Pith. sign in

REVIEW 3 major objections 5 minor 64 references

The revolution in strong lensing discoveries from Euclid

T0 review · 3 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Euclid's first quick data release yields about 500 high-quality strong lens candidates, and the search pipeline projects 15,000 lenses by DR1 and roughly 110,000 over the full mission.

desk verdict A clear collaboration overview of Euclid Q1 strong lens results, where the real new content is the area-scaled forecasts; those forecasts need statistical backing before they are quoted as fact. read the letter →

arxiv 2508.14624 v1 pith:GQ7RT3IN submitted 2025-08-20 astro-ph.GA

Natalie E. P. Lines , Tian Li , Thomas E. Collett , Philip Holloway , James W. Nightingale , Karina Rojas , Aprajita Verma , Mike Walmsley This is my paper
classification astro-ph.GA
keywords stronggravitationallensingEuclidQ1machinelearningcitizensciencegalaxy-galaxylenseslenscandidatesEinsteinringssurveyastronomy
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper reports the first strong-lens harvest from Euclid's Quick Data Release 1 (Q1): about 500 high-quality galaxy-galaxy strong lens candidates in 63 square degrees, found by ranking a million objects with five machine-learning models, having citizen scientists check 100,000 cutouts, and having experts grade the survivors. The authors argue that this proves the search strategy works at Euclid scale: the best model ranks lenses so efficiently that the top million candidates in the full survey would contain the great majority of lenses. Scaling the Q1 result by area, they forecast 15,000 strong lenses in the first major data release and roughly 110,000 over the six-year mission, an increase of two orders of magnitude over the few thousand lenses known today. If correct, this turns strong lensing from a sparse-data field into a statistical science for dark matter, galaxy evolution, and cosmology.

What carries the argument

The machinery is a three-stage discovery pipeline: machine-learning ranking of lens candidates (several network architectures, led by a model fine-tuned from a galaxy-morphology pre-trained network), citizen-science review of the top-ranked cutouts, and expert grading of the survivors, with automated lens modelling as an independent confirmation step. The load-bearing element is the ranking purity: the false positives are mostly chance alignments of elongated galaxies rather than spiral or ring impostors, so the top of the ranking is dense with true lenses. That purity is what lets the team extrapolate from 100,000 inspected cutouts in Q1 to a top-million inspection over the full mission.

What would settle it

Take one or more full Euclid tiles from Q1 and have expert astronomers visually inspect every source in them, without using the machine-learning ranking; if that complete census finds substantially more high-quality lenses than the ranked pipeline found in the same tiles, the completeness assumption behind the area-scaled forecasts fails. A complementary test is to obtain spectra for all 500 candidates and measure how many show multiple images at the same redshift.

Watch

Extended reading notes

Core claim

The central claim is that Euclid's quick-release data already contain the largest high-resolution sample of strong lens candidates ever assembled, and that the discovery pipeline is mature enough to extrapolate to the full survey. In Q1, a 63 square degree region with 30 million sources, the combination of machine-learning ranking, crowd-sourced visual inspection, and expert grading produced about 500 high-quality candidates, nearly all new. Automated mass modelling of over 300 systems reconstructed the background sources as coherent single objects, indicating they are almost certainly real lenses. Because the false-positive population is dominated by chance alignments rather than by spiral

Load-bearing premise

The whole projection rests on Q1's 63 square degrees being a representative, near-complete sample of the full Euclid survey, so that lens counts scale linearly with area; if Q1 missed a meaningful fraction of lenses or sits in an atypical patch of sky, the 15,000 and 110,000 numbers shrink.

