English

Sharp High-dimensional Central Limit Theorems for Log-concave Distributions

Probability 2023-05-30 v4

Abstract

Let X1,,XnX_1,\dots,X_n be i.i.d. log-concave random vectors in Rd\mathbb R^d with mean 0 and covariance matrix Σ\Sigma. We study the problem of quantifying the normal approximation error for W=n1/2i=1nXiW=n^{-1/2}\sum_{i=1}^nX_i with explicit dependence on the dimension dd. Specifically, without any restriction on Σ\Sigma, we show that the approximation error over rectangles in Rd\mathbb R^d is bounded by C(log13(dn)/n)1/2C(\log^{13}(dn)/n)^{1/2} for some universal constant CC. Moreover, if the Kannan-Lov\'asz-Simonovits (KLS) spectral gap conjecture is true, this bound can be improved to C(log3(dn)/n)1/2C(\log^{3}(dn)/n)^{1/2}. This improved bound is optimal in terms of both nn and dd in the regime logn=O(logd)\log n=O(\log d). We also give pp-Wasserstein bounds with all p2p\geq2 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 pp-Wasserstein distance for every p2p\geq2. 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

R2 v1 2026-06-25T01:19:34.859Z