[SciPy-user] Solver with n-dimentional steps

[David Warde-Farley]

On 15-Jun-08, at 4:18 AM, Yosef Meller wrote:

> I'm trying to use scipy.optimize.fsolve(function, x0, args) with a  
> function
> that has an input vector with length in the tens of thousands (about  
> 3*4000).
> This turns out to be impractical, because fsolve runs the function  
> after
> taking a step in each direction in turn. Runing the function takes  
> about 1/10
> second, so in the thousands of runs it's way too much.

It's doing finite differences to estimate the gradient. This is pretty  
much unavoidable if you're only giving it those three arguments. Have  
you thought about supplying fprime as a function to analytically  
compute the gradient?

David