A search for black holes with metal-poor stellar companions: I. Survey sample selection and single epoch radial velocity follow-up

TL;DR

This study investigates the link between metal-poor stellar companions and the formation of massive stellar-mass black holes by assembling a Gaia-selected sample of ~890 candidates with astrometric deviations, RV variability, and low metallicity. Through single-epoch high-resolution spectroscopy with MIKE and APF, the authors measure RVs to identify large RV amplitudes or Gaia RV discrepancies, isolating a set of promising BH-candidate targets for future multi-epoch orbital modeling once Gaia DR4 epoch data become available. They quantify contamination from hot stars, LPVs, and hierarchical triples, and compare their findings to other compact-object catalogs, while outlining population-level inferences on BH occurrence rates in metal-poor environments. The work establishes a well-characterized BH-candidate sample and demonstrates a strategy to leverage Gaia DR4 and ground-based RVs to constrain the demographics of BHs with metal-poor companions and to test formation scenarios linking low metallicity to the most massive remnants.

Abstract

Stellar-mass black holes (BHs) above $30 M_\odot$ are predicted to form from low-metallicity progenitors, but direct detections of such systems in the Milky Way remain scarce. Motivated by the recent discovery of Gaia BH3, a $33 M_\odot$ BH with a very metal-poor giant companion, we conduct a systematic search for additional systems. Approximately 900 candidates are identified with Gaia as having significant deviations from single-star astrometric motion, evidence of RV variability, and low metallicities inferred from Gaia XP spectra. We obtain single epoch high-resolution spectra for over 600 of these sources with Magellan/MIKE and Lick/APF and measure independent RVs with $\approx 1$ km s$^{-1}$ precision. After removing contaminants such as hot stars, pulsators, eclipsing binaries, and hierarchical triples, we identify about 15 promising candidates with large RV amplitudes or offsets from the Gaia reported values. This program establishes a well-characterized sample of BH candidates for detailed orbital modeling once Gaia DR4 epoch astrometry and RVs are released in late 2026; multi-epoch RV follow-up is ongoing. Together, the Gaia and ground-based data will place new constraints on the demographics of BHs with metal-poor companions and test theoretical predictions linking low metallicity to the formation of the most massive stellar remnants.

Figures (9)

