Development and Applications of Composite and Low-Cost Approaches in Molecular Crystal Structure Prediction

Abstract

Despite significant progress made in the last twenty years, the crystal structure prediction (CSP) of organic molecular solids remains challenging, as the demand to predict more complex crystal structures increases. On the one hand, relative energies between candidate crystal structures generated during a CSP protocol must be calculated accurately; on the other, the complexity of the crystal-energy landscape imposes stringent limitations on the method’s computational cost. While plane-wave density-functional theory (DFT) methods have become the workhorse for the final stages of CSP protocols, due to their balance between high accuracy and efficiency, they remain prohibitively expensive during the early and intermediate stages.

The primary aim of this thesis is the development of composite approaches for CSP, which comprise a geometry optimization using a low-cost method followed by a single- point energy calculation using plane-wave DFT with the exchange-hole dipole moment (XDM) dispersion model. The composite approaches were first tested on small molecular solids; assessment based on their abilities to produce absolute lattice energies was found to be misleading, and relative lattice energies provided a much better indicator of performance in a CSP context. To allow use of the XDM dispersion model with low-level methods, it was implemented in the SIESTA code, which uses numerical finite-support local orbitals to reduce the computational cost of the calculation. Composite approaches making use of the same DFT-D method both for low- and high-level DFT frameworks yielded the best ac- curacy, while remaining significantly cheaper than performing full geometry optimizations with plane-wave DFT. The composite approaches were then successfully employed for CSP of organic molecules with applications ranging from chiral organic semiconductors to pharmaceutical solids.

Secondary objectives of this thesis sought to offer insight as to whether certain classes of solid-state materials are not appropriate benchmarks for method validation, and whether DFT-D methods are always suitable to describe all molecular crystals of interest. In particular, using compounds that form polytypes, e.g., crystalline aspirin, to validate com- putational methods was found to be inadvisable due to their high geometric and energetic similarity. Also, delocalization error, an often-overlooked limitation of most DFT methods, affected the correct identification of the protonation site in multicomponent acid-base crystals. This error greatly affects the reliability of these methods for validation of experi- mental (or the prediction of new) crystal structures.

Overall, the work presented in this dissertation provides appropriate methodological and benchmarking tools to accelerate the intermediate stages of CSP protocols, while retaining high levels of accuracy and reliability in the crystal-energy landscapes generated, ultimately enabling the study of increasingly complex molecular crystals.