AMICO galaxy clusters in KiDS-DR3: the impact of estimator statistics on the luminosity-mass scaling relation

TL;DR

This study compares $L^1$ and $L^2$ regression methods for weak lensing shear analysis around galaxy clusters, highlighting the importance of estimator choice on mass-luminosity scaling relations in precision cosmology.

Contribution

It introduces a comparison between $L^1$ and $L^2$ estimators for shear inference, demonstrating the impact on derived cluster masses and scaling relations in KiDS-DR3 data.

Findings

$L^1$ estimator shows comparable efficiency to weighted mean, especially at higher S/N.

A small bias (~3%) exists between the two methods in surface density profiles.

The shear signal yields a power-law slope of ~1.24 in the luminosity-mass relation.

Abstract

As modern-day precision cosmology aims for statistical uncertainties of the percent level or lower, it becomes increasingly important to reconsider estimator assumptions at each step of the process, and their consequences on the statistical variability of the scientific results. We compare $L^1$ regression statistics to the weighted mean, the canonical $L^2$ method based on Gaussian assumptions, for inference of the weak gravitational shear signal from a catalog of background ellipticity measurements around a sample of clusters, in many recent analyses a standard step in the process. We use the shape measurements of background sources around 6925 AMICO clusters detected in the KiDS 3rd data release. We investigate the robustness of our results and the dependence of uncertainties on the signal-to-noise ratios of the background source detections. Using a halo model approach, we derive lensing masses from the estimated excess surface density profiles. The highly significant shear signal allows us to study the scaling relation between the $r$-band cluster luminosity $L_{200}$, and the derived lensing mass $M_{200}$. We show the results of the scaling relations derived in 13 bins in $L_{200}$, with a tightly constrained power law slope of $\sim 1.24\pm 0.08$. We observe a small, but significant relative bias of a few percent in the recovered excess surface density profiles between the two regression methods, which translates to a $1\sigma$ difference in $M_{200}$. The efficiency of $L^1$ is at least that of the weighted mean, relatively increasing with higher signal-to-noise shape measurements. Our results indicate the relevance of optimizing the estimator for infering the gravitational shear from a distribution of background ellipticities. The interpretation of measured relative biases can be gauged by deeper observations, while increased computation times remain feasible.