Public Data Release of the FIRE-2 Cosmological Zoom-in Simulations of Galaxy Formation

We generated all simulations using Gizmo ²⁰ (Hopkins 2015), a multimethod gravity plus (magneto)hydrodynamics code. We used the MFM mode for hydrodynamics, a quasi-Lagrangian Godunov method that provides adaptive spatial resolution while maintaining exact conservation of mass, energy, and momentum, excellent angular momentum conservation, and accurate shock capturing. Thus, the method provides advantages of both SPH and Eulerian adaptive mesh refinement methods. Gizmo solves gravity using an improved version of the Tree-PM solver from GADGET-3 (Springel 2005), using fully adaptive and conservative gravitational force softening for gas cells that matches their hydrodynamic resolution.

All of these simulations use the same FIRE-2 physics model (Hopkins et al. 2018a), with minor exceptions that we describe below. Briefly, FIRE-2 incorporates radiative cooling and heating across 10−10¹⁰ K, including free–free, photoionization and recombination, Compton, photoelectric and dust collisional, cosmic ray, molecular, metal-line, and fine-structure processes, self-consistently tracking 11 elements (hydrogen, helium, carbon, nitrogen, oxygen, neon, magnesium, silicon, sulfur, calcium, and iron). This includes photoionization and heating from a redshift-dependent, spatially uniform ultraviolet background (Faucher-Giguère et al. 2009), which reionizes the universe at z ≈ 10. ²¹ The modeling of ionization also includes approximations for self-shielding of dense gas and radiation from local sources based on the LEBRON scheme (Hopkins et al. 2020a).

Star formation occurs in self-gravitating gas (following Hopkins et al. 2013) that also is molecular and self-shielding (following Krumholz & Gnedin 2011), Jeans unstable, and exceeds a minimum density threshold, n_SF > 1000 cm⁻³. FIRE-2 follows several stellar-feedback mechanisms, including (i) local and long-range momentum flux from radiation pressure in the ultraviolet and optical (single scattering), as well as reradiated light in the infrared; (ii) energy, momentum, mass, and metal injection from core-collapse + Type Ia supernovae and stellar mass loss (dominated by O, B, and asymptotic giant branch stars); and (iii) photoionization and photoelectric heating. FIRE-2 models every star particle as a single stellar population with a single age and metallicity, and tabulates all feedback event rates, luminosities and energies, mass-loss rates, and other quantities directly from stellar evolution models (STARBURST99 v7.0; Leitherer et al. 1999, 2014), assuming a Kroupa (2001) initial mass function for stars across 0.1–100 M_⊙.

Core-collapse supernovae, Type Ia supernovae, and stellar winds generate and deposit metals into surrounding gas cells. FIRE-2 adopts the following models: (i) for stellar winds, rates from STARBURST99 and yields from a compilation of van den Hoek & Groenewegen (1997), Marigo (2001), and Izzard et al. (2004); (ii) for core-collapse supernovae, rates from STARBURST99 and nucleosynthetic yields from Nomoto et al. (2006); and (iii) for Type Ia supernovae, rates from Mannucci et al. (2006) and yields from Iwamoto et al. (1999). FIRE-2 initializes abundances in gas (typically at z ≈ 99) for all elements i (beyond H and He) to a floor of $\left[{{\rm{M}}}_{i}/{\rm{H}}\right]\approx -4$, to prevent numerical problems in cooling. All simulations in this data release (except the Massive Halo suite) include an explicit model for unresolved turbulent diffusion of metals in gas (Hopkins 2017; Su et al. 2017a; Escala et al. 2018).

For more details on the physics and numerics of the FIRE-2 simulations, see Hopkins (2015) for the Gizmo simulation code, Hopkins et al. (2018a) for the FIRE-2 physics model, Hopkins et al. (2018b) for more details on modeling mechanical feedback, and Hopkins et al. (2020a) for more details on modeling radiative feedback.

We release the FIRE-2 simulations that include the base set of FIRE-2 physics, as described above. These simulations do not include any additional physics, including the optional models in Hopkins et al. (2018a). Specifically, the simulations that we release:

• 1.  
  do not include magnetohydrodynamics (MHD) or anisotropic conduction and viscosity; recent implementations in FIRE-2 suggest that these processes do not significantly change galaxy-wide properties (Su et al. 2017a; Hopkins et al. 2020b);
• 2.  
  do not model self-consistent cosmic-ray injection and transport, beyond assuming a fixed heating rate from cosmic rates in dense gas (see Chan et al. 2019; Ji et al. 2020; Hopkins et al. 2021);
• 3.  
  do not model self-consistent creation and destruction of dust, beyond simply assuming that dust traces gas-phase metallicity (see Choban et al. 2022);
• 4.  
  use the LEBRON method to model radiative transfer in the optically thin limit (beyond the local gas kernel); they do not model radiation hydrodynamics via methods such as flux-limited diffusion or M1, though these approaches are unlikely to change galaxy-wide properties significantly (see Hopkins et al. 2020a); and
• 5.  
  only the Massive Halo suite models the growth of supermassive black holes, and no simulation models feedback from an active galactic nucleus (AGN; see Wellons et al. 2023; Mercedes-Feliz et al. 2023).

In many cases, these additional models remain under active development and exploration within the FIRE collaboration, and we anticipate including simulations that model them in future data releases. We caution users about interpreting properties that may be sensitive to these physical processes: for example, the lack of MHD will underestimate small-scale magnetic pressure in the ISM, which could bias the properties of a structure like a GMC; see Section 2.4 for more discussion.

