aeroimperial-optimization / mpyalmip Goto Github PK

Hello everyone!

I am looking through Van Der Pol example and try to increase the value of degPvals up to 12 as was done in the arXiv paper. I have several questions:

1. For the set of parameters specified in the example, the solver gives a value U=5.82, but returns a code pdINF, which means that the solution is expected to be infeasible. How can we trust such a result?
2. For the degPvals=10 I found a strange behavior. For most of the runs the solver founds only the trivial result U=0, but for about 10% of runs it founds U=5.09 (with the same starting parameters). So, the output of the solver is stochastic. If we change parameters opts.sdpa_gmp.epsilonDash = 1.0e-20 and opts.sdpa_gmp.precision = 150 then the result becomes stable, but the return code is still pdINF.
3. For degPvals=12 , opts.sdpa_gmp.epsilonDash = 1.0e-20, opts.sdpa_gmp.precision = 150 the matlab stops working after the solver run.

Is this all expected behavior or may be there are some problems related to the fact that I use sdpa_gmp on Windows? If everything is OK, so what is the trick? How we can get robust estimates for degPvals=12? @giofantuzzi could you comment, please?

Do you have any plans to make mpYALMIP available under WINDOWS? Would a WINDOWS version entail any inherent difficulties vs. LINUX? Would you expect it to work under MATLAB R2014A (YALMIP does)?

Thanks.

What's the greatest number of digits you have tried under mpYALMIP or in any other version?

Have you thought about modifying SDPA-GMP to use arbitrary precision outward rounded interval or radial arithmetic, so that rock-solid reliable calculations could be performed in a high enough precision to get rigorous, yet useful, results to difficult problems?