Blackwell Optimality in the Class of All Policies in Markov Decision Chains witha Borel State Space and Unbounded Rewards

摘要

This paper is the second part of the authors' study of Blackwell optimal policiesin Markov decision chains with a Borel state space and unbounded rewards. The authors prove that a stationary policy is Blackwell optimal in the class of all history-dependent policies if it is Blackwell optimal in the class of stationary policies. The authors also develop recurrence and drift conditions which ensure ergodicity and intergrability assumptions made in the previous paper, and which are more suitable for applications. As an example the authors study a cash-balance model.