Stan differential equation functionality development priorities

,

Variadic arguments in ODE function

I’m working on this now. I want to have it done for the next release (roughly 2 months out?). There’s a design doc here and discussion here.

Adjoint ODEs might be able to hit the next release too. I’ve historically been unnecessarily pessimistic about the difficulty of doing that. Anyway if these are gonna go in the next release I probably need to finish the math stuff this month.

DDE/SDE/other diff eq type solvers

No developments in core Stan that I know of. @Charles_Driver is the SDE guy I know off the top of my head so he might have opinions here.

Non-linear Newton-Krylov root solvers for non-linear ODEs
Additional ODE solvers

I would like more ODE solvers. DifferentialEquations.jl seems to be the gold standard. If there are any steps we can do to identify and close the gaps, I’m in for it.

PDE solvers

Of most use to me are introductions of the non-linear root solving

If you want to do the variadic version of the algebra solver, that would be great. Or anything really.

If you want to set up the adjoint stuff, that’d also be helpful, but you’ll have to deal with the work in progress variadic ODEs branch.

@jtimonen has done some experiments with fixed-step size ODE solvers for correcting approximate solves. People using Stan have found it useful to just do their own solvers: Costs/benefits of solving large ODE systems as systems of difference equations/in discrete time - #7 by martinmodrak , so we were looking at ways they could diagnose these approximation.

1 Like