🔭 astrophysics

KiDS-1000: Improved constraints on cosmology, intrinsic alignments and baryonic feedback from clipped cosmic shear

Using the fourth data release of the Kilo-Degree Survey (KiDS-1000), this study employs "clipped" shear correlation functions modeled with Gaussian process emulators to simultaneously constrain cosmological parameters, intrinsic alignments, and baryonic feedback, achieving significantly tighter constraints on $S_8$ and $w_0$ compared to conventional methods while remaining consistent with the standard $\Lambda$ CDM model.

Original authors: Benjamin Giblin, Joachim Harnois-Déraps, Louise Paquereau, Catherine Heymans

Published 2026-08-10

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Benjamin Giblin, Joachim Harnois-Déraps, Louise Paquereau, Catherine Heymans

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 the universe as a giant, invisible ocean made of dark matter, stretching from the Big Bang to today. We can't see this ocean directly, but we can see how it bends the light from distant galaxies, much like how a funhouse mirror distorts your reflection. This bending of light is called "weak gravitational lensing," and by measuring the tiny, stretched shapes of billions of galaxies, astronomers can map out the invisible structures of the cosmos. This map helps us answer the biggest questions in physics: What is the universe made of? How fast is it expanding? And what is this mysterious "dark energy" pushing everything apart?

However, looking at this cosmic ocean is tricky. The water isn't perfectly clear; it's filled with "bubbles" of normal matter (like stars and gas) that interact in messy ways, and the galaxies themselves have a tendency to line up in patterns that aren't caused by gravity. These messy bits are called "systematics," and they act like static on a radio, making it hard to hear the true signal of the universe's shape. For a long time, scientists have tried to filter out this static, but the most interesting parts of the universe—the dense, clumpy regions where galaxies form—are also the noisiest parts. This paper is about a clever new way to turn down the volume on the noise without losing the music.

The Paper's Story: Cutting the Clutter

This paper, written by a team of astronomers using data from the "KiDS-1000" survey, introduces a technique called "clipping" to get a clearer picture of the universe. Think of the universe's matter distribution like a mountain range. Most of the time, we look at the whole landscape, including the tiny pebbles and the massive, jagged peaks. But the massive peaks are where the "noise" (like gas and stars pushing back against gravity) is strongest, making it hard to measure the true shape of the mountains.

The authors decided to try something radical: they "clipped" the map. Imagine taking a pair of scissors and snipping off the very tops of the highest mountains, flattening them out. By removing these extreme high-density regions, they effectively filtered out the messy, hard-to-model parts of the universe. They then compared this "clipped" map with the original, "unclipped" map.

The team didn't just guess that this would work; they built a massive digital playground to test it first. They used supercomputer simulations to create thousands of fake universes, complete with all the messy bits like galaxy alignments and gas feedback. They trained a smart computer program (an "emulator") on these simulations to learn how the clipped and unclipped maps should look under different rules of physics. This allowed them to model the messy parts of the real data with much higher precision than before.

What They Found

When they applied this method to the real data from 21 million galaxies, the results were a significant improvement. By combining the "clipped" and "unclipped" views, they tightened their measurements of the universe's properties. Specifically, they found that the uncertainty in a key number called $S_8$ (which measures how "clumpy" the universe is) shrank by 16%. They also improved the measurement of the dark energy equation of state ( $w_0$ ) by 24%.

One of the most exciting discoveries was that this method helped them understand the "messy bits" better. The clipping technique acted like a spotlight, helping them place a strict upper limit on how strong the feedback from stars and black holes (baryonic feedback) could be. It also improved their understanding of how galaxies align with each other (intrinsic alignments) by 27%, a huge leap forward in controlling for that specific type of noise.

The Verdict

The authors found that their new, clipped measurements are consistent with the standard model of cosmology ( $\Lambda$ CDM) and match up well with previous major studies, even though they used a completely different way of calculating the numbers. They didn't find a new, strange universe; instead, they found a sharper, more precise version of the one we already knew.

