Stratification by Fitness in Evolutionary Algorithms: Structural Theory and Experimental Results

We present the Fitness-Stratified Genetic Algorithm (FSGA), a diversity mechanism
that partitions the population into fitness-ranked strata and restricts competition to the offspring’s
destination stratum. This paper summarizes the current status of the project: a simplified analyt-
ical model, theoretical results already established, practical variants, and experimental evidence
on standard binary benchmarks. Current results include a mean-field selection equilibrium, con-
ditional runtime bounds on ONE M AX and JUMPm , within-generation diversity preservation, and
almost sure convergence. Empirically, crowding improves performance on rugged NK landscapes,
while stratum-dependent mutation rates accelerate the search on smooth landscapes. We also iden-
tify the main open problems needed to turn the current theory into a complete runtime analysis and
to bring the method back to combinatorial domains.