First-order algorithm with ${\mathcal{O}({\rm ln}(1{/}\epsilon))}$ convergence for ${\epsilon}$-equilibrium in two-person zero-sum games

Abstract

We propose an iterated version of Nesterov’s first-order smoothing method for the two-person zero-sum game equilibrium problem

$$\min_{x \in Q_1}\max_{y \in Q_2} {x}^{\rm T}{Ay} = \max_{y \in Q_2} \min_{x \in Q_1} {x}^{\rm T}{Ay}.$$

This formulation applies to matrix games as well as sequential games. Our new algorithmic scheme computes an ${\epsilon}$-equilibrium to this min-max problem in ${\mathcal {O}\left(\frac{\|A\|}{\delta(A)} \, {\rm ln}(1{/}\epsilon)\right)}$ first-order iterations, where δ(A) is a certain condition measure of the matrix A. This improves upon the previous first-order methods which required ${\mathcal {O}(1{/}\epsilon)}$ iterations, and it matches the iteration complexity bound of interior-point methods in terms of the algorithm’s dependence on ${\epsilon}$. Unlike interior-point methods that are inapplicable to large games due to their memory requirements, our algorithm retains the small memory requirements of prior first-order methods. Our scheme supplements Nesterov’s method with an outer loop that lowers the target ${\epsilon}$ between iterations (this target affects the amount of smoothing in the inner loop). Computational experiments both in matrix games and sequential games show that a significant speed improvement is obtained in practice as well, and the relative speed improvement increases with the desired accuracy (as suggested by the complexity bounds).