Randomized Hamiltonian Monte Carlo

Abstract

Tuning the durations of the Hamiltonian flow in Hamiltonian Monte Carlo (also called Hybrid Monte Carlo) (HMC) involves a tradeoff between computational cost and sampling quality, which is typically challenging to resolve in a satisfactory way. In this article we present and analyze a randomized HMC method (RHMC), in which these durations are i.i.d. exponential random variables whose mean is a free parameter. We focus on the small time step size limit, where the algorithm is rejection-free and the computational cost is proportional to the mean duration. In this limit, we prove that RHMC is geometrically ergodic under the same conditions that imply geometric ergodicity of the solution to underdamped Langevin equations. Moreover, in the context of a multi-dimensional Gaussian distribution, we prove that the sampling efficiency of RHMC, unlike that of constant duration HMC, behaves in a regular way. This regularity is also verified numerically in non-Gaussian target distributions. Finally we suggest variants of RHMC for which the time step size is not required to be small.

40 pages, to be published in Annals of Applied Probability

References in corpus (2)

• Numerical integrators for the Hybrid Monte Carlo method
• Compressible Generalized Hybrid Monte Carlo

Cited by in corpus (38)

• Sampling Can Be Faster Than Optimization
• On the Geometric Ergodicity of Hamiltonian Monte Carlo
• Geometric integrators and the Hamiltonian Monte Carlo method
• Coupling and Convergence for Hamiltonian Monte Carlo
• Discontinuous Hamiltonian Monte Carlo for discrete parameters and discontinuous likelihoods
• On the convergence of Hamiltonian Monte Carlo
• A generalized class of strongly stable and dimension-free T-RPMD integrators
• Adaptive multi-stage integrators for optimal energy conservation in molecular simulations
• Efficient Learning of the Parameters of Non-Linear Models using Differentiable Resampling in Particle Filters
• Convergence of unadjusted Hamiltonian Monte Carlo for mean-field models
• On explicit $L^2$-convergence rate estimate for piecewise deterministic Markov processes in MCMC algorithms
• Does Hamiltonian Monte Carlo mix faster than a random walk on multimodal densities?
• Simulating Coulomb gases and log-gases with hybrid Monte Carlo algorithms
• Weighted $L^2$-contractivity of Langevin dynamics with singular potentials
• Couplings and quantitative contraction rates for Langevin dynamics
• Couplings for Andersen Dynamics
• Two-scale coupling for preconditioned Hamiltonian Monte Carlo in infinite dimensions
• Birth-death dynamics for sampling: Global convergence, approximations and their asymptotics
• Subgeometric hypocoercivity for piecewise-deterministic Markov process Monte Carlo methods
• Peskun-Tierney ordering for Markov chain and process Monte Carlo: beyond the reversible scenario
• Cores for Piecewise-Deterministic Markov Processes used in Markov Chain Monte Carlo
• Quantum-Inspired Hamiltonian Monte Carlo for Bayesian Sampling
• Complexity of zigzag sampling algorithm for strongly log-concave distributions
• Non-reversible lifts of reversible diffusion processes and relaxation times
• Concentration Inequalities and UQ Bounds for Hypocoercive MCMC Samplers
• Hamiltonian Monte Carlo on Symmetric and Homogeneous Spaces via Symplectic Reduction
• HMC: avoiding rejections by not using leapfrog and some results on the acceptance rate
• Cayley Splitting for Second-Order Langevin Stochastic Partial Differential Equations
• The shifted ODE method for underdamped Langevin MCMC
• Practical Bayesian System Identification using Hamiltonian Monte Carlo
• An Introduction to Hamiltonian Monte Carlo Method for Sampling
• Unadjusted Hamiltonian MCMC with Stratified Monte Carlo Time Integration
• The connections between Lyapunov functions for some optimization algorithms and differential equations
• Bayesian inference methodology to characterize the dust emissivity at far-infrared and submillimeter frequencies
• Accelerating MCMC Algorithms
• Well-posedness and long time behavior of singular Langevin stochastic differential equations
• Is the NUTS algorithm correct?
• An entropic approach for Hamiltonian Monte Carlo: the idealized case