[gmx-users] replica exchange

Xavier Periole x.periole at rug.nl
Tue Nov 1 17:20:11 CET 2005

rob yang wrote:

> Hi gmxers,
> I have a general question about the implementation behind RE in 3.3.  
> if I were to run the following command:
> mpirun -np 2  mdrun_mpi -s input.tpr -np 2 -replex 1000
> where there are 2 .tpr files already made in the directory: 
> input0.tpr, input1.tpr
> and I have defined 2 independent computers linked by network.
> would gromacs run the two sim independently on the two computers and 
> only communicate with each other every 1000 steps? or
> would gromacs try to run 1 sim parallelly on both computers first and 
> then the other? ie: constant communication between the 2 computers 
> through network which would definately be slower.

first case.

> In case the above command makes no sense, here's what I want to do:
> given two sims differred only by the temperature, I'd like to run RE 
> using two boxes with the swapping rate being every 1000 steps. 
> Ideally, I'd like to have one sim per box, which only talks to each 
> other at the swapping time.
> p.s.  another q on how to analyze the results.  Should I concat 
> input0.trr and input1.trr into a single file at the end of the run? Or 
> treat them as two separate and independent runs?
That is your choice !! Depend on what your are interested in !! And 
mixing temperatures is not generally
what is interesting. One advantages of REMD is to generate properties 
over a range of temperatures...

> thank you very much in advance.
> rob


 Xavier Periole - Ph.D.

 Dept. of Biophysical Chemistry / MD Group   
 Univ. of Groningen
 Nijenborgh 4
 9747 AG Groningen
 The Netherlands
 Tel: +31-503634329
 Fax: +31-503634800
 email: x.periole at rug.nl
 web-page: http://md.chem.rug.nl/~periole

More information about the gromacs.org_gmx-users mailing list