Studies And Improvements Of CALYPSO Structure Prediction Method

The information of structures occupied a critical role in understanding thephysical and chemical properties of materials. Experimentally, severaltechniques (e.g., X-ray diffraction and neutron diffraction) can be used forstructural determination. These techniques have been developed extremely well.However, it happens frequently that experiments fail to identify the structuredue to low-quality diffraction data, particularly at extreme conditions (e.g.,high pressure). Therefore, the development of theoretical structure predictionmethod is greatly necessary. Besides, theoretical structure prediction not onlycan assist experiment to determine structures of matters, but also can predictstructures with novel properties to be helpful to experimental synthesis.Theoretical structure prediction with only known information of chemicalcomposition becomes a hot topic. Recently, several structure predictionmethods have been proposed. The methodology for structural prediction(CALYPSO) based on Particle Swarm Optimization (PSO) has been proposedby our group. Due to high efficiency, the CALYPSO structural predictionmethod is one of the most efficient structure prediction methods. In this thesis,some improvement strategies were proposed on the basis of previouslydeveloped CALYPSO method.1. It is low efficiency to stochastically generate initial structures for largesystem because of unreasonable atomic distance of totally random structures.To solve aforementioned problem, a new method for producing initialstructures was proposed in this thesis. This method is combination of smallstructural unit to large structural unit. The tests illustrate that the technique isvery powerful to find the global structure of large system. 2. The global variant of PSO promotes exploitation since all particles areattracted by the same best position, thereby, converging faster. While localvariant of PSO has better exploration properties. For purpose of utilizingadvantages of both global PSO and local PSO, we implemented the newalgorithm combining the exploration and exploitation properties of both thelocal and global PSO variants to CALYPSO in this thesis.3. Artificial Bee Colony (ABC) algorithm as an emerging swarm intelligentalgorithm firstly had been applied to function optimization problems byKaraboga D. Currently, it has been widely applied in many fields (e.g.,analogue simulation and combinatorial pptimization). Here, we implemented itto structure prediction and the test results indicate that ABC algorithm isefficient for structural determination.