamma$/$Z$/$W$ pairs, $Z\gamma$ and, if $m_\psi < m_s/2$, $\bar{\psi}\psi$ final states. We will focus on gluon fusion production, although VBF could provide extremely interesting distinct signatures.
The diphoton excess reported in [@ATLAS-CONF-2015-081; @CMS-PAS-EXO-15-004] appears at an invariant mass around $750$ GeV, with a 3.6$\sigma$ (2.0$\sigma$) and 2.6$\sigma$ (1.2$\sigma$) local (global) significance for ATLAS and CMS respectively. A preliminary fit performed in [@Falkowski:2015swt] points, at 95% confidence level (CL), to cross section values $\sigma(pp \rightarrow s) \times {\rm BR}(s \rightarrow \gamma\gamma) \sim 1-5$ fb assuming a width $\Gamma_s = 5$ GeV and $\sigma(pp \rightarrow s) \times {\rm BR}(s \rightarrow \gamma\gamma) \sim 2-12$ fb for a larger width $\Gamma_s = 40$ GeV when the ATLAS and CMS Run-1 and Run-2 results are combined.
One of the cleanest signatures of a new heavy scalar resonance described by the Lagrangian would be a peak in the dijet or four-lepton invariant mass distributions. Currently the ATLAS and CMS collaborations do not provide dijet limits at $\sqrt{s} = 13$ TeV for masses as low as 750 GeV, as the presentation of their results starts at $m_s \sim$ 1 TeV. The $\sqrt{s} = 8$ TeV ATLAS and CMS analyses presented in [@Khachatryan:2015sja; @Aad:2014aqa] set a limit of $\sigma_{jj} < 1$ pb for a 1 TeV resonance coupling dominantly to $gg$ (for a mass of 750 GeV the limit shown by ATLAS is of the order of 10 pb).
Passing to EW gauge boson final states, ATLAS sets the limits $\sigma_{ZZ} \lesssim 12$ fb [@Aad:2015kna] and $\sigma_{WW} \lesssim 40$ fb [@Aad:2015agg] for a 750 GeV particle decaying into $ZZ/WW$ pairs. For the same mass the ATLAS search for a resonance decaying into a $Z\gamma$ final state places an upper bound of $\sigma_{\gamma Z} \leq 3.5$ fb [@Aad:2014fha], at a CM energy of 8 TeV. On the diphoton side, both ATLAS and CMS have presented upper bounds for the production cross section of a diphoton resonance at $\sqrt{s} = 8$ TeV, setting a limit $\sigma_{\gamma \gamma} \leq 2$ fb [@Khachatryan:2015qba; @Aad:2015mna].
Dark matter and a (pseudo-)scalar portal at 750 GeV {#sec:darkmatter}
---------------------------------------------------
Another interesting possibility arising from the Lagrangian of Eq. is that the fermion $\psi$ could be responsible (also even partially) for the DM abundance observed in the Universe. It has already been shown that assuming standard thermal freeze-out the DM abundance observed by WMAP9 [@Hinshaw:2012aka] and Planck [@Ade:2013zuv] can indeed be obtained in this setup for a wide range of $\psi$ masses [@Backovic:2015fnp; @Mambrini:2015wyu]. As reference values for the DM density, we consider the 3$\sigma$ range from the (CMB+BAO+H$_0$) WMAP 9-year results $$\label{eq:relicbound}
\Omega h^2 = 0.1153 \pm 0.0057 \ .$$ The $CP$ properties of $s$ are crucial for the predicted relic density. In the $CP$-even case, the thermally averaged self-annihilation cross section $\langle \sigma v \rangle$ is velocity-suppressed, which amounts to large $y_\psi$ coupling values being required in order to achieve the observed DM abundance. When $s$ is odd under $CP$, this velocity suppression is lost and smaller values of $y_\psi$ are sufficient to satisfy the bound . For reasonable values of $\Lambda_3$, such that the 8 TeV LHC dijet bounds described in the previous paragraph are satisfied, the predicted relic abundance is found to be prohibitively large for $m_\psi \lesssim 200$ GeV in the $CP$-even case, and $m_\psi \lesssim 100$ GeV in the $CP$-odd one, unless non-perturbative values are considered for $y_\psi$.
Additional constraints come from direct detection (DD) and indirect detection (ID) experiments, with the predictions again depending strongly on the transformation properties of $s$ under $CP$. DD constraints, and in particular the LUX [@Akerib:2013tjd] results, are relevant in the $CP$-even case. We find that, depending on the assumptions adopted for the quark (and, consequently, gluon) content of the nucleon, and taking the couplings of $s$ to the SM quarks to be identically zero, a lower limit can be set on the DM mass which ranges between $\sim200$ and $\sim 300$ GeV. Some more details on our DD computations are given in App. \[app:DD\]. ID constraints on the other hand are ineffective in the $CP$-even case, due to the velocity suppression in $\langle \sigma v \rangle$.
The situation is inversed when $s$ is $CP$-odd. The Lagrangian yields a negligible spin-independent scattering cross section off nuclei. Instead, in this case it is ID which becomes relevant. The strongest bounds come from the six-year Fermi satellite searches for DM annihilation-induced continuum gamma-rays from dwarf spheroidal galaxies [@Ackermann:2015zua] and for gamma-ray lines from the galactic centre [@Ackermann:2015lka]. Additional constraints could also arise from the AMS-02 searches for antiprotons [@AMS02antiprotons] as extracted, for example, in [@Giesen:2015ufa], which we nonetheless find to be weaker for the DM mass range of our interest. A more detailed discussion of ID constraints and perspectives can be found in [@Park:2015ysf]. For low values of $\Lambda_{1,2} \lesssim 50$ GeV, the gamma-ray line searches dominate and can exclude DM masses up to $\sim 200$ GeV, depending also on the assumptions for the underlying DM halo profile in the Milky Way. Continuum gamma-ray searches give comparable but slightly weaker bounds.
For reasons of clarity, throughout the subsequent discussion we will ignore DM detection constraints. The indicative numbers quoted previously, although subject to uncertainties, show nonetheless that DD and ID could provide valuable information on scenarios relating the putative diphoton excess with DM.
Analysis {#sec:analysis}
========
For our analysis, we calculate diphoton and dijet cross sections at the 13 TeV LHC, as well as monojet production for $\sqrt{s}=$ 8 TeV, mediated by the $s$ resonance as described by the Lagrangian . In particular we consider the processes: $$\begin{split}
& pp\to s \to \gamma\gamma,\\
& pp\to s \to j j,\\
& pp\to s \to \psi \psi j\\
\end{split}
\label{eq:LHC}$$ We moreover compute the relic abundance of $\psi$ assuming standard thermal freeze-out.
Analysis setup {#sec:tools}
--------------
The model described in Eqs. and has been implemented in the `UFO` format [@Degrande:2011ua] through the `Feynrules` package [@Alloul:2013bka] and event samples have been generated through `MadGraph5_aMC@NLO` [@Alwall:2014hca]. In particular, the 13 TeV $p p \rightarrow \gamma \gamma$ and $p p \rightarrow j j$ cross sections were computed at parton level and convoluted with the CTEQ6L1 [@Pumplin:2002vw] parton distribution functions[^2]. Furthermore, we have also calculated the width of the resonance within the same set up. DM observables have been computed with the [micrOMEGAs4.1]{} package [@Belanger:2014vza], with the exception of the spin-independent WIMP-nucleon scattering cross section, which has been calculated analytically as described in App. \[app:DD\].
In order to exploit the constraints arising from 8 TeV data on monojet signatures, we have used a recast version of the ATLAS monojet search ATLAS-SUSY-2013-21 [@Aad:2014nra] [^3], implemented in the `MadAnalysis5` [@Conte:2014zja] package and described on the Public Analysis Database (PAD) [@Dumont:2014tja]. This recast analysis is publicly available online at [@atlasmonojet], together with a validation note [@atlasmonojeturl]. This analysis targeted decays of the SUSY partner of the top quark, the stop, into a charm quark and neutralino final state, for a compressed stop-neutralino spectrum. The search tags the emission of a hard initial state radiation jet recoiling against the $E_T^{\rm miss}$.
The generated parton level events for the process $pp\to \psi \psi j$ were hadronised with the `PYTHIA6` [@Sjostrand:2006za] package. A merging scale of 30 GeV was used to perform the Matrix element-Parton Shower matching (ME-PS) [@Hoche:2006ph] between the 0 and 1 jet samples. A fast detector simulation was performed with the `MadAnalysis5` tuned version of the `Delphes3` [@deFavereau:2013fsa] package as described in [@Dumont:2014tja]. Jets were reconstructed using `FastJet` [@Cacciari:2011ma], via an $\rm anti$-$k_{T}$ [@Cacciari:2008gp] algorithm with a cone size of 0.4 and they are required to have $p_{T}>20$ GeV. Furthermore we have used the ATLAS `AUET2B` tune [@ATL-PHYS-PUB-2011-009] to simulate underlying events.
The reconstructed events were finally passed through the aforementioned recast ATLAS monojet analysis [@Aad:2014nra], which consists of three signal regions targeting ($p_{T}^{j}$, $E_{T}^{\rm miss}$) threshold combinations of (280, 200), (340, 340) and (450, 450) GeV respectively. To obtain the constraints arising from the ATLAS monojet analysis, we have used the [exclusion-CLs.py]{} module implemented in the `MadAnalysis5` package. This module determines, given the number of signal, expected and observed background events, together with the background uncertainty (the latter three directly taken from the experimental publications), the most sensitive signal region (SR) of the analysis and the exclusion CL using the CLs prescription [@Read:2000ru; @Read:2002hq] for the most sensitive SR.
For our analysis we scanned over $\Lambda_3$ and $ y_{\psi}$ for discrete values of $\Lambda_1, \Lambda_2$ and DM masses $m_{\psi}$, setting $\Lambda_1= \Lambda_2 \equiv \Lambda_{1,2}$ for simplicity. In particular the following parameter scan was performed[^4]
- $m_\psi = 50,\;150,\;250,\;350$ and $450$ GeV,
- $\Lambda_{1,2}$= 20, 50, 200 and 400 GeV,
- $\Lambda_3 \in [200, 3000]$ GeV and $y_\psi \in [0.05, 4\pi]$.
Note that we have chosen to study relatively extreme values for $\Lambda_{1,2}$, since the behaviour of the various observables in the intermediate regime can be inferred via an interpolation between the values we consider. We should also point out that especially for the $\Lambda_{1,2} = 20$ and 50 GeV scenarios, substantial cross sections into $ZZ$ and $WW$ final states are predicted over a significant fraction of the parameter space, which are in direct conflict with the corresponding limits quoted in Sec. \[sec:implications\]. We have explicitly verified that all of our scenarios with $\sigma_{\gamma\gamma} < 12$ fb, *i.e.* within the region preferred by the observed diphoton excess, are consistent with the relevant bounds on $\sigma_{ZZ/WW}$. Throughout the subsequent discussion, although these bounds will be omitted for clarity, the reader should keep in mind that $ZZ/WW$ searches are (at least) in tension with all parameter space regions characterised by $\sigma_{\gamma\gamma} \gtrsim 18$ fb. This tension can be relaxed, for example, by considering scenarios with $\Lambda_2\gg\Lambda_1$.
Results {#sec:results}
-------
We first consider the regime where $m_{\psi} < m_{s}/2$. This region is particularly interesting as it can in principle account for the potentially large width of the resonance through decays into the invisible state $\psi$ [@Mambrini:2015wyu; @Backovic:2015fnp]. Motivated by the comments on the DM density made in Sec. \[sec:darkmatter\], we choose to present our results for the cases $m_\psi = 250$ and 350 GeV. For $m_\psi = 50$ GeV, it is simply impossible to reproduce the observed DM abundance for perturbative values of $y_\psi$. For $m_\psi = 150$ GeV, it is possible to do so in the $CP$-odd case but only at the cost of large values for $y_\psi$ which amount to an exceedingly large width $\Gamma_s$ (this regime is also in quite strong tension with indirect searches for gamma-ray lines). We will nonetheless comment on our findings for these cases later on.
Our main results are presented in Fig. \[fig:monojet250\] for $m_\psi = 250$ GeV and in Fig. \[fig:monojet350\] for $m_\psi = 350$ GeV, for the values $\Lambda_{1,2} = 20, 50, 200, 400$ GeV in the top left, top right, bottom left and bottom right panels respectively. The predicted 13 TeV production cross sections for the dijet (blue contours) and diphoton (red regions) final states are shown, along with the total width of the resonance (green contours). The 95% CL monojet constraints derived at 8 TeV from the recast search as described in Sec. \[sec:tools\] are also overlaid (black contours). Finally, where possible, a blue (green) band satisfying the DM bound for the $CP$-even ($CP$-odd) case is shown. All cross sections are given in fb and masses/widths in GeV. Note that the results for extremely large widths should be interpreted with care. In this regime in fact a full momentum-dependent width ought to be used in the resonance propagator when performing the calculation.
![Predictions for $p p \rightarrow s \rightarrow \gamma \gamma$ (red band) and $p p \rightarrow s \rightarrow j j$ (blue contours) cross sections at $\sqrt{s} = 13$ TeV, overlaid with 8 TeV monojet constraints (black line) and the width of the resonance $s$ (green contours). The mass of the invisible fermion $\psi$ is fixed at $m_\psi = 250$ GeV and $\Lambda_{1,2} = 20, 50, 200, 400$ GeV in the top left, top right, bottom left and bottom right panels respectively. Monojet constraints are derived at 95% C.L. The blue (green) band shows regions of parameter space compatible with the observed DM density for a scalar (pseudoscalar) mediator.\[fig:monojet250\]](./plots/mpsi_250_l12_20.pdf "fig:"){width="50.00000%"} ![Predictions for $p p \rightarrow s \rightarrow \gamma \gamma$ (red band) and $p p \rightarrow s \rightarrow j j$ (blue contours) cross sections at $\sqrt{s} = 13$ TeV, overlaid with 8 TeV monojet constraints (black line) and the width of the resonance $s$ (green contours). The mass of the invisible fermion $\psi$ is fixed at $m_\psi = 250$ GeV and $\Lambda_{1,2} = 20, 50, 200, 400$ GeV in the top left, top right, bottom left and bottom right panels respectively. Monojet constraints are derived at 95% C.L. The blue (green) band shows regions of parameter space compatible with the observed DM density for a scalar (pseudoscalar) mediator.\[fig:monojet250\]](./plots/mpsi_250_l12_50.pdf "fig:"){width="50.00000%"}\
![Predictions for $p p \rightarrow s \rightarrow \gamma \gamma$ (red band) and $p p \rightarrow s \rightarrow j j$ (blue contours) cross sections at $\sqrt{s} = 13$ TeV, overlaid with 8 TeV monojet constraints (black line) and the width of the resonance $s$ (green contours). The mass of the invisible fermion $\psi$ is fixed at $m_\psi = 250$ GeV and $\Lambda_{1,2} = 20, 50, 200, 400$ GeV in the top left, top right, bottom left and bottom right panels respectively. Monojet constraints are derived at 95% C.L. The blue (green) band shows regions of parameter space compatible with the observed DM density for a scalar (pseudoscalar) mediator.\[fig:monojet250\]](./plots/mpsi_250_l12_200.pdf "fig:"){width="50.00000%"} ![Predictions for $p p \rightarrow s \rightarrow \gamma \gamma$ (red band) and $p p \rightarrow s \rightarrow j j$ (blue contours) cross sections at $\sqrt{s} = 13$ TeV, overlaid with 8 TeV monojet constraints (black line) and the width of the resonance $s$ (green contours). The mass of the invisible fermion $\psi$ is fixed at $m_\psi = 250$ GeV and $\Lambda_{1,2} = 20, 50, 200, 400$ GeV in the top left, top right, bottom left and bottom right panels respectively. Monojet constraints are derived at 95% C.L. The blue (green) band shows regions of parameter space compatible with the observed DM density for a scalar (pseudoscalar) mediator.\[fig:monojet250\]](./plots/mpsi_250_l12_400.pdf "fig:"){width="50.00000%"}
A first observation that can be made is that in both the $m_\psi = 250$ GeV and 350 GeV cases, the width of the resonance is fairly independent of $\Lambda_3$, especially when $\Gamma_s \gtrsim 10$ GeV. This behaviour can be understood from the fact that in most of the parameter space at hand, $\Gamma_s$ is completely dominated by the invisible (and, to a lesser extent, EW gauge boson) contribution unless $y_\psi$ and $\Lambda_3$ simultaneously attain small values. The dijet and diphoton cross sections, on the other hand, depend both on $y_\psi$ and $\Lambda_3$. The dijet cross sections are sizeable for smaller $\Lambda_3$ scales, due to the increase in the $s$ production cross section, but also for smaller values of $y_{\psi}$. A similar behaviour is present in the diphoton cross section which moreover increases, as expected, with decreasing $\Lambda_{1,2}$. The $y_\psi$ dependence of the two cross sections is due to both the increase in BR$(s\rightarrow gg/\gamma\gamma)$ and to the decrease of the total width of the resonance. In order to get a feeling of the impact that dijet searches could have on our parameter space, we can naively extrapolate the existing 13 TeV constraints presented in [@ATLAS-CONF-2015-042; @Khachatryan:2015dcf] for a minimal resonance mass of 1.5 TeV down to 750 GeV, assuming that the limit remains constant. Such a – very aggressive – extrapolation would amount to a limit of the order of a few pb, which could be strong enough to probe part of the $m_\psi = 350$ GeV scenario of Fig. \[fig:monojet350\]. However, a dedicated experimental study is required in order to make any concrete statement.
Leaving monojet constraints aside for the moment, we see that in the $m_\psi = 250$ GeV case (Fig. \[fig:monojet250\]) the requirements for a a substantial diphoton cross section and a large resonance width $\Gamma_s > 20$ GeV can be reconciled in substantial parts of the parameter space, except for the case $\Lambda_{1,2} = 400$ GeV where the predicted diphoton cross section is too low. The relic abundance constraint for $\psi$ significantly reduces the available parameter space, although it is still possible to accomodate all three requirements assuming a $CP$-even scalar for $\Lambda_{1,2} = 20$ or 50 GeV (the latter at the price of a slightly larger width) and a $CP$-odd scalar when $\Lambda_{1,2} = 200$ GeV. Note that DM is underabundant (overabundant) above (below) the blue and green bands. The imposition of the monojet constraints has an important impact on the parameter space, excluding $\Lambda_3$ values below $\sim 500$ GeV regardless of the value of $y_\psi$, unless $y_\psi \lesssim~0.25$. This behaviour can be understood by the fact that for sufficiently large values of $y_\psi$, the branching ratio into $\psi$ pairs is basically unity and the monojet cross section essentially only depends on $\Lambda_3$, except for its dependence on the total width of $s$. The only surviving region for the parameter choices shown in Fig. \[fig:monojet250\] where all requirements can be (approximately) reconciled is for $\Lambda_{1,2} = 200$ GeV, a $CP$-odd scalar $s$ and $\Lambda_3 \gtrsim 500$ GeV. Interestingly, though, by comparing the $\Lambda_{1,2} = 20$ and 50 GeV cases, we can deduce that all requirements can also be rendered compatible assuming a $CP$-even scalar for $\Lambda_{1,2}$ values around 30 GeV and for $\Lambda_3$ values above the monojet exclusion bounds. Besides, if the relic abundance requirement is dropped, then for sufficiently large $\Lambda_3$ values the low $\Lambda_{1,2}$ scenarios can generically account for a broad resonance with a large enough diphoton cross section.
![Predictions for $p p \rightarrow s \rightarrow \gamma \gamma$ (red band) and $p p \rightarrow s \rightarrow j j$ (blue contours) cross sections at $\sqrt{s} = 13$ TeV, overlaid with 8 TeV monojet constraints (black line) and the width of the resonance $s$ (green contours). The mass of the invisible fermion $\psi$ is fixed at $m_\psi = 350$ GeV and $\Lambda_{1,2} = 20, 50, 200, 400$ GeV in the top left, top right, bottom left and bottom right panels respectively. Monojet constraints are derived at 95% C.L. The DM abundance can be reproduced for very low $y_\psi$ values of $\sim 0.07$ and $\sim 0.02$ in the scalar and pseudoscalar cases respectively and the corresponding points are omitted for clarity. \[fig:monojet350\]](./plots/mpsi_350_l12_20.pdf "fig:"){width="50.00000%"} ![Predictions for $p p \rightarrow s \rightarrow \gamma \gamma$ (red band) and $p p \rightarrow s \rightarrow j j$ (blue contours) cross sections at $\sqrt{s} = 13$ TeV, overlaid with 8 TeV monojet constraints (black line) and the width of the resonance $s$ (green contours). The mass of the invisible fermion $\psi$ is fixed at $m_\psi = 350$ GeV and $\Lambda_{1,2} = 20, 50, 200, 400$ GeV in the top left, top right, bottom left and bottom right panels respectively. Monojet constraints are derived at 95% C.L. The DM abundance can be reproduced for very low $y_\psi$ values of $\sim 0.07$ and $\sim 0.02$ in the scalar and pseudoscalar cases respectively and the corresponding points are omitted for clarity. \[fig:monojet350\]](./plots/mpsi_350_l12_50.pdf "fig:"){width="50.00000%"}\
![Predictions for $p p \rightarrow s \rightarrow \gamma \gamma$ (red band) and $p p \rightarrow s \rightarrow j j$ (blue contours) cross sections at $\sqrt{s} = 13$ TeV, overlaid with 8 TeV monojet constraints (black line) and the width of the resonance $s$ (green contours). The mass of the invisible fermion $\psi$ is fixed at $m_\psi = 350$ GeV and $\Lambda_{1,2} = 20, 50, 200, 400$ GeV in the top left, top right, bottom left and bottom right panels respectively. Monojet constraints are derived at 95% C.L. The DM abundance can be reproduced for very low $y_\psi$ values of $\sim 0.07$ and $\sim 0.02$ in the scalar and pseudoscalar cases respectively and the corresponding points are omitted for clarity. \[fig:monojet350\]](./plots/mpsi_350_l12_200.pdf "fig:"){width="50.00000%"} ![Predictions for $p p \rightarrow s \rightarrow \gamma \gamma$ (red band) and $p p \rightarrow s \rightarrow j j$ (blue contours) cross sections at $\sqrt{s} = 13$ TeV, overlaid with 8 TeV monojet constraints (black line) and the width of the resonance $s$ (green contours). The mass of the invisible fermion $\psi$ is fixed at $m_\psi = 350$ GeV and $\Lambda_{1,2} = 20, 50, 200, 400$ GeV in the top left, top right, bottom left and bottom right panels respectively. Monojet constraints are derived at 95% C.L. The DM abundance can be reproduced for very low $y_\psi$ values of $\sim 0.07$ and $\sim 0.02$ in the scalar and pseudoscalar cases respectively and the corresponding points are omitted for clarity. \[fig:monojet350\]](./plots/mpsi_350_l12_400.pdf "fig:"){width="50.00000%"}
We now turn our attention to Fig. \[fig:monojet350\], which corresponds to $m_{\psi} = 350$ GeV. In this case, the reduction of phase space for the $s \rightarrow \psi\psi$ decay generically leads to smaller widths and, consequently, larger diphoton (and dijet) cross sections with respect to the $m_{\psi} = 250$ GeV scenario. The monojet constraint shown again as a black line rules out most of the parameter space with $\Lambda_{3} \lesssim 650$ GeV. For this value of $m_{\psi}$ the observed relic density is obtained for $y_{\psi} \sim 0.07$ (0.02) for the $CP$-even ($CP$-odd) case which lies at the lower edge of our plots and the corresponding points are not shown. The relic density constraint is fully incompatible with a large width but can be reconciled with the diphoton excess for sufficiently large $\Lambda_{1,2,3}$ values. Conversely, when DM constraints are dismissed, a substantial diphoton cross section is compatible with a large invisible width for sufficiently low $\Lambda_{1,2}$ values.
A comparison of the excluded regions for $\psi$ masses of 250 and 350 GeV from the monojet searches shows that for large values of $y_{\psi}$, the limits are stronger in the latter case. This is due to a reduction of the total width as $m_\psi$ increases, leading to an enhancement of the total cross section. However for small values of $y_{\psi}$, where the total width is sufficiently small in both scenarios, the exclusion is stronger for the 250 case as compared to 350 due to the higher kinematic acceptance of the monojet search for smaller $\psi$ masses.
The regime between $m_\psi = 250$ and 350 GeV can be understood as an interpolation between the results presented in Figs. \[fig:monojet250\] and \[fig:monojet350\]. Indeed, for such intermediate masses we expect that it is still possible to reconcile a broad diphoton resonance $s$ with the correct DM relic density assuming a $CP$-even scalar $s$. This should happen in particular for relatively low values of $\Lambda_{1,2} \sim 20$ – $50$ GeV. Referring for example to the top right panel of Fig.\[fig:monojet250\], increasing $m_\psi$ would amount to smaller values of the $y_\psi$ coupling being required in order to reproduce the observed relic abundance as the “funnel region” is gradually approached. Schematically, the blue band would then move downwards, towards larger diphoton cross sections and more reasonable $\Gamma_s$ values of the order of 10 to 45 GeV.
In the case $m_\psi < 250$ GeV, on the other hand, the opposite behaviour is expected. For slightly smaller $\psi$ masses (but larger than 150 GeV according to our findings), it is now the $CP$-odd case which can be relevant. Referring again to the top right panel of Fig. \[fig:monojet250\], decreasing $m_\psi$ would amount to larger values of the $y_\psi$ coupling being required in order to achieve the correct relic density. The green band would then move upwards and become compatible with the width and diphoton cross section requirements. Besides, we remind the reader that such a configuration could face severe problems with gamma-ray line searches.
![Predictions for $p p \rightarrow s \rightarrow \gamma \gamma$ (red band) and $p p \rightarrow s \rightarrow j j$ (blue contours) cross sections at $\sqrt{s} = 13$ TeV, overlaid with 8 TeV monojet constraints (black line) and the width of the resonance $s$ (green contours). The mass of the invisible fermion $\psi$ is fixed at $m_\psi = 450$ GeV and $\Lambda_{1,2} = 300, 500$ GeV in the left and right panel respectively. Monojet constraints are derived at 95% C.L. The blue (green) band shows regions of parameter space compatible with the observed DM density for a scalar (pseudoscalar) mediator.\[fig:monojetoffshell\]](./plots/mpsi_450_l12_300.pdf "fig:"){width="48.00000%"}![Predictions for $p p \rightarrow s \rightarrow \gamma \gamma$ (red band) and $p p \rightarrow s \rightarrow j j$ (blue contours) cross sections at $\sqrt{s} = 13$ TeV, overlaid with 8 TeV monojet constraints (black line) and the width of the resonance $s$ (green contours). The mass of the invisible fermion $\psi$ is fixed at $m_\psi = 450$ GeV and $\Lambda_{1,2} = 300, 500$ GeV in the left and right panel respectively. Monojet constraints are derived at 95% C.L. The blue (green) band shows regions of parameter space compatible with the observed DM density for a scalar (pseudoscalar) mediator.\[fig:monojetoffshell\]](./plots/mpsi_450_l12_500.pdf "fig:"){width="48.00000%"}
Given the extremely preliminary nature of the diphoton excess, we have no a priori reason to consider only large width scenarios. Therefore, we also consider two examples with $2 m_\psi> m_s$, necessarily leading to a narrow width for the resonance $s$. In this case the invisible final state is therefore produced via an off-shell mediator. In Fig. \[fig:monojetoffshell\], we present the results for $m_{\psi} = 450$ GeV with scale choices of $\Lambda_{1,2} = 300$ and 500 GeV (left and right panel respectively). As illustrated in Figs. \[fig:monojet250\] and \[fig:monojet350\], the LHC monojet cross sections do not depend drastically on the scale $\Lambda_{1,2}$, hence we derived the constraints for $\Lambda_{1,2} = 500$ GeV, and have used them for the $\Lambda_{1,2} = 300$ GeV case as well. The expected diphoton cross sections in this case can easily exceed $10$ fb, the width of the resonance is smaller, and the monojet search excludes a much smaller region of parameter space, as is expected. The relic density band, once again shown in blue (green) for the $CP$-even ($CP$-odd) case, passes very well through the regions of preferred parameter space and one can obtain the correct DM abundance while within the LHC bounds.
Summary and conclusions {#sec:outro}
=======================
Motivated by the recent hint of a possibly broad excess in the diphoton channel at the LHC, in this work we studied monojet constraints on potential invisible decays of a scalar particle with a mass of $\sim 750$ GeV. We examined the extent to which it is possible to reconcile these constraints with the preferred diphoton cross section values, a large resonance width and, eventually, the relic DM abundance in the Universe in case the invisible decay product is stable on cosmological timescales. We have also presented predictions for the dijet production cross section at the 13 TeV LHC.
We showed that monojet searches already place important constraints on interpretations of the putative 750 GeV diphoton resonance as a portal to a DM sector. Nevertheless for limited regions of the parameter space it is still possible to accommodate all requirements. These regions will be probed, assuming the diphoton excess persists in the LHC data, in the next few years from a combination of LHC analyses and direct/indirect DM detection searches.
Once either the DM or the large width requirements are dropped, it is much easier to reconcile the remaining conditions. Concretely, a broad resonance can still be explained through invisible decays without conflicting monojet searches, whereas a narrow resonance can easily mediate the DM-SM interactions. Additional interesting signatures not considered in this work include multijets (along the lines of [@Buchmueller:2015eea]), $\gamma Z$ and four- or two-lepton final states as well as, in the case of strong coupling to EW gauge bosons, VBF production of the resonance.
In any case, within the next few months it will become clear whether the 750 GeV “excess” constitutes merely a statistical fluctuation or a sign of – long sought for – physics beyond the Standard Model.
Acknowledgements {#sec:ack}
================
AG and SK are supported by the ‘New Frontiers’ program of the Austrian Academy of Sciences. The work of DS is supported by the French ANR, project DMAstroLHC, ANR-12-BS05-0006, and by the Investissements d’avenir, Labex ENIGMASS. We thank Giorgio Arcadi, Yann Mambrini and Alberto Mariotti for several useful discussions.
Some comments on potential UV completions {#uvcomplete}
=========================================
In order to get a feeling of the type of NP that could give rise to interactions like the ones described by the Lagrangians of Eqs. and and the corresponding values of $\Lambda_{1,2,3}$ used in the analysis, we assume a set of additional vector-like fermions $f$ charged under the SM gauge group that couple to $s$ through Yukawa-type terms.
Fermions transforming according to the fundamental representation of $SU(3)_c$ will generate a partial width $\Gamma(s \rightarrow gg)$ as [@Spira:1995rr; @Djouadi:2005gi] $$\label{ggdec}
\Gamma_{\rm UV}(s \rightarrow gg) = \frac{\alpha_{s}^{2} m _{s}^{3}}{72 \pi^3} \left|\frac{3}{4}\sum_{f} \frac{y_{f}}{m_f} F^{s^{\pm}}_{1/2}(\tau_f)\right|^{2}$$ where $\alpha_s$ is the strong coupling constant, $m_s$ the resonance mass, $y_{f}$ and $m_f$ the Yukawa couplings and masses of the heavy fermions and $F^{s^{\pm}}_{1/2}(\tau)$ the loop form factor for the CP even and CP odd case respectively, which reads $$\begin{aligned}
F^{s^{+}}_{1/2}(\tau_f) & = &\frac{2}{\tau_f^{2}}[\tau_f + (\tau_f - 1)f(\tau_f)] \\
F^{s^{-}}_{1/2}(\tau_f) & = & 2 \tau_f^{-1}f(\tau_f)\end{aligned}$$ with $\tau_f \equiv m_s^2/(4 m_f^2)$. For heavy coloured fermions, that is assuming $\tau_f\le1$, the function $f(\tau_f)$ is given by $$\begin{aligned}
f(\tau_f) & = & \arcsin^{2}\sqrt{\tau_f}, \quad \tau_f \le 1.\end{aligned}$$
The corresponding expression for $\Gamma(s \rightarrow gg)$ obtained from the Lagrangians of Eqs. and , on the other hand, reads $$\Gamma_{\rm EFT}(s \rightarrow gg) = \frac{\alpha_s^2}{8\pi} \frac{m_s^3}{\Lambda_3^2}.$$
Then, by matching the two expressions we can obtain the value of $\Lambda_3$ as a function of the fermion masses, their Yukawa couplings and their multiplicities. Assuming for simplicity that all fermions couple identically to $s$ and that there are $N_f$ copies of them, we get $$\Lambda_3 = 3\pi \frac{m_f}{N_f y_f}\frac{4}{3}\frac{1}{ \left| F^{s^{\pm}}_{1/2}(\tau_f) \right|}.$$ When $m_f \gtrsim m_s$, the form factor $F$ becomes $\left| F^{s^{+}}_{1/2}(\tau_f) \right| \simeq 4/3$ for a CP-even $s$ and $\left| F^{s^{-}}_{1/2}(\tau_f) \right| \simeq 2$ for a CP-odd one. We can then write $$\label{eq:l3uv}
\Lambda_3=\begin{cases}
\frac{3 \pi m_f}{N_f y_f} \qquad \text{(scalar)}\\
\frac{2 \pi m_f}{N_f y_f} \qquad \text{(pseudoscalar)}.
\end{cases}$$ If we assume coloured fermions with a mass of 1 TeV, a value compatible with the latest experimental limits on heavy quark masses [^5] [@CMS:2015alb], for $N_f=1$ and $y_T=1$, Eq. leads to a large value for the scale $\Lambda_3 \gtrsim 6$ TeV. Lower values, down to $\sim$ 400 GeV, can be obtained assuming higher fermion multiplicities and/or larger couplings to the resonance $s$. For example for $N_f=y_f=5$ we obtain $\Lambda_3=$ 400 GeV and 250 GeV for the CP even and CP odd case respectively.
Similarly, in the EW sector the decay width $\Gamma(s \rightarrow \gamma\gamma)$, assuming the process is mediated by loops of fermions $f$, reads [@Djouadi:2005gi] $$\label{eq:l12uv}
\Gamma_{\rm UV}(s \to \gamma\gamma) = \frac{\alpha^2 m_s^3} {256\pi^3}
\bigg| \sum_f N^c_f Q_f^2 \frac{y_{f}}{m_f} F^{s^{\pm}}_{1/2}(\tau_f)\bigg|^2$$ where all the factors follow from Eq. apart from the fine structure constant $\alpha$, the color factor $N^c_f$ and the electric charges of the fermions running in the loop, $Q_f$. In our effective description, taking $\Lambda_1 = \Lambda_2 \equiv \Lambda_{1,2}$, the corresponding expression becomes $$\Gamma_{\rm EFT}(s \to \gamma\gamma) = \frac{\alpha^2 m_s^3}{16 \pi \Lambda_{1,2}^2}.$$ We can then establish the correspondence $$\Lambda_{1,2} = \frac{4\pi m_f}{N_f Q_f^2 y_f \left| F^{s^{\pm}}_{1/2}(\tau_f) \right|}$$
The form factor $F$ attains its maximal value close to the threshold $m_f\sim m_s/2$ (note that one has to consider $m_f \gtrsim m_s/2$ so as to avoid the tree level decay of $s$ into a pair of heavy fermions). The explicit value is $\left| F^{s^{+}}_{1/2}(\tau_f) \right| \simeq 2$ and $\left| F^{s^{-}}_{1/2}(\tau_f) \right| \simeq 5$ for the CP-even and CP-odd cases respectively. Taking then $m_f\sim m_s/2$ and assuming the heavy fermions to be neutral under $SU(3)_c$ and, again for simplicity, to all couple identically to $s$ we obtain $$\label{eq:l3uvbis}
\Lambda_{1,2}\sim\begin{cases}
\frac{2350~\text{GeV}}{\left(N_f Q_f^2 y_f \right)} & \text{scalar}.\\
\frac{950~\text{GeV}}{\left(N_f Q_f^2 y_f \right)} & \text{pseudoscalar}.
\end{cases}$$ It is then clear that, at least for $Q_f=1$, achieving the lowest $\Lambda_{1,2}$ scales we consider in our analysis (20 GeV) is quite difficult in such a picture involving vector-like fermions, for both the cases of a CP even or CP odd scalar, even if the perturbativity limits are saturated for each fermion. Note, however, that $\Lambda_{1,2}$ needs not be interpreted as coming from such a type of UV completion but could instead parametrize some appropriate strong dynamics. Besides, for the higher values of $\Lambda_{1,2}$ considered in our analysis perturbative embeddings of the Lagrangians and can be envisaged fairly easily. For example, taking again $N_f=y_f=5$, we obtain $\Lambda_{1,2}=100$ and 40 GeV for the CP even and CP odd case respectively. Note that even if the theory is perturbative at the input scale, renormalization group evolution of the couplings may lead to the apparition of Landau poles at scales of a few TeV. A discussion of such effects can be found in [@Franceschini:2015kwy].
Some more details on direct detection {#app:DD}
=====================================
For convenience, we recall here the formalism relevant to the computation of the DM-nucleon spin-independent scattering cross section, following closely Ref. [@Hisano:2010ct]. Integrating out the scalar $s$ in Eq. , we obtain an effective coupling of $\psi$ pairs to gluons described, to lowest order, by the Lagrangian $${\cal{L}_{\rm eff}} = f_G \bar{\psi} \psi G_{\mu\nu} G^{\mu\nu}$$ where in our conventions the coefficient $f_G$ is given by $$f_G \equiv \frac{y_{\psi}}{2} \frac{\alpha_s}{4\Lambda_3} \frac{1}{m_s^2} \ .$$ The spin-independent scattering cross section is then simply computed by $$\sigma_{\rm SI} = \frac{4}{\pi} \mu_{\psi N}^2 \left| f_p \right|^2$$ where the amplitude $f_p$ reads $$f_p = m_p \frac{8\pi}{9\alpha_s} f_G f_{TG}$$ and $f_{TG}$ is the gluon form-factor. The latter can be related to the standard $f_{Tq}$ quantities that describe the “quark content” of the nucleon, $f_{TG} = 1 - \sum_{q = u,d,s} f_{Tq}$. The constraints quoted in Sec. \[sec:darkmatter\] are based on the choice $f_{Tu} = 0.0153$, $f_{Td} = 0.0191$ and $f_{Ts} = 0.0447$ (which is also the default choice in the public code [micrOMEGAs]{}).
It should be noted that the cross section depends quite strongly on the choices for the $f_{Tq}$ quantities. For example, older computations of $\sigma_{\rm SI}$ used a much larger value for $f_{Ts}$, which would decrease the predicted cross section. All recent lattice simulations point to values close to the ones we have used. Furthermore, the spin-independent cross section changes quite drastically once couplings to quarks are turned on. In particular, as also pointed out in [@Mambrini:2015wyu], couplings to heavy quarks tend to cancel out the gluon contribution. It is then clear that the behaviour of $\sigma_{\rm SI}$ in a UV-complete model could indeed be fairly different than the one predicted by the Lagrangian .
[^1]: For an earlier study of such interactions see, for example, [@Jaeckel:2012yz].
[^2]: Given the preliminary nature of the excess seen in the early 13 TeV data, the main uncertainties do not come from the analysis setup but rather from the experimental side. In this respect these details are given for completeness and to render our analysis more transparent.
[^3]: Other dedicated DM searches for $j + E_{T}^{\rm miss}$ final states exist and can also be used. These searches, e.g. [@Khachatryan:2014rra], contain several signal regions corresponding to different $j + E_{T}^{\rm miss}$ cuts. The cuts on the analysis used in this study are nonetheless comparable to the ones used in DM searches.
[^4]: Since the monojet analysis is expected to have a mild dependence on the $CP$ properties of the mediator [@Buckley:2014fba; @Harris:2014hga; @Chala:2015ama], for simplicity we have only performed our computations for the scalar case.
[^5]: For a consistent UV completion it is important to mention the necessity to decay these NP states. This can be achieved by introducing a linear mixing between the heavy quarks and the SM fermions, for example the top quark. While this introduces a certain degree of model dependence in the discussion, we assume this mixing to be small enough so that the $s f\bar f$ interaction does not cause a large $s\to t \bar t$ decay rate, while leaving the previous discussion on the loop induced $ggs$ coupling unaffected.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We re-examine the likelihood for alien civilizations to develop communication technology on the basis of the general assumption that life elsewhere could have a non-carbon chemical foundation. We particularized the discussion to a complex silicon-based biochemistry in a nitrogen solvent, and elaborate on the environment in which such a chemistry is feasible, and if so, on what scales. More concretely, we determine the region outside the habitable zone where such organisms can grow and flourish and after that we study how our findings impact the recently derived upper limit on the fraction of living intelligent species that develop communication technology $\langle \xi_{\rm biotec} \rangle$. We also compare this new restriction on $\langle \xi_{\rm biotec} \rangle$ with that resulting from the extension of the habitable zone to accommodate subsurface exolife, originating in planets with subsurface (water) oceans.'
author:
- 'Luis A. Anchordoqui'
- 'Susanna M. Weber'
date: August 2019
title: Upper limit on the fraction of alien civilizations that develop communication technology
---
=1
Introduction
============
day on which the Earth began to move is a seminal moment in world history. In the XVI century, Copernicus’ theory of heliocentrism transformed a millennia-old worldview with the shocking revolutionary idea that we humans are not at the center of the universe. We may be again at the verge of an exciting moment in the history of mankind. In the last few decades, the so-called “exoplanet” revolution has shown us that the universe is awash in alien worlds and that we humans may not be alone. The heedful quest for “exolife” has been dubbed the [*cosmic modesty conjecture*]{}: “The richness of the universe teaches us modesty and guides us to search for both primitive and intelligent forms of life elsewhere without prejudice” [@Loeb:2017cm].
In its most restrictive incarnation the cosmic modesty conjecture calls for carbon-based organisms, operating in a water-based medium, with higher forms (perhaps) metabolizing oxygen. All forms of life on Earth share this same basic biochemistry. Indeed, the concept of life based around anything other than carbon certainly seems outlandish at first. However, our present knowledge of physics does not guarantee this fact, rather falling into what Sagan referred to as “carbon chauvinism” [@Sagan]. In this work we re-examine the likelihood for alien civilizations to develop communication technology. We adopt an extreme viewpoint of the cosmic modesty conjecture and argue that life is a sort-of “nanotechnology phenomenon” of “molecular automaton” style, and that it is the liquid nature where life evolves what determines the biochemistry of non-terrestrial life (rather than a certain biochemistry being required for life’s existence).
The number of intelligent civilizations in our galaxy at any given time capable of releasing detectable signals of their existence into space can be cast in a quite simple functional form, $$N = \Gamma_\star \ f_{\rm p} \ n_{\rm e} \ f_{\ell} \ f_{\rm i} \ f_{\rm c} \ L_\tau \,,
\label{Drake}$$ where $\Gamma_\star$ is the average rate of star formation, $f_{\rm p}$ is the fraction of stars with planetary systems, $n_{\rm e}$ is the number of planets (per solar system) with a long-lasting ($\sim 4~{\rm Gyr}$) ecoshell, $f_\ell$ is the fraction of suitable planets on which life actually appears, $f_{\rm i}$ is the fraction of living species that develop intelligence, $f_{\rm c}$ is the fraction of intelligent species with communications technology, and $L_\tau$ is the length of time such civilizations release detectable signals into space (i.e. the lifetime of the [*communicative phase*]{}) [@Drake].
A more compact form of (\[Drake\]) can be obtained by considering separately its astrophysical and biotechnological factors $$N = \langle \zeta_{\rm astro} \rangle \, \langle \xi_{\rm biotec} \rangle \, L_\tau \, ,
\label{Dshort}$$ where $\langle \zeta_{\rm astro} \rangle = \Gamma_\star \, f_{\rm p} \, n_{\rm e}$ represents the production rate of habitable planets with long-lasting ecoshell (determined through astrophysics) and $\langle \xi_{\rm biotec} \rangle = f_\ell \, f_{\rm i} \, f_{\rm c}$ represents the product of all chemical, biological and technological factors leading to the development of a technological civilization [@Prantzos:2013hg]. $\langle \cdots \rangle$ indicates average over all the multiple manners civilizations can arise, grow, and develop such technology, starting at any time since the formation of our Galaxy in any location inside it. This averaging procedure must be regarded as a crude approximation because the characteristics of the initial conditions in a planet and its surroundings may affect $f_\ell$, $f_{\rm i}$, and $f_{\rm c}$ with high complexity. In a recent study we estimated the production rate of exoplanets where carbon-based organisms operating in a water-based medium can flourish and the rate of planetary catastrophes which could threaten the evolution of life on the surface of these worlds [@Anchordoqui:2017eog]. Armed with these estimates we used our current measurement of $N = 0$ to set an upper limit on $\langle \xi_{\rm biotec} \rangle$. In this work we extend our study by considering that life elsewhere could have a non-carbon chemical foundation.
The layout of the paper is as follows. We begin in Sec. \[sec:2\] with an overview of the circumstellar habitable zone. In Sec. \[sec:3\] we explore wether life may develop outside the habitable zone. In Sec. \[sec:4\] we combine the odds of finding exolife different from us with the non observation of artificial signals from beyond Earth to revise the upper limit on the average fraction of living intelligent species that develop communication technology. We conclude in Sec. \[sec:5\] with some implications of our results.
The Habitable Zone {#sec:2}
==================
planet $r_p$ ($10^9$ m) $\mathscr{F}_{r_p}$ (W m$^{-2}$) $\alpha_p$ $T_p$ (K) $T_{p_{\rm measured}}$ (K) $T_{p_{\rm surface}}$ (K) $\tau$ (Earth days)
------------------ ---------------------------- -------------------------------------------- ---------------------- --------------------- -------------------------------------- ------------------------------------- -------------------------------
Venus 108 2632 0.77 227 230 760 243
Earth 150 1367 0.30 255 250 288 1.00
Mars 228 589 0.24 211 220 230 1.03
Jupiter 780 51 0.51 103 130 134 0.41
Life needs of stars for at least two reasons:
- stars are required to synthesize heavy elements (such as carbon, oxygen, $\cdots$, iron) out of which rocky planets and the molecules of life are made;
- stars maintain a source of heat to power the chemistry of life on the surface of their planets.
For example, the Earth receives almost all of its energy from the Sun. We can think of the Sun as a blackbody radiator at temperature $T_\odot \simeq 5,777~{\rm K}$. The Stefan-Boltzmann law gives us a way of calculating the power radiated per unit area, $${\rm Sun's \ emitted \ radiation \ per \ unit \ area} = \sigma T_\odot^4,
\label{uno}$$ where $\sigma = 5.67 \times 10^{-8}~{\rm W m^{-2} K^{-4}}$. At the present time in its evolution the Sun emits energy at a rate of $$\begin{aligned}
{\rm Solar \ luminosity} \equiv L_\odot & = & 4 \pi \ R_\odot^2 \ \sigma \ T_\odot^4 \nonumber \\
& \simeq & 3.87 \times 10^{26}~{\rm W} \, ,
\label{dos}\end{aligned}$$ where we have assumed that the Sun is a sphere of radius $R_\odot = 6.95 \times 10^8~{\rm m}$.
The flux of solar energy at the Earth, generally refer to as the “solar constant”, depends on the distance of the Earth from the Sun, $r_\oplus$, and is given by the inverse square law $${\rm Solar \ constant} \equiv \mathscr{F}_{r_\oplus} = \frac{L_\odot}{4\pi r_\oplus^2} \, .
\label{tres}$$ Obviously, due to variations in the orbit of the Earth the solar constant is not actually constant. The estimate set out in Table \[tabla1\], $\mathscr{F}_{r_\oplus} = 1,367~{\rm W m}^{-2}$, corresponds to the average value which results from the average Earth-Sun distance of $r_\oplus = 1.5 \times 10^{11}~{\rm m}$.
type of surface albedo (%)
----------------------------- --------------------------
ocean 2 - 10
forest 6 - 18
cities 14 - 18
grass 7 - 25
soil 10 - 20
grassland 16 - 20
desert (sand) 35 - 45
ice 20 -70
cloud (thin, thick stratus) 3, 60 -70
snow (old) 40 -60
snow (fresh) 75 - 95
: Albedos for different surfaces. \[tabla2\]
Since water is essential for life as we know it, the search for biosignature gases naturally focuses on planets located in the [*habitable zone*]{} of their host stars, which is defined as the orbital range around the star within which surface liquid water could be sustained. Each star is surrounded by a [*habitable zone*]{}.
To estimate the orbital range of the habitable zone in the solar system we consider the energy balance of any planet intercepting the solar energy flux and radiating energy away. Suppose that a planet of radius $R_p$ absorbs all incident light (100%) from the Sun. The planet blocks out an area of $\pi R_p^2$, and so this is the area we will use to find the power that is absorbed by the planet. Note that we do not use the whole surface area of the spherical planet, since the whole planet does not intercept the sunlight all at the same time. We also do not use half of the whole surface area of the spherical planet, since even though half of the planet is in “daylight”, the areas that are not perpendicular to the rays of sunlight should not be given as much importance as the areas that are perpendicular to the rays of sunlight. We need to use the two dimensional area blocked out by the spherical planet, and since a sphere always casts a circular shadow, we need only find the area of a circle of radius $R_p$. The power input to the planet is given by $${\rm Solar \ power \ incident \ on \ planet} = \mathscr{F}_{r_p} \ \pi \ R_p^2 \, ,$$ and so for planet Earth we have $$\begin{aligned}
{\rm Solar \ power \ incident\ on \ Earth} & = & \mathscr{F}_{r_\oplus} \ \pi \ R_\oplus^2 \nonumber \\ & \simeq & 1.74 \times 10^{17}~{\rm W} \, .\end{aligned}$$ However, not all of this radiation is absorbed by the planet, but rather a significant fraction is reflected. The ratio of reflected to incident solar energy is called the albedo, $\alpha_p$, which depends on the nature of the reflecting surface (it is large for clouds and light surfaces such as deserts, snow, and ice), see Table \[tabla2\]. Under the present terrestrial conditions of cloudiness and snow and ice cover, on average a fraction $\alpha_\oplus \sim 0.30$ of the incoming solar radiation at the Earth is reflected back to space. The solar radiation absorbed by a planet is then $$\mathscr{F}_{p,{\rm abs}} = (1 -\alpha_p) \ \mathscr{F}_{r_p} \ \pi R_p^2 \, .
\label{seis}$$ Estimates of exoplanet’s albedos are given in [@DelGenio]. In equilibrium, the total flux radiated by the planet into space must balance the radiation it absorbes. Now, we assume that the planets emits in all directions like a blackbody of uniform temperature $T_p$ (known as the “effective planetary temperature”, or “emission temperature”) the Stefan-Boltzmann law gives: $${\rm Planet's\ emitted \ radiation \ per \ unit \ area} = \sigma T_p^4,$$ and so the total radiation emitted by the planet is $${\rm Planet's\ emitted \ radiation} \equiv \mathscr{F}_{p,{\rm em}} = 4 \pi\ R_P^2 \ \sigma \ T_p^4 \,.
\label{ocho}$$ Note that (\[ocho\]) predicts the temperature one would infer by looking at the planet if a black body curve were fitted to the measured spectrum of outgoing radiation, and therefore this can be taken as the definition of the emission temperature. We can now equate (\[seis\]) and (\[ocho\]) to obtain $$T_p = \left[ \frac{\mathscr{F}_{r_p} (1 -\alpha_p)}{4\sigma} \right]^{1/4} \, .
\label{Tp}$$ It is noteworthy that the radius of the Earth has cancelled out. This means that $T_p$ depends only on the planetary albedo and the distance of the planet from the Sun. Substituting for the values given in Table \[tabla1\] we find that the Earth has an emission temperature of $T_\oplus \sim 255~{\rm K}$.
Using (\[dos\]) and (\[tres\]) we estimate the flux of solar energy for any planet in the solar system, $$\mathscr{F}_{r_p} = \frac{R_\odot^2 \sigma T_\odot^4}{r^2_p }\,.
\label{diez}$$ Substituting (\[diez\]) into (\[Tp\]) and rearranging we obtain $$r_p = \frac{R_\odot \, T_\odot^2 \, \sqrt{1 - \alpha_p} }{2 \ T_p^2} \, .
\label{once}$$ Using (\[once\]) it is straightforward to estimate the range of distances that determine the habitable zone by requiring $273 < T_p/{\rm K} < 373$.
In Table \[tabla1\] we show several parameters for some of the planets in the solar system and we compare the approximate measured values with those computed using (\[Tp\]). In general, there is a good agreement. Jupiter, however, is an exception due to the fact that about one-half of the energy input comes from the gravitational collapse of the planet.
A point worth noting at this juncture is that the temperature of a planet as derived in (\[Tp\]) has small corrections due to atmospheric effects. More concretely, the so-called “natural greenhouse effect” is the process by which radiation from a planet’s atmosphere warms the planet’s surface to a temperature above what it would be without this atmosphere. Radiatively active gases (i.e., greenhouse gases) in a planet’s atmosphere radiate energy in all directions. Obviously, such a greenhouse effect is not the same on all planets, and differs dramatically based on the thickness and composition of the atmosphere. Three planets that show how dramatically the conditions of a planet can change with the different levels of the greenhouse effect are Venus, Earth, and Mars. For example, $T_\oplus$ is nearly 40 K cooler than the globally averaged observed surface temperature. Without the greenhouse effect, the temperature on Earth would be too cold for life.
There are four main naturally occurring gases that are responsible for the greenhouse effect: water vapor, carbon dioxide, methane and nitrous oxide. Of these gases, water vapor has the largest effect. Once these gases absorb energy, the gas particles begin to vibrate and they radiate energy in all directions, including approximately 30% of it back towards Earth.
The other two important greenhouse gases are ozone and halocarbons. Despite the fact that most of the greenhouse gases occur naturally in the atmosphere, some are man-made and the most well-known of these are fluorocarbons. Since the industrial revolution, human activities have also resulted in an increase in natural greenhouse gases, especially carbon dioxide. An increase in these gases in the atmosphere enhances the atmosphere’s ability to trap heat, which leads to an increase in the average surface temperature of the Earth.
For a really strong greenhouse effect, we should look at Venus. Venus is similar to Earth in terms of size and mass, but its surface temperature is about 760 K. This is hot enough to melt lead! The Venusian atmosphere is mainly made up of carbon dioxide, a greenhouse gas.
In summary, the habitable zone is the orbital range around a star within which surface liquid water could be sustained. Since water is essential for life as we know it, the search for biosignature gases naturally focuses on planets located in the habitable zone of their host stars. The habitable zone of the solar system looks like a ring around the Sun. Rocky planets with an orbit within this ring may have liquid water to support life. The habitable zone around a single star looks similar to the habitable zone in our Solar System. The only difference is the size of the ring. If the star is bigger than the Sun it has a wider zone, if the star is smaller it has a narrower zone. It might seem that the bigger the star the better. However, the biggest stars have relatively short lifespans, so the life around them probably would not have enough time to evolve [@Loeb:2016vdd]. The habitable zones of small stars face a different problem. Besides being narrow they are relatively close to the star. A hypothetical planet in such a region would be tidally locked [@Seager]. That means that one half of it would always face the star and be extremely hot, while the opposite side would always be facing away and freezing. Such conditions are not very favorable for life.
The frequency $\eta_\oplus$ of terrestrial planets in and the habitable zone of solar-type stars can be determined using data from the [*Kepler*]{} mission [@Borucki:2010zz; @Borucki:2011nn; @Batalha:2012gh]. Current estimates suggest $0.15^{+0.13}_{-0.06} < \eta_\oplus < 0.61^{+0.07}_{-0.15}$ [@Dressing:2013mid; @Kopparapu:2013xpa].
Life outside the Habitable Zone {#sec:3}
===============================
It has been pointed out that life outside the habitable zone may be possible on planets with subsurface oceans [@Lingam:2017]. Allowing for the possibility of subsurface ocean worlds yields a frequency of planets $\eta \sim 1$. Now, because the planets with subsurface oceans outside the habitable zone are more common than rocky planets in the habitable zone, one may wonder why do we find ourselves on the latter. The answer to this question most likely stems from the fact that “we” refers to an intelligent, conscious and technologically sophisticated species. In other words, albeit the probability of life on subsurface worlds may be non-negligible, it is quite plausible that the likelihood of technological life could instead be selectively lowered. In this section we explore the possibility that life elsewhere could have a non-carbon chemical foundation; e.g., in the spirit of [@Reynolds; @Wells] we envision a race of intelligent silicon-based life forms.
Considering that Earth is the only reference point we have when studying life, it is unsurprising that biochemistry has always been connected to the elements of carbon, hydrogen, oxygen, and nitrogen. Moreover, carbon can form bonds with many other non-metals, as well as large polymers. These unique qualities have led many to argue that carbon is a pre-requisite for the existence of even very simple life. However, this must not necessarily be true. It has long been suspected that silicon and germanium can enter into some of the same kind chemical reactions than carbon does [@Sagan]. Recent research in both chemistry and astrobiology has shown that it is theoretically quite feasible for silicon to form complex, self-replicating systems similar to the ones that produced the first, simple forms of life on Earth [@Bains; @Schulze-Makuch]. More concretely:
- Silicon is able to form stable covalent bonds with itself, as well as stable compounds with carbon and oxygen [@West]. These structures can form many diverse systems, including ring systems, which could be analogs to sugars, a key component of biochemistry on Earth. This stability is a prerequisite for building the complex chemical structures that support life on Earth, making silicon a strong contender.
- Silanols, the silicon containing analogues of alcohols have surprising solubility properties, with diisobutylsilane diol being soluble in water and hexane [@Lickiss]. Solubility is another crucial factor in the development of life, since having a solvent and a substance is the model for early development of life that we see on Earth.
- Silicon’s [*chiral*]{} properties. All life on Earth is made of molecules that twist in the same direction, that is they have an inherent [*handedness*]{}. In other words, each of life’s molecular building blocks (amino acids and sugars) has a twin: not an identical one, but a mirror image. On Earth, the amino acids characteristic of life are all “left-handed” in shape, and cannot be exchanged for their right-handed doppelgänger. Meanwhile, all sugars characteristic of life on Earth are “right-handed.” The opposite hands for both amino acids and sugars exist in the universe, but they just are not utilized by any known biological life form. (Some bacteria can actually convert right-handed amino acids into the left-handed version, but they cannot use the right-handed ones as is.) This phenomenon of biological shape selection is called [*chirality*]{} – from the Greek for handedness. We say that both sugars and amino acids on Earth are homochiral: one-handed.
Though we are still unsure why it is that the molecules of carbon-based life choose only one orientation, it seems reasonable to require that in order for silicon to replicate the processes that originated life on Earth the molecules must also be chiral, and exist in a left- or right-handed forms in potential living environments. There is certainly reason to be optimistic: an observation of chirality in noncrystalline silica chiral nano-ribbons has been reported in [@Okazaki].
- Silicon’s high reactivity is a barrier to forming complex structures on Earth, as this high rate of reaction leaves little time for construction. However, this only holds true for environments with a climate similar to earth. On the outskirts of the solar system, where the reactivity of carbon is severely impacted by the drop in temperature, silicon’s high reaction rate could be the key to the development of life in these cryogenic environments, allowing it to flourish where carbon based life would be impossible.
One probable environment for silicon life is liquid nitrogen [@Bains]. Nitrogen is one of the few substances that can still dissolve silicon at very cold temperatures, as solubi. Additionally, silicon is able to form stable covalent bonds with nitrogen, as well as with itself.
The habitable zone for silicon life would then depend on the area around a star in which nitrogen is a liquid. Neptune’s moon, Triton, has been considered a candidate for surface level nitrogen lakes [@Broadfoot]. Triton is the only large satellite in the solar system to circle a planet in a retrograde direction, i.e. in a direction opposite to the rotation of the planet. The retrograde orbit and Triton’s relatively high density suggest that this satellite may have been captured by Neptune as it traveled through space several billion years ago. If this were the case, tidal heating could have melted Triton in its originally eccentric orbit, and the satellite might have been liquid for as long as one billion years after its capture by Neptune. However, presently Triton is quite cold, with a surface temperature of 38 K, and an extremely thin atmosphere (the atmospheric pressure at Triton’s surface is about 14 microbars, 1/70,000th the surface pressure on Earth). Nitrogen ice particles might form thin clouds a few kilometers above the surface. Hence, even though the surface temperature is below the freezing point of liquid nitrogen it is reasonable to assume that the albedo of a hypothetical planet that could support silicon life will be similar to that of Triton, $\alpha_{\rm Triton} \sim 0.6$ [@McEwen].
Next, using (\[once\]) we determine the habitable zone of silicon-based life for a main sequence star like our sun, with temperature $T_\odot$ and radius $R_\odot$. We take the planetary surface temperature in between the boiling and freezing point of liquid nitrogen, $63.15 < T_p/{\rm K} < 77.36$. Plugging in these values in (\[once\]) we find that for a main sequence star like our sun, the habitable zone of silicon-based life stretches from 1.24 billion km to 1.85 billion km from the star. We can now estimate what planets within the solar system fall into the silicon habitable zone during all parts of their orbit. The two planets closest to the silicon habitable zone are Saturn and Uranus. Saturn has a perihelion of 1.35 billion km and aphelion of 1.51 billion km, meaning that it is within the proper distance range for silicon biochemistry. However, Saturn is mostly a gas planet, and thus unsuitable for supporting any life. Uranus, on the other hand, has a perihelion of 2.75 billion km and aphelion 3.00 billion km, making it too cold for surface lakes or oceans of nitrogen. This result is also in agreement with the commonly accepted surface temperature of Uranus, roughly 57 K [@Lunine:1993fz], which is below the freezing point of nitrogen.
We expand our focus to include ultra-cool stars, such as TRAPPIST-1A, as they are the most common stars in the Milky Way, and thus their orbiting planets are representative of “average” star systems. More concretely, M-dwarfs like Proxima Centauri and TRAPPIST-1 are 10 times more abundant than the Sun [@Chabrier:2003ki; @Robles:2008ww] and have stellar lifetimes that are about 100 to 1000 times greater [@Tarter:2006dy; @Adams:1996xe; @Loeb:2016vps]. Furthermore, exoplanets around these stars are easier to detect (the transit signals produced by Earth-sized planets are 80 times stronger than the signal produced by similar planets transiting a Sun-like star) and their atmospheres can be analyzed via transit spectroscopy, thus enabling the ready detection of biomarkers [@Fujii]. However, various physical mechanisms could act in concert to suppress the likelihood of Earth-based life on M-dwarf exoplanets relative to their counterparts around solar-type stars [@Lingam:2018dky]. Nevertheless, this may not be the case for silicon-based alien life forms. Herein, we evaluate the TRAPPIST-1 system as representative of ultra cool stars, for which $T_\star = 2,511~{\rm K}$, and $R_\star = 84,179.7~{\rm km}$ [@VanGrootel]. We generalize (\[once\]) substituting $T_\odot$ by $T_\star$ and $R_\odot$ by $R_\star$, to find that the habitable zone for ultra-cool dwarf stars encompasses a distance range between $1.6$ million km to $3.0$ million km from the planet’s star, whereas for silicon-based life on nitrogen lakes the habitability circumstellar region spans the orbital range within $28$ million km and $42$ million km. This seems to indicate the frequency of planets hosting any form of life must be extended. As for subsurface ocean worlds, we may take $\eta \sim 1$ for intelligent, conscious and technologically sophisticated species.
Upper Limit on $\bm{\langle \xi_{\rm biotec} \rangle}$ {#sec:4}
======================================================
Next, in line with our stated plan, we derive an upper limit on $\langle \xi_{\rm biotec} \rangle$. The present day star formation rate in the Galaxy is estimated to be $\dot M_\star = 1.65 \pm 0.19~M_\odot~{\rm yr}^{-1}$ [@Licquia:2014rsa; @Chomiuk:2011fc]. This estimate has been derived assuming the Kroupa initial mass function (IMF) [@Kroupa:2002ky; @Kroupa:2003jm]. The shape of this IMF is lognormal-like and exhibits a peak around $M/M_\odot \approx 0.4$ [@Lada], suggesting there are roughly 2 stars per $M_\odot$. Altogether, this yields $\Gamma_\star \approx 3~{\rm yr}^{-1}$. Before proceeding, we pause to note that the average star formation rate in the Galaxy could be about 4 times the current rate [@Kennicutt:2012ea]. Now, only 10% of these stars are appropriate for harboring habitable planets. This is because the mass of the star $M_\star < 1.1 M_\odot$ to be sufficiently long-lived (with main sequence lifetimes larger than 4.5 Gyr) and $M_\star > 0.7 M_\odot$ to possess circumstellar habitable zones outside the tidally locked region [@Seager]. The production rate of habitable planets is then $$\zeta \sim 0.045 \left(\frac{\Gamma_\star}{3~{\rm yr^{-1}}}\right) \left(\frac{\eta}{0.15} \right)~{\rm yr}^{-1} \,,$$ with $0.15 < \eta <1$ and $3 <\Gamma_\star/{\rm yr}^{-1} < 12$.
Now, a habitable planet must survive and remain in a habitable zone to present day. The potential hazard of nearby gamma-ray bursts (GRBs) has been estimated elsewhere [@Anchordoqui:2017eog]. Long GRBs are associated with supernova explosions. When nuclear fuel is exhausted at the center of a massive star, thermal pressure can no longer sustain gravity and the core collapses on itself. If this process leads to the formation of a rapidly spinning black hole, accreted matter can be funneled into a pair of powerful relativistic jets that drill their way through the outer layers of the dying star. If the jet is pointing towards Earth, its high-energy emission is seen as a GRB.
The luminosity of long GRBs – the most powerful ones – is so intense that they are observed about once a day from random directions in the sky. The physical conditions in the dissipation region produce a heavy flux of photons with energies above about 100 keV. If one GRB were to happen nearby, the intense flash of gamma rays illuminating the Earth for tens of seconds could severely damage the thin ozone layer that absorbs ultraviolet radiation from the Sun. Indeed a fluence of $100~{\rm kJ/m^2}$ would create a depletion of 91% of this life-protecting layer on a timescale of a month, via a chain of chemical reactions in the atmosphere. This would be enough to cause a massive life-extinction event [@Ruderman; @Thorsett:1995us; @Dar:1997he; @Thomas:2004vj; @Thomas:2005gpa; @Melott:2003rs; @Piran:2014wfa].
The critical time $t_c$ for life to arise and evolve becomes the dominant uncertainty on estimating the probability of having at least one lethal GRB, with a critical fluency $100~{\rm kJ/m}^2$. Life has been evolving on Earth for close to 4 Gyr [@Mojzsis; @Dodd], but complex life is well under 1 Gyr old, and intelligent life is only a Myr old at most. In what follows we adopt $t_c =1~{\rm Gyr}$ and 4 Gyr as critical time intervals for life evolution [@Loeb:2016vps]. With this in mind, the probability for a GRB to destroy an entire alien race $/\!\!\!\zeta$ is estimated to be [@Anchordoqui:2017eog] $$0.044 \alt /\!\!\!\zeta \alt 0.22 \, .$$ To derive the upper limit on $\langle \xi_{\rm biotect} \rangle$ we must take as fiducial parameters those giving the possible lowest value of $\langle \zeta_{\rm astro} \rangle \sim \zeta /\!\!\!\zeta$, i.e. $$\langle \zeta_{\rm astro} \rangle \sim 2 \times 10^{-3} \left(\frac{\Gamma_\star}{3~{\rm yr}^{-1}}\right) \left(\frac{\eta}{0.15} \right) \left(\frac{/\!\!\!\zeta}{0.044}\right)~{\rm yr}^{-1} \, .
\label{zetaastro}$$
Finally, to determine the upper bound on $\langle \xi_{\rm biotec} \rangle$ we must decide on the possible minimum $L_\tau$. Herein we consider $L_\tau > 0.3~{\rm Myr}$ such that $c L_\tau \gg$ propagation distances of Galactic scales ($\sim 10~{\rm kpc}$). This would provide enough time to receive electromagnetic (and/or high-energy neutrino [@Learned:2008gr]) signals from any advanced civilization living in the Milky Way which is trying to communicate with us.
The non observation of artificial signals from beyond Earth prevents an estimate of the event rate. Our best estimate of the proportion of cases which have a signal (i.e. an event) is zero, but there will be uncertainty in this estimate. Just because we have not seen an event yet does not mean we will never see one. We need a confidence interval for this estimate. If a corresponding hypothesis test is performed, the confidence level (CL) is the complement of the level of statistical significance, e.g, a 95% confidence interval reflects a significance level of 0.05. Because the number of events observed is zero, we cannot use the usual standard error estimate for the confidence interval. Instead, we use a small sample confidence interval for the estimate, based on the exact probabilities of the Binomial distribution. In the absence of background, $N < 3.09$ determines the 95% confidence interval [@Feldman:1997qc].
Models with associated parameters in the right-hand-side of (\[Dshort\]) predicting more than 3.09 events are excluded at 95% CL. Then substituting (\[zetaastro\]) into (\[Dshort\]) with $L_\tau \sim 0.3~{\rm Myr}$ we obtain $$\langle \xi_{\rm biotec} \rangle < 5 \times 10^{-3} \left(\frac{3~{\rm yr}^{-1}}{\Gamma_\star}\right) \left(\frac{0.15}{\eta} \right) \left(\frac{0.044}{/\!\!\!\zeta}\right)$$ at the 95% CL.
Conclusion {#sec:5}
==========
We have re-examined the likelihood for alien civilizations to develop communication technology on the basis of the general assumption that life elsewhere could have a non-carbon chemical foundation. We derived a conservative upper bound on the average fraction of living intelligent species of Earth-like planets that develop communication technology: $\langle \xi_{\rm biotec} \rangle < 5 \times 10^{-3}$ at the 95% CL. The upper limit can be up to a factor of 6 more restrictive if we assume there may be non-carbon based living organisms.
The Breakthrough Listen Initiative, announced in July 2015 as a 10-year 100M USD program, is the most comprehensive effort in history to quantify the distribution of advanced, technologically capable life in the universe [@Isaacson; @Lipman; @Gajjar:2019rvn]. The search for extrasolar biomolecular building blocks and molecular biosignatures of extinct extraterrestrial life will soon be extended to the lunar surface [@Lingam:2019]. The ability to detect alien life may still be years or more away, but the quest is underway.
[999]{}
A. Loeb, [ Cosmic modesty]{}, arXiv:1706.05959 \[astro-ph.EP\].
C. Sagan, [ The cosmic connection]{}, (Anchor Press, 1973). ISBN: 9780385004572
F. D. Drake, [ The radio search for intelligent extraterrestrial life]{}, in [*Current aspects of exobiology*]{}, (Eds. G. Mamikunian and M. H. Briggs, Elsevier Ltd., 1965), p.323. ISBN: 978-1-4832-0047-7
N. Prantzos, [ A joint analysis of the Drake equation and the Fermi paradox]{}, Int. J. Astrobiol. [**12**]{}, 246 (2013) doi:10.1017/S1473550413000037 \[arXiv:1301.6411\].
L. A. Anchordoqui, S. M. Weber and J. F. Soriano, [ Is there anybody out there?]{}, PoS ICRC [**2017**]{}, 254 (2018) doi:10.22323/1.301.0254 \[arXiv:1706.01907 \[astro-ph.HE\]\].
A. D. Del Genio [*et al.*]{}, [ Albedos, equilibrium temperature and surface temperatures of habitable planets]{}, arXiv:1812.06606.
A. Loeb, [ On the habitability of our Universe]{}, arXiv:1606.08926 \[astro-ph.CO\].
S. Seager, [ Exoplanet habitability]{}, Science, [**340**]{}, 577 (2013) doi: 10.1126/science.1232226
W. J. Borucki [*et al.*]{} \[Kepler Collaboration\], [ Kepler planet-detection mission: Introduction and first results]{}, Science [**327**]{}, 977 (2010). doi:10.1126/science.1185402
W. J. Borucki [*et al.*]{} \[Kepler Collaboration\], [ Characteristics of planetary candidates observed by Kepler, II: Analysis of the first four months of data]{}, Astrophys. J. [**736**]{}, 19 (2011) doi:10.1088/0004-637X/736/1/19 \[arXiv:1102.0541\].
N. M. Batalha [*et al.*]{} \[Kepler Collaboration\], [ Planetary candidates observed by Kepler, III: Analysis of the first 16 months of data]{}, Astrophys. J. Supp. [**204**]{}, 24 (2013) doi:10.1088/0067-0049/204/2/24 \[arXiv:1202.5852\].
C. D. Dressing and D. Charbonneau, [ The occurrence rate of small planets around small stars]{}, Astrophys. J. [**767**]{}, 95 (2013) doi:10.1088/0004-637X/767/1/95 \[arXiv:1302.1647\].
R. k. Kopparapu, [ A revised estimate of the occurrence rate of terrestrial planets in the habitable zones around Kepler M-dwarfs]{}, Astrophys. J. [**767**]{}, L8 (2013) doi:10.1088/2041-8205/767/1/L8 \[arXiv:1303.2649\]. M. Lingam and A. Loeb, [ Subsurface exolife]{}, Int. J. Astrobiology [**18**]{} 112 (2019) doi:10.1017/S1473550418000083 arXiv:1711.09908 \[astro-ph.EP\].
J. E. Reynolds, [ Opening address to the British Association for the Advancement of Science]{}, Nature [**48**]{}, 477 (1893).
H. G. Wells, [ Another basis for life]{}, Saturday Review, p. 676 (December 22, 1894).
D. Schulze-Makuch and L. N. Irwin, [ The prospect of alien life in exotic forms on other worlds]{}, Naturwissenschaften [**93**]{}, 155 (2006) doi:10.1007/s00114-005-0078-6
W. Bains, [ Many chemistries could be used to build living systems]{}, Astrobiology [**4**]{}, 137 (2004) doi:10.1089/153110704323175124
R. West, [ Multiple bonds to silicon: 20 years later]{}, Polyhedron [**21**]{}, 467 (2002) doi: 10.1016/S0277-5387(01)01017-8.
P. Lickiss, [ The synthesis and structures of organosilanols]{} Adv. Inorg. Chem. [**42**]{}, 174 (1995).
Y. Okazaki, T. Buffeteau, E. Siurdyban, D. Talaga, N. Ryu, R. Yagi, E. Pouget, M. Takafuji, H. Ihara, and R. Oda, [ Direct observation of siloxane chirality on twisted and helical nanometric amorphous silica]{}, Nano Lett. [**16**]{}, 6411 (2016) doi:10.1021/acs.nanolett.6b02858
A. L. Broadfoot, S. K. Atreya, J. L. Bertaux, J. E. Blamont, A. J. Dessler, T. M. Donahue, W. T. Forrester, D. T. Hall, F. Herbert, J. B. Holberg, D. M. Hunter, V. A. Krasnopolsky, S. Linick, J. I. Lunine, J. C. McConnell, H. W. Moos, B. R. Sandel, N. M. Schneider, D. E. Shemansky, G. R. Smith, D. F. Strobel, R. V. Yelle, [ Ultraviolet spectrometer observations of Neptune and Triton]{} Science [**246**]{}, 1459 (1989) doi:10.1126/science.246.4936.1459.
A. S. McEwen, [ Global color and albedo variations on Triton]{} Geophysical Research Letters [**17**]{}, 1765 (1990) doi:10.1029/GL017i010p01765
J. I. Lunine, [ The atmospheres of uranus and neptune]{}, Ann. Rev. Astron. Astrophys. [**31**]{}, 217 (1993). doi:10.1146/annurev.aa.31.090193.001245 G. Chabrier, [ Galactic stellar and substellar initial mass function]{}, Publ. Astron. Soc. Pac. [**115**]{}, 763 (2003) doi:10.1086/376392 \[astro-ph/0304382\]. J. A. Robles, C. H. Lineweaver, D. Grether, C. Flynn, C. A. Egan, M. B. Pracy, J. Holmberg and E. Gardner, [ A comprehensive comparison of the Sun to other stars: searching for self-selection effects]{}, Astrophys. J. [**684**]{}, 691 (2008) doi:10.1086/589985 \[arXiv:0805.2962 \[astro-ph\]\]. J. C. Tarter [*et al.*]{}, [ A re-appraisal of the habitability of planets around M dwarf stars]{}, Astrobiology [**7**]{}, 30 (2007) doi:10.1089/ast.2006.0124 \[astro-ph/0609799\]. F. C. Adams and G. Laughlin, [ A dying universe: The Long term fate and evolution of astrophysical objects]{}, Rev. Mod. Phys. [**69**]{}, 337 (1997) doi:10.1103/RevModPhys.69.337 \[astro-ph/9701131\]. A. Loeb, R. A. Batista and D. Sloan, [ Relative likelihood for life as a function of cosmic time]{}, JCAP [**1608**]{}, 040 (2016) doi:10.1088/1475-7516/2016/08/040 \[arXiv:1606.08448 \[astro-ph.CO\]\]. Y. Fujii, D. Angerhausen, R. Deitrick, S. Domagal-Goldman, J. L. Grenfell, Y. Hori, S. R. Kane, E. Pallé, H. Rauer, N. Siegler, K. Stapelfeldt, and K. B. Stevenson, [ Exoplanet biosignatures: observational prospects]{} Astrobiology [**18**]{}, 739 (2018) doi:10.1089/ast.2017.1733
M. Lingam and A. Loeb, [ Physical constraints for the evolution of life on exoplanets]{}, Rev. Mod. Phys. [**91**]{}, 021002 (2019) doi:10.1103/RevModPhys.91.021002 \[arXiv:1810.02007 \[astro-ph.EP\]\].
V. Van Grootel, C. S. Fernandes, M. Gillon, E. Jehin, J. Manfroid, R. Scuflaire, A. J. Burgasser, A. Burdanov, L. Delrez, B. O. Demory, J. de Wit, D. Queloz, A. H. M. J. Triaud [ Stellar parameters for Trappist-1]{} Astrophys. J. [**853**]{}, 30 (2018) doi:10.3847/1538-4357/aaa023 arXiv:1712.01911.
T. C. Licquia and J. A. Newman, [ Improved estimates of the Milky Way’s stellar mass and star formation rate from hierarchical bayesian meta-analysis]{}, Astrophys. J. [**806**]{}, 96 (2015) doi:10.1088/0004-637X/806/1/96 \[arXiv:1407.1078\]. L. Chomiuk and M. S. Povich, [ Toward a unification of star formation rate determinations in the Milky Way and other galaxies]{}, Astron. J. [**142**]{}, 197 (2011) doi:10.1088/0004-6256/142/6/197 \[arXiv:1110.4105\].
P. Kroupa, [ The initial mass function of stars: Evidence for uniformity in variable systems]{}, Science [**295**]{}, 82 (2002) doi:10.1126/science.1067524 \[astro-ph/0201098\].
P. Kroupa and C. Weidner, [ Galactic - field IMFs of massive stars]{}, Astrophys. J. [**598**]{}, 1076 (2003) doi:10.1086/379105 \[astro-ph/0308356\].
C. J. Lada, [ Star formation in the local Milky Way]{}, arXiv:1508.02711.
R. C. Kennicutt, Jr. and N. J. Evans II, [ Star formation in the Milky Way and nearby galaxies]{}, Ann. Rev. Astron. Astrophys. [**50**]{}, 531 (2012) doi:10.1146/annurev-astro-081811-125610 \[arXiv:1204.3552\].
M. A. Ruderman, [ Possible consequences of nearby supernova explosions for atmospheric ozone and terrestrial life]{}, Science [**185**]{}, 1079 (1974).
S. E. Thorsett, [ Terrestrial implications of cosmological gamma-ray burst models]{}, Astrophys. J. [**444**]{}, L53 (1995) doi:10.1086/187858 \[astro-ph/9501019\].
A. Dar, A. Laor and N. J. Shaviv, [ Life extinctions by cosmic ray jets]{}, Phys. Rev. Lett. [**80**]{}, 5813 (1998) doi:10.1103/PhysRevLett.80.5813 \[astro-ph/9705008\].
B. C. Thomas, C. H. Jackman, A. L. Melott, C. M. Laird, R. S. Stolarski, N. Gehrels, J. K. Cannizzo and D. P. Hogan, [ Terrestrial ozone depletion due to a Milky Way gamma-ray burst]{}, Astrophys. J. [**622**]{}, L153 (2005) doi:10.1086/429799 \[astro-ph/0411284\]. B. C. Thomas [*et al.*]{}, [ Gamma-ray bursts and the Earth: Exploration of atmospheric, biological, climatic and biogeochemical effects]{}, Astrophys. J. [**634**]{}, 509 (2005) doi:10.1086/496914 \[astro-ph/0505472\].
A. Melott [*et al.*]{}, [ Did a gamma-ray burst initiate the late Ordovician mass extinction?]{}, Int. J. Astrobiol. [**3**]{}, 55 (2004) doi:10.1017/S1473550404001910 \[astro-ph/0309415\]. T. Piran and R. Jimenez, [ Possible role of gamma-ray bursts on life extinction in the Universe]{}, Phys. Rev. Lett. [**113**]{}, 231102 (2014) doi:10.1103/PhysRevLett.113.231102 \[arXiv:1409.2506\].
S. J. Mojzsis, G. Arrhenius, K. D. McKeegan, T. M. Harrison, A. P. Nutman, and C. R. L. Friend, [ Evidence for life on Earth before 3,800 million years ago]{}, Nature [**384**]{}, 55 (1996) doi:10.1038/384055a0.
M. S. Dodd, D. Papineau, T. Grenne, J. F. Slack, M. Rittner, F. Pirajno, J. O’Neil, and C. T. S. Little, [ Evidence for early life in Earth’s oldest hydrothermal vent precipitates]{}, Nature [**543**]{}, 60 (2017) doi:10.1038/nature21377.
J. G. Learned, S. Pakvasa and A. Zee, [ Galactic neutrino communication]{}, Phys. Lett. B [**671**]{}, 15 (2009) doi:10.1016/j.physletb.2008.11.057 \[arXiv:0805.2429\]. G. J. Feldman and R. D. Cousins, [ A Unified approach to the classical statistical analysis of small signals]{}, Phys. Rev. D [**57**]{}, 3873 (1998) doi:10.1103/PhysRevD.57.3873 \[physics/9711021\].
H. Isaacson, A. P. V. Siemion, G. W. Marcy, M. Lebofsky, D. C. Price, D. MacMahon, S. Croft, D. DeBoer, J. Hickish, D. Werthimer, S. Sheikh, G. Hellbourg, J. E. Enriquez [ The breakthrough listen search for intelligent life: Target selection of nearby stars and galaxies]{} arXiv:1701.06227 \[astro-ph.IM\].
D. Lipman, H. Isaacson, A. P. V. Siemion, M. Lebofsky, D. C. Price, D. MacMahon, S. Croft, D. DeBoer, J. Hickish, D. Werthimer, G. Hellbourg, J. E. Enriquez, N. Gizani, [ The breakthrough listen search for intelligent life: Searching Boyajian’s star for laser line emission]{}, arXiv:1812.10161 \[astro-ph.IM\].
V. Gajjar [*et al.*]{}, [ The breakthrough listen search for extraterrestrial intelligence]{}, arXiv:1907.05519 \[astro-ph.IM\]. M. Lingam and A. Loeb, [ Searching the Moon for extrasolar material and the building blocks of extraterrestrial life]{}, arXiv:1907.05427 \[astro-ph.EP\].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Mean field dynamos may explain the origin of large scale magnetic fields of galaxies, but controversy arises over the extent of dynamo quenching by the growing field. Here we explain how apparently conflicting results may be mutually consistent, by showing the role of magnetic helicity conservation and boundary terms usually neglected. We estimate the associated magnetic energy flowing out of the Galaxy but emphasize that the mechanism of field escape needs to be addressed.'
author:
- 'Eric G. Blackman'
- 'George B. Field'
title: 'Mean Field Dynamo Saturation: Toward Understanding Conflicting Results'
---
\#1[\[\#1\]]{}
\#1[(\#1)]{}
\#1[(\[\#1\])]{} Œ[|[**E**]{}]{} \#1[\#1]{} \#1[[*\#1*]{}]{} \#1[[*\#1*]{}]{}
=
\#1 1.25in .125in .25in
Field Growth and Constraining the Turbulent EMF
===============================================
Unlike the turbulent amplification of small scale magnetic energy to near equipartition with the the kinetic energy spectrum, the mean field dynamo (MFD) field generation (c.f. Parker 1979, Kulsrud 1999) on scales $>$ turbulent input scale is controversial (c.f. Field et al. 1999). The MFD equation is $\partial_t\bbb=\curl \lb\bfv \ts \bfb \rb+\lambda\nabla^2 \bbb + \curl (\bbV\ts \bbb)$, with the turbulent EMF $\lb\bfv\ts\bfb\rb
=\alpha\bbb-\beta\curl\bbb$, and pseudo-scalar $\alpha$ and scalar $\beta$. How well MFD growth applies when the when the dynamic magnetic backreaction is included depends on the survival of $\lb\bfv\ts\bfb\rb$. Blackman & Field (2000a) used Ohm’s law and mean field theory (e.g.${\bf B}={\bfb}+{\overline {\bf B}}$; $\langle\bfb\rangle$ =0) to constrain the dynamic value of $\lb \bfv\ts \bfb\rb$ analytically. Deriving $\lb\bfv\ts\bfb\rb\cdot{\bbB}
/c = -\eta\lb\bfj\cdot\bfb\rb +\lb\bfe\cdot\bfb\rb$ and then expanding the fluctuating electric field $\bfe$ into its potentials, gives $\lb\bfe\cdot\bfb\rb = -\partial_t\lb\bfa\cdot\bfb\rb/2c+\nabla\cdot
\lb\bfa\ts \bfe -\phi\bfb\rb$. Thus, for $\lb\bfv\ts\bfb\rb$ not to be resistively limited, there must be time variation of $\lb \bfa\cdot\bfb\rb$, or non-vanishing boundary terms. When such terms vanish, helical turbulence without mean field gradients gives $\alpha \le (b/{\bar B})^2 \alpha_0/R_m$, where $\alpha_0$ is the kinematic value of the psuedoscalar coefficient $\alpha$, and $R_m$ is the magnetic Reynolds number. There is thus an ambiguity in interpreting all existing numerical experiments suggesting $\alpha$ quenching (e.g. Cattaneo & Hughes 1996); the quenching might not be dynamical, but may be due to boundary conditions.
Magnetic Helicity Escape, Dynamo Action, & Coronal Activity
===========================================================
The above result highlights the role of total magnetic helicity $H^M = \int_V \A\cdot \B\, d^3x$ (Elsässer 1956), where $V$ is a volume of integration, and $\A$ is the vector potential. That MFD growth involves a magnetic helicity inverse cascade was demonstrated by Pouquet (1976). The $\alpha$ effect conserves $H^M$ by pumping a positive (negative) amount to scales $>L$ (the outer turbulent scale) and a negative (positive) amount to scales $\ll L$. Brandenburg’s (2000) simulations confirm this inverse cascade and the role of $H^M$ conservation.
A large-scale field can be generated only as fast as $H_M$ can be removed or dissipated. Presently, simulations have invoked boundary conditions for which the growth of large scale field is resisitvely limited. Large $R_m$ systems must rely on open boundary conditions. To see this, note that $H^M$ satisfies $
{\partial_t} (\A\cdot\B)+c \nabla\cdot (\E\times \A+A_0\B) =
-2c\E\cdot \B
$ where $\E = -{\BV\over c}\times \B $. Consider two cases. Case (1): The mean scale = universal scale, or the integration is over periodic boundaries. Then boundary terms vanish, so $\partial_t \lb{\bf A}\cdot {\bf B} \rb=
-2c \lb{\bf E}\cdot {\bf B}\rb=
-2c {\bbE}\cdot {\bbB}-2c \lb{\bfe}\cdot {\bfb}\rb=\eta \lb {\bf J}\cdot{\bf B}\rb$ and $\partial_t ({\bbA}\cdot \bbB) =-2c {\overline {\bf E}}\cdot {\bbB}; \ \ \partial_t \lb{\bfa}\cdot {\bfb} \rb=-2c \lb {\bfe}\cdot {\bfb}\rb$. This is the case of section 1. Dynamo action is resistively limited. Case (2): The system (e.g. Galaxy or Sun) mean volume V $<<$ universal volume. Here we must use the gauge invariant relative helicity $H^M_R$ inside and outside of the spherical or disk rotator (Berger & Field 1984). The integral over the the universal volume then satisfies $\partial_t\int_{U}{\bf A} \cdot {\bf B}d^3x=\partial_t{H}^M_{R,in}+\partial_t{H}^M_{R,out}=-2c\int_U{\bf E}\cdot{\bf B}d^3x\simeq 0$. The formulae for the $H^M_R$ of the mean and fluctuating quantities inside the rotator are $\partial_t {{H}}_{R,in}(\bbB)=
-2c\int_{in}{\bbE\cdot\bbB} d^3x
+2c\int_{S_{in}}({\bbA}_p\times
{\bbE})\cdot d{\bf S}
$ and $
\partial_t{{H}}_{R,in}({\bfb})=
-2c\int_{in}{\lb\bfe\cdot\bfb\rb} d^3x
+2c\int_{S_{in}}\lb{\bf a}_p\times
{\bfe}\rb\cdot d{\bf S}.
\label{r4aae}
$ In a steady state, $\partial_tH_{R,in}=0=\partial_tH_{R,out}$. This and ${\bf E}\cdot{\bf B}\simeq 0$ imply that the above surface terms above must be equal and opposite. Moreover, the surface term balances the $\bbE\cdot \bbB$ term in the $\partial_tH^M_{R,in}$ equation. The boundary term can thus allow for a significant turbulent EMF because the latter is contained in $\bbE\cdot \bbB$. Dynamo action unrestricted by resisitivity is possible only in case (2). This is consistent with Pouquet et al. (1976) and Brandenburg (2000).
If $H_M$ flows through the boundary, then so does magnetic magnetic energy. Blackman & Field (2000b) showed that a typical minimum power leaving the system when a MFD is operating is given by $\dot E^M
\ge
{k_{\rm min}\over
8\pi} |\dot H^M|\nn
= {k_{\rm min}\over 6} \left| \lra{\alpha\ob^2}\right|V
$, where $V$ is the system volume. Dynamos operating in the Sun, accretion disks, and the Galaxy would then lead to a net escape of magnetic energy and small and large scale magnetic helicity. Coronal activity from the emergence and dissipation of helical magnetic flux is thus a prediction of the MFD in all of these cases, and is observed directly in the Sun (c.f. Pevtsov et al. 1999). For the Galaxy, ${\dot E}^M \gsim ({\pi R^2})\alpha \ob^2
\sim10^{40} ({R / 12{\rm kpc}})^2 ({\alpha / 10^5{\rm cm/s}})
({\ob/ 5\ts 10^{-6}{\rm G}})^2 {\rm erg/s}$, in each hemisphere. Blackman & Field (2000b) discuss how this relation may be consistent with coronal energy input rates required by Savage (1995) and Reynolds et al. (1999).
Open Questions
==============
An MFD unlimited by resitivivity requires the helicity to flow through the boundary AND that there be some mechanism that enables this flow. Thus there are two separate issues. Even if the boundary conditions allow it, does it actually happen? One may have to include the dynamics of buoyancy or winds to fully demonstrate the non-resistive MFD. Note that turbulent diffusion of the mean magnetic field (not necessarily the actual field) across the boundary is required to maintain a quadrupole field in the Galaxy with a net flux inside the disk. Similarly, for the Sun, the solar cycle requires net diffusion through the boundary. The flow of helicity would appeal to the same dynamics needed by these constraints.
The analytic and numerical studies that we have seen which show catastrophic suppression of the dynamo coefficients, or resistively limited dynamo action, either (1) invoke periodic boundary conditions, and/or (2) are 2-D, or (3) do not distinguish between zeroth order isotropic components of the turbulence and the higher order anistropic perturbations for a weak mean field (Blackman & Field 1999). This means that there always seems to be an alternative explanation. and the observed suppression is then ambiguous as we have described. This does not mean that some of the physical concepts found in the strong suppression results are invalid, but just that they may be valid only for the restricted cases considered. For example, the observation that the Lagrangian chaos properties of the flow are changed in the presence of a weak mean field for turbulence in a periodic box (e.g. Cattaneo et al. 1996) needs to be understood in the relation to the imposed boundary conditions, and the shape of the magnetic energy spectrum (e.g. dominated at small or large scales?). Along these lines, note that the helicity constraint is global, but also becomes a constraint for any sub-volume of a periodic box once the system is fully mixed.
References
==========
Blackman, E. G. & Field, G. B. 1999, ApJ [**521**]{} 597.
Blackman, E. G. & Field, G. B. 2000a, ApJ [**534**]{} 984.
Blackman, E. G. & Field, G. B. 2000b, MNRAS, in press, astro-ph/9912459.
Berger M.C. & Field G.B., 1984, JFM, [**147**]{} 133.
Brandenburg A., 2000, submitted to ApJ, astro-ph/0006186.
Cattaneo, F., & Hughes, D.W. 1996, Phys. Rev. E. [**54**]{}, 4532.
Cattaneo, F., & Hughes, D.W., Kim E.-J, 1996, PRL [**76**]{}, 2057.
Elsässer, W. M. 1956, Rev. Mod. Phys. [**23**]{}, 135.
Field, G. 1986, in [*Magnetospheric Phenomena in Astrophysics*]{}, R. Epstein & W. Feldman, eds., AIP Conf. Proc. 144 (Los Alamos: Los Alamos Sci. Lab.), 324.
Field, G. B., Blackman, E. G., & Chou, H. 1999, ApJ [**513**]{}, 638.
Frisch, U., Pouquet, A., Léorat, J. & Mazure, A. 1975, JFM [**68**]{}, 769.
Krause F. & Rädler K.-H., 1980, [ *Mean-field magnetohydrodynamics and dynamo theory*]{}, (New York: Pergamon).
Kulsrud R., 1999, ARAA, [**37**]{} 37.
Parker, E. N., [*Cosmical Magnetic Fields*]{} (Oxford: Clarendon Press).
Pevtsov, A., Canfield, R. C., & Brown, M. R. eds. 1999, [*Magnetic Helicity in Space and Laboratory Plasmas*]{}, (American Geophysysical Union).
Pouquet, A., Frisch, U., & Leorat, J. 1976, JFM [**77**]{}, 321.
Reynolds R.J., Haffner L.M., Tufte S.L., 1999, ApJ [**525**]{} 21.
Savage B.D., 1995, in [*The Physics of the Interstellar Medium and Intergalactic Medium*]{}, A. Ferrara, C.F. McKee, C. Heiles, & P.R. Shapiro eds. ASP conf ser vol 60. (San Francisco: PASP) p233.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'For the training of face detection network based on R-CNN framework, anchors are assigned to be positive samples if intersection-over-unions (IoUs) with ground-truth are higher than the first threshold(such as 0.7); and to be negative samples if their IoUs are lower than the second threshold(such as 0.3). And the face detection model is trained by the above labels. However, anchors with IoU between first threshold and second threshold are not used. We propose a novel training strategy, Precise Box Score(PBS), to train object detection models. The proposed training strategy uses the anchors with IoUs between the first and second threshold, which can consistently improve the performance of face detection. Our proposed training strategy extracts more information from datasets, making better utilization of existing datasets. What’s more, we also introduce a simple but effective model compression method(SEMCM), which can boost the performance of face detectors further. Experimental results show that the performance of face detection network can consistently be improved based on our proposed scheme.'
author:
- |
Ce Qi$^1$ Xiaoping Chen$^1$ Pingyu Wang$^1$ Fei Su$^{1,2}$\
$^1$School of Information and Communication Engineering\
$^2$Beijing Key Laboratory of Network System and Network Culture\
Beijing University of Posts and Telecommunications, Beijing, China\
\
bibliography:
- 'egbib.bib'
title: 'Precise Box Score: Extract More Information from Datasets to Improve the Performance of Face Detection'
---
Introduction
============
Face detection, which is the basis of face alignment and face recognition, plays an important role in face related tasks. More accurate face detection and face bounding boxes will also benefit the performance of face alignment and face recognition.
Many research works [@zhu2012face; @li2013probabilistic; @li2014efficient; @mathias2014face; @chen2014joint; @yang2014aggregate; @li2015convolutional; @yang2015facial; @yang2015convolutional; @zhu2017cms; @jiang2017face; @wang2017face; @wang2017detecting; @Hu_2017_CVPR; @najibi2017ssh] have been done to improve the performance of face detectors. However, there is still big gap between humans and current face detectors, especially in the scenarios of small faces or occluded faces. The gap becomes bigger in case of resource constraint environment for the trading of the complexity and the required speed and memory. Some good performance face detectors are usually slow and high memory foot-prints(*e.g.* it takes more than 1 second in [@Hu_2017_CVPR] per image and HyperNet [@kong2016hypernet] is slow and has big model size).
One way to make the face detection models efficient is to use more powerful networks or design a specific network architecture. But this kind of strategy is not elegant and may cannot be used in other detection tasks. Recently, Hu *et al.* [@Hu_2017_CVPR] gets state-of-the-art results on the WIDER FACE detection benchmark [@yang2016wider] by using a similar approach to the Region Proposal Networks(RPN) [@ren2015faster] to directly detect faces. To boost the performance, it introduces an image pyramid as an integral part of the method, which is not that time efficient.
Another way is using more training data and data augmentation. AFLW [@koestinger2011annotated], which is a relatively smaller face detection dataset, is usually used as train set before. Face detection model’s performance will be boosted when using the WIDER FACE dataset [@yang2016wider], a relatively bigger face detection dataset. What’s more, data augmentation such as flipping and blurring will also help the final accuracy.
When training the R-CNN style face detector, anchors are assigned to be positive samples if intersection-over-unions (IoUs) with ground-truth are higher than the first threshold(such as 0.7); and to be negative samples if their IoUs are lower than the second threshold(such as 0.3). The object detection model is trained by the above labels, meaning the positive labels and negative labels are set by hand roughly. And the anchors with IoUs between first and second threshold are not used, which loses much information from detection dataset.
In this paper, we show that when training face detection model based on R-CNN framework, the original anchors assignment strategy is not appropriate and loses much information from original face detection dataset. As is shown in Fig. \[fig:orignal training strategy\], original training strategy has three weaknesses: (a) choosing thresholds roughly; (b) setting positive and negative labels with 1 or 0 roughly; (c) the information of anchors with IoUs between 0.7 and 0.3 is not used. So, we propose a novel training strategy, called Precise Box Score(PBS), to train face detection models. The proposed training strategy uses the anchors more effectively, meaning more information from face detection dataset will be used for training. What’s more, we also introduce a simple but effective model compression method(SEMCM), which can boost the performance of face detectors further. The experimental results show that when using the proposed novel training strategy and model compression method, the performance of face detection model can consistently be improved.
The rest of the paper is organized as follows. Section \[sec:related work\] provides an overview of the related works. Section \[sec:Novel Training strategy: Precise Box Score(PBS)\] introduces the proposed training strategy: Precise Box Score(PBS) and the new architecture designed for PBS. Section \[sec:A Simple but Effective Model Compression Method(SEMCM)\] describes a simple but effective model compression method(SEMCM), which can improve the performance of face detector and reduce the model size. Section \[sec:Experiments\] presents the experiments and Section \[sec:Conclusion\] gives the conclusion.
Related Work {#sec:related work}
============
Face Detection
--------------
Face detection are the basis of face related tasks. And there are many works have been done to improve the performance of face detection. Before the re-emergence of convolutional neural networks(CNN), many traditional methods [@viola2001rapid; @zhu2012face; @li2013probabilistic; @li2014efficient; @mathias2014face; @chen2014joint; @yang2014aggregate] have been proposed for face detection. The most successful traditional method is Viola-Jones [@viola2001rapid] detector. However, most of the traditional methods use hand-crafted features, which limit the performance of face detector. Following the success of CNN [@krizhevsky2012imagenet], the performance of face detection is improved significantly, for the discriminative features of CNN. Recently, many CNN-based works have been done for face detection, such as [@li2015convolutional; @yang2015facial; @yang2015convolutional; @zhu2017cms; @Hu_2017_CVPR].
R-CNN Style Face Detector
-------------------------
The idea of detecting and localizing objects in two stages is widely used in object detection, such as Faster R-CNN [@ren2015faster] and R-fcn [@dai2016r]. Face detector [@jiang2017face; @wang2017face; @wang2017detecting] with R-CNN style can obtain good accuracy. However, unlike the object detection with many classes, face detection detects only one class. Single stage face detectors [@zhu2017cms; @Hu_2017_CVPR; @najibi2017ssh] also work well, which detect faces directly from the early convolutional layers with bounding box classification and regression. Most of the single stage face detection methods are more similar to the object proposal algorithm which is used as the first stage in detection pipeline. These kind of algorithms generally regress a set of anchors toward faces and assign scores to different anchors according to the intersection-over-unions (IoUs) between anchors and ground truth bounding boxes.
Main Focus of Face Detection Research
-------------------------------------
For single stage face detectors, there are many works have been done to boost the performance of face detection. Most of the methods focus on scale invariance and context modeling. Scale invariant can make detection of different scale of faces easier and context information can do help to hard classified faces.
For general object detection, ION [@bell2016inside] uses skip pooling and RNN(recurrent neural networks) for context modeling and scale invariance. FPN [@dollar2014fast] employs skip connections and multiple shared RPN from different convolutional layers. The same methods also be used for face detection. CMS-RCNN [@zhu2017cms] employs skip connection, too. Hu [@Hu_2017_CVPR] uses image pyramids and context modeling to improve the performance.
There are also some other methods focusing on object loss functions of detection, such as Unitbox [@yu2016unitbox] and Grid loss [@opitz2016grid]. Some other researchers do efforts on non-maximum suppression(NMS), a post processing step. Soft-NMS [@bodla2017improving] uses a very simple but effective way to improve the NMS. Authors in [@hosang2016convnet] use a convolutional network to guide the NMS after detection.
Our training strategy(PBS) focuses on how to extract more information from current face detection dataset to improve the performance of face detection.
Novel Training Strategy: Precise Box Score (PBS) {#sec:Novel Training strategy: Precise Box Score(PBS)}
================================================
Most of the works about detection are related to network architecture, loss function or post processing step. The main focus of the proposed training strategy is the input data step of model, just say, how to extract more information from current face detection datasets.
Our novel training strategy(PBS) is designed for detection network using anchors. Details are shown in Fig. \[fig:PBS\_train\_test\].
{width="95.00000%"}
General Architecture
--------------------
Fig. \[fig:General architecture of our face detection network(FDN)\] shows the general architecture we use, called Face Detection Network(FDN). It is a fully convolutional network which performs bounding box classification and regression simultaneously. For localization, just like RPN in [@ren2015faster], our face detection network(FDN), regresses a set of predefined bounding boxes(anchors), to approximate the ground-truth bounding boxes. And the operations of bounding box classification and regression are added on the top of feature map with stride 16. The scales FDN used are 4, 8, 16 and 32. And, we only consider anchors with aspect ratio of 1:1 to reduce the number of total anchor boxes and fit the truth that most face boxes have aspect ratios of 1:1. Exactly, if the top of feature map with stride 16 has size $W_i \times H_i$, there would be $W_i \times H_i \times K $ anchors, where $K$ equals to $4$ in our setting.
For the reason of FDN with all convolutional network, the input images can be set to any size. And the total model size is small. Note that there are two different network architectures in Fig. \[fig:General architecture of our face detection network(FDN)\]. Details will be described as follows.
Precise Box Score(PBS) {#subsec:Precise Box Score(PBS)}
----------------------
In this section, we will discuss the training strategy of R-CNN style detector, such as Faster R-CNN [@ren2015faster], R-fcn [@dai2016r], Hu *et al.* [@Hu_2017_CVPR] and SSH [@najibi2017ssh]. Furthermore, the details of our novel training strategy(PBS) will be given.
**General training strategy of R-CNN style detector:** For the training of R-CNN style detector, anchor boxes are introduced to serve as reference of multiple scales and aspect ratios. Classification and Regression are done simultaneously for anchors to do region proposal. When training, anchors are assigned a binary class label(of being an object or not). The conditions of assigning positive labels to anchors are: (i) anchors with the highest Intersection over Unions(IoUs) overlap with a ground-truth box, or (ii) anchors that have IoUs higher than $0.7$ with any ground-truth box. And anchors are assigned to be negative labels if their IoUs are lower than $0.3$ with all ground-truth boxes. Anchors that are neither positive nor negative do not contribute to the training. The detail of general training strategy of R-CNN style detector is shown in Fig. \[fig:orignal training strategy\] and the formula is illustrated in Eq. (\[equ:orignal training strategy\]). Note that the figure and formula above approximately illustrate the general training strategy of R-CNN style detector, for the existing of principle (i) described above.
$$\label{equ:orignal training strategy}
label_{anchor}(IoU) =
\begin{cases}
$0$& \text{$IoU<0.3$}\\
$1$& \text{$IoU>0.7$}
\end{cases}$$
**Weakness of general training strategy:** As shown in Fig. \[fig:orignal training strategy\], according to the general training strategy of R-CNN style described above, we can find that: (1) the positive anchors threshold($0.7$) and negative anchors threshold($0.3$) are set to certain numbers roughly. The two thresholds in Faster R-CNN are $0.7$ and $0.3$ respectively. And the two thresholds in R-fcn are the same numbers, $0.5$. (2) the labels are binary class, meaning the labels are $0$ or $1$. The rough binary label loses much information from detection dataset. For example, the anchor with IoU $0.75$ is different with the anchor with IoU $1.0$. The latter is more like a positive sample than the former. (3) the anchors with IoUs between first and second thresholds are not used.
**Precise Box Score(PBS):** To overcome the three weaknesses of general training strategy described above, we propose a novel training strategy, Precise Box Score(PBS). PBS will choose the best thresholds and use precise float point numbers as labels, when the precise float point numbers are the outputs of a designed function using IoUs as inputs, as illustrated in Fig. \[fig:our new training strategy\]. It will firstly choose the best thresholds through experiments and then choose a best function to translate IoUs to labels. Detailed steps are as follows:
\(1) Using Eq. (\[equ:pbs\_just\_adjust\_pos\_bound\]) to choose the best thresholds through experiments(Note that $Bound_{pos} = 0.3, 0.4, 0.5, 0.7$ when $Bound_{neg}=0.3$, and $Bound_{pos}=0.2$ when $Bound_{neg}=0.1$). The best thresholds are represented by $Bound_{best\_pos}$ and $Bound_{best\_neg}$.
\(2) Choosing the best function to translate IoUs to labels, based on best thresholds $Bound_{best\_pos}$ and $Bound_{best\_neg}$ obtained in step (1). Three classes of functions are used, as illustrated in Eq. (\[equ:pbs\_add\]), Eq. (\[equ:pbs\_split\_one\]) and Eq. (\[equ:pbs\_split\_two\]). Eq. (\[equ:pbs\_add\]) adds a shift variable A($A \ge 0$) to the IoU for the positive label, and when the positive label is bigger than $1$, it will be set to $1$. This function roughly translates the IoUs to labels. Eq. (\[equ:pbs\_split\_one\]) limits some IoUs($Bound_{best\_pos} \textless IoU \textless Bound_1$) to $Score_1$. And Eq. (\[equ:pbs\_split\_two\]) uses more variables and do more precise limitation to the function of IoU. Details will be shown in experiments in Section \[subsec:Precise Sigmoid+PBS on FDDB\].
$$\label{equ:pbs_just_adjust_pos_bound}
label_{anchor}(IoU) =
\begin{cases}
$0$& \text{$IoU \textless Bound_{neg}$}\\
$1$& \text{$IoU \textgreater Bound_{pos}$}
\end{cases}$$
$$\label{equ:pbs_add}
\resizebox{0.9\hsize}{!}
{$
label_{anchor}(IoU) =
\begin{cases}
$0$& \text{$IoU \textless Bound_{best\_neg}$}\\
$IoU+A$& \text{IoU \textgreater $Bound_{best\_pos}, (IoU+A) \textless 1$}\\
$1$& \text{$IoU \textgreater Bound_{best\_pos}, (IoU+A) \ge 1$}
\end{cases}
$}$$
$$\label{equ:pbs_split_one}
\resizebox{0.9\hsize}{!}
{$
label_{anchor}(IoU) =
\begin{cases}
$0$& \text{$IoU \textless Bound_{best\_neg}$}\\
$$Score_1$$& \text{$Bound_{best\_pos} \textless IoU \textless Bound_1$}\\
$1$& \text{$IoU \ge Bound_1$}
\end{cases}
$}$$
$$\label{equ:pbs_split_two}
\resizebox{0.9\hsize}{!}
{$
label_{anchor}(IoU) =
\begin{cases}
$0$& \text{$IoU \textless Bound_{best\_neg}$}\\
$$Score_1$$& \text{$Bound_{best\_pos} \textless IoU \textless Bound_1$}\\
$$Score_2$$& \text{$Bound_1 \le IoU \textless Bound_2$}\\
$1$& \text{$IoU \ge Bound_2$}
\end{cases}
$}$$
New Architecture Designed for Precise Box Score(PBS)
----------------------------------------------------
As introduced above, the proposed novel training strategy, Precise Box Score(PBS), uses precise float point numbers as labels, not simply uses binary labels. So, softmax with cross entropy loss, as shown in Eq. (\[equ:softmax with cross entropy loss\]), which is designed for binary labels, is not appropriate for the precise float point number labels.
$$\label{equ:softmax with cross entropy loss}
L_{s}=-\frac{1}{N}\sum_{i} logP(y_i|x_i)=-\frac{1}{N}\sum_{i} log\frac{e^{f_{y_i}}}{\sum_{j} e^{f_j}}$$
where $x_i$ and $y_i \in [1...C]$ denote the $ith$ input data and its corresponding label, respectively. $f_j$ denotes the $jth$ element of the softmax input vector $f$, and $j \in [1...C]$. $N$ is the number of training images. $C$ is the number of class. We design a new architecture designed for Precise Box Score(PBS), as illustrated in Fig. \[fig:FDN with precise-sigmoid\]. Because of two reasons, the new architecture is needed:
\(a) Our proposed novel training strategy, Precise Box Score(PBS), uses precise float point numbers as labels, not simply uses binary labels.
\(b) The precise float point number labels the PBS used are in $[0,1]$.
So, the designed new architecture for PBS replaces the softmax(with cross entropy loss)with sigmoid(with euclidean loss), as shown in Eq. (\[equ:sigmoid with euclidean loss\]), which is called Precise Sigmoid. The new architecture, which uses sigmoid(with euclidean loss), will output the numbers in $[0,1]$, while with the loss for precise float point number.
The formula of sigmoid with euclidean loss(Precise Sigmoid) is shown in Eq. (\[equ:sigmoid with euclidean loss\]).
$$\label{equ:sigmoid with euclidean loss}
L_{PreciseSigmoid}=\frac{1}{2 \times N}\sum_{i} {(\frac{1}{1+e^{-x_i}}-y_i)}^2$$
where $L_{PreciseSigmoid}$ denotes the Precise Sigmoid Loss, new loss designed for PBS in the new architecture. $x_i$ and $y_i \in [1...C]$ denote the $ith$ input data and its corresponding label. $N$ is the number of training images. $C$ is the number of class.
The loss of our face detector using new architecture is:
$$\label{equ:new final loss}
L=\lambda L_{PreciseSigmoid} + L_{Regression}$$
where $L_{PreciseSigmoid}$. denotes the Precise Sigmoid Loss. $L_{Regression}$ denotes the SmoothL1 Loss [@ren2015faster] used for bounding box regression.
**Note:** we know that in mathematics, the sigmoid with euclidean loss may generate gradient vanishing. To solve this problem, we train the new architecture based on the parameters of a model pretrained by softmax with cross entropy loss. Experiments will be given in Section \[sec:Experiments\] to demonstrate the effectiveness of this method, which can avoid the gradient vanishing of sigmoid with euclidean loss. The experimental results for demonstration are shown in Table \[table:accuracy comparison of precise sigmoid and softmax\].
The benefits of new architecture are as follows:
\(a) Training phase: using precise float point numbers in $[0,1]$ as labels satisfies the training request of PBS. (b) Testing phase: the outputs of sigmoid are in $[0,1]$, which can be used as scores of face boxes directly.
\(c) It is effective to reduce the params of models through sigmoid, which makes the training and testing faster.
Superiority of Precise Sigmoid+Precise Box Score(PBS)
-----------------------------------------------------
\(a) Using the labels of PBS with Precise Sigmoid is the full implementation of proposed new training strategy.
\(b) Using the precise float point number labels can extract more information from detection dataset, which can help the training.
\(c) Under PBS, models can output precise and appropriate scores for bounding boxes, which can benefit the post processing of NMS(bounding boxes with lower IoUs with ground-truth get lower scores).
A Simple but Effective Model Compression Method(SEMCM) {#sec:A Simple but Effective Model Compression Method(SEMCM)}
======================================================
{width="95.00000%"}
After using the Precise Box Score(PBS) to improve the performance of face detection model, we propose a simple but effective model compression method(SEMCM) for one stage face detector. There are also some other works for model compression [@han2015deep; @hinton2015distilling; @luo2016face; @li2017mimicking]. [@han2015deep] is a general but relatively complex model compression method. [@hinton2015distilling; @luo2016face; @li2017mimicking] are relatively simple model compression methods, while [@hinton2015distilling; @luo2016face] are designed for image classification or face recognition. [@li2017mimicking] is designed for object detection, but this method is used for two stage detector and uses the information of Region Of Interest(ROI), which does not exist in one stage detector.
SEMCM is designed for one stage detector, as illustrated in Fig. \[fig:overall architecture of SEMCM\]. One stage detector uses classification and regression results for anchors simultaneously to get the final results. SEMCM uses the output from pretrianed big model directly and the output will be used as supervision signals for the training of small model. Note that the output feature maps of big model and small model should be in same dimension in width, height and channel. So, the most convenient way to get a trainable small model is just downsampling the channels of all layers in big model, except the output layers used for classification and regression. The details of training steps of SEMCM are described as follows:
\(1) A big model is trained through our proposed training strategy(PBS). (2) A small model is obtained by downsampling the channels of all layers in big model, except the output layers used for classification and regression. (3) The small model is trained simply by general training strategy with some iterations. SEMCM also works for half-trained small model. (4) Frozen the parameters of pretrained big model in (1), and train the small model supervised by the output feature maps of big model. (5) **Note:** to make the training of SEMCM stable. Two new layers are designed, to guide how to use the output feature maps of big model to supervise the training of small model, as illustrated in Fig. \[fig:overall architecture of SEMCM\]. And the original training signals are also added to guarantee the good performance of SEMCM.
Experiments {#sec:Experiments}
===========
Experimental Setup
------------------
The new architecture models(shown in Fig. \[fig:FDN with precise-sigmoid\]) trained with PBS will start the training from a pretrained network trained through softmax with cross entropy loss. Using a pretrained model will solve the problem of gradient vanishing of sigmoid with euclidean loss for PBS. All anchors have aspect ratio of 1:1. For FDDB [@fddbTech], anchors with scales {4, 8, 16, 32} are used on the feature map with total stride $16$. The training and testing are both single scale, meaning we rescale the shorter side of the image up to 600 pixels while keeping the longer side below 1000 pixels without changing the aspect ratio. For WIDER FACE dataset [@yang2016wider], all settings follow the SSH [@najibi2017ssh]. During inference, model outputs 300 top scoring boxes and NMS with threshold of $0.3$ is performed on the boxes to get the final detection results.
The goal of our experiments is to demonstrate the effectiveness of PBS and SEMCM, so we use relatively simple networks to verify the effectiveness of our methods. We use one stage face detector with main bone of ZF-net [@zeiler2014visualizing], VGG\_CNN\_M\_1024 [@chatfield2014return], VGG16 [@simonyan2014very] and ZF-24-net. Note that the ZF-24-net is a network designed for SEMCM. ZF-24-net has the same architecture as ZF-net, except all layers’ channel reduced by $\frac{1}{4}$(not inculde layers for classification and regression), meaning ZF-24-net is a smaller network compared with ZF-net. The four designed one stage face detectors’ model size is 17.3M, 30.9M, 68.3M and 1.1M, respectively.
Datasets
--------
FDDB [@fddbTech] and WIDER FACE [@yang2016wider] are used in our experiments.
**FDDB [@fddbTech]:** FDDB contains 2845 images with 5171 annotated faces. We use this dataset only for testing. **WIDER FACE [@yang2016wider]:** WIDER FACE contains 32, 203 images with 393, 703 annotated faces, 158, 989 of which are in the train set, 39, 496 in the validation set and rest are in the test set. The validation and test set are divided into “easy”, “medium”, “hard” subsets cumulatively(*i.e.* the “hard” set contains all images). This is one of the most challenging public face detection datasets, with wide variety of face scales and occlusion. By default, we train our models on the train set of WIDER FACE [@yang2016wider] and evaluate on the validation set of WIDER FACE or FDDB [@fddbTech].
Ablation study of loss weight for Precise Sigmoid
-------------------------------------------------
We firstly do experiments on FDDB [@fddbTech] for loss weight $\lambda$ in Eq. (\[equ:new final loss\]) of the proposed new architecture(shown in Fig. \[fig:FDN with precise-sigmoid\]) for PBS. When doing the experiments for loss weight of $L_{PreciseSigmoid}$ in Eq. (\[equ:new final loss\]), we use the original R-CNN style training strategy, as described in Section \[subsec:Precise Box Score(PBS)\], and the two thresholds are set to $0.7$ and $0.3$. All anchors have aspect ratio of 1:1 with scales {4, 8, 16, 32}, which are used on the feature map with total stride $16$.
We use the train set of WIDER FACE for training and FDDB for testing. Different networks are used as main bone of one stage detector, to find an optimal loss weight for different networks. The results are shown as follows. Table \[table:accuracy on FDDB with different loss weight of zf-net\],\[table:accuracy on FDDB with different loss weight of vgg\_m\_cnn\_1024\] and \[table:accuracy on FDDB with different loss weight of zf\_24\] give the results of one stage detector with main bone of ZF-net [@zeiler2014visualizing], VGG\_CNN\_M\_1024 [@chatfield2014return] and ZF-24-net. the accuracy is measured when the false positive is $1000$, $500$, $100$ respectively on FDDB. The “$-$” in three tables means the model cannot converge well.
From Table \[table:accuracy on FDDB with different loss weight of zf-net\],\[table:accuracy on FDDB with different loss weight of vgg\_m\_cnn\_1024\] and \[table:accuracy on FDDB with different loss weight of zf\_24\], the optimal loss weight $\lambda$ equals to $300$. So, for the new architecture with Precise Sigmoid, we use $\lambda=300$ for $L_{PreciseSigmoid}$ by default.
Precise Sigmoid+PBS on FDDB {#subsec:Precise Sigmoid+PBS on FDDB}
---------------------------
**The comparable performance of Precise Sigmoid**: we also compare the performance of our Precise Sigmoid with softmax **by using original R-CNN style training strategy**, as shown in Table \[table:accuracy comparison of precise sigmoid and softmax\]. The results show the comparable performance of Precise Sigmoid and softmax. Furthermore, better results on VGG16 based network of Precise Sigmoid prove the gradient vanishing problem can be solved. **The effectiveness of Precise Sigmoid+PBS:** the Precise Sigmoid is designed for PBS, and experimental results will show the effectiveness of Precise Sigmoid+PBS. The experiments are done with ZF-net as main bone of face detector.
Fig. \[fig:Detailed experimental results of different Precise Box Score(PBS)\] shows the detailed experimental results of Precise Sigmoid+PBS, and the PBS strategy of “split\_0.4\_0.8\_0.5\_0.9” gets the best accuracy. Next, results in Table \[table:accuracy comparison of precise sigmoid+PBS and softmax\] prove the effectiveness of PBS. The network using Precise Sigmoid+PBS can consistently get accuracy gain compared with conventional softmax+original training strategy.
![Detailed experimental results of different Precise Box Score(PBS) on FDDB. “pos0.7+all=1” denotes the using of Eq. (\[equ:pbs\_just\_adjust\_pos\_bound\]) with $Bound_{pos}=0.7$. “pos0.4+add0.6” denotes the using of Eq. (\[equ:pbs\_add\]) with $Bound_{pos}=0.4$ and $A=0.6$. “pos0.4+split\_0.7\_0.8” demotes the using of Eq. (\[equ:pbs\_split\_one\]) with $Bound_{best\_pos}=0.4$, $Bound_1=0.7$ and $Score_1=0.8$. “pos0.4+split\_0.4\_0.8\_0.5\_0.9” denotes the using of Eq. (\[equ:pbs\_split\_two\]) with $Bound_{best\_pos}=0.4$, $Bound_1=0.4$, $Score_1=0.8$, $Bound_2=0.5$ and $Score_2=0.9$.[]{data-label="fig:Detailed experimental results of different Precise Box Score(PBS)"}](discROC_detail_pbs_zf.png){width="46.00000%"}
Precise Sigmoid+PBS on WIDER FACE {#subsec:Precise Sigmoid+PBS on WIDER face}
---------------------------------
We also evaluate the performance of Precise Sigmoid+PBS on WIDER FACE. SSH [@najibi2017ssh] is used as baseline, using the train set of WIDER FACE for training. The training and testing are both single scale, meaning we rescale the shorter side of the image up to 1200 pixels while keeping the longer side below 1600 pixels without changing the aspect ratio. The settings of scale and aspect ratio follow the SSH’s.
Table \[table:Comparison of original SSH with SSH trianed by our Precise Sigmoid+PBS\] compares the original SSH with SSH trained by our Precise Sigmoid+PBS. SSH uses two rough IoU thresholds, 0.5 and 0.3. In experiments, we adjust the threshold for positive samples on the architecture of Precise Sigmoid. We also use the empirical best parameters of PBS(“pos0.4+split\_0.4\_0.8\_0.5\_0.9”, the results of Fig \[fig:Detailed experimental results of different Precise Box Score(PBS)\]), to demonstrate the effectiveness of Precise Sigmoid+PBS. Table \[table:Comparison of original SSH with SSH trianed by our Precise Sigmoid+PBS\] shows the effectiveness of Precise Sigmoid+PBS:
\(a) SSH trained by Precise Sigmoid+PBS outperforms the original SSH [@najibi2017ssh] by 0.3%, 0.6% and 0.8% in “easy”, “medium”, “hard” subsets of WIDER FACE respectively.
\(b) current result of SSH trained by Precise Sigmoid+PBS is just simply using the setting of “pos0.4+split\_0.4\_0.8\_0.5\_0.9”, which is the empirical parameters of PBS, as shown in Fig. \[fig:Detailed experimental results of different Precise Box Score(PBS)\]. Specific adjustment of parameters of PBS may make the model get even better result.
SEMCM on FDDB
-------------
There, we use the one stage face detector with main bone of ZF-net as teacher model. And the one stage face detector with main bone of ZF-24-net as student model. As illustrated in Fig. \[fig:overall architecture of SEMCM\], we conduct the SEMCM as described in Section \[sec:A Simple but Effective Model Compression Method(SEMCM)\]. The teacher model(ZF-net) has accuracy of 94.6, 93.3 and 83.4 when the false positive on FDDB is 1000, 500 and 100, and the model is trained by Precise Sigmoid+PBS. The student model(ZF-24-net) is all layers’ channel reduced by $\frac{1}{4}$ from teacher model(ZF-net), except the layers for classification and regression. The student model(ZF-24-net) is half-trained with accuracy of 88.4, 86.5 and 77.4.
Table \[table:Results of SEMCM\] and Fig \[fig:Results of SEMCM\] shows that: after using SEMCM, the performance of small student model can be raised to 92.2, 91.0, 84.5, even better than the accuracy of the small student model trained with Precise Sigmoid+PBS. The detailed accuracy gain is(the accuracy shown next is when false positive on FDDB is 1000, 500 and 100 respectively.):
\(a) The small model has model size of 1M. And it is trained from the performance of 88.4, 86.5 and 77.4, which is a half-trained model.
\(b) The small model is raised to the accuracy of 92.2, 91.0, 84.5 from 88.4, 86.5 and 77.4. The gain is 3.8, 4.5, 7.1, relative to the half-trained model for initialization. The gain is 1.6, 2.1, 2.3, relative to the model trained with softmax. The gain is 0.5, 0.8, 1.4, relative to the model trained with Precise Sigmoid+PBS.
\(c) The small model is raised to the accuracy of 92.2, 91.0, 84.5, even better than the small model trained with Precise Sigmoid+PBS, proving the complementary of Precise Sigmoid+PBS and SEMCM.
![Results of SEMCM.[]{data-label="fig:Results of SEMCM"}](discROC_semcm_zf_24.png){width="40.00000%"}
Qualitative Results
-------------------
Some qualitative results of three face detectors are shown in Fig \[fig:Qualitative results\]. The three models have model size of 1.1M, 17.3M and 98.1M, respectively. The first one stage face detector of ZF-24-net as main bone is trained by Precise Sigmoid+PBS+SEMEM. The second one stage face detector of ZF-net as main bone is trained by Precise Sigmoid+PBS. And the third model, SSH, is trained by Precise Sigmoid+PBS.
Conclusion {#sec:Conclusion}
==========
We propose a novel training strategy, Precise Box Score(PBS), which can extract more information from detection dataset and benefit the post-processing of NMS for the precise bounding box scores. And a new architecture, Precise Sigmoid, is introduced for the implementation of PBS. We do experiments using one stage face detector on FDDB to explore how to design the function of PBS. Further more, a simply but effective model compression method(SEMCM) is proposed for one stage face detector, which can boost the performance of face detection further. Experiments demonstrate: (a) Precise Sigmoid+PBS can consistently improve the performance of face detection, and (b) the complementary of Precise Sigmoid+PBS and SEMCM.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Extremely Compact Stars (ECS) contain trapped null geodesics. When such objects enter the evolution period admitting geodetical motion of neutrinos, certain part of neutrinos produced in their interior will be trapped influencing their neutrino luminosity and thermal evolution. We study neutrino trapping in the braneworld ECS, assuming uniform distribution of neutrino emissivity and massless neutrinos. We give the efficiency of the neutrino trapping effects in the framework of the simple model of the internal spacetime with uniform distribution of energy density, and external spacetime described by the Reissner-Nordström geometry characterized by the braneworld “tidal” parameter $b$. For $b < 0$ the external spacetime is of the black-hole type, while for $b > 0$ the external spacetime can be of both black-hole and naked-singularity type. Then the ECS surface radius $R$ can be located also above the unstable (outer) photon circular orbit. Such basically new types of the spacetimes strongly alter the trapping phenomena as compared to the standard case of $b = 0$. It is shown that the neutrino trapping effects are slightly lowered by the presence of physically more plausible case of $b < 0$, as compared to the standard internal Schwarzschild spacetime, while they can be magnified by positive tidal charges if $b < 1$ and lowered for $b > 1$. However, potential astrophysical relevance of the trapping phenomena is strongly enhanced for negative tidal charges enabling a significant enlargement of the ECS surface radius to values coherent with recent observations.'
author:
- Zdeněk Stuchlík
- Jan Hladík
- |
\
Martin Urbanec
date: 'Received: date / Accepted: date'
title: Neutrino trapping in braneworld extremely compact stars
---
Introduction
============
One of the promising approaches to the higher-dimensional gravity theories is represented by the braneworld models, where the observable universe is a 3-brane (domain wall) to which the matter fields are confined, while the gravity field enters the extra spatial dimensions of size that could strongly exceed the Planck length scale [@Ark-Dim-Dva:1998:]. Such models provide an elegant solution to the hierarchy problem of electroweak and quantum gravity scales, as these become of the same order ($\sim \mathrm{TeV}$) due to large extra dimensions and could be well tested on the planned supercollider experiments [@Dim-Lan:2001:]. Gravity can be localized near the brane at low energies even with an infinite size extra dimension with warped spacetime satisfying the 5D Einstein equations containing negative cosmological constant [@Ran-Sun:1999:] and an arbitrary energy-momentum tensor allowed on the brane [@Shi-Mae-Sas:1999:]. The Randall-Sundrum model gives 4D Einstein gravity in low energies and the Newtonian limit appears on the 3-brane with high accuracy. Significant deviations from the Einstein gravity occur at very high energies, in the early universe and in vicinity of compact objects as black holes and neutron stars [@Maar:2004:].
Recently, no exact solution of the full 5D Einstein equations is know, but there is a variety of astrophysically plausible special solutions of the 4D effective Einstein equations constrained to the brane. Such solutions describe black holes with spherical symmetry [@Dad-Maar-Pap-Rez:2000:PHYSR4:] or with axial symmetry [@Ali-Gum:2005:] and compact objects that could represent neutron (quark) stars [@Ger-Maar:2001:]. The black hole spacetimes are determined by geometry of the spherical Reissner-Nordström (R-N) and axial Kerr-Newman (K-N) type where the electric charge squared is substituted by the braneworld “tidal charge” parameter representing the tidal (Weyl tensor) effects of the bulk space onto the 4D black hole structure. Astrophysically relevant properties of the braneworld black holes were studied in a series of papers devoted to both motion of matter in their vicinity [@Stu-Kot:2009:; @Ali-Tal:2009:; @Abdu-Ahme:2010:PHYSR4; @Mam-Hak-Toj:2010:MPLA:; @Mor-Ahme-Abdu-Mam:2010:ASS:] and optical phenomena [@Sche-Stu:2009:a; @Sche-Stu:2009:b; @Bin-Nun:2010:PHYSR4:a; @Bin-Nun:2010:PHYSR4:b].
In the simple model of spherically symmetric stars with uniform energy density profile a variety of special solutions having asymptotically Schwarzschildian character and satisfying the braneworld boundary conditions were found [@Ger-Maar:2001:]. The most popular is the one with external spacetime described by the Reissner-Nordström geometry with the braneworld tidal charge parameter reflecting the tidal effects of the bulk and related to the energy density and brane tension. Its properties were extensively studied both in the weak field limit [@Boh-Har-Lob:2008:CLAQG:; @Boh-Ris-Har-Lob:2010:CLASQG] and strong field limit when some restrictions on the brane tension were implied from the data of kHz QPOs observed in low mass binary systems with a neutron star [@Kot-Stu-Tor:2008:CLASQG:]. Here we focus our attention to the interior of the compact objects, namely to the phenomena related to trapping of neutrinos in the so called Extremely Compact Stars (ECS) admitting (by definition) existence of trapped null geodesics [@Stu-Tor-Hle-Urb:2009:]. Usually, the braneworld tension is assumed positive while the related tidal charge has to be negative [@Ger-Maar:2001:], but the negative tension and related positive tidal charge are not excluded [@Dad-Maar-Pap-Rez:2000:PHYSR4:], so we consider here both positive and negative tidal charge of the compact object. Note that exterior of neutron stars with positive tidal charges can be described by both black-hole and naked-singularity types of the external R-N spacetime (for details see [@Kot-Stu-Tor:2008:CLASQG:]).
In the standard ($b = 0$) internal Schwarzschild spacetimes of uniform energy density [@Schw:1916:SITBA:; @Stu:2000:ACTPS2:] with radius $R < 3GM/c^2$, bound null geodesics exist being concentrated around the stable circular null geodesic [@Abr-Mil-Stu:1993:PHYSR4:; @Stu-etal:2001:PHYSR4:]. From the behaviour of the effective potential of null geodesics in the exterior, vacuum Schwarzschild spacetimes, determining the unstable null circular geodesics at the radius $r_{\mathrm{ph}} = 3GM/c^2$ (see, e.g., [@Mis-Tho-Whe:1973:Gra:]), we can conclude that any spherically symmetric, static non-singular interior spacetime with radius $R < r_{\mathrm{ph}}$ admits existence of bound null geodesics. The trapped null geodesics are then governed by both internal and external barriers of the null geodesics effective potential. [^1] The realistic equations of state admitting the existence of the extremely compact objects were found and investigated for neutron stars, quark stars and Q-stars [@Bah-Lyn-Sel:1990:ApJ:; @Mil-Sha-Nol:1998:MNRAS:; @Nil-Cla:2000:GRRelStarsPolyEOS:; @Stu-Tor-Hle-Urb:2009:; @Hle-Stu-Mra:2004:RAGtime4and5:CrossRef; @Oest:2001:RAGtime2and3:]. For spacetimes with a non-zero tidal charge, the situation is more complex as we shall demonstrate in the following.
The existence of bound null geodesics in ECS has interesting astrophysical consequences. For example, using the notion of the optical reference geometry [@Abr-Pra:1990:; @Stu-Hle-Jur:2000:; @Stu:1990:] it was shown that trapped modes of gravitational waves could influence some instabilities in these objects [@Abr-etal:1997:CLAQG:]. Trapping of neutrinos can be of high importance in the interior of the ECS: first, it will suppress the neutrino flow as measured by distant observers and, second, it can influence cooling of the ECS in a layer extending from some radius depending on details of their structure up to the surface radius. The cooling process could even be realized in a “two-temperature” regime, when the temperature profile in the interior of the star with no trapped neutrinos differs from the profile established in the external layer with trapped neutrinos [@Stu-Tor-Hle-Urb:2009:] modifying substantially the standard picture of the neutron star structure as given in [@Gle:1992:PHYSR4:; @Gle:2000:CompactStars:; @Hae-Zdu:1986:NATURE:; @Web-Gle:1992:ASTRJ2:; @Web:1999:Pul:]. In a different context, trapping of neutrinos inside rotating neutron stars has been discussed in [@Mallick-etal:2009:arXiv:0905.3605:].
