Application of genetic algorithms in the development of catalytic inorganic materials
Fritz-Haber-Institute of Max-Planck-Gesellschaft, Faradayweg 4-6, Berlin D-14195, Germany
This workshop contribution deals with various aspects of catalyst discovery and development by high-throughput experimentation. First, a randomised first generation of catalytic materials of different compositions is prepared; the choice of chemical elements leading to the various compositions is based on fundamental knowledge and intuition as well as accounting for the unexpected (serendipity). The first generation is then tested for catalytic performance. Mainly those materials that meet a set of objective functions, specifically activity, selectivity and/or yield for an individual catalytic reaction, are chosen for designing the second generation by applying a genetic algorithm (GA); the experimental procedure is then iteratively repeated until no further improvement of the objective functions is observed. In general, at least 5 to 10 generations each consisting of approximately 50 to 100 different catalytic materials are required for finding an optimal composition of the catalytic material.
By relating the output data of the objective function to the descriptors of the materials (i.e., composition and other properties) via an artificial neural network (ANN) insights in the relationships between catalytic performance and descriptors may be obtained. Furthermore, visualization of the results is facilitated.
Besides these basic principles in the application of a GA for catalytic-materials development also aspects of high-throughput experimentation are discussed, which are required for preparing and testing the large number of materials needed in the application of GA.
Presentation: Invited at E-MRS Fall Meeting 2007, Genetic algorithms for beginners, by Manfred Baerns
See On-line Journal of E-MRS Fall Meeting 2007
Submitted: 2007-05-25 23:08 Revised: 2009-06-07 00:44