[gmx-users] Parameterizing a force field

[Mark Abraham]

Lee-Ping wrote:
> Hi there,
> 
> I'd like to fit a set of force field parameters (100+) to a set of
> several ab initio force calculations.  I've noticed that ffscan is
> tailored to this purpose, but I don't know how to use it, so I wrote a
> script that tries to fit the force field parameters using some numerical
> optimization techniques.
> 
> Every time I vary one of the force field parameters, I need to call
> grompp and mdrun and run a single-point calculation to get the force.
> (a small detail: I use "rerun" so I only need to call mdrun once for a
> batch of geometries with the parameters fixed).  The optimization
> techniques all work by varying the force field parameters and comparing
> the forces that come out; thus, I need to call grompp and mdrun many
> thousands of times, with the only thing modified being the force field
> parameters.
> 
> It turns out that repeatedly calling mdrun and grompp uses up a lot of
> computer resources, and takes up most of the time involved in the
> optimization.  Is there a method out there for rapidly running several
> force calculations on the same molecular geometry/topology, with only
> the force field parameters modified (perhaps with only a single call of
> grompp or mdrun)?

Technical point - you want to run several energy calculations, not force 
calculations - but you know that :-)

ffscan is the tool you want to use, but other than reading its man page 
and then its code, I can only suggest searching the mailing list for 
help with its use. If you come up blank, please ask again, and hopefully 
a developer can come up with some documentation. This is a little-used 
area of gromacs, so please be patient. :-)

Mark