On the Probability Distributions of Ellipticity

TL;DR

This paper derives an exact 2D probability distribution for ellipticity measurements under Gaussian noise, highlighting biases in estimators and implications for weak lensing survey calibration.

Contribution

It generalizes the known ratio distribution to the multivariate case for ellipticity, providing a new analytical tool for weak lensing data analysis.

Findings

Bias in maximum likelihood ellipticity estimates identified.

Calibration strategies for upcoming surveys quantified.

Stokes parameters offer unbiased shear estimates with high variance.

Abstract

In this paper we derive an exact full expression for the 2D probability distribution of the ellipticity of an object measured from data, only assuming Gaussian noise in pixel values. This is a generalisation of the probability distribution for the ratio of single random variables, that is well-known, to the multivariate case. This expression is derived within the context of the measurement of weak gravitational lensing from noisy galaxy images. We find that the third flattening, or epsilon-ellipticity, has a biased maximum likelihood but an unbiased mean; and that the third eccentricity, or normalised polarisation chi, has both a biased maximum likelihood and a biased mean. The very fact that the bias in the ellipticity is itself a function of the ellipticity requires an accurate knowledge of the intrinsic ellipticity distribution of the galaxies in order to properly calibrate shear measurements. We use this expression to explore strategies for calibration of biases caused by measurement processes in weak gravitational lensing. We find that upcoming weak lensing surveys like KiDS or DES require calibration fields of order of several square degrees and 1.2 magnitude deeper than the wide survey in order to correct for the noise bias. Future surveys like Euclid will require calibration fields of order 40 square degree and several magnitude deeper than the wide survey. We also investigate the use of the Stokes parameters to estimate the shear as an alternative to the ellipticity. We find that they can provide unbiased shear estimates at the cost of a very large variance in the measurement. The python code used to compute the distributions presented in the paper and to perform the numerical calculations are available on request.