COSMOS-Web galaxy groups: Evolution of red sequence and quiescent galaxy fraction

This study analyzes COSMOS-Web galaxy groups from z=0 to z=3.7 using machine learning and multi-method approaches to reveal that quiescent galaxy fractions build up steadily from z=1.5–2 with earlier growth in richer groups, identifies a rare overdensity at z=3.4, and finds that while X-ray faint groups have lower quiescent fractions, the red sequence's slope and scatter remain stable over approximately 12 billion years.

The Gist

Imagine the universe as a giant, bustling city that has been growing and changing for over 13 billion years. In this city, galaxies are like neighborhoods. Some neighborhoods are quiet, orderly, and full of older residents who have stopped having children (these are quiescent galaxies). Others are chaotic, noisy, and full of young families building new things (these are star-forming galaxies).

This paper is a report from a team of astronomers who went on a deep-space road trip to study how these "galaxy neighborhoods" (specifically groups and clusters) evolved over time. They used a powerful new telescope, the James Webb Space Telescope (JWST), to look back in time, all the way to when the universe was just a toddler.

Here is the story of their journey, broken down into simple parts:

1. The Map and the Detective (The Data)

To find these neighborhoods, the team used a special digital tool called AMICO. Think of AMICO as a very smart detective who looks at a crowded room and says, "These people are standing close together and seem to know each other; they must be a group!"

Usually, detectives look for people wearing the same color shirt to find a group. But this detective is special: it doesn't look at colors. It just looks at where people are standing and how old they seem. This is crucial because the astronomers wanted to see how the "quiet neighborhood" (the Red Sequence) formed naturally, without the detective accidentally picking only the quiet people to begin with.

They looked at the COSMOS field, a patch of sky that is like a "super-library" of the universe, filled with data from X-rays to radio waves.

2. Teaching the Computer to Spot the "Quiet" Neighbors (Machine Learning)

The astronomers needed to sort the galaxies into "Quiet" (Red) and "Busy" (Blue). In the past, they used simple rules, like "If a galaxy is this specific shade of red, it's quiet." But the universe is messy, and simple rules often miss the nuance.

So, they taught a computer using Machine Learning.

• The Analogy: Imagine you are teaching a child to distinguish between a sleeping cat and a playing dog. You don't just say "if it's still, it's a cat." You show the child thousands of examples, pointing out the ears, the fur, the posture, and the context.
• The Result: The computer (using an algorithm called XGB) learned to look at the "light" coming from the galaxies in different colors (like red, blue, and infrared). It became a master sorter, correctly identifying quiet galaxies with over 93% accuracy, even when some data was missing (like a blurry photo).

3. The Great Transformation (The Findings)

The team wanted to know: When did the galaxies stop having babies (star formation) and start settling down?

• The Timeline: They found that in the richest, most crowded neighborhoods (the biggest groups), the "quiet" galaxies started showing up around 10 billion years ago (redshift $z \approx 2$).
• The Trend: It happened faster in the big, crowded cities. The richer the group, the sooner the galaxies turned "red" and stopped forming stars. It's like how a busy city center might get old and quiet faster than a small, isolated village.
• The "Ghost" Overdensity: The most exciting discovery was a tiny, super-dense cluster of quiet galaxies at $z = 3.4$ (over 11.5 billion years ago). This is like finding a fully formed, quiet retirement community in the middle of a construction site. It's one of the oldest "red sequences" ever seen, suggesting that some galaxies grew up and settled down incredibly fast.

4. The "X-Ray" Connection (Why are some quiet?)

The team also checked if these groups were glowing with X-rays (which happens when gas is super hot and dense).

• The Discovery: Groups that glowed brightly in X-rays had more quiet galaxies. Groups that were "X-ray faint" (dim) had fewer quiet galaxies.
• The Metaphor: Think of the universe as a web of strings (filaments).
  • X-ray Bright Groups are like the intersections of the web (nodes). They are in the middle of the action, swallowing up galaxies that have already been "pre-processed" (quieted down) in the strings leading to them.
  • X-ray Faint Groups are like the strings themselves (filaments). They are on the outskirts, picking up fresh, noisy galaxies from the empty space that haven't been quieted down yet.