The approximation of free, geodetical motion of neutrinos in the internal spacetime could be used when the mean free path of neutrinos $\lambda > R$. Neutrinos have inelastic scatter on electrons (muons) and elastic scatter on neutrons. The scatter cross section on electrons (neutrons) $\sigma_\mathrm{e}$ ($\sigma_\mathrm{n}$) determines the mean free path by formula $\lambda = \left(\sigma_{i} n_{i}\right)^{-1}$ where $n_i$ ($i = \mathrm{e, n}$) denotes the number density of electrons (neutrons). It was shown [@Sha-Teu:1983:BHWDNS:] that $$\lambda_\mathrm{e}\sim 9\times 10^7\left(\frac{\rho_\mathrm{nucl}}{\rho} \right)^{4/3}\left(\frac{100~\mathrm{keV}}{E_\nu}\right)^{3}~\mathrm{km},$$ while $$\lambda_\mathrm{n}\sim 300\frac{\rho_\mathrm{nucl}}{\rho} \left(\frac{100\mathrm{keV}}{E_\nu}\right)^{2}~\mathrm{km}.$$ There is $\lambda_\mathrm{e} \gtrsim 10$ km for $E_\nu \lesssim 20$ MeV and $\lambda_\mathrm{n} \gtrsim 10$ km for $E_\nu \lesssim 500$ keV. Therefore, in a few hours old neutron star, see [@Lat-Pra:2007:PhysRep:; @Sha-Teu:1983:BHWDNS:], at temperatures $T \lesssim 10^9$ K ($E_\nu \sim 100$ keV), the neutrino motion could be considered geodetical through whole the internal spacetime.
Bound neutrinos with mean free path${}\gg R$ will slow down the cooling. Of course, they will be re-scattered due to finiteness of the mean free path. An eventual scattering of trapped neutrinos will cause change of their impact parameter, therefore, some of them will escape the ECS, suppressing thus the slow down of the cooling process in the region of neutrino trapping. Clearly, the scattering effect of the trapped neutrinos is a complex process deserving sophisticated numerical code based on the Monte Carlo method (we expect modelling of this effect in future).
The cooling process deserves sophisticated analytical estimates and detailed numerical simulations. As a first step in considering the role of trapped neutrinos in ECS, efficiency of the neutrino trapping effect was studied in the simple case of the internal Schwarzschild spacetime with uniform distribution of energy density (but a nontrivial pressure profile) and isotropic and uniform distribution of local neutrino luminosity, when all the calculations can be realized in terms of elementary functions only (see [@Stu-Tor-Hle-Urb:2009:]). [^2] The influence on the neutrino luminosity of the star is given by a luminosity trapping coefficient relating the total number of trapped neutrinos and the total number of radiated neutrinos (per unit time of distant observers). The influence on the cooling process is given by two “cooling” trapping coefficients: a “local” one given by ratio of trapped and radiated neutrinos at any radius where the trapping occurs, and the “global” one giving ratio of trapped and radiated neutrinos (per unit time of distant observers) integrated over whole the region where the trapping occurs.
In the present paper we study the role of the braneworld “tidal charge” parameter on the existence of ECS and estimate its role in the efficiency of the neutrino trapping process using the same trapping coefficients as those introduced in [@Stu-Tor-Hle-Urb:2009:]. We consider the trapping effect in the simple case of objects with uniform distribution of energy density. The external field is given by the Reissner-Nordström geometry [@Ger-Maar:2001:]. This special solution is most extensively studied in the literature, since its external field is identical to those of the black hole solution with spherical symmetry [@Dad-Maar-Pap-Rez:2000:PHYSR4:] — therefore, the results of our study could be efficiently compared to the results obtained in studies of other phenomena [@Boh-Har-Lob:2008:CLAQG:; @Kot-Stu-Tor:2008:CLASQG:]. We consider both positive and negative values of the braneworld tidal parameter — usually, only the negatively valued tidal charges (corresponding to positive tension of the brane) are discussed [@Ger-Maar:2001:], but the inverse situation with positive tidal charge is not excluded and will be treated here, since it could demonstrate some new interesting effects.
Our paper is organized as follows. In Section \[SECbrwns\], we summarize properties and matching conditions of the internal uniform energy density spacetime and the Reissner-Nordström external spacetime. In Section \[SECogeoi\], null geodesics of both the internal and external spacetime are described in terms of properly given effective potential and the ECS are classified according to the properties of the trapping region of the null geodesic motion. In Section \[SECtrappn\], the trapping of neutrinos is discused. In Section \[SECeffnt\], the trapping efficiency coefficients are defined for both the total neutrino luminosity and neutrino cooling process, and determined for all kinds of the braneworld ECS. In Section \[SECconcl\], concluding remarks are presented. Throughout the paper, we shall use the high-energy units with $\hbar = c = k_\mathrm{B} = 1$, if not stated otherwise. For simplicity, we assume zero rest energy of neutrinos, isotropic emission at each radius, and the period of evolution of the compact stars, when the temperature is low enough to describe the motion of neutrinos by the null geodesics of the spacetime.
Braneworld neutron stars {#SECbrwns}
========================
We describe the braneworld neutron stars by the simple model of uniform energy density interior and the “tidally” charged exterior. We give the internal and external geometry and the matching conditions. Accordingly, we determine conditions on the existence of ECS depending on the tidal parameter characterizing the external spacetime.
Matching conditions of internal and external spacetimes
-------------------------------------------------------
In the standard Schwarzschild coordinates and the high-energy units, the line element of the spherically symmetric spacetimes reads
$$\mathrm{d}s^2 = -A^{2}(r)\mathrm{d}t^2 + B^{2}(r)\mathrm{d}r^2 + r^{2}\mathrm{d}\Omega.\label{EQmetric}$$
The internal solution is characterized by the uniform energy density distribution — $\varrho = \mathrm{const}$, and by the tension of the brane — $\lambda$. We assume $\varrho > 0$, but we allow both possibilities $\lambda > 0$, $\lambda < 0$ for completeness. The internal geometry is matched to the external geometry at the surface of the star $r = R$. The line element of the internal solution is given by the metric coefficients $A^{-}(r)$, $B^{-}(r)$ that are determined by [@Ger-Maar:2001:]
$$A^{-}(r) = \frac{\Delta(R)}{\left(1+p(r)/\varrho\right)}\label{EQamin}$$
and $$\left(B^{-}(r)\right)^2 = \frac{1}{\Delta^{2}(r)} = \left[1-\frac{2G M}{r}\left(\frac{r}{R}\right)^3 \left(1+\frac{\varrho}{2\lambda}\right)\right]^{-1},$$ where $M = \frac{4}{3}\pi\varrho R^3$.
The pressure radial profile is given by $$\frac{p(r)}{\varrho} = \frac{\left[\Delta(r) - \Delta(R)\right]\left(1 + \varrho/\lambda\right)}{3\Delta(R) - \Delta(r) + \left[3\Delta(R) - 2\Delta(r)\right]\left(\varrho/\lambda\right)}.$$ For $r = R$, there is $p(r)/\varrho = 0$. The maximum of the pressure profile is at $r = 0$.
The reality condition on the metric coefficient $B^{-}(r)$ (taken at $r = R$) implies a relation between $\lambda$, $\varrho$ and $R$ that can be expressed in the form $$\label{EQRm2GM}
\frac{G M}{R-2G M} \geq \frac{\varrho}{\lambda}.$$
Considering the restriction $R > 2 G M$ ($R < 2 G M$), we can see that the reality condition (\[EQRm2GM\]) is satisfied for all $\lambda < 0$ (forbidden for all $\lambda > 0$), while for positive tension $\lambda > 0$ (negative tension $\lambda < 0$), we obtain a limit on the positive (negative) tension given by [@Ger-Maar:2001:] $$\lambda \geq \left(\frac{R - 2 G M}{G M}\right)\varrho .$$ (Notice that considering possibility of $R < 2 G M$, we obtain a limit on the negative tension.)
The line element of the external geometry is given by the metric coefficients $A^{+}(r)$, $B^{+}(r)$ that are determined by $$\left( A^{+}(r)\right)^{2} = \left( B^{+}(r)\right)^{-2} = 1 - \frac{2G\mathcal{M}}{r} + \frac{q}{r^2},$$ where, due to the matching conditions on the neutron star surface, i.e., $A^{-}(R) = A^{+}(R)$, $B^{-}(R) = B^{+}(R)$, the external mass parameter $\mathcal{M}$ and external tidal charge parameter $q$ are related to the internal geometry parameters $\varrho$, $\lambda$ (and $M$) by the relations
$$q = -3G M R\frac{\varrho}{\lambda}$$
$$\mathcal{M} = M \left(1-\frac{\varrho}{\lambda}\right).$$
For $\lambda > 0$, the tidal charge $q < 0$ and $\mathcal{M} < M$, while for $\lambda < 0$, there is $q > 0$ and $\mathcal{M} > M$. Notice that for $\lambda > 0$, the condition $\varrho < \lambda$ has to be satisfied in order to have $\mathcal{M} > 0$.
For our purposes, it is convenient to express the internal spacetime coefficients using the parameters of the external spacetime that can be directly determined from observations of accretion and optical phenomena in vicinity of the neutron stars. Since the matching conditions imply the relations $$\label{EQq}
q = - \frac{3G M R \varrho}{\lambda} = \frac{3G \mathcal{M}R}{\left(1 - \lambda/\varrho\right)},$$
$$\frac{\lambda}{\varrho} = 1-\frac{3G\mathcal{M}R}{q},$$
the internal metric can be expressed in terms of the external parameters $\mathcal{M}$, $R$, $q$ and a new parameter $$\label{EQX}
X \equiv \frac{q}{3G \mathcal{M}R}$$ in the form $$\left(B^{-}(r)\right)^2 = \Delta^{-2}(r) = \left[1 - \frac{2G\mathcal{M}r^2}{R^3} \left(1 - \frac{3}{2}X\right)\right]^{-1}$$ $$A^{-}(r) = \frac{\Delta(R)\left[3\Delta(R)\left(2X - 1\right)+\Delta(r)\left(1 - 3X\right)\right]}{2\Delta(R)\left(2X - 1\right) - \Delta(r) X}.$$ The pressure profile can be expressed in the form
$$\frac{p(r)}{\varrho} = \frac{\left[\Delta(r) - \Delta(R)\right] \left(2X - 1\right)}{\left[3\Delta(R) - \Delta(r)\right] \left(X - 1\right) + \left[3\Delta(R) - 2\Delta(r)\right] X}.$$
The observational restrictions obtained from the measurements outside the neutron star then can be applied to the internal parameters using the relations (\[EQq\])–(\[EQX\]).
Limit on existence of uniform density stars
-------------------------------------------
The limit on the existence of the uniform density spherical configuration is related to their compactness and is determined by the condition of pressure finiteness in their center. We then find the limit on compactness of the star given in terms of the gravitational mass related to the internal geometry [@Ger-Maar:2001:]: $$\label{EQGMR}
\frac{G M}{R}\leq \frac{4}{9}\left[\frac{1 + \frac{5}{4}\frac{\varrho}{\lambda}}{\left(1 + \frac{\varrho}{\lambda}\right)^2}\right].$$ The lowest order correction is given by $$\frac{G M}{R}\leq \frac{4}{9}\left[1 - \frac{3}{4}\frac{\varrho}{\lambda}\right].$$ Using the external gravitational mass parameter $\mathcal{M}$, the compactness limit (\[EQGMR\]) is transformed to the form $$\frac{G \mathcal{M}}{R}\leq \frac{4}{9}\left[\frac{\left(1 + \frac{5}{4}\frac{\varrho}{\lambda}\right)\left(1 - \frac{\varrho}{\lambda}\right)}{\left(1 + \frac{\varrho}{\lambda}\right)^2}\right].$$ Introducing the gravitational radius $$r_\mathrm{g} \equiv G \mathcal{M}$$ and dimensionless braneworld tidal charge $$b \equiv \frac{q}{r_\mathrm{g}^{2}},$$ we can put $$X = \frac{q}{3G\mathcal{M}R} = \frac{1}{3}b\frac{r_\mathrm{g}}{R}$$ and we arrive to the relations $$\Delta^{2}(R) = 1 - 2\frac{r_\mathrm{g}}{R} + b \left(\frac{r_\mathrm{g}}{R}\right)^2$$ $$\Delta^{2}(r) = 1 - \frac{r_\mathrm{g}}{r}\left(\frac{r}{R}\right)^3 \left(2 - b\frac{r_\mathrm{g}}{R}\right)$$ In terms of dimensionless units ($r_\mathrm{g} = 1$), the pressure function reads $$\frac{p}{\varrho}\left(r, R, b\right) = \frac{\left(\Delta(r) - \Delta(R)\right) \left(2b - 3R\right)}{3 \Delta(R) \left(2b - 3R\right) - 3\Delta(r) \left(b - R\right)}.$$ The pressure increases monotonously with radius decreasing. The central pressure is given by $$\frac{p}{\varrho}\left(r = 0, R, b\right) = \frac{3R - 2b}{3R \left(\sqrt{b + (R - 2)R} + R - 3\right) + 6b}.$$ The reality condition of the central pressure reads $b + (R-2) R\geq 0$ and is equivalent to the condition $\Delta(R)\in \mathbb{R}$. The pressure in the center of the star must be finite and positive. These conditions imply the limits $R_\mathrm{min}(b)$ on the existence of uniform density stars. The central pressure $p(r = 0, R, b)$ diverges when the surface radius $R$ satisfies the condition $$3R \left(\sqrt{b + (R-2) R} + R-3\right) + 6b = 0.$$ This leads to a cubic equation relative to $R$ which gives one solution that is relevant for $b\leq 1$ and reads $$\begin{aligned}
R & = & \frac{1}{4} \left(\frac{(b-9) (b-1)}{\sqrt[3]{8 \sqrt{(b-1)^2 b^3}+b \left[b (b+17)-45\right]+27}} + \right. \nonumber \\
& & + \left.\sqrt[3]{8 \sqrt{(b-1)^2 b^3}+b \left[b (b+17)-45\right]+27}+b+3\right).
\end{aligned}$$ The solution is depicted on Figure \[FIGzone\] as the part of $R_\mathrm{min}(b)$ denoted by $p \rightarrow \infty$. For $b = 0$, we arrive at the standard condition $R > \frac{9}{4}$ (see [@Stu-Tor-Hle-Urb:2009:]). For $b > 1$ the relevant limit is given by the condition of positiveness of the central pressure. It reads $$R_\mathrm{min}(b| b > 1) = \frac{2}{3}b$$ and is denoted by $p \rightarrow 0$. For smaller $R$ the central pressure is not positive.
Note that for $3/4 < b < 1$ there is a region of surface radii given by the condition $2b/3 < R < 1-\sqrt{1-b}$ where $p(r = 0)$ is positive. However, the star surface is located under the inner horizon of the external spacetime that belongs to the black hole type R-N spacetime. Such a configuration has to be hidden under the inner horizon of the external spacetime and is irrelevant for our considerations.
![Classification of the extremely compact stars (ECS). The parameter space $(b-R)$ is separated into five Zones I–IV${}_\mathrm{a, b}$; their properties are given in Section \[SECngaep\].[]{data-label="FIGzone"}](Fig1.pdf){width="0.8\hsize"}
Extremely compact neutron stars
-------------------------------
The ECS (neutron, quark, or hybrid) are defined by the existence of trapped null geodesics in the interior of the compact star. In the case of internal uniform density Schwarzschild spacetimes ($b = 0$), such objects appear just when the surface of the compact star is located under the photon circular geodesic of the external vacuum Schwarzschild spacetime located at $r_\mathrm{ph} = 3r_\mathrm{g}$ [@Stu-Tor-Hle-Urb:2009:]. In the braneworld uniform density compact stars this simple rule does not hold and the situation is more complex due to different character of the external spacetime and its relation to the internal spacetime.
Assuming the external spacetime to be of the Reissner-Nordström (R-N) type with the tidal charge substituting the electric charge squared appearing in the standard R-N spacetimes, the photon circular geodesics are given by the condition $$r^2 - 3r_\mathrm{g}r + 2q = 0.$$
However, it is convenient to use dimensionless tidal charge $b$ and dimensionless radial coordinate defined by $$r/r_\mathrm{g}\rightarrow r.$$
For negative tidal charges ($b < 0$) external spacetimes of the black-hole type are allowed only [@Kot-Stu-Tor:2008:CLASQG:]. The radii of the photon circular orbits are given by $$r_{\mathrm{ph}\pm} = \frac{3}{2}\left(1 \pm \sqrt{1-\frac{8}{9}b}\right). \label{EQph}$$ However, only the outer photon circular geodesic at $r_\mathrm{ph+}$ is physically relevant and we see immediately that $(b < 0)$ there is $r_{\mathrm{ph}}>3$.
For braneworld neutron stars with a positive tidal charge ($b > 0$) both black-hole and naked singularity R-N spacetimes are relevant [@Kot-Stu-Tor:2008:CLASQG:]. In the black-hole type spacetimes ($b \leq 1$), the photon circular orbits are given by the relation (\[EQph\]), but only the outer solution corresponding to an unstable orbit can be astrophysically relevant since the surface of the compact star has to be located above radius of the outer horizon. In the naked singularity spacetimes ($b > 1$) two photon circular orbits are can be astrophysically relevant when $b < 9/8$, the inner one being stable, the outer one — unstable. In the spacetimes with $b > 9/8$, no photon circular orbits can exist. Therefore, in the braneworld spacetimes, the existence of ECS is governed by the interplay of the behaviour of the internal and external effective potential of the motion and will be determined and classified in the following section.
Null geodesics of internal and external spacetimes {#SECogeoi}
==================================================
In terms of the tetrad formalism the metric (\[EQmetric\]) reads $$\mathrm{d} s^{2} = -[\omega^{(t)}]^{2}
+[\omega^{(r)}]^{2}
+[\omega^{(\theta)}]^{2}
+[\omega^{(\phi)}]^{2},$$ where $$\omega^{(t)} = A(r)\,\mathrm{d} t,\quad
\omega^{(r)} = B(r)\,\mathrm{d} r,\quad
\omega^{(\theta)} = r\,\mathrm{d} \theta,\quad
\omega^{(\phi)} = r\sin\theta\,\mathrm{d} \phi.$$ Since there is $e^{\mu}_{(\alpha)} = \left[\omega^{(\alpha)}_{\mu}\right]^{-1}$, the tetrad of 4-vectors is given by $$\vec{e}_{(t)} = \frac{1}{A(r)} \frac{\partial}{\partial t},\quad
\vec{e}_{(r)} = \frac{1}{B(r)} \frac{\partial}{\partial r},\quad
\vec{e}_{(\theta)} = \frac{1}{r} \frac{\partial}{\partial \theta},\quad
\vec{e}_{(\phi)} = \frac{1}{r\sin\theta}\frac{\partial}{\partial \phi}.$$ Tetrad components of 4-momentum of a test particle or a photon are determined by the projections $p_{(\alpha)} = p_{\mu}e^{\mu}_{(\alpha)}$, $p^{(\alpha)} = p^{\mu}\omega_{\mu}^{(\alpha)}$ which give quantities measured by the local observers.
Null geodesics and effective potential {#SECngaep}
--------------------------------------
We consider the period of evolution and cooling of ECS when their temperature falls down enough that the motion of neutrinos can be considered free, i.e., geodetical. We can assume this period starts at the moment when mean free path of neutrinos becomes to be comparable to the radius $R$, i.e., in hours after the gravitational collapse creating the compact object [@Gle:1992:PHYSR4:; @Gle:2000:CompactStars:; @Hae-Zdu:1986:NATURE:; @Web:1999:Pul:; @Sha-Teu:1983:BHWDNS:]. In fact, there are arguments that this condition starts to be fulfilled about 50 s after collapse to a proto-neutron star [@Lat-Pra:2007:PhysRep:; @Lat-Pra:2004:SCIENCE:]. Weak interaction of ultrarelativistic (massless) neutrinos implies their motion along null geodesics obeying the equations ($\lambda$ is an affine parameter) $$\frac{\mathrm{D} p^{\mu}}{\mathrm{d} \lambda} = 0,\qquad
p^{\mu}p_{\mu} = 0. \label{EQgeod}$$
Due to the existence of two Killing vector fields: the temporal $\partial/\partial t$ one, and the azimuthal $\partial/\partial \phi$ one, two conserved components of the 4-momentum must exist: $$E=-p_{t}\quad\mbox{(energy),}\qquad
L=p_{\phi}\quad\mbox{(axial angular momentum)}. \label{EQconserv}$$ Moreover, the motion plane is central. For a single-particle motion, one can set $\theta=\pi/2=\mathrm{const}$, choosing the equatorial plane.
The motion along null-geodesics is independent of energy (frequency) and can conveniently be described in terms of the impact parameter $$\ell = \frac{L}{E}. \label{EQimpar}$$ Then (\[EQgeod\]) yields the relevant equation governing the radial motion in the form $$(p^{r})^{2} = A^{-2}(r)B^{-2}(r)E^{2}
\left(1-A^{2}(r)\frac{\ell^{2}}{r^{2}}\right). \label{EQgovmot}$$ Clearly, the energy $E$ is irrelevant and can be used for rescalling of the affine parameter $\lambda$. The radial motion is restricted by an effective potential related to the impact parameter $\ell$, and defined by the relations $$\ell^{2} \leq V{}_{\mathrm{eff}} =\left\{
\begin{array}{lll}
V{}_{\mathrm{eff}}^{\mathrm{int}} = \displaystyle\frac{r^2}{(A^{-}(r))^2} & \quad\mbox{for} & r\leq R\\
V{}_{\mathrm{eff}}^{\mathrm{ext}} = \displaystyle\frac{r^2}{(A^{+}(r))^2} & \quad\mbox{for} & r > R.
\end{array}\right.$$ $V{}_{\mathrm{eff}}^{\mathrm{int}}$ is the effective potential of the null-geodetical motion in the internal braneworld spacetime and $V{}_{\mathrm{eff}}^{\mathrm{ext}}$ is the effective potential of the null-geodetical motion in the external, vacuum R-N spacetime. We shall see in the following that due to the bulk space tidal effects on the matching of the internal and external spacetimes (see [@Ger-Maar:2001:] for details) we obtain a non-standard variety of relations of the effective potentials in the ECS interior and exterior.
Using the dimensionless radial coordinate expressed in terms of the gravitational radius ($r/G\mathcal{M}\rightarrow r$), and dimensionless tidal charge $b$, we obtain the relations
$$V{}_{\mathrm{eff}}^{\mathrm{int}} = \frac{r^2 R^2 \left[b Y - 2R (2b - 3R) Z\right]^2}{9 \left[b + (R-2) R\right] \left[(R - b) + R (2b -3R) Z\right]^2}, \label{EQellsb}$$
$$V{}_{\mathrm{eff}}^{\mathrm{ext}} = \frac{r^4}{b + (r - 2) r},$$
where $$\begin{aligned}
Y &\equiv & R^2 \Delta(r) = \sqrt{r^2(b-2R)+R^4},\\
Z &\equiv & R \Delta(R) = \sqrt{b+(R-2) R}.
\end{aligned}$$
![The effective potential $V_{\mathrm{eff}}$ of both internal (dashed lines) and external spacetimes (full lines) given for the black-hole type ($b < 1$) and naked-singularity ($b > 1$) type of the external spacetime. All characteristic cases of its behaviour are presented for both ECS and non-extreme compact stars.[]{data-label="FIGVdiff"}](Fig2a.pdf "fig:"){width="\hsize"} ![The effective potential $V_{\mathrm{eff}}$ of both internal (dashed lines) and external spacetimes (full lines) given for the black-hole type ($b < 1$) and naked-singularity ($b > 1$) type of the external spacetime. All characteristic cases of its behaviour are presented for both ECS and non-extreme compact stars.[]{data-label="FIGVdiff"}](Fig2c.pdf "fig:"){width="\hsize"}
![The effective potential $V_{\mathrm{eff}}$ of both internal (dashed lines) and external spacetimes (full lines) given for the black-hole type ($b < 1$) and naked-singularity ($b > 1$) type of the external spacetime. All characteristic cases of its behaviour are presented for both ECS and non-extreme compact stars.[]{data-label="FIGVdiff"}](Fig2b.pdf "fig:"){width="\hsize"} ![The effective potential $V_{\mathrm{eff}}$ of both internal (dashed lines) and external spacetimes (full lines) given for the black-hole type ($b < 1$) and naked-singularity ($b > 1$) type of the external spacetime. All characteristic cases of its behaviour are presented for both ECS and non-extreme compact stars.[]{data-label="FIGVdiff"}](Fig2d.pdf "fig:"){width="\hsize"}
![The effective potential $V_{\mathrm{eff}}$ of both internal (dashed lines) and external spacetimes (full lines) given for the black-hole type ($b < 1$) and naked-singularity ($b > 1$) type of the external spacetime. All characteristic cases of its behaviour are presented for both ECS and non-extreme compact stars.[]{data-label="FIGVdiff"}](Fig2e.pdf){width=".49\hsize"}
Circular null geodesics, located at $r_\mathrm{c(i)}$ and $r_\mathrm{c(e)}$, respectively, are given by the local extrema of the effective potential ($\partial {V_{\mathrm{eff}}}/\partial {r} = 0$). In the internal spacetime we have to solve a nontrivial equation $r$: $$\begin{aligned}
\lefteqn{\mathrm{d}\left(V{}_\mathrm{eff}^\mathrm{int}(r, R, b)\right)/\mathrm{d}r \equiv}\nonumber \\
& & \left[2 r R^2 \left(b (Y - 4 R Z) + 6 R^2 Z\right) \left\{-b^3 \left(r^2 (Y-4 R Z)+8 R^2 Y\right) + b^2 R\times \right.\right. \nonumber \\
& & \times \left(r^2 (3 Y-14 R Z)+R^2 \left(6 R^2 Z - 9 R Y + 40 Y\right)\right)+b R^2 \left(R^2(-13 R^2 Z + \right.\nonumber \\
& & \left.\left.\left.\left. + 25 R Y- 66 Y\right)- 2 r^2 (Y- 6 R Z)\right)+ 6 R^5 \left(R^2 Z-3 R Y+ 6 Y\right)\right\}\right]\times \nonumber \\
& &\times \left[9 \left(b+(R-2) R\right) Y \left(R (Y-3 R Z)-b \left(Y-\nonumber 2 R Z\right)\right)^3\right]^{-1} = 0.\\
\end{aligned}$$ This equation can be solved by using numerical methods and determines the loci $r_\mathrm{c(i)}(R, b)$ of the internal circular null geodesics that are stable and correspond to a local maximum of the internal effective potential. From the radius $r_\mathrm{c(i)}$ we easily obtain the impact parameter $\ell^{2}_{\mathrm{c(i)}}(R, b)$ using equation (\[EQellsb\]). The existence of the internal stable circular null geodesics is allowed only for the surface radius $R$ limited by values explicitly given by $$R_\mathrm{I\pm} = \frac{3}{2} \pm \sqrt{\frac{9}{4} - \frac{5}{3}b}$$ that has to satisfy naturally also the condition $R > R_\mathrm{min}(b)$.
In the external spacetime, the photon circular geodesics and the related local extrema of the effective potential are given by equation (\[EQph\]) and their properties were discussed in the previous section.
When we take into account the (non-)existence of the local extrema of the internal/external part of the effective potential $V_\mathrm{eff}$, we obtain the whole region of existence of ECS. The function $R{}_\mathrm{max}^\mathrm{ECS}(b)$ is determined by the relation $$R{}_\mathrm{max}^{\mathrm{ECS}} =\left\{
\begin{array}{lll}
r_\mathrm{ph} \equiv \frac{1}{2} \left(3 + \sqrt{9-8 b}\right) & \quad\mbox{for} & b \leq 0\\
R_\mathrm{I+} \equiv \frac{3}{2} + \sqrt{\frac{9}{4} - \frac{5}{3}b} & \quad\mbox{for} & 0 < b < 27/20
\end{array}
\right.,$$ while for $R{}_\mathrm{min}^\mathrm{ECS}(b)$ we have $$R{}_\mathrm{min}^{\mathrm{ECS}} = \left\{
\begin{array}{lll}
R_\mathrm{min} & \quad\mbox{for} & b \leq 1\\
R_\mathrm{I-} \equiv \frac{3}{2} - \sqrt{\frac{9}{4} - \frac{5}{3}b} & \quad\mbox{for}& 1 < b < 27/20
\end{array}
\right..$$ The region of ECS in the parameter space is represented in Figure \[FIGzone\].
Typical behaviour of the effective potential of the null-geodetical motion $V{}_{\mathrm{eff}}$ is demonstrated in Figure \[FIGVdiff\] for appropriatelly chosen values of the tidal charge $b$. Here, and henceforth, we express the radii in units of the gravitational radius $r_g = G\mathcal{M}$. The selection of the tidal charge values used in Figure \[FIGVdiff\] demonstrates the full classification of the behavior of the effective potential in both internal and external spacetimes. When the effective potential in the internal or external spacetime (or in both of them) has a local extreme corresponding to a photon circular geodesic, trapped null geodesics can appear if the surface radius $R$ is properly chosen giving an ECS. On the other hand, we find an ordinary compact star if the surface is chosen in such a way that no local extrema of the effective potentials exist (see Figure \[FIGVdiff\]e). Separation of the zones of compact stars of different character, both extremely compact and ordinary compact, in the parameter space $(b-R)$, is determined by the functions $R_\mathrm{min}(b)$, $R_\mathrm{max}(b)$ and $r_\mathrm{ph}(b)$ where the last function governs radius of the photon circular geodesics in the external spacetime (both of them for $1 < b < 9/8$). The subdivision of the Zone IV in the tidal charge range $1 < b < 9/8$ is given by the relation of the magnitude of the effective potential at the surface of the ECS and the magnitude of the external effective potential at its local minimum, and is determined numerically (see Figure \[FIGzone\]).
We divided the region of the parameter space $R-b$ corresponding to the existence of ECS into five zones. The classification is based on the existence of local maxima/minima of $V{}^{\mathrm{int/ext}}_{\mathrm{eff}}$ in the following way (see Figures \[FIGzone\] and \[FIGbehzoneALL\])
- [*Zone I* there exist maximum of $V{}^{\mathrm{int}}_{\mathrm{eff}}$ and minimum of $V{}^{\mathrm{ext}}_{\mathrm{eff}}$, both located at $r \neq R$, $\mathrm{min} V{}^{\mathrm{ext}}_{\mathrm{eff}} < V{}^{\mathrm{ext}}_{\mathrm{eff}}(r = R)$, and there is no local maximum of $V{}^{\mathrm{ext}}_{\mathrm{eff}}$ at $r > R$;]{}
- [*Zone II* there exist maximum of $V{}^{\mathrm{int}}_{\mathrm{eff}}$ at $r = R$ and minimum of $V{}^{\mathrm{ext}}_{\mathrm{eff}}$ at $r \neq R$, $\mathrm{min} V{}^{\mathrm{ext}}_{\mathrm{eff}} < \mathrm{max} V{}^{\mathrm{int}}_{\mathrm{eff}}$, and there is no local maximum of $V{}^{\mathrm{ext}}_{\mathrm{eff}}$ at $r > R$;]{}
- [*Zone III* there exist maximum of $V{}^{\mathrm{int}}_{\mathrm{eff}}$ at $r < R$ and minimum of $V{}^{\mathrm{ext}}_{\mathrm{eff}}$ at $r = R$; there is no local maximum of $V{}^{\mathrm{ext}}_{\mathrm{eff}}$ at $r > R$;]{}
- *Zone IV* there exist both the local maximum and local minimum of $V{}^{\mathrm{ext}}_{\mathrm{eff}}$ at $r > R$, this zone can be divided into two parts
- [minimum of $V{}^{\mathrm{ext}}_{\mathrm{eff}} < V{}^{\mathrm{int}}_{\mathrm{eff}}\left(r = R\right)$;]{}
- [ minimum of $V{}^{\mathrm{ext}}_{\mathrm{eff}} > V{}^{\mathrm{int}}_{\mathrm{eff}}\left(r = R\right)$.]{}
Trapping of neutrinos {#SECtrappn}
=====================
In ECS, some part of produced neutrinos is prevented from escaping these static objects because of stable circular null geodesics existing in the internal spacetime, and/or because of unstable circular null geodesics existing in the external spacetime (Figure \[FIGbehzoneALL\]). For external braneworld naked-singularity spacetimes with tidal charge in the interval $1 < b < 9/8$, an additional stable circular null geodesic exists that can be important for appearance and structure of accretion discs ([@Stu-Schee:2010:CLAQG:]), but it is irrelevant for trapping of neutrinos radiated by ECS.
The relation $\partial {V{}^{int}_{\mathrm{eff}}}/\partial {r} = 0$ determines location of the stable circular null geodesics of the internal spacetime and implies the impact parameter which corresponds to the local maximum of the effective potential $V{}_{\mathrm{eff}}^{\mathrm{int}}$ at $r_{\mathrm{c(i)}}$ to be given by $\ell_{\mathrm{c(i)}}^{2}$.
For $b \leq 9/8$, the local minimum of $V{}_{\mathrm{eff}}^{\mathrm{ext}}$ is located at $$r_{\mathrm{c(e)}}=\frac{1}{2} \left(3 + \sqrt{9 - 8 b}\right)$$ and corresponds to the unstable circular null geodesics of the Reissner-Nordström external spacetime, with the impact parameter determined by $$\ell_{\mathrm{c(e)}}^{2} = \frac{\left(3 + \sqrt{9 - 8 b}\right)^4}{8 \left(3 - 2 b + \sqrt{9 - 8 b}\right)}.$$ For the ECS spacetimes corresponding to all of the Zones I–IV${}_\mathrm{a, b}$ of the parameter space $(b-R)$, typical behaviour of the effective potentials $V{}_{\mathrm{eff}}^{\mathrm{int}}$ and $V{}_{\mathrm{eff}}^{\mathrm{ext}}$ is represented in Figure \[FIGbehzoneALL\] where the regions of trapped neutrinos are shown explicitly.
![Schematic classification of the effective potential $V{}_\mathrm{eff}$ of the ECS for all the Zones I–IV$_\mathrm{a, b}$ of the parameter space $(b-R)$ as depicted on Figure \[FIGzone\].[]{data-label="FIGbehzoneALL"}](Fig3a.pdf "fig:"){width="\hsize"} ![Schematic classification of the effective potential $V{}_\mathrm{eff}$ of the ECS for all the Zones I–IV$_\mathrm{a, b}$ of the parameter space $(b-R)$ as depicted on Figure \[FIGzone\].[]{data-label="FIGbehzoneALL"}](Fig3c.pdf "fig:"){width="\hsize"}
![Schematic classification of the effective potential $V{}_\mathrm{eff}$ of the ECS for all the Zones I–IV$_\mathrm{a, b}$ of the parameter space $(b-R)$ as depicted on Figure \[FIGzone\].[]{data-label="FIGbehzoneALL"}](Fig3b.pdf "fig:"){width="\hsize"} ![Schematic classification of the effective potential $V{}_\mathrm{eff}$ of the ECS for all the Zones I–IV$_\mathrm{a, b}$ of the parameter space $(b-R)$ as depicted on Figure \[FIGzone\].[]{data-label="FIGbehzoneALL"}](Fig3d.pdf "fig:"){width="\hsize"}
![Schematic classification of the effective potential $V{}_\mathrm{eff}$ of the ECS for all the Zones I–IV$_\mathrm{a, b}$ of the parameter space $(b-R)$ as depicted on Figure \[FIGzone\].[]{data-label="FIGbehzoneALL"}](Fig3e.pdf){width="0.499\hsize"}
Regions of trapping
-------------------
Bound neutrinos (depicted by the gray filled areas in Figure \[FIGbehzoneALL\]) may be fully trapped inside the internal spacetime, or may, in some spacetimes, appear outside the interior of the ECS being trapped by its strong gravitational field, and entering the interior again; note that the trapped neutrinos moving outside the compact star occur in the spacetimes of the Zones I, II and IV${}_\mathrm{a}$. Therefore, we divide the trapped neutrinos into two families:
- *“Internal” bound neutrinos* (depicted using light gray filled area) with impact parameter between $\ell^{2}_{\mathrm{int}}(R)$ and $\ell^{2}_{\mathrm{c(i)}}$; motion of these neutrinos is restricted to the interior of the ECS. The internal bound neutrinos appear exclusively in spacetimes of the Zones III and IV${}_\mathrm{b}$, but they occur also in the spacetimes of the Zones I and IV${}_\mathrm{a}$.
- *“External” bound neutrinos* (depicted by the dark gray filling) with impact parameter between $\ell^{2}_{\mathrm{c(e)}}$ and $\ell^{2}_{\mathrm{int}}(R)$; such bound neutrinos may leave the ECS interior, but they re-enter it. In spacetimes of the Zone II all bound neutrinos are of this type.
![Location of the radii relevant for the trapping of neutrinos $r_\mathrm{b(i)}$, $r_\mathrm{b(e)}$ and $r_\mathrm{c(e)}$ given as functions of the surface radius $R$ for characteristic values of the tidal charge $b$. The range of the radius $R$ corresponding to the Zones I–IV of the parameter space $(b-R)$ is depicted in all presented figures.[]{data-label="FIGradius"}](Fig4a.pdf "fig:"){width="\hsize"} ![Location of the radii relevant for the trapping of neutrinos $r_\mathrm{b(i)}$, $r_\mathrm{b(e)}$ and $r_\mathrm{c(e)}$ given as functions of the surface radius $R$ for characteristic values of the tidal charge $b$. The range of the radius $R$ corresponding to the Zones I–IV of the parameter space $(b-R)$ is depicted in all presented figures.[]{data-label="FIGradius"}](Fig4c.pdf "fig:"){width="\hsize"}
![Location of the radii relevant for the trapping of neutrinos $r_\mathrm{b(i)}$, $r_\mathrm{b(e)}$ and $r_\mathrm{c(e)}$ given as functions of the surface radius $R$ for characteristic values of the tidal charge $b$. The range of the radius $R$ corresponding to the Zones I–IV of the parameter space $(b-R)$ is depicted in all presented figures.[]{data-label="FIGradius"}](Fig4b.pdf "fig:"){width="\hsize"} ![Location of the radii relevant for the trapping of neutrinos $r_\mathrm{b(i)}$, $r_\mathrm{b(e)}$ and $r_\mathrm{c(e)}$ given as functions of the surface radius $R$ for characteristic values of the tidal charge $b$. The range of the radius $R$ corresponding to the Zones I–IV of the parameter space $(b-R)$ is depicted in all presented figures.[]{data-label="FIGradius"}](Fig4d.pdf "fig:"){width="\hsize"}
The pericentre and the impact parameter of the “external” marginally bound neutrinos ($r_{\mathrm{b(e)}}$; Zones I, II, IV${}_\mathrm{a}$) can be determined from the condition $$V{}^{\mathrm{int}}_{\mathrm{eff}}(r = r_{\mathrm{b(e)}}, R, b) = \frac{\left(3 + \sqrt{9-8 b}\right)^4}{8 \left(3 - 2 b+\sqrt{9-8 b}\right)}\label{EQrbe}$$ while the pericentre and the impact parameter of the “internal” marginally bound neutrinos ($r_{\mathrm{b(i)}}$; Zones I, III, IV${}_\mathrm{a}$, IV${}_\mathrm{b}$) are given by the condition $$V{}^{\mathrm{int}}_{\mathrm{eff}}(r = r_{\mathrm{b(i)}}, R, b) = \frac{R^4}{b + (R - 2) R},\label{EQrbi}$$ see Figures \[FIGbehzoneALL\], \[FIGradius\] for the graphical representation. For completeness, we show in Fig. \[FIGradius\] also loci $r_\mathrm{c(i)}$ of the stable circular null geodesic and the surface radius $R$ of the ECS.
Directional angles
------------------
Considering (without loss of generality, as stated above equation (\[EQimpar\])) an equatorial motion, we can define the *directional angle* relative to the outward pointed radial direction measured in the emitter system (i.e., the local system of static observers in the internal spacetime) by the standard relations $$\sin\psi = \frac{p^{(\phi)}}{p^{(t)}}, \qquad
\cos\psi = \frac{p^{(r)}}{p^{(t)}},$$ where $$p^{(\mathrm{\alpha})} = p^{\mu}\omega_{\mu}^{(\mathrm{\alpha})},\qquad
p_{(\mathrm{\alpha})} = p_{\mu}e^{\mu}_{(\mathrm{\alpha})}$$ are the neutrino momentum component as measured by the static observers. Besides conserving components (\[EQconserv\]), and $p_{\theta} = 0$, equation (\[EQgovmot\]) implies $$p_{r} = \pm E A^{-1}B\left(1-A^2 \frac{\ell^{2}}{r^{2}}\right)^{1/2}.$$ For the directional angles we thus obtain relations $$\sin\psi = A\frac{\ell}{r},\qquad \cos\psi = \pm\left(1-\sin^{2}\psi\right)^{1/2},\label{EQcossin}$$ where $A$ is given by equation (\[EQamin\]).
The directional angle limit for the bound neutrinos is determined for spacetimes in the Zones I, II and IV${}_\mathrm{a}$ by the impact parameter $\ell_{\mathrm{c(e)}}^{2}$. We arrive to the relation $$\begin{aligned}
\cos\psi_{\mathrm{e}} = \pm \frac{1}{2} \sqrt{4-\frac{9 \left(\sqrt{9-8 b}+3\right)^4 Z^2 \left[(R-b) Y+R (2 b-3 R) Z\right]^2}{2 \left(-2 b+\sqrt{9-8 b}+3\right) r^2 R^2 \left[b Y+2 R (3 R-2 b) Z\right]^2}}.\label{EQsicose}
\end{aligned}$$ The interval of relevant radii is given by $r \in (r_{\mathrm{b(e)}}, R)$.
The directional angle limit for the “internal” bound neutrinos is determined by equation (\[EQcossin\]), where for $\ell$ we use $\ell = \ell_\mathrm{int}(R)$. In the Zone II spacetimes, there are none neutrinos bound only in their interior. On the other hand, for spacetimes in the Zones III and IV${}_\mathrm{b}$, this limit represents the total limit on all bound neutrinos. For the internal neutrinos we arrive to the relation $$\begin{aligned}
\cos\psi_{\mathrm{i}} = \pm \sqrt{1-\frac{9 R^2 \left[(R-b) Y + R (2 b-3 R) Z\right]^2}{r^2 \left[b Y + 2 R (3 R-2 b) Z\right]^2}}.
\end{aligned}$$ Apparently, the condition $\psi_\mathrm{i} > \psi_\mathrm{e}$ holds at any given radius $r < R$ of the ECS spacetimes belonging to the Zones of the parameter space where both values have good meaning.
Local escaped to produced neutrinos ratio
-----------------------------------------
![Schematic illustration of the bound-escape ratio at a radius $r \in (r_{\mathrm{b(e)}}, R)$ of an internal spacetime. Direction of the neutrino motion with respect to the static observers is related to $e_{(r)}$ giving the outward oriented radial direction. In the spacetimes of the Zones III and IV${}_\mathrm{b}$ only $\psi_\mathrm{i}$ cone is relevant, while for spacetime of the Zone II only angle $\psi_\mathrm{e}$ is relevant.[]{data-label="FIGbound"}](Fig5.pdf){width=".45\hsize"}
We assume that neutrinos are locally produced by isotropically emitting sources. Then escaped-to-produced-neutrinos ratio depends on a geometrical argument only. It is determined by the solid angle $2\Omega$ corresponding to escaping neutrinos (also inward emitted neutrinos must be involved because even these neutrinos can be radiated away), see Figure \[FIGbound\].
![Profiles of the local trapping factor $\beta (r/R, b)$ in the ECS with external spacetime of a black-hole type. They are constructed for fixed characteristic values of the brane parameter $b$ and sequences of the surface radius $R$ corresponding to the ECS. The upper figure represents the case of negative tidal charges ($b = -1.2$), while the lower figure represents the case of positive tidal charges ($b = 0.8$). The gray line connects the local maxima of the $\beta (r/R, b)$ profiles. The $\beta (r/R, b)$ profiles constructed for parameters belonging to different zones of the parameter space are separated by thick lines. In the regions of different zones the profiles are constructed for surface radii $R$ spaced by equal distances in the interval of maximal and minimal values of $R$ of ECS.[]{data-label="FIGbetaA"}](Fig6a.pdf "fig:"){width=".95\hsize"} ![Profiles of the local trapping factor $\beta (r/R, b)$ in the ECS with external spacetime of a black-hole type. They are constructed for fixed characteristic values of the brane parameter $b$ and sequences of the surface radius $R$ corresponding to the ECS. The upper figure represents the case of negative tidal charges ($b = -1.2$), while the lower figure represents the case of positive tidal charges ($b = 0.8$). The gray line connects the local maxima of the $\beta (r/R, b)$ profiles. The $\beta (r/R, b)$ profiles constructed for parameters belonging to different zones of the parameter space are separated by thick lines. In the regions of different zones the profiles are constructed for surface radii $R$ spaced by equal distances in the interval of maximal and minimal values of $R$ of ECS.[]{data-label="FIGbetaA"}](Fig6b.pdf "fig:"){width=".95\hsize"}
![Profiles of the local trapping factor $\beta (r/R, b)$ in the ECS with external spacetime of a naked-singularity type, constructed for fixed characteristic values of the brane parameter $b$ and sequences of the surface radius $R$ corresponding to the ECS. The upper figure corresponds to the case when two circular null geodesics of the external spacetime can exist ($b = 1.08$), while the lower figure corresponds only to the spacetimes where no circular null geodesic exists in the external spacetime ($b = 1.2$). The gray line connects the local maxima of the $\beta (r/R, b)$ profiles. The figures are constructed in the same way as those presented for the black-hole external spacetimes.[]{data-label="FIGbetaB"}](Fig7a.pdf "fig:"){width=".95\hsize"} ![Profiles of the local trapping factor $\beta (r/R, b)$ in the ECS with external spacetime of a naked-singularity type, constructed for fixed characteristic values of the brane parameter $b$ and sequences of the surface radius $R$ corresponding to the ECS. The upper figure corresponds to the case when two circular null geodesics of the external spacetime can exist ($b = 1.08$), while the lower figure corresponds only to the spacetimes where no circular null geodesic exists in the external spacetime ($b = 1.2$). The gray line connects the local maxima of the $\beta (r/R, b)$ profiles. The figures are constructed in the same way as those presented for the black-hole external spacetimes.[]{data-label="FIGbetaB"}](Fig7b.pdf "fig:"){width=".95\hsize"}
Let $N_\mathrm{p}$, $N_\mathrm{e}$ and $N_\mathrm{b}$ denote, respectively, the number of produced, escaped and trapped neutrinos per unit time of an external static observer at infinity. In order to determine the global correction factors $$\mathcal{E}(R, b)\equiv \frac{N_{\mathrm{e}}(R, b)}{N_{\mathrm{p}}(R, b)},\qquad
\mathcal{B}(R, b)\equiv \frac{N_{\mathrm{b}}(R, b)}{N_{\mathrm{p}}(R, b)} = 1 - \mathcal{E}(R, b),$$ it is necessary to introduce the local correction factor for trapped neutrinos at a given radius $r \in (r_{\mathrm{b(e)}}, R)$ (or $r \in (r_{\mathrm{b(i)}}, R)$, if relevant).
In the general case of non-isotropic emission of neutrinos, the escaped-to-produced ratio is given by the expression $$\Omega_{\mathrm e}(\Psi_{\mathrm e}) = \int\limits_0^{\Psi_{\mathrm e}}\int\limits_0^{2\pi}p(\Psi)\sin\Psi \mathrm{d}\Psi \mathrm{d}\phi$$ with $p(\Psi)$ being a directional function of the emission (scattering) process. Because of the assumption of isotropic emission of neutrinos in the frame of the static observers, the escaping solid angle $\Omega_\mathrm{e} (\Psi_\mathrm{e})$ determines fully the ratio of escaped-produced neutrinos and is given by $$\Omega_{\mathrm e}(\Psi_{\mathrm e}) = \int\limits_0^{\Psi_{\mathrm e}}\int\limits_0^{2\pi}\sin\Psi \mathrm{d}\Psi\mathrm{d} \phi
= 2 \pi (1-\cos\Psi_{\mathrm e}).$$ The escaping correction factor $$\epsilon(r, R, b) = \frac{\mathrm{d} N_{\mathrm{e}}(r, R, b)}{\mathrm{d} N_{\mathrm{p}}(r, R, b)}
= \frac{2\Omega(\psi_{\mathrm{e}}(r, R, b))}{4\pi} = 1 - \cos\psi_{\mathrm{e}}(r, R, b),$$ while the complementary factor for trapped neutrinos $$\beta(r, R, b) = 1 - \epsilon(r, R, b)
= \frac{\mathrm{d} N_{\mathrm{b}}(r, R, b)}{\mathrm{d} N_{\mathrm{p}}(r, R, b)} = \cos\psi_{\mathrm{e}}(r, R, B).$$ Notice that we consider production and escaping rates at a given radius $r$, but the radius $R$ of the compact object enters the relation as it determines the escaping directional angle. The coefficient $\beta(r, R, b)$ determines local efficiency of the neutrino trapping, i.e., the ratio of the trapped and produced neutrinos at any given radius $r \in(r_{\mathrm{b(e)}}, R)$ or $r \in(r_{\mathrm{b(i)}}, R)$. Its profile is shown for several representative values of the tidal charge $b$ and related sequences of the surface radius $R$ in Figures \[FIGbetaA\], \[FIGbetaB\]. The local maxima of the function $\beta(r, R, b)$ (with $R$ and $b$ being fixed) are given by the condition $\partial \beta / \partial r = 0$ which is satisfied at radius $r = r_{\mathrm{c(i)}}$ implying coincidence with the radius of the stable circular null geodesic, as anticipated intuitively. For the Zone II the maxima are naturally located at the surface radius. In Figures \[FIGbetaA\], \[FIGbetaB\], the maxima are depicted explicitly.
We can see that the local trapping factor in any given ECS spacetime reaches its maximal values at the radius corresponding to the stable circular null geodesic and decreases as the radius falls to the centre of the ECS. Nevertheless, the trapping region can extend down to the region near the centre only for the most compact ECS with external spacetime of the black-hole type (for both negative and positive tidal charges), having surface radius close the minimal allowed surface radius. The maximum of the $\beta(r, R, b)$ profile then reaches value of $\beta \sim 1$ close to the centre, where the stable null circular geodesic is located in such spacetimes. On the other hand, for positively tidally charged ECS with external spacetime of the naked-singularity type, the $\beta(r, R, b)$ profile cannot reach the central part and its maximum is much smaller than $\beta = 1$.
Neutrino production rates
-------------------------
Generally, the neutrino production is a very complex process depending on detailed structure of an extremely compact object. We can express the locally defined neutrino production rate in the form $$\mathcal{I}(r\{\mathcal{A}\}) = \frac{\mathrm{d} \mathcal{N}(r\{\mathcal{A}\})}{\mathrm{d} \tau(r)},$$ where d$\mathcal{N}$ is the number of interactions at radius $r$, $\tau$ is the proper time of the static observer at the given $r$, $\{\mathcal{A}\}$ is the full set of quantities relevant for the production rate. We can write that $$\mathrm{d} \mathcal{N}(r) = n(r)\Gamma(r)\mathrm{d} V(r),$$ where $n(r)$, $\Gamma(r)$ and $\mathrm{d} V(r)$ are the number density of particles entering the neutrino production processes, the neutrino production rate and the proper volume element at the radius $r$, respectively. Both $n(r)$ and $\Gamma(r)$ are given by detailed structure of the extremely compact objects, $\mathrm{d} V(r)$ is given by the spacetime geometry.
Here, considering the uniform energy density braneworld stars (for requirements of more realistic model see, e.g., [@Oest:2001:RAGtime2and3:; @Web:1999:Pul:]), we shall assume the local production rate to be proportional to the energy density, i.e., we assume uniform production rate as measured by the local static observers; of course, from the point of view of static observers at infinity, the production rate will not be distributed uniformly. (According to [@Gle:1992:PHYSR4:; @Gle:2000:CompactStars:; @Hae-Zdu:1986:NATURE:], such toy model could be reasonable good starting point for more realistic calculations.)
In the internal spacetime we can thus write the local neutrino production rate in the form $$\mathcal{I}(r) = \frac{\mathrm{d} \mathcal{N}}{\mathrm{d} \tau} \propto \rho = \mathrm{const}.$$ The local neutrino production rate related to the distant static observers is then given by the relation including the time-delay factor $$I = \frac{\mathrm{d} N}{\mathrm{d} t} = \mathcal{I} A(r, R, b).$$ The number of neutrinos produced at a given radius in a proper volume $\mathrm{d} V$ per unit time of a distant static observer (that are governed by the internal metric coefficients) is given by the relation $$\mathrm{d} N_{\mathrm{p}}(r, R, b) = I(r, R, b)\,\mathrm{d} V(r) = 4\pi\mathcal{I}\,A(r, R, b)\,B(r, R, b)\,r^{2}\,\mathrm{d} r.$$ Integrating through whole the compact object (from $0$ to $R$), we arrive to the global neutrino production rate in the form $$N_{\mathrm{p}}(R, b) = 4\pi\mathcal{I} \int_{0}^{R} A^{-}(r, R, b)B^{-}(r, R, b)r^2\,\mathrm{d} r.$$ In an analogical way, we can give the expressions for the global rates of escaping and trapping of the produced neutrinos: $$\begin{aligned}
N_{\mathrm{e}}(R,b) &=& 4\pi\mathcal{I}
\int_{r_{\mathrm {b(e)}}}^{R} (1-\cos\psi_{\mathrm{e}}(r, R, b))A^{-}(r, R, b)B^{-}(r, R, b) r^{2}
\,\mathrm{d} r + \nonumber \\
& & + N_{\mathrm{p}}(r_{\mathrm {b(e)}}),\\
N_{\mathrm{b}}(R,b) &=& 4\pi\mathcal{I} \int_{r_{\mathrm{b(e)}}}^{R} \cos\psi_{\mathrm{e}}(r, R, b)A^{-}(r, R, b)B^{-}(r, R, b) r^{2}\,\mathrm{d} r,
\end{aligned}$$ where $r_{\mathrm{b(e)}}$ is the radius given by equation (\[EQrbe\]) and $\cos\Psi_{\mathrm{e}}(r, R, b)$ is determined by equation (\[EQsicose\]). In ECS spacetimes with parameters belonging to the Zones where $r_{\mathrm{b(e)}}$ has not good meaning, its role takes $r_{\mathrm{b(i)}}$, while we replace $\cos\psi_{\mathrm{e}}$ for $\cos\psi_{\mathrm{i}}$. In such ECS spacetimes all bound neutrinos are the internal ones.
Efficiency of neutrino trapping {#SECeffnt}
===============================
In order to characterize the trapping of neutrinos in extremely compact stars, we introduce some coefficients giving the efficiency of the trapping effect in connection to the total neutrino luminosity and the cooling process in the period of the evolution of the star corresponding to the geodetical motion of neutrinos.
Trapping coefficient of total neutrino luminosity
-------------------------------------------------
The influence of the trapping effect on the total neutrino luminosity of ECS can be appropriately given by the coefficient $\mathcal B_L$ relating the number of neutrinos produced inside the whole compact star during unit time of distant observers and the number of those produced neutrinos that will be captured by the extremely strong gravitational field of the star. The total luminosity trapping coefficient is therefore given by the relation $$\mathcal{B}_\mathrm{L}(R,b) = \frac{\int_{r_{\mathrm{b(e)}}}^{R}A^{-}(r,R,b)B^{-}(r, R, b)\cos\psi_{\mathrm{e}}(r, R,b)
\,r^{2}\,\mathrm{d} r}{\int_{0}^{R} A^{-}(r,R,b) B^{-}(r,R,b) r^{2}\,\mathrm{d} r},$$ where $$A^{-}B^{-}=\frac{3 R Z \left[(R-b) Y + R (2 b - 3 R) Z\right]}{Y \left[2 R (2 b - 3 R) Z - b Y\right]},$$ and the complementary luminosity “escaping” coefficient is determined by the simple formula $$\mathcal{E}_\mathrm{L}(R, b) = 1 - \mathcal{B}_\mathrm{L}(R, b).$$
We can, moreover, define other global characteristic coefficients. For the “internal” neutrinos with motion restricted to the interior of the star, we introduce a coefficient $$\mathcal{Q}_\mathrm{L}(R, b) = \frac{\int_{r_{\mathrm{b(i)}}}^{R}A^{-}(r, R, b)B^{-}(r, R, b)\cos\psi_{\mathrm{i}}(r, R, b)\,r^{2}\,\mathrm{d} r}{\int_{0}^{R} A^{-}(r,R,b) B^{-}(r,R,b) r^{2}\,\mathrm{d} r}$$ and for the “external” neutrinos, we can use a complementary coefficient $$\mathcal{X}_\mathrm{L} = \frac{N_{\mathrm{ext}}}{N_{\mathrm{p}}} = \mathcal{B}_\mathrm{L} - \mathcal{Q}_\mathrm{L}.$$
![The total luminosity trapping coefficient $\mathcal{B}_\mathrm{L}(R, b)$ (full lines) and the internal luminosity trapping coefficient $\mathcal{Q}_\mathrm{L}(R, b)$ (dashed lines) as functions of the surface radius $R$ for fixed characteristic values of the tidal charge $b$. The left plot is given for negative tidal charges $-1.2 \leq b\leq 0$, the right plot for positive tidal charges $0\leq b < 27/20$. The values of $b$ are equally spaced with interval of $0.2$ for $b < 1$ and $0.7$ for $b > 1$.[]{data-label="FIGBQa"}](Fig8a.pdf "fig:"){width=".49\hsize"}![The total luminosity trapping coefficient $\mathcal{B}_\mathrm{L}(R, b)$ (full lines) and the internal luminosity trapping coefficient $\mathcal{Q}_\mathrm{L}(R, b)$ (dashed lines) as functions of the surface radius $R$ for fixed characteristic values of the tidal charge $b$. The left plot is given for negative tidal charges $-1.2 \leq b\leq 0$, the right plot for positive tidal charges $0\leq b < 27/20$. The values of $b$ are equally spaced with interval of $0.2$ for $b < 1$ and $0.7$ for $b > 1$.[]{data-label="FIGBQa"}](Fig8b.pdf "fig:"){width=".49\hsize"}
The results are illustrated for the coefficients $\mathcal{B}_\mathrm{L}(R, b)$, $\mathcal{Q}_\mathrm{L}(R, b)$ in Figure \[FIGBQa\]. The behaviour of the total luminosity trapping factor is qualitatively the same for ECS with external spacetimes of the black-hole type for both negative and positive tidal charges, but it has qualitatively different character for the black-hole and naked-singularity ECS spacetimes. In the black-hole-type ECS it monotonously increases with decreasing surface radius for any fixed tidal charge (both negative and positive), while in the naked-singularity-type ECS spacetimes it reaches a maximum between the centre and the surface radius $R$, and then it decreases to zero value for surface radius approaching the minimal value of $R$.
For the black-hole-type ECS with negative tidal charges the maximum of the coefficient $\mathcal{B}_\mathrm{L}(R, b)$ slightly decreases with decreasing $b$, while for positively tidally charged ECS it increases with increasing $b$ approaching $\mathcal{B}_\mathrm{L}(R, b) = 1$ for $b \rightarrow 1$. We observe a similar behavior also for $\mathcal{Q}_\mathrm{L}(R, b)$, but the magnitude of this coefficient is smaller in comparison with $\mathcal{B}_\mathrm{L}(R, b)$. In the naked-singularity-type ECS (with positive tidal charges) the local maximum of $\mathcal{B}_\mathrm{L}(R, b)$ (and $\mathcal{Q}_\mathrm{L}(R, b)$) strongly decreases with $b$ increasing; of course, for ECS in the Zone III of the space of parameters, the parameter $\mathcal{Q}_\mathrm{L}(R, b)$ is not defined.
Trapping coefficient of neutrino cooling process
------------------------------------------------
The efficiency of the influence of neutrino trapping on the cooling process is most effectively described by the local coefficient of trapping $b_\mathrm{c}$ relating the trapped and produced neutrinos at a given radius of the star that is defined by the relation $$b_\mathrm{c} (r, R, b) \equiv \beta (r, R, b).$$ The local cooling coefficient is therefore given in Figures \[FIGbetaA\], \[FIGbetaB\] for appropriately chosen tidal charges $b$ and related sequences of the surface radius $R$. All the properties of the local trapping coefficient discussed above are thus relevant for the local cooling phenomena.
![The global cooling coefficient related to trapping of neutrinos in the active zone of trapping $\mathcal{B}_\mathrm{C}(R, b)$. For tidal charge $b$ fixed for values with the same interval as in the previous figure and the full range of $R$ corresponding to the ECS is scanned.[]{data-label="FIGBC"}](Fig9.pdf){width=".7\hsize"}
Further, the cooling process can be appropriately described in a complementary manner by a global coefficient for trapping, restricted to the “active” zone, where the trapping of neutrinos occurs. The global cooling coefficient is thus defined by the relation $$\mathcal{B}_\mathrm{c}(R, b) \equiv \frac{\int_{r_\mathrm{b(e, i)}}^R A^{-}(r, R, b)B^{-}(r, R, b)\cos\psi_{\mathrm{e, i}}(r, R, b)\mathrm{d} r}
{\int_{r_\mathrm{b(e, i)}}^R A^{-}(r, R, b)B^{-}(r, R, b) r^{2}\mathrm{d} r}.$$ The global “cooling” coefficient $\mathcal{B}_\mathrm{c} (R, b)$ is presented in Figure \[FIGBC\] — for few characteristic values of the tidal charge $b$, both negative and positive, the full range of the ESC surface radius is scanned.
The behaviour of the global cooling factor $\mathcal{B}_\mathrm{c}(R, b)$ is of the same character as for the total luminosity factor $\mathcal{B}_\mathrm{L}(R, b)$ in both black-hole and naked-singularity-type ECS and for both negative and positive tidal charges. Its magnitude is nearly twice higher than those of the total luminosity factor for the surface radii $R$ close to their maximal value, but it is nearly equal to the total luminosity factor as $R$ approaches its minimal value.
Conclusions {#SECconcl}
===========
We have demonstrated that the braneworld ECS can be separated into five classes corresponding to five Zones of the parameter space $(b-R)$ where trapping of neutrinos and their motion in the external field of the ECS are realized in qualitatively different way. There are fundamental differences of the trapping effect in the field of ECS with negative and positive tidal charges. The differences are related to both the surface radius of the ECS and the trapping coeffients of both kind — for the total luminosity and the cooling process. Moreover, there is a strong difference of the trapping efficiency in the ECS with positive tidal charge when related to the black-hole spacetimes and the naked-singularity spacetimes. In the first case, the trapping is very efficient and the global trapping coefficient can approach the value of $1$ for the spacetimes with $b \rightarrow 1$, while in the second case the coefficients strongly decrease for tidal charges growing above the critical value of $b = 1$.
There is an astrophysically very important effect related to the surface radius of the ECS with the negative and positive tidal charges. For $b < 0$, the surface radius of ECS can significantly overcome the limiting value of $R = 3$ valid for $b = 0$. On the other hand, for $b > 0$, the surface radius of ECS decreases under the standard value of $R = 3$ — for ECS with $b \sim 1$, there is $R_\mathrm{max} \sim 2.3$ that is substantially lower than the values indicated by observations, and for $b > 1$ the ECS surface radius decreases further.
In the ECS spacetimes with negative charge there is a crucial effect related to the increase of the surface radius of the ECS allowed for decreasing $b < 0$. For tidal charges slightly under the value of $b \sim -1$, the ECS surface radius approaches $R = 4$ strongly overcoming the limit of $R = 3$ corresponding to the standard Schwarzschild spacetimes and enters the region of values commonly accepted for observed neutron stars that can be lowered down to $R = 3.5$ [@Lat-Pra:2007:PhysRep:; @Cot-Pae-Men:2002:NATURE:]. In such cases the coefficient of total luminosity can be of quite significant value for astrophysically realistic radii of ECS. For example, in the case of ECS with $R = 3.5$ and $b = -1.2$, there is $\mathcal{B}_\mathrm{L}(R, b) \sim 0.023$. The global cooling coefficient can approach even the value of $\mathcal{B}_\mathrm{c}(R, b) \sim 0.077$ under such conditions, indicating thus very strong and significant effects of trapping for braneworld negatively charged ECS, significantly overcoming the related values corresponding to the standard (tidally non-charged) spacetimes [@Stu-Tor-Hle-Urb:2009:]. Because the effect of trapping of neutrinos is cumulative, one can expect its relevance in realistic models of ECS to be strongly enhanced by negative tidal charges. Note that the conditions on the tidal charge magnitude implied by the high-frequency quasiperiodic oscillations observed in some low-mass-X-ray- binaries containing neutron or quark stars put the limit $b<1.2$ (assuming their mass $m \sim 2 M_{\odot}$) [@Kot-Stu-Tor:2008:CLASQG:]. This limit allows for sufficient increasing of the ECS surface radius with negative tidal charges enabling strong effects of neutrino trapping in the interior of such ECS with observationally acceptable radius. The expected lowering of the neutrino luminosity in observed neutron stars can be about $\sim 10\,\%$ and could be observationally tested.
We can conclude that the negative tidal charges cause slight decrease of the efficiency of the trapping phenomena in the field of ECS, but they significantly shift these phenomena to the region of astrophysically relevant situations due to the strong increase of the surface radius of such ECS, shifting the radius to values acceptable observationally. On the other hand, the positive tidal charges generally enhance the trapping effects for the ECS spacetimes of the black-hole type, but the surface radius of such ECS is strongly shifted to the regions excluded observationally.
The present work was supported by the Czech grants MSM 4781305903, LC 06014, GAČR 205/09/H033 and the internal grant SGS/2/2010. One of the authors (Z.S.) would like to express his gratitude to the Czech Committee for collaboration with CERN and the Theory Division of CERN for perfect hospitality.
[99]{} Abdujabbarov, A., Ahmedov, B.: Test particle motion around a black hole in a braneworld. Phys. Rev. D **81**, 044022 (2010)
Abramowicz, M.A., Anderson, N., Bruni, M., Ghosh, P., Sonego, S.: Gravitational waves from ultracompact stars: the optical geometry view of trapped modes. Class. Quantum Grav. **14**, L189–L194 (1997)
Abramowicz, M.A., Miller, J.C., Stuchl[í]{}k, Z.: Concept of radius of gyration in general relativity. Phys. Rev. D **47** 1440–1447 (1993)
Abramowicz, M.A., Prasanna, A.R.: Centrifugal Force Reversal Near a Schwarzschild Black-Hole. Mon. Not. R. Astron. Soc. **245**, 720 (1990)
Aliev, A.N., G[ü]{}mr[ü]{}k[ç]{}[ü]{}o[ğ]{}lu, A.E.: Charged rotating black holes on a 3-brane. Phys. Rev. D **71**, 104027 (2005)
Aliev, A.N., Talazan, P.: Gravitational effects of rotating braneworld black holes. Phys. Rev. D **80**, 044023 (2009)
Arkani-Hamed, N., Dimopoulos, S., Dvali, G.: The hierarchy problem and new dimensions at a millimeter. Phys. Lett. B **429**, 263–272 (1998)
Bahcall, S., Lynn, B.W., Selipsky, S. B.: New models for neutron stars. Astrophys. J. **362**, 251–255 (1990)
B[ö]{}hmer, C.G., De Risi, G., Harko, T., Lobo, F.S.N.: Classical tests of general relativity in brane world models. Class. Quantum Grav. **27**, 185013 (2010)
B[ö]{}hmer, C.G., Harko, T., Lobo, F.S.N.: Solar system tests of brane world models. Class. Quantum Grav. **25**, 045015 (2008)
Bin-Nun, A.Y.: Relativistic images in Randall-Sundrum II braneworld lensing. Phys. Rev. D **81**, 123011 (2010)
Bin-Nun, A.Y.: Gravitational lensing of stars orbiting Sgr A\* as a probe of the black hole metric in the Galactic center. Phys. Rev. D **82**, 064009 (2010)
Cottam, J., Paerls, F., Mendez, M.: Gravitationally redshifted absorption lines in the X-ray burst spectra of a neutron star. Nature **420**, 51–54 (2002)
Dadhich, N., Maartens, R., Papadopoulos, P., Rezania, V.: Black holes on the brane. Phys. Lett. B **487**, 1–6 (2000)
Dimopoulos, S., Landsberg, G.: Black Holes at the Large Hadron Collider. Phys. Rev. Lett. **87**, 161602 (2001)
Germani, C., Maartens, R.: Stars in the braneworld. Phys. Rev. D **64**, 124010 (2001)
Glendenning, N.K.: First-order phase transitions with more than one conserved charge: Consequences for neutron stars. Phys. Rev. D **46**, 1274–1287 (1992)
Glendenning, N.K.: Compact Stars: Nuclear Physics, Particle Physics, and General Relativity. Springer, New York (2000)
Haensel, P., Zdunik, J.L.: A submillisecond pulsar and the equation of state of dense matter. Nature **340**, 617–619 (1986)
Hledík, S., Stuchlík, Z., Mrázová, K.: Comparison of general relativistic polytropic and adiabatic fluid spheres with a repulsive cosmological constant. In: Hledík, S., Stuchlík, Z. (eds.) Proceedings of RAGtime 4/5: Workshops on black holes and neutron stars, Opava, 14–16/13–15 October 2002/03, pp. 75–89. Silesian University in Opava, Opava (2004)
Kotrlov[á]{}, A., Stuchl[í]{}k, Z., T[ö]{}r[ö]{}k, G.: Quasiperiodic oscillations in a strong gravitational field around neutron stars testing braneworld models. Class. Quantum Grav. **25**, 225016 (2008)
Lattimer, J., Prakash, M.: The physics of neutron stars. Science **304**, 536–542 (2004)
Lattimer, J., Prakash, M.: Neutron star observations: Prognosis on Equation of state constraints. Phys. Rep. **442**, 109–165 (2007)
Maartens, R.: Brane-World Gravity. Living Rev. Relativity **7**, 7 (2004)
Mallick, R., Bhattacharyya, A., Ghosh, S.K., Raha, S.: General Relativistic effect on the energy deposition rate for neutrino pair annihilation above the equatorial plane along the symmetry axis near a rotating neutron star. ArXiv e-prints 0905.3605v2 \[astro-ph.HE\] (2009)
Mamadjanov, A.I., Hakimov, A.A., Tojiev, S.R.: Quantum Interference Effects in Spacetime of Slowly Rotating Compact Objects in Braneworld. Mod. Phys. Lett. A **25**, 243–256 (2010)
Miller, J., Shahbaz, T., Nolan, L. A.: Are Q-stars a serious threat for stellar-mass black hole candidates? Mon. Not. R. Astron. Soc. **294**, L25–L29 (1998)
Misner, C.W., Thorne, K.S., Wheeler, J.A.: Gravitation. San Francisco, W. H. Freeman (1973)
Morozova, V.S., Ahmedov, B.J., Abdujabbarov, A.A., Mamadjanov, A.I.: Plasma magnetosphere of rotating magnetized neutron star in the braneworld. Astrophys. & Space Sci. **330**, 257–266 (2010)
Nilsson, U.S., Ugla, C.: General Relativistic Stars: Polytropic Equations of State. Annals of Physics **286**, 292–319 (2000)
stgaard, E.: Internal structure of neutron stars. In: Hled[í]{}k, S., Stuchl[í]{}k, Z. (eds.) Proceedings of RAGtime 2/3: Workshops on black holes and neutron stars, Opava, 11–13/8–10 October 2000/01, pp. 73–102. Silesian University in Opava, Opava (2001)
Randall, L., Sundrum, R.: An Alternative to Compactification. Phys. Rev. Lett. **83**, 4690–4693 (1999)
Schee, J., Stuchl[í]{}k, Z.: Profiles of emission lines generated by rings orbiting braneworld Kerr black holes. Gen. Relativ. Gravit. **41**, 1795–1818 (2009)
Schee, J., Stuchl[í]{}k, Z.: Optical Phenomena in the Field of Braneworld Kerr Black Holes. Int. J. Mod. Phys. D **18**, 983–1024 (2009)
Schwarzschild, K.: Über das Gravitationsfeld einer Kugel aus inkompressibler Flüssigkeit nach der Einsteinschen Theorie. Sitzungsber. K. Preuss. Akad. Wiss., Phys.–Math. Kl. 424–434 (1916)
Shapiro, S.L., Teukolsky, S.A.: Black Holes, White Dwarfs and Neutron Stars: The Physics of Compact Objects. Wiley–VCH, New York (1983)
Shiromizu, T., Maeda, K.-I., Sasaki, M.: The Einstein equations on the 3-brane world. Phys. Rev. D **62**, 024012 (2000)
Stuchl[í]{}k, Z.: Note on the properties of the Schwarzschild-de-Sitter spacetime. Bull. Astronom. Inst. Czechoslovakia **41**, 341–343 (1990)
Stuchlík, Z.: Spherically symmetric static configurations of uniform density in spacetimes with a non-zero cosmological constant. Acta Phys. Slovaca, **50**, 219–228 (2000)
Stuchl[í]{}k, Z., Hled[í]{}k, S., Jur[á]{}ň, J.: Optical reference geometry of Kerr-Newman spacetimes. Class. Quantum Grav. **17**, 2691–2718 (2000)
Stuchlík, Z., Hledík, S., Šoltés, J., [Ø]{}stgaard, E.: Null geodesics and embedding diagrams of the interior Schwarzschild–de Sitter spacetimes with uniform density. Phys. Rev. D **64**, 044004 (2001)
Stuchl[í]{}k, Z., Kotrlov[á]{}, A.: Orbital resonances in discs around braneworld Kerr black holes. Gen. Relativ. Gravit. **41**, 1305–1343 (2009)
Stuchl[í]{}k, Z., Schee, J.: Appearance of Keplerian discs orbiting Kerr superspinars. Class. Quantum Grav. **27**, 215017 (2010)
Stuchl[í]{}k, Z., T[ö]{}r[ö]{}k, G., Hled[í]{}k, S., Urbanec, M.: Neutrino trapping in extremely compact objects: I. Efficiency of trapping in the internal Schwarzschild spacetimes. Class. Quantum Grav. **26**, 035003 (2009)
Weber, F.: Pulsars as Astrophysical Laboratories for Nuclear and Particle Physics. Taylor & Francis, London (1999)
Weber, F., Glendenning, N.K.: Application of the improved Hartle method for the construction of general relativistic rotating neutron star models. Astrophys. J. **390**, 541 (1992)
[^1]: In principle, the bound null geodesics could exist also in objects having $R > 3GM/c^2$, e.g., in some composite polytropic spheres [@Nil-Cla:2000:GRRelStarsPolyEOS:].
[^2]: The internal Schwarzschild spacetime can well represent the spacetime properties of realistic ECS [@Gle:1992:PHYSR4:; @Gle:2000:CompactStars:; @Hae-Zdu:1986:NATURE:] which are crucial in estimating the role of neutrino trapping. Such a simple spacetime could serve quite well as a test bed for realistic models of ECS.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We address the problem of estimating time and frequency shifts of a known waveform in the presence of multiple measurement vectors (MMVs). This problem naturally arises in radar imaging and wireless communications. Specifically, a signal ensemble is observed, where each signal of the ensemble is formed by a superposition of a small number of scaled, time-delayed, and frequency shifted versions of a known waveform sharing the same continuous-valued time and frequency components. The goal is to recover the continuous-valued time-frequency pairs from a small number of observations. In this work, we propose a semidefinite programming which exactly recovers $s$ pairs of time-frequency shifts from $L$ regularly spaced samples per measurement vector under a minimum separation condition between the time-frequency shifts. Moreover, we prove that the number $s$ of time-frequency shifts scales linearly with the number $L$ of samples up to a log-factor. Extensive numerical results are also provided to validate the effectiveness of the proposed method over the single measurement vectors (SMVs) problem. In particular, we find that our approach leads to a relaxed minimum separation condition and reduced number of required samples.'
author:
- 'Maral Safari, Sajad Daei, Farzan Haddadi [^1]'
bibliography:
- 'mypaperbibe1.bib'
title: 'Off-the-grid Recovery of Time and Frequency Shifts with Multiple Measurement Vectors'
---
Atomic norm minimization, super-resolution, multiple measurement vectors.
Introduction
============
er the past few years, there has been a growing interest in using super-resolution, a tool for recovering the high-resolution information from low-pass data. This technique is shown to be useful in many applications such as radar imaging [@handbook2008mi], astronomy [@puschmann2005super], communication systems [@luo2006low], geophysics [@khaidukov2004diffraction], microscopy [@mccutchen1967superresolution] and also in the direction of arrival (DOA) estimation [@sayed2005network], [@stoica2005spectral] in which the aim is to estimate the directions of narrow-band sources by an array of sensors.
In this work, we study the problem of using an antenna array to estimate the time delays and Doppler (frequency) shifts of a known waveform. This problem naturally occurs in active radar imaging [@activradarimaging; @heckel2016super] and multi-path channel identification in wireless communications [@heckel2016super]. More precisely, in these applications, a known waveform $x(t)$ is transmitted and reflections from moving sources are received at the $R$-element antenna array. Writing in mathematical terms, we observe a signal ensemble $$\begin{aligned}
\label{eq.rel1}
y_m(t)=\sum_{j=1}^sb_{jm}x(t-\overline{\tau}_j){\rm e}^{i2\pi \overline{\nu}_j t},~ m=1,..., R,
\end{aligned}$$ at the array where $R$ is the number of array elements, $b_{jm}$ is the attenuation factor corresponding to the time-Doppler shifts $(\overline{\tau}_j, \overline{\nu}_j)$, and $s$ is the number of moving sources. In active radar imaging, estimating delay and Doppler shifts provides valuable information about the location and relative velocity of the targets in the scene. Besides, in wireless communications[@heckel2016super], model represents a scenario where a mobile user is rapidly moving and sends a known training sequence to a base station (BS) for channel estimation and equalization purposes. In case that the communication channel is frequency selective, the signal arrives at the BS with multiple different delays and Doppler shifts. Estimating the delays and Doppler shifts is necessary for BS in order to remove the inter-symbol interference.
By taking regularly spaced samples of $y(t)$ in model (see Section for a detailed description), we have a measurement vector at each antenna composed of $L$ samples. Considering a $R$-element antenna array, we encounter multiple measurement vectors (MMVs) by assuming that the delay-Doppler pairs remain fixed at the output of array. It is worth mentioning that the aforementioned method for accessing MMVs is different from what has been considered in the literature (see for example [@atomicmmv; @yang2016exact]), since, there, MMVs refer to multiple snapshots in the time domain. However, here we assume $R$ measurements are observed via an $R$-element antenna array. It is also possible to have multiple measurements in time (alternatively meant to be multiple snapshots) by choosing the probing signal $x(t)$ to be periodic[^2]. There are, however, a few constraints imposed by practical scenarios: The probing signal $x(t)$ has finite band-width $B$, the received signals at the array are only observed during a finite interval of length $T$, the delay-Doppler pairs has finite support, i.e. $(\overline{\tau}_j, \overline{\nu}_j)\in [-\tfrac{T}{2},\tfrac{T}{2}]\times [-\tfrac{B}{2},\tfrac{B}{2}]$. By the latter assumptions, and since the effective support of the probing signal in the time and frequency domain must be greater than the delay-Doppler shifts, $x(t)$ and $y(t)$ must be both approximately time- and band-limited. Hence, the natural resolution limit, i.e. the accuracy up to which $(\overline{\tau}_j, \overline{\nu}_j)$ can be uniquely resolved, is $\tfrac{1}{B}$ and $\tfrac{1}{T}$ in the delay and Doppler directions, respectively. This resolution limit can be achieved by using a standard digital matched filter in order to identify the delay-Doppler pairs. In this paper, we show that this resolution limit can be broken by assuming that the delay-Doppler pair $(\overline{\tau}_j, \overline{\nu}_j)$ can take any continuous values in $[-\tfrac{T}{2},\tfrac{T}{2}]\times [-\tfrac{B}{2},\tfrac{B}{2}]$ and is not constrained to be on a predefined domain of grids which is the case in the well-known theory of compressed sensing (CS) [@candes2006robust]. Specifically, using the on-grid assumption in CS, $\ell_{1,2}$ minimization can be applied to recover the unknowns $(\overline{\tau}_j, \overline{\nu}_j)$ from MMVs. However, most CS-based methods[^3] including $\ell_{1,2}$ minimization needs incoherence property which does not generally hold when the grids are fine and hence we encounter an unavoidable basis mismatch between on-the-grid and true delay-Doppler pairs. Our goal in this paper is to estimate the continuous delay-Doppler pairs $(\overline{\tau}_j, \overline{\nu}_j), j=1,..., s$ from these MMVs. To achieve this goal, we propose general atomic norm problems (inspired by [@chandrasekaran2012convex]) for two-dimensional (2D) super-resolution in the noise-free and noisy cases equipped with MMVs. To the best of the authors’ knowledge, super-resolving 2D continuous parameters from MMVs using the concepts of atomic norm minimization has not been addressed before and indeed most of the prior works can be seen as a special case of it. Further, our proposed problems can be viewed as a continuous counterpart of $\ell_{1,2}$ minimization in CS. However, unlike $\ell_{1,2}$ which is designed for recovering one on-grid parameter, our framework, instead, is seeking to recover two continuous off-grid unknowns (i.e. delay-Doppler pairs). We show that our proposed atomic norm problems can be efficiently solved using semidefinite programming and the continuous delay-Doppler shifts[^4] can be exactly recovered. Moreover, we theoretically prove that if we take $L$ noise-free samples per measurement vector, up to $\mathcal{O}(\tfrac{L}{\log(L^6 T)})$ delay-Doppler pairs can be exactly recovered with high probability using our proposed method provided that a certain minimum separation condition between the time-frequency shifts is satisfied. Numerical results also demonstrate that our proposed approach leads to improved (relaxed) minimum separation condition compared to the case of single measurement vector (SMV) problem proposed in [@heckel2016super] in both noise-free and noisy cases. Besides, we show that under a fixed minimum separation between the delay-Doppler pairs, the number of required samples for successful and robust recovery decreases.
Related Works and Key Differences
---------------------------------
Conventional subspace-based methods such as MUSIC and ESPRIT [@schmidt1982signal; @ottersten1991performance], assume that the signal amplitudes $b_{jm}$ are uncorrelated and the covariance matrix corresponding to samples of each array element is low-rank. The performance of these approaches are prone to be corrupted against noise and correlations between sources $b_{jm}$.
The theory of super-resolution using convex optimization is first initiated by Candes et al. in [@candes2014towards]. They propose a problem for recovering off-grid time-domain spikes from low-pass Fourier measurements in the SMV case. They show that the continuous spikes in the time domain can be exactly recovered by solving semidefinite programming provided that a minimum separation between sources is satisfied.
Tang et al. in [@tang2013compressed] study super-resolution problem in the framework of CS. They propose an atomic norm minimization where the frequency spikes of a signal are recovered from its partial time-domain samples. Their work can be regarded as the continuous counterpart of $\ell_{1}$ minimization in CS. The difference with [@candes2014towards] lies in the fact that they consider only partial random observations rather than full observations in [@candes2014towards].
Hyder et al. propose a non-convex algorithm based on smoothed $\ell_{0,2}$[^5] to extract the joint (common) sparsity pattern of 1D signals from MMVs. They investigate both narrow-band and broad-band DOA estimation and show that using MMVs allows a larger sensor spacing than the smallest half-wavelength of the signal in case of broad-band signal. Unlike the subspace-based methods, this method does not require the number of 1D sources in advance. However, the DOAs are assumed on-the-grid which somewhat limits the applicability of the method. Moreover, an issue associated with this non-convex approach is that its stability against measurement noise is not assured. Yang et al. in [@yang2016exact] study recovering signals which share the same 1D frequency parameter from MMVs. They propose an atomic norm framework to solve this problem and study the advantage of using MMVs over SMV. They show that the availability of MMVs results in relaxed minimum separation condition and reduced number of required measurements. The observed signal is scaled samples of the sum of $s$ sinusoids and coincides with model when the delays are ignored or alternatively when we have only frequency shifts. However, most of the proof techniques in [@yang2016exact] can not be applied to model . In particular, our atomic framework and choice of atoms (which turns to be a multiplication of two Dirichlet kernel) are completely different from that in [@yang2016exact], since model deals with two unknown parameters in two different domains (i.e. time and frequency shifts). In fact, 1D methods (e.g. in [@candes2014towards; @ottersten1991performance]) can not be used for proofs in 2D case.
Li et al. in [@atomicmmv] investigate the benefit of incorporating MMVs into off-the-grid frequency estimation and denoising. They develop an atomic norm minimization that aims to estimate and denoise spectrally sparse signals from partial noisy MMVs. They provide numerical experiments showing the performance gain of their method over SMV when the number of measurement vectors increases. They obtain a performance guaranty to denoise MMV signals which share common 1D frequency components. Their analysis and proofs extend the results of [@tang2013compressed] for MMV signals for 1D frequency estimation. Again, their proof approach fails to apply for estimating 2D time-frequency shifts.
Heckel et al. in [@heckel2016super] tackle the problem of identifying time-frequency shifts in radar scenario. The observed signal at receiver is a scaled superposition of time and frequency shifted versions of a known waveform. Thus, their model can be regarded as a special case of our model where only one element is used at the array. Specifically, an atomic norm approach is provided to recover the continuous delay-Doppler pairs using SMV. As opposed to what is done in [@heckel2016super], we benefit from the common atomic sparsity pattern of MMVs at the outputs of the sensor array. Our work can be viewed as an extension of the SMV work [@heckel2016super] to the MMV case. However, this nontrivial generalization comes with major mathematical differences, of which we can mention the uniqueness proof of our proposed atomic problem that deals with vector-valued dual polynomials which is much more challenging than the scalar-valued polynomial used in [@heckel2016super].
There also exist other works that might be somehow relevant to our work such as [@heckel2016mimo] which deals with recovering three continuous parameters in multiple input multiple output (MIMO) scenario. However, besides the fact that their model is different from ours, their work does not use the availability of MMVs.
Outline and Notations
---------------------
Throughout the paper, scalars are denoted by lowercase letters, vectors by lowercase boldface letters, and matrices by uppercase boldface letters. The $k$th element of a vector $\bm{x}$ is denoted by $x(k)$. The absolute value of a scalar, the element-wise absolute value of a vector and the cardinality of a set are shown by $|\cdot|$. The infinity norm is $\|\bm{z}\|_{\infty}=\underset{k}{\max}~|z_{k}|$. In addition, $\|\cdot\|_{1}$, $\|\cdot\|_{2}$ and $\|\cdot\|_F$ are reserved for $\ell_{1}$, $\ell_{2}$ and Frobenius norms, respectively. We define $\|\bm{A}\|:=\max_{\|\bm{v}\|_2=1}\|\bm{Av}\|_2$ and $\|\bm{A}\|_{2,\infty}$ := $\underset{j}{\max} \|\bm{a}_j\|_{2}$, where $\bm{a}_j$ denotes the j-th row of a matrix $\bm{A} $. The operator $\langle\cdot,\cdot \rangle_{\mathbb{R}}$ stands for the real part of the inner product of two vectors. We use a 2D index for vectors or matrices. Indeed, by $[{\bm{z}}]_{(k,l)}, k,l=-N,...,N$, we mean that ${\bm{z}}=[z_{(-N,-N)},z_{(-N,-N+1)},...,z_{(-N,N)},z_{(-N+1,-N)},...,z_{(N,N)}]$. The operators $\rm{tr}(\cdot)$ and $(\cdot)^{\rm H}$ represent the trace and Hermitian of a matrix, respectively. $x^{*}$ is the conjugate of $x$. To show that $\bm{A}$ is a positive semidefinite matrix, we write $\bm{A}\succeq 0$. $\mathds{E}[\cdot]$ and $\mathds{P}[\cdot]$ denote the expectation and probability of an event, respectively. $ {{\mathbb{S}}^{R-1}}=\{\bm{\varphi} \in \mathbb{C}^{R \times 1}:{{\left\| \bm{\varphi} \right\|}_{2}}=1\} $ denote the unit complex or real sphere. Finally, we use numerical constants $c,\tilde{c},c',c_1,c_2, ...$ which take on different values at different places.
System Model and Recovery via Convex Optimization {#section.systemmodel}
=================================================
As we assumed earlier, $(\bar\tau_j,\bar{\nu}_j)\in[\tfrac{-T}{2},\tfrac{T}{2}]\times[\tfrac{B}{2},\tfrac{B}{2}]$. Based on $2BT$-Theorem [@slepian1976bandwidth],[@durisi2012sensitivity], we can take samples of $y_m(t)$ in the interval $[\tfrac{-T}{2},\tfrac{T}{2}]$ at rate $\tfrac{1}{B}$. So, we totally have $L:=BT$ samples[^6] of the form (see Appendix \[proof.equivalence\] for a detailed derivation): $$\begin{aligned}
\label{eq.sampled_y}
&{{y}_{p{m}}}=\sum\limits_{j=1}^{s}{{{b}_{j{m}}}}\sum\limits_{k,l=-N}^{N}{{{{D}}_{_{N}}}}(\tfrac{l}{L}-{{\tau }_{j}}){D}_{N}(\tfrac{k}{L}-{{\nu }_{j}}){{x}_{p-l}}{{e}^{i2\pi (\tfrac{kp}{L})}} \text{ }\nonumber\\
&p=-N,...,N \quad L=2N+1,\quad m=1,...,R, \end{aligned}$$ where $$\begin{aligned}
\label{eq.diric}
{{{D}}_{N}}(t):=\tfrac{1}{L}\sum\limits_{k=-N}^{N}{{{\rm e}^{i2\pi tk}}}\end{aligned}$$ is the Dirichlet kernel. $\tau_j:=\tfrac{\overline{\tau}_j}{T}$ and $\nu_j:=\tfrac{\overline{\nu}_j}{B}$ are the normalized time-frequency shifts, respectively. $x_l$ is the $l$-th sample of the probing signal $x(t)$ and is assumed to be $L$-periodic. It is easy to verify that $(\tau_j,\nu_j)\in[-\tfrac{1}{2},\tfrac{1}{2}]^2$. Due to the periodicity property, without loss of generality, we assume that $(\tau_j,\nu_j)\in[0,1]^2$. Define atoms $\bm{a}\in\mathbb{C}^{L^2\times 1}$ with elements $$\begin{aligned}
\label{eq.a_r}
&{{[\bm{a}(\bm{r})]}_{(k,l)}}={{{D}}_{N}}(\tfrac{l}{L}-\tau ){{{D}}_{N}}(\tfrac{k}{L}-\nu ),\quad \bm{{r} }=[\tau,\nu]^{\top},\\
&k,l=-N,...,N. \nonumber\end{aligned}$$ By using the definition of Dirichlet kernel in , the atoms $\bm{a}\in\mathbb{C}^{L^2\times 1}$ can also be reformulated as $$\begin{aligned}
\label{eq.ar}
\bm{a}(\bm{r})=\bm{F}^{\rm H}\bm{f}(\bm{r}),\end{aligned}$$ where $[\bm{f}(\bm{r})]_{(m,n)}:={\rm e}^{-i2\pi (m\tau+n\nu)}$, and $\bm{F}^{\rm H}$ is the inverse 2D discrete Fourier transform whose entries are given by $$\begin{aligned}
[\bm{F}^{\rm H}]_{(k,l),(m,n)}:=\tfrac{1}{L^2}{\rm e}^{i2\pi(\tfrac{m k+n l}{L})}.\end{aligned}$$ The relation can be reformulated in matrix form as $$\begin{aligned}
\label{eq.input_output_rel}
\bm {Y}=\bm{G}\bm{X}\in\mathbb{C}^{L\times R},\end{aligned}$$ where $$\begin{aligned}
\label{signal}
&\bm{X}=\sum\limits_{j=1}^{s}{\bm{a}({{\bm{r}}_{j}}\text{)}{{\bm{b}}_{j}^{\rm H}}=\sum\limits_{j=1}^{s}{{{c}_{j}}}}\bm{a(}{{\bm{r}}_{j}}\text{,}{{\bm{\varphi} }_{j}})=\sum\limits_{j=1}^{s}{{{c}_{j}}}\bm{a(}{{\bm{r}}_{j}})\bm{\varphi}_{j}^{\rm H},\nonumber\\
& \bm{r} _{j}:=[\tau_{j},\nu_j]^{\top},\end{aligned}$$ ${{\bm{b}}_{j}^{\rm H}}=[{{b}_{j1}},...,{{b}_{jR}}]\in {\mathbb{C}}^{1 \times R}$ is the attenuation vector, ${{c}_{j}}={{\left\| {\bm{b}_{j}} \right\|}_{2}}>0$, and ${{\bm{\varphi}}_{j}}=c_{j}^{-1}{\bm{b}_{j}}\in\mathbb{S}^{R-1}$. Here, $\bm{G}\in {{\mathbb{C}}^{L\times {{L}^{2}}}}$ is the Gabor matrix whose elements are given by $$\begin{aligned}
\label{gabor}
[\bm{G}]_{p,(k,l)}:={{x}_{p-l}}{{\rm e}^{i2\pi (\frac{kp}{L})}},\quad k,l,p=-N,...,N.\end{aligned}$$ We observe from that $\bm{X}\in \mathbb{C}^{L^{2}\times R} $ is a sparse combination of a few matrix atoms $\bm{a}(\bm{r}_j,\bm{\varphi}_j), j=1,..., s$ belonging to the atomic set $$\begin{aligned}
\mathcal{A} :=\{\bm{a(r,\varphi)}:= \bm{a}(\bm{r})\bm{\varphi}^{\rm H}:\quad \bm{ r}\in {{[0,1]}^{2}},\quad \bm{\varphi} \in {{\mathbb{S}}^{R-1}}\}.\end{aligned}$$ Hence, to extract $\bm{X}\in\mathbb{C}^{L^2\times R}$ from the underdetermined observations $\bm{Y}\in\mathbb{C}^{L\times R}$, inspired by [@chandrasekaran2012convex], we propose the atomic norm minimization $$\begin{aligned}
\label{atommin}
\min_{\bm{Z}\in\mathbb{C}^{L^2\times R}}\|\bm{Z}\|_{\mathcal{A}}~\text{ subject to } \bm {Y}=\bm{G Z},\end{aligned}$$ where the atomic norm $\|\bm{X}\|_{\mathcal{A}}$ is defined as $$\begin{aligned}
\label{atomic}
&{{\left\| \bm{X} \right\|}_{\mathcal{A}}}:=\inf \{t>0: \bm{X}\in {t\,
{\rm conv}(\mathcal{A})}\}=\nonumber\\
&\inf \{\sum\limits_{j}{{{c}_{j}}:\bm{X}}=\sum\limits_{j}{{{c}_{j}}\bm{a}(\bm{r}_j,\bm{\varphi}_j)}, {{c}_{_{j}}}>0,{{\bm{r}}_{j}}\in {{[0,1]}^{2}}\}, \end{aligned}$$ and ${\rm conv}(\mathcal{A})$ denotes the convex hull of $\mathcal{A}$.
Main Result
===========
In the following, we state our main result which provides conditions for exact recovery of $\bm{X}$ in .
\[thm.main\] Suppose that the entries of the probing signal $x_l$ , $l=-N,...,N$, are i.i.d. random variables distributed as $\mathcal{N} (0,\tfrac{1}{L})$ where $L=2N+1$. Let $\bm{Y}\in\mathbb{C}^{L\times R}$ be the observed matrix at the $R$-element antenna array as in i.e. $$\begin{aligned}
\bm{Y}=\bm{G X}\in\mathbb{C}^{L\times R},~\bm{X}=\sum\limits_{{{\bm{r}}_{j}}\in \mathcal{S}}{{{c}_{j}}\bm{a}(\bm{r}_j,\bm{\varphi}_j)}\end{aligned}$$ with $L>1024$. Here, $\bm{G}$ is the Gabor matrix defined in and $\mathcal{S}$ is the location of delay-Doppler pairs corresponding to $\bm{X}$. Fix $\delta >0$. Assume that ${{\bm{\varphi} }_{j}}, j=1,..., s$ are independent and uniformly distributed on the unit sphere $\mathbb{S}^{R-1}$ with $\mathds{E}[{{\bm{\varphi} }_{j}}]=0$ and that the set of time-frequency shifts $ \mathcal{S} =\{{{\bm{r}}_{1}},{{\bm{r}}_{2}},...,{{\bm{r}}_{s}}\}\subset {{[0,1]}^{2}}$ obeys the minimum separation condition $$\begin{aligned}
\label{sep}
&\max(\Big|\tau_j-\tau_{j'}\Big|,\Big|\nu_j-\nu_{j'}\Big|)\ge\tfrac{2.38}{N} ,\\
&\forall(\tau_j,\nu_j),(\tau_{j'},\nu_{j'})\in \mathcal{S}\quad \text{with}\quad j\ne j'.
$$ Moreover, assume that $$\begin{aligned}
s\le c\tfrac{L}{\log^3(\tfrac{L^6R}{\delta})}
$$ where $c$ is a constant. Then, with probability at least $1-\delta $, $\bm{X}$ is the unique solution of .
Proof. See Appendix \[construction.dual.poly\].
The proof of Theorem \[thm.main\] is built upon constructing a certain dual certificate for . In the following proposition, we describe the desired form of a valid vector-valued dual certificate which guaranties the optimality of $\bm{X}$ in .
\[prop.optimality\] Assume that $\bm {Y}=\bm{G}\bm{X}$ with $\bm{X}=\sum_{\bm{r}_j\in\mathcal{S}}c_j\bm{a}(\bm{r}_j,\bm{\varphi}_j)$ where $\mathcal{S}$ is the location of delay-Doppler pairs corresponding to $\bm{X}$. If there exists a vector-valued dual polynomial $\bm{q}: [0,1]^2\rightarrow \mathbb{C}^{R\times 1}$, $$\begin{aligned}
\label{eq.dual_pol_vector}
\bm{q}(\bm{r})=\bm{\Lambda}^{\rm H}\bm{G}\bm{a}(\bm{r})
\end{aligned}$$ satisfying $$\begin{aligned}
\label{dual}
\begin{array}{cc}
\bm{q}(\bm{r}_j)=\bm{\varphi}_j,&\bm{r}_j\in\mathcal{S},\\
\|\bm{q}(\bm{r})\|_2\le 1,& \bm{r}\in[0,1]^2\setminus \mathcal{S},
\end{array}
\end{aligned}$$ then $\bm{X}$ is the optimal solution of .
The problem involves finding infinitely many variables and can not be directly solved. To practically solve , we first obtain its dual formulation obtained from a standard Lagrangian approach (e.g. see [@bertsekas2003convex Chapter 6]): $$\begin{aligned}
\label{eq.dual_prob}
\max_{\bm{\Lambda}\in\mathbb{C}^{L\times R}}\,{\rm Re}~ {{\langle \text{ }\!\!\bm{\Lambda},\bm{Y} \rangle }_{{F}}}\text{ subject to }\left\| {{\bm{G}}^{\rm H}}\bm{\Lambda}\right\|_{\mathcal{A}}^{d}\le 1,\end{aligned}$$ where $\bm{\Lambda}=[{{\bm{\Lambda}}_{pm}}]\in {{\mathbb{C}}^{L\times R}}, m=1,..., R, p=-N,..., N$ and $$\begin{aligned}
\|\bm{V}\|_{\mathcal{A}}^d:=\sup_{\|\bm{Z}\|_{\mathcal{A}}\le 1} {\rm Re}\langle \bm{V}, \bm{Z} \rangle\end{aligned}$$ is the dual norm. Hence, using , we have $$\begin{aligned}
\label{eq.dualnorm_def}
&\|\bm{G}^{\rm H}\bm{\Lambda}\|_{\mathcal{A}}^d=\sup_{\substack{\|\bm{\varphi}\|_2=1\\\bm{r}\in[0,1]^2}}\langle\bm{G}^{\rm H}\bm{\Lambda}, \bm{a}(\bm{r})\bm{\varphi}^{\rm H}\rangle_F=\nonumber\\
&\sup_{\substack{\|\bm{\varphi}\|_2=1\\\bm{r}\in[0,1]^2}}\langle \bm{\varphi}, (\bm{F}\bm{G}^{\rm H}\bm{\Lambda})^{\rm H}\bm{f}(\bm{r})\rangle=\sup_{\bm{r}\in[0,1]^2}\|(\bm{F}\bm{G}^{\rm H}\bm{\Lambda})^{\rm H}\bm{f}(\bm{r})\|_2,\end{aligned}$$ where we used Holder inequality in the last step. Hence, the constraint of becomes equivalent to $$\begin{aligned}
\label{eq.constraint1}
&\|(\bm{F}\bm{G}^{\rm H}\bm{\Lambda})^{\rm H}\bm{f}(\bm{r})\|_2^2=\sum_{m=1}^R\bigg|\sum_{k,l=-N}^N[\bm{F}\bm{G}^{\rm H}\bm{\Lambda}]_{(k,l),m}{\rm e}^{i2\pi(k\tau+l\nu)}\bigg|^2\nonumber\\
&\le 1,~\forall \bm{r}\in [0,1]^2.\end{aligned}$$ By replacing , the dual problem involves infinitely many constraints. The following proposition which is an adaptation of [@fernandez2016super Proposition 2.4] and [@dumitrescu2017positive Corollary 4.27] provides a tractable sufficient condition for the constraint .
\[prop.boundedlemma\] Let $\bm{P}=[P_{(k,l),m}]$ be a matrix in $\in\mathbb{C}^{L^2\times R}$ with $k,l=-N,..., N,~ m=1,..., R, L=2N+1$. If $$\begin{aligned}
&\sum_{m=1}^R|\sum_{k,l=-N}^N P_{(k,l),m}{\rm e}^{i2\pi(k\tau+l\nu)}|^2\le 1,~~\forall r \in [0,1]^2,
\end{aligned}$$ then there exists a Hermitian positive semidefinite matrix $\bm{Q}\in \mathbb{C}^{L^2\times L^2}$ obeying $$\begin{aligned}
&\begin{bmatrix}
\bm{Q}& \bm{P}\\
\bm{P}^{\rm H}&\bm{I}_{R}
\end{bmatrix}\succeq 0,
{\rm trace}((\bm{\Theta}_k\otimes \bm{\Theta}_l)\bm{Q})=\delta_{(k,l)},\nonumber\\
&\forall k,l=-N, ..., N,
$$ where $$\begin{aligned}
\delta_{(k,l)}:=\left\{\begin{array}{cc}
1, & (k,l)=(0,0),\\
0, & {\rm o.w.}
\end{array}\right\}\end{aligned}$$ is the indicator function and $\bm{\Theta }_{k}$ stands for the Toeplitz matrix composed of ones on the $k$-th diagonal and zeros elsewhere.
By exploiting Proposition \[prop.boundedlemma\], the dual problem is relaxed to the following semidefinite programming (SDP): $$\begin{aligned}
\label{eq.sdp}
&\max_{\substack{\bm{\Lambda}\in\mathbb{C}^{L\times R}\\\bm{Q}\in\mathbb{C}^{L^2\times L^2}, \bm{Q}\succeq 0}}\,{\rm Re}{{\left\langle \text{ }\!\!\bm{\Lambda},\bm{Y} \right\rangle }_{{F}}}\text{ subject to }\nonumber\\
&\begin{bmatrix}
\bm{Q}& \bm{F}\bm{G}^{\rm H}\bm{\Lambda}\\
\bm{\Lambda}^{\rm H}\bm{G}\bm{F}^{\rm H}&\bm{I}_{R}
\end{bmatrix}\succeq 0,
{\rm trace}((\bm{\Theta}_k\otimes \bm{\Theta}_l)\bm{Q})=\delta_{(k,l)},\nonumber\\
&\forall k,l=-N, ..., N,\end{aligned}$$
The problem is only a relaxation for . In fact, the size of $\bm{Q}$ could be larger than $L^2\times L^2$, since the sum of squares expression of a bivariate positive trigonometric polynomial with degree $(L,L)$ might have factors with degree greater than the minimum degree $(L,L)$. However, as simulation results of [@heckel2016super; @dumitrescu2017positive] indicate, relaxations of minimal degree often lead to optimal solutions in practice[^7].
When the measurements are contaminated with noise, i.e. $\bm{Y}=\bm{G}\bm{X}+\bm{W}$, we solve the following problem: $$\begin{aligned}
\underset{{\bm{Z}}}{\min} \|{\bm{Z}}\|_\mathcal{A} \quad {\rm subject ~ to}\text{ }\|\mathbf{Y}-\bm{G}{\bm{Z}}\|_{F}\le \eta,\end{aligned}$$ where $\eta$ is an upper-bound on the standard deviation of noise. Moreover, the SDP problem in this case takes the form $$\begin{aligned}
\label{sdp_noisy}
&\max_{\substack{\bm{\Lambda}\in\mathbb{C}^{L\times R}\\\bm{Q}\in\mathbb{C}^{L^2\times L^2}, \bm{Q}\succeq 0}}{\rm Re}~\langle\bm{\bm{\Lambda}},\mathbf{Y} \rangle_{F}-\eta\|\bm{\bm{\Lambda}}\|_F\quad {\rm subject ~to} \nonumber\\
&\begin{bmatrix}
\bm{Q}& \bm{F}\bm{G}^{\rm H}\bm{\Lambda}\\
\bm{\Lambda}^{\rm H}\bm{G}\bm{F}^{\rm H}&\bm{I}_{R}
\end{bmatrix}\succeq 0,
{\rm tr}((\bm{\Theta}_k\otimes \bm{\Theta}_l)\bm{Q})=\delta_{(k,l)},\nonumber\\
&\forall k,l=-N, ..., N,\end{aligned}$$ Now, we are ready to state the procedure of finding delay-Doppler pairs from the dual solution in both noiseless and noisy cases. Write the vector-valued dual polynomial $$\begin{aligned}
\label{eq.dual_pol}
\bm{q}(\bm{r})=\widehat{\bm{\Lambda}}^{\rm H}\bm{G}\bm{a}(\bm{r}),\end{aligned}$$ where $\widehat{\bm{\Lambda}}$ is the solutions of the SDP problems and . Proposition \[prop.optimality\] suggests that an estimate $\widehat{\mathcal{S}}$ of the set of delay-Doppler pairs $\mathcal{S}$ can be obtained from $$\begin{aligned}
\label{eq.support_estimates}
\widehat{\mathcal{S}}=\{\bm{r}\in[0,1]^2\big|\|\bm{q}(\bm{r})\|_2=1\}.
\end{aligned}$$
Simulation Results {#section.simulation}
==================
In this section, we consider the benefits of using MMVs in both noiseless and noisy cases. Let the probing signal $x_l, ~l=-N,...,N$ and the coefficients $\bm{b}_{jm}$ be drawn from i.i.d. uniform distribution on the complex unit sphere. Choose $N=4$. First, in top and bottom of Figure \[fig1\], we consider two time-frequency shifts in locations $\bm{r}_1=(0.2,0.5)$ and $\bm{r}_2=(0.8,0.5)$ with $R=1$ and $R=10$, respectively. We implement the problem via SDPT3 in CVX package [@cvx] and plot $\|\bm{q}(\bm{r})\|_2$ where $\bm{q}(\bm{r})$ is obtained from . Then, according to , we estimate time-frequency shifts by checking where $\|\bm{q}(\bm{r})\|_2$ achieves one. As it turns out from top and bottom images of Figure \[fig1\], while the SMV case ($R=1$) fails and find spurious sources, the MMV case ($R=10$) localize the delay-Doppler pairs correctly. This in turn shows that using more antenna arrays improves the recovery performance for a fixed and weak time-frequency minimum separation. In the second experiment, we check a case where the sources are closer to each other as shown in top and bottom images of Figures \[fig2\]. Again, we can see the superiority of using MMV ($R=30$) over SMV. All in all, one can infer from Figures \[fig1\] and \[fig2\] that under a fixed number of measurements $N=4$, benefiting more MMVs can make the required minimum separation condition weaker (alternatively leading to a more relaxed condition). In the third experiment, we examine the noisy case where we consider complex noise $\bm{W}$ with i.i.d. Gaussian elements such that the signal to noise ratio (SNR) defined as $\text{SNR}=\tfrac{\|\bm{Y}\|_{F}^{2}}{\|\bm{W}\|_{F}^{2}}$ is equal to $10\,$dB. We choose two delay-Doppler pairs $\bm{r}_1=(0.2,0.5)$ and $\bm{r}_2=(0.8,0.5)$ and implement with $\eta=0.8$. As shown in top and bottom images of Figure \[fig3\], increasing the array elements (from $R=1$ in the top image to $R=50$ in the bottom image) can improve the recovery performance of delay-Doppler pairs.
[.5]{} ![Noiseless case. The true sources are located at $\bm{r}_1=(0.2,0.5)$, $\bm{r}_2=(0.8,0.5)$. We set $N=4$ and solve . Top and bottom images show $\|\bm{q}(\bm{r})\|_2$ for SMV ($R=1$) and MMV ($R=10$) cases, respectively. Red markers show where $\|\bm{q}(\bm{r})\|_2$ equals one.[]{data-label="fig1"}](T=1ed "fig:")
[.5]{} ![Noiseless case. The true sources are located at $\bm{r}_1=(0.2,0.5)$, $\bm{r}_2=(0.8,0.5)$. We set $N=4$ and solve . Top and bottom images show $\|\bm{q}(\bm{r})\|_2$ for SMV ($R=1$) and MMV ($R=10$) cases, respectively. Red markers show where $\|\bm{q}(\bm{r})\|_2$ equals one.[]{data-label="fig1"}](noiseless,T=100ed "fig:")
[.4]{} ![Noiseless case. The true sources are located at $\bm{r}_1=(0.2,0.2)$ and $\bm{r}_2=(0.3,0.3)$. We set $N=4$ and solve . Top and bottom images show $\|\bm{q}(\bm{r})\|_2$ for SMV ($R=1$) and MMV ($R=30$) cases, respectively. Red markers show where $\|\bm{q}(\bm{r})\|_2$ equals one.[]{data-label="fig2"}](minom-sep-T=1ed "fig:")
[.4]{} ![Noiseless case. The true sources are located at $\bm{r}_1=(0.2,0.2)$ and $\bm{r}_2=(0.3,0.3)$. We set $N=4$ and solve . Top and bottom images show $\|\bm{q}(\bm{r})\|_2$ for SMV ($R=1$) and MMV ($R=30$) cases, respectively. Red markers show where $\|\bm{q}(\bm{r})\|_2$ equals one.[]{data-label="fig2"}](mini-sep-T=100 "fig:")
[.5]{} ![Noisy case. The true sources are located at $\bm{r}_1=(0.2,0.2)$, $\bm{r}_2=(0.3,0.3)$. We set $N=4, \eta=0.8, {\rm SNR}=10{\rm dB}$ and solve . Top and bottom images show $\|\bm{q}(\bm{r})\|_2$ for SMV ($R=1$) and MMV ($R=50$) cases, respectively. Red markers show where $\|\bm{q}(\bm{r})\|_2$ equals one.[]{data-label="fig3"}](noisy-T=1,snr=10ed "fig:")
[.5]{} ![Noisy case. The true sources are located at $\bm{r}_1=(0.2,0.2)$, $\bm{r}_2=(0.3,0.3)$. We set $N=4, \eta=0.8, {\rm SNR}=10{\rm dB}$ and solve . Top and bottom images show $\|\bm{q}(\bm{r})\|_2$ for SMV ($R=1$) and MMV ($R=50$) cases, respectively. Red markers show where $\|\bm{q}(\bm{r})\|_2$ equals one.[]{data-label="fig3"}](noisy-minied "fig:")
Equivalence of and {#proof.equivalence}
===================
By sampling at rate $\tfrac{1}{B}$, the equation can be displayed as follows: $$\begin{aligned}
\label{24}
&y_m(\tfrac{p}{B})=\sum_{j=1}^sb_{jm}x(\tfrac{p}{B}-\overline{\tau}_j){\rm e}^{i2\pi \overline{\nu}_j \tfrac{p}{B}}, p=-N,..., N,
$$ we know that (by applying the discrete Fourier transform (DFT) and inverse DFT (IDFT) to $x$): $$\begin{aligned}
\label{25}
&x(\tfrac{p}{B}-\tfrac{\tau_jTB}{B})=x(\tfrac{p}{B}-\tfrac{\tau_jL}{B})=x[p-\tau_jL]\nonumber\\
&=\tfrac{1}{L}\sum_{k=-N}^{N}\Big(\sum_{l=-N}^{N}x[l]e^{-\tfrac{i2\pi kl}{L}}\Big){\rm e}^{-\tfrac{i2\pi \tau_jLk}{L}}e^{\tfrac{i2\pi kp}{L}}.\end{aligned}$$ Substituting into leads to $$\begin{aligned}
\label{eq.rel6}
&y(\tfrac{p}{B})=\tfrac{1}{L}\sum_{j=1}^{s}b_{jm}\sum_{k=-N}^{N}\Big(\sum_{l=-N}^{N}x[l]e^{-\tfrac{i2\pi kl}{L}}\Big){\rm e}^{-\tfrac{i2\pi \tau_jLk}{L}}\nonumber\\&{\rm e}^{\tfrac{i2\pi kp}{L}}{\rm e}^{i2\pi {\nu}_jp}=\sum_{j=1}^{s}b_{jm}\tfrac{1}{L}\sum_{k,l=-N}^{N}x[l]{\rm e}^{\tfrac{i2\pi k(p-l)}{L}}\nonumber\\
&{\rm e}^{i2\pi [p\nu_j-k\tau_j]}=\sum_{j=1}^{s}b_{jm}{\rm e}^{i2\pi p \nu_j }\tfrac{1}{L}\sum_{l=-N}^{N}\sum_{k=-N}^{N}x[l]\nonumber\\
&{\rm e}^{i2\pi [\tfrac{p-l}{L}-\tau_j]k}=\sum_{j=1}^{s}b_{jm}{\rm e}^{i2\pi p\nu_j}\tfrac{1}{L}\sum_{n=p-N}^{p+N}\sum_{k=-N}^{N}x[p-n],\nonumber\\
&{\rm e}^{i2\pi [\tfrac{n}{L}-\tau_j]k} ~~~ p=-N,..., N, ~ m=1,..., R.\end{aligned}$$ By using the definition , the fact that $$\begin{aligned}
\sum_{k=-N}^{N}D_N(\tfrac{k}{L}-\nu_j){\rm e}^{\tfrac{i2\pi pk}{L}}={\rm e}^{i2\pi p\nu_j},\end{aligned}$$ and the periodicity property of $x_l$, becomes $$\begin{aligned}
&y_m(\tfrac{p}{B})=\sum_{j=1}^{s}b_{jm}{\rm e}^{i2\pi p\nu_j}\tfrac{1}{L}\sum_{l=-N}^{N}\sum_{k=-N}^{N}x[l-p]{\rm e}^{i2\pi [\tfrac{l}{L}-\tau_j]k}\nonumber\\
&=\sum_{j=1}^{s}b_{jm}{\rm e}^{i2\pi p\nu_j}\sum_{l=-N}^{N}x[l-p]D_N(\tfrac{l}{L}-\tau_j)\nonumber\\
&=\sum_{j=1}^{s}b_{jm}\sum_{k=-N}^{N}D_N(\tfrac{k}{L}-\nu_j){\rm e}^{\tfrac{i2\pi pk}{L}}\sum_{l=-N}^{N}x[l-p]\nonumber\\
&\cdot D_N(\tfrac{l}{L}-\tau_j)=\nonumber\\
&\sum_{j=1}^{s}b_{jm}\sum_{k,l=-N}^{N}D_N(\tfrac{k}{L}-\nu_j)D_N(\tfrac{l}{L}-\tau_j)x_{l-p}{\rm e}^{\tfrac{i2\pi pk}{L}}.\end{aligned}$$
Proof of Proposition \[prop.optimality\]
========================================
We begin with $\bm{\Lambda}$ which lies in the feasible set of , since due to and the assumptions , we have: $$\begin{aligned}
\label{rel5}
&\|\bm{G}^{\rm H}\bm{\Lambda}\|_{\mathcal{A}}^d=\sup_{\substack{ \bm{r}\in[0,1]^2}}\|\bm{q}(\bm{r})\|_2 \le 1.\end{aligned}$$ We proceed by writing $$\begin{aligned}
&\|\bm{X}\|_{\mathcal{A}}\ge \|\bm{G}^{\rm H}\bm{\Lambda}\|_{\mathcal{A}}^d\|\bm{X}\|_{\mathcal{A}}\stackrel{({\textup{\uppercase\expandafter{\romannumeral1}}})}{\ge}{\rm Re}~\langle \bm{G}^{\rm H}\bm{\Lambda}, \bm{X} \rangle_F={\rm Re}~\langle \bm{\Lambda}, \bm{Y} \rangle_F=\nonumber\\
&{\rm Re}~\langle \bm{\Lambda}, \bm{G} \sum_{\bm{r}_j\in\mathcal{S}}c_j\bm{a}(\bm{r}_j)\bm{\varphi}_j^{\rm H} \rangle_F=\sum_{\bm{r}_j\in\mathcal{S}}{\rm Re}~c_j\langle \bm{\varphi}_j, \bm{\varphi}_j \rangle
\stackrel{({\textup{\uppercase\expandafter{\romannumeral2}}})}{=}\nonumber\\
&\sum_{\bm{r}_j\in\mathcal{S}}c_j=\|\bm{X}\|_{\mathcal{A}},\end{aligned}$$ where the inequality $({\textup{\uppercase\expandafter{\romannumeral1}}})$ is due to , and the equality $({\textup{\uppercase\expandafter{\romannumeral2}}})$ stems from the assumptions in . The latter relation shows that all the inequalities must be turned into equality. Thus, ${\rm Re}\langle \bm{\Lambda},\bm{Y}\rangle=\|\bm{X}\|_{\mathcal{A}}$ which in turn shows that $(\bm{X},\bm{\Lambda})$ are primal-dual optimal solutions. For uniqueness, we argue by contradiction and assume that there exists another optimal primal solution $\widehat{\bm{X}}=\sum_{\bm{r}_j\in\widehat{\mathcal{S}}}\widehat{c}_j\bm{a}(\bm{r}_j)\widehat{\bm{\varphi}}_j^{\rm H}$ where $\widehat{S}\neq \mathcal{S}$. It holds that $$\begin{aligned}
&\|\widehat{\bm{X}}\|_{\mathcal{A}}={\rm Re}~\langle \bm{\Lambda}, \bm{G}\widehat{\bm{X}}\rangle={\rm Re}~\langle \bm{\Lambda}, \bm{G} \sum_{\bm{r}_j\in\widehat{\mathcal{S}}}\widehat{c}_j\bm{a}(\bm{r}_j)\widehat{\bm{\varphi}}_j^{\rm H}\rangle=\nonumber\\
&\sum_{\bm{r}_j\in {\mathcal{S}}}{\rm Re}~\widehat{c}_j\langle \widehat{\bm{\varphi}}_j, \bm{q}(\bm{r}_j)\rangle+
\sum_{\bm{r}_j\in \widehat{\mathcal{S}}\setminus \mathcal{S}}{\rm Re}~\widehat{c}_j\langle \widehat{\bm{\varphi}}_j, \bm{q}(\bm{r}_j)\rangle
<\sum_{\bm{r}_j\in\widehat{\mathcal{S}}}\widehat{c}_j\nonumber\\
&=\|\widehat{\bm{X}}\|_{\mathcal{A}},\end{aligned}$$ where we used the assumptions in in the last inequality. Hence, we have a contradiction and $\widehat{\mathcal{S}}=\mathcal{S}$. As a consequence, since $\bm{a}(\bm{r}_j),~ \bm{r}_j\in\mathcal{S}$ are linearly independent, the optimal primal solution is unique.
Proof of Theorem \[thm.main\] {#construction.dual.poly}
=============================
In this section, we prove Theorem \[thm.main\] by constructing 2D vector-valued dual polynomial $\bm{q}$ satisfying . Without loss of generality, we assume that $N$ is even and define the squared Fejer kernel $$\begin{aligned}
K(t):=\tfrac{1}{M}\sum\limits_{k=-N}^{N}{{{g}_{k}}{{\rm e}^{i2\pi tk}}}:={{\left( \tfrac{\sin (M\pi t)}{M\sin (\pi t)} \right)}^{4}},\text{ }M=\tfrac{N}{2}+1,\end{aligned}$$ where $g_k$ is the discrete convolution of two triangular functions. First, in the following, we construct a deterministic dual polynomial satisfying which is later used in our analysis: $$\begin{aligned}
\label{determin}
& \overline{\bm{q}}(\mathbf{r)=}\sum\limits_{j=1}^{s}{{ \overline{\bm{\alpha} }}_{j}}{ \overline{G}(\bm{r}-{{\bm{r}}_{j}})}+{{ \overline{\bm{\beta }}}_{j}}{{ \overline{G}}^{(1,0)}}(\bm{r}-{{\bm{r}}_{j}}) +\nonumber \\
& \overline{\bm{\gamma }}_{j}{{ \overline{G}}^{(0,1)}}(\bm{r}-{{\bm{r}}_{j}}),\end{aligned}$$ where the coefficients ${{ \overline{\bm{\alpha} }}_{j}},{{ \overline{\bm{\beta} }}_{j}},{{ \overline{\bm{\gamma} }}_{j}}\in {{\mathbb{C}}^{R\times 1}}$, $\overline{G}^{(m,n)} = \tfrac{\partial^{m}\partial^{n}\overline{G}}{\partial \tau^{m}\partial \nu^{n}} $ and $ \overline{G}(\bm{r}):=K(\tau )K(\nu )$. An important requirement for the condition to hold, is that $ \overline{\bm{q}}(\bm{r})$ reaches the local maxima by choosing the specific coefficients ${{ \overline{\bm{\alpha}}}_{j}}, {{ \overline{\bm{\beta} }}_{j}}, {{ \overline{\bm{\gamma}}}_{j}}$ satisfying $$\begin{aligned}
\label{eq.conditions_qbar}
& \overline{\bm{q}}({\bm{r}_j})=\bm{\varphi}_j, & \forall \bm{r}_j \in \mathcal{S} \nonumber\\ & \overline{\bm{q}}^{(1,0)}({\bm{r}_j})=\bm{0}\in \mathbb{C}^{R\times 1}, &\forall \bm{r}_j \in \mathcal{S}\nonumber\\
& \overline{\bm{q}}^{(0,1)}({\bm{r}_j})=\bm{0}\in \mathbb{C}^{R\times 1}, &\forall \bm{r}_j \in \mathcal{S},\end{aligned}$$ where ${{ \overline{\bm{q}}}^{(m,n)}}(\bm{r}):=\tfrac{\partial^{m}\partial^{n}\overline{\bm{q}}}{\partial \tau^{m}\partial \nu^{n}} $. Now, we construct the random polynomial $\bm{q}(\bm{r})$ with function $G_{(m,n)}(\bm{r},\bm{r}_j),m,n=0,1$ as $$\begin{aligned}
\label{dual-poly}
&\bm{q}(\mathbf{r)=}\sum\limits_{j=1}^{s}{{\bm{\alpha} }_{j}}{{{G}_{^{(0,0)}}}(\bm{r},{{\bm{r}}_{j}})}+{{\bm{\beta} }_{j}}{{G}_{^{(1,0)}}}(\bm{r},{{\bm{r}}_{j}})\nonumber\\
&+ {{\bm{\gamma} }_{j}}{{G}_{^{(0,1)}}}(\bm{r},{{\bm{r}}_{j}}),\end{aligned}$$ where the coefficients $\bm{\alpha}_k,\bm{\beta}_k,\bm{\gamma}_k$ are such that: $$\begin{aligned}
\label{shart}
&{\bm{q}}({\bm{r}_j})=\bm{\varphi}_j,\quad\quad\quad\quad\quad~~ \forall \bm{r}_j \in \mathcal{S} \nonumber\\ &{\bm{q}}^{(1,0)}({\bm{r}_j})=\bm{0}\in \mathbb{C}^{R\times 1},\quad \forall \bm{r}_j \in \mathcal{S}\nonumber\\
&{\bm{q}}^{(0,1)}({\bm{r}_j})=\bm{0}\in \mathbb{C}^{R\times 1},\quad \forall \bm{r}_j \in \mathcal{S}\end{aligned}$$ and $\overline{G}:=\mathds{E}G$ is the expectation of kernel $G$. The dual polynomial $\bm{q}$ in can be regarded as a random version of $\overline{\bm{q}}$ in with the randomness introduced by $\bm{x}$.
Choice of coefficients
----------------------
We choose the coefficients ${{\bm{\alpha} }_{j}},{{\bm{\beta} }_{j}},{{\bm{\gamma} }_{j}}$ to construct $\bm{q}({\bm r})$ such that holds with high probability. Writing in matrix form, yields $$\begin{aligned}
\underbrace{\left[ \begin{matrix}
{{ \overline{\bm{D}}}^{(0,0)}} & \kappa^{-1}{{ \overline{\bm{D}}}^{(1,0)}} & \kappa^{-1}{{ \overline{\bm{D}}}^{(0,1)}} \\
-\kappa^{-1}{{ \overline{\bm{D}}}^{(1,0)}} & -\kappa^{-2}{{ \overline{\bm{D}}}^{(2,0)}} & -\kappa^{-2}{{ \overline{\bm{D}}}^{(1,1)}} \\
-\kappa^{-1}{{ \overline{\bm{D}}}^{(0,1)}} & -\kappa^{-2}{{ \overline{\bm{D}}}^{(1,1)}} & -\kappa^{-2}{{ \overline{\bm{D}}}^{(0,2)}} \\
\end{matrix} \right]}_{ \overline{\bm{D}}}\left[ \begin{matrix}
\overline{\bm{\alpha} } \\
\kappa \overline{\bm{\beta} } \\
\kappa \overline{\bm{\gamma} } \\
\end{matrix} \right]=\left[ \begin{matrix}
\overline{\bm{\Phi}} \\
\bm{0} \\
\bm{0} \\
\end{matrix} \right],\end{aligned}$$ where $(\kappa^{2}=\left| {K}''(0) \right|=\sqrt{\tfrac{\pi^2}{3}(N^2+4N)},K(0)=1)$ and ${{\left[ {{ \overline{\bm{D}}}^{(m,n)}} \right]}_{\kappa,j}}:={{ \overline{G}}^{(m,n)}}({{\bm{r}}_{\kappa}}-{{\bm{r}}_{_{j}}})$, $\overline{\bm{\Phi}}=[\bm{\overline{\varphi}}_1, \bm{\overline{\varphi}}_2,\hdots,\bm{\overline{\varphi}}_s]^{\top} \in \mathbb{C}^{s\times R}$ , $\bm{\overline{\alpha}}=[\bm{\overline{\alpha}}_{1},\bm{\overline{\alpha}}_{2},\hdots,\bm{\overline{\alpha}}_{s}]^{\top} \in \mathbb{C}^{s\times R}$, $\bm{\overline{\beta}}=[\bm{\overline{\beta}}_{1}, \bm{\overline{\beta}}_{2},\hdots, \bm{\overline{\beta} }_{s}]^{\top} \in \mathbb{C}^{s\times R}$, $\bm{\overline{\gamma}}=[\bm{\overline{\gamma}}_1,\bm{\overline{\gamma}}_2,\hdots,\bm{\overline{\gamma}}_s]^{\top}\in \mathbb{C}^{s\times R}$ and $ \overline{\bm{D}}$ is symmetric because $ \overline{\bm{D}}^{(0,0)},~\overline{\bm{D}}^{(1,1)}, \overline{\bm{D}}^{(2,0)},~\overline{\bm{D}}^{(0,2)}$ are symmetric and $ \overline{\bm{D}}^{(1,0)},~\overline{\bm{D}}^{(0,1)}$ are antisymmetric. ${ \overline{\bm{D}}}$ is invertible and also the coefficients can be obtained as $$\begin{aligned}
\left[ \begin{matrix}
\overline{\bm{\alpha} } \\
\kappa \overline{\bm{\beta} } \\
\kappa \overline{\bm{\gamma} } \\
\end{matrix} \right]={{ \overline{\bm{D}}}^{-1}}\left[ \begin{matrix}
\bm{\Phi} \\
\bm{0 } \\
\bm{0 } \\
\end{matrix} \right]= \overline{\mathbf{L}}\bm{\Phi},\end{aligned}$$ where $ \overline{\mathbf{L}}\in \mathbb{C}^{3s\times s}$ is the first $ s $ columns of ${ \overline{\bm{D}}^{-1}}$.
[@heckel2016super Proposition 8.2] ${{ \overline{\bm{D}}}}$ is invertible and $$\begin{aligned}
\left\| \mathbf{I}- \overline{\bm{D}} \right\|\le 0.19808,\end{aligned}$$ $$\begin{aligned}
\left\| \overline{\bm{D}} \right\|\le 1.19808,\end{aligned}$$ $$\begin{aligned}
\left\| {{ \overline{\bm{D}}}^{-1}} \right\|\le 1.24700.\end{aligned}$$
Next, we choose the coefficients ${{\bm{\alpha} }},{{\bm{\beta} }},{{\bm{\gamma} }}\in {{\mathbb{C}}^{s\times R}}$ such that the conditions hold. First, write in matrix form as $$\begin{aligned}
\label{36}
\underbrace{\left[ \begin{matrix}
\bm{D}_{(0,0)}^{(0,0)} & \kappa^{-1}\bm{D}_{(1,0)}^{(0,0)} & \kappa^{-1}\bm{D}_{(0,1)}^{(0,0)} \\
-\kappa^{-1}\bm{D}_{(0,0)}^{(1,0)} & -\kappa^{-2}\bm{D}_{(1,0)}^{(1,0)} & -\kappa^{-2}\bm{D}_{(0,1)}^{(1,0)} \\
-\kappa^{-1}\bm{D}_{(0,0)}^{(0,1)} & -\kappa^{-2}\bm{D}_{(1,0)}^{(0,1)} & -\kappa^{-2}\bm{D}_{(0,1)}^{(0,1)} \\
\end{matrix} \right]}_{\bm{D}}\left[ \begin{matrix}
\bm{\alpha} \\
{{\kappa}}\bm{\beta} \\
{{\kappa}}\bm{\gamma} \\
\end{matrix} \right]=\left[ \begin{matrix}
\bm{\Phi} \\
\bm{0 } \\
\bm{0 } \\
\end{matrix} \right],\end{aligned}$$ where $\Big[\bm{D}^{(m,n)}_{(m',n')}\Big]_{j,k}
:=G^{(m,n)}_{(m',n')}(\bm{r}_j,\bm{r}_k)$, $\bm{\Phi}=[\bm{\varphi}_1, \bm{\varphi}_2,\hdots,\bm{\varphi}_s]^{\top} \in \mathbb{C}^{s\times R}$ , $\bm{\alpha}=[\bm{\alpha}_{1},\bm{\alpha}_{2},\hdots,\bm{\alpha}_{s}]^{\top}$, $\bm{\beta}=[\bm{\beta}_{1}, \bm{\beta}_{2},\hdots, \bm{\beta}_{s}]^{\top}$, $\bm{\gamma}=[\bm{\gamma}_1,\bm{\gamma}_2,\hdots,\bm{\gamma}_s]^{\top}$, where $\bm{\alpha}_j$, $\bm{\beta}_j$, $\bm{\gamma}_j \in \mathbb{C}^{R \times 1}$ with $j={1,\hdots, s}$. To prove the existence of coefficients $\bm{\alpha},\bm{\beta}, \bm{\gamma}$, we show that $\bm{D}$ in is invertible with high probability. Define the event $$\begin{aligned}
{{\zeta }_{\xi }}=\{\left\| \bm{D}-\overline{\bm{D}} \right\|\le \xi \}.\end{aligned}$$ If ${{\zeta }_{\xi }}$ occurs with $\xi \in \left( 0,\tfrac{1}{4} \right]$, $\bm{D}$ is invertible since $$\begin{aligned}
\left\| \mathbf{I}-\bm{D} \right\|\le \left\| \bm{D}- \overline{\bm{D}} \right\|+\left\| \overline{\bm{D}}-\mathbf{I} \right\|\le \xi +0.1908\le 0.4408.\end{aligned}$$ Hence, $\bm{\alpha}_j,\bm{\beta}_j,\bm{\gamma}_j$ can be given as $$\begin{aligned}
\label{coefficient}
\left[ \begin{matrix}
{\bm{\alpha} } \\
\kappa{\bm{\beta} } \\
\kappa{\bm{\gamma} } \\
\end{matrix} \right]={{{\bm{D}}}^{-1}}\left[ \begin{matrix}
\bm{\Phi} \\
\bm{0 } \\
\bm{0 } \\
\end{matrix} \right]={\mathbf{L}}\bm{\Phi},\end{aligned}$$ where $\mathbf{L} \in \mathbb{C}^{3s\times s}$ is the first $s$ columns of $\bm{D}^{-1}$. To proceed, we use the following important lemma about the concentration of $\bm{L}$ around $\overline{\bm{L}}$:
\[lem.L\_bounds\]([@heckel2016super Lemma 8.4]) If the event ${{\zeta }_{\xi }}$ with $\xi \in (0,\tfrac{1}{4}]$ occurs, then we have $$\begin{aligned}
\| \mathbf{L} \|\le 2.5,
\end{aligned}$$ $$\begin{aligned}
\| \mathbf{L}- \overline{\mathbf{L}}\|\le 2.5\xi.
