LSC-Virgo Burst Analysis Working Group

Navigation

Burst Group Home
Wiki
ExtTrig Home
CBC Group Home
LSC, LIGO, Virgo

Documents

'How-to' docs
Agendas/minutes
[2008, 2007, earlier]
Talks/Posters [pre-wiki]
Papers
Paper plans
White papers
ViewCVS

Investigations

Analysis projects
Old notebook [General, S2, S3, S4, S5]
Virgo workarea
External collabs

Review

Main review page
Telecons

Segment Lists in the S4 LIGO-GEO Analysis

Description of the Problem

There are two issues with the segments lists used for the LIGO-GEO analysis:

  1. Apparently 10 days of bad GEO data were included into the iWB-CP analysis. Also there is a bug in Siong Heng's script which handles Laura Cadonati's trigger tables. The result of the bug is that the H1/H2 consistency test was applied incorrectly.
  2. Even after removal of the 10 days from the iWB-CP segments, the iWB-CP and cWB segment lists disagree at the level of a few hours (1%). In particular, there is about 1% of data analysed by cWB that was not analysed by iWB-CP, despite the fact the the cWB analysis was run later after additional DQ flags had been created.
We need to correct the segment lists and update results (trigger lists, efficiencies, etc.) before releasing the paper.

Original Post by Klimenko

(The following section is the version 1.6 of this page taken from CVS. It is included here for reference.)

Description of the Problem

Apparently 10 days of bad GEO data were included into the iWB-CP analysis. Also there is a bug in Siong's script which handles Laura's trigger tables. The result of the bug is that the H1/h2 consistency test is applied incorrectly. To address this problems the following work has been done:

a) Igor and Sergey calculated the live time and compared segments used by both pipelines.

There is a difference of about 3 hours between the final iWB+CP and cWB segments. The difference is due to different segment lists from GEO used in the analysis. The incoherent pipeline was run much earlier then the coherent pipeline. The following iWB+CP GEO segment list was used in the analysis. It requires that only GEO segments with duration greater then 300 sec are used. For the coherent analysis we used segment lists posted on the following web page: May, 2006, GEO segments. We did not use >300 seconds requirement in the cWB analysis. Igor has checked that the difference between cWB segments and iWB-CP segments has no segments longer then 300 seconds.

b) Given the common segment list new trigger tables were generated for the incoherent analysis and new efficiency plots produced (Laura and Siong)

Segment Lists

iWB-CP Pipeline

Igor Yakushin explained the method for determining the iWB-CP segment lists in this email . Four different input segment lists are used: category 1 and category 2 segments for H1-H2-L1 and for H1-H2-L1-G1. (It is not clear why H1-H2-L1 was used.) Yakushin and Sutton verified independently that the combined files ligo-geo.run.jobs.txt, ligo-geo.pp.txt, and segment.lst are correctly constructed from the four input segment lists. Briefly, the procedure is

The final list, segment.lst, contains the ten days of data that were supposed to be left out of the analysis (times before GPS 7XXXXXXXX).

H1H2L1 H1H2L1G1 intersection
cat 1 file: S4H1H2L1v07_segmentlist.txt file: S4H1H2L1G1v07_nochi2cut.txt file: ligo-geo.run.jobs.txt
livetime: 1393209 s / 387.00 hr livetime: 1392636 s / 386.84 hr livetime: 1328898 s / 369.14 hr
min dur: 1 s min dur: 300 s min dur: 284 s
cat 2 file: S4H1H2L1v07_segveto.out file: S4H1H2L1G1v07_chi2cut.txt file: ligo-geo.pp.txt
livetime: 1355934 s / 376.65 hr livetime: 1292647 s / 359.09 hr livetime: 1221178 s / 339.22 hr
min dur: 1 s min dur: 300 s min dur: 1 s
cat 1 + cat 2 file: segment.lst
livetime: 1202645 s / 334.07 hr
min dur: 1 s

File descriptions:

