The reputation-based model with the three archetypal games.

The density of cooperators, ρC, A, the density of soft strategies in game B, ρd, B, the normalized payoff difference of cooperators and defectors in game B, Δπ, B, and the correlation between the strategy of the individuals in the two games, 〈sAsB〉c, D, as a function of the probability of error in inferring the PD strategy of the opponent, η. Here, from top to bottom, the game B is the Snow Drift, the Battle of the Sexes, and the Leader game. The payoff values used for the games are presented in Table 1. The lines show the result of the replicator-mutator dynamics, and the markers show the results of simulations. The solid blue line shows the equilibrium fixed point, which occurs starting from an unbiased initial condition in which the density of all the strategies are equal, and the dashed red line shows the non-equilibrium fixed point, which can occur for certain initial conditions. The system is bistable for small recognition noise, η and both a cooperative fixed point with a high level of cooperation A and soft strategies B, and a defective fixed point with a low level of cooperation and soft strategies are possible. In the cooperative phase, cooperators receive a higher payoff from game B C. Moreover, the strategies of individuals show an anti-correlation in the cooperative fixed point D, resulting from the fact that defectors play softly with cooperators and cooperators play hard with defectors in this fixed point. For the simulations, a sample of 80 simulations, in a population of size N = 10000 is used. The simulations start from random initial conditions. In each simulation, the dynamics settle in one of the two fixed points. The markers show the averages, and the error bars show the standard deviation in the sample of simulations that settle in the given fixed point. The size of the markers is proportional to the number of times that the given fixed point occurs. Here, the mutation rate, ν = 0.005. The simulations are run for 20000 time steps, and an average over the last 1000 time steps is taken.