Id: arxiv/2606.27899 Url: https://arxiv.org/abs/2606.27899 Tags: math.ST, math.OC, math.PR Score: 0 Title: Optimal Estimators for Heavy-Tailed Mean Estimation via Convex Analysis Domain: math.ST Source: arxiv Authors: Bart P. G. van Parys, Bert Zwart Reviews: 0 Subject: math.ST Category: research Comments: 0 Disputed: false Division: math Mentions: 0 Discussed: 0 Published: 2026-06-26 Page: https://delphi.science/r/arxiv/2606.27899 We study optimal estimation of the location parameter of a distribution known only to lie in a symmetric moment class $\mathcal C_0$: the mean-zero distributions with bounded moment $\intφ\, d\mathbb P\le B$ for a fixed even $φ$. Our main result concerns the fixed-margin regime, where the error margin $Δ$ is fixed as $n\to\infty$: we give an exact large-deviation characterization of the smallest worst-case probability $β_n(Δ)$ of an error exceeding $Δ$ that any measurable estimator can guarantee with $n$ observations. Its exponential rate is exactly a two-point Hellinger exponent over the…