5

THE PROGENITORS OF GAMMA-RAY BURSTS

One day I undertook a tour through the country, and the diversity and beauties of nature I met with in this charming season, expelled every gloomy and vexatious thought.

—Daniel Boone, from “Boone’s Narrative,"
appended to John Filson, The Discovery, Settlement,
and Present State of Kentucke, 1784, 54–55

The physics of the central engine (§2.3) is constrained by basic measurements of GRB variability, inferred luminosity and energy release, and event spectra. This has led to the conclusion that a newly formed black hole or rapidly spinning neutron star lies at the heart of most distant GRBs: popular-level promotions of GRBs have noted, not unreasonably, that a GRB is the birth cry of a newborn black hole. But the tale we spin about GRB central engines may be more aptly seen as the denouement of an epic story about the tragic life cycle of the object(s) that produces the catastrophic event leading to the GRB. Studying GRB progenitors, then, is at once a quest to uncover the sorts of the preexplosion astrophysical entities that lead to a GRB and to understand how those entities evolved during their lifetime. Since clearly not all objects in the Universe make GRBs, we wish to understand what sets GRB progenitors apart from non-GRB progenitors. Are the progenitors of GRBs, like Romeo and Juliet, fated to die so catastrophically from the outset, or are they just undistinguished actors who took some unfortunate turns in their life path?

By now it should be evident that there are few absolute certainties with GRBs; so it should come as little surprise that the understanding of the progenitors is no exception. One clear standout is the progenitors of Soft Gammaray Repeaters which are very obviously neutron stars. There are a number of corroborating lines of evidence for this progenitor association: (1) Some well-localized SGRs are associated with supernova remnants, suggesting they are byproducts of recent SNe (and many types of SNe are thought to leave behind NSs); (2) there is quiescent X-ray emission from the sites of SGRs, similar to a class of NSs called “anomalous X-ray pulsars”; (3) Galactic SGRs tend to be found in the Galactic plane (see figure 1.3), where most young NSs reside; and (4) the ringdown emission after SGR pulses is periodic, with periods comparable to that of slowly rotating NSs (few seconds). Moreover, there is good evidence that very high magnetic field strengths set the progenitors of SGRs apart from other neutron stars. The magnetic field strengths are inferred by observation of a slightly decreasing period between successive events, directly revealing the rate at which the progenitor appears to be spinning more slowly with time. One can calculate the amount of energy required to “brake” the progenitor’s spin,¹ which, in turn, implicates magnetic fields of strength > 10¹⁵ Gauss, tens of thousands of times higher than typical pulsar fields. The precise origin of the high-energy burst and the cause for the diversity of luminosities and light curves are not well understood.

5.1 A Massive-Star Origin

If magnetars are the most securely known progenitor of a subclass of GRBs, the next in line are massive stars, very likely the origin of some long-duration GRBs. We have seen (§4.2.1, §4.2.2) that the locations of long-duration GRBs around star-forming galaxies provide strong (but circumstantial) evidence for a massivestar origin of such events. Since the afterglows of long bursts appear to obliterate the incriminating evidence of their progenitor (§4.1.1), the best evidence comes in the aftermath of the GRB, as the afterglow is fading and the results of having blown up a star become manifest. We discuss this observational evidence in §5.1.2 but now turn our attention to physical models of massive-star progenitors.

5.1.1 Collapsars and Friends

The complexity of the life cycle of stars can be summarized as a long struggle between the attractive force of gravity on large scales and the repulsive forces of pressure on atomic scales. At high density and temperature, such as exists in the cores of stars, nuclear-fusion processes act as an abundant heat source, serving to keep the pressure high and halt the star from falling in on itself. For most of the life of stars, like a hot cold war, a happy and stable stalemate between the two forces persists. However, all stars eventually run out of nuclear fuel, causing the inner portions to contract. For stars less massive than about 8 M at birth, the postnuclear burning phase contraction is halted by electron degeneracy pressure. An inert core of mass ~0.1–1 M—a white dwarf—remains essentially forever.² For stars more massive than about 8 M at birth, the crush of gravity is so intense that electron degeneracy pressure is not enough to halt contraction. Instead, there is an implosion of the core, and, by various proposed mechanisms, this causes a massive shockwave to propagate outward through the collapsing star, causing it to explode. Densities and temperatures are so high in the regions of the exploding star that new elements, some radioactive, are quickly formed. Energy from the radioactive decay of some elements is deposited into the exploding material, and a supernova is born.

The specific type of supernova created in exploding massive stars depends both on the composition and size of the star before it explodes and the circumburst medium into which the outflowing material will propagate. The least massive “core-collapse” supernova progenitors have large hydrogen envelopes before explosion and leave behind an NS remnant. Higher-mass progenitors (> 25 M at birth) probably also leave behind an NS but, prior to explosion, have lost their hydrogen (a Type Ib SN) and, in the more massive cases, helium (a Type Ic SN). These hydrogen-stripped stars are called Wolf-Rayet stars, and there are many of them seen in the Milky Way. Understanding precisely how and why these stars explode is a continuing pursuit (outward-moving shocks from the creation of a newly formed NS and/or pressure from neutrinos are the main culprits). Also active is the art of modeling the diverse physical states in the supernova explosions themselves that give rise to the diversity of supernovae observed.

