Superconductivity in metal-insulator composite bands: A realization of negative-U pairing in purely repulsive systems

We consider the multiband, purely repulsive Hubbard model on a class of systems in which there are two (or more) kinds of orbitals per unit cell such that one metallic band interacts with the other insulating band. From both perturbational canonical transformation and quantum Monte Carlo studies we have shown that superconductivity due to nonretarded attraction (negative U) is a generic feature of this class of metal-insulator composite many-body systems. Remarkably, the superconductivity arises, under appropriate conditions, for various types of insulating bands (a charge-transfer insulator, Mott-Hubbard insulator, etc.). Superconductivity is unambiguously identified from the pairing correlation function, which agrees with that of the attractive Hubbard model surprisingly well even in the nonperturbative regime. The superconductivity revealed here is in a sense an opposite limit to conventional superconductivity with retarded attraction from boson-exchange mechanisms. We also show that the present mechanism is distinct from existing multiband models such as d-${\mathit{p}}_{\mathrm{\sigma }}$ models. As we increase the magnitude of the attraction in the present model by changing some parameters (e.g., to make the accompanying band metallic), the similarity with the attractive Hubbard model becomes eventually degraded, so that the optimum condition for superconductivity is determined by this kind of tradeoff. We also discuss the relevance of the present model to high-${\mathit{T}}_{\mathit{c}}$ superconductivity in cuprates and fullerenes.