TNG-Cluster: Cosmological Cluster Simulation Suite

• TNG-Cluster is a suite of 352 high-resolution simulations that detail massive galaxy clusters using zoom-in techniques and a standardized physical model.
• The simulation employs TNG300-1-like resolution and advanced MHD methods to accurately capture ICM thermodynamics, AGN feedback, and multiwavelength observables.
• It provides actionable insights into cluster dynamics, cool-core classification, and protocluster evolution, bridging scale-specific simulation with observational benchmarks.

TNG-Cluster is a suite of 352 cosmological magnetohydrodynamical zoom-in simulations of massive galaxy clusters that extends the IllustrisTNG program to the rarest high-mass halos, while preserving the IllustrisTNG physical model and TNG300-1-like numerical resolution. It was designed to increase the statistical sampling of systems with $\log(M_{200c}/M_\odot)\gtrsim 14.3$–15.4 at $z=0$, thereby enabling population-level studies of the intracluster medium (ICM), brightest cluster galaxies, supermassive black holes, satellite evolution, and multiwavelength observables from X-ray and Sunyaev–Zeldovich signals to radio synchrotron emission (Nelson et al., 2023). In subsequent work, the suite has also been used as a high-mass complement to TNG300 for rare-halo and high-redshift studies, including AGN host halos before Cosmic Noon (Kapahtia et al., 11 Jun 2026).

1. Conception, sample, and simulation design

TNG-Cluster was built from a dark-matter-only parent simulation of side length $1\,\mathrm{Gpc}$, a volume described as approximately thirty-six times larger than TNG300, and re-simulates 352 cluster regions selected at $z=0$ (Nelson et al., 2023). The target selection includes all halos above $M_{200c}\ge 10^{15}\,M_\odot$ and a randomized subset across $10^{14.3}$–$10^{15}\,M_\odot$ chosen to produce roughly flat-in-log mass sampling at the high-mass end (Nelson et al., 2023). In the combined TNG300+TNG-Cluster sample, the number of halos with $M>10^{14}\,M_\odot$ at $z=0$ rises to 636, compared to 280 in TNG300 alone, which is central to its role in rare-halo statistics (Kapahtia et al., 11 Jun 2026).

The suite adopts the standard spherical-overdensity definitions

$$M_{\Delta c}=\frac{4\pi}{3}\,\Delta\,\rho_c(z)\,R_{\Delta c}^3,$$

with $z=0$0, so that $z=0$1 and $z=0$2 denote the masses enclosed within radii where the mean density equals $z=0$3 or $z=0$4 times the critical density, respectively (Nelson et al., 2023). These conventions structure nearly all later TNG-Cluster analyses, including cluster thermodynamics, gas motions, cool-core classification, and high-redshift progenitor tracking (Lehle et al., 2023).

A distinctive technical feature is the reconstruction of the 352 zooms into a “virtual box” with TNG-like snapshots, catalogs, and merger trees, allowing analyses to proceed with the same workflows used for previous IllustrisTNG releases (Nelson et al., 2023). The introductory overview reports 100 snapshots from $z=0$5 to $z=0$6, halo finding with Friends-of-Friends, substructure identification with SUBFIND, and merger trees with SubLink (Nelson et al., 2023). This design makes the suite simultaneously a targeted zoom program and a statistically organized cluster census.

2. Numerical methodology and physical model

TNG-Cluster uses the moving-mesh code AREPO, solving gravity with Tree-PM, hydrodynamics with a finite-volume Godunov scheme on a Voronoi mesh, and ideal MHD self-consistently (Nelson et al., 2023). In the magnetic-field study, the MHD solver is described as using Powell 8-wave divergence control, with a homogeneous primordial seed field of $z=0$7 comoving Gauss that is subsequently amplified by collapse, turbulence, and shear (Lehle et al., 16 Jul 2025). The high-resolution regions match TNG300-1, with mean gas cell mass $z=0$8, dark matter particle mass $z=0$9, and collisionless softening $1\,\mathrm{Gpc}$0; gas softening is adaptive with a minimum of 370 comoving pc (Nelson et al., 2023).

The galaxy-formation model is the IllustrisTNG model without recalibration, including radiative cooling and heating, star formation, stellar evolution and enrichment, stellar feedback, black-hole seeding and growth, and dual-mode AGN feedback (Nelson et al., 2023). In the AGN-host analysis, the shared subgrid setup with TNG300 is described more explicitly: SMBHs are seeded with mass $1\,\mathrm{Gpc}$1 in FoF halos above $1\,\mathrm{Gpc}$2, accrete through Eddington-limited Bondi accretion, and grow by mergers; feedback proceeds in a thermal mode at high accretion rates and a kinetic mode at low accretion rates, with a mass-dependent Eddington-ratio threshold governing the transition (Kapahtia et al., 11 Jun 2026). The overview paper also notes a minor technical update adopted from TNG50, namely a star-formation-timescale safety cap for very dense gas, with secondary consequences for the most massive BCGs and SMBHs (Nelson et al., 2023).

