FLAMINGO Cosmological Simulations

• FLAMINGO simulations are a suite of very large hydrodynamical models that integrate galaxy formation physics, cluster analysis, and survey observables.
• They employ cutting-edge techniques including SWIFT and SPHENIX with Gaussian-process emulator calibration to robustly model baryonic feedback and cosmological variations.
• The resource provides extensive data products and forward-modeling capabilities for studying cosmic phenomena like weak lensing, cluster scaling relations, and anisotropy tests.

FLAMINGO is the Virgo Consortium’s Full-hydro Large-scale structure simulations with All-sky Mapping for the Interpretation of Next Generation Observations: a suite of very large cosmological hydrodynamical simulations built to connect galaxy formation physics, hot gas in groups and clusters, and survey-scale large-scale-structure observables within a single calibrated framework. The simulations are run with SWIFT, use the SPHENIX hydrodynamics scheme, include explicit neutrino particles, and couple baryonic subgrid models to observational calibration targets such as the present-day galaxy stellar mass function and low-redshift cluster gas fractions. FLAMINGO was designed simultaneously as a theoretical laboratory and as a survey-facing simulation backbone for cosmology, clusters, baryonic feedback, and galaxy formation (Schaye et al., 2023, Helly et al., 27 Apr 2026).

1. Origins and scientific rationale

FLAMINGO was introduced to address a specific regime of modern cosmology in which baryonic physics can no longer be treated as a small perturbation. On the scales probed by weak lensing, CMB lensing, Sunyaev–Zel’dovich measurements, cluster abundances, and related large-scale-structure observables, feedback from star formation and active galactic nuclei redistributes gas, alters halo masses, changes cluster gas fractions, and suppresses both the halo mass function and the matter power spectrum by up to about $20\%$ (Schaye et al., 2023). The project therefore combines very large simulation volumes with full hydrodynamics and a controlled set of astrophysical and cosmological variations.

The fiducial cosmology adopted in the main FLAMINGO papers is based on DES Y3 constraints and is quoted as

$$\Omega_{\rm m}=0.306,\quad \Omega_{\rm b}=0.0486,\quad \sigma_8=0.807,\quad H_0=68.1,\quad n_s=0.967,$$

with a minimal neutrino mass of $0.06$ eV (Kugel et al., 2023). This cosmological setup anchors a suite intended not only to reproduce low-redshift calibration data, but also to propagate physically plausible baryonic uncertainty into survey observables.

A defining design principle is that FLAMINGO is not a single tuned run. It is a family of runs spanning multiple numerical resolutions, feedback calibrations, cosmologies, and dark-matter models. That structure is central to the project’s role in precision cosmology, because the same framework is used to test cluster scaling relations, harmonic-space large-scale-structure observables, weak-lensing peaks, X-ray and SZ cross-correlations, rare high-redshift galaxy populations, quasar statistics, and full-sky secondary CMB foregrounds (Schaye et al., 2023).

2. Numerical framework and calibration strategy

The suite is organized around three numerical resolutions: m8, m9, and m10, whose baryonic particle masses differ by factors of $8$. The flagship hydrodynamical runs are L1_m8, a $1\,\mathrm{Gpc}$ box at high resolution, and L2p8_m9, a $2.8\,\mathrm{Gpc}$ box at intermediate resolution. The latter follows $2\times 5040^3$ baryonic and dark-matter particles plus $2800^3$ neutrino particles, corresponding to approximately $3\times 10^{11}$ resolution elements, and was presented as the largest hydrodynamical cosmological simulation run to $z=0$ (Schaye et al., 2023). In the project overview, the m8, m9, and m10 baryonic particle masses are given as approximately $$\Omega_{\rm m}=0.306,\quad \Omega_{\rm b}=0.0486,\quad \sigma_8=0.807,\quad H_0=68.1,\quad n_s=0.967,$$0, $$\Omega_{\rm m}=0.306,\quad \Omega_{\rm b}=0.0486,\quad \sigma_8=0.807,\quad H_0=68.1,\quad n_s=0.967,$$1, and $$\Omega_{\rm m}=0.306,\quad \Omega_{\rm b}=0.0486,\quad \sigma_8=0.807,\quad H_0=68.1,\quad n_s=0.967,$$2, respectively, with m9 having a maximum proper softening of $$\Omega_{\rm m}=0.306,\quad \Omega_{\rm b}=0.0486,\quad \sigma_8=0.807,\quad H_0=68.1,\quad n_s=0.967,$$3 kpc (Schaye et al., 2023).

