Abstract : We give a policy iteration algorithm to solve zero-sum stochastic games with finite state and action spaces and perfect information, when the value is defined in terms of the mean payoff per turn. This algorithm does not require any irreducibility assumption on the Markov chains determined by the strategies of the players. It is based on a discrete nonlinear analogue of the notion of reduction of a super-harmonic function.
https://hal.inria.fr/inria-00144146
Contributeur : Stephane Gaubert
<>
Soumis le : mardi 1 mai 2007 - 17:25:52
Dernière modification le : vendredi 25 mai 2018 - 12:02:03
Document(s) archivé(s) le : mercredi 7 avril 2010 - 03:30:50
Jean Cochet-Terrasson, Stéphane Gaubert. Policy iteration algorithm for zero-sum stochastic games with mean payoff. Comptes Rendus Mathématique, Elsevier Masson, 2006, 343 (5), pp.377-382. 〈10.1016/j.crma.2006.07.011〉. 〈inria-00144146〉
