Re: [DuMux] 1D grid and MPI

[Dmitry Pavlov]

Dear Timo,

Thank you. I will go ahead with the Newton.EnableShiftCriterion = 1 for now.

Best regards,

Dmitry


On 6/30/22 13:51, Timo Koch wrote:

Dear Dmitry,

this is a known current bug 
(https://git.iws.uni-stuttgart.de/dumux-repositories/dumux/-/issues/864)
There are some ideas for a fix but due to limited developer resources 
we are currently focussing on a better linear solver interface that 
includes a fix for this, instead of a temporary fix.
A draft is here: 
https://git.iws.uni-stuttgart.de/dumux-repositories/dumux/-/merge_requests/2113


What helps:
* You can either use a criterion that does not use the residual 
(relative shift criterion) in the Newton.
* Or you have to compute the residual norm by using the correct 
consistent parallel ScalarProduct.
For how to set up the ScalarProduct have a look at MR2113 
(https://git.iws.uni-stuttgart.de/dumux-repositories/dumux/-/merge_requests/2113) and/or 
the parallel solvers (e.g. 
https://git.iws.uni-stuttgart.de/dumux-repositories/dumux/-/blob/master/dumux/linear/istlsolverfactorybackend.hh)


Best regards,
Timo



On 29 Jun 2022, at 15:57, Dmitry Pavlov  
wrote:


Hello,

I have a 1D radial mode porous flow simulation implemented via

Dune::YaspGrid<1, Dune::EquidistantOffsetCoordinates >

and

    struct GGTraits : public 
BoxDefaultGridGeometryTraits

    { using Extrusion = RotationalExtrusion<0>; };

I am using the Box method. My program works well when ran as single. 
With MPI, however, I get the following message from fvassembler.hh:


Warning: norm calculation adds entries corresponding to
overlapping entities multiple times. Please use the norm
function provided by a linear solver instead.

and the solver does not converge because the residual does not change.

I am seeking advice.


Best regards,

Dmitry


___
DuMux mailing list
DuMux@listserv.uni-stuttgart.de
https://listserv.uni-stuttgart.de/mailman/listinfo/dumux
___
DuMux mailing list
DuMux@listserv.uni-stuttgart.de
https://listserv.uni-stuttgart.de/mailman/listinfo/dumux

Re: [DuMux] 1D grid and MPI

[Timo Koch]

Dear Dmitry,

this is a known current bug 
(https://git.iws.uni-stuttgart.de/dumux-repositories/dumux/-/issues/864 
)
There are some ideas for a fix but due to limited developer resources we are 
currently focussing on a better linear solver interface that includes a fix for 
this, instead of a temporary fix.
A draft is here: 
https://git.iws.uni-stuttgart.de/dumux-repositories/dumux/-/merge_requests/2113 


What helps:
* You can either use a criterion that does not use the residual (relative shift 
criterion) in the Newton.
* Or you have to compute the residual norm by using the correct consistent 
parallel ScalarProduct.
For how to set up the ScalarProduct have a look at MR2113 
(https://git.iws.uni-stuttgart.de/dumux-repositories/dumux/-/merge_requests/2113
 
)
 and/or the parallel solvers (e.g. 
https://git.iws.uni-stuttgart.de/dumux-repositories/dumux/-/blob/master/dumux/linear/istlsolverfactorybackend.hh
 
)

Best regards,
Timo



> On 29 Jun 2022, at 15:57, Dmitry Pavlov  wrote:
> 
> Hello,
> 
> I have a 1D radial mode porous flow simulation implemented via
> 
> Dune::YaspGrid<1, Dune::EquidistantOffsetCoordinates >
> 
> and
> 
> struct GGTraits : public BoxDefaultGridGeometryTraits
> { using Extrusion = RotationalExtrusion<0>; };
> 
> I am using the Box method. My program works well when ran as single. With 
> MPI, however, I get the following message from fvassembler.hh:
> 
> Warning: norm calculation adds entries corresponding to
> overlapping entities multiple times. Please use the norm
> function provided by a linear solver instead.
> 
> and the solver does not converge because the residual does not change.
> 
> I am seeking advice.
> 
> 
> Best regards,
> 
> Dmitry
> 
> 
> ___
> DuMux mailing list
> DuMux@listserv.uni-stuttgart.de
> https://listserv.uni-stuttgart.de/mailman/listinfo/dumux

___
DuMux mailing list
DuMux@listserv.uni-stuttgart.de
https://listserv.uni-stuttgart.de/mailman/listinfo/dumux

[DuMux] 1D grid and MPI

[Dmitry Pavlov]

Hello,

I have a 1D radial mode porous flow simulation implemented via

Dune::YaspGrid<1, Dune::EquidistantOffsetCoordinates >

and

    struct GGTraits : public BoxDefaultGridGeometryTraits
    { using Extrusion = RotationalExtrusion<0>; };

I am using the Box method. My program works well when ran as single. 
With MPI, however, I get the following message from fvassembler.hh:


Warning: norm calculation adds entries corresponding to
overlapping entities multiple times. Please use the norm
function provided by a linear solver instead.

and the solver does not converge because the residual does not change.

I am seeking advice.


Best regards,

Dmitry


___
DuMux mailing list
DuMux@listserv.uni-stuttgart.de
https://listserv.uni-stuttgart.de/mailman/listinfo/dumux