DAE mass-matrix form: how much of OrdinaryDiffEq do I need to depend on? #2513
Comments
|
you need OrdinaryDiffEqNonlinearSolve and OrdinaryDiffEqRosenbrock for singular me matrices |
|
Yeah it has a special error message for this, I wonder why it's not hit here. |
Just to confirm this, Installing OrdinaryDiffEqNonlinearSolve alone isn't enough, I also have to import it. But then it does run as it should. |
Based on a close reading of https://github.com/SciML/OrdinaryDiffEq.jl/blob/a849f328da10f39be9f4b60fa13aa712375061e3/lib/OrdinaryDiffEqCore/src/initialize_dae.jl, here is what I have pieced together:
Indeed, if I adjust the example so that we use julia> pr = ODEProblem(dae_f, [1.0, -1.0], (0.0, 1.0), initialzealg=OrdinaryDiffEqRosenbrock.SciMLBase.NoInit())
julia> solve(pr, Rosenbrock23())
retcode: Success
Interpolation: specialized 2nd order "free" stiffness-aware interpolation
t: 14-element Vector{Float64}:
0.0
1.0e-6
1.1e-5
0.00011099999999999999
0.0011109999999999998
0.011110999999999996
0.06683274154492563
0.12255448308985126
0.23458930091053964
0.346624118731228
0.5123338263199448
0.6780435339086617
0.8849787889379794
1.0
u: 14-element Vector{Vector{Float64}}:
[1.0, -1.0]
[0.9999990000005, -0.9999990000005]
[0.9999890000604997, -0.9999890000604997]
[0.9998890061602316, -0.9998890061602316]
[0.998889616891569, -0.998889616891569]
[0.9889504591072803, -0.9889504591072803]
[0.9353450167194357, -0.9353450167194357]
[0.884645223878678, -0.884645223878678]
[0.7908389642917913, -0.7908389642917913]
[0.7069797592982714, -0.7069797592982714]
[0.5989062717031488, -0.5989062717031488]
[0.5073535947357113, -0.5073535947357113]
[0.412364026329937, -0.412364026329937]
[0.36753665558720594, -0.36753665558720594]
And if I pass it an invalid initial condition, it will still solve, with a correct solution for the algebraic variable except at the initial condition (which I guess might be expected, but is somewhat disconcerting). julia> pr = ODEProblem(dae_f, [1.0, 0.0], (0.0, 1.0), initialzealg=OrdinaryDiffEqRosenbrock.SciMLBase.NoInit())
julia> solve(pr, Rosenbrock23())
retcode: Success
Interpolation: specialized 2nd order "free" stiffness-aware interpolation
t: 14-element Vector{Float64}:
0.0
1.0e-6
1.1e-5
0.00011099999999999999
0.0011109999999999998
0.011110999999999996
0.0668326888361821
0.1225543776723642
0.23458911814501143
0.34662385861765865
0.5123335359943582
0.6780432133710577
0.8849784513102263
1.0
u: 14-element Vector{Vector{Float64}}:
[1.0, 0.0]
[0.9999990000005, -0.9999990000005006]
[0.9999890000604997, -0.9999890000604995]
[0.9998890061602316, -0.9998890061602318]
[0.998889616891569, -0.9988896168915689]
[0.9889504591072803, -0.9889504591072802]
[0.9353450660390021, -0.935345066039002]
[0.8846453171711408, -0.8846453171711407]
[0.7908391089561135, -0.7908391089561134]
[0.7069799433906985, -0.7069799433906983]
[0.5989064458099173, -0.598906445809917]
[0.5073537576078662, -0.5073537576078659]
[0.41236416579338064, -0.4123641657933806]
[0.3675366555969062, -0.3675366555969061]
So it seems to me like the default init shouldn't be |
|
Contemplating how this would be fixed: should there be a method in Oh, but we were supposed to hit an error, and in this case I managed to get a solution without an error. So maybe the error needs to move further up the call stack to |
|
We can check |
|
Okay, took a stab at a pull request which makes a change of this sort. I'm still relatively inexperienced with this kind of contribution, so I'm happy to make whatever changes are necessary. (Haven't done the code formatting step yet, but I assume that more changes will be needed anyway to my solution before it will get merged.) A preliminary test indicates that with the above MWE, that pull request triggers the appropriate error if OrdinaryDiffEqNonlinearSolve isn't installed, but still triggers the error if ODENonlinearSolve is loaded, so I know I have it wrong. Will take a look later. |
Describe the example
In the switch from [email protected] to [email protected], some tooling for solving DAEs in mass-matrix form is no longer exported.
I notice that on the upgrade from 1.1 to 1.2, OrdinaryDiffEqNonlinearSolve and some other related packages are removed as dependencies.
If I just need to depend on another subset of SciML, I don't mind, just need to know what to import (and maybe this should make it into the documentation at https://docs.sciml.ai/DiffEqDocs/stable/tutorials/dae_example/ ?)
Minimal Reproducible Example 👇
Error & Stacktrace⚠️
With OrdinaryDiffEqRosenbrock v1.2
Not Working Environment (please complete the following information):
using Pkg; Pkg.status()using Pkg; Pkg.status(; mode = PKGMODE_MANIFEST)versioninfo()Working Environment (please complete the following information):
using Pkg; Pkg.status()using Pkg; Pkg.status(; mode = PKGMODE_MANIFEST)versioninfo()Additional context
The text was updated successfully, but these errors were encountered: