Hydration Behavior of Quartz Surfaces Revealed by Molecular Dynamics Simulations with a Novel Machine Learning Potential

Understanding the hydration behavior of quartz surfaces is critical for elucidating interfacial properties in geochemical and engineering systems. However, capturing these processes via simulation poses a significant challenge, as the high computational cost of conventional density functional theory (DFT) methods renders them unsuitable for simulating large interfacial systems over long time scales. In this work, a machine learning potential (MLP) was developed using an active learning strategy and a high-quality data set to enable large-scale molecular dynamics simulations of the quartz–water interface at ab initio accuracy. The trained potential achieves high accuracy in energy and force predictions across diverse structures and significantly outperforms empirical potentials in both precision and generalization. Driven by this potential, nanosecond-scale machine learning molecular dynamics (ML-MD) simulations have been performed to elucidate the surface reconstruction behavior of cleaved α-quartz (101) and the temporal evolution of surface hydroxylation. The results show that surface reconstruction induces a pronounced structural response that varies with depth in bulk atoms, and hydroxylation proceeds via water dissociation and proton transfer mechanisms, yielding a final silanol density of 5.169 ± 0.416 nm–2, which is consistent with the thermogravimetric analysis (TGA) results. Additionally, we characterize the layered structure and hydrogen-bonding network of interfacial water, revealing a hierarchical transition from surface-anchored to bulk-like water. This study demonstrates the potential of ML-MD as a powerful and efficient framework for simulation of complex mineral–water interfacial processes with ab initio accuracy at reduced computational cost.