False "Cycle found" (algebraic loop) error with SUBSYSTEM containing multiple INTEGRATORs feeding a shared OUTPORT

[KhilyV3]

Hi,

I'm running into a "Cycle found" error at compile() time on a diagram that shouldn't contain a real algebraic loop, and I'd like to check whether this is expected behaviour or a known limitation.

Setup: A SUBSYSTEM containing a chain of blocks with two INTEGRATORs (e.g. a simple double-integrator plant: force → acceleration (algebraic) → integrate → velocity → integrate → position), whose OUTPORT exposes both the algebraic signal (acceleration) and the purely-integrated signals (velocity, position). Outside the subsystem, the position output feeds back into an algebraic function (e.g. a gravity model) whose output eventually feeds back into the subsystem's force input — the classic "state feedback through an algebraic nonlinearity" pattern.

Physically/mathematically there's no algebraic loop here, since the feedback path passes through two integrators (states), exactly the kind of loop Simulink resolves without any issue since integrators have no direct feedthrough.

However, bd.compile() raises a RuntimeError: could not compile system with a "Cycle found" trace that goes through the subsystem's INPORT/OUTPORT blocks, even though the actual signal that closes the loop (position) only depends on an INTEGRATOR's state, not on the algebraic (acceleration) branch.

I found a workaround: excluding the algebraic (feedthrough) output from the subsystem's OUTPORT and recomputing it separately outside the subsystem avoids the false positive, which supports the idea that this is a false detection rather than a genuine algebraic loop.

Question: Is this a known limitation of the current algebraic-loop detection, and if so, is a fix/patch planned? Happy to share a minimal reproducible script if useful.

Thanks for the great library!

[petercorke]

It'll depend a bit on what version you're running. v1.2.x fixed this, but the releases were broken so you might be on something older. Let me know. It'd help if you could attach a minimal example that shows the problem and I can test. I'd urge you to upgrade to 1.3.0 just released. Thanks for your interest.

[petercorke]

did that sort your problem?

[KhilyV3]

Hello,
Unfortunately, I have upgraded to version 1.4.0 and I still have this problem. Here is a minimal example which reproduces my error.

import bdsim
import numpy as np

sim = bdsim.BDSim()

ss = sim.blockdiagram("Subsystem")
inport = ss.INPORT(1)
outport = ss.OUTPORT(2)
int = ss.INTEGRATOR(0)
ss.connect(inport,int,outport[1])
ss.connect(int,outport[0])

bd = sim.blockdiagram()
scope1 = bd.SCOPE(title="Integrated output")
scope2 = bd.SCOPE(title="Feedthrough output")
block = bd.SUBSYSTEM(ss)
bd.connect(block[0],scope1)
bd.connect(block[1],scope2)
bd.connect(block[0],block)

bd.compile()

Which, if I am not mistaken, boils down to this diagram :

Console Output :

  ERROR: cycle found:
    Block.function('Subsystem/inport.0', type=inport, nin=1, nout=1)
    Block.function('Subsystem/outport.0', type=outport, nin=2, nout=2)
    Block.function('Subsystem/inport.0', type=inport, nin=1, nout=1)
Traceback (most recent call last):
  File "bdsim\minimal_example.py", line 21, in <module>
    bd.compile()
    ~~~~~~~~~~^^
  File "bdsim\.venv\Lib\site-packages\bdsim\blockdiagram.py", line 575, in compile
    raise RuntimeError("could not compile system")
RuntimeError: could not compile system

However, there is no real algebraic loop here. The loop is between the integrator's output and its input. But it seems that it is shadowed by the subsystem.
If you remove the subsystem and create the exact same system directly :

import bdsim
import numpy as np

sim = bdsim.BDSim()

bd = sim.blockdiagram()
scope1 = bd.SCOPE(title="Integrated output")
scope2 = bd.SCOPE(title="Feedthrough output")
int = bd.INTEGRATOR(0)
bd.connect(int,scope1)
bd.connect(int,scope2)
bd.connect(int,int)

bd.compile()

Console Output :

>>> Start simulation: T = 5, dt = None, max_step = 0.05
  Hybrid system solver: 1 continuous state variable, 0 discrete state variables
    x0 =  [1.e-05]

bdsim ◉◉◉◉◉◉◉◉◉◉◉◉◉◉◉◉◉◉◉◉◉◉◉◉◉◉◉◉◉◉◉◉ 100.0% - 0s

<<< Simulation complete
  block diagram evaluations: 710
    of which ydot calls:     608  (remaining are post-integration replay + sink passes)
  bd.evaluate() mean time:   16.7 us/call  (total 11.9 ms)
  clocktime speedup:         52.71x  (simulated 5s in 94.9 ms)
  integration time points:   102
  scheduled event intervals: 1  (each bounded by a clock tick, scheduled event, or terminal marker)
  integrator wall time:       0.012 s  (solve_ivp only; excludes post-integration replay)
  zero-crossing events:      0

Everything works fine.