The physical model includes radiative cooling and heating, star formation, stellar mass loss and chemical enrichment, stellar feedback, black-hole growth, AGN feedback, and massive neutrinos (Schaye et al., 2023). In the machine-learning calibration paper, SWIFT is described as using the Fast Multipole Method for gravity, the SPHENIX SPH scheme, and neutrinos evolved with the $$\Omega_{\rm m}=0.306,\quad \Omega_{\rm b}=0.0486,\quad \sigma_8=0.807,\quad H_0=68.1,\quad n_s=0.967,$$4 method (Kugel et al., 2023). FLAMINGO also includes two AGN feedback implementations: a fiducial thermal model and an alternative jet-like kinetic model (Schaye et al., 2023).

Calibration is one of the project’s distinctive features. Rather than tuning by hand, FLAMINGO uses Gaussian-process emulators trained on 32-node Latin hypercube simulation sets at each resolution, with MCMC used to fit the emulator predictions to the observed $$\Omega_{\rm m}=0.306,\quad \Omega_{\rm b}=0.0486,\quad \sigma_8=0.807,\quad H_0=68.1,\quad n_s=0.967,$$5 galaxy stellar mass function and cluster gas fractions (Kugel et al., 2023). The calibration explicitly includes observational systematics such as stellar-mass bias, cosmic-variance bias, and hydrostatic mass bias. The best-fitting intermediate-resolution biases reported in the calibration study are a stellar-mass bias of $$\Omega_{\rm m}=0.306,\quad \Omega_{\rm b}=0.0486,\quad \sigma_8=0.807,\quad H_0=68.1,\quad n_s=0.967,$$6 dex, a cosmic-variance bias of $$\Omega_{\rm m}=0.306,\quad \Omega_{\rm b}=0.0486,\quad \sigma_8=0.807,\quad H_0=68.1,\quad n_s=0.967,$$7, and a hydrostatic mass bias implemented by dividing X-ray masses by $$\Omega_{\rm m}=0.306,\quad \Omega_{\rm b}=0.0486,\quad \sigma_8=0.807,\quad H_0=68.1,\quad n_s=0.967,$$8 (Schaye et al., 2023). The result is a weak-convergence strategy: the subgrid parameters differ across resolutions, but each resolution is calibrated to the same observational targets over its resolved mass range (Kugel et al., 2023).

3. Simulation families, model variations, and public release

The public FLAMINGO release describes a suite of 22 hydrodynamical simulations and 16 gravity-only simulations, totaling more than 2.3 petabytes of data (Helly et al., 27 Apr 2026). Most production runs use $$\Omega_{\rm m}=0.306,\quad \Omega_{\rm b}=0.0486,\quad \sigma_8=0.807,\quad H_0=68.1,\quad n_s=0.967,$$9 volumes at m9 resolution, while the fiducial m9 model also exists in the larger $0.06$0 box. The release also includes FLAMINGO-10k, a gravity-only run with $0.06$1 particles in a $0.06$2 box, corresponding to nearly $0.06$3 evolved particles (Helly et al., 27 Apr 2026).

A major strength of the suite is the controlled variation of both astrophysical and cosmological assumptions. Astrophysical families include gas-fraction shifts such as fgas+2sigma, fgas-2sigma, fgas-4sigma, and fgas-8sigma; stellar-mass-function shifts such as M*-sigma; combined variants such as M*-sigma_fgas-4sigma; thermal and jet AGN implementations; and a NoCooling run in which radiative cooling, star formation, and feedback are turned off (Helly et al., 27 Apr 2026). Cosmological variants include the fiducial D3A cosmology, Planck, the low-$0.06$4 LS8 cosmology, several massive-neutrino models, and decaying dark matter runs (Helly et al., 27 Apr 2026).

