Abstract:It is of significant interest in many applications to sample from a high-dimensional target distribution $\pi$ with the density $\pi(\text{d} x) \propto e^{-U(x)} (\text{d} x) $, based on the temporal discretization of the Langevin stochastic differential equations (SDEs). In this paper, we propose an explicit projected Langevin Monte Carlo (PLMC) algorithm with non-convex potential $U$ and super-linear gradient of $U$ and investigate the non-asymptotic analysis of its sampling error in total variation distance. Equipped with time-independent regularity estimates for the associated Kolmogorov equation, we derive the non-asymptotic bounds on the total variation distance between the target distribution of the Langevin SDEs and the law induced by the PLMC scheme with order $\mathcal{O}(d^{\max\{3\gamma/2 , 2\gamma-1 \}} h |\ln h|)$, where $d$ is the dimension of the target distribution and $\gamma \geq 1$ characterizes the growth of the gradient of $U$. In addition, if the gradient of $U$ is globally Lipschitz continuous, an improved convergence order of $\mathcal{O}(d^{3/2} h)$ for the classical Langevin Monte Carlo (LMC) scheme is derived with a refinement of the proof based on Malliavin calculus techniques. To achieve a given precision $\epsilon$, the smallest number of iterations of the PLMC algorithm is proved to be of order ${\mathcal{O}}\big(\tfrac{d^{\max\{3\gamma/2 , 2\gamma-1 \}}}{\epsilon} \ \cdot \ln (\tfrac{d}{\epsilon}) \cdot \ln (\tfrac{1}{\epsilon}) \big)$. In particular, the classical Langevin Monte Carlo (LMC) scheme with the non-convex potential $U$ and the globally Lipschitz gradient of $U$ can be guaranteed by order ${\mathcal{O}}\big(\tfrac{d^{3/2}}{\epsilon} \cdot \ln (\tfrac{1}{\epsilon}) \big)$. Numerical experiments are provided to confirm the theoretical findings.

Ergodicity of Langevin Dynamics and its Discretizations for Non-smooth Potentials

Subgradient Langevin Methods for Sampling from Non-smooth Potentials

Non-asymptotic entropic bounds for non-linear kinetic Langevin sampler with second-order splitting scheme

Convergence of kinetic Langevin samplers for non-convex potentials

Contraction and Convergence Rates for Discretized Kinetic Langevin Dynamics

Projected Langevin Monte Carlo algorithms in non-convex and super-linear setting

Appropriate State-Dependent Friction Coefficient Accelerates Kinetic Langevin Dynamics

Exact potentials in multivariate Langevin equations

Penalized Langevin dynamics with vanishing penalty for smooth and log-concave targets

Multilevel-Langevin pathwise average for Gibbs approximation

Convergence of Kinetic Langevin Monte Carlo on Lie groups

Exponential ergodicity of Lévy driven Langevin dynamics with singular potentials

Projected Langevin dynamics and a gradient flow for entropic optimal transport

Analysis of Primal-Dual Langevin Algorithms

Concentration of the Langevin Algorithm's Stationary Distribution

Efficient Sampling on Riemannian Manifolds via Langevin MCMC

From Optimization to Sampling via Lyapunov Potentials

Langevin Monte Carlo for strongly log-concave distributions: Randomized midpoint revisited

Sampling and estimation on manifolds using the Langevin diffusion