Optimization strategies developed on NiO for Heisenberg exchange coupling calculations using projector augmented wave based first-principles DFT+U+J

Abstract

High-performance batteries, heterogeneous catalysts and next-generation photovoltaics often centrally involve transition metal oxides (TMOs) that undergo charge or spin-state changes. Demand for accurate DFT modeling of TMOs has increased in recent years, driving improved quantification and correction schemes for approximate DFT's characteristic errors, notably those pertaining to self-interaction and static correlation. Of considerable interest, meanwhile, is the use of DFT-accessible quantities to compute parameters of coarse-grained models such as for magnetism. To understand the interference of error corrections and model mappings, we probe the prototypical Mott-Hubbard insulator NiO, calculating its electronic structure in its antiferromagnetic I/II and ferromagnetic states. We examine the pronounced sensitivity of the first principles calculated Hubbard U and Hund's J parameters to choices concerning Projector Augmented Wave (PAW) based population analysis, we reevaluate spin quantification conventions for the Heisenberg model, and we seek to develop best practices for calculating Hubbard parameters specific to energetically meta-stable magnetic orderings of TMOs. Within this framework, we assess several corrective functionals using in situ calculated U and J parameters, e.g., DFT+U and DFT+U+J. We find that while using a straightforward workflow with minimal empiricism, the NiO Heisenberg parameter RMS error with respect to experiment was reduced to 13%, an advance upon the state-of-the-art. Methodologically, we used a linear-response implementation for calculating the Hubbard U available in the open-source plane-wave DFT code Abinit. We have extended its utility to calculate the Hund's exchange coupling J, however our findings are anticipated to be applicable to any DFT+U implementation.