The data products are correspondingly broad. Full particle snapshots are released at

$0.06$5

with additional reduced and downsampled snapshots (Helly et al., 27 Apr 2026). Halo and galaxy catalogues use HBT-HERONS for subhaloes and merger trees and SOAP for halo and galaxy properties. Lightcone products comprise HEALPix all-sky maps, particle lightcones, and halo lightcones. The HEALPix maps include mass, star formation rate, weak-lensing convergence, CMB lensing, thermal SZ, kinetic SZ, anisotropic screening / optical depth, dispersion measure, cosmic infrared background, radio point source emission, and diffuse X-ray emission, released at $0.06$6 and downsampled $0.06$7 (Helly et al., 27 Apr 2026).

Because full downloads are often impractical, the release implements a web-based HDF5-serving interface that allows browsing, metadata inspection, and selective extraction of complete files, individual datasets, slices, or subsets. The service is supported by the Python client hdfstream, compatibility with swiftsimio, and a browser-based user interface (Helly et al., 27 Apr 2026). In practice, this turns FLAMINGO from a project-specific simulation campaign into a durable community resource.

4. Galaxy clusters, hot baryons, and feedback phenomenology

Galaxy groups and clusters are one of FLAMINGO’s primary design targets. In the $0.06$8 flagship run, the box contains at $0.06$9 more than two million haloes above $8$0, including $8$1 haloes above $8$2, $8$3 above $8$4, and $8$5 above $8$6 (Braspenning et al., 2023). This statistical power underlies a series of cluster-oriented studies on X-ray profiles, selection effects, cluster counts, dynamical state, and baryon–black-hole coupling.

In the comparison to X-ray observations, FLAMINGO predicts cluster scaling relations and radial profiles of temperature, density, pressure, entropy, and metallicity out to $8$7. The main result is that the temperature, density, pressure, and entropy profiles of the fiducial model are in excellent agreement with X-ray observations, while the metallicities in the core are too high by about $8$8 dex (Braspenning et al., 2023). The profiles evolve close to self-similarly once standard virial scalings are removed, with metallicity as the main exception. The same study also shows that weighting is not a minor technical choice: X-ray-weighted profiles have lower core temperatures and entropies and higher core densities than volume-weighted profiles, especially inside $8$9 (Braspenning et al., 2023).

Feedback implementation materially changes core structure. Relative to thermally driven AGN feedback calibrated to the same gas fraction at $1\,\mathrm{Gpc}$0, kinetic jet feedback produces a hotter core with higher entropy, lower density, and a cool-core fraction suppressed by more than $1\,\mathrm{Gpc}$1 (Braspenning et al., 2023). Stronger feedback calibrated to lower gas fractions raises temperatures and entropies, lowers pressures, and shifts gas-density structure to larger radii. In a separate analysis of the gas-fraction–halo-mass relation, FLAMINGO shows that the correlation between central black-hole mass and gas fraction changes sign across halo mass: it is negative for $1\,\mathrm{Gpc}$2 and positive for $1\,\mathrm{Gpc}$3, with the reversal attributed to early gas expulsion followed by re-accretion in earlier-forming systems (Costello et al., 20 Oct 2025).

The suite has also been used to quantify cluster-selection systematics. In the $1\,\mathrm{Gpc}$4 run, samples selected by X-ray luminosity, Compton-$1\,\mathrm{Gpc}$5, or richness are all skewed toward lower masses than a true mass-selected sample (Kugel et al., 2024). At $1\,\mathrm{Gpc}$6, observable cuts corresponding to median masses between $1\,\mathrm{Gpc}$7 and $1\,\mathrm{Gpc}$8 give nearly unbiased median masses for all selection methods, but X-ray selection becomes biased at higher masses. For cuts corresponding to median masses below $1\,\mathrm{Gpc}$9 at $2.8\,\mathrm{Gpc}$0 and for all masses at $2.8\,\mathrm{Gpc}$1, only Compton-$2.8\,\mathrm{Gpc}$2 selection yields nearly unbiased median masses (Kugel et al., 2024).

