Some Tractable Win-Lose Games

Determining a Nash equilibrium in a $2$-player non-zero sum game is known to be PPAD-hard (Chen and Deng (2006), Chen, Deng and Teng (2009)). The problem, even when restricted to win-lose bimatrix games, remains PPAD-hard (Abbott, Kane and Valiant (2005)). However, there do exist polynomial time tractable classes of win-lose bimatrix games - such as, very sparse games (Codenotti, Leoncini and Resta (2006)) and planar games (Addario-Berry, Olver and Vetta (2007)). We extend the results in the latter work to $K_{3,3}$ minor-free games and a subclass of $K_5$ minor-free games. Both these classes of games strictly contain planar games. Further, we sharpen the upper bound to unambiguous logspace, a small complexity class contained well within polynomial time. Apart from these classes of games, our results also extend to a class of games that contain both $K_{3,3}$ and $K_5$ as minors, thereby covering a large and non-trivial class of win-lose bimatrix games. For this class, we prove an upper bound of nondeterministic logspace, again a small complexity class within polynomial time. Our techniques are primarily graph theoretic and use structural characterizations of the considered minor-closed families.

💡 Research Summary

The paper investigates the computational complexity of finding Nash equilibria in two‑player win‑lose (0‑1) bimatrix games, a problem that remains PPAD‑hard even when restricted to this binary payoff setting. Building on earlier work that identified polynomial‑time solvable subclasses—very sparse games and planar games—the authors broaden the tractable landscape by focusing on graph‑theoretic properties of the underlying payoff graph.

A win‑lose game can be represented by a bipartite directed graph G = (R ∪ C, E), where rows R and columns C correspond to the pure strategies of the two players and an edge (r, c) indicates that the row player receives a payoff of 1 against column c (and symmetrically for the column player). The central insight is that the structural restrictions on G imposed by forbidden minors dramatically simplify the equilibrium computation.

The first major result concerns games whose payoff graph is K₃,₃‑minor‑free. By invoking Wagner’s theorem, the authors show that such graphs are either planar or can be made planar by removing a constant‑size set of edges. This near‑planarity enables a reduction of the equilibrium problem to a series of alternating‑path searches that can be carried out by a deterministic logspace machine with a unique accepting computation (the complexity class UL). The algorithm proceeds by nondeterministically guessing a candidate mixed strategy, verifying in logspace that no unilateral deviation yields a higher payoff, and using the uniqueness property to guarantee that the guessed strategy is indeed a Nash equilibrium.

The second contribution extends the analysis to a subclass of K₅‑minor‑free graphs. Using the Robertson‑Seymour structure theorem, the authors decompose any such graph into a planar “core” together with a bounded‑size collection of non‑planar “gadgets.” The planar core is handled by the same UL technique as above, while each gadget, because of its constant size, can be exhaustively explored within nondeterministic logspace (NL). By stitching together the solutions for the core and the gadgets at the separating vertices, the overall equilibrium can be found in NL.

A particularly striking achievement is the identification of a larger family that simultaneously excludes both K₃,₃ and K₅ as minors. The authors prove that any graph in this family must contain a small separator that splits the graph into planar components. They then apply a divide‑and‑conquer strategy: each component is solved using the UL algorithm, and the separator vertices are processed by a nondeterministic verification step that ensures consistency across components. The entire procedure stays within NL, demonstrating that even though the class contains graphs with both K₃,₃ and K₅ as minors, the equilibrium problem does not inherit the full PPAD‑hardness of the general case.

Technically, the paper leverages three key ideas: (1) bounded tree‑width of minor‑free graphs to compress dynamic‑programming tables into logarithmic space; (2) separator‑based decomposition that isolates the difficult non‑planar parts; and (3) a “proof‑certifiable” selection mechanism that allows a nondeterministic machine to guess a strategy and then verify its optimality using only logspace resources. These ideas collectively lower the known upper bound from polynomial time to unambiguous logspace (UL) for K₃,₃‑minor‑free games and to nondeterministic logspace (NL) for the broader minor‑closed families.

The paper concludes with several avenues for future work: extending the UL result to the full class of K₅‑minor‑free games, exploring other forbidden‑minor families such as K₄,₄, and adapting the structural approach to games with more than two payoff values. By demonstrating that sophisticated graph‑theoretic tools can dramatically reduce the computational burden of equilibrium computation, the work opens a promising line of research at the intersection of algorithmic game theory and graph minor theory.