THE ULTRA-LONG GAMMA-RAY BURST 111209A: THE COLLAPSE OF A BLUE SUPERGIANT?

Swift/XRT data were taken as Level 1 event files from the Swift Data Center archive at NASA/GSFC and were processed with the task xrtpipeline version 0.12.6, applying the most updated calibrations. Data acquisition started in Windowed Timing (WT) mode and continued in this mode up to the fourth orbit (T₀ + 22700 s). XRT then switched to the Photon Counting (PC) mode for the remaining of the observation (up to 26 days after the trigger). Pile-up can be observed in WT mode during the first orbit (i.e., up to T₀ + 2000 s), and in PC between T₀ + 22700 s and T₀ + 30900 s. As a consequence, spectra and light curves were extracted using the methods of Romano et al. (2006) and Vaughan et al. (2006), excluding a central circle of radius 1 pixels from the point-spread function (PSF) in WT mode; for the PC mode, the radius was set dynamically between 7 and 1 pixels as a function of the flux of the afterglow, and the results were checked with the PSF profile of the XRT.

All spectra and light curves were then extracted using standard filtering and screening criteria (including bad columns and bad pixels), and corrected for telescope vignetting and aperture filtering using the ancillary response files (ARFs) generated by the task xrtmkarf. We then rebinned all files to either match the spectral resolution when the data quality was high enough, or to obtain at least 30 counts per bin.

We retrieved the Observation Data Files from the XMM-Newton archive and reprocessed them using the XMM-SAS version 12.0.1 and the latest available calibration files associated with this version of the SAS. The raw events files were processed using the tasks emchain, epchain, rgsproc, omichain, omfchain. The PN instrument observed GRB 111209A in Full Window mode, while the MOS cameras setting was on Small Window. Because of the brightness of the afterglow, we checked for pile-up in the data. We used the task epatplot for this purpose, and confirmed that the data were free from pile-up, even if a significant fraction of out off time events were observed. Because of this latter effect, the background regions have been chosen, when possible, far away from the source and corrected for the difference of off-axis angle using the ARFs. In the special case of the PN instrument, the instrumental background is highly variable depending on the position on the instrument—this was taken into account by using a region where the instrumental background was compatible with the position of the source.

We then checked for flares of solar protons which can induce a huge increase of the background count rate, and found some marginal events where the background rate was increased by a factor of five. Due to the brightness of the afterglow, these events are not very significant for the following analysis: when maximal, the background represents around 6% of the total signal at the position of the afterglow.

The event files were then filtered for good photons, using FLAG == 0 and Pattern filtering (less than 4 for the PN, less than 12 for the MOS). Spectra were extracted using the task especget, light curves with the task evselect.

GRB 111209A X-ray emission indicates an initial shallow decay phase. Assuming a power-law model between T₀ + 0.425 ks and T₀ + 10.4 ks, the best-fit decay index is α_p = 0.544 ± 0.003. This phase ends very late in comparison with typical X-ray emissions. At the end of this phase, a gradual steepening drives the light curve to the so called "steep decay" phase, thought to indicate the end of the prompt emission and is characterized by very fast decrease of the count rate. During this phase and the following one, the light curve can be modeled with a double broken power law with a steep decay index of α_s = 4.9 ± 0.2, a plateau with decay index of α_f = 0.5 ± 0.2, and a final decay index of α_a = 1.51 ± 0.08.

In the following, all models fitted to the data were multiplied by two components in order to take into account our Galaxy and the host galaxy photoelectric absorption from metals. The equivalent hydrogen column density of our Galaxy was set to N_H = 1.48 × 10²⁰ cm⁻² (Kalberla et al. 2005). The second absorption component was left free to vary at the redshift of the burst, assuming solar metallicity. We found no temporal variation of the intrinsic N_H, with a mean of (2.5 ± 0.4) × 10²¹ cm⁻².

