{"id":"arxiv/2610.30109","url":"https://arxiv.org/abs/2610.30109","tags":["math.ST","cs.IT","cs.LG"],"score":0,"title":"Minimax Optimality of the NPMLE for Gaussian Mixtures in Hellinger Distance","domain":"math.ST","source":"arxiv","authors":["Shayan Hundrieser","Tudor Manole"],"reviews":0,"subject":"math.ST","summary":"We analyze the problem of estimating univariate Gaussian mixture densities using the nonparametric maximum likelihood estimator (NPMLE) based on $n$ independent and identically distributed samples. We show that the unconstrained NPMLE achieves the minimax optimal rate $\\log(n)/n$ in squared Hellinger distance, when the true mixing measure is sub-Gaussian. In addition, we show that when the true mixing measure has compact support, the NMPLE constrained to a compact interval also achieves the minimax optimal rate, which is slightly faster and of order $\\log(n)/(n \\log \\log n)$. Our proof…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-10-21","reactions":{}}