🔭 astrophysics

How to count clustered galaxies

This paper presents an empirical method that corrects the systematic bias in galaxy number counts caused by clustering in confusion-limited submillimetre surveys by combining 1- and 2-point statistics, and applies it to revise Herschel-SPIRE observations of the GOODS-N field.

Original authors: Yunting Wang, Ryley Hill, Douglas Scott, Tessa Vernstrom

Published 2026-05-12

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Yunting Wang, Ryley Hill, Douglas Scott, Tessa Vernstrom

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

The Big Problem: The "Crowded Room" Effect

Imagine you are trying to count the number of people in a very large, dark room. You can't see everyone clearly because the lights are dim and the room is full of fog.

The Easy Part: If someone is standing right in front of you and shouting, you can easily count them. In astronomy, these are the bright, nearby galaxies that telescopes can see clearly.
The Hard Part: Most galaxies are faint and far away. They are like people whispering in the back of the room. Because the telescope's "eye" (its beam) is a bit blurry, these faint whispers blend together into a single, low hum of background noise. Astronomers call this "confusion noise."

For decades, astronomers have used a statistical trick called P(D) analysis to estimate how many of these faint, whispering galaxies are out there. They look at the "hum" of the background noise and try to figure out how many individual voices created it.

The Mistake: Assuming Everyone is Alone

The standard P(D) method makes a big assumption: It assumes galaxies are scattered randomly, like raindrops falling on a sidewalk. If you assume they are random, the math works out pretty well.

But galaxies aren't random. They are social. They hang out in groups, clusters, and families, much like people at a party tend to stand in small groups rather than being evenly spaced out across the dance floor.

The authors of this paper discovered that because galaxies are clustered, the standard math gets it wrong.

When galaxies cluster together, their faint whispers merge into a louder shout.
The standard math mistakes this "group shout" for a single, very loud person.
The Result: The old method overestimates the number of bright galaxies and underestimates the number of faint ones. It's like thinking there are fewer people in the room because you counted the groups as single individuals.

The Solution: Listening to the "Group Dynamics"

To fix this, the authors developed a new method that listens to two different things at once:

The Volume (1-point statistics): How loud is the hum in each spot? (This is the standard P(D) method).
The Pattern (2-point statistics): How are the groups arranged? Are they clumped together or spread out?

The Analogy:
Imagine you are trying to guess how many people are in a room by listening to the noise.

Old Method: You just measure the total volume. If it's loud, you guess there are many people. But you don't realize that if everyone is huddled in one corner, the volume is high even if there are fewer people.
New Method: You measure the volume and you look at the pattern of the sound waves to see if people are huddled in groups. Once you know they are huddled, you can adjust your math to realize, "Ah, that loud spot isn't one giant person; it's a group of five."

What They Did

Simulated a Universe: They used a super-computer to create a fake universe (based on the "SIDES" simulation) where they knew the exact number of galaxies. They created two versions: one where galaxies were random, and one where they were clustered.
Tested the Math: They ran the old P(D) method on the clustered fake universe. As predicted, the old method got the count wrong.
Created a Correction: They figured out a mathematical "correction formula." This formula looks at how clumpy the data is (using the "group pattern" measurement) and adjusts the volume count to get the true number of people.
Applied to Real Data: They took real telescope data from the GOODS-N field (a deep patch of sky observed by the Herschel Space Telescope) at three different colors (wavelengths): 250, 350, and 500 micrometers.

The Results: How Much Was Wrong?

The correction made a huge difference, especially for the "blurriest" view (500 micrometers):

At 500 micrometers: The old method was counting galaxies as if there were 1.6 times more of them than there actually were around a specific brightness level. It was significantly overestimating the crowd.
At 350 and 250 micrometers: The error was smaller because the telescope's "eye" was sharper (less blurry) at these wavelengths, but the correction was still necessary.

Why This Matters

This paper provides a new, more accurate way to count the faintest galaxies in the universe.

For Cosmology: It helps scientists understand how galaxies form and evolve. If you count the galaxies wrong, your models of the universe's history are wrong.
For Future Telescopes: This method isn't just for the Herschel telescope. The authors say it can be used for current and future telescopes (like SCUBA-2 and the upcoming CCAT) that look at the universe in "foggy" submillimeter light.

In short: The authors realized that because galaxies like to hang out in groups, the old way of counting them was like miscounting a crowd by ignoring the groups. They built a new calculator that accounts for the "clumping," giving us a much more accurate census of the universe's faintest inhabitants.

