The IACOB project

Recent spectroscopic and high-angular resolution surveys of O-type stars have provided fresh insights on the multiplicity properties of main sequence massive stars (Sana, 2017; Barbá et al., 2017; Vanbeveren & Mennekens, 2017, and references therein). On one hand, the observed fraction of spectroscopic binary/multiple systems with at least one O star component reported by several medium and large size spectroscopic surveys ranges between 42% and 68% (see, e.g., Barbá et al., 2010; Sana et al., 2012, 2013; Chini et al., 2012; Sota et al., 2014; Kobulnicky et al., 2014; Aldoretta et al., 2015; Almeida et al., 2017). On the other hand, as indicated by Sana et al. (2014), when information from high-angular resolution surveys is also considered, the observed multiplicity fraction of Galactic O-type stars with close-by companions can reach 90%.

These observational results have certainly reinforced the long-standing assertion that binary interaction should not be overlooked when investigating massive star evolution (e.g., Vanbeveren, 1993). Indeed, based on empirical arguments, Sana et al. (2012) proposed that 25% of massive binary systems will interact while both components are still on the main sequence, and nearly 70

In parallel, the second decade of the 21st century witnessed the first direct observation of gravitational waves originating from a pair of merging black holes (Abbott et al., 2016). Since then, several tens of detections have been announced, including several cases associated with the merging of two neutron stars (Abbott et al., 2017) and a potential event produced by a neutron star-black hole binary merger (Ackley et al., 2020).

Theoreticians and modelers are working diligently to investigate the impact that binary interaction may have on our current understanding of massive star evolution (e.g., Song et al., 2013, 2016, 2018; de Mink et al., 2013; de Mink & Mandel, 2016; Eldridge et al., 2017; Wang et al., 2020). They are also studying the various possible pathways toward the merging of massive black holes and neutron stars (e.g., Marchant et al., 2016, 2017; Eldridge & Stanway, 2016; Eldridge et al., 2019; Mandel & de Mink, 2016; de Mink & Mandel, 2016; Stevenson et al., 2017).

Slowly but surely, we are also increasing the amount of available information about the orbital and physical properties of O-type binaries (e.g., Almeida et al., 2017; Barbá et al., 2017; Sana, 2017; Martins et al., 2017; Maíz Apellániz et al., 2019; Trigueros Páez et al., 2021; Mahy et al., 2022; Shenar et al., 2022). This is allowing us to construct empirical distributions of binary periods, mass ratios, and eccentricities using statistically significant samples of binary and multiple systems in different environments. Additionally, we can identify hidden companions spectroscopically, which could be associated with the donor of a post-interaction binary system after becoming a stripped star, a black hole, or a neutron star.

The time is ripe to conduct similar observational studies in other phases of the evolution of massive stars beyond the main sequence, such as the so-called blue and red supergiants. This step is crucial and necessary if we aim to empirically constrain any theoretical attempt to connect the early evolutionary phases of massive binaries (i.e., the O and B main sequence stars) with their potential end products, represented by black hole and neutron star binaries that lead to gravitational wave emission if they merge at some point in their lifetimes.

In this paper, we primarily focus on the more direct descendants, from an evolutionary standpoint, of main sequence O-type stars: the B-type supergiants. Despite their importance, the multiplicity properties of this domain of massive star evolution have been only vaguely explored to date (e.g., Dunstall et al., 2015; McEvoy et al., 2015). We present results from the multi-epoch spectroscopic analysis of a sample of 61 evolved O- and B-type stars (mostly supergiants, but also a few giants and bright giants) and 13 O-type subgiants/dwarfs for which we have more than 4500 high-resolution spectra gathered over a period of about 12 years. As we will illustrate throughout this paper, the ubiquitous existence of pulsational-type phenomena in this region of the Hertzsprung-Russell diagram makes the correct identification of spectroscopic binaries in O- and B-type supergiants (OB-Sgs) more challenging than in O giants (Gs) and dwarfs (DWs). In this case, a more complex observational strategy is required to avoid spurious conclusions about the percentage of binaries resulting from the misidentification of pulsating stars as single-line spectroscopic binary (SB1) systems.

