Cooperation and Phase Separation Driven by a Coevolving Snowdrift Game

We investigate the cooperative behavior and the phase separation in a coevolving system. Agents in the system constructed by a regular random network initially play the snowdrift game with their neighbors. They try to obtain a better competing environment by imitating a neighbor's more successful strategy or cutting the connection to a defective neighbor and randomly rewiring to another agent so as to seek a better neighborhood. The dynamic process of strategy imitation and relationship among agents due to rewiring neighbors may drive the system into different states. The simulation results show that there are three different phases in the q–r plane, where q is the rewiring probability and r is the cost-to-benefit ratio. One is a static phase of a pure cooperative cluster with a few isolated defectors. The other two belong to active phases with one of a main mixed-strategy cluster and the other of a pure defective state. We find that a simple mean field theory can predict correctly the static phase and the active phase of the main mixed-strategy cluster. The theoretical boundary line between the two phases is in good agreement with the simulation result.