Martin Schütz has developed a software package for nonlinear, mixed-integer optimization based on genetic algorithms and evolution strategies.
Multilayer coatings are of interest in a variety of applications where filters for electromagnetic radiation with a specific reflection and absorption behaviour are required. As a simple example in optics, consider the problem of designing a multilayer coating which reflects heat radiation while letting through all visible light radiation - light bulbs designed according to this principle would not emit heat but just the light. More important applications exist in medicine technique, radar and sonar systems.
In cooperation with an industrial customer (Philips GmbH, Germany), Martin Schütz has developed a software tool that combines the adaptive learning capabilities of genetic algorithms and those of evolution strategies in order to achieve a performance (measured in terms of the error between desired and actual reflection behaviour) better than existing commercial software tools for multilayer coating design.
Tests on a reference problem from literature (and problems of particular interest to the customer) yielded coatings of very high quality for a wide range of different optical thicknesses of the filter design (corresponding to the number of layers in the filter). Moreover, the method does not need, unlike many other methods, an initial design provided by an expert, but synthesizes a high-quality design completely from scratch.