🔭 astrophysics

UNIONS-3500 Weak Lensing: II. B-mode validation for cosmic shear

This paper validates the cosmic shear $B$ -mode systematics for the UNIONS-3500 survey using three distinct statistical frameworks, identifying and mitigating detector-level additive biases through specific scale and sample cuts to establish robust null tests essential for future Stage-IV cosmological inference.

Original authors: C. Daley, A. Guinot, S. Guerrini, F. Hervas-Peters, L. W. K. Goh, C. Murray, M. Kilbinger, A. Wittje, M. J. Hudson, H. Hildebrandt, L. van Waerbeke, A. W. McConnachie

Published 2026-04-06

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: C. Daley, A. Guinot, S. Guerrini, F. Hervas-Peters, L. W. K. Goh, C. Murray, M. Kilbinger, A. Wittje, M. J. Hudson, H. Hildebrandt, L. van Waerbeke, A. W. McConnachie

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

The Great Cosmic Shear Detective Story: Cleaning Up the Universe's Mirror

Imagine you are trying to take a perfect, crystal-clear photograph of a distant galaxy. But there's a problem: you are looking through a dirty window. The window has smudges, scratches, and a weird pattern of dust that wasn't there when you bought it.

In the world of astronomy, this "window" is the Earth's atmosphere and the telescope's own camera. The "smudges" are systematic errors. The "distant galaxies" are the clues we need to understand how the universe is built.

This paper is about UNIONS-3500, a massive project mapping the northern sky to study Weak Gravitational Lensing.

1. The Cosmic Mirror (What are they looking at?)

Gravity is invisible, but it bends light. When light from a distant galaxy passes through a massive cluster of dark matter, the galaxy's image gets stretched and distorted, like a funhouse mirror.

Astronomers call this Cosmic Shear. By measuring how millions of galaxies are slightly squashed or stretched, they can map the invisible "scaffolding" of the universe.

However, there's a catch. The universe is supposed to stretch galaxies in a very specific, smooth way (called E-modes). If the stretching looks chaotic or swirling (called B-modes), it usually means something is wrong with the telescope or the data, not the universe itself.

The Analogy: Imagine you are trying to measure the wind by watching leaves blow.

E-modes: The leaves all blow gently in the same direction. This is the real wind (the universe).
B-modes: The leaves are swirling in tiny, chaotic circles. This isn't wind; it's the fan in your room blowing on them (the telescope error).

2. The Three Detective Tools

The scientists in this paper didn't just look at the data once. They used three different "detective kits" to check for these swirling errors (B-modes):

The Correlation Map (Pure E/B): They looked at how galaxy shapes relate to each other at different distances.
The COSEBI Filter: Think of this as a musical equalizer. It breaks the data down into different "notes" (modes). If there's a bad noise, it shows up as a specific, repeating pattern in the notes.
The Power Spectrum (Harmonic Space): This looks at the data from a "frequency" perspective, like looking at a sound wave instead of the sound itself.

The Problem: Sometimes, one detective says, "Everything is clean!" while another says, "I see a weird pattern!" This happens because each tool looks at the data differently. If you only use one tool, you might miss a sneaky error hiding in the blind spot of the others.

3. The "CCD Ghost" (The Main Villain)

When the team first looked at their data, they found a spooky pattern. It looked like a repeating ripple, like a drumbeat, across the entire sky.

They realized this wasn't the universe. It was their camera.

The Camera: The telescope uses a giant digital sensor made of 40 smaller chips (CCDs).
The Glitch: Each chip had a tiny, fixed imperfection. It was like if every tile in a mosaic floor was slightly tilted the same way.
The Result: Because the camera has 40 tiles, the error repeated every time the telescope pointed at a new spot. This created a "ghost" pattern in the data that looked like swirling B-modes.

The Fix: The scientists realized that if they ignored the very smallest distances between galaxies (where the camera tiles are most visible) and the very largest distances (where the sky is too fuzzy), they could cut out the "ghost."

