Mixing Time Guarantees for Unadjusted Hamiltonian Monte Carlo
This work addresses theoretical guarantees for sampling efficiency in high-dimensional Bayesian inference, offering incremental improvements in mixing time bounds.
The paper provides quantitative upper bounds on the mixing time of unadjusted Hamiltonian Monte Carlo (uHMC) for two model classes, showing logarithmic dependence on dimension, and achieves ε-approximation with O(d^{3/4}ε^{-1/2}log(d/ε)) or O(d^{1/2}ε^{-1/2}log(d/ε)) gradient evaluations.
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 $\varepsilon$-accurate approximation of the target distribution $μ$ in total variation distance can be achieved by uHMC for a broad class of models with $O\left(d^{3/4}\varepsilon^{-1/2}\log (d/\varepsilon )\right)$ gradient evaluations, and for mean field models with weak interactions with $O\left(d^{1/2}\varepsilon^{-1/2}\log (d/\varepsilon )\right)$ gradient evaluations. The proofs are based on the construction of successful couplings for uHMC that realize the upper bounds.