I want to model the combination of a physical (and thus continuous-time) system with a (discrete-time) controller. The controller algorithm is executed, say, every second. The output of the controller (that thus changes every second) is the input to the physical system. I tried the straightforward …

[image] asprionj: What would be a good way to implement this? Basically, it would have to be something like tstops every second, but then pause the solver and let the discrete function (the controller algorithm) execute and change parameters and/or some state variables with zero autonomous dynam…

Just found out about PeriodicCallback – seems to be exactly what I’m looking for…? Edit: yes, that’s exactly it. Can even modify the parameters; awesome.

Hrm, one more thing though: CVODE_BDF works exceptionally well for my problem – it’s like 10x faster than any other solver. But it does not implement the callback interface, right?

[image] asprionj: Hrm, one more thing though: CVODE_BDF works exceptionally well for my problem – it’s like 10x faster than any other solver. But it does not implement the callback interface, right? It does implement the callback interface. (If you’d like to help donate this as a test proble…

Here’s a “minimal-as-possible” example: using DifferentialEquations using Plots f(x, p, t) = p .- x # the system ("unit first-order lag") ctrl_f(e, x) = 0.85(e + 1.2x) # the controller (proportional-integral) ctrl_x = [0.0] # array to record (discrete) state of controller function ctrl_cb(int) # …

[image] ChrisRackauckas: (If you’d like to help donate this as a test problem in the future to help us keep improving benchmarks, we would appreciate it!) I cannot donate the original problem, sorry – it comprises quite a lot of critical know-how from my job. I can try to come up with a gener…

Sundials doesn’t initialize its callbacks. That should get changed.

Callback initialization was added to Sundials master @asprionj . I’m just going to finish up its progress bar support and add a release.

Tested the use-case here with CVODE_BDF() from the latest Sundials Release 3.5.0 , it works. Thanks again Chris! Will apply it to the original problem tomorrow and see how performance is compared to the native solvers. (Larger, computationally intensive model, and much longer simulation.)

DiffEqs: Hybrid (continuous+discrete) system / periodic callback

Specific Domains Modelling & Simulations

ChrisRackauckas May 2, 2019, 7:40pm 2

Yup, tstops + DiscreteCallback is a good way to do it. Or just use the integrator interface.

Problem about using DiscreteCallback function in solving ODE problem

Topic		Replies	Views
Mixing time-discrete and time-continuous systems in a simulation Modelling & Simulations question	21	2417	April 17, 2025
Why are only ContinuousCallbacks exposed in ODESystem? Modelling & Simulations modelingtoolkit	11	801	June 13, 2022
OrdinaryDiffEq: unexpected results in DEDataArray Modelling & Simulations question , diffeq	10	880	April 21, 2021
DAE mass matrix formulation fails but implicit formulation works Modelling & Simulations question	19	751	April 7, 2024
Discrete callback + Saving callback for changing parameters in DifferentialEquations doesn't work as intended Numerics question , diffeq , differentialequation	11	225	September 12, 2024

DiffEqs: Hybrid (continuous+discrete) system / periodic callback

Related topics