LOCALIZATION AND BROADBAND FOLLOW-UP OF THE
GRAVITATIONAL-WAVE TRANSIENT GW150914
The LIGO Scienti
fi
c Collaboration and the Virgo Collaboration, the Australian Square Kilometer Array
Path
fi
nder
(
ASKAP
)
Collaboration, the BOOTES Collaboration, the Dark Energy Survey and the Dark Energy
Camera GW-EM Collaborations, the Fermi GBM Collaboration, the GRAvitational Wave Inaf TeAm
(
GRAWITA
)
,
the INTEGRAL Collaboration, the Intermediate Palomar Transient Factory
(
iPTF
)
Collaboration, the
InterPlanetary Network, the J-GEM Collaboration, the La Silla
–
QUEST Survey, the Liverpool Telescope
Collaboration, the Low Frequency Array
(
LOFAR
)
Collaboration, the MASTER Collaboration, the MAXI
Collaboration, the Murchison Wide-
fi
eld Array
(
MWA
)
Collaboration, the Pan-STARRS Collaboration, the
PESSTO Collaboration, the Pi of the Sky Collaboration, the SkyMapper Collaboration, the Swift Collaboration,
the TAROT, Zadko, Algerian National Observatory, and C2PU Collaboration, the TOROS Collaboration,
and the VISTA Collaboration
(
See the Supplement, Abbott et al.
2016b
, for the full list of authors.
)
Received 2016 February 29; revised 2016 April 26; accepted 2016 May 2; published 2016 July 20
ABSTRACT
A gravitational-wave
(
GW
)
transient was identi
fi
ed in data recorded by the Advanced Laser Interferometer
Gravitational-wave Observatory
(
LIGO
)
detectors on 2015 September 14. The event, initially designated G184098
and later given the name GW150914, is described in detail elsewhere. By prior arrangement, preliminary estimates
of the time, signi
fi
cance, and sky location of the event were shared with 63 teams of observers covering radio,
optical, near-infrared, X-ray, and gamma-ray wavelengths with ground- and space-based facilities. In this Letter we
describe the low-latency analysis of the GW data and present the sky localization of the
fi
rst observed compact
binary merger. We summarize the follow-up observations reported by 25 teams via private Gamma-ray
Coordinates Network circulars, giving an overview of the participating facilities, the GW sky localization
coverage, the timeline, and depth of the observations. As this event turned out to be a binary black hole merger,
there is little expectation of a detectable electromagnetic
(
EM
)
signature. Nevertheless, this
fi
rst broadband
campaign to search for a counterpart of an Advanced LIGO source represents a milestone and highlights the broad
capabilities of the transient astronomy community and the observing strategies that have been developed to pursue
neutron star binary merger events. Detailed investigations of the EM data and results of the EM follow-up
campaign are being disseminated in papers by the individual teams.
Key words:
gravitational waves
–
methods: observational
1. INTRODUCTION
In 2015 September, the Advanced Laser Interferometer
Gravitational-wave Observatory
(
LIGO; Aasi et al.
2015
)
made
the
fi
rst direct detection of an astrophysical gravitational-wave
(
GW
)
signal that turned out to be from a binary black hole
(
BBH
)
merger. The LIGO Hanford and Livingston sites are
the
fi
rst two nodes of a growing global network of highly
sensitive GW facilities, soon to include Advanced Virgo
(
Acernese et al.
2015
)
, KAGRA, and LIGO
–
India. Some of
the most promising astrophysical sources of GWs are also
expected to produce broadband electromagnetic
(
EM
)
emis-
sion and neutrinos. This has cr
eated exciting new opportu-
nities for joint broadband EM observations and multi-
messenger astronomy.
In a compact binary coalescence
(
CBC
)
event, a tight binary
comprised of two neutron stars
(
NSs
)
, two black holes
(
BHs
)
,
or a NS and a BH experiences a runaway orbital decay due to
gravitational radiation. In a binary including at least one NS
—
a
binary neutron star
(
BNS
)
or neutron star
–
black hole
(
NSBH
)
merger
—
we expect EM signatures due to energetic out
fl
ows at
different timescales and wavelengths. If a relativistic jet forms,
we may observe a prompt short gamma-ray burst
(
GRB
)
lasting
on the order of one second or less, followed by X-ray, optical,
and radio afterglows of hours to days duration
(
e.g., Eichler
et al.
1989
; Narayan et al.
1992
; Nakar
2007
; Berger
2014
;
Fong et al.
