The Three Hundred Project: Cluster ReSimulations

• The Three Hundred Project is a suite of 324 galaxy cluster zoom re-simulations that integrate full baryonic physics and semi-analytic models for comprehensive astrophysical analyses.
• It employs diverse hydrodynamical codes, including GADGET-X, GADGET-MUSIC, and GIZMO-Simba, to capture gas dynamics, star formation, and feedback mechanisms across scales.
• The project provides robust scaling relations, merger dynamics, and survey-facing methodologies, serving as a benchmark for studies in precision cosmology and cluster astrophysics.

The Three Hundred Project is a suite of 324 galaxy-cluster zoom re-simulations drawn from the MultiDark Planck 2 dark-matter-only run and re-simulated with full baryonic physics, together with matched semi-analytic realizations on the same halo backbone, for cosmological and astrophysical applications (Cui et al., 2018). Its stated objectives include providing a mass-complete sample of the most massive clusters at $z=0$ and their environments, enabling precision cosmology via cluster abundance and internal structure, advancing understanding of cooling, star formation, and AGN/SN feedback on cluster scales, and offering direct comparisons between hydrodynamical and semi-analytic models (Cui et al., 2018). Subsequent work has used the same framework to study baryon fractions, gas and stellar profiles, cluster outskirts, mergers, hydrostatic-mass bias, radio relics, mass mapping, splashback structure, and spectroscopic mass estimators (Rasia et al., 27 May 2025).

1. Scope, sample construction, and scientific remit

The baseline project re-simulates the 324 most massive $z=0$ clusters in the MDPL2 parent simulation, a $1\,h^{-1}\,\mathrm{Gpc}$ cube with $3840^3$ particles in a Planck-2015 cosmology (Cui et al., 2018). For each target, a $15\,h^{-1}\,\mathrm{Mpc}$-radius region is regenerated with multi-level refinement, yielding a mass-complete sample of the most massive clusters with $M_{200}>6.4\times10^{14}\,h^{-1}\,M_\odot$ at $z=0$, while also resolving $\simeq 5500$ haloes down to $M_{200}=10^{13}\,h^{-1}\,M_\odot$ without low-resolution contamination (Cui et al., 2018). Snapshots have been stored at 128 epochs between $z=17$ and $z=0$0, so the project supports both low-redshift statistical analyses and progenitor tracking over a large fraction of cosmic time (Cui et al., 2018).

The project is not restricted to a single hydrodynamical realization. Later extensions add the \textsc{Gizmo-Simba} run, which places the SIMBA galaxy-formation model in the same zoom framework and emphasizes the co-evolution of galaxies, gas, and black holes in cluster environments (Cui et al., 2022). Other derivative studies explicitly treat the suite as a platform for cluster astrophysics, cluster-galaxy evolution, mass-observable calibration, and survey forecasting (Ansarifard et al., 2019). This suggests that “The Three Hundred Project” functions less as a single simulation product than as a common experimental infrastructure for cross-code, cross-method, and survey-facing cluster research.

2. Numerical implementations and data infrastructure

The original hydrodynamical realizations were run with \textsc{GADGET-X} and \textsc{GADGET-MUSIC}, both using the same initial conditions but different SPH formulations and sub-grid prescriptions (Cui et al., 2018). \textsc{GADGET-X} uses a “modern” SPH scheme with a Wendland-C4 kernel, artificial conductivity and viscosity, metal-dependent cooling, star formation and SN feedback, and dual-mode AGN feedback; \textsc{GADGET-MUSIC} uses an entropy-conserving SPH formulation, metal-independent cooling, Springel & Hernquist star formation, and no AGN feedback (Cui et al., 2018). The \textsc{Gizmo-Simba} extension instead uses the GIZMO meshless finite-mass solver with SIMBA sub-grid physics, including $z=0$1-regulated star formation, two-phase winds, torque-limited cold-gas accretion, Bondi hot accretion, kinetic AGN winds, and X-ray heating (Cui et al., 2022).

Implementation	Type	Distinctive features
\textsc{GADGET-X}	Full hydro	Modern SPH, metal-dependent cooling, AGN feedback
\textsc{GADGET-MUSIC}	Full hydro	Classical SPH, no AGN feedback
\textsc{Gizmo-Simba}	Full hydro	MFM hydro, SIMBA feedback, kinetic jets
Galacticus / SAG / SAGE	SAMs	Same MDPL2 dark-matter backbone

The baseline high-resolution masses in the original hydrodynamical project are $z=0$2 and $z=0$3, with physical Plummer-equivalent softening $z=0$4 below $z=0$5–2 and comoving above (Cui et al., 2018). The emulator work later introduces “3K”, “7K”, and “15K” dark-matter-only realizations, with corresponding GIZMO hydrodynamical runs at 3K and 7K, to transfer cluster-galaxy statistics from hydro to higher-resolution DMO calculations (Gómez et al., 4 Apr 2025).

