Abstract

The broad-line region (BLR) is a sub-parsec region in an Active Galactic Nucleus (AGN) where dense, optically thick gas clouds produce broad emission lines (BELs). BELs are important probes of physical processes near the central supermassive black hole (SMBH); however, a complete understanding of the BLR’s structure and dynamics remains lacking. The observed diversity in BEL profile shapes suggests substantial differences in BLR properties across AGN. The availability of large-scale spectroscopic databases provides a valuable opportunity to systematically study these properties in large samples of AGN. They also offer the opportunity to identify rare but important systems, such as SMBH binaries and recoiling SMBHs (RBHs), the anticipated products of galaxy mergers. However, the sheer volume of data requires efficient, robust analytical methods to extract information. This dissertation begins by revisiting SMBHB and RBH candidates and characterizing their BEL profiles using measurements of velocity shifts, asymmetry, and kurtosis. Comparisons with a control sample of typical AGN show substantial overlap between their distributions, indicating that kinematic signatures alone are insufficient to identify SMBHB and RBH candidates within the broader AGN population. This result motivates a more general investigation into the origins of BEL profile diversity by developing a scalable machine learning framework (MLF) for large spectroscopic databases. The MLF combines dimensionality reduction through non-negative matrix factorization, clustering, and spectral modeling with the physically motivated cloud-ensemble photoionization code \texttt{BELMAC}. This integration allows us to gain insights into BLR structure and dynamics by linking ML outputs to physically interpretable BLR dynamics.  Applied to three distinct samples, the MLF reveals five dominant spectral components: two correspond to the core and wings of the BEL profile, one to narrow emission lines, and the remaining two control the continuum slope. Clustering analysis reveals that the primary driver of the observed diversity is the relative strength of the BEL core and wing components. This suggests that the BLR itself may comprise at least two physically distinct components, a scenario previously proposed and now emerging from large-scale analysis. This work demonstrates the feasibility of combining the interpretability of traditional analysis techniques with the scalability and robustness of ML.

Publication Date

8-2026

Document Type

Dissertation

Student Type

Graduate

Degree Name

Astrophysical Sciences and Technology (Ph.D.)

Department, Program, or Center

Physics and Astronomy, School of

College

College of Science

Advisor

Andrew Robinson

Advisor/Committee Member

Moumita Das

Advisor/Committee Member

Jeyhan Kartaltepe

Campus

RIT – Main Campus

Download

Share

COinS