The PAU Survey: Dissecting strong-lensing systems with narrow-band photometric redshifts

arXiv:2611.14229 · astro-ph.GA, astro-ph.IM · Submitted 2026-11-10 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Astrophysics Radio. Generated commentary on the latest astrophysics papers.

Vera: We're looking at the paper: "The PAU Survey".

Jocelyn: Here's the gist of the paper. The PAU Survey presents a proof-of-concept demonstration using eight known strong-lens systems to assess agreement with spectroscopic redshifts and evaluate the contributions of narrow-band photometry and foreground-light subtraction, which is crucial for validating lens candidates by testing the lens-source redshift ordering and providing accurate source redshifts in favorable cases The pipeline involves three main steps: broad-band lens modelling, image-based region selection, and multi-band photometry with redshift estimation The broad-band lens modelling fits one reference broad-band image per system using PyAutoLens, representing the deflector light by a multi-Gaussian expansion (MGE) The method uses an elliptical power-law mass profile with external shear to reconstruct the source on a regularised pixel grid Data and Sample Description The analysis utilizes broad-band imaging from the Hyper Suprime-Cam Subarray Strategic Program (HSC) and the Canada-FranceHawaii Telescope Legacy Survey (CFHTLS) for broad bands, and narrow-band imaging from the Physics of the Accelerating Universe Survey (PAUS) The parent sample of galaxy-scale strong lensing systems contains 58 systems drawn from several surveys, with 30 systems from the Strong Lensing Legacy Survey Redshift Recovery and Agreement All eight lens peaks agree with spectroscopy within δz ≤ 0.014, where δz = (zb − zspec)/(1 + zspec) Four source peaks, all at zspec s ≤ 1.23, have δz ≤ 0.025 Symmetric subtraction lowers the median source δz from 0.336 to 0.071 At the common descriptive threshold δz < 0.05, PAUS NB+BB yields eight lens and four source matches, compared with three and two for EAZY using broad bands alone Validation and Future Prospects The results show that lens-model-informed narrow-band photometry can recover component redshift information even when source light is blended with the deflector The results demonstrate that useful redshift information can be recovered from blended lens and source light in favorable cases by symmetric-region foreground subtraction Future applications can benefit from larger lens samples, broader wavelength coverage, and improved spatial resolution Wider narrow-band surveys such as J-PAS offer opportunities to extend the approach to larger samples and support follow-up prioritisation The present evidence establishes feasibility and quantifies best-fit agreement in known lenses The posterior intervals still require independent calibration, and candidate-validation performance remains to be tested on samples containing both lenses and non-lenses Larger spectroscopic samples, image simulations, and joint propagation of photometric and modelling uncertainties are the next steps towards assessing the reliability and wider applicability of the method The present demonstration provides a practical starting point for developing this use of multi-band imaging in strong-lens research The data products and reproduction scripts for all eight systems will be made publicly available on GitHub Appendix Results Summary The appendix shows results for the remaining four systems, including J1404+5200 and SL2SJ0217-0513 The source peak offsets are shown together with their broad redshift distributions in Figure A.1 The ordering probability is P(zsat > zl) = 0.537, indicating that the arcs surrounding the lens galaxies are more likely to originate from background sources than foreground objects The neighboring object’s distribution favours lower redshifts than the arc’s, illustrating the additional information available from separate component photometry The source posterior peaks at 1.240, while the satellite posterior peaks at 0.721 and has a broad distribution that includes the lens redshift of 0.

Paper summary: Vera: So, this paper is called "The PAU Survey: Dissecting strong-lensing systems with narrow-band photometric redshifts". Basically, they're taking a sample of known strong lensing systems and trying to see if they can use narrow-band photometry to figure out the redshifts of those sources more accurately than before.

Jocelyn: Right, so the main idea is that by using this narrow-band information, you can test the ordering of source redshifts and get better estimates for those source redshifts in cases where it's favorable.

Subrahmanyan: That sounds like a way to check if we’re getting our source redshift order right, which is really important for understanding how structure forms in the universe.

