The CCSD(T) model with Cholesky decomposition of orbital energy denominators

A new implementation of the coupled cluster singles and doubles with approximate triples correction method [CCSD(T)] using Cholesky decomposition of the orbital energy denominators is described. The new algorithm reduces the scaling of CCSD(T) from N7 to N6, where N is the number of orbitals. The Cholesky decomposition is carried out using simple analytical expressions that allow us to evaluate a priori the order in which the decomposition should be carried out and to obtain the relevant parts of the vectors whenever needed in the calculation. Several benchmarks have been carried out comparing the performance of the conventional and Cholesky CCSD(T) implementations. The Cholesky implementation shows a speed-up factor larger than O2/V, where O is the number of occupied and V the number of virtual orbitals, and in general at most 5 vectors are needed to get a precision of μEh. We demonstrate that the Cholesky algorithm is better suited for studying large systems. © 2010 Wiley Periodicals, Inc.