We consider the problem of computing stationary points in min-max optimization, with a particular focus on the special case of computing Nash equilibria in (two-)team zero-sum games. We first show that computing $\epsilon$-Nash equilibria in $3$-player \emph{adversarial} team games -- wherein a team of $2$ players competes against a \emph{single} adversary -- is \textsf{CLS}-complete, resolving the complexity of Nash equilibria in such settings. Our proof proceeds by reducing from \emph{symmetric} $\epsilon$-Nash equilibria in \emph{symmetric}, identical-payoff, two-player games, by suitably leveraging the adversarial player so as to enforce symmetry -- without disturbing the structure of the game. In particular, the class of instances we construct comprises solely polymatrix games, thereby also settling a question left open by Hollender, Maystre, and Nagarajan (2024). We also provide some further results concerning equilibrium computation in adversarial team games. Moreover, we establish that computing \emph{symmetric} (first-order) equilibria in \emph{symmetric} min-max optimization is \textsf{PPAD}-complete, even for quadratic functions. Building on this reduction, we further show that computing symmetric $\epsilon$-Nash equilibria in symmetric, $6$-player ($3$ vs. $3$) team zero-sum games is also \textsf{PPAD}-complete, even for $\epsilon = \text{poly}(1/n)$. As an immediate corollary, this precludes the existence of symmetric dynamics -- which includes many of the algorithms considered in the literature -- converging to stationary points. Finally, we prove that computing a \emph{non-symmetric} $\text{poly}(1/n)$-equilibrium in symmetric min-max optimization is \textsf{FNP}-hard.
翻译:我们考虑在极小极大优化中计算平稳点的问题,特别关注计算(双)团队零和博弈中纳什均衡这一特例。我们首先证明,在3人对抗性团队博弈——其中由2名玩家组成的团队与单个对抗者竞争——中计算ε-纳什均衡是CLS完全的,从而解决了此类设定下纳什均衡的计算复杂性。我们的证明通过从对称、相同收益的双人博弈中的对称ε-纳什均衡进行归约来实现,通过恰当地利用对抗性玩家来强制对称性,同时不破坏博弈的结构。特别地,我们所构造的实例类仅包含多矩阵博弈,从而也解决了Hollender、Maystre和Nagarajan(2024)遗留的一个问题。我们还提供了一些关于对抗性团队博弈中均衡计算的进一步结果。此外,我们证明了即使在二次函数情形下,计算对称极小极大优化中的对称(一阶)均衡也是PPAD完全的。基于此归约,我们进一步证明,即使在ε = poly(1/n)的情况下,计算对称的6人(3对3)团队零和博弈中的对称ε-纳什均衡也是PPAD完全的。作为一个直接推论,这排除了存在对称动力学——包括文献中考虑的许多算法——收敛到平稳点的可能性。最后,我们证明在对称极小极大优化中计算一个非对称的poly(1/n)-均衡是FNP难的。