LIGO Data Analysis: Status and Plans

LIGO Data Analysis: Status and PlansPatrick BradyUniversity of Wisconsin-Milwaukee

LIGO Scientific Collaboration

LIGO Scientific Collaboration - TAMA Symposium

Outline of talkFirst LIGO science data run (S1)Summary of runOrganization of LIGO data analysis effortsStatus of each effortDetailed discussion of burst/inspiral analysis Emphasis because of TAMA/LIGO coincidence analysis

Plans for second LIGO science data run (S2)Online analysis plans/statusTools used in control room

Summary


LIGO Run ScheduleScience runs are interspersed with engineering runs and commissioning activitiesNowJan 2002FebMarJunMayAprJulAugOctSepNovJan 2003FebDecMarJunMayAprJulAugOctSepNovE7E8S2S3S1E10, strain noise per root HzE9


LIGO sensitivity at start of S1S1 run23 Aug 9 Sep, 2002

Standard candle2 x 1.4 Msun optimally oriented binaryS1 reach includesMilky WayLMCSMC


Data from S1 runH1: 235 HoursH2: 298 HoursL1: 170 Hours3x: 95.7 HoursS1 total run time: 408 hours = 17 daysH1 (4km, Hanford) duty cycle 57.6%H2 (2km, Hanford) duty cycle 73.1%L1 (4km, Livingston) duty cycle 41.7%Triple coincidence duty cycle 23.4%Green bands with black borders indicate locked segments


Data analysis groupsBurst and other transients: Sam Finn and Peter Saulson (co-chairs)http://www.ligo.caltech.edu/~ajw/bursts/bursts.htmlContinuous WaveMichael Landry and Mariallessandra Papa (co-chairs)http://www.lsc-group.phys.uwm.edu/pulgroup/InspiralPatrick Brady and Gabriel Gonzalez (co-chairs)http://www.lsc-group.phys.uwm.edu/iulgroup/StochasticPeter Fritschel and Joe Romano (co-chairs)http://feynman.utb.edu/~joe/research/stochastic/upperlimits


Burst Group ActivitiesSearch for bursts of unknown origin/waveformGenerate event triggers using SLOPE, TFCLUSTERS, POWERVeto triggers due to instrumental artifacts Determine upper limit on rate as function of strainMonte Carlo by simulated injections of astrophysical motivated signals (Zwerger et al) and other burst waveforms

Search for bursts associated with GRBs.Triggered analysis of on-source timesResult by comparison of on-source versus off-source distributionsFirst EM triggered search with LIGO


Continuous Wave Group AnalysisKnown pulsar searchesCatalog of known pulsarsHeterodyne narrow BW folding dataCoherent frequency domain search using Hough transformAll sky unbiasedSum short power spectra (no doppler correction)Wide area searchHierarchical Hough transform code is under developmentDemodulation is functioning and used in known pulsar searchDemodulation points on sky under controlEfficient positioning of spindown/sky points under development


Inspiral Group ActivitiesBinary Neutron Star SearchBread n butter source for LIGODetermined upper limit on the rate of BNS inspirals in the universeBlack hole MACHO binary search (0.5

Stochastic Group ActivitiesAnalytic calculation of expected upper limits (~100 hrs): W for LHO 2k-LHO 4k will provide the most stringent direct observational upper limit to dateCoherence measurements of GW channels show little coherence for LLO-LHO 2k correlationsInvestigation of effect of line removal for LHO 2km-LHO 4km correlations (e.g., reduction in instrumental correlated noise)Injection of simulated stochastic signals into the data and extraction from the noise to validate end-to-end capability of analysisCorrelations between LLO with ALLEGRO bar detectorALLEGRO was rotated into 3 different positions during earlier E7 runAnalysis in progress


Playground Data SetRepresentative sample of data distributed over runNot used in determining astrophysical resultsUsed to tune thresholds Determine veto cutsGain experience without introducing bias into upper limit analysesCharacteristics~9 hours of triple coincident data was selected for this purposeChosen by Gabriela Gonzalez in consultation with others on siteRepresentative of broad range of instrumental behavior Which groups used it?Inspiral groupBurst groupStochastic group


Burst Trigger GenerationOne hour summary plot ConfidenceLog( probability that trigger is caused by Gaussian noise)Large negative value is loud burstTime-frequency plotGreen: 0 > confidence > -100Red: confidence < -100 No. of eventsTriggers per 10 secondsEvent definitionDifferent for TFCLUSTER, SLOPE, POWERTrigger in 200-400Hz band with duration ~100 sec


