Begin Rosenbrock refactor

[ParamThakkar123]

Checklist

• Appropriate tests were added
• Any code changes were done in a way that does not break public API
• All documentation related to code changes were updated
• The new code follows the
  contributor guidelines, in particular the SciML Style Guide and
  COLPRAC.
• Any new documentation only uses public API

Additional context

Add any other context about the problem here.
solves a part of #233

[ParamThakkar123]

@ChrisRackauckas @oscardssmith I am getting a error here saying RosenbrockCache has no field du1. This error occurs in derivative_utils.jl. I changed the line to dus so that it matches with the fields present in RosenbrockCache but that creates and error in NLNewtonCache.

Just had a question, how do I make a change so that this error gets solve and doesn't propagate or create errors in other caches ?

[ParamThakkar123]

@ChrisRackauckas @oscardssmith I am getting a error here saying RosenbrockCache has no field du1. This error occurs in derivative_utils.jl. I changed the line to dus so that it matches with the fields present in RosenbrockCache but that creates and error in NLNewtonCache.

Just had a question, how do I make a change so that this error gets solve and doesn't propagate or create errors in other caches ?

I managed to solve the other ones. But needed your suggestion on this one

[ChrisRackauckas]

du1 is used in the Jacobian calculation, so it's required for all methods that have a build_jac_config.

[ParamThakkar123]

du1 is used in the Jacobian calculation, so it's required for all methods that have a build_jac_config.

Should I be adding that explicitly as a field there ??

[ChrisRackauckas]

yes

[oscardssmith]

I've pushed a bunch of fixes to this. Specifically, du, du1 and du2 shouldn't have been a Vector, H should be a matrix, and the size of the loops should be based on the size of A rather than a fixed constant.

[ParamThakkar123]

Got it. I think I made similar mistakes in tsit. I will make a fix there too.

[ChrisRackauckas]

Got it. I think I made similar mistakes in tsit. I will make a fix there too.

No, as mentioned in other places, Tsit is completely different. It already has ExplicitRK. It's more about feature completeness of ExplicitRK and its performance.

[ParamThakkar123]

Okay, @ChrisRackauckas @oscardssmith I will start working on other solvers then. Which solvers should be the next one ?? Though I started working on SDIRK and Tsit on my own but just wanted to know if you have any priorities with respect to solvers regarding which ones should be done first

[oscardssmith]

stats look identical, and performance looks marginally better somehow? I think this is ready to merge.

function rober(u, p, t)
   y₁, y₂, y₃ = u
   k₁, k₂, k₃ = p
   [-k₁ * y₁ + k₃ * y₂ * y₃,
       k₁ * y₁ - k₃ * y₂ * y₃ - k₂ * y₂^2,
       k₂ * y₂^2]
end
prob_rober = ODEProblem(rober, [1.0, 0.0, 0.0], (0.0, 1e3), (0.04, 3e7, 1e4))

# PR
julia> @benchmark sol = solve(prob_rober, Rodas4())
BenchmarkTools.Trial: 10000 samples with 1 evaluation.
 Range (min … max):  190.246 μs … 127.780 ms  ┊ GC (min … max):  0.00% … 99.72%
 Time  (median):     220.651 μs               ┊ GC (median):     0.00%
 Time  (mean ± σ):   281.783 μs ±   1.310 ms  ┊ GC (mean ± σ):  13.86% ±  8.56%

   ██▅▄▃▂▁▂▃▃▁                                                  ▂
  █████████████▇▇▆▇▆▅▆▅▄▃▄▁▃▁▄▄▃▁▃▄▁▃▄▃▄▄▁▅▄▄▄▅▅▄▆▆▅▄▆▅▄▆▇▆▅▆▆▆ █
  190 μs        Histogram: log(frequency) by time       1.02 ms <
# v 6.89
BenchmarkTools.Trial: 10000 samples with 1 evaluation.
 Range (min … max):  203.744 μs … 129.049 ms  ┊ GC (min … max):  0.00% … 99.74%
 Time  (median):     234.780 μs               ┊ GC (median):     0.00%
 Time  (mean ± σ):   277.780 μs ±   1.306 ms  ┊ GC (mean ± σ):  12.14% ±  8.67%

      ▄▆██▇▆▅▄▃▂▂▁▁              ▁▂▁▁▁▁▂▁                       ▂
  ▆▆▆███████████████▇▆▇██▇▆▅▆▆▅▆██████████▇█▇▇▅▄▄▅▁▃▄▄▃▅▄▃▄▄▁▄▅ █
  204 μs        Histogram: log(frequency) by time        447 μs <

 Memory estimate: 452.92 KiB, allocs estimate: 5914.

[ParamThakkar123]

@oscardssmith there's an undef error saying du is not defined in stiff_addsteps

[ParamThakkar123]

Should I push a commit to fix it now

[oscardssmith]

@ParamThakkar123 where was that error? I didn't see it...

Edit... Oh I do. Your fix is slightly wrong. updated fix incoming.

[ParamThakkar123]

@oscardssmith here's a snapshot of the error. It can be seen in the buildkite tests inside integratorII_v1 tests

[oscardssmith]

@ParamThakkar123 please stop pushing incorrect fixes. du is not supposed to be in the ConstantCache.

[ParamThakkar123]

@oscardssmith okay sorry for that

[oscardssmith]

@ChrisRackauckas good to merge?

[oscardssmith]

ok, ready to merge?

[ChrisRackauckas]

There's still a lot more to do of course with Rosenbrock, since there's now two generic caches (generic_rosenbrock + this one), and this only got the 4th order.

[ParamThakkar123]

Yes @ChrisRackauckas . I am working on that part too. Will start with a new PR addressing this in an hour or two.