\end{aligned}$$
The following lemma provides conditions that ${\zeta }_{\xi }$ occurs with high probability. We use this lemma to complete our proof.
\[2\]([@heckel2016super Lemma 8.6]) If $$\begin{aligned}
L\ge s\tfrac{{{c}_{1}}}{{{\xi }^{2}}}{{\log }^{2}}\tfrac{18{{s}^{2}}}{\delta},
\end{aligned}$$ then, $$\begin{aligned}
\mathds{P} [{{\zeta }_{\xi }}] \ge 1-\delta .\end{aligned}$$
Showing that TEXT and TEXT are close on a grid
-----------------------------------------------
The goal of this section is to show that $\bm{q}(\bm{r})\text{ }$ and $\overline{\bm{q}}(\bm{r})\text{ }$ are close in a set of grid points $\Omega$.
\[3\] Let $\Omega \subset {{[0,1]}^{2}}$ be a finite set of points. Fix $0<\epsilon \le 1$ and $\delta {>0}$. If $$\begin{aligned}
&L\ge \tfrac{s}{{{\epsilon }^{2}}}\max ({{c}_{2}}{{\log }^{2}}(\tfrac{12s\left| \Omega \right|}{\delta })\log (\tfrac{2(R+1)\left| \Omega \right|}{\delta }),\\
&{{c}_{3}}\log (\tfrac{(R+1)\left| \Omega \right|}{\delta })\log (\tfrac{18{{s}^{2}}}{\delta })),\nonumber\end{aligned}$$ then, $\mathds{P}\left[ \underset{\bm{r}\in \Omega }{\mathop{\max}}\,\tfrac{1}{\kappa^{m+n}}{{\left\| {{\bm{q}}^{(m,n)}}(\bm{r})-{{\overline{\bm{q}}}^{(m,n)}}(\bm{r}) \right\|}_{2}}\le \epsilon \right]\ge 1-4\delta $.
It is straightforward to verify that $(m,n)$-th partial derivative of the dual polynomial $\bm{q}(\bm{r})$ can be written as $$\begin{aligned}
\label{45}
&\tfrac{1}{\kappa^{m+n}}{{\bm{q}}^{(m,n)}}(\bm{r})=\sum\limits_{j=1}^{s}{G_{(0,0)}^{(m,n)}}(\bm{r},{{\bm{r}}_{j}}){{\bm{\alpha} }_{j}}\nonumber\\
&+\tfrac{1}{{{\kappa}}}G_{(1,0)}^{(m,n)}(\bm{r},{{\bm{r}}_{j}}){{\kappa}}{{\bm{\beta} }_{j}}+\tfrac{1}{{{\kappa}}}G_{(0,1)}^{(m,n)}(\bm{r},{{\bm{r}}_{j}}){{\kappa}}{{\bm{\gamma} }_{j}}\nonumber \\ &={{({{\bm{w}}^{(m,n)}}(\bm{r}))}^{\rm H}}\mathbf{L}\bm{\Phi},\end{aligned}$$ where $$\begin{aligned}
&{{\text{(}{{\bm{w}}^{(m,n)}})}^{\rm H}}(\bm{r}):=\tfrac{1}{\kappa^{m+n}}[G_{(0,0)}^{(m,n)}(\bm{r},{{\bm{r}}_{1}}),...,G_{(0,0)}^{(m,n)}(\bm{r},{{\bm{r}}_{s}}),\nonumber\\
&\tfrac{1}{{\kappa}}G_{(1,0)}^{(m,n)}(\bm{r},{{\bm{r}}_{1}}),...,\tfrac{1}{{{\kappa}}}G_{(1,0)}^{(m,n)}(\bm{r},{{\bm{r}}_{s}}),\nonumber\\
&\tfrac{1}{{{\kappa}}}G_{(0,1)}^{(m,n)}(\bm{r},{{\bm{r}}_{1}}),...,\tfrac{1}{{{\kappa}}}G_{(0,1)}^{(m,n)}(\bm{r},{{\bm{r}}_{s}})].\end{aligned}$$ Due to $\mathds{E}\left[ G_{({m}',{n}')}^{(m,n)}(\bm{r},{{\bm{r}}_{j}}) \right]={{ \overline{G}}^{(m+{m}',n+{n}')}}(\bm{r}-{{\bm{r}}_{j}})$, it holds that $\mathds{E}\left[ {{\bm{w}}^{(m,n)}}(\bm{r}) \right]={{ \overline{\bm{w}}}^{(m,n)}}(\bm{r})$, where $$\begin{aligned}
&{{\text{(}{{ \overline{\bm{w}}}^{(m,n)}})}^{\rm H}}(\bm{r}):=\tfrac{1}{\kappa^{m+n}}[{{ \overline{G}}^{(m,n)}}(\bm{r}-{{\bm{r}}_{1}}),...,{{ \overline{G}}^{(m,n)}}(\bm{r}-{{\bm{r}}_{s}}),\nonumber\\
&\tfrac{1}{{{\kappa}}}{{ \overline{G}}^{(m+1,n)}}(\bm{r}-{{\bm{r}}_{1}}),...,\tfrac{1}{{{\kappa}}}{{ \overline{G}}^{(m+1,n)}}(\bm{r}-{{\bm{r}}_{s}}),\nonumber\\
&\tfrac{1}{{{\kappa}}}{{ \overline{G}}^{(m,n+1)}}(\bm{r}-{{\bm{r}}_{1}}),...,\tfrac{1}{{{\kappa}}}{{ \overline{G}}^{(m,n+1)}}(\bm{r}-{{\bm{r}}_{s}})].\end{aligned}$$ Now, we can decompose as follows $$\begin{aligned}
&\tfrac{1}{\kappa^{m+n}}{{\bm{q}}^{(m,n)}}(\bm{r})=(\bm{w}^{(m,n)})^{{\rm H}}(\bm{r})\mathbf{L}\bm{\Phi}={{\text{(}{{ \overline{\bm{w}}}^{(m,n)}})}^{\rm H}}(\bm{r}) \overline{\mathbf{L}}\bm{\Phi}\nonumber\\ &-\underbrace{{{({{\bm{w}}^{(m,n)}}(\bm{r})-{{ \overline{\bm{w}}}^{(m,n)}})}^{\rm H}}\mathbf{L}\bm{\Phi} }_{\bm{I}_{1}^{(m,n)}(\bm{r})}+\underbrace{{{\text{(}{{ \overline{\bm{w}}}^{(m,n)}}\text{)}}^{\rm H}}(\mathbf{L}- \overline{\mathbf{L}})\bm{\Phi} }_{\bm{I}_{2}^{(m,n)}(\bm{r})}\nonumber\\
&=\tfrac{1}{\kappa^{m+n}}{{ \overline{\bm{q}}}^{(m,n)}}(\bm{r})+{\bm{I}_{1}^{(m,n)}(\bm{r})}+{\bm{I}_{2}^{(m,n)}(\bm{r})}.\end{aligned}$$ The following lemmas show that ${\bm{I}_{1}^{(m,n)}(\bm{r})}$ and ${\bm{I}_{2}^{(m,n)}(\bm{r})}$ are small on a set of grid points $\Omega $ with high probability.
\[4\] Consider $\Omega \subset [0,1]^{2}$ as a finite set of points and assume that $m+n \le 2$. Then, we have $$\begin{aligned}
\mathds{P}\left[ \underset{\bm{r}\in \Omega }{\mathop{\max }}\,{{\left\| \bm{I}_{1}^{(m,n)}(\bm{r}) \right\|}_{2}}\ge \epsilon \right]\le \delta +\mathds{P}\left[ {{{ \overline{\zeta}}}_{1/4}} \right],~ \forall \delta,\epsilon>0\end{aligned}$$ provided $ L\ge \tfrac{{{c}_{2}}}{{{\epsilon }^{2}}}s{{\log }^{2}}(\tfrac{12s\left| \Omega \right|}{\delta })\log (\tfrac{2(R+1)\left| \Omega \right|}{\delta })$.
Proof. See Appendix \[proof.of.lemma2\].
\[lem5\] Let $\Omega \subset [0,1]^{2}$ be a finite set of grid points and $m+n \le 2$. For all $\xi,\epsilon,\delta>0$, with $ \xi\le\tfrac{\epsilon }{c_3\sqrt{\log ((R+1)\tfrac{|\Omega|}{\delta} )}},$ where $c_3 \le \tfrac{1}{4}$, it holds that $\mathds{P}\left[ \underset{\bm{r}\in \Omega }{\mathop{\max }}\,{{\left\| \bm{I}_{2}^{(m,n)} \right\|}_{2}}\ge \epsilon \left| {{\zeta }_{\xi }} \right. \right]\le \delta$.
Proof. See Appendix \[proof.of.lemma3\].
Now we can complete the proof of Lemma \[3\] by writing $$\begin{aligned}
&\mathds{P}\left[ \underset{\bm{r}\in \Omega }{\mathop{\text{max}}}\,\tfrac{\text{1}}{\kappa^{m+n}}{{\left\| {{\bm{q}}^{(m,n)}}(\bm{r})-{{{ \overline{\bm{q}}}}^{(m,n)}}(\bm{r}) \right\|}_{2}}\ge 2\epsilon \right]\nonumber\\
&=\mathds{P}\left[ \underset{\bm{r}\in \Omega }{\mathop{\text{max}}}\,\tfrac{\text{1}}{\kappa^{m+n}}{{\left\| \bm{I}_{\text{1}}^{(m,n)}(\bm{r})+\bm{I}_{2}^{(m,n)} \right\|}_{2}}\ge 2\epsilon \right]\nonumber\\
&\le\mathds{P}\left[ \underset{\bm{r}\in \Omega }{\mathop{\text{max}}}\,\tfrac{\text{1}}{\kappa^{m+n}}{{\left\| \bm{I}_{\text{1}}^{(m,n)}(\bm{r}) \right\|}_{2}}\ge \epsilon \right]+\mathds{P}\left[ {{\overline{\zeta }}_{\xi }} \right]\nonumber\\
&+\mathds{P}\left[ \underset{\bm{r}\in \Omega }{\mathop{\text{max}}}\,\tfrac{\text{1}}{\kappa^{m+n}}{{\left\| \bm{I}_{\text{2}}^{(m,n)}(\bm{r}) \right\|}_{2}}\ge \epsilon \left| {{\zeta }_{\xi }} \right. \right]\le 4\delta,\end{aligned}$$ where we used the union bound and Lemmas \[lem5\], \[4\]. By choosing $\xi=\epsilon c_3\log^{-\tfrac{1}{2}}((R+1)\tfrac{|\Omega|}{\delta} ) $, the condition in Lemma \[2\] becomes $L\ge s(\tfrac{c_1}{\epsilon c_3^{2}})\log (2(R+1))\log(\tfrac{18s^{2}}{\delta})$ where $c:=\tfrac{c_1}{c_3^{2}}$.
Showing that TEXT and TEXT are close for all TEXT
--------------------------------------------------
In this part, benefiting Lemma \[3\], we want to show that ${\bm{q}}^{(m,n)}(\bm{r})$ is close to $\overline{\bm{q}}^{(m,n)}(\bm{r})$ for all $\bm{r}\in[0,1]^{2}$ with high probability which is given in the following lemma.
\[lem.nearness\_all\_r\] Let $\epsilon,\delta>0$. It holds that $$\begin{aligned}
\label{dif_q(r)}
\underset{\bm{r}\in[0,1]^{2},(m,n):m+n\le2}{\max}\tfrac{1}{\kappa^{m+n}}\|\bm{q}^{(m,n)}(\bm{r})- \overline{\bm{q}}^{(m,n)}(\bm{r})\|_2\le\epsilon
\end{aligned}$$ with probability at least $1-\delta$ provided that $$\begin{aligned}
L\ge \tfrac{sc}{\epsilon^{2}}\log^{3}\Big(\tfrac{c'RL^{6}}{\delta\epsilon^{2}}\Big).
\end{aligned}$$
To prove, we first choose a set of sufficiently fine points in $\Omega$ such that $$\begin{aligned}
\max_{\bm{r}\in [0,1]^2}\min_{\bm{r}_g\in \Omega}\|\bm{r}-\bm{r}_g\|_{\infty}\le \tfrac{\epsilon}{3\tilde{c}L^{\tfrac{5}{2}}R^{\tfrac{1}{2}}}\end{aligned}$$ with cardinality $$\begin{aligned}
|\Omega|=\Big(\tfrac{3\tilde{c}L^{\tfrac{5}{2}}R^{\tfrac{1}{2}}}{\epsilon}\Big)^{2}=\tfrac{c'L^{5}R}{\epsilon^{2}}.\end{aligned}$$ By using the union bound over all six pairs $(m,n)$ obeying $m+n \le 2$ and following Lemma \[3\], we find that
$$\begin{aligned}
\label{dif_q(rg)}
&\mathds{P}\Big\{\underbrace{\underset{\bm{r}_g\in \Omega,m+n\le 2}{\max}\tfrac{1}{\kappa^{m+n}}\|\bm{q}^{(m,n)}(\bm{r}_g)- \overline{\bm{q}}^{(m,n)}(\bm{r}_g)\|_2\le\tfrac{\epsilon}{3}}_{\mathcal{E}_1}\Big\}\nonumber\\
&\ge 1-\frac{\delta}{2}.\end{aligned}$$
To prove that the same result holds for all $\bm{r}\in[0,1]^2$, it is necessary to show that $$\begin{aligned}
\label{q-event}
\mathds{P}\Big\{ \underbrace{\underset{\bm{r}\in [0,1]^{2},m+n\le 2}{\max}\tfrac{1}{\kappa^{m+n}}\|\bm{q}^{(m,n)}(\bm{r})\|_2\le\tfrac{\tilde{c}}{2}{L}^{\tfrac{3}{2}} }_{\mathcal{E}_2}\Big\}\ge 1-\frac{\delta}{2},\end{aligned}$$ which is proved in Appendix \[proof.equation\_norm\_q\]. We also have $\mathds{P}\{\mathcal{E}_1\cap\mathcal{E}_2\}\ge 1-\delta$. Since the event $\mathcal{E}_1\cap \mathcal{E}_2$ implies the event (see Appendix \[proof.whyimply\] for the reason), Lemma \[lem.nearness\_all\_r\] is concluded.
Showing that TEXT for all TEXT {#proof.offsupoort}
------------------------------
We begin with defining the following sets: $$\begin{aligned}
\Omega_{\rm far} := \forall \bm{r}\in [0,1]^{2} :\min_{\bm{r}_j\in \mathcal{S}}\|\bm{r}-\bm{r}_j\|_{\infty}\ge \tfrac{0.2447}{N}\label{eq.far}\\
\Omega_{\rm close}:= \forall \bm{r}\notin \mathcal{S},\bm{r}_j \in \mathcal{S} :0\le \|\bm{r}-\bm{r}_j\|_{\infty}\le \tfrac{0.2447}{N}\label{eq.near}.\end{aligned}$$ The former argument implies that the points are far from $\bm{r}_j$ while the latter include points that are close to it. In order to show that $\bm{q}(\bm{r})$ in satisfies , it is enough to show that $\|\bm{q}(\bm{r})\|_2 \le 1$ for $\forall \bm{r}\in \Omega_{\rm far}$ and $\forall \bm{r}\in \Omega_{\rm close}$. To proceed, suppose that $L\ge \tfrac{sc}{\epsilon^{2}}\log^{3}\Big(\tfrac{c'RL^{6}}{\delta\epsilon^{2}}\Big)$.
\[omega-far\] $\|\bm{q}(\bm{r})\|_2\le 1, \forall \bm{r}\in \Omega_{\rm far}$ with high probability.
To prove this, take $\epsilon=0.002$ in to reach $$\begin{aligned}
\|\bm{q}^{(m,n)}(\bm{r})- \overline{\bm{q}}^{(m,n)}(\bm{r})\|_2 \le 0.002. \end{aligned}$$ By the triangular inequality, we have $$\begin{aligned}
&\|\bm{q}^{(m,n)}(\bm{r})\|_2\le\|\bm{q}^{(m,n)}(\bm{r})- \overline{\bm{q}}^{(m,n)}(\bm{r})\|_2+\| \overline{\bm{q}}^{(m,n)}(\bm{r})\|_2\nonumber\\
&\le0.9978,\end{aligned}$$ which verifies that $\|\bm{q}(\bm{r})\|_2\le 1$ for far $\bm{r}$. For near $\bm{r}$, we state the following lemma.
\[lem.near\] $\|\bm{q}(\bm{r})\|_2\le 1, \forall \bm{r}\in \Omega_{\rm close}$ with high probability.
To prove this, assume without loss of generality that $\bm{0}\in \mathcal{S} $ i.e. $\|\bm{r}\|_{\infty}\le\tfrac{0.2447}{N}$. A sufficient condition for Lemma \[lem.near\] to hold, is to show that the Hessian matrix of $\|\bm{q}(\bm{r})\|_2^{2}$, i.e., $$\begin{aligned}
\tfrac{1}{\kappa^{2}}\bm{H}=\tfrac{1}{\kappa^{2}} \begin{bmatrix}
\tfrac{\partial^{2}\|\bm{q}(\bm{r})\|_2^{2}}{\partial\tau^{2}}& \tfrac{\partial^2\|\bm{q}(\bm{r})\|_2^{2}}{\partial\tau\partial\nu}\\
&\\
\tfrac{\partial^2 \|\bm{q}(\bm{r})\|_2^{2}}{\partial\tau\partial\nu}&\tfrac{\partial^{2}\|\bm{q}(\bm{r})\|_2^{2}}{\partial\nu^{2}}
\end{bmatrix} \end{aligned}$$ is negative definite for all near $\bm{r}$. For this to hold, we should have $$\begin{aligned}
\label{74}
\tfrac{1}{\kappa^{2}}{\rm tr}(\bm{H})=\tfrac{\partial^{2}}{\partial\tau^{2}}\tfrac{1}{\kappa^{2}}\|\bm{q}(\bm{r})\|_2^{2}+\tfrac{\partial^{2}}{\partial\nu^{2}}\tfrac{1}{\kappa^{2}}\|\bm{q}(\bm{r})\|_2^{2} < 0\end{aligned}$$ and $$\begin{aligned}
\label{determinan}
&\tfrac{1}{\kappa^{2}}{\rm det}(\bm{H})=\Big(\tfrac{\partial^{2}}{\partial\tau^{2}}\tfrac{1}{\kappa^{2}}\|\bm{q}(\bm{r})\|_2^{2}\Big)\Big(\tfrac{\partial^{2}}{\partial\nu^{2}}\tfrac{1}{\kappa^{2}}\|\bm{q}(\bm{r})\|_2^{2}\Big)\nonumber\\
&-\Big(\tfrac{\partial^2}{\partial\tau\partial\nu}\tfrac{1}{\kappa^{2}}\|\bm{q}(\bm{r})\|_2^{2}\Big)^{2} > 0.\end{aligned}$$ To find $\|\bm{q}(\bm{r})\|_2$ and its derivatives, we need to borrow some bounds from [@candes2014towards Section, C.2] which states that $\forall \bm{r}\in \Omega_{{\rm close}}$ and $N\ge 512$, we have $$\begin{aligned}
\overline{G}(\bm{r})\ge 0.8113,\quad | \overline{G}^{(1,0)}(\bm{r})|\le 0.8113,\nonumber\\
\overline{G}^{(2,0)}(\bm{r})\le -2.097N^{2},\quad | \overline{G}^{(1,1)}(\bm{r})|\le 0.6531N,\nonumber\\
| \overline{G}^{(2,1)}(\bm{r})|\le 2.669N^{2},\quad | \overline{G}^{(3,0)}(\bm{r})|\le 8.070N^{3}.\end{aligned}$$ Introduce $$\begin{aligned}
\overline{Z}^{(m',n')}(\bm{r}):=\sum_{\bm{r}_j\in \mathcal{S} \setminus {{0}}}| \overline{G}^{(m',n')}(\bm{r}-\bm{r}_j)|.\end{aligned}$$ Again, based on [@candes2014towards Table C.1], it holds that $$\begin{aligned}
\overline{Z}^{(0,0)}(\bm{r}) \le 6.405\times 10^{-2},\quad \overline{Z}^{(1,0)}(\bm{r}) \le 0.1047N,\nonumber\\
\overline{Z}^{(2,0)}(\bm{r}) \le0.4019N,\quad \overline{Z}^{(1,1)}(\bm{r})\le 0.1642N^{2},\nonumber\\
\overline{Z}^{(2,1)}(\bm{r}) \le 0675N^{2},\quad \overline{Z}^{(3,0)}(\bm{r})\le 1.574N^{3},\end{aligned}$$ and we also have [@candes2014towards Section C.1] $$\begin{aligned}
\| \overline{\bm{\alpha}}_j\|_{2}\le \alpha_{\max}=1+5.577\times 10^{-2}\nonumber\\
\| \overline{\bm{\alpha}}_j\|_{2} \ge \alpha_{\min}=1-5.577\times 10^{-2}\nonumber\\
\| \overline{\bm{\beta}}_j\|_{2}\le \beta_{\max}=\tfrac{2.93}{N}\times 10^{-2}\nonumber\\
\| \overline{\bm{\gamma}}_j\|_{2}\le \gamma_{\max}=\tfrac{2.93}{N}\times 10^{-2}.\end{aligned}$$ We use the aforementioned formulas to obtain the bounds $$\begin{aligned}
&\| \overline{\bm{q}}(\bm{r})\|_2=\|\sum_{j=1}^{s} \overline{G}^{(0,0)}(\bm{r}-\bm{r}_j) \overline{\bm{\alpha}}_j + \overline{G}^{(1,0)} (\bm{r}-\bm{r}_j) \overline{\bm{\beta}}_j
\nonumber\\
&+ \overline{G}^{(0,1)}(\bm{r}-\bm{r}_j) \overline{\bm{\gamma}}_j\|_2\le \alpha_{\max}\Big(| \overline{G}^{(0,0)}(\bm{r})|+ \overline{Z}^{(0,0)}(\bm{r})\Big)\nonumber\\
&+ 2\beta_{\max}\Big(| \overline{G}^{(1,0)}(\bm{r})|+ \overline{Z}^{(1,0)}(\bm{r})\Big)\le 1.295 + \tfrac{0.0475}{N}.\end{aligned}$$ For the derivatives of $\bm{q}$, we have: $$\begin{aligned}
\label{q(1,0)}
&\| \overline{\bm{q}}^{(1,0)}\|_2\le \alpha_{\max}\Big(| \overline{G}^{(1,0)}(\bm{r})|+ \overline{Z}^{(1,0)}(\bm{r})\Big)\nonumber\\
&+ \beta_{\max}\Big(| \overline{G}^{(2,0)}(\bm{r})|+ \overline{Z}^{(2,0)}(\bm{r})\Big)+\gamma_{\max}\Big(| \overline{G}^{(1,1)}(\bm{r})|\nonumber\\
&+ \overline{Z}^{(1,1)}(\bm{r})\Big)\le 0.08874 +0.2148N. \end{aligned}$$ Other derivatives can be obtained using similar steps as follows: $$\begin{aligned}
\label{bound}
&\| \overline{\bm{q}}^{(1,1)}\|_2\le 0.846N +0.213N^{2},\nonumber\\
&\| \overline{\bm{q}}^{(2,0)}\|_2\le 0.5025N +3.8845N^{2}.\end{aligned}$$ Now, we proceed by writing $$\begin{aligned}
\label{80}
&\tfrac{\partial^{2}}{\partial\tau^{2}}\|\tfrac{1}{\kappa}\bm{q}(\bm{r})\|^{2}_2=\tfrac{\partial}{\partial\tau}\tfrac{1}{\kappa^{2}}\langle \bm{q}^{(1,0)}(\bm{r}),\bm{q}(\bm{r})\rangle=2\| \tfrac{1}{\kappa}\bm{q}^{(1,0)}(\bm{r})\|^{2}_{2}\nonumber\\
&+ \tfrac{2}{{\kappa}^{2}}{\rm Re}\Big[\Big(\bm{q}^{(2,0)}(\bm{r})\Big)^{\rm H}\bm{q}(\bm{r})\Big],\end{aligned}$$ where the first term can be bounded as $$\begin{aligned}
\label{81}
&\| \tfrac{1}{\kappa}\bm{q}^{(1,0)}(\bm{r})\|^{2}_{2}\le \| \tfrac{1}{\kappa}\Big(\bm{q}^{(1,0)}(\bm{r})- \overline{\bm{q}}^{(1,0)}(\bm{r})\Big)\|^{2}_{2}\nonumber\\
&+\| \tfrac{1}{\kappa} \overline{\bm{q}}^{(1,0)}(\bm{r})\|^{2}_{2}\le \epsilon^{2}+0.0141.\end{aligned}$$ The first inequality above comes from the triangular inequality while the last one is based on Lemma \[lem.nearness\_all\_r\], \[q(1,0)\] and the fact that ${\kappa}^{2} \ge \tfrac{\pi^{2}}{3}N^{2}$. The second term in can be bounded by: $$\begin{aligned}
\label{82}
&\tfrac{1}{{\kappa}^{2}}{\rm Re}\Big[\Big(\bm{q}^{(2,0)}(\bm{r})\Big)^{\rm H}\bm{q}(\bm{r})\Big]={\rm Re}\Big[\tfrac{2}{{\kappa}^{2}}\Big(\bm{q}^{(2,0)}(\bm{r})\nonumber\\
&- \overline{\bm{q}}^{(2,0)}(\bm{r})+ \overline{\bm{q}}^{(2,0)}(\bm{r})\Big)^{\rm H}(\bm{q}(\bm{r})- \overline{\bm{q}}(\bm{r})+ \overline{\bm{q}}(\bm{r}))\Big]=\nonumber\\
&{\rm Re}\Big[\tfrac{1}{{\kappa}^{2}}\Big(\bm{q}^{(2,0)}(\bm{r})- \overline{\bm{q}}^{(2,0)}(\bm{r})\Big)^{\rm H}(\bm{q}(\bm{r})- \overline{\bm{q}}(\bm{r}))\Big]\nonumber\\
&+{\rm Re}\Big[\tfrac{1}{{\kappa}^{2}}\Big( \overline{\bm{q}}^{(2,0)}(\bm{r})\Big)^{\rm H} \overline{\bm{q}}(\bm{r})\Big]\nonumber\\
&+{\rm Re}\Big[\tfrac{1}{{\kappa}^{2}}\Big(\bm{q}^{(2,0)}(\bm{r})- \overline{\bm{q}}^{(2,0)}(\bm{r})\Big)^{\rm H} \overline{\bm{q}}(\bm{r})\Big]\nonumber\\
&+{\rm Re}\Big[\tfrac{1}{{\kappa}^{2}}\Big( \overline{\bm{q}}^{(2,0)}(\bm{r})\Big)^{\rm H}(\bm{q}(\bm{r})- \overline{\bm{q}}(\bm{r}))\Big]\nonumber\\
&\le \epsilon^{2}-0.307+1.129\epsilon +1.181\epsilon\nonumber\\
&\le \epsilon^{2}+ 2.31\epsilon - 0.307.\end{aligned}$$ Substituting and into , yields to $$\begin{aligned}
\label{86}
\tfrac{1}{{\kappa}^{2}} {\rm tr}(\bm{H})\le 8\epsilon^{2}+ 9.24\epsilon - 1.1712.\end{aligned}$$ It is straightforward to verify that the above term is negative by setting $\epsilon\le 0.1$. Next, we prove . The second term in can be written as $$\begin{aligned}
\label{87}
&\tfrac{\partial}{\partial\tau \partial\nu}\|\tfrac{1}{\kappa} \bm{q}(\bm{r})\|_2^{2}=\tfrac{2}{{\kappa}^{2}}\langle \bm{q}^{(1,0)}(\bm{r}),\bm{q}^{(0,1)}(\bm{r}) \rangle\nonumber\\
&+\tfrac{2}{{\kappa}^{2}}\langle \bm{q}^{(1,1)}(\bm{r}),\bm{q}(\bm{r}) \rangle.\end{aligned}$$ The upper-bound of the first term in is obtained by $$\begin{aligned}
\label{88}
&\tfrac{1}{{\kappa}^{2}}\langle \bm{q}^{(1,0)}(\bm{r}),\bm{q}^{(0,1)}(\bm{r}) \rangle={\rm Re} \Big[\tfrac{1}{{\kappa}^{2}}\Big(\bm{q}^{(1,0)}(\bm{r})- \overline{\bm{q}}^{(1,0)}(\bm{r})\Big)^{\rm H}\nonumber\\
&\Big(\bm{q}^{(0,1)}(\bm{r})- \overline{\bm{q}}^{(0,1)}(\bm{r})\Big) \Big]+{\rm Re} \Big[\tfrac{1}{{\kappa}^{2}}\Big( \overline{\bm{q}}^{(1,0)}(\bm{r})\Big)^{\rm H} \overline{\bm{q}}^{(0,1)}(\bm{r}) \Big]\nonumber\\
&+{\rm Re}\Big[\tfrac{1}{{\kappa}^{2}}\Big(\bm{q}^{(1,0)}(\bm{r})- \overline{\bm{q}}^{(1,0)}(\bm{r})\Big)^{\rm H} \overline{\bm{q}}^{(0,1)}(\bm{r}) \Big]\nonumber\\
&+{\rm Re}\Big[\tfrac{1}{{\kappa}^{2}}\Big( \overline{\bm{q}}^{(1,0)}(\bm{r})\Big)^{\rm H}\Big(\bm{q}^{(0,1)}(\bm{r})- \overline{\bm{q}}^{(0,1)}(\bm{r})\Big) \Big]\nonumber\\
&\le \epsilon^{2} + 0.238\epsilon +0.0736,\end{aligned}$$ where the last inequality follows from , and the fact that ${\kappa}^{2} \ge \tfrac{\pi^{2}}{3}N^{2}$. By using similar steps as in , we reach $$\begin{aligned}
\label{89}
\tfrac{1}{\kappa^{2}}\langle \bm{q}^{(1,1)}(\bm{r}),\bm{q}(\bm{r})\rangle \le \epsilon^{2} +0.195\epsilon +0.0736.\end{aligned}$$ Substituting and into yields to $$\begin{aligned}
\label{moshtagh2bodi}
\tfrac{\partial^2}{\partial\tau \partial\nu}\|\tfrac{1}{{\kappa}}\bm{q}(\bm{r})\|_2^{2}\le 4\epsilon^{2}+ 2.865\epsilon +0.175.\end{aligned}$$ By using the bound obtained for and and setting $\epsilon=0.05$, is satisfied. Finally, based on Lemmas \[omega-far\] and \[lem.near\], we can show that $\|\bm{q}(\bm{r})\|_2 \le 1, \forall \bm{r}\in [0,1]^{2}\ \setminus \mathcal{S}$.
Proof of Lemma \[4\] {#proof.of.lemma2}
====================
Set $\epsilon=2.5ab$ and $\Delta \bm{w}^{(m,n)}:=\bm{w}^{(m,n)}(\bm{r})-\overline{\bm{w}}^{(m,n)}(\bm{r})$. For all $a,b\ge 0$ we have $$\begin{aligned}
\label{eq.rel4}
&\mathds{P}[\underset{\mathbf{ r} \in \Omega}{\max}\| \bm{I}_1^{(m,n)}(\bm{r})\|_2 \ge 2.5ab]=\mathds{P}\Big[\underset{\bm{r}\in\Omega}{\max}\|(\Delta \bm{w}^{(m,n)})^{\text{H}} \mathbf{L} \bm{\Phi}\|_2 \nonumber \\
&\ge 2.5ab\Big] \le \mathds{P}\Big[\underset{\bm{r} \in \Omega }{\bigcup}\Big\{\|(\Delta \bm{w}^{(m,n)})^{\text{H}} \mathbf{L} \bm{\Phi}\|_2 \ge \| \mathbf{L}^{\text{H}}(\Delta \bm{w}^{(m,n)}) \|_2b\Big\}\nonumber\\
& \cup \Big\{\|\mathbf{L}^{\text{H}}(\Delta \bm{w}^{(m,n)})\|_2\ge 2.5a\big\} \Big]\le \mathds{P}\Big[\underset{r\in\Omega}{\bigcup}\Big\{\|(\Delta \bm{w}^{(m,n)})^{\text{H}} \mathbf{L} \bm{\Phi}\|_2 \nonumber\\
&\ge \| \mathbf{L}^{\text{H}}\Delta \bm{w}^{(m,n)} \|_2b\Big\}\cup\Big\{\|\Delta \bm{w}^{(m,n)}\|_2\ge a \Big\}\cup \Big\{\|\mathbf{L}\|\ge2.5\Big\}\Big]\nonumber\\
&\le \mathds{P}[\|\mathbf{L}\|\ge2.5]+\sum_{\bm{r} \in \Omega}\Big(\mathds{P}\Big[\|(\Delta \bm{w}^{(m,n)})^{\text{H}} \mathbf{L} \bm{\Phi}\|_2\nonumber\\
&\stackrel{({\textup{\uppercase\expandafter{\romannumeral1}}})}{\ge} \| \mathbf{L}^{\text{H}}(\Delta \bm{w}^{(m,n)}) \|_2b\Big]+\mathds{P}\Big[\|\Delta \bm{w}^{(m,n)}\|_2\ge a\Big]\Big)\nonumber\\
&\le \mathds{P}[ \overline{\zeta}_{\tfrac{1}{4}}]+(R+1)|\Omega|e^{-\tfrac{b^2}{8}}+\mathds{P}\Big[\|\Delta \bm{w}^{(m,n)}\|_2\ge a\Big]\nonumber\\
&\le \mathds{P}[ \overline{\zeta}_{\tfrac{1}{4}}]+\tfrac{\delta}{2}+\mathds{P}\Big[\|\Delta \bm{w}^{(m,n)}\|_2\ge a \Big],\end{aligned}$$ where $({\textup{\uppercase\expandafter{\romannumeral1}}})$ comes from the Hoeffding’s inequality given below.
[@yang2016exact](Heoffding’s inequality for vectors) Consider the rows of $\bm{\Phi} \in \mathbb{C}^{s\times R}$ be i.i.d. on the complex hyper-sphere $\mathbb{S}^{R-1}$ with zero mean. Then, for all $\mathbf{L}^{\text{H}}\Delta \bm{w}^{(m,n)} \in \mathbb{C}^{s},~\mathbf{L}^{\text{H}}\Delta \bm{w}^{(m,n)}\ne \bm{0}$ and $b\ge0$, $$\begin{aligned}
\mathds{P}(\|(\Delta \bm{w}^{(m,n)})^{\text{H}} \mathbf{L} \bm{\Phi}\|_2\ge \| \mathbf{L}^{\text{H}}\Delta \bm{w}^{(m,n)} \|_2b)\le(R+1) e^{\tfrac{-b^{2}}{8}}.
\end{aligned}$$ By using $(R+1)|\Omega|e^{\tfrac{-b^{2}}{8}}\le\tfrac{\delta}{2}$, we obtain $b=\sqrt{8 \log2(R+1)\tfrac{|\Omega|}{\delta}}$ .
By using [@heckel2016super Section 8.3.1] and choosing $\mathds{P}[\|\Delta \bm{w}^{(m,n)}\|_2 \ge a]\le \tfrac{\delta}{2}$ and $a=\sqrt{\tfrac{3s}{L}}12^{\tfrac{3}{2}}c_1(\tfrac{c^{2}_2}{c})$, we proceed as follows: $$\begin{aligned}
&\mathds{P}\Big[\underset{\bm{r}\in \Omega}{\text{max}}\|\bm{I}_1^{(m,n)}(\bm{r})\|_2\ge 360c_1(\tfrac{c_2^{2}}{\sqrt{c}}) \sqrt{\tfrac{s}{L}}\nonumber\\
&\log\Big(\tfrac{12s|\Omega|}{\delta}\Big)\sqrt{\log\Big(\tfrac{2(R+1)|\Omega|}{\delta}\Big)}\Big]\le \delta+\mathds{P}(\overline{\zeta}_{\frac{1}{4}})=2\delta\end{aligned}$$
Proof of Lemma \[lem5\]: {#proof.of.lemma3}
========================
By the union bound: $$\begin{aligned}
&\mathds{P}\Big[\underset{\bm{r}\in\Omega}{\text{max}}\|\bm{I}_2^{(m,n)}(\bm{r})\|_2\ge \epsilon~ \big|~ \zeta_\xi\Big]\nonumber\\
&\le\sum_{\bm{r}\in\Omega}\mathds{P}\Big[\|( \overline{\bm{w}}^{(m,n)}(\bm{r}))^{\text{H}}(\mathbf{L}- \overline{\mathbf{L}})\bm{\Phi} \|_2\ge \epsilon~ \big|~ \zeta_\xi\Big]\nonumber\\
&\le\sum_{\bm{r}\in\Omega}\mathds{P}\Big[\|( \overline{\bm{w}}^{(m,n)}(\bm{r}))^{\text{H}}(\mathbf{L}- \overline{\mathbf{L}})\bm{\Phi}\|_2\nonumber\\& \ge \|(\mathbf{L}- \overline{\mathbf{L}})^{\text{H}} \overline{\bm{w}}^{(m,n)}(\bm{r})\|_2\tfrac{\epsilon}{(c_2\xi)} \Big]\le |\Omega|(R+1)e^{-\tfrac{(\tfrac{\epsilon}{(c_2\xi)})^{2}}{8}}\le\delta.\end{aligned}$$
Proof of {#proof.equation_norm_q}
---------
We first find an upper-bound for $\|\bm{q}^{(m,n)}(\bm{r})\|_2$ as follows: $$\begin{aligned}
\label{bound_q}
&\tfrac{1}{\kappa^{m+n}}\|\bm{q}^{(m,n)}(\bm{r})\|_2=\|(\bm{w}^{(m,n)})^{\rm H}\mathbf{L}\bm{\Phi}\|_2\le\|\mathbf{L}\| \|\bm{\Phi}\|\|\bm{w}^{(m,n)}\|_2\nonumber\\
&\le\|\mathbf{L}\| \|\bm{\Phi}\|_F\|\bm{w}^{(m,n)}\|_2\le\|\mathbf{L}\| \sqrt{s} \sqrt{3s} \|\bm{w}^{(m,n)}\|_\infty \nonumber\\
& \le\|\mathbf{L}\|s \sqrt{3}\underset{j,({m}',{n}')\in \{(0,0,),(1,0),(0,1)\}}{\max} \tfrac{|G_{({m}',{n}')}^{(m,n)}(\bm{r},{{\bm{r}}_{j}})|}{\kappa^{m+{m}'+{n}'+n}},\end{aligned}$$ where we used the fact that $\|\bm{\Phi}\|\le \|\bm{\Phi}\|_F$ and for all $\bm{r}$ and all $\bm{r}_j$ we have [@heckel2016super Equation 8.66]: $$\begin{aligned}
\label{bound_G}
\tfrac{|G_{({m}',{n}')}^{(m,n)}(\bm{r},{{\bm{r}}_{j}})|}{\kappa^{m+{m}'+{n}'+n}}\le c_1 12^{\tfrac{3}{2}}\sqrt{L}\|\bm{x}\|^{2}_2.
$$ By replacing into and using $s\le L$, we have $\tfrac{\|\bm{q}^{(m,n)}(\bm{r})\|_2}{\kappa^{m+n}}\le 72 c_1L^{\tfrac{3}{2}}\|\mathbf{L}\|\|\bm{x}\|^{2}_2$. Thus, by taking $\tfrac{\tilde{c}}{2}=(2.5)\cdotp(3)\cdotp(72)c_1$, $$\begin{aligned}
&\mathds{P}\Big[\underset{\bm{r}\in[0,1]^{2},m+n\le2}{\max}\|\bm{q}^{(m,n)}(\bm{r})\|_2\ge \tfrac{\tilde{c}}{2} L^{\tfrac{3}{2}}\Big]\nonumber\\
&\le \mathds{P}[\|\mathbf{L}\|\|\bm{x}\|^{2}_2\ge(2.5)\cdotp(3)]\nonumber\\
&\le \mathds{P}[\|\mathbf{L}\|\ge2.5]+\mathds{P}[\|\bm{x}\|^{2}_2\ge 3]\le \tfrac{\delta}{2},\end{aligned}$$ where we used the fact that $\mathds{P}[\|\bm{x}\|^{2}_2\ge 3] \le \tfrac{\delta}{4}$ [@heckel2016super Equation 8.69]. The last inequality follows from $\mathds{P}[\|\mathbf{L}\|\ge 2.5]\le \mathds{P}[ \overline{\zeta}_{\tfrac{1}{4}}]\le\tfrac{\delta}{4}$ (from Lemma \[lem.L\_bounds\]).
The reason that $\mathcal{E}_1$, $\mathcal{E}_2$ imply {#proof.whyimply}
=======================================================
Let $\bm{r}_g$ be the closest points in $\Omega$ to $\bm{r}$ with respect to $\ell_{\infty}$-measure. By the triangle inequality: $$\begin{aligned}
\label{64}
& \tfrac{1}{\kappa^{m+n}}\|\bm{q}^{(m,n)}(\bm{r})- \overline{\bm{q}}^{(m,n)}(\bm{r})\|_2\nonumber\\
&\le \tfrac{1}{\kappa^{m+n}}\Big[ \|\bm{q}^{(m,n)}(\bm{r})-{\bm{q}}^{(m,n)}(\bm{r}_g)\|_2 \nonumber\\
&+ \|\bm{q}^{(m,n)}(\bm{r}_g)- \overline{\bm{q}}^{(m,n)}(\bm{r}_g)\|_2\nonumber\\
& +\| \overline{\bm{q}}^{(m,n)}(\bm{r}_g)- \overline{\bm{q}}^{(m,n)}(\bm{r})\|_2\Big].
\end{aligned}$$ Next, we obtain upper-bounds for the composing terms of the above relation, separately. For the first term, we have: $$\begin{aligned}
\label{65}
&\|\bm{q}^{(m,n)}(\bm{r})-{\bm{q}}^{(m,n)}(\bm{r}_g)\|_2\le \sqrt{R}~\underset{i}{\max}|\bm{q}^{(m,n)}(\bm{r})-\nonumber\\
&{\bm{q}}^{(m,n)}(\bm{r}_g)|_i.\end{aligned}$$ We proceed by writing $$\begin{aligned}
\label{bernestein}
&|\bm{q}^{(m,n)}(\bm{r})-{\bm{q}}^{(m,n)}(\bm{r}_g)|_i=|\bm{q}^{(m,n)}(\tau,\nu)-\bm{q}^{(m,n)}(\tau,\nu_g)\nonumber\\
&+\bm{q}^{(m,n)}(\tau,\nu_g)-\bm{q}^{(m,n)}(\tau_g,\nu_g)|_i\nonumber\\
&\le|\bm{q}^{(m,n)}(\tau,\nu)-\bm{q}^{(m,n)}(\tau,\nu_g)|_i\nonumber\\
&+|\bm{q}^{(m,n)}(\tau,\nu_g)-\bm{q}^{(m,n)}(\tau_g,\nu_g)|_i\nonumber\\
&\le|\nu-\nu_g|\underset{z}{\sup}|\bm{q}^{(m,n+1)}(\tau,z)|_i\nonumber\\ &+|\tau-\tau_g|\underset{z}{\sup}|\bm{q}^{(m+1,n)}(z,\nu_g)|_i\nonumber\\
&\le |\nu-\nu_g|2\pi N \underset{z}{\sup}\|\bm{q}^{(m,n)}(\tau,z)\|_2 \nonumber\\
&+ |\tau-\tau_g|2\pi N \underset{z}{\sup}\|\bm{q}^{(m,n)}(z,\tau)\|_2,\end{aligned}$$ where in the last step, we used Bernstein’s polynomial inequality [@harris1996bernstein Corollary 8]. Substituting into , we reach $$\begin{aligned}
\label{ber2}
&\tfrac{1}{\kappa^{m+n}}\|\bm{q}^{(m,n)}(\bm{r})-\bm{q}^{(m,n)}(\bm{r}_g)\|_2 \nonumber\\
&\le \tfrac{\tilde{c}}{2}L^{\tfrac{5}{2}}R^{\tfrac{1}{2}}(|\tau -\tau_g|+|\nu-\nu_g|)\nonumber\\
&\le \tilde{c}L^{\tfrac{5}{2}}R^{\tfrac{1}{2}}\|\bm{r}-\bm{r}_g\|_{\infty}\le \tfrac{\epsilon}{3}.\end{aligned}$$ Similarly, we have $$\begin{aligned}
\label{bound_q_ber}
\tfrac{1}{\kappa^{m+n}}\| \overline{\bm{q}}^{(m,n)}(\bm{r}_g)- \overline{\bm{q}}^{(m,n)}(\bm{r})\|_2\le\tfrac{\epsilon}{3}.\end{aligned}$$
Substituting , , into leads to $$\begin{aligned}
\tfrac{1}{\kappa^{m+n}}\|\bm{q}^{(m,n)}(\bm{r})- \overline{\bm{q}}^{(m,n)}(\bm{r})\|_2\le \epsilon\end{aligned}$$ for all $(m,n):m+n\le 2$ and for all $\bm{r}\in [0,1]^{2}$.
[^1]: M. Safari, S. Daei and F. Haddadi are with the School of Electrical Engineering, Iran University of Science & Technology.
[^2]: See [@heckel2016super Appendix H] for a detailed discussion.
[^3]: See e.g. [@daei2019exploiting; @daei2019distribution].
[^4]: Throughout, we occasionally use time-frequency shifts instead of delay-Doppler shifts.
[^5]: A function that promotes the number of blocks with non-zero $\ell_2$ norm.
[^6]: Without loss of generality, we assume that $L$ is an odd integer.
[^7]: See [@heckel2016super Section 6] and [@dumitrescu2017positive Remark 3.6] for a detailed discussion.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'Marcus Thelander Andrén, Bo Bernhardsson, Anton Cervin and Kristian Soltesz[^1]'
bibliography:
- 'paper.bib'
title: '**On Event-Based Sampling for LQG-Optimal Control** '
---
[^1]: This work has been supported by the Swedish Research Council. The authors are with the Department of Automatic Control, Lund University, Sweden, and are members of the LCCC Linnaeus Center and the ELLIIT Excellence Center at Lund University. [e-mail: {marcus, bob, anton, kristian}@control.lth.se]{}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Markov chain Monte Carlo (MCMC) sampling is an important and commonly used tool for the analysis of hierarchical models. Nevertheless, practitioners generally have two options for MCMC: utilize existing software that generates a black-box “one size fits all" algorithm, or the challenging (and time consuming) task of implementing a problem-specific MCMC algorithm. Either choice may result in inefficient sampling, and hence researchers have become accustomed to MCMC runtimes on the order of days (or longer) for large models. We propose an automated procedure to determine an efficient MCMC algorithm for a given model and computing platform. Our procedure dynamically determines blocks of parameters for joint sampling that result in efficient sampling of the entire model. We test this procedure using a diverse suite of example models, and observe non-trivial improvements in MCMC efficiency for many models. Our procedure is the first attempt at such, and may be generalized to a broader space of MCMC algorithms. Our results suggest that substantive improvements in MCMC efficiency may be practically realized using our automated blocking procedure, or variants thereof, which warrants additional study and application.'
author:
- |
Daniel Turek$^{*1,2}$, Perry de Valpine$^{2}$,\
Christopher J. Paciorek$^{1}$, Clifford Anderson-Bergman$^{1,2}$
bibliography:
- '/Users/dturek/Dropbox/Berkeley/References/ZoteroLibrary.bib'
title: 'Automated Parameter Blocking for Efficient Markov-Chain Monte Carlo Sampling'
---
0
[0]{}
$^*$Corresponding Author\
[email protected]\
\
$^1$Department of Statistics\
University of California, Berkeley\
\
$^2$Department of Environmental Science, Policy, and Management\
University of California, Berkeley\
1
[0]{}
Automated Parameter Blocking for Efficient Markov-Chain Monte Carlo Sampling
**Keywords:**\
MCMC, Metropolis-Hastings, Block sampling, Integrated autocorrelation time, Mixing.
Introduction {#sec:introduction}
============
Markov chain Monte Carlo (MCMC) has become a core computational method for many statistical analyses. These include routine Bayesian analyses, but also hybrid algorithms that use MCMC as one component, such as Monte Carlo Expectation Maximization [MCEM; @Caffo2005] or data cloning [@Lele2007]. Nevertheless, the automated generation of black-box MCMC algorithms, as occurs in generally available software, does not necessarily result in efficient MCMC sampling. Analysts are thereby accustomed to MCMC run times measured in minutes, hours or even days for large hierarchical models. Computation time is frequently the limiting factor, either limiting the range of models considered, or limiting the potential for performing diagnostics and comparisons using methods such as bootstrapping [@Efron1994], cross validation [@Gneiting2007], or calibration of posterior predictive p-values [@Hjort2006], among others. Therefore, any widely applicable improvements to MCMC performance may greatly improve the practical analyses of large hierarchical models.
Among the many MCMC sampling algorithms developed to improve MCMC efficiency, one of the most basic approaches has been block sampling: jointly updating multiple dimensions of a target distribution simultaneously [@Roberts1997; @Sargent2000]. When one or more dimensions of the posterior distribution are correlated, joint sampling of these dimensions (with any variety of block samplers) can increase sampling performance relative to updating each dimension independently [*e.g.,* @Liu1994]. Despite wide recognition of the usefulness of this basic idea for designing efficient MCMC algorithms, there has been no automated method for choosing blocks to optimize – or at least greatly improve – performance. Here we develop such a method.
Existing theoretical work comparing block samplers to univariate samplers [@Mengersen1996; @Roberts1996a; @Roberts1997a among others] has provided many insights but falls short of providing a complete assessment of MCMC efficiency for several reasons. First, it uses MCMC convergence rates as the metric for comparison, without consideration of the computational demands of block sampling. Instead, our viewpoint is that any measure of MCMC efficiency must incorporate both the convergence rate and the computational requirements of achieving improvements in convergence rate. This may give a different picture of the actual efficiency of a sampling algorithm. Accelerated convergence at an extreme computational cost is obviously not optimal. Second, the computational requirements of different steps will vary greatly across platforms, depending on such factors as processor, memory architecture, use of efficient linear algebra packages, etc. Therefore, even if theoretical comparisons were extended to incorporate aspects of computation, the best block sampling scheme would remain platform-dependent. It is important to recognize that computational costs affect not only the proposal step – such as the cost of generating a multivariate normal proposal – but also model computations and density evaluations. Some parts of hierarchical models may inherently involve expensive computations, which can impact the relative efficiency of different blocking schemes. Third, existing theories and methods presume that some wise, manual selection of blocks may be feasible, based for example on an understanding of the model structure, which leads to understanding which posterior dimensions may be correlated. In general, however, it is difficult to know *a priori* which dimensions will be correlated, which is one purpose of automating a procedure like MCMC in the first place.
Here, we present a procedure for the automated exploration of MCMC blocking schemes, seeking a highly efficient MCMC algorithm specific to the hierarchical model and computing environment at hand. This represents a higher level of automated algorithm generation than is provided by existing software, which serve to produce “one size fits all" MCMC algorithms. The family of BUGS packages [WinBUGS, JAGS, and OpenBUGS; @Lunn2000; @Plummer2011; @Lunn2012] assigns samplers based on local characteristics of each model parameter, using a combination of Gibbs sampling, adaptive rejection sampling, slice sampling, and, in limited cases, block sampling. Other MCMC packages including ADMB [@Skaug2006] and Stan [@StanDevelopmentTeam2014] use Hamiltonian MCMC sampling [@Neal2011], which may generally be more efficient but nevertheless represents a static approach to MCMC algorithm generation. Yet other promising methods such as Langevin sampling [@Marshall2012] are not incorporated into software commonly used by practitioners. For simplicity, we restrict our attention to univariate and blocked adaptive random walk sampling. However, the main concept of exploring the space of parameter blocks to improve MCMC efficiency generalizes to allow the use of other sampling methods.
In section \[sec:efficiency\], we examine the pros and cons of univariate versus block random walk sampling, both in terms of algorithmic and computational efficiencies. From these considerations we conclude that the combination of samplers that yields optimal MCMC efficiency (defined as an MCMC algorithm’s rate of generating effectively independent samples) will be model- and platform-specific. Section \[sec:autoblock\] introduces a procedure for automated blocking of hierarchical model parameters, designed to maximize the resulting MCMC algorithm efficiency. The main idea of this procedure is to iteratively cluster model parameters based upon empirical posterior correlations, then intelligently subdivide the hierarchical clustering tree (a dendrogram) to determine blockings of parameters that result in efficient MCMC sampling. A series of simulated and real data examples given in section \[sec:performance\] demonstrate that this procedure can improve MCMC efficiency many-fold over statically defined MCMC algorithms and can dynamically generate algorithms comparable in performance to model-specific, hand-tuned algorithms. We close with a discussion in section \[sec:discussion\].
MCMC Algorithms: Definitions and Efficiency {#sec:efficiency}
===========================================
In this section, we first define a space of valid MCMC algorithms. Then, we examine two dominant contributors to the efficiency of an MCMC algorithm: the algorithmic capability to produce effectively-independent samples from the target distribution, and the computational demands of the algorithm in generating MCMC samples; these are composed to form our metric of overall MCMC efficiency. Drawing upon existing asymptotic theory of MCMC sampling, the scaling of computational costs of different sampling schemes, and on several illustrative examples, we argue that achieving an optimally efficient MCMC algorithm for a specific model by pure theory is prohibitively difficult. That conclusion motivates our approach of computationally optimizing – or at least greatly improving – MCMC performance through automated exploration of a space of valid MCMC algorithms.
MCMC Definition {#mcmcdef}
---------------
We assume a given, fixed, hierarchical model $\mathcal{M}$, which may be represented as a directed acyclic graph where vertices represent top-level model parameters, latent states, or fixed observations (data), and edges represent dependencies between these components. Denote the set of all top-level model parameters and latent states (the unknown components for which we may seek inferences) as $\Theta = \{ \theta_1, \ldots, \theta_d \}$, which will be referred to as the parameters of $\mathcal{M}$.
An MCMC algorithm may be defined in terms of its sampling scheme over $\Theta$. Let $b$ be any non-empty subset (“block") of $\Theta$, and $u \in U$ be any valid MCMC sampling (or “updating") method such as slice sampling or conjugate Gibbs sampling [see @Gilks2005 for a broad overview of MCMC sampling methods]. We define a valid MCMC sampler $\psi = u(b)$ as the application of $u$ to $b$, where $b$ satisfies any assumptions of $u$ (*e.g.,* conjugacy). In addition to satisfying standard properties of Markov chains [see, for example, @Robert2004], we define a valid MCMC algorithm as any set of samplers $\Psi = \{ \psi_1, \ldots, \psi_k \}$, where $\psi_i = u_i(b_i)$ for $i = 1, \ldots, k$, satisfying $\cup_{i=1}^k b_i = \Theta$; that is, the MCMC algorithm updates each model parameter in at least one sampler. We represent the chain of samples generated from successive applications of $\Psi$ as $X^{(0)}, X^{(1)}, \ldots$, where sample $X^{(i)}$ implies model state $\Theta = X^{(i)}$, $X^{(0)}$ is the set of initial values, $X^{(i)} = (X^{(i)}_1, \ldots, X^{(i)}_d)$, and $X_k = \{ X^{(0)}_k, X^{(1)}_k, \ldots \}$ is the scalar chain of samples of $\theta_k$ (for $k = 1, \ldots, d$).
This paper focuses attention on the restricted set of sampling methods $U_0 = \{ u_\text{scalar}, u_\text{block} \}$, where $u_\text{scalar}$ denotes univariate adaptive random walk Metropolis-Hastings sampling [hereafter, scalar sampling; @Metropolis1953; @Hastings1970], and $u_\text{block}$ denotes the multivariate generalization of this algorithm [hereafter, block sampling; @Haario1999], with the practical restriction that any $\psi = u_\text{block}(b)$ satisfies $| b | > 1$. The $u_\text{scalar}$ algorithm adaptively tunes the proposal scale, while $u_\text{block}$ additionally tunes the proposal covariance [@Roberts2009]. All scalar and block samplers asymptotically achieve the theoretically optimal scaling of proposal distributions (and therefore acceptance rates, and mixing) as derived in @Roberts1997a, and implement adaptation routines as set out in @Shaby2011.
For hierarchical model $\mathcal{M}$ with parameters $\Theta$, our studies focus almost exclusively on the set of MCMC algorithms $\boldsymbol{\Psi}_\mathcal{M}$, which contains all algorithms of the form $\Psi = \{ \psi_1, \ldots, \psi_k \}$, where $\psi_i = u_i(b_i)$, each $u_i \in U_0$, and the sets $b_i$ form a partition of $\Theta$. We specifically name two algorithms in $\boldsymbol{\Psi}_\mathcal{M}$ which are boundary cases. The first consists of $d$ scalar samplers: $\Psi_\text{scalar} = \{ \psi_1, \ldots, \psi_d \}$, where each $\psi_i = u_\text{scalar}(\theta_i)$. The second has a single block sampler for all parameters: $\Psi_\text{block} = \{ u_\text{block}(\Theta) \}$. Implicit is the assumption that each $\theta_i$ is continuous-valued, which is the case throughout this paper.
Algorithmic Efficiency {#sec:algorithmiceff}
----------------------
We first consider MCMC algorithmic efficiency, independent of any computational requirements. This measure of efficiency solely represents the best mixing, or equivalently the least autocorrelation, or the highest effective sample size, without consideration for the computational (time) requirements of generating a set of samples. After reviewing the definition of MCMC algorithmic efficiency which is based upon integrated autocorrelation time, we study the use of $\Psi_\text{scalar}$ or $\Psi_\text{block}$ for particular choices of $\mathcal{M}$, and quantify the effects on this measure of efficiency.
As in @Roberts2001, we define MCMC algorithmic efficiency as the effective sample size divided by the chain length. This represents the rate of production of effectively independent samples per MCMC sample. The effective sample size (ESS) of an MCMC chain is defined as $\text{ESS} = N / \tau$, where $N$ is the chain length and $\tau$ is the integrated autocorrelation time. For a scalar chain of samples $X_0, X_1, \ldots$, which is assumed to have converged to its stationary distribution, @Straatsma1986 define the integrated autocorrelation time as $\tau = 1 + 2 \sum_{i=1}^{\infty} \text{cor} (X_0, X_i)$. $\tau$ may be interpreted as the number of MCMC samples required, on average, for an independent sample to be drawn. Our measure of algorithmic efficiency is thus $\tau^{-1}$, the number of effective samples per actual sample [@Thompson2010]. $\tau^{-1}$ also characterizes the speed at which expectations of arbitrary functions of the sample values approach their stationary values [@Roberts1997], and no less satisfies the natural intuition that larger values indicate better performance.
For MCMC algorithm $\Psi$ acting on model $\mathcal{M}$ with parameters $\Theta$, we define the algorithmic efficiency of each $\theta \in \Theta$ as $A(\Psi, \theta) = \tau^{-1}$, where $\tau$ is the integrated autocorrelation time of the samples of $\theta$ generated from repeated application of $\Psi$. Overloading notation, we define the algorithmic efficiency of MCMC algorithm $\Psi$ as $A(\Psi) = \text{min}_{\theta \in \Theta} \, A(\Psi, \theta)$. This definition is motivated by noting that often an MCMC produces seemingly good mixing of many model dimensions but poor mixing of just a few dimensions. In this case, the poorly mixing dimensions will limit the validity of the entire posterior sample (although this is not universally true of all model structures). Therefore, we take the conservative approach, and our general aim is to maximize the algorithmic efficiency for the parameter exhibiting the slowest mixing.
We now study the potential for algorithmic *inefficiency* that may result from scalar or block sampling, by examining situations in which each are particularly ill-suited.
### Efficiency loss from block sampling {#efficiency-loss-from-block-sampling .unnumbered}
Consider MCMC algorithm $\Psi_\text{block} \in \boldsymbol{\Psi}_\mathcal{M}$. Application of $\Psi_\text{block}$ generates a random proposal vector $X^* \in \mathbb{R}^d$, then jointly accepts or rejects $X^*$. In the framework of the sampling method $u_\text{block}$, $X^* \sim \text{N}_d(\mu, \sigma^2_d \Sigma)$, where $\mu$ and $\Sigma$ vary according to current state of the MCMC chain and properties of the target stationary distribution, but not proportionally to $d$. @Roberts1997a show that in order to achieve the asymptotically optimal acceptance rate, and therefore sample chain mixing, $\sigma^2_d \propto d^{-1}$. As a consequence of this attenuation in the proposal variance, the rate at which $\Psi_\text{block}$ explores the space of $\mathbb{R}^d$, and accordingly $A(\Psi_\text{block})$, is proportional to $d^{-1}$. This result applies equivalently to block samplers $\psi = u_\text{block}(b)$ acting on subsets $b \subset \Theta$, where the algorithmic efficiency (for the elements of $b$) achieved by application of $\psi$ is inversely proportional to the number of elements in $b$.
This result has an important implication on block sampling. All other factors being equal (*e.g.*, effect of posterior correlations), a block sampler of dimension $k$ must generate $k$ times more samples to have the same effective sample size as those samples produced through scalar sampling [@Roberts2001]. This inefficiency is inherent to block sampling and scales with the dimension of any block sampler.
### Efficiency loss from scalar sampling {#efficiency-loss-from-scalar-sampling .unnumbered}
To study the potential loss of algorithmic efficiency which may result from scalar sampling under $\Psi_\text{scalar} \in \boldsymbol{\Psi}_\mathcal{M}$, we consider correlated posterior distributions. It is well-understood that strong posterior correlations can retard the speed of convergence of MCMC sampling [*e.g.,* @Roberts1997; @Gilks2005], and that block sampling can alleviate this. However, the nature of this inefficiency is not characterized, in particular as a function of the degree of correlation and number of dimensions exhibiting correlation. We undertake a simulation study, to gauge how these factors effect algorithmic efficiency. Consider target distribution $\text{N}_d(0, \Sigma)$, where the covariance (equivalently, correlation) matrix $\Sigma$ consists of 1s on the main diagonal and $\rho$ elsewhere, $|\rho| < 1$. We consider the algorithmic efficiencies of individual model parameters under scalar sampling, $A(\Psi_\text{scalar}, \theta)$ for $\theta \in \Theta$.
Empirically, we observe that each $A(\Psi_\text{scalar}, \theta)$ tends toward zero as $\rho$ approaches one, or as $d$ diverges ($\rho \neq 0$). The nature of these relationships is characterized on a log-log scale in Figure \[fig:algorithmicEfficiency\] (left), where the horizontal axis plots $-\log(1-\rho)$, such that positive horizontal shifts represent $\rho$ exponentially approaching the boundary $\rho=1$, or perfect correlation between parameters. As one would expect, when $\rho=0$ all values of $d$ yield identical algorithmic efficiency. However, when $\rho > 0$ we enter a linear regime where each $A(\Psi_\text{scalar}, \theta)$ exponentially decays towards zero. Even for fixed $d$, algorithmic efficiency under $\Psi_\text{scalar}$ can be arbitrarily poor as $\rho$ approaches unity.
![MCMC algorithmic efficiencies for different values of model dimension (d), and intra-group correlation ($\rho$). The quantity $-\log(1-\rho)$ is plotted on the horizontal axis. Model structures include constant off-diagonal elements (equal to $\rho$) in the induced correlation matrix (left), and exponentially decaying correlations (right).[]{data-label="fig:algorithmicEfficiency"}](algorithmicEfficiency)
It may be extreme to assume a target distribution with arbitrarily large blocks of parameters that exhibit arbitrarily high pairwise-correlation. As an alternative, we consider the same multivariate normal form, but with elements of $\Sigma$ given as $\sigma_{i,j} = \rho^{|i-j|}$, $|\rho|<1$, as might occur in the covariance structure of a spatial model [@Banerjee2003 p.27]. The algorithmic efficiency of the first model parameter – since elements of $\Theta$ are no longer exchangeable – is shown in Figure \[fig:algorithmicEfficiency\] (right), where it displays similar attenuation as in the prior example. Most notably, we now observe the incremental effect of $d$ diminishing as $d$ increases, consistent with the covariance structure.
Through a combination of theory and simulated examples, we observe that the algorithmic efficiency achieved under $\Psi_\text{scalar}$ or $\Psi_\text{block}$ will depend non-trivially upon the model dimension, and the extent and structure of the posterior correlation. We have only considered two simple, highly modular and systematic posterior correlation structures. In practice, any model of interest will demonstrate a substantially more complex correlation structure (which is unknown, anyway), making a full analytical study of MCMC algorithmic efficiency difficult. No less, we have only considered the two boundary-case algorithms $\Psi_\text{scalar}$ and $\Psi_\text{block}$ in $\boldsymbol{\Psi}_\mathcal{M}$. Our desire to derive general results for algorithmic efficiency will be complicated substantially further when we consider the complete set $\boldsymbol{\Psi}_\mathcal{M}$.
Computational Efficiency {#sec:computation}
------------------------
While section \[sec:algorithmiceff\] considered the efficiency of MCMC algorithm $\Psi$ at producing independent samples without regard for computation time, we now consider the computational requirements of $\Psi$, measured in units of algorithm runtime per MCMC iteration. We focus on computations for model density evaluations. There are also book-keeping and loop-iteration costs, which we label as algorithm overhead. These overhead terms comprise a small fraction of total computation, and we safely disregard them. Denote the computational requirement of $\Psi$ as $C(\Psi)$, and again overload notation to define $C(\psi)$ as the computational requirement of a single sampler $\psi$. For $\Psi = \{ \psi_1, \ldots, \psi_k\}$, $C(\Psi) = \sum_{i=1}^k C(\psi_i)$. As far as we are aware, an analysis of MCMC efficiency which incorporates $C(\Psi)$ has not been carried out to date. Literature which does address MCMC efficiency typically recognizes that a computational aspect exists, but then focuses solely on $A(\Psi)$, *e.g.*, @Roberts1997.
We now present an accounting of the main contributions to $C(\Psi_\text{scalar})$ and $C(\Psi_\text{block})$, general for any $\mathcal{M}$. To support our accounting, we denote the set of all fixed and known components of $\mathcal{M}$ (*e.g.,* observations, other data) as $Y$, which is disjoint from the unknown set of parameters $\Theta = \{ \theta_1, \ldots, \theta_d\}$. For each $\theta_i \in \Theta$, let $x_i \subset \Theta \cup Y$ be the set of model components which immediately depend on $\theta_i$, or the set of direct descendants of $\theta_i$ in the model graph introduced in section \[mcmcdef\]. Finally, we denote the computational cost of evaluating the density functions corresponding to any subset $x \subset \Theta \cup Y$ as $\text{dens}(x)$.
### Scalar Sampling Computation {#scalar-sampling-computation .unnumbered}
On each iteration of $\Psi_\text{scalar} = \{ \psi_1, \ldots, \psi_d \}$, each sampler $\psi_i$ will incur computational cost $C(\psi_i) = \text{dens}(\theta_i) + \text{dens}(x_i) + O(1)$. The trailing constant term represents generation of the proposal value and making the MH decision (generation from normal and uniform distributions, respectively). The total computational requirement of $\Psi_\text{scalar}$ is thus $$C(\Psi_\text{scalar}) = \sum_{i=1}^d C(\psi_i) = \sum_{i=1}^d \text{dens}(\theta_i) + \sum_{i=1}^d \text{dens}(x_i) + O(d).$$ Note that under $\Psi_\text{scalar}$, each density evaluation $\text{dens}(\theta_i)$ must occur independently. This is true even when the evaluation of a particular $\text{dens}(\theta_i)$ term necessitates the calculation of a subsuming multivariate density – in the most extreme case, $\text{dens}(\Theta)$. Thus, in the worst case, a single MCMC iteration of $\Psi_\text{scalar}$ could incur $d$ evaluations of the entire joint density of $\Theta$. A similar computational explosion can result from the calculation of each $\text{dens}(x_i)$ term.
### Block Sampling Computation {#block-sampling-computation .unnumbered}
We now consider the components of $C(\Psi_\text{block})$. On each iteration of $\Psi_\text{block}$, the sole sampler $u_\text{block}(\Theta)$ requires evaluation of the complete prior and likelihood densities, $\text{dens}(\Theta \cup Y)$. This is notably different from the density evaluation terms appearing in $C(\Psi_\text{scalar})$, in that it incurs only a *single* evaluation of the complete joint model density. In addition, the adaptation routine of $u_\text{block}(\Theta)$ requires a Cholesky factorization of the adapted covariance matrix, which requires $O(d^3)$ operations to calculate in full generality [@Trefethen1997 p.176]. This factorization occurs every AI iterations, where AI is the adaptation interval of $u_\text{block}$, and therefore has an amortized computational cost of $O(d^3 / \text{AI})$. Each iteration of $u_\text{block}$ requires generating a $d$-dimensional multivariate normal proposal, which requires $O(d^2)$ operations, and performing a single constant-time MH decision, which is dropped as a lower-order term. The total amortized computational requirement of $\Psi_\text{block}$ may be written as $$C(\Psi_\text{block}) = \text{dens}(\Theta \cup Y) + O(d^3 / \text{AI}) + O(d^2).$$
### Timing Comparison {#timing-comparison .unnumbered}
We have seen that $C(\Psi_\text{block})$ is at least quadratic in $d$, and technically cubic but with a small leading coefficient. Depending on the distributional structure of $\Theta$, the density evaluations comprising $C(\Psi_\text{scalar})$ may be unwieldy. The relative magnitude of these competing terms is difficult to intuitively gauge, so to gain practical insight, we perform a timing study of the $\Psi_\text{scalar}$ (All Scalar) and $\Psi_\text{block}$ (All Blocked) algorithms. Three models involving no likelihood components are considered, with prior structures on $\Theta$ given as:
- $\theta_i \sim \text{N}(\mu, \sigma)$ for each $\theta_i \in \Theta$
- $\theta_i \sim \text{Gamma}(\alpha, \beta)$ for each $\theta_i \in \Theta$
- $\Theta \sim \text{N}_d(\mu, \Sigma)$
Figure \[fig:computationalRequirement\] presents timing results measured in seconds per 10,000 iterations, plotted against dimension $d$, without consideration of algorithmic efficiency (section \[sec:algorithmiceff\]). There are a number of interesting features, which we briefly summarize. $C(\Psi_\text{scalar})$ is $O(d)$ when each $\theta_i$ independently follows a univariate distribution, and therefore $\sum_{i=1}^d \text{dens}(\theta_i) = \text{dens}(\Theta) \leq d \cdot K$, where $K = \text{max}_{\theta \in \Theta} \; \text{dens}(\theta)$. For all practical purposes, it appears that $C(\Psi_\text{block})$ is $O(d^2)$, as the cubic coefficient $1 / \text{AI} = 1 / 200$ is relatively small. $C(\Psi_\text{block})$ is largely resilient to changes in the underlying distribution of $\Theta$; univariate gamma densities are more costly than normal densities, as we would expect, and as for $C(\Psi_\text{scalar})$; and the multivariate normal structure even slightly more so. Perhaps most striking, $C(\Psi_\text{scalar})$ is $O(d^3)$ when the underlying distribution of $\Theta$ is multivariate, since each multivariate normal density evaluation is $O(d^2)$, which occurs $d$ times for each iteration of $\Psi_\text{scalar}$. Although both are cubic in $d$, $C(\Psi_\text{scalar})$ dwarfs $C(\Psi_\text{block})$ due to the difference in the leading cubic coefficients.
![MCMC runtimes for the All Scalar and All Blocked algorithms, for univariate normal, univariate gamma, and multivariate normal model structures.[]{data-label="fig:computationalRequirement"}](computationalRequirement)
Overall Efficiency {#sec:overallefficiency}
------------------
We have examined both the algorithmic and computational efficiency of MCMC algorithms, each of which fundamentally affect overall MCMC efficiency. We define the overall efficiency of $\Psi$ simply as $E(\Psi) = A(\Psi) / C(\Psi)$. We consider this to be a sensible measure of the overall efficiency of an MCMC algorithm, since $E(\Psi)$ may be interpreted as the number of effective samples produced per unit of MCMC algorithm runtime, for the slowest mixing model parameter. If we can construct $\Psi$ to maximize $E(\Psi)$, then $\Psi$ is the most time-efficient MCMC algorithm for generating effectively independent samples to approximate the true, joint posterior distribution of $\Theta$.
That being said, from our examination of algorithmic and computational efficiency, it is not immediately clear how to balance tradeoffs between $A(\Psi)$ and $C(\Psi)$ to maximize $E(\Psi)$. We have generally considered the two boundary-case algorithms $\Psi_\text{scalar}$ and $\Psi_\text{block}$, but a huge number of intermediate algorithms exist. For $\Psi \in \boldsymbol{\Psi}_\mathcal{M}$, we may gain useful insights regarding the factors affecting $E(\Psi)$ in terms of the properties of each $\psi_i$. $C(\Psi) = \sum_{i=1}^k C(\psi_i)$, and the values $A(\Psi, \theta_i)$ which determine $A(\Psi)$ each result from a single application of scalar or block sampling – although this neglects the phenomenon where improving $A(\Psi, \theta_i)$ may affect $A(\Psi, \theta_j)$, $i \neq j$. However, finding a global optimum $\Psi_\text{opt} = \text{argmax}_{\Psi \in \boldsymbol{\Psi}_\mathcal{M}} E(\Psi)$ poses a combinatorial challenge. Instead of seeking $\Psi_\text{opt}$, we now propose an iterative procedure to navigate $\boldsymbol{\Psi}_\mathcal{M}$, with the aim of maximizing $E(\Psi)$ to the degree possible.
Automated Blocking {#sec:autoblock}
==================
In this section, we propose an iterative, self-tuning procedure for automated blocking of hierarchical model parameters to produce an efficient MCMC algorithm. This procedure uses the empirical posterior correlation to cluster groups of correlated parameters into sampling blocks. A hierarchical clustering tree of model parameters is constructed, and subsequently cut at some height (selected using a finite search) to produce parameter groups each exhibiting a minimal intra-group posterior correlation. This procedure is iterated, so that as MCMC efficiency improves, the empirical posterior correlations are more accurate, and the resulting tree and parameter groups stabilize. The end-result is a partition of the model parameters, which uniquely specifies an MCMC algorithm $\Psi \in \boldsymbol{\Psi}_\mathcal{M}$ employing scalar and block sampling, for which the overall efficiency $E(\Psi)$ (section \[sec:overallefficiency\]) is increased to the degree possible. We also discuss more sophisticated approaches, but our heuristic approach allows huge efficiency gains in some cases and establishes the basic procedure.
Procedure
---------
Assume a given, fixed, hierarchical model $\mathcal{M}$, with parameters $\Theta = \{ \theta_1, \ldots, \theta_d \}$. Algorithm \[alg:autoblock\] presents pseudocode for our automated blocking procedure, which produces MCMC algorithm $\Psi_\text{AutoBlock} \in \boldsymbol{\Psi}_\mathcal{M}$. Subscripting indices $j$ and $k$ are understood to take all values in $1, \ldots, d$.
$i \gets 0$ $\Psi_0 \gets \Psi_\text{scalar}$ *loop*: $i \gets i + 1$ Generate samples of $\Theta$ under $\Psi_{i-1}$, where $X_j$ represents the sample chain of $\theta_j$ Discard initial 50% of each chain $X_j$ $\rho_{j,k} \gets \text{cor}(X_j, X_k)$ $d_{j,k} \gets 1 - | \rho_{j,k} |$ Construct distance matrix $D$ from elements $d_{j,k}$ Construct hierarchical clustering tree $T$ from $D$ $\boldsymbol{\Psi}_\text{cand} \gets \{ \Psi(T_{\text{cut}=0}), \Psi(T_{\text{cut}=0.1}), \Psi(T_{\text{cut}=0.2}), \ldots, \Psi(T_{\text{cut}=1}) \}$ $\Psi_{i} \gets \text{argmax}_{\Psi \in \boldsymbol{\Psi}_\text{cand}} E(\Psi)$ **if** $E(\Psi_{i}) > E(\Psi_{i-1})$ **and** $\Psi_{i} \neq \Psi_{i-1}$ **then goto** *loop* $\Psi_\text{AutoBlock} \gets \Psi_i$ **return** $\Psi_\text{AutoBlock}$
The procedure begins with the initial MCMC algorithm $\Psi_0 = \Psi_\text{scalar}$, or scalar sampling of all model parameters; lacking prior information, this is used as the starting point for gaining insight about the posterior correlation structure. Subsequent iterations are based upon the empirical posterior correlation produced in the previous iteration, and, to a varying degree, will institute blocks for parameter groups exhibiting sufficiency high intra-group correlations.
Prior to calculating the empirical correlation terms $\rho_{j,k}$, we discard the seemingly excessive and somewhat arbitrary initial 50% of all samples. This should not be confused with a traditional “burn-in," whose purpose is to “forget" the initial state and ensure convergence to the target distribution. Instead, discarding these initial samples allows all adaptive scalar and block samplers ample time to self-tune, and thereby achieve their theoretically optimal algorithmic efficiency. The choice of 50% is largely arbitrary, and excessive in most cases, and could almost certainly be relaxed without affecting algorithm performance.
Empirical correlations are transformed into distances using the transformation $d_{j,k} = 1 - | \rho_{j,k} |$. The form of this transformation is selected to induce several properties for elements of the distance matrix $D$: the main diagonal consists of zeros; strong correlation results in $d \approx$ 0; weak or zero correlation results in $d \approx$ 1; and correlations of $\rho$ and $-\rho$ result in the same distance.
We use the R function `hclust` to create the hierarchical tree $T$ from the distance matrix $D$. The default “complete linkage" clustering [@Everitt2011 chapter 4] is appropriate, since this ensures that all parameters within each cluster have a minimum absolute pairwise correlation. At height $h \in [0,1]$ in $T$, the absolute correlation between parameter pairs (within clusters) is at least $1 - h$.
We use the R function `cutree` for cutting the hierarchical clustering tree $T$ at a specified height $h \in [0,1]$ to produce disjoint parameter groupings, which may be used to define parameter blocks for the purpose of MCMC sampling. We justify this means of generating parameter sampling blocks, insofar as to increase algorithmic efficiency we strive to group *correlated* parameters into sampling blocks – the exact effect of cutting $T$ at any particular height.
We define the MCMC algorithm $\Psi(T_{\text{cut}=h}) \in \boldsymbol{\Psi}_\mathcal{M}$ as the unique MCMC algorithm defined by scalar and/or block sampling applied to the parameter blocks that result from cutting $T$ at height $h$. We note that for all $T$, $\Psi(T_{\text{cut}=0}) = \Psi_\text{scalar}$, and $\Psi(T_{\text{cut}=1}) = \Psi_\text{block}$.
There is no universally optimal value of the cut height $h$, as our findings in section \[sec:efficiency\] imply that any $h \in [0,1]$ may maximize the efficiency $E(\Psi(T_{\text{cut}=h}))$ for a particular model $\mathcal{M}$. We recognize that a combination of distinct cut heights applied to different branches of $T$ may produce the maximal efficiency, but we do not consider such strategies herein.
Rather than attempting to infer what might be an appropriate cut height for model $\mathcal{M}$, we consider a range of potential cut heights, and the resulting MCMC algorithms. These comprise $\boldsymbol{\Psi}_\text{cand}$, the candidate set of MCMC algorithms for a particular iteration. This approach allows the blocking procedure to adjust according to the model, the posterior correlation structure, and the underlying computational architecture. The MCMC algorithm for each iteration $(i \geq 1)$ is selected from among $\boldsymbol{\Psi}_\text{cand}$ as that which produces the maximal overall efficiency.
To estimate the integrated autocorrelation time, and hence the algorithmic efficiency, of a chain of MCMC samples, we use the `effectiveSize` function in the R `coda` package [@Plummer2006]. The approach underlying this function – using a fitted autoregressive model to estimate the spectral density at frequency zero – has been seen to converge fastest among several methods to the true integrated autocorrelation time [@Thompson2010].
As $E(\Psi_i)$ increases through the course of iterations, improved estimates of the posterior correlation are produced, giving the potential for more refined parameter blockings, and thus progressive increases in $E(\Psi_i)$ in subsequent iterations. This iterative procedure continues until either (1) $\Psi_{i} = \Psi_{i-1}$ (identical algorithms are selected on consecutive iterations), or (2) $E(\Psi_{i}) < E(\Psi_{i-1})$ (efficiency decreases between iterations). In practice, our procedure typically halts with terminating condition (1). This may be concurrent with terminating condition (2), on account of stochastic variation in sampling and/or runtime.
We select the output from our automated blocking procedure as $\Psi_\text{AutoBlock}$, the MCMC algorithm selected in the final iteration. In our experience, $\Psi_\text{AutoBlock}$ is typically identical to the MCMC algorithm of the second-to-last iteration; that is, the procedure has converged to a stationary state. If a situation arises where the final iteration produces a different MCMC algorithm with efficiency inferior to that of the previous iteration, then prudence would suggest a thoughtful examination of the posterior samples, empirical correlation matrices, properties of the adapted samplers, convergence diagnostics, etc.
Automated Blocking Performance {#sec:performance}
==============================
We now compare the performance of MCMC algorithms produced using the automated blocking procedure of section \[sec:autoblock\] against various static MCMC algorithms. First, we describe the computing environment in which our analyses are performed. We then describe a broadly representative suite of example models, and present the performance results of automated and static MCMC algorithms for each. A public Github repository containing scripts for reproducing our results may be found at `https://github.com/danielturek/automated-blocking-examples`.
Computing Environment
---------------------
Since one of our points is that optimal design of MCMC algorithms depends on the computing environment, we briefly summarize the software tools and computing platform used. All statistical models and MCMC algorithms were built using the NIMBLE package [@NIMBLEDevelopmentTeam2014] for R [@RCoreTeam2014]. NIMBLE allows hierarchical models to be defined within R using the BUGS model declaration syntax introduced by the BUGS project [@Lunn2000; @Lunn2012]. MCMC algorithms in NIMBLE are written using NIMBLE’s domain specific language for specifying hierarchical model algorithms. This language is an enhanced subset of R (interfaced through an R session) which is compiled into C++ code, which is subsequently compiled and run.
As a result, the examples here use highly efficient code generated automatically for each model and algorithm. Of particular importance is that matrix operations are done via the highly optimized C++ Eigen library [@Guennebaud2010]. Finally, the high-level programmability provided by R facilitated the dynamic exploration of MCMC algorithms. Examples were run using R version 3.1.2, using the BLAS (Basic Linear Algebra Subprograms) provided by R for multivariate density calculations and simulation, and running under Macintosh OSX version 10.9.5 on a 2.5 GHz Intel Core i7 processor.
Model Descriptions
------------------
We tested the automated blocking procedure on seven examples, including real-data and toy models, and compared the results against standard MCMC algorithms. When there are obvious “hand-tuned" algorithms that a seasoned MCMC practitioner would consider for a particular model, we included those as well. For the toy models, the goal was to construct posterior distributions with specific correlation structures as described below. In these cases the models are simply prior distributions without any likelihood component.
### Varying Size Blocks of Fixed Correlation {#varying-size-blocks-of-fixed-correlation .unnumbered}
This model structure demonstrates parameter groups of varying size, where each group exhibits a fixed intra-group pairwise correlation. The model contains $N=64$ parameters, half of which are grouped to have pairwise posterior correlation of $\rho$. This is accomplished using a prior multivariate normal distribution with appropriate covariance (equivalently, correlation) matrix, which in the absence of a likelihood term fully determines the posterior distribution for these 32 parameters. Similarly, additional disjoint groups of correlated parameters are constructed of sizes 16, 8, 4, and 2. The remaining two parameters are uncorrelated to any others, specified using univariate normal prior distributions. We consider three values for the intra-group correlation, $\rho$ = 0.2, 0.5, and 0.8. As these models have no likelihood, we are using MCMC to sample from the prior distribution. We note that the dependence structure is the same as the block-diagonal covariance structure (with the blocks having compound symmetry) obtained when analytically integrating over the exchangeable prior means of clustered random effects. This example thereby mimics the structure found in basic multilevel hierarchical models, albeit without the explicit computational expense of a likelihood calculation.
### Fixed Size Blocks of Varying Correlation {#fixed-size-blocks-of-varying-correlation .unnumbered}
The next model structure exhibits fixed size groupings of parameters, with posterior correlations ranging between 0 and 0.9. Each such model contains $N = 10n$ parameters. Again employing multivariate normal distributions, we induce nine disjoint groupings of $n$ parameters each, having intra-group pairwise correlations of 0.1, 0.2, $\ldots$, and 0.9. The remaining $n$ parameters are fully uncorrelated. Three such models of this structure are constructed, using the values $n$ = 2, 5, and 10. As in the previous models, these do not include any likelihood term.
### Random Effects Model {#random-effects-model .unnumbered}
We select the “litters" model from among the original example models provided with the MCMC package WinBUGS. This random effects model contains two groups of 16 binomial observations. Within each group, the binomial probabilities are modeled as random effects arising from a beta distribution. The particular parameterization of the beta distributions (in terms of $\alpha$ and $\beta$, rather than $\mu$ and $\sigma$) results in strong correlations between each $\alpha_i, \beta_i$ pair. The WinBUGS manual comments upfront that this model exhibits slow mixing. We consider an informed MCMC algorithm, which blocks each $\alpha_i, \beta_i$ pair. In addition, the beta-binomial conjugacy relationships permit use of cross-level sampling, where we jointly sample top-level parameters and conjugate latent states, as used by @Rue2005 [p.141-143].
### Auto-Regressive Model {#auto-regressive-model .unnumbered}
We select the “ice” model from among the examples provided with WinBUGS as an auto-regressive [AR; @Harvey1993] example, which is also analyzed in @Breslow1993. The data contains 77 incident counts of breast cancer occurring in Iceland, which are modeled as Poisson counts. Explanatory variables include age group, year of birth (represented using 11 cohorts ranging between 1840 and 1949), and the total person-years for the subjects in each group. The model uses second-order AR smoothing of birth cohort effects.
### Linear Gaussian State Space Models {#linear-gaussian-state-space-models .unnumbered}
We construct two linear Gaussian state space models [@Durbin2012] each consisting of 100 latent states and observations. State transitions are governed by a first order AR process, and we seek inferences about the transition process, and the system and observation noise. We consider two equivalent parameterizations of the state transition process. First, in terms of the intercept and mean of the AR process, which have largely uncorrelated posteriors (independent parameterization), and second, in terms of the intercept and autocorrelation, which are known to be highly correlated (correlated parameterization). For the correlated parameterization, we consider an informed MCMC algorithm, which blocks the intercept and autocorrelation parameters. We deliberately include this inferior parameterization, to assess MCMC performance in the case of known strong posterior correlation. In practice, an analyst may not know which model parameterization(s) will produce uncorrelated posterior dimensions.
### Spatial Model {#spatial-model .unnumbered}
We consider a spatially dependent hierarchical model. The data consist of 148 measurements of scallop abundance at various locations off the New York and New Jersey coastline, and was collected by the Northeast Fisheries Science Center of the National Marine Fisheries Service in 1993. The data set is publicly available at `http://www.biostat.umn.edu/~brad/data/myscallops.txt`, and is analyzed in @Banerjee2003, pages 44-65. Following @Banerjee2003, we model the mean log-abundance as multivariate normal with covariance that decays exponentially as a function of distance. The covariance is given by $\text{cov}(g_i,g_j) = \sigma^2 \exp(-d_{i,j}/\rho)$, where the observations are modeled as Poisson counts $y_i \sim \text{Poisson}(\text{exp}(g_i))$, and $d_{i,j}$ is the distance between observations $y_i$ and $y_j$. Since this covariance structure induces a trade-off between $\sigma$ and $\rho$, we expect these parameters to be correlated in the posterior distribution.
### Generalized Linear Mixed Model {#generalized-linear-mixed-model .unnumbered}
We include a reasonably sized generalized linear mixed model [GLMM; @Gelman2006 chapter 6]. We make use of the Minnesota Health Plan dataset available in @Waller1997 and follow the analysis of @Zipunnikov2006. The dataset contains 968 counts of senior-citizen clinical visits, which are modeled as Poisson counts. The linear predictor contains fixed and random effects, using a variety of covariates and including several interaction terms.
Performance Results
-------------------
We present three quantities to gauge the performance of MCMC algorithm $\Psi$. Rather than algorithmic efficiency $A(\Psi)$, for convenience of interpretation we present the proportional quantity ESS = 10,000 $A(\Psi)$, where ESS denotes effective sample size. This scaling of $A(\Psi)$ has a natural interpretation as the number of effective samples (for the slowest mixing parameter) which result from a chain of 10,000 MCMC samples. Similarly, to represent the computational requirement $C(\Psi)$, we present the proportional quantity Runtime = 10,000 $C(\Psi)$, interpretable as the time (in seconds) required to generate 10,000 MCMC samples. We directly present the overall MCMC efficiency as Efficiency = ESS / Runtime = $A(\Psi) / C(\Psi) = E(\Psi)$, which is independent of any scaling, and maintains the intuitive interpretation as the number of effective samples generated per second of algorithm runtime (again, for the slowest mixing parameter). MCMC sampling is performed using a fixed random number seed and identical initial values for each model, so identical MCMC algorithms will produce identical sample chains, and hence ESS, but not necessarily Runtime or Efficiency on account of discrepancies in algorithm runtime. We observe the automated procedure producing the same MCMC algorithm across repeated experiments, with numerical results for Runtime and Efficiency varying less than 5% from those presented herein.
For each example model $\mathcal{M}$, we present results for MCMC algorithm $\Psi_\text{block}$ denoted as “All Blocked," and those of $\Psi_\text{scalar}$ as “All Scalar," noting that $\Psi_\text{scalar}$ also represents the initial state ($0^{th}$ iteration) of the automated blocking procedure. The maximally efficient algorithm generated via automated blocking is presented as “Auto Blocking," which will generally represent a dynamically determined blocking scheme. We also present a third static MCMC algorithm, which is not necessarily a member of $\boldsymbol{\Psi}_\mathcal{M}$ on account of the possible use of conjugate sampling. This algorithm assigns block samplers to groups of parameters arising from multivariate distributions, scalar samplers to parameters arising from univariate distributions, and assigns conjugate samplers whenever the structure of $\mathcal{M}$ permits; this static algorithm may be more representative of default MCMC algorithms provided by software packages, and is denoted as “Default." Finally, for several example models we include an informed blocking of the model parameters, based upon expert or prior knowledge, which is referred to as “Informed Blocking." Results for the random effects model also include the “Informed Cross-Level" MCMC algorithm which makes use of cross-level sampling, which is not in $\boldsymbol{\Psi}_\mathcal{M}$.
### Varying Size Blocks of Fixed Correlation {#varying-size-blocks-of-fixed-correlation-1 .unnumbered}
The left pane of Figure \[fig:contrivedModels\] displays the Efficiency performance for the model structures containing varying sized blocks of fixed correlation. For $\rho=0.2$, the Auto Blocking algorithm selects cut height $h = 0$, which corresponds to re-selecting the algorithm All Scalar. Since this MCMC algorithm is identical to the initial state, the automated procedure terminates there. The All Blocked scheme actually runs faster, but the algorithmic efficiency loss inherent to large block sampling dominates, resulting in Efficiency approximately four times lower. For larger values of $\rho$, the All Scalar algorithm suffers progressively more since it fails to institute any blocking in the presence of increasing correlations. For $\rho = 0.5$ and 0.8, Auto Blocking algorithm selects cut heights $h=0.6$ and $h=0.3$, respectively, which each exactly place all correlated terms into sampling blocks. In every case, the slowest mixing parameter is from among the largest correlated group of 32 parameters.
![Efficiency results for two contrived model structures: varying sized blocks of fixed correlation (left), and fixed sized blocks of varying correlation (right).[]{data-label="fig:contrivedModels"}](contrivedModels)
### Fixed Size Blocks of Varying Correlation {#fixed-size-blocks-of-varying-correlation-1 .unnumbered}
The right pane of Figure \[fig:contrivedModels\] presents results for the model structure containing fixed size parameter groupings with correlations between 0 and 0.9. For each size model, the automated blocking procedure selects a particular cut height (and hence, MCMC algorithm) twice consecutively, thus terminating on the third iteration. The cut heights selected for models $N$ = 20, 50, and 100 are $h$ = 0.5, 0.8, and 0.9, respectively (not shown), progressively pushing more of the correlated parameter groupings into sampling blocks. The Auto Blocking algorithm produces increases in Efficiency by factors of 4.5, 7, and 21 in the three models, over the static All Scalar and All Blocked algorithms.
### Random Effects Model {#random-effects-model-1 .unnumbered}
In the random effects model (Table \[tab:exampleModels\]), automated blocking generates an MCMC algorithm identical to the Informed Blocking algorithm (blocking each $\alpha_i$, $\beta_i$ pair), which produces a tenfold improvement in Efficiency over the most efficient static algorithm – for this model, All Scalar sampling. The cut height $h = 0.1$ indicates that only the $\alpha_i$, $\beta_i$ pairs exhibit posterior correlations above 0.9. The Informed Cross-Level algorithm requires a substantially longer Runtime and produces a high ESS, which results in nearly identical Efficiency as the efficiently blocked Auto Blocking algorithm.
### Auto-Regressive Model {#auto-regressive-model-1 .unnumbered}
In the auto-regressive model (Table \[tab:exampleModels\]), an AR process value exhibited the slowest mixing under All Scalar sampling. When all 24 model parameters (AR process values, fixed effects, and one hyper-parameter) are blocked, the algorithm Runtime is nearly halved. This decrease in Runtime is largely due to the dependency structure inherent to the AR process. Scalar sampling of AR process values requires nearly a three-fold increase in density evaluations of the process values (since it’s a second-order AR process) relative to All Blocked sampling. In addition to the improved Runtime, the All Blocked sampling of the correlated AR process values increases their individual algorithmic efficiencies, and the slowest mixing parameter is among the fixed effects. The Efficiency under All Blocked sampling is over double that of All Scalar sampling. The automated blocking procedure identifies a blocking scheme which blocks together all AR process values and fixed effects (23 total; cut height $h$ = 0.4), and performs univariate sampling of the single hyper-parameter. This has a similar Runtime to All Blocked sampling, but increases algorithmic efficiency for all parameters. The resulting overall Efficiency under the Auto Blocking MCMC algorithm is over three times that of All Scalar sampling.
Model MCMC Scheme ESS Runtime Efficiency
------- ---------------------- -------- --------- ------------
All Blocked 0.4 0.29 1.3
Default 1.1 1.19 1.0
All Scalar 2.1 0.51 4.2
Informed Blocking 19.0 0.50 38.2
Informed Cross-Level 101.3 2.64 38.5
Auto Blocking 19.0 0.48 39.2
All Blocked 8.9 0.3 27.3
All Scalar 6.5 0.6 11.5
Auto Blocking 12.7 0.3 37.5
All Blocked 0.3 0.8 0.4
Default 27.6 4.6 6.0
All Scalar 20.2 1.3 15.7
Auto Blocking 29.1 1.3 22.4
All Blocked 0.6 0.7 0.8
Default 1.7 4.9 0.4
All Scalar 1.1 1.3 0.8
Informed Blocking 18.4 1.2 15.6
Auto Blocking 26.1 1.2 20.9
All Blocked 0.2 5.71 0.04
Default 0.4 10.86 0.04
All Scalar 171.3 83.87 2.0
Auto Blocking 1208.0 78.62 15.4
All Blocked 2.2 44.3 0.05
All Scalar 60.9 22.6 3.0
Auto Blocking 60.9 22.6 3.0
: MCMC performance results for the suite of example models. Effective sample size (ESS) is measured in effective samples per 10,000 iterations, Runtime is presented as seconds per 10,000 iterations, and Efficiency is in units of effective samples produced per second of algorithm runtime.[]{data-label="tab:exampleModels"}
### Linear Gaussian State Space Models {#linear-gaussian-state-space-models-1 .unnumbered}
Table \[tab:exampleModels\] presents results for both parameterizations of the state space model. In the Independent parameterization, the observation noise parameter is the slowest mixing, in all except the All Blocked algorithm. The All Blocked algorithm runs quickly, but is limited by the extremely low algorithmic efficiency of the AR process intercept parameter. The Default algorithm assigns conjugate normal samplers to each latent state, resulting in high algorithmic efficiency but a substantially longer Runtime, which diminishes the overall Efficiency. Auto Blocking (cut height $h$ = 0.8) creates a block of six parameters containing five latent states and the observation noise, and a disjoint block of the two AR process parameters. This combination, unlikely to be discovered though any combination of prior knowledge or trial and error, produces a 40% increase in Efficiency over All Scalar sampling, which is the most efficient static MCMC algorithm.
We suspect the intercept and autocorrelation parameters of the AR process to be correlated in the naïve parameterization of the state space model. The All Blocked algorithm once again runs quickly, but is limited by the ESS of the AR process intercept. The Default algorithm is again slow due to conjugate sampling, but similar to the All Scalar algorithm, produces low algorithmic efficiency of the correlated AR process parameters. The Auto Blocking algorithm (cut height $h$ = 0.1) selects the same parameter block as in the Informed Blocking algorithm (AR process intercept and autocorrelation), and additionally a block containing the observation noise and a latent state. The Runtimes are, accordingly, nearly identical, however the ESS of the observation noise, the limiting parameter, increases. Automated blocking produces Efficiency over 20 times higher than the All Blocked algorithm, which is the most efficient static algorithm, and 25% higher than the Informed Blocking algorithm. It is important to note that the automated blocking procedure overcame the sampling inefficiencies introduced by this naïve parameterization, without requiring user intervention.
### Spatial Model {#spatial-model-1 .unnumbered}
MCMC performance results for the spatial model (Table \[tab:exampleModels\]) display several interesting trade-offs in MCMC efficiency. The spatial model contains 148 latent parameters jointly following a multivariate normal distribution, and three top-level parameters that govern this distribution ($\mu$, $\sigma$, and $\rho$). As there are no conjugate relationships between parameters, the sole difference between the All Blocked algorithm and the Default algorithm is the inclusion of these top-level parameters in the large sampling block. Therefore, the five second difference in Runtime can be attributed to three fewer multivariate density evaluations (per MCMC iteration) under the All Blocked algorithm. However, under either algorithm, the blocked sampling of latent parameters produces extremely low ESS values of 0.2 and 0.4 among the latent parameters. The minimal ESS value increases by a factor of two when the top-level parameters are removed from the large block sampler, and thereby achieve better mixing.
The All Scalar algorithm frees all latent parameters from block sampling. Each scalar sampler requires its own, independent, evaluation of the latent multivariate density, hence the Runtime of the All Scalar algorithm increases dramatically. That being said, the ESS values of the slowest mixing latent parameters under the All Blocked and Default algorithms both increase to approximately 4000 (not shown). $\rho$ is the slowest mixing parameter under the All Scalar algorithm, with ESS increased from 139.6 (under the Default algorithm) to 171.3, even though it underwent scalar sampling in both cases; this is another example of the slowest mixing parameter affecting the algorithmic efficiency of other model parameters.
The automated blocking procedure selects cut height $h = 0.1$, which produces a single block containing $\rho$ and $\sigma$; this indicates an empirical posterior correlation of at least 0.9 between these parameters. The ESS of $\rho$ increases to approximately 1500 (not shown). A latent parameter once again produces the slowest mixing with ESS of 1208, which produces nearly a tenfold increase in Efficiency relative to the All Scalar algorithm. The Runtime of the Auto Blocking algorithm decreases slightly compared to the All Scalar algorithm, since the single block sampler induces one fewer evaluation of the latent multivariate density.
### Generalized Linear Mixed Model {#generalized-linear-mixed-model-1 .unnumbered}
We first note that our GLMM model is by far the largest example considered, containing nearly 2000 stochastic model components (including observations); so we anticipate comparatively low MCMC Efficiencies regardless, since MCMC algorithms simply take time to carry out all model calculations. For this model (Table \[tab:exampleModels\]), the automated procedure converges on the All Scalar algorithm, which is the same as its initial state, and which produces overall MCMC Efficiency of about 3. In hindsight this result may not surprise us, since the fully exchangeable nature of the random effects in this model does not induce correlations among the sampled parameters for this particular dataset. Correspondingly, for a large number of un-correlated random effects, and in the absence of multivariate distributions, univariate sampling produces the highest Efficiency. We also note that the All Blocked algorithm, which consists of a single block sampler of dimension 858, has Runtime approximately twice that of the All Scalar algorithm, and produces an overall Efficiency of approximately 0.05.
![Efficiencies of MCMC algorithms for the suite of example models.[]{data-label="fig:exampleModels"}](exampleModels)
### Efficiency Gains from Automated Blocking {#efficiency-gains-from-automated-blocking .unnumbered}
In Figure \[fig:exampleModels\], we present the overall Efficiencies achieved for our suite of example models (excluding the two contrived model structures). The Auto Blocking algorithm consistently out-performs any static algorithm in terms of Efficiency, ranging between roughly a 50% increase to several orders of magnitude of improvement. The exception is the GLMM example, in which Auto Blocking matches the All Scalar algorithm identically. We observe variation in the relative Efficiencies among the static algorithms, reinforcing our notion that overall MCMC efficiency is highly dependent upon hierarchical model structure, and attempting to infer what might be an efficient MCMC algorithm for a particular problem is, in general, difficult.
Discussion {#sec:discussion}
==========
We have presented a general automated procedure for determining an “efficient" MCMC algorithm for hierarchical models. Our procedure is a greedy, iterative algorithm, which traverses a finite and well-defined set of MCMC algorithms. This is the first such automated MCMC-generating procedure of its kind, so far as we are aware. Using a suite of example models, we have observed that our automated procedure generates improvements in efficiency (relative to static MCMC algorithms) ranging between one and three orders of magnitude. In each case, the automated procedure produced an MCMC algorithm at least as efficient as any model-specific MCMC algorithm making use of prior knowledge or expert opinion. In all examples, our iterative procedure terminated within four iterations, although it is plausible that for more complex models it would proceed longer.
Our study has been confined to a single dimension of a much broader problem. We have strictly considered combinations of scalar and blocked adaptive Metropolis-Hastings sampling, with a small number of exceptions only for the purpose of comparison (*e.g.*, the use of conjugate sampling). No less, we have restricted ourselves to non-overlapping sampling: each model parameter may only be sampled by a single MCMC sampler function. We may instead view the domain of our problem (automated determination of an efficient MCMC algorithm) as a broader space of MCMC algorithms. This space may permit a wide range of sampling algorithms not considered herein: auxiliary variable algorithms such as slice sampling [@Neal2003], or derivative-based sampling algorithms such as Hamiltonian Monte Carlo [@Duane1987], among many possibilities. The resulting combinatorial explosion in the space of MCMC algorithms makes any process of trial-and-error, or an attempt at comprehensive exploration, futile. It is for this reason we seek to develop an automated procedure for determining an efficient MCMC algorithm, which may not be globally, maximally efficient, but provides non-trivial improvements in efficiency, nonetheless.
It should be noted that aspects of the problem addressed herein superficially resemble, but are fundamentally different in nature from hierarchical clustering, or sparse covariance estimation. Granted, our automated procedure firstly utilizes an empirical covariance matrix generated from MCMC sampling chains. However, whereas sparse covariance estimation seeks to estimate the non-zero elements of the underlying covariance structure [@Cai2011], our procedure concerns the non-trivial (correlated) elements, with little concern for the smaller entries. Our blocking algorithm also makes use of the complete linkage clustering algorithm, for determining groupings of correlated model parameters. Clustering algorithms have been applied to a wide variety of problems [@Xu2005], but not to parameters of hierarchical models specifically with the aim of accounting for trade-offs between MCMC algorithmic efficiency and computational requirements, to produce a *computationally efficient* MCMC algorithm. This is a fundamentally different goal than merely producing groupings of “similar" parameters (given some measure of similarity), as is generally the goal in most clustering applications. Thereby, the existing literature on these subjects is related, but not intimately applicable to our problem at hand. A deeper consideration of these topics may be worthwhile, but we consider it beyond the scope of this paper.
Reasonably straightforward improvements could be made to our automated blocking procedure, which is presented as a sensible first approach that addresses the factors affecting MCMC algorithm efficiency. By design, our procedure natively accounts for differences in system platform or architecture that may affect the relative efficiencies of MCMC algorithms. We can envision a wide variety of possible extensions to our algorithm, ranging from only re-blocking the slowest mixing parameter on each iteration, to permitting cuts at different heights on distinct branches of the hierarchical clustering tree. Our procedure is intended as a proof-of-concept for the automated generation of efficient MCMC algorithms, and to serve as a starting point for subsequent research.
0
[0]{}
Acknowledgements {#acknowledgements .unnumbered}
================
This work was supported by the NSF under grant DBI-1147230.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Partial dependence curves (FPD) introduced by @PDP, are an important model interpretation tool, but are often not accessible to business analysts and scientists who typically lack the skills to choose, tune, and assess machine learning models. It is also common for the same partial dependence algorithm on the same data to give meaningfully different curves for different models, which calls into question their precision. Expertise is required to distinguish between model artifacts and true relationships in the data.
In this paper, we contribute methods for computing partial dependence curves, for both numerical ([<span style="font-variant:small-caps;">StratPD</span>]{}) and categorical explanatory variables ([<span style="font-variant:small-caps;">CatStratPD</span>]{}), that work directly from training data rather than predictions of a model. Our methods provide a direct estimate of partial dependence, and rely on approximating the partial derivative of an unknown regression function without first fitting a model and then approximating its partial derivative. We investigate settings where contemporary partial dependence methods—including FPD, ALE, and SHAP methods—give biased results. Furthermore, we demonstrate that our approach works correctly on synthetic and plausibly on real data sets. Our goal is not to argue that model-based techniques are not useful. Rather, we hope to open a new line of inquiry into nonparametric partial dependence.
author:
- |
Terence Parr\
University of San Francisco\
[[email protected]]{}
- |
James D. Wilson\
University of San Francisco\
[[email protected]]{}
bibliography:
- 'stratpd.bib'
title: 'Technical Report: Partial Dependence through Stratification'
---
Introduction
============
Partial dependence, the isolated effect of a specific variable or variables on the response variable, $y$, is important to researchers and practitioners in many disparate fields such as medicine, business, and the social sciences. For example, in medicine, researchers are interested in the relationship between an individual’s demographics or clinical features and their susceptibility to illness. Business analysts at a car manufacturer might need to know how changes in their supply chain affect defect rates. Climate scientists are interested in how different atmospheric carbon levels affect temperature.
For an explanatory matrix, $\bf X$, with a single (column) variable, $x_1$, a plot of the $y$ against $x_1$ visualizes the marginal effect of feature $x_1$ on $y$ exactly. Given two or more features, one can similarly plot the marginal effects of each feature separately, however, the analysis is complicated by the interactions of the variables. Variable interactions and codependencies between features result in marginal plots that do not isolate the specific contribution of a feature of interest to the response. For example, a marginal plot of sex (male/female) against body weight would likely show that, on average, men are heavier than women. While true, men are also taller than women on average, which likely accounts for most of the difference in average weight. It is unlikely that two “identical” people, differing only in sex, would be appreciably different in weight.
Rather than looking directly at the data, there are several partial dependence techniques that interrogate fitted models provided by the user: Friedman’s original partial dependence [@PDP] (which we will denote FPD), functional ANOVA [@fanova], Individual Conditional Expectations (ICE) [@ICE], Accumulated Local Effects (ALE) [@ALE], and most recently SHAP [@shap]. Model-based techniques dominate the partial dependence research literature because interpreting the output of a fitted model has several advantages. For example, models have a tendency to smooth over noise. Models act like analysis preprocessing steps, potentially reducing the computational burden on model-based partial dependence techniques; e.g., ALE is $O(n)$ for the $n$ records of $\bf X$. Model-based techniques are typically model-agnostic, though for efficiency, some provide model-specific optimizations, as SHAP does. Partial dependence techniques that interrogate models also provide insight into the models themselves; i.e., how variables affect model behavior. It is also true that, in some cases, a predictive model is the primary goal so creating a suitable model is not an extra burden.
Model-based techniques do have two significant disadvantages, however. The first relates to their ability to tease apart the effect of codependent features because models are sometimes required to extrapolate into regions of nonexistent support or even into nonsensical observations; e.g., see discussions in @ALE and @fanova. As we demonstrate in [Section \[sec:experiments\]]{}, using synthetic and real data sets, model-based techniques can vary in their ability to isolate variable effects in practice. Second, there are vast armies of business analysts and scientists at work that need to analyze data, in a manner akin to exploratory data analysis (EDA), that have no intention of creating a predictive model. Either they have no need, perhaps needing only partial dependence plots, or they lack the expertise to choose, tune, and assess models (or write software at all).
![Meaningfully different results from a single partial dependence technique, FPD/ICE, applied to the same data but different models. Plots of number of bathrooms versus rent price using New York City apartment rent data [@rent] with $n=10,000$ of \~50k. (a) marginal plot, (b) plot derived from random forest, (c) plot derived from gradient boosted machine, and (d) plot derived from ordinary least squares regression. Hyper parameters were tuned using 5-fold cross validation grid search over several hyper parameters. Keras model trained by experimentation: single hidden layer of 100 neurons, 500 epochs, batch size of 1000, batch normalization, and 30% dropout. [<span style="font-variant:small-caps;">StratPD</span>]{} gives a plausible roughly result that rent goes up linearly with the number of bathrooms. $R^2$ were computed on 20% validation sets.[]{data-label="fig:baths_price"}](bathrooms_vs_price.pdf)
Even in the case where a machine learning practitioner is available to create a fitted model for the analyst, hazards exist. If a fitted model is unable to accurately capture the relationship between features and $y$ accurately, for whatever reason, then partial dependence does not provide any useful information to the user. To make interpretation more challenging, there is no definition of “accurate enough.” Also, given an accurate fitted model, business analysts and scientists are still peering at the data through the lens of the model, which can distort partial dependence curves. Separating visual artifacts of the model from real effects present in the data requires expertise in model behavior (and optimally in the implementation of model fitting algorithms).
Consider the combined FPD/ICE plots shown in [Figure \[fig:baths\_price\]]{} derived from several models (random forest, gradient boosting, linear regression, deep learning) fitted to the same New York City rent data set [@rent]. The subplots in [Figure \[fig:baths\_price\]]{}(b)-(e) present starkly different partial dependence relationships and it is unclear which, if any, is correct. The marginal plot, (a), drawn directly from the data shows a roughly linear growth in price for a rise in the number of bathrooms, but this relationship is biased because of the dependence of bathrooms on other variables, such as the number of bedrooms. (e.g., five bathroom, one bedroom apartments are unlikely.) For real data sets with codependent features, the true relationship is unknown so it is hard to evaluate the correctness of the plots. (Humans are unreliable estimators, which is why we need data analysis algorithms in the first place.) Nonetheless, having the same algorithm, operating on the same data, give meaningfully different partial dependences is undesirable and makes one question their precision.
Experts are often able to quickly recognize model artifacts, such as the stairstep phenomenon in [Figure \[fig:baths\_price\]]{}(b) and (c) inherent to decision tree-based methods trying unsuccessfully to extrapolate. In this case, though, the stairstep is more accurate than the linear relationship in (d) and (e) because the number of bathrooms is discrete (except for “half baths”). The point is that interpreting model-based partial dependence plots can be misleading, even for experts.
An accurate mechanism to compute partial dependences that did not peer through fitted models would be most welcome. Such partial dependence curves would be accessible to users, like business analysts, who lack the expertise to create suitable models. (One can imagine a spreadsheet plug-in that produced partial dependence curves.) A mechanism that did not rely on a user-provided model would also reduce the chance of plot misinterpretation due to model artifacts and could even help machine learning practitioners to choose appropriate models based upon relationships exposed in the data.
In this paper, we propose a strategy, called [<span style="font-variant:small-caps;">strat</span>ified <span style="font-variant:small-caps;">p</span>artial <span style="font-variant:small-caps;">d</span>ependence]{} ([<span style="font-variant:small-caps;">StratPD</span>]{}), that computes partial dependences directly from training data $({\bf X}, {\bf y})$, rather than through the predictions of a fitted model, and that does not presume mutually-independent features. As an example, [Figure \[fig:baths\_price\]]{}(f) shows the partial dependence plot computed by [<span style="font-variant:small-caps;">StratPD</span>]{}. [<span style="font-variant:small-caps;">StratPD</span>]{} operates very much like a “model-free” ALE, at least for numerical variables. Our technique is also based upon the notion of an idealized partial dependence: integration over the partial derivative of $y$ with respect to the variable of interest for the smooth function that generated $({\bf X}, {\bf y})$. As that function is unknown, we estimate the partial derivatives from the data non-parametrically. Colloquially, the approach examines changes in $y$ across $x_j$ while holding [$x_{\texttt{\char`\\} j}$]{} constant or nearly constant ([$x_{\texttt{\char`\\} j}$]{} denotes all variables except $x_j$). To hold [$x_{\texttt{\char`\\} j}$]{} constant, we use a single decision tree to partition feature space, a concept used by @rfimp and @RFunsup for conditional permutation importance and observation similarity measures, respectively. Our second contribution, [<span style="font-variant:small-caps;">CatStratPD</span>]{}, computes partial dependence curves for categorical variables that, unlike existing techniques, does not assume adjacent category levels are similar. Both [<span style="font-variant:small-caps;">StratPD</span>]{} and [<span style="font-variant:small-caps;">CatStratPD</span>]{} are quadratic in $n$, in the worst case (like FPD), though [<span style="font-variant:small-caps;">StratPD</span>]{} behaves linearly on real data sets. Our prototype is currently limited to regression, isolates only single-variable partial dependence, and cannot identify interaction effects (as ICE can). The software is available via Python package [stratx]{} with source at [github.com/parrt/stratx]{}, including the code to regenerate images in this paper.
We begin by describing the proposed stratification approach in [Section \[sec:stratpd\]]{} then compare [<span style="font-variant:small-caps;">StratPD</span>]{} to related (model-based) work in [Section \[sec:related\]]{}. In [Section \[sec:experiments\]]{}, we present partial dependence curves generated by [<span style="font-variant:small-caps;">StratPD</span>]{} and [<span style="font-variant:small-caps;">CatStratPD</span>]{} on synthetic and real data sets, contrast the plots with those of existing methods, and use synthetic data to highlight possible bias in some model-based methods.
Partial dependence without model predictions {#sec:stratpd}
============================================
In special circumstances, we know the precise effect of each feature $x_j$ on response $y$. Assume we are given training data pair ($\bf X, y$) where ${\bf X} = [x^{(1)}, \ldots, x^{(n)}]$ is an $n \times p$ matrix whose $p$ columns represent observed features and ${\bf y}$ is the $n \times 1$ vector of responses. For any smooth function $f:\mathbb{R}^{p} \rightarrow \mathbb{R}$ that precisely maps each ${x^{(i)}}$ row vector to response $y^{(i)}$, ${y^{(i)}} = f({x^{(i)}})$, the partial derivative of $y$ with respect to $x_j$ gives the change in $y$ holding all other variables constant. Integrating the partial derivative then gives the [*idealized partial dependence*]{} of $y$ on $x_j$, the isolated contribution of $x_j$ to $y$:
[**Definition 1**]{} The [*idealized partial dependence*]{} of $y$ on feature $x_j$ for continuous and smooth generator function $f:\mathbb{R}^{p} \rightarrow \mathbb{R}$ evaluated at $x_j = z$ is the cumulative sum up to $z$:
$$\label{eq:pd}
\text{\it PD}_j(z) = \int_{min(x_j)}^z \frac{\partial f}{\partial x_j} dx_j$$
$\text{\it PD}_j(z)$ is the value contributed to $f$ by $x_j$ at $x_j = z$ and $\text{\it PD}_j(min(x_j))=0$. The underlying generator function is unknown, and so other approaches begin by estimating $f$ with a fitted model, $\hat{f}$. Instead, we estimate the partial derivatives of the true function, ${\partial f}/{\partial x_j}$, from the raw training data then integrate to obtain $PD_j$. We note that ALE also derives partial dependence by estimating and integrating across partial derivatives (e.g., see Equation 7 in @ALE) but does so using local changes in predictions from $\hat{f}$ rather than $f$. The advantages of this $PD_j$ definition are that it ([*i*]{}) does not require a fitted model or its predictions and ([*ii*]{}) is insensitive to codependent features.
The key idea is to stratify [$x_{\texttt{\char`\\} j}$]{} feature space into disjoint regions of observations where all [$x_{\texttt{\char`\\} j}$]{} variables are approximately matched across the observations in that region. Within each [$x_{\texttt{\char`\\} j}$]{} region, any fluctuations in the response variable are likely due to the variable of interest, $x_j$. In the ideal case, the values for each [$x_{\texttt{\char`\\} j}$]{} variable within a region are identical and, because only $x_j$ is changing, $y$ changes should be attributed to $x_j$, noise, or irreducible error. (More on this shortly.) Estimates of the partial derivative within a region are computed discretely as the changes in $y$ values between unique and ordered $x_j$ positions: $(\bar{y}^{(i+1)} - \bar{y}^{(i)})/(x_j^{(i+1)} - x_j^{(i)})$ for all $i$ in a region such that $x_j^{(i)}$ is unique and $\bar{y}^{(i)}$ is the average $y$ at unique $x_j^{(i)}$. This amounts to performing piecewise linear regression through the region observations, one model per unique pair of $x_j$ values, and collecting the model $\beta_1$ coefficients to estimate partial derivatives. The overall partial derivative at $x_j=z$ is the average of all slopes, found in any region, whose range $[x_j^{(i)},x_j^{(i+1)})$ spans $z$. One could apply a nonparametric method to smooth through the discontinuities at $x_j$ points within a leaf, but this has not proven necessary in practice. The partial dependence curve points are often the result of two averaging operations, one within and one across regions, which tends to smooth the curve; e.g., see [Figure \[fig:interactions\]]{}(d) below.
Stratification occurs through the use of a decision tree fit to ([${\bf X}_{\texttt{\char`\\} j}$]{}, $\bf y$), whose leaves aggregate observations with equal or similar [$x_{\texttt{\char`\\} j}$]{} features. The [$x_{\texttt{\char`\\} j}$]{} features can be numerical variables or label-encoded categorical variables (assigned a unique integer). [<span style="font-variant:small-caps;">StratPD</span>]{} only uses the tree for the purpose of partitioning feature space and never uses predictions from any model. See [Algorithm \[alg:StratPD\]]{} for more details.
For the stratification approach to work, decision tree leaves must satisfactorily stratify [$x_{\texttt{\char`\\} j}$]{}. If the [$x_{\texttt{\char`\\} j}$]{} observations in each region are not similar enough, the relationship between $x_j$ and $y$ is less accurate. Regions can also become so small that even the $x_j$ values become equal, leaving a single unique $x_j$ observation in a leaf. Without a change in $x_j$, no partial derivative estimate is possible and these nonsupporting observations must be ignored (e.g., the two leftmost points in [Figure \[fig:partitioning\]]{}(a)). A degenerate case occurs when identical or nearly identical $x_j$ and $x_j'$ variables exist. Stratifying $x_j$ as part of [$x_{\texttt{\char`\\} j}$]{} would also match up $x_j'$ values, leading to both exhibiting flat curves, as if the decision tree were trained on $(\bf X, y)$ not ([${\bf X}_{\texttt{\char`\\} j}$]{}, $\bf y$). Our experience is that using the collection of leaves from a random forest, which restricts the number of variables available during node splitting, prevents partitioning from relying too heavily on either $x_j$ or $x_j'$. Some leaves have observations that vary in $x_j$ or $x_j'$ and partial derivatives can still be estimated.
[<span style="font-variant:small-caps;">StratPD</span>]{} uses a hyper parameter, [min\_samples\_leaf]{}, to control the minimum number of observations in each decision tree leaf. Generally speaking, smaller values lead to more confidence that fluctuations in $y$ are due solely to $x_j$, but more observations per leaf prevent [<span style="font-variant:small-caps;">StratPD</span>]{} from missing nonlinearities and make it less susceptible to noise. As the leaf size grows, however, one risks introducing contributions from [$x_{\texttt{\char`\\} j}$]{} into the relationship between $x_j$ and $y$. At the extreme, the decision tree would consist of a single leaf node containing all observations, leading to a marginal not partial dependence curve.
[<span style="font-variant:small-caps;">StratPD</span>]{} uses another hyper parameter called [min\_slopes\_per\_x]{} to ignore any partial derivatives estimated with too few observations. Dropping uncertain partial derivatives greatly improves accuracy and stability. Partial dependences computed by integrating over local partial derivatives are highly sensitive to partial derivatives computed at the left edge of any $x_j$’s range because imprecision at the left edge affects the entire curve. This presents a problem when there are few samples with $x_j$ values at the extreme left (see, for example, the $x_j$ histogram of [Figure \[fig:yearmade\]]{}(d)). Fortunately, sensible defaults for [<span style="font-variant:small-caps;">StratPD</span>]{} (10 observations and 5 slopes) work well in most cases and were used to generate all plots in this paper.
For categorical explanatory variables, [<span style="font-variant:small-caps;">CatStratPD</span>]{} uses the same stratification approach, but cannot apply regression of $y$ to non-ordinal, categorical $x_j$. Instead, [<span style="font-variant:small-caps;">CatStratPD</span>]{} groups leaf observations by category and computes the average response per category in each leaf. Consider a single leaf and its $p$-dimensional average response vector $\bar{\bf y}$. Then choose a random reference category, [*refcat*]{}, and subtract that category’s average value from $\bar{\bf y}$ to get a vector of relative deltas between categories: $\Delta {\bf y}$ = $\bar{\bf y} - \bar{\bf y}_{\it refcat}$. The $\Delta {\bf y}$ vectors from all leaves are then merged via averaging, weighted by the number of observations per category, to get the overall effect of each category on the response. The delta vectors for two leaves, $\Delta {\bf y}$ and $\Delta {\bf y}'$, can only be merged if there is at least one category in common. [<span style="font-variant:small-caps;">CatStratPD</span>]{} initializes a running average vector to an arbitrary starting leaf’s $\Delta {\bf y}$ and then makes multiple passes over the remaining vectors, merging any vectors with a category in common with the running average vector. Observations associated with any remaining, unmerged leaves must be ignored. [<span style="font-variant:small-caps;">CatStratPD</span>]{} uses a single hyper parameter [min\_samples\_leaf]{} to control stratification. Both [<span style="font-variant:small-caps;">StratPD</span>]{} and [<span style="font-variant:small-caps;">CatStratPD</span>]{} have an optional hyper parameter called [ntrials]{} (default is 1) that averages the results from multiple bootstrapped samples, which can reduce variance.
Stratification of high-cardinality categorical variables tends to create small groups of category subsets, which complicates the averaging process across groups. (Such $\Delta {\bf y}$ vectors are sparse and we use [*NaN*]{}s to represent missing values.) If both groups have the same reference category, merging is a simple matter of averaging the two delta vectors, where [*mean*]{}([*z,NaN*]{}) = [*z*]{}. For delta vectors with different reference categories and at least one category in common, one vector is adjusted to use a randomly-selected reference category, $c$, in common: $\Delta {\bf y}' = \Delta {\bf y}' - \Delta {\bf y}_c' + \Delta {\bf y}_c$. That equation adjusts vector $\Delta {\bf y}'$ so $\Delta {\bf y}_c'=0$ then adds the corresponding value from $\Delta {\bf y}$ so $\Delta {\bf y}_c' = \Delta {\bf y}_c$, which renders the average of $\Delta {\bf y}$ and $\Delta {\bf y}'$ meaningful. See [Algorithm \[alg:CatStratPD\]]{} for more details.
[<span style="font-variant:small-caps;">StratPD</span>]{} and [<span style="font-variant:small-caps;">CatStratPD</span>]{} both have theoretical worst-case time complexity of $O(n^2)$ for $n$ observations. For [<span style="font-variant:small-caps;">StratPD</span>]{}, stratification costs $O(p n \,log n)$, computing $y$ deltas for all observations among the leaves has linear cost, and averaging slopes across unique $x_j$ ranges is on the order of $|unique({\bf X}_j)| \times n$ or $n^2$ when all ${\bf X}_j$ are unique in the worst case. [<span style="font-variant:small-caps;">StratPD</span>]{} is, thus, $O(n^2)$ in the worst case. [<span style="font-variant:small-caps;">CatStratPD</span>]{} also stratifies in $O(p n \,log n)$ and computes category deltas linearly in $n$ but must make multiple passes over the $|T|$ leaves to average all possible leaf category delta vectors. In practice, three passes is the max we have seen (for high-cardinality variables), so we can assume the number of passes is some small constant to get a tighter bound. Averaging two vectors costs $|unique({\bf X}_j)|$, so each pass requires $|T| \times |unique({\bf X}_j)|$. The number of leaves is roughly $n / {\tt\small min\_samples\_leaf}$ and, worst-case, $|unique({\bf X}_j)|=n$, meaning that merging dominates [<span style="font-variant:small-caps;">CatStratPD</span>]{} complexity leading to $O(n^2)$. Experiments show that our prototype is fast enough for practical use (see [Section \[sec:experiments\]]{}).
Related work {#sec:related}
============
The mechanisms most closely related to [<span style="font-variant:small-caps;">StratPD</span>]{} and [<span style="font-variant:small-caps;">CatStratPD</span>]{} are FPD, ICE, SHAP, and ALE, which all define partial dependence in terms of impact on estimated models, $\hat{f}$, rather than the unknown true function $f$. Let $x_S$ be the subset of features of interest where $S \subset F = \{1, 2, .., p\}$. @PDP defines the partial dependence as an expectation conditioned on the remaining variables:
$${\it FPD}_S(x_S) = \operatorname{\mathbb{E}}[\hat{f}(x_S,{\bf X}_{{\texttt{\char`\\}}S})],$$
where the expectation can be estimated by $\frac{1}{n} \sum_{i=1}^n \hat{f}(x_S, x_{{\texttt{\char`\\}}S}^{(i)})$. The Individual Conditional Expectation (ICE) plot [@ICE] estimates the partial dependence of the prediction $\hat{f}$ on $x_S$, or single variable $x_j$, across individual observations. ICE produces a curve from the fitted model over all values of $x_j$ while holding [$x_{\texttt{\char`\\} j}$]{} fixed: $\hat{f}^{(i)}_j = \widehat{f}(\{x_j^{(k)}\}_{k = 1}^n, x_{{\texttt{\char`\\}}j}^{(i)})$; the FPD curve for $x_j$ is the average over all $x_j$ ICE curves. The motivation for ICE is to identify variable interactions that average out in the FPD curve.
The SHAP method from @shap has roots in [*Shapley regression values*]{} [@shapley-regression] and calculates the average marginal effect of adding $x_j$ to models, $\hat{f}_S$, trained on all possible subsets of features:
$$\label{eq:shap}
\phi_j(\hat{f},x_F) = \sum_{S \subseteq F {\texttt{\char`\\}}\{j\}}\
\frac{|S|!(|F|-|S|-1)!}{|F|!}\
[ \hat{f}_{S \cup \{j\}}(x_{S \cup \{j\}}) - \hat{f}_S(x_S) ]$$
To avoid training a combinatorial explosion of models with the various feature subsets, $\hat{f}_S(x_S)$, SHAP reuses a single model fitted to $(\bf X, y)$ by running simplified feature vectors into the model. As @manyshap describes, there are many possible implementations for simplified vectors. One is to replace “missing” features with their expected value or some other baseline vector (BShap in @manyshap). SHAP uses a more general approach (“interventional” mode) that approximates $\hat{f}_S(x_S)$ with $\operatorname{\mathbb{E}}[\hat{f}(x_{S},{\bf X}'_{{\texttt{\char`\\}}S}) | {\bf X'}_S = x_S]$ where ${\bf X}'$ is called the [*background set*]{} and users can pass in, for example, a single vector with ${\bf X}_{{\texttt{\char`\\}}S}$ column averages or even the entire training set, $\bf X$, which is what we will assume (and is called CES($\hat{D}$) in @manyshap where $\hat{D}={\bf X}$). To further reduce computation costs, SHAP users typically explain a small subsample of the data set, but with potentially a commensurate reduction in the explanatory resolution of the underlying population. SHAP has model-type-dependent optimizations for linear regression, deep learning, and decision-tree based models.
For efficiency, SHAP approximates $\operatorname{\mathbb{E}}[\hat{f}(x_{S},{\bf X}_{{\texttt{\char`\\}}S}) | {\bf X}_S = x_S]$ with $\operatorname{\mathbb{E}}[\hat{f}(x_{S},{\bf X}_{{\texttt{\char`\\}}S})]$, which assumes feature independence. [@janzing2019feature] argues that “[*unconditional*]{} \[as implemented\] rather than [*conditional*]{} \[as defined\] expectations provide the right notion of dropping features.” But, using the unconditional expectation makes the inner difference of [Equation \[eq:shap\]]{} a function of codependency-sensitive FPDs:
$$\begin{split}
\hat{f}_{S \cup \{j\}}(x_{S \cup \{j\}}) - \hat{f}_S(x_S) & = \operatorname{\mathbb{E}}[\hat{f}(x_{S \cup \{j\}},{\bf X}_{{\texttt{\char`\\}}(S \cup \{j\})})] - \operatorname{\mathbb{E}}[\hat{f}(x_{S},{\bf X}_{{\texttt{\char`\\}}S})]\\
& = \text{\it FPD}_{S \cup \{j\}}({\bf x}) - \text{\it FPD}_{S}({\bf x})
\end{split}$$
If the individual contributions are potentially biased, averaging the contributions of many such feature permutations might not lead to an accurate partial dependence. Even if the conditional expectation is used, @manyshap points out that SHAP is sensitive to the sparsity of $\bf X$ because condition ${\bf X}_S = x_S$ will find few or no training records with the exact $x_S$ values of some input vector.
The goal of ALE [@ALE] is to overcome the bias in previous model-based techniques arising from extrapolations of $\hat{f}$ far outside the support of the training data in the presence of codependent variables. ALE partitions range $[\text{min}({\bf X}_j) \,..\, \text{max}({\bf X}_j)]$ for variable $x_j$ into $K$ bins and estimates the “uncentered main effect” (Equation 15) at $x_j = z$ as the cumulative sum of the partial derivatives for all bins up to the bin containing $z$. ALE estimates the partial derivative of $\hat{f}$ at $x_j=z$ as $\operatorname{\mathbb{E}}[\hat{f}(b_k, {\bf X}_{{\texttt{\char`\\}}j}) - \hat{f}(b_{k-1}, {\bf X}_{{\texttt{\char`\\}}j}) \, | \, x_j \in (b_{k-1},b_k]]$ for bin $b_k$ that contains $z$. They also extend ALE to two variables by partitioning feature space into $K^2$ rectangular bins and computing the second-order finite difference of $\hat{f}$ with respect to the two variables for each bin.
Another related technique that integrates over partial derivatives to measure $x_j$ effects is called Integrated Gradients (IG) from @intgrad. Given a single input vector, $\bf x$, to a deep learning classifier, IG integrates over the gradient of the model output function $\hat{f}$ at points along the path from a baseline vector, $\bf x'$, to $\bf x$. IG can be seen as computing the partial dependence of $\hat{f}$ at a single $\bf x$, but using multiple $\bf x$ vectors would yield an $x_j$ partial dependence curve (relative to a baseline $\bf x'$).
[<span style="font-variant:small-caps;">StratPD</span>]{} is like a “model-free” version of ALE in that [<span style="font-variant:small-caps;">StratPD</span>]{} also defines partial dependence as the cumulative sum of partial derivatives, but we estimate derivatives using response values directly rather than $\hat{f}$ predictions. An advantage to estimating partial dependence via fitted models is that $\hat{f}$ removes variability from the potentially noisy response values, $\bf y$. However, in practice, this requires expertise to choose and tune an appropriate model for a data set. The fact that different models can lead to meaningfully different curves for the same data can lead to misinterpretation. Also, expertise is often required to distinguish between model artifacts and interesting visual phenomena arising from the data.
ALE partitions $x_j$ into bins then fixes [$x_{\texttt{\char`\\} j}$]{} as it shifts $x_j$ to bin edges to compute finite differences, as depicted in [Figure \[fig:partitioning\]]{}(b) for $p=2$. The wedges on the $x_1$ axis indicate the $x_1$ points of the computed partial dependence curve. [<span style="font-variant:small-caps;">StratPD</span>]{} partitions [$x_{\texttt{\char`\\} j}$]{} into regions of (hopefully) similar observations and computes finite differences between the average $y$ values at unique $x_j$ values in each region, as depicted in [Figure \[fig:partitioning\]]{}(a). The leftmost two observations are ignored as there is no change in $x_1$ in that leaf. The shaded area illustrates that the partial derivative at any $x_j=z$ is the average of all derivatives spanning $z$ across [$x_{\texttt{\char`\\} j}$]{} regions. [<span style="font-variant:small-caps;">StratPD</span>]{} assumes all points within a region are identical in [$x_{\texttt{\char`\\} j}$]{}, effectively projecting points in [$x_{\texttt{\char`\\} j}$]{} space onto a hyperplane if they are not. ALE shifts $x_j$ values in a small neighborhood and [<span style="font-variant:small-caps;">StratPD</span>]{} depends on a suitable [min\_samples\_leaf]{} hyper parameter to prevent [$x_{\texttt{\char`\\} j}$]{} points in a regions from becoming too dissimilar. [<span style="font-variant:small-caps;">StratPD</span>]{} automatically generates more curve points in areas of high $x_j$ density, but ALE is more efficient.
