I think supplying a constant gradient will lead to some convergence issues as the " landscape" (hence the gradient) of your objective function changes with each iteration in your optimization. So if you are better off using a derivative free algorithm (such as BOBYQA) if there are issues with your gradient.
Regards, Ngoy Sent from my Windows Phone From: Steven G. Johnson Sent: 2014/01/23 09:14 PM To: [email protected] Subject: Re: [NLopt-discuss] Nonlinear constrained optimization using SLSQP returns exception "nlopt roundoff-limited" On Jan 23, 2014, at 11:07 AM, Tobias Schmidt <[email protected]> wrote: > OK, in my understanding the optimization should be executable with a > gradient-based algorithm if I set *grad* to a constant value instead of a > calculated one. I don't think that most of the algorithms will converge with an incorrect gradient. Certainly the convergence proofs assume a correct gradient. What is the point of using a gradient-based algorithm if you don't supply the gradient? _______________________________________________ NLopt-discuss mailing list [email protected] http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/nlopt-discuss _______________________________________________ NLopt-discuss mailing list [email protected] http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/nlopt-discuss
