Jiutian-1G Cosmological Simulation

• Jiutian-1G is a gravity-only cosmological N-body simulation that evolves 6144³ dark-matter particles in a 1000 h⁻¹ Mpc box to study large-scale structure.
• It employs sophisticated halo-finding and merger-tree techniques (SubFind and HBT+) to construct detailed mock catalogs for galaxy formation and weak-lensing analyses.
• The simulation uses Planck-2018 ΛCDM parameters and meticulous numerical methods to balance fine mass resolution with extensive volume for extragalactic survey forecasts.

Jiutian-1G is a gravity-only cosmological $N$-body simulation in the Jiutian suite developed for the China Space Station Telescope (CSST) extragalactic surveys. It is one of the two primary large-volume, high-resolution runs, evolving $6144^3$ dark-matter particles in a periodic box of side $1000\,h^{-1}\,\mathrm{Mpc}$, and it is used to construct halo catalogs, subhalo merger trees, semi-analytical galaxy catalogs, weak-lensing light cones, and survey forecasts for CSST (Han et al., 27 Mar 2025, Tan et al., 5 Nov 2025, Xiong et al., 2024). Within the broader Jiutian program, Jiutian-1G occupies the regime where large-scale clustering statistics and comparatively fine halo mass resolution are simultaneously required, especially for galaxy-population studies and mock-catalog construction.

1. Position within the Jiutian simulation program

The Jiutian suite is described as a hybrid simulation program for CSST extragalactic surveys with four complementary modules: primary runs with the fiducial concordance cosmology, emulator runs exploring parameter uncertainties around the fiducial cosmology, reconstruction runs intended to recover the observed Universe position by position, and extension runs employing cosmologies beyond the standard model (Han et al., 27 Mar 2025). Jiutian-1G belongs to the primary-run module.

Within that primary module, Jiutian-1G and Jiutian-2G both employ $6144^3$ particles but differ in volume: a $1\,\mathrm{Gpc}/h$ box for Jiutian-1G and a $2\,\mathrm{Gpc}/h$ box for Jiutian-2G. The numeral in the name denotes the box length in $\mathrm{Gpc}/h$. Jiutian-1G trades larger volume for finer mass resolution, which the project documentation identifies as advantageous for low-mass halo statistics and high-resolution galaxy formation studies at $z\lesssim2.5$, while Jiutian-2G is used to sample cosmic variance and rare massive structures to $z=5$ (Tan et al., 5 Nov 2025).

This positioning matters because the Jiutian project uses primary runs as the substrate for multiple downstream pipelines. These include two independent subhalo and merger-tree constructions, semi-analytical galaxy models, subhalo abundance matching, and lensing-oriented products, so Jiutian-1G is not only a standalone $N$-body experiment but also a central numerical input to a multi-product survey simulation framework (Han et al., 27 Mar 2025).

2. Numerical specifications and cosmological setup

The primary-run descriptions of Jiutian-1G agree on its basic numerical design and on its role as a large-volume, high-particle-count dark-matter simulation (Han et al., 27 Mar 2025, Tan et al., 5 Nov 2025).

Quantity	Value	Note
Box side length	$6144^3$0	periodic comoving cube
Particle number	$6144^3$1	dark-matter particles
Particle mass	$6144^3$2	primary-run value
Snapshots	$6144^3$3	from $6144^3$4 to $6144^3$5
Force softening	$6144^3$6	primary-run specification

In the primary-run documentation, the particle mass is written as

$6144^3$7

which makes explicit that the mass resolution is set by the mean density, the cubic volume, and the particle count (Tan et al., 5 Nov 2025).

The main cosmological parameters are given as Planck 2018 values: $6144^3$8, $6144^3$9, $1000\,h^{-1}\,\mathrm{Mpc}$0, $1000\,h^{-1}\,\mathrm{Mpc}$1, and $1000\,h^{-1}\,\mathrm{Mpc}$2; where specified, $1000\,h^{-1}\,\mathrm{Mpc}$3 is also listed (Han et al., 27 Mar 2025, Xiong et al., 2024). A separate subhalo-convergence summary reports slightly different Planck 2018 numbers, namely $1000\,h^{-1}\,\mathrm{Mpc}$4, $1000\,h^{-1}\,\mathrm{Mpc}$5, $1000\,h^{-1}\,\mathrm{Mpc}$6, $1000\,h^{-1}\,\mathrm{Mpc}$7, and $1000\,h^{-1}\,\mathrm{Mpc}$8 (Xu, 11 Aug 2025). This suggests that Jiutian-1G is consistently framed as a Planck-2018-based $1000\,h^{-1}\,\mathrm{Mpc}$9CDM run, while some secondary summaries present rounded or alternate parameter values.

The weak-lensing forecast paper also quotes the mean interparticle spacing as $6144^3$0 and states that adopting a Plummer-equivalent softening of $6144^3$1 yields the force resolution (Xiong et al., 2024). This differs from the $6144^3$2 force softening given in the primary-run references (Han et al., 27 Mar 2025, Tan et al., 5 Nov 2025). The discrepancy is part of the published record and is best read as a difference in how the force scale is summarized across applications.

