🔭 astrophysics

Comparing cosmic shear nulling methods for Stage-IV surveys

This paper evaluates three nulling strategies—BNT transform, LU factorization (LUnul), and cross-correlation with LSS tracers—for mitigating baryon feedback bias in Stage-IV cosmic shear surveys, finding that while all effectively reduce biases in cosmological parameters like $S_8$ , they differ significantly in their trade-offs between bias reduction, information preservation, and theoretical rigor.

Original authors: Naomi Clare Robertson, Alex Hall

Published 2026-05-06

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Naomi Clare Robertson, Alex Hall

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

Imagine you are trying to take a crystal-clear photograph of a distant mountain range (the universe) to measure its shape and size. However, there is a thick, swirling fog (baryon feedback) swirling around the base of the mountains. This fog is caused by energetic processes like exploding stars and supermassive black holes. While the fog is fascinating in its own right, it distorts the view of the mountains, making your measurements of the mountain's true shape inaccurate.

This paper is about three different strategies to "dissipate the fog" so astronomers can get a truer picture of the universe, specifically for the massive, next-generation telescopes (Stage-IV surveys) that will soon be scanning the entire sky.

Here is a breakdown of the problem and the three solutions the authors tested, using simple analogies:

The Problem: The "Fog" of Baryon Feedback

In astronomy, "cosmic shear" is a way of measuring how the gravity of invisible dark matter bends the light from distant galaxies. It's like looking at a funhouse mirror to see how much the glass is warped.

However, the "fog" (baryon feedback) changes how matter clumps together on small scales. If you try to measure the universe using these small, foggy details, your math gets messed up, and you might conclude the universe is expanding at the wrong speed or has the wrong amount of dark energy.

The usual fix is to simply throw away the "foggy" parts of the photo (cutting out the small scales). But this is like throwing away half your photo to get a clear image; you lose a lot of valuable information and your measurements become less precise.

The Three "Fog-Clearing" Methods

The authors tested three clever mathematical tricks to remove the fog's influence without throwing away the whole photo.

1. The "LU Nulling" Method (The Aggressive Filter)

The Analogy: Imagine you have a choir singing a song. The fog is a loud, low-frequency hum that everyone is singing along to. The LU Nulling method is like a sound engineer who takes the microphones, subtracts the hum from every singer's voice, and then recombines them.

How it works: It uses a mathematical technique (LU factorization) to look at the data and mathematically cancel out the specific parts of the signal that come from the "foggy" small scales.
The Result: It is the most aggressive and effective at removing the bias (the distortion). It can look at smaller, fuzzier details than the other methods without getting the answer wrong.
The Catch: In its eagerness to remove the fog, it also removes some of the "good" sound (information). It's very accurate, but the final measurement isn't as sharp or precise as it could be.

2. The "BNT Transform" (The Rearranger)

The Analogy: Imagine the photo of the mountains is taken through a lens that smears the foreground (the fog) into the background (the mountains). The BNT Transform is like a digital tool that reorganizes the pixels. It doesn't delete anything; it just rearranges the data so that the "foreground fog" and "background mountains" are separated into different layers.

How it works: It reorganizes the data so that the signal from nearby galaxies (where the fog is) is separated from the signal of distant galaxies. This makes the relationship between the "size" of the detail and its location much clearer.
The Result: It keeps almost all the information (it's "lossless"). It's a very rigorous, scientifically solid method.
The Catch: It works best if you have a huge number of "slices" of the universe to work with. Since the current setup only has a few slices (6 bins), this method didn't show a massive advantage over the standard way of doing things in this specific test, though it promises to be very powerful for future surveys with more data.

3. The "Cross-Correlation" Method (The Reference Check)

The Analogy: Imagine you are trying to measure the wind speed on a mountain, but the wind is gusting wildly near the ground (the fog). To get a true reading, you ask a friend standing on a nearby hill (a different set of galaxies) to tell you how much the wind is blowing there. You then use their report to subtract the local wind noise from your own measurement.

