Machine Learning Study of the Magnetic Ordering in Two-dimensional Materials

The advent of two-dimensional (2D) materials opened new arenas for magnetic compounds, even when classical theories discourage their examination. The recent experimental discovery of 2D magnetic materials has inspired fundamental questions about the physical mechanism behind the magnetic ordering. Despite the huge interest raised by 2D magnetic materials, there are no a priori rules or trends allowing the prediction and understanding of this class of functional compounds. Here, we perform a machine learning study to predict and understand the tendency in the space of atomic species and 2D structures for a material to be classified as i) magnetic or non-magnetic using a random forest algorithm coupled to the SHAP method for model interpretability; and ii) ferromagnetic or antiferromagnetic based on the SISSO method. The ML models (i.e., a materials map that is function of the composition, atomic properties, and crystal symmetry) provide an accuracy higher than 85%. We find classification rules discriminating the existence of magnetism as well as the magnetic ordering type in terms of the strength of the spin-orbit coupling, the type of transition metal, the indirect bonding among transitions metals, and atomic properties of the constituent atoms, indicating new routes for experimental exploration.