Strong Gravitational Lens Candidate

Updated 31 January 2026

Strong gravitational lens candidates are astrophysical systems that produce multiple or significantly distorted images through gravitational deflection of background light.
Researchers utilize visual inspection, spectroscopic blending, and automated machine learning pipelines to distinguish genuine lensing events from contaminants.
High-resolution imaging and detailed spectroscopic confirmation of these candidates enable precise studies of galaxy structure, dark matter distribution, and cosmological parameters.

A strong gravitational lens candidate is an astrophysical system—typically a galaxy or cluster of galaxies—suspected of producing multiple or highly distorted images of background sources due to the deflection of light in a gravitational potential well, operating in the strong lensing regime. Such candidates manifest as multiply imaged quasars, arcs, rings, or complex image configurations in imaging/spectroscopic data. Their identification underpins diverse research programs in galaxy structure, dark matter substructure, time-delay cosmography, and early-universe galaxy studies.

1. Observational Signatures and Candidate Definition

The defining criterion of a strong gravitational lens candidate is evidence of multiple, highly magnified or significantly distorted background source images, originating from the intersection of a critical curve (where the projected surface mass density exceeds the critical density, $\kappa > 1$ ) with the sky plane. Canonical signatures include:

Galaxy-scale Lenses: Einstein rings, multiple arcs, or crossed image patterns (quads, doubles) around early-type galaxies.
Cluster-scale Lenses: Giant arcs and arclets (exceeding $10''$ in length, e.g., PLCK G287.0+32.9, $\theta_E \sim 42''$ (Zitrin et al., 2017)), radial features, image multiplicity in cluster core vicinities.
Lensed Quasars: Multiple point sources with consistent colors and spectra, separated by fractions of an arcsecond to several arcseconds; flux ratios constrained by simple mass models (Chan et al., 2014, Krone-Martins et al., 2018).
Spectroscopic Indicators: Blended spectra with two distinct redshift systems, corresponding to a foreground deflector and a background emission-line galaxy (e.g., [O II], H $\alpha$ ; (Talbot et al., 2022, Karp et al., 3 Dec 2025)).

Stringent candidates exhibit confirmed or highly probable lensing morphologies and at least preliminary photometric or spectroscopic redshift separation of putative lens and source.

2. Methodologies for Candidate Identification

Strong lens candidates are identified via a combination of photometric survey imaging, color/magnitude selection, spectroscopic searches, and increasingly, automated machine learning driven pipelines. Key methodologies include:

Visual Inspection: Systematic manual scanning of large image sets for arc/ring morphologies, human ranking by lensing “grade” (A: definite, B: probable, C: possible) (O'Donnell et al., 2021, Nagam et al., 13 Feb 2025, Storfer et al., 8 May 2025).
Color and Morphology Preselection: Isolation of massive galaxies using color cuts (e.g., $g{-}r$ and $r{-}i$ ), size, and surface-brightness thresholds, followed by searching for multiple blue knots (candidate arcs) nearby (O'Donnell et al., 2021, Storfer et al., 8 May 2025).
Spectroscopic Blending: Detection of multiple redshift components in single-fiber (SDSS, DESI) or IFU spectra (MaNGA, DEVILS)—commonly via high-S/N emission-line residuals after subtraction of the foreground galaxy’s PCA-modeled continuum. Lensed background lines ([O II], H $\beta$ , [O III]) are sought at impact parameters $\lesssim \theta_E$ (Talbot et al., 2022, Smith, 2016, Holwerda et al., 2021, Karp et al., 3 Dec 2025).
Machine Learning and Automated Algorithms:
- CNN/Transformer-based: Deep convolutional/residual/transformer models, trained on simulated and real lens/non-lens image cutouts, achieve ROC-AUC $\gtrsim 0.94$ and can process $O(10^6)$ images with low FPR (Metcalf et al., 2018, Grespan et al., 2024, Nagam et al., 13 Feb 2025, Storfer et al., 8 May 2025).
- Arc-Finders, SVMs: Early pipelines used morphological filtering (elongation, curvature), Gabor/shapelet features, and SVM classifiers (Metcalf et al., 2018, Chan et al., 2014).
- Astrometric/Photometric Combinations: High-precision astrometry (e.g., Gaia DR2) coupled with colors and supervised learning (ERT) rapidly isolates quasar-lens configurations (Krone-Martins et al., 2018).

