Discovery of a Double Blue Straggler Sequence in M15: New Insight into the Core-collapse Process

In the color–magnitude diagram (CMD) of any stellar aggregates blue straggler stars (BSSs) typically appear as a sparse and scattered population of stars lying in a region at brighter magnitudes and bluer colors with respect to the cluster main-sequence turnoff (MS-TO) stars (Sandage 1953; Ferraro et al. 1997, 2003, 2012, 2018; Piotto et al. 2004; Lanzoni et al. 2007; Leigh et al. 2007; Simunovic & Puzia 2016).

Their location in the CMD, combined with further observational evidence (Shara et al. 1997; Gilliland et al. 1998; Ferraro et al. 2006; Fiorentino et al. 2014), suggests that they are hydrogen burning stars more massive than the parent cluster stars at the MS-TO. These exotica have been suggested to be powerful tools to investigate the internal dynamical evolution of the parent cluster (Ferraro et al. 2012, 2018; Alessandrini et al. 2016; Lanzoni et al. 2016). BSSs are generated through two main formation channels: (1) mass-transfer (MT) phenomena and/or coalescence in primordial binary systems (McCrea 1964), and (2) direct stellar collisions (Hills & Day 1976). While the first process (which is driven by the secular evolution of binary systems) is common to any stellar environment (low- and high-density stellar systems, as globular and open clusters, dwarf galaxies, the Galactic field), the second one requires a high-density environment, where the rate of stellar collisions is increased. Note that a collisional environment can also affect the efficiency of the MT formation channel, since dynamical interactions involve single and binary stars thus favoring the tightening of binaries. The congested central regions of high-density globular clusters (GCs) are the ideal habitat where collisions can occur. Hence in such an environment both the BSS channels can be active (Ferraro et al. 1993, 1995), although the MT formation channel seems to be the most active one (see Davies et al. 2004; Knigge et al. 2009).

Although a few intriguing spectroscopic features have been interpreted as the fingerprints of the MT formation channels (see Ferraro et al. 2006), BSSs produced by different formation processes still appear photometrically indistinguishable. In this respect, a promising route of investigation was opened by the discovery of two distinct sequences of BSSs in the post core-collapse (CC) cluster M30 by Ferraro et al. (2009). After this discovery, a similar feature has been detected in NGC 362 (Dalessandro et al. 2013) and NGC 1261 (Simunovic et al. 2014). In only one case has a double BSS sequence been detected in a young cluster in the Large Magellanic Cloud (Li et al. 2018a, 2018b). However, it has been argued (Dalessandro et al. 2019a, 2019b) that the observed bifurcation could be an artifact due to field star contamination.

The main characteristic ubiquitously detected in these clusters is the presence of a narrow BSS sequence on the blue side of the CMD, separated through a clear-cut gap from a more scattered population of BSSs on the red side. The narrowness of the blue sequence demonstrates that it is populated by a nearly coeval population of stars with different masses, generated over a relatively short period of time. Moreover, the perfect agreement of the blue sequence locus with collisional isochrones (Sills et al. 2009) suggests that these objects have been formed through collisions. Instead, the red sequence is by far too red to be reproduced by collisional models of any age. All these facts support the hypothesis that the origin of the blue BSS sequence is related to a phenomenon able to enhance the probability of collisions over a relatively short period of time, thus promoting the formation of collisional BSSs. Ferraro et al. (2009) proposed that this phenomenon is the cluster CC. This hypothesis has been recently confirmed by numerical simulations (Portegies Zwart 2019).

In this context, Ferraro et al. (2009) also suggested that the properties of the blue-BSS sequence (essentially its extension in luminosity) can be used to date the epoch of CC. In short, while the MS-TO luminosity is a sensitive indicator of the age of the cluster, the luminosity of TO point of the blue BSS sequence can be used to determine the epoch at which the collisional event has promoted the formation of this stellar population. In fact the comparison with collisional models allowed the dating of the epoch of the CC event: approximately 2 Gyr ago in M30 (Ferraro et al. 2009) and NGC 1261 (Simunovic et al. 2014), and 0.2 Gyr ago in NGC 362 (Dalessandro et al. 2013).

