Cross entropy minimization in uninvadable states of complex populations (Q1179450)

Under recombination, phenotypic constraints, or frequency-dependent selection, maximization of average fitness is not a necessary outcome of selection. Therefore, evolutionary game theory is focused on phenotypes and considers equilibrium states characterized by stability rather than optimality. The paper presents a dynamical model adapted to evolutionary games that maximizes cross negentropy (``order stabilizing effects''). Let \((X,{\mathcal X})\) be the measurable space of strategies and \(\mathcal P\) a convex set of probability measures on it. Let \(\mathcal L\) be the (Banach) \(L\)-space of the experiment \((X,{\mathcal X},{\mathcal P})\) in the sense of \textit{L. Le Cam} [see ``Asymptotic methods in statistical decision theory'' (1986; Zbl 0605.62002)]. The replicator dynamics \(\mathcal R\) is defined on \(\mathcal L\) such that, if state \(Q\in{\mathcal P}\) favours strategies in a set \(A\in {\mathcal X}\), then the frequency of strategies belonging to \(A\) should increase with time, and vice-versa. Let \(P\in{\mathcal P}\) be a strongly uninvadable state (defined in the paper and compared with uninvadable and evolutionary stable states) and \((Q(t)\); \(0\leq t\leq T)\) a trajectory according to \(\mathcal R\) starting in a suitable vicinity (under the Kullback-Leibler distance \(K_{Q:P}\) of \(Q\) w.r.t. \(P\), which is not a metric) of \(P\). The author proves that the cross entropy \(K_{Q:P}\) decreases monotonically with time \(t\). The paper ends with special results for evolutionary mixed strategy games and its particular case of population games (where response depends on state \(P\) only through the average strategy of the population).