  1. S4H1H2L1v07_segmentlist.txt : Taken from WaveBurst CVS. Header info:
    # Segment list file created Thu Aug 04 11:50:22 AM EDT 2005 by dicredic on ulysses.phy.syr.edu
    # Input: S4H1H2L1v07_segs.txt (run S4, ifos H1H2L1, version 07)
    # Don't analyze data with the following flags: H1:INJECTION_BURST H1:INJECTION_INSPIRAL H1:INJECTION_PULSTART H1:INJECTION_STOCHASTIC H1:OUTSIDE_S4 H1:PRELOCKLOSS_30 H1:SEISMIC_0D9_1D1 H1:ADC_OVERFLOW H1:JET H2:INJECTION_BURST H2:INJECTION_INSPIRAL H2:INJECTION_PULSTART H2:INJECTION_STOCHASTIC H2:OUTSIDE_S4 H2:PRELOCKLOSS_30 H2:SEISMIC_0D9_1D1 H2:ADC_OVERFLOW H2:JET H2:WIND_OVER_35MPH L1:INJECTION_BURST L1:INJECTION_INSPIRAL L1:INJECTION_PULSTART L1:INJECTION_STOCHASTIC L1:NO_DATA L1:NO_RDS L1:OUT_OF_LOCK L1:OUTSIDE_S4 L1:PRELOCKLOSS_30 L1:ADC_OVERFLOW
    # Total duration of segments to be analyzed: 1393209
    
  2. S4H1H2L1G1v07_nochi2cut.txt: Taken from ldas-pcdev1.ligo-la.caltech.edu:/archive/home/igor/S4/offline/LG_S4_seglists/. Header info:
    # Segment list file created Mon Aug 08 16:02:07 BST 2005 by siong on gravitar.astro.gla.ac.uk
    # Input: S4H1H2L1G1v07_segs.txt (run S4, ifos H1H2L1G1, version 07)
    # Minimum segment length to analyze: 300
    # Don't analyze data with the following flags: H1:INJECTION_BURST H1:INJECTION_INSPIRAL H1:INJECTION_PULSTART H2:INJECTION_BURST H2:INJECTION_INSPIRAL H2:INJECTION_PULSTART L1:INJECTION_BURST L1:INJECTION_INSPIRAL L1:INJECTION_PULSTART L1:NO_DATA L1:NO_RDS L1:OUT_OF_LOCK
    # Total duration of segments to be analyzed: 1392636
    
  3. S4H1H2L1v07_segveto.out : Taken from ldas-pcdev1.ligo-la.caltech.edu:/archive/home/igor/S4/offline/. No header info.

  4. S4H1H2L1G1v07_chi2cut.txt : Taken from ldas-pcdev1.ligo-la.caltech.edu:/archive/home/igor/S4/offline/LG_S4_seglists/. Header info:
    # Segment list file created Mon Aug 08 16:02:25 BST 2005 by siong on gravitar.astro.gla.ac.uk
    # Input: S4H1H2L1G1v07_segs.txt (run S4, ifos H1H2L1G1, version 07)
    # Minimum segment length to analyze: 300
    # Don't analyze data with the following flags: G1:CALIB_CHISQR_V1 H1:INJECTION_BURST H1:INJECTION_INSPIRAL H1:INJECTION_PULSTART H2:INJECTION_BURST H2:INJECTION_INSPIRAL H2:INJECTION_PULSTART L1:INJECTION_BURST L1:INJECTION_INSPIRAL L1:INJECTION_PULSTART L1:NO_DATA L1:NO_RDS L1:OUT_OF_LOCK
    # Total duration of segments to be analyzed: 1292647
    
  5. segment.lst : Taken from Igor Yakshin's trigger set page (runs 17, 17z) at http://touro.ligo-la.caltech.edu/~igor/S4/p3/OUTPUT_LIGO-GEO.quadruple.2/ No useful header info.
The final list of analysed segments, segment.lst, contains data from one 600 sec period that is not included in the file iWB_segment_with10days.lst originally posted by Klimenko. The times are:
   794896891   794897190         299
   794897191   794897210          19
   794897212   794897374         162
   794897376   794897381           5
   794897383   794897491         108
It is not yet clear if this data was actually analysed.

cWB Pipeline

The coherent analysis used the following segment lists:

data set file livetime min duration
category 1 4-col format (cWB CVS) 979085 s / 271.97 hr 300 s
category 2 4-col format (cWB CVS) 2-col format (elog) 1005414 s / 279.28 hr 1 s
intersection cWB_cat1_and_cat2.txt 935750 s / 259.93 hr 1 s

Apparent discrepancy: The livetime quoted on the cWB results page (``Search sample'' section) is 597 sec shorter than the amount in the combined cWB_cat1_and_cat2.txt file above. The number from the results page is supposed to list the amount of data processed by jobs that finished successfully, and should account for edge effects, etc. It appears at first glance that at least one cWB job failed.

