Moreover, thanks to simulation reuse enabled by SBI, they could perform the combined analysis using *zero* calls to a Boltzmann solver

To illustrate their method, they combine the full Planck CMB likelihoods with a 3x2pt simulator for a Stage IV galaxy survey (e.g. Euclid), and test evolving dark energy. They find that future 3x2pt data *alone* could detect dynamical dark energy preferred by DESI at the *5sigma* level

This is particularly relevant for legacy cosmology likelihoods such as Planck, which can't easily be reformulated as forward simulators, complicating its integration into simulation-based combined analyses

Applying it to the case of dark matter (DM) spikes around massive black holes (BHs), they show that a fully relativistic treatment is needed to accurately model the environmental dephasing of gravitational wave (GW) signals from EMRIs

He will use it to conduct six months of research at DESY, the German laboratory for high-energy physics in Berlin.

Congrats to all the authors!
Oleg Savchenko, Florian List, Guillermo Franco Abellán, Noemi Anau Montel, Christoph Weniger

Owing to its simplicity, the method is significantly faster than other proposed algorithms (training only takes ~1h on a GPU, and sampling just ~3s for 1000 samples), while still being highly precise and interpretable

The method relies on the assumption that the multi-million dimensional posterior of cosmic ICs can be described by a simple Gaussian, where the mean and covariance have been trained on the Quijote N-body simulations suite

As an application to non-gaussian posteriors, the authors also test a model of 2-body decaying dark matter, finding that Stage IV surveys can improve current bounds by up to 1 order of magnitude!

They are able to recover the posterior distribution of all parameters with a speedup factor of ~40-60 compared to MCMC!