FERMI-LAT OBSERVATIONS OF THE LIGO EVENT GW150914

The ∼2.4 sr field of view (FoV) and broad energy coverage from 20 MeV to >300 GeV of the Large Area Telescope (LAT; Atwood et al. 2009) on the Fermi Gamma-ray Space Telescope mission make it a powerful instrument to monitor the sky for high-energy transients. As the LAT surveys the entire sky every 3 hr, it is sensitive to transient emission from a variety of sources, including stellar-mass compact objects (neutron stars—NSs, and black holes—BHs) over timescales from milliseconds to years, including those predicted to be associated with gravitational waves (GWs). Current GW detectors are sensitive to signals from the merging of compact objects. Some of these mergers, like the putative progenitors of short Gamma-Ray Bursts (sGRBs; Paczynski 1986; Eichler et al. 1989; Narayan et al. 1992; Rezzolla et al. 2011; Metzger & Berger 2012), emit both a short-lived γ-ray signal (≲2 s) immediately after the merger ("prompt" emission) and a long-lived and broadband "afterglow" signal lasting minutes to hours. If the GRB happens to be in the FoV at the time of the trigger, the LAT can detect the short-lived prompt emission phase. If the GRB is outside the FoV, because of its survey capability, the LAT can still detect the GRB by measuring its temporally extended afterglow emission as soon as it enters the FoV. The LAT has detected high-energy γ-ray emission from >130 GRBs to date (Vianello et al. 2016), including ∼10 sGRBs. Given the uncertainty in theoretical predictions for counterparts to GW sources and the demonstrated emission of high-energy γ-rays from systems containing compact objects, searching the LAT data for γ-ray counterparts to GW events is extremely important.

The era of GW astronomy began with the first science run ("O1") of the recently upgraded Laser Interferometer Gravitational-wave Observatory (LIGO; Abramovici et al. 1992; Abbott et al. 2009) from 2015 September to 2016 January. The Virgo Observatory (Caron et al. 1999; Acernese et al. 2009) will soon be added to the network for the second science run in late 2016. The GW frequency range that LIGO and Virgo are sensitive to is expected to be dominated by mergers of compact stellar-mass objects that are most likely remnants of stellar evolution: two neutron stars (NS–NS), two black holes (BH–BH), or an NS and a BH. The sensitivity and horizon distance of the GW network to these mergers scales with the masses of the systems; therefore, the accessible volume of the universe for NS mergers is significantly smaller than that of BH mergers. When those mergers include at least one NS, an electromagnetic (EM) counterpart is predicted to accompany the merger signal in the form of a sGRB. The EM outcome of a BH–BH merger is less well understood. Finding the counterpart of a GW event is important for understanding the nature of the source. It also has an additional yet less evident benefit: it significantly improves the accuracy with which all parameters (distance, mass, spin, inclination, etc.) can be estimated. This is obtained by better constraining the localization of the event, which is normally poorly estimated using only the GW signal (Sathyaprakash & Schutz 2009). Fermi-LAT can localize a transient source with sub-degree accuracy, a very big improvement with respect to a typical localization region from a GW detector, which will typically cover hundreds of square degrees.

On 2015 September 14 at 09:50:45 UTC, the LIGO Hanford and Livingston installations detected a coincident signal within 10 ms from GW150914, a high-significance trigger in the engineering run just prior to the start of O1. The trigger was determined to be consistent with a waveform predicted by general relativity from the inspiral and merger of a stellar-mass binary BH system, with constituent BHs of masses around 29 ${M}_{\odot }$ and 36 ${M}_{\odot }$ (Abbott et al. 2016b). The GW luminosity expected theoretically for a massive BH–BH merger leads to an estimate of around 400 Mpc (i.e., z ∼ 0.09; Abbott et al. 2016a) for the distance of the source. This observation provides evidence for the existence of isolated and binary stellar-mass BHs and is the first observation of such a system merging. Two days later, on 2015 September 16, LIGO notified the EM partner observatories operating within a Memorandum of Understanding (MOU). The EM partner observatories executed follow-up programs (Abbott et al. 2016c) to search for a counterpart within the 750 deg² localization region ($\sim 90\%$ confidence), which was later refined to 601 deg² (Abbott et al. 2016c). Fermi was operating in normal survey mode at the time of the trigger. Hence, the LAT autonomously observed the entire LIGO localization region within ∼70 minutes of the GW trigger, independently of any notification from LIGO, in the high-energy γ-ray band. The LAT Collaboration reported a preliminary search throughout the LIGO localization area that did not reveal any new γ-ray sources (Omodei et al. 2015). The results of a search of the data of the other instrument on board Fermi, the Gamma-ray Burst Monitor (GBM, Meegan et al. 2009) and the evidence for a weak counterpart are discussed separately in Connaughton et al. (2016).

