On the accept-reject mechanism for Metropolis-Hastings algorithms

DOI10.1214/23-AAP1948MaRDI QIDQ6139681zbMATH OpenOpenAlexFDO

Authors N. Glatt-Holtz, Justin Krometis, Cecilia F. Mondaini

Publication date 19 January 2024

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

Full work available at URL https://arxiv.org/abs/2011.04493

zbMATH Keywords

Hamiltonian Monte CarloMarkov chain Monte Carlo (MCMC) algorithmsMetropolis-Hastings algorithms, sampling on abstract state spacessurrogate trajectory methods

Mathematics Subject Classification ID

Monte Carlo methods (65C05) Numerical methods for Hamiltonian systems including symplectic integrators (65P10)

Abstract: This work develops a powerful and versatile framework for determining acceptance ratios in Metropolis-Hastings type Markov kernels widely used in statistical sampling problems. Our approach allows us to derive new classes of kernels which unify random walk or diffusion-type sampling methods with more complicated "extended phase space" algorithms based around ideas from Hamiltonian dynamics. Our starting point is an abstract result developed in the generality of measurable state spaces that addresses proposal kernels that possess a certain involution structure. Note that, while this underlying proposal structure suggests a scope which includes Hamiltonian-type kernels, we demonstrate that our abstract result is, in an appropriate sense, equivalent to an earlier general state space setting developed in [Tierney, Annals of Applied Probability, 1998] where the connection to Hamiltonian methods was more obscure. Altogether, the theoretical unity and reach of our main result provides a basis for deriving novel sampling algorithms while laying bare important relationships between existing methods.

Review text

The main contribution of this work is the presentation of a fairly general method to define reversible Markov chains with respect to a given target measure, from a given proposal kernel on an extended state space, by defining a suitable acceptance probability. This extends the classical Metropolis-Hastings settings, and more precisely it is shown to be equivalent to the general settings of a work of Tierney in 1998. By contrast to the latter, the present method is written in a way which is reminiscent of the Hamiltonian Monte Carlo algorithm, which may be easier to interpret and thus to apply. The method is meant to be very versatile, covering in particular infinite-dimensional problems or non-separable Hamilonian schemes. Half of the work is concerned with examples, unifying many popular samplers and introducing some variations, motivated in particular by the use of so-called ``surrogate dynamics which can be used in place of the standard Hamiltonian dynamics with a reduce numerical cost.

Recommendations

• On Irreversible Metropolis Sampling Related to Langevin Dynamics
• Spectral gaps for a Metropolis-Hastings algorithm in infinite dimensions
• Hybrid Monte Carlo on Hilbert spaces
• Markov chain Monte Carlo algorithms with sequential proposals
• Irreducibility and geometric ergodicity of Hamiltonian Monte Carlo

Cites work

• A Bayesian Approach to Estimating Background Flows from a Passive Scalar
• A computational framework for infinite-dimensional Bayesian inverse problems. II: stochastic Newton MCMC with application to ice sheet flow inverse problems
• A note on Metropolis-Hastings kernels for general state spaces
• A stochastic Newton MCMC method for large-scale statistical inverse problems with application to seismic inversion
• Advanced MCMC methods for sampling on diffusion pathspace
• An analysis of infinite dimensional Bayesian inverse shape acoustic scattering and its numerical approximation
• An introduction to manifolds. 2nd revised ed.
• Analysis of SPDEs arising in path sampling. I: The Gaussian case
• Analysis of SPDEs arising in path sampling. II: The nonlinear case
• Bayesian data analysis.
• Classical Dynamics
• Emulation of higher-order tensors in manifold Monte Carlo methods for Bayesian inverse problems
• Equation of state calculations by fast computing machines
• Error bounds for Metropolis-Hastings algorithms applied to perturbations of Gaussian measures in high dimensions
• Exponential convergence of Langevin distributions and their discrete approximations
• FEM-based discretization-invariant MCMC methods for PDE-constrained Bayesian inverse problems
• Geometric integrators and the Hamiltonian Monte Carlo method
• Geometric MCMC for infinite-dimensional inverse problems
• Geometric Numerical Integration
• Gradient flows in metric spaces and in the space of probability measures
• Hamiltonian Monte Carlo acceleration using surrogate functions with random bases
• scientific article; zbMATH DE number 5954252 (Why is no real title available?)
• scientific article; zbMATH DE number 1324223 (Why is no real title available?)
• scientific article; zbMATH DE number 1522714 (Why is no real title available?)
• Hybrid Monte Carlo on Hilbert spaces
• Invariant measures of stochastic partial differential equations and conditioned diffusions
• Inverse problems: a Bayesian perspective
• MCMC METHODS FOR DIFFUSION BRIDGES
• MCMC methods for functions: modifying old algorithms to make them faster
• MCMC using Hamiltonian dynamics
• Measure theory. Vol. I and II
• Mixing rates for Hamiltonian Monte Carlo algorithms in finite and infinite dimensions
• Modified Hamiltonian Monte Carlo for Bayesian inference
• Monte Carlo sampling methods using Markov chains and their applications
• Monte Carlo strategies in scientific computing.
• Neural network gradient Hamiltonian Monte Carlo
• On perturbed proximal gradient algorithms
• On the Probabilistic Cauchy Theory for Nonlinear Dispersive PDEs
• Periodic nonlinear Schrödinger equation and invariant measures
• Precomputing strategy for Hamiltonian Monte Carlo method based on regularity in parameter space
• Randomness and nonlinear evolution equations
• Reversible jump Markov chain Monte Carlo computation and Bayesian model determination
• Riemann manifold Langevin and Hamiltonian Monte Carlo methods. With discussion and authors' reply
• Sampling conditioned diffusions
• Signal processing problems on function space: Bayesian formulation, stochastic PDEs and effective MCMC methods
• Simulating Hamiltonian Dynamics
• Spectral gaps for a Metropolis-Hastings algorithm in infinite dimensions
• Statistical and computational inverse problems.
• Stochastic Equations in Infinite Dimensions
• The Bouncy Particle Sampler: A Non-Reversible Rejection-Free Markov Chain Monte Carlo Method
• The Convergence of Markov Chain Monte Carlo Methods: From the Metropolis Method to Hamiltonian Monte Carlo
• The Multiple-Try Method and Local Optimization in Metropolis Sampling
• The no-U-turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo
• Two-scale coupling for preconditioned Hamiltonian Monte Carlo in infinite dimensions

Cited in

(6)

• HMC: reducing the number of rejections by not using leapfrog and some results on the acceptance rate
• Two Metropolis--Hastings Algorithms for Posterior Measures with Non-Gaussian Priors in Infinite Dimensions
• A large deviation principle for the empirical measures of Metropolis-Hastings chains
• Numerical unique ergodicity of monotone SDEs driven by nondegenerate multiplicative noise
• GIST: Gibbs self-tuning for locally adaptive Hamiltonian Monte Carlo
• Large deviations for independent Metropolis Hastings and Metropolis-adjusted Langevin algorithm

This page was built for publication: On the accept-reject mechanism for Metropolis-Hastings algorithms

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6139681)