🔭 astrophysics

Extensive analysis of reconstruction algorithms for DESI 2024 baryon acoustic oscillations

This paper evaluates multigrid, iterative FFT, and iterative FFT particle reconstruction algorithms for DESI's first-year data using realistic mocks across three tracer samples, finding that multigrid and iterative FFT methods agree within 0.4% on BAO scales while recommending against the iterative FFT particle approach and providing optimal settings for future DESI observations.

Original authors: X. Chen, Z. Ding, E. Paillas, S. Nadathur, H. Seo, S. Chen, N. Padmanabhan, M. White, A. de Mattia, P. McDonald, A. J. Ross, A. Variu, A. Carnero Rosell, B. Hadzhiyska, M. M. S Hanif, D. Forero-Sánche

Published 2026-03-02

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: X. Chen, Z. Ding, E. Paillas, S. Nadathur, H. Seo, S. Chen, N. Padmanabhan, M. White, A. de Mattia, P. McDonald, A. J. Ross, A. Variu, A. Carnero Rosell, B. Hadzhiyska, M. M. S Hanif, D. Forero-Sánchez, S. Ahlen, O. Alves, U. Andrade, S. BenZvi, D. Bianchi, D. Brooks, E. Chaussidon, T. Claybaugh, A. de la Macorra, Biprateep Dey, K. Fanning, S. Ferraro, A. Font-Ribera, J. E. Forero-Romero, C. Garcia-Quintero, E. Gaztañaga, S. Gontcho A Gontcho, G. Gutierrez, C. Hahn, K. Honscheid, S. Juneau, R. Kehoe, D. Kirkby, T. Kisner, A. Kremin, M. E. Levi, A. Meisner, J. Mena-Fernández, R. Miquel, J. Moustakas, A. Muñoz-Gutiérrez, F. Nikakhtar, N. Palanque-Delabrouille, W. J. Percival, F. Prada, I. Pérez-Ràfols, M. Rashkovetskyi, G. Rossi, R. Ruggeri, E. Sanchez, C. Saulder, D. Schlegel, M. Schubnell, A. Smith, D. Sprayberry, G. Tarlé, D. Valcin, M. Vargas-Magaña, B. A. Weaver, S. Yuan, R. Zhou

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. Billions of years ago, sound waves rippled through this ocean, leaving behind a specific, repeating pattern in how matter clumps together. Astronomers call this pattern Baryon Acoustic Oscillations (BAO). It's like a "standard ruler" painted across the cosmos. By measuring the size of this ruler at different times in the universe's history, we can figure out how fast the universe is expanding and what dark energy is doing.

However, there's a problem. Over billions of years, gravity has acted like a heavy hand, smearing out that crisp ruler pattern. Galaxies have drifted from their original spots, blurring the lines and making the ruler look fuzzy and hard to measure. This is where Reconstruction comes in.

The Core Problem: The Cosmic Blur

Think of the early universe as a perfectly organized classroom where every student (galaxy) sits in a specific seat. Over time, the bell rings, and the students start running around, bumping into each other, and changing seats. If you take a photo now, you can't easily tell who sat where originally.

Reconstruction is the process of trying to reverse the chaos. It's like a detective trying to figure out where every student started so they can put them back in their original seats, restoring the perfect classroom order. If we can do this, the "ruler" becomes sharp again, and our measurements become incredibly precise.

The Three Detectives (Algorithms)

To solve this "who sat where" mystery, the DESI (Dark Energy Spectroscopic Instrument) team needed a computer algorithm. The paper compares three different "detectives" (algorithms) to see which one does the best job of reversing the cosmic drift:

The Multigrid (MG) Detective: This detective uses a "V-cycle" strategy. Imagine looking at the classroom from a high balcony (coarse grid) to see the big picture, then walking down to the floor (fine grid) to check specific details. They bounce back and forth between these views to quickly zero in on the solution. It's a classic, robust method.
The Iterative FFT (iFFT) Detective: This detective works in "frequency space." Instead of looking at individual students, they look at the waves of movement. They make a guess, check how wrong it is, and then tweak their guess over and over (iteratively) until the waves line up perfectly. It's like tuning a radio until the static disappears.
The Iterative FFT Particle (iFFTP) Detective: This one is similar to the second detective, but instead of just tweaking the math, it physically moves the "students" (galaxies) in the simulation during every step of the process to fix the errors.

