[Getdp] Re: Users group on GetDP ?

Christophe Geuzaine Christophe.Geuzaine at ulg.ac.be
Mon Apr 9 23:13:29 CEST 2001


Hello,

I just fell back on your e-mail, which was hiding itself on top of my
inbox mail folder, and for which my reply was waiting in the draft box
for almost one month :-(

(I don't know if my collegue Patrick provided some answers to your
questions---he's in Brazil for the moment.)

> 
> I was wondering if there is a users group where I can put forward
> some questions. Your documentation is fairly complete, and thanks
> to your examples I manage to get the computation going, but evidently
> not every option is clarified in detail. Instead of bothering you,
> I'd like to contact other users, keeping you as a last resort :) .

I have created 2 mailing lists for getdp. They are still a bit
experimental, but it would be nice if you could try them out. They are
accessible at http://www.geuz.org/mailman/listinfo.

> Also, occasionally I could contribute my calculations to a pool of
> examples.
> Perhaps also some of my notes might help another newbie user, especially
> 
> on background litterature. In aeronautics the finite volume method is
> commonly used. Standard works on Finite Elements in our domain do not
> treat edge based elements for divergence free solutions.
> An exchange between users on the use of boundary conditions would be
> quite interesting.

Yes, you're right: it is definitely something which could help new (and
even more experienced) users.

> 
> My actual problem is to extract an integral value from a solution.
> In an electrostatic problem in a conductor, I impose the potential
> at conducting surfaces, and solve for the potential in the material.
> This gives me the electric field, and hence the current densities.
> After reading the manual and trying many combinations of postprocessing,
> 
> I didn't manage to compute the surface integral of the current density
> at the conducting surface, to obtain the total current.

Did you try 'Integral' instead of 'Local' in the PostProcessing field?
(note that the fields must be defined on the surfaces).

> Any hint is appreciated.
> Also if possible some clarification on the use of the Print function
> to extract the value of a variable at a certain node. I didn't get the
> Printf function to work.

Yes, 'Printf' is a bit of a hack right now, and is only useful to debug
expressions. There is no direct way to access the value at a certain
node, except using 'Print[system-name, {dof-number,dof-number,...} ,
File "file"]' directly after a 'Solve' in a Resolution. But we will add
such a feature in the PostProcessing.

> Since it seems I'm shamelessly presenting my wish-list, a detailed
> example on the network options would be welcome.

Ok.

> I like your explanation on p. 13 of "Dof" ; on the other hand, I can
> only
> approximately understand the terms 0-form ...2-form on p. 44 (section
> 4.4).
> What is meant by a 3-form is quite a mystery. Do you have some
> references
> for an interested engineer ?

P-forms are a construction of differential geometry. Roughly stated, in
a three-dimensional space, to vector fields like the magnetic field $h$
or the electric field $e$ belonging to $H(curl}$, for which it makes
sense to compute the circulation on a contour, correspond differential
forms of degree one (called 1-forms), whereas to vector fields like the
magnetic flux density $b$ or the current density $j$ belonging to
$H(div)$, and for which it makes sense to compute the flux across a
surface, correspond differential forms of degree two (2-forms). In a
similar way, to scalar fields like the electric potential $v$ or the
charge density $q$, belonging respectively to $H1$ and $L2$, which are
evaluated locally or integrated over a volume, correspond differential
forms of degree zero and three (0-forms and 3-forms). Differential
geometry then defines a unique derivation operator (the exterior
derivative $d$), transforming (p-1)-forms to p-forms, from whom the
grad, curl and div operator are the representents in three dimensions.

You can find a good introfuction to differential geometry in

@book{schutz-geodiff-80,
  author =	 "B. Schutz",
  title =	 "Geometrical Methods of Mathematical Physics",
  publisher =	 CAMBRIDGE,
  year =	 1980
}

A more complete book is

@book{flanders-geodiff-89,
  author =	 "H. Flanders",
  title =	 "Differential Forms with Applications to the Physical
                  Sciences",
  publisher =	 DOVER,
  year =	 1989
}



> Are examples available for the use of the Green functions, section 4.2.3
> ?

Nope, but I should add some.

> Do you have an example on how to use your gnuplot files with gnuplot ?
> Gnuplot is a software I don't use very often, and I didn't manage to
> read
> the files.
> 

The basic thing to know is the "using" command. To plot a 2D graph using
columns 3 and 8, you should type something like

gnuplot> plot "file" using 3:8 with lines

> As a side remark, I didn't find the files mentioned in chapter 6
> of the manual on the webside, so I extracted them from the .pdf version
> from the manual (version 0.76).
> It appears that there is just a typo on p. 65 in Jacobian_Lib.pro :
> "JacobianMethod" should be "Jacobian".

Ok, thanks!

> 
> Just for curiosity, I was wondering if you plan to include upwind
> biased weighting functions for the computation of fluid dynamics.
> I have no intentions to use it in the near future, it is just
> that I have been working in a related field.

Yes, we have a collegue working on that topic. But maybe he'll come up
with a more general/radical solution, i.e. directly introduce a generic
discontinuous Galerkin method.

> Another question is if you plan to issue parallel versions of the code,
> either in multi-threading or using a parallel library ?
> 

We are using PETSc as a parallel solver in our development release, but
this is still a bit experimental.

> I apologize for the long mail and many questions, but I send you these
> remarks in an effort to contribute constructively to your very useful
> software.

No problem. 


With all my apologies for the late answer,


Christophe

-- 
Christophe Geuzaine

Tel: 32 (0) 4 366 37 10    http://geuz.org
Fax: 32 (0) 4 366 29 10    mailto:Christophe.Geuzaine at ulg.ac.be