{"id":"arxiv/2602.00474","url":"https://arxiv.org/abs/2602.00474","tags":["stat.ML","cs.LG","math.NA"],"score":0,"title":"Persistent-Transient Policy Evaluation for Markov Chains via Minimal Peripheral Quotients","domain":"stat.ML","source":"arxiv","authors":["Yang Xu","Vaneet Aggarwal"],"reviews":0,"subject":"stat.ML","summary":"We study fixed-policy evaluation for finite Markov chains that may be reducible and periodic. Classical evaluation methods with gain and bias decomposition are not always diagnostic: the gain records only invariant Cesàro averages, while persistent phase-dependent behavior is absorbed into the bias together with genuinely transient effects. We identify the real peripheral invariant subspace $\\mathcal{K}(P)$ of the transition matrix $P$ as the source of this ambiguity. Quotienting by $\\mathcal{K}(P)$ is the minimal exact quotient that removes all non-decaying modes and makes the remaining…","category":"research","comments":0,"disputed":false,"division":"stat","mentions":0,"discussed":0,"published":"2026-01-31","reactions":{}}