Model Validation and Units

ModelingToolkit.jl provides extensive functionality for model validation and unit checking. This is done by providing metadata to the variable types and then running the validation functions which identify malformed systems and non-physical equations. This approach provides high performance and compatibility with numerical solvers.

Assigning Units

Units may assigned with the following syntax.

using ModelingToolkit, Unitful
@variables t [unit = u"s"] x(t) [unit = u"m"] g(t) w(t) [unit = "Hz"]

@variables(t, [unit = u"s"], x(t), [unit = u"m"], g(t), w(t), [unit = "Hz"])

@variables(begin
t, [unit = u"s"],
x(t), [unit = u"m"],
g(t),
w(t), [unit = "Hz"]
end)

# Simultaneously set default value (use plain numbers, not quantities)
@variable x=10 [unit = u"m"]

# Symbolic array: unit applies to all elements
@variable x[1:3] [unit = u"m"]

Do not use quantities such as 1u"s", 1/u"s" or u"1/s" as these will result in errors; instead use u"s", u"s^-1", or u"s"^-1.

Unit Validation & Inspection

Unit validation of equations happens automatically when creating a system. However, for debugging purposes one may wish to validate the equations directly using validate.

Inside, validate uses get_unit, which may be directly applied to any term. Note that validate will not throw an error in the event of incompatible units, but get_unit will. If you would rather receive a warning instead of an error, use safe_get_unit which will yield nothing in the event of an error. Unit agreement is tested with ModelingToolkit.equivalent(u1,u2).

Example usage below. Note that ModelingToolkit does not force unit conversions to preferred units in the event of nonstandard combinations – it merely checks that the equations are consistent.

using ModelingToolkit, Unitful
@parameters τ [unit = u"ms"]
@variables t [unit = u"ms"] E(t) [unit = u"kJ"] P(t) [unit = u"MW"]
D = Differential(t)
eqs = eqs = [D(E) ~ P - E/τ,
                0 ~ P       ]
ModelingToolkit.validate(eqs) #Returns true
ModelingToolkit.validate(eqs[1]) #Returns true
ModelingToolkit.get_unit(eqs[1].rhs) #Returns u"kJ ms^-1"

An example of an inconsistent system: at present, ModelingToolkit requires that the units of all terms in an equation or sum to be equal-valued (ModelingToolkit.equivalent(u1,u2)), rather that simply dimensionally consistent. In the future, the validation stage may be upgraded to support the insertion of conversion factors into the equations.

using ModelingToolkit, Unitful
@parameters τ [unit = u"ms"]
@variables t [unit = u"ms"] E(t) [unit = u"J"] P(t) [unit = u"MW"]
D = Differential(t)
eqs = eqs = [D(E) ~ P - E/τ,
                0 ~ P       ]
ModelingToolkit.validate(eqs) #Returns false while displaying a warning message

User-Defined Registered Functions and Types

In order to validate user-defined types and registered functions, specialize get_unit. Single-parameter calls to get_unit expect an object type, while two-parameter calls expect a function type as the first argument, and a vector of arguments as the second argument.

using ModelingToolkit
# Composite type parameter in registered function
@parameters t
D = Differential(t)
struct NewType
    f
end
@register dummycomplex(complex::Num, scalar)
dummycomplex(complex, scalar) = complex.f - scalar

c = NewType(1)
MT.get_unit(x::NewType) = MT.get_unit(x.f)
function MT.get_unit(op::typeof(dummycomplex),args)
    argunits = MT.get_unit.(args)
    MT.get_unit(-,args)
end

sts = @variables a(t)=0 [unit = u"cm"]
ps = @parameters s=-1 [unit = u"cm"] c=c [unit = u"cm"]
eqs = [D(a) ~ dummycomplex(c, s);]
sys = ODESystem(eqs, t, [sts...;], [ps...;], name=:sys)
sys_simple = structural_simplify(sys)

Unitful Literals

In order for a function to work correctly during both validation & execution, the function must be unit-agnostic. That is, no unitful literals may be used. Any unitful quantity must either be a parameter or variable. For example, these equations will not validate successfully.

using ModelingToolkit, Unitful
@variables t [unit = u"ms"] E(t) [unit = u"J"] P(t) [unit = u"MW"]
D = Differential(t)
eqs = [D(E) ~ P - E/1u"ms"   ]
ModelingToolkit.validate(eqs) #Returns false while displaying a warning message

myfunc(E) = E/1u"ms"
eqs = [D(E) ~ P - myfunc(E) ]
ModelingToolkit.validate(eqs) #Returns false while displaying a warning message

Instead, they should be parameterized:

using ModelingToolkit, Unitful
@parameters τ [unit = u"ms"]
@variables t [unit = u"ms"] E(t) [unit = u"kJ"] P(t) [unit = u"MW"]
D = Differential(t)
eqs = [D(E) ~ P - E/τ]
ModelingToolkit.validate(eqs) #Returns true

myfunc(E,τ) = E/τ 
eqs = [D(E) ~ P - myfunc(E,τ)]
ModelingToolkit.validate(eqs) #Returns true

It is recommended not to circumvent unit validation by specializing user-defined functions on Unitful arguments vs. Numbers. This both fails to take advantage of validate for ensuring correctness, and may cause in errors in the future when ModelingToolkit is extended to support eliminating Unitful literals from functions.

Other Restrictions

Unitful provides non-scalar units such as dBm, °C, etc. At this time, ModelingToolkit only supports scalar quantities. Additionally, angular degrees (°) are not supported because trigonometric functions will treat plain numerical values as radians, which would lead systems validated using degrees to behave erroneously when being solved.

Troubleshooting & Gotchas

If a system fails to validate due to unit issues, at least one warning message will appear, including a line number as well as the unit types and expressions that were in conflict. Some system constructors re-order equations before the unit checking can be done, in which case the equation numbers may be inaccurate. The printed expression that the problem resides in is always correctly shown.

Symbolic exponents for unitful variables are supported (ex: P^γ in thermodynamics). However, this means that ModelingToolkit cannot reduce such expressions to Unitful.Unitlike subtypes at validation time because the exponent value is not available. In this case ModelingToolkit.get_unit is type-unstable, yielding a symbolic result, which can still be checked for symbolic equality with ModelingToolkit.equivalent.

Parameter & Initial Condition Values

Parameter and initial condition values are supplied to problem constructors as plain numbers, with the understanding that they have been converted to the appropriate units. This is done for simplicity of interfacing with optimization solvers. Some helper function for dealing with value maps:

remove_units(p::Dict) = Dict(k => Unitful.ustrip(ModelingToolkit.get_unit(k),v) for (k,v) in p)
add_units(p::Dict) = Dict(k => v*ModelingToolkit.get_unit(k) for (k,v) in p)

Recommended usage:

pars = @parameters τ [unit = u"ms"]
p = Dict(τ => 1u"ms")
ODEProblem(sys,remove_units(u0),tspan,remove_units(p))