Résumé | We study all-switches strategy improvement algorithms for parity, mean-payoff, discounted-payoff, and simple stochastic games. While these algorithms are now known to take exponential time in the worst case, we follow a recent line of work on the simplex method, and study them from a computational complexity point of view. We show that it is PSPACE-complete to decide the following problems: (1) given an edge e, will all-switches strategy improvement ever switch e? (2) given an edge e, is e in the optimal strategy found by all-switches strategy improvement? |