[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [ff3d-users] (no subject)

From: Stephane Del Pino
Subject: Re: [ff3d-users] (no subject)
Date: Thu, 8 Feb 2007 00:13:31 +0100
User-agent: KMail/1.9.5

Le mercredi 7 février 2007 13:52, Benoit DESJARDINS a écrit :
> Hello
> As a big fan of freefem in 2D for almost ten years, I am trying to learn
> about ff3d. I got a couple of problems and questions that some of you may
> already have encountered. many thanks in advance
> Benoit

Hello Benoit,
> Questions about FF3D
> 1) Mesh
>      a. GMSH
>              i. Freefem3D used to work fine with GMSH type format. Since
> the V2 version of GMSH, file formats have changed,   which makes GMSH
> incompatible with Freefem3D. The format differences are described in the
> pdf documentation on GMSH web site.
Thanks for the information, I did not know. I will try to add this new format 
soon. I still have a bug to fix for the surface mesh generation when multiple 
boundaries are given. I need also to finish the localization of variables (up 
to now language variables are global: once they are defined they exist until 
the end of the code, there is not a memory leak, it is just a bad behavior of 
the language). Then I will do this.

>             ii. Did anyone try to take account of boundary conditions with
> GMSH meshes (identify parts of surface boundary with specific boundary
> conditions).
Yes. You just use the reference number associated to your boundary:
For PDE description:
        u = g on ref; // ref is of course an integer
and in variational formula, do
        int[ref](u*v)+... = int[ref](g*v); // for instance to describe Robin BC

>      b. In the case of Cartesian structured meshes, is it possible to use
> meshes with geometrical progressions.
Yes. But you have to be careful: for optimization means, ff3d makes difference 
between cartesian uniform structured meshes and other kind of hexahedra 
meshes. If you build a cartesian structured mesh with geometric progression, 
it will be treated the same way as unstructured hexahedra meshes...

The big problem with this is that interpolation of fem function defined on 
that kind of meshes is not implemented (*).

Any way, you can use tetrahedra mesh:
        mesh m = structured(n,a,b);
        m = tetrahedrize(m);
        m = transform(m,[f1,f2,f3]);
where [f1,f2,f3] is a tranformation field

>      c. Do adaptmesh – like functions exist in FF3D like in 2D (I am a big
> fan of 2D adaptmesh function) ?
Well, up to now, it is not possible. Moreover, in my job I am working on mesh 
adaptation, collaborating with P. Frey (University of Paris 6) and C. 
Dobrzinsky (University of Bordeaux). So, I cannot put their adaptation 
technique into ff3d without their agreement.
On a second hand, frontal adaptation based on the resolution of 
Hamilton-Jacobi equation could be a good alternative. I should work on that 
topic with P. Hoch, but we need to find the time to do it...

So this might be possible in an undefined future...

> 2) Inputs
>      a. Can one read a scalar function on a given mesh (like .bb formats) ?
        function u = read(medit,"u.bb",M); // M is the mesh where u is defined

> 3) Outputs
>      a. In case of vtk postprocessing files (or other formats), one would
> like to write a list of scalar + vectors in a single file.
> save(vtk,fichier_toto,{a,b,c,d,e},M)) ?
This is not possible. I put it on the todo list ;-)

> 4) On a longer term, will the code be parallelized? Mesh partitioning for
> domain decomposition ?
I hope so. I need to finish a big task before version 1.0 which is allow the 
use of mixed finite elements. After that I will start version 2.0 which will 
be parallel: linear algebra parallelism only, I do not think that ff3d is 
adapted to domain decomposition...
The question is when will this be achieved?

Best regards,

(*) in fact I do not know how to do it, by the way if someone knows, I am 
interested... My problem is that I am not able to invert the conform 
transformation: I have try to use a Newton algorithm but it does not always 
converge ...

reply via email to

[Prev in Thread] Current Thread [Next in Thread]