A GLOBULAR CLUSTER TOWARD M87 WITH A RADIAL VELOCITY < − 1000 km s⁻¹: THE FIRST HYPERVELOCITY CLUSTER

The known Galactic hypervelocity stars all have positive velocities (Brown et al. 2012), as expected if they have been ejected by the central SMBH into the halo. A highly negative velocity for an ejected star would be observed only in the unlikely event that a star was ejected toward the Sun, and quite recently (≲ 8 Myr, given a distance of ∼8 kpc). Our object has a galactic latitude of 74°, and thus is not in the direction of the Galactic Center. In our Hectospec sample, the next most negative velocities are > − 300 km s⁻¹, above the expected escape velocity for Galactic halo stars (Kenyon et al. 2008). Thus the other negative-velocity objects we observe are likely to be stars.

If the object were in a different part of the sky, we could take more seriously the exotic possibility that it might be a hypervelocity star from a nearby galaxy (such as M31; Sherwin et al. 2008). However, its velocity and position are very implausible for an origin from M31 or other nearby galaxies.

If the object is in Virgo, consider that M87 has a systemic velocity of 1307 ± 8 km s⁻¹ (Smith et al. 2000), while the galaxy cluster itself has a mean of 1050 ± 35 km s⁻¹ (Binggeli et al. 1993), so that our object's velocity with respect to M87 and Virgo are about 2300 and 2100 km s⁻¹, respectively.

The object could be confirmed as a GC if resolved with high-resolution optical data. For a first constraint on the half-light radius r_h, we used an archival i-band CFHT image of the field with seeing ∼065. We measured r_h for both HVGC-1 and, as a comparison, several luminous and large (r_h ∼ 20 pc) star clusters in the same field (Brodie et al. 2011). The estimates were made as discussed in Strader et al. (2011b): using ishape (Larsen 1999) we fit King models with fixed c = 30 convolved with a point-spread function made from bright stars near HVGC-1. The r_h ∼ 20 pc objects are clearly resolved and we reproduce the published r_h estimates to within ∼15%–20%. HVGC-1 shows modest evidence for being resolved with r_h ∼ 6 pc. While we believe this measurement is too small to be reliable, the fitting—and the comparison to known objects—suggests an upper limit of r_h ∼ 10–15 pc. This is consistent with a GC, but rules out a larger, more distant galaxy infalling to Virgo.

We have obtained photometry of HVGC-1 using ground-based CFHT images taken for the Next Generation Virgo Cluster Survey (NGVS; Ferrarese et al. 2012) and the associated NGVS-IR K-band survey, but separately processed by CADC (Gwyn 2008). Foreground extinction corrections were taken from Schlafly & Finkbeiner (2011).

Muñoz et al. (2014) have found that in the optical/IR the best separation between single stars and composite old stellar populations is with uiK photometry. Figure 2 shows a two-color i − K versus u − i diagram using CADC photometry of objects spectroscopically classified as stars or GCs and with photometric uncertainties <0.05 mag. Here we consider an object a foreground star if its Hectospec velocity is <250 km s⁻¹; objects with velocities between 500 and 2300 km s⁻¹ are designated as GCs. As found in Muñoz et al. (2014), the GC and star sequences cleanly separate, with only a few outliers. These may be objects with problematic photometry.

Zoom In Zoom Out Reset image size

With u − i = 1.67 ± 0.01 and i − K = −0.01 ± 0.03, HVGC-1 is firmly situated in the GC sequence, strongly suggesting that it is a GC rather than a single star. It has g₀ = 20.57, consistent with a massive GC if at the distance of M87 (see below).

The Hectospec spectrum of HVGC-1 shows strong lines of Ca ii H&K, moderate strength Balmer lines, and a weak G-band (Figure 3, right). There are no emission lines, other than night-sky line residuals. Overall the spectrum looks like an early G star or an intermediate-metallicity GC (Caldwell et al. 2011), not at all like the B and A high-velocity stars of Brown et al. (2012).

Zoom In Zoom Out Reset image size

Rose (1985) discussed the use of line index ratios to search for young stars in composite stellar populations. These ratios can also be used to distinguish single stars from simple stellar populations. In Figure 3 (left), we plot the ratio of central intensities of Ca ii H+H to Ca ii K (the index "Ca ii") against the ratio of Hγ to the G-band at 4300 Å, though this latter choice is not essential.

