Source REconstruction of Arcs behind Cluster Halos (SourceREACH): A New Source Reconstruction Algorithm Optimized for Giant Arcs and Galaxy Cluster Lenses

Charles Keeton; Lana Eid

arxiv: 2511.11952 · v2 · pith:2HJ2TBFHnew · submitted 2025-11-14 · 🌌 astro-ph.CO

Source REconstruction of Arcs behind Cluster Halos (SourceREACH): A New Source Reconstruction Algorithm Optimized for Giant Arcs and Galaxy Cluster Lenses

Lana Eid , Charles Keeton This is my paper

Pith reviewed 2026-05-25 07:52 UTC · model grok-4.3

classification 🌌 astro-ph.CO

keywords source reconstructiongravitational lensinggalaxy clustersgiant arcsK nearest neighbor regressionde-lensingAbell 370

0 comments

The pith

A new algorithm reconstructs background sources from giant arcs by de-lensing observed pixels then applying regression to the source plane.

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

The paper introduces SourceREACH, a source reconstruction method tailored to giant arcs lensed by galaxy clusters. It works by first deconvolving the image with the point spread function, then de-lensing individual pixels into the source plane, and finally applying interpolation or regression with smoothing to recover the source while respecting the position-dependent resolution caused by varying magnification. Tests on both simulated data and the real giant arc in Abell 370 show that K Nearest Neighbor Regression produces the best tradeoff between suppressing noise and retaining compact source features. A sympathetic reader would care because such reconstructions supply direct information on the background galaxy's structure and on the cluster's mass distribution, both of which are otherwise computationally expensive to extract from extended lensed images.

Core claim

The SourceREACH algorithm deconvolves the observed image by the point spread function, de-lenses the image pixels, and uses interpolation or regression with smoothing to determine the model source. By operating on de-lensed points the method automatically accounts for the varying resolution across the source plane. When different interpolation and regression techniques are compared on mock data and on the giant arc in Abell 370, K Nearest Neighbor Regression supplies the best balance of noise smoothing and preservation of compact detail in the source.

What carries the argument

De-lensing of image pixels into the source plane followed by K Nearest Neighbor Regression to produce the model source while handling spatially varying resolution.

If this is right

Giant arcs can be modeled with reduced computational cost compared with traditional pixelated source reconstruction methods.
K Nearest Neighbor Regression yields source models that suppress noise without erasing compact features in high-resolution cluster-lens data.
The approach automatically incorporates the position-dependent magnification when mapping image pixels back to the source plane.
The same pipeline works on both simulated arcs and real observations such as the Abell 370 system.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same de-lensing-plus-regression logic could be applied to other strong-lensing configurations once an accurate lens model exists.
Adding an explicit PSF convolution step inside the forward model would extend usability to data sets where PSF blurring cannot be ignored.
The method's speed advantage may allow routine analysis of the much larger samples of arcs expected from upcoming wide-field surveys.

Load-bearing premise

The method assumes point spread function effects are not significant and that an accurate lens model is already available for the de-lensing step.

What would settle it

Re-running the reconstruction on the same Abell 370 arc data after deliberately inserting a known PSF kernel or after replacing the lens model with a deliberately inaccurate one would show whether the recovered source matches expectations only when the assumptions hold.

Figures

Figures reproduced from arXiv: 2511.11952 by Charles Keeton, Lana Eid.

**Figure 1.** Figure 1: Top left: Data image created from the HST F814W filter from the Hubble Frontier Fields program. Top right: Data arc with light from galaxies near the arc modeled and subtracted (except for two along the left portion of the arc, which were masked instead; see Paper 1 for more discussion). Bottom left: Masked and cleaned data arc, which was the input to pixsrc for our previous optimization runs. Bottom right… view at source ↗

**Figure 2.** Figure 2: Results from our de-lensing source reconstruction applied to the idealized mock arc with no noise and no PSF. Top Row: De-lensed source prior to smoothing and isolated mock arc. Remaining Rows: Smoothing method tests on the same set of pixels as above: Radial Basis Function Interpolation, K Nearest Neighbors Regression, Decision Tree Regression, and Random Forest Regression. First Column: Model source, wit… view at source ↗

**Figure 3.** Figure 3: Similar to [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

**Figure 4.** Figure 4: Results from our de-lensing source reconstruction applied to the cleaned data in the F814W filter for the observed arc in Abell 370. The structure of the figure is the same as in [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗

**Figure 5.** Figure 5: Similar to [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗

**Figure 6.** Figure 6: Similar to [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗

**Figure 7.** Figure 7: Similar to [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗

**Figure 8.** Figure 8: Similar to [PITH_FULL_IMAGE:figures/full_fig_p013_8.png] view at source ↗

**Figure 9.** Figure 9: Similar structure as [PITH_FULL_IMAGE:figures/full_fig_p014_9.png] view at source ↗

**Figure 10.** Figure 10: Abell 370 data arc source reconstruction for the HFF F814W, F606W, and F435W filters, predicted and smoothed with the KNN Ball Tree option with 6 nearest neighbors. We display here the source reconstruction in the left column, the reconvolved model arc in the middle column, and the residuals showing deconvolved data minus model arc with no PSF reconvolution in the right column [PITH_FULL_IMAGE:figures/fu… view at source ↗

