Guy Moore1,2,Matthew K. Horton1,Kristin Persson1
Lawrence Berkeley National Laboratory1,University of California, Berkeley2
Guy Moore1,2,Matthew K. Horton1,Kristin Persson1
Lawrence Berkeley National Laboratory1,University of California, Berkeley2
Moving from the atomistic picture of magnetism to larger length scale models is an important challenge for the design and discovery of technologically relevant magnetic materials. This problem requires increased care for correlated electron systems, such as transition metal oxides, in which multi-body interactions are difficult to model using conventional density functional theory (DFT). In the extension to DFT+U+J, Hubbard U and Hund J values account for on-site coulomb interactions between localized electrons. These Hubbard U and Hund J values are computed using a customized linear response computational workflow suitable for high-throughput DFT applications. We present a framework for obtaining magnetic exchange constants from DFT+U+J using the established single-particle Green’s function approach, which can be used to study finite-temperature behavior of lattice models using Monte Carlo methods. The Heisenberg exchange constants are highly sensitive to two important prerequisites: the magnetic ground-state, as well as U and J values. We explore the sensitivity of the magnetic ground state and resulting exchange constants to U and J values. Additionally, we have implemented and benchmarked a “source-free” exchange-correlation magnetic field in VASP. This source-free functional, paired with a custom particle swarm optimization strategy, SpinPSO, has resulted in improved agreement with a variety of magnetic ground-states measured by neutron diffraction. This ground-up computational approach will allow for the discovery of magnetic materials with technological applications ranging from spintronics to cost-effective magnetocaloric materials for magnetic refrigeration.