Thanks for your answers!
Patrick, indeed, after looking at the code, the difference arose between
versions 1.16 and 2.0. However, I can't use gsl_multifit_linear because I
also need the rank to compute the unbiased estimate of the error variance,
as well as the p value.
The rank was returned
Hello,
Positive definite does not mean all the matrix entries are positive.
It means all the eigenvalues of the matrix are positive. In your case,
both H and S have negative eigenvalues, so neither matrix is positive
definite. In this case, you need to use the Real Generalized
Nonsymmetric