Robert Bickley

Topic

Deep learning-enabled studies of galaxy mergers and supermassive black hole evolution

Department of Physics and Astronomy

Date & location

• Tuesday, August 13, 2024
• 9:00 A.M.
• Clearihue Building, Room B017

Examining Committee

Supervisory Committee

• Dr. Sara Ellison, Department of Physics and Astronomy, University of Victoria (Supervisor)
• Dr. Luc Simard, Department of Physics and Astronomy, UVic (Member)
• Dr. Hossen Teimoorinia, Department of Physics and Astronomy, UVic (Member)
• Dr. Alexandra Branzan Albu, Department of Electrical and Computer Engineering, UVic (Outside Member)

External Examiner

• Dr. Dale Kocevski, Physics and Astronomy Department, Colby College

Chair of Oral Examination

• Dr. Andrew Wender, Department of Political Science, UVic

Abstract

When the smooth evolution of an isolated galaxy is punctuated by a merger event with a companion of similar mass, theory and observations indicate that a metamorphosis will begin. Dramatic changes in the morphologies and kinematics of merging galaxies are thought to funnel gas towards their centres, leading to elevated star formation rates and supermassive black hole (SMBH) accretion rates. The transformation brought about by mergers appears to be the missing link between the two main types of galaxies – blue star-forming spiral galaxies, and red quiescent elliptical galaxies – observed in the Universe. Simulations predict that galaxies are experiencing the most rapid changes immediately after coalescence (when the merging companions are no longer distinct objects), but observational samples of post-merger galaxies predating this work are generally incomplete (small, and possibly not representative of the post-merger class) or contaminated.

In this work, I present the methodological details of an updated post-merger identification effort using a simulation-trained convolutional neural network (CNN a type of automated machine vision tool) to flag galaxies that are very likely to be post-mergers. I present a proof-of-concept feasibility study using mock observations of simulated galaxies (Chapter 2) before applying the CNN to classify real images of galaxies in the low-redshift Universe (Chapter 3). The CNN classification effort is followed by a manual quality control exercise, which finally leads to the identification of large (with some 100s of galaxies each), pure, and defensible post-merger samples from two different imaging surveys: the Canada France Imaging Survey (CFIS), and the Dark Energy Camera Legacy Survey (DECaLS). With the post-merger samples in hand, I also present on the demographics and evolutionary characteristics of post-merger galaxies using multiple astronomical surveys for multi-wavelength characterization.

I find that star-forming post-mergers are elevated by a factor of ~ 2 in their star formation rates relative to star-forming non-merger galaxies (Chapter 4). I also find that active galactic nuclei (AGN; the observable phenomena associated with SMBH accretion) are more common by a factor of 2–4 in post-mergers compared to non-mergers, and that those AGN appear to be about twice as luminous as AGN in non-mergers (Chapter 5). Finally, I use new X-ray observations from the extended ROentgen Survey with an Imaging Telescope Array (eROSITA) space mission to verify that AGN are unusually common in post-mergers, and to characterize the strength of the connection between mergers, SMBH, and AGN obscuration (Chapter 6). In each result, I also compare the characteristics of the new post-merger samples to statistically identified groups of galaxy pairs that are presumed pre-mergers. Close galaxy pairs are somewhat more likely to experience elevated star formation, SMBH accretion, and obscuration than their isolated peers, but the results for galaxy pairs are generally weaker than for post-mergers.

Together, the results of my studies indicate that the amplitude of transformation seen in post-mergers is unique in the low-redshift Universe. Looking forward, I project the viability of future astronomical surveys for post-merger identification, and find something rather unexpected: while next-generation observatories will offer an opportunity for marginal improvement in identifying the remnants of major galaxy mergers, imaging that is already available (CFIS, DECaLS) is well suited to the task (Chapter 7). I therefore posit that the present generation of astronomers studying galaxy mergers can use forthcoming surveys like Euclid and the Legacy Survey of Space and Time (LSST) to answer more difficult and granular questions about the impact of mergers on galaxy evolution.