Dormant BH candidates from Gaia DR3 summary diagnostics

TL;DR

This study demonstrates that Gaia DR3 summary statistics (ruwe, RV scatter, and photometric variability) can identify dormant stellar-mass compact objects in binaries without requiring full orbital solutions. A gaiamock-based forward model combined with MCMC inference maps observed Gaia diagnostics to companion mass $M_\mathrm{CO}$ and orbital period $P$, applied to ~21k RGB giants to produce a robust RGB+BH candidate catalogue and a smaller MS+BH appendix. The method yields 556 RGB+BH candidates with $M_\mathrm{CO}\gtrsim3\,M_\odot$ and shows higher completeness than DR3 NSS astrometric solutions in the intermediate period range (roughly 100–2000 days), though spectroscopic follow-up remains essential for validation and for exploiting Gaia DR4's extended baseline. The work lays a path toward a substantially expanded census of dormant black holes and neutron stars in the Galaxy and highlights the importance of future Gaia data releases for confirming these systems.

Abstract

We present a rigorous identification of candidates for dormant black holes (BHs) and neutron stars (NSs) in binaries using summary statistics from Gaia DR3, rather than full orbital solutions. Although Gaia astrometric orbits have already revealed a small sample of compact object binaries, many systems remain undetected due to stringent quality cuts imposed on the published orbits. Using a forward-modelling framework that simulates Gaia observables, in particular the renormalised unit weight error (ruwe) and radial velocity (RV) scatter, we infer posterior distributions for companion mass and orbital period via MCMC sampling, marginalising over nuisance orbital parameters. We validate our approach by comparing the predicted masses and periods against full orbit solutions from DR3, and by successfully recovering known compact object binaries as promising candidates. The method is best suited for systems with red giant primaries, which have more reliable Gaia RV scatter and a light centroid more likely dominated by one component, compared to main-sequence stars. And they are less likely to be triples with short-period inner binaries, which produce confounding signatures. We apply the method to three million giants and identify 556 systems with best-fit companion masses $\gtrsim 3\,M_\odot$. Recovery simulations suggest our selection method is substantially more sensitive than the DR3 non-single-star catalogue, particularly for binaries with periods below 1 year and above $\sim 6$ years. These candidates represent promising targets for spectroscopic follow-up and Gaia DR4 analysis to confirm the presence of compact objects. Candidate main-sequence stars with massive companions face a larger set of confounding effects. Therefore, we present an analogous catalogue of 279 additional `main sequence' candidates only as an appendix.

Figures (15)

• Figure 1: The strict quality cuts for Gaia DR3 excised many binaries (some with BH companions) from the published Gaia NSS catalogue. The figure illustrates the cut on period vs. parallax over parallax uncertainty for binaries in the DR3 astrometric catalogue and the dormant BH binaries Gaia BH 1, 2 Chakrabarti+2023El-Badry+2023bEl-Badry+2023.
• Figure 2: Predicted vs. catalogued Gaia DR3 summary statistics for known binaries, with the predictions based on forward modelling of their orbital parameters. Top panel: Comparison between predicted and catalogue values of the astrometric goodness of fit of the single-star astrometric solution, ruwe, for binaries with published orbital solutions. Bottom panel: Predicted vs. observed RV "errors" for spectroscopic binaries (SB1) in Gaia DR3. In both plots, the blue line shows the median trend, while the dashed blue lines indicate the 16th and 84th percentiles. The 1:1 relation is marked in black. Lower subpanels show the logarithmic residuals between observed and predicted values, with side histograms illustrating their normalised distribution. The mean bias ($\mu$) and scatter ($\sigma$) of the residuals are indicated.
• Figure 3: Posterior inference for two example binary systems using our forward-modelling approach: a luminous stellar binary (DR3 4609711364265062016, left panels) and a compact-object binary, Gaia BH 2 (DR3 5870569352746779008, right panels). Top row: Posterior samples in the space of companion mass vs. orbital period, along with the marginalised posterior distribution of the companion mass. Orange star symbols mark the true (injected) parameters of each system; blue stars denote the inferred best-fit parameters (posterior-weighted mean). Dashed ellipses indicate the 1$\sigma$ and 2$\sigma$ confidence contours derived from the covariance matrix. Bottom row: predicted observables (ruwe versus RV uncertainty $\sigma_\text{RV}$) for the same posterior samples, colour-coded by the companion mass. The observed values are again shown as orange stars. In the stellar case, low-mass companions reproduce the observed signatures well; in the BH case, a high companion mass is required to explain the elevated ruwe and moderate $\sigma_\text{RV}$, consistent with the true binary configurations. Much of the variance in predicting the observables results from the random sampling of RV epochs, which can produce varying observables also for the same binary parameters.
• Figure 4: CMD of the RGB candidate sample (2D histogram), overlaid on all stars within 100 pc from Gaia DR3 (grey points). Highlighted are confirmed BH binaries Gaia BH 2 and BH 3, and the candidate LAMOST BH. Neutron star binaries (triangles) and Gaia BH 1 are shown for reference, but lie outside our RGB selection.
• Figure 5: Distributions of inferred stellar masses and radii for the RGB candidate sample. Estimates are obtained using an XGBoost gradient boosting model trained on stellar radii and masses derived from asteroseismology. The giant stars in the sample have median masses of $1.3\,\text{M}_\odot$ and typical radii around $10\,\text{R}_\odot$.