Minutes of 2008-Feb-14 S5 QPipeline Review Teleconference
Shourov Chatterji, Jonah Kanner, Isabel Leonor, Dave Reitze
Minutes by Jonah Kanner.
S5 H1H2 scatterplots and significance
An investigation on a modified definition of H1H2 significance suggest that correlated energy, or perhaps a more complicate cut in the correlated vs. coherent energy plane may be a useful definition of significance.
S5 H1H2 1 year triggers
Updated distributions for time-lagged H1H2 triggers from the full 1st year analysis have been posted. Note that the zero lag triggers on this page are only from the 1 day playground.
As decided last week, the timeline has been rearranged to focus first on the triple coincident analysis in the context of the 1 day playground. At the same time, trigger production will continue for the H1H2 full 1 year analysis.
S5 H1H2L1 1 day playground analysis
A first look at the distribution of triple "coincident" triggers is posted here.
Dave: Why don't we walk through the plots that are passed around on e-mail Shourov: We'll start with scatter plots of H1H2 energies ~shourov/s5qreview/scaatter/ellipse These are plotted in terms of energy instead of energy: snr = sqrt(2*energy) Dave: The blue lines are isocontours? Shourov: It's a family of curves parametrized by a single value to define a significance We can see why correlated energy has been doing well as a statistic - in the plots a cut on correlated energy does about as well separating glitches and injections as can be done. This ellipse cut behaves about the same as a cut on correlated energy, but I thought the downward curve