Igor Yakushin has double-checked the livetime by examining the actual output files. He confirms that the original number of 935750 sec from the cat 1 + cat 2 lists is correct. He suspects that the report page was generated before the last job completed, which is why the contribution of one job was missed. For more details see his email.

Recommendation: In future analyses (including S5), both the input segment lists and the list of times actually processed successfully should be stored in CVS. Otherwise we will continue to run into problems with ``lost'' data.

Intersection of iWB-CP and cWB Segments

(Klimenko)

Taking the iWB-CP segment list to be iWB_segment_with10days.lst (with livetime 333.90 hr), the intersection of all three lists (cWB cat 1 AND cWB cat 2 AND iWB) is a common segment list with total live time 257.03 hr.

There is a difference of about 3 hours between the final iWB+CP and cWB segments. The difference is due to different segment lists from GEO used in the analysis. The incoherent pipeline was run much earlier then the coherent pipeline. The following iWB+CP GEO segment list was used in the analysis. It requires that only GEO segments with duration greater then 300 sec are used. We did not use >300 seconds requirement in the cWB analysis. Igor has checked that the difference between cWB segments and iWB-CP segments has no segments longer then 300 seconds.

Update (Sutton, 2008/05/28)

As noted, the revised iWB-CP segment list segment.lst includes data from one 600 sec segment that does not appear in iWB_segment_with10days.lst. Using segment.lst gives a new common segment list common_segments_v2.lst that is larger by 577 sec, all from this additional segment.

data set file livetime min duration
intersection: iWB and cWB common_segments_v2.lst 925883 s / 257.19 hr 1 s

Since the new common segment list is a superset of the old common segment list that was used to re-compute efficiencies and trigger sets, and since the amount of additional data is so small, there is no obvious need to re-compute trigger sets and efficiencies with the new common list.

Production Triggers

Common-segment production triggers

The WBCP triggers that were processed by CorrPower and which fall within the original common_segments_257h.lst segments list can be found here. The triggers were selected from the original 3544 triggers in the previous trigger file using this python script. A total of 2702 triggers are found to be within the common data segment times (same number as the Matlab script). They have been checked to correspond exactly to the ones selected to fall within the common data segment times by LG_plot_S4_scatter_hist.m.

Sutton has independently verified that the 3544 iWB-CP triggers all fall in the cat 1 + cat 2 iWB-CP segment lists. He also verified that the 2702 iWB-CP triggers were drawn correctly from the original common_segments_257h.lst. An additional three iWB-CP triggers fall in the updated common_segments_v2.lst file that were not in the original list:

