LIGO-G Z Status of the LIGO-AURIGA Joint Burst Analysis L. Cadonati for the LIGO-AURIGA joint working group LSC meeting – Ann Arbor, June

LIGO-G Z 2 Scope of this talk Describe “tuning in action” for the first LIGO-AURIGA coincidence search: an all-sky, trigger-based burst search performed over the last two weeks of S3 in coincidence with AURIGA’s run 331. We use a tuning data set obtained with 49 sets of time lags between AURIGA, LHO and LLO data and discuss how the analysis thresholds affect the final background rate and the detection efficiency for a class of simulated waveforms. We choose here to focus on sine and cosine Gaussians at f 0 =900 Hz, Q=9. Francesco Salemi (AURIGA) is planning to present this material at the Amaldi Conference, on behalf of the LSC and the AURIGA Collaboration (except for slides marked as “not Amaldi”)

LIGO-G Z 3 Sensitivity Spectra single-sided PSD best performance during the 331 and the S3 run AURIGA run 331: Dec – Jan LIGO S3 run: Oct – Jan AURIGA 331: poor data quality (unmodeled excess noise) LIGO S3: glitchy, noise variability Main scope of analysis of this data set: study of IFO-bar methodologies on actual data.

LIGO-G Z 4 Data Selection Criteria LIGO’s data selection criteria »Applied all data quality flags selected for the LIGO-only analysis (dust included) »Require all three interferometers in operation The veto implemented on the AURIGA side to compensate run- specific effects »Wide-band glitches »Epoch veto based on Monte Carlo detection efficiency Coincidence run (after removal of 10% playground data set): 352 h After AURIGA wide-band (4%) and epoch (42%) veto: 211 h LIGO H1H2-noL1 (with DQ flags): 193 h Intersection (AU-H1-H2): 116 h LIGO triple coincidence (with DQ flags): 61 h Intersection (AU-H1-H2-L1): 36 h

LIGO-G Z 5 Analysis method From the white paper: method 2, “eyes wide open” No assumptions are made on direction or waveform. The r-statistic test (within CorrPower ) is applied to the LIGO interferometers around the time of the AURIGA triggers. Single-IFO h rss at each LIGO detector is estimated with the burst parameter estimation code (Note: this portion of the analysis is not mature for presentation at AMALDI) Efficiency for classes of waveforms and source population is performed through Monte Carlo simulation, LIGO-style. The accidental rate (background) is estimated from “off-source” data sets, where LLO and LHO data streams are time-shifted (different lags) with respect to the AURIGA data stream.

LIGO-G Z 6 Minimal Thresholds AURIGA: for the analysis of run 331 data, the exchange threshold is as low as allowed before hitting the Gaussian wall: SNR > 4.5 (empirically chosen). Since the noise performance in run 331 was far from stationary Gaussian, the threshold to be used in an hypothetical IGEC-style search would have been SNR>7. LIGO: from first principles, events with  in Gaussian noise are consistent with the null hypothesis (no correlation) at the 0.1% level or higher: these events can already be discarded. S3 was a “glitchy” run (the LIGO-only used the threshold    !); we decided to use as starting point/minimal threshold that of S2 (    ). The tuning procedure will establish which direction we need to move away from these minimal thresholds. The tuning procedure is still in progress: we have explored 4-detector coincidences and are working on the AU-H1-H2 triple coincidence. We reserve to freeze thresholds after we have explored both sectors and remain blind to the final result until that point. In order to exemplify the procedure, today we will discuss background and detection efficiency for a sample waveform and a trial set of thresholds and how these quantity are affected by the thresholds choice.

LIGO-G Z 7 Diagnostic plots of AURIGA triggers AURIGA trigger generation: delta matched filter. »Events characterized by time, its uncertainty, H, SNR Rate (SNR>4.5) vs “live hours” in the AURIGA dataset SNR vs time

LIGO-G Z 8 More AURIGA diagnostics 20 sec “bump” Auto-coincidences (100 ms window) Plateaux reached at ~ 300 sec Bump at 20 sec disappears increasing the SNR threshold SNR>7 SNR>6 SNR>5 SNR>4.5

LIGO-G Z 9 LIGO events Apply r-statistic test in CorrPower »Auriga arrival time ± uncertainty ± 27 ms (=flight time) »Max uncertainty =100ms, typical <<100ms Use  to make a statement on the coherence between the 3 LIGO IFOs. Impose H1-H2 consistency criteria: »Sign of the H1-H2 correlation »Amplitude cut between H1 and H2 ( Hz vs broadband? This is not in a mature state, but is already showing promising results in the rejection of background events)

