Séminaire Univers /
Seminar Universe

The Lyman alpha forest is one of the most powerful probes of the matter distribution on small scales and at high redshift. Extracting cosmological information from it, however, requires accurate modelling not only of the forest itself but also of the contaminants imprinted on the same spectra. Two of the most important are metal absorption lines and high-column-density (HCD) absorbers, both of which can bias cosmological inferences if not properly accounted for. In the first part of this talk, I will give an introduction to the Lyman alpha forest and how it is modelled. I will then discuss my work on metal contamination, and recent work using the IllustrisTNG simulations to characterise HCD systems from a halo-centred perspective (https://arxiv.org/abs/2609.19235). In the second part of this talk, I will turn to a broader challenge: how to robustly connect cosmological simulations with real observations when simulations are imperfect representations of the data. I will present machine learning approaches to tackle this problem, including domain adaptation and contrastive learning, and discuss how they can help build models that generalise from simulations to observations.