The Electromagnetic Counterpart of the Binary Neutron Star Merger LIGO/Virgo GW170817. VIII. A Comparison to Cosmological Short-duration Gamma-Ray Bursts

In order to enable an effective comparison to the counterpart to GW170817, we utilize data from all previous short GRBs with redshift measurements. We compile redshift measurements from Berger (2014) for bursts discovered over 2004–2013. To update this sample, we also include redshifts for new bursts from GCN circulars (Hartoog et al. 2014; Cucchiara & Levan 2016; Levan et al. 2016), recent papers on individual events (Fong et al. 2016; Troja et al. 2016; Selsing et al. 2017), and additional optical spectroscopy which will be described in an upcoming work (W. Fong et al. 2017, in preparation). Our total sample of short GRBs with redshifts comprises 36 events and spans a redshift range of $\approx 0.12\mbox{--}2.6$, with a median of $\langle z\rangle \approx 0.46$ (Figure 1).

Zoom In Zoom Out Reset image size

We note that, by default, requiring that the events have redshifts only includes bursts that are well localized to ≲few arcsec uncertainty via the detection of an X-ray afterglow. However, we previously explored any potential bias within the Swift short GRB population introduced by requiring an X-ray afterglow based on a smaller sample of events (Fong et al. 2013). We found that the large majority of bursts that did not have detected X-ray afterglows were affected by observing constraints that would generally be decoupled from any burst or host galaxy properties. Thus, we do not expect the requirement of a redshift to introduce any discernible bias in this sample.

In order to compare the X-ray emission of on-axis short GRB afterglows to the X-ray counterpart for GW170817 (Evans et al. 2017; Margutti et al. 2017), we utilize the sample in Fong et al. (2015) which covers short GRBs discovered between 2004 November and 2015 March (references for individual data sets therein). The data were taken primarily with the X-Ray Telescope (XRT; Burrows et al. 2005) on board the Swift satellite (Gehrels et al. 2004), the Chandra X-Ray Observatory, and XMM-Newton. To supplement this data set, we gather all Swift/XRT observations from the light curve repository (Evans et al. 2007, 2009) for short GRBs with detected afterglows between 2015 March and 2017 August. In order to enable a direct comparison to the X-ray counterpart to GW170817, we use the burst redshifts to convert to X-ray luminosity. Our sample of short GRBs with X-ray afterglows comprises 36 events, and the light curves are shown in Figure 2.

Zoom In Zoom Out Reset image size

We compile radio observations for short GRBs in a similar manner to that for the X-ray observations, starting with the sample from Fong et al. (2015). We update this sample to include five additional events with observations from the Karl G. Jansky Very Large Array (VLA) under Programs 15A-235 (PI: Berger) and 17A-218 (PI: Fong). Of the bursts with redshifts, four have previously reported detections (Berger et al. 2005; Soderberg et al. 2006; Fong et al. 2014, 2015), while two additional detections are included in the updated sample. In total, the sample includes six bursts with detections and 19 additional events with upper limits. We convert the data to luminosity via the burst redshifts. The resulting radio light curves and upper limits, along with the results of radio monitoring following GW170817 at 6 and 10 GHz with the VLA (Alexander et al. 2017), are shown in Figure 2.

There have been three previous claims of kilonovae in the literature: an NIR excess following GRB 130603B ($z=0.356;$ Berger et al. 2013; Tanvir et al. 2013), and optical excesses following GRB 050709 ($z=0.16;$ Jin et al. 2016) and GRB 060614¹⁵ ($z=0.125;$ Jin et al. 2015; Yang et al. 2015). We note an additional optical excess was reported following GRB 080503 (Perley et al. 2009); however, since this burst does not have a known redshift, we do not include it in this discussion. To enable a comparison of the luminosity and temporal behavior of the optical/NIR counterpart to GW170817 (Chornock et al. 2017; Cowperthwaite et al. 2017; Nicholl et al. 2017), we collect observations corresponding as close as possible to the rest-frame r- and H-bands for each burst (Berger et al. 2013; Tanvir et al. 2013; Jin et al. 2015, 2016).

