A convincing interpretation of mixed-strategy equilibria describes the
m as steady states in a large population in which players use pure str
ategies but the population as a whole mimics a mixed strategy. I study
the conditions under which an evolutionary, stochastic learning proce
ss converges to the appropriate distribution over pure strategies in t
he population. I find that not all mixed equilibria can be justified a
s the result of an evolutionary process even if the equilibrium is uni
que. For symmetric 2 x 2 and 3 x 3 games I give necessary and sufficie
nt conditions for convergence, which are related to the concept of an
ESS, and for n x n games I give a sufficient condition. (C) 1997 Acade
mic Press.