In 1993 Stan Woosley of the University of California, Santa Cruz, published a seminal paper highlighting the properties of a massive star (15 M before exploding) that undergoes core collapse.³ But instead of the creation of an NS at the center, he posited that a black hole could be formed under certain conditions. In this case, the usual mechanisms for exploding the star would not be strong enough to do so. This is why Woosley referred to this event as a “failed” Type Ib supernova. Mass from the star would instead flow inward to the newly formed black hole. If the star was rotating, then matter would fall freely along the rotation axis toward the black hole on timescales of seconds; however, along the equatorial plane, matter would be held up in an accretion disk by a centripetal force. This model, then, sets the stage for the central-engine scenario (§2.3) thought to power a long-duration GRB: a small region around a black hole where gravitational potential energy from an appreciable amount of mass (> 0.1 M) could be tapped over the course of seconds to tens of seconds.

Aside from the basics of the central engine, recall that relativistic outflow is a crucial requirement of the fireball (and afterglow) model. But too much matter entrained in the outflow—a so-called dirty fireball—would bog down the acceleration and stop the outflow from reaching relativistic speeds. To this end, it would seem that just about the worst place in the Universe to launch a pristine fireball would be at the center of a collapsing star. Indeed, the very crucial question of whether Woosley’s model could also allow for relativistic outflow would not be answered for several more years. In 1999, Woosley and his student Andrew MacFadyen presented the results of a detailed numerical simulation that followed the collapse to a black hole, the dynamics of the accretion disk, and the fate of the energy deposited in the region around the polar axis.⁴ They, and subsequently others, showed that the polar regions^* become sufficiently evacuated of matter (after a few seconds) to allow the deposited energy to begin to propagate outward into those regions. The infalling matter outside the polar regions creates an effective barrier that funnels the outflowing material into a jet. This jet, which will last as long as matter continues to feed the central black hole, in turn pushes aside any remaining matter. By the time the jets emerge, simulations have shown that they can reach the requisite Lorentz factors (Γ ~ 100; see footnote on page 56) and that they are highly collimated (opening angles of a few degrees, as inferred from afterglow modeling; see §3.2).

The launching of the jet has consequences for the star not entirely anticipated in the original 1993 picture. First, the jet will interact with the matter in the jet boundary, creating shocks within the outflow. Therefore, even if the energy input is constant near the black hole (which it should not be, in general), these shocks lead to an unsteady relativistic “wind,” also a basic requirement of the GRB production mechanism in the internal-shock scenario. Second, the jet is thought to carry enough energy that, even if a small fraction of it couples to the matter in the jet boundary, it will be enough to explode the star. During this disruption, explosive nucleosynthesis (particularly of the element ⁵⁶Ni) can occur in the dense regions of the jet boundaries and also (more likely) from a wind produced in the outskirts of the accretion disk. Explosion of a star and nucleosynthesis of⁵⁶Ni are the ingredients of a supernova, so, contrary to the original “failed Ib supernova” scenario, it became clear on theoretical grounds that the more sophisticated collapsar model indeed predicted both a GRB (from the polar jets) and a supernova.

5.1.2 Connection to Supernovae

The particular flavor of supernova that should accompany a long-duration GRB in the collapsar progenitor model is somewhat of a postdiction: by 1999, GRB 980425 had already been phenomenologically connected to a strange supernova SN 1998bw. That supernova was very bright, comparable to some of the brightest supernovae ever observed at the time. This indicated a large amount of ⁵⁶Ni synthesis: at least 0.5 M in just that element alone. The SN spectrum also showed broad undulating features indicative of high-velocity outflow (owing to large Doppler shifts in the ejecta), at least several tens of thousands of kilometers per second (an order of magnitude faster than most other SNe). Importantly, the spectrum showed no strong evidence for the presence of hydrogen or helium in the outflowing material. These observations led to the observational classification of SN 1998bw as a broad-lined Type Ic (Ic-BL) supernova. The radio light curve was unlike that seen in any SN to date and was modeled as having arisen from mildly relativistic outflow in the outer layers of the ejecta.⁵

Theoretical models explaining individual SNe derive from an admixture of basic physical equations, intuition, and computationally intensive supercomputer modeling of the complex interplay between light, gas, and various heating sources. For SN 1998bw, a model⁶ developed to explain the brightness, apparent velocity, elemental abundances, and time evolution suggested that a hydrogenand helium-stripped star (a so-called carbon-oxygen star) of mass ~14 M produced a whopping 0.7 M of⁵⁶Ni; the star would have been ~40 M at birth. Since the total energy coupled to the ejecta was E_SN ≈ 10⁵² ergs (roughly a factor of ten larger than the typical energies of SNe), SN 1998bw was advanced as a new class of exploding star called “hypernovae,” connoting especially energetic supernovae.

The progenitor of SN 1998bw appeared to be very similar to what Woosley had envisioned in 1993. But the substantive difference in the outcome (aside from the fact that an SN did indeed occur) was that a very weak GRB was produced. Indeed, GRB 980425 was at least three orders of magnitude less energetic than the “typical” cosmological long-duration GRB with known redshift. The extreme nature of both the GRB and SN threw many for a loop, but the collapsar proponents quickly saw this odd couple as possible bookends of a more typical population. In this case, much of the energy reservoir available during core collapse was coupled to the protons and neutrons in the star, giving rise to the observed fast-moving (but mostly nonrelativistic) ejecta in the SN; a minority fraction of the energy was coupled to the mildly relativistic material (as manifested by the radio light curve) and the highly relativistic material (as manifested by the GRB). But in other cases, presumably with cosmological long-duration GRBs, there could be a different partitioning of the energy. By 1999, Woosley and MacFadyen suggested that all long-duration GRBs should be accompanied by supernovae with qualities similar to SN 1998bw.

