Inferring stellar metallicity and elemental abundances from kinematic and spectroscopic data using machine learning

V. Adibekyan, B. M. T. B. Soares, S. G. Sousa, N. C. Santos, E. Delgado-Mena, I. Minchev, R. Chertovskih, Zh. Martirosyan, G. Israelian, A. A. Hakobyan · 2026

Read Inferring stellar metallicity and elemental abundances from kinematic and spectroscopic data using machine learning by V. Adibekyan, B. M. T. B. Soares, S. G. Sousa, N. C. Santos, E. Delgado-Mena, I. Minchev, R. Chertovskih, Zh. Martirosyan, G. Israelian, A. A. Hakobyan (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-18131.txt.

Introduction

V. Adibekyan's 2026 study uses machine learning to infer star chemistry from motion and spectra where direct measures fail, notably for cool stars. The team trains on a large APOGEE sample of red giants and checks the models on a separate HARPS sample of near dwarfs. Motion alone recovers only a small share of the spread in iron content, with vertical orbit height the best single clue. Adding iron content sharpens forecasts of carbon, oxygen, magnesium, and silicon well beyond a plain baseline.

Catalog record

CREATORV. Adibekyan, B. M. T. B. Soares, S. G. Sousa, N. C. Santos, E. Delgado-Mena, I. Minchev, R. Chertovskih, Zh. Martirosyan, G. Israelian, A. A. Hakobyan
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE75238
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.18131

