First demonstration of early warning gravitational wave alerts
Ryan Magee, Deep Chatterjee, Leo P. Singer, Surabhi Sachdev, Manoj Kovalam, Geoffrey Mo, Stuart Anderson, Patrick Brady, Patrick Brockill, Kipp Cannon, Tito Dal Canton, Qi Chu, Patrick Clearwater, Alex Codoreanu, Marco Drago, Patrick Godwin, Shaon Ghosh, Giuseppe Greco, Chad Hanna, Shasvath J. Kapadia, Erik Katsavounidis, Victor Oloworaran, Alexander E. Pace, Fiona Panther, Anwarul Patwary, Roberto De Pietri, Brandon Piotrzkowski, Tanner Prestegard, Luca Rei, Anala K. Sreekumar, Marek J. Szczepa?czyk, Vinaya Valsan, Aaron Viets, Madeline Wade, Linqing Wen, John Zweizig
DDraft version February 15, 2021
Typeset using L A TEX preprint2 style in AASTeX63
First demonstration of early warning gravitational wave alerts
Ryan Magee,
1, 2, 3
Deep Chatterjee,
4, 5, 6
Leo P. Singer And Surabhi Sachdev
2, 3
These authors contributed equally to this work—Manoj Kovalam,
8, 9
Geoffrey Mo,
10, 11
Stuart Anderson, Patrick Brady, Patrick Brockill, Kipp Cannon, Tito Dal Canton, Qi Chu,
8, 9
Patrick Clearwater,
8, 14
Alex Codoreanu,
8, 14
Marco Drago,
15, 16
Patrick Godwin,
2, 3
Shaon Ghosh, Giuseppe Greco,
18, 19
Chad Hanna,
2, 3, 20
Shasvath J. Kapadia, Erik Katsavounidis,
10, 11
Victor Oloworaran,
8, 9
Alexander E. Pace,
2, 3
Fiona Panther,
8, 9
Anwarul Patwary,
8, 9
Roberto De Pietri,
22, 23
Brandon Piotrzkowski, Tanner Prestegard, Luca Rei, Anala K. Sreekumar,
8, 9
Marek J. Szczepa´nczyk, Vinaya Valsan, Aaron Viets, Madeline Wade, Linqing Wen
8, 9
And John Zweizig LIGO, California Institute of Technology, Pasadena, CA 91125, USA Department of Physics, The Pennsylvania State University, University Park, PA 16802, USA Institute for Gravitation and the Cosmos, The Pennsylvania State University, University Park, PA 16802, USA Center for Astrophysical Surveys, National Center for Supercomputing Applications, Urbana, IL, 61801, USA Illinois Center for Advanced Studies of the Universe, Department of Physics, University of Illinois atUrbana-Champaign, Urbana, IL 61801, USA Leonard E. Parker Center for Gravitation, Cosmology, and Astrophysics, University of Wisconsin-Milwaukee,Milwaukee, WI 53201, USA Astroparticle Physics Laboratory, NASA Goddard Space Flight Center, Mail Code 661, Greenbelt, MD 20771, USA Australian Research Council Centre of Excellence for Gravitational Wave Discovery (OzGrav) Department of Physics, University of Western Australia, Crawley WA 6009, Australia LIGO Laboratory, Massachusetts Institute of Technology, Cambridge, MA 02139, USA Department of Physics and Kavli Institute for Astrophysics and Space Research, Massachusetts Institute ofTechnology, Cambridge, MA 02139, USA Research Center for the Early Universe, The University of Tokyo, 113-0033, Japan Universit´e Paris-Saclay, CNRS/IN2P3, IJCLab, 91405 Orsay, France Gravitational Wave Data Centre, Swinburne University, Hawthorn VIC 3122, Australia Universit`a di Roma “La Sapienza,” I-00185 Roma, Italy INFN, Sezione di Roma, I-00185 Roma Department of Physics and Astronomy, Montclair State University, Montclair, NJ, 07043 Universit`a degli Studi di Urbino ”Carlo Bo”, I-61029 Urbino, Italy INFN, Sezione di Firenze, I-50019 Sesto Fiorentino, Firenze, Italy Department of Astronomy and Astrophysics, The Pennsylvania State University, University Park, PA 16802, USA International Centre for Theoretical Sciences, Tata Institute of Fundamental Research, Bangalore 560089, India Dipartimento di Scienze Matematiche, Fisiche e Informatiche, Universit`a di Parma, I-43124 Parma, Italy INFN, Sezione di Genova, I-16146 Genova, Italy University of Florida, Gainesville, FL 32611, USA Department of Physics, Kenyon College, Gambier, OH 43022, USA a r X i v : . [ a s t r o - ph . H E ] F e b Magee, Chatterjee, Singer, Sachdev, et al.
