Anomalous delocalization of resonant states in graphene & the vacancy magnetic moment

Carbon atom vacancies in graphene give rise to a local magnetic moment of $\sigma+\pi$ origin, whose magnitude is yet uncertain and debated. Partial quenching of π magnetism has been ubiquitously reported in periodic first principles calculations, with magnetic moments scattered in the range 1.0–2.0 µB, slowly converging to the lower or the upper end, depending on how the diluted limit is approached. By contrast, (ensemble) density functional theory calculations on cluster models neatly converge to the value of $2~\mu_\textrm{B}$ when increasing the system size. This stunning discrepancy has sparked a debate about the role of defect–defect interactions and self-doping, and about the importance of the self-interaction-error in the density-functional-theory description of the vacancy-induced states. Here, we settle this puzzle by showing that the problem has a fundamental, mono-electronic origin which is related to the special (periodic) arrangement of defects that results when using the slab-supercell approach. Specifically, we report the existence of resonant states that are anomalously delocalized over the lattice and that make the π midgap band unphysically dispersive, hence prone to self-doping and quenching of the π magnetism. Hybrid functionals fix the problem by widening the gap between the spin-resolved π midgap bands, without reducing their artificial widths. As a consequence, while reconciling the magnetic moment with expectations, they predict a spin-splitting which is one order of magnitude larger than found in experiments.