5. The "Red Line" (The Red Sequence)

Astronomers look at a chart where the color of a galaxy is plotted against its brightness. The quiet galaxies form a tight, straight line called the Red Sequence.

• The Question: Does this line change as we look further back in time? Does it get steeper or wobbly?
• The Answer: Surprisingly, no. The line stayed remarkably straight and consistent for 12 billion years. This tells us that once a galaxy joins the "quiet club," it stays quiet and behaves very predictably, regardless of how old the universe is.

Summary: What does this all mean?

This paper is like a history book for galaxy neighborhoods. It tells us that:

1. Nature vs. Nurture: While a galaxy's own mass matters, the environment (living in a crowded group) is a huge factor in making a galaxy "quiet."
2. Fast Track to Maturity: In the early universe, the biggest groups managed to turn their galaxies "red" and stop their star formation much faster than we expected.
3. The Cosmic Web Matters: Where a group sits in the cosmic web (a busy intersection vs. a quiet string) determines how many quiet galaxies it has.

The astronomers used the most powerful telescope ever built and the smartest computer tricks to prove that even in the chaotic, baby universe, some galaxies knew how to grow up and settle down very early in life.

Technical Summary

Here is a detailed technical summary of the paper "COSMOS-Web galaxy groups: Evolution of red sequence and quiescent galaxy fraction" by Toni et al. (2025).

1. Problem and Scientific Context

The study addresses the formation and evolution of quiescent (red, passive) galaxies within galaxy groups and low-mass clusters across cosmic time. While the "red sequence" (RS) and quiescent fractions are well-characterized in massive local clusters and up to intermediate redshifts ($z \sim 1$), there is a lack of understanding regarding:

• How the first galaxies quenched and built up the red sequence at high redshifts ($z > 1.5$).
• The evolution of these properties specifically in galaxy groups, which dominate the halo mass function and host the bulk of the universe's galaxies, rather than just massive clusters.
• The interplay between intrinsic galaxy properties (mass) and environmental factors (density, X-ray emission, large-scale structure) in driving quenching.

The authors aim to characterize red fractions and red sequence parameters in COSMOS groups spanning $\sim 12$ Gyr of cosmic history ($z = 0$ to $z = 3.7$).

2. Methodology

Data Sets

• COSMOS-Web Group Catalog: The primary sample consists of 1,678 group and protocluster core candidates detected in the COSMOS-Web field using the AMICO (Adaptive Matched Identifier of Clustered Objects) algorithm. This catalog spans $z = 0.08$ to $3.7$ and is based on the deep COSMOS2025 photometric catalog (JWST NIRCam + UltraVISTA).
• AMICO-COSMOS: A secondary sample of 622 groups in the full COSMOS field (based on COSMOS2020) with associated X-ray properties (flux, luminosity, significance) used to study environmental effects.
• Training Data: The COSMOS2015 catalog was used as the ground truth for training machine learning models to avoid data leakage and overfitting.

Machine Learning Classification

To distinguish between quiescent and star-forming galaxies, the authors developed a probabilistic classification framework:

• Algorithms: They tested XGBoost (XGB, a non-linear gradient boosting tree) and Linear Discriminant Analysis (LDA).
• Features: Rest-frame magnitudes in bands $NUV, u, r, i, z, Y, J, H, K$. Feature importance analysis identified $r, NUV, J,$ and $K$ as the most constraining.
• Handling Imbalance: Since quiescent galaxies are the minority class, the authors utilized SMOTE (Synthetic Minority Over-sampling Technique) for LDA and tested XGB with and without missing value imputation (using a flag value of -99.9).
• Performance: The XGB model with missing value imputation achieved the best balance, with an F1-score $> 93\%$ for quiescent galaxies and well-calibrated probabilities.