For cluster-count cosmology, FLAMINGO is used as a benchmark against the conventional pipeline that combines a dark-matter-only halo mass function with a power-law observable–mass relation and lognormal scatter. In SZ-selected mock samples, standard halo-mass-function prescriptions differ significantly from one another and from FLAMINGO’s dark-matter-only result, and for Simons Observatory-like surveys the study concludes that dramatic improvements are needed in the halo mass function, the functional form of the scaling relation, the treatment of scatter, and the prior on baryonic effects (Kugel et al., 2024). FLAMINGO has likewise been used to study dynamical state diagnostics: the X-ray centroid shift $2.8\,\mathrm{Gpc}$3 is found to be the most reliable proxy for recent accretion, while the radius of minimum gas-density scatter shifts outward with feedback strength and is proposed as a possible feedback diagnostic (Magnus et al., 12 Sep 2025).

5. Large-scale structure, weak lensing, and cosmological inference

A central FLAMINGO theme is the comparison between hydrodynamical predictions and low-redshift large-scale-structure observables. In harmonic-space analyses of cosmic shear, CMB lensing, and the thermal SZ effect, FLAMINGO shows that realistic baryonic feedback modifies the observables but is not sufficiently large to remove the $2.8\,\mathrm{Gpc}$4 tension (McCarthy et al., 2023). The simulations reproduce the CMB lensing power spectrum well in standard high-$2.8\,\mathrm{Gpc}$5 cosmologies, while the cosmic shear power spectrum, the tSZ power spectrum, and cross-spectra involving shear, CMB lensing, and tSZ remain in varying degrees of tension with CMB-specified $2.8\,\mathrm{Gpc}$6CDM (McCarthy et al., 2023). Lower-$2.8\,\mathrm{Gpc}$7 cosmologies and larger neutrino masses suppress power and can improve agreement for some probes, but no single adjustment simultaneously resolves cosmic shear, tSZ, and CMB lensing.

FLAMINGO also shows that baryonic suppression is not fully separable from cosmology. Defining

$2.8\,\mathrm{Gpc}$8

the simulations exhibit non-factorizable cosmology dependence at the several-percent level. For modest cosmology changes around Planck, the non-factorizable correction reaches up to $2.8\,\mathrm{Gpc}$9 on $2\times 5040^3$0, and the paper models the relevant coupling through the combination

$2\times 5040^3$1

arguing that the combined suppression from baryons and non-baryonic growth suppression is greater than the sum of its parts, especially for decaying dark matter (Elbers et al., 2024).

Several observationally matched studies push this logic into specific probes. In the comparison to Planck + ACT kinetic SZ stacking on BOSS galaxies, the fiducial FLAMINGO feedback model calibrated to X-ray gas fractions is strongly disfavored, while the kSZ data prefer substantially stronger gas ejection; this stronger feedback can help reduce the $2\times 5040^3$2 tension for cosmic shear, although it worsens agreement with X-ray group measurements and only marginally alleviates tSZ-based discrepancies (McCarthy et al., 2024). In the X-ray–cosmic-shear cross-correlation, the signal is most sensitive to haloes with $2\times 5040^3$3, but the effects of baryonic feedback and cosmology are largely degenerate, and unresolved AGN contamination becomes a decisive uncertainty in interpreting the DES–ROSAT measurement (McDonald et al., 2 Feb 2026).

Weak-lensing non-Gaussian observables exhibit a related structure. For Euclid-like convergence maps, FLAMINGO shows that high-significance peaks with $2\times 5040^3$4 primarily trace haloes with $2\times 5040^3$5; the abundance of peaks is sensitive to baryonic feedback, but the shape of the redshift distribution is largely insensitive to baryonic physics and does change with cosmology, implying a route to simultaneous feedback calibration and cosmological inference (Broxterman et al., 2024). In a separate study of the weak-lensing scattering transform, the baryonic transfer function—defined as the ratio of hydrodynamical to dark-matter-only scattering coefficients—is found to be nearly insensitive to cosmology while exhibiting feedback-driven suppression of up to $2\times 5040^3$6 on scales corresponding to $2\times 5040^3$7; however, Euclid-like shape noise reduces the baryonic signature to about $2\times 5040^3$8 even after $2\times 5040^3$9 arcmin smoothing (Marinichenko et al., 10 Oct 2025).