All of the simulations that we release used the same FIRE-2 physics model (with minor variations as we describe). While this provides a self-consistent suite, it does not allow a user to explore the effects of different astrophysical models or model parameters. That said, we released the cosmological initial conditions and Gizmo configuration and parameter files for nearly all of these simulations, and a version of the Gizmo source code is publicly available. Therefore, although the code including the full FIRE-2 physics is not presently in the public domain, users have access to tools necessary to (re)run simulations with model variants, including restarting simulations from the released snapshots.

All FIRE-2 cosmological simulations zoom in on a selected region at high resolution, embedded within a lower-resolution cosmological background (see Oñorbe et al. 2014). We first ran low-resolution dark-matter-only simulations within uniform-resolution cosmological boxes, then we selected regions of interest at z = 0 for the Core suite, z = 1 for the Massive Halo suite, or z = 5 for the High Redshift suite. We then chose a spherical volume centered on one halo (or a pair of halos) of interest. For most FIRE-2 simulations, this region extends 4–8 R_200m around the halo(s), where R_200m is the radius within which the mean density of the halo is 200 times the mean matter density of the universe. We then traced the particles in this region back to z ≈ 99 and regenerated the encompassing convex hull at high resolution using MUSIC (Hahn & Abel 2011). We resimulated the zoom-in region at high resolution, including dark matter, gas, and star formation, while the lower-resolution cosmological box that encompasses it contains only dark matter at low resolution. As a result of using a convex hull to set/encompass the initial-condition volume (to help ensure its regularity and smoothness), the geometry of the zoom-in region at lower redshifts can be irregular and nonspherical. By design, the primary halo(s) in each zoom-in region have zero contamination from low-resolution dark matter out to at least R_200m and typically much farther.

Each cosmological simulation zoom region is typically one to a few megaparsecs in size. Except for the ELVIS on FIRE simulations, we centered each zoom-in region on a single primary halo that we chose to be cosmologically isolated from halos of similar or greater mass, purely to limit computational cost. (Oñorbe et al. 2014 showed via dark-matter-only simulations that the Lagrangian volume of the initial conditions of a halo does not bias its properties at z = 0, though the effects of the initial conditions on galaxy properties in baryonic simulations remain less explored.) Thus, an important caveat is that these simulations do not fairly sample the full range of cosmological environments. For example, they do not sample the densest regions that a halo can inhabit, and there are no satellites of massive galaxy groups or clusters, nor "splashback" galaxies that ever orbited within them. Furthermore, these simulations do not sample the lowest-density regions that probe the typical intergalactic medium (IGM).

We chose most primary halos at particular mass scales, e.g., M_200m(z = 0) ∼ 10⁹, 10¹⁰, 10¹¹, and 10¹² M_⊙ for the Core suite, so the primary halos/galaxies in these simulations do not fairly sample the full halo/galaxy mass function. In particular, we chose these systems initially based on their final dark-matter halo mass, so while the selection function of halo masses is well defined, the selection function of galaxy stellar masses is not, because of scatter in the relation between galaxy stellar mass and halo mass. So, a set of primary galaxies at a given stellar mass does not necessarily sample the full range of halo masses that could form such galaxies.

Analyzing zoom-in simulations is different than analyzing a larger-volume, uniform-resolution cosmological simulation, like Illustris or EAGLE. In particular, while each zoom-in simulation contains particles across the entire cosmological volume (typically 86–172 Mpc along each spatial dimension), the volume outside of the zoom-in region contains only low-resolution dark-matter particles. Generally, a user should analyze only high-resolution particles (which is straightforward, because Gizmo stores low-resolution dark-matter particles as a separate particle type; see Section 4.2) that are safely within the zoom-in region. As a simulation progresses, the zoom-in region inevitably develops a boundary region at its edge that contains overlapping high- and low-resolution particles, so one must use caution in analyzing high-resolution particles near the edge of the zoom-in region. To make this easier, our default Rockstar halo HDF5 files (see Section 4.5) include the total mass of low-resolution dark-matter particles within each halo. We recommend analyzing only galaxies within halos whose fraction of total mass in low-resolution particles is less than a few percent.

2.4.1. High Redshifts

2.4.2. Numerical Limitations in Resolution

2.4.3. Known Tensions with Observations

Table 1 lists the Core suite of FIRE-2 simulations run to z = 0, including the properties of each primary halo/galaxy at z = 0. We release 39 full snapshots across z = 0−10. Specifically, we release 19 snapshots across z = 1−10 spaced every Δz = 0.5, nine snapshots across z = 0.1−1 spaced every Δz = 0.1, and 11 snapshots spaced every Δt ≈ 2.2 Myr just prior to and including z = 0. Table 1 also lists the published article that introduced each simulation at the stated resolution. We request anyone who uses a given simulation to cite its relevant publication.

