Re: [Getfem-users] getFem problem, export bug possibly and mailing list issue

[Renard Yves]

Umut Tabak <address@hidden> a écrit :

> Dear Professor Renard,
>
> First, of all, I would like let you know, that I am registered for  
> the Getfem users mailing list, but none of my posts appear on the  
> mailing list pages on the gna server. I confirmed my membership but  
> as said, my messages do not appear online. Do you have any idea what  
> I should do.

This is probably because you use a different mail address than the one  
you give to register yourself.


> I have 1 question and possibly one inconsistency in the matrix  
> market export of the matrices that are formed by the getFem++.
>
> 1.)  I am trying to use the library for my Ph.D. research project in  
> order to assemble my system matrices (For the moment I am concerned  
> with interior vibroacoustic problems so I need the assemble the  
> structural and fluid parts of the matrices as well as the coupling  
> terms which result in an unsymmetric system, most probably, you are  
> familiar with this problem). I followed the documentation and could  
> assemble some matrices for the solution of the helmholtz equation on  
> a, say, 3d domain where I import the mesh from Gmsh. This type of  
> problem is a scalar problem since only the pressures are the  
> unknowns. I set the dimension of the mesh to 1 with the set_qdim()  
> method and use FEM_QK(3,1) without problems, though the mesh is 3D  
> but with set_qdim(1) I can get the right results which I could nor  
> understand.
>
> If I try to use the linear elasticity assemble function on page 27  
> on the manual, namely, asm_stiffness_matrix_for_linear_elasticity,  
> with mesh dimension 1, I get a wrong qdim error for the mesh_fem, if  
> I use  set_qdim(3) and for instance with FEM_PK(3,1) and  
> IM_TETRAHEDRON(5) I get
>
> Error in /usr/local/include/getfem/getfem_assembling.h, line 577 :
> invalid data mesh fem (Qdim=1 required)

This means that you give the elasticity coefficients on a vectorial  
fem. You need two fems to call this assembly procedure a first one for  
the unknown which as to be vectorial (set_qdim(N)) and a second one,
which is often a P0 method if your coefficients are constant, to  
describe the Lamé coefficients. Note that
in the current devloppement version (you can get via gna.org) there is  
an assembly procedure, namely  
asm_stiffness_matrix_for_homogeneous_linear_elasticity which does not  
require a fem for the lamé coefficient.

> Elasticity problem is a vectorial problem, is not the dimension 3? I  
> could not understand where I making a mistake I guess I have  
> conceptual misunderstanding at some point. Could you help me on this  
> problem?
>
> 2.)  I am not sure if this is a bug or not. When I form my matrices  
> for the scalar problem explained above. I export them in the matrix  
> market format using the gmm++ functions, but sometimes in the  
> export, if the matrix dimension are high, say 13690, I get 'nan'  
> entries in the exported matrix. I tried the hb export format, and it  
> seems to work correctly for the same problem, I do not get any 'nan'  
> entries in the matrices exported. If you like I can send you these  
> files for your further experimentation.

Yes please, and if possible the smallest program producing these files.

Regards,

Yves.

> With the kindest regards,
>
> Umut