Stochastic better-reply dynamics in finite games (Q926202)

The author studies a model where individuals from finite populations are repeatedly drawn to play a finite game and in every period choose a weakly better reply (not necessarily a best reply) to a sample distribution from a finite history of past play. It is shown that if the sample is small enough, the dynamics converges globally to sets of states that span MCUBR sets (minimal closed sets under better replies). In particular, there is convergence almost surely to a strict Nash equilibrium from any initial state if and only if all MCUBR sets are singleton.