AI and the World's Hard Problems · Physics · arXiv · 2026-10-19
Physics-Informed Machine Learning for Response Tensors in Crystalline Magnets: Application to Kagome Metals
Predicting response tensors from magnetic textures is a key challenge in noncollinear magnets, and existing descriptors were limited to first-order multipole expansions or simple magnetic structures. The authors built a complete symmetry-invariant descriptor basis for arbitrary crystal structures and modified principal component analysis to derive uncorrelated, hierarchically stable order parameters. Applied to the anomalous Hall conductivity of a magnetic kagome lattice, the squared regression coefficients followed a Zipf-like power law in the regime where linear regression works well.