Design and analysis of an asymmetric mutation operator

Evolutionary algorithms as general randomized search heuristics typically perform a random search that is biased only by the fitness of the search points encountered. In practical applications the use of biased variation operators suggested by problem-specific knowledge may speed-up the search considerably. Problems defined over bit strings of finite length often have the property that good solutions have only very few one-bits or very few zero-bits. One specific mutation operator that is tailored towards such situations is defined and analyzed. The assets and drawbacks of this mutation operator are discussed. This is done by presenting analytical results on illustrative example functions as well as on function classes