Abstract

A large fraction of white dwarfs (between 25-50%) exhibit traces of heavy elements in their atmospheres, likely from the recent or ongoing accretion of rocky or icy extrasolar material. Through spectroscopic observations of these “polluted” white dwarfs, it is possible to infer the bulk composition of their accreted material and learn about extrasolar compositions more broadly. To date, there are more than 1000 known polluted white dwarfs, yet less than a few dozen have been characterised in detail with high-resolution spectroscopy. The scarcity of polluted white dwarfs with well-known photospheric abundances is partly due to the nature of conventional spectral modelling techniques, which typically involve manual, time-intensive, and iterative work, as well as proprietary atmospheric models that are not easily available to the astrophysical community. In our group, we aim to overcome these limitations by developing a fast and reliable Machine Learning (ML) pipeline to accurately determine the main physical and chemical properties of polluted white dwarfs from their spectra. In addition to designing new ML techniques for the study of photospheric pollution, we also seek to expand the population of polluted white dwarfs and increase the number of such objects with precise spectra. To this end, we are leveraging massive databases such as Gaia EDR3 and LAMOST and are acquiring high-resolution data with multiple astronomical facilities, including Magellan/MIKE and Keck/ESI. In this talk, I will give an overview of how polluted white dwarfs can be used to learn about extrasolar compositions, present our LAMOST /Gaia catalogue of polluted white dwarfs, discuss our ML pipeline, and justify how ML tools can open the door to a statistical understanding of extrasolar geochemistry.