Flow with variables (NonFixed ocp)

[jbcaillau]

@ocots the function returned by Flow applied to an ocp should depend on the variable (when there is one, i.e. when the ocp is NonFixed) For instance, in this case (where tf is variable).

[ocots]

@jbcaillau Actually, you can always give or not an extra variable to the function returned by Flow:

https://github.com/control-toolbox/CTFlows.jl/blob/8999240c9e813a82e6381a71720c5760a3d68c55/src/hamiltonian.jl#L45

If the problem is Fixed then the variable is not used (this case should be unauthorized). If the problem is NonFixed then the variable is used. In the case of the Goddard problem, there is a variable (the final time tf) that we can pass to the function returned by Flow (or not). Note that this variable is not used to compute neither the dynamics/cost nor the feedback control hence we are not compelled to pass it. In this case the variable takes the default value which is Real[]. I understand that this can be weird but it would be strange to have to write something like that:

xf, pf = f0(t3, x3, p3, tf, tf)

There is a test where you can see that we can provide a variable (I need to add tests):

https://github.com/control-toolbox/CTFlows.jl/blob/8999240c9e813a82e6381a71720c5760a3d68c55/test/test_flow_hamiltonian.jl#L22

[jbcaillau]

Thanks @ocots. Exactly. Two points below.

Actually, you can always give or not an extra variable to the function returned by Flow:

https://github.com/control-toolbox/CTFlows.jl/blob/8999240c9e813a82e6381a71720c5760a3d68c55/src/hamiltonian.jl#L45

If the problem is Fixed then the variable is not used (this case should be unauthorized).

Point 1. Yes, when the problem is Fixed, calling a flow with a variable should not be possible.

If the problem is NonFixed then the variable is used. In the case of the Goddard problem, there is a variable (the final time tf) that we can pass to the function returned by Flow (or not). Note that this variable is not use to compute neither the dynamics/cost nor the feedback control hence we are not compelled to pass it. In this case the variable takes the default value which is Real[]. I understand that this can be weird but it would be strange to have to write something like that:

xf, pf = f0(t3, x3, p3, tf, tf)

Point 2. From the user point of view, the two particular cases when either t0 or tf is the only variable, it is strange to force the user to write such a thing. So it is good to give the user the possibility not to add the variable t0 or tf during the call in these two specific cases (dim $1$ variable equal to either t0 or tf). This can easily be taken care of with a check in __variable helper. The typical situation being the free final time case. Conversely, I would not advocate talking care of the case when the variable is a dim $2$ one and equals [ t0, tf ].

There is a test where you can see that we can provide a variable (I need to add tests):

https://github.com/control-toolbox/CTFlows.jl/blob/8999240c9e813a82e6381a71720c5760a3d68c55/test/test_flow_hamiltonian.jl#L22

• the standard behaviour is (i) to forbid (= throw an exception) passing a variable to a Fixed problem, (ii) to force passing a variable to a NonFixed = Variable problem
• two exceptions to (ii): for a dim $1$ variable equal either to t0 or tf, allow not to give t0 or tf when calling a flow
• note that when t0 or tf is the (only) variable, passing Real[] as the default when the variable is omitted does not work since the computation of the flow might actually depend on the variable (tf in the dynamics, for instance...) so the value (of t0 or tf) must be retrieved by __variable

Side notes:

• add alias Variable = NonFixed, keeping NonFixed for compatibility and homogeneity with NonAutonomous. Conversely, I do not think relevant to add the alias NonVariable = Fixed...
• being Autonomous and / or Variable are traits and should be implemented as such (to be done)

[ocots]

@jbcaillau top priority.

[jbcaillau]

@ocots agreed. I thought you would do it, but I can take care of it: let me know. In any case, I first need to smoothen some things in https://github.com/control-toolbox/CTBase.jl/issues/166#issuecomment-2183913880

[ocots]

Actually I forgot. But I can take care of it.

[jbcaillau]

Actually I forgot. But I can take care of it.

maybe you can. maybe not. 🤭

[ocots]

@jbcaillau Still forgotten.

[jbcaillau]

@ocots no worry. fine as is, but we ought to take care of it these days.