Local methods are particularly useful for the calculation of weak intermolecular interactions since the basis set superposition error (BSSE) is largely reduced [1,13,14] and counterpoise corrections are usually not necessary (provided the BSSE of the underlying Hartree-Fock is small). However, one must be careful to define the domains properly and to include all intermolecular pairs at the highest computational level. A convenient way to define appropriate domains and pair lists is to use the option INTERACT=1. If this option is given, individual molecules are identified automatically and all intermolecular pairs are automatically treated as strong pairs and included in the LCCSD. Similarly, appropriate triples lists are generated for LCCSD(T) calculations. It is required that all orbital domains are located on individual molecules. Note however that the inclusion of the intermolecular pairs strongly increases the number of strong pairs and triples, and therefore high-level calculations can become very expensive.
For calculations of interaction potentials of weakly interacting systems, the domains of the subsystems should be determined at a very large distance and saved using the SAVE=record option on the LOCAL or MULTP directive, or the SAVE directive (see section 32.8.3). If the asymptotic energy is not needed it is sufficient to do this initial calculation using option DOMONLY=1). These domains should then be reused in the subsequent calculations at all other intermolecular distances by using the START=record option or the START directive (see section 32.8.4). Only in this way the basis set superposition error is minimized and normally negligible (of course, this does not affect the BSSE for the SCF, and therefore the basis set should be sufficiently large to make the SCF BSSE negligible).
Usually, diffuse basis functions are important for obtaining accurate intermolecular interactions. When diffuse basis sets are used, it may happen that the Pipek-Mezey localization does not yield well localized orbitals. This problem can in most cases be overcome by using the directive
as described in section 32.9.3
A final warning concerns local density fitting (see sections 32.10 and 15): local fitting must not be used in counterpoise calculations, since no fitting functions would be present on the dummy atoms and this can lead to large errors.
For examples and discussions of these aspects see Refs. [21-23]