We supplement these detections with any optical/NIR observations following short GRBs on timescales of $\gtrsim 1$ day after the burst. To this end, we retrieve Hubble Space Telescope (HST) observations (PI: Tanvir; Program 14237) of the short GRB 160821B from the Mikulski Archive for Space Telescopes archive. We utilize observations taken with the Wide Field Camera 3 (WFC3) in the F160W filter, corresponding to rest-frame H-band at the redshift of the burst ($z=0.16;$ Levan et al. 2016), on 2016 September 14 UT (∼23 days after the burst). We used the astrodrizzle task as part of the Drizzlepac software package (Gonzaga 2012) to create final drizzled image, using final_scale = 0065 pixel⁻¹ and final_pixfrac = 0.8. We use standard tasks in IRAF to perform aperture photometry of faint point sources in the field to calculate a $3\sigma$ limit of ${m}_{160{\rm{W}}}\,\gtrsim$ 26.0 mag.

We add to this sample optical and NIR upper limits following 14 additional events with redshifts. Compared to the sample in Fong et al. (2015), we note the addition of a deep limit following GRB 050509B of $r\gtrsim 25.7$ mag at $\approx 25.9\,\mathrm{hr}$ after the burst (Cenko et al. 2005; Bloom et al. 2006). Although this limit corresponds to rest-frame V-band, we note that the expected V − r color based on observations of the optical counterpart to GW170817 at $\approx 1.5$ days after the event (Cowperthwaite et al. 2017) is negligible, $\lesssim 0.2$ mag. Thus, we still include this point in our sample.

For all of the bursts with detections of or limits on kilonova emission, we use the burst redshifts to convert apparent magnitude to K-corrected absolute magnitudes. The data for the previous short GRBs, as well as Gemini-South H-band observations of the counterpart to GW170817 (Cowperthwaite et al. 2017), are shown in Figure 3. In order to enable a direct comparison to early short GRB optical limits, we also employ the initial DECam observation at i-band at $\approx 0.47$ days, and correct for the observed r − i color of $\approx 0.2$ mag at $\approx 1.4$ days (Cowperthwaite et al. 2017). We note that this is conservative, as the source has a blue color at $\lesssim 1$ day and 0.2 mag is likely an upper limit on the r − i color.

Zoom In Zoom Out Reset image size

In order to obtain a complete sample of short GRB host properties, we collect all of the available values for host redshifts, galaxy type, rest-frame B-band luminosity (L_B), stellar mass (M_*), stellar population age (τ), and star formation rate (SFR) from large samples (Leibler & Berger 2010; Fong et al. 2013; Berger 2014), as well as papers on individual objects (Perley et al. 2012; de Ugarte Postigo et al. 2014; Fong et al. 2016; Troja et al. 2016; Selsing et al. 2017). We supplement this sample with values for the redshifts and stellar population properties of eight additional short GRB hosts discovered over 2014–2017, the details of which will be described in an upcoming work (W. Fong et al. 2017, in preparation). For these bursts, we infer stellar masses from a combination of broadband SED fitting, and K-band luminosity and B − K color relations (Bell & de Jong 2001; Mannucci et al. 2005). We calculate rest-frame B-band luminosities from host photometry. We normalize these values to the characteristic luminosity (L*) of the evolving galaxy luminosity function over $z\sim 0$ to $z\approx 2$, depending on the redshift of each host (Blanton et al. 2003; Willmer et al. 2006; Marchesini et al. 2007). The median of each stellar population property is listed in Table 1. The total sample of 36 events provides a comprehensive comparison sample for NGC 4993.

Table 1.  Comparison of Stellar Population Properties

Property	SGRB SF Median	SGRB Ell. Median	SGRB Total Median	NGC 4993
(1)	(2)	(3)	(4)	(5)
Stellar Mass (log(${M}_{* }/{M}_{\odot }$)	9.65 (0.90)	10.85 (0.33)	10.10 (0.72)	10.65
Rest-frame B-band Luminosity (LB/L*)	0.70 (1.00)	1.0 (0.80)	0.85 (0.94)	4.2
Star Formation Rate (M⊙ yr−1)	1.1 (0)	≲0.2 (0)	≲0.5 (0)	0.01
Stellar Population Age (Gyr)	0.26 (1.00)	1.3 (1.00)	0.50 (1.00)	11.2
Projected Physical Offset δR (kpc)	5.5 (0.27)	20.7 (0.17)	6.9 (0.24)	2.1
Fractional Flux	0.14 (0.75)	0.21 (0.80)	0.19 (0.76)	0.54

Note. Columns correspond to: (1) stellar population property, (2) median for star-forming short GRB hosts, (3) median for elliptical short GRB hosts, (4) median for total short GRB host population, and (5) value for NGC 4993 from Blanchard et al. (2017). In each column, the numbers in parentheses represent the fraction of events in each class with values below that of NGC 4993.

