Predicting Material Properties by Integrating Combinatorial Experiments, Ab-initio Calculations, and Machine Learning in the Fe_xCo_yNi_1-x-y Ternary

Combinatorial experiments allow rapid mapping of composition-structure-property relationships across large compositional phase space. There are, however, physical properties which are difficult to perform rapid quantitative mapping. For instance, it is nontrivial to extract detailed information regarding spin and orbital magnetic moments of compounds in composition spreads. We propose an alternative approach which combines combinatorial experimentation, ab-initio calculations, and machine learning. In order to decipher the detailed magnetic properties across Fe-Co-Ni composition spread, we perform structural phase distribution mapping of the spread by X-ray diffraction, use Korringa Kohn Rostoker – Coherent Potential Approximation for ab-initio calculations, and apply non-negative matrix factorization. As a result, the predicted mapping of saturation magnetization by the proposed approach was well-consistent to the mapping of Kerr-rotation in combinatorial magnet-optic Kerr effect experiments. The result indicates that our approach has a potential to enable rapid mapping of more various physical properties by integrating combinatorial experiments, ab-initio calculations, and machine learning.