[Gmsh] structured and unstructured meshes bug

Christophe Geuzaine cag32 at case.edu
Wed Mar 22 12:59:34 CET 2006


Michel Aubourg wrote:
> Hi Christophe,
> 
> Thank you very much for this answer.
> 
> I did't find that in the documentation (the HTML one build from sources 
> distribution) where we can read that the surfaces are meshed before the 
> volumes and that the so generated triangles are forced to be faces of the 
> tetrahedra created later.
> I understood that the mesh of an internal surface was uniquely defined, which 
> is not the case; I suggest to up-to-date this documentation to avoid 
> disapointment for future users.

I'll add something in the docs. Note that this can *only* happen when 
you try to mix structured and unstructured volume meshes. In the usual 
case (unstructured volume meshes, with or without structured surface 
meshes), this will *never* happen.

> 
> In another way I join a very short file for which the use of "isotropic" gives 
> of course bad mesh and "netgen" crashes; what is wrong ?

As I said in my last email, the "netgen" algorithm requires the boundary 
of the volume to be oriented with outward pointing normals. This is not 
automatically ensured by gmsh at the moment, so you have to do it by 
hand. See attached file for a correct definition of your geometry.


> 
> I used gmsh 1.60.1 and 1.64.1 both compiled alternatively with gcc-3.4.3, 
> gcc-3.3.6 and gcc-4.0.1.
> 
> Best,
> 
> Michel
> 
> Le Mercredi 22 Mars 2006 02:11, vous avez écrit :
>> Michel Aubourg wrote:
>>> Re-bonjour,
>>>
>>> Avec le fichier joint cela ira mieux.
>> Hi Michel - Unfortunately, that's inherent to the Delaunay technique as
>> implemented in the default "isotropic" algorithm. You can use the
>> combined Delaunay+advancing front technique from the "netgen" algorithm
>> to fix this. (Add e.g. "Mesh.Algorithm3D = 4;" at the beginning of your
>> file; beware that your volume boundary has to be defined with exterior
>> pointing normals when using this algorithm.)
>>
>> Best,
>>
>> Christophe
>>
>>



-- 
Christophe Geuzaine
Assistant Professor, Case Western Reserve University, Mathematics
http://www.case.edu/artsci/math/geuzaine
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: test_corrected.geo
URL: <http://www.geuz.org/pipermail/gmsh/attachments/20060322/ce86a032/attachment.geo>