[gmx-users] Charge group moving too much

[Mark Abraham]

Jennifer Williams wrote:
> Hello,
> 
> One of my simulations just crashed after 660,000 steps with the 
> following message:
> 
> Step 660560:
> The charge group starting at atom 4100 moved than the distance allowed 
> by the domain decomposition (1.500000) in direction X
> distance out of cell 2.356293
> Old coordinates:    1.752    0.447    0.751
> New coordinates:    3.593    1.532    0.674
> Old cell boundaries in direction X:    0.000    2.116
> New cell boundaries in direction X:    0.000    2.122
> 
> -------------------------------------------------------
> Program mdrun, VERSION 4.0.5
> Source code file: domdec.c, line: 3651
> 
> Fatal error:
> A charge group moved too far between two domain decomposition steps
> This usually means that your system is not well equilibrated
> -------------------------------------------------------
> 
> "Why Weren't You at My Funeral ?" (G. Groenhof)
> 
> Error on node 0, will try to stop all the nodes
> Halting parallel program mdrun on CPU 0 out of 8
> 
> I am modelling flexible alkyl chains which are anchored to a frozen 
> silica wall. This particular atom (4100) is at the edge of the periodic 
> boundary condition so I am guessing that what happens is that while the 
> anchored side stays where it is, the other free end wanders into the 
> next cell and this perhaps triggers the error message? If this is indeed 
> the problem, is there some way to stop the program stopping with an 
> error message when this happens?

The old and new coordinates reported aren't consistent with that, and 
the error message isn't "this bond is too long", it's "this group of 
atoms moved way too much last integration".

> I did do an energy minimisation before starting and my structure looks 
> OK. It seems strange that it would crash only after 660000 steps.

Indeed.

> Also what is meant by the following? I am using NVT so my simulation 
> cell should not change.
> 
> Old cell boundaries in direction X:    0.000    2.116
> New cell boundaries in direction X:    0.000    2.122
> 
> 
> My cell is about 4.3 x 4.3 x 7.5 nm
> I am using gromacs-4.0.5

I'm mystified, but I know little about interactions with frozen groups. 
I don't think Vitaly's suggestion will help in the context of a 
simulation that runs for hundreds of thousands of steps.

Whether or not you can reproduce the crash will be useful data. If so, 
construct a run that will stop just before it, and then continue on to 
the crash with nstxout = 1 so you can see whether these atoms are really 
moving lots, or perhaps a bug was provoked.

Mark