Auriga Galaxy Formation Model

• Auriga galaxy formation model is a cosmological framework using Arepo, simulating Milky Way–mass and dwarf haloes with detailed baryonic subgrid physics.
• The model emphasizes efficient cooling in the CGM and non-local kinetic wind feedback that drives recycling, yielding higher stellar masses and realistic disc features.
• Auriga’s robust yet resolution-sensitive approach produces thin, dynamically rich discs with prominent bars and spirals, making it a critical reference for galaxy evolution studies.

Searching arXiv for recent and core Auriga papers to ground the article. The Auriga galaxy formation model is a cosmological galaxy formation framework implemented in the moving-mesh magnetohydrodynamics code Arepo and used primarily for zoom-in simulations of Milky Way–mass haloes, with later extensions to dwarf galaxies and star-cluster modelling. In the literature surveyed here, it appears as a full-physics model coupling gravity, ideal MHD, radiative cooling, a multiphase interstellar medium, stochastic star formation, stellar evolution, supernova-driven winds, supermassive black holes, and AGN feedback, all within a $\Lambda$CDM cosmology (Grand et al., 2024). The model is defined not only by its numerical platform but also by a distinctive subgrid treatment of baryonic physics, especially its kinetic, isotropic wind feedback and its regulation of circumgalactic-medium cooling, which have been shown to shape stellar masses, disc structure, bars, spiral arms, satellites, and baryon cycling in Milky Way analogues (Hu et al., 23 Mar 2026).

1. Definition and numerical framework

Auriga is a suite of cosmological zoom-in simulations of Milky Way–mass haloes run with Arepo, a moving-mesh MHD code (Hu et al., 23 Mar 2026). In the broader augmented project, the same galaxy formation model is applied to 40 Milky Way-mass halos and 26 dwarf galaxy-mass halos in a Planck 2013 $\Lambda$CDM cosmology, with initial redshift $z_{\rm init}=127$ and parent initial conditions derived from the EAGLE dark-matter-only volume (Grand et al., 2024). The target Milky Way–mass systems span

$0.5 < M_{200} / [10^{12}\,{\rm M}_\odot] < 2,$

while additional suites extend to lower halo masses (Grand et al., 2024).

At the standard Milky Way resolution level, Auriga typically uses baryonic mass resolution of order $5\times 10^4\,M_\odot$ and dark-matter particle mass of order $3\times 10^5$–$4\times 10^5\,M_\odot$, with higher-resolution variants improving mass resolution by a factor of 8 (Sante et al., 20 Feb 2025). The model includes ideal MHD with a uniform comoving seed field of strength $10^{-14}\,\mathrm{G}$ at $z=127$, which is subsequently amplified by structure formation and disc dynamics (Sante et al., 20 Feb 2025). Magnetic fields are evolved self-consistently and produce large-scale ordered azimuthal configurations in discs by $z=0$, while the global galaxy properties remain only weakly affected because equipartition is reached relatively late (Pakmor et al., 2017).

The published data release emphasizes that Auriga follows the coupled evolution of dark matter, gas, stars, supermassive black holes, and magnetic fields, and that the framework has been shown to produce numerically well-converged galaxy properties for Milky Way–mass systems (Grand et al., 2024). A plausible implication is that Auriga is intended as a unified galaxy formation model rather than a family of mass-specific prescriptions.

2. Baryonic subgrid physics

Auriga adopts a comprehensive galaxy formation model descended from the Vogelsberger–Marinacci–Grand lineage and includes primordial and metal-line cooling, a UV background, a subgrid multiphase ISM, stochastic star formation, stellar evolution and enrichment, kinetic stellar feedback, and AGN feedback (Grand et al., 2024). In the comparative Milky Way analogue study, the most explicit distinguishing feature is the kinetic wind model of Vogelsberger et al. (2013) (Hu et al., 23 Mar 2026).

In this scheme, stellar feedback is implemented by creating “wind particles” that are launched kinematically and isotropically (Hu et al., 23 Mar 2026). These particles are temporarily decoupled from the dense ISM, travel into lower-density gas, and later recouple once they have moved sufficiently far or reached a low-density region, depositing mass, metals, and both thermal and kinetic energy back into the gas (Hu et al., 23 Mar 2026). The resulting feedback is non-local: energy avoids immediate radiative loss in dense star-forming gas and is thermalized in the outer ISM or CGM. In comparison studies against EAGLE-like or APOSTLE-like models, this is contrasted directly with thermal supernova feedback that heats dense gas in situ to a fixed temperature (Hu et al., 23 Mar 2026).