The paper is structured as follows. In Sect. 2, we define the working sample and present the observational material. The tools and methods used to extract information about the stellar parameters and perform the radial velocity (RV) measurements are presented in Sect. 3. Results of our investigation of the percentage of spectroscopic binaries in the B-Sg domain are presented in Sect. 4 and discussed in Sect. 5. The main conclusions and some guidelines for future investigations are summarized in Sect. 6.

The bulk of the observations considered in this paper were gathered within the framework of the IACOB project (Simón-Díaz et al., 2011b, 2015, 2020). This includes multi-epoch, high-resolution spectra obtained with the FIES instrument (Telting et al., 2014) attached to the 2.56-meter Nordic Optical Telescope (NOT), and the HERMES spectrograph (Raskin et al., 2011) attached to the 1.2-meter Mercator telescope during several observing runs allocated between 2008 and 2018. As we have indicated in other papers of the IACOB series, although HERMES provides a somewhat larger resolving power ($R$ = 85 000) than the mid- and high-resolution mode of FIES observations ($R$ = 46 000 and 67 000, respectively), the combined use of both instruments has allowed us to build an almost homogeneous spectroscopic data set regarding the quality of the spectra, long-term stability of the wavelength calibration, and accuracy of the zero point in radial velocity. Generally speaking, these observations allow us to achieve accuracies in RV measurements ranging from several hundred m s^-1 to a few km s^-1, mainly depending on the broadening of the diagnostic lines.

By the time the study presented in this paper was performed (late in 2020), the IACOB spectroscopic database¹¹1http://research.iac.es/proyecto/iacob/iacobcat/ comprised more than 8500 spectra of about 1000 Northern Galactic O- and B-type stars (see a more detailed description of the IACOB database and the main drivers for the compilation of such a huge spectroscopic dataset in Simón-Díaz et al., 2015, 2020). In this study, we mainly focus on the sample of O-type stars, B supergiants (B-Sgs), and early B giants (early B-Gs) for which we had more than four epochs available. However, we also make use of the complete dataset to identify double-line spectroscopic binaries (SB2), which are detectable in some cases even with access to only one or two epochs (see Sect. 4).

Table LABEL:table:B1 lists our working sample of stars. This sample comprises 13 O-type dwarfs and subgiants (LCs V and IV), 10 mid/late O-type giants and supergiants (LCs III, II and I), 46 B-Sgs (LCs I and II) and 5 early B-Gs (LC III).

Figure 1 depicts the location of these 74 stars in a spectroscopic Hertzsprung-Russell diagram (sHRD, Langer & Kudritzki, 2014). To serve as a reference, we also present in the background a grayscale representation of the probability density distribution of the location of 575 Galactic massive stars in this diagram, following the method described by Castro et al. (2014). As expected, our sample of stars mainly covers the hot evolutionary stages of stars with masses in the range of approximately 15–80 M_⊙, following the single-star non-rotating evolutionary tracks from Ekström et al. (2012).

Figure 2 shows two histograms indicating the number of stars with a certain number of epochs (top) and the characteristic time-span of the compiled observations (bottom). As seen in the top panel of Fig. 2, we have access to 5 – 25 spectra for approximately $\sim$60% of the stars. This is the usual number of epochs considered in most previous works in the literature hunting for spectroscopic binaries in the OB star domain (e.g., Chini et al., 2012; Sana et al., 2013; Dunstall et al., 2015). However, there is also a non-negligible number of stars with a significantly larger number of epochs available, with some reaching more than one hundred.

Figure 10 (complemented with the bottom panel of Fig. 2) provides a global overview of the time coverage of the observations. As a natural consequence of the way the IACOB spectroscopic database has been built²²2By the time of writing this paper, the database included observations gathered during more than 120 nights distributed between 2008 and 2020. Some of them are associated with observing runs devoted to the investigation of spectroscopic signatures of stellar oscillations in O stars and B-Sgs (e.g., Simón-Díaz et al., 2010, 2017, 2018; Aerts et al., 2017, 2018)., our observations are far from being homogeneous both in terms of total time-span and duty-cycle. However, this must not be considered as a caveat for the study presented here, especially considering that for all the stars we have a minimum time span of approximately 2 months and a minimum of 5 epochs.