![Comparison of [<span style="font-variant:small-caps;">StratPD</span>]{}, [<span style="font-variant:small-caps;">CatStratPD</span>]{} and ALE for $p=2$ with continuous $x_2$ and continuous $x_1$ in (a), (b) and categorical $x_1$ in (c), (d). F=false and T=true. Vertical dashed lines are ALE bin edges and horizontal dashed lines are regions partitioned by a decision tree fit to ([${\bf X}_{\texttt{\char`\\} j}$]{}, ${\bf y}$). Whereas ALE shifts $x_1$ observations to bin edges holding $x_2$ exactly constant (and asks for predictions at those points), [<span style="font-variant:small-caps;">StratPD</span>]{} assumes $x_2$ values are the same and uses known $y$ values. The small wedges on $x_1$ axis indicate partial dependence curve points. The shades of green and blue indicate $y$ values. The leftmost two observations in (a) and the lowermost observation in (c) are ignored as finite differences are not defined. There is no curve value for the rightmost $x_1$ value in (a) due to forward differencing.[]{data-label="fig:partitioning"}](partitioning.pdf)
Using decision trees for the purpose of partitioning feature space as [<span style="font-variant:small-caps;">StratPD</span>]{} does was previously used by @rfimp to improve permutation importance for random forests by permuting $x_j$ only within the observations of a leaf. Earlier, @RFunsup defined a similarity measure between two observations according to how often they appear in the same leaf in a random forest. Rather than partitioning [$x_{\texttt{\char`\\} j}$]{} space like [<span style="font-variant:small-caps;">StratPD</span>]{}, those techniques partitioned all of $x$ space.
The model-based techniques under discussion treat boolean and label-encoded categorical variables (encoded as unique integers) as numerical variables, even though there is no defined order, as depicted in [Figure \[fig:partitioning\]]{}(d). ALE does, however, take advantage of the lack of order to choose an $x_j$ order that reduces “extrapolation outside the data envelope” by measuring the similarity of [${\bf X}_{\texttt{\char`\\} j}$]{} sample values across $x_j$ categories. Adjacent category integers, though, could still represent the most semantically different categories, so any shift of an observation’s category to extrapolate is risky. Even the smallest possible extrapolation can conjure up nonsensical observations, such as pregnant males, as we demonstrate in [Figure \[fig:pregnant\]]{} below where FPD, SHAP, and ALE underestimate pregnancy’s effect on body weight. (For boolean $x_j$, ALE behaves like FPD.) See [@stopperm] for more on the dangers of permuting features.
In contrast, [<span style="font-variant:small-caps;">CatStratPD</span>]{} uses a different algorithm for categoricals and computes differences between the average $y$ for all categories within the leaf to a random reference category; see [Figure \[fig:partitioning\]]{}(c). These leaf delta vectors are then merged across leaves to arrive at an overall delta vector relating the relative effect of each category on $y$. One could argue that [<span style="font-variant:small-caps;">StratPD</span>]{} also extrapolates because [$x_{\texttt{\char`\\} j}$]{} could include categorical variables and not all [$x_{\texttt{\char`\\} j}$]{} records would be identical. But, our approach only assumes [$x_{\texttt{\char`\\} j}$]{} values are similar and uses known training $y$ values for finite differences, rather than asking a model to make prediction for nonsensical records, which could be wildly inaccurate. Also, the decision tree would, by definition, likely partition [$x_{\texttt{\char`\\} j}$]{} space into regions that can be treated similarly, thus, grouping semantically similar categorical levels together.
Experimental results {#sec:experiments}
====================
In this section, we demonstrate experimentally that [<span style="font-variant:small-caps;">StratPD</span>]{} and [<span style="font-variant:small-caps;">CatStratPD</span>]{} compute accurate partial dependence curves for synthetic data and plausible results for a real data set. Experiments also provide evidence that existing model-based techniques can provide meaningfully-biased curves. We begin by comparing the partial dependence curves from popular techniques on synthetic data with complex interactions.[^1]
![Partial dependence plots of $n=2000$ data generated from noiseless $y = x_1^2 + x_1 x_2 + 5 x_1 sin(3 x_2) + 10$ where $x_1,x_2,x_3 \sim U(0,10)$ and $x_3$ does not affect $y$. The model is a random forest with 10 trees trained on all data ($R^2=0.997$). SHAP used all $\bf X$ as background data and ALE used $K=300$. The curves were generated from same 2000 data points that the model was trained on.[]{data-label="fig:interactions"}](interactions.pdf)
[Figure \[fig:interactions\]]{} illustrates that FPD, SHAP, ALE, and [<span style="font-variant:small-caps;">StratPD</span>]{} all suitably isolate the effect of independent individual variables on the response for noiseless data generated via: $y = x_1^2 + x_1 x_2 + 5 x_1 sin(3 x_2) + 10$ for $x_1,x_2,x_3 \sim U(0,10)$ where $x_3$ does not affect $y$. The shapes of the curves for all techniques look similar except that [<span style="font-variant:small-caps;">StratPD</span>]{} starts all curves at $y=0$ (as could the others). SHAP’s curves have the advantage that they indicate the presence of variable interactions. To our eye, [<span style="font-variant:small-caps;">StratPD</span>]{}’s curves are smoothest despite not having access to model predictions.
Models have a tendency to smooth out noise and a legitimate concern is that, without the benefit of a model, [<span style="font-variant:small-caps;">StratPD</span>]{} could be adversely affected. [Figure \[fig:noise\]]{} demonstrates [<span style="font-variant:small-caps;">StratPD</span>]{} curves for noisy quadratics generated from $y = x_1^2 + x_2 + 10 + \epsilon$ where $\epsilon \sim N(0,\sigma)$ $\epsilon$ and, at $\sigma=2$, 95% of the noise falls within \[0,4\] (since $2\sigma = 4$), meaning that the signal-to-noise ratio is at best 1-to-1 for $x_1^2$ in $[-2,2]$. For zero-centered Gaussian noise and this data set, [<span style="font-variant:small-caps;">StratPD</span>]{} appears resilient, though $\sigma=2$ does show considerable variation across runs. The superfluous noise variable $x_3$ in [Figure \[fig:interactions\]]{} also did not confuse [<span style="font-variant:small-caps;">StratPD</span>]{}.
![The effective noise on [<span style="font-variant:small-caps;">StratPD</span>]{} for $y = x_1^2 + x_1 + 10 + \epsilon$ where $x_1,x_2 \sim U(-2,2)$, $\epsilon \sim N(0,\sigma)$ with $\sigma \in \{0,0.5,1,2\}$. The 95% interval for amplitude of the noise in (d) for $\sigma=2$ is the same as the signal.[]{data-label="fig:noise"}](noise.pdf)
Turning to categorical variables, [Figure \[fig:statetemp\]]{} presents partial dependence curves for FPD, ALE, and [<span style="font-variant:small-caps;">CatStratPD</span>]{} derived from a noisy synthetic weather data set, where temperature varies in sinusoidal fashion over the year and with different baseline temperatures per state. (The vertical “smear” in the FPD plot shows the complete sine waves but from the side, edge on.) Variable [state]{} is independent and all plots identify the baseline temperature per state correctly.
![$y = base[x_{\it state}] + 10 sin(\frac{2\pi}{365}x_{\it dayofyear}+\pi) + \epsilon$ where $\epsilon \sim N(\mu=0, \sigma=4)$. The [*base*]{} temperature per state is $\{{\it AZ}=90, {\it CA}=70, {\it CO}=40, {\it NV}=80, {\it WA}=60\}$. Sinusoids in (a) are the average of three years’ temperature data.[]{data-label="fig:statetemp"}](dayofyear_vs_temp.pdf "fig:") ![$y = base[x_{\it state}] + 10 sin(\frac{2\pi}{365}x_{\it dayofyear}+\pi) + \epsilon$ where $\epsilon \sim N(\mu=0, \sigma=4)$. The [*base*]{} temperature per state is $\{{\it AZ}=90, {\it CA}=70, {\it CO}=40, {\it NV}=80, {\it WA}=60\}$. Sinusoids in (a) are the average of three years’ temperature data.[]{data-label="fig:statetemp"}](state_vs_temp_pdp.pdf "fig:") ![$y = base[x_{\it state}] + 10 sin(\frac{2\pi}{365}x_{\it dayofyear}+\pi) + \epsilon$ where $\epsilon \sim N(\mu=0, \sigma=4)$. The [*base*]{} temperature per state is $\{{\it AZ}=90, {\it CA}=70, {\it CO}=40, {\it NV}=80, {\it WA}=60\}$. Sinusoids in (a) are the average of three years’ temperature data.[]{data-label="fig:statetemp"}](state_ale.pdf "fig:") ![$y = base[x_{\it state}] + 10 sin(\frac{2\pi}{365}x_{\it dayofyear}+\pi) + \epsilon$ where $\epsilon \sim N(\mu=0, \sigma=4)$. The [*base*]{} temperature per state is $\{{\it AZ}=90, {\it CA}=70, {\it CO}=40, {\it NV}=80, {\it WA}=60\}$. Sinusoids in (a) are the average of three years’ temperature data.[]{data-label="fig:statetemp"}](state_vs_temp_stratpd.pdf "fig:")
The primary goal of the stratification approach proposed in this paper is to obtain accurate partial dependence curves in the presence of codependent variables. To test [<span style="font-variant:small-caps;">StratPD</span>]{}/[<span style="font-variant:small-caps;">CatStratPD</span>]{} and discover potential bias in existing techniques, we synthesized a body weight data set generated by the following equation with nontrivial codependencies between variables:
$$\label{eq:weight}
{
\begin{array}{rll}
y & = &120 + 10(x_{height} - min(x_{height})) + 40x_{pregnant} - 1.5x_{education}\\[2pt]
\vspace{-10pt}\\[2pt]
\multicolumn{2}{r}{\text{where}} & x_{sex} \sim Bernoulli(\{M,F\}, p=0.5)\\[5pt]
& & x_{pregnant} = \begin{cases}
Bernoulli(\{0,1\},p=0.5) & \text{ if } x_{sex} = F\\
0 & \text{ if } x_{sex}=M\\
\end{cases}\\[15pt]
& & x_{height} = \begin{cases}
5*12+5+ \epsilon & \text{ if } x_{sex}=F,~ \epsilon \sim U(-4.5,5)\\
5*12+8 + \epsilon & \text{ if } x_{sex}=M,~ \epsilon \sim U(-7,8)\\
\end{cases}\\[15pt]
& & x_{education} = \begin{cases}
12 + \epsilon & \text{ if } x_{sex}=F,~ \epsilon \sim U(0,8)\\
10 + \epsilon & \text{ if } x_{sex}=M,~ \epsilon \sim U(0,8)\\
\end{cases}
\end{array}
}$$
The partial derivative of $y$ with respect to $x_{height}$ is 10 (holding all other variables constant), so the optimal partial dependence curve is a line with slope 10. [Figure \[fig:heightweight\]]{} illustrates the curves for the techniques under consideration, with ALE and [<span style="font-variant:small-caps;">StratPD</span>]{} giving the sharpest representation of the linear relationship. ([<span style="font-variant:small-caps;">StratPD</span>]{}’s curve is drawn on top of the SHAP plots using the righthand scale.) The FPD and both SHAP plots also suggest a linear relationship, albeit with a little less precision. The ICE curves in [Figure \[fig:heightweight\]]{}(a) and “fuzzy” SHAP curves have the advantage that they alert users to variable dependencies or interaction terms. On the other hand, the kink in the partial dependence curve and other visual phenomena could confuse less experienced machine learning practitioners and certainly analysts and researchers in other fields (our primary target communities).
![Partial dependence plots of response body weight on feature $x_{height}$ using $n$=2000 synthetic observations from [Equation \[eq:weight\]]{}. FPD and SHAP have “kinks” at the maximum female height. SHAP defines feature importance as the average SHAP value magnitude, which overemphasizes importance for heights below 70 inches here. The male/female ratio is 50/50, half of the women are pregnant, and pregnancy contributes 40 pounds. SHAP interrogated an RF tuned via 5-fold cross validation grid search (OOB $R^2$ 0.999) and explained all 2000 samples; the interventional case used 100 observations as background data. ALE used $K=300$.[]{data-label="fig:heightweight"}](height_vs_weight_pdp.pdf "fig:") ![Partial dependence plots of response body weight on feature $x_{height}$ using $n$=2000 synthetic observations from [Equation \[eq:weight\]]{}. FPD and SHAP have “kinks” at the maximum female height. SHAP defines feature importance as the average SHAP value magnitude, which overemphasizes importance for heights below 70 inches here. The male/female ratio is 50/50, half of the women are pregnant, and pregnancy contributes 40 pounds. SHAP interrogated an RF tuned via 5-fold cross validation grid search (OOB $R^2$ 0.999) and explained all 2000 samples; the interventional case used 100 observations as background data. ALE used $K=300$.[]{data-label="fig:heightweight"}](weight_tree_path_dependent_shap.pdf "fig:") ![Partial dependence plots of response body weight on feature $x_{height}$ using $n$=2000 synthetic observations from [Equation \[eq:weight\]]{}. FPD and SHAP have “kinks” at the maximum female height. SHAP defines feature importance as the average SHAP value magnitude, which overemphasizes importance for heights below 70 inches here. The male/female ratio is 50/50, half of the women are pregnant, and pregnancy contributes 40 pounds. SHAP interrogated an RF tuned via 5-fold cross validation grid search (OOB $R^2$ 0.999) and explained all 2000 samples; the interventional case used 100 observations as background data. ALE used $K=300$.[]{data-label="fig:heightweight"}](weight_interventional_shap.pdf "fig:") ![Partial dependence plots of response body weight on feature $x_{height}$ using $n$=2000 synthetic observations from [Equation \[eq:weight\]]{}. FPD and SHAP have “kinks” at the maximum female height. SHAP defines feature importance as the average SHAP value magnitude, which overemphasizes importance for heights below 70 inches here. The male/female ratio is 50/50, half of the women are pregnant, and pregnancy contributes 40 pounds. SHAP interrogated an RF tuned via 5-fold cross validation grid search (OOB $R^2$ 0.999) and explained all 2000 samples; the interventional case used 100 observations as background data. ALE used $K=300$.[]{data-label="fig:heightweight"}](height_ale.pdf "fig:")
Even for experts, explaining this behavior requires some thought, and one must distinguish between model artifacts and interesting phenomena. The discontinuity at the maximum female height location arises partially from the model having trouble extrapolating for extremely tall pregnant women. Consider one of the upper ICE lines in [Figure \[fig:heightweight\]]{}(a) for a pregnant woman. As the ICE line slides $x_{height}$ above the maximum height for a woman, the model leaves the support of the training data and predicts a [*lower*]{} weight as height increases (there are no pregnant men in the training data). ALE’s curve is straight because it focuses on local effects, demonstrating that the lack of sharp slope-10 lines for FPD and SHAP cannot be attributed simply to a poor choice of model.
![Partial dependence bar charts for boolean $x_{pregnant}$ with the same model from [Figure \[fig:heightweight\]]{}. Only [<span style="font-variant:small-caps;">CatStratPD</span>]{} gets the correct 40 pound contribution per [Equation \[eq:weight\]]{}.[]{data-label="fig:pregnant"}](pregnant_vs_weight_pdp.pdf "fig:") ![Partial dependence bar charts for boolean $x_{pregnant}$ with the same model from [Figure \[fig:heightweight\]]{}. Only [<span style="font-variant:small-caps;">CatStratPD</span>]{} gets the correct 40 pound contribution per [Equation \[eq:weight\]]{}.[]{data-label="fig:pregnant"}](pregnant_vs_weight_shap.pdf "fig:") ![Partial dependence bar charts for boolean $x_{pregnant}$ with the same model from [Figure \[fig:heightweight\]]{}. Only [<span style="font-variant:small-caps;">CatStratPD</span>]{} gets the correct 40 pound contribution per [Equation \[eq:weight\]]{}.[]{data-label="fig:pregnant"}](pregnant_2_ale.pdf "fig:") ![Partial dependence bar charts for boolean $x_{pregnant}$ with the same model from [Figure \[fig:heightweight\]]{}. Only [<span style="font-variant:small-caps;">CatStratPD</span>]{} gets the correct 40 pound contribution per [Equation \[eq:weight\]]{}.[]{data-label="fig:pregnant"}](pregnant_vs_weight_stratpd.pdf "fig:")
Also, SHAP defines $x_j$ feature importance as the average magnitude of the $x_j$ SHAP values, which introduces a paradox. The spread of the SHAP values alerts users to variable interactions, but allows contributions from other variables to leak in, thus, potentially leading to less precise estimates of $x_{height}$’s importance. The average SHAP magnitude skews upward, in this case, because of the contributions from pregnant women.
It is conceivable that a more sophisticated model (in terms of extrapolation) could sharpen the FPD and SHAP curves for $x_{height}$. There is a difference, however, between extrapolating to a meaningful but unsupported vector and making predictions for nonsensical vectors arising from variable codependencies. Techniques that rely on such predictions make the implicit assumption of variable independence, introducing the potential for bias. Consider [Figure \[fig:pregnant\]]{} that presents the partial dependence results for categorical variable $x_{pregnant}$ (same data set). The weight gain from pregnancy is 40 pounds per [Equation \[eq:weight\]]{}, but only [<span style="font-variant:small-caps;">CatStratPD</span>]{} identifies that exact relationship; FPD, SHAP, and ALE show a gain of 30 pounds.
[<span style="font-variant:small-caps;">CatStratPD</span>]{} stratifies persons with the same or similar sex, education, and height into groups and then examines the relationship between $x_{pregnant}$ and $y$. If a group contains both pregnant and nonpregnant females, the difference in weight will be 40 pounds in this noiseless data set (if we assume identical [$x_{\texttt{\char`\\} j}$]{}). FPD and ALE rely on computations that require fitted models to conjure up predictions for nonsensical records representing pregnant males (e.g., $\hat{f}(x_j = pregnant, {\bf X}_{{\texttt{\char`\\}}j})$). Not even a human knows how to estimate the weight of a pregnant male. SHAP, per its definition, does not require such predictions, but in practice for efficiency reasons, SHAP approximates $\operatorname{\mathbb{E}}[\hat{f}(x_{j} = pregnant,{\bf X}_{{\texttt{\char`\\}}j}) | {\bf X}_j = x_j]$ with $\operatorname{\mathbb{E}}[\hat{f}(x_{j} = pregnant,{\bf X}_{{\texttt{\char`\\}}j})]$, which does not restrict pregnancy to females. As discussed above, there are advantages to all of these model-based techniques, but this example demonstrates there is potential for partial dependence bias.
![Partial dependence curves of bulldozer [YearMade]{} versus [SalePrice]{} for FPD/ SHAP, ALE, and [<span style="font-variant:small-caps;">StratPD</span>]{}. $n$=20,000 observations drawn from \~362k. SHAP interrogated a random forest (OOB $R^2=0.85$) to explain 1000 training observations with 100 observations as background data. ALE used $K=300$. Hyper parameters were tuned using 5-fold cross validation grid search over several hyper parameters.[]{data-label="fig:yearmade"}](bulldozer_YearMade_pdp.pdf "fig:") ![Partial dependence curves of bulldozer [YearMade]{} versus [SalePrice]{} for FPD/ SHAP, ALE, and [<span style="font-variant:small-caps;">StratPD</span>]{}. $n$=20,000 observations drawn from \~362k. SHAP interrogated a random forest (OOB $R^2=0.85$) to explain 1000 training observations with 100 observations as background data. ALE used $K=300$. Hyper parameters were tuned using 5-fold cross validation grid search over several hyper parameters.[]{data-label="fig:yearmade"}](bulldozer_YearMade_shap.pdf "fig:") ![Partial dependence curves of bulldozer [YearMade]{} versus [SalePrice]{} for FPD/ SHAP, ALE, and [<span style="font-variant:small-caps;">StratPD</span>]{}. $n$=20,000 observations drawn from \~362k. SHAP interrogated a random forest (OOB $R^2=0.85$) to explain 1000 training observations with 100 observations as background data. ALE used $K=300$. Hyper parameters were tuned using 5-fold cross validation grid search over several hyper parameters.[]{data-label="fig:yearmade"}](bulldozer_YearMade_ale.pdf "fig:") ![Partial dependence curves of bulldozer [YearMade]{} versus [SalePrice]{} for FPD/ SHAP, ALE, and [<span style="font-variant:small-caps;">StratPD</span>]{}. $n$=20,000 observations drawn from \~362k. SHAP interrogated a random forest (OOB $R^2=0.85$) to explain 1000 training observations with 100 observations as background data. ALE used $K=300$. Hyper parameters were tuned using 5-fold cross validation grid search over several hyper parameters.[]{data-label="fig:yearmade"}](bulldozer_YearMade_stratpd.pdf "fig:")
The stratification approach also gives plausible results for real data sets, such as the bulldozer auction data from @bulldozer. [Figure \[fig:yearmade\]]{} shows the partial dependence curves for the same set of techniques as before on feature [YearMade]{}, chosen as a representative because it is very predictive of sale price. The shape and magnitude of the FPD, SHAP, ALE, and [<span style="font-variant:small-caps;">StratPD</span>]{} curves are similar, indicating that older bulldozers are worth less at auction, which is plausible. The [<span style="font-variant:small-caps;">StratPD</span>]{} curve shows 10 bootstrapped trials where the heavy black dots represent the partial dependence curve and the other colored curves describe the variability.
![Partial dependence curves of bulldozer [ProductSize]{} versus [SalePrice]{} for FPD/ICE, SHAP, ALE, and [<span style="font-variant:small-caps;">StratPD</span>]{}. $n$=20,000 observations drawn from \~362k. Same model, setup as in [Figure \[fig:yearmade\]]{}. The [<span style="font-variant:small-caps;">StratPD</span>]{} dots are the average of 10 bootstrapped trials and the partial dependence for [ProductSize]{} is missing because a forward difference is unavailable at the right edge.[]{data-label="fig:ProductSize"}](bulldozer_ProductSize_pdp.pdf "fig:") ![Partial dependence curves of bulldozer [ProductSize]{} versus [SalePrice]{} for FPD/ICE, SHAP, ALE, and [<span style="font-variant:small-caps;">StratPD</span>]{}. $n$=20,000 observations drawn from \~362k. Same model, setup as in [Figure \[fig:yearmade\]]{}. The [<span style="font-variant:small-caps;">StratPD</span>]{} dots are the average of 10 bootstrapped trials and the partial dependence for [ProductSize]{} is missing because a forward difference is unavailable at the right edge.[]{data-label="fig:ProductSize"}](bulldozer_ProductSize_shap.pdf "fig:") ![Partial dependence curves of bulldozer [ProductSize]{} versus [SalePrice]{} for FPD/ICE, SHAP, ALE, and [<span style="font-variant:small-caps;">StratPD</span>]{}. $n$=20,000 observations drawn from \~362k. Same model, setup as in [Figure \[fig:yearmade\]]{}. The [<span style="font-variant:small-caps;">StratPD</span>]{} dots are the average of 10 bootstrapped trials and the partial dependence for [ProductSize]{} is missing because a forward difference is unavailable at the right edge.[]{data-label="fig:ProductSize"}](bulldozer_ProductSize_ale.pdf "fig:") ![Partial dependence curves of bulldozer [ProductSize]{} versus [SalePrice]{} for FPD/ICE, SHAP, ALE, and [<span style="font-variant:small-caps;">StratPD</span>]{}. $n$=20,000 observations drawn from \~362k. Same model, setup as in [Figure \[fig:yearmade\]]{}. The [<span style="font-variant:small-caps;">StratPD</span>]{} dots are the average of 10 bootstrapped trials and the partial dependence for [ProductSize]{} is missing because a forward difference is unavailable at the right edge.[]{data-label="fig:ProductSize"}](bulldozer_ProductSize_stratpd.pdf "fig:")
As a second example, consider the curves for feature [ProductSize]{} in [Figure \[fig:ProductSize\]]{}. All plots have roughly the same shape but the [<span style="font-variant:small-caps;">StratPD</span>]{} plot lacks a dot for [ProductSize]{}=5 because forward differences are unavailable at the right edge. (We anticipate switching switching to a central difference to avoid this issue with low-cardinality discrete variables.) It also appears that [<span style="font-variant:small-caps;">StratPD</span>]{} considers [ProductSize]{}’s 3 and 4 to be worth less than suggested by the other techniques, though [<span style="font-variant:small-caps;">StratPD</span>]{}’s might be in line with the average SHAP plot values.
![Time to compute partial dependence curve for up to 30,000 observations for 40 numerical and 11 categorical variables. $p=20$, bulldozer $p=14$, and flight $p=17$. The Numba just-in-time compiler is used to improve performance, but timing does not include compilation; users do experience this “warm-up” time.[]{data-label="fig:timing"}](timing.pdf)
And, finally, an important consideration for any tool is performance, so we plotted execution time versus data size (up to 30,000 observations) for three real Kaggle data sets: rent, bulldozer, and flight arrival delays [@flights]. [Figure \[fig:timing\]]{} shows growth curves for 40 numerical variables and 11 categorical variables grouped by type of variable. For these data sets, [<span style="font-variant:small-caps;">StratPD</span>]{} takes 1.2s or less to process 30,000 records for any numerical $x_j$, despite the potential for quadratic cost. [<span style="font-variant:small-caps;">CatStratPD</span>]{} typically processes categorical variables in less than 2s but takes 13s for the high-cardinality categorical [ModelID]{} of bulldozer (which looks mildly quadratic). These elapsed times for our prototype show it to be practical and competitive with FPD/ICE, SHAP, and ALE. If the cost to train a model using (cross validated) grid search for hyper parameter tuning is included, [<span style="font-variant:small-caps;">StratPD</span>]{} and [<span style="font-variant:small-caps;">CatStratPD</span>]{} outperform these existing techniques (as training and tuning is often measured in minutes).
Conclusion and future work
==========================
In this paper, we contribute a method for computing partial dependence curves, for both numerical and categorical explanatory variables, that does not use predictions from a fitted model. Working directly from the data makes partial dependences accessible to business analysts and scientists not qualified to choose, tune, and assess machine learning models. For experts, it can provide hints about the relationships in the data to help guide their choice of model. Our experiments show that [<span style="font-variant:small-caps;">StratPD</span>]{} and [<span style="font-variant:small-caps;">CatStratPD</span>]{} are fast enough for practical use and correctly identify partial dependences for synthetic data and give plausible curves on real data sets. [<span style="font-variant:small-caps;">StratPD</span>]{} relies on two important hyper parameters (with broadly applicable defaults) but model-based techniques should include the hyper parameters of the required fitted model for a fair comparison. Our goal here is not to argue that model-based techniques are not useful. Rather, we are pointing out potential issues and hoping to open a new line of nonparametric inquiry that experiments have shown to be applicable in situations and accurate in cases where model-based techniques are not. An obvious next goal for this approach is a version for classifiers and to extend the technique to two variables, paralleling ALE’s second order derivative approach.
Appendix
========
Pseudocode for [<span style="font-variant:small-caps;">StratPD</span>]{} and [<span style="font-variant:small-caps;">CatStratPD</span>]{}.\
$T$ := Decision tree regressor fit to ([${\bf X}_{\texttt{\char`\\} j}$]{}, $\bf y$) with hyper parameter: ${\it min\_samples\_leaf}$ ${\bf ux} := \{x_j^{(i)}\}_{i \in 1..n}$, the unique and ordered $x_j$ in ${\bf X}_j$\
Let $\bf c$ and $\delta$ be vectors of length $|\bf ux|$\
($\triangleright$ [**]{})[each $x \in {\bf ux}$]{}[ ${\bf slopes}_x$ := \[$slope$ for $(a,b,slope) \in \bf D$ if $x \ge a$ and $x <b$\]\
${\bf c}_x$ := $|{\bf slopes}_x|$ ${\delta}_x$ := $\overline{\bf slopes}_x$ ]{} ${\bf \delta}$ := ${\bf \delta}[{\bf c} \ge min\_slopes\_per\_x]$ ${\bf ux}$ := ${\bf ux}[{\bf c} \ge min\_slopes\_per\_x]$ ${\bf pd}_x$ := ${\bf ux}^{(k+1)} - {\bf ux}^{(k)}$ for $k = 1..|{\bf ux}|-1$\
${\bf pd}_y$ := \[0\] + cumulative\_sum($\delta \times {\bf pd}_x$) \[alg:StratPD\]
$n_{cats} := |\{x_j^{(i)}\}_{i \in 1..n}|$, $n_{leaves} := |T|$\
$T$ := Decision tree regressor fit to ([${\bf X}_{\texttt{\char`\\} j}$]{}, $\bf y$) with hyper parameter: ${\it min\_samples\_leaf}$ Let $\Delta Y$ be a $n_{cats} \times n_{leaves}$ matrix whose columns, $\Delta Y_L$, are vectors of leaf category deltas\
Let $C$ be a $n_{cats} \times n_{leaves}$ matrix whose columns, $C_L$, are vectors for leaf category counts\
($\triangleright$ [**]{})[each leaf $L \in T$]{}[ ${\bf x}_L := \{x_j^{(i)}\}_{i \in L}$, the unique $x_j$ categories in $L$ $\bar{\bf y}_L^{(k)} := \operatorname{\mathbb{E}}[ y^{(i)} \,|\, (x_j^{(i)}=x_L^{(k)}, y^{(i)})] \text{ for } k = 1..|{\bf x}_L|, i \in L$ $C_L^{(k)} := |i \in L : x_j^{(i)}={\bf x}_L^{(k)}|$ ${\it refcat}$ := random category chosen from ${\bf x}_L$\
$\Delta {Y}_L$ = $\bar{\bf y}_L - \bar{\bf y}_L^{\it refcat}$\
]{} $\Delta {\bf y}$, [**c**]{} := $\Delta {Y}_1$, $C_1$ $completed$ := $\{L_1\}$; $work$ := $\{L_2 .. L_{n_{leaves}}\}$; ($\triangleright$ [**]{})[$|work| > 0$ and $|completed|>0$]{}[ completed := $\emptyset$\
$work := work {\texttt{\char`\\}}{\it completed}$\
]{}\
\[alg:CatStratPD\]
[^1]: All simulations in this section were run on a 4.0 Ghz 32G RAM machine running OS X 10.13.6 with SHAP 0.34, scikit-learn 0.21.3, XGBoost 0.90, TensorFlow 2.1.0, and Python 3.7.4; ALEPlot 1.1 and R 3.6.3. A single random seed was used across simulations for graph reproducibility purposes.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A radio labeling of a connected graph $G$ is a function $f:V(G) \to \mathbb Z_{0}^+$ such that for every pair of vertices $u,v \in V(G)$, we have ${\lvertf(u)-f(v)\rvert} \ge {\normalfont \text{diam}}(G) + 1 - d(u,v)$ where ${\normalfont \text{diam}}(G)$ denotes the diameter of $G$ and $d(u,v)$ the distance between vertices $u$ and $v$. Let ${\normalfont \text{span}}(f)$ be the difference between the greatest label and least label assigned to $V(G)$. Then, the *radio number* of a graph ${\normalfont \text{rn}}(G)$ is defined as the minimum value of ${\normalfont \text{span}}(f)$ over all radio labelings of $G$. There have been few results on the radio number of the grid graph: (blank) gave an upper and lower bound for square grids , and (blank) were unable to completely determine the radio number of the ladder graph (a 2 by $n$ grid). In this paper, we completely determine the radio number of the grid graph, characterizing five subcases of the problem and providing a closed form solution to each. These results have implications in the optimization of radio frequency assignment in wireless networks such as cell towers and environmental sensors.'
author:
- 'Tian-Shun Allan Jiang'
date: 'August 1, 2013'
title: The Radio Number of Grid Graphs
---
Introduction {#sec:intro}
============
Given a graph $G$, a *graph labeling* is a function that takes the vertices of a graph $V(G)$ to a subset of the integers, subject to certain constraints. One of the most well-known graph labeling problems is determining the chromatic number of a graph $G$, the smallest number of colors needed to color $V(G)$ so that no two adjacent vertices share the same color. The subject of our investigation, the radio labeling, is an extension of the chromatic number problem, as the label of each vertex is constrained by all other vertices of the graph, not just adjacent vertices.
Radio labeling is motivated by the channel assignment problem introduced by Hale in [@hale; @original; @problem], who defined a whole class of graph labeling problems designed to optimize wireless network efficiency. Suppose a wireless network is composed of a set of radio stations, all separated by some geographic distance. If we treat each radio station as a vertex, and assign an edge between two stations if they are geographically “close," then the radio labeling of the graph is closely related to the network’s optimal frequency assignment. Finding the radio number of a network’s graph is essential to reduce the range of required frequencies used by the network. This efficient usage of wireless frequency spectra helps reduce crowding on heavily used frequency bands, such as the 2.4 GHz band which is used for wireless internet, cell phones, microwaves, and bluetooth, and thus has the potential to improve the quality of wireless communications.
The fact that determining ${\normalfont \text{rn}}(G)$ in general is NP-complete (blank) motivates us to determine the radio number on special classes of graphs. We chose to investigate the grid graph for its closeness to real-world applications. In Section \[sec:prelim\] of this paper, we introduce some definitions and existing work. Then, in Sections \[sec:utn\] and \[sec:bump\] we investigate the *upper traceable number* and *bump* on grid graph $G_{a,b}$, two important metrics in determining ${\normalfont \text{rn}}(G_{a,b})$. In Section \[sec:grid\] we determine the value of ${\normalfont \text{rn}}(G_{a,b})$ with $a,b \ge 3$, while in Section \[sec:ladder\] we determine the value of the ladder graph ${\normalfont \text{rn}}(G_{2,b})$, which is a special case.
Preliminaries {#sec:prelim}
=============
For ease of notation, we write ${\normalfont \text{diam}}(G)$ as $D$. Now, Li [@m-ary; @trees] gives the following definition:
A radio labeling of a connected graph $G$ is a mapping $f : V(G) \to \mathbb Z_{0}^+$ such that ${\lvertf(u) - f(v)\rvert} \ge D +1 - d(u,v)$ for each pair of distinct vertices $u,v \in V(G)$, where $d(u, v)$ is the distance between $u$ and $v$. The span of $f$ is defined as $\max_{u,v \in V(G)} {\lvertf(u) - f(v)\rvert}$, and the radio number of $G$ is the minimum span of a radio labeling of $G$.
Based on this definition we observe the following fact:
\[fact:no repeat\] In a radio labeling, no two vertices may have the same label.
By Fact \[fact:no repeat\], given any radio labeling $f$ on a graph $G$, we may find a unique ordering of vertices by increasing label.
Let this ordering on $f$ be $s_f: u_{1}, u_{2}, \ldots u_{n}$. Notice that there exist multiple $f$’s with ordering $s_f$ (ie. there is a many-to-one relationship between $f$ and $s_f$). Conversely, when given an ordering $s_f$, we are interested in the set of corresponding $f$ with minimum span.
For convenience, let $f_i = f(u_{i+1}) - f(u_{i})$ and $d_i = d(u_{i+1},u_{i})$. Then, by the definition of a radio labeling we have $$f_i \ge D + 1 - d_i.$$ By taking the sum of all inequalities for $1 \le i \le n-1$, we have $${\normalfont \text{span}}(f) = \sum_{i=1}^{n-1}f_i \ge (n-1)(D+1) - \sum_{i=1}^{n-1}d_i.$$ Now, let ${\normalfont \text{rl}}(G)$ be the set of radio labelings on $G$. Since the radio number is the minimum span of a labeling, we get
The lower bound of the radio number on any connected graph $G$ is $${\normalfont \text{rn}}(G) \ge (n-1)(D+1) - \max_{f \in {\normalfont \text{rl}}(G)}\left(\sum_{i=1}^{n-1} d_i \right).$$
To ease notation, the $\max$ and $\min$ functions are always maximizing or minimizing an expression over the set ${\normalfont \text{rl}}(G)$. For completeness, we also present an upper bound on the radio number on any connected graph $G$.
\[lem:upper bound\] ${\normalfont \text{rn}}(G) = O(n^2)$. In particular, $${\normalfont \text{rn}}(G) \le D \cdot (n-1) \le (n-1)^2.$$
Since $D \le n-1 \implies D \cdot (n-1) \le (n-1)^2$, it remains to be shown that ${\normalfont \text{rn}}(G) \le D \cdot (n-1).$
If the vertices of $G$ are in any order $v_{1},v_{2},\ldots,v_{n}$, let $f(v_{i})=D \cdot (i-1)$ for $1 \le i \le n-1$. This radio labeling has span $D \cdot (n-1)$ as specified.
We now introduce some terminology to more accurately describe ${\normalfont \text{rn}}(G)$:
A *bump* is defined in relation to two consecutive vertices in the ordering $s_f$. $$b(u_{i+1},u_{i}) := f_i - (D + 1 - d_i).$$
For convenience, we denote $d(u_{i+1},u_{i})$ with $b_i$.
A vertex $u_i$’s set of *tightness neighbors* ${\normalfont \text{tn}}(u_i)$ consists of all vertices $u_j \in V(G)$ such that $${\lvertf(u_i)-f(u_j)\rvert} = D+1 - d(u_i,u_j).$$
\[rem:additional value\] Notice that $f_i = D+1 - d_i + b_i$. This shows that $b_i > 0$ if and only if $u_{i-1} \not\in {\normalfont \text{tn}}(u_i)$.
Using the definition of the bump, we get
Given a simple, connected graph $G$, $${\normalfont \text{rn}}(G) = (n-1)(D+1) - \max\left(\sum_{i=1}^{n-1}\left(d_i-b_i\right)\right) \ge (n-1)(D+1) - \left(\max\sum_{i=1}^{n-1}d_i - \min\sum_{i=1}^{n-1}b_i\right).$$
We discuss $\max\left(\sum_{i=1}^{n-1}d_i\right)$, also known as the *upper traceable number* $t^+(G)$ in Section \[sec:utn\] and $\min\left(\sum_{i=1}^{n-1}b_i\right)$ in Section \[sec:bump\].
Upper Traceable Number of Grid Graph {#sec:utn}
====================================
Let $G_{a,b}$ denote the grid graph with $a$ vertices in the $x$-direction and $b$ vertices in the $y$-direction. For ease of notation, given an ordering $s$, let $d(s) = \sum_{i=1}^{n-1}d_i$. We establish the following theorem on $t^+(G_{a,b})$:
\[thm:utn grid\] $$t^+(G_{a,b}) =
\begin{cases}
\dfrac{a^2b+b^2a}{2}-1, & \normalfont\text{if }a,b\text{ are even, }a,b > 2
\\[1em]
\dfrac{a^2b+b^2a-a}{2}-1, & \normalfont\text{if }a\text{ is even, }b \text{ is odd}
\\[1em]
\dfrac{a^2b+b^2a-a-b}{2}-1, & \normalfont\text{if }a,b\text{ are odd, }a,b > 2
\end{cases}$$
We prove the theorem with the following Lemma:
\[lem:max x distance\] Let $d_x(s)$ denote the sum of the $x$-components of each distance $d_i$ in ordering $s$. Then, $$\max(d_x(s)) =
\begin{cases}
\dfrac{a^2b}{2}-1, & \normalfont\text{if }a\text{ is even}\\[1em]
\dfrac{(a^2-1)b}{2}, & \normalfont\text{if }a\text{ is odd}
\end{cases}$$
First, we construct an ordering $s^*$ that attains the value of $\max(d_x(s))$ given in Lemma \[lem:max x distance\] above. Denote the $x$-value of a vertex $u_i$ by $x_i$.
**Case 1: a is even**
![Constructing an ordering $s$ that attains $\max(d_x(s))$ on $G_{6,4}$[]{data-label="fig:max x even"}](G_{6,4}_region_arrows.png)
1. $x_1 = \frac{a}{2}$ and $x_n = \frac{a}{2}+1$.
2. Each successive $u_{i+1}$ alternates from Region A to Region B, where Region A consists of all vertices such that $x_j \le \frac{a}{2}$, and Region B consists of all vertices such that $x_k \ge \frac{a}{2}+1$. See Figure \[fig:max x even\].
**Case 2: a is odd**
![Example of regions A, B and X on $G_{5,4}$[]{data-label="fig:max x odd"}](G_{5,4}_region_arrows.png)
1. $x_1 = x_n = \frac{a+1}{2}$.
2. There does not exist a pair $u_i, u_{i+1}$ which are both in Region A or both in Region B, where Region A consists of all vertices such that $x_j < \frac{a+1}{2}$, and Region B consists of all vertices such that $x_k > \frac{a+1}{2}$, and Region X consists of all vertices such that $x_l = \frac{a+1}{2}$. See Figure \[fig:max x odd\].
Now, we establish an upper bound on the value of $d_x(s)$. Since $d_x(u_{i+1},u_{i}) = {\lvertx_{i+1}-x_{i}\rvert}$, one vertex contributes a positive value and the other a negative value to the distance calculation. Notice that we can represent distances by labeling the graph with $+$ and $-$ signs, as in Figure \[fig:+- distance\]. Looking at $d_x(s) = \sum_{i=1}^{n-1}{\lvertx_{i+1}-x_i\rvert}$, we see that the sum has $2$ of each $x_2$ through $x_{n-1}$, and 1 of each term $x_1$ and $x_n$. Half of the terms in the sum are positive, and the other half are negative. In fact, the value of $d_x(s)$ can be modeled as such: given multisets $A = \{a_1,a_2,\ldots,a_{2n}\}$ and $B = \{b_1,b_2,\ldots,b_{2n}\}$, where $$A = (\{-1, 0 , 1\}, \{(-1, n-1),(0, 2),(1, n-1)\}) \text{ and}$$ $$B = (\{1,2,3,\ldots,n\}, \{(1,2),(2,2),(3,2),\ldots,(a,2)\}),$$ any value of $d_x(s)$ can be modeled by the quantity $a_1b_1+a_2b_2+\ldots+a_{2n}b_{2n}$. By the rearrangement inequality, the maximum sum of $d_x(s) = a_1b_1+a_2b_2+\ldots+a_{2n}b_{2n}$ occurs when $A$ and $B$ are both sorted in nondecreasing order, giving us an upper bound.
![Using $+$ and $-$ signs to represent $x$-component distance. Above, we have $d_x(X,Y) = -1 + 3 = 2$[]{data-label="fig:+- distance"}](+-distance_example_labeled.png)
When depicting this maximum sum using $+$ and $-$ signs on the grid graph, there are two configurations based on whether $a$ is even or odd. *When $a$ is even:*
- Region B contains all $+$ signs, and Region A contains all $-$ signs.
- The two vertices with 1 sign have $x$-values of $\frac{a}{2}$ and $\frac{a}{2}+1$ respectively.
*When $a$ is odd:*
- Region B contains all $+$ signs, and Region A contains all $-$ signs.
- There are an equal number of $+$ and $-$ signs in Region X, and the two vertices with 1 label lie in Region X.
Notice that if we represent $d_x(s^*)$ with $+$ and $-$ signs (from the construction given above), the $+$ and $-$ sign representation coincides with the $+$ and $-$ signs in the upper bound given here. In fact, the $+$ and $-$ sign representation given here may only be achieved by the construction $s^*$ given above. Since we constructed cases that achieve the upper bound of $d_x(s)$, we have proved that our construction attains $\max(d_x(s))$ and we are done.
Now that we have established orderings $s^*$ that attain $\max(d_x(s))$, we use these results in the $x$ and $y$ directions to prove Theorem \[thm:utn grid\].
*Proof of Theorem \[thm:utn grid\].* **Case A - $a,b$ are even:** Using the fact that $t^+(G_{a,b}) = \max(d(s)) \le \max(d_x(s)) + \max(d_y(s))$ because $d(s) = d_x(s) + d_y(s)$, we show that $t^+(G_{a,b}) = \max(d_x(s)) + \max(d_y(s)) - 1$ for $a,b$ even.
For sake of contradiction, we assume that we can achieve $\max(d(s)) = \max(d_x(s)) + \max(d_y(s))$. Then, applying the results from Lemma \[lem:max x distance\], we split the grid into 4 quadrants and consider the consequences of combining conditions to simultaneously achieve $\max(d_x(s))$ and $\max(d_y(s))$, where the *$x$-median* is the set of vertices with $y_j = \frac{b+1}{2}$, the *$y$-median* is the set of vertices with $x_j = \frac{a+1}{2}$, the $1^\text{st}$ *quadrant* is the set of vertices with $x_j > \frac{a+1}{2}$ and $y_j > \frac{b+1}{2}$, the $2^\text{nd}$ *quadrant* is the set of vertices with $x_j < \frac{a+1}{2}$ and $y_j > \frac{b+1}{2}$, the $3^\text{rd}$ *quadrant* is the set of vertices with $x_j < \frac{a+1}{2}$ and $y_j < \frac{b+1}{2}$, and the $4^\text{th}$ *quadrant* is the set of vertices with $x_j > \frac{a+1}{2}$ and $y_j < \frac{b+1}{2}$. See Figure \[fig:odd x odd quadrants\] below. WRONG.
![Example $G_{8,6}$ split into four quadrants[]{data-label="fig:even x even quadrants"}](8by6quadrants.png)
1. The starting and ending vertex $u_1$ and $u_n$ lie on opposing central vertices of the graph, ie vertices $(\frac{a}{2}+1,\frac{b}{2}+1)$ and $(\frac{a}{2},\frac{b}{2})$ or $(\frac{a}{2},\frac{b}{2}+1)$ and $(\frac{a}{2}+1,\frac{b}{2})$.
2. Vertices $u_{i}$ and $u_{i+1}$ only alternate from quadrants 1 to 3 or from quadrants 2 to 4.
Clearly, condition (2) above cannot be satisfied, as $u_1$ starts in quadrant 1, and there is no way to reach quadrants 2 or 4 without breaking condition (2). Thus, we show how to find an ordering $s$ such that $d(s) = \max(d_x(s)) + \max(d_y(s)) - 1$. Starting with $u_1$ on vertex $(\frac{a}{2}+1,\frac{b}{2}+1)$, we choose vertices alternating from quadrants 1 and 3, until all vertices in quadrant 3 have been chosen. Now we must break condition (2), and choose the next vertex in either quadrant 2 or 4. Without loss of generality, we choose a vertex in quadrant 4, picking any vertex with a $y$-value of $\frac{b}{2}$. We then continue choosing vertices by alternating between quadrants 2 and 4, and make sure that the final vertex, $u_n$ lies on vertex $(\frac{a}{2},\frac{b}{2}+1)$. Notice that condition (1) has also been violated.
Considering the ordering $s$ we have just created, notice that $s$ achieves $\max(d_x(s))$. Thus, we show that $s$ attains $\max(d_y(s)) - 1$, using the same $+$ sign $-$ sign analysis. In this case, we have a $+$ sign in Region A with $y$-value $\frac{b}{2}$, and two vertices with 1 $+$ sign in Region B with $y$-value $\frac{b}{2}+1$. Comparing this configuration to the achieving maximum $y$-distance, we have moved a $+$ sign from the $\frac{b}{2}+1$ level to the $\frac{b}{2}$ level, resulting in a net decrease of 1. Thus, $t^+(G_{a,b}) = \max(d_x(s)) + \max(d_y(s)) - 1$.
**Case B - $a$ is even, $b$ is odd:** For sake of contradiction, assume that we can achieve $\max(d(s)) = \max(d_x(s)) + \max(d_y(s))$. In this case, we are able to show that $t^+(G_{a,b}) = \max(d_x(s)) + \max(d_y(s))$ when $a$ is even, $b$ is odd, and $a \ge 4$. That means there exists an ordering $s$ that satisfies the combined conditions below:
1. The starting and ending vertex $u_1$ and $u_n$ lie in the $x$-median, with $x_1 = \frac{a}{2}$ and $x_n = \frac{a}{2} + 1$
2. There does not exist a pair $u_i, u_{i+1}$ such that $x_i,x_{i+1} \ge \frac{a}{2} + 1$, or $x_i,x_{i+1} \le \frac{a}{2}$, or $y_i,y_{i+1} > \frac{b+1}{2}$, or $y_i,y_{i+1} < \frac{b+1}{2}$.
We construct this ordering $s$ by referring to quadrants shown in Figure \[fig:even x odd quadrants\].
![Example $G_{8,7}$ split into four quadrants and a median in the odd dimension[]{data-label="fig:even x odd quadrants"}](8by7quadrants.png)
So, we start with $u_1$ on $(\frac{b+1}{2}, \frac{a}{2})$, and then choose $u_2$ in quadrant 1. Then, we proceed to choose vertices, alternating between quadrants 1 and 3, until all of the vertices in quadrant 3 have been used. Then, we choose vertices on the $x$-median, choosing vertices on opposite sides of the dotted line, and leaving vertex $(\frac{b+1}{2}, \frac{a}{2}+1)$ unused, as it is reserved for $u_n$. Notice that this requires that there are at least 4 vertices in the median section, explaining the condition that $a \ge 4$ (and why ladder graphs are a special case). When all median vertices on the left side of the dotted line are chosen, we proceed to choose a vertex in quadrant 4. Then, we choose vertices alternating between quadrants 2 and 4 until all vertices in quadrant 2 have been used. We then choose the final vertex, $u_n = (\frac{b+1}{2}, \frac{a}{2}+1)$. Since this ordering $s$ attains $\max(d_x(s))$ and $\max(d_y(s))$, we are done.
**Case C - $a,b$ are odd:** Using the fact that $t^+(G_{a,b}) = \max(d(s)) \le \max(d_x(s)) + \max(d_y(s))$ because $d(s) = d_x(s) + d_y(s)$, we show that $t^+(G_{a,b}) = \max(d_x(s)) + \max(d_y(s)) - 1$ for $a,b$ odd.
First, we show that $\max(d(s)) \neq \max(d_x(s)) + \max(d_y(s))$. Applying the results from Lemma \[lem:max x distance\], we split the grid into 4 quadrants and two medians, where the *$x$-median* is the set of vertices with $y_j = \frac{b+1}{2}$, the *$y$-median* is the set of vertices with $x_j = \frac{a+1}{2}$, the $1^\text{st}$ *quadrant* is the set of vertices with $x_j > \frac{a+1}{2}$ and $y_j > \frac{b+1}{2}$, the $2^\text{nd}$ *quadrant* is the set of vertices with $x_j < \frac{a+1}{2}$ and $y_j > \frac{b+1}{2}$, the $3^\text{rd}$ *quadrant* is the set of vertices with $x_j < \frac{a+1}{2}$ and $y_j < \frac{b+1}{2}$, and the $4^\text{th}$ *quadrant* is the set of vertices with $x_j > \frac{a+1}{2}$ and $y_j < \frac{b+1}{2}$. See Figure \[fig:odd x odd quadrants\] below.
![Example $G_{7,7}$ split into four quadrants and two medians[]{data-label="fig:odd x odd quadrants"}](7by7quadrants.png)
Now, combining requirements to attain $\max(d_x(s))$ and $\max(d_y(s))$, each subsequent $u_{i+1}$ is under the following combined restrictions:
1. The starting and ending vertex $u_1$ and $u_n$ reside in the intersection of the two medians.
2. There does not exist a pair $u_i, u_{i+1}$ such that $x_i,x_{i+1} > \frac{a+1}{2}$, or $x_i,x_{i+1} < \frac{a+1}{2}$, or $y_i,y_{i+1} > \frac{b+1}{2}$, or $y_i,y_{i+1} < \frac{b+1}{2}$.
Clearly, condition (1) cannot be satisfied. Thus, we have shown that $\max(d(s)) \neq \max(d_x(s)) + \max(d_y(s))$ as desired. We now claim that we can achieve $\max(d(s)) = \max(d_x(s)) + \max(d_y(s)) - 1$ by letting $u_1 = (\frac{a+1}{2}, \frac{b+1}{2})$ and $u_n = (\frac{a+1}{2}, \frac{b+1}{2} - 1)$.
Since it is not difficult to construct an ordering $s$ that has start and end vertices on $(\frac{a+1}{2}, \frac{b+1}{2})$ and $(\frac{a+1}{2}, \frac{b+1}{2} - 1)$ while following condition (2) above, we focus on showing that such an ordering has distance $d(s) = \max(d_x(s)) + \max(d_y(s)) - 1$. Since $d_x(s) = \max(d_x(s))$ in the ordering above, we show that $d_y(s) = \max(d_y(s)) - 1$.
Again, consider the representation of $d_y(s)$ using $+$ and $-$ signs. The only condition we did not follow to maximize $d_y(s)$ is that the vertex $(\frac{a+1}{2}, \frac{b+1}{2} - 1)$ in Region B has only 1 $-$ sign. Compared to an ordering which maximizes $y$-distance, the change is that we have moved the minus sign which should have been in Region B to the median. The net change is $-1$, so $d_y(s) = \max(d_y(s)) - 1$.
Bumps on Grid Graph {#sec:bump}
===================
Bumps are the distinguishing factor between the upper traceable number problem and the radio number problem. Recall from Remark \[rem:additional value\] that a bump $b_i$ is the additional value on $f(u_{i+1})$ than is required by its tightness neighbor constraint from $u_i$. A vertex $u_k$’s set of tightness neighbor constraints allows us to classify which vertex makes $b_k > 0$. Then, for the greatest $c>1$ such that $u_{k-c} \in {\normalfont \text{tn}}(u_k)$, we say $u_k$ has a $(k-c)$–bump. **The $(k-2)$–bump:** Consider vertices $u_{k},u_{k-1},u_{k-2}$ for $3 \le k \le n$.
Then, in a $(k-2)$–bump on vertex $u_k$, the tightness constraint on $u_k, u_{k-2}$ is greater than the sum of tightness constraints from $u_{k-2}, u_{k-1}$ and $u_{k-1}, u_{k}$, giving us $$D+1 - d(u_{k},u_{k-2}) > (D+1 - d_{k-1} + b_{k-1}) + (D+1 - d_{k-2} + b_{k-2}).$$
Simplifying, we get $d_{k-1} + d_{k-2} - d(u_{k},u_{k-2}) > D+1 + b_{k-1} + b_{k-2}$. The left hand side expression $d_{k-1} + d_{k-2} - d(u_{k},u_{k-2})$ has an special interpretation on the grid graph. Consider the bounding rectangle formed using vertices $u_{k-2}$ and $u_{k}$ as two corners. Then, define $d_\text{rect}$ to be the minimum distance from any vertex within the bounding rectangle to vertex $u_{k-1}$. Then, we have $d_{k-1} + d_{k-2} - d(u_{k},u_{k-2}) = 2d_\text{rect}$. Substituting this into the inequality, we see that the following condition necessarily creates a $(k-2)$–bump: $$d_\text{rect} > \frac{D+1 + b_{k-1} + b_{k-2}}{2}.$$
**The $(k-3)$–bump:** There is no configuration of vertices $u_{k},u_{k-1},u_{k-2},u_{k-3}$ such that $b_i > 0$. For sake of contradiction, let us assume that such a configuration exists. Then, we have $$f(u_{k}) - f(u_{k-3}) = D+1 - d(u_{k},u_{k-3}) > f_{k-1} + f_{k-2} + f_{k-3}.$$
The expression simplifies to $d_{k-1} + d_{k-2} + d_{k-2} - d(u_{k},u_{k-3}) > 2(D+1) + b_{k-1} + b_{k-2} + b_{k-3}$. By using a similar method to evaluate $d_{k-1} + d_{k-2} - d(u_{k},u_{k-2})$ on the $(k-2)$–bump, we see that $d_{k-1} + d_{k-2} + d_{k-2} - d(u_{k},u_{k-3}) = 2d(u_{k-1},u_{k-2})$. Clearly, then $2d(u_{k-1},u_{k-2}) > 2(D+1) + b_{k-1} + b_{k-2} + b_{k-3} > 2(D+1) \implies d(u_{k-1},u_{k-2}) > D+1$. Since the diameter is the maximum distance on $G$, this is a contradiction. Thus, our original assumption that there exists configuration of $u_{k},u_{k-1},u_{k-2},u_{k-3}$ such that $b_i > 0$ was false.
Radio Number of Grid Graph {#sec:grid}
==========================
Radio Number of Ladder Graph {#sec:ladder}
============================
Conclusion {#sec:conclusion}
==========
[10]{}
Vladoi, Alexandru, *Wireless networks: a labeling problem* (2010). Selected Senior Projects Spring 2010. Paper 24.
D. Liu and X. Zhu. *Multi-level distance labelings for paths and cycles.* SIAM J. Discrete Math, 19:281–293, 2005.
D. Liu. *Radio number for trees.* Discrete Mathematics, 308(7):1153 – 1164, 2008.
M. Kchikech, R. Khennoufa, O. Togni, *Radio k-labeling for cartesian products of graphs*, Electronic Notes in Discrete Mathematics, 22 (2005), 347-352.
Saha, Laxman, Panigrahi, Pratima, and Kumar, Pawan. “Improved bounds for radio -chromatic number of hypercube .." *International Journal of Mathematics and Mathematical Sciences 2011* (2011): Article ID 961649, 7 p.-Article ID 961649, 7 p.. <http://eudml.org/doc/232091>.
Xiangwen Li, Vicky Mak, and Sanming Zhou. 2010. Optimal radio labellings of complete m-ary trees. *Discrete Appl. Math.* 158, 5 (March 2010), 507-515. DOI=10.1016/j.dam.2009.11.014 http://dx.doi.org/10.1016/j.dam.2009.11.014
W. K. Hale, *Frequency assignment: theory and applications*, Proc. IEEE, 68 (1980), 1497 – 1514.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We experimentally investigated the optical responses of a superconducting niobium resonator. It was found that, with increasing radiation power, the resonance frequency increases monotonically below around 500 mK, decreases monotonically above around 1 K and exhibits a nonmonotonic behavior at around $700$ mK. These observations show that one can operate the irradiated resonator in three temperature regimes, depending on whether two-level system (TLS) effects or kinetic inductance effects dominate. Furthermore, we found that the optical responses at ultra-low temperatures can be qualitatively regarded as a photon-induced thermalization effect of TLSs, which could be utilized to achieve thermal sensitive photon detections.'
author:
- Yiwen Wang
- Pinjia Zhou
- Lianfu Wei
- Haijie Li
- Beihong Zhang
- Miao Zhang
- Qiang Wei
- Yurong Fang
- Chunhai Cao
bibliography:
- 'aipsamp.bib'
nocite: '[@*]'
title: 'Photon-induced thermal effects in superconducting coplanar waveguide resonators'
---
In recent years superconducting coplanar waveguide (CPW) resonators have attracted extensive attentions, due to its important applications in solid-state quantum information processings and sensitive photon detections [@pkday1; @gao1; @Baselmans1; @Mazin1]. Particularly, high-quality factor resonators [@Barends1; @Vissers1; @Megrant1] (typically $Q >
10^{5}$) operating well below transition temperature $T_{c}$ have been demonstrated to be suitable for serving as so-called microwave kinetic inductance detectors (MKIDs) [@Zmuidzinas]. According to Mattis-Bardeen theory [@Zmuidzinas2], the absorption of photons (or increasing bath temperature) will decrease the electron-pair density in superconducting film, causing an increase in the kinetic inductance [@Mazin2] and thus a decrease in the resonance frequency. Meanwhile, the dissipation increases due to quasiparticle excitations, leading to a reduction in the quality factor. Therefore by probing the changes in the frequency or dissipation of a MKID, the radiation light signals can be detected. Much work to date have been done to optimize resonator geometries and materials as well as a better understanding of the noise properties in order to improve the detection sensitivity of MKID [@gao2; @Barends2; @Leduc].
In this paper we try to demonstrate another possible photon detection approach with a superconducting resonator. This approach is based on the thermalization of two-level systems (TLSs) in dielectrics rather than the kinetic inductance effects in the metal film. With a high-quality factor niobium resonator, we studied its optical responses by performing transmission measurements at low temperatures in a broad range of $20$ mK to $1.9$ K, well below the niobium transition temperature $T_{c} = 9.2$ K. We found that the irradiated niobium resonator can be operated in three temperature regimes: with increasing radiation power, the resonance frequency increases monotonically below $\sim 500$ mK, decreases monotonically above $\sim 1$ K and exhibits an interesting nonmonotonicity at temperatures around $700$ mK. Similar crossover behavior with bath temperature for a unirradiated resonator have been reported [@gao3; @Barends3; @gao4; @Lindstrom; @gao5]. Our observations can be qualitatively explained by the temperature dependent permittivity (due to TLSs in dielectrics) and temperature dependent kinetic inductance. At sufficiently low temperatures, TLS effects dominate over kinetic inductance effects, suggesting that the photon-induced thermalization effects of TLSs in dielectrics could be utilized to achieve thermal sensitive photon detections. Note that this is a different photon detection approach from MKID, which typically requires a superconducting material and geometry with high kinetic inductance fraction [@Porch].
Our CPW resonator [@Wang] was fabricated by magnetron sputtering and photolithography. A $160$ nm thick niobium film was deposited on a $500$ $\mu$m thick high-resistivity silicon substrate. Contact exposure as well as reflect ion etching were used to generate desired patterns on the film. The microscope photograph of the key structure of the measured sample is shown in Fig. 1(a). The resonator is actually a meandered $\lambda /4$ transmission line, which shorts to the ground plane at one end and capacitively couples to a feedline at another end. The feedline is $20$ $\mu$m wide with a gap of $15$ $\mu$m to the ground plane. The center line of the resonator is $7$ $\mu$m wide with a gap of $7$ $\mu$m. The total length of the resonator is about $15.814$ mm and the relative permittivity of Si substrate is $11.9$, giving a theoretical fundamental resonance frequency $f_{r} = 1.8665$ GHz.
[ff1.eps]{} (18,800)[**(a)**]{} (500,800)[**(b)**]{} (485,390)[**(c)**]{} (68,730)[$\footnotesize 400\mu m$]{} (210,70)[Feedline]{} (123,389)[CPW Resonator]{} (906,800)[Nb]{} (490,5)[{width="4.35cm"}]{}
The resonator chip is glued and wire-bonded to a gold-plated oxygen-free copper box equipped with microwave connections. This sample box is mounted at the cold plate of the mixing chamber in a dilution refrigerator with base temperature less than $20$ mK. The temperature was measured using a calibrated R1101 resistance thermometer which is placed close to the sample box. To radiate the CPW resonator, a single-mode optical fiber is set up from the room temperature down to the sample box. A room temperature laser source is connected to the top end of the fiber and generates a steady stream of photons at wavelength $635$ nm. The bottom end of the fiber is carefully aligned and fixed so that the incoming photons can hit the short end area of the resonator, where the standing wave current distribution is maximum. The fiber end is about $100$ $\mu$m vertically away from the chip surface and the irradiation area (indicated by the red circle in Fig. 1(b)) is about $20$ $\mu$m in diameter. Therefore, part of the incoming photons are incident on the superconducting film and part on the bare substrate in the gap. Additionally, not all the photons are absorbed by the chip, as some photons could be reflected and scattered into the environment.
The device transmission amplitudes $S_{21}$ as a function of frequency were measured with an Agilent E5071C vector network analyzer. Attenuators and DC blocks are positioned appropriately to suppress circuit noises. The microwave driving power reaching the chip is estimated to be about $-65$ dBm, which is fixed for all the measurements in this letter. At temperature $20$ mK, a resonance dip was found at $f_{r1} = 1.836149$ GHz, shown in Fig. 1(c). Note that the resonance curves have been calibrated to measure $0$ dB off resonance to remove the effects of attenuation in the circuits. By fitting the $\lvert S_{21}\lvert ^{2}$ data to a skewed Lorentzian model [@Gao-phd], one can obtain the loaded quality factor of the resonator $Q_{1}$ = $3.7$ $\times 10^{5}$. We then radiated the resonator with a power $137$ nW and measured the transmissions when the system reaches thermal and electric equilibrium. As shown in Fig. 1(c), the resonance frequency shifts to a higher value $f_{r2}
= 1.836163$ GHz and the quality factor decreases to $Q_{2}$ = $2.8$ $\times 10^{5}$. The decrease in quality factor is expected but the increase in resonance frequency is in contrary to the prediction of Mattis-Bardeen theory.
We now investigate the optical responses of the resonator with different radiation powers. To this aim we varied the light intensity and measured corresponding microwave transmissions through the resonator at base temperature $T$ = $20$ mK. Fig. 2 (blue circles) shows the relative resonance frequency shift as a function of the radiation power on the short end area. Here, the relative frequency shift is defined as $\Delta f_{r}/f_{r} = [f_{r}(P) -
f_{r}(P=0)]/f_{r}(P=0)$, with $P$ being the total radiation power entering the sample box and $f_{r}(P)$ the corresponding resonance frequency. It is shown that the frequency shift increases monotonically with increasing radiation power in the range of $0.4$ nW to $2.2$ $\mu$W. The inset shows the same plot except the power-axis is in logarithmic scale. One can see that the frequency shift increases approximately linearly with logarithmic power above a crossover point (at $\approx $ $1$ nW), indicated by the intersection of two red dotted lines.
{width="7.5cm"}
Qualitatively, the radiation power dependence observed here is similar to the temperature dependence reported in a few recent experiments [@gao3; @Barends3; @gao4; @Lindstrom; @gao5], where the measured resonance frequency shift shows approximately a logarithmic increase with temperature well below $T_{c}$. Such behaviors were also verified in our measurements of the relative frequency shift versus bath temperature, as shown in Fig. 3.
[f6.eps]{}
Here, $\Delta f_{r}/f_{r} = (f_{r}(T) -f_{r}(T_{0}))/f_{r}(T_{0})$ with the reference temperature $T_{0} =$ $20$ mK. Note that the experiments can not be explained by Mattis-Bardeen theory, which predicts a frequency shift opposite in sign. Instead, the observed temperature dependence can be attributed to the TLS effects in dielectrics [@Martinis2; @Martinis3]. In fact, TLSs are abundant in amorphous materials and can be present in substrate, the surface oxides on the metal film and the interface between metal and substrate [@Martinis]. At low temperatures, the unsaturated TLSs with electric dipole moments can interact resonantly with the microwave field, giving a temperature dependent variation of permittivity [@Phillips]. Changing in permittivity affects the capacitance per unit length and thus the resonance frequency. Assuming a uniform distribution of TLSs, the relative resonance frequency shift is given by [@gao3] $$\frac{\Delta
f_{r}}{f_{r}}=C\left\{\ln\left(\frac{T}{T_{0}}\right)-\left[g(T,\omega)-g(T_{0},\omega)\right]\right\},$$ where $C$ is the unique fitting parameter, $T_{0} =$ $20$ mK the reference temperature and $g(T,\omega)
=\text{Re}\left[\Psi(\frac{1}{2}+\hbar\omega/2\pi ik_BT)\right]$ with $\Psi$ being the complex digamma function. Fits to the above equation ($C =$ $1.15$ $\times 10^{-5}$) can explain the observed temperature dependence below $1.9$ K, where the resonance frequency increases monotonically with temperature. This suggests that at low temperatures $T << T_{c}$, which is the case here, the thermal effects on the kinetic inductance are relatively weak [@Meservey]. Additionally, considering the geometry of our resonator, its kinetic inductance fraction is estimated to be very small (less than $0.02$). Therefore the kinetic inductance effects are negligible for $T << T_{c}$ and TLS effects become prominent, which is the reason why Mattis-Bardeen theory does not describe the dominating loss mechanism at ultra-low temperatures. However, the resonance frequency starts to drop at around $\sim 2$ K, indicating that TLS effects saturate and kinetic inductance becomes dominant as temperature approaching $T_{c}$. Similar experimental results have been observed in a NbTiN resonator covered with SiO$_x$ layer [@Barends3].
We believe that the above TLS effects can also explain the observed radiation power dependence at $T =$ $20$ mK. Experimentally, photons irradiated on both metal film and substrate will mainly heat the chip and increase its effective temperature, although the bath temperature does not change. At ultra-low temperatures(e.g., $T =$ $20$ mK), the incident photons can excite TLSs in the substrate, especially those in the exposed substrate surface (i.e., SiO$_x$ in our device) and nearby the irradiated area. To further verify it is indeed the thermalization of TLSs in the substrate dominates the radiation power dependence of resonance frequency at ultra-low temperatures, we also directly radiated on a small area of the bare substrate, which is about $5$ mm away from the resonator. We performed the relevant transmission measurements at $20$ mK and obtained a similar radiation power dependence, as shown in Fig. 2 (black squares). The similar power dependence implies that the thermalization of TLSs is the main factor in shifting the resonace frequency. Moreover, the photon responsivity is weaker when radiating on the bare substrate of a certain distance away from the resonator than that when radiating on the short end area. This is reasonable and can be explained by a nonuniform temperature distribution around the irradiated area. The effective temperature at the RF active component of the resonator is lower when the $20$ $\mu$m thermal source is moved $5$ mm away.
[f5.eps]{} (30,980)[**(a)**]{} (30,480)[**(b)**]{}
We then investigated the optical responses of the resonator at different bath temperatures. Fig. 4(a) shows the relative resonance frequency shift as a function of radiation power, taken at seven selective bath temperatures in a broad range of $20$ mK - $1900$ mK. Here, the relative frequency shift is defined as $\Delta f_{r}/f_{r}
= (f_{r}(P,T) - f_{r}(P=0,T))/f_{r}(P=0,T)$, with $T$ being the bath temperature and $P$ the radiation power. Note that the reference frequency $f_{r}(P=0,T)$ varies with temperature $T$. At lower temperatures (i.e., $20$ mK, $160$ mK, $273$ mK and $425$ mK), the frequency shift increases monotonically with increasing radiation power, indicating the TLS effects dominate in this temperature range. Apparently, the resonator responsivity to photons is strongest at the lowest temperature $T =$ $20$ mK. For the temperature at $700$ mK, we observed an interesting phenomenon that the resonance frequency exhibits a nonmonotonic behavior with increasing radiation power: the frequency firstly goes down and then goes up, which can be seen more clearly in Fig. 4(b). This suggests that the kinetic inductance effects start to play in role at the temperature around $700$ mK and the TLS effects are less pronounced than those at ultra-low temperatures. With lower radiation power, the incident photons break Cooper pairs near the irradiated area, increasing the kinetic inductance and decreasing the resonance frequency. However, with enhancing radiation power, the chip is heated to excite the unsaturated TLSs and the TLS effects may become dominated over the kinetic inductance effects again, causing the resonance frequency increase. Furthermore, for higher temperatures, e.g. at $1000$ mK and $1900$ mK, the frequency decreases monotonically with radiation power. As temperatures increase, the TLS effects tend to fully saturate. Therefore only kinetic inductance changes with incident photons in high temperature regime. In summary, we experimentally studied the optical responses of a superconducting niobium coplanar waveguide resonator at various temperatures. Our results show that one can operate an irradiated Nb resonator in three different temperature regimes below $T_{c}$. At ultra-low temperatures (below $\sim 500$ mK), TLS effects dominate while at higher temperatures (above $\sim$ $1$ K), TLS effects saturate and kinetic inductance effects dominate. In the middle regime (e.g., around $700$ mK), both TLSs and kinetic inductance take effects. Therefore, instead of maximizing the kinetic inductance effects to improve the performance of MKID for sensitive photon detections, the TLS effects could also be utilized to achieve weak light detections at ultra-low temperatures. Furthermore, lower bath temperatures lead to stronger photon responses.
Although the sensitivity of the TLS-based photon detections demonstrated here is much lower than that of a MKID, the relevant resonator is relatively easy to be fabricated since high kinetic inductance fraction is not required. The performance of proposed TLS-based weak light detections could be improved by minimizing the heat capacity of the chip. The chip size can be designed and fabricated as small as possible so that a certain radiation power can lead to a bigger effective temperature change of the resonator. Also, an antireflection-coating can be used to reduce photon reflections and achieve a high optical coupling efficiency. In addition, increasing the density of TLSs in dielectrics will enhance the TLS-resonator coupling and thus in principle may improve the device sensitivity to incident photons.
This work was supported in part by the National Natural Science Foundation (Grant No. 61301031, No. 11174373, No. 11204249), and the National Fundamental Research Program of China (Grant No. 2010CB923104). We thank Profs. Y. Yu and P. H. Wu for kind supports and discussions.
[1]{} P. K. Day, H. G. LeDuc, B. A. Mazin, A. Vayonakis, and J. Zmuidzinas, Nature(London) **425**, 817 (2003). J. Gao, M. R. Visser, M. O. Sandberg, F. C. S. da Silva, S. W. Nam, D. P. Pappas, D. S. Wisbey, E. C. Langman, S. R. Meeker, B. A. Mazin, H. G. Leduc, J. Zmuidzinas, and K. D. Irwin, Appl. Phys. Lett. **101**, 142602 (2012). J. Baselmans, S. Yates, P. Diener, and P. Visser, J. Low Temp. Phys. **167**, 360 (2012). B. A. Mazin, B. Bumble, S. R. Meeker, K. Brien, S. McHugh, and E. Langman, Optics Express **20(2)**, 1503 (2012). R. Barends, N. Vercruyssen, A. Endo, P. J. de Visser, T. Zijlstra, T. M. Klapwijk, P. Diener, S. J. C. Yates, and J. J. A. Baselmans, Appl. Phys. Lett. **97**, 023508 (2010). M. R. Vissers, J. Gao, D. S. Wisbey, D. A. Hite, C. C. Tsuei, A. D. Corcoles, M. Steffen, and D. P. Pappas, Appl. Phys. Lett. **97**, 232509 (2010). A. Megrant, C. Neill, R. Barends, B. Chiaro, Yu Chen, L. Feigl, J. Kelly, Erik Lucero, Matteo Mariantoni, P. J. J. Malley, D. Sank, A. Vainsencher, J. Wenner, T. C. White, Y. Yin, J. Zhao, C. J. Palmstrom, John M. Martinis, and A. N. Cleland, Appl. Phys. Lett. **100**, 113510 (2012). J. Zmuidzinas, Annu. Rev. Cond. Mater. Phys. **3**, 169 (2012). J. Gao, J. Zmuidzinas, A. Vayonakis, P. Day, B. Mazin, and H. Leduc, J. Low Temp. Phys. **151**, 157 (2008). B. A. Mazin, “Microwave kinetic inductance detectors,” Ph.D. thesis (California Institute of Technology, 2004). J. Gao, J. Zmuidzinas, B. A. Mazin, Henry G. LeDuc, and Peter K. Day, Appl. Phys. Lett. **90**, 102507 (2007). R. Barends, N. Vercruyssen, A. Endo, P. J. de Visser, T. Zijlstra, T. M. Klapwijk, and J. J. A. Baselmans, Appl. Phys. Lett. **97**, 033507 (2010). H. Leduc, B. Bumble,P. Day, B. Eom, J. Gao, S, Golwala, B. Mazin, S. McHugh, A. Merrill, D. Moore, O. Noroozian, A. Turner, and J. Zmuidzinas, Appl. Phys. Lett. **97**, 102509 (2010). J. Gao, M. Daal, A. Vayonakis,S. Kumar, J. Zmuidzinas, B. Sadoulet, B. A. Mazin, P. K. Day, and H. G. Leduc, Appl. Phys. Lett. **92**, 152505 (2008). R. Barends, H. L. Hortensius, T. Zijlstra, J. J. A. Baselmans, S. J. C. Yates, J. R. Gao, and T. M. Klapwijk, Appl. Phys. Lett. **92**, 223502 (2008). S. Kumar, J. Gao, J. Zmuidzinas, B. A. Mazin, H. G. LeDuc, and P. K. Day, Appl. Phys. Lett. **92**, 123503 (2008). T. Lindstrom, J. E. Healey, M. S. Colclough, C. M. Muirhead, and A. Ya. Tzalenchuk, Phys. Rev. B **80**, 132501 (2009). D. S. Wisbey, J. Gao, M. R. Vissers, F. C. S. da Silva, J. S. Kline, L. Vale, and D. P. Pappas, J. Appl. Phys. **108(9)**, 093918 (2010). A. Porch, P. Mauskopf, S. Doyle, and C. Duscombe, IEEE Trans. Appl. Superconduct. **80**, 552 (2005). H. Li, Y. Wang, L. Wei, P. Zhou, Q. Wei, C. Cao, Y. Fang, Y. Yang, and P. Wu, Chin. Sci. Bull. **58(20)**, 2413 (2013). J. Gao, Ph. D. thesis, Caltech, 2008. John M. Martinis, K. B. Cooper, R. McDermott, Matthias Steffen, Markus Ansmann, K. D. Osborn, K. Cicak, Seongshik Oh, D. P. Pappas, R. W. Simmonds, and Clare C. Yu, Phys. Rev. Lett. **95**, 210503 (2005). R. W. Simmonds, K. M. Lang, D. A. Hite, S. Nam, D. P. Pappas, and John M. Martinis, Phys. Rev. Lett. **93**, 077003 (2004). J. Wenner, R. Barends, R. Bialczak, Y. Chen, J. Kelly, E. Lucero, M. Mariantoni, A. Megrant, P. Malley, D. Sank, A. Vainsencher, H. Wang, T. White, Y. Yin, J. Zhao, A. Cleland, and J. M. Martinis, Appl. Phys. Lett. **99**, 113513 (2011). W. A. Phillips, Rep. Prog. Phys. **50**, 1657 (1987) R. Meservey and P. M. Tedrow, J. Appl. Phys. **40**, 2028 (1969)
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The properties of scalar mesons and glueballs at finite temperature are analyzed through a [*bottom-up*]{} holographic approach. We focus on the spectral functions and mass spectra. A discussion on hadron dissociation and deconfinement phase transition is also put forward.'
author:
- Floriana Giannuzzi
title: Finite temperature hadrons from holographic QCD
---
Since its appearence in 1998, the AdS/CFT correspondence [@Maldacena:1997re] has been considered as a very promising tool for studying the non-perturbative regime of QCD, by relating it to a weakly-coupled theory. In particular, the correspondence can be better applied to the finite temperature case, one of the main reason being that in this limit the theory is no more conformal.
The conjecture states that type IIB string theory in a $AdS_5 \times S^5$ space, where $AdS_5$ is a five-dimensional anti-de Sitter space and $S^5$ is a five-dimensional sphere, is dual to $\mathcal{N}=4$ Super-Yang-Mills theory in a four-dimensional Minkowski space. It is also known as the holographic conjecture since the gauge theory can be constructed through a projection of the gravity theory on the boundary of the $AdS$ space [@Witten:1998qj]. According to the correspondence, a five-dimensional field $\phi$, whose boudary value is $\phi_0$, is related to a four-dimensional operator $\mathcal{O}$ by: $$\label{relAdSCFT}
Z_{\mathcal{S}}[\phi_0(x)]=\left< \mbox{e}^{ \int_{\partial AdS_{d+1}} \phi_0(x) \, \mathcal{O}(x)} \right>_{\mbox{\tiny CFT}} \,;$$ strictly speaking, the generating functional on the left-hand side of (\[relAdSCFT\]), computed in $\phi_0$, is equal to the generating functional of the correlation function of the operator whose source is $\phi_0$, on the right-hand side. In Poincaré coordinates, the $AdS$ space is characterized by the metric $$ds^2=\frac{R^2}{z^2}(dt^2-d\vec x^2-dz^2)\,,$$ where $z$ is called holographic coordinate. The boundary of the space is at $z=0$.
Up to now the dual theory of QCD has not been found yet. From a phenomenological point of view, people are trying to construct an [*ad hoc*]{} theory in a five-dimensional $AdS$ space such that its projection on the four-dimensional boundary can reproduce as many as possible QCD properties. One of such [*bottom-up*]{} approaches developed so far is the Soft Wall model [@Karch:2006pv], in which conformal symmetry (proper of $AdS$ spaces) is broken by inserting a factor $e^{-c^2\,z^2}$ in the action, with $c$ a mass scale, here fixed from the $\rho$ meson mass: $c=m_\rho/2=388$ MeV [@Karch:2006pv].
In this holographic picture, temperature effects are introduced by modifying the metric of the anti-de Sitter space. In this respect, one can either impose a periodicity of Euclidean time, in which the temperature is the inverse of the compactification radius, or introduce a black hole in the metric along the fifth dimension, $z$, such that the temperature is related to the inverse of the position of the horizon of the black hole. From now on, the former case will be referred to as the “Thermal-AdS” model, whereas the latter case as “AdS-Black Hole” model. Therefore, “Thermal-AdS” model is characterized by the following metric: $$\label{metricTH}
ds^2=\frac{R^2}{z^2} \left( d\tau^2+d\vec{x}^2+dz^2 \right) \qquad 0<\tau<\beta'=1/T\;,$$ where $\tau$ is the Euclidean time and $T$ is the temperature. On the other hand, the “AdS-Black Hole” metric is given by: $$\label{metricBH}
ds^2=\frac{R^2}{z^2} \left(f(z) d\tau^2+d\vec{x}^2+\frac{dz^2}{f(z)} \right) \qquad f(z)=1-\frac{z^4}{z_h^4}\;,$$ where $z_h$ is the position of the horizon of the black hole, such that $$0<z<z_h=1/(\pi T) \;;$$ in this case, the metric is smooth and complete if and only if the Euclidean time is periodic [@Witten:1998zw], with period $\beta=1/T=\pi z_h$.\
To find out what is the model that can better holographically describe QCD at finite temperature, one can either analyze both models separately and compare their outcomes with predictions coming from other approaches to QCD, or introduce a criterion for determining which metric is the stable one, for instance, by comparing the corresponding free energies and choosing the model with the lowest one. Here we consider both possibilities and compute spectral functions of scalar mesons and scalar glueballs.
Let us start from analyzing each model separately. From the point of view of spectral functions, “Thermal-AdS” is completely analogous to the zero-temperature model, and it yields the same results: finite-temperature spectral functions are therefore expected to be characterized by zero-width peaks at fixed positions, as at $T=0$.\
“AdS-Black Hole”, instead, deserves more attention. In the following, it will be shown how to compute spectral functions in a particular case, i.e. for scalar glueballs; however the procedure is general and can be extended to any other observable. Scalar glueballs have been investigated in the Soft Wall model at zero temperature in [@Colangelo:2007pt]. The five-dimensional field which is dual to the QCD operator $\beta(\alpha_s) \mbox{ Tr}(G^2(x))$, where $\beta(\alpha_s)$ is the Callan-Symanzik function, is a scalar massless field $X(x,z)$, standing the relation between the mass and the conformal dimension ($\Delta$) of a ($p$-form) operator [@Witten:1998qj]: $$m_5^2 R^2= (\Delta-p) (\Delta+p-4) \,.$$ The five-dimensional action for this field in the Soft Wall model is: $$\label{actionglu}
\mathcal{S}=\frac{1}{2 k} \int d^5x\, \sqrt{g} \mbox{ e}^{-c^2\, z^2}\, g^{MN}\, \partial_M X \partial_N X$$ where $g$ is the determinant of the metric and $k$ is a parameter which makes the action dimensionless. In order to compute spectral functions we move to the Fourier space by defining $$X(x,z)=\int d^4q\,\, \mbox{e}^{iq\cdot x} \tilde{X}(q,z) \,,$$ and we write the Fourier transformed field $\tilde{X}(q,z)=K(q,z) \tilde{X}_0(q)$ as the product of the bulk-to-boundary propagator $K(q,z)$ and the source $ \tilde{X}_0(q)$. From the action (\[actionglu\]) one can derive the equation of motion for $K(q,z)$ $$K''(q,z)- \frac{4-f(z)+2\,c^2\,z^2\,f(z)}{z\, f(z)} K'(q,z)+\left( \frac{q_0^2}{f(z)^2}-\frac{\vec{q}^2}{f(z)} \right) K(q,z)=0 \,;$$ for semplicity, the case $\vec q=0$ (rest frame of the glueball) will be considered. The first boundary condition is $K(q,0)=1$, since the source $\tilde X_0$ is defined as the value of the field $\tilde X$ at $z=0$; the second one is that the bulk-to-boundary propagator must behave as a “[*falling in*]{}” solution near the horizon of the black hole: $$K(q,u)\xrightarrow[u\rightarrow1]{}(1-u)^{-i \, q_0\, z_h/4}$$ where $u=z/z_h$. The latter condition allows us to get in the end the retarded Green’s function. Standing Eq. (\[relAdSCFT\]), the two-point correlation function can be computed by functionally deriving twice the action (\[actionglu\]) with respect to the source $\tilde X_0(q)$, thus obtaining: $$\Pi(q_0^2)=\left. \frac{\delta^2 \mathcal{S}}{\delta X_0 \delta X_0} \right|_{X_0=0}= \left. \frac{1}{2k} \frac{R^3 f(u)}{u^3 z_h^4} \mbox{e}^{-c^2\, z^2} K(q,u) \partial_u K(q,u) \right|_{u=0}\,.$$ The spectral function is the imaginary part of the Green’s function $\rho(q_0^2)=\Im(\Pi(q_0^2))$; the first two peaks of the spectral function are shown in Fig. \[specfunglu\] for four values of temperature [@Colangelo:2009ra]. The figure shows that at low temperatures the spectral function is characterized by very narrow peaks which become broader and move towards smaller values of mass as the temperature is increased; at $T\gtrsim 44$ MeV we can find no more peaks in the spectral function, so bound states do not exist anymore. We can also notice that excited states dissociate at lower temperatures than the ground state.
![Spectral function of scalar glueballs in the model with black hole (“AdS-Black Hole”) at $T=21$ MeV (solid blue line), $T=25$ MeV (dashed purple line), $T=29$ MeV (dotted yellow line), and $T=44$ MeV (dot-dashed green line).[]{data-label="specfunglu"}](specfunc_glueb.eps){width="8cm"}
By fitting each peak through a Breit-Wigner function $$\rho(x)=\frac{a\, m\, \Gamma\, x^b}{(x-m^2)^2+m^2 \Gamma^2} \,,$$ it is possible to find how the squared masses and widths of glueballs vary with temperature; they are shown in Fig. \[masswidglu\]. As expected, we find that the masses decrease while increasing temperature, starting from the value at $T=0$, $m^2_n=4 c^2(n+2)$, while the widths increase.
![Squared mass (left panel) and width (right panel) of scalar glueballs versus temperature in the model with black hole (“AdS-Black Hole”).[]{data-label="masswidglu"}](M2glueball "fig:"){width="6.cm"} ![Squared mass (left panel) and width (right panel) of scalar glueballs versus temperature in the model with black hole (“AdS-Black Hole”).[]{data-label="masswidglu"}](Widglueball "fig:"){width="6.cm"}
The qualitative behavior we have found for spectral functions, masses and widths is similar to the one observed in lattice simulations [@lattice], but the dissociation temperature for the ground-state glueball, i.e. the temperature at which the first peak in the spectral function disappears, turns out to be much lower than the one found in lattice studies.\
Similar results can be observed in the scalar-meson sector [@Colangelo:2009ra]; the two main differences are that the dissociation temperature for the ground state (around 75 MeV) is higher than the one found for glueballs, and that the ground-state squared mass, although decreasing with temperature from its value at $T=0$ [@Colangelo:2008us], at a certain temperature starts growing until the dissociation. The mass and width of scalar mesons versus temperature are plotted in Fig. \[masswidmes\]. In general, one can see that the behavior of the spectral function in Fig. \[specfunglu\] is quite universal, since it is similar for glueballs, scalar mesons and also vector mesons [@Fujita:2009wc].
![Squared mass (left panel) and width (right panel) of scalar mesons versus temperature in the model with black hole (“AdS-Black Hole”).[]{data-label="masswidmes"}](M2scalar "fig:"){width="6.cm"} ![Squared mass (left panel) and width (right panel) of scalar mesons versus temperature in the model with black hole (“AdS-Black Hole”).[]{data-label="masswidmes"}](Widscalar "fig:"){width="6.cm"}
A further possibility for studying holographic QCD at finite temperature is to make the metric a dynamical quantity, so that it can vary with temperature. To determine which metric between “AdS-Black Hole” (\[metricBH\]) and “Thermal-AdS” (\[metricTH\]) should be used for each value of the temperature, one can compute and compare the corresponding free energies and choose the one with the lowest result [@Witten:1998zw]. In the Soft Wall model [@Herzog:2006ra], the free energy in “Thermal-AdS” is $$V_{TH}(\epsilon)=\frac{4R^3}{\kappa^2}\int_0^{\beta'} d\tau \int_\epsilon^{\infty} dz\,\, \frac{1}{z^5} \mbox{ e}^{-c^2\, z^2}\,,$$ and in “AdS-Black Hole” $$V_{BH}(\epsilon)=\frac{4R^3}{\kappa^2}\int_0^{\pi z_h} d\tau \int_\epsilon^{z_h} dz\,\, \frac{1}{z^5} \mbox{ e}^{-c^2\, z^2} \,,$$ where $\epsilon\to 0$ has been introduced in order to regularize the two quantities. It turns out that at low ([*resp.*]{} high) temperatures “Thermal-AdS” ([*resp.*]{} “AdS-Black Hole”) is the right metric to use. A first order Hawking-Page phase transition [@HPage] between the two metrics occurs at $T_c\approx 191$ MeV [@Herzog:2006ra], and it has been identified with the deconfinement transition of QCD. As a matter of fact, in this framework the low-temperature spectral function has to be computed using the “Thermal-AdS” model, so it is characterized by zero-width peaks at fixed positions for every $T<T_c$. On the other hand, at temperatures higher than the critical one Fig. \[specfunglu\] shows that the spectral function computed in “AdS-Black Hole” has no peaks, so hadrons have already melted. Thus, in this model dissociation of scalar glueballs takes place together with deconfinement as a first order phase transition. This is also what happens to scalar mesons.
Some concluding remarks are then in order. We have analyzed two possible phenomenological models describing finite-temperature QCD in a holographic framework, looking at scalar-glueball spectral functions. If we use a model with a non-dynamical metric, in which the temperature is introduced through a black hole, we get a realistic qualitative description of the behavior of hadrons in a hot medium, but with a scale of temperature different from the one predicted by other models of QCD. If we let the metric change, describing the deconfinement transition on the boundary of the $AdS$ space as a Hawking-Page phase transition between two metrics in the bulk, the resulting spectral function in the confined phase is always equal to the one at zero temperature, while becoming suddenly flat in the deconfined phase. The first description can better simulate how hadron properties could change in a medium, but it seems to fail from a quantitative point of view. The second description may reproduce hadron properties in the limit of large $N_c$. Therefore, the finite-temperature holographic representation of QCD in this [*bottom up*]{} approach still needs much efforts, and slight modifications of the model may be required. Recently, some developments of the “AdS-Black Hole” model have been put forward. In [@Colangelo:2010pe] the deconfinement transition in the chemical potential-temperature space is investigated in the Soft Wall model, using, as order parameter, the behavior of the static quark-antiquark potential at large distances. Furthermore, in [@Grigoryan:2010pj] the Authors construct a slightly different model, whose parameters are fitted from the masses and decay constants of $J/\psi$ and $\psi'$, thus finding higher dissociation temperatures also in the light-meson sector.
**Acknowledgments**
This work was supported, in part, by the EU contract No. MRTN-CT-2006-035482, “FLAVIAnet” and by the grant “*Borse di ricerca in collaborazione internazionale*” by *Regione Puglia, Italy*. I thank the IPPP, Durham, for hospitality during the completion of this work.
Finally, I would like to join the Organizers of QCD@Work 2010 in warmly remembering Beppe Nardulli as a man devoted to science.
[99]{}
J. M. Maldacena, Adv. Theor. Math. Phys. [**2**]{}, 231 (1998) \[Int. J. Theor. Phys. [**38**]{}, 1113 (1999)\] \[arXiv:hep-th/9711200\].
E. Witten, Adv. Theor. Math. Phys. [**2**]{}, 253 (1998) \[arXiv:hep-th/9802150\].
A. Karch, E. Katz, D. T. Son and M. A. Stephanov, Phys. Rev. D [**74**]{}, 015005 (2006) \[arXiv:hep-ph/0602229\].
E. Witten, Adv. Theor. Math. Phys. [**2**]{}, 505 (1998) \[arXiv:hep-th/9803131\].
P. Colangelo, F. De Fazio, F. Jugeau and S. Nicotri, Phys. Lett. B [**652**]{}, 73 (2007) \[arXiv:hep-ph/0703316\]; Int. J. Mod. Phys. A [**24**]{}, 4177 (2009) \[arXiv:0711.4747 \[hep-ph\]\]; H. Forkel, Phys. Rev. D [**78**]{}, 025001 (2008) \[arXiv:0711.1179 \[hep-ph\]\]; H. Boschi-Filho and N. R. F. Braga, Eur. Phys. J. C [**32**]{}, 529 (2004) \[arXiv:hep-th/0209080\]; JHEP [**0305**]{}, 009 (2003) \[arXiv:hep-th/0212207\].
P. Colangelo, F. Giannuzzi and S. Nicotri, Phys. Rev. D [**80**]{}, 094019 (2009) \[arXiv:0909.1534 \[hep-ph\]\].
N. Ishii, H. Suganuma and H. Matsufuru, Phys. Rev. D [**66**]{}, 014507 (2002) \[arXiv:hep-lat/0109011\]; Phys. Rev. D [**66**]{}, 094506 (2002) \[arXiv:hep-lat/0206020\]; X. F. Meng, G. Li, Y. Chen, C. Liu, Y. B. Liu, J. P. Ma and J. B. Zhang, Phys. Rev. D [**80**]{}, 114502 (2009) \[arXiv:0903.1991 \[hep-lat\]\].
P. Colangelo, F. De Fazio, F. Giannuzzi, F. Jugeau and S. Nicotri, Phys. Rev. D [**78**]{}, 055009 (2008) \[arXiv:0807.1054 \[hep-ph\]\].
M. Fujita, K. Fukushima, T. Misumi and M. Murata, Phys. Rev. D [**80**]{}, 035001 (2009) arXiv:0903.2316 \[hep-ph\].
C. P. Herzog, Phys. Rev. Lett. [**98**]{}, 091601 (2007) \[arXiv:hep-th/0608151\].
S. W. Hawking and D. N. Page, Commun. Math. Phys. [**87**]{}, 577 (1983) \[arXiv:hep-th/9803131\].
P. Colangelo, F. Giannuzzi and S. Nicotri, arXiv:1008.3116 \[hep-ph\].
H. R. Grigoryan, P. M. Hohler and M. A. Stephanov, Phys. Rev. D [**82**]{}, 026005 (2010) \[arXiv:1003.1138 \[hep-ph\]\].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'An analysis is made of the relation between quantum theory and classical mechanics, in the context of the limit $\hbar \to 0$. Several ways in which this limit may be performed are considered. It is shown that Schrödinger’s equation for a single particle moving in an external potential $V$ does not, except in special cases, lead, in this limit, to Newton’s equation of motion for the particle. This shows that classical mechanics cannot be regarded as emerging from quantum mechanics—at least in this sense—upon straightforward application of the limit $\hbar \to 0$.'
author:
- |
U. Klein[^1]\
University of Linz\
Institute for Theoretical Physics\
A-4040 Linz, Austria\
title: 'What is the limit $\hbar \to 0$ of quantum theory?'
---
American Journal of Physics, in press
Introduction {#sec:Introduction}
============
Dirac’s famous book [@dirac:principles_p88] on quantum theory states that “...classical mechanics may be regarded as the limiting case of quantum mechanics when $\hbar$ tends to zero.” In quantum mechanics a single particle in an external potential is described by Schrödinger’s equation, $$\label{eq:SPSCHROE}
\bigg[ \frac{\hbar}{i} \frac{\partial}{\partial t}
- \frac{\hbar^{2}}{2 m }
\sum_{k=1}^{3} \left( \frac{\partial}{\partial x_k} \right)^{2}
+V(x,t)
\bigg]\psi(x,t)=0
\mbox{.}$$ Thus, Dirac’s statement would imply that Newton’s second law, $$\label{eq:NSEDSCHL}
\frac{d}{dt} m r_k(t) = p_k(t),
\hspace{0.5cm}
\frac{d}{dt} p_k(t) = -\frac{\partial V(x,t)}{\partial x_k}\Big|_{x=r(t)},$$ should follow from Schrödinger’s equation in the limit $\hbar \to 0$.
Nobody has ever performed a general exact calculation showing that Eq. implies Eq. in the limit $\hbar \to 0$. This conflict leads frequently to the statement that this limit cannot be understood in such a simple way. But this answer is not really satisfying. The attitude of the scientific community with regard to this point—which is extremely important for the interpretation of quantum theory (QT) as well as for more general questions such as the problem of reductionism—is somewhat inconsistent. On the one hand, Dirac’s dictum—which has been approved by other great physicists—is considered to be true. On the other hand it cannot be verified. Since the beginnings of QT, a never-ending series of works deal with this question, but the deterministic limit of QT, in the sense of the above general statement, has never been attained. Frequently, isolated “classical properties” which indicate asymptotic “nearness” of deterministic physics, or structural similarities (such as those between Poisson brackets and commutators) are considered as a substitute for the limit. The point is that in most of these papers (see e.g. [@rowe:classical; @werner:classical; @allori_zanghi:classical] to mention only a few) the question “what is the limit $\hbar \to 0$ of quantum theory?” is *not* studied. It is taken for granted that the final answer to this question has already been given (by Dirac and others) and the remaining problem is just how to confirm or illustrate it by concrete calculations. But none of these attempts is satisfying.
In the present paper the question formulated in the title will be studied without knowing the answer. A detailed step-by-step style of exposition has been chosen in order to understand this singular limit. In fact, the paper has been written with the idea in mind to provide an in-depth answer to a student’s question concerning the mathematical details of the transition from Eq. (\[eq:SPSCHROE\]) to Eq. (\[eq:NSEDSCHL\]). The questions how to perform a limit and how to characterize the relations between different (related) physical theories are closely connected to the basic question of how to characterize physical theories themselves. We take a pragmatic position with respect to this question and characterize a physical theory simply by the set of its predictions. This leads automatically to a reasonable definition of limit relations between different physical theories.
We start by discussing, in Section \[sec:two-exampl-class\], two well-understood concrete limiting relations between two pairs of classical physical theories. These classical limiting relations, referred to as *standard limit* and *deterministic limit*, define possible meanings of the phrase “the limit $\hbar \to 0$” in QT. The term “deterministic limit” is closely related to the notion of a “deterministic theory” used in this paper. The latter is (in contrast to an “indeterministic theory”) able to make deterministic predictions (with probability 1) on single events. In other words, the predictions of a deterministic theory can be verified in single experiments, in contrast to experiments on statistical ensembles, which are required to verify predictions of an indeterministic theory. In this context the term probabilistic is a synonym for indeterministic. Note that an indeterministic theory is, as a rule, based on deterministic equations, i.e., on equations whose solutions at a time $t$ are uniquely determined by initial values at an earlier time $t_0$. It is the physical meaning of the variables that makes a theory deterministic or indeterministic.
In Section \[sec:stand-limit-quant\] we use the variables introduced by Madelung to obtain the standard limit of QT, previously found by Rosen,[@rosen:classical_quantum] Schiller,[@schiller:quasiclassical] and others. In Section \[sec:formal\_standard-limit-quant\] we derive the deterministic limit of the standard limit of QT. We find that Ehrenfest’s relations, which have not been taken into account in previous treatments,[@rosen:classical_quantum; @cohn:quantum; @kobe:comments; @nikolic:classical; @gondran:discerned] provide the missing link between the standard limit (field) theory and the trajectory equations of Newtonian mechanics (NM). In Section \[sec:formal-limit-quant\] we investigate the deterministic limit of QT and conclude that this limit does not exist. In Section \[sec:combi-limit-quant\] we try to reconstruct the states of NM from QT by combining the deterministic limit and the standard limit. In this way we are indeed able to identify a class of (three) potentials and states that allow for a transition from QT to NM in the limit $\hbar \to 0$. These include among others the coherent states of the harmonic oscillator.[@schroedinger:continuous] In Section \[sec:are-all-potentials\] we try to extend this process to arbitrary potentials. The obtained results are discussed and interpreted in Section \[sec:discussion\] and the final conclusion is drawn in Section \[sec:conclusion\].
Two examples of limit relations in classical physics {#sec:two-exampl-class}
====================================================
As our first example we consider the relation between relativistic mechanics and NM. As is well known, in relativistic mechanics a new fundamental constant, the speed of light $c$, appears, which is absent (infinitely large) in NM. Otherwise, the mathematical structures of both theories are similar. The basic equations of relativistic mechanics differ from Eqs. (\[eq:NSEDSCHL\]) only by factors of $\gamma$, which depend on $v/c$ and disappear (reduce to 1) if $c$ becomes large: $$\label{eq:THLOFSG}
\gamma=\sqrt{1-\frac{v^{2}}{c^{2}}},\hspace{1cm}
\lim_{c \rightarrow \infty} \gamma= 1
\mbox{.}$$ The relation between relativistic mechanics and NM may be summarized as follows:
- Both theories have the same mathematical structure: ordinary differential equations for trajectories. A new fundamental constant $c$ appears in relativistic mechanics.
- The limit $1/c \rightarrow 0$ transforms the basic equations of relativistic mechanics into the basic equations of NM; the same is true for the solutions of these equations.
We see that relativistic mechanics and NM provide a perfect realization of a limit relation (NM is the limit theory of relativistic mechanics) or a covering relation (relativistic mechanics is the covering theory of NM). The significant feature is the appearance of a new fundamental constant which allows for a transition between two different theories of the same mathematical type. We will refer to the type of limit relation encountered in this first example as *standard* limit relation.
Our second example concerns the relation between NM and a probabilistic version of NM, which can be constructed according to the following well-known recipe. We consider a phase space probability density $\rho(x,p,t)$ and assume that the total differential of $\rho$ vanishes: $$\label{eq:WEAT23TTD}
d\rho(x,p,t)=
\frac{\partial \rho}{\partial x_k} dx_k+
\frac{\partial \rho}{\partial p_k} dp_k+
\frac{\partial \rho}{\partial t} dt =0
\mbox{.}$$ This means that $\rho$ is assumed to be constant along arbitrary infinitesimal changes of $x_k,\,p_k,\,t$. Next we postulate that the movement in phase space follows classical mechanics, i.e., we set $dx_k=(p_k/m)dt$ and $dp_k=-(\partial V/\partial x_k)dt$. This leads to the partial differential equation (Liouville equation) $$\label{eq:AI3SP7DFNM}
\frac{\partial \rho}{\partial t}+
\frac{p_k}{m} \frac{\partial \rho}{\partial x_k}
-\frac{\partial V}{\partial x_k}\frac{\partial \rho}{\partial p_k} =0
\mbox{,}$$ which has to be solved by choosing initial values $\rho(x,p,0)$ for the new dynamical variable $\rho(x,p,t)$. The relation between the probabilistic version (of NM) and NM may be summarized as follows:
- The probabilistic version and NM have different mathematical structures; the probabilistic version is ruled by a partial differential equation, NM by an ordinary differential equation. No new constant appears in the probabilistic version.
- The probabilistic version and NM belong to fundamentally different epistemological categories. NM is a deterministic theory. The probabilistic version is a probabilistic (indeterministic) theory; predictions about individual events cannot be made because the initial values for individual particles are unknown.
The absence of a new fundamental constant prevents a simple transition between the two theories as found in our first example. Nevertheless, a kind of limit relation can be established by means of appropriate (singular) *initial values*. A probability density that is sharply peaked at $t=0$ retains its shape at later times. Inserting the Ansatz $$\label{eq:IM9ESA2HON}
\rho(x,p,t) = \delta^{(3)}(x-r(t))\,\delta^{(3)}(p-p(t))$$ into Eq. (\[eq:AI3SP7DFNM\]), it is easily shown that admissible particle trajectories $r_k(t)$, $p_k(t)$ are just given by the solutions of Newton’s equations (\[eq:NSEDSCHL\]). Thus, NM can be considered as a limit theory of the probabilistic version in the sense that the manifold of solutions of a properly (with regard to singular initial values) generalized version of the probabilistic version leads to NM. This limit relation is weaker than the one encountered in our first example, because there is no mapping of individual solutions. By allowing for singular solutions we have essentially constructed the *union* of the deterministic theory NM and the original probabilistic version of NM; it is then no surprise that the generalized probabilistic version theory contains NM as a special case. Considered from a formal point of view, however, the (generalized) probabilistic version is a perfect covering theory since its manifold of solutions is larger than that of NM. We shall refer to the kind of limit relation found in this second example as a *deterministic* limit relation.
The standard limit of quantum theory {#sec:stand-limit-quant}
====================================
Let us now compare QT and NM \[Eqs. and \] in the light of the above examples. The two theories obviously differ in their mathematical structures; this indicates the possibility to obtain NM from QT by means of a deterministic limiting process. However, in addition, a new fundamental constant ($\hbar$) appears in QT; this indicates the possibility to obtain NM from QT by means of a standard limiting process. Thus, the limiting process that leads from QT to NM is either nonexistent or more complex than either of the above examples.
Let us try, as a first step, to perform the standard limit of QT, as defined by the first example in Section \[sec:two-exampl-class\]. Taking the limit $\hbar \to 0$ in Eq. produces a nonsensical result. This indicates that the real and imaginary parts of $\psi$ are not appropriate variables with regard to this limiting process, probably because they become singular in the limit $\hbar \to 0$. Thus, a different set of dynamical variables, with regular behavior in the limit, should be chosen. There is convincing evidence, from various physical contexts, that appropriate variables, denoted by $\rho$ and $S$, are defined by the transformation $$\label{eq:HDUI9J2UI}
\psi = \sqrt{\rho}\,\mathrm{e}^{i S/\hbar}
\mbox{.}$$ This transformation has been introduced by Madelung. [@madelung:quantentheorie] Note that using these variables in a meaningful limiting process requires that the modulus of $\psi$ remains regular while its phase diverges like $\hbar^{-1}$ for small $\hbar$. This singular behavior, which was noted very early, [@vanvleck:correspondence] is the behavior of the majority of “well-behaved” quantum states. Other, more singular, states may, however, behave in a different manner and will then require a different factorization in terms of $\hbar$. This may also lead to different equations in the limit $\hbar \to 0$; an example will be given in Section \[sec:combi-limit-quant\].
In terms of the new variables, Schrödinger’s equation takes the form of two coupled nonlinear differential equations. The first is a continuity equation which does not contain $\hbar$, $$\label{eq:CA2IH3TMF}
\frac{\partial \rho}{\partial t}+\frac{\partial}{\partial x_{k}}
\frac{\rho}{m} \frac{\partial S}{\partial x_{k}}=0
\mbox{.}$$ The second equation contains $\hbar$ as a proportionality factor in front of a single term: $$\label{eq:QHKUZ24MF}
\frac{\partial S}{\partial t}+ \frac{1}{2m}
\sum_{k}
\left( \frac{\partial S}{\partial x_{k}} \right)^{2} +V -
\frac{\hbar^2}{2m}\frac{\triangle\sqrt{\rho}}{\sqrt{\rho}}
=0
\mbox{.}$$ Equation is referred to as quantum Hamilton-Jacobi equation (QHJ). The $\hbar$-dependent “quantum term” in Eq. describes the influence of $\rho$ on $S$. (It is frequently denoted as “quantum potential,” which is an extremely misleading nomenclature because a potential is, as a rule, an externally controlled quantity.) Its physical meaning, as interpreted by the present author, has been discussed in more detail elsewhere. [@klein:schroedingers]
In the limit $\hbar \to 0$ the quantum term disappears. Thus the standard limit of QT is given by two partial differential equations: the continuity equation , which depends on $\rho$ and $S$, and the Hamilton-Jacobi (HJ) equation, $$\label{eq:QHCL14MF}
\frac{\partial S}{\partial t}+ \frac{1}{2m}
\sum_{k}
\left( \frac{\partial S}{\partial x_{k}} \right)^{2} +V
=0
\mbox{,}$$ which depends only on $S$. The two equations and , which will be referred to as probabilistic Hamilton-Jacobi theory (PHJ), constitute the classical limit of Schrödinger’s equation or single-particle QT. Clearly, this limit does not agree with the trajectory equations of NM.
Much confusion has been created by the fact that the Hamilton-Jacobi formulation of classical mechanics allows the determination of particle trajectories with the help of the HJ equation. From the fact that this equation can be obtained from QT in the limit $\hbar \to 0$ it is often concluded, neglecting the continuity equation, that classical mechanics is the $\hbar \to 0$ limit of QT. However, the limit $\hbar \to 0$ of QT does not provide us with the theory of canonical transformations, which is required to actually calculate particle trajectories. Note also that for exactly those quantum-mechanical states that are most similar to classical states (e.g., coherent states, see Section \[sec:combi-limit-quant\]), the classical limit of QHJ *differs* from HJ. There *is* in fact a connection between the PHJ and NM, but this requires a second limiting process, as will be explained in Section \[sec:formal\_standard-limit-quant\].
Both the PHJ and its (standard) covering theory QT are probabilistic theories, which provide statistical predictions (probabilities and expectation values) if initial values for $S$ and $\rho$ are specified. Although we now have partial differential equations, the relation between QT and PHJ resembles in essential aspects the relation between relativistic mechanics and NM.
The deterministic limit of the standard limit of quantum theory {#sec:formal_standard-limit-quant}
===============================================================
The deterministic limit of the classical limit PHJ of QT is of considerable interest for the present problem, despite the fact that the PHJ no longer contains $\hbar$. Existence of a deterministic limit implies that $\rho(x,t)$ takes the form of a delta function peaked at trajectory coordinates $r_k(t)$ \[which, hopefully, should then be solutions of the classical Eqs. (\[eq:NSEDSCHL\])\]. Thus, adopting a standard formula, we may write $\rho(x,t)$ as an analytic function $$\label{eq:AFT18FO6TD}
\rho_{\epsilon}(x,t)= \left(\frac{1}{\pi \epsilon} \right)^{3/2}
\exp \left\{-\frac{1}{\epsilon}\sum_{k=1}^{3}\left[x_k-r_k(t) \right]^{2} \right\},$$ which represents $\delta^{(3)}(x-r(t))$ under the integral sign in the limit $\epsilon \to 0$: $$\label{eq:D7HLMM9F}
\lim_{\epsilon \rightarrow 0}\rho_{\epsilon}(x,t)=\delta^{(3)}(x-r(t))
\mbox{.}$$ In order to check whether or not this deterministic representation of $\rho(x,t)$ is compatible with the basic equations of PHJ, we insert the Ansatz (\[eq:AFT18FO6TD\]) into the continuity equation (\[eq:CA2IH3TMF\]) and calculate the derivatives. After some rearrangement, Eq. (\[eq:CA2IH3TMF\]) takes the form [@nikolic:classical] $$\label{eq:EZ4SRW8MGB}
\rho_{\epsilon}(x,t)
\left\{
\left[x_k-r_k(t)\right]
\left(p_k(t)-\frac{\partial S(x,t)}{\partial x_k}\right)
+ \frac{\epsilon}{2}
\frac{\partial}{\partial x_k} \frac{\partial}{\partial x_k} S
\right\}=0
\mbox{,}$$ where $p_k(t)=m\dot{v}_k(t)$. At this point we recall that in the PHJ a momentum field $p_k(x,t)$, defined by $$\label{eq:JU25ZT7SD}
p_k(x,t)=\frac{\partial S(x,t)}{\partial x_k}$$ exists. The trajectory momentum $p_k(t)$ should be clearly distinguished from this field momentum $p_k(x,t)$.
In the limit $\epsilon \to 0$, $\rho_{\epsilon}$ becomes a distribution and both sides of Eq. (\[eq:EZ4SRW8MGB\]) have to be integrated over three-dimensional space in order to obtain a mathematically well-defined expression. The first term in the bracket vanishes as a consequence of the term $x_k-r_k(t)$ (at this point we start to disagree with Ref. [@nikolic:classical]). The second term vanishes too for $\epsilon \to 0$ provided the second derivative of $S$ is regular at $\epsilon=0$. But this can safely be assumed since the equation for $S$ does not contain $\rho$. We conclude that the singular (deterministic) Ansatz (\[eq:AFT18FO6TD\]) is a valid solution of PHJ for arbitrary $S$.
The present theory is incomplete since differential equations for the particle trajectories $r_k(t)$ still have to be found. Generally, two conditions must be fulfilled in order to define particle coordinates in a probabilistic theory, namely (i) the limit of sharp (deterministic) probability distributions must be a valid solution, and (ii) an evolution law for the time-dependent expectation values must exist. We have just shown that the first (more critical) condition is fulfilled; Ehrenfest-like relations corresponding to the second condition exist in almost all statistical theories. For the PHJ these take exactly the same form as in QT, namely [@kocis:ehrenfest; @klein:statistical] $$\begin{aligned}
\frac{d}{dt} \overline{x_k} & = & \frac{\overline{p_k}}{m}
\mbox{,}
\label{eq:FIRUZT2TSCHL}\\
\frac{d}{dt} \overline{p_k} & = &
\overline{-\frac{\partial V(x,t)}{\partial x_k}}
\label{eq:SEUZT3TSCHL}
\mbox{,}\end{aligned}$$ where average values such as $\overline{x_k}$ are defined according to the standard expression $$\label{eq:DE1HFDZ8EV}
\overline{x_k}(t) =\int \mathrm{d^{3}} x\, \rho(x,t)\, x_k
\mbox{.}$$ From Eq. and the continuity equation we obtain the useful relation $$\label{eq:AZGR345FFDE}
\overline{p_k}(t) = \int \mathrm{d^{3}} x\, \rho(x,t)\,
\frac{\partial S(x,t)}{\partial x_k}
\mbox{.}$$ Since we have shown that the deterministic limit for $\rho$ is a valid solution of PHJ, we may now use Eq. and obtain in the limit $\epsilon \to 0$ the identification of trajectory quantities, $$\label{eq:FI21OTR89QU}
\overline{x_k}(t) = r_k(t), \qquad \overline{p_k}(t) = m\dot{r}_k(t) =p_k(t)
\mbox{,}$$ from the definitions of the expectation values. The differential relations connecting these quantities follow from Ehrenfest’s theorem and agree with the basic Eqs. of NM. A completely different type of physical law has emerged from the field theoretic relations of the PHJ theory. Thus, classical mechanics is, indeed, contained in PHJ as a deterministic limit, in analogy to the second example of Section \[sec:two-exampl-class\].
Equation takes in this limit the form $$\label{eq:TRA12TRAQU}
p_k(t)=\frac{\partial S(x,t)}{\partial x_k}\Big|_{x=r(t)}=\frac{\partial S(r(t),t)}{\partial r_k(t)}
\mbox{,}$$ which provides an interesting link between a particle variable and a field variable. We expect for consistency that this link admits a derivation of the equation for $\dot{p}$ \[see Eqs. \] from the (field-theoretic) HJ equation. This is indeed the case. We calculate the derivative of the HJ equation with respect to $x_i$, change the order of derivatives with respect to $x_i$ and $t$, and project the resulting relation on the trajectory points $x_k=r_k(t)$. This leads to the equation $$\label{eq:NEI8LTRS1F}
\frac{\partial}{\partial t}\Big|_{2.}
\frac{\partial S(r(t),t)}{\partial r_i(t)} +
\frac{1}{m} \frac{\partial S(r(t),t)}{\partial r_k(t)}
\frac{\partial^{2} S(r(t),t)}{\partial r_i(t) \partial r_k(t)}+
\frac{\partial V(r(t),t)}{\partial r_i(t)} =0
\mbox{, }$$ where the notation indicates that the time derivative operates on the second argument of $S$ only. Using now Eq. and the definition of particle momentum, we see that the first two terms of Eq. agree exactly with the (total) time derivative of $p_i(t)$ and Eq. becomes the second Newton equation. This establishes the connection between the PHJ equations and trajectory differential equations mentioned in Section \[sec:stand-limit-quant\], and completes our treatment of the deterministic limit of the PHJ theory. This derivation of NM seems to be new; it is based on several interesting papers [@rosen:classical_quantum; @cohn:quantum; @kobe:comments; @nikolic:classical; @gondran:discerned] reporting important steps in the right direction.
The deterministic limit of quantum theory {#sec:formal-limit-quant}
=========================================
In QT, the coupling term in QHJ \[see Eq. \] prevents a deterministic limit of the kind found for the PHJ. To see this, we start from the assumption that a quantum mechanical system exists that admits a solution of the form of Eq. for arbitrary $t$. Inserting Eq. into Eqs. and leads to two equations. The first is the continuity equation, which takes the same form as before. The second is the QHJ, which takes the form $$\label{eq:QHIMIR23MF}
\frac{\partial S}{\partial t}+ \frac{1}{2m}
\sum_{k}
\left( \frac{\partial S}{\partial x_{k}} \right)^{2} +V +
\frac{1}{2m}
\left( \frac{\hbar}{\epsilon} \right)^{2}
\left\{ 3 \epsilon- \sum_{k=1}^{3} \left[x_k-r_k(t) \right]^{2}\right\}
=0
\mbox{.}$$ Equation shows that the coupling term diverges (for finite $\hbar$) in the limit $\epsilon \to 0$. Consequently, there is no reason to expect that the second derivative of $S$ with respect to $x_k$ \[see Eq. \] is regular at $\epsilon \to 0$ and that a delta-function-like $\rho(x,t)$, as given by Eq. , can be a solution of Eqs. and . Thus, the deterministic limit of QT (if it exists) cannot be obtained in the same way as in the PHJ.
We next consider several concrete solutions of QT that lead to probability densities similar to Eq. . As a first example we consider an ensemble of free particles that are distributed at $t=0$ according to a probability density centered at $r_k(0)=0$ \[set $r_k(t)=0$ in Eq. \]. The initial value for $S(x,t)$ is given by $S(x,0)=p_{0,k}x_k$, i.e., $S(x,t)$ fulfills at $t=0$ the deterministic relation . These initial values describe for small $\epsilon$ a localized, classical particle in the sense that there is no uncertainty with respect to position or momentum. A calculation found in many textbooks leads to the following solution of Schrödinger’s equation for $\rho$: $$\rho(x,t)= \left(\frac{1}{\pi A(t)} \right)^{\frac{3}{2}}
\exp \left\{-\frac{1}{A(t)}\sum_{k=1}^{3}
\left[ x_k - r_k(t) \right]^{2} \right\},
\label{eq:ALG5SC9IDON}$$ where $mr_k(t)=p_{0,k}t$ and $A(t)=A_{f}(t)=\epsilon [1+(\hbar/\epsilon)^{2}(t/m)^{2} ]$. We see from Eq. that the peak of $\rho$ moves in agreement with NM, but the width of the wave packet increases with increasing time as well as with *decreasing* $\epsilon$. A complete localization can be achieved only at $t=0$. At later times the quantum uncertainty, due to the finite $\hbar$, dominates the behavior of the ensemble completely, despite our choice of “deterministic” initial conditions.
As a second example, we consider an ensemble of particles moving in a harmonic oscillator potential $V(x)=(m\omega^{2}/2)x_kx_k$, using exactly the same initial conditions as in the above example of force-free motion. The result for $\rho$ takes the same form as for the force-free ensemble \[see Eq. \], but with $mr_k(t)=(p_{0,k}/\omega) \sin \omega t$ and $$\label{eq:JW2SD6LH9Z}
A(t)=A_{h}(t)=\epsilon
\left[ \cos^{2}\omega t +
\frac{1}{m^{2}\omega^{2}} \left( \frac{\hbar}{\epsilon}\right)^{2}\sin^{2}\omega t \right].$$ The width $A_{h}(t)$ increases again with decreasing $\epsilon$ and prevents again a deterministic limit. We mention, without going into details, [@ter_haar:problems] that a third example showing the same behavior can be found, namely an ensemble of particles moving under the influence of a constant force.
The three examples considered in this section correspond to three potentials proportional to $x_k^{n}$, where $n=0,1,2$. For these potentials the expectation values of the corresponding forces fulfill the relation $\overline{F_k(x)}=F_k(\overline{x})$. Therefore, equations of motion for $\overline{x_k}$ and $\overline{p_k}$ exist as a consequence of Ehrenfest’s theorem. Despite these classical features, even these “optimal” states do not permit a deterministic limit of QT. We conclude, in agreement with common wisdom, that this limit does not exist.
The combined limit of quantum theory {#sec:combi-limit-quant}
====================================
Let us summarize what has been achieved so far. In Section \[sec:stand-limit-quant\] the limit $\hbar \to 0$ has been taken for arbitrary quantum states (including wave packets with fixed width $\epsilon$). The result of this first “standard limiting process” was a classical statistical theory referred to as probabilistic Hamilton-Jacobi theory (PHJ). In Section \[sec:formal\_standard-limit-quant\] the limit $\epsilon \to 0$ of PHJ has been taken. The result of this second “deterministic limit” was Newtonian mechanics (NM). Therefore NM is a subset of the classical limit PHJ of quantum theory (QT), but NM is not *the* classical limit of QT, since we cannot neglect almost all of the (statistical) states of PHJ. Thus, the two limiting processes performed in this order have not led us from QT to NM in the sense that NM can be said to be the classical limit $\hbar \to 0$ of QT. In Section \[sec:formal-limit-quant\] it has been shown that inverting the order of the two limiting processes (first $\epsilon \to 0$ then $\hbar \to 0$) does not solve the problem either since the limit $\epsilon \to 0$ (with $\hbar$ fixed) does not exist. The two limiting processes clearly do not commute. Thus, it is impossible to obtain NM as the classical limit of QT, no matter which order of the two (separate) limiting processes is chosen.
Fortunately, we have still the option to *combine* both limits. That is, we could assume that the width of the wave packets is a monotonic function of $\hbar$. This means that the localization of wave packets (the deterministic limit) and the change of the basic equations of QT (the standard limit) take place *simultaneously* in the limit $\hbar \to 0$. Such states seem artificial from the point of view of experimental verification since the numerical value of $\hbar$ is not under our control. Nevertheless, a construction of NM from QT along these lines would certainly provide a kind of justification for Dirac’s claim that QT reduces to NM in the limit $\hbar \to 0$. Note also that the subject of our study is essentially of a formal nature. We are asking whether or not all predictions of NM can be obtained, by means of some limiting process $\hbar \to 0$, from the basic equations of QT. There are no in-principle constraints on how to perform this limit.
A brief look at the above examples for $A(t)$ shows that a *linear* relation between $\hbar$ and $\epsilon$ seems most promising. Thus, we set $$\label{eq:DRMI45WWLG}
\epsilon=k\hbar
\mbox{,}$$ were $k$ is an arbitrary constant. In order to use a notation similar to that in Section \[sec:formal\_standard-limit-quant\] \[see Eq. \], $\epsilon$ will be used instead of $\hbar$ as small parameter; it may be identified with $\hbar$ in most of the following relations. Let us perform the identification for the two examples considered in Section \[sec:formal-limit-quant\], with potentials $V(x)=V_{f}(x)=0$ and $V(x)=V_{h}(x)=(m\omega^{2}/2)x_kx_k$, respectively. Using the same initial conditions as in Section \[sec:formal-limit-quant\], the solutions for $\rho$ and $S$ of Eqs. and take essentially the same form in both cases, namely $$\begin{aligned}
\rho(x,t)&=& \left(\frac{1}{\pi \epsilon(t)} \right)^{3/2}
\exp \left\{-\frac{1}{\epsilon(t)}\sum_{k=1}^{3}
\left[ x_k - r_k(t) \right]^{2} \right\},
\label{eq:HZ46TRE9LPL}\\
S(x,t)&=& \frac{m}{4}\frac{\dot{\epsilon}(t)}{\epsilon(t)}
\sum_{i=1}^{3}\left[x_i-r_i(t) \right]^{2}-\frac{1}{2}p_{k}(t)r_{k}(t)
+p_{k}(t)x_k - \sum_{i=1}^{3}\frac{\epsilon}{2k}\tan^{-1}
\frac{r_{i}(t)}{k p_{i}(t)}
\label{eq:SO8PER34DFL}
\mbox{.}\end{aligned}$$ The solutions ($\rho_{f},\,S_{f}$) and ($\rho_{h},\,S_{h}$) for particle ensembles in force-free regions and linear-force fields may be obtained from Eqs. and by using different widths $\epsilon(t)=\epsilon_{f}(t)$ and $\epsilon(t)=\epsilon_{h}(t)$, given by $$\label{eq:JUZA2GR9WQ}
\epsilon_{f}(t) = \epsilon \left(1+\frac{t^{2}}{k^{2}m^{2}} \right), \qquad
\epsilon_{h}(t)=\epsilon
\left[ \cos^{2}\omega t +
\frac{1}{k^{2}m^{2}\omega^{2}} \sin^{2}\omega t \right],$$ and different trajectory components $r_k(t)=r_k^{(f)}(t)$ and $r_k(t)=r_k^{(h)}$ (as well as momentum components $p_k(t)=m \dot{r}_k(t)$), given by $$\label{eq:J19UHT45POI}
r_k^{(f)}(t)=\frac{p_{0,k}}{m}t , \qquad
r_k^{(h)}=\frac{p_{0,k}}{m \omega t}\sin \omega t.$$ As Eq. shows, both widths are time-dependent; for the free-particle ensemble the width increases quadratically, while for the bounded motion of the harmonic oscillator it varies periodically. However, both widths vanish in the limit $\epsilon \to 0$ for arbitrary (finite) times $t$. This means that the deterministic probability density we were looking for is, in fact, created in this limit. The solutions for $S$ are well-behaved at $\epsilon=0$. The limiting process in the continuity equation can be performed in a similar way as in Section \[sec:formal\_standard-limit-quant\] (an additional term due to the time dependence of $\epsilon(t)$ is regular at $\epsilon=0$). The remaining steps—the derivation of Newton’s equations and their field-theoretic derivation from the QHJ—can be performed in the same way as in Section \[sec:formal\_standard-limit-quant\]. Note also that the QHJ is regular at $\epsilon=0$ and *differs* in this limit from the HJ equation \[cf. the discussion following Eq. \]. In view of a recent discussion [@ling:limit; @kazandjian:limit] it should be noted that this field-theoretic limit is not equivalent to its projection on the trajectory.
The above solutions, with $\epsilon$ and $\hbar$ considered as *independent* parameters (as in Section \[sec:formal-limit-quant\]), have been reported many times in the literature. It has also been pointed out that the special value $\epsilon=\hbar/m\omega$, in the harmonic oscillator example, produces the coherent states found by Schrödinger.[@schroedinger:continuous] On the other hand, the relevance of the weaker statement $\epsilon=k\hbar$ for the classical limit problem has apparently not been recognized. It is not necessary to restrict oneself to the coherent states of the harmonic oscillator (the special case $k=1/m\omega$) in order to obtain deterministic motion; the latter may be obtained for a much larger class of force-free states, harmonic oscillator states, and constant-force states (this last example has not been discussed explicitly) as shown above. Summarizing this section, we found three potentials $V(x) \sim x^{n},\,n=0,1,2$, which allow for a derivation of NM from QT in the limit $\hbar \to 0$. For these potentials, equations of motion for $\bar{x}_k,\ \bar{p}_k$ exist, as mentioned already. Home and Sengupta [@home_sengupta:] have shown that for these potentials the form of the quantum-mechanical solution may be obtained with the help of the classical Liouville theorem.
Can the combined limit be performed for all potentials? {#sec:are-all-potentials}
=======================================================
We know now that three potentials exist which, for properly chosen initial wave packets, lead to deterministic equations of motion in the limit $\hbar \to 0$. We shall refer to such potentials for brevity as *deterministic potentials*. A (complete) reconstruction of NM from QT requires that *all* (or almost all) potentials are deterministic. In this section we ask if this can be true.
As a first point we note that the probability density $\rho(x,t)$ of all deterministic wave packets takes, by definition, a very specific functional form, namely one that reduces, like Eq. , in the limit $\hbar \to 0$ to a delta function. This fixes essentially one of our two dynamic variables; we have two differential equations for a single unknown variable $S(x,t)$. It seems unlikely that this overdetermined system of equations admits solutions for $S(x,t)$ for *arbitrary* potentials $V$.
The second point to note is that the existence of a deterministic limit fixes not only the functional form of $\rho$ but also its argument. Let us assume that a deterministic solution for $\rho$ and $S$, with $\rho(x,t)$ taking the form of Eq. with unspecified $\epsilon(t)$, exists. The probability density $\rho(x,t)$ depends necessarily on $\vec{x}-\vec{r}(t)$, where the position vector $\vec{r}(t)$ is a solution of Newton’s equation for the same potential $V(x)$ that occurs in the Schrödinger equation. The crucial point is that this dependence is not created by the limiting process $\hbar \to 0$ but is already present for finite $\hbar$, in the exact quantum-mechanical solution. For given initial conditions it has been created, so to say, by the quantum-theoretical formalism. This implies that $\vec{r}(t)$ describes for finite $\hbar$ not the time dependence of a particle trajectory but of a position expectation value. Since $\vec{r}(t)$ is (again for finite $\hbar$) the solution of Newton’s equations, such equations for expectation values must already be present in the quantum-theoretical formalism. Of course, in the deterministic limit $\hbar \to 0$ the difference between particle trajectories and expectation values vanishes, but the important point is that Newton’s equations must hold already for finite $\hbar$. We conclude that the existence of the equations of motion of NM for position expectation values is a necessary condition for the existence of deterministic potentials.
This line of reasoning leads to a mathematical condition for deterministic potentials. Let us assume that we have a deterministic potential $V^\textrm{(det)}(x)$ in our quantum-theoretical ($\hbar$ finite) problem. We calculate the expectation value $\overline{x_k}(t)$ as defined by Eq. using the deterministic probability density Eq. . We obtain $\overline{x_k}(t)=r_{k}(t)$, i.e., the expectation value follows the time dependence of the trajectory \[the peak of $\rho (x,t)$\]. The latter must fulfill Newton’s equation with the force derived from $V^\textrm{(det)}(x)$; otherwise the deterministic limit could not exist. Using these facts in Ehrenfest’s equations, Eqs. and , we obtain immediately the integral equation for deterministic potentials $$\label{eq:EFU29EDIREL}
F^\textrm{(det)}_{k}(r)=\int \mathrm{d^{3}} x\,
\delta_{\epsilon}^{(3)}(x-r)
F^\textrm{(det)}_{k}(x)
\mbox{,}$$ where $F^\textrm{(det)}_{k}(x)=-\partial V^\textrm{(det)}(x) / \partial x_{k}$. The quantity $\rho(x,t)$ has been renamed $\delta_{\epsilon}^{(3)}(x-r(t))$ in order to show the convolution-type structure of the equation. Note that Eq. is only for $\epsilon > 0$ a constraint for $V^\textrm{(det)}(x)$. It is easy to see that a particular solution is given by $V^\textrm{(det)}(x)=a+b_kx_k+
c_{i,k}x_ix_k$, where the coefficients may depend on time. This is essentially a linear combination of the three deterministic potentials $x^{n},\,n=0,1,2$, found already in the last section. According to a theorem by Titchmarsh [@titchmarsh:fourier_integrals] (a simple proof may be obtained with the help of the theory of generalized functions;[@lighthill:generalized] see the Appendix), other solutions of Eq. do not exist. This theorem shows that the “combined limit” cannot be performed for all potentials. Although the present treatment does not cover all conceivable physical situations, the results obtained so far imply already definitively that the limit $\hbar \to 0$ of QT does not generally agree with NM.
Discussion {#sec:discussion}
==========
In our first limiting procedure, which is appropriate for “well-behaved” quantum states, we found that QT agrees in the limit $\hbar \to 0$ with a classical statistical theory referred to as PHJ. The latter contains as a limiting case the deterministic states ruled by NM. Let us stress once again that the fact that these deterministic states are contained in QT does not mean that NM is the limit of QT. This limit is PHJ, which contains a much larger number of (probabilistic) states not belonging to NM. In a second attempt, the limit $\hbar \to 0$ was simultaneously applied to the wave-packet width and in the basic equations. We found that almost all states do *not* admit a transition from QT to NM in the limit $\hbar \to 0$.
Our final result is then that NM cannot be obtained from QT, at least by means of a mathematical limiting process $\hbar \to 0$. This result has been obtained in the framework of standard QT using the Schrödinger picture to describe the quantum dynamics. One may ask whether this conclusion is specific for these choices, or remains true for other dynamical pictures, such as the Heisenberg picture, and for other formulations of QT, such as Feynman’s path-integral formulation. Recall that the present approach is (as discussed in Section \[sec:Introduction\]) solely based on predictions, i.e., the *numerical output* of the quantum theoretical formalism. These numbers do not depend on a particular picture of quantum dynamics. They are also independent from the choice of a particular formulation of QT since all formulations of QT must lead to the same predictions.
Let us illustrate the last point with a discussion of the path-integral formulation of QT.[@feynman.hibbs:quantum] The central quantity of this approach, the propagator, is an infinite sum of terms of the form $\sim\exp iS/\hbar$, where $S$ is the classical action and each term in the sum is to be evaluated along a different path between the initial and final space-time points $x_0$ and $x_1$. In the classical limit $\hbar \to 0$ the dominating contribution to the sum comes from the classical path between $x_0$ and $x_1$, which extremizes $S$. The fact that this path obeys the differential equations of NM is sometimes interpreted in the sense of a transition from QT to NM, which the path-integral formulation reveals in a particular rigourous and straighforward way. Such an interpretation is not justified. The form of the propagator says nothing about whether or not a particle is really present at the initial space-time point $x_0$; it tells us just what will happen *given that* a particle occupies the point $x_0$ with certainty. The second variable of QT, the probability density $\rho$, must also be taken into account; it enters the initial state and makes the final state uncertain despite the deterministic form of the propagator. A general classification scheme for probabilistic theories, taking the different roles of initial values and evolution equations into account, has been reported recently. [@klein:statistical] It is a general feature of classical statistical theories that the time evolution in the event space (configuration or phase space) is deterministic and the impossibility of making deterministic predictions (on single events) is solely due to uncertainty in the initial values. This classical feature is also visible in the phase-space theory defined by Eq. and (though in a less explicit way) in the configuration-space theory PHJ \[see Eqs. and \]. It is this feature, and not the transition from QT to NM, that is most explicit in the path-integral formalism. The classical limit of Feynman’s version of QT is equivalent to the classical limit of Schrödinger’s version of QT (the PHJ), since both versions are equivalent.
Our final result is in disagreement with Dirac’s statement quoted at the beginning of our study. Dirac discusses the problem of the classical limit in Section 31 of his book. He formulates the following general principle:
> For any dynamical system with a classical analogue, a state for which the classical description is valid as an approximation is presented in quantum mechanics as a wave packet,… so in order that the classical description be valid, the wave packet should remain a wave packet and should move according to the laws of classical dynamics. We shall verify that this is so.