Recent Monte Carlo simulations of synthetic MT-BSSs (Jiang et al. 2017) showed that the blue-BSS sequence can also be populated by MT-BSSs. This finding by itself does not invalidate the possibility that the blue BSS sequence might be mainly populated by collisional BSSs, since the easiest way to reproduce the observed narrowness of the blue sequence and the adjacent gap is still the hypothesis that the vast majority of BSSs along this sequence formed almost simultaneously, instead, as a result of a continuous process extending over a long time interval. Conversely, the result of Jiang et al. (2017) might explain the presence of the few W Uma stars (which are close binaries) actually detected by Ferraro et al. (2009) and Dalessandro et al. (2013) along the blue BSS sequence in M30 and in NGC 362. In this framework it is important to stress that, although the location in the CMD cannot be used to individually distinguish collisional from MT-BSSs, it mirrors the occurrence of the two formation mechanisms: when a narrow blue sequence is detected, this is mainly populated by collisional BSSs, with some possible contamination from MT-BSSs (as the WUma variables detected along the blue sequences); the red side, instead, cannot be reproduced by any of the available collisional models (Sills et al. 2009) and is mainly populated by MT-BSSs with some contamination from evolved BSSs generated by collisions.

In this paper we present the discovery of a double BSS sequence in the post-CC cluster M15 (NGC 7078). Indeed, M15 is one of the most massive (log M = 6.3M_⊙, several 10⁶M_⊙) and dense ($\mathrm{log}{\rho }_{0}=6.2{M}_{\odot }\,{\mathrm{pc}}^{-3}$) clusters of the Galaxy. As emphasized by Leigh et al. (2013), while binary evolution seems to be the dominant channel for BSS formation, the fact that M15 has one of the densest and most massive cores among Galactic GCs, makes it the ideal environment in which to expect the collisional formation channel to be relevant. This fact has been recently confirmed by Leigh et al. (2019) where the authors foresee the collisions via 2+2 interactions to play a crucial role on the formation of BSSs in post-CC clusters. Since the very first detections of a central cusp in its surface brightness profile (Djorgovski & King 1986; Lauer et al. 1991; Stetson 1994) it has been considered as a sort of "prototype" of post-CC clusters, and it is still cataloged as such in all the most recent GC compilations (see the updated version of the Harris Catalog). The BSS sequence studied here shows a complex structure that might provide crucial information on the CC phenomenon.

In this work we use a set of archival high resolution images obtained with the Advanced Camera for Survey (ACS) on board the Hubble Space Telescope (HST). The core of the cluster was observed with the High Resolution Channel (HRC) mode of the ACS in the F220W and F435W filters (hereafter, U₂₂₀ and B₄₃₅, respectively). The HRC provides a supreme spatial resolution, with a pixel scale of ∼0.028 × 0.025 arcsec² pixel⁻¹ in a field of view of 29'' × 26''. A total of 13 images of 125 s of exposure time each were acquired in the B₄₃₅ filter under the program GO-10401 (PI Chandar), while 8 × 290 s exposures were taken with the U₂₂₀ filter under the program GO-9792 (PI Knigge). These images were already used in Dieball et al. (2007) and Haurberg et al. (2010) to probe the stellar population in the innermost regions of the cluster. As described in Haurberg et al. (2010), the HRC images were combined using a dedicated software developed by Jay Anderson and similar to DrizzlePack.⁹ In short, the software combines single raw images of a given band to improve the sampling of the point-spread function (PSF). A 3'' × 3'' section of a super sampled frame thus obtained in the B₄₃₅ band is shown in Figure 1 (left panel), where it is also compared to a single raw exposure (right panel). As immediately visible in the figure, the software allows us to reduce the pixel size to ∼0.0125 × 0.0125 arcsec² pixel⁻¹ on the combined images, thus doubling the effective spatial resolution in both the U₂₂₀ and the B₄₃₅ filters. This image processing is crucial as it allows us to resolve the stars in the very central region of the cluster, where the stellar crowding is too severe even for the resolving power of the HRC (see also Figure 1 from Haurberg et al. 2010).

