Bug #2330

Lost particles while sorting Gromacs 2016.4

Added by Soumadwip Ghosh over 2 years ago. Updated over 2 years ago.

Target version:
Affected version - extra info:
Affected version:


Software Inconsistency Error:

Lost particles while sorting (6 out of 8)

Happens during NPT equilibration on 16 processors of standard GPUs. Affected version is 2016.4. The job however runs on an older gromacs 5.1.0 (double precision)without any hitches. Another thing to note is that after the NPT step (run with 5.1.0) is over if one creates a new run.tpr file for the production run using the 2016.4 version it runs fine. The production simulation runs fine on the native gromacs 5.1.0 version as well. It seems like the NPT step is getting screwed up in 2016.4. The system consists of a membrane protein in detergent micelle having a total number of atoms ~ 88747.

npt5k.tpr (3.48 MB) npt5k.tpr Soumadwip Ghosh, 12/05/2017 11:42 PM

Associated revisions

Revision 011eec72 (diff)
Added by Berk Hess over 2 years ago

Add grompp note for PR pcouple + position restraints

Refs #2330

Change-Id: I299e39247d69214d9b54c47874d522ac7b00a30e


#1 Updated by Mark Abraham over 2 years ago

Thanks for the report. Can we please have a log file from one run that works in 5.1 (note that 5.1.0 does not exist), and one which does not in 2016.4?

#2 Updated by Mark Abraham over 2 years ago

My general suggestion to you is that pressure coupling with parrinello-rahmann is known to be unstable when the volume has a significant change - thus might work or not with different versions or inputs arbitrarily. Does that apply in your case? What happens if you follow general practice and correct the volume with berendsen pressure coupling first?

#3 Updated by Berk Hess over 2 years ago

  • Status changed from New to Rejected

You are running Parrinello-Rahman pressure coupling together with position restraints. This is very likely the source of instability. You should not use Parrinello-Rahman for equilibration and certainly not combined with position restraints.
Please try this with Berendsen pressure coupling instead. I am marking this issue as rejected, but please reply when it also fails with Berendsen.
We could consider adding a grompp warning when combining position restraints with Parrinello-Rahman, but I could imagine valid uses. A note is another option.
Another issue is that we get a software inconsistency error instead of a nice error message. But that is often difficult to avoid with an exploding system.

#4 Updated by Gerrit Code Review Bot over 2 years ago

Gerrit received a related patchset '1' for Issue #2330.
Uploader: Berk Hess ()
Change-Id: gromacs~release-2018~I299e39247d69214d9b54c47874d522ac7b00a30e
Gerrit URL:

Also available in: Atom PDF