FLAMINGO Hydrodynamic Simulation Suite

• FLAMINGO is a cosmological hydrodynamic simulation suite that models large-scale structure using full baryonic physics, explicit massive neutrinos, and Gpc-scale volumes.
• It integrates matched hydrodynamic and gravity-only runs to accurately capture baryon effects on matter clustering, weak lensing, and galaxy cluster observables.
• The suite employs advanced numerical methods and machine-learning calibration to reproduce key observables such as the galaxy stellar mass function, SZ, and CMB anisotropies.

FLAMINGO is a suite of large-volume cosmological hydrodynamical simulations constructed for late-time large-scale structure, cluster cosmology, weak lensing, and survey-scale mock observables. It combines full baryonic physics, explicit massive neutrinos, matched gravity-only counterparts, and Gpc-scale volumes, and it has been used to study the matter power spectrum, galaxy clusters, intrinsic alignments, weak-lensing peaks, quasars, high-redshift quiescent galaxies, and secondary cosmic microwave background anisotropies (Helly et al., 27 Apr 2026).

1. Scientific scope and design philosophy

FLAMINGO is designed as a “late-time Universe” cosmological laboratory with three stated aims: to predict large-scale structure observables at the precision required by Euclid, LSST, DESI, and CMB experiments; to model baryonic effects self-consistently in large volumes; and to explore cosmology–baryon degeneracies, including massive neutrinos and decaying dark matter (Helly et al., 27 Apr 2026). This places the suite in the regime of survey-scale hydrodynamics rather than small-box, high-resolution galaxy formation experiments.

That design choice is central to its scientific role. The suite targets halo statistics, baryon backreaction on clustering, cluster gas physics, weak lensing, SZ and X-ray observables, CIB, radio foregrounds, and lightcone-based synthetic skies. In this sense, FLAMINGO is both a simulation campaign and a mock-observable engine. A recurring methodological theme is that the same hydrodynamical volume is used to generate multiple observables from the same underlying matter, gas, stellar, and black-hole distributions, rather than stitching together independent templates (Yang et al., 10 Dec 2025).

The suite is also explicitly built around matched hydrodynamical and gravity-only realizations. That pairing underlies several of its principal uses, most notably the baryonic response of the matter power spectrum and the construction of baryonic transfer functions for weak-lensing and CMB observables (Schaller et al., 2024).

2. Numerical framework and calibration

All FLAMINGO simulations use the open-source SWIFT code. Gravity is solved with a fourth-order fast multipole method for short-range forces coupled to a particle–mesh scheme for long-range forces, while hydrodynamics is evolved with the SPHENIX flavour of smoothed particle hydrodynamics (Helly et al., 27 Apr 2026). Massive neutrinos are simulated explicitly as particles with the $\delta f$ method, and initial conditions are generated with blackMonofonIC using higher-order LPT and a three-fluid formalism with separate transfer functions for CDM, baryons, and neutrinos (Helly et al., 27 Apr 2026).

The subgrid model includes radiative cooling and photo-heating, star formation, stellar evolution and chemical enrichment, supernova feedback, black-hole seeding and growth, and AGN feedback. The suite includes both a thermal AGN implementation and jet-mode runs in which feedback is injected as collimated bipolar outflows (Helly et al., 27 Apr 2026). Across the published analyses, these modules are described as EAGLE-like in spirit, but they are recalibrated for FLAMINGO’s resolution and cosmological objectives (Baker et al., 2024).

Calibration is performed with machine-learning methods. Four subgrid parameters associated with stellar and AGN feedback are tuned to reproduce the $z=0$ galaxy stellar mass function and low-redshift cluster gas fractions (Helly et al., 27 Apr 2026). This is a weak-convergence strategy: each numerical resolution is calibrated independently to the same observational targets, so comparisons between m8, m9, and m10 are not pure resolution tests of a fixed physical model. A direct implication is that high-redshift predictions, especially at the massive end, are effectively extrapolations away from the low-redshift calibration regime (Baker et al., 2024).