The following calculation is intended to show that such a wave packet always exists. Unfortunately, a systematic investigation of different classes of potentials or initial values is not performed. Instead, Dirac imposes several conditions for the considered wave packets, formulated verbally or in the form of inequalities, which he assumes to be true for arbitrary potentials but which need not necessarily be true. He arrives at the canonical equations of motions for the peaks of supposedly arbitrary wave packets. In reality, these conditions impose strong restrictions on the forms of initial values and potentials and can be fulfilled only in a few very special cases. Quantum-mechanical solutions for the three “deterministic potentials,” where Ehrenfest’s relations agree with NM, are often used to demonstrate so-called “classical behavior” of wave packets. It should be borne in mind that this behavior is not generic but represents the exception(s) from the rule.
After the discovery of QT, the community was shocked by the breakdown of NM in the microscopic world and it seemed inconceivable that NM should not even survive as the classical limit of QT. Schrödinger, like Dirac, considered this as evident, and wrote at the end of his famous paper about coherent states:[@schroedinger:continuous]
> We can definitely foresee that, in a similar way, wave groups can be constructed which move round highly quantized Kepler ellipses and are the representation by wave mechanics of the hydrogen electron.
The coherent states of the harmonic oscillator have been generalized to arbitrary potentials in various ways,[@zhang_et_al:coherent] but none of these generalizations admits a clear transition to the classical (deterministic) limit. Special attention was, of course, devoted to the Coulomb potential, but despite intense research, Schrödinger’s idea could not be realized and this chapter has apparently already been closed.[@zlatev_et_al:possibility]
The classical limit of QT is the PHJ, a classical statistical theory defined by Eqs. and . The limit $\hbar \to 0$ transforms a quantum probabilistic theory into a classical probabilistic theory. The behavior of the uncertainty relation illustrates this conclusion in a simple way. For $\hbar \to 0$ it takes the form $$\label{eq:D3TR12GUPR}
\Delta x \,\Delta p \geq 0
\mbox{,}$$ which means that in the classical limit the uncertainty product is in general different from zero; a detailed comparison has been reported by Devi and Karthik. [@devi_karthik:uncertainity] Almost all states of PHJ will show uncertainties; the equality sign in Eq. just indicates that the transition to the deterministic limit (as performed in Section \[sec:formal\_standard-limit-quant\]) is not forbidden.
The classical limit plays an important role in the prolonged discussion about the proper interpretation of QT. In the years after the discovery of QT, a number of dogmas were established, which have been repeated since then so many times that they are considered today as “well-established” scientific facts. One of these dogmas states that “QT provides a complete description of individual particles.” It is hard to understand how a probabilistic theory could provide a “complete” description of individual events. But one should first analyze the possible meanings of the term “complete.” A detailed analysis shows that this term is ambiguous.[@klein:completeness] It may mean “no better theory exists” (metaphysical completeness) or “all facts that can be observed can be predicted” (predictive completeness). The still prevailing (Copenhagen) standard interpretation claims that QT is complete in both respects. Einstein, Podolsky, and Rosen (EPR), showed that QT is predictive-incomplete. [@einstein.podolsky.ea:can] In the last paragraph of their paper,[@einstein.podolsky.ea:can] the authors expressed their *belief* that QT is metaphysical-incomplete. EPR’s proof of predictive incompleteness was correct and could not be attacked, so metaphysical incompleteness was attacked instead. The EPR paper was misinterpreted as if the authors had claimed they had proven metaphysical incompleteness. Further consequences of this misinterpretation are discussed elsewhere.[@klein:completeness] The important point to note is that metaphysical completeness is a philosophical term; physics can test only predictive completeness (by comparison with observation). Thus, let us concentrate on the question of predictive completeness of QT; in the remaining part of this section the term completeness will be used in this sense.
The fact that NM does not emerge from QT, at least in the sense of the mathematical limit $\hbar \to 0$, presents a painful obstacle to the completeness dogma. Every statistical theory, whether classical or quantum, is unable to predict individual events and is therefore, by its very definition, incomplete. How can quantum theory be complete if its classical limit is incomplete? In order to eliminate this problem, Bohr created the “correspondence principle.” It postulates that quantum states become similar to states of NM for large values of $S/\hbar$. However, this principle is not in agreement with the structure of Schrödinger’s equation. For large $S/\hbar$ the quantum term in Eq. becomes negligibly small and QT becomes similar to PHJ and not to NM; similarity with NM requires *in addition* a sharply peaked $\rho$. The breakdown of Bohr’s correspondence principle in concrete situations has been reported many times in the literature; see e.g. Cabrera and Kiwi [@cabrera_kiwi:large] and Diamond. [@diamond:classically]
Let us finally discuss the logical implication leading from Eq. to Eq. , at the very beginning of the present paper, from the point of view of the interpretation of QT. If Eq. really presents a (complete) description of a single electron, then the fact that the complete classical description, Eq. , follows from Eq. in the limit $\hbar \to 0$ seems to be evident, not for mathematical but for interpretational reasons. But we have mathematically shown that Eq. does not follow from Eq. . Thus, the premise in the above interpretation of the implication “Eq. $\to$ Eq. ” cannot be true, i.e., the assumption that Eq. provides a (complete) description of a single electron cannot be true. Therefore, our result provides an argument in favor of the statistical interpretation, [@ballentine:statistical] which claims that QT provides a complete description not of single particles but of statistical ensembles only.
Conclusion {#sec:conclusion}
==========
This work has compared the predictions of QT for vanishing $\hbar$ with the predictions of NM. Generally, the predictions of a physical theory are a logical consequence of (a) the mathematical form of its basic equations, and (b) the set of initial values. Both aspects have been studied in this work, in order to take into account all possible ways to obtain predictions of NM from QT in the limit $\hbar \to 0$. This comparison has been performed for the simplest and most significant situation of single particles in external potentials. Our conclusion is that NM does not emerge from QT, at least in the sense of the straighforward application of the limit $\hbar \to 0$. This mathematical result should be taken into account in considerations about the interpretation of QT.
Appendix {#sec:appendix}
========
Using the theory of generalized functions,[@lighthill:generalized] the integral equation can be solved quickly. We discuss for simplicity the one-dimensional integral equation, $$\label{eq:AODVOT8EE}
f(x)=\int d x\, k(x-y) f(y)
\mbox{,}$$ where $k(x)$ is a normalized Gaussian. Introducing the Fourier transforms $F(k)$ and $K(k)$ of $f(x)$ and $k(x)$, Eq. takes the form $$\label{eq:FTRVOT78E}
\int d k\, F(k) \left[ 1- \sqrt{2\pi} K(k)\right] =0
\mbox{.}$$ Nontrivial solutions for $F(k)$ \[and $f(x)$\] can exist only if the factor in brackets has zeros. This is the case, since the Fourier transform of a normalized Gaussian is again a normalized Gaussian and therefore the bracketed quantity vanishes at $k=0$ with a leading quadratic term. Thus, the Fourier transform of every solution of must obey $$\label{eq:SHZT45IU9}
F(k)k^{2}=0$$ for all $k$. This implies that the Fourier transform of $k^{2}F(k)$, which is proportional to the second derivative $f''(x)$ of $f(x)$, vanishes too. This implies $f(x) = a +bx$, in agreement with a theorem by Titchmarsh.[@titchmarsh:fourier_integrals] \[The solution for $F(k)$ is a linear combination of $\delta(k)$ and $\delta'(k)$.\]
[10]{}
P. A. M. Dirac, *The Principles of Quantum Mechanics*, 4th ed.(Oxford University Press, Oxford, 1958), p. 88.
V. Allori and N. Zanghi, “On the classical limit of quantum mechanics,” Found. Phys. [**39**]{}, 20–32 (2009).
E. G. P. Rowe, “Classical limit of quantum mechanics [(]{}electron in a magnetic field[)]{},” Am. J. Phys. [**59**]{}, 1111–1117 (1991).
R. F. Werner and M. P. H. Wolff, “Classical mechanics as quantum mechanics with infinitesimal $\hbar$,” Phys. Lett. A [**202**]{}, 155–159 (1995).
N. Rosen, “The relation between classical and quantum mechanics,” Am. J. Phys. [**32**]{}, 597–600 (1964).
R. Schiller, “Quasi-classical theory of the nonspinning electron,” Phys. Rev. [**125**]{}, 1100–1108 (1962).
J. Cohn, “Quantum theory in the classical limit,” Am. J. Phys. [**40**]{}, 463–467 (1972).
M. Gondran and A. Gondran, “The two limits of the [S]{}chr[ö]{}dinger equation in the semi-classical approximation,” in *Foundations of Probability and Physics - 6*, edited by M. D’Ariano et al. (AIP Conference Proceedings, Melville and New York, 2012) p. 318, see also arXiv:1107.0790 \[quant-ph\]
D. H. Kobe, “Comments on the classical limit of quantum mechanics,” Am. J. Phys. [**42**]{}, 73–74 (1973).
H. Nikolic, “Classical mechanics without determinism,” Found. Phys. Lett. [**19**]{}, 553–566 (2006).
E. Schr[ö]{}dinger, *Collected Papers on Wave Mechanics*, 3rd ed.(Chelsea Publishing, New York, 1982), p. 41.
E. Madelung, “Quantentheorie in hydrodynamischer [F]{}orm,” Z. Phys. [**40**]{}, 322–326 (1926).
J. H. [Van Vleck]{}, “The correspondence principle in the statistical interpretation of quantum mechanics,” Proc. Natl. Acad. Sci. U.S. [**14**]{}, 178–188 (1928).
U. Klein, “[Schr[ö]{}dinger’s]{} equation with gauge coupling derived from a continuity equation,” Found. Phys. [**39**]{}, 964 (2009).
L. Kocis, “Ehrenfest theorem for the [H]{}amilton-[J]{}acobi equation,” Acta Phys. Polon. A [**102**]{}, 709–716 (2002).
U. Klein, “The statistical origins of quantum mechanics,” Physics Research International 808424 (2011), [<http://www.hindawi.com/journals/phys/2010/808424.html>](<http://www.hindawi.com/journals/phys/2010/808424.html>).
D. ter Haar, *Selected Problems in Quantum Mechanics* (Infosearch Limited, London, 1964).
L. Kazandjian, “The $\hbar \to 0$ limit of the [S]{}chr[ö]{}dinger equation,” Am. J. Phys. [**74**]{}, 557 (2006).
Song Ling, “On the $\hbar \to 0$ limit of the [S]{}chr[ö]{}dinger equation,” J. Chem. Phys. [**96**]{}, 7869–7870 (1992).
D. Home and S. Sengupta, “Classical limit of quantum mechanics,” Am. J. Phys. [**51**]{}, 265–267 (1983).
E. C. Titchmarsh, *Introduction to the Theory of Fourier Integrals*, 2nd ed. (Oxford, London, 1948), Theorem 146 on p. 305.
M. J. Lighthill, *An Introduction to Fourier Analysis and Generalized Functions* (Cambridge University Press, Cambridge, 1958).
R. P. Feynman and A. R. Hibbs, *Quantum Mechanics and Path Integrals*, (McGraw-Hill, New York, 1965).
L. Zhou and L. M. Kuang, “Coherent states: theory and some applications,” Rev. Mod. Phys. [**62**]{}, 867–927 (1990).
I. Zlatev, W. Zhang, and D. H. Feng, “Possibility that [S]{}chr[ö]{}dinger’s conjecture for the hydrogen-atom coherent states is not attainable,” Phys. Rev. A [**50**]{}, R1973–R1975 (1994).
A. R. Usha Devi and H. S. Karthik, “Uncertainty relations in the realm of classical dynamics,” arXiv:1108.2682 \[quant-ph\].
U. Klein, “Is the individuality interpretation of quantum theory wrong ?,” arXiv:1207.6215 \[quant-ph\], see also [<http://statintquant.net>](<http://statintquant.net>).
A. Einstein, B. Podolsky, and N. Rosen, “Can quantum-mechanical description of physical reality be considered complete?,” Phys. Rev. [**47**]{}, 777–780 (1935).
G. G. Cabrera and M. Kiwi, “Large quantum-number states and the correspondence principle,” Phys. Rev. A [**36**]{}, 2995–2998 (1987).
J. J. Diamond, “Classically forbidden behavior of the quantum harmonic oscillator for large quantum numbers,” Am. J. Phys. [**60**]{}, 912–916 (1992).
L. E. Ballentine, “The statistical interpretation of quantum mechanics,” Rev. Mod. Phys. [**42**]{}, 358–381 (1970).
[^1]: [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
author:
- |
S. Valcke$^{1}$[^1], S. De Rijcke$^{1}$[^2], E. Rödiger$^2$& H. Dejonghe$^1$\
$^{1}$Sterrenkundig Observatorium, Ghent University, Krijgslaan 281, S9, 9000 Gent, Belgium\
$^2$Jacobs University Bremen, PO Box 750 561, 28725 Bremen, Germany
bibliography:
- 'biblio.bib'
date: 'Accepted . Received ; in original form '
title: 'Kelvin-Helmholtz instabilities in Smoothed Particle Hydrodynamics'
---
\[firstpage\]
hydrodynamics – instabilities – methods: numerical keywords
Acknowledgments {#acknowledgments .unnumbered}
===============
We thank the referee for the valuable and constructive comments. We also thank Justin Read for making the code to compute the $E_0$ errors publicly available[^3]. SV acknowledges and is grateful for the financial support of the Fund for Scientific Research – Flanders (FWO). The SPH simulations were run on our local computer cluster ITHILDIN. The <span style="font-variant:small-caps;">FLASH</span> software used for one of the simulations in this paper was developed by the DOE supported ASCI/Alliances Center for Astrophysical Thermonuclear Flashes at the University of Chicago.
\[lastpage\]
[^1]: Doctoral Fellow of the Fund for Scientific Research – Flanders, Belgium (FWO). E-mail: [email protected]
[^2]: E-mail: [email protected]
[^3]: http://www.astrosim.net/code/doku.php?id=codesonline:nbody:osphpatch
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present sharp quantitative weighted norm inequalities for the Hardy-Littlewood maximal function in the context of Locally Compact Abelian Groups, obtaining an improved version of the so-called Buckley’s Theorem. On the way, we prove a precise reverse Hölder inequality for Muckenhoupt $A_\infty$ weights and provide a valid version of the “open property" for Muckenhoupt $A_p$ weights.'
address: 'Departamento de Matemática, Facultad de Ciencias Exactas y Naturales, Universidad de Buenos Aires, Ciudad Universitaria Pabellón I, Buenos Aires 1428 Capital Federal Argentina'
author:
- Victoria Paternostro
- Ezequiel Rela
title: 'Improved Buckley’s theorem on LCA groups'
---
[^1]
Introduction and main results
=============================
The study of weighted norm inequalities for maximal type operators is one of the central topics in harmonic analysis that began with the celebrated theorem of Muckenhaupt [@Muckenhoupt:Ap]. It states that the class of weights (nonnegative locally integrable functions) characterizing the boundedness of the Hardy-Littlewood maximal function $M$ on the weighted space $L^p(\mathbb{R}^n,w dx)$ is the so-called Muckenhaupt $A_p$ class (see below for the precise definitions). It is important to remark that Muckenhaupt’s result is qualitative, that is, it does not provide any precise information of how the operator norm of $M$ depends on the underlying weight in $w\in A_p$. The first quantitative result on the boundeness for the maximal function in $\mathbb{R}^n$ dates back to the 90’s, is due to Buckley [@Buckley] and gives the best possible power dependence on the $A_p$ constant $[w]_{A_p}$. More precisely, Buckley proved that $$\label{eq:Buckley}
\|M\|_{L^p(\mathbb{R}^n,w dx)\to L^p(\mathbb{R}^n,w dx)}\le C [w]^{\frac{1}{p-1}}_{A_p}, \qquad 1<p<\infty.$$ Recently a simpler and elegant proof was presented by Lerner [@Lerner-Elementary] who used a very clever argument composing weighted versions of the maximal function. Later, finer improvements have been found. In particular, there is in [@HPR1] a sharp mixed bound valid in the context of spaces of homogeneous type.
Our purpose here is to obtain sharp quantitative norm estimates in the context of Locally Compact Abelian groups (LCA groups). The modern approach to this problem is to use a sharp version of the reverse Hölder inequality (RHI) with a precise quantitative expression for the exponent to derive a proper open property for the $A_p$ claseses. Then an interpolation type argument allows to prove the desired bound.
In the rest of the introduction we first described in details the context where we will work in and then properly state the results that we will prove.
Muckenhoupt weights and maximal function on LCA groups
------------------------------------------------------
In the euclidean setting the standard way to introduce $A_p$ weights is by considering averages over cubes, balls or more general families of convex sets. In any case, the family is built using some specific metric. In our context of LCA groups we lack of such concept. However there are many LCA groups where we do have the possibility of consider a family of *base sets* satisfying the other fundamental property of the basis of cubes or balls: any point has a family of decreasing base sets shrinking to it and, in addition, the whole space can be covered by the increasing union of such family.
In order to properly defined the $A_p$ classes let us fix an LCA group $G$ with a measure $\mu$ that is inner regular and such that $\mu(K)<\infty$ for every compact set $K\subset G$. Notice that $\mu$ does need to be the Haar measure because we do not assume $\mu$ to be translation invariant. The reader can find a comprehensive treatment of Harmonic Analysis on LCA groups in [@HR-vol2; @HR-vol1; @rudinLCA]. The general assumption on the group will be that it admits a sequence of neighborhoods of $0$ with certain properties that we described in the next definition (cf. ).
A collection $\{U_i\}_{i\in\mathbb{Z}}$ is a covering family for $G$ if
1. $\{U_i\}_{i\in\mathbb{Z}}$ is an increasing base of relatively compact neighborhoods of $0$, $\bigcup_{i\in\mathbb{Z}}U_i=G$ and $\bigcap_iU_i=\{0\}$.
2. There exists a positive constant $D\ge 1$ and an increasing function $\theta:\mathbb{Z}\to\mathbb{Z}$ such that for any $i\in \mathbb{Z}$ and any $x\in G$
- $i\le\theta(i)$
- $U_i - U_i\subset U_{\theta(i)}$
- $\mu(x+ U_{\theta(i)})\le D\mu (x+U_i)$.
We will refer to the third condition as the *doubling* property of the measure $\mu$ with respect to $\theta$ and we will call $D$ the doubling constant. In the case of $\mathbb{R}^n$ equipped with the natural metric and measure, we can consider the family of dyadic cubes of sidelength $2^i$ or the euclidean balls $B(x,2^i)$ for $i\in \mathbb{Z}$. The doubling constant of the Lebesgue measure in this context is $2^n$ and the function $\theta$ can be taken to be $\theta(i)=i+1$. Therefore, the intuition here is that the index $i$ in the above definition can be seen as a sort of *radius* or *size* of the given set $U_i$.
For each $x\in G$, the set $x+U_i$ will be called *base set* and the collection of all base sets will be denoted by $$\label{eq:basesets}
\mathcal{B}:=\left\{x+U_i : x\in G, i\in \mathbb{Z}\right\}.$$
The notion of base sets allows to define a direct analogue of the Hardy Littlewood maximal function: $$\label{eq:Maximal}
Mf(x) = \sup_{x\in U\in\mathcal{B} } \avgint_{U} |f|\ d\mu :=\sup_{x\in U\in\mathcal{B} } \frac1{\mu(U)} \int_{U} |f|\ d\mu.$$
As we already mentioned, our purpose here is to prove sharp weighted norm inequalities for this operator in $L^p(G,w d\mu)$, where $w$ is a weight on $G$ . Firstly, recall that the celebrated Muckenhoupt’s Theorem asserts that the class of weights characterizing the boundedness of $M$ on $L^p(\mathbb{R}^n,w dx)$, $p>1$ is the Muckenhoupt $A_p$ class defined in $\mathbb{R}^n$ by
$$\label{eq:A_p-Rn}
[w]_{A_p(\mathbb{R}^n,dx)}:=\sup_{Q} \left(\avgint_Q w\,d\mu\right) \left(\avgint_Q w^{1-p'}\,d\mu\right)^{p-1} <\infty.$$
Here $p'$ denotes the conjugate exponent of $p$ defined by the condition $\frac{1}{p}+\frac{1}{p'}=1$. In the case of LCA groups the analogue of is obtained by replacing the cubes by base sets. More precisely, a weight $w$ is an $A_p=A_p(G,d\mu)$ weight if $$\label{eq:A_p}
[w]_{A_p}:=\sup_{U\in\mathcal{B}} \left(\avgint_U w\,d\mu\right) \left(\avgint_U w^{1-p'}\,d\mu\right)^{p-1} <\infty.$$
The limiting case of , when $p=1$, defines the class $A_1$; that is, the set of weights $w$ such that $$[w]_{ A_1 }:=\sup_{U\in \mathcal B}\bigg(\avgint_U w\,d\mu \bigg) \operatorname*{ess\,sup}_{U} (w^{-1})<+\infty,$$ which is equivalent to $w$ having the property $$Mw(x)\le [w]_{A_1}w(x)\qquad \mu\text{-a.e. } x \in G.$$ As in the usual setting of $\mathbb{R}^n$ we will also often refer to $\sigma:=w^{1-p'}$ as the *dual weight* for $w$. It is easy to verify that $w\in A_p$ if and only if $\sigma\in A_{p'}$.
The family of $A_p$ classes is increasing and this motivates the definition of the larger class $A_\infty$ as the union $A_\infty=\bigcup_{p\ge 1} A_p$. There are many characterizations of the class $A_\infty$ (see [@DMRO-Ainfty] or the more classical reference [@gra04]). Some of them are given in terms of the finiteness of some $A_\infty$ constant suitably defined. The classical definition consists in taking the limit on the $A_p$ constant as $p$ goes to infinity, namely: $$\label{eq:Ainfty-exp}
(w)_{A_\infty}:=\sup_{U\in\mathcal{B}}\left (\avgint w d\mu \right )\exp{\avgint_U \log(w^{-1}) d\mu}.$$ However, the modern tendency is to consider the so-called Fujii-Wilson constant implicitly introduced by Fujii in [@Fujii], and later rediscovered by Wilson, [@Wilson:87; @Wilson-LNM], and here we choose to follow this approach by defining the $A_\infty$ constant as $$\label{eq:Ainfty-definition}
[w]_{A_\infty}:=\sup_{U\in \mathcal B}\frac{1}{w(U)}\int_U M(w\chi_U)\ d\mu,$$ where $w(U)=\int_Uw\ d\mu$.
Our contribution {#subsec:results}
----------------
As we have already seen, there is a proper -and natural- way to define the $A_p$ and $A_\infty$ classes on an LCA groups having covering families. In contrast with the case $p<\infty$, it is not immediate that the weight $w$ belongs to $A_\infty$ when any of constants defined on and is finite. In fact, a weight $w$ is in $A_\infty$ (that is, in some $A_p$) if and only if it satisfies the reverse Hölder inequality which says that $$\left(\avgint_U w^{r}\ d\mu\right)^{1/r}\leq C \avgint_{\widehat{U}} w\ d\mu$$ for some $r>1$ and where $\widehat{U}$ is an open set defined in terms of $U$ (in the euclidean case $\widehat{U}=U$ and in the case of spaces of homogeneous type, it is a dilation of $U$). This is a very well known result in the qualitative case, but it was proved recently in [@HPR1] a sharp quantitative result in terms of $[w]_{A_\infty}$ in the context of spaces of homogeneous type.
Our first result is the following version of the RHI. Note that, as in [@HPR1], we are able to precisely describe the exponent $r$ in term of the constant $[w]_{A_\infty}$.
\[thm:SharpRHI\]
Let $w\in A_\infty$. Define the exponent $r(w)$ as $$r(w)=1+\frac{1}{4D^{10}[w]_{A_\infty}-1},$$ where $D$ is the doubling constant. Then, for a fixed $U=x_0+U_{i_0}\in\mathcal{B}$, we have that the following inequality holds $$\label{eq:RHI}
\left(\avgint_U w^{r(w)}\ d\mu\right)^{1/r(w)}\leq 2D^2 \avgint_{\widehat{U}} w\ d\mu,$$ where $\widehat{U}$ is the union of the base sets $\{x+U_i:\, x\in U,\,\,i\leq i_0\}$.
Once we have proven such RHI, we are able to provide a quantitative open property for $A_p$ classes. It is very well known that the $A_p$ classes are open in the sense that if $w\in A_p$ for some $p>1$, then $w$ also belongs to some $A_{p-\varepsilon}$ for some $\varepsilon>0$. But the best possible $\varepsilon$ in this property is not completely characterized. Another related interesting and still open question (even in the euclidean setting) is to determine, given a weight $w\in A_\infty$, the smallest $p>1$ such that $w\in A_p$. There are some estimates in [@HagPar-Embeddings] but there is no proof of its sharpness.
Here we will deduce from Theorem \[thm:SharpRHI\] an open property for $A_p$ classes in LCA groups with some control on the constants. More precisely, given $w\in A_p$ for $1<p<\infty$ we will obtain that $w\in A_{p-\varepsilon}$ for $\varepsilon=\frac{p-1}{C[\sigma]_{A_\infty}}$ with $C=4D^{10}$. Further, $[w]_{A_{p-\varepsilon}}\le 2^{p-1}D^{4p-2}[w]_{A_p}$ (se Lemma \[lem:OpenProperty\]).
In a recent article [@Sauer] Sauer proved a weighted bound for the maximal function for LCA groups following Lerner’s approach. Additionally, it is asked there if is it possible to obtain the sharp result from Buckley in this general setting. In our main theorem we answer this question by the affirmative and moreover, we provide a better mixed bound. By a mixed bound we understand a bound that depends on $[w]_{A_p}$ and $[w]_{A_\infty}$ of the form $\varphi([w]_{A_p}[w]_{A_\infty})$ where $\varphi$ is some nonegative function, typically a power function. Since we always have that $[w]_{A_\infty}\le [w]_{A_p}$, usually mixed type bounds are sharper than estimates involving only the $A_p$ constant.
A result in this direction was obtained in [@HPR1] where the authors proved an improvement of Buckley’s result in terms of mixed bounds for spaces of homogeneous type, namely $$\|M\|_{L^p_w\to L^p_w}\le C\left ([w]_{A_p}[\sigma]_{A_\infty}\right )^{\frac{1}{p}}\le C [w]_{A_p}^{\frac{1}{p-1}}.$$
Our main result provides an extension of the above estimate to the context of LCA groups and we will obtain it as a consequence of the RHI and the open property. We remark here that the lack of geometry in this setting constitutes a major obstacle to overcome.
\[thm:CorSharpBuckley\] Let $M$ be the Hardy-Littlewood maximal function defined in and let $1<p<\infty$. Then there is a structural constant $C>0$ such that $$\label{eq:improvedBuckley}
\|Mf\|_{L_w^p(G)} \le C \left ([w]_{A_p}[\sigma]_{A_\infty}\right )^\frac{1}{p}.$$ In particular, $$\label{eq:Buckley2}
\|M\|_{L^p(w)} \leq C [w]_{A_p}^{\frac{1}{p-1}}.$$
Outline
-------
The paper is organized as follows. In Section \[sec:prelim\] we give some preliminary results. We prove the engulfing property in this context that will be used several times along the paper. We also define the local maximal function, prove a crucial covering lemma (Lemma \[lem:CZ\]) and show a localization property of the local maximal function. In Section \[sec:proofs\] we give the proofs of the results described in Section \[subsec:results\]
Preliminaries {#sec:prelim}
=============
In this section we provide some properties of covering families that we will use. Furthermore, we will introduce a local maximal function which will be crucial to prove the RHI.
As we already mentioned in the introduction the family of dyadic cubes of sidelength $2^i$ or the euclidean balls $B(x,2^i)$ for $i\in \mathbb{Z}$ are covering families for $G=\mathbb{R^d}$. Other examples are presented below.
\(1) When $G=\mathbb{T}=\{e^{2\pi i t}:\, t\in[-\frac1{2},\frac1{2}) \}$ with the Haar measure consider $U_k\subseteq G$ defined as $U_0=\mathbb{T}$ and for $k\in\mathbb{N}$, $U_k=\{0\}$ and $U_{-k}=\{e^{2\pi i t}:\, |t|<\frac1{2^{k+1}} \}$. Then, $\{U_k\}_{k\in\mathbb{Z}}$ is a covering family for $\mathbb{T}$ with $\theta(k)=k+1$ and $D=2$.
\(2) For $G=\mathbb{Z}$ take $U_i=\{k\in\mathbb{Z}:\, |k|\leq 2^{i-1}\}$ for $i\geq 1$ and $U_i=\{0\}$ otherwise. Then $\{U_i\}_{i\in\mathbb{Z}}$ is a covering family for $\mathbb{Z}$ with $\theta(i)=i+1$ and $D=2$.
\(3) Let $G$ be an LCA group with Haar measure $\mu$ and let $H$ be a compact and open subgroup of $G$ with $\mu(H)=1$. Consider an expansive automorphism $A:G\to G$ with respect to $H$, which means that $H\subsetneq AH$ and $\bigcap_{i<0}A^iH=\{0\}$. If additionally, $G=\bigcup_{i\in\mathbb{Z}}A^iH$, then $\{A^iH\}_{i\in\mathbb{Z}}$ is a covering family for $G$. Indeed. Since $H\subsetneq AH$ and $H$ is a group, $A^iH-A^iH=A^iH\subseteq A^{i+1}H$ so $\theta(i)=i+1$. To see that the doubling property is satisfied, note that $\mu_A$ defined as $\mu_A(B):=\mu(AB)$ for $B\subseteq G$ a Borel set, is a Haar measure on $G$. Thus, there is a positive number $\alpha$ such that $\mu_A=\alpha\mu$. The constant $\alpha$ is the so-called [*modulus of $A$*]{} and is denoted by $\alpha=|A|$. Then, $\mu(A^{i+1}H)=\mu_A(A^{i}H)=|A|\mu(A^{i}H)$ for $i\in\mathbb{Z}$. Observe that $G/H$ is discrete and $AH/H$ is finite, so $AH$ is the union of finitely many cosets of the quotient $G/H$, say $\{H+s_j\}_{j=1}^r$. Therefore, $|A|=|A|\mu(H)=\mu(AH)=r$ and since $H\subsetneq AH$, $r\geq2$. Thus we can take $D=|A|\geq 2$. A structure of this type is considered in [@BB04] for constructing wavelets on LCA groups with open and compact subgroups.\
For a concrete example of this situation, consider the $p$-adic group $G=\mathbb{Q}_p$ where $p\geq 2$ is a prime number consisting of all formal Laurent series in $p$ with coefficients $\{0, 1, . . . , p-1\}$, that is, $$\mathbb{Q}_p=\left\{\sum_{n\geq n_0} a_np^n:\, n_0\in\mathbb{Z},\,\, a_n\in \{0, 1, . . . , p-1\}\right\}.$$ As a compact and open subgroup we can consider $H=\mathbb{Z}_p$ which is $$\mathbb{Z}_p=\left\{\sum_{n\geq 0} a_np^n:\, a_n\in \{0, 1, . . . , p-1\}\right\}.$$ Take $A:\mathbb{Q}_p\to \mathbb{Q}_p$ the automorphism defined as $A(x)=p^{-1}x$. Then, $A$ is expansive with respect to $\mathbb{Z}_p$ and it can be easily checked that $\mathbb{Q}_p=\bigcup_{i\in\mathbb{Z}}A^i\mathbb{Z}_p$. Then, $\{A^i\mathbb{Z}_p\}_{i\in\mathbb{Z}}$ is a covering family for $\mathbb{Q}_p$ and in this case, $D=|A|=p$.
Let $\{U_i\}_{i\in\mathbb{Z}}$ be a fixed covering family for $G$. From now on, we assume the sets $U_i$ to be symmetric. This is not a restriction at all because one can always consider the new family of base sets formed by the difference sets $U_i-U_i$ which increases the doubling constant from $D$ to $D^2$. We denote $2U_i:=U_i-U_i=U_i+U_i$.
Any covering family has the so-called *engulfing* property:
\[lem:engulfing-property\] Let $U,V$ be two base sets such that $U=x+U_i$ and $V=y+U_j$ with $i\le j$ and $x,y\in G$. If $U\cap V\neq\emptyset$, then $x+U_i\subset y+U_{\theta^2(j)}$.
There are two point $u_i\in U_i$ and $u_j\in U_j$ such that $x+u_i=y+u_j$. Then $x=y+u_i-u_j\in y+ U_j-U_j\subset y+U_{\theta(j)}$ and therefore $x+U_i\subset y+U_{\theta(j)}+U_{\theta(j)}\subset y+U_{\theta^2(j)}$ (recall that we assume that the base sets are symmetric).
\[rem:jota\] For a given $V\in \mathcal{B}$, where $\mathcal{B}$ is the base of $G$ defined as in we will denote by $j(V)\in\mathbb{Z}$ the maximum integer such that $V=x+U_{j(V)}$ for some $x\in G$. To see that such a number exists, let us define $N(V)=\{j\in \mathbb{Z}: \exists \,x\in G , V=x+U_j\}$ and show that $\sup N(V)<\infty$. If $\sup N(V)=\infty$, we could find a sequence $\{x_n\}_{n\in \mathbb{N}}$ of points in $G$ and a sequence of integer indices $\{i_n\}_{n\in \mathbb{N}}$ such that $i_n\to\infty$ as $n\to\infty$ and $$V=x_n+U_{i_n} \qquad \text{ for all } n\in\mathbb{N}.$$
By compactness of $\overline{V}$ we can assume (relabelling) that the sequence converges to some $x\in G$, which we can assume to be the origin. Now we claim that, for any $j\in \mathbb{N}$, there is some $m\in \mathbb{N}$ such that $U_j\subset x_m+U_{i_m}$ and from this fact would follow that $\mu(V)=\infty$, but this implies that $\infty =\mu(V)\le \mu(\overline{V})<\infty$ whis is a contradiction. To verify the claim, fix $U_j$ and choose $n_0$ such that $x_n\in U_j$ and $i_n\geq j$ for all $n\ge n_0$. Then we have that $$U_j\cap x_n + U_{i_n}\neq \emptyset$$ for all $n\ge n_0$. Furthermore, the above still holds if we replace $x_n$ by any $x_m$ with $m\ge n \ge n_0$ since $x_m\in U_j$ and $x_m\in x_m + U_{i_n}$. Therefore by the engulfing property (see e.g. Lemma \[lem:engulfing-property\]) we obtain that $$U_j\subset x_m + U_{\theta^2(i_n)} \subset x_m + U_{i_m}$$ for any $m$ such that $i_m\ge \theta^2(i_n)$.
In order to introduce the local maximal function, we first define a local base for a fixed base set $U$.
\[def:localBase\] Let $U\in\mathcal{B}$ be a fixed base set and let $k:=j(U)$. The local base $\mathcal{B}_U$ is defined as $$\label{eq:localBase}
\mathcal{B}_U:=\left \{y+U_j: y\in U, j\le k\right \}.$$ We also defined the [*enlarged*]{} set $\widehat{U}$ by the formula $$\label{eq:U-hat}
\widehat{U}:=\bigcup_{V\in \mathcal{B}_U}V.$$
\[lem:engulfing\] Let $U=x+U_{k}$ be a fixed base set in $\mathcal{B}$ and set $k=j(U)$. We then have the following geometric properties:
1. If $V\in \mathcal{B}_U$ then $V\subset x+U_{\theta(k)}$.
2. For any $z\in U$, we have that $$\label{eq:engulfing1}
\widehat{U}\subset z+U_{\theta^2(k)},$$ where $\widehat{U}$ is as in . As a consequence of this last property, we obtain $$\label{eq:engulfing2}
\mu(\widehat{U})\le \mu(z+U_{\theta^2(k)})\le D^2\mu(z+U_k)$$ for any $z\in U$. In particular, since $U=x+U_k$, $\mu(\widehat{U})\le D^2\mu(U)$.
$(a)$. Let $V=y+U_j$ with $j\le k$ and take any $z\in V$. Then $z=y+u_j$ with $u_j\in U_j\subset U_k$. Since $y\in U$ we can write $y=x+u_k$, $u_k\in U_k$. Then we have that $z=x+u_j+u_k\in x+U_k+U_k\subset x+U_{\theta(k)}$.
$(b)$. Let $V\in \mathcal{B}_U$, $V=y+U_j$ with $y\in U$, $j\le k$. By Lemma \[lem:engulfing-property\], since $V\cap U\neq \emptyset$, we have that $V\subset x+U_{\theta(k)}$. Take any $z\in U$, $z=x+u_k$, $u_k\in U_k$. Then $$V\subset x+U_{\theta(k)}= z-u_k+U_{\theta(k)}\subset z-U_k+U_{\theta(k)}\subset z-U_{\theta(k)}+U_{\theta(k)}\subset z+U_{\theta^2(k)}.$$
We now define the local maximal function as follows $$\label{eq:localMaximal}
M_Uf(y):=\sup_{y\in V\in\mathcal{B}_U}\avgint_V |f(z)| \ d\mu(z)$$ for any $y\in\widehat{U}$ and and $M_Uf(y)=0$ otherwise.
\[rem:lebesgue\] $(a)$ In [@HR-vol2 Theorem 44.18], it is proven a version of the Lebesgue Differentiation Theorem with respect to the Haar measure for LCA groups having a $D'$-sequences (cf. [@HR-vol2 Definition 44.10]). A careful reading of the proof of [@HR-vol2 Theorem 44.18] reveals that the result is still true with the obvious changes for measures which are not translation invariant. Thus, since a covering family is in particular a $D'$-sequence, we have that the Lebesgue Differentiation Theorem holds in our context.
$(b)$ As a consequence of the Lebesgue Differentiation Theorem, we have the elementary but important property of the local maximal function:\
$f(x)\le M_Uf(x)$ $\mu$-almost everywhere $x\in U$.
Consider now, for a fixed $U\in\mathcal{B}$, the level set for the local maximal function acting on a weight $w$ at scale $\lambda>0$: $$\label{eq:Omegalambda}
\Omega_\lambda:=\left \{x\in \widehat{U}: M_Uw(x)>\lambda\right \}.$$
A key instrument will be a Calderón-Zygmund decomposition of $\Omega_\lambda$. We will obtain it by using an adapted version of a covering lemma from [@EG77 Lemma 2.2.1]. Although the proof follows standard arguments, we include it here for completeness. When $w$ be a nonnegative and locally integrable function on $G$ and $V\subseteq G$ is relatively compact we denote the average of $w$ on $V$ as $w_V$; that is, $w_V=\avgint_{V}w\ d\mu$.
\[lem:CZ\] Let $U\in\mathcal{B}$ be a fixed base set in $G$ and let $w$ be a nonnegative and integrable function supported on $\widehat{U}$. For $\lambda>w_{\widehat{U}}$, define $\Omega_\lambda$ as in . If $\Omega_\lambda$ is nonempty, there exists a finite or countable index set $Q$ and a family $\{y_i+U_{{\alpha_i}}\}_{i\in Q}$ of pairwise disjoint base sets from $\mathcal{B}_U$ such that
1. The sequence $\{{\alpha_i}\}_{i\in Q}$ is decreasing.
2. $\displaystyle\bigcup_{i\in Q} y_i+U_{{\alpha_i}}\subset\Omega_\lambda\subset \displaystyle\bigcup_{i\in Q} y_i+U_{\theta^2(\alpha_i)}$.
3. For any $i\in Q$, we have that $$\lambda < \avgint_{y_i+U_{{\alpha_i}}} w\ d\mu.$$
4. Given $r>{\alpha_i}$ for some $i\in Q$, then $$\label{eq:big-average}
\avgint_{y_i+U_r}w\ d\mu \le D^2\lambda.$$
Suppose that there is no finite sequence of points in $\Omega_\lambda$ such that the conclusion holds (in that case, there is nothing to prove). For $x\in \Omega_\lambda$, define $$\alpha(x)=\max \left \{j\in\mathbb{Z}: \exists\, V=y+U_j \in \mathcal{B}_U, x\in V, \avgint_{V}w\ d\mu>\lambda\right \}.$$ Since $V=y+U_j \in \mathcal{B}_U$ implies $j\leq j(U)$, we have that $\alpha$ is well-defined. Consider now, for each $x\in \Omega_\lambda$ a base set $V_x\in \mathcal{B}_U$, $V_x:=y_x+U_{\alpha(x)}$ such that $x\in V_x$. In other words, one of the base sets in $\mathcal{B}$ containing the point $x$ where the map $\alpha$ attains its value. Observe that in particular, $\alpha(y_x)\geq\alpha(x)$.
We start by picking $x_1$ as a extremal point for $\alpha$, that is $\alpha(x_1)\ge \alpha(x)$ for all $x\in\Omega_\lambda$. Put $\alpha_1=\alpha(x_1)$ and $y_1:=y_{x_1}$ such that $V_{x_1}=y_1+U_{\alpha_1}$. Note that, since $\alpha_1\le\alpha(y_1)\le\alpha(x_1)=\alpha_1$, we also have that $\alpha(y_1)=\alpha_1$. Now suppose that we have chosen the first $n$ points $y_1,\dots, y_n$ and their respective base sets $U_{\alpha_1},\dots,U_{\alpha_n}$ such that
- the sets $V_j:= y_j+U_{\alpha_j}$, $1\le j\le n$ are pairwise disjoint,
- $\alpha_j:=\alpha(y_j)\ge \alpha(x)$ for all $x\in A_{j-1}$, where $$A_j:=\Omega_\lambda\setminus \bigcup_{\ell\le j} y_\ell +U_{\theta^2(\alpha_\ell)}\qquad 1\le j \le n.$$
Since we are assuming that this procedure never ends, we have that $A_j\neq\emptyset$ for all $1\leq j\leq n$. Therefore we can choose $ x_{n+1}\in A_n$ such that $\alpha_{n+1}:=\alpha(x_{n+1})\geq \alpha(x)$ for all $x\in A_n$. This means that there is base set $V_{n+1}:=y_{n+1}+U_{\alpha_{n+1}}$ and in particular $w_{V_{n+1}}>\lambda$ and $\alpha(y_{n+1})=\alpha_{n+1}$. Note that this construction produces a decreasing sequence $\{\alpha_n\}_{n\in \mathbb{N}}$. Let’s see that $V_{n+1}\cap V_j=\emptyset$ for all $1\le j \le n$. Supposing that this is not the case, we could find $u\in U_{\alpha_{n+1}}$ and $v\in U_{\alpha_j}$ for some $j\le n$ such that $$y_{n+1}+u=y_j+v.$$ Since $x_{n+1}\in V_{n+1}$, we have that for some $z\in U_{\alpha_{n+1}}$, $$x_{n+1}=y_{n+1}+z=y_j+v-u+z\in y_j+ U_{\alpha_j}-U_{\alpha_{n+1}}+U_{\alpha_{n+1}}.$$ Since $U_{\alpha_{n+1}}\subset U_{\alpha_{j}}$ and trivially $U_{\alpha_{j}}\subset U_{\theta(\alpha_{j})}$, we get that $$x_{n+1}\in y_j+U_{\theta^2(\alpha_{j})},$$ which is a contradiction by the choice of $x_{n+1}$.
We are left to prove that this procedure exhausts the set $\Omega_\lambda$. If not, there is a point $x\in A_n$ with $\alpha(x)\le \alpha_n$ for all $n\ge 1$. Define the set $S$ as $$S:=\{y_n: n\in\mathbb{N}\}.$$ Since $$S\subset \{z\in \Omega_\lambda: \alpha(z)\ge\alpha(x)\}\subset \widehat{U}$$ and $\widehat{U}$ is contained in some base set (see item (b) in Lemma \[lem:engulfing\]) we conclude that $S$ is relatively compact.
By monotonicity of $\alpha$, we have that $U_{\alpha_n}\subset U_{\alpha_1}$. Therefore the set $$F:=\bigcup_n (y_n+U_{\alpha_n})\subset S+U_{\alpha_1}$$ is also relatively compact and this implies that $\mu(\overline{F})<\infty$. Now consider $N\in \mathbb{Z}$ such that $\overline{S}\subset U_N$ and an integer $r>0$ such that $\theta^r(\alpha(x))\ge N$. Then we have that for any $n\in\mathbb{N}$, $y_n\in S\subset U_N\subset U_{\theta^r(\alpha(x))}$ and thus $0\in y_n+U_{\theta^r(\alpha(x))}$. Further, we obtain that $$U_N=0+U_N\subset y_n+U_{\theta^r(\alpha(x))}+U_N\subset y_n+2U_{\theta^r(\alpha(x))}\subset y_n+U_{\theta^{r+1}(\alpha(x))}.$$ The doubling property shows that $$\mu(U_N)\le D^{r+1}\mu(y_n +U_{\alpha(x)})$$ and this implies that $$\mu(F)=\sum_n\mu(y_n+U_{\alpha_n})\ge \sum_n\mu(y_n+U_{\alpha(x)})
\ge D^{-(r+1)}\sum_n\mu(U_N)=\infty.$$ This contradicts the condition $\mu(\overline{F})<\infty$ and we conclude with the proof of items (a), (b) and (c) of the lemma.
We prove now item (d). Towards to control the average on $y_i+ U_r$ we consider two cases: first we consider $r\leq k:=j(U)$. Then $y_i+ U_{r}\in \mathcal{B}_U$ and by maximality we have $\displaystyle \avgint_{y_i+ U_r} w \ d\mu\leq\lambda$. Indeed, if not we would have that $\alpha_i=\alpha(y_i)\geq r > \alpha_i$. Second, in case $r> k$, we have that $\theta^2(r)> \theta^2(k)$ and thus, by Lemma \[lem:engulfing\], $y_i+U_{\theta^2(r)}\supset
y_i+U_{\theta^2(k)}\supset\widehat{U}$. Therefore, since $w=0$ a.e $\widehat{U}^c$, we have $$\avgint_{y_i+ U_r }w \ d\mu\leq \frac{\mu(\widehat{U})}{\mu(y_i+ U_r)}\avgint_{\widehat{U}}w \ d\mu\leq D^2\lambda.$$ The lemma is now completely proven.
Now we present a localization argument for the local maximal function $M_U$. The idea is better understood when considering the usual dyadic maximal function $M^d_Q$ localized on a cube $Q$ in $\mathbb{R}^n$. Suppose that the level set $\Omega_\lambda=\{x\in Q: M^d_Qw(x)>\lambda\}$ for $\lambda>w_Q$ is decomposed into dyadic subcubes of $Q$ such that $Q=\bigcup_i Q_i$ and the cubes $Q_i$ are maximal with respect to the condition $w_{Q_i}>\lambda$. Then the conclusion is that for any $x\in Q_i$ the equality $M^d_Qw(x)=M^d_Q(w\chi_{Q_i})(x)$ holds. In this more general setting, the analogous result is contained in the following lemma which has not a direct proof as in the dyadic case.
For simplicity in the exposition, we introduce the following notation. Given a base set of the form $V=y+U_j$ we denote by $V^*$ the dilation of $V$ by $\theta$, i.e. $V^*=y+U_{\theta(j)}$. Further iterations of this operation are defined recursively, that is, $V^{**}=(V^*)^*$ and $V^{n*}$ for $n$ iterations of the dilation operation.
\[lem:local-max\] Let $U\in\mathcal{B}$ be a fixed base set and consider $w=w\chi_{\widehat{U}}$ a nonnegative and integrable function on $\widehat{U}$ where $\widehat{U}$ is as in . For $\lambda>w_{\widehat{U}}$ let $\Omega_\lambda$ defined as above and let $\{V_i\}_{i\in Q}=\{y_i+U_{\alpha_i}\}_{i\in Q}$ be the C-Z decomposition of $\Omega_\lambda$ given by Lemma \[lem:CZ\]. Then, for $L=D^6$ any $i\in Q$ and any $x\in V_i^{**}\cap \Omega_{L\lambda}$ we have $$\label{eq:local-max}
M_Uw(x)\le M_U(w\chi_{V_i^{4*}})(x).$$
Let $x\in V^{**}_i\cap\Omega_{L\lambda}$. Then there exists $V\in\mathcal{B}_U$, $V=y+U_j$, with $y\in U$ and $j\le j(U)$ such that $x\in V$ and $w_V>L\lambda$. We claim that $j\le \theta^2(\alpha_i)$. To see that this is in fact true, suppose towards a contradiction, that $j>\theta^2(\alpha_i)$. Then, $V\subset y_i+U_{\theta^2(j)}$. Indeed, if $z\in V$ then $z=y+w$ with $w\in U_j$. On the other hand, since $x\in V^{**}_i\cap V$, $x=y_i+u=y+v$ with $u\in U_{\theta^2(\alpha_i)}$ and $v\in U_j$. Then $$z=y+v-v+w=x -v+w=y_i+ u-v+w.$$ Since $U_{\theta^2(\alpha_i)} \subset U_j$, we obtain that $z\in y_i+ U_j+U_{\theta(j)}\subset y_i+ U_{\theta^2(j)}$. As a consequence, $$\avgint_V w\ d\mu\le \frac{\mu(y_i+ U_{\theta^2(j)})}{\mu(V)}\avgint_{y_i+ U_{\theta^2(j)}}w \ d\mu.$$ We note that since $\theta^2(\alpha_i)<j$, $x\in V\cap V^{**}_i\subset V\cap(y_i+U_j)$ and then, by the engulfing property we have that $y_i+U_j\subset y+U_{\theta^2(j)}$. Thus, using the doubling property of the measure $\mu$ we obtain $$\frac{\mu(y_i+ U_{\theta^2(j)})}{\mu(y+U_j)}\leq D^2 \frac{\mu(y_i+ U_j)}{\mu(y+U_j)}
\leq D^2\frac{\mu(y+ U_{\theta^2(j)})}{\mu(y+U_j)}\leq D^4.$$
Furthermore, since $\theta^2(j)\geq j>\theta^2(\alpha_i)\geq\alpha_i$, by item (4) in Lemma we have that $$\avgint_{y_i+ U_{\theta^2(j)}}w \ d\mu\leq
D^2\lambda$$ and we can conclude that $$L\lambda<\avgint_V w\ d\mu\le D^6\lambda=L\lambda,$$ which gives a contradiction. Hence, the claim $j\le \theta^2(\alpha_i)$ holds.
Now, using Lemma \[lem:engulfing-property\] we have that $V\subset V^{4*}_i$ and then $$\avgint_Vw\ d\mu=\avgint_Vw\chi_{V^{4*}_i}\ d\mu\le M(w\chi_{V^{4*}_i})(x)$$ which proves inequality .
Proof of the main results {#sec:proofs}
=========================
We present here the proof of Theorem \[thm:SharpRHI\].
Step 1. We start with the following estimate for the local maximal function. Let $U=x_0+U_k$ be a fixed base set. We claim that, for $\varepsilon=\frac{1}{4D^{10}[w]_{A_\infty}-1}$, we have that $$\label{eq:Maximal-vs-w}
\avgint_{\widehat U} (M_Uw)^{1+\varepsilon} \ d\mu \le 2 [w]_{A_\infty}\left (\avgint_{\widehat U} w \ d\mu\right )^{1+\varepsilon}.$$ Recall that we may assume that the weight $w$ is supported on $\widehat U$. Let $\Omega_\lambda$ defined as in . We write the norm using the layer cake formula as follows $$\begin{aligned}
\int_{\widehat U} (M_Uw)^{1+\varepsilon} \ d\mu & = & \int_0^{\infty} \varepsilon \lambda^{\varepsilon-1}M_Uw(\Omega_\lambda)\ d\lambda\\
& = & \int_0^{w_{\widehat U}} \varepsilon \lambda^{\varepsilon-1}M_Uw(\Omega_\lambda)\ d\lambda
+ \int_{w_{\widehat U}}^\infty \varepsilon \lambda^{\varepsilon-1}M_Uw(\Omega_\lambda)\ d\lambda\\
& = & I + II.\end{aligned}$$
The first term is easily controlled by using the $A_\infty$ constant of $w$ (see ): $$\begin{aligned}
I\le M_Uw(\widehat U)w_{\widehat U}^\varepsilon & = & w_{\widehat U}^\varepsilon\int_{\widehat U}M_Uw \ d\mu \\
& \le & w_{\widehat U}^\varepsilon\int_{y+U_{\theta^2(k)}}M_U(w\chi_{y+U_{\theta^2(k)}}) \ d\mu\\
& \le & w_{\widehat U}^\varepsilon [w]_{A_\infty}w(y+U_{\theta^2(k)})\\
& = & w_{\widehat U}^\varepsilon [w]_{A_\infty}w(\widehat{U})\end{aligned}$$ where $y\in U$ and we used Lemma \[lem:engulfing\] and the definition of $[w]_{A_\infty}$.
Now, for each $\lambda>w_{\widehat U}$ we consider $\{V_i\}_{i\in Q}$ the C-Z decomposition of $\Omega_\lambda$ from Lemma \[lem:CZ\] to control $II$. We have that $$\begin{aligned}
M_Uw(\Omega_\lambda)\le \sum_i M_Uw(V_i^{**}).\end{aligned}$$ For any $i\in Q$ we write $V^{**}_i=V_1\cup V_2$ with $V_1:=V^{**}_i\cap\Omega_{L\lambda}$ and $V_2:=V^{**}_i\setminus\Omega_{L\lambda}$ where $L=D^6$. Thus, by Lemma \[lem:local-max\] and the $A_{\infty}$ property we have $$\begin{aligned}
M_Uw(V_i^{**}) &=& \int_{V_1}M_Uw \ d\mu + \int_{V_2}M_Uw \ d\mu\\
& \le & \int_{V_1}M_U(w\chi_{V^{4*}_i})(x) \ d\mu + L\lambda\mu(V_2)\\
& \le & [w]_{A_{\infty}} w(V^{4*}_i) + L\lambda\mu(V^{4*}_i) = \left([w]_{A_{\infty}} w_{V^{4*}_i}+ L\lambda\right)\mu(V^{4*}_i)\\\\
& \le & \left([w]_{A_{\infty}} \lambda D^2+ L\lambda\right)D^4\mu(V_i)\le 2[w]_{A_{\infty}} \lambda D^{10} \mu(V_i),\end{aligned}$$ where in the last inequality we have used and the doubling property of $\mu$. This gives $$\begin{aligned}
M_Uw(\Omega_\lambda) & \le & \sum_i M_Uw(V_i^{**})\le 2[w]_{A_{\infty}} \lambda D^{10}\sum_i \mu(V_i)\\
& \le & 2[w]_{A_{\infty}} \lambda D^{10}\mu(\Omega_\lambda).\end{aligned}$$ Thus, $$\begin{aligned}
II & \le & 2[w]_{A_{\infty}} D^{10} \int_0^\infty \varepsilon \lambda^\varepsilon\mu(\Omega_\lambda)\ d\lambda\\
& = & 2[w]_{A_{\infty}} D^{10}\frac{\varepsilon}{\varepsilon+1}\int_{\widehat U}M_Uw^{1+\varepsilon}\, d\mu.\end{aligned}$$ Therefore, gathering all the estimations and averaging over $\widehat U$, we have that $$\left(1- 2[w]_{A_{\infty}} D^{10} \frac{\varepsilon}{\varepsilon+1}\right)\avgint_{\widehat U}M_Uw^{1+\varepsilon}\, d\mu \le w_{\widehat U}^{1+\varepsilon}.$$ Choosing $\varepsilon\le \frac{1}{4[w]_{A_{\infty}} D^{10}-1}$ we get that $1- 2[w]_{A_{\infty}} D^{10} \frac{\varepsilon}{\varepsilon+1}\ge \frac{1}{2}$ and we obtain the desired estimate .
Step 2. Now we proceed to prove the main estimate . By Remark \[rem:lebesgue\] we have that $w(x)\le M_Uw(x)$ holds on $U$. Then we obtain $$\int_U w^{1+\varepsilon}\ d\mu\le \int_{U} (M_Uw)^\varepsilon w \ d\mu\le \int_{\widehat{U}} (M_Uw)^\varepsilon w \ d\mu.$$ Once again we use the layer cake formula combined with the C-Z decomposition of $\Omega_\lambda$ and proceeding in a similar way as above we obtain $$\begin{aligned}
\int_{\widehat U} (M_Uw)^{\varepsilon} w \ d\mu & = & \int_0^{\infty} \varepsilon \lambda^{\varepsilon-1}w(\Omega_\lambda)\ d\lambda\\
& = & \int_0^{w_{\widehat U}} \varepsilon \lambda^{\varepsilon-1}w(\Omega_\lambda)\ d\lambda
+ \int_{w_{\widehat U}}^\infty \varepsilon \lambda^{\varepsilon-1}w(\Omega_\lambda)\ d\lambda\\
& \le & w(\widehat U)w_{\widehat U}^\varepsilon + \int_{w_{\widehat U}}^\infty \varepsilon \lambda^{\varepsilon-1}\sum_i w(V_i^{**})\ d\lambda\\
& \le & w(\widehat U)w_{\widehat U}^\varepsilon + D^2\int_{w_{\widehat U}}^\infty \varepsilon \lambda^{\varepsilon}\sum_i \mu(V_i^{**})\ d\lambda\\
& \le & w(\widehat U)w_{\widehat U}^\varepsilon + D^4\int_{w_{\widehat U}}^\infty \varepsilon \lambda^{\varepsilon}\sum_i \mu(V_i)\ d\lambda\\
& \le & w(\widehat U)w_{\widehat U}^\varepsilon + D^4\int_{0}^{\infty}\varepsilon \lambda^{\varepsilon} \mu(\Omega_\lambda)\ d\lambda\\
& \le & w(\widehat U)w_{\widehat U}^\varepsilon + \frac{D^4\varepsilon}{\varepsilon+1}\int_{\widehat U}(M_Uw)^{1+\varepsilon}\ d\mu\\\end{aligned}$$ Therefore, averaging over $U$, using that $\mu(\widehat{U})\le D^2 \mu(U)$ and we have $$\avgint_U w^{1+\varepsilon}\ d\mu\le D^2 w_{\widehat U}^{\varepsilon+1} + \frac{2D^6\varepsilon[w]_{A_\infty}}{\varepsilon+1}\left(\avgint_{\widehat U}w\ d\mu\right)^{1+\varepsilon}.$$ By our previous choice of $\varepsilon$, $\frac{2D^6\varepsilon[w]_{A_\infty}}{\varepsilon+1}\le \frac{2D^{10}\varepsilon[w]_{A_\infty}}{\varepsilon+1}\le \frac{1}{2}$ and we conclude that $$\avgint_U w^{1+\varepsilon}\ d\mu \le 2D^2 \left(\avgint_{\widehat U}w\ d\mu\right)^{1+\varepsilon}.$$
We present now some classical applications of the RHI to weighted norm inequalities for maximal functions. One crucial property of $A_p$ classes is the well known open condition. In the next lemma we provide a quantitative version of it.
\[lem:OpenProperty\] For $1<p<\infty$, let $w\in A_p$ . Then, for $\varepsilon=\frac{p-1}{C[\sigma]_{A_\infty}}$ with $C=4D^{10}$ and $\sigma=w^{1-p'}$, we have that $w\in A_{p-\varepsilon}$. Further, $$[w]_{A_{p-\varepsilon}}\le 2^{p-1}D^{4p-2}[w]_{A_p}.$$
Let $w\in A_p$. The $A_{p-\varepsilon}$ condition for $w$ takes the form $$\sup_{U\in \mathcal{B}}\left (\avgint_U w\ d\mu\right )\left (\avgint_U w^{1-{(p-\varepsilon)'}}\ d\mu\right )^{p-\varepsilon-1}<\infty.$$ Recall that the dual weight of $w$, $\sigma=w^{1-p'}$ is also in $A_\infty$. Therefore it satisfies a RHI with exponent $r(\sigma)$ given by Theorem \[thm:SharpRHI\]. Choose $\varepsilon$ such that $1-(p-\varepsilon)'=(1-p')r(\sigma) $, namely $\varepsilon=\frac{p-1}{r(\sigma)'}$ which is equivalent to the condition $r(\sigma)=\frac{p-1}{p-\varepsilon-1}$. Then we obtain $$\begin{aligned}
\left (\avgint_U w^{1-(p-\varepsilon)'\ d\mu}\right )^{p-\varepsilon-1} & = & \left (\avgint_U \sigma^{(1-p')r(\sigma)}\ d\mu\right )^{\frac{p-1}{r(\sigma)}}\\
&\le & \left (2D^2 \avgint_{\widehat{U}} \sigma\ d\mu\right)^{p-1},\end{aligned}$$ for any $U\in \mathcal{B}$. Now, for $U=x+U_k \in \mathcal{B}$, recall that $U^{**}=x+U_{\theta^2(k)}$ and that $\widehat{U}\subset U^{**}$. Then we have that $$\begin{aligned}
\left (\avgint_U w\ d\mu\right )\left (\avgint_U w^{1-{(p-\varepsilon)'}}\ d\mu\right )^{p-\varepsilon-1}
& \le & C \left (\avgint_{U^{**}} w\ d\mu\right )\left (\avgint_{U^{**}} \sigma \ d\mu\right )^{p-1}\end{aligned}$$ with $C=2^{p-1}D^{4p-2}$. We conclude that $$[w]_{A_{p-\varepsilon}}\le 2^{p-1}D^{4p-2} [w]_{A_p}.$$
In what follows we will need the fact that the maximal function $M$ maps $L^{q,\infty}_w(G)$ to itself with operator norm bounded by $C[w]^{\frac{1}{q}}_{A_q}$ for some $C>0$. Without presenting any details on weak norms and Lorentz spaces, we include here a quantitative estimate on the size of level sets of the maximal function.
\[lem:weak\] Let $1\le q<\infty$ and let $M$ the maximal function defined in . Then, for any $f\in L_w^q(G)$, we have that $$\label{eq:weakM}
\sup_{\lambda>0}\lambda^q w(\{x\in G: Mf(x)>\lambda\})\le D^{2q} [w]_{A_q}\|f\|^q_{L^q_w}.$$
For any locally integrable function $f$ and any $\lambda>0$ let $\Omega_\lambda$ be the level set $\Omega_\lambda:=\{x\in G: Mf(x)>\lambda\}$. We also define some sort of *truncated* maximal operator as follows: for any $K\in \mathbb{Z}$, let $M_K$ the averaging operator given by $$\label{eq:truncated-maximal}
M_Kf(x)=\sup_{V\in\mathcal{B}_K(x)}\avgint_V |f(z)| d\mu,$$ where the supremum is taken over the subfamily $\mathcal{B}_K$ of $\mathcal{B}$ consisting of all base sets of the form $y+U_i$ with $y\in G$ and $i\le K$ containing the point $x$, i.e.: $$\label{eq:BK(x)}
\mathcal{B}_K(x):=\{V=y+U_i: x\in V, i\le K\}.$$ For each $K$ we consider the corresponding level set $\Omega^K_\lambda:=\{x\in G: M_Kf(x)>\lambda\}$. We clearly have that the family $\{\Omega^K_\lambda\}$ is increasing in $K$ and also $\Omega_\lambda=\bigcup_K\Omega^K_\lambda$. We therefore may compute the value of $w(\Omega_\lambda)$ as the limit of $w(\Omega^K_\lambda)$. In addition, we recall that the group $G$ is $\sigma$-compact since $G=\bigcup_{r\in \mathbb{Z}} \overline{U_r}$. We will use again a limiting argument to compute $w(\Omega^K_\lambda)$ as the limit of $w(\Omega_\lambda^K\cap U_r)$ with $r\to +\infty$.
Now for $K\in \mathbb{Z}$ fixed, choose $r\in\mathbb{Z}$ such that $r\ge K$. A simple Vitali’s covering lemma can be applied now to $\Omega_\lambda^K\cap U_r$. We want to select a countable subfamily of disjoint base sets whose dilates cover $\Omega^K_\lambda\cap U_r$. More precisely, suppose that the set $\Omega^K_\lambda\cap U_r$ is nonempty. For each $x\in\Omega^K_\lambda\cap U_r$ there exists a base set $V_x$ of the form $V_x=y_x+U_{i_x}$ such that $$\label{eq:simple-vitali-1}
\avgint_{V_x}|f(z)|\ d\mu>\lambda.$$ Since we have that $i_x\le K$ for all $x\in \Omega^K_\lambda\cap U_r$, there is some $i_{1}=\max\{i_x\}$. We start the recursive selection procedure by picking one of this largest base sets as $V_1=y_1+U_{i_1}$. Now suppose that the first $V_1,V_2,\dots,V_k$ sets have been selected. We pick $V_{k+1}$ verifying that $V_{k+1}=y_{k+1}+U_{i_{k+1}}$ where $i_{k+1}=\max\{i_x: y_x+U_{i_x}\cap V_j=\emptyset, j=1,\dots, k\}$.
This process generates a sequence of disjoint base sets $\{V_k\}$. We note that the index sequence $\{i_k\}$ goes to $-\infty$ as $k$ goes to infinity. If not, since it is decreasing, there would be some $i_0=i_k$ for all $k\ge k_0$. Then we have that $ V_k\cap U_r\neq \emptyset$ and $i_k\le K\le r$ and by the engulfing property, $V_k\subset U_r^{**}$ for all $k\ge k_0$. In particular, the set $S=\{y_k: k\ge k_0\}\subset U_r^{**}$ is relatively compact. Then, considering the set $$F=\bigcup_{k\ge k_0} V_k\subset S+U_{i_0}$$ and proceeding as in Lemma \[lem:CZ\] we get a contradiction.
We claim now that $$\label{eq:simple-vitali-2}
\Omega_\lambda^K\cap U_r\subset \bigcup_{k\in \mathbb{N}} V_k^{**}.$$ To verify it, consider some $x\in \Omega^K_\lambda\cap U_r$ and the corresponding $V_x=y_x+U_{i_x}$. Suppose first that $V_x$ intersects some $V_k$. Let $k_0$ the smallest $k\in\mathbb{N}$ such that $V_x\cap V_k\neq\emptyset$. Then we have that $i_x\le i_{k_0}$, since $i_{k_0}$ was selected as the largest index among all the sets $V_x$ disjoint from $V_1,\dots, V_{k_0-1}$ (and by hypothesis $V_x$ is one of them). Then the engulfing property yields $$V_x=y_x+U_{i_x}\subset y_{k_0}+U_{\theta^2(i_{k_0})}=V_{k_0}^{**}.$$ We are left to consider the case when $V_x\cap V_k=\emptyset$ for all $k\in \mathbb{N}$. But in this case, we would have that $i_x\le i_k$ for all $k$ and this is a contradiction since we saw that $i_k\to -\infty$.
Summing up, we find a countable collection of base sets $\{V_k\}_k$ such that $$\avgint_{V_k}f\ d\mu >\lambda \qquad \text{ and } \qquad \Omega^K_\lambda\cap U_r\subset \bigcup_k V_k^{**}.$$
Then we can compute $$\begin{aligned}
\lambda^q w(\Omega^K_\lambda\cap U_r)& \le & \sum_k \lambda^q w(V_k^{**})\\
& \le & \sum_k w(V_k^{**})\left (\avgint_{V_k} w^{-\frac{1}{q}}w^{\frac{1}{q}}|f|\right )^q\\
& \le & \sum_k \frac{w(V_k^{**})}{\mu(V_k)^q} \left (\int_{V_k}w^{1-q'}\ d\mu\right )^{q-1} \left (\int_{V_k} |f|^qw\ d\mu\right )\\
& \le & D^{2q}\sum_k \frac{w(V_k^{**})}{\mu(V_k^{**})^q} \left (\int_{V_k^{**}}w^{1-q'}\ d\mu\right )^{q-1} \left (\int_{V_k} |f|^qw\ d\mu\right )\\
& \le & D^{2q}[w]_{A_q}\sum_k \int_{V_k} |f|^qw\ d\mu\\
& \le & D^{2q}[w]_{A_q}\|f\|^q_{L^q_w}.\end{aligned}$$ From this estimate we conclude that $$\lambda^q w(\Omega_\lambda)\le D^{2q}[w]_{A_q}\|f\|^q_{L^q_w}$$ for any $\lambda >0$.
Now we are able to present the proof of the sharp version of Buckley’s Theorem for the maximal function $M$ on $L^p(G)$, $p>1$.
The idea is to use a sort of interpolation type argument, exploiting the sublinearity of the maximal operator $M$ and the weak type estimate for $M$ from Lemma \[lem:weak\]. For any $f\in L^p_w(G)$ and any $t>0$, define the truncation $f_t:=f\chi_{\{|f|>t\}}$. Then, an easy computation of the averages defining $M$ gives that $$\{x\in G: Mf(x)>2t\}\subset \{x\in G: Mf_t(x)>t\}.$$
Now we compute the $L^p_w$ norm as follows $$\begin{aligned}
\|Mf\|^p_{L_w^p(G)} &=& \int_0^\infty pt^{p-1}w(\{x\in G: Mf(x)>t\}) dt\\
&=&2^p \int_0^\infty pt^{p-1}w(\{x\in G: Mf(x)>2t\}) dt\\
&\le &2^p \int_0^\infty pt^{p-1}w(\{x\in G: Mf_t(x)>t\}) dt.\end{aligned}$$
We recall the open property for Muckenhaupt weights: any $w\in A_p$ also belongs to $A_{p-\varepsilon}$ for some explicit $\varepsilon>0$ (see Lemma \[lem:OpenProperty\]). Using the estimate of Lemma \[lem:weak\] for $q=p-\varepsilon$, we obtain
$$\begin{aligned}
\|Mf\|^p_{L_w^p(G)} &\leq&2^p p D^{2(p-\varepsilon)}[w]_{A_{p-\varepsilon}}\int_0^\infty t^{\varepsilon-1}\int_G f^{p-\varepsilon}_t(x) w(x)\ d\mu dt \nonumber \\
& = & \frac{2^p p D^{2(p-\varepsilon)}[w]_{A_{p-\varepsilon}}}{\varepsilon}\int_G |f(x)|^p w\ d\mu \nonumber\\
& \leq & \frac{p2^{2p-1}D^{6p-2}[w]_{A_{p}}}{\varepsilon}\|f\|_{L^p_w(G)}^p, \label{eq:eqfinal}\end{aligned}$$
where in the last inequality we have used Lemma \[lem:OpenProperty\]. Noticing that in Lemma \[lem:OpenProperty\], $\varepsilon=\frac{p-1}{4D^{10}[\sigma]_{A_\infty}}$ we finally conclude from that $$\|Mf\|_{L_w^p(G)} \le C \left ([w]_{A_p}[\sigma]_{A_\infty}\right )^\frac{1}{p}\|f\|_{L^p_w(G)}$$ and the proof of is complete.
Finally, since $[\sigma]_{A_\infty}\leq[\sigma]_{A_{p'}}=[w]_{A_p}^{p'-1}$, follows from .
[DMRO16]{}
John J. Benedetto and Robert L. Benedetto, *A wavelet theory for local fields and related groups*, J. Geom. Anal. **14** (2004), no. 3, 423–456.
Stephen M. Buckley, *Estimates for operator norms on weighted spaces and reverse [J]{}ensen inequalities*, Trans. Amer. Math. Soc. **340** (1993), no. 1, 253–272.
Javier Duoandikoetxea, Francisco J. Mart[í]{}n-Reyes, and Sheldy Ombrosi, *On the [$A_\infty$]{} conditions for general bases*, Math. Z. **282** (2016), no. 3-4, 955–972. [MR ]{}[3473651]{}
R. E. Edwards and G. I. Gaudry, *Littlewood-[P]{}aley and multiplier theory*, Springer-Verlag, Berlin-New York, 1977, Ergebnisse der Mathematik und ihrer Grenzgebiete, Band 90. [MR ]{}[0618663 (58 \#29760)]{}
Nobuhiko Fujii, *Weighted bounded mean oscillation and singular integrals*, Math. Japon. **22** (1977/78), no. 5, 529–534.
Loukas Grafakos, *Classical and modern [F]{}ourier analysis*, Pearson Education, Inc., Upper Saddle River, NJ, 2004.
Paul Hagelstein and Ioannis Parissis, *Weighted [S]{}olyanik estimates for the [H]{}ardy-[L]{}ittlewood maximal operator and embedding of [${A}_\infty$]{} into [${A}_p$]{}*, J. Geom. Anal. **26** (2016), no. 2, 924–946.
Tuomas Hyt[ö]{}nen, Carlos P[é]{}rez, and Ezequiel Rela, *Sharp [R]{}everse [H]{}ölder property for [$A_\infty$]{} weights on spaces of homogeneous type*, J. Funct. Anal. **263** (2012), no. 12, 3883–3899.
Edwin Hewitt and Kenneth A. Ross, *Abstract harmonic analysis. [V]{}ol. [II]{}: [S]{}tructure and analysis for compact groups. [A]{}nalysis on locally compact [A]{}belian groups*, Die Grundlehren der mathematischen Wissenschaften, Band 152, Springer-Verlag, New York-Berlin, 1970. [MR ]{}[0262773 (41 \#7378)]{}
[to3em]{}, *Abstract harmonic analysis. [V]{}ol. [I]{}*, second ed., Grundlehren der Mathematischen Wissenschaften \[Fundamental Principles of Mathematical Sciences\], vol. 115, Springer-Verlag, Berlin-New York, 1979, Structure of topological groups, integration theory, group representations. [MR ]{}[551496]{}
Andrei K. Lerner, *An elementary approach to several results on the [H]{}ardy-[L]{}ittlewood maximal operator*, Proc. Amer. Math. Soc. **136** (2008), no. 8, 2829–2833.
Benjamin Muckenhoupt, *Weighted norm inequalities for the [H]{}ardy maximal function*, Trans. Amer. Math. Soc. **165** (1972), 207–226.
Walter Rudin, *Fourier analysis on groups*, Interscience Tracts in Pure and Applied Mathematics, No. 12, Interscience Publishers (a division of John Wiley and Sons), New York-London, 1962. [MR ]{}[0152834 (27 \#2808)]{}
Jonas Sauer, *An extrapolation theorem in non-[E]{}uclidean geometries and its application to partial differential equations*, J. Elliptic Parabol. Equ. **1** (2015), 403–418. [MR ]{}[3487344]{}
J. Michael Wilson, *Weighted inequalities for the dyadic square function without dyadic [$A_\infty$]{}*, Duke Math. J. **55** (1987), no. 1, 19–50.
Michael Wilson, *Weighted [L]{}ittlewood-[P]{}aley theory and exponential-square integrability*, Lecture Notes in Mathematics, vol. 1924, Springer, Berlin, 2008.
[^1]: Both authors are partially supported by grants UBACyT 20020130100403BA, CONICET-PIP 11220150100355 and PICT 2014-1480.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study the limit of first eigenfunctions of (discrete) $p$-Laplacian on a finite subset of a graph with Dirichlet boundary condition, as $p\to 1.$ We prove that up to a subsequence, they converge to a summation of characteristic functions of Cheeger cuts of the graph. We give an example to show that the limit may not be a characteristic function of a single Cheeger cut.'
address:
- 'Huabin Ge, Department of Mathematics, Beijing Jiaotong University, Beijing, 100044, China'
- 'Bobo Hua, School of Mathematical Sciences, LMNS, Fudan University, Shanghai 200433, China'
- 'Wenfeng Jiang, School of Mathematics (Zhuhai), Sun Yat-Sen University, Zhuhai, China'
author:
- Huabin Ge
- Bobo Hua
- Wenfeng Jiang
title: '**A note on limit of first eigenfunctions of $p$-Laplacian on graphs**'
---
[^1]
[^2]
[^3]
Introduction
============
The spectral theory of linear Laplacian, called Laplace-Beltrami operator, on a domain of an Euclidean space or a Riemannian manifold is extensively studied in the literature, see e.g. [@CH53]. The $p$-Laplacians are nonlinear generalizations of the linear Laplacian, which corresponds to the case $p=2$. These are nonlinear elliptic operators which possess many analogous properties as the linear Laplacian.
A graph consists of a set of vertices and a set of edges. The Laplacian on a finite graph is a finite dimensional linear operator, see e.g. [@Chung10], which emerges from the discretization of the Laplace-Beltrami operator of a manifold, the Cayley graph of a discrete group, data sciences and many others. Compared to continuous Laplacians, one advantage of discrete Laplacians is that one can calculate the eigenvalues of the Laplacian on a finite graph which are in fact eigenvalues of a finite matrix.
Cheeger [@Cheeger70] defined an isoperimetric constant, now called Cheeger constant, on a compact manifold and used it to estimate the first non-trivial eigenvalue of the Laplacian, see also [@K]. These were generalized to graphs by Alon-Milman [@Alon-Milman] and Dodziuk [@Dodziuk84] respectively. These estimates, called Cheeger estimates, are very useful in the spectral theory. The spectral theory for discrete $p$-Laplacians was studied by [@Yamasaki79; @Amghibech03; @Takeuchi03; @Hein-Buhler10; @Keller-Mugnolo16]. It turns out that this theory unifies these constants involved in the Cheeger estimate. The Cheeger constant of a finite graph is in fact the first non-trivial eigenvalue of $1$-Laplacian, see Chang, Hein et al. [@Chang16; @Chang-Shao-Zhang17; @Hein-Buhler10]. So that the Cheeger estimate can be regarded as the eigenvalue relation for $p$-Laplacians, $p=1$ and $p=2.$
In this paper, we study first eigenvalues and eigenfunctions of the $p$-Laplacian on a subgraph of a graph with Dirichlet boundary condition. Let $G= (V, E)$ be a simple, undirected, locally finite graph, where $V$ is the vertex set and $E$ is the edge set. Two vertices $x,y$ are called neighbors, denoted by $x\sim y$, if $\{x,y\}\in E.$ Let $$\mu:V \to (0, \infty), x\mapsto \mu_x,$$ be the vertex measure on $V$ and $$w: E \to (0, \infty), \{x,y\}\mapsto w_{xy}=w_{yx},$$ be the edge measure on $E.$ The quadruple $(V,E,\mu,w)$ is called a weighted graph. For a subset of $V$ (a subset of $E,$ resp.) we denote by $|\cdot|_\mu$ ($|\cdot|_w,$ resp.) the $\mu$-measure ($w$-measure, resp.) of the set. Let $\Omega$ be a finite subset of $V.$ We denote by $C_0(\Omega)$ the set of function on $V$ which vanishes on $V\setminus \Omega.$
For $p>1,$ we define the $p$-Laplacian with Dirichlet boundary condition, called Dirichlet $p$-Laplacian, on $\Omega$ as $$\Delta_p u(x)=\frac{1}{\mu_x}\sum_{y\in V: y\sim x}w_{xy}|u(y)-u(x)|^{p-2}(u(y)-u(x)),\quad x\in \Omega, u\in C_0(\Omega).$$ We call the pair $(\lambda, u)\in {{\mathbb R}}\times C_0(\Omega)$ satisfying $$\label{defi:eigenequation}-\Delta_p u(x)=\lambda |u(x)|^{p-2}u(x),\quad \forall x\in \Omega$$ the eigenvalue and the eigenfunction of Dirichlet $p$-Laplaican. The smallest eigenvalue of Dirichlet $p$-Laplacian is called the first eigenvalue, denoted by $\lambda_{1,p}(\Omega)$, and the associated eigenfunction is called the first eigenfunction, denoted by $u_p.$
For $p\geq 1,$ the $p$-Dirichlet energy is defined as $$E_p(u):=\sum_{x,y\in V, y\sim x}w_{xy}|u(y)-u(x)|^p,\quad u\in C_0(\Omega),$$ and the modified $p$-Dirichlet energy is defined as $$\widetilde{E_p}(u):=\frac{E_p(u)}{|u|^p},\quad \forall u\in C_0(\Omega)\setminus \{0\}.$$ We consider the variational problem $$\label{defil:variation}\widetilde{E_p}:C_0(\Omega)\setminus\{0\}\to {{\mathbb R}}.$$ One is ready to see that for $p>1,$ critical points and critical values of the problem satisfy . The spectral theory for $1$-Laplacian was developed by Chang and Hein. Eigenvalues and eigenfunctions of Dirichlet $1$-Laplacian are defined as critical values and critical points of the variational problem for $p=1.$ We denote by $\lambda_{1,1}(\Omega)$ the first eigenvalue for Dirichlet $1$-Laplacian on $\Omega.$ This yields the Rayleigh quotient characterization, $p\geq 1,$ $$\label{eq:Rayleigh}\lambda_{1,p}(\Omega)=\inf_{u\in C_0(V), u\neq 0}\widetilde{E_p}(u).$$ First eigenfunctions of Dirichlet $p$-Laplacian have some nice properties, see [@BH].
For a finite connected subset $\Omega$ of $V$ and $p>1,$ a first eigenfunction $u$ on $\Omega$ has fixed sign, i.e. $u>0$ on $\Omega$ or $u<0$ on $\Omega.$ Moreover, the first eigenvalue is simple, i.e. for any two first eigenfunctions $u$ and $v$ on $\Omega,$ $u=c v$ for some $c\neq 0.$
By this result, there is a unique first eigenfunction on $\Omega$ satisfies $$u>0,\quad \mathrm{on}\ \Omega,\quad |u|_p=1,$$ which is called the normalized first eigenfunction of $p$-Laplacian.
We introduce the definition of the Cheeger constant on $\Omega.$ For any subset $D\in V,$ the boundary of $D$ is defined as $\partial D:=\{\{x,y\}\in E: x\in D, y\in V\setminus D\}.$
The Cheeger constant on $\Omega$ is defined as $$h(\Omega)=\inf_{D\subset \Omega} \frac{|\partial D|_w}{|D|_\mu}.$$ A subset $D$ of $\Omega$ is called a Cheeger cut if $$\frac{|\partial D|_w}{|D|_\mu}=h(\Omega).$$
We prove the main result in the following.
\[thm:main1\] Let $\Omega$ be a finite connected subset of $V.$ For any sequence $\{p_i\}_{i=1}^\infty$ satisfying $p_i>1, p_i\to 1$, let $u_i$ be the corresponding normalized first eigenfunction of $p_i$-Laplacian. Then there is a subsequence $\{p_{i_k}\}$ such that $\{u_{i_k}\}_k$ converges $$\lim_{k \to \infty}u_{i_k}=\sum_{n=1}^N c_n{\mathds{1}}_{A_n} .$$ where $A_n,$ $1\leq n\leq N$, are Cheeger cuts of $\Omega$ satisfying $A_{N}\subsetneq A_{N-1}\subsetneq \cdots \subsetneq A_1, $ ${\mathds{1}}_{A_n}$ are characteristic functions on $A_n$ and $c_n>0.$
We have the following corollary.
\[coro:1\]For a finite connected subset $\Omega$ of $V,$ suppose that the Cheeger cut of $\Omega$ is unique. Then $$\lim_{p \to \infty}u_{p}=\frac{1}{|A|_\mu} {\mathds{1}}_A,$$ where $A$ is the Cheeger cut of $\Omega.$
Concerning with these results, we have the following open problems.
\[prob:1\] In general case, is it true that the normalized first eigenfunction $u_p$ converges, as $p\to 1$?
\[prob:2\] What are the limits of the normalized first eigenfunctions in Theorem \[thm:main1\].
For Problem \[prob:1\], since there is no uniqueness for the Cheeger cuts, see e.g. Example \[exam:1\] in Section \[sec:example\], one needs new ideas to prove the result. For Problem \[prob:2\], one might hope that the limit of a sequence of normalized first eigenfunctions is a characterization function of a single Cheeger cut, as in Corollary \[coro:1\]. By investigating Example \[exam:1\] in Section \[sec:example\], we show that this is not true in general. This indicates that the result in Theorem \[thm:main1\] cannot be improved to the characterization function of a single Cheeger cut.
The paper is organized as follows: In next section, we prove the main result, Theorem \[thm:main1\]. In Section \[sec:example\], we construct an example to show the sharpness of Theorem \[thm:main1\].
Proof of Theorem \[thm:main1\]
==============================
Let $(V,E,\mu,w)$ be a weighted graph and $\Omega$ is a connected subset of $V,$ i.e. for any $x,y\in \Omega$ there is a path, $x=x_0\sim x_1\sim \cdots \sim x_k=y$, connecting $x$ and $y$ with $x_i\in \Omega,\forall 1\leq i\leq k-1.$ We need some lemmas.
$$\lim_{p\to 1}\lambda_{1,p}(\Omega)=\lambda_{1,1}(\Omega).$$
Note that for any $u\in C_0(\Omega),$ $u\neq 0,$ $$\widetilde{E_p}(u)\to \widetilde{E_1}(u),\quad p\to 1.$$ Hence the lemma follows from the Rayleigh quotient characterization .
\[lem:1.2\] For any $p\geq 1,$ $$\lambda_{1,p}(\Omega)\leq h(\Omega).$$
As $\Omega$ is a finite subset, we choose $D\subset \Omega$ such that $
\frac{|\partial D|_w}{|D|_\mu}=h(\Omega).
$ Consider the characteristic function on $D,$ $${\mathds{1}}_D(x)=\left\{\begin{array}{ll}1,& x\in D,\\
0,& x\in V\setminus D,\end{array}\right.$$ Then by , $$\lambda_{1,p}(\Omega)\leq \widetilde{E_p}({\mathds{1}}_D)=\frac{|\partial D|_w}{|D|_\mu}=h(\Omega).$$ This proves the lemma.
\[lem:equivalent\]
$\lambda_{1,1}(\Omega)= h(\Omega).$
By Lemma \[lem:1.2\], it suffices to prove that $\lambda_{1,1}(\Omega)\geq h(\Omega).$ For any $u\in C_0(\Omega)$ with $|u|_1=1,$ let $g=|u|.$ For any $\sigma\geq 0,$ set $\Omega_{\sigma}:=\{x\in \Omega: g(x)>\sigma\}$ and $G(\sigma):=|\partial \Omega_\sigma|_w.$ Then $$G(\sigma)=\sum_{e=\{x,y\}\in E,g(x)\leq \sigma< g(y)}w_{xy}.$$
Hence $$\begin{aligned}
\int_{0}^{+\infty}G(\sigma) d\sigma&=&\int_{0}^{+\infty}\sum_{e=\{x,y\}\in E,g(y)>g(x)}w_{xy} {\mathds{1}}_{[g(x),g(y))}(\sigma)d\sigma\notag\\
&=&\sum_{e=\{x,y\}\in E,g(y)>g(x)}w_{xy}\int_{0}^{+\infty} {\mathds{1}}_{[g(x),g(y))}(\sigma)d\sigma\notag\\
&=& \sum_{e=\{x,y\}\in E,g(y)>g(x)}w_{xy}|g(x)-g(y)|=E_1(g). \notag\\\end{aligned}$$ Moreover, by the definition of $h(\Omega)$, $\Omega_{\sigma}\leq h(\Omega) | \Omega_{\sigma}|_\mu,\ \forall \sigma\geq 0.$ This yields that $$\begin{aligned}
\label{main-proof-2}
E_1(u)\geq E_1(g)=&\int_{0}^{+\infty}G(\sigma)d\sigma\geq h(\Omega) \int_{0}^{+\infty}| \Omega_{\sigma}|_\mu\\
=&h(\Omega)|u|_1=h(\Omega).\notag\end{aligned}$$ By taking the infimum over $u\in C_0(\Omega)$ with $|u|_1=1$ on the left hand side, we prove the lemma by .
By the definition, $u$ is called a first eigenfunction of $1$-Laplacian on $\Omega$ if $$\widetilde{E_1}(u)=\inf_{v\in C_0(V), v\neq 0}\widetilde{E_1}(v)=\lambda_{1,1}(\Omega).$$ Since $\widetilde{E_1}(|u|)\leq \widetilde{E_1}(u),$ $|u|$ is also a first eigenfunction of $1$-Laplacian on $\Omega.$
\[lem:structure\] Let $u$ be a nonnegative first eigenfunction of $1$-Laplacian on $\Omega.$ Then for any $\sigma\geq 0,$ $\Omega_\sigma:=\{x\in \Omega: u>\sigma\}$ is a Cheeger cut. Moreover, $$\label{eq:structure1}
u=\sum_{n=1}^N c_n{\mathds{1}}_{A_n} .$$ where $A_n,$ $1\leq n\leq N$, are Cheeger cuts of $\Omega$ satisfying $A_{N}\subsetneq A_{N-1}\subsetneq \cdots \subsetneq A_1,$ and $c_n>0.$
Let $\{a_i\}_{i=0}^M$ be the range of the function $u,$ such that $$0=a_0<a_1<\cdots<a_M=\max_{x\in V}u(x).$$ Then for any $\sigma\in [a_i, a_{i+1}),$ $0\leq i\leq M-1,$ $$\Omega_\sigma=\Omega_{a_i}.$$ By the proof of Lemma \[lem:equivalent\], see , $$\begin{aligned}
h(\Omega)|u|_1&=&\lambda_{1,1}(\Omega)|u|_1=E_1(u)=\int_{0}^{+\infty}G(\sigma)d\sigma=\sum_{i=0}^{M-1}(a_{i+1}-a_i)|\partial \Omega_{a_i}|_w\\
&\geq& h(\Omega) \sum_{i=0}^{M-1}(a_{i+1}-a_i)|\Omega_{a_i}|_\mu=h(\Omega)|u|_1.\end{aligned}$$ Hence for any $0\leq i\leq M-1,$ $$|\partial \Omega_{a_i}|_w=h(\Omega)|\Omega_{a_i}|_\mu.$$ So that for any $\sigma\geq 0,$ $\Omega_\sigma$ is a Cheeger cut.
For the other assertion, note that $$u=a_1{\mathds{1}}_{\Omega_{a_0}}+(a_2-a_1){\mathds{1}}_{\Omega_{a_1}}+\cdots+(a_M-a_{M-1}){\mathds{1}}_{\Omega_{M-1}}.$$ This proves the result.
Now we are ready to prove Theorem \[thm:main1\].
Note that $$|u_i|_{p_i}=1,\quad \widetilde{E_1}(u_i)=\lambda_{1,p_i}(\Omega).$$ For any $x\in \Omega,$ $|u_i(x)|\leq |u_i|_{p_i}=1.$ Since $\Omega$ is a finite set, there is a subsequence $\{p_{i_k}\}_k$ of $\{p_i\}_i$ and $u\in C_0(\Omega)$ such that $$u_{i_k}\to u,\quad \mathrm{pointwise\ on\ \Omega}.$$ Hence $$u\geq 0,\quad |u|_1=\lim_{k\to \infty}|u_{i_k}|_{p_{i_k}}=1,$$$$\quad E_1(u)=\lim_{k\to \infty}E_1(u_{i_k})=\lim_{k\to \infty}\lambda_{1,p_{i_k}}(\Omega)=\lambda_{1,1}(\Omega).$$ This yields that $u$ is a nonnegative normalized $1$-Laplacian eigenfunction. The theorem follow from Lemma \[lem:structure\].
This yields the following corollary.
It suffices to prove that for any sequence $p_i\to 0,$ $p_i>1,$ there is a subsequence $p_{i_k}$ such that $$\lim_{k \to \infty}u_{p_{i_k}}=\frac{1}{|A|_\mu} {\mathds{1}}_A,$$ where $A$ is the Cheeger cut of $\Omega.$ By Lemma \[lem:structure\], there is a subsequence $p_{i_k}$ and a nonnegative normalized first eigenfunction of $1$-Laplacian $u$ such that $$\lim_{k \to \infty}u_{p_{i_k}}=u.$$ Since the Cheeger cut of $\Omega$ is unique, by , $u=c{\mathds{1}}_A.$ By $|u|_1=1,$ $c=\frac{1}{|A|_\mu}.$ This proves the corollary.
An example {#sec:example}
==========
In this section, we give an example to show that the limit of the first eigenfunction of $\Delta_p$ may not be the characteristic function of a single set.
\[exam:1\]Consider the graph $G$ as in Fig. \[Fig:1\] such that edge weight $w_{x,y}=1$ for any $x\sim y$ and vertex weight $\mu_{x_1}=\mu_{x_2}=2,\mu_{y_1}=\mu_{y_2}=4.$ Let $\Omega=\{x_1,x_2,y_1,y_2\}.$ By the enumeration, the Cheeger cuts for $h(\Omega)$ are $$\{x_1,x_2\},\{x_1,x_2,y_1\},\{x_1,x_2,y_2\},\{x_1,x_2,y_1,y_2\}.$$
at (0,0)[![Example $G$[]{data-label="Fig:1"}](Cheeger-1-v2 "fig:"){width="0.84\linewidth"}]{}; at (-3.2, -.5)[$y_1$]{}; at (-1, -.5)[$x_1$]{}; at (1.15, -.5)[$x_2$]{}; at (3.2, -.5)[$y_2$]{};
We calculate the normalized first eigenfunctions $u_p$ of $p$-Laplacian on $\Omega$ for $p>1.$ One is ready to see that there is a symmetry, $T:\Omega\to \Omega,$ such that $$T(x_1)=x_2,T(x_2)=x_1,T(y_1)=y_2,T(y_2)=y_1,$$ perserving the Dirichlet boundary condition. Since the first eigenfunction is positive and unique up to constant multiplication, $u_p(x_1)=u_p(x_2),u_p(y_1)=u_p(y_2),$ see [@BH]. For convenience, we write $u_p$ as a vector $$u_p=(u_p(x_1),u_p(x_2),u_p(y_1),u_p(y_2)),$$ and by scaling we set $$v_p=\frac{1}{u_p(x_1)}u_p=(1,1,t_p,t_p).$$ Note that $v_p$ is also a first eigenfunction of $p$-Laplacian on $\Omega.$
Setting $x=t_p,$ by the eigen-equation, $$\left\{\begin{array}{ll}-\frac{1}{2}|1-x|^{p-2}(x-1)=\lambda_{1,p},\\
-\frac14(3|x|^{p-2}(-x)+\frac{1}{2}|1-x|^{p-2}(1-x))=\lambda_{1,p} x^{p-1}.
\end{array}\right.$$ By the first equation, $x<1.$ Plugging the first equation into the second and dividing it by $x^{p-1}$, we get $$2(1-x)^q+(\frac1x-1)^q-3=0,$$ where $q=p-1.$
Set $$f(x,q)=2(1-x)^q+(\frac1x-1)^q-3,\quad \forall x\in (0,1], q>0.$$ For any $q>0,$ since $f(\cdot, q)$ is monotonely decreasing, $f(1,q)=-3$ and $f(x,q)\to +\infty, x\to 0+$, there is a unique solution to $f(x,q)=0$ denoted by $x_q.$ Note that $x_q=t_p.$
For fixed $x\in (0,1),$ we write $a=1-x,$ $b=\frac1x-1,$ and set $$g(q)=f(x,q)=2 a^q+b^q-3.$$ For $q\geq 0,$ $$g'(q)=2 a^q\ln a+b^q\ln b ,\quad g''(q)=2 a^q(\ln a)^2+ b^q(\ln b)^2\geq 0,$$ where the derivatives are understood as right derivatives for $q=0.$ Hence $g(p)$ is convex in $(0,+\infty),$ which yields that for $q\geq 0,$ $$\label{eq:exam1}g(q)\geq g(0)+g'(0)q= \ln(a^2 b) q.$$ One can show that
$$\left\{\begin{array}{ll}a^2 b>1,& x<\hat{x},\\
a^2 b=1,& x=\hat{x},\\
a^2 b<1,& x>\hat{x},\\
\end{array}\right.$$
where $\hat{x}$ is the real solution of $(1-x)^3=x,$ given by $$\hat{x}=1-\sqrt[3]{\frac{\sqrt{93}+9}{18}}+\sqrt[3]{\frac{\sqrt{93}-9}{18}}\approx 0.31767.$$ Hence by , for any $x<\hat{x},$ $$g(q)\geq \ln(a^2 b) q>0,\quad q>0.$$ Hence $x_q\geq \hat{x}.$
For any $x>\hat{x},$ by Taylor expension of $g(q)$ at $q=0,$ $$g(q)= \ln(a^2 b) q+o(q),\quad q\to 0.$$ Hence by $a^2b<1,$ for sufficiently small $q,$ $$g(q)<0,$$ this yields that $x_q<x.$ This implies that $\limsup_{q\to 0} x_q\leq x.$ By passing to the limit $x\to \hat{x}+,$ we get $$\limsup_{q\to 0} x_q\leq \hat{x}.$$ Combining it with $x_q\geq \hat{x},$ $$\lim_{q\to 0} x_q=\hat{x}.$$
This yields that as $p\to 1,$ $$v_p=(1,1, t_p,t_p)\to (1,1, \hat{x},\hat{x}).$$ This yields that $$u_p=\frac{v_p}{|v_p|_p}\to \frac{1}{4+8\hat{x}}(1,1, \hat{x},\hat{x})\approx (0.15287,0.15287,0.04856,0.04856).$$ Hence, the limit is not a characteristic function on a single set.
[50]{}
*N. Alon and V. D. Milman*, $\lambda_1$, isoperimetric inequalities for graphs, and superconcentrators. J. Combin. Theory Ser. B, **38**(1985), 73–88.
*S. Amghibech*, Eigenvalues of the discrete p-Laplacian for graphs. Ars Combin., **67**(2003), 283–302.
*K. C. Chang*, Spectrum of the 1-Laplacian and Cheeger’s constant on graphs, J. Graph Theory, **81**(2016), no.2, 167–207.
*K. C. Chang, S. H. Shao, D. Zhang*, Cheeger’s cut, maxcut and the spectral theory of $1$-Laplacian on graphs, Sci. China Math. (2017), no. 11, 1–18.
*J. Cheeger*, A lower bound for the smallest eigenvalue of the Laplacian, Problems in analysis (Papers dedicated to Salomon Bochner, 1969), Princeton Univ. Press, Princeton, N. J., 1970, 195–199.
*F. Chung*, Spectral Graph Theory, Betascript Publishing, 2010:212.
*R. Courant and D. Hilbert*, Methods of mathematical physics. Vol. I. Interscience Publishers, Inc., New York, N.Y., 1953.
*J. Dodziuk*, Difference equations, isoperimetric inequality and transience of certain random walks, Trans. Amer. Math. Soc., **284**(1984), 787–794.
*M. Hein, T. Bühler*, An inverse power method for nonlinear eigenproblems with applications in 1-spectral clustering and sparse PCA. In Advances in Neural Information Processing Systems (NIPS), 847–855, Cambridge, MA, 2010. MIT Press.
*B. Hua and L. Wang*, [Dirichlet $p$-Laplacian eigenvalues and Cheeger constants on symmetric graphs]{}, preprint.
*B. Kawohl*, [Isoperimetric estimates for the first eigenvalue of the p-Laplace operator and the Cheeger constant.]{}, Commentationes Mathematicae Universitatis Carolinae (2003) Volume: 44, Issue: 4, page 659-667 ISSN: 0010-2628.
*M. Keller, D. Mugnolo*, General Cheeger inequalities for $p$-Laplacians on graphs, Nonlinear Analysis, **147**(2016), 80–95.
*H. Takeuchi*, The spectrum of the p-Laplacian and $p$-harmonic morphisms on graphs, Illinois J. Math., **47**(2003), 939–955.
*M. Yamasaki*, Discrete potentials on an infinite network, Memoirs of the Faculty of Literature and Science, Shimane University, **13**(1979), 31–44.
[^1]: H. Ge is supported by NSFC (China) under grant no. 11871094.
[^2]: B. Hua is supported by NSFC (China) under grant no. 11831004 and grant no. 11826031.
[^3]: W. Jiang is supported by Chinese Universities Scientific Fund(74120-31610002)
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We propose a picture of the fluctuations in branching random walks, which leads to predictions for the distribution of a random variable that characterizes the position of the bulk of the particles. We also interpret the $1/\sqrt{t}$ correction to the average position of the rightmost particle of a branching random walk for large times $t\gg 1$, computed by Ebert and Van Saarloos, as fluctuations on top of the mean-field approximation of this process with a Brunet-Derrida cutoff at the tip that simulates discreteness. Our analytical formulas successfully compare to numerical simulations of a particular model of branching random walk.'
author:
- |
A. H. Mueller${}^1$, S. Munier${}^2$[^1]\
\
*${}^{(1)}$Department of Physics, Columbia University, New York, USA\
*${}^{(2)}$Centre de physique théorique, École Polytechnique, CNRS, Palaiseau, France.**
title: '**Phenomenological picture of fluctuations in branching random walks**'
---
Introduction
============
The goal of this work is to better understand the distribution of the particles generated by a branching random walk process after some large evolution time.\
Our initial motivation for addressing this problem comes from particle physics [@Iancu:2004es] (for a review, see [@Munier:2009pc]). In the context of the scattering of hadrons at large energies, high-occupation quantum fluctuations dominate some of the scattering cross sections currently measured for example at the LHC. These quantum fluctuations can be thought of as being built up, as the hadrons are accelerated, by the successive branchings first of their constituant quarks into quark-gluon pairs, and then of the gluons into pairs of gluons, with some diffusion in their momenta. The dynamics of these gluons is actually exactly the kind of branching diffusion process that we are going to address in this work. Therefore, results that do not depend on the detailed properties of the particular branching random walk considered may be transposed to particle physics, and give quantitative insight into hadronic scattering cross sections.
Of course, the applications of branching random walks are much wider than particle physics. Branching random walks may for example generate Cayley trees which would represent the configuration space of directed polymers in random media [@DS:1988].\
Although our discussion will be very general, for definiteness, we shall consider a simple model for a branching random walk (BRW) in continuous time $t$ and one-dimensional space $x$, defined by two elementary processes: Each particle diffuses independently of the others with diffusion constant 1, and may split into two particles at rate 1, in such a way that the mean particle density $\langle n(t,x)\rangle$ obeys the equation \_t n(t,x)=\_x\^2 n(t,x)+n(t,x). \[eq:meannBRW\] A particular realization of this BRW is represented in Fig. \[fig:plotBRW\].\
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![\[fig:plotBRW\][(a)]{} One realization of the continuous BRW up to time $t=8$. To guide the eye, we also plot the theoretical (truncated) mean position of the boundaries of the BRW, namely $\pm\bar X_t=\pm 2t\mp\frac32 \ln t$ (dashed lines). [(b)]{} Distribution of the particles at times $t=2,6$ and $8$ for this particular realization in bins of size 1 ($\log_{10}$ scale on the vertical axis). We see the bulk building up a smoother (more “deterministic”) distribution as time elapses, while the low-density tails remain noisy. Also, for this realization, the distribution is skewed towards negative values of $x$, due to an accidentally large drift in the initial stages, whose memory is kept throughout the evolution. ](plotBRW.eps "fig:"){width=".45\textwidth"} ![\[fig:plotBRW\][(a)]{} One realization of the continuous BRW up to time $t=8$. To guide the eye, we also plot the theoretical (truncated) mean position of the boundaries of the BRW, namely $\pm\bar X_t=\pm 2t\mp\frac32 \ln t$ (dashed lines). [(b)]{} Distribution of the particles at times $t=2,6$ and $8$ for this particular realization in bins of size 1 ($\log_{10}$ scale on the vertical axis). We see the bulk building up a smoother (more “deterministic”) distribution as time elapses, while the low-density tails remain noisy. Also, for this realization, the distribution is skewed towards negative values of $x$, due to an accidentally large drift in the initial stages, whose memory is kept throughout the evolution. ](plotdensity.eps "fig:"){width=".45\textwidth"}
(a) (b)
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Several properties of BRW are known. In particular, in any given realization of the stochastic process, for large enough times, the forward part of the distribution of the particles looks like an exponential $e^{-x}$ (scaled by an appropriate time-dependent constant, also depending on the particular realization considered) up to fluctuations effectively concentrated at its low-density tip. We shall call this exponential part the “front.”
Then, one can also establish rigorously [@McK:1975; @B:1983] that the probability $Q(t,x)$ that all particles sit at a position smaller than $x$ obeys a nonlinear partial differential equation which reads \_t Q(t,x)=\_x\^2 Q(t,x)-Q(t,x)+Q\^2(t,x). \[eq:FKPP\] This is a version of the Fisher-Kolmogorov-Petrovsky-Piscounov (FKPP) equation [@F:1937; @KPP:1937]. (For an extensive review, see Ref. [@VS:2003], and for more applications of the FKPP equation, see e.g. Ref. [@Majumdar:2002]). If the BRW starts at time $t=0$ with a single particle located at $x=0$, then the initial condition is $Q(t=0,x)=\theta(x)$.
With such an initial condition, the solution of the FKPP equation tends to a so-called “traveling wave”. The position of a FKPP traveling wave, which is related to the average position of the rightmost particle in the BRW, is known in the large-time limit: $$\left[\text{FKPP front position}\right]=2t-\frac{3}{2}\ln t
+\text{const}
+\frac{C_X}{\sqrt{t}}+\cdots\ \
\text{with}\ \
C_X=-3\sqrt{\pi},
\label{eq:FKPPpos}$$ where the last term was found by Ebert and Van Saarloos [@EVS:2000]. (The additive constant depends on the way one defines the position of the front. It is uninteresting for our purpose.) Note that the Ebert-Van Saarloos term is a decreasing but [*positive*]{} contribution to the front velocity. Equation (\[eq:FKPPpos\]) may easily be extended to different branching diffusion models by appropriately replacing some numerical constants (see below, Sec. \[sec:numerics\]).
More generally, if $N(t)$ is the number of particles at time $t$, and $\{x_i(t)\}$ is the set of their positions in a given realization, then G\_t(x)\_[i=1]{}\^[N(t)]{} f(x-x\_i(t))\[eq:G\] for any given function $f$ satisfies the same FKPP equation as $Q$, the initial condition being the function $f(x)$ itself in the case of a BRW starting with one single particle at the origin. If $f$ is a monotonous function of $x$ such that $f(x)\underset{x\rightarrow-\infty}{\longrightarrow}0$ and $f(x)\underset{x\rightarrow+\infty}{\longrightarrow}1$, and if $f$ reaches 1 fast enough, namely $1-f(x)\underset{x\rightarrow+\infty}{\sim}e^{-\gamma x}$ with $\gamma>1$, then the traveling wave solution holds and the front position is still given by Eq. (\[eq:FKPPpos\]). Another interesting particular case is the “critical case” when $f$ is such that $\gamma=1$ exactly. Then, $$\left[\text{FKPP front position}\right]_\text{critical}=2t-\frac{1}{2}\ln t
+\text{const}
+{\frac{C_Y}{\sqrt{t}}}+\cdots
\label{eq:FKPPposcrit}$$ where $C_Y$ is a constant that we shall determine later on (see Sec. \[sec:det\]).\
There also exists a theorem established by Lalley and Sellke [@LS:1987] that gives the asymptotic (large time) shape of the distribution of the position of the rightmost particle in a frame whose origin is at position $\ln Z$, where $Z$ is some random variable that depends on the realization and may be thought of as a characterization of the position of the bulk of the particles in the BRW. (Its precise definition will be given later on). More recently [@BD:2009; @BD:2011; @Aidekon:2013], the distribution of the distances between the foremost particles was derived with the help of the solution to the FKPP equation with some peculiar initial condition.\
In this paper, we propose a phenomenological picture of the fluctuations in BRW, and we derive within this picture some new statistical properties of a random variable similar to $\ln Z$. (Appendix \[sec:Z\] also lists some properties of $\ln Z$ itself.)\
In Sec. \[sec:model\], we shall introduce our phenomenological picture for branching random walks. Section \[sec:distrib\] is devoted to deriving the quantitative predictions of this model for a particular random variable that can characterize the early-time fluctuations of the branching random walk. The computation of a few free constant parameters requires us to solve deterministic equations: This is explained in Sec. \[sec:det\]. Numerical checks are in order since our analytical results are based on conjectures: This is done in Sec. \[sec:numerics\]. In light of our phenomenological model, we shall then come back to the discussion of the Ebert-Van Saarloos result on the $1/\sqrt{t}$ correction to the position of FKPP fronts (Sec. \[sec:EVS\]). Conclusions are given in Sec. \[sec:conclusion\].
\[sec:model\] Phenomenological description of branching random walks
====================================================================
Picture
-------
The picture of the fluctuations in branching random walks (BRWs) that we have in mind is the following. There are essentially three types of fluctuations that may affect the position of the front or of the foremost particle.
1. First, there are fluctuations occurring at very early times ($t\sim 1$), when the system consists in a few particles. They have a large (of order 1) and lasting impact on the position of the front or of the rightmost particle. The main effect is given by the random waiting time of the first particle before it splits into two particles, during which it diffuses, but the subsequent waiting times of the latter two particles also contribute, etc... until the system contains a large enough number of particles that makes it partly “deterministic”. We do not believe that there is a simple way to compute the effect of these fluctuations, since there is no large parameter in the problem which would allow for sensible approximations.
2. Once the system contains many particles, which happens say at time $t_0^\prime\gg 1$, it enters a “mean-field” regime: In a first approximation, its particle density obeys a deterministic evolution with a moving absorptive boundary at a position that we shall call $\bar X_t$ (and symmetrically at $-\bar X_t$), set in such a way that the particle density is 1 at some fixed distance of order 1 to the left or to the right of this boundary, respectively. These boundaries simulate the discreteness of the particles. This is the Brunet-Derrida cutoff [@BD:1997], and it was shown to correctly represent the leading effect of the noise on the position of the front in the context of the stochastic FKPP equation.
From now on, we shall focus on the right boundary. (The right and the left halves of the BRW essentially decouple once the system has grown large enough). The large-time expression of the shape of the particle density near the right boundary reads, in such a model $$\psi_{\bar X_t}(x,t)=\left[\alpha(x-\bar X_t)+\beta\right]
\exp\left(\bar X_t-x-\frac{(\bar X_t-x)^2}{4t}\right)
\theta(\bar X_t-x)
\label{eq:shape}$$ in the region $1\ll\bar X_t-x\lesssim \sqrt{t}$, where $$\bar X_t=2t-\frac32\ln t+\frac{C_{\bar X}}{\sqrt{t}}
\label{eq:Xt}$$ is, up to an uninteresting non-universal additive constant, the position of the tip of the front, namely of the right boundary. The Heaviside step function $\theta$ enforces the fact that the particle density is 0 to the right of $\bar X_t$. $C_{\bar X}$, $\alpha<0$ and $\beta$ are constants undetermined at this stage. $\psi$ is essentially a decreasing exponential supplemented with a Gaussian and a linear prefactors. The $t$-dependence enters explicitly as the width of the Gaussian, and implicitly through the position $\bar X_t$ of the absorptive boundary. (There are corrections to the shape of the front at order $1/\sqrt{t}$, namely to the function $\psi_{\bar X}$ itself, but it turns out that we do not need to take them into account in our model, except for the determination of one overall numerical constant: We will come back to the derivation of Eq. (\[eq:shape\]) and of its corrections in Sec. \[sec:det\].)
The fluctuations on top of this essentially deterministic front we have just described must take place in the tip region, where the particle density is low. We shall assume that a single fluctuation effectively gives the dominant correction to the deterministic evolution, and that the distribution $p(\delta)$ of the position $\delta$ of this fluctuation with respect to the tip of the front is exponential: $$p(\delta)=C_1 e^{-\delta}.
\label{eq:proba}$$ We have found (see below) that these fluctuations bring a contribution of order $1/\sqrt{t}$ to the average position both of the front and of the rightmost particle in the BRW.
3. Finally, there are tip fluctuations occurring at very late times, say between $t-\bar t_0$ and $t$, where $\bar t_0$ is of order 1. They are also distributed as $e^{-\delta}$. They obviously add noise to the position of the tip of the front, but they do not have an effect on the bulk of the particle distribution since they do not have time to develop their own front at time $t$.
This picture is parallel to the phenomenological model for front fluctuations proposed in Ref. [@Brunet:2005bz] in the context of the stochastic FKPP problem.
\[sec:var\]Variables
--------------------
To arrive at quantitative predictions for the behavior of the BRW, we need to introduce random variables that characterize the realizations. We shall discuss the following ones:
- $X_t$, the position of the rightmost particle,
- $Y_t=\sqrt{t}\sum_i e^{x_i(t)-2t}$, where the sum goes over all the particles in the system,
- $Z_t=\sum_i \left[2t-x_i(t)\right]e^{x_i(t)-2t}$.
Throughout, we shall denote by $\langle A \rangle$ the statistical average (over realizations) of a given variable $A$ in the full stochastic model, and by $\bar{A}$ the value of this variable in a mean-field approximation of the same model with a discreteness cutoff at the tip. (These notations have already been used in Eq. (\[eq:meannBRW\]) and Eqs. (\[eq:shape\]), (\[eq:Xt\]) respectively.) Discrete sums over the particles will often be replaced by integrals wherever the particle density is large enough.\
Let us briefly comment on the random variables we have just introduced.
- As already mentioned, $\langle X_t\rangle$ is related to the solution of the FKPP equation with the step function as an initial condition.
- The average $\langle \ln Y_t\rangle$ tends to a constant at large $t$. In addition, in any given event, the random variable $\ln Y_t$ itself tends to a constant, which has some distribution (which we do not know how to compute) which may be used to characterize the early-time fluctuations. Note that an appropriate generating function of the moments of Y\_t\_i e\^[x\_i(t)]{}=Y\_t \[eq:defYt\] also obeys the FKPP equation, but with the “critical” initial condition discussed in the Introduction. We will come back to the latter fact in Sec. \[sec:det\]. Also, in the context of directed polymers in random media, $\tilde Y_t$ is the partition function and $\langle\ln \tilde Y_t\rangle$ the free energy averaged over the disorder [@DS:1988].
- $Z_t$ is the variable used by Lalley and Sellke in the theorem alluded to in the Introduction. However, we are not going to focus on this variable in the body of this paper, since we found it has many drawbacks for our purpose. First, a practical drawback: Although $Z_t$ tends almost surely to a positive constant when $t\rightarrow+\infty$ [@LS:1987], it takes negative values at finite times, with finite probability; $\ln Z_t$ is then undefined in these particular realizations. Second, a theoretical drawback: It turns out that the finite-time corrections to the moments of $\ln Z_t$ are very sensitive to the initial fluctuations, the ones that are not computable analytically. We shall nevertheless quote a few results on the distribution of $\ln Z_t$ in Appendix B.
In some intuitive sense, $\ln Y_t$ and $\ln Z_t$ characterize the position of the “front” of a particular realization of the evolution at time $t$.\
The variables $\ln Y_t$ and $\ln Z_t$ keep the memory of the initial fluctuations. Therefore, we shall not attempt to compute the distribution of the fluctuations in $\ln Y$ accumulated over the whole history of the BRW, but instead the fluctuations of this variable between two large times $t_0$ and $t$, in order to have a quantity that is independent of the very early times at which there is no mean-field regime.
\[sec:distrib\]Statistics of $f\equiv \ln Y_t-\ln Y_{t_0}$ in the phenomenological picture
===========================================================================================
Here, starting from the phenomenological model defined in Sec. \[sec:model\], we shall deduce new results on the distribution $p(f)$ of the variable $f\equiv \ln Y_t-\ln Y_{t_0}$ (and on its moments) for $t,t_0,t-t_0\gg 1$, up to one single constant for the moments of order larger than 2, and up to an additional constant for the first moment. Throughout, we shall aim at the accuracy ${\cal O}\left(1/\sqrt{t_0},1/\sqrt{t},1/\sqrt{t-t_0}\right)$ for $p(f)$ and neglect higher powers of these expansion variables.
\[sec:distrib1\]Effect of a fluctuation on $\ln Y$
--------------------------------------------------
We first compute $\bar Y_t$, namely the variable $Y_t$ in the mean-field approximation with the cutoff in the tail. Using the definition of the variable $Y_t$ in Sec. \[sec:var\] and using Eqs. (\[eq:shape\]),(\[eq:Xt\]), we find $$\bar Y_t=\sqrt{t}\int_{-\infty}^{+\infty}dx\, \psi_{\bar X_t}(x,t)e^{x-2t}=-2\alpha
\left(1+\frac{C_{\bar Y}}{\sqrt{t}}\right),
\label{eq:Y0}$$ at first order in $1/\sqrt{t}$. $C_{\bar Y}$ is a constant literally equal to $C_{\bar X}-\frac{\beta\sqrt{\pi}}{2\alpha}$ in this calculation, but there would also be other contributions to $C_{\bar Y}$ that we cannot get from the large-$t$ asymptotic shape of the front exhibited in Eq. (\[eq:shape\]). We shall postpone the full calculation of $C_{\bar Y}$ to Sec. \[sec:det\].
It turns out that the term of order $1/\sqrt{t}$ in Eq. (\[eq:Y0\]) generates ${\cal O}(1/t)$ contributions to the distributions and to the moments that we shall address. Hence it is enough for our purpose to keep no more than the constant term, namely we write |Y\_t-2. \[eq:Y00\] We perform a more complete calculation in Appendix \[sec:details\], keeping the subleading terms, in order to demonstrate that this a priori approximation is indeed accurate enough.
Let us consider a fluctuation occurring at time $t_1\gg 1$ at a distance $\delta$ from the tip of the deterministic front. From Eq. (\[eq:Xt\]), at time $t>t_1$ such that $t-t_1\gg1$, this fluctuation has developed its own front whose tip sits at position $$\bar X_{\delta,t}=\bar X_{t_1}+\delta+\bar X_{t-t_1}
=\bar X_t+\delta-\frac32 \ln\frac{t_1(t-t_1)}{t}.
\label{eq:Xdelta}$$ There would of course be terms proportional to $1/\sqrt{t_1}$, $1/\sqrt{t}$ and $1/\sqrt{t-t_1}$ also here, but we again anticipate that they would eventually lead to corrections of higher order to the quantities of interest. We refer the reader to Appendix A for the details.
The shape of the front generated by this fluctuation will eventually have the form $C\times \psi_{\bar X_{\delta,t}}(x,t-t_1)$, where $C$ is a constant that we cannot determine since it is related to some “average” shape of the fluctuation. With this extra fluctuation, $Y_t$ has the following expression: Y\_t=\_[-]{}\^[+]{}dx \_[|X\_t]{}(x,t)e\^[x-2t]{} +C\_[-]{}\^[+]{}dx \_[|X\_[,t]{}]{}(x,t-t\_1)e\^[x-2t]{} . The first term is just $\bar Y_t$: We replace it by Eq. (\[eq:Y00\]). The second term is integrated in the same way as the first one, using the expression (\[eq:Xdelta\]) for $\bar X_{\delta,t}$. We find Y\_t=-2(1+C ). Thus the forward shift in $\ln Y_t$ induced at time $t$ by such a fluctuation occurring at time $t_1$ reads Y\_t=Y\_t-|Y\_t =(1+C ). \[eq:shift\] Note that in the asymptotic limit of interest, at first glance, the nontrivial term in this expression seems to be of order $1/t_1^{3/2}$, thus, if $t_1\sim t_0$, it is smaller than our accuracy goal. However, it is enhanced by the $e^{\delta}$ factor, which turns out to be large.
Probability distribution and moments
------------------------------------
We may convert the conjectured probability of a forward fluctuation of size $\delta$ (Eq. (\[eq:proba\])) into the probability distribution of the difference of $\ln Y$ between two times $t_0$ and $t>t_0$ by simple changes of variables. We first discuss the variable fY\_t-Y\_[t\_0]{}. The fundamental observation is that a fluctuation may essentially have two opposite effects on $\delta f\equiv \delta\ln Y_t-\delta\ln Y_{t_0}$. If it occurs after time $t_0$, then it gives a positive contribution. If instead it occurs before $t_0$, it generates a negative $\delta f$. Now we observe that the difference between $\delta f$ and $f$ reads $\ln {\bar Y_t}/{\bar Y_{t_0}}$, which is of order $1/\sqrt{t},1/\sqrt{t_0}$ and thus, we may trade $\delta f$ for $f$ (see Appendix \[sec:details\] for more details).\
Let us first address the case in which the fluctuation occurs between $t_0$ and $t$. Using Eq. (\[eq:shift\]) together with the distribution (\[eq:proba\]), the probability that the size of the shift in $\delta f$ induced by a fluctuation at time $t_1$ is less than some $F$ reads P(f<F;t\_1)=C\_1( 1- ). We shall always assume that $F$ is finite, and the ordering $t,t-t_1,t_1\gg 1$. The probability distribution of $f$ then reads p(f;t\_1)=|\_[F=f]{}= . \[eq:pf01\] The rate of the fluctuations is assumed constant in time, thus the distribution of $f$ results from a simple integration over $t_1$ from $t_0$ to $t$ with uniform measure. It reads p(f)=2CC\_1 . \[eq:pf1\]
Exactly in the same way, we may compute the probability distribution of $f$ when the fluctuation occurs at a time smaller than $t_0$. In this case, the effect on $f$ of a fluctuation of size $\delta$ reads f=. Using the same method, we find p(f)=2C C\_1 ( 1- ) . \[eq:pf2\] The integral over $t_1$ which has to be performed to arrive at these expressions is dominated by values of $t_1$ of the order of $t_0$. This helps us to understand a posteriori why we were allowed to drop terms of order $1/\sqrt{t}$ and $1/\sqrt{t_1}$ in Eqs. (\[eq:Y0\]) and (\[eq:Xdelta\]), respectively, although we were aiming at such accuracy for $p(f)$.
The probability distribution given in Eq. (\[eq:pf1\]) and (\[eq:pf2\]) cannot be normalized and the first moment $\langle f\rangle$ is also divergent. We shall compute the latter separately in the next section.\
An analytic continuation of the generating function for the moments of $f$ can be obtained from Eq. (\[eq:pf1\]) and (\[eq:pf2\]) by a direct calculation. We get e\^[f]{} = 2CC\_1 { -(-)+()+ }, \[eq:moments\] keeping in mind that this formula can be used only for moments of second order or higher.
The analytical structure of Eq. (\[eq:moments\]) is particularly simple. There are poles on the positive real $\nu$-axis, fully contained in the first term $-\nu\psi(-\nu)$: They correspond to positive values of $f$. All the other poles, contained in the remaining terms, are located on the real negative axis, and correspond to negative values of $f$.\
A comment is in order on the conjectured probability distribution (\[eq:proba\]) of the tip fluctuations that we used in the above derivation. Actually, we omitted a time-dependent Gaussian factor of the form $e^{-\delta^2/(4t_1)}$, which would cut off the exponential distribution of $\delta$ at a distance $2\sqrt{t_1}$ ahead of the tip of the front, and thus modify the distribution of $f$ for large positive $f$. However, numerically, we do not find evidence for such a modification: It seems that Eq. (\[eq:pf1\]) has a more general validity. We do not have a good explanation for this surprising fact in the context of our phenomenological model for fluctuations. But it turns out that a different calculation of the positive $f$ fluctuations outlined in Appendix \[sec:exact\] does not have such limitations.
\[sec:mu1\] Correction to the first moment of $f$ due to the fluctuations
-------------------------------------------------------------------------
Since it is not possible to use Eq. (\[eq:moments\]) to get the first moment of $f$, we shall arrive at its expression through a direct calculation. We must take into account the expansion (keeping terms at least as large as $1/\sqrt{t}$, $1/\sqrt{t_0}$, $1/\sqrt{t-t_0}$) of the density of particles in the deterministic limit with a discreteness cutoff, and, in addition, the effect of the fluctuations which intermittently speed up the evolution. We have already guessed that there is an ${\cal O}(1/\sqrt{t})$ contribution to the deterministic evolution (see Eq. (\[eq:Y0\])), but a full calculation will eventually be needed. Here, we shall simply denote by $C_{\bar Y}$ its coefficient.\
The average of $f=\ln Y_t-\ln Y_{t_0}$ over realizations has thus a mean-field contribution, and a contribution from the fluctuations which in turn can be decomposed in positive and negative contributions $\mu_1^+$ and $\mu_1^-$ respectively. We shall evaluate the latter in this section.
We write \_1=Y\_t-Y\_[t\_0]{} =C\_[|Y]{} (-) +\_1\^[+]{}-\_1\^[-]{}. \[eq:c1\] Using Eq. (\[eq:proba\]) and Eq. (\[eq:shift\]), we get the expression \_1\^[+]{}=\_[t\_0\^]{}\^[t]{}dt\_1 \_0\^[+]{} dC\_1 e\^[-]{} (1+C ) for the positive part of the contribution at $t$ of the fluctuations, and \_1\^[-]{}=\_[t\_0\^]{}\^[t\_0]{}dt\_1 \_0\^[+]{} dC\_1 e\^[-]{} ( [1+C ]{}) subtracts the effect at $t_0$ of the fluctuations occurring at $t_1<t_0$. We have introduced a time $t_0^\prime$ of order 1 as a lower bound in these integrals in order to make these expressions finite. The physical meaning of this cutoff is clear: Before $t_0^\prime$, there is no mean-field regime because the whole system consists in a few particles only.
Let us start with the computation of $\mu_1^+$. It is useful to perform the change of variables = , u\_=e\^[-]{}\^[3/2]{}, \[eq:change\] which leads to the following expression of $\mu_1^{+}$: \_1\^[+]{}=\_\^1 \_0\^[u\_0()]{}du\_(1+), \[eq:mu1+\] where $u_0(\lambda)=\frac{t^{3/2}}{C}\lambda^{3/2}\sqrt{1-\lambda}$. $u_0$ is large compared to 1, except when $1-\lambda$ is of order $1/t^3$. But the contribution of the region $[1-1/t^3,1]$ in the $\lambda$-integration is smaller than $\sim 1/t^{3/2}$, and hence negligible. So we may always assume $u_0\gg 1$.
The integral over $u_\delta$ is performed analytically, and the large-$u_0$ limit may eventually be taken: \_0\^[u\_0]{}du\_(1+)=(1+u\_0)(1+u\_0)-u\_0u\_0 u\_0. The remainder reads \_1\^+=, \[eq:mu1+I0I1\] where \_0=\_[t\_0\^/t]{}\^1 d\^[-3/2]{}(1-)\^[-1/2]{}, \_1=\_[t\_0\^/t]{}\^1 d\^[-3/2]{}(1-)\^[-1/2]{} (\^[3/2]{}). \[eq:defI0\] ${\cal I}_0$ and ${\cal I}_1$ can be performed with the help of the change of variable $\lambda=\sin^2\theta$: \_0=2\_\^ =2 () =2, while for ${\cal I}_1$, a further integration by parts is needed to get rid of the log. We eventually arrive at the following exact expression for (\[eq:mu1+I0I1\]): \_1\^[+]{}=2CC\_1{ - }. \[eq:c1+\] The term $\mu_1^-$ is the same as the term $\mu_1^+$ except for the replacement $t\rightarrow t_0$.
Since we are neglecting terms of relative order $t_0/t$, $t_0^\prime/t_0$ and higher, we may expand the expressions for $\mu_1^+$ and $\mu_1^-$. The difference $\mu_1^+-\mu_1^-$ then reads \_1\^+-\_1\^-=4CC\_1(-). Equation (\[eq:c1\]) eventually leads to the following expression for $\mu_1$: \_1=(4C C\_1-C\_[|Y]{})(-). \[eq:mu1\]
We note a very important property of this result: It does not depend on $t_0^\prime$. If it did, then we would loose predictivity because $t_0^\prime$ is the arbitrary time after which we declare that the fluctuations are small enough for our calculation to apply. (This would not be true at the next order in $1/\sqrt{t}$, $1/\sqrt{t_0}$).
\[sec:det\]Deterministic calculations
=====================================
In this section, we first review the Ebert-Van Saarloos method [@EVS:2000] to compute the order $1/\sqrt{t}$ correction to the mean position of the rightmost particle in the BRW $\langle X_t\rangle$. We extend the method to the position of the right boundary in the deterministic model with discreteness cutoffs $\bar X_t$, and eventually adapt it to $\langle \ln Y_t\rangle$.
The calculations presented here will enable us to determine the remaining unknown constants, namely $C_{\bar X}$ (see Eq. (\[eq:Xt\])), $C_{\bar Y}$ (Eq. (\[eq:Y0\])), and $CC_1$. The latter two constants appear in particular in Eqs. (\[eq:pf1\]), (\[eq:pf2\]), (\[eq:moments\]) and (\[eq:mu1\]).
\[sec:noncritical\] Ebert-Van Saarloos calculation and its extension
--------------------------------------------------------------------
The original calculation of Ebert and Van Saarloos aimed at finding properties of the solutions to the FKPP equation \_t (t,x)=\_x\^2(t,x)+(t,x)-\^2(t,x) \[eq:FKPPphi\] for $\phi\ll 1$, with a steep enough initial condition, e.g. $\phi(t=0,x)=\theta(-x)$. This equation is actually the same as Eq. (\[eq:FKPP\]), with the correspondence $\phi(t,x)=1-Q(t,x)$. The nonlinearity can essentially be viewed as a moving absorptive boundary on a linear partial differential equation, the position of the boundary being set in such a way that $\phi$ has a maximum at a fixed height.
To determine the value of the constants $C_{\bar X}$ and $C_{\bar Y}$, we need to address a branching diffusion equation with a nonlinearity that forces $\phi$ to go to zero over a distance of order 1 at the right of the point at which $\phi(t,x)=1$, and which therefore acts as a tip cutoff. In terms of a smooth equation, we may write, for example, \_t (t,x)=\_x\^2(t,x)+ \[eq:tipcutoffsmooth\] and study the properties of the solutions to this equation in the region $\phi\gg 1$.
In both cases, the equation can be linearized in the respective domain of interest, and one gets \_t (t,x)=\_x\^2(t,x)+(t,x). We shall assume that the nonlinear term is equivalent to a right-moving absorptive boundary at the accuracy at which we want to address the problem. (This assumption was better motivated by Ebert and Van Saarloos in their discussion of what they call the “interior expansion” [@EVS:2000]). In the first case, we study the function $\phi$ to the right of the boundary; in the second case, we study the function to the left.
#### Solution to the linearized equation with an appropriate boundary condition.
Near the boundary, at large times, the function $\phi$ reads (t,x)\~(+)e\^[-]{}, \[eq:phiinterior\] where $\xi=x-\text{[position of the cutoff]}\sim x-2t+\frac32\ln t+\dots$ and this is valid for $1\ll \xi\ll \sqrt{t}$. According to Ebert-Van Saarloos [@EVS:2000], the large-$t$ corrections to this shape are of the form $1/t$ (there is no term of order $1/\sqrt{t}$). All these features should not depend on whether we address Eq. (\[eq:FKPPphi\]) or Eq. (\[eq:tipcutoffsmooth\]) above, except for the signs of $\alpha$ and $\xi$.
We write (t,x)=e\^[--z]{}g(t,z) , where $z=\frac{\xi^2}{4t}$, and the following ansatz are taken:
&=x-2t+32t++\
g(t,z)&= g\_[-12]{}(z)+g\_0(z)+
\[eq:ansatzg\] The variable $\xi$ may be positive in the linear domain (it is the case for the usual Ebert-Van Saarloos solution) or negative: Therefore, we write $\xi=\varepsilon\sqrt{4t\,z}$, where $\varepsilon=\pm 1$. The ansatz for the front position contained in $\xi$ already incorporates the two known [@B:1983] dominant terms at large $t$, namely $2t-\frac32\ln t$. The $-\frac{2c}{\sqrt{t}}$ term was new in Ref. [@EVS:2000].
Thanks to these ansatz, the original equation splits into a hierarchy of equations for the functions $g$. The first two equations of this set read
&z g\_[-12]{}\^+(12-z) g\_[-12]{}\^+12 g\_[-12]{}=0 ,\
&z g\_[0]{}\^+(12-z) g\_[0]{}\^+g\_[0]{}= c g\_[-12]{}-32 (g\_[-12]{}-g\^\_[-12]{}) .
\[eq:hierarchy\] The first equation of the hierarchy is the Kummer equation z+(b-z)-aw=0 with $w=g_{-\frac12}$, $a=-\frac12$ and $b=\frac12$. Two independent solutions are, for example, the two Kummer functions (or ${}_1F_1$ hypergeometric functions) M(a,b,z) z\^[1-b]{}M(a-b+1,2-b,z) namely, in our case, M(-12,12,z) M(0,32,z). The latter is just the elementary function $\sqrt{z}$, while the former diverges like $-e^{z}/(2z)$ for large $z$, and has thus to be discarded. Hence the solution reads g\_[-12]{}=2\^ \[eq:g-12sol\] where $\alpha^\prime$ is a constant, arbitrary at this stage.
As for the second equation in Eq. (\[eq:hierarchy\]) whose solution is the function $g_0$, it is an inhomogeneous Kummer differential equation. A basis for the solutions of the homogeneous part is, for example, the set of the two functions M(-1,12,z)=1-2z M(-12,32,z). We need to find a particular solution of the full equation. We define $y\equiv\sqrt{z}$; The equation for $g_0$ then reads -2y+4g\_0 = 8\^( -32y\^2 +cy +34) \[eq:diffg0\] and we may look for solutions in terms of a series: g\_0(y)=\_[k=0]{}\^[+]{}a\_k y\^k . Inserting this expression into the differential equation (\[eq:diffg0\]), we get the following relations between the coefficients of the series: a\_[k+2]{}=a\_k , a\_2=-2a\_0+3[\^]{} , a\_3=-+ , a\_4=-[\^]{} . The free parameters are $a_0$ and $a_1$. We may choose them in such a way that $a_{2,3}=0$: We therefore set $a_0=\frac32{\alpha^\prime}\varepsilon$ and $a_1=4{\alpha^\prime} c$. Then a\_[2n]{}=-32 , , where we used the duplication formula $\Gamma(2n+1)=\frac{2^{2n}}{\sqrt{\pi}}
\Gamma(n+\frac12)\Gamma(n+1)$. Switching back to the variable $z$, the final expression for the particular solution reads g\_0\^[sp]{}(z)=32[\^]{}+4[\^]{} c -F\_2(z) F\_2(z)= \_[n=2]{}\^z\^n , which, except for the sign factors $\varepsilon$, is the Ebert-Van Saarloos result [@EVS:2000]. Following again Ref. [@EVS:2000], we write the solution for $g_0$ as g\_0(z)=32[\^]{}+4[\^]{} c -F\_2(z)+k\_0(1-2z) +l\_0M( -12,32,z ) \[eq:g0sol\] and inserting (\[eq:g0sol\]) together with (\[eq:g-12sol\]) into (\[eq:ansatzg\]),(\[eq:phiinterior\]), we would get the expression of $\phi$ up to the constants $\alpha^\prime,c,k_0,l_0$. We are now going to determine them from a matching procedure.
