help with GetDP needed

[Samuel Kvasnica]

This is a multi-part message in MIME format.
--------------A17AD55816F90C3760C83C7A
Content-Type: text/plain; charset=iso-8859-2
Content-Transfer-Encoding: 7bit

Christophe Geuzaine wrote:

> Maybe I made a mistake in my mail ? Anyway, you should give in the
> PostProcessing the same jacobian method as in the Formulation. The name
> refers to the name of the jacobian method : for example, it should be
> 'test' in this case (not 'Vol')

No no, sorry, this was o.k. I just didn't notice that that's the name of jacobian method,
not name of jacobian.

> Well, the radius computed by getdp is R=0.45, which is equal to Rext
> (but R<Rext should hold, not R<=Rext). Just add a small quantity to the
> Rext value, to assure the inequality.

o.k., with Rext=450.0000001e-3 problem dissapeared,but similar problem at R=0.35 appeared.
So I've set also Rint to 349.99999999e-3 and it works now !

> > and many 'Unknown variable: nan' parser errors after trying to open generated file with gmsh.
> > Any idea ?
>
> No, this is strange.

this problem still remains. It appeared after I started to use VolAxi. .pos file for Az really
contains
some NaN's. B and H makes no problems, just Az.  Could these be some discontinuities on the axis ?
Maybe axis should be excluded from postprocessing someway ?

> No, I didn't implement higher order interpolation stuff yet. But why do
> you need such a fine mesh ? This magnetostatic problem is very simple,
> and besides geometrical singularities, there are no discretisation
> problems in it. So, below a certain threshold, mesh refinement (or
> higher interpolation) will just smooth the answer and not get you closer
> to the theoretical value. If you absolutely want to have 'ultra-smooth'
> field patterns, you could as well perform a laplacian smoothing to the
> solution inside the domain as a post-processing step...

Well, look at the attached picture, there are still some ugly steps of the order of ~1mm.
As you see symetry is very good - these 2 curves are the same curve mirrored around
z=0. Actually what I need is not finer mesh but - computing on rectangular mesh or a
better interpolation in postprocessing or a laplacian smoothing as you said. Is there such
a possibility to do it during postprocessing directly with getdp ?


Thanks,

Sam