The methodology is often hierarchical: preselection narrows candidate pools, followed by ML or model-based filters, culminating with expert visual vetting and, for robust grades, spectroscopic confirmation.

3. Quantitative Selection Criteria

Quantitative thresholds define "candidacy" and control contamination:

Einstein Radius ($10''$0): A key parameter; candidates typically must show image separations $10''$1 (seeing-limited) or lower for space-based surveys (Euclid: $10''$2). Empirically, spectroscopically detected samples peak at $10''$3 (Karp et al., 3 Dec 2025, Smith, 2016, Talbot et al., 2022).
Photometric/Spectroscopic Consistency: For lensed quasar candidates, image pairs/groups must satisfy near-identical colors and spectroscopic redshifts ($10''$4 as in LAMOST J1606+2900 (Chen et al., 2023)), near-constant flux ratios, and the absence of plausible alternatives (e.g., interacting quasars).
Emission-Line S/N and Blend Statistic: Blended spectra require secondary emission lines at S/N $10''$5 (multi-line) or $10''$6 ([O II] doublet), significant blend statistics ($10''$7 fitted threshold as in (Holwerda et al., 2021)), and $10''$8 to exclude star-forming companions.
Neural Net Thresholds: CNN/transformer-based pipelines require high-confidence scores ($10''$9–0.9) to flag lens candidates (Grespan et al., 2024, Storfer et al., 8 May 2025, Nagam et al., 13 Feb 2025).
Probabilistic Lensing Region: In single-fiber spectroscopic searches, a lensing probability $\theta_E \sim 42''$ 0 is computed by integrating the likelihood of the [O II] detection originating within the Einstein radius versus beyond, with $\theta_E \sim 42''$ 1 median $\theta_E \sim 42''$ 2 (Karp et al., 3 Dec 2025).

Common practice is to construct a hierarchical grading structure, prioritizing high-purity at the expense of completeness for follow-up.

4. Confirmation and Follow-up Strategies

Confirmation of strong lens candidates requires additional data and modeling:

High-Resolution Imaging: HST, JWST, or adaptive optics observations resolve arcs, rings, and multiple images, allowing for model fitting and elimination of contaminants such as ring galaxies or mergers (Salmon et al., 2018, Nagam et al., 13 Feb 2025, Zitrin et al., 2017).
Spectroscopic Confirmation: Independent measurements of $\theta_E \sim 42''$ 3 and $\theta_E \sim 42''$ 4 (from absorption and emission features respectively), or identification of background emission lines spatially offset with respect to the lens, lift degeneracy with interlopers (Holwerda et al., 2021, Karp et al., 3 Dec 2025, Huang et al., 22 Sep 2025).
Lens Modeling: Application of parametric models (SIS, SIE, NFW) plus external shear to the observed image position and flux configuration; success is gauged via $\theta_E \sim 42''$ 5 fit to positions/fluxes (Chan et al., 2014, Krone-Martins et al., 2018, Nagam et al., 13 Feb 2025). Critical for sub-milliarcsecond astrometry (e.g., Gaia quadruple quasar candidates).
Spectral Decomposition: In integral-field datasets, narrow-band imaging of residuals after continuum subtraction reveals arcs at predicted Einstein radii (Talbot et al., 2022, Smith, 2016, Talbot et al., 2018).
Time-Delay and Variability Monitoring: For lensed quasars or supernovae, correlated variability and measured time delays confirm the lens and enable cosmographic studies.

