Robust yet precise: perturbative forward modeling for Large-Scale Structure cosmology at the field-level

• Applicant: Dr. Julia Stadler
• Subject Area: Astrophysics and Astronomy
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 545534254

Over the coming decade, a new generation of spectroscopic galaxy surveys, such as DESI and Euclid, is mapping the cosmological Large Scale Structure (LSS) at unprecedented volume and depth. The data can transform our understanding of the origin and evolution of the Universe, but fully unlocking its potential is challenging. State-of the-art analyses lose information by first compressing the data into summary statistics, foremost the power- and bispectrum, which are then compared to theory predictions. To fully exhaust the constraining power and maximize the chances for discoveries, novel analysis strategies are required. Crucially, these strategies need to address theoretical uncertainties in the galaxy bias relation, which originate from the formation and evolution of galaxies on small scales and can mislead cosmological conclusions. This project will establish a novel analysis strategy for modern galaxy surveys to simultaneously maximize their constraining power and provide robust control over theoretical errors. The crucial aspects of the approach are, first, the comparison between model and observations at the level of the galaxy density field, which avoids information loss in the data compression. This is facilitated by a forward model that predicts the large-scale galaxy density for any given realization of the primordial perturbations and the cosmological parameters. Second, the forward model relies on a perturbative description of galaxy bias (EFTofLSS) as presently the only approach that ensures rigorous control over theoretical uncertainties. Early work by the PI and others has demonstrated the feasibility and potential of forward-modeling; now the PI will lead a team of one postdoc and two PhD students to elevate the technique to the complexity of observational data. They will develop a fast and differentiable forward model that describes all relevant cosmological, observational, and systematic effects. Furthermore, they will enhance the inference algorithm to meet the required efficiency and robustness. All developments are verified against numerical simulations and mock data challenges. The project culminates in the analysis of DESI data, with the goal to derive at least twice as precise parameter constraints, compared to the state-of-the-art, and to provide insights into the initial conditions and the structure formation history. The heighten precision and spatial information can uncover novel and unexpected phenomena; additionally it will enable many interesting follow-up studies and cross-correlations between the matter density with further cosmological and astrophysical probes. The forward model and the systematic mitigation strategies developed will benefit analyses beyond the project scope. As such, the proposed research is an integral contribution to maximizing the scientific return from the next generation of LSS surveys with modern statistical and Machine Learning techniques.

• DFG Programme: Emmy Noether Independent Junior Research Groups