Stable Orbits of Planetary Systems

Courtesy NASA/JPL-Caltech

Hello,

Today we have an entry in the field of Astronomy. It is specifically about planetary systems around other stars and their orbits.

Astronomers detect planets around other stars and collect data. There are enough data points to confirm that they have indeed detected a planet but not enough to determine the exact orbit of that planet around its host star. Throw in multiple planets around a single host star and you have a planetary system for which you do not know the stable orbital configuration.

Knowing the stable orbital configuration allows astronomers to predict planetary positions and movements which would in turn be useful for making observations (such as atmospheric composition) and help bolster or weaken theories of exoplanets.

How can ML help?

Previously, orbital configurations would have to be simulated over many billions of orbits using brute-force techniques in order to find stable configurations. These would take many hours even on modern supercomputers. Machine Learning helps with reducing this massive computational burden.

The technique works like this: Instead of simulating a given configuration over a billion orbits, it is done over 10,000 orbits which takes only a fraction of a second. From this short snippet, 10 summary metrics  are calculated that capture the system’s resonant dynamics. Finally, using these 10 features a machine learning algorithm is trained and then used to predict whether the configuration would remain stable if the simulation were to be kept running for one billion orbits. This new model is called SPOCK - Stability of Planetary Orbital Classifcations Klassifier and it helps eliminate "fast instabilities".

How good/fast is it?

Simulation runs that would take tens of thousands of hours using the traditional brute-force approach will now only be a matter of minutes using SPOCK. It is 100,000 times faster than the previous approach thus breaking the computational bottleneck.

Here's the research paper that describes the techniques and methodology used in this advancement:

 Predicting the long-term stability of compact multiplanet systems

Hope you enjoyed this brief snippet on Data Science for Science!