Local Correlation Calculations Using Standard and Renormalized Coupled-Cluster Methods

摘要

This article discusses our recent effort toward the extension of the linear scaling local correlation approach, termed 'cluster-in-molecule' and abbreviated as CIM [S. Li, J. Ma, and Y. Jiang, Comput. Chem. 23, 237 (2002); S. Li, J. Shen, W. Li, and Y. Jiang, Chem. Phys. 125, 074109 (2006)1 to the coupled-cluster (CC) theory with singles and doubles (CCSD) and CC methods with singles, doubles, and non-iterative triples, including the standard CCSD(T) approach and the completely renormalized CR-CC(2,3) scheme [P. Piecuch and M. W/och, J. Chem. Phys. 123, 224105 (2005); P. Piecuch, M. Wloch, J.R. Gour, and A. Kinal, Chem. Phys. Lett. 418, 467 (2006)]. As in the earlier CIM work that dealt with the second-order many-body perturbation theory and CC doubles approach, the main idea of the CIM-CCSD, CIM-CCSD(T), and CIM-CR-CC(2,3) methods is the realization of the fact that the total correlation energy of a large system can be obtained as a sum of contributions from the occupied orthonormal localized molecular orbitals and their respective occupied and unoccupied orbital domains. The CIMCCSD, CIM-CCSD(T), and CIM-CR-CC(2,3) methods pursued in this work are characterized by high computational efficiency in both the CIM and CC parts, enabling calculations for much larger systems than previously possible. This is achieved by combining the natural linear scaling and embarrassing parallelism of the CIM ansatz with the vectorized CC codes that rely on recursively generated intermediates and fast matrix multiplication routines. By comparing the results of the canonical and CIM-CC calculations for normal alkanes and water clusters, it is demonstrated that the CIM-CCSD, CIM-CCSD(T), and CIM-CR-CC(2,3) approaches recover the corresponding canonical CC correlation energies to within 0.1 % or so, while offering linear scaling of the computer costs with the systeth size and savings in the computer effort by orders of magnitude. By examining the dissociation of dodecane into C 1 1H23 and CH3 and several lowest-energy structures of the (H 20)clusters, it is shown that the CIM-CC methods accurately reproduce the relative energetics of the corresponding canonical CC calculations.