Three Hundred Simulations Overview

• The Three Hundred simulations are a coordinated set of zoom-in galaxy cluster re-simulations featuring full-physics baryonic modeling alongside dark-matter-only and semi-analytic approaches.
• They employ multiple hydrodynamical models (Gadget-X, GADGET-MUSIC, GIZMO-SIMBA) and semi-analytic methods to enable robust code comparisons and isolate baryonic effects.
• The suite offers publicly accessible data products that support cluster mass calibration, weak-lensing studies, and in-depth analyses of galaxy populations, substructure, and intra-cluster light.

The Three Hundred simulations are a coordinated set of zoom-in galaxy-cluster re-simulations built to combine cluster-scale statistics with full-physics baryonic modeling. In the project’s basic construction, 324 of the most massive halos identified in the MDPL2 dark-matter-only simulation are re-simulated inside spherical regions of radius $15\,h^{-1}\,\mathrm{Mpc}$, producing a matched laboratory of hydrodynamical, dark-matter-only, and semi-analytic realizations of the same cluster population. The suite was introduced as a public catalogue for cosmological and astrophysical applications and later expanded into a multi-model program that includes Gadget-X, GADGET-MUSIC, GIZMO-SIMBA, and semi-analytic galaxy catalogues, with applications ranging from cluster scaling relations and weak-lensing mass calibration to intra-cluster light, subhalo structure, and mock survey support (Cui et al., 2018, Cui et al., 2022).

1. Project architecture and sample definition

The parent simulation for The Three Hundred is MDPL2, described as a periodic box of side length $1\,h^{-1}\,\mathrm{Gpc}$ with $3840^3$ dark-matter particles and Planck cosmology. The core sample consists of the 324 most massive cluster-scale halos at $z=0$, selected at roughly $M_{200}\gtrsim 10^{14.8}\,M_\odot$ or, equivalently in the project description, with a virial-mass threshold around $M_{\rm vir}\gtrsim 8\times10^{14}\,M_\odot$. Each target is re-simulated in a large Lagrangian region extending to $15\,h^{-1}\,\mathrm{Mpc}$, large enough to include not only the central cluster but also surrounding groups and filaments, which is why the suite supports studies of preprocessing and cluster environments as well as the main halo population (Cui et al., 2018, Cui et al., 2022).

A mass-complete subsample is defined by comparison with the parent MDPL2 halo mass function. At $z=0$, the reported completeness limits are $M_{200}=6.42\times10^{14}\,M_\odot$ and $M_{500}=4.60\times10^{14}\,M_\odot$. In Euclid-oriented analyses, the zoom ensemble is treated as probing a $1\,h^{-1}\,\mathrm{Gpc}$0 volume, and the cluster selection is summarized as “a complete sample of massive clusters beyond $1\,h^{-1}\,\mathrm{Gpc}$1 at redshift 0 (1)” (Cui et al., 2018, Muñoz et al., 2023).

The project is not numerically monolithic. Some analyses describe the cluster database as evolved in 128 snapshots from $1\,h^{-1}\,\mathrm{Gpc}$2 to $1\,h^{-1}\,\mathrm{Gpc}$3, while GIZMO-based studies use 129 snapshots from $1\,h^{-1}\,\mathrm{Gpc}$4 to $1\,h^{-1}\,\mathrm{Gpc}$5. This difference reflects the specific realization or post-processing pipeline being used rather than a change in the underlying scientific role of the suite, namely a statistically large, matched cluster sample with resolved baryonic structure and controlled model variation (Paliwal et al., 2021, Walker et al., 10 Aug 2025).

2. Numerical implementations and model families

A defining feature of The Three Hundred is that the same initial conditions are realized with multiple hydrodynamical and semi-analytic models. This allows code-to-code comparisons that isolate the effect of baryonic prescriptions from halo-to-halo variance. The original public release emphasized two hydrodynamical runs, Gadget-X and GADGET-MUSIC, together with three semi-analytic models—Galacticus, SAG, and SAGE—applied to the same underlying dark-matter-only merger trees. Later work added the GIZMO-SIMBA hydrodynamical suite, thereby extending the project into a genuinely multi-code cluster laboratory (Cui et al., 2018, Cui et al., 2022).

