Hubbard U Corrections for 3d Transition-Metal Oxides within the SCAN+U Framework in Density Functional Theory

Abstract

Redox-active 3d transition-metal oxides (TMOs) are crucial ingredients for multiple sustainable energy applications, including solar cells, batteries, catalysis, and solar thermochemical water splitting. However, any predictive modeling, such as those employing density functional theory, needs to describe accurately the energetics of redox reactions involving transition metals, if new candidate materials are to be identified in a reliable fashion. In the present work, we identify optimal U values for other 3d TMOs, specifically Ti, V, Cr, Co, Ni, and Cu, within the SCAN+U framework. We determine optimal U values of 2.5, 1.0, 3.0, and 2.5 eV for Ti, V, Co, and Ni oxides, respectively, while Cr and Cu oxides best reproduce redox thermodynamics without any U correction at all. While the U values required for Ti, V, Co, and Ni are lower than those needed within the generalized gradient approximation (GGA)+U or local density approximation (LDA)+U approaches, inclusion of U makes non-negligible improvements in ground-state property evaluations of these oxides. Here we also validate our optimal U values by performing a number of transferability checks for each 3d material. We further apply the optimized SCAN+U framework with and without long-range dispersion corrections to evaluate the cathode properties of layered LiMO2 (M = V, Cr, Mn, Fe, Co, Ni, and Cu) systems. Specifically, we assess the predicted interlayer spacing, topotactic voltages, stabilities, and electronic structure compared to both experiment and DFT-PBE-based calculations. We find that SCAN-based functionals predict larger voltages due to underestimating the stabilities of the MO2 systems. Furthermore, adding dispersion corrections to PBE has a greater effect on voltage predictions than with SCAN, indicating that DFT-SCAN – despite being a ground-state theory – captures some short and medium range dispersion interactions and does so better than DFT-PBE. The predicted electronic structure is consistent between SCAN-based and PBE-based functionals, with the exception of LiFeO2.