When Streams Curve Away: A Test of Dark Matter from Extragalactic Stellar Stream Populations

When Streams Curve Away: a Test of Dark Matter from Extragalactic Stellar Stream Populations

N. Starkman, J. Nibauer, S. Pearson, S. Wu^‡, L. Necib

The Astrophysical Journal·2026submitted

When Streams Curve Away: a Test of Dark Matter from Extragalactic Stellar Stream PopulationsWhen Streams Curve Awaysubmitted

$Likelihood ratio between CDM and SIDM plotted against the observed fraction of streams with a convex segment, for catalogs of 100, 1,000 and 10,000 streams. The curves steepen with catalog size; at 10,000 streams, either predicted rate is strongly favored.$

The Astrophysical JournalApJ·2026submitted·N. Starkman, J. Nibauer, et al.N. Starkman, J. Nibauer, S. Pearson, et al.N. Starkman, J. Nibauer, S. Pearson, S. Wu^‡, L. Necib·The Astrophysical JournalApJ·2026submitted

The Astrophysical JournalApJ·2026submitted

How often stellar streams curve away from their host galaxy’s center can tell cold from self-interacting dark matter at up to 5σ, with catalogs the size Euclid, Rubin and Roman will provide.

How often stellar streams curve away from their host galaxy’s center can tell cold from self-interacting dark matter at up to 5σ, with catalogs the size Euclid, Rubin and Roman will provide. In a spherical halo a stream cannot curve this way in projection; in a triaxial halo, which cold dark matter predicts, it can. The test needs only the stream’s projected track and the host’s center: no distances, velocities or model of the stream’s progenitor.

1st |

student^‡ 4th | 20262026Context ›

Software

@misc{Starkman+:2026:stream-convexity-rate,
  author        = {Starkman, Nathaniel and Nibauer, Jacob and Pearson, Sarah and Wu, Sirui and Necib, Lina},
  title         = {{When Streams Curve Away: a Test of Dark Matter from Extragalactic Stellar Stream Populations}},
  journal       = {The Astrophysical Journal},
  year          = {2026},
  month         = {sep},
  eprint        = {2609.40057},
  archivePrefix = {arXiv},
  primaryClass  = {astro-ph.GA},
  adsurl        = {https://ui.adsabs.harvard.edu/abs/2026arXiv260940057S},
  abstract      = {The nature of dark matter sets the shape of galactic halos: collisionless cold dark matter (CDM) predicts triaxial halos, while scattering in self-interacting dark matter (SIDM) rounds them toward sphericity. We show that a large catalog of projected stellar stream tracks can statistically constrain halo shapes, and therefore the particle nature of dark matter, without detailed modeling of any individual stream or gravitational potential. We develop a new geometric diagnostic for stellar streams: if the curvature of any segment along a stream points away from the host galaxy's projected center of mass, then that stream falsifies broad classes of potentials, including spherical ones. We call such a segment convex. The diagnostic needs only a projected stream track and a host-center estimate, requiring no kinematics, distances, progenitor model, or fit to individual host potentials. Halo triaxiality sets the rate at which streams show convex segments, making this rate an observable of the halo-shape distribution and the underlying dark-matter model. We forecast that upcoming survey-scale stream morphology catalogs can discriminate between CDM and SIDM at up to $5\sigma$.},
  note          = {submitted}
}

Most of a galaxy’s mass is dark matter, and what dark matter is sets the shape of the halo it forms. If dark matter is cold and collisionless (CDM), halos are triaxial: squashed differently along each of three axes. If its particles scatter off one another (self-interacting dark matter, SIDM), the collisions can round halos to be more spherical.

Stellar streams trace a halo’s shape. A stellar stream forms when the host galaxy’s tides pull apart a satellite galaxy or globular cluster; its stars spread out along nearly the same orbit. In a spherical halo, a stream observed by an external view always appears to bend toward the centre of its galaxy. In a triaxial halo it need not: a segment of it can bend away. We call such a segment convex. Finding one needs only the stream’s path on the sky and where its galaxy’s centre is: no velocities, no distances, and no model of the stream or the halo.

Two simulated streams in a triaxial halo. Drag the globe of viewing directions to turn it, or drag a stream’s square to see the stream from there: red marks where it bends away from the galaxy’s centre. Inside the globe, distances from the centre are on a square-root scale: a point r from the centre is drawn at √(r / 90 kpc) of the globe’s radius. The sky panels are to scale. The full version adds the two test orbits →

Open · FullDownload HTML

How often streams show a convex segment therefore follows how triaxial halos are. In our simulations, streams in CDM halos have a convex segment longer than $10 kpc$ about twice as often as streams in rounder SIDM halos. With enough streams, counting the convex ones tells the two apart.

Paper Figure 6. Likelihood-ratio

Z_{LR}

comparison of CDM and SIDM from the observed rate

λ^_{obs} = K / N

, the

K

streams containing a convex segment out of a catalog of

N

, for

N = 1 0^{2}, 1 0^{3}, 1 0^{4}

(solid). The dashed curve is the

N \approx 2300

streams projected for Euclid by Starkman et al. (2026). Curves use the predicted rates

(λ_{C}, λ_{SI}) \approx (1.3%, 0.67%)

. The statistic

Z_{LR}

measures relative model preference: positive values favor CDM (blue quadrant), negative values favor SIDM (green quadrant), and larger absolute values mean stronger preference. Shading marks the quadrant in which each model is favored; vertical lines mark the two predicted rates, from Figure 5. With

\sim 1 0^{3}

streams, a rate near either prediction begins to favor its corresponding model;

\sim 1 0^{4}

streams strongly distinguish between the two.

OpenDownload HTML · PDF · PNG

Today’s catalogs hold about a hundred streams beyond the Milky Way. Surveys such as Euclid, Rubin and Roman should find thousands (at least $2300$ in Euclid alone), enough to separate CDM from SIDM at up to 5σ.