Stellar Pinball: A Computer Reads the Rules of Chaos ✛ ✎

Original: "Predicting the unpredictable: binary--single scattering with machine learning"

· Ahmad Farhani Asl, David Fonseca Mota, Dennis Fremstad, Fatemeh Rahimi

arXiv:2607.16763v1 · 2026-07-18 · arXiv non-exclusive · independent analysis · 2 min

A program trained on a million examples predicts the outcome of a three-star encounter with 88% accuracy.

In the hearts of globular clusters, stars are packed so tightly that binary pairs constantly collide with singles. Each encounter is like a pinball strike: the balls fly apart, swap places, or lock into unexpected combinations. This chaos governs the entire life of the cluster — from the birth of black holes and neutron stars to the generation of gravitational waves, ripples in spacetime.

Why is predicting the outcome so hard? Over a hundred years ago, mathematician Henri Poincaré showed that the three-body problem under gravity is chaotic. The slightest nudge changes the finale completely. But astrophysicists decided to take a different approach: they ran nearly a million computer simulations of such encounters and fed the results to a learning algorithm. The algorithm sought patterns where the human eye sees only disorder. The key turned out to be a simple quantity — the 'stiffness' of the pair, i.e., how tightly the stars hold together compared to the impact force. If stiffness is high, the pair is more likely to survive or swap partners.

The model correctly predicted the outcome in 88% of cases. And in 99% of instances, the correct answer was among its top two guesses, and it honestly reported when it was uncertain.

Errors didn't occur randomly; they happened at a thin boundary between two outcomes: exchange and a temporary triple union. There, fate is decided by tiny random fluctuations, like a ball teetering on a ridge between two holes. It is in this zone that chaos is fundamentally unpredictable. Such a 'chaos reader' speeds up calculations by hundreds of thousands of times and allows more accurate modeling of the lives of globular clusters, predicting how many gravitational waves will reach Earth, and even how stars evolve and new pairs form within them. Stellar pinball gains rules, and its clatter becomes a meaningful rhythm.

Whole graph →

In some configurations, three stars form a temporary hierarchical tandem that drifts through the cluster for millions of years before breaking apart.

In Liu Cixin’s famous novel 'The Three-Body Problem,' the Trisolaran civilization suffers from the unpredictable orbits of three suns. Astronomers face a similar problem at the scale of star clusters, and now some of that unpredictability has been tamed by an algorithm.

s = \frac{E_{\text{bind}}}{E_{\text{kin}}} = \frac{G m_1 m_2 / (2a)}{\frac{1}{2} \mu v_\infty^2}

Binary hardness: if s is large, the pair survives the encounter; if small, it is disrupted.

\lambda = \lim_{t\to\infty} \lim_{\delta_0\to 0} \frac{1}{t} \ln \frac{|\delta(t)|}{|\delta_0|}

Lyapunov exponent — the rate of exponential divergence of trajectories in a chaotic system. A positive value in the three-body problem gives rise to unpredictability.

Scientists

Henri Poincaré William Herschel Jan Oort

Concepts

xgboost overfitting three body problem prediction prediction accuracy probabilistic forecast confusion matrix binomial distribution training classifier forecast skill training data statistical uncertainty bayesian hierarchical model labeled data predictive learning neural networks Stochastic modeling learning rate stochastic model selection

I am the author

Stellar Pinball: A Computer Reads the Rules of Chaos ✛ ✎