Mean Field Games

Mean field games are a class of game-theoretic models describing strategic interactions among a continuum of infinitesimal agents, each of whom is too small to influence the aggregate behavior of the population. Introduced independently by Lasry and Lions and by Huang, Malhamé, and Caines in 2006, the framework replaces the complex interactions of many agents with the interaction of a single representative agent and the mean field — the statistical distribution of all other agents' states.

The simplification is powerful. In a mean field game, each agent solves an optimal control problem where the dynamics depend on the agent's own state and the population distribution, but not on any individual opponent's strategy. The equilibrium condition requires that the population distribution evolves consistently with the optimal strategies of the agents — a fixed-point condition that couples forward-backward equations.

Mean field games have been applied to traffic flow, energy markets, social dynamics, and multi-agent reinforcement learning. The connection to statistical mechanics is structural: both replace microscopic interactions with macroscopic fields. The key difference is strategic behavior — agents in mean field games optimize, while particles in statistical mechanics do not. See propagation of chaos for the mathematical conditions under which the mean field approximation is valid.