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

TL;DR

This paper develops sum-of-squares hierarchies to certify concavity and monotonicity in complex multiplayer games, providing polynomial-time methods to verify these properties and approximate games, despite the NP-hardness of certification.

Contribution

It introduces sum-of-squares hierarchies for certifying concavity and monotonicity in polynomial games, and proposes methods to approximate and analyze such games efficiently.

Findings

Sum-of-squares hierarchies can certify concavity and monotonicity in polynomial games.

Almost all concave/monotone games are certified at some finite hierarchy level.

Polynomial-time algorithms approximate the closest SOS-concave/monotone game to any given game.

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 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 imperfect recall extensive-form games.