6. High-redshift galaxies, quasars, and rare populations

FLAMINGO’s large volumes are especially valuable for rare high-redshift populations whose observed number densities are low enough that cosmic variance itself becomes a central modeling issue. In the comparison to JWST JADES spectroscopy, the $2800^3$0 L1_m8 run is used as the theoretical benchmark for massive quiescent galaxies at $2800^3$1. The observed number densities are found to be roughly ten times higher than the mean FLAMINGO prediction in the $2800^3$2–4.5 range, and cosmic variance is ruled out at the $2800^3$3 level (Baker et al., 2024). Within the simulation, the vast majority of the stellar mass in these quiescent systems forms in situ, the ex-situ contribution stays below $2800^3$4 even for the most massive systems, and the galaxies undergo only about one and a half major dry mergers by $2800^3$5 and about one by $2800^3$6 on average, leading the authors to argue that repeated major dry mergers are not a viable generic quenching channel (Baker et al., 2024).

A related study divides the same $2800^3$7 high-resolution volume into $2800^3$8 independent $2800^3$9 sub-volumes, approximately JWST-field scale, and finds that cosmic variance substantially exceeds the Poisson expectation. At $3\times 10^{11}$0, the total variance in the number of haloes with $3\times 10^{11}$1 is $3\times 10^{11}$2–$3\times 10^{11}$3 times the Poisson expectation, and the variance in the most massive halo per JWST-like field is about twice the Poisson prediction at $3\times 10^{11}$4 (Lim et al., 12 Nov 2025). The same paper reports a pronounced large-scale conformity: ranking sub-volumes by the stellar mass of their most massive galaxy, $3\times 10^{11}$5, reveals coherent modulation of the stellar-to-halo mass relation across the full $3\times 10^{11}$6 volume. The stellar fraction of the most massive galaxies peaks at $3\times 10^{11}$7 at $3\times 10^{11}$8 (Lim et al., 12 Nov 2025).

For dusty star-forming systems, FLAMINGO is used as the main cosmological volume for modeling submillimeter galaxies through parametric relations calibrated from radiative-transfer calculations. With the H13 flux prescription, FLAMINGO reproduces observed SMG number counts and redshift distributions without invoking a top-heavy initial mass function (Kumar et al., 31 Jan 2025). In that model, SMGs with $3\times 10^{11}$9 mJy contribute up to $z=0$0 of the cosmic star-formation-rate density at $z=0$1, and the TolTEC Ultra Deep Survey is forecast to detect approximately $z=0$2 sources over $z=0$3 at $z=0$4 mm above $z=0$5, capturing about $z=0$6 of the cosmic SFRD at $z=0$7 (Kumar et al., 31 Jan 2025).

The large $z=0$8 box is likewise crucial for luminous quasars. FLAMINGO reproduces the observed bolometric quasar luminosity function at low redshift and for faint quasars with $z=0$9, but significantly underpredicts the bright end at $$\Omega_{\rm m}=0.306,\quad \Omega_{\rm b}=0.0486,\quad \sigma_8=0.807,\quad H_0=68.1,\quad n_s=0.967,$$00–3 (Ding et al., 28 Oct 2025). Adding a $$\Omega_{\rm m}=0.306,\quad \Omega_{\rm b}=0.0486,\quad \sigma_8=0.807,\quad H_0=68.1,\quad n_s=0.967,$$01 dex log-normal luminosity scatter boosts bright-quasar counts by upscattering lower-luminosity systems and improves agreement with observations, although the enhancement is then driven mainly by lower-mass black holes radiating above the Eddington limit (Ding et al., 28 Oct 2025). The same study finds that quasar clustering is reproduced across $$\Omega_{\rm m}=0.306,\quad \Omega_{\rm b}=0.0486,\quad \sigma_8=0.807,\quad H_0=68.1,\quad n_s=0.967,$$02, but the clustering strength is underpredicted at $$\Omega_{\rm m}=0.306,\quad \Omega_{\rm b}=0.0486,\quad \sigma_8=0.807,\quad H_0=68.1,\quad n_s=0.967,$$03 (Ding et al., 28 Oct 2025).

