3.1 Prompt emission

During the final stages of the merger of a compact binary, the system is expected to launch a highly relativistic jet that interacts with itself and the surrounding medium (the fireball model for GRBs; e.g. Piran Reference Piran1999). Collisions of material moving at different velocities within the jet will lead to internal shocks, giving short-lived bursts of gamma-rays that we detect as the SGRB prompt emission. As the accretion timescale is expected to be < 2 s (Metzger, Piro, & Quataert Reference Metzger, Piro and Quataert2008), the GRBs associated with compact binary mergers are typically shorter in duration than those associated with core-collapse supernovae (explaining the observed distribution of GRBs; Kouveliotou et al. Reference Kouveliotou, Meegan, Fishman, Bhat, Briggs, Koshut, Paciesas and Pendleton1993). However, the division between these two populations is not easily identifiable from the prompt gamma-ray emission alone (e.g. Bromberg et al. Reference Bromberg, Nakar, Piran and Sari2013; Qin et al. Reference Qin2013).

A number of Fermi-LAT GRBs have shown > GeV emission (even at redshifts as distant as z ≈ 4). Two bursts were observed with gamma-ray photons reaching energies up to 94 GeV (GRB 130427A) and 62 GeV (GRB 131231A)—this supports the suggestion that the photon energies may extend higher than previously assumed (Bouvier et al. Reference Bouvier, Gilmore, Connaughton, Otte, Primack and Williams2011; Inoue et al. Reference Inoue, Granot and O’Brien2013). Significantly, these discoveries have not been limited to LGRBs, with SGRBs also showing high-energy photons and GeV emission often continuing for tens of seconds beyond the initial burst. The fact that Fermi-LAT discovered a photon of energy 31 GeV during the prompt phase of GRB 090510 (Ackermann et al. Reference Ackermann2010) is promising for co-ordinated observations between GW detectors and ground-based CTAs (Bartos et al. Reference Bartos2014) operating at > 10 GeV. Additionally, SGRBs with time-extended emission have recently been cited as promising targets for CTAs (Veres & Mészáros Reference Veres and Mészáros2014).

One exciting possibility is the observation of prompt optical flashes. So far, these emissions have only been observed in LGRBs (Racusin et al. Reference Racusin2008; Akerlof et al. Reference Akerlof1999; Vestrand et al. Reference Vestrand2014). An early optical emission correlated with the prompt gamma-rays could indicate a common origin related to the internal shocks (Vestrand et al. Reference Vestrand2005).

A number of studies have suggested that compact binary mergers could generate prompt coherent radio emission (e.g. Totani Reference Totani2013). Such mechanisms include excitation of the plasma surrounding a compact binary merger by GWs (Moortgat & Kuijpers Reference Moortgat, Kuijpers, Chen, Bloom, Madejski and Patrosian2005), from a dynamically generated magnetic field after the merger (Pshirkov & Postnov Reference Pshirkov and Postnov2010), or from the onset of the collision of the forward shock with the surrounding medium (Usov & Katz Reference Usov and Katz2000; Sagiv & Waxman Reference Sagiv and Waxman2002). However, the detectability of emission from these processes will be dependent upon the scattering by the surrounding environment (Macquart Reference Macquart2007). Nonetheless, these studies suggest compact binary mergers are an interesting contender for the progenitors of FRBs (Lorimer et al. Reference Lorimer, Bailes, McLaughlin, Narkevic and Crawford2007; Thornton et al. Reference Thornton2013), which are currently unknown.

3.2 Energy injection at late times

Plateaus and flares in X-ray light curves following GRBs are signatures of ongoing energy injection. This could be caused by late-time accretion onto a central BH (unlikely in the compact binary scenario; see discussion in Rowlinson et al. Reference Rowlinson, O’Brien, Metzger, Tanvir and Levan2013), or from ongoing energy injection from the spindown of a newly born NS. Indeed, recent studies (e.g. Giacomazzo & Perna Reference Giacomazzo and Perna2013; Zhang Reference Zhang2013; Lasky et al. Reference Lasky, Haskell, Ravi, Howell and Coward2014) have shown that the merger of two NSs could result in a supramassive NS; a star with a mass greater than the non-rotating maximum mass but supported from further collapse through rotation (Cook, Shapiro, & Teukolsky Reference Cook, Shapiro and Teukolsky1994).

Around 60% of X-ray afterglow light curves of SGRBs observed by the Swift satelliteFootnote ⁴ (Gehrels et al. Reference Gehrels2004) have shown plateaus lasting 100–10 000 s after the burst; these have been attributed to EM spin-down emissions from protomagnetars (Rowlinson et al. Reference Rowlinson2010, Reference Rowlinson, O’Brien, Metzger, Tanvir and Levan2013) formed via the merger of two NSs (Dai & Lu Reference Dai and Lu1998; Zhang & Mészáros Reference Zhang and Mészáros2001). Observations of the plateau phase can also be used to constrain the NS equation of state, with GW observations of the inspiral phase significantly aiding this endeavour (Lasky et al. Reference Lasky, Haskell, Ravi, Howell and Coward2014).

If the post-merger remnant is an NS, early optical afterglow as bright as 17th magnitude in R band (assuming a distance of ~ 300 Mpc; see Section 6.2) could be produced from dissipation of a wide-beamed protomagnetar wind Zhang (Reference Zhang2013). This magnetar wind could launch ejecta at relativistic speeds which would interact with the surrounding medium and produce a bright broadband afterglow from synchrotron radiation (Gao et al. Reference Gao, Ding, Wu, Zhang and Dai2013a).

GW emission may also accompany an afterglow plateau if a millisecond magnetar is born from the collision. Multiple mechanisms for generating such GWs exist in nascent NSs, including secular bar modes (e.g. Lai & Shapiro Reference Lai and Shapiro1995; Shibata & Karino Reference Shibata and Karino2004; Corsi & Mészáros Reference Corsi and Mészáros2009a), r-modes (Andersson Reference Andersson1998; Andersson & Kokkotas Reference Andersson and Kokkotas2001), and magnetic-field-induced stellar deformations (Cutler Reference Cutler2002; Haskell et al. Reference Haskell, Samuelsson, Glampedakis and Andersson2008; Dall’Osso et al. Reference Dall’Osso, Giacomazzo, Perna and Stella2015). Such emission could be observable by aLIGO out to ≲ 100 Mpc (Corsi & Mészáros Reference Corsi and Mészáros2009a; Fan et al. Reference Fan, Wu and Wei2013; Dall’Osso et al. Reference Dall’Osso, Giacomazzo, Perna and Stella2015). In fact, the X-ray light curve itself can be used to constrain the total GW emission from these systems (Lasky & Glampedakis, in preparation).