Component	Method	Noted features
Gadget-X	modern SPH hydrodynamics	metal-dependent cooling, star formation, stellar feedback, AGN feedback
GADGET-MUSIC	classical SPH hydrodynamics	no AGN feedback
GIZMO-SIMBA	GIZMO meshless finite-mass / meshless hydrodynamics with SIMBA physics	$1\,h^{-1}\,\mathrm{Gpc}$6-based star formation, two-phase winds, black-hole accretion/feedback
Galacticus, SAG, SAGE	semi-analytic models on matched DMO trees	model comparison and emulator construction

GIZMO-SIMBA was introduced as a new hydrodynamic component of The Three Hundred, using GIZMO in meshless finite-mass mode and importing the Simba galaxy-formation model. The implementation is described as including $1\,h^{-1}\,\mathrm{Gpc}$7-based star formation, torque-limited cold accretion onto black holes, kinetic bipolar AGN feedback, and on-the-fly dust tracking, with a mild re-calibration to recover $1\,h^{-1}\,\mathrm{Gpc}$8 cluster stellar properties at The Three Hundred resolution. A central result of the model-comparison program is that broadly similar $1\,h^{-1}\,\mathrm{Gpc}$9 stellar observables can arise from different histories: Gizmo-Simba forms stars earlier, quenches more abruptly, ejects more gas, and leaves older and redder stellar populations than Gadget-X (Cui et al., 2022).

The project’s matched-code design also supports direct robustness tests. For example, intra-cluster-light analyses use Gadget-X and Gizmo-Simba on exactly the same initial conditions, making it possible to distinguish trends driven by hierarchical cluster growth from trends driven by hydrodynamical method or subgrid physics. This is especially important where the two models agree on integrated structure but differ on the timing of star formation or the survival of stripped gas (Contreras-Santos et al., 2024, Contreras-Santos et al., 22 Sep 2025).

3. Data products, observables, and mock-observation infrastructure

The Three Hundred was introduced not only as a simulation suite but also as a public data resource. The project states that the initial conditions, simulation outputs, halo catalogues, full hydrodynamical cluster dataset, and semi-analytical catalogues are publicly available. Companion work was explicitly expected to provide a multi-wavelength mock-observation database including radio/SZ, optical, X-ray, and lensing products, and later analyses use exactly such synthetic observables as core scientific inputs (Cui et al., 2018).

A particularly explicit realization of this observational interface appears in the NIKA2 Sunyaev-Zeldovich Large Program twin-sample analysis. There, The Three Hundred provides synthetic X-ray, optical, gravitational lensing, radio, and SZ Compton-parameter maps for clusters selected to match the observational sample in redshift, mass, and integrated Compton parameter $3840^3$0. These products are used in a like-for-like way with real NIKA2, Planck, ACT, X-ray, optical, and lensing data, so that the same mass-proxy pipelines can be applied to both simulated and observed clusters. The intention is not merely visual comparison: SZ maps are used to measure $3840^3$1, X-ray photon-count maps to infer gas density and hydrostatic mass, optical data to estimate velocity dispersion, and lensing $3840^3$2-maps to probe projected mass (Paliwal et al., 2021).

Halo and subhalo identification across the suite is commonly carried out with AHF, and several analyses rely on the fact that AHF can identify host halos, subhalos, and nested substructures in both hydrodynamical and dark-matter-only realizations. In particle-tracking studies this catalogue layer is essential because it provides the operational definitions of the brightest cluster galaxy, the intra-cluster light, satellite membership, and the distinction between material bound to the main halo and material bound to substructure (Contreras-Santos et al., 2024, Walker et al., 10 Aug 2025).

