On-the-fly assessment of diffusion barriers of disordered transition metal oxyfluorides using local descriptors

Disorder plays an increasingly important role in the design and development of high-performance battery materials and other clean energy materials like thermoelectrics and catalysts. However, conventional computational design approaches based on the thermodynamic properties of statistically averaged structures are unable to predict the accessible energy and power densities of such materials. Kinetic properties like ionic diffusion within locally resolved atomic structures is needed to perform longer time and length scale simulations like kinetic Monte Carlo in order to accurately estimate kinetic properties like power densities in battery electrodes. Here, we present and demonstrate a fast, on-the-fly, approach to calculate local diffusion barrier as a function of only the local atomic structure using machine learning and cluster expansion, particularly for Li-ions in lithium-rich transition metal oxyfluorides and the disordered rock salt (DRS) Li_2-xVO₂F electrodes.