HEP-Astro Seminar | Mapping the Dark Matter in the Milky Way with Stars: The Signal We Already Have

Dark matter has so far revealed itself in exactly one way: through gravity. In this talk, I will argue that this makes the distribution of dark matter our only positive signal — and that mapping it precisely may be our most direct route to understanding what dark matter is. Galactic dynamics is no longer data-limited but model-limited, and the assumption most in need of revision, underlying nearly all of our models, is that galaxies are in equilibrium. I will present recent work leveraging Gaia DR3 that reveals a declining stellar rotation curve, show how underestimated errors in Jeans modeling bias these results, and how simulations let us build a more robust treatment that avoids tensions with other probes of the density profile. I will then turn to dwarf galaxies, where the kinematic data will not substantially improve and only better inference can extract the physics: I will present GraphNPE, a graph neural network method that recovers dark matter density profiles from sparse stellar kinematics, revealing a hint of a core in Boötes I, consistent with self-interacting dark matter. Finally, I will turn to the local velocity distribution, where the Milky Way's merger history imprints substructure that the Standard Halo Model does not capture, and use mutual information to show that this history is itself progressively erased from stellar dynamics — setting a fundamental limit on what the Galaxy can ever tell us. Together, these results outline a program of inference that assumes no more than the data supports and extracts no less than it contains, arriving just as Gaia DR4 and the next generation of spectroscopic surveys come online.