Re: [Help-gsl] Gsl-1.8 compiled and running under MSYS-1.10+second order derivative

[Rene Girard]

Brian,

  I did as you recommended i.e. I tried 

  f_x.function = (double (*) (double, void *)) &g_x;
 and 
   f_x.function =  &g_x;

but I got the following when I compiled the 
program given below:

rm -f ./t_dn2.o
rm -f ./t_dn2
gcc  -O2 -Wall -ansi -I/usr/local/include -c -o t_dn2.o t_dn2.c
t_dn2.c: In function ‘g_x’:
t_dn2.c:61: warning: dereferencing ‘void *’ pointer
t_dn2.c:61: error: void value not ignored as it ought to be
make: *** [t_dn2.o] Error 1

If instead of 
- - - - - - -  program follows - - - - - - - - 
/*
 * Programme t_dn2.c: Le programme t_dn2.c donne une methode d'utilisation 
 *                    des fonctions de la librarie GSL pour le calcul de la 
 *                    derivee partielle du premier ordre d'une fonction a deux 
variables 
 *                    z = f(x,y) par rapport a x. Dans ce programme on 
considere 
 *                    la fonction test:
 * 
 *                          f(x,y) = (sin(pi*x/2)*sin(pi*y/2))^2
 *
 */
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <gsl/gsl_math.h>
#include <gsl/gsl_deriv.h>


double g_x(double, void *);

int main(void)
 {
  gsl_function f_x;
  const double pi = 4.0*atan(1.0);     
  double ro,r1,dx,er1,x,y;
  double p1[2];
  int i;

  y = 0.65; 

  dx = 0.1;

  p1[0] = 1.0e-04;
  p1[1] = y;

/*  f_x.function = (double (*) ()) &g_x; */
  f_x.function = &g_x;
  f_x.params = &p1[0];

  x = -dx;
  printf("f(x,y) = (sin(pi*x/2)*sin(pi*y/2))^2\n");
  for(i = 1; i <= 11; i++)
   {
    x += dx;
    gsl_deriv_central(&f_x,x,p1[0],&r1,&er1);
    ro = pi*(sin(pi*p1[1]/2.0)*sin(pi*p1[1]/2.0))*
         cos(pi*x/2.0)*sin(pi*x/2.0);      
    printf("Pour x = %11.6f et y = %5.2f\n",x,y);
    printf("f_x           = %14.10f +/- %14.10f\n",r1,er1);
    printf("Valeur exacte = %14.10f\n",ro); 
   }

  return 0;

 } /* fin du programme t_dn2.c */    

double g_x(double x, void *params)
 {
  const double pi = 4.0*atan(1.0);    
  double yo,y1;

  yo = (sin(pi*x/2.0)*sin(pi*params[1]/2));
  y1 = yo*yo;

  return y1;  
 }
- - - - - - - - - - - - - - - - - - - - - - -
If I go back to the verison that worked and put the cast

(double (*)(double, double *))
in the following statement 

  f_x.function = (double (*) (double, double *)) &g_x;

I get this
t_dn2.c: In function ‘main’:
t_dn2.c:35: warning: assignment from incompatible pointer type
gcc -o ./t_dn2 ./t_dn2.o -I/usr/local/include -L/usr/local/lib -s               
-lgsl -lgslcblas -lm

Again if I revert to the cast

  f_x.function = (double (*) ( )) &g_x;

No problem! It compiles and gives the correct result.

Now I am wondering if I am not understanding correctly  the purpose and usage 
of the argument "void *params" which for me is to be able to pass parameters to 
the function used to evaluate the numerical derivative.

This use of the argument "void *params" allows me to calculate cross derivative 
of a 2 variables function: d_xy f(x,y). I tried it and it works.

I wish to be able to follow your recommendation but as you can see there is 
something perhaps that I am not understanding.

Note: I use gcc 4.1.1 to compile the program

Regards

Rene



Brian Gough <address@hidden> wrote: Rene Girard wrote:

> double g1(double, double *);
> 
> I would like to bring to your attention that when I
> compile that simple example and
> I set the pointer to the function from we want to have
> the derivative with the following
> statement:
>  
>   f.function = &g1; (like in the example in Chap. 27)
> 
> I get the following warning:
> 
> "warning: assignment from incompatible pointer type"
> 
> to remove the warning I had to cast the pointer as
> follow:
> 
>  f.function = (double (*) ()) &g1;

Hello,
Thanks for your email.  I think your warning can be avoided without a 
cast by using the prototype double g1(double, void *) which matches the 
definition in GSL.  Otherwise, the appropriate cast would be (double 
(*)(double, void *)). See the header file gsl_math.h for the definition 
of the gsl_function type.

-- 
best regards,

Brian Gough
(GSL Maintainer)

Network Theory Ltd,
Publishing the GSL Manual - http://www.network-theory.co.uk/gsl/manual/


                
---------------------------------
The best gets better. See why everyone is raving about the All-new Yahoo! Mail.