Is there a guide-line for choosing methods/basis-set in generating new parameters from QM calculations? - This might be important, since mixing of basis and/or methods ususally yield different results, and hence consistency may be sacrificed.
When generating new parameters to be consistent with the CHARMM22 or CHARMM27
additive models there are two areas to consider; nonbonded parameters versus
internal parameters. With respect to internal parameters, use the highest level
of theory accessible for the compounds you are working on. This will give the
best geometries, vibrations and, most importantly, conformational energies.
For the nonbond parameters; meaning the charges, the HF/6-31G* level
should be used to calculate interactions of your model compound with water.
The resulting interaction energies (properly scaled, see our papers or my
web page) and distances are then used as the target data to adjust the
charges. Since the HF/6-31G* level of theory has been used for the
remainder of the force field charge optimization, it is important that be used
for new compounds. Note that it is generally safe to transfer the Lennard-
Jones parameters from appropriate atoms already in the force field.
it's certainly a relief to learn that parameters for new compounds should be optimized at the same level of theory. However, in some cases the effort in spending intensive CPU:time pays off very little when compared to smaller basis, and for metals of third and fourth row, the basis-sets are highly limited. A good choice in this aspect seems to be the MP2/LANL2DZ combination