Theory of evolution strategies - a tutorial

Evolution strategies form a class of evolutionary optimization procedures the behavior of which is comparatively well understood theoretically, at least for simple cases. An approach commonly adopted is to view the optimization as a dynamical process. Local performance measures are introduced so as to arrive at quantitative results regarding the performance of the algorithms. By employing simple fitness models and approximations exact in certain limits, analytical results can be obtained for multi-parent strategies including recombination and even for some simple strategies employing self-adaptation of strategy parameters. This tutorial outlines the basic algorithms and the methods applied to their analysis. Local performance measures and the most commonly used fitness models are introduced and results from performance analyses are presented. Basic working principles of evolution strategies — namely the evolutionary progress principle, the genetic repair principle, and the phenomenon of speciation — are identified and discussed. Finally, open problems and areas of future research are outlined.