Latte simulations: the Milky Way on FIRE

flying through a Latte cosmological simulation of a Milky Way-like galaxy (m12f), showing its evolution over the last 1 billion years (credit: Phil Hopkins)


flying through another Latte simulation (m12i)

Formation histories, over 13.8 billion years, of 3 galaxies in the Latte suite of FIRE-2 cosmological simulation of Milky Way-like galaxies (m12i, m12f, m12m), showing stars (first version) and gas (second version) out to 400 kpc, including satellite dwarf galaxies. (credit: Shea Garrison-Kimmel)

synthetic real-color image of stars in a Latte cosmological simulation of a Milky Way-like galaxy (m12f), showing thin disk morphology, spiral structure, clustered young star-forming regions, and dust extinction (credit: Phil Hopkins)
Latte: overview

I am leading the Latte suite of FIRE-2 simulations of Milky Way-mass galaxies (Wetzel et al 2016), which is part of the FIRE simulation project.

Our Milky Way provides the best laboratory to study the physics of galaxy formation and to test the nature of dark matter. Excitingly, a new generation of observations is revolutionizing our view of the Milky Way: the Gaia satellite missionHubble Space Telescope (HST) and James Webb Space Telescope (JWST) , Dark Energy Survey (DES), Large Synoptic Survey Telescope (LSST), and a multitude of ground-based spectroscopic surveys (such as RAVESDSS-APOGEE and SDSS-5, GALAH, Gaia-ESO, LAMOST, PRISTINESubaru PFS, 4MOST, WEAVE) are measuring full orbital phase-space distributions (3D position + 3D velocity) and multi-species elemental abundances for 100’s of millions of stars in the Milky Way and its satellite dwarf galaxies. A critical challenge remains in interpreting these unprecedented datasets: the lack of sufficiently realistic (cosmological) theoretical modeling tools.

A primary goal of the Latte simulations is to understand the formation of our own Milky Way galaxy in a cosmological context, by self-consistently modeling the dark-matter + gaseous halo, stellar + gaseous disk, bulge, giant molecular clouds (GMCs), star clusters, stellar halo, and satellites dwarf galaxies of Milky Way-like systems across their full 13.8 billion-year formation. We seek to use these simulations to address key problems in:
galactic archaeology – reconstructing the Milky Way’s entire formation history in full 6-D
near-field cosmology – using stellar and gas dynamics to measure the distribution, and test the particle nature, of dark matter around us

To accomplish this, the Latte simulations are pushing the resolution frontier, simulating Milky Way-like galaxies at ultra-high resolution, with particle masses of 7070 solar masses and spatial resolution down to 1 pc. Thus, Latte achieves high dynamic range to resolve phase structure in the inter-stellar medium (ISM), allowing gas to collapse into resolved giant molecular clouds (GMCs) and allowing stars to form within them in clusters. This resolution also allows us to model star particles self-consistently as single-age mono-abundance stellar populations, facilitating comparisons to stellar populations across the Milky Way. Furthermore, our new Ananke framework (Sanderson et al 2019) allows us to generate synthetic stellar surveys of the Latte simulations to compare directly against the Milky Way.

This combination of (1) ultra-high resolution, (2) state-of-the-art FIRE-2 physics model, and (3) Ananke synthetic survey generation together make the Latte simulation suite an unprecedented theoretical tool for understanding the formation of the Milky Way (as well as Andromeda and similar-mass galaxies) in a cosmological context.

synthetic real-color image of stars in a Latte cosmological simulation (m12i), showing a Milky Way-like galaxy at center surrounded by a realistic population of satellite dwarf galaxies and a stellar halo that includes tidal streams and shells (credit: Phil Hopkins)
Latte: in detail

In more detail, the Latte simulations are a suite of individual (isolated) Milky Way-mass halos that we simulated using the same resolution, cosmology, and FIRE-2 physics model. We selected them from a cosmological volume (from the AGORA code comparison project) of periodic box length 85.5 Mpc with ΛCDM cosmology: Ω_Λ = 0.728, Ω_matter = 0.272, Ω_baryon = 0.0455, h = 0.702, σ_8 = 0.807, n_s = 0.961. From this volume, we select halos at z = 0 based only on their mass (M_200m = 1 – 2 x 10^12 Msun) and an isolation criterion (no neighboring halos of similar mass within at least 5 x R_200m) to limit computational cost. In particular, our selection is agnostic to any halo properties beyond mass and isolation, including formation history, concentration, spin, or subhalo population.

Each Latte simulation represents a cosmological zoom-in region around a single Milky Way-mass halo. This zoom-in region is simulated at full resolution with dark matter, gas, and stars, embedded within the full cosmological volume, which is simulated self-consistently but at low resolution and with only dark matter. The zoom-in region around each MW-mass halo is typically a few Mpc in size, and all of the Latte simulations have zoom-in regions that are uncontaminated by low-resolution dark matter out to at least 600 kpc at z = 0.

The Latte simulation suite has particle mass resolution:

  • 35,000 Msun for dark-matter
  • 7070 Msun for gas
  • 7070 Msun initially for stars (~5000 Msun at z = 0 because of stellar mass loss)

and spatial (force) resolution:

  • 40 pc for dark matter
  • 4 pc for stars
  • 1 pc (minimum) for gas

Wetzel et al 2016 introduced the Latte simulations and showed that they produce a realistic population of satellite dwarf galaxies. Since then, these simulations have been used in over 40 published articles.

We have shown that the Latte simulations produce galaxies that agree with many observed properties of the Milky Way, Andromeda, and similar-mass galaxies at z = 0, including their satellite populations, without any ‘fine-tuning’ of our FIRE-2 physics model, including:

Comparison with the Milky Way and nearby galaxies
Comparison with stellar halos
Comparison with satellite dwarf galaxies


We also have used the Latte simulations to make predictions for:

predictions for the Milky Way and nearby galaxies
predictions for stellar halos
predictions for the dark-matter halo

Read more about Latte and the broader suite of FIRE-2 simulations at the FIRE project website.

Milky Way-like image, looking inward from the solar position, from an Ananake synthetic Gaia survey of a Latte simulation (m12m) (credit: Robyn Sanderson)
Latte: Publicly available data

Full simulation snapshots at z = 0 from 3 of the Latte simulations (m12f, m12i, m12m), including all dark matter, gas, and star particle data, are publicly available via yt Hub at Also available are synthetic Gaia DR2-like surveys created from these 3 simulations.

Binary Black Hole catalog

Our public release of Latte also includes data from our model for the binary black hole population in/around the Milky Way (Lamberts et al 2018), as modeled using one of the Latte simulations (m12i). This represents the first combination of a high-resolution cosmological simulation of a Milky Way-mass galaxy with a binary population synthesis model in this context.  Our model predicts ~1 million binary black holes in/around the Milky Way at z = 0, with a mean mass of 28 Msun. Access our binary black hole catalog via yt Hub at

9 Ananke synthetic Gaia DR2-like surveys from 3 Latte cosmological simulations of Milky Way-like galaxies (credit: Robyn Sanderson)
Ananke: synthetic gaia surveys of Latte simulations

We developed the Ananke framework (Sanderson et al 2019) to generate realistic synthetic stellar surveys from cosmological baryonic simulations. We used Ananke on the Latte simulations to create a set of synthetic stellar surveys that resemble the Gaia satellite mission‘s Data Release 2. We generated 3 synthetic surveys (each using different solar positions) from each of 3 Latte simulations, for a total of 9 synthetic surveys.

Each synthetic Gaia DR2-like survey features:

  • Cosmological formation histories with self-consistent dark-matter and gas accretion, satellite galaxies, and tidal streams/shells.
  • Spatially resolved stellar populations and young star clusters in realistically thin and dynamically complex star-forming disks. The Latte simulations provide sufficient resolution to model kinematically cold, single-age, mono-abundance stellar populations at mass resolution ~5000 Msun at z = 0, comparable to (massive) star clusters. Furthermore, the gas spatial resolution (as small as 1 pc) and temperature range (as cold as 10 K) resolve the formation of giant molecular clouds (GMCs) and the clustered formation of young stars inside of them.
  • Complete stellar populations synthesized using a Kroupa stellar initial mass function (IMF) and PARSEC model stellar isochrones.
  • Self-consistent dust extinction calculated using a constant dust-to-metals ratio, documented for easy adaptation to other extinction models. This provides realistic correlations between (young) stars and gas, including dust extinction near star-forming regions.
  • Simulated Gaia DR2-like uncertainties based on survey averages, with intrinsic values provided for use with more sophisticated error models.

The result is a self-consistent, dust-extincted synthetic survey of each simulated galaxy that leaves intact important observational relationships between gas, dust extinction, stellar populations, and dark matter. Read more about these synthetic Gaia DR2-like surveys and access them via yt Hub at

Spin around an ELVIS on FIRE cosmological simulation of a Local Group-like system (Romeo & Juliet), showing stars. (credit: Shea Garrison-Kimmel)

Formation history of Romeo & Juliet, showing stars (top) and gas (bottom)

elvis: the Local Group on FIRE

In addition to the Latte suite of individual Milky Way-mass galaxies, our full FIRE-2 simulations suite also includes simulations of Local Group-like volumes, which include a pair of Milky Way + Andromeda-mass galaxies at their measured masses, relative distances and velocities. These Local Group-like volumes extend out to several Mpc to model the low-mass (dwarf) galaxy population on the Local Volume.  Shea Garrison-Kimmel led this “ELVIS on FIRE” simulation suite (which is an extension of his previous ELVIS suite of dark-matter-only simulations). ELVIS on FIRE has similar (even better) resolution than the Latte suite: baryonic particles of just 3500 – 4000 solar masses. See Garrison-Kimmel et al 2019 for more details about ELVIS on FIRE.