We began with an analysis of the prompt spectral emission between T₀ and T₀ + 2500 s, seen by Swift/XRT. GRB 111209A shows a strong flare at T₀ + 1200 s. Since flares are known to have highly variable spectra, this was excluded in order to investigate the underlying spectral continuum. The spectrum was of exceptionally high quality, and was rebinned in order to obtain at least 400 counts per bin. The background subtracted energy spectrum between T₀ + 425 s and T₀ + 1000 s is inconsistent with a simple power-law or cutoff power-law model. The best-fitting model was that based on a broken power law added to a blackbody component. The high-energy segment of the broken power law is consistent with the BAT spectrum at the same time (Γ = 1.48 ± 0.03; Palmer et al. 2011), but harder than the time-averaged Konus-Wind best-fit model. We note however that the following spectra (extracted between 5.0 and 47.8 ks) are softer, consistent with the time averaged Konus-Wind spectrum. We therefore assume that the spectral position of the break (i.e., E_break) varies with time, and that this break is due to the crossing of a specific frequency within the observed band. The blackbody component is very soft (see Table 1 and Figure 2), and accounts for about 0.01% of the total flux in the 0.5–10.0 keV band. Its luminosity is ∼1.6 × 10⁵⁰ erg s⁻¹. Moreover, this component is not detected at later times.

Zoom In Zoom Out Reset image size

Table 1. X-Ray Spectral Analysis of GRB 111209A

Instrument	Start	End	Spectral	Γ	E0	Second Spectral	$\chi _{\nu }^2$
Observing	Time	Time	Model			Component	(dof)
Mode	(s)	(s)			(keV)	(eV or energy index)
Swift/WT	425	1000	BKN+BB	1.17 ± 0.03	3.4 ± 0.4	30 ± 5	0.98 (181)
				1.52 ± 0.06
Swift/WT	5025	7825	PL	1.58 ± 0.03			1.13 (590)
Swift/PC	22700	25130	PL	1.8 ± 0.1			0.87 (50)
Swift/PC	28780	30880	PL	1.9 ± 0.1			0.78 (32)
XMM-Newton	55073	61416	2PL	0.9 ± 0.7		2.7 ± 0.2	0.89 (151)
XMM-Newton	61416	66416	2PL	$1.2^{+0.7}_{-0.9}$		$2.8^{+0.4}_{-0.2}$	0.84 (111)
XMM-Newton	66416	71416	2PL	$1.9^{+0.4}_{-0.8}$		$3.3^{+0.9}_{-0.5}$	0.86 (95)
XMM-Newton	71416	76416	2PL	$1.7^{+0.5}_{-0.9}$		$3.1^{+0.8}_{-0.4}$	1.05 (83)
XMM-Newton	76416	91416	2PL	$1.7^{+0.5}_{-0.7}$		$3.0^{+0.5}_{-0.3}$	0.89 (228)
XMM-Newton	91416	108599	2PL	$2.3^{+0.3}_{-0.8}$		$3.4^{+1.3}_{-0.6}$	1.04 (234)

Download table as:  ASCIITypeset image

The XMM-Newton monitoring period covered the transition from the steep decay phase to the "normal" decay phase, for an interval of 53,000 s. Because of the greater sensitivity of XMM-Newton in comparison with the Swift/XRT, we were able to extract the spectra in six different time intervals (see Table 1). A simple power-law model was first used, giving an acceptable value of the χ² statistic value. However, the residuals of the fit clearly showed systematic excesses at the hard end of the spectrum, and thus a double component model was preferred. The inclusion of a second component produced an F-test probability (the probability that the decrease of the χ² statistic value is due to a spurious effect) of the order of 10⁻⁹. The spectral fit clearly indicates that this second component is not dominating, and provides a small contribution to the total observed flux.

We extracted a spectral energy distribution (SED) during the XMM-Newton observation using published results from Kann et al. (2011) in the optical, in order to confirm no thermal emission was present during this phase. Kann et al. (2011) already noted that the GROND data excluded a thermal component, and indeed we have found an optical spectral index of β = 1.1 ± 0.2, using the optical data alone. Using the X-ray to optical data, we found that an absorbed broken power law fitted the data well. We note that we can exclude the presence of a thermal emission in the far UV or the soft X-ray parts of the broad band spectrum.

