Application of genetic algorithms to model the structure of molecular crystals (in English)
Polimery 2014, No 7-8, 542
In the case of complex organic molecules, very often happens that the powder diffraction pattern shows only few crystalline reflections and therefore powerful methods like the Rietveld procedure cannot work properly. Then in order to find the model of the structure, for which the calculated diffraction pattern shows an acceptable agreement with the experimental data, one may use one of the "artificial intelligence” techniques. Among them, the genetic algorithms (GA) method is one of the most widely used in a variety of scientific problems. Our idea was to use the unit cell (with its atoms content) as a phenotype of the GA. Each individual represents a random crystal structure. Its genotype contains: lattice constants: a, b and c, lattice angles: a, b, and g, and orientation of molecule vs. the crystallographic axes. Its diffraction pattern is calculated and compared with the reference data. We have prepared a computer program named CrystalFinder. In its present shape, the CrystalFinder program is able to reproduce the crystalline structure of simple "toy” molecular crystals built of molecules containing up to ca twenty atoms. Usually 200—300 generations are sufficient to get a very good agreement between the reference diffraction pattern and the diffractogram obtained by the program as the best fitted individual. It follows that genetic algorithms can be used to model the structure of molecular crystals and more complex systems containing macromolecules.