FMR analysis by machine learning leads to remarkable insights into the magnetic anisotropy of [Formula: see text] thin films.

The traditional approach to analyzing ferromagnetic resonance spectroscopy (FMR) data can produce inconsistent material parameters when measurements are analyzed at broadband and fixed-frequency conditions separately [Nat. Comm. 8, 234 (2017), Figs. 4 and 5 ]. Machine learning-based global optimization addresses this issue by simultaneously analyzing all FMR data, independent of frequency. Through a comprehensive reanalysis of published data and analysis of independent measurements on epitaxial [Formula: see text] thin films, we demonstrate that this method yields identical magnetic anisotropy parameters at both broadband and fixed-frequency conditions. In contrast, traditional fitting methods produce differences up to 7% when applied to broadband and fixed-frequency measurements separately. This methodology also enables direct extraction of fundamental parameters, such as the g-factor and magnetization, from FMR data alone, with results consistent with independent measurements. By leveraging measurements for all frequencies, the machine learning approach facilitates self-consistent and frequency-independent material evaluation and effectively distinguishes intrinsic properties from measurement artifacts.