{"id":"arxiv/2610.06308","url":"https://arxiv.org/abs/2610.06308","tags":["math.ST","cs.DS","cs.LG","math.NA"],"score":0,"title":"Sharp dimensional analysis of midpoint methods for Langevin sampling","domain":"math.ST","source":"arxiv","authors":["Fan Chen","Sinho Chewi","Jianfeng Lu","Matthew S. Zhang"],"reviews":0,"subject":"math.ST","summary":"We study deterministic and randomized midpoint discretizations of Langevin dynamics for a target $π\\propto e^{-V}$, where $0 \\prec αI\\preceq\\nabla^2V\\preceqβI$ and $κ=β/α$. To achieve $\\sqrtα\\,W_2\\leqslant\\varepsilon$, we show that deterministic Heun uses at most $\\widetilde O(κ^{4/3}d^{1/3}\\varepsilon^{-2/3})$ gradient queries, and underdamped exponential midpoint uses $\\widetilde O(κ^{5/4}d^{1/4}\\varepsilon^{-1/2})$. The proofs exploit cancellation at stationarity and smoothing using techniques from Malliavin calculus, outperforming previous upper bounds based on standard couplings. At…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-10-05","reactions":{}}