A Framework to Analyze Multiscale Sampling MCMC Methods

Abstract

We consider the theoretical analysis of Multiscale Sampling Methods, which are a new class of gradient-free Markov chain Monte Carlo (MCMC) methods for high dimensional inverse differential equation problems. A detailed presentation of those methods is given, including a review of each MCMC technique that they employ. Then, we propose a two-part framework to study and compare those methods. The first part identifies the new corresponding state space for the chain of random fields, and the second assesses convergence conditions on the instrumental and target distributions. Three Multiscale Sampling Methods are then analyzed using this new framework.