Download table as:  ASCIITypeset image

We first compare the radio and X-ray observations of the counterpart to GW170817 to those following on-axis short GRBs. In the X-ray band, the counterpart is observed to brighten over $\approx 2.4\mbox{--}15.4$ days, indicative of a weak on-axis jet or an off-axis afterglow observed at an angle ${\theta }_{\mathrm{obs}}$ from the jet axis (Margutti et al. 2017). At the time of the first detection reported in Margutti et al. (2017) at $\approx 15.4$ days, the X-ray counterpart had a luminosity of $\approx 1.1\times {10}^{39}$ erg s⁻¹ ($0.3\mbox{--}10$ keV). The median luminosity of the five short GRBs that have X-ray detections at similar rest-frame times is $\approx 3\times {10}^{42}$ erg s⁻¹ (Figure 2); thus, the X-ray counterpart to GW170817 is $\approx 3000$ times less luminous. Moreover, the X-ray counterpart to GW170817 is $\approx 50$ times less luminous than the faintest known X-ray emission from a short GRB. Given that the depths of late-time observations following short GRBs are essentially at the limits of current X-ray observatories, Figure 2 demonstrates that the counterpart to GW170817, if shifted to the redshifts of short GRBs, would not have been detected.

The radio afterglows of short GRBs provide a complementary comparison sample. Alexander et al. (2017) report monitoring of the radio counterpart to GW170817 spanning $\approx 0.6\mbox{--}40$ days, with deep limits of $\gtrsim (2\mbox{--}5)\times {10}^{35}$ erg s⁻¹ until $\approx 19$ days and detections of $\approx (2.1\mbox{--}2.6)\times {10}^{35}$ erg s⁻¹ thereafter (Figure 2). Overall, the radio counterpart to GW170817 is a factor of $\gtrsim {10}^{4}$ times fainter than detected radio afterglows, which have a median of $\approx 4\times {10}^{39}$ erg s⁻¹. The faintest reported detection of an on-axis radio afterglow was following one of the most nearby known short GRBs, GRB 160821B (z = 0.16), with $\approx 1.2\times {10}^{38}$ erg s⁻¹ (Figure 2), $\approx 500$ times more luminous than the radio counterpart to GW170817, and on very different timescales. Based on the radio monitoring by Alexander et al. (2017), the radio counterpart to GW170817 at the distance of even the most nearby known short GRBs would not have been detected.

Margutti et al. (2017) and Alexander et al. (2017) demonstrate that the temporal behavior of the X-ray and radio counterparts can be jointly explained by an off-axis jet. In particular, they find that for a jet opening angle of 15°, set to the median for short GRBs (Fong et al. 2015), the best-fit family of solutions has a range of circumburst densities, $\approx {10}^{-4}\mbox{--}0.01$ cm⁻³, jet energies of ${10}^{49}\mbox{--}{10}^{50}$ erg, and observer angles, ${\theta }_{\mathrm{obs}}\approx 20\mbox{--}40^\circ$, where the range of solutions is dominated by the uncertainty in the microphysical parameters (Alexander et al. 2017; Margutti et al. 2017). The range of allowed energies and densities inferred from the radio and X-ray counterparts to GW170817 is fully consistent with those inferred from on-axis short GRB afterglows, which have medians of $\approx 8\times {10}^{49}$ erg and $\approx (0.3\mbox{--}1.5)\times {10}^{-2}$ cm⁻³, respectively (Fong et al. 2015). This remarkable similarity hints that viewing angle effects are the dominant, and possibly the only, factor causing the difference in the observed radio and X-ray behavior between this event and on-axis short GRBs.