Table 1. Core Suite of 23 Primary Galaxies/Halos across 20 Different Simulations to z = 0

Name	M200m	R200m	Mstar,90	mbaryon	mdm	${\epsilon }_{\mathrm{gas},\ \min }$	star	dm	Ndm	Size	Cosmology	Reference
	(M⊙)	(kpc)	(M⊙)	(M⊙)	(M⊙)	(pc)	(pc)	(pc)		(GB)
m09	2.60 × 109	44	1.5 × 104	32	160	0.4	0.7	14	2.31 × 108	17	N	W19
m10q	8.23 × 109	64	4.8 × 106	33	160	0.4	0.7	14	1.24 × 108	11	A	W19
m10v	1.08 × 1010	70	3.1 × 105	33	160	0.1	0.7	14	2.94 × 108	25	A	W19
m11b	4.65 × 1010	114	4.5 × 107	2100	10,000	1.0	2.6	26	2.24 × 107	3.1	A	C18
m11i	7.77 × 1010	133	9.2 × 108	7100	39,000	1.0	4.0	40	4.59 × 106	0.9	P	E18
m11q	1.63 × 1011	174	3.7 × 108	880	4400	0.5	2.0	20	1.30 × 108	11	A	H18
m11e	1.68 × 1011	171	1.4 × 109	7100	39,000	1.0	4.0	40	1.12 × 107	2.0	P	E18
m11h	2.07 × 1011	184	3.6 × 109	7100	39,000	1.0	4.0	40	1.58 × 107	2.8	P	E18
m11d	3.23 × 1011	213	3.9 × 109	7100	39,000	1.0	4.0	40	1.61 × 107	2.8	P	E18
m12z	9.25 × 1011	307	2.0 × 1010	4200	21,000	0.4	3.2	33	1.27 × 108	14	Z	G19a
m12w	1.08 × 1012	319	5.7 × 1010	7100	39,000	1.0	4.0	40	7.33 × 107	8.4	P	S20
m12r	1.10 × 1012	321	1.7 × 1010	7100	39,000	1.0	4.0	40	6.03 × 107	6.2	P	S20
m12i	1.18 × 1012	336	6.3 × 1010	7100	35,000	1.0	4.0	40	7.05 × 107	7.2	A	W16
m12c	1.35 × 1012	351	5.8 × 1010	7100	35,000	1.0	4.0	40	1.51 × 108	16	A	G19a
m12b	1.43 × 1012	358	8.5 × 1010	7100	35,000	1.0	4.0	40	7.45 × 107	8.4	A	G19a
m12m	1.58 × 1012	371	1.1 × 1011	7100	35,000	1.0	4.0	40	1.41 × 108	17	A	H18
m12f	1.71 × 1012	380	7.9 × 1010	7100	35,000	1.0	4.0	40	9.62 × 107	9.1	A	G17
Juliet	1.10 × 1012	321	3.8 × 1010	3500	19,000	0.7	4.4	32	3.06 × 108	33	P	G19a
Romeo	1.32 × 1012	341	6.6 × 1010	3500	19,000	0.7	4.4	32	3.06 × 108	33	P	G19a
Louise	1.15 × 1012	333	2.6 × 1010	4000	20,000	0.4	2.7	31	3.80 × 108	42	E	G19a
Thelma	1.43 × 1012	358	7.1 × 1010	4000	20,000	0.4	2.7	31	3.80 × 108	42	E	G19a
Remus	1.22 × 1012	339	4.6 × 1010	4000	20,000	0.4	3.8	31	2.97 × 108	33	E	G19b
Romulus	2.08 × 1012	406	9.1 × 1010	4000	20,000	0.4	3.8	31	2.97 × 108	33	E	G19b

Notes. We release 39 full snapshots across z = 0–10. Each cosmological simulation zooms in on a single isolated halo, except the last set (ELVIS on FIRE suite) for which each simulation zooms in on a LG-like MW+M31-mass pair (Romeo and Juliet, Thelma and Louise, Romulus and Remus). We simulated m09, m10q, m10v, and m11b without spurious cosmic-ray heating at z ≳ 10 (see Section 3.1). We list the following properties for each galaxy/halo at z = 0. name: this generally indicates the (log) halo mass, to order of magnitude. M_200m and R_200m: total mass and spherical radius within which the mean density is 200× the matter density of the universe. M_star,90: stellar mass within a spherical radius that encloses 90% of the stellar mass within 20 kpc. m_baryon and m_dm: initial masses of baryonic (gas or star) and dark-matter particles; gas cells can be up to 3 times more massive than this, because they gain mass from stellar ejecta/winds; for star particles this represents the typical mass at formation, but because of stellar mass loss the typical star particle is ≈30% smaller than this. ${\epsilon }_{\mathrm{gas},\min }$: minimum adaptive force softening (Plummer equivalent) for gas cells (equals the hydrodynamic smoothing kernel). _star and _dm: force softening (Plummer equivalent) for star and dark-matter particles. N_dm: number of high-resolution dark-matter particles in the zoom-in region; the total number of particles (including gas and stars) is approximately twice this. Size: total size in GB of each simulation snapshot (for some simulations, a snapshot is stored across multiple file blocks). Cosmology: cosmological parameters used in the simulation, as follows: A ("AGORA": Ω_m = 0.272, Ω_Λ = 0.728, Ω_b = 0.0455, h = 0.702, σ₈ = 0.807, n_s = 0.961); P ("Planck": Ω_m = 0.31, Ω_Λ = 0.69, Ω_b = 0.048, h = 0.68, σ₈ = 0.82, n_s = 0.97); N (Ω_m = 0.266, Ω_Λ = 0.734, Ω_b = 0.044, h = 0.71, σ₈ = 0.801, n_s = 0.963); E (Ω_m = 0.266, Ω_Λ = 0.734, Ω_b = 0.0449, h = 0.71, σ₈ = 0.801, n_s = 0.963). Z (Ω_m = 0.2821, Ω_Λ = 0.7179, Ω_b = 0.0461, h = 0.697, σ₈ = 0.817, n_s = 0.9646). Reference: the published article that introduced the simulation at this resolution. We request any user of a given simulation to cite the accompanying article: W16: Wetzel et al. (2016); G17: Garrison-Kimmel et al. (2017); H18: Hopkins et al. (2018a); C18: Chan et al. (2018); E18: El-Badry et al. (2018a); W19: Wheeler et al. (2019); G19a: Garrison-Kimmel et al. (2019a); G19b: Garrison-Kimmel et al. (2019b); S20: Samuel et al. (2020).

