Reducing Self-Interaction Error in Transition-Metal Oxides with Different Exact-Exchange Fractions for Energy and Density

Abstract: DFT is vital for materials discovery, and at the base of extensive molecular and materials databases, chemical reaction predictions, and machine learning potentials. The widespread use of DFT in chemistry and materials science aims for "chemical accuracy," but this is limited by the unknown exchange and correlation (XC) functional. A meta-GGA, the restored regularized strongly constrained and appropriately normed, r r2SCAN XC functional, fulfils 17 exact constraints of the XC energy. r2SCAN still appears inadequate at predicting material properties of strongly correlated compounds. Inaccuracies of r2SCAN arise from functional and density-driven errors, linked to the self-interaction error. We introduce a new method, r2SCANY@r2SCANX, for simulating transition metal oxides accurately. r2SCANY@r2SCANX utilizes different fractions of exact exchange: X to set the electronic density, and Y to set the energy density functional approximation. r2SCANY@r2SCANX addresses functional-driven and density-driven inaccuracies. Using just one or two universal parameters, r2SCANY@r2SCANX enhances the r2SCAN predictions of the properties of 18 correlated oxides, outperforming the highly parameterized DFT+U method. The O2 overbinding in r2SCAN (~0.3 eV/O2) reduces to just ~0.03 eV/O$_2$ with any X in r2SCAN10@r2SCANX. Uncertainties for oxide oxidation energies and magnetic moments are reduced by r2SCAN10@r2SCAN50, minimizing r2SCAN density-driven errors. The computationally efficient r2SCAN10@r2SCAN is nearly as accurate as the hybrid r2SCAN10 for oxidation energies. Thus, accurate energy differences can be achieved by rate-limiting self-consistent iterations and geometry optimizations with the efficient r2SCAN. Subsequently, expensive hybrid functionals can be applied in a fast-to-execute single post-self-consistent calculation, as in r2SCAN10@r2SCAN, which is 10 to 300 times faster than r2SCAN10.