English

Complexity Lower Bounds for Nonconvex-Strongly-Concave Min-Max Optimization

Optimization and Control 2021-04-20 v1 Machine Learning Machine Learning

Abstract

We provide a first-order oracle complexity lower bound for finding stationary points of min-max optimization problems where the objective function is smooth, nonconvex in the minimization variable, and strongly concave in the maximization variable. We establish a lower bound of Ω(κϵ2)\Omega\left(\sqrt{\kappa}\epsilon^{-2}\right) for deterministic oracles, where ϵ\epsilon defines the level of approximate stationarity and κ\kappa is the condition number. Our analysis shows that the upper bound achieved in (Lin et al., 2020b) is optimal in the ϵ\epsilon and κ\kappa dependence up to logarithmic factors. For stochastic oracles, we provide a lower bound of Ω(κϵ2+κ1/3ϵ4)\Omega\left(\sqrt{\kappa}\epsilon^{-2} + \kappa^{1/3}\epsilon^{-4}\right). It suggests that there is a significant gap between the upper bound O(κ3ϵ4)\mathcal{O}(\kappa^3 \epsilon^{-4}) in (Lin et al., 2020a) and our lower bound in the condition number dependence.

Keywords

Cite

@article{arxiv.2104.08708,
  title  = {Complexity Lower Bounds for Nonconvex-Strongly-Concave Min-Max Optimization},
  author = {Haochuan Li and Yi Tian and Jingzhao Zhang and Ali Jadbabaie},
  journal= {arXiv preprint arXiv:2104.08708},
  year   = {2021}
}

Comments

20 pages, 1 figure

R2 v1 2026-06-24T01:17:14.683Z