MTK system based on symbolic arrays fails: type Array has no field lhs

Specific Domains Modelling & Simulations
1

I struggle a little bit with a relatively simple MTK model that contains matrix vector multiplications. My example code is the following:

using ModelingToolkit

Am = [1 0 3; 2 0 1; 0 1 4]
Bm = [4 5; 3 2; 1 4]

@parameters t
Dt = Differential(t)

@assert size(Am, 1) == size(Bm, 1)
nr_of_components = size(Am, 1) # 3
nr_of_incoming_feeds = size(Am, 2) # 3
nr_of_independent_rates = size(Bm, 2) # 2

@parameters begin
    A[1:nr_of_components, 1:nr_of_incoming_feeds] # [3 x 3] matrix
    B[1:nr_of_components, 1:nr_of_independent_rates] # [3 x 2] matrix
end

@variables begin
    u_1(t)[1:nr_of_incoming_feeds] # [2 x 1] vector
    c(t)[1:nr_of_components] # [3 x 1] vector
    q_i(t)[1:nr_of_independent_rates] # [2 x 1] vector
end

eqns = [
    Dt.(c) ~ A * u_1 + B * q_i * c[1] 
]
# This gives:
# 1-element Vector{Symbolics.Arr{Any, 1}}:
# (broadcast(~, broadcast(Differential(t), c(t)), broadcast(+, A*u_1(t), B*broadcast(*, q_i(t), Ref((c(t))[1])))))[Base.OneTo(3)]

@named mdl = ODESystem(eqns)

With this I get the error:

ERROR: type Array has no field lhs
Stacktrace:
 [1] getproperty(x::Vector{Equation}, f::Symbol)
   @ Base ./Base.jl:37
 [2] ODESystem(eqs::Vector{Symbolics.Arr{Any, 1}}, iv::Num; kwargs::@Kwargs{name::String})
   @ ModelingToolkit ~/.julia/packages/ModelingToolkit/Gpzyo/src/systems/diffeqs/odesystem.jl:255
 [3] top-level scope
   @ ~/.julia/packages/ModelingToolkit/Gpzyo/src/systems/abstractsystem.jl:1127

Does somebody knows what this error means and how to fix it? The dimensions of the matrices and vectors should be correct.
Is it possible to solve this system using symbolic arrays?

Thanks in advance!

2

See the updated docs on symbolic arrays:

For now I highly recommend the array of structs form though in the near future I will be working on making the non-scalarized form much better.

3

Hey @ChrisRackauckas, thanks again for the advice. Just wanted to point out that these types of vectorized models now work perfectly fine after the v9 update. Keep up the great work, highly appreciate it!