Data-driven studies of magnetic two-dimensional materials

Abstract

We use a data-driven approach to study the magnetic and thermodynamic properties of van der Waals (vdW) layered materials. We investigate monolayers of the form $$\hbox {A}_2\hbox {B}_2\hbox {X}_6$$ A 2 B 2 X 6 , based on the known material $$\hbox {Cr}_2\hbox {Ge}_2\hbox {Te}_6$$ Cr 2 Ge 2 Te 6 , using density functional theory (DFT) calculations and machine learning methods to determine their magnetic properties, such as magnetic order and magnetic moment. We also examine formation energies and use them as a proxy for chemical stability. We show that machine learning tools, combined with DFT calculations, can provide a computationally efficient means to predict properties of such two-dimensional (2D) magnetic materials. Our data analytics approach provides insights into the microscopic origins of magnetic ordering in these systems. For instance, we find that the X site strongly affects the magnetic coupling between neighboring A sites, which drives the magnetic ordering. Our approach opens new ways for rapid discovery of chemically stable vdW materials that exhibit magnetic behavior.

Machine learning assisted hierarchical filtering: a strategy for designing magnets with large moment and anisotropy energy

“…In application of ML to magnetic properties reported in the literature, the models are often trained on a rather narrow class of materials, reducing the scope of their applicability. For example, Rhone et al 6 used only tri-chalcogenides with a particular crystal structure. Miyazato et al 15 used 2D materials with general formulas AB and AB 2 only.…”

“…Reduced error pruning decision trees along with the random subspace technique performed the best with an MAE of 0.09 eV per atom. Rhone et al, 6 although working with a uniform set of compounds all having the stoichiometry A 2 B 2 X 6 (A = TM; B = Si, Ge, P; X = S, Se, Te), and the same lattice structure, found rather large MAE between 0.33 eV per atom to 0.46 eV per atom in different models, but large R 2 values of B0.9. The best performing model, in terms of MAE, to our knowledge is the MEGNet model introduced by Chen et al 8 which gave MAE of 0.028 eV per atom on a dataset of 60 000 materials obtained from the Materials Project.…”

“…Second, magnetic moment may be a more difficult target quantity to predict. Rhone et al 6 worked with a uniform set of compounds having the general formula A 2 B 2 X 6 , as already mentioned. An Extra Trees regression model gives R 2 = 0.95 for the X = Te compounds.…”

“…In recent years, machine learning (ML) has emerged as a powerful means of predicting material properties without having to perform experiments or DFT calculations for every material, provided a sufficiently large database is available. It has already been applied to a wide range of issues in materials research including band gaps, 4 stability, [5][6][7][8] atomic dynamics at grain boundaries, 9 vibrational free energies and entropies, 10 screening battery electrodes, 11 to mention only a few. Relevant to our work, ML has been applied to the study of magnetic properties in a few published works.…”

“…Based on high throughput computations and ML, new magnetic materials with high T c were predicted in the Heusler family by Sanvito et al 13 Long et al 14 applied ML to design of ferromagnetic materials by classifying magnetic materials as ferro-(FM) or anti-ferromagnets (AFM), and predicting T c of the FMs. Rhone et al 6 applied ML to predict formation energies and magnetic moments of TM trichalcogenide layers. Miyazato et al 15 used ML to predict new 2D magnetic materials.…”