Some plateaus following SGRBs exhibit an extremely steep decay phase, commonly interpreted as the collapse of the nascent NS to a BH (Troja et al. Reference Troja2007; Lyons et al. Reference Lyons2010; Rowlinson et al. Reference Rowlinson2010, Reference Rowlinson, O’Brien, Metzger, Tanvir and Levan2013). Such collapse could potentially produce an FRB when the magnetic field lines snap as they cross the BH horizon (Falcke & Rezzolla Reference Falcke and Rezzolla2014; Zhang Reference Zhang2014), which is expected to occur ≲ 5 × 10⁴ s after the merger (Ravi & Lasky Reference Ravi and Lasky2014). A low-latency GW trigger could enable prompt follow-ups to test this connection (Chu et al. Reference Chu, Howell, Rowlinson, Gao, Zhang, Tingay, Boer and Wen2015).

3.3 Afterglow

As the relativistic jet propagates, it collides with the medium surrounding the progenitor resulting in a forward shock travelling into the surrounding medium, and a reverse shock propagating back up the jet (e.g. Sari, Piran, & Halpern Reference Sari, Piran and Halpern1999; Rees & Meszaros Reference Rees and Meszaros1992). These shock fronts produce multi-wavelength synchrotron emission, initially peaking in the X-ray and moving through the different wavelengths to radio as it fades. The typical afterglow of GRBs is attributed to the forward shock emission and the brightness of this afterglow is dependent upon a number of parameters, including the density of the surrounding medium. Therefore, in a low-density environment, the forward shock component is expected to be relatively faint.

The multi-wavelength afterglows of SGRBs have been observed and are typically fainter than those of LGRBs (Berger Reference Berger2007; Gehrels et al. Reference Gehrels2008; Nysewander, Fruchter, & Pe’er Reference Nysewander, Fruchter and Pe’er2009; Kann et al. Reference Kann2011). This is consistent with SGRBs being less energetic than LGRBs and with their locations in lower density environments. The reverse shock has also been observed for SGRB 051221A (e.g. Soderberg et al. Reference Soderberg2006).

3.4 Kilonova

A ‘kilonova’ is been predicted to form after the merger of two NSs. This faint optical transient is powered by the radioactive decay of the ejected neutron rich matter (Li & Paczyński Reference Li and Paczyński1998; Rosswog Reference Rosswog2005; Metzger et al. Reference Metzger2010) and could reach around 21–23 mag in the optical and 21–24 mag in the near-infrared (NIR) for a source at 200 Mpc (Tanaka & Hotokezaka Reference Tanaka and Hotokezaka2013). Recent optical and NIR follow-up observations of GRB 130603B have provided the most conclusive evidence to date of this scenario, reinforcing the theory that compact object mergers are the progenitors of SGRBs (Tanvir et al. Reference Tanvir, Levan, Fruchter, Hjorth, Hounsell, Wiersema and Tunnicliffe2013; Berger et al. Reference Berger, Fong and Chornock2013). These observations have added significantly to other observational evidence in support of this scenario (Berger et al. Reference Berger2005; Bloom et al. Reference Bloom2006; Berger Reference Berger2009). Coincident EM and GW observations could confirm that SGRBs are indeed the result of coalescing compact binaries.

An additional prompt EM emission related to the kilonova mechanism has also recently been suggested by Metzger et al. (Reference Metzger, Bauswein, Goriely and Kasen2015). This has been inspired by studies that suggest a small fraction of the ejected neutron-rich matter can expand rapidly enough to avoid r-process capture (Bauswein, Goriely, & Janka Reference Bauswein, Goriely and Janka2013). The suggestion is that β-decay from free neutrons in the outermost layers of this ejecta could power optical emission on a timescale of hours after the merger, peaking at around magnitude 22 in the U-band for a source. For a source at 200 Mpc, this signal would peak at around magnitude 22 in the U-band and would act as a precursor to a kilonova.

5.1 Instrument sensitivity

The output from a single GW detector consists of a time series data stream, s(t), composed of the detector response to a GW signal, h(t), and the detector noise n(t):

(1) $$\begin{equation} s(t) = h(t) + n(t). \end{equation}$$

In general, h(t) will be a linear combination of the two orthogonal transverse polarizations, $h_{+,\times }$, weighted by the dimensionless detector antenna pattern functions for the two polarizations F _{+, ×}:

(2) $$\begin{equation} h(t) = F_{+}(t,\theta ,\phi ,\psi )h_{+}(t) + F_{\times }(t,\theta , \phi ,\psi )h_{\times }(t) \,, \end{equation}$$

which describe the detector sensitivity to radiation of different polarisations, incident from different directions (Schutz & Tinto Reference Schutz and Tinto1987; Tinto Reference Tinto1987; Jaranowski, Krolak, & Schutz Reference Jaranowski, Krolak and Schutz1998). The angles, θ and ϕ, represent the direction to the source and ψ is the polarisation angle of the wave.

A GW detector can follow the phase of a GW signal, so the time series is generally represented in the frequency domain by the strain amplitude spectral density, $\tilde{h}(f)$. This quantity is defined through the power spectral density $S_{s}(f)=\tilde{s}^{*}(f)\tilde{s}(f)$, with $\tilde{s}(f)$ the Fourier transform of the time series. Similarly, one can define a signal power spectral density, S_h(f), and a noise power spectral density, S_n(f). The strain amplitude spectral density is given by

(3) $$\begin{equation} \tilde{h}(f) = \sqrt{S_{s}(f)}, \end{equation}$$

with dimensions of Hz^−1/2 (Thorne, Reference Thorne, Hawking and Israel1987). This quantity is often used in plots to display the sensitivity of GW interferometers.

5.2 GW detector networks

A single GW detector cannot determine the polarisation state or source direction of a transient signalFootnote ⁵. To obtain source localisation, a widely separated network of GW detectors is essential. Such a network may employ techniques such as coincidence analysis, in which individual events from different detectors are correlated in time (Arnaud et al., Reference Arnaud2002), or coherent analysis, in which synchronised detector outputs are merged before searching for a common pattern (Finn, Reference Finn2002). By effectively resolving the different times of arrival of GW events between members of a network, coherent network analysis enables a detector array to become an all-sky monitor with good angular resolution over all source directions.

