Controlling chaos in distributed systems

A simple and robust procedure for freezing out chaotic behavior in systems composed of interacting agents making decisions based on imperfect and delayed information is described. It is based on a reward mechanism whereby the relative number of computational agents following effective strategies is increased at the expense of the others. This procedure, which generates a diverse population out of an essentially homogeneous one, is able to control chaos through a series of dynamical bifurcations into a stable fixed point. Stability boundaries are computed and the minimal amount of diversity required in the system is established