Zoom In Zoom Out Reset image size

We first performed a standard PSF fitting photometry on the two super sampled U₂₂₀ and B₄₃₅ frames using DAOPHOTII (Stetson 1987). The catalog was calibrated into the VEGAMAG system using the recipe from Sirianni et al. (2005) and adopting the most recent zero-points available through the ACS Zero-point Calculator.¹⁰ All the post-MS stellar evolutionary sequences typical of a GC are well distinguishable on the first version of the CMD shown in the left panel of Figure 2. Still, we notice a number of objects falling in the region brighter than the Sub-Giant Branch (SGB) and bluer than the Red Giant Branch (RGB).

Zoom In Zoom Out Reset image size

The origin and the reliability of these stars (sometime called "yellow stragglers") is a well known problem and it has been discussed in many papers in the literature. The vast majority of these objects have been demonstrated to be optical blends (see, e.g., Figure 8 in Ferraro et al. 1992 and Figure 7 in Ferraro et al. 1991, where a few examples of deblending are illustrated). Here, we used the package ROMAFOT (Buonanno et al. 1983) to manually inspect the quality of the residuals of the initial PSF fitting process, and to iteratively improve the quality of the PSF fitting for a selection of individual problematic stars. ROMAFOT is a software developed to perform accurate photometry in crowded fields. While the PSF fitting procedure with ROMAFOT requires a higher level of manual interaction by the user with respect to DAOPHOTII, it has the unique advantage of a graphical interface that allows the user to improve the local deconvolution of stellar profiles (see also Monelli et al. 2015). Indeed the results of the analysis with ROMAFOT fully confirm the findings by Ferraro et al. (1991, 1992): most of the objects lying in the yellow stragglers region turned out to be blends. For the sake of illustration, in the left panel of Figure 2 we highlight four spurious sources (open symbols) due to poor PSF fitting. As shown in the right-hand panel, each of these sources is indeed the blend of an MS-TO and an SGB star (solid symbols). These sources are also highlighted with the same symbols on the B₄₃₅ images shown in Figure 3. In Figure 2 we also plot the equal-mass binary sequence (dashed line) and the adopted BSS selection box. It is evident that the conservative BSS selection discussed in Section 3 (considering only BSSs brighter than U = 19.8) prevents any significant contamination from unresolved binaries and blends.

Zoom In Zoom Out Reset image size

The iterative deblending process described above was used to optimize the PSF modeling of all the objects lying outside the well recognizable canonical evolutionary sequences in the CMD, including all the BSSs. In Figure 4 we show the final CMD obtained at the end of the accurate deblending procedure. It should be noted that the HRC data set that we use in this paper allows us to resolve stars with a separation larger than ∼0012, which translates into a separation of ∼125 au at the distance of M15 (10.4 kpc). Hence, even after our deblending procedure, it is still possible that a few unresolved binaries and blends with separation smaller that 125 au are still populating the "yellow straggler region." On the other hand, some of these objects could also be BSSs that are evolving from their main-sequence (core hydrogen burning) stage to a more evolved evolutionary phase (the SGB). Still, given the overall low number of BSSs, and taking into account that the SGB stage at the typical BSS mass (∼1.2 M_⊙) is quite rapid, we expect just a few evolved BSSs in this part of the CMD.

Zoom In Zoom Out Reset image size

The CMD shown in Figure 4 demonstrates the outstanding quality of the data, which allows us to properly sample the hot stellar population in the core of M15 and to detect MS stars well below the MS-TO. Thanks to the use of the U₂₂₀ filter, the hot horizontal-branch (HB) stars describe a very narrow and well defined sequence. This fact per se demonstrates already the excellent photometric quality of the data (see also Dieball et al. 2007). The same applies to the RGB stars whose sequence spans almost 2.5 mag in color on the CMD. As expected, the use of the UV filter clearly suppresses the contribution of the RGB stars to the total cluster's light, making this plane optimal for sampling the hottest stellar populations in the cluster (see also Ferraro et al. 1997, 2003, 2018; Raso et al. 2017).

