On the accuracy of ab initio force constant calculations with respect to basis sets

Abstract

Gradients and stretching force constants for N{sub 2}, O{sub 2}, FH, and F{sub 2} were calculated at the experimentally determined equilibrium bond lengths. Several large basis sets, including the atomic natural orbital (ANO) basis proposed by Almloef and Taylor, were used in multireference CISD calculations. The superposition contributions to the gradients and force constants were obtained from counterpoise calculations. The variations in the gradients and force constants obtained with these basis sets of TZ2P quality and higher were in general at the 1% level or greater. The superposition error in the MR CISD correlation energy was large enough in most cases to shift the equilibrium geometries by several thousands of an angstrom. Their results contribute to the growing body of information suggesting that the prediction of quadratic stretching force constants to within 1% and equilibrium bond lengths to within {plus minus} 0.002 {angstrom} is beyond the predictive ability of TZ2P (or TX2P + f) basis sets. When such accuracy is claimed with these relatively small basis sets, it is very likely due to a cancellation of errors.