Tracing Cosmic Origins: Inferring Initial Conditions from Galaxy Surveys
Jens Jasche
Imminently arriving galaxy surveys offer unprecedented opportunities to test cosmology and fundamental physics through the rich phenomenology predicted for cosmic structure. However, traditional data analysis methods often rely on limited statistical summaries, overlooking crucial information embedded in the complex, three-dimensional distribution of matter. To address these limitations, physics-informed field-level inference is emerging as a powerful alternative approach. This method connects the early and late-time universe by jointly inferring cosmic initial conditions and mapping nonlinear density and velocity fields through a generative cosmic structure model. By employing dynamic, nonlinear models of structure formation during inference, we achieve the most comprehensive characterisation of cosmic structure, its origins, and its dynamic evolution. In addition to promising improved cosmological parameter constraints, this approach opens new avenues for studying cosmology and fundamental physics using large-scale cosmic structures. In this presentation, I will briefly introduce the concepts of field-level inference and Bayesian physical forward modelling, illustrating their promise for cosmological parameter inference and tests of primordial non-Gaussianity. I will showcase the approach across a range of data and illustrate how these techniques can be used to test fundamental physics, such as probing the particle nature of dark matter or predicting the anisotropy in the neutrino sky. Finally, I will demonstrate that the Bayesian field-level inference approach allows us to recover the initial conditions over a vast dynamic range of cosmic scales, culminating in a detailed reconstruction of the initial conditions for the Local Group. This reconstruction provides a plausible and causal formation history of the Milky Way-Andromeda galaxy pair, accurately recovering their properties, including masses, rotation curves, and relative velocities.
Event Timeslots (1)
Mon
-