15 175 107 stars. Each has now been assigned an age. But no number is exact: a star’s age is like trying to guess a person’s age from a photograph. You’d say “about 30,” not “exactly 30 years and 5 days.” It’s the same here: the number is shrouded in a cloud of uncertainty. This label is vital for galactic evolution: without it, the map of the Milky Way remains flat. Back in the last century, Cecilia Payne-Gaposchkin figured out what stars are made of, and Hans Bethe unraveled the source of their energy, so we know how they live and age. But age can't be measured directly. It is inferred from stellar evolution, or, for a few, from a star’s “quiver” (as was done by the Kepler telescope under William Borucki). However, for most stars we only have photometry and spectroscopy from Gaia. From these data we extract temperature, surface gravity, and metallicity — which hold the imprint of age.
The new approach uses Machine Learning. It was trained on 135 569 stars whose ages were already fairly well known thanks to the APOGEE project. Unlike a simple calculator, this algorithm outputs not a single number but a whole fan of possible values, as if saying: “most likely this, but with a spread.” The catch is that even the reference ages aren't perfect; they have their own uncertainty. So the calculation was tweaked: the known error of each reference star was added to the algorithm's own uncertainty. But the input data (temperature, chemical composition) are also measured with inaccuracies. To understand how these inaccuracies creep into the result, numerical simulation were used: for each star, 200 slightly altered parameter sets were run through and the results averaged. Thus a model was born that doesn't just name an age, but honestly reports the fuzziness of the estimate.
Verification showed: the average deviation of predicted age from true age is only –0,119 billion years, with a typical scatter of 1,200 billion years. The algorithm was then applied to 15 175 107 stars from the Gaia catalog with suitable parameters. For each star, we now have not only an age but also three types of error: from the method itself, from inaccurate input data, and the total. The typical total error is about 1,4 billion years, roughly a quarter of the Earth's age. Comparison with other methods confirms reliability. Now astronomers can select stars with acceptable precision to study star formation in different parts of the Galaxy. This allows us not to hide uncertainty but to work with it — to chart a course through the fog. Gazing through this probabilistic haze, researchers are beginning to discern the outlines of the Milky Way's history.
The famous star “Methuselah” was initially assigned an age of 16 billion years — older than the Universe itself. The error was later corrected, and now the new method helps avoid such mishaps by showing the fuzziness of each estimate.
Finding a star’s exact age from its light resembles the tricorder from Star Trek. It's not yet instantaneous analysis, but a catalog of 15 million stars is a big step toward such a tool.