Astronomers have developed a machine learning algorithm which will predict the orbits of any planets will quickly become unstable. Based on the modeling computer program evaluates the most important orbital characteristics of celestial bodies with a precision of 85 percent predicted, they will survive after a billion revolutions around the star. In contrast to the predecessor, which one simulation took up to 10 hours, the algorithm does the job in seconds, and in the future the program will be able to help in the search for potentially habitable worlds. Preprint of article accepted for publication in the journal Proceedings of the National Academy of Sciences, available on the website arXiv.org.
The evaluation of the stability of planetary systems is one of the oldest qualitative problems of celestial mechanics. In the framework of Newtonian gravitational theory a system of two bodies is stable, but if it is added to another object, it can lead to the ejection from the system one of the bodies. the Main problem is the orbital resonances — a situation when the periods of rotation of two or more celestial bodies in the ratio of small natural numbers. Such resonances are found even in the Solar system: for example, Saturn and Jupiter are almost in the exact resonance 2:5 — that is, two revolutions of Saturn around the Sun correspond to three revolutions of Jupiter and Pluto is in orbital resonance 2:3 with Neptune.
Astronomer Daniel Tamayo (Tamayo Daniel) from Princeton University, along with colleagues used data on the orbital resonances, which lead to instability of the system of two planets to create a machine learning algorithm which will predict the instability of the system of three and more bodies. as all cases are impossible to reach, the researchers focused on so-called “fast instabilities” — scenarios where collisions between planets occur quite early on the age of the system. In work, the researchers defined the system as stable, if included in her celestial body is able to make 109 revolutions around the stars and not crash into each other. Such a large number of revolutions due to the fact that today the majority of discovered planets are close to their parent stars.
Astronomers have generated 100 thousand compact techplanet systems, 80 percent of which was used for learning algorithm is called Stability of Planetary Orbital Configurations Klassifier (SPOCK). In the first phase, the program simulates only ten thousand revolutions of the planets around the star, which significantly saves computational time. Based on these simulations, the algorithm extracts the ten key system parameters describing the resonance dynamics, and predicts the stability of orbits for a billion revs.
Through this approach, SPOCK determines the stability of the orbits of the planets in 100 thousand times faster than previous algorithms. Tamayo said that although he and his colleagues have not “decided” the General problem of stability of planets, SPOCK securely identificeret fast instability in compact systems, which, according to the researchers, the most important in characterizing the stability of planetary systems.
In recent years, artificial intelligence is increasingly used in astronomy. With the help of scientists trying to predict the probability of the presence of life on extrasolar planets, to look for protoplanetary disks and to record a quick radio bursts.