Download table as:  ASCIITypeset image

Except for the last set, we name the simulations according to the (log) mass of the primary host halo at z = 0 (the letter in the name is arbitrary). We selected these halos at z = 0 based on their dark-matter halo mass, and an additional isolation criterion of having no neighboring halos of similar mass (typically ≳30%) within at least (typically) ≈5 R_200m, motivated purely by limiting computational cost.

The bottom two sets in Table 1 represent our suite of MW/M31-mass galaxies. Simulations named m12* (except m12z) we generated as part of the Latte suite (introduced in Wetzel et al. 2016) of halos with M_200m(z = 0) = 1−2 × 10¹² M_⊙. We reemphasize that their selection was agnostic to any halo properties beyond mass, including formation history, concentration, spin, or satellite/subhalo population. m12z is similar, although at slightly lower mass and better resolution. Simulations in the last set are part of the ELVIS on FIRE suite of LG-like MW+M31-mass pairs (Romeo and Juliet, Thelma and Louise, Romulus and Remus). Their selection at z = 0 consists of (Garrison-Kimmel et al. 2019a) (i) two neighboring halos, each with a mass M_200m = 1−3 × 10¹² M_⊙, (ii) total pair mass of M_200m = 2–5 × 10¹² M_⊙, (iii) halo center separation of 600–1000 kpc, (iv) relative halo radial velocity v_rad < 0km s⁻¹, and (v) no other massive halo within 2.8 Mpc of either host center. These criteria do not constrain the larger-scale environment around these halos.

Given that users may be interested in comparing these simulations against the MW (and M31), we note that, among the Latte suite, m12i, m12f, m12m, and m12b are probably the most similar to the MW across a range of properties: Sanderson et al. (2020) showed that m12i, m12f, and m12m have broadly similar stellar masses, scale radii, scale heights, and gas fractions as the MW. Among the ELVIS on FIRE suite, the thinnest, most MW-like disks are Romeo, Romulus, and Remus. Relative to the Latte suite, the galaxies in the ELVIS on FIRE suite tend to form stars earlier (Santistevan et al. 2020), form larger disks (Bellardini et al. 2022), and their disks start to form/settle earlier (Yu et al. 2021, F. McCluskey et al. 2023, in preparation). In particular, Romeo is the earliest-forming galaxy/disk in the suite. This may be relevant given that the MW's disk shows evidence for early formation (e.g., Belokurov & Kravtsov 2022; Conroy et al. 2022).

As Table 1 shows, these simulations used similar but slightly different assumed cosmologies (generally for comparison with specific previous studies), encompassing the ranges Ω_m = 0.266–0.31, Ω_Λ = 0.69–0.734, Ω_b = 0.044–0.48, σ₈ = 0.801–0.82, n_s = 0.961–0.97, and w = −1, generally consistent with Planck Collaboration et al. (2020). Some simulations used the cosmological box from the AGORA project (Kim et al. 2014). The differences in growth histories from differing cosmological parameters are generally small compared with halo-to-halo variations.

Table 2 lists the Massive Halo suite of FIRE-2 simulations run to z = 1, including the properties of each primary halo/galaxy at z = 1. We release 19 full snapshots across z = 1–10, spaced every Δz = 0.5. We request any user of these simulations to cite Anglés-Alcázar et al. (2017b).

Table 2. Massive Halo Suite of Four Primary Galaxies/Halos Simulated to z = 1; we Release 19 Full Snapshots across z = 1–10

Name	Mvir	Rvir	Mstar	mbaryon	mdm	${\epsilon }_{\mathrm{gas},\ \min }$	star	dm	Ndm	Size	Cosmology	Reference
	(M⊙)	(kpc)	(M⊙)	(M⊙)	(M⊙)	(pc)	(pc)	(pc)		(GB)
A1	3.92 × 1012	247	2.75 × 1011	3.3 × 104	1.7 × 105	0.7	7	57	3.52 × 107	7.3	Z	AA17
A2	7.75 × 1012	310	4.10 × 1011	3.3 × 104	1.7 × 105	0.7	7	57	1.13 × 108	23	Z	AA17
A4	4.54 × 1012	260	2.34 × 1011	3.3 × 104	1.7 × 105	0.7	7	57	6.44 × 107	13	Z	AA17
A8	1.27 × 1013	359	5.36 × 1011	3.3 × 104	1.7 × 105	0.7	7	57	1.42 × 108	29	Z	AA17

Notes. Properties are as in Table 1, with the following exceptions. We list galaxy/halo properties at z = 1. M_vir and R_vir refer to the evolving virial overdensity definition for a halo from Bryan & Norman (1998), and M_star is the total stellar mass within 0.1 R_vir. We request any user of these simulations to cite Anglés-Alcázar et al. (2017b, AA17).

