Oxide-water interfaces are crucial in various physicochemical phenomena with applications in climate science, ocean chemistry, catalysis, electrochemistry, and gas sensing, and they are commonly investigated using surface-specific sum-frequency generation (SFG) spectroscopy. In this work, we explored both the Al-terminated and O-terminated α-Al₂O₃(0001)-water interface using ab initio molecular dynamics (AIMD) simulations, establishing clear relationships between vibrational sum-frequency generation (vSFG) spectra and distinct water structures at the interface. We also captured critical interfacial water reactive events, specifically water dissociation processes, occurring exclusively at the Al-terminated α-Al₂O₃(0001)-water interfaces. However, despite the detailed insights provided by AIMD, its computational expense poses a significant limitation to its widespread use. To address this challenge, we present an efficient approach for simulating solid-water interfaces using a machine learning (ML)-based deep neural network (DNN). AIMD data from both the Al-terminated and O-terminated Al₂O₃(0001)-water interfaces were utilized to train the ML-based force fields. We evaluated the performance of the DNN models by comparing predicted forces with those obtained from density functional theory (DFT), using root mean square error (RMSE) and coefficient of determination (R²) as benchmarks. The interfacial water structure—characterized by the density profile, hydrogen bonding classes, water orientation, and geometric order parameters—was validated using AIMD data, and the water dissociation processes over longer timescales (nanosecond) were predicted. Our results demonstrate that this new method successfully significantly accelerates AIMD simulations while maintaining accuracy, thus enables the calculation of SFG spectra with significantly reduced computational cost. These advancements pave the way for more efficient investigations of complex solid-water systems.

Christopher Arges

Appearances