Achieving good directional sensitivity is of paramount importance for GW/EM associations. For the sources considered in this review, directional sensitivity is determined through triangulation of arrival timesFootnote ⁶. To maximise the time delays, and hence improve directionality, it is advantageous that a network be as geographically widely separated as possible (Sathyaprakash, Reference Sathyaprakash2004) and as such, a number of detectors are planned to join the aLIGO/AdV network throughout the next decade.

The Japanese observatory KAGRAFootnote ⁷, should begin operations by around 2018–19 (Kuroda & the LCGT Collaboration, Reference Kuroda2010); at design sensitivity, this detector could improve the directional precision of a aLIGO/AdV network by a factor of 1.5–2 and the detection rate by a similar factor (Fairhurst, Reference Fairhurst2011; Chu et al., Reference Chu, Howell, Rowlinson, Gao, Zhang, Tingay, Boer and Wen2015). LIGO-India operating at aLIGO sensitivity will be added to the aLIGO/AdV network by 2022—by then, BNSs will be detectable out to ~ 200 Mpc and up to 400 events are possible per year (Abadie et al., Reference Abadie2010b). An Indian detector will improve the angular resolution sufficiently to increase the percentage of GW sources detected within 5 deg² from 3–7% to 17% (Aasi et al., Reference Aasi2013b).

It has long been recognised that a GW detector in Australia would add the longest baseline to the proposed advanced detector network (e.g. Cavalier et al., Reference Cavalier2006; Blair et al., Reference Blair2008; Wen & Chen, Reference Wen and Chen2010). For example, adding an Australian detector to an aLIGO/AdV three detector network can reduce the error in solid angle to tens of arcminutes for high signal-to-noise ratio (SNR) signals (Wen & Schutz, Reference Wen, Schutz, Blair, Howell, Ju and Zhao2012), dramatically improving the ability to localise GW sources for multi-wavelength follow-up observations. This scenario could be realised when third-generation observatories such as the ‘Einstein gravitational wave Telescope’ (ET)Footnote ⁸ become a reality in the next decade (Hild, Chelkowski, & Freise, Reference Hild, Chelkowski and Freise2008; Hild et al., Reference Hild, Chelkowski, Freise, Franc, Morgado, Flaminio and DeSalvo2010, Reference Hild2011). The optimal site for a detector in the southern hemisphere been shown to be Western Australia (Schutz, Reference Schutz2011), the current home to an 80-m baseline prototype GW detectorFootnote ⁹.

5.3 The GW false alarm rate

The false alarm rate (FAR) is the rate that false positives appear above a given SNR threshold, and is dependent on the number of glitches (non-stationary transients) in the GW data stream. It is a critical measure as it determines whether a candidate should be considered for follow-up. For well-modelled sources, the background of false alarms is at a level close to that of Gaussian noise. For unmodelled sources—typically short-duration transients—the data quality has a greater effect on detection confidence. One therefore sets the threshold high enough so that noise generated false alarms are negligible. Given that the probability, P(h)dh, of observing an event with an amplitude in the range h to h + dh is given by a Gaussian distribution of standard deviation σ, the probability of obtaining a FAR greater than a given threshold, ρ, is

(4) $$\begin{equation} P(h |h > z) = \frac{1}{\sqrt{2 \pi } \sigma } \int ^{\infty }_{\rho } \rm {exp}\left( \frac{-h^{2}}{2 \sigma ^2} \right) \rm {d}h \,. \end{equation}$$

To be 99% confident that a GW has been detected, one can set an SNR ~ 8 which is equivalent to a FAR of 1 in 100 years of observation (3 × 10⁻¹⁰ Hz). To see this, one can approximate number of noise instances during that period. If the detector output sampling rate is 1 kHz and the output is processed through ~ 10³ filters, in 100 years we get P(h|h > z) = (3 × 10¹⁵)⁻¹, yielding ρ ~ 8 which is our required SNR (see Sathyaprakash & Schutz, Reference Sathyaprakash and Schutz2009, for a detailed discussion of this argument). For a network of three equivalent detectors combined SNR, ρ_c is given as

(5) $$\begin{equation} \rho _{c} = \sqrt{\sum _{i}{\rho ^{2}_{i}}}, \end{equation}$$

where ρ_i represents the SNR in the ith detector (Cutler & Flanagan, Reference Cutler and Flanagan1994). This shows that for a network of three equivalent detectors, to dismiss false alarms at a level 3 × 10⁻¹⁰ Hz requires ρ_c ~ 12.

6.1 Detection of inspiralling compact objects

The expected GW signals from CBCs takes on the well-modelled chirp form shown in Figure 2. The figure shows how the signal increases in both amplitude and frequency towards merger; as it does so it sweeps across the sensitive bandwidth of advanced GW interferometric detectors.

Figure 2. Top: The predicted chirp waveform of a coalescing compact binary system 40 s before merger. As the signal increases in both amplitude and frequency towards merger, it will sweep across the sensitive bandwidth of advanced GW interferometric detectors. After the merger, the signal will show a ring down phase (not shown in this plot) which will take the form of an increasingly damped sinusoid. Bottom: The final 50 ms before merger.

For such well-modelled signals, the most efficient signal detection method to extract signals from noisy detector data is matched filtering, in which a template, representing the predicted waveform as a function of time is correlated with the output of a detector (Helstrom, Reference Helstrom1968). A matched signal will produce an output much greater than that expected for pure noise with an optimal SNR given as

(6) $$\begin{equation} \rho = 2\left[ \int _{0}^\infty \, \mathrm{d}f\frac{ \left|\tilde{h}(f)\right|^2 }{S_h(f)} \right]^{1/2}. \end{equation}$$

For well-modelled sources, matched filtering enhances the value of the signal by a factor $\sqrt{n}$, where n is the number of cycles used in the integration. As inspiralling systems approach merger, even though the rest frame GW amplitude will increase, the number of cycles in each frequency bin, n = f ²(df/dt)⁻¹, gets smaller; therefore, the detected signal will decrease. This means that for inspiralling systems, rather than solely base the predicted amplitude of the radiation as a true indicator of the detectability, we include a measure of the observed cycles. The value of n increases with the compactness of the system as it approaches merger and if observed from a frequency of 10 Hz until merger, could produce n ~ 10⁴ cycles—effectively improving the detectability by a factor of 100. However, to achieve such gains, a GW data-stream would have to be filtered by a large number of templates (of order ~ 10⁴$–$10⁵) in near real time—the significant challenges in both theoretical modelling and computational efficiency to achieve this cannot be underestimated.

