Classification and parameterisation of a large Gaia sample of white dwarfs using XP spectra

TL;DR

This paper presents a large-scale classification and parameter estimation of 100,000 Gaia white dwarfs using XP spectra, machine learning, and spectral fitting to derive their physical properties and analyze the white dwarf population.

Contribution

It introduces a new catalog of white dwarfs with spectral classification and physical parameters derived from Gaia XP spectra, employing machine learning and automated spectral fitting techniques.

Findings

Successfully classified and parameterized 100,000 white dwarfs.

Demonstrated the effectiveness of machine learning in spectral classification.

Provided insights into the properties and distribution of Gaia white dwarfs.

Abstract

The latest Gaia data release in July 2022, DR3, added a number of important data products to those available in earlier releases, including radial velocity data, information on stellar multiplicity and XP spectra of a selected sample of stars. While the normal Gaia photometry (G, GBP and GRP bands) and astrometry can be used to identify white dwarfs with high confidence, it is much more difficult to parameterise the stars and determine the white dwarf spectral type from this information alone. The availability of the XP spectra and synthetic photometry presents an opportunity for more detailed spectral classification and measurement of effective temperature and surface gravity of Gaia white dwarfs. A magnitude limit of G < 17.6 was applied to the routine production of XP spectra for Gaia sources, which would have excluded most white dwarfs. We created a catalogue of 100,000 high-quality white dwarf identifications for which XP spectra were processed, with a magnitude limit of G < 20.5. Synthetic photometry was computed for all these stars, from the XP spectra, in Johnson, SDSS and J-PAS, published as the Gaia Synthetic Photometry Catalogue - White Dwarfs (GSPC-WD). We have now taken this catalogue and applied machine learning techniques to provide a classification of all the stars from the XP spectra. We have then applied an automated spectral fitting programme, with chi-squared minimisation, to measure their physical parameters (effective temperature and log g) from which we can estimate the white dwarf masses and radii. We present the results of this work, demonstrating the power of being able to classify and parameterise such a large sample of 100, 000 stars. We describe what we can learn about the white dwarf population from this data set. We also explore the uncertainties in the process and the limitations of the data set.