2015
)
. Rapid neutron capture in the sub-relativistic
ejecta
(
e.g., Lattimer & Schramm
1976
)
is hypothesized to
produce a kilonova or macronova, an optical and near-infrared
signal lasting hours to weeks
(
e.g., Li & Paczy
ń
ski
1998
)
.
Eventually,wemayobservearadioblastwavefromthis
sub-relativistic out
fl
ow, detectable for months to years
(
e.g.,
Nakar & Piran
2011
)
. Furthermore, several seconds prior to
or tens of minutes after merger, we may see a coherent radio
burst lasting milliseconds
(
e.g., Hansen & Lyutikov
2001
;
Zhang
2014
)
.Inshort,aNSbinarycanproduceEM
radiation over a wide range of wavelengths and timescales.
On the other hand, in the case of a stellar-mass BBH, the
current consensus is that no signi
fi
cant EM counterpart
emission is expected except for those in highly improbable
environments pervaded by large ambient magnetic
fi
elds or
baryon densities.
The
fi
rst campaign to
fi
nd EM counterparts triggered by low-
latency GW event candidates was carried out with the initial
The Astrophysical Journal Letters,
826:L13
(
8pp
)
, 2016 July 20
doi:10.3847
/
2041-8205
/
826
/
1
/
L13
© 2016. The American Astronomical Society. All rights reserved.
1
LIGO and Virgo detectors and several EM astronomy facilities
in 2009 and 2010
(
Abadie et al.
2012a
,
2012b
; Evans et al.
2012
; Aasi et al.
2014
)
. In preparing for Advanced detector
operations, the LIGO and Virgo collaborations worked with the
broader astronomy community to set up an evolved and greatly
expanded EM follow-up program.
381
Seventy-four groups with
access to ground- and space-based facilities joined, of which 63
were operational during Advanced LIGO
ʼ
s
fi
rst observing run
(
O1
)
. Details of the 2009 to 2010 EM follow campaign and
changes for O1 are given in Section 1 of the Supplement
(
Abbott et al.
2016b
)
.
After years of construction and commissioning, the Advanced
LIGO detectors at Livingston, Louisiana, and Hanford,
Washington, began observing in 2015 September with about
3.5 times the distance reach
(
<
40 times the sensitive volume
)
of
the earlier detectors. A strong GW event was identi
fi
ed shortly
after the pre-run calibration process was completed. Deep
analysis of this event, initially called G184098 and later given
the name GW150914, is presented in Abbott et al.
(
2016e
)
and
companion papers referenced therein. In this paper we describe
the initial low-latency analysis and event candidate selection
(
Section
2
)
, the rapid determination of likely sky localization
(
Section
3
)
, and the follow-up EM observations carried out by
partner facilities
(
Sections
4
and
5
)
. For analyses of those
observations, we refer the reader to the now-public Gamma-ray
Coordinates Network
(
GCN
)
circulars
382
and to a number of
recent papers. We end with a brief discussion of EM counterpart
detection prospects for future events.
2. DATA ANALYSIS AND DISCOVERY
As con
fi
gured for O1, four low-latency pipelines continually
search for transient signals that are coincident in the two
detectors within the 10 ms light travel time separating them.
Coherent WaveBurst
(
cWB; Klimenko et al.
2016
)
and
Omicron
+
LALInference Burst
(
oLIB; Lynch et al.
2015
)
both
search for unmodeled GW bursts
(
Abbott et al.
2016f
)
.
GSTLAL
(
Cannon et al.
2012
; Messick et al.
2016
)
and
Multi-Band Template Analysis
(
MBTA; Adams et al.
2015
)
search speci
fi
cally for NS binary mergers using matched
fi
ltering. Because CBC waveforms can be precisely computed
from general relativity, GSTLAL and MBTA are more
sensitive to CBC signals than the burst search pipelines are.
All four detection pipelines report candidates within a few
minutes of data acquisition.
LIGO conducted a series of engineering runs throughout
Advanced LIGO
ʼ
s construction and commissioning to prepare
to collect and analyze data in a stable con
fi
guration. The eighth
engineering run
(
ER8
)
began on 2015 August 17 at 15:00 and
critical software was frozen by August 30.
