Stochastic Games via Mathematical Programming

Abstract

This chapter is devoted to the introduction of the theory of finite state/action Stochastic Games. These games can be regarded as competitive Markov decision processes where there are two or more controllers, usually called players, whose fortunes are coupled either because the probability transitions are coupled or because their rewards are coupled, or both. It is assumed that the players have complete knowledge of these coupling functions but that they behave “noncooperatively,” that is, they choose their controls without any collusion and with the single-minded purpose of each maximizing her/his own payoff criterion.