Re: [Getfem-users] nonlinear brick

[Yves Renard]

Dear Roman,

There is at least two nonlinear working new bricks. The nonlinear elasticity 
brick (getfem_nonlinear_elasticity.cc) and the contact with friction brick   
(getfem_Coulomb_friction.cc). But they are not present in getfem++ 4.0.0.
They are in the svn version. You can also download the archive here (this is a 
svn snapshot, not a stable release) :

http://math.univ-lyon1.fr/homes-www/renard/temp/getfem-4.0.tar.gz

I apologize, I did not yet documented how to build a nonlinear brick ! The 
most important things :

model::build_version nl  indicates weither the right hand side or the tangent 
matrix is to be assembled.

const model::varnamelist &varl  contains the variables whose value can be used 
to assemble the rhs or tangent term (the value is obtained with 
md.real_variable(varl[i]) as for the data).

Your term
 double a = vecl[0][0];
cannot be correct. vecl is to be filled by  asm_real_tangent_terms. It is 
initialized with 0.


Yves.


On mercredi 24 mars 2010, Roman Putanowicz wrote:
> Dear All,
>
> Could someone, please, point me to an example involving nonlinear brick
> using the new brick system?
>
> Actually, I am trying to get the real filling of GetFEM++ internals.
> I just thought to use GetFEM++ to solve a problem which boils down
> to a single quadratic scalar equation. I know that this is not what
> GetFEM++ was meant for, but in principle GetFEM++ allows to model board
> kind of mathematical models so why not this :).
>
> I have defined the following brick:
>
> namespace getfem {
>   struct my_quadratic_eq_brick : public virtual_brick {
>     void asm_real_tangent_terms(const model &md,
>                               getfem::size_type ib,
>                               const model::varnamelist &varl,
>                               const model::varnamelist &datal,
>                               const model::mimlist &mims,
>                               model::real_matlist &matl,
>                               model::real_veclist &vecl,
>                               model::real_veclist &vecl_sym,
>                               size_type region,
>                               model::build_version nl) const {
>
>       GMM_ASSERT1(matl.size() == 1, "my_quadratic_eq_brick has only one
> term"); GMM_ASSERT1(mims.size() == 0,
>         "my_quadratic_eq_brick does not need mesh_im");
>       GMM_ASSERT1(varl.size() == 1 && datal.size() == 1,
>         "Wrong number of varibles for my_quadratic_eq_brick");
>
>       /* get equation coefficients */
>       const model_real_plain_vector COEFF = md.real_variable(datal[0]);
>
>       gmm::dense_matrix<double> mm(1,1);
>       double a = vecl[0][0];
>       mm(0,0) = COEFF[0]*a*2.0 + COEFF[1];
>       gmm::copy(mm, matl[0]);
>     }
>
>     my_quadratic_term_brick(bool symmetric_, bool coercive_) {
>       set_flags("Quadratic term brick",
>                 false /* is linear*/,
>                 symmetric_ /* is symmetric */, coercive_ /* is coercive */,
>                 true /* is real */, true /* is complex */);
>     }
>   };
>
>   size_type add_my_quadratic_eq_brick
>   (model &md, const std::string &varname, const std::string &dataname) {
>     pbrick pbr = new my_quadratic_term_brick(false, false);
>     model::termlist tl;
>     tl.push_back(model::term_description(varname, varname, true));
>     model::varnamelist dl;
>     dl.push_back(dataname);
>     dl.push_back(varname);
>     return md.add_brick(pbr, model::varnamelist(1, varname), dl, tl,
>                       model::mimlist(), size_type(-1));
>   }
> } /* namespace getfem */
>
>
> Well, so far so good.
>
> Now a bit of code which uses the brick:
>
>   md.add_fixed_size_variable("x", 1);
>   std::vector<scalar_type> F(1);
>   std::vector<scalar_type> coeff(3);
>   coeff[0] = 7.0;
>   coeff[1] = 11.0;
>   coeff[2] = 2.0;
>   md.add_initialized_fixed_size_data("coeff", coeff);
>
>   md.first_iter();
>
>   F[0] = 2.0;
>   gmm::copy(F, md.set_real_variable("x"));
>
>   getfem::size_type id = add_my_quadratic_eq_brick(md, "x", "coeff");
>
>   md.assembly(getfem::model::BUILD_MATRIX);
>
>   std::cout << md.real_tangent_matrix();
>
> And here is the problem. On standard output I get the 1 by 1 matrix
> whose only element value is 11 (the coefficient at linear term in the
> equation). So it seems that the "x" vector is uninitialised.
> Is there any mechanism that ensures that the tangent matrix
> will be built on the basis of supplied guessed solution?
>
> I thought that this will be done with these two lines
>
> F[0] = 2.0;
> gmm::copy(F, md.set_real_variable("x"));
>
> but apparently I miss something.
>
> I would appreciate any hint.
>
> Regards
>
> Roman



-- 

  Yves Renard (address@hidden)       tel : (33) 04.72.43.87.08
  Pole de Mathematiques, INSA-Lyon             fax : (33) 04.72.43.85.29
  20, rue Albert Einstein
  69621 Villeurbanne Cedex, FRANCE
  http://math.univ-lyon1.fr/~renard

---------