A Machine Learning Approach to Decipher the Origin of Magnetic Anisotropy in Three-Coordinate Cobalt Single-Ion Magnets

Abstract

Single Molecule Magnets (SMMs) emulate permanent magnets and are highly regarded for their role in compact information storage and molecular spintronics. Their behavior is primarily governed by magnetic anisotropy, expressed through parameters like the axial zero-field splitting (D) and orientation of magnetic anisotropy (gx, gy, gz) in mononuclear transition metal complexes. Low-coordinate mononuclear transition metal complexes stand out for their substantial anisotropy and higher blocking temperatures. However, understanding the intricate interplay between these parameters poses a significant challenge, often beyond traditional magneto-structural correlations. Hence, machine learning (ML) tools have been embraced to address these complexities. By employing an ML model based on Co-ligand bond length and angle relative to the pseudo-C3 axis, this study effectively rationalizes variations in D values, g-factors, and rhombic anisotropy, crucial for determining magnetic properties. Leveraging a dataset of 627 molecules, the research explores ML's potential in predicting magnetic anisotropy parameters in three-coordinate Co(II) complexes, achieving a minimal mean absolute error (MAE) of approximately 17 cm⁻¹ and high accuracy levels exceeding 95% for classification tasks. These insights offer valuable guidance for the development of innovative single-ion magnets.