Efficient discovery of large magnetic anisotropy with thermodynamically stable materials via multi-objective Bayesian optimization

This study presents an efficient magnetic material exploring method with large magnetocrystalline anisotropy (MCA) and thermodynamic stability by combining multi-objective Bayesian optimization (MBO) with first-principles calculations. Focusing on six-layer tetragonal ordered alloys of rare-earth-free elements (Fe, Co, Ni, Cu), our approach optimizes material properties and process conditions within high-dimensional spaces. Explanatory variables derived from elemental and structural information were employed to enhance Bayesian optimization. Utilizing Gaussian process regression and dimensionality-reduced descriptors, we achieved efficient and robust exploration for precomputed dataset. As a result, MBO efficiently suggested the following three materials – Fe/Fe/Fe/Ni/Fe/Ni, Fe/Co/Fe/Co/Fe/Ni, and Fe/Co/Co/Fe/Ni/Ni – that exhibit thermodynamic stability while demonstrating an MCA value up to four times larger than that of L1₀-FeNi. The developed methodology accelerates material discovery and reduces resource demands, offering a scalable materials solution for optimizing functionality and process conditions across diverse material systems. This study accelerates advanced magnetic materials discovery by integrating multi-objective Bayesian optimization and first-principles calculations, uncovering rare-earth-free alloys with exceptional magnetocrystalline anisotropy and thermodynamic stability, enabling scalable materials development.