Efficient Probabilistic Modeling of Crystallization at Mesoscopic Scale

Crystallization processes at the mesoscopic scale, where faceted, dendritic growth, and multigrain formation can be observed, are of particular interest within materials science. These processes are highly nonlinear, stochastic, and sensitive to small perturbations of system parameters and initial conditions. Traditional numerical models of these systems are computationally demanding. To address this, we introduce the Crystal Growth Neural Emulator (CGNE), a machine learning emulator that efficiently models crystallization using autoregressive latent variable models, which improves inference time by a factor of 11 compared to numerical simulations. To validate simulation quality, we compare morphological properties of crystals to those from numerical simulations, and find that CGNE substantially improves simulation fidelity and diversity over existing probabilistic models.