[gmx-users] problem with center of mass removal, rest-group, all gmx versions

[Michael Brunsteiner]

David van der Spoel wrote:

> If your system contains 6508 this is the whole system and that should 
> not be the cause of the lincs problem.

obviously this is how far you got in reading my mail ;) ... since
a few lines further down (see below) i wrote explicitly that my
system is much larger than 6508 atoms ... anyway, I don't think
the COM removal is the reason why my system blows up, this is probably
merely a book-keeping problem, but its strange anyway, i wonder where
gmx takes this number from... a domain decomposition issue ?? (i run 
parallel jobs on four nodes, and my system has 26030 atoms, which is
about four times as many atoms as gmx seems to believe there are ...



>
> in the mdp-documentation it says:
>
> comm_grps:
>     group(s) for center of mass motion removal, default is the whole system
>
> I only now noticed that i never set comm_grps explicitly,
> assuming that the default (System) is ok, the comm_mode i typically
> set to Linear and nst_comm to either 100 or 1, or I leave these
> commands out altogether assuming that the defaults are ok ...
> system blows up in each case.
> My system has much more than 6508 atoms, and none of the groups in the
> ndx file contains 6508 atoms, and i double and triple checked that
> there is not a sigle atom that is NOT member of some group in the
> ndx file ... so what is this "rest" group??  it is apparently
> made automatically by grompp or mdrun. I can imagine that constraining
> the center of mass of a (possibly randomly chosen) subgroup of
> atoms can cause instabilities, has anybody observed something
> like this before ??
> I have seen a couple of mails in this list in which people describe
> similar problems, but found none with a solution/explanation.
>
> I should add that i run gromcas on an AMD64 PC,
> linux 2.6.9. I also tried a 32 bit machine, and I tried gmx 3.3.2,
> 3.3.3, 4.0 and 4.0.2 ... always the same problem.
>
> any help is appreciated! (sample mdp file below)
>
> cheers,
> Michael