The search was on for SNe associated with cosmological GRBs. I presented evidence⁷ for an SN-like bump in the light curve following the event GRB 980326. While the timescales were similar to SN 1998bw and the colors appeared consistent, no easily recognizable signature of an SN was seen in the low-quality spectrum taken around the time of maximum light of the bump. Likewise, since no redshift was known for the GRB or host galaxy, the evidence for an SN was far from ironclad. Shortly after publishing our results, a credible SN-like bump was found in the reanalyzed afterglow light curves of GRB 970228. Over the next several years, evidence of bumps was found buried in many GRB afterglows, with events like GRB 011121 showing strong photometric evidence for a 1998bw-like supernova. Some events appeared even brighter than SN 1998bw and some fainter (by as much as an order of magnitude in peak flux).

The smoking gun that would convince most remaining skeptics of the connection between GRBs and the death of massive stars came in 2003. GRB 030329 was an incredibly bright event with a bright afterglow, originating from one of the lowest redshifts to date (z = 0.16). By most metrics, it was an ordinary long burst of duration T₉₀ ≈ 23 seconds.⁸ Yet, as the afterglow faded, a set of broad spectroscopic features—subtle at first, then overwhelming—developed. Removing a model for an afterglow component from spectra taken eight days after the event, Tom Matheson and Krzysztof Stanek (both then at Harvard) and collaborators found striking evidence that these features resembled the high-velocity features of SN 1998bw (figure 5.1). Other groups confirmed the remarkable finding: another broad-lined Type Ic supernova was associated with a GRB, but this time it resembled the other cosmological events that dominate the observed rate of GRBs.⁹

By July 2010, three additional events (from low redshift, when the SN would be expected to be detectable) showed strong spectroscopic evidence for an SN component, and many more had strong photometric evidence for such. However, most Swift GRBs do not have associated SNe, and the explanation for this is straightforward: even if SNe accompany all long-duration GRBs, these types of SNe appear exceedingly faint at optical wavebands when they occur beyond redshift of z ≈ 1. This is due to the suppression of the blue and ultraviolet portions of their spectra (redshifted into the optical bandpass for z > 1) due to heavy metals (e.g., iron) in the ejecta. Moreover, since GRB-SNe are about as bright as the galaxies they occur in, we contend with both afterglow light and host galaxy light in searching for the SNe.

Figure 5.1. Comparison of the spectral evolution SN 1998bw and SN 2003dh. Solid lines show SN 2003dh, associated with GRB 030329, and dotted lines show SN 1998bw taken at similar times since the GRB trigger. Times after the GRB are given in the restframe, correcting for cosmological-time dilation. The remarkable similarity between the two supernovae showed that at least some cosmological GRBs arise from the same type of progenitor as the oddball GRB 980425. Adapted from J. Hjorth et al., Nature 423, 847 (2003).

Figure 5.2. The optical observations of long-duration GRBs without associated SNe to deep levels. The gray points are the detected points in the afterglow light curve; the points with a downward-facing arrow are nondetections (“upper limits”) at the position of the afterglow, meaning that any source at that position would need to be fainter. Overplotted are three GRBs as they would have appeared at the redshift of the GRB (SN 2006aj, associated with GRB 060218; SN 1998bw, associated with GRB 980425; SN 2002ap, a faint broad-lined Ic SN). All SNe like these fiducial events are ruled out by the data. Adapted from J. P. U. Fynbo et al., Nature 444, 1047 (2006).

5.1.3 Long-Duration Events without Supernovae

This notion that all long-duration bursts could be associted with Type Ic supernovae persisted for years. But in the summer of 2006, two GRBs were discovered by Swift at sufficiently low redshift that the SNe, if they were of similar brightness to SN 1998bw, should have been easily detected. Instead, despite concerted efforts to find the SN peaks in GRB 060505 and GRB 060614, no such signatures were found in deep images (see figure 5.2). Using some prior inference on the lack of dust obscuration from afterglow fitting, a Danish-led collaboration (which included me) suggested that any contemporaneous SN must have been fainter at its peak brightness than about −14 magnitude; this magnitude translates into a luminosity less than about one-tenth to one-hundredth the peak luminosity typical of most SNe, essentially ruling out all known types of SNe.¹⁰ Some suggested that the events were simply “short-burst impostors” and, therefore, on physical grounds that we should not have expected an SN (see §5.2.2). Others suggested that the GRB might have originated from much higher redshift than the putative host galaxies, appealing to a chance coincidence of the alignment of those distant GRBs with foreground galaxies.¹¹

My take, which is by no means universally accepted, is that instead of invoking the need for “long-duration short bursts” (syntactically discordant, at the very minimum) or multiple chance superpositions with foreground galaxies, the natural explanation for such events¹² would be that they did indeed arise from the collapse of massive stars, but instead of producing the necessary conditions to blow up the star and produce a supernova, only a GRB was produced. In a sense, I suggest that Woosley’s original “failed supernova” hypothesis may have been realized^*with these events.

