of 24
All-sky search for short gravitational-wave bursts in the third Advanced
LIGO and Advanced Virgo run
R. Abbott
etal.
*
(The LIGO Scientific Collaboration, the Virgo Collaboration, and the KAGRA Collaboration)
(Received 8 July 2021; accepted 12 October 2021; published 23 December 2021)
This paper presents the results of a search for generic short-duration gravitational-wave transients in data
from the third observing run of Advanced LIGO and Advanced Virgo. Transients with durations of
milliseconds to a few seconds in the 24
4096 Hz frequency band are targeted by the search, with no
assumptions made regarding the incoming signal direction, polarization, or morphology. Gravitational
waves from compact binary coalescences that have been identified by other targeted analyses are detected,
but no statistically significant evidence for other gravitational wave bursts is found. Sensitivities to a variety
of signals are presented. These include updated upper limits on the source rate density as a function of the
characteristic frequency of the signal, which are roughly an order of magnitude better than previous upper
limits. This search is sensitive to sources radiating as little as
10
10
M
c
2
in gravitational waves at
70
Hz from a distance of 10 kpc, with 50% detection efficiency at a false alarm rate of one per century.
The sensitivity of this search to two plausible astrophysical sources is estimated: neutron star
f
modes,
which may be excited by pulsar glitches, as well as selected core-collapse supernova models.
DOI:
10.1103/PhysRevD.104.122004
I. INTRODUCTION
The third observing run (O3) of the Advanced LIGO
[1]
and Advanced Virgo
[2]
detectors started on April 1, 2019,
and ended on March 27, 2020. During O3, tens of
gravitational waves (GWs) from compact binary coalescence
(CBC) were detected
[3
6]
. In addition to CBCs, there are
several plausible sources of short-duration GW transients
(GW bursts) that have not yet been observed, such as core-
collapse supernovae (CCSNe), neutron star excitations,
nonlinear memory effects, or cosmic string cusps and kinks
[7
11]
. Additional source populations could exist that are
yet to be predicted. For these reasons, GW burst searches
capable of detecting a wide range of signal waveforms
provide a unique opportunity for new discoveries.
All-sky searches look for signals arriving at any time from
any sky direction. GW searches may use signal models
(targeted search) or remain agnostic about the signal
morphology (generic search). Targeted analyses include
searches for CBCs
[3
5,12]
and cosmic strings
[11]
.
Generic all-sky searches look for short-duration GW tran-
sients, up to a few seconds duration
[13]
, and for longer GW
transients, up to
10
3
s duration
[14]
.
This paper presents results of the generic all-sky search
that is sensitive to the widest range of morphologies of short
duration GW bursts during O3. The generic all-sky search is
also sensitive to some CBC events
[13]
, but these are not the
primary targets of this analysis, and details of CBC
detections during O3 are given elsewhere
[3,4]
.Oncethe
CBC candidates are excised, this search produces a null
result.
This null result is interpreted in terms of sensitivities to a
wide variety of generic morphologies, similarly to what was
done in previous observing runs, O1
[15]
and O2
[13]
.The
current analysis improves on previous upper limits due to
improvements in detector sensitivity and a longer observa-
tion run. In addition, this paper includes the interpretation of
results in terms of two expected astrophysical sources:
CCSNe and neutron star
f
modes. Since no tuning of the
generic all-sky search is performed, these results should be
considered conservative. The sensitivity of the search to five
CCSNe waveform models is presented, both versus distance
and for a Galactic distribution of sources. GWemission from
CCSNe is expected in the frequency band below 1 kHz.
Neutron star
f
modes may be excited by pulsar glitches and
are expected to emit GWs in the frequency range 2
3kHz.
The search sensitivity is tested for two equations of state and
masses in the range
1
2
M
.
The analyses described here use the final LIGO-Virgo
calibration
[16
19]
and data quality
[20]
information and
their results supersede those from searches for GW bursts
that were deployed in low latency during O3. The latter
provided prompt public alerts for follow-up observations by
other telescopes
[21]
, analyzing near real-time data streams
with preliminary calibration and data quality information.
