[gmx-users] Pull code with shuffle/sort options

[Steve Fiedler]

Dear all,

I have encountered noticeable differences while using the pull code with 
the -shuffle -sort options in parallel, as opposed to running in serial 
or in parallel without the options. Using -shuffle -sort, the dynamics 
of the molecule of interest appears to be unaffected by the pulling 
potential. Without the options, the molecule can be seen to be correctly 
pulled in the desired direction. I'm using version 3.3.1, and this 
problem is present in both single and double precision versions.

One idea is that deshuf.ndx should be modified to include the pulled 
molecule reference group.  A search of the mailing list archive and 
manual didn't find this issue addressed.   Suggestions would be 
appreciated. 

A representative pull.ppa file used:
verbose = no
runtype = umbrella
group_1 = NNC
reference_group = REF_MCS
reftype = com
pulldim = N N Y
k1 = 5000.0
pos1 = 0 0 0.0

and the programs are invoked with:

grompp_d -sort -shuffle -np 4
mpirun -np 4 mdrun_d -v -pi pull.ppa -pn index.ndx

Regards,

Steve Fiedler