We iteratively improve the accuracy of semiempirical methods for the evaluation of noncovalent interactions. This is carried out using a database of reference DFT+D and coupled-cluster (CC) intermolecular interaction energies calculated for a set of specially chosen small molecules representing important functional groups.
Corrections to the semiempirical models are evaluated from this data set, and analytical correction functions are obtained through fittings. Subsequent data sets containing larger molecules, with more complex functional units, are added and the corresponding fits carried out. After validation, the method is applied to the study of other systems.