Dimension-Independent Convergence of Underdamped Langevin Monte Carlo in KL Divergence

Shiyuan Zhang Qiwei Di Xuheng Li Quanquan Gu · 2026

Read Dimension-Independent Convergence of Underdamped Langevin Monte Carlo in KL Divergence by Shiyuan Zhang Qiwei Di Xuheng Li Quanquan Gu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-02429.txt.

Description

Shiyuan Zhang, Qiwei Di, Xuheng Li, and Quanquan Gu prove dimension-free sampling bounds in this 2026 paper. Langevin samplers draw from probability distributions defined by an energy function. The underdamped version adds momentum and works well in high dimensions in practice. Past guarantees, though, grew with dimension and became useless when dimension was huge.

Their result is the first KL divergence bound for discretized underdamped sampling with no dimension factor. The bound depends instead on the trace of a curvature bound for the energy. When that trace is far below the dimension, iteration costs improve sharply. The proof refines a local error framework into a dimension-free form. Overdamped methods look worse in exactly those favorable regimes. An open theoretical gap is now closed.

Catalog record

CREATORShiyuan Zhang Qiwei Di Xuheng Li Quanquan Gu
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE157256
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.02429

Browse the full library · About the Tower of Babel