ABSTRACTGravitational-wave observations became commonplace in Advanced LIGO-Virgo’s re-cently concluded third observing run. 56 non-retracted candidates were identified andpublicly announced in near real time. Gravitational waves from binary neutron starmergers, however, remain of special interest since they can be precursors to high-energyastrophysical phenomena like γ -ray bursts and kilonovae. While late-time electromag-netic emissions provide important information about the astrophysical processes within,the prompt emission along with gravitational waves uniquely reveals the extreme mat-ter and gravity during - and in the seconds following - merger. Rapid communication ofsource location and properties from the gravitational-wave data is crucial to facilitatemulti-messenger follow-up of such sources. This is especially enabled if the partnerfacilities are forewarned via an early-warning (pre-merger) alert. Here we describethe commissioning and performance of such a low-latency infrastructure within LIGO-Virgo. We present results from an end-to-end mock data challenge that detects binaryneutron star mergers and alerts partner facilities before merger. We set expectationsfor these alerts in future observing runs. Keywords:
Compact binary stars(283), Computational methods(1965), Gamma-raybursts(629), Gravitational wave astronomy(675), Gravitational wave detec-tors(676), Neutron stars(1108) INTRODUCTIONThe field of gravitational-wave astronomy hasexploded in the years following the first directobservation of gravitational waves (GWs) froma binary black hole (BBH) merger (Abbott et al.2016). Since then, LIGO-Virgo have published49 candidate events, many of which were identi-fied in low-latency ; these include 2 binary neu-tron star (BNS) and 2 neutron star–black hole(NSBH) candidates (Abbott et al. 2020a). Thedetection of GWs from compact binaries, espe-cially from BBHs, has become routine. GWsfrom BNS and NSBH mergers, however, re-main rare. BNS and NSBH mergers are of spe-cial interest due to the possibility of counter-part electromagnetic (EM) signals. For BNSmergers in particular, it has long been hypothe-sized that the central engine (post-merger) canlaunch short gamma-ray bursts (SGRBs) (Lat- Some of the 56 have not yet appeared in a LIGO-Virgopublication. timer & Schramm 1976; Lee & Ramirez-Ruiz2007), kilonovae (Li & Paczynski 1998; Metzgeret al. 2010), and radio waves and X-rays postmerger (Nakar & Piran 2011; Metzger & Berger2012). In the special case of the presence of amagnetized NS, it can also lead to GRB pre-cursors before the merger (Metzger & Zivancev2016).Although the improvement in AdvancedLIGO-Virgo’s sensitivity was paralleled by anal-ogous advancements in the field of time-domainastronomy, the first observed BNS merger,GW170817 (Abbott et al. 2017c), remains theonly realization of multi-messenger astronomy(MMA) with GWs. The coincident observationof GWs followed by an SGRB, GRB 170817A,and the kilonova AT 2017gfo, (Abbott et al.2017d) bore evidence to the several-decade-oldhypothesis that compact object mergers wereprogenitors of these exotic transients. The jointobservations also contributed greatly to ourunderstanding of fundamental physics (Abbott irst demonstration of early warning gravitational-wave alerts ∼ ∼ ∼ Swift -BAT has recently also demonstrated thepotential to respond autonomously to extremelylow-latency triggers in the future, with the in-troduction of an on-board sub-threshold triggerrecovery algorithm (GUANO, Tohuvavohu et al.