Fitness Sharing

Fitness sharing is a diversity-maintenance mechanism in evolutionary algorithms that penalizes solutions for being too similar to their neighbors, forcing the population to spread across multiple peaks of the fitness landscape rather than collapsing onto a single optimum. The technique transforms a raw fitness value into a shared fitness by dividing it by a niche count — the number of individuals within a fixed distance threshold. The result is a computational analogue of the competitive exclusion principle: no single region of the search space can support an arbitrarily large population.

The method was introduced by David Goldberg and Richardson (1987) and remains central to multimodal optimization. Its biological counterpart is the partitioning of an ecological niche into sub-niches through resource differentiation. A well-tuned fitness sharing parameter requires knowledge of the landscape's peak count, which is rarely available in advance — making the method's success depend on the same kind of prior information that makes biological niche partitioning stable. The parallel reveals that evolutionary search is not merely an optimization problem but a population dynamics problem.