Editorial extensions

If this is right

  • The catalogue will contain roughly 15,000 strong lenses in Euclid's first major data release (DR1, public in late 2026), and about 110,000 over the full mission if Q1 is representative.
  • The full-sample top-million visual inspection is projected to yield about 75,000 new lenses, a conservative lower limit that combining multiple models can raise.
  • Around 1,000 compound lenses and 2,500 dark-matter subhalo detections are expected from the full survey, enabling statistical tests of dark-matter models.
  • The rare classes found despite no targeted training—compound and edge-on disk lenses—indicate the pipeline generalizes to lens morphologies it was not built to find.
  • With this sample size, strong lensing can move from individual-object studies to population constraints on galaxy mass profiles, the initial mass function, cosmic shear, and dark energy, including time-delay cosmography with lensed supernovae and quasars.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the Q1 selection is representative, the bottleneck shifts from discovery to confirmation: spectroscopic follow-up of tens of thousands of candidates will require prioritization, since Euclid's own spectra are too shallow to measure most lens redshifts.
  • The discovery of edge-on disk and compound lenses without targeted training suggests the pre-trained morphology model generalizes to rare lens shapes; the same pipeline may surface other untrained rare classes once a dedicated search is added.
  • The false-positive mix, dominated by chance alignments, may change with sky depth and environment, so periodic retraining on newly confirmed lenses will be needed to hold the 110,000 projection.
  • If the 110,000 forecast holds, time-delay cosmography with lensed supernovae and quasars becomes a statistical sample rather than a handful of systems, changing how the Hubble-constant tension is tested when paired with a wide, fast-cadence ground-based survey.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper reports on the strong-lens search in Euclid's first quick data release (Q1, 63 deg²), claiming about 500 high-quality strong-lens candidates discovered through a combination of five machine-learning models, a Space Warps citizen-science campaign, and expert visual inspection. It highlights rare systems (compound lenses, edge-on disk lenses) and states that automated lens modelling with PyAutoLens confirms most candidates. The central forward-looking claim, made in the Abstract and the Future section, is that scaling Q1 results by area yields 15,000 lenses in DR1 and ~110,000 by the end of the mission, with an additional estimate of 75,000 lenses if the top million ranked objects in the full survey are inspected. The paper argues that Euclid will increase the number of known strong lenses by two orders of magnitude, enabling new science.

Significance. If the pipeline and the projections hold, this work represents a major milestone: a large, high-purity sample of galaxy-scale strong lenses with Euclid's 0.16″ resolution, on a path to sample sizes that are unprecedented in the field. The paper's strengths are its concrete integration of ML ranking, citizen-science triage, and expert vetting, and its emphasis on rare and previously underrepresented lens populations. The headline numbers are falsifiable and will be tested by DR1. However, the significance hinges on the validity of the area-scaling extrapolation and on the internal completeness/purity estimates, which are not derived in this paper but are inherited from companion papers by the same collaboration. As a standalone manuscript, it does not provide the statistical support needed to independently substantiate the '110,000 strong lenses' forecast.

major comments (3)
  1. [Future] The projections '15 000 strong lenses in DR1' and '110 000 by the end of the full mission' rest entirely on the statement 'Scaling up the results from Q1 by area.' No statistical derivation, uncertainty budget, or discussion of Q1's representativeness is given. The paper does not state where Q1 lies (Galactic latitude, extinction, PSF properties) or how the field was selected relative to the full Euclid Wide Survey. A 20% systematic difference in lens surface density would change the full-mission number by ~22,000, which is material to the 'revolution' claim. Please provide the derivation behind these numbers, give uncertainties, and explicitly discuss whether Q1 is a fair sample of the survey.
  2. [Future] The sentence about retraining—'preliminary retraining efforts using the Q1 lenses and false positives have already produced a model that can discover the first quarter of the lenses in a Q1-size sample with a purity of 94%'—needs a description of the evaluation protocol. If the test set is the same Q1 sample that supplied the training examples, or a subset selected by the same ML rankings, the 94% purity is vulnerable to circularity. Please state the test-set construction, whether the Q1 catalogue of 500 lenses was used as ground truth, and how contamination from unconfirmed candidates was handled. This result is load-bearing for the claim that future searches can overcome the visual-inspection bottleneck.
  3. [500 strong lenses found] The completeness claim—'a statistical analysis based on a representative sample of the Q1 data suggests that almost all of the highest-quality strong lenses were found'—is a keystone for the area-scaling argument, but it is not supported in this paper. The analysis is in the companion paper [54], which uses the same models and the same Q1 data. Even if that analysis is valid for Q1, it establishes only completeness within Q1, not representativeness of Q1 for the full survey. Please either (a) state the completeness fraction and the method used to estimate it, so the reader can judge it here, or (b) clearly label the 15,000/110,000 numbers as collaboration-internal forecasts that depend on the untested assumption that Q1 is representative.
minor comments (5)
  1. [Authors] Affiliation 4 appears twice in the author list; the duplicated School of Engineering entry should be removed.
  2. [Figure 2 caption] 'James Web Space Telescope' should be 'James Webb Space Telescope'; also 'PanSTARSS' should be 'Pan-STARRS.'
  3. [The challenge] The sentence 'a network could classify it either as ‘lens’ as ‘non-lens’' appears to have a typo; it should read 'as either “lens” or “non-lens.”'
  4. [Introduction] The statement 'only 1 in roughly 10 000 massive galaxies can lens a background source into multiple images' is given without a citation. Please add a reference to [25] or a related population estimate.
  5. [Figure 2] The axis labels in the figure are typeset with unusual spacing ('1 0 0 1 0 1 ...'); this is likely a rendering artifact, but the published version should use standard scientific notation.

