By Ilana MacDonald, Dunlap Institute for Astronomy & Astrophysics

Determining the ages of stars is fundamental to understanding many areas of astronomy. Despite this, it remains an extremely challenging problem to solve, as stellar ages cannot be ascertained by simply observing stars. University of Toronto astronomers have developed a machine-learning model to help solve this problem.

This research, published today in The Astrophysical Journal, was led by PhD candidate Phil Van-Lane of the David A. Dunlap Department of Astronomy & Astrophysics, along with Prof. Josh Speagle and Prof. Gwen Eadie, both jointly appointed between the Department of Statistical Sciences and the David A. Dunlap Department of Astronomy & Astrophysics.

Stars tend to form in clusters, which allows one to assume that members of the same star cluster share the same age. By observing the evolutionary stages of higher mass stars, which progress more rapidly than those of lower mass stars, researchers can often determine the age of all stars in the cluster. However, this also means identifying the age of lone stars is far more challenging since there are no other stars of the same origin to serve as points of comparison.

One method that has most recently been used to determine the age of a star is by examining its rotational speed. As stars age, their spin tends to slow down due to the interaction of the star’s magnetic field with its stellar wind. Though well understood qualitatively, this phenomenon is difficult to quantify.

Gyrochronology, the process of deriving a star’s age from its rotation period, has been studied since 1972, when Dr. Andrew Skumanich first introduced the concept. Although many analytical models have been developed over the past decades to describe how the rotation speed of stars changes over time, none of them have been able to accurately solve this problem. Even though it is known that stars do slow down over time, this phenomenon cannot be easily described by a simple mathematical formula.

With the advent of large amounts of data from stellar surveys such as Kepler, K2, TESS, and GAIA, it has now become possible to use machine learning techniques to find a solution to this problem. The new model introduced by Van-Lane and his collaborators, called ChronoFlow, is the first to use state-of-the-art machine learning techniques.

In this research, Van-Lane and his colleagues assembled the largest catalogue of rotating stars in clusters to date, with about 8,000 stars in over 30 clusters of various ages. Their new model, ChronoFlow, uses this dataset and machine learning to describe how the speed at which a star rotates changes as it ages. Because their model uses machine learning, it can predict ages of stars with an accuracy previously impossible to achieve with analytical models.

 

“Our methodology can be likened to trying to guess the age of a person,” says Speagle, who guided the project from start to finish. “In astronomy, we don’t know the ages of every star. We know that groups of stars have the same age, so this would be like having a bunch of photos of people at 5 years old, 15 years old, 30 years old, and 50 years old, then having someone hand you a new photo and ask you to guess how old that person is. It’s a tricky problem!”

Using the dataset of the ages and rotation rates of stars in clusters, ChronoFlow has learned to  estimate the ages of other stars with remarkable precision. This is because it models how rotation rates of populations of stars are expected to evolve over time.

“The first ‘wow’ moment was in the proof-of-concept phase when we realized that this technique actually showed a lot of promise,” continues Van-Lane, “and was able to model the continuous evolution of stellar rotation periods.”

These results will have important implications across many aspects of astronomy. Knowing stellar ages is necessary to understanding not only how stars work, but also modeling how exoplanets form and evolve, and learning about the history of the evolution of our own Milky Way as well as that of other galaxies.

The success of ChronoFlow in this scenario demonstrates that machine learning models are highly effective in addressing astrophysical problems when trained on datasets like these star clusters, which contain unevenly distributed data collected from many different telescopes. This suggests that machine learning may provide valuable insights into other astrophysical problems.

The model will be available to the public, along with documentation and tutorials which provide steps for anyone to infer the ages of stars from observations. The code can be found here: https://github.com/philvanlane/chronoflow.

 

About the Dunlap Institute for Astronomy & Astrophysics

The Dunlap Institute for Astronomy & Astrophysics in the Faculty of Arts & Science at the University of Toronto is an endowed research institute with over 80 faculty, postdocs, students, and staff, dedicated to innovative technology, groundbreaking research, world-class training, and public engagement.

The research themes of its faculty and Dunlap Fellows span the Universe and include optical, infrared and radio instrumentation, Dark Energy, large-scale structure, the Cosmic Microwave Background, the interstellar medium, galaxy evolution, cosmic magnetism and time-domain science.

The Dunlap Institute, the David A. Dunlap Department of Astronomy & Astrophysics, and other researchers across the University of Toronto’s three campuses together comprise the leading concentration of astronomers in Canada, at the leading research university in the country.

###

For more information, contact:

Ilana MacDonald
Public Outreach, Communications, and Events Strategist
Dunlap Institute, University of Toronto
media@dunlap.utoronto.ca