Divide-and-conquer-based linear-scaling approach for traditional and renormalized coupled cluster methods with single, double, and noniterative triple excitations

We have reported the divide-and-conquer (DC)-based linear-scaling correlation treatment of coupled-cluster method with single and double excitations (CCSD) [Kobayashi and Nakai, J. Chem. Phys. 129, 044103 (2009)]. In the DC-CCSD method, the CCSD equations derived from subsystem orbitals are solved for each subsystem in order to obtain the total correlation energy by summing up subsystem contributions using energy density analysis. In this study, we extend the DC-CCSD method for treating noniterative perturbative triple excitations using CCSD T₁ and T₂ amplitudes, namely, CCSD(T). In the DC-CCSD(T) method, the so-called (T) corrections are also computed for each subsystem. Numerical assessments indicate that DC-CCSD(T) reproduces the CCSD(T) results with high accuracy and significantly less computational cost. We further extend the DC-based correlation method to renormalized CCSD(T) [Kowalski and Piecuch, J. Chem. Phys. 113, 18 (2000)] for avoiding the divergence that occurs in multireference problems such as bond dissociation.