Re: [Help-gsl] Nonlinear Least-Squares Fitting

[Patrick Alken]

Hello,

  Yes this is a bug, thanks for reporting it. The problem was I had 
defined the time variable to be in [0, N], where N is the number of 
points to fit. Then, for some values of N, after the first iteration the 
solver finds a negative value for lambda, and so exp(-lambda * N) gets 
very large, resulting in nan values in the expb_f function.

  I have now replaced the time variable with:

t_i = i * TMAX / (N - 1)

so that t \in [0,TMAX] and you can increase N to anything you want. I 
left TMAX = 40. I will update the example program in the repository. In 
the meantime I have attached the corrected source code. I tested it up 
to N = 1000. Please let me know if you have any further problems.

Thanks,
Patrick

On 01/11/2018 04:50 AM, Anna Russo Russo wrote:
> Hi!
>
> I've run the "Exponential Fitting Example", from the GSL 2.4 documentation,
> and I noticed that, if the number of data points is greater than 56, the
> solver terminates with status "exceeded max number of iterations". It
> happens even if I allow it to perform a considerably larger number of
> iterations: after the first iteration the parameters are so different from
> the right ones that the value of the function to be minimized is 'nan' and
> they don't change in the next iterations.
>
> What may be the cause for this behavior?
>
> Thanks for reading,
>
> Anna

nlfit.c
Description: Text Data