Review Committee Meeting Wednesday 30 November 2005 09:00 PST / 12:00 EST
Minutes: Wednesday 30 November 2005 09:00 PST / 12:00 EST
MinutesWe went through the presentation, highlighting "results" that had already been approved and identifying "new results" for focused attention.
- Amplitude cut on slide 12. Concerns:
- Why is a ratio of 3 needed rather than 2 (as with the LIGO-only
search)? Does this mean that the hrss estimation code used here
- We have not yet reviewed the hrss estimation code. While H1-recovered vs. H2-recovered plots look reassuring, recovered vs. injected plots show a bias. This has not yet been understood. (Is it a problem? Is it because estimation is done after whitening?) See slide 29.
- We resolved that we are still not comfortable showing results using
the amplitude cut.
- Why is a ratio of 3 needed rather than 2 (as with the LIGO-only search)? Does this mean that the hrss estimation code used here is problematic?
- Choice of thresholds on slide 13 (and as shown on later slides).
- AU-H1-H2-L1 thresholds seem OK. But why is the gamma threshold
for AU-H1-H2 chosen to be 9? Why not 10? 10 appears to be more at
the "foot" of the histogram.
Response: the choice actually gives similar expected false rates.
- Philosophical point: the purpose of adding additional detectors
is to drive the false rate to practically zero. Why is this not being
exploited in this analysis?
Response: instruments too glitchy for this to be a practical goal.
- AU-H1-H2-L1 thresholds seem OK. But why is the gamma threshold for AU-H1-H2 chosen to be 9? Why not 10? 10 appears to be more at the "foot" of the histogram.
- Slide 16 and some following slides. Cluser of high-gamma
- All of these are coming from a single GPS time. This time
corresponds to a glitch in the DARM_CTRL_EXC_DAQ channel.
- This time is not marked in the V03 data quality flags, but it
is in the V02 data quality flags. This was an oversight. It is described
as a calibration line loss. Similar problems have been seen since S3 and
are being addressed.
- Decision: the V02 data quality flags that mark this and similar
events should be applied. This will cause minimal extra dead-time and
will presumably not have an adverse effect on efficiency. These events
should not be shown at GWDAW.
- All of these are coming from a single GPS time. This time corresponds to a glitch in the DARM_CTRL_EXC_DAQ channel.
- Slide 19: Poissonian check. Has there been a check of whether
time between AURIGA triggers follows an exponential distribution?
- See slide 27. Notice the 20 sec "bump" in autocoincidence.
This is due to some environmental effect.
- Is there an effective monitor? Answer: no, especially not for
this epoch of AURIGA data.
- What are the time-slides between AURIGA and LIGO? Answer: various,
random eighths of second between 7s and 100s. Whatever causes the
20 sec bump would probably get smeared out in time-slides with LIGO.
Resulting LIGO-AURIGA triggers are probably quite Poissonian.
- See slide 27. Notice the 20 sec "bump" in autocoincidence. This is due to some environmental effect.
- Slide 24: Uncomfortable with exercise.
- Why just one draw from set of lags? Why not repeat to get a
Answer: could get a distribution but need to discard draws so that each pair is distinct.
- Dislike proposed statistical analysis because:
- Use of background in FC procedure may not be justified: background might be "polluted" by true GW signals!
- Would not be comfortable having a FC interval bounded away from zero. This has been avoided in previous LIGO results by declaring that upper bound only would be used. New method is now being adopted by LIGO.
- Not happy with test of null hypothesis method of deciding on a detection. This means that we may have a detection without further analysis. It also precludes making a detection of one gold-plated event.
- Statistical methods have not been presented to the burst group. They are too fresh.
- Why not just open the box?
If box is opened without a mature pipeline specification (including statistical analysis), a rate limit would be inadmissable for presentation or publication. Group should only do this if they do not intend to publish a rate limit.
- Why just one draw from set of lags? Why not repeat to get a distribution?
- R0 cut: is it safe in S3 (i.e., are there any sign errors)? Answer: the
calibration team has retroactively understood all sign issues in previous
science runs and has determined that the R0 cut as it is being applied to S3
data is safe.
- MDC frames: are they being correctly applied in the AURIGA analysis
system? Are they correctly checked?
- AURIGA efficiency curves with MDC frames give excellent agreement
with their efficiency curves with their own software injection code.
- Patrick Sutton has spot-checked (with an independent code) the MDC
frames to verify that what is being produced is correct.
- AURIGA efficiency curves with MDC frames give excellent agreement with their efficiency curves with their own software injection code.
- Summary: we are comfortable with the presentation except that (i)
the amplitude cut should not be included (and plots suitably modified),
(ii) the additional data quality flag should be adopted (and plots suitably
modified), (iii) the statistical interpretation should be omitted.
- Discussion of LIGO-GEO results postponed to next meeting.