The suite adopts a Planck cosmology, with the overview paper listing $1\,\mathrm{Gpc}$3, $1\,\mathrm{Gpc}$4, $1\,\mathrm{Gpc}$5, $1\,\mathrm{Gpc}$6, $1\,\mathrm{Gpc}$7, and $1\,\mathrm{Gpc}$8 (Nelson et al., 2023). Additional on-the-fly diagnostics include Monte Carlo tracer particles, shock finding, and SMBH feedback episode recording, which later papers use to study accretion shocks, radio relics, and the origin of cool gas (Nelson et al., 2023). This combination of fixed subgrid physics and expanded high-mass statistics is fundamental to TNG-Cluster’s role as a controlled extension of IllustrisTNG rather than a model variant (Kapahtia et al., 11 Jun 2026).

3. Intracluster-medium thermodynamics and cluster core demographics

A central scientific use of TNG-Cluster is the thermodynamic characterization of the hot ICM. Radial profiles of entropy, temperature, density, and pressure show large diversity in cluster cores and much smaller scatter in the outskirts, while mass ordering is strong outside the core and weak within it (Lehle et al., 2023). The cool-core study defines entropy as

$1\,\mathrm{Gpc}$9

and cooling time as

$z=0$0

using non-star-forming, cooling gas with $z=0$1 (Lehle et al., 2023). Across the full $z=0$2 sample, the distributions of central cooling time, entropy, electron density, cuspiness, and X-ray concentration are described as unimodal and continuous, with strong cool-core and non-cool-core systems occupying the extremes rather than defining disjoint classes (Lehle et al., 2023). Using central cooling time within $z=0$3 as the fiducial diagnostic, the $z=0$4 fractions are reported as $z=0$5, $z=0$6, and $z=0$7 (Lehle et al., 2023).

The suite also predicts a thermodynamic history in which the ICM had a cooler past. Tracking the main progenitors of the 352 clusters since $z=0$8, the cool-gas study finds that the cool ICM mass increases with redshift at fixed cluster mass, that lower-mass and higher-redshift clusters are more susceptible to cooling, and that kinetic SMBH feedback progressively shifts the temperature distribution, reducing the cool ICM content from the inside out (Rohr et al., 2024). At low redshift, most cool ICM gas lies in and around satellites, whereas at $z=0$9 filamentary accretion also contributes significantly (Rohr et al., 2024). The same work relates the decline of cool ICM mass since $M_{200c}\ge 10^{15}\,M_\odot$0 to both the fall in the number of gas-rich satellites and the cumulative action of kinetic SMBH feedback (Rohr et al., 2024).

A broader census of cool gas within clusters, using the combined TNG300+TNG-Cluster sample of 632 clusters at $M_{200c}\ge 10^{15}\,M_\odot$1, finds that cool gas with $M_{200c}\ge 10^{15}\,M_\odot$2 is present in every cluster within $M_{200c}\ge 10^{15}\,M_\odot$3, but usually constitutes only a small fraction of the total gas mass, from $M_{200c}\ge 10^{15}\,M_\odot$4 up to a few per cent (Staffehl et al., 3 Mar 2025). Most of this cool gas lies in the outskirts and is bound to infalling satellites and other halos; only a minority of clusters host substantial central cool reservoirs, and those are preferentially associated with present-day cool cores (Staffehl et al., 3 Mar 2025). Taken together, these results define a TNG-Cluster picture in which hot-gas thermodynamics, multiphase structure, and cool-core status are tightly linked but are not reducible to a single bimodal core taxonomy.

4. Gas kinematics, AGN feedback, and structure formation in the ICM

The suite has been used extensively to map ICM kinematics from the innermost core to the virial boundary. In the “atlas of gas motions,” radial-velocity probability distributions in cores and intermediate regions are reported as approximately Gaussian and centered near zero, whereas the outskirts show asymmetric, double-peaked distributions in which the second peak traces cosmic accretion (Ayromlou et al., 2023). Velocity structure functions are close to Kolmogorov-like slopes over most radii, but steepen significantly within $M_{200c}\ge 10^{15}\,M_\odot$5, indicating more disturbed small-scale core dynamics (Ayromlou et al., 2023). A dedicated Reynolds-decomposition analysis of 352 clusters at $M_{200c}\ge 10^{15}\,M_\odot$6 further separates coherent bulk motions from turbulence and finds that, in cluster centers, turbulence contributes less than half of the total velocity dispersion for most clusters, with typical turbulent dispersions of $M_{200c}\ge 10^{15}\,M_\odot$7–$M_{200c}\ge 10^{15}\,M_\odot$8 and sub-percent turbulent pressure support (Saha et al., 3 Jun 2026). The same study identifies a characteristic “U-shaped” radial profile in which $M_{200c}\ge 10^{15}\,M_\odot$9 peaks in the center, reaches a minimum at $10^{14.3}$0–$10^{14.3}$1, and rises again toward the outskirts, with SMBH feedback as the principal driver of core turbulence (Saha et al., 3 Jun 2026).

