Formation energies of charged defects in two-dimensional materials -- resolution of long-standing difficulties (1911.11192v1)

Abstract: Formation energies of charged point defects in semiconductors are calculated using periodic supercells, which entail a divergence arising from long-range Coulombic interactions. The divergence is typically removed by the so-called jellium approach. Recently, Wu, Zhang and Pantelides [WZP, Phys. Rev. Lett. 119, 105501 (2017)] traced the origin of the divergence to the assumption that charged defects are formed by physically removing electrons from or adding electrons to the crystal, violating charge neutrality, a key principle of statistical mechanics that determines the Fermi level. An alternative theory was constructed by recognizing that "charged" defects form by trading carriers with the energy bands, whereby supercells are always charge-neutral so that no divergence is present and no ad-hoc procedures need to be adopted for calculations. Here we give a more detailed exposition of the foundations of both methods and show that the jellium approach can be derived from the statistical-mechanics-backed WZP definition by steps whose validity cannot be assessed a priori. In particular, the divergence appears when the charge density of band carriers is dropped, leaving a supercharged crystal. In the case of charged defects in two-dimensional (2D) materials, unphysical fields appear in vacuum regions. None of these pathological features are present in the reformulated theory. Finally, we report new calculations in both bulk and 2D materials. The WZP approach yields formation energies that differ from jellium values by up to ~1 eV. By analyzing the spatial distribution of wave functions and defect potentials, we provide insights into the inner workings of both methods and demonstrate that the failure of the jellium approach to include the neutralizing electron density of band carriers, as is the case in the physical system, is responsible for the numerical differences between the two methods.