Auriga’s ISM follows the Springel & Hernquist two-phase model in which gas above a threshold density enters a pressurized star-forming phase (Sante et al., 20 Feb 2025). One paper states explicitly that when gas reaches densities

$\Lambda$0

it enters a subgrid two-phase ISM model with hot and cold phases in pressure equilibrium and an effective equation of state (Sante et al., 20 Feb 2025). Another Auriga overview gives the star-formation threshold as

$\Lambda$1

with star formation realized stochastically and calibrated to a Kennicutt–Schmidt-like relation (Grand et al., 2024). In the context of the comparison paper, the ISM is operationally defined as gas that has at least once satisfied the star formation criterion, i.e. gas with $\Lambda$2 at least once (Hu et al., 23 Mar 2026).

A useful decomposition of stellar mass growth employed in the Milky Way analogue comparison is

$\Lambda$3

where $\Lambda$4 is the total gas mass ever accreted into the halo, $\Lambda$5 is the fraction of accreted gas that ever cools into the star-forming ISM, $\Lambda$6 is the retained cold-baryon fraction, and $\Lambda$7 is the efficiency of converting cold baryons into stars (Hu et al., 23 Mar 2026). This decomposition is central to how Auriga’s behavior is interpreted: the main differences from comparison models arise not inside the star-forming ISM, but in the amount of gas that cools from the CGM into the ISM.

3. Circumgalactic regulation and the baryon cycle

A recurring conclusion across Auriga studies is that the model produces a CGM that is comparatively favorable to cooling, recycling, and long-lived baryon retention. In the APOSTLE–Auriga comparison, the dominant systematic difference is that Auriga has a much higher efficiency of cooling and condensation from the CGM into the ISM, quantified by

$\Lambda$8

at $\Lambda$9 (Hu et al., 23 Mar 2026). By contrast, the total amount of gas ever accreted into the halo differs by at most $z_{\rm init}=127$0 between the matched runs, and the cold baryon retention factor satisfies

$z_{\rm init}=127$1

in both models at $z_{\rm init}=127$2 (Hu et al., 23 Mar 2026). This implies that Auriga’s higher stellar masses are driven primarily by enhanced CGM$z_{\rm init}=127$3ISM supply rather than by a more efficient star-formation law once gas is cold.

The Local Group baryon-cycle comparison with EAGLE sharpens this picture. There, Auriga predicts that the Milky Way is almost baryonically closed, whereas EAGLE predicts that only half of the expected baryons reside within the halo (Kelly et al., 2021). At $z_{\rm init}=127$4, within $z_{\rm init}=127$5, the primary Auriga haloes in that study have

$z_{\rm init}=127$6

while the matched EAGLE haloes have

$z_{\rm init}=127$7

(Kelly et al., 2021). The paper attributes this to differences in the energy injection method from supernovae to gas: EAGLE’s thermal heating drives halo-wide hot outflows at high redshift that both eject baryons and impede fresh accretion, whereas in Auriga gas accretion is almost unaffected by feedback (Kelly et al., 2021).

The same study shows that, among baryons initially associated with the halo’s Lagrangian region, Auriga has very little permanently impeded gas—gas that never enters the halo—whereas EAGLE has a substantial permanently prevented component (Kelly et al., 2021). Auriga’s missing baryons are predominantly temporarily ejected and then recycled. This supports the description of Auriga as a fountain-dominated baryon cycle rather than a strongly preventive feedback model.

Auriga’s CGM itself is highly diverse. In a sample of isolated Milky Way–mass haloes, the covering fractions of common ions span broad ranges and correlate with stellar mass, AGN luminosity, and disc fraction (Hani et al., 2019). The paper finds that the covering fractions of hydrogen and metals positively correlate with stellar mass, that the covering fractions of H I, C IV, and Si II anticorrelate with AGN luminosity due to ionization effects, and that the covering fractions of H I, C IV, and Si II positively correlate with disc fraction because outflows populate the CGM with cool and dense gas (Hani et al., 2019). This suggests that the Auriga model links CGM phase structure to the morphology and feedback state of the central disc, not only to halo mass.