(2020)). By the beginning of Advanced LIGO-Virgo’s fourth observing run (O4), it is expectedthat established missions and observatories willbe joined by next generation facilities like the Rubin Observatory (Ivezi´c et al. 2019). Thisgreatly improves the chances of performing tar-geted followup observations of prompt, or evenprecursor (Troja et al. 2010; Tsang et al. 2012),emission from compact binary mergers providedthat pre-merger alerts can be issued.LIGO-Virgo has since streamlined the alertprocess (see Fig. 3). Advanced LIGO’s and Ad-vanced Virgo’s third observing run (O3) sawthe dawn of autonomously distributed Prelimi-nary GCN Notices (LIGO Scientific Collabora-tion 2019) , which allowed LIGO-Virgo to no-tify the world of candidate signals within 7 . +92 − minutes of observation. To further enable EM-GW observations, we can leverage the long-lived nature of BNSs in the sensitive band ofadvanced ground-based GW detectors to makepre-merger detections (Cannon et al. 2012; Chuet al. 2016). This was recently demonstratedby Sachdev et al. (2020) and Nitz et al. (2020).The early detection and communication of GWsfrom BNSs aims to facilitate EM follow-up ef-forts by further reducing the latency of alertsand improving prospects of capturing the ini-tial spectra.In this letter we describe the commissioningand performance of the low-latency sub-systemwithin Advanced LIGO-Virgo that is able toprovide pre-merger alerts for electromagneti-cally bright compact binaries. We begin by de-scribing the end-to-end low-latency workflow inSection 2, from the time of data acquisition tothe dissemination of public alerts. We then as-sess the performance of a subset of this infras-tructure in a mock data challenge described inSection 3, with special emphasis placed on pre-merger alerts. We demonstrate that Prelimi-nary GCN Notices can be distributed with truenegative latencies: partner observatories receive https://gcn.gsfc.nasa.gov/ The 95% reported here is severely impacted by severalhigh latency events that evaded automated procedures.
Magee, Chatterjee, Singer, Sachdev, et al. sky localizations and source information beforethe binary has completed its merger. We reporton the improved latencies at each step of theworkflow, and set expectations for pre-mergeralerts in O4 and next generation detectors inSection 4. ANALYSISThe low-latency workflow begins with data ac-quisition at each interferometer. The digital sig-nal from the output photodiode is initially cali-brated by a pipeline that runs on the set of com-puters that directly control the interferometer.The calibrated data, while produced with near-zero latency, are not yet accurate enough foruse by low-latency gravitational-wave searches.The data are broadcast to a set of computerswhere a GStreamer-based pipeline corrects thestrain data to achieve the required level of ac-curacy (Viets et al. 2018). This pipeline writesthe calibrated strain data to a proprietary LIGOframe data format and then transfers them tocomputing sites. There, the calibrated dataare ingested by the complete set of low-latencyfull bandwidth GW pipelines: cWB (Klimenko& Mitselmakher 2004; Klimenko et al. 2005,2006, 2011, 2016), GstLAL (Sachdev et al.2019; Hanna et al. 2020; Messick et al. 2017),MBTAOnline (Adams et al. 2015), PyCBCLive (Nitz et al. 2018; Dal Canton et al. 2020),and SPIIR (Luan et al. 2012; Hooper et al. 2012;Liu et al. 2012; Guo et al. 2018; Chu 2017). Forthe first time, we also incorporate two matched-filter based pipelines focused on pre-merger de-tection into our workflow: GstLAL (Sachdevet al. 2020; Cannon et al. 2012) and SPIIR (Chuet al. 2020). All detection pipelines analyze thedata for GWs and assign significances to can-didate triggers. Candidates that are assignedfalse alarm rates (FARs) less than one per hour are uploaded to the GRAvitational-wave Can- No trials factor is applied to the candidate uploadthreshold. didate Event DataBase (GraceDB) alongsidedata required downstream in the alert process.After candidates are uploaded, the task man-ager GWCelery interacts with low-latencysearches and GraceDB to orchestrate a num-ber of parallel and interconnected processeswhich, in the event of a discovery, culminates inthe dissemination of GCN Notices. GWCeleryprovided the semi-automated infrastructure forpublic alerts in O3, as well as for the mock datachallenge reported here. The