In this paper, we describe LAT observations of the localization area of GW150914 around the time of the trigger, including a dedicated search for an EM γ-ray counterpart. No candidate counterparts were found. We describe the details of the data analysis in Section 2, discuss the implications of the counterpart search and prospects for future GW triggers in Section 3, and conclude in Section 4.

We performed two complementary sets of searches for transient high-energy γ-ray emission: automated searches (Section 2.1) that are performed routinely on all LAT data and targeted searches in the LIGO localization region (Section 2.2) on short and long time baselines that exploit the full sensitivity of the standard LAT analysis chain. In Appendices A and B we provide more details on the various analysis steps.

2.1. Automated Searches

2.2. Search in the LIGO Localization Region

The LIGO Scientific Collaboration reported results from a Bayesian parameter estimation analysis of GW150914 under the assumption that the signal arises from a compact binary coalescence (CBC) using the latest offline calibration of the GW strain data. The most accurate localization map for this event (LALInference) is based on a Bayesian Markov-Chain Monte Carlo and nested sampling to forward model the full GW signal, including spin precession and regression of systematic calibration errors. The localization probability is primarily in the southern portion of the annulus determined by the arrival time difference between LIGO Hanford and LIGO Livingston of ∼7 ms.

Given the uncertainty on EM signals from the merging of two BHs, we searched different time windows by carrying out two customized analyses of the LAT data. Both analyses are based on the standard maximum likelihood analysis technique used for LAT data and are summarized in Appendix A. In all of our searches we included in the likelihood model all sources (point-like and extended) from the LAT source catalog (3FGL; Acero et al. 2015) as well as the Galactic and isotropic diffuse templates provided by the Fermi-LAT Collaboration.⁷⁶ We used the Pass 8 P8_TRANSIENTR010E_V6 event class and the corresponding instrument response functions. These searches are described in the following sections.

2.2.1. Short-baseline Search

This search focuses on the hours immediately after the GW trigger t_GW (2015 September 14, 09:50:45 UTC). The LAT can detect long and short GRB afterglows up to thousands of seconds after the end of the prompt emission (De Pasquale et al. 2010; Ackermann et al. 2013c; Vianello et al. 2015). Thus, a search in this time window is the most likely to find a counterpart to GW150914 if it is similar to a sGRB. Fermi was in normal survey mode operations around t_GW, rocked $50^\circ$ North from the orbital plane. We consider a point in the sky observable by LAT if it is within the $65^\circ$ radius FoV and has an angle with respect to the local zenith smaller than $100^\circ$. The latter requirement is used to exclude contamination from γ-ray emission from the Earth's atmosphere. The coverage is the integral of the probability densities of all points in the LIGO localization probability map observable by LAT at a given time and it is shown in Figure 1 as a function of time. While the coverage was between 50% and 90% in the hours before the trigger (the dashed line), at t_GW the LAT was unfortunately viewing the opposite side of the sky from the LIGO localization region. The coverage was zero until $\sim {t}_{\mathrm{GW}}+4200$. The time interval ${t}_{\mathrm{GW}}+4442\mbox{--}4867\;{\rm{s}}$ (T₁) had a coverage >90%, while during ${t}_{\mathrm{GW}}+4867$–10,000 s coverage varied between 50% and 98%, decreasing back to zero at around ${t}_{\mathrm{GW}}+10$ ks. We searched for a transient source in the time interval having more than 90% coverage (T₁) and we did not find any significant excess.

Zoom In Zoom Out Reset image size

We then derived upper limits for the γ-ray flux of GW150914. Because the sensitivity of the LAT depends strongly on the angle from the source to the boresight of the instrument, the continuous variation of the LAT viewing direction in survey mode makes any flux limit for a particular source position time-dependent. Flux limits are also sensitive to astrophysical backgrounds, particularly in the Galactic plane, so that positions along the LIGO arc will have different flux limits even for the same observing conditions. These effects mean that flux limits vary according to both the time of observation and the position in the localization region. We show a map of the derived upper limits (95% confidence level) for the γ-ray flux of GW150914 in the band 100 MeV − 1 GeV in the left-hand panel of Figure 2 and a histogram of the upper limits in the right-hand panel, both for interval T₁. Assuming a power-law spectrum for the source with a photon index of $\alpha =-2$, which is typical for GRB afterglows at LAT energies, the upper limits we find have a median of $1.7\times {10}^{-9}$ erg cm⁻²s⁻¹, and 5% and 95% percentiles of $0.9\times {10}^{-9}$ and $3.7\times {10}^{-9}$ erg cm⁻²s⁻¹, respectively. These upper limits are only weakly dependent on the choice of α as shown in the right-hand panel of Figure 2. We now describe the upper limits analysis in more detail.

