Constriant in ADNLPModels package

Specific Domains Optimization (Mathematical)
,
1

Hi everyone,

I am trying to do a nonlinear optimization with six parameters using the ADNLModles package. For example:

function constriants(para)
    cons=1-para[1]^2-para[2]^2-para[3]^2+2*para[1]*para[2]*para[3]
    return([para[1];para[2];para[3];cons])
end

lb=[-0.999;-0.999;-0.999;0.001]
ub=[0.999;0.999;0.999;Inf]

function obj(para)
         println(para)
          value
return (value)

opt=ADNLPModel(obj,para,constriants,lb,ub)

ipopt(opt)

And I printed out every parameter in the optimization routine, and I found that sometimes the searching area was outside the constraints, and the optimization crashed.

I was wondering if that’s a bug in the package or if there is another optimization package I could use for nonlinear constraints. I would appreciate any advice on making this function work.

Thanks！

2

Welcome. I’ve moved your question to the Optimization category since it may be more easily seen by someone who can answer it.

Since it sounds like you are solving a small problem, can you post your actual code?

3

Hi, answering from the phone.

Ipopt is responsible here, it doesn’t guarantee non-bounds constraints are satisfied in intermediate steps. See https://stackoverflow.com/a/39125549
This question was asked kn the past, you can check for more information.
One thing to try is the Percival.jl solver. It is also compliant with adnlpmodels.

4

Thanks for moving my post! I updated the constraints part in my code (my objective function is too long).

5

Thanks! It’s exactly my problem. I will try Percival solver.