A new analysis of how stars move in our Galaxy finds that models that add an invisible dark matter halo fit the data far better than popular alternatives that change the laws of gravity. The authors report that Modified Newtonian Dynamics (MOND) is strongly disfavored at more than 13σ (sigma) and another modified theory, Scalar–Tensor–Vector Gravity (STVG), is disfavored at more than 4σ. In contrast, common dark matter halo models (Navarro–Frenk–White and Einasto profiles) produce a consistent fit to both kinds of measurements the team used.

The study combines two independent kinds of measurements that probe different directions of the Galaxy’s gravitational field. One is the radial rotation curve — how fast stars orbit at different distances from the center — built from a sample of 120,309 red giant branch stars assembled from APOGEE spectroscopy plus Gaia, 2MASS and WISE photometry. These distances use improved spectrophotometric parallaxes and cover the anti-center sector from about 6 to 27.5 kiloparsecs from the center. The other constraint comes from a vertical mapping of the gravitational potential obtained from the so-called phase-space “snail” seen in Gaia data. That snail is a spiraling pattern in the vertical position–velocity plane that reveals non-equilibrium motions; the authors use it to measure the vertical restoring force without assuming the Galaxy is in steady state. The vertical potentials are measured around effective radii 7.8 to 11.0 kpc.

To compare theories, the team built a detailed baryonic model of the Milky Way. The stellar disk is taken from recent work on mono-abundance populations and follows a broken-exponential form with a flat inner density and a measured half-light radius of 5.75 ± 0.38 kpc. Bulge and gas components were set to recent observational values. For gravity they tested two dark matter halo shapes (NFW and Einasto) and two modified-gravity frameworks: quasi-linear MOND (QUMOND) with three versions of the interpolation function and a fixed characteristic acceleration a0 = 1.2×10−10 m s−2, and STVG with free strength and range parameters. They numerically solved the relevant equations on an axisymmetric cylindrical grid, scanned parameters with Markov Chain Monte Carlo, and combined the rotation-curve and vertical-potential likelihoods. The authors also included conservative systematic errors: rotation-curve uncertainties varying from about 5% to 20% with radius and up to ~5% systematic uncertainty for the vertical potential.