The FLAMINGO simulations data release

Abstract: We describe the public release of $>2.3$ petabytes of data from the FLAMINGO cosmological simulations. The suite consists of hydrodynamical simulations that include radiative cooling, star formation, stellar mass loss and the resulting chemical enrichment, supernova feedback, and two implementations of AGN feedback. Neutrinos are simulated explicitly using particles. Data products include snapshots, halo and galaxy catalogues, HEALPix all-sky lightcone maps, particle data for lightcone maps, and power spectra. The FLAMINGO set includes 22 hydrodynamical simulations. In addition, there are 16 gravity-only simulations, including the $10080^3$ particles FLAMINGO-10k run, with initial conditions that match those of the corresponding hydrodynamical runs. The fiducial hydrodynamical simulations span three numerical resolutions that have each been calibrated to reproduce the present-day galaxy stellar mass function and gas fractions in low-redshift clusters. Other simulations systematically vary the galaxy stellar mass function, cluster gas fractions, cosmology (including neutrino masses), and/or the nature of dark matter, in volumes of 1Gpc$^3$. The release includes hitherto unpublished simulations that use extra dark matter particles. While we provide a facility for downloading complete simulation outputs, we recognise that for many users this will not be possible due to limited local storage or network bandwidth. We implement a web service that enables users to explore available outputs and selectively download datasets or parts of datasets.

• The paper introduces an extensive suite of hydrodynamical and gravity-only simulations that accurately model cosmic structures and baryonic effects.
• The paper details the use of state-of-the-art computational methods, including SWIFT code and explicit neutrino treatment, to calibrate galaxy formation and feedback processes.
• The paper provides over 2.3 petabytes of public data products, enabling robust analysis and validation of survey cosmology, galaxy evolution, and observational systematics.

The FLAMINGO Simulations: Architecture, Scope, Data Products, and Implications

Overview and Motivation

The FLAMINGO simulations, described in "The FLAMINGO simulations data release" (2604.24324), set a new standard for cosmological hydrodynamical simulations targeting large-scale structure and multiwavelength synthetic sky modeling. The project comprises an extensive suite—22 hydrodynamical and 16 gravity-only simulations—executed with the SWIFT code, incorporating state-of-the-art solvers for gravity (4th-order fast-multipole, particle-mesh) and hydrodynamics (SPHENIX), as well as explicit massive neutrino treatment via the $\delta f$ method. The simulation suite is designed for self-consistent predictions of galaxy formation, cluster thermodynamics, and observable signatures across cosmic time and broad spatial scales, with attention to calibration against low-redshift observables.

The scientific premise of FLAMINGO is that next-generation cosmological surveys (Euclid, LSST, DESI, etc.) demand precise theoretical modeling—specifically, models that self-consistently couple baryonic physics, feedback processes, and neutrino effects. On non-linear and quasi-linear scales, baryonic effects can induce power spectrum modifications at several percent level and drive selection effects in galaxy and cluster cosmology. Pure dark-matter N-body simulations are insufficient for controlling these systematics.

Simulation Suite and Calibration Strategy

Multi-Physics and Flexible Cosmology

FLAMINGO covers three mass resolutions—m8, m9, and m10—differentiated by baryonic particle mass spans ($1.34\times10^8$ to $8.56\times10^9~M_\odot$). The suite leverages volumetric coverage from $1~\mathrm{Gpc}^3$ to $22~\mathrm{Gpc}^3$, enabling treatment of both rare objects (massive clusters, bright quasars) and cosmic variance-mitigated statistics. Each simulation tracks dark matter, baryons, and neutrinos (with reduced sampling). Multiple cosmological parameter sets are explored, including variants of $\Lambda$CDM, enhanced $\sum m_\nu$, and decaying dark matter.

The galaxy formation and feedback models are independently calibrated at each resolution via Markov Chain Monte Carlo (MCMC) guided by a Gaussian process emulator, following the approach pioneered in [Kugel et al., (Kugel et al., 2023)]. Calibration targets are the $z=0$ stellar mass function (GAMA DR4) and cluster gas fractions from combined X-ray and lensing data. This results in distinct parameter choices for stellar and AGN feedback at each resolution. Some models systematically shift calibration targets ($\Delta m_*$, $\Delta f_{gas}$) to sample the plausible range of baryonic effects.

Subgrid Physics Configuration

• Radiative cooling/heating: Metal-line cooling and photoionization implemented following [Ploeckinger et al., 2020].
• Star formation and entropy regulation: Based on [Schaye et al., 2008], with an imposed ISM entropy floor.
• Stellar evolution and chemical enrichment: Metal yields and delayed feedback per [Wiersma et al., 2009].
• Supernova feedback: Kinetic implementation per [Chaikin et al., 2023].
• AGN feedback: Two options—thermal energy injection and kinetic jet-mode [Booth & Schaye 2009; Husko et al., 2022]. Both coupled to a standard black hole seeding/accretion scheme.

Explicit attention is paid to potential numerical artifacts, e.g., spurious energy transfer between baryons and dark matter when particle masses differ substantially, mitigated in a super-sampled dark matter run.

Data Products and Accessibility

The FLAMINGO data release exceeds 2.3 petabytes, hosted at the DiRAC Memory Intensive Service. The data products are intended to maximize community exploitation and benchmarking for observables covering galaxy populations, clusters, and cosmic large-scale structure.

