Coherent GRB Search in the WSR1 Data with XPipeline

1
Coherent GRB Search in the WSR1 Data with
X-Pipeline

• Patrick Sutton Michal Was
• LIGO-Caltech Ecole Normale Superieure

2
Outline

• Coherent Statistics for GWB Detection
• geometric interpretation
• detection statistics
• consistency tests
• X-Pipeline Triggered Searches
• Triggered GRB search in the project IIB / WSR1
  data set
• analysis procedure
• trigger characteristics
• loudest event upper limits
• compare LIGO-Virgo to LIGO-only

3
Coherent Analysis Basic Formalism

• Form vector of whitened data from one frequency
  bin

• Implicit time shift for sky position W.
• GWB parameters sky position W(q,f) (known for
  GRB case) and amplitudes h(f), hx(f).
• GWB can only appear in F, Fx directions.

4
Geometric View One TF Pixel
Work in dominant polarization frame (Klimenko
et al, PRD 05)
(noise only)
(signal plus noise)
(signal plus noise)
5
Likelihood or Energy Measures
Likelihoods used in this analysis
Hard Constraint
Null Energy
Incoherent Energy
Klimenko, Mohanty, Rakhmanov, Mitselmakher, PRD
72 122002 (2005) J. Phys. Conf. Ser. 32 12
(2006) Chatterji, Lazzarini, Stein, Sutton,
Searle, Tinto, PRD 74 082005 (2006)
6
X Pipeline

• MatLab based coherent analysis package.
• http//www.ligo.caltech.edu/psutton/protected/xpi
  peline/xpipeline.html
• Code available in matapps CVS.
• In GWB detection mode
• Time shifts data, FFTs, constructs time-freq.
  likelihood maps.
• Chooses sky position that maximizes summed
  likelihood in each time slice.
• Scripted for easy analysis of GRBs
• ./grb.py --params-file grb.ini --grb-time
  841896355 --right-ascension 217.5
  --declination -28.8 --detector H1
  --detector L1 --detector H2 --detector V1

7
WSR1 GRB Search

• No GRBs during WSR1 coincident operation
• There was a GRB a few minutes before the official
  start, but Virgo was not operating in science
  mode until about 10 hours later.
• therefore make up a GRB to serve as a test
  case
• GPS trigger time 841896355
• right ascension 217.5255
• declination -28.7510
• This was in the middle of the longest 5x
  (H1-H2-L1-G1-V1) coincidence segment. The sky
  position was chosen to favor GEO Virgo
• Site F_2F_x2
• H 0.2598
• L 0.3364
• V 0.8356
• G 0.7691

8
Analysis procedure data

• Follow procedure similar to that used to date in
  LIGO GRB searches (Leonor et al.)
• Data sets
• on-source data /- 1 minute around GRB time
• off-source data All H1-H2-L1-V1 coincident data
  within /-12 hours of GRB trigger for background
  estimation (about 16 hours total).
• simulations Add GWB signals to on-source data
• no data quality flags applied yet
• Divide data into overlapping pieces of duration
  1/256 sec.
• FFT each piece, compute likelihoods summed over
  frequency bins 512,1536 Hz (5 bins).

9
Analysis procedure likelihoods

• Analyse all data, using hard constraint
  likelihood as the detection statistic (Leonor
  uses cross-correlation).
• Record all hard constraint likelihoods above some
  low threshold (10 Hz false rate).
• Compare null vs. incoherent energy as consistency
  test to remove glitches (Chatterji et al. PRD).
• Use off-source and simulation results to tune the
  null vs. incoherent consistency test.

10
Simulated GWBs

• Lazarus black-hole merger waveforms (Baker et al.
  02)
• Equal-mass, non-spinning 55 Mo.
• Circularly polarized (were looking down the GRB
  axis)
• Not a real waveform for GRBs! Just a test
  waveform in-band.

11
Likelihood Distributions
Events from off-source data Color-coded by hard
constraint likelihood Glitches lies on diagonal
loud glitches
Gaussian background
12
Likelihood Distributions
Lazarus GWBs Real GWB signals lie above the
diagonal. hrss 5e-22 / Hz1/2 hrss 2.5e-22 /
Hz1/2
loud glitches
Gaussian background
13
Likelihood Distributions
(Zoomed-in view) Threshold on ratio Einc/Enull
to remove glitches. Optimum value (1.12)
determined by computing expected upper limit for
off-source data. Will try more sophisticated
cuts in the future.
loud glitches
Gaussian background
14
On-Source Results

• Repeat entire analysis for H1-H2-L1 network
• Use exactly the same tuning procedure for
  Einc/Enull cut.
• LIGO-only 90 Upper Limit
• hrss 4.6 x 10-22 Hz-1/2
• Using Virgo data improved amplitude sensitivity!
• GRB position favored Virgo by a factor 2.5.

• Practice Upper Limit
• Look at on-source data.
• Find largest hard constraint likelihood surviving
  Einc/Enull cut (loudest event).
• Upper limit on GWB amplitude is the smallest
  amplitude such that 90 of injections survive
  Einc/Enull cut and are louder than loudest event.
• LIGO-Virgo 90 Upper Limit
• hrss 3.7 x 10-22 Hz-1/2

15
Summary

• Used X-Pipeline to demonstrate a preliminary, but
  complete, fully coherent triggered analysis.
• From raw strain data to loudest event upper
  limit.
• Coherent statistics for detection and glitch
  rejection.
• Analysed 2 different network configurations
  (LIGO-Virgo LIGO-only).
• Combining Virgo WSR1 data with LIGO data gave an
  improved amplitude upper limit for a GRB
  favorably positioned for Virgo.
• More study needed to determine scientific value
  of LIGO-Virgo joint analyses
• Used atypical GRB, but Virgo data improving since
  WSR1

16
Future Plans

• Finish script job processing infrastructure
• mainly post-processing codes, combining multiple
  time-frequency resolutions, automated tuning.
• Finish implementation and testing of clustering
  algorithm for time-frequency maps
• using J. Sylvestres generalized clusters
  algorithm (TFClusters)
• improve sensitivity
• Run over S5 GRB set