Greedy transformation of evolutionary algorithm search spaces for scheduling problems

Many scheduling algorithms search the space of possible solutions (schedules), but some instead search the space of permutations of the set of jobs, employing a greedy algorithm to map any such permutation to a schedule that can be evaluated by the fitness function. The search algorithm is thus simplified because knowledge about problem domain details is encapsulated in the greedy algorithm that constructs schedules, and the fitness function that evaluates them. The variety of types of algorithms for which this sort of "greedy transformation" has proven effective, and the range of successful applications, prompts us to look more closely at how such transformations may also make good solutions easier to find. In this paper we experimentally evaluate some characteristics of search spaces under greedy transformations as a first step toward understanding why this technique is effective.