In Figure 3 (as in Figure 2), we use the M87 candidates classified by velocity into star or GC categories. There is a clear separation between the line-index ratio sequences. Adding confirmed M31 GCs (Caldwell et al. 2011) and Galactic halo stars observed with Hectospec strengthens this conclusion. Foreground stars are mostly in the upper part of the diagram, overlapping the more metal-rich M31 and Virgo GCs (these objects have few hot stars, so the Ca ii index offers no discrimination). For HVGC-1, the Ca ii ratio is around 0.5, and though the uncertainties are large, this ratio is lower than observed for any known foreground star in our sample. Rather, the index ratios are within the range for confirmed GCs in M87 and M31.

The combined photometric and spectroscopic data provide strong evidence that HVGC-1 is a GC.

Given this interpretation, we can calculate basic properties of the cluster. Using the calibration of Lick Fe indices from GC integrated light spectra with [Fe/H] provided in Caldwell et al. (2011), we derive a spectroscopic [Fe/H] estimate of [Fe/H] = −0.9 ± 0.3. This value is consistent with the intermediate-metallicity appearance of the spectrum (moderate Balmer/metal lines), though the photometry in Figure 2 suggests a somewhat lower metallicity. From the photometry we estimate V₀ = 20.33, which would imply M_V = −10.76 and a mass of ∼3.4 × 10⁶ M_☉ for a distance of 16.5 Mpc and assuming M/L_V = 2 (Strader et al. 2011a). However, it is possible the cluster is somewhat closer than this distance (see Section 5.5), which would make it less massive. The expected σ for a typical GC size (2.5 pc) would be only σ = 24 km s⁻¹; we constrain the Hectospec value to be ≲ 80 km s⁻¹.

All galaxy clusters have an important stellar component in the form of "intracluster light": stars stripped from the outer parts of galaxies during the mergers or encounters that occur during the assembly of cluster-central galaxies like M87 (e.g., Purcell et al. 2007). Given their extended spatial distribution, GCs are expected to be stripped along with field stars, and have been found in intergalactic space in clusters including Virgo (Lee et al. 2010). Here we consider whether a GC with a relative velocity of −2300 km s⁻¹ is consistent with the tail of a virialized distribution of objects formed during the assembly of Virgo.

A simple argument that HVGC-1 is unlikely to be in the tail of a distribution is that there are no objects at less extreme outlying velocities: the most extreme positive velocity is <2800 km s⁻¹, less than 1500 km s⁻¹ offset from the M87 systemic, and there are no other non-stellar objects with velocities < − 300 km s⁻¹, 1600 km s⁻¹ from systemic. Such objects should be much more common than GCs like HVGC-1 if one is observing the tail of a virialized distribution.

As another approach, we use a simulation of the formation of a Virgo-like cluster (M_vir ∼ 10¹⁴ M_☉), which includes a central "brightest cluster galaxy" (Martizzi et al. 2012). In Figure 4, we plot the model one-dimensional velocity distribution for both stars and dark matter projected within a 100 kpc radius cylinder around the cluster center. The stellar and dark matter distributions are close to Gaussian, with some lumps in the stellar distribution due to individual galaxies. Both the stars and dark matter have sharp cutoffs well short of a >2300 km s⁻¹ relative velocity, with literally zero particles in the simulation at such an extreme velocity. This comparison should not be over-interpreted, as the simulation is not a perfect match to Virgo, but it does suggest that HVGC-1 is unlikely to be part of a virialized distribution of intracluster GCs.

Zoom In Zoom Out Reset image size

While HVGC-1 is probably not a normal intracluster GC, it is possible that it was recently given a "kick" through a three-body interaction with M87 and a subhalo. Velocity outliers are sometimes observed in simulations of galaxy formation due to such interactions. For example, Sales et al. (2007) suggested that the extreme radial velocity of the Galactic satellite Leo I could be explained if it were ejected as the lighter member of a bound pair of satellites on its first approach to the Galaxy. These simulations find that this process can generate velocities up to ∼3 times the virial velocity (see also Ludlow et al. 2009).

The virial velocity of the group-scale halo surrounding M87 is only ∼600 km s⁻¹. However, that of Virgo is probably in the range ∼900–1300 km s⁻¹ (Strader et al. 2011b), consistent with producing an object like HVGC-1 on the first pericenter passage of an infalling subhalo. However, these extreme velocities are only expected to be observable for a short time after the impulse, so this scenario predicts that HVGC-1 must still be relatively close to the center of M87. The subhalo itself would be observable as a galaxy in the close vicinity of M87.

