update to MTKv10

[baggepinnen]

• fix: ensure => nothing overrides defaults SciML/ModelingToolkit.jl#3714
• fix: fix namespacing of variables of variables SciML/ModelingToolkit.jl#3711
• failure in update to MTKv10 #76 (comment)

[baggepinnen]

@AayushSabharwal

A docs example fails with the error below, it appears to be related to symbolic arrays?

julia> prob = ODEProblem(structural_simplify(closed_loop), [F.xd => 0], (0.0, 8.0))
ERROR: ArgumentError: (Cgs₊A(Cgs₊v(t)))[1, 2] is present in the system but (Cgs₊A(Cgs₊v(t)))[1, 2] is not an unknown.
Stacktrace:
 [1] TearingState(sys::System; quick_cancel::Bool, check::Bool, sort_eqs::Bool)
   @ ModelingToolkit ~/.julia/packages/ModelingToolkit/v0QS9/src/systems/systemstructure.jl:376
 [2] __mtkcompile(sys::System; simplify::Bool, inputs::Vector{…}, outputs::Vector{…}, disturbance_inputs::Vector{…}, sort_eqs::Bool, kwargs::@Kwargs{…})
   @ ModelingToolkit ~/.julia/packages/ModelingToolkit/v0QS9/src/systems/systems.jl:79
 [3] __mtkcompile
   @ ~/.julia/packages/ModelingToolkit/v0QS9/src/systems/systems.jl:62 [inlined]
 [4] mtkcompile(sys::System; additional_passes::Vector{…}, simplify::Bool, split::Bool, allow_symbolic::Bool, allow_parameter::Bool, conservative::Bool, fully_determined::Bool, inputs::Vector{…}, outputs::Vector{…}, disturbance_inputs::Vector{…}, kwargs::@Kwargs{})
   @ ModelingToolkit ~/.julia/packages/ModelingToolkit/v0QS9/src/systems/systems.jl:37
 [5] mtkcompile
   @ ~/.julia/packages/ModelingToolkit/v0QS9/src/systems/systems.jl:30 [inlined]
 [6] structural_simplify(sys::System; kwargs::@Kwargs{})
   @ ModelingToolkit ./deprecated.jl:105
 [7] structural_simplify(sys::System)
   @ ModelingToolkit ./deprecated.jl:103
 [8] top-level scope
   @ REPL[29]:1
Some type information was truncated. Use `show(err)` to see complete types.

The unknowns do appear to contain this strange variable though

Unknowns (37): see unknowns(closed_loop)
  duffing₊x(t) [defaults to 0]
  duffing₊v(t) [defaults to 0]
  duffing₊u₊u(t): Inner variable in RealInput u
  duffing₊y₊u(t): Inner variable in RealOutput y
  (Cgs₊x(t))[1] [defaults to 0.0]: State variables of gain-scheduled statespace system Cgs
  (Cgs₊x(t))[2] [defaults to 0.0]: State variables of gain-scheduled statespace system Cgs
  (Cgs₊x(t))[3] [defaults to 0.0]: State variables of gain-scheduled statespace system Cgs
  Cgs₊v(t): Scheduling variable of gain-scheduled statespace system Cgs
  Cgs₊scheduling_input₊u(t): Inner variable in RealInput scheduling_input
  (Cgs₊A(v(t)))[1, 1] [defaults to -141.421]
  (Cgs₊A(v(t)))[2, 1] [defaults to 128.0]
  (Cgs₊A(v(t)))[3, 1] [defaults to 0.0]
  (Cgs₊A(v(t)))[1, 2] [defaults to -78.125]
  (Cgs₊A(v(t)))[2, 2] [defaults to 0.0]
  (Cgs₊A(v(t)))[3, 2] [defaults to 4.0]
  (Cgs₊A(v(t)))[1, 3] [defaults to 0.0]
  (Cgs₊A(v(t)))[2, 3] [defaults to 0.0]
  (Cgs₊A(v(t)))[3, 3] [defaults to 0.0]
  (Cgs₊B(v(t)))[1, 1] [defaults to 128.0]
  (Cgs₊B(v(t)))[2, 1] [defaults to 0.0]
  (Cgs₊B(v(t)))[3, 1] [defaults to 0.0]
  (Cgs₊C(v(t)))[1, 1] [defaults to 181.953]
  (Cgs₊C(v(t)))[1, 2] [defaults to 6.47795]

[AayushSabharwal]

Is that meant to be a variable of the system?

[baggepinnen]

Yes, they are defined like this

@variables v(t) [
        description = "Scheduling variable of gain-scheduled statespace system $name",
] 
@variables A(v)[1:nx, 1:nx]

v is a function of t while A is a function of v. This worked as I expected in MTKv9, but MTKv10 does not like it

[AayushSabharwal]

It's a namespacing bug. The unknown is (Cgs₊A(v(t)))[1, 2] but in the namespaced equation it shows up as (Cgs₊A(Cgs₊v(t)))[1, 2]

[baggepinnen]

On MTK master, initialization now fails in the batch_linearization.md file with

julia> controllersy, ssy, ops3, resolved_ops3 = trajectory_ss(closed_loop, closed_loop.r, closed_loop.u, sol; t=timepoints, verbose=true, loop_openings=[closed_loop.y]);
┌ Warning: Initialization system is overdetermined. 1 equations for 0 unknowns. Initialization will default to using least squares. `SCCNonlinearProblem` can only be used for initialization of fully determined systems and hence will not be used here. To suppress this warning pass warn_initialize_determined = false. To make this warning into an error, pass fully_determined = true
└ @ ModelingToolkit ~/.julia/dev/ModelingToolkit/src/problems/initializationproblem.jl:106
ERROR: Initialization algorithm SciMLBase.OverrideInit failed with  ...
Stacktrace:
  [1] error(s::String)
    @ Base ./error.jl:35
  [2] (::ModelingToolkit.LinearizationFunction{…})(u::Vector{…}, p::MTKParameters{…}, t::Float64)
    @ ModelingToolkit ~/.julia/dev/ModelingToolkit/src/linearization.jl:305
  [3] solve(prob::LinearizationProblem{ModelingToolkit.LinearizationFunction{…}, Float64}; allow_input_derivatives::Bool)
    @ ModelingToolkit ~/.julia/dev/ModelingToolkit/src/linearization.jl:442

From Aayush:

Okay so the issue is that simplifiction artificially inserts variable ~ 0 equations for the initialization, essentially SciML/ModelingToolkit.jl#3302 The TL;DR is that when running alias_elimination! MTK somehow adds those equations for the system and I’m yet to figure out why.