First row benchmark tests of the parametrized configuration interaction with parameter X (PCI-X) scheme

Abstract

A recently suggested scheme termed parametrized configuration interaction with parameter X (PCI‐X) for scaling the correlation energy has been applied on a benchmark test consisting of 32 first row molecules. Several different methods like Mo/ller–Plesset second‐order perturbation theory (MP2), modified coupled pair functional method (MCPF), averaged coupled pair functional method (ACPF), coupled cluster singles and doubles (CCSD), and CCSD with a perturbational estimate of triple excitations [CCSD(T)] have been tested using systematically chosen basis sets ranging from double zeta (DZ) to very large atomic natural orbital (ANO) sets containing several sets of d and f functions. For each method and basis set the scaling parameter is optimized. The scaling does in all cases lead to large, sometimes dramatic, improvements of the results. Typically, using a single reference state method like MCPF the average absolute deviation compared to experiments for the benchmark goes from an unscaled value of about 20 kcal/mol down to about 2 kcal/mol. For MCPF and similar methods no improvement of the results is obtained going beyond the DZ+polarization (DZP) level. Significant improvements due to scaling occurs even at the highest level using the CCSD(T) method and the largest basis set. For medium size basis sets the present scaling is far superior to the extrapolation schemes used in the Gaussian‐1 and ‐2 (G1 and G2) theories.