Public accessibility is a formal part of the project. The released products include roughly $z=0$6 of simulations and $z=0$7 of halo catalogues, with full-physics snapshots, AHF halo catalogues, merger trees, stellar-population outputs, project website access, and CosmoSim database entries; the introductory paper states that all simulations and derived data products are publicly available (Cui et al., 2018).

3. Operational definitions and theoretical quantities

The project uses standard overdensity definitions,

$z=0$8

with common choices $z=0$9 or $1\,h^{-1}\,\mathrm{Gpc}$0 (Cui et al., 2018). Baryonic fractions are defined as

$1\,h^{-1}\,\mathrm{Gpc}$1

and scaling relations are generally written as $1\,h^{-1}\,\mathrm{Gpc}$2, with $1\,h^{-1}\,\mathrm{Gpc}$3 (Cui et al., 2018).

Several later analyses extend this common language to the intracluster medium and cluster outskirts. The entropy profile is defined by

$1\,h^{-1}\,\mathrm{Gpc}$4

and two operational definitions of the shock radius are used: the peak-entropy radius $1\,h^{-1}\,\mathrm{Gpc}$5, where $1\,h^{-1}\,\mathrm{Gpc}$6 is maximal, and the slope-minimum radius $1\,h^{-1}\,\mathrm{Gpc}$7, where $1\,h^{-1}\,\mathrm{Gpc}$8 attains its minimum (Zhang et al., 2024). The splashback radius $1\,h^{-1}\,\mathrm{Gpc}$9 is measured from the spherically averaged 3D mass-density profile as the radius where $3840^3$0 is minimal (Zhang et al., 2024). In baryon-fraction work, the hot-gas fraction is defined using gas particles with $3840^3$1 that are not flagged as star-forming, and the relative fraction is written as

$3840^3$2

with $3840^3$3 (Rasia et al., 27 May 2025).

The project has also been used to study the cluster fundamental plane,

$3840^3$4

where $3840^3$5 and $3840^3$6 are the NFW scale radius and scale mass, respectively (Díaz-García et al., 2021). Across these applications, the project repeatedly uses profile minima, profile curvature, and self-similar normalization to define physically motivated cluster boundaries and observables.

4. Baryonic structure, scaling relations, and self-similarity

The introductory analysis found that the modelled clusters are generally in reasonable agreement with observations with respect to baryonic fractions and gas scaling relations at $3840^3$7, while also showing model-dependent differences such as overly massive central galaxies and galaxy colours shifted to bluer $3840^3$8 by about $3840^3$9 dex at the peak position (Cui et al., 2018). Gas scaling relations down to $15\,h^{-1}\,\mathrm{Mpc}$0 are a particularly robust result: the $15\,h^{-1}\,\mathrm{Mpc}$1–$15\,h^{-1}\,\mathrm{Mpc}$2 slopes are $15\,h^{-1}\,\mathrm{Mpc}$3 for \textsc{GADGET-MUSIC} and $15\,h^{-1}\,\mathrm{Mpc}$4 for \textsc{GADGET-X}, versus the non-radiative self-similar value $15\,h^{-1}\,\mathrm{Mpc}$5, and the $15\,h^{-1}\,\mathrm{Mpc}$6–$15\,h^{-1}\,\mathrm{Mpc}$7 slopes are $15\,h^{-1}\,\mathrm{Mpc}$8–$15\,h^{-1}\,\mathrm{Mpc}$9, consistent with Planck and weak-lensing measurements (Cui et al., 2018). The same study argues that gas scaling relations are remarkably robust to details of feedback prescriptions, at least above $M_{200}>6.4\times10^{14}\,h^{-1}\,M_\odot$0 (Cui et al., 2018).

A more profile-oriented analysis shows that gas density profiles in both hydrodynamical runs match observed trends well, including reduced scatter at large radii, while \textsc{GADGET-X} reproduces the observed temperature-profile shape more successfully than \textsc{GADGET-MUSIC}; by contrast, cumulative stellar density profiles from semi-analytic models are in better agreement with SDSS than the hydrodynamical runs, which show relatively higher profiles (Li et al., 2020). The same paper finds that gas density and metallicity display the strongest self-similarity beyond $M_{200}>6.4\times10^{14}\,h^{-1}\,M_\odot$1, whereas temperature and stellar mass density show larger scatter and stronger dependence on dynamical state and cool-core status (Li et al., 2020).

