Half a Million M Dwarf Stars Characterized Using Domain-Adapted Spectral Analysis

TL;DR

This study presents a large catalog of atmospheric parameters for over half a million M dwarf stars derived from low-resolution spectra using a novel domain adaptation technique, improving parameter accuracy and providing a valuable resource for stellar research.

Contribution

The paper introduces Cycle-StarNet, a new domain adaptation method that enhances spectral analysis of M dwarfs, enabling accurate parameter estimation from low-quality data and creating a comprehensive stellar catalog.

Findings

Reduced fitting residuals from 2.0 to 1.68 times flux uncertainty.

Achieved systematic offsets of 12K in Teff, -0.04 dex in log g, -0.06 dex in [M/H].

Catalog includes stellar parameters, classifications, activity, and binary flags.

Abstract

We present fundamental atmospheric parameters (Teff and log g) and metallicities ([M/H]) for 507,513 M dwarf stars using low-resolution spectra (R~1800) from LAMOST DR10. By employing Cycle-StarNet, an innovative domain adaptation approach, we successfully bridge the gap between theoretical PHOENIX synthetic spectra and observed LAMOST spectra, enabling parameter measurements even for lower signal-to-noise data (S/N>5). The fitting residual analysis shows a reduction from 2.0 times to 1.68 times the flux uncertainty. Comparing with available literature values, we find systematic offsets and precisions of 12$\pm$70 K in Teff, -0.04$\pm$0.17 dex in log g, and -0.06$\pm$0.20 dex in [M/H]. The precision improves for higher quality spectra (S/N>50) to 47 K, 0.12 dex, and 0.14 dex respectively. The metallicity consistency between wide binaries shows a scatter of 0.24 dex, improving to 0.15 dex at S/N>50. We provide a comprehensive catalog including stellar parameters, spectral classifications, activity indicators, and binary/variability flags, establishing a resource for studies of the most numerous stellar population. The complete catalog is available at https://doi.org/10.5281/zenodo.14030249.