Crystal Nucleation Analysis from the Time Evolution of Local Particle Environments

Crystallization is consequential to many applications, including pharmaceutical production, flow assurance, and climate modeling. Effective control over crystallization relies on an understanding of the possible structures that can form during nucleation and growth. Molecular simulations allow a more fine-grained approach to discovering important, though possibly short-lived, intermediate structures, but their characterization from atomic coordinates is often difficult. We combine general features of the local atomic arrangements with a deep learning model to discover the unique structures that form during crystal nucleation. While many previous mechanistic studies have relied on features that describe the entire crystal nucleus, such as its size, shape, and composition, we focus on the evolution of the atoms involved in the formation of the nucleus in the feature space to describe nucleation processes. Understanding the role of how local atomic environments evolve allows further control of nucleation processes, with applications in polymorph selection.