**Figure 11.** Figure 11: Left: RGB source reconstruction using the F814W, F606W, and F435W HFF filters. Corresponding caustics are overlaid in yellow; a 1 kpc scale bar is shown for reference. Right: Model arc created using source reconstruction and reconvolved with PSF in each filter. Corresponding critical curves and data mask are overlaid in yellow and green, respectively; a 1 arcsecond scale bar is shown for reference. jib et… view at source ↗

read the original abstract

We introduce a new algorithm designed for use with extended lensed images, specifically giant arcs lensed by galaxy clusters. These highly magnified images contain important information about both the mass distribution of the cluster and the properties of the background source, but modeling them requires significant computational effort. Our new source reconstruction methodology is designed to be accurate and efficient for high-resolution observations in which point spread function effects are not significant. The overall process deconvolves the observed image by the point spread function, de-lenses the image pixels, and uses interpolation or regression with smoothing to determine the model source. By working with de-lensed points, the method accounts for varying resolution across the source plane. We evaluate the speed and accuracy of different interpolation and regression methods using both mock data and real data for the giant arc in Abell 370. We find that utilizing K Nearest Neighbor Regression results in the best balance of noise smoothing and preservation of compact detail in the source.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

De-lensing pixels first then running KNN regression is the core new step, but the performance claim rests on an untested perfect-lens-model assumption and lacks any numbers.

read the letter

The paper's main contribution is a straightforward pipeline that maps image pixels back to the source plane using an existing lens model, then applies regression or interpolation to reconstruct the source while accounting for the uneven sampling that magnification produces. They test several options on mocks and on the Abell 370 giant arc and conclude that K-nearest-neighbor regression gives the best trade-off between noise suppression and retention of compact features. That workflow is presented as new for this class of systems, and the focus on efficiency for high-resolution data without PSF complications is reasonable given the target use case of giant arcs in cluster fields. Running both simulated and real data is also a positive step. The limitation that stands out is the absence of any quantitative results. The abstract and stress-test note give no chi-squared values, no residual maps, no timing benchmarks against existing codes, and no direct comparison to standard pixelated source reconstruction methods. Without those numbers it is difficult to know whether the claimed advantage is real or just an artifact of the particular test images. A second and more structural issue is the reliance on an already-accurate lens model. The method de-lenses the pixels first, so any error in the convergence or shear map shifts the source-plane points systematically. The mocks use the exact input model, and the Abell 370 example presumably uses a published model treated as truth; neither case probes how the ranking among regression methods changes when the lens model has the 5-10% uncertainties that are typical for cluster-scale work. This is a specialized methods paper aimed at groups already doing cluster lensing who need a fast source estimate once the mass model is in hand. It is not a foundational advance, but the idea is clear enough that a serious referee could usefully push for the missing benchmarks and a sensitivity test to lens-model errors. I would send it to review rather than desk-reject.

Referee Report

2 major / 1 minor

Summary. The paper introduces SourceREACH, an algorithm for source reconstruction from giant arcs in galaxy cluster lenses. It deconvolves the observed image by the PSF, de-lenses pixels using a supplied lens model, and applies interpolation or regression with smoothing to reconstruct the source while accounting for varying resolution across the source plane. Evaluation on mock data and real Abell 370 observations identifies K Nearest Neighbor Regression as providing the optimal balance of noise smoothing and detail preservation.

Significance. If the performance ranking is placed on quantitative footing and the dependence on an accurate input lens model is tested, the approach could supply a computationally efficient reconstruction tool for high-resolution cluster-lensing data where PSF effects are negligible.

major comments (2)

[Abstract and Section 3] Abstract and evaluation section: the claim that KNN regression 'results in the best balance of noise smoothing and preservation of compact detail' is presented without any quantitative metrics (e.g., RMS residuals, structural similarity indices, or cross-validation scores), error bars, or explicit comparison baselines against the other tested methods. This leaves the headline performance conclusion only weakly supported.
[Section 3] Section 3: the entire evaluation pipeline (both mock and Abell 370) assumes an exact, error-free lens model for the de-lensing step. No sensitivity analysis is reported for realistic cluster-scale lens-model uncertainties (typically 5–10 % in convergence), which would systematically displace source-plane points and could change the relative ranking of the regression methods. Mock tests use the input model by construction and therefore do not probe this failure mode.

minor comments (1)

[Abstract] Abstract: the statement that the algorithm is 'designed for high-resolution observations in which point spread function effects are not significant' is not accompanied by any quantitative criterion or validation test for when this approximation holds.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive report. The comments identify opportunities to strengthen the quantitative basis of our performance claims and to examine robustness to lens-model errors. We address each major comment below and will revise the manuscript accordingly.

read point-by-point responses

Referee: [Abstract and Section 3] Abstract and evaluation section: the claim that KNN regression 'results in the best balance of noise smoothing and preservation of compact detail' is presented without any quantitative metrics (e.g., RMS residuals, structural similarity indices, or cross-validation scores), error bars, or explicit comparison baselines against the other tested methods. This leaves the headline performance conclusion only weakly supported.

