StarStream on Gaia: Stream discovery and mass loss rate of globular clusters

TL;DR

Gaia DR3 enables a comprehensive 6D view of the Milky Way's globular cluster streams. The authors introduce StarStream, a physics-based, morphology-agnostic detection framework that uses a KDE mixture model and a particle-spray stream generator to identify GC streams and quantify membership with probabilistic scores. They detect 87 GC streams, including 34 high-quality cases, enabling the first unbiased estimates of GC mass loss rates that typically range from 1 to 100 M_⊙ Myr^{-1} and show a positive correlation with cluster size in certain regimes, consistent with BH-rich disruption scenarios. These results imply that tidal streams are common around GCs and provide new observational constraints for GC evolution and Milky Way potential models, with public data and code releases to facilitate further study and validation.

Abstract

We apply the automatic stellar stream detection algorithm StarStream to Gaia Data Release 3 and identify 87 stellar streams associated with Galactic globular clusters (GCs), including 34 high-quality cases with median completeness and purity both exceeding 50%, as estimated from modeling mock streams. These detections double the number of known GC streams, and increase the fraction of GCs with tidal streams at high Galactic latitudes (|b| > 30 degree) to 75%. In contrast to visual expectations, many new streams are wide or short, or misaligned with their progenitors' orbits. Taking advantage of the unbiased density measurements enabled by our method, we also estimate the mass loss rate for the progenitor GCs. We find that several low-mass, large-size clusters have enhanced mass loss rates, indicating that they are approaching complete tidal disruption.

Figures (6)

• Figure 1: Detection quality metrics of StarStream by chen_starstream_2025. Upper row: Purity (magenta) and completeness (cyan) as functions of the progenitor's extinction $A_V$ (left) and background density as characterized by $N_{\rm bg}$ within the $10^\circ$ search radius (right). Lower row: Number of detections in the null test ($N_{\rm null}$, red) as a function of $A_V$ and background density. We also show the number of actual detection when applying StarStream to MW GCs as blue lines, with individual detections shown as circles. Shaded regions represent the 25%--75% ranges, smoothed by a Gaussian kernel with bandwidth $=0.2$ dex for $A_V$ and $0.4$ dex for $N_{\rm bg}$. We show our threshold for high-quality detection, $A_V<0.6$ and $N_{\rm bg}<6\times10^6$, as vertical dashed lines. We also show the horizontal line to indicate the minimum selection threshold $N_{\rm detect}=10$.
• Figure 2: Detections of stream members (blue circles) around 34 MW GCs in the high-quality sample ($A_V<0.6$ and $N_{\rm bg}<6\times10^6$). We show these streams in the great circle frame ($\phi_1$--$\phi_2$) centered on the progenitor GC. Streams are placed in the descending order of the length $r_{90}$. Each star is color-coded by the stream probability, as indicated by the colorbar. The tidal radius of the GC is shown as the brown circle. We show orbits of progenitor GCs as solid brown curves, projected in the same great circle frame. For comparison, we also show the simulated streams (gray symbols).
• Figure 3: Similar to Fig. \ref{['fig:mw_streams']}, but for the color--magnitude space $G$ vs. $\rm BP-RP$.
• Figure 4: Similar to Fig. \ref{['fig:mw_streams']}, but for extended streams with $r_{90}>10^\circ$. Only streams with more than 10 extended detections outside the original $10^\circ$ search radius are shown here, with the extended detections shown as open symbols. The original search radius is marked as dashed circles in each panel.
• Figure 5: Mass loss rates of 34 streams in the high-quality sample, plotted against $M_{\rm GC}$ (left), $\Omega_{\rm tid}$ (middle), and $r_{\rm h}$ (right), with uncertainties shown as errorbars. The best-fit relation for these measurements are shown as light blue shaded regions. For comparison, we also show the BHs and no BHs models from gieles_mass-loss_2023 and the best-fit relations in chen_stellar_2025 as magenta and gray shaded regions, respectively. Note that $\Omega_{\rm tid}$ in chen_stellar_2025 is smaller by a constant $\sqrt{2}$, which we have accounted for in the comparison here.