reinit! now ignores given state (2nd argument) for an MTK-derived ODEIntegrator.

[Datseris]

Minimal Reproducible Example 👇

using ModelingToolkit, OrdinaryDiffEqDefault

@independent_variables t
D = Differential(t)
@mtkmodel Roessler begin
    @parameters begin
        a = 0.2
        b = 0.2
        c = 5.7
    end
    @variables begin
        x(t) = 1.0
        y(t) = 0.0
        z(t) = 0.0
        nlt(t) # nonlinear term
    end
    @equations begin
        D(x) ~ -y -z
        D(y) ~ x + a*y
        D(z) ~ b + nlt
        nlt ~ z*(x - c)
    end
end

@mtkbuild roessler_model = Roessler()

prob = ODEProblem(roessler_model)
integ = OrdinaryDiffEqDefault.init(prob)
reinit!(integ, SVector(5, 5, 0.0))
integ

julia> integ
t: 0.0
u: 3-element SVector{3, Float64} with indices SOneTo(3):
 1.0
 0.0
 0.0

There are no error messages. The reinit! function simply now ignores the second argument and does not re-initialize the integrator at the given state. This is breaking for DynamicalSystems.jl, so I hope there is an easy fix!

Disclaimer: I have not checked whether this problem persists even for normally created ODEProblems.

• Output of using Pkg; Pkg.status()

 pkg> st -m ModelingToolkit OrdinaryDiffEqDefault
Status `C:\Users\datse\.julia\environments\NLD\Manifest.toml`
  [961ee093] ModelingToolkit v9.65.0
  [50262376] OrdinaryDiffEqDefault v1.3.0

• Output of versioninfo()

1.10.8.

Add any other context about the problem here.

[AayushSabharwal]

reinit! reruns initialization as well, thus overriding the state. Passing reinit_dae = false will disable this behavior and respect the provided u0.

[ChrisRackauckas]

We should at least warn if someone does a numerical reinitialization on a problem with a symbolic reinit function.

[Datseris]

reinit! reruns initialization as well, thus overriding the state. Passing reinit_dae = false will disable this behavior and respect the provided u0.

Why isn't the reinit_dae = false the default? If it was, this wouldn't be a breaking change, as this behavior did not occur in other MTK v9 versions. Please do add the warning, yes.

[Datseris]

Furthermore the docstring of reinit! should be changed to reflect that in fact the 2nd position argument will not be the new state.

[AayushSabharwal]

I agree with the warning. Initialization is there because the user might be solving for some parameters from unknowns, or the provided u may not be consistent.

[ChrisRackauckas]

Why isn't the reinit_dae = false the default?

Because then it's almost guaranteed to not work. It has always been true because you can't solve a DAE without it?

[Datseris]

"It has always been like that" is not my experience here. Can you acknowledge that this is breaking? I am not creating a DAE above, I am creating an ODE. The same code that used to work does not.

[ChrisRackauckas]

It has always done that? It was using BrownBasicInit.