Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

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

TL;DR

This paper develops a six-probe ($6\times2$pt) framework that combines DES Year 3 weak lensing and galaxy clustering with Planck PR4 CMB lensing (and cross-correlations) to jointly constrain cosmology and baryonic feedback. Baryonic physics is modeled with PCA in the ANTILLES simulation space and calibrated using a baryon-mass-fraction observation-based prior, implemented via a neural emulator to enable extensive robustness testing. The analysis yields precise constraints on $S_8$ and $ar{Y}_b$-driven baryonic feedback, finding a preference for weak feedback over strong scenarios like Illustris or OWLS AGN T8.7, and demonstrates consistency with external primary CMB and geometric priors while showing modest residual tension in $S_8$ under some combinations. The approach highlights the power of ML-accelerated, multi-probe analyses for constraining small-scale baryonic effects and paves the way for applying these methods to future surveys such as DESI, DESI-like LSST data, and enhanced CMB datasets.

Abstract

We combine weak lensing, galaxy clustering, cosmic microwave background (CMB) lensing, and their cross-correlations (so-called 6$\times$2pt) to constrain cosmology and baryonic feedback scenarios using data from the Dark Energy Survey (DES) Y3 Maglim catalog and the Planck satellite PR4 data release. We include all data points in the DES Y3 cosmic shear two-point correlation function (2PCF) down to 2.$^\prime$5 and model baryonic feedback processes via principal components (PCs) that are constructed from the ANTILLES simulations. We find a tight correlation between the amplitude of the first PC $Q_1$ and mean normalized baryon mass fraction $\bar{Y_\mathrm{b}}=\bar{f}_\mathrm{b}/(Ω_\mathrm{b}/Ω_\mathrm{m})$ from the ANTILLES simulations and employ an independent $\bar{Y_\mathrm{b}}$ measurement from Akino et al. (2022) as a prior of $Q_1$. We train a neural network $6\times2$pt emulator to boost the analysis speed by $\mathcal{O}(10^3)$, which enables us to run an impressive number of simulated analyses to validate our analysis against various systematics. For our 6$\times$2pt analysis, we find $S_8=0.8073\pm0.0094$ when including a $Q_1$ prior from $\bar{Y_\mathrm{b}}$ observations. This level of cosmological constraining power allows us to put tight constraints on the strength of baryonic feedback. We find $Q_1=0.025^{+0.024}_{-0.029}$ for our 6$\times$2pt analysis and $Q_1=0.043\pm{0.016}$ when combining with external information from Planck, ACT, DESI. All these results indicate weak feedback, e.g., the tensions to Illustris ($Q_1=0.095$) and OWLS AGN T8.7 ($Q_1=0.137$) are 2.9$σ$-3.3$σ$ and 4.7$σ$-5.9$σ$, respectively.

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

TL;DR

This paper develops a six-probe (pt) framework that combines DES Year 3 weak lensing and galaxy clustering with Planck PR4 CMB lensing (and cross-correlations) to jointly constrain cosmology and baryonic feedback. Baryonic physics is modeled with PCA in the ANTILLES simulation space and calibrated using a baryon-mass-fraction observation-based prior, implemented via a neural emulator to enable extensive robustness testing. The analysis yields precise constraints on and -driven baryonic feedback, finding a preference for weak feedback over strong scenarios like Illustris or OWLS AGN T8.7, and demonstrates consistency with external primary CMB and geometric priors while showing modest residual tension in under some combinations. The approach highlights the power of ML-accelerated, multi-probe analyses for constraining small-scale baryonic effects and paves the way for applying these methods to future surveys such as DESI, DESI-like LSST data, and enhanced CMB datasets.

Abstract

We combine weak lensing, galaxy clustering, cosmic microwave background (CMB) lensing, and their cross-correlations (so-called 62pt) to constrain cosmology and baryonic feedback scenarios using data from the Dark Energy Survey (DES) Y3 Maglim catalog and the Planck satellite PR4 data release. We include all data points in the DES Y3 cosmic shear two-point correlation function (2PCF) down to 2.5 and model baryonic feedback processes via principal components (PCs) that are constructed from the ANTILLES simulations. We find a tight correlation between the amplitude of the first PC and mean normalized baryon mass fraction from the ANTILLES simulations and employ an independent measurement from Akino et al. (2022) as a prior of . We train a neural network pt emulator to boost the analysis speed by , which enables us to run an impressive number of simulated analyses to validate our analysis against various systematics. For our 62pt analysis, we find when including a prior from observations. This level of cosmological constraining power allows us to put tight constraints on the strength of baryonic feedback. We find for our 62pt analysis and when combining with external information from Planck, ACT, DESI. All these results indicate weak feedback, e.g., the tensions to Illustris () and OWLS AGN T8.7 () are 2.9-3.3 and 4.7-5.9, respectively.

