Re: [Help-gsl] LM non-lin fit

[Brian Gough]

Jim Cadien writes:
 > I have been using the gsl_multifit_fdfsolver_lmsder.  My data are 
 > such that the errors in the independent variable are as large as 
 > in the dependent variable.  I see no way in the gsl routines to 
 > use those independent errors in the fitting.  Has anyone done 
 > this, and is it make a significant difference to the fit?

LMSDER can do this if you provide the appropriate chisq function.

Taking the example in the manual as a starting point, it just needs
the orthogonal distance from (Xi, Yi) to the fitted function, summed
over i (i.e. for all points).  You'd need to modify the derivatives
accordingly.

-- 
Brian Gough