Zoom In Zoom Out Reset image size

We considered all γ-rays with reconstructed energies between 100 MeV and 100 GeV and arriving during the interval T₁. We then considered the LALInference probability map provided by LIGO, which is a HEALPix map with NSIDE = 512 (Górski et al. 2005), corresponding to a typical pixel size of ∼011. The point-spread function (PSF) of the LAT has a 68% containment radius at 1 GeV of ∼1°. To save computation time, we resampled the map to NSIDE = 256, which corresponds to a typical pixel size of ∼02. We considered the 1,1463 pixels in the resampled map contained within the 90% containment region provided by LIGO. Let us denote ${{\boldsymbol{v}}}_{i}$ the direction of the center of the i-th pixel. For each i-th pixel we performed an independent likelihood analysis (see Appendix 2.1), considering a region of interest (ROI) of 8° radius centered in ${{\boldsymbol{v}}}_{i}$ and testing for a new source at the position ${{\boldsymbol{v}}}_{i}$. For each likelihood analysis we required the zenith angle of the events to be no more than $100^\circ$. We did not find any new source with a test statistic (TS) above our adopted threshold of 25, corresponding to $\sim 5\sigma$ (pre-trials). We then computed the 95% confidence level upper limit for the flux of a source at each ${{\boldsymbol{v}}}_{i}$. To obtain upper limits reasonably independent of the photon index α adopted in the analysis, we need to choose an energy range small enough. We chose the range 100 MeV–1 GeV for the upper limits, which provides the largest photon statistic and the maximum sensitivity for sources similar to GRBs. The right-hand panel of Figure 2 shows that our measurement is indeed largely independent of the choice of α.

Connaughton et al. (2016) reported the weak transient γ-ray source GW150914-GBM lasting ∼1 s, 0.4 s after the LIGO trigger on GW150914. GW150914-GBM is consistent with being due to a low-fluence sGRB at an unfavorable viewing geometry to the GBM detectors, although this is not expected from a BH–BH merger. Assuming the two signals have a common origin, the combined LIGO and GBM observations reduce the 90% confidence region from 601 deg² to 199 deg². Within the combined LIGO/GBM localization and in the time interval T₁, the most significant excess in the LAT data has TS = 18. We estimate for this excess a p-value of ∼0.05, taking into account the number of trials of our analysis, which corresponds to a significance that is well below our threshold of $5\sigma$. The excess has a spectrum well-modeled with a power law with a soft photon index $\alpha =-3.2\pm 0.8$ and it is located close the limb of the Earth (which indeed has a soft spectrum). Therefore, we consider this excess in the LAT data very likely to be either a statistical fluctuation or due to Earth limb contamination and therefore unrelated to GW150914 or GW150914-GBM.

2.2.2. Long-baseline Search

In this second search we considered data gathered during a two-month interval centered on t_GW. To increase the number of γ-rays, we included all photons with energies between 60 MeV and 100 GeV. Because the PSF at 60 MeV is broad, we applied a zenith cut of $95^\circ$ to further limit Earth limb contamination. We looked both for a long-duration signal of the order of one day as well as for a short-duration signal, but not necessarily in strict temporal coincidence with the LIGO trigger. To this end, we covered the entire 90% probability region provided by LIGO with a set of nine partially overlapping ROIs, each with a radius of $10^\circ$. Figure 3 shows the locations of these ROIs and the confidence contours obtained from the LALInference probability map. They are overlaid on a sky map of the γ-rays detected by the LAT over the interval t_GW to ${t}_{\mathrm{GW}}+10,000\;{\rm{s}}$. In Table 1 we provide the location of the center of each ROI, listing all of the LAT sources from the 3FGL catalog (Acero et al. 2015) within each ROI along with their associations and their classes, using the same notation as the 3FGL catalog.

