Emerging computational and machine learning methodologies for proton-conducting oxides: materials discovery and fundamental understanding

This review presents computational and machine learning methodologies developed during a 5-year research project on proton-conducting oxides. The main goal was to develop methodologies that could assist in materials discovery or provide new insights into complex proton-conducting oxides. Through these methodologies, three new proton-conducting oxides, including both perovskite and non-perovskites, have been discovered. In terms of gaining insights, octahedral tilt/distortions and oxygen affinity are found to play a critical role in determining proton diffusivities and conductivities in doped barium zirconates. Replica exchange Monte Carlo approach has enabled to reveal realistic defect configurations, hydration behavior, and their temperature dependence in oxides. Our approach ‘Materials discovery through interpretation’, which integrates new insights or tendencies obtained from computations and experiments to sequential explorations of materials, has also identified perovskites that exhibit proton conductivity exceeding 0.01 S/cm and high chemical stability at 300 ∘C. This review presents new computational and machine learning methodologies that identified three new proton-conducting oxides and two fast proton conductors. It also provides new insights into hydration, proton diffusion, and proton conduction.