One important aspect of well-modelled inspiralling systems is that a detection can be made tens of seconds before the merger if enough cycles can be detected to boost the SNR (Manzotti & Dietz, Reference Manzotti and Dietz2012; Cannon et al., Reference Cannon2012). Figure 2 illustrates this concept showing a chirp signal 40 s before the merger phase. This scenario could allow a low-latency alert to be sent out to EM facilities as near real-time as possible to catch a prompt EM signature; the combination of EM and GW data in this regime would provide valuable insight into the inner workings of such cataclysmic events.

It is also worth noting that GWs can provide an independent measure of luminosity distance, d _L (Schutz, Reference Schutz1986). During the inspiral phase, the GW strain, and the rate of change of GW frequency are given as

(7)$$\begin{equation} \begin{array}{c}h \propto \mathcal {M}_z^{5/3} f^{2/3}d_{{\rm L}} \\ \dot{f} \propto \mathcal {M}_z^{5/3}f^{11/3}, \end{array} \end{equation}$$

where $\mathcal {M}_z=\left(1+z\right)\mathcal {M}$ is the redshifted chirp mass, $\mathcal {M} = \left( m_{1} m_{2} \right)^{3/5}/( m_{1} + m_{2})^{1/5}$, and m ₁, m ₂ are the component masses of the binary. Therefore, if one can determine the redshift through, for example, a galaxy association, one can measure the redshift–luminosity distance relation independent of the cosmic distance ladder. A recent series of papers has reinvigorated this topic by introducing novel methods for breaking the redshift-chirp mass degeneracy with future GW observations (Messenger & Read, Reference Messenger and Read2012; Taylor, Gair, & Mandel, Reference Taylor, Gair and Mandel2012; Taylor & Gair, Reference Taylor and Gair2012; Nissanke et al., Reference Nissanke, Holz, Dalal, Hughes, Sievers and Hirata2013; Messenger et al., Reference Messenger, Takami, Gossan, Rezzolla and Sathyaprakash2014).

Although matched filtering is the optimal strategy for Gaussian, stationary noise, high-amplitude transients due to instrumental and environmental artefacts can render GW data to be non-stationary and non-Gaussian. Therefore, one must employ robust methods that can reject instrumental artefacts and retain the true GW events.

One such method is the χ² veto that is a powerful consistency test used to reject false alarms (Allen, Reference Allen2005). This method uses the fact that the quantity ρ is an integral over all frequencies and therefore not sensitive to the contributions from different frequency regions of the broadband signal. One can split the signal spectrum into n bins of equal SNR contribution, and draw a comparison with the expected value in each bin (based on the model template). A true GW event will have power accumulated approximately equally in each of n bins; a noise glitch will have power unevenly distributed and will yield a large χ² value.

6.2 The detection range and rates of coalescing compact objects

In the GW domain, detector sensitivity is generally based on the detection range of BNSs—the most likely events for detection. The inspiral horizon distance, D _H, is the distance to which an optimally orientated and located equal mass binary can be detected with a SNR equal to 8. For a system with reduced mass, $\mathcal {\mu } = \left( m_{1} m_{2} \right)/( m_{1} + m_{2})$, this distance is approximated as (Singer et al., Reference Singer2014)

(8) $$\begin{equation} D_{\mathrm{H}} = \frac{G^{5/6}M^{1/3}\mu ^{1/2}}{c^{3/2}\pi ^{2/3}\rho }\sqrt{\frac{5}{6} \int ^{ f_{{\rm U}} }_{ f_{{\rm L}} } \frac{ f^{-7/3} }{S_{n}(f) } } \rm {d}f \,, \end{equation}$$

where G is Newton’s gravitational constant, c is the speed of light, M = m ₁ + m ₂ is the total of the system masses, S_n(f) the power spectral density of the detectors noise curve, and f the signal frequency. The lower limiting frequency of the integral, f _L is equal to 10 Hz for aLIGO; the upper limiting frequency can be approximated by the last stable orbit of a Schwarzchild BH, $4400[\rm M_{\odot }/ (m_{1} + m_{2})]\,\rm {Hz}$.

To calculate approximate values of D _H, a simpler approximation is given by

(9) $$\begin{equation} D_{\mathrm{H}} = C(M) \left( \frac{M}{\rm M_{\odot }} \right)^{1/3} \left( \frac{\mu }{\rm M_{\odot }} \right)^{1/2} \left( \frac{1}{\rho } \right), \end{equation}$$

where C(M) gives the value of the integral over S_n(f) in Equation (8) for different M; these are calculated for different observing epochs using the sensitivity curves expected for early aLIGO configurationsFootnote ¹⁰. The values of C(M) can then be conveniently read off Figure 3 for the different observing runs of aLIGO/AdV. Tabulated values of C(M) are provided in Table A1.

Figure 3. To easily approximate the maximum detection ranges for different types of coalescing compact objects, values of C(M) given in Equation (9) are provided by the curves for different values of the combined masses. The curves represent the values of the integral in Equation (8) for the different values of upper frequency and for the different aLIGO observing scenarios as shown in Table 1.

An average range can be obtained by scaling D _H by a factor 2.26 (Singer et al., Reference Singer2014). This range assumes a uniform distribution of source sky locations and orientations. A standard figure of merit used by aLIGO/AdV is the SenseMon Range which is the average detectable range for two 1.4 M_⊙ NSs (Abadie et al., Reference Abadie2010a). An additional scaling is given through the association of a GRB with a face on merger which provides an on-source time in which to search for a GW event; this increases the sensitivity by a factor of 1.5 (and the corresponding rate of events by a factor 3; Schutz, Reference Schutz2011) in comparison with an all-sky/all-time search (Kochanek & Piran, Reference Kochanek and Piran1993); therefore, the average orientation average distance of 197 Mpc (see Table 1) becomes 300 Mpc. Thus, spatially and temporally coincident EM observations enable GW searches to dig deeper into the noise and therefore extend the detection horizon (Was et al., Reference Was, Sutton, Jones and Leonor2012).

