Re: [Bug-gsl] bug in compute_gradient_direction?

[Yajun Wang]

Hi:

I modified the example provided in gsl manual in the section of
"Nonlinear Least-Squares Fitting". The original example have 40 terms
to minimize, now I change it to 2. Though the error produced is not
exactly the error from my program, I think the basically idea is the
gsl behavior when the number of terms is less than the number of
parameters: (N < p).

The error is as follows:
""""""
gsl: covar.c:51: ERROR: Jacobian be rectangular M x N with M >= N
Default GSL error handler invoked.
""""""

I suspect that gsl always expect N>= p, which might be reasonable. But
the code should inform the user earlier, using some checkers. Because
the error such as above is confusing for the users.

Nevertheless, in my own program, I add one more dummy term to raise N
up to p, which works quite well. Thanks for your work.

regards,
yalding


On 3/28/06, Brian Gough <address@hidden> wrote:
> Yajun Wang writes:
>  > I want to use the Nonlinear Least Squares Fitting. I am working on the
>  > sum of 3-dimensional formulas. However, the program quits quietly
>  > whenever I have only 2 formulas to solve.
>
> Please can you send a example program which we can compile to
> reproduce the problem -- thanks.
>
> --
> Brian Gough
>
> Network Theory Ltd,
> Publishing the GSL Manual - http://www.network-theory.co.uk/gsl/manual/
>

main.c
Description: Text document