A mathematical framework coupling individual optimal-control problems with a mean-field description of the population distribution, yielding a self-consistent equilibrium when each agent optimally responds to the population mean while the population evolves under the aggregate of those responses.

Principle

Combine Hamilton–Jacobi–Bellman equations (individual optimality) with a forward Kolmogorov/Fokker–Planck equation for the distribution; seek a fixed point where the optimal controls computed from the value function drive the distribution that generated those controls.

Demonstration

Modeling crowd motion in a corridor: each pedestrian minimizes travel time and discomfort given the density field, yielding a coupled HJB–Fokker–Planck system whose solution predicts evacuation patterns and congestion formation as a mean-field Nash equilibrium.

Misapplication

Using MFG when the population is small, highly heterogeneous, or when strategic interactions are dominated by a few strong links invalidates the mean-field limit and can produce misleading equilibria that ignore important finite-player effects.

Consequence

Provides tractable Nash-equilibrium approximations for very large populations, permits decentralized strategy synthesis and insight into aggregate effects of individual optimization, and can guide mechanism design when N is large.

Reversal

Finite-player differential games and centralized mean-field control differ: finite-N games retain idiosyncratic strategic effects, while mean-field control (central planner) solves a social optimization problem rather than a Nash equilibrium.

Boundary

Applies in the limit of many symmetric or weakly heterogeneous agents with negligible individual impact; excludes small-N strategic settings, strongly networked interactions, or problems where common noise and coarse heterogeneity break the mean-field assumption without modification.

Semantic Tension

Tension with mean-field control (centralized optimal control), agent‑based simulation, and multi-agent reinforcement learning: MFG focuses on decentralized equilibrium structure in the large-N limit rather than finite-agent optimal coordination or learning dynamics.

Synthesis

Mean-field games marry optimal control and kinetic-like population dynamics into a fixed-point problem: individual optimization and population evolution must be consistent, producing Nash-like equilibria that approximate strategic behavior in very large populations.