Re: nonlinear fit with weights

[Brian Gough <bjg at network-theory dot co dot uk>]

M Atakan Gurkan writes:
 > I am trying to use the nonlinear multifit functions in GSL with a
 > weight. Different points in my data set has different error bars
 > and I would like to use this information while making the fit.  I
 > basically copied and modified the example from the manual, then I
 > set values of sigma[] to reflect my error bars. However the only
 > effect of changing sigma was to change the error estimates in the
 > fit, it did not change the fit itself.  I think I am doing
 > something completely wrong. I read the manual, but could not find
 > anything that looks helpful. Is it possible at all to use different
 > weights for different data points (as in, e.g., xmgrace)? And if
 > so, how?

The gsl_multifit_fdf routines have been tested against the NIST
standard reference datasets, which use variable weights (see
multifit/test*.c for examples). So I think that GSL itself is ok.

-- 
Brian Gough