Machine learning potential for predicting atomic structures of impurity-segregated grain boundaries in Al2O3.
Author：Tatsuya Yokoi
Affiliation：Department of Materials Physics, Graduate School of Engineering, Nagoya University
Abstract：This work developed machine-learning potentials trained on DFT data to predict energetically favorable atomic structures of impurity-segregated grain boundaries in Al2O3. Grain boundary segregation of substitutional trivalent cations was examined. It was found that our machine-learning potentials accurately predicted the relationship between atomic structures and grain boundary energies. This enabled us to systematically explore low-energy atomic structures without performing direct DFT calculations of individual atomic structures, significantly reducing the computational cost. The lowest-energy structure was in good agreement with experimental images observed by electron microscopy.
Affiliation：Department of Materials Physics, Graduate School of Engineering, Nagoya University
Abstract：This work developed machine-learning potentials trained on DFT data to predict energetically favorable atomic structures of impurity-segregated grain boundaries in Al2O3. Grain boundary segregation of substitutional trivalent cations was examined. It was found that our machine-learning potentials accurately predicted the relationship between atomic structures and grain boundary energies. This enabled us to systematically explore low-energy atomic structures without performing direct DFT calculations of individual atomic structures, significantly reducing the computational cost. The lowest-energy structure was in good agreement with experimental images observed by electron microscopy.
Posted : March 31,2026
