Predicting the crystal structure of organic molecular materials

Abstract

This paper describes a novel method for predicting the crystal structure of organic molecular materials which employs a series of successive approximations to focus on structures of high probability, without resorting to a brute force search and energy minimization of all possible structures. The problem of multiple local minima is overcome by assuming that the crystal structure is closely packed, thereby eliminating 217 of the 230 possible space groups. Configurations within the 13 remaining space groups are searched by rotating the reference molecule about Cartesian axes in rotational increments of 15°. Initial energy minimization is performed using (6–12) Lennard–Jones pair potentials to produce a set of closely packed structures. The structures are then refined with the introduction of a Coulombic potential calculated using molecular multipole moments. This method has successfully located local minima which correspond to the observed crystal structures of several saturated and unsaturated hydro-C atoms with no a priori information provided. For large polycyclic aromatic hydrocarbons, additional refinements of the energy calculations are required to distinguish the experimental structure from a small number of closely packed structures. Our methodology for a priori crystal structure prediction represents the most efficient algorithm presented to date, in a field where the first successes have only been described within the past year and have been few and far between. Since our algorithm is capable of locating a large number of reasonable structures with similar energy in a short period of time, and is more likely to locate a minimum corresponding to the experimental structure, our program provides a superior framework to determine the level of theory required to calculate the intermolecular potential. For all but highly asymmetric hydrocarbons, however, distinguishing the observed structure from a large number of highly probable structures requires more rigorously calculated intermolecular interactions than pair potentials, plus an ad hoc electrostatic potential, and is thus beyond the scope of this paper. All calculations were performed on the Ohio Supercomputer Center's Cray Y-MP.