As described elsewhere (see, e.g., Simón-Díaz & Herrero, 2014; Holgado et al., 2018), all spectra were normalized and corrected for Earth’s motion in a homogeneous manner using our own routines implemented in IDL and the information provided in the headers of the FITS files. In all cases, we used the interstellar Na i lines at 5890 and 5895 Å as a sanity check of the stability of the zero point in radial velocity for all the spectra of a given target (which we found to be better than 1 km s^-1).

In addition to the main HERMES and FIES spectroscopic datasets described above, in this paper, we also benefit from additional time-series spectroscopy (with a resolution of approximately $R\approx 80,000$) obtained with the Hertzsprung-SONG 1m telescope (Grundahl et al., 2007, 2014; Fredslund Andersen et al., 2019). These observations have been gathered in the framework of a spin-off project of IACOB, in which we are using this robotic telescope to obtain long-term and high-cadence observations of a sample of bright northern O stars and B-Sgs. A specific description of this higher cadence spectroscopic dataset, as well as its scientific exploitation, will be presented in subsequent papers (see also Simón-Díaz et al., 2017). In the meanwhile, for the sake of the discussion presented in Sect. 5.1, we have used some of the gathered time-series. In particular, they allow us to provide some illustrative examples of the type of line-profile variability detected in several presumably single pulsating OB giants and supergiants.

Following the arguments presented in Sect. 3.2, we use the peak-to-peak amplitude of the radial velocity distribution (RV_pp) resulting from the Gaussian fit of a specific diagnostic line to quantify the detected variability of our program stars.

Table LABEL:table:B1 summarizes the total number of considered spectra, as well as the final values of RV_pp for each star in the sample. The table is organized by bins of increasing RV_pp and, within each bin, by spectral type, ranging from early O-dwarfs to late B-Sgs. Column 4 indicates the diagnostic line used for the line-broadening (Sect.3.1) and RV (Sect.3.2) analysis. Uncertainties associated with the measurement of RVs are approximately 1 – 3 km s^-1 (depending on the line-broadening), except for a few stars with large projected rotational velocities ($v$ sin $i$ $\geq$ 200 km s^-1), where the uncertainties may reach up to 5 – 10 km s^-1.

The top left panel in Fig. 6 provides a global overview of the results of our study in the spectroscopic HR diagram. The size of the symbol associated with each star is proportional to the measured RV_pp (except for those targets where RV_pp is larger than 30 km s^-1, which are highlighted with a star symbol). Further notes about the various symbols and colors used in this figure can be found in the corresponding caption.

This panel is complemented by two additional panels presenting the distribution of detected RV_pp as a function of T_eff (bottom panel) and log $\mathcal{L}$ (right panel). As described in Sect. 5, the utility of these two latter panels is twofold. On the one hand, they illustrate the difficulties imposed by the ubiquitous presence of spectroscopic variability due to stellar oscillations in the O and B supergiant domain, complicating the clear identification of single line spectroscopic binaries based solely on the RV_pp quantity. On the other hand, they provide the first statistically significant empirical map of the characteristic amplitude of radial velocity variability associated with pulsational phenomena in the entire O star and B supergiant domain.

We note that, although we have included a few O-type dwarfs and subgiants (13) in our study for comparison purposes regarding the detected variability associated with stellar oscillations, our primary focus henceforth will be on the results obtained for the more evolved OB-Sgs (56) and early B-Gs (5). For a detailed summary of results from a more in-depth study of spectroscopic binarity among Galactic O-type stars, we refer the reader to Barbá et al. (2017) and references therein.

