Machine learning-driven discovery of hard magnetic materials using high-throughput computation and screening

We present a machine-learning-driven framework for discovering high-performance rare-earth-free hard magnetic materials integrating machine learning, a universal graph deep-learning interatomic potential, and density functional theory validation. Key contributions include the identification of FeCo-based ternary alloys with remarkable magnetic properties, such as uniaxial anisotropy constant, K 1 , Curie temperature, T C , and saturation magnetization, M S . Notable examples include Fe 6 CoB 2 and FeCo 5 B, which exhibit K 1 values of 1.76 MJ/m 3 and 1.00 MJ/m 3 , respectively, with M S above 1.3 T, and T C exceeding 600 K. These properties align with the needs of high-temperature and high-performance applications. The universal graph deep-learning interatomic potential M3GNet accelerates the structural relaxation process, enabling the efficient screening of 48,000 candidate structures, while density functional theory validates the top performers with energy product ( B H ) max reaching more than 600 kJ/m 3 . Our study highlights a scalable, efficient pipeline for advancing the discovery of permanent magnets, reducing reliance on rare-earth elements. • A machine learning pipeline predicts magnetic properties and structural stability of new hard magnets. • Accelerated discovery of FeCo-based ternary alloys with high magnetocrystalline anisotropy and Curie temperatures.