Released Data Types

• Snapshots: Full, reduced (halo-centric), and randomly downsampled particle data at 13 designated redshifts per simulation, with additional more frequent outputs for a subset of runs.
• Halo and Galaxy Catalogues: Hierarchical, merger-traced catalogues constructed with HBT-HERONS [Han et al., 2018], analyzed with SOAP [mcgibbon25]. Aperture definitions include various spherical and projected truncations, supporting cross-simulation matching and controlled analyses of baryonic versus DMO runs.
• Lightcones and All-sky Maps: Integrated and tomographic HEALPix maps for mass, SFR, weak and CMB lensing, thermal/kinetic SZ, X-ray, IR, radio, and DM observables, constructed to facilitate synthetic sky comparisons in the survey context.
• Power Spectra: Shot-noise-corrected 3D (auto-, cross-) power spectra for total matter, CDM, gas, star/black hole, and electron pressure fields at $1.34\times10^8$0 redshifts per simulation, allowing quantification of baryonic suppression/enhancement effects.
• Initial Conditions: HDF5-format initial conditions (or configuration files) for each run, supporting reproducibility and zoom-in resimulations.

Infrastructure for Data Access

To circumvent the barrier imposed by dataset size, a web service with slice/field-based HDF5 and messagepack streaming is provided, along with a Python client, integration with swiftsimio, and a web-based browser. This system supports partial data retrievals—targeting, for example, halo-centric cutouts, or tomographic sky maps—enabling usage without supercomputing-scale storage or bandwidth. The architecture ensures compatibility with traditional HPC workflows and supports interactive exploration by non-collaboration users.

Model Performance and Known Systematics

Observational Consistency and Discrepancies

The fiducial calibrated models robustly reproduce the $1.34\times10^8$1 stellar mass function and X-ray cluster gas fractions across the calibration mass range, and the cosmic star formation history is matched up to $1.34\times10^8$2 for m9. FLAMINGO improves the abundance of high-SFR galaxies (e.g., SMGs) relative to earlier generations. The models also display realistic scaling relations and thermodynamic profiles for clusters, including at high redshift, though central metallicities are overpredicted.

Outstanding tensions include:

• Overly large sizes for low-mass galaxies ($1.34\times10^8$3)
• Undersuppression of star formation in the highest-mass galaxies ($1.34\times10^8$4)
• Insufficient numbers of massive, quiescent galaxies at high-redshift
• Discrepant feedback efficiency for the group regime: stacked kSZ and X-ray analyses favor models with low gas fractions, but individually X-ray selected groups fit better to the fiducial model. This highlights systematics in selection—kSZ and X-ray stacks probe different mass/redshift regimes.

The suite enables clear quantification of baryonic and feedback effects on the matter power spectrum, lensing, and SZ observables, supporting the modeling of systematics for survey cosmology.

Numerical Convergence

Mass and spatial resolution are varied over nearly two orders of magnitude. Feedback models are independently re-calibrated at each resolution. However, because different resolutions necessarily imply different effective physical models (e.g., the scale of energy injection events), convergence failing outside calibrated ranges is not strictly a numerical artifact, but sometimes a reflection of model incompleteness or calibration incompleteness.

A specific bug (misapplied star formation threshold, incorrect black hole repositioning due to not removing the BH potential) affects some intermediate-resolution models, leading to differences in quenched fractions and AGN feedback efficiency across model variants.

Implementation Bugs and Guidance

Documented bugs and caveats include missing satellite AGN due to BH repositioning on gas-poor apocenters, epoch-independent AGN heating parameter in SOAP, and corrections to the cosmological scaling of kSZ/DM maps. Remediation strategies and masks are provided for affected observables.

Theoretical and Practical Implications

FLAMINGO sets a new scale for the systematic exploration of baryonic, neutrino, and cosmological parameter space in self-consistent hydro simulations. With the companion DMO suite and the released calibration variations, the dataset supports construction of emulators and response functions for next-generation weak lensing, clustering, and cross-correlation analyses.

Practically, the public release—delivering full-resolution data, partial cutouts, and synthetic sky maps—enables broad community usage for both survey pipeline validation and method development. FLAMINGO is directly relevant for covariance modeling, mitigation of baryonic uncertainties, modeling selection and projection effects, and as a testbed for new machine learning inference procedures in large-scale structure.

Theoretically, the suite will drive advances in our understanding of the interplay between feedback, baryon cycling, and large-scale structure. The inclusion of independently calibrated feedback variants and cosmologies supports marginalization over baryonic systematics in cosmological constraints in a manner not possible with smaller, non-systematically-constructed simulation sets.

An open direction is the integration and cross-validation with semi-analytic models and survey simulation frameworks; ongoing and future analysis will likely produce further value-added products, including galaxy photometry, mock image cubes, and improved merger trees, building towards a comprehensive virtual universe platform.

Conclusion

The FLAMINGO simulations set a benchmark for large-scale cosmological hydrodynamical experiment design, focusing on systematic coverage of baryonic physics, feedback calibration, and cosmological parameter space. The 2.3+ PB public data release, combined with flexible, user-oriented access, positions the suite as a central resource for theoretical astrophysics and cosmological data analysis pipelines. The demonstrated strengths and documented systematics make FLAMINGO an essential component in the quantification and mitigation of baryonic effects on cosmological inference, as well as a fertile ground for future theoretical developments in the modeling of the observable universe (2604.24324).