THEORETICAL-STUDIES OF HYDROGEN-BONDED COMPLEXES USING SEMIEMPIRICAL METHODS

Recently proposed modified MNDO methods (MNDO/M and MNDO-PM3) for the study of hydrogen-bonded complexes are tested in four simple systems: (H2O)2, NH4 + · H2O, formamide water and formamide-methanol dimers. A comparison is made with AM1 since this method, intended to be a general improvement of MNDO, was already shown to fail in the former two dimers. Bifurcated hydrogen-bonded structures were always preferred over linear ones. AM1 also shows this artifact in the formamide-water dimer but not in the formamide-methanol one. This is due to the inability of -CH3 to bind to the carbonyl oxygen. In this case the AM1 optimized global minimum is in excellent agreement with the Hartree-Fock ab initio DZP optimized structure. Moreover, when it is physically impossible to form bifurcated hydrogen bonds, AM1 is in good agreement with ab initio results. MNDO/M and MNDO-PM3 (an improvement on AM1 ) do not give any bifurcated structure either in the former two dimers or in the formamide complexes. MNDO/M and MNDO-PM3 structures are in agreement with ab initio and experimental ones in almost all cases, except for the global minima in both the formamide-water and the formamide- methanol dimers. Singly hydrogen-bonded structures are predicted instead of the experimentally determined and ab initio calculated correct doubly hydrogen-bonded ones. MNDO-PM3, although not entirely correct, gives structures which are more correct than the MNDO/M ones. In conclusion, neither of these semiempirical methods is entirely correct. Overall, MNDO-PM3 seems to be more trustworthy than AM1 or MNDO/M, although these methods may be used when no bifurcated hydrogen bonds (AM1) or cycles (MNDO//M) exist. Failures of MNDO-PM3 indicate that a better semiempirical method is still needed for studies of solvation and hydrogen- bonded clusters. © 1990.