Likelihood-free inference with the Dark Energy Survey

Likelihood-free inference with the Dark Energy Survey

PI: Niall Jeffrey

Cosmological simulations run on the UK’s DiRAC high-performance computing system have enabled a step change in how the Dark Energy Survey extracts cosmology from weak gravitational lensing. The Gower Street simulation suite, generated on DiRAC, provides the large, high-fidelity mock universes needed for modern simulation-based inference, allowing the Dark Energy Survey to compare observed lensing maps directly to forward-modelled predictions rather than relying on approximate analytic methods. This approach has delivered the strongest weak-lensing-only constraints on dark energy from DES to date, significantly improving the precision of wCDM cosmological parameters and demonstrating that map-level inference can unlock more information than traditional techniques. These results establish DiRAC-enabled simulation-based inference as a key route to fully exploiting next-generation cosmological surveys. With further DiRAC projects we are now extending the available models of dark energy, to enable new analyses with DES and the new European Space Agency’s Euclid mission.