Id: arxiv/2608.06155 Url: https://arxiv.org/abs/2608.06155 Tags: math.DS, cs.LG, math.NA, math.ST, stat.ML Score: 0 Title: Verifiable Regularity Criterion for Conditional Expectation Operators and Conditional Mean Embeddings with Applications to Nonparametric Regression, Bayesian Inverse Problems, and Koopman Operators Domain: math.DS Source: arxiv Authors: Maximiliano Hertel, Ilja Klebanov, Manuel Schaller, Karl Worthmann Reviews: 0 Subject: math.DS Category: research Comments: 0 Disputed: false Division: math Mentions: 0 Discussed: 0 Published: 2026-08-06 Page: https://delphi.science/r/arxiv/2608.06155 Conditional expectation operators (CEOs) and their associated conditional mean embeddings (CMEs) play a central role across applied mathematics and machine learning, appearing in nonparametric regression, Bayesian inverse problems, and Koopman operator theory. A fundamental question is when a CEO maps a function space on $\mathcal{Y}$ into a prescribed function space on $\mathcal{X}$, particularly a reproducing kernel Hilbert space (RKHS). We show that such mapping properties are characterized by the regularity of the Radon--Nikodym density of the conditional law, and establish a simple,…