Inferring the Impacts of Baryonic Feedback from Kinetic Sunyaev-Zeldovich Cross-Correlations

TL;DR

This work addresses how baryonic feedback alters small-scale matter clustering and its potential to bias cosmological inferences such as $S_8$. It introduces a fully data-driven power-spectrum emulator that maps the shape of $R(k)=P(k)/P_{\mathrm{DMO}}(k)$ to the observed kSZ cross-spectrum and galaxy clustering, using templates derived from the FLAMINGO hydrodynamical simulations. The approach combines PCA/LDA feature extraction on $C_\ell^{gg}$ and $C_\ell^{\hat{b}b}$ with a multi-layer perceptron to predict $R(k)$ from nine inputs, achieving sub-percent accuracy for $k\le 5\,h/\mathrm{Mpc}$ and $0.2\le z\le1.25$, and validating robustness to unseen feedback and cosmology. Forecasts for a Simons Observatory–like CMB dataset plus DESI-like galaxies suggest a $\sim 50\sigma$ detection of the kSZ signal and enable rapid, joint inference of baryonic suppression, offering a path to break degeneracies with new physics and improve small-scale cosmological constraints.

Abstract

The complex processes of baryonic feedback associated with galaxy evolution are still poorly understood, and their impact on the clustering of matter on small scales remains difficult to quantify. While many fitting functions and emulators exist to model the matter power spectrum, their input parameters are not directly observable. However, recent studies using hydrodynamical simulations have identified a promising correlation between the gas content of halos and changes to the matter power spectrum from feedback. Building on these findings, we create the first fully data-driven power spectrum emulator. We utilize the kinematic Sunyaev-Zeldovich (kSZ) effect, a secondary anisotropy in the cosmic microwave background, as a tracer of free electrons in and around halos. We train a neural network to learn the mapping between the suppression of the matter power spectrum and the shape of the kSZ power spectrum extracted with a radial velocity template. We train and validate our algorithm using the FLAMINGO suite of hydrodynamical simulations, which encompasses a wide range of feedback models. Our emulator can reconstruct the matter power spectrum at the sub-percent level for scales $k\leq 5\;h/$Mpc and $0.2\leq z \leq 1.25$ directly from the data. Our model is robust and retains percent-level accuracy even for feedback models and cosmological parameter values not seen during training (except in a few extreme cases drastically different from the fiducial model). Due to its robustness, our algorithm offers a new way to identify the sources of suppression in the matter power spectrum, breaking the degeneracies between baryonic feedback and new physics. Finally, we present a forecast for reconstruction of the matter power spectrum combining maps of the microwave background anisotropies from a Simons Observatory-like experiment and galaxy catalogs from the Dark Energy Spectroscopic Instrument.

Figures (10)

• Figure 1: Example of maps used with the template method. All the maps are based on the simulation L1_ m9 at redshift $z=0.6$. By zooming-in, we can observe the correlated features of the Doppler $b$ map and the galaxy template. (Upper left) Galaxy radial velocity. (Upper right) Galaxy density. (Center left) Doppler $b$ template. (Center right) True Doppler $b$. (Lower left) 20 by 20 degree region of template map. (Lower right) 20 by 20 degree region of true Doppler $b$ map. The $v_r, \;\delta_g$, and $\delta_g v_r$ maps have been smoothed with a Gaussian with a FWHM of 5 arcminutes for ease of visualization.
• Figure 2: Variation in the matter power spectrum suppression at $k=1\;h$/Mpc as a function of the kSZ cross-correlation at $\ell=5000$ and redshift $z=0.3$. The template is formed of galaxies with stellar masses above $10^{11.3}\;\mathrm{M}_\odot$. The dashed grey line is the least-squares linear fit using the log-polynomial model.
• Figure 3: (Top) Doppler $b$ cross-correlation spectra from 100 HODs at redshift $z=0.6$ for the fiducial feedback model (L1$\_$m9) along with the log-polynomial fits. (Center) Mean Doppler $b$ template cross-correlation power spectra. The weaker AGN feedback denoted fgas$+2\sigma$ is less suppressed than the fiducial model and the more pronounced feedback mechanisms (L1_ m9, fgas$-4\sigma$ Jets+fgas$-4\sigma$). The dashed lines represent the mean of log-polynomial fits to the cross-spectra. (Bottom) Power spectrum ratio at $z=6$ for the four models shown above. The dashed lines denote the mean of the $R(k)$ predictions from our emulator. In the two bottom plots, the wavenumbers and angular multipoles are related through the Limber approximation (see Eq. \ref{['eq:limber']}). All panels are at redshift $z=0.6$.
• Figure 4: Workflow (from left to right) of our MLP emulator taking as input the galaxy auto-spectrum, the kSZ cross-spectrum, and the redshift. The full-size hidden layers with 1024 neurons are partially hidden for clarity. The numbers indicate the shapes of the arrays after each operation.
• Figure 5: Test-set residuals from the neural network prediction. The test amounts for 5% of the total dataset and includes samples from all redshift slices and feedback models. The solid line represents the median error over the validation samples while the shaded area denotes the 68% confidence interval. The error is calculated as $R_\mathrm{model}(k)/R_\mathrm{true}(k)-1$.