{"id":"arxiv/2610.00103","url":"https://arxiv.org/abs/2610.00103","tags":["math.PR","cs.DS"],"score":0,"title":"Exact Universality of Online Discrepancy","domain":"math.PR","source":"arxiv","authors":["Sunghyeon Jo","Taekyun Lee"],"reviews":0,"subject":"math.PR","summary":"We study online vector balancing with $N$ random vectors in $\\mathbb{R}^M$ revealed sequentially, where each vector must be assigned an irrevocable sign upon arrival. The goal is to minimize the expected $\\ell^\\infty$ norm of the final signed sum. For i.i.d. entries with mean zero, variance one, and a finite fourth moment, we prove that, as $M/N\\toα\\in(0,\\infty)$, the optimal value divided by $\\sqrt N$ converges to a limit $R_α$ independent of the entry distribution. This limit is the stochastic control value identified for Gaussian inputs by Fiedler, Jackson, Lacker, and Niles-Weed. In…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-09-08","reactions":{}}