🔭 astrophysics

XShooter DESI Lens Program: Sample characterization

This paper presents a VLT/XShooter spectroscopic characterization of 67 lens systems identified by the DESI Legacy Imaging Surveys, successfully measuring redshifts for the majority of lenses and sources to establish a representative distribution for calibrating future large-scale cosmological analyses.

Original authors: Eric Jullo, Christophe Boghossian, Luderic Chapel, Felipe Urcelay, Christopher Storfer, Xiaosheng Huang, Raphael Gavazzi, Jens-Kristian Krogager, Aleksandar Cikota

Published 2026-07-10

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Eric Jullo, Christophe Boghossian, Luderic Chapel, Felipe Urcelay, Christopher Storfer, Xiaosheng Huang, Raphael Gavazzi, Jens-Kristian Krogager, Aleksandar Cikota

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the universe as a giant, cosmic funhouse filled with invisible funhouse mirrors. These aren't the wavy, silly mirrors you find at a carnival; they are massive clusters of galaxies that bend light so powerfully they create "strong gravitational lenses." These lenses stretch and warp the images of distant galaxies behind them, turning them into glowing arcs, rings, or even multiple copies of the same object.

For a long time, finding these cosmic mirages was like trying to spot a specific firefly in a dark forest. But recently, huge sky surveys have started using super-smart computer brains (deep learning) to scan the darkness, finding thousands of these lens candidates. The problem? Just because the computer says, "Hey, that looks like a lens!" doesn't mean we know exactly what it is or how far away it really is. To use these mirrors to measure the secrets of the universe, we need to know the distance to both the mirror (the lens) and the object being reflected (the source).

That's where this paper comes in. The authors, led by Eric Jullo, decided to play detective with a very powerful tool: the XShooter instrument on the Very Large Telescope (VLT) in Chile. They picked 67 of these mysterious candidates that live in the southern sky and pointed the telescope at them to take a "spectral fingerprint." Think of a spectrum as a barcode that tells you exactly how fast an object is moving away from us, which translates directly into its distance.

The Big Catch and the Big Win
The team faced a tricky challenge. Usually, astronomers point their slits at one bright object. But here, they had to stretch their slit across two different galaxies at once: the nearby lens galaxy and the far-away source galaxy. It's like trying to read the license plates on two cars driving at different speeds while they are both zooming past you in the rain. It made the data messy and hard to clean up.

Despite the difficulty, they managed to get a clear "barcode" for 58 of the lens galaxies and 57 of the background source galaxies. That's a success rate of about 82% for the lenses and 85% for the sources. They found that the lens galaxies are mostly relatively close (less than 1 billion light-years away, or redshift $z < 1$ ), while the background sources are much further away, stretching all the way out to $z = 4.34$ . The middle ground for the sources is around $z = 1.54$ .

What They Found (and What They Didn't)
Here is the most exciting part: The authors were worried that maybe they could only measure the distance for the "easy" targets—like the super-bright ones—and that the faint, hard-to-see ones were being left out. If that were true, their map of the universe would be biased, like a census that only counted people wearing bright red hats.

But after checking their data, they found no bias. The distance distribution of the sources they measured looks exactly like the true distribution of the whole population. It's as if they managed to grab a handful of marbles from a jar, and the mix of colors in their hand perfectly matched the mix in the jar. This means the data they collected is a trustworthy representative of the whole group.

They also spotted some cool details in the light:

7 sources showed signs of spinning disks, like a cosmic merry-go-round, visible in the way their light shifted.
2 sources had emission lines that looked like they were being blown away, suggesting powerful outflows of gas.
They even found a few systems where the lens galaxy is actually a pair of galaxies, or where the source is a quasar (a super-bright black hole feeding frenzy).

The Verdict
The paper doesn't claim to have solved the universe's biggest mysteries today. Instead, it does something just as important: it proves that the computer-generated lists of lens candidates are real and that we can reliably measure their distances.

The authors conclude that because they found no bias in their measurements, this new catalog of 57 source redshifts is ready to be used as a "training set" for future, even bigger surveys like Euclid and LSST. It's like they've built a high-quality map of a small neighborhood that proves the GPS algorithms work, so now the big trucks can drive through the whole country with confidence. They haven't found the treasure yet, but they've confirmed the map is accurate enough to start the real hunt.

