Quasars acting as Strong Lenses Found in DESI DR1

TL;DR

The paper presents a data-driven pipeline to identify quasars acting as strong gravitational lenses in DESI DR1, enabling direct estimation of host-galaxy masses via the Einstein radius and helping to probe the redshift evolution of SMBH–host scaling relations. It combines a CNN classifier trained on realistic mock lens spectra (QSO+ELG blends) with two redshift-estimation approaches (Redrock and a regression CNN refined by a Gaussian fit to the $[$OII$]$ doublet), achieving near state-of-the-art performance ($\text{AUC} \approx 0.99$) and robust background redshift recovery across SNR regimes. The methodology yields seven Grade A lens candidates in the blind sample and demonstrates scalability to future spectroscopic surveys, significantly increasing the candidate pool beyond previously known associations. This approach offers a fast, scalable pathway to assemble statistically meaningful samples of QSO lenses for direct host-mass measurements and SMBH–host evolution studies, aided by the $1.5''$ DESI fiber capturing blended foreground and background light.

Abstract

Quasars acting as strong gravitational lenses offer a rare opportunity to probe the redshift evolution of scaling relations between supermassive black holes and their host galaxies, particularly the $M_{\mathrm{BH}}$--$M_{\mathrm{host}}$ relation. Using these powerful probes, the mass of the host galaxy can be precisely inferred from the Einstein radius $θ_{\mathrm{E}}$. Using 812{,}118 quasars from DESI DR1 ($0.03 \leq z \leq 1.8$), we searched for quasars lensing higher-redshift galaxies by identifying background emission-line features in their spectra. To detect these rare systems, we trained a convolutional neural network (CNN) on mock lenses constructed from real DESI spectra of quasars and emission-line galaxies (ELGs), achieving a high classification performance (AUC = 0.99). We also trained a regression network to estimate the redshift of the background ELG. Applying this pipeline, we identified seven high-quality (Grade~A) lens candidates, each exhibiting a strong [O\,\textsc{ii}] doublet at a higher redshift than the foreground quasar; four candidates additionally show H$β$ and [O\,\textsc{iii}] emission. These results significantly expand the sample of quasar lens candidates beyond the twelve identified and three confirmed in previous work, and demonstrate the potential for scalable, data-driven discovery of quasars as strong lenses in upcoming spectroscopic surveys.

Figures (4)

• Figure 1: Diagram of the full pipeline illustrating the sequence of steps used to identify quasar (QSO) lens systems in the DESI DR1 dataset. The process integrates convolutional neural networks for classification, redshift-based selection criteria, and visual inspection to refine candidates based on score thresholds, emission-line strengths, and model fits. Futher explanation for the cuts and inspection can be found in section \ref{['RedshiftTraining']}.
• Figure 2: $F_1$ Score, calculated for different threshold values for both Phase 1 and Phase 2. Our final adopted threshold maximizes the F1 score with an adopted value of 0.70 for Phase 1 and Phase 2.
• Figure 3: True Positive Rate vs False Positive Rate computed on the validation sample.
• Figure 4: True redshifts of the background object $z_{ELG}$ versus the recovered redshifts $z_{\rm True}$ grouped by the SNR of the [OII] line. The shaded region corresponds to a redshift error $\Delta z =0.1$ from the true redshift value.