Forward modeling for XRISM-like spectroscopy shows that this kinematic complexity has direct observational consequences. In Perseus-like TNG-Cluster systems, mock Resolve spectra yield typical core line-of-sight velocity dispersions of $10^{14.3}$2, consistent with subsonic turbulence and with turbulent pressure fractions below $10^{14.3}$3 of the thermal pressure (Truong et al., 2023). Yet the same clusters often host high-velocity, anisotropic outflows and bulk motions of small amounts of super-virial hot gas moving at up to thousands of $10^{14.3}$4, so integrated line widths do not isolate true turbulence without additional spatial or kinematic filtering (Truong et al., 2023). This point is sharpened in the bulk-versus-turbulence study, which argues that using total line broadening as a proxy for turbulence can overstate turbulent energy by a factor of about four in cluster cores because the typical relation is $10^{14.3}$5 (Saha et al., 3 Jun 2026).

AGN feedback-driven structure formation is another defining TNG-Cluster result. In mock Chandra images of all 352 clusters at $10^{14.3}$6, X-ray cavities arise naturally from episodic kinetic SMBH feedback, with about $10^{14.3}$7 of halos hosting cavities, a total of 233 cavities in one snapshot, and cavity morphologies that include single, paired, and multiple systems (Prunier et al., 2024). Cavities are underdense, hot, often approximately in pressure equilibrium with the local ICM, and in about one quarter of cases are bordered by bright rims associated with weak shocks (Prunier et al., 2024). A direct comparison to a volume-limited Chandra sample found similar cavity incidence, comparable sizes and morphologies, and cavity powers estimated in the observational manner of $10^{14.3}$8–$10^{14.3}$9, supporting the interpretation that cavity heating in TNG-Cluster is quantitatively realistic despite the absence of explicit bipolar relativistic jets or cosmic rays (Prunier et al., 3 Mar 2025).

The same AGN model also produces weak X-ray shocks. In a mass-matched sample of 100 simulated clusters, 50 shocks were identified in 30 systems; their Mach numbers are predominantly below 2, with median $10^{15}\,M_\odot$0, and the inferred shock powers are typically comparable to cooling luminosities of $10^{15}\,M_\odot$1–$10^{15}\,M_\odot$2 (Prunier et al., 29 Sep 2025). This suggests that, within the TNG-Cluster framework, shocks and cavities are complementary heating channels: cavities dominate nearer the SMBH, while shocks act more isotropically and at somewhat larger radii (Prunier et al., 29 Sep 2025).

Magnetic-field evolution constitutes a further layer of feedback-regulated ICM structure. TNG-Cluster predicts central magnetic-field strengths spanning roughly $10^{15}\,M_\odot$3–$10^{15}\,M_\odot$4 across the population, with SCC systems having systematically stronger central fields than WCC or NCC systems at fixed mass (Lehle et al., 16 Jul 2025). The main topological result is that $10^{15}\,M_\odot$5 cool-core clusters show preferentially tangential magnetic fields at a characteristic scale of $10^{15}\,M_\odot$6, whereas the full population is otherwise broadly isotropic at most radii and redshifts (Lehle et al., 16 Jul 2025). The proposed explanation is trapping of internal gravity waves in strongly stratified cool cores, rather than a population-wide effect of AGN outflows or merger-driven sloshing (Lehle et al., 16 Jul 2025).

5. Multiwavelength science and high-redshift applications

Because TNG-Cluster resolves both the ICM and the hierarchical assembly of rare halos, it has been used well beyond low-redshift cluster thermodynamics. In radio post-processing, the suite yields a library of approximately 300 radio relics across $10^{15}\,M_\odot$7–1, reproducing double relics, single relics, and inverted relics, and predicting that extremely large relics above $10^{15}\,M_\odot$8 occur predominantly in massive mergers with $10^{15}\,M_\odot$9 (Lee et al., 2023). A later statistical study of double relics, combining TNG-Cluster and TNG300, finds that the relic axis usually aligns with the collision axis within about $M>10^{14}\,M_\odot$0 and that the separation of a double relic pair provides a tight estimator of the time since collision,

$M>10^{14}\,M_\odot$1

with an accuracy of roughly $M>10^{14}\,M_\odot$2 (Lee et al., 24 Oct 2025). This suggests that TNG-Cluster can be used not only to populate relic catalogs but also to calibrate merger-dynamical inference.

