[gmx-users] Re: Problem with REMD

[Mark Abraham]

On 26/07/2012 8:17 PM, ppxasjsm wrote:
> Hi Mark,
>
>> You're not the first to make this wrong assumption, but GROMACS actually
>> exchanges coordinates among (sets of) cpus, which write to the same
>> output file, i.e. the trajectories written by mdrun are already at
>> constant temperature. (That's inefficient because you have to re-do the
>> domain decomposition, and is likely an artefact of the pre-existing
>> mdrun -multi functionality from which the REMD implementation derives. I
>> have a private GROMACS version off 4.5.5 that scales better to large
>> processor sets by exchanging only T (and some minor details) and only
>> doing pairwise communication during exchange attempts - however if it
>> sees the light of day, it would be in at least GROMACS 5.0.)
> Ok if i understand this correctly, my output files md0.xtc, md1.xtc ...
> correspond to the original temperatures set in md0.mdp, md1.mdp etc..., so I
> can just straight away build my Ramachandran diagran from these.
>
> Then I don't quite understand what the script demux.pl actually does

It does the converse of the operation you thought it did.

>   and why
> I need replica_index.xvg and replica_temp.xvg and what  trjcat -f *.xtc
> -demux replica_index.xvg gives me.

Nobody said you did need them.

> It seems very counterintuitive to change configurations rather than
> temperatures, it would be nice if this was explicitly stated in the REMD
> gromacs reference page.

http://www.gromacs.org/Documentation/How-tos/REMD#Post-Processing says this.

> @Mark, if you do have a version, where temperatures are exchanged instead,
> would it be somehow possible to obtain this?

No, I don't have the time right now to make even an alpha-quality 
version available and roughly documented. It'd only be useful if you 
were running on hundreds of processors for hundreds of nanoseconds, and 
likely only under noisy network conditions, anyway.

>
> Thanks for resolving this.
>
> One last question regarding this. The energy, which is used for calculating
> the exchange probabilities, does this correspond to the potential energy,
> which can be extracted from g_energy -f file.edr < 11 0, or is it a
> different energy?

If nstenergy is a suitable value wrt mdrun -replex, it will be the 
potential energy in the .edr file, which you can approximately verify 
from the .log file.

Mark