The public suite adopts a fiducial DES-Y3 cosmology, denoted “D3A,” with $h=0.681$, $\Omega_{\rm m}=0.306$, $\Omega_\Lambda=0.694$, $\Omega_{\rm b}=0.0486$, $\sum m_\nu c^2=0.06\,\mathrm{eV}$, $A_s=2.099\times10^{-9}$, $n_s=0.967$, $\sigma_8=0.807$, and $z=0$0 (Helly et al., 27 Apr 2026). It also includes Planck-like cosmologies, low-$z=0$1 variants, multiple neutrino-mass choices, and decaying-dark-matter runs (Helly et al., 27 Apr 2026).

3. Simulation suite and released data

The public FLAMINGO release contains 22 hydrodynamical simulations and 16 gravity-only simulations, amounting to more than 2.3 petabytes of data (Helly et al., 27 Apr 2026). The fiducial hydrodynamical suite spans three numerical resolutions—m8, m9, and m10—and a flagship large-volume run at m9 resolution.

Run	Volume and particles	Mass resolution and softening
L1_m8	$z=0$2; $z=0$3; $z=0$4	$z=0$5; $z=0$6; $z=0$7; $z=0$8
L1_m9	$z=0$9; $h=0.681$0; $h=0.681$1	$h=0.681$2; $h=0.681$3; $h=0.681$4; $h=0.681$5
L1_m10	$h=0.681$6; $h=0.681$7; $h=0.681$8	$h=0.681$9; $\Omega_{\rm m}=0.306$0; $\Omega_{\rm m}=0.306$1; $\Omega_{\rm m}=0.306$2
L2p8_m9	$\Omega_{\rm m}=0.306$3; $\Omega_{\rm m}=0.306$4; $\Omega_{\rm m}=0.306$5	same $\Omega_{\rm m}=0.306$6, $\Omega_{\rm m}=0.306$7, and softening as L1_m9

The released products include snapshots, halo and galaxy catalogues, HEALPix all-sky lightcone maps, particle lightcones, halo lightcones, and on-the-fly power spectra (Helly et al., 27 Apr 2026). Full hydro and DMO snapshots are available at a reduced set of redshifts for most runs, while L1_m9 and its DMO companion retain the full 78–79 snapshot sequence (Helly et al., 27 Apr 2026). Lightcone products extend to $\Omega_{\rm m}=0.306$8 or $\Omega_{\rm m}=0.306$9 for maps and to $\Omega_\Lambda=0.694$0 for halo lightcones (Helly et al., 27 Apr 2026).

Halo and galaxy post-processing is pipeline-dependent across analyses. The public release provides HBT-HERONS merger trees and SOAP halo catalogues, while some cluster-focused studies use VELOCIraptor plus SOAP (Helly et al., 27 Apr 2026). This distinction matters because some published results depend explicitly on the adopted halo or subhalo definitions.

Access is organized around remote, selective reading rather than bulk download. The release provides a web interface, the hdfstream API, and direct integration with swiftsimio, including spatial cutouts and remote HDF5-style access (Helly et al., 27 Apr 2026). That architecture reflects the practical reality that the full data volume is too large for most local-storage workflows.

4. Matter clustering, lensing, intrinsic alignments, and full-sky observables

A central FLAMINGO quantity is the baryonic response of the matter power spectrum,

$\Omega_\Lambda=0.694$1

Using matched hydro/DMO runs, FLAMINGO finds that this response is only well converged for simulation volumes in excess of $\Omega_\Lambda=0.694$2, and it packages the result in a Gaussian-process emulator that reproduces the simulations to better than one per cent up to $\Omega_\Lambda=0.694$3 and up to $\Omega_\Lambda=0.694$4 for the feedback models present in the suite (Schaller et al., 2024). The response strengthens toward low redshift, affects progressively larger scales, and is enhanced by lower gas fractions, lower stellar masses at fixed gas fraction, and jet-mode AGN feedback (Schaller et al., 2024).

