Density functional plus dynamical mean field theory of the metal-insulator transition in early transition metal oxides

TL;DR

This study combines density functional theory and dynamical mean-field theory to analyze the metal-insulator transition in early transition-metal oxides, highlighting the importance of charge-transfer physics and the impact of double-counting corrections.

Contribution

It demonstrates the effectiveness of DFT+DMFT with full charge self-consistency in predicting properties of transition-metal oxides, and clarifies the role of charge-transfer effects and model choices.

Findings

Standard DFT+DMFT predicts some materials as metals incorrectly.

Charge-transfer physics significantly influences the properties of early transition-metal oxides.

Proper double-counting correction aligns theoretical predictions with experimental spectra.

Abstract

The combination of density functional theory and single-site dynamical mean-field theory, using both Hartree and full continuous-time quantum Monte Carlo impurity solvers, is used to study the metal-insulator phase diagram of perovskite transition-metal oxides of the form $AB$O$_3$ with a rare-earth ion $A$=Sr, La, Y and transition metal $B$=Ti, V, Cr. The correlated subspace is constructed from atomiclike $d$ orbitals defined using maximally localized Wannier functions derived from the full $p$-$d$ manifold; for comparison, results obtained using a projector method are also given. Paramagnetic DFT+DMFT computations using full charge self-consistency along with the standard "fully localized limit" (FLL) double counting are shown to incorrectly predict that LaTiO$_3$, YTiO$_3$, LaVO$_3$ and SrMnO$_3$ are metals. A more general examination of the dependence of physical properties on the mean $p$-$d$ energy splitting, the occupancy of the correlated $d$ states, the double-counting correction, and the lattice structure demonstrates the importance of charge-transfer physics even in the early transition-metal oxides and elucidates the factors underlying the failure of the standard approximations. If the double counting is chosen to produce a $p$-$d$ splitting consistent with experimental spectra, single-site dynamical mean-field theory provides a reasonable account of the materials properties. The relation of the results to those obtained from "$d$-only" models in which the correlation problem is based on the frontier orbital $p$-$d$ antibonding bands is determined. It is found that if an effective interaction $U$ is properly chosen the $d$-only model provides a good account of the physics of the $d^1$ and $d^2$ materials.