Forecast for the detectability of patchy hydrogen reionization in WEAVE-QSO measurements of the Lyman- alpha forest power spectrum at redshift z at least 4

The paper presents the first detailed forecasts for the detectability of patchy hydrogen reionization in the one-dimensional Lyman-α forest power spectrum to be measured by the WEAVE-QSO survey. Using the Sherwood-relics reionization simulations and a WEAVE-QSO survey configuration, the authors generate mock spectra in four redshift bins, z = 4.0, 4.2, 4.4, and 4.6, in which relic ionization and temperature fluctuations from patchy hydrogen reionization enhance the Lyalpha forest power spectrum on large scales (i.e., at wavenumber k ∼ 10−3 s km−1). Their Lyalpha forest pipeline forecasts the power spectrum covariance by considering sample size, spectral resolution, noise subtraction, continuum placement, metal contamination, and damping wings from high-column density absorbers. Applying this covariance forecast within a Bayesian parameter inference framework, they find that the signature of patchy hydrogen reionization should be detectable at a significance of ≃ 4.5sigma. The forthcoming WEAVE-QSO 1D power spectrum measurements should therefore be able to directly detect and characterize the large-scale relic imprint of patchy hydrogen reionization in the Lyalpha forest power spectrum at z ≥ 4.

The mock spectra are based on the Sherwood-Relics simulation suite and processed with the WEAVEify pipeline to include the main observational ingredients relevant for WEAVE-QSO spectra. A realistic high-redshift QSO population is constructed using luminosity-function-based Monte Carlo sampling, and an r-band magnitude cut of m r < 21.9 is adopted. For a 10 000 deg2 survey area, this selection yields expected QSO numbers of 2942, 1822, 1385, and 1155 in the four redshift bins considered here.

The covariance matrix estimate includes statistical errors and systematic contributions from spectral resolution, noise subtraction, continuum misplacement, uncorrelated metal contamination, and the incomplete identification of high column density absorbers with NHI > 1019 cm−2. Within the adopted maximum wavenumber cut, k < 0.05 s km−1, the estimated relative uncertainties remain below 10 per cent on the largest scales in all redshift bins. Averaged over all bins and scales below this cut, the total uncertainty is 5.2 per cent. Over the range −2.6 ≲ log10 (k/s km−1) ≲ −1.7, the uncertainties are dominated by the statistical contribution and the average uncertainty is 3.1 per cent.

Using this forecast covariance, a mock PLyalpha data vector containing the fiducial patchy reionization signal is constructed and tested against an IGM heated and ionised by a spatially uniform UV background. Allowing one independent patchy reionization amplitude parameter, Ap, in each redshift bin improves the fit from chi2 /dof = 65.28/32 to chi2 /dof = 36.07/28, corresponding to Δchi2 = 29.21 for four additional patchy amplitude parameters and a ≃ 4.5sigma preference for additional power at large scales from patchy reionization. This result remains robust to changes in the prior adopted for IGM thermal parameters and is driven almost entirely by the lowest-k bin from each redshift bin.

The paper concludes that a WEAVE-QSO measurement of the high-redshift Lyalpha forest power spectrum should be capable of detecting and characterizing the large-scale relic imprint of patchy reionization, provided that the observational covariance is controlled at the level forecast here. The framework developed provides a route for assessing the sensitivity of current and future massive spectroscopic surveys to the timing of reionization through large-scale PLyalpha measurements.

Improvement 1: Physics-Aware Surrogate Modeling for Reionization Forecasts

• What to improve: Replace generic interpolation or black-box ML models with a hybrid AI system that embeds the Sherwood-Relics simulation physics (patchy reionization amplitude, IGM thermal history, UV background fluctuations) as differentiable priors.

• What the improved AI can do: Given a survey configuration (area, magnitude limit, redshift bins), it can instantly predict the 1D Lyα forest power spectrum and its covariance without running expensive hydrodynamical simulations, enabling rapid exploration of survey designs (e.g., optimal redshift binning, exposure time, or target selection) for detecting patchy reionization.

