Dynamic population games

In this paper, we define a new class of dynamic games played in large populations of anonymous agents. The behavior of agents in these games depends on a time-homogeneous type and a time-varying state, which are private to each agent and characterize their available actions and motifs. We consider finite type, state, and action spaces. On the individual agent level, the state evolves in discrete-time as the agent participates in interactions, in which the state transitions are affected by the agent's individual action and the distribution of other agents' states and actions. On the societal level, we consider that the agents form a continuum of mass and that interactions occur either synchronously or asynchronously, and derive models for the evolution of the agents' state distribution. We characterize the stationary equilibrium as the solution concept in our games, which is a condition where all agents are playing their best response and the state distribution is stationary. At least one stationary equilibrium is guaranteed to exist in every dynamic population game. Our approach intersects with previous works on anonymous sequential games, mean-field games, and Markov decision evolutionary games, but it is novel in how we relate the dynamic setting to a classical, static population game setting. In particular, stationary equilibria can be reduced to standard Nash equilibria in classical population games. This simplifies the analysis of these games and inspires the formulation of an evolutionary model for the coupled dynamics of both the agents' actions and states.