The evolution paper extends these conclusions to $M_{200}>6.4\times10^{14}\,h^{-1}\,M_\odot$2, finding that both \textsc{Gadget-X} and \textsc{Gizmo-Simba} agree with observations mostly at outer radii $M_{200}>6.4\times10^{14}\,h^{-1}\,M_\odot$3, consistent with self-similarity, but diverge in the centre, where \textsc{Gizmo-Simba} often produces large entropy cores and higher central temperatures while \textsc{Gadget-X} remains closer to low-redshift X-ray samples (Li et al., 2023). In the baryon-fraction analysis, simple power-law mass trends are found to be inadequate: high masses show a near-constant plateau, low masses show a steep decline, and the transition near $M_{200}>6.4\times10^{14}\,h^{-1}\,M_\odot$4–$M_{200}>6.4\times10^{14}\,h^{-1}\,M_\odot$5 is better captured by either a parabolic or logarithmic form in the logarithmic plane, with stronger redshift evolution at smaller radii (Rasia et al., 27 May 2025). A plausible implication is that cluster baryon-content calibration depends on both mass scale and aperture, not merely on a single global slope.

5. Outskirts, accretion shocks, splashback, and infall processing

The project has been used extensively to characterize the physical boundary of clusters. In the shock–splashback study based on 324 \textsc{GIZMO-Simba-7k} clusters, the median stacked radii are

$M_{200}>6.4\times10^{14}\,h^{-1}\,M_\odot$6

and individual-cluster fits give

$M_{200}>6.4\times10^{14}\,h^{-1}\,M_\odot$7

in $M_{200}>6.4\times10^{14}\,h^{-1}\,M_\odot$8 (Zhang et al., 2024). Both $M_{200}>6.4\times10^{14}\,h^{-1}\,M_\odot$9 and $z=0$0 anti-correlate with $z=0$1 and with recent fractional mass growth, while $z=0$2 tends to be larger for clusters with higher recent accretion rates (Zhang et al., 2024). The paper summarizes this as the shock radius lying systematically outside the splashback radius by $z=0$3–$z=0$4 (Zhang et al., 2024).

A complementary stellar-splashback analysis dynamically decomposes clusters into orbiting and infalling material and finds that the truncation radius $z=0$5, associated with the splashback feature, coincides for stars and dark matter, with $z=0$6, while the stellar orbiting profile declines more steeply, with $z=0$7 versus $z=0$8 (Walker et al., 10 Aug 2025). The fitted relation

$z=0$9

has $\simeq 5500$0, $\simeq 5500$1, and $\simeq 5500$2, and projected stellar profiles recover $\simeq 5500$3 with RMS scatter $\simeq 5500$4 (Walker et al., 10 Aug 2025). This suggests that diffuse stellar-light measurements can act as a proxy for recent mass accretion.

The same outer-cluster environment hosts strong preprocessing and backsplash phenomena. Infalling haloes within $\simeq 5500$5 lose $\simeq 5500$6 of their gas by $\simeq 5500$7 and then rapidly lose the remainder by $\simeq 5500$8 in 3D, with the projected 100% gas-loss radius shifting inward to $\simeq 5500$9 (Mostoghiu et al., 2021). Separately, backsplash galaxies—defined as systems that previously passed within $M_{200}=10^{13}\,h^{-1}\,M_\odot$0 but are now outside it—constitute 58% of all galaxies between $M_{200}=10^{13}\,h^{-1}\,M_\odot$1 and $M_{200}=10^{13}\,h^{-1}\,M_\odot$2 at $M_{200}=10^{13}\,h^{-1}\,M_\odot$3, with 95% of clusters lying between 21% and 85%; the fraction is higher in dynamically relaxed clusters and builds rapidly at $M_{200}=10^{13}\,h^{-1}\,M_\odot$4 (Haggar et al., 2020). Taken together, these studies treat the outskirts not as a single boundary but as a set of coupled transitions in gas thermodynamics, collisionless orbits, and galaxy environmental history.