At high redshift, TNG-Cluster extends the rare-halo statistics of TNG300 into the progenitors of present-day clusters. For AGNs and quasars at $M>10^{14}\,M_\odot$3–7, the combined TNG300+TNG-Cluster analysis finds a highly non-linear median AGN luminosity–halo mass relation that flattens or turns over at $M>10^{14}\,M_\odot$4 for at least $M>10^{14}\,M_\odot$5–6, and predicts that quasars with $M>10^{14}\,M_\odot$6–$M>10^{14}\,M_\odot$7 typically inhabit halos of $M>10^{14}\,M_\odot$8–$M>10^{14}\,M_\odot$9 at $z=0$0–6 (Kapahtia et al., 11 Jun 2026). The same work emphasizes that the scatter in luminosity at fixed halo mass is much larger than the scatter in halo mass at fixed luminosity, implying only weak coupling between halo growth and instantaneous SMBH accretion (Kapahtia et al., 11 Jun 2026).

Protocluster studies use the same rare-halo ancestry in a different way. Defining protoclusters as the set of galaxies that will reside within the virialized region of a $z=0$1 TNG-Cluster halo, one analysis shows that observational incompleteness in stellar mass or star-formation rate seriously biases recovery of the highest galaxy-density peak at $z=0$2: stellar-mass-limited and SFR-limited samples recover the true peak within $z=0$3 in $z=0$4 of cases (Baxter et al., 4 Apr 2025). The same study finds that at $z=0$5 the highest galaxy number-density peaks generally do not coincide with the highest stellar-mass or dark-matter density peaks, and argues that apertures of about $z=0$6 arcmin are needed to capture the progenitors of the most massive present-day clusters, rather than the $z=0$7 arcmin apertures commonly used in spectroscopic confirmation (Baxter et al., 4 Apr 2025). TNG-Cluster therefore functions both as a low-redshift cluster laboratory and as a controlled progenitor ensemble for pre-virialized environments.

The suite has also been used to study core-state evolution itself. A population analysis of cool-core transformations identifies frequent entropy-state changes after cluster “formation,” with the typical halo undergoing about two to three transformations, most of them toward higher-entropy cores (Lehle et al., 3 Mar 2025). Mergers within roughly $z=0$8 of an entropy increase occur more often than in shuffled control samples, but repeated kinetic AGN injections are presented as the more direct mechanism by which central entropy is cumulatively raised (Lehle et al., 3 Mar 2025). This suggests that TNG-Cluster connects instantaneous AGN phenomenology, long-term core demographics, and assembly histories within a single statistical framework.

6. Data-analytic uses, observational positioning, and limitations

TNG-Cluster has also become a dataset for methodological work. In the ERGO-ML study, 7,968 intrinsic X-ray maps derived from 352 clusters, up to eight snapshots in $z=0$9, and three orthogonal projections were used to train a nearest-neighbour contrastive learning model (Chadayammuri et al., 2024). The resulting eight-dimensional latent representation organizes the population continuously from relaxed to merging systems and from centrally peaked to flatter X-ray morphologies, and supports few-percent prediction accuracy for several mass-related quantities as well as by-example analog retrieval (Chadayammuri et al., 2024). This use case treats TNG-Cluster not merely as a simulation suite but as a structured benchmark for morphology–physics inference.

Relative to earlier cluster simulations, the suite occupies a specific niche: it combines the statistical breadth of a large parent volume with TNG300-1-like baryonic resolution, ideal MHD, and an unchanged IllustrisTNG physical model (Nelson et al., 2023). That combination allows direct comparison across wavelengths and redshifts, but it also fixes the model assumptions. Multiple TNG-Cluster studies note that the baseline implementation does not include cosmic rays, anisotropic thermal conduction, or other plasma microphysics, and that turbulence, cold clouds, and small-scale core structure are only partially resolved at the adopted mass resolution (Prunier et al., 2024). Analyses that depend on filtering choices, deprojection, or forward modeling—such as the bulk–turbulence decomposition or XRISM spectral inference—explicitly show sensitivity to methodology and projection, even when the ensemble-level trends remain robust (Saha et al., 3 Jun 2026).

A broader implication is that TNG-Cluster is best interpreted as a controlled, high-statistics realization of the IllustrisTNG galaxy-formation model in the cluster regime. Its central-property distributions are often continuous rather than bimodal, its feedback-generated structures arise without explicit relativistic jets or cosmic rays, and its agreement with observations is commonly strongest in population-level scaling relations, radial trends, and incidence fractions rather than in every detailed morphology (Lehle et al., 2023). This suggests that the suite is especially valuable for questions of demographic structure, covariance among observables, and causal plausibility across assembly history, AGN activity, and ICM response.