3. Initial conditions, solver, and time integration

The run begins at an initial redshift $6144^3$3 (Han et al., 27 Mar 2025, Tan et al., 5 Nov 2025, Xiong et al., 2024). The linear matter power spectrum at $6144^3$4 is computed with CAMB, and the primary-run references describe the initial particle displacements as generated with the Zel’dovich approximation, including one summary that specifies a $6144^3$5 grid and another that refers to a glass-like pre-initial particle load (Han et al., 27 Mar 2025, Tan et al., 5 Nov 2025). By contrast, the CSST weak-lensing study states that second-order Lagrangian perturbation theory is used to set the particle displacements (Xiong et al., 2024). A plausible implication is that the literature around Jiutian-1G contains multiple implementation summaries for the initial-displacement stage.

The gravitational evolution is attributed to LGadget-3 or L-Gadget3, described as a private Gadget-3 branch or as a memory- and communication-optimized variant of GADGET-2/3 (Han et al., 27 Mar 2025, Tan et al., 5 Nov 2025). The weak-lensing study gives a more explicit algorithmic description: the code uses a hybrid TreePM scheme, with long-range forces solved on a regular particle-mesh grid, typically of size $6144^3$6 cells, via Fourier methods, and short-range interactions handled with a Barnes-Hut oct-tree (Xiong et al., 2024). The subhalo-convergence study characterizes the evolution more generically as a Gadget-3/4 tree-PM code with a standard adaptive, hierarchical time-step scheme and fixed comoving softening length (Xu, 11 Aug 2025).

Time integration is described in the weak-lensing application as symplectic leapfrog, in the “kick-drift-kick” form, for the particle equations of motion

$6144^3$7

with gravitational softening to suppress two-body scattering (Xiong et al., 2024). The same study states that Jiutian-1G outputs 128 snapshots down to $6144^3$8 with typical time spacing $6144^3$9, which is used for light-cone slicing and non-linear growth tracking (Xiong et al., 2024).

4. Halo identification, subhalo tracking, and merger trees

Halo and subhalo reconstruction in Jiutian-1G is central to its use in mock-catalog production. Friends-of-Friends halo catalogs are built at each snapshot using linking length $1\,\mathrm{Gpc}/h$0 times the mean particle separation (Han et al., 27 Mar 2025, Tan et al., 5 Nov 2025). The Jiutian primary-run framework explicitly uses two independent subhalo and merger-tree pipelines: SubFind with LHaloTrees, and hbt+ (Hierarchical Bound-Tracing) (Han et al., 27 Mar 2025).

In the SubFind + LHaloTrees branch, subhaloes are identified as locally overdense, self-bound structures within FoF halos, and merger trees are constructed by linking subhaloes across snapshots (Han et al., 27 Mar 2025). In the HBT+ branch, the method is time-domain rather than snapshot-local: it tracks each halo’s particles continuously, gives merger trees through uninterrupted track IDs, and records subhalo depth levels, where level-1 infalls directly into the host and level-2 infalls into a level-1 first (Han et al., 27 Mar 2025).

The subhalo-convergence analysis provides a more formal HBT+ definition. At each snapshot, new subhalos form when a smaller FoF halo merges into a larger one, and each subhalo is defined by the self-bound particles that survive iterative unbinding:

$1\,\mathrm{Gpc}/h$1

Even after a subhalo falls below the resolution limit, HBT+ continues to follow its most-bound particle as an orphan, thereby preserving identity and merger history (Xu, 11 Aug 2025).

For Jiutian-1G-based mock catalogs, the resolution threshold is operationally important. One project summary states that HBT+ formally resolves subhalos down to 20 particles, corresponding to $1\,\mathrm{Gpc}/h$2, but that most analyses adopt a conservative soft limit of 100 particles, approximately $1\,\mathrm{Gpc}/h$3, to ensure robust subhalo structure and merger histories (Tan et al., 5 Nov 2025). This threshold governs where galaxy stellar-mass functions and luminosity functions in the mocks should be interpreted as complete.

5. Galaxy, lensing, and survey-mock pipelines

Jiutian-1G is used as the dark-matter substrate for several distinct mock-catalog pipelines. The Jiutian overview lists four sets of mock galaxy light-cone catalogs built on the primary runs: a GAEA-hbt+ light-cone catalog, an LGalaxies-Line emission-line catalog, an LGalaxies-Lensing weak and strong lensing catalog, and a SHAM-DESI empirical catalog (Han et al., 27 Mar 2025).

In the GAEA-hbt+ pipeline, the GAlaxy Evolution and Assembly (GAEA) semi-analytical model is grafted onto hbt+ merger trees. Star-formation histories are passed to the neural-network-accelerated stellar population synthesizer StarDuster, and light-cone construction uses on-the-fly interpolation plus the Blic lightcone builder. The outputs include positions, velocities, stellar, gas, and metal masses, spectral energy distributions, emission lines, and merger-history links (Han et al., 27 Mar 2025). The dedicated CSST catalog paper further states that StarDuster is trained on radiative transfer simulations to account for detailed galaxy geometry in modeling dust obscuration, and that BLiC generates galaxy light-cones up to $1\,\mathrm{Gpc}/h$4 by interpolating galaxy properties over time with an optimized interpolation scheme (Tan et al., 5 Nov 2025).

