Strategy updating rules and strategy distributions in dynamical multiagent systems

摘要

In the evolutionary version of the minority game, agents update theirstrategies (gene-value $p$) in order to improve their performance. Motivated byrecent intriguing results obtained for prize-to-fine ratios which are smallerthan unity, we explore the system's dynamics with a strategy updating rule ofthe form $p \to p \pm \delta p$ ($0 \leq p \leq 1$). We find that the strategydistribution depends strongly on the values of the prize-to-fine ratio $R$, thelength scale $\delta p$, and the type of boundary condition used. We show thatthese parameters determine the amplitude and frequency of the the temporaloscillations observed in the gene space. These regular oscillations are shownto be the main factor which determines the strategy distribution of thepopulation. In addition, we find that agents characterized by $p={1 \over 2}$(a coin-tossing strategy) have the best chances of survival at asymptoticallylong times, regardless of the value of $\delta p$ and the boundary conditionsused.