[gmx-developers] Re: NMR refinement

[Bogdan Costescu]

On Thu, Jan 17, 2013 at 5:22 PM, santhosh <mesanthu at gmail.com> wrote:
> 1). The simulation is run with -multi N, N simulations running parallelly.
> (This is not replica exchange but a NMR refinement so -repex would *NOT*
> be
> there).
> In the custom code I require the replicaID to be sent as a identification
> and I am doing that by cr->ms->sim in md.c, (commrec.h has a struct
> t_commrec which has gmx_multisim_t which has sim as the identification of
> the replica). Please confirm that its right.

Sorry, I haven't looked at multisim, from just a quick look it seems
right. Maybe someone else who is familiar with this part of the code
can comment.

> 2). After every 1 step, I need to collect the current configuration of
> selected atoms.

This formulation is quite vague. Do you mean that you need to collect
the coordinates of the atoms  ? If so, I mentioned already the
trajectory writing code which can serve as inspiration.

> ( For now I am thinking of making nstxtcout be 1 and use
> trajconv to extract the last frame from that custom code) and compute
> force
> using custom code.

Be careful with writing out XTC. It seems you use the coordinates to
calculate forces - are the calculated forces not much sensitive to the
precision of coordinates ? If they are, then using full precision
trajectories (TRR) might be required.

> Since there is no place I am trying to sync both the replica's, both would
> run independently and the custom force computation would be based on what
> ever is the last written frame of the other replica(Need not be same step
> in
> the replicas??).
>
> Is that right ? And any hints on trying to make sure that the replica's
> are
> always in sync(same number of steps run in both the simulations) like NMR
> refinement protocol is.

It's not clear (at least to me) from the above paragraph whether you
do want the replicas to run in sync... Anyway, hopefully someone else
can offer more help in this area.

Cheers,
Bogdan