[gmx-users] Crashing abnormally

[Anton Feenstra]

Amadeu wrote:
> **Fatal error: ci = 1777 should be in 0 .. 1727 [FILE nsgrid.c, LINE 218]
[...]

> I would appreciate if anyone could comment on this error and perhaps
> provide some input/feedback what may be happening. Is it possible
> there is something to do with hardware/memory problem?

The possiblility of hw/mem problems you can eliminate by re-running the
exact same run (same .tpr, same mdrun binary, same CPU's, same nodes).
If it fails again, it wasn't hw/mem, since these errors are essentially
random.

It seems likely, you encountered some 'spurious' crashes, where actually
nothing went wrong but by chance some atoms were accelerated too much for
the integration timestep to handle correctly (but in principle still
physically sensible). This explains why you didn't see the same error
in a single precision run, since different round-off errors will cause
the runs to diverge rapidly (within 10 ps or so, usually).

My attitude to this type of errors is very pragmatic. I try a re-start
with slightly rounded (in the sixth decimal) coordinates. If that
succeeds, I assume the error was due to breaking the timestep-imposed
speed-limit. If the error persists, there is something wrong in my
simulated system (e.g., strange conformations, close contacts, or
even wrong charges, wrong topology or other problems with ff parameters).

I've posted the procedure I use several times before, but once more won't
hurt. Here it comes:
1) dump the last .trr (coordinate + velocities) frame into a .gro file
with higher precision, using:
# trjconv -f traj.trr -s topol.tpr -o frame.gro -ndec 6 -dump 1234
(where '1234' stands for the time of your last complete .trr frame, that
you can find with 'gmxcheck -f traj.trr')
2) convert the dumped frame.gro back to .trr format, using:
# trjconv -f frame.gro -o frame.trr
3) generate new.tpr file with restart coordinates & velocities with tpbconv:
# tpbconv -s topol.tpr -o new.tpr -f frame.trr
(possibly, you'll want to adjust the remaining # steps with one of tpbconf
options.)
4) Re-start your run with the new.tpr file.


-- 
Groetjes,

Anton
  _____________ _______________________________________________________
|             |                                                       |
|  _   _  ___,| K. Anton Feenstra                                     |
| / \ / \'| | | Dept. of Pharmacochem. - Vrije Universiteit Amsterdam |
|(   |   )| | | De Boelelaan 1083 - 1081 HV Amsterdam - Netherlands   |
| \_/ \_/ | | | Tel: +31 20 44 47608 - Fax: +31 20 44 47610           |
|             | Feenstra at few.vu.nl - www.few.vu.nl/~feenstra/         |
|             | "If You See Me Getting High, Knock Me Down" (RHCP)    |
|_____________|_______________________________________________________|