Exploring homogeneous ice nucleation through molecular dynamics simulations with ab initio accuracy

Probing the homogeneous ice nucleation process from scratch at the molecular level is of great value but still experimentally unachievable. Previous theoretical simulations have found that ice originates from the low-mobile region with increasing abundance and persistence of tetrahedrally coordinated water molecules. However, a detailed microscopic picture of how the disordered hydrogen-bond network rearranges itself into an ordered network is still unclear. In this work, we use a deep neural network (DNN) model to ``learn'' the interatomic potential energy from quantum-mechanical data, thereby allowing for large-scale and long molecular dynamics simulations with ab initio accuracy. Using the forward flux sampling, a $36\text{\ensuremath{-}}\textmu{}\mathrm{s}$-long simulation is performed to achieve complete homogeneous ice nucleation trajectories from scratch. We demonstrate that the nucleation mechanism and dynamics at the molecular resolution are deeply affected by the structural and dynamical heterogeneities in supercooled water. In distinct contrast to previous understanding, we find that ice formation occurs in coexisting high-mobile and low-mobile regions of supercooled water, with low-mobile, highly ordered regions first transforming into high-mobile regions before they crystallize into ice. Specifically, imperfectly coordinated (IC) water molecules with high mobility pave the way for the hydrogen-bond network rearrangement through the mechanisms of IC status transfer, annihilation, and generation, leading to the growth or shrinkage of the ice nucleus; the hydrogen-bond network formed by perfectly coordinated (PC) molecules stabilizes the nucleus, thus preventing it from vanishing and growing. Ice is born through competition and cooperation between IC and PC molecules. In the end, the nucleation rate at supercooling 46 K is calculated in a direct manner, and the result agrees well with experiments.