On the common population diversity measures in metaheuristics and their limitations

Abstract

Maintaining population diversity is one of the fundamental challenges for metaheuristic algorithms. With the emergence of adaptive and self-adaptive methods, the population diversity is frequently used as an indicator of the population state and feedback for the adaptive mechanism. In literature, several methods for quantification of the population diversity were proposed over the years. However, expressing the overall complexity of a metaheuristic population state by a single number inherently leads to simplification and distortion. As we show in this paper, lower diversity value does not automatically mean worse conditions for the emerging of new feasible solutions.