383
The rest of ER8
was to be used to calibrate the detectors, to carry out diagnostic
studies, to practice maintaining a high coincident duty cycle,
and to train and tune the data analysis pipelines. Calibration
was complete by September 12 and O1 was scheduled to begin
on September 18. On 2015 September 14, cWB reported a
burst candidate to have occurred at 09:50:45 with a network
signal-to-noise ratio
(
S
/
N
)
of 23.45 and an estimated false
alarm rate
(
FAR
)
<
0.371 yr
−
1
based on the available
(
limited
at that time
)
data statistics. Also, oLIB reported a candidate
with consistent timing and S
/
N. No candidates were reported at
this time by the low-latency GSTLAL and MBTA pipelines,
ruling out a BNS or NSBH merger.
Although the candidate occurred before O1 of
fi
cially
began, the LIGO and Virgo collaborations decided to send
an alert to partner facilites because the preliminary FAR
estimate satis
fi
ed our planned alert threshold of 1 month
−
1
.
Although we had not planned to disseminate real-time GCN
notices before the formal start of O1, most of the computing
infrastructure was in place. Basic data quality checks were
done within hours of GW150914; both interferometers were
stable and the data stream was free of artifacts
(
Abbott et al.
2016c
)
. A cWB sky map was available 17 minutes after the
data were recorded and a LALInference Burst
(
LIB
)
sky map
was available after 14 hr. After extra data integrity checks and
an update to the GCN server software, these two sky maps
were communicated to observing partners in a GCN circular
nearly two days after the event occurred
(
GCN
18330
)
. Mass
estimates were not released in this initial circular,
Figure 1.
Timeline of observations of GW150914, separated by band and relative to the time of the GW trigger. The top row shows GW information releases. The
bottom four rows show high-energy, optical, near-infrared, and radio observations, respectively. Optical spectroscopy and narrow-
fi
eld radio observations are
indicated with darker tick marks and boldface text. Table
1
reports more detailed information on the times of observations made with each instrument.
381
See program description and participation information at
http:
//
www.ligo.
org
/
scientists
/
GWEMalerts.php
.
382
All circulars related to GW150914 are collected at
http:
//
gcn.gsfc.nasa.
gov
/
other
/
GW150914.gcn3
.
383
All dates and times are in UT.
2
The Astrophysical Journal Letters,
826:L13
(
8pp
)
, 2016 July 20
Abbott et al.
and observers may have assumed the event was associated
with a BNS system or a GW burst
(
e.g., from a nearby core-
collapse supernova; SN
)
. The knowledge that GW150914
was consistent with a BBH inspiral and merger was only
shared later on October 3
(
GCN
18388
)
. Figure
1
shows the
chronology of the GW detection alerts and follow-up
observations.
The data were re-analyzed of
fl
ine with two independent
matched-
fi
lter searches using a template bank that includes
both NS binary and BBH mergers. The waveform was
con
fi
rmed to be consistent with a BBH merger and this
information was shared with observers about three weeks after
the event
(
GCN
18388
)
. The FAR was evaluated with the data
collected through 20 October, reported to be less than 1 in 100
years
(
GCN
18851
; Abbott et al.
2016d
)
, and ultimately
determined to be much lower. The
fi
nal results of the of
fl
ine
search are reported in Abbott et al.
(
2016e
)
.
3. SKY MAPS
We produce and disseminate probability sky maps using a
sequence of algorithms with increasing accuracy and
computational cost. Here, we compare four location estimates:
the prompt cWB and LIB localizations that were initially
shared with observing partners, plus the rapid BAYESTAR
localization and the
fi
nal localization from LALInference. All
four are shown in Figure
2
.
cWB performs a constrained maximum likelihood estimate
of the reconstructed signal on a sky grid
(
Klimenko et al.
2016
)
weighted by the detectors
’
antenna patterns
(
Essick et al.
2015
)
and makes minimal assumptions about the waveform morph-
ology. With two detectors, this amounts to restricting the signal
to only one of two orthogonal GW polarizations throughout
most of the sky. LIB performs Bayesian inference assuming the
signal is a sinusoidally modulated Gaussian
(
Lynch
et al.
2015
)
. While this assumption may not perfectly match
the data, it is
fl
exible enough to produce reliable localizations
for a wide variety of waveforms, including BBH inspiral-
merger-ringdown signals
(
Essick et al.
2015
)
. BAYESTAR
produces sky maps by triangulating the times, amplitudes, and
phases on arrival supplied by all the CBC pipelines
(
Singer &
Price
2016
)
. BAYESTAR was not available promptly because
the low-latency CBC searches were not con
fi
gured for BBHs;
the localization presented here is derived from the of
fl
ine CBC
Figure 2.
