Towards model-free stellar chemical abundances. Potential applications in the search for chemically peculiar stars in large spectroscopic surveys

Abstract

Chemical abundance determinations from stellar spectra are challenged by observational noise, limitations in stellar models, and departures from simplifying assumptions. While traditional and supervised machine learning methods have made remarkable progress in estimating atmospheric parameters and chemical compositions within existing physical models, these factors still constrain our ability to fully exploit the vast data sets provided by modern spectroscopic surveys. We aim to develop a self-supervised, disentangled representation learning framework that extracts chemically meaningful features directly from spectra, without relying on externally imposed label catalogs. We build a variational autoencoder-based representation learning model with physics-inspired structure: multiple decoders each focus on spectral regions dominated by a particular element, enforcing that each latent dimension maps to a single abundance. To evaluate the potential application of our framework, we trained and validated the model on low-resolution, low signal-to-noise synthetic spectra focusing on $[Fe/H]$, $[C/Fe]$, and $[\alpha /Fe]$. We then demonstrate how the trained model can be used to flag stars as chemically enhanced or depleted in these abundances based on their position within the latent distribution. Our model successfully learns a representation of spectra whose axes correlate tightly with the target abundances ($r=0.92\pm 0.01$ for $[Fe/H]$, $r=0.92\pm 0.01$ for $[C/Fe]$, $r=0.82\pm 0.02$ for $[\alpha /Fe]$). The disentangled representations provide a robust means to distinguish stars based on their chemical properties, offering an efficient and scalable solution for large spectroscopic surveys.

16 pages, 11 figures, accepted on A&A

References in corpus (24)

• A grid of MARCS model atmospheres for late-type stars I. Methods and general properties
• The GALAH Survey: Scientific Motivation
• Determining stellar atmospheric parameters and chemical abundances of FGK stars with iSpec
• Carbon Enhanced Metal-Poor Stars. I. Chemical Compositions of 26 Stars
• Carbon-Enhanced Metal-Poor Star Frequencies in the Galaxy: Corrections for the Effect of Evolutionary Status on Carbon Abundances
• Modern stellar spectroscopy caveats
• Gaia Data Release 3: Analysis of RVS spectra using the General Stellar Parametriser from spectroscopy
• The Galaxy and its stellar halo: insights on their formation from a hybrid cosmological approach
• Probing the Formation of the First Low-Mass Stars with Stellar Archaeology
• The Gaia-ESO Public Spectroscopic Survey: Implementation, data products, open cluster survey, science, and legacy
• The Stagger-grid: A grid of 3D stellar atmosphere models - III. The relation to mixing-length convection theory
• The Gaia-ESO Public Spectroscopic Survey: Motivation, implementation, GIRAFFE data processing, analysis, and final data products
• Four-hundred Very Metal-Poor Stars Studied with LAMOST and Subaru. II. Elemental abundances
• Searching for changing-state AGNs in massive datasets -- I: applying deep learning and anomaly detection techniques to find AGNs with anomalous variability behaviours
• On the Relative Ages of the $\alpha$-Rich and $\alpha$-Poor Stellar Populations in the Galactic Halo
• A variational encoder-decoder approach to precise spectroscopic age estimation for large Galactic surveys
• Formation of carbon-enhanced metal-poor stars as a consequence of inhomogeneous metal mixing
• Autoencoding Galaxy Spectra I: Architecture
• Disentangled Representation Learning for Astronomical Chemical Tagging
• Predicting metallicities and carbon abundances from Gaia XP spectra for (carbon-enhanced) metal-poor stars
• Disentanglement Learning for Variational Autoencoders Applied to Audio-Visual Speech Enhancement
• Using autoencoders and deep transfer learning to determine the stellar parameters of 286 CARMENES M dwarfs
• Disentangling stellar atmospheric parameters in astronomical spectra using Generative Adversarial Neural Networks
• M3DIS -- A grid of 3D radiation-hydrodynamics stellar atmosphere models for stellar surveys. II. Carbon-enhanced metal-poor stars