🔭 astrophysics

Influence of photometric galaxies redshift distribution in BAO estimation

This study utilizes the DES Y3 LRG catalog to demonstrate that incorporating realistic photometric redshift probability distributions into cosmological models significantly impacts Baryon Acoustic Oscillation (BAO) constraints, revealing that the ANNz2 estimator with Gaussian-like samples best recovers the fiducial Planck 18 $\Lambda$ CDM model while DNF excels for dark energy equation of state parameters, ultimately recommending the use of multiple photo-z algorithms to better understand systematic effects in future surveys.

Original authors: Paula S. Ferreira, Ribamar R. R. Reis

Published 2026-04-06

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Paula S. Ferreira, Ribamar R. R. Reis

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 Picture: Mapping the Universe's "Fingerprint"

Imagine the universe as a giant, expanding loaf of raisin bread. A long time ago, sound waves rippled through the hot soup of particles that made up the early universe. When the universe cooled down, those sound waves stopped, leaving a specific "fingerprint" or pattern in the way galaxies are arranged today.

Scientists call this pattern Baryon Acoustic Oscillations (BAO). It's like a cosmic ruler. If we can measure the distance between galaxies that match this specific pattern, we can figure out how fast the universe is expanding and what it's made of (like dark energy).

The Problem: The "Blurry" Camera

To measure this cosmic ruler, astronomers need to know exactly how far away every galaxy is.

Spectroscopic surveys (the gold standard) are like taking a high-resolution photo of a galaxy's light spectrum. It's slow and expensive, but the distance is precise.
Photometric surveys (the focus of this paper) are like taking a photo through different colored filters (red, blue, green). It's fast and can capture millions of galaxies at once, but the distance estimate is "blurry."

Think of it this way: If you look at a streetlight through a foggy window, you know it's a light, but you aren't 100% sure if it's 100 meters away or 150 meters away. You have a Probability Distribution Function (PDF). Instead of saying "It is at 120 meters," the camera says, "It's probably around 120 meters, but it could be anywhere between 100 and 140."

The Experiment: Testing the "Blur"

The authors of this paper wanted to see how this "blur" affects our ability to find the cosmic ruler (BAO). They used data from the Dark Energy Survey (DES), which has millions of galaxies.

They tested four different "algorithms" (computer programs) that try to guess the distance of these galaxies based on the blurry photos:

ANNz2 (A neural network, like a brain learning from examples).
BPZ (A template matcher, like matching a fingerprint to a database).
ENF & DNF (Nearest neighbor methods, like saying "This galaxy looks just like that one I already know the distance to").

They also tried two different ways of cleaning up the data:

The "Gaussian" Cut: Keeping only the galaxies where the distance guess looks like a smooth, perfect bell curve (very predictable).
The "Small Peaks" Cut: Keeping galaxies where the guess doesn't have weird, confusing multiple peaks (like a mountain range with two tops instead of one).

The Findings: Who Got the Best Focus?

The researchers ran simulations and real data to see which algorithm and which data cut could find the "cosmic ruler" most accurately.

1. The "Smooth" vs. The "Noisy"
Imagine trying to hear a song in a quiet room versus a noisy room.

When they used the ANNz2 algorithm and kept only the "smooth" (Gaussian) guesses, the result was very close to the true answer. It was like listening to the song in a quiet room.
When they used other algorithms (like DNF) or kept all the "noisy" data, the ruler shifted. The "fingerprint" of the universe appeared in the wrong place.

2. The Trade-off
There is a catch. To get that "smooth" data, they had to throw away a lot of galaxies.

ANNz2 was the winner because it could find the ruler even with a smaller number of galaxies, provided the data was "smooth."
DNF is a powerful tool, but it needs a huge amount of data to work well. If you cut the data down to make it "smooth," DNF loses its signal and the ruler disappears.

3. The "Double-Blind" Test
The authors also tested a new mathematical method to combine the data. They found that if you don't account for the "blur" (the PDF shape) in your math, you get the wrong answer. It's like trying to measure a room with a ruler that stretches and shrinks depending on the temperature; you have to correct for the temperature, or your measurement is wrong.

The Conclusion: A Team Effort

The main takeaway is that not all distance guesses are created equal.

ANNz2 is currently the best "photographer" for this specific job because it handles the "blur" better than the others.
However, relying on just one algorithm is risky. It's like having only one map to navigate a new city; if that map has a mistake, you get lost.

The Recommendation:
Future space surveys should use multiple algorithms at the same time. By comparing the results from different "guessers," scientists can spot errors, understand the "systematic noise" (the foggy window), and get a much clearer picture of the universe's expansion.

In a Nutshell

This paper is a quality control check for the tools we use to map the universe. It tells us that while taking "blurry" photos of millions of galaxies is great for speed, we need to be very careful about how we interpret that blur. If we choose the right computer program (ANNz2) and the right way to clean the data, we can still find the universe's hidden ruler and understand how our cosmos is evolving.

1. Problem Statement

Baryon Acoustic Oscillations (BAO) serve as a standard ruler for measuring cosmic expansion. While spectroscopic surveys provide precise redshifts, photometric surveys (like the Dark Energy Survey, DES) offer larger sky coverage and higher object counts, which are crucial for large-scale structure studies. However, photometric redshifts ( $z_p$ ) suffer from significant uncertainties and non-Gaussian errors.

The core problem addressed is how the shape of the redshift Probability Distribution Function (PDF) for individual galaxies affects the estimation of the BAO feature in the transverse correlation function ( $\xi(s_\perp)$ ). Standard analyses often assume Gaussian redshift errors, but real photo-z estimators produce complex, sometimes multi-modal PDFs. The authors investigate whether different photo-z estimation algorithms and specific sample selection cuts (based on PDF shape) introduce systematic biases in the inferred BAO position and cosmological parameters.

