Emulation for lensing and clustering observables of the cosmological large-scale structure

Recent developments in observational technologies open exciting opportunities to map out the detailed structure of the universe. Remarkably, the unique combination of imaging and spectroscopic galaxy surveys is now becoming well established as a standard analysis methodology for precision cosmology. While the former can access directly the underlying clustering of mass dominated by dark matter projected on the sky through the weak gravitational lensing effect, the latter provides us with the three dimensional map of the structure traced by galaxies. One can mitigate the galaxy-bias uncertainty, which has been the major obstacle for cosmology based on galaxy surveys, by jointly analyzing these effects. We still need, however, a robust and versatile theoretical and statistical framework to interpret these datasets. The Dark Quest project, launched in 2015, is a structure formation simulation campaign precisely
for this purpose. We have developed an emulation tool, dubbed as Dark Emulator, based on a large database of simulated dark matter halos in virtual universes with different cosmologies efficiently sampled in six-dimensional parameter space. Dark Emulator employs a simple machine-learning architecture with Gaussian process at its core. It makes predictions of various statistical measures of dark matter halos, both lensing and clustering observables, for a given cosmological parameters in a few seconds on laptop computers without running a new simulation. This AI-aided tool, once supplemented with recipes for the halo-galaxy connection, is therefore applicable to real-data analyses as the theoretical template, which typically requires hundreds of thousands of function calls in the course of parameter inference. I will introduce this project and report the status of its application to Subaru HSC data.
We are looking forward to seeing you online.