Technical Summary: XShooter DESI Lens Program: Sample Characterization

Problem Statement
Large-scale cosmological imaging surveys, such as the DESI Legacy Imaging Surveys, are detecting orders of magnitude more strong gravitational lens systems than previously known. While these datasets promise robust constraints on cosmology and galaxy evolution, the utility of these samples depends on the preliminary characterization of their spectroscopic redshifts. Specifically, there is a need to verify whether the redshift distributions of lensed sources derived from these large surveys are representative of the true population or if they suffer from selection biases introduced by the detection algorithms or follow-up observation strategies.

Methodology
The authors conducted a VLT/XShooter observation program targeting 96 "Grade A" lens systems located in the Southern hemisphere (Declination < −20°) from the DESI Legacy Imaging Surveys catalog (Huang et al. 2021; Storfer et al. 2024). These systems were originally identified using deep residual neural networks (ResNet) applied to $grz$ imaging.

Observations: The program utilized the XShooter spectrograph, configured with a $11'' \times 1.6''$ slit to cover both the lens galaxy and the lensed source simultaneously. Observations were conducted in SLIT, NODDING mode with 3 arcsec nodding. Exposure times were set to 1400s for the UVB and VIS arms and split into 5 $\times$ 300s sub-exposures for the NIR arm, totaling approximately 1 hour per Observation Block.
Data Reduction: Data were reduced using the standard ESO reflex pipeline (version 2.11.5) and CPL (version 7.3.2). Custom Python scripts were developed to isolate lens and source fluxes from 2D spectrograms and collapse them into 1D spectra.
Redshift Estimation: Redshifts were estimated manually using the Marz web interface. The authors graded the quality of redshift measurements ( $Q_z$ $Q_{z}$ ) on a scale of 1 to 4:
- $Q_z=4$ : Several emission lines or very good continuum detection.
- $Q_z=3$ : Good continuum detection (e.g., Balmer break, Ca/H absorption).
- $Q_z=2$ : One emission line with uncertain continuum.
- $Q_z=1$ : Very uncertain continuum detection.

Key Results
The program successfully obtained spectroscopic redshifts for 67 observed systems out of the 96 targeted.

Success Rates: Redshifts were measured for 58 lenses (82% efficiency) and 57 sources (85% efficiency). 45 systems yielded redshifts for both the lens and the source.
Redshift Distributions:
- Lenses: Most lens redshifts fall in the range $z < 1$ (median $z \approx 0.5$ ). The majority (35%) were derived from continuum features ( $Q_z=3$ ), while only 12% were based on emission lines ( $Q_z=4$ ). The highest measured lens redshift was $z=1.02$ .
- Sources: Source redshifts span a broad range from $0.28 < z < 4.34$ , with a median of $z_s = 1.54$ . Unlike lenses, 70% of source redshifts were derived from emission lines ( $Q_z=4$ ).
Magnitude Dependence: The success rate for measuring lens redshifts is strongly dependent on magnitude. For lenses with $z_{AB} < 20$ , the success rate is 93%, whereas it drops to 40% for lenses with $18 < z_{AB} < 20$ . Faint lenses ( $z_{AB} > 20$ ) are often satellite galaxies in small groups. In contrast, source redshifts were successfully measured for sources of any magnitude, down to $z_{AB} = 25.2$ .
Physical Characterization: The authors identified specific physical phenomena in a subset of sources:
- Two sources show indications of outflows based on the shape of emission lines.
- Seven sources exhibit rotating disks, identified via velocity gradients in [OII] or H $\alpha$ lines.
Bias Analysis: By comparing the color distributions ( $g-r$ and $r-z$ ) of the measured sample against the parent sample, the authors found no particular bias associated with the redshift measurement operation.

Significance and Claims
The paper claims that the measured source redshift distribution is likely representative of the true distribution of sources in the DESI Legacy Imaging Surveys. Consequently, this dataset can be utilized to:

Calibrate analyses in large imaging surveys.
Serve as an updated training set for other future surveys, specifically Euclid and LSST.
Support forward-model analyses in cosmology and galaxy evolution studies.

The authors emphasize that the lack of bias in their measurement operation validates the use of these spectroscopic follow-ups to characterize the statistical properties of lensed sources detected by deep-learning pipelines in wide-field surveys.

More like this