A straightforward a posteriori method for reduction of density-fitting error in coupled-cluster calculations

Abstract: We present a simple method for \emph{a posteriori} removal of a significant fraction of the density-fitting error from the calculated total coupled-cluster energies. The method treats the difference between the exact and density-fitted integrals as a perturbation, and simplified response-like equations allow to calculate improved amplitudes and the corresponding energy correction. The proposed method is tested at the coupled-cluster singles and doubles level of theory for a diverse set of moderately-sized molecules. On average, error reductions by a factor of approximately ten and twenty are observed in double-zeta and triple-zeta basis sets, respectively. Similar reductions are observed in calculations of interaction energies of several model complexes. The computational cost of the procedure is small in comparison with the preceding coupled-cluster iterations. The applicability of the method is not limited to the density-fitting approximation; in principle, it can be used in conjunction with an arbitrary decomposition scheme of the electron repulsion integrals.