Zoom In Zoom Out Reset image size

Table 1.  ROIs and the 3FGL Sources They Contain

ROI (R.A., decl.)	R.A.	decl.	3FGL Name	Association	Classa
	deg	deg
ROI0 (150.0, −6.4)	147.223	0.363	3FGL J0948.8+0021	PMN J0948+0022	NLSY1
	148.255	−8.663	3FGL J0953.0−0839	PMN J0953−0840	bll
ROI1 (153.0, −48.0)	155.790	−57.759	3FGL J1023.1−5745	LAT PSR J1023−5746	PSR
	159.735	−53.186	3FGL J1038.9−5311	MRC 1036−529	bcu
	164.499	−52.455	3FGL J1057.9−5227	PSR J1057−5226	PSR
ROI2 (146.0, −63.0)	136.218	−57.570	3FGL J0904.8−5734	PKS 0903−57	bcu
	143.481	−62.534	3FGL J0933.9−6232
	154.730	−58.946	3FGL J1018.9−5856	1FGL J1018.6−5856	HMB
	155.790	−57.759	3FGL J1023.1−5745	LAT PSR J1023−5746	PSR
	157.123	−58.320	3FGL J1028.4−5819	PSR J1028−5819	PSR
	158.926	−67.334	3FGL J1035.7−6720
	161.129	−57.630	3FGL J1044.5−5737	LAT PSR J1044−5737	PSR
	161.277	−59.692	3FGL J1045.1−5941	Eta Carinae	BIN
	162.067	−58.535	3FGL J1048.2−5832	PSR J1048−5832	PSR
ROI3 (131.0, −68.0)	122.811	−75.492	3FGL J0811.2−7529	PMN J0810−7530	bll
	143.481	−62.534	3FGL J0933.9−6232
ROI4 (115.0, −70.0)	90.313	−70.609	3FGL J0601.2−7036	PKS 0601−70	fsrq
	98.942	−75.293	3FGL J0635.7−7517	PKS 0637−75	fsrq
	101.088	−67.223	3FGL J0644.3−6713	PKS 0644−671	bcu
	105.158	−66.173	3FGL J0700.6−6610	PKS 0700−661	bll
	122.811	−75.492	3FGL J0811.2−7529	PMN J0810−7530	bll
ROI5 (98.0, −71.0)	81.650	−68.420	3FGL J0526.6−6825e	LMC	GAL
	90.313	−70.609	3FGL J0601.2−7036	PKS 0601−70	fsrq
	98.942	−75.293	3FGL J0635.7−7517	PKS 0637−75	fsrq
	101.088	−67.223	3FGL J0644.3−6713	PKS 0644−671	bcu
	105.158	−66.173	3FGL J0700.6−6610	PKS 0700−661	bll
	122.811	−75.492	3FGL J0811.2−7529	PMN J0810−7530	bll
ROI6 (80.0, −71.0)	79.189	−62.121	3FGL J0516.7−6207	PKS 0516−621	bll
	81.650	−68.420	3FGL J0526.6−6825e	LMC	GAL
	90.313	−70.609	3FGL J0601.2−7036	PKS 0601−70	fsrq
	98.942	−75.293	3FGL J0635.7−7517	PKS 0637−75	fsrq
	101.088	−67.223	3FGL J0644.3−6713	PKS 0644−671	bcu
ROI7 (63.0, −66.0)	47.478	−60.963	3FGL J0309.9−6057	PKS 0308−611	fsrq
	76.780	−61.050	3FGL J0507.1−6102	PKS 0506−61	fsrq
	79.189	−62.121	3FGL J0516.7−6207	PKS 0516−621	bll
	81.650	−68.420	3FGL J0526.6−6825e	LMC	GAL
ROI8 (52.0, −57.0)	39.195	−61.600	3FGL J0236.7−6136	PKS 0235−618	fsrq
	47.478	−60.963	3FGL J0309.9−6057	PKS 0308−611	fsrq

Note.

^aClass names are: NLSY1: narrow Line Seyfert 1 galaxies; bll: BL Lac type of blazar; PSR: pulsar, identified by pulsations; bcu: blazar candidate of uncertain type; HMB: high-mass binary; BIN: binary; fsrq: FSRQ type of blazar; GAL: normal galaxy. From the 3FGL catalog (Acero et al. 2015).