4. Cluster masses, baryons, and cosmological calibration

The Three Hundred has been used extensively for cluster mass definitions, baryonic fractions, and scaling-relation calibration. Standard overdensity conventions are used throughout, with

$3840^3$3

In matched halo comparisons, baryonic effects on $3840^3$4 are reported as small, with a median difference of about $3840^3$5 relative to the dark-matter-only reference, while $3840^3$6 is typically $3840^3$7–$3840^3$8 higher in the hydrodynamical runs below $3840^3$9. Dynamical state is characterized by the standard relaxation indicators

$z=0$0

with relaxed clusters satisfying $z=0$1, $z=0$2, and $z=0$3 (Cui et al., 2018).

Gas and SZ scaling relations are among the project’s most stable outputs. The temperature–mass and $z=0$4–mass relations were found to be in reasonable agreement with observations at $z=0$5, and the $z=0$6–$z=0$7 relation is described as showing only weak sensitivity to subgrid baryonic differences down to $z=0$8. By contrast, stellar observables remain more model-dependent: central galaxies are often too massive, and cluster galaxies are systematically bluer than observed, with $z=0$9 lower by about $M_{200}\gtrsim 10^{14.8}\,M_\odot$0–$M_{200}\gtrsim 10^{14.8}\,M_\odot$1 dex at the peak of the distribution (Cui et al., 2018).

A more recent extension concerns baryon and hot-gas fraction evolution. Using about 300 simulated massive clusters with median mass $M_{200}\gtrsim 10^{14.8}\,M_\odot$2 at $M_{200}\gtrsim 10^{14.8}\,M_\odot$3, the project modeled $M_{200}\gtrsim 10^{14.8}\,M_\odot$4 and $M_{200}\gtrsim 10^{14.8}\,M_\odot$5 at $M_{200}\gtrsim 10^{14.8}\,M_\odot$6, $M_{200}\gtrsim 10^{14.8}\,M_\odot$7, and $M_{200}\gtrsim 10^{14.8}\,M_\odot$8 from $M_{200}\gtrsim 10^{14.8}\,M_\odot$9 to $M_{\rm vir}\gtrsim 8\times10^{14}\,M_\odot$0. The central conclusion is methodological: a power law poorly describes these relations across the full mass range, because it cannot simultaneously represent high-mass flattening, low-mass depletion, and the intermediate transition. Quadratic and logarithmic forms in the logarithmic plane describe the curvature more accurately, and the evolution is strongest at smaller radii, especially for the hot-gas fraction (Rasia et al., 27 May 2025).

The suite is also used directly for mass calibration. In the NIKA2 LPSZ “twin sample” program, simulated clusters are matched to about 50 SZ-selected targets at $M_{\rm vir}\gtrsim 8\times10^{14}\,M_\odot$1 using mass, redshift, and $M_{\rm vir}\gtrsim 8\times10^{14}\,M_\odot$2. Three matched synthetic samples are reported: $M_{\rm vir}\gtrsim 8\times10^{14}\,M_\odot$3, selected by total mass; $M_{\rm vir}\gtrsim 8\times10^{14}\,M_\odot$4, selected by hydrodynamic mass; and $M_{\rm vir}\gtrsim 8\times10^{14}\,M_\odot$5, selected by $M_{\rm vir}\gtrsim 8\times10^{14}\,M_\odot$6. The reported average ratios are close to unity: $M_{\rm vir}\gtrsim 8\times10^{14}\,M_\odot$7

$M_{\rm vir}\gtrsim 8\times10^{14}\,M_\odot$8

The aim is to cross-correlate scaling relations such as $M_{\rm vir}\gtrsim 8\times10^{14}\,M_\odot$9, $15\,h^{-1}\,\mathrm{Mpc}$0, $15\,h^{-1}\,\mathrm{Mpc}$1, and $15\,h^{-1}\,\mathrm{Mpc}$2 so as to reduce scatter in inferred mass and relate mass bias to redshift, morphology, and dynamical state (Paliwal et al., 2021).

