ModelingToolkit -- changes with `remake`?

Specific Domains Modelling & Simulations
1

I have a @mtkbuild model named hp_na_ns_oL which uses some register_symbolic function u_u_v(t) that I specify via a constant value function u_v_const(). I want to simulate the system to steady state, and use the steady state as initial value in the next simulation – where I intend to change u_u_v(t) from being constant to u_u_v(t) = u_v_step(t).

I think the following used to work in the past:

tspan = (0,1e3)
prob_na_ns_oL = ODEProblem(hp_na_ns_oL, [], tspan)
u_u_v(t) = u_v_const(t)
sol = solve(prob_na_ns_oL, Rodas5())
x_ss = sol[end]
# remake problem with new initial state and new time span
tspan = (0,60)
prob_na_ns_oL = remake(prob_na_ns_oL, u0 = x_ss, tspan=tspan)

However, when I solve this “remake” model, it does not start in steady state! (Steady state is reached long before t=1e3…)

Has there been a “recent” change (i.e., last few months?) with remake? When I read the documentation, it mentions other functions in addition to remake, such as setp, setu, and replace. But the documentation seems a little sparse.

Any experience on how to set the initial states?

2

Open an issue to discuss.

The issue here is that if you have an initialization problem then it is solved to start the simulation. You must’ve been in a case where you didn’t have an initialization problem before, so it kind of worked, but now there’s always an initialization problem, and so that’s likely the thing involved here.

@cryptic.ax maybe we should put a warning on that remake call if there’s an initialization problem? Something to make it a bit more transparent.

3

I have opened an issue for question/discussion. I have received some suggestions, but there are still some problems I face. More in the MTK issues section.