Dielectric anisotropy in the GW space-time method

TL;DR

This paper extends the GW space-time method to better handle dielectric anisotropy in periodic systems, improving numerical stability and convergence for strongly anisotropic materials like slabs and multilayers.

Contribution

It introduces an anisotropic extension of the GW space-time method and efficient computation of non-local pseudopotential contributions, enhancing calculations for anisotropic systems.

Findings

Improved convergence in GW calculations for anisotropic materials.

Enhanced numerical stability for slab and multilayer systems.

More efficient treatment of dielectric anisotropy in reciprocal space.

Abstract

Excited-state calculations, notably for quasiparticle band structures, are nowadays routinely performed within the GW approximation for the electronic self-energy. Nevertheless, certain numerical approximations and simplifications are still employed in practice to make the computations feasible. An important aspect for periodic systems is the proper treatment of the singularity of the screened Coulomb interaction in reciprocal space, which results from the slow 1/r decay in real space. This must be done without introducing artificial interactions between the quasiparticles and their periodic images in repeated cells, which occur when integrals of the screened Coulomb interaction are discretised in reciprocal space. An adequate treatment of both aspects is crucial for a numerically stable computation of the self-energy. In this article we build on existing schemes for isotropic screening and present an extension for anisotropic systems. We also show how the contributions to the dielectric function arising from the non-local part of the pseudopotentials can be computed efficiently. These improvements are crucial for obtaining a fast convergence with respect to the number of points used for the Brillouin zone integration and prove to be essential to make GW calculations for strongly anisotropic systems, such as slabs or multilayers, efficient.