Weak-lensing applications take the calibration program further. Hydrodynamical weak-lensing projections out to $15\,h^{-1}\,\mathrm{Mpc}$3 show that weak-lensing masses are biased low on average relative to true 3D mass, that the bias depends on mass and redshift, and that the differences between GadgetX and GIZMO-SIMBA are modest when each model is compared to its own true mass. The same framework yields weak-lensing mass–richness relations broadly consistent with SDSS redMaPPer when a stellar-mass cut of $15\,h^{-1}\,\mathrm{Mpc}$4 is used. A plausible implication is that The Three Hundred provides not only intrinsic cluster physics but also simulation-based priors for survey-scale cluster cosmology (Giocoli et al., 23 Jan 2025).

5. Galaxy populations, substructure, and dense-environment dynamics

Because the zoom regions extend well beyond the central halo, The Three Hundred supports galaxy studies in cluster outskirts and high-density environments. Early comparisons already showed that satellite stellar mass functions are highly model-dependent: Gadget-MUSIC overproduces massive satellites, Gadget-X underproduces low-mass satellites, Galacticus overproduces low-mass satellites largely because of orphan galaxies, and none of the compared models fully solves the color discrepancy relative to observed clusters (Cui et al., 2018).

Later work used the suite to study observational close-pair selection in cluster environments. In the galaxy-pair analysis, galaxies within $15\,h^{-1}\,\mathrm{Mpc}$5 of the cluster center, with $15\,h^{-1}\,\mathrm{Mpc}$6, are projected onto the XY plane and selected with thresholds $15\,h^{-1}\,\mathrm{Mpc}$7 and $15\,h^{-1}\,\mathrm{Mpc}$8. “True” pairs are defined by the bound-pair criterion inherited from Haggar et al.: $15\,h^{-1}\,\mathrm{Mpc}$9 The results show an explicit purity–completeness trade-off: the tightest projected cuts can reach about $z=0$0–$z=0$1 purity but only about $z=0$2 completeness, whereas the loosest cuts recover about $z=0$3–$z=0$4 of true pairs at purity as low as about $z=0$5–$z=0$6. A Random Forest classifier trained on seven selected properties raises the default $z=0$7 kpc, $z=0$8 sample from about $z=0$9 purity and about $M_{200}=6.42\times10^{14}\,M_\odot$0 completeness to about $M_{200}=6.42\times10^{14}\,M_\odot$1 purity with about $M_{200}=6.42\times10^{14}\,M_\odot$2 completeness relative to the true-pair observational sample. The most important discriminants are size, mass, spin parameter, gas content, and stellar shape (Contreras-Santos et al., 2023).

The suite also demonstrates that massive dark matter-deficient satellites can arise naturally in a standard $M_{200}=6.42\times10^{14}\,M_\odot$3CDM cluster environment. For satellites inside $M_{200}=6.42\times10^{14}\,M_\odot$4 at $M_{200}=6.42\times10^{14}\,M_\odot$5, with $M_{200}=6.42\times10^{14}\,M_\odot$6 and excluding the BCG, the relevant diagnostic is the stellar-to-total bound-mass ratio $M_{200}=6.42\times10^{14}\,M_\odot$7. The simulations contain objects with $M_{200}=6.42\times10^{14}\,M_\odot$8 and stellar masses around $M_{200}=6.42\times10^{14}\,M_\odot$9. Their origin is traced to multiple pericentre passages that strip the extended dark-matter halo while leaving the compact stellar component comparatively intact. In the full sample, there are 302 galaxies with $M_{500}=4.60\times10^{14}\,M_\odot$0 and $M_{500}=4.60\times10^{14}\,M_\odot$1, roughly $M_{500}=4.60\times10^{14}\,M_\odot$2 of all satellite galaxies, and 9 more extreme systems with $M_{500}=4.60\times10^{14}\,M_\odot$3 and $M_{500}=4.60\times10^{14}\,M_\odot$4, each in a different host cluster. The reported Spearman trends,

$M_{500}=4.60\times10^{14}\,M_\odot$5

show that more dark matter-deficient satellites have typically completed more orbits and reached smaller pericentres (Contreras-Santos et al., 2024).