That same CSST catalog is reported to reproduce observed stellar mass functions, luminosity functions, gas mass fractions, the galaxy size-mass relation, and galaxy clustering across a range of galaxy mass and redshift, while also presenting the photometric and redshift distributions expected for CSST surveys (Tan et al., 5 Nov 2025). In this context, Jiutian-1G is used jointly with Jiutian-2G to test convergence across two resolutions.

The weak-lensing and clustering forecast for the CSST photometric survey uses Jiutian-1G differently. It constructs a partial-sky light cone to $1\,\mathrm{Gpc}/h$5 covering $1\,\mathrm{Gpc}/h$6, adopts a special-viewing-angle strategy to reduce box-replication artifacts, slices the cone into planes of thickness $1\,\mathrm{Gpc}/h$7, and performs multi-lens-plane ray tracing with LensTools (Xiong et al., 2024). Galaxy populations in that study are produced with an improved version of the L-Galaxies semi-analytic model, building on FoF/SUBFIND halo catalogs and using Bruzual and Charlot (2003) stellar population synthesis with a Chabrier IMF and Charlot and Fall (2000) dust attenuation. The mock selection then applies the CSST $1\,\mathrm{Gpc}/h$8 point-source limit $1\,\mathrm{Gpc}/h$9 and assigns photometric redshifts with Gaussian error $2\,\mathrm{Gpc}/h$0 (Xiong et al., 2024).

An important technical distinction is that the Jiutian primary-run reference states that on-the-fly lightcone outputs are not produced for the 1G run, whereas lightcones are available on the fly for the $2\,\mathrm{Gpc}/h$1 and $2\,\mathrm{Gpc}/h$2 runs (Han et al., 27 Mar 2025). Published Jiutian-1G applications therefore rely on post-processing light-cone generation rather than native on-the-fly 1G lightcone outputs.

6. Convergence properties, correction schemes, and limitations

The most detailed published numerical-convergence analysis for Jiutian-1G focuses on subhalo abundance and phase-space structure using HBT+ catalogs and comparison to a higher-resolution Jiutian-300 run (Xu, 11 Aug 2025). The key variable is the surviving subhalo peak mass function (SPMF), with

$2\,\mathrm{Gpc}/h$3

Within $2\,\mathrm{Gpc}/h$4, the SPMF is fitted by a double-Schechter form,

$2\,\mathrm{Gpc}/h$5

reported to achieve approximately $2\,\mathrm{Gpc}/h$6 accuracy over $2\,\mathrm{Gpc}/h$7, $2\,\mathrm{Gpc}/h$8, and $2\,\mathrm{Gpc}/h$9 (Xu, 11 Aug 2025).

For direct convergence, the same study states that surviving subhalos converge only when $\mathrm{Gpc}/h$0 exceeds 5000 particles. For Jiutian-1G, this threshold is

$\mathrm{Gpc}/h$1

Below that scale, the paper argues that orphan recovery is necessary. It applies the merger-timescale model of Jiang et al. (2008), for which

$\mathrm{Gpc}/h$2

with $\mathrm{Gpc}/h$3, $\mathrm{Gpc}/h$4, and $\mathrm{Gpc}/h$5. Among the tested prescriptions, this model is reported to outperform Boylan-Kolchin (2008), Lacey and Cole (1993), Simha and Cole (2017), and Xu and Jing (2025) when applied to HBT+ catalogs (Xu, 11 Aug 2025).

With Jiang08 orphan treatment, Jiutian-1G recovers the SPMF to $\mathrm{Gpc}/h$6 error down to the 5000-particle threshold, and real-space spatial clustering and velocity dispersion agree with Jiutian-300 to $\mathrm{Gpc}/h$7--$\mathrm{Gpc}/h$8 down to $\mathrm{Gpc}/h$9--$z\lesssim2.5$0 (Xu, 11 Aug 2025). Below $z\lesssim2.5$1, significant deviations remain. The study attributes these to numerical disruption of low-energy orbits and notes that more sophisticated orphan dynamics are required.

The limitations are sharper in redshift space. Even when real-space positions and velocities appear converged at $z\lesssim2.5$2--$z\lesssim2.5$3, unresolved small-$z\lesssim2.5$4 pairs leak to much larger redshift-space separations through Fingers-of-God. The monopole $z\lesssim2.5$5 reaches only $z\lesssim2.5$6 agreement at scales $z\lesssim2.5$7--$z\lesssim2.5$8, while the quadrupole and hexadecapole remain more than $z\lesssim2.5$9 discrepant until $z=5$0 (Xu, 11 Aug 2025). The paper therefore recommends modified multipoles with a $z=5$1 cut for redshift-space distortion studies.

These results define the practical operating domain of Jiutian-1G. The simulation provides large statistical volume and sufficient resolution for many halo, galaxy, and lensing applications, but subhalo-based inference on very small scales remains limited by numerical convergence and orphan modeling.