[gmx-users] Updating charges on-the-fly

[Mark Abraham]

On 18/10/2011 5:40 AM, J. Nathan Scott wrote:
> Hello fellow GMX users,
>
> I've been digging in the archives and haven't yet found a good 
> response to this question, so I thought I'd ask again, since I see 
> from a post just last month others are still interested in this 
> subject as well. Is it possible in Gromacs to periodically 
> update/change the charges on a given molecule or residue during the 
> course of a simulation?

Not currently, but it would not be hard to do.

> What I am envisioning is having a production simulation running, and 
> then every "x" amount of time, the current checkpoint file gets read 
> and processed by an external routine that calculates charges on the 
> residue or molecule of interest, which are then fed back into Gromacs 
> to continue the run. Is it possible to execute an external program 
> from within a Gromacs simulation? I know this functionality must be in 
> there somewhere, since Gromacs can communicate with external QM 
> software, but I'm not sure if a charge modification scheme could make 
> use of the same built-in I/O methods.

GROMACS is a normal C program, and can execute an external program. 
Obviously this will still be inefficient if GROMACS was running in 
parallel, since you have to have an I/O phase before and after the 
external program and probably the other processors cannot do anything 
useful. You would then have to re-distribute the charge vector over any 
GROMACS parallel nodes.

>
> Is such a scheme possible? I realize an alternative method would be to 
> stop the run, do some calculation to recalculate the charges, and then 
> perhaps use some simple code to modify the .top file directly, and 
> finally grompp out a new tpr file and continue the run. Lather, rinse, 
> repeat. I'm guessing this could be rather slow because of needing to 
> call grompp over and over and the need to continually restart the 
> simulation.

This would be slower still, but reliable and easy for proving the method.

>
> For what it's worth, my interest lies mainly in excited state 
> chromophores and I'd be using modified ZINDO code to calculate charges 
> every x number of steps. This calculation is *very* fast, even for 
> large chromophores, therefore my focus on other performance bottlenecks.

Ideally, GROMACS code would be linked with ZINDO and call the correct 
function directly. This shouldn't be too bad since you merely need an 
array of coordinates (and some other data which can be hard-coded 
somehow) as input, and the charges out. You should probably get one or 
other of the crude hacks working first, to judge how much value the 
speed-up might have.

Mark

>
> I would be extremely grateful for any feedback on this subject, 
> particularly from anyone who has done a similar thing successfully and 
> cares to share any tips or code.
>
> -- 
> ----------
> J. Nathan Scott, Ph.D.
> Postdoctoral Fellow
> Department of Chemistry and Biochemistry
> Montana State University
>
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20111018/bfd3190f/attachment.html>