The SALT survey of helium-rich hot subdwarfs: unsupervised classification and kinematic analysis

TL;DR

This study uses unsupervised machine learning techniques on spectroscopic data from SALT to classify helium-rich hot subdwarfs, revealing new spectral groupings and their kinematic properties, thus advancing understanding of their evolution.

Contribution

Introduces a PCA and spectral clustering-based classification system for helium-rich hot subdwarfs, uncovering new spectral groups and refining existing classification schemes.

Findings

Identified 6 major spectral clusters corresponding to traditional classes.

Revised classification for very hot sdOs based on PCA results.

Found higher halo star fractions in He-sdOs and Hot-sdOs.

Abstract

Hydrogen-deficient stars form sequences across the HR diagram from cool R CrB stars to helium-rich hot subdwarfs and PG1159 stars, tracing multiple evolution pathways. The Southern African Large Telescope (SALT) has been used to conduct a medium-resolution spectroscopic survey of 697 helium-rich hot subdwarfs and related stars. Using 587 stars from the full SALT sample, we define an unsupervised data-led classification system based on Principal Components Analysis (PCA) and Spectral Clustering (SC), designed as a data-discovery tool to identify possible new groups and sequences within the data. Using the first 3 principal components (PCs), we identify 6 major clusters of hot subdwarf spectra, identified with the traditional spectral classes of classical sdB and sdO stars, helium-rich He-sdO stars and very hot subdwarfs. Two clusters covering large volumes of PC space include a) He-sdB and He-sdOB stars and b) intermediate iHe-sdB and iHe-sdOB stars. Most spectra in these major clusters form connected sequences in PC space. Using a second stage of PCA and SC, we identify sub-clusters within 4 major clusters, particularly in the He-sdB/OB and iHe-sdB/OB clusters. In comparison with traditional MK-like classification, we find that the PC clusters are reasonably well separated, with some overlap at cluster boundaries. For very hot sdOs, where the number of standards is small, PC classification has led to a revision of the Drilling MK-like system. Gaia DR3 measurements have been used to determine kinematic parameters for the PC-defined classes. Although disk stars account for a significant fraction of all classes, He-sdOs and Hot-sdOs include a higher fraction of halo stars.