# Replicator Dynamics **Entity class:** Evolutionary game-theory model **Domain:** Population change / reinforcement learning **Maturity:** Developed ## Definition **Replicator dynamics** describe how the frequencies of strategies in a population change according to their relative performance against the population's current mix. A strategy performing better than the population average tends to gain share; one performing worse tends to lose share. The state is a distribution of strategies, not a list of individual intentions, and the payoffs can themselves change as that distribution changes. This makes the dynamics useful for [[wiki/Non-Transitive Games|cyclic games]], where a strategy's advantage can disappear as it becomes common. ## The reinforcement-learning bridge Börgers and Sarin showed that a specific reinforcement model, Cross learning, converges in a continuous-time limit to the replicator dynamics of evolutionary game theory. This is a correspondence under stated learning and limiting assumptions, not an identity between every algorithm that learns within a lifetime and natural selection across generations. Soma, Bouteiller, Hamann and Beltrame later used related replicator equations to connect local imitation rules with a swarm-level learning rule in a simplified nest-choice model, which is developed at [[wiki/Maynard-Cross Learning|Maynard-Cross Learning]]. The bridge is valuable precisely because the units and update rules remain identifiable: action probabilities in one interpretation, population proportions in another. [[wiki/Replicator|Replicator]] names a copying entity or pattern; **replicator dynamics** name a mathematical process governing the changing proportions of strategies. [[articles/Rock, Paper, Scissors|Rock, Paper, Scissors]] invokes the latter in its account of reinforcement learning, social populations and cyclic competition. ## Sources - [Börgers and Sarin, *Journal of Economic Theory* (1997)](https://doi.org/10.1006/jeth.1997.2319) — reinforcement learning and the continuous-time replicator limit. - [Bloembergen et al., *Journal of Artificial Intelligence Research* (2015)](https://doi.org/10.1613/jair.4818) — evolutionary analysis of multi-agent learning. - [Soma et al., *Swarm Intelligence* (2026)](https://link.springer.com/article/10.1007/s11721-026-00264-7) — local imitation, replicator dynamics and macro-level reinforcement learning.