Good convergence to solution, but overly slow decrease in objective function #562
-
Are there options that will make IPOPT less timid when moving into an infeasible region to find better objective values? I'm seeing IPOPT converge to a good solution that I know to be correct, and progress toward the solution is very steady throughout the solve. That being said, the progress is slower than I'd expect and the number of iterations seems high. For context, the problem being solved is a highly sparse optimal control problem. Output is below. Any ideas/suggestions are welcome!
|
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 2 replies
-
There are plenty of Ipopt options one could play with: https://coin-or.github.io/Ipopt/OPTIONS.html Maybe it is the nonconvexity of the problem (lg(rg) not '-') that prevents Ipopt from doing fast progress. A better starting point might help. Updating to a current Ipopt version or changing the linear solver will probably not help, but who knows. |
Beta Was this translation helpful? Give feedback.
There are plenty of Ipopt options one could play with: https://coin-or.github.io/Ipopt/OPTIONS.html
You may want to avoid the ones marked as "advanced". mu_strategy, nlp_scaling_method, and other options related to step computation and line search could be interesting.
Maybe it is the nonconvexity of the problem (lg(rg) not '-') that prevents Ipopt from doing fast progress. A better starting point might help.
Updating to a current Ipopt version or changing the linear solver will probably not help, but who knows.