Multi-scale weak lensing detection of galaxy clusters with source redshift tomography

This paper investigates the potential of source redshift tomography to enhance weak lensing galaxy cluster detection using a multi-scale wavelet method, finding that while tomographic combinations do not outperform a single optimal redshift bin due to the accumulation of spurious detections and reduced purity, the approach remains viable despite challenges from large-scale structure contamination and photometric redshift errors.

The Gist

Here is an explanation of the paper, translated into everyday language with some creative analogies.

The Big Picture: Finding Invisible Mountains

Imagine the universe is a vast, dark ocean. Most of the "stuff" in this ocean is Dark Matter, which is invisible to our eyes. However, we know it's there because it has gravity. Just like a heavy rock sitting in a river bends the water flowing around it, a massive cluster of dark matter bends the light coming from galaxies behind it.

Astronomers call this Weak Gravitational Lensing. It's like looking at a funhouse mirror that slightly stretches and distorts the background scenery. By measuring how much the background galaxies are stretched, we can map out where the invisible dark matter "mountains" (clusters) are hiding.

The Problem: The "Crowded Room" Effect

The main challenge in this paper is a problem called dilution.

Imagine you are trying to hear a specific person singing in a crowded room.

• The Singers: These are the distant galaxies behind the dark matter cluster. Their light gets stretched (distorted) by the cluster.
• The Crowd: These are the galaxies in front of the cluster. Their light hasn't passed through the cluster yet, so it isn't stretched.

If you look at everyone in the room (all galaxies, near and far) and try to calculate the average distortion, the people who aren't singing (the foreground crowd) drown out the signal. They "dilute" the effect, making the invisible mountain look smaller and harder to find.

The Proposed Solution: The "Redshift Filter"

The authors wanted to test a clever trick called Source Redshift Tomography.

Think of the universe as a giant, multi-layered cake.

• Layer 1 (Bottom): Galaxies very close to us.
• Layer 2 (Middle): Galaxies in the middle distance.
• Layer 3 (Top): Galaxies very far away.

The "Redshift" is basically a measure of how far away a galaxy is (and how old its light is). The idea is: If we only look at the top layer of the cake (the farthest galaxies), we can ignore the people standing in front of the cluster.

By cutting off the "bottom layers" of the cake (ignoring nearby galaxies), we should get a clearer, stronger signal of the dark matter clusters. The paper tests if using multiple layers (combining different slices of the cake) is even better than just picking one good slice.

The Experiment: Virtual Universes

Since we can't actually cut up the real universe, the authors built Virtual Universes (simulations) on supercomputers. They created three types of test worlds:

1. The Ideal World: Perfect, isolated dark matter blobs with no background noise.
2. The "Realistic" World: Dark matter blobs injected into a universe full of other messy structures (like cosmic filaments).
3. The "Messy" World: A full, complex simulation where everything interacts naturally.

They also simulated two types of "eyes" looking at these worlds:

• Perfect Eyes: Knowing the exact distance to every galaxy.
• Blurry Eyes (Euclid-like): Knowing the distance only roughly, with some errors (like taking a photo with a shaky hand).

The Surprising Results

The team expected that using multiple slices (combining data from different distance layers) would be the ultimate winner, giving them the most complete list of clusters.

But here is the twist:

• The "One Good Slice" Winner: They found that picking just one specific slice of the cake (ignoring galaxies closer than a certain distance) worked almost as well as combining all the slices. It was the sweet spot: far enough to avoid the "crowd," but close enough to still have plenty of galaxies to look at.
• The "Too Many Slices" Trap: When they tried to combine multiple slices to get even more data, something went wrong.
  • The Analogy: Imagine you are trying to find a specific person in a crowd by asking three different groups of people to point them out. Each group has their own "noise" (random mistakes). If you combine the lists from all three groups, you don't just get more correct answers; you also get more false alarms. The "noise" from each group adds up, creating fake clusters that don't exist.

The Conclusion: Quality Over Quantity

The paper concludes that for future telescopes (like the Euclid mission), the best strategy isn't to throw everything at the problem by combining every possible data slice.

Instead, the most efficient method is to be selective.

1. Cut the noise: Ignore the nearby galaxies that dilute the signal.
2. Pick the sweet spot: Use a specific distance range (around $z=0.4$) where the signal is strongest and the noise is manageable.
3. Don't over-combine: Trying to merge too many different distance layers actually creates more "fake" detections, lowering the reliability of the final list.

In short: It's better to have a clean, focused list of 100 real clusters than a messy list of 150 clusters where 50 of them are fake. The "one good slice" approach is the most reliable way to map the invisible mountains of the universe.

Technical Summary

Here is a detailed technical summary of the paper "Multi-scale weak lensing detection of galaxy clusters with source redshift tomography."

1. Problem Statement

Galaxy clusters are vital cosmological probes, but their detection via Weak Lensing (WL) is limited by the dilution effect. In WL, foreground galaxies (those between the observer and the cluster) are not lensed but are included in the source sample, diluting the shear signal and lowering the Signal-to-Noise (S/N) ratio.

• Current Limitations: Traditional WL detection uses the full source population ($z_s \geq 0$), suffering from maximum dilution. While excluding foregrounds ($z_s > z_{s,min}$) improves S/N for high-redshift lenses, it reduces source density, lowering sensitivity for low-redshift lenses.
• Proposed Solution: The paper investigates source redshift tomography using the $z_{s,min}$-cut technique. This involves creating multiple WL mass maps using different minimum source redshift thresholds and combining the resulting peak detections to maximize the total number of detected clusters while mitigating dilution.
• Context: The study builds upon the multi-scale wavelet detection algorithm (Leroy et al., 2023) and evaluates its potential improvement under Euclid-like survey conditions (depth $z_s \approx 3$, high source density).

