Predicting Crystallization Outcomes of Chiral Molecules Using Simple Molecular Models.

다국어 초록 (Multilingual Abstract)

The crystallization of molecules is a key process in manufacturing pesticides, fertilizers, dyes, pharmaceutical compounds, explosives, and many other materials. Moreover, the spatial arrangement of molecules in the solid state determines many of a compound’s physical properties, including its stability, hardness, hygroscopicity, and solubility. Therefore, reliably predicting the solid form of organic and inorganic molecules is of considerable practical importance across chemistry and material science. Over the last decades, computers have been used with increasing success to predict molecular crystals. Most current methods of computational crystal structure prediction (CSP) rank different polymorphs according to their lattice energy or free energy. The most advanced CSP methods now routinely furnish many of the polymorphs found in experiments. However, CSP has still not replaced expensive and time-consuming experimental screening procedures because many of the lowest-energy crystal structures predicted by CSP do not form in experiments. In addition, CSP currently cannot distinguish between molecules that crystallize easily and those that tend to form amorphous solids. These limitations of CSP are partly due to the numerical effort associated with calculating molecular interactions: Accurate free-energetic ranking of polymorphs requires numerically cumbersome electronic structure calculations or highly specialized force fields that must be generated on a case-by-case basis. The number and complexity of polymorphs that can be predicted for a given molecule are therefore limited. More importantly, most CSP methods ignore all kinetic aspects of molecular crystallization, which are at the heart of polymorphism and amorphization. While the pertinent microscopic processes involved in crystal nucleation and growth are well understood, directly evaluating the crystallization rates of many different polymorphs in molecular dynamics computer simulations is essentially impossible due to extreme computational cost. The practical impact of in-silico CSP on the targeted manufacturing of molecular crystals, therefore, remains limited.In this work, we develop a family of simple models of chiral molecules capable of reproducing the full range of crystallization behavior observed in real molecules. We develop several methods for efficiently simulating the crystallization dynamics of these models and for the fast identification of their crystal structures, including crystals with complex and large unit cells that are usually ignored by current CSP methods. Due to the numerical efficiency of our models, we thoroughly investigate the crystallization dynamics and polymorph landscapes of thousands of different molecules. Our study, for the first time, provides a statistically meaningful correlation of the thermodynamics and kinetics of molecular crystallization across a wide range of molecular shapes and interactions, allowing us to address several long-standing questions: What fraction of molecules do not form the thermodynamically stable crystal? Can the rates of crystallization of different polymorphs be estimated without numerically strenuous dynamical simulations? How can the likelihood of crystallization and chiral separation be enhanced? Why do molecules tend to form structurally simple polymorphs? This work can serve as guiding principles for the synthetic design of molecules and the prudent choice of crystallization conditions. Most importantly, we suggest conceptually simple and numerically efficient ways of accounting for kinetic effects in computational crystal structure prediction.