Download table as:  ASCIITypeset image

We selected these halos from the A-series of the FIRE-1 MassiveFIRE suite (Feldmann et al. 2016, 2017) to cover a range of formation histories for halo mass M_vir ≈ 10^12.5 M_⊙ at z = 2. Refer to Feldmann et al. (2016, 2017) regarding the selection strategy and halo growth histories of the MassiveFIRE simulations.

In addition to being resimulated with the FIRE-2 model, these Massive Halo simulations include a model for the growth of massive black holes, based on gravitational torques between the stellar and gas components (Hopkins et al. 2011; Anglés-Alcázar et al. 2017a). However, these simulations do not include AGN feedback from black holes, so they form overly massive galaxies with ultradense nuclear stellar distributions at late times (see Section 2.4). This is the key reason we simulated these galaxies only to z = 1.

Unlike all other FIRE-2 simulations in this data release, these Massive Halo simulations do not include a model for subgrid turbulent diffusion of metals in gas.

Table 3 lists the High Redshift suite of FIRE-2 simulations run to z = 5, including the properties of each primary halo/galaxy at z = 5. We release 11 full snapshots across z = 5–10, spaced every Δz = 0.5. We request any users to cite Ma et al. 2018a, 2019, and/or Ma et al. 2020a.

Table 3. High Redshift Suite of 22 Primary Galaxies/Halos Simulated to z = 5; we Release 11 Full Snapshots across z = 5−10

Name	Mvir	Rvir	Mstar	mbaryon	mdm	${\epsilon }_{\mathrm{gas},\ \min }$	star	dm	Ndm	Size	Cosmology	Reference
	(M⊙)	(kpc)	(M⊙)	(M⊙)	(M⊙)	(pc)	(pc)	(pc)		(GB)
z5m12b	8.7 × 1011	51.2	2.6 × 1010	7100	3.9 × 104	0.42	2.1	42	3.5 × 107	8.3	P	M18
z5m12c	7.9 × 1011	49.5	1.8 × 1010	7100	3.9 × 104	0.42	2.1	42	3.3 × 107	8.6	P	M19
z5m12d	5.7 × 1011	44.5	1.2 × 1010	7100	3.9 × 104	0.42	2.1	42	2.3 × 107	7.6	P	M19
z5m12e	5.0 × 1011	42.6	1.4 × 1010	7100	3.9 × 104	0.42	2.1	42	1.9 × 107	6.3	P	M19
z5m12a	4.5 × 1011	41.1	5.4 × 109	7100	3.9 × 104	0.42	2.1	42	1.7 × 107	4.8	P	M18
z5m11f	3.1 × 1011	36.4	4.7 × 109	7100	3.9 × 104	0.42	2.1	42	1.2 × 107	4.0	P	M19
z5m11e	2.5 × 1011	33.6	2.5 × 109	7100	3.9 × 104	0.42	2.1	42	9.6 × 106	2.6	P	M18
z5m11g	2.0 × 1011	31.2	1.9 × 109	7100	3.9 × 104	0.42	2.1	42	7.7 × 106	2.6	P	M19
z5m11d	1.4 × 1011	27.5	1.6 × 109	7100	3.9 × 104	0.42	2.1	42	4.8 × 106	1.7	P	M18
z5m11h	1.0 × 1011	24.9	1.6 × 109	7100	3.9 × 104	0.42	2.1	42	3.8 × 106	1.6	P	M19
z5m11c	7.6 × 1010	22.7	9.5 × 108	891	4900	0.28	1.4	21	2.0 × 107	9.1	P	M20
z5m11i	5.2 × 1010	20.0	2.8 × 108	891	4900	0.28	1.4	21	1.3 × 107	6.3	P	M20
z5m11b	4.0 × 1010	18.3	1.7 × 108	891	4900	0.28	1.4	21	1.1 × 107	5.2	P	M18
z5m11a	4.2 × 1010	18.6	1.2 × 108	954	5200	0.28	1.4	21	1.0 × 107	5.3	P	M18
z5m10f	3.3 × 1010	17.2	1.6 × 108	954	5200	0.28	1.4	21	8.4 × 106	3.8	P	M18
z5m10e	2.6 × 1010	15.8	3.9 × 107	954	5200	0.28	1.4	21	7.5 × 106	3.3	P	M18
z5m10d	1.9 × 1010	14.2	4.8 × 107	954	5200	0.28	1.4	21	4.7 × 106	2.2	P	M18
z5m10c	1.3 × 1010	12.7	5.6 × 107	954	5200	0.28	1.4	21	3.2 × 106	1.5	P	M18
z5m10b	1.2 × 1010	12.4	3.4 × 107	954	5200	0.28	1.4	21	3.1 × 106	1.4	P	M18
z5m10a	6.6 × 109	10.0	1.5 × 107	119	650	0.14	0.7	10	1.2 × 107	9.1	P	M20
z5m09b	3.9 × 109	8.4	2.8 × 106	119	650	0.14	0.7	10	7.5 × 106	3.8	P	M18
z5m09a	2.4 × 109	7.1	1.6 × 106	119	650	0.14	0.7	10	4.8 × 106	2.3	P	M18

Notes. Properties are as in Table 1, with the following exceptions. We list galaxy/halo properties at z = 5. M_vir and R_vir refer to the evolving virial overdensity definition for a halo from Bryan & Norman (1998), and M_star is the total stellar mass within R_vir. We request any user of these simulations to cite Ma et al. (2018a, 2019) and/or Ma et al. (2020a).

