Parity Games, Imperfect Information and Structural Complexity

Abstract

We address the problem of solving parity games with imperfect information on finite graphs of bounded structural complexity. It is a major open problem whether parity games with perfect information can be solved in PTIME. Restricting the structural complexity of the game arenas, however, often leads to efficient algorithms for parity games. Such results are known for graph classes of bounded tree-width, DAG-width, directed path-width, and entanglement, which we describe in terms of cops and robber games. Conversely, the introduction of imperfect information makes the problem more difficult, it becomes EXPTIME-hard. We analyse the interaction of both approaches. We use a simple method to measure the amount of "unawareness"' of a player, the amount of imperfect information. It turns out that if it is unbounded, low structural complexity does not make the problem simpler. It remains EXPTIME-hard or PSPACE-hard even on very simple graphs. For games with bounded imperfect information we analyse the powerset construction, which is commonly used to convert a game of imperfect information into an equivalent game with perfect information. This construction preserves boundedness of directed path-width and DAG-width, but not of entanglement or of tree-width. Hence, if directed path-width or DAG-width are bounded, parity games with bounded imperfect information can be solved in PTIME. For DAG-width we follow two approaches. One leads to a generalization of the known fact that perfect information parity games are in PTIME if DAG-width is bounded. We prove this theorem for non-monotone DAG-width. The other approach introduces a cops and robbers game (with multiple robbers) on directed graphs, considered Richerby and Thilikos forundirected graphs. We show a tight linear bound for the number of additional cops needed to capture an additional robber.