astronomy/ai in a supporting role/Astronomy and Astrophysics 2022 · v2

Astronomers map the mass of galaxy cluster Abell 2744 using lensed images

Researchers built a model of how mass is spread through the galaxy cluster Abell 2744, using archival telescope data. A neural network sorted 23 of the 225 cluster galaxies that went into the model.

spectrum · one line per step, placed by what the step does · bright lines used AI

New high-precision strong lensing modeling of Abell 2744
Astronomy and Astrophysics, 2022

doi:10.1051/0004-6361/202244575 · record aix-00251 v2 · checked 2026-10-09

ai-supportingrole of AI

AI was for: Classification
Model family: Convolutional neural network
Checked by: Held-out
Code: not reported

AI processed or interpreted data, but the main finding does not rest on it.

read as

The science is explained before the AI appears. Switch to field specialist to go straight to the method.

Assumes the discipline and goes straight to the method.

The diagram, the record and what the paper did not report are identical in both modes. Only the framing changes — never the evidence.

Introduction by AIxSci · plain language

What this research was about

A galaxy cluster is a vast collection of galaxies bound together by gravity, and its mass bends the light of anything lying behind it. Where the bending is strong, a single distant galaxy can appear two or three times over, in different spots on the sky. Working backwards from those repeated images to the mass that caused them is possible, but fiddly. You need to be sure which smudges on an image are the same object seen twice, how far away each one is, and which galaxies genuinely belong to the cluster rather than sitting in front of or behind it.

The team set out to build a model of the mass in Abell 2744 from archival observations: multi-band images from the Hubble Space Telescope and spectra from the MUSE instrument on the Very Large Telescope in Chile. Spectra reveal distance, so they could confirm 90 multiple images coming from 30 background sources, and use those as the constraints the model must reproduce. They also measured the internal motions of stars in 85 cluster galaxies. The model was then fitted with the LensTool software, and judged by how closely its predicted image positions matched the real ones.

Where AI came in

Membership of the cluster is easiest to settle from a spectrum, but spectra are expensive and not every galaxy has one. The researchers used an existing convolutional neural network — a type of program trained to recognise patterns in images — to decide membership from Hubble picture cutouts alone. It had been trained on roughly 3,300 labelled examples drawn from 14 other clusters. Tested against the galaxies in Abell 2744 whose membership was already known from spectra, it recovered 88 per cent of them and was right 95 per cent of the time.

It added 23 galaxies to the 202 already confirmed by spectra, giving 225 in total. Each counts as a small lump of mass within the cluster, so the network was standing in for the conventional colour-based sorting astronomers use when spectra are missing. Everything else was done without machine learning: the distances and stellar motions came from fitting template spectra, the multiple images were picked out by eye, and the mass distribution itself was fitted statistically in LensTool.

Written by AIxSci from the checked record below, to give context for readers outside the field. It is not part of the record.

The work

Technical · from the record

The authors built a parametric strong lensing model of the galaxy cluster Abell 2744 from archival HST imaging and VLT/MUSE spectroscopy, constrained by 90 spectroscopically confirmed multiple images from 30 background sources. A convolutional neural network applied to multi-band HST cutouts selected 23 photometric cluster members, which were added to 202 spectroscopically confirmed members to give a 225-galaxy subhalo component. Stellar velocity dispersions measured for 85 member galaxies were used to set the member scaling relations. The reference model, which represents the cluster environment with three clumps fixed on the brightest outer galaxies, reproduces the observed image positions with a root-mean-square displacement of 0.37 arcseconds, compared with 0.44 arcseconds for the variant based on weak-lensing clumps.

How AI was used

A convolutional neural network classifier, developed on multi-band HST image cutouts with spectroscopic labels from approximately 3,300 samples across 14 CLASH and HFF clusters at redshifts 0.2 to 0.6, was applied to A2744 to identify bright photometric cluster members that lack spectroscopic redshifts. Its selections were appended to the spectroscopically selected member catalog, and the combined member list defined the galaxy-scale subhalo mass component of the lens model, whose velocity dispersions and truncation radii were tied to galaxy luminosity through scaling relations. All other steps used non-learned methods: redshift and velocity-dispersion measurement by template fitting, visual identification of multiple images and lensed clumps, and Bayesian optimisation of the parametric mass distribution in LensTool.

