Systematic MRD-CI calculations using the AM1 Hamiltonian have been carried out for two polyenes and eight aromatic hydrocarbons ranging from benzene to ovalene (C32H14). Twenty singlet-singlet excitation energies in these compounds were calculated and compared with experimental data and ab initio STO-3G results. On an absolute scale, the AMl/MRD-CI approach underestimates the excitation energies to states with dominant covalent character by an average of 1.1 eV, whereas the errors for ionic states are between -1.0 and 1.0 eV. The STO-3G calculated data are much too high by = 1 eV and = 5 eV, respectively. The inclusion of mr-correlation effects through second-order Epstein-Nesbet perturbation theory combined with the use of localized orbitals leads to a significant improvement of the ab initio calculated state energies. In an analogous AM1 treatment, negligible corrections for the mr correlations are found, which is attributed to the implicit account in the parameters and approximation of the semiempirical Hamiltonian. The possible error sources of the calculational methods are discussed. 0 1994 by John Wiley & Sons, Inc.
ground state properties including heats of formation, molecular geometries, dipole moments, and thermodynamic properties of large organic molecules. For excited states, only a few of such calodern semiempirical valence-electron meth-culations are available,6-" whereas the older M ods like MND0,I MNDOC,2 AMl,3 or CNDO/S12 and INDO/S13 methods are still in com-PM3,* which are based on the well-founded neglect mon use. The latter two methods have the imporof diatomic differential overlap approximation tant disadvantage that they are only parameterized (NDD0),5 have been very successful in calculating to calculate the excitation energies (i.e., an accurate