How it works: This method uses a separate map of galaxy positions (the "friend on the hill") to identify and subtract the contribution of the nearby, foggy structures from the main lensing data.
The Result: It is very good at reducing the bias for the most important measurements (like $S_8$ ).
The Catch: It requires you to have that extra "friend" (data from galaxy clustering). Also, if your friend and you are standing too close together (overlap between lens and source), the method gets confused and less effective.

What Did They Find?

The authors ran a simulation to see which method gives the best balance between accuracy (not being wrong) and precision (being very sure of the answer).

The Trade-off: In all cases, trying to remove the fog's influence makes the measurements slightly less precise. You can't have it both ways perfectly.
The Winner for Accuracy: The LU Nulling method was the best at stopping the measurements from being biased. It allowed them to look at smaller scales without getting the wrong answer.
The Winner for Information: The BNT Transform kept the most information intact, though it needed more data slices to really shine.
The Special Case: The Cross-Correlation method worked well but required extra data that isn't always available in the same way.

The Bottom Line

The paper concludes that while we can't perfectly model the "fog" of baryon feedback yet, these three mathematical "fog-clearing" tools are effective. They allow future telescopes to use more of the data they collect without getting tricked by the small-scale chaos of the universe.

The authors suggest that for the upcoming massive surveys, using these methods (especially LU Nulling or BNT) will help scientists get unbiased answers about Dark Energy, even if they have to throw away a little bit of statistical "sharpness" to do it. It's a trade-off: a slightly fuzzier picture is better than a sharp picture that tells you the wrong story.

Technical Summary: Comparing Cosmic Shear Nulling Methods for Stage-IV Surveys

Problem Statement
Cosmic shear measurements from Stage-IV surveys (e.g., Euclid, Rubin Observatory) aim to constrain Dark Energy physics at the percent level. However, these measurements are contaminated by baryon feedback processes (e.g., AGN feedback, supernovae), which suppress the matter power spectrum at small scales ( $k \gtrsim 10 \, \mathrm{Mpc}^{-1}$ ). Current hydrodynamical simulations predict suppression amplitudes varying significantly (10% to >30%), and sub-grid physics prescriptions remain uncertain. Standard analysis mitigates this by applying angular scale cuts (limiting $\ell_{\max}$ ), but this approach is suboptimal because a given angular scale integrates contributions from a wide range of physical 3D scales due to the broad radial lensing kernel. Consequently, aggressive scale cuts are required to ensure unbiased cosmological constraints, leading to a significant loss of statistical precision. This paper investigates whether "nulling" transformations applied to the data vector can more effectively isolate and remove high- $k$ (small-scale) contributions compared to standard scale cuts.

Methodology
The authors perform a Fisher forecast analysis comparing three distinct nulling strategies against a standard, un-nulled cosmic shear analysis. The analysis utilizes a mock data vector based on a Euclid-like survey setup with six tomographic source bins, a galaxy number density of 30 galaxies/arcmin $^2$ , and a sky area of 15,000 deg $^2$ . Baryon feedback is modeled using HMCode2020 with a fiducial AGN heating parameter ( $\Theta_{\mathrm{AGN}} = 7.8$ ). The authors assume a $w_0w_a$ CDM cosmology and fix all systematics (e.g., photometric redshifts, shear bias) to zero to isolate the impact of baryon feedback.