major subsystemsinclude: • The listener for LVAlert, which isa publish-subscribe system used byGraceDB to push machine-readable noti-fications about its state. • The Superevent Manager, which clustersand merges related candidates into su-perevents . • The client functionality to interact withGraceDB. • The GCN listener that listens for no-tices from external facilities to spot co-incidences with GW candidates. • The External Trigger Manager, whichcorrelates gravitational-wave events withGRB, neutrino, and supernova events. • The GCN broker that disseminates GWcandidate information for external con-sumption. • The Orchestrator, which executes the per-(super)event annotation workflow.After candidate events are uploaded bydetection pipelines, they are localized via https://gracedb.ligo.org/ https://gwcelery.readthedocs.io/ https://emfollow.docs.ligo.org/userguide/analysis/superevents.html https://gracedb-sdk.readthedocs.io irst demonstration of early warning gravitational-wave alerts S/N ≈ ≈
13 S/N ≈ ∼
400 deg. ∼
80 deg. ∼
30 deg. EW DetectionGW detectionpipelinesData acquisitionCalibration GCNSkymaps CoincidencesDQ checksSource prop.Superevent manager
GWCelery
GraceDB
Figure 1.
This upper half of the figure illustrates the complete pipeline and interaction of the various(sub)systems, mentioned in Sec. 2, responsible for disseminating early warning alerts. The waveform evo-lution with time is shown in the bottom half along with the dependence of the sky-localization area onthe cutoff time of the early-warning templates and the accumulated S/N during the binary inspiral. Thewaveforms, time to merger, S/N, and localizations in this figure are qualitative.
BAYESTAR (Singer & Price 2016), given aprobability of having an electromagnetic coun-terpart (Chatterjee et al. 2020), and assigneda source-category based astrophysical probabil-ity under the assumption that astrophysical andterrestrial triggers occur as independent Pois-son processes (Kapadia et al. 2020). Events arechecked for temporal and, when possible, spa-tial coincidences with gamma-ray bursts or neu- trino bursts using the RAVEN pipeline (Urban2016). A joint significance is calculated to de-cide whether the joint candidate should be pub-lished.BAYESTAR was optimized in order to sup-port early warning localizations which led toa median run time of 0.5 s per event for earlywarning triggers and 1.1 s per event for fullbandwidth triggers. The latter is a 4 . × Magee, Chatterjee, Singer, Sachdev, et al. speedup compared to usual O3 performance.The significant changes included rearrangementof loops to improve memory access patterns andmake better use of x86 64 vector instructions,changes to the input data handling to distin-guish properly between the merger time and thecutoff time of early warning templates, and theredesign of the reconstruction filter that is usedto sample the SNR time series for likelihoodevaluation to use a lower sample rate. To mitigate the effect of noise transients, basicdata quality checks are also performed for everycandidate uploaded to GraceDB. In particular,specific state vectors are checked to ensure thatcandidate events occur during times when therelevant detectors are in observing mode andto verify that there are no coincident hardwareinjections.A qualitative overview of entire pipeline andthe various (sub)systems mentioned above is il-lustrated in Fig. 1. A heuristic waveform evo-lution and the effect of different early-warningtemplate cutoff times on the accumulated S/Nand the sky-localization is also shown. RESULTSTo demonstrate the robustness of the alert in-frastructure, we describe the results of a mockdata challenge carried out between 11 June 20201700 UTC and 19 June 2020 1700 UTC. Datapreviously collected during O3 were replayed asa mock low-latency analysis. We note that sincethe challenge relied on previously collected data,it was impossible to test the full low-latencyworkflow; notably, data