Crystal Structure Prediction based on Statistical Potentials

Organic molecule crystals are becoming more and more important in applications like piezoelectricity, ferroelectricity and pigments. These properties depend on the molecule and on the crystal structure. For this reason much effort is being made to predict the crystal structure of organic molecules. We have developed a new algorithm differing mainly in three features from other approaches (simulated annealing, Monte Carlo etc.). First, we analyze just one molecule for proper symmetry operations building up the crystal; second, the program works in a discrete space; and finally the scoring function (energy function) is derived statistically from known crystal structures and tabulated. Our program computes a list of crystal structures weighted according to our scoring function. The new algorithm FlexCryst is currently implemented for the four space groups P1, Pbar1, P2₁, and P2₁2₁2₁. The three latter space groups are widespread in nature. The algorithm computes structural models of acceptable quality and shows excellent time performance. During our validation we found the experimental structure among the structures proposed by the algorithm in 123 of 129 cases for P1, in 66 of 95 cases for Pbar1, in 73 of 100 cases for P2₁, and in 94 of 98 cases for P2₁2₁2₁. The performance depends on the space group. In the case of P1 the run time per molecule is about two minutes and increases up to roughly one hour for the space group P2₁.