Extreme Binaries in a Target Rich Environment

Mistry, D (2025) Extreme Binaries in a Target Rich Environment. Doctoral thesis, Liverpool John Moores University.

Preview

Text 2025mistryphd.pdf - Published Version Available under License Creative Commons Attribution Non-commercial. Download (19MB) | Preview

Abstract

Cataclysmic Variables (CVs) are a class of binary star systems consisting of a white dwarf accreting matter from a main sequence companion star. The diversity in their physical properties causes them to exhibit a wide range of observable phenomena, including dwarf nova outbursts and novae explosions, making them crucial for the study of binary evolution and accretion physics. The discovery and characterisation of CVs have been greatly facilitated by wide-field time domain surveys, such as the Catalina Real-Time Transient Survey (CRTS) and the Zwicky Transient Facility (ZTF). They can detect significant changes in the brightness of astrophysical objects on various timescales to generate alerts. Due to the continuing advancements in survey technology, alert rates are on the rise, with the Rubin Observatory expected to generate of order 107 alerts per night. Given the large alert rates, classifying these events by the class of astrophysical transient responsible can no longer be solely performed manually. Moreover, follow-up facilities are too few to characterise all events, therefore, follow-up time will be reserved for the rarest of events. Source classification and the search for the rarest of events in this deluge of alerts requires the automation provided by Machine Learning (ML). ML-based source classification is an active research field, with the distinction between many classes of transient possible (e.g., supernovae, active galactic nuclei, and variable star subtypes). However, ML-based searches for CVs is an underdeveloped field; furthermore, an emphasis on the identification/ classification of CV subtypes with ML is unexplored. Given the diversity present within the CV transient class and the under-representation of certain subtypes, such as those with strongly magnetic white dwarfs and ultra-short period helium accreting CVs, a ML-based pipeline specifically purposed for such a task is much needed. The objective of this research has been to address this gap, whilst also identifying the factors that hinder this objective. On this pathway transient sources published by the Gaia Science Alerts program (GSA) were explored with ML. Utilising the technique of light curve feature extraction and with the aid of source metadata from the Gaia survey a ML model based on the Random Forest algorithm was produced. It is capable of distinguishing CVs from supernovae, active galactic nuclei and young stellar objects with a 92% precision score (fraction of those predicted as CV belonging to the class). Of 13,280 sources within GSA without an assigned transient classification, the model predicts the CV class for ∼2800, of which spectroscopic confirmation has been acquired for 15 so far. During the next research phase, the higher cadence, multi-band survey of the Zwicky Transient Facility (ZTF) was explored. A two-stage ML pipeline was developed that comprises and alerts filtering stage aimed at removing non-CVs, followed by an ML classifier tasked with dividing the filtered sources into their CV subtypes based on features extracted from their light curves in combination with Gaia DR3 data. During the month of June 2023 alone, 51 candidates of the CV class were discovered, 14 of which are candidates of either the rare AM CVn or polar CV subtypes. Representations of the ML classifier’s prediction patterns, input into the Generative Topographic Mapping algorithm, indicate the influence of CV evolutionary factors. CV evolution and the consequential blending of boundaries that separate CV subtypes from one another, is found to be a major factor in the difficulty of distinguishing between CV subtypes. To conclude the research, dimensionality reduction techniques were explored with the ZTF dataset. The findings reaffirm the view that CV evolution plays a major factor in the difficulty in distinguishing between subtypes, as do the intricacies of the ZTF survey photometry. In addition, the reduced dimensionality representations were found to be particularly valuable in approximating a subtype classification, with distinct locations of strongly eclipsing CVs as well as polars a particular highlight.

Item Type:	Thesis (Doctoral)
Uncontrolled Keywords:	Machine Learning; Cataclysmic Variables; Time Domain Astronomy
Subjects:	Q Science > QB Astronomy Q Science > QC Physics
Divisions:	Astrophysics Research Institute
Date of acceptance:	4 February 2025
Date of first compliant Open Access:	24 June 2025
Date Deposited:	24 Jun 2025 08:53
Last Modified:	24 Jun 2025 08:53
DOI or ID number:	10.24377/LJMU.t.00025758
Supervisors:	Copperwheat, C, Darnley, M and Olier, I
URI:	https://researchonline.ljmu.ac.uk/id/eprint/25758

View Item