🔭 astrophysics

Owl-z: a Bayesian tool to select z \geq 7 quasars

Owl-z is a versatile Bayesian tool designed to identify high-redshift (z ≥ 7) quasars and distinguish them from contaminants like brown dwarfs and intermediate-redshift galaxies in wide-field surveys, demonstrating high performance and adaptability for optimizing follow-up observations in the Euclid mission and beyond.

Original authors: Meriam Ezziati, Roser Pello, Jean-Gabriel Cuby, Pierre Pudlo, François-Xavier Dupé, Jean-Charles Lambert, Jean-Charles Cuillandre, Olivier Ilbert, Sylvain de la Torre, Stéphane Arnouts, Eric Jullo, Da

Published 2026-01-15

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Meriam Ezziati, Roser Pello, Jean-Gabriel Cuby, Pierre Pudlo, François-Xavier Dupé, Jean-Charles Lambert, Jean-Charles Cuillandre, Olivier Ilbert, Sylvain de la Torre, Stéphane Arnouts, Eric Jullo, Daming Yang

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, cosmic ocean. Deep within this ocean, there are lighthouses called quasars. These are super-bright beacons powered by massive black holes, and the oldest ones (from when the universe was very young, over 13 billion years ago) are incredibly hard to find. They are so far away that their light has stretched out, making them look very red, almost invisible to our eyes.

The problem? The ocean is full of "fog" and "decoys." There are other objects, like brown dwarfs (failed stars that are dim and cool) and ancient galaxies, that look suspiciously similar to these ancient quasars. If you try to find the real lighthouses just by looking at their color, you'll get lost in the fog.

This paper introduces a new tool called Owl-z. Think of Owl-z as a super-smart, digital detective that uses a method called Bayesian statistics. Instead of just guessing, Owl-z acts like a seasoned investigator who weighs the odds.

Here is how it works, using a few simple analogies:

1. The Detective's Toolkit (The Bayesian Method)

Imagine you are trying to identify a person in a crowd.

The Old Way (Color-Cutting): You might say, "If they are wearing a red hat, they are the suspect." But many innocent people wear red hats too. This leads to mistakes.
The Owl-z Way: Owl-z doesn't just look at the hat. It looks at the whole picture. It asks: "How likely is it that this person is a suspect given their hat, their height, their location, and how many suspects usually hang out in this area?"
It calculates a probability score. It doesn't just say "Yes" or "No"; it says, "There is a 99% chance this is a real quasar, and only a 1% chance it's a brown dwarf."

2. The Three Suspects

Owl-z is trained to distinguish between three types of cosmic "suspects":

The Target: High-redshift quasars (the ancient lighthouses).
The Imposter #1: Brown dwarfs (cool, dim stars that look red).
The Imposter #2: Old, dusty galaxies that happen to look red because of their distance.

Owl-z uses a massive library of "mugshots" (spectral templates) of what these objects should look like. It compares the actual data from telescopes against these mugshots to see which one fits best.

3. The "F-Measure" Scorecard

How do we know if Owl-z is good at its job? The authors use a score called the F-measure.

Think of this like a report card for a student.
Completeness: Did the student find all the right answers? (Did we find all the quasars?)
Purity: Did the student avoid picking wrong answers? (Did we avoid picking up fake quasars?)
The F-measure combines these two. A high score means Owl-z is finding the real quasars without getting tricked by the imposters.

4. What the Paper Found

The authors tested Owl-z using two main methods:

The "Re-Identification" Test: They fed Owl-z data from quasars we already know exist. Owl-z successfully found them all, proving it can recognize the real deal.
The "Simulation" Test: They created a fake universe in a computer with millions of fake quasars and imposters. They let Owl-z loose to see how well it performed.

The Results:

Bright Sources: When the quasars are bright (like a lighthouse in clear weather), Owl-z is nearly perfect. It finds them easily and rarely gets confused.
Faint Sources: When the quasars are very dim (like a lighthouse in a thick fog), it gets harder. The "fog" (noise) makes it harder to tell the difference between a real quasar and a brown dwarf.
The Sweet Spot: Owl-z works best for objects that are about two "magnitudes" (a measure of brightness) brighter than the telescope's limit. For these, it provides a very reliable list of candidates for astronomers to study further.

5. The "Adjustable Dial"

One of the coolest features of Owl-z is that it has a dial.

If astronomers want to be super careful and only look at the absolute most likely candidates (high purity), they can turn the dial up. They might miss a few faint quasars, but they won't waste time on fakes.
If they want to cast a wide net to make sure they don't miss anything (high completeness), they can turn the dial down, accepting that they might have to check a few more fakes later.
The paper shows that by adjusting this dial based on how bright and how far away an object is, astronomers can get the best possible results.

Summary

Owl-z is a powerful, open-source software tool designed to help astronomers find the oldest, most distant quasars in the universe. It acts like a smart filter that separates the real cosmic lighthouses from the confusing fog of brown dwarfs and old galaxies. While it works best on bright objects, it is flexible enough to be tuned for different needs, making it a vital tool for upcoming massive sky surveys like the Euclid mission. It doesn't just find things; it tells you how confident it is that what it found is real.