4. The Great Filter Test

The team tried different versions of their data catalog (a list of galaxies):

Version A (Loose): Included every galaxy they could find. Result: Failed the test. The "ghost" was too loud.
Version B (Masked): They covered up the bright stars (which cause glare). Result: Failed the test. Covering up stars actually made the "ghost" pattern worse in some ways because it created weird gaps in the map.
Version C (Relaxed Flags): They included galaxies that were partially cut off or blended together. Result: Failed the test.
Version D (The Winner - Size Cut): They kept only the galaxies that were big and clear enough to measure perfectly, ignoring the tiny, blurry ones. Result: Success! All three detective tools agreed: "No more ghosts. The data is clean."

5. The Big Lesson

The most important takeaway from this paper is don't trust just one test.

In the past, scientists might have used just one method to check their data. If that one method said "Pass," they would move on. This paper shows that different methods can hide errors from each other.

The Metaphor: Imagine you are checking a car for safety. You check the brakes, then the tires, then the engine. If you only check the brakes, you might miss a flat tire. You need to check everything to be sure the car is safe.

By demanding that all three detective tools pass the test simultaneously, the UNIONS team found the "sweet spot" where the data is clean enough to trust.

The Conclusion

This paper is a victory lap for the UNIONS-3500 project. They successfully cleaned their "dirty window," identified the "ghost" caused by their camera's tiles, and found the perfect settings to ignore the noise.

Now, they can look at the universe with confidence, knowing that the patterns they see are real cosmic structures, not just smudges on their lens. This paves the way for the next generation of telescopes (like Euclid and LSST) to map the universe with even greater precision.

1. Problem Statement

Cosmic shear measurements, which map the distribution of dark matter via the distortion of distant galaxy shapes, are currently limited by systematic errors rather than statistical noise (Stage-III sensitivity).

The Core Issue: In the absence of cosmological B-modes (curl-like shear patterns) at current sensitivity, any detected B-mode power is an unambiguous indicator of residual observational systematics (e.g., Point Spread Function (PSF) leakage, detector defects, or astrometric errors).
The Challenge: Different statistical frameworks (real-space correlation functions, COSEBIs, and harmonic-space power spectra) weight angular scales and treat "ambiguous modes" (patterns that cannot be uniquely classified as E or B on a finite interval) differently. Consequently, a dataset might pass a null test in one framework while failing in another.
Goal: To validate the UNIONS-3500 survey (2,894 deg², $n_{eff} \approx 5.0$ arcmin $^{-2}$ ) by identifying and mitigating systematic contamination to ensure the data is suitable for precision cosmological inference (Paper III and IV).

2. Methodology

The authors employed a multi-framework approach to validate the shear catalog, utilizing three distinct E/B-separable statistics to cross-check for systematics.

A. Data Variants

The analysis tested four versions of the ShapePipe v1.4 catalog to isolate the source of contamination:

Initial: Loose size cuts ( $r_{h,gal}/r_{h,psf} > 0.5$ ).
Size-cut (Fiducial): Tightened size cuts ( $r_{h,gal}/r_{h,psf} > 0.707$ ) to remove unreliable shape measurements.
Masked: Size-cut catalog plus masking of bright ( $r<12$ ) and faint ( $12<r<18$ ) stellar halos.
Relaxed-flags: Size-cut catalog with relaxed SExtractor flags to include deblended objects.

B. Statistical Frameworks

The team computed B-mode diagnostics using three complementary methods:

Pure-mode Correlation Functions ( $\xi^E/B_\pm$ ): Decomposes the measured shear correlation functions into pure E, pure B, and ambiguous components using integral transforms (Schneider et al. 2022).
COSEBIs (Complete Orthogonal Sets of E/B-mode Integrals): Compresses the correlation functions into discrete orthogonal modes ( $B_n$ ). This method is sensitive to oscillatory patterns in angular space.
Harmonic-space Power Spectra ( $C^{BB}_\ell$ ): Estimates the B-mode power spectrum directly from the catalog using a catalog-based estimator (Wolz et al. 2025) to avoid pixelization artifacts.

C. Validation Metrics

Probability-to-Exceed (PTE): For each statistic, the authors computed the PTE of the B-mode data vector against a null hypothesis (zero B-modes) across a 2D grid of scale-cut boundaries ( $\theta_{min}, \theta_{max}$ or $\ell_{min}, \ell_{max}$ ).
Covariance: Gaussian-only covariances were used (conservative approach), derived from semi-analytical propagation and simulations (GLASS).
Blinding: The analysis was conducted within a blinded framework, with scale cuts selected based on B-mode stability rather than cosmological signal optimization.

