GetDP

[Samuel Kvasnica]

> You could use higher order interpolation, e.g. by adding new basis
> functions to the nodal basis functions used in your problem. For
> example, the following line could be added in the function space (this
> defines hierachical basis functions associated with edges):
>
> { Name se2 ; NameOfCoef ae2 ; Function HBF_PerpendicularEdge_2E ;
>   Support Domain_Mag ; Entity EdgesOf[ All ] ; }
>
> Constraints should of course also be defined on these functions.

I added this but results seem to be even more hairy (see pictures b_1f with 1 base function, b_2f with HBF_...
added).
Note that althought b_1f looks quite good, the symetry is somewhat broken (also compared with experimetal
data).
That would be no problem for final presentation but I need to feed an PIC plasma simulation with this data and
I'd like
to have a bit better results. Should I use brute force and increse mesh density again or is it just some
problem with interpolation used ? And again -I'm not able to increase node density too much since it segfaults
with smaller 'characteristic lenght' more and more.

The problem is I'm using this part of getDP as a black box and don't understand how it is working at all. Well,
for somebody
with experience in this field would be getDP manual enough, but for me it is too brief. Could you recommend me
some
good source of info to get just light overview ?


> You should probably adapt the solving technique by editing the
> SOLVER.PAR file (e.g. invrease de number on non-zero elements kept in
> the ILU factorization which serve as a preconditioner to the krylov
> accelerator).

this worked, it doesn't diverge now.

> Yes, I agree. BTW, I would like to have suggestions to produce some
> simple post processing formats. How do you think data should be
> presented in the default format? A bit like with the 'gnuplot2' option?
> Or should the numbers be more strongly formatted (e.g. with the same
> number of digits)? Or...?

Well, the best would be to support some standard data format like netCDF. I'm planning to implement
it also in my data acquisition software (http://www.iaee.tuwien.ac.at/sensor/samplin). But anyway,
ascii format is inevitably needed and the 'gnuplot2' is quite ok. As I said, with awk one can filter out
anything needed. Regarding number format, you could add an string option accepting printf like
arguments e.g. "%f10.5".


> > And the last one:  Is there a trivial way to display magnetic field line by giving coordinates of 1 point?

>
> ???

Imagine e.g. permanent magnet. I'd select a point somewhere on S pole and want to diplay magnetic field curve
flowing
from this point towards N pole. Yes, actually for 2D the z component of A could be used but it would be nice if
e.g. the color of
the line would correspond to the strenght of field. Or a possibility to generate just coordinates of points
sitting on one field curve.

Regards,

Sam

--
--------------------------------------------------------------
Mag.Samuel Kvasnica
IAEE TU-WIEN
Gusshausstrase 27-29 / 359   email: kvasnica at iaee.tuwien.ac.at
1040 Wien                    tel:   +43-1-58801-35914
Austria                      fax:   +43-1-58801-35997
--------------------------------------------------------------