Force fields and point charges for crystal structure modeling

Molecular simulation is increasingly used by chemical engineers and industrial chemists in process and product development. In particular, the possibility to predict the structure and stability of potential polymorphs of a substance is of tremendous interest to the pharmaceutical and specialty chemicals industry. Molecular mechanics modeling relies on the use of parametrized force fields and methods of assigning point charges to the atoms in the molecules. In commercial molecular simulation software, a wide variety of such combinations are available, and there is a need for critical assessment of the capabilities of the different alternatives. In the present work, the performance of several molecular mechanics force fields combined with different methods for the assignment of atomic point charges have been examined with regard to their ability to calculate absolute crystal lattice energies and their capacity to identify the experimental structure as a minimum on the potential energy hypersurface. Seven small, aromatic monomolecular. crystalline compounds are used in the evaluation. It is found that the majority of the examined methods cannot be used to reliably predict absolute lattice energies. The most promising results were obtained with the Pcff force field using integral charges, and the Dreiding force field using Gasteiger charges, both of which performed with an accuracy of the same order of magnitude as the variations in experimental lattice energies. Overall, it has been observed that the best results are achieved if the same force field method is used to relax the crystal structure and calculate the energy, and to optimize and calculate the energy of the gas phase molecule used for the correction for changes in molecular geometry. The Pcff and Compass force fields with integral charges have been found to predict relaxed structures closest to the experimental ones. In addition, five different methods for determining point charges fitted to the electrostatic potential (ESP charges), available in the same software, have been evaluated. For each method, the molecular geometries of 10 small, organic molecules were optimized, and ESP charges calculated and analyzed for linear correlation with a set of reference charges of an accepted standard method, HF/6-31G*. Dmol-3 gives charges that correlate well with the reference charge. The charges from Vamp are not linearly scalable to the HF/6-31G*-level, which is attributed partly to the geometry optimization but mainly to the calculation of the ESP and the subsequent charge fit.