Molecular mechanism of ice nucleation on feldspar

Abstract

Understanding how water transforms into ice at complex surfaces is central to a wide range of natural and technological processes, yet molecular simulations of this transformation have largely been restricted to idealized surfaces or used models with limited predictive power. Here, we employ advanced molecular simulation tools to study ice nucleation on feldspar, the most abundant mineral in Earth's crust and one of the main ice nucleating particles in the atmosphere. We develop a machine learning interatomic potential trained on ab initio electronic structure calculations, achieving quantum accuracy across a broad range of feldspar-water interfaces. Molecular simulations driven by this potential reveal that the feldspar (110) surface uniquely templates interfacial water into an arrangement resembling the structure of water over ice. Combining this approach with enhanced-sampling and seeded methods, we directly observe nucleation of cubic ice at the (110) feldspar surface, characterize the critical nucleus, and demonstrate that its orientation relative to the mineral surface is consistent with experimental observations. On this basis, we identify the (110) surface as the dominant active site, overturning the currently accepted mechanism, which attributes feldspar's exceptional ice nucleation ability to the (100) surface exposed at defects. These results provide new insight into a principal pathway for atmospheric ice formation. More broadly, they demonstrate the power of ab initio machine learning simulation for the in silico prediction of ice nucleation at surfaces, and clarify the connection between interfacial water structure and the ice-nucleation ability of realistic surfaces.