Alcock-Paczynski Blinding Scheme for the Ly-$\alpha$ Forest Analysis

Technical Summary: Alcock-Paczynski Blinding Scheme for Lyα Forest Analysis

Problem Statement
As cosmological experiments like the Dark Energy Spectroscopic Instrument (DESI) enter an era of high precision, the risk of confirmation bias—where analysts unconsciously favor results aligning with prior expectations—becomes a critical concern. While blinding strategies are established in supernovae and weak lensing analyses, and widely used in galaxy surveys, their application to the Lyman-α (Lyα) forest presents unique challenges. The Lyα forest analysis involves extracting a flux fluctuation field (delta field) from quasar spectra, a process complicated by the entanglement of signals from the quasar continuum, intergalactic medium (IGM) absorption, metal lines, and Lyman-β contamination. Existing blinding methods often operate at the parameter inference stage or on correlation functions, which may not protect against bias introduced during pipeline development or when using alternative analysis codes. Furthermore, a naive application of cosmological transformations to the Lyα forest is difficult because different absorption features (e.g., Lyα vs. metal lines) originate from distinct redshifts but appear at the same observed wavelength, making a consistent shift non-trivial.

Methodology
The authors propose a catalog-level blinding scheme based on a modified Alcock-Paczynski (AP) test. The core concept involves intentionally modifying the inferred expansion history of the Universe to hide the true cosmological parameters. The method is implemented as follows:

1. Theoretical Framework: The scheme mimics an alternative cosmological model (specifically a flat $\Lambda$CDM model with a shifted matter density $\Omega_m$ or Hubble constant $H_0$) by altering the comoving distance $R(z)$ between objects. This is achieved by mapping the redshifts of the reference cosmology ($\Omega_{ref}$) to a blinded cosmology ($\Omega_{bld}$) such that $R(z'|\Omega_{ref}) = R(z|\Omega_{bld})$.
2. Implementation in Lyα Forest: Unlike galaxy surveys where discrete redshifts are shifted, the Lyα forest requires shifting a continuous field. The authors implement the blinding after the extraction of the delta field ($\delta(\lambda)$) but before the calculation of correlation functions.
   • Quasar Redshifts: The redshifts of the quasars (QSOs) are shifted using the derived function $f(z)$.
   • Wavelength Shift: The delta field in wavelength space is shifted according to $\lambda' = \lambda + g(\lambda)$, where $g(\lambda)$ is derived from the redshift shift function. This assumes all absorption arises from the Lyα emission line ($\lambda_{Ly\alpha} \approx 1216$ Å) to define the mapping, acknowledging that this assumption introduces limitations for metal lines.
   • Rebinning: Due to the non-linear nature of the wavelength shift, the forest range changes. The blinded delta field is resampled onto a linear grid (in comoving distance or logarithmic wavelength) to prevent unblinding via grid spacing comparisons.
3. Pipeline Integration: The scheme is integrated into the PICCA code, the standard tool for DESI Lyα analyses. The blinding is applied immediately after the iterative delta extraction process (which fits the continuum) and before the computation of 1D and 3D correlation functions. This ensures the full analysis pipeline, including covariance estimation and systematics treatment, is developed using blinded data.

Key Contributions

• Catalog-Level Blinding: The paper introduces a blinding strategy that operates at the catalog level (on the delta field and QSO redshifts) rather than at the final parameter level, offering stronger protection against confirmation bias during pipeline development.
• Handling Continuous Fields: The authors adapt the discrete AP blinding used for galaxies to the continuous absorption field of the Lyα forest, providing a specific algorithm for wavelength-space shifting.
• Validation Across Complexity: The method is validated on a hierarchy of datasets:
  • Idealized, low-noise synthetic catalogs (Catalog A: Lyα only; Catalog B: Lyα + Lyβ + metals).
  • Realistic synthetic catalogs (Catalog C: DESI Year 1 expectations with noise).
  • Real data from DESI Data Release 1 (DR1).
• Cross-Correlation Consistency: The study demonstrates that the blinding can be consistently applied to both auto-correlations (Lyα×Lyα) and cross-correlations (Lyα×QSO), preserving the internal consistency required for combined analyses.

Results

• BAO Peak Shift: The method successfully shifts the Baryon Acoustic Oscillation (BAO) peak position in the 3D correlation functions to the values expected for the blinded cosmology (e.g., a $\sim 5\%$ change in matter content).
• Posterior Integrity: For low-noise mocks, the shape of the posterior distributions for the BAO parameters ($\alpha_{\parallel}, \alpha_{\perp}$) remains largely intact, with only a shifted central value consistent with the blinded cosmology within $1\sigma$. For realistic DESI DR1 data, the posterior width changes by less than 5%, indicating a mild dependency on the blinding (likely due to rebinning) but no significant degradation in statistical power.
• Metal Contamination Limitation: The authors identify a specific limitation: in the 1D correlation function, metal absorption features (which do not follow the Lyα-based cosmological remapping) appear as shifted structures. This could theoretically allow an analyst to deduce the direction of the blinding. However, the paper notes this effect is negligible in the 3D correlation functions used for primary cosmological constraints and can be mitigated by secondary blinding strategies for the 1D case.
• Computational Efficiency: The blinding prescription adds approximately 5% to the wall-clock time of the pipeline, scaling linearly with the number of quasars, making it computationally feasible for large datasets.

Significance and Claims
The paper claims that this catalog-level AP blinding scheme is a viable and robust method for cosmological inference using the Lyman-α forest. Its primary significance lies in enabling the development and validation of the entire analysis pipeline on blinded data, thereby minimizing confirmation bias. The authors emphasize that the method is particularly effective when used in conjunction with cross-analyses of other tracers (like QSOs) using the same blinding strategy. While acknowledging the limitation regarding metal lines in 1D correlations, the study concludes that the method effectively hides the underlying cosmology in 3D correlation analyses without providing obvious clues to the analyst, making it suitable for current and future DESI data releases.