{"id":"arxiv/2602.05259","url":"https://arxiv.org/abs/2602.05259","tags":["math.ST","stat.ML"],"score":0,"title":"An Asymptotic Law of the Iterated Logarithm for $\\mathrm{KL}_{\\inf}$","domain":"math.ST","source":"arxiv","authors":["Ashwin Ram","Aaditya Ramdas"],"reviews":0,"subject":"math.ST","summary":"The population $\\mathrm{KL}_{\\inf}$ is a fundamental quantity that appears in lower bounds for (asymptotically) optimal regret of pure-exploration stochastic bandit algorithms, and optimal stopping time of sequential tests. Motivated by this, an empirical $\\mathrm{KL}_{\\inf}$ statistic is frequently used in the design of (asymptotically) optimal bandit algorithms and sequential tests. While nonasymptotic concentration bounds for the empirical $\\mathrm{KL}_{\\inf}$ have been developed, their optimality in terms of constants and rates is questionable, and their generality is limited (usually to…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-02-05","reactions":{}}