The suite has also been used to calibrate fast predictions for non-linear CMB lensing in the presence of baryons and neutrinos. In that context, FLAMINGO hydrodynamic runs show that baryonic suppression can reduce the lensing power by of order 10%, and the combination of perturbation theory, matter-power emulation, and an FLAMINGO-calibrated suppression model reproduces the hydrodynamic result to 4% at $\Omega_\Lambda=0.694$5 (Upadhye et al., 2023). A notable theoretical result from those tests is that the scale-dependent suppression from neutrinos and from baryons approximately factorizes (Upadhye et al., 2023).

At the map level, FLAMINGO produces self-consistent full-sky realizations of CMB lensing, thermal and kinetic SZ, CIB, radio point sources, and anisotropic screening. The tSZ field is defined through the standard Compton-$\Omega_\Lambda=0.694$6 line-of-sight integral,

$\Omega_\Lambda=0.694$7

and the resulting mock skies reproduce a wide range of observational constraints while preserving cross-correlations among components that are generated from the same hydrodynamical realization (Yang et al., 10 Dec 2025).

Weak-lensing peak statistics provide another direct application. In Euclid-like convergence maps, FLAMINGO shows that SNR$\Omega_\Lambda=0.694$8 peaks primarily trace $\Omega_\Lambda=0.694$9 haloes, and that the shape of the redshift distribution of those peaks is insensitive to baryonic physics but does change with cosmology (Broxterman et al., 2024). This is important because the peak-height distribution and the peak-redshift distribution scale differently with cosmology and baryonic feedback, so the two observables can be combined to constrain cosmology while calibrating baryonic systematics (Broxterman et al., 2024).

On the galaxy-shape side, FLAMINGO has been used to measure intrinsic alignments for a 4,941,492-galaxy $\Omega_{\rm b}=0.0486$0 LRG-like sample in the 2.8 Gpc hydrodynamic box. Both NLA and TATT fit the measured two-point statistics well, but a mass-dependent TATT model, TATT-M, is found to be very strongly preferred over NLA, and feedback variations do not significantly change the alignment amplitude once halo mass is controlled for (Herle et al., 22 Jan 2026).

5. Clusters, galaxies, and black holes

Cluster astrophysics is one of the most mature FLAMINGO applications. The suite contains unprecedented numbers of massive systems, with more than $\Omega_{\rm b}=0.0486$1 groups and more than $\Omega_{\rm b}=0.0486$2 clusters in the mass range used for comparison with X-ray observations (Braspenning et al., 2023). In the fiducial model, the temperature, density, pressure, and entropy profiles of clusters are in excellent agreement with observations, while metallicities in the core are too high (Braspenning et al., 2023). The thermodynamic profiles show almost no evolution relative to self-similar expectations, except that metallicity decreases with redshift (Braspenning et al., 2023).

The feedback variations elucidate how FLAMINGO’s subgrid physics maps into cluster structure. Stronger feedback calibrated to lower gas fractions yields higher temperatures, entropies, and metallicities but lower pressures; compared to thermally driven AGN feedback, kinetic jet feedback calibrated to the same gas fraction at $\Omega_{\rm b}=0.0486$3 produces a hotter core with higher entropies and lower densities, and therefore a smaller fraction of cool-core clusters (Braspenning et al., 2023). These results are central because the cluster-gas calibration anchors much of FLAMINGO’s baryonic response on cosmological scales.