LIGO-G Z 10 Testing correlation of LIGO (lagged) data around the time of AURIGA triggers A background set reserved for tuning purposes: 49 time lags between AU, LLO and LHO No H1-H2 consistency cuts applied yet Triggers below SNR=4.5 are not exchanged Minimal threshold 1.4 million triggers in the background dataset ~200 triggers above minimal thresholds AURIGA SNR (red=all blue=AU-H1-H2-L1) LIGO  (49 lags, AU-H1-H2-L1)

LIGO-G Z 11 Testing correlation of LIGO (lagged) data around the time of AURIGA triggers (NOT to be shown at Amaldi) A background set reserved for tuning purposes: 49 time lags between AU, LLO and LHO No H1-H2 consistency cuts applied yet Triggers below SNR=4.5 are not exchanged Minimal threshold 1.4 million triggers in the background dataset ~200 triggers above minimal thresholds

LIGO-G Z 12 Tuning the Amplitude Cut (NOT to be shown at Amaldi) Ratio = max(H1/H2, H2/H1) Ratio = 2 vs ratio = 3: 5% loss in detection efficiency, small gain in falses. Preliminary choice: ratio=3 ratio=3 Red: background events Black: simulated signals (sine/cosine gaussians, gaussians, damped sinusoids) ratio=3  ≥ 4  ≥ 6

LIGO-G Z 13  histograms before and after H1-H2 amplitude cut (NOT to be shown at Amaldi) Surviving events are being followed up (glitches at L1,H1 in random coincidence?) amplitude cut promising…  distribution for all background events in the set (1.4 million events) Surviving R0>0 but not amplitude cut (ratio=3) Surviving both H1H2 consistency cuts Events rejected by the R0>0 cut

LIGO-G Z 14 Scatter plot of false events before and after H1-H2 amplitude cut (NOT to be shown at Amaldi) 1.4 million triggers were in the background dataset 3 events left with the preliminary tuning parameters, corresponding to 0.07 events expected at zero lag, 0.5 mHz background rate

LIGO-G Z 15 Tuning Summary Plot (2-d ROC) using Sine/Cosine Gaussians f 0 =900Hz Q=9 Trial thresholds for illustration The response is dominated by AURIGA sensitivity/noise: better increase LIGO threshold to suppress falses and preserve efficiency Background events contour lines 1, 5, 10 evts/49 lags Preliminary background events contour lines are conservative: ongoing studies of H1-H2 amplitude cut are promising for suppression of false events.

LIGO-G Z 16 Preliminary sensitivity Using the trial tuning parameters: »  >6 »SNR_AURIGA>4.5 »No H1-H2 amplitude cut (not yet mature) Sine/cosine gaussians: »Efficiency of the joint search is dominated by AURIGA »A factor 2 worse than LIGO-only at similar frequency »AURIGA alone: sensitivity improved by a factor <2 due to the lower SNR threshold (with the non-gaussianity of the run 331 data set, a threshold of SNR 7-8 would be required by the IGEC protocol) Tested damped sinusoids and gaussian pulses too – not reported here.

LIGO-G Z 17 Efficiency curve for sine/cosine gaussians with trial set of tuning parameters Tuning tests performed so far indicate that the efficiency curve changes by at most 10% as the tuning parameters are changed in the range that is being considered. SNR_auriga=4.5  0 =6 is not critical for the efficiency (see tuning plot) Preliminary studies on the introduction of an H1-H2 amplitude cut would give at most a 5-10% loss in hrss_50. AURIGA: 7.5e-20 Hz -0.5 at SNR=7 8.5e-20 Hz -0.5 at SNR=8 (threshold range required by the IGEC protocol for run 331) 2.3e-20 Hz -0.5 LIGO only at 850Hz Waveburst ETG  0 =10 Efficiency curve for the joint analysis, dominated by AURIGA: 50% efficiency at ~5.5e-20 Hz -0.5

LIGO-G Z 18 Outlook Step needed to complete the “all-sky” S3-331 coincidence analysis: »Complete characterization and implementation of an H1-H2 amplitude cut for false events suppression. »Repeat analysis of triple coincidence AU-H1-H2 »Fix thresholds and open the box for zero-lag coincidences Still exploring possibilities for a targeted, directional joint analysis.

