Denoising Diffusion Delensing Delight: Reconstructing the Non-Gaussian CMB Lensing Potential with Diffusion Fashions
Authors: Thomas Flöss, William R. Coulton, Adriaan J. Duivenvoorden, Francisco Villaescusa-Navarro, Benjamin D. Wandelt
Summary: Optimum extraction of cosmological info from observations of the Cosmic Microwave Background critically depends on our means to precisely undo the distortions attributable to weak gravitational lensing. On this work, we reveal using denoising diffusion fashions in performing Bayesian lensing reconstruction. We present that score-based generative fashions can produce correct, uncorrelated samples from the CMB lensing convergence map posterior, given noisy CMB observations. To validate our strategy, we evaluate the samples of our mannequin to these obtained utilizing established Hamiltonian Monte Carlo strategies, which assume a Gaussian lensing potential. We then transcend this assumption of Gaussianity, and prepare and validate our mannequin on non-Gaussian lensing knowledge, obtained by ray-tracing N-body simulations. We reveal that on this case, samples from our mannequin have correct non-Gaussian statistics past the ability spectrum. The tactic supplies an avenue in the direction of extra environment friendly and correct lensing reconstruction, that doesn’t depend on an approximate analytic description of the posterior likelihood. The reconstructed lensing maps can be utilized as an unbiased tracer of the matter distribution, and to enhance delensing of the CMB, leading to extra exact cosmological parameter inference.
