[gmx-users] Mdrun box nan error

[Erik Lindahl]

Hi Natalie,

> On Fri, 2004-05-07 at 20:01, Nathalie R. Basdevant wrote:
>> David,
>> So I have tried your suggestion to launch the same tpr twice. The 
>> error (box
>> nan) is NOT reproducible. That's the only error I get, the 'atoms' 
>> thing
>> referred to the accompanying LINCS error.
>> So that leaves compiler and hardware then. I'll investigate...

Just to get this clear - are you launching *exactly* the same command 
twice, with *exactly* the same input files, but you get different 
results?

In that case it cannot be a compiler issue, since it should be running 
the same code. It is either a bug in the operating system, or in the 
hardware. However, a lower level of compiler optimization might avoid 
triggering the bug...

I have contacts both HP and Intel, so I'm sure we can get some help to 
debug it. What we would need from you is:

1. A simple (small) system where it happens, and preferrably after a 
small number of steps. You can use gmxdump to compare trajectories and 
see if they differ.

2. Every single piece of information you have about your operating 
system, version, hardware, memory, compiler (+version), etc.



Cheers,

Erik