Student: Kiyam Lin
First supervisors: Benjamin Joachimi (Astro), Jason McEwen (MSSL)
Second supervisor: Andrew Pontzen (Astro)
Entry year: October 2020
Study of the cosmological large-scale structure will yield unique insights into the fundamental physics of the Universe, shedding light on the properties of dark matter and the mass scale of neutrinos, and permitting tests of general relativity and possible extensions of the cosmological standard model. These insights will come from the joint analysis of multiple probes mapping a large fraction of the sky, in particular the distribution of galaxies, gravitational lensing by the large-scale matter distribution, and late-time modifications of the cosmic microwave background.
All of these probes trace the matter distribution indirectly, such that the link between observables and matter has to be modelled jointly with the cosmology. The highly non-linear evolution of structure formation in the late-time Universe complicates matters further: the statistical properties of the observables are generally not known in closed form, and it is not clear which observables saturate the information that can be extracted from the survey maps.
Traditional likelihood analysis approaches struggle in this situation. Therefore, forward-modeling the data with detailed suites of simulations and jointly inferring the statistical properties of the data with the model parameters holds great promise. This PhD project will address the critical elements in this novel approach, creating fast simulations, finessing the choice and compression of observables, and tailoring likelihood-free inference methods. The student will work with state-of-the-art techniques in cosmology, statistics, and machine learning to infer dark matter and dark energy properties from the forthcoming DESI survey and the ESO Kilo-Degree Survey, as well as early LSST data.
The past decade has seen tremendous progress in cosmology, underpinned by an array of precision observational probes, painting a picture of a Universe consisting of roughly 4 per cent ordinary matter, 26 per cent "dark matter" and 70 per cent even more mysterious "dark energy". In this picture, galaxies form and evolve within dark matter halos, which themselves grow via gravitational instability from small density fluctuations present in the early Universe. However, important questions remain about the way dark matter halos emerge from these simple initial seeds of cosmic structure. A theoretical understanding of this process is an essential step towards unravelling the intricate connection between halo and galaxy formation. Recent work has shown that modern machine learning techniques may have the capacity to provide new insights into the process of dark matter halo formation, in some cases contradicting long-established human intuition. This project will develop powerful techniques based on deep-learning algorithms to shed light on the dark matter halo-galaxy connection.
Funding source: "Cosmoparticle" funding from Physics and Astronomy