TNG50: High-Resolution Galaxy Simulations

Updated 28 July 2025

TNG50 simulations are the highest-resolution realization in the IllustrisTNG suite, capturing the detailed evolution of dark matter, cosmic gas, stars, and supermassive black holes in a ~52 Mpc³ volume.
They employ the AREPO moving-mesh code to solve magnetohydrodynamics and integrate comprehensive galaxy formation physics, including stellar and black hole feedback mechanisms.
The simulation produces extensive data products and synthetic observables that enable robust statistical studies of galaxy structure, kinematics, and the multiphase circumgalactic medium.

The TNG50 simulations represent the highest-resolution realization within the IllustrisTNG suite, designed to model the coupled cosmological evolution of dark matter, cosmic gas, stars, and supermassive black holes while incorporating a modern treatment of galaxy formation physics and magnetohydrodynamics. The simulation’s key traits—its spatial fidelity, extensive baryonic physics, broad statistical reach, and accessible data infrastructure—make it a central reference for the paper of the internal structure, dynamics, and multi-component environments of galaxies across cosmic time.

1. Simulation Scope, Numerical Setup, and Data Products

TNG50 simulates a $\sim(51.7\,\mathrm{Mpc})^3$ comoving volume, employing $2160^3$ (gas and dark matter) resolution elements, attaining baryon mass resolution of $8.5\times10^4\,M_\odot$ and gas cell sizes routinely reaching 70–140 pc in star-forming regions (Nelson et al., 2018). Such resolution is more than two orders of magnitude finer in mass than TNG100 or TNG300, enabling detailed modeling of sub-kpc galaxy structure unresolved in prior cosmological boxes (Pillepich et al., 2019).

The simulation is run with the AREPO moving-mesh code, which advances the fully coupled equations of self-gravitating, ideal magnetohydrodynamics (MHD), outlined by conservation equations for mass, momentum (including Lorentz and Maxwell stresses), energy, and the evolution of magnetic fields with second order accuracy in both space and time.

A comprehensive model for galaxy formation is implemented, including:

Primordial and metal-line radiative cooling within a time-varying UV background
Stochastic star formation for gas above a density threshold and multiphase ISM pressurization
Stellar feedback (Type Ia/II SNe, AGB mass loss), chemodynamics, and galactic–scale outflows
Black hole growth with both “quasar” (thermal) and “kinetic wind” feedback modes
MHD seeding and amplification from a primordial seed field.

Outputs include: 100 full snapshots across $z=127$ to 0; “mini” snapshots for focused time sampling; satellite/group catalogs (FoF/Subfind); merger trees (SubLink/LHaloTree); high-time-resolution subboxes; auxiliary catalogs with tracer particles, stellar circularities, angular momenta, and synthetic observables (Nelson et al., 2018).

2. Galaxy Structural and Kinematical Evolution

TNG50’s refinement allows for statistically robust paper of galaxy disks and morphologies, previously only feasible in zoom-ins (Pillepich et al., 2019).

Structural properties (e.g., stellar half-mass and half-light radii, disk heights/thicknesses) are measured for both stellar and star-forming gas components. Stellar disks down to $<1$ kpc radii and vertical heights below 300 pc are routinely resolved; gaseous disks are even thinner and flatter.
Disk “flatness” and intrinsic 3D oblate shapes become more prevalent with both increasing mass and cosmic time ( $z\lesssim5$ ), as quantified by sphericity and intermediate-to-major axis ratios.
The kinematic transition from dispersion-dominated (low $V/\sigma$ ) structures at high redshift to dynamically “cold” (high $V_{\max}/\sigma > 2$ –10) settled disks at $z\sim0$ is reproduced for star-forming gas. Stellar kinematics retain higher dispersions (lower $V/\sigma$ ), reflecting collisionless dynamics and past heating events.
Evolutionary trends are robust across mass and redshift: gas disks are more rotationally supported and thinner than stellar ones, and disk settling is directly predicted by a decline in velocity dispersion over time.

Comparisons with realistic mock observations and integral-field data indicate that simulated galaxies exhibit considerable line-of-sight velocity asymmetries and non-circular motions. The typical central dark matter fraction within $R_e$ at $z=2$ is $f^v_\mathrm{DM}(<R_e)\sim0.32\pm0.10$ , in reasonable agreement with—but systematically higher than—observed high- $z$ disks when analyzed at fixed $R_e$ (Übler et al., 2020).

