# Vortices Instead of Equilibria in MinMax Optimization: Chaos and Butterfly Effects of Online Learning in Zero-Sum Games

arXiv: https://arxiv.org/abs/1905.08396 · HTML: https://mlsanity.com/paper/1905.08396/

## Key facts

- Cited by 25 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 21 May 2019.

## Built on by (most cited first, 25 of 25)

- [Near-Optimal No-Regret Learning in General Games](https://mlsanity.com/paper/2108.06924/) (2021)
- [Survival of the strictest: Stable and unstable equilibria under regularized learning with partial information](https://mlsanity.com/paper/2101.04667/) (2021)
- [Autobidders with Budget and ROI Constraints: Efficiency, Regret, and Pacing Dynamics](https://mlsanity.com/paper/2301.13306/) (2023)
- [Neural Network Weights Do Not Converge to Stationary Points: An Invariant Measure Perspective](https://mlsanity.com/paper/2110.06256/) (2021)
- [Fast Last-Iterate Convergence of Learning in Games Requires Forgetful Algorithms](https://mlsanity.com/paper/2406.10631/) (2024)
- [Alternating Mirror Descent for Constrained Min-Max Games](https://mlsanity.com/paper/2206.04160/) (2022)
- [Reevaluating Policy Gradient Methods for Imperfect-Information Games](https://mlsanity.com/paper/2502.08938/) (2025)
- [Learning in Matrix Games can be Arbitrarily Complex](https://mlsanity.com/paper/2103.03405/) (2021)
- [Online Learning in Periodic Zero-Sum Games](https://mlsanity.com/paper/2111.03377/) (2021)
- [Solving Min-Max Optimization with Hidden Structure via Gradient Descent Ascent](https://mlsanity.com/paper/2101.05248/) (2021)
- [Lyapunov Exponents for Diversity in Differentiable Games](https://mlsanity.com/paper/2112.14570/) (2021)
- [Last-Iterate Convergence Properties of Regret-Matching Algorithms in Games](https://mlsanity.com/paper/2311.00676/) (2023)
- [Nash, Conley, and Computation: Impossibility and Incompleteness in Game Dynamics](https://mlsanity.com/paper/2203.14129/) (2022)
- [Online Optimization in Games via Control Theory: Connecting Regret, Passivity and Poincar'e Recurrence](https://mlsanity.com/paper/2106.04748/) (2021)
- [The Replicator Dynamic, Chain Components and the Response Graph](https://mlsanity.com/paper/2209.15230/) (2022)
- [The Update-Equivalence Framework for Decision-Time Planning](https://mlsanity.com/paper/2304.13138/) (2023)
- [Evolutionary Dynamics and Phi-Regret Minimization in Games](https://mlsanity.com/paper/2106.14668/) (2021)
- [On the Stability of Learning in Network Games with Many Players](https://mlsanity.com/paper/2403.15848/) (2024)
- [EMAgnet: Parameter-Space EMA Regularization for Policy Gradient Self-Play in Large Games](https://mlsanity.com/paper/2606.23995/) (2026)
- [Constants of Motion: The Antidote to Chaos in Optimization and Game Dynamics](https://mlsanity.com/paper/2109.03974/) (2021)
- [Chaos persists in large-scale multi-agent learning despite adaptive learning rates](https://mlsanity.com/paper/2306.01032/) (2023)
- [Data-Scarce Identification of Game Dynamics via Sum-of-Squares Optimization](https://mlsanity.com/paper/2307.06640/) (2023)
- [Last-iterate Convergence for Symmetric, General-sum, 2 × 2 Games Under The Exponential Weights Dynamic](https://mlsanity.com/paper/2502.08063/) (2025)
- [Langevin Multiplicative Weights Update with Applications in Polynomial Portfolio Management](https://mlsanity.com/paper/2502.19210/) (2025)
- [Local and Global Stability in Performative Reinforcement Learning](https://mlsanity.com/paper/2609.06467/) (2026)

## Abstract (arXiv)

We establish that algorithmic experiments in zero-sum games "fail miserably" to confirm the unique, sharp prediction of maxmin equilibration. Contradicting nearly a century of economic thought that treats zero-sum games nearly axiomatically as the exemplar symbol of economic stability, we prove that no meaningful prediction can be made about the day-to-day behavior of online learning dynamics in zero-sum games. Concretely, Multiplicative Weights Updates (MWU) with constant step-size is Lyapunov chaotic in the dual (payoff) space. Simply put, let's assume that an observer asks the agents playing Matching-Pennies whether they prefer Heads or Tails (and by how much in terms of aggregate payoff so far). The range of possible answers consistent with any arbitrary small set of initial conditions blows up exponentially with time everywhere in the payoff space. This result is robust both algorithmically as well as game theoretically: 1) Algorithmic robustness: Chaos is robust to agents using any of a general sub-family of Follow-the-Regularized-Leader (FTRL) algorithms, the well known regret-minimizing dynamics, even when agents mix-and-match dynamics, use different or slowly decreasing step-sizes. 2) Game theoretic robustness: Chaos is robust to all affine variants of zero-sum games (strictly competitive games), network variants with arbitrary large number of agents and even to competitive settings beyond these. Our result is in stark contrast with the time-average convergence of online learning to (approximate) Nash equilibrium, a result widely reported as "(weak) convergence to equilibrium".