Subhalo structure remains a site of active tension. In comparisons to strong-lensing-inferred cluster subhalos, GIZMO-SIMBA reproduces the cumulative subhalo mass function more successfully than GADGET-X at low masses, but neither model fully matches the observed $M_{500}=4.60\times10^{14}\,M_\odot$6–$M_{500}=4.60\times10^{14}\,M_\odot$7 relation. At fixed $M_{500}=4.60\times10^{14}\,M_\odot$8, the simulated subhalos still have too small $M_{500}=4.60\times10^{14}\,M_\odot$9, especially below $1\,h^{-1}\,\mathrm{Gpc}$00. The residuals correlate with half-mass radius, stellar age, baryon fraction, cluster-centric distance, and relaxation state, indicating that compactness, stripping history, and host dynamical state all matter (Srivastava et al., 2023).

6. Intra-cluster light, cluster boundaries, and resolution extension

The intra-cluster light has become one of the most intensively studied components in The Three Hundred. In the operational definition used for the full-physics cluster sample, the brightest cluster galaxy is separated by a fixed 50 kpc spherical aperture centered on the cluster halo center, and the ICL is defined as the remaining stellar particles bound to the main cluster halo but not bound to any subhalo. Within $1\,h^{-1}\,\mathrm{Gpc}$01, the ICL contains $1\,h^{-1}\,\mathrm{Gpc}$02–$1\,h^{-1}\,\mathrm{Gpc}$03 of the total stellar mass, with median $1\,h^{-1}\,\mathrm{Gpc}$04, while the BCG contributes around $1\,h^{-1}\,\mathrm{Gpc}$05; the BCG fraction is summarized as below $1\,h^{-1}\,\mathrm{Gpc}$06 for Gizmo-Simba and below $1\,h^{-1}\,\mathrm{Gpc}$07 for Gadget-X. No significant dependence on halo mass is found across roughly $1\,h^{-1}\,\mathrm{Gpc}$08 to $1\,h^{-1}\,\mathrm{Gpc}$09, but the dynamical-state dependence is clear: the 50 most relaxed clusters have ICL fractions of roughly $1\,h^{-1}\,\mathrm{Gpc}$10 in Gadget-X and $1\,h^{-1}\,\mathrm{Gpc}$11 in Gizmo-Simba, whereas the 50 most disturbed clusters drop to about $1\,h^{-1}\,\mathrm{Gpc}$12 and $1\,h^{-1}\,\mathrm{Gpc}$13, respectively (Contreras-Santos et al., 2024).

The same analysis shows that the ICL is a regular tracer of the dark-matter distribution. Up to $1\,h^{-1}\,\mathrm{Gpc}$14, the volumetric-density ratio is described by a power law,

$1\,h^{-1}\,\mathrm{Gpc}$15

with $1\,h^{-1}\,\mathrm{Gpc}$16 and $1\,h^{-1}\,\mathrm{Gpc}$17. The best-fit parameters are $1\,h^{-1}\,\mathrm{Gpc}$18, $1\,h^{-1}\,\mathrm{Gpc}$19 for Gadget-X and $1\,h^{-1}\,\mathrm{Gpc}$20, $1\,h^{-1}\,\mathrm{Gpc}$21 for Gizmo-Simba. The velocity-dispersion ratio is fit by

$1\,h^{-1}\,\mathrm{Gpc}$22

with $1\,h^{-1}\,\mathrm{Gpc}$23, $1\,h^{-1}\,\mathrm{Gpc}$24 for Gadget-X and $1\,h^{-1}\,\mathrm{Gpc}$25, $1\,h^{-1}\,\mathrm{Gpc}$26 for Gizmo-Simba. This suggests that the ICL can be used to infer dark-matter density and kinematic structure, although the exact normalization depends on dynamical state (Contreras-Santos et al., 2024).

