Galaxy cluster SZ detection with unbiased noise estimation: an iterative approach

Multi-frequency matched filters (MMFs) are routinely used to detect galaxy clusters from CMB data through the thermal Sunyaev-Zeldovich (tSZ) effect, leading to cluster catalogues that can be used for cosmological inference. In order to be applied, MMFs require knowledge of the cross-frequency power spectra of the noise in the maps. This is typically estimated from the data and taken to be equal to the power spectra of the data, assuming the contribution from the tSZ signal of the detections to be negligible. Using both analytical arguments and Planck-like mock observations, we show that doing so causes the MMF noise to be overestimated, inducing a loss of signal to noise. Furthermore, the MMF cluster observable (the amplitude y(0)($) over cap or the signal to noise q) does not behave as expected, which can potentially bias cosmological inference. In particular, the observable becomes biased with respect to its theoretical prediction and displays a variance that also differs from its predicted value. We propose an iterative MMF (iMMF) approach designed to mitigate these effects. In this approach, after a first standard MMF step, the noise power spectra are reestimated by masking the detections from the data, delivering an updated iterative cluster catalogue. Applying our iMMF to our Planck-like mock observations, we find that the aforementioned effects are completely suppressed. This leads to a signal-to-noise gain relative to the standard MMF, with more significant detections and a higher number of them, and to a cluster observable with the expected theoretical properties, thus eliminating any potential biases in the cosmological constraints.

Automated summary

Multi-frequency matched filters (MMFs) are commonly used to detect galaxy clusters from CMB data, but estimating the cross-frequency power spectra of the noise incorrectly leads to overestimated MMF noise and biases in cosmological inference. We propose an iterative MMF (iMMF) approach that improves the estimates by reestimating the noise power spectra after the first MMF step. Applying the iMMF to Planck-like mock observations suppresses the mentioned effects and leads to more significant detections, a higher number of them, and a cluster observable with the expected theoretical properties.