Exploring Sustainable High-Entropy Magnetic Materials Using Active Learning

Rare-earth-free and precious-metal-free magnets have important applications in advanced magnetic recording media, permanent magnets, and spintronic devices. High-entropy alloys (HEAs) have emerged as a promising platform to achieve strong magnetic anisotropy, motivating our efforts in using active learning to guide ab initio calculations through the vast parameter space of HEAs. Using L1₀ phase MnAl, with a favorably high uniaxial anisotropy of 2.1 MJ/m³, as a base lattice, we explore the compositional space of (MnFeCoNi)Al for high magnetic anisotropy energy, high saturation magnetization, and moderate Curie temperature near 500 K, driven by Gaussian process. Through Bayesian optimization, we vary the concentrations of elements in the HEAs to achieve the desired properties. We demonstrate a smooth profile of magnetic properties from bulk DFT calculations. Ideal compositions predicted by the Gaussian process are selected to undergo manual optimization of their lattice constants and a recalculation of their magnetic properties. The trend in properties is found to be consistent with the original data. A correlation is observed between the concentration of Mn in the material and the lattice constants of the relaxed cell, which leads to an intriguing engineering strategy for heat-assisted magnetic recording media.