Long runtimes for structural_simplify on large chemical reaction network model (potential future benchmark case?)

[TorkelE]

The model in question can be found here: https://github.com/SciML/Catalyst_PLOS_COMPBIO_2023/tree/master/Benchmarks/Data (it is the BCR.net file). It is quite large, about 1100 unknowns.

Currently, structural_simplify has very long runtimes when applied to this model. I imagine that this might make sense, as structural_simplify haven't been optimised for this kind of model yet. However, it might make sense to add it as a future benchmark for evaluating structural_simplifys performance?

To retrieve the model run

using Catalyst, ReactionNetworkImporters
BCR_loaded = loadrxnetwork(BNGNetwork(), "BCR.net")
BCR = complete(BCR_loaded.rn)

# Without remove conserved.
begin
    @time osys = convert(ODESystem, BCR; remove_conserved = false) # 3.4 seconds.
    @time complete(osys) # 1.9 seconds.
    @time structural_simplify(osys) # 471 seconds.
end

(I tried priming the model through Catalyst's remove_conserved, but that didn't help)

Actually simulating the model is fairly straightforward and quick

using OrdinaryDiffEq
oprob_js = ODEProblem(BCR, BCR_loaded.u0, 1000.0, BCR_loaded.p; jac = true, sparse = true)
@time sol = solve(oprob_js, Rodas5P()) # 11 seconds