Docs suggestions

[paulflang]

1. Here, I think it should be rober! instead of rober.
2. In the global structural identifiability tutorial:
   a. The docs say that "We can see that all states (except $x_7$) and all parameters are locally identifiable", despite the results don't show any states.
   b. Why does it say that "We can see that only parameters a, g are unidentifiable", when the result suggest only g is unidentifiable.
   c. In the case of inputs u1 and u2 It is unclear whether these inputs are known perturbations (to improve identifiability) or are unknown and shall be estimated (reduces identifiability). In case of the former, the example should test if g is now identifiable. In the case of the latter, it would still be interesting to know why we are not checking for all parameters.
   d. AFAIK structural identifiability methods scale poorly to large models. Maybe it would be good to add a sentence on scalability.
3. The example for DAE Index Reduction uses structural_simplify(dae_index_lowering(traced_sys)), despite another place states that this is superfluous, as structural_simplify calls it internally. Maybe worth clarifying.

[ChrisRackauckas]

@AayushSabharwal let's make sure we take this into account with the overhaul

[bjarthur]

what is the latest on scaling to large models? 2.d. in the OP. i have heard too that the time to compile is quadratic in the number of equations. is that still the case?

[ChrisRackauckas]

Structural identifiability scaling is very different from MTK scaling. That is dependent on whether it's global or local.

[ChrisRackauckas]

As for other large model pieces, JuliaSimCompiler achieves very good scaling, and we have added new CSE passes. We'll be adding even more CSE in a few weeks too. So I think we'll take stock of where we are at in just a few months.

https://docs.sciml.ai/SciMLBenchmarksOutput/stable/ModelingToolkit/ThermalFluid/ this is the benchmark we use to track this scaling.

[bjarthur]

thanks for the link to the benchmarks. i also found these and noticed that many of the plots are completely blank:

https://docs.sciml.ai/SciMLBenchmarksOutput/stable/ModelingToolkit/RCCircuit/

[bjarthur]

and on this page the "go to benchmarks.sciml.ai" links back to the very same page.

[bjarthur]

and perhaps it's better documented elsewhere, but it took me awhile to figure out that JSIR in those plots refers to the julia sim compiler. that could perhaps be explained a bit better on the benchmarks page. the difference with MTK in those plots being that JSIR is not open source, right? you need a license to JuliaSIM?

[ChrisRackauckas]

Yes though it's free for academic use.

i also found these and noticed that many of the plots are completely blank:

Yes someone needs to go through and figure out what happened to the RC circuit one.