KIPAC Tea Talk: Mujica: Reconstruct Cosmological Initial Conditions from Incomplete Galaxy Surveys with Statistically Regularized Optimization / Intrinsic alignments in the FLAMINGO simulations and consequences for weak lensing

Dong Abstract: Reconstructing cosmological initial conditions from incomplete galaxy redshift surveys is a high-dimensional inverse problem in which weakly constrained regions can make likelihood optimization unstable. We introduce Mujica, an optimization-based reconstruction method that integrates modified L-BFGS updates, rank-order matching, and explicit control of large-scale mode amplitudes. We evaluate Mujica against a baseline maximum-likelihood estimator (MLE) and a standard maximum a posteriori (MAP) approach using mock lightcones derived from the Millennium Simulation and associated semi-analytic galaxy catalogs, spanning a range of survey geometries and depths. Across the configurations considered, Mujica substantially reduces the low-fidelity failure modes seen in the baseline methods while retaining strong agreement with the underlying matter field and a snapshot-based estimate of the initial conditions within the observed volume. We additionally present ablation experiments that isolate the roles of individual algorithmic components and discuss the implications of applying this class of reconstruction methods to incomplete galaxy surveys.

Herle Abstract: Intrinsic alignments in the FLAMINGO simulations and consequences for weak lensing Weak gravitational lensing has emerged as one of the most powerful probes of cosmology, uniquely tracing the total matter distribution and directly probing the growth of large-scale structure. As a late-time observable, it provides key sensitivity to dark energy that complements early-Universe measurements of the Cosmic Microwave Background. Forthcoming surveys such as Euclid, Roman and the Legacy Survey of Space and Time will deliver unprecedented weak lensing measurements, but fully exploiting their statistical power requires sub-percent control of astrophysical systematics. A dominant contaminant is the intrinsic alignment (IA) of galaxies, which correlates galaxy shapes independent of lensing and can bias cosmological constraints if not accurately modelled. In this talk, I present new constraints on IA from the FLAMINGO simulations, analysing millions of Luminous Red Galaxy analogues. By jointly modelling galaxy clustering and alignments, we obtain some of the most precise constraints from hydrodynamic simulations to date. While commonly used IA models like NLA and TATT describe the data well, we introduce a new mass-dependent extension, TATT-M, that provides a significantly improved and more efficient description. This model has been adopted as the fiducial IA model for the Euclid DR1 weak lensing analysis. We then investigate the physical drivers of IA. We find that baryonic feedback does not change the alignment signal beyond its effect on stellar mass, whereas halo assembly history plays a key role: galaxies in earlier-forming haloes exhibit systematically stronger alignments, providing the first clear evidence of assembly bias in IA. We also explore the redshift evolution of the alignment signal, and show that it is more complex than assumed in current models. Finally, I introduce a method to robustly infer halo shapes from sparse satellite populations by correcting for sampling noise, enabling less biased measurements and extending such studies to lower-mass systems. Together, these results establish a simulation-driven framework for intrinsic alignment modelling, directly informing precision weak lensing analyses in the era of next generation weak lensing surveys.