🔭 astrophysics

ComPACT: Mass-Redshift Properties of the galaxy cluster catalogue

This paper presents the ComPACT catalogue, a deep-learning-based collection of 2,962 Sunyaev-Zel'dovich galaxy cluster candidates derived from ACT+Planck maps, which confirms approximately 60% of candidates and significantly expands the known population of high-redshift, high-mass clusters.

Original authors: S. Voskresenskaia, N. Lyskova, I. Zaznobin, A. Meshcheryakov

Published 2026-05-20

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: S. Voskresenskaia, N. Lyskova, I. Zaznobin, A. Meshcheryakov

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the universe as a giant, dark ocean. Most of the water is invisible, but hidden within it are massive, glowing islands made of hot gas and dark matter. These islands are galaxy clusters, the largest structures in the cosmos. For a long time, finding them was like trying to spot a specific type of fish in a stormy sea using only a basic flashlight.

This paper introduces a new, super-powered tool called ComPACT (which stands for a specific machine-learning project) that acts like a high-tech, deep-learning sonar. Here is what the researchers did and found, explained simply:

1. The Problem: Finding Invisible Islands

Galaxy clusters are hard to see because they are far away and often faint. Traditional methods used "match filters"—think of these as looking for a specific shape, like a perfect circle. If a cluster was slightly distorted or faint, the old tools missed it.

2. The Solution: A "Smart Eye"

The team trained a Convolutional Neural Network (CNN)—a type of artificial intelligence that learns by looking at thousands of examples, much like a child learning to recognize a cat by seeing many pictures of cats.

The Data: They fed this AI a combined map of the sky from two powerful telescopes: ACT (Atacama Cosmology Telescope) and Planck.
The Trick: Instead of looking for a perfect circle, the AI learned to recognize the complex, messy "fingerprint" of a galaxy cluster, even if it was faint or weirdly shaped.

3. The Catch: Is the AI Hallucinating?

The AI found 2,962 potential clusters. But, just like a metal detector that beeps at every soda can and every coin, the AI might have found some "fake" signals.

The Verification: The team had to go back and check these 2,962 candidates using other telescopes (optical and infrared) to see if they were real.
The Result: About 60% of the AI's suggestions were real galaxy clusters. That means they confirmed roughly 1,784 new or known clusters. The other 40% were likely just noise or false alarms.

4. Measuring the "Weight" and "Distance"

Once they confirmed the clusters were real, they needed to know two things:

How far away are they? (Redshift): They used light from the galaxies inside the clusters to calculate distance. They found clusters ranging from very close neighbors to ones so far away their light has been traveling for billions of years.
How heavy are they? (Mass): They used the heat signature of the gas inside the clusters to estimate their weight. They found masses ranging from "small" (0.25 times the mass of our Sun's galaxy) to "giant" (13 times that mass).

5. The Big Discoveries

New Giants: The AI found five massive, distant clusters that no one knew about before. These are like finding five new, huge islands in a part of the ocean we thought was empty.
Better at High Altitudes: The AI was particularly good at finding clusters that are very far away (high redshift) and very heavy. Traditional methods often miss these because they are faint, but the AI's "smart eye" could see them.
Efficiency: The new catalogue is more complete than previous ones. If you imagine previous surveys as a net with big holes, ComPACT is a net with much smaller holes, catching more of the "fish" (clusters) that were slipping through before.

Summary

The researchers built a smart computer program that scanned the sky for galaxy clusters. It found thousands of candidates, confirmed over 1,700 of them as real, and measured their distance and weight. Most importantly, it found five massive, distant giants that were previously hidden, proving that using AI to look at the universe is a powerful way to discover things we missed with older tools.

What they did NOT do:
The paper does not claim this technology can be used for medical imaging, climate change monitoring, or finding exoplanets. It is strictly about mapping galaxy clusters in the universe.

Technical Summary: ComPACT Mass–Redshift Properties

Problem and Motivation
Galaxy clusters serve as critical probes for cosmology and galaxy evolution, yet detecting them, particularly at high redshifts or with lower signal-to-noise ratios (SNR), remains challenging. Traditional Sunyaev–Zeldovich (SZ) surveys, such as those conducted by Planck, ACT, and SPT, rely on matched-filter techniques that depend on predefined cluster templates. These methods may miss systems that deviate from standard templates or fall below detection thresholds. While machine learning (ML) has shown promise in signal extraction, there is a need to rigorously characterize the completeness, purity, and physical properties (mass and redshift) of cluster catalogues generated by deep learning (DL) approaches to validate their utility for cosmological studies.

