Theoretical study of the geometrical and vibrational properties of aniline by different semiempirical and ab initio methods: The ground state, the T 1 triplet state and the D + radical cation AIP Conf.Direct dynamics calculations are performed, using the semiempirical neglect of diatomic differential overlap ͑NDDO͒ molecular orbital theory, to explore the level of electronic structure theory required to accurately describe the product energy partitioning when formaldehyde dissociates into hydrogen and carbon monoxide. Trajectories are initiated at the saddlepoint and are propagated for the short time needed to form products, by obtaining the energy and gradient directly from the NDDO theory. The resulting product energy partitioning is compared to available experimental data and the findings of two previous trajectory studies, including one ab initio trajectory study at the HF/6-31G** level of theory ͓Chem. Phys. Lett. 228, 436 ͑1994͔͒. The MNDO, AM1, and PM3 semiempirical Hamiltonians are studied, as well as Hamiltonians based on specific reaction parameters ͑SRP͒. For the latter, the original PM3 and AM1 parameters are adjusted to reproduce some ab initio potential energy surface properties, such as stationary points and part of the reaction path. A series of NDDO-SRP Hamiltonians are chosen by fitting different features of a HF/6-31G** potential energy surface. Only qualitative agreement is found between the product energy distributions of the NDDO-SRP Hamiltonians and that of the HF/6-31G** Hamiltonian. This result is consistent with the well known difficulty of reproducing a HF/6-31G** Hamiltonian with a NDDO-SRP model, since dynamic correlation is not treated in ab initio SCF, but is incorporated into semiempirical methods. Trajectory results with NDDO-SRP Hamiltonians, which reproduce a few experimental and/or high-level ab initio stationary points, are in poor agreement with the experimental product energy partitioning. Reparameterizing the NDDO Hamiltonian is laborious, and only a few properties of the potential energy surface can be reproduced at the same time. This indicates the limitations of the NDDO-SRP approach, which might be well suited for locally interpolating ab initio data, but not for quantitatively describing global potential energy surfaces.