7. Lightcones, isotropy tests, and full-sky sky-model construction

From the outset, FLAMINGO was built to produce survey-like observables through on-the-fly past-lightcone generation for multiple observers (Schaye et al., 2023). This capability supports both direct mock-observation pipelines and broader tests of cosmological assumptions. One example is the isotropy analysis based on the $$\Omega_{\rm m}=0.306,\quad \Omega_{\rm b}=0.0486,\quad \sigma_8=0.807,\quad H_0=68.1,\quad n_s=0.967,$$04 L2p8_m9 run, in which 1728 simulated lightcones were constructed to emulate the methodology of Migkas et al. for galaxy-cluster scaling relations. Using $$\Omega_{\rm m}=0.306,\quad \Omega_{\rm b}=0.0486,\quad \sigma_8=0.807,\quad H_0=68.1,\quad n_s=0.967,$$05, $$\Omega_{\rm m}=0.306,\quad \Omega_{\rm b}=0.0486,\quad \sigma_8=0.807,\quad H_0=68.1,\quad n_s=0.967,$$06, and $$\Omega_{\rm m}=0.306,\quad \Omega_{\rm b}=0.0486,\quad \sigma_8=0.807,\quad H_0=68.1,\quad n_s=0.967,$$07 relations, the study finds that an M21-like anisotropy emerges in FLAMINGO with probability

$$\Omega_{\rm m}=0.306,\quad \Omega_{\rm b}=0.0486,\quad \sigma_8=0.807,\quad H_0=68.1,\quad n_s=0.967,$$08

while a bulk-flow interpretation at $$\Omega_{\rm m}=0.306,\quad \Omega_{\rm b}=0.0486,\quad \sigma_8=0.807,\quad H_0=68.1,\quad n_s=0.967,$$09 reaches approximately $$\Omega_{\rm m}=0.306,\quad \Omega_{\rm b}=0.0486,\quad \sigma_8=0.807,\quad H_0=68.1,\quad n_s=0.967,$$10 (He et al., 2 Apr 2025). The same analysis concludes that statistical noise accounts for over $$\Omega_{\rm m}=0.306,\quad \Omega_{\rm b}=0.0486,\quad \sigma_8=0.807,\quad H_0=68.1,\quad n_s=0.967,$$11 of the anisotropy amplitude in each lightcone, with peculiar velocities contributing less than $$\Omega_{\rm m}=0.306,\quad \Omega_{\rm b}=0.0486,\quad \sigma_8=0.807,\quad H_0=68.1,\quad n_s=0.967,$$12, and that tighter scaling relations yield stronger isotropy constraints (He et al., 2 Apr 2025).

FLAMINGO lightcones have also been used to construct a self-consistent full-sky secondary-CMB and foreground suite including CMB lensing, thermal SZ, kinetic SZ, cosmic infrared background, radio point sources, and anisotropic screening / optical depth (Yang et al., 10 Dec 2025). A major point of that work is methodological: the components are not painted independently onto a dark-matter-only scaffold, but arise from the same baryonic simulation, so their cross-correlations are physically linked by the common underlying gas, star-formation, and black-hole distributions. The paper concludes that the hydrodynamical mock skies perform at least as well as previous dark-matter-only mock frameworks in matching observational constraints while retaining self-consistency across components (Yang et al., 10 Dec 2025).

In this sense, FLAMINGO’s legacy is twofold. It is a calibrated simulation suite for baryonic cosmology and cluster physics, and it is an all-sky forward-modeling infrastructure with publicly released lightcones, maps, catalogues, and power spectra (Helly et al., 27 Apr 2026). A plausible implication is that its long-term importance lies not only in any individual result—on $$\Omega_{\rm m}=0.306,\quad \Omega_{\rm b}=0.0486,\quad \sigma_8=0.807,\quad H_0=68.1,\quad n_s=0.967,$$13, cluster profiles, high-redshift quiescent galaxies, or bright quasars—but in the way it standardizes a common hydrodynamical platform for comparing cosmology, feedback, and survey observables across traditionally separate subfields.