CustomUExtFunction within CustomExactFunction*

[Tomas Svaton]

Hello everybody,

I have taken a look on the class UExtFunction. Could someone suggest me how to implement the following issue:

class CustomUExtFunction : public UExtFunction<double>

{

public:

  CustomUExtFunction(double mu, double lambda, CustomExactFunction* MyFun) : UExtFunction<double>(), mu(mu), lambda(lambda), MyFun(MyFun)

  {

  }

  virtual void value (int n, Func<double>** u_ext, Func<double>* result, Geom<double>* geometry) const

  {

    for(int i = 0; i < n; i++)

    {

      double ux = u_ext[0]->dx[i];

      double uy = u_ext[0]->dy[i];

      double vx = u_ext[1]->dx[i];

      double vy = u_ext[1]->dy[i];

      double dS11dE11 = mu * MyFun * ux;

      result->val[i] = dS11dE11;

    }

  };

  virtual void ord(Func<Hermes::Ord>** u_ext, Func<Hermes::Ord>* result) const

  {

    result->val[0] = u_ext[mu]->val[0];

  };

  int mu;

  CustomExactFunction* MyFun;

};

The problem is in the incompatibility of the datatypes on the line:

double dS11dE11 = -2.0/3.0 * mu * MyFun * ux;

This would permit to implement more complex weak forms with non constant coefficients over the domain.

Thanks for any help or advice.

Kind regards,

Tomas.

[Lukas Korous]

Hello Tomas,

could you send description of the class CustomExactFunction?

Thanks,
Lukas

> --
> You received this message because you are subscribed to the Google Groups
> "hermes2d" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to hermes2d+u...@googlegroups.com.
> To post to this group, send email to herm...@googlegroups.com.
> Visit this group at http://groups.google.com/group/hermes2d.
> For more options, visit https://groups.google.com/groups/opt_out.



--
Lukáš Korous

[Tomas Svaton]

Hello Lukas,

I am sorry about my rough question. This is the custom function in definitions.cpp

double CustomExactFunction::val(double x, double y)

{

  double scaleNu = 10;

  double scaleE = 10;

  double Eln = E * (1-1.0/scaleE) * y + E/scaleE;

  double nuln = nu * (1.0-1.0/scaleNu) * y + nu/scaleNu;

  return (Eln * nuln) / ((1 + nuln) * (1 - 2*nuln));

}

double CustomExactFunction::dx(double x, double y)

{

  return 0.0;

}

double CustomExactFunction::ddxx(double x, double y)

{

  return 0.0;

}

and its declaration definititon.h

class CustomExactFunction

{

public:

  CustomExactFunction(double E, double nu) : E(E), nu(nu) {};

  double val(double x, double y);

  double dx(double x, double y);

  double ddxx(double x, double y);

protected:

  double E, nu;

};

Thank you very much once more.

Kind regards,

Tomas.

[Lukas Korous]

OK then,

you can probably just write:


...

double dS11dE11 = mu * MyFun->val(geometry->x[i], geometry->y[i]) * ux;

...

Am I right?

Lukas

[Tomas Svaton]

Thank you very much, Lukas. 

You are true ant it works.

Kind regards,

Tomas.

[Lukas Korous]

Cool,

send some pictures-results, I would be interested :-).

Regards,
Lukas

[Tomas Svaton]

Hi Lukas,

I thought I understood the way, but there is still something missing to me in the implementation and I have still ask you. Could you take a look on it, please?

I have this code in main.cpp (like is in devel):

...

UExtFunctionSharedPtr<double> fn_0(new CustomUExtFunction(0));

DiscreteProblem<double> dp(&wf, spaces);

wf.set_u_ext_fn(fn_0);

...

My question is: How could I access the variable fn_0 via definitions.cpp (eventually via definitions.h) and use it in the definitions of custom weak forms, e.g. in cycles like

double result = 0;

for (int i = 0; i < n; i++)

{

  result += - wt[i] * (mu * fn_0->val[i] * u->dx[i] * v->dx[i]);  // here function fn_0=sqrt(u_1 + u_2)

}

return result;

I did not found it in devel. I use the scheme (maybe obsolete now) "definitions.cpp & definitions.h & main.cpp" and I have touched the original class which is in devel inthe following way (to be splitted into definitions.cpp and definitions.h):

I have created an example of function in definitions.cpp

CustomUExtFunction::CustomUExtFunction(int index) : UExtFunction<double>(), index(index) {}

void CustomUExtFunction::value(int n, Func<double>** u_ext, Func<double>* result, Geom<double>* geometry) const

{

  for(int i = 0; i < n; i++)

  {

    result->val[i] = Hermes::pow(u_ext[0]->val[i] + u_ext[1]->val[i],1.0/2.0);

    result->dx[i] = u_ext[index]->dx[i];

    result->dy[i] = u_ext[index]->dy[i];

  }

}

void CustomUExtFunction::ord(Func<Hermes::Ord>** u_ext, Func<Hermes::Ord>* result) const

{

  result->val[0] = u_ext[index]->val[0];

};

then the interface in definitions.h

class CustomUExtFunction : public UExtFunction<double>

{

public:

  CustomUExtFunction(int index);

  virtual void value(int n, Func<double>** u_ext, Func<double>* result, Geom<double>* geometry) const;

  virtual void ord(Func<Hermes::Ord>** u_ext, Func<Hermes::Ord>* result) const;

protected :

  int index;

};

Anyway I enclose the most recent obtained results with Hermes. I hope you will enjoy them.

I will be glad for any advice.

Thank you in advance.

Kind regards,

Tomas.

[Tomas Svaton]

Hi Lukas,

Stop, stop, stop, please. I am sorry about a little bit chaos. I am trying now another way that I did not realised is much more simple.

I will keep you informed.

Anyway thank you very much.

Kind regards,

Tomas.

You received this message because you are subscribed to a topic in the Google Groups "hermes2d" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/hermes2d/W9j8CoZ1tyw/unsubscribe.
To unsubscribe from this group and all its topics, send an email to hermes2d+u...@googlegroups.com.

[Lukas Korous]

Hi,

sorry for delay, it's a rather busy time now.

Hi,

basically, the behavior - not documented yet - sorry, is that whatever
you put to the "u_ext functions" through

wf.set_u_ext_fn(fn_0);

appears in the front of the ext array in the value method, so you would write:

for (int i = 0; i < n; i++)
{

result += - wt[i] * (mu * ext[0]->val[i] * u->dx[i] * v->dx[i]);
}

And your scheme is just fine, it is not obsolete at all.
Nice pictures.
Lukas