3.1. A Double Sequence of BSS

A surprising feature clearly visible in the CMD of Figure 4 is the presence of two distinct, parallel and well separated sequences of BSSs. The two sequences are almost vertical in the CMD, in a magnitude range 20 < U₂₂₀ < 18, and located at (U₂₂₀–B₄₃₅) ∼ 0.8 and ∼1.2. A similar feature was previously detected using optical bands in the GCs M30, NGC 362, and NGC 1261 (Ferraro et al. 2009; Dalessandro et al. 2013; Simunovic et al. 2014, respectively). This is the first time that such a feature has been detected in a UV-CMD.

We show in Figure 5 a portion of the CMD zoomed on the BSS region. Hereafter, we will consider as bona fide BSSs those stars populating the region of the CMD in the color ranges explicitly mentioned above and with U ≲ 19.8. As discussed above, such a conservative magnitude threshold guarantees that the BSS sample is negligibly affected by residual blended sources and unresolved binaries (see the right panel in Figure 2). In the left panel we show with a dashed line the fit to the bluest sequence of BSSs. We take this line as a reference and we then calculate the distance in (U₂₂₀–B₄₃₅) color of the BSSs observed at U₂₂₀ < 19.8 (black dots). In the rightmost panel we show the histogram of the distribution of the color's distances. This is clearly not unimodal: at least two peaks, with a clear gap in between, are well distinguishable. To assess the statistical significance of such bimodality, we used the Gaussian mixture modeling (GMM) algorithm presented by Muratov & Gnedin (2010), which works under the assumption that the multimodes of a distribution are well modeled via Gaussian functions. We found that the separation of the means of the two peaks relative to their widths is D = 4.71 ± 0.78. The parametric bootstrap probability of drawing such a value of D randomly from a unimodal distribution is 3.6%. The probability of drawing the measured kurtosis is 0.4%. As such, all three statistics clearly indicate that the observed BSS color distribution in the UV-CMD is bimodal. The GMM code also provides the user with the probability distribution of each element to belong to a given peak. We find that 27 and 15 stars have a probability >98% to belong to the bluest and reddest peak, respectively. Their location in the CMD is shown in Figure 6. The two stars shown as black dots in the figure have a probability of 96% (36%) and 69% (31%) to belong to the bluest (reddest) sequence. We also used the Dip test of unimodality (Hartigan & Hartigan 1985) to further investigate the significance of the deviation from unimodality of the BSS color distribution shown if Figure 5. The Dip test has the benefit of being insensitive to the assumption of Gaussianity. We found that the observed BSS color distribution has 98.5% probability to deviate from a unimodal distribution. The tests described above hence indicate that the bimodality of the BSS sequence that we have found in the UV-CMD of M15 is a statistically solid result.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In order to further prove the presence of the observed bimodality in the BSS color distribution, we retrieved a number of high resolution images of the core of M15 obtained with the HST/WFC3 (GO-13297; PI: Piotto). These images, obtained using the F275W and F336W filters were already used in Raso et al. (2017) to investigate the dynamical age of the cluster through the study of the BSS' radial distribution (see also Lanzoni et al. 2016). We stress here that the plate scale of ∼004 offered by the WFC3 did not allow the authors to firmly detect the double BSS sequence in the core of the cluster as reported in this work. Interestingly enough, hints for the presence of a double BSS sequence are visible in the CMD of M15 shown in Figure 22 of Piotto et al. (2015).

We used the photometric catalog obtained by Raso et al. (2017) to resolve the BSS in a region outside the area covered by our HRC data out to a radius r = 80''. Moreover, we used several hundreds of stars in common between the WFC3 and our HRC catalog to accurately register the position of the stars in the HRC catalog to the WFC3 position. We then used ALLFRAME (Stetson 1994) to obtain accurate PSF fitting photometry of the F275W and F336W images of the WFC3 using as input catalog of stellar centroids the coordinates of HRC stars registered on the WFC3 coordinate system.

We show on Figure 7 the CMDs obtained with the two data sets. Clearly the separation of the BSS into a blue and red sequences according to the selection done in the HRC plane, holds also in the CMD obtained with the WFC3.