In this regard, Fig. 7 depicts the percentage distribution of this specific sample of evolved high mass stars as a function of the measured value of RV_pp. In this case, we assume a bin size of 0.5 km s^-1 and cut the distribution at $\sim$30 km s^-1, leaving out five early B-Sgs (of luminosity class II) and one mid B-Sg with peak-to-peak amplitudes of the radial velocity curves ranging from $\sim$40 to $210$ km s^-1 (see top left panel in Fig. 6 and Table LABEL:table:B1).

Lastly, for completeness, we summarize in Table 1 the percentage of SB2 systems detected by visual inspection of all spectra available in the IACOB spectroscopic database at the time of performing this study (see Sect. 2). We focus particularly on three groups of stars of interest for this study: first, the late O- and early/mid/late B-type supergiants (OB-Sgs), which are the main stellar group discussed hereafter. For comparative purposes, we also consider the O-type dwarfs and giants in the sample as one group⁵⁵5This information has been extracted from Table E.2 in Holgado et al. (2020)., along with the early B-type giants as the third group.

In summary, the percentage of SB2 systems is much smaller among the OB-Sg sample compared to their less evolved counterparts, the O-type dwarfs/giants, and the less luminous early B-Gs. Notably, no SB2 systems with a mid- or late-B Sg component were found. This result will be further discussed, along with the percentage of detected SB1 systems in the OB-Sg domain in Sect. 5.2.

In light of the stability of the FIES and HERMES spectrographs, which is better than approximately 1 km s^-1, one could reasonably consider tagging all stars in the sample with a measured RV_pp $\gtrsim$ 2 – 3 km s^-1 as single line spectroscopic binaries. Under this assumption, the percentage of detected SB1 stars in our sample of 61 OB-Sgs and late-O/early-B giants would be approximately 90 %.

However, the situation is more complex, as one must consider several other sources of spectroscopic variability present in the blue supergiant domain, including pulsations and/or wind-variability phenomena. See Simón-Díaz et al. (2010, 2018); Aerts et al. (2017, 2018); Burssens et al. (2020) for further discussion and references. For example, this is illustrated in Fig. 7 (see also Dunstall et al., 2015), where we demonstrate that the final percentage of stars among our sample of OB-Sgs which can be definitively identified as SB1 critically depends on the chosen threshold in RV_pp used to minimize the number of false positives arising from intrinsic sources of variability. For instance, assuming a threshold in RV_pp for clear spectroscopic binary detection of 5 km s^-1 would result in a percentage of around 65%; however, this percentage drops to approximately 15% if the threshold is set at 20 km s^-1. Therefore, before drawing any firm conclusions, we need to better understand the impact of intrinsic stellar variability in the OB-Sg domain.

Thanks to the intensive observational efforts of several groups over the last two decades, we are gradually gaining a comprehensive empirical understanding of spectroscopic binarity in the early evolutionary stages of stars with masses exceeding approximately 15 to 20 solar masses. This includes both Main Sequence O-type stars and blue supergiants.

Our study reveals a significant finding: there is mounting evidence indicating a substantial decrease, by a factor $\sim$4 – 5, in the incidence of detected spectroscopic binaries when transitioning from O-type stars to the domain of blue supergiants (B-Sgs). This trend is evident in both our Galactic B-Sg sample and the LMC study by Dunstall et al. (2015). Notably, there is a notable scarcity of spectroscopic B-Sg binaries with mid and late spectral types. Furthermore, the occurrence of SB2 systems among B-Sgs is minimal, comprising less than 5% of the total, in contrast to the higher incidence observed among O-type dwarfs, which have been reported to reach levels as high as 40% (Barbá et al., 2017).

In qualitative terms, a significant portion of the identified double-line spectroscopic binaries, particularly those with at least one O-type star component, are anticipated to transition to single-line (SB1) systems as the more massive component evolves. This transformation can occur under various scenarios: (a) the initially more massive O-type star evolves into a more luminous B-Sg without substantial interaction with its lower-mass companion, (b) it sheds enough mass through mass transfer to become a fainter, stripped lower-mass star, or (c) it survives a supernova explosion as a neutron star or black hole, with the binary system remaining intact. In each of these cases, the contrast in luminosity between the two binary components increases, making the detection of clear SB2 systems more challenging and providing a natural explanation for the scarcity of SB2 systems in the B-Sg domain.

