[gmx-users] REMD: after replica exchange breakdown

[Mark Abraham]

On 11/05/2011 10:59 PM, SebastianWaltz wrote:
> On 05/11/2011 05:20 AM, Mark Abraham wrote:
>> On 05/11/2011 02:32 AM, SebastianWaltz wrote:
>>> Dear GROMACS User,
>>>
>>> I'm running a REMD MD of 32 replicas. It worked for the first few
>>> replica exchanges. After 30ps again two replicas were exchanged and I
>>> received the Error:
>>> x particles communicated to PME node y are more than a cell length out
>>> of the domain decomposition cell of their charge group
>> The only rationale I can see is that a velocity was scaled after
>> exchange that was now large enough to lead to the normal sort of
>> exploding system. Was the exploding replica going up in temperature?
>> How did you equilibrate the systems?
> I did a NPT equilibration for each replica which leads obviously to
> different volumes and looks like to cause some problems. If I use the
> same volume for each replica it works fine.

Yeah, that'd be it. Do the NPT equilibration at your target temperature, 
and then construct the other replicas to have that volume, before 
equilibrating each replica at its temperature under NVT.

>>> I am using NVT assembles for each replica with a exchanging interval of
>>> 5ps. For each replica I use 4 processors. Constraint are only the bonds
>>> to hydrogens (constraints = hbonds). Using PME for electro static.
>> Information like GROMACS version, simulation system dimensions and
>> contents would be useful. What replica temperature distribution exists?
> I am using version 4.0.7. My system consists of a small peptide soluted
> in ~1500 rigid CHCl3 molecules. System equilibrated with NPT and NVT
> production run seems to work only if I increase the number of replicas
> (32->64) meaning that I also decrease the difference of the volumes. The
> temperature is exponentially distributed from 200 to 450K.

Seems OK.

Mark

>>> When running each replica for its own everything is working fine,
>>> nothing is blowing up like one would expect after getting the above
>>> Error.
>> Good trouble-shooting :) You'd have been told to try this otherwise...
>>
>> Mark
>>
> Thanks a lot
>
> Sebastian