Games on graphs: A minor modification of payoff scheme makes a big difference

Abstract

Various social dilemma games that follow different strategy updating rules have been studied on many networks.The reported results span the entire spectrum, from significantly boosting,to marginally affecting,to seriously decreasing the level of cooperation.Experimental results that are qualitatively different from theoretical prediction have also been reported.It is widely believed that the results are largely determined by three elements,including payoff matrices of the underlying 2*2 games,the way that the strategic states of the players are updated and the structure of the networks.Here we discuss the impact of a seemly non-essential mechanism -- what we refer to as a "payoff scheme". Specifically, in each round after the states of all of the players are determined,the payoff scheme is how each player's payoff is calculated.In addition to the two conventions in which either the accumulated or the averaged payoff is calculated from playing with all of the neighboring players,we here study the effects of calculating the payoff from pairing up with one random player from among the neighboring players. Based on probability theory, in a situation of uncorrelated events, the average payoff that involves all of the neighbors should,in principal,be equivalent to the payoff from pairing up with one neighbor.However,our simulation of games on graphs shows that, in many cases,the two payoff schemes lead to qualitatively different levels of cooperation.This finding appears to provide a possible explanation for a wide spectrum of observed behaviors in the literature.We have also observed that results from the randomly-pairing-one mechanism are more robust than the involving-all-neighbours mechanism because,in the former case, neither the other three main elements nor the initial states of the players have a large impact on the final level of cooperation compared with in the latter case.