Table 1. The expected observing scenarios for the aLIGO/AdV era based on Aasi et al. (Reference Aasi2013b). The available detectors are labelled: H: aLIGO-Handford; L: aLIGO-Livinstone; V: AdV. The aLIGO/AdV detectors will be at design sensitivity by 2019. The expected average ranges for NS/NS and NS/BH inspirals are given in Mpc as well as the horizon distances in parenthesis; these are calculated using Equation (8) along with the sensitivity noise curves for each of the different observing epochs given in https://dcc.ligo.org/LIGO-T1200307 and assuming masses of 1.4 M_⊙ and 10 M_⊙ for NSs and BHs, respectively. The detection rates are estimated using the calculated horizon distances along with Equation (19) of Kopparapu et al. (Reference Kopparapu, Hanna, Kalogera, O’Shaughnessy, González, Brady and Fairhurst2008) which is valid for horizon distances greater than 50 Mpc; we obtain estimates in agreement with upper range of the plausible estimates given in Abadie et al. (Reference Abadie2010b).

Table 1 shows that by 2016–2017 aLIGO/AdV will be accessible to NS/NS inspirals beyond the Coma cluster (100 Mpc). Beyond 2017, with rates of order 20 yr⁻¹, detections can be expected. The estimates provided in Table 1 assume a realistic event rate estimates for CBC sources (Abadie et al., Reference Abadie2010b); corresponding numbers that assume plausible pessimistic rate estimates can be obtained by scaling the detection rate estimates down by an order of magnitude. Adopting the latter estimates, there is still a reasonable chance of an NS/NS inspiral and merger detection during 2017.

7.1 Detecting unmodelled burst sources

Transients that are not well modelled due to their highly complex emissions are also targets for GW detectors; many such unmodelled bursts could be associated with LGRBs.

All sky burst searches aim to cast the widest possible net and utilise signal processing algorithms that are as robust as possible; no assumptions are made on the time of arrival, the signals origin or direction. Detection algorithms typically look for signals above a background noise level that are consistent in across multiple detectors; such algorithms often use time–frequency domain methods that look for excesses in time–frequency maps. For example, X-PIPELINE combines data from arbitrary detectors in a network and searches for clusters of pixels with energies significantly greater than background (Sutton et al., Reference Sutton2010). Searches are best employed in networks of detectors using coherent analyses, as described in Section 5.2. By combining amplitude and phase information from separate detectors in a network, the combined GW signal will increase coherently while the uncorrelated noise can be eliminated. The coherent WaveBurst (cWB) is the primary analysis pipeline for identifying burst signals in low latency (Klimenko et al., Reference Klimenko, Mohanty, Rakhmanov and Mitselmakher2005).

7.2 The detection range and rates of burst events

For unmodelled burst sources, the detection strategies are independent of waveform morphology. Therefore, an effective sensitive rangeFootnote ¹¹ for a narrow-band source can be estimated by considering the total energy emitted in GWs assuming a peak emission frequency, f ₀, for a given SNR ρ (Sutton, Reference Sutton2013):

(10) $$\begin{equation} D_{\mathrm{Eff}} \approx \left( \frac{G}{2 \pi ^{2} c^{3} } \right)^{1/2} \left( \frac{ 1 }{ S(f_{0}) f_{0}^{2} } \right)^{1/2} \left( \frac{ E_{\rm {GW}} }{ \rho ^{2} } \right)^{1/2}. \end{equation}$$

One can determine a convenient approximation of D _Eff:

(11) $$\begin{equation} D_{\mathrm{Eff}} \approx C_{\mathrm{B}}(f_{0}) \left( \frac{ E_{\rm {GW}} }{ \rho ^{2} } \right)^{1/2}, \end{equation}$$

for which, as in Section 6.2, values of C _B can be derived using the projected sensitivity noise curves for different epochs of observation for aLIGO. Values of C _B can be read from Figure 4 for a given f ₀; tabulated values of C _B are provided in Table B1.

Figure 4. To easily approximate the maximum detection ranges for different types of GW burst events, the curves of the function C_B(f) given in Equation (11) are provided for different values of the peak GW frequency. The curves represent the first two components of Equation (10) and are shown for different aLIGO observing scenarios as shown in Table 1.

Although there is significant uncertainty in these estimations, as will be discussed later in Section 8.1, such approximations can provide constraints on the global parameters of burst populations such as GRBs.

8.1 GRB searches

A number of searches have been conducted using LIGO data for coincident GRB events (Abadie et al., Reference Abadie2012b, Reference Abadie2012c; Aasi et al., Reference Aasi2013a; Abbott et al., Reference Abbott2008c; Predoi et al., Reference Predoi and Hurley2012). The recent GRB search of Abadie et al. (Reference Abadie2012b) used 154 GRBs observed during the LIGO and Virgo science runs of 2009–2010 and used both modelled and unmodelled searches in a time-window around the recorded time of the GRB and from the same directions on the sky. For unmodelled bursts, the X-PIPELINE” method was used to conduct a coherent search, assuming an optimistic emission in GWs of order 10⁻²M_⊙ and peak emission frequencies of 150 and 300 Hz. Modelled searches were conducted on the sample of short-duration GRBs by combining the data coherently and using template banks corresponding to probable parameters for coalescing systems of NSs and/or BHs (Harry & Fairhurst, Reference Harry and Fairhurst2011), also yielding exclusion distances—the distance beyond which the source must be to avoid detection. The median exclusion distances were 17 Mpc at 150 Hz for the unmodelled search and 16 Mpc for the modelled.

While no GW events were found, none of the observed GRBs fell within the exclusion distance; the closest to date was the llGRB 980425 at 36 Mpc (z ~ 0.0085). However, in the advanced detector era, null detections will yield exclusion distances useful to constrain models. For example, two llGRBs observed by Swift were at 145 Mpc (GRB 060218) and 264 Mpc (GRB 100316D). Such distances mean that some of the more extreme emission models can be put to the test using GW data.

As discussed earlier in Section 4, following the collapse of a massive star, long-lived ( ~ 10–1000 s) GW bursts may be produced from rotational instabilities in the protoneutron star remnant (Corsi & Mészáros, Reference Corsi and Mészáros2009a; Piro & Thrane, Reference Piro and Thrane2012; Piro & Ott, Reference Piro and Ott2011) or in the resulting accretion disk (van Putten, Reference van Putten2008; Piro & Pfahl, Reference Piro and Pfahl2007). In either case, the signal is expected to be narrow band with a slowly evolving frequency.