Inspiral Trigger Generation: TemplatesUse template based matched filtering algorithmTemplate waveforms for non-spinning binaries2.0 post-Newtonian approx.Computational efficiencyStationary phase approximation to Fourier TransformDiscrete set of templates labeled by M1, M21.0 Msun < m1, m2 < 3.0 Msun2110 templatesAt most 3% loss in SNR


Inspiral Trigger GenerationAS_Q processed in chunks of 256 seconds down-sampled to 4096HzEach chunk is divided into 7 segments of 64 seconds overlapped by 32 secondsFor each template: Compute the SNR: large values indicate that GW channel correlates well with the templateIf SNR > 6.5, compute a2 : small values indicate that SNR was accumulated in a manner consistent with an inspiral signal. If a2 < 5.0, record triggerTriggers are clustered within duration of each template, but multiple templates can trigger at same time.


Dealing with Non-Gaussian SpurionsExample at LIGO LouisiannaCattle Guard(Gonzalez, Chickarmane, Saulson during E7)How to deal with them?Auxiliary channels vetoes Can physical cause be tracked? Use PEM & other channelsPotential vetoesEvaluated vetoes generated by several tools


Veto InvestigationsTuned veto SNR thresholds and windows using the playground data; focused on eliminating the highest-SNR candidates without introducing much deadtime Best channels turned out to be auxilliary interferometer channels Example from inspiral analysis (similar for burst)Livingston 4km:Net deadtime 2.8%Hanford 4km:Net deadtime 0.3%Hanford 2km: Net deadtime 3.2%Must check that veto conditions would not veto a real gravitational wave!Studied coupling using hardware injections in differential control & end mass excitationsFound some surprise couplings which may involve abandoning certain vetoes


Effectiveness of Vetoesfor S1 Playground DataSo, how do triggers & vetoes fit together in analysis pipeline?


H1 Sees?Coincidence?TriggersMatched FilterGW + InjectionsVeto TriggersAuxiliary ChannelsFilterH1 cleanNot L1L1 cleanNot H1Event CandidatesNoYes-YesDumpNoL1 & H1clean


Sample Population Monte CarloBinary Neutron star population

Mass distribution derived from population synthesis models

Spatial distribution out to 200kpc including Milky Way, LMC and SMC

LMC and SMC contribute about 12% of a Milky Way equivalent Galaxy

Signals injected into data stream and used to determine efficiency of pipeline to detection of BNS population


Testing with Hardware InjectionInspiral Injections into hardware


Testing with Hardware InjectionsPre-run: 2 x 1.4 Msun into L1:LSC-DARM_CTRL-EXCPre-run: 2 x 1.4 Msun into L1:LSC-ETMX-EXC

T_injT0SNRD_effD_inj%error714029925.4249714029925.431617.3445.753.6-17.2971030015.42449714030015.431635.6322.326.7-19.73714030105.4249 714030105.4314 53.83 13.8 13.4 2.90


LIGO First Science Run SynopsisCompact object inspiraling waveformsBNS coverage will include the Milky Way, plus LMC, SMCBlack hole MACHO search under way Bursts/transient events96 hours of 3X coincidence2 different (complementary) filters applied to datafrequency-time clustering algorithm, time-domain slope detectorEfficiency using astrophysically motivated SNe waveforms and other.Continuous wave sourcesInitial searches target known EM sources, e.g.:-PSR J1939+2134 (P= 1.557 ms, search and analysis in progress)Sco X-1 (in progress - 500 Hz - 600 Hz, multi-parameter search)Stochastic backgroundLimiting sensitivity for W will be better than previous direct GW observational determinations with resonant bars (narrowband)


Plans for S2 S2: 14 Feb 14 April 2003Working groups have well defined primary analysis path for S2 dataPlan to extend/enhance methodology from S1Some new tools are under development, e.g coherent multi-detector analysisVeto development will be a major focus of the S2 detector characterization effortDesire to understand the instrumental origin of glitchesSignificant effort has been put into hardware injection plan Daily hardware injections of astrophysical signals to help calibrate systematics


Plans for S2 Trigger generators to run in real-timeBurst searches: TFCLUSTERS, POWER, SLOPEInspiral searchesKnown pulsar demodulationExternal trigger searchesIssues that caused problems in S1 but have been fixedAccurate, real-time calibration data availableOn-line monitoring tool to provide alarms to control room


ConclusionLIGO Scientific Operation Started in Aug. 2002!Analysis has been proceedingFirst results should be announced in Mar 2003

Second run scheduled 14 Feb - 15 Apr 2003Sensitivity should be almost 10x better than S1Moving towards real-time analysis environment

Looking forward to the TAMA/LIGO coincidence effort


Documents

LIGO Data Analysis: Status and Plans