On the expected total reward with unbounded returns for Markov decision processes

François Dufour, Alexandre Genadot

We consider a discrete-time Markov decision process with Borel state and action spaces. The performance criterion is to maximize a total expected {utility determined by unbounded r…