🔭 astrophysics

Constraining baryonic feedback and cosmology from DES Y3 and Planck PR4 6 $\times$ 2pt data. I. $Λ$ CDM models

By combining DES Y3 and Planck PR4 6 $\times$ 2pt data with a neural network emulator, this study constrains cosmology and baryonic feedback, finding that the results favor weak feedback scenarios that are in significant tension with certain hydrodynamical simulations like Illustris and OWLS.

Original authors: Jiachuan Xu, Tim Eifler, Elisabeth Krause, Vivian Miranda, Jaime Salcido, Ian McCarthy

Published 2026-02-10

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Jiachuan Xu, Tim Eifler, Elisabeth Krause, Vivian Miranda, Jaime Salcido, Ian McCarthy

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 solve a massive, cosmic jigsaw puzzle. You have pieces from two different sets: one set describes the "skeleton" of the universe (Dark Matter and gravity), and the other set describes the "flesh and blood" (Baryons, like gas and stars).

The problem? The "flesh" is messy. It doesn't just sit where gravity tells it to; it gets blown around by exploding stars and supermassive black holes, like wind scattering leaves in a park. If you don't account for that "wind," your map of the universe's skeleton will be wrong.

This paper is about a team of scientists who have developed a high-tech way to account for that cosmic wind so they can see the skeleton more clearly.

1. The Problem: The "S8 Tension"

Scientists are currently having a bit of an argument called the $S_8$ tension.

Group A looks at the "Early Universe" (using light from the Big Bang, via the Planck satellite) and predicts how "clumpy" the universe should be today.
Group B looks at the "Late Universe" (using weak gravitational lensing, via the Dark Energy Survey) and sees a universe that is much smoother and less clumpy than expected.

It’s like two people looking at a photo of a crowd: one person predicts a dense, packed stadium, but the other person sees people spread out comfortably. Is the prediction wrong, or is something "smoothing out" the crowd?

2. The Culprit: Baryonic Feedback (The Cosmic Wind)

The scientists suspect the "smoothing" is caused by Baryonic Feedback. Think of supermassive black holes at the centers of galaxies as giant leaf blowers. As they gobble up matter, they blast energy back out, pushing gas away from the center of galaxies. This "wind" pushes matter around, making the universe look less clumpy to our telescopes.

3. The Solution: The "Smart Emulator" and the "Cheat Sheet"

To solve this, the researchers did two brilliant things:

The Smart Emulator (The AI Assistant): Calculating how gravity and "wind" interact is incredibly slow and difficult for even the best supercomputers. To speed things up, the team trained a Neural Network (AI). Instead of doing the math from scratch every time, the AI acts like a "cheat sheet" that can instantly predict how the universe should look under different conditions. This made their work 1,000 times faster.
The Baryon Cheat Sheet (The Prior): They used X-ray data from galaxy clusters to measure exactly how much "flesh" (baryon mass) is actually present in these cosmic structures. By knowing how much matter is there, they could better estimate how strong the "wind" must be.

4. The Results: A Smoother Universe

By combining all their data (weak lensing, galaxy clustering, and the Cosmic Microwave Background) into one giant "6x2pt" analysis, they found:

The Wind is relatively weak: The "leaf blowers" (black holes) are active, but they aren't blowing the universe apart as violently as some older models suggested.
The Tension remains, but it's manageable: Their results for how clumpy the universe is ( $S_8$ ) are much closer to the "Early Universe" predictions than previously thought. They found that when you account for the "wind" correctly, the two groups of scientists are actually starting to agree.

Summary in a Nutshell

If the universe is a forest, scientists have been trying to map the trees (Dark Matter) while ignoring the wind (Baryons) blowing the leaves around. This paper provides a much better "wind sensor" and a super-fast AI to help them map the forest accurately, proving that the universe is behaving mostly as we expected, once you account for the messy business of exploding stars and hungry black holes.

Technical Summary: Constraining Baryonic Feedback and Cosmology from DES Y3 and Planck PR4 $6\times2\text{pt}$ Data

1. Problem Statement

The paper addresses two interconnected challenges in modern precision cosmology:

The $S_8$ Tension: There is a persistent discrepancy between the amplitude of matter fluctuations ( $S_8 \equiv \sigma_8 \sqrt{\Omega_m/0.3}$ ) measured by low-redshift weak lensing surveys (which tend to find lower values) and the predictions from primary Cosmic Microwave Background (CMB) data (which find higher values).
Baryonic Feedback Uncertainty: On small scales, baryonic processes (such as AGN feedback and stellar winds) redistribute matter, suppressing the matter power spectrum $P(k)$ . If not modeled accurately, this suppression can bias cosmological parameter estimation, potentially mimicking or masking new physics.

