{"id":"arxiv/2608.30221","url":"https://arxiv.org/abs/2608.30221","tags":["math.ST"],"score":0,"title":"Cubic-Root Gaussian Approximation under Unrestricted Covariance","domain":"math.ST","source":"arxiv","authors":["Zijun Gao","Weihan Zhang"],"reviews":0,"subject":"math.ST","summary":"For Gaussian approximation over high-dimensional rectangles under unrestricted covariance, Chernozhukov et al. (2023b) conjectured that the $n^{-1/4}$ rate, up to logarithmic factors, is near-optimal. We show that, under the coordinatewise subexponential condition with scale $B_n$ and the marginal variance lower bound condition with constant $b$ in Chernozhukov et al. (2023b), the approximation error in dimension $d$ is bounded by \\begin{align*} C_b\\min\\left\\{ 1,\\, \\left(\\frac{B_n^2}{n}\\right)^{1/3}\\{\\log(2dn)\\}^{7/3} + \\frac{B_n}{\\sqrt n}\\{\\log(2dn)\\}^{5/2} \\right\\}. \\end{align*} In…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-08-31","reactions":{}}