3. Feedback, Outflows, and the Multi-phase CGM

The feedback model in TNG50 employs locally prescribed, simple subgrid recipes for both stellar and AGN-driven outflows, yet complex multiphase wind structures and scaling emerge at galactic/halo scales (Nelson et al., 2019).

Supernova (SN) feedback drives kinetic winds where the launch speed scales with local dark matter velocity dispersion, and the wind mass-loading parameter is set by local gas metallicity and prescribed energy conversion: $\eta_M^{\mathrm{SN}} = 2(1-\tau_w)e_w/v_w^2$ .
Black hole feedback operates thermo- or kinetically, with energetic injection rates set by local accretion physics and Eddington ratio. Massive/quenching galaxies ( $M_\star\gtrsim10^{10.5}\,M_\odot$ ) are dominated by slow, kinetic-mode AGN feedback, producing extremely fast multi-phase outflows exceeding $3000$ km/s and significant mass-loading out to tens of kiloparsecs.
Outflows are observed to become naturally collimated (bipolar, despite isotropic injection), with opening angles and structure set by hydrodynamic interaction with the ISM/CGM.

In the CGM, TNG50 robustly predicts the emergence of abundant, cold ( $T\sim10^4$ K) cloudlets, which are stabilized more by magnetic ( $P_B=B^2/8\pi$ ) than thermal ( $P_{\rm gas}$ ) pressure ( $\beta\ll1$ ) (Nelson et al., 2020, Ramesh et al., 2023). These clouds, often clustered near satellites or in situ cooled by strong local overdensities, have properties and covering fractions resembling those measured in absorption-line surveys. The simulation demonstrates that bulk thermal instabilities and ram-pressure stripping both contribute to this multiphase CGM structure.

4. Barred Galaxies, Pattern Speeds, and Disk–Halo Interaction

TNG50 resolves bar formation and evolution in unprecedented detail, enabling studies of cosmic and mass-dependent bar frequency, morphology, and dynamical connection to the DM halo (Ansar et al., 2023, Semczuk et al., 15 Jul 2024, Gonçalves et al., 24 Jun 2025).

Bar strength ( $A_2/A_0$ ) is measured via Fourier decomposition; bar pattern speed evolution ( $\Omega_p$ ) is tracked in the presence of gas inflow, AGN feedback, and halo properties.
Strongly barred galaxies at $z=0$ preferentially inhabit halos with low central spin ( $\lambda$ ), showing a robust anti-correlation: $A_2/A_0 > 0.4$ is linked to lower $\lambda$ and less DM angular momentum on 5–10 kpc scales. This trend weakens at higher redshift ( $z=1$ ).
Bars lose pattern speed over time via dynamical friction with the central halo. The presence of cold gas (suppressed by AGN feedback in massive galaxies) can slow or avert this decline.
A dark matter “shadow bar” forms, closely aligned with the stellar bar, confirming angular momentum transfer from disk to halo.
Mock imaging across IR/optical/UV bands demonstrates that bar length and ellipticity increase in bluer wavelengths, primarily in star-forming galaxies, reflecting the predominance of young stellar populations at small bars scales (Gonçalves et al., 24 Jun 2025).

5. Morphological and Environmental Diversity: Satellites, Disk Thickness, and Planarity

TNG50 enables a statistical census of galaxy morphology, satellite abundance, and structural features down to the dwarf regime.

The simulation predicts a strong positive correlation between rotational support (e.g., $\kappa_{\mathrm{rot}}$ ) and stellar mass: dwarfs (log $\,M_\star/M_\odot<9$ ) are dispersion-supported, clump-dominated ( $>$ 80% of stars in $r<1$ kpc), and show nearly constant half-light radii, while more massive systems form prominent rotationally supported disks (Celiz et al., 2 May 2025).
Star formation in dwarfs occurs predominantly in unresolved baryonic clumps—a likely numerical artifact due to subgrid equation-of-state and SN wind implementation—underscoring the need for caution when interpreting low-mass galaxy morphologies.
TNG50 robustly reproduces observed abundance and diversity of satellite galaxies around MW/M31 analogs, with no missing satellites problem. Satellite number correlates with host mass, luminosity, and assembly history, and the presence of Magellanic Cloud analogs emerges naturally (Engler et al., 2021, Pillepich et al., 2023).
Statistically, thin disk fractions and apparent galaxy shapes, as measured in synthetic Hyper Suprime-Cam images, are in precise agreement with real HSC-SSP data at $M_\star>10^{9.5}\,M_\odot$ —contradicting prior claims of a “thin disk deficit” in $\Lambda$ CDM (Xu et al., 27 Jul 2024). At lower masses, simulated disks appear thicker due to numerically-induced dynamical heating.
For satellite systems, roughly 11% (up to 27% in lower resolution) exhibit coherent, thin, planar structures (mean heights $\sim$ 5.2 kpc); such planes have similar fractions as observationally estimated. In-plane satellites have longer formation times and more active ISM cycles than out-of-plane counterparts (Caiyu et al., 17 Dec 2024).