All theoretical studies of collapsars have noted the importance of the angular momentum of the precollapse star. With too little momentum, an accretion disk feeding the central black hole cannot be supported long enough to power a jet through the collapsing star. Spinning too fast, the accretion disk is insufficient to feed the black hole and power the GRB. Since the total angular momentum before collapse is a complex function of many competing physical mechanisms, it is essentially a free parameter in the prescription for events that follow core collapse. One possible explanation for the origin of GRB 060505 and GRB 060614, and more generally in the diversity of GRB-SN brightnesses and light curves, is simply that the differences in the angular momentum affect the total ⁵⁶Ni production.

To be sure, rotation of long-GRB progenitors has yet definitively to be established observationally as a driver of GRB-SN diversity. Indeed, on the theoretical front, the amount of rotation required for the nominal collapsar model to produce a GRB and an SN may be difficult to arrange in practice. In envisioning the full life cycle of the massive star that leads to the pre-core-collapse progenitor, we require that a significant amount of mass be lost from birth to death (perhaps as much as 25–40 M). Most stellar mass loss happens in stellar winds, and Wolf-Rayet stars (in particular) lose copious amounts of mass, at a rate of about 1 M every ten thousand years! But mass ejected to large distances is very effective at slowing up what remains of the star.¹³ Therefore, “normal” channels of mass loss for massive stars might lead to evolved stars that are spinning too slowly and thus are incapable of making GRBs. To this end, some have advocated that a productive mass-loss channel involves the interaction of the progenitor with a binary companion star. In this case mass can be flung from the progenitor, but the angular momentum can remain relatively high (where the spin rate would now be dictated by the orbital period of the binary system). How angular momentum is maintained in GRB progenitors is an open question. It is also an open question how important binary-star interactions are for GRB/SN production.

5.2 Mergers of Compact Objects

Historically, there were many attractive features of merging compact stars as GRB progenitors.¹⁴ First, we have known for decades that such systems exist in the Milky Way and that, in the case of some observed NS–NS binaries, the merger will happen in less than the current age of the Universe. Second, the modeled (albeit uncertain) rates of coalescence of such systems approximately match the inferred rate of GRBs. Third, the variability timescales in GRBs are naturally explained by the light crossing time of the two objects at the time of merger. Fourth, the total mass-energy available to be accreted into a newly formed black hole is larger than that required to produce the total energy released in the GRB event itself plus the afterglow. Indeed, merging NSs were the leading explanation for cosmological GRBs until a pernicious Mother Nature decided to reveal long-duration GRBs as arising from an entirely different beast.

By 2004, with the localization of short bursts missing from the observational trophy chest, degenerate merger events remained a viable progenitor scenario. Indeed, one of the “hopes” of Swift—vindicated soon after launch by observation—would be that regular Swift localizations of short bursts would help reveal their nature. Though we are not nearly as certain as the progenitors of SGRs or long-duration bursts, observational evidence is mounting that this subclass may indeed be due to such progenitors.

5.2.1 Models

The stellar graveyard consists of just a few classes of remains or compact “remnants”: white dwarfs, neutron stars, and black holes.^15* Getting to the central engine configurations in models involving such objects is tantamount to envisioning a scenario where these objects play primary roles both in rapidly supplying the necessary mass and in serving to channel that mass into the energy that powers a GRB (see figure 1.4 for an illustration). Since the pathways to creation of these remnants are numerous, the degenerate progenitor models must also chart the long-term evolution of the objects that ultimately lead to the central engine.

The progenitor scenario most discussed in the literature involves the coalescence or merger of two neutron stars. One channel, probably the dominant mechanism, for the creation of this merger event starts the story with a binary system of two newly formed high-mass stars. Unlike the low rate of human twin birth, it is very common for stars to be formed in binary systems. Since stars born in binaries tend to have similar masses, this also implies that the stars will undergo a similar arc of development. If one star undergoes a core-collapse SN, it will leave behind a neutron star. We believe that NSs born in SN explosions undergo a “kick” during their creation that sends them flying off in some random direction relative to where the presupernova star was heading. In some cases, the NS will have a large-enough velocity simply to bid its sibling adieu and disrupt the binary. But in many cases, the velocity of the NS will not exceed the critical velocity needed to escape the gravitational pull of the system. In this case, the binary system will persist. What happens then depends on the state of the other star and the details of the binary orbit.¹⁶ In some cases, the companion star may also explode as a core-collapse supernova, leaving behind an NS. In most cases, the binary system will disrupt but with the conditions of the kick just right, the NS–NS binary will survive.

At this point, the neutron stars orbit around each other in a highly stable configuration, interacting only through their mutual gravitational attraction. Isaac Newton would have posited that these two objects could dance the “Do-Si-Do” for eternity. But General Relativity introduces a certain fatigue to the system: over time, the binary orbit begins to decay as energy is radiated through gravitational waves. Eventually the binary decays to the point where the NSs collide and a GRB central engine is born. Much of their mass goes into making a spinning black hole with mass 2–3 M. Some of the mass forms an accretion disk (lasting < 1 sec) that feeds the black hole, and some of the mass (0.1–0.2 M) is flung out.

