Hi Georg,
I also would say that the El2PAmgTraits come closest to what you need
since the el2p model is also based on PDELab. The traits basically
distinguish between parallel and sequential settings. As soon as you
take a parallel grid manager, the parallel setting is employed, even for
sequential runs. The standard AmgTraits from dumux/linear additionally
distinguish between box and cell-centered. The traits will just set the
correct types. If things compile, you should be fine from that perspective.
What could help regarding the convergence would be to tweak the
parameters for the AMG. They are currently hardwired in
dumux/linear/amgbackend.hh line 123ff. Some explanations are given at
http://www.dune-project.org/doc/doxygen/dune-istl-html/class_dune_1_1_amg_1_1_parameters.html
The basic problem is that the AMG doesn't know anything about the fact
that your problem/matrix is a coupled one that consists of several
meaningful subproblems/submatrices. A solver that acknowledges this will
most likely be superior. We have started working on this by means of a
Master thesis. As soon as we know more, I can let you know.
Kind regards
Bernd
On 02/29/2016 10:00 AM, [email protected] wrote:
Hi Bernd,
I tried running my multidomain model with the AMG-Solver (not yet in parallel).
First error that popped up was about AmgTraits not being set. As a first try I
used the AmgTraits defined in the el2ppropertydefaults. However, it does not
converge. For larger time steps I get warnings from UMFPACK that the matrix is
singular. For smaller timesteps the solver does not converge. Maybe I could get
the solver to converge if I change some of these traits? I am not sure what
these traits should be for a multidomain application (MType, VType, Comm,
LinearOperator, ScalarProduct and Smoother). Is there some place I could find
information on what to use here?
Kind regards
Georg
-----Ursprüngliche Nachricht-----
Von: Dumux [mailto:[email protected]] Im Auftrag von
Bernd Flemisch
Gesendet: Mittwoch, 24. Februar 2016 15:32
An: DuMuX User Forum
Betreff: Re: [DuMuX] Multidomain in parallel
Hi Georg,
I think that there is no test case for the multidomain models. You can have a
look at test_box1pwithamg and the corresponding problem file 1ptestproblem.hh
in test/implicit/1p (or test/porousmediumflow/1p/implicit in the current
master) to see how to enable another solver, this time the AMGBackend:
#include <dumux/linear/amgbackend.hh>
SET_TYPE_PROP(OnePTestBoxProblemWithAMG, LinearSolver,
Dumux::AMGBackend<TypeTag> );
I am unsure whether this works in combination with a multidomain setting,
please try.
Kind regards
Bernd
On 02/23/2016 03:51 PM, [email protected] wrote:
Hi Christoph,
We ran our multidomain application in parallel using SuperLU as solver.
However, this lead to an increase in wall time. Since, for now, we do not solve
Stokes flow it might be possible to use an iterative solver for our fuel cell
models and speed up the simulation through parallelization. Is there a test
case using an iterative solver?
Kind regards
Georg
-----Ursprüngliche Nachricht-----
Von: Dumux [mailto:[email protected]] Im Auftrag
von Christoph Grüninger
Gesendet: Dienstag, 16. Februar 2016 16:40
An: [email protected]
Betreff: Re: [DuMuX] Multidomain in parallel
Hi Georg,
sure, you can use MultiDomain / MultiDomaingrid in parallel. The best
documentation about parallel computing should be the grid howto. You
can download it from the website, see
http://www.dune-project.org/doc/grid-howto/grid-howto.pdf
or build it yourself from the dune-grid-howto module.
In DuMuX there should be tests that run in parallel, too. But not coupled ones.
In DuMuX we introduced the MultiDomain stuff for coupling Stokes with Darcy.
Stokes leads to a saddle-point problem that could not be efficiently solved by
an iterative solver. All direct solvers are not parallel. That's why we never
tried to run the whole problem in parallel. If your assembly is slow, it still
might be worth a try.
Bye
Christoph
--
The method has been developed for use on a high-speed electronic computer and
would be impractical for hand-solution purposes.
[Harlow & Welch 1965]
_______________________________________________
Dumux mailing list
[email protected]
https://listserv.uni-stuttgart.de/mailman/listinfo/dumux
--
_______________________________________________________________
Bernd Flemisch phone: +49 711 685 69162
IWS, Universität Stuttgart fax: +49 711 685 60430
Pfaffenwaldring 61 email: [email protected]
D-70569 Stuttgart url: www.hydrosys.uni-stuttgart.de
_______________________________________________________________
_______________________________________________
Dumux mailing list
[email protected]
https://listserv.uni-stuttgart.de/mailman/listinfo/dumux
_______________________________________________
Dumux mailing list
[email protected]
https://listserv.uni-stuttgart.de/mailman/listinfo/dumux
--
_______________________________________________________________
Bernd Flemisch phone: +49 711 685 69162
IWS, Universität Stuttgart fax: +49 711 685 60430
Pfaffenwaldring 61 email: [email protected]
D-70569 Stuttgart url: www.hydrosys.uni-stuttgart.de
_______________________________________________________________
_______________________________________________
Dumux mailing list
[email protected]
https://listserv.uni-stuttgart.de/mailman/listinfo/dumux
