Sandu projection

[SKopecz]

This implements Sandu's projection method, see #124.

[JoshuaLampert]

Would it make sense and be feasible to put JuMP.jl in an extension as it is not very lightweight and not necessary for every user, such that we save the additional loading time if no Sandu projection is used?

[SKopecz]

Is 25a8127 what you had in mind?

[JoshuaLampert]

Yes, looks good. Thanks!

[codecov]

Codecov Report

❌ Patch coverage is 98.75000% with 1 line in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
ext/JuMPExt.jl	97.77%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

[coveralls]

Pull Request Test Coverage Report for Build 18007485686

Details

• 79 of 80 (98.75%) changed or added relevant lines in 2 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+0.04%) to 97.805%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
ext/JuMPExt.jl	44	45	97.78%

Totals
Change from base Build 17996129247:	0.04%
Covered Lines:	2050
Relevant Lines:	2096

---

💛 - Coveralls

[SKopecz]

This is a scaled version of the stratospheric reaction problem and can be solved using SanduProjection and Clarabel with default settings.

Unfortunately, scaling destroys the "conservation structure" of a PDS, i.e. terms that canceled before scaling, do not afterwards. I was not able to find a PDS representation of the scaled statospheric problem, which could be solved as good as the original one by MPRK schemes. That's why there is only an ODE version of this problem.

[SKopecz]

I would have expected that using StaticArrays is not necessary in the extension, as it is used in the parent package.

[SKopecz]

The code

using PositiveIntegrators
using OrdinaryDiffEqRosenbrock: ROS2
using JuMP: Model
using Clarabel
using Profile

AT = linear_invariants_stratreac_scaled()
b = AT * prob_ode_stratreac_scaled.u0
cb = SanduProjection(Model(Clarabel.Optimizer), AT, b, 1.0; save = false)

function doit(args...; n = 1)
    for _ in 1:n
        solve(prob_ode_stratreac_scaled, ROS2();
              save_everystep = false, callback = cb)
    end
end

@profview doit(1)
@profview doit(10^6)

generates the following flame graph.

There seems to be quite some runtime dispatch. But most of it does not seem to be related to SanduProjection. Zooming into the discrete callback section, we see:

As I see it, the runtime dispatch is a problem of Clarabel.

What do you think?

[JoshuaLampert]

Thanks! I left some comments and suggestions below.

[JoshuaLampert]

Thanks!

[JoshuaLampert]

Thanks!

[ranocha]

Thanks