Christopher Rackauckas

Tuples don't have vector semantics which is what causes the issue. Here's the amount of type piracy that you'd need to do in order to do this: ```julia using DifferentialEquations,...

However, to support stiff ODE solvers... you're asking for linear algebra on tuples. Not impossible, but 🤷‍♂ . Do you have a proposed fix for: ```julia u = (λ_0,v_0) J...

We can also get auto-switch algorithms to work, if they never switch to the implicit method: ```julia using StaticArrays OrdinaryDiffEq._vec(x::Tuple) = SVector(x) Base.adjoint(x::Tuple) = adjoint(SVector(x)) Base.:-(x::NTuple{N},y::NTuple{N}) where {N} = ntuple(i->x[i]...

For future reference to myself, playing around: ```julia using DifferentialEquations, Plots V = x -> x^2/2 # Potential Vp = x -> x # Force λ_0 = 2.3 # initial...

Using an automatic conversion to `TupleVec`, I can get it to be good for non-stiff ODEs. It's not robust though, and fails at the stiff ODE solver in some deep...

For now, I added an informative error: ```julia julia> solve(prob,Tsit5()) ERROR: Tuple type used as a state. Since a tuple does not have vector properties, it will not work as...

That's a Volterra integral equation. The solution will be in NeuralPDE.jl (https://neuralpde.sciml.ai/dev/) and @ashutosh-b-b is working on these solvers right now. We can use this as the example.

Yes, NeuralPDE.jl is still our preferred solver for IDEs.

Can you provide a StatProfiler flamegraph? https://github.com/tkluck/StatProfilerHTML.jl That would highlight where the potential issues are.

If you did: ```julia function noop!(integrator) u_modified!(integrator,false) end ``` does it go away? I assume a lot of it may just be running more initializations (especially for stiff ODE solvers...