#### Matching conditions.
We now match with the shape of the so-called “interior” region at $z\ll 1$. This means that $\phi$ just obtained should have the same small-$z$ expansion as the limiting form of $\phi$ in Eq. (\[eq:phiinterior\]). Hence we need to impose g\_[-12]{}(z)\~2 g\_0(z)\~+[O]{}(z). The first constraint is solved by setting $\alpha^\prime=\alpha$. As for the second one, it means in particular that there should be no term proportional to $\sqrt{z}$ in $g_0(z)$. This requirement leads to the equations +k\_0= , 4c+[l\_0]{}=0. \[eq:require\]
Now we must also check the behavior at $z\rightarrow+\infty$. We need the expansion of the functions $M$ and $F_2$ for $z\rightarrow\infty$. Let us start with $M$. We shall use the integral representation M(a,b,z)= \_0\^1 due\^[zu]{} u\^[a-1]{}(1-u)\^[b-a-1]{}. We change the variable for $u$ to $1-u$ in the integral, and we expand the $(1-u)^{a-1}$ factor near $u=0$: M(a,b,z)= e\^z\_0\^1 du e\^[-zu]{}u\^[b-a-1]{}\_[k=0]{}\^[+]{} u\^k. We then notice that we may extend the integral to $+\infty$ without adding exponentially-enhanced terms. Finally, we perform the remaining integration over $u$. The result reads
M(a,b,z)&=e\^zz\^[a-b]{} \_[k=0]{}\^[+]{} +o(e\^z)\
&=e\^z z\^[a-b]{}\_2F\_0(1-a,b-a;1/z)+o(e\^z).
Setting $a=-1/2$ and $b=3/2$, we write M(-12,32,z) \~-14 e\^[z]{}z\^[-3/2]{} \_[k=0]{}\^[+]{}(1+k)z\^[-k]{} =-14 e\^z z\^[-3/2]{}\_2F\_0(32,2;;). We now turn to $F_2$. We write the following integral representation: 12 F\_2(z)=\_[0]{} . This representation may be checked by expanding the exponential in the integral and performing the integration over $u$. For large $z$, the two rightmost terms do not play any role since they are not exponentially enhanced. We may now treat the first term exactly in the same way as in the case of the Kummer function $M$. After taking the $\eta\rightarrow 0$ limit, which is finite once all non-exponentially enhanced terms have been discarded, we get F\_2\~2e\^[z]{}z\^[-3/2]{}\_[k=0]{}\^[+]{} (k+32)(k+1)z\^[-k]{}. Up to an overall constant, all terms are identical to the ones in the expansion of the $M$ function.
Requiring the cancellation of these exponentially-enhanced terms in the expression (\[eq:g0sol\]) for $g_0$ leads to the equation 32+=0 . Using this equation and the second equation in (\[eq:require\]), one determines the value of $c$: c=32 . Hence this constant is [*positive*]{} for the Ebert-Van Saarloos solution of the FKPP equation, but is [*negative*]{} when one computes the position of the tip of a front with a discretness cutoff.
#### Matched solution.
All in all, we get (t,x)=e\^[--z]{}{ ++( 3-2)z +6 -F\_2(z) } \[eq:phinoncrit\] with =x-2t+32t+3. The first two terms in $\phi$, namely $e^{-\xi-z}(\alpha\xi+\beta)$, give back Eq. (\[eq:shape\]). The next terms are finite-time corrections.
Identifying $\xi$ with $x-\langle X_t\rangle$ and setting $\varepsilon=+1$, we recover the value of C\_[X]{}=-3 \[eq:CX\] already derived by Ebert and Van Saarloos (see Eq. (\[eq:FKPPpos\])). With $\xi=x-\bar X_t$ and $\varepsilon=-1$, we read off this formula the value of the constant C\_[|X]{}=3. \[eq:CMFX\] We can also deduce the value of $C_{\bar Y}$ by using the definition of the variable $Y_t$ given in Sec. \[sec:var\] and the shape of the mean-field particle distribution (\[eq:phinoncrit\]): |Y\_t=\_[-]{}\^[+]{}dx (t,x)e\^[x-2t]{}, which, after replacement by the expression (\[eq:phinoncrit\]) and setting $\varepsilon=-1$, becomes $$\begin{gathered}
\bar Y_t=e^{3\sqrt{\pi/t}}
\bigg\{
-2\alpha-\frac{1}{\sqrt{t}}
\int_0^{+\infty}\frac{dz}{\sqrt{z}}e^{-z}
\bigg[
\beta(1-2z)\\
-3\alpha z-6\alpha\sqrt{\pi z}
\left\{1-M\left(-\frac12,\frac32,z\right)\right\}
+\frac{3\alpha}{2}F_2(z)
\bigg]
\bigg\}.\end{gathered}$$ The term proportional to $\beta$ is zero after the integration, and the other terms give numerical constants. We finally find, at order $1/\sqrt{t}$, =(-2)+ , C\_[|Y]{}=32. \[eq:CMFY\]
\[sec:critical\] Solution of the deterministic FKPP equation with the critical initial condition
-------------------------------------------------------------------------------------------------
Let us consider a generating function of the moments of the $\tilde Y_t$ variable: G\_t(x)=e\^[-Y\_t e\^[-x]{}]{}= \_[i=1]{}\^[N(t)]{} e\^[-e\^[-(x-x\_i(t))]{}]{}. Defining $f(x)=e^{-e^{-x}}$, $G_t(x)$ has exactly the form shown in Eq. (\[eq:G\]) and thus $\phi(t,x)\equiv 1-G_t(x)$ solves the FKPP equation (\[eq:FKPPphi\]), $\partial_t\phi=\partial_x^2\phi +\phi-\phi^2$. If the initial condition for the underlying branching random walk is a single particle at position $x=0$, (t=0,x)=1-e\^[-e\^[-x]{}]{} and then, the position of the FKPP traveling wave is given by Eq. (\[eq:FKPPposcrit\]). In this section, we shall address this case using the Ebert-Van Saarloos method in order to obtain the $1/\sqrt{t}$ correction to the latter and some analytic features of $\phi$. Indeed, from the expression of $\phi$, we may in principle compute the moments of $\ln\tilde Y_t$, using the identity Y\_t\^= - \_[-]{}\^[+]{}dxe\^[x]{}. \[eq:MomentsFromPhi\]
#### General solution of the linearized equation in a moving frame.
Following Ebert-Van Saarloos, we define =x-2t-\_t (t,x)=e\^[-]{}(t,). \[eq:defpsi\] The linearized FKPP equation for $\psi$ reads \_t(t,)=\_\^2(t,)+\_t(\_-1)(t,). Next, we take the ansatz $\chi_t=-\frac12\ln t-\frac{2 c}{\sqrt{t}}$, and introduce the variable $z=\frac{\xi^2}{4t}$. The function $g(t,z)$ is $\psi(t,\xi)$ expressed with the help of $z$, and we look for solutions in the form g(t,z)=g\_[-12]{}(z)+g\_0(z). \[eq:defg\] We are led to the following hierarchical set of equations (compare to Eq. (\[eq:hierarchy\])):
&zg\_[-12]{}\^(z)+(z+12)g\_[-12]{}\^(z)=0\
&zg\_0\^(z)+(z+12)g\_0\^(z)+12 g\_0(z) =c g\_[-12]{}(z)+12 g\_[-12]{}\^(z).
The solution reads
&g\_[-12]{}(z)=b+a()\
&g\_0(z)= +2c+ az e\^[-z]{}\_2F\_2(1,1;[32]{},2;z)
\[eq:solgen\] where $\operatorname{erf},\operatorname{erfi}$ are the error functions defined by (x)=\_0\^x dt e\^[-t\^2]{}, (x)=-i (ix), and $a,b,c_1,c_2$ are integration constants to be determined. The terms in the first square brackets in Eq. (\[eq:solgen\]) correspond to the general solution of the homogeneous equation for $g_0$, while the next two terms represent a particular solution of the full equation as can easily be checked.
#### Matching conditions.
Because of the initial condition, the tail of the front at $\xi\rightarrow \infty$ has the exact shape (t,x2t1)=e\^[-(x-2t)]{} at any time. In particular, there is no overall constant. Comparing to Eqs. (\[eq:defpsi\]),(\[eq:defg\]), this condition means that g(t,z)=+2c+[O]{}(1/). Let us expand our solution (\[eq:solgen\]) for $g(t,z)$ for large $t,z$: $$\begin{gathered}
g(t\rightarrow\infty,z\rightarrow \infty)=
\sqrt{t}\left[b+a\sqrt{\pi}+{\cal O}(e^{-z})\right]
+2c(b+a\sqrt{\pi})\\
+\frac12(c_1+a\sqrt{\pi})\left(
\frac{1}{\sqrt{z}}+\frac{1}{2z^{3/2}}+\cdots
\right)
+{\cal O}(e^{-z}).\end{gathered}$$ The identification with the expected asymptotic form leads to the conditions: b+a=1, [2c]{}(b+a)=2c. The second condition is trivial once the first one is satisfied.
We also impose that all terms that are not exponentially suppressed cancel, which is realized by setting c\_1+a=0.
We turn to the limit $z\rightarrow 0$. The condition (\[eq:phiinterior\]) reads, in terms of the $g$-function g(t,z0)=+\[eq:match0\] which in particular forbids constant terms and terms proportional to $\sqrt{z}$. Since the small-$z$ expansion of our solution reads g(t,z0)=\
+2cb+c\_2+(4ac+c\_1)+[O]{}(z) we see that $b$ needs to be set to 0 and $c=-c_1/(4a)$.
Putting everything together, we find that all constraints are solved by the choice a=, b=0, c=, c\_1=-1, c\_2=. Note that the coefficient $\alpha$ in Eq. (\[eq:match0\]) is determined to be $\alpha=2/\sqrt{\pi}$, while in the noncritical case, it is a free parameter.
#### Matched solution.
All in all, our solution reads (t,x)= e\^[-]{}, \[eq:phi\] with =x-2t+t+ , z=. The $1/\sqrt{t}$ term is identical to the one in the “pushed front” calculation of Ref. [@EVS:2000], see Appendix G, Eq. (G18) therein, although the front solution chosen in that work is different, see Eq. (G7).
We can now deduce from this calculation the average value $\mu_1$ of $\ln Y_t-\ln Y_{t_0}$ by expanding the exact formula Eq. (\[eq:MomentsFromPhi\]) in powers of $\nu$ and keeping the coefficient of the term of order $\nu$: Y\_t=Y\_t+2t-12t=-(1)-\_[-]{}\^[+]{}dxx (t,x). \[eq:logYfromphi\] We find \_1=Y\_t-Y\_[t\_0]{}=( -). Identifying the latter equation to Eq. (\[eq:mu1\]) and taking into account the value of $C_{\bar Y}$ already computed in Eq. (\[eq:CMFY\]), we finally obtain a determination of $CC_1$: CC\_1=. \[eq:CC1\]
\[sec:numerics\]Complete results and numerical checks
=====================================================
Since the new results we have obtained rely in an essential way on a model for fluctuations and hence on a set of conjectures, we need to check them with the help of numerical simulations in order to get confidence in the validity of our picture. In the first part of this section, we shall list the formulas we have obtained but extending them to more general BRW models. Then, we define a model that is convenient for numerical implementation in Sec. \[sec:modelnum\], and we test our results against numerical simulations of this particular model in Secs. \[sec:checkdet\] and \[sec:checkstat\].
Parameter-free predictions for a general branching diffusion
------------------------------------------------------------
We now extend our results to general branching diffusion kernels. In the continuous case, we write the equation for the average particle density as \_t[n(t,x)]{}= (-\_x)n(t,x)where $\chi(-\partial_x)$ is the operator that represents the branching diffusion. The eigenfunctions are the exponential functions $e^{-\gamma x}$, and the corresponding eigenvalues are $\chi(\gamma)$. In the case discussed in the previous sections, $\chi(-\partial_x)=\partial_x^2+1$ and $\chi(\gamma)=\gamma^2+1$. We introduce $\gamma_0$ which solves $\chi^\prime(\gamma_0)=\chi(\gamma_0)/\gamma_0$. Then in the case studied so far, $\gamma_0=1$ and $\chi(\gamma_0)=\chi^\prime(\gamma_0)=\chi^{\prime\prime}(\gamma_0)=2$.
We can also address the discrete time and space case, which is useful in particular for numerical simulations. We write =(-\_x)n(t,x)where now $\delta_x$ is some finite difference operator, such as \_x f(x)=. In this case, $t$ and $x$ take their values on lattices of respective spacing $\Delta t$ and $\Delta x$. Again, the eigenfunctions of the kernel are the exponential functions.
The generalization of our previous results to an arbitrary BRW relies on the fact that at large times, the “wave number” $\gamma_0$ dominates and the kernel eigenvalue $\chi(\gamma)$ may be expanded to second order around $\gamma_0$ [@VS:2003]. We then essentially use dimensional analysis to put in the appropriate process-dependent factors. We list here the generalized expressions without detailed justifications.
With the general kernel, the FKPP front position reads (see Eq. (\[eq:FKPPpos\])) X\_t=’(\_0)t-t+ -+The position of the tip of the front in the mean-field model with a discreteness cutoff reads instead |X\_t=’(\_0)t-t+ ++This expression generalizes Eq. (\[eq:Xt\]) with $C_X$ computed in Sec. \[sec:noncritical\] (see Eq. (\[eq:CMFX\])).
The relevant variable that characterizes the fluctuations of the position of the bulk of the particles is $\frac{1}{\gamma_0}\ln Y_t$. We have computed its value in the deterministic model with a tip cutoff: = + . \[eq:logYdettip\] This equation generalizes Eq. (\[eq:CMFY\]).
The stochasticity that we found tractable analytically is related to the fluctuations of the difference of this variable at two distinct large times $t_0$ and $t$: f= ( Y\_t-Y\_[t\_0]{} ). Its first moment reads \_1=f = ( - ). \[eq:mu1fulldet\] The probability distribution of the fluctuations reads p(f)=
& ,\
& .
\[eq:pf\] This formula is the generalized form of Eqs. (\[eq:pf1\]) and (\[eq:pf2\]). A generating function of the moments of order larger than 2 can be written as e\^[\_0f]{} = { -(-)+()+ }. For example, expanding this generating function, we find that the moments of order $k\geq 2$ read \_k= m\_k, \[eq:muk\] where the $m_k$’s are numerical constants. The first ones read
m\_2&= -2+\^2 2\
m\_3&= 32(3)+ -2+\^2 2+\^3 2\
m\_4&= 3(22-)(3) + -2 -\^2 2 -2\^3 2 +\^4 2
or in numbers, $m_2=4.06013\cdots$, $m_3=6.56570\cdots$, $m_4=26.9902\cdots$.
\[sec:modelnum\]Model suitable for a numerical implementation
-------------------------------------------------------------
For simplicity of the implementation, we considered a discretized branching diffusion model. At each time step, a particle on lattice site $x$ (with lattice spacing $\Delta x=1$) has the probability $\Delta t$ to give birth to another particle on the same site, $\Delta t$ to move to the site $x+1$, $\Delta t$ to move to the site $x-1$, and $1-3\Delta t$ to stay unchanged at the same site. The eigenfunctions of the corresponding diffusion kernel are the exponential functions $e^{-\gamma x}$, and the eigenvalues read ()=. The discretization in time is chosen to be $\Delta t=0.01$. The relevant parameters for this model are $$\gamma_0=0.91338\cdots\ ,\ \ \chi^\prime(\gamma_0)=2.05412\cdots
\ ,\ \ \chi''(\gamma_0)=2.79893\cdots
\label{eq:numbers}$$
\[sec:checkdet\]Check of the deterministic analytical results
-------------------------------------------------------------
We solve the equivalent of the deterministic FKPP equation with the critical initial condition. For our discretized model, the FKPP equation becomes the finite difference equation l\_[x+1]{}(t+t)=l\_x(t)+{1+t} with the initial condition $l_x(t=0)=\ln\left[1-\exp(-e^{-\gamma_0 x})\right]$. Here $x$ is an integer that labels the sites of the lattice. $l_x(t)$ is the logarithm of the equivalent of $\phi$ defined in Sec. \[sec:det\]. The use of a logarithmic variable avoids problems with numerical accuracy in the region $\phi\rightarrow 0$, upon which the solution depends crucially.
First, we integrate the solution according to Eq. (\[eq:logYfromphi\]) in order to get $\frac{1}{\gamma_0}\langle\ln Y_t\rangle$ . The analytical expectation for the model which is implemented is given in Eq. (\[eq:mu1fulldet\]) with the numerical inputs (\[eq:numbers\]): Y\_t=-. The numerical calculation is shown in Fig. \[fig:plotpos\], and is in perfect agreement with the analytical formula. In order to estimate more quantitatively the quality of this agreement, we fit a function of the form f(t)=c\_0++ + , \[eq:fit\] where the $c$’s are the free parameters. The value of $c_{\frac12}$ which we get from the fit is $c_{\frac12}=0.8918$, which is very close to the expected value from our analytical calculation.
![\[fig:plotpos\] $\frac{1}{\gamma_0}\langle \ln Y_t\rangle$ from the numerical solution of the FKPP equation with the “critical” initial condition, as a function of $1/\sqrt{t}$. (The constant term is subtracted.) One sees that it converges to the analytical result Eq. (\[eq:mu1fulldet\]) (with $t_0\rightarrow +\infty$; straight line) for $t\rightarrow +\infty$.](plotpos.eps){width=".8\textwidth"}
Next, we solve the deterministic FKPP equation with a tip cutoff. In practice, the latter cutoff is implemented as a smooth nonlinearity, as in Eq. (\[eq:tipcutoffsmooth\]). More precisely, the equation we solve numerically is the following: l\_x(t+t)=l\_x(t) +{1 +t} with $l_x(t=0)=-|x|$. Here, $l_x(t)$ is the logarithm of the number of particles on site $x$ at time $t$. The logarithmic scale for the evolved function is useful here because of the exponential growth of the number of particles with time. Also in this case, the result is in excellent agreement with the analytical expectation (see Fig. \[fig:plotcutoff\]), which, for the considered model, should read (see Eq. (\[eq:logYdettip\])) = + The fit of the same function $f(t)$ as before to the numerical data gives $c_{\frac12}=2.7120$ which, again, is very close to the analytical estimate.
![\[fig:plotcutoff\] $\frac{1}{\gamma_0}{\ln \bar Y_t}$ from the numerical solution of the branching diffusion equation with a cutoff as a function of $1/\sqrt{t}$. (The constant is subtracted.) Again, the numerical solution converges to the analytical result (Eq. (\[eq:logYdettip\]); straight line) as $t$ gets large. ](plotcutoff.eps){width=".8\textwidth"}
\[sec:checkstat\]Check of the statistics of $f$
-----------------------------------------------
We now use a Monte-Carlo implementation of the stochastic model of a branching random walk described above in order to test the probability distribution of $f$ given in Eq. (\[eq:pf\]).
The implementation is quite straightforward, except maybe that after a few timesteps, the number of particles $n_x$ in the central bins (typically $|x|\leq \chi^\prime(\gamma_0) t$) becomes very large. To handle such large particle numbers, we further evolve these bins in a deterministic way. (In practice in the code, we set the limit between stochastic and deterministic evolution at $n_x=10^6$.) Such an approximate treatment was tested before in a similar context, see e.g. Refs. [@Moro:2004; @BD:1999; @BD:2001]. As in the deterministic case discussed above, we also switch to logarithmic variables, $l_x\equiv\ln n_x$, in order to be able to handle the large particle numbers in a standard double-precision computer representation. Of course, the low-density tails of the system are treated fully stochastically.
The result for the distribution of $f$ is displayed in Fig. \[fig:plotdistributionf\] compared to the analytical formulas (\[eq:pf\]). We see an excellent agreement between the outcome of our model and the numerical data.
![\[fig:plotdistributionf\] Distribution of $f$ for $t=1000$ and two different values of $t_0$. The numerical data (points with statistical error bars and bin width) are compared to Eq. (\[eq:pf\]) (continuous lines) ($\log_{10}$ scale on the vertical axis). ](plot_p.eps){width=".95\textwidth"}
We can also compute numerically the first few moments of the variable $f$ and plot them against $t_0$ (Fig. \[fig:plotmoments\]).
![\[fig:plotmoments\] Moments of $f=\ln Y_t/Y_{t_0}$ of order 2 to 4 for $t=500$ as a function of $t_0$ in logarithmic scales. The numerical data (full lines) are compared to the analytical calculation in Eq. (\[eq:muk\]) (dashed lines) ($\log_{10}$ scale on both axes). There are about $4\times 10^5$ realizations in the statistical ensemble used to perform the averages. ](plot_moments.eps){width=".95\textwidth"}
Here again, there is a good agreement between the analytical result and the numerical calculation, although more statistics would be needed in order to reach a good accuracy for the moments of order 3 and 4.
\[sec:EVS\]Stochastic interpretation of the $1/\sqrt{t}$ corrections to the position of FKPP fronts
===================================================================================================
So far, we have essentially discussed the statistics of the $\ln Y$ random variable in the light of our phenomenological picture of BRW. We are now going to address the average of the position of the rightmost particle $\langle X_t\rangle$ which is also the position of the FKPP front, and whose expression at order $1/\sqrt{t}$ was obtained in Ref. [@EVS:2000].
Correction to $\langle X_t\rangle$ due to fluctuations
------------------------------------------------------
As in the case of $\langle f\rangle$, in our picture, the average value of the position of the front is given by the deterministic evolution of the bulk of the particles, supplemented by a contribution from fluctuations in the low-density region. We may write \_1\^X\_t-X\_[t\_0]{}= 2(t-t\_0)-32 +C\_[|X]{} (-) +\_1\^[+]{}-\_1\^[-]{}. \[eq:c1prime\] In this section, we shall compute $\mu_1^{\prime +}-\mu_1^{\prime -}$. $\mu_1^{\prime +}$ is the contribution at time $t$ of fluctuations that occur all over the range of time and $\mu_1^{\prime -}$ is the contribution at time $t_0$ of fluctuations that have occurred before $t_0$: \_1\^[+]{}=\_0\^t dt\_1\_0\^[+]{} dX\_t , \_1\^[-]{}=\_0\^[t\_0]{} dt\_1\_0\^[+]{} dX\_[t\_0]{} , where the appropriate regulators will be introduced later. $\delta X_t$ is the contribution to the shift of the position of the tip of the front at time $t$ of a fluctuation of size $\delta$ occurring at $t_1$. Let us now evaluate $\delta X_t$.
When a fluctuation occurs at time $t_1$ at position $\delta$ ahead of the tip $\bar X_{t_1}$ of the regular front, then it develops its own front by independent branching diffusion. The resulting density of particles at time $t\gg t_1$ becomes the sum of two terms, and therefore has the shape \_[|X\_t+X\_t]{}(x,t)=\_[|X\_t]{}(x,t)+C\_[|X\_[,t]{}]{}(x,t) where $\psi$ is given by Eq. (\[eq:shape\]) and $\bar X_{\delta,t}$ by Eq. (\[eq:Xdelta\]). Using the latter equations, one is led to X\_t=.
The calculation of $\mu_1^{\prime +}$ and $\mu_1^{\prime -}$ proceeds exactly as in the case of $\mu_1^+$ and $\mu_1^-$ in Sec. \[sec:mu1\]. $\mu^{\prime+}_-$ is still given by an equation of the form of (\[eq:mu1+\]), but with the replacements ${\cal I}_0\rightarrow {\cal I}^\prime_0$ and ${\cal I}_1\rightarrow {\cal I}^\prime_1$, where now \_0\^=\_[t\_0\^/t]{}\^[1-|t\_0/t]{} d\^[-3/2]{}(1-)\^[-3/2]{} , \_1\^=\_[t\_0\^/t]{}\^[1-|t\_0/t]{} d\^[-3/2]{}(1-)\^[-3/2]{} . Note that in the present case, late times need to be cutoff in order to ensure the convergence of the integrals: We pick some arbitrary $\bar t_0\ll t$ say of order one.
The same change of variable as before may be used: $\lambda=\sin^2\theta$, then \_0\^=8\_\^ , \_1\^=24\_\^ . After performing the integrals and expanding in the limit of small $\bar t_0$, $t_0^\prime$ compared to $t$, one gets \_1\^[+]{}=2CC\_1{ (+3) +(+3) - }. The main difference with respect to Eq. (\[eq:c1+\]) (once the relevant expansions have been performed) is the presence of $\bar t_0$-dependent terms and of an extra factor 3 in the last term. As before, $\mu_1^{\prime-}$ is deduced from the above formula by replacing $t$ by $t_0$. We then see that in the difference $\mu_1^{\prime+}-\mu_1^{\prime-}$, the $\bar t_0$ and $t_0^\prime$ dependences cancel.
As for the moments of order $n\geq 2$, they are found to depend on $\bar t_0$, that is, on the late-time fluctuations, as they should, since $X_t$ is the position of the rightmost particle, which experiences a stochastic motion of size 1 over time scales of order 1.
Recovering the Ebert-Van Saarloos term
--------------------------------------
Putting everything together, namely the value of $C_{\bar X}$ from Eq. (\[eq:CMFX\]) and the value of $\mu_1^{\prime+}-\mu_1^{\prime-}$ just computed, we find the interesting expression X\_[t]{}-X\_[t\_0]{}= \_1\^= 2(t-t\_0)-32+ -. The terms that grow with $t$ and $t_0$ are the usual deterministic terms from Bramson’s classical solution [@B:1983]. Then, the next terms, under the square brackets, are respectively the deterministic correction to the position of the discreteness cutoff in the mean-field model, and the correction due to fluctuations. We see that the latter is exactly twice the former, with a minus sign. The sum of these two terms gives back the Ebert-Van Saarloos correction for $\left\langle X_t-X_{t_0}\right\rangle$, see Eq. (\[eq:FKPPpos\]).
In other words, the mismatch between $\bar X_t$, the position of the tip of the front in the deterministic model with a cutoff, and $\langle X_t\rangle$, the mean position of the rightmost particle in the full stochastic model, is exactly due to the very fluctuations we have been analyzing in this paper.
\[sec:conclusion\]Conclusions
=============================
Some time ago, we proposed a model for the fluctuations of stochastic pulled fronts [@Brunet:2005bz], which are realizations of the stochastic FKPP (sFKPP) equation (for a review, see Ref. [@Panja:2004]). Equations in the class of the sFKPP equation may be thought of, for instance, as representing the dynamics of the particle number density in a branching-diffusion process in which there is in addition a nonlinear selection/saturation process that effectively limits the density of particles. The realizations of such equations are stochastic traveling waves. The stochasticity comes from the discreteness of the number of particles. In this context, the (deterministic) FKPP equation represents the mean-field (or infinite number of particle) limit of the full dynamics.
Expansions about the mean-field solution were considered already a long time ago; see, e.g., Ref. [@BHP:1994] where the so-called $\Omega$ expansion (see Ref. [@VanKampen]) was applied to study fluctuations in the context of reaction-diffusion processes. Later, we could obtain new analytical results thanks to a phenomenological model [@Brunet:2005bz]. The picture encoded in our model was the following: Most of the time, the traveling wave front propagates deterministically, obeying the ordinary deterministic FKPP equation supplemented with a cutoff in the tail, accounting for discreteness by making sure that the number density of particles reaches 0 rapidly whenever it drops below 1. Brunet and Derrida had shown [@BD:1997] that such a cutoff correctly represents the main effect of the noise on the velocity of the front. On top of that, in our model, there are some rare fluctuations consisting in a few particles randomly sent far ahead of the tip of the front, which upon further evolution build up a new front that completely takes over the old one. A positive correction to the front velocity was found, and the cumulants of the front position were computed (see Ref. [@Brunet:2005bz]).\
In the present paper, we have considered a simple branching random walk, without any selection mechanism. We have used exactly the same ingredients as the ones conjectured in the model for stochastic fronts, namely deterministic evolution with a cutoff and fluctuations consisting in a few particles randomly sent ahead the tip of the front at a distance distributed exponentially. We were also able to arrive at a quantitative characterization of the fluctuations of the front in these processes.\
There are however a few important differences between the branching random walk and the stochastic FKPP front. First, the initial fluctuations are never “forgotten” in the BRW case. This is because of the absence of a selection mechanism able to “kill” the front and let it be periodically regenerated by fluctuations. Therefore, we could only compute the effect of the fluctuations on the front position between two large times $t_0$ and $t$. Next, while it was quite straightforward to define a proper front position in the sFKPP case (as for example the integral of the normalized particle density from position say 0 to $+\infty$), it is more tricky for the simple branching random walk. We were led to consider the variables $\ln Y$ and $\ln Z$ (introduced in Sec. \[sec:var\]). Our main result is the distribution of the variable $\ln Y_t/Y_{t_0}$ given in Eq. (\[eq:pf\]), where $t_0$ and $t$ are two large times such that $t_0,t,t-t_0\gg 1$. Interestingly enough, the distribution of the positive values of this variable is identical (up to an overall factor) to the distribution of the front fluctuations in the sFKPP case. The same holds true for the distribution of $\ln Z_t/Z_{t_0}$ to which we dedicate Appendix \[sec:Z\].
We were also able to discuss the average of the position of the rightmost particle, but not its higher moments since they are sensitive to the very late-time fluctuations which are not properly described in our model. As for the average position, we could nevertheless propose an appealing interpretation of the ${\cal O}(1/\sqrt{t})$ correction to the front position computed by Ebert and Van Saarloos in Ref. [@EVS:2000].\
There are still many open questions. Maybe the most outstanding one on the technical side would be to try and compute the statistics of $\ln Y_t/Y_{t_0}$ (and of $\ln Z_t/Z_{t_0}$) exactly, instead of relying on a phenomenological picture involving conjectures. We outlined such a calculation in Appendix \[sec:exact\], based on the evaluation of a generating function, but without being able to complete it.
Acknowledgements {#acknowledgements .unnumbered}
================
The idea at the origin of Appendix \[sec:exact\] is due to Prof. B. Derrida. We thank him also for very helpful discussions, and for his reading of the manuscript. We acknowledge support from “P2IO Excellence Laboratory”, and from the US Department of Energy, Grant No. DE-FG02-92ER40699.
\[sec:details\] Details of the calculation of the probability distribution of $f$
=================================================================================
In this appendix, we go back to the calculations that lead to Eqs. (\[eq:shift\]), (\[eq:pf1\]) and (\[eq:pf2\]), but keeping the subleading terms that we neglected a priori in Sec. \[sec:distrib1\] in order to simplify the presentation.
The exact evaluation of $\bar Y_t$ starting from its definition given in Eq. (\[eq:Y0\]), in which one inserts Eqs. (\[eq:shape\]), (\[eq:Xt\]), makes use of the basic Gaussian integral \_[-]{}\^0 dx(x+)e\^[-]{}=-2t+. We immediately arrive at Eq. (\[eq:Y0\]), which may also be rewritten at order $1/\sqrt{t}$ as |Y\_t=-2e\^. We now add a fluctuation occurring say at time $t_1$. It develops a front whose tip sits, at time $t$, at position |X\_[,t]{}=|X\_t+-32 +C\_[|X]{}( +- ), which is Eq. (\[eq:Xt\]) supplemented with the subleading terms. Keeping all the latter, we see that Eq. (\[eq:shift\]) just needs to be replaced by Y\_t=. As for the probability distribution of the fluctuations in Eq. (\[eq:pf01\]), it becomes p(f;t\_1)= e\^[+C\_[|Y]{}(-)]{} , which has to be integrated over $t_1$. We recall that after integration over $t_1$, the obtained expression will be correct at order $1/t_1$, $1/(t-t_1)$, $1/t$, hence only the first nontrivial terms are relevant in the expansion of the exponential.
In the absence of ${\cal O}(1/(t-t_1))$ terms, the integration region could be chosen to be $[t_0,t]$ as in Sec. \[sec:distrib\]. Now however we have a non-integrable singularity at $t_1=t$ which needs to be cut off. Hence we write p(f)=\_[t\_0]{}\^[t-[|t\_0]{}]{}dt\_1 p(f;t\_1) =CC\_1 e\^[-]{} \_[t\_0]{}\^[t-[|t\_0]{}]{} e\^[+ ]{} where ${\bar t_0}$ is an arbitrary time interval whose length is of the order of 1.
Let us compute the integral \_[t\_0]{}\^[t-[|t\_0]{}]{} e\^[+ ]{} appearing in the previous expression. We expand the exponential to lowest order, and hence we get the three terms = [J]{}\_0+C\_[|X]{}[J]{}\_[1]{}\^[(1)]{}+C\_[|Y]{}[J]{}\_[1]{}\^[(2)]{}, where ${\cal J}_0$ (which is essentially the same integral as ${\cal I}_0$ in Eq. (\[eq:defI0\])) gives back the lowest-order result in Eq. (\[eq:pf1\]): \_0=\_[t\_0]{}\^[t-[|t\_0]{}]{} =2-2 2+[O]{}(1/t). \[eq:I0exp\] As for the two other terms, \_[1]{}\^[(1)]{}= \_[t\_0]{}\^[t-[|t\_0]{}]{},\_[1]{}\^[(2)]{}= \_[t\_0]{}\^[t-[|t\_0]{}]{} are new contributions which are subleading, as is easy to demonstrate from an exact calculation of these integrals. We start with the computation of ${\cal J}_{1}^{(1)}$: \_[1]{}\^[(1)]{}= ( + ) - ( + ). Since $\operatorname{arctanh}\sqrt{1-x}\underset{x\rightarrow 0}{\sim}-\frac12\ln x$, it is clear that the largest terms in ${\cal J}_{1}^{(1)}$ are at most of order $\ln (t/t_0)/t$ and $1/t_0$. As for ${\cal J}_{1}^{(2)}$, \_[1]{}\^[(2)]{}= 2( - ) + ( - ) The second term is divergent for ${\bar t_0}\rightarrow 0$. It gives the dominant contribution at large $t$: ${\cal J}_{1}^{(2)}\sim \ln (t/{\bar t_0})/t$. The other terms are also subleading, of order $1/\sqrt{t t_0}$ and $1/t$.
Hence we see that at order ${\cal O}(1/\sqrt{t},1/\sqrt{t_0},1/\sqrt{t-t_0})$, ${\cal J}$ boils down to the first term in the expansion of ${\cal J}_0$ in Eq. (\[eq:I0exp\]).
Lastly, we have already noticed in Sec. \[sec:distrib\] that f-f=C\_[|Y]{} (-), thus replacing $\delta f$ by $f$ in $p(\delta f)$ brings about only subleading contributions.
All in all, we have justified the approximations that led to Eq. (\[eq:pf1\]). From a very similar calculation, we would also recover Eq. (\[eq:pf2\]).
\[sec:Z\]Statistics of $f_Z\equiv\ln Z_t-\ln Z_{t_0}$
=====================================================
In the same way as for the variable $f=\ln Y_t-\ln Y_{t_0}$, we may try to get the statistics of $f_Z=\ln Z_t-\ln Z_{t_0}$ from our phenomenological model. The variable $Z_t$ is of interest since it is used in a mathematical theorem to characterize what we call the position of the front in each realization, however, as we shall see, we cannot obtain full analytical formulas for the first moment of $f_Z$ as in the case of $f$. Moreover, as was already commented above, the $Z_t$ variable has properties that make it awkward for numerical simulations.
The first step is to compute $Z_t$ in the mean-field approximation with a tip cutoff. The result reads |Z\_t=-2(1+). This formula is analogous to Eq. (\[eq:Y0\]), but there is now a slightly stronger $t$-dependence, $\propto\ln t/\sqrt{t}$, which we are able to determine completely from the leading-order shape of the particle distribution. We have dropped terms of order $1/\sqrt{t}$ and higher.
The effect of a fluctuation occurring at time $t_1$ on $\ln Z_t$ is Z\_t=\[eq:shiftZ\] (Compare to Eq. (\[eq:shift\])). The following approximate formula can now be written for the distribution of $f_Z$: p(f\_Z)=dt\_1\_0\^[+]{}dp(), where $\delta \ln Z$ is given by Eq. (\[eq:shiftZ\]), while $p(\delta)$ is the probability distribution (\[eq:proba\]). The bounds on the integral over $t_1$ depend on whether $f_Z$ is positive or negative. Indeed, positive values of $f_Z$ are generated by fluctuations occurring at $t_1$ between $t_0$ and $t$, while fluctuations before $t_0$ (namely between the times $t_0^\prime$ at which we declare that the system contains a large number of particles and $t_0$) give rise to negative values of $f_Z$.
The distribution of positive $f_Z$ is quite easy to compute. It is enough to recognize that the terms of order $1/\sqrt{t-t_1}$ and $1/\sqrt{t}$ inside the square bracket give subleading contributions to $p(f_Z)$. It turns out that the final result is very similar to $p(f)$ (see Eq. (\[eq:pf1\])), except for the detailed form of the $t_0$ and $t$ dependence: p(f\_Z>0)=2CC\_1 ( - ) . Inserting the value of the constant $CC_1$ previously determined (see Eq. (\[eq:CC1\])) and going to a general branching diffusion kernel, we get p(f\_Z>0)= ( - ) . \[eq:pfZ\]
Negative values of $f_Z$ are more complicated to deal with since we can no longer neglect the $1/\sqrt{t-t_1}$ term in Eq. (\[eq:shiftZ\]) a priori. Performing the change of variable $u\equiv t_1^{3/2}/(C e^\delta)$ and expanding for large $t$ and $t_0$, the equation for $p(f_Z)$ simplifies to p(f\_Z<0)=CC\_1\_[t\_0\^]{}\^[t\_0]{} \_0\^ du . Due to the Dirac $\delta$-function, we see that $p(f_Z)=0$ as soon as $f_Z<\frac{1}{\sqrt{\pi} u_0}\left(\frac{1}{\sqrt{t}}-\frac{1}{\sqrt{t_0}}\right)$, where $u_0$ solves $\ln\left(Cu_0\right)=1+\frac{1}{u_0}$, and hence is of order 1. This means that $p(f_Z)$ is of higher-order in powers of $1/\sqrt{t}$ and $1/\sqrt{t_0}$ when $f_Z<0$.
Our formula for the distribution, Eq. (\[eq:pfZ\]), successfully compares to the numerical data, see Fig. \[fig:plotdistributionfZ\]. We also see that the distribution of negative values of $f_Z$ is indeed sharply suppressed (compare to the distribution of $f$ in Fig. \[fig:plotdistributionf\]).\
![\[fig:plotdistributionfZ\] Distribution of $f_Z$ for $t=1000$ and two different values of $t_0$. The numerical data (points with statistical error bars and bin width) are compared to Eq. (\[eq:pfZ\]) (continuous lines). ($\log_{10}$ scale on the vertical axis). ](plot_pZ.eps){width=".95\textwidth"}
As for the mean of $f_Z$, we found that it depends on the arbitrary time $t_0^\prime$ roughly as $1/\sqrt{t_0^\prime}$, and thus is not calculable.
\[sec:exact\]Generating function for the moments
================================================
In this section, we are going to find the form of the large positive $f$-fluctuations from a generating function, hence from a deterministic calculation.
General framework and exact formulas
------------------------------------
We can write the following identity: ()\^= \_0\^[+]{} [du]{} [dv]{} ()\^ e\^[-u Y\_t-v Y\_[t\_0]{}]{} , \[eq:momentsexacts\] see Eq. (\[eq:defYt\]) for the definition of $\tilde Y$. This equation follows from the integral representation of the $\Gamma$ function, and is suitable for series expansions in $\nu$, which eventually lead to the moments of $\ln \tilde Y_t/\tilde Y_{t_0}$. For some calculations outlined below, it will prove useful to change $\tilde u$ and $\tilde v$ to the variables u=u , v=v since Eq. (\[eq:momentsexacts\]) then holds in the very same form (just up to the replacements $\{\tilde u,\tilde v,\tilde Y_t,\tilde Y_{t_0}\}\rightarrow
\{u,v,Y_t,Y_{t_0}\}$) directly for the moments of $\ln Y_t/Y_{t_0}=f$.
Let us introduce the generating function G\_[t\_0]{}(x)=e\^[-(u Y\_t+v Y\_[t\_0]{})e\^[-x]{}]{} . It is the value of this function at zero, $G_{t_0}(0)$, from which one computes the generating function in Eq. (\[eq:momentsexacts\]), which reads ()\^= \_0\^[+]{} [du]{} [dv]{} ()\^. \[eq:momentsexacts2\] The function $G_{t_0}(x)$ may also be written as G\_[t\_0]{}(x)=\_[i=1]{}\^[N(t\_0)]{} g\_(x-x\_i(t\_0))e\^[-v e\^[-(x-x\_i(t\_0))]{}]{} , \[eq:Gexact\] where $\tau\equiv t-t_0$ is a parameter in this equation, and g\_(x)e\^[-uY\_e\^[-x]{}]{}. In this form, it is clear that $G_{t_0}(x)$ obeys the FKPP equation (with time variable $t_0$), with the initial condition $g_\tau(x)e^{-\tilde v e^{-x}}$. But $g_\tau(x)$ may also be written as g\_(x)=\_[i=1]{}\^[N()]{}e\^[-u e\^[-(x-x\_i())]{}]{}, which makes it obvious that it also obeys the FKPP equation (with time variable $\tau$), with the initial condition $g_0(x)=e^{-\tilde u e^{-x}}$.
So far, these formulas are exact and should in principle enable the computation of the moments of $f$, from some hopefully limited knowledge of the properties of the solutions to the FKPP equation.
We have not been able to fully compute the generating function. However, we can use the systematic solution to FKPP for the evolution of $g$, and a mean-field approximation for $G$: Interestingly enough, this turns out to be enough to compute the positive fluctuations of $f$.
Approximate solution: Moments of $f>0$
--------------------------------------
In this section, we shall consider the stronger limit $t\gg t_0\gg 1$.
Let us treat the evolution from the initial time $t=0$ to time $t_0$ in the mean-field approximation with a tip cutoff: This means that we assume a distribution of particles at time $t_0$ given by Eq. (\[eq:shape\]). Then the product over the particles in Eq. (\[eq:Gexact\]) becomes the exponential of an integral over the spatial coordinate weighted by the particle density: G\_[t\_0]{}(x)=, \[eq:Gapprox\] where $\bar X_{t_0}=2t_0-\frac32\ln t_0$. We have dropped the $\beta$ term in the form of the particle distribution as well as the $1/\sqrt{t_0}$ term in $\bar X_{t_0}$ since they would eventually give subleading contributions, of order $1/t_0$, at large $t_0$.
We see that the Gaussian under the integral makes sure that the range of integration in the variable $x^\prime-\bar X_{t_0}$ is effectively $[-2\sqrt{t_0},0]$.
We turn to the $g_{t-t_0}$. We know that it obeys the FKPP equation with the critical initial condition. Hence the solution can be deduced from Eq. (\[eq:phi\]). However, since $t_0\ll t$, defining $\xi=x-\ln u-2(t-t_0)+\frac12\ln(t-t_0)$, we may expand the solution for $1\ll \xi\ll\sqrt{t-t_0}$, namely 1-g\_[t-t\_0]{}() e\^[-]{}. We have dropped the term of order $1/\sqrt{t-t_0}$ in $\xi$.
We shall now proceed with the integration in Eq. (\[eq:Gapprox\]). Keeping only the term of order $1/\sqrt{t_0}$ and switching to the $u$, $v$ variables, we find G\_[t\_0]{}(0)=e\^[-2(u+v)]{}( 1+u ). Inserting this expression into Eq. (\[eq:momentsexacts2\]) (with $\tilde u$, $\tilde v$ being replaced by $u$, $v$), we now perform the integrals over $u$ and $v$. The exact result is e\^[f]{} = 1+ . Remarkably, if we invert this equation for the probability distribution of $f$ by performing an appropriate contour integration over $\nu$, we exactly recover Eq. (\[eq:pf\]) for the case $f>0$ (in the limit $t\rightarrow+\infty$, and up to replacements of the parameters in (\[eq:pf\]): $\gamma_0\rightarrow 1$, $\chi^{\prime\prime}(\gamma_0)\rightarrow 2$). Note that the constant $CC_1$ which appeared in the phenomenological model is determined without any further calculation in the present approach. The case $f<0$ however cannot be obtained unless we were able to release the mean-field approximation for the evolution between $t=0$ and $t=t_0$.
[0]{}
E. Iancu, A. H. Mueller and S. Munier, Phys. Lett. B [606]{}, 342 (2005). S. Munier, Phys. Rept. [473]{}, 1 (2009). B. Derrida, H. Spohn, J. Stat. Phys. 51, 817-840 (1988).
H. P. McKean, Commun. Pure Appl. Math. 28, 323 (1975).
M. D. Bramson, Mem. Am. Math. Soc. 44, 285 (1983).
R. A. Fisher, Ann. Eugenics [ 7]{}, 355 (1937). A. Kolmogorov, I. Petrovsky, and N. Piscounov, Moscou Univ. Bull. Math. [ A1]{}, 1 (1937). W. van Saarloos, Phys. Rep. 386, 29-222 (2003).
S. N. Majumdar, P.L. Krapivsky, Physica A 318, 161 (2003).
U. Ebert and W. van Saarloos, Physica D 146, 1-99 (2000).
S. P. Lalley, T. Sellke, Ann. Prob. 15, No. 3, 1052-1061 (1987).
E. Brunet, B. Derrida, Europhys. Lett. 87, 60010 (2009).
E. Brunet, B. Derrida, J. Stat. Phys. 143, 420-446 (2011). E. Aidekon, J. Berestycki, E. Brunet, Z. Shi, Probab. Theory Related Fields 157, 405 (2013).
E. Brunet, B. Derrida, Phys. Rev. E 56 2597-2604 (1997).
E. Brunet, B. Derrida, A. H. Mueller and S. Munier, Phys. Rev. E [ 73]{}, 056126 (2006). E. Moro, Phys. Rev. [ E69]{} (2004) 060101(R); Phys. Rev. [ E70]{} (2004) 045102(R).
E. Brunet, B. Derrida, Comp. Phys. Comm. [ 121-122]{} (1999) 376. E. Brunet, B. Derrida, J. Stat. Phys. [ 103]{} (2001) 269. D. Panja, Phys. Rep. 393, 87-174 (2004).
H.P. Breuer, W. Huber and F. Petruccione, Physica D 73 (1994) 259–273.
N. G. van Kampen, Stochastic Processes in Physics and Chemistry, Third Edition (North Holland, 2007).
[^1]: Corresponding author. Email: [[email protected]]{}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this paper we study a transport-diffusion model with some logarithmic dissipations. We look for two kinds of estimates. The first one is a maximum principle whose proof is based on Askey theorem concerning characteristic functions and some tools from the theory of $C_0$-semigroups. The second one is a smoothing effect based on some results from harmonic analysis and sub-Markovian operators. As an application we prove the global well-posedness for the two-dimensional Euler-Boussinesq system where the dissipation occurs only on the temperature equation and has the form $\frac{{|\textnormal{D}|}}{\log^\alpha(e^4+{|\textnormal{D}|})}$, with $\alpha\in[0,\frac12]$. This result improves the critical dissipation $(\alpha=0)$ needed for global well-posedness which was discussed in [@HKR].'
address: |
IRMAR, Université de Rennes 1\
Campus de Beaulieu\
35 042 Rennes cedex\
France
author:
- Taoufik Hmidi
---
Introduction
============
The first goal of this paper is to study some mathematical problems related to the following transport-diffusion model with logarithmic dissipations $$\label{trdif}
\left\{
\begin{array}{ll}
\partial_{t}\theta+v\cdot\nabla\theta+ \kappa \frac{{|\textnormal{D}|}^\beta}{\log^\alpha(\lambda+{|\textnormal{D}|})} \theta=0, (t,x)\in{\mathbb{R}}_+\times{\mathbb{R}}^d\\
\textnormal{div}\,v=0\\
\theta_{| t=0}=\theta_{0}.
\end{array} \right.$$ Here, the unknown is the scalar function $\theta$, the velocity $v$ is a time-dependent vector field with zero divergence and $\theta_0$ is the initial datum. The parameter $\kappa\geq0$, $\lambda>1$ and $\alpha,\beta\in{\mathbb{R}}.$ The operator $\frac{{|\textnormal{D}|}^\beta}{\log^\alpha(\lambda+{|\textnormal{D}|})}$ is defined through its Fourier transform
$
\mathcal{F}({\frac{{|\textnormal{D}|}^\beta}{\log^\alpha(\lambda+{|\textnormal{D}|})}f})(\xi)=\frac{|\xi|^\beta}{\log^\alpha(\lambda+|\xi|)}(\mathcal{F}f)(\xi)
.$
We will discuss along this paper some quantitative properties for this model, especially two kinds of information will be established: maximum principle and some smoothing effects. We notice that the special case of the equation corresponding to $\alpha=0$ and $\beta\in[0,2]$ appears naturally in some fluid models like quasi-geostrophic equations or Boussinesq systems. In this context A. Córdoba and D. Córdoba [@cc] established the [*a priori*]{} $L^p$ estimates: for $p\in[1,\infty]$ and $t\geq0$ $$\label{eq-max}
\|\theta(t)\|_{L^p}\le\|\theta_0\|_{L^p}.$$ We remark that the proof in the case $p=+\infty$ can be obtained from the following representation of the fractional Laplacian ${|\textnormal{D}|}^\beta$, $${|\textnormal{D}|}^\beta f(x)=c_d\int_{{\mathbb{R}}^d}\frac{f(x)-f(y)}{|x-y|^{d+\beta}}dy.$$ Indeed, one can check that if a continuous function reaches its maximum at some point $x_0$ then ${|\textnormal{D}|}^\beta f(x_0)\geq0$ and hence we conclude as for the heat equation. Our first main result si a generalization of the result of [@cc] to
\[max-princ0\] Let $\kappa\geq0, d\in\{2,3\},\beta\in]0,1],\alpha\geq0,\lambda\geq e^{\frac{3+2\alpha}{\beta}}$ and $ p\in[1,\infty]$. Then any smooth solution of satisfies $$\|\theta(t)\|_{L^p} \le \|\theta_0\|_{L^p}.$$
The restriction on the parameter $\beta$ is technical and we believe that the above theorem remains true for $\beta\in]1,2[$ and $\alpha>0.$
Let us discuss the proof in the special case of $v\equiv0.$ The equation is reduced to the fractional heat equation
$
\partial_{t}\theta+ \kappa\, \mathcal{L }\, \theta=0\quad\hbox{with}\quad\mathcal{L}:=\small{\frac{{|\textnormal{D}|}^\beta}{\log^\alpha(\lambda+{|\textnormal{D}|})}}\cdot
$
The solution is explicitly given by the convolution formula $$\theta(t,x)=K_t\star\theta_0(x)\quad\hbox{with}\quad \widehat{K_t}(\xi)=e^{-t \frac{|\xi|^\beta}{\log^\alpha(\lambda+|\xi|)}}.$$ We will show that the family $(K_t)_{t\geq0}$ is a convolution semigroup of probabilities which means that $\mathcal{L}$ is the generator of a Lévy semigroup. Consequently, this family is a $C_0$-semigroup of contractions on $L^p$ for every $p\in[1,\infty[.$ The important step in the proof is to get the positivity of the kernel $K_t$. For this purpose we use Askey’s criterion for characteristic functions, see Theorem \[rad1\]. We point out that the restrictions on the dimension $d$ and the values of $\beta $ are due to the use of this criterion. Now to deal with the full transport-diffusion equation we use some results from the theory of $C_0-$semigroups of contractions.
The second estimate that we intend to establish is a generalized Bernstein inequality. Before stating the result we recall that for $q\in{\mathbb{N}}$ the operator $\Delta_q$ is the frequency localization around a ring of size $2^q$, see next section for more details. Now our result reads as follows,
\[coer10\] Let $d\in\{1,2,3\},\beta\in]0,1],\alpha\geq0,\lambda\geq e^{\frac{3+2\alpha}{\beta}}$ and $ p\in]1,\infty[.$ Then we have for $q\in{\mathbb{N}}$ and $f\in \mathcal{S}({\mathbb{R}}^d),$
${
2^{q\beta}(q+1)^{-\alpha}\|\Delta_q f\|_{L^p}^p\le C{\displaystyle{\int}}_{{\mathbb{R}}^d} \Big(\frac{{|\textnormal{D}|}^{\beta}}{\log^{\alpha}(\lambda+{|\textnormal{D}|})}\Delta_qf\Big)\,|\Delta_q f|^{p-2}\Delta_qfdx.
}$
where $C$ is a constant depending on $p,\alpha,\beta$ and $\lambda$.
The proof relies on some tools from the theory of Lévy operators or more generally sub-Markovians operators combined with some results from harmonic analysis.
1. When $\alpha=0$ then the above inequality is valid for all $\beta\in[0,2].$ The case $\beta=2$ was discussed in [@dan; @plan]. The remaining case $\beta\in[0,2[$ was treated by Miao [*et al.*]{} in [@cmz] but only for $p\geq2.$
2. The proof for the case $p=2$ is an easy consequence of Plancherel identity and does not require any assumption on the parameters $ \alpha,\beta$ and $\lambda$.
The second part of this paper is concerned with an application of Theorems \[max-princ0\] and \[coer10\] to the following Boussinesq model with general dissipation $$\label{bintro7}
\left\{
\begin{array}{ll}
\partial_{t}v+v\cdot\nabla v+\nabla \pi=\theta e_{2},\,\, (t,x)\in{\mathbb{R}}_+\times{\mathbb{R}}^2 \\
\partial_{t}\theta+v\cdot\nabla\theta+ \kappa\, \mathcal{L}\,\theta=0\\
\textnormal{div}\,v=0\\
v_{| t=0}=v_{0}, \quad \theta_{| t=0}=\theta_{0}.
\end{array} \right.$$ Here, the velocity field $v$ is given by $v=(v^1,v^2)$, the pressure $\pi$ and the temperature $\theta$ are scalar functions. The force term $\theta e_{2}$ in the velocity equation, with $e_2$ the vector $(0,1)$, models the effect of the gravity on the fluid motion. The operator $\mathcal{L} $ whose form may vary is used to take into account anomalous diffusion in the fluid motion. From mathematical point of view, the question of global well-posedness for the inviscid model, corresponding to $ \kappa = 0,$ is extremely hard to deal with. We point out that the classical theory of symmetric hyperbolic quasi-linear systems can be applied for this system and thus we can get the local well-posedness for smooth initial data. The significant quantity that one need to bound in order to get the global existence is the $L^\infty$-norm of the vorticity defined by $\omega=
\mbox{curl } v = \partial_{1} v^2 - \partial_{2} v^1.$ Now we observe from the first equation of that $\omega$ solves the equation $$\label{vortintro}
\partial_{t} \omega + v \cdot \nabla \omega = \partial_{1} \theta.$$ The main difficulty encountered for the global existence is due to the lack of strong dissipation in the temperature equation: we don’t see how to estimate in a suitable way the However, the situation in the viscous case, $\kappa>0$ and $ \mathcal{L}= -\Delta$, is well-understood and the question of global existence is solved recently in a series of papers. In [@Cha], Chae proved the global existence and uniqueness for initial data $(v_0,\theta_0)\in H^s\times H^s,$ with $s>2,$ see also [@Hou]. This result was improved by the author and Keraani in [@hk] to initial data $v_0\in B_{p,1}^{\frac{2}{p}+1}$ and $\theta_0\in B_{p,1}^{-1+\frac2p}\cap L^r, r>2$. The global existence of Yudovich solutions for this system was treated in [@dp1]. We also mention that in [@dp], Danchin and Paicu constructed global strong solutions for a dissipative term of the form $ \partial_{11}\theta$ instead of $\Delta\theta.$ In the papers [@H-Z; @HKR] we try to understand the lower dissipation $\mathcal{L}={|\textnormal{D}|}^\alpha$ needed for global existence. In [@H-Z] the author and Zerguine proved the global well-posedness when $\alpha\in]1,2[.$ The proof relies on the fact that the dissipation is sufficiently strong to counterbalance the possible amplification of the vorticity due However the case $\alpha=1$ is not reached by the method and this value of $\alpha$ is called critical in the sense that the dissipation and the amplification of the vorticity due have the same order.\
In [@HKR] we prove that there is some hidden structure leading to global existence in the critical case. More precisely, we introduced the mixed quantity $\Gamma=\omega+\frac{\partial_1}{{|\textnormal{D}|}}\theta$ which satisfies the equation $$\partial_t\Gamma+v\cdot\nabla\Gamma=-[\mathcal{R},v\cdot\nabla]\theta.
\quad\hbox{with}\quad \mathcal{R}:={\frac{\partial_1}{{|\textnormal{D}|}}}.$$ As a matter of fact, the problem in the framework of Lebesgue spaces is reduced to the estimate the commutator between the advection $v\cdot\nabla$ and Riesz transform $\mathcal{R}$ which is homogenous of degree zero. Since Riesz transform is a Calderón-Zygmund operator then using in a suitable way the smoothing effects for $\theta$ we can get a global estimate of $\|\omega(t)\|_{L^p}$. One can then use this information to control more strong norms of the vorticity like $\|\omega(t)\|_{L^\infty}$ or $\|\omega(t)\|_{B_{\infty,1}^0}$. For more discussions about the global well-posedness problem concerning other classes of Boussinesq systems we refer to [@HKR1; @MX].
Our goal here is to relax the critical dissipation needed for global well-posedness by some logarithmic factor. More precisely we will study the logarithmically critical Boussinesq model
$$\label{Bouss}
\left\{
\begin{array}{ll}
\partial_{t}v+v\cdot\nabla v+\nabla \pi=\theta e_{2} \\
\partial_{t}\theta+v\cdot\nabla\theta + \frac{{|\textnormal{D}|}}{\log^\alpha(\lambda+{|\textnormal{D}|})} \theta = 0 \\
\textnormal{div}\,v=0\\
v_{| t=0}=v^{0}, \quad \theta_{| t=0}=\theta^{0}.
\end{array} \right.$$
Before stating our result we will need some new definitions. First, we define the logarithmic Riesz transform $\mathcal{R}_\alpha$ by $\mathcal{R}_\alpha=\frac{\partial_1}{{|\textnormal{D}|}}{\tiny}{\log^{\alpha}(\lambda+{|\textnormal{D}|})}.$ Second, for a given $\alpha\in{\mathbb{R}}$ we define the function spaces $\{\mathcal{X}_p\}_{1\le p\le\infty}$ by $$u\in \mathcal{X}_p\Leftrightarrow \|u\|_{\mathcal{X}_p}:=\|u\|_{B_{\infty,1}^0\cap L^p}+\|\mathcal{R}_\alpha u\|_{B_{\infty,1}^0\cap L^p}<\infty.$$
Our result reads as follows (see section \[preliminaries\] for the definitions and the basic properties of Besov spaces).
\[theo1\] Let $\alpha\in[0,\frac12], \lambda\geq e^{4}$ and $p\in]2, \infty[$. Let $v_0\in {B}_{\infty, 1}^{1}\cap \dot{W}^{1,p}$ be a divergence free vector-field of ${\mathbb{R}}^2$ and $\theta_0\in \mathcal{X}_p$. Then there exists a unique global solution $(v, \theta)$ to the system with $$v\in L^\infty_{\textnormal{loc}}\big({\mathbb{R}}_+;{B}_{\infty, 1}^{1}\cap \dot{W}^{1,p}),
\qquad \theta\in L_{\textnormal{loc}}^{\infty}\big({\mathbb{R}}_+; \mathcal{X}_p\big)\cap \widetilde L_{\textnormal{loc}}^1({\mathbb{R}}_+; {B}_{p, \infty}^{1,-\alpha}).$$
The proof shares the same ideas as the case $\alpha=0$ treated in [@HKR] but with more technical difficulties. We define $\mathcal{R}_\alpha=\frac{\partial_1}{{|\textnormal{D}|}}{\tiny}{\log^{\alpha}(\lambda+{|\textnormal{D}|})}$ and $\Gamma=\omega+\mathcal{R}_\alpha\theta.$ Then we get $$\partial_t\Gamma+v\cdot\nabla\Gamma=-[\mathcal{R}_\alpha,v\cdot\nabla]\theta.$$ To estimate the commutator in the framework of Lebesgue spaces we use the para-differential calculus combined with Theorems \[max-princ0\] and \[coer10\].
1. We point out that for global well-posedness to the generalized Navier-Stokes system in dimension three, Tao proved in a recent paper [@Tao] that we can improve the dissipation ${|\textnormal{D}|}^{\frac52}$ to $\frac{{|\textnormal{D}|}^{\frac52}}{\log^{\frac12}(2+{|\textnormal{D}|})}\cdot$
2. The space $\mathcal{X}_ p$ is less regular than the space $B_{\infty,1}^\varepsilon\cap B_{p,1}^{\varepsilon}, \forall\varepsilon>0.$ More precisely, we will see in Proposition \[embed\] that $B_{\infty,1}^\varepsilon\cap B_{p,1}^{\varepsilon}\hookrightarrow \mathcal{X}_p.$
3. If we take $\theta=0$ then the system is reduced to the two-dimensional Euler system. It is well known that this system is globally well-posed in $H^s$ for $s>2$. The main tool for global existence is the BKM criterion [@bkm] ensuring that the development of finite-time singularities for Kato’s solutions is related to the blowup of the $L^\infty$ norm of the vorticity near the maximal time existence. In [@vis] Vishik extended the global existence of strong solutions to initial data belonging to Besov spaces $B_{p,1}^{1+2/p}.$ Notice that these spaces have the same scale as Lipschitz functions and in this sense they are called critical and it is not at all clear whether BKM criterion can be used in this context.
4. Since $
B_{r,1}^{1+2/r}\hookrightarrow {B}_{\infty, 1}^{1}\cap \dot{W}^{1,p}$ for all $ r\in [1,+\infty[$ and $p>\max\{r,2\}$, then the space of initial velocity in our theorem contains all the critical spaces $B_{p,1}^{1+2/p}$ except the biggest one, that is $B_{\infty,1}^{1}$. For the limiting case we have been able to prove the global existence only up to the extra assumption $\nabla v_0\in L^p$ for some $p\in]2, \infty[$. The reason behind this extra assumption is the fact that to obtain a global $L^\infty$ bound for the vorticity we need before to establish an $L^p$ estimate for some $p\in]2,\infty[$ and it is not clear how to get rid of this condition.
5. Since $\nabla v\in L^{1}_{\rm loc}({\mathbb{R}}_+;L^\infty)$ then we can propagate all the higher regularities: critical (for example $v_0\in B_{p,1}^{1+2/p}$ with $p<\infty$) and sub-critical (for example $v_0\in H^{s}$,
Notations and preliminaries {#preliminaries}
===========================
Notations
---------
Throughout this paper we will use the following notations.
$\bullet$ For any positive $A$ and $B$ the notation $A\lesssim B$ means that there exist a positive harmless constant $C$ such that $A\le CB$.
$\bullet$ For any tempered distribution $u$ both $ \hat u$ and $\mathcal F u$ denote the Fourier transform of $u$.
$\bullet$ Pour every $p\in [1,\infty]$, $\|\cdot\|_{L^p}$ denotes the norm in the Lebesgue space $L^p$.
$\bullet$ The norm in the mixed space time Lebesgue space $L^p([0,T],L^r(\mathbb R^d)$ is denoted by $\|\cdot\|_{L^p_TL^r}$ (with the obvious generalization to $\|\cdot\|_{L^p_T\mathcal X} $ for any normed space $\mathcal X$).
$\bullet$ For any pair of operators $P$ and $Q$ on some Banach space $\mathcal{X}$, the commutator $[P,Q]$ is given by $PQ-QP$.
$\bullet$ For $p\in[1,\infty]$, we denote by $\dot{W}^{1,p}$ the space of distributions $u$ such that $\nabla u\in L^p.$
Functional spaces
-----------------
Let us introduce the so-called Littlewood-Paley decomposition and the corresponding cut-off operators. There exists two radial positive functions $\chi\in \mathcal{D}({\mathbb{R}}^d)$ and $\varphi\in\mathcal{D}({\mathbb{R}}^d\backslash{\{0\}})$ such that
- $\displaystyle{\chi(\xi)+\sum_{q\geq0}\varphi(2^{-q}\xi)=1}$;$\quad \displaystyle{\forall\,\,q\geq1,\, \textnormal{supp }\chi\cap \textnormal{supp }\varphi(2^{-q})=\varnothing}$
- $ \textnormal{supp }\varphi(2^{-j}\cdot)\cap
\textnormal{supp }\varphi(2^{-k}\cdot)=\varnothing,$ if $|j-k|\geq 2$.
For every $v\in{\mathcal S}'({\mathbb{R}}^d)$ we set $$\Delta_{-1}v=\chi(\hbox{D})v~;\, \forall
q\in{\mathbb{N}},\;\Delta_qv=\varphi(2^{-q}\hbox{D})v\quad\hbox{ and }\;
S_q=\sum_{ j=-1}^{ q-1}\Delta_{j}.$$ The homogeneous operators are defined by $$\dot{\Delta}_{q}v=\varphi(2^{-q}\hbox{D})v,\quad \dot S_{q}v=\sum_{j\leq q-1}\dot\Delta_{j}v,\quad\forall q\in{\mathbb{Z}}.$$ From [@b] we split the product $uv$ into three parts: $$uv=T_u v+T_v u+R(u,v),$$ with $$T_u v=\sum_{q}S_{q-1}u\Delta_q v,\quad R(u,v)=\sum_{q}\Delta_qu\tilde\Delta_{q}v \quad\hbox{and}\quad \tilde\Delta_{q}=\sum_{i=-1}^1\Delta_{q+i}.$$
For $(p,r)\in[1,+\infty]^2$ and $s\in{\mathbb{R}}$ we define the inhomogeneous Besov as the set of tempered distributions $u$ such that $$\|u\|_{B_{p,r}^s}:=\Big( 2^{qs}
\|\Delta_q u\|_{L^{p}}\Big)_{\ell ^{r}}<+\infty.$$ The homogeneous Besov space $\dot B_{p,r}^s$ is defined as the set of $u\in\mathcal{S}'({\mathbb{R}}^d)$ up to polynomials such that $$\|u\|_{\dot B_{p,r}^s}:=\Big( 2^{qs}
\|\dot\Delta_q u\|_{L^{p}}\Big)_{\ell ^{r}({\mathbb{Z}})}<+\infty.$$ For $s,s'\in{\mathbb{R}}$ and $p,r\in[1,\infty]$ we define the generalized Besov space $ B_{p,r}^{s,s'}$ as the set of tempered distributions $u$ such that $$\|u\|_{B_{p,r}^{s,s'}}:=\Big(2^{qs}(|q|+1)^{s'}\|\Delta_q u\|_{L^p}\Big)_{\ell^r}<\infty$$ [Let $T>0$]{} we denote by $L^\rho_{T}B_{p,r}^{s,s'}$ the space of distributions $u$ such that $$\|u\|_{L^\rho_{T}B_{p,r}^{s,s'}}:= \Big\|\Big( 2^{qs}(|q|+1)^{s'}
\|\Delta_q u\|_{L^p}\Big)_{\ell ^{r}}\Big\|_{L^\rho_{T}}<+\infty.$$ We say that $u$ belongs to the space $\widetilde L^\rho_{T}{B_{p,r}^{s,s'}}$ if $$\|u\|_{ \widetilde L^\rho_{T}{B_{p,r}^s}}:= \Big( 2^{qs}(|q|+1)^{s'}
\|\Delta_q u\|_{L^\rho_{T}L^p}\Big)_{\ell ^{r}}<+\infty.$$ By a direct application of the Minkowski inequality, we have the following links between these spaces. Let $ \varepsilon>0,$ then $$L^\rho_{T}B_{p,r}^s\hookrightarrow\widetilde L^\rho_{T}{B_{p,r}^s}\hookrightarrow{L^\rho_{T}}{B_{p,r}^{s-\varepsilon}}
,\,\textnormal{if}\quad r\geq \rho,$$ $${L^\rho_{T}}{B_{p,r}^{s+\varepsilon}}\hookrightarrow\widetilde L^\rho_{T}{B_{p,r}^s}\hookrightarrow L^\rho_{T}B_{p,r}^s,\, \textnormal{if}\quad
\rho\geq r.$$
We will make continuous use of Bernstein inequalities (see [@che1] for instance).
\[lb\] There exists a constant $C$ such that for $q,k\in{\mathbb{N}},$ $1\leq a\leq b$ and for $f\in L^a({\mathbb{R}}^d)$, $$\begin{aligned}
\sup_{|\alpha|=k}\|\partial ^{\alpha}S_{q}f\|_{L^b}&\leq& C^k\,2^{q(k+d(\frac{1}{a}-\frac{1}{b}))}\|S_{q}f\|_{L^a},\\
\ C^{-k}2^
{qk}\|{\Delta}_{q}f\|_{L^a}&\leq&\sup_{|\alpha|=k}\|\partial ^{\alpha}{\Delta}_{q}f\|_{L^a}\leq C^k2^{qk}\|{\Delta}_{q}f\|_{L^a}.\end{aligned}$$
Maximum principle
=================
Our task is to establish some useful estimates for the following equation generalizing $$\label{transport-d}
\left\{ \begin{array}{ll}
\partial_t\theta+v\cdot\nabla\theta+\frac{{|\textnormal{D}|}^\beta}{\log^\alpha(\lambda+{|\textnormal{D}|})}\theta=f\\
\textnormal{div}\, v=0\\
\theta_{| t=0}=\theta_{0},
\end{array} \right.$$ Two special problems will be studied: the first one deals with $L^p$ estimates that give in particular Theorem \[max-princ0\]. However the second one consists in establishing some logarithmic estimates in Besov spaces with index regularity zero. The first main result of this section generalizes Theorem \[max-princ0\].
\[max-princ\] Let $p\in[1,\infty],\beta\in]0,1],\alpha\geq0$ and $\lambda\geq e^{\frac{3+2\alpha}{\beta}}$. Then any smooth solution of satisfies $$\|\theta(t)\|_{L^p}\le\|\theta_0\|_{L^p}+\int_0^t\|f(\tau)\|_{L^p}d\tau.$$
The proof will be done in two steps. The first one is to valid the result for the free fractional heat equation. More precisely we will establish that the semigroup $e^{t\mathcal{L}}$, with $\mathcal{L}:=\frac{{|\textnormal{D}|}^\beta}{\log^\alpha(\lambda+{|\textnormal{D}|})}$, is a contraction in Lebesgue spaces $L^p$, for every $p\in[1,\infty[$ of course under suitable conditions on the parameters $\alpha,\beta,\lambda$. This problem is reduced to show that $\|K_t\|_{L^1}\le1$. This is equivalent to $K_t\in L^1$ and $K_t\geq0.$ As we will see, to get the integrability of the kernel we do not need any restriction on the value of our parameters. Nevertheless, the positivity of $K_t$ requires some restrictions which are detailed in The second step is to establish the $L^p$ estimate for the system and for this purpose we use some results about Lévy operators or more generally sub-Markovians operators.
Definite positive functions
---------------------------
As we will see there is a strong connection between the positivity of the kernel $K_t$ introduced above and the notion of definite positive functions. We will first gather some well-known properties about definite positive functions and recall some useful criteria for characteristic functions. Second and as an application we will show that the kernel $K_t$ is positive under suitable conditions on the involved parameters.
Let $f:{\mathbb{R}}^d\to{\mathbb{C}}$ be a complex-valued function. We say that $f$ is [ definite positive]{} if only if for every integer $n\in{\mathbb{N}}^*$ and every set of points $\{x_j,j=1,...,n\}$ of ${\mathbb{R}}^d$ the matrix $(f(x_j-x_k))_{1\le j,k\le n}$ is positive Hermitian, that is, for every complex numbers $\xi_1,...,\xi_n$ we have $$\sum_{j,k=1}^nf(x_j-x_k)\xi_j\bar\xi_k\geq 0.$$
We will give some results about definite positive functions.
1. From the definition, every definite positive function $f$ satisfies $$f(0)\geq 0,\quad f(-x)=\overline{f(x)},\quad |f(x)|\le f(0).$$
2. The continuity of a definite positive function $f$ at zero gives the continuity everywhere. More precisely we have $$|f(x)-f(y)|\le 2f(0)\big(f(0)-f(x-y)\big).$$
3. The sum of two definite positive functions is also definite positive and according to Shur’s lemma the product of two definite positive functions is also definite positive and therefore the class of definite positive functions is a convex cone closed under multiplication.
4. Let $\mu$ be a finite positive measure then its Fourier-Stielitjes transform is given by $$\widehat\mu(\xi)=\int_{{\mathbb{R}}^d}e^{-ix\cdot\xi}d\mu(x).$$ It is easy to see that $\widehat{\mu}$ is a definite positive function. Indeed $$\begin{aligned}
\sum_{j,k=1}^n\widehat\mu(x_j-x_k)\xi_j\bar \xi_k&=&\int_{{\mathbb{R}}^d}\Big(\sum_{j,k=1}^ne^{-ix\cdot x_j}\xi_je^{ix\cdot x_k}\bar \xi_k\Big)d\mu(x)\\
&=&\int_{{\mathbb{R}}^d}\Big|\sum_{j=1}^ne^{-ix\cdot x_j}\xi_j\Big|^2d\mu(x)\\
&\geq&0.\end{aligned}$$
The converse of the last point $(4)$ is stated by the following result due to Bochner, see for instance Theorem 19 in [@Bochner].
Let $f:{\mathbb{R}}^d\to{\mathbb{C}}$ be a continuous definite positive function, then $f$ is the Fourier transform of a finite positive Borel measure.
Hereafter we will focus on a the class of radial definite positive functions. First we say that $f:{\mathbb{R}}^d\to{\mathbb{R}}$ is radial if $f(x)=F(|x|)$ with $F:[0,+\infty[\to{\mathbb{C}}$. There are some criteria for radial functions to be definite positive. For example in dimension one the celebrated criterion of Pólya [@Polya] states that if $F:[0,+\infty[\to{\mathbb{R}}$ is continuous and convex with $F(0)=1$ and $\lim_{r\to+\infty}F(r)=0$ then $f(x)=F(|x|) $ is definite positive. This criterion was extended in higher dimensions by numerous authors [@Ask; @Gneit; @Trig]. We will restrict ourselves to the following one due to Askey [@Ask].
\[rad1\] Let $d\in{\mathbb{N}},\,\,F:[0,+\infty\to{\mathbb{R}}$ be a continuous function such that
1. $F(0)=1,$
2. the function $r\mapsto (-1)^dF^{(d)}(r)$ exists and is convex on $]0,+\infty[,$
3. $lim_{r\to+\infty}F(r)=\lim_{r\to+\infty} F^{(d)}(r)=0.$
Then for every $k\in\{1,2,..,2d+1\} $ the function $x\mapsto F(|x|)$ is the Fourier transform of a probability measure on ${\mathbb{R}}^k$.
As an application of Askey’s theorem we have that $x\mapsto e^{-t|x|^\beta}$ is definite positive for all $t>0,$ $\beta\in]0,1]$ and $d\in{\mathbb{N}}$. Indeed, the function $F(r)=e^{-tr^\beta}$ is completely monotone, that is, $(-1)^kF^{(k)}(r)\geq0,\,\forall \,r>0,\, k\in{\mathbb{N}}.$ Although the case $\beta\in]1,2]$ can not be reached by this criterion the result is still true.
We will now see that the perturbation of the above function by a logarithmic damping is also definite positive. More precisely, we have
\[rad2\] Let $\alpha,t\in[0,+\infty[\times]0,+\infty[, \beta\in]0,1],\,\lambda\geq e^{\frac{3+2\alpha}{\beta}}$ and define $f:{\mathbb{R}}^d\to{\mathbb{R}}$ by $$f(x)=e^{-t\frac{|x|^\beta}{\log^{\alpha}(\lambda+|x|)}}.$$ Then $f$ is a definite positive function for $d\in\{1,2,3\}.$
1. It is possible that the above result remains true for higher dimension $d\geq4$ but we avoid to deal with this more computational case. We think also that the radial function associated to $f$ is completely monotone.
2. The lower bound of $\lambda$ is not optimal by our method. In fact we can obtain more precise bound but this seems to be irrelevant.
We write $f(x)=F(|x|)$ with $$F(r)=e^{-t\phi(r)}\quad\hbox{and}\quad \phi(r)=\frac{r^\beta}{\log^{\alpha}(\lambda+r)}\cdot$$ The function $F$ is smooth on $]0,\infty[$ and assumptions $(1)$ and $(3)$ of Theorem \[rad1\] are satsified. It follows that the function $f$ is definite positive for $d\in\{1,2,3\}$ if $$F^{(3)}(r)\le0.$$ Easy computations give for $r>0$, $$F^{(3)}(r)=\Big[-t\,\phi^{(3)}(r)+3t^2\,\phi^{\prime}(r)\phi^{(2)}(r)-t^3\,(\phi^{\prime}(r))^3\Big]F(r).$$ We will prove that $$\phi^{\prime}(r)\geq0,\, \phi^{(2)}(r)\le0\quad\hbox{and}\quad \phi^{(3)}(r)\geq0.$$ This is sufficient to get $F^{(3)}(r)\le0,\forall\, r>0.$ The first derivative of $\phi$ is given by $$\begin{aligned}
\phi^{\prime}(r)&=&\frac{\beta\, r^{\beta-1}}{\log^{\alpha}(\lambda+r)}-\frac{\alpha\, r^{\beta}}{(\lambda+r)\log^{\alpha+1}(\lambda+r)}\\
&=&\frac{ r^{\beta-1}}{(\lambda+r)\log^{\alpha+1}(\lambda+r)}\Big(\beta\lambda\log(\lambda+r)+r\big(\beta\log(\lambda+r)-\alpha\big) \Big).\end{aligned}$$ We see that if $\lambda$ satisfies $$\label{c1}
\lambda\geq e^{\frac\alpha\beta}$$ then $\phi^{\prime}(r)\geq0$. For the second derivative of $\phi$ we obtain $$\begin{aligned}
\phi^{(2)}(r)&=&-
\frac{\beta(1-\beta)r^{\beta-2}}{\log^{\alpha}(\lambda+r)}-\frac{2\alpha\beta r^{\beta-1}}{(\lambda+r)\log^{1+\alpha}(\lambda+r)}\\
&+&\frac{\alpha r^\beta}{(\lambda+r)^2\log^{\alpha+1}(\lambda+r)}+\frac{\alpha(\alpha+1)r^\beta}{(\lambda+r)^2\log^{\alpha+2}(\lambda+r)}\\
&=&\frac{r^{\beta-2}}{\log^{\alpha}(\lambda+r)}\Bigg[ -\beta(1-\beta)-\frac{2\alpha\beta r}{(\lambda+r)\log(\lambda+r)}\\
&+&\frac{\alpha r^2}{(\lambda+r)^2\log(\lambda+r)}+\frac{\alpha(\alpha+1) r^2}{(\lambda+r)^2\log^2(\lambda+r)}\Bigg]. \end{aligned}$$ Since $\frac{ r^2}{(\lambda+r)^2}\le\frac{ r}{\lambda+r}\le1,$ then $$\begin{aligned}
\phi^{(2)}(r)&\le&\frac{r^{\beta-2}}{\log^{\alpha}(\lambda+r)}\Bigg[(1-\beta)\Big(-\beta+\frac{2\alpha}{\log(\lambda+r)}\Big)-\frac{\alpha r}{(\lambda+r)\log(\lambda+r)}\Big(1-\frac{\alpha+1}{\log(\lambda+r)}\Big)\Bigg]\\
&\le&\frac{r^{\beta-2}}{\log^{\alpha}(\lambda+r)}\Bigg[(1-\beta)\Big(-\beta+\frac{2\alpha}{\log\lambda}\Big)-\frac{\alpha r}{(\lambda+r)\log(\lambda+r)}\Big(1-\frac{\alpha+1}{\log\lambda}\Big)\Bigg].\end{aligned}$$ Now we choose $\lambda$ such that $$-\beta+\frac{2\alpha}{\log\lambda}\leq0\quad\hbox{and}\quad 1-\frac{\alpha+1}{\log\lambda}\geq0$$ which is true despite $\lambda$ satisfies $$\label{cond1}
\max(e^{\frac{2\alpha}{\beta}},e^{\alpha+1})\le\lambda.$$ Under this assumption we get $$\phi^{(2)}(r)\le0,\forall r>0.$$ Similarly we have $$\begin{aligned}
\phi^{(3)}(r)&=&\alpha(\alpha+1)r^{\beta-1}\frac{\log^{-\alpha-3}(\lambda+r)}{(\lambda+r)^2}\Big[3\lambda\beta\log(\lambda+r)+r\big(3\beta\log(\lambda+r)-(2+\alpha)\big)\Big]\\&+&
\alpha r^{\beta-2}\frac{\log^{-2-\alpha}(\lambda+r)}{(\lambda+r)^3}\Big[r^2\Big(-3(1+\alpha)+(-3\beta^2+6\beta-2)\log(\lambda+r)\Big)\\
&+&\log(\lambda+r)\Big(\lambda\beta(9-6\beta)r+3\lambda^2\beta(1-\beta) \Big) \Big]\\&+&
(2-\beta)(1-\beta)\beta r^{\beta-3}\log^{-\alpha}(\lambda+r)\\&=&
I_1+I_2+I_3+I_4.\end{aligned}$$ it is easy to see that $I_3$ and $I_4$ are nonnegative. On the other hand we have $$\begin{aligned}
I_1+I_2&=&3\lambda\beta\alpha(\alpha+1)r^{\beta-1}\frac{\log^{-\alpha-2}(\lambda+r)}{(\lambda+r)^2}\\&+&
\alpha r^{\beta}\frac{\log^{-2-\alpha}(\lambda+r)}{(\lambda+r)^3}\Bigg[-3(1+\alpha)+(\alpha+1)(\lambda+r)\Big(3\beta-\frac{2+\alpha}{\log(\lambda+r)}\Big)\\&+&(-3\beta^2+6\beta-2)\log(\lambda+r)\Bigg]\end{aligned}$$ Since $-3\beta^2+6\beta-2\geq-2,$ for $\beta\in[0,1]$ and $-\frac{\log x}{x}\geq -\frac{\log\lambda}{\lambda}, \forall x\geq \lambda\geq e,$ then $$\begin{aligned}
I_1+I_2&\geq&
\alpha(\alpha+1) r^{\beta}\frac{\log^{-2-\alpha}(\lambda+r)}{(\lambda+r)^3}\Bigg[-3+(\lambda+r)\Big(3\beta-\frac{2+\alpha}{\log\lambda}-2\frac{\log(\lambda+r)}{(\alpha+1)(\lambda+r)}\Big)\Bigg]\\
&\geq&\alpha(\alpha+1) r^{\beta}\frac{\log^{-2-\alpha}(\lambda+r)}{(\lambda+r)^2}\Bigg[3\beta-\frac{3}{\lambda}-\frac{2+\alpha}{\log\lambda}-\frac{2\log\lambda}{(\alpha+1)\lambda}\Big)\Bigg].\end{aligned}$$ We can check that $$\log\lambda\le\lambda\quad\hbox{and}\quad \log^2\lambda\le\lambda, \,\forall\lambda\geq e.$$ Thus $$\begin{aligned}
I_1+I_2&\geq&\alpha(\alpha+1) r^{\beta}\frac{\log^{-2-\alpha}(\lambda+r)}{(\lambda+r)^2}\Big[3\beta-\frac{1}{\log\lambda}\big(5+\alpha+\frac{2}{\alpha+1} \big)\Big]\\
&\geq&\alpha(\alpha+1) r^{\beta}\frac{\log^{-2-\alpha}(\lambda+r)}{(\lambda+r)^2}\Big[3\beta-\frac{7+\alpha}{\log\lambda}\Big].\end{aligned}$$ We choose $\lambda$ such that $$3\beta-\frac{7+\alpha}{\log\lambda}\geq 0.$$ It follows that $I_1+I_2$ is nonnegative if $$\label{c3}
\lambda\geq e^{\frac{7+\alpha}{3\beta}}.$$ Remark that the assumptions , and are satisfies under the condition $$\lambda\geq e^{\frac{3+2\alpha}{{\beta}}}.$$ Finally, we get: $\quad \forall \alpha\geq0,\, \beta\in]0,1],\,\lambda\geq e^{\frac{3+2\alpha}{{\beta}}},$ $$\forall r>0,\quad\phi^{(3)}(r)\geq0.$$ This achieves the proof.
More precise informations about the kernel $K_t$ will be listed in the following lemma.
\[kernel\] Let $\lambda\geq 2$ and denote by $K_t$ the element of $\mathcal{S}'({\mathbb{R}}^d)$ such that $$\widehat{K_t}(\xi)=e^{-t\frac{|\xi|^\beta}{\log^{\alpha}(\lambda+|\xi|)}}.$$ Then we have the following properties.
1. For $(t,\alpha,\beta)\in]0,\infty[\times{\mathbb{R}}\times]0,\infty[$ the function $K_t$ belongs to $\in L^1\cap C_0$.
2. For $d\in\{1,2,3\},(t,\alpha,\beta)\in]0,+\infty[\times[0,\infty[\times]0,1]$ and $\lambda\geq e^{\frac{3+2\alpha}{\beta}},$ we have $$\quad K_t(x)\geq 0,\,\forall x\in {\mathbb{R}}_+\quad\hbox{and}\quad \|K_t\|_{L^1}=1.$$
[**$(1)$**]{} By definition we have $$K_t(x)=(2\pi)^{-d}\int_{{\mathbb{R}}^d}e^{-t\frac{|\xi|^\beta}{\log^{\alpha}(\lambda+|\xi|)}} e^{i\,x\cdot\xi}d\xi.$$ Let $\mu\geq0,$ then integrating by parts we get $$|x|^{\mu}x_j^dK_t(x)=(-2i\pi)^{-d}\int_{{\mathbb{R}}^d}\partial_{\xi_j}^d\big(e^{-t\frac{|\xi|^\beta}{\log^{\alpha}(\lambda+|\xi|)}}\big) |x|^{\mu}e^{i\,x\cdot\xi}d\xi.$$ On the other hand we have $$|x|^{\mu}e^{i\,x\cdot\xi}={|\textnormal{D}|}^{\mu}e^{i\,x\cdot\xi},$$ here ${|\textnormal{D}|}$ is a fractional derivative on the variable $\xi$. Thus we get $$|x|^{\mu}x_j^dK_t(x)=(-2i\pi)^{-d}\int_{{\mathbb{R}}^d}{|\textnormal{D}|}^\mu\partial_{\xi_j}^d\big(e^{-t\frac{|\xi|^\beta}{\log^{\alpha}(\lambda+|\xi|)}}\big)e^{i\,x\cdot\xi}d\xi.$$ Now we use the following representation for ${|\textnormal{D}|}^\mu$ when $\mu\in]0,2]$ $${|\textnormal{D}|}^\mu f(x)=C_{\mu,d}\int_{{\mathbb{R}}^{d}}\frac{f(x)-f(x-y)}{|y|^{d+\mu}}dy.$$ It follows that $$|x|^{\mu}|x_j^dK_t(x)|\le C_{\mu,d}\int_{{\mathbb{R}}^{2d}}\frac{|\mathcal{K}_j(\xi)-\mathcal{K}_j(\xi-y)|}{|y|^{d+\mu}}dyd\xi$$ with $$\mathcal{K}_j(\xi):=\partial_{\xi_j}^d\big(e^{-t\frac{|\xi|^\beta}{\log^{\alpha}(\lambda+|\xi|)}}\big).$$ Now we decompose the integral into two parts $$\begin{aligned}
\int_{{\mathbb{R}}^{2d}}\frac{|\mathcal{K}_j(\xi)-\mathcal{K}_j(\xi-y)|}{|y|^{d+\mu}}dyd\xi&=&\int_{|y|\geq\frac{|\xi|}{2}}\frac{|\mathcal{K}_j(\xi)-\mathcal{K}_j(\xi-y)|}{|y|^{d+\mu}}dyd\xi\\
&+&\int_{|y|\leq\frac{|\xi|}{2}}\frac{|\mathcal{K}_j(\xi)-\mathcal{K}_j(\xi-y)|}{|y|^{d+\mu}}dyd\xi\\
&=&I_1+I_2.\end{aligned}$$ To estimate the first term we use the following estimate that can be obtained by straightforward computations $$\begin{aligned}
|\mathcal{K}_j(\xi)|&\le& C_{t,\alpha,\beta}\frac{|\xi|^{\beta-d}}{\log^{\alpha}(\lambda+|\xi|)}e^{-t\frac{|\xi|^\beta}{\log^{\alpha}(\lambda+|\xi|)}}\\
&\le&C_{t,\alpha,\beta}{|\xi|^{\beta-d}}e^{-\frac12t\frac{|\xi|^\beta}{\log^{\alpha}(\lambda+|\xi|)}}.\end{aligned}$$ Hence we get under the assumption $\mu\in]0,\beta[,$ $$\begin{aligned}
I_1&\le& C_{t,\alpha,\beta}\int_{|\xi|\le 2|y|}\frac{1}{|y|^{d+\mu}}\Big({|\xi|^{\beta-d}}e^{-\frac12t\frac{|\xi|^\beta}{\log^{\alpha}(\lambda+|\xi|)}}+{|\xi-y|^{\beta-d}}e^{-\frac12t\frac{|\xi-y|^\beta}{\log^{\alpha}(\lambda+|\xi-y|)}}\Big)d\xi dy\\
&\le&C_{t,\alpha,\beta}\int_{|\xi|\le 3|y|}\frac{1}{|y|^{d+\mu}}{|\xi|^{\beta-d}}e^{-\frac12 t\frac{|\xi|^\beta}{\log^{\alpha}(\lambda+|\xi|)}}d\xi dy\\
&\le&C_{t,\alpha,\beta}\int_{{\mathbb{R}}^d}\frac{1}{|\xi|^{d+\mu-\beta}}e^{-\frac12 t\frac{|\xi|^\beta}{\log^{\alpha}(\lambda+|\xi|)}}d\xi\\
&\le& C_{t,\alpha,\beta}.\end{aligned}$$ To estimate the second term we use the mean-value Theorem $${|\mathcal{K}_j(\xi)-\mathcal{K}_j(\xi-y)|}\le|y|\sup_{\eta\in[\xi-y,\xi]}|\nabla \mathcal{K}_j(\eta)|.$$ On the other hand we have $$\begin{aligned}
|\nabla \mathcal{K}_j(\eta)|&\le& C_{t,\alpha,\beta}|\eta|^{\beta-d-1}e^{-\frac12t\frac{|\eta|^\beta}{\log^{\alpha}(\lambda+|\eta|)}}.\end{aligned}$$ Now since $|y|\le\frac12|\xi|$ then for $\eta\in[\xi-y,\xi]$ we have $$\frac12|\xi|\le|\eta|\le\frac52|\xi|.$$ This yields $${|\mathcal{K}_j(\xi)-\mathcal{K}_j(\xi-y)|}\le C_t|y||\xi|^{\beta-d-1}e^{-Ct|\xi|^{\frac\beta2}}.$$ Therefore we find for $\mu\in]0,\beta[\cap]0,1[$, $$\begin{aligned}
I_2&\le& C_{t,\alpha,\beta}\int_{|y|\le\frac12|\xi|}\frac{1}{|y|^{d+\mu-1}}|\xi|^{\beta-d-1}e^{-Ct|\xi|^{\frac\beta2}}dyd\xi\\
&\le&C_{t,\alpha,\beta}\int_{{\mathbb{R}}^2}\frac{1}{|\xi|^{d+\mu-\beta}}e^{-Ct|\xi|^{\frac\beta2}}d\xi\\
&\le&C_{t,\alpha,\beta}.\end{aligned}$$ Finally we get $$\hbox{for}\quad j=1,..,d,\quad |x|^\mu|x_j|^d|K_t(x)|\le C_{t,\alpha,\beta}.$$ Since $K_t\in C_0$ then $$\big(1+|x|^{d+\mu}\big)|K_t(x)|\le C_t.$$ This proves that $K_t\in L^1({\mathbb{R}}^d)$.
[**$(2)$**]{} Using Theorem \[rad1\] and Proposition \[rad2\] we get $K_t\geq0$. Since $K_t\in L^1$ then $$\|K_t\|_{L^1}=\widehat{K_t}(0)=1.$$
Now we define the propagator $e^{-t\frac{{|\textnormal{D}|}^\beta}{\log^\alpha(\lambda+{|\textnormal{D}|})}}$ by convolution $$e^{-t\frac{{|\textnormal{D}|}^\beta}{\log^\alpha(\lambda+{|\textnormal{D}|})}}f=K_t\star f.$$ We have the following result.
\[cor1\] Let $\alpha\geq0,\beta\in]0,1],\,\lambda\geq e^{\frac{3+2\alpha}{\beta}}$ and $p\in[1,\infty]$. Then $$\|e^{-t\frac{{|\textnormal{D}|}^\beta}{\log^\alpha(\lambda+{|\textnormal{D}|})}} f\|_{L^p}\le\|f\|_{L^p},\quad\forall\,\, f\in L^p.$$
From the classical convolution inequalities combined with Lemma \[kernel\] we get $$\begin{aligned}
\|e^{-t\frac{{|\textnormal{D}|}^\beta}{\log^\alpha(\lambda+{|\textnormal{D}|})}} f\|_{L^p}&\le&\|K_t\|_{L^1}\|f\|_{L^p}\\
&\le&\|f\|_{L^p}.\end{aligned}$$
Structure of the semigroup $(e^{-t\frac{{|\textnormal{D}|}^\beta}{\log^\alpha(\lambda+{|\textnormal{D}|})}})_{t\geq0}$
----------------------------------------------------------------------------------------------------------------------
We will first recall the notions of $C_0-$semigroup and sub-Markovian generators. First we introduce the notion of strongly continuous semigroup.
\[def-sem\] Let $X$ be a Banach space and $(T_t)_{t\geq0}$ be a family of bounded operators from $X$ into $X$. This family is called a strongly continuous semigroup on $X$ or a $C_0$-semigroup if
1. $T_0=\hbox{Id},$
2. for every $t,s\geq0,\, T_{t+s}=T_tT_s,$
3. for every $x\in X, \lim_{t\to 0^+}\|T_t x-x\|=0.$
If in addition the semigroup satisfies the estimate $$\|T_t\|_{\mathcal{L}(X)}\leq 1,$$ then it is called a $C_0$-semigroup of contractions.
For a given $C_0$-semigroup of contractions $(T_t)_{t\geq0}$ we define its domain $\mathcal{D}(A)$ by $$\mathcal{D}(A):=\Big\{f\in X;\lim_{t\to0^+}\frac{T_tf-f}{t}\,\,\textnormal{exists in}\,\, X \Big\},$$
$$Af=\lim_{t\to0^+}\frac{T_tf-f}{t},\quad f\in \mathcal{D}(A).$$ It is well-known that the operator $A$ is densely defined: its domain $\mathcal{D}(A)$ is dense
We introduce now the special case of sub-Markovian semigroups.
\[defis2\] Let $X=L^p({\mathbb{R}}^d),$ with $p\in[1,\infty[ $ and $d\in{\mathbb{N}}^*.$ Let $(T_t)_{t\geq0}$ be a $C_0$-semigroup of contractions on $X.$ It is said a sub-Markovian semigroup if
1. If $f\in X, f(x)\geq 0,$ a .e. then for every $t\geq0$, $T_t f(x)\geq 0,$ a.e.,
2. If $f\in X, |f|\leq1$ then for every $t\geq0,\,|T_tf|\le1$.
Denote by $L^p_+:=\Big\{f\in L^p; f(x)\geq0, a.e\Big\}$. Then we have the following classical result.
\[deurl\] Let $A$ be a nonnegative self-adjoint operator Then we have the equivalence between
1. $\forall t>0,\, f\in L^2_+\Rightarrow e^{-tA}f\in L^2_+.$
2. $f\in \mathcal{D}(A^{\frac12})\Rightarrow |f|\in\mathcal{D}(A^{\frac12})\quad\hbox{and}\quad\|A^{\frac12}|f|\|_{L^2}\le\|A^{\frac12}f\|_{L^2}$
Now we will establish the following result.
\[props23\] Let $d\in\{1,2,3\},\, p\in[1,\infty[,\alpha\geq0, \beta\in]0,1]$ and $ \lambda\geq e^{\frac{3+2\alpha}{\beta}}$. then
1. The family $(e^{-t\mathcal{L}})_{t\geq0}$ defines a $C_0$-semigroup of contractions in $L^p({\mathbb{R}}^d).
$
2. The family $(e^{-t\mathcal{L}})_{t\geq0}$ defines a sub-Markovian semigroup in $L^p({\mathbb{R}}^d).
$
3. The operator $(e^{-t\mathcal{L}})_{t\geq0}$ satisfies the Beurling-Deny theorem described in Theorem $\ref{deurl}$.
[**($1$)**]{} For $f\in L^p$ we define the action of the semigroup to this function by $$e^{-t\mathcal{L}}f(x)=K_t\star f(x),$$ where $\widehat{K_t}(\xi)=e^{-t\frac{|\xi|}{\log^\alpha(\lambda+|\xi|)}}.$ From Corollary \[cor1\] we have that the semigroup maps $L^p$ to itself for every $p\in [1,\infty]$ and $$\|K_t\star f\|_{L^p}\le\|f\|_{L^p}.$$ The points $(1)$ and $(2)$ of the Definition \[def-sem\] are easy to check. It remains to prove the third point concerning the strong continuity of the semigroup. Since $\|K_t\|_{L^1}=1$ and $K_t\geq0,$ then for $\eta>0$ we have $$\begin{aligned}
K_t\star f(x)-f(x)&=&\int_{{\mathbb{R}}^d}K_t(y)(f(x-y)-f(x))dy\\
&=&\int_{|y|\le\eta}K_t(y)(f(x-y)-f(x))dy\\
&+&\int_{|y|\geq\eta}K_t(y)(f(x-y)-f(x))dy\\
&=&\hbox{I}_1(x)+\hbox{I}_2(x).\end{aligned}$$ The first term is estimated as follows $$\begin{aligned}
\|\hbox{I}_1\|_{L^p}&\le&\int_{|y|\le\eta}K_t(y)\|f(\cdot-y)-f(\cdot)\|_{L^p}dy\\
&\le&\sup_{|y|\le\eta}\|f(\cdot-y)-f(\cdot)\|_{L^p}.\end{aligned}$$ For the second term we write $$\|\hbox{I}_2\|_{L^p}\le 2\|f\|_{L^p}\int_{|y|\geq \eta}K_t(y)dy.$$ Combining these estimates we get $$\|K_t\star f-f\|_{L^p}\le\sup_{|y|\le\eta}\|f(\cdot-y)-f(\cdot)\|_{L^p}+2\|f\|_{L^p}\int_{|y|\geq \eta}K_t(y)dy.$$ It is well-know that for every $p\in[1,\infty[$ we have $$\lim_{\eta\to 0^+}\sup_{|y|\le\eta}\|f(\cdot-y)-f(\cdot)\|_{L^p}=0.$$ Thus for a given $\varepsilon>0$ we can find $\eta> 0$ small enough such that $$\sup_{|y|\le\eta}\|f(\cdot-y)-f(\cdot)\|_{L^p}\le\varepsilon.$$ Now to conclude the proof it suffices to prove that $$\lim_{t\to0^+}\int_{|y|\geq \eta}K_t(y)dy=0.$$ This assertion is a consequence of the following result $$K_t\overset{t\to 0^+}{\rightharpoonup} \delta_0.$$ To prove the last one we write for $\phi\in\mathscr{S},$ $$\begin{aligned}
\langle K_t,\phi\rangle&=&\frac{1}{(2\pi)^d}\langle \widehat{K_t},\widehat\phi\rangle\\
&=&\frac{1}{(2\pi)^d}\int_{{\mathbb{R}}^d}e^{-t|\xi|^\alpha}\widehat\phi(\xi)d\xi.\end{aligned}$$ We can use now Lebesgue theorem and the inversion Fourier transform leading to $$\lim_{t\to 0^+}\langle K_t,\phi\rangle=\frac{1}{(2\pi)^d}\int_{{\mathbb{R}}^d}\widehat\phi(\xi)d\xi=\phi(0).$$ Finally we get that $(K_t\star)_{t\geq0}$ defines a $C_0$-semigroup of contractions for every $p\in[1,\infty[$. [**($2$)**]{} From the Definition \[defis2\] and the first part of Proposition \[props23\] it remains to show that
1. For $f\in L^p$ with $ f(x)\geq 0 ,$ a.e. we have $e^{-t\mathcal{L}}f(x)\geq 0$
2. For $f\in L^p$ with $ |f(x)|\leq 1$, a.e. we have $|e^{-t\mathcal{L}}f(x)|\le 1$
The proof is a direct consequence of the explicit formula $$e^{-t\mathcal{L}}f(x)=K_t\star f(x),$$ where according to Lemma \[kernel\] we have $K_t\geq0$ and $\|K_t\|_{L^1}=1$. [**[($3$)]{}**]{} It is not hard to see that the operator $\frac{{|\textnormal{D}|}^\beta}{\log^\alpha(\lambda+{|\textnormal{D}|})}$ is a nonnegative self-adjoint operator of $L^2$. This operator satisfies the first condition of Theorem \[deurl\] since the kernel $K_t$ is positive.
The following result gives in particular Theorem \[max-princ\].
\[maxi\]Let $
A$ be a generator of a $C_0$-semigroup of contractions, then
1. Let $p\in[1,\infty[$ and $u\in \mathcal{D}({A})$. then $$\int_{{\mathbb{R}}^2}A u\,|u|^{p-1}\textnormal{sign}\, u\,dx\leq 0.$$
2. Let $\theta$ be a smooth solution of the equation $$\partial_t \theta+v\cdot\nabla\theta-A\theta=f$$where $v$ is a smooth vector-field with zero divergence and $f$ a smooth function. Then for every $p\in[1,\infty]$ $$\|\theta(t)\|_{L^p}\le\|\theta_0\|_{L^p}+\int_0^t\|f(\tau)\|_{L^p}d\tau.$$
[**[(1)]{}**]{} We introduce the operation $[h,g]$ between two functions by $$[h,g]=\|g\|_{L^p}^{2-p}\int_{{\mathbb{R}}^2}h(x)|g(x)|^{p-1}\textnormal{sign}\, g(x)dx.$$ Now, we define the function $\psi:[0,\infty[\to {\mathbb{R}}$ by $$\psi(t)=\Big[e^{t A}u,u\Big].$$ We have $\psi(0)=\|u\|_{L^p}^2$ and from Hölder inequality combined with the fact that the operator $e^{tA}$ is a contraction on $L^p$ we get $$\begin{aligned}
\psi(t)&\le&\|e^{t A}u\|_{L^p}\|u\|_{L^p}\\
&\le&\|u\|_{L^p}^2.\end{aligned}$$ Thus we find $\psi(t)\le\psi(0), \forall t\geq 0$. Therefore we get $\lim_{t\to0^+}\frac{\psi(t)-\psi(0)}{t}\leq 0.$ This gives $$\int_{{\mathbb{R}}^2}Au(x)|u(x)|^{p-1}\textnormal{sign}\, u(x)dx\leq0.$$
[**2)**]{} Let $p\in[1,\infty[$ then multiplying the equation by $|\theta|^{p-1}\textnormal{sign} \theta$ and integrating by parts using $\textnormal{div} v=0$ we get $$\frac1p\frac{d}{dt}\|\theta(t)\|_{L^p}^p+\int_{{\mathbb{R}}^2}|A\theta(x)\theta(x)|^{p-1}\textnormal{sign}\, \theta(x)dx\le\|f(t)\|_{L^p}\|\theta(t)\|_{L^p}^{p-1}.$$ Using Proposition \[maxi\] we find $$\frac1p\frac{d}{dt}\|\theta(t)\|_{L^p}^p\le \|f(t)\|_{L^p}\|\theta(t)\|_{L^p}^{p-1}.$$ By simplifying $$\frac{d}{dt}\|\theta(t)\|_{L^p}\le \|f(t)\|_{L^p}.$$ Integrating in time we get for $p\in[1,\infty[$ $$\|\theta(t)\|_{L^p}\le\|\theta_0\|_{L^p}.$$ Since the estimates are uniform on the parameter $p$ then we can get the limit case $p=+\infty.$
Logarithmic estimate
--------------------
Let us now move to the last part of this section which deals with some logarithmic estimates generalizing the results of [@vis; @hk]. First we recall the following result of propagation of Besov regularities.
\[prop-Bes\] Let $\kappa\geq0$ and $A$ be a $C_0$ semigroup of contractions on $L^m({\mathbb{R}}^d)$ for every $m\in[1,\infty[.$ We assume that for every $q\in{\mathbb{N}}\cup\{-1\}$, the operator $\Delta_q$ and $A$ commute on a dense subset of $L^p$. Let $(p,r)\in[1,\infty]^2, s\in]-1,1[$ and $\theta$ be a smooth solution of $$\partial_t\theta+v\cdot\nabla\theta-\kappa A\theta=f.$$ Then we have $$\|\theta\|_{\widetilde L^\infty_tB_{p,r}^s}\lesssim e^{CV(t)}\Big(\|\theta_0\|_{B_{p,r}^s}+\int_0^te^{-CV(\tau)}\|f(\tau)\|_{B_{p,r}^s}d\tau \Big),$$ where $V(t)=\|\nabla v\|_{L^1_t L^\infty}$ and $C$ a constant depending only on $s$ and $d$.
We set $\theta_q:=\Delta_q\theta$ then by localizing in frequency the equation of $\theta$ we get $$\partial_t\theta_q+v\cdot\nabla\theta_q-\kappa A\theta_q=-[\Delta_q, v\cdot\nabla]\theta+f_q.$$ Using Proposition \[maxi\] we get $$\|\theta_q(t)\|_{L^p}\le\|\theta_q(0)\|_{L^p}+\int_0^t\|[\Delta_q, v\cdot\nabla]\theta(\tau)\|_{L^p}d\tau+\int_0^t\|f_q(\tau)\|_{L^p}d\tau.$$ On the other hand we have the classical commuator estimate, see [@che1] $$\|[\Delta_q, v\cdot\nabla]\theta\|_{L^p}\le C 2^{-qs}c_q\|\nabla v\|_{L^\infty} \|\theta\|_{B_{p,r}^s},\quad \|(c_q)\|_{\ell^r}=1.$$ Thus $$\|\theta(t)\|_{B_{p,r}^s}\le\|\theta_0\|_{B_{p,r}^s}+C\int_0^t\|\nabla v\|_{L^\infty} \|\theta\|_{B_{p,r}^s}+\int_0^t\|f(\tau)\|_{B_{p,r}^s}d\tau.$$ It suffices now to use Gronwall inequality
Now we will show that for the index regularity $s=0$ we can obtain a better estimate with a linear growth on the norm of the velocity.
\[thmlog\] Let $v$ be a smooth divergence free vector-field on ${\mathbb{R}}^d$. Let $\kappa\geq0$ and $A$ be a generator of $C_0$-semigroup of contractions on $L^p({\mathbb{R}}^d)$ for every $p\in[1,\infty[.$ We assume that for every $q\in{\mathbb{N}}$, the opeartors $\Delta_q$ and $A$ commute on a dense subset of $L^p$. Let $\theta$ be a smooth solution of $$\partial_t\theta+v\cdot\nabla\theta-\kappa A\theta=f.$$ Then we have for every $p\in [1,\infty]$ $$\|\theta\|_{\widetilde L^\infty_tB_{p,1}^0}\leq C\Big( \|\theta_0\|_{B_{p,1}^0}+\|f\|_{L^1_tB_{p,1}^0}\Big)\Big( 1+\int_0^t\|\nabla v(\tau)\|_{L^\infty}d\tau \Big),$$ where the constant $C$ does not depend on $p$ and $\kappa.$
We mention that the result is first proved in [@vis] for the case $\kappa=0$ by using the special structure of the transport equation. In [@H-K2] Keraani and the author generalized Vishik’s result for a transport-diffusion equation where the dissipation term has the The method described in [@H-K2] can be easily adapted here for our model.
Let $q\in{\mathbb{N}}\cup\{-1\}$ and denote by $\overline\theta_q$ the unique global solution of the initial value problem $$\label{R_{Q}}\left\lbrace
\begin{array}{l}
\partial_t \overline\theta_q+v\cdot\nabla \overline\theta_q-\kappa A\overline\theta_q=\Delta_{q}f,
\\
{\overline\theta_q}_{|t=0}=\Delta_{q}\theta^{0}.\\
\end{array}
\right.$$ Using Proposition \[prop-Bes\] with $s=\pm\frac12$ we get $$\|\overline\theta_q\|_{\widetilde L^\infty_tB_{p,\infty}^{\pm\frac12}}\lesssim
\big(\|\Delta_{q} \theta_0\|_{B_{p,\infty}^{\pm\frac12}}+
\|\Delta_{q}f\|_{L^1_{t}B_{p,\infty}^{\pm\frac12}}\big) e^{CV(t)
},$$ where $
V(t)=\|\nabla v\|_{L^1_{t}L^\infty}.
$ Combined with the definition of Besov spaces this yields $$\label{t1}
\|\Delta_{j}\overline\theta_q\|_{L^\infty_tL^p}\lesssim 2^{-\frac12|j-q|}
\big(\| \Delta_{q} \theta_0\|_{L^p}+\|\Delta_{q}f\|_{L^1_{t}L^p}\big) e^{CV(t)}.$$ By linearity and again the definition of Besov spaces we have $$\begin{aligned}
\label{t2}
\|\theta\|_{\widetilde L^\infty_tB_{p,1}^0}\leq
\sum_{|j-q|\geq N}
\|\Delta_{j}\overline\theta_q\|_{L^\infty_tL^p}+\sum_{|j-q|< N}
\|\Delta_{j}\overline\theta_q\|_{L^\infty_tL^p},\end{aligned}$$ where $N\in \Bbb N$ is to be chosen later. To deal with the first sum we use (\[t1\]) $$\begin{aligned}
\nonumber\sum_{|j-q|\geq N}
\|\Delta_{j}\overline\theta_q\|_{L^\infty_tL^p}&\lesssim& 2^{-N/2}\sum_{q\geq-1}\big(\|\Delta_{q}\theta_0\|_{L^p}+\|\Delta_{q}f\|_{L^1_{t}L^p}\big)e^{CV(t)}
\\
&\lesssim& 2^{-N/2}
\big(\|\theta ^0\|_{B_{p,1}^0}+\|f\|_{L^1_{t}B_{p,1}^0}\big)e^{CV(t)}.\end{aligned}$$ We now turn to the second sum in the right-hand side of (\[t2\]).
It is clear that $$\nonumber\sum_{|j-q|< N}
\|\Delta_{j}\overline\theta_q\|_{L^\infty_tL^p}\lesssim \sum_{|j-q|< N}
\|\overline\theta_q\|_{L^\infty_tL^p}.$$ Applying Proposition \[propmax\] to the system (\[R\_[Q]{}\]) yields $$\|\overline\theta_q\|_{L^\infty_tL^p}\leq \|\Delta_q\theta_0\|_{L^p}+\|\Delta_{q}f\|_{L^1_{t}L^p}.$$ It follows that $$\sum_{|j-q|<N}\|\Delta_{j}\overline\theta_q\|_{L^\infty_tL^p}\lesssim N\big(\|\theta ^0\|_{B_{p,1}^0}+\|f\|_{L^1_{t}B_{p,1}^0}\big).$$ The outcome is the following $$\|\theta\|_{\widetilde L^\infty_tB_{p,1}^0}\lesssim\big(\|\theta ^0\|_{B_{p,1}^0}+\|f\|_{L^1_{t}B_{p,1}^0}\big)\Big(2^{- N/2}e^{CV(t)}+N\Big).$$ Choosing $$N=\Big[\frac{2C
V(t)}{ \log 2}\Big]+1,$$ we get the desired result.
Combining Propositions \[thmlog\] and \[props23\] we get,
\[thmlog\] Let $v$ be a smooth divergence free vector-field on ${\mathbb{R}}^d$, with $d\in\{2,3\}.$ Let $\kappa,\alpha\geq 0,\beta\in]0,1],\,\lambda\geq e^{\frac{3+2\alpha}{\beta}}$, $ p\in[1,\infty]$ and $\theta$ be a smooth solution of $$\partial_t\theta+v\cdot\nabla\theta+\kappa{|\textnormal{D}|}^\beta\log^{-\alpha}(\lambda+{|\textnormal{D}|})\theta=f.$$ Then we have $$\|\theta\|_{\widetilde L^\infty_tB_{p,1}^0}\leq C\Big( \|\theta_0\|_{B_{p,1}^0}+\|f\|_{L^1_tB_{p,1}^0}\Big)\Big( 1+\int_0^t\|\nabla v(\tau)\|_{L^\infty}d\tau \Big),$$ where the constant $C$ depends only on $\lambda$ and $\alpha$.
Proof of Theorem \[coer10\]
===========================
Bernstein inequality
--------------------
This section is devoted to the generalization of the classical Bernstein inequality described in for more general operators.
\[Bernst\] Let $\alpha\in{\mathbb{R}},\beta>0$ and $ \lambda\geq2.$ Then there exists a constant $C$ such that for every $f\in\mathcal{S}({\mathbb{R}}^d)$ and for every $q\geq-1$ and $ p\in[1,\infty]$ we have $$\Big\|\Delta_q \big(\frac{{|\textnormal{D}|}^\beta}{\log^\alpha(\lambda+{|\textnormal{D}|})} f\big)\Big\|_{L^p}\le C2^{q\beta}{(|q|+1)^{-\alpha}} \|\Delta_qf\|_{L^p}.$$ Moreover $$\Big\| S_q \big(\frac{{|\textnormal{D}|}^\beta}{\log^\alpha(\lambda+{|\textnormal{D}|})} f\big)\Big\|_{L^p}\le C2^{q\beta}{(|q|+1)^{-\alpha}} \|S_qf\|_{L^p}.$$
\[rmq6\] The first result of Proposition \[Bernst\] remains true for more general situation where $q\in{\mathbb{N}}$ and the operator ${|\textnormal{D}|}^\beta$ is replaced by $a(\textnormal{D})$ with $a(\xi)$ a homogeneous distribution of order $\beta\in{\mathbb{R}}$ that is $a\in C^\infty({\mathbb{R}}\backslash\{0\})$ and for every $\gamma\in {\mathbb{N}}^d$ $$|\partial_\xi^\gamma a(\xi)|\le C|\xi|^{\beta-|\gamma|}.$$
It is easy to see that $$\Delta_q \big(\frac{{|\textnormal{D}|}^\beta}{\log^\alpha(\lambda+{|\textnormal{D}|})} f\big)=K_q\star\Delta_qf,$$ with $$\widehat{K_q}(\xi)=\frac{\tilde{\phi}(2^{-q}\xi)|\xi|^{\beta}}{{\log^\alpha(\lambda+|\xi|)}}$$ and $\tilde{\phi}$ is a smooth function supported in the ring $\{\frac14\le|x|\le 3\}$ and taking the value 1 on the support of the function $\phi$ introduced in section \[preliminaries\] . By Fourier inversion formula and change of variables we get $$\begin{aligned}
K_q(x)&=&c_d\int_{{\mathbb{R}}^d}e^{ix\cdot\xi}\frac{\tilde{\phi}(2^{-q}\xi)|\xi|^{\beta}}{{\log^\alpha(\lambda+|\xi|)}}d\xi\\
&=&c_d2^{q\beta}2^{qd}\int_{{\mathbb{R}}^d}e^{i2^qx\cdot\xi}\frac{\tilde{\phi}(\xi)|\xi|^{\beta}}{{\log^\alpha(\lambda+2^q|\xi|)}}d\xi\\
&:=&c_d2^{q\beta}2^{qd}\tilde{K_q}(2^q x),\end{aligned}$$ with $$\tilde{K_q}(x)=\int_{{\mathbb{R}}^d}e^{ix\cdot\xi}\frac{\tilde{\phi}(\xi)|\xi|^{\beta}}{{\log^\alpha(\lambda+2^q|\xi|)}}d\xi.$$ Obviously we have $$\|{K_q}\|_{L^1}=c_d2^{q\beta}\|\tilde{K_q}\|_{L^1}.$$ Hence to prove Proposition \[Bernst\] it suffices to establish $$\label{be1}
\|\tilde{K_q}\|_{L^1}\le C{(q+1)^{-\alpha}}.$$ From the definition of $\tilde{K_q}$ we see that $$\tilde{K_q}(x)=\int_{{\mathbb{R}}^d}e^{ix\cdot\xi}\frac{\tilde{\psi}(\xi)}{{\log^\alpha(\lambda+2^q|\xi|)}}d\xi$$ where $\tilde\psi$ belongs to Schwartz class and supported in $\{\frac14\le|x|\le 3\}$. By integration by parts we get for $j\in\{1,2,...,d\}$ $$x_j^{d+1}\tilde{K_q}(x)=(-i)^{d+1}\int_{\frac14\le|\xi|\le 3}e^{ix\cdot\xi}\,\partial_{\xi_j}^{d+1}\Big(\frac{\tilde{\psi}(\xi)}{{\log^\alpha(\lambda+2^q|\xi|)}}\Big)d\xi.$$ Now we claim that $$\Big|\partial_{\xi_j}^{d+1}\Big(\frac{\tilde{\psi}(\xi)}{{\log^\alpha(\lambda+2^q|\xi|)}}\Big) \Big|\le C_{\lambda,\alpha,d}\frac{g(\xi)}{{\log^\alpha(\lambda+2^{q})}},$$ where $g\in \mathcal{S}({\mathbb{R}}^d).$ This is an easy consequence of Leibniz formula and the following fact $$\begin{aligned}
\Big|\partial_{\xi_j}^{n}\Big(\frac{1}{{\log^\alpha(\lambda+2^q|\xi|)}}\Big)\Big|&\le& \sum_{l,k=1}^nc_{l,k}\Big(\frac{2^q}{\lambda+2^q|\xi|}\big)^l\frac{1}{{\log^{\alpha+k}(\lambda+2^q|\xi|)}}\\
&\le&\frac{C_{\lambda,\alpha,n}}{{\log^{\alpha}(\lambda+2^{q})}},\quad\hbox{for}\quad \frac14\le|\xi|\le2.\end{aligned}$$ Thus we get for $j\in\{1,..,d\}$ $$|x_j|^{d+1}|\tilde{K_q}(x)|\le {C}{{\log^{-\alpha}(\lambda+2^{q})}},\,\forall x\in{\mathbb{R}}^d.$$ It follows that $$|x|^{d+1}|\tilde{K_q}(x)|\le {C}{{\log^{-\alpha}(\lambda+2^{q})}},\,\forall x\in{\mathbb{R}}^d.$$ It is easy to see that $\tilde{K_q}$ is continuous and $$|\tilde{K_q}(x)|\le{C}{{\log^{-\alpha}(\lambda+2^{q})}}$$ Consequently, $$|\tilde{K_q}(x)|\le {C}{{\log^{-\alpha}(\lambda+2^{q})}}(1+|x|)^{-d-1},\,\forall x\in{\mathbb{R}}^d.$$ This yields $$\begin{aligned}
\|\tilde{K_q}\|_{L^1}&\le &{C}{{\log^{-\alpha}(\lambda+2^{q})}}\\
&\le& C(q+1)^{-\alpha}.\end{aligned}$$ This concludes the proof of the first case $q\in{\mathbb{N}}$.
We can write in this case the kernel $K_{-1}$ as $$\begin{aligned}
K_{-1}(x)&=&\int_{{\mathbb{R}}^d}e^{ix\cdot\xi}\frac{\tilde{\chi}(\xi)|\xi|^{\beta}}{{\log^\alpha(\lambda+|\xi|)}}d\xi\\
&=&\int_{{\mathbb{R}}^d}e^{ix\cdot\xi}{{\chi_1}(\xi)}d\xi,\end{aligned}$$ where $\tilde\chi$ is a smooth compactly supported function taking the value $1$ on the support of the function $\chi$ introduced in section \[preliminaries\]. The function $\chi_1$ is given by ${\chi_1}(\xi)=\frac{\tilde{\chi}(\xi)|\xi|^{\beta}}{{\log^\alpha(\lambda+|\xi|)}}.$ We can see by easy computations that $\tilde{\chi}$ is smooth outside zero and satisfies for every $\gamma\in{\mathbb{N}}^d,$ $$|\partial_\xi^\gamma\tilde{\chi}(\xi)|\le C_\gamma|\xi|^{\beta-|\gamma|},\quad\forall \xi \neq0.$$ Using Mikhlin-Hörmander theorem we get $$|K_{-1}(x)|\le C|x|^{-d-\beta}.$$ Since $K_{-1}$ is continuous at zero then we have $|K_{-1}(x)|\le C (1+|x|)^{-d-\beta}.$ This proves that $K_{-1}\in L^1$.
To prove the second estimate we use the first result combined with the following $$\begin{aligned}
\Big\|S_q\big(\frac{{|\textnormal{D}|}^\beta}{\log^\alpha(\lambda+{|\textnormal{D}|})} f\big)\Big\|_{L^p}&\le&\sum_{j=-1}^{q+1}\Big\|\Delta_j \big(\frac{{|\textnormal{D}|}^\beta}{\log^\alpha(\lambda+{|\textnormal{D}|})} S_q f\big)\Big\|_{L^p}\\&\le& C\|S_q f\|_{L^p}\sum_{j=-1}^{q+1}2^{j\beta}(|j|+1)^{-\alpha}.\end{aligned}$$ Since $\beta>0$ then the last series diverges and $$\sum_{j=-1}^{q+1}2^{j\beta}(|j|+1)^{-\alpha}\le C2^{q\beta} (|q|+1)^{-\alpha}.$$ This can be deduced from the asymptotic behavior $$\int_{1}^{x} e^{\beta t}t^{-\alpha}dt\approx \frac1\beta e^{\beta x}x^{-\alpha},\quad\hbox{as}\quad x\to+\infty.$$
As a consequence of Proposition \[Bernst\] we get the following result which describes discuss the action of the logarithmic Riesz transform $\mathcal{R}_\alpha=\frac{\partial_1\log^{\alpha}(\lambda+{|\textnormal{D}|})}{{|\textnormal{D}|}}$ on Besov spaces.
\[embed\] Let $\alpha\in{\mathbb{R}},\lambda>1$ and $p\in[1,\infty]$. Then the map $$(\textnormal{Id}-\Delta_{-1})\mathcal{R}_\alpha :B_{p,r}^{s,\alpha}\to B_{p,r}^s$$ is continuous.
Generalized Bernstein inequality
--------------------------------
The main goal of this section is to prove Theorem \[coer10\]. Some preliminaries lemmas will be needed. The first one is a Stroock-Varopoulos inequality for sub-Markovian operators. For the proof see [@Lisk; @Lisk1].
Let $p>1$ and $A$ be a sub-Markovian generator, then we have $$4\frac{p-1}{p^2}\|A^{\frac12}(|f|^{\frac{p}{2}}\textnormal{sign }f)\|_{L^2}^2\le\int_{{\mathbb{R}}^d} (Af)\,|f|^{p-1}\textnormal{sign}\,fdx\leq C_p\|A^{\frac12}(|f|^{\frac{p}{2}}\textnormal{sign }f)\|_{L^2}^2.$$ Moreover the generator $A$ satisfies the first Deurling-Deny condition $$4\frac{p-1}{p^2}\|A^{\frac12}(|f|^{\frac{p}{2}})\|_{L^2}^2\le\int_{{\mathbb{R}}^d} (Af)\,|f|^{p-1}\textnormal{sign }fdx.$$
Combining this result with Proposition \[props23\] we get,
\[coer\] Let $, p>1, \beta\in]0,1],\alpha\geq0$ and $\lambda\geq e^{\frac{3+2\alpha}{\beta}}$. Then we have $$4\frac{p-1}{p^2}\Big\|\frac{{|\textnormal{D}|}^{\frac\beta2}}{\log^{\frac\alpha2}(\lambda+{|\textnormal{D}|})}(|f|^{\frac{p}{2}})\Big\|_{L^2}^2\le\int_{{\mathbb{R}}^d} \Big(\frac{{|\textnormal{D}|}^{\beta}}{\log^{\alpha}(\lambda+{|\textnormal{D}|})}f\Big)\,|f|^{p-1}\textnormal{sign }fdx.$$
We will make use of the following composition results,
\[compos-bes\]
1. Let $\mu\geq1$ and $s\in[0,\mu[\cap [0,2[.$ Then $$\||f|^\mu\|_{B_{2,2}^s}\le C\|f\|_{B_{2\mu,2}^s}\|f\|_{B_{2\mu,2}^0}^{\mu-1}$$
2. $\mu\in]0,1], p,q\in[1,\infty]$ and $0<s<1+\frac1p.$ Then $$\||f|^\mu\|_{B_{\frac p\mu,\frac q\mu}^{s\mu}}\le C\|f\|_{B_{p,q}^s}^\mu.$$
We point out that the first estimate is a particular case of a general result due to Miao [*et al.*]{}, see [@cmz]. The second one is established by Sickel in [@Si], see also Theorem 1.4 of [@kateb].
Next we will recall the following result proved in [@cmz; @dan; @plan],
\[coer01\] Let $d\geq1,\,\beta\in]0,2]$ and $p\geq2.$ Then we have for $q\in{\mathbb{N}}$ and $f\in \mathcal{S}({\mathbb{R}}^d),$ $$2^{q\beta}\|\Delta_q f\|_{L^p}^p\le C\int_{{\mathbb{R}}^d} \big({{|\textnormal{D}|}^{\beta}}\Delta_qf\big)\,|\Delta_q f|^{p-1}\textnormal{sign }\Delta_qfdx.$$ where $C$ depends on $p$ and $\beta$. Moreover, for $\beta=2$ we can extend the above inequality to $p\in]1,\infty[$.
Now we will restate and prove Theorem \[coer10\].
\[coer1\] Let $d\in\{1,2,3\}, \beta\in]0,1],\alpha\geq0, \lambda\geq e^{\frac{3+2\alpha}{\beta}}$ and $p>1.$ Then we have for $q\in{\mathbb{N}}$ and $f\in \mathcal{S}({\mathbb{R}}^d),$ $$2^{q\beta}(q+1)^{-\alpha}\|\Delta_q f\|_{L^p}^p\le C\int_{{\mathbb{R}}^d} \Big(\frac{{|\textnormal{D}|}^{\beta}}{\log^{\alpha}(\lambda+{|\textnormal{D}|})}\Delta_qf\Big)\,|\Delta_q f|^{p-1}\textnormal{sign }\Delta_qfdx.$$ where $C$ depends on $p,\alpha$ and $\lambda$.
Using Corollary \[coer\] it suffices to prove $$C^{-1}2^{q\beta}(q+1)^{-\alpha}\|\Delta_q f\|_{L^p}^p\le \Big\|\frac{{|\textnormal{D}|}^{\frac\beta2}}{\log^{\frac\alpha2}(\lambda+{|\textnormal{D}|})}(|\Delta_qf|^{\frac{p}{2}})\Big\|_{L^2}^2.$$ We will use an idea of Miao [*et al.*]{} [@cmz]. Let $N\in {\mathbb{N}}$ then we have $$\|{|\textnormal{D}|}(|f_q|^{\frac{p}{2}})\|_{L^2}\le\|S_N{|\textnormal{D}|}(|f_q|^{\frac p2})\|_{L^2}+\|\|(\textnormal{Id}-S_N){|\textnormal{D}|}(|f_q|^{\frac p2})\|_{L^2}.$$ It is clear that for $s\geq0$ $$\label{sick2}
\|\|(\textnormal{Id}-S_N){|\textnormal{D}|}(|f_q|^{\frac p2})\|_{L^2}\le C 2^{-N s}\||f_q|^{\frac p2}\|_{ B_{2,2}^{1+s}}.$$ We have now to deal with fraction powers in Besov spaces. We will treat differently the cases $p> 2$ and $p\le2$.
Combining Lemma \[compos-bes\]-(1) with Bernstein inequality we get under the assumption $0<s<\min(\frac p2-1,2),$ $$\begin{aligned}
\||f_q|^{\frac p2}\|_{ B_{2,2}^{1+s}}&\le &C\|f_q\|_{B_{p,2}^{0}}^{\frac p2-1}\|f_q\|_{B_{p,2}^{1+s}}\\
&\le& C2^{q(1+s)}\|f_q\|_{L^p}^{\frac p2}.\end{aligned}$$ Using Lemma \[compos-bes\]-(2) and Bernstein inequality, we get for $0<s<\frac{p-1}{2},$ $$\begin{aligned}
\||f_q|^{\frac p2}\|_{ B_{2,2}^{1+s}}&\le &C\|f_q\|_{B_{p,p}^{\frac{2+2s}{p}}}^{\frac p2}\\
&\le& C2^{q(1+s)}\|f_q\|_{L^p}^{\frac p2}.\end{aligned}$$
It follows from (\[sick2\]) and the previous inequalities that there exists $s_p>0$ such that $$\|\|(\textnormal{Id}-S_N){|\textnormal{D}|}(|f_q|^{\frac p2})\|_{L^2}\le C2^{-N s}2^{q(1+s)}\|f_q\|_{L^p}^{\frac p2}.$$ On the other hand Proposition \[Bernst\] gives $$\begin{aligned}
\|S_N{|\textnormal{D}|}(|f_q|^{\frac p2})\|_{L^2}&\le &\Big\|S_N{{|\textnormal{D}|}^{1-\frac\beta2}}{\log^{\frac\alpha2}(\lambda+{|\textnormal{D}|})}\big(\frac{{|\textnormal{D}|}^{\frac\beta2}}{\log^{\frac\alpha2}(\lambda+{|\textnormal{D}|})}(|f_q|^{\frac p2})\big)\Big\|_{L^2}\\
&\le&C 2^{N(1-\frac\beta2)}N^{\frac\alpha2}\Big\|\frac{{|\textnormal{D}|}^{\frac\beta2}}{\log^{\frac\alpha2}(\lambda+{|\textnormal{D}|})}(|f_q|^{\frac p2})\big)\Big\|_{L^2}.\end{aligned}$$ Therefore we get $$\|{|\textnormal{D}|}(|f_q|^{\frac{p}{2}})\|_{L^2}\le C2^{-N s}2^{q(1+s)}\|f_q\|_{L^p}^{\frac p2}+C 2^{N(1-\frac\beta2)}N^{\frac\alpha2}\Big\|\frac{{|\textnormal{D}|}^{\frac\beta2}}{\log^{\frac\alpha2}(\lambda+{|\textnormal{D}|})}(|f_q|^{\frac p2})\big)\Big\|_{L^2}.$$ According to Proposition \[coer01\] we have for $p\in]1,\infty[$ $$C_p2^q\|f_q\|_{L^p}^{\frac p2}\le
\|{|\textnormal{D}|}(|f_q|^{\frac{p}{2}})\|_{L^2}.$$ Combining both last estimates we get $$2^q\|f_q\|_{L^p}^{\frac p2}\le C2^{s(q-N)}\,2^{q}\|f_q\|_{L^p}^{\frac p2}+C 2^{N(1-\frac\beta2)}N^{\frac\alpha2}\Big\|\frac{{|\textnormal{D}|}^{\frac\beta2}}{\log^{\frac\alpha2}(\lambda+{|\textnormal{D}|})}(|f_q|^{\frac p2})\big)\Big\|_{L^2}$$ We take $N=q+N_0$ such that $C2^{-N_0 s}\le\frac12$ . Then we get $$\|f_q\|_{L^p}^{\frac p2}\le C 2^{-q\frac\beta2}(q+1)^{\frac\alpha2}\Big\|\frac{{|\textnormal{D}|}^{\frac\beta2}}{\log^{\frac\alpha2}(\lambda+{|\textnormal{D}|})}\big(|f_q|^{\frac p2}\big)\Big\|_{L^2}.$$ This gives the desired result.
Commutator estimates
====================
We will establish in this section some commutator estimates. The following result was proved in [@HKR].
\[commu\]Given $(p,m)\in[1,\infty]^2$ such that $p\geq m'$ with $m'$ the conjugate exponent of $m$. Let $f,g$ and $h$ be three functions such that $\nabla f\in L^p, g\in L^m$ and $xh\in L^{m'}$. Then, $$\|h\star(fg)-f(h\star g)\|_{L^p}\leq \|xh\|_{L^{m'}}\|\nabla f\|_{L^p}\|g\|_{L^{m}}.$$
Now we will prove the following lemma.
\[conv\] Let $(a_n)_{n\in{\mathbb{Z}}}$ be a sequence of strictly nonnegative real numbers such that $$M:=\max\Big(\sup_{n\in{\mathbb{Z}}} a_n^{-1}\sum_{j\le n}a_j,\sup_{n\in{\mathbb{Z}}} a_n\sum_{j\geq n}a_j^{-1}\Big)<\infty.$$ Then for every $p\in[1,\infty]$ the linear operator $T:\ell^p\to \ell^p$ defined by $$T((b_n)_{n\in{\mathbb{Z}}})=\Big(\sum_{j\le n}a_j a_n^{-1} b_j\Big)_{n\in{\mathbb{Z}}}$$ is continuous and $\|T\|_{\mathcal{L}(\ell^p)}\le M.$
By interpolation it suffices to prove the cases $p=1$ and $p=+\infty$. Let’s start and denote ${\bf{b}}=(b_n)_{n\in{\mathbb{Z}}}.$ Then from Fubini lemma and the hypothesis $$\begin{aligned}
\|T{\bf b}\|_{\ell^1}&\le& \sum_{n\in{\mathbb{Z}}}\sum_{j\le n}a_j a_n^{-1} |b_j|\\
&\le& \sum_{j\in{\mathbb{Z}}}|b_j|a_j\sum_{n\geq j}a_n^{-1} \\
&\le& M\|{\bf b}\|_{\ell^1}.