6. Observational Predictions: Luminosity Functions, Tully-Fisher Relation, and Synthetic Imaging

TNG50 is extensively post-processed with the SKIRT radiative transfer code, enabling direct comparison of synthetic galaxy photometry and spectral observables to data (Trcka et al., 2022, Baes et al., 28 Feb 2025).

Full SEDs and broadband luminosities (UV to submm) are predicted for thousands of TNG50 galaxies. Calibration to the DustPedia sample ensures that model parameters (e.g., dust–to–metal ratio, PDR clearing timescales) reproduce observed luminosity/SED scaling relations. Modest discrepancies at the faint/bright LF ends are sensitive to aperture and orientation choices.
The multi-wavelength Tully–Fisher relation steepens and tightens from NUV ( $a = -7.46\pm0.14$ mag dex $^{-1}$ , $\sigma_{\perp, {\rm FUV}} \sim 0.116$ ) to IRAC $a = -9.66\pm0.09$ $a = - 9.66 \pm 0.09$ mag dex $^{-1}$ , $\sigma_{\perp, [4.5]} \sim 0.045$ " title="" rel="nofollow" data-turbo="false" class="assistant-link">4.5. Incorporating secondary trends (e.g., $u-r$ colour or sSFR dependence) further reduces scatter, with the modified relation having constant intrinsic tightness across UV–MIR bands (Baes et al., 28 Feb 2025).
Luminosity functions in 14 bands match observations at the “knee” to within $\pm0.04$ dex; discrepancies are largest in UV and at extreme luminosity ends. A lower-resolution run (TNG50-2, 8 $\times$ coarser) agrees better with data, reflecting the original calibration basis of subgrid models.

7. Data Access, Analysis Tools, and Scientific Cautions

All TNG50 data—including full and mini snapshots, group catalogs, merger trees, tracer and synthetic observables—are public, with cumulative volume exceeding 1.1 PB (Nelson et al., 2018). Web-based APIs allow fine-grained data queries: users can retrieve cutouts, search catalogs by arbitrary constraints, follow merger tree branches, and obtain on-demand visualizations (e.g., gas/stellar density maps, scaling relation plots). A cloud-hosted JupyterLab interface offers a remote, browser-based, near-native analytical environment with direct access to the full simulation data, eliminating the need to download multi-terabyte files.

Key cautions for TNG50 data users:

Even at this resolution, ultra-faint dwarf galaxies or nuclear substructure ( $M_\star\lesssim10^5~M_\odot$ ) remain unresolved. Results at the lowest-mass and smallest-scale regimes require care.
Subhalo identification in environments with strong baryonic perturbations can yield artifacts (e.g., transient disk clumps misclassified as satellites), flagged in public catalogs.
Gas temperature corrections were required for underdense IGM (density-dependent correction) and cells near the cooling floor (reset to $10^4$ K), applied at $z\leq 5$ .
Convergence varies with resolution; numerical artifacts (e.g., “clump-disc” star formation in dwarfs, as above) may affect the lowest-mass/size analyses. Users must account for resolution-related systematics in science applications (Celiz et al., 2 May 2025, Trcka et al., 2022).

In summary, TNG50 establishes a new standard for large-volume, high-fidelity galaxy formation simulations within a $\Lambda$ CDM cosmology. By integrating nuanced baryonic physics, MHD, and robust data/analysis infrastructure, it enables precise, physically motivated, and statistically meaningful studies of galaxy and circumgalactic structure and evolution. The public availability of the full dataset, comprehensive synthetic observations, and modular query tools allows the broader community to rigorously test, challenge, and refine models of cosmic structure formation.