Download table as:  ASCIITypeset image

For the first analysis of the second search, we divided the data in 10 ks time bins. For each time bin and for each ROI, we calculated a TS map (see Appendix B) and determined the location of the grid position with the maximum TS (TS_max). We considered the position of TS_max as the location of a possible counterpart and we ran an unbinned likelihood analysis adding a point source at the position of TS_max. This gave a value of TS_src (which is normally similar to TS_max). In these maps derived from low-statistics data, single high-energy γ-rays can cause a high value of TS_max. To reduce the number of false positives from random coincidences, we required that the number of photons ${N}_{\gamma }$ that have a probability larger than 0.9 to be associated with the candidate counterpart to be greater than 2. No excesses met this requirement. In Appendix B we present the TS maps for the 9 ROIs for the time interval 0–10 ks since t_GW (the lack of exposure from 0–4 ks is accounted for in the exposure calculation). We repeated the same analysis considering time bins of one day and again did not find any significant excess.

We also considered the possibility of excesses over shorter timescales (<1 hr), similar to the typical duration of high-energy emission from GRBs (Ackermann et al. 2013c; Vianello et al. 2015), but not in temporal coincidence with the GW trigger. We calculated the entry and exit times for each ROI in the FoV of the LAT (a "FoV passage"), requiring that the distance between the LAT boresight and the center of the ROI be <60°. In standard survey mode the duration of a FoV passage varies from a few hundred seconds to nearly one hour. Because we do not know if an EM signal would be in temporal coincidence with the GW signal, we searched for possible excesses in every passage, corresponding to a total of 6615 passages for each ROI. We did not detect any significant excess in any of the passages before or after t_GW for any ROI.

Figure 4 displays the four TS maps with the highest TS_src and ${N}_{\gamma }\gt 2$. The first map (TS${}_{\mathrm{src}}=21$; top left) is from ROI₀ and corresponds to the passage from 22:15:53 to 22:20:30 UTC on 2015 October 10 (17.5 days after the trigger time t_GW). The second map (top right) is from ROI₃ in the passage from 17:47:41 to 17:56:41 UTC on 2015 August 21 (20.7 days before t_GW) and has also a TS${}_{\mathrm{src}}=21$. The third map (TS${}_{\mathrm{src}}=22$; bottom left) is from ROI₇ in the interval from 21:35:48 to 21:50:01 UTC on 2015 August 20 (24.5 days before t_GW). Finally, the last map (bottom right) is from ROI₈ during 12:59:40 to 13:39:41 UTC on 2015 September 17 (3.1 days after the trigger LIGO trigger time t_GW) has the highest value of TS with TS${}_{\mathrm{src}}=23$. The peak TS values correspond to low-energy γ rays in random coincidence with high-energy γ rays, and the highest values obtained in this analysis should not be considered indicative of a possible EM counterpart. Moreover, the above search involves a large number of trials. To estimate their impact on the peak TS values we performed a Monte Carlo study described in the next section.

Zoom In Zoom Out Reset image size

To validate our interpretations of TS values we performed a detailed Monte Carlo simulation of two months of data (the same interval used in our analysis). The actual pointing history of the satellite was used; therefore, the correct exposure of the sky was automatically taken into account. All of the sources from the 3FGL catalog were kept fixed at their 3FGL catalog fluxes. As a result, the simulation is suitable for computing the distribution of TS under the null hypothesis that no transient signal is present. With the simulated data we repeated exactly the same analysis used on real data that we described in the previous section.

In Figure 5 we compare the distribution of TS_src obtained from flight data (the filled lines) and Monte Carlo (the dashed lines). We note that the Monte Carlo distributions are a good match to the distributions of the TS_src values obtained from the flight data and the good absolute agreement is consistent with no statistically significant transient counterpart being present in the flight data. Also, given the large number of trials, relatively high values of TS can be obtained in Monte Carlo simulations even if no transient signal was added. In the flight data we found four cases with TS > 20 and with ${N}_{\gamma }\gt 2$, and this must be compared with the nine cases we obtain when we analyzed the simulated data. In other words, we expected nine false positives with TS${}_{\mathrm{src}}\gt 20$ (and ${N}_{\gamma }\gt 2$) in two months of flight data observed in 4.

Zoom In Zoom Out Reset image size

The most promising astrophysical GW sources in the frequency range of LIGO/Virgo are the mergers of compact object binaries with NS and/or stellar-mass BH constituents. The detection of GW150914 is undoubtedly a major breakthrough in astronomy, being the first detection of GWs and the first detection of a merging binary BH system.

