Simulations and modeling for precision cosmology

par Zarija Lukic

vendredi 30 janv. 2026, 11:00 → 12:00 Europe/Paris

Auditorium Vivargent (LAPTh)

Ongoing and upcoming cosmology surveys aim to address some of the deepest questions in fundamental physics—from the nature of dark energy and dark matter to tests of gravity and particle physics on the largest observable scales. Interpreting these data, which probe multiple cosmic epochs and physical scales, is a formidable challenge: it is fundamentally an inverse problem in which we infer the underlying physics from imperfect, biased, and noisy observations. In this talk I will describe our simulation and modeling program at Berkeley Lab's Computational Cosmology Center to support inference using Lyman-α forest and intergalactic medium measurements. We have developed Nyx, a highly efficient and scalable cosmological hydrodynamics code, to generate high-fidelity realizations needed to connect theory to data, such as DESI observations. However, producing suites of full-physics simulations at the required accuracy remains computationally expensive, motivating complementary approaches. I will discuss our work on emulators—surrogate models trained on finite ensembles of simulations—as well as new hybrid strategies that combine physical modeling with generative ML/AI approaches.