Specialised searches for long-lived GW transients associated with GRBs were conducted but have yielded no candidate detections (Aasi et al., Reference Aasi2013a). Sensitivity studies suggest that advanced detectors could detect such signals at distances of 44 Mpc (Thrane & Coughlin, Reference Thrane and Coughlin2014). There are significant theoretical uncertainties, but the rate of long-lived GW bursts may be sufficiently high for detections with advanced detectors (Piro & Thrane, Reference Piro and Thrane2012). EM counterparts might include jet-powered type II supernovae, a luminous red nova-like event, or an ‘un-nova’ (Piro & Thrane, Reference Piro and Thrane2012).

8.2 Individual GRB searches

GW searches based on the short-hard GRBs 051103 and 070201 were able to provide some insight into the hosts and emission mechanisms. In the case of GRB 051103, GW data supported evidence that this event was a giant flare of a Soft Gamma-ray Repeater (SGR) (Ofek et al., Reference Ofek2006; Frederiks et al., Reference Frederiks, Palshin, Aptekar, Golenetskii, Cline and Mazets2007; Hurley et al., Reference Hurley2010). Triangulation by the inter-planetary network (IPNFootnote ¹³) suggested that the bright short hard GRB 051103 was in the nearby M81 galaxy (3.6 Mpc). Whether it was from an SGRB (its duration was 0.17 s) or an SGR, giant flare was uncertain (Ofek et al., Reference Ofek2006; Hurley et al., Reference Hurley2010). The energy release, ~ 5 × 10⁴⁸ erg assuming it occurred in M81, is a factor of 10 times brighter than the brightest SGR giant flare observed (SGR 1806-20; Hurley, Boggs, & Smith, Reference Hurley2005; Hurley et al., Reference Hurley2010). Given a typical SGRB energy release of ~ 10⁵⁰ erg, for an SGRB origin to be compatible, the event could have been a background event to M81 or one would need to invoke a fainter population of short-hard GRBs (Lipunov et al., Reference Lipunov2005; Hurley et al., Reference Hurley2010).

Follow-up GW searches were performed for both modelled (assuming an inspiralling coalescing binary compact object) and unmodelled bursts (assuming events such as an associated star-quake in a magnetar) (Abadie et al., Reference Abadie2012c). Only the former signal would have been detectible at the distance of M81 (Levin & van Hoven, Reference Levin and van Hoven2011; Zink, Lasky, & Kokkotas, Reference Zink, Lasky and Kokkotas2012); the analysis and null result exclude a BNS merger in M81 as the progenitor with 98% confidence. If the event occurred in M81, the analysis supports the hypothesis of an SGR giant flare producing GRB 051103, which is therefore the most distant extragalactic magnetar observed. Similarly, the study of GRB 070201 (Abbott et al., Reference Abbott2008a) observed in M31, provided evidence that this burst did not result from a BNS merger from M31 and is likely to be an SGR giant flare. Given our understanding of SGR giant flares from our own Galaxy, it is statistically unlikely that both GRBs 051103 and 070201 were extragalactic SGR giant flares (Chapman, Priddey, & Tanvir, Reference Chapman, Priddey and Tanvir2009); hence, it is likely that one or both are classical SGRBs from background galaxies.

9.1 The GW detection pipeline

The main objective of the GW detection pipeline is to identify the most statistically significant GW triggers in the data stream, determine the most probable sky positions and relay the information to partner EM observational facilities as fast as possible—the general strategy was previously referred to as LOOC-UPFootnote ¹⁴ (Kanner et al., Reference Kanner2008; Shawhan, Reference Shawhan, Peck, Seaman and Comeron2012). The advanced detector era will see significant improvements in speed; and when a forth detector comes on line, coordinate reconstruction. The basic processes involved in sending out GW triggers to EM partners can be generalised as follows:

• Low-latency data analysis: For well-modelled CBC signals, matched filtering (see Section 7.1) is applied to the data using a bank of templates; these are based on the most probable ranges of source parameters e.g. component masses, inclination angles etc. Events above a defined SNR are recorded as triggers. Unmodelled burst searches are also conducted using techniques that are designed to detect a wide range of signals.

• Position reconstruction: Timing triangulation using the differences in the arrival times at each detector in a network can localise the source on the sky (Fairhurst, Reference Fairhurst2009). At the expense of speed, tighter confidence regions can be determined through more time intensive methods such as coherent analysis. The latter would be beneficial for the optical or radio follow-ups of GRB afterglows.

• Host Galaxy identification: As the positional errors are typically larger than the FoV of most EM instruments (typically tens of square degrees), the probability of a successful follow-up can be improved by using catalogues of nearby galaxies and globular clusters to apply statistical weight on individual tiles (typically 0.4° × 0.4°) of an error box (Nuttall & Sutton, Reference Nuttall and Sutton2010; Fan, Messenger, & Heng, Reference Fan, Messenger and Heng2014; Bartos, Crotts, & Marka, Reference Bartos, Crotts and Marka2015). We note that the final aLIGO detection horizon will extend to regions beyond which typical galaxy catalogues have good completeness. Additionally, sources with large galactic offsets could prove problematic (Tunnicliffe et al., Reference Tunnicliffe2014).

• FAR estimation: The statistical significance of a GW trigger is given through its FAR already discussed in Section 5.3. The FAR will identify high significance events that should be considered for follow-up. The FAR represents the average rate at which detector noise fluctuations create false positives with an equal or greater value than the detection statistic or SNR. The rate of background triggers it typically estimated by applying a number of artificial time-shifts of varying durations to the data streams of different detectors in a network around the time of the event—the time shifts remove any correlations from possible GW signals. By sampling different alignments of the statistical fluctuations, a measure of the background rate is obtained that sets the value of the FAR around any GW trigger. A typical FAR threshold adopted to send out alerts during O1 is around one event each month of livetimeFootnote ¹⁵

• Send out VOEvent: To rapidly communicate the information required by EM facilities for follow-up the VOEventFootnote ¹⁶ standard will be adopted (Williams et al., Reference Williams, Barthelmy, Denny, Graham, Swinbank, Peck, Seaman and Comeron2012). This is recognised as the standard syntax for fast dissemination of machine-readable information on astrophysical transients. There are currently different implementations of the VOEvent Transport ProtocolFootnote ¹⁷ that have been adopted by NASA and ESA space-based observatories including Swift and Fermi and will be used by the Square Kilometer Array pathfinder telescopes, The Low Frequency Array (LOFAR), ASKAP, and MeerKAT. The technical content of a VOEvent alert sent out by aLIGO/AdV for a CBC event should include estimates of the FAR (in Hz), chirp mass the maximum distance (in Mpc); for burst events. content will include central frequency, duration and an estimate of the energy fluence at Earth. Rather than a singular RA/Dec position, the sky position of a GW source will be provided by way of a probability sky map which can be multimodal and non-Gaussian.

