Certifying Concavity and Monotonicity in Games via Sum-of-Squares Hierarchies

Vincent Leon, Iosif Sakos, Ryann Sim, Antonios Varvitsiotis

Keywords: concave games, monotone games, sum of squares, verifying convexity

TL;DR: We introduce SOS-concave and SOS-monotone games, which are subclasses of concave/monotone games that are verifiable via semidefinite programming.

Abstract: Concavity and its refinements underpin tractability in multiplayer games, where players independently choose actions to maximize their own payoffs which depend on other players’ actions. In *concave* games, where players' strategy sets are compact and convex, and their payoffs are concave in their own actions, strong guarantees follow: Nash equilibria always exist and decentralized algorithms converge to equilibria. If the game is furthermore *monotone*, an even stronger guarantee holds: Nash equilibria are unique under strictness assumptions. Unfortunately, we show that *certifying* concavity or monotonicity is NP-hard, already for games where utilities are multivariate polynomials and compact, convex basic semialgebraic strategy sets—an expressive class that captures extensive-form games with imperfect recall. On the positive side, we develop two hierarchies of sum-of-squares programs that certify concavity and monotonicity of a given game, and each level of the hierarchies can be solved in polynomial time. We show that almost all concave/monotone games are certified at some finite level of the hierarchies. Subsequently, we introduce the classes of SOS-concave/monotone games, which globally approximate concave/monotone games, and show that for any given game we can compute the closest SOS-concave/monotone game in polynomial time. Finally, we apply our techniques to canonical examples of extensive-form games with imperfect recall.

Primary Area: Theory (e.g., control theory, learning theory, algorithmic game theory)

Submission Number: 9275