[Getdp] Simple electrostatic simulation yields nan values everywhere

Théo Mathurin theo.mathurin at univ-artois.fr
Wed Sep 18 10:48:03 CEST 2019

```Well, thank you for the explanation. It helps make sense of these weird
outcomes.

It's also good to know that the results I got in my tests with round
conductors were not correct.

Le mer. 18 sept. 2019 à 10:02, Christophe Geuzaine <cgeuzaine at uliege.be> a
écrit :

>
>
> > On 18 Sep 2019, at 09:51, Théo Mathurin <theo.mathurin at univ-artois.fr>
> wrote:
> >
> > Hello Christophe,
> >
> >
> > That indeed solved the problem. Still, given this blunder I don't
> understand why the computation worked with round conductors.
>
> The Jacobian transformation you use ("Vol") assumes the dimension of your
> element (here a line, i.e. 1D) is the same as the dimension of the problem
> (which is 2D here, not 1D). In effect it will thus keep the "x-coordinate"
> only: for a square conductor this will lead to null Jacobians for the
> vertical sides; on round conductors if no element is strictly aligned with
> the y-axis it will give something nonzero. In all cases the result is wrong
> and you should use "Sur" for integrals on line elements in 2D.
>
> Christophe
>
>
> >
> > Best regards,
> >
> > Théo Mathurin
> >
> > Research engineer
> > Laboratoire Systèmes Electrotechniques et Environnement (LSEE) -
> Université d'Artois
> >
> > Le mar. 17 sept. 2019 à 21:44, Christophe Geuzaine <cgeuzaine at uliege.be>
> a écrit :
> >
> > Hi Théo,
> >
> > You are integrating your volume stiffness matrix on DomainCC_Ele
> >
> >   Galerkin { [ eps[] * Dof{d v} , {d v} ] ; In DomainCC_Ele ; Jacobian
> >
> > but you define DomainCC_Ele to also include the (inner) boundary
> Skin_ins_conds
> >
> >     DomainCC_Ele += Region[{Skin_ins_conds}] ;
> >
> > Remove this and you should be fine.
> >
> > Christophe
> >
> >
> > > On 17 Sep 2019, at 17:34, Théo Mathurin <theo.mathurin at univ-artois.fr>
> wrote:
> > >
> > > Hello everyone,
> > >
> > > I've encountered a strange issue with my electrostatic computation.
> I'm using the V formulation with the electric potential v saved as a
> post-operation output (see pro file attached).
> > >
> > > The geometry is desperately simple: two conductors in a square
> surrounding medium, each with its insulation. When the conductors are
> round, everything's fine and the electric potential is computed correctly.
> But when their shape is changed to rectangular, for some reason I get nan
> values everywhere in the domain.
> > >
> > > My msh file is generated using the Gmsh Python API. Attached is the
> corresponding Python script, where one can switch the shape from circles to
> rectangles. For debugging purposes, the script is written so that the
> minimal amount of code that is not shared by the two cases is isolated
> behind a conditional statement.
> > >
> > > Apparently, the problem arises when and only when one or several of
> the left or right insulation outer curves are straight lines. I have
> checked that there is nothing wrong with physical regions.
> > >
> > > In both cases I'm using the same pro file, and the command to run
> GetDP is the following:
> > >
> > > getdp -pro electrostatics.pro -msh test_geom.msh -solve EleSta_vf
> -pos get_v
> > >
> > > Note that I was able to carry out the very same computation
> successfully in a different, much more complicated geometry with
> rectangular conductors. I sincerely have no idea as to the root of this
> seemingly intractable problem, and therefore any help would be greatly
> appreciated. If needed, I can also provide the msh files I get when I'm
> running my Python script.
> > >
> > > Other potentially relevant information below
> > > - Python API from Gmsh 4.4.1 Linux 64 bits SDK
> > > - GetDP 3.2.0 Linux 64 bits
> > > - Python 3.7
> > > - Ubuntu 18.04.3
> > >
> > > Best regards,
> > >
> > > Théo Mathurin
> > >
> > > Research engineer
> > > Laboratoire Systèmes Electrotechniques et Environnement (LSEE) -
> Université d'Artois
> > >
> > > <test_geom.py><electrostatics.pro
> >_______________________________________________
> > > getdp mailing list
> > > getdp at onelab.info
> > > http://onelab.info/mailman/listinfo/getdp
> >
> > —
> > Prof. Christophe Geuzaine
> > University of Liege, Electrical Engineering and Computer Science
> > http://www.montefiore.ulg.ac.be/~geuzaine
> >
> >
> >
>
> —
> Prof. Christophe Geuzaine
> University of Liege, Electrical Engineering and Computer Science
> http://www.montefiore.ulg.ac.be/~geuzaine
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://onelab.info/pipermail/getdp/attachments/20190918/e4a77bbd/attachment-0001.html>
```