Speaker
Description
In this talk I will present an overview of current cosmological emulator development efforts in Euclid, aimed at enabling efficient and accurate parameter inference in the era of high-precision cosmology. Cosmological emulators are surrogate models trained on high-fidelity theoretical predictions from Boltzmann solvers, or from numerical simulations. Once trained, these models can reproduce complex observables with percent-level accuracy while reducing the computational cost of evaluation by several orders of magnitude. A single emulator prediction can be produced in milliseconds rather than minutes, making these tools essential for modern Bayesian inference methods, which require millions of model evaluations across high-dimensional cosmological parameter spaces.
I will focus in particular on CosmoPower emulators designed to predict both large-scale structure and cosmic microwave background observables in multi-probe cosmological analyses. For large-scale structure, this includes emulators of the linear and nonlinear matter power spectrum used in combined analyses of cosmic shear, galaxy clustering, and galaxy–galaxy lensing. At the linear level, accurate emulation requires reproducing the transfer function and growth history across a wide cosmological parameter space, while at nonlinear scales additional complexity arises from gravitational collapse and baryonic feedback processes.
I will also present recent work on CMB power spectrum emulators, which provide fast predictions for temperature and polarization spectra and enable efficient joint analyses combining CMB and large-scale structure probes. Finally, I will discuss ongoing efforts to extend emulator coverage beyond the standard ΛCDM models, with emphasis on training strategies, validation, and the propagation of emulator uncertainties into cosmological constraints.