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

TL;DR

This paper introduces a novel method, r$^2$SCANY@r$^2$SCANX, which uses different fractions of exact exchange to reduce self-interaction errors in DFT calculations for transition-metal oxides, improving accuracy over existing methods.

Contribution

The paper proposes a new approach that employs different exact exchange fractions for density and energy, enhancing predictions for strongly correlated transition-metal oxides.

Findings

r$^2$SCANY@r$^2$SCANX outperforms existing methods on 20 oxides.

It reduces prediction errors for magnetic moments and band gaps.

The method improves upon the state-of-the-art DFT+U approach.

Abstract

Density functional theory (DFT) in chemistry and materials science aims for "chemical accuracy," but this goal is challenged by the need to approximate the exact exchange-correlation (XC) energy functional. The r$^2$SCAN, meta-generalized gradient approximation to the XC functional fulfills 17 exact constraints of the XC energy, and has significantly boosted prediction accuracy for molecules and materials. However, r$^2$SCAN remains inadequate at predicting properties of open \textit{d} and \textit{f} transition-metal strongly correlated compounds, such as band gaps, magnetic moments, and oxidation energies. Prediction inaccuracies of r$^2$SCAN energies arise from functional and density-driven errors, mainly resulting from the DFT self-interaction error. We propose the r$^2$SCANY@r$^2$SCANX method to mitigate the self-interaction error of XC functionals for the accurate simulations of electronic, magnetic, and thermochemical properties of transition metal oxides. r$^2$SCANY@r$^2$SCANX uses different fractions of exact Hartree-Fock exchange: X for the electronic density and Y for the density functional approximation of the total energy, thereby simultaneously addressing functional-driven and density-driven inaccuracies. Building just on 1 (or maximum 2) parameters that apply unchanged to \emph{s-p}-bonded systems, we demonstrate that, r$^2$SCANY@r$^2$SCANX improves upon the r$^2$SCAN predictions for 20 highly correlated oxides and even outperforms the highly parameterized DFT(r$^2$SCAN)+\emph{U} method -- the state-of-the-art approach to predict strongly correlated materials. Prediction uncertainties for oxidation energies and magnetic moments of transition metal oxides are significantly reduced by r$^2$SCAN10@r$^2$SCAN50 and band gaps with r$^2$SCAN10@r$^2$SCAN. r$^2$SCAN10@r$^2$SCAN50 diminishes the density-driven error of the energy in r$^2$SCAN and r$^2$SCAN10.