Sequential versus parallel cooperative coevolutionary (1+1) EAs

Differences in computing environments often suggest variations in algorithms that are more natural for specific contexts, and that realize performance benefits as a result (e.g., sequential versus parallel computing environments). This is often true of traditional evolutionary algorithms, but is perhaps even more true of coevolutionary algorithms. This article analyzes two such variants of cooperative (1+1) EA. Using traditional run time analysis tools, it is proved that both variants are equivalent for separable objective functions and often have the same difficulties when facing some problems that are inseparable across population boundaries; however, very different performance on some inseparable functions can also be shown. This is done analytically and empirically using a carefully designed example problem.