Resolving the odd-even oscillation of water dissociation at rutile TiO₂(110)-water interface by machine learning...

Aqueous rutile TiO₂(110) is the most widely studied water-oxide interface, and yet questions about water dissociation are still controversial. Theoretical studies have systematically investigated the influence of the slab thickness on water dissociation energy (E_diss) at 1 monolayer coverage using static density functional theory calculation and found that E_diss exhibits odd-even oscillation with respect to the TiO₂ slab thickness. However, less studies have accounted for the full solvation of an aqueous phase using ab initio molecular dynamics due to high computational costs in which only three, four, and five trilayer models of rutile(110)-water interfaces have been simulated. Here, we report Machine Learning accelerated Molecular Dynamics (MLMD) simulations of defect-free rutile(110)-water interfaces, which allows for a systematic study of the slab thickness ranging from 3 to 17 trilayers with much lower costs while keeping ab initio accuracy. Our MLMD simulations show that the dissociation degree of surface water (α) oscillates with the slab thickness and converges to ∼2% as the TiO₂ slab becomes thicker. Converting α into dissociation free energy (ΔA_diss) and comparing with dissociation total energy E_diss calculated with a single monolayer of water, we find that the full solvation of the interfaces suppresses surface water from dissociating. It is interesting to note that the machine learning potential trained from the dataset containing exclusively the five trilayer TiO₂ model exhibits excellent transferability to other slab thicknesses and further captures the oscillating behavior of surface water dissociation. Detailed analyses indicate that the central plane in odd trilayer slabs modulates the interaction between double trilayers and, thus, the bonding strength between terminal Ti and water, which affects pKa of surface water and water dissociation degree.