Understanding hydrogen diffusivity in amorphous titania: A combined density functional theory, machine learning, and graph theory study

Hydrogen diffusion plays an important role in understanding how materials chemically degrade over time. Open questions regarding the underlying diffusion pathways and whether they lead to the trapping of Hydrogen provide insight into properties such as the incubation time towards hydrogen embrittlement. For the case of polycrystalline materials, answering these questions are non-trivial, as hydrogen can diffuse through two unique environments: (1) bulk crystalline region and (2) grain boundary region. In this work we approximate the grain boundaries using the amorphous bulk phase due to similarities in the short-range atomic order. Density functional theory (DFT) was used to generate amorphous configurations of titania via ab initio molecular dynamics. A spectrum of hydrogen binding energies was then calculated using a multitude of oxygen sites with varying coordination number. Classical molecular dynamics were then performed on a larger titania system using a machine learning force field. A graph-based order parameter was then used to characterize the classically derived amorphous phase space and was mapped onto the oxygen coordination number. Our results qualitatively indicate that hydrogen diffusivity can be tailored to either inhibit or induce diffusion based on the specific local oxygen environments present in the amorphous phase.