3. Key Results

A. Detection of Systematic Contamination

Initial Catalogs: All catalog versions showed significant B-mode failures on the full angular range ($1-250$ arcmin).
COSEBI Oscillations: A distinct oscillatory pattern was observed in the COSEBI $B_n$ modes (up to $n=20$ ), with the first mode ( $n=1$ ) exceeding $4\sigma$ . This pattern is characteristic of repeating additive shear bias (a constant ellipticity offset per CCD) seen in previous MegaCam surveys like CFHTLenS.
Scale Dependence: The contamination was localized to CCD angular scales (6–14 arcmin).
- Pure-mode: Showed excesses at small and large scales.
- Harmonic: Showed low-level positive offsets and outliers at specific multipoles.

B. Impact of Catalog Refinements

Size Cuts: Tightening the galaxy size cut ( $r_{h,gal}/r_{h,psf} > 0.707$ ) was the most effective mitigation. It suppressed the B-mode power in $\xi^B_-$ and harmonic space, allowing the Fiducial (Size-cut) catalog to pass all null tests.
Stellar Masking: Contrary to expectations, masking stellar halos contracted the acceptable PTE region. The irregular gaps introduced by masking increased the fraction of ambiguous modes, destabilizing the pure-mode estimates and causing failures in $\xi^B_+$ and $\xi^B_-$ , despite passing in harmonic space.
Relaxed Flags: Including deblended objects led to failures in $\xi^B_-$ and harmonic space, suggesting these objects contribute to the systematic signal.

C. Scale Cuts and Final Validation

The authors adopted scale cuts of 12–83 arcmin (configuration space) and 300 < $\ell$ < 1600 (harmonic space).

Result: With these cuts, the Fiducial catalog passed all three statistics simultaneously (Minimum PTE = 0.18).
Suppression: The lower cut ($12$ arcmin) effectively excludes the CCD-scale separations where the repeating additive bias is strongest, suppressing the oscillatory COSEBI signature to $<1\sigma$ .

4. Key Contributions

Multi-Framework Validation: The paper demonstrates that relying on a single statistic is insufficient. A dataset can pass a harmonic-space test while failing a real-space or COSEBI test because the filter functions weight contaminated scales differently. The authors establish a rigorous protocol requiring simultaneous passage across all frameworks.
Identification of MegaCam Systematics: The study confirms that the oscillatory B-mode pattern observed in UNIONS is consistent with the detector-level additive bias previously identified in CFHTLenS, linking it to the MegaCam camera's focal plane geometry.
Catalog Optimization: The analysis proves that tightening galaxy size cuts is more effective for B-mode mitigation than stellar masking for this specific dataset, as masking introduced new ambiguities.
Methodological Insight: The authors show that transforming harmonic-space bandpowers into COSEBIs yields the same oscillatory failure, proving that the sensitivity to systematics is driven by the filter functions, not the basis (real vs. harmonic space).

5. Significance

Stage-IV Readiness: As surveys move to Stage-IV (e.g., Euclid, LSST), systematic errors will dominate the error budget. This paper provides a blueprint for "multi-framework validation," ensuring that scale cuts are robust against different statistical projections of the same underlying contamination.
UNIONS-3500 Cosmology: The validated scale cuts and catalog selection are directly applied to the cosmological constraints presented in the companion papers (Paper III and IV), ensuring that the resulting $S_8$ measurements are not biased by the identified CCD-scale systematics.
Diagnostic vs. Correction: The authors argue against modeling and subtracting B-modes as a bias. Instead, they treat B-modes as a diagnostic: if a scale cut removes the B-mode contamination, it inherently removes the corresponding E-mode contamination without needing to assume a specific E/B ratio model.

In summary, this paper successfully validates the UNIONS-3500 shear catalog by identifying a detector-level systematic, demonstrating the necessity of cross-statistic validation, and establishing conservative scale cuts that enable robust cosmological inference.