[Ipopt] Query about the objective function gradient

Seth Watts watts2 at illinois.edu
Thu Feb 1 10:02:16 EST 2018

Hi Chinmay -

If your objective function is differentiable, then you can get around
implementing an analytical gradient by instead implementing a finite
difference approximation. Since this will require at least one objective
function evaluation per decision variable, this could be quite slow
depending on the problem size, but it is a possible way forward. (I have
done this when my objective function depended on the output of a third
party code.) You could also look into automatic differentiation code that
could determine the gradient of your objective. However, if your objective
is non-differentiable within your feasible region, you will need to find
another optimization algorithm.

- Seth

On Thu, Feb 1, 2018 at 9:39 AM, Chinmay Rajhans Official <
rajhanschinmay2 at gmail.com> wrote:

> To: all,
> I am trying to solve nonlinear optimization problem using IPOpt solver
> inside MATLAB.
> The objective function is too complicated.
> Can giving of objective gradient be made optional?
> Or
> can the code run without supplying of the gradient of the objective
> function?
> Thank you.
> Chinmay
> _______________________________________________
> Ipopt mailing list
> Ipopt at list.coin-or.org
> https://list.coin-or.org/mailman/listinfo/ipopt
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20180201/0c48ea89/attachment-0001.html>

More information about the Ipopt mailing list