{"id":"arxiv/2611.13930","url":"https://arxiv.org/abs/2611.13930","tags":["math.ST","math.NA"],"score":0,"title":"Lengthscale Estimation for the Gaussian Kernel and Deterministic Response Functions","domain":"math.ST","source":"arxiv","authors":["Subhendu Pramanick","Toni Karvonen"],"reviews":0,"subject":"math.ST","summary":"The uncertainty estimates Gaussian processes output rely on the chosen covariance kernel and its hyperparameters. We study the fixed-domain asymptotics of the maximum likelihood and cross-validation estimators of the lengthscale parameter of the popular infinitely smooth Gaussian kernel when the observations are produced by a fixed univariate response function. No theoretical results for Gaussian lengthscale estimation exist in this setting. We prove lower bounds on the lengthscale estimators that depend on the regularity of the response function, as measured by the rate of decay of its…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-11-10","reactions":{}}