[Gmsh] Question / potential bug - characteristic length meshing issue

[Christophe Geuzaine]

Keith Johnson wrote:

> Greetings,
> 
> I am fairly new with respect to gmsh, and have been following the
> distribution list mail for several months now.  Most of the questions
> and answers are very helpful.
> 
> I have run into a problem which can be reduced to a simple case of
> looking at a 2d / 3d mesh on a finite thickness cylindrical disk with a
> hole (compressed geom file attached).
> 
> When I set the characteristic length as the one that is presently
> active, I have problems with surfave and volume mesh generation.  If I
> set the characteristic length, lcart2, to any of the other two values,
> every thing works. 
> 
> This is only a small part of a much larger model, but my present
> difficulty can be reduced to this simple case.
> 
> It is highly possible that I am not doing something correctly, but it
> does not make sense that two values of characteristic length work fine
> and the third does not.
> 
> The case, here is that the smallest characteristic length that works is
> not sufficient discretization.
> 
> Have I bumped into a bug, or am I not doing something correctly?

No, it's a limitation of our projection technique (we always generate
the 2D grids on planes, and then we project the results on the curved
surfaces).

If you take a look at the log, you will see that we couldn't compute a
good "mean" projection plane for surface 18 (cf. "using rough algo").
When this happens, the only solution at the moment is to reduce the
curvature of the problematic surface (in you case, you could e.g. do two
extrusions instead of one).

Christophe

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