[gmx-users] parameters for torsion potental

[Victor Rosas Garcia]

Hello everybody,

Let's say that I have a torsion potential, calculated at an ab initio level
and I want to develop parameters for GROMACS.
1) As I understand, I have to develop a topology (used acpype for that)
2) Then I took the topology and found all the dihedrals that include my
torsion of interest, and set all those parameters to zero.
3) I then took all the structures used in the ab initio calculations, those
where the torsion of interest is systematically varied, and converted them
to PDB format.
4) then used all the pdb's to run single points with GROMACS and obtained a
"torsion" potential, that looks quite different from the ab initio
potential.
5) so then I got the difference between the ab initio potential and the
GROMACS potential, diff = Eabinitio - Egromacs
6) now I understand that, to get my parameters, I have to fit the energy
difference curve to some kind of analytic function, whether a
Ryckaert-Bellemans or a Fourier series.

My questions are:
a) Is the procedure I outlined correct or am I missing something?
b) when fitting the energy differences to the analytic function, does the
fitted function have to overlap the energy differences? or is it enough
that the curves be parallel?
c) are there any rules of thumb on how to choose the analytic function or
when to use tabulated potentials? e.g. in one case, I got a very shallow ab
initio potential, so that even when fitting to a 16-term Fourier series (a
type 9 dihedral), the fitted potential zigzags above and below the ab
initio potential, thus defining several unphysical barriers to rotation.
This makes me think a tabulated potential would be more adequate.

Any ideas?

Have a great day

Victor