          #  Event ID  Tstart            Duration[s]  G1_lag[s] H1_lag[s] H2_lag[s] L1_lag[s]   Tpeak          Gamma_20  Gamma_50  Gamma_100  Gamma    hrssC_20  hrssC_50  hrssC_100  hrssC      dur [ms]     H1-H2[ms] H2-H1[ms] H1-L1[ms] L1-H1[ms] H2-L1[ms] L1-H2[ms]  R:H1-H2   R:H2-H1   R:H1-L1   R:L1-H1   R:H2-L1   R:L1-H2  R0:H1-H2  R0:H1-L1  R0:H2-L1  Gamma:H1-H2 Gamma:H1-L1 Gamma:H2-L1  H1_tpeak   H1_terr    H1_hrss    H1_ehrss    H1_fc    H1_BW     H1_SNR   H1_Fmin  H1_Fmax      H1_Tmin      H1_Tmax     H2_tpeak   H2_terr    H2_hrss    H2_ehrss    H2_fc    H2_BW     H2_SNR   H2_Fmin  H2_Fmax      H2_Tmin      H2_Tmax     L1_tpeak   L1_terr    L1_hrss    L1_ehrss    L1_fc    L1_BW     L1_SNR   L1_Fmin  L1_Fmax      L1_Tmin      L1_Tmax            INstart           INdur        INinjtime     INlagH1    INlagH2    INlagL1        INfc       INbw       GS         H1ampl       H2ampl       L1ampl     eventID     veto
(line 1527) S04-G1H1H2L1-waveb-23-bckg-001420  794896951.4375     0.1250       0.000     0.000     0.000   -15.625 794896951.4805      1.5229    1.1771    0.5033    1.5229    1.23e-21  1.49e-21  1.81e-21  3.11e-21    2.00e+01      0.9766   -0.9766    0.8545   -0.8545    7.9346   -7.9346  2.01e-01  2.09e-01  2.51e-01  2.67e-01  2.07e-01 -2.57e-01 -2.09e-02  5.79e-02  3.90e-02    0.0000    3.0618    1.5069   794896951.4684   0.0000  0.000e+00  0.000e+00     0.00     0.00         0.00   960.00  1024.00 794896951.4678 794896951.5140   794896951.5167   0.0001  1.017e-21  3.769e-22     0.00  3086.00         3.04  1000.00  4086.00 794896951.4726 794896951.5312   794896935.9074   0.0002  7.520e-22  3.206e-22     0.00  3086.00         2.65  1000.00  4086.00 794896951.4725 794896951.5334   794896951.4375     0.1250          0.0000      0.0000     0.0000     0.0000   -15.6250     992.000     64.000     1.623e+00    3.183e-21    6.213e-22    1.038e-21    8.400e-22     2066
(line 1528) S04-G1H1H2L1-waveb-23-bckg-001421  794897067.0938     0.0781       0.000     0.000     0.000   -15.625 794897067.2139      1.1760    1.5664    0.6787    1.5664    1.11e-21  1.49e-21  1.92e-21  4.11e-21    5.00e+01     -0.6104    0.6104   -2.4414    2.4414   -4.2725    4.2725  1.18e-01  1.24e-01 -1.39e-01  1.26e-01 -1.69e-01 -1.45e-01  1.64e-02 -2.37e-02 -6.05e-02    1.8485    0.0000    2.8507   794897067.1201   0.0000  0.000e+00  0.000e+00     0.00     0.00         0.00   864.00   960.00 794897067.1148 794897067.1508   794897067.1163   0.0000  0.000e+00  0.000e+00     0.00     0.00         0.00   864.00   960.00 794897067.1153 794897067.1552   794897051.4877   0.0000  0.000e+00  0.000e+00     0.00     0.00         0.00   864.00   960.00 794897067.1138 794897067.1545   794897067.0938     0.0781          0.0000      0.0000     0.0000     0.0000   -15.6250     912.000     96.000     1.667e+00    4.341e-21    5.219e-22    1.012e-21    6.515e-22     2067
(line 2187) S04-G1H1H2L1-waveb-23-bckg-002036  794897080.2656     0.0547       0.000     0.000     0.000    37.500 794897080.2665      1.1822    0.5232    0.7101    1.1822    1.08e-21  1.43e-21  1.63e-21  3.56e-21    2.00e+01      0.4883   -0.4883  -10.2539   10.2539  -10.1318    9.6436  2.23e-01  2.23e-01  2.21e-01  2.38e-01 -1.73e-01  1.71e-01  2.23e-01  1.95e-02 -9.83e-02    2.3712    1.1754    0.0000   794897080.3097   0.0001  1.208e-21  4.819e-22     0.00  2870.00         4.00  1216.00  4086.00 794897080.2780 794897080.3096   794897080.2991   0.0000  0.000e+00  0.000e+00     0.00     0.00         0.00  1152.00  1280.00 794897080.2802 794897080.3005   794897117.7799   0.0000  0.000e+00  0.000e+00     0.00     0.00         0.00  1152.00  1280.00 794897080.2795 794897080.3132   794897080.2656     0.0547          0.0000      0.0000     0.0000     0.0000    37.5000    1216.000    128.000     1.678e+00    4.869e-21    9.729e-22    8.530e-22    7.409e-22     2068
These three additional triggers fall into the first segment that was missed in the original common list. The first of the three has R0<0 and so fails this cut. The other two triggers pass the R0 and H1-H2 amplitude cuts. Note that they are all background triggers with Gamma < 1.6. We can probably safely ignore these.

