Finite-size supercell correction schemes for charged defect calculations

Various schemes for correcting the finite-size supercell errors in the case of charged defect calculations are analyzed and their performance for a series of defect systems is compared. We focus on the schemes proposed by Makov and Payne (MP), Freysoldt, Neugebauer, and Van de Walle (FNV), and Lany and Zunger (LZ). The role of the potential alignment is also assessed. We demonstrate a connection between the defect charge distribution and the potential alignment, which establishes a relation between the MP and FNV schemes. Calculations are performed using supercells of various sizes and the corrected formation energies are compared to the values obtained by extrapolation to infinitely large supercells. For defects with localized charge distributions, we generally find that the FNV scheme improves upon the LZ one, while the MP scheme tends to overcorrect except for point-charge-like defects. We also encountered a class of defects, for which all the correction schemes fail to produce results consistent with the extrapolated values. This behavior is found to be caused by partial delocalization of the defect charge. We associate this effect to hybridization between the defect state and the band-edge states of the host. The occurrence of defect charge delocalization also reflects in the evolution of the defect Kohn-Sham levels with increasing supercell size. We discuss the physical relevance of the latter class of defects.