Heterogeneous units and ModelingToolkit

I am trying to use the ModelingToolkit together with Unitful to model a coupled system of two ODEs with different units. The following code fails on ODEProblem with a DimensionError unless both units in u0 are the same:

using ModelingToolkit
using DifferentialEquations
using Plots
using Unitful
@parameters t k1 k2
@variables y(t) z(t)
@derivatives D'~t
eqs = [D(y) ~ k1*y, D(z) ~ k2*z]
de = ODESystem(eqs)
u0 = [y => 1.5u"V", z => 3.1u"N"]
p = [k1 => 1.0u"1/s", k2 => 1.0u"1/s"]
tspan = (0.0u"s",1e-5u"s")
prob = ODEProblem(de,u0,tspan, p, jac=false) # ERROR: DimensionError: V and 3.1 N are not dimensionally compatible.
sol = solve(prob, Euler(), dt=1e-6u"s")

I know that heterogeneous variables are an issue, as discussed here: DifferentialEquations.jl and systems with heterogenous units
The problem in that case was the adaptive stepping that can’t calculate a vector norm if the vector has heterogeneous units.
The suggested solution was to use ArrayPartition:
https://github.com/SciML/DifferentialEquations.jl/issues/147

It suspect that I am running into a similar problem, even though in my case the code fails at ODEProblem(), not on solve().

What is the best approach to use units in my case?

The solution here is to type the variables and generate the code as un-united. This isn’t documented yet but will be documented once the front end part of it is ready.

Thanks, Chris!

Did you get this working, if so, would you post your code?

I didn’t get it to work, but I haven’t checked the latest versions. At the time, it still needed some changes to ModelingToolkit.

1 Like

We still want to finish the union of SymbolicUtils and MTK first before heading back and making an interface to the unit parts (which exist, but are buried)

1 Like