Halo Stars Help Put the Milky Way’s Mass Near 1.3 Trillion Suns

Yanrui Zhou and colleagues report a Milky Way virial mass of approximately 1.3 trillion times the Sun’s mass, combining observations of halo stars with an earlier satellite-galaxy analysis. Their study also demonstrates a major complication: measurement errors can make the Galaxy’s outer regions appear more massive unless their effects are calibrated.

The work was submitted to arXiv on November 3, 2026. It is presented as a preprint; the supplied landing page does not establish peer-reviewed publication.

Following stars beyond the disk

The researchers used observations from the Dark Energy Spectroscopic Instrument Milky Way Survey. Their dataset covers the survey’s first three observing years and is described as an internal dataset intended for public release through DESI DR2, not an already public DR2 catalog.

After selection cuts, the analysis included 1,863 blue horizontal branch stars and 2,282 RR Lyrae stars, spanning distances of roughly 20–80 kiloparsecs from the Galactic center. Both stellar classes are useful distance indicators and bright enough to trace the outer stellar halo.

DESI supplies line-of-sight velocities, while Gaia DR3 supplies proper motions—the stars’ apparent movement across the sky. Combined with distance estimates, these measurements describe positions and motions in three dimensions. The authors also excluded the sky footprint of the Sagittarius stream to reduce contamination by stars whose shared motion could violate the model’s equilibrium assumption.

A flexible model, not an assumption-free one

The team applied the empirical distribution function method, or emPDF. Instead of imposing a particular mathematical form on the stars’ distribution, it reconstructs that distribution from the observations under trial gravitational models.

Its guiding idea is that, in a steady-state system, the observed distribution should resemble the distribution obtained by averaging the stars’ motions over their orbits. The researchers seek the gravitational potential that makes those descriptions agree.

That flexibility does not remove every assumption. This analysis assumes spherical symmetry and represents the total mass distribution with a Navarro–Frenk–White profile. It excludes the inner 20 kiloparsecs because a single such profile cannot adequately describe the central disk and other complex structures.

The fitted parameters describe virial mass and concentration. Here, virial mass means the mass inside a radius where the average density is 200 times the universe’s critical density. Concentration describes that radius relative to the profile’s characteristic scale radius.

Simulated observations reveal a bias

To test the method, Zhou and colleagues used AuriDESI mock catalogs derived from six simulated Milky Way-like galaxies. Each was viewed from four observer positions, with repeated realizations incorporating observational errors. These are simulated observations used for calibration, not additional measurements of our Galaxy.

The tests showed that realistic errors tend to inflate the inferred outer mass and lower the fitted concentration. Proper-motion uncertainties are especially important because their effects on inferred velocities grow with distance. The current emPDF implementation does not explicitly incorporate measurement errors into its likelihood; the authors instead correct the results using the mock-derived biases.

They also estimate additional uncertainty from the mocks. Differences between simulated galaxies contribute more scatter than changing the observer’s direction within a galaxy. Four views of each system therefore do not amount to 24 independent galaxy histories.

Why satellites sharpen the answer

After calibration, the two stellar populations give compatible results. Their separate best-fit virial masses are approximately 1.1 trillion solar masses for blue horizontal branch stars and 1.4 trillion for RR Lyrae stars, with substantial overlapping uncertainties.

The stars constrain enclosed mass most directly near their median distances, around 27 kiloparsecs. Their virial-mass estimates extend beyond the observed stellar range through the adopted profile, making the distinction between local constraints and model-based extrapolation important.

For the combined result, the authors added an earlier analysis of 28 satellite galaxies using a simulation-informed distribution function. Satellites supply complementary information at larger radii, helping separate combinations of mass and concentration that the stars alone cannot distinguish well.

The joint best fit is log10(M200c/M☉) = 12.12, with formal statistical uncertainties of −0.05 and +0.05, and reported total uncertainties of −0.04 and +0.06. Those total bounds correspond approximately to 1.2–1.5 trillion solar masses.

The precision remains conditional on the modeling and calibration. The combination assumes the three constraints are independent, the satellite method relies on a simulation-informed orbital distribution, and the stellar calibration samples only six simulated galaxies. The authors identify larger simulation suites, explicit measurement-error modeling and more general Galactic potentials as future improvements—not completed results.