Next we compare the luminosity and temporal evolution of the optical/NIR counterpart to GW170817 (Chornock et al. 2017; Cowperthwaite et al. 2017; Nicholl et al. 2017; Soares-Santos et al. 2017) to previous claims of kilonovae from short GRBs (Figure 3). We also investigate whether previous searches following short GRBs were sensitive enough to detect excess emission with similar luminosities to the counterpart of GW170817.

In the rest-frame optical band, the two claimed detections of kilonovae, GRB 050709 (Jin et al. 2016) and GRB 060614 (Jin et al. 2015; Yang et al. 2015), are $\gtrsim 3\mbox{--}5$ times more luminous than the counterpart to GW170817 at $\gtrsim 3$ days. At NIR wavelengths, the single detection following GRB 130603B at $\approx 7$ days had a luminosity $\approx 30$ times greater than the expected level of the afterglow at that time (Berger et al. 2013; Tanvir et al. 2013). This significant NIR excess, in conjunction with the extremely red color, represents the most convincing case of a kilonova following a cosmological short GRB. The comparison in Figure 3 demonstrates that the NIR excess following GRB 130603B was substantially more luminous, $\approx 3$ times brighter, than the NIR counterpart to GW170817 at the same epoch (Figure 3). Moreover, all three claimed kilonovae remain luminous for longer after the merger in their respective rest-frame bands.

At the most basic level, the larger observed luminosities and sustained brightness for the three short GRBs could be explained by differences in the masses and velocities of ejected material during and subsequent to the mergers (e.g., Metzger et al. 2010; Barnes & Kasen 2013). For both GRB 050709 and GRB 060614, the inferred ejecta masses and velocities are $\approx 0.05\mbox{--}0.1\,{M}_{\odot }$ and $0.1\mbox{--}0.2c$ (Yang et al. 2015; Jin et al. 2016). Using single-component models available at the time (Barnes & Kasen 2013), the luminosity of the kilonova following GRB 130603B translated to an ejecta mass of $\approx 0.03\mbox{--}0.08\,{M}_{\odot }$ for velocities of $0.1\mbox{--}0.3c$ (Berger et al. 2013).

However, it has been suggested that there may be an additional early "blue" kilonova component as the result of lanthanide-free material from disk winds. This component may be amplified by the formation of a hypermassive NS following the merger that is at least temporarily stable to collapse (Metzger & Fernández 2014; Kasen et al. 2015; Metzger 2017). Indeed, observations of the counterpart to GW170817 have demonstrated that single-component models are likely inadequate for explaining kilonova behavior. The combination of multi-band optical/NIR photometry of the counterpart to GW170817 (Cowperthwaite et al. 2017), high-cadence UV/optical spectroscopy (Nicholl et al. 2017), and high-cadence NIR spectroscopy (Chornock et al. 2017), show evidence for "blue" and "red" kilonova components with overall lower inferred ejecta masses when compared to values inferred from short GRBs: $0.01\,{M}_{\odot }$ and $\approx 0.03\mbox{--}0.04\,{M}_{\odot }$, respectively. Still, the comparison between potential kilonovae following short GRBs and the optical/NIR counterpart to GW170817 suggests that there may be a broad range of kilonova luminosities, colors, and timescales.

Finally, we utilize optical/NIR upper limits following short GRBs to explore whether such searches would have been sensitive to the optical/NIR counterpart to GW170817, had it originated at higher redshifts. There are numerous optical limits following short GRBs at $\lesssim 1$ day from afterglow searches (Fong et al. 2015), and the deepest limit is following GRB 050509B, with ${M}_{r}\approx -14.3$ at $\approx 0.9$ days (Cenko et al. 2005; Bloom et al. 2006). This is $\approx 1$ mag deeper, or $\approx 2.2$ times less luminous, than the optical brightness of the counterpart to GW170817 (Cowperthwaite et al. 2017; Soares-Santos et al. 2017) interpolated to the same rest-frame time. Thus, a "blue" kilonova of similar luminosity to GW170817 can be confidently ruled out for GRB 050509B (see also Hjorth et al. 2005a; Metzger et al. 2010), which also likely originated from an elliptical host galaxy at a low redshift of z = 0.225 (Gehrels et al. 2005; Bloom et al. 2006). We note that since this event was detected as an on-axis short GRB, it is unlikely that there was significant obscuration of any "blue" component by a "red" kilonova component, which is expected to be more equatorial in geometry (Metzger & Fernández 2014).

