[ff3d-users] mesh adaptation with ff3d

[Stephane Del Pino]

Hello Emilio.

Emilio Melero García wrote:
> The variables are local now. So far i have not tried to refine the mesh 
> so it is not clear to me now how this should be done... How does it help 
> the variables being local??

Well. Local variables are a more standard behavior regarding to other 
programming languages. It is a bit technical, but before, variables were 
created once for all: even if the variable was declared in a loop. This 
introduced a major problem for femfunction variables since they were 
always defined on the same mesh. This means that the variable always 
leaved on the mesh used for their creation. And this made refinement not 
possible.

Shortly just considere that previously changing a variable mesh was not 
possible, now it is...

> And i would like to take the opportunity to raise a question...What is 
> the best approach to improve the numerical accuracy of ff3d 
> computations?. Is it always to use a finer mesh (larger computing cost)?
Well, in fact, to get a better solution you need to have a fine mesh 
where it is needed or to have a better approximation.
If I remember your case, if it is possible for you to define new meshes 
that are finer near gradients with gmsh, it could be a good starting 
point. I am not a specialist of gmsh, but, if it can generate meshes 
based on a metric definition it should work with ff3d. The computational 
loop would look like :

generate a first mesh with your favorite mesher

do {
  load the mesh
  solve the problem on the mesh
  get some adaptation criterion (gradient,...)
  re-generate the mesh with your mesher using the adaptation criterion
} while(...);

I will have time next week to work on ff3d, I will try to do such a 
procedure ...

After proceeding to some research on the web, I am not sure that gmsh or 
netgen can do dynamic mesh adaptation, does anyone have information on 
that ? Maybe Benoit ?

Best regards,
Stephane.