Ab-initio-based transferable potential for sodalites

In this paper we derive a potential energy function for the frame atoms and the Na+ ions in sodalite-type zeolites. We use Coulomb interactions between the charges on the frame atoms and the ions, a Buckingham potential for the pairwise interactions, and a three-body term that couples bending and stretching for the O-T-O unit (T = Si, Al). This potential ascribes realistic charges to the O atoms of the framework, with the charges on Si and Al fixed to ensure transferability of the potential to all possible Si/Al ratios. The remaining potential parameters are chosen to ensure (1) accurate fits to ab initio surfaces of model clusters calculated at the MP2-6-31G* level of theory, (2) accurate predictions as to NaCl-SOD, and (3) a reasonable infrared spectrum for NaCl-SOD and silica-SOD. We find that potentials that mimic bending with a purely angular dependent term give both poor fits to ab initio surfaces and vibrational frequencies in NaCl-SOD. Three-body potentials that include a coupling to bond lengths and pair potential schemes that include O-O interactions result in force fields that give both accurate fits to ab initio data, accurate structural predictions, and reasonable vibrational frequencies. The force field supplements the pairwise additive terms with a nearest neighbor O-T-O three-body potential that enforces the tetravalent sp(3) hybridization of the T atoms. The potential gives bond lengths that are accurate to within 0.01 Angstrom, lattice constants to within 1%, vibrational frequencies that are within 3% of experiment, and bond angles that are accurate to within 1 degrees of their X-ray-determined values.