Authors: We agree that the current text relies on visual comparison of reconstructed sources. In the revised manuscript we will add explicit quantitative metrics for the mock-data tests, including RMS residuals to the true source, structural similarity index (SSIM), and cross-validation scores where appropriate. These quantities will be reported for every interpolation and regression method together with error bars obtained from multiple independent noise realizations, thereby placing the ranking of KNN regression on a firmer numerical footing. revision: yes
Referee: [Section 3] Section 3: the entire evaluation pipeline (both mock and Abell 370) assumes an exact, error-free lens model for the de-lensing step. No sensitivity analysis is reported for realistic cluster-scale lens-model uncertainties (typically 5–10 % in convergence), which would systematically displace source-plane points and could change the relative ranking of the regression methods. Mock tests use the input model by construction and therefore do not probe this failure mode.

Authors: The referee correctly observes that our present tests assume a perfect lens model. We will add a dedicated sensitivity subsection that perturbs the convergence map by 5 % and 10 % (consistent with typical cluster-lens uncertainties) and re-runs the full reconstruction pipeline on the mock data. The resulting changes in RMS, SSIM, and method ranking will be quantified and discussed. For the Abell 370 observations we will reference published lens-model uncertainties and comment on their expected influence on the source-plane sampling. revision: yes

Circularity Check

0 steps flagged

No significant circularity in algorithmic evaluation on external data

full rationale

The paper describes an algorithmic procedure (de-lensing followed by interpolation or regression) and evaluates its performance empirically on mock data and real observations of Abell 370. The central finding that KNN regression offers the best noise/detail balance is a direct outcome of those external tests rather than any parameter fitted inside the method's own equations or any derivation that reduces to its inputs by construction. No self-definitional steps, fitted-input predictions, or load-bearing self-citations appear in the presented logic; the assumption of an accurate input lens model is an external precondition, not a circular element within the paper.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The algorithm rests on standard gravitational lensing assumptions and the availability of a pre-existing lens model; no new physical entities or many fitted parameters are introduced beyond the choice of regression method.

axioms (2)

domain assumption An accurate lens mass model is available to map observed pixels back to the source plane.
The de-lensing step presupposes a known deflection field from the cluster.
domain assumption Point spread function effects can be neglected or pre-corrected.
Explicitly stated as the regime for which the algorithm is designed.

pith-pipeline@v0.9.0 · 5703 in / 1348 out tokens · 47291 ms · 2026-05-25T07:52:27.318893+00:00 · methodology

discussion (0)

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

First Statistical Study of Over 100 Magnified Stellar Events at Redshift $z \approx 0.725$ with JWST
astro-ph.GA 2026-04 unverdicted novelty 6.0

Over 100 caustic-crossing stellar events identified in the Dragon galaxy at z≈0.725 with JWST data yield a stellar luminosity function slope β=2.18 and confirm parity asymmetry.

Reference graph

Works this paper leans on

1 extracted references · 1 canonical work pages · cited by 1 Pith paper

[1]

B., et al., 2017, A&A, 600, A90 Caminha G

Acebron A., et al., 2024, ApJ, 976, 110 Birrer S., 2021, ApJ, 919, 38 Birrer S., Amara A., 2018, Physics of the Dark Universe, 22, 189 Birrer S., Amara A., Refregier A., 2015, ApJ, 813, 102 Bradaˇ c M., Schneider P., Lombardi M., Erben T., 2005, A&A, 437, 39 Bradaˇ c M., et al., 2009, ApJ, 706, 1201 Bradley L., et al., 2022, astropy/photutils: 1.5.0, doi:...

work page doi:10.5281/zenodo.6825092 2024

[1] [1]

B., et al., 2017, A&A, 600, A90 Caminha G

Acebron A., et al., 2024, ApJ, 976, 110 Birrer S., 2021, ApJ, 919, 38 Birrer S., Amara A., 2018, Physics of the Dark Universe, 22, 189 Birrer S., Amara A., Refregier A., 2015, ApJ, 813, 102 Bradaˇ c M., Schneider P., Lombardi M., Erben T., 2005, A&A, 437, 39 Bradaˇ c M., et al., 2009, ApJ, 706, 1201 Bradley L., et al., 2022, astropy/photutils: 1.5.0, doi:...

work page doi:10.5281/zenodo.6825092 2024