Closed svigerske closed 5 years ago
Comment by @svigerske created at 2017-09-15 14:33:22
This looks likes something is rather missing in the CGPenalty implementation. So try not to use this option.
Maybe you can come up with a better starting point, if the default one leads to Ipopt concluding infeasibility.
Issue created by migration from Trac.
Original creator: kamilova
Original creation time: 2017-09-15 10:08:13
Assignee: ipopt-team
Version: 3.12
Dear all,
I am struggling with my Ipopt implementation. At this point I am confident that I am supplying the objective function, gradient, constraint function and Jacobian correctly. I do not have any second order information so I use a Hessian approximation with Ipopt. As it turns out, my problem would always go into restoration phase and then I would get the output
EXIT: Converged to a point of local infeasibility. Problem may be infeasible.
I started playing with the options, and set:
hessian_approximation limited-memory hessian_approximation_space all-variables slack_bound_frac 0.5 slack_bound_push 0.5 limited_memory_update_type sr1 nlp_scaling_method gradient-based expect_infeasible_problem yes mu_strategy adaptive derivative_test first-order corrector_type affine
These seem to improve the optimisation behaviour. I wanted to try a different line search method, since in a previous optimisation implementation (of the same problem) I had a lot of issues with the penalty function used which lead my points to not make any progress due to the tiniest stepsizes. I wanted to activate cg-penalty, and the optimisation seems to work a lot better, but is terminated early, with the error message:
Exception of type: IpoptException in file "IpBacktrackingLineSearch.cpp" at line 545: Exception message: No Restoration Phase given to this Backtracking Line Search Object!
EXIT: Some uncaught Ipopt exception encountered.
An error occurred. The error code is -100
I fear that it might be one of the other options that I have changed, or maybe I didn't add something I should have.
Any advice on the matter would be greatly appreciated.
Thank you for your time,
Alissa