title: Inferring the assembly and merger histories of galaxies with the IllustrisTNG simulations and machine learning
creator: Eisert, Lukas
subject: ddc-520
subject: 520 Astronomy and allied sciences
description: This thesis presents an investigation into galaxy formation and evolution, utilizing cutting-edge cosmological simulations and machine learning methodologies. Galaxy data from the full cosmological simulations TNG50 and TNG100 within the IllustrisTNG project are employed, and machine learning techniques are trained to extract the assembly and merging history of these simulated galaxies. These machine-learning models can then be applied to observational data, offering a novel method to connect simulations and observations. Initially, this is achieved by using scalars representing integrated observable galaxy features, from which we are able to infer scalars describing the assembly and merging history accurately. However, scalars encompass only a fraction of the complete observational data (images, spectra, IFUs, etc.), and accurately reconstructing scalars from simulations with the same observable bias as observations is not trivial. To address this, contrastive learning is employed to perform representation learning on both survey-realistic mocks of TNG and observed Hyper Suprime-Cam (HSC) image data (in r, g, and i bands), ensuring comparability between simulated and observed images. Remarkably, our findings demonstrate sufficient similarity between the simulated and observed images to justify the idea of simulation-based inference with images. Subsequently, an inference model trained on TNG data is successfully applied to HSC data, allowing the retrieval of information regarding the ex-situ fraction and the time and mass of the last major merger undergone by a galaxy. This interdisciplinary approach merges the domains of cosmological simulations, observational astronomy, and machine learning, offering a new perspective on galaxy formation and evolution. The developed methodologies not only enhance our ability to interpret observational data but also enable the assessment of the realism of cosmological simulations.
date: 2024
type: Dissertation
type: info:eu-repo/semantics/doctoralThesis
type: NonPeerReviewed
format: application/pdf
identifier: https://archiv.ub.uni-heidelberg.de/volltextserverhttps://archiv.ub.uni-heidelberg.de/volltextserver/35162/1/PhD_Thesis.pdf
identifier: DOI:10.11588/heidok.00035162
identifier: urn:nbn:de:bsz:16-heidok-351622
identifier:   Eisert, Lukas  (2024) Inferring the assembly and merger histories of galaxies with the IllustrisTNG simulations and machine learning.  [Dissertation]     
relation: https://archiv.ub.uni-heidelberg.de/volltextserver/35162/
rights: info:eu-repo/semantics/openAccess
rights: http://archiv.ub.uni-heidelberg.de/volltextserver/help/license_urhg.html
language: eng