SMBp Group Meeting: Bruno Regaldo-Saint Blancard (CCM)

Thursday Jun 29, 2023, 2:00 PM → 3:00 PM America/New_York

3rd Floor Classroom/3-Flatiron Institute (162 5th Avenue)

Speaker: Bruno Regaldo-Saint Blancard (CCM)

 

Topic: Simulation-Based Inference for Cosmology: Galaxy Clustering Analysis with SimBIG and the Wavelet Scattering Transform

Abstract: The dynamical evolution of the Universe is surprisingly well understood under the lens of the standard model of cosmology, or ΛCDM model. The spatial distribution of galaxies constitutes a prime observable to constrain the few parameters of this model. Indeed, galaxies trace the large-scale structure of the Universe, and their non-Gaussian clustering statistics result from the nonlinear evolution of the Universe. We conduct a Bayesian inference of the ΛCDM parameters Ωm, Ωb, h, ns, and σ8 from BOSS CMASS galaxy catalogs by combining the wavelet scattering transform (WST) with a simulation-based inference approach enabled by the SimBIG forward model. We design a set of reduced WST statistics that leverage symmetries of redshift-space data. Posterior distributions are estimated with a conditional normalizing flow trained on 20,000 forward modeled SimBIG simulations. We validate the posterior estimates using simulation-based calibration (SBC) and assess generalization and robustness to the change of forward model using a suite of test simulations. We find that, when probing spatial scales down to k_max = 0.5 h/Mpc our inference pipeline passes the SBC test and demonstrates robustness to the change of forward for all parameters but σ8. We solve robustness issues with σ8 by dropping the WST coefficients that probe scales beyond k ~ 0.3 h/Mpc. When applied to BOSS CMASS data, while using conservative priors, the robust model yields constraints that improve previous results obtained with power spectrum measurements with k_max = 0.5 h/Mpc by tightening 90% credible intervals on Ωm, Ωb, h, ns, and σ8 by factors 1.6, 1.7, 1.3, 1.4, and 1.3, respectively. However, we still raise concerns on these results, since a careful examination of these predictions across different normalizing flow architectures indicates a form of model specification that will be addressed in follow-up works.