The rest of this paper is organized as follows: Sec.
II
reviews the dataset used for these analyses. Section
III
describes the two search algorithms deployed and their
*
Full author list given at the end of the article.
PHYSICAL REVIEW D
104,
122004 (2021)
2470-0010
=
2021
=
104(12)
=
122004(24)
122004-1
© 2021 American Physical Society
results (
III A
), and discusses the loudest candidate events
remaining after excluding the recognized CBC candidates
(
III B
). Section
IV
discusses the sensitivity of this all-sky
search and sets new rate-density limits for transient GW
events other than CBC, as well as the sensitivity to CCSN
modelsandtoneutronstar
f
modes. Finally, Sec.
V
summ-
arizes the results and implications from this minimally
modeled search for GW transients.
II. O3: THE THIRD ADVANCED-DETECTOR
OBSERVING RUN
A. Dataset
The O3 dataset extends from April 1, 2019 to March 27,
2020. A commissioning break between October 1, 2019
and November 1, 2019 separates the first six-month epoch
(O3a) of the observing run from the second epoch (O3b).
Figure
1
shows typical spectral sensitivities of the detec-
tors. The Hanford-Livingston (HL) network is analyzed
during times where these two detectors operated in coinci-
dence. In addition, results for the Hanford-Virgo network
(HV) and the Livingston-Virgo network (LV) are presented
for times when data from either of the LIGO detectors is not
available. See Sec.
III
for an explanation of why the two
detector network is preferred over the three detector
Hanford-Livingston-Virgo (HLV) network for this search.
During the six months of O3a, 130.2 days of data were
collected at Hanford, 138.5 days of data were collected at
Livingston, and 139.5 days of data were collected at Virgo.
The amount of data actually analyzed is reduced by requiring
coincidence between two detectors, removing poor periods
of data quality as described in Sec.
II B
, and requiring at least
200 seconds of continuous observation-quality data. This
results in the following total amounts of analyzed data:
104.9 days for the HL network, 14.8 days for the HV, and
25.6 days for the LV network.
During O3b, data were collected for 115.7 days at
Hanford, for 115.5 days at Livingston, and for 113.2 days
at Virgo. The actual analyzed data amounts are 93.4 days
for HL, 17.8 days for HV, and 14.8 days for the LV
network.
The calibration uncertainties for the LIGO detectors in
the 20
2000 Hz frequency range are
<
7%
in amplitude,
<
4
°inphase,
<
1
μ
s in timing for O3a
[16]
,and
<
12%
in amplitude,
<
10
° in phase,
<
1
μ
sintimingforO3b
[17]
. The calibration uncertainties for Virgo in most of the
20
2000 Hz frequency range are
<
5%
in amplitude,
<
2
°
in phase, and
<
10
μ
s in timing for all of O3
[18,19]
.
These uncertainties are not expected to have a significant
impact on the search presented here. However, they can
contribute to the systematic uncertainties associated with
the efficiency numbers quoted in Sec.
IV
.TheO3aGW
strain data used in this paper are part of the O3a data
release through the Gravitational Wave Open Science
Center
[22]
, and can be found at
[23]
.
B. Data quality
The LIGO and Virgo detectors are affected by various
sources of terrestrial noise that can interfere with the
detection of GWs
[24,25]
. In addition to the primary channel
recording GWs, the interferometers use a large number of
auxiliary channels that observe either the external environ-
ment
[26,27]
, or the interferometer itself. Through the use of
auxiliary channels, it is possible to substantially reduce the
impact of noise transients by discarding (vetoing) a small
percentage of observing time during which noise contami-
nation is present
[28]
. A brief discussion of some of the most
relevant data quality issues is presented in this section, but
much more detail on these issues and their mitigation can be
found in Ref.
[20]
.