Improvement 2: Uncertainty-Aware Covariance Emulation

• What to improve: Train a neural network to emulate the full covariance matrix (statistical + systematic contributions) as a function of survey parameters (QSO number, spectral resolution, noise, continuum misplacement, metal contamination, damped Lyα systems).

• What the improved AI can do: For any proposed survey (e.g., DESI-II, PFS, or future space missions), it can output a realistic covariance matrix in milliseconds, allowing Bayesian model comparison between uniform reionization and patchy reionization scenarios—without manual pipeline re-runs. This enables real-time sensitivity forecasts and adaptive survey planning.

Improvement 3: Anomaly Detection for Relic Signals in Noisy Spectra

• What to improve: Use a transformer-based or convolutional network trained on mock WEAVE-QSO spectra (with injected patchy reionization signals) to detect the large-scale power excess (k 1e-3 s/km) even when individual spectra have low SNR or continuum errors.

• What the improved AI can do: Automatically flag QSO spectra that carry the strongest relic reionization signal, prioritize them for deeper follow-up, and provide a per-spectrum likelihood of containing patchy reionization imprints—reducing the need for full sample stacking and improving detection significance beyond the 4.5σ forecast.

Improvement 4: Active Learning for Optimal Redshift Bin Selection

• What to improve: Implement an active learning loop that uses the forecast covariance and mock data to iteratively propose which redshift bins or k-ranges to observe next, maximizing the expected Δχ2 for patchy reionization detection.

• What the improved AI can do: Given limited observing time, it can dynamically allocate QSO targets across z = 4.0–4.6 to minimize total uncertainty, potentially boosting the detection significance from 4.5σ to >5σ by focusing on the lowest-k bins where the signal is strongest, while accounting for real-time data quality.

Improvement 5: Robust Bayesian Inference with Neural Posterior Estimation

• What to improve: Replace MCMC sampling in the Bayesian parameter inference (for Ap and IGM thermal parameters) with a neural posterior estimator (e.g., normalizing flows) trained on the mock spectra and covariance.

• What the improved AI can do: For a real WEAVE-QSO dataset, it can instantly produce full posterior distributions for the patchy reionization amplitude per redshift bin, including degeneracies with thermal parameters, and provide a calibrated detection significance—without the computational cost of thousands of MCMC chains. This enables near-real-time analysis during survey operations.

Improvement 6: Transfer Learning to Other Surveys and Cosmological Probes

• What to improve: Pre-train a model on the Sherwood-Relics simulations and WEAVE-QSO pipeline, then fine-tune it on other large-scale structure probes (e.g., 21-cm power spectrum, CMB lensing) that also carry reionization imprints.

• What the improved AI can do: Jointly constrain the timing and topology of reionization by combining Lyα forest data with future 21-cm observations (e.g., SKA) or CMB-S4, providing a multi-probe consistency check and breaking degeneracies that single-probe analyses cannot—ultimately yielding a more precise measurement of the reionization history.

We present the first detailed forecasts for the detectability of patchy hydrogen reionization in the one-dimensional Ly alpha forest power spectrum to be measured by the WEAVE-QSO survey. Using the Sherwood-relics reionization simulations and a WEAVE-QSO survey configuration, we generate mock spectra in four redshift bins, z=4.0,4.2,4.4, and 4.6, in which relic ionization and temperature fluctuations from patchy hydrogen reionization enhance the Ly alpha forest power spectrum on large scales (i.e., at wavenumber k about 10-3, s,km-1). Our Ly alpha forest pipeline forecasts the power spectrum covariance by considering sample size, spectral resolution, noise subtraction, continuum placement, metal contamination, and damping wings from high-column density absorbers. Applying our covariance forecast within a Bayesian parameter inference framework, we find that the signature of patchy hydrogen reionization should be detectable at a significance of 4.5 sigma. The forthcoming WEAVE-QSO 1D power spectrum measurements should therefore be able to directly detect and characterize the large-scale relic imprint of patchy hydrogen reionization in the Ly alpha forest power spectrum at z at least 4.