Probing the LDA-1/2 method as a starting point for G₀W₀ calculations

Employing the $G_0W_0$ approximation of Hedin's $GW$ approach one can obtain quasi-particle energies of extended systems and molecules with good accuracy. However, for many materials, semi-local exchange-correlation functionals are unsatisfactory starting points for $G_0W_0$ calculations. Hybrid functionals often improve upon them, but at a substantially higher computational cost. As an alternative, we suggest the LDA-1/2 method, which provides reasonable band gaps, without being computationally involved. In this work, we systematically compare 3 starting points for $G_0W_0$: LDA, PBE0, and LDA-1/2. A selection of solids is chosen for this benchmark: C, Si, SiC, AlP, LiF, MgO, Ne, Ar, GaN, GaAs, CdS, ZnS, and ZnO. We demonstrate that LDA-1/2 is a good starting point in most cases, reducing the mean absolute error of band gaps by 50% when compared to the other 2 functionals.