Electron correlation and properties of many-electron systems

Abstract

Different computational methods of quantum chemistry, which are currently employed in accurate calculations of atomic and molecular properties, are discussed and their advantages and disadvantages are investigated in calculations of electric properties of Be, FH, and H2O. A complete fourth-order many-body perturbation theory (MBPT) approach based on the coupled Hartree–Fock (CHF) solution for the perturbed problem reveals that in several cases the fourth-order correlation correction to the CHF result is dominated by those terms which involve triply substituted intermediate states. Moreover, in some cases a complete fourth-order MBPT result appears to be inferior to those obtained in approximate MBPT treatments which involve only certain classes of intermediate states. Hence, a complete fourth-order MBPT approach may not be sufficient for accurate predictions of atomic and molecular properties. The applicability of the complete active space (CAS) SCF method is studied in calculations of electric properties of FH. Although this method offers some advantages over the MBPT approach, its usefulness may be limited by the requirement of rather large active orbital spaces which have to be used in order to achieve sufficiently high accuracy. The CAS SCF results can be improved by using multireference CI methods with reference functions built of the CAS SCF natural orbitals. A good performance of the MBPT method limited to singly and doubly substituted intermediate states suggests that reasonable results can also be obtained in similar configuration interaction (CI) calculations, provided the latter are corrected for the erratic treatment of unlinked clusters. It is shown that the unlinked cluster corrections due to Davidson and Siegbahn result in reasonable estimates of the total correlation contribution to different properties. Because of its fairly routine character the CI method limited to single and double substitutions may in this way provide useful data with relatively small computational effort.