Zoom In Zoom Out Reset image size

3.2. Variable BSS

3.3. A Collisional Sequence

3.4. An Additional Intriguing Feature

3.5. BSS Radial Distribution

We plot in Figure 8 the cumulative radial distribution of the BSSs belonging to the blue and the red sequences (blue/solid and red/dashed lines, respectively). As explained in Section 3 we used a photometric catalog obtained with the WFC3 to extend the BSS radial distribution out to a radius r = 80''. As expected the BSSs are significantly more segregated than normal MS stars (black dotted line), regardless of which population they belong to. As also found in the GCs M30 and NGC 362, the red sequence appears to be more segregated than the blue one. As discussed in Dalessandro et al. (2013), this difference may offer further support for a difference formation history of the two populations. In short the blue-BSSs, born as a consequence of increase of collisions during the CC, have been also kicked outward during collisional interactions while most of the red BSSs sank into the cluster center because of dynamical friction and did not suffer significant hardening during the CC phase. The Kolgomorov–Smirnov test applied to the cumulative radial distributions suggests that the blue and the red sequences do not belong to the same parental population at an only 1.5σ level of significance. We stress here that this is likely due to the small number statistics of the two populations. The inset in the figure shows the cumulative radial distributions of the BSS along the two branches of the blue sequence. The BSSs populating the youngest collisional sequence (dashed line) appear more centrally segregated than the oldest ones (solid line). Although the two subpopulations include only a few stars each, the statistical significance of such a difference turns out to be of the order of 1.5–2σ.

Zoom In Zoom Out Reset image size

As previously discussed, the presence of a double sequence of BSSs is a feature that was already discovered using optical photometry in the core of the GCs M30, NGC 362, and NGC 1261.

Ferraro et al. (2009) first suggested that the presence of two parallel sequences of BSSs in the CMD can be explained by the coexistence of BSSs formed through two different formation mechanisms: the blue sequence would be originated mainly by collisions, while the red sequence derives from a continuous MT formation process (see also Dalessandro et al. 2013 and Xin et al. 2015). This scenario has been recently confirmed by a set of simulations that follow the evolution of a sample of BSSs generated by stellar collisions (Portegies Zwart 2019). This work concludes that the blue BSS chain detected in M30 can be indeed populated by collisional BSSs originated (over a short timescale) during the CC phase, thus finally proving that CC can be at the origin of a narrow sequence of nearly coeval Blue BSSs. Conversely, the photometric properties of BSSs generated by MT processes in binary systems are currently controversial, since different models seem to predict different MT-BSS distributions in the CMD (see the case of the models computed for M30 by Xin et al. 2015 and Jiang et al. 2017). In addition, even admitting that MT-BSSs may extend to the blue side of the BSS distribution in the CMD, it is unclear if and how MT processes alone could produce two split sequences, distinctly separated by a gap. Since at the moment there are no consolidated and specific models of BSS formation through MT for M15, here we focus on the blue sequence that, in this cluster, shows the most intriguing features. The nice match between the blue-BSS locus in the CMD and the collisional isochrones of Sills et al. (2009) strongly suggests that this population could be produced during a dynamical phase characterized by a high rate of stellar collisions. Moreover, in striking similarity with previous cases, the fact that the blue BSSs in the CMD appear so nicely aligned along the collisional isochrone MS suggest that they were mostly generated over a relatively short timescale (within a few 10⁸ Myr), thus representing an essentially coeval population. The superb photometric quality of the data presented here allows us to make a significant step forward in our understanding of this phenomenon. It is particularly interesting that in M15 we are able to distinguish possible substructures along the collisional sequence. The two branches discovered here, in fact, can be interpreted as two generations of collisional BSSs, possibly formed during two distinct episodes of intense dynamical interactions. Since M15 is a core-collapsed cluster (see, e.g., Djorgovski & King 1986; Chernoff & Djorgovski 1989; Murphy et al. 2011), such episodes can quite possibly be associated with the cluster's structural properties at the time of CC and during the dynamical stages following this event. In particular, after reaching a phase of deep CC, clusters may undergo a series of core oscillations driven by gravothermal effects and binary activity (see, e.g., Bettwieser & Sugimoto 1984; Cohn et al. 1989; Heggie & Ramamani 1989; Takahashi & Inagaki 1991; Makino 1996; McMillan & Engle 1996; Heggie & Giersz 2009; Breen & Heggie 2012), characterized by several distinct phases of high central density followed by more extended stages during which a cluster rebounds toward a structure with lower density and a more extended core. Interestingly, Grabhorn et al. (1992) performed detailed numerical simulations, including post-collapse core oscillations, able to reproduce the contemporary presence of a large core radius (13 pc) and a cusp in the stellar density profile of M15 as revealed by early HST observations from Lauer et al. (1991).

