Crystal Structure Prediction (CSP) comprehends two challenging tasks: efficient structure sampling, to obtain all possible polymorphs of a crystal; and accurate lattice energy determination, to enable distinction between polymorphs. The latter remains the main challenge of successful CSP. Herein, we present a new benchmark database of lattice energies created employing our recently developed CrystaLattE; a software tool that automates dual-level parallel computation of lattice energies of molecular crystals. Our approach uses the many-body expansion approach and is designed with the aim of computing lattice energies accurately enough to discriminate between low-energy polymorphs of organic crystals. This database displays a hierarchical, multiscale scheme, featured in CrystaLattE, that employs the CCSD(T)/CBS method for the most relevant interactions and MP2/aug-cc-pVDZ for more distant interactions. Our database includes small organic molecules contained in other crystal databases and expands them with an additional set of molecular crystals relevant in pharmaceutical and energy-research applications. Based on the data from this set, we discuss the predominance of the two-body interactions contribution in contrast with that of higher-order terms in the many-body expansion. We also analyze how higher-order terms may become more important depending on the nature of the interaction in crystal lattices. Our results greatly impact the field of CSP, by providing an automated method to compute highly-accurate energies that can complement an efficient structure sampling code.

Automated multiscale methods for benchmark-level lattice energies of molecular crystals with CrystaLattE

2019 Spring National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.