4. Milky Way analogues: masses, discs, and morphology

When applied to Milky Way–mass zoom-ins, Auriga forms well-defined disc galaxies with flat rotation curves, realistic H I discs, and structural scaling relations that compare well with observations (Sante et al., 20 Feb 2025). In the matched comparison against APOSTLE, Auriga galaxies have halo masses

$z_{\rm init}=127$8

with counterpart differences below 0.1 dex, but stellar masses that are systematically larger by roughly $z_{\rm init}=127$9–$0.5 < M_{200} / [10^{12}\,{\rm M}_\odot] < 2,$0 dex (Hu et al., 23 Mar 2026). The paper summarizes this as a systematic $0.5 < M_{200} / [10^{12}\,{\rm M}_\odot] < 2,$1 dex higher stellar mass in Auriga galaxies at fixed halo mass (Hu et al., 23 Mar 2026).

These galaxies also have higher stellar surface densities at all radii and are often slightly more compact, with somewhat smaller effective radii and larger Sérsic indices in matched pairs (Hu et al., 23 Mar 2026). The stellar surface density is fit with

$0.5 < M_{200} / [10^{12}\,{\rm M}_\odot] < 2,$2

and the photometric disc component with

$0.5 < M_{200} / [10^{12}\,{\rm M}_\odot] < 2,$3

(Hu et al., 23 Mar 2026).

Auriga discs are generally more prominent than their matched APOSTLE counterparts. Using a kinematic decomposition based on orbital circularity

$0.5 < M_{200} / [10^{12}\,{\rm M}_\odot] < 2,$4

with $0.5 < M_{200} / [10^{12}\,{\rm M}_\odot] < 2,$5 defining the kinematic disc, all Auriga galaxies in that comparison have kinematic

$0.5 < M_{200} / [10^{12}\,{\rm M}_\odot] < 2,$6

and three of four matched Auriga systems have higher disk-to-total ratios than their APOSTLE twins (Hu et al., 23 Mar 2026). Auriga discs are also systematically older at a given radius, indicating earlier disc formation (Hu et al., 23 Mar 2026).

Bars and spiral arms are especially important morphological outcomes of the Auriga model. The comparative study reports that Auriga discs show more prominent bars and spiral arms, and that these structures are associated with younger, more metal-rich stars in the age and metallicity maps (Hu et al., 23 Mar 2026). The dedicated bar study expands this substantially: in 39 Milky Way–mass discs, bars are identified through the $0.5 < M_{200} / [10^{12}\,{\rm M}_\odot] < 2,$7 Fourier mode

$0.5 < M_{200} / [10^{12}\,{\rm M}_\odot] < 2,$8

with barred galaxies satisfying $0.5 < M_{200} / [10^{12}\,{\rm M}_\odot] < 2,$9 and visual confirmation (Fragkoudi et al., 2024). In that sample, the bar fraction at $5\times 10^4\,M_\odot$0 is approximately $5\times 10^4\,M_\odot$1, decreasing with redshift and plateauing around $5\times 10^4\,M_\odot$2 at $5\times 10^4\,M_\odot$3 (Fragkoudi et al., 2024).

Barred Auriga galaxies tend to be more baryon-dominated, to assemble their stellar mass earlier, and to have lower Toomre $5\times 10^4\,M_\odot$4 at the epoch of bar formation (Fragkoudi et al., 2024). A central result is that barred galaxies are more baryon-dominated at all redshifts, and galaxies that are baryon-dominated but remain unbarred have higher ex-situ bulge fractions (Fragkoudi et al., 2024). This connects bar formation in Auriga to the detailed balance between in-situ disc growth and merger-built bulges.

Spiral structure in Auriga is also diverse. In the high-resolution Auriga Superstars extension, the pattern-speed profiles show that several classical spiral theories are realized in different systems or even in the same system at different times: large-scale kinematic density waves, manifold spirals, dynamic co-rotating spirals, and overlapping modes (Grand et al., 16 Feb 2026). The same galaxy may show qualitative evolution of its spiral pattern-speed profile on sub-gigayear timescales (Grand et al., 16 Feb 2026). This indicates that Auriga does not impose a single spiral mechanism; rather, spiral structure emerges from the interaction of cosmological perturbations, bars, and disc dynamics.

