possibly usable reference cvodes implementation

[Sebastian Weber]

Hi!

Maybe the cantera project is a good starting point for integrating cvodes into Stan. I think the license allows so and the code looks fairly clean:

https://github.com/Cantera/cantera/tree/2334410b5bd6a576169b56ce8596bce4a55b4541/src/numerics

and

https://github.com/Cantera/cantera/blob/2334410b5bd6a576169b56ce8596bce4a55b4541/src/numerics/CVodesIntegrator.cpp

see there the cvodes integrator classes. Looks like nice code which exposes most major options important for ODE integration and also enables LAPACK when available which should give a decent speedup (2-3x) when enabled.

The license is "new BSD" from the project home:

https://code.google.com/p/cantera/

So it should be good to go into Stan?

Sebastian

[Bob Carpenter]

Yes, BSD is OK. Overall, though, I'd prefer not to depend on
these intermediate packages. I think the main issue is getting
the builds working everywhere, so I'm not sure we need this kind
of intermediate layer.

Also, this particular one has a C-like interface from C++; I'd
much rather get a proper C++ designed interface.

I'm curious what Daniel or Michael think --- they've spent a lot
of time looking at the Sundials code so have a better idea than me
of what this interface might be contributing.

- Bob

> --
> You received this message because you are subscribed to the Google Groups "stan development mailing list" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to stan-dev+u...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

[Michael Betancourt]

They don’t do anything Tom Haber didn’t already do.
I’ll have an implementation soon.