[gmx-developers] Shift for MARTINI force field?

[Berk Hess]

David van der Spoel wrote:
> Michael Shirts wrote:
>> I noticed this in a paper from van Gundsteren's group:
>> http://dx.doi.org/0.1039/b818713d.
>>
>> "Since this shift function was erroneously described in the GROMACS
>> manual and also erroneously implemented in the GROMACS  program, as
>> has been extensively investigated and reported in ref. 25, we did not
>> want to use exactly the same (inconsistent)
>> shift function, but the shift function as implemented in the  GROMOS05
>> biomolecular simulation software."
>>
>> Just checking, this has been addressed now, correct?
>
> Interesting, for your information I published the correct shift
> function in 2006 (unnoted by Van Gunsteren) in
> http://dx.doi.org/10.1021/ct0502256 (note that this also contains the
> first publication of the OPLS switch function :) along with a whole
> range of other commonly used shift functions). I think (but I'm
> biased) that the equations in that paper are correct, and I'm pretty
> sure the implementation at that time (gmx 3.3) matches the equations,
> but this could easily be checked by running mdrun -debug with shift
> functions, since this will dump the tables, don't have time right now,
> any volunteers? That means that the only thing to fix is the manual.
The code gives (and has always given, as far as I know) continuous
potential, force and second derivative.
The formula in the manual does not describe correctly what the code does.
In that sense the manual is erroneous.
Also note that the dispersion correction is always correct, since it use
the spline table data as input.
But I don't see how the code is erroneous.
There is surely no discontinuity.
I would say that the minimum demands on a shifting function are that the
potential
is continuous and differentiable. Since a shifting function is non-physical,
you are completely free to choose the shape of it as you want.

Berk