Hi Greg,
The problem I see is that simulations of any kind are always designed to be efficient. (Like Global illumination is not physically correct, it is physically plausible, as light has no problems bouncing a billion times around by itself. If we would simulate that with a standard computer, it might take years.)
With this example, we try to find a sweet spot that satisfies the eye. Well, it is better than that, but I like to make a point. From that perspective, each iteration will be less precise, like a copy of a copy, which can be counteracted in many ways.
First, you set up the main hanger; since these are always in balance, it doesn’t matter how much weight is hanging on each side since the balance determines the main idea. So the first central part can be set in motion, cached, baked, or exported (Alembic), and re-imported.
Which eliminates any dynamic but leaves the results of the dynamic. (At the moment, it is animation data of any kind; the objects can be made into Collision objects or just left alone.)
The next iteration builds up on stable data. The new dynamic can be backed, cached or exported, and re-imported. Again we have on level two stable data. (For these step by step, a good idea is to set the Dampening high and then quickly lower, which tames the system.)
Then we build levels three, four, five, or how many level A. Calder had in his work? Some go beyond a dozen.
I know that the system is more complex than that. The last element in the mobile will affect the whole system if one blows shortly on that element, which would move the whole system and, in return, the initiator of that movement. This is how I read Calder’s work as an artist (meaning) and from the point of a social structure (consequence) as an architect reading static/dynamic to create. A. Calder would have ignored that, for him it was an object. Early on he used a motor to drive his objects, after doing it manually. later the motor idea was left, and natural movements were used. So, the idea to animate here one or the other way is plausible: Artist’s intention is certainly key.
Of course, we can look at it in many ways, but the complexity is given, no matter what, and that is kind of the problem for the system here.
So far, I have had no luck getting access to get these two systems working bilaterally. The Bullet system has XPresso nodes, so you could get some information from there, a one-way road.
Bounding boxes are based on the Shape you set up. If you do not specify those, a flat sphere represents your Torus. Bullet allows for Proxy models, low poly, to save time.
If your setup shows instabilities, perhaps go to the Expert Tab and change the steps/ iterations.
Enjoy your exploration; as usual, there is no failure, only gained experience.