Quiescent Fraction Estimation

Three methods were employed to calculate the quiescent fraction ($f_q$) within groups:

1. Pure Membership: Summing membership probabilities weighted by quiescent probabilities (model-dependent).
2. Secondary Association Subtraction: Correcting for galaxies associated with multiple detections.
3. Cylinder Background Subtraction: A model-independent technique subtracting the density of field galaxies within a cylinder ($0.5$ Mpc/h radius) around the group center.

Red Sequence Detection

• Observed Frame: Used matched rest-frame photometry (selecting band pairs bracketing the 4000 Å break) to define the Color-Magnitude Diagram (CMD).
• Fitting: Applied weighted least squares (WLS) with $3\sigma$ clipping to identify the ridgeline.
• Rest-Frame: Cross-validated using LePhare rest-frame magnitudes ($J$ and $K_s$).

3. Key Contributions

1. Robust ML Classification for High-$z$ Groups: Demonstrated that XGBoost classifiers trained on rest-frame magnitudes can effectively identify quiescent galaxies in deep, high-redshift photometric catalogs, handling missing data and class imbalance better than traditional cuts or linear methods.
2. Model-Independent Analysis: Successfully applied a cylinder-based background subtraction method to derive quiescent fractions independent of the AMICO detection model assumptions (NFW profile, Schechter function).
3. High-Redshift Red Sequence Characterization: Provided the first extensive characterization of red sequence parameters (slope, scatter, zero-point) in galaxy groups from $z=0$ to $z=3.7$, extending beyond the limits of previous massive cluster studies.
4. Discovery of a High-$z$ Overdensity: Identified a rare, compact overdensity of quiescent galaxies at $z = 3.4$, potentially one of the most distant early red sequences observed.

4. Key Results

• Quiescent Fraction Evolution:

  • The quiescent population in groups builds up steadily starting from $z \sim 1.5 - 2$.
  • Richness Dependence: The growth is faster and occurs earlier in the richest groups ($\lambda^* > 20$) compared to poorer groups.
  • X-ray Dependence: Groups with significant X-ray emission (X-ray bright) have, on average, higher quiescent fractions than X-ray faint groups.
  • Environmental Link: X-ray faint groups are more likely located in cosmic filaments, where pre-processing of galaxies is lower. X-ray bright groups are found in cosmic nodes, where accreted galaxies are more likely pre-processed and quenched.

• Red Sequence Parameters:

  • Ridgeline Position: The average observed colors of the red sequence are consistent with passively evolving elliptical galaxy models.
  • Slope and Scatter: No significant evolution in the slope or scatter of the red sequence was found over the studied redshift range ($z=0$ to $3.7$). The scatter remains below$0.10$ mag for most detections, consistent with local cluster properties.
  • High-$z$ Detection: At $z = 3.4$, a group (ID CW117) was found containing 3 quiescent galaxies within $\sim 250$ kpc/h. Their colors align with evolutionary synthesis models for a galaxy formed at $z_f = 8$ with a burst at $z=5$.

5. Significance

• Cosmic Quenching Timeline: The results confirm that the mechanisms quenching galaxies in dense environments (groups) are active as early as $z \sim 2$, with the richest systems quenching first. This supports the "downsizing" scenario where massive systems evolve earlier.
• Environment vs. Mass: The study highlights the critical role of the large-scale structure (filaments vs. nodes) in galaxy evolution. The lower quiescent fraction in X-ray faint (filament) groups suggests that environmental pre-processing is a key driver of quenching before galaxies reach the cluster core.
• JWST Capabilities: This work demonstrates the power of JWST deep fields (COSMOS-Web) combined with advanced algorithms (AMICO + ML) to probe the earliest stages of galaxy group assembly and the formation of the red sequence, pushing the boundaries of known high-redshift quiescent systems.
• Future Directions: The detection of the $z=3.4$ overdensity underscores the need for spectroscopic confirmation to validate these high-redshift red sequences and understand the precise formation history of the earliest quiescent populations.