Abstract:Parallel Markov Chain Monte Carlo (pMCMC) algorithms generate clouds of proposals at each step to efficiently resolve a target probability distribution. We build a rigorous foundational framework for pMCMC algorithms that situates these methods within a unified 'extended phase space' measure-theoretic formalism. Drawing on our recent work that provides a comprehensive theory for reversible single proposal methods, we herein derive general criteria for multiproposal acceptance mechanisms which yield ergodic chains on general state spaces. Our formulation encompasses a variety of methodologies, including proposal cloud resampling and Hamiltonian methods, while providing a basis for the derivation of novel algorithms. In particular, we obtain a top-down picture for a class of methods arising from 'conditionally independent' proposal structures. As an immediate application, we identify several new algorithms including a multiproposal version of the popular preconditioned Crank-Nicolson (pCN) sampler suitable for high- and infinite-dimensional target measures which are absolutely continuous with respect to a Gaussian base measure. To supplement our theoretical results, we carry out a selection of numerical case studies that evaluate the efficacy of these novel algorithms. First, noting that the true potential of pMCMC algorithms arises from their natural parallelizability, we provide a limited parallelization study using TensorFlow and a graphics processing unit to scale pMCMC algorithms that leverage as many as 100k proposals at each step. Second, we use our multiproposal pCN algorithm (mpCN) to resolve a selection of problems in Bayesian statistical inversion for partial differential equations motivated by fluid measurement. These examples provide preliminary evidence of the efficacy of mpCN for high-dimensional target distributions featuring complex geometries and multimodal structures.

Processus aléatoires et applications -- Algorithmes MCMC et vitesse de convergence

Processus aléatoires et applications

An introduction to random matrix theory

Notes on Randomized Algorithms

An invitation to adaptive Markov chain Monte Carlo convergence theory

A First Course in Monte Carlo Methods

A calculus for Markov chain Monte Carlo: studying approximations in algorithms

On the concentration properties of Interacting particle processes

Continuous Martingales and Brownian Motion

Rademacher complexity for Markov chains : Applications to kernel smoothing and Metropolis-Hasting

Lectures on Probability Theory

Deviation inequalities, Moderate deviations and some limit theorems for bifurcating Markov chains with application

Interacting Markov chain Monte Carlo methods for solving nonlinear measure-valued equations

Convergence of Langevin-simulated annealing algorithms with multiplicative noise

On Convergence Rate Bounds for a Class of Nonlinear Markov Chains

Fluctuations of Interacting Markov Chain Monte Carlo Methods

Introduction to the theory of stochastic processes and Brownian motion problems

Air Markov Chain Monte Carlo

A Tutorial on Advanced Dynamic Monte Carlo Methods for Systems with Discrete State Spaces

Parallel MCMC Algorithms: Theoretical Foundations, Algorithm Design, Case Studies

PDMP characterisation of event-chain Monte Carlo algorithms for particle systems