The two branches of blue-BSS sequence might be the outcome of the increased rate of collisions during two separate density peaks in the post-CC evolution. In order to provide a general illustration of the possible history of the cluster's structural evolution leading to the observed blue-BSS branches, we show in Figure 9 the time evolution of the 1% Lagrangian radius (top panel) and of the collisional parameter (bottom panel) as obtained from a Monte Carlo simulation run with the MOCCA code (Giersz et al. 2008). The collisional parameter ${\rm{\Gamma }}={\rho }_{c}^{2}{r}_{c}^{3}/{\sigma }_{c}$ (where ${\rho }_{c},\,{r}_{c},\mathrm{and}\,{\sigma }_{c}$ are, respectively, the cluster's central density, core radius and central velocity dispersion) is often used to provide a general quantitative measure of the rate of stellar collisions in a GC (see, e.g., Davies et al. 2004). The simulation presented here starts from the initial structural properties discussed in Hong et al. (2017), with 7 × 10⁵ stars, no primordial binaries and an initial half-mass–radius equal to 2 pc. We point out that the adopted simulation and initial conditions do not consist of a detailed model of M15, but just illustrate the general evolution of a cluster reaching a phase of deep CC and undergoing core oscillations in the post-CC phase.

Zoom In Zoom Out Reset image size

As shown in Figure 9, the initial deep CC (clearly distinguishable at t/t_CC = 1) leads to the largest increase in the value of Γ, and is followed by several distinct recollapse episodes leading to secondary peaks of different magnitudes in the collisional parameter (all smaller than the initial peak associated with the first collapse). The time interval between the various peaks in Γ is not constant and, for the simulation presented here, typically ranges from approximately a few hundred million years to approximately a billion years. Although we strongly emphasize that models exploring in detail the actual BSS production in each collapse episode, along with a proper characterization of the collision events and the number of stars involved, are necessary (see, e.g., Banerjee 2016), here we suggest that the BSS populations observed along the two branches of the blue-BSS sequence in M15 might be produced during the initial deep collapse and during one of the subsequent collapse events, one sufficiently deep to trigger a significant number of collisions and BSS production.

Note that a similar argument was used by Simunovic et al. (2014) to explain the presence of a few bright BSSs located on the blue side of the blue-BSS sequence in NGC 1261. However, this cluster does not show the typical signature of CC, and for this reason the authors suggest that it is currently experiencing a bounce state.

We used an exceptional set of high resolution ACS/HRC images to explore the stellar population in the collapsed core of the GC M15. A high-precision UV-CMD has revealed the existence of two clear-cut sequences of BSSs. In particular, we discovered that the blue sequence, which should be populated by collisional BSSs in the interpretative scenario proposed by Ferraro et al. (2009), consists of two distinct branches nicely reproduced by collisional isochrones of 2 and 5.5 Gyr. We interpret these features as the observational signature of two major collisional episodes suffered by M15, likely connected to the collapse of its core: the first one (possibly tracing the beginning of the CC process) occurred approximately 5.5 Gyr ago, while the most recent one dates back 2 Gyr ago. This result reinforces the evidence of the strong link existing between the observational properties of the BSS population in a cluster and its dynamical history (Ferraro et al. 2009, 2012, 2018), and it opens a window to gather a deeper understanding of CC and post-CC evolution, as well as the link between these dynamical phases and a cluster's stellar content.

We are grateful to the anonymous referee for precious comments and suggestions. We are grateful to Jay Anderson and Nathalie C. Haurberg for providing us with the processed ACS/HRC images. This research used the facilities of the Canadian Astronomy Data Centre operated by the National Research Council of Canada with the support of the Canadian Space Agency. F.R.F. is grateful to the Indiana University for the hospitality during his stay in Bloomington, when part of this work was performed.