1. Searching for gravitational wave bursts with the new global detector network
2007 May 3
Searching for gravitational wave bursts with the new global detector network
Shourov K. Chatterji
INFN Sezioni di Roma / Caltech LIGO Laboratory
for the LIGO Scientific and Virgo Collaborations
7th Eduardo Amaldi Conference on Gravitational Waves
2007 July 8-14
Sydney, Australia
Shourov K. Chatterji - EGO Seminar - Cascina, Italy

2. Abstract
EGO Seminar May 3

3. Overview Benefits of joint data analysis
Agreement on joint data analysis
Previous studies based on coincidence methods
Investigations with simulated data
Current investigations with real data
Coherent methods for joint data analysis
Special case of collocated detectors
General case of multiple non-aligned detectors
Maximum likelihood statistics
Null stream consistency tests
Source reconstruction
Hierarchical search approach
Conclusions
EGO Seminar May 3

4. Benefits of a global network
Improved sky coverage
Less likely that event occurs in null of detector network
Improved duty cycle
More likely that at least one detector observes an event
EGO Seminar May 3

5. Benefits of a global network
Increased signal to noise ratio
Coherently sum signals from multiple detectors
Improved detection confidence
Multi-detector coincidence greatly reduces false rate
Coherent consistency tests can differentiate between gravitational-wave signals and instrumental glitches
Improved source reconstruction
“Inverse problem” requires three non-aligned detectors
Provides sky position and both polarizations of waveform
Permits comparison with theory
This is where the science is!
Shared best practices
Learn from each other’s approaches
EGO Seminar May 3

6. Agreement on joint data analysis
Recognizing the benefits of joint data analysis, the LIGO Scientific and Virgo collaborations have entered into an agreement to jointly analyze data from the GEO, LIGO, and Virgo detectors
Sharing of data started in May 2007 when Virgo commenced its first long science run in coordination with LIGO’s fifth science run
A joint run plan committee is coordinating detector operation and commissioning schedules to improve the prospects for detection
Joint data analysis meetings are already taking place
Joint data analysis exercises with simulated and small amounts of real data have already been performed for the inspiral, burst, and stochastic analysis
EGO Seminar May 3

7. Project 2b
~72 hours of coincident data from the G1, H1, H2, L1, and V1 detectors from September 2006 (LIGO’s fifth science run and Virgo’s first weekly science run)
Data shifted in time by a secret amount
Perform “end to end” analysis using one of the methods
Produce upper limits for a variety of simulated burst signals
Study coherent methods
Identify options and issues for the upcoming joint analysis
1.2 days
1.4 days
1.89 days
2.1 days
2.2 days
1.8 days H1 V1 L1
EGO Seminar May 3

8. Coincident analysis Applied the excess power search algorithm
Searches for statistically significant signals using a time-frequency decomposition of the data
Tests for coincident events between multiple detectors
Searched from 800 to 2000 Hz
EGO Seminar May 3

9. Pairwise coincident sensivity
Searching for gravitational wave bursts with the new global detector network
2007 May 3
Pairwise coincident sensivity
Based on previous investigations, we study the union of all pairwise coincident triggers
The detection threshold in each detector was set to maximize detection efficiency for all simulated waveforms at a false event rate of 1 uHz.
The resulting efficiencies are compared with the efficiency of using only the two 4km LIGO detectors.
EGO Seminar May 3
Shourov K. Chatterji - EGO Seminar - Cascina, Italy

10. Coherent analysis Naturally handles arbitrary networks of detectors
Coherent sum:
Find linear combinations of detector data that maximize signal to noise ratio
data = response x signal noise
X = F h + n
coherent null
N-2 dimensional null space
detector data
Null sum:
Linear combinations of detector data that cancel the signal provide useful consistency tests.
coherent sum
2 dimensional signal space
EGO Seminar May 3

