Loading
Fetching the latest data…
Fetching the latest data…
Set a cloud of bodies going under nothing but gravity and one competition decides everything that follows: gravity pulling inward against rotation and random motion holding outward. It is the oldest question in astrophysics and it is visible in a few hundred steps — turn the rotation up and the same cloud stops collapsing. This is also the module where the platform’s determinism promise was hardest to keep, and seeing why is worth a step of its own.
Before you start — None. Nothing here needs more than the idea that things attract each other. The advanced tier states every formula and the numerical method if you want to check them.
With no rotation, a cloud of bodies simply falls inward. What happens next is not the collapse itself but what the system does with the energy it gained on the way down — and the answer is usually to throw a member out.
Do — Open the cold collapse and read clustering against bound fraction. Then raise the rotation slider and watch the contraction shrink without touching anything else.
The same cloud, same seed, same bodies, same steps — spun up. It contracts far less, and the reason has nothing to do with the simulation: rotation carries angular momentum that gravity cannot dispose of, so the cloud settles into something extended instead of a point.
Do — Open the rotationally supported preset and compare its clustering with the cold one. One slider, one difference, and it is the difference between a disc and a lump.
Energy conservation grades the arithmetic, not the result. A cloud spun so hard that nothing stays bound will often score near-perfectly on it — the bodies fly apart in straight lines, which is the easiest thing a fixed timestep ever has to follow. A cloud that collapses beautifully can score badly, because a close pass is the hardest.
Do — Open the spun-apart preset: bound fraction 0, energy conservation near 100. If a single number could tell you whether a run went well, this module would only need one gauge.
Two bodies orbiting each other have a closed-form solution known since Newton. Three do not, and not for want of trying — the motion is chaotic, so arbitrarily close starting conditions end up arbitrarily far apart. That is a theorem about the problem, not a limitation of the computer.
Do — Open the three-body preset and change the seed by one. The starting clouds are nearly identical and the histories are not. Do it again with another seed to be sure it is not one unlucky pair.
Chaos is exactly what makes this module hard to verify. Every other module here contains its floating-point risk by rounding one approximated operation; that works when the operation runs once, and fails when it runs tens of thousands of times in a system that amplifies differences exponentially. So this engine has no approximated operation at all — no square root, no exponent operator, no trigonometry. The reciprocal square root is built from multiplication and division by Newton iteration, which every processor computes identically.
Do — Save a run, then open it from the archive. The server re-ran the whole trajectory and got your numbers exactly — for a chaotic system, which is only possible because nothing in the engine is allowed to be approximate.
Gravity does not create or destroy energy, so if the simulation says the total changed, the simulation is what changed it. This grades the arithmetic, not the outcome — a cloud that flew apart can score full marks here.
Try — Open the spun-apart preset. Nothing stays bound and energy conservation reads near 100 — the integrator did its job on a system that simply came apart. Those are two different questions and this module refuses to merge them.
How much of your cloud is still held together. Bodies get flung out — that is not a failure of the simulation, it is how a gravitating system gets rid of energy.
Try — Raise the spin until bodies start leaving. The ones that go are usually the outermost, and the ones that stay end up closer together than they started.
Did the cloud pull itself together? This compares how spread out it is at the end against how spread out it began.
Try — Run the cold collapse and the rotationally supported preset back to back. Same seed, same bodies, same steps — the only difference is spin, and it is the difference between contracting and not.
A settled system has a particular relationship between its motion and its gravity: twice the kinetic energy cancels the potential. This scores how close yours is to that.
Try — Watch this against clustering. A cloud mid-collapse scores badly here even though it is contracting beautifully — being on the way somewhere is not the same as having arrived.
How many bodies have found a partner and are orbiting it. Pairs are the smallest structure gravity makes.
Try — Run the three-body preset and change the seed by one. Two nearly identical starts diverge completely — that is not noise, it is the reason the three-body problem has no general solution.
Surfacing sources you can verify is a deliberate anti-pseudoscience measure, not a bibliography. Nothing on this page asks to be taken on trust.
Glossary — every term used above, defined once.