Comparative ab initio and semi-empirical study of hydrogen bonded complexes of NH 3 and H 2O

Abstract

Recently developed semiempirical methods (AM1, MNDO-PM3 and MNDO/M) are tested by comparison with ab initio Hartree-Fock and Møller-Plesset 4-31G and 6-311G (d,p) calculations on cuts of the potential energy hyperfaces of the hydrogen bonded dimers H 2O·H 2O, NH 3·NH 3 and NH 3·H 2O. AM1 gives a bifurcatedly hydrogen-bonded complex for the water dimer, as already published, whereas it gives trifurcated structures for the other two dimers. There is no indication of a quasi-linear hydrogen bonded dimer in these last two cases, in agreement with the ab initio calculations. Moreover, there is a secondary minimum in the ammonia dinner surface, with a cyclic structure, which closely resembles that obtained in the ab initio calculations. MNDO-PM3 gives the quasi-linear hydrogen bonded structure of the water dimer as the correct one. For NH 3·NH 3 the global minimum is no longer the trifurcated structure (which is, however, a secondary minimum) but a cyclic one. However, the ring is so closed that hydrogen atoms of both monomers are almost head-to-head. For NH 3·H 2O a further minimum appears, which is now the global one, corresponding to the linearly hydrogen-bonded structure of the dimer. The aspect of this attractor, as well as the aspect of the cut in the PES of the water dimer, suggests that undue importance has been given to linear hydrogen bonds in the parametrization of MNDO-PM3. MNDO/M behaves in a very similar way to MNDO-PM3 and, owing to its better definition, the latter should be preferred. In conclusion, none of these methods is truly reliable for calculations on weakly bound dimers. MNDO-PM3 is better than AMI for the water and the ammonia-water dimers but not for the ammonia dimer; MNDO/M is no better than MNDO-PM3.