2. Methodology

Datasets

The authors tested their method on three levels of increasing physical complexity:

1. Idealized NFW Mocks (LSS-free): Spherically symmetric Navarro-Frenk-White (NFW) haloes injected into noise-only fields. This isolates the algorithm's performance without Large Scale Structure (LSS) noise.
2. NFW Injection in N-body Fields: NFW haloes injected into realistic N-body simulation fields (DEMNUni-Cov). This retains LSS fluctuations and projection effects while keeping halo properties controlled.
3. Full N-body Simulations: Detection performed directly on the DEMNUni-Cov simulation fields, where haloes are self-consistently formed (including substructure, non-NFW profiles, and correlated matter).

Note: Both true redshifts and Euclid-like photometric redshifts (with scatter $\sigma_z = 0.05(1+z)$ and 10% catastrophic outliers) were tested.

Detection Algorithm

• Multi-scale Wavelet Transform: The convergence map ($\kappa$) is decomposed using a starlet transform (undecimated wavelet) into scales $W_2, W_3, W_4$ (targeting radii $\approx 1.2, 2.3, 4.7$ arcmin).
• Peak Detection: Peaks are identified in each scale. A scale-dependent threshold is applied using B-mode maps to maintain a constant false detection rate.
• Merging: Detections across scales are hierarchically merged to avoid double-counting the same overdensity.
• Tomographic Binning: Eleven overlapping source redshift bins were created by varying $z_{s,min}$ from 0 to 1.0 (step 0.1).
• Combination Strategies: The authors tested combining detections from 1, 2, 3, or 4 distinct $z_{s,min}$ bins.
• Matching: Detected peaks were matched to simulation haloes using a Signal-to-Noise (S/N) criterion rather than just angular distance, selecting the halo with the highest theoretical S/N if multiple candidates exist within the matching radius.

Performance Metrics

Performance was evaluated using Completeness ($C$) vs. Purity ($P$) curves. The primary metric for ranking methods was the Area Under the Curve (AUC) in the high-purity regime ($P \in [0.85, 0.95]$).

3. Key Contributions

1. Refinement of Multi-scale Detection: Improved the merging of multi-scale peaks and the association of detections with haloes by prioritizing theoretical S/N over simple angular proximity.
2. Systematic Tomographic Analysis: Provided a comprehensive empirical test of the $z_{s,min}$-cut technique across varying levels of simulation realism (from idealized to full N-body) and redshift uncertainties.
3. Identification of the "Spurious Accumulation" Limit: Demonstrated that while tomography reduces dilution, the accumulation of independent spurious detections across overlapping bins creates a dominant limitation that negates the benefits of multi-bin strategies in current survey depths.

4. Key Results

Single Bin vs. Integrated Approach

• Significant Gain: A single source redshift bin with an intermediate cut ($z_{s,min} \approx 0.4$) significantly outperforms the standard integrated approach ($z_{s,min}=0$).
  • In N-body simulations with photometric redshifts, this single-bin strategy increased completeness by 8% at 90% purity compared to the integrated method.
  • This confirms that removing foregrounds to reduce dilution is beneficial, provided a sufficient source density remains.

Multi-Bin Strategies (2, 3, or 4 bins)

• No Net Gain at Fixed Purity: Contrary to expectations, combining multiple bins did not yield further improvements in completeness at a fixed purity level compared to the best single-bin strategy.
• The Mechanism of Failure:
  • While merging bins shifts the detection curve toward higher completeness, it simultaneously shifts it toward lower purity.
  • Because source bins overlap significantly, they share correlated noise, but the unique galaxies in each bin generate independent shape noise.
  • Merging the catalogs (taking the union) causes spurious detections (noise peaks) from different bins to accumulate. This accumulation lowers the overall purity of the combined catalog.
• Threshold vs. Purity Comparison: If one compares methods at a fixed detection threshold (S/N), multi-bin strategies appear to offer large gains (e.g., +23% to +37% completeness). However, this is misleading because the purity drops significantly. When compared at fixed purity, the gains vanish.

Impact of Realism

• LSS Contamination: The presence of LSS (in N-body fields) reduced the potential gains of the tomographic approach compared to idealized NFW mocks.
• Photometric Redshifts: Photo-z errors further reduced the efficiency, though the single-bin strategy remained robust.
• Scale Dependence: The results held true for both multi-scale and single-scale (W3 only) detection modes, indicating the limitation is inherent to the tomographic binning strategy, not the wavelet method.

5. Significance and Conclusions

• Optimal Strategy: For Euclid-like surveys, the optimal strategy is not to combine multiple tomographic bins, but to select a single optimal source redshift bin (specifically $z_{s,min} \approx 0.4$). This balances the reduction of foreground dilution with the retention of sufficient source density.
• Dominant Limitation: The primary barrier to multi-bin tomography is the accumulation of spurious detections. As bins are added, the "noise floor" of false positives rises faster than the true signal gain.
• Future Directions:
  • Deeper Surveys: The technique might become viable for future, deeper surveys (accessing higher $z_s$ and higher densities) where bins can be made narrower, reducing noise correlation.
  • Spurious Flagging: Developing methods to systematically flag and remove spurious detections that appear in multiple bins could recover the potential benefits of multi-bin merging.
  • Blending: The study highlights that a significant fraction of detections (up to 30% in multi-scale cases) are "blended" (associated with multiple haloes along the line of sight), complicating mass measurements and requiring careful handling in cosmological analyses.

Final Verdict: While source redshift tomography via $z_{s,min}$ cuts is a powerful tool for improving weak lensing cluster detection, combining multiple bins does not currently offer advantages over a single, well-chosen bin due to the accumulation of false positives. The most effective approach for near-future surveys is a single-bin optimization.