Understanding the origin of structure in the Universe
Lead Research Organisation:
University College London
Department Name: Physics and Astronomy
Abstract
The early universe is a "laboratory" for testing physics at very high energies, up to a trillion times greater than the energies reached by the Large Hadron Collider. The origin of structure in the universe is deeply tied to this extreme physics, which is imprinted in the primordial ripples seen in the cosmic microwave background (the leftover heat of the Big Bang), and the large scale structure of the universe traced by galaxies. The main purpose of the project is to understand the physics of the early universe, creating innovative algorithms to exploit novel observables, and applying them to next generation CMB data from Planck and large scale structure data, especially DES. The student will be part of the European Research Council Starting Grant Project CosmicDawn.
Publications
Lucie-Smith L
(2018)
Machine learning cosmological structure formation
in Monthly Notices of the Royal Astronomical Society
Lucie-Smith Luisa
(2018)
Machine learning cosmological structure formation
in ArXiv e-prints
Studentship Projects
| Project Reference | Relationship | Related To | Start | End | Student Name |
|---|---|---|---|---|---|
| ST/N50449X/1 | 30/09/2015 | 30/03/2021 | |||
| 1641684 | Studentship | ST/N50449X/1 | 30/09/2015 | 31/10/2019 | Luisa Lucie-Smith |
| Description | We developed a novel approach based on machine learning which aims to provide new physical insights into the physics driving halo formation. Dark matter halos are the fundamental building blocks of cosmic large-scale structure. Improving our theoretical understanding of their structure, evolution and formation is an essential step towards understanding how galaxies form, which in turn will allow us to fully exploit the large amount of data from future large-volume galaxy surveys. We trained a machine learning algorithm to learn cosmological structure formation directly from N-body simulation. The algorithm infers the relationship between the initial conditions and the final dark matter halos, based on inputs describing different properties of the local environment surrounding the dark matter particles in the initial conditions. Using this method we are able to investigate which aspects of the early-Universe density field impact the formation of the final dark matter halos. When the algorithm was trained on spherical overdensities around dark matter particles in the initial conditions, we found that it matched predictions based on spherical collapse analytic approximations such as EPS theory. When providing the algorithm with additional information on the tidal shear field (motivated by ellipsoidal collapse approximations such as that of Sheth-Tormen theory), the classification performance of the machine learning was not enhanced. We concluded that the linear density field contains sufficient information to predict the formation of dark matter haloes at the accuracy of existing spherical and ellipsoidal collapse analytic frameworks. This finding allowed for novel insights into the role of the tidal shear field in the formation of dark matter halos, which differ from existing interpretations of the Sheth-Tormen theory. |
| Exploitation Route | In general, our approach can be extended to yield physical understanding of other complex non-linear processes in the context of cosmological structure formation and beyond. We are currently extending our machine learning framework to a setup where we predict individual halo masses and test whether the tidal shear field and other physical properties become informative in predicting final halo mass. In general, our approach can be used outside of academia as a tool to gain understanding of any machine learned mapping. |
| Sectors | Digital/Communication/Information Technologies (including Software) Environment Other |
| Description | We expect further non-academic impact in the coming six months. |
| Description | Andrew Pontzen |
| Organisation | University College London |
| Country | United Kingdom |
| Sector | Academic/University |
| PI Contribution | 2nd supervisor |
| Collaborator Contribution | 2nd supervisor |
| Impact | Paper on "machine learning cosmological structure formation" |
| Start Year | 2015 |
| Description | Hiranya Peiris |
| Organisation | University College London |
| Country | United Kingdom |
| Sector | Academic/University |
| PI Contribution | Primary supervisor |
| Collaborator Contribution | Primary supervisor |
| Impact | Publication of "Machine learning cosmological structure formation" |
| Start Year | 2015 |