A correction method for vertical transition levels (VTLs) involving defect states calculated with a supercell technique is formulated and its effectiveness is systematically verified with 10 defects in prototypical materials: MgO, GaN, cubic-BN, and 3C-SiC. Within point-charge approximations, the corrected VTLs for cation vacancies are found to well reproduce their dilute limits, while they do not work as satisfactorily for anion vacancies. In contrast, when the finite defect size contribution is incorporated, the mean absolute errors are reduced and become less than 0.1 eV with moderate size supercells in all cases. Our correction scheme is a powerful technique because it is adaptive for evaluating various quantities at fixed geometry, as represented by those relevant to the generalized Koopmans' theorem.Point defects determine various physical properties in solids. One of the most important properties is the optical property. For instance, ZnO is known to show green luminescence, which is seemingly attributed to point defects [1]. Another example is SrTiO 3 , known to show blue-light emission at room temperature after Ar + -irradiation, whose origin is presumably the emergence of oxygen vacancies [2]. In addition, we can investigate defects with deep states, which often degrade the device performance, from the photo-absorption and -emission spectra. It is generally considered that absorption (emission) of a photon by a defect promotes (demotes) an electron to the excited (ground) state, most probably without altering the atomic configuration, based on the Franck-Condon principle [3]. This is a consequence of the fact that electrons are much lighter than nuclei. Thus, we can represent the optical transition by a vertical arrow in the configuration coordination diagram, and its transition energy is given as an optical transition level or vertical transition level (VTL) with respect to the valence band maximum (VBM) or the conduction band minimum (CBM) [4,5].First-principles calculations have become powerful tools to understand and predict the defect-mediated optical transitions. In the calculations of extended systems with defects, these levels are almost always evaluated by a supercell approach nowadays, where a charged defect interacts with its periodic images and background charge, which erroneously modifies the total energies of charged defect supercells [6,7]. Methods to correct the energies to the dilute limit are well established, as represented by the scheme proposed by Freysoldt, Neugebauer, and Van de Walle (FNV) [8], and its extension to anisotropic systems and/or relaxed geometries (eFNV) [9]. The correction energy of the eFNV scheme is written aswhere E PC is the point charge (PC) correction energy, * kumagai@msl.titech.ac.jp ∆V PC,q/b | far is the potential difference between defect-induced and PC potentials at a region outside of the defect in the supercell, and q is the defect charge state. The second term is referred to as an alignment-like term, and is approximately equal to the Makov-Payne thir...