English

Multilevel-Langevin pathwise average for Gibbs approximation

Numerical Analysis 2024-02-13 v2 Numerical Analysis Probability

Abstract

We propose and study a new multilevel method for the numerical approximation of a Gibbs distribution π\pi on Rd\mathbb{R}^d, based on (overdamped) Langevin diffusions. This method inspired by \cite{mainPPlangevin} and \cite{giles_szpruch_invariant} relies on a multilevel occupation measure, i.e.i.e. on an appropriate combination of RR occupation measures of (constant-step) Euler schemes with respective steps γr=γ02r\gamma_r = \gamma_0 2^{-r}, r=0,,Rr=0,\ldots,R. We first state a quantitative result under general assumptions which guarantees an \textit{ε\varepsilon-approximation} (in a L2L^2-sense) with a cost of the order ε2\varepsilon^{-2} or ε2logε3\varepsilon^{-2}|\log \varepsilon|^3 under less contractive assumptions. We then apply it to overdamped Langevin diffusions with strongly convex potential U:RdRU:\mathbb{R}^d\rightarrow\mathbb{R} and obtain an \textit{ε\varepsilon-complexity} of the order O(dε2log3(dε2)){\cal O}(d\varepsilon^{-2}\log^3(d\varepsilon^{-2})) or O(dε2){\cal O}(d\varepsilon^{-2}) under additional assumptions on UU. More precisely, up to universal constants, an appropriate choice of the parameters leads to a cost controlled by (λˉU1)2λU3dε2{(\bar{\lambda}_U\vee 1)^2}{\underline{\lambda}_U^{-3}} d\varepsilon^{-2} (where λˉU\bar{\lambda}_U and λU\underline{\lambda}_U respectively denote the supremum and the infimum of the largest and lowest eigenvalue of D2UD^2U). We finally complete these theoretical results with some numerical illustrations including comparisons to other algorithms in Bayesian learning and opening to non strongly convex setting.

Keywords

Cite

@article{arxiv.2109.07753,
  title  = {Multilevel-Langevin pathwise average for Gibbs approximation},
  author = {Maxime Egéa and Fabien Panloup},
  journal= {arXiv preprint arXiv:2109.07753},
  year   = {2024}
}
R2 v1 2026-06-24T06:01:09.697Z