9.2 Communicating GW triggers for EM follow-up

If the search pipelines find a candidate signal, it is recorded in the GW Candidate event Database (GraceDBFootnote ¹⁸). If its FAR is above threshold, a series of VOEvents are issued. The initial VOEvent will contain only basic information such as the event time, FAR and the GW detectors that have recorded the event. Subsequent VOEvents will contain the information discussed above including skymaps which will provide the probability that the event came from a particular region of sky. The VOEvent will contain a link to the sky map provided in the HEALPixFootnote ¹⁹ format. The first skymap will be a rapid localisation skymap determined by the BAYESTARFootnote ²⁰ pipeline (Singer et al., Reference Singer2014). This localisation information can be available within 10 s of seconds after detection (Singer, Reference Singer2015). After further analysis (of order hours), refined full parameter estimation skymaps will be provided using the more rigorous but computationally demanding stochastic samplers in the LALINFERENCE pipelineFootnote ²¹ that utilises detailed estimates of masses and spins (Berry et al., Reference Berry2015).

The morphology of the skymaps are dependent on the location of the source in the sky relative to the GW detector networks antenna pattern function. Some of the probability maps will consist of a single elongated arc which can cover several hundred square degrees, whilst others consist of two or more degenerate arcs. The degeneracy is a result of the two detector networks limited sensitivity to source polarisation (Schutz, Reference Schutz2011; Klimenko et al., Reference Klimenko2011). Figure 5 shows two example skymapsFootnote ²² typical of that expected from the period 2015–17 which will consist of a two detector network and later a third AdV at lower sensitivity (around 36 Mpc range as compared to around 100 Mpc for the aLIGO instruments). The plot shows both a single mode and a bimodal skymaps which will occur in almost equal numbers during this run.

Figure 5. Typical GW source skymaps expected from science runs between 2015 and 2017. The maps are Mollweide projections in geographical coordinates and show (a) two degenerate arcs totalling 820 deg² (event ♯10405) and (b) a single elongated arc of 692 deg² (event ♯790258). Both events have a network SNR of 12.7 and the true location of the events are shown by stars. The skymaps are taken from the website repository http://www.ligo.org/scientists/first2years/.

We note that for two-detector detections during this period both the BAYESTAR and stochastic sampler pipelines are expected to produce compatible localisation regions (Singer et al., Reference Singer2014; Berry et al., Reference Berry2015). For the case of 2016–17 with three-detectors in operation, triggers in all three instruments can provide confidence regions of tens of degrees, although this will occur in less than 17% of events (Singer et al., Reference Singer2014). If AdV records an SNR less than 4, as BAYESTAR only considers triggers above SNR = 4, it will ignore the third instrument; in this case the stochastic sampler could provide an improved estimate, with up to 50% smaller area, although within hours latency rather than seconds. By 2019, with aLIGO and AdV running at design sensitivity, up to 25% of coalescing binary sources are expected to be localised within 20 deg² (Aasi et al., Reference Aasi2013b).

To fully exploit the scientific promise of rapid GW triggered follow-ups, the signal processing will have to be conducted as close to real time as possible (low latency). This a tremendously complex task and is highly computationally demanding. A number of pipelines have been proposed and tested (Abadie et al., Reference Abadie2012a; Buskulic et al., Reference Buskulic2010; Cannon et al., Reference Cannon2012); this has been a particular focus for Australian facilities (Luan et al., Reference Luan, Hooper, Wen and Chen2011; Hooper et al., Reference Hooper, Chung, Luan, Blair, Chen and Wen2012). Pipelines are presently able to make detections in under a minute (Urban, personal communication), but the effort to get this down to as low as possible will continue throughout the GW multi-messenger era.

10.1 The first follow-up programme: 2009–2010

The first EM follow-up of GW triggers was performed during 2009–2010Footnote ²³ using the low-latency pipelines cWB, Omega, and MBTA (see Abadie et al., Reference Abadie2012d). GW data from the LIGO/Virgo network was calibrated and sent to the LIGO computing centre at Caltech for analysis within a minute. Although triggers were generated within 6 min, additional manual checks were performed to further verify the data quality and conditions at each detector site—these latter steps extended the total latency to around 10–30 min for each alert. As mentioned in Section 10.1, the strategy in this pilot study was referred to as LOOC-UP (Kanner et al., Reference Kanner2008; Shawhan, Reference Shawhan, Peck, Seaman and Comeron2012).

A total of 10 EM instruments were employed for LOOK-UP including Swift, LOFAR, ROTSE, TAROT, QUEST, the Liverpool Telescope, PTF, and Pi of the Sky; Australian participation was provided in the optical through SkyMapper (Keller et al., Reference Keller2007) and the Zadko Telescope (Coward et al., Reference Coward2010). Both instruments responded to GW triggers at a rate of around one per week, with nine and five tiles per trigger respectively; in total eight alerts were followed up (Aasi et al., Reference Aasi2014a). The main latency bottleneck during LOOK-UP was the manual checks on data quality and conditions. To allow alerts to be sent out significantly faster, automation was highlighted as an important prerequisite for coincident detection in the advanced detector era.

10.2 Multi-messenger astrophysics during the advanced detector era

The number of Australian facilities with involvement has increased for the advanced detector era. In addition to Zadko and SkyMapper, a number of other instruments have MoUs with the aLIGO/AdV event follow-up programme. The full complement is given in Table 2, along with their relevant specifications.

Table 2. The properties of a selection of the Australian instruments with MoUs in place for aLIGO/AdV follow-ups [1] Tingay et al. (Reference Tingay2013); [2] Murphy et al. (Reference Murphy2013); [3] Tinney et al. (Reference Tinney, Moorwood and Iye2004); [4] Keller et al. (Reference Keller2007); [5] Coward et al. (Reference Coward2010); [6] http://goto-observatory.org/; [7] Lennarz et al. (Reference Lennarz, Chadwick, Domainko, Parsons, Rowell and Tam2013); [8] Acharya et al. (Reference Acharya2013); Bartos et al. (Reference Bartos2014). ♭ Approximated using Figure 5 of Funk, Hinton, & CTA Consortium (Reference Funk and Hinton2013a). † Sensitivity in survey mode based on Bartos et al. (Reference Bartos2014). Exposure time includes an estimate of the required slewing times to tile a 1000 deg² area using convergent pointing mode.