5. Vertical structure, thick discs, and satellite systems

Auriga produces thin stellar discs in Milky Way–mass haloes. For the kinematic disc component, the vertical density profile is well described by

$5\times 10^4\,M_\odot$5

with scale heights that increase with radius and remain below $5\times 10^4\,M_\odot$6 kpc inside $5\times 10^4\,M_\odot$7 kpc in both Auriga and APOSTLE (Hu et al., 23 Mar 2026). A model-independent thickness measure,

$5\times 10^4\,M_\odot$8

agrees with the fitted $5\times 10^4\,M_\odot$9 model, validating that description (Hu et al., 23 Mar 2026). This is consistent with the broader conclusion that Auriga forms thin, rotationally supported stellar discs despite its energetic feedback and cosmological merger histories.

The thick-disc formation study adds a chemically resolved perspective. Using 24 spiral galaxies from the Auriga zoom-in sample, it finds that thick discs are older, more metal-poor, and more [Mg/Fe]-enhanced than thin discs, but also internally complex, with contributions from in-situ formation, accreted gas, and ex-situ stars (Pinna et al., 2023). Across the sample, thick-disk regions host 7–61% of their stellar mass in accreted stars, with a median accreted fraction of about 28% (Pinna et al., 2023). Thick disks thus emerge in Auriga as composite structures produced by the interplay between internal enrichment and external gas and stellar accretion (Pinna et al., 2023).

The model’s treatment of satellites also has characteristic consequences. In the APOSTLE–Auriga comparison, satellites within 400 kpc and with

$3\times 10^5$0

are systematically more massive in Auriga, with the most massive satellite in each halo typically about 0.3 dex more massive than in the corresponding APOSTLE halo (Hu et al., 23 Mar 2026). Yet in three of four halo pairs, the total number of satellites above $3\times 10^5$1 is similar between the models (Hu et al., 23 Mar 2026). This reflects two countervailing effects explicitly discussed in the paper: a more massive central galaxy in Auriga produces stronger tidal fields that can destroy more subhaloes, but Auriga dwarfs form more stars at fixed halo mass (Hu et al., 23 Mar 2026).

The satellite stellar mass–metallicity relation in both simulations lies about 0.25 dex above the observed relation of Local Group dwarfs (Hu et al., 23 Mar 2026). Auriga’s very low-mass satellites appear less extreme than APOSTLE’s, which the authors link indirectly to more efficient metal mixing in the Arepo-based framework (Hu et al., 23 Mar 2026). Both models also lack a significant population of faint, blue, low-mass star-forming satellites, though the discrepancy is more pronounced in APOSTLE (Hu et al., 23 Mar 2026).

6. Extensions, robustness, and interpretive uses

Auriga has increasingly been used not only as a forward galaxy-formation model but also as an inference backbone. In GalactiKit, the Auriga cosmological MHD simulations define the forward mapping between progenitor merger properties at infall and the chemo-dynamical properties of their debris at $3\times 10^5$2 (Sante et al., 20 Feb 2025). Using simulation-based inference with Masked Autoregressive Flows, the study shows that the combined use of $3\times 10^5$3, $3\times 10^5$4, [Fe/H], and [$3\times 10^5$5/Fe] from stellar debris can recover infall times to about $3\times 10^5$6 Gyr and stellar masses to about $3\times 10^5$7 dex (Sante et al., 20 Feb 2025). This indicates that Auriga’s merger histories, enrichment histories, and dynamical evolution encode a sufficiently informative mapping to support Galactic archaeology applications.

The model’s numerical robustness has also been quantified directly. A study of seven realizations of the same Milky Way–like halo, differing only in the random numbers used by stochastic star formation, wind launching, and AGN bubble placement, finds that global galaxy properties at $3\times 10^5$8—including stellar mass, star formation history, bulge and disc masses, and disc radius and height—change by less than 10% between realizations (Pakmor et al., 17 Jul 2025). By contrast, the present-day star formation rate can vary by a factor of two, and detailed internal morphology such as bar strength can differ (Pakmor et al., 17 Jul 2025). This establishes that the Auriga model is globally robust at fixed resolution, while remaining sensitive in chaotic fine-grained details.

Resolution changes matter more than stochastic variability. The same robustness study shows that lowering mass resolution by a factor of 8 reduces stellar mass by 21.1%, lowers outer halo stellar mass by 39.4%, makes discs 28.6% thicker, and lowers total stellar angular momentum by 33.6%, all shifts larger than the intrinsic run-to-run scatter (Pakmor et al., 17 Jul 2025). This is important for interpreting cross-resolution comparisons: in Auriga, numerical resolution induces systematic changes that are larger than stochastic differences (Pakmor et al., 17 Jul 2025).

