Going beyond $S_8$: fast inference of the matter power spectrum from weak-lensing surveys

TL;DR

This paper introduces a fast, flexible framework to extract the scale-dependent matter power spectrum from weak lensing data, revealing potential tensions with Planck ΛCDM predictions and enabling model-agnostic tests of structure growth.

Contribution

The authors develop a novel method to infer the matter power spectrum from weak lensing data, allowing for deviations from ΛCDM and applying it to multiple datasets with public code and data.

Findings

15-30% suppression in galaxy-lensing matter power spectrum at intermediate scales

Combined dataset tensions reach up to 4 sigma with Planck ΛCDM

CMB lensing data consistent with ΛCDM at large scales

Abstract

Weak lensing surveys are often summarized by constraints on the derived parameter ${S_8\equiv\sigma_8\sqrt{\Omega_{\rm m}/0.3}}$, obscuring the rich scale and redshift information encoded in the data, and limiting our ability to identify the origin of any tensions with $\Lambda$CDM predictions from the cosmic microwave background. In this work, we introduce a fast and flexible framework to extract the scale-dependent matter power spectrum $P(k, z)$ from cosmic shear and CMB lensing measurements, parameterizing deviations from the Planck $\Lambda$CDM prediction as a free function $\alpha(k)$. Using public data from DES Y3, KiDS-1000, HSC Y3, and ACT DR6, we constrain $\alpha(k)$ with fast Hamiltonian Monte Carlo inference, employing multipoles up to $\ell_{\rm max}\sim2000$ for the galaxy lensing surveys. Our results show a consistent 15-30% suppression in the matter power spectrum at intermediate scales ($k \sim 0.1-1{\rm Mpc}^{-1}$) in galaxy-lensing data relative to a Planck $\Lambda$CDM prediction with a CDM-only (no baryonic feedback) power spectrum, with combined tensions reaching up to $4\sigma$. This is under a fixed cosmology and with analytic marginalization over shear and redshift calibration uncertainties. In contrast, ACT CMB lensing is consistent with $\Lambda$CDM at ${k\lesssim 0.1 {\rm Mpc}^{-1}}$. We validate our method using mock data, quantify consistency between datasets, and demonstrate how the resulting $\alpha(k)$ likelihoods can be used to test specific models for the power spectrum. All code, data products, and derived likelihoods are publicly released. Our results highlight the importance of reporting lensing constraints on $P(k, z)$ and pave the way for model-agnostic test of growth of structure with upcoming surveys such as LSST, Euclid, and Roman.