Machine Learning Approach for Data Analysis of Magnetic Orbital Moments and Magnetocrystalline Anisotropy in Transition-Metal Thin Films on MgO(001)

Abstract

Using the Least Absolute Shrinkage and Selection Operator (LASSO) technique, we analyze a long-standing issue in the field of magnetism: the relationship between orbital magnetic moments and magnetocrystalline anisotropy (MCA) energy in transition-metal thin films. Our LASSO regression utilizes the data obtained from first principles calculations for single slabs with six atomic-layers of binary Au-Fe, Au-Co, and Fe-Co films on MgO(001). In the case of Fe-Co thin films, we have successfully regressed the MCA energy against the anisotropy of orbital moments along the in-plane and the perpendicular plane directions, giving a linear behavior. For the Au-Fe and Au-Co thin films, however, our data-driven analysis shows no relation between the MCA energy and the anisotropy of orbital moments.

Transition-Metal Interlink Neural Network: Machine Learning of 2D Metal–Organic Frameworks with High Magnetic Anisotropy

“…In recent years, machine learning (ML) techniques have been rapidly developed and widely used in physics, chemistry, and materials science. , Based on massive databases constructed from either experimental or computational means, ML could provide credible results with much fewer computational requirements compared to conventional ab initio approaches. − Nowadays, the commonly used ML algorithms in the computational materials science include artificial neural networks (ANNs), Gaussian approximation potentials, and kernel ridge regression . Current targets for the predicted material properties include electronic band gap, magnetic moment, hardness, bulk modulus, and Young’s modulus. , In addition, the data-driven mode of ML is also widely used to design functional materials by providing some guidance rules. This has already shown potential in accelerating the discovery of new catalysts, − perovskites, ,− photovoltaic materials, and spintronic materials .…”

“…38 Current targets for the predicted material properties include electronic band gap, magnetic moment, hardness, bulk modulus, and Young's modulus. 39,40 In addition, the data-driven mode of ML is also widely used to design functional materials by providing some guidance rules. This has already shown potential in accelerating the discovery of new catalysts, 41−43 perovskites, 28,44−46 photovoltaic materials, 47 and spintronic materials.…”