[Getdp] Mag. Vec Potential

[Kubicek Bernhard]

Dear all,

as promised I send a small system which I try to solve with a tree
gauged magnetic vector potential. 
The system was created, because my final-goal calculation does not work
at all:

We try to simulate arc discharges, in a geometry simmilar to
coolmesh.png. The top surface is a symmetry boundary, the red and blue
things are conducting copper electrodes, green and grey-blue is some
air. One of the (visible) small side faces of the electrodes we insert a
source current density, the (visible) small side face of the opposed
electrode is on potential zero. In the "air", we have a
temperature-field imported into GetDP up to 30000 Kelvin. The plasma is
modelled via a temperature dependent conductivity. All this is coupled
with a commercial CFD code, to simulate the arcs behaviour transiently,
e.g. the pinch forces, the resistive heating, fluid pressure waves,
movement...

All this workes nicely for our "playground" geometries e.g.
http://www.geuz.org/getdp/wiki/Gallery . However, in the attached mesh
we are more or less forced to use a hexahedral mesh. The electrodes are
so thin that they would force an enourmous ammout of tetrahedras for
some useful meshing. With the shown structured hexahedral grid, we hope
to use longish "bricks" instead of "cubeish" cells. Also I think it is
an advantage that and the violet region, the mesh bends around the
electrodes in a similar way as the magnetic field. 
However, we experience very bad behaviour even for the calculation of
the current density. However, the magnetic field is completely invalid.
Which I don't understand, as my very beautiful mesh is indeed very
beautiful. The big difference can be the hexahedral mesh, oposed to the
tetrahedral mesh we use usually.

That is why I created the attached system, to trace down the problem's
source. 
// The gambit mesh are excluded because of their size, you can download
it here: 
// http://kariert.org/files/gambit.gmsh

Basically, it is a infinitely long square wire with a constant electric
current "enclosed" in a box. The geometry is meshed either tetrahedral
or hexahedral using gmsh. To get some hexahedral grid using the
transfinite algorithm, the 8 outer volumes were chosen. Alternatively, I
have one more tetrahedral mesh created using converters from Gambit,
Fluent's commercial mesher, to pinpoint mesh problems (mesh quality is a
bit better than with gmsh2, reflected in less strange spots around "bad"
cells, see tetra* images). However, due to the conversion, the region
numbers are different, and therefore there are two blocks in the .pro's
region definition of which one needs to uncommented.

Attached are a couple of pngs showing the calculation results. The
tetrahedral ones compare to some extend with the analytical solution
(which should have a max norm of approx 0.27). Apart from a a tube where
there are very high fields, at the border between the volumes. Some
similar tubes on the outer sides can be seen, if the gauge condition is
not applied in the d* faces. It looks like if somehow a current is
searching a way to flow backwards. Knowing that the static h-field can
only exist if there are no current-sources in the calculated regime (
rot h =j -> div j==div rot h==0), we still think that things should be
ok if the are current sources on boundary surfaces? 
We find this highly strange. 

Furthermore, the hexahedral calculation is not even close in their
magnitude. Our idea that maybe on the boundary faces that cut the wire,
one could "help" the a-potential by declaring a source-h-field also
failed quite impressively.

If someone has some ideas on what is going on and how we could improve
the calculations, that would be _extremely_ nice. Personally, I am under
a lot of pressure to make this work asap, but I am out of ideas and
knowledge.

Very nice greetings,
and thank you in advance, also for reading this long mail,  

 Bernhard

PS:
Once more, my mail was to big for the list (400kb).
All attached files can be found here:
http://kariert.org/files/afield.zip