transfer and calibra-tion latencies are not included ( ∼ The early warning templates are Nyquist critically sam-pled which could lead to ringing artifacts. bandwidth triggers used the same FAR thresh-old set throughout O3 for public alerts (1 per2 months) . At fixed FAR, the astrophysi-cal probability (Kapadia et al. 2020) associatedwith pre-merger analyses is lower than for fullbandwidth analyses. Due to this fact, combinedwith our chosen higher FAR threshold for early-warning alerts, we issued retraction circularsfor early warning candidates that were not alsoidentified by the full bandwidth analyses. Therewere no retraction criteria set for full bandwidthtriggers.During the mock data challenge, eight candi-dates were published via the test GCN. 3 candi-dates were identified by only the full bandwidthanalyses and were distributed via notice and cir-cular (LIGO Scientific Collaboration 2020a,b,c).The remaining 5 public candidates were iden-tified only by the early warning pipelines andwere distributed via GCN notices to subscribersof test alerts. None of these 5 candidates wereobserved in the full bandwidth analyses; theywere therefore subsequently retracted (LIGOScientific Collaboration 2020d,e,f,g,h). Out ofthe 5 retracted triggers, 4 came from the Gst-LAL early warning pipeline, while 1 was issuedby the SPIIR early warning pipeline. An au-thentication issue prevented the SPIIR pipelinefrom issuing additional events past the FARthreshold. A summary of the 5 early warningalerts is given in Table 1.Although only 5 pre-merger candidates passedthe early warning public alert threshold, Gst-LAL and SPIIR uploaded 82 and 141 earlywarning candidate events, respectively, toGraceDB. We use the metadata associated withthese uploads to produce Fig. 2. From theevents crossing threshold we see that the maxi-mum delivery time from event upload is 15s, in- A trials factor is applied on top of this threshold to ac-count for the two early warning and four full bandwidthmatched filter pipelines irst demonstration of early warning gravitational-wave alerts C u m u l a t i v e d i s t r i b u t i o n GstLALGstLALGstLALGstLALGstLALGstLAL
Event uploadedPSD uploadedClassificationLocalizationGCN sentGCN received60 40 20 0 20 40 60
Time from merger (seconds) C u m u l a t i v e d i s t r i b u t i o n SPIIRSPIIRSPIIRSPIIRSPIIRSPIIR
Time from event upload (seconds)
Figure 2.
Latencies associated with early warning uploads from the GstLAL (top) and SPIIR (bottom)pipelines. Design differences between the pipelines lead to distinct distributions for the time before mergerat which a candidate is identified. The left panels indicate that ∼
85% and ∼
35% of the uploaded Gst-LAL and SPIIR candidates, respectively, are localized prior to merger. The right panels demonstrate thatdespite differences in latencies associated with event identification, the scatter of the remaining processes isremarkably similar. dependent of pipeline. This enables ∼
85% and ∼
35% of the GstLAL and SPIIR candidates,respectively, to be localized before merger. LOOKING AHEADEarly warning alerts using real data have notyet been released by the LIGO-Virgo collab-oration. Despite the steady improvement ofthe alert infrastructure (Figure 3), there remainseveral areas for improvement in the processingof data and production of alerts if the collab-oration decides to pursue pre-merger triggers. As previously mentioned, low-latency data cali-bration is currently a two step process; the near-zero-latency pipeline is corrected by a secondaryGStreamer-based pipeline. Work is underwayto reduce this to a single calibration step toreduce latency by O (seconds). The calibrateddata are transferred from the detector sites tothe computing clusters in ∼ with https://kafka.apache.org/ Magee, Chatterjee, Singer, Sachdev, et al. +1 h+1 d+2 min+5 min+10 min-30 s-10 s+10 s+30 s
O1 O2 O3
Thiswork
First event createdFirst GCN sent T i m e a f t e r m e r g e r Figure 3.