To address specific data quality problems, tens of
different data quality vetoes defining times to be removed
from the search are constructed and applied to the
analyses described in this paper. The most significant
data quality issues successfully discarded by these vetoes
are high signal-to-noise ratio (SNR) glitches associated
with light intensity dips in both LIGO interferometers,
radio frequency beatnotes (also known as whistles), and a
single half-hour period of high amplitude violin mode
resonances of the LIGO Hanford suspension system. The
effectiveness of each data quality veto is determined based
on the ratio of the percentage of glitches rem
oved to
amount of observation time vetoed. An additional stage of
automated statistical vetoes using the hveto
[29]
algorithm
is subsequently applied using the same procedure as in O2
[13]
. Hveto uses a hierarchical method to produce a
ranked list of statistically significant vetoes generated
by applying a specific list of SNR thresholds and time
windows to a subset of LIGO
s auxiliary channels.
FIG. 1. Representative amplitude spectral density of the three
detectors
strain sensitivity (LIGO Livingston 5 September 2019
20
53 UTC, LIGO Hanford 29 April 2019 11
47 UTC, Virgo 10
April 2019 00
34 UTC).
R. ABBOTT
et al.
PHYS. REV. D
104,
122004 (2021)
122004-2
Between 1% and 2% of the total observation time per
interferometer is discarded due to data quality issues, with
precise breakdowns provided in Ref.
[20]
. A complete list
of vetoes used in this search with brief descriptions of
eachisgiveninRef.
[30]
.
Unfortunately, these vetoes do not suppress all nonas-
trophysical features of the data. As interferometer sensitivity
has improved, light scattering has become more prominent at
low frequencies
[31,32]
. Scattering noise was significantly
reduced in the latter part of the run, but it remained a
prominent feature throughout much of O3, especially during
periods of high anthropogenic or seismic activity. Because of
the large amount of time with light scattering present and the
lack of straightforward and consistent auxiliary channel
witnesses, most light-scattering glitches are not vetoed.
Another prominent noise type that is not vetoed by
standard methods are blip glitches
[33]
. These have
recurred in both LIGO interferometers throughout the
advanced detector era. Blips are short-duration, low
quality factor (
Q
) glitches which occur at the rate of
several per day. As these blips do not have clear auxiliary
witnesses or known origin, and are not clearly morpho-
logically distinct from some astrophysical models of
interest, they must be handled by the search algorithms
themselves. During O3 a new population of loud single-
pulse bliplike glitches was observed. The origin of these
glitches is not known. See Sec.
III A 1
for more details on
the handling of this glitch class.
III. UNMODELED GW TRANSIENT ANALYSES
Using the three-detector HL
V network generally ena-
bles a more accurate reconstruction of both the structure
of the GW signal and its sky location than is possible with
a two-detector network. However, for purposes of detec-
tion, the sensitivity of the HLV network is not better than
the HL network for the O3 analyses described in this
paper. The generic all-sky search for GW bursts cannot
rely on assumptions about the GW polarization state.
Since the two LIGO interferometers are nearly coaligned
and therefore sensitive to similar linear combinations of
the GW polarization components over most sky direc-
tions, Hanford and Livingston generally detect a given
GW with comparable SNRs. Virgo, by contrast, typically
senses a different linear combination of GW polariza-
tions. In O3 the LIGO interferometers have better
sensitivity than Virgo (see Fig.
1
), and for many source
directions the difference in detector orientation enhances
this disparity.
In addition, there is a negligible loss in detection
efficiency when narrowing the analysis of HL-only data
to search for the GW polarization projection that best
matches the network from each sky direction. This allows
us to implement stricter requirements on the signal coher-
ence between the Hanford and Livingston detectors and
results in a more effective rejection of noise transients.
This advantage is not shared by analyses of networks
involving Virgo due to its misalignment with the LIGO
detectors. To make full use of Virgo data, the analysis has to
either open the search to both GW polarization components
over the sky, or relax the requirements on the signal
coherence between participating detectors. The distribution
of non-Gaussian noise outliers present in all detectors in
O3 is thus more effectively mitigated in coherent analyses of
the HL network than in analyses with networks including
Virgo, and this affects the resulting detection efficiency.
The analyses described in this paper therefore use the HL
network rather than HLV because we are interested in
maximizing detection probability.
