[Paper Review] The AGEL Survey: Spectroscopic Confirmation of Strong Gravitational Lenses in the DES and DECaLS Fields Selected Using Convolutional Neural Networks

Kim‐Vy Tran, Anishya Harshan+14 more|arXiv (Cornell University)|May 11, 2022

Galaxies: Formation, Evolution, PhenomenaPhysics and Astronomy65 references32 citations

TL;DR

This paper presents spectroscopic confirmation of 68 strong gravitational lenses from the AGEL survey, using convolutional neural networks (CNNs) to identify candidates in deep DES and DECaLS imaging. The method achieves an 88% success rate (68/77 confirmed), with lens systems at higher redshifts (zdeflectors: 0.21–0.89, zsources: 0.88–3.55) than previous surveys, enabling robust studies of mass evolution and galaxy formation.

ABSTRACT

We present spectroscopic confirmation of candidate strong gravitational lenses using the Keck Observatory and Very Large Telescope as part of our ASTRO 3D Galaxy Evolution with Lenses (AGEL) survey. We confirm that 1) search methods using Convolutional Neural Networks (CNN) with visual inspection successfully identify strong gravitational lenses and 2) the lenses are at higher redshifts relative to existing surveys due to the combination of deeper and higher resolution imaging from DECam and spectroscopy spanning optical to near-infrared wavelengths. We measure 104 redshifts in 77 systems selected from a catalog in the DES and DECaLS imaging fields (r<22 mag). Combining our results with published redshifts, we present redshifts for 68 lenses and establish that CNN-based searches are highly effective for use in future imaging surveys with a success rate of 88% (defined as 68/77). We report 53 strong lenses with spectroscopic redshifts for both the deflector and source (z_src>z_defl), and 15 lenses with a spectroscopic redshift for either the deflector (z_defl>0.21) or source (z_src>1.34). For the 68 lenses, the deflectors and sources have average redshifts and standard deviations of 0.58+/-0.14 and 1.92+/-0.59 respectively, and corresponding redshift ranges of (0.21<z_defl<0.89) and (0.88<z_src<3.55). The AGEL systems include 41 deflectors at zdefl>0.5 that are ideal for follow-up studies to track how mass density profiles evolve with redshift. Our goal with AGEL is to spectroscopically confirm ~100 strong gravitational lenses that can be observed from both hemispheres throughout the year. The AGEL survey is a resource for refining automated all-sky searches and addressing a range of questions in astrophysics and cosmology.

Motivation & Objective

• To spectroscopically confirm strong gravitational lens candidates selected via convolutional neural networks (CNNs) in deep DES and DECaLS imaging fields.
• To extend the redshift reach of confirmed strong lenses beyond previous surveys by leveraging deeper, higher-resolution imaging and optical-to-near-infrared spectroscopy.
• To establish a high-purity, spectroscopically confirmed sample of ~100 bright strong lenses (r ≤ 22 mag) observable from both hemispheres year-round.
• To provide a benchmark dataset for refining automated all-sky lens searches and improving CNN-based detection pipelines.
• To enable detailed studies of mass density profiles, dark matter halos, and high-redshift galaxy evolution through spatially resolved multiwavelength follow-up.

Proposed method

• Utilized pre-identified lens candidates from CNN-based searches applied to public DES and DECaLS imaging (r ≤ 22 mag), leveraging high angular resolution and depth.
• Conducted spectroscopic follow-up using the Keck Observatory and Very Large Telescope (VLT) to measure redshifts for both lens deflector and background source components.
• Combined new spectroscopic redshifts with literature values to establish complete redshift measurements for 68 confirmed systems (53 with both components, 15 with one component).
• Validated photometric redshifts from existing surveys (e.g., BPZ) against spectroscopic measurements, finding a mean absolute difference of |Δz| = 0.03 ± 0.02.
• Acquired high-resolution Hubble Space Telescope imaging for a subset of systems to resolve sub-kiloparsec structures and probe dark matter substructure.
• Used a hybrid approach of CNN selection followed by visual inspection and spectroscopic confirmation to ensure high purity and reliability in lens identification.

Experimental results

Research questions

• RQ1 Can convolutional neural networks (CNNs) effectively identify strong gravitational lens candidates in deep optical imaging with high success and purity?
• RQ2 What is the redshift distribution of spectroscopically confirmed strong lenses in the DES and DECaLS fields, and how does it compare to previous surveys?
• RQ3 To what extent do photometric redshifts from existing surveys agree with spectroscopic redshifts for lens deflector and source components?
• RQ4 How do the redshifts and morphologies of AGEL lenses compare to those in earlier surveys like SLACS, BELLS, and CASSOWARY in terms of depth and resolution?
• RQ5 What is the potential of the AGEL sample for probing the evolution of mass density profiles and dark matter halos with cosmic time?

Key findings

• The AGEL survey achieved a spectroscopic confirmation success rate of 88% (68 out of 77 candidates), demonstrating the high reliability of CNN-based lens detection in deep imaging.
• The confirmed lens systems span a wide redshift range: deflector redshifts from zdeflectors = 0.21 to 0.89 (mean: 0.58 ± 0.14), and source redshifts from zsources = 0.88 to 3.55 (mean: 1.92 ± 0.59).
• 41 of the 68 confirmed lenses have deflector redshifts ≥ 0.5, making them ideal for studying the evolution of mass density profiles with redshift.
• Photometric redshifts from existing surveys (e.g., BPZ) show excellent agreement with spectroscopic measurements, with a mean absolute difference of |Δz| = 0.03 ± 0.02.
• The combination of deeper imaging (DECam) and optical-to-near-infrared spectroscopy enabled the detection of higher-redshift systems than previous surveys, including sources at z > 3.
• The AGEL sample includes 53 systems with spectroscopic redshifts for both deflector and source, and 15 with redshifts for only one component, significantly expanding the available high-redshift lensing sample.