2. Methodology

The study utilizes the Dark Energy Survey Year 3 (DES Y3) Luminous Red Galaxy (LRG) catalog. The authors compare four distinct photo-z estimation algorithms:

ANNz2: Artificial Neural Networks (training-based).
BPZ: Bayesian Photometric Redshift (template fitting).
ENF: Euclidean Neighbourhood Fitting.
DNF: Directional Neighbourhood Fitting.

Data Processing & Selection:

Sample Selection: The authors applied two specific selection criteria to the catalogs ("Pz Cats") to isolate different PDF shapes:
1. Gaussian Selection: Selecting galaxies where the PDF is nearly Gaussian (using statistical moments).
2. Small Peaks (Sp) Selection: Selecting galaxies with minimal noise, defined as having secondary peaks no larger than 30% of the main peak.
3. Full Sample: No selection cuts applied.
Mock Validation: CoLoRe simulations based on Planck 18 $\Lambda$ CDM cosmology were used to validate the pipeline and test bias functions.
Correlation Function Modeling:
- Parametric Approach: The authors fitted a polynomial model with a Gaussian peak to the projected correlation function $\xi(s_\perp)$ to extract the shift parameter $\alpha$ .
- Cross-Bin Approach: Adapting the method from [24], they computed the transverse correlation function by crossing photo-z bins. They utilized the conditional distribution $f(z|z_p)$ (derived from matched spectroscopic samples) as a kernel window function within the CAMB code to generate the theoretical power spectrum and correlation function.
- Line-of-Sight Treatment: They tested the inclusion vs. exclusion of line-of-sight terms ( $\mu$ ) in the angular correlation function, finding that including all angles (transverse to the line of sight) preserves the BAO signal better than strict exclusion.

Statistical Analysis:

Bayes Factor: The authors calculated the Bayes Factor ( $B_{AA'}$ ) to compare the likelihood of different models (varying $H_0$ , $\Omega_b h^2$ , $w_0$ , and galaxy bias) against a fiducial ideal model and the DNF full-sample result.
Error Estimation: Asymmetric error bars were derived using bootstrap methods and mode-based confidence intervals rather than assuming Gaussian posteriors.

3. Key Contributions

Realistic PDF Integration: The study moves beyond the assumption of Gaussian redshift errors by incorporating the actual, complex PDFs generated by different estimators directly into the cosmological model.
Algorithm Benchmarking: A comprehensive comparison of ANNz2, BPZ, ENF, and DNF under identical statistical conditions and selection cuts, specifically focusing on their ability to recover the BAO signal.
Selection Cut Impact: Demonstration that selecting galaxies based on PDF shape (Gaussian vs. Low Noise) significantly alters the effective redshift ( $z_{eff}$ ) and the resulting BAO constraints.
Systematic Bias Quantification: Identification of how "spurious" PDFs (multi-modal or noisy) can shift the inferred BAO position, potentially biasing cosmological constraints even if the effective redshifts appear compatible.

4. Key Results

ANNz2 Superiority: The ANNz2 estimator consistently outperformed the others.
- The ANNz2 Gaussian sample selection yielded a shift parameter $\alpha$ closest to the fiducial Planck 18 $\Lambda$ CDM model and the DES Y3 spectroscopic results.
- ANNz2 remained robust across all three selection criteria (Full, Gaussian, Small Peaks), whereas other estimators failed to detect the BAO signal or showed significant shifts when sample sizes were reduced.
DNF Performance: DNF performed well only when the full sample (maximum statistics) was used. When statistical cuts were applied (Gaussian or Small Peaks), the BAO feature disappeared or became unreliable, indicating DNF relies heavily on large training sets to compensate for individual PDF imperfections.
PDF Shape Influence: The shape of the galaxy redshift PDF directly shifts the BAO feature position.
- ENF and BPZ showed significant deviations or failure to detect the signal in selected subsamples.
- BPZ Gaussian samples showed weak evidence against the fiducial model when varying the dark energy equation of state ( $w_0$ ).
Bin Width Sensitivity: The study confirmed that bin width ( $\Delta z$ ) and bin separation critically affect BAO detection. Thinner bins at high redshifts lead to a loss of correlation between shells, washing out the BAO signal.
Bayes Factor Evidence:
- ANNz2 Gaussian provided the strongest evidence (Definite to Strong) against the ideal Gaussian model and the DNF model across various parameter variations ( $H_0$ , $\Omega_b$ , $w_0$ , bias).
- DNF showed "Weak" or "Definite" evidence only when the full sample was used; performance degraded significantly with sample cuts.

5. Significance and Recommendations

Systematic Control: The paper demonstrates that ignoring the specific shape of photo-z PDFs can lead to biased cosmological constraints. The "best" estimator depends on the sample size and the specific selection cuts applied.
Future Survey Strategy: The authors recommend that future photometric survey collaborations (e.g., LSST, Euclid, Roman) incorporate multiple photo-z estimation algorithms into their cosmological inference pipelines. Relying on a single estimator may mask systematic effects.
Algorithm Selection: For the DES Y3 data and similar training sets, ANNz2 is identified as the optimal estimator for BAO analysis due to its robustness against PDF shape variations and statistical loss.
Methodological Validation: The study validates the use of cross-bin correlation functions with realistic $f(z|z_p)$ kernels as a reliable method for extracting transverse BAO signals from photometric data, provided the binning strategy is optimized for the redshift range.

In conclusion, this work highlights that the fidelity of BAO measurements in photometric surveys is inextricably linked to the quality and shape of the redshift PDFs, and that algorithm selection must be tailored to the specific statistical power of the dataset to avoid systematic biases.