Improved description of intermolecular interactions in hybrid ab initio/Empirical force-field models for crystal structure prediction

The prediction of polymorphism is important, since crystal physicochemical properties are crystal structure dependent. Crystal Structure Prediction (CSP) seeks to predict the putative polymorphs of a system from a minimal description of molecular connectivity. In this thesis, I study routes to improve CSP accuracy in hybrid ab initio/empirical force-field (HAIEFF) models, through developments in the description of intermolecular electrostatic interactions and residual intermolecular interactions. I first seek to parametrize more-accurate empirical force-fields (FFs). To this end, I contribute to the CE-755 dataset: a chemically-diverse set of high-quality organic molecular crystal data derived using dispersion-corrected periodic Density Functional Theory. Employing the CE-755, I fit seven FFs for four electrostatic models, totaling 28 independent parametrizations. Improved energy and geometry prediction accuracy is achieved by combining new atom types with modified FF forms that mitigate the unphysical traits of the Buckingham potential. The new FFs are reasonably transferable and predictions of relative polymorph stability in the CE-755 are marginally enhanced. The parameter estimation results indicate that describing induction is vital for HAIEFF model accuracy. As such, the Self-Consistent Electronic Response to Point-charge (SCERP) induction model was implemented into our group’s CSP programs. Single-point evaluations of induction energies using SCERP and the polarizable continuum model (PCM) suggest that the former is sounder. As before, SCERP-compatible FFs are parametrized, through which an effective set of atom-pair specific Tang-Toennies damping factors are proposed. Finally, I perform CSP studies for several hydrogen bonding molecular systems. Generally, good CSP results are obtained with the newly parametrized FFs, although further improvement from using SCERP is tenuous. Several notable deficiencies with both PCM and SCERP are uncovered, particularly as it pertains to the modeling of crystals with polar unit cells. I offer some guidance on addressing these flaws, as well as broader direction for the future development of HAIEFF models.