Verifying the stability of planetary systems using a machine learning algorithm

Dan Tamayo (Photo courtesy Ken Jones)
Dan Tamayo (Photo courtesy Ken Jones)

An article published in “The Astrophysical Journal Letters” describes the development and the application of machine learning algorithms to verify the stability of planetary systems. A team of researchers at the University of Toronto Scarborough led by Dan Tamayo experimented this new approach to this type of astronomical research by creating a method a thousand times faster than conventional ones.

In recent decades the study of exoplanets has become an increasingly important branch of astronomy thanks to the development of better and better instruments to discover them. Today there are several methods to identify exoplanets but they can provide us with only incomplete information. This makes verifying the stability of the planetary systems that host them much more complex because of the remaining variables.

Supercomputers are getting better and better as well so their computing power keeps on increasing but there are scientific problems that still require long processing times due to the enormous amounts of data that need to be considered. Luckily, advances in information technology also concern software and the progress occurred in artificial intelligence led among other things to the development of machine learning algorithms.

Big companies are already using machine learning algorithms in many ways. Google created a number of them, for example to handla spam in Gmail or to improve its Translate service. The advantage of this type of algorithm is that a system that implements one can learn through a training and starting from the data that are provided then can carry out evaluations and generate new contents without the need of additional programming.

In the field of astronomy and in particular of planetary systems, simulations are an application to celestial mechanics of the n-body problem. Dan Tamayo’s team developed a machine learning algorithm to solve that kind of problem that turned out to be a thousand times faster than direct simulations.

Some weeks were been needed to train the machine learning system but it was worth it. The results show that this type of approach is useful in the field of astronomy too, in this case in the study of planetary systems. Figuring out if they’re stable means to better understand their formation mechanisms and their evolution. These findings could open the door to other developments of this type.

Leave a Reply

Your email address will not be published. Required fields are marked *