1. LIGO Scientific Collaboration meeting
S5 Q Pipeline Status
Shourov K. Chatterji
Caltech
LIGO Scientific Collaboration meeting
MIT
2006 November 5

2. S5 Q Pipeline search
The Q Pipeline is being applied to search for untriggered bursts in S5 data through 2006 April 3.
There are two primary components of this analysis
H1H2 double coincident search for combined excess signal energy followed by H1H2 null stream consistency test
H1H2L1 triple coincident search for time frequency coincidence between H1H2 triggers and L1 triggers
Upper limits will be determined using the loudest event statistic
The loudest N events will be followed up QScans and a fully coherent follow-up if data is available from a sufficient number of detectors
LSC meeting 2006 November 5

3. Q Pipeline
Multiresolution time-frequency search for statistically significant excess signal energy
Projects whitened data onto an overlapping basis of sinusoidal Gaussians characterized by central time, central frequency, and Q (ratio of central frequency to bandwidth)
The template bank is constructed using a maximum mismatch approach similar to the matched filtering approach
The search is equivalent to a matched filter search for waveforms that are sinusoidal Gaussians after whitening
The reported normalized energy is measure of event significance and is simply twice the squared SNR that would be reported by a matched filter search
LSC meeting 2006 November 5

4. H1H2 Null stream veto The collocated H1H2 search has two components
H+: The optimal linear combination of H1 and H2 that maximizes the SNR of potential signals
Weighting factors are inversely proportional to power spectral density in each frequency bin
Normalized energy, Z+, is twice the quadrature sum of SNR in both detectors
H-: The H1 – H2 null stream, which should be consistent with noise for real gravitational-waves
To avoid calibration uncertainties, test normalized energy, Z-, against that expected null stream normalized energy, Z0, assuming incoherent glitches in the two detectors
Z- > a + b Z0
LSC meeting 2006 November 5

5. Analysis code
The QScan and QPipeline codes have been merged into a single code base, which is now available in the LSC MatApps CVS repository [links: code, documentation]
This code base is now installated on most LSC computing clusters and LIGO laboratory general computing networks.
Running QPipeline is now as simple as running QScan
ssh ldas-pcdev1.ligo.caltech.edu
mkdir –p ~/qpipelinetest
cd ~/qpipelinetest
~shourov/qscan/bin/qpipeline.sh \ \ .
It’s easy! Try it now!
LSC meeting 2006 November 5

6. Summary of livetime and duty cycleH1
H1H2
90.1 (60.1%)
3.3 (2.2%)
34.4 (22.9%)
5.1 (3.4%)
55.8 (37.2%)
12.7 (8.5%)
10.2 (6.8%)
7.1 (4.7%) H2 L1
Livetime in days through April 3, 2006
Duty cycle relative to start of S5 at LHO
LSC meeting 2006 November 5

7. Data quality and vetoes
The following data quality flags were used to define the segment lists for trigger generation
No other data quality flags have been applied.
No auxiliary channel vetoes have been applied.
OUT_OF_LOCK
OUT_OF_SCIENCE_MODE
PRE_LOCKLOSS_120
MISSING_RDS_C02_LX
Injection
LSC meeting 2006 November 5

8. Search parameters Using h(t) data Using burst MDCs Search space
channelNames: {'H1:LSC-STRAIN', 'H2:LSC-STRAIN'}
frameTypes: {'H1_RDS_C02_LX', 'H2_RDS_C02_LX'}
timeShifts: [+0 -1]
injectionNames: {'H1:GW-H', 'H2:GW-H'}
injectionTypes: {'SG2_S5_A', 'SG2_S5_A'}
injectionFactors: [ ]
reSampleFrequency:
blockDuration:
minimumBlockOverlap: 8
qRange: [4 64]
frequencyRange: [ ]
maximumMismatch:
analysisMode: 'collocated'
outlierFactor:
falseRate: e0
maximumSignificants: 1e5
maximumTriggers: e3
durationInflation:
bandwidthInflation: 1.0
Using h(t) data
Using burst MDCs
Search space
LSC meeting 2006 November 5

9. Post-processing parameters
High threshold veto
H+ events are excluded if they fall within two seconds of very inconsistent tiles (Z- > Z0)
Low threshold veto
H+ events are excluded if they overlap inconsistent tiles (Z- > Z0).
Low frequency, low Q cut
A significant fraction of all glitches are observed to fall in the lowest frequency, lowest Q template.
H+ events are excluded if they have a central frequency less than 100 Hz and a Q less than 14.
LSC meeting 2006 November 5

10. Trigger production
Trigger production and post-processing is visible online
Detectors
Injections
Time Shift
Amplitude x 2.5x10-21 H1 - H2 L1
H1H2
0, 0
0, -1
SG2_S5_A
1/16, 1/8, 1/4, 1/2, 1, 2, 4, 8, 16
GA2_S5_A
WNB2_S5_A +1
LSC meeting 2006 November 5

11. Distribution of H1H2 significance
Loudest event Z+ = 176
LSC meeting 2006 November 5

12. Loudest H1H2 event Loudest event is low Q burst in H1
No clear glitch in H2
Why did this pass the H1H2 null stream veto?
Potential room for improved veto tuning? H1 H2
LSC meeting 2006 November 5

13. Efficiency for sinusoidal Gaussians
LSC meeting 2006 November 5

14. Efficiency for sinusoidal Gaussians
LSC meeting 2006 November 5

15. Efficiency for Gaussians
LSC meeting 2006 November 5

16. Efficiency for white noise bursts
LSC meeting 2006 November 5

17. Predicted (“time lag”) upper limits
Decreased efficiency
Increased livetime
Crossover at 7e-22 and 0.08 events per day
LSC meeting 2006 November 5

18. Outlook Further tuning will likely improve performance
Anything that eliminates loud events will improve the resulting upper limit
Triple coincidence (at the expense of livetime)
Application of data quality flags and vetoes based on auxiliary or environmental channels
Concern about potential zero lag H1H2 surprise
Use S4 as a playground data set
Trigger production complete at 17 different time lags including zero lag
Use one day of S5 as a playground data set
Once tuning is complete, the box will be opened for the analysis through April 3.
LSC meeting 2006 November 5