Abstract
Recent surveys are providing a wealth of data about the Universe at large scales. Models of large-scale structure (LSS) formation are crucial for understanding and analyzing this data. The development of relativistic N-body codes has enabled the direct study of relativistic effects in simulations. From the literature, these effects are predicted to be small, but it is not obvious that they can be neglected or added onto Newtonian simulations for the scales approaching the cosmological horizon. I investigate this further by simulating LSS on scales close to the cosmological horizon using both relativistic and Newtonian gravity theories, aiming to find their difference. I propose a method for studying these differences more directly by combining the gravitational potential from two different redshifts. The differences between the simulations are probed using the bispectrum as a higher-order statistic, and machine learning techniques aimed at classifying the simulations both at the field level and by their power spectra. The bispectral analysis shows significant detection of relativistic effects in the equilateral, squeezed and stretched limit for scales approaching the horizon. The machine learning classification shows a ∼95% accuracy for the field-level images and ∼80% accuracy for the power spectra of the same images. This indicates that relativistic effects are present and detectable, and parts of this information are lost when calculating power spectra. The results are promising and support further investigation of the relativistic effects present in large-scale structure formation, especially using higher order statistics and field-level analysis to fully understand these effects.
Professor David F. Mota, Institute of Theoretical Astrophysics, UiO
Professor Julian Adamek, Universität Zürich
Intern. assessor: Professor Boris V. Gudiksen, Institute of Theoretical Astrophysics, UiO
Extern. assessor: Dr. Guilherme Brando de Oliveira, M.P.I Astrophysical and Cosmological Relativity