LSC-Virgo Burst Analysis Working Group


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


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


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


Main review page

Review Committee Meeting Monday 20 February 2006 09:00 PST / 12:00 EST

Minutes: Monday 20 February 2006 09:00 PST / 12:00 EST

Agenda and Contact Info


  1. Survey of reviewing tasks for APS presentations to be completed before the LSC meeting
  2. Continue the review of the GRB GWB search

Contact Info

AccuConference teleconferencing service:
   Phone: 1-800-704-9896, participant code: 038621#
   International callers ++1-404-920-6472 with same code


  1. Survey of reviewing tasks.
    • As of now I have talk overviews for:
      • S5 Untriggered Search (Igor)
      • S4 Untriggered Search (Peter)
      • GRB Search (Isabel)
      • SGR 1806-20 Search (Luca)
      but not
      • Q-Pipeline (Shourov)
      • Results II (Sutton)
      Patrick Sutton's talk will contain material that is in other presentations so it is not important to get this. Shourov has just circulated technical information about the Q-pipeline search and a link to a broad overview of what he wishes to present. ACTION: Shourov will put together a talk overview in the next couple of days.
    • Need to commission two new search reviews:
      1. Q-Pipeline. Internal code review is being conducted by Isabel. ACTION: Isabel will begin to compile some report describing the internal review actions. Review webpage now contains a section for the Q-pipeline review.
      2. SGR 1806-20 Search. Need to initiate both an internal code review and a full external review. Code is quite minimal (few hundred lines of Matlab). Shourov will do internal code review "in his spare time". Review committee needs detailed technical documentation describing the search algorithm, details, validity/sanity checks, etc. ACTION: Luca will compile such a technical document so that we can kick-off the review.
    • We will not be discussing the review priority in this meeting. Suggestion is to do it in Tuesdays Burst telecon since the Burst group should be setting the overall priority when possible.
  2. Continue the review of the GRB GWB search
    • Cross correlation statistic: Eq. (7) is not quite right... the quantities h1(i) and h2(i) are the mean-subtracted timeseries in Matlab. Shouldn't matter much because the data is high-pass filtered (but it does kill one degree of freedom).
    • Should there be a little slop in the timing? CorrPower allows around 1ms difference in start time. Apparently this was to account for calibration inaccuracies (esp. when running on uncalibrated data). Szabi says that the intrinsic timing uncertainty is not significant. Biggest worry is that a slightly incorrect phase calibration might occur. Note that the simulations inject with the same response phase as the analysis uses so it would be insensitive to any errors. Thus need to assess this as a systematic error. ACTION: systematic uncertainties need to propogate possible phase errors in calibration (if they don't already).
    • Figure 14 shows that even for very short signals (~2ms) the optimal cc duration is greater than 10ms! This seems counterintuitive.... Isabel claims it is just the 1/sqrt(n) effect and has nothing to do with non-Gaussian noise in the detector. In that case, should be able to derive the length from the statistical properties of the cc stat. Perhaps it is a overlap issue (want to make sure that signal is entirely within the duration of the cc statistic)? Probably not since 2ms is much shorter than the "best" length of 15ms. This should be looked at more but the bottom line here is that 25ms does a pretty good job of detecting the SG signals and a 100ms integration would pick up longer signals. (What about signals much longer than this, like 1s? A: do a different search.)
    • Analysis is repeated off-source to get a distribution of the max cc statistic for both the 25ms and 100ms integration lengths. Use this to assign a probability to each loudest event. Distributions are the same whether time slides are done or not. The probability assigned should be uniformly distributed in 0 to 1 in the case of no signal. Plot cumulative distribution of these probabilities for all 54 IFO pairs for the GRBs. Null hypothesis should be a straight line. Results are shown in Figs. 17 and 18.

      A single loud signal should shift the point on the lower left far to the left (and hence above the straight line curve). This would be used to test for a signal.

      More generally, an excess of semi-unlikely events would cause points to depart from the straight line somewhere nearer the middle. But would this not also cause a departure for the point on the lower left? Under what situation would the "least probable" point be consistent with the null hypothesis but yet there is an excess in the middle of the plot? ACTION: Isabel will show plausible scenarios of what we might see if there are a number of very-marginal signals and what curves these scenarios might produces. A graphical illustration of the method (even cartoon graphics) would make things clearer.

      Note: Soumya's method provides an alternative approach.

  3. AOB: How much longer do we expect to be going through Isabel's document? A: Probably one more telecon, but we'll need to revisit periodically to follow up on issues raised.
$Id: minutes-2006-02-20.html,v 1.2 2006/02/21 03:59:19 jolien Exp $