Re: [ff3d-users] Re: Re: Reading GMSH hexahedral mesh.

[Stephane Del Pino]

Hello.

First, I think I have fixed the interpolation problem for unstructured 
Q1 meshes. It is quite slow but it seems to work in the cvs tree...

Cécile Giorla wrote:
> Because FF3D doesn't allow to refined mesh,
Note that I am not sure that there existes *free* adaptive mesh 
refinement packages in 3D. If you have anyother information about that, 
I am interested in...

> I thought importing one from 
> GMSH would be the best way to deal with this kind of problem.
> What do you think it would be best in this case?
Well, it is not easy to answer. I think it depends on what equation you 
want to solve:
- if it is just scalar problems, using unstructured Q1 meshes will be 
just as fast as unstructured tetra.
- if you indent to solve a time dependent problems such as 
Navier-Stokes, you will be penalized by the cost of the interpolation on 
unstructured hexahedral meshes: ff3d uses interpolation in many places 
to improve its genericity, but sometimes it has a cost...

For instance, for NS equations: you will need interpolation for
the convective term, this can be really costly: the problem is that you 
have to interpolate for each quadrature point of each element, each 
function of the second member and the viscosity term. Each localization 
being independent even if the functions leave on the same mesh. This is 
written in a general way, not in an optimal way. However, if some 
optimization have already been performed, there are still a lot missing 
and localization in a Q1 unstructured mesh is *very* expensive.

I think that if you can use a P1 mesh, it is probably the "best" solution...

In order to get ideas fixed, try to solve a time dependent problem using 
the 3 strategies:
- structured Q1 (more efficient)
- unstructured P1
- unstructured Q1 (less efficient).

I hope that helps,
Best regards,
Stephane.