Characterisation of the non-Gaussianity of radio and IR point-sources at CMB frequencies

TL;DR

This paper characterizes the non-Gaussianity of radio and IR point sources in CMB frequencies, proposing a method to predict the bispectrum and analyzing their impact on primordial non-Gaussianity measurements.

Contribution

It introduces a simple prescription to infer the angular bispectrum from the power spectrum of point sources and assesses their impact on f_NL bias in CMB data.

Findings

IR bispectrum peaks for squeezed triangles

Clustering enhances IR bispectrum significantly at certain scales

IR sources can mimic primordial local bispectrum, biasing f_NL measurements

Abstract

This study, using publicly available simulations, focuses on the characterisation of the non-Gaussianity produced by radio point sources and by infrared (IR) sources in the frequency range of the cosmic microwave background from 30 to 350 GHz. We propose a simple prescription to infer the angular bispectrum from the power spectrum of point sources considering independent populations of sources, with or without clustering. We test the accuracy of our prediction using publicly available all-sky simulations of radio and IR sources and find very good agreement. We further characterise the configuration dependence and the frequency behaviour of the IR and radio bispectra. We show that the IR angular bispectrum peaks for squeezed triangles and that the clustering of IR sources enhances the bispectrum values by several orders of magnitude at scales l \sim 100. At 150 GHz the bispectrum of IR sources starts to dominate that of radio sources on large angular scales, and it dominates over the whole multipole range at 350 GHz. Finally, we compute the bias on f_NL induced by radio and IR sources. We show that the positive bias induced by radio sources is significantly reduced by masking the sources. We also show, for the first time, that the form of the IR bispectrum mimics a primordial 'local' bispectrum f_NL. The IR sources produce a negative bias which becomes important for Planck-like resolution and at high frequencies (Delta f_NL ~ -6 at 277 GHz and Delta f_NL \sim -60-70 at 350 GHz). Most of the signal being due to the clustering of faint IR sources, the bias Delta f_NL^IR is not reduced by masking sources above a flux limit and may, in some cases, even be increased due to the reduction of the shot-noise term.