Circularity Check

1 steps flagged · score 5.0 of 10

One supporting ML performance claim is circular as written; the core area-scaled forecast is an extrapolation, not a circular derivation.

  1. fitted input called prediction [Future section, first paragraph (retraining claim)]
    "preliminary retraining efforts using the Q1 lenses and false positives have already produced a model that can discover the first quarter of the lenses in a Q1-size sample with a purity of 94%, doubling the purity compared to the previous version of the network."

    The model is retrained on Q1 lens and false-positive labels, and the reported metric is discovery performance on a Q1-size sample. The text does not state that this sample is disjoint from the training labels. As written, the 94% purity could measure the model's ability to re-identify objects it was trained on, not its ability to generalize to unseen Euclid data. The paper then uses this number to argue that 'the rate of improvement in machine learning performance can make up for limitations in the data volume that can be visually inspected,' so a training-set performance statistic is being presented as predictive evidence.

full rationale

The paper's central projection—15,000 lenses in DR1 and 110,000 for the full mission—is an area extrapolation from the Q1 count (500 lenses over 63 deg^2). That is arithmetic scaling, not a circular derivation: the Q1 count is an observed input and the survey area is a known mission parameter. The main weakness of this projection is the unstated assumption that Q1 is representative of the full Euclid Wide Survey; this is a validity risk, not a circularity. The mission forecast from Collett (2015) is an independent a priori prediction, and confirming it with Q1 data is legitimate even though the author is a co-author. The completeness and purity estimates are cited to companion Euclid papers by overlapping authors, but those are data papers rather than bare assertions, so self-citation alone is not the circularity. The one genuine circular step is the retraining claim in the Future section: a model trained on Q1 labels is evaluated on a Q1-size sample with no stated held-out split, so the reported 94% purity is consistent with measuring rediscovery of the training data. Because this step is secondary to the area-scaled forecast, the overall circularity is moderate rather than fundamental.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The paper itself performs no new derivation; its projections rest on the transfer of completeness and purity estimates from companion Q1 papers and on an implicit assumption that Q1 is representative. The ledger records the explicit and implicit inputs that the headline numbers depend on.

free parameters (2)
  • Q1 lens surface density = ~7.9 lenses per square degree (500 lenses / 63 deg^2)
    Used as the basis for the 15,000 and 110,000 projections; it is measured from a non-random visual sample and has no quoted uncertainty in this paper.
  • Top-million rank threshold = 1,000,000 images
    Chosen by hand in the Future section to estimate 75,000 lenses; the resulting count depends on this cutoff and on the ML model's purity at that rank, neither of which is derived here.
assumptions (4)
  • domain assumption Q1 data are representative of the full Euclid survey and lens density scales linearly with area.
    Future section: 'Scaling up the results from Q1 by area, Euclid is on track to discover 15,000 strong lenses...'. No treatment of field selection, depth variation, or survey coverage is given.
  • domain assumption Expert visual inspection plus PyAutoLens SIE and shear fits certify the 500 objects as strong lenses.
    Q1 section: automated modelling on over 300 systems indicates they are almost certainly lenses, but many candidates require spectroscopy. The catalogue is treated as strong lenses despite candidate status.
  • domain assumption The completeness estimate from companion paper [54], based on a representative Q1 sample, is valid.
    500 strong lenses found section: 'a statistical analysis based on a representative sample of the Q1 data suggests that almost all of the highest-quality strong lenses were found [54]'; the method is not reproduced here.
  • domain assumption The rarity rate and Euclid survey parameters from [25] and [26] are correct.
    Introduction uses 1 in 10,000 massive galaxies and 0.16 arcsecond resolution to frame the forecast; these are accepted inputs from previous work, not checked here.

