[gmx-users] Re: Parameterizing a force field (Mark Abraham)

[Mark Abraham]

Lee-Ping wrote:
> Thank you both for the response!  I looked at the -j option in mdrun,
> but it looks like that has about as much documentation as ffscan (i.e.
> none at all).  I also fruitlessly checked the mailing list archives for
> the past year and a half; there are some questions about ffscan, but
> none of which contained instructions on how to run it.  
> 
> The next step is to look inside the gromacs source code, but there are a
> lot of different files and I'm not sure as to which pieces of source
> code describe how the input files are parsed.  If you could tell me
> which files to look at, I would appreciate it a lot.  Thanks again!

Mostly, files that parse files end in io.c

> Specifically:
> 1) What piece of code do I look at to see how mdrun parses the wham.gct
> file (turned on by the -j switch)?

Don't know. Start with src/kernel/mdrun.c and see what it calls to parse 
this file.

> 2) Which file describes how ffscan parses its inputs?

Start with src/kernel/ffscan.c and see what it calls :-)

> Also: If ffscan performs a grid scan, is it really able to fit 100+
> parameters in a reasonable period of time (couple of hours)?  Would a
> numerical optimization procedure for the parameters be a possible
> improvement over this?

ffscan doesn't do a fit, per se, it just does a grid scan and reports 
suitable values (see man ffscan). It'd have to be heaps faster than your 
alternative scripts calling grompp and mdrun to do your series of single 
point energy evaluations.

Mark