Download table as:  ASCIITypeset image

We generated these simulations for studying galaxies at the Epoch of Reionization (see Ma et al. 2018a, 2018b, 2019, 2020a). We selected these halos across a mass range of M_vir ≈ 10⁹–10¹² M_⊙ at z = 5 from cosmological volumes of size (11 Mpc)³ and (43 Mpc)³. Including both the primary galaxy and all lower-mass (satellite) galaxies within each zoom-in region, this High Redshift suite contains about 2000 resolved galaxies at z = 5.

The FIRE-2 simulations are available via the Flatiron Institute Data Exploration and Comparison Hub (FlatHUB), at the following website: http://flathub.flatironinstitute.org/fire. FlatHUB provides two ways to access the data. First, using the website above, a user can click on the "Browse" box to access each suite and simulation via the browser. We recommend this method to browse the available data and download a small amount of it. Second, the FlatHUB website above also provides a Globus ID for transferring via Globus. ²² We recommend using Globus, especially when transferring a large amount of data.

Within a given simulation directory, all simulation snapshots are in a directory named output/. Gizmo stores each snapshot via HDF5 file(s). For simulations with fewer particles, we store each snapshot as a single HDF5 file, named snapshot_NNN.hdf5, while we split larger simulations into multiple HDF5 files within a directory named snapdir_NNN/.

NNN is the snapshot index, which increases with time and ranges from 0 to 600. For example, at integer redshifts the snapshot indices are as follows: z = 0 (600), z = 1 (277), z = 2 (172), z = 3 (120), z = 4 (88), z = 5 (67), z = 6 (52), z = 7 (41), z = 8 (33), z = 9 (26), and z = 10 (20). Each simulation directory contains a file named snapshot_times.txt that lists the index, scale factor, redshift, age of the universe (in billions of years), lookback time (in billions of years), and time spacing since the previous snapshot (in millions of years) of all snapshots written (up to 600). Each suite at its top level also contains a file named snapshot_times_public.txt that lists only the snapshots that we publicly release.

Each snapshot contains four types of particle species: gas cells (stored as type 0), stars (stored as type 4), and dark matter at high resolution (stored as type 1), all of which exist only in the zoom-in region, as well as low-resolution dark matter (stored as type 2) that exists across the entire cosmological box.

Each snapshot file contains an HDF5 header that includes useful information about the simulation and the contents of the snapshot, including the number of particles of each species, units, cosmological parameters, and so on. Two of the most important (for unit conversions below) are the scale factor of the snapshot, a, stored in the header as Time, and the dimensionless Hubble parameter, h, stored in the header as HubbleParam.

Below we list all properties stored for each particle species, along with their units within the snapshot file. However, we strongly encourage anyone to use one of the publicly available python reader/analysis packages that we list in Section 5, which automatically convert all quantities into more common and useful units. For more extensive documentation on the contents of snapshots, refer to the Gizmo Users Guide ²³ (see Section 6).

Each Gizmo snapshot stores the following properties for all particles, with the following names and units. Any quantities listed without units are dimensionless.

• 1.  
  ParticleIDs: indexing starts at 0 (not necessarily unique for star particles and gas cells; see below).
• 2.  
  ParticleChildIDsNumber and ParticleIDGenerationNumber: star particles and gas cells have these two additional IDs to track them uniquely. As a gas cell inherits mass from stellar feedback, to ensure mass balancing Gizmo splits it into two if it exceeds 3 times its initial mass, which means that multiple gas cells and/or star particles can have the same ParticleID (a star particle inherits its IDs from its progenitor gas cell). Thus, Gizmo stores these two additional IDs, initialized to 0 at the start of the simulation. Each time a gas cell splits in two, one cell retains the same ParticleChildIDsNumber, the other cell gets ParticleChildIDsNumber +=2 ^{ParticleIDGenerationNumber} . Both cells then get ParticleIDGenerationNumber +=1. Because Gizmo stores ParticleChildIDsNumber as a 32-bit integer, this allows for a maximum of 30 splittings, then ParticleChildIDsNumber aliases back to 0 and is no longer unique.
• 3.  
  Coordinates (h⁻¹ kpc comoving): 3D positions; multiply by the scale factor a to convert to physical position.
• 4.  
  Velocities $(\sqrt{a}$ km s⁻¹): 3D velocities; multiply by $\sqrt{a}$ to convert to physical/peculiar velocity.
• 5.  
  Masses (10¹⁰ h⁻¹ M_⊙): multiply by 10¹⁰ h⁻¹ to convert to M_⊙.
• 6.  
  Potential (km² s⁻²): gravitational potential with arbitrary normalization (stored for most MW-mass simulations).
• 7.  
  OStarNumber: simulations in the Core suite that used baryonic mass resolution ≈30 M_⊙ (m09, m10q, m10v) used stochastic sampling of massive stars (>8 M_⊙), so this indicates the number of such stars in a given star particle.

Star particles and gas cells also store their elemental abundances:

• 1.  
  Metallicity: 11 elemental abundances, stored as linear mass fractions, with the following order/indices: H (0), He (1), C (2), N (3), O (4), Ne (5), Mg (6), Si (7), S (8), Ca (9), Fe (10). ²⁴

Star particles also store the following:

• 1.  
  StellarFormationTime: scale factor at formation.

