Homogeneous ice nucleation in an ab initio machine-learning model of water

• DOI: 10.1073/pnas.2207294119
• Publication Date: 2022-08-08T19:16:07Z
• Authors: Pablo M. Piaggi; Jack Weis; Athanassios Z. Pa...; Pablo G. Debenedetti; Roberto Car
• Keywords: Supercooling; Classical nucleation theory; Supersaturation; Ice nucleus

Molecular simulations have provided valuable insight into the microscopic mechanisms underlying homogeneous ice nucleation. While empirical models been used extensively to study this phenomenon, based on first-principles calculations so far proven prohibitively expensive. Here, we circumvent difficulty by using an efficient machine-learning model trained density-functional theory energies and forces. We compute nucleation rates at atmospheric pressure, over a broad range of supercoolings, seeding technique systems up hundreds thousands atoms simulated with ab initio accuracy. The key quantity is size critical cluster (i.e., such that has equal probabilities growing or melting given supersaturation), which together equations classical rates. find for our moderate supercoolings are in good agreement experimental measurements within error calculation. also impact properties as thermodynamic driving force, interfacial free energy, stacking disorder calculated