[gmx-users] For non-zero entries for table x, the forces deviate from minus the numerical derivative of the potential

[Douwe Pollmann]

Dear Gromacs users,

I am currently trying to implement a user-defined dihedral into a coarse
grained model (in version 5.1.4), but I get the following error:

'For the 35988 non-zero entries for table 0 in tables/table_d4.xvg the
forces deviate on average 196% from minus the numerical derivative of the
potential'.

The three columns in table_d4 are the angle [-180, 180], the potential
energy (in kJ/mol) and minus the derivative (first I had +derivative, which
resulted in an error or 204%). If I am correct, Gromacs interpolates its
own function through the data points given in table_d4, but the result is
far from correct in this case.

I have tried to use -debug during mdrun to get the interpolated values of
Gromacs, but I only get the values (ctab.xvg, dtab.xvg and rtab.xvg) of the
non-bonded potentials, while this dihedral is of course a bonded potential.

In section 6.10.1 of the manual (version 5.1.4) it is mentioned that
'Gromacs stores A0, A1, A2 and A3', the parameters of the interpolated
cubic spline. My question is if any of you knows if it is possible to get
these parameters, or if it is possible to get the functions interpolated by
Gromacs, to check where the mistakes are made.
If this is not possible at all, do you have any ideas how I can decrease
the deviation of the forces? 0% deviation is not needed, but 200% is a bit
too much..

Thank you very much in advance!

Kind regards,
Douwe Pollmann