Thread Subject:

Objective function or lagrangian gradient input to fmincon?

Hello,

It is not clear from the help files whether the gradient that fmincon accepts should be the gradient of the objective function or of the lagrangian.

This page (fmincon help):
http://www.mathworks.com/help/optim/ug/fmincon.html#f186882
makes it clear in the "Hessian" section that the hessian fmincon accepts is the hessian of the lagrangian.

But the description of the inputs seems to imply that the gradient supplied is of just the objective function, not the lagrangian: "[x,fval,exitflag,output,lambda,grad] = fmincon(...) returns the value of the gradient of fun at the solution x."

So, the gradient doesn't need to be of the lagrangian; just the objective function?

Thanks!

"George Shoukry" wrote in message <kdi3jg$bte$1@newscl01ah.mathworks.com>...
> So, the gradient doesn't need to be of the lagrangian; just the objective function?
>

Simple and logical:

The gradient implemented of the objective function for must be the objective function alone. The gradient implemented of the constraint function for must be the constraint function alone.

Bruno

On 1/21/2013 2:17 AM, Bruno Luong wrote:
> "George Shoukry" wrote in message
> <kdi3jg$bte$1@newscl01ah.mathworks.com>...
>> So, the gradient doesn't need to be of the lagrangian; just the
>> objective function?
>>
>
> Simple and logical:
>
> The gradient implemented of the objective function for must be the
> objective function alone. The gradient implemented of the constraint
> function for must be the constraint function alone.
> Bruno

To add to what Bruno said, you could check the documentation for
objectives and constraints:
http://www.mathworks.com/help/optim/ug/writing-objective-functions.html#bsj1e55
http://www.mathworks.com/help/optim/ug/writing-constraints.html#brhkghv-16

Alan Weiss
MATLAB mathematical toolbox documentation