REG-CCSD(T): A New Efficient CCSD(T) Approach Accelerated by the Regularized Fock and Intermediate Excitation Matrix

In this work, the REG-CCSD(T) method, which is the extension of the coupled cluster (CC) with single (S), double (D), and perturbative triple (T) excitations accelerated by regularizing (REG) the Fock and intermediate excitation matrix, is newly developed for molecules based on the canonical CCSD(T) references. The new technique performs the outstanding efficiency of the CCSD(T) method by adding the regularization parameter to solve the positive definiteness and invertibility issues of the matrix during the calculations. The accuracy of the REG-CCSD(T) method is extensively tested on several open-shell and closed-shell molecules, with a mean absolute deviation (MAD) value of calculated energy as only 4.06 × 10^-4 kcal/mol compared with the results of the canonical CCSD(T) method. The REG-CCSD(T) method with density-fitting REGDF-CCSD(T), with frozen-core approximations REGFC-CCSD(T), and the REGDF-FC-CCSD(T) method were also developed in this work as the extensions of the REG-CCSD(T) approach, respectively. The average time (in seconds) for the calculation of each benchmarked complex is ranked as REGFC-CCSD(T) (1.23) < REG-CCSD(T) (1.56) < REGDF-FC-CCSD(T) (2.70) < REGDF-CCSD(T) (2.76) < CCSD(T) (16.32), using the cc-pVDZ basis set and 28 parallel cores for calculations. Among these five CCSD(T) methods, the REG-CCSD(T) approach is recommended since it balances the accuracy and efficiency of calculations. This work provides a new efficient CCSD(T) approach named REG-CCSD(T) and its several variants, which could be used in the calculations of quantum chemistry requiring high accuracy and efficiency with affordable computational costs.