Given the argument presented above, one might naively anticipate a rise in the proportion of detected SB1 systems among B-Sgs compared to their less evolved O-star counterparts. However, this assertion appears to contradict empirical observations. Yet, we must consider that the situation is more intricate from both an evolutionary and observational standpoint.

On one hand, the majority of close massive binaries originate from either case A or case B mass transfer and typically manifest as binaries with a B-type supergiant component only intermittently in later stages of evolution (see, e.g., Wang et al., 2020). Moreover, a considerable fraction of these binaries may merge during main sequence evolution, estimated to be around 10% (de Mink et al., 2014). Additionally, some B-Sgs have been hypothesized to be in a post-red supergiant phase of evolution, either evolving as single stars (e.g., Ekström et al., 2012) or resulting from a merger between a red supergiant and a lower mass main sequence companion (e.g., Menon et al., 2019, 2024).

On the other hand, wider binaries may evolve to a scenario where the O-type component transitions into a mid/late B-Sg without filling its Roche lobe. These systems typically have long orbital periods and consequently exhibit relatively low-amplitude RV curves. As a result, intrinsic variability due to stellar oscillations, which, as discussed in Sect. 5.1, can induce RV variations of up to 20 – 25 km s^-1, can hinder the clear identification of orbital motion in these systems.

While more work is still needed to reach firm conclusions, it becomes evident that investigating the incidence of binarity in the B-Sg domain can provide valuable empirical insights into several open questions in the field of massive stars. These include the percentage of merger events occurring during the main sequence evolution of O-type stars (de Mink et al., 2014), the exact location of the Terminal Age Main Sequence (McEvoy et al., 2015), the potential post-red supergiant evolutionary nature of B-Sgs (de Mink et al., 2014), and ultimately, the impact of binaries on massive star evolution (Langer, 2012; Wang et al., 2020), including intermediate phases in which an OB star has a black hole and/or neutron star companion (Langer et al., 2020).

In this regard, we want to emphasize once more the need for caution when combining and interpreting the information provided by various available studies. This caution is necessary not only due to potential observational biases affecting the samples being considered but also because of the impact that intrinsic variability has on the detection of actual spectroscopic binaries, especially among OB-Sgs.

Indeed, the second important message we want to highlight from our study is that defining the exact percentage of spectroscopic binaries in the OB-Sg domain requires a better understanding and characterization of the amplitude of spectroscopic variability produced by the various sources of intrinsic variability predicted to occur in these types of stars.

Crucially, we remark that depending on the origin of this intrinsic variability, there may be significant dependencies on metallicity. Hence, any empirical investigation performed in the Milky Way (like the one presented in this paper) might not necessarily be directly extrapolated to other metallicity environments. For example, at present, there is an intense debate about the main physical driver of the detected stellar variability (both photometric and spectroscopic) and the so-called macroturbulent broadening in the entire massive OB star domain (see, e.g., Simón-Díaz et al., 2017; Grassitelli et al., 2015a; Bowman et al., 2020, and references therein). Depending on the relative impact that the convectively driven internal gravity waves predicted to originate in the stellar core (Aerts & Rogers, 2015) and/or the subsurface convective layers (Cantiello et al., 2009; Grassitelli et al., 2015a, b; Cantiello & Braithwaite, 2019; Lecoanet et al., 2019) can have on the stochastic low-frequency variability, which is increasingly detected in blue massive OB stars (Blomme et al., 2011; Buysschaert et al., 2015; Aerts et al., 2017, 2018; Simón-Díaz et al., 2018; Bowman et al., 2019, 2020), stellar metallicity may play an important role.

This is, therefore, a crucial research avenue that will undoubtedly shape our future understanding of the prevalence of binary systems throughout the evolution of massive stars. In this pursuit, it is imperative to consider other anticipated sources of stellar variability, such as coherent heat-driven modes (Godart et al., 2017), and intrinsic variations in the stellar wind. As exemplified by Burssens et al. (2020), the TESS mission presents a unique opportunity for advancements in this field.