Jacobian of function that solves an ODE internally

[JMmontilla]

I am trying to compute the jacobian of a function that solves an Ode internally (using DiffEquations.jl), the function has the following structure:

function fConstraint(x::Vector)::Vector sol = calc_data(x) #ode solving #println(x) res = [ sol[1][2]-data["hgoal"]*1e3/datos_prob["r12"], -sol[1][2], sol[2][2]+data["tspan"][2]*0.99, -sol[1][1]+data["hmmin"]*1e3/datos_prob["r12"] ] end

The function is used to compute the restrictions in an optimization problem, and I am using a solver that requires the gradient of the constraints. Is there a way to do this with ForwardDiff? I ask because I get an enormous error code that starts like this:

ForwardDiff.jacobian(fConstraint, [0.001,0.1,0.1])
ERROR: MethodError: no method matching Float64(::ForwardDiff.Dual{ForwardDiff.Tag{typeof(fConstraint),Float64},Float64,3})
Closest candidates are:
Float64(::Real, ::RoundingMode) where T<:AbstractFloat at rounding.jl:194
Float64(::T<:Number) where T<:Number at boot.jl:741
Float64(::Int8) at float.jl:60
...

, and I guess it is because of the internal ode, otherwise it is a very simple function and it computes without problems (I have tried with another much simpler function that only uses sums and it computes the jacobian well).

Thanks in advance :)

[JMmontilla]

The code doen't look well sorry, it is more like:

function fConstraint(x::Vector)::Vector
sol = calc_data(x)
#println(x)
res = [ sol[1][2]-data["hgoal"]*1e3/datos_prob["r12"],
-sol[1][2],
sol[2][2]+data["tspan"][2]*0.99,
-sol[1][1]+data["hmmin"]*1e3/datos_prob["r12"] ]
end

with correct indentation (the function itself works fine)

[ChrisRackauckas]

I have no idea what you're asking. Do you have code that can be ran?

[JMmontilla]

This code shows what I was trying to explain:
testJacobian.jl.zip
Thank you very much for taking interest¡

[ChrisRackauckas]

You're trying to autodiff a code with a continuous callback, which means that time points at which things occur is differentiable too, and you have to make sure that time is Dual in that case. You do this by making tspan be convert.(eltype(x),data["tspan"]). This fixes the errors.

[JMmontilla]

You're trying to autodiff a code with a continuous callback, which means that time points at which things occur is differentiable too, and you have to make sure that time is Dual in that case. You do this by making tspan be convert.(eltype(x),data["tspan"]). This fixes the errors.

Thank you very much!. I would have never figured that out by myself. Does the change have any effect on performance (ODE-wise)?

[ChrisRackauckas]

Just compile time.