On the tractability of Nash equilibrium

Ioannis Avramopoulos

2024-03-10

Abstract:In this paper, we propose a method for solving a PPAD-complete problem [Papadimitriou, 1994]. Given is the payoff matrix $C$ of a symmetric bimatrix game $(C, C^T)$ and our goal is to compute a Nash equilibrium of $(C, C^T)$. In this paper, we devise a nonlinear replicator dynamic (whose right-hand-side can be obtained by solving a pair of convex optimization problems) with the following property: Under any invertible $0 < C \leq 1$, every orbit of our dynamic starting at an interior strategy of the standard simplex approaches a set of strategies of $(C, C^T)$ such that, for each strategy in this set, a symmetric Nash equilibrium strategy can be computed by solving the aforementioned convex mathematical programs. We prove convergence using results in analysis (the analytic implicit function theorem), nonlinear optimization theory (duality theory, Berge's maximum principle, and a theorem of Robinson [1980] on the Lipschitz continuity of parametric nonlinear programs), and dynamical systems theory (such as the LaSalle invariance principle).

Optimization and Control,Computer Science and Game Theory

What problem does this paper attempt to address?

This paper aims to solve the problem of computing Nash equilibria in symmetric bimatrix games. Specifically, the paper proposes a method to solve a PPAD - complete problem (Papadimitriou, 1994), that is, given the payoff matrix \(C\) of the symmetric bimatrix game \((C, C^T)\), the goal is to compute a Nash equilibrium of this game.

### Main contributions of the paper

1. **Proposed method**:
- The author designs a nonlinear replicator dynamics equation, and its right - hand side can be obtained by solving a pair of convex optimization problems.
- This dynamic equation has the following property: under any condition of \(0 < C \leq 1\), every trajectory starting from an interior strategy of the standard simplex will approach a set of strategies, and each of these strategies can be used to compute a symmetric Nash equilibrium strategy by solving the above - mentioned convex optimization problems.

2. **Convergence proof**:
- The author uses analysis (the analytic implicit function theorem), nonlinear optimization theory (duality theory, Berge's maximum principle, Robinson's Lipschitz continuity theorem on parametric nonlinear programming) and dynamical systems theory (such as LaSalle's invariance principle) to prove convergence.

3. **Technical details**:
- The paper details how to construct optimization problems through the relative entropy function and a series of constraint conditions, and proves the relationship between the solutions of these optimization problems and Nash equilibria.
- The author also discusses the connection between this method and other related works, such as the Multiplicative Weights Update Method and the Lemke - Howson algorithm.

### Key terms

- **Nash Equilibrium**: In a multi - player game, if no player can obtain a better payoff by unilaterally changing their strategy, then this combination of strategies is called Nash equilibrium.
- **Symmetric Bimatrix Game**: A special bimatrix game in which the payoff matrices of the two players are transposes of each other.
- **Replicator Dynamics**: A dynamical model that describes the change in the frequency of strategies in a population.
- **Convex Optimization Problem**: A class of optimization problems whose objective function and constraint conditions are convex, having good theoretical properties and efficient solution algorithms.

### Conclusion

By proposing a new nonlinear replicator dynamics equation, the paper successfully solves the problem of computing Nash equilibria in symmetric bimatrix games. This method is not only of great theoretical significance but also provides new tools and ideas for practical applications.