Compatibility, complexity and compression: advancing statistical techniques for cosmology
The standard model of cosmology ΛCDM provides an excellent description of a wide range of cosmological observations, yet persistent open questions concerning cosmic acceleration, inflation, and discrepancies in inferred values of the Hubble-Lemaître parameter continue to motivate further investigat...
Guardado en:
| Autor principal: | |
|---|---|
| Formato: | Book/Monograph Tesis |
| Lenguaje: | inglés |
| Publicado: |
Heidelberg
28 Jul. 2026
|
| DOI: | 10.11588/heidok.00039106 |
| Materias: | |
| Acceso en línea: | Resolving-System, kostenfrei: https://nbn-resolving.org/urn:nbn:de:bsz:16-heidok-391062 Resolving-System, kostenfrei: https://doi.org/10.11588/heidok.00039106 Verlag, kostenfrei, Volltext: http://www.ub.uni-heidelberg.de/archiv/39106 Langzeitarchivierung Nationalbibliothek, kostenfrei: https://d-nb.info/1414756704/34 
|
| Notas de Autor: | put forward by Benedikt Schosser ; Betreuer: Björn Malte Schäfer ; referees: Prof. Dr. Björn Malte Schäfer [und ein weiterer Gutachter] |
| Sumario: | The standard model of cosmology ΛCDM provides an excellent description of a wide range of cosmological observations, yet persistent open questions concerning cosmic acceleration, inflation, and discrepancies in inferred values of the Hubble-Lemaître parameter continue to motivate further investigations. Challenging ΛCDM requires reliable parameter inference, consistency checks across multiple experiments, statistical comparison between different models, and trustworthy machine learning methods. This thesis develops information geometric and statistical tools for these stages of modern cosmological inference. First, information geometry is used to quantify the distance between data sets and thereby provide new insight into the Hubble tension. Second, a framework is introduced to sample from the posterior over polynomial dark energy equation of state models through a Markov chain Monte Carlo scheme on the discrete model space, enabling Bayesian inference of models. Third, neural-network-based compression methods are demonstrated to outperform classical approaches for forecasted 21 cm data, yielding constraints on fundamental physics that are competitive with Planck data. Finally, information geometric analyses of neural network latent spaces improve their interpretability and provide insight into the underlying physical processes. Together, these results pave a way to a principled approach for statistical analysis for cosmology and thereby towards resolving the open questions of ΛCDM. |
|---|---|
| Descripción Física: | Online Resource |
| DOI: | 10.11588/heidok.00039106 |