The Experiment: Testing the Detectives

The authors ran these three detectives against realistic simulations of the universe (called "mocks") using three different types of cosmic "tracers":

ELG (Emission Line Galaxies): A crowded, busy classroom (high number density).
QSO (Quasars): A sparse, quiet classroom with very few students (low number density, high redshift).
BGS (Bright Galaxies): A classroom where the students are spread out over a huge area, making it hard to see who is looking at whom (wide-angle effects).

They asked: Do these detectives agree on where the students started? Do they restore the ruler pattern equally well?

The Verdict

1. The "Standard" Detectives (MG and iFFT) are Twins.
The Multigrid (MG) and the Iterative FFT (iFFT) detectives agreed almost perfectly. Their results were so similar that the difference was less than 0.4%—essentially indistinguishable noise.

The Winner: The team chose iFFT for the official DESI data analysis. Why? Because it's much faster. It's like choosing a sports car over a reliable truck; both get you to the destination, but the sports car gets there in a fraction of the time.

2. The "Risky" Detective (iFFTP) Failed.
The iFFTP detective had a major flaw. Because it physically moved the "students" around during the process, it sometimes pushed them right off the edge of the map (the survey boundary). This caused errors, especially at the edges of the universe being studied.

The Result: It was unstable and gave different, worse answers than the other two. The authors recommend against using this method until it's fixed.

3. The "Wide Angle" Challenge.
The BGS sample (the wide-angle one) was the hardest test. Because the "line of sight" changes so much across the sky for these galaxies, the math gets tricky. Even here, the two good detectives (MG and iFFT) stayed in sync, though they showed slightly more disagreement than in the other samples. Crucially, this small disagreement didn't mess up the final measurement of the cosmic ruler.

The Big Picture: Why This Matters

This paper is essentially a "quality control" report. Before the DESI team releases their final, groundbreaking results on the expansion of the universe, they needed to prove that their choice of algorithm (iFFT) wasn't introducing hidden errors.

The Takeaway:

We have successfully "un-blurred" the cosmic ruler.
The iFFT algorithm is fast, accurate, and safe to use.
The iFFTP algorithm is too glitchy for now.
Even in the most difficult scenarios (low density, wide angles), the method works.

By confirming that these algorithms work, the paper paves the way for DESI to measure the universe's expansion history with sub-percent precision, helping us solve the mystery of Dark Energy—the invisible force pushing the universe apart. It's like ensuring our telescope is perfectly focused before we try to read the smallest text in the universe's history book.

1. Problem Statement

Baryon Acoustic Oscillations (BAO) serve as a "standard ruler" for measuring the expansion history of the universe. However, the precision of BAO measurements is degraded by gravitational nonlinear evolution (bulk flows and cluster collapse), which smears the BAO peak in the correlation function and suppresses oscillations in the power spectrum.

BAO Reconstruction is a standard procedure used to mitigate this degradation by estimating the displacement field of matter from the observed density field and shifting galaxies and random points back toward their initial positions. While the "standard" reconstruction framework is well-established, applying it to real survey data like the Dark Energy Spectroscopic Instrument (DESI) presents specific challenges:

Changing Lines of Sight (LOS): Unlike idealized periodic boxes, real surveys have varying LOS directions, making the plane-parallel approximation invalid.
Algorithmic Divergence: Several algorithms exist to solve the linearized continuity equation in redshift space under these conditions (Multigrid, Iterative FFT, Iterative FFT Particle), but their consistency and performance on realistic DESI-like data had not been comprehensively compared.
Systematic Uncertainties: As DESI aims for sub-percent precision, it is critical to determine if differences between reconstruction algorithms introduce biases in BAO parameter estimation ( $\alpha_{iso}$ and $\alpha_{AP}$ ).

2. Methodology

The authors conducted a comprehensive analysis using AbacusSummit N-body simulations tailored to DESI requirements (Planck 2018 $\Lambda$ CDM cosmology). They tested three specific reconstruction algorithms:

Multigrid (MG): Solves the linearized continuity equation in configuration space using a V-cycle multigrid method with a damped Jacobi smoother. It handles varying LOS by solving the differential equation directly on a grid.
Iterative FFT (iFFT): Solves the equation in Fourier space. It iteratively estimates the real-space density by removing Redshift Space Distortions (RSDs) via potential updates, without moving galaxy positions during the iteration.
Iterative FFT Particle (iFFTP): A variation of iFFT where RSDs are removed by physically moving galaxy particles at each iteration (previously used in eBOSS).

Test Samples:
The study focused on three representative DESI tracers covering extreme regimes:

Emission Line Galaxies (ELG): High number density ( $0.8 < z < 1.1$ ).
Quasars (QSO): High redshift, low number density, large volume ( $1.6 < z < 2.1$ ).
Bright Galaxy Sample (BGS): Low redshift, testing "wide-angle" effects where LOS varies significantly ( $0.1 < z < 0.4$ ).