Maximizing the science return from GW detections requires the identification and study of coincident EM counterparts, which would help resolve degeneracies associated with the inferred binary parameters. For example, a strong γ-ray signal may only be seen at small binary inclination relative to the sky, such that the jet (along the direction of the total angular momentum) is pointed toward us. A high-significance counterpart association would also decrease the significance threshold necessary for a confident GW detection, thereby effectively increasing the distance to which the GW signal can be detected and the searchable volume as the cube of the distance. It would help identify the host galaxy and thereby constrain or measure the merger redshift, setting the luminosity scale and allowing an independent measurement of the Hubble constant or other cosmological parameters (Cannon et al. 2012; Metzger & Berger 2012). The complementary information encoded in the EM signal (spectral and temporal evolution, energetics, and inferred environment) is likewise essential to unravel the astrophysical context of the coalescence event.

The most commonly hypothesized EM counterpart of an NS–NS/NS–BH merger is a sGRB, powered by accretion onto one of the two central compact objects (Paczynski 1986; Eichler et al. 1989; Narayan et al. 1992; Rezzolla et al. 2011; Metzger & Berger 2012), which launches relativistic jets that produce a short (<2 s) bright flash of keV–MeV peak γ rays followed by a broadband longer-lasting afterglow. The LAT detects approximately two sGRBs per year, with >100 MeV afterglows lasting up to hundreds of seconds after the trigger (Ackermann et al. 2010, 2013c), seeded by sGRB triggers from the GBM and Swift-BAT. The most prominent example of a LAT-detected sGRB is GRB 090510, a very luminous burst with ${{\rm{E}}}_{\mathrm{iso}}\sim {10}^{53}$ erg s⁻¹, which simultaneously triggered Swift and GBM as well as causing a LAT on-board transient source trigger, resulting in a LAT localization being circulated within seconds. The LAT detected both a spike during the prompt γ-ray emission and an extended afterglow lasting hundreds of seconds, which is consistent with the observed optical and X-ray emission (De Pasquale et al. 2010; Giuliani et al. 2010). Figure 6 shows the >100 MeV γ-ray light curve of GRB 090510 scaled to z = 0.09, the nominal redshift of GW150914 inferred from the GW observations, in comparison to the LAT upper limits described in Section 2.2.1. The rate of sGRB is highly uncertain, and one as bright as GRB90510 located at z = 0.09 is a rare event. Presuming that NS–NS mergers are the progenitors of sGRB events, estimates of the number of sGRBs within z < 0.1 (see, for example, Guetta & Piran 2006) span a broad range of approximately one in every five to 100 years, depending on various assumptions concerning the progenitor characteristics and their population. In the bright sGRB sample detected by Swift (D'Avanzo et al. 2014), 1 sGRB has redshift 0.122, suggesting that the rate of sGRB at very low redshift is not negligible. If GW150914 would have had a luminosity and a high-energy γ-ray light curve similar to GRB 090510 and been more favorably placed relative to the LAT boresight at the trigger time, it would have been easily detectable by the LAT during observations similar to those described in this paper. A GRB at the redshift of the LIGO GW source entering the FoV of Fermi-LAT within the first 100 s would have been detected if it were more than an order of magnitude less luminous than GRB 090510.

Zoom In Zoom Out Reset image size

LAT-detected long GRBs have also shown similar behavior, with a prompt spike contemporaneous with sub-MeV emission followed by long-lived emission lasting from minutes to hours with similar time dependence as radio-to-X-ray afterglows (Ackermann et al. 2013b, 2014). Although the detection rate of LAT sGRBs is low, the low redshifts of potentially detectable GW sources and therefore their potentially bright EM counterpart emission, the uncertainty in the observational signatures of GW sources and the continuous observations of the entire sky during survey operations make searching the LAT data for counterpart sources worthwhile.

3.1. EM Counterparts to BH Binary Mergers

The Fermi-LAT is uniquely capable of searching for high-energy γ-ray counterparts to sources detected by GW observatories. We use this capability to undertake a detailed search in the regular LAT survey data for a counterpart temporally and spatially coincident with the LIGO trigger on GW150914. Although GW150914 was not in the LAT FoV at the trigger time, the LAT observed the entire region within ∼70 minutes of the GW trigger. We searched on short and long timescales for evidence of a transient γ-ray source contemporaneous with GW150914. No LAT counterpart is detected and upper limits have been set on the GeV γ-ray flux within the LIGO localization.