Gas cells also store the following:

• 1.  
  Density (10¹⁰ h² a⁻³ M_⊙ kpc⁻³): defined via the cell's mass and the cell's SmoothingLength.
• 2.  
  InternalEnergy (km² s⁻²): specific internal energy; use to compute temperature.
• 3.  
  SmoothingLength (h⁻¹ kpc comoving): full extent of the neighbor interaction kernel (radius of compact support).
• 4.  
  ElectronAbundance: mean number of free electrons per hydrogen nucleus.
• 5.  
  NeutralHydrogenAbundance: fraction of hydrogen that is neutral.
• 6.  
  Star FormationRate (M_⊙ yr⁻¹): instantaneous rate of star formation.

Black hole particles (only modeled in the Massive Halo suite) also store the following:

• 1.  
  BH_Mass(10¹⁰ h⁻¹ M_⊙): mass of the black hole (not necessarily the total mass of the particle; see below).
• 2.  
  BH_Mass_AlphaDisk (10¹⁰ h⁻¹ M_⊙): mass in the subgrid viscous accretion disk.
• 3.  
  BH_Mdot (10¹⁰ M_⊙ yr⁻¹): instantaneous rate of accretion.
• 4.  
  BH_AccretionLength (h⁻¹ kpc comoving): full extent of the neighbor accretion radius (kernel length).
• 5.  
  BH_NProgs: cumulative number of black holes that merged into this one (0 if none).

Each simulation directory contains the following files that specify the settings that Gizmo used when compiling and running the simulation: gizmo_config.h lists the compile-time configuration settings and gizmo_parameters.txt lists the run-time parameters.

For each simulation, we include its cosmological initial conditions ²⁵ in a directory named initial_condition/. This contains the MUSIC configuration files, which list the full cosmological parameters, and the initial-condition file at z ≈ 99, named *.ics.

Each simulation includes a catalog of (sub)halos and their galaxies at each snapshot, within a directory named halo/.

For each simulation, we generated (sub)halo catalogs via Rockstar (Behroozi et al. 2013), We provide these as our default and recommended galaxy/halo catalogs. We used a slightly modified version ²⁶ of Rockstar-Galaxies, ²⁷ which is a version of Rockstar with support for multimass and multispecies particles. We used the same Rockstar parameters for all simulations, and we provide the input configuration file that we used, named rockstar_config.txt, within the top-level directory of each suite.

We ran Rockstar-Galaxies using only dark-matter particles, because we found this led to better numerical stability, especially for subhalos. We therefore place these files in a directory named rockstar_dm/, to reinforce that we generated the halo catalogs using only dark-matter information. Thus, any (sub)halo properties in the catalog are measured using only the dark-matter particles (ignoring stars and gas). We then assigned star particles to these (sub)halos in post-processing, using HaloAnalysis ²⁸ (Wetzel & Garrison-Kimmel 2020a); for more details, see Samuel et al. (2020). We store these (sub)halo catalogs in a converted HDF5 format, named halo_NNN.hdf5, and corresponding galaxy stellar and star-particle information for each (sub)halo is in star_NNN.hdf5, where NNN is the snapshot index, all within a directory named catalog_hdf5/. The Appendix lists the halo/galaxy properties in these HDF5 files.

We strongly recommend using these HDF5 halo/galaxy files. For completeness, however, we also provide the ASCII text files, out_NNN.list, that Rockstar directly outputs in a separate directory named catalog/; see the documentation from Rockstar-Galaxies regarding the contents of these files.

For the Massive Halo and High Redshift suites, we also generated (sub)halo catalogs using the Amiga Halo Finder (AHF; Knollmann & Knebe 2009), which reside in the directory named AHF/ within halo/. We ran AHF simultaneously on all particles, including dark matter, gas, and stars. AHF uses an isodensity contour to identify a halo center, and we defined the halo boundary via a spherical overdensity with a virial radius given by the redshift-dependent virial overdensity definition of Bryan & Norman (1998). The AHF catalogs are in simple text format and contain many properties for (sub)halos, including stellar and gaseous properties; see the AHF file header for more information.

For the Core suite to z = 0, each simulation also contains, within a directory named track/, HDF5 files named star_gas_pointers_NNN.hdf5. Each file contains, for every star particle and gas cell at snapshot 600 (z = 0), a pointer to its index in the particle or cell array at a previous snapshot NNN. Therefore, one can use these pointers easily to track where a star particle or gas cell was in a previous snapshot, or between any two snapshots. We generated these pointers as one cannot simply use ParticleIDs alone to match/track particles, because multiple gas cells and/or star particles can have the same ParticleIDs. Rather, one needs to use ParticleIDs plus ParticleChildIDsNumber and ParticleIDGenerationNumber (see Section 4.2). Thus, the pointers in star_gas_pointers_NNN.hdf5 merely simplify this particle tracking for a user; see the GizmoAnalysis package ²⁹ (Section 5) for more details on using them.

We describe two methods to locate the primary galaxy/halo within each simulation, specifically its position and velocity.

First, one can use the Rockstar (or AHF) halo/galaxy catalogs to find the center, either via the dark matter (using the halo information in halo_NNN.hdf5) or via the stars (using the stellar information in star_NNN.hdf5). To define the primary host, one should use the most massive halo within the zoom-in region that is uncontaminated by low-resolution dark-matter particles. The publicly available HaloAnalysis package for reading the halo/galaxy catalogs (see Section 5) automatically assigns the primary host halo(s) this way during read in (see the Appendix). The (sub)halo catalogs also provide the best way to identify the coordinates of all other (satellite) galaxies within the zoom-in region.

