Re: What is the best approach - 'sqp' or 'fsolve' ?

[Guido Walter Pettinari]

Hi Jaroslev,

thank you for your answer!

Since I have to minimize just one function - not a system - and I have no 
constraints, I guess I am better off with 'fminunc'. I am using Octave 3.2.3; 
do you deem the 'fminunc' included there is robust enough?

I would like to ask you one more question about optimization. Which 'fzero' 
function would you recommend, the one in Octave core or the one in the 
optimization package?

Thank you very much for your time!

Cheers,

Guido

On Jan 28, 2010, at 12:40 , Jaroslav Hajek wrote:

> On Thu, Jan 28, 2010 at 12:27 PM, Guido Walter Pettinari
> <address@hidden> wrote:
>> Hello world!
>> 
>> I want to minimise the function f ( x ) with respect to the parameters 
>> contained in "x".
>> I was wondering which of the following approach is better/faster/more 
>> accurate:
>> 
>> 1) use the "sqp" optimisation routine, providing gradient + Hessian of f(x);
>> 
>> 2) use the "fsolve" routine to solve the system:
>> gradient(f) = 0,
>> providing the Jacobian.
>> 
>> Thank you for your attention.
>> 
> 
> In most cases, definitely the first. By treating the gradient as
> independent residuals, you ignore the structure of the problem.
> You may also consider trying fminunc (esp. if using tip Octave), but
> that currently can't work with user hessians.
> 
> -- 
> RNDr. Jaroslav Hajek, PhD
> computing expert & GNU Octave developer
> Aeronautical Research and Test Institute (VZLU)
> Prague, Czech Republic
> url: www.highegg.matfyz.cz