Re: Fzero for functions that are never negative?

[Przemek Klosowski]

On 10/18/2012 05:16 PM, Joza wrote:
> As regards the fsolve function, is it possible to specify the x-convergence
> test for EACH ITERATION? I know can input the absolute convergence error for
> "TolX", but I would rather test each iteration, which will depend on the
> value obtained in the previous iteration. Is that possible?

You can very simply modify almost all(*) Octave built-in functions, 
because they are often implemented in Octave/Matlab language. If you 
'type fzero', you'll see the implementation of the function 'fzero'; you 
can edit it to do what you want and redefine fzero() or create my_fzero().

> I'm also curious, does it happen often that simple self-made algorithms can
> provide better answers than Octave's built-in functions?

For a specific case, yes, of course---the Octave built-ins are by design 
handling a general case, so if you can figure out a shortcut that works 
better for your specific case, it'll beat the built-in.

I don't know whether it 'happens often', but new algorithms do appear 
even though a lot of smart people worked for a long time to implement 
Octave's built-ins. If you can think of a self-made algorithm that's 
better than the current one, please contribute it---that's the beauty of 
Free Software.


(*) some Octave functions are implemented in C/C++/Fortran; it's 
possible to modify them as well but it's a little more involved.