Title:Detection of the significant impact of source clustering on higher-order statistics with DES Year 3 weak gravitational lensing data

Abstract:We demonstrate and measure the impact of source galaxy clustering on higher-order summary statistics of weak gravitational lensing data. By comparing simulated data with galaxies that either trace or do not trace the underlying density field, we show this effect can exceed measurement uncertainties for common higher-order statistics for certain analysis choices. Source clustering effects are larger at small scales and for statistics applied to combinations of low and high redshift samples, and diminish at high redshift. We evaluate the impact on different weak lensing observables, finding that third moments and wavelet phase harmonics are more affected than peak count statistics. Using Dark Energy Survey Year 3 data we construct null tests for the source-clustering-free case, finding a $p$-value of $p=4\times10^{-3}$ (2.6 $\sigma$) using third-order map moments and $p=3\times10^{-11}$ (6.5 $\sigma$) using wavelet phase harmonics. The impact of source clustering on cosmological inference can be either be included in the model or minimized through \textit{ad-hoc} procedures (e.g. scale cuts). We verify that the procedures adopted in existing DES Y3 cosmological analyses (using map moments and peaks) were sufficient to render this effect negligible. Failing to account for source clustering can significantly impact cosmological inference from higher-order gravitational lensing statistics, e.g. higher-order N-point functions, wavelet-moment observables (including phase harmonics and scattering transforms), and deep learning or field level summary statistics of weak lensing maps. We provide recipes both to minimise the impact of source clustering and to incorporate source clustering effects into forward-modelled mock data.