Mixing time guarantees for unadjusted Hamiltonian Monte Carlo

Nawaf Bou-Rabee, Andreas Eberle

Research output: Contribution to journalArticlepeer-review

1 Scopus citations

Abstract

We provide quantitative upper bounds on the total variation mixing time of the Markov chain corresponding to the unadjusted Hamiltonian Monte Carlo (uHMC) algorithm. For two general classes of models and fixed time discretization step size h, the mixing time is shown to depend only logarithmically on the dimension. Moreover, we provide quantitative upper bounds on the total variation distance between the invariant measure of the uHMC chain and the true target measure. As a consequence, we show that an e-accurate approximation of(the target distribution) µ in total variation distance can be achieved by uHMC: (i) for a broad class of models with O gradient evaluations; and (ii) for mean field models with weak interactions with O d3/4e-1/2 log(d/e) ) gradient ( d1/2 e-1/2 log(d/e) evaluations. The proofs are based on the construction of successful couplings for uHMC that realize the upper bounds.

Original languageEnglish (US)
Pages (from-to)75-104
Number of pages30
JournalBernoulli
Volume29
Issue number1
DOIs
StatePublished - Feb 2023

All Science Journal Classification (ASJC) codes

  • Statistics and Probability

Keywords

  • couplings
  • Hamiltonian Monte Carlo
  • MCMC
  • mixing time
  • variational integrators

Fingerprint

Dive into the research topics of 'Mixing time guarantees for unadjusted Hamiltonian Monte Carlo'. Together they form a unique fingerprint.

Cite this