The search for short GW bursts is sensitive to CBC
sources, especially binary black hole coalescences
[13]
,
and hence a fraction of them are found by the analyses
presented here. The discussion of the astrophysical proper-
ties and implications of the detected CBC events is the
subject of other LIGO Scientific and Virgo Collaboration
catalog papers (see Ref.
[3]
for O3a results). Search results
in this paper initially include GWs from CBCs, but known
CBC events are excised in a subsequent step, and dis-
cussion here is limited to candidate events that are not
found by targeted searches for such sources.
A. Search algorithms
In order to make the results of the search more robust, two
independently developed search algorithms are deployed:
coherent WaveBurst (CWB) and BayesWave (BW). The
CWB algorithm is used to analyze the entire dataset. BW is
computationally more intensive, thus it is only used to follow
up a subset of the dataset identified by CWB in order to
provide a partly independent measurement of the candidates
significance. Both of these algorithms and their results are
described below.
1. Coherent WaveBurst
Coherent WaveBurst is an algorithm based on the
maximum-likelihood-ratio statistic over all sky directions
applied to excesses of signal power in the time-frequency
domain representation of the strain data from the network
of detectors
[34
36]
. The analysis uses the Wilson-
Daubechies-Meyer wavelet transform at various time-
frequency resolutions
[37]
. Multiple resolutions allow
adaptation of time-frequency characterization to the signal
features. Coherent WaveBurst is routinely used in LIGO-
Virgo searches and reconstruction of GW transients
[13,15]
.
In this work the low and high frequency parts of the
parameter space are separately covered by two analyses.
The same procedure was also done for O1
[15]
and O2
[13]
.
The clusters of wavelets which fall above the noise floor
of the detectors and pass the internal thresholds of the
pipeline are referred as
triggers
.
The low-frequency analysis covers the frequency range
between 16 and 1024 Hz. Triggers with mean reconstructed
ALL-SKY SEARCH FOR SHORT GRAVITATIONAL-WAVE
...
PHYS. REV. D
104,
122004 (2021)
122004-3
frequency below 24 and 32 Hz are rejected for O3a and
O3b, respectively, to avoid contamination from loud and
frequent low-frequency glitches. The HV and LV networks
are analyzed only when one of the LIGO interferometers is
unavailable, i.e., there is no overlap in time of dataset for
any of the networks considered and the live times for each
network are mutually exclusive.
The requirement on the signal coherence across detectors
is implemented as a threshold on the network correlation
coefficient (referred to as
c
c
in Ref.
[34]
), which is the
fraction of coherent energy in the network of detectors.
Triggers are required to pass
c
c
thresholds of 0.8 for the
analysis of the HL network and 0.5 for the HV and LV
networks, since Virgo is not coaligned with the two LIGO
detectors. These thresholds were chosen to optimize overall
performance over the set of signal models listed in Table
I
The triggers obtained after passing the frequency rejection
and network correlation coefficient threshold are further
divided into three different, mutually exclusive bins, referred
to as LF1, LF2, and LF3. The choice of the bins is based on
the background triggers
morphologies, and the goal is to
isolate background triggers that are loud and frequent to a
small part of the parameter space. LF1 contains triggers with
most of the signal energy confined to a single oscillation.
In O3 a population of such short-duration blip glitches
dominates the tail of the background trigger distribution
and hence the LF1 bin is introduced in the O3 search to
confine these glitches (see Sec.
II B
). LF2 contains the
remaining triggers that are characterized by
Q
3
,also
resembling blip glitches. LF3 contains the higher
Q
low-
frequency triggers and shows the cleanest background
distribution. Unlike O1 and O2, nonstationary spectral lines
do not contribute significantly to the background in O3.
The low-frequency CWB analysis is performed separately
for O3a and O3b. The background distribution of triggers is
calculated by time-shifting the data of one detector with
respect to the other detector by an amount that breaks any
correlation between detectors for a real signal. The HL
networkistime-shiftedtoobtainatotalbackgroundlivetime
of around 2000 years. For the HV and LV networks, around
1000 years of background are generated using all coincident
data. The use of full coincident time for the HV and LV
networks is necessary because the exclusive livetime is not
sufficient to produce such large background statistics.