The paper suggests that this "clipping" trick is a powerful tool that moves beyond simple theory and works on real, messy data. While there are still some small disagreements with other studies regarding the exact alignment of galaxies, the overall picture is solid. The authors conclude that as we get better data from future telescopes, this method of "trimming the peaks" will be a vital tool for unlocking the secrets of the dark universe, turning a blurry, noisy photo into a crystal-clear portrait of our cosmic home.

Technical Summary: KiDS-1000: Improved constraints on cosmology, intrinsic alignments and baryonic feedback from clipped cosmic shear

Problem Statement
The Standard Model of Cosmology ( $\Lambda$ CDM) faces persistent tensions regarding the values of key parameters, particularly the Hubble constant ( $H_0$ ) and the matter clustering amplitude ( $S_8$ ). While weak gravitational lensing (cosmic shear) is a primary probe for these parameters, traditional analyses relying on two-point statistics (e.g., shear correlation functions) are sub-optimal because they extract information only up to the second order. The late-time matter distribution probed by lensing is highly non-Gaussian, containing higher-order information that two-point statistics fail to fully utilize. Furthermore, the small scales where this non-Gaussian information resides are heavily contaminated by baryonic feedback and intrinsic alignments (IA), making robust modeling difficult. Previous attempts to use higher-order statistics have struggled to jointly model these systematics with sufficient accuracy. Additionally, while "clipping" (removing high-density regions from the projected field) has been proposed to decorrelate Fourier modes and improve information extraction from two-point statistics, previous applications to real lensing data (e.g., Giblin et al. 2018) lacked comprehensive modeling of systematics such as intrinsic alignments, baryonic feedback, and photometric redshift uncertainties.

Methodology
This paper presents the first cosmological constraints from the KiDS-1000 survey using clipped shear correlation functions, employing a fully forward-modeling approach based on $N$ -body simulations and Gaussian Process (GP) emulators.

Data and Simulations: The analysis utilizes the fourth data release of the Kilo-Degree Survey (KiDS-1000), covering 777 deg $^2$ with 21 million galaxies. The authors rely on the cosmoSLICS and SLICS suites of dark-matter-only and hydrodynamical $N$ -body simulations. These simulations offer significantly larger volumes and higher resolution than previous datasets used for clipped lensing.
Clipping Procedure: The methodology follows the approach of Giblin et al. (2018). Convergence ( $\kappa$ ) maps are generated from ellipticity maps using the Kaiser-Squires mass mapping technique. Regions exceeding a threshold $\kappa_c = 0.010$ are "clipped" (set to the threshold value), creating a clipped map ( $\kappa_{clip}$ ) and a residual map ( $\Delta\kappa$ ). The residual map is inverted to obtain residual ellipticities, which are subtracted from the observed ellipticities to produce clipped ellipticities.
Systematic Modeling: A critical advancement is the comprehensive modeling of systematics. The authors train GP emulators on variations of the simulations to predict the dependence of both clipped and unclipped shear correlation functions on:
- Cosmology: A 4-dimensional $w$ CDM parameter space ( $\Omega_m, S_8, h, w_0$ ).
- Intrinsic Alignments (IA): Modeled using the non-linear linear alignment (NLA) model with amplitude $A_{IA}$ .
- Baryonic Feedback (BF): Modeled using hydrodynamical magneticum simulations, scaled by a free nuisance parameter $b_{bary}$ .
- Photometric Redshifts ( $z$ ): Modeled via shifts ( $\Delta z$ ) to the redshift distributions.
- Source-Lens Clustering (SLC): Investigated and found to be negligible for this statistic.
Emulation Strategy: GP emulators are trained to predict the cosmological and systematic biases relative to a fiducial model. The emulators are validated via leave-one-out cross-validation, showing sub-dominant errors compared to statistical noise. The emulator uncertainty is folded into the covariance matrix to prevent bias.
Parameter Inference: The analysis uses a Gaussian likelihood (with Hartlap correction for the inverse covariance) and the nautilus sampler to explore the 11-dimensional parameter space (4 cosmological + 7 nuisance). The data vector consists of auto- and cross-correlations for five tomographic redshift bins, measuring both clipped and unclipped $\xi_+$ and $\xi_-$ , resulting in a vector of length 540.