FLAMINGO also resolves the AGN and quasar population at survey volumes. In the $\Omega_{\rm b}=0.0486$4 L2p8_m9 run, it reproduces the observed quasar luminosity function at low redshift and for faint quasars, and it reproduces the observed clustering across $\Omega_{\rm b}=0.0486$5 for $\Omega_{\rm b}=0.0486$6 (Ding et al., 28 Oct 2025). At the same time, it significantly underpredicts the abundance of bright quasars at $\Omega_{\rm b}=0.0486$7–3, and the bright-end boost obtained by adding a 0.75 dex log-normal luminosity scatter is driven primarily by low-mass black holes radiating above the Eddington limit in the post-processed luminosities (Ding et al., 28 Oct 2025). The simulation also underpredicts quasar clustering at $\Omega_{\rm b}=0.0486$8, consistent with other models (Ding et al., 28 Oct 2025).

6. High-redshift tests, tensions, and documented caveats

FLAMINGO’s large volumes make it possible to test rare high-redshift systems. For protoclusters during the epoch of reionization, the suite predicts that, when considering galaxies more massive than $\Omega_{\rm b}=0.0486$9, progenitors similar to those of the Coma cluster make a dominant contribution to the cosmic star-formation rate density at high redshift, and that the onset of suppression of star formation in protocluster environments occurs as early as $\sum m_\nu c^2=0.06\,\mathrm{eV}$0 (Lim et al., 2024). The same analysis shows that aperture choices strongly bias inferred integrated properties: total masses can be biased by factors of 2 to 5 at $\sum m_\nu c^2=0.06\,\mathrm{eV}$1–7 and by an order of magnitude at $\sum m_\nu c^2=0.06\,\mathrm{eV}$2, which is sufficient to remove the reported $\sum m_\nu c^2=0.06\,\mathrm{eV}$3 tension with the number density of structures found in recent JWST observations (Lim et al., 2024).

A much sharper tension appears in the abundance of massive quiescent galaxies. Using the L1_m8 hydrodynamical run as the theoretical counterpart to JADES spectroscopy, FLAMINGO underpredicts the number density of massive quiescent galaxies by about a factor of 2 at $\sum m_\nu c^2=0.06\,\mathrm{eV}$4 and by nearly an order of magnitude at $\sum m_\nu c^2=0.06\,\mathrm{eV}$5, with cosmic variance ruled out as the sole explanation at greater than 3$\sum m_\nu c^2=0.06\,\mathrm{eV}$6 (Baker et al., 2024). For the quiescent systems that do form, FLAMINGO predicts that most of their stellar mass formed in situ, with ex-situ fractions remaining below about 40%, and only about 1.5 major dry mergers by $\sum m_\nu c^2=0.06\,\mathrm{eV}$7 and about 1 by $\sum m_\nu c^2=0.06\,\mathrm{eV}$8 (Baker et al., 2024). In the interpretation advanced in that study, the discrepancy points to missing or miscalibrated physics in the efficiency and timing of star formation and AGN feedback at $\sum m_\nu c^2=0.06\,\mathrm{eV}$9 (Baker et al., 2024).

The suite also comes with documented technical caveats. The data-release paper emphasizes that each resolution level is independently recalibrated, so comparisons between m8, m9, and m10 are comparisons of different effective physical models, not the same model at different resolution (Helly et al., 27 Apr 2026). It also lists known issues, including a black-hole repositioning bug in most m9 non-Jet runs, a star-formation-threshold bug in m9 non-Jet runs for primordial gas, the use of gas-only anchors for BH recentering, thermal AGN feedback artifacts in snapshot-derived thermodynamic quantities, and a corrected scale-factor error in the original kSZ and dispersion-measure HEALPix maps (Helly et al., 27 Apr 2026). These caveats do not negate the scientific reach of FLAMINGO, but they are part of the proper interpretation of specific runs and derived catalogues.

Taken together, FLAMINGO occupies a distinctive position among contemporary simulation suites: it is explicitly optimized for large-scale, survey-facing cosmology with full hydrodynamics, yet it is rich enough to expose where current subgrid prescriptions succeed, where they remain degenerate, and where they fail under direct confrontation with new data.