Update (Sutton, 2008/06/02)

None of the zero-lag triggers fall into the segment that is missing from the original common_segments_257h.lst segments list.

Plot scripts and trigger selection

The plot scripts and accompanying data can be found in the paper CVS archive. The trigger counts in Table 1 is determined by the LG_plot_S4_scatter_hist.m script. The script outputs the following numbers:

  Number of triggers within common segment list before H1-H2 cuts (amplitude and R0): 2702
  Number of background triggers after H1-H2 cuts (amplitude and R0): 881
  Number of background triggers with Gamma > 3: 1
  Number of background triggers with Gamma > 4: 0

  Perform KS test and spit out P value.
  H = 0 means null hypothesis is upheld.
  H = 1 means null hypothesis is rejected.

    H = 0
    P = 0.17965807867471
Update (Sutton, 2008/06/02)

I have independently verified these numbers. They are produced using the original common_segments_257h.lst segments list. I also examined the plotting script LG _plot_S4_scatter_hist.m and it looks fine. However, I have no record of how the combined livetime vs lag file was produced, and so cannot verify that it is correct. This file is used for weighting the contribution of different lags in the histograms in Figure 4b.

Simulation Triggers

Common Simulation Triggers

Given the original common_segments_257h.lst file, new simulation trigger tables were generated by Igor for the incoherent analysis (script used for the trigger generation). The triggers are located in the CIT directory:

  /archive/home/igor/OUTPUT_S4_LG_SG21_run24a.merged/iWB/resampled_revised 

New efficiency plots will be produced by Laura and Siong. They can be compared to the efficiency tables produced by Igor (script used for generation of tables).


Update (Cadonati, 2008/05/04)

Injection lists

  1. Igor's original link with the triggers from iWB, the segments analyzed and the injections in those segments (warning: old lsc password)
    Laura's copy of the injection list from Igor
    reformatted version of the list for input in Laura's code. Used: FormatInjectionList.pl It is essentially only to make the list readable in a browser, with correct column alignment.
    This is the segment list

  2. resampled by Brennan Hughey to address the issue of the cos(theta) distribution at the time of MDC production
    This is the segment list

  3. is the set of common injections with cWB, a subset of the list produced by Brennan.
    Laura's copy of the list, the file actually used to derive efficiencies.
    This is the segment list

Working areas and efficiency tables, logs

  1. The original injection list: not resampled, using Igor's initial injection list.
    Efficiency table, plot, Fit log and parameters

  2. The resampled injection list: using the list from Brennan Hughey.
    Efficiency table, plot, Fit log and parameters
    To complete the table on the paper, we needed efficiencies at lower gamma thresholds - results are:
    Efficiency table, Fit logs: gammaThr=0 gammaThr=3 gammaThr=4

  3. The final injection list: after resampling and intersection with the segment list common for iWB and cW, using the list from Igor.
    Efficiency table, plot
    Fit logs: gammaThr=0 gammaThr=3 gammaThr=4

Update (email from Klimenko, 2008/05/29)

Hi Patrick,
the attached plot shows how the loud events are rejected by different
selection cuts.

I have selected SG at 1053Hz (type=3). Total 564 injections are done
for each strain bin. This is shown by a histogram with no fill.
The colored histograms show events detected in each bin for
3 different cases:

1) red - no cuts (only a cut on abs(Tinjected - Tdetected)<0.1 is applied) 
  efficiency for loud events is very close to 100%
2) green - Lk>36 is applied - some injected events are lost but not the
loud ones.
3) blue -  + sqrt(rho_eff)>3.4 - now some loud events are lost.

So the major contribution cames from the small value of cc reconstructed 
for some loud events. This is not surprizing that this is happening at
minimum of the detector sensitivities where the effect of the regulator
is more visible. Also cc depends on how the correlated energy is
reconstructed. More accurate procedure was implemented for later version
and it is described in the most recent cWB note.
Sergei's plot

Update (Sutton, 2008/06/03)

I have looked at the antenna repsonses of the missed loud injections for both pipelines as reported on Igor's efficiency tables. The situation is reversed from studies using the older versions of the tables:

$Id$