Calibrating magnification bias for the $E_G$ statistic to test general relativity

TL;DR

This paper investigates how magnification bias affects the $E_G$ statistic used to test general relativity, proposing a calibration method to mitigate its impact especially in photometric surveys, supported by simulation results.

Contribution

It introduces a calibration technique for magnification bias in the $E_G$ statistic, enhancing its reliability in testing gravity with future surveys.

Findings

Magnification bias has a small effect for spectroscopic surveys.

The proposed calibration method effectively reduces bias within simulation errors.

Calibration accuracy depends on redshift distribution and magnification bias measurement precision.

Abstract

We assess the effect of magnification bias on the $E_G$ statistic for probing gravity. $E_G$, a statistic constructed from power spectrum estimates of both weak lensing and redshift space distortions (RSD), directly tests general relativity (GR) while in principle being independent of clustering bias. This property has motivated its recent use in multiple tests of GR. Recent work has suggested that the magnification bias of galaxies due to foreground matter perturbations breaks the bias-independence of the $E_G$ statistic. The magnitude of this effect is very sensitive to the clustering and magnification biases of the galaxy sample. We show that for realistic values of the clustering and magnification biases, the effect for magnification bias is small relative to statistical errors for most spectroscopic galaxy surveys but large for photometric galaxy surveys. For the cases with significant magnification bias, we propose a method to calibrate magnification bias using measurements of the lensing auto-power spectrum. We test this calibration method using simulations, finding that our calibration method can calibrate $E_G$ from 2-4 times the simulation error to well within the errors, although its accuracy is sensitive to the precision of the measured redshift distribution and magnification bias, but not the clustering bias. This work gives strong evidence that this method will work increasingly well in future CMB lensing surveys.