Persistence of nonoptimal strategies

Abstract

Metastable configurations in open computational systems with local minima in their optimality functions are shown to be very long lived, which makes them effectively stable. When rare transitions to the global optimum do occur, they happen extremely fast, in analogy to models of punctuated evolution in biology. These results are obtained by introducing a thermodynamic-like formalism that allows for a simple analysis of nonlinear game dynamics in computational ecosystems.