Linear-Quadratic Mean Field Games with Common Noise: A Direct Approach

TL;DR

This paper develops a direct approach to solve linear-quadratic mean field games with common noise, reducing complexity with Riccati equations and establishing decentralized strategies as approximate Nash equilibria.

Contribution

It introduces a novel method to handle state coupling in mean field games with common noise, simplifying analysis via Riccati equations and solution properties.

Findings

Explicit decentralized strategies constructed for all players.

Strategies form an $\\epsilon$-Nash equilibrium.

Extended results to infinite-horizon case with algebraic Riccati equations.

Abstract

This paper investigates a linear-quadratic mean field games problem with common noise, where the drift term and diffusion term of individual state equations are coupled with both the state, control, and mean field terms of the state, and we adopt the direct approach to tackle this problem. Compared with addressing the corresponding mean field teams problem, the mean field games problem with state coupling presents greater challenges. This is not only reflected in the explosive increase in the number of adjoint equations when applying variational analysis but also in the need for more Riccati equations during decoupling the high-dimensional forward-backward stochastic differential equations system. We take a different set of steps and ingeniously utilize the inherent properties of the equations to address this challenge. First, we solve an $N$-player games problem within a vast and finite population setting, and obtain a set of forward-backward stochastic differential equations by variational analysis. Then, we derive the limiting forward-backward stochastic differential equations by taking the limit as $N$ approaches infinity and applying the law of large numbers. Based on the existence and uniqueness of solutions to backward stochastic differential equations, some variables in the equations are identically zero, which significantly reduces the complexity of the analysis. This allows us to introduce just two Riccati equations to explicitly construct decentralized strategies for all participants. Moreover, we demonstrate that the constructed decentralized strategies constitute an $\epsilon$-Nash equilibrium strategy for the original problem. We also extend the results to the infinite-horizon case and analyze the solvability of algebraic Riccati equations. Finally, numerical simulations are provided to illustrate the preceding conclusions.