Ponente
Descripción
Galaxy Surveys provide an important cosmological probe to study dark energy and structure formation in the Universe. The cosmological content is however encoded in biased tracers of the non-linear evolved cosmic density field. The analysis of such data is complex and requires non-Gaussian posterior distribution functions (PDFs). Sampling such PDFs can be done within a Bayesian framework at the expense of heavy computations using, for instance, a Hamiltonian Monte Carlo Sampling algorithm. We suggest here to dramatically speed up these schemes by implementing a higher order Hamiltonian Monte Carlo scheme, demonstrating, with a lognormal-Poisson model, that the correlation length between samples and the computational cost can be reduced by an order of magnitude. This scheme has been implemented in a Bayesian framework, including structure formation with Lagrangian perturbation theory, yielding accurate reconstructions of the primordial density field associated to a distribution of galaxies, as obtained from galaxy surveys.
