Reliability of HF/IQA, B3LYP/IQA, and MP2/IQA data in interpreting the nature and strength of interactions

Mainly focusing on the B3LYP level, the reliability of level of theory (LoT)-dependent IQA-defined energy terms (self- and additive atomic energies, interaction energy and its components) in interpreting interactions was investigated at three LoTs using default settings in AIMAll software explicitly implementing the actual B3LYP exchange-correlation functional. Reliability was quantified using relative errors (REs) defined as, for example, RE = B3LYP/IQA(computed) - B3LYP/IQA(expected), using reference CCSD/BBC1/IQA data to obtain the LoT/IQA(expected) terms. On average, B3LYP produced the most accurate IQA energies among the LoTs investigated, affording REs an order of magnitude smaller than those at the HF level. The B3LYP/IQA description of the O4H6 and O3O4 interactions in glycol conformers compared well with the CCSD/BBC1/IQA-generated picture. Exceptionally reliable data were obtained at the B3LYP level for changes in the IQA energies computed for structural changes in glycol. The FAMSEC-based interpretation produced exact qualitative description and perfectly quantitatively comparable values to CCSD/BBC1/IQA data. The ΔEIQA = ΔE criterion (representing changes between the final and any suitably selected initial structure of a molecular system) is reported to 'validate' or predict the usefulness of changes in the LoT/IQA computed energy terms with respect to interaction interpretability. This was supported by (i) the smallest errors in the IQA energy changes being obtained at MP2/Müller, despite EIQA largely overestimating E (by -170 kcal mol-1 due to large REs in self-atomic energies) and (ii) reasonable HF-generated FAMSEC descriptors, regardless of the largest REs in the HF/IQA data. Finally, adding Grimme's D3 empirical dispersion correction had no significant effect on the REs.