how reviews work

0 comments
Cite this review

Pith. "Pith review of The revolution in strong lensing discoveries from Euclid." pith.science (2026). https://pith.science/paper/GQ7RT3IN

@misc{pith2026250814624,
  author       = {Pith},
  title        = {Pith review of: The revolution in strong lensing discoveries from Euclid},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GQ7RT3IN}},
  note         = {Machine review of arXiv:2508.14624}
}
read the original abstract

Strong gravitational lensing offers a powerful and direct probe of dark matter, galaxy evolution and cosmology, yet strong lenses are rare: only 1 in roughly 10,000 massive galaxies can lens a background source into multiple images. The European Space Agency's Euclid telescope, with its unique combination of high-resolution imaging and wide-area sky coverage, is set to transform this field. In its first quick data release, covering just 0.45% of the full survey area, around 500 high-quality strong lens candidates have been identified using a synergy of machine learning, citizen science and expert visual inspection. This dataset includes exotic systems such as compound lenses and edge-on disk lenses, demonstrating Euclid's capacity to probe the lens parameter space. The machine learning models developed to discover strong lenses in Euclid data are able to find lenses with high purity rates, confirming that the mission's forecast of discovering over 100,000 strong lenses is achievable during its 6-year mission. This will increase the number of known strong lenses by two orders of magnitude, transforming the science that can be done with strong lensing.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

64 extracted references · 53 canonical work pages

  1. [27]

    Euclid Collaboration: Walmsley, M. et al. Euclid Quick Data Release (Q1): The Strong Lensing Discovery Engine A – System overview and lens catalogue. arXiv e-prints arXiv:2503.15324 (2025)

    arXiv 2025

  2. [54]

    Euclid Collaboration: Lines, N.E.P. et al. Euclid Quick Data Release (Q1). The Strong Lensing Discovery Engine C – Finding lenses with machine learning.arXiv e-prints arXiv:2503.15326 (2025)

    arXiv 2025

  3. [59]

    Euclid Collaboration: Li, T. et al. Euclid Quick Data Release (Q1). The Strong Lensing Discovery Engine D – Double-source-plane lens candidates.arXiv e-prints arXiv:2503.15327 (2025)

    arXiv 2025

  4. [61]

    Euclid Collaboration: Holloway, P. et al. Euclid Quick Data Release (Q1). The Strong Lensing Discovery Engine E – Ensemble classification of strong grav- itational lenses: lessons for Data Release 1. arXiv e-prints arXiv:2503.15328 (2025)

    arXiv 2025

  5. [1]

    Nebulae as Gravitational Lenses

    Zwicky, F. Nebulae as Gravitational Lenses. Phys. Rev.51, 290–290 (1937)

    work page 1937

  6. [2]

    Lens-Like Action of a Star by the Deviation of Light in the Gravitational Field

    Einstein, A. Lens-Like Action of a Star by the Deviation of Light in the Gravitational Field. Science 84, 506–507 (1936)

    work page 1936

  7. [3]

    O’Riordan, C.M. et al. Euclid: A complete Einstein ring in NGC 6505. Astron. Astrophys. 694, A145 (2025)

    work page 2025

  8. [4]

    J., Treu, T., Birrer, S

    Shajib, A. J., Treu, T., Birrer, S. & Sonnenfeld, A. Dark matter haloes of massive elliptical galaxies at z∼ 0.2 are well described by the Navarro-Frenk-White profile. Mon. Not. R. Astron. Soc.503, 2380–2405 (2021)

    work page 2021

Show all 64 references
  1. [5]

    B., Ellis, R

    Newman, A. B., Ellis, R. S. & Treu, T. Luminous and Dark Matter Profiles from Galaxies to Clusters: Bridging the Gap with Group-scale Lenses. Astrophys. J. 814, 26 (2015)

    2015

  2. [6]

    & Cautun, M

    Sonnenfeld, A. & Cautun, M. Statistical strong lensing. I. Constraints on the inner structure of galaxies from samples of a thousand lenses. Astron. Astrophys. 651, A18 (2021). 11

    2021

  3. [7]

    Wang, K. et al. Measuring the Stellar-to-Halo Mass Relation at ∼ 1010 Solar masses, using space-based imaging of galaxy-galaxy strong lenses. arXiv e-prints arXiv:2501.16139 (2025)