As stated previously, this paper focuses on the properties of the progenitor of GRB 111209A, and the nature of this second component will be investigated in more detail in a forthcoming paper (G. Stratta et al., in preparation). We note however that this is the first time, to date, that such a double non-thermal component has been detected during the steep decay phase. Several models can account for it (e.g., see the model of Barniol Duran & Kumar 2009). The optical light curve features a rebrightening during the XMM-Newton observation: this component may be linked to the prompt-to-afterglow transition, to the optical bump, or even be totally unrelated to these components.

The estimation of the burst duration using classical methods (i.e., T90) is complicated by two factors: it is well known that the duration depends on the observing band (the softer the band, the longer the detected signal); and for this event the prompt phase is split into several orbits, with gaps in the signal. Significantly, the event was recorded by Konus-Wind before T₀, when the burst was not in the field of view of Swift/BAT. Therefore, all attempts to use classical methods are flawed. To give an estimate of the total duration for comparison with other bursts, we thus estimated the burst start and stop epochs T_start and T_end as follows:

• 1.  
  For T_start, we use the start of the detection by Konus-Wind, about 5400 s before T₀.
• 2.  
  For T_end, we use the time of the start of the sharp decay, about 20,000 s after T₀.

The choice of the T_end value is motivated by the fact that the sharp decline is usually interpreted as the high-latitude emission of the prompt phase, and its start time corresponds to the true end of the prompt phase (Kumar & Panaitescu 2000). Before the steep decay phase, the initial XRT and BAT light curves are usually well correlated, indicating a common origin. This is also the case for GRB 111209A (see Figure 1). With these assumptions, the total duration of GRB 111209A is at least 25,000 s.

We extensively surveyed the Swift, BATSE, and Konus-Wind catalogs to find other ultra-long bursts. The most complete and homogeneous catalog of prompt burst properties is the BATSE 4B catalog (Paciesas et al. 1999). From this catalog, only few bursts have a duration larger than 1000 s (the super-long ones), and none have a duration longer than 10,000 s. For the case of Swift, we also searched in the XRT light curve repository (Evans et al. 2007) to obtain the start time of the sharp decay for all bursts, and found no burst with such a large duration. Thus, GRB 111209A is to date the longest confirmed GRB ever observed, and its duration is exceptional.

We note, however, for completeness, that one event, longer than GRB 111209A, has a debated origin. Swift J164449.3+573451 (hereafter J1644+57) has been proposed by several groups to be a tidal disruption event (e.g., Burrows et al. 2011; Bloom et al. 2011; Levan et al. 2011; Tchekhovskoy et al. 2013). Nevertheless Quataert & Kasen (2012), even though agreeing that the most plausible origin was a tidal disruption event, have shown theoretically that the explosion of a massive star could produce a similar event.

It is possible that selection effects make only the brightest part of ultra-long burst light curves detectable, thus reducing their intrinsic durations and making ultra-long GRBs indistinguishable from normal bursts. We investigated this selection effect in two ways.

First, we constrained the maximum distance from which this burst could be detected. We extracted data from the Konus-Wind light curve, rescaling it to a given distance and then added the background of the instrument (adding statistical fluctuations in the final signal to take into account Poisson statistics). We used two signal levels: the original one, and a second level an order of magnitude fainter. The light curve was examined for a burst with a rate trigger threshold of 5σ. We used two different bin sizes for the light curve (1 and 2 minutes), the latter simulating more complex trigger methods where an image trigger can be considered. The simulation was performed 100 times for all redshifts between 0.3 and 2.2 (with a step of 0.1). Results are presented in Figure 3, and are quite clear: GRB 111209A cannot be detected at high-z. In fact, this event can be detected only within a limiting redshift of z = 1.4.

Zoom In Zoom Out Reset image size

Second, for positive detections (at z < 1.4), we computed the expected duration of the burst, using the standard GRB procedure to obtain T90. Again, this is done for two distinct temporal resolutions and flux thresholds. Results are presented in Figure 4. Surprisingly, we find that the duration is larger than 10000 s in all cases. Thus, an ultra-long burst that could trigger an instrument would always be detected as an ultra-long GRB. The lack of such other bursts in ∼30 yr of archives indicates a very rare event in the local universe. Such events may have been more common in the distant universe, but their faintness makes their detection unfeasible.

