Two-scale criteria for Poincaré and log-Sobolev inequalities with applications to Markov chain Monte Carlo

Abstract

Given a collection of distributions $\{P_y\}$ and a mixing distribution $\rho$ supported over ${\mathbb{R}}^d$, we propose new sufficient conditions under which the mixture / joint distribution satisfies a Poincaré or log-Sobolev inequality. We develop these sufficient conditions in a unified manner using the framework of $\Phi$-Sobolev inequalities (Chafaï, 2004). The conditions that we develop in this work are satisfied by a variety of Markov chains, and consequently allows us to characterise the evolution of these functional inequalities for iterates generated by simulating these Markov chains. As a result, we obtain an clean error analysis for estimating a broad class of functionals using Markov chain Monte Carlo strategies along these Markov chains.

v2: discusses a broader condition that supersedes previous conditions