A critical comparison of mindo/3, mndo, am1, and pm3 for a model problem - carbon clusters c2-c10. An ad hoc reparametrization of mndo well suited for the accurate prediction of their spectroscopic constants

Abstract

Dissociation energies and potential energy surface features for the carbon clusters C2 to C-10 are compared with ab initio or experimental results for the semiempirical methods MINDO/3, MNDO, AM1 and PM3. Quite surprisingly, MINDO/3 gives a rather good account of the various structures and electronic states, unlike the other three methods. MINDO/3 tends towards systematic overestimate of binding energies, the other methods to systematic gross underestimates. Reparametrization of the diatomic parameters alpha, beta-s, beta-p for exact reproduction of the experimental data for C3 results in much improved values for binding energies, but fails to correct the state splittings. Also reparametrizing U(ss), U(pp), xi-s, and xi-p to reproduce the ab initio linear-rhombic energy difference in C4 results in a much improved description of the other states. For the linear structures, computed harmonic frequencies with the latter parameters are in surprisingly good agreement with experimental or correlated ab initio data, where available; experimental values are consistently overestimated by about 40 cm-1. Other results are comparable in quality to good ab initio treatments. The experimental IR bands at 2128 and 1892 cm-1, formerly assigned to C-9, should be reassigned to linear C7. The intense 1997 cm-1 feature almost certainly belongs to C-9; bands at 1952 and 1197 cm-1 both belong to linear C6. Tentative assignments of bands in the 1600-1850 cm-1 region to various cyclic structures of C6, C-8, and C-10 have been made. As such, this suggests a new and promising procedure for the theoretical study of large molecules in general, and of large clusters in particular.