Merritt et al. (2009) and O'Leary & Loeb (2009) predicted the existence of "hypercompact" stellar systems in galactic halos, comprised of an SMBH and a population of bound stars. These are the result of asymmetric kicks due to gravitational wave emission during the close interactions of binary BHs. However, the predicted kick velocities are generally lower than observed for HVGC-1, and the internal velocity dispersions are expected to be a significant (>0.2) fraction of the kick velocities. Such a large dispersion (>400–1000 km s⁻¹) far exceeds the upper limit ≲ 80 km s⁻¹(Section 3.2). Further, HVGC-1 is metal-poor, while hypercompact stellar systems—originating in galaxy centers—should be metal-rich. A related scenario posits the ejection of a hypercompact stellar system in three-body SMBH interactions during multiple galaxy mergers (Kulkarni & Loeb 2012), but has a similar observational prediction (metal-rich cluster with high σ), inconsistent with HVGC-1.

When a galaxy with an SMBH is accreted by a central cluster galaxy like M87, the BH will sink to the center via dynamical friction and form a binary SMBH. Stars that pass close enough to this binary can be ejected as hypervelocity stars (Yu & Tremaine 2003; Holley-Bockelmann et al. 2005). Assuming a 1:10 mass ratio, a primary mass of 6.6 × 10⁹ M_☉ (Gebhardt et al. 2011), and using the Yu (2002) formula, we find that a binary with a separation of ≲ 1.7 pc can eject an object with a velocity >2300 km s⁻¹. The allowed separation scales with the mass ratio, so separations up to ∼4.5 pc are feasible at fixed total mass.

The same process can apply to GCs, with the important caveat that tidal effects are important. The tidal radius of a 2 × 10⁶ M_☉ GC passing within 1 pc of the M87 SMBH is less than 0.1 pc, so nearly all of the stars would be stripped except for the dense central core. If instead the GC were initially more massive (≳ 10⁷ M_☉), and the binary had a 1:3 mass ratio, then the tidal radius would be 0.3–0.4 pc for a distance of 2–3 pc to the BH. If the cluster were relatively dense, most of its stars would still be stripped, but the core, perhaps with >10⁶ M_☉, could survive and be ejected. Numerical simulations to investigate this possibility are desirable.

We note that Batcheldor et al. (2010) argue that M87's SMBH is displaced from the stellar center of the galaxy, either because the BH is currently a ∼ 1:10 mass ratio binary or because it is undergoing a damped oscillation after a kick from a BH merger. This can be taken as extremely speculative evidence for the recent existence of a binary BH at the center of M87. More concretely, GC kinematics provide evidence for a recent minor merger (Strader et al. 2011b; Romanowsky et al. 2012).

A slight variation on this scenario would be a three-body encounter between a single BH and a binary GC, directly analogous to the formation of Galactic hypervelocity stars. Young binary star clusters are known (e.g., Mucciarelli et al. 2012), though these have short coalescence times and no old binary clusters have been discovered.

While the tangential motion of HVGC-1 is unknown, its radial velocity is so extreme that is it reasonable to assume its tangential motion is smaller than its radial motion. Thus HVGC-1 is likely to be much further from the center of M87 than the projected distance of ∼85 kpc. If we assume that it originated at the center of M87, then we can calculate its inferred total velocity and compare it to the implied escape velocity for a given halo model.

Under these assumptions, we find that the total velocity of HVGC-1 is easily above that of the escape velocity of Virgo for most published halo models (e.g., Karachentsev et al. 2014; Rines & Diaferio 2006), which have virial masses of 4–8 × 10¹⁴ M_☉.⁷ HVGC-1 is below escape velocity only in the unlikely case that (1) its three-dimensional distance is close to its projected distance; (2) the impulse was along our line of sight; and (3) Virgo's mass is at the upper end of the allowed range.

Therefore we can conclude that HVGC-1 either will or has escaped from the Virgo Cluster following a three-body interaction—making it the first known hypervelocity GC.

The nature of this interaction remains unclear. A distance to HVGC-1 would help constrain its origin; such a measurement may be possible using deep Hubble Space Telescope imaging. At its current motion of >2.4 Mpc Gyr⁻¹ it may have already left the Virgo Cluster and be sailing out into intercluster space.