Value of the constraints at the solution

[joseph-gergaud]

We want to solve an optimization problem with contraints and we would like for diagnostic to have the value of the constraints at the solution. Is it possible ?

[tmigot]

The returned structure by all the JuliaSmoothOptimizers is a GenericExecutionStats, see documentation here SolverCore.GenericExecutionStats.

The particular field you might be interested in are:

- `dual_feas`: The dual feasibility norm at solution (default: Inf);
- `primal_feas`: The primal feasibility norm at solution (default: 0.0 if unconstrained, Inf otherwise);

[joseph-gergaud]

Yes, I have seen this, but we don't have de value of the constraints (Ipopt return these values)

[abelsiqueira]

You should be able to evaluate the constraints value at the solution with something like

using NLPModels
cx = cons(nlp, output.solution)

The NLPModels defines the API, cons is the function that evaluates constraints, nlp is the problem and output is the object returned by the solver.

[joseph-gergaud]

When I execute the following program

using NLPModelsIpopt, ADNLPModels
nlp = ADNLPModel(x -> sum(x.^2), ones(3));

f(x) = 0.5*((x[1]-1)^2 + x[2]^2)
c(x) = [-x[1]+x[2]^2]
x0 = [1.;1.]
println("f(x0) = ", f(x0))
nlp = ADNLPModel(f, x0, c, [0.], [0.])

output = ipopt(nlp, print_level = 5)


println("solution = ", output.solution)

println("cons(nlp, output.solution) = ", cons(nlp, output.solution))

I obtain :

EXIT: Optimal Solution Found.
solution = [0.5000000001101497, 0.7071067813550226]
ERROR: LoadError: UndefVarError: cons not defined
Stacktrace:
[1] top-level scope
@ ~/control-toolbox/CTDirect.jl/work/toto.jl:15
[2] include(fname::String)
@ Base.MainInclude ./client.jl:476
[3] top-level scope
@ REPL[6]:1
in expression starting at /Users/gergaud/control-toolbox/CTDirect.jl/work/toto.jl:15

And if write :

println("ADNLPModels.cons(nlp, output.solution) = ", ADNLPModels.cons(nlp, output.solution))

It's ok I obtain

EXIT: Optimal Solution Found.
solution = [0.5000000001101497, 0.7071067813550226]
ADNLPModels.cons(nlp, output.solution) = [1.2811007810142883e-10]

I think you have to export the fonction cons

Thanks,

[abelsiqueira]

Yes, the cons function is exported only in NLPModels, so you would have to using NLPModels to access it or qualify it as you have done.
Either way, I'm glad it works. Let us know if you need more help 👍

[joseph-gergaud]

Thanks a lot. We can close the discussion