Zoom In Zoom Out Reset image size

We estimated the rate of ultra-long GRBs in the local universe using the properties of GRB 111209A. Taking into account the effect of the redshift on Swift's sensitivity and k-correction (Coward et al. 2012; Howell & Coward 2013), this rate density can be inferred from the measured peak flux. We used the brightest epoch of the GRB 111209A prompt phase, as seen by Konus-Wind, scaled to the Swift energy band. This is then scaled to the limiting detection flux of Swift to obtain a maximal redshift for detection for Swift, z_lim (see Coward et al. 2012, for details). The corresponding volume of detectability is then:

$$\begin{equation} V_{{\rm max}} = \int ^{z_{{\rm lim}}}_{0} \frac{dV}{dz}dz. \end{equation} \tag{ 1 }$$

Using the cosmological model provided in the Introduction, the maximum detectable volume corresponding to z_lim = 1.4 is V_max = 292 Gpc³. Following Guetta & Della Valle (2007); Coward et al. (2012) we calculate a rate density

$$\begin{equation} R = \frac{1}{V_{\rm {max}}} \frac{1}{T} \frac{1}{\Omega } \frac{1}{\eta _{z}}\,, \end{equation} \tag{ 2 }$$

with T the maximum observation-time for the sample (7 yr), Ω = 0.17 the sky coverage of the instrument and η_z = 0.3 the fraction of Swift bursts with a measured redshift. We obtain $9^{+ 27}_{-8} \times 10^{-3}$ Gpc⁻³ yr⁻¹ where the errors are the 90% Poisson confidence limits (Gehrels 1986).

As only four out of 644 Swift bursts detected between the launch of the satellite and 2012 March have prompt emission durations greater than some thousands of seconds, we obtain the final rate for ultra-long GRBs to be of the order of 4 × R/644 ∼ 6 × 10⁻⁵ Gpc⁻³ yr⁻¹. This is significantly lower than that of the normal long GRB rate density (see Howell & Coward 2013).

We retrieved from the literature the prompt emission information available for the nine super-long bursts (with duration larger than 1000 s) listed in Table 2. The fact that GRB 111209A is harder and more fluent than other bursts is another indicator of its uniqueness. In Figure 5 we also present these bursts in the duration–fluence plane. GRB 111209A appears as an outlier in the duration distribution, but not in the fluence distribution, where other super-long GRBs can have a similar (even if slightly smaller) fluence. We note that supernova shock breakouts have a much smaller fluence, weakening a possible link between this kind of progenitor and GRB 111209A. Once corrected for distance (when possible—several bursts lacked a good follow-up campaign and thus a redshift measurement), this burst also appears to be more energetic than the others.

Zoom In Zoom Out Reset image size

We also investigated the X-ray afterglow properties of these bursts. As shown in previous works using infrared, optical, and X-ray data, the luminosity of afterglows tends to cluster in several groups (Boër & Gendre 2000; Gendre & Boër 2005; Liang & Zhang 2006; Nardini et al. 2006; Gendre et al. 2008a, 2008b). If one includes the X-ray light curve of GRB 111209A in the sample of Gendre et al. (2008a; see Figure 6), the event is shown to be marginally under luminous. However, this could be interpreted as a normal consequence of the duration of the prompt phase. In this phase, internal shocks dissipate the energy of the fireball. The longer the prompt phase lasts, the less energy remains to be dissipated within the following external shock.

Zoom In Zoom Out Reset image size

The first hypothesis is that the exceptional duration of GRB 111209A is due to a component usually not present in normal GRBs. Two super long GRBs (GRB 060218, z = 0.033, and GRB 100316D, z = 0.059) were associated with a supernova shock breakout (Campana et al. 2006; Starling et al. 2011). These two events are also classified as sub-energetic (Howell & Coward 2013), with very soft X-ray spectra, and their prompt spectra (including the steep decay) show a thermal emission, which is responsible for the long duration. We also do observe a thermal component at the start of the XRT observation; however this component disappears very soon, and most of the prompt phase is free of thermal emission. We can therefore discard this hypothesis.

Starling et al. (2012) have found in several GRBs associated with a supernova a thermal component. They explain it as a hot plasma cocoon associated with the jet. However, they observe this component in the steep decay phase (where we do not see any thermal emission from GRB 111209A using XMM-Newton data), which is harder and dimmer than the one we detect at the start of the prompt phase. Thus, this model cannot be applied to GRB 111209A.

