Restoration Failed!

[crossfi...@gmail.com]

Hi,

I am working on implementing and solve a problem using collocation method. This is a NLP problem and I use the IPOPT-solver. With a small number of collocation points and integration steps I am able to solve the problem. However, the resolution is not good. When I try to increase the number of collocation points I get Restoration Failed! The only difference I do is going from 1 to 2 collocation points. Searching for a solution I could only relate this to the problem being infeasible. I know the problem is feasible since I just solve it with a fewer number of collocation points. Any clue what is going on? 

[Joel Andersson]

Hi!

What do you mean by increasing the number of collocation points? You mean increasing the order of the interpolating polynomial? If IPOPT fails with "Restoration Failed", it does not mean that the problem is infeasible, it means that IPOPT fails to find an infeasible starting point (I guess), which is not the same thing. I would have a look at the initial guess of the NLP. With a better initial guess, you might be able to avoid this. You can also try to first just simulate the system using IPOPT, that is fix the initial conditions (if they are not fixed already) and fix all controls and parameters.

Best regards,

Joel

[crossfi...@gmail.com]

I have been occupied for a while, but I thought I report back my solution to this problem. What solved it for me was to replace the default solver "mumps" with "ma27" obtained from http://www.hsl.rl.ac.uk/ipopt/ 

[Joel Andersson]

OK, yeah, makes sense. Mumps does not work very well. ma27 and the other HSL solvers are much better in general.

Joel

[Henrik Nord]

Hi! I also have som error with restoration. How did you change the default solver to ma27?

[Henrik Nord]

I try to write:

opts = struct;

opts.ipopt.linear_solver('ma27');

But get the error: Unrecognized field name "linear_solver".

[Joel Andersson]

Try

opts = struct

opts.ipopt.linear_solver = 'ma27';

Joel