The fraction of robust candidates confirmed varies widely by selection method and depth, but high-purity spectroscopic blends routinely achieve $\theta_E \sim 42''$ 6 confirmation rates (Holwerda et al., 2021).

5. Astrophysical and Cosmological Applications

A validated strong gravitational lens candidate sample supports a broad range of scientific goals:

Dark Matter and Galaxy Structure: Image positions and flux ratios tightly constrain the inner mass profile and substructure abundance (flux-ratio anomalies) (Karp et al., 3 Dec 2025, Krone-Martins et al., 2018).
Cosmography: Time delays measured in lensed quasar systems provide a direct probe of $\theta_E \sim 42''$ 7 independent of local distance ladders or CMB calibrations (O'Donnell et al., 2021, Karp et al., 3 Dec 2025).
High-Redshift Galaxy Studies: Lensing magnification enables the detection and spatial/kinematic study of galaxies at $\theta_E \sim 42''$ 8–10, extending the reach of JWST and ALMA (Salmon et al., 2018, Zitrin et al., 2017).
Transient Magnification: Multiply lensed supernovae facilitate time-delay measurements and tests of lens model accuracy (Karp et al., 3 Dec 2025).
Statistical Lens Samples: Next-generation surveys (Euclid, Rubin, SKA) will assemble samples of $\theta_E \sim 42''$ 9– $\alpha$ 0 confirmed lenses, enabling precision measurements of the evolution of galaxy mass structures, dark energy, and the matter power spectrum (Serjeant, 2014, Nagam et al., 13 Feb 2025, Storfer et al., 8 May 2025).

Derived lens parameters—Einstein mass, enclosed mass-to-light ratio, total mass profiles—serve as probes for the stellar initial mass function and dark matter fraction in galaxies (Smith, 2016, Bussmann et al., 2013).

6. Future Prospects and Challenges

Survey-scale lens candidate identification faces key challenges and opportunities:

Scalability: Automated ML pipelines, notably deep CNNs and transformer encoders, are essential for processing $\alpha$ 1 image cutouts; ensemble models and domain adaptation (fine-tuning on real data) are critical for minimizing false positives (Grespan et al., 2024, Metcalf et al., 2018, Nagam et al., 13 Feb 2025).
Purity vs. Completeness: There is a fundamental trade-off; maximizing high-confidence detection (purity) is preferable for expensive follow-up studies, but modeling selection functions is crucial for statistical cosmology (Nagam et al., 13 Feb 2025, Metcalf et al., 2018).
False Positive Control: Rings, mergers, and edge-on spirals are the dominant contaminants; including realistic "hard negatives" in training sets substantially reduces FPR (Grespan et al., 2024).
Spectroscopic Confirmation Bottlenecks: The need for spatially resolved spectroscopy, especially for small Einstein-radii systems and higher redshifts, remains a rate-limiting step—multiobject spectrographs and IFU surveys are needed (Karp et al., 3 Dec 2025, Talbot et al., 2022, Holwerda et al., 2021).
Photometric Biases: Selection depends on survey depth, seeing, and input assumptions in simulations, requiring careful calibration to ensure unbiased lens statistics (Storfer et al., 8 May 2025, Metcalf et al., 2018).
Cataloging and Data Releases: Systematic release of candidates, false positives, and metadata (redshifts, grades, imaging) facilitates community validation and future methodological refinement (Grespan et al., 2024, Storfer et al., 8 May 2025).

Ongoing efforts are converging on hierarchical, hybrid pipelines—preselection using color/magnitude, machine learning filtering, human vetting, and targeted spectroscopy—that are robust to diverse lens morphologies and survey conditions (Nagam et al., 13 Feb 2025, Metcalf et al., 2018, Storfer et al., 8 May 2025). These frameworks are fundamental to realizing the full scientific promise of strong gravitational lens candidates in the era of petabyte-scale astronomical surveys.