GRB 101225A, another super-long burst (Racusin et al. 2010), shows a strong optical thermal component. The lack of a redshift measurement for this burst made difficult to draw any conclusions on the explosion mechanism and two very different models were proposed: a minor-body tidal disruption by a neutron star in our galaxy (Campana et al. 2011), and the merger of a neutron star with a helium star in a distant galaxy (Thöne et al. 2011).

Apart from the lack of any strong thermal emission at optical (and X-ray) wavelengths, GRB 111209A differs from GRB 101225A as it is at least 10³ times more energetic (using the most distant scenario of Thöne et al. 2011), suggesting a different origin for GRB 111209A.

In recent years, evidence has emerged that magnetars could account for long GRB properties (e.g., Metzger 2010). This class of progenitor provides a natural explanation of extended emissions as long lasting energy injection (Dall'Osso et al. 2011; Metzger et al. 2011). Indeed, if the typical duration of a long GRB powered by a magnetar is of the order of 40–100 s, atypical values of the magnetic field and spin period could allow a duration of ∼25, 000 s. However, such atypical values cannot fit both the observed values of E_iso and E_p. For instance, using the formalism of Metzger et al. (2011), reproducing the observed values of E_iso and duration would lead to E_p ∼ 180 keV, that is a factor of three lower than the rest frame peak energy of GRB 111209A (520 keV). Trying to conciliate these values results in nonphysical solutions (e.g., a neutron star spinning at a value exceeding the breakup value, P_lim = 0.96 ms, Lattimer & Prakash 2004).

Another possibility would be to consider the long duration of the prompt phase as a plateau phase (i.e., assume the true prompt phase has been missed by all instruments). However, the magnetar model indicates that once the plateau phase is finished the decay should be monotonic with a decay index of 1–2 (Dall'Osso et al. 2011). This is in contradiction with our findings from the light curve (decay of α = 4.9 ± 0.2 with the presence of a following plateau).

Troja et al. (2007) has proposed a solution (later extended by Lyons et al. 2010) to this problem, with a two-stage progenitor: the collapsar would produce a magnetar that would collapse into a black hole a few hundreds of seconds later. Under this scenario one could expect a double plateau, i.e., a non-monotonic decay after the first plateau phase. The first plateau, due to the energy injection by the magnetar, would then be the observed prompt signal. However, again, one needs atypical values of the magnetic field and spin period to explain the observed properties of this event, and again this would produce inconsistent values of E_iso and E_p, not observed in this event. Significantly, the "first plateau" has also been observed in the gamma-ray band (it is the prompt signal), which is itself unique, making GRB 111209A a very peculiar event.

We thus conclude that a magnetar progenitor for GRB 111209A is not supported by the data.

Swift J1644+57 was at first claimed to be a GRB. It is thus possible that GRB 111209A is not a GRB but a tidal disruption event. This solution is, however, not very convincing. Firstly, no host galaxy has been detected, while one may expect a massive black hole to lie in a galaxy bright enough to be detected at small redshift. Second, the light curve should decay as t^−5/3 (Lodato et al. 2009). This is however not the case (see Section 4). In fact, Swift J1644+57 did not show the characteristic GRB light curve behavior (so called steep-flat-steep, Zhang 2007) clearly visible in the case of GRB 111209A. For these reasons we can discard this hypothesis.

The main difficulty in explaining the nature of the progenitor of GRB 111209A is its duration. Woosley & Heger (2012) investigated scenarios to provide a long duration event with an energy budget typical of a long GRB. They proposed four scenarios:

• 1.  
  single supergiant stars with less than 10% solar metallicity;
• 2.  
  supergiant systems in tidally locked binaries;
• 3.  
  pair instability in very high mass stars resulting in collapse to a black hole;
• 4.  
  helium stars in tidally locked binaries.