Comparison of different GW sky maps, showing the 90% credible level contours for each algorithm. This is an orthographic projection centered on the
centroid of the LIB localization. The inset shows the distribution of the polar angle
θ
HL
(
equivalently, the arrival time difference
Δ
t
HL
)
.
3
The Astrophysical Journal Letters,
826:L13
(
8pp
)
, 2016 July 20
Abbott et al.
search. LALInference performs full forward modeling of the
data using a parameterized CBC waveform which allows for
BH spins and detector calibration uncertainties
(
Veitch
et al.
2015
)
. It is the most accurate method for CBC signals
but takes the most time due to the high dimensionality. We
present the same LALInference map as Abbott et al.
(
2016g
)
,
with a spline interpolation procedure to include the potential
effects of calibration uncertainties. The BAYESTAR and
LALInference maps were shared with observers on 2016
January 13
(
GCN
18858
)
, at the conclusion of the O1 run.
Since GW150914 is a CBC event, we consider the LALInfer-
ence map to be the most accurate, authoritative, and
fi
nal
localization for this event. This map has a 90% credible region
with area 630 deg
2
.
All of the sky maps agree qualitatively, favoring a broad,
long section of arc in the southern hemisphere and to a lesser
extent a shorter section of nearly the same arc near the equator.
While the majority of LIB
ʼ
s probability is concentrated in the
Southern hemisphere, a non-trivial fraction of the 90%
con
fi
dence region extends into the northern hemisphere. The
LALInference sky map shows much less support in the
northern hemisphere which is likely associated with the
stronger constraints available with full CBC waveforms. The
cWB localization also supports an isolated hot spot near
α
∼
9
h
,
δ
∼
5
°
, where the detector responses make it possible to
independently measure two polarization components. In this
region, cWB considers signals not constrained to have the
elliptical polarization expected from a compact binary merger.
Quantitative comparisons of the four sky maps can be found
in Section
2
of the Supplement
(
Abbott et al.
2016b
)
. The main
feature in all of the maps is an annulus with polar angle
θ
HL
determined by the arrival time difference
Δ
t
HL
between the two
detectors. However, re
fi
nements are possible due to phase as
well as amplitude consistency and the mildly directional
antenna patterns of the LIGO detectors
(
Kasliwal & Nissanke
2014
; Singer et al.
2014
)
. In particular, the detectors
’
antenna
Figure 3.
Footprints of observations in comparison with the 50% and 90% credible levels of the initially distributed GW localization maps. Radio
fi
elds are shaded in
red, optical
/
infrared
fi
elds are in green, and the XRT
fi
elds are indicated by the blue circles. The all-sky
Fermi
GBM, LAT,
INTEGRAL
SPI-ACS, and MAXI
observations are not shown. Where
fi
elds overlap, the shading is darker. The initial cWB localization is shown as thin black contour lines and the LIB localization as
thick black lines. The inset highlights the
Swift
observations consisting of a hexagonal grid and a selection of the a posteriori most highly ranked galaxies. The
Schlegel et al.
(
1998
)
reddening map is shown in the background to represent the Galactic plane. The projection is the same as in Figure
2
.
4
The Astrophysical Journal Letters,
826:L13
(
8pp
)
, 2016 July 20
Abbott et al.
patterns dominate the modulation around the ring for
unmodeled reconstructions through a correlation with the
inferred distance of the source
(
Essick et al.
2015
)
. As shown in
Figure
2
, the algorithms all infer polar angles that are consistent
at the 1
σ
level.
4. FOLLOW-UP OBSERVATIONS
Twenty-
fi
ve participating teams of observers responded to
the GW alert to mobilize satellites and ground-based telescopes
spanning 19 orders of magnitude in EM wavelength. Observa-
tions and archival analysis started shortly after the candidate
was announced, two days after the event was recorded. Most
facilities followed tiling strategies based on the cWB and LIB
sky maps. Some groups, considering the possibility of a NS
merger or core-collapse SN, selected
fi
elds based on the areal
density of nearby galaxies or targeted the Large Magellanic
Cloud
(
LMC
)(
e.g., Annis et al.
