Discovery of rare-earth-free magnetic ternary compounds using machine learning assisted adaptive genetic algorithms

Finding new materials with desired properties is a challenging task owing to the vast number of possible compositions and crystal structures. In order to address this problem, we outline a feedback loop scheme consisting of machine learning assisted high-throughput first-principles calculations and adaptive genetic algorithm. Our scheme enables efficient and accurate predictions of materials properties through a wide range of compositional and structural space, allowing the fast discovery of materials with desired properties. We illustrate the procedure to a ternary Fe-Co-B system, where we discovered hundreds of new metastable Fe-Co-B structures across the ternary phase space. Many of many of these new structures possess promising magnetic properties that can be used as rare-earth-free magnets.