A Benchmark of GW Methods for Azabenzenes: Is the GW Approximation Good Enough?

TL;DR

This paper evaluates various GW approximation methods for azabenzenes, comparing their accuracy in predicting electronic spectra against experimental data, and finds that less self-consistent methods sometimes outperform more complex ones.

Contribution

It provides a comprehensive benchmark of GW methods for azabenzenes, revealing that higher self-consistency does not always lead to better accuracy.

Findings

G0W0 and scGW0 often outperform fully self-consistent scGW in spectral predictions.

The accuracy of GW methods varies significantly depending on the starting point.

Less self-consistent GW methods can sometimes provide better agreement with experiments.

Abstract

Many-body perturbation theory in the GW approximation is a useful method for describing electronic properties associated with charged excitations. A hierarchy of GW methods exists, starting from non-self-consistent G0W0, through partial self-consistency in the eigenvalues (ev-scGW) and in the Green function (scGW0), to fully self-consistent GW (scGW). Here, we assess the performance of these methods for benzene, pyridine, and the diazines. The quasiparticle spectra are compared to photoemission spectroscopy (PES) experiments with respect to all measured particle removal energies and the ordering of the frontier orbitals. We find that the accuracy of the calculated spectra does not match the expectations based on their level of self-consistency. In particular, for certain starting points G0W0 and scGW0 provide spectra in better agreement with the PES than scGW.