Evolutionary crystal structure prediction proved to be a powerful approachfor studying a wide range of materials. Here, we present a specificallydesigned algorithm for the prediction of the structure of complex crystalsconsisting of well-defined molecular units. The main feature of this newapproach is that each unit is treated as a whole body, which drasticallyreduces the search space and improves the efficiency, but necessitates theintroduction of new variation operators described here. To increase diversityof the population of structures, the initial population andpart($\scriptsize{\sim}$20%) of the new generations are generated using spacegroup symmetry combined with random cell parameters and random positions andorientations of molecular units. We illustrate the efficiency and reliabilityof this approach by number of tests (ice, ammonia, carbon dioxide, methane,benzene, glycine and butane-1,4-diammonium dibromide). This approach easilypredicts the crystal structure of methane \emph{A} containing 21 methanemolecules (105 atoms) per unit cell. We demonstrate that this new approach hasalso a high potential for the study of complex inorganic crystals on theexamples of a complex hydrogen storage material Mg(BH$_4$)$_2$ and elementalboron.
展开▼