As a follow up to my post (pasted below), I am providing some more details. 

1. I am applying forces on heavy atoms (water oxygens and protein
non-hydrogens). 
2. I tried switching of SHAKE. Results improve marginally, but the 8
processor data are still wrong.   

3. There is no pressure coupling.  I have left a vapor layer of a few
nanometers to allow for any 
density changes. 

I hope this further information helps in debugging. 

Thanks
Sumanth Jamadagni

Dear Gromacs community, 

 I modified the pull code in GROMACS to apply forces on atoms that are in 
a certain region of 
space.  It compiles error free and runs fine to give me the expected
results. I then recompiled with 
mpi support. The cluster I am using has 8 cores/processors per node. When 
I use 1, 2 or 4 nodes 
to run, I get identical results, but the results are very different when
I run parallel on all 8 cores. 

1. I am not sure how the forces I apply in the pull code are parallelized. 
2. Is this easy for me to edit in the source code or is it best if I use  
only upto 4 processors for my 
job? 
3. Will the number of processors that I am able to use change depending
on the system size? 

I currently have on protein and about 7500 water molecules to give a
total of about 25000 atoms. I am using SHAKE with constraints on h-bonds. 

Thanks for any help!
Sumanth 


Sumanth N Jamadagni
Graduate Student 
Isermann Dept of Chemical and Biological Engg
Rensselaer Polytechnic Institute

jamads@rpi.edu
(Cell)518-598-2786

http://boyle.che.rpi.edu/~sumanth