Building a comprehensive picture of stellar death for the era of synoptic surveys

Gagliano, Alexander Thomas

Permalink

https://hdl.handle.net/2142/120099 Copy

Description

Title

Building a comprehensive picture of stellar death for the era of synoptic surveys

Author(s)

Gagliano, Alexander Thomas

Issue Date

2023-04-27

Director of Research (if dissertation) or Advisor (if thesis)

Narayan, Gautham S

Doctoral Committee Chair(s)

Fields, Brian D

Committee Member(s)

• Shen, Yue
• Foley, Ryan

Department of Study

Astronomy

Discipline

Astronomy

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• supernovae
• machine learning
• neural networks
• galaxy surveys
• Vera C. Rubin Observatory
• active learning

Abstract

Nearly a century after their interpretation as the terminal stages of stellar evolution, countless questions still surround the physics powering supernovae. Without the ability to observe a star at the precise moment of its demise, our efforts to trace observed phenomenology back to nature of the terminal progenitor system are limited. This thesis attempts to clarify this connection through a detailed analysis of an explosion’s local environment and signatures of interaction detected within the first few days of an explosion. We emphasize the value of these early-signatures through a comprehensive analysis of the nearby SN Ic 2020oi, and reveal its nature as the detonation of a low-mass (∼9.5 M⊙) binary progenitor enshrouded with ∼0.1 M⊙ of circumstellar material. To quantify the statistical leverage offered by a transient’s host galaxy, we construct the Galaxies HOsting Supernovae and other Transients (GHOST) catalog of 16,175 supernovae and the photometric properties of their associated host galaxies from the first Data Release of the Pan-STARRS 3-π survey. We use a random forest classification model to distinguish between Type-Ia and Type-II SNe with ∼68% accuracy, and release a series of software tools that can be used to improve host-galaxy association at scale. Next, we forecast the host-galaxy correlations that will be revealed in deep upcoming surveys extending to z < 3, and release the Simulated Catalog of Optical Transients and Correlated Hosts (SCOTCH) for benchmarking upcoming classification algorithms. We use this catalog to train a ‘First Impressions’ classifier, consisting of a recurrent neural network trained on synthetic samples and validated on supernovae from the Zwicky Transient Facility Bright Transient Sample (ZTF BTS). This classifier achieves a total accuracy of 83% within the first three days, a first in the literature; at thirty days, the precision and recall achieved are comparable to full-phase networks in the literature with more complex architectures. On the precipice of the Vera C. Rubin Observatory’s unprecedented discovery rates, computational techniques able to leverage physical correlations will be essential for further clarifying progenitor physics and identifying objects of interest for targeted follow-up campaigns.

Graduation Semester

2023-05

Type of Resource

Thesis

Copyright and License Information

Copyright 2023 Alexander Gagliano

Owning Collections