Electrostatics-based finite-size correction for first-principles point defect calculations

Abstract: Finite-size corrections for charged defect supercell calculations typically consist of image-charge and potential alignment corrections. A wide variety of schemes for both corrections have been proposed for decades. Regarding the image-charge correction, Freysoldt, Neugebauer, and Van de Walle (FNV) recently proposed a novel method that enables us to accurately estimate the correction energy a posteriori through alignment of the defect-induced potential to the model charge potential [Freysoldt et al., Phys. Rev. Lett. 102, 016402 (2009)]. This method, however, still has two issues in practice. Firstly, it uses planar-averaged potential for determining the potential offset, which cannot be readily applied to relaxed system. Secondly, the long-range Coulomb interaction is assumed to be screened by a macroscopic dielectric constant. This is valid only for cubic systems and can bring forth huge errors for defects in anisotropic materials. In this study, we use the atomic site electrostatic potential as a potential marker instead of the planar-averaged potential, and extend the FNV scheme by adopting the point charge model in an anisotropic medium for estimating long-range interactions. We also revisit the conventional potential alignment correction and show that it is fully included in the image-charge correction and therefore unnecessary. In addition, we show that the potential alignment corresponds to a part of first-order and full of third-order image-charge correction; thus the third-order image-charge contribution is absent after the potential alignment. Finally, a systematic assessment of the accuracy of the extended FNV correction scheme is performed for a wide range of material classes. The defect formation energies calculated using around 100-atom supercells are successfully corrected even after atomic relaxation within a few tenths of eV compared to those in the dilute limit.