\end{aligned}$$ For the case $p=+\infty$ we write $$\begin{aligned}
\|T{\bf b}\|_{\ell^\infty}&\le& \sup_{n\in{\mathbb{Z}}}\sum_{j\le n}a_j a_n^{-1} |b_j|\\
&\le& \|{\bf b}\|_{\ell^\infty}\sup_{n\in{\mathbb{Z}}}a_n^{-1}\sum_{j\leq n}a_j \\
&\le& M\|{\bf b}\|_{\ell^\infty}.
\end{aligned}$$ This completes the proof.
The goal now is to study the commutation between the following operators $$\mathcal{R}_\alpha=\frac{\partial_1}{{|\textnormal{D}|}}\log^\alpha(\lambda+{|\textnormal{D}|})\quad\hbox{and}\quad v \cdot \nabla.$$ Recall that $B_{\infty,2}^{s,s'}$ is the space given by the set of tempered distributions $u$ such that $$\|u\|_{B_{\infty,r}^{s,s'}}=\|\big(2^{qs}(|q|+1)^{s'}\|\Delta_qu\|_{L^\infty}\big)_q\|_{\ell^r}.$$ The main result of this section reads as follows.
\[propcom\] Let $\alpha\in{\mathbb{R}},\lambda>1$, $v$ be a smooth divergence free vector-field and $\theta $ be a smooth scalar function.
1. For every $(p,r)\in [2,\infty[\times[1,\infty]$ there exists a constant $C=C(p,r)$ such that $$\|[\mathcal{R_\alpha}, v\cdot\nabla]\theta\|_{B_{p,r}^0}\le C \|\nabla v\|_{L^p}\big(\|\theta\|_{B_{\infty,r}^{0,\alpha}}+\|\theta\|_{L^p}\big).$$
2. For every $(r,\rho)\in[1,\infty]\times]1,\infty[$ and $\epsilon >0$ there exists a constant $C=C(r,\rho,\epsilon)$ such that
$$\|[\mathcal{R_\alpha}, v\cdot\nabla]\theta\|_{B_{\infty,r}^0}\le C (\|\omega\|_{L^\infty}+\|\omega\|_{L^\rho})\big(\|\theta\|_{B_{\infty,r}^\epsilon}+\|\theta\|_{L^\rho}\big).$$
[**$(1)$**]{} We split the commutator into three parts according to Bony’s decomposition [@b], $$\begin{aligned}
\nonumber[\mathcal{R}_\alpha, v\cdot\nabla]\theta&=&\sum_{q\in{\mathbb{N}}}[\mathcal{R}_\alpha, S_{q-1}v\cdot\nabla]\Delta_q\theta+\sum_{q\in{\mathbb{N}}}[\mathcal{R}_\alpha, \Delta_qv\cdot\nabla]S_{q-1}\theta\\
\nonumber&+&\sum_{q\geq-1} [\mathcal{R}_\alpha, \Delta_qv\cdot\nabla]\widetilde{\Delta}_q\theta\\
\nonumber&=& \sum_{q\in{\mathbb{N}}}\mbox{I}_q+\sum_{q\in{\mathbb{N}}}\mbox{II}_q+\sum_{q\geq-1}\mbox{III}_q\\
&=&\mbox{I}+\mbox{II}+\mbox{III}.\end{aligned}$$ We start with the estimate of the first term $\mbox{I}$. It is easy to see that there exists $\tilde\varphi\in\mathcal{S}$ whose spectum does not meet the origin such that $$\mbox{I}_q(x)=h_q\star( S_{q-1}v\cdot\nabla\Delta_q\theta)-S_{q-1}v\cdot(h_q\star \nabla\Delta_q\theta),$$ where $$\widehat{h_q}(\xi)=i\tilde\varphi(2^{-q}\xi)\frac{\xi_1}{|\xi|}\log^\alpha(\lambda+|\xi|).$$ Applying Lemma \[commu\] with $m=\infty$ we get $$\begin{aligned}
\nonumber
\|\mbox{I}_q\|_{L^p}&\lesssim &\|xh_q\|_{L^1} \|\nabla S_{q-1} v\|_{L^p}\|\Delta_q\nabla\theta\|_{L^\infty}
\\
\label{x1}
&\lesssim& 2^q\|xh_q\|_{L^1}\|\Delta_q\theta\|_{L^\infty}\|\nabla v\|_{L^p}.\end{aligned}$$ We can easily check that $$\|xh_q\|_{L^1}= 2^{-q}\|x\tilde{h}_q\|_{L^1}\quad\hbox{with}\quad \widehat{\tilde{h}_q}(\xi)=i\tilde\varphi(\xi)\frac{\xi_1}{|\xi|}\log^\alpha(\lambda+2^q|\xi|).$$ We can get by a similar way to the proof of Proposition \[Bernst\] $$\|\tilde h_q\|_{L^1}\le C (1+|q|)^\alpha.$$ Thus estimate (\[x1\]) becomes $$\|\mbox{I}_q\|_{L^p}\le C (1+|q|)^\alpha\|\Delta_q\theta\|_{L^\infty}\|\nabla v\|_{L^p}.$$ Combined with the trivial fact $$\Delta_j\sum_{q}\mbox{I}_q= \sum_{|j-q|\le 4}\mbox{I}_q$$ this yields $$\begin{aligned}
\|\mbox{I}\|_{B_{p,r}^0}
&\lesssim&
\Big(\sum_{q\geq-1}\|\mbox{I}_q\|_{L^p}^r\Big)^{\frac1r}
\\
&\lesssim&\|\nabla v\|_{L^p}
\|\theta\|_{B_{\infty,r}^{0,\alpha}}.\end{aligned}$$
Let us move to the second term $\mbox{II}$. As before one writes $$\mbox{II}_q(x)=h_q\star( \Delta_q v\cdot\nabla S_{q-1}\theta)-\Delta_q v\cdot(h_q\star \nabla S_{q-1}\theta),$$ and then we obtain the estimate $$\begin{aligned}
\nonumber\|\mbox{II}_q\|_{L^p}&\lesssim& 2^{-q}(1+|q|)^\alpha\|\Delta_q \nabla v\|_{L^p}\| S_{q-1}\nabla\theta\|_{L^\infty}\\
&\lesssim&\|\nabla v\|_{L^p}\sum_{j\le q-2}\frac{2^{j}(1+|j|)^{-\alpha}}{2^{q}(1+|q|)^{-\alpha}}((1+|j|)^\alpha\|\Delta_j\theta\|_{L^\infty}).\end{aligned}$$ Combined with Lemma \[conv\] this yields $$\|\mbox{II}\|_{B_{p,r}^0}\lesssim \|\nabla v\|_{L^p}\|\theta\|_{B_{\infty,r}^{0,\alpha}}.$$
Let us now deal with the third term $\mbox{III}$. Using the fact that the divergence of $\Delta_q v$ vanishes, then we can write $\mbox{III}$ as $$\begin{aligned}
\mbox{III}&=&\sum_{q\geq 2} \mathcal{R}_\alpha \textnormal{div} (\Delta_qv\, \widetilde{\Delta}_q\theta)- \sum_{q\geq 2} \textnormal{div}(\Delta_qv\,\mathcal{R}_\alpha\widetilde{\Delta}_q\theta)+\sum_{q\leq 1}[\mathcal{R}_\alpha, \Delta_{q}v\cdot\nabla]\widetilde{\Delta}_{q}\theta\\&=&J_1+J_2+J_3.\end{aligned}$$ Using Remark \[rmq6\] we get $$\begin{aligned}
\big\|\Delta_j\mathcal{R}_\alpha\textnormal{div} (\Delta_qv\, \widetilde{\Delta}_q\theta)\big\|_{L^p}\lesssim 2^j(1+|j|)^\alpha\|\Delta_q v\|_{L^p}\|\widetilde{\Delta}_q\theta\|_{L^\infty}.\end{aligned}$$ and since $q\geq2$ $$\begin{aligned}
\big\|\Delta_j \textnormal{div}(\Delta_qv\, \mathcal{R}_\alpha\widetilde{\Delta}_q\theta)\big\|_{L^p}&\lesssim & 2^j\|\Delta_q v\|_{L^p}\|\mathcal{R}_\alpha\widetilde{\Delta}_q\theta\|_{L^\infty}
\\
&\lesssim& 2^j(1+|q|)^{\alpha}\|\Delta_q v\|_{L^p}\|\widetilde{\Delta}_q\theta\|_{L^\infty}.\end{aligned}$$ Therefore we get $$\begin{aligned}
\|\Delta_j (J_1+J_2)\|_{L^p}&\lesssim& \sum_{q\in{\mathbb{N}}\atop q\geq j-4}2^j(1+|q|)^\alpha\|\Delta_q v\|_{L^p}\|\widetilde{\Delta}_q\theta\|_{L^\infty}\\
&\lesssim&\|\nabla v\|_{L^p} \sum_{q\in{\mathbb{N}}\atop q\geq j-4}2^{j-q}(1+|q|)^\alpha\|{\Delta}_q\theta\|_{L^\infty},\end{aligned}$$ where we have again used Bernstein inequality to get the last line. It suffices now to use Lemma \[conv\] $$\|J_1+J_2\|_{B_{p,r}^0}\lesssim \|\nabla v\|_{L^p}\|\theta\|_{B_{\infty,r}^{0,\alpha}}.$$ For the last term $J_3$ we can write $$\sum_{-1\leq q\leq 1}[\mathcal{R}_\alpha, \Delta_{q}v\cdot\nabla]\widetilde{\Delta}_{q}\theta(x)=\sum_{q\leq 1}[\textnormal{div }\widetilde{\chi}(\textnormal{D})\mathcal{R}_\alpha, \Delta_{q}v]\widetilde{\Delta}_{q}\theta(x),$$ where $\widetilde{\chi}$ belongs to $\mathcal{D}({\mathbb{R}}^d)$. From the proof of Proposition \[Bernst\] we get that $\textnormal{div }\widetilde{\chi}(\textnormal{D})\mathcal{R}_\alpha$ is a convolution operator with a kernel $\tilde{h}$ satisfying $$|\tilde{h}(x)|\lesssim (1+|x|)^{-d-1}.$$ Thus $$J_3= \sum_{q\leq 1} \tilde h\star( \Delta_qv\cdot\tilde\Delta_q\theta)-\Delta_qv\cdot(\tilde h\star \tilde\Delta_q\theta).$$ First of all we point out that $\Delta_j J_3=0$ for $j\geq 6$, thus we just need to estimate the low frequencies of $J_3$. Noticing that $x\tilde h$ belongs to $L^{p'}$ for $p'>1$ then using Lemma \[commu\] with $m=p\geq 2$ we obtain $$\begin{aligned}
\|\Delta_j J_3\|_{L^p}&\lesssim& \sum_{q\le 1} \|x\tilde h\|_{L^{p'}} \|\Delta_q \nabla v\|_{L^p}\|\widetilde{\Delta}_q\theta\|_{L^p}\\
&\lesssim& \| \nabla v\|_{L^p}\sum_{-1\leq q\leq 1}\|{\Delta}_q\theta\|_{L^p}.\end{aligned}$$ This yields finally $$\|J_3\|_{B_{p,r}^0}\lesssim \|\nabla v\|_{L^p}\|\theta\|_{L^p}.$$ This completes the proof of the first part of Theorem \[propcom\].
[**$(2)$**]{} The second part can be done in the same way so we will give here just a shorten proof. To estimate the terms ${\rm I}$ and ${\rm II}$ we use two facts: the first one is $\|\Delta_q\nabla u\|_{L^\infty}\approx \|\Delta_q \omega\|_{L^\infty}$ for all $q\in \mathbb N.$ The second one is $$\begin{aligned}
\|\nabla S_{q-1} v\|_{L^\infty}& \lesssim&\|\nabla\Delta_{-1}v\|_{L^\infty}+\sum_{j=0}^{q-2}\|\Delta_j\nabla v\|_{L^\infty}\\
&\lesssim&\|\omega\|_{L^\rho}+ q\|\omega\|_{L^\infty}.\end{aligned}$$ Thus becomes $$\|\hbox{I}_q\|_{L^\infty}\le \|\omega\|_{L^\infty}(1+|q|)^{1+\alpha}\|\Delta_q\theta\|_{L^\infty}$$ and by Proposition \[embed\] $$\begin{aligned}
\|\hbox{I}\|_{B_{\infty,r}^0}&\le&\|\omega\|_{L^\infty}\|\theta\|_{B_{\infty,r}^{0,1+\alpha}}\\
&\le&\|\omega\|_{L^\infty}\|\theta\|_{B_{\infty,\infty}^\epsilon}.\end{aligned}$$ The second term $\hbox{II}$ is estimated as follows $$\begin{aligned}
\|\hbox{II}\|_{B_{\infty,r}^0}&\le&\|\omega\|_{L^\infty}\|\theta\|_{B_{\infty,r}^{0,\alpha}}\\
&\le&\|\omega\|_{L^\infty}\|\theta\|_{B_{\infty,\infty}^\epsilon}\end{aligned}$$
For the remainder term we do strictly the same analysis as before except for $J_3$: we apply Lemma \[commu\] with $p=\infty$ and $m=\rho$ leading to $$\begin{aligned}
\nonumber
\|\Delta_j J_3\|_{L^p}&\lesssim& \sum_{q\le 1} \|x\tilde h\|_{L^{\rho'}} \|\Delta_q \nabla v\|_{L^\infty}\|\widetilde{\Delta}_q\theta\|_{L^\rho}\\
&\lesssim& \| \nabla v\|_{L^\rho}\sum_{-1\leq q\leq 1}\|{\Delta}_q\theta\|_{L^\rho}\\
&\lesssim& \|\omega\|_{L^\rho}\|\theta\|_{L^\rho}.\end{aligned}$$ This ends the proof of the theorem.
Smoothing effects
=================
In this section we will describe some smoothing effects for the model (\[transport-d\]) and focus only on the case $\beta=1.$ Remark that we can obtain similar results for the case $\beta\in]0,1].$ $$\left\{ \begin{array}{ll}
\partial_{t} \theta+v\cdot\nabla \theta+\frac{{{|\textnormal{D}|}}}{\log^\alpha(\lambda+{|\textnormal{D}|})}\vert \theta=f\\
\theta_{| t=0}=\theta^{0}.
\end{array} \right. \tag{${\textnormal{TD}}$}$$ We intend to prove the following smoothing effect.
\[thm99\] Let $\alpha\geq 0, \lambda\geq e^{3+2\alpha}, d\in\{2,3\}$ and $v$ be a smooth divergence-free vector field of ${\mathbb{R}}^d$ with vorticity $\omega$. Then, for every $p\in ]1,\infty[$ there exists a constant $C$ such that $$\sup_{q\in{\mathbb{N}}}2^q(1+q)^{-\alpha}\|\Delta_q \theta\|_{L^1_tL^p}\leq C\| \theta_0\|_{L^p}+C\|\theta_0\|_{L^\infty} \|\omega\|_{L^1_tL^p},$$ for every smooth solution $\theta$ of ${\rm (TD)}$ with zero source term $f$.
For the sake of simplicity we state the result of smoothing effect only for $\beta=1$ but the result remains true under the hypothesis of Proposition \[coer1\].
We start with localizing in frequencies the equation: for $q\geq-1$ we set $\theta_q:=\Delta_q\theta. $ Then $$\partial_t\theta_q+v\cdot\nabla\theta_q+\frac{{{|\textnormal{D}|}}}{\log^{\alpha}(\lambda+{|\textnormal{D}|})}\theta_q=-[\Delta_q, v\cdot\nabla]\theta.$$ Recall that ${\theta}_q$ is real function since the functions involved in the dyadic partition of the unity are radial. Then multiplying the above equation by $|\theta_q|^{p-2}{\theta}_q,$ integrating by parts and using Hölder inequalities we get $$\frac1p\frac{d}{dt}\|\theta_q\|_{L^p}^p+\int_{{\mathbb{R}}^2}\Big(\frac{\vert\textnormal{D}\vert}{\log^{\alpha}(\lambda+{|\textnormal{D}|})}\theta_q\Big) |\theta_q|^{p-2}{\theta}_qdx\leq \|\theta_q\|_{L^p}^{p-1}\|[\Delta_q, v\cdot\nabla]\theta\|_{L^p}.$$ Using Proposition \[coer1\] we get for $q\geq0$ $$c 2^{q}(1+q)^{-\alpha}\|\theta_q\|_{L^p}^p\le\int_{{\mathbb{R}}^2}\Big(\frac{\vert\textnormal{D}\vert}{\log^{\alpha}(\lambda+{|\textnormal{D}|})}\theta_q\Big) |\theta_q|^{p-2}{\theta}_qdx,$$ where $c$ depends on $p$. Inserting this estimate in the previous one we obtain $$\frac1p\frac{d}{dt}\|\theta_q\|_{L^p}^p+c2^q(1+q)^{-\alpha}\|\theta_q\|_{L^p}^p\lesssim \|\theta_q\|_{L^p}^{p-1} \|[\Delta_q, v\cdot\nabla]\theta\|_{L^p}.$$ Thus we find $$\label{est}
\frac{d}{dt}\|\theta_q\|_{L^p}+c2^q(1+q)^{-\alpha} \|\theta_q\|_{L^p}\lesssim \|[\Delta_q, v\cdot\nabla]\theta\|_{L^p}.$$ To estimate the right hand-side we will use the following result, see Proposition 3.3 of [@HKR]. $$\|[\Delta_q, v\cdot\nabla]\theta\|_{L^p}\lesssim \|\nabla v\|_{L^p}\|\theta\|_{B_{\infty,\infty}^{0}}.$$
Combined with this lemma yields $$\begin{aligned}
\frac{d}{dt}\big( e^{ct2^q(1+q)^{-\alpha}}\|\theta_q(t)\|_{L^p}\big)&\lesssim& e^{ct2^q(1+q)^{-\alpha}} \|\nabla v(t)\|_{L^p}\|\theta(t)\|_{B_{\infty,\infty}^0}\\
&\lesssim& e^{ct2^q(1+q)^{-\alpha}} \|\omega(t)\|_{L^p}\|\theta_0\|_{L^\infty}.\end{aligned}$$ To get the last line, we have used the conservation of the $L^\infty$ norm of $\theta$ and the classical fact $$\|\nabla v \|_{L^p } \lesssim \| \omega \|_{L^p}\qquad\forall p\! \in ]1, +\infty[.$$ Integrating the differential inequality we get for $q\in{\mathbb{N}}$ $$\begin{aligned}
\|\theta_q(t)\|_{L^p}&\lesssim& \|\theta_q^0\|_{L^p} e^{-ct2^q(1+q)^{-\alpha}}+\|\theta_0\|_{L^\infty}\int_0^t e^{-c(t-\tau)2^q(1+q)^{-\alpha}} \|\omega(\tau)\|_{L^p}d\tau.\end{aligned}$$ Integrating in time yields finally $$\begin{aligned}
2^q(1+q)^{-\alpha}\|\theta_q\|_{L^1_tL^p}&\lesssim& \|\theta_q^0\|_{L^p} +\|\theta_0\|_{L^\infty}\int_0^t \|\omega(\tau)\|_{L^p}d\tau\\
&\lesssim&\|\theta_0\|_{L^p} +\|\theta_0\|_{L^\infty}\int_0^t \|\omega(\tau)\|_{L^p}d\tau,\end{aligned}$$ which is the desired result.
Proof of Theorem \[theo1\]
===========================
Throughout this section we use the notation $\Phi_k$ to denote any function of the form $$\Phi_k(t)= C_{0}\underbrace{ \exp(...\exp }_{k\,times}(C_0t)...),$$ where $C_{0}$ depends on the involved norms of the initial data and its value may vary from line to line up to some absolute constants. We will make an intensive use (without mentionning it) of the following trivial facts $$\int_0^t\Phi_k(\tau)d\tau\leq \Phi_k(t)\qquad{\rm and}\qquad \exp({\int_0^t\Phi_k(\tau)d\tau})\leq \Phi_{k+1}(t).$$
\[sectionbouss\] The proof of Theorem \[theo1\] will be done in several steps. The first one deals with some [*[*a priori* ]{}* ]{} estimates for the equations . In the second one we prove the uniqueness part. Finally, we will discuss the construction of the solutions at the end of this section.
[A priori estimates]{}
----------------------
In Theorem \[theo1\] we deal with critical regularities and one needs to bound the Lipschitz norm of the velocity in order to get the global persistence of the initial regularities. For this purpose we will proceed in several steps: one of the main steps is to give an $L^\infty$-bound of the vorticity but due to some technical difficulties related to Riesz transforms this will not be done in a straight way. We prove before an $L^p$ estimate for the vorticity with $2<p<\infty$.
### $L^p$-estimate of the vorticity
We intend now to bound the $L^p$-norm of the vorticity and to describe a smoothing effect for the temperature.
\[max-pro\] Let $\alpha\in[0,\frac12], \lambda\geq e^{3+2\alpha}$ and $p\in]2,\infty[$. Let $(v,\theta)$ be a solution of with $\omega^0\in L^p,\, \theta_0\in L^p\cap L^\infty$ and $ \mathcal{R}_\alpha\theta_0\in L^p$. Then for every $\epsilon>0$ $$\|\omega(t)\|_{L^p}+\|\theta\|_{L^1_t B_{p,1}^{1-\epsilon}}\le \Phi_2(t).$$
Applying the transform $\mathcal{R}_\alpha$ to the temperature equation we get $$\label{bm}
\partial_t\mathcal{R}_\alpha\theta+v\cdot\nabla\mathcal{R}_\alpha\theta+\frac{|\textnormal{D}|}{\log^\alpha(\lambda+{|\textnormal{D}|})}\mathcal{R}_\alpha\theta=-[\mathcal{R}_\alpha, v\cdot\nabla]\theta.$$ Since $\frac{|\textnormal{D}|}{\log^\alpha(\lambda+{|\textnormal{D}|})}\mathcal{R}_\alpha=\partial_1,$ then the function $\Gamma:=\omega+\mathcal{R}_\alpha\theta$ satisfies $$\label{bm1}
\partial_t\Gamma+v\cdot\nabla\Gamma=-[\mathcal{R}_\alpha, v\cdot\nabla]\theta.$$ According to the first part of Proposition \[propcom\] applied with $r=2$, $$\big\|[\mathcal{R}_\alpha,v\cdot\nabla]\theta\big\|_{B_{p,2}^{0}}\lesssim\|\nabla v\|_{L^p}\big(\|\theta\|_{B_{\infty,2}^{0,\alpha}}+\|\theta\|_{L^p} \big).$$ Using the classical embedding $B_{p,2}^0\hookrightarrow L^p$ which is true only for $p\in[2,\infty)$ $$\big\|[\mathcal{R}_\alpha,v\cdot\nabla]\theta\big\|_{L^p}\leq\|\nabla v\|_{L^p}\big(\|\theta\|_{B_{\infty,2}^{0,\alpha}}+\|\theta\|_{L^p}\big).$$ Since ${\rm div }\,v=0$ then the $L^p$ estimate applied to the transport equation gives $$\|\Gamma(t)\|_{L^p}\leq\|\Gamma^0\|_{L^p}+\int_0^t\|[\mathcal{R}_\alpha, v\cdot\nabla]\theta(\tau)\|_{L^p}d\tau.$$ Applying Theorem \[max-princ\] to (\[bm\]) yields $$\|\mathcal{R}_\alpha\theta(t)\|_{L^p}\le\|\mathcal{R}_\alpha\theta_0\|_{L^p}+\int_0^t\|[\mathcal{R}_\alpha, v\cdot\nabla]\theta(\tau)\|_{L^p}d\tau.$$ We set $f(t):=\|\omega(t)\|_{L^p}+\|\mathcal{R}_\alpha\theta(t)\|_{L^p}$. Then from the previous estimates we get $$\begin{aligned}
f(t)&\lesssim& \|\Gamma_0\|_{L^p}+\|\mathcal{R}_\alpha\theta_0\|_{L^p}+\int_0^t\|\nabla v(\tau)\|_{L^p}\big(\|\theta(\tau)\|_{B_{\infty,2}^{0,\alpha}}+\|\theta\|_{L^p}\big) d\tau
\\
&\lesssim&f(0)+\int_0^t f(\tau)\big(\|\theta(\tau)\|_{B_{\infty,2}^{0,\alpha}}+\|\theta_0\|_{L^p}\big) d\tau.\end{aligned}$$ We have used here two estimates: the Calderón-Zygmund estimate: for $p\in(1,\infty)$ $$\|\nabla v\|_{L^p}\le C\|\omega\|_{L^p}.$$ The second one is $\|\theta(t)\|_{L^p}\le\|\theta_0\|_{L^p}$ described in Theorem \[max-princ\].
According to Gronwall lemma we get $$\label{gr1}
f(t)\lesssim f(0)e^{C\|\theta_0\|_{L^p} t}e^{C\|\theta\|_{L^1_tB_{\infty,2}^{0,\alpha}}}.$$ Let $N\in{\mathbb{N}}$, then Bernstein inequalities and Theorem \[max-princ\] give $$\begin{aligned}
\nonumber\|\theta\|_{L^1_tB_{\infty,2}^{0,\alpha}}&\le&t\Big\|\Big(\sum_{q< N}(1+|q|)^{2\alpha}\|\Delta_q\theta\|_{L^\infty}^2\Big)^{\frac12}\Big\|_{L^\infty_t}+\|(\hbox{Id}-S_N)\theta\|_{L^1_t B_{\infty,1}^{0,\alpha}}\\
\nonumber&\lesssim&t\|\theta\|_{L^\infty_{t,x}}N^{\frac12+\alpha}+\sum_{q\geq N }(1+|q|)^\alpha\|\Delta_q\theta\|_{L^1_tL^\infty}\\
\nonumber&\lesssim&
t\|\theta_0\|_{L^\infty}N^{\frac12+\alpha}+\sum_{q\geq N }(1+|q|)^\alpha\|\Delta_q\theta\|_{L^1_tL^\infty}\\
&\lesssim& N^{\frac12+\alpha}\|\theta_0\|_{L^\infty} t+\sum_{q\geq N }2^{q\frac2p}(1+|q|)^\alpha\|\Delta_q\theta\|_{L^1_tL^p}.\end{aligned}$$ Using Theorem \[thm99\] then for $p>2$ and $0<\varepsilon<1-\frac2p$ we obtain $$\begin{aligned}
\nonumber
\sum_{q\geq N }(1+|q|)^\alpha2^{q\frac2p}\|\Delta_q\theta\|_{L^1_tL^p}&\lesssim&\sum_{q\geq N }(1+|q|)^{2\alpha}2^{q(\frac2p-1)}\Big( \|\theta_0\|_{L^p}+\|\theta_0\|_{L^\infty}\|\omega\|_{L^1_tL^p}\Big)
\\
&\lesssim&\sum_{q\geq N }2^{q(\frac2p+\varepsilon-1)}\Big( \|\theta_0\|_{L^p}+\|\theta_0\|_{L^\infty}\|\omega\|_{L^1_tL^p}\Big)\\
&\lesssim& \|\theta_0\|_{L^p}+2^{N(-1+\varepsilon+\frac2p)}\|\theta_0\|_{L^\infty}\|\omega\|_{L^1_tL^p}.\end{aligned}$$ Consequently, $$\begin{aligned}
\label{es43}
\nonumber\|\theta\|_{L^1_tB_{\infty,2}^{0,\alpha}}\lesssim N^{\frac12+\alpha}\|\theta_0\|_{L^\infty} t+ \|\theta_0\|_{L^p}+2^{N(-1+\varepsilon+\frac2p)}\|\theta_0\|_{L^\infty}\|\omega\|_{L^1_tL^p}.\end{aligned}$$ We choose $N$ as follows $${N= \Bigg[\frac{\log\big(e+\|\omega\|_{L^1_tL^p}\big)}{(1-\varepsilon-2/p)\log2}\Bigg].}$$ This yields $$\|\theta\|_{L^1_tB_{\infty,2}^{0,\alpha}}\lesssim\|\theta_0\|_{L^\infty\cap L^p}+\|\theta_0\|_{L^\infty} t\log^{\frac12+\alpha}\Big(e+\int_0^t\|\omega(\tau)\|_{L^p}d\tau\Big).$$ Combining this estimate with (\[gr1\]) we get $$\begin{aligned}
\label{ineq23}
\nonumber\|\theta\|_{L^1_tB_{\infty,2}^{0,\alpha}}&\lesssim& \|\theta_0\|_{L^\infty\cap L^p}+\|\theta_0\|_{L^\infty} t\, \log^{\frac12+\alpha}\Big(e+Cf(0)e^{C\|\theta_0\|_{L^p}t}e^{C\|\theta\|_{L^1_tB_{\infty,2}^{0,\alpha}}}\Big)\\
&\le&{C}_0\log^{\frac12+\alpha}(e+f(0))\,(1+t^{\frac32+\alpha})+C\|\theta_0\|_{L^\infty} t\|\theta\|_{L^1_tB_{\infty,2}^{0,\alpha}}^{\frac12+\alpha},\end{aligned}$$ where ${C}_0$ is a constant depending on $\|\theta_0\|_{L^p\cap L^\infty}$.
We will use the following lemma.
Let $a,b> 0$ and $\alpha\in[0,1[$. Let $x\in{\mathbb{R}}_+$ be a solution of the inequality $$(\star)\quad x\le a+b x^\alpha.$$ Then there exists $C:=C(\alpha)$ such that $$x\le C_\alpha(a+b^{\frac{1}{1-\alpha}}).$$
We set $y=a^{-1}x$. Then the inequality $(\star)$ becomes $$y\le 1+ba^{\alpha-1}y^\alpha.$$ We will look for a number $\mu>0$ such that $y\le e^\mu.$ Then it suffices to find $\mu$ such that $$1+ba^{\alpha-1}e^{\mu\alpha}\le e^\mu.$$ It suffices also to find $\mu$ such that $$(1+ba^{\alpha-1})e^{\mu\alpha}= e^\mu.$$ This gives $e^{\mu}=(1+ba^{\alpha-1})^{\frac{1}{1-\alpha}}$. Now we can use the inequality: for every $t,s\geq0$ $$(t+s)^{\frac{1}{1-\alpha}}\le C_\alpha(t^{\frac{1}{1-\alpha}}+s^{\frac{1}{1-\alpha}}).$$ It follows that $$y\le C_\alpha(1+a^{-1}b^{\frac{1}{1-\alpha}}).$$ This yields $$x\le C_\alpha (a+b^{\frac{1}{1-\alpha}})$$
Applying this lemma to (\[ineq23\]) we get for every $t\in{\mathbb{R}}_+$ $$\begin{aligned}
\label{eq34}
\nonumber\|\theta\|_{L^1_tB_{\infty,2}^{0,\alpha}}&\le& C_0 (t^\frac32+t^{\frac{2}{1-2\alpha}})\\
\nonumber&\le&C_0 (1+t^{\frac{2}{1-2\alpha}})\\
&\le&\Phi_1(t).\end{aligned}$$ It follows from (\[gr1\]) $$\begin{aligned}
\label{gr6}
\nonumber f(t)&\le& C_0e^{C_0t^{\frac{2}{1-2\alpha}}}\\
&\le&\Phi_2(t)\end{aligned}$$ Applying Theorem \[thm99\] and (\[gr6\]) we get for every $\epsilon>0, q\in{\mathbb{N}}$ $$\begin{aligned}
\label{gr4}
\nonumber2^{q}(1+|q|)^{-\alpha}\|\Delta_q\theta\|_{L^1_tL^p}&\leq& C_0e^{C_0t^{\frac{2}{1-2\alpha}}}\\
&\le&\Phi_2(t).\end{aligned}$$ The estimate (\[ineq23\]) becomes $$\|\theta\|_{L^1_tB_{\infty,2}^{0,\frac12}}\le {C}_0\log(e+f(0))(1+t^{2})+C\|\theta_0\|_{L^\infty} t\|\theta\|_{L^1_tB_{\infty,2}^{0,\frac12}},$$ with ${C}_0$ a constant depending on $\|\theta_0\|_{L^p\cap L^\infty}$. Hence if we choose $t$ small enough such that $$\label{eq35}
C\|\theta_0\|_{L^\infty} t=\frac12,$$ then $$\|\theta\|_{L^1_tB_{\infty,2}^{0,\alpha}}\le {C}_0\log(e+f(0)).$$ From we get that $$f(t)\le{C}_0(e+f(0))^{C_0}.$$ Now let $t$ be a given positive time and choose a partition $(t_i)_{i=1}^N$ of $[0,t]$ such that $$\label{parti}
C\|\theta_0\|_{L^\infty} (t_{i+1}-t_i)\approx\frac12.$$ Set $a_i:=\int_{t_i}^{t_{i+1}}\|\theta(\tau)\|_{B_{\infty,2}^{0,\frac12}}d\tau$ and $b_i=f(t_i)$. Thus reproducing similar computations to (\[ineq23\]) yields $$\begin{aligned}
a_i\le{C}_0\log(e+b_i)(1+(t_{i+1}-t_i)^{2})+C\|\theta_0\|_{L^\infty}(t_{i+1}-t_i)a_i.\end{aligned}$$ Hence we get $$\label{recc1}
a_i\le {C}_0\log(e+b_i).$$ The analogous estimate to (\[gr1\]) is $$\begin{aligned}
\label{recc2}
\nonumber b_{i+1}&\lesssim& b_i e^{C(t_{i+1}-t_i)\|\theta_0\|_{L^p}}e^{Ca_i}\\
&\le&C_0 b_i e^{Ca_i}.\end{aligned}$$ Combining (\[recc1\]) and (\[recc2\]) yields $$b_{i+1}\le C_0(e+b_i)^{C_0}.$$ By induction we can prove that for every $i\in\{1,..,N\}$ we have $$b_i\le C_0e^{\exp{C_0 i}}$$ and consequently from $$a_i\le C_0e^{C_0 i}.$$ It follows that $$\begin{aligned}
\|\theta\|_{L^1_tB_{\infty,2}^{0,\frac12}}&=&\sum_{i=^1}^Na_i\\
&\le&C_0e^{C_0 N}\\
&\leq& C_0e^{C_0 t}.\end{aligned}$$ We have used in the last inequality the fact that $$N\approx C_0 t$$ which is a consequence of (\[parti\]). We have also obtained $$f(t)\le C_0e^{\exp{C_0 t}}.$$ It is not hard to see that from one can obtain that for every $s<1$ $$\label{x5}
\|\theta\|_{ L^1_tB_{p,1}^s}\leq \|\theta\|_{ \widetilde{ L}^1_tB_{p,\infty}^{1,-\alpha}} \leq \Phi_2(t).$$
This ends the proof of Proposition \[max-pro\].
Combining with Bernstein inequalities and the fact that $p>2$ this yields $$\label{x9}
\|\theta\|_{ L^1_tB_{\infty,1}^\epsilon}\leq
\Phi_2(t),$$ for every $\epsilon <1-\frac2p$.
### $L^\infty$-bound of the vorticity.
We will prove the following result.
\[pr0\] Let $\alpha\in[0,\frac12], \lambda\geq e^{3+2\alpha}, p\in]2,\infty[$ and $(v,\theta)$ be a smooth solution of the system such that $\omega^0,\theta_0,\mathcal{R}_\alpha\theta_0\in L^p\cap L^\infty$. Then we have $$\label{x10}
\|\omega(t)\|_{L^\infty}+\|\mathcal{R}_\alpha\theta(t)\|_{L^\infty}\le \Phi_3(t)$$ and $$\label{x11}
\|v(t)\|_{L^\infty}\le \Phi_4(t).$$
By using the maximum principle for the transport equation , we get $$\|\Gamma(t)\|_{L^\infty}\leq\|\Gamma^0\|_{L^\infty}+\int_0^t\|[\mathcal{R}_\alpha, v\cdot\nabla]\theta(\tau)\|_{L^\infty}
d\tau.$$ Since the function $\mathcal{R}_\alpha\theta$ satisfies the equation $$\label{rtheta}
\big(\partial_t+v\cdot\nabla+{|\textnormal{D}|}\log^{-\alpha}(\lambda+{|\textnormal{D}|})\big)\mathcal{R}_\alpha\theta=-[\mathcal{R}_\alpha,v\cdot\nabla]\theta,$$ then using Theorem \[max-princ\] we get $$\|\mathcal{R}_\alpha\theta(t)\|_{L^\infty}\leq\|\mathcal{R}_\alpha\theta(t)\|_{L^\infty}+\int_0^t\|[\mathcal{R}_\alpha, v\cdot\nabla]\theta(\tau)\|_{L^\infty}
d\tau.$$ Thus we obtain $$\begin{aligned}
\|\Gamma(t)\|_{L^\infty}+\|\mathcal{R}_\alpha\theta(t)\|_{L^\infty}&\leq&\|\Gamma^0\|_{L^\infty}+\|\mathcal{R}_\alpha\theta_0\|_{L^\infty}+2\int_0^t\|[\mathcal{R}_\alpha, v\cdot\nabla]\theta(\tau)\|_{L^\infty}
d\tau
\\
&\leq & C_0+\int_0^t \|[\mathcal{R}_\alpha, v\cdot\nabla]\theta(\tau)\|_{B_{\infty,1}^0}
d\tau.\end{aligned}$$ It follows from Theorem \[max-princ\], Proposition \[propcom\]-(2) and Proposition \[max-pro\] $$\begin{aligned}
\|\omega(t)\|_{L^\infty}+\|\mathcal{R}_\alpha\theta(t)\|_{L^\infty}&\lesssim&C_0+\int_0^t\|\omega(\tau)\|_{L^\infty\cap L^p}\big(\|\theta(\tau)\|_{B_{\infty,1}^\epsilon}+\|\theta(\tau)\|_{L^p}\big)
d\tau
\\
&\lesssim&C_0+\|\omega\|_{L^\infty_t L^p}\big(\|\theta\|_{L^1_tB_{\infty,1}^\epsilon}+t\|\theta_0\|_{L^p}\big)\\
&+&\int_0^t\|\omega(\tau)\|_{L^\infty}\big(\|\theta(\tau)\|_{B_{\infty,1}^\epsilon}+\|\theta_0\|_{L^p}\big)
d\tau.\end{aligned}$$ Let $0<\epsilon<1-\frac2p$ then using we get $$\|\omega(t)\|_{L^\infty}+\|\mathcal{R}_\alpha\theta(t)\|_{L^\infty}\lesssim\Phi_2(t)+\int_0^t\|\omega(\tau)\|_{L^\infty}\big(\|\theta(\tau)\|_{B_{\infty,1}^\epsilon}+\|\theta_0\|_{L^p}\big)d\tau.$$ Therefore we obtain by Gronwall lemma and $$\begin{aligned}
\|\omega(t)\|_{L^\infty}+\|\mathcal{R}_\alpha\theta(t)\|_{L^\infty}\leq \Phi_3(t).\end{aligned}$$
[**]{} Let $N\in{\mathbb{N}}$ to be chosen later. Using the fact that $\|\dot\Delta_qv\|_{L^\infty}\approx 2^{-q}\|\dot\Delta_q\omega\|_{L^\infty}$, then we get$$\begin{aligned}
\|v(t)\|_{L^\infty}&\leq& \|\chi(2^{N}{|\textnormal{D}|})v(t)\|_{L^\infty}+\sum_{q\geq-N}2^{-q}\|\dot\Delta_q\omega(t)\|_{L^\infty}\\
\\
&\le& \|\chi(2^{N}{|\textnormal{D}|})v(t)\|_{L^\infty}+2^N\|\omega(t)\|_{L^\infty}.\end{aligned}$$ Applying the frequency localizing operator to the velocity equation we get $$\chi(2^{N}{|\textnormal{D}|})v=\chi(2^{N}{|\textnormal{D}|})v_0+\int_0^t\mathcal P\chi(2^{N}{|\textnormal{D}|})\theta(\tau)d\tau+\int_0^t\mathcal P\chi(2^{N}{|\textnormal{D}|}){\rm div}(v\otimes v)(\tau)d\tau,$$ where $\mathcal P$ stands for Leray projector. From Lemma \[lb\], Calderón-Zygmund estimate and the uniform boundness of $\chi(2^{N}{|\textnormal{D}|})$ we get $$\begin{aligned}
\int_0^t\|\chi(2^{N}{|\textnormal{D}|})\mathcal P\theta(\tau)\|_{L^\infty}d\tau&\lesssim&2^{-N\frac2p} \int_0^t\|\theta(\tau)\|_{L^p}d\tau\\
&\lesssim&t\|\theta_0\|_{L^p}.\end{aligned}$$ Using Proposition \[cor1\]-(2) we find $$\int_0^t\|\mathcal P\chi(2^{N}{|\textnormal{D}|}){\rm div}(v\otimes v)(\tau)\|_{L^\infty}d\tau\lesssim 2^{N}\int_0^t\|v(\tau)\|_{L^\infty}^2d\tau.$$ The outcome is $$\begin{aligned}
\|v(t)\|_{L^\infty}&\lesssim& \|v_0\|_{L^\infty}+t \| \theta_0\|_{L^p}+ 2^{-N}\int_0^t \| v(\tau)\|_{L^\infty}^2d\tau+2^N\|\omega(t)\|_{L^\infty}
\\
&\lesssim& 2^{-N}\int_0^t \| v(\tau)\|_{L^\infty}^2d\tau+ 2^N \Phi_3(t) \end{aligned}$$ Choosing judiciously $N$ we find $$\|v(t)\|_{L^\infty}\le\Phi_3(t)\Big(1+\Big(\int_0^t\|v(\tau)\|_{L^\infty}^2d\tau\Big)^{\frac12}\Big).$$From Gronwall lemma we get $$\|v(t)\|_{L^\infty}\le \Phi_4(t).$$
### Lipschitz bound of the velocity
Now we will establish the following result.
\[pr10\] Let $\alpha\in[0,\frac12],\lambda\geq e^{3+2\alpha}, p\in]2,\infty[$ and $(v,\theta)$ be a smooth solution of the system with $\omega^0,\theta_0,\mathcal{R}_\alpha\theta_0\in B_{\infty,1}^0\cap L^p$. Then $$\|\mathcal{R}_\alpha\theta(t)\|_{B_{\infty,1}^0}+\|\omega(t)\|_{B_{\infty,1}^0}+\|v(t)\|_{B_{\infty,1}^1}\leq \Phi_4(t).$$
Applying Corollary \[thmlog\] to the equations and , we obtain $$\label{eqlog}
\|\Gamma(t)\|_{B_{\infty,1}^0}+\|\mathcal R_\alpha\theta(t)\|_{B_{\infty,1}^0} \lesssim\Big(C_0+\big\|[\mathcal{R}_\alpha, v\cdot\nabla]\theta\big\|_{L^1_tB_{\infty,1}^0}\Big)\Big(1+\|\nabla v\|_{L^1_tL^\infty} \Big).$$ Thanks to Theorem \[propcom\], Propositions \[pr0\], \[max-pro\] and we get $$\begin{aligned}
\big\|[\mathcal{R}_\alpha, v\cdot\nabla]\theta\big\|_{L^1_tB_{\infty,1}^0}&\lesssim &\int_0^t(\|\omega(\tau)\|_{L^\infty}+\|\omega(\tau)\|_{L^p})\big(\|\theta(\tau)\|_{B_{\infty,1}^\epsilon}+\|\theta(\tau)\|_{L^p}\big)d\tau
\\
&\lesssim & \Phi_3(t).\end{aligned}$$ By easy computations we get $$\begin{aligned}
\nonumber
\|\nabla v\|_{L^\infty}&\le& \|\nabla\Delta_{-1}v\|_{L^\infty}+\sum_{q\in{\mathbb{N}}}\|\Delta_q\nabla v\|_{L^\infty}
\\
\nonumber
&\lesssim& \|\omega\|_{L^p}+\sum_{q\in{\mathbb{N}}}\|\Delta_q\omega\|_{L^\infty}
\\
\label{x8}
&\lesssim& \Phi_2(t)+\|\omega(t)\|_{B_{\infty,1}^0}.\end{aligned}$$ Putting together and leads to $$\begin{aligned}
\|\omega(t)\|_{B_{\infty,1}^0}\leq \|\Gamma(t)\|_{B_{\infty,1}^0}+\|\mathcal R_\alpha\theta(t)\|_{B_{\infty,1}^0}\leq \Phi_3(t)\Big(1+\int_0^t\|\omega(\tau)\|_{B_{\infty,1}^0}d\tau\Big).\end{aligned}$$ Thus we obtain from Gronwall inequality $$\label{x12}
\|\omega(t)\|_{B_{\infty,1}^0}+\|\mathcal R_\alpha\theta(t)\|_{B_{\infty,1}^0}\leq\Phi_4(t).$$ Coming back to we get $$\begin{aligned}
\nonumber
\|\nabla v(t)\|_{L^\infty}\leq \Phi_4(t).\end{aligned}$$ Let us move to the estimate of $v$ in the space $B_{\infty,1}^1$. By definition we have $$\|v(t)\|_{B_{\infty,1}^1}\lesssim \|v(t)\|_{L^\infty}+\|\omega(t)\|_{B_{\infty,1}^0}.$$ Combined with and this yields $$\|v(t)\|_{B_{\infty,1}^1}\leq \Phi_4(t)$$ The proof of is now achieved.
Uniqueness
----------
We will show that the Boussinesq system has a unique solution in the following function space $$\mathcal{E}_T= (L^\infty_TB_{\infty,1}^0\cap L^1_TB_{\infty,1}^1)\times (L^\infty_TL^p\cap \widetilde L^1_T B_{p,\infty}^{1,-\alpha}),\quad 2<p<\infty.$$ Let $(v^1,\theta^1)$ and $(v^2,\theta^2)$ be two solutions of belonging to the space $\mathcal{E}_T$ and denote $$v=v^2-v^1,\quad\theta=\theta^2-\theta^1.$$ Then we get $$\left\{
\begin{array}{ll}
\partial_t v+v^2\cdot\nabla v=-\nabla \pi-v\cdot\nabla v^1+\theta e_2\\
\partial_t\theta+v^2\cdot\nabla \theta+\frac{|\textnormal{D}|}{\log^\alpha(\lambda+{|\textnormal{D}|})}\theta=-v\cdot\nabla\theta^1\\
v_{| t=0}=v_0, \quad \theta_{| t=0}=\theta_0.
\end{array} \right.$$ According to Proposition \[prop-Bes\] we have $$\|v(t)\|_{B_{\infty,1}^0}\le Ce^{CV_1(t)}\Big(\|v_0\|_{B_{\infty,1}^0}+\|\nabla \pi\|_{L^1_tB_{\infty,1}^0}+\| v\cdot\nabla v^1\|_{L^1_tB_{\infty,1}^0}+\|\theta\|_{L^1_tB_{\infty,1}^0}\Big),$$ with $V_1(t)=\|\nabla v^1\|_{L^1_tL^\infty}.$ Straightforward computations using the incompressibility of the flows gives $$\begin{aligned}
\nabla \pi&=&-\nabla\Delta^{-1}\textnormal{div }(v\cdot\nabla(v^1+v^2))+\nabla\Delta^{-1}\partial_2\theta\\
&=&\hbox{I}+\hbox{II}.\end{aligned}$$ To estimate the first term of the RHS we use the definition $$\|\hbox{I}\|_{B_{\infty,1}^0}\lesssim\|(\nabla\Delta^{-1}\textnormal{div })\textnormal{div }\Delta_{-1}(v\otimes (v^1+v^2))\|_{L^\infty}+\|v\cdot\nabla(v^1+v^2)\|_{B_{\infty,1}^1}$$ From Proposition 3.1-(2) of [@HKR] and Besov embeddings we have $$\begin{aligned}
\|(\nabla\Delta^{-1}\textnormal{div })\textnormal{div }\Delta_{-1}(v\otimes (v^1+v^2))\|_{L^\infty}&\lesssim& \|v\otimes (v^1+v^2)\|_{L^\infty}\\
&\lesssim&\|v\|_{B_{\infty,1}^0} \|v^1+v^2\|_{B_{\infty,1}^0}.\end{aligned}$$ Using the incompressibility of $v$ and using Bony’s decomposition one can easily obtain $$\|v\cdot\nabla(v^1+v^2)\|_{B_{\infty,1}^0}\lesssim \|v\|_{B_{\infty,1}^0}\|v^1+v^2\|_{B_{\infty,1}^1}.$$ Putting together these estimates yields $$\label{s12}
\|\hbox{I}\|_{B_{\infty,1}^0}\lesssim \|v\|_{B_{\infty,1}^0}\|v^1+v^2\|_{B_{\infty,1}^1}.$$ Let us now show how to estimate the second term $\hbox{II}$. By using Besov embeddings and Calderón-Zygmund estimate we get $$\begin{aligned}
\|\hbox{II}\|_{B_{\infty,1}^0}&\lesssim& \|\nabla\Delta^{-1}\partial_2\theta\|_{B_{p,1}^{\frac2p}}\\
&\lesssim& \|\theta\|_{B_{p,1}^{\frac2p}}.
\end{aligned}$$ Combining this estimate with (\[s12\]) yields $$\begin{aligned}
\label{1170}
\nonumber\|v(t)\|_{B_{\infty,1}^0}\lesssim e^{CV(t)}\Big(\|v_0\|_{B_{\infty,1}^0}&+&\int_0^t\| v(\tau)\|_{B_{\infty,1}^0}\big[1+\|(v^1,v^2)(\tau)\|_{B_{\infty,1}^1}\big]d\tau\Big)\\&+&e^{CV(t)}
\|\theta\|_{L^1_tB_{p,1}^{\frac2p}},
\end{aligned}$$ where $V(t):=\|(v^1,v^2)\|_{L^1_tB_{\infty,1}^1}.$
Now we have to estimate $\|\theta\|_{L^1_tB_{p,1}^{\frac2p}}$. By applying $\Delta_q$ to the equation of $\theta $ and arguing similarly to the proof of Theorem \[thm99\] we obtain for $q\in{\mathbb{N}}$ $$\begin{aligned}
\nonumber\|\theta_q(t)\|_{L^p} &\lesssim& e^{-ct2^q(1+q)^{-\alpha}}\|\theta_q^0\|_{L^p}+\int_0^t e^{-c2^q(1+q)^{-\alpha}(t-\tau)} \|\Delta_q(v\cdot\nabla \theta^1)(\tau)\|_{L^p}d\tau\\
&+&\int_0^te^{-c2^q(1+q)^{-\alpha}(t-\tau)} \|\big[v^2\cdot\nabla,\Delta_q\big]\theta(\tau) \|_{L^p}d\tau.\end{aligned}$$ Remark, first, that an obvious Hölder inequality yields that for every $\varepsilon\in[0,1]$ there exists an absolute constant $C$ such that $$\int_0^t e^{-c\tau2^q(1+q)^{-\alpha}}d\tau\leq C t^{1-\varepsilon}2^{-q\varepsilon}(1+q)^{\alpha\,\varepsilon},\qquad \forall\, t\geq 0.$$ Using this fact and integrating in time $$\begin{aligned}
\label{113}
\nonumber2^{q\frac2p}\|\theta_q\|_{L^1_tL^p}& \lesssim & (q+1)^{\alpha}2^{q(-1+\frac2p)}\|\theta_q^0\|_{L^p}\\
\nonumber&+& t^{1-\varepsilon} (q+1)^{\alpha\varepsilon}2^{q(-\varepsilon+\frac2p)}\int_0^t \Big( \|\Delta_q(v\cdot\nabla \theta^1)(\tau)\|_{L^p}+ \|[v^2\cdot\nabla,\Delta_q]\theta(\tau) \|_{L^p}\Big)d\tau\\
&=& (q+1)^{\alpha}2^{q(-1+\frac2p)}\|\theta_q^0\|_{L^p}+{\rm I}_q(t)+{\rm II}_q(t).\end{aligned}$$ Using Bony’s decomposition we get easily $$\begin{aligned}
\|\Delta_q(v\cdot\nabla \theta^1)(t)\|_{L^p}& \lesssim& \|v(t)\|_{L^\infty}\sum_{j\leq q+2} 2^{j} \|\Delta_j\theta^1(t)\|_{L^p}\\
&+& 2^q \|v(t)\|_{L^\infty}\sum_{j\geq q-4} \|\Delta_j\theta^1(t)\|_{L^p}\\
&\lesssim&\|v(t)\|_{L^\infty}\sum_{j\leq q+2}(1+|j|)^\alpha \big(2^{j}(1+|j|)^{-\alpha} \|\Delta_j\theta^1(t)\|_{L^p}\big)\\
&+& \|v(t)\|_{L^\infty}\sum_{j\geq q-4} 2^{q-j}(1+|j|)^{\alpha} \big(2^j(1+|j|)^{-\alpha} \|\Delta_j\theta^1(t)\|_{L^p}\big).
\end{aligned}$$ Integrating in time we get $$\begin{aligned}
\label{114}
\nonumber {\rm{I}}_q (t) & \lesssim& t^{1-\varepsilon} \|v\|_{L^\infty_tL^\infty}2^{q(\frac2p-\varepsilon)}(q+1)^{1+\alpha(1+\varepsilon)}\|\theta^1\|_{\widetilde L^1_tB_{p,\infty}^{1,-\alpha}}\\
\nonumber&+& t^{1-\varepsilon} \|v\|_{L^\infty_tL^\infty}\|\theta^1\|_{\widetilde L^1_tB_{p,\infty}^{1,-\alpha}}2^{q(\frac2p+1-\varepsilon)}(q+1)^{\alpha(1+\varepsilon)}\sum_{j\geq q-4}2^{-j} (1+|j|)^{\alpha} \\
&\lesssim&t^{1-\varepsilon} \|v\|_{L^\infty_tL^\infty}2^{q(\frac2p-\varepsilon)}(q+1)^{1+\alpha(1+\varepsilon)}\|\theta^1\|_{\widetilde L^1_tB_{p,\infty}^{1,-\alpha}}.
\end{aligned}$$ To estimate the term ${\rm II}_q$ we use the following classical commutator ( since $2/p<1$), see [@che1] $$\|[v^2\cdot\nabla,\Delta_q]\theta \|_{L^p}\lesssim 2^{-q\frac2p}\|\nabla v^2\|_{L^\infty}\|\theta\|_{B_{p,1}^{\frac2p}}.$$
Thus we obtain, $$\label{115}
{\rm II}_q (t) \lesssim t^{1-\varepsilon}(q+1)^{\alpha\varepsilon} 2^{-q\varepsilon}\|\nabla v^2\|_{L^\infty_tL^\infty}\|\theta\|_{L^1_tB_{p,1}^{\frac2p}}.$$
We choose $\varepsilon\in]0,1[$ such that $\frac2p-\varepsilon<0,$ which is possible since $p>2.$ Then combining (\[113\]), (\[114\]) and (\[115\]) we get $$\begin{aligned}
\|\theta\|_{L^1_tB_{p,1}^{\frac2p}}&\lesssim & \|\theta_0\|_{L^p}+t^{1-\varepsilon} \|v\|_{L^\infty_tL^\infty}\|\theta^1\|_{\widetilde L^1_tB_{p,\infty}^{1,-\alpha}}+t^{1-\varepsilon}\|\nabla v^2\|_{L^\infty_tL^\infty}\|\theta\|_{L^1_tB_{p,1}^{\frac2p}}.\end{aligned}$$ It follows that there exists small $\delta>0$ such that for $t\in[0,\delta]$ $$\|\theta\|_{L^1_tB_{p,1}^{\frac2p}}\lesssim \|\theta_0\|_{L^p}+t^{1-\varepsilon} \|v\|_{L^\infty_tL^\infty}\|\theta^1\|_{\widetilde L^1_tB_{p,\infty}^{1,-\alpha}}.$$ Plugging this estimate into we find $$\|v\|_{L^\infty_tB_{\infty,1}^0}\lesssim e^{CV(t)}\Big(\|v_0\|_{B_{\infty,1}^0}+\|\theta_0\|_{L^p}+t\| v\|_{L^\infty_tB_{\infty,1}^0}+ t^{\varepsilon} \|v\|_{L^\infty_tL^\infty}\|\theta^1\|_{\widetilde L^1_tB_{p,\infty}^{1,-\alpha}}\Big).$$ If $\delta$ is sufficiently small then we get for $t\in[0,\delta]$ $$\label{uni1}
\|v\|_{L^\infty_tB_{\infty,1}^0}\lesssim \|v_0\|_{B_{\infty,1}^0}+\|\theta_0\|_{L^p}.$$ This gives in turn $$\label{uni2}
\|\theta\|_{L^1_tB_{p,1}^{\frac2p}}\lesssim \|v_0\|_{B_{\infty,1}^0}+\|\theta_0\|_{L^p}.$$ This gives in particular the uniqueness on $[0,\delta]$. Iterating this argument yields the uniqueness in $[0,T]$.
Existence
---------
We consider the following system $$\left\{ \begin{array}{ll}
\partial_{t}v_n+v_n\cdot\nabla v_n+\nabla \pi_n=\theta_n e_{2}\\
\partial_{t}\theta_n+v_n\cdot\nabla\theta_n+\frac{\vert \textnormal{D}}{\log^\alpha(\lambda+{|\textnormal{D}|})}\theta_n=0\\
\textnormal{div}v_n=0\\
{v_n}_{| t=0}=S_nv^{0}, \quad {\theta_n}_{| t=0}=S_n\theta^{0}.
\end{array} \right. \tag{B$_{n}$}$$ By using the same method as [@hk] we can prove that this system has a unique local smooth solution $(v_n,\theta_n)$. The global existence of these solutions is governed by the following criterion: we can push the construction beyond the time $T$ if the quantity $\|\nabla v_n\|_{L^1_T L^\infty}$ is finite. Now from the [*a priori*]{} estimates the Lipschitz norm can not blow up in finite time and then the solution $(v_n,\theta_n)$ is globally defined. Once again from the [*a priori*]{} estimates we have for $2<p<\infty$ $$\|v_n\|_{ L^\infty_TB_{\infty,1}^1}+\|\omega_n\|_{L^\infty_T L^p}+\|\theta_n\|_{L^\infty_T\mathcal{X}_p}\leq \Phi_4(T).$$ The space $\mathcal{X}_p$ was introduced before the statement of Theorem \[theo1\] It follows that up to an extraction the sequence $(v_n,\theta_n)$ is weakly convergent to $(v,\theta)$ belonging to $L^\infty_TB_{\infty,1}^1\times L^\infty_T\mathcal{X}_p,$ with $\omega\in L^\infty_T L^p$. For $(n,m)\in{\mathbb{N}}^2$ we set $v_{n,m}=v_n-v_m$ and $\theta_{n,m}=\theta_n-\theta_m$ then according to the estimate (\[uni1\]) and (\[uni2\]) we get for $T=\delta$ $$\|v_{n,m}\|_{L^\infty_TB_{\infty,1}^0}+\|\theta_{n,m}\|_{L^1_TB_{p,1}^{\frac2p}}\lesssim \|S_nv_0-S_m v_0\|_{B_{\infty,1}^0}+\|S_n\theta_0-S_m \theta_0\|_{L^p}.$$ This shows that $(v_n,\theta_n)$ is a Cauchy sequence in the Banach space $L^\infty_TB_{\infty,1}^0\times L^1_TB_{p,1}^{\frac2p}$ and then it converges strongly to $(v,\theta).$ This allows to pass to the limit in the system $({\rm B}_n)$ and then we get that $(v,\theta)$ is a solution of the Boussinesq system (\[Bouss\]).
[10]{}
R. Askey, [*Radial Characteristic Functions.*]{} University of Wisconsin-Madison, Mathematics Research Center, 1262, 1973. <span style="font-variant:small-caps;">J. T. Beale, T. Kato and A. Majda</span>, *Remarks on the Breakdown of Smooth Solutions for the 3-D Euler Equations*. Commun. Math. Phys. **94**, p. 61-66, (1984). S. Bochner, [*Lectures on Fourier integrals.*]{} Translated by Morris Tenenbaum and Harry Pollard. Annals of Mathematics Studies, No. 42 Princeton University Press, Princeton, N.J. 1959. <span style="font-variant:small-caps;">J.-M. Bony</span>, *Calcul symbolique et propagation des singularités pour les équations aux dérivées partielles non linéaires*. Ann. de l’Ecole Norm. Sup., **14**, p. 209-246, (1981).
<span style="font-variant:small-caps;">D. Chae</span>,[*Global regularity for the $2$-D Boussinesq equations with partial viscous terms,*]{} Advances in Math., [**203**]{}, 2 (2006) 497-513. <span style="font-variant:small-caps;">J.-Y. Chemin</span>, *Perfect incompressible fluids*. Oxford University Press 1998.
<span style="font-variant:small-caps;">Q. Chen, C. Miao, Z, Zhang</span>, *A new Bernstein inequality and the 2D dissipative quasi-geostrophic equation*, Commun. Math. Phys. [**271**]{}, 821-838 (2007). <span style="font-variant:small-caps;">A. Córdoba, D. Córdoba</span>,[ *A maximum principle applied to quasi-geostrophic equations*]{}; Comm. Math. Phys., **249**, p. 511-528 (2004). R. Danchin: [*Local theory in critical spaces for compressible viscous and heat conductive gases,*]{} Communications in Partial Differential Equations, 26, pages 1183-1233 (2001). R. Danchin, M. Paicu, *Global well-posedness issues for the inviscid Boussinesq system with Yudovich’s type data.* arXiv:0806.4081\[math.AP\] 25 Jun (2008). <span style="font-variant:small-caps;">R. Danchin, M. Paicu</span>, *Global existence results for the anistropic Boussinesq system in dimension two*. arXiv:0809.4984v1 \[math.AP\] 29 Sep (2008). T. Gneiting, [*Criteria of Pólya type for radial positive definite functions.*]{} Proc. Amer. Math. Soc. 129 (2001), no. 8, 2309-2318. <span style="font-variant:small-caps;">T. Hmidi, S. Keraani</span> , *On the global well-posedness of the Boussinesq system with zero viscosity*, to appear in Indiana. J. Math.
<span style="font-variant:small-caps;">T. Hmidi, S. Keraani</span>, [*Incompressible viscous flows in borderline Besov spaces* ]{}, Arch. for Rational Mech. and Analysis [**189**]{} (2008), no 2, 283-300.
<span style="font-variant:small-caps;">T. Hmidi, S. Keraani and F. Rousset</span>, *Global well-posedness for Euler-Boussinesq system with critical dissipation*, arXiv:0903.3747v1 \[math.AP\].
<span style="font-variant:small-caps;">T. Hmidi, S. Keraani, F. Rousset</span>, *Global well-posedness for Navier-Stokes-Boussinesq system with critical dissipation*, arXiv:0908.0894v1 \[math.AP\].
<span style="font-variant:small-caps;">T. Hmidi, M. Zerguine</span>, [*On the global well-posedness of the Euler-Boussinesq system with fractional dissipation*]{}, Preprint. T. Y. Hou, C. Li, [*Global well-posedness of the viscous Boussinesq equations*]{}. Discrete and Continuous Dynamical Systems, ${\bf 12,}$ 1 (2005) 1-12. <span style="font-variant:small-caps;">A. Kiselev, F. Nasarov, A. Volberg</span>, *Global well-posedness for the critical $2D$ dissipative quasi-geostrophic equation*. Invent. Math. **167**, no. 3, 445-453, (2007). D. Kateb, [*On the boundedness of the mapping$f\to|f|^\mu, \mu>1$ on Besov spaces*]{}, Math. Nachr. 248-249, 110-128 (2003). G. Letac, Q. I. Rahman, A factorisation of the Askey’s characteristic function $(1-\|t\|_{2n+1})_+^{n+1}$, Ann. Inst. H. Poincaré Probab. Statist., 22 (1986), 169-174.
V. A. Liskevich, M.A. Perelmuter, Yu. A. Semenov, *Form-bounded perturbations of generators of sub-Markovian semigroups.* Acta Appl. Math. 44 (1996), no. 3, 353–377.
V. A. Liskevich, Yu. A. Semenov, [ *Some problems on Markov semigroups.*]{} Schrödinger operators, Markov semigroups, wavelet analysis, operator algebras, 163–217, Math. Top., 11, Akademie Verlag, Berlin, 1996. <span style="font-variant:small-caps;">C. Miao, L. Xue</span>, [*On the global well-posedness of a class of Boussinesq-Navier-Stokes suystems*]{}, arXiv:0910.0311v1 \[math.AP\].
F. Planchon, [*Sur une inégalité de type Poincaré*]{}, Comptes-Rendus de l’AcadŽmie des Sciences de Paris, [330]{} (2000), no. 1, 21-23.
G. Polyá, [*Remarks on Characteristic Functions,*]{} Proceedings of the Berkeley Symposium on Mathematical Statistics and Probability, University of California Press, 1949, 115-123 W. Sickel, [*Boundedness properties of the mapping $f\to|f|^\mu, 0\mu<1$ in the framework of Besov spaces,*]{} unpublished manuscript. T. Tao, [*Global regularity for a logarithmically supercritical hyperdissipative Navier-Stokes equation*]{}, arXiv:0906.3070v4. <span style="font-variant:small-caps;">H. Triebel</span>, *Theory of function spaces*. (Leipzig 1983). R. M. Trigub, [*A criterion for a characteristic function and a Polyá type criterion for radial functions of several variables,*]{} Theory Probab. Appl., 34 (1989), 738-742 <span style="font-variant:small-caps;">M. Vishik</span>, *Hydrodynamics in Besov Spaces*. Arch. Rational Mech. Anal 145, p. 197-214, (1998).
| {
"pile_set_name": "ArXiv"
} |
---
author:
- Antonio Ortiz
- Gyula Bencédi
- Héctor Bello
- Satyajit Jena
bibliography:
- 'references.bib'
title: 'Jet effects in high-multiplicity pp events'
---
Introduction
============
The study of the high-multiplicity pp events has become important because we need to understand the origin of the fluid-like features which have been found in such small systems [@Ortiz:2015iha; @Adam:2016dau; @Armesto:2015ioy; @Bautista:2015kwa].
In this work we concentrate on the radial flow signatures, which not only hydrodynamical models can explain. Namely, the effect has been also found in [@Sjostrand:2014zea] and it is attributed to multi-parton interactions (MPI) and color reconnection (CR) via boosted color strings [@Ortiz:2013yxa]. For high-multiplicity events, the blast-wave parametrization, a hydro inspired model, has been found to fit very well the transverse momentum ([$p_{\rm{T}}$]{}) spectra of different particle species [@Cuautle:2015kra]. Although, the quality of the fits become worse for low-multiplicity events, we see that the parameter related to the average transverse expansion velocity ($\langle\beta_{\rm T}\rangle$) increases with increasing multiplicity. This effect is qualitatively similar to what has been seen at the LHC [@Abelev:2013haa].
In , color reconnection was originally introduced in order to explain the rise of the average [$p_{\rm{T}}$]{}with the event multiplicity. In short, the model allows the interaction among the partons which originate from MPI and initial-/final-state radiation. There are different implementations, e.g., the default MPI-based model of 8.212 introduces a probability which is the largest for a low-[$p_{\rm{T}}$]{}system to be reconnected with one of a harder [$p_{\rm{T}}$]{}scale. And the interaction between two systems of high-[$p_{\rm{T}}$]{}scales is not allowed. Such a soft-hard interaction also suggests that jets may play a role in the observed radial flow-like patterns as highlighted in [@Cuautle:2015kra; @Ortiz:2015cma].
In this work, the role of jets in high-multiplicity pp collisions is investigated using 8.212. The inclusive [$p_{\rm{T}}$]{}spectra of identified particles are studied for events with and without jets, where the jets are reconstructed using the anti-$k_{\rm T}$ algorithm implemented in FastJet [@Cacciari:2011ma].
Results
=======
Proton-proton collisions at $\sqrt{s}=7$TeV were simulated with 8.212 using the tune Monash 2013 [@Skands:2014pea]. Events were classified according with their event multiplicity ($N_{\rm ch}$) and leading jet [$p_{\rm{T}}$]{}($p_{\rm T}^{\rm jet}$). All the observables were calculated counting particles within $|\eta|$$<$1. For the jet finder only detectable particles (including charged and neutral particles) are considered within cone radius of 0.4, while for the [$p_{\rm{T}}$]{}spectra and event multiplicity only charged particles are taken into account.
To investigate on the radial flow-like effects in jets we first study the proton-to-pion ratio in low-multiplicity events and as a function of $p_{\rm T}^{\rm jet}$ (see Fig. \[fig:1\]). It is worth noticing that events without jets[^1] dominate for momenta below 2GeV/$c$, while at larger momenta, jets start playing a more important role. In addition, a bump at intermediate [$p_{\rm{T}}$]{}is observed in all the event classes. The [$p_{\rm{T}}$]{}, where the peak emerges, increases with increasing $p_{\rm T}^{\rm jet}$. This structure resembles one observed in the different colliding systems at the LHC [@Adam:2015kca; @Ortiz:2015iha] and which sometimes is referred as a “flow peak” [@Ortiz:2013yxa]. This effect is the same in events generated with and without color reconnection.
![(Color online). Leading jet [$p_{\rm{T}}$]{}dependence of the proton-to-pion ratio as a function of transverse momentum for low-multiplicity pp collisions at $\sqrt{s}=7$TeV. The results for the different $p_{\rm T}^{\rm jet}$ intervals (markers) are compared with the inclusive case (solid line). Events with (left) and without (right) color reconnection are shown.[]{data-label="fig:1"}](./PicturesCR/PtoPiVsJetPtMultBin0 "fig:"){width="45.00000%"} ![(Color online). Leading jet [$p_{\rm{T}}$]{}dependence of the proton-to-pion ratio as a function of transverse momentum for low-multiplicity pp collisions at $\sqrt{s}=7$TeV. The results for the different $p_{\rm T}^{\rm jet}$ intervals (markers) are compared with the inclusive case (solid line). Events with (left) and without (right) color reconnection are shown.[]{data-label="fig:1"}](./PicturesnoCR/PtoPiVsJetPtMultBin0 "fig:"){width="45.00000%"}
![(Color online). Leading jet [$p_{\rm{T}}$]{}dependence of the transverse momentum spectra for low (top) and high (bottom) multiplicity pp collisions at $\sqrt{s}=7$TeV. The blast-wave parametrization is shown with solid lines.[]{data-label="fig:2"}](./PicturesCR/Jetbin0/hspectraNchBin0 "fig:"){width="32.00000%"} ![(Color online). Leading jet [$p_{\rm{T}}$]{}dependence of the transverse momentum spectra for low (top) and high (bottom) multiplicity pp collisions at $\sqrt{s}=7$TeV. The blast-wave parametrization is shown with solid lines.[]{data-label="fig:2"}](./PicturesCR/Jetbin2/hspectraNchBin0 "fig:"){width="32.00000%"} ![(Color online). Leading jet [$p_{\rm{T}}$]{}dependence of the transverse momentum spectra for low (top) and high (bottom) multiplicity pp collisions at $\sqrt{s}=7$TeV. The blast-wave parametrization is shown with solid lines.[]{data-label="fig:2"}](./PicturesCR/Jetbin4/hspectraNchBin0 "fig:"){width="32.00000%"} ![(Color online). Leading jet [$p_{\rm{T}}$]{}dependence of the transverse momentum spectra for low (top) and high (bottom) multiplicity pp collisions at $\sqrt{s}=7$TeV. The blast-wave parametrization is shown with solid lines.[]{data-label="fig:2"}](./PicturesCR/Jetbin0/hspectraNchBin5 "fig:"){width="32.00000%"} ![(Color online). Leading jet [$p_{\rm{T}}$]{}dependence of the transverse momentum spectra for low (top) and high (bottom) multiplicity pp collisions at $\sqrt{s}=7$TeV. The blast-wave parametrization is shown with solid lines.[]{data-label="fig:2"}](./PicturesCR/Jetbin2/hspectraNchBin5 "fig:"){width="32.00000%"} ![(Color online). Leading jet [$p_{\rm{T}}$]{}dependence of the transverse momentum spectra for low (top) and high (bottom) multiplicity pp collisions at $\sqrt{s}=7$TeV. The blast-wave parametrization is shown with solid lines.[]{data-label="fig:2"}](./PicturesCR/Jetbin4/hspectraNchBin5 "fig:"){width="32.00000%"}
The blast-wave analysis of the [$p_{\rm{T}}$]{}spectra has been performed using the same particle species and [$p_{\rm{T}}$]{}intervals described in [@Cuautle:2015kra]. Figure \[fig:2\] shows that the hydro model can describe the [$p_{\rm{T}}$]{}spectra when jets with momentum above 5GeV/$c$ are part of the event. Actually, a $\langle \beta_{\rm T}\rangle$ of $\approx$0.5 can be achieved when the jet [$p_{\rm{T}}$]{}is larger than 20GeV/$c$. Contrarily, the model does not describe the spectra in events without jets. This result is consistent with the spherocity analysis reported in [@Cuautle:2015kra], where it was argued that the fast parent parton being a boosted system can mimic radial flow too. The same analysis was also implemented for high-multiplicity events, in that case, thanks to color reconnection, the quality of the fit improves in events without jets, however a small $\langle \beta_{\rm T} \rangle$ ($\approx$0.37) is obtained and it increases up to $\approx$0.51 when a high-[$p_{\rm{T}}$]{}jet is identified in the event. Actually, when a high-[$p_{\rm{T}}$]{}jet was required, a very weak multiplicity dependence of $\langle \beta_{\rm T} \rangle$ is observed.
Summary
=======
In summary, we have studied the role of jets in the radial flow-like features of . We have found that even in low-multiplicity events the blast-wave model is able to describe the [$p_{\rm{T}}$]{}spectra of different particle species only when jets are part of the event. At high-multiplicity, $\langle \beta_{\rm T} \rangle$ can be very small in events without jets ($\approx$0.37). The interaction of jets with the soft component is therefore important to produce the observed effects in . This seems to be a promising tool which could be exploited by the experiments in order to understand better the LHC data.
Acknowledgements {#acknowledgements .unnumbered}
================
The authors acknowledge the useful discussions with Guy Paić, Eleazar Cuautle, Peter Christiansen and Gergely Barnaf[ö]{}ldi. Support for this work has been received from CONACYT under the grant No. 260440; from DGAPA-UNAM under PAPIIT grants IA102515, IN105113, IN107911 and IN108414; and OTKA under the grant NK106119. The EPLANET program supported the mobility from Mexico to Europe and vis.
[^1]: Events without jets are those where the jet finder can not reconstruct one with $p_{\rm T}^{\rm jet}$$>$5GeV/$c$.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Spatially resolved photocurrent measurements on carbon nanotube field-effect transistors (CNFETs) operated in various transport regimes are reported. It is demonstrated that the photocurrents measured at different biasing conditions provide access to the electronic band structure profile of the nanotube channel. A comparison of the profiles with the device switched into *n*- or *p*-type states clearly evidences the impact of chemical doping from the ambient. Moreover, we show that scanning photocurrent microscopy constitutes an effective and facile technique for the quantitative determination of the Schottky barrier height in such devices.'
address:
- 'Max-Planck-Institute for Solid State Research, Heisenbergstrasse 1, 70569 Stuttgart, Germany'
- 'Max-Planck-Institute for Solid State Research, Heisenbergstrasse 1, 70569 Stuttgart, Germany'
- 'Max-Planck-Institute for Solid State Research, Heisenbergstrasse 1, 70569 Stuttgart, Germany'
- 'Max-Planck-Institute for Solid State Research, Heisenbergstrasse 1, 70569 Stuttgart, Germany'
- 'University of Siegen, Adolf-Reichwein-Strasse 2, 57068 Siegen, Germany'
- 'University of Siegen, Adolf-Reichwein-Strasse 2, 57068 Siegen, Germany'
- 'Max-Planck-Institute for Solid State Research, Heisenbergstrasse 1, 70569 Stuttgart, Germany'
- 'Max-Planck-Institute for Solid State Research, Heisenbergstrasse 1, 70569 Stuttgart, Germany'
- 'Institut de Physique des Nanostructures, École Polytechnique Fédérale de Lausanne, CH-1015 Lausanne, Switzerland'
author:
- 'Eduardo J. H. Lee'
- Kannan Balasubramanian
- Jens Dorfmüller
- Ralf Vogelgesang
- Nan Fu
- Alf Mews
- Marko Burghard
- Klaus Kern
bibliography:
- 'apssamp.bib'
title: Electronic Band Structure Mapping of Nanotube Transistors by Scanning Photocurrent Microscopy
---
Carbon nanotube field-effect transistors (CNFETs) have been extensively studied for applications in electronics and optoelectronics [@cnfet; @freitagpc]. Despite significant progress in this research field, a range of important device features remain to be fully elucidated, such as the nature of the metal-CNT interface [@interface]. In this context, scanning probe microscopies can provide relevant local information from devices [@freitagsgm; @defect1]. A well-suited technique to explore the band structure profile within CNT devices is scanning photocurrent microscopy (SPCM), which so far revealed dominant photocurrent generation at the CNT-metal contacts [@bal1; @bal2]. Here we report the detailed SPCM characterization of CNFETs in different charge transport regimes, and proof that optical excitation of the nanotubes is the origin of the detected photocurrents. The SPCM data enable convenient access to the height of the Schottky barriers at the contacts. Furthermore, the obtained band profiles are in excellent agreement with current models of CNTFET device operation.
SWCNTs were synthesized via chemical vapor deposition (CVD), following the procedure reported by Choi et al [@choi]. AuPd electrical contacts (15nm thickness) were defined on top the tubes using standard e-beam lithography. Photocurrent and reflection images were simultaneously acquired through a confocal optical microscope coupled to an electrical measurement set-up [@bal1]. Photo-illumination was carried out by a HeNe laser (E $\sim$ 1.96 eV, spot size $\sim$ 0.5 $\mu$m) with a focused beam intensity of $\sim$ 100 kW/cm$^2$. The photocurrent measurements were performed under ambient conditions, on a total of 11 different devices.
Figure 1(a) shows the transfer characteristic of a CNFET displaying slightly asymmetric ambipolar behavior. The photocurrent images of the device taken at zero bias in both the *p*- and *n*-type regimes, presented in Fig. 1(b), exhibit enhanced photocurrent responses close to the metal contacts. By contrast, no detectable photoresponse is observed along the entire CNFET device in the OFF state. This difference can be explained by Schottky barriers at the contacts prevalent in the ON states, whereas the flat bands in the OFF-state do not support acceleration of the photo-generated carriers \[see insets of Fig. 1(b)\]. The gate-induced modulation of the SPCM images solidifies the previous conclusion that built-in electric fields are responsible for the photocurrent generation [@bal1; @bal2]. Further support for such interpretation stems from the sign of the photocurrent lobes in the various regimes, as exemplified by the line profiles in Fig. 1(c). Based upon the convention that electrons flowing out of the source contact are measured as a positive current, the signs of the photocurrents at the contacts in the ON states of the device are consistent with the direction of the built-in fields there.
![\[fig:epsart\] (a) Conductance *vs*. gate voltage plot of a CNFET (b) Zero-bias SPCM images (white corresponds to positive current). The dashed lines indicate the edges of the source (S) and drain (D) electrical contacts. The insets depict the respective band diagrams. (c) Photocurrent line profiles taken along the nanotube.](fig1)
Although the photoresponses in the *p*- and *n*-type regimes are qualitatively similar, they exhibit some subtle differences. While in the *p*-type regime, the photocurrent peaks appear almost exactly at the tube/contact interface, in the *n*-type regime they are offset by $\sim$ 0.25 $\mu$m towards the center of the CNT channel. Such a behavior was observed in all of the investigated samples. It demonstrates that the band structure of the device in the *p*-type regime is not just a mirror image of that in the *n*-type regime. To elucidate this further, a series of zero-bias photocurrent images were acquired while sweeping the gate voltage from the *n*-type regime to the *p*-type regime, as depicted in Fig. 2(a). It is apparent that while tuning the device from the *n*-type regime to the OFF state, the photocurrent lobes shift gradually towards the center of the channel, become broader and considerably decrease in intensity. The very low intensity (few tens of pA) features along the channel in the OFF state, observed for some of the samples, may be ascribed to weak local electric fields similar to those observed in metallic CNTs [@bal2]. Such fields could arise due to, e.g. defect sites [@defect1]. When the device is further tuned to the *p*-type regime, the lobes appear suddenly at the contacts without showing any gradual movements. This observation is independent of the speed and direction of the gate voltage sweep.
The trend observed upon switching from the *n*-type regime to the OFF state is indicative of lowering and widening of the Schottky barriers. These changes are consistent with theoretical simulations, according to which the Schottky barrier width is closely related to the magnitude of the local electric fields at the contacts [@heinzeprl]. The sudden appearance of the *p*-type lobes in comparison to the gradual movement and broadening of the *n*-type lobes evidences that in the former regime the gate voltage has a much weaker effect on the width of the Schottky barriers. The overall behavior of the CNFET is in close accord with the doped-CNFET model recently proposed by Chen and Fuhrer [@chendoping]. It involves chemical *p*-type doping upon nanotube exposure to ambient conditions, thereby introducing a strong band-bending confined to a very small width $W_{doped}$ for *p*-type operation \[Fig. 2(b)\] [@tersoffprl]. By comparison, in the *n*-type regime charge depletion occurs over an effective barrier width $W_{eff}$ of the order of magnitude of the gate oxide thickness *t*, which can be effectively modulated by the gate voltage. This situation is different from intrinsic-CNFETs for which in both regimes the bands are bent over a depletion length *W* $\sim$ *t*. Within the doping model, the occurrence of the *p*-type photocurrent lobes at the contacts and the $\sim$ 0.25 $\mu$m shift of the *n*-type lobes \[Figs. 1(c) and 2(a)\] can be related to the presence of Schottky barriers with respective widths of $W_{doped}$ and $W_{eff}$. Moreover, the broadening of the lobes upon transition from the *n*-type regime to the OFF-state is attributable to an increase of $W_{eff}$, whereas in the *p*-type regime, the absence of such a behavior indicates that the Schottky barrier width is confined to $W_{doped}$. Therefore, it is evidenced that the effect of exposure to ambient conditions is not restricted to changes in the metal work function.
Towards the task of determining the band structure profile of CNFETs with the aid of SPCM it has to be ensured that the observed photocurrent signals indeed originate from photoexcitation of the nanotube. Previous CNFET photoconductivity studies have reported photovoltage generation at the Si/SiO$_{2}$ interface upon excitation with energies above the Si band gap [@freitagpc; @marcusjap]. To rule out such contributions in the SPCM response, measurements were performed in the *p*-type regime with laser excitation at $\lambda_{exc}$ = 1.6 $\mu$m (E $\sim$ 0.78 eV, spot size $\sim$ 2 $\mu$m). The obtained SPCM images displayed the same principal features as in the experiments with $\lambda_{exc}$ = 633 nm, proving that the photoresponse predominantly arises from optical excitation of the nanotubes.
 Zero bias photocurrent response upon transition from the *n*- to *p*-type. Color scales were normalized such that white (black) corresponds to the maximum (minimum) detected current. (b) Band diagrams depicting the intrinsic- and doped-CNFET models. (c) Gate voltage dependence of the drain current (dotted curve) and of the intensity of the photocurrent lobes (solid lines). The inset shows the activation barrier energy at zero bias (dots) and the logarithm of the resistance (red line) as a function of gate voltage. $V_{th}$ and $V_{fb}$ correspond to the threshold and flat-band voltages.](fig2)
In order to determine the Schottky barrier height for holes $\Phi_{Bp}$, the gate dependence of the drain current $I_{d}$ and the photocurrent at the contacts $I_{ph}$ was evaluated, following a similar procedure as reported for Si nanowires [@sinw]. The two $I_{ph}$ curves in Fig. 2(c) cross each other at $V_{G}$ = $V_{fb}$ = 6.9 V, where $I_{ph}$ is approximately zero. At this gate voltage, the device is in the OFF state with flat bands at the contacts. Moreover, when $V_{G}$ equals the threshold voltage, the drain current reaches zero, corresponding to the situation where the valence band is approximately aligned at the Fermi level [@appenzeller]. The Schottky barrier height can be calculated as $\Phi_{Bp} =
\alpha_{p}\cdot(V_{fb} - V_{th})$, where $\alpha_{p}$ is the gate coupling parameter which can be estimated from the subthreshold slope obtainable from the drain current curve in the *p*-type regime [@alpha]. For the sample in Fig. 2, the subthreshold slope is $\sim$300 mV/decade, yielding an $\alpha_{p}$ value close to 0.2, from which $\Phi_{Bp}$ is determined to be $\sim$180 meV. Similar values have been obtained for other samples. In order to independently determine $\Phi_{Bp}$, we have performed temperature dependent $I-V$ measurements and analyzed the data within the framework of thermionic emission theory, as described in previous works [@martelprl; @chendoping]. The measured activation barrier energies at zero bias are shown as a function of $V_{G}$ in the inset of Fig. 2(c). In the ON states, the activation energy significantly underestimates the barrier height, since charge injection occurs via thermal-assisted tunneling. On the other hand, in the OFF state, thermionic emission is the dominant process, such that the activation energy approaches $\Phi_{Bp}$. Due to sensitivity limitations, a lower limit of $\sim$170 meV was estimated for the barrier height, which is in close agreement to the value obtained by SPCM measurements.
Having characterized the devices at zero bias, we now address the effect of finite drain-source bias on the band structure profile of CNFETs. Figure 3(a) displays a series of SPCM images taken at various bias voltages with the device operated in the *n*-type regime. Although a somewhat smaller ratio of photocurrent to the dark current is found in the *p*-type regime, the following discussion applies equally well in this case. The images disclose that an applied voltage leads to the enhancement of one of the photocurrent lobes, depending on the sign of the bias. For sufficiently large bias, a single lobe is observed. To better visualize the CNFET band profile, we integrate the photocurrent signal along the length of the tube. This approach is justified by assuming that the photocurrent is directly proportional to the built-in electric field [@footnote]. In Fig. 3(b), the resulting qualitative electrostatic potential profiles are depicted for different bias values, with the source taken as ground and the drain being lifted (lowered) upon application of a negative (positive) potential. It follows from the line profiles that the center of the channels remains approximately flat irrespective of the applied bias. Moreover, application of a more positive (negative) bias results in a voltage drop predominantly occurring at the source (drain). The band profile changes with increasing bias closely following the behavior expected for the *n*-type unipolar operation mode of a CNFET, confirming that the charge transport through the CNT is governed by the contacts. Furthermore, the photocurrent at the lobes $I_{ph}$ is plotted as a function of $V_{ds}$ in Fig. 3(c). The $I_{ph}$-$V_{ds}$ curves are characteristic of reverse-biased Schottky diodes under photoexcitation [@nearfield].
 Effect of bias voltage on the photoresponse in the *n*-type ON state. (b) CNFET electrostatic potential profiles taken along the tube. (c) Bias dependence of the photocurrent response at the contacts.](fig3)
Finally, the effect of gate voltage on the photoresponse of biased CNFETs was investigated. To this end, a series of images similar to those in Fig. 2 were recorded at an applied bias of $V_{ds}$ = +0.7 V, which are collected in Fig. 4(a). At the starting point of $V_{G}$ = 10 V in the *n*-type regime, a single lobe at the source contact is observed, in close correspondence to the situation in Fig. 3(a). Upon decreasing $V_{G}$, the intensity of this lobe gradually decreases and clear photocurrent signals emerge along the tube between the contacts. Around $V_{G}$ = 3.7 V, *p*-type operation sets in and a single lobe emerges at the drain contact, analogous to the observation in Fig. 3(a). The set of SPCM images provides a complete picture of the photoconductivity in CNFETs, in close agreement with a previously proposed model [@bal03]. A major conclusion is that true photoconductivity from an individual CNT is only observable in the OFF state of the device. In the ON states, the local photoresponse is similar to that obtained from a reverse-biased Schottky diode.
 SPCM images depicting the transition from the *n*- to the *p*-type regimes taken at $V_{ds}$ = 0.7V. (b) Electrostatic potential profiles obtained from the corresponding photocurrent images. (c) Band diagrams describing the different transport regimes of a biased CNFET.](fig4)
It can be recognized from the corresponding electrostatic potential profiles in Fig. 4(b) that in the *p*- (*n*-) type regime, the voltage drops are concentrated at the drain (source) contact, as opposed to the OFF-state, in which the potential drop is distributed along the entire channel. Closer inspection of the profile at $V_{G}$ = 4.3 V suggests that the voltage drop occurring in the vicinity of the contacts is symmetric and stronger than in the middle of the CNT. In total, the band profiles agree very well with the widely accepted model for the operating mechanism of CNFETs, wherein the band structure at the contacts is modulated by the fields induced by gate and drain-source voltages [@appenzeller]. The band profiles are also consistent with the model used to interpret electroluminescence experiments [@freitagprl]. Specifically, in the OFF state, the voltage drops at the contacts are symmetric, thus leading to balanced injection of electrons and holes, followed by light emission from the center of the devices. Upon moving away from the OFF state to the *p*- (*n*-) type regime, the voltage is found to drop mostly at the drain (source) contact, whereby hole (electron) injection is favored. Although the drain bias used in the present experiment is not ideally suited for observing electroluminescence (where $V_{ds}$ is required to be twice as high as $V_{G}$), the extracted band profiles describe the underlying scenario impressively well.
In summary, SPCM has been demonstrated to be a versatile tool for CNFET characterization. The obtained photocurrent images confirm the relevance of chemical doping in CNFETs. Moreover, it has been shown that SPCM allows facile estimation of the Schottky barrier height, for whose determination reliable and straightforward methods are still lacking. The gained values of 150-200 meV are in good agreement with theoretical predictions. Furthermore, SPCM experiments have revealed that photoconductivity can be observed in the OFF state, whereas in the ON regimes the local photoresponse is dominated by the contacts, which behave as photoexcited reverse-biased Schottky diodes.
Ph. Avouris, MRS Bull. **29**, 403 (2004).
M. Freitag, Y. Martin, J. A. Misewich, R. Martel and Ph. Avouris, Nano Lett. **3**, 1067 (2003).
Z. H. Chen, J. Appenzeller, J. Knoch, YM Lin, Ph. Avouris, Nano Lett. **5**, 1497 (2005).
M. Freitag, M. Radosavljevic, Y. X. Zhou, A. T. Johnson, W. F. Smith, Appl. Phys. Lett. **79**, 3326 (2001).
M. Bockrath, W. Liang, D. Bozovic, J. H. Hafner, C. M. Lieber, M. Tinkham and H. Park, Science **291**, 283 (2001).
K. Balasubramanian, Y. Fan, M. Burghard, K. Kern, M. Friedrich, U. Wannek and A. Mews, Appl. Phys. Lett. **84**, 2400 (2004).
K. Balasubramanian, M. Burghard, K. Kern, M. Scolari, A. Mews, Nano Lett. **5**, 507 (2005).
H. C. Choi, W. Kim, D. Wang and H. Dai, J. Phys. Chem. B **106**, 12361 (2002).
S. Heinze, J. Tersoff, R. Martel, V. Derycke, J. Appenzeller and Ph. Avouris, Phys. Rev. Lett. **89**, 106801 (2002).
Y. F. Chen and M. S. Fuhrer, Nano Lett. **6** 2158 (2006).
F. Leonard and J. Tersoff, Phys. Rev. Lett **89**, 179902 (2002).
M. S. Marcus, J. M. Simmons, O. M. Castellini, R. J. Hamers and M. A. Eriksson, J. Appl. Phys. **100**, 084306 (2006).
Y. Ahn, J. Dunning and J. Park, Nano Lett. **5**, 1367 (2005).
J. Appenzeller, J. Knoch, V. Derycke, R. Martel, S. Wind and Ph. Avouris, Phys. Rev. Lett. **89**, 126801 (2002).
S. Rosenblatt, Y. Yaish, J. Park, J. Gore, V. Sazanova and P. L. McEuen, Nano Lett. **2**, 89 (2002).
R. Martel, V. Derycke, C. Lavoie, J. Appenzeller, K. K. Chan, J. Tersoff and Ph. Avouris, Phys. Rev. Lett. **87**, 256805 (2001).
The electrostatic potential profile is valid only in the region between the contacts. Moreover, it is broadened by convolution with the Airy pattern of the incident laser beam.
Y. Gu, E. S. Kwak, J. L. Lensch, J. E. Allen, T. W. Odom and L. J. Lauhon, Appl. Phys. Lett. **87**, 043111 (2005).
K. Balasubramanian and M. Burghard, Semicond. Sci. Technol. **21**, S22 (2006).
M. Freitag, J. Chen, J. Tersoff, J. C. Tsang, Q. Fu, J. Liu and Ph. Avouris, Phys. Rev. Lett. **93**, 076803 (2004).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Child-Langmuir law expresses the IV relation between two parallel plates. It has been found that in reality the ordinary Child Langmuir Law does not hold true. It shows some geometric dependence on the system.In this report capacitors of various thickness and septation has been studied both in macro and nano scale to propose a geometrical correction to the law. More over from the studies, a new type of radiation detector was prepared which can detect radiation in small volumes as compared with normal radiation detectors.'
---
\
\
**Saptarshi Ghosh$^{1\dagger}$ and H.C.Verma$^{2\ast}$\
*${^1}$ Discipline of Physics, Indian Institute of Technology Indore, Indore 453552, India\
*${^2}$ Department of Physics, Indian Institute of Technology, Kanpur 208016, India\
${^{\dagger}}$ Email: [email protected] ${^{\ast}}$ Email: [email protected]****
[**keywords:**]{} Child-Langmuir law, Nano-Capacitors
Introduction
============
In modern days physicists are trying to study local effects inside different types of materials. These become effective in Nano meter and Micro meter range. So the importance of studying the Nano structures and designing them for different purposes is of immense importance.With increasing use of Nano/Micro scale structures, in scientific researches and in industrial applications, the importance of the field of Nano Science has increased.The increasing use of Micro structures in present day circuitry of different appliances has reduced their size dramatically and has given them more compact look.For this reason, the interest now lies in further reducing the size of the materials and that makes Nano Technology and Nano Science to be one of the forefront of active research.\
Nuclear radiations $\alpha, \beta, \gamma $ particles have been used for a variety of material modifications. Detecting their presence is therefore quite important task. The present day detectors are capable of detecting such radiations in a large volume. With circuitry shrinking to nano and micro scale , it will be very useful to have particle radiation detectors with smaller dimensions. The present work is a step in this direction.\
The report is divided in five sections. Firstly,The project framework and the main motivation behind the studies has been discussed.Then the studies on Macro and Nano scale is presented along with the fabrication techniques and experiment details. Later the Nano Radiation Detector (NRD) model is presented along with the experimental data.At the end Further improvement upon the present NRD model has been suggested.
Project Framework
=================
The main objective of the project is the nano radiation detector.Various studies have been made for detector designs. In last semester, due to technical problems macro capacitors and nano capacitors could not be studied.In section 3 of the report, these objectives were fulfilled. The proposed detector is based on passes of current through the medium in a small size capacitor and hence IV characteristic of such a capacitor has to be studied. Though Child-Langmuir Law gives some insight but the theoretical factors are to be studied more systematically. We have done some work in the direction and the results are presented in this report.\
The effects of $\alpha$ Particles on IV characteristic is also reported and this shows the feasibility of out setup to be developed as $\alpha$ particle detectors.
Geometry effect on I-V characteristic of a gap between parallel plate capacitor
===============================================================================
####
\
##### 3.1.1 Fabrication
\
The system used to perform the study mainly consists of a vacuum chamber and a Turbo Molecular Pump (TMP) assembly. For studying the geometric effect both the chamber and the assembly were designed from scratch.\
The chamber was cylindrical in shape. It had two electrodes for measurement in the vacuum.The electrodes were insulated from the body of the chamber by a thick sheet of perspex. Also rubber O-rings were used to prevent vacuum leaking. It also provided freedom of the restricted movement of the electrodes (as shown in Fig \[\[fig:electrode\]\]).Proper grooving was made on the electrodes (as shown in the fig \[\[fig:chamber\]\]) as per our necessity. The diagram of the whole chamber is shown in fig \[\[fig:chamber\]\].Several square sized copper plates (dimensions : $2\times2cm^2,3\times3cm^2,4\times4cm^2,5\times5cm^2$) were made from the workshop of physics department.These plates and the electrodes was designed in such a way that the inter plate distance can be varied in vacuum inside the chamber.\
![Vacuum Chamber With parallel plate Capacitor[]{data-label="fig:chamber"}](chamber.eps){width="40.00000%"}
Next the vacuum pump assembly was designed. In light of the failure of diffusion pump in previous semester, a turbo molecular pump and a rotary pump (as backing pump) was used in the system. The number of junction points were reduced to minimize the chance of leaking. The chamber was connected to TMP via a plus type of structure shown in the fig\[\[fig:system\]\]. All the sensors were mounted on one hand of the plus connector. Metal O-ring was used in between the TMP and the connector. The whole system was connected to a Kethley-6430 system and a computer to measure the data electronically without disturbing the system. The whole setup was shown in Fig \[\[fig:system\]\]\
![TMP Vacuum Assembly[]{data-label="fig:system"}](system.eps){width="20.00000%"}
![Vacuum Chamber With parallel plate Capacitor[]{data-label="fig:electrode"}](electrode.eps){width="20.00000%"}
\
##### 3.1.2 Experiment
\
The rotary could evacuate the chamber up to 1.1 Torr. Then the TMP was used to take the vacuum up to order of $10^{-4}$ Torr. The whole experiment was done in the vacuum of this magnitude.\
Firstly the capacitor plates were screwed with the electrodes. Proper connection was made between the computer system and the whole vacuum assembly so that whole system can be operated digitally. After evacuating the chamber to desired value, the kethley setup was switched on to start the measurement. The IV measurement was done with three different (0.5 cm, 1.0 cm, 1.5 cm ) separations between the capacitor plates.\
The whole process of measurement was repeated for each set of copper plates. Two sets of data were taken to check the inconsistency of the measurement. The analysis of the experimental data is shown in the next section.
##### 3.1.3 Result & Discussion
\
The graphs obtained in the measurement are shown below.Fig \[\[fig:2x2\]\],Fig \[\[fig:3x3\]\], Fig \[\[fig:4x4\]\],Fig \[\[fig:5x5\]\].
![IV characteristic for 2x2 cm copper plates[]{data-label="fig:2x2"}](2x2.eps){width="30.00000%"}
\
![IV characteristic for 3x3 cm copper plates[]{data-label="fig:3x3"}](3x3.eps){width="30.00000%"}
\
![IV characteristic for 4x4 cm copper plates[]{data-label="fig:4x4"}](4x4.eps){width="30.00000%"}
\
![IV characteristic for 5x5 cm copper plates[]{data-label="fig:5x5"}](5x5.eps){width="30.00000%"}
\
These IV characteristic curves have shown some interesting results. .These interesting results can be analyzed well when we keep separation between the plates fixed.Following graphs are plotted for different fixed values of separations.Fig\[\[fig:0.5\]\],fig\[\[fig:1\]\],fig\[\[fig:1.5\]\]
![IV characteristic for copper plates\
(separation 0.5cm)[]{data-label="fig:0.5"}](pfive.eps){width="30.00000%"}
\
![IV characteristic for copper plates\
(separation 1.0cm)[]{data-label="fig:1"}](one.eps){width="30.00000%"}
\
![IV characteristic for copper plates\
(separation 1.5cm)[]{data-label="fig:1.5"}](onepfive.eps){width="30.00000%"}
\
From these graphs one can easily notice that as indicated in the standard Child-Langmuir law, that the current varies with three and half power of the voltage, is not entirely correct. Foe example in Fig \[8\] ,for the same separation ,The IV curves are different for different plate dimensions.One needs to introduce a geometric factor in the expression. Such correction have been proposed in literature. Defining a geometric factor as **area/$separations^2$**.The current is supposed to be proportional to this factor.Since all other plates are of square shape,We will be using **g=length of a plate/Separation between plates** as a geometric parameter. As it is observed for small sizes the geometric correction is to be re-looked. For example, if we look at graphs in Fig \[8\], we see that the current largest for $2\times2cm^2$ plates and smallest for $5\times5cm^2$ for the same separation of 0.5 $cm^2$. According to geometrical correction proposed in the literature (ref \[1\]), the current should increase by a factor of 6.25 for a fixed voltage. In contrary, we observe the current to decrease.\
With Sufficient data the geometric dependence can be obtained from these graphs. However with 4 sizes of plates and 3 separations taken one can only have 12 data points for the plotting I vs g graph as shown in fig \[\[fig:gdep\]\].Two points (g=2,g=4) coincide with each other.Only Ten data points are shown in the graph.
![geometric effect on Space charge limited current\
(at 50v)[]{data-label="fig:gdep"}](gdep.eps){width="30.00000%"}
\
It is very hard to obtain a power dependence from only twelve data points. However we consider geometric factor to be g and it varies to the power a.So overall we assume the current equation to be $$I=cg^{a}v^{b}$$ where b can be 3/2 according to Child-Langmuir law or it may not be.We want to keep this freedom in the equation. From this equation, $$a=\frac{g}{I}\frac{dI}{dg}$$ We have calculated slope at all the points except terminal ones and from there we get the value of a which varies roughly from -0.2 to -0.3 .The result is quite surprising.This geometric factor can be used to fit the data we obtained for the Nano capacitor measurement. However as mentioned in the introduction, if one can modify the standard Child-Langmuir law by introducing a geometric factor and if the same power holds for low dimensions them it can be immensely helpful for production of many low dimensional electronic equipments.\
####
\
##### 3.2.1 Fabrication
\
In this section, The process and the designing techniques will be discussed. The Nano Capacitors was made using shadow masking technique.\
Firstly,The mask of necessary dimensions were made at Meera lasers,a Chennai based laser drilling company. They made the mask as per our design shown in the fig \[\[fig:mask\]\]on tin base. Four masks were made of width $50\mu m$,$40\mu m$,$30\mu m$,$20\mu m$ . only one mask is shown in the figure \[\[fig:mask\]\].\
![Mask prepared on Tin base[]{data-label="fig:mask"}](mask.eps){width="30.00000%"}
\
Then the substrate was prepared for deposition. Si wafers were oxidized to produce $SiO_2$ substrate. First, the Si wafers were cleaned by RCA technique.Then it was being heated to $1000^0$C in pure oxygen environment for oxidation for almost 24 hours.This work had been done in the Semiconductor Lab at IIT Kanpur. The $SiO_2$ was chosen as substrate because of its good adhesion property with gold film.\
The capacitors were made by depositing gold on the $SiO_2$ substrate. Thermal Evaporator was used for the deposition. A detailed description of the thermal evaporation technique can be found in the previous semester report. The Nano wires of thickness 80 $\mu$m was prepared for the measurement.\
Now Due to unavailability of the FIB machine of IBC,IITK,Somewhat crude technique was used to create the ’Nano’ capacitor. A fine surgical blade was used for that purpose. A sharp cut was made by the surgical blade edge in the fine wire prepared by the shadow masking. So a small gap of the width similar to the thickness of the blade was created (as shown in the following fig\[\[fig:setup\]\] ). This gap was used as the capacitor for our measurement.
![The nano Capacitor[]{data-label="fig:setup"}](setup.eps){width="30.00000%"}
\
##### 3.2.2 Experimentation
\
A contact pin of the shape as shown in the fig \[14\]was used for making contact with the thin flim capacitor. This rare contact pin was provided by Prof. R.S. Anand of EE dept. of IIT kanpur. Using the pin, the capacitor was connected to a standard bread board. Standard bread board wires were used to make connection with the kethley meter.\
![contact pins used[]{data-label="fig:cpin"}](cpin.eps){width="3.00000%"}
\
The measurement was done on capacitors with widths 30$\mu$m and 40$\mu$m respectively. Due to technical difficulties measurement could not be done on other capacitors. Two sets of data were taken on each capacitor. The measurement was done in atmosphere.
##### 3.2.3 Result & Discussion
\
The results are shown in the following figures.Fig\[\[fig:30\]\],Fig\[\[fig:40\]\]
![IV characteristic for Nano Capacitor (width 30 $\mu$m)[]{data-label="fig:30"}](30.eps){width="30.00000%"}
\
![IV characteristic for Nano Capacitor (width 40 $\mu$m)[]{data-label="fig:40"}](40.eps){width="30.00000%"}
\
##### 3.2.4 Discussion
\
From the graphs it can be clearly seen that it does not follow child-Langmuir law. A similar graph was shown in the previous semester report. Although in that case the gap was much larger than the present case. The similarity of the data from previous semester also indicates that they may follow the same geometric power as of macro scale study.\
However main motive of this study is for the nano radiation detector. To distinguish the signal from the background noise we need to know the IV curve of the concerned detector without any alpha source. This data will provide the background band in the detection.Any signal which strength is stronger than the back ground can be considered as a Alpha particle hit which we will discuss below.
Nano Radiation Detector
=======================
The main idea behind the nano radiation detector is very simple. The IV characteristic between two capacitor plates are given by standard or modified Child-Langmuir law. Now this Current was generated by space charge limited current. Now if a alpha particle hits the gap in between the plates then ionization caused by the alpha particle will give rise to a lot more charge carriers. So it will result in a larger current which should be significantly distinguishable from the background signal. By proper electronics this signal can be counted and a local alpha particle distribution can be prepared which may be helpful in many scientific applications.
##### 4.1 Fabrication
\
The Same capacitors used previously was used here as the detectors.
##### 4.2 Experimentation
\
The experimental setup was also same as before. Only the Alpha particle sources was kept near the setup for radiation. Americium-241 (${}^{241}_{95}Am$ ) was used for the Alpha radiation source. All safety measures including radiation batch was taken during the experimentation.
##### 4.3 Result & Discussion
\
Two types of measurement were done in the experiment. Two sets of data were taken for current against voltage. Five sets of data were taken as current against time. Only one I-T graph is shown here because this is the only graph where a signal was detected.
![IV characteristic of nano radiation detector[]{data-label="fig:nrdiv1"}](nrdiv1.eps){width="30.00000%"}
\
![IV characteristic of nano radiation detector[]{data-label="fig:nrdiv2"}](nrdiv2.eps){width="30.00000%"}
\
![IT characteristic of nano radiation detector[]{data-label="fig:nrdit"}](nrdit.eps){width="30.00000%"}
\
##### 4.4 Discussion
\
Now Am-241 is a strong alpha particle source. So one may expect to get a continuous hits i.e. a large number of signals. But experimental data shows opposite to the expected result. The reason behind this is the non-standardization of the equipment. Firstly the alpha source was kept just besides of the detector. So the radiation is getting spread out over a lot of area. Again we are using a single capacitor as the detector. The dimension of the wire is in micron range.Moreover the cut made on the detector also in order of microns. So it can be very easily understood that the probability of hitting event of a alpha particle at exact gap point is very low.This is the reason behind the very low hit rate. How ever this problem can be solved using a grid of capacitors which is discussed in next section.
Future Agenda
=============
From the previous discussion and the experimental result it is clear that to continue with only one capacitor for the detector is not likely to be a useful exercise. So we propose a grid of capacitors as shown below in the fig \[\[fig:grid\]\]\
![Proposed grid nano radiation detector[]{data-label="fig:grid"}](grid.eps){width="20.00000%"}
\
This type of grid will be able to detect a large number of hits of alpha particles. Moreover to ignore the capacitive effects of the pads one can use the nano capacitor study to ignore the normal background signals. The accuracy can be further increased by using the same in some special Gas environment so that the dead time can be reduced. Since the dimension of the system is very small, the dead time will be much much smaller as compared to the standard detectors.Also as we have seen in the experimental data that the signal gets stronger at higher voltages which is expected.One can use the $Ga^{+}$ irradiation data from the my colleague Avishek Kr. Basu’s report to increase the voltage tolerance of nano wires of the grid.However the effect of the irradiation on the detector signal remained to be studied. We sincerely hope that some future project student will complete the work.
Acknowledgments
===============
I am thankful to my guide Prof H.C.Verma for his nice guidance and advises in my project.I will remain thankful to him for taking me as his student .I also express my gratitude to Prof. R.S. Anand for his help. I am also thankful to Dr. Nobin Banerji,Krishnasamy S, Sandipji for their help and support.I also thank my college avishek for his help in completing the project.Lastly i must not forget to acknowledge all my friends and professors at IIT Kanpur for their time to time valuable suggestions during discussion hours..
[99]{}
R. J. Umstattd and C. G. Carr,[*A simple physical derivation of Child-Langmuir space-charge-limited emission using vacuum capacitance*]{}, American Journal of Physics [**73**]{}, 160 (2005).
N.S Rajput,A.K Singh,H.C. Verma, [*Role of the substrate in electical transport characteristics of focused ion beam fabricated nano gap electrode*]{}, J.Appl.Phys. [**112**]{} 024310 (2012).
Takashi Nagase,Kenji Gamo,Tohru Kubota,Shinro Mashiko,“Direct fabrication of nano-gap electrodes by focused ion beam etching”, Thin Solid Films [**499(1-2)**]{} 279-284 (2006).
I.B.I Prenitzer,F.A. Stevie,L.A. Giannuzzi,“Introduction to Focussed ion beam instrumentation theory, techniques and practice”
Nan Yao,“Focused Ion beam systems Basics and Application”
S.S. Kapoor,V.S. Ramamurthy,“Nuclear Radiation Detectors”
Glenn F.Knoll,“Radiation Detection and Measurement”
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A de Bruijn cycle is a cyclic listing of length $\ca$, of a collection of $\ca$ combinatorial objects, so that each object appears exactly once as a set of consecutive elements in the cycle. In this paper, we show the power of de Bruijn’s original theorem, namely that the cycles bearing his name exist for $n$-letter words on a $k$-letter alphabet for all values of $k,n$, to prove that we can create de Bruijn cycles for the assignment of elements of $[n]=\{1,2,\ldots,n\}$ to the sets in any labeled subposet of the Boolean lattice; de Bruijn’s theorem corresponds to the case when the subposet in question consists of a single ground element. The landmark work of [@cdg] extended the agenda of finding de Bruijn cycles to possibly the next most natural set of combinatorial objects, namely $k$-subsets of $[n]$. In this area, important contributions have been those of [@h] and [@r]. Here we follow the direction of [@bg], who proved that, in a suitable encoding, de Bruijn cycles can be created for the subsets of $[n]$ of size in the interval $[s,t]; 0\le s<t\le n$. In this paper we generalize this result to exhibit de Bruijn cycles for words with weight between $s$ and $t$, where these parameters are suitably restricted.'
author:
- |
André Alexander Campbell & Anant P. Godbole\
Department of Mathematics and Statistics\
East Tennessee State University
- |
Bill Kay\
Department of Mathematics and Computer Science\
Emory University
title: Contributions to the Theory of de Bruijn Cycles
---
\[thm\][Claim]{} \[thm\][Definition]{} \[thm\][Lemma]{} \[thm\][Corollary]{} \[thm\][Remark]{} \[thm\][Proposition]{} \[thm\][Example]{} \[thm\][equation]{} \[section\] \[thm\][Question]{} \[1\][\#1]{}
Introduction
============
A de Bruijn cycle is a cyclic listing of length $\ca$, of a collection of $\ca$ combinatorial objects, so that each object appears exactly once as a set of consecutive elements in the cycle. For example, the cyclic list 11101000 encodes each of the eight binary three letter words so that each appears exactly once. Viewed differently, and using a different coding, we see that the same list encodes all the 8 subsets of $[3]=\{1,2,3\}$, with the convention that, e.g., the string 101 represents the subset $\{1,3\}$ – whose characteristic vector it happens to be. de Bruijn’s theorem states that this example is not an anomaly, and that de Bruijn cycles exist for the collection of $n$-letter words on the $k$-letter alphabet $\{0,1,\ldots,k-1\}$, and thus, using a different coding, for all the multisets of $[n]=\{1,2,\ldots,n\}$ where each element may appear at most $k-1$ times in a multiset. [*A de Bruijn cycle is most often called a Universal Cycle (or U-cycle) in general contexts, but we shall not do so in this paper.*]{}
Consider the poset $\cb=\cb_n$, the so-called Boolean Lattice of all subsets of $[n]$ ordered by inclusion. A [*subposet*]{} of $\cb$ is defined in this paper to be a collection of subsets of $\cb$ with the same Hasse diagram that is, moreover, consistent with the Hasse diagram of $\cb$. For example, the W-poset or a 4-chain are each consistent with $\cb_3$ but not with $\cb_2$. In Section 2 of this paper, we show the versatility of de Bruijn’s original theorem by showing that we can create de Bruijn cycles for the assignment of elements of $[n]=\{1,2,\ldots,n\}$ to the sets in [*any labeled*]{} subposet of the Boolean lattice; de Bruijn’s theorem corresponds to the case when the subposet in question consists of a single ground element. This result is equivalent to showing that de Bruijn cycles exist for the possible ways of filling in the elements of a collection of sets, labeled $A,B,C$, etc., whose Venn Diagram has a shape with specified intersection patterns.
The landmark work of [@cdg] extended the agenda of finding de Bruijn cycles to possibly the next most natural set of combinatorial objects, namely $k$-subsets of $[n]$. Here the authors conjectured that de Bruijn cycles for the $k$-subsets of $[n]$ existed if and only if $n$ was sufficiently large and $n\big|{n\choose k}$, which is an obvious necessary condition for the existence of the cycle in the natural coding (where, e.g., the set $\{a,b,c\}$ is encoded in the string by some permutation of its elements) since in this case each element of the set must appear equally often. In this area, important contributions have been those of [@h] and [@r]. In [@h], Hurlbert makes significant progress for the case of $k=3,4,6$; the case of $k=4$ is reduced to verification of two ground cases in [@r]. Here we follow the direction of [@bg], who proved that, in a suitable encoding, de Bruijn cycles can be created for the subsets of $[n]$ of size in the interval $[s,t]; 0\le s<t\le n$. For example, the string 1110011010 is a de Bruijn cycle of all the 2- and 3-subsets of \[4\] if one uses a sliding window of length 4 and the characteristic vector coding for subsets. Once again, we remind the reader that setting $s=0$ and $t=n$ and using a full (length $n$) encoding window for a subset yields de Bruijn’s theorem! Thus the result in [@bg] can be viewed as a restricted version of de Bruijn’s theorem. In this paper we generalize this result to exhibit de Bruijn cycles for words of length $n$ over a $k$-ary alphabet and having weight between $s$ and $t$, where the parameters $s$ and $t$ are suitably restricted. Viewed differently, this proves existence of de Bruijn cycles for multisets having between $s$ and $t$ elements, and where each element of $[n]$ may appear at most $k-1$ times in the multiset. Setting $k=2$ enables us to retrieve the result in [@bg] on subsets of cardinality between $s$ and $t$. Results for de Bruijn cycles of multisets using the “classical" coding may be found in [@hjz].
A good survey of some recent and not-so-recent papers on de Bruijn cycles may be found in [@hh]. See also [@bks], where de Bruijn cycles are proved to exist for graphs, hypergraphs, and such – using the simple and far-reaching device of invoking de Bruijn’s theorem, where it all began, together with using a suitable encoding. And, last but not least, the reader is urged to study the very recent [@dg] and order it for his/her library.
Labeled Subposets of the Boolean Lattice, a.k.a. Venn Diagrams with Specified Intersection Patterns
===================================================================================================
A poset $(P,\le)$ is a set $P$ with a relation $\le$ on $P$ that is reflexive, transitive, and antisymmetric. The Hasse diagram of a poset is a representation of $(P,\le)$ using a pictorial device that represents its [*transitive reduction*]{}. Specifically, an “upward edge" is drawn from $x$ to $y$ if $x<y$ and there is no $z$ such that $x<z<y$. The main result of this section is the following:
Let $(P,\le)$ be a fixed poset with elements of $P$ being subsets of $[n]$ and $A\le B$ if $A\subseteq B$. Assume furthermore that the Hasse diagram of $P$ is consistent with that of the Boolean Lattice $\cb_n$. There are then $\alpha^n$ assignments of the elements of $[n]$ to the elements of $P$, where $\alpha$ is the number of antichains in $P$. Furthermore, there exists a de Bruijn cycle of these assignments.
We start with the first part of the result. Given any element $j\in[n]$ it is clear that if $j\in A$ for some $A\in P$, we must have $j\in B$ for each $B\ge A$, since $\le$ is the inclusion relation. It thus follows that the number of ways to assign element $j$ to the sets in $P$ is equal to the number of ways of labeling the elements of $P$ with zeros and ones so that if $A$ is labeled by a 1, then so is each $B\supseteq A$; this includes the case where all the elements of $P$ are labeled with zeros. We call a coloring $c:P\rightarrow\{0,1\}$ [*unitarily up-closed*]{} if $c(A)=1\Rightarrow c(B)=1$ for each $B\in P$ with $B\ge A$. We will prove below that the set of unitarily up-closed colorings is in bijection with the antichains of $(P,\le)$; note that $\emptyset$ is considered to be an antichain in $P$, as are all one-element sets:
For one direction, let $c$ be a unitarily up-closed coloring on $(P,\le)$. Let $M_c := \{A\in P : c(A) = 1\ {\rm and\ }
c(B) = 0\ {\rm for\ every\ } B\le A\}$ i.e., the set of elements of $P$ minimal with respect to receiving the value 1 under $c$. Clearly, $M_c$ forms an antichain, since if there exist $A,B\in M_c$ with $A\le B$, then $A$ does not meet the constructive criteria of $M_c$. This provides a map from colorings to antichains. To see that the map is injective, let $c$ and $d$ be two different unitarily up-closed colorings, and suppose that $M_c = M_d$. Fix $A\in P$. Suppose that $c(A) = 1$. If $A$ is minimal, then we have that $d(A)$ = 1 since $M_c = M_d$. If $A$ is not minimal, it is above some $B\in M_c$. $B\in M_d$ by hypothesis, and since $B\le A$, $d(A) = 1$ as $d$ is unitarily up-closed. Now suppose that $c(A) = 0$. Since $A$ is not above [*any*]{} element of $M_c$, $A$ is also not above any element of $M_d$. Hence $d(A) = 0$, as if $d(A) = 1$ we would have that $A$ was above some minimal element under $d$.
To see the reverse, notice that for a fixed antichain $\ca$, we can define a unitarily closed up-coloring $c=c_\ca$ by the rule $c(A) = 1$ iff $B\le A$ for some $B\in\ca$, 0 otherwise. We claim that $M_c = \ca$, and hence this provides a $1-1$ inverse to our $1-1$ function. Since no element below any element of $\ca$ receives color 1, clearly $\ca\subseteq M_c$. On the other hand, every element $A$ with $c(A)=1$ is above some element of $\ca$, and since $M_c$ contains every element of $\ca$, no other element receiving color 1 can belong to $M_c$, and we are done.
It follows that the “fate" of each element of $[n]$, i.e., which sets in $P$ it belongs to, can be determined in $\alpha$ ways, and thus there are $\alpha^n$ assignments of the elements of $[n]$ to the sets in $P$.
To prove the second part, we note that each of the $\alpha$ ways of assigning element $j$ to the sets in $P$ can be coded using letters from an alphabet $\Lambda$ of size $\alpha$. By de Bruijn’s theorem, there exists a cycle of the elements of $\Lambda^{[n]}$, and thus any such de Bruijn cycle may be viewed as a de Bruijn cycle of the assignment of elements of $[n]$ to the sets in $P$. This completes the proof; specific examples follow.
[**Examples.**]{} The two letter words on a three letter alphabet $\{a,b,c\}$ may be cycled as follows: $$cc \rightarrow ca \rightarrow aa \rightarrow ab \rightarrow bb \rightarrow bc \rightarrow cb \rightarrow ba \rightarrow ac.$$ We let the letters of the alphabet code for the “allowable configurations" corresponding to the three antichains of the 2-chain with Hasse diagram
(max) at (0,4) [$B$]{}; (b) at (0,3) [$A$]{}; (max) – (b);
This is done by using a coding such as $$a={{0}\atop{0}}; \quad b={{1}\atop{0}}; \quad c={{1}\atop{1}},$$ where, e.g., ${{1}\atop{0}}$ indicates that $j\in B; j\not\in A$. Thus the de Bruijn cycle $ccaabbcba$ above may be written as $${{1}\atop{1}}\quad {{1}\atop{1}}\quad {{0}\atop{0}}\quad {{0}\atop{0}}\quad {{1}\atop{0}}\quad {{1}\atop{0}}\quad {{1}\atop{1}}\quad {{1}\atop{0}}\quad {{0}\atop{0}}$$ To decode the cycle, we read the above stack from left to right in sliding groups of two columns at a time, the first and second columns describing to which sets elements $1$ and $2$ belong. Since $n=2$, this is all we need. The actual de Bruijn cycle of assignments of $\{1,2\}$ to the two sets in the poset is then as follows:
(max) at (0,4) [$ \{1,2\} $]{}; (b) at (0,3) [$ \{1,2\} $]{}; (max) – (b);
(max) at (0,4) [$ \{1\} $]{}; (b) at (0,3) [$ \{1\} $]{}; (max) – (b);
(max) at (0,4) [$ {\emptyset } $]{}; (b) at (0,3) [$ {\emptyset } $]{}; (max) – (b);
(max) at (0,4) [$ \{2\} $]{}; (b) at (0,3) [$ {\emptyset} $]{}; (max) – (b);
(max) at (0,4) [$ \{1,2\} $]{}; (b) at (0,3) [$ {\emptyset } $]{}; (max) – (b);
(max) at (0,4) [$ \{1,2\} $]{}; (b) at (0,3) [$ \{2\} $]{}; (max) – (b);
(max) at (0,4) [$ \{1,2\} $]{}; (b) at (0,3) [$ \{1\} $]{}; (max) – (b);
(max) at (0,4) [$ \{1\} $]{}; (b) at (0,3) [$ {\emptyset } $]{}; (max) – (b);
(max) at (0,4) [$ \{2\} $]{}; (b) at (0,3) [$ \{2\} $]{}; (max) – (b);
The same may be done for a poset with any Hasse diagram. If, e.g., we have the W-poset with 13 antichains, then
\(i) We start with a de Bruijn cycle for all the $n$-letter words on an alphabet such as $\{A,B,\ldots, M\}$.
\(ii) We then rewrite this as an assignment of the elements of $\{1,2\ldots,n\}$ to the five sets in the poset, using a sliding stack of height 5 and width $n$.
\(iii) Finally, we may “draw" the $13^n$ realizations of the poset, i.e., the assignments of $n$ elements to 13 sets in a Venn diagram.
Note, however, that going from steps (i) to (ii) to (iii) leads to a sequential loss of data compression, and thus (i) gives the recommended de Bruijn cycle.
Words with Weights in Prescribed Intervals
==========================================
The main result of this section is the following:
Let $k,n\in{\mathbb Z}^+$. Consider $n$ letter words $w=(w_1,w_2,\ldots,w_n)$ on the $k$-letter alphabet $\Lambda=\{0,1,\ldots,k-1\}$, and define the weight $h(w)$ of $w$ by $h(w)=\sum_{i=1}^nw_i$. Let $s,t\in{\mathbb Z}^+$ satisfy $0\le s<s+k-1\le t\le n(k-1)$. Let $\cw$ be the collection of all such words with weights between $s$ and $t$. Then there exists a de Bruijn cycle of the elements of $\cw$.
We create a digraph $D=(V,E)$ with vertex set equal to the set of all $n-1$-letter words over the alphabet $\Lambda$ and having weights between $s-(k-1)$ and $t$. A directed edge is drawn from vertex $v_1$ to vertex $v_2$ if (i) the last $n-2$ letters of $v_1$ are the same as the first $n-2$ letters of $v_2$ and if (ii) the edge labeled by the concatenation of the letters in $v_1$ and $v_2$ yields an $n$-letter word with weight between $s$ and $t$.
It is clear that the indegree $i(v)$ and outdegree $o(v)$ of any vertex $v$ are both equal to the number of prefixes (or suffixes) that yield an edge label of legal weight; if, e.g., $h(v)=s-(k-1)$ or $h(v)=t$, then $i(v)=o(v)=1$. We will next show that $D$ has a single weakly connected component; this will show that it is Eulerian[@west], and the Eulerian path will give the required de Bruijn cycle.
Weak connectedness will established by showing that there exists a path between any vertex in $D$ and a special sink vertex $SV$ of weight $s$ that consists of a uniquely determined number $a\ge1$ of letters $x$ followed by $n-1-a\ge0$ letters that are all $x+1$. The rest of the proof of Theorem 2 consists of demonstrating four embedded lemmas, that state that we can (i) connect any vertex of weight $\ge s+1$ to one of weight $s$; (ii) connect any vertex of weight $\le s-1$ to one of weight $s$; (iii) connect any vertex of weight $s$ to one with weight $s$ and letters $x, x+1$; and (iv) connect any vertex with the right number of $x$s and $x+1$s to $SV$.
Let $v$ have weight $h(v)$ in the range $[s+1,t]$. Then there is a path from $v$ to another vertex of weight $s$.
We first cycle any zeros in the front of the vertex to the end. These steps may be taken without changing the weight of the vertex, and so that the weight of the words representing the edges is also maintained at $h(v)$. The resulting vertex, $v'$, has weight $h(v')=h(v)$ that may be as high as $t$, and has first letter different from zero. If $h(v')=t$, then the only allowable letter we can add, when the first letter is dropped, is zero. This maintains the edge weight at $t$, while leading to the new vertex weight $h(v'')$ satisfying the conditions $t>h(v'')\ge t-(k-1)\ge s$. If $h(v')=t-r\ge s+1$, we replace the first letter $f$ by $\min\{f-1, r\}$. This either reduces the vertex weight by one or changes it to $t-r-f+r=t-f\ge t-(k-1)\ge s$. The edge weight might increase from $t-r$ to $t$, but this is OK. We now see that $v''$ either has weight $s$ or, if not, has weight smaller than $h(v')$. We repeat the above process until we reach a vertex with weight $s$.
Let $v$ have weight $h(v)$ in the range $[s-(k-1),s-1]$. Then there is a path from $v$ to another vertex of weight $s$.
If $h(v)=s-(k-1)$, we drop the first letter and add the letter $k-1$. This makes the edge label equal to $s$, and the vertex weight does not decrease. If the vertex label has increased, we have made progress towards our target. If it has not, due to the first letter in $v$ being $k-1$, we repeat the process till we get to a vertex with first letter smaller than $k-1$. If the starting weight is $h(v)=s-r; 1\le r\le k-2$, we drop all letters $k-1$ at the front of the word and append $k-1$s to the end until the first letter is smaller than $k-1$, thus arriving at the vertex $v'$. We now add the letter $\max\{r, f+1\}$, where $f$ is the first letter of the vertex $v'$. This leads to the edge weight becoming either $s$ (if $r\ge f+1$) or $s-r+f+1\le t$ (if $r\le f+1$). The vertex weight goes up by 1, if $f+1\ge r$ and becomes $s-f$ if $f+1\le r$. We now iterate the above process until the vertex weight becomes $s$.
Let $v=(v_1,\ldots,v_{n-1})$ have weight $h(v)=s$. Then there is a path from $v$ to another vertex of weight $s$ and consisting entirely of letters that are either $x$ or $x+1$ (and at least one letter $x$).
Type $A$ letters are those in the set $\{0,1,\ldots,x\}$, while type $B$ letters belong to the set $\{x+1,\ldots,k-1\}$. Let $a$ and $b$ be the desired number of $x$s and $x+1$s. Depending on how many type $A$ and type $B$ letters $v$ has, we will need the path from $v$ to decrease a certain number of type $B$ letters to either $x$ or $x+1$, and to increase some or all of the type $A$ letters to either $x$ or $x+1$. Start with $v_1$. If $v_1$ is a type $A$ letter we change it to either $x$ or $x+1$ as needed. This possibly increases the vertex and edge weights, and the condition $t\ge s+(k-1)$ keeps both weights legal. (The weights don’t have to increase since we may replace the $x$ with an $x$, causing no change in vertex weight; or $x$ might equal 0, in which case there is no change in edge weight.) If $v_1$ is of type $B$ we again change it to either $x$ or $x+1$, causing a possible drop in vertex weight. We now repeat the process with the first letter $v_2$ of the new vertex $v'$ [*as long as the new vertex weight $h(v')$ satisfies*]{} $t-h(v')\ge x$ or $h(v')-(s-(k-1))\ge x$. No edge traversal in the digraph is undertaken that leads to a vertex that is closer than $x$ in weight from the two extreme vertex weights, namely $s-(k-1)$ and $t$. If such a “dangerous" occurrence is imminent, we abort such a move, cycling instead until we have the opportunity to increase a “dangerously low weight" or decrease a “dangerously high vertex weight." The process is repeated until we have the required numbers $a,b$ of symbols $x$ and $x+1$ respectively. An example is given after the proof of Corollary 3.
Let $v$ have weight $h(v)=s$ and be composed entirely of $x$s and $x+1$s as in Lemma 3. Then there is a path from $v$ to the sink vertex $SV=(x,x,\ldots,x,x+1,\ldots,x+1)$ having the same number of $x$s and $x+1$s as $v$.
The proof is similar to that of Lemma 5. We first identify both $x$s and $x+1$s in $v$ that are “out of place". Clearly the number of out of place $x$s must equal the number of out of place $x+1$s, and the latter all appear before the former. We cycle the letters of the word until we arrive at the first out of place $x+1$, delete it, and add an $x$ at the end. This decreases the vertex weight by one. We continue cycling the word and replacing $x+1$s by $x$s until the vertex weight is within $x$ of the minimum legal weight, i.e. $s-(k-1)$. The next phase is to increase the vertex weight by replacing (cyclically) out of place $x$s by $x+1$s, until the vertex weight is within $x$ of the maximum allowable, i.e. $t$. We alternate this process until the two letters are all cyclically in the right places, and then cycle until we reach $SV$.
Lemmas 3, 4, 5, and 6 together complete the proof of Theorem 2.
We define a [*Redundant de Bruijn Cycle*]{} of a collection of $\ca$ combinatorial objects to be a cycle of length $\ca'>\ca$ so that each object appears exactly once as a set of consecutive elements in the cycle, and thus so that $\ca'-\ca$ consecutive elements are redundant objects of another type.
For each $t\ge k-1$, there exists a redundant de Bruijn cycle of the $\ca(n,k,t)$ $n$-letter weight $t$ words over $\{0,1,\ldots,k-1\}$ that is of length $\ca'(n,k,t)=\ca(n,k,t)(1+o(1)); k,t$ fixed, $n\to\infty$.
We set $s=t-(k-1)\ge0$ in Theorem 2, and obtain a de Bruijn cycle of all words with weight between $t-(k-1)$ and $t$. The length of this cycle is $$\sum_{j=t-(k-1)}^t \ca(n,k,j)=\ca(n,k,t)(1+o(1)),$$ since for fixed $t$ and large $n$, $\ca(n,k,j)$ is a monotone function of $j$; $0\le j\le t$ with $\ca(n,k,j)/\ca(n,k,j+1)\to0,\ n\to\infty$. This completes the proof.
[**An Example: The Algorithm in Lemma 5 in Action.**]{} We choose $v$ to be $(0,0,0,2,2,5,5,5,3,3)$; $s=25, t=30, k=6, n=11$. The target vertex is $SV=(2,2,2,2,2,3,3,3,3,3)$. Legal vertex weights are between 20 and 30, and legal edge weights are between 25 and 30. We declare a dangerous situation to be one in which further progress would lead to vertex weights of 20, 21, 29, or 30 – unless we cycle to a vertex with appropriate first letter. Below, red numbers represent vertex weights, blue numbers represent edge weights, and the symbol D represents “Danger." We proceed as follows:
[0,0,0,2,2,5,5,5,3,3]{} [ 25]{}\
$\downarrow$ [ 28]{}\
[0,0,2,2,5,5,5,3,3,3]{} [ 28, D]{}\
$\downarrow$ [ 28]{}\
[0,2,2,5,5,5,3,3,3,0]{} [ 28, D]{}\
$\downarrow$ [ 28]{}\
[2,2,5,5,5,3,3,3,0,0]{} [ 28, D]{}\
$\downarrow$ [ 30]{}\
[2,5,5,5,3,3,3,0,0,2]{} [ 28, D]{}\
$\downarrow$ [ 30]{}\
[5,5,5,3,3,3,0,0,2,2]{} [ 28]{}\
$\downarrow$ [ 30]{}\
[5,5,3,3,3,0,0,2,2,2]{} [ 25]{}\
$\downarrow$ [ 27]{}\
[5,3,3,3,0,0,2,2,2,2]{} [ 22, D]{}\
$\downarrow$ [ 27]{}\
[3,3,3,0,0,2,2,2,2,5]{} [ 22, D]{}\
$\downarrow$ [ 25]{}\
[3,3,0,0,2,2,2,2,5,3]{} [ 22, D]{}\
$\downarrow$ [ 25]{}\
[3,0,0,2,2,2,2,5,3,3]{} [ 22, D]{}\
$\downarrow$ [ 25]{}\
[0,0,2,2,2,2,5,3,3,3]{} [ 22]{}\
$\downarrow$ [ 25]{}\
[0,2,2,2,2,5,3,3,3,3]{} [ 25]{}\
$\downarrow$ [ 28]{}\
[2,2,2,2,5,3,3,3,3,3]{} [ 28, D]{}\
$\downarrow$ [ 30]{}\
[2,2,2,5,3,3,3,3,3,2]{} [ 28, D]{}\
$\downarrow$ [ 30]{}\
[2,2,5,3,3,3,3,3,2,2]{} [ 28, D]{}\
$\downarrow$ [ 30]{}\
[2,5,3,3,3,3,3,2,2,2]{} [ 28, D]{}\
$\downarrow$ [ 30]{}\
[5,3,3,3,3,3,2,2,2,2]{} [ 28]{}\
$\downarrow$ [ 30]{}\
[3,3,3,3,3,2,2,2,2,2]{} [ 25]{}\
$\downarrow$ [ 28]{}\
[3,3,3,3,2,2,2,2,2,3]{} [ 25]{}\
$\downarrow$ [ 28]{}\
[3,3,3,2,2,2,2,2,3,3]{} [ 25]{}\
$\downarrow$ [ 28]{}\
[3,3,2,2,2,2,2,3,3,3]{} [ 25]{}\
$\downarrow$ [ 28]{}\
[3,2,2,2,2,2,3,3,3,3]{} [ 25]{}\
$\downarrow$ [ 28]{}\
[2,2,2,2,2,3,3,3,3,3]{} [ 25]{}
thus arriving at the sink vertex.
