Particle Physics Seminars- SB/BNL Joint Cosmo Seminar

Current (e.g. DES) and future (e.g. LSST, Euclid) experiments aim to convert multiband images of the sky into precise constraints on cosmological models, neutrino masses, and modifications of general relativity. This standard path for this inference involves making point estimates of the galaxies' redshifts (from observed colors) and weak gravitational lensing distortions (from observed morphologies), then combining these into various cross-correlations and other summary statistics that are compared to numerical simulations of the Universe. These estimators require a slew of empirical corrections to various biases, and have yet to demonstrate accuracies sufficient to reduce biases below systematic errors. I describe two steps to greatly simplify this process and eliminate the need for simulation-based calibration of estimators: first, a practical means to estimate the joint posterior probability of a galaxies' redshift and line-of-sight lensing; second, a method to sample from the posterior distribution of all mass distributions and cosmologies conditional on the galaxy density and lensing data. The main advantages of the new scheme include improved lensing and photo-z accuracy (to the required part-per-thousand level), recovery of non-Gaussian information that is lost in the usual 2-point summary statistics, and correct propagation of uncertainties (including photo-z uncertainties) into the cosmological inferences.