Genetic and Random Search Methods in Optimal Shape Design Problems

We describe the application of two global optimization methods, namely of genetic and random search type algorithms in shape optimization. When the so-called fictitious domain approaches are used for the numerical realization of state problems, the resulting minimized function is non-differentiable and stair-wise, in general. Such complicated behaviour excludes the use of classical local methods. Specific modifications of the above-mentioned global methods for our class of problems are described. Numerical results of several model examples computed by different variants of genetic and random search type algorithms are discussed.