Predicting magnetic anisotropy energies using site-specific spin-orbit coupling energies and machine learning: Application to iron-cobalt nitrides

We perform high-throughput first-principles calculations to predict the magnetic anisotropy energies of a variety of iron-cobalt nitrides. We illustrate the efficacy of a spatial decomposition technique that divides the total magnetic anisotropy energy into spin-orbit coupling energy contributions from individual sites. The spatial decomposition scheme that we utilized works for a wide range of magnetic anisotropy energies. We also construct a machine-learning model by combining the site-specific spin-orbit coupling energies with structural details on each atomic site. We adopt the same approach to predicting site-specific magnetic moments. We show the capability of our machine-learning model to accelerate computational screening of candidate materials with high magnetization and large magnetic anisotropy energy.