AR-2019-2020

Figure 12: Demonstration of regularised deconvolution for a fiducial injected map (a) The raw ‘dirty’ map (b) is convolved with the antenna pattern functions of the detectors and needs to be deconvolved to estimate the true sky. Maximum likelihood estimation with the standard conjugate gradient method, that performs well in these situations, yields a noisy result. The Bayesian regularisation method, with a generic prior to prefer a smoother, less fluctuating, solution in this case, produces a much better match between the recovered and the injected map. The dirty map has poorer match with the source map for weak sources, especially for point like ones, which is why it is not optimal to use it directly without deconvolution. [Phys. Rev. D 100, 043541, arXiv:1905.08276 [gr-qc].]