Uncovering Giant Magnetic Anisotropy in Co(II) Single-Ion Magnets via Combined Ab Initio and Machine-Learning Approaches

Single-ion magnets, particularly low-coordinate Co(II) complexes, are promising for high-density data storage; however, achieving large effective energy barriers while retaining strong magnetic anisotropy remains an unresolved challenge. In this work, we systematically extracted three-coordinate Co(II) complexes from reported X-ray crystal structures and subsequently expanded this set to 1053 complexes, thereby capturing a broad chemical diversity that spans homoleptic {CoX₃} species (X = C, N, O, S) and heteroleptic motifs such as {CoX₂Y} and {CoCClX} (X, Y = C, N, O, S, Cl, Br, I). The D values in this data set range from +95 to -222 cm^-1, with E/D ratios from 0 to 0.32, capturing diverse magnetic behavior. We have developed a machine learning (ML) model using geometric parameters such as bond lengths, angles, and deviations from ideal C_3V geometry, etc. that predicts D, E/D, and g-factors with over 95% accuracy, with a mean absolute error of ∼12 cm^-1 and aligns well with both CASSCF results and available experimental data. Remarkably, our models also uncovered over 40 new Co(II) complexes with large negative D values (>180 cm^-1) and low E/D ratios (0.01-0.07), demonstrating the capability of the ML-driven approach to accelerate the discovery of next-generation ambiently stable SIMs.