[Ipopt] enforcing monotone decrease of mu

Martin Neuenhofen martinneuenhofen at googlemail.com
Mon Feb 6 06:28:26 EST 2017

Attendum: Did anyone else observe that ipopt v 3.12.3 in Matlab even with
the options

options.ipopt.mu_strategy = 'adaptive';+
options.ipopt.mu_max = 1e-1

uses values for mu that are 10^0.3 ?
And, indeed, this spoils my overall convergence because for small mu-s it
occurs that Ipopt pointlessly improves performance whilst searching in a
completely infeasible area.

Is there an option to enforce a maximum value for mu without any exception
that Ipopt can exploit to still choose it larger?

Kind regards

2017-02-06 10:08 GMT+00:00 Martin Neuenhofen <
martinneuenhofen at googlemail.com>:

> Hi all,
> I am using ipopt with the option.ipopt.mu_strategy = 'monotone' but from
> by iteration count I observe that lg(mu) does increase: It starts at -1 and
> after some iterations becomes -0.0 . Can one enforce a truly monotone
> behaviour? Setting the upper limit to -1 works for this case but may it
> happen that the solver first decreases in a sequence like  -1, -2, -3 and
> then comes back to -2 or can this only happen because the solver tries to
> find a suitable initial mu?
> I am thankful for Professor Miller's feedback to my questions on Ipopt's
> convergence measures. It would be of great help for me if someone could
> come back to my latter questions.
> Kind regards
> Martin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20170206/859beaa7/attachment.html>

More information about the Ipopt mailing list