Regime estimation for parameter space of small system

[ChrisRackauckas]

@SidxA

i) I had the feeling some inplace static equations had roughly the same consumption for one solve, but is this now the right choice for so many iterations?

depends. It can be better just because the cache is smaller though.

@SidxA

ii) I had the feeling that the trajectory() out of DynamicalSystems.jl had a slightly different behaviour, which I’d like to test but did’nt manage to implement the terminal callback with something like

that’d be new to me.

@SidxA

iii) Im relatively new to numerical integration and am rather insecure about the choice of solving algorithm, is the standard Tsit5() a good choice here? Is it reasonable to try out for higher tolerances explicitely? The numbers usually stay pretty low and I believe there are no huge slopes to expect.

Best thing to do is just try things. There’s a bunch of suggestions in https://diffeq.sciml.ai/latest/solvers/ode_solve/#Recommended-Methods-1, and yes try different tolerances.

@SidxA

iv)Do I have to specify some kind of multithreading or does julia automate the computation on the different cores? Would it be good to try out some form such that the solver is done one after another using the same space and storing the evaluation somewhere else because intermediate results are not required later? Regarding that which variables are better static like the initial conditions here for instance?

If using pure DiffEq here, you can use https://diffeq.sciml.ai/latest/features/ensemble/ which will parallelize it.

@SidxA

v) Is there property to check whether solver got terminated maybe by the callback, easier then comparing the final time-value?

sol.retcode == :Terminated

@SidxA

vi) Is the usage of the minimum() a good choice in the callback to define some lower threshold or would it increase the performance by putting the callback aside and looking later through the values? The point is that in some cases after the threshold is reached another equation goes through the roof ~e80.

minimum is fine. Yes, it sounds like the equation could have bad behavior near zero, and extinctions in ecological models are definitely an example of this.

@SidxA

set_parameter!()

a bit expendable here as I already call the parameters in the ODEproblem?

Not sure what you mean.