Emulating non-LTE ultraviolet spectra in late type stars with neural networks
File(s)
Author(s)
Chang, Tze-En
Type
Thesis
Abstract
Magnetic activity in late-type stars drives variability in their ultraviolet output, with significant implications for stellar physics and exoplanetary atmospheres. Accurate modelling of this variability requires non-local thermodynamic equilibrium (NLTE) radiative transfer, but direct calculations are computationally intensive, particularly when applied to three-dimensional magnetohydrodynamic (MHD) simulations. In this thesis, I develop and apply neural network emulators to reproduce NLTE ultraviolet spectra of cool stars, trained on MURaM simulations spanning quiet and magnetically active stellar conditions. The emulators achieve high accuracy across both G- and K-type stars, capturing centre-to-limb variation, facular contrasts, and the spectra of both quiet and active atmospheres, while reducing computational costs by orders of magnitude. Analysis with integrated gradients identifies the temperature stratification as a key contributor to the network predictions, linking the emulation process to the underlying physics. Overall, this work shows that neural networks provide a practical and efficient method for modelling stellar NLTE spectra, enabling systematic exploration of stellar variability and its observational consequences.
Version
Open Access
Date Issued
2025-10-04
Date Awarded
01/02/2026
License URL
Advisor
Unruh, Yvonne
Publisher Department
Department of Physics
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)