Minutes of 2008-Jan-31 S5 QPipeline Review Teleconference
Shourov Chatterji, Erik Katsavounidis, Isabel Leonor, Dave Reitze
Minutes by Isabel Leonor.
Minutes from previous meetings are now posted in the S5 QPipeline review web pages.
Relevant links and agenda information for the next call will also be posted in the same location.
According to the proposed timeline, this weeks milestones included a code freeze, and completed H1H2 playground analysis including simulated signal injections with and without calibration uncertainties.
The code freeze is not in place yet. This is a critical task, but efforts were focused on finishing the H1H2 playground analysis. A code freeze is still anticipated this week, and remains a high priority.
The H1H2 1 day playground analysis has been completed.
Last week's results were also extended to include 200 time lags (every 0.5 second from -50 to +50 seconds).
All of the available MDC waveforms (Gaussians, sinusoidal Gaussians with Qs of 3, 9, and 100, and white noise bursts) have been tested. Efficiency curves and upper limits have been constructed for each waveform, after thresholding at the significance of the loudest zero lag event. Results were produced with and without the null stream veto applied. Results were also produced for the Q 9 sinusoudal Gaussians waveforms when injected with a 10 percent greater amplitude in H2 than H1. These results indicate that the null stream veto does not adversely effect detection efficiency at any amplitude, and correctly handles a 10 percent calibration amplitude uncertainty.
In anticipation of next month's focus on the triple coincident search, trigger production is already complete for the L1 raw and simulated signal injection data sets.
Trigger production for S4 has also started in order to provide a larger zero lag playground set, provide a useful means of comparing against previous searches, and hopefully test procedures for combining detector networks. Currrently, trigger production is complete for S4 H1H2 zero lag and some H1H2 non-zero lags. A first set of distributions are available.
Time-lag trigger production for the full S5 1st year analysis will begin this week once an updated segment list is constructed.
* Shourov has set up the webpage for the QPipeline review minutes and agenda: http://www.lsc-group.phys.uwm.edu/bursts/review/projects/s5-qpipe/minutes/ * Shourov made an agenda for today's telecon: http://www.lsc-group.phys.uwm.edu/bursts/review/projects/s5-qpipe/minutes/minutes_20080131.html * Shourov: No code freeze yet. * Shourov walked us through results of S5 H1H2 1-day playground analysis. Erik proposed at the last review telecon that the number of time lags be increased. This has been done, according to Shourov. He now uses 200 time lags, every 0.5 second from -50 to +50 seconds. * We discussed the resulting distributions in this page: http://ldas-jobs.ligo.caltech.edu/~shourov/s5qreview/distributions/ The left-hand plots are cumulative distributions of the coherent SNR, and the right-hand plots are differential distributions. Most of the discussion focused on the loudest event found after the veto cuts were applied, i.e. the black distribution. Erik: Has loudest event been investigated? Is there a veto that can be relevant to this loudest event? Dave: Is there a glitch taxonomy? Shourov: No, but he would love to have something like that (Shourov is co-chair of the Glitch Group). Erik: It's hard to do taxonomy from GW signal alone; need to look at auxiliary channels. * Shourov then showed us the time lag plot on the same page. Erik to Shourov: Don't you expect more correlations near or at zero lag compared to other time slides? What if you look at a lower SNR of 12? Shourov: Doesn't think it will change. The time lag plot shows that the null stream veto is very effective at getting rid of glitches. The combined effect of both null stream and signal veto cut is effective. * Shourov discussed the efficiency curves on this page: http://ldas-jobs.ligo.caltech.edu/~shourov/s5qreview/simulations/ Dave: How many injections? Shourov: 55 injections per waveform (not much). Dave: For Q=100, why is green curve low? Shourov: Matlab recycles colors. This green line is for 2000 Hz. This is probably at the edge of the search space. Also, the best curve, corresponding to 153 Hz flattens out. Q=100 is slightly off search space because this actually corresponds to a QPipeline Q' = 141. Shourov: Right-hand plots are UL plots. Shourov: Calibration uncertainty does not seem to have effect on efficiencies. Dave: How do these efficiencies compare to other searches? Shourov: Can't directly compare because this is only for the 1-day playground. Erik: Looking at hrss50 plotted with noise spectrum (on the same page), SNR of 40-50 divided by 3 (for all-sky search) = 13-18, which relates to loudest event of search. Shourov: Remapping distributions should have effect. Hope to discuss next week. Shourov: For 1-year search, how much louder will loudest event be? Dave: Only Category 1 flags have been applied here? Erik: Need to think about DQ cuts more. He (Erik) will work on this. Shourov: Will think about DQ flags. Erik: Need to look at outliers, since studying these wil probably provide good lessons for choosing DQ flags. Finally, Shourov showed results from running H1H2 pipeline on S4 data. Plots of cumulative and differential distributions are here: http://ldas-jobs.ligo.caltech.edu/~shourov/s5qreview/distributions_s4/distributions_cumulative_zoom_medium.png http://ldas-jobs.ligo.caltech.edu/~shourov/s5qreview/distributions_s4/distributions_histogram_zoom_medium.png Isabel: What is causing the peak in the differential distribution? Shourov: Not sure. It might be effect of null stream veto.