Automated galaxy–galaxy strong lens modelling: No lens left behind

Abstract

The distribution of dark and luminous matter can be mapped around galaxies that gravitationally lens background objects into arcs or Einstein rings. New surveys will soon observe hundreds of thousands of galaxy lenses, and current, labour-intensive analysis methods will not scale up to this challenge. We develop an automatic, Bayesian method which we use to fit a sample of 59 lenses imaged by the Hubble Space Telescope. We set out to leave no lens behind and focus on ways in which automated fits fail in a small handful of lenses, describing adjustments to the pipeline that ultimately allows us to infer accurate lens models for all 59 lenses. A high success rate is key to avoid catastrophic outliers that would bias large samples with small statistical errors. We establish the two most difficult steps to be subtracting foreground lens light and initialising a first, approximate lens model. After that, increasing model complexity is straightforward. We put forward a likelihood cap method to avoid the underestimation of errors due to pixel discretization noise inherent to pixel-based methods. With this new approach to error estimation, we find a mean $\sim 1{{\%}}$ fractional uncertainty on the Einstein radius measurement which does not degrade with redshift up to at least z = 0.7. This is in stark contrast to measurables from other techniques, like stellar dynamics, and demonstrates the power of lensing for studies of galaxy evolution. Our PyAutoLens software is open source, and is installed in the Science Data Centres of the ESA Euclid mission.

Holismokes

“…The latter is the best match to our networks, and is comparable in performance, as mentioned in S21a. There is also currently a lot of work going into automated modeling without machine learning (e.g., Nightingale et al 2018Nightingale et al , 2021Rojas et al 2022;Savary et al 2022;Ertl et al 2023;Etherington et al 2022;Gu et al 2022;Schmidt et al 2023), which typically performs better than neural networks but requires significantly longer run times of hours to days. We refer to Schuldt et al (2022) for a direct comparison between the network presented here and traditionally obtained models for real HSC lenses.…”

Modeling Strong Lenses from Wide-Field Ground-Based Observations in KiDS and GAMA

“…Forming robust quantitative goodness-offit metrics is currently an open problem in automated lens modeling. Etherington et al (2022) explored this problem using P A L with much higher resolution images and found that none did a particularly satisfactory job. We take the Bayesian evidence to be the reference figure of merit and follow this with a blind visual inspection of the image, fit, and spectrum of each modeled candidate.…”

Mass Reconstruction of Galaxy-scale Strong Gravitational Lenses Using Broken Power-law Model

“…For most of galaxy-scale lenses, their Einstein radii are typically smaller than their half-light radii, making it difficult to clearly identify the relatively faint lensed images. In order to subtract the foreground light distributions, parametric light profiles, e.g., Sérsic or double Sérsic profiles (Bolton et al 2008;Shajib et al 2019;Birrer et al 2020;Etherington et al 2022), are usually adopted as fitting models. However, these profiles with a limited number of parameters may result in undesired residuals, especially for the light distributions with complex angular structures.…”

“…Usually, the Einstein radius is deemed to be such a quantity (Schneider et al 2006;Treu 2010). Its measurement error is typically of the order of a few per-cent (Bolton et al 2008;Shu et al 2015Shu et al , 2016Shajib et al 2021;Etherington et al 2022). However, there are also works showing surprising results.…”