    2025

  4. [8]

    Nightingale, J. W. et al. Abell 1201: detection of an ultramassive black hole in a strong gravitational lens. Mon. Not. R. Astron. Soc.521, 3298–3322 (2023)

    2023

  5. [9]

    Melo-Carneiro, Carlos R. et al. Unveiling a 36 Billion Solar Mass Black Hole at the Centre of the Cosmic Horseshoe Gravitational Lens. arXiv e-prints arXiv:2502.13788 (2025)

    2025 arXiv

  6. [10]

    B., Fleury, P., Larena, J

    Hogg, N. B., Fleury, P., Larena, J. & Martinelli, M. Measuring line-of-sight shear with Einstein rings: a proof of concept.Mon. Not. R. Astron. Soc.520, 5982–6000 (2023)

    2023

  7. [11]

    & Amara, A

    Birrer, S., Refregier, A. & Amara, A. Cosmic Shear with Einstein Rings. Astrophys. J. Lett.852, L14 (2018)

    2018

  8. [12]

    Etherington, A. et al. Strong gravitational lensing’s ’external shear’ is not shear. Mon. Not. R. Astron. Soc.531, 3684–3697 (2024)

    2024

  9. [13]

    E., Krawczyk, C

    Li, T., Collett, T. E., Krawczyk, C. M. & Enzi, W. Cosmology from large pop- ulations of galaxy-galaxy strong gravitational lenses. Mon. Not. R. Astron. Soc. 527, 5311–5323 (2024)

    2024

  10. [14]

    Vegetti, S., Koopmans, L. V. E., Bolton, A., Treu, T. & Gavazzi, R. Detection of a dark substructure through gravitational imaging. Mon. Not. R. Astron. Soc. 408, 1969–1981 (2010)

    1969

  11. [15]

    Vegetti, S. et al. Gravitational detection of a low-mass dark satellite galaxy at cosmological distance. Nature 481, 341–343 (2012)

    2012

  12. [16]

    Hezaveh, Y.D. et al. Detection of Lensing Substructure Using ALMA Observa- tions of the Dusty Galaxy SDP.81. Astrophys. J.823, 37 (2016)

    2016

  13. [17]

    & Vegetti, S

    Minor, Q., Gad-Nasr, S., Kaplinghat, M. & Vegetti, S. An unexpected high con- centration for the dark substructure in the gravitational lens SDSSJ0946+1006. Mon. Not. R. Astron. Soc.507, 1662–1683 (2021)

    2021

  14. [18]

    J., Enzi, W

    Ballard, D. J., Enzi, W. J. R., Collett, T. E., Turner, H. C. & Smith, R. J. Gravitational imaging through a triple source plane lens: revisiting the ΛCDM- defying dark subhalo in SDSSJ0946+1006. Mon. Not. R. Astron. Soc.528, 7564– 7586 (2024)

    2024

  15. [19]

    Despali, G. et al. Detecting low-mass haloes with strong gravitational lensing II: constraints on the density profiles of two detected subhaloes. arXiv e-prints arXiv:2407.12910 (2024). 12

    2024 arXiv

  16. [20]

    Enzi, W. J. R., Krawczyk, C. M., Ballard, D. J. & Collett, T. E. The overcon- centrated dark halo in the strong lens SDSS J0946+1006 is a subhalo: evidence for self interacting dark matter? Mon. Not. R. Astron. Soc.(2025)

    2025

  17. [21]

    & Marshall, P

    Orban de Xivry, G. & Marshall, P. An atlas of predicted exotic gravitational lenses. Mon. Not. R. Astron. Soc.399, 2–20 (2009)

    2009

  18. [22]

    Treu, T. et al. The SWELLS survey - I. A large spectroscopically selected sample of edge-on late-type lens galaxies. Mon. Not. R. Astron. Soc.417, 1601–1620 (2011)

    2011

  19. [23]

    Acevedo Barroso, J.A. et al. Searching for strong lensing by late-type galaxies in UNIONS. arXiv e-prints arXiv:2503.10610 (2025)

    2025

  20. [24]

    Wong, K.C. et al. H0LiCOW - XIII. A 2.4 per cent measurement of H 0 from lensed quasars: 5.3σ tension between early- and late-Universe probes. Mon. Not. R. Astron. Soc.498, 1420–1439 (2020)

