Discovery of 15 2D magnets with perpendicular magnetic anisotropy using ML with physical information

1/ AI utilization method: supervised learning classifier including octahedral crystal-field information
2/ Application task: classification of candidate two-dimensional magnets with perpendicular magnetic anisotropy
3/ Reported result: screened more than 10,000 candidates, proposed 15 new PMA materials, and reported 89% classification accuracy
4/ The researchers used a machine learning classifier that included octahedral crystal-field information as descriptors to identify two-dimensional magnets exhibiting perpendicular magnetic anisotropy (PMA).
5/ With this method, more than 10,000 candidates were screened, 15 new PMA materials were proposed, and Ba₂VTe₂S₂ was mentioned as a promising candidate among them.
6/ The classification accuracy was reported as 89%.
7/ From a materials research perspective, this result shows that candidate discovery can be accelerated not by simple data-driven classification alone, but by reflecting physical knowledge about the structural and electronic environment in descriptors.
8/ In particular, for the search for two-dimensional magnets targeting rare magnetic anisotropy, this serves as an example in which physics-informed feature design improves the efficiency of high-throughput screening.
9/ However, the reported performance and candidate proposals are results for the given data and the defined search space, and do not mean that the proposed materials are actually synthesizable or that the desired magnetic properties can be reproduced experimentally.
10/ Classification accuracy alone cannot fully determine the distribution of false positives or out-of-domain performance, and final validation requires separate calculations and experiments. [1][2]