From: Gustavo Seabra <>
Date: Thu, 18 Jan 2018 10:22:47 -0300

Unfortunately, this is the expected behavior for the DFTB code in sander.
The code runs all in processor, and is not parallel. This has to do with
the diagonalization routines, that are not parallel.

If you are used to DFTB+, maybe you could try the new interface that uses
it as an external QM code.

Gustavo Seabra
Em 18 de jan de 2018 07:00, "daniele selli" <>
> Dear Amber Users,
> I am trying to use the QMMM implementation in AMBER with the DFTB
> semiempirical method for the QM part.
> When I run simple tests, like water dimer (one molecule DFTB and the other
> one tip3p) optimization or a short MD simulation of a 15x15x15 ang water
> box (3 molecules QM and 110 MM tip3p) I have no problem.
> The code behaves well and I got reasonable results.
> The problems come when I try to work with bigger systems!
> I can still menage 5865 water molecules box, 113 of which are QM, but I
> noticed that the code (sander.MPI) doesn't scale passing from 1 node (28
> cores) to 2 nodes (56 cores) connect with infiniband of course.
> Is there a reason for that? There is some special flag to enable during the
> compilation or keyword to insert in the input of the qmmm calculation?
> Then, if I have 6539 water molecules in total, 959 of which are QM the
> calculation is essentially going nowhere. It is stacked and I cannot obtain
> a single SCC cycle waiting a lot of time. This means that with the present
> implementation of AMBER+SCC_DFTB one cannot handle systems with a QM region
> so big or, again, I am missing something?
> I have been using DFTB since a lot of time and ~1000 atoms is a normally
> manageable size (using the DFTB+ code).
> Thank you very much for you willingness and hopefully help,
> Best Regards
> Daniele Selli
> _______________________________________________
> AMBER mailing list
AMBER mailing list
Received on Thu Jan 18 2018 - 05:30:03 PST
Custom Search