Wasserstein mixing time of the unadjusted Langevin algorithm

DOI10.48550/arXiv.2608.02430MaRDI QIDQ7284607arXivFDO

Authors Francesco Pedrotti, Peter A. Whalley

Publication date 3 August 2026

Published in arXiv

Tldr We provide new estimates in Wasserstein distance for the asymptotic bias of the unadjusted Langevin algorithm, implying an improved mixing time bound of $κ\sqrt{d}/ε$. Our analysis leverages techniques from transportation cost and concentration inequalities to derive a bound that improves upon previous work by a factor of √d/ε.

Full work available at URL https://doi.org/10.48550/arXiv.2608.02430

arXiv classification

cs.LGmath.NAmath.PRstat.CO

Cites work

• Analysis and geometry of Markov diffusion operators
• Analysis of Langevin Monte Carlo via convex optimization
• Asymptotic bias of inexact Markov chain Monte Carlo methods in high dimension
• Central limit theorem for additive functionals of reversible Markov processes and applications to simple exclusions
• Convergence of Unadjusted Langevin in High Dimensions: Delocalization of Bias
• Further and stronger analogy between sampling and optimization: Langevin Monte Carlo and gradient descent
• High-dimensional Bayesian inference via the unadjusted Langevin algorithm
• Isoperimetric inequalities in high-dimensional convex sets
• Log-concavity and strong log-concavity: a review
• Logarithmic bounds for isoperimetry and slices of convex sets
• Markov chains and stochastic stability
• Nonasymptotic convergence analysis for the unadjusted Langevin algorithm
• On the Lipschitz properties of transportation along heat flows
• Rapid convergence of the unadjusted Langevin algorithm: isoperimetry suffices
• Sampling as optimization in the space of measures: The Langevin dynamics as a composite optimization problem
• Spectral monotonicity under Gaussian convolution
• Sqrt(d) Dimension Dependence of Langevin Monte Carlo
• User-friendly guarantees for the Langevin Monte Carlo with inaccurate gradient

Cited in

(1)

• Theoretical guarantees for stochastic gradient Langevin dynamics

This page was built for publication: Wasserstein mixing time of the unadjusted Langevin algorithm

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7284607)