The structure of low-complexity Gibbs measures on product spaces

DOI10.1214/19-AOP1352zbMATH Open1444.60006arXiv1810.07278MaRDI QIDQ2189463FDOQ2189463

Tim Austin

Publication date: 15 June 2020

Published in: The Annals of Probability (Search for Journal in Brave)

Abstract: Let $K_1$, $dots$, $K_n$ be bounded, complete, separable metric spaces. Let $lambd{a}_i$ be a Borel probability measure on $K_i$ for each $i$. Let $f:pro{d}_i{K}_{i}omathbbR$ be a bounded and continuous potential function, and let ${\displaystyle }$mu(d mathbf{x}) propto e^{f(mathbf{x})}lambda_1(d x_1)cdots lambda_n(d x_n)${\displaystyle }$ be the associated Gibbs distribution. At each point $mathbfxinpro{d}_i{K}_i$, one can define a `discrete gradient' $ablaf(mathbfx,,cdot,)$ by comparing the values of $f$ at all points which differ from $mathbfx$ in at most one coordinate. In case $pro{d}_i{K}_i={-1,1}^{n}subsetmathbb{R}^n$, the discrete gradient $ablaf(mathbfx,,cdot,)$ is naturally identified with a vector in $mathbb{R}^n$. This paper shows that a `low-complexity' assumption on $ablaf$ implies that $mu$ can be approximated by a mixture of other measures, relatively few in number, and most of them close to product measures in the sense of optimal transport. This implies also an approximation to the partition function of $f$ in terms of product measures, along the lines of Chatterjee and Dembo's theory of `nonlinear large deviations'. An important precedent for this work is a result of Eldan in the case $pro{d}_i{K}_i={-1,1}^n$. Eldan's assumption is that the discrete gradients $ablaf(mathbfx,,cdot,)$ all lie in a subset of $mathbb{R}^n$ that has small Gaussian width. His proof is based on the careful construction of a diffusion in $mathbb{R}^n$ which starts at the origin and ends with the desired distribution on the subset ${-1,1}^n$. Here our assumption is a more naive covering-number bound on the set of gradients $ablaf(mathbfx,,cdot,):mathbfxinpro{d}_i{K}_i$, and our proof relies only on basic inequalities of information theory. As a result, it is shorter, and applies to Gibbs measures on arbitrary product spaces.

Full work available at URL: https://arxiv.org/abs/1810.07278

Recommendations

• Gaussian-width gradient complexity, reverse log-Sobolev inequalities and nonlinear large deviations
• Decomposition of mean-field Gibbs distributions into product measures
• Quantitative approximate independence for continuous mean field Gibbs measures
• Stein's method for discrete Gibbs measures
• A transportation approach to the mean-field approximation

zbMATH Keywords

Gibbs measuresnonlinear large deviationsdual total correlationgradient complexitymixtures of product measures

Mathematics Subject Classification ID

Optimal transportation (49Q22) Probability measures on topological spaces (60B05) Measures of information, entropy (94A17) Lattice systems (Ising, dimer, Potts, etc.) and systems on graphs arising in equilibrium statistical mechanics (82B20) Stochastic processes (60G99)

Cites Work

• Elements of Information Theory
• Real Analysis and Probability
• Probability in Banach spaces. Isoperimetry and processes
• Title not available (Why is that?)
• I-divergence geometry of probability distributions and minimization problems
• Measure concentration for a class of random processes
• Bounding \(\bar d\)-distance by informational divergence: A method to prove measure concentration
• A simple proof of the blowing-up lemma (Corresp.)
• Concentration of measure and isoperimetric inequalities in product spaces
• Title not available (Why is that?)
• Transportation cost for Gaussian and other product measures
• Nonlinear large deviations
• On the variational problem for upper tails in sparse random graphs
• Nonnegative entropy measures of multivariate symmetric correlations
• Concentration of measure inequalities for Markov chains and \(\Phi\)-mixing processes.
• Information inequalities and concentration of measure
• Large deviations for random graphs. École d'Été de Probabilités de Saint-Flour XLV -- 2015
• Linear dependence structure of the entropy space
• Decomposition of mean-field Gibbs distributions into product measures
• Measure concentration and the weak Pinsker property
• Gaussian-width gradient complexity, reverse log-Sobolev inequalities and nonlinear large deviations
• Upper tails and independence polynomials in random graphs
• Exponential random graphs behave like mixtures of stochastic block models

Cited In (14)

• Large deviations for the largest eigenvalue of Gaussian networks with constant average degree
• Upper tails via high moments and entropic stability
• Large deviations of subgraph counts for sparse Erdős-Rényi graphs
• Nonlinear large deviation bounds with applications to Wigner matrices and sparse Erdős-Rényi graphs
• Multi-variate correlation and mixtures of product measures
• Upper tail for homomorphism counts in constrained sparse random graphs
• A transportation approach to the mean-field approximation
• Taming correlations through entropy-efficient measure decompositions with applications to mean-field approximation
• Nonlinear large deviations: beyond the hypercube
• Spectral edge in sparse random graphs: upper and lower tail large deviations
• Upper tail of the spectral radius of sparse Erdös-Rényi graphs
• Upper Tail Large Deviations of Regular Subgraph Counts in Erdős‐Rényi Graphs in the Full Localized Regime
• Replica symmetry in upper tails of mean-field hypergraphs
• A Dimension-Free Reverse Logarithmic Sobolev Inequality for Low-Complexity Functions in Gaussian Space

This page was built for publication: The structure of low-complexity Gibbs measures on product spaces

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2189463)