Re: [Bug-gsl] bug in gsl_multifit_wlinear

[Brian Gough]

Richard Mott writes:
 > There is a bug in gsl_multifit_wlinear (and presumably in 
 > gsl_multifit_linear too). The problem manifests itself when the design 
 > matrix X is not of full rank. The documentation says that this case is 
 > handled correctly in the singular value decompostion by setting the tiny 
 > SVD components to 0. In fact his doesn't happen. Instead the function 
 > returns a huge and non-sensical chisq value. The block of code in 
 > question is

Thanks, yes, the handling of near-zero singular values could be
improved... I've added an entry to the BUGS file for this.

-- 
Brian Gough

Network Theory Ltd,
Publishing Free Software Manuals --- http://www.network-theory.co.uk/