What's the state of IDA(S) for sensitivity analysis?

[chaozg]

Hi there,

I recently discovered Sundials.jl and have been impressed with the performance of IDA(S) for my DAE problem. Thank you very much for your work on this!

I'm curious about the current state of IDA(S) for forward/adjoint sensitivity analysis. Specifically, can we perform sensitivity analysis with IDA(S) using the DAEProblem interface? If so, is there a demo or example that I could study?

Thanks again for your efforts, and I appreciate any guidance you can provide.

Regard,
Charlie

[ChrisRackauckas]

IIRC can already use IDA in SciMLSensitivity's ForwardSensitivity tooling. For adjoints, we just need to add a dispatch for DAEProblem in SciMLSensitivity's handling, and by doing so we don't need to have anything specific to Sundials there. We are currently missing that interface, but OrdinaryDiffEq's FBDF in mass matrix form is usually a faster way to handle DAEs and we'd recommend that for now.