Speaker
Description
Two-point correlation functions are a standard tool in cosmology. However, their estimators have intrinsically non-Gaussian likelihoods, even for perfectly Gaussian random fields. This non-Gaussianity is most evident on the large scales targeted by stage-IV weak lensing surveys. We present a framework for computing exact correlation-function likelihoods for the masked spin-2 fields of cosmic shear. These likelihood distributions show significant skewness and can systematically shift parameter inferences. We provide exact solutions for the one-dimensional marginals and introduce a Gaussian copula model that efficiently captures the dependencies in high-dimensional correlation-function data and which can straightforwardly be used in Bayesian analyses. Copula-based approaches open a path to flexible, accurate likelihood modeling not only for weak lensing but for a wider range of cosmological probes. More generally, our results illustrate how both correlation structure and likelihood shape impact posterior constraints, highlighting the need for careful modeling of non-Gaussian likelihoods.