[gmx-developers] PME-Bug with triclinic boxes

[Oliver Lange]

Hi developers,

I have a problem with the PME-Bug with triclinic boxes.
I have read that the CVS-Version has a fix for that bug, but there doesn't seem to be
a stable release with this bugfix in it.
Could you distribute a bugfix patch against version 3.1.4
or is such a patch available already?

Regards
Oliver Lange

David wrote:
> On Sat, 2003-11-15 at 11:02, Martina Bertsch, PhD wrote:
> 
>>Dear GROMACS Developers,
>>
>>When running a CG-EM on a system composed of a protein embedded in a
>>lipid bilayer, with no constraints and FLEX_SPC (see attached *em.mdp), I 
>>get the following error message:
>>
>>Fatal error: ci = -2147483648 should be in 0 .. 7999 [FILE nsgrid.c, LINE 210]
>>
>>I figured that the ci is -(231) and the range of acceptable values is
>>from 0 to 7999, for a total of 20*20*20 = 8000 cells in my grid, as shown 
>>in the log file. I thought there could be calculations associated with the 
>>grid that resulted in numbers greater than the maximum cardinality of the 
>>machine word size (232), causing some kind of numerical overflow.
>>
>>A number of users have reported this problem in various contexts, and the only 
>>solution that I have found is the fix for the minimize.c program, suggested by 
>>Dr. Graham Smith (see the May 28, 2002 message to gmx-developers). Dr. Smith 
>>actually   said
>> he used to get the same error message when running CG-EM.
>>I implemented the fix and successfully recompiled.
>>
>>However, when I try running CG-EM with the same data set, I still get the same message:
>>
>>Fatal error: ci = -2147483648 should be in 0 .. 7999 [FILE nsgrid.c, LINE 210]
>>
> 
> The error comes from coordinates being NaN (not a number). The physical
> reason for this is that particles come too close to each other. Your
> best bet is probably to check your starting structure once more: see if
> you have lipids close to your protein. If you do steepest descents EM it
> will print which atom has the largest force on it, so tht might be a
> good starting point.
> 
> 
> Furthermore you are using a cut-off with longer Vanderwaals than
> electrostatic. Is that really what you intend to do?
> 

-- 
Oliver Lange
MPI for biophysical Chemistry
olange at gwdg.de
Fax: +495512011089
Tel: +495512011306