The high-frequency analysis covers the frequency range
1024
4096 Hz. The analysis is carried out in the frequency
band 512
4096 Hz but only triggers with mean reconstructed
frequencies
1
kHz are kept. For this analysis only the HL
network is considered, as Virgo is significantly less sensitive
than the LIGO interferometers in the high-frequency region
(a factor of
5
above 1000 Hz, see Fig.
1
). Similarly to the
low-frequency analysis, a network correlation coefficient
threshold of 0.8 is used for the high-frequency part of the
analysis. No division of background triggers into analysis
bins is required for this analysis. However, during the first
part of O3 run until May 16, 2019, there were background
triggers dominating the tail with central frequency
f
0
>
3400
Hz; for this part of the run only the triggers with
central frequency
3400
Hz are admitted. The full fre-
quency range is considered for all times from May 16
onward. As a result, the high-frequency CWB analysis is
divided into three parts, the first two parts are in O3a (before
and after May 16, 2019, see above), and the third part
corresponds to all of O3b. Total background live times of
around 1000 years are produced for O3.
The significance of each trigger is calculated by comparing
the coherent network SNR
η
c
[34]
with the background
distribution of the bin and the network to which the trigger
belongs. The
inverse false alarm rate
(IFAR) is calculated for
each observed trigger. The IFAR for the low-frequency
analysis is penalized by a trials factor of 3 corresponding
to the three analysis bins LF1, LF2, and LF3. The criteria for
a significant detection of an event is set at IFAR
100
years.
The analysis results for the CWB low-frequency region are
showninFig.
2
. The loudest candidate event in the HL
network after excluding known CBCs
[3]
occurred at UTC
2019-09-28 02
11
45. This candidate has an IFAR of
0.53 years. The second most significant candidate in this
network occurred at UTC time 2019-08-04 08
35
43, with an
IFAR of 0.19 years. The loudest candidate for the HVand LV
TABLE I. The
h
rss
values (in units of
10
22
Hz
1
=
2
) for which
50% detection efficiency is achieved with an IFAR of 100 years
for each of the injected signal morphologies. The SG waveforms
reported in this table have circular polarization:
>
40
indicates
that 50% detection efficiency is not achieved for the maximum
h
rss
used in this injection set, and
-
denotes waveforms not
analyzed by BW.
h
rss
10
22
Hz
1
=
2
)
O3a
O3b
Morphology
cWB BW cWB BW
Gaussian pulses (linear)
τ
GA
¼
0
.
1
ms
18.1 - 8.2 -
τ
GA
¼
2
.
5
ms
25.2 - 10.5 -
Sine-Gaussian wavelets (circular)
f
0
¼
70
Hz,
Q
¼
3
1.1
>
40
1.1
>
40
f
0
¼
70
Hz,
Q
¼
100
1.0
>
40
1.0
>
40
f
0
¼
235
Hz,
Q
¼
100
0.8 2.5 0.8 3.7
f
0
¼
554
Hz,
Q
¼
8
.
9
1.0
>
40
1.1
>
40
f
0
¼
849
Hz,
Q
¼
3
1.5
>
40
1.6
>
40
f
0
¼
1304
Hz,
Q
¼
9
1.9 - 1.9 -
f
0
¼
1615
Hz,
Q
¼
100
2.2 - 2.4 -
f
0
¼
2000
Hz,
Q
¼
3
3.2 - 3.1 -
f
0
¼
2477
Hz,
Q
¼
8
.
9
3.8 - 3.7 -
f
0
¼
3067
Hz,
Q
¼
3
5.6 - 5.0 -
White-noise bursts
f
low
¼
100
Hz,
Δ
f
¼
100
Hz,
τ
WNB
¼
0
.
1
s 0.9 2.6 1.0 3.4
f
low
¼
250
Hz,
Δ
f
¼
100
Hz,
τ
WNB
¼
0
.
1
s 0.9 2.2 1.0 3.5
f
0
¼
750
Hz,
Δ
f
¼
100
Hz,
τ
WNB
¼
0
.
1
s 1.5 2.8 1.5 3.9
R. ABBOTT
et al.
