MACHINE LEARNING REPRESENTATION FOR ATOMIC ENERGIES IN MAGNETIC MATERIALS

Abstract

In this study, we propose machine learning models, including linear regression, LASSO regression, and Ridge regression, for fast estimating atomic energies in a magnetic system. In our method, the total energy of a magnetic system contains chemical energy and magnetic energy. The chemical energy is approximated as the summation of atomic energy which is the interaction energy with its surrounding chemical environment within a certain cutoff radius. Atomic energy is decomposed into two-body terms which are expressed as a linear combination of basis functions. The magnetic energy is also approximated as the summation of atomic magnetic energy. The machine learning models, trained with crystal bcc-Fe data, can fast estimate the total energy of the system in both magnetic and non-magnetic states. Result from these models were analyzed and compared with calculated results by density functional theory (DFT). Model evaluation metrics including MSE, MAE and R² indicated that Ridge regression gives the best results. Results from our machine learning models show good agreement with DFT calculations.