Assessment of model chemistries for noncovalent interactions

In the present study, we report tests of 57 model chemistry methods for calculating binding energies of 31 diverse van der Waals molecules arranged in five databases of noncovalent interaction energies. The model chemistries studied include wave function theory (WFT), density functional theory (DFT), and combined wave function-density-functional-theory (CWFDFT), and they include methods whose computational effort scales (for large systems) as N⁷, N⁶, N⁵, and N⁴, where N is the number of atoms. The model chemistries include 2 CWFDFT N⁷ models, 4 multilevel WFT N⁷ models, 5 single-level WFT N⁷ models, 4 CWFDFT N ⁶ models, 3 multilevel WFT N⁶ models, 11 single-level WFT N⁶ models, 5 CWFDFT N⁵ models, 10 single-level WFTN ⁵ models, 4 multilevel WFT N⁵ models, 4 single-level DFT N⁴ models, and 5 single-level WFT N⁴ models. We draw the following conclusions based on the mean absolute errors in 31 noncovalent binding energies: (1) MCG3-MPW gives the best performance for predicting the binding energies of these noncovalent complexes. (2) MCQCISD-MPWB and MCQCISD-MPW are the best two N⁶ methods. (3) M05-2X is the best single-level method for these noncovalent complexes. These four methods should facilitate useful calculations on a wide variety of practical applications involving hydrogen bonding, charge-transfer complexes, dipole interactions, weak (dispersion-like) interactions, and π⋯π stacking. If a user is interested in only a particular type of noncovalent interactions, though, some other methods, may be recommended for especially favorable performance/cost ratios. For example, BMC-CCSD has an outstanding performance for hydrogen bonding, and PWB6K has an outstanding cost-adjusted performance for dipole interaction calculations on very large systems. We also show that M05-2X performs well for interactions of amino acid pair residues.