• Figure 1: Left panel: RV measurements of Gaia BH3 taken with Gaia's Radial Velocity Spectrometer (black points) along with the orbital model determined by Panuzzo:2024 (dashed line). Gaia BH3's RVs in the Gaia DR3 window showed a roughly monotonic increase of $\approx 40$ km s$^{-1}$; RV quadrature occurred during the Gaia DR4 window, revealing the full RV amplitude of $\approx 110$ km s$^{-1}$. Data in Gaia DR5 will nearly close the orbit. Right panel: In general, Gaia DR3 catalogs did not contain epoch RV measurements, only a median RV. This is shown as the horizontal black line; the reported uncertainty is the shaded region. Ground-based follow-up observations of Gaia BH3 in 2024-2025 (red line) would be $15-25$ km s$^{-1}$ different from the Gaia DR3 median RV. This suggests large RV differences from the Gaia DR3 RV such that ground-based follow-up could be used to identify massive BHs in binaries with long orbital periods.
• Figure 2: Right panel: RV semi-amplitude for star + star and star + WD binaries drawn from the population synthesis simulations of El-Badry:2024c. Left panel: RV semi-amplitude of binaries drawn from the Gaia NSS SB1 catalog. For binaries where the orbital period is longer than 3 years, a threshold of K = 15 km s$^{-1}$ will eliminate most stellar binaries and improve the BH detection rate. The main contaminant will be binaries with shorter orbital periods; these can be ruled out with a few epochs of RV follow-up. The number of binaries within each period range differs between the left and right panels because the Gaia NSS processing pipeline is more sensitive to binaries with orbital periods $P < 3$ years, because the temporal baseline of Gaia DR3 is about 3 years.
• Figure 3: Targets, color-coded by their classifications from the observed RVs as discussed in § \ref{['sec:Results']}. Sources with reliable RVs are "Good" (black triangles), those that have reliable Gaia RVs but are classified as variable stars are "Var" (red $+$), those that do not have reliably measured RVs are "Bad" (yellow dots). Gaia BH3 is marked as the pink star. Sources that have orbital solutions from the Gaia NSS catalog are "Orbit" (light blue $\times$) and we do not obtain RV follow-up of those sources. The gray points are a random subset of sources in the gaia_source catalog that have $\varpi > 1$ mas, $\varpi/\sigma_\varpi > 5$, and rv_method_used = 1. Top left: extinction-corrected CMD. Top right: $\Delta \textrm{RV}_{\textrm{Gaia}}$ vs. $\texttt{ruwe}$. Bottom left: on-sky distribution. Bottom right: [M/H] vs. $T_{\textrm{eff}}$ inferred from the XP spectra, as reported in Andrae:2023b.
• Figure 4: Results for the reliably measured RVs; sources that are variable stars are shown in red. Top row: Difference between $v_{\textrm{Gaia}}$ and the follow-up APF/MIKE RV (right), as a function of $v_{\textrm{Gaia}}$ (left panel). The y-error uncertainty is $\sigma_{v_{\textrm{Gaia}}}$ and the APF/MIKE RV error added in quadrature. Bottom row: Difference between $v_{\textrm{Gaia}}$ and the follow-up APF/MIKE RV, normalized by $0.5 \Delta \textrm{RV}_{\textrm{Gaia}}$ (i.e., the RV semi-amplitude robust; left), as a function of $v_{\textrm{Gaia}}$ (right). The y-error uncertainty is $\sigma_{v_{\textrm{Gaia}}}$ and the APF/MIKE RV error added in quadrature, normalized by $0.5 \Delta \textrm{RV}_{\textrm{Gaia}}$. Points that fall within the two gray horizontal lines are sources that have follow-up RVs that fall within Gaia's reported $\Delta \textrm{RV}_{\textrm{Gaia}}$, assuming that amplitude is centered on $v_{\textrm{Gaia}}$. In both the top and bottom panels, there are a few variable sources that fall outside the RV ranges of these plots.
• Figure 5: Gaia's rv_amplitude_robust ($\Delta RV_{Gaia}$) vs. photometric variability in Gaia DR3 $\Delta G$ for Gaia-identified Cepheids (left panel), RR Lyrae (middle panel), and long-period variables (LPVs, right panel). The Gaia variable source catalogs contain $1.5 \times 10^4$ Cepheids, $2.7 \times 10^5$ RR Lyrae, and $1.7 \times 10^6$ LPVs, however, Gaia RVs are only available for bright stars at $G \lesssim 13$, so the samples plotted are only 10%, 0.3%, 15% of the Cepheids, RR Lyrae, and LPVs, respectively. The blue points show these measurements. The black stars show the overlap with our BH3-like sample. The RV variation in our sample of Cepheids and RR Lyrae can all be explained by pulsations as they fall well within the roughly linear trend between RV and photometric variation; binarity is not needed to explain the RVs. LPVs typically have RV semi-amplitudes $\lesssim 10$ km s$^{-1}$Arenou:2023; $>99\%$ of the Gaia LPVs with RV measurements have $\Delta RV < 20$ km s$^{-1}$ which is consistent with this. LPV RV pulsational amplitudes do not show a strong dependence with $\Delta G$ nor $\texttt{ruwe}$. LPVs are a heterogeneous class consisting of fundamental mode pulsators (Miras), long secondary periods, and ellipsoidal binaries Lebzelter:2023Trabucchi:2023. About half of our sample classified as LPVs have $\Delta RV > 20$ km s$^{-1}$, which potentially points to binarity but is difficult to definitively say.