Paper Structure

This paper contains 32 sections, 19 equations, 17 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Data
  3. Dark Energy Survey Year 3 data
  4. Planck PR4 CMB Lensing
  5. X-ray measurements
  6. Modeling Baryonic Physics
  7. Baryon feedback simulations
  8. Calibrating baryonic feedback from baryon mass fraction observations
  9. Machine Learning Accelerated Inference
  10. Theory models, covariances - Cobaya-CosmoLike Architecture
  11. Data vector modeling
  12. Covariance modeling
  13. Neural network details and performance
  14. Architecture
  15. Training
  16. ...and 17 more sections

Figures (17)

  • Figure 1: Left panels: matter power spectrum ratio between (i) hydrosims and gravity-only $N$-body sims and (ii) alternative nonlinear power spectrum models and Halofit. The top (bottom) panel shows the ratio at $z=0$ ($z=1$). We show the ratios of ANTILLES simulations in gray solid lines. We show three special ANTILLES simulations in pink, green, and cyan solid lines, whose feedback strength is defined as strong, mean, and weak compared to observation in AEO+22 (see Sec. \ref{['sec:fb_Q1']}). Other hydrosims are shown in dotted-dashed lines: BAHAMAS $\mathrm{log}\,T_\mathrm{AGN}=7.8$ (blue), $\mathrm{log}\,T_\mathrm{AGN}=7.6$ (orange), $\mathrm{log}\,T_\mathrm{AGN}=8.0$ (green), TNG100 (red), Illustris (purple), and Eagle (brown). Black lines show the contrast between Halofit SPJ+03TH13BMSP14 and alternative nonlinear $P(k)$ models: BACCOemu (solid), CosmicEmu (dashed), EuclidEmu2 (dotted-dashed), and HMCode2020 (dotted). The thick black lines show the $P(k)$ ratios in the native $k$ range of the alternative models and the thin black lines show their spline extrapolation. Right panels: $\xi_\pm^{44}(\vartheta)$ difference due to baryonic feedback and nonlinear power spectrum modeling. We show the first three PCs in solid yellow/red/teal lines, which are padded along the $y$-axis for better visualization. Hydrosims are shown with the same colors and line styles as the left panels. We only show one alternative $P(k)$ model, EuclidEmu2, in the left panels.
  • Figure 2: The amplitudes of the first two PCs ($Q_1$ and $Q_2$) of the hydrodynamical simulations used in this work. The smallest data points are ANTILLES simulations that are used to generate the PCs, and are colored based on their mean normalized baryon fraction $\bar{Y}_\mathrm{b}$ of halos within $10^{13}$-$10^{14}$$M_\odot$. Note that $\bar{Y}_\mathrm{b}$ correlates with $Q_1$ tightly. The medium data points are the 11 hydrosims used in XEM+23. The largest data points are the three ANTILLES simulations that are reserved from PC generation, corresponding to the weak (blue cross), mean (green triangle), and strong (red pentagon) feedback scenarios.
  • Figure 3: The scaling relation between $Q_1$ and the mean baryon fraction in halos of mass $M_\mathrm{h}\in[10^{13},\,10^{14}]\,M_\odot$, normalized by the cosmic baryon fraction $\Omega_\mathrm{b}/\Omega_\mathrm{m}$. The gray data points are ANTILLES simulations used to generate PCs, and the colored data points are the weak (blue cross), mean (green triangle), and strong (red pentagon) feedback scenarios. The relation between $Q_1$ and $\bar{Y}_\mathrm{b}$ can be fitted as the black dashed line. We show the $\bar{Y}_b$ observed in AEO+22 in the horizontal solid line and its uncertainty in the shaded gray region. Alternative estimate of $\bar{Y}_b$ based on the latest eROSITA $f_\mathrm{gas}$ measurement PBM+24 is shown in the horizontal purple line. We note that this is not a rigorous estimate since it assumes that the HSC-XXL sample and the eROSITA sample have the same stellar mass fraction, and we only aim to show the impact qualitatively.
  • Figure 4: The architecture of the neural network emulator for a single probe. The input cosmological and nuisance parameters are masked and normalized such that only relevant parameters are passed into the input layer of size $N_\mathrm{in}^\mathrm{reduced}\times256$, which applies a linear affine transform to the input data. Three $256\times256$ residual networks (ResMLP) layers are appended to the input layer, each consisting of two dense layers with a residual connection. Then the $256\times N_\mathrm{out}^\mathrm{reduced}$ output layer is appended to the ResMLP layers to predict a specific probe in $6\times2$pt.
  • Figure 5: The distribution of $\Delta\chi^2$ between the $6\times2$pt model vectors evaluated by CoCoA and the ResMLP128 emulator over the validation dataset, which has $T=64$. Over the 10,000 samples in the validation dataset, only 3.06 percent of the samples give $\Delta\chi^2>0.2$.
  • ...and 12 more figures