In the following sections, starting from low-energy observational instruments up to high energy, we discuss these different facilities and their potential contribution towards the multi-messenger era. As discussed in Section 9.2, the greatest challenge that will face EM facilities will be contending with the large error regions which could often consist of two or more degenerate arcs—we cannot be certain of the exact error regions we will have to overcome. We can however consider two epochs in the following:

• Early epoch: This epoch includes the early and mid observing runs from 2015 to 2017 as given in Table 1. The median error regions will be in the range 230–500 deg² (Singer et al., Reference Singer2014)–we conservatively adopt the larger value of 500 deg² for our approximations. Near the end of this epoch, as AdV joins the two aLIGO detectors, one could expect to observe less than 12% of sources within 20 deg² (Aasi et al., Reference Aasi2013b), but it is safe to assume that the vast majority of the expected small sample of detections will have error regions of order 100 s of deg².

• Late epoch: During this epoch, aLIGO and AdV will be approaching design sensitivity. One can now expect of order 10–30% of the detections to be localised within 20 deg² (Aasi et al., Reference Aasi2013b). Chu et al. (Reference Chu, Howell, Rowlinson, Gao, Zhang, Tingay, Boer and Wen2015) have shown that assuming a three detector aLIGO/AdV network 100% of sources can be localised within 50 deg²—we will therefore conservatively adopt this value. We note that the inclusion of KAGRA in 2018–19 could improve the situation in terms of localisation; Chu et al. (Reference Chu, Howell, Rowlinson, Gao, Zhang, Tingay, Boer and Wen2015) further show that including this detector to expand the aLIGO/AdV network will allow 100% of sources to be localised to within 30 deg².

10.3 The radio domain

10.4 The optical domain

10.4.2 Coordinated optical observations of GW triggers

The relative sparsity of automated telescopes in the Southern hemisphere implies that instruments such as AAT, GOTO, SkyMapper, and Zadko can play an important role in GW follow-ups. This bias has been seen in the sky distribution of Swift triggered GRB optical afterglows (see for example, Figure 5 of Coward et al., Reference Coward2010). For the case of GRBs, this can hamper the sampling of light curves that last order ~ hours. Hence, both the longitude and latitude of the Australian optical facilities fill a niche space for follow-up.

In Figure 6, we examine the performance of the larger FoV Australian instruments shown in Table 2 (we have omitted AAT due to its smaller FoV) in terms of obtaining the optical afterglow of an SGRB associated with a NS/NS merger. The general formalism is given in Coward et al. (Reference Coward, Branchesi, Howell, Lasky and Böer2014) and considers a measure of the decay of the afterglow with time and a derived luminosity function. The plot is a good illustration of the capabilities of different instruments for rapid response follow-ups. The plot shows that in terms of GW follow-up of SGRBs associated with NS mergers, the first configuration of GOTO (assuming an exposure time of ~ 7.5 min for Phase 1 and 2 instruments) will be comparative with that of SkyMapper; both are well equipped for follow-ups and can achieve efficiencies of the order of 80–90% that of facilities such as BlackGEM (Ghosh & Nelemans, Reference Ghosh and Nelemans2015), Pan-STARRS (Hodapp et al., Reference Hodapp and Oschmann2004), and ZTF (Smith et al., Reference Smith, Ramsay, McLean and Takami2014). Zadko performs well in comparison to the fast response and wider FoV TAROT telescopes because of its sensitivity (TAROTs limit is 18 mag in the R-band and it has a FoV of 3.5 deg²).

Figure 6. A density plot of coincident GW-Optical detection efficiency to recover an SGRB (fading) optical afterglow in the imaging time versus telescope limiting magnitude plane. This plot, adapted from Coward et al. (Reference Coward, Branchesi, Howell, Lasky and Böer2014), shows the Australian optical instruments that have MoUs in place for aLIGO/AdV follow-ups. The total imaging time is the product of the number of tiles required to cover a uniform GW error box for a particular instruments FoV and exposure time. The efficiency, shown by the shaded regions is calculated by considering an optical afterglow luminosity function for SGRBs coupled with limiting magnitude and total imaging time of each instrument. We show results for two scenarios: early epoch (lhs: 500 deg²) and late epoch (rhs: 50 deg²). The Australian facilities Zadko and SkyMapper as well as GOTO (Phase 1, P1 and Phase 2, P2 which will include a second instrument in Australia), Pan-STARRS, BlackGEM and ZTF; three facilities expected to perform with high efficiency in follow-ups during the advanced detector era—their imaging [time/limiting magnitude] combinations result in their performance being far better the assumed parameter space shown for the late epoch. The efficiencies can be scaled by the expected detection rates and other caveats related to follow-up. We note that GOTO (both P1 and P2) and SkyMapper can make an important contributions to the follow-up programme in both epochs. Zadko can make a niche contribution during the latter stages of the advanced detector era as the error regions and detection rates improve.

We note here the coincident detection efficiencies considered in this section ignores a number of other factors including crowded star fields in the Galactic plane and Galactic dust obscuration. Other factors include expertise in dealing with false positives and the ability to apply optimum tiling strategies—it does however supply a gauge of how well Australian optical facilities can compete in this area.

Follow-up searches for optical r-process kilonova detections could also play an important role in the multi-messenger era. For a source at 200 Mpc, the predictions of Tanaka & Hotokezaka (Reference Tanaka and Hotokezaka2013) suggest that the flux should reach around 21–23 mag in the optical and 21–24 mag in the NIR JHK bands (in AB magnitudes). Although the AAT would seem well suited to NIR follow-ups, the small FoV of this instrument may make detections difficult for the large error box of a aLIGO/AdV network (10–100 deg² ). However, this instrument could be useful as part of a hierarchical strategy, providing deep follow-up of targets obtained from a larger FoV telescope. The large FoV of SkyMapper is well suited but would require an event with m < 21 mag. The final configuration of GOTO ( m = 21 mag with a large FoV (18–38 deg²) suggests this facility could be efficient for follow-up. In fact, dedicated instruments with a wide FoV such as GOTO should play an important role in the multi-messenger era as the first stage of a coordinated follow-up strategy, refining positions for smaller FoV EM instruments.

10.5 Ground-based follow-ups in gamma-rays