Vera: Exactly. They use eight known strong-lens systems as a proof of concept, checking how well the narrow-band photometry and foreground-light subtraction work when testing that lens source redshift ordering and giving them accurate source redshifts in favorable cases.

Jocelyn: And they found some really interesting agreement, saying all eight lens peaks agreed with spectroscopy within a delta z less than zero point zero one four.

Subrahmanyan: That level of agreement is tight, especially when you look at the source peaks that are all at a redshift less than one point two three and still have a delta z less than zero point zero two five.

Vera: Before this paper, they were getting median source delta z values around zero point three three six, but with symmetric subtraction, they lowered that to zero point zero seven one.

Jocelyn: And when you apply that common threshold of delta z less than zero point zero five, the PAUS narrow-band plus broad-band analysis yielded eight lens and four source matches compared to just three and two for EAZY using only broad bands.

Subrahmanyan: That jump from two to eight matches is significant. It suggests that this technique can recover component redshift information even when the source light is blended with the deflector, which is a tricky situation.

Vera: That's what they show, that lens-model informed narrow-band photometry can actually recover component redshift information even when the source light is blended with the deflector.

Jocelyn: It’s showing that useful redshift information can be recovered from blended lens and source light in favorable cases by symmetric-region foreground subtraction.

Subrahmanyan: I wonder what this implies for structure formation, specifically how we interpret the relative distances between different structures when they are overlapping in the sky.

Vera: It suggests that if you have good multi-band data, you can get better constraints on these cosmological scales than just relying on broad bands alone.

Jocelyn: But they're also being careful about what this means, saying that the posterior intervals still require independent calibration and candidate validation performance remains to be tested on samples containing both lenses and non-lenses.

Subrahmanyan: So, it’s a proof of concept showing feasibility in known systems, but they aren't claiming it's ready for massive applications yet.

Vera: Exactly. The next steps involve larger spectroscopic samples, image simulations, and joint propagation of photometric and modelling uncertainties to assess the reliability and wider applicability of this method.

Jocelyn: So that’s where the discussion moves next: what does this paper actually mean for the broader field? Let's talk about the title and authors of "The PAU Survey: Dissecting strong-lensing systems with narrow-band photometric redshifts".

Subrahmanyan: From a theoretical standpoint, it's about proving that our observational tools can disentangle complex gravitational lensing effects using multi-wavelength data.

Vera: It’s really about showing that combining different parts of the light spectrum, like the broad bands and the narrow bands from PAUS, gives us a better handle on the source redshifts when they are close together.

Jocelyn: The authors are essentially demonstrating how this specific combination of photometric analysis and model fitting helps them validate those lens candidates by testing that source redshift ordering.

Subrahmanyan: For cosmology, this means we can start getting more precise constraints on distances in systems where the lensing geometry is particularly complex, like these strong-lensing configurations they studied.

Vera: It provides a practical starting point for developing this use of multi-band imaging in strong-lens research and gives us a concrete way to test these ideas before we deploy them on much larger samples.

Jocelyn: So the takeaway is that the method works well in these specific, controlled environments, but it needs more rigorous testing on bigger samples to be truly useful for mapping out cosmic structure.

Conclusion: Vera: So, this paper is about using narrow-band photometry to figure out source redshifts in strong lenses better than before, focusing on eight known systems.

Jocelyn: Right, and the title itself is pretty descriptive: "The PAU Survey: Dissecting strong-lensing systems with narrow-band photometric redshifts."

Subrahmanyan: It really zeroes in on how this specific technique helps us disentangle the source and the lens, which is key for cosmology.

Vera: Exactly. The authors are showing that by combining broad bands and these narrower PAUS bands, you can recover component redshift information even when the source light is blended with the deflector light.

Jocelyn: And they’re doing this using a specific methodology—broad-band lens modeling with a multi-Gaussian expansion to reconstruct the source image on a grid.

Subrahmanyan: That modeling part is what lets them test that crucial lens-source redshift ordering, which tells us about the geometry of the universe.

Vera: They found they agree with spectroscopy within a delta z less than zero point zero one four for all eight lens peaks, which is tight when you consider the source redshifts are all below one point two three.

