Optimal composition of atomic orbital basis sets for recovering static correlation energies

Static correlation energies (Estat) are calculated in a range of basis sets for a chemically diverse collection of atoms and molecules. The reliability of a basis set in capturing Estat is assessed according to the following: mean and maximum absolute deviations from near-exact Estat estimates, monotonic convergence to the complete basis set limit, and ability to capture Estat accurately independent of changes in geometry, molecular size, and electronic configuration. Within the polarization and correlation-consistent basis set series, triple-ζ basis sets are the smallest that can reliably capture Estat. The cc-pVTZ basis set performs particularly well, recovering Estat to chemical accuracy for all atoms and molecules in our data set. A series of customized basis sets are constructed by stripping polarization functions from, and swapping polarization functions among, existing basis sets. Basis sets without polarization functions are incapable of accurately recovering Estat. Basis sets with a near-complete set of s, p, and d functions can approach chemical accuracy in maximum absolute error. However, this may be achieved at lower computational cost by using a well balanced triple-ζ basis set including f functions, along with a smaller number of s, p, and d functions. Recommended basis sets for calculating Estat with increasing accuracy at increasing computational cost are 6-311G(2d,2p), cc-pVTZ, and cc-pVQZ stripped of g functions.