Fast optimal CMB power spectrum estimation with Hamiltonian sampling

Abstract

We present a method for fast optimal estimation of the temperature angular power spectrum from observations of the cosmic microwave background. We employ a Hamiltonian Monte Carlo (HMC) sampler to obtain samples from the posterior probability distribution of all the power spectrum coefficients given a set of observations. We compare the properties of the HMC and the related Gibbs sampling approach on low-resolution simulations and find that the HMC method performs favourably even in the regime of relatively low signal-to-noise. We also demonstrate the method on high-resolution data by applying it to simulated WMAP data. Analysis of a WMAP-sized data set is possible in a around eighty hours on a high-end desktop computer. HMC imposes few conditions on the distribution to be sampled and provides us with an extremely flexible approach upon which to build.