An accurate and practical method for inference of weak gravitational lensing from galaxy images

TL;DR

This paper presents a highly accurate Bayesian Fourier Domain method for weak gravitational lensing inference from galaxy images, effectively correcting biases and achieving unprecedented precision in shear measurement.

Contribution

The paper extends the BFD formalism to correct for selection biases and demonstrates its high accuracy and efficiency on simulated data, outperforming previous methods.

Findings

Achieved shear measurement accuracy of m=0.0021±0.0004 on simulated data

Implemented an efficient algorithm measuring ~10 galaxies/second/core

Validated the method's robustness with deep sky exposure simulations

Abstract

We demonstrate highly accurate recovery of weak gravitational lensing shear using an implementation of the Bayesian Fourier Domain (BFD) method proposed by Bernstein & Armstrong (2014, BA14), extended to correct for selection biases. The BFD formalism is rigorously correct for Nyquist-sampled, background-limited, uncrowded image of background galaxies. BFD does not assign shapes to galaxies, instead compressing the pixel data D into a vector of moments M, such that we have an analytic expression for the probability P(M|g) of obtaining the observations with gravitational lensing distortion g along the line of sight. We implement an algorithm for conducting BFD's integrations over the population of unlensed source galaxies which measures ~10 galaxies/second/core with good scaling properties. Initial tests of this code on ~10^9 simulated lensed galaxy images recover the simulated shear to a fractional accuracy of m=0.0021+-0.0004, substantially more accurate than has been demonstrated previously for any generally applicable method. Deep sky exposures generate a sufficiently accurate approximation to the noiseless, unlensed galaxy population distribution assumed as input to BFD. Potential extensions of the method include simultaneous measurement of magnification and shear; multiple-exposure, multi-band observations; and joint inference of photometric redshifts and lensing tomography.