    2020

  21. [25]

    Collett, T. E. The Population of Galaxy-Galaxy Strong Lenses in Forthcoming Optical Imaging Surveys. Astrophys. J.811, 20 (2015)

    2015

  22. [26]

    Euclid Collaboration: Mellier, Y. et al. Euclid: I. Overview of the Euclid mission. Astron. Astrophys.697, A1 (2025)

    2025

  23. [28]

    Nightingale, J. et al. The COSMOS-Web Lens Survey (COWLS) I: Discovery of >100 high redshift strong lenses in contiguous JWST imaging. arXiv e-prints arXiv:2503.08777 (2025)

    2025 arXiv

  24. [29]

    Mahler, G. et al. The COSMOS-Web Lens Survey (COWLS) II: depth, resolu- tion, and NIR coverage from JWST reveal 17 spectacular lenses. arXiv e-prints arXiv:2503.08782 (2025)

    2025

  25. [30]

    Faure, C. et al. First Catalog of Strong Lens Candidates in the COSMOS Field. Astrophys. J. Suppl. Ser.176, 19–38 (2008)

    2008

  26. [31]

    J., Treu, T

    Gavazzi, R., Marshall, P. J., Treu, T. & Sonnenfeld, A. RINGFINDER: Automated Detection of Galaxy-scale Gravitational Lenses in Ground-based Multi-filter Imaging Data. Astrophys. J.785, 144 (2014)

    2014

  27. [32]

    Sonnenfeld, A. et al. Survey of Gravitationally-lensed Objects in HSC Imaging (SuGOHI). I. Automatic search for galaxy-scale strong lenses. Publ. Astron. Soc. Jpn 70, S29 (2018)

    2018

  28. [33]

    Sonnenfeld, A. et al. Survey of Gravitationally-lensed Objects in HSC Imaging 13 (SuGOHI). VI. Crowdsourced lens finding with Space Warps. Astron. Astrophys. 642, A148 (2020)

    2020

  29. [34]

    Ca˜ nameras, R. et al. HOLISMOKES. VI. New galaxy-scale strong lens candidates from the HSC-SSP imaging survey. Astron. Astrophys.653, L6 (2021)

    2021

  30. [35]

    Shu, Y. et al. HOLISMOKES. VIII. High-redshift, strong-lens search in the Hyper Suprime-Cam Subaru Strategic Program. Astron. Astrophys.662, A4 (2022)

    2022

  31. [36]

    Wong, K.C. et al. Survey of Gravitationally lensed objects in HSC Imaging (SuGOHI). VIII. New galaxy-scale lenses from the HSC SSP. Publ. Astron. Soc. Jpn 74, 1209–1219 (2022)

    2022

  32. [37]

    Jaelani, A.T. et al. Survey of gravitationally lensed objects in HSC imaging (SuGOHI) - X. Strong lens finding in the HSC-SSP using convolutional neural networks. Mon. Not. R. Astron. Soc.535, 1625–1639 (2024)

    2024

  33. [38]

    Schuldt, S. et al. HOLISMOKES: XIII. Strong-lens candidates at all mass scales and their environments from the Hyper-Suprime Cam and deep learning. Astron. Astrophys. 693, A291 (2025)

    2025

  34. [39]

    Schuldt, S. et al. HOLISMOKES XVI: Lens search in HSC-PDR3 with a neural network committee and post-processing for false-positive removal. arXiv e-prints arXiv:2503.07733 (2025)

    2025 arXiv

  35. [40]

    Jacobs, C. et al. An Extended Catalog of Galaxy-Galaxy Strong Gravitational Lenses Discovered in DES Using Convolutional Neural Networks. Astrophys. J. Suppl. Ser. 243, 17 (2019)

    2019

  36. [41]

    Rojas, K. et al. Search of strong lens systems in the Dark Energy Survey using convolutional neural networks. Astron. Astrophys.668, A73 (2022)

    2022

  37. [42]

    Gonzalez, J. et al. Discovering Strong Gravitational Lenses in the Dark Energy Survey with Interactive Machine Learning and Crowd-sourced Inspection with Space Warps. arXiv e-prints arXiv:2501.15679 (2025)

    2025 arXiv

  38. [43]

