MULTIPOPULATION GENETIC ALGORITHM FOR SIMULATION OF THE CRYSTAL STRUCTURE FROM X-RAY DIFFRACTION DATA
Igor S Yakimov, Alexander Zaloga, Sergey Burakov, Eugeny С Semenkin
Abstract
A multipopulation genetic algorithm for a crystal structure solution from the X-ray powder diffraction data is proposed. Individual genetic algorithms are executed on different units of the computing cluster. The local optimization is performed periodically by the full-profile structure analysis (Rietveld method). The best trial structures are accumulated on the control unit for migration back to the routine compute units. The work of multi-population algorithm is discussed on 3 example of test structures with 9-10 independent atoms. The reliability of the structure search increases in a half order of magnitude more due to migration.