As the sensitivity of LIGO and Virgo improve over the next few years, their detection horizon for NS–NS and NS–BH binary mergers and thus the likelihood of sGRB coincidence, will increase greatly. The merger rates are highly uncertain and depend on the populations and evolution of binaries and opening angles of sGRB jets. The coincident detection of prompt γ-ray signals with on-axis or potentially off-axis afterglow signals will teach us about the physics of binary mergers. With the discovery of GW150914, the search for EM-GW coincidences enters a new phase. It is important to test the strong expectation that stellar BH mergers do not radiate much light; it is equally important to refine the techniques that will be needed to associate sGRBs with GW events from NS mergers. The approach described in this communication is well-suited to achieve these twin goals when LIGO detects more GW events.

The Fermi LAT Collaboration acknowledges generous ongoing support from a number of agencies and institutes that have supported both the development and the operation of the LAT as well as scientific data analysis. These include the National Aeronautics and Space Administration and the Department of Energy in the United States, the Commissariat à l'Energie Atomique and the Centre National de la Recherche Scientifique/Institut National de Physique Nucléaire et de Physique des Particules in France, the Agenzia Spaziale Italiana and the Istituto Nazionale di Fisica Nucleare in Italy, the Ministry of Education, Culture, Sports, Science and Technology (MEXT), the High Energy Accelerator Research Organization (KEK) and Japan Aerospace Exploration Agency (JAXA) in Japan, and the K.A. Wallenberg Foundation, the Swedish Research Council, and the Swedish National Space Board in Sweden. Additional support for science analysis during the operations phase is gratefully acknowledged from the Istituto Nazionale di Astrofisica in Italy and the Centre National d'Études Spatiales in France.

The standard tools for Fermi-LAT analysis, the Fermi ScienceTools, are available for download from the FSSC⁷⁷ , where the LAT data can also be downloaded. In all analyses presented in this paper, we used the Pass 8 data class P8R2_TRANSIENT010E_V6 and the ST v10r0p5. The ScienceTools are based on the standard maximum likelihood analysis: a model summarizing knowledge about the sources of γ rays in a particular region of the sky is convolved with the instrument response and optimized over its parameters to maximize the likelihood that the model best represents the data. The details on how to perform such an analysis with LAT data are described on the FSSC website; here we summarize the main steps. We start by selecting all the data within a given energy range and contained within a ROI. In the case of the unbinned likelihood analysis used in this paper, the ROI is circular and is therefore characterized by a center and a radius. Because the upper layers of the Earth's atmosphere are a bright source of γ rays that are very difficult to model in the likelihood analysis, we need to further reduce the Earth Limb contamination in the data. At the altitude of the nearly circular Fermi orbit, the limb is always seen by the LAT at an angle of $\sim 113^\circ$ from the zenith direction. We therefore remove all photons with zenith angles larger than a threshold that depends on the minimum energy used in the analysis (because the PSF is larger at lower energies). In this paper, we define the size of the ROI, the energy range, and the zenith angle limit used in each analysis in the respective sections.

A.1. Source Significance

A.2. TS Maps and Trial Factors

For the TS maps of the long-baseline search we use the largest grid fitting inside each of the 9 ROIs. We model the background by taking into account the isotropic component (defined in the template file iso_P8R2_TRANSIENT010E_V6_v06.txt), which includes the extragalactic diffuse γ radiation and residual charged-particle contamination, and the Galactic diffuse γ-ray emission (using the spatial and spectral template in gll_iem_v06.fits), which is the result of interaction of cosmic rays with the gas and the EM field of the Milky Way. The normalization of the first is left free to vary in order to accommodate orbital variations, while the Galactic diffuse emission model is held fixed to its nominal value. All the point sources from the 3FGL catalog (Acero et al. 2015) are also included with their parameters fixed.

In Figure 7 we display the TS maps for each region integrated from t_GW to t_GW + 10 ks. The 3FGL sources have been overlaid as yellow stars, while the white circles represent the Fermi-LAT γ rays (the size of the circle is proportional to the reconstructed energy of the event).

Zoom In Zoom Out Reset image size

For completeness, we report the Fermi-LAT sources in the 3FGL catalog that are present in each ROI with, where available, source and class associations. As the integration times in our analysis are much shorter than the integration time of the catalog, we select only those sources with an average significance (Signif_Avg in the 3FGL) >20, which roughly corresponds to an average flux in the range 10⁻⁹–10⁻⁷ ph cm⁻² s⁻¹ in the energy range 0.1–1 GeV.

