Machine Learning Force Field beyond the Limits of Classical and First-Principles Molecular Dynamics Simulations: The Case of Kaolinite Hydration

Abstract

The understanding of the interaction between mineral surfaces and water holds great significance in unraveling the physical chemistry of surfaces in beneficiation processes. Within this study, we employ theoretical simulations to explore the adsorption mechanism of water molecules on the kaolinite surface, a clay mineral commonly found in iron ores. Albeit ab initio simulations offer key insights into the microscopic aspects of water adsorption, such as dissociation mechanisms or the hydrophilic/hydrophobic nature of the surface, the analysis of water structuring is limited by insufficient statistical sampling beyond the interfacial monolayer. By extending size and simulation time, the machine learning (ML) potential model is able to reproduce the experimental enthalpy profile by increasing the hydration film and provides converged densities with a well-defined minimum between the first water layer and water molecules beyond. Furthermore, the time correlation function combined with the ML method allows for the observation of dynamic occurrences on surfaces, including residence lifetime and the exchange of water molecules between hydration layers. In addition to offering valuable insights into the hydration mechanisms of kaolinite surfaces, for the first time we validate the use of the ML force field to overcome the limitations of both first-principles and classical molecular dynamics simulations for investigating adsorption phenomena.