Scenarios (2) and (4) lead, assuming a typical values of 1% $\dot{M}c^2$ for the energy conversion efficiency, to luminosities orders of magnitude too faint in comparison to the that observed in the case of GRB 111209A. Only a 100% energy conversion efficiency would marginally reconciliate these numbers. As for scenario (3), at a distance of z = 0.677, one would expect a bright optical supernova easily detectable with large aperture telescopes. Despite follow-up of GRB 111209A by the Hubble Space Telescope and the Very Large Telescope, no evidence of a supernova was claimed (Levan 2012). The above arguments leave us with scenario (1), a single supergiant star with low metallicity. We note that Quataert & Kasen (2012) used the same kind of progenitor to discuss the link between Swift J1644+57 and a GRB.

The hypothesized progenitors of long duration GRBs are Wolf–Rayet stars (stars with the outer layers expelled during stellar evolution). When these layers are still present, as in low-metallicity supergiant stars with weak stellar winds, the stellar envelope may fall back and accretion can fuel the central engine for a much longer time. In this scenario, blue supergiant stars can produce GRBs with prompt emission lasting about 10⁴ s (Woosley & Heger 2012).

Theoretically, a larger star, such as a red supergiant star could also be considered, assuming that the outer layers are expelled in order to maintain the free-fall time scale to the order of 10⁴ s. For instance, a very weak explosion could eject the convective envelope of a red supergiant without producing an observable supernova, (e.g., Dexter & Kasen 2012). However, in such a case, this layer of matter is still surrounding the progenitor of the GRB, making the surrounding medium density profile much more complex. As shown by e.g., Chevalier et al. (2004), these complex geometries should imprint their mark on the afterglow light curve, and in fact do (see for instance GRB 050904, Gendre et al. 2007). Here we do not see this complex afterglow behavior, supporting the suggestion of a simple blue supergiant star.

If one considers the ratio of normal GRBs to ultra-long GRBs, this is shown in the Swift archives to be of the order of 200:1. In comparison, the fraction of blue supergiants compared to Wolf–Rayet stars is typically 10:1 (Eldridge et al. 2008)—therefore ultra-long GRBs should be more frequent than normal long GRBs. At first consideration, these numbers are difficult to reconcile. However, the greatest uncertainty of supergiant collapsars is whether matter with high angular momentum will be ejected by mass loss or will fall into the center of the star (Woosley & Heger 2012). It is plausible that for most of blue supergiant stars there is no fall back of the external layers, and thus no long emission. Hence, the observed and the expected rates may be reconciled by assuming the external layers of most blue supergiants do not fall into the stellar core. A rare combination of exceptionally large rotation and small mass loss at the end of stellar evolution could lead to a GRB 111209A like event. These properties can be both achieved in case of a low-metallicity progenitor. This would also help reconcile the numbers, as most of the local universe blue supergiant stars are not observed with low metallicity. One should then consider if it is possible to have these low-metallicity stars in the local universe.

The most putative hypothesis on the above consideration is the presence in the local universe of a galaxy with a very low metallicity. In fact, blue supergiants have short lifetimes, and the only way of having a low-metallicity blue supergiant is to form it from low-metallicity gas. We must thus discuss the possibility of having very low metallicity galaxies in the local universe. If such galaxies should have been common in the early universe, the hierarchical galaxy formation model implies that these have merged to form the larger galaxies observed today (see, e.g., Klypin et al. 1999). However, as pointed out by several authors, such a mechanism is not efficient at 100% (e.g., Klypin et al. 1999; Diemand et al. 2007), and galaxies with a very low metallicity should still exist in the local universe. In fact, several of those are known (Morales-Luis et al. 2011), and one object (I Zw 18) has a measured oxygen abundance of 0.05 solar one (Izotov et al. 1997). These are local objects (with z < 0.1), accounting for about 0.1% of all the local galaxies (Morales-Luis et al. 2011), thus reinforcing the idea that GRB 111209A is a very rare event.

Furthermore, as noted by Morales-Luis et al. (2011), the luminosity–metallicity relationship (Lequeux et al. 1979) implies that low-metallicity galaxies cannot be observed in the more distant universe. The fact that the host galaxy of GRB 111209A has not been resolved by the Hubble Space Telescope (Levan 2012) would then support this hypothesis: only the GRB afterglow was visible, and GRB 111209A traces the location of a putative (very) metal poor galaxy at large distance (z = 0.677). At this distance, this galaxy would not have been detected without the GRB which occurred in it.