Learning Purified Mixed Equilibria

better understand when mixed equilibria might arise within populations of interact acting agents, we examine a model of smoothed fictitious play that is designed to capture Harsanyi's "Purification", view of mixed equilibria in a setting with a large population of agents. Our analysis concerns the local stability of equilibria when the degree of heterogeneity in the population is small. In 2 x 2 games our model is easy to analyze and yields the same conclusions as have previous models. Our primary focus is on 3 x 3 games where we provide a general characterization of which equilibria are locally stable, and discuss its implications in several particular cases. Among our conclusions are that learning can sometimes provide a justification for mixed equilibria outside of 2 x 2 games, that whether an equilibrium is stable or unstable is often dependent on the distribution of payoff heterogeneity in the population, that the totally mixed equilibria of zero sum games are generically stable, and that under a "balanced perturbation" condition the equilibria of symmetric games are generically unstable.