The GALAH Survey: A New Sample of Extremely Metal-Poor Stars Using A Machine Learning Classification Algorithm

TL;DR

This paper uses a machine learning algorithm on GALAH survey spectra to identify 54 extremely metal-poor stars, demonstrating a scalable method for future large-scale stellar surveys.

Contribution

It introduces a novel machine learning approach to efficiently identify EMP stars from large spectroscopic datasets, including a significant number of main sequence candidates.

Findings

Identified 54 EMP candidates with [Fe/H] ≤ -3.0

Discovered 6 stars with [Fe/H] ≤ -3.5

Sample's metallicity distribution aligns with previous studies

Abstract

Extremely Metal-Poor (EMP) stars provide a valuable probe of early chemical enrichment in the Milky Way. Here we leverage a large sample of $\sim600,000$ high-resolution stellar spectra from the GALAH survey plus a machine learning algorithm to find 54 candidates with estimated [Fe/H]~$\leq$~-3.0, 6 of which have [Fe/H]~$\leq$~-3.5. Our sample includes $\sim 20 \%$ main sequence EMP candidates, unusually high for \emp surveys. We find the magnitude-limited metallicity distribution function of our sample is consistent with previous work that used more complex selection criteria. The method we present has significant potential for application to the next generation of massive stellar spectroscopic surveys, which will expand the available spectroscopic data well into the millions of stars.