Computational investigation of surface freezing in a molecular model of water

Water freezes in a wide variety of low-temperature environments, from meteors and atmospheric clouds to soil and biological cells. In nature, ice usually nucleates at or near interfaces, because homogenous nucleation in the bulk can only be observed at deep supercoolings. Although the effect of proximal surfaces on freezing has been extensively studied, major gaps in understanding remain regarding freezing near vapor-liquid interfaces, with earlier experimental studies being mostly inconclusive. The question of how a vapor-liquid interface affects freezing in its vicinity is therefore still a major open question in ice physics. Here, we address this question computationally by using the forward-flux sampling algorithm to compute the nucleation rate in a freestanding nanofilm of supercooled water. We use the TIP4P/ice force field, one of the best existing molecular models of water, and observe that the nucleation rate in the film increases by seven orders of magnitude with respect to bulk at the same temperature. By analyzing the nucleation pathway, we conclude that freezing in the film initiates not at the surface, but within an interior region where the formation of double-diamond cages (DDCs) is favored in comparison with the bulk. This, in turn, facilitates freezing by favoring the formation of nuclei rich in cubic ice, which, as demonstrated by us earlier, are more likely to grow and overcome the nucleation barrier. The films considered here are ultrathin because their interior regions are not truly bulk-like, due to their subtle structural differences with the bulk.