A history of end-to-end latencies across public alerts in the first three observing runs and themock data challenge presented here (Abbott et al. 2019a). an additional ∼ . is attributed to the choice todistribute data via frame files. A number ofimprovements are under development to reducethis latency budget.Reductions to the noise budget at frequencies (cid:46)
30 Hz will improve the possibility of detectionpipelines identifying signals long before merger.We estimate that if the noise floor below 30 Hzremains unchanged from O3, the recovered S/None minute and 30 seconds before merger will be ∼
50% and ∼
20% less, respectively, than if thedetectors reach the previously projected O4 sen-sitivity. The effect is less severe for early warn-ing times just before merger, but low frequencynoise is a major barrier to advance alerts.Figures 2 and 3 demonstrate that the GWalert system is capable of providing GW alertsbefore merger, but they do not consider theprospects for detection from an astrophysical Four seconds for Virgo data. source population. We generate a population ofsimulated BNS signals, henceforth referred to as injections , using the
TaylorF2 (Sathyaprakash& Dhurandhar 1991; Blanchet et al. 1995,2005; Buonanno et al. 2009) waveform model.Both source-frame component masses are drawnfrom a Gaussian distribution between 1 . M (cid:12) < m , m < . M (cid:12) with mean mass of 1 . M (cid:12) and standard deviation of 0 . M (cid:12) , modeled af-ter observations of galactic BNSs ( ¨Ozel & Freire2016) . The neutron stars in the populationare non-spinning, motivated by the low spinsof BNSs expected to merge within a Hubbletime (Burgay et al. 2003; Zhu et al. 2018).The signals are distributed uniformly in co-moving volume up to a redshift of z = 0 . Note that if GW190425 is a BNS, then galactic measure-ments are not representative of neutron star masses. irst demonstration of early warning gravitational-wave alerts − Time before coalescence (s)10 − D e t ec t i o n s ( y r − ) (a) − − E x p ec t e dnu m b e r o f e v e n t s p e r y e a r −
90% credible area (deg )10 − − − C u m u l a t i v ee v e n t s a s a f r a c t i o n o f t o t a l e v e n t s
60 s45 s30 s15 s10 s0 s (b)
Figure 4. (4a) Projected O4 early warning detection rate assuming 0 second (blue) and 25 second (red)end-to-end latencies from the GW alert system. The worst case scenario assumes 5 seconds for calibrationand data transfer, 5 seconds for pipeline analysis, and 15 seconds for event upload and GCN creation. Therate of expected detections was estimated from a simulated data set assuming a 100% detector duty cycle forthe 4-detector HLVK network. The uncertainty bands reflect the (5% , KAGRA at their projected O4 sensitivities. We simulate the results of an early warningmatched-filtering pipeline by considering 6 dif-ferent discrete frequency cut-offs: 29 Hz, 32 Hz,38 Hz, 49 Hz, 56 Hz, and 1024 Hz to analyzesignal recovery at (approximately) 58 s, 44 s,28 s, 14 s, 10 s, and 0 s before merger, moti-vated by Sachdev et al. (2020). We calculatethe network S/N of each injection at each fre-quency cut-off and consider the events that passan S/N cut-off of 12.0 as ‘detected’. We thencalculate the sky posteriors for each of the de- https://dcc.ligo.org/LIGO-T2000012/public tected signals by using BAYESTAR (Singer &Price 2016). We use the most recent BNS lo-cal merger rate from Abbott et al. (2020b) of320 +410 − Gpc − yr − to estimate the number ofevents detected per year in the detector net-work. In Figure 4a we see that our optimisticscenario predicts 5 +7 − GCN will be received 1second before merger per year, while our pes-simistic scenario predicts O (1) GCN will be re-ceived 1 second before merger per year consid-ering the higher end of the BNS rate. Figure 4bpredicts that ∼ ∼
