Ice is present everywhere on Earth and has an essential role in several areas, such as cloud physics, climate change and cryopreservation. The role of ice is determined by its formation behaviour and associated structure. However, these are not fully understood¹. In particular, there is a long-standing debate about whether water can freeze to form cubic ice—a currently undescribed phase in the phase space of ordinary hexagonal ice^2,3,4,5,6. The mainstream view inferred from a collection of laboratory data attributes this divergence to the inability to discern cubic ice from stacking-disordered ice—a mixture of cubic and hexagonal sequences^7,8,9,10,11. Using cryogenic transmission electron microscopy combined with low-dose imaging, we show here the preferential nucleation of cubic ice at low-temperature interfaces, resulting in two types of separate crystallization of cubic ice and hexagonal ice from water vapour deposition at 102 K. Moreover, we identify a series of cubic-ice defects, including two types of stacking disorder, revealing the structure evolution dynamics supported by molecular dynamics simulations. The realization of direct, real-space imaging of ice formation and its dynamic behaviour at the molecular level provides an opportunity for ice research at the molecular level using transmission electron microscopy, which may be extended to other hydrogen-bonding crystals.

冰存在于地球上的各个角落，并且在云物理、气候变化和低温保存等多个领域发挥着重要作用。冰的作用取决于其形成行为和相关结构。然而，这些还没有被完全理解¹ 。特别是，关于水是否可以冻结形成立方冰（普通六方冰相空间中目前尚未描述的相）存在长期争论^2,3,4,5,6 。从实验室数据收集中推断出的主流观点将这种差异归因于无法区分立方冰和堆积无序冰（立方体和六边形序列的混合物^7,8,9,10,11 ）。使用低温透射电子显微镜结合低剂量成像，我们在这里展示了立方冰在低温界面的优先成核，导致在 102 K 的水蒸气沉积中产生立方冰和六方冰两种类型的单独结晶。我们识别了一系列立方冰缺陷，包括两种类型的堆积无序，揭示了分子动力学模拟支持的结构演化动力学。在分子水平上实现冰形成及其动态行为的直接、真实空间成像为使用透射电子显微镜在分子水平上进行冰研究提供了机会，这可能会扩展到其他氢键晶体。