Applying grids of advanced theoretical stellar spectra to observations
Miss Ella Wang1, Dr Thomas Nordlander1,2
1The Australian National University, Canberra, Australia, 2ASTRO 3D, Canberra, Australia
Measuring accurate and precise elemental abundances in stars is necessary to understand the history of the Milky Way. However, these measurements are computationally intensive due to the modelling required. We have computed a grid of synthetic spectra for lithium using advanced 3D non-local thermodynamic equilibrium (NLTE) radiative transfer. Furthermore, we investigate different interpolation methods on this grid so we are able to measure the lithium abundance for any observed Sun-like star. We find that Kriging (a type of Gaussian process) is able to reproduce line shapes better than feedforward neural networks and predicts the lithium abundance to a precision ~3% (0.014 dex), similar to the modelling precision. We create an additional feedforward neural network to interpolate only the line strength (bypassing the line shape) to a precision of 0.01 dex.
Ella is an astrophysics student at the Australian National University and ASTRO3D. They are interested in applications of machine learning to spectral datasets, with a focus on 3D NLTE synthetic spectra and the GALAH and Gaia RVS surveys.