High Dimensional Beam Inference II: Inference of a Perturbed HERA Beam from Simulated Visibility Data

Abstract Accurate beam modeling is important in many radio astronomy applications. In this paper, we focus on beam modeling for 21-cm intensity mapping experiments using radio interferometers, though the techniques also apply to single dish experiments with small modifications. In 21-cm intensity mapping, beam models are usually determined from highly detailed electromagnetic simulations of the receiver system. However, these simulations are expensive, and therefore have limited ability to describe practical imperfections in the beam pattern. We present a fully analytic Bayesian inference framework to infer a beam pattern from the interferometric visibilities assuming a particular sky model and that the beam pattern for all elements is identical, allowing one to capture deviations from the ideal beam for relatively low computational cost. We represent the beam using a sparse Fourier-Bessel basis on a projection of the hemisphere to the unit disc, but the framework applies to any linear basis expansion of the primary beam. We test the framework on simulated visibilities from an unpolarized sky, ignoring mutual coupling of array elements. We successfully recover the simulated, perturbed power beam when the sky model is perfect. Briefly exploring sky model inaccuracies, we find that beam inferences are sensitive to them, so we suggest jointly modeling uncertainties in the sky and beam in related inference tasks.