    Huang, X. et al. Finding Strong Gravitational Lenses in the DESI DECam Legacy Survey. Astrophys. J.894, 78 (2020)

    2020

  39. [44]

    Huang, X. et al. Discovering New Strong Gravitational Lenses in the DESI Legacy Imaging Surveys. Astrophys. J.909, 27 (2021)

    2021

  40. [45]

    Storfer, C. et al. New Strong Gravitational Lenses from the DESI Legacy Imaging Surveys Data Release 9. Astrophys. J. Suppl. Ser.274, 16 (2024)

    2024

  41. [46]

    Strong lensing in UNIONS: Toward a pipeline from discovery to modeling

    Savary, E et al. Strong lensing in UNIONS: Toward a pipeline from discovery to modeling. Astron. Astrophys.666, A1 (2022). 14

    2022

  42. [47]

    Li, R. et al. High-quality Strong Lens Candidates in the Final Kilo-Degree Survey Footprint. Astrophys. J.923, 16 (2021)

    2021

  43. [48]

    Ca˜ nameras, R. et al. HOLISMOKES. II. Identifying galaxy-scale strong grav- itational lenses in Pan-STARRS using convolutional neural networks. Astron. Astrophys. 644, A163 (2020)

    2020

  44. [49]

    Walsh, D., Carswell, R. F. & Weymann, R. J. 0957+561 A, B: twin quasistellar objects or gravitational lens? Nature 279, 381–384 (1979)

    1979

  45. [50]

    Marshall, P.J. et al. SPACE W ARPS - I. Crowdsourcing the discovery of gravitational lenses. Mon. Not. R. Astron. Soc.455, 1171–1190 (2016)

    2016

  46. [51]

    More, A. et al. SPACE W ARPS- II. New gravitational lens candidates from the CFHTLS discovered through citizen science. Mon. Not. R. Astron. Soc.455, 1191–1210 (2016)

    2016

  47. [52]

    Garvin, E.O. et al. Hubble Asteroid Hunter. II. Identifying strong gravitational lenses in HST images with crowdsourcing. Astron. Astrophys.667, A141 (2022)

    2022

  48. [53]

    Gravitational lenses and lens candidates identified from the COSMOS field

    Jackson, N. Gravitational lenses and lens candidates identified from the COSMOS field. Mon. Not. R. Astron. Soc.389, 1311–1318 (2008)

    2008

  49. [55]

    Euclid Collaboration: Rojas, K. et al. Euclid Quick Data Release (Q1) The Strong Lensing Discovery Engine B – Early strong lens candidates from visual inspection of high velocity dispersion galaxies. arXiv e-prints arXiv:2503.15325 (2025)

    2025 arXiv

  50. [56]

    Walmsley, M. et al. Zoobot: Adaptable deep learning models for galaxy morphology. J. Open Source Softw.8, 5312 (2023)

    2023

  51. [57]

    Pearce-Casey, R. et al. Euclid: Searches for strong gravitational lenses using con- volutional neural nets in Early Release Observations of the Perseus field. Astron. Astrophys. 696, A214 (2025)

    2025

  52. [58]

    Nightingale, J.W. et al. ‘PyAutoLens’: Open-source strong gravitational lensing. J. Open Source Softw.6, 2825 (2021)

    2021

  53. [60]

    M., Despali, G., Vegetti, S., Lovell, M

    O’Riordan, C. M., Despali, G., Vegetti, S., Lovell, M. R. & Molin´ e,´A. Sensitivity of strong lensing observations to dark matter substructure: a case study with Euclid. Mon. Not. R. Astron. Soc.521, 2342–2356 (2023). 15

    2023

  54. [62]

    Ivezi´ c,ˇZ. et al. LSST: From Science Drivers to Reference Design and Anticipated Data Products. Astrophys. J.873, 111 (2019)

    2019

  55. [63]

    Sainz de Murieta, A. et al. Find the haystacks, then look for needles: the rate of strongly lensed supernovae in galaxy-galaxy strong gravitational lenses. Mon. Not. R. Astron. Soc.535, 2523–2537 (2024)

    2024

  56. [64]

    Shajib, A.J. et al. Strong gravitational lenses from the Vera C. Rubin Observatory. Philos. Trans. R.Soc. A.383, 20240117 (2025). 16

    2025

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.