Open baggepinnen opened 4 weeks ago
In MTK v8, the call to init
is very fast
julia> @btime init(prob, OrdinaryDiffEq.Rodas5P())
61.636 μs (361 allocations: 74.64 KiB)
and produces a zero residual (last two equations)
julia> prob.f(res.u, prob.p, 0)
9-element Vector{Float64}:
-0.0
0.00042497264388805433
-0.0013812304089120954
0.11162675106913678
-0.0
-0.0033437500000000004
0.0
-6.622879952788985e-9
3.034388078286968e-8
The new init produces huge residuals
The linearization workflow below took
on MTKv8 while on master it takes
Most of this time appears to be in this call.
It also appears as if the result is wrong in MTK v9
I have other examples where the regression is over 100x