[gmx-developers] nrexcl definition

[Paul Whitford]

Thanks.  I'll get the latest version and try again.

I have a related question:  When I got the other message, I also
receive this before gromacs shuts down.

-------------------------------------------------------
Program mdrun_mpi, VERSION 3.3.99_development_20071104
Source code file: domdec_top.c, line: 174

Fatal error:
10680 of the 44415 bonded interactions could not be calculated because
some atoms involved moved further apart than the multi-body cut-off
distance (0 nm) (option -rdd) or the non-bonded cut-off distance (1.5
nm)
-------------------------------------------------------


It says the interaction "could not be calculated because..."  they are
too far apart.  If I use the flag   "-rdd 1.5" and suppress the
messages and the signal to kill, then at each step will it not
calculate the bonded terms only at steps where the distance exceeds
1.5(which is fine for my purposes)?  Or, once the bonded interaction
is beyond 1.5 it will it not calculate the interaction ever again?

I would surpress the messages by removing

    if ((int)(nb + 0.5) != cr->dd->nbonded_global)
      dd_print_missing_interactions(fplog,cr,(int)(nb + 0.5));

from stat.c.  Perhaps there is a flag that will supress the messages,
rather than editing the source.

thanks

-Paul


> In the cvs version, did the definition of nrexcl change? I am using a
> homemade potential and I have always used nrexcl=3 (with gromacs 3.3.2).
> If
> I use this same definition with the cvs version, I get errors saying
>
> A list of missing interactions:
> n831(0): LJ-14 of 15302 missing 10680
> n831(0):
> The first 10 missing interactions, except for exclusions:
> n831(0): LJ-14 atomsn831(0): 2059n831(0): 234n831(0):
>
>
> When I change nrexcl to 4, the messages go away(I want there to be no
> LJ-14
> interactions, which this seems to do). I think the definition of nrexcl
> changed somewhere, but I haven't seen documentation of it.
> Was there a change, or does it look like I'm doing something incorrectly?
> thanks
>
> -Paul

No, nothing has changed.
But there were some bugs in the domain decomposition which could
lead to problems. I fixed these on January 29th. Somebody else
also reported incorrect interaction prints. I suspect that these
came from memory corruption, but I am not completely sure.
If you were not running the latest CVS, please try.
If you were, could you send me a tpr file that causes problems?

But setting nrexcl to 3 (or 4) does NOT remove LJ-14.
LJ-14 are listed separately in the topology in the [ pairs ] section.

Berk.