Duality and Markovian strategies (Q1972604)

This paper is concerned with finitely repeated two player zero-sum stochastic games with lack of information on one side. Some properties of the optimal strategies and the values are obtained by using the dual game defined by \textit{B. De Meyer} [Math. Oper. Res. 21, 209-236 (1996; Zbl 0846.90142)]. It is proved that in the original game the uninformed player has an optimal behavior strategy that does not depend of his past own moves; moreover at each stage it depends only on the current stochastic state, on the stage and on a state variable that can be updated using its last value, the last move of the opponent and the current stochastic state. In the last section, some extensions are stated; in particular the result is extended to the case of games with incomplete information on both sides.