Efficient calculation of the Coulomb matrix and its expansion around k=0 within the FLAPW method

TL;DR

This paper presents an efficient, accurate method for calculating the Coulomb matrix within the FLAPW framework, including an analytic expansion around k=0 to handle divergences, improving electronic structure calculations.

Contribution

It introduces a novel analytic expansion of the Coulomb matrix around k=0 in the FLAPW method, avoiding projection onto plane waves and enhancing computational efficiency.

Findings

Efficient Coulomb matrix calculation with improved accuracy.

Successful application to electron-energy-loss spectra of nickel.

Good agreement with experimental spectra, including core transitions.

Abstract

We derive formulas for the Coulomb matrix within the full-potential linearized augmented-plane-wave (FLAPW) method. The Coulomb matrix is a central ingredient in implementations of many-body perturbation theory, such as the Hartree-Fock and GW approximations for the electronic self-energy or the random-phase approximation for the dielectric function. It is represented in the mixed product basis, which combines numerical muffin-tin functions and interstitial plane waves that are here expanded with the Rayleigh formula. The resulting algorithm is very efficient in terms of both computational cost and accuracy and is superior to an implementation with the Fourier transform of the step function. In order to allow an analytic treatment of the divergence at k=0 in reciprocal space, we expand the Coulomb matrix analytically around this point without resorting to a projection onto plane waves. We then apply a basis transformation that diagonalizes the Coulomb matrix and confines the divergence to a single eigenvalue. At the same time, response matrices like the dielectric function separate into head, wings, and body with the same mathematical properties as in a plane-wave basis. As an illustration we apply the formulas to electron-energy-loss spectra for nickel at different k vectors including k=0. The convergence of the spectra towards the result at k=0 is clearly seen. Our treatment also allows to include transitions from core states that give rise to a shallow peak at high energies and lead to good agreement with experiment.