Metrics:

Displacement Field: Comparison of vector differences between algorithms.
Power Spectrum: Monopole and quadrupole comparisons (RecSym and RecIso conventions).
Propagator: Cross-correlation with the initial density field to measure linear density recovery.
BAO Fitting: Direct fitting of BAO parameters ( $\alpha_{iso}, \alpha_{AP}$ ) to check for biases.
Computational Efficiency: Runtime comparison.

3. Key Contributions

Algorithm Benchmarking: Provided the first rigorous comparison of MG, iFFT, and iFFTP on realistic DESI mocks with varying LOS and survey boundaries.
Optimal Configuration: Established optimal grid resolutions (4 $h^{-1}$ Mpc) and random particle densities (20 $\times$ galaxy density) for DESI reconstruction.
Convention Sensitivity: Demonstrated that the RecSym convention (where randoms are shifted identically to galaxies, preserving RSDs) is less sensitive to algorithmic differences than the RecIso convention (which attempts to remove RSDs).
Recommendation on iFFTP: Identified a critical flaw in the iFFTP algorithm regarding survey boundaries and recommended against its use without further development.

4. Key Results

A. Algorithm Consistency (MG vs. iFFT)

Agreement: MG and iFFT show excellent agreement. In the RecSym convention, differences in the post-reconstruction power spectrum are < 0.4% for both monopole and quadrupole across all tracers.
Convergence: iFFT converges rapidly; 3 iterations are sufficient for DESI precision.
LOS Dependence: Differences are slightly larger along the line of sight and at survey boundaries, but remain well within the error budget of DESI DR1 and Y5 (Year 5) data.
Tracer Specifics:
- QSO: Shows the best agreement between MG and iFFT (likely due to higher redshift reducing LOS variation).
- BGS: Shows the largest displacement differences (up to ~32% fractional difference in LOS displacement at boundaries) due to wide-angle effects, yet the impact on the power spectrum and BAO fits remains negligible (< 0.4%).

B. Performance of iFFTP

Divergence: iFFTP exhibits significant divergence compared to MG and iFFT.
- ELG: Differences reach ~1% (monopole) and ~10% (quadrupole) with 3 iterations, worsening with more iterations.
- Boundary Issues: The algorithm moves galaxies outside survey boundaries during iterations without matching randoms, causing density estimation errors.
Recommendation: The authors recommend against using iFFTP for BAO analysis in its current form due to instability and lack of convergence.

C. Reconstruction Conventions (RecSym vs. RecIso)

RecSym: Less sensitive to algorithmic errors. Differences between MG and iFFT are minimal.
RecIso: More sensitive to displacement errors, particularly on large scales where RSD removal is attempted. Differences between algorithms are larger (up to ~2% in power spectrum along the LOS), though still small in absolute terms.
Conclusion: RecSym is preferred for robustness against systematics.

D. BAO Constraints

Bias Check: Despite noticeable differences in displacement fields and power spectra (especially for BGS and iFFTP), the resulting BAO parameter fits ( $\alpha_{iso}, \alpha_{AP}$ ) are unbiased and consistent across all algorithms.
Precision: The differences in BAO constraints between MG and iFFT are negligible for DESI DR1 and Y5 precision goals.

E. Computational Efficiency

iFFT is significantly faster: On the Perlmutter cluster, iFFT (3 iterations) took ~150 seconds, while MG took ~1300 seconds for the same task.
Decision: Due to its speed and high agreement with MG, DESI DR1 adopted iFFT as the default reconstruction algorithm.

5. Significance

This paper provides the foundational validation for the reconstruction pipeline used in the DESI 2024 BAO analysis (DR1) and future data releases.

Systematic Budget: It confirms that the choice of reconstruction algorithm (MG vs. iFFT) does not contribute significantly to the systematic error budget for BAO measurements.
Robustness: It demonstrates that BAO constraints are robust against algorithmic variations, even in challenging regimes like wide-angle effects (BGS) or low number densities (QSO).
Future Proofing: By identifying the flaws in iFFTP and establishing the superiority of RecSym, the study guides future analyses (e.g., Euclid, Roman) to avoid similar pitfalls and ensures that reconstruction remains a reliable tool for sub-percent precision cosmology.

In summary, the work validates the use of the iFFT algorithm with the RecSym convention as the optimal, efficient, and robust choice for DESI BAO reconstruction, ensuring that the survey can achieve its goal of measuring dark energy evolution with unprecedented precision.