Cubic scaling algorithm for the random phase approximation: Self-interstitials and vacancies in Si

The random phase approximation (RPA) to the correlation energy is among the most promising methods to obtain accurate correlation energy differences from diagrammatic perturbation theory at modest computational cost. We show here that a cubic system size scaling can be readily obtained, which dramatically reduces the computation time by one to two orders of magnitude for large systems. Furthermore, the scaling with respect to the number of k points used to sample the Brillouin zone can be reduced to linear order. In combination, this allows accurate and very well-converged single-point RPA calculations, with a time complexity that is roughly on par or better than for self-consistent Hartree-Fock and hybrid-functional calculations. The present implementation enables new applications. Here, we apply the RPA to determine the energy difference between diamond Si and beta-tin Si, the energetics of the Si self-interstitial defect and the Si vacancy, the latter with up to 256 atom supercells. We show that the RPA predicts Si interstitial and vacancy energies in excellent agreement with experiment. Si self-interstitial diffusion barriers are also in good agreement with experiment, as opposed to previous calculations based on hybrid functionals or range-separated RPA variants.