How often this NS–NS merger scenario occurs is not precisely known; but, given that we know of ∼6 NS–NS systems in the Milky Way that appear to have formed this way, it seems rather clear that it does happen. Since we also have observational evidence of the binary fraction of massive stars and of the distribution of kick velocities that NSs receive on birth (during supernovae), attempts have been made to predict the rate of NS–NS mergers using such observational ingredients. Current estimates place this rate (in the Milky Way) to be between one event per million years to one event per hundred thousand years.¹⁷ Other “channels” for NS–NS production may be important in the overall rate of coalescence. For instance, in some sort of cosmic square dance, neutron stars populating dense regions of stars may actually exchange binary partners. Globular clusters—conglomerations of millions of stars and thousands of neutron stars—serve as the main dance floor for NS–NS binaries formed this way.

There is a rich literature describing other viable compact objects as GRB progenitors. The double supernova channel, for example, may also produce neutron star–black hole (NS–BH) binaries that will also coalescence after orbital decay due to gravitational-wave emission. In this case, one of the two initial stars may be massive enough to produce a BH directly when it explodes as a supernova.¹⁸ Or the first neutron star could accrete enough matter from the second star to collapse to a BH before the second star explodes as a supernova.

5.2.2 A priori expectations

Since the viable compact merger scenarios essentially lead to the same (short-lived) central engine, there is no obvious observational path to distinguishing between the progenitor possibilities from the GRBs or afterglows. However, there are several sets of observations that should winnow down the choices among the degenerate-binary merger scenarios. Most striking are the different a priori expectations of degenerate-binary merger scenarios relative to the expectations from collapsars. First and foremost, none of the compact progenitor scenarios should lead to a bright Type Ic SN, so the presence of a contemporaneous SN would rule out such scenarios for that GRB.^* We can view the absence of an SN as a necessary but not sufficient condition for confirming such progenitors.

Host Galaxies: For some of the possible progenitors, particularly NS–NS mergers, the broad distribution of merger times since birth implies by the time of the GRB, that the general population of stars in the host galaxy will have evolved appreciably. In particular, since star formation appears to happen in “bursts” of time lasting tens-of-million to hundreds-of-million years, the star-forming episode that gave rise to the progenitors will have subsided, and the stars in the host will have aged significantly. Thus, GRBs from NS–NS mergers are expected to be associated with more of a diverse population of host galaxies than those hosting collapsars. This diversity would obviously be reflected in average stellar age and, most manifestly, in galaxy color.

Locations: Since NS–NS progenitors also experience a systemic velocity “kick” during formation of upward of several hundred kilometers per second (v_kick ≈300 km/s), we might expect them to travel far from their birth site in the time τ (∼10⁸⁻⁹ yr) until merger. This is quite different from collapsars that should not travel far before core collapse. In linear distance, this is

l ≈ ν_kick × τ = 30 to 300 kpc.    (5.1)

This distance traveled can be much larger than the size of the host galaxy; in some cases v_kick may be larger than the velocity required to escape the gravitational tug of the host galaxy.¹⁹ Therefore, we naturally expect NS–NS binaries to be loosely distributed around their host galaxies and, in some fraction of the cases, may be far from the light of their true host. NS–NS binaries literally evaporate from galaxies. Other variants, like NS–BH binaries, may receive less of a kick during formation and thus, like high-school sweethearts who get married quickly and never leave their hometown, will die near to their birth site.²⁰ Likewise, short-lived progenitors, even if they travel fast, might not make it far from their birth site until the coalescence. Given these considerations, we expect that some GRBs from compact binary progenitors should not be significantly offset from the light of their host galaxy. However, since we do not know from ab initio models precisely what the merger time nor the kick distributions will be, we cannot predict precisely what the “offset distribution” of events from such progenitors should be. But compared to collapsars, most models do generally predict a wider separation of such progenitors from galaxy hosts.^*

Short-lived Transients: If some events do indeed occur far from the stellar birth site, they might also occur far from the regions where there is an appreciable density of ambient gas. Under the assumption that external shocks dominate the emission of afterglows, this implies that such events should have a relatively weak afterglow. Indeed, in the simplest models, for a fixed energy in the external shock, the peak brightness of the afterglow at a fixed wavelength should scale as the square root of the ambient density. Given that the density in the intergalactic medium can be millions of times less than the density in the interstellar medium of galaxies, we might expect that afterglows from such progenitors could be upward of a thousand times fainter than afterglows from events produced within galaxies.²¹

In most merger models, the mass that does not flow into the newly formed black hole may instead get ejected from the system. In the case of a shredded NS, we expect the flung-off mass to undergo a rapid burst of nucleosynthesis, given that it is highly compressed and rich in neutrons. The decay of these rapidly produced radioactive isotopes of heavy elements serves as a powerful heating source for the outflowing material. Just as in a supernova, the heat generated by radioactive decay is trapped by the dense ejecta until the material expands to a much larger radius. It takes a few weeks until an ordinary supernova reaches maximum brightness, but these events, given that there is much less mass involved, will reach peak brightness about one day after the coalescence. The nominal expectations for such a short-lived radioactive-powered transient was worked out by Bohdan Paczyski and Li-Xin Li²² in 1998. More detailed modeling of the nucleosynthetic processes that drive these “Li-Paczyski mini-supernovae” has been carried out recently in much more detail.²³ The conclusion is that there should indeed be a short event reaching peak visual magnitude of −14, about one hundred times fainter than SN 1998bw. Saying “Li-Paczyski mini-supernovae” is a bit of a mouthful (say it ten times, fast!), so several alternative names have been proffered. Since the peak of the event is expected to be approximately one thousand times brighter than a nova event, I prefer the term “kilonova,” advanced by my colleague Eliot Quataert at the University of California, Berkeley.

