AutoLens: Automated Modeling of a Strong Lens's Light, Mass and Source

TL;DR

AutoLens is an automated software suite that models galaxy-scale strong gravitational lenses by simultaneously analyzing the lens's light and mass, reconstructing the source, and automatically selecting model complexity, all without user intervention.

Contribution

It introduces the first fully automated pipeline for modeling strong lenses, capable of handling complex source and lens profiles with Bayesian model comparison.

Findings

High accuracy in recovering light, mass, and source profiles.

Effective on simulated data resembling Hubble and Euclid observations.

Automated model selection adapts to data complexity.

Abstract

This work presents AutoLens, the first entirely automated modeling suite for the analysis of galaxy-scale strong gravitational lenses. AutoLens simultaneously models the lens galaxy's light and mass whilst reconstructing the extended source galaxy on an adaptive pixel-grid. The method's approach to source-plane discretization is amorphous, adapting its clustering and regularization to the intrinsic properties of the lensed source. The lens's light is fitted using a superposition of Sersic functions, allowing AutoLens to cleanly deblend its light from the source. Single component mass models representing the lens's total mass density profile are demonstrated, which in conjunction with light modeling can detect central images using a centrally cored profile. Decomposed mass modeling is also shown, which can fully decouple a lens's light and dark matter and determine whether the two component are geometrically aligned. The complexity of the light and mass models are automatically chosen via Bayesian model comparison. These steps form AutoLens's automated analysis pipeline, such that all results in this work are generated without any user-intervention. This is rigorously tested on a large suite of simulated images, assessing its performance on a broad range of lens profiles, source morphologies and lensing geometries. The method's performance is excellent, with accurate light, mass and source profiles inferred for data sets representative of both existing Hubble imaging and future Euclid wide-field observations.