2. Methodology

The authors perform a joint "6×2pt" analysis, combining six two-point correlation functions (2PCFs) to maximize constraining power and break degeneracies:

Probes: Cosmic shear ( $\xi_{\pm}$ ), projected galaxy clustering ( $w_g$ ), CMB lensing band-power ( $C_\kappa$ ), and their cross-correlations (galaxy-galaxy lensing $\gamma_t$ , galaxy-CMB lensing $w_{g\kappa}$ , and galaxy shear-CMB lensing $w_{s\kappa}$ ).
Datasets: Dark Energy Survey (DES) Year 3 (Maglim and Metacalibration catalogs) and Planck satellite PR4 CMB lensing maps.
Baryonic Modeling (PCA Approach): To mitigate feedback effects without assuming a specific physical model, the authors use Principal Component Analysis (PCA). They construct PCs from the ANTILLES simulation suite (400 hydrodynamical simulations). The model vector is adjusted by varying the amplitudes ( $Q_i$ ) of these PCs.
Baryon Mass Fraction Prior: A key innovation is the use of an external prior on the first PC ( $Q_1$ ). They establish a nonlinear scaling relation between $Q_1$ and the mean normalized baryon mass fraction ( $\bar{Y}_b$ ) derived from X-ray/optical observations (Akino et al. 2022).
Machine Learning Acceleration: To handle the high dimensionality and the need for extensive systematic testing, they train a neural network emulator (ResMLP). This speeds up the MCMC inference by approximately $10^3$ , allowing for hundreds of simulated analyses to validate robustness.

3. Key Contributions

Upgraded Pipeline: Significant improvements over previous studies (e.g., X24) by using the more constraining DES Y3 Maglim catalogs, Planck PR4 data, and the superior ANTILLES simulation suite.
Self-Consistent Baryon Calibration: They successfully linked the PCA-based feedback parameters to physically observable quantities (baryon mass fractions in $10^{13}–10^{14} M_\odot$ halos), providing a robust way to incorporate astrophysical information into cosmological fits.
Systematics Validation: The study provides an exhaustive assessment of the "crosstalk" between different systematics, including Intrinsic Alignment (IA), Point Spread Function (PSF) bias, and nonlinear matter power spectrum modeling (e.g., Halofit vs. EuclidEmulator2).

4. Results

Cosmological Constraints:
- The baseline $6\times2\text{pt}$ analysis (with the $\bar{Y}_b$ prior) yields $S_8 = 0.8073 \pm 0.0094$ .
- When combined with external geometric and CMB probes (Planck/ACT and DESI DR2 BAO + BBN), the constraints tighten significantly to $S_8 = 0.8079 \pm 0.0073$ and $\Omega_m = 0.2995 \pm 0.0034$ .
Baryonic Feedback Strength:
- The analysis finds evidence for weak baryonic feedback. The amplitude of the first PC is measured at $Q_1 = 0.025^{+0.024}_{-0.029}$ (for $6\times2\text{pt}$ alone).
- This result strongly excludes high-feedback models like Illustris ( $Q_1 = 0.095$ ) and OWLS AGN T8.7 ( $Q_1 = 0.137$ ) at the $2.9\sigma$ to $5.9\sigma$ levels.
$S_8$ Tension Status: The results show good consistency (within $\sim 1.35\sigma$ ) between the $6\times2\text{pt}$ analysis and primary CMB measurements (P-ACT), suggesting that the $S_8$ tension is not resolved by baryonic feedback alone.

5. Significance

This paper demonstrates that by combining multiple probes and using advanced machine-learning-accelerated PCA modeling, one can push cosmological constraints to much smaller scales (down to $2.5'$ ) without being biased by baryonic physics. It provides a benchmark for future surveys (DES Y6, LSST, Euclid), highlighting that while current systematics are controlled, the next generation of data will require even more precise and robust mitigation strategies to avoid "crosstalk" between astrophysics and cosmology.