Gravitational Waves: The last on our list of a priori expectations from compact mergers is gravitational waves (GWs). Just as accelerating charged particles produce light, GWs are produced by accelerating mass (in certain geometric configurations). GWs are ripples that deform space and time and travel at the speed of light. The slow procession to coalescence of widely separated compact binaries occurs because gravitational waves gradually carry away energy from the system. But in the last second before coalescence, the rate of acceleration increases dramatically, and a tremendous burst of gravitational waves emerges, hastening the coalescence. For a distant observer, the GRB would be then be accompanied by a (near) simultaneous gravitational-wave event. Given the masses of the coalescing objects and the distance to the event, the intensity and temporal development of the gravitational-wave signal can be calculated reasonably precisely. As GRB-SNe were for collapsars, the discovery of a GRB and concurrent gravitational-wave event is considered the observational smoking gun that would unequivocally reveal the precise nature of the progenitor of that event (i.e., a compact binary merger). We discuss this possibility further in §6.5.

5.2.3 Confrontation with Observation

As a group, observations of long-duration GRBs do not appear to reflect these a priori expectations of compact-binary merger progenitors. But, as shown throughout our discussion of the environments of GRBs (chapter 4), the types of galaxies that are associated with short-duration GRBs and the locations of the events around those galaxies appear to reflect some of the compact-binary merger-model expectations. For example, whereas no long-duration GRBs appear to be associated with galaxies dominated by old stars, 10–20 percent of short-duration GRBs are associated with old-stellar-population galaxies (such as ellipticals). Similarly, as shown in figure 5.3, the locations of short-duration bursts appear to be more widely distributed around their associated hosts. However, unlike with long-duration GRBs, the association of a given short burst with a given host galaxy tends to be more uncertain. The reason for this is partly due to the observational bias of poor afterglow localizations²⁴ and partly, it seems, intrinsic. Indeed, viewing the question from the model perspective, if we posit that some fraction of merger events should happen far from their host galaxies (§5.2.2), we expect that a fraction of events should have ambiguous (or incorrect) host galaxy associations and thus incorrect offset measurements. This is like letting a bunch of cats loose in a crowd and, after some time has passed, saying that the owner of each cat is the person it happens to be nearest to.²⁵ The short GRB 060502b is emblematic of the difficulty of host association.²⁶ The event was 90 milliseconds (0.09 sec) in duration, had an X-ray localization of 4.4 arcseconds, and had no detected optical afterglow. In the error region there are a few faint (presumably distant) galaxies that could be the host. But 17 arcseconds away (i.e., relatively nearby) there is a bright-red galaxy made up of lots of old stars. At the distance of that galaxy, the angular separation on the sky corresponds to a physical separation of 73 ± 19 kpc in projection. This is a distance easily obtainable for NS–NS binary coalescence (equation 5.1) and comparable to that inferred for GRB 050509b from a very similar elliptical galaxy. Without a redshift of the event (which could have easily disproven the elliptical galaxy association), we are left with an almost unpalatable uncertainty about the host and location from it. Everything we assert in astrophysics is probabilistic in nature, but connecting short-duration bursts to galaxies is a particularly uncomfortable exercise.

Figure 5.3. The cumulative location distribution of short- and long-duration GRBs around their putative host galaxies. The twenty (pre-Swift) long-duration GRBs (light gray line; from J. S. Bloom, S. R. Kulkarni, and S. G. Djorgovski, AJ 123, 1111 [2000]) all lie within about 10 kpc from the apparent centers of their hosts. In contrast, short-duration bursts (even selecting those with the most clear associations to individual galaxies) appear to reside at larger distances from their hosts. Depicted in smooth curves are three different ab initio models for the location distribution of merging NS–NS binaries, showing reasonable agreement with the observed short-burst population. Adapted from W. Fong, E. Berger, and D. B. Fox, ApJ 708, 9 (2010).

To date no short-duration bursts have had detected SN emission, but only a handful of short bursts (with reasonably secure redshift) have had deep-enough follow-up to rule out SNe beyond a reasonable doubt. Of course, the absence of SNe in these events is certainly tantalizing but by no means definitive. Likewise, kilonovae—one rather clear predicted signature of compact mergers—have not been definitively identified. There was a curious short-duration GRB (080503) that showed evidence for a one-day timescale optical event (see figure 5.4) that nominally fit kilonova models.²⁷ However, since no redshift of the source could be established, the intrinsic peak brightness could not be calculated. More importantly, there was detected X-ray emission that appeared to track the optical event at around one day after the GRB, something not predicted in kilonova models. Our conclusion from this event was that a kilonova was indeed a possible explanation but not the only viable model. Taking the short-burst population as a whole, no other searches reported to date have been sensitive enough to detect such a faint event, so it is difficult to draw any strong conclusions. Future searches for kilonovae (both triggered by GRBs and perhaps found in untriggered optical surveys) should prove most illuminating.