2016
)
. Had the BBH nature of
the signal been promptly available, most groups would not
have favored local galaxies because LIGO
ʼ
s range for BBH
mergers is many times larger than that for BNSs. Figure
3
displays the footprints of all reported observations. The
campaign is summarized in Table
1
in terms of instruments,
depth, time, and sky coverage. Some optical candidate
counterparts were followed up spectroscopically and in the
radio band as summarized in Table
2
. The overall EM follow-
up of GW150914 consisting of broadband tiled observations
Table 1
Summary of Tiled Observations
Facility
/
Area
Contained Probability
(
%
)
Instrument
Band
a
Depth
b
Time
c
(
deg
2
)
cWB LIB BSTR
d
LALInf GCN
Gamma-ray
Fermi
LAT
20 MeV
–
300 GeV
1.7
×
10
−
9
(
every
3hr
)
L
100 100
100
100
18709
Fermi
GBM
8 keV
–
40 MeV
0.7
–
5
×
10
−
7
(
0.1
–
1 MeV
)
(
archival
)
L
100 100
100
100
18339
INTEGRAL
75 keV
–
1 MeV
1.3
×
10
−
7
(
archival
)
L
100 100
100
100
18354
IPN
15 keV
–
10 MeV
1
×
10
−
7
(
archival
)
L
100 100
100
100
L
X-ray
MAXI
/
GSC
2
–
20 keV
1
×
10
−
9
(
archival
)
17900
95
89
92
84
19013
Swift
XRT
0.3
–
10 keV
5
×
10
−
13
(
gal.
)
2.3, 1, 1
0.6 0.03 0.18
0.04
0.05
18331
2
–
4
×
10
−
12
(
LMC
)
3.4, 1, 1
4.1
1.2
1.9
0.16
0.26
18346
Optical
e
DECam
i
,
zi
<
22.5,
z
<
21.5
3.9, 5, 22
100
38
14
14
11
18344
,
18350
iPTF
RR
<
20.4
3.1, 3, 1
130
2.8
2.5
0.0
0.2
18337
KWFC
ii
<
18.8
3.4, 1, 1
24
0.0
1.2
0.0
0.1
18361
MASTER
C
<
19.9
−
1.1, 7, 7
710
50
36
55
50
18333
,
18390
,
18903
,
19021
Pan-STARRS1
ii
<
19.2
−
20.8
3.2, 21, 42
430
28
29
2.0
4.2
18335
,
18343
,
18362
,
18394
La Silla
–
QUEST
g
,
rr
<
21
3.8, 5, 0.1
80
23
16
6.2
5.7
18347
SkyMapper
i
,
vi
<
19.1,
v
<
17.1
2.4, 2, 3
30
9.1
7.9
1.5
1.9
18349
Swift
UVOT
uu
<
19.8
(
gal.
)
2.3, 1, 1
3
0.7
1.0
0.1
0.1
18331
uu
<
18.8
(
LMC
)
3.4, 1, 1
18346
TAROT
C
R
<
18
2.8, 5, 14
30
15
3.5
1.6
1.9
18332
,
18348
TOROS
C
r
<
21
2.5, 7, 90
0.6 0.03 0.0
0.0
0.0
18338
VST@ESO
rr
<
22.4
2.9, 6, 50
90
29
10
14
10
18336
,
18397
Near Infrared
VISTA@ESO
Y
,
J
,
K
S
J
<
20.7
4.8, 1, 7
70
15
6.4
10
8.0
18353
Radio
ASKAP
863.5 MHz
5
–
15 mJy
7.5, 2, 6
270
82
28
44
27
18363
,
18655
LOFAR
145 MHz
12.5 mJy
6.8, 3, 90
100
27
1.3
0.0
0.1
18364
,
18424
,
18690
MWA
118 MHz
200 mJy
3.5, 2, 8
2800
97
72
86
86
18345
Notes.
a
Band: photon energy, optical or near-infrared
fi
lter
(
or C for clear un
fi
ltered light
)
, wavelength range, or central frequency.
b
Depth: gamma
/
X-ray limiting
fl
ux in erg cm
−
2
s
−
1
;5
σ
optical
/
IR limiting magnitude
(
AB
)
; and 5
σ
radio limiting spectral
fl
ux density in mJy. The reported values
correspond to the faintest
fl
ux
/
magnitude of detectable sources in the images.
c
Elapsed time in days between start of observations and the time of GW150914
(
2015 September 14 09:50:45
)
, number of repeated observations of the same area, and
total observation period in days.
d
BAYESTAR.
e
Searches for bright optical transients were also done by BOOTES-3 and Pi of the Sky. Details are given in the Supplement
(
Abbott et al.
2016b
)
.
5
The Astrophysical Journal Letters,
826:L13
(
8pp
)
, 2016 July 20
Abbott et al.