Geometric Approaches to Crystal Packing and Property Prediction

The lack of a robust model for large scale labelling of crystal packings has led to difficulty in the analysis of large crystal databases for the purpose of property prediction through machine learning (ML)-based algorithms. Here we seek to more accurately and systematically describe the topology of a crystal with a crystal matching algorithm to enable the comparison of molecular packings in a given structure with those in a large database of crystals. The method facilitates the identification of structurally similar packings, which can guide ML models in property prediction as polymorphs are identified. We also investigate the crystallization of hard particle systems through Monte Carlo simulations to determine the impact of particle shape on crystal formation. This approach offers a cost-effective method to observe and understand crystallization dynamics, potentially providing a powerful tool for accelerating CSP and aid in ML-assisted material discovery.