Technical Summary: How to Count Clustered Galaxies

Problem Statement
Obtaining robust galaxy number counts is essential for understanding galaxy evolution and refining subgrid physics in cosmological simulations. However, at submillimetre wavelengths, surveys are often limited by "confusion noise," where faint, unresolved galaxies contribute collectively to background fluctuations rather than being individually detected. Standard statistical methods, specifically P(D) fluctuation analysis, are employed to recover counts in these confusion-limited regimes. A fundamental limitation of the standard P(D) framework is its assumption that galaxies are Poisson-distributed (randomly distributed) on the sky. In reality, galaxies are spatially clustered. This clustering distorts the flux histogram (the 1-point statistic) of the map, leading to systematic biases in the derived galaxy number counts. Previous attempts to address this have relied on analytical formulations that struggle with realistic count models or empirical corrections based on specific simulations that may not perfectly mirror observed fields.

Methodology
The authors propose an empirical, data-driven framework that simultaneously utilizes 1-point and 2-point statistics to correct for clustering bias. The methodology proceeds as follows:

Simulation and Bias Characterization: Using the Simulated Infrared Dusty Extragalactic Sky (SIDES) catalogue, the authors generate simulated Herschel-SPIRE maps with varying degrees of clustering strength (characterized by the clustering scale $\theta_0$ ). They compare the flux histograms of these clustered maps against "randomised" maps (where galaxy positions are shuffled) to quantify how clustering alters the probability density function (PDF) of pixel fluxes.
Correction Function Development: The difference between the clustered and randomised histograms is parameterized using an asymmetric "W-shaped" correction function. This function combines second-order polynomials and skewed Gaussians on either side of the histogram peak. The parameters of this function are derived as simple linear or log-linear relations to the characteristic clustering scale $\theta_0$ .
Clustering Strength Estimation: To apply the correction to real data without prior knowledge of the counts, the authors estimate the clustering strength directly from the map's 2-point statistics. They calculate the angular power spectrum of the map, subtract the shot-noise component, and fit the remaining power-law slope to determine the characteristic clustering scale $\theta_0$ .
Iterative Correction: The estimated $\theta_0$ is used to generate the specific correction function for that dataset. This function is applied to the observed flux histogram to produce a "clustering-corrected" histogram.
P(D) Fitting: The corrected histogram is then fed into a standard P(D) analysis using a Markov chain Monte Carlo (MCMC) approach with a node-based model to derive the unbiased galaxy number counts.

Key Results
The authors applied this methodology to deep Herschel-SPIRE observations of the GOODS-N field at 250, 350, and 500 $\mu$ m.

Magnitude of Bias: Clustering significantly inflates the apparent number counts at 500 $\mu$ m, where the beam size is largest. At a flux density of approximately 10 mJy, the uncorrected P(D) analysis overestimates the counts by a factor of 1.6. The effect is less severe at 350 $\mu$ m and 250 $\mu$ m due to smaller beam sizes.
Correction Efficacy: When the correction is applied to simulated maps, the recovered galaxy number counts show a dramatic improvement in goodness-of-fit ( $\chi^2$ dropping from 10,581 to 915) compared to the true underlying catalogue counts.
Observational Application: The revised counts for the GOODS-N field show excellent agreement with deep, direct counts from SCUBA-2 observations at 450 $\mu$ m. The corrected counts also align well with SIDES simulations.
Comparison to Literature: The results differ from previous P(D) analyses (e.g., Glenn et al. 2010) that lacked clustering corrections, particularly at the faint end. The authors also note discrepancies with stacking analyses (e.g., Béthermin et al. 2012) and multiwavelength deblending results (Wang et al. 2019), attributing the latter to potential incompleteness in the prior catalogues used for deblending.

Significance and Claims
The paper claims to present a robust, generalised tool for extracting intrinsic galaxy number counts from confusion-limited maps. By combining 1-point (flux histogram) and 2-point (power spectrum) statistics, the method maximizes the information extracted from the data without requiring external priors from other wavelengths. The authors assert that this approach enables unbiased exploitation of current and future far-infrared and submillimetre surveys, including those from SCUBA-2 and the upcoming Cerro Chajnantor Atacama Telescope (CCAT). The methodology is presented as broadly applicable to any dataset with well-characterised beam and noise properties, offering a solution to a long-standing systematic bias in submillimetre astronomy.