JuMP Interface, Incorrect Status Returned

[ccoffrin]

There seems to be an issue when using Optim as a solver for JuMP in a constrained optimization setting.

If the algorithm terminates but does not satisfy the constraints it currently returns,

• termination_status of LOCALLY_SOLVED
• primal status of FEASIBLE_POINT

Per this guide, https://jump.dev/MathOptInterface.jl/stable/manual/solutions/#Local-search-solvers

In this situation it should probably return,

• termination_status of LOCALLY_INFEASIBLE
• primal status of INFEASIBLE_POINT

The issue can be reproduced by running this example, https://github.com/lanl-ansi/rosetta-opf/blob/main/debug/optim-jump.jl

CC @blegat

[pkofod]

Thanks for reporting it!

[blegat]

We are assuming that converged means LOCALLY_SOLVED:

Optim.jl/src/MOI_wrapper.jl

Lines 337 to 338 in 1a649e8

	elseif converged(model.results)
	return MOI.LOCALLY_SOLVED

Is this an incorrect assumption or should converged be false here ?

[ccoffrin]

I am surely not on expert in Optim but based on my brief experience using it, I suspect this function should look something like,

if converged, 
    check if the constraints are satisfied up to the prescribed tolerance
    if yes return `LOCALLY_SOLVED` otherwise return `LOCALLY_INFEASIBLE`

[pkofod]

Is it possible to give a reproducer here? Then it'd be a bit easier to look into.

[ccoffrin]

It's not a steam lined example but this code can be used to reproduce,

The issue can be reproduced by running this example, https://github.com/lanl-ansi/rosetta-opf/blob/main/debug/optim-jump.jl