SPEEDING UP LANGEVIN DYNAMICS BY MIXING

  • Alexander Christie
  • , Yuanyuan Feng*
  • , Gautam Iyer
  • , Alexei Novikov
  • *Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

Abstract

We study an overdamped Langevin equation on the d-dimensional torus with sta tionary distribution proportional to (Formula Presented). When U has multiple wells, the mixing time of the associated process is exponentially large (Formula Presented). We add a drift to the Langevin dynamics (without changing the stationary distribution) and obtain quantitative estimates on the mixing time. Our main result shows that the mixing time of the Langevin system can be made arbitrarily small by adding a drift that is sufficiently mixed. We provide one construction of a mixing drift, and our main result can be applied by using this drift with a large amplitude. For numerical purposes, it is useful to keep the size of the imposed drift small, and we show that the smallest allowable amplitude ensures that the mixing time is (Formula Presented), which is an order of magnitude smaller than (Formula Presented).

Original languageEnglish
Pages (from-to)1696-1743
Number of pages48
JournalMultiscale Modeling and Simulation
Volume23
Issue number4
DOIs
StatePublished - 2025

Keywords

  • Langevin Monte Carlo
  • enhanced dissipation
  • mixing

Fingerprint

Dive into the research topics of 'SPEEDING UP LANGEVIN DYNAMICS BY MIXING'. Together they form a unique fingerprint.

Cite this