Dear Jean-Paul,
I've compiled the new DEALII version with your changes, but I am unsure of
how to use the changes correctly.
If I understand correctly, you can now set Teuchos parameters after you
have initialized the TrilinosWrappers::PreconditionAMG.
This is achieved by creating a Teuchos::ParameterList object and using it
in the set_parameters public function of the
Preconditioner::AdditionalParameters.
So for instance in my case it would be something like:
// Still need to figure out the syntax for the additional_parameters
Teuchos::ParameterList additional_parameter("smoother: params");
And then I would set the parameters in the preconditioner additional
parameters
I have two things i am unsure:
1) For the set the additional parameters I need a
std::unique_ptr<Epetra_MultiVector> &distributed_constant_modes . I presume
this vector is created in the initialization from the std::vector of
constant boolean modes? Is there a way to grab it again and reuse it? I do
not understand why I need to specify the constant modes for every change of
parameters.
2) Is there any documentation on the various parameter list. What I would
like to change is the atol of the ILU smoother and ILU preconditions. I
think this will be in the smoother: params, but I am unsure how to proceed
from there. I have found the MULUE documentation, but it seems to be more
tailored around their XML interface?
Thank you so much,
Bruno
On Tuesday, 12 March 2019 05:06:03 UTC-4, Jean-Paul Pelteret wrote:
>
> Dear Bruno,
>
> I believe that my proposed changes are working now, so you could always
> pull that branch from my repository and use it immediately. That might be
> the most convenient way for you to move forward.
>
> Thanks for being willing to share your findings wrt. AMG parameters. I
> think that would be really useful. I also think that encapsulating a
> parameter study for NS + AMG/GMG within a tutorial could be interesting and
> valuable. Maybe you could open up an issue on the GitHub repository and we
> could all discuss the specifics there.
>
> Best,
> Jean-Paul
>
> On 11 Mar 2019, at 14:25, Bruno Blais <[email protected] <javascript:>>
> wrote:
>
> Dear Jean-Paul and Wolfgang,
> Is it better if I try to go with the Teuchos way or should I wait until
> Jean-Paul finishes the merge on his branch ? I think the latter might be a
> good option.
> In all cases, I would be more than glad to share my experience in
> parameters tuning for the AMG solver with ILU type of smoother and coarse
> solve.
>
> On a slightly related note. Hopefully I will have finished the basis of my
> GLS stabilized navier-stokes solver by June or so. It already works pretty
> well in parallel with GMRES + ILU, but I want to test more what can be done
> in terms of multigrid and AMG.
> In a way, it is very reminescent of a similar work done in MOOSE, but
> using Trilinos, P4est and deal.II. The MOOSE paper is here:
> https://www.sciencedirect.com/science/article/pii/S0965997817310591
>
> I am more than willing to share everything about the solver. Do you
> believe this could be something that could make an interesting Tutorial? I
> could wrap-it around the study of the flow around a cylinder or a backward
> facing step to have some validation. Verification would be done using MMS.
> Best
> Bruno
>
> --
> The deal.II project is located at http://www.dealii.org/
> For mailing list/forum options, see
> https://groups.google.com/d/forum/dealii?hl=en
> ---
> You received this message because you are subscribed to the Google Groups
> "deal.II User Group" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to [email protected] <javascript:>.
> For more options, visit https://groups.google.com/d/optout.
>
>
>
--
The deal.II project is located at http://www.dealii.org/
For mailing list/forum options, see
https://groups.google.com/d/forum/dealii?hl=en
---
You received this message because you are subscribed to the Google Groups
"deal.II User Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