A concurrent gravitational-wave event—the cleanest and most distinct prediction from compact mergers—has not be confirmed observationally. But since current gravitational-wave detectors are not sensitive enough to NS–NS events beyond the distance of the very nearest galaxies (such as Andromeda), the rate of NS–NS (or NS–BH) mergers is thought to be simply too low to detect even one event per decade with the current setups. The good news is that more sensitive searches are on the horizon (see §6.5) and that we may indeed detect definitive gravitational-wave signatures of short-duration GRB progenitors as early as ∼2015.

Figure 5.4. Kilonova model fits to the curious late-time emission from short-duration GRB 080503. The characteristic timescale of one day for the time to peak is related to the velocity of the outflow and the mass of the ejected material from an NS–NS merger. The luminosity/magnitude is related to the total mass of the ejected material and the radioactive heating rate (a quantity difficult to determine theoretically). Unfortunately, the redshift of GRB 080503 could not be inferred from nearby associations, so two models are plotted (“low z”: d = 124 Mpc; “high z”: d = 2.7 Gpc). The basic characteristics of the expected kilonova light curve are well matched, although in detail there are discrepancies (not, at this point, all that worrying given the large uncertainties in the models). Adapted from D. A. Perley et al., ApJ 696, 1871 (2009).

The evidence, albeit circumstantial, continues to mount that at least some short-duration GRBs have different progenitors than do long-duration GRBs, and it seems clear that these progenitors are associated with an older stellar population than collapsars are. There is no unambiguous evidence to date to implicate NS–NS or NS–BH binaries as short-duration GRB progenitors, but the general consensus by 2010 was that this class of progenitors seem to be the most likely.²⁸

While kilonovae and afterglows are a natural consequence of the properties of the outflow, in the NS–NS or NS–BH scenarios, all known timescales for the mass flow in the merger configuration are very short: there are no natural timescales for the inflowing matter of more than a few seconds. The a priori expectations, then, are that we should see no significant emission on tens-to-hundreds of second timescales and beyond. Yet long tails of emission extending to hundreds of seconds after the GRB are often observed in what otherwise appear (in the initial spike) to be ordinary short bursts.²⁹ Moreover, the late-time X-ray flare in the short GRB 050709 (§3.1.1) finds no natural explanation in the NS–NS or NS–BH models. To be sure, many short bursts show neither long-lived tails following the prompt emission nor X-ray flares. On the other hand, those short bursts that do exhibit such behavior remind us to keep an open mind about the progenitors of all short-duration GRBs.

5.3 Extragalactic Magnetars

The bright flare from SGR 0525−66 during the March 5th event was outclassed by even brighter flares from other SGRs. The brightest known giant flare from an SGR—an event from SGR 1806−20—occurred on December 27, 2004. It had a total energy release of few×10⁴⁶ erg, a factor of more than one hundred brighter than the March 5th event (but still a factor of one hundred less energetic than the underluminous GRB 980425). At the luminosity of the very brightest giant flares from magnetars, we should be able to see such flares from nearby galaxies. But at such large distances we would only see the initial hard-spectrum spike of emission and miss, below the detection threshold, the characteristic periodic ringdown seen in more nearby Galactic SGRs flare events. That is, giant flares from SGRs in other galaxies would appear like a lot like short-hard GRBs.

Even with a few nearby examples, the typical energies and the volumetric rate of giant flares from SGRs are difficult to pin down precisely. The events are rare (probably no more than once per 5000 years per source), and the distances to most SGRs is uncertain (leading to an uncertainty in the intrinsic brightness of detected events). By correlating the BATSE and IPN samples with nearby galaxies, it appears that a few percent (and no more than 10 percent) of the short events observed with BATSE may have occurred from extragalactic magnetars. No Swift event has been convincingly associated with a “local universe” galaxy (< 200 Mpc)³⁰ but, during the Swift era, two IPN short-hard bursts were localized close to very nearby neighbors of the Milky Way: GRB 051103 (near galaxy M81 at a distance of 3.5 Mpc) and GRB 070201A (near Andromeda, at a distance of 770 kpc). Neither of these events showed any evidence for periodicity in the light curves. On statistical grounds, it is likely that at least one of these events did indeed originate from its putative host.

Accepting that at least one event originated from nearby, what was the progenitor? Without a supernova detected to very deep limits, we can certainly say they were not due to an ordinary collapsar. By great fortune, the Laser Interferometer Gravitational-wave Observatory (LIGO) project was collecting science data at the time of GRB 070201A. The LIGO team was able to put very useful limits on the lack of gravitational waves from the direction of Andromeda around the time of the GRB. The LIGO data were enough to rule out NS–NS and NS–BH mergers to a high degree of confidence up to the distance of Andromeda. This leaves extragalactic magnetars as the most likely suspect. The implied energy release at the distance of Andromeda was indeed close to that inferred from the December 27th event from SGR 1806−20, providing some confirming evidence. However, there was no obvious supernova remnant in the IPN error region of GRB 070201A. This suggests either that the NS traveled far from its birth remnant or that the magnetar was produced by other means.³¹

5.4 Classification Challenges

Nature has clearly found numerous ways to express herself on the canvas of high-energy photons.^* It is, in many respects, our calling to try to uncover the physics and origins of these different progenitors and the resulting burst mechanisms. But to understand the various progenitors in individual classes, we presuppose that we can indeed classify events physically. Instead, all we know for sure is what we observe—all we know we can do is classify phenomenologically. Physical classification is not just the ultimate goal: it has immediate consequences on the scientific process itself. In particular, resources for following up the bevy of well-localized GRBs are scarce, and it is clear that most of the juiciest science comes only when follow-up is robust. Given some foreknowledge of the likely physical classification (i.e., “this event might be very low redshift and thus allow us to observe the GRB-SN in great detail” or “this event might be very high redshift”), different researchers will respond with their resource allocations in different ways.