6. Mergers, radio relics, and galaxy evolution in dense environments

Cluster mergers are another major theme. A merger catalogue based on rapid mass jumps in the main-progenitor history identifies 178 merger events with $M_{200}=10^{13}\,h^{-1}\,M_\odot$5 and $M_{200}=10^{13}\,h^{-1}\,M_\odot$6, reduced to 164 after requiring $M_{200}=10^{13}\,h^{-1}\,M_\odot$7 and $M_{200}=10^{13}\,h^{-1}\,M_\odot$8 to be defined; the inferred mass ratios peak at $M_{200}=10^{13}\,h^{-1}\,M_\odot$9, confirming that the selected events are major mergers (Contreras-Santos et al., 2022). The same work finds that BCG stellar growth during mergers is dominated by ex-situ accretion, which provides $z=17$0 of $z=17$1, while in-situ star formation provides $z=17$2, and BCGs in mergers form in median around 70 per cent more stars than those in a control sample (Contreras-Santos et al., 2022). Observable consequences include a median increase in SDSS-$z=17$3 luminosity of about 20 per cent during mergers and a slower increase of the $z=17$4 colour (Contreras-Santos et al., 2022).

A radio-emission analysis uses 555 major-merger events with $z=17$5 to model merger-induced shock fronts and their synchrotron output (Nuza et al., 2024). The median radio luminosity light curve rises sharply after core passage, peaks after $z=17$6–$z=17$7, and its peak enhancement relative to the emission at core passage is $z=17$8 for groups and $z=17$9–$z=0$00 for massive clusters (Nuza et al., 2024). The peak relic power follows

$z=0$01

with scatter $z=0$02–1 dex at fixed mass (Nuza et al., 2024). Most merger orbits are fairly radial, with a median opening angle of $z=0$03 before collision and a median impact parameter of about $z=0$04 (Nuza et al., 2024).

These studies collectively connect the project’s dynamical merger histories to directly modelled observables. The evidence in the data is that mergers reshape both the stellar content of BCGs and the non-thermal emission of the outer intracluster medium, while the details depend on orbit, mass, and accretion geometry.

7. Inference, emulation, and survey-facing methodologies

A substantial part of the project’s later impact lies in methods development. For hydrostatic-equilibrium masses, an analysis of more than 300 simulated massive clusters finds raw median biases at $z=0$05 of $z=0$06 and $z=0$07, with the underestimation increasing in irregular, high-clumpiness systems (Ansarifard et al., 2019). By combining X-ray azimuthal scatter or ellipticity with the asymptotic density or pressure slope, the proposed corrections bring the medians to 1.00 and reduce the skewness of the SZ bias distribution from $z=0$08 to $z=0$09 (Ansarifard et al., 2019). The stated goal is to reduce and symmetrize the HE-mass bias distribution for cosmological analyses of X-ray and SZ surveys (Ansarifard et al., 2019).

The suite has also been used as supervised-learning training data. A multiview U-Net trained on idealized SZ, X-ray, and stellar surface-density maps from The Three Hundred predicts total projected mass maps with median cylindrical mass bias at $z=0$10 of about 1% and scatter of about 3% for the best multiview model, while reproducing the 2D mass-map power spectrum to $z=0$11 down to $z=0$12 (Andres et al., 2023). The authors explicitly note that the method depends on the physics implemented in the hydrodynamic simulations, but the near-unbiased integrated masses indicate that the simulated map pairs form a useful benchmark for non-lensing reconstruction strategies (Andres et al., 2023).

Semi-analytic emulation pushes this idea further. A calibrated emulator based on the SAMs SAGE and SAG is trained against 7K-GIZMO cluster galaxies and then run on DMO versions of The Three Hundred at three resolutions; SAG is found to emulate the hydrodynamical galaxy populations more successfully than SAGE because of its explicit treatment of orphan galaxies (Gómez et al., 4 Apr 2025). The resulting 15K-SAG realization extends the cumulative stellar mass function to $z=0$13, about an order of magnitude below 7K-GIZMO, while a full 7K hydro region requires $z=0$14 CPU-h and the SAM emulator runs in $z=0$15 CPU-h per region (Gómez et al., 4 Apr 2025). This is presented as a fast and efficient emulator of hydrodynamical cluster simulations (Gómez et al., 4 Apr 2025).

Finally, the project has been used to forecast spectroscopic mass-profile recovery for upcoming surveys. In a caustic-method study targeting CATARSIS, a non-iterative filling-factor treatment gives median $z=0$16 or 1.12, depending on the assumed anisotropy model, whereas the iterative approaches reduce the median to 0.94 or 0.93 and lower the scatter to 0.18–0.20 (Córdoba et al., 2 Jun 2026). CATARSIS itself is described as targeting 16 clusters at $z=0$17, aiming for 500–1000 secure redshifts per cluster within $z=0$18, with forecast per-cluster $z=0$19 biases $z=0$20 and scatter $z=0$21 (Córdoba et al., 2 Jun 2026). In this survey-facing mode, The Three Hundred Project serves as a controlled environment for turning dynamical, thermodynamical, and photometric mock data into mass-calibration algorithms and observational strategy tests.