[gmx-users] Replica exchange shell script

[Mark Abraham]

Sukit Leekumjorn wrote:
> Thank you very much Mark
> 
> I got REMD to run now but I ended up having the same problem as the 
> other users had in the past.  Below is the error message.  I got this to 
> run for 2 nodes.  For 4, 6, and 8 nodes, the below error shows up in the 
> log files.  I did try several suggestions from the mailing list, but I 
> could not get  it to runs more than 2 nodes.  Any suggestion in fixing 
> this problem would be greatly appreciated.  Here is the executable 
> command line I used:

Is there anything about your hardware that might be having an effect 
after 2 nodes/processors?

> export EXEC="$FF/mdrun_3.3_gcc_mpi -multi -np 8 -replex 200 -s dpp.tpr 
> -deffnm dpp"
> 
> I did try removing -multi but did not work.  I also "grompp" the .mdp in 
> the double precision mode (floating problem), but still did not work.

What was your .mdp file and how did you call grompp?

> Repl      0     1     2     3     4     5     6     7 Repl  T 340.0 
> 345.0 350.0 355.0 360.0 365.0 370.0 375.0

As an aside, a linear increase of temperatures will see relatively many 
swaps between higher pairs of temperatures than lower pairs, because the 
distribution of pairs of ensembles that differ by Delta T overlap more 
with increasing T. You might want this for some reason I can't think of 
at the moment, but an exponential increase of temperatures will see a 
(more) linear distribution of exchange frequencies over the T range. You 
also get to access higher temperatures for a given ratio of the lowest 
two temperatures. Generating an exponential sequences is easy of course, 
with T3 = T2 * (T2/T1)

Mark