Methodology
This study analyzes the ComPACT catalogue, a dataset of 2,962 SZ-selected galaxy cluster candidates identified using a convolutional neural network (CNN) applied to combined Atacama Cosmology Telescope (ACT) and Planck intensity maps. The analysis focuses on validating the catalogue and deriving physical properties through the following steps:

Data Sources: The study utilizes combined ACT+Planck intensity maps (90, 150, 220 GHz) and Compton- $y$ maps (from both ACT+Planck and Planck full-mission NILC) for mass estimation. Optical/infrared photometry from the DESI Legacy Imaging Surveys (DECaLS DR9) and WISE is used for redshift estimation.
Redshift Determination: A hierarchical strategy is employed:
1. Cross-matching: Candidates are matched against existing spectroscopic and photometric catalogues (e.g., ACT DR6, PSZ2, SPT-DEEP) within a 5 arcmin radius.
2. Photometric Estimation: For unmatched candidates, two independent algorithms are applied:
  - Zaznobin method: A modified algorithm adapted from X-ray/SZ studies, utilizing infrared luminosity distributions and photometric redshifts from the DESI Legacy Surveys. It employs a two-stage process (preliminary and refined redshift) with specific reliability thresholds ( $p_1 > 0.978, p_2 > 0.8$ ).
  - zCluster method: An algorithm estimating redshifts via weighted summation of galaxy photometric redshift probability distributions, utilizing a contrast parameter $\delta$ to assess reliability ( $\delta \ge 3$ ).
Mass Estimation: Masses ( $M_{500c}$ ) are derived using SZ scaling relations. For unmatched objects, the signal-to-noise ratio (SNR) is calculated in the Compton- $y$ maps. The integrated Compton- $Y$ parameter ( $Y_{cyl}$ ) is measured within a 10 arcmin aperture, scaled to match Planck measurements, and converted to $M_{500c}$ using the relation from Planck Collaboration et al. (2014).
Completeness Assessment: The catalogue's completeness is evaluated by injecting synthetic clusters (based on the Universal Pressure Profile model) into real ACT+Planck maps and processing them through the DL detection pipeline.

Key Contributions and Results

Catalogue Validation: Approximately 60% of the 2,962 ComPACT candidates are confirmed as galaxy clusters. The confirmation rate is higher (63.2%) within the ACT DR5 footprint compared to outside it (38.2%), largely due to the availability of DESI Legacy Imaging Surveys photometry.
Redshift and Mass Coverage:
- Redshifts are determined for 1,771 clusters (60% of the sample), spanning $0.007 < z < 1.7$ . This includes 116 new redshift measurements, 69 derived via zCluster and 47 via the Zaznobin method.
- Masses are estimated for 56% of the sample (1,659 objects), ranging from $0.25 \times 10^{14} M_\odot$ to $13.1 \times 10^{14} M_\odot$ . This includes 158 new mass determinations (114 from ACT+Planck maps and 44 from Planck maps).
Completeness and Sensitivity: Simulations indicate that at fixed purity, the ComPACT catalogue achieves higher completeness than the ACT DR5 catalogue ($SNR > 3$), particularly for massive clusters at high redshifts ( $z > 0.8$ ). The cumulative completeness for clusters with $M_{500c} > 10^{14} M_\odot$ is estimated at 77% (interpreted as an upper limit due to the lack of SZcat selection in the simulation).
High-Mass Systems: The study identifies five previously unreported massive clusters ( $M_{500c} > 6 \times 10^{14} M_\odot$ ) at $z > 0.7$ . These discoveries increase the known population of such high-redshift, high-mass systems by approximately 10%. The catalogue also recovers well-known systems like the Bullet Cluster and El Gordo.

Significance
The paper concludes that the ComPACT catalogue effectively expands the population of SZ-selected clusters, particularly in the high-mass and high-redshift regimes where traditional matched-filter methods may be less sensitive. The results demonstrate the effectiveness of deep-learning approaches for cluster detection in microwave data, offering a valuable resource for future cosmological studies and the investigation of baryonic processes in dense regions. The work validates the DL-based detection pipeline and provides a robust set of new redshift and mass measurements for the community.