The DREAMS Project: Disentangling the Impact of Halo-to-Halo Variance and Baryonic Feedback on Milky Way Dark Matter Density Profiles

The Big Picture

Imagine trying to find a book in a library where you don’t know whether the shelves are organized alphabetically, by color, or at random. That’s roughly the challenge facing physicists hunting for dark matter in our galaxy. To detect it, they need to know precisely how dark matter is spread across the Milky Way: how densely it packs at the center, how thinly it spreads at the edges. Scientists call this the dark matter density profile, and getting it right is essential.

The stakes are real. Direct detection searches and gamma-ray telescopes target the galactic center, where the highest concentration of dark matter is expected. But dark matter doesn’t emit light, and its distribution depends on the turbulent history of galaxy formation over cosmic time. If the assumed map is wrong, conclusions about dark matter’s properties could be off by factors of ten or a hundred.

The DREAMS Project team ran 1,024 simulated Milky Way-mass galaxies to get at this problem. They systematically varied the physics of stellar explosions and black hole growth, then measured how sensitive the dark matter density profiles were to each change.

Key Insight: The dominant source of uncertainty in dark matter density profiles is not the physics of how stars explode or black holes grow. It’s the cosmic luck of which particular galaxy you happen to inhabit. Galaxy-to-galaxy variation swamps everything else.

How It Works

The DREAMS Project (Dark Matter and Astrophysics with Machine Learning and Simulations) built its simulations on the IllustrisTNG model, one of the most widely used frameworks for galaxy formation. It treats dark matter, gas, stars, and supermassive black holes together in a fully self-consistent simulation.

The team varied several parameters across all 1,024 runs:

• Supernova wind speed and energy, controlling how aggressively stellar explosions blow gas out of galaxies
• Black hole feedback efficiency and accretion rate, governing how actively galactic nuclei heat their surroundings
• Two cosmological parameters (the matter power spectrum amplitude and slope)

They sampled this space using a Latin hypercube strategy, which spreads test cases evenly across a multi-dimensional parameter space so no two simulations are clones of each other. Each simulation also started from a different random initial condition, capturing both physics variation and the intrinsic scatter built into the universe’s structure.

When the team compared dark matter density profiles across the full sample, halo-to-halo variance consistently dominated over any systematic shift from the feedback parameters. This variance is the natural scatter in how galaxies form: different merger histories, different proximity to galaxy clusters, the basic randomness of structure formation.

Turn up the supernova winds. Turn down the black hole feedback. Tweak the cosmology. The profiles barely budge compared to how much they vary from halo to halo.

One exception stood out. When supernova wind energies were cranked to extreme values, feedback became so powerful it suppressed star formation almost entirely. The remaining halos were nearly collisionless, dominated by dark matter, with almost no baryons (ordinary matter like gas, stars, and dust) left to influence the structure. But these extreme halos don’t produce realistic Milky Way analogs.

A long-standing debate in the field asks whether baryonic feedback creates a flat dark matter core, with roughly constant density at the center, or preserves the sharp cusp that dark matter-only simulations universally predict. In a cusp, density rises steeply inward.

Some simulations with explosive, episodic “bursty” stellar feedback do create cores. Rapid fluctuations in the gravitational potential repeatedly scour out the center. The DREAMS halos tell a different story. Regardless of feedback variation, all 1,024 simulations are consistent with adiabatic contraction, where dark matter responds smoothly to the slow accumulation of baryons at the center. This makes the inner profile denser rather than hollow. IllustrisTNG’s smooth feedback model simply doesn’t generate the violent potential fluctuations needed to carve out a core.

Why It Matters

Every dark matter search using the Milky Way as its laboratory depends on knowing two things: how much dark matter passes through a given volume near Earth, and how fast those particles move relative to our solar system. For years, researchers have worried that uncertainties in baryonic feedback (notoriously hard to model) might be silently dominating the error budget.

The DREAMS results push back on that concern. For Milky Way-mass halos in the IllustrisTNG framework, feedback uncertainty is secondary. The bigger problem is irreducible halo-to-halo variance: our Milky Way is just one realization of a random process. That kind of uncertainty can be addressed statistically, by building better priors over the population of Milky Way-like galaxies and folding in observational constraints from stellar motions and satellite populations.

The work also raises a pointed question for the broader simulation community: do other widely-used models (FIRE, EAGLE, Simba) agree? If those models predict cores where DREAMS predicts cusps, reconciling them matters before dark matter searches can claim real confidence in their assumed density profiles.

Bottom Line: Out of 1,024 simulated Milky Ways with widely varying physics, it’s not the supernova prescription or the black hole model that most shapes the dark matter profile. It’s simply which galaxy you are. This shifts the uncertainty problem for dark matter detection toward population-level statistical approaches.

---