Models for Stellar Spectroscopy

Most of the light that we observe comes from stars. Our knowledge about the Universe thus depends critically on an understanding of the properties of stars. Stellar spectroscopic modelling involves interpreting stellar spectra to extract information about stellar physical properties, including atmospheric temperature, elemental abundances, mass and radius. In order to extract this information, precise theoretical and computational models are required. These computational models can require large amounts of computing power to solve, and in the past, simplifying assumptions (such as local thermodynamic equilibrium in stellar atmospheres and spherical symmetry) have been made in order to render the problem tractable. The Astrophysical Spectroscopy and Stellar Populations group works on developing sophisticated computational models that go beyond these simplifying assumptions, in order to improve the accuracy of stellar atmospheric models. The Milky Way group works on an alternative, purely empirical approach, known as data-driven spectroscopy, in which machine-learning models are first trained using datasets of thousands of stars with known physical parameters, and then applied to determine the parameters of millions of stars with observed spectra.

The work described above is enabled by astrometric, spectroscopic and photometric surveys, which have vastly increased the number of stars with measured positions, velocities, spectra and fluxes over the past decade.