Jocelyn: But there’s a huge caveat here because they show that symmetric subtraction really lowers the median source delta z from about zero point three three six down to just zero point zero seven one.

Subrahmanyan: That drop is significant because it shows that this method, when done right, can provide much cleaner redshift estimates for the source population.

Vera: And they found that when you use this common threshold of delta z less than zero point zero five, the PAUS narrow-band plus broad-band analysis yielded eight lens and four source matches.

Jocelyn: That’s a real jump compared to just using broad bands alone, which only gave them three lens and two source matches.

Subrahmanyan: It suggests that this combination of photometric analysis and model fitting can be a powerful tool for candidate validation in the field.

Vera: They do stress that the posterior intervals still need independent calibration, and they have to test how well it works on samples with both lenses and non-lenses.

Jocelyn: So, this paper proves feasibility in these eight systems, but they are setting the stage for much bigger tests.

Subrahmanyan: And what they're saying is that larger spectroscopic samples and joint propagation of all those uncertainties are the next big steps to really test how reliable this method is for mapping cosmic structure.

Vera: It’s a practical starting point, showing how this multi-band imaging approach can be used in strong-lens research.

Jocelyn: The authors have made the data and reproduction scripts public on GitHub so anyone can check the results themselves.

Subrahmanyan: So, it’s a good demonstration of how to use multi-band imaging to get better constraints on distances in complex lensing environments.

Vera: Next time we talk, we’re going to look at how this specific method compares when you move beyond just eight systems and try to apply it across much larger samples.

Shuaibo Geng, Pablo Renard, Qiuhan He, David Navarro-Gironés, Magherita Grespan, Agnieszka Pollo, Enrique Gaztañaga, Marek Biesiada, Ricard Casas, Juan García-Bellido, Pau Tallada-Crespí

National Centre for Nuclear Research Warsaw Poland University of Hunan Normal University Astronomical Observatory of the Jagiellonian University Institute of Space Sciences CSIC Institut d’Estudis Espacials de Catalunya Leiden Observatory Kapteyn Astronomical Institute Sub-department of Astrophysics University of Portsmouth Instituto de Física Teórica CSIC Universidad Autónoma de Madrid Centro de Investigaciones Energéticas Medioambientales y Tecnológicas

astro-ph.GA, astro-ph.IM

Submitted: 2026-11-10

Updated: 2026-11-10

Comments: 17 pages, 7 figures

Project page: https://hsc-niji.github.io

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 83/100

The gist: The PAU Survey presents a proof-of-concept demonstration using eight known strong-lens systems to assess agreement with spectroscopic redshifts and evaluate the contributions of narrow-band…

Key concepts

Broad-band Lens Modelling
This step uses software to fit a mathematical model (a multi-Gaussian expansion) to the image of the deflecting galaxy. It tries to reconstruct what the background source image should look like on a grid, assuming a specific mass profile for the lens.
Narrow-band Photometry
This involves using specific color filters in imaging surveys to get precise measurements of light at particular wavelengths. This data is used alongside broad-band images to estimate the redshift of the source and help distinguish between foreground objects and background sources.
Redshift Agreement ($\delta z$)
This measures how close the estimated redshift from a photometric method is to the actual spectroscopic redshift. A small $\delta z$ value indicates high agreement, showing that the photometric technique accurately identifies the source's distance relative to known measurements.
Symmetric Subtraction
This technique involves removing light from regions around lens galaxies before analyzing the remaining image. This helps lower the estimated redshift uncertainty for source peaks by effectively subtracting foreground or blended light contributions.