The three methods compared are:

LU Nulling (LUnul): A new implementation of a method originally proposed by Huterer & White (2005). It treats the Limber integral as a discrete matrix multiplication. By performing an LU factorization of the kernel matrix, the method constructs linear combinations of the tomographic power spectra. This effectively cancels the common high- $k$ contributions (which are similar across bins at low redshift) while retaining differences at lower $k$ . The method allows for a "soft" cut where specific rows of the transformed data vector are discarded based on a tolerance parameter ($tol=0.01$) for high- $k$ contributions.
BNT Transform: The Bernardeau-Nishimichi-Taruya transform operates on the lensing field (shear maps). It uses a linear, invertible transformation of the tomographic bin indices to localize the lensing kernels in redshift space. This creates a cleaner mapping between angular multipole $\ell$ and 3D wavenumber $k$ , suppressing cross-correlations between non-adjacent bins.
Cross-Correlation: Adapted from CMB lensing literature, this method uses a foreground galaxy clustering sample (low-redshift lens bins) to decorrelate the shear signal. By subtracting the projected density field (weighted by the cross-correlation spectrum) from the shear field, the contribution from low-redshift, small-scale structures is suppressed. In this study, the two lowest redshift bins are repurposed as lenses to "clean" the three highest redshift source bins.

Key Contributions

Implementation of LUnul: The authors provide a public implementation of the LU factorization method for cosmic shear, a technique not previously explored in the literature since its initial proposal.
Application of Cross-Correlation to Cosmic Shear: The paper adapts the cross-correlation nulling technique, previously used for CMB lensing, to the context of Stage-IV cosmic shear combined with photometric galaxy clustering.
Systematic Comparison: The study offers a direct comparison of these three distinct approaches (field-based, statistics-based, and tracer-based) within a unified Fisher forecast framework, evaluating their efficacy in reducing bias while retaining information.

Results
The analysis evaluates the trade-off between parameter bias (accuracy) and error bars (precision) across a range of maximum wavenumbers ( $k_{\max}$ ):

Bias Reduction: All three methods effectively reduce the bias on cosmological parameters ( $S_8, \Omega_m, w_0, w_a$ ) caused by unmodeled baryon feedback compared to the standard case.
LUnul Performance: This method is the most aggressive in suppressing high- $k$ modes. It achieves the lowest bias (reducing $S_8$ bias to $<0.5\sigma$ at $k_{\max} = 0.3 \, h/\mathrm{Mpc}$ ), outperforming other methods in accuracy. However, this comes at the cost of precision; the constraining power (Figure of Merit, FoM) is significantly lower than the standard case at comparable scale cuts because it discards more information.
BNT Performance: The BNT transform preserves more information than LUnul and has a rigorous theoretical grounding. However, in this specific setup with only six tomographic bins, it shows only marginal improvement over the standard case. The authors note that BNT is known to be more effective with a larger number of bins.
Cross-Correlation Performance: This method is effective at reducing absolute biases in $S_8$ but requires additional information from galaxy clustering. While it produces tighter error bars (due to the inclusion of density field information), the relative bias (bias divided by error) remains high. Furthermore, it is less constraining on Dark Energy parameters ( $w_0, w_a$ ) compared to other methods. The method is limited by the overlap between lens and source samples in photometric surveys.
Parameter Sensitivity: $S_8$ is consistently the most biased parameter across all methods due to its sensitivity to the amplitude of the power spectrum.

Significance and Claims
The paper concludes that while nulling techniques can significantly mitigate the bias introduced by baryon feedback, they generally do so at the expense of statistical precision. The authors claim that:

LUnul is the most robust method for removing bias and pushing to smaller scales, though it requires careful numerical tuning and sacrifices precision.
BNT is theoretically sound and preserves information but may require more tomographic bins (typical of future Stage-IV data releases) to realize its full potential over standard cuts.
Cross-correlation offers a viable path to reduce bias using existing 3 $\times$ 2-pt data but is limited by the difficulty of perfectly separating lens and source populations in photometric surveys.

The authors emphasize that these methods provide an alternative to complex baryon feedback modeling. They assert that Stage-IV surveys will likely need to employ strong scale cuts or rely on accurate baryon models regardless of the nulling choice, but these nulling strategies offer a way to construct data vectors that are more immune to small-scale astrophysical uncertainties. The results reinforce the established trade-off between accuracy (bias reduction) and precision in cosmic shear analyses.