[Gmsh] questions on a 3D mesh

[Christophe Geuzaine]

Joumana EL-GHARIB wrote:
> Hello,
> 
> We use GMSH for 3D mesh construction.
> 
> We are encoutering two kinds of problems :
> 
> 1- In the windows 1.44 version it is possible to design the geometry but
> not to do the meshing,
> while it is perfectly possible in the Linux 1.44 version.
> 
> 2-In the Linux version (the one where the meshing is possible), it is not
> possible
> to change the mesh size (the only we obtained is 40000 elements) while we
> are looking to have
> a mesh with less elements. Do you have ideas to control  such 3D mesh
> refinement ?
> 

Joumana - Here is what I obtain when I mesh your model with the version
distributed on the web (Linux, gcc 2.95.3, -O2):

gmsh test.geo -clscale 1.0 : 42191 elements
gmsh test.geo -clscale 1.2 : failed
gmsh test.geo -clscale 1.4 : failed
gmsh test.geo -clscale 1.6 : failed
gmsh test.geo -clscale 1.8 : failed
gmsh test.geo -clscale 2.0 : 5929 elements
gmsh test.geo -clscale 2.2 : 4726 elements
gmsh test.geo -clscale 2.4 : failed
gmsh test.geo -clscale 2.6 : 3065 elements

But if I use my current development version (Linux, gcc 2.96, -O2), I get:

gmsh test.geo -clscale 1.0 : 41604 elements
gmsh test.geo -clscale 1.2 : 35011 elements
gmsh test.geo -clscale 1.4 : 14618 elements
gmsh test.geo -clscale 1.6 : 10183 elements
gmsh test.geo -clscale 1.8 : 7466 elements
gmsh test.geo -clscale 2.0 : 5831 elements
gmsh test.geo -clscale 2.2 : 4738 elements
gmsh test.geo -clscale 2.4 : 3516 elements
gmsh test.geo -clscale 2.6 : 3037 elements

I also get similar results using the latest gcc version (3.3). The
problem thus probably comes from our non-robust implementation of
various geometrical predicates, combined with bogus gcc optimizations.

Just download the latest sources (http://geuz.org/gmsh/src/) and
recompile on your machine using a recent version of gcc. In any case, I
will build the next Linux binary release (gmsh 1.45) using gcc 3.3.

I don't have access to a Windows machine at the moment, so I cannot test
that version.

Christophe

-- 
Christophe A. Geuzaine
Applied and Computational Mathematics, Caltech
geuzaine at acm.caltech.edu - http://geuz.org