Re: [Getfem-users] different mesh_fem definitions for different regions

[Renard Yves]

Umut Tabak <address@hidden> a écrit :

> Renard Yves wrote:
> > Dear Umut,
> >
> > If your region is a part of a mesh (not a boundary)
> Dear Professor Renard,
>
> Yes, there are two volumes that are mentioned in the previous mail,  
> as the solid and fluid domains that have hexahedral meshes and the  
> interface surface between these two domains,
>
> > , you can simply defines the mesh fem by setting a finite element  
> > method only on the element of this region.
> This is what I also thought after I read the manual a bit more and  
> looking at the code documentation.
> > You can also define a mesh_fem on the whole mesh but specify the  
> > region to the assembly procedures (or to the brick if you use the  
> > new bricks in Getfem++ 4.0).
> This point is still not clear to me, maybe I should try on some  
> example codes, but I think you are suggesting to assign the element  
> properties to regions by using vectors that specify the elements in  
> that region, and use the assembly over the whole mesh_fem but using  
> the regions defined and since the fem_descriptors are assigned using  
> the region information, the matrices will be assemble correctly. I  
> guess, this is the reasonable explanation, correct me if I am wrong  
> or if you have additions. Excuse me if there are some points there  
> are inconsistent, I am a new user of the library eventually.

The last argument of assembly procedures is the mesh region. If it is  
ommited, the assembly is done on the whole region. If it is specified,  
the assembly is done only on the mesh region. Of course, the problem  
in that case is that the vector and matricies obtained by the assembly  
are on the finite element method defined on the whole mesh. So you  
have to restrict the matrices and vectors to the degrees of freedom on  
the considered region.

Probably the simplest way to do is the first one. Define the finite  
element method only on the element of the mesh region (loop over the  
element of the mesh region with an mr_visitor object and call  
mf.set_finite_element(cv, pf) ).



> I have not read about the bricks yet, but Andriy Andreykiv, who is  
> also using getFem++ quite often for his research and sending posts  
> here, has pointed that bricks can be quite versatile by hiding the  
> implementation and boundary condition application details for  
> modelling.

With a good understanding of the brick system, you can really go  
faster in building finite element modelisation, especially for  
coupling problems. But this mean to understand how to build a brick.  
The new bricks are easier to use than the old ones.


> > The third mean is indeed to use a  partial_mesh_fem object.  
> > However, the implementation in Getfem++ 3.1 is not efficient at  
> > all. The implementation
> > in Getfem++ 4.0 has been simplified and optimized.
> I am using the svn copy for the moment, so could you make the same  
> comment for the svn copy or should we wait for the next stable  
> release of the library.

You are right. Getfem++ 4.0 is not released for the moment (I wait to  
have upgrade enough old bricks into new ones to release Getfem++ 4.0).  
I wanted to talk about the current svn version.


Yves.