Hi,

On Tuesday 08 April 2008 10:46:37 Leopold Palomo-Avellaneda wrote:
> I have been lately working with the gsl library, using the SVD
> decomposition, but I am having real trouble, so i thought that i was
> probably doing something wrong and you could help me.

yes, you are doing something wrong: You calculate the pseudoinverse in a 
completely different way in the C++ program and in the Matlab script and 
expect to get the same results. In the C++ program, you do it using 
gsl_linalg_LU_decomp () and gsl_linalg_LU_invert () which goes badly wrong 
because the third diagonal entry of the matrix S is very close to zero. In 
the Matlab script, you apparently thought of this problem and did it this 
way:
 
% As S is diagonal, que can compute its inverse by dividing inverting its
% diagonal values
for i = 1:max
   if (S(i,i) > 0.0000000001)
       S(i,i) = 1/S(i,i);
   else
       S(i,i) = 0;
   end
 end

The "if (S(i,i) > 0.0000000001)" check avoids the huge matrix entries which 
caused the problems in your C++ program. I've attached a fixed version of 
your program which reproduces the results of the Matlab script (at least up 
to the limited precision of the Matlab results).
 
Frank

P.S.: Apparently, my first reply was rejected by the list server because the 
Google Mail web interface for some reason decided to send it as text/html.