Second, we more commonly use, and therefore most strongly recommend, an iterative zoom-in method using star particles to identify the primary galaxy. Typically we start by measuring the mean center-of-mass position of all star particles. We then keep those within a sphere of some large initial radius (≈1 Mpc) around this center, and using only those star particles we recompute the center position. We iteratively shrink this sphere by ≈50% in radius each time and recompute the center, until the spherical radius drops below some threshold, such as ∼10 pc. Then, we typically compute the center-of-mass velocity of all star particles within some fixed radius of this center, typically ≲8 kpc. The publicly available GizmoAnalysis package for reading snapshots (see Section 5) automatically uses this approach to assign the position and velocity of the primary galaxy(s) to the particle catalog during read in.

Furthermore, the GizmoAnalysis package can identify the orientation of the galaxy, that is, the direction of the disk, if a user sets assign_hosts_rotation = True in the function gizmo.io.Read.read_snapshots(). Specifically, we first identify star particles within 10 kpc of the center of the primary galaxy, and we keep only those that are within a radius that encloses 90% of this total stellar mass, to help remove possible galaxy mergers. Among these, we keep only the 25% youngest, which generally are the most disk-like, and using these we measure the moment-of-inertia tensor to identify the three principal axes of the galaxy. We use this moment-of-inertia tensor to rotate the coordinates into the frame of the disk.

For the Core suite to z = 0, we used the particle-tracking pipeline in GizmoAnalysis to record the coordinates of the primary galaxy(s) at every snapshot. Specifically, we record all star particles within the primary host halo at z = 0, and using only these star particles that end up as part of the host today, we compute the position, velocity, and moment of inertia tensor of the primary galaxy(s) at all previous snapshots. We store these properties in a file name host_coordinates.hdf5 within track/. Specifically, we store the primary galaxy position and velocity in arrays named host.position (kpc comoving) and host.velocity (km s⁻¹), with shape N_snapshot × N_host × N_dimension, where N_snapshot is the total number of snapshots (typically 600); N_host is the number of primary galaxies, which is 1 for all simulations except the ELVIS on FIRE LG-like simulations, for which it is 2; and N_dimension = 3. The rotation tensor, named host.rotation, has shape N_snapshot × N_host × N_dimension × N_dimension.

Thus, when analyzing simulations from the Core suite, we recommend a user to read host_coordinates.hdf5 and use these values to locate the primary galaxy(s) at any snapshot.

For the Core suite to z = 0, the file named host_coordinates.hdf5 within track/ also contains, for each star particle at z = 0, its "formation" coordinates, measured at the snapshot immediately after it formed. Given the snapshot time spacing, this is always ≲25 Myr and more typically ≈10 Myr after formation. We measure the formation coordinates as the 3D Cartesian x, y, z position and velocity centered on the primary galaxy and aligned with its disk orientation (principal component axes of its moment of inertia tensor; see above) at first snapshot after each star particle formed, which can be different for each star particle. Specifically, star.form.host.distance (kpc physical) and star.form.host.velocity (km s⁻¹) have shape N_particle × N_dimension, where N_particle is the number of star particles at z = 0 and N_dimension = 3. (For the ELVIS on FIRE simulations, host_coordinates.hdf5 also stores star.form.host2.distance and star.form.host2.velocity for the second host galaxy.) Thus, one can use these formation coordinate to explore how the positions and orbits of star particles at z = 0 have changed since their formation. One also can use them to identify star particles that formed "ex situ," in another galaxy outside of the primary galaxy, using any desired cut on distance and/or velocity.

To make the identification of ex situ stars even easier, for the Core suite to z = 0, each simulation contains a text file named star_exsitu_flag_600.txt inside track/ that lists, for every star particle at z = 0, a binary flag that is 1 if the star particle formed ex situ, that is, outside of the primary galaxy in another lower-mass galaxy. We define a star particle as "ex situ" following Bellardini et al. (2022), if it formed at a spherical distance d_form > 30 kpc comoving (>30a kpc physical, where a is the expansion scale factor) from the center of the primary galaxy.

The FIRE project website ³⁰ links to several additional public data sets that relate to these FIRE-2 simulations:

• 1.  
  MUSIC cosmological initial-condition files for most of these simulations: http://www.tapir.caltech.edu/~phopkins/publicICs.
• 2.  
  Synthetic Gaia DR2-like surveys for three MW-mass galaxies (m12i, m12f, m12m) from Sanderson et al. (2020): http://ananke.hub.yt. ³¹ We also provide synthetic SDSS-APOGEE catalogs of radial velocities and elemental abundances (Nikakhtar et al. 2021), available as part of the Sloan Digital Sky Survey Data Release 17 ³² (Abdurro'uf et al. 2022), query through CasJobs. ³³
• 3.  
  Catalogs and properties of stellar streams and their progenitor galaxies for the MW-mass simulations from Panithanpaisal et al. (2021): https://flathub.flatironinstitute.org/sapfire.
• 4.  
  Multipole basis expansion models of the mass distribution for the MW-mass halos, from Arora et al. (2022): http://physics.upenn.edu/dynamics/data/pot_models.
• 5.  
  Properties of predicted binary black holes in the m12i MW-mass galaxy from Lamberts et al. (2018): http://ananke.hub.yt.
• 6.  
  Animations, images, and other visualizations: http://www.tapir.caltech.edu/~phopkins/Site/animations.