[gmx-developers] Error with octahedral box and domain decomposition in CVS-version

Jochen Hub jhub at gwdg.de
Tue May 20 09:02:40 CEST 2008


Hi,

when running a octahedral system of 31779 atoms on 16 or 8 CPUs with 
domain decomposition I am getting the errors like shown below. Is that a 
known problem or has it even been fixed in the current  CVS version? 
Should I submit a bugzilla? Playing with the -dd parameters or switching 
-dbl on an off did not help. Btw, the charge group which seems to move 
too rapidly is always a water molecule. The system (protein+water+ions) 
is OK, it is runnig without problems in gromacs 3.3.1.

Cheers, Jochen


Making 3D domain decomposition 2 x 2 x 2
starting mdrun 'Protein'
100000000 steps, 200000.0 ps.

step 0

Step 30:
The charge group starting at atom 21975 moved more than the cut-off 
(1.000000) in direction x
distance out of cell -1.110520
Old coordinates:    0.602    1.641    0.007
New coordinates:   -1.668    4.849    5.566
Old cell boundaries in direction x:    0.000    3.257
New cell boundaries in direction x:    0.000    3.225
-------------------------------------------------------
Program mdrun, VERSION 3.3.99_development_20070618
Source code file: domdec.c, line: 1806

Fatal error:
A charge group move too far between two domain decomposition steps
-------------------------------------------------------



-- 
************************************************
Dr. Jochen Hub
Max Planck Institute for Biophysical Chemistry
Computational biomolecular dynamics group
Am Fassberg 11
D-37077 Goettingen, Germany
Email: jhub[at]gwdg.de
Tel.: +49 (0)551 201-2312
************************************************ 




More information about the gromacs.org_gmx-developers mailing list