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

Minutes of 2008-Apr-03 S5 QPipeline Review Teleconference

Attendance

Shourov Chatterji, Jonah Kanner, Dave Reitze

Minutes by Jonah Kanner.

Announcements

Discussion


http://ldas-jobs.ligo.caltech.edu/~shourov/s5qreview/triggers/s5/box.html

Dave:  At some future date, we should consider the status of the review.  From my perspective, I've seen no major problems so far.

Shourov:  These are the one year open box results.
The black dots represent 1000, .5s time shifts.  Vertical band on Livingston triggers arises because of a lack of null stream.
Other cut is an attempt to follow countour of glitches.  Triggers that "fall off the bottom" of H-L plot motivates looking at 
H1H2 only case for detections.  On those, simple cut on correlated energy.  H1H2 cut corresponds to where diagonal H-L cut
intercepts x-axis.

Dave:  You didn't say much about the thresholds.  Should I assume:  you looked at non-zero lag triggers, and drawn thresholds to 
likely exceed zero lag events

Shourov:  that's right.  the burst group goes for 5 events in 100 time lags - in these plots we see 1000 time lags.  
There are 63 unique triple coincidence triggers in these plots
  -  the h1h2 only plots have a more conservative threshold due to the problems with shifting h1h2.

   -  The recoverable injections in the triple coincidence plot that go off the bottom suggest that there are recoverable 
      injections (signals) that won't have an L1 trigger

Dave:  Each dot represents one or more than one

Shourov:  Actually, there can be multiple dots for a single event.  There are 63 UNIQUE triple coincidence events above
threshold.

Jonah:  Do we really want to consider H1H2 event candidates?  Could it make our upper limit logic more complicated.

Shourov:  Actually, it makes sense to include this for upper limit.  For detection, it might be hard.  
About 3% of injections are recoverable by H1H2, but not by L1.

Dave:  It would be hard to claim a detection w/ H1H2

Shourov:  I think it would depend on making a strong case with other factors.  Signal morphology matters - if you see a 
distinct shape in the Q plane, and it does not "look like" a glitch, it might be a good case

Dave:  What about cosmic rays?

Shourov:  I think Ray Frey is looking at that.

Shourov:  Moving down the page, we see zero-lag triggers at various stages of cuts.  There is nothing above the blue cuts.

Dave:  Looking at category 3 on the left hand plot (H1H2) - we see an event that looks like an outlier in coherent energy

S:  If you look back at the injections, that dot falls well below the injection "line" (ratio of corr/coh energies for inj.). 
 The ratio of corr to coherent energies suggests that this is not a real signal. 
 One thing that makes it hard to follow-up events is deciding what statistic to use for "loudest."  I made the choice to use
the H1H2 correlated energy as my measure of loudness, even for triple coincidence events.  

S:  At the bottom, there is a table of the loudest events.  Making we can pick out this event.

S:  Maybe what I should do is follow-up loudest 100 in corr and 100 loudest in coherent energy

D:  you should follow it up - see what happened

S:  Further down page, we see data for H1H2 when L1 is not in science mode.  I keep the thresholds exactly the same.  
    Again, nothing passing threshold.  I produce the same distribution plots as I did for playground.

D:  Looking at histogram, why does FAR go down when you get to lower frequencies, the FAR seems to go down.  
    That doesn't seem to make sense.  H1H2 sig, all triggers, plot on the upper right hand side

S:  i think its due to thresholds of the search.  trigger production had some baseline.  

D:  Ok, I think i have to digest that

S:  The next plot - the probability of getting that result, peaks.  you can see the error bars on the black dots get bigger as you
get into the tail.  

D:  there is an interesting statement - equinox event - is 30 microhertz - for H1H2 only - 
but with an L1 consistency we get one FA every 13 years

S:  that's a rough estimate

S:  Next we see the high freq. H1H2 triggers - they look much more consistent - the probability of the excess only goes to 1/100

S:  Following this, we have sigmoid efficiency curves.  We can see this both for H1H2 and for H1H2L1.  These are plotted for the 
chosen thresholds. 
  The first section has triple coincident times, the 2nd section has only double coincident

D:  I think Peter Shawhan argued that the points on the sigmoid curves should be more dense.  I remember you argued 
aganist that.

S:  Yes, these hrss50% marks are rough.  What should be done is to put it through sigmoid fitting routines with more
accurate estimates.  
The ultimate goal is the upper limit curves, which involves combining results of multiple pipelines.  
this requires actually running at the same scale factors for multiple pipelines.  The result does not depend on sigmoid fits.
sadly, the Q amps. do not align w/ the cWB scale factors

D:  Is there a plan for that?

S:  I claimed that Q pipeline should not have to re-run all the results - i claimed that cWB would have an easier time of it

D:  is there consensus?

S:  we're at a stalemate right now.  i have a few suggestions, which i am passing around to burst mailing list, etc.  scientifically,
 we want to have as many points as possible.  but, at what cost??

S:  The re-running is essentially the last thing that needs to be done for the first year analysis.

S:  Block normal opened the box, and there's nothing there.  block normal is now at the point we're they are starting to 
run simulations.

J:  why set thresholds at all?  what happened to loudest event?

S:  it was the combing pipelines that motivated it

S:  I still think it makes sense to push forward with significance and re-mapping.

D:  Ok.  So we're do we stand?  what is left to do for the 1st year analysis?

S:  Modulo re-running with new scale factors, i think the first year is done.

D: what i would like to do, is go back and look over minutes, and see if there are other things left to do.  for example, that 
there were non-integer time lags and such.  i want to - and i hope others do if they have the time - is 
to look over things a final time again.  speaking personally, i think things look very good and there are not major issues. 
 but, i would like to hear what jonah and isabel have to say about that.
for the next meeting, we should be prepared to address anything outstanding.

J:  yes, that seems fine.

D:  there is the 2nd year analysis, of course, and that will be a lot more work.  the intention is to bring virgo into that, right?

S:  well, i;ve heard some skepticism for that

D:  Well, the idea is to settle the equinox once and for all.  you don't need virgo for that.  Again, i think that for next week we should be prepared to give this a final pass through.  I'm of the opinion we should try and 
move this forward and make a report. 

J:  It sounds like we are setting as an action item going back through our notes to be ready for a final pass through next week.  

Action Items

$Id: minutes_20080403.html,v 1.1 2008/04/24 17:19:39 leonor Exp $