Is it possible to have ComponentArray with differenty types?

[paolo-mgi]

I would like to integrate a system of differential equations in which the evolution of a real component depends upon an indicator of the remaining complex components.

test_u0 = ComponentArray(v=1.0 + 0.0*im, 0.0+im*0.0], m= 1.0)
function test_psi!(du, u, p, t)
du.v = -im * u.v
du.m = u.m * real(dot(u.v,u.v))
end
test_prob = ODEProblem(test_psi!, test_u0, (0.0, tf), p)
test_sol=solve(test_prob, Vern7())

Is it possible to use ComponentArray so that u.m is Float64 at every step? The code above returns

`ComponentVector{ComplexF64}(v = ComplexF64[0.9803381875004509 - 0.19732470228080312im, 0.0 + 0.0im], m = ComplexF64[1.2197284031519102 + 0.0im])'

I understand that ArrayPartition

psi0= [1.0 + 0.0im, 0.0 + 0.0im]
ArrayPartition(psi0,1.0)

yields the desired output but I do not understand how to pass this output to an ODEProblem
Many thanks

[jonniedie]

You could technically do this by doing

test_u0 = ComponentArray{Union{Float64, ComplexF64}(v=[1.0 + 0.0*im, 0.0+im*0.0], m= 1.0)

But it’s probably going to be slightly slower than keeping everything complex and converting m to a real number when you unpack it. There’s unfortunately not much I can do on the ComponentArrays side to make heterogeneous arrays faster when the solvers need to blindly iterate over the elements. But keeping the Union to only two options should make things reasonably fast.

[paolo-mgi]

Please bear with me but I thik I do not understand the syntax correctly. If I enter

test_u0 = ComponentArray{Union{ComplexF64,Float64}}(v=[1.0 + 0.0*im, 0.0+im*0.0], m= 1.0)

I get

ComponentVector{Union{Float64, ComplexF64}}(v = Union{Float64, ComplexF64}[1.0 + 0.0im, 0.0 + 0.0im], m = 1.0 + 0.0im)

which does not look to me the desired result. Many thanks!

[jonniedie]

Interesting. The solver must be doing something to promote the types then. I don’t think there’s much we can do about that. It looks like ComponentArrays is converting that type. But either way, is there any reason you can’t use a complex number for m and just convert it to real whenever you want to use it? That will be the most performant option.

[paolo-mgi]

I can definitely work with complex, as you suggest. I was just curious. ComponentArrays allowed me to finally get a quantum trajectory code to work properly :). I wanted only to further test its functionalyity. Many thanks for your work!