29
11

Concentration of the Langevin Algorithm's Stationary Distribution

Abstract

A canonical algorithm for log-concave sampling is the Langevin Algorithm, aka the Langevin Diffusion run with some discretization stepsize η>0\eta > 0. This discretization leads the Langevin Algorithm to have a stationary distribution πη\pi_{\eta} which differs from the stationary distribution π\pi of the Langevin Diffusion, and it is an important challenge to understand whether the well-known properties of π\pi extend to πη\pi_{\eta}. In particular, while concentration properties such as isoperimetry and rapidly decaying tails are classically known for π\pi, the analogous properties for πη\pi_{\eta} are open questions with direct algorithmic implications. This note provides a first step in this direction by establishing concentration results for πη\pi_{\eta} that mirror classical results for π\pi. Specifically, we show that for any nontrivial stepsize η>0\eta > 0, πη\pi_{\eta} is sub-exponential (respectively, sub-Gaussian) when the potential is convex (respectively, strongly convex). Moreover, the concentration bounds we show are essentially tight. Key to our analysis is the use of a rotation-invariant moment generating function (aka Bessel function) to study the stationary dynamics of the Langevin Algorithm. This technique may be of independent interest because it enables directly analyzing the discrete-time stationary distribution πη\pi_{\eta} without going through the continuous-time stationary distribution π\pi as an intermediary.

View on arXiv
Comments on this paper