H. Chan, M. Cherukara, B. Narayanan, T. Loeffler, C. Benmore, S. Gray, S. Sankaranarayanan
Argonne National Laboratory,
United States
Keywords: molecular dynamics, force field, water
Summary:
Formation and growth of grains of ice is ubiquitous, influencing naturally occurring phenomena and processes happening at the nanoscale, like intracellular freezing. Despite the exponential growth in computing resources, it remains a grand challenge to simulate phase transitions and dynamical processes in deeply supercooled systems due to limitations imposed by system sizes and timescales which is further compounded by their sluggish kinetics. Here, we applied machine learning to develop a set of interatomic potentials for water. These coarse-grained models accurately describe the structure and thermodynamic anomalies of both water and ice, and they are two orders of magnitude cheaper computational cost than existing atomistic models. Furthermore, our workflow navigates efficiently through the high-dimensional parameter space to even improve upon existing high-quality coarse-grained models. Multi-million molecule nucleation and grain growth simulations are performed using the models. New machine learning based techniques are developed to analyze the grain size distribution of the simulations to elucidate the molecular mechanism of ice nucleation.