PHYS. REV. D
104,
122004 (2021)
122004-4
networks is an HV event at UTC time 2019-04-30 00
49
32,
with an IFAR of 12 years. Though none of these meet the
IFAR threshold of 100 years for a detection, investigations
into these loudest remaining candidates are discussed further
in Sec.
III B
.
The results for the high-frequency CWB analysis are
shown in Fig.
3
, the loudest event has an IFAR of 0.3 years.
2. BayesWave
BW
[38
40]
is a Bayesian algorithm modeling both
GW signals and non-Gaussian noise transients as sums of
sine-Gaussian wavelets. The number of wavelets used is
marginalized over using a transdimensional reversible jump
Markov chain Monte Carlo algorithm. The detection statistic
used is the natural logarithm of the signal-to-glitch Bayes
factor (ln
B
S
;
G
), i.e., the Bayes factor between the signal
model consisting of Gaussian noise and an astrophysical
signal coherent across detectors; and the glitch model, which
describes the data as Gaussian noise and glitches modeled
independently in each detector. Thus a positive ln
B
S
;
G
indicates that the presence of a GW signal is favored, while
anegativeln
B
S
;
G
shows support for the event being a glitch.
Due to the transdimensional sampling it requires, ana-
lyzing the entire O3 dataset with BW is computationally
prohibitive. Thus BW is used as a follow-up to the CWB
pipeline, similarly to previous observing runs
[13,15]
.By
doing so an additional measurement of IFAR for the
candidates followed up by BW is acquired, thus making
the search presented in this paper more robust against
potential shortcomings of individual algorithms. BW fol-
lowed up CWB candidates in the low-frequency analysis,
treating all the search bins as a single bin, and using a
threshold of
η
c
¼
9
.
90
. This is the same threshold as the
one used in the O2 analysis
[13]
, and was chosen as the
lowest value that results in a computationally manageable
number of background triggers. BW uses the same back-
ground dataset as CWB from time slides.
A total of 22 CWB candidates are above the
η
c
threshold,
19 of which are known CBC candidate events described in
recent or future publications. This is fewer than found by
CWB, because not all CBC candidates passed the BW
follow-up threshold. The combined results from all detector
networks are shown in Fig.
4
in terms of the cumulative
distribution of their IFAR values. The three candidate
events remaining after removing the known CBC candidate
events are discussed in the previous section. None of these
is found with high enough significance in BW to be
considered a likely GW event. Section
III B
discusses these
candidate burst events.
B. Candidate events
1. Surviving non-CBC candidates
The three non-CBC candidate events with
η
c
values
above 9.90, a high enough value to trigger BW follow-up,
are discussed individually below. They are identified by the
FIG. 2. Cumulative number of events versus IFAR found by the
CWB low-frequency analysis using all O3 data for the HL
network (top panel), and the HV and LV networks combined
(bottom panel). Circular points show results for all data and
triangular points show after times around all known compact
binary coalescence sources have been excised. The solid line
shows the expected mean value of the background, given the
analyzed time. The shaded regions show the 1, 2, and
3
σ
Poisson
uncertainty regions.
FIG. 3. Cumulative number of events versus IFAR found by the
CWB high-frequency analysis (triangular points) using all O3
data for the HL network (Virgo is not used for high-frequency
search). The solid line shows the expected mean value of the
background, given the analyzed time. The shaded regions show
the 1, 2, and
3
σ
Poisson uncertainty regions.
ALL-SKY SEARCH FOR SHORT GRAVITATIONAL-WAVE
...
PHYS. REV. D
104,
122004 (2021)
122004-5
time at which they occurred. In each case, the statistical
significance is not high enough to claim the candidate as a
GW event. Though none of these candidates are vetoed by
data quality procedures, data quality concerns for each case
are discussed.
2019-09-28 02
11
45 UTC: the most significant HL CWB
candidate has an inverse false alarm rate of 0.53 years in a
CWB all-sky search and 0.8 years in the BW follow-up. This
initially appeared in the low-latency CBC-focused CWB
analysis but was manually rejected in near-real time as most
probably being caused by a glitch in the Livingston detector
[41]
. It does not pass signal consistency cuts specific to the
version of that search focused on CBCs, described in
Ref.
