NERV: Neural-network Enhanced Reconstruction of the UniVerse with Application to Baryon Acoustic Oscillations in the BOSS DR12 Galaxy Sample
Abstract
We present the first application of neural-network-based baryon acoustic oscillation (BAO) reconstruction to real galaxy survey data, restoring the acoustic signature damped by nonlinear structure growth. {\texttt{NERV}} ({\bf N}eural-network {\bf E}nhanced {\bf R}econstruction of the Uni{\bf V}erse) augments standard reconstruction with a convolutional neural network trained on cubic N-body simulations, and explicitly accounts for realistic observational effects including the curved-sky geometry, the redshift-dependent selection function, and finite survey boundaries, by tessellating the survey volume into local patches. We validate the method on the \textsc{MultiDark-Patchy} mock catalogs, recovering unbiased BAO dilation parameters. Applied to the BOSS DR12 galaxy sample, NERV improves the precision of the BAO distance measurements significantly. These results establish neural reconstruction as a practical component of BAO analyses for ongoing surveys such as DESI, with the potential to substantially tighten constraints on the cosmic expansion history and the nature of dark energy.
Related
- Model-independent and reconstruction-free evidence for cosmic acceleration from DESI baryon acoustic oscillations
- Little Red Dot Cosmology: A Matter-Era Baryon Acoustic Oscillations Probe of ΛCDM
- Joint Constraints on the Hubble Constant and Cosmic Baryon Density from Localized Fast Radio Bursts and DESI DR2 Baryon Acoustic Oscillations
- Probing Minihalos in the Late Dark Ages through Velocity Acoustic Oscillations in the 21-cm Power Spectrum
- Baryon Acoustic Oscillations from the C iv Forest with DESI DR2
BibTeX
@misc{zang2026nerv,
title = {{NERV: Neural-network Enhanced Reconstruction of the UniVerse with Application to Baryon Acoustic Oscillations in the BOSS DR12 Galaxy Sample}},
author = {Shi-Hui Zang and Hong-Ming Zhu and Tian-Xiang Mao and Ue-Li Pen},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.15943},
eprint = {2609.15943},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.15943},
}
RIS
TY - UNPB
TI - NERV: Neural-network Enhanced Reconstruction of the UniVerse with Application to Baryon Acoustic Oscillations in the BOSS DR12 Galaxy Sample
AU - Shi-Hui Zang
AU - Hong-Ming Zhu
AU - Tian-Xiang Mao
AU - Ue-Li Pen
PY - 2026
DA - 2026/09/14
PB - arXiv
DO - 10.48550/arxiv.2609.15943
UR - https://arxiv.org/abs/2609.15943
AB - We present the first application of neural-network-based baryon acoustic oscillation (BAO) reconstruction to real galaxy survey data, restoring the acoustic signature damped by nonlinear structure growth. {\texttt{NERV}} ({\bf N}eural-network {\bf E}nhanced {\bf R}econstruction of the Uni{\bf V}erse) augments standard reconstruction with a convolutional neural network trained on cubic N-body simulations, and explicitly accounts for realistic observational effects including the curved-sky geometry, the redshift-dependent selection function, and finite survey boundaries, by tessellating the survey volume into local patches. We validate the method on the \textsc{MultiDark-Patchy} mock catalogs, recovering unbiased BAO dilation parameters. Applied to the BOSS DR12 galaxy sample, NERV improves the precision of the BAO distance measurements significantly. These results establish neural reconstruction as a practical component of BAO analyses for ongoing surveys such as DESI, with the potential to substantially tighten constraints on the cosmic expansion history and the nature of dark energy.
ER -