LIGO-G Z 19

LIGO-G Z EXTRA SLIDES Talk given at GWDAW by LC on behalf of the LIGO-AURIGA joint working group

LIGO-G Z 21 MOU for S3/AU1 burst analysis signed in July 2004 »develop methodologies for bar/interferometer searches »implement on real data a time coincidence, triggered based searches »explore coherent methods Joint Working Group composition: »AURIGA: G. Prodi, L. Baggio, A. Mion, A. Ortolan, S. Poggi, F. Salemi, G. Vedovato »LSC: L. Cadonati, S. Heng, W. Johnson, P. Sutton, M. Zanolin Target analysis completion: spring 2005 AURIGA-LIGO First Coincidence Run AURIGA AU1 run: Dec – Jan LIGO S3 run: Oct – Jan hours 4-fold coincidence 175 hours 3-fold coincidence

LIGO-G Z 22 Sensitivity Spectra single-sided PSD best performance during the AU1 and the S3 run

LIGO-G Z 23 Time Variability h rss of Sine-Gaussian pulses with: f 0 =900Hz Q=9 SNR  =1

LIGO-G Z 24 Geometry Considerations Circular polarization: f AU f LHO f 2 = F F x 2 antenna pattern F  = f cos(2  )  = angle between wave frame and earth frame Linear polarization:  f 2 =F + 2 +F x 2 (  independent) LIGO Hanford AURIGA

LIGO-G Z 25 Sky Coverage for LLO+LHO+AURIGA assumption: 1.gravitational waves are linearly polarized 2.AURIGA sensitivity is 1/3 of the LLO/LHO sensitivities required: 1.F (  ) > 0.1 at LLO, LHO 2.F (  ) > 0.3 at AURIGA the color scale represents the percentage of detectable polarizations

LIGO-G Z 26 Scientific Motivation for a Joint Burst Analysis 1.False alarm rate suppression 2.Increased effective observation times »i.e. combining all possible coincidences of 3 out of 4 or more detectors 3.Increased detection confidence »waveform parameters can be estimated with at least 3 locations and enough time resolution (Gursel-Tinto, 1989)

LIGO-G Z 27 General Strategy for a Trigger-based Analysis Data quality and veto performed “at home” Exchange SINGLE DETECTOR triggers »peak time and its error »homogenously defined amplitude A (BW Hz and broadband) and its error Exchange A T (minimum detectable) versus time Blind analysis »all tuning is performed on time-shifted data before “opening the box” Compare the measured test statistic to its background distribution »background measured on a different set of time shifts Result interpretation »directional analysis (optimized) »all-sky search (using Monte Carlo for source position distribution) »Results interpreted for sine-gaussians, damped sinusoids, Lazarus waveforms for BH-BH mergers (10-20 M ๏ )

LIGO-G Z 28 Method 1: Directional Analysis IGEC-style analysis, on a telescope of directions (see talk by Poggi) New: account for polarizations Use consistent definition of amplitude in all detectors. Threshold on the event amplitude at each detector, scaled by the antenna pattern for the given direction. Require coincidence of at least 3 detectors.

LIGO-G Z 29 Method 2: “Eyes Wide Open” No assumptions are made on direction or waveform. A CorrPower search (see poster) is applied to the LIGO interferometers around the time of the AURIGA triggers. Efficiency for classes of waveforms and source population is performed through Monte Carlo simulation, LIGO-style (see talks by Zweizig, Yakushin, Klimenko). The accidental rate (background) is obtained with unphysical time-shifts between data streams.

LIGO-G Z 30 Sample Performance for Method 2 LIGO: SIMULATED white gaussian noise matching median BLRMS during the AURIGA-LIGO coincidence run AURIGA: Monte Carlo average detection efficiency using  -filter on sine-gaussians with f 0 =900Hz Q=9 and SNR>4 using ACTUAL DATA (no epoch veto) PRELIMINARY

LIGO-G Z x /sqrt(Hz) 1.5x /sqrt(Hz) AURIGA on real data LIGO on white gaussian noise uniform polarization all-sky distribution over the AU1/S3 2-week period SNR=4 threshold on AURIGA Correlation confidence threshold on CorrPower as in the S2 analysis (  >4) Network efficiency ~ product of the two curves Sample Performance for Method 2 LIGO and AURIGA efficiency for sine-gaussians with f 0 =900Hz Q=9 The LIGO curve was obtained on simulated data: a performance degradation is expected on real data

LIGO-G Z 32 Summary and Outlook A working group for the joint burst search in LIGO and AURIGA has been formed, with the purpose to: »develop methodologies for bar/interferometer searches, to be tested on real data »time coincidence, triggered based search on a 2-week coincidence period (Dec 24, 2003 – Jan 9, 2004) »explore coherent methods Simulations and methodological studies are in progress. Planning on first event list exchange within the next two months, for a trigger-based burst analysis.