Fitting to magnetic forces improves the reliability of magnetic Moment Tensor Potentials

Computational Materials Science. 2024. Vol. 245. Article 113331.

Kotykhov A., Gubaev K., Sotskov V., Hodapp M., Shapeev A., Ivan S. Novikov

We propose a novel method for fitting machine-learning interatomic potentials with magnetic degrees of freedom, namely, magnetic Moment Tensor Potentials (mMTP). The main feature of the methodology consists in fitting mMTP to magnetic forces (negative derivatives of energies with respect to magnetic moments) derived from spin-polarized density functional theory calculations. We test our method on the bcc Fe-Al system with different composition. Specifically, we calculate formation energies, equilibrium lattice parameter, and total cell magnetization. Our findings demonstrate a precise match between values calculated with mMTP and those obtained by DFT at zero temperature. Additionally, using molecular dynamics, we estimate the finite temperature lattice parameter and capture the cell expansion as was previously revealed in experiment. We demonstrate that mMTPs fitted to magnetic forces, increase the relaxation reliability, which is the percent of successfully relaxed structures (i.e. with almost zero forces, stresses, and magnetic moments after the optimization of geometry). Eventually, we show that the proposed methodology can provide an accurate and reliable mMTP with reduced number of computationally complex spin-polarized density functional theory calculations.

Keywords: magnetism, magnetic machine-learning interatomic potentials, constrained density functional theory, molecular dy-namics