On February 5 at 2 PM, Maarja Bussov will defend her PhD dissertation "Clustering analysis for astrophysical structures". A successful defence will gain her a doctoral degree in physics from the University of Tartu.
The defence is held virtually on Zoom. The meeting ID is 922 7772 1302 and passcode 654321.
Prof. Elmo Tempel (Tartu Observatory, University of Tartu), Prof. Radu S. Stoica (University of Lorraine, CNRS, IECL)
Dr. Pekka Heinämäki (Department of Physics and Astronomy, University of Turku)
In this PhD thesis, two classes of astrophysical datasets – large scale galaxy redshift surveys and large supercomputer simulations of fully-kinetic turbulent plasma – are studied with clustering algorithms. In the first part we investigate the most dominant structure element of the Universe: the galaxy filaments. Majority of galaxies in the Universe reside in these galaxy filaments, which are long bridges connecting spherical high-density regions of galaxies and border immense voids almost without galaxies. Mapping the structure from observational galaxy datasets is of utmost importance for understanding the objects residing inside them, that is, galaxies and the intergalactic medium. In this work, we reveal a hidden pattern in the locations of galaxies residing inside these structures, which sheds light on environmental effects governing the evolution of galaxies. Then, we trace the detected galaxy filaments with a new observational dataset of galaxies, and prove the detected network. This motivates the use of these new datasets in the future modeling of the Universe. In the second part of this thesis we study images originating from simulations of turbulent magnetically dominated plasma, which models the physical phenomena observed in galaxy clusters, black hole accretion disks, solar corona, and even in fusion reactors. Physical phenomena responsible for the excitation of particles inside the plasma are not yet fully understood. In order to understand the underlying physics, the physical structures inside the plasma need to be detected. We apply an unsupervised machine learning algorithm on these images; and detect the physical structures pixel-by-pixel, including those responsible for the ejection of particles. We also develop an ensemble framework to improve the accuracy of the results. This thesis demonstrates the great potential and value of clustering analysis tools, from a wide spectrum of concepts, for revealing and understanding physical phenomena.