Technical Summary of "Owl-z: a Bayesian tool to select z ≳7 quasars"

Problem Statement
The identification of high-redshift ( $z \gtrsim 7$ ) quasars is critical for understanding the early Universe, the genesis of supermassive black holes, and the epoch of reionization. However, photometric selection of these objects in wide-field surveys is challenging due to their faintness and the severe contamination from interloping populations. Specifically, the red colors of high- $z$ quasars, caused by the Lyman- $\alpha$ break, are mimicked by brown dwarfs (spectral types M, L, T) and intermediate-redshift early-type galaxies ( $1 \lesssim z \lesssim 2$ ). Traditional color-cut methods often struggle to distinguish these populations, particularly at lower redshifts ( $7 < z < 8$ ) where the color spaces of quasars and contaminants overlap significantly. With the advent of massive surveys like Euclid, there is a need for efficient, optimized statistical tools to select high- $z$ quasar candidates while minimizing the demand for costly spectroscopic follow-up.

Methodology
The paper introduces Owl-z, an open-source Bayesian code designed to calculate the posterior probability that a source belongs to one of three modeled populations: high- $z$ quasars ( $7 \le z \le 12$ ), intermediate-redshift early-type galaxies ( $1 < z < 2$ ), or brown dwarfs (MLT stars, spectral types M6–T9).

Bayesian Formalism: The code employs Bayes' theorem to compute the posterior probability $P(t|D)$ for a source type $t$ given photometric data $D$ . This is driven by the prior surface density of the populations and the integrated likelihood (evidence) of the data given the source type.
Population Modeling:
- Quasars: Modeled using a collection of spectral energy distribution (SED) templates (smoothed composites and parametric models) and a double power-law luminosity function (LF) derived from Willott et al. (2010) and Matsuoka et al. (2023).
- Brown Dwarfs: Modeled using observed spectral libraries (LHS and SpeX prism) and WISE photometry. The spatial distribution is modeled as a thin disk with an exponential vertical and horizontal distribution, restricted to the thin disk approximation for the EWS depth, with surface densities dependent on Galactic latitude.
- Contaminant Galaxies: Modeled using early-type galaxy SEDs (SSP models) and a luminosity function derived from the E-COSMOS catalogue (Euclidised COSMOS2022), fitted with a Schechter function.
Likelihood Calculation: The likelihood of the data is computed as a mixture of multivariate densities centered on model fluxes, accounting for Gaussian photometric errors and non-detections (using forced photometry or upper limits).
Implementation: Written in Python (utilizing NumPy, SciPy, Astropy, and Cython for speed), the code takes photometric data and Galactic coordinates as input and outputs probabilities for each population, along with best-fit parameters (redshift, magnitude, spectral type).

Key Contributions

Development of Owl-z: A flexible, fast, and public Bayesian tool specifically tuned for the Euclid Wide Survey (EWS) but adaptable to other surveys (e.g., LSST, Roman).
Comprehensive Contaminant Modeling: Explicit modeling of both brown dwarf populations (including the impact of Galactic latitude and thin/thick disk approximations) and intermediate-redshift early-type galaxies, which are identified as the primary contaminants.
Validation on Real Data: Successful re-identification of all known spectroscopically confirmed quasars at $z > 7$ from various surveys (DECaLS, Pan-STARRS, HSC, UKIDSS, WISE), demonstrating the code's versatility across different photometric datasets.
Performance Metrics: Introduction of a global performance metric, the F-measure, which combines completeness (True Positives / (True Positives + False Negatives)) and purity (True Positives / (True Positives + False Positives)) to evaluate selection efficiency.

Results

Re-identification: Owl-z achieved a probability $P_q > 0.99$ for all eight tested spectroscopically confirmed quasars at $z > 7$ , correctly distinguishing them from MLT dwarf contaminants.
Completeness and Purity (Simulations): Using simulated EWS data ( $7 \le z \le 12$ $7 \leq z \leq 12$ , $18 \le H_E \le 24.5$ $18 \leq H_{E} \leq 24.5$ ):
- Bright Sources: For sources brighter than $H_E \approx 22$ , Owl-z reaches full performance (high F-measure) with high completeness and purity, largely independent of redshift.
- Faint Sources: Performance degrades for fainter sources ( $H_E > 22$ ) due to low signal-to-noise ratios (SNR) and increased confusion with contaminants.
- Redshift Dependence: Completeness is lower at $7 < z < 8$ due to confusion with late-type dwarfs. At $z \approx 10$ , a dip in completeness is observed due to confusion with early-type galaxies.
- Photometric Redshifts: The code provides accurate photometric redshift estimates ( $z_{out}$ ) with a Pearson correlation of 0.97 against true redshifts for $H_E < 24$ , and a low outlier fraction ( $\sim 2.5\%$ for $\Delta z > 0.1(1+z)$ ).
Threshold Optimization: The study demonstrates that the selection probability threshold ( $\zeta$ ) can be tuned post-processing. A low threshold ( $\zeta = 0.1$ ) is sufficient for bright sources to maximize completeness, while a higher threshold ( $\zeta \approx 0.9$ ) is required for faint sources to maintain purity against early-type galaxy contamination.
Robustness: The code's performance is relatively uniform across the EWS footprint, showing only minor variations due to Galactic coordinates or SNR differences. The inclusion of thick-disk brown dwarfs in simulations had a negligible impact on the F-measure for the EWS depth, validating the thin-disk approximation for this survey.

Significance
The paper claims that Owl-z is a powerful public tool for optimizing the selection of high- $z$ quasars in Euclid and other wide-field surveys. By providing a robust method to distinguish quasars from contaminants, it facilitates the optimization of photometric and spectroscopic confirmation campaigns. The ability to adjust selection thresholds based on magnitude and redshift allows observers to balance the trade-off between finding more candidates (completeness) and ensuring their reliability (purity). The authors emphasize that while the tool is tuned for Euclid, its Bayesian framework and adaptability make it a versatile solution for the upcoming era of large-scale cosmological surveys.