We present a comprehensive dataset developed for machine learning-based reconstruction of the dark matter field from mock observations of the DESI Bright Galaxy Survey (BGS). The dataset is derived from the high-resolution Jiutian N-body simulation with Planck 2018 cosmology. It includes: 
(1)  real-space dark matter density fields
(2) real-space dark matter velocity fields (three components)
(3) redshift-space galaxy density fields

Designed to reproduce the observational characteristics of the DESI BGS, the dataset incorporates key features including: 
(i) geometric selection effects to mimic the DESI survey footprint.
(ii) flux-limited sampling with a z-band apparent magnitude cut of m_z < 19.0. 
(iii) redshift-space distortions accounting for both the large-scale Kaiser effect and small-scale Finger-of-God effect.
(iv)DESI BGS sky mask covering approximately 9625 deg² in the Northern Galactic Cap (NGC). 
(v) redshift range: 0.1 < z < 0.4. 

To reproduce a realistic lightcone geometry, simulation boxes were randomly rotated and shifted prior to stacking, and a fixed rotation of 58 degrees in right ascension was applied to the coordinate system to maximize survey volume coverage within a periodic box of side length 1564 h⁻¹ Mpc. All fields are interpolated onto a 512³ Cartesian mesh using the Cloud-in-Cell (CIC) scheme to generate consistent gridded representations of both galaxy and dark matter distributions. 

The dataset includes 10 mock realizations produced through independent rotations and translations, enabling robust training, validation, and uncertainty estimation. This dataset forms the foundation of the DarkAI reconstruction framework, which aims to recover the three-dimensional dark matter density, velocity, and tidal fields directly from redshift-space galaxy distributions. It serves as a valuable resource for benchmarking cosmological inference pipelines and for evaluating field-level deep learning models under observationally realistic conditions. For further details, please refer to the associated publication: “DarkAI: Reconstructing the density, velocity and tidal field of dark matter from DESI-like bright galaxy sample” (arXiv:2501.12621).

This Version 2 dataset differs from Version 1, which contained only a single galaxy field sample.
Please note that the current release does not include the effects of fiber assignment, which will be incorporated in a future public release.