Drift analysis

Drift analysis, introduced to the field of evolutionary computation by He and Yao ten years ago, quickly became one of the strongest tools to prove upper and lower bounds on the run-times of evolutionary algorithms. It has, however, the reputation of being difficult to use, both because it relies on deeper mathematical tools and because it needs a clever guess of a potential function. In this tutorial, after presenting the classical results, I will focus on the recently developed multiplicative drift analysis method. It often is easier to employ and yields stronger results, e.g., run-time bounds that hold with high probability. I will end with a number of open problems of different difficulties. The intended audience of the tutorial has some basic experience in theory, though no particular prerequisites are required.