Spectroscopic Quasar Anomaly Detection (SQuAD) I: Rest-Frame UV Spectra from SDSS DR16

TL;DR

This paper applies anomaly detection algorithms to SDSS DR16 quasar spectra, identifying 1,888 anomalous quasars with diverse spectral features and categorizing them into multiple groups based on emission lines, metallicity, and reddening.

Contribution

It introduces a novel application of PCA and hierarchical K-Means clustering to identify and categorize quasar anomalies in a large spectroscopic dataset, including BAL and reddening features.

Findings

Identified 1,888 anomalous quasars in SDSS DR16.

Categorized anomalies into 10 broad groups based on spectral features.

Provided a value-added catalog of detected anomalies.

Abstract

We present the results of applying anomaly detection algorithms to a quasar spectroscopic sub-sample from the SDSS DR16 Quasar Catalog, covering the redshift range 1.88 < z < 2.47. Principal Component Analysis (PCA) was employed for dimensionality reduction of the quasar spectra, followed by hierarchical K-Means clustering in a 20-dimensional PCA eigenvector hyperspace. To prevent broad absorption line (BAL) quasars from being identified as the primary anomaly group, we conducted the analysis with and without them, comparing both datasets for a clearer identification of other anomalous quasar types. We identified 1,888 anomalous quasars, categorized into 10 broad groups. The anomalous groups include C IV Peakers-quasars with extremely strong and narrow C IV emission lines; Excess Si IV emitters-quasars where the Si IV line is as strong as the C IV line; and Si IV Deficient anomalies, which exhibit significantly weaker Si IV emission compared to typical quasars. The anomalous nature of these quasars is attributed to lower Eddington ratios for C IV Peakers, super-solar metallicity for Excess Si IV emitters, and sub-solar metallicity for Si IV Deficient anomalies. Additionally, we identified four groups of BAL anomalies: Blue BALs, Flat BALs, Reddened BALs, and FeLoBALs, distinguished primarily by the strength of reddening in these sources. Further, among the non-BAL quasars, we identified three types of reddened anomaly groups classified as heavily reddened, moderately reddened, and plateau-shaped spectrum quasars, each exhibiting varying degrees of reddening. The detected anomalies are presented as a value-added catalog.