Galactic and Extragalactic Astronomy in the Age of Large Sky Surveys

Author: ORCID icon
Cheng, Xinlun, Astronomy - Graduate School of Arts and Sciences, University of Virginia
Majewski, Steven, AS-Astronomy (ASTR), University of Virginia

Large-scale sky surveys have been and are the backbone of many revolutionary discoveries in astronomy and astrophysics. With huge strides in computer science and statistics, sky surveys are generating ever increasing amounts of data, posing both opportunities and challenges to astronomers in terms of proposing new astrophysical models to explain observed phenomena and machine learning models for data analysis and data mining. In this thesis, I will cover several such areas of progress used in my research. I will first present a characterization of the Galactic warp using the phase space and chemical information contained within millions of stars provided by Gaia and APOGEE, and our physics modeling attempt is the first to discover precession of the warp through stellar kinematics. Next we examine previously adopted assumptions in describing the density and velocity dispersion profile of the Galactic disk against a database formed by combining Gaia and APOGEE database, and we conclude that the traditional analytical methods of interpreting and modeling the stellar kinematics within such data sets are no longer adequate. A similar chemo-dynamical exploration can be applied to nearby galaxies --- including dwarf galaxies --- and we do so in the case of the Large Magellanic Cloud (LMC). Here we discover the existence of a kinematically and chemically distinct population of stars at the southern edge of the LMC. The discovered population has patterns consistent with an origin deriving from an interaction between the LMC, the Small Magellanic Cloud, and the Milky Way. Lastly, we present a novel deep learning technique, contrastive self-supervised learning, applied to mitigate the lack of annotated training set in many astrophysically interesting problems. The pre-trained deep learning model exhibits a strong zero-shot learning capability and has great accuracy after fine-tuning when applied to the problem of searching for white dwarf - main sequence binaries within a large survey database of optical stellar spectra. The research areas covered in this dissertation serve as demonstrations of recent progress gained by the application of new approaches applied to large-scale astronomical survey data.

PHD (Doctor of Philosophy)
Milky Way Galaxy, Galactic Kinematics and Dynamics, Galactic Disk, Sky Surveys, Large Magellanic Cloud, Artificial Intelligence, Deep Learning
Issued Date: