The DREAMS Project: Disentangling the Impact of Halo-to-Halo Variance and Baryonic Feedback on Milky Way Satellite Galaxies

The Big Picture

Imagine trying to understand traffic patterns in a city by studying just ten intersections, then wondering whether your conclusions apply to the whole metropolis. That’s the challenge astronomers face when modeling the Milky Way’s satellite galaxies: tiny, faint dwarf galaxies orbiting our home galaxy, each one a window into the universe’s darkest and smallest structures. For decades, researchers have built computer simulations to predict what these satellites should look like. But a nagging question persists: when simulations disagree with observations, is it because our physics models are wrong, or just because every Milky Way is a little different?

The DREAMS Project goes after this question directly. By simulating 1,024 Milky Way-mass host galaxies (far more than most previous efforts), researchers led by Jonah Rose and Mariangela Lisanti at Princeton set out to measure how much the assumed physics of galaxy formation matters compared to the natural diversity of galaxies themselves. It turns out the messiness of the cosmos, not the messiness of our equations, drives most of what we see in satellite galaxy populations.

Key Insight: The natural variation from galaxy to galaxy swamps the uncertainty from how we model star formation and stellar explosions. Satellite galaxy predictions are more stable than many expected, and observations primarily reflect the diverse formation histories of individual galaxies rather than flaws in the underlying model.

How It Works

Each of the 1,024 host galaxies was simulated using the TNG galaxy-formation model, best known from the IllustrisTNG suite. It’s a widely used physics recipe that specifies how gas cools, stars form, supernovae explode, and black holes grow. To probe the role of physics assumptions, the team ran simulations across a grid of baryonic and cosmological parameters, adjusting everything from how much energy supernovae pump into surrounding gas to the rate of cosmic expansion.

For each simulated galaxy, the researchers tracked four properties of the satellite population:

• Satellite stellar mass function: how many satellites exist at each mass scale
• Radial distribution: how far satellites orbit from their host
• Inner slope of the dark matter density profile: the steepness of the dark matter concentration at each satellite’s core
• Stellar half-light radius: how physically extended each satellite’s stars are

The team compared these predictions against two benchmarks: the SAGA Survey, which catalogs satellite galaxies around 101 nearby Milky Way analogs in the real universe, and a companion set of N-body (dark matter-only) DREAMS simulations that strip away baryonic physics entirely. By stacking results across all 1,024 hosts and across all parameter variations, they could separate the signal of physics uncertainty from the noise of galaxy-to-galaxy variation.

Plot the predicted satellite stellar mass function for different physics settings, and the curves pile nearly on top of one another. But the cloud of galaxy-to-galaxy variation engulfs them all. No matter how you tune the supernova model or cosmological parameters within observationally allowed ranges, the diversity of galaxy formation histories produces a far wider spread than the model uncertainty itself.

One parameter does stand out. Supernova wind energy (the total kinetic energy injected into surrounding gas when massive stars explode) leaves a measurable imprint. Cranking up wind energy suppresses satellite stellar masses and produces more spatially extended stellar distributions. The physics is intuitive: stronger winds push gas outward before it can collapse into stars, leaving satellites less centrally concentrated.

Wind speed, by contrast, has only a minor effect. Parameters governing black holes, cosmic expansion, and other processes show essentially no measurable signal.

Why It Matters

Satellite galaxies are among the best laboratories for probing dark matter on sub-galactic scales. Competing theories (warm dark matter, self-interacting dark matter, various ultralight candidates) all predict different satellite populations. Before anyone can use observations to rule models in or out, we need to know how much observed scatter in satellite properties comes from dark matter physics versus galaxy formation physics.

DREAMS settles that on the galaxy-formation side: the physics uncertainty is small. If observed satellites depart from standard ΛCDM predictions, exotic dark matter is a more likely explanation than poorly modeled stellar explosions or black hole activity.

There’s a practical upshot for computational cosmology, too. Rather than spending compute on ever-finer resolution for a handful of systems, future surveys may gain more from expanding the statistical sample of hosts. Diversity dominates the total prediction uncertainty; model tuning does not.

The DREAMS approach, with its large ensemble and systematic parameter variation, is a proof of concept for moving past one-galaxy-at-a-time simulation. And the comparison against the SAGA Survey supports the idea that the TNG model captures the essential physics at this mass scale.

Bottom Line: With 1,024 simulated Milky Way analogs, the DREAMS Project shows that satellite galaxy properties hold up against uncertainty in baryonic physics. The cosmos’s own diversity, not our model imprecision, is the main source of theoretical scatter when comparing simulations to observations.