English

Solving Zero-Sum Games with Fewer Matrix-Vector Products

Optimization and Control 2025-09-05 v1 Data Structures and Algorithms Computer Science and Game Theory

Abstract

In this paper we consider the problem of computing an ϵ\epsilon-approximate Nash Equilibrium of a zero-sum game in a payoff matrix ARm×nA \in \mathbb{R}^{m \times n} with O(1)O(1)-bounded entries given access to a matrix-vector product oracle for AA and its transpose AA^\top. We provide a deterministic algorithm that solves the problem using O~(ϵ8/9)\tilde{O}(\epsilon^{-8/9})-oracle queries, where O~()\tilde{O}(\cdot) hides factors polylogarithmic in mm, nn, and ϵ1\epsilon^{-1}. Our result improves upon the state-of-the-art query complexity of O~(ϵ1)\tilde{O}(\epsilon^{-1}) established by [Nemirovski, 2004] and [Nesterov, 2005]. We obtain this result through a general framework that yields improved deterministic query complexities for solving a broader class of minimax optimization problems which includes computing a linear classifier (hard-margin support vector machine) as well as linear regression.

Keywords

Cite

@article{arxiv.2509.04426,
  title  = {Solving Zero-Sum Games with Fewer Matrix-Vector Products},
  author = {Ishani Karmarkar and Liam O'Carroll and Aaron Sidford},
  journal= {arXiv preprint arXiv:2509.04426},
  year   = {2025}
}

Comments

FOCS 2025

R2 v1 2026-07-01T05:21:39.867Z