Map-level baryonification: unified treatment of weak lensing two-point and higher-order statistics

TL;DR

Baryonic feedback biases weak-lensing statistics on small scales, complicating cosmological inference. The authors implement map-level baryonification via the BCM to post-process dark-matter-only simulations, enabling joint modeling of two-point and higher-order statistics in DES-Y3-like weak-lensing data using FLAMINGO calibration. They develop an emulator to rapidly predict BCM-induced changes across a 10-parameter space and perform MCMC inference to fit multiple hydrodynamic variants, achieving ~2% accuracy for statistics up to ell ~ 2000. The results show that both full and reduced-parameter BCM variants can robustly reproduce non-Gaussian lensing statistics, supporting simulation-based inference that marginalizes baryonic uncertainties and paves the way for tighter cosmological constraints from upcoming surveys.

Abstract

Precision cosmology benefits from extracting maximal information from cosmic structures, motivating the use of higher-order statistics (HOS) at small spatial scales. However, predicting how baryonic processes modify matter statistics at these scales has been challenging. The baryonic correction model (BCM) addresses this by modifying dark-matter-only simulations to mimic baryonic effects, providing a flexible, simulation-based framework for predicting both two-point and HOS. We show that a 3-parameter version of the BCM can jointly fit weak lensing maps' two-point statistics, wavelet phase harmonics coefficients, scattering coefficients, and the third and fourth moments to within 2% accuracy across all scales $\ell < 2000$ and tomographic bins for a DES-Y3-like redshift distribution ($z \lesssim 2$), using the FLAMINGO simulations. These results demonstrate the viability of BCM-assisted, simulation-based weak lensing inference of two-point and HOS, paving the way for robust cosmological constraints that fully exploit non-Gaussian information on small spatial scales.

Figures (11)

• Figure 1: Right: redshift distribution of the four tomographic bins of the DES Y3 wak lensing sample considered here. Left: Power spectra of the 4 noiseless, full-sky convergence maps produced from the FLAMINGO DMO simulation, using the DES Y3 redshift distributions, compared to predictions from EuclidEmu Knabenhans2019.
• Figure 2: Summary statistics of the FLAMINGO $\kappa$ maps (Sec. \ref{['sec:sim']}). Rows correspond to different statistics, and columns to redshift bins (modeled after DES Y3 data). The y-axis shows the FLAMINGO statistics relative to the statistics of the DMO simulation of the same initial condition. On the x-axis, spatial scales decreases to the right. ST and WPH are ordered by the wavelet size ($j$) as a proxy for their characteristic scale. In FLAMINGO, baryon feedback physics suppresses all statistics on all scales, and is the strongest in low redshift bins.
• Figure 3: Summary statistics of the baryonified (colored) and the FLAMINGO fiducial variant's (black, dashed) convergence map, focusing on the first redshift bin. For the baryonified maps, a single BCM parameter is varied at a time. Rows correspond to different BCM parameters, and columns to different statistics. The y-axis shows the statistics of both the baryonified and FLAMINGO maps, relative to those of the DMO simulation with the same initial condition. On the x-axis, spatial scale decreases to the right. ST and WPH statistics are ordered by wavelet size ($j$), used here as a proxy for characteristic scale. The BCM flexibly generates a range of baryonic signatures in the statistics, clustering around the FLAMINGO results. The variation of other parameters in Tab. \ref{['tab:params_descrp']} is shown in Fig. \ref{['fig:variation_7param']} in App. \ref{['appx:sensitivity']}.
• Figure 4: The emulator mean relative error, grouped by statistics and by redshift bins. The mean absolute (relative) error (MAE) is shown in text. The error is evaluated by averaging across the test set across the baryon parameter space. The emulator achieves sub-percent accuracy for all statistics besides the low redshift moments. The higher-order moment statistics are intrinsically nosier since it involves multiplications of random variables. We remind the reader that the emulator is used to efficiently explore the BCM parameter space when comparing to FLAMINGO statistics. However, the final assessment of how well the BCM reproduces a given FLAMINGO statistic is based on running the full baryonification pipeline at the emulator-derived best-fit parameters.
• Figure 5: The residuals error of the BCM when fitted to the FLAMINGO fiducial variant. Rows correspond to different statistics, and columns to redshift bins. The y-axis shows the ratio between the statistics of the best-fit baryonified map to that of the fiducial variant. On the x-axis, spatial scales decreases to the right. ST and WPH are ordered by the wavelet size ($j$) as a proxy for their characteristic scale. Colors indicate the number of free parameters in the BCM fit. The shaded region marks $\pm$2% for visual reference. For all n-parameter models, the BCM flexibly reproduces the baryonic features in the FLAMINGO fiducial variant simulation to percent level. The prediction is robust even with only 3 degrees of freedoms. Differences in accuracy between the models stem from how we estimate the best-fit parameter, as discussed in Sec. \ref{['sec:reduced_param']}. The residuals for the strongest (fgas$-8\sigma$) and weakest (fgas$+2\sigma$) feedback variants are shown in Fig. \ref{['fig:feedback+']} and Fig. \ref{['fig:feedback-']} in App. \ref{['appx:residuals']}.