Guo, Ningyuan;
(2024)
Deep learning insights into dark matter halo formation.
Doctoral thesis (Ph.D), UCL (University College London).
Preview |
Text
GuoNingyuan-PhDThesis-accepted_version.pdf - Accepted Version Download (4MB) | Preview |
Abstract
Unprecedented amounts of high-quality data will become available in the coming years from surveys such as Euclid and Rubin. They will allow stringent tests of the cosmological model and shed light on the nature of dark energy and dark matter. Understanding what determines properties of dark matter halos, the building blocks of cosmic structure, is crucial to developing robust and accurate theoretical models of quantities needed for cosmological inference, such as the halo mass function (HMF). This understanding is made difficult by the highly non-linear nature of halo formation. This thesis develops a novel approach to gaining physical insights on dark matter halo formation using deep learning. The approach takes advantage of deep learning models’ ability to learn highly non-linear mappings, and combines this with techniques to interpret the model to extract new physical understanding. The approach is first developed using a synthetic image dataset: we investigate the use of an encoder-decoder architecture to learn an interpretable low-dimensional latent representation that captures underlying factors generating the dataset. We also build upon the novel approach of using mutual information to interpret the latent representation in a way that enables discovering physical factors not known a priori. The framework is then used to determine the information required to model the present-day HMF to the percent-level accuracy required by forthcoming surveys over a wCDM+N_eff parameter space. We find that three independent latent variables capture all relevant information. Interpreting these variables reveals that in addition to mass variance, the recent linear growth history and N_eff are also needed to accurately predict the HMF. The framework can be extended to additionally model the HMF’s redshift evolution. The latent variables our framework discovers can be used to improve halo mass function modelling over a large cosmological parameter space and redshift range to meet future survey demands.
| Type: | Thesis (Doctoral) |
|---|---|
| Qualification: | Ph.D |
| Title: | Deep learning insights into dark matter halo formation |
| Open access status: | An open access version is available from UCL Discovery |
| Language: | English |
| Additional information: | Copyright © The Author 2024. Original content in this thesis is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) Licence (https://creativecommons.org/licenses/by-nc/4.0/). Any third-party copyright material present remains the property of its respective owner(s) and is licensed under its existing terms. Access may initially be restricted at the author’s request. |
| UCL classification: | UCL UCL > Provost and Vice Provost Offices > UCL BEAMS UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of Maths and Physical Sciences UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of Maths and Physical Sciences > Dept of Physics and Astronomy |
| URI: | https://discovery-pp.ucl.ac.uk/id/eprint/10202670 |
Archive Staff Only
 |
View Item |
