[gmx-users] Local Coordinates with DomDec

Bradley Treece btreece at andrew.cmu.edu
Mon Oct 9 16:50:58 CEST 2017

Hi All,

I have a question about the handling of coordinates in local arrays with
domain decomposition. It may take a moment to get to my actual question,
but I want to make sure to be thorough with my explanation. Also, if I need
to point this question toward a different mailing list, please let me know.

By referencing the data structure reference page, I am aware that the
mdatoms structure contains nr and homenr indicating the number of local and
home atoms dealt with in the arrays x, f, and etc. I also know how to use
the commrec to get back the global indices. This part may not be crucial to
my question, but I also would like to check that I have this right. I
searched high and low for the mdatoms->start integer but couldn't find it
anywhere. I assume, and through careful checks, feel confident that this is
part of an older version and it is the case that home atoms are contained
in the first homenr atoms of an array; meaning the start parameter was
somehow tied to the location of home atoms in the arrays but is now

My question is in regard to local atoms that are not part of the home set.
Many of the position and force values for these behave as expected, but
occasionally the three vectors for both are only zeros (almost as if they
were uninitialized). I know for topological reasons that these atoms should
not have coordinates of zero and presume that their forces should not be
zero. I also have verified that they do fall below the nr cutoff of the
index. I was wondering if anyone could point me to why this might be the
case or correct any misunderstanding I have for how domdec manages these

For reference, most of my testing takes place within the do_force function
of src/gromacs/mdlib/sim_util.cpp  but I have done some checks elsewhere in
the code to verify that it is not an issue isolated to this part of the
code. I am also running 5.1.1 in case there are any version dependencies to
take into account.

Thanks for your help,

Bradley Treece

More information about the gromacs.org_gmx-users mailing list