11. Coherent analysis Applied the Coherent WaveBurst search algorithm
Considered the performance of different network configurations
Studied the frequency band from 256 to 2048 Hz
Both the sensitivity and the stationarity of a detector impacts the detection efficiency of the coherent waveburst anlaysis
Signal amplitude required for 50 percent detection efficiency
network
sg361q9
sg849q9
sg1615q9
live time
sec
H1xH2
11x10-22
16x10-22
31x10-22
182772
L1xH1xH2
8x10-22
14x10-22
37x10-22
157599
L1xH1xH2xV1
9x10-22
17x10-22
40x10-22
104062
L1xH1xH2xG1
41x10-22
140351
L1xH1xH2xV1xG1
42x10-22
102907
EGO Seminar May 3

12. Comparison of detection efficiencies
Searching for gravitational wave bursts with the new global detector network
2007 May 3
Comparison of detection efficiencies
Detection efficiencies of pairwise coincident search shows good performance relative to coherent search from 800 to 2k Hz
Potentially due to larger frequency range of coherent search or effect of non-stationarities on coherent search
EGO Seminar May 3
Shourov K. Chatterji - EGO Seminar - Cascina, Italy

13. Note
The following slides on coherent consistency testing refer to work that is not yet complete.
If results are available in time to be approved for Amaldi, they will be included here.
If not, these simulated results will be used to illustrate the method.

14. Coherent consistency tests
Coherent methods also off the possibility of consistency tests
Test null stream for consistency with noise over sky
c2 / DOF consistency with a GWB as a function of direction for a simulated supernova (~1 kpc)
Interference fringes
from combining signal in two detectors.
True source location:
intersection of fringes
c2 / DOF ~ 1
GWB: Dimmelmeier et al. A1B3G3 waveform,
Astron. Astrophys (2002) , SNR = 20
Network: H1-L1-Virgo, design sensitivity
EGO Seminar May 3

15. Example consistency sky maps
Simulated gravitational-wave burst
Simulated coincident glitch
EGO Seminar May 3

16. Source position recovery
Given a statistically significant and consistent signal
What direction on the sky maximizes the coherent sum?
What direction on the sky minimizes the null stream?
Sky position estimates typically fall along rings of constant time delay between sites, leading to poor direction.
Other statistics or image processing techniques may be needed
Null energy Enull varies slowly along rings of constant time delay with respect to any detector pair.
Noise fluctuations often move minimum away from source location (pointing error)
source
location
EGO Seminar May 3

17. Coherent sky position recovery example
Comparison of sky position recovery by Coherent WaveBurst algorithm for LIGO only and LIGO/GEO/Virgo networks.
H1xL1
L1xH1xH2xV1xG1
EGO Seminar May 3

18. Waveform recovery
Given a statistically significant and consistent signal and a best guess sky position
Solve the matrix equation x = Fh + n for the waveforms h
Find the psuedo-inverse R such that RF = I
Network: H1-L1-GEO
GWB: Zwerger-Muller A4B1G4, SNR=40 [Astron. Astrophys (1997)]
Recovered signal (blue)
is a noisy, band-passed
version of simulated GWB signal (red)
Injected GWB signal has hx = 0.
Recovered hx (green) is just noise.
EGO Seminar May 3

19. Summary
Coincident and coherent analysis tools are now available to take advantage of data from networks of detectors
The union of pairwise coincident triggers from the LIGO/Virgo network provides an improvement in detection efficiency and coverage over the LIGO only network
Coherent methods naturally handle arbitrary networks of detectors and provide increased detection confidence, consistency tests, and parameter estimation.
Coincident and coherent analysis methods complement eachother and can be used as part of a hierarchical search
We are ready to start analyzing joint GEO, LIGO, and Virgo data
EGO Seminar May 3

20. Outlook Most recent official detector noise spectrums will go here
Tentative timeline for detector operations including enhanced and dvanced detectors will go here.
EGO Seminar May 3

21. End

22. Initial joint working group
Need to plan for joint analysis as soon as possible
First meeting held in June 2004
Investigations using simulated detector noise
Develop common language to describe searches
Learn from and gain confidence in search algorithms
Demonstrate ability to analyze each other’s data
Confront different sample frequencies, file formats, etc.
Quantify the benefit of the LIGO/Virgo network
Investigations using real detector noise
Learn to handle data quality and veto information
Confront non-ideal behavior of real detectors
Identify options for future joint analysis
EGO Seminar May 3

