On Fri, Feb 1, 2013 at 6:52 PM, Brian Radak <radak004.umn.edu> wrote:
> Hi all,
>
> I am trying to run QM/MM umbrella sampling using HREMD. However, all of the
> simulations crash and the first replica in the groupfile gives the
> following error:
>
> Creation of partners array failed! Bad REMD setup!
>
Since this is an error that's only possible in the developers version I'm
moving it to the developer list instead.
> Here's an example line from the groupfile:
>
> -O -i r0/QMMM_1.inp -o r0/QMMM_1.out -p
> ../initial/ets_C41pC75p_Mg_QMMM_initial.prmtop -c r0/QMMM_0.rst7 -r
> r0/QMMM_1.rst7 -x r0/QMMM_1.nc -rem 3 -remlog QMMM_1.remlog
>
> Poking around in the source code it looks like the error message is
> triggered when checking that all replicas have 2 defined replicas. Under
> what circumstances would/should this check fail? It seems like a pretty
> foolproof process based off of reading the groupfile.
>
So what's really happening here is that the replica index is grabbed for
the current dimension (this is how I know it's Amber developer only), the
neighbors (or partners) of the array are determined by the replica one
higher and one lower (as defined in the groupfile for H-REMD), and then all
indices are gathered into a table so each replica can pick out the ranks
that its partners are in the REMD communicator.
The only way I can really see this failing in one dimension is if the index
array (either index_list or replica_indexes) is corrupted at some point --
I've seen weird behavior with specific compiler/MPI combinations in rare
instances (OpenMPI with ifort comes to mind). Can you modify the code to
print out these tables on each thread? Printing them to a different file
(something like write(666+remd_rank, *) blah) will help narrow down what
data isn't correct on which replica.
Also, do the test cases pass?
Worst case scenario we can debug it next week.
--Jason
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Fri Feb 01 2013 - 19:00:04 PST
