Basis Set Dependence of Interaction Energies Computed Using Composite Post-MP2 Methods.

Abstract

We report the performance of composite post-MP2 ab initio methods with small basis sets for description of noncovalent interactions, using the S66 data set as a benchmark. For three representative complexes, it is shown that explicitly correlated coupled cluster (CCSD-F12a) methods yield interaction energies ca. 0.1 kcal/mol from the complete basis set limit with aug-cc-pVDZ. Triple excitations are not explicitly correlated in this approach, but we show that scaling the perturbative triples via the (T*) approximation improves agreement with benchmark values. Across the entire S66 data set, this approach results in a root-mean-square error (RMSE) of 0.13 kcal/mol or 3%, with well-balanced description of all classes of complex. The basis set dependence of traditional CCSD(T) interaction energies is examined, and the small 6-31G*(0.25) basis set is found to give particularly accurate results (RMSE = 0.15 kcal/mol, or 4%). We also employ spin component scaling (SCS) of CCSD-F12a data, which gives slightly better accuracy than CCSD(T*)-F12a if contributions from same- and opposite-spin pairs are optimized for this data set (RMSE = 0.08 kcal/mol, or 2%). Interpolation of local MP2 and MP3 is also shown to accurately reproduce benchmark data with both aug-cc-pVDZ (RMSE = 0.18 kcal/mol or 5%) and 6-31G*(0.25) (RMSE = 0.13 kcal/mol or 4%).