Unconventional defect configurations in aluminum oxide

Electronic structure calculations have been instrumental in characterizing point defects in materials, including formation energies, diffusivities, optical properties, and more. Accurate results rely on accurate atomic configurations, which are not always straightforward to obtain: complex total energy surfaces with several local minima at distinct defect geometries challenge first-principles predictions for, e.g., metal oxides. In this work, we evaluate several approaches for sampling candidate defect sites within density functional theory, including Bayesian inference and a method based on the Voronoi decomposition of the pure crystal. We also investigate the influence of exchange and correlation, spin polarization, and different schemes for correcting finite size effects for charged defects. We discover a series of complex and unconventional defect geometries in aluminum oxide, in some cases identifying lower energy interstitial geometries than previously known, and in other cases finding that rearranging a point defect into a defect cluster is energetically favorable. Sometimes, the lowest energy geometry even depends on defect charge state. These findings demonstrate the necessity for thorough and systematic defect geometry optimization and have profound implications for poorly understood defect migration paths and diffusion processes in aluminum oxide, including the possibility of mechanisms mediated by charge-transfer processes.