Sharp High-dimensional Central Limit Theorems for Log-concave Distributions
Abstract
Let be i.i.d. log-concave random vectors in with mean 0 and covariance matrix . We study the problem of quantifying the normal approximation error for with explicit dependence on the dimension . Specifically, without any restriction on , we show that the approximation error over rectangles in is bounded by for some universal constant . Moreover, if the Kannan-Lov\'asz-Simonovits (KLS) spectral gap conjecture is true, this bound can be improved to . This improved bound is optimal in terms of both and in the regime . We also give -Wasserstein bounds with all and a Cram\'er type moderate deviation result for this normal approximation error, and they are all optimal under the KLS conjecture. To prove these bounds, we develop a new Gaussian coupling inequality that gives almost dimension-free bounds for projected versions of -Wasserstein distance for every . We prove this coupling inequality by combining Stein's method and Eldan's stochastic localization procedure.
Keywords
Cite
@article{arxiv.2207.14536,
title = {Sharp High-dimensional Central Limit Theorems for Log-concave Distributions},
author = {Xiao Fang and Yuta Koike},
journal= {arXiv preprint arXiv:2207.14536},
year = {2023}
}
Comments
37 pages. Some typos are corrected