Magnetic anisotropy is a fundamental prerequisite for stabilizing long-range order in two-dimensional (2D) magnets, yet its evaluation via first-principles methods is computationally prohibitive for large-scale material discovery. Here, an octahedral-field-informed machine learning framework is reported, specifically designed to identify 2D magnets with perpendicular magnetic anisotropy (PMA). By incorporating crystal-field-informed descriptors, this model transcends traditional composition-based heuristics, achieving an 89% classification accuracy while accelerating the discovery process by 6 orders of magnitude. High-throughput screening of over 10 000 candidates reveals 15 novel PMA materials, with ${\text{Ba}}_2{\text{VTe}}_2{\mathrm{S}}_2$ identified as a prominent candidate. Specifically, axial elongation along the out-of-plane direction by Ba cations modulates the $d$-orbital splitting, resulting in a giant magnetic anisotropy energy of 5.65 meV per V atom. This work provides an interpretable framework for rapid discovery of 2D PMA magnets and highlights how physics-informed descriptors enable rational design of low-dimensional magnetic materials.

Close

PrevNext

View figure in article

• Magnetic anisotropy
• Spintronics
• 2-dimensional systems
• Ferromagnets
• Magnetic semiconductors
• Ab initio molecular dynamics
• Density functional calculations
• Machine learning