A Potential Weakness Of The Genetic Algorithm In Multimodal Optimizations

The genetic algorithms are robust and stable evolutionary computation techniques in the sense that they mostly provide a solution. This fact is undisputedly an important positive feature, particularly in the situation when no feasible solution is available. On the other hand, this robustness and the fact that living nature utilizes analogous mechanisms for its development and propagation sometimes lead to unrealistic expectations and exaggerated faith in their capabilities. In particular, when they are used as function optimizers they sometimes tend to demonstrate also their weak points. In this article, we demonstrate the limits of the genetic algorithm in a multimodal model task. The task consists in the identification of a de Bruijn sequence of a given length between general binary sequences. Generally, the multimodality of the fitness function can lead to the stagnation of the genetic algorithm well below the desirable fitness function values which hinders the genetic algorithm from finding the optimized solution of the task under consideration.