23. Project 1A 3 hours of simulated H1 and V1 detector noise
A variety of simulated waveforms:
Binary neutron star inspirals
Unmodeled Bursts:
Core-collapse supernovae
Gaussians
Sinusoidal Gaussians
Applied a variety of LSC and Virgo search algorithms
Characterized and compared search algorithm performance
Burst and inspiral results presented at GWDAW 9 and published in corresponding conference proceedings
Burst: Class. Quant. Grav. 22, S1293 (2005)
Inspiral: Class. Quant. Grav. 22, S1149 (2005)
EGO Seminar May 3

24. Project 1A burst results
EGO Seminar May 3

25. Project 1B
24 hours of simulated triple coincident (H1, L1, and V1) data
Injection populations of simulated waveforms
Binary neutron star inspirals from M87 and NGC 6744
Supernovae in the direction of the galactic center
Demonstrated benefit of union of two detector networks
Demonstrated sky position reconstruction and astrophysical parameter estimation
Joint burst and inspiral results presented at Amaldi 6 and published in corresponding conference proceedings
J. Phys. Conf. Ser. 32, 212 (2006)
EGO Seminar May 3

26. Project 1b results
For coincident searches, union of double coincident detector networks provides improved performance
Burst results for simulated supernovae in the direction of the galactic center at a 1 mHz false rate:
Inspiral results for simulated signals from M87 and NGC 6744 at an SNR threshold of 6:
EGO Seminar May 3

27. Timing based recovery of sky position
1 ms Gaussian signals recovered within ~1 degree of true position using Gaussian matched filter and timing based recovery of position
Rings of constant time of arrival delay between detectors
Source location
Plane of detectors
Mirror image source location
EGO Seminar May 3

28. Project 1 results
Comprehensive papers describing work on simulated data
Expand on previous work incorporating lessons learned
Burst: arXiV:gr-qc/
Investigate robustness over large signal space
Investigate intersection and union of algorithms
Inspiral: arXiV:gr-qc/
Detailed comparison of analyses pipelines
Bayesian estimation of source parameters
EGO Seminar May 3

29. Project 2a First exchange of real data occurred on 2006 June 15
~3 hours of non-coincident G1, H1, and V1 data including:
Periods of lock loss
Periods of good and bad data quality
Hardware injections of burst and inspiral waveforms
Data quality and veto segment information also exchanged
Applied LSC and Virgo search algorithms
Results shared at a working group meeting on 2006 June in Orsay, France
EGO Seminar May 3

30. Project 2b
~72 hours of coincident data from the G1, H1, H2, L1, and V1 detectors from September 2006 (LIGO’s fifth science run and Virgo’s first weekly science run)
Data shifted in time by a secret amount
Perform “end to end” analysis using one of the methods
Produce upper limits for a variety of simulated burst signals
Study coherent methods
Identify options and issues for the upcoming joint analysis
Results to be presented at Amaldi7 in Sydney, Australia
EGO Seminar May 3

31. Coincident and coherent analysis
Coincident methods
Test for time and/or frequency coincidence between detectors
Performance constrained by least sensitive detector
Difficult to select detector dependent significant thresholds for optimal performance
Coherent methods
Test linear combinations of detector data based on:
Frequency dependent sensitivity of each detector
Expected response of each detector to gravitational waves from a particular direction on the sky
Naturally handles arbitrary numbers of detectors
Naturally handles detectors with different sensitivities
Three different applications of coherent methods:
Event detection
Consistency testing
Source reconstruction
EGO Seminar May 3

32. General case of non-aligned detectors
Y. Gursel & M. Tinto PRD (1989)
“Near Optimal Solution to the Inverse Problem for GWBs”.
Full solution requires at least three non-aligned detectors
First solution of the problem for the three detector case
Procedure:
Use a linear combination of time shifted signals from two detectors to predict the signal in the third detector
For each direction on the sky, compute the prediction error
Find the direction on the sky where the prediction error is minimized
Yields best fit sky position
Also yields both polarizations of gravitational waveforms
This is equivalent to the maximum likelihood approach
EGO Seminar May 3