English

Learning Zero-Sum Linear Quadratic Games with Improved Sample Complexity and Last-Iterate Convergence

Systems and Control 2025-08-19 v4 Computer Science and Game Theory Machine Learning Systems and Control

Abstract

Zero-sum Linear Quadratic (LQ) games are fundamental in optimal control and can be used (i)~as a dynamic game formulation for risk-sensitive or robust control and (ii)~as a benchmark setting for multi-agent reinforcement learning with two competing agents in continuous state-control spaces. In contrast to the well-studied single-agent linear quadratic regulator problem, zero-sum LQ games entail solving a challenging nonconvex-nonconcave min-max problem with an objective function that lacks coercivity. Recently, Zhang et al. showed that an~ϵ\epsilon-Nash equilibrium (NE) of finite horizon zero-sum LQ games can be learned via nested model-free Natural Policy Gradient (NPG) algorithms with poly(1/ϵ)(1/\epsilon) sample complexity. In this work, we propose a simpler nested Zeroth-Order (ZO) algorithm improving sample complexity by several orders of magnitude and guaranteeing convergence of the last iterate. Our main results are two-fold: (i) in the deterministic setting, we establish the first global last-iterate linear convergence result for the nested algorithm that seeks NE of zero-sum LQ games; (ii) in the model-free setting, we establish a~O~(ϵ2)\widetilde{\mathcal{O}}(\epsilon^{-2}) sample complexity using a single-point ZO estimator. For our last-iterate convergence results, our analysis leverages the Implicit Regularization (IR) property and a new gradient domination condition for the primal function. Our key improvements in the sample complexity rely on a more sample-efficient nested algorithm design and a finer control of the ZO natural gradient estimation error utilizing the structure endowed by the finite-horizon setting.

Keywords

Cite

@article{arxiv.2309.04272,
  title  = {Learning Zero-Sum Linear Quadratic Games with Improved Sample Complexity and Last-Iterate Convergence},
  author = {Jiduan Wu and Anas Barakat and Ilyas Fatkhullin and Niao He},
  journal= {arXiv preprint arXiv:2309.04272},
  year   = {2025}
}
R2 v1 2026-06-28T12:16:09.148Z