At first blush, it would seem that the two loci in the hardness–duration space provide a direct observational conduit to two distinct physical classes: long-soft bursts are collapsars, and short-hard bursts are due to, for instance, NS–NS mergers. Indeed, in recent years, calling an event a “short burst” (even if it has a long-duration tail of high-energy emission!) has become synonymous with “it originates from a compact binary merger” (and vice versa); likewise, calling an event a “long burst” is shorthand for saying it “originated in the core collapse of a massive star” (and vice versa). But in the presence of extragalactic magnetars “polluting” the short-hard burst locus, we see a clear example of where this one-to-one mapping between what is observed and what is inferred breaks down. In the long-soft burst locus, we also have possible evidence (in GRB 060505 and GRB 060614; §5.1.3) of pollution by a noncollapsar population. Other long-duration events also hint at a greater physical diversity. For instance, GRB 070610 (T₉₀ = 8.3 sec) was localized near the Galactic plane, and its afterglow flickered strangely for days. We do not yet know what the progenitor was, but it almost certainly originated from within the Milky Way and was unlikely to be a collapsar.³² Even if a given GRB could be somehow classified intrinsically by its high-energy emission alone, the range of redshifts at which GRBs originate induces some important observational blurring. For example, a short-hard burst originating from high redshift might appear long and soft. Likewise, a long-duration GRB with an initial spike of emission might only have the initial spike detected above the instrumental threshold, leading to a classification as a short-duration GRB.^*

Some have introduced new metrics on GRB light curves as a way to disambiguate between classes, such as focusing on the temporal and spectral properties of the initial pulse of the event. But clearly a GRB is so much more than the initial deluge of gamma-ray light: it is the afterglow, it is the supernova that may or may not follow, it is the emission in gravitational waves and neutrinos. Indeed, we have seen that the total energy release in the GRB itself (E_γ ≈ 10⁴⁸⁻⁵¹ erg) is, in the case of long-duration bursts, probably less than the energy associated with the GRB-SN (E_SN ∼ 10⁵² erg). Moreover, in many progenitor models, this release is almost certainly less than the energy radiated in the form of neutrinos and gravitational waves (which have not yet been detected). So trying to devise a physical classification for a GRB from gamma rays alone might be like classifying the character of a person by studying her nose.

We have been looking at a deep intermingling between what is observed and what is inferred about underlying causes. This healthy conflation lies at the heart of many other fields in astronomy (supernova and quasar classification spring to mind). Indeed, the challenges of classification, especially in the face of so much diversity, might be taken as an essential component of astrophysics, a manifestation of the inexorable connection between observation and theory.

*The polar axis is the direction that is perpendicular to the rotation of the accretion disk, probably similar to the overall polar axis of the progenitor precollapse. The two polar regions are the areas around the polar axis both above and below the central engine (think of two ice-cream cones attached at their tips at the central engine)

*When Albert Einstein looked at cosmological solutions to his General Relativistic field equations, the solutions he found told him that the Universe was either expanding or contracting, something that contradicted the observations of the day. So he added in a “cosmological constant” to create a steady universe. Soon after it was shown that the Universe was indeed expanding, Einstein is famously quoted as saying that the addition of that constant was the “biggest blunder” in his professional career. Six decades later, evidence for a nonzero cosmological constant was uncovered by detailed measurements of SNe (see §6.5). This essentially vindicated Einstein’s tweaked equation and added deep irony to his “biggest blunder” exclamation. With affinity and affection, I think it is apt to call this interpretation of supernova-less long-duration GRBs as “Woosley’s biggest blunder.”

^*If these end states of stars are arm-wrestling matches between gravity and opposing forces: white dwarfs (supported by “electron-degeneracy pressure”; §5.1.1) and neutron stars (in part supported by “neutron-degeneracy pressure”) represent an eternal stalemate of sorts. But with black holes, gravity has gone over the top to pin its opponents: all pressure support is insufficient to counteract the crush of gravity; therefore, all the mass of a black hole is concentrated at one point. It is thought that quantum mechanics plays a role in halting the concentration of all mass to an infinitesimally small region of space. So the mass in a BH may be concentrated in a finite volume.

^*Of course, as the discussion in §5.1.3 asserts, the absence of an SN does not rule out massive stars, nor does it require degenerate-binary merger progenitors.

^*Still, locations alone certainly cannot be the ultimate indicator of collapsar or compact merger progenitors: we know, for instance, that core-collapse supernovae can happen in small satellite regions of star formation far from galaxies. Likewise, some collapsars might be found far from the detectable light of distant galaxies.

^*This is even more true if we do not restrict ourselves to events that look similar to GRBs (i.e., those events populating the traditional T₉₀–E_peak plane). Among transient events, we know that terrestrial lightning produces bursts of gamma rays as do events near the surface of the Sun. Some massive black holes at the centers of galaxies produce episodes of high-energy gamma-ray emission, and so forth.

^*Of course, comparing events using restframe quantities relies on a redshift measurement, which is not always possible.