Where is `OrdinaryDiffEq.ode_interpolation`'s output type defined?

cadojo · July 18, 2021, 3:27am

I’ve defined my own subtype of DenseArray called CartesianState. The source code is available here. The following functionality works…

state = CartesianState(randn(6))
similar(state)
SymbolicUtils.Code.create_array(::Type{<:CartesianState}, ...)
# others

I’d like to use my custom array as a type for DifferentialEquations solvers. I can use my new array type to make an ODEProblem, and solving the problem returns a valid ODESolution.

The ODESolution u field is a Vector of CartesianState values – great! But when I interpolate between time steps via (::ODESolution)(t), I get a Vector{Float64} back.

I’ve traced this back to OrdinaryDiffEq.ode_interpolation, but I can’t figure out where that function’s output type is determined. Does anyone know offhand?

ChrisRackauckas · July 18, 2021, 9:56am

It should just be defined as the result of a broadcast operation:

github.com

SciML/OrdinaryDiffEq.jl/blob/master/src/dense/generic_dense.jl#L484


      
                          cache.caches[id.alg_choice[i₊]], idxs, deriv)
                  else
                      _ode_addsteps!(ks[i₊], ts[i₋], timeseries[i₋], timeseries[i₊], dt, f, p,
                          cache) # update the kcurrent
                      val = ode_interpolant(Θ, dt, timeseries[i₋], timeseries[i₊], ks[i₊], cache,
                          idxs, deriv)
                  end
              end
          
              val
          end
          
          """
          ode_interpolation!(out,tval::Number,ts,timeseries,ks)
          
          Get the value at tval where the solution is known at the
          times ts (sorted), with values timeseries and derivatives ks
          """
          function ode_interpolation!(out, tval::Number, id::I, idxs, deriv::D, p,
              continuity::Symbol = :left) where {I, D}
              @unpack ts, timeseries, ks, f, cache = id