On the Verification and Computation of Strong Nash Equilibrium

Computing equilibria of games is a central task in computer science. A large number of results are known for \emph{Nash equilibrium} (NE). However, these can be adopted only when coalitions are not an issue. When instead agents can form coalitions, NE is inadequate and an appropriate solution concept is \emph{strong Nash equilibrium} (SNE). Few computational results are known about SNE. In this paper, we first study the problem of verifying whether a strategy profile is an SNE, showing that the problem is in $\mathcal{P}$. We then design a spatial branch–and–bound algorithm to find an SNE, and we experimentally evaluate the algorithm.

💡 Research Summary

The paper addresses the verification and computation of Strong Nash Equilibrium (SNE), a solution concept that extends the classic Nash equilibrium (NE) by requiring resilience against coordinated deviations by any coalition of players. While NE guarantees that no single player can profitably deviate, SNE demands that no group of players can jointly change their strategies to make every member strictly better off, making it a stricter and often non‑existent equilibrium concept.

Verification Complexity
The authors first show that checking whether a given mixed‑strategy profile x is an SNE can be reduced to two sub‑problems: (1) verifying that x is an NE, which is already known to be polynomial‑time via linear constraints; and (2) confirming that for every coalition N′⊆N, the restriction of x to the members of N′ is weakly Pareto‑efficient when the strategies of the remaining players are fixed. To handle the second part, they construct a fictitious “agent i_f” that plays against the real agents. The expected utility of each real agent i under a candidate deviation can be expressed as a linear function of the mixed strategies of the coalition. The verification then becomes a min‑max problem where the fictitious agent chooses an action that maximizes the minimum utility gain across coalition members. By leveraging results from prior work on multi‑agent min‑max problems (specifically, when the number of actions of the maximizing agent is bounded by the number of players), they prove that this min‑max problem can be solved in polynomial time when the number of players n is a constant. Consequently, the overall SNE verification problem lies in P for fixed n, and the existence problem lies in NP.

Verification Algorithm
Algorithm 2 enumerates all non‑empty coalitions (2ⁿ − 1 of them) and, for each coalition, calls Algorithm 1 (the Pareto‑efficiency verifier). Algorithm 1 enumerates all joint supports of size n, solves a linear program derived from the min‑max formulation, and checks whether a Pareto‑dominating deviation exists. If any coalition admits such a deviation, the algorithm returns the coalition and a dominating mixed strategy; otherwise it confirms that x is an SNE. Because n is treated as a constant, the exponential factor is absorbed into a polynomial bound.

Computation via Spatial Branch‑and‑Bound
For actually finding an SNE, the authors design a spatial branch‑and‑bound framework specialized to two‑player games (though the ideas extend to more players). The procedure iterates between: (i) invoking an NE oracle (e.g., MIP‑Nash, a mixed‑integer programming formulation for NE) to obtain a candidate equilibrium; and (ii) applying the polynomial‑time SNE verification to test the candidate. If verification fails, the identified coalition and its Pareto‑dominating deviation are used to prune the current region of the strategy space, thereby reducing the search space for subsequent iterations. This alternating process continues until an SNE is found or the space is exhausted.

Experimental Evaluation
The authors observe that the widely used NE benchmark suite GAMUT is unsuitable for SNE testing because its instances either lack SNEs or admit only pure‑strategy SNEs, which are trivially easy to detect. To remedy this, they develop a generator that produces games where only mixed‑strategy SNEs exist. These instances involve two players with 6–10 actions each and payoff matrices constructed to ensure that any SNE must be mixed. Experiments show that the branch‑and‑bound algorithm converges significantly faster than a naïve exhaustive search, especially when the identified coalition yields a strong Pareto improvement, allowing large portions of the space to be discarded.

Mixed‑Strategy Multilateral Deviations
Through two illustrative examples (Figures 2 and 3), the paper demonstrates that restricting attention to pure‑strategy deviations is insufficient for SNE verification. In one example, a profile is not weakly Pareto‑efficient despite no pure‑strategy coalition improving both players; a mixed‑strategy deviation does. In another, a pure‑strategy deviation improves the sum of utilities but not both players simultaneously, while a mixed deviation does. These examples underline that SNE conditions cannot be captured by a finite set of linear constraints based solely on pure strategies, unlike NE.

Conclusions and Future Work
The paper makes three main contributions: (1) it establishes that SNE verification is polynomial‑time for a fixed number of players; (2) it proposes a practical branch‑and‑bound algorithm for computing SNEs in two‑player games; and (3) it clarifies the necessity of considering mixed‑strategy multilateral deviations in SNE analysis. Open directions include extending the algorithmic framework to games with more than two players, devising tighter finite‑constraint characterizations of SNE (if possible), and developing approximation methods for large‑scale or real‑time applications. Overall, the work bridges a theoretical gap in the complexity of SNE verification and provides a concrete computational tool for researchers studying cooperative deviations in strategic games.