On the Geometric Ergodicity of Hamiltonian Monte Carlo

Abstract

We establish general conditions under which Markov chains produced by the Hamiltonian Monte Carlo method will and will not be geometrically ergodic. We consider implementations with both position-independent and position-dependent integration times. In the former case we find that the conditions for geometric ergodicity are essentially a gradient of the log-density which asymptotically points towards the centre of the space and grows no faster than linearly. In an idealised scenario in which the integration time is allowed to change in different regions of the space, we show that geometric ergodicity can be recovered for a much broader class of tail behaviours, leading to some guidelines for the choice of this free parameter in practice.

arXiv comment29 pages + supplement (included in arXival as Appendix), 1 figure. Corrigendum related to Theorem 5.14 and Corollary 2.3 included as Appendix D

References in corpus (16)

• The No-U-Turn Sampler: Adaptively Setting Path Lengths in Hamiltonian Monte Carlo
• General state space Markov chains and MCMC algorithms
• A Short History of Markov Chain Monte Carlo: Subjective Recollections from Incomplete Data
• Renewal theory and computable convergence rates for geometrically ergodic Markov chains
• Randomized Hamiltonian Monte Carlo
• Geometric integrators and the Hamiltonian Monte Carlo method
• Geodesic Monte Carlo on Embedded Manifolds
• Optimizing The Integrator Step Size for Hamiltonian Monte Carlo
• A General Metric for Riemannian Manifold Hamiltonian Monte Carlo
• On the convergence of Hamiltonian Monte Carlo
• Information-geometric Markov Chain Monte Carlo methods using Diffusions
• Hamiltonian Monte Carlo Without Detailed Balance
• Some things we've learned (about Markov chain Monte Carlo)
• Extra Chance Generalized Hybrid Monte Carlo
• Identifying the Optimal Integration Time in Hamiltonian Monte Carlo
• Error bounds for Metropolis-Hastings algorithms applied to perturbations of Gaussian measures in high dimensions

Cited by in corpus (24)

• A Conceptual Introduction to Hamiltonian Monte Carlo
• Rapid Mixing of Hamiltonian Monte Carlo on Strongly Log-Concave Distributions
• Discontinuous Hamiltonian Monte Carlo for discrete parameters and discontinuous likelihoods
• Diagnosing Suboptimal Cotangent Disintegrations in Hamiltonian Monte Carlo
• On the convergence of Hamiltonian Monte Carlo
• Modified Hamiltonian Monte Carlo for Bayesian inference
• Convergence of unadjusted Hamiltonian Monte Carlo for mean-field models
• Fast mixing of Metropolized Hamiltonian Monte Carlo: Benefits of multi-step gradients
• Does Hamiltonian Monte Carlo mix faster than a random walk on multimodal densities?
• Expert Elicitation and Data Noise Learning for Material Flow Analysis using Bayesian Inference
• Approximations of Geometrically Ergodic Reversible Markov Chains
• Couplings for Andersen Dynamics
• Two-scale coupling for preconditioned Hamiltonian Monte Carlo in infinite dimensions
• On the Ergodicity, Bias and Asymptotic Normality of Randomized Midpoint Sampling Method
• Geometric Methods for Sampling, Optimisation, Inference and Adaptive Agents
• Geometry-Aware Hamiltonian Variational Auto-Encoder
• Multiple projection MCMC algorithms on submanifolds
• Recycling intermediate steps to improve Hamiltonian Monte Carlo
• HMC: avoiding rejections by not using leapfrog and some results on the acceptance rate
• Markov Chain Monte Carlo Methods for Estimating Systemic Risk Allocations
• Magnetic Manifold Hamiltonian Monte Carlo
• A Unifying and Canonical Description of Measure-Preserving Diffusions
• Understanding MCMC Dynamics as Flows on the Wasserstein Space
• Convergence of position-dependent MALA with application to conditional simulation in GLMMs