Projector self-consistent DFT+U using non-orthogonal generalized Wannier functions

TL;DR

This paper introduces a self-consistent DFT+U method using non-orthogonal generalized Wannier functions as projectors, improving the definition of correlated subspaces and enabling large-scale calculations.

Contribution

It proposes a novel self-consistent approach for DFT+U using NGWFs, reducing arbitrariness and enhancing applicability to large systems.

Findings

Converges reliably with iterative refinement of projectors

Produces ground-state properties comparable to hydrogenic projectors

Applicable within linear-scaling DFT frameworks for large systems

Abstract

We present a formulation of the density-functional theory + Hubbard model (DFT+U) method that is self-consistent over the choice of Hubbard projectors used to define the correlated subspaces. In order to overcome the arbitrariness in this choice, we propose the use of non-orthogonal generalized Wannier functions (NGWFs) as projectors for the DFT+U correction. We iteratively refine these NGWF projectors and, hence, the DFT+U functional, such that the correlated subspaces are fully self-consistent with the DFT+U ground-state. We discuss the convergence characteristics of this algorithm and compare ground-state properties thus computed with those calculated using hydrogenic projectors. Our approach is implemented within, but not restricted to, a linear-scaling DFT framework, opening the path to DFT+U calculations on systems of unprecedented size.