Optical and NIR searches following all of the remaining short GRBs were not sensitive enough or did not occur on the proper timescales to detect a kilonova with similar luminosities to that of GW170817. If the luminosity of the counterpart to GW170817 is representative of kilonovae from NS mergers, then searches following cosmological short GRBs will only probe the brightest end of the kilonova luminosity function. An added complication to additional kilonova detections is that $\approx 30 \%$ of on-axis short GRBs have optical afterglow emission that dominates at early times, with a median of ${M}_{r}\approx -18.8$ mag at 10 hr (Berger 2014; Fong et al. 2015). Thus, if the luminosity of GW170817 is representative of most kilonovae, the most promising route to accurately mapping the luminosity function is via Advanced LIGO/Virgo triggers.

The locations of short GRBs with respect to their host galaxies are an important probe of their progenitor systems, as they contain key information about their local environments at the time of explosion. Past studies have shown that short GRBs have a median projected physical offset ($\delta R$) of $\approx 5\,\mathrm{kpc}$ (Fong et al. 2010; Fong & Berger 2013), and span a wide range of $\approx 1\mbox{--}75\,\mathrm{kpc}$ (Berger 2010; Tunnicliffe et al. 2014). Moreover, as a population, their locations are only very weakly correlated with their underlying host light distributions, indicating that they do not trace the distributions of their hosts' stellar mass or star formation. In general, their locations demonstrate that the progenitors must migrate from their birth sites to their explosion sites, and have been used as one of the primary arguments that they originate from NS–NS/NS–BH progenitors (Berger 2010; Fong et al. 2010; Fong & Berger 2013; Tunnicliffe et al. 2014).

In this vein, Blanchard et al. (2017) use HST optical and near-IR data and find that the counterpart to GW170817 is located at a projected physical offset of $\approx 2.1\,\mathrm{kpc}$ from the center of NGC 4993. This value is low compared to short GRBs, with only $\approx 17 \% \mbox{--}27 \%$ of events located closer to their host centers, regardless of host galaxy type (Figure 4 and Table 1). While the median offset for short GRBs in elliptical hosts is somewhat large, $\approx 21\,\mathrm{kpc}$, this is based on only six events which span the full range of $\approx 1\mbox{--}50\,\mathrm{kpc}$ (Figure 4).

Zoom In Zoom Out Reset image size

A complementary tool to physical offsets is to examine the location of the counterpart to GW170817 with respect to NGC 4993's light distribution ("fractional flux"). Using this method, Blanchard et al. (2017) find that the counterpart lies on a region of average brightness in the rest-frame optical band, with a fractional flux value of 0.54, indicating that there is nothing unusual about this position with respect to its host stellar mass distribution. However, this value is high compared to short GRBs, with $\approx 75 \% \mbox{--}80 \%$ of events on lower brightness levels (Figure 4 and Table 1). Together with the small projected physical offset and old stellar population age of $\approx 11.2\,\mathrm{Gyr}$, this implies that the progenitor of GW170817 may have had a comparatively small kick velocity compared to those inferred for short GRBs, or happened to be on a part of its trajectory that was close to its host at the time of merger. However, since we have no a priori information on where the system was born, we can only place an upper limit on the kick velocity using the velocity dispersion of the host galaxy, which Blanchard et al. (2017) find to be $\approx 150\mbox{--}200$ km s⁻¹. Furthermore, if the location of the counterpart to GW170817 is representative, NS–NS mergers may be more correlated to their host stellar mass distribution than implied from short GRBs alone.

