88 dev adding models for exa

[jbcaillau]

close #88

[ocots]

@jbcaillau I am working on #139

[jbcaillau]

@ocots what is this NaN in target xf for the steering use case?

[ocots]

@ocots what is this NaN in target xf for the steering use case?

Good catch. I don't know. Does it make sense to let the first component free at the final time. I have removed the NaN for the other models. Actually I think that Yassin wrote a NaN to have in JuMP:

@constraint(model, x2[N + 1] == xf[2])
@constraint(model, x3[N + 1] == xf[3])
@constraint(model, x4[N + 1] == xf[4])

instead of

@constraint(model, x2[N + 1] == xf[1])
@constraint(model, x3[N + 1] == xf[2])
@constraint(model, x4[N + 1] == xf[3])

[ocots]

@jbcaillau There is sometimes a difference between my local CI and on runners.

Here, there is a fail for the space shuttle problem: https://github.com/control-toolbox/OptimalControlProblems.jl/actions/runs/17739285652/job/50412586297?pr=137#step:4:2976

What is strange, is that there is a fail for the "quick" test but not for the "solution" test. Both make the same calls to the solvers with the same init and parameters.

The problem seems to come from:

OptimalControlProblems.jl/ext/OptimalControlModels/space_shuttle.jl

Line 163 in 5c46373

h_dot = v * sin(γ)

[jbcaillau]

@ocots nice job. a lot of things, bravo 👍🏽

[jbcaillau]

@ocots i see the call here, don't see anything special

[jbcaillau]

@ocots important: be sure to use Mumps for the linear solver on CPU, both for Ipopt / MadNLP and adnlp / exa.

[ocots]

@ocots important: be sure to use Mumps for the linear solver on CPU, both for Ipopt / MadNLP and adnlp / exa.

This is the case.

[ocots]

@jbcaillau We will need to add some documentation about the _s problems. There is no mention of them for the moment.