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Inferring stellar metallicity and elemental abundances from kinematic and spectroscopic data using machine learning V. Adibekyan, B. M. T. B. Soares, S. G. Sousa, N. C. Santos, E. Delgado-Mena, I. Minchev, R. Chertovskih, Zh. Martirosyan, G. Israelian, A. A. Hakobyan Abstract Context. Elemental abundances of FGK stars can be derived routinely from high-resolution optical spectra, but this remains considerably more difficult for cooler stars. Yet, even estimates of the chemical composition of planet-host stars is central to studies of planet formation and interior composition. Machine-learning methods offer a practical route to infer otherwise inaccessible abundances from more widely available stellar data. Aims. We investigate how much information on stellar metallicity and selected elemental abundances is encoded in kinematic and orbital properties, and whether these empirical relations can be used to estimate abundances and abundance ratios relevant for exoplanet-host characterisation. Methods. We use a large APOGEE DR17 sample of red giant stars as the main training set and an independent HARPS sample of nearby FGK dwarfs for external validation. From astrometry and radial velocities we derive Galactic velocities, orbital parameters, and stellar population-membership probabilities. We benchmark several machine-learning regressors, optimise the strongest models, and analyse feature importance using gain-based metrics, permutation importance, single-feature models, and SHAP values. We also explored the prediction of C and O from Mg, Si, and [Fe/H], and derived simple empirical relations between selected abundance ratios (Fe/Si, Mg/Si, C/O, and Fe/O) and metallicity. Results. Kinematic information alone recovers only a limited fraction of the variance in stellar metallicity, with a clear performance ceiling at RMSE ∼ \sim 0.20 dex. The most informative predictor is the maximum vertical orbital excursion, Z max Z_{\max} , followed by radial orbital parameters. When [Fe/H] is combined with kinematic information, the abundances of C, O, Mg, and Si are predicted significantly more accurately than with the baseline approximation [ X / H ] = [ Fe / H ] \mathrm{[X/H]}=\mathrm{[Fe/H]} . In contrast, when predicting C and O from Mg, Si, and [Fe/H], most of the predictive power is already contained in the elemental abundances themselves, with Mg being the dominant contributor, and the addition of kinematic information provides little improvement. The trained models reproduce the main abundance trends associated with Galactic chemical evolution. We find that the slopes of the relations between Fe/Si, Mg/Si, C/O, and Fe/O and metallicity differ slightly between the HARPS and APOGEE samples, with fractional differences generally below 17%. Conclusions. Stellar kinematics and orbital parameters contain only limited information on stellar metallicity, with predictive performance primarily constrained by the intrinsic information content of the input features rather than by the choice of model architecture. Nevertheless, when combined with metallicity, these parameters enable accurate predictions of elemental abundances, including those that are difficult to determine directly. This provides a practical framework for extending chemical characterisation to stars for which detailed abundance determinations are challenging. Implications for exoplanet host stars 1 Introduction Determining stellar chemical abundances is fundamental for a wide range of astrophysical studies, from Galactic chemical evolution to the formation and characterisation of planetary systems. In the context of exoplanets, the chemical composition of the host star is of particular interest because it reflects, to first order, the composition of the protoplanetary disk. Stellar metallicity is known to correlate with the occurrence of giant planets ( 65 ; 26 ; 5 ) , while abundance ratios such as Mg/Si, Fe/Si, and C/O are relevant for the mineralogy, internal structure, and bulk composition of planets ( 13 ; 20 ; 75 ; 4 ; 18 ) . Large surveys such as the Apache Point Observatory Galactic Evolution Experiment ( 1 , APOGEE;) and the GALactic Archaeology with HERMES survey ( 16 , GALAH;) have enabled precise determinations of chemical abundances for hundreds of thousands of stars. However, accurate abundance determinations remain challenging for several stellar regimes and chemical species. In cool stars, particularly at optical wavelengths, spectra are strongly affected by molecular absorption and severe line blending, which complicate both the determination of stellar parameters and detailed abundance analyses ( 76 ; 58 ; 47 ; 49 ) . This problem is especially acute for M dwarfs, for which even metallicity determinations remain difficult and are often based on indirect calibrations or limited spectral diagnostics ( 56 ; 64 ; 48 ; 7 ; 21 ) . Even when high-resolution near-infrared spectra are available, multi-element abundance determinations for M dwarfs still require careful modelling and remain limited to relatively small samples ( 70 ; 73 ; 74 ; 57 ) . The determination of C and O abundances from optical spectra is also challenging for stars cooler than 5200 K ( 10 ; 72 ; 18 ; 11 ) , because the lines become weak at these temperatures and blends become increasingly important. The growing volume of stellar datasets motivate the use of data-driven methods. Machine-learning (ML) techniques have been widely applied to derive stellar parameters and abundances from spectra and photometric data, capturing complex and non-linear relations that are difficult to model explicitly ( 8 ; 54 ; 23 ; 43 ) . Recent studies have shown that ML models can recover metallicity for cool stars from photometric data in M dwarfs ( 21 ) , further illustrating the potential of these methods in regimes where classical analyses face limitations. The connection between stellar chemistry and dynamics is an important ingredient of Galactic archaeology, since stars with different birth radii, ages, and evolutionary histories populate different regions of chemo-dynamical space ( 2 ; 12 ; 52 ; 61 ; 32 ) . In particular, metallicity and abundance ratios have been shown to correlate with orbital properties reflecting the coupled chemical and dynamical evolution of the Galactic disc ( 12 ) . Recent studies have also highlighted the strong connection between stellar abundances, age, and Galactic dynamics. Using APOGEE red clump stars in the low \alpha disk, 55 showed that many elemental abundance ratios can be predicted with high precision from stellar age and metallicity alone, implying that much of the chemical information is encoded in these two quantities. Their results also emphasize the strong coupling between the chemical and dynamical evolution of the Galactic disk and provide additional motivation for exploring data driven relations between abundances and stellar dynamical properties. In this work, we investigate how well stellar metallicity and individual elemental abundances can be inferred from kinematic and orbital information, and whether these relations can be exploited to estimate abundances that are particularly relevant for exoplanet studies. We first examine the predictive power of stellar dynamics for metallicity (relevant for M dwarfs), and then explore the extent to which metallicity and kinematics can recover the abundances of C, O, Mg, and Si. Finally, we test whether C and O can be predicted from Mg, Si, and [Fe/H], with optional inclusion of kinematic information. This approach is particularly relevant for extending chemical characterisation to stars for which direct abundance determinations, especially of C and O, are difficult. Finally, we apply our models to stars hosting planets and investigate the resulting abundance distributions. This allows us to extend chemical studies of planet-host stars to samples where direct abundance determinations are unavailable, providing new insights into the chemical conditions

Read the complete text