Preview

PDF, English Download (32MB) | Terms of use

Abstract

Modelling techniques for stellar atmospheres are undergoing continuous improvement. In this thesis, I showcase how these methods are used for spectroscopic analysis and for modelling time-dependent molecular formation and dissociation. I first use CO5BOLD model atmospheres with the LINFOR3D spectrum synthesis code to determine the photospheric solar silicon abundance of 7.57 ± 0.04. This work also revealed some issues present in the cutting-edge methods, such as synthesised lines being overly broadened. Next, I constructed a chemical reaction network in order to model the time-dependent evolution of molecular species in (carbon-enhanced) metal-poor dwarf and red giant atmospheres, again using CO5 BOLD. This was to test if the assumption of chemical equi librium, widely assumed in spectroscopic studies, was still vaild in the photospheres of metal-poor stars. Indeed, the mean deviations from chemical equilibrium are below 0.2 dex across the spectroscopically relevant regions of the atmosphere, though deviations increase with height. Finally, I implemented machine learning methods in order to remove noise and line blends from spectra, as well as to predict the equilibrium state of a chemical reaction network. The methods used and developed in this thesis illustrate the importance of both conventional and machine learning modelling techniques, and merge them to further improve accuracy, precision, and efficiency.