[3]
, but remains in the more general burst analysis at
lower significance. Instrumental investigations into possible
origins focused on magnetic noise in the station at the end of
Hanford
s X arm, but magnetic coupling was ruled out as a
significant contributor to the power of the putative signal.
The morphology in the Livingston detector resembles Tomte
glitches
[24,42]
appearing at other times, while there is little
power in the Hanford detector. The significant difference in
the relative amplitude between Hanford and Livingston
would mean that, if astrophysical, this candidate event
would have to originate from the
5%
of the sky where
Hanford has negligible sensitivity but Livingston
ssensi-
tivity is significant.
2019-08-04 08
35
43 UTC: the second most significant
low-frequency HL CWB candidate, at an IFAR of 0.19 years,
was also initially identified in a low-latency CWB targeted
search for binary black hole coalescences, but it did not meet
the significance threshold to generate a public alert. BW
follow-up finds an IFAR of 12.2 years, making this the
most significant non-CBC outlier in that analysis. It occurred
less than a second after an SNR
60
series of glitches
in Livingston, which are themselves too loud to be
astrophysical. These glitches morphologically resemble
the repeating blips class of glitches
[42]
occurring at other
times in both LIGO interferometers. Its close proximity to
these glitches makes it impossible to discount an instru-
mental origin, though it is not vetoed by any auxiliary
witness channel. As a follow-up study to the low-latency
search, BW was used to model the glitches occurring just
before the candidate event, and that model was subtracted
from the data in order to produce a deglitched data stream
[40]
. It was found that this glitch subtraction lowered the
SNR but had negligible effect on the reconstructed mor-
phology of the candidate.
2019-04-30 00
49
32 UTC: an additional candidate is
identified in the HV O3a CWB search, a less sensitive
network than HL, at an IFAR of 12.29 years. It was not
identified as a trigger of interest in the low-latency search.
The BW follow-up gives an IFAR of 2.4 years for this
trigger. The presence of blip glitches in Hanford less than a
second prior to the candidate and the resemblance to a blip
glitch in the Hanford interferometer lead to similar data
quality concerns as the previous trigger.
2. Low-latency-only candidates
In the low-latency search described in Sec.
I
, public
alerts were generated for burst search candidates with
significance exceeding an IFAR of 4 years. Two candidate
events crossed this significance threshold in the low-latency
CWB search but do not appear in the version of the analysis
presented in this paper, as explained below.
S191110af: this was a high-frequency (
1780
Hz) HL
CWB candidate that generated a public alert
[43]
basedonits
significance in the low-latency CWB analysis. Follow-up of
the candidate shortly after it was identified indicated that it
was due to a faulty piezoelectric transducer at Hanford. This
candidate event does not appear in the analysis described in
this paper, as times strongly affected by this noise were
vetoed
[20]
.Itisnolongerofastrophysicalinterest.
S200114f: this HL candidate generated a public alert
[44]
based on its significance in the low-latency unmod-
eled CWB all-sky search, but it is not found in the analysis
as described in this paper because it fails an internal CWB
consistency cut (the network correlation coefficient
c
c
<
0
.
8
,seeSec.
III A 1
) requiring the signal to be
correlated between the two LIGO detectors. It is further
discussed in the O3 intermediate mass black hole search
paper
[4]
.
IV. ASTROPHYSICAL INTERPRETATION
OF THE RESULTS
In order to place the search results in an astrophysical
context, it is necessary to measure detection efficiency for
plausible signals. This is accomplished by injecting simulated
signals (via software) into the detector data and recovering
them using the search methods described in previous
FIG. 4. Cumulative number of events versus IFAR found by the
BW follow-up to the CWB low-frequency analyses using all O3
data (circular points) and after times around all compact binary
coalescence sources have been excised (triangular points). The
solid line shows the expected background, given the analyzed
time. The shaded regions show the 1, 2, and
3
σ
Poisson
uncertainty regions.
R. ABBOTT
et al.
PHYS. REV. D
104,
122004 (2021)
122004-6