Auriga has also been extended with subgrid globular-cluster formation and evolution. The AuriGLOBES model implements star-cluster formation in tidally compressive, high-pressure gas and includes enhanced mass loss from compact object remnants heating (Guerra et al., 29 Jun 2026). In that extension, the resulting globular-cluster populations reproduce the empirical GC system mass–halo mass relation within a $3\times 10^5$9 scatter and require both formation in compressive tides and enhanced remnant-driven mass loss to transform an initial Schechter mass function into the observed globular-cluster mass function (Guerra et al., 29 Jun 2026). This constitutes a substantial generalization of Auriga from galaxy formation alone to coupled galaxy–cluster population modelling.

A notable counterpoint is provided by the older globular-cluster-candidate analysis, which used old star particles as a proxy for GC formation sites. That study concluded that Auriga’s old stellar population is too metal-rich and too radially extended to be reconciled with the observed GC systems of the Milky Way and M31 under simple cluster formation and destruction assumptions (Halbesma et al., 2019). The AuriGLOBES extension can be read as a response to that limitation: rather than relying on old field stars as GC proxies, it introduces explicit cluster-formation physics within the same host galaxy model.

7. Physical interpretation and limitations

Across these studies, the Auriga galaxy formation model is consistently interpreted as a kinetic-wind, moving-mesh implementation whose distinctive behavior lies in how it regulates the CGM and therefore the supply of star-forming gas (Hu et al., 23 Mar 2026). Its characteristic chain of causation is: kinetic, isotropic winds plus Arepo hydrodynamics produce weaker preventive feedback and more recycling; this raises the fraction of accreted gas that cools into the ISM; that in turn yields higher stellar masses, denser discs, earlier disc formation, stronger bars and spiral arms, and more massive satellites (Hu et al., 23 Mar 2026).

Several strengths recur in the literature. Auriga produces thin disc galaxies in $4\times 10^5\,M_\odot$0 haloes, early and dynamically rich discs with bars and spirals, plausible satellite counts above $4\times 10^5\,M_\odot$1, realistic H I discs and star-forming main-sequence behavior, and broadly successful magnetic, kinematic, and structural properties across a wide mass range (Hu et al., 23 Mar 2026). The public data release emphasizes that the model compares well with the Tully–Fisher relation, the star-forming main sequence, and H I gas fraction and disc thickness, and that gas discs build rotation and settle into increasing $4\times 10^5\,M_\odot$2 in rough agreement with some H$4\times 10^5\,M_\odot$3 observations (Grand et al., 2024).

The limitations are equally clear. In matched Milky Way–mass comparisons, Auriga overproduces stellar mass relative to APOSTLE by about $4\times 10^5\,M_\odot$4 dex at fixed halo mass and yields high stellar surface densities and central concentrations (Hu et al., 23 Mar 2026). Satellite metallicities are too high by about $4\times 10^5\,M_\odot$5 dex relative to Local Group dwarfs (Hu et al., 23 Mar 2026). The globular-cluster proxy study found the old stellar component too metal-rich and too radially extended for realistic Milky Way/M31 GC systems (Halbesma et al., 2019). The CGM study indicates that Auriga’s AGN luminosities at $4\times 10^5\,M_\odot$6 may be too high relative to many observed $4\times 10^5\,M_\odot$7 galaxies, which can over-ionize low ions and elevate O VI columns (Hani et al., 2019). The robustness study shows that detailed morphology is seed-sensitive in marginally unstable systems, and that resolution dependence is systematic and non-negligible (Pakmor et al., 17 Jul 2025).

A broader interpretive point follows from the controlled comparison papers. Because Auriga and its comparators can be run on identical initial conditions, differences in final galaxies can be causally tied to the subgrid model, especially the feedback implementation (Hu et al., 23 Mar 2026). This makes Auriga an important reference model for understanding how Milky Way analogues depend on CGM cooling, wind recycling, and baryonic angular-momentum regulation. The accumulated evidence suggests that Auriga’s most consequential modeling choice is not a radical star-formation law inside the ISM, but its comparatively permissive treatment of how gas moves from halo scales into the star-forming disc (Hu et al., 23 Mar 2026).