dolfin team mailing list archive

[Johan Jansson <johanjan@xxxxxxxxxxxxxxxx>]

On Mon, Jul 26, 2004 at 05:37:04PM -0400, Bing Jian wrote:
> 
> Hi, Dear Dolfin folks,
> 
>    I just tried the poisson and poisson-multigrid in
> /src/demo/solvers.  Here are some notes, questions
> and suggestions:
> 
>    (1) I note there are already precompiled
> executables in each directory. However, when I run
> those binary programs on my Mandrake, it gives segmentation
> fault after the iterations converge.
> 
>    (2) So I just did "make clean" and "make" again. Since
> I copied these 2 directories out of the installation position,
> when I did make, I need to modify the prefix directory for
> configuration. I don't know if it would be better to organize the
> build process by using CMake (www.cmake.org), like what
> itk (www.itk.org) does.
> 

Are you using the version from Anders' thesis? I think it would be
better to download from the web:

http://www.phi.chalmers.se/pub/software/dolfin/v0.4/

or use CVS (code not always guaranteed to work):

cvs -d :pserver:anonymous@xxxxxxxxxxxxxxxxxxxxxx:/cvs checkout -P dolfin

I think the problem with your prefix directory is solved by running
"./configure" from the DOLFIN root.


>    (3) After re-building, I can run the solvers successfully
> and generate the poisson.m. I think it's very fast! So great!
> For visualization, in matlab with pde toolbox, I just use
> pdesurf('points','cells','u') after loading poisson.m to
> visualize the solution,  is it right? Seems the solution
> solved by multi-grid solver is smoother than another one,
> and its error is smaller (0.13529)

The Poisson module is 3D, and I don't think pde toolbox can display
such solutions/meshes. If you try the convection diffusion module,
then you can just load convdiff.m and plot: pdesurf(points, cells,
u{5}) where the cell number 5 is time point number 5.

For the vector wave module, you can load wavevector.m and then plot:
pdesurf(points, cells, u{50}(:, 1)) for component 1 and
pdesurf(points, cells, u{50}(:, 2)) for component 2 at time point 50.

> 
>    (4) As for the mesh, how to generate the mesh.xml.gz,
> using matlab or gid?  If the domain of the problem is
> a rectangle, can I just use regular mesh? And how to
> store the vectorized solution u into 2d array?
> 

I haven't looked much at the mesh part, but it seems the scripts in
"src/utils/" will do what you need. You have "writeinp.m" which
converts from matlab to INP, and then "inp2xml". Perhaps there are
even better ways.

  Johan