[gmx-users] That darned nsgrid.c error

[Berk Hess]

>Fatal error: ci = -1 should be in 0 .. -1 [FILE nsgrid.c, LINE 210]
>Error on node 0, will try to stop all the nodes
>[0] MPI Abort by user Aborting program !
>[0] Aborting program!
>Fatal error: ci = -1 should be in 0 .. -1 [FILE nsgrid.c, LINE 210]
>Error on node 1, will try to stop all the nodes
>[1] MPI Abort by user Aborting program !
>[1] Aborting program!
>
>I've had this type of error happen for all sorts of reasons ... from
>needing to recompile the fftw libraries to having a poorly relaxed system.
>The test system ran before the upgrade very smoothly.  It also runs
>successfully on a single processor and on other machines (an IBM SP on
>multiple processors for example)... it is well relaxed. Can someone list
>the reasons that this particular error could occur for me?  I would really
>appreciate your help.

This error and that other darned error: "Only triclinic boxes with the first 
vector parallel to ..."
occur often when the system becomes unstable. No further checks are made 
during dynamics,
except for the large pressure sacling check. This means that when there are 
very large or NAN forces,
for instance because of a bad starting structure or a too large timestep, 
usually only one of these
two errors appears.

In this case there is another hint, namely: "should be in 0 .. -1".
 From this one can see that there are zero grid cells. However, there is a 
check before this
in the code to check that the box vectors are longer than twice the cut-off 
length.
So if the code is compiled and linked correctly there can never be zero grid 
cells.
Because of this I would guess that there is some library incompatibility 
problem.
Such as different libraries compiled with different MPI versions.

Berk.

_________________________________________________________________
Tired of spam? Get advanced junk mail protection with MSN 8. 
http://join.msn.com/?page=features/junkmail