20% ( ∼ . ∼ Magee, Chatterjee, Singer, Sachdev, et al. of the detectable events ( ∼ at O4 sensitivities. This highlights theneed for continued latency improvements in ad-vance of O4 to maximize the potential of cap-turing prompt emission.In the design sensitivity era with three de-tectors, Sachdev et al. (2020) have shown thatabout half of the total detectable BNSs will befound 10 s before merger, and about 2% will beidentified before merger and localized to within100 deg . Sachdev et al. (2020) used the Gst-LAL pipeline in an early warning configurationto assign FARs to simulated BNS signals to es-timate these rates. We extend this to includeKAGRA in the detector network, but we esti-mate rates based on a fiducial S/N cut-off of12. We find that our zero-latency scenario im-proves to ∼ ∼ ∼ (cid:46) , ∼ ∼ > Note that the estimated BNS rate at the time of Sachdevet al. (2020) was approximately three times larger thanthe updated rate presented in Abbott et al. (2020b) et al. (2018) found that the Einstein Telescopecan alert observers up to 20 hours in advance for58% of detectable BNS at 200 Mpc and 100% at40 Mpc. The majority of these signals will bewell localized. A similar study by Akcay (2019)with a S/N detection threshold of 15 found thatthe Einstein Telescope will provide early noticefor O (10 ) BNS mergers next decade.ACKNOWLEDGMENTSWe are grateful to B.S. Sathyaprakash forreviewing our manuscript and providing use-ful comments. We thank the LIGO Labora-tory for use of its computing facility to makethis work possible, and we gratefully acknowl-edge the support of NSF grants PHY-0757058and PHY-0823459. C.H. gratefully acknowl-edges the support of NSF grant OAC-1841480.DC acknowledges NSF grant no. PHY-1700765and PHY-1912649, and is supported by theIllinois Survey Science Fellowship of the Cen-ter for Astrophysical Surveys (CAPS) at theUniversity of Illinois Urbana-Champaign. S.S.is supported by the Eberly Research Funds ofPenn State, The Pennsylvania State Univer-sity, University Park, Pennsylvania. G. M.is supported by the National Science Founda-tion (NSF) through award PHY-1764464 to theLIGO Laboratory. MK, QC, FP, LW, AP, AS,VO acknowledge the funding from AustralianResearch Council (ARC) Centre of Excellencefor Gravitational Wave Discovery OzGrav un-der grant CE170100004. Facilities:
LIGO, EGO:Virgo irst demonstration of early warning gravitational-wave alerts Software: astropy (Astropy Collaborationet al. 2013), numpy (Harris et al. 2020), mat-plotlib (Hunter 2007), iPython (Perez & Granger 2007), pandas (Wes McKinney 2010), gwpy(Macleod et al. 2020), celery (Solem & contribu-tors 2020)APPENDIX
Table 1.
A summary of the 5 early warning alert information and latencies from the mock data challengedescribed in Sec. 3. Among the 5, MS200619bf was reported by the SPIIR pipeline, while the others werereported from GstLAL. The latencies are broken down in steps of the event being uploaded into GraceDB,the superevent being created, the skymap being available for the preferred event, and the notice beingacknowledged by GCN.
Superevent Date (UTC) FAR Latency GCNsEvent Superevent Skymap NoticeMS200615h 2020-06-15 00:35:40 2.02e-06 -2.9 -1.9 0.1 7.1 https://gcn.gsfc.nasa.gov/gcn3/27951.gcn3MS200618aq 2020-06-18 05:47:05 1.78e-07 -53.1 -52.1 -50.1 -35.1 https://gcn.gsfc.nasa.gov/gcn3/27990.gcn3MS200618bq 2020-06-18 11:00:59 3.50e-06 -16.9 -21.9 -11.9 -2.9 https://gcn.gsfc.nasa.gov/gcn3/27987.gcn3MS200618bx 2020-06-18 12:17:08 3.76e-06 -63.3 -62.3 -59.3 -51.3 https://gcn.gsfc.nasa.gov/gcn3/27988.gcn3MS200619bf 2020-06-19 10:24:43 1.91e-06 -41.0 -40.0 -35.0 -27.0 https://gcn.gsfc.nasa.gov/gcn3/27989.gcn3
REFERENCES