Characterizing the host galaxy environments of short GRBs is key to understanding the parent stellar populations in which these explosive transients were born. From fitting the optical and NIR SED of NGC 4993, Blanchard et al. (2017) find that NGC 4993 has a stellar mass of log(${M}_{* }/{M}_{\odot })=10.65$, rest-frame B-band luminosity of $\approx 2\times {10}^{10}\,{L}_{\odot }$ ($4.2{L}^{* }$ when compared to the local galaxy luminosity function Blanton et al. 2003), stellar population age of $\approx 11.2\,\mathrm{Gyr}$, and SFR of $0.01\,{M}_{\odot }$ yr⁻¹ (Table 1). Moreover, their analysis from surface brightness profile fitting indicates an elliptical morphology and half-light radius of $\approx 3\mbox{--}4\,\mathrm{kpc}$. Capitalizing on the past decade of short GRB host galaxy studies (Berger 2009; Fong et al. 2013; Berger 2014), we can now compare the stellar population of NGC 4993 with the rest-frame B-band luminosities, stellar masses, stellar population ages, and SFRs of short GRB hosts.

The elliptical morphology, along with the low level of ongoing star formation in NGC 4993, is consistent with an early-type galaxy designation, which comprise only $\approx 1/3$ of short GRB hosts; the majority of short GRBs occur in star-forming galaxies (Berger 2009; Fong et al. 2013). NGC 4993 is among the most optically luminous host galaxies, superceded only by two other elliptical galaxies at low redshift: GRB 050509B ($z=0.225;$ Gehrels et al. 2005; Bloom et al. 2006) and GRB 150101B ($z=0.134;$ Fong et al. 2016). However, its half-light radius is at the median level compared to short GRBs of 3.5 kpc, suggesting a comparatively compact morphology.

NGC 4993 also has by far the oldest stellar population age of $\approx 11.2\,\mathrm{Gyr}$, with the short GRBs in our sample spanning $\approx 0.03\mbox{--}4.4\,\mathrm{Gyr}$. Using stellar population age as a proxy for the delay time between the formation and merger of the progenitor system, this suggests a very broad range of delay times, as predicted from population synthesis models of NS–NS binaries (Belczynski et al. 2006). For a delay time distribution of the form $P(\tau )\propto {\tau }^{n}$, the stellar population ages of short GRBs, together with the relative burst rates in elliptical and star-forming galaxies, suggest that $n\approx -1$ (Zheng & Ramirez-Ruiz 2007; O'Shaughnessy et al. 2008; Leibler & Berger 2010; Fong et al. 2013). However, if there continues to be an increasing number of detected binary neutron star mergers with long delay times of $\gtrsim 5\,\mathrm{Gyr}$, the shape of the delay time distribution will shift toward values of $n\gtrsim -1$ (e.g., Zheng & Ramirez-Ruiz 2007). Additional events detected within the Advanced LIGO/Virgo volume will be necessary in order to provide a low-redshift anchor and delineate the delay time distribution.

In terms of SFRs, short GRB star-forming hosts span a range of $\approx 0.1\mbox{--}30\,{M}_{\odot }$ yr⁻¹ (Berger 2009, 2014; Perley et al. 2012; Berger et al. 2013), with a median of $\approx 1.1\,{M}_{\odot }$ yr⁻¹, comparable to that of the Milky Way (Table 1). Short GRB hosts with no previous evidence of ongoing star formation had limits of $\lesssim 0.05\mbox{--}1.5\,{M}_{\odot }$ yr⁻¹ (Berger 2009, 2014; Table 1). In this context, NGC 4993 also has the lowest measured SFR of any short GRB host galaxy, enabled by its proximity, and is well below the most constraining limits for cosmological hosts. This low SFR is also commensurate with its old stellar population age (Blanchard et al. 2017).

Finally, the inferred stellar masses for the short GRBs in our sample have a median of log $({M}_{* }/{M}_{\odot })\approx 10.1$ (Table 1). NGC 4993 thus has a comparatively high stellar mass relative to the short GRB population, with log(${M}_{* }/{M}_{\odot })=10.65$, as $\approx 70 \%$ of hosts have lower stellar masses. Compared to the stellar masses of elliptical galaxies, NGC 4993 is well below average, with only $\approx 1/3$ of short GRB elliptical hosts having lower stellar masses.

In summary, NGC 4993 is superlative in its large optical luminosity, old stellar population age, and low SFR compared to the short GRB host population. It is average in terms of its size, and above average in terms of its stellar mass.