Stellar age determination using deep neural networks: Isochrone ages for 1.3 million stars, based on BaSTI, MIST, PARSEC, Dartmouth and SYCLIST evolutionary grids

TL;DR

A model-driven deep learning approach to age determination, by training neural networks on stellar evolutionary grids, which retrieves the same ages as a bayesian approach like SPInS, for only a fraction of the computational cost, with a 60,000 speedup factor for a typical star.

Abstract

We aim to develop a model-driven deep learning approach to age determination, by training neural networks on stellar evolutionary grids. Contrary to the usual data-driven deep learning approach of using prior age estimates as training data, our method has the potential for a wider and less biased range of application. The low computational cost of deep learning methods compared to bayesian isochrone-fitting allows for a broad analysis of large spectroscopic catalogues. We train multilayer perceptrons on different stellar evolutionary grids to map [M/H], MG, (GBP - GRP) to stellar age $τ$. We combine Gaia photometry and parallaxes, metallicities and $α$ elements from spectroscopic surveys and extinction maps, which are passed through the neural networks to estimate stellar ages. We apply our method to the LAMOST DR10, GALAH DR3 & DR4 and APOGEE DR17 spectroscopic surveys, for which we estimate the ages using the BaSTI tracks, along with other stellar evolutionary models. We leverage this novel technique to study, for the first time, differences in age estimates from several evolutionary grids applied on very large datasets. In addition, we date 13 open clusters and one globular cluster and find a median absolute deviation with literature ages of 0.20 Gyr. Along with the stellar ages catalogues from our estimates, we release NEST (Neural Estimator of Stellar Times), a python package to estimate stellar age based on this work, as well as a web interface. We show that, when using the same evolutionary grid, our method retrieves the same ages as a bayesian approach like SPInS, for only a fraction of the computational cost, with a 60,000 speedup factor for a typical star. This model-driven deep learning technique thus opens up the way for broad galactic archeology studies on the largest datasets available today and in the near future with upcoming surveys such as 4MOST.

Figures (13)

• Figure 1: Top panels, leftmost panel: Age comparison of Casamiquela_24's 35K sub-giant sample age estimates and results from the NN trained on BaSTI. Top panels, other panels: Age estimates from the NN trained on BaSTI compared to those from NNs trained on (from left to right) MIST, PARSEC, Dartmouth and SYCLIST evolutionary grids. Running medians are shown in white lines, as well as a one-to-one black dashed line to guide the eye. Bottom panels: Running median of the difference $\Delta\tau$ between the age estimates $\tau$ of C24 and our BaSTI NN (leftmost panel) and between our BaSTI NN and other NNs (other panels) in solid lines, and associated standard deviations in shades areas.
• Figure 2: Left panel: CMD of cluster NGC 2447 members (star symbols, coloured by our age estimates). Superimposed are BaSTI isochrones in black, and the isochrone corresponding to the cluster age given by our method in pink, with its given uncertainties in filled area. Right panel: Age comparison of our (pink symbols) and Casamiquela_24's age estimates (mode in red and median in green) and literature ages as defined in C24, coloured by number of members. NGC 2447 is highlighted in cyan.
• Figure 3: Age comparison of the "golden" samples as defined in Section \ref{['sec:age_comp']} used in this work with age estimates from the literature. From top to bottom: StarHorse Starhorse_catalog, Nataf, Xiang_2025, Kordopatis & Sanders. The number of stars is shown in each panel in the top left corner, a one-to-one dashed line is provided to guide the eye and the running median and $\pm1\sigma$ is plotted in white full and dashed lines.
• Figure 4: Abundance versus age planes for our samples, with age estimated with the NN trained on the BaSTI grid. Odd rows: Age versus [M/H], even rows: Age versus [$\alpha$/Fe]. The number of stars in each sample is annotated in the upper panel corners. Lighter colours correspond to higher density regions, and the means and $\pm1\sigma$ of the distributions are plotted as white full and dashed lines.
• Figure 5: Abundance versus age trends for each of our samples, with age estimated with the NN trained on the BaSTI grid. Coloured lines and shaded areas are the means and $\pm1\sigma$ of the distributions, as shown in white in Fig. \ref{['fig:age_abundance']}.