The shape of the work

Structural · the record, drawn

AI stepNo AI↤ what the AI stood in for

1Acquisition
no AI

Assemble archival imaging and spectroscopy

Obtaining raw data, whether by measurement, download or retrieval.

we use archival observations from the MUSE integral field spectrograph, mounted on the VLTwhere the paper describes this · verbatim

in the paper

2Preparation
no AI

Build secure multiple-image catalog

Cleaning, filtering, normalising or labelling data already obtained.

we construct our sample by considering only secure systems, that are spectroscopically confirmed by our VLT/MUSE analysis with a QF value ≥2where the paper describes this · verbatim

in the paper

3Preparation
no AI

Select spectroscopic cluster members

Cleaning, filtering, normalising or labelling data already obtained.

Spectroscopically confirmed cluster members are identified as those galaxies, brighter than mF160W=24where the paper describes this · verbatim

in the paper

4Inference
AI

Classify photometric cluster members with CNN

Running a trained model over new data to predict, classify or score. The AI stood in for conventional algorithm.

based on a convolution neural network (CNN) technique, which identifies cluster members using multi-band HST image cutoutswhere the paper describes this · verbatim

in the paper

5Preparation
no AI

Measure internal stellar velocity dispersions

Cleaning, filtering, normalising or labelling data already obtained.

Velocity dispersions are then measured using the publicly available software Penalized Pixel-Fitting method, over the wavelength range [3700-5700] Åwhere the paper describes this · verbatim

in the paper

6Optimisation
no AI

Optimise parametric cluster mass model

Iterative search over a space.

We develop a new lens model of A2744 using the publicly available software LensTool, which reconstructs the total mass distribution of a galaxy clusterwhere the paper describes this · verbatim

in the paper

7Validation
no AI

Assess model against observed image positions

Testing outputs against ground truth.

we also consider and quote the root-mean-square separation between the observed and model-predicted positions of the multiple imageswhere the paper describes this · verbatim

in the paper

What the record says

Technical · every part carries its own basis

+ in the paper~ our reading− not reported

How to read the quotations. A quotation shows where the paper describes something. It does not quote every value beside it: one passage locates a part of the work, and values without their own quotation are our reading of that passage.

~Role of AI

AI in a supporting roleour reading

A CNN supplied 23 of the 225 cluster member galaxies entering the lens model; the mass reconstruction itself is a parametric Bayesian fit with LensTool, so the headline result does not rest on the learned model

+What the AI was for

Classificationin the paper

identified with a convolution neural network methodology with a high degree of puritywhere the paper describes this · verbatim

+Model families

Convolutional neural networkin the paper

+How it was taught

Supervisedin the paper

+Models named

convolutional neural network cluster-member classifier (CLASH-VLT methodology) · Off the shelfin the paper

+How results were checked

Held-outin the paper

When tested on the spectroscopic sample of A2744, we measure a completeness level of 88% and a high degree of purity, equal to 95%.where the paper describes this · verbatim

−Code · weights · data

code not reportedweights not reporteddata not reportednot reported

−Compute

not reportednot reported

What this paper did not report

Technical · absence is published deliberately

Reported as not stated — 6 items

CodeWhether the code is available is not stated.
Trained model weightsWhether the trained model is available is not stated.
DataWhether the data are available is not stated.
ComputeThe hardware or time used is not stated.
How many were testedThe paper gives no count of what was tested.
Version of convolutional neural network cluster-member classifier (CLASH-VLT methodology)Which version of the model was used is not stated.

About this article

Record aix-00251, version 2, checked by a person on 2026-10-09. The record describes the paper; it does not assess whether the paper's findings are right. The paper is published under CC-BY; quotations are at most 25 words. How we work · Report an error