Christopher Rackauckas

It would be good to run similar tests on BCR (@isaacsas ) to see if that behaves the same. If it's standard, then one trick might be to: - Lower...

`sol = solve(prob,Tsit5())` is fine. It looks like IterativeSolvers.jl is having issues with the matrix-free form... @YingboMa could you look into this?

I think @YingboMa said he has this one.

```julia using OrdinaryDiffEq A = [1. 0 0 -5 4 -2 4 -3 -4 0 0 1 5 -2 2 3] u0 = rand(4,2) tspan = (0.0,1.0) function f(du,u,p,t) du...

It's just the methods like Rosenbrock which don't do nonlinear solving because it doesn't have a tolerance for the nonlinear solver, which is used to set the GMRES default. @YingboMa...

Yeah, this is more of an upstream issue with the differentiation libraries. No autodiff library currently does this, and we need a good check on the numerical diff libraries to...

Yes. That adds a few allocations to the solving though, so it might slow it down.

`integrator(t,Val{1})` probably does it (`Val{1}` means first derivative from the interpolant).

>I guess, this can be seen as a bug in IDA, because IDA has du at the initial point internally stored and could just return it. Yeah 🤷‍♂ >However, then...

I think we need to dig into Sundials and see if we can grab more cache vectors that it's already made. That's probably better than adding another register, since there's...