Vedant Puri

@gaurav-arya , would you be able to help with this?

It is a mistake for the simple reason that ```julia julia> A(B(C(u, p, t), p, t), p, t) ≈ (A ∘ B ∘ C)(u, p, t) false julia> A(B(C(u, p,...

Correct

We should always use `L(du, u, p, t), L(u, p, t)`. The `*, mul!` are for convenience.

yeah i agree. Then we can simplify the API from `MatrixOp(A; update_func = .., update_func! = ..)`, to just having one kwarg `MatrixOp(A; update_func = ..)`, where `update_func` can have...

Also, I'd like to remove `u` dependence in `update_coefficients/update_func` for the reason below. https://github.com/SciML/SciMLOperators.jl/blob/master/src/interface.jl#L42-L54 It leads to unsafe behaviour that we have no way to protect against. Removing `u` from...

thanks

what you could do is pass the state vector `u` via the `p` parameter, and apply the operator to some other vector `v`, ```julia update_coefficients!(L, u, p, t) = update_coefficients!(L,...

does something like this work? ```julia function Ajac(Jv, v, u, p, t;A=A) SciMLOperators.update_coefficients!(A, u, p, t) mul!(Jv, A, v) end ``` where `A isa SciMLOperator`

With current TensorProductOperator implementation, https://github.com/SciML/SciMLOperators.jl/blob/c9488881763bf838898f212f3aa071e4e987bc53/src/sciml.jl#L684-L724 reference computation: - a single component matvec. ```julia using LinearAlgebra, BenchmarkTools u = rand(12, 12 * 100) v = rand(12, 12 * 100) A =...