Experimental Investigation of Scatter Search for Graph Coloring

Scatter search [9, 10] is an evolutionary approach related to the tabu search metaheuristic [8]. It is based on strategies proposed in the 1960s for combining decision rules and constraints. This approach has only been applied recently to a few optimization problems. Applications of this method include, for instance, vehicle routing [20] and unconstrained optimization [17]; see also [9] for more references. Like other population-based methods, scatter search uses combination of configurations to generate new configurations which can replace others in the population. But the way combinations and replacements are made differs from the traditional strategies used in pure genetic algorithms. Combinations operate on multiple parents, and replacements rely not only on the improvement of an objective function but also on the improvement of the population diversity. Furthermore, scatter search works with a small set of configurations, usually less than twenty, and uses deterministic heuristics as much as possible in place of randomization to make a decision.