bug in mdrun -rerun in gromacs 3.2.1 and 3.3; energies depend on box sizes and initial configuration in tpr file
I've been using the CVS version of Gromacs 3.2.1 from Nov. 2004 for some free energy calculations
(specifically, I've been using the version with the correct PME for free energy calculations), and I've
discovered a bug, which it turns out was previously reported on the list here, (http://
www.gromacs.org/pipermail/gmx-users/2003-December/008429.html) but apparently never fixed
(see this thread for recent discussion: http://www.gromacs.org/pipermail/gmx-users/2005-October/
Specifically, the bug affects mdrun -rerun and makes it so mdrun -rerun uses box information and
other information from the tpr file it begins from, NOT just the information from the trr (or xtc) file it is
run on. This means that the energies (almost all of the energies) from mdrun -rerun are sensitive to the
details of the intial coordinate .gro file used to generate the rerun tpr file. I've further found that this
sensitivity extends beyond just box information. For example, if I use two different .gro files for fairly
different configurations of the same system (with different box sizes) I get significantly different
energies in my .ene and .log files (when reprocessing a full precision trr file). If I change the box sizes
in the two .gro files to match, and repeat the whole test, energies are closer to agreeing, but still
Per David van der Spoel's request, I have uploaded a short (5 frame) trr file for a system I've tried this
on, as well as an initial tpr file I generated the trr file with, and a second tpr file I used to start the
-rerun case. These are here (http://dmobley.com/uploads/ligand_rerun.tpr), here (http://
dmobley.com/uploads/ligand_test0.tpr), and here (http://dmobley.com/uploads/ligand_test0.trr). The
"test0" is the initial run, and the "rerun" is the tpr file for mdrun -rerun. All options for the two are
identical; the only difference is the .gro files they were generated with.
These are generated with Gromacs 3.3 as I just wanted to make sure the bug wasn't fixed in Gromacs
3.3, although I am mainly using the Gromacs 3.2.1 CVS version with PME suitable for free energy. If
possible, I would like to get a bug fix for my 3.2.1 CVS version, as I'm using the Pande group's AMBER
force field, and their force field is not compatible with the new hdb file format in 3.3.
It would be nice to have this fixed as soon as possible, as we can't proceed on our project until it is
Thanks for your help.
#2 Updated by David van der Spoel over 14 years ago
Created an attachment (id=6)
Initialized code for rerun changed.
Please drop attached md.c in your src/kernel directory of a 3.3 installation
and recompile. The initialization in the case of rerun was wrong, based on the
coordinates and the box in the tpr file. Please compare the results of rerun
with your original simulation.