Further Research
================
It would be interesting to improve Theorem 2 so that the range of the weight of the word is as small as possible, or else to prove that the range in Theorem 2 is the best possible. Also, one might ask how and to what extent one can show existence of de Bruijn cycles for unordered posets. Last but not least, can results be proved for (labelled as well as unlabelled) subposets of mother posets other than the Boolean Lattice?
Acknowledgments
===============
The research of AG was supported by NSF Grant 1004624.
[99]{} A. Blanca and A. Godbole (2011). “On Universal Cycles for new classes of combinatorial structures," [*SIAM J. Discrete Math.*]{} [**25**]{}, 1832–1842. G. Brockman, B. Kay, E. Snively (2010). “On universal cycles of labeled graphs," [*Electr. J. Combinatorics*]{} [**17**]{}, Paper R4. F. Chung, P. Diaconis, and R. Graham (1992). “Universal cycles for combinatorial structures," [*Discrete Math.*]{} [**110**]{}, 43–59. N. G. De Bruijn (1946). “A combinatorial problem," [*Nederl. Akad. Wetensch. Proc.*]{} [**49**]{}, 758–764. P. Diaconis and R. Graham (2011). [*Magical Mathematics: The Mathematical Ideas that Animate Great Magic Tricks*]{}, Princeton University Press. G. Hurlbert (1994). “On universal cycles for $k$-subsets of an $n$-set," [*SIAM J. Discrete Math.*]{} [**7**]{}, 598–604. V. Horan and G. Hurlbert (2013+). “Universal Cycles for weak orders," Preprint, see [http://arxiv.org/pdf/1203.5169.pdf]{} G. Hurlbert, T. Johnson, and J. Zahl (2009). “On universal cycles for multisets," [*Discrete Math.*]{} [**309**]{}, 5321–5327. Y. Rudoy (2013+). “An inductive approach to constructing Universal Cycles on the $k$-subsets of $[n]$", Preprint, see [http://arxiv.org/pdf/1209.4662v1.pdf]{} D. West (1996). *Introduction to Graph Theory*, Prentice Hall, New Jersey.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Compliant and soft hands have gained a lot of attention in the past decade because of their ability to adapt to the shape of the objects, increasing their effectiveness for grasping. However, when it comes to grasping highly flexible objects such as textiles, we face the dual problem: it is the object that will adapt to the shape of the hand or gripper. In this context, the classic grasp analysis or grasping taxonomies are not suitable for describing textile objects grasps. This work proposes a novel definition of textile object grasps that abstracts from the robotic embodiment or hand shape and recovers concepts from the early neuroscience literature on hand prehension skills. This framework enables us to identify what grasps have been used in literature until now to perform robotic cloth manipulation, and allows for a precise definition of all the tasks that have been tackled in terms of manipulation primitives based on regrasps. In addition, we also review what grippers have been used. Our analysis shows how the vast majority of cloth manipulations have relied only on one type of grasp, and at the same time we identify several tasks that need more variety of grasp types to be executed successfully. Our framework is generic, provides a classification of cloth manipulation primitives and can inspire gripper design and benchmark construction for cloth manipulation.'
author:
- 'Júlia Borràs, Guillem Alenyà, Carme Torras, [^1] [^2]'
title: '**A Grasping-centered Analysis for Cloth Manipulation**'
---
Cloth manipulation, grasping taxonomy, dexterous manipulation, robot grippers.
Introduction {#sec:introduction}
============
Robot manipulation in human environments has experienced great progress in recent years [@torras2016service]. However, efforts have been focused mostly on rigid objects, and core capabilities such as grasping, placing, or handing to a person still remain a hard and unsolved problem when dealing with challenging objects such as textiles. Indeed, textiles present many additional challenges with respect to rigid object manipulation, including difficult perception, complexity in modeling the object and predicting its behavior [@jimenez2017visual], but it is of special importance in the case of domestic and service robotics as textiles are present in many contexts of our daily lives.
As a consequence, this topic is gaining a lot of attention from EU Research committees. This work is framed in the ERC advanced grant CLOTHILDE, started in 2018, which aims to develop a theory of cloth manipulation based on a topology-rich, compact cloth representation, advanced cloth perception, probabilistic motion planning and learning from demonstrations. In the project, we are studying how to define manipulation complexity and/or performance measures in terms of used grasps, computational time and quality of goal attainment. However, for that to be possible, we need a benchmark. Finding a proper benchmark and replicable experiments is challenging for robot manipulation in general [@Bonsignorio2015; @quispe2018taxonomy] and, in particular for cloth-like objects, benchmarks are almost inexistent. We believe that a proper classification of the types of grasps and manipulation primitives used in handling textile objects is one of the steps necessary to develop a future such benchmark.
In this work, we analyze and classify the types of grasps and the manipulation primitives that have appeared in literature, focusing on manipulations for handling clothes, including sorting, folding, unfolding, picking & placing, hanging, etc. This is one of the three demanding applications addressed in the CLOTHILDE project, the other two being fitting elastic covers and helping handicapped people to dress, which we leave out of the scope of this paper.
Despite being a relatively novel topic, there is already an extensive literature about robots perceiving and dealing with clothes in household tasks. A good recent review covering cloth-like objects, among others, can be found in [@sanchez2018robotic]. One of the main conclusions of the study is that manipulation skills still remain underdeveloped compared to vision or control for cloth manipulation and that improved end-effectors need to be developed to better handle this kind of objects. Indeed, there are only a few works focusing specifically on grasping of cloth-like objects, that is, on how and in how many different ways cloth can be grasped, and most of them have been developed for an industrial application and are not suitable for domestic environments [@Koustoumpardis2004; @MOULIANITIS1999].
{width="\textwidth"}
. \[fig:grippers\]
On the contrary, grasping has been studied extensively for rigid objects [@Feix2016Grasp; @prattichizzo2008grasping]. However, most works assume not only that the object is rigid, but also that it is fixed inside the hand and can be held in any direction. This is not true for clothes in almost any case. Furthermore, grasp classification has been highly dependent on object shape and size, while this distinction does not make much sense for clothes because cloth will conform to the shape of the gripper. Note also that rigid-object grasping research has started only recently to consider extrinsic contacts [@dafle2014extrinsic; @Chavan-Dafle2015; @Eppner2015], that is, the use of contacts with the environment to improve grasp success. Instead, cloth manipulation has been utilizing extrinsic contacts since the very beginning, because they are badly needed to deal with this type of extremely deformable materials. It is worth highlighting that, in the context of rigid objects, environmental contacts are viewed as exploitable constraints to ease their grasping [@Eppner2015], but not as grasps [*per se*]{} as they don’t fix their pose. Instead, such contacts can be considered grasps for nonrigid objects, since they just partly confine their shape in the same way as grippers do. Finally, cloth manipulation is mostly bimanual and most of the grasps have to be used in couples. In conclusion, we believe that further study of grasping centered on clothes, and considering all the aspects mentioned above, is necessary and could improve the robustness, autonomy and versatility of cloth manipulation by robots.
Another interesting insight of the review in [@sanchez2018robotic] is that the literature shows a clear tendency to develop techniques focused on a specific task and, thus, finding general solutions remains a major open issue. Identifying manipulation primitives that are common across different tasks can be of crucial relevance to move forward in this direction.
In this work we pave the way towards this end by analyzing previous attempts at robotizing specific cloth manipulation tasks under a common framework derived from three distinguishing traits of textiles: 1) their conforming to the gripper geometry (dual of soft manipulation), 2) the possibility to treat environmental contacts as grasps, thus allowing for a unified treatment, and 3) the need of two simultaneous grasps (bimanual handling) to accomplish most tasks. These traits permit abstracting from the particularities of existing grippers and defining some handy geometries and their combinations, inspired by the concepts of virtual fingers and opposition spaces from early neuroscience works on hand prehension [@iberall1986opposition]. The proposed framework delineates a well-defined landscape, which permits identifying grasps missing in previous research that may supply the versatility we are aiming at, leading to alternative ways of performing a given task or even the capacity to tackle new tasks. Other far-reaching possibilities are the design of grippers suitable to accomplish a given repertoire of tasks, and building benchmarks to assess the performance of perception, control and manipulation algorithms using some type of gripper to perform a precisely specified subset of tasks.
The paper is structured as follows. analyzes all the grippers used for cloth manipulation. In we propose a framework to characterize textile grasps and manipulations, which is then applied in to systematize grasps and manipulations tackled in literature. In we show an experiment to assess the influence of grasp type on task performance, and discuss the implications of our analysis and several possible applications. Finally, in we provide conclusions and outline our future work.
Gripper designs for manipulating textiles {#sec:grippers}
=========================================
shows, as far as the authors know, all the grippers and robotic hands used in the literature focusing on cloth manipulation and presenting robotic experiments. Next we will describe how these grippers are used and what kinds of grasps they can perform.
First, we observe that when cloth is manipulated by sliding a point contact on a table, no real gripper is needed (.**a**). Of course, then the functionality is limited to flattening clothes on a table. Most papers tackling general cloth manipulation use simple off-the-shelf parallel grippers like . **b** to **g**, from different robot platforms like PR2 robot, HIRO, Nekonote robot arm, Baxter and HRP2 robot, respectively. They all have relatively small planar fingertips and perform pinch grasps. Others use generic robotic hands such as -**q** to **t**, corresponding to the RobotiQ, Barret and Shadow hands respectively. In these cases, the hands usually perform a pinch grasp between the thumb and 2 fingers, except in the case of [@ramisa2012using; @Monso2012] that use different finger configurations from the Barrett hand to pick and place crumpled clothes, discussed in .
It is interesting to note how those tasks that require a bit more of support use grippers that perform a grasp between two lines. These are . **m**, **o** and **p**. The linear grip in **p** is used to grasp all the side of a T-shirt, therefore simplifying a bimanual folding operation to using a single hand. The gripper in . **m** is used in [@jia2017manipulating] to maintain tension between grasps and therefore a larger part of the cloth can be controlled. Finally, the gripper in .**o** was used for the task in [@colome2018dimensionality]. Colomé and Torras apply learning techniques to fold a T-shirt by rotating the shoulders of the T-shirt in the air. The gripper in the figure performs a grasp between two lines of very short length, and it could almost be considered as a pinch grasp. However, later they redesigned the gripper into the version shown at the top left of . The grasp between longer lines enables to transmit more torque effectively to the cloth, reducing the training time. A comparison of manipulation motions using different gripper geometries will be presented in .
Finally, we will pay a bit more of attention to the grippers specially designed for manipulating clothes in domestic environments (.**h** to **p** and **u** and **v**, that is, the right column of the table). Some of these are parallel grippers with some modification at the fingertips, like .**h** that has a ball bearing to ease sliding, or .**k** and **l** that use rolling fingertips to allow sliding along or pulling upwards a pinched cloth, respectively. .**i** is effectively a parallel gripper with a very thin finger to slide under flat clothes, and a touch sensor at the other fingertip. A linear brush was added at the side to be able to perform a flattening operation by sliding it on top of the cloth, as shown in the left image in .**i**. Designs **j** and **n** show a similar idea: to have small grippers at the fingertips of a parallel gripper in **j** or joined by a prismatic actuator in **n**. These designs allow in-hand manipulation such as grasp gaiting, or edge tracing with a single gripper in the case of **n**, although such a big long gripper would not be practical in domestic environments. .**u** shows an under-actuated gripper specially designed for pinching and clamping edges and corners of a flat cloth laying on a table. In addition, it could perform the same pinch grasp as the multifingered robot hands that we mentioned before. Finally, the hand in .**v** is a humanoid-like 3-fingered hand with a palm, and it is specially designed to be able to slide on a cloth with a flat hand, perform a pinch between the palm and one finger or a pinch between two fingers. These operations are identified as requirements for cloth manipulation in [@ono2001picking].
In conclusion, off-the-shelf grippers can only perform pinch grasps, which suffice to execute most of the tasks. Other specially-designed grippers exhibit a higher versatility, but more testing is needed to validate their usefulness for a wide range of cloth manipulation tasks.
A framework proposal for robotized clothes handling {#sec:framework}
===================================================
Characterizing grasps in terms of opposing geometric entities {#subsec:grasping_framework}
-------------------------------------------------------------
Grasping analysis and taxonomies have been highly influenced by neuroscience works in the 80’s that defined parameters to characterize hand prehensile postures in terms of virtual fingers and opposition spaces. Iberall’s works [@iberall1986opposition; @arbib1985coordinated] established that, to grasp an object, at least two opposing forces against the object surface are needed. The opposition space is defined by the possible directions of the two forces. The concept of virtual finger (VF) groups fingers or hand surfaces that apply forces in the same direction into functional units. In other words, each finger and hand surfaces working in a grasp are assigned to one of the VF, and most grasps can be defined by just two VFs exerting opposing forces. This permits the abstraction from the mechanics of the hand and its fingers.
{width="75.00000%"}
Iberall’s works were centered on the human hand, and they described hand postures as collections of VFs able to apply effective forces against the object. She showed that standard prehensile classifications like Cutkosky’s [@cutkosky1989grasp] could be captured with this framework and classified in terms of VFs and opposition directions, as done later by Feix et al. [@Feix2016Grasp].
Although traditionally the concept of VF has been applied to the human hand, the framework could be much widely used to characterize grasping by artificial means as well [@iberall1997human]. Indeed, it is shown in [@iberall1991parameterizing] that a grasp could be fully described and analyzed in terms of the parameters that describe VFs: parameters of the grasping contact surface patch (length, width and orientation w.r.t. the palm), the orientation of the force the VF could exert and the sensor information.
[|l|c|c|]{} Reference &
---------
Gripper
in
---------
: Grasps used in literature[]{data-label="tab:graspsInPapers"}
& Grasps used\
[@cuen2008action] & a & [P$\Pi_e$]{}\
[Maitin-Shepard et al. 2010 ]{} [@maitin2010cloth] & b & [PP]{}, 2[PP]{}, 2[$\Pi\Pi_e$]{}, 2[PP]{}+[$\Pi_e$]{}\
[Bersch et al. 2011 ]{} [@bersch2011bimanual] & b & [PP]{}, 2[PP]{}, 2[$\Pi\Pi_e$]{}\
[Miller et al. 2012 ]{} & b & [PP]{}+[$\Pi_e$]{}, 2[PP]{}+[$\Pi_e$]{}\
[Lee at al. 2015 ]{}[@lee2015learning] & b & 2[PP]{}, 2[PP]{}+[$\Pi_e$]{}\
[Yuba et al. 2017 ]{} [@yuba2017unfolding] & c & [PP]{}, 2[PP]{}, 2[PP]{}+[$\Pi_e$]{}\
[Koishihara et al. 2017 ]{} [@koishihara2017hanging] & c & [PP]{}, [PP]{}+[L$_i$]{}$^1$\
[Willimon et al. 2011 ]{} [@willimon2011model] & d & [PP]{}+[$\Pi_e$]{}\
[Moriya et al. 2018 ]{} [@moriya2018method] & e & [PP]{}\
[Li et al. 2015 ]{} [@li2015regrasping; @li2015folding] & f & [PP]{}, 2[PP]{}, 2[PP]{}+[$\Pi_e$]{}, [PP]{}+[$\Pi_e$]{}\
[Kita et al. 2011 ]{} [@kita2011clothes] & g & [PP]{}, 2[PP]{},\
[Osawa et al. 2007 ]{} [@osawa2007unfolding] & h & [PP]{}, 2[PP]{},\
[Doumanoglou et al. 2016 ]{} [@Doumanoglou2016] & i & [PP]{}, 2[PP]{}, 2[PP]{}+[$\Pi_e$]{}, [P$\Pi_e$]{}+[L$\Pi_e$]{}\
[Sun et al. 2015 ]{} [@sun2015accurate] & i & [PP]{}+[$\Pi_e$]{}, 2[PP]{}+[$\Pi_e$]{}\
[Sahari et al. 2010 ]{} & j,k & [PP]{}, 2[PP]{}, sh 2[PP]{}\
[Hamajima et al. 2000 ]{} [@hamajima2000planning] & l & [PP]{}, sh 2[P$\Pi_e$]{}\
[Kabaya et al. 1998 ]{} [@kabaya1998service] & l & [PP]{}, sh 2[P$\Pi_e$]{}\
[Jia et al. 2017 ]{} [@jia2017manipulating] & m & 2[LL]{}, [LL]{}+[$\Pi_e$]{}\
[Shibata et al. 2012 ]{} [@shibata2009wiping; @shibata2012fabric] & n & sh 2[PP]{}, sh 2[PP]{}+[$\Pi_e$]{}\
[Colome and Torras 2018 ]{} [@colome2018dimensionality] & o & 2[LL]{}, 2[LL]{}+[$\Pi_e$]{}\
[Sugiura et al. 2010 ]{} [@Foldy2010] & p & [LL]{}+[$\Pi_e$]{}\
[Yamakawa et al. 2011 ]{} [@yamakawa2011motion] & q & 2[PP]{}\
[Ruan et al. 2018 ]{} [@ruan2018accounting] & r & 2[PP]{}+[$\Pi_e$]{}\
[Balaguer and Carpin 2011 ]{} [@balaguer2011combining] & s & 2[PP]{}, 2[PP]{}+[$\Pi_e$]{}\
[Monso et al. 2012 ]{} [@Monso2012] & s & [PP]{}, sh2[PP]{}, [L$\Pi$]{}$^2$\
[Ramisa et al. 2012 ]{} [@ramisa2012using] & s & [PP]{}\
[Twardon et al. 2015 ]{} [@twardon2015interaction] & t & [PP]{}, 2[PP]{},\
[Koustoumpardis et al. 2014 ]{} & u & [PP]{}, sh 2[P$\Pi_e$]{}\
[Ono and Takase 2007 ]{} [@ono2007better; @ono2001picking] & v & [PP]{}, sh 2[P$\Pi_e$]{}, [P$\Pi_e$]{}, [$\Pi\Pi_e$]{}\
Total grasp instances & & 63\
Notes\
$1.$ Line contact is achieved by grasping a hanger with the second gripper.\
$2.$ From the finger configurations in -[**s**]{}, the far left configuration leads to a sh2[PP]{}, where the first [PP]{} is between one finger and the right side of thumb and the second [PP]{} is achieved with the other finger and the left side of the thumb. For the next to the right image, when fingers close in symmetric configuration, the achieved grasp is a pinch that is either a [PP]{} grasp or a sh2[PP]{} grasp as before. With the configuration on the right, the hand achieves a [L$\Pi$]{} grasp where the line is formed with the 3 aligned fingertips that will exert force against the plane of the palm. All grasps are used to pick crumpled clothes.
We propose to define textile grasps in terms of the number and geometry of VFs. In other words, [*geometric virtual fingers*]{} define the shape of the contact patch and the orientation with respect to each VF. Implicitly, this also determines the possible directions of the opposition forces. For textiles, we can characterize most of the grasps using only three geometric entities: points (P), lines (L) and planes ($\Pi$); and *we define a grasp as a set of opposing geometric entities*. We believe this definition makes sense in the context of pure shapeless objects that will conform to the geometry of the grasping device, also implying the shape of the object will not influence the definition of the grasp.
[|l|lll|]{} **Task Name** & **Start** &[**Steps**]{} & **End**\
1\. (a) Unfold in the air & (H1) [PP]{} at p &
[l l]{} ${\bf 1)}$ (H2) [ ]{}Gr v: [PP]{} & ${\bf 2)}$ (H1) [ ]{}Rlse [PP]{}\
${\bf 3)}$ (H1) [ ]{}Gr v: [PP]{} &
& 2[PP]{} unfd\
1. (b) Unfold in the air & (H1) [PP]{} at p &
[l l]{} ${\bf 1)}$ (H1) [ ]{}Add [$\Pi_e$]{}: [PP]{}+[$\Pi_e$]{}& ${\bf 2)}$ (H1) [ ]{}Slide to reveal corner\
${\bf 3)}$ (H2) [ ]{}Gr v: 2[PP]{}+[$\Pi_e$]{}& ${\bf 4}$ [ ]{}Rm [$\Pi_e$]{}: 2[PP]{}\
${\bf 5)}$ Repeat from ${\bf 1)}$ to ${\bf 4)}$ changing hands
& 2[PP]{} unfd\
1\. (c) Unfold in the air tracing & [$\Pi_e$]{} crumpled &
[l l]{} ${\bf 1)}$ (H1) [ ]{}Gr v: [PP]{} +[$\Pi_e$]{}& ${\bf 2}$ (H1) [ ]{}Rm [$\Pi_e$]{}: [PP]{}\
${\bf 3)}$ (H2)[ ]{}Gr e near H1: 2[PP]{} & ${\bf 4)}$ [ ]{}Trace edge until H2 finds v
& 2[PP]{} unfd\
2. Fold on table & [$\Pi_e$]{} flat or fd &
[l l]{} ${\bf 1)}$ [ ]{}Gr 2 v: 2[PP]{} + [$\Pi_e$]{}& ${\bf 2)}$ [ ]{}Fold ${\bf 3)}$ [ ]{}Rlse 2[PP]{} on tbl
& [$\Pi_e$]{} fd\
3. (a) Fold in the air with table & 2[LL]{} unfd &
[l l ]{} ${\bf 1)}$ [ ]{}Pronation rot. each H & ${\bf 2)}$ [ ]{}Add [$\Pi_e$]{}: 2[LL]{}+[$\Pi_e$]{} folding\
${\bf 3)}$ [ ]{}Rlse 2[LL]{} from inside fold &
& [$\Pi_e$]{} fd\
3. (b) Fold in the air dynamic & 2[PP]{} unfd &
[l l ]{} ${\bf 1)}$ [ ]{}Fast move cloth until bottom vertices move up &\
${\bf 2)}$ [ ]{}Regrasp 2[PP]{} to add the bottom vertexs &
& 2[PP]{} fd\
4. Place on table flat or folding & 2[PP]{} unfd &
[l l]{} ${\bf 1)}$ [ ]{}Add [$\Pi_e$]{}: 2[PP]{} + [$\Pi_e$]{}& ${\bf 2)}$ [ ]{}Rlse 2[PP]{} on tbl
& [$\Pi_e$]{} flat or fd\
5. (a) Flatten cloth & [$\Pi_e$]{} p.flat &
[l ]{} ${\bf 1)}$ [ ]{}Gr 2{[L$\Pi_e$]{}, [P$\Pi_e$]{}}\
${\bf 2)}$ [ ]{}Slide H1, hold H2
& [$\Pi_e$]{}(p)flat\
5. (b) Flatten cloth & [$\Pi_e$]{} p.flat &
[l ]{} ${\bf 1)}$ [ ]{}Gr 2[$\Pi\Pi_e$]{}\
${\bf 2)}$ [ ]{}Slide H1 & H2 in opp. dir.
& [$\Pi_e$]{} (p)flat\
5. (c) Flatten cloth & [$\Pi_e$]{} p.flat &
[l l]{} ${\bf 1)}$ [ ]{}Gr {v, e}: $\{1, 2\}$[PP]{}+[$\Pi_e$]{}\
${\bf 2)}$ [ ]{}Pull H(s) to flatten &${\bf 3)}$ [ ]{}Rlse $\{1, 2\}$[PP]{}on table
& [$\Pi_e$]{}(p)flat\
5. (d) Flatten cloth &
-------------------
[$\Pi_e$]{}p.flat
(1 v folded)
-------------------
&
[l l]{} ${\bf 1)}$ (H1) [ ]{}Gr p [sh 2P$\Pi_e$]{} to hold cloth & ${\bf 2)}$ (H2) [ ]{}Gr v: [PP]{} while H1 holds\
${\bf 3)}$ (H1) [ ]{}Rlse [sh 2P$\Pi_e$]{} & ${\bf 4)}$ (H2) [ ]{}Unfold v and rls [PP]{}
& [$\Pi_e$]{} (p)flat\
6. Separate from pile (pick & place) & [$\Pi_e$]{} crumpled &
[l l]{} ${\bf 1)}$ [ ]{}Gr p w. [PP]{} or [$\Pi\Pi$]{} & ${\bf 2)}$ [ ]{}Rlse
& [$\Pi_e$]{} crumpled\
7. Pick & place folded cloth & [$\Pi_e$]{} fd &
[l l]{} ${\bf 1)}$ [ ]{}Gr 2[PP]{} from 2 sides and lift & ${\bf 2)}$ [ ]{}Rlse 2[PP]{} on table
& [$\Pi_e$]{} fd\
8. Clamp (p)flat cloth (pick up) & [$\Pi_e$]{} (p)flat &
[l l]{} ${\bf 1)}$ [ ]{}Move gripper finger under cloth\
${\bf 2)}$ [ ]{}Grasp cloth at e/v: [PP]{} + [$\Pi_e$]{} & ${\bf 3)}$ [ ]{}Rm [$\Pi_e$]{}
& [PP]{} at e/v\
9\. Pinch (p)flat cloth (pick up) & [$\Pi_e$]{} (p)flat &
[l l]{} ${\bf 1)}$ [ ]{}Gr [sh 2P$\Pi_e$]{} & ${\bf 2)}$ [ ]{}Bring P contacts together\
${\bf 3)}$ [ ]{}Rgr to [PP]{} + [$\Pi_e$]{}& ${\bf 4)}$ [ ]{}Rm [$\Pi_e$]{}
& [PP]{} at p\
10. Pick single layer from stack & [$\Pi_e$]{} flat &
[l ]{} ${\bf 1)}$ (H1) [ ]{}Gr p [P$\Pi_e$]{} near edge\
${\bf 2)}$ (H1) [ ]{}Push and slide inwards cloth to create crease and hold\
${\bf 3)}$ (H2) [ ]{}Grasp edge [PP]{} on crease and [ ]{}Rlse H1
& [PP]{} at p\
11\. Grasp gaiting & (H1) [PP]{} at p &
[l l]{} ${\bf 1)}$ (H2) [ ]{}Gr p [PP]{} near H1 & ${\bf 2)}$ (H1) [ ]{}Rlse [PP]{}\
${\bf 3)}$ Iterate &
& [PP]{} at p\
12. Hang & 2[PP]{} unfd &
[l l]{} ${\bf 1)}$ [ ]{}Add [P$_e$]{}: 2[PP]{} + [P$_e$]{}& ${\bf 2)}$ [ ]{}Rlse 2[PP]{}
& [P$_e$]{}\
13. Put hanger in T-shirt & (H1) [PP]{} at p &
[l ]{} ${\bf 1)}$ (H2) [ ]{}Insert side of hanger in neck hole, adding a [L]{}\
${\bf 2)}$ (H2) [ ]{}Pull between [L]{} and [PP]{} to insert other side of hanger\
& [L]{}\
14.Japanese style T-shirt fold & [$\Pi_e$]{} flat &
-----------------------------------------------------------------------------------------------
${\bf 1)}$ [ ]{}Gr$^2$: 2[PP]{} + [$\Pi_e$]{}
${\bf 2)}$ [ ]{}H1 moves to opposite edg, H2 holds
${\bf 3)}$ (H1) [ ]{}Add edge p to [PP]{} ${\bf 4)}$ [ ]{}Rm [$\Pi_e$]{}
${\bf 5)}$ [ ]{}shake$^3$ ${\bf 6)}$ [ ]{}Add [$\Pi_e$]{} ${\bf 7)}$ [ ]{}Rlse 2[PP]{} on tbl
-----------------------------------------------------------------------------------------------
& [$\Pi_e$]{} fd\
Notation: v: vertex/corner of cloth, e: point on the edge of cloth, p: any point interior of cloth\
Abbreviations: H: hand, F: fold, Fd: Folded, Unf: unfold, Unfd: Unfolded, Gr: Grasp, Rm: Remove, edg: edge, Rlse: Release, Tr: Trace, sh: single hand, bm: bimanual, p.flat: partially flat.
We can easily see at this point that the environment geometries play a very relevant role as grasping agents to increase functionalities of the grippers. Therefore, we will consider environmental contacts as extra fingers, so that a geometric VF can be either intrinsic, that is, defined by the contact patch of the gripper/hand, or extrinsic, that is, defined by contact surfaces from the environment. Any extrinsic VF will be denoted with the subindex “$_e$”. In this context, and as done also in [@dafle2014extrinsic], we consider as a grasp an object lying on a table. That is, the table is an extrinsic planar VF, denoted [$\Pi_e$]{}.
Because cloth manipulation is intrinsically bimanual, we will denote bimanual grasps with the addition symbol, in a way that the flattening operation in .**i** is represented by [P$\Pi_e$]{}+[L$\Pi_e$]{}, and if the grasps are symmetric in both hands, like for instance two hands holding a cloth in the air with two pinch grasps, we simply write 2[PP]{}. We assume that any double grasp is bimanual, and we mark it as a single handed one otherwise, using “sh”. For instance, the grasp in .**n** is defined as sh 2[PP]{}. Similarly, a single grasp is assumed to be single handed, but can be marked as bimanual using “bm” if it is performed by two hands.
In the context of our framework, we define both prehensile and non-prehensile grasps. Any combination of geometries of different dimensions will lead to a non-prehensile grasp, while the same dimensions lead to prehensile ones. In general, according to definitions in literature [@bullock2012hand], even when the two geometries are the same, like a [PP]{} grasp or a [$\Pi\Pi$]{} grasp, they perform a non-prehensile grasp because generally they will not fix the position of the entire object with respect to the hand. However, in the context of our framework, we will consider these as prehensile grasps (prehensile inside the limits of the hand), to distinguish them from the other “less” prehensile grasps. Finally, any grasp involving only extrinsic geometries, like [$\Pi_e$]{}, is also non-prehensile because it assumes the second opposition force is gravity.
In we will apply this framework to describe all the grasps used by the grippers in , summarized and classified in .
Manipulation primitives as grasp transitions {#subsec:grasping_manipulation}
--------------------------------------------
Similarly as in [@dafle2014extrinsic], after having established the repertoire of grasps we will deal with, we can consider as manipulation primitives either regrasping actions or motions with a fixed grasp. Within this framework, we define six types of manipulation primitives:
- [ ]{}Adding/removing extrinsic contacts. These are regrasp actions where the origin/destination has an extrinsic VF while the destination/origin doesn’t, respectively.
- [ ]{}Grasping. A regrasp from a single extrinsic VF grasp (first row in ) to any other grasp in the figure.
- [ ]{}Releasing. The opposite of grasping, that is, a regrasp from any grasp to a single extrinsic VF grasp.
- [ ]{}Regrasping. Pure regrasp action, that is, any transition between grasps that does not belong to the two previous types.
- [ ]{}Sliding. Manipulation where the initial and end grasps are the same, but the cloth has moved with respect to the grasp without losing contact completely. Examples include edge tracing or flattening on a table.
- [ ]{}In-grasping motions. Single hand or bimanual grasps that are fixed and moved by the arms without performing any regrasp but changing the state of the cloth. Examples include applying tension between two [PP]{} grasps or the motion to fold a cloth after is grasped.
Then, we can define a task as a sequence of these types of manipulations. In we will describe all the tasks addressed in literature as sequences of manipulation primitives of the above six types.
Analyzing cloth manipulation tasks using the proposed framework {#sec:analyzingManipulations}
===============================================================
Gripper capabilities in terms of the proposed framework {#subsec:applicationOnGrasps}
-------------------------------------------------------
The use of geometric virtual fingers is general in the sense that it can be applied to the human hand by mapping its five fingers and palm to a geometric arrangement of VFs, but it can also be applied to other generic grippers like the ones listed in the previous section. Revisiting the grippers in we realize that if we define a point as a relatively small flat surface, most of the grippers above perform a [PP]{} grasp (a pinch) (.**b**-**g**, **q**-**t**, **i** and **u**) or a sh 2[PP]{} grasp (.**j** and **n**). Some of them perform a [PP]{} grasp with additional abilities to roll over clothes (.**h, k** and **l**). In addition, the grippers in .**m** and **p** perform an [LL]{} grasp.
[[Osawa et al. 2007 ]{} [@osawa2007unfolding]]{} [[Maitin-Shepard et al. 2010 ]{} [@maitin2010cloth]]{} [[Kita et al. 2011 ]{} [@kita2011clothes]]{} [[Bersch et al. 2011 ]{} [@bersch2011bimanual]]{} [[Li et al. 2015 ]{} [@li2015regrasping; @li2015folding]]{} [[Doumanoglou et al. 2016 ]{} [@Doumanoglou2016]]{} [[Customano-Towner et al. 2011 ]{} [@cusumano2011bringing]]{} [[Sahari et al. 2010 ]{} [@sahari2010Clothes; @sahari2010edge; @salleh2008inchworm]]{} [[Shibata et al. 2012 ]{} [@shibata2009wiping; @shibata2012fabric]]{} [[Yuba et al. 2017 ]{} [@yuba2017unfolding]]{} [[Sugiura et al. 2010 ]{} [@Foldy2010]]{} [[Miller et al. 2012 ]{}[@miller2012geometric; @miller2011parametrized; @van2010gravity; @lakshmanan2013constraint]]{} [[Lee at al. 2015 ]{}[@lee2015learning]]{} [[Jia et al. 2017 ]{} [@jia2017manipulating]]{} [[Colome and Torras 2018 ]{} [@colome2018dimensionality] ]{} [[Yamakawa et al. 2011 ]{} [@yamakawa2011motion]]{} [[Balaguer and Carpin 2011 ]{} [@balaguer2011combining]]{} [[Ruan et al. 2018 ]{} [@ruan2018accounting]]{} [[Cu[é]{}n Roch[í]{}n et al. 2008 ]{} [@cuen2008action]]{} [[Willimon et al. 2011 ]{} [@willimon2011model]]{} [[Sun et al. 2015 ]{} [@sun2015accurate]]{} [[Hamajima et al. 2000 ]{} [@hamajima2000planning]]{} [[Monso et al. 2012 ]{} [@Monso2012]]{} [[Ramisa et al. 2012 ]{} [@ramisa2012using]]{} [[Moriya et al. 2018 ]{} [@moriya2018method]]{} [[Kabaya et al. 1998 ]{} [@kabaya1998service]]{} [[Koustoumpardis et al. 2014 ]{} [@koustoumpardis2014underactuated; @Koustoumpardis2017]]{} [[Ono and Takase 2007 ]{} [@ono2007better; @ono2001picking]]{} [[Twardon et al. 2015 ]{} [@twardon2015interaction] ]{} [[Koishihara et al. 2017 ]{} [@koishihara2017hanging]]{} [[Bell et al. 2010 ]{} [@bell2010flexible]]{}
--------------------------- --- -------------------------------------------------- -------------------------------------------------------- ---------------------------------------------- --------------------------------------------------- ------------------------------------------------------------- ----------------------------------------------------- --------------------------------------------------------------- ---------------------------------------------------------------------------------------- ----------------------------------------------------------------------- ------------------------------------------------ ------------------------------------------- ------------------------------------------------------------------------------------------------------------------------ -------------------------------------------- ------------------------------------------------- -------------------------------------------------------------- ----------------------------------------------------- ------------------------------------------------------------ ------------------------------------------------- ------------------------------------------------------------ ---------------------------------------------------- --------------------------------------------- ------------------------------------------------------- ----------------------------------------- ------------------------------------------------ ------------------------------------------------- -------------------------------------------------- --------------------------------------------------------------------------------------------- ---------------------------------------------------------------- --------------------------------------------------------- ---------------------------------------------------------- -----------------------------------------------
1\. (a) Unfold in the air 6 x x x x x x
1\. (b) Unfold in the air 1 x
1\. (c) Unfold sliding 3 x x x
2\. Fold on table 8 x$^1$ x x x x x$^3$ x
3\. Fold in the air 1 x$^1$ x
3\. Dynamic fold 1 x
4\. Place on table 5 x$^1$ x$^1$ x x x x
5\. (a) Flatten cloth 2 x x$^4$
5\. (b) Flatten cloth 1 x$^1$
5\. (c) Flatten cloth 2 x x
5\. (d) Flatten cloth 1 x
6\. Separate from pile 3 x x x
7\. Pick & place fd cl 2 x$^1$ x$^5$
8\. Clamp (p)flat cl 5 x x$^2$ x x x
9\. Pinch (p)flat Clamp 3 x x x
10\. Pick single layer 1 x
11\. Grasp gaiting 1 x
12\. Hang 1 x
13\. Put hanger 1 x
14.Japanese fold 1 x
TT 1 5 1 3 3 4 1 2 2 2 1 1 1 1 1 1 1 1 1 2 1 1 1 1 1 2 3 3 1 1 1
$1.$ Open loop procedure. $2.$ Manipulation to first bend the edge to be able to clamp. $3.$ Applying controlled tension between grasps. $4.$ Single handed (without H2 holding). $5.$ Single handed.
If we consider the extrinsic geometries they use, the grippers in .**b** and **v** can perform a [$\Pi\Pi_e$]{} grasp, the gripper in .**i** a [L$\Pi_e$]{} and [P$\Pi_e$]{} grasps and the gripper in .**u, l** and **v** a [sh 2P$\Pi_e$]{} grasp, necessary to perform a pinch of a cloth that is lying flat on a table. The gripper .**u** is specially designed to be able to perform simultaneously a [PP]{} and a [sh 2P$\Pi_e$]{} grasp, used to grasp a folded corner while holding the cloth under. The most versatile gripper in terms of possible textile grasps is the one in .**v** from [@ono2007better] that can perform the grasps [$\Pi\Pi_e$]{}, [P$\Pi$]{}, [PP]{} and [sh 2P$\Pi_e$]{}. summarizes the relationship between all the reviewed works, what gripper they use and how many grasps they can realize.
shows all these grasps represented in a schematic way, as well as the grasp combinations appearing in the reviewed bibliography. Note that bimanual robots performing cloth manipulation tasks in the literature have dealt with up to 5 virtual fingers, leading to up to three couples of opposition forces.
Encoding previously tackled tasks within the framework {#subsec:applicationOnTasks}
------------------------------------------------------
In we list all the tasks that we have identified in literature related to cloth folding. For each task, we define the corresponding sequence of primitives and label each primitive with its type, using the nomenclature proposed in . In addition, shows what works address which tasks.
Unfold in the air (task 1(a)) consist in grasping a crumpled cloth and regrasp it until its gasped flat by two significant points. It has been solved in [@osawa2007unfolding; @maitin2010cloth; @kita2011clothes; @bersch2011bimanual; @li2015regrasping; @Doumanoglou2016]. These are all vision-based solutions where grasping points have to be located to regrasp the cloth by the corners (or other grasping points) to permit its recognition. This task relies on vision and recognition algorithms, more than on the manipulation itself. A good review of such algorithms can be found in [@sanchez2018robotic]. As regards to manipulation, what is important is the gripper ability to grasp the correct point without changing the configuration of the cloth during the approach. In a similar way (task 1(b)), some authors ease the recognition and grasping of cloth corners by sliding the cloth on a table [@cusumano2011bringing], leading to a slightly modified strategy of the same task (task 1(b)). The same objective of tasks 1(a) and (b) is accomplished in task 1(c) simplifying the vision requirements and increasing the challenge of manipulation. It is based on localizing only one corner on a table (provided it is visible), grasp it, and reach for another corner with the second hand by sliding the gripper along the edge of the cloth without losing it [@yuba2017unfolding]. To increase the chances of having at least one corner visible, the authors previously slide the cloth on the table, similarly as in task 1(b). This edge tracing manipulation can be faster because it reduces the number of grasping points that need to be localized, but the manipulation of “pinching and sliding” is of high complexity, in many cases requiring specially designed grippers [@sahari2010Clothes; @shibata2012fabric]. Unfolding in the air is a crucial task to recognize the manipulated cloth and set the canonical cloth state to start any other manipulation, therefore, it is one of the most popular tasks, appearing 10 times in .
Once the cloth is unfolded in the air, placing it on the table (task 4) can be done with an open-loop procedure, but challenges may arise at planning time if relying on a cloth model that needs to accurately deal with friction and collisions [@ruan2018accounting]. Moreover the task may require taking into account complex dynamics to ensure the cloth is laid flat [@balaguer2011combining].
Only three of the tasks involve folding itself: fold on table, fold in the air and dynamic fold in the air (tasks 2, 3(a) and 3(b)). The first has been the second most attempted task in literature. Fold on table consists in grasping two vertices and folding the cloth until the desired fold is attained [@Doumanoglou2016; @li2015folding; @miller2012geometric]. Planning such folds requires recognizing the cloth and deciding where the folding lines should go [@miller2012geometric] but also to compute the hand trajectories to realize the manipulation [@li2015folding]. A big linear gripper can simplify the manipulation [@Foldy2010] but is not general enough to deal with several pieces of cloth of different sizes. Another challenge is to monitor how the cloth is being folded in a controlled manner, for instance, with vision [@jia2017manipulating] or with force control manipulation maintaining tension between hands [@lee2015learning]. The task of folding in the air is tackled in [@colome2018dimensionality] and [@bersch2011bimanual]. It presents several manipulation challenges that are not fully solved in the papers; for instance, the grasping and releasing is not addressed in [@colome2018dimensionality] and the task is done as an open-loop procedure in [@bersch2011bimanual] without assessing if the T-shirt is properly folded. This approach to folding promises to be much faster than the other, but it may require different procedures depending on the cloth item. The dynamic fold done in [@yamakawa2011motion] is also fast, but entails a very complex and risky manipulation, and it requires a high-velocity hardware.
Cloth flattening on the table has been attempted by 7 authors, making it the third most popular task, which has been tackled with several different approaches (task 5(a to d)). Some works slide surfaces of the grippers on top of the cloth [@maitin2010cloth; @Doumanoglou2016; @cuen2008action] and others grasp edges or vertices and pull them [@willimon2011model; @sun2015accurate]. The challenges are to understand the state of the cloth, localize the creases and define a direction of sliding or pulling to flatten the cloth. A more challenging task from the manipulation point of view would be that of unfolding a corner that has remained folded. Such grasp is difficult because it is easy to grasp at the same time the cloth that lies under the corner. A specially designed gripper for this task is presented in [@koustoumpardis2014underactuated] (.**u**) which uses one finger to hold the cloth underneath and another finger to grasp the corner on top.
Another important task is pick and place. Regarding crumpled clothes, pick and place becomes challenging when only one cloth needs to be picked up from a bunch of several tangled clothes (task 6). Again, the main challenge in this task comes from vision [@ramisa2012using; @hamajima2000planning]. Picking up folded clothes, a task that would be very relevant in any domestic environment, was done in [@maitin2010cloth] using an open-loop procedure and it is approached in more detail in [@moriya2018method], but only using one hand. The paper focuses on localizing the edge of the folded piece that doesn’t have loose layers and, therefore, it can be picked without being unfolded.
Other popular tasks are those of clamping and pinching (tasks 8 and 9). A pinch consists in grasping the cloth from one side, while in a clamp each finger is in a different side of the cloth. The main challenge for the clamp is to go under the cloth before grasping while, for a pinch, a [sh 2P$\Pi_e$]{} grasp needs to be performed and then the cloth dragged by bringing the two point contacts together. The same strategy for pinching could also lead to a clamp [@koustoumpardis2014underactuated]. Works like [@shibata2012fabric] create first a crease in the cloth to facilitate clamping. Note that a pinch or a clamp is necessary to grasp the cloth to fold it on a table (task 2), but not all the reviewed works addressing task 2 appear in for clamping or pinching because some of them don’t care if the grasp was a clamp or a pinch as long as it was close to the edge [@maitin2010cloth; @miller2012geometric; @li2015folding]. However, this distinction may be more relevant when the quality of the end result becomes more important. Nowadays, it is taken into account in papers whose main focus is to design a gripper for textile manipulation [@koustoumpardis2014underactuated; @kabaya1998service; @Le2013development].
The last five tasks in the list appear only in one paper. Picking a single layer is approached in [@ono2001picking] by pressing on the pile to create more spacing allowing to clamp one single layer. However, this approach has been solved in industry with vacuum grippers quite efficiently. Grasp gaiting could substitute the “pinch and slide” to trace an edge, so as to reduce the risk of losing the cloth [@sahari2010edge]. However, it will probably turn out to be too slow compared to sliding to constitute a practical solution.
Finally, the tasks of hanging clothes or putting hangers inside garments are of obvious utility in domestic and commercial environments and they will probably attract more attention in the near future.
[|l|r|]{} **Primitive** & **\#**\
\
Add [P$_e$]{} to 2[PP]{} & 1\
Add [$\Pi_e$]{} to 2[LL]{}, 2[PP]{} or [PP]{} & 1,2,2\
Rm [$\Pi_e$]{} from 2[PP]{}+[$\Pi_e$]{} & 2\
Rm [$\Pi_e$]{} from [PP]{}+[$\Pi_e$]{} & 2\
&\
Total & 10\
\
Gr [P$\Pi_e$]{} or sh 2[P$\Pi_e$]{} & 1,2\
Gr 2[P$\Pi_e$]{} or 2[$\Pi\Pi_e$]{} & 1,1\
Gr [$\Pi\Pi$]{} & 1\
Gr 2[PP]{} & 1\
Gr 2[PP]{} + [$\Pi_e$]{} & 2\
Gr [PP]{} + [$\Pi_e$]{} & 4\
&\
Total & 13\
[|l|r|]{} **Primitive** & **\#**\
\
Release [PP]{}, 2[PP]{} or [$\Pi\Pi$]{} & 1,1,1\
Release 2[LL]{} from inside a fold & 1\
Release [PP]{} or 2[PP]{} on a table & 2, 4\
Release [P$\Pi_e$]{} or sh 2[P$\Pi_e$]{} & 1,1\
&\
Total & 12\
\
From 2[PP]{} to [PP]{} and viceversa & 2,4\
From [PP]{}+[$\Pi_e$]{}to 2[PP]{}+[$\Pi_e$]{}& 1\
From [PP]{} to [PP]{}+[L]{} & 1\
From [PP]{}: Add point to [PP]{}& 2\
From sh 2[P$\Pi_e$]{} to [PP]{}+sh 2[P$\Pi_e$]{}& 1\
From sh 2[P$\Pi_e$]{} to [PP]{} + [$\Pi_e$]{} & 1\
&\
Total & 12\
[|l|r|]{} **Primitive** & **\#**\
\
in 2[PP]{}+[$\Pi_e$]{}: H1 folds & H2 holds or both Hs fold & 1,1\
in [PP]{}+[$\Pi_e$]{}: Pull to flatten or unfold the gr. corner & 1,1\
in 2[PP]{}: Shake to move hanging v. & 2\
in 2[LL]{}: Pronation rotation of each H & 1\
in [P$\Pi_e$]{}: Push inwards cloth (to create crease) & 1\
in sh 2[P$\Pi_e$]{}: Bring P contacts together & 1\
in [PP]{}+Ł: Apply tension between hands & 1\
in [$\Pi_e$]{}: Move gripper finger under cloth & 1\
Total & 11\
\
in 2[$\Pi\Pi_e$]{}: Slide H1 & H2 in opp. dir. & 1\
in 2[L$\Pi_e$]{} or 2[P$\Pi_e$]{}: Slide H1, hold H2 & 1\
in 2[PP]{}: Trace edge & 1\
in [PP]{}+[$\Pi_e$]{}: Slide cloth on table to reveal corner & 1\
Total & 4\
![Setup of the experiments with three different grasp types to fold a T-shirt in the air. The left column shows frames of the task using the [LL]{}grasp, the middle column for the [L$\Pi$]{}grasp and the right one for the [PP]{}grasp. Bottom line shows the trajectories corresponding to the right gripper of one trial, with color code representing acceleration norm. []{data-label="fig:experimentSetup"}](figs/Figure3.pdf){width="\linewidth"}
Applications of the framework {#sec:applications}
=============================
Study of manipulation parameters under different grasp types when folding a T-shirt in the air {#subsec:experiment}
----------------------------------------------------------------------------------------------
So far, we assume and claim that choosing one type of grasp or another among those we have defined could greatly influence the performance of a task. To study this influence in more detail we have designed a simple experiment where we recorded motion data of a subject *wearing* different types of grippers, using the XSens suit and performing a cloth manipulation task. The hands were rigidly holding the grippers, preventing relative motion between hand and gripper.
We compared the three grippers that are shown at the top of . The first is the evolution of the gripper that appears in -o, it is a parallel gripper where each closing part consists of two points. Assuming two points form a line, we can define that grasp as a [LL]{}. The next to the right gripper consists of 2 rigid lines at the bottom, forming a plane, and a linear finger on top. Therefore, that gripper can perform a [L$\Pi$]{} grasp. Finally, the last gripper performs a simple [PP]{} grasp. All the grippers are used passively with a human-in-the-loop manipulation, and we assume the cloth is pre-grasped. The task performed is described as the task 3 in , “Fold a T-shirt in the air”, but in this case there is no release step.
We repeated the tasks 10 times, wearing the three grippers. We collected upper body motion data with the Xsens suit, using an extra sensor attached to the right gripper to directly record its position, velocity and acceleration. This extra sensor is part of the Xsens suit and is usually used to record motion of a manipulated object or a tool, like for instance, a sword. The bottom graphics in show example trajectories corresponding to one trial of each gripper, each point colored with the corresponding linear acceleration norm. In -(top) we show the different velocity profiles for each gripper. Each curve and shaded area represents the mean and standard deviation of all the trial curves, aligned to match the maximum and minimum peaks. At the bottom, we compare the maximum acceleration peaks for the trials performed with the different grippers.
The results of the experiment show evidence on how grasp type can influence the execution of a task. When the gripper provides more support surface along the T-shirt shoulders, the task of folding requires less acceleration, indicating that the dynamics of the task is greatly simplified. We cannot observe a significant difference between the [LL]{} gripper and the [L$\Pi$]{} gripper, they both provide sufficient support and therefore, the acceleration peak is similar. Instead, we can appreciate differences in the trajectories (bottom of ) because the rotation pivot point is aligned with the rotation axis of the wrist for the [LL]{} gripper, while with the other gripper the user tends to rotate it about the abducted fingertip.
Finally, at the last frames of the task shown in , we can observe that the quality of the fold is also different depending on the gripper used, the results being clearly worst for the [PP]{} grasp. We could observe this, but we haven’t included the results because we do not have a fair measure to compare qualities of fold.
Increasing versatility and task repertoire {#subsec:discussion}
------------------------------------------
With the experiment in the previous section we have seen how the dynamics of the task of “Fold a T-shirt in the air” can be greatly simplified by just using the appropriate grasp type. This can reduce the training times when applying learning techniques, and also change the quality of the result. But there are more examples where improving manipulation skills can reduce the complexity of many of the perception and pose estimation algorithms. For instance, the task “Unfold in the air” may seem relatively easy by looking at the manipulation primitives used, but finding and localizing the two corners and successfully grasping them is a difficult endeavor that usually involves complex vision and recognition algorithms. A way of reducing the complexity of recognizing the second grasp corner is for the second hand to grasp an edge point close to the first grasp point, and then trace the edge until the next corner is reached, which can potentially be faster than looking for the second corner. This is at the expense of gaining manipulation complexity, meaning that more attention has to be put on touch sensors and increasing contact surfaces to gain more tactile feedback. There are many other examples of the importance of versatile grasping abilities. The task “Fold on table”, for instance, has been always addressed in a very similar way: grasping two corners laying on the table and performing a coordinated motion with the arms. However, this may not be the best approach and sometimes it doesn’t yield very good end results. Other approaches could consist in folding with one hand while the other holds the cloth by the folding line. However, for that to happen we probably need a linear grasp on the edge and a planar or linear hold along the folding line, which is not possible with most of the off-the-shelf existing grippers.
![(Top) Comparison between the norms of the linear velocity (left) and the angular velocity (right) of the grippers during the executions of the trials. (Bottom) Comparison between peak accelerations (left) and angular acceleration (right) of the grippers during execution of the fold in the air.[]{data-label="fig:results"}](figs/Figure4.pdf){width="\linewidth"}
Moreover, this approach to folding can’t be applied to large clothes like bed sheets or tablecloths. We humans usually fold them in the air, starting from an unfolded 2[PP]{} grasp, to bring the two corners together, regrasp them in a single hand and use the other hand to trace the edge until the end of the fold, to end up again in a 2[PP]{} grasp where the cloth has one extra fold. When several of these folds have been done, the last one to finish can be done on the table. For big bed sheets, this same strategy is started with 4 hands (2 people). This is of course a complex manipulation that requires more versatile grippers to be executed successfully and more manipulation skills. However, this kind of operations, or something equivalent, will be needed eventually if robots are to fold big textile objects.
Alternative grasps could also be needed depending on the cloth material. As an example, the task of picking and placing a folded garment (task 7 in ) has been implemented in [@maitin2010cloth] using a 2[PP]{} grasp. This was successful for the folded towels used in the paper, but they required to be flattened after being placed on the table. We believe that attempting this task with only point contacts would not work for a thinner textile material, as it would unfold while transferring it. Using a planar contact under the garment can provide better support to achieve the task, for instance a 2[P$\Pi$]{} or 2[L$\Pi$]{} grasp, and prevent it from unfolding while transferring it or having to flatten it after begin placed. In [@moriya2018method] the task is performed with a single [PP]{} grasp using a parallel gripper. Provided the correct edge is grasped, one can lift the folded cloth without unfolding it, but to place it back one needs to perform a motion similar to “Place flat on table” (task 4) to prevent unfolding the cloth. This motion requires free space to allow the arm to move. However, if the task requires to place the folded clothes inside a shelf, as it may be expected for folded clothes, that space for the motion may not be available. Therefore, here using a 2[P$\Pi$]{} or 2[L$\Pi$]{} grasp becomes again relevant to reduce the workspace volume needed to perform a task.
Extrinsic contacts are fundamental in cloth folding, for instance to perform a pinch. Other strategies have been explored for rigid objects in [@Eppner2015] and they should be explored and evaluated in the context of cloth manipulation. For instance, bringing the object to the table edge to facilitate clamping. This has only been considered in [@koustoumpardis2014underactuated] as a scenario to evaluate a novel gripper for clothes, but should receive more attention to evaluate its usefulness with respect to other strategies.
In conclusion, we believe that to achieve a more varied repertoire of manipulations, increase success ratios and improve the quality of results, we need, among other things, to be capable of executing more diverse and sophisticated grasps.
Providing guidance for gripper design {#subsec:gripperDesign}
-------------------------------------
The definition of grasps for textiles using geometric virtual fingers permits expanding the taxonomy in with as many combinations of geometries one can think of. In the previous section we have already identified some tasks that could use other geometries, like “Pick & place folded cloth” could use a plane to support the safe transfer of the object or “Fold a T-shirt in the air” benefits from a linear geometry that can transmit more torque to bend in the shoulders of the T-shirt.
collects all the manipulation primitives appearing as task steps in , which add up to 62 primitives. The table classifies them by type and counts the multiplicity of each one. We can see how most of the primitives appear only once, indicating the complexity and particularity of each manipulation task. This may explain the lack of general solutions that are found in cloth manipulations, as pointed out in [@sanchez2018robotic].
It is striking that the implementation of manipulation primitives has so strongly relied on the PP grasp so far. This is probably due to the fact that generic parallel-jaw grippers are the most available and, when designing grippers to handle textiles, they were tailored to a very specific application. More versatile grippers able to deal with cloth could achieve many more primitives with different and more tailored functionalities to ease reaching the objective of every task.
Identifying all the individual primitives that we want a gripper to execute can help to identify the requirements each grasp will need in terms of friction, force or precision, which are to be taken into account when we design grippers. For instance, sliding on cloth or releasing from inside a fold require low friction, but hanging a cloth with a [PP]{} grasp may require a lot of force, or increased friction. Grasping a cloth corner requires precision while a 2[P$\Pi$]{} to pick up a folded cloth may not require to grasp a particular point.
In our work [@donaire2020versatile; @donaire2019ma] we defined gripper requirements using our framework, selecting three tasks and extracting their required grasp geometries. We selected to implement a [LL]{} geometry with one linear finger on top, and two bottom finger that can abduct to morph the gripper into a [L$\Pi$]{} geometry, similar to the gripper in the top middle of . We aimed at providing the skills for performing three tasks: picking and placing folded clothes using a [L$\Pi$]{} grasp, the edge tracing task and folding a T-shirt in the air with this same [L$\Pi$]{} grasp. To execute all the primitives we needed variable friction capacities, so we took inspiration from the design in [@spiers2018variable]. Our design method consisted in implementing progressive simple motor-less prototypes to test the geometries, like the one shown in the top-middle picture in , and to better define the requirements for each primitive. Finally, we implemented a fully functional prototype that could execute the required tasks.
Aiming at versatility in the design of grippers to handle textiles is a very challenging goal which, up to the authors’ knowledge has been addressed only by very few authors before and was solved only partially [@ono2007better; @koustoumpardis2014underactuated], probably because there hasn’t been a deep study of the grasping and manipulation variety needed to manipulate clothes. Our framework offers a systematic way to explore grasp combinations, with potential to identify the most useful to enhance manipulation primitives, thus suggesting versatile end-effector designs, including multigrip and reconfigurable grippers able to reshape themselves into virtual fingers with different geometric contacts.
Identifying and characterizing manipulation primitives
------------------------------------------------------
Following , a given task can be segmented according to the used grasps, to obtain different segments that correspond to manipulation primitives like the ones in . Combined with motion capture approaches, this can provide large amounts of labeled manipulation data, provided we can either recognize grasp type from the data (using data gloves) or by manually labeling them.
Assigning semantic tags for each of the primitives, as it is done in , allows to learn semantic sequencing of steps within tasks, with applications to autonomous high-level task planning. On the lower level, having data about trajectories, velocities and accelerations for each of the primitives enables to learn motion representations, for instance DMPs, for each of them. Therefore, this permits defining motion building blocks that can be sequenced to perform a given task. Similar approaches have been successfully applied in literature for rigid object manipulation [@aksoy2011learning; @aein2018library] or whole-body human motions [@kulic2012incremental].
Our framework opens the door to apply these kind of methods to cloth manipulation tasks. In order to successfully implement this kind of approaches fully from collecting data to the execution level, several additional problems need to be solved, including efficient action representation for each of the primitives and efficient cloth state representation, perception and estimation. In our research group we are working in parallel on several of these problems in the context of the ERC Clothilde project.
Benchmarking manipulation
-------------------------
One of the great challenges of benchmarking manipulation is to understand the complexities of all the possible skills that need to be evaluated, thus classification efforts are needed to set up the grounds for a proper benchmark. Several examples exist in literature where classifications like grasping taxonomies [@Feix2016Grasp], ontologies of actions [@worgotter2013simple], or task taxonomies [@quispe2018taxonomy] are used to evaluate robot performance in carrying out tasks, mostly involving rigid objects. We are working on using the proposed framework to classify and evaluate the complexity of a given task. Assessing the complexity of a task is useful to rank tasks in a benchmark and to enable their evaluation, specially if the evaluation has to be done qualitatively. In other words, if we can just state if a task was successfully completed or not, a quantification of the complexity of the task can attribute value to a solution.
In addition, our framework puts special effort in abstracting from the human hand shape and the robot embodiment, which is of special importance when benchmarking manipulation, as many research labs have different robotic systems solving similar problems, and comparison of results must be independent of the used platform.
Conclusions and future research {#sec:conclusions}
===============================
We have proposed a framework to characterize and systematize grasps, manipulation primitives and tasks for the versatile handling of clothes by robots. Grasps have been redefined in terms of prehension geometries, that we called geometric virtual fingers. This generic definition abstracts from the particular robotic embodiment that is used and can easily inspire gripper designs. Because cloth will conform to the shape of the grasping geometry, this grasp definition is not influenced by object shape or size. Instead, our definition focuses on the geometric shape of the gripper contact area where the robot has to transmit forces to the hanging part of the cloth. In the specialized literature only some combinations of geometries have appeared, the [PP]{} being the most prominent one by far. Similarly, the most used grippers can only implement a basic pinch.
We have identified several tasks that are solved in previous works using pinch grasps, which could be improved if more versatile grippers were used. The following types of improvements are envisaged:
- [*More efficient learning and execution of a given task.*]{} For example, when folding a T-shirt in the air, an L contact geometry could help bend each shoulder, thus transmitting the appropriate dynamics to the hanging part of the shirt, and reducing required dynamics of the robot arm.
- [*Higher quality of the end result.*]{} A clothing item could be better folded if a folding line were rigidly maintained, either intrinsically or extrinsically, or tension were applied between parts of the cloth.
- [*Wider range of textile materials for which a task could be completed.*]{} Providing larger support surfaces or multiple well-placed contacts could ensure a suitable end result for objects of varied fabrics.
- [*Increased repertoire of robotized tasks.*]{} For instance, it is not possible to fold a big piece of cloth (e.g., a bed sheet) or turn inside out a shirt with state-of-the-art procedures.
- [*Reduced workspace requirements.*]{} Clothes could be placed on a tight shelf or a drawer if a gripper implementing 2[P$\Pi$]{} or 2[L$\Pi$]{} grasps were available, capable of compressing a pile of folded clothes.
- [*Reduced auxiliary technical requirements.*]{} Dependency on complex vision methods could be reduced by relying on robust manipulation primitives such as sliding to trace edges or flatten a cloth lying on a table.
In addition, our framework considers both intrinsic and extrinsic geometries, allowing for the definition of extrinsic dexterous manipulations in a natural way, just as additional re-grasping operations, that is, particular instances of manipulation primitives as appearing in Table IV.
Our analysis has shown that many relevant cloth manipulation tasks have been only barely addressed so far or require major improvements. Identifying the manipulation primitives and the grasps required by each task following the presented framework will help determine the kind of gripper and the actions a robot needs to be able to perform. Moreover, such systematic task specification is a step towards establishing a benchmark for cloth manipulation, in which alternative strategies could be explored and evaluated in terms of success ratios and quality of end results.
We are currently carrying out a human study –using a motion capture system and some hand-held grippers– to evaluate what grasps humans use to execute a range of cloth handling tasks. Such grasps will then be encoded in terms of geometric virtual fingers and the study will serve to validate and refine our framework, by perhaps expanding the list of grasps and tasks, so as to ultimately generate a complete benchmark of cloth manipulation primitives.
[10]{}
C. Torras, “Service robots for citizens of the future,” [*European Review*]{}, vol. 24, no. 1, pp. 17–30, 2016.
P. Jim[é]{}nez, “Visual grasp point localization, classification and state recognition in robotic manipulation of cloth: An overview,” [*Robotics and Autonomous Systems*]{}, vol. 92, pp. 107–125, 2017.
F. Bonsignorio and A. P. del Pobil, “Toward replicable and measurable robotics research \[from the guest editors\],” [*[IEEE]{} Robotics [&]{} Automation Magazine*]{}, vol. 22, no. 3, pp. 32–35, 2015.
A. H. Quispe, H. B. Amor, and H. I. Christensen, “A taxonomy of benchmark tasks for robot manipulation,” in [*Robotics Research*]{}, pp. 405–421, Springer, 2018.
J. Sanchez, J.-A. Corrales, B.-C. Bouzgarrou, and Y. Mezouar, “Robotic manipulation and sensing of deformable objects in domestic and industrial applications: a survey,” [*The International Journal of Robotics Research*]{}, vol. 37, no. 7, pp. 688–716, 2018.
P. Koustoumpardis and N. Aspragathos, “A review of gripping devices for fabric handling,” in [*Int. Conf. on Intelligent Manipulation and Grasping*]{}, 2004.
V. Moulianitis, A. Dentsoras, and N. Aspragathos, “A knowledge-based system for the conceptual design of grippers for handling fabrics,” [*[AI]{} [EDAM]{}*]{}, vol. 13, no. 01, 1999.
S. Cu[é]{}n Roch[í]{}n, J. Andrade-Cetto, and C. Torras, “Action selection for robotic manipulation of deformable objects,” in [*Frontier Science Conference Series for Young Researchers: Experimental Cognitive Robotics (FSCYR: ECR)*]{}, ESF-JSPS, 2008.
J. Maitin-Shepard, M. Cusumano-Towner, J. Lei, and P. Abbeel, “Cloth grasp point detection based on multiple-view geometric cues with application to robotic towel folding,” in [*IEEE Int. Conf. on Robotics and Automation*]{}, pp. 2308–2315, 2010.
H. Yuba, S. Arnold, and K. Yamazaki, “Unfolding of a rectangular cloth from unarranged starting shapes by a dual-armed robot with a mechanism for managing recognition error and uncertainty,” [*Advanced Robotics*]{}, vol. 31, no. 10, pp. 544–556, 2017.
B. Willimon, S. Birchfield, and I. Walker, “Model for unfolding laundry using interactive perception,” in [*IEEE/RSJ Int. Conf. on Intelligent Robots and Systems*]{}, pp. 4871–4876, 2011.
Y. Moriya, D. Tanaka, K. Yamazaki, and K. Takeshita, “A method of picking up a folded fabric product by a single-armed robot,” [*ROBOMECH Journal*]{}, vol. 5, no. 1, p. 1, 2018.
Y. Li, Y. Yue, D. Xu, E. Grinspun, and P. K. Allen, “Folding deformable objects using predictive simulation and trajectory optimization,” in [ *IEEE/RSJ Int. Conf. on Intelligent Robots and Systems*]{}, pp. 6000–6006, 2015.
Y. Kita, F. Kanehiro, T. Ueshiba, and N. Kita, “Clothes handling based on recognition by strategic observation,” in [*2011 11th IEEE-RAS Int. Conf. on Humanoid Robots (Humanoids)*]{}, pp. 53–58, 2011.
F. Osawa, H. Seki, and Y. Kamiya, “Unfolding of massive laundry and classification types by dual manipulator,” [*Journal of Advanced Computational Intelligence and Intelligent Informatics*]{}, vol. 11, no. 5, pp. 457–463, 2007.
A. Doumanoglou, J. Stria, G. Peleka, I. Mariolis, V. Petrik, A. Kargakos, L. Wagner, V. Hlavac, T.-K. Kim, and S. Malassiotis, “Folding clothes autonomously: A complete pipeline,” [*[IEEE]{} Transactions on Robotics*]{}, vol. 32, no. 6, pp. 1461–1478, 2016.
M. J. Thuy-Hong-Loan Le, A. Landini, M. Zoppi, D. Zlatanov, and R. Molfino, “On the development of a specialized flexible gripper for garment handling,” [*Journal of Automation and Control Engineering*]{}, vol. 1, no. 3, 2013.
K. S. M. Sahari, H. Seki, Y. Kamiya, and M. Hikizu, “Edge tracing manipulation of clothes based on different gripper types,” [*Journal of Computer Science*]{}, vol. 6, no. 8, pp. 872–879, 2010.
T. Kabaya and M. Kakikura, “Service robot for housekeeping: Clothing handling,” [*Journal of Robotics and Mechatronics*]{}, vol. 10, pp. 252–257, 1998.
K. Hamajima and M. Kakikura, “Planning strategy for task of unfolding clothes,” [*Robotics and Autonomous Systems*]{}, vol. 32, no. 2-3, pp. 145–152, 2000.
B. Jia, Z. Hu, J. Pan, and D. Manocha, “Manipulating highly heformable materials using a visual feedback dictionary,” [*arXiv preprint arXiv:1710.06947*]{}, 2017.
M. Shibata and S. Hirai, “Fabric manipulation utilizing contacts with the environment,” in [*IEEE Int. Conf. on Automation Science and Engineering*]{}, pp. 442–447, 2012.
A. Colom[é]{} and C. Torras, “Dimensionality reduction for dynamic movement primitives and application to bimanual manipulation of clothes,” [*IEEE Transactions on Robotics*]{}, vol. 34, no. 3, pp. 602–615, 2018.
Y. Sugiura, T. Igarashi, H. Takahashi, T. Gowon, C. L. Fernando, M. Sugimoto, and M. Inami, “Foldy: Graphical teaching for garment-folding robot,” in [*18 Workshop on interactive system and software (WISS 2010)*]{}, pp. 7–12, 2010. (in Japanese).
A. Namiki, Y. Imai, M. Ishikawa, and M. Kaneko, “Development of a high-speed multifingered hand system and its application to catching,” in [ *Proceedings 2003 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2003)(Cat. No. 03CH37453)*]{}, vol. 3, pp. 2666–2671, IEEE, 2003.
Y. Yamakawa, A. Namiki, and M. Ishikawa, “Motion planning for dynamic folding of a cloth with two high-speed robot hands and two high-speed sliders,” in [*2011 IEEE International Conference on Robotics and Automation*]{}, pp. 5486–5491, IEEE, 2011.
M. Ruan, D. Mc Conachie, and D. Berenson, “Accounting for directional rigidity and constraints in control for manipulation of deformable objects without physical simulation,” in [*2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)*]{}, pp. 512–519, IEEE, 2018.
A. Ramisa, G. Alenya, F. Moreno-Noguer, and C. Torras, “Using depth and appearance features for informed robot grasping of highly wrinkled clothes,” in [*IEEE Int. Conf. on Robotics and Automation*]{}, pp. 1703–1708, 2012.
P. Monsó, G. Alenyà, and C. Torras, “Pomdp approach to robotized clothes separation,” in [*IEEE/RSJ Int. Conf. on Intelligent Robots and Systems*]{}, 2012.
B. Balaguer and S. Carpin, “Combining imitation and reinforcement learning to fold deformable planar objects,” in [*IEEE/RSJ Int. Conf. on Intelligent Robots and Systems*]{}, pp. 1405–1412, 2011.
L. Twardon and H. Ritter, “Interaction skills for a coat-check robot: Identifying and handling the boundary components of clothes,” in [*IEEE Int. Conf. on Robotics and Automation*]{}, pp. 3682–3688, 2015.
P. N. Koustoumpardis, K. X. Nastos, and N. A. Aspragathos, “Underactuated 3-finger robotic gripper for grasping fabrics,” in [*Robotics in Alpe-Adria-Danube Region (RAAD), 2014 23rd Int. Conf. on*]{}, pp. 1–8, 2014.
E. Ono, K. Kitagaki, and M. Kakikura, “Picking up a piece of fabric from layers by a hand with 3 fingers and a palm,” in [*IEEE/RSJ Int. Conf. on Intelligent Robots and Systems*]{}, vol. 2, pp. 931–936, 2001.
E. Ono and K. Takase, “On better pushing for picking a piece of fabric from layers,” in [*IEEE Int. Conf. on Robotics and Biomimetics*]{}, pp. 589–594, 2007.
T. Feix, J. Romero, H.-B. Schmiedmayer, A. M. Dollar, and D. Kragic, “The [GRASP]{} taxonomy of human grasp types,” [*[IEEE]{} Transactions on Human-Machine Systems*]{}, vol. 46, no. 1, pp. 66–77, 2016.
D. Prattichizzo and J. C. Trinkle, “Grasping,” in [*Springer handbook of robotics*]{}, pp. 671–700, Springer, 2008.
N. C. Dafle, A. Rodriguez, R. Paolini, B. Tang, S. S. Srinivasa, M. Erdmann, M. T. Mason, I. Lundberg, H. Staab, and T. Fuhlbrigge, “Extrinsic dexterity: In-hand manipulation with external forces,” in [*IEEE Int. Conf. on Robotics and Automation*]{}, pp. 1578–1585, 2014.
N. Chavan-Dafle and A. Rodriguez, “Prehensile pushing: In-hand manipulation with push-primitives,” in [*IEEE/RSJ Int. Conf. on Intelligent Robots and Systems*]{}, 2015.
C. Eppner, R. Deimel, J. [Á]{}lvarez-Ruiz, M. Maertens, and O. Brock, “Exploitation of environmental constraints in human and robotic grasping,” [*The International Journal of Robotics Research*]{}, vol. 34, no. 7, pp. 1021–1038, 2015.
T. Iberall, “Opposition space as a structuring concept for the analysis of skilled hand movements,” [*Generation and modulation of action patterns*]{}, vol. 15, pp. 158–173, 1986.
M. Arbib, T. Iberall, and D. Lyons, “Coordinated control programs for movements of the hand,” [*Experimental brain research*]{}, pp. 111–129, 1985.
M. R. Cutkosky, “On grasp choice, grasp models, and the design of hands for manufacturing tasks,” [*IEEE Transactions on robotics and automation*]{}, vol. 5, no. 3, pp. 269–279, 1989.
T. Iberall, “Human prehension and dexterous robot hands,” [*The International Journal of Robotics Research*]{}, vol. 16, no. 3, pp. 285–299, 1997.
T. Iberall, C. Torras, and C. MacKenzie, “Parameterizing prehension: A mathematical model of opposition space,” in [*Proceedings of the third COGNITIVA symposium on At the crossroads of artificial intelligence, cognitive science, and neuroscience*]{}, pp. 635–642, North-Holland Publishing Co., 1991.
C. Bersch, B. Pitzer, and S. Kammel, “Bimanual robotic cloth manipulation for laundry folding,” in [*IEEE/RSJ Int. Conf. on Intelligent Robots and Systems*]{}, pp. 1413–1419, 2011.
S. Miller, J. Van Den Berg, M. Fritz, T. Darrell, K. Goldberg, and P. Abbeel, “A geometric approach to robotic laundry folding,” [*The International Journal of Robotics Research*]{}, vol. 31, no. 2, pp. 249–267, 2012.
M. Cusumano-Towner, A. Singh, S. Miller, J. F. O’Brien, and P. Abbeel, “Bringing clothing into desired configurations with limited perception,” in [*IEEE Int. Conf. on Robotics and Automation*]{}, pp. 3893–3900, 2011.
S. Miller, M. Fritz, T. Darrell, and P. Abbeel, “Parametrized shape models for clothing,” in [*IEEE Int. Conf. on Robotics and Automation*]{}, pp. 4861–4868, 2011.
J. Van Den Berg, S. Miller, K. Goldberg, and P. Abbeel, “Gravity-based robotic cloth folding,” in [*Algorithmic Foundations of Robotics IX*]{}, pp. 409–424, Springer, 2010.
K. Lakshmanan, A. Sachdev, Z. Xie, D. Berenson, K. Goldberg, and P. Abbeel, “A constraint-aware motion planning algorithm for robotic folding of clothes,” in [*Experimental Robotics*]{}, pp. 547–562, Springer, 2013.
A. X. Lee, H. Lu, A. Gupta, S. Levine, and P. Abbeel, “Learning force-based manipulation of deformable objects from multiple demonstrations,” in [ *IEEE Int. Conf. on Robotics and Automation*]{}, pp. 177–184, 2015.
Y. Koishihara, S. Arnold, K. Yamazaki, and T. Matsubara, “Hanging work of t-shirt in consideration of deformability and stretchability,” in [ *Information and Automation (ICIA), 2017 IEEE International Conference on*]{}, pp. 130–135, IEEE, 2017.
Y. Li, D. Xu, Y. Yue, Y. Wang, S.-F. Chang, E. Grinspun, and P. K. Allen, “Regrasping and unfolding of garments using predictive thin shell modeling,” in [*Robotics and Automation (ICRA), 2015 IEEE International Conference on*]{}, pp. 1382–1388, IEEE, 2015.
L. Sun, G. Aragon-Camarasa, S. Rogers, and J. P. Siebert, “Accurate garment surface analysis using an active stereo robot head with application to dual-arm flattening,” in [*IEEE Int. Conf. on Robotics and Automation*]{}, pp. 185–192, 2015.
K. S. M. Sahari, H. Seki, Y. Kamiya, and M. Hikizu, “Clothes manipulation by robot grippers with roller fingertips,” [*Advanced Robotics*]{}, vol. 24, no. 1-2, pp. 139–158, 2010.
K. Salleh, H. Seki, Y. Kamiya, and M. Hikizu, “Inchworm robot grippers for clothes manipulation,” [*Artificial Life and Robotics*]{}, vol. 12, no. 1-2, pp. 142–147, 2008.
M. Shibata, T. Ota, and S. Hirai, “Wiping motion for deformable object handling,” in [*IEEE Int. Conf. on Robotics and Automation*]{}, pp. 134–139, 2009.
P. N. Koustoumpardis, K. X. Nastos, and N. A. Aspragathos, “A 3-finger robotic gripper for grasping fabrics based on cams-followers mechanism,” in [ *Advances in Service and Industrial Robotics*]{}, pp. 612–620, Springer International Publishing, 2017.
I. M. Bullock, R. R. Ma, and A. M. Dollar, “A hand-centric classification of human and robot dexterous manipulation,” [*IEEE transactions on Haptics*]{}, vol. 6, no. 2, pp. 129–144, 2012.
M. P. Bell, [*Flexible object manipulation*]{}. PhD thesis, Dartmouth College, 2010.
S. Donaire, J. Borràs, G. Alenyà, and C. Torras, “A versatile gripper for cloth manipulation (submitted),” 2020.
S. Donaire C[ó]{}nsul, “M[à]{} rob[ò]{}tica multifuncional per a tasques dom[è]{}stiques,” [B.S.]{} thesis, Universitat Polit[è]{}cnica de Catalunya, 2019.
A. J. Spiers, B. Calli, and A. M. Dollar, “Variable-friction finger surfaces to enable within-hand manipulation via gripping and sliding,” [*IEEE Robotics and Automation Letters*]{}, vol. 3, no. 4, pp. 4116–4123, 2018.
E. E. Aksoy, A. Abramov, J. D[ö]{}rr, K. Ning, B. Dellen, and F. W[ö]{}rg[ö]{}tter, “Learning the semantics of object–action relations by observation,” [*The International Journal of Robotics Research*]{}, vol. 30, no. 10, pp. 1229–1249, 2011.
M. J. Aein, E. E. Aksoy, and F. W[ö]{}rg[ö]{}tter, “Library of actions: Implementing a generic robot execution framework by using manipulation action semantics,” [*The International Journal of Robotics Research*]{}, p. 0278364919850295, 2018.
D. Kuli[ć]{}, C. Ott, D. Lee, J. Ishikawa, and Y. Nakamura, “Incremental learning of full body motion primitives and their sequencing through human motion observation,” [*The International Journal of Robotics Research*]{}, vol. 31, no. 3, pp. 330–345, 2012.
F. W[ö]{}rg[ö]{}tter, E. E. Aksoy, N. Kr[ü]{}ger, J. Piater, A. Ude, and M. Tamosiunaite, “A simple ontology of manipulation actions based on hand-object relations,” [*IEEE Transactions on Autonomous Mental Development*]{}, vol. 5, no. 2, pp. 117–134, 2013.
[Júlia Borràs]{} is a Mathematician and Computer Scientist since 2004 and 2006, respectively, and she obtained her European Ph.D. degree in 2011 working on kinematics and reconfiguration designs for the Stewart-Gough parallel platform. She worked abroad for 6 years as a postdoc, two years at Prof. Aaron Dollar GrabLab group from Yale University and four years at the Karlsruhe Institute of Technology (KIT) with prof. Tamim Asfour H2T group. She has worked on parallel robots, underactuated robot hands, grasping, dextrous manipulation, whole-body motion analysis, humanoid robot locomotion, novel designs for robotic hands and grippers, and robotic cloth manipulation. In 2018 she was awarded a Ramon y Cajal scholarship, one of the most prestigious senior postdoctoral scholarships in Spain. Recently, she has earned a tenured position at the Spanish Scientific Research Council (CSIC).
[Guillem Alenyà]{} holds a tenured position at Spanish Scientific Research Council CSIC. He received a PhD degree (Doctor Europeus) from UPC in 2007 with a work on mobile robot navigation using active contours while he was supported by a EU-FP6 Marie-Curie scholarship. He has been visitor at KIT-Karlsruhe, INRIA-Grenoble and BRL-Bristol. He has coordinated numerous scientific and technological transfer projects involving image understanding, task learning and plan execution tasks. He has published more than 100 articles in relevant venues in the areas of robotics, computer vision, and artificial intelligence.
[Carme Torras]{} is Research Professor at the Institut de Robòtica i Informàtica Industrial (CSIC-UPC) in Barcelona, where she leads a research group on assistive and collaborative robotics. She received M.Sc. degrees in Mathematics and Computer Science from the University of Barcelona and the University of Massachusetts, respectively, and a Ph.D. degree in Computer Science from the Technical University of Catalonia (UPC). Prof. Torras has published six research books and more than three hundred papers in robotics, machine learning, geometric reasoning, and neurocomputing. She has supervised 19 PhD theses and led 16 European projects, the latest being her ERC Advanced Grant project CLOTHILDE – Cloth manipulation learning from demonstrations. Prof. Torras is IEEE and EurAI Fellow, member of Academia Europaea and the Royal Academy of Sciences and Arts of Barcelona. She has served as Senior Editor of the IEEE Transactions on Robotics, and Associate Vice-President for Publications of the IEEE Robotics and Automation Society. Convinced that science fiction can help promote ethics in AI and robotics, one of her novels - winner of the Pedrolo and Ictineu awards - has been translated into English with the title The Vestigial Heart (MIT Press, 2018) and published together with online materials to teach a course on “Ethics in Social Robotics and AI”.
[^1]: The research leading to these results receives funding from the European Research Council (ERC) from the European Union Horizon 2020 Programme under grant agreement no. 741930 (CLOTHILDE: CLOTH manIpulation Learning from DEmonstrations) and is also supported by the Spanish State Research Agency through the María de Maeztu Seal of Excellence to IRI (MDM-2016-0656) and the “Ramon y Cajal” Fellowship RYC-2017-22703.
[^2]: The authors are with Institut de Robòtica i Informàtica Industrial, CSIC-UPC, Llorens i Artigas 4-6, 08028 Barcelona, Spain. [{jborras, galenya, torras}@iri.upc.edu]{}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'This paper proposes key instance selection based on video saliency covering objectness and dynamics for unsupervised video object segmentation (UVOS). Our method takes frames sequentially and extracts object proposals with corresponding masks for each frame. We link objects according to their similarity until the $M$-th frame and then assign them unique IDs (i.e., instances). Similarity measure takes into account multiple properties such as ReID descriptor, expected trajectory, and semantic co-segmentation result. After $M$-th frame, we select $K$ IDs based on video saliency and frequency of appearance; then only these key IDs are tracked through the remaining frames. Thanks to these technical contributions, our results are ranked third on the leaderboard of UVOS DAVIS challenge.'
author:
- |
Donghyeon Cho\
SK T-Brain\
- |
Sungeun Hong\
SK T-Brain\
- |
Sungil Kang\
SK T-Brain\
- |
Jiwon Kim\
SK T-Brain\
bibliography:
- 'egbib.bib'
title: Key Instance Selection for Unsupervised Video Object Segmentation
---
Introduction
============
Given a mask in the first frame, semi-supervised video object segmentation (SVOS) is a task of generating masks in the subsequent frames. After the first SVOS challenge [@Perazzi_cvpr2016], the SVOS has been gradually getting attention. Also, datasets are continuously updated [@DAVIS_2017] or newly constructed [@Xu_youtubeVOS]. Recently, interactive video object segmentation (IVOS) [@DAVIS_2018] and unsupervised video object segmentation (UVOS) [@DAVIS_2019] were introduced as new challenges. This paper tackles UVOS which does not require any human supervision.
The basic approach of the UVOS is to estimate a mask of the first frame, then use it to apply conventional SVOS methods [@Luiten_cvprw2018; @Luiten_2018accv]. However, not only is it greatly influenced by the results of the first frame, it is not guaranteed that there are all targets in the first frame. To resolve these problems, we can continuously assign a new ID (i.e., instance) to an object that satisfies certain criteria. However, this increases not only the number of IDs indiscriminately but also time complexity. Therefore, it is recommended to fix the appropriate $K$, the maximum number of IDs. In this paper, we propose a method to select $K$ IDs by exploring video saliency of several frames at the beginning of the video. Compared to random $K$ instance selection, our method can capture main objects well, even in case of small $K$ as shown in [Fig. \[fig:topk\]]{}-(a).
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -- --
![Segmentation results with respect to $K$ instances on a ‘scooter-black’ video. Top: random instance selection. Bottom: our key instance selection.[]{data-label="fig:topk"}](figures/topk/rnd/5/00010.png "fig:"){width="0.32\linewidth"} ![Segmentation results with respect to $K$ instances on a ‘scooter-black’ video. Top: random instance selection. Bottom: our key instance selection.[]{data-label="fig:topk"}](figures/topk/rnd/10/00010.png "fig:"){width="0.32\linewidth"} ![Segmentation results with respect to $K$ instances on a ‘scooter-black’ video. Top: random instance selection. Bottom: our key instance selection.[]{data-label="fig:topk"}](figures/topk/rnd/15/00010.png "fig:"){width="0.32\linewidth"}
![Segmentation results with respect to $K$ instances on a ‘scooter-black’ video. Top: random instance selection. Bottom: our key instance selection.[]{data-label="fig:topk"}](figures/topk/our/5/00010.png "fig:"){width="0.32\linewidth"} ![Segmentation results with respect to $K$ instances on a ‘scooter-black’ video. Top: random instance selection. Bottom: our key instance selection.[]{data-label="fig:topk"}](figures/topk/our/10/00010.png "fig:"){width="0.32\linewidth"} ![Segmentation results with respect to $K$ instances on a ‘scooter-black’ video. Top: random instance selection. Bottom: our key instance selection.[]{data-label="fig:topk"}](figures/topk/our/15/00010.png "fig:"){width="0.32\linewidth"}
[(a) $K$=5 ]{} [(b) $K$=10 ]{} [(c) $K$=15]{}
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -- --
The main contributions of this paper are as follows. Compared to adding new IDs continuously until the end of the video, our key instance selection approach significantly reduces time as well as memory complexity. Also, our model can capture regions of importance better than random instance selection. When we measure the similarity between assigned IDs and extracted proposals, we use all set of positive ReID descriptors for each ID. Unlike conventional methods that use optical flow [@Ilg_cvpr2017] to handle large appearance changes in a frame sequence, we use semantic co-segmentation [@oh_2018cvpr; @Oh_cvprw2018]. Finally, we automatically search hyperparameters through Meta AI system developed by T-Brain under scalable environments (e.g., 144 GPUs).
0
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -- -- -- -- --
![(a) An Input image at $t=1$. (b) A ground truth mask at $t=1$. (c) An estimated mask at $t=1$. (d) An estimated mask at $t=90$. Note that there is only one object needed to track but multiple objects are captured by instance segmentation. In the future frame, an estimated target mask is damaged by a detected non-target object.[]{data-label="fig:teaser"}](figures/teasure0.jpg "fig:"){width="0.44\linewidth"} ![(a) An Input image at $t=1$. (b) A ground truth mask at $t=1$. (c) An estimated mask at $t=1$. (d) An estimated mask at $t=90$. Note that there is only one object needed to track but multiple objects are captured by instance segmentation. In the future frame, an estimated target mask is damaged by a detected non-target object.[]{data-label="fig:teaser"}](figures/teasure1.png "fig:"){width="0.44\linewidth"}
[(a) ]{} [(b) ]{}
![(a) An Input image at $t=1$. (b) A ground truth mask at $t=1$. (c) An estimated mask at $t=1$. (d) An estimated mask at $t=90$. Note that there is only one object needed to track but multiple objects are captured by instance segmentation. In the future frame, an estimated target mask is damaged by a detected non-target object.[]{data-label="fig:teaser"}](figures/teasure2.png "fig:"){width="0.44\linewidth"} ![(a) An Input image at $t=1$. (b) A ground truth mask at $t=1$. (c) An estimated mask at $t=1$. (d) An estimated mask at $t=90$. Note that there is only one object needed to track but multiple objects are captured by instance segmentation. In the future frame, an estimated target mask is damaged by a detected non-target object.[]{data-label="fig:teaser"}](figures/00089.png "fig:"){width="0.44\linewidth"}
[(c) ]{} [(d) ]{}
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -- -- -- -- --
Method {#sec:pipeline}
======
Our objective is to assign IDs to the proposals without additional inputs such as a mask of the first frame. For this, we use an object pool that manages assigning, adding, and deleting IDs. [Fig. \[fig:overview\]]{} illustrates the overview of the proposed method.
As a first step, we perform instance segmentation on the current frame and propagate the masks obtained in the previous frame. We then extract features and assign IDs to candidates that satisfy the criteria of the online tracker linked with object pool. Here, if the candidate matches the existing ID in the object pool, we assign the matched ID to the candidate, and the online tracker is updated; otherwise, we assign a new ID to the candidate, then this new ID is added to the object pool. Adding a new ID is performed only up to the $M$-th frame. At the $M$-th frame, $K$ instances among the accumulated IDs are selected and finally, those selected IDs are tracked in the remaining frames.
![Overview of the proposed method.[]{data-label="fig:overview"}](figures/overall_flow "fig:"){width="0.99\linewidth"}\
#### Candidate Generation:
Given a frame at time $I^{t}$ ($t \in T$), we perform instance segmentation to get bounding box and mask by using Mask R-CNN [@He_iccv2017] and DeepLab [@Chen_eccv2018]. Motion blur or occlusion, which often occurs in videos, can result in poor segmentation results for certain frames. To address this issue, we use masks propagated from the previous frame as mask candidates. Concretely, we utilize RGMP [@oh_2018cvpr; @Oh_cvprw2018] in our experiments.
#### ID Assignment:
To assign an ID to each candidate, we compute specific scores by comparing the candidates and the registered IDs in the object pool. The first score is $S_{iou}(l,n)$ which is IOU between a mask from ID ($l \in L$) and a mask of the candidate ($n \in N$). Here, $L$ refers to the number of IDs registered in the object pool and $N$ is the number of candidates. We omit notation time $t$ for simplicity. The second score is $S_{traj}(l,n)$ that measures how far the candidate is from the predicted bounding box of ID as $$\begin{aligned}
S_{traj}(l,n) =1 - \min{(\frac{dot(\vec{b}_{l},\vec{b}_{n})}{\alpha_{traj}}, 1)},
\label{eq:track}
\end{aligned}$$ where $\alpha_{traj}$ is a normalization factor. Also, $\vec{b}_{l}$ and $\vec{b}_{n}$ are vectors from bounding boxes of ID and candidate, respectively. The third score $S_{reid}(l,n)$ considers a distance of ReID descriptors [@Osep_2017] between ID $l$ and candidate $n$. Unlike [@Luiten_cvprw2018; @Luiten_2018accv], our method use the nearest ReID descriptor among the all set of positive ReID descriptors of ID $l$ as $$\begin{aligned}
S_{reid}(l,n) = 1 - \min{(\frac{\min_{j} ||d_{l}(j) - r_{n}||}{\alpha_{reid}}, 1)},
\label{eq:reid}
\end{aligned}$$ where $r_{n}$ is a ReID descriptor for candidate $n$ and $d_{l}(\cdot)$ are positive ReID descriptors for ID $l$. The last score is a relative ReID score $S_{rel}(l,n)$: $$\begin{aligned}
S_{rel}(l,n) = \frac{S_{reid}(l,n)}{\max_{l}S_{reid}(l,n)}.
\label{eq:relative}
\end{aligned}$$ Total score is weighted summation of above four scores: $$\begin{aligned}
\begin{split}
S_{total}(l,n) = w_{iou} \cdot S_{iou}(l,n) + w_{traj} \cdot S_{traj}(l,n) &\\ + w_{reid} \cdot S_{reid}(l,n) + w_{rel} \cdot S_{rel}(l,n),
\end{split}
\label{eq:total}
\end{aligned}$$ where $w_{iou}$, $w_{traj}$, $w_{reid}$, and $w_{rel}$ are weight factors of each term. Finally, we assign ID $l$ to the candidate object as follows: $$\begin{aligned}
\hat{n}=
\begin{cases}
{\mathop{\rm argmax}\limits}_{n} S_{total}(l,n),& \text{if } \geq \tau_{c}\\
\mathrm{None}, & \text{otherwise}
\end{cases}
,
\label{eq:assign}
\end{aligned}$$ where $\tau_{c=\left \{ 1,2 \right \}}$ is a threshold value for cutting off object with low confidence. Before selecting the $K$ instances ($t<=M$) as in [Sec. \[sec:K\_selection\]]{}, $c$ is 1, and after that $c$ is 2. If one of candidates is assigned to ID $l$, then $d_{\hat{n}}$ is added to pool of positive ReID descriptors for ID $l$. Also $\vec{b}_{l}$ is updated by using $\vec{b}_{\hat{n}}$.
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -- --
![Video saliency results. Note that a motorcycle with motion is only focused among a lot of objects in a scene.[]{data-label="fig:video_sal"}](figures/sal/img1.jpg "fig:"){width="0.32\linewidth"} ![Video saliency results. Note that a motorcycle with motion is only focused among a lot of objects in a scene.[]{data-label="fig:video_sal"}](figures/sal/img2.jpg "fig:"){width="0.32\linewidth"} ![Video saliency results. Note that a motorcycle with motion is only focused among a lot of objects in a scene.[]{data-label="fig:video_sal"}](figures/sal/img3.jpg "fig:"){width="0.32\linewidth"}
![Video saliency results. Note that a motorcycle with motion is only focused among a lot of objects in a scene.[]{data-label="fig:video_sal"}](figures/sal/sal1.png "fig:"){width="0.32\linewidth"} ![Video saliency results. Note that a motorcycle with motion is only focused among a lot of objects in a scene.[]{data-label="fig:video_sal"}](figures/sal/sal2.png "fig:"){width="0.32\linewidth"} ![Video saliency results. Note that a motorcycle with motion is only focused among a lot of objects in a scene.[]{data-label="fig:video_sal"}](figures/sal/sal3.png "fig:"){width="0.32\linewidth"}
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -- --
#### Key Instances Selection: {#sec:K_selection}
Basically, the pipeline mentioned above is iterated at each frame. There is ID addition process at the beginning of the video. Especially, in the first frame, object candidates with high confidence are added as new IDs. From the second frame and $M$-th frame, new IDs are added when object candidates have high objectness score and are not overlapped with objects assigned to existing IDs. After $M$-th frame, we select at most $K$ IDs. As selection criteria, we use weighted summation of video saliency score [@Wang_tip2018] and frequency of each ID as $$\begin{aligned}
S_{sel}(l) = w_{sal} \cdot S_{sal}(l) + w_{freq} \cdot S_{freq}(l),
\label{eq:selection}
\end{aligned}$$ where $w_{sal}$ and $w_{freq}$ are weight factors of each term. $S_{sal}(l)$ and $S_{freq}(l)$ are computed over frames. [Fig. \[fig:video\_sal\]]{} shows effectiveness of video saliency based approach. The frequency of each ID means the number of each ID’s appearance up to the $M$-th frame. Because there are several IDs that are not connected to any proposal at a certain time by $\tau_{1}$, $S_{freq}(l)$ can vary by ID.
#### Hyperparameter Search:
We perform hyperparameter search for $w_{iou}$, $w_{traj}$, $w_{reid}$, $w_{rel}$, $\tau_{1}$, $\tau_{2}$, $w_{sal}$ and $w_{freq}$ on DAVIS validation dataset. Global mean ($\mathcal{J}$$\&$$\mathcal{F}$) of results is $0.599$ and searched weighting factors are $0.12, 0.575, 0.3, 0.0065, 0.55, 0.35, 0.5$ and $1.0$, respectively.
Experiments
===========
We evaluate the proposed method on UVOS DAVIS challenge dataset [@DAVIS_2019], which contains 30 videos without a mask from the first frame in each video. We directly submit our results to the CodaLab site [@CodaLab] to get segmentation results. Evaluation metrics are Region Jaccard ($\mathcal{J}$) and Boundary $F$ measure ($\mathcal{F}$) for each instance. As shown in [Table. \[tab:results\_challenge\]]{}, our method achieves the third rank with respect to Global Mean as well as all the other specific metrics.
[Fig. \[fig:Result\]]{} shows qualitative results on DAVIS validation set. In the first row of [Fig. \[fig:Result\]]{}, our method well captures *blackswan* over time. The proposed method also shows robust video segmentation results even in the relatively dynamic *Parkour* example. In addition, our method faithfully works on multiple object segmentation (from the third row to the last low in [Fig. \[fig:Result\]]{}). Note that changes in scale (*lab-coat*) and appearance (*mbike-trick*) are handled appropriately by our method.
The proposed key instance selection method is quantitatively compared with random instance selection according to the $K$ as shown in [Table. \[tab:top\_k\]]{}. Since the proposed scheme focuses on semantically meaningful regions, it shows promising results even at a small $K$.
-------------------------------------------------------------- -------------------------------------------------------------- -------------------------------------------------------------- -------------------------------------------------------------- -------------------------------------------------------------- -- --
{width="0.192\linewidth"} {width="0.192\linewidth"} {width="0.192\linewidth"} {width="0.192\linewidth"} {width="0.192\linewidth"}
{width="0.192\linewidth"} {width="0.192\linewidth"} {width="0.192\linewidth"} {width="0.192\linewidth"} {width="0.192\linewidth"}
{width="0.192\linewidth"} {width="0.192\linewidth"} {width="0.192\linewidth"} {width="0.192\linewidth"} {width="0.192\linewidth"}
{width="0.192\linewidth"} {width="0.192\linewidth"} {width="0.192\linewidth"} {width="0.192\linewidth"} {width="0.192\linewidth"}
{width="0.192\linewidth"} {width="0.192\linewidth"} {width="0.192\linewidth"} {width="0.192\linewidth"} {width="0.192\linewidth"}
{width="0.192\linewidth"} {width="0.192\linewidth"} {width="0.192\linewidth"} {width="0.192\linewidth"} {width="0.192\linewidth"}
-------------------------------------------------------------- -------------------------------------------------------------- -------------------------------------------------------------- -------------------------------------------------------------- -------------------------------------------------------------- -- --
Conclusion
==========
We have presented key instance selection for unsupervised video object segmentation (UVOS). Our method allows maximum $K$ instances to be tracked by considering video saliency and frequency of appearance. Focusing on objects that are in the spotlight enables to reduce time complexity. In addition, objects in a frame sequence are linked by specific scores from ReID descriptor, trajectories, and co-segmentation result. Finally, we perform hyperparameter search to find out the optimal hyperparameters in our model by Meta AI system. Our method showed competitive results in UVOS DAVIS challenge.
Acknowledgements
================
We thank Youngsoon Lee for help to use a Meta AI system which is effective in searching hyperparameters.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The optical conductivity of graphene strained uniaxially is studied within the Kubo-Greenwood formalism. Focusing on inter-band absorption, we analyze and quantify the breakdown of universal transparency in the visible region of the spectrum, and analytically characterize the transparency as a function of strain and polarization. Measuring transmittance as a function of incident polarization directly reflects the magnitude and direction of strain. Moreover, direction-dependent selection rules permit identification of the lattice orientation by monitoring the van-Hove transitions. These photoelastic effects in graphene can be explored towards atomically thin, broadband optical elements.'
author:
- 'Vitor M. Pereira'
- 'R. M. Ribeiro'
- 'N. M. R. Peres'
- 'A. H. Castro Neto [^1]'
bibliography:
- 'graphene\_optical.bib'
title: Optical Properties of Strained Graphene
---
Introduction
============
Transparent flexible electronics is currently a much sought technology, with applications that can range from foldable displays and electronic paper, to transparent solar cells. Graphene, on account of its ultimate thickness, large transparency [@Nair:2008], high mechanical resilience under strong stress/bending cycling, and excellent electronic mobility [@Bolotin:2008], has been promptly ranked among the best placed materials to achieve those technologies [@Ruoff:2009; @Bae:2010]. Such goals require a thorough understanding of the interplay between the different aspects that will unavoidably be present in such devices, namely how sensitive the dielectric and optical properties of graphene are to gating and straining. At the same time, broadband optical elements that can be scaled down to the nanoscale, and easily integrated into photonic/photoelectronic circuits, are equally appealing scenarios in current nanotechnology.
The optical absorption response of graphene has been recently given thorough attention on both the experimental [@Nair:2008; @Basov:2008; @Heinz:2008; @Kuzmenko:2008; @Heinz:2010] and theoretical fronts [@Ando:2002; @Peres:2006; @Falkovsky:2007; @Mischenko:2009; @Sheehy:2009; @Gusynin:2009; @Peres:2010; @RMP:2009; @Peres-RMP:2010]. One of the distinguishing features of undoped (or lightly doped) graphene arises from the constancy of its transparency, ${\ensuremath{\mathcal{T}}}(\omega)$, which is controlled by electron-hole excitation processes, and universal: ${\ensuremath{\mathcal{T}}}(\omega) \approx 1 - \pi \alpha$ ($\alpha\simeq 1/137$ being the fine structure constant) [@Nair:2008]. This universality is a consequence of the particle hole symmetry of graphene’s spectrum, combined with the cancellation of the frequency dependencies of the matrix element and vanishing density of states. It holds throughout a broad spectral range comprising the frequencies between the Fermi energy, $\mu$, and the vicinity of the van Hove singularity (VHS) at the $M$ point in the Brillouin zone (BZ). In undoped graphene this covers a band spanning the visible region, down to the far IR.
Here we analyze and quantify how the optical absorption associated with direct optical transitions is affected by strain. The strain-induced anisotropy leaves a clear signature in the optical response of the system, modulating its transmittance, reflectance and absorption, while simultaneously rotating the polarization of incoming light. Tailoring of graphene systems on the basis of such properties extends the concept of strain engineering in graphene from electronic structure and transport [@Pereira:2009a; @Guinea:2010] to the optical domain as well. Such effects are described next. Related work on effects of strain in graphene’s optical response has been reported in Refs. [@Pellegrino:2009; @Pellegrino:2010; @Sinner:2010].
![ Lattice orientation in direct and reciprocal space, overlaid with several quantities used in the text. The bottom illustration shows the relative orientation of the lattice with respect to a general tension direction $\theta$. []{data-label="fig:Fig-1"}](Figs/Fig-1){width="45.00000%"}
Strain-Induced Anisotropy
=========================
We concentrate on uniform uniaxial deformations of free graphene, which provide the highest degree of anisotropy for given amount of strain, ${{\ensuremath{\varepsilon}}}$. Within a tight-binding approximation for the $\pi$-band subsystem, strain impacts electronic motion via the modification of the Slater-Koster parameters $t_i \equiv t({\ensuremath{\bm{\delta}}}_i)$ in the Hamiltonian $$H = \sum_{{{\ensuremath{\bm{R}}}},{\ensuremath{\bm{\delta}}}} t({\ensuremath{\bm{\delta}}}) \,
a^\dagger_{{{\ensuremath{\bm{R}}}}}b^{\phantom{\dagger}}_{{{\ensuremath{\bm{R}}}}+{\ensuremath{\bm{\delta}}}}
+ \text{H. c.}
\,.
\label{eq:Hamiltonian}$$ Here ${{\ensuremath{\bm{R}}}}$ denotes a position in the Bravais lattice; ${\ensuremath{\bm{\delta}}}_{1,2,3}$ connects site ${{\ensuremath{\bm{R}}}}$ to its neighbors; $a({{\ensuremath{\bm{R}}}})$ and $b({{\ensuremath{\bm{R}}}})$ are the field operators in sublattices A and B. In what follows, we characterize the interplay between strain and electronic structure in the same framework described in detail in [ref. ]{} . In summary, this entails the assumption that hopping amplitudes vary with distance as $t(r) \simeq t e^{-\lambda(r-a)}$, with $t\approx 2.7$eV, $a = 1.42$Å, $\lambda a \approx 3-4$ [@Pereira:2009b; @Pellegrino:2010], and we disregard bond bending effects, since they are not significant in this effective description [@Ribeiro:2009]. Bond deformations are uniform and given to linear order in terms of the strain tensor ${{\ensuremath{\varepsilon}}}_{ij}$: $\bm{{\ensuremath{\bm{\delta}}}} = (\bm{1}+\bm{{{\ensuremath{\varepsilon}}}})\cdot\bm{{\ensuremath{\bm{\delta}}}}^0$. This approximation was shown to describe with good accuracy both the threshold deformation for the Lifshitz and metal-insulator transition at large deformations [@Pereira:2009b; @Ni:2009; @Choi:2010], and the behavior of $t_i(\delta)$ when compared to *ab-initio* calculations [@Ribeiro:2009]. Since the hexagonal lattice is elastically isotropic, ${{\ensuremath{\varepsilon}}}_{ij}$ can be fully parametrized by the amount of uniaxial strain, ${{\ensuremath{\varepsilon}}}$, its direction $\theta$ with respect to a zig-zag ([$\mathcal{ZZ}$]{}) direction, and Poisson’s ratio, $\nu\approx
0.16$ [@Pereira:2009b]. Our Cartesian directions are such that a [$\mathcal{ZZ}$]{} edge coincides with the $x$ axis ([fig. \[fig:Fig-1\]]{}). Tension along $\theta=0$ \[[$\mathcal{ZZ}$]{}\] and $\theta=\pi/2$ \[armchair ([$\mathcal{AA}$]{})\] defines representative directions that will be recalled frequently. We shall disregard the strain-induced modification of the reciprocal lattice, as it is not relevant for the optical absorption. Since here we are not interested in large deformations, we further approximate $t_i \equiv t(\delta_i) \approx t - t \lambda (\delta_i - a)$.
The low energy Hamiltonian appropriate for optical processes below the UV can be derived conventionally [@RMP:2009], by expanding around the *shifted* Dirac points: ${{\ensuremath{\bm{k}}}}=
{{\ensuremath{\bm{k}}}}_D + {{\ensuremath{\bm{q}}}}$ ($q\ll k_D$). For arbitrary $t_i$ the nonequivalent Dirac points lie at ${{\ensuremath{\bm{k}}}}_{D,\zeta} = \zeta
(A-B,\,A+B)$, with $ A=\frac{1}{\sqrt{3}a} \cos^{-1}\frac{t_1^2-t_2^2-t_3^2}{2t_2 t_3}, $ $ B=\frac{1}{3a} \cos^{-1}\frac{t_3^2-t_1^2-t_2^2}{2t_1 t_2} $, and $\zeta=\pm1$ is the valley index, labeling the two nonequivalent Dirac points. In their vicinity, the electronic dispersion is given by $$\begin{gathered}
E^2_{{\ensuremath{\bm{q}}}}\simeq
\frac{9}{4} q_y^2a^2 t_2^2
+\frac{3\sqrt{3}}{2}q_xq_ya^2 (t_3^2-t_1^2) \\
+\frac{3}{4}q_x^2a^2 (2t_1^2-t_2^2+2t_3^2)
+ {\ensuremath{\mathcal{O}}}(q^3)
\label{eq:Eq-general}
.\end{gathered}$$ The Fermi surface is consequently an ellipse, with principal axes that will be rotated with respect to ${\ensuremath{\mathcal{O}}}x y$ in the general situation. [eq. ]{} can be cast compactly as $E^2_{{\ensuremath{\bm{q}}}}=
\hbar^2v_F^2 \times {{\ensuremath{\bm{q}}}}.\bm{\eta}^2.{{\ensuremath{\bm{q}}}}$, where $\hbar v_F
\equiv 3ta / 2$. Diagonalization of $\bm{\eta}^2$ yields the principal velocities $v_\pm^2=v_F^2\eta^2_\pm$, with $$\begin{gathered}
\eta^2_\pm =
\frac{t_1^2+t_2^2+t_3^2}{3t^2}
\\ \pm \frac{2}{3t^2}
\sqrt{t_1^4+t_2^4+t_3^4-t_1^2t_2^2-t_2^2t_3^2-t_1^2t_3^2}
,\end{gathered}$$ and the principal directions: $$\tan\varphi_\pm =
\frac{t_3^2-t_1^2}{\sqrt{3}(\eta_\pm^2t^2-t_2^2)}
.$$ These directions define the slow/longitudinal ($-,l$) and fast/transverse ($+,t$) directions of strained graphene, insofar as the direction-dependent Fermi velocity in the elliptical Fermi surface is smallest along the one, and largest along the other. In the principal coordinate system, the effective Hamiltonian of valley $\zeta=\pm$ reads $$H_{\zeta} \simeq
\zeta v_F \tau_1\eta_l p_l + v_F \tau_2\eta_t p_t
\label{eq:H-eff-strain}
,$$ where $\tau_{1,2}$ are Pauli matrices acting on the $(A,B)$ sublattice space. Coupling to a light wave described by the physical vector potential $\bm{A}(t)=\bm{A_0} \exp(-i\omega t)$ is achieved by the minimal substitution ${{\ensuremath{\bm{p}}}}\to{{\ensuremath{\bm{p}}}}+e\bm{A}$ in [eq. ]{}. The frequency-dependent conductivity is extracted from the linear response to $\bm{A}_0$.
For future reference, the anisotropy parameters will be expressed directly in terms of the longitudinal deformation ${{\ensuremath{\varepsilon}}}$, to first order. For general tension along $\theta$ with respect to ${\ensuremath{\mathcal{O}}}x$ we have
\[eq:aniso-explicit\] $$\begin{gathered}
\eta_{t} \simeq 1 + a\lambda\nu{{\ensuremath{\varepsilon}}},\quad
\eta_{l} \simeq 1 - a\lambda{{\ensuremath{\varepsilon}}}\label{eq:aniso-explicit-etas}
,\\
\delta {{\ensuremath{\bm{k}}}}_{D,\zeta} \simeq
\zeta \lambda{{\ensuremath{\varepsilon}}}\frac{1+\nu}{2}(\cos2\theta,-\sin2\theta)
\label{eq:aniso-explicit-Dirac}
,\\
\tan\varphi_{l} \simeq
\Bigl[1 - a\lambda{{\ensuremath{\varepsilon}}}\frac{1+\nu}{8} (1+2\cos4\theta)
\Bigr]\tan\theta
,\\
\tan\varphi_t \simeq
\Bigl[1 + a\lambda{{\ensuremath{\varepsilon}}}\frac{1+\sigma}{8} (1+2\cos 4\theta)
\Bigr]\tan\bigl(\theta+\frac{\pi}{2} \bigr)
.
\label{eq:aniso-vs-strain}\end{gathered}$$
As intuition dictates, the slow/longitudinal axis is coincident with the tension axis ($\varphi\approx\theta$). From the strain-induced correction to the density of states (DOS) in the vicinity of the Dirac points follows immediately: $\rho(E)\simeq \rho^\text{iso}(E) / (\eta_l \eta_y)$, where the isotropic DOS reads $\rho^\text{iso}(E)=2|E|/(\pi\hbar^2v_F^2)$. Notice that, even though the effective longitudinal (transverse) velocity increases (decreases), the net effect in the DOS is always a slope enhancement, because $\eta_l \eta_t \approx 1 - a\lambda(1-\nu){{\ensuremath{\varepsilon}}}<1$. This means that strain modifies the Fermi energy and/or electron density.
Conductivity of Strained Graphene
=================================
Kubo’s form of the frequency dependent conductivity reads $$\sigma_{\alpha\beta}(\omega)
= \frac{ig_sq^2}{A\omega} \sum_{{{\ensuremath{\bm{k}}}}\lambda\lambda'}
\frac{
v^{{{\ensuremath{\bm{k}}}}\lambda\lambda'}_\alpha
v^{{{\ensuremath{\bm{k}}}}\lambda'\lambda}_\beta
(n_{{{\ensuremath{\bm{k}}}}\lambda}-n_{{{\ensuremath{\bm{k}}}}\lambda'})
}{
\hbar\omega-({\epsilon}_{{{\ensuremath{\bm{k}}}},\lambda'} -{\epsilon}_{{{\ensuremath{\bm{k}}}},\lambda})+i0^+
}
\label{eq:sigma-Kubo}
,$$ where $g_s=2$ is the spin degeneracy, $A$ the total area, $\lambda,\lambda'=\pm$, and $v^{{{\ensuremath{\bm{k}}}}\lambda\lambda'}_\alpha \equiv
\langle\bm{k},\lambda\vert v_{\alpha}\vert\bm{k},\lambda'\rangle$ are the matrix elements of the velocity operator $v_\alpha=i\hbar^{-1}[H,x_\alpha]$ in the momentum eigenbasis. For the general tight-binding Hamiltonian , these matrix elements read explicitly $$\begin{gathered}
\langle{{\ensuremath{\bm{k}}}},\lambda\vert\bm{v}\vert{{\ensuremath{\bm{k}}}}',\lambda'\rangle =
-\frac{\lambda}{\hbar}\delta_{{{\ensuremath{\bm{k}}}},{{\ensuremath{\bm{k}}}}'}\delta_{\lambda\lambda'}
\sum_\alpha t_\alpha {\ensuremath{\bm{\delta}}}_\alpha \sin\left[{{\ensuremath{\bm{k}}}}.{\ensuremath{\bm{n}}}_\alpha
-\theta_{{\ensuremath{\bm{k}}}}\right]\\
-\frac{i\lambda}{\hbar}\delta_{{{\ensuremath{\bm{k}}}},{{\ensuremath{\bm{k}}}}'}\left(1-\delta_{
\lambda\lambda'}\right)\sum_\alpha t_\alpha{\ensuremath{\bm{\delta}}}_\alpha
\cos\left[{{\ensuremath{\bm{k}}}}.{\ensuremath{\bm{n}}}_{\alpha}-\theta_{{\ensuremath{\bm{k}}}}\right]
\label{eq:v-matrix}
,\end{gathered}$$ with $\theta_{{{\ensuremath{\bm{k}}}}} = \arg \bigl[\sum_i t_i \exp(i{{\ensuremath{\bm{k}}}}.{\ensuremath{\bm{n}}}_i)\bigr]$. Contributions with $\lambda=\lambda'$ are associated with *intra*-band transitions, and $\lambda\ne\lambda'$ with *inter*-band, direct transitions. In this form the conductivity can be directly calculated at the tight-binding level. In order to proceed fully analytically we work in the Dirac approximation , and consider a clean system at zero temperature, thus retaining only the inter-band contribution. In the principal coordinate system defined by the slow/longitudinal and fast/transverse axes the relevant matrix elements become:
\[eq:v-matrix-Dirac\] $$\begin{aligned}
|v^{{{\ensuremath{\bm{k}}}},\lambda,-\lambda}_l|^2 & \approx
v_F^2\eta_l^2\sin^2\theta_{{{\ensuremath{\bm{k}}}}_D}
,\\
|v^{{{\ensuremath{\bm{k}}}},\lambda,-\lambda}_t|^2 &\approx
v_F^2\,\eta_t^2\cos^2\theta_{{{\ensuremath{\bm{k}}}}_D}
,\\
v^{{{\ensuremath{\bm{k}}}},\lambda,-\lambda}_l v^{{{\ensuremath{\bm{k}}}},-\lambda,\lambda}_t &\approx
\zeta\, v_F^2\tfrac{\eta_l\eta_t}{2} \sin(2\theta_{{{\ensuremath{\bm{k}}}}_D})
.\end{aligned}$$
The anisotropy is explicitly encoded both in the parameters $\eta_{\pm}$, and in the phase $\theta_{{{\ensuremath{\bm{k}}}}_D}$.
From here and [eq. ]{} it is straightforward to obtain the frequency dependent conductivity $\sigma_{\alpha\beta} = \sigma_{\alpha\beta}^\prime + i
\sigma_{\alpha\beta}^{\prime\prime}$. Its real part reads $$\sigma_{ll}^\prime(\omega) \simeq
\frac{\eta_l}{\eta_t} \times \sigma_0 \times
\Bigl[
f\bigl(-\tfrac{\hbar\omega}{2}-\mu\bigr)
- f\bigl(\tfrac{\hbar\omega}{2}-\mu\bigr)
\Bigr]
\label{eq:Re-sigma}$$ for the longitudinal conductivity, and $\sigma_{tt}^\prime(\omega) = (\eta_t/\eta_l)^2\,\sigma_{ll}
^\prime(\omega)$ for the transverse component. The isotropic (and universal) value is $\sigma_0=e^2/(4\hbar)$, and $f(x)$ represents the Fermi-Dirac occupation function. The imaginary part, $\sigma''(\omega)$, reads $$\sigma_{ll}^{\prime\prime}(\omega) \simeq
\frac{\eta_l}{\eta_t} \times \frac{\sigma_0}{\pi}\times
\log\biggl|\frac{2|\mu|-\omega}{2|\mu|+\omega}\biggr|
\label{eq:Im-sigma}$$ when $T=0$, and $\sigma_{tt}^{\prime\prime}(\omega) =
(\eta_t/\eta_l)^2\,\sigma_{ll}^{\prime\prime}$. The off-diagonal components $\sigma_{lt}$ are zero, as one expects from symmetry and the absence of magnetic fields. This result shows that $\sigma^\prime_{ii}(\omega)$ is still constant within the region of validity of the Dirac approximation, and for $2\mu \lesssim \hbar\omega \ll T$, being only renormalized by the anisotropy factors $\eta_{\pm}$. The degree of anisotropy is controlled by $\eta_l/\eta_t$, which is a sensible result because $\sigma_{ii}(\omega)\propto |v^{{{\ensuremath{\bm{k}}}},\lambda,-\lambda}
_i|^2$, and the ratio reflects the quotient between the Fermi velocities along the principal strain directions. From Eqs. and , and recalling the expressions in [eq. ]{}, we can express the strain induced corrections to the full isotropic conductivity, $\sigma(\omega)=\sigma^\prime(\omega)+i\sigma^{\prime\prime}(\omega)$, in linear order in the deformation as $$\sigma_{ll,tt} (\omega)
\simeq
\sigma^\text{iso} (\omega) \times
\bigl(1 \mp 2 |\delta k_D| a \bigr)
\label{eq:sigma-strain}
,$$ where $\sigma^\text{iso} (\omega)$ represents the full frequency-dependent conductivity in the absence of strain. For example, tension along [$\mathcal{ZZ}$]{} ($\theta=0$) yields a decrease in $\sigma_{xx}$, and an increase of the same magnitude for $\sigma_{yy}$. Substitution of the material parameters applicable to free-standing graphene in and yields an anisotropy factor $(\sigma_t - \sigma_l)/\sigma^\text{iso} = 4 |\delta k_D| a \sim
8{{\ensuremath{\varepsilon}}}$. This is sufficiently marked to be detectable by conventional optical means, like absorption in the visible or IR, which we discuss below.
{width="90.00000%"}
Ab-Initio Optical Conductivity
==============================
Given the approximations used, it is legitimate to question the validity of the analytical corrections written in [eq. ]{}. To that end, we have extracted the optical conductivity of uniaxially strained graphene from first-principles as well. Density Functional Theory (DFT) calculations were performed with the *ab-initio* spin-density functional code <span style="font-variant:small-caps;">aimpro</span> [@Rayson:2008]. We used the GGA in the scheme of Perdew, Burke, and Ernzerhof [@Perdew:1996]. Core states were accounted for by dual-space separable pseudopotentials [@Hartwigsen:1998]. The valence states are expanded over a set of $s$, $p$, and $d$-like localized, atom-centered Gaussian functions. The BZ was sampled according to the scheme proposed by Monkhorst-Pack [@Monkhorst:1998]. We used a convergence-tested grid of $20\times 20\times \times1$ points for the self-consistent calculations. In equilibrium we obtained the optimized lattice parameter $a = 1.42$Å.
Uniaxial strain was applied with relaxation as described earlier in Ref. . For each strained configuration we extracted the dielectric function within the long-wavelength dipole approximation [@Fall:2008], and from it the optical conductivity. The results for $\sigma^\prime_{xx}(\omega)$ and $\sigma^\prime_{yy}(\omega)$ so obtained are shown in [fig. \[fig:Fig-2\]]{}, for representative uniaxial strains applied along [$\mathcal{ZZ}$]{} ($\theta=0$) and [$\mathcal{AA}$]{} ($\theta=\pi/2$).
The main panel of [fig. \[fig:Fig-2\]]{}(a) shows the renormalization of the real part of $\sigma_{xx}$, which corresponds to $\sigma_{ll}$ for [$\mathcal{ZZ}$]{} and $\sigma_{tt}$ for [$\mathcal{AA}$]{}. It is clear that for visible and IR frequencies the conductivity remains roughly constant in $\omega$, but its magnitude depends on the amount of strain. The strain dependence is shown in detail in the inset, at $\omega=1\,$eV. The perfect linearity in ${{\ensuremath{\varepsilon}}}$ of the *ab-initio* results up to at least ${{\ensuremath{\varepsilon}}}\approx 10\%$ shows that the analytical result of [eq. ]{} is indeed quite dependable for a wide range of stretching. We must point out, however, that the linear slopes quoted in [fig. \[fig:Fig-2\]]{} cannot immediately be used to extract the bandstructure parameter $\lambda$ using [eq. ]{}. That is because the *ab-initio* calculation naturally includes the relaxation and deformation of the lattice, which, as advanced in the very beginning, we chose to ignore, not to encumber the discussion by more complicated expressions which do not change the main results.
Taking the lattice deformation into account leads to a correction of the expansion of the dispersion around the Dirac points. This leads to a renormalization of entries in the matrix $\bm{\eta}^2$. It can be shown straightforwardly that such changes amount to replacing $\lambda a\to\lambda a - 1$ in $\bm{\eta}^2$, and, consequently, in every ensuing result. In this way, the *ab-initio* slopes quoted in [fig. \[fig:Fig-2\]]{} correspond to $\lambda \simeq 2.42\,\text{\AA}^{-1}$ ([$\mathcal{ZZ}$]{}) and $\lambda \simeq 2.56\,\text{\AA}^{-1}$ ([$\mathcal{AA}$]{}) for the inset of panel (a), and $\lambda \simeq 2.97\,\text{\AA}^{-1}$ ([$\mathcal{ZZ}$]{}) and $\lambda \simeq 2.17\,\text{\AA}^{-1}$ ([$\mathcal{AA}$]{}) for the inset of panel (b). This rather nicely agrees with the estimates quoted earlier that put $\lambda a\sim3-4$.
van Hove Singularities
======================
In the UV band, $\sigma(\omega)$ is dominated by direct transitions between the $M$ points in the BZ, which coincide with VHS in the electronic dispersion. From the point of view of , these transitions occur at momenta ${{\ensuremath{\bm{k}}}}$ coinciding precisely with $M_{1,3}=(\pm\pi/\sqrt{3},\,\pi/3)$ and $M_2=(2\pi/3,\,0)$ (see [fig. \[fig:Fig-1\]]{} for our convention regarding the $M$ points). The corresponding resonant frequencies are therefore given by $\omega_{Mi} = 2 |E({{\ensuremath{\bm{k}}}}=M_i)|$, and read $\omega_{M1} = 2|t_1+t_2-t_3|$, $\omega_{M2} = 2|t_1-t_2+t_3|$, and $\omega_{M3} = 2|-t_1+t_2+t_3|$. In the most general situation $t_1\ne t_2\ne t_3$ the VHS split. Linearly expanding the hoppings in the magnitude of uniaxial strain, and defining $\omega_{Mi}\approx2t+ta\lambda{{\ensuremath{\varepsilon}}}\Delta\omega_{Mi}$, such splitting acquires the explicit form
\[eq:M-splitting\] $$\begin{aligned}
\Delta\omega_{M1,3} & \simeq
\nu-1+(1+\nu)\bigl(\cos2\theta\pm\sqrt{3}\sin2\theta\bigr)
,\\
\Delta\omega_{M2} & \simeq
\nu-1-2(1+\nu)\cos2\theta
.\end{aligned}$$
But, as is obvious from the *ab-initio* results in [fig. \[fig:Fig-2\]]{}, this splitting is not always visible in the absorption spectrum. That is because the velocity matrix elements impose a modulation of the strength associated with these transitions. In order to extract this effect we need to abandon the Dirac approximation (\[eq:H-eff-strain\],\[eq:v-matrix-Dirac\]), and work with the full-tight binding bandstructure and matrix elements. In take, for example, the matrix elements at ${{\ensuremath{\bm{k}}}}=M_2$, which read
\[eq:v-M2\] $$\begin{aligned}
|v_x^{M_2}|^2 & =
\frac{3a^2}{4\hbar^2}(t_1-t_3)^2
,\\
|v_y^{M_2}|^2 & =
\frac{1a^2}{4\hbar^2}(t_1+2t_2+t_3)^2
,\\
v_x^{M_2}v_y^{M_2} & =
-\frac{\sqrt{3}a^2}{4\hbar^2}(t_1+2t_2+t_3)(t_1-t_3)
.\end{aligned}$$
When expanded in powers of strain, $|v_x^{M_2}|^2$ will be zero to linear order, irrespective of the strain direction. This suppresses the $M_2$ singularity in $\sigma'(\omega)$ leaving only the peaks related to $M_{1,3}$, whose associated matrix elements are finite. Since the geometry chosen in [fig. \[fig:Fig-2\]]{} considers only $\theta=0$ and $\pi/2$, $M_1$ and $M_3$ are still degenerate, as per , and thus only one peak should survive in $\sigma^\prime_{xx}$, precisely as seen *ab-initio* in [fig. \[fig:Fig-2\]]{}(a). On the other hand, there is no such selection rule arising from the velocity matrix elements when computing $\sigma^\prime_{yy}$, which allows the splitting between $M_2$ and $M_{1,3}$ to be observed, as [fig. \[fig:Fig-2\]]{}(b) clearly demonstrates. Eqs. also explain why in [fig. \[fig:Fig-2\]]{}(a) the strain-induced shift of the peak is less pronounced for tension along [$\mathcal{ZZ}$]{}($\theta=0$), than [$\mathcal{AA}$]{} ($\theta=\pi/2$): $\Delta\omega_{M1}({\ensuremath{\mathcal{ZZ}}}) = \nu\ \Delta\omega_{M1}({\ensuremath{\mathcal{AA}}})$. And similarly, Eqs. account for the different relative intensity of the peaks associated with $M_2$ and $M_{1,3}$ in [fig. \[fig:Fig-2\]]{}(b).
It is important to stress here that these results are not tied to a particular coordinate system. We can easily express Eqs. in any coordinate system, rotated by $\varphi$ with respect to ${\ensuremath{\mathcal{O}}}x y$, and conclude that the longitudinal matrix element $|v_{x'}^{M_i}|^2$ always vanishes when $\varphi$ coincides with a [$\mathcal{ZZ}$]{} direction. For example, $|v_{x'}^{M_2}|^2$ vanishes for $\varphi=0$, $|v_{x'}^{M_3}|^2$ vanishes for $\varphi=\pi/3$, and $|v_{x'}^{M_1}|^2$ vanishes for $\varphi=2\pi/3$. Suppression of a van Hove peak in the longitudinal conductivity therefore singles out one of the [$\mathcal{ZZ}$]{} directions of the lattice. This is consistent and explains the numerical calculations of Ref. , and has an important consequence: by monitoring the splitting in the absorption peaks associated with VHS, and by inspecting the selection rule just described, *one can measure simultaneously: the magnitude of strain, its direction, and the direction of the underlying lattice with respect to the laboratory coordinate system*. This provides an optical means to perform the same measurements that have been made by exploring the splitting of the $G$ peak in the Raman spectrum of strained graphene [@Ferrari:2009; @Hone:2009].
![ Illustration of the behavior of the transmittance as a function of incident polarization measured in the laboratory frame. According to , the phase shift $\theta$ defines the direction of strain, and the amplitude its magnitude. []{data-label="fig:Fig-3"}](Figs/Fig-3){width="45.00000%"}
Transparency and Dichroism
==========================
The asymmetry in the conductivity tensor will result in a certain degree of dichroism, as the absorbance of linearly polarized light will depend on the polarization direction with respect to the slow/fast axes. Treating graphene as a 2D conducting sheet, and solving the associated Fresnel equations, we can extract the degree of polarization rotation for normal incidence on graphene in vacuum. In the visible and IR where [eq. ]{} stands that would be $$\frac{\tan\phi_T}{\tan\phi_I} =
\frac{2+c\mu_0\sigma_{ll}}{2+c\mu_{0}\sigma_{tt}}
\approx
1-4|\delta k_D|a \frac{c\mu_0\sigma_0}{2+c\mu_0\sigma_0}
\label{eq:Polarization}
,$$ where $\phi_{T,I}$ are the transmitted and incident polarizations measured with respect to the slow/longitudinal axis, and $c\mu_0\sigma_0=\pi\alpha\approx 0.02$. Likewise, the transmittance for linear polarization becomes $${\ensuremath{\mathcal{T}}}\approx 1 - \pi\alpha\bigl[1-2|\delta k_D|a\,\cos2\phi_I\bigr]
\label{eq:Transmittance}$$ An important consequence of this periodic modulation is that it allows a direct determination of both the strain direction and its magnitude, as follows. In some laboratory coordinates is transformed by making $\phi_I\to\phi_I-\varphi_l$. The amplitude of ${\ensuremath{\mathcal{T}}}$ as a function of polarization direction determines the amount of strain, while the phase shift $\varphi_l$ fixes the direction. This is illustrated in [fig. \[fig:Fig-3\]]{}. The corrections to both polarization and transmittance are weighted by $\pi\alpha$, and will be necessarily small. But, one the one hand, the modulation amplitude is roughly $\sim 8{{\ensuremath{\varepsilon}}}\pi\alpha$ ($0.16{{\ensuremath{\varepsilon}}}$) and transmittance can be routinely measured within $0.1\%$ of precision. On the other hand, the transmittance of multilayer graphene is, to a great extent, cumulative [@Nair:2008; @Kuzmenko:2008]. This implies that the same results apply for multilayer graphene, provided one replaces $\pi\alpha\to
N\pi\alpha$ in , where $N$ accounts for the number of graphene layers. The effect is thus naturally enhanced in multilayers, as it is in the vicinity of the VHS (or any resonance, for that matter). Another interesting application is that, if strain can be controlled with precision, an expression like allows, by means of a simple optical experiment, the measurement of the bandstructure parameter $\lambda = d\log t/dr$, whose knowledge is crucial for the characterization of all strain-induced effects on the bandstructure.
Discussion
==========
The photoelastic effect in undoped graphene has been quantified, and shown to possess features that might be appealing in the development of atomically thin optical elements. One of such characteristics is the frequency independent response in a very large frequency range, which remains valid when the system is anisotropically strained. This constancy and predictability is a relevant feature for broadband applications. The degree of anisotropy induced by strain is determined by how much the Dirac point is displaced from its position at $K/K'$ in the BZ: $|\delta k_D|$ . This opens several possibilities, such as: monitoring optical absorption as a function of strain to characterize the band parameters of graphene; or monitoring the transmittance as a function of incident polarization in order to measure the magnitude and direction of strain in graphene devices. This last example could find applications in completely passive, transparent, strain sensors.
Even if the magnitude of the photoelastic effect in graphene at visible or IR frequencies is limited by its small natural absorption of $\pi\alpha\sim 2\%$, it is nonetheless significant for an atomically thin membrane. Additional versatility is provided by the fact that the effect can be naturally amplified by stacking multilayers, or that the optical absorption can be radically affected by electronic doping as well, on account of Pauli blocking [@Ando:2002; @Basov:2008].
A determination of the lattice orientation cannot be made from the optical absorption at low energies, simply due to the isotropy of the Dirac dispersion. By contrast, the electronic dispersion at energies close to the VHS fully reflects the symmetry of the underlying lattice. Therefore, by measuring the optical response at frequencies resonant with the van Hove transitions, and analyzing the strain-induced splitting and selection rules of the absorption peaks, one can extract the amount of strain, its direction, and the lattice orientation.
Another consequence of the strain-induced tunability of the VHS is that it can have a significant import in current efforts to elevate the Fermi level of graphene up to the VHS [@Barbaros:2010; @Kim:2010; @Ye:2010]. The ability to achieve this, and thereby dramatically increase the electronic DOS, is expected to facilitate many-body instabilities and the establishment of correlated phases, such as superconductivity, or charge/spin density waves [@Gonzalez:2008; @Rotenberg:2010]. Strain engineering of graphene can, in this respect, work as a facilitator and provide added tunability. For example, from it follows that the energy of the VHS can be reduced by uniaxial strain to values as low as $E/t \simeq 1-(3+\nu)a\lambda{{\ensuremath{\varepsilon}}}/2$. Using $\lambda a \sim 3$ for estimate purposes we can write $E/t \simeq 1-5{{\ensuremath{\varepsilon}}}$, so that the reduction for 10% strain can be as much as 50% in the position of the VHS. On top of this we would need to include excitonic corrections that are known to renormalize the VHS further down in energy [@Louie:2009], and which we neglect in our treatment.
Finally, the photoelastic effect is the basis of many optical and mechano-optical devices and applications at the macro-scale. The characteristics of graphene in this respect, which we just surfaced, might provide a valuable route towards the downscaling of those concepts to the realm of atomically thin optical elements, and their application at the nanoscale.
AHCN acknowledges the partial support of the U.S. DOE under grant DE-FG02-08ER46512, and ONR grant MURI N00014-09-1-1063.
[^1]: On leave from Physics Department, Boston University.
| {
"pile_set_name": "ArXiv"
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.