Two strategies are useful for associating EM detections of transients with GW sources. The first is to search the GW data for counterparts to EM events. This was done archivally using GRB triggers for the years 2006 to 2011 (Aasi et al. 2014) and the initial LIGO data runs, yielding no credible candidate sources (with a $\lt 20$ Mpc horizon distance). This approach is not very efficient, as most of the EM events are outside the current volume sampled by the initial LIGO runs, which increases the trials factors and diminishes the GW sensitivity.

The second approach is to search for EM bursts (e.g., sGRB) related to GW events. The most promising way to find a counterpart is via a prompt trigger from the burst itself using wide FoV instruments like GBM or Swift-BAT. Given its large FoV and sensitivity to energies from 8 keV to 40 MeV, GBM is the most prolific detector of sGRBs currently operating, with ∼40 to 80 detections per year, and therefore it is very well-suited for EM counterpart searches (Connaughton et al. 2016). Yet the LAT offers several capabilities that make it a unique resource.

In the normal mode of Fermi GRB detection and measurement, GBM issues triggers on board for both short and long GRBs (as well as solar flares, terrestrial γ-ray flashes, magnetar flares, and other short, hard X-ray/soft γ-ray transients), and approximately half of the GBM GRBs occur in the LAT FoV. However, in cases of high peak brightness, GBM initiates an automated repoint of the Fermi spacecraft to optimize LAT observations over the next 2.5 hr. Regardless of whether an automated pointing is triggered, even during normal sky survey operations, the LAT will observe the entire sky within two orbits (∼3 hr), fully covering 100% of any GW localization contour, making follow-up observations of GW triggers both automatic and routine. Neither instrument on Fermi has observing constraints limiting the viewing near the Sun or the moon.

If the LAT were to detect a counterpart to a GW trigger, it has the distinct advantage of providing smaller localization uncertainties than GBM. Typical GBM localizations have radii of several degrees plus systematic uncertainties of 3°–14° (Connaughton et al. 2015), while LAT localizations have radii ∼01–1°. A LAT detection would substantially reduce the sky area for follow-up observations at other wavelengths, some of which require extensive tiling campaigns over days after a trigger to cover a significant fraction of the GW localization region (Evans et al. 2016; Abbott et al. 2016c). Although the LAT obtained an on-board localization for short GRB 090510 (Ackermann et al. 2010) within 15 seconds, LAT localizations typically have an ∼8 hr latency for data transmission and ground processing.

The LAT has also recently (as of 2015 June) benefited from a major upgrade to the event reconstruction analysis pipeline, known as Pass 8 (Atwood et al. 2013). Pass 8 improves the LAT sensitivity due to increased photon acceptance, especially at low (<100 MeV) and high energies (>10 GeV), and reduces localization radii by $\sim 30\%$. Combined with new analysis pipelines that search for transients on all timescales (Vianello et al. 2015), the LAT is now better suited to discover counterparts to GW sources in both automated pipelines and the specialized searches described in Section 2.

The LIGO EM follow-up partners conducted large follow-up campaigns in the optical, radio, and X-ray to search for a counterpart to GW150914 (Abbott et al. 2016c). Similar campaigns will occur for other GW triggers no matter the likely progenitor and regardless of whether it is observed along the putative jet axis. Although no orphan GRB afterglows (which are expected to be associated with off-axis observing angles) have conclusively been detected, models (Granot et al. 2002; van Eerten et al. 2010) predict delayed, fainter transient emission from the afterglow itself, an isotropic optical-near-infrared kilonova signature (Li & Paczyński 1998). Other potential EM counterparts have been theorized, including late radio emission from mildly relativistic material that is dynamically ejected during the merger and drives a shock into the external medium (Nakar & Piran 2011), and radio-to-X-ray (or γ-ray) emission over seconds to days from relativistic ejecta created as the NSs collide during their merger (Kyutoku et al. 2014). Given the capability of the LAT to detect transients on timescales from milliseconds to years, careful searches for possible afterglow emission within the LIGO localization area over relevant timescales after the GW trigger are essential and require no change to the Fermi observing strategy.

High-energy γ-ray emission from GRBs observed by the LAT above 100 MeV lasts much longer (minutes to hours) than the prompt emission observed by the GBM (Ackermann et al. 2013c) for both long and short GRBs (Abdo et al. 2010; De Pasquale et al. 2010). Therefore, even if a localization probability region from LIGO/Virgo is outside the FoV of the LAT during the time of the trigger, the instrument could detect temporally extended emission in the minutes following the prompt signal when the region re-enters its FoV. This could result in a much more precise localization of the EM counterpart.