Terminology

Summary

The PAU Survey presents a proof-of-concept demonstration using eight known strong-lens systems to assess agreement with spectroscopic redshifts and evaluate the contributions of narrow-band photometry and foreground-light subtraction, which is crucial for validating lens candidates by testing the lens-source redshift ordering and providing accurate source redshifts in favorable cases

Methodology Overview

The pipeline involves three main steps: broad-band lens modelling, image-based region selection, and multi-band photometry with redshift estimation The broad-band lens modelling fits one reference broad-band image per system using PyAutoLens, representing the deflector light by a multi-Gaussian expansion (MGE) The method uses an elliptical power-law mass profile with external shear to reconstruct the source on a regularised pixel grid

Data and Sample Description

The analysis utilizes broad-band imaging from the Hyper Suprime-Cam Subarray Strategic Program (HSC) and the Canada-FranceHawaii Telescope Legacy Survey (CFHTLS) for broad bands, and narrow-band imaging from the Physics of the Accelerating Universe Survey (PAUS) The parent sample of galaxy-scale strong lensing systems contains 58 systems drawn from several surveys, with 30 systems from the Strong Lensing Legacy Survey

Redshift Recovery and Agreement

All eight lens peaks agree with spectroscopy within δz ≤ 0.014, where δz = (zb − zspec)/(1 + zspec) Four source peaks, all at zspec s ≤ 1. p. 023, have δz ≤ 0. p. 0025 Symmetric subtraction lowers the median source δz from 0. p. 0336 to 0.071 At the common descriptive threshold δz < 0. p. 005, PAUS NB+BB yields eight lens and four source matches, compared with three and two for EAZY using broad bands alone

Validation and Future Prospects

The results show that lens-model-informed narrow-band photometry can recover component redshift information even when source light is blended with the deflector The results demonstrate that useful redshift information can be recovered from blended lens and source light in favorable cases by symmetric-region foreground subtraction Future applications can benefit from larger lens samples, broader wavelength coverage, and improved spatial resolution Wider narrow-band surveys such as J-PAS offer opportunities to extend the approach to larger samples and support follow-up prioritisation

Conclusions

The present evidence establishes feasibility and quantifies best-fit agreement in known lenses The posterior intervals still require independent calibration, and candidate-validation performance remains to be tested on samples containing both lenses and non-lenses Larger spectroscopic samples, image simulations, and joint propagation of photometric and modelling uncertainties are the next steps towards assessing the reliability and wider applicability of the method The present demonstration provides a practical starting point for developing this use of multi-band imaging in strong-lens research The data products and reproduction scripts for all eight systems will be made publicly available on GitHub

Appendix Results Summary

The appendix shows results for the remaining four systems, including J1404+5200 and SL2SJ0217-0513 The source peak offsets are shown together with their broad redshift distributions in Figure A.1 The ordering probability is P(zsat > zl) = 0. p. 9537, indicating that the arcs surrounding the lens galaxies are more likely to originate from background sources than foreground objects The neighboring object’s distribution favours lower redshifts than the arc’s, illustrating the additional information available from separate component photometry The source posterior peaks at 1. p. 9240, while the satellite posterior peaks at 0. p. 3721 and has a broad distribution that includes the lens redshift of 0. p. 9

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Component Redshift Estimation from Blended Light: The improved system can recover component redshift information even when source light is blended with deflector light by using lens-model-informed narrow-band photometry.

  2. Redshift Ordering Probability Assessment: The system can calculate the redshift ordering probability Pord, which is defined as P(zs > zl) = Z pl(zl) / (Z ps(zs)) dzs / dzl, to provide information for vetting lens candidates.

  3. Foreground Contamination Mitigation: The system incorporates symmetric-background subtraction to lower the median source |delta z| from 0.336 to 0.071, significantly improving accuracy by mitigating foreground contamination that dominates inferred colours and model-based source photometry.

  4. Multi-band Constraint Utilization: The system utilizes a combination of broad- and narrow-band imaging fluxes with BCNz to estimate redshifts, demonstrating that lens-model-informed narrow-band photometry can recover component redshift information even when source light is blended with the deflector.

  5. Candidate Prioritization for Spectroscopic Follow-up: The system can provide a source redshift distribution p(zs) which predicts a range of observed wavelengths for candidate emission lines, informing target priorities and instrument selection for spectroscopic follow-up.

  6. Component Classification via Neighbouring Objects: The system can perform a neighboring-object test to extract the redshift posterior of an object outside the main arc, which illustrates the additional information available from separate component photometry.

Sources

Related papers