Key Contributions

First Systematics-Controlled Clipping Analysis: This is the first application of clipping to real weak lensing data that simultaneously models intrinsic alignments, baryonic feedback, photometric redshift uncertainties, and source-lens clustering.
Forward Modeling with Simulations: Unlike previous analyses that relied on theoretical prescriptions (e.g., hmcode) for unclipped statistics, this work forward-models both clipped and unclipped statistics using $N$ -body simulations, providing an independent validation of the KiDS-1000 cosmic shear results (Asgari et al. 2021).
Tomographic Clipping: The analysis introduces tomographic redshift binning within a clipped lensing framework for the first time, expanding the data vector significantly compared to previous proof-of-concept studies.
Handling Masking Bias: By replicating the KiDS-1000 survey footprint (including masking) directly in the simulation tiles, the authors forward-model the masking bias, allowing the use of both $\xi_+$ and $\xi_-$ components of the clipped statistic, which were previously biased by masking.

Results
The combination of clipped and unclipped probes yields tighter constraints on cosmological and nuisance parameters compared to the unclipped approach alone:

Cosmological Constraints: The analysis yields $\Omega_m = 0.263^{+0.035}_{-0.038}$ $Ω_{m} = 0.26 3_{- 0.038}^{+ 0.035}$ , $S_8 = 0.724^{+0.027}_{-0.027}$ $S_{8} = 0.72 4_{- 0.027}^{+ 0.027}$ , and $w_0 = -1.24^{+0.26}_{-0.28}$ $w_{0} = - 1.2 4_{- 0.28}^{+ 0.26}$ .
- The uncertainty on $S_8$ is tightened by 16%.
- The uncertainty on $w_0$ is tightened by 24%.
- The Figure of Merit (FoM) in the $\Omega_m-S_8$ plane improves by a factor of 1.2.
Systematic Constraints:
- The intrinsic alignment amplitude ( $A_{IA}$ ) constraint improves by 27%, reaching a precision of ~39% (compared to ~90% in the unclipped analysis).
- The baryonic feedback parameter ( $b_{bary}$ ), which is unconstrained in the unclipped analysis, is constrained to an upper limit of $b_{bary} < 0.97$ in the combined analysis.
Consistency: The results are consistent with the $\Lambda$ CDM model and the main KiDS-1000 cosmic shear analysis (Asgari et al. 2021), despite using a completely independent simulation- and emulator-based framework. The goodness-of-fit ( $p$ -values) is acceptable for all analyses.
Tensions: The $S_8$ constraint is in mild tension (1.8 $\sigma$ ) with the A21 result and 3.0 $\sigma$ with Planck CMB results, though the authors note that differences in redshift calibration and survey specifications in newer data (KiDS-Legacy) may shift these values.

Significance and Claims
The paper claims that clipping transforms cosmic shear from a "proof-of-concept" technique into a competitive and robust lensing statistic. By filtering out high-density regions, clipping allows two-point statistics to extract additional cosmological information from larger, linear scales where baryonic contamination is weaker, while the complementary information from the unclipped statistic helps break degeneracies.

The authors emphasize that the complementary nature of the clipped and unclipped probes allows for the decoupling of scale-dependent processes. This is evidenced by the significant improvement in constraining the intrinsic alignment amplitude and placing an upper limit on baryonic feedback strength—parameters that are difficult to constrain with unclipped statistics alone. The work demonstrates that simulation-based forward modeling can successfully control systematics in higher-order or modified lensing statistics, validating the approach for future Stage-IV surveys like Euclid and LSST. The authors modestly conclude that while challenges remain (e.g., emulator inaccuracies at small scales), the remaining obstacles to fully utilizing clipping are likely short-lived given upcoming advancements in data quality and simulation resources.

More like this