Particle tracking establishes the origin of that diffuse component. Half of the present-day ICL mass is typically in place between $1\,h^{-1}\,\mathrm{Gpc}$27 and $1\,h^{-1}\,\mathrm{Gpc}$28, and by $1\,h^{-1}\,\mathrm{Gpc}$29 only about $1\,h^{-1}\,\mathrm{Gpc}$30 to $1\,h^{-1}\,\mathrm{Gpc}$31 of the final ICL mass has assembled. The dominant channel is stripping from subhalos after infall into the host cluster. Within that channel, $1\,h^{-1}\,\mathrm{Gpc}$32–$1\,h^{-1}\,\mathrm{Gpc}$33 of the ICL comes from contributors with infall stellar mass above $1\,h^{-1}\,\mathrm{Gpc}$34, and merger-ratio accounting gives medians of about $1\,h^{-1}\,\mathrm{Gpc}$35 from major mergers, $1\,h^{-1}\,\mathrm{Gpc}$36–$1\,h^{-1}\,\mathrm{Gpc}$37 from minor mergers, and $1\,h^{-1}\,\mathrm{Gpc}$38–$1\,h^{-1}\,\mathrm{Gpc}$39 from smooth accretion or very minor mergers. The in-situ channel is strongly model-dependent, with medians of 12 per cent in Gadget-X and 2 per cent in Gizmo-Simba; the non-central in-situ component in Gadget-X is interpreted as star formation in gas stripped from high-velocity infalling satellites rather than star formation in a long-lived diffuse medium (Contreras-Santos et al., 22 Sep 2025).

The stellar outskirts of the clusters also encode the splashback boundary. In the GIZMO run, the orbiting and infalling components are separated dynamically, and the orbiting profile is fit with a truncated Einasto-like form. The central result is that the truncation radius $1\,h^{-1}\,\mathrm{Gpc}$40 coincides for stars and dark matter on average, while the stellar profile has a systematically steeper truncation. Both components follow a consistent $1\,h^{-1}\,\mathrm{Gpc}$41–$1\,h^{-1}\,\mathrm{Gpc}$42 relation, where

$1\,h^{-1}\,\mathrm{Gpc}$43

and the projected stellar profile can, in principle, recover the splashback radius with scatter of about $1\,h^{-1}\,\mathrm{Gpc}$44. This suggests that stellar halo or intra-cluster-light profiles may serve as a proxy for the physical boundary of the cluster and for recent accretion history (Walker et al., 10 Aug 2025).

The same combination of hydrodynamical realism and matched dark-matter-only realizations underlies the project’s survey-facing extensions. For Euclid-oriented cluster finding, the current hydrodynamical mass resolution of $1\,h^{-1}\,\mathrm{Gpc}$45 is reported as insufficient to characterize the luminosity function down to Euclid depth, with the H-band luminosity function dropping around magnitude $1\,h^{-1}\,\mathrm{Gpc}$46 while Euclid is expected to reach about magnitude 24; the inferred requirement is $1\,h^{-1}\,\mathrm{Gpc}$47. Semi-analytic emulators built on matched DMO trees address this limitation. A SAGE calibration based on Particle Swarm Optimization generated about 10 million galaxies from the 324 regions of 7k-DMO and about 1 million galaxies from the 3 available 15k-DMO regions, while a later SAGE-versus-SAG emulator study found that SAG better reproduces hydrodynamical galaxy properties because of its treatment of orphan galaxies. In that formulation, SAG reaches dwarf galaxies down to stellar masses of $1\,h^{-1}\,\mathrm{Gpc}$48, about an order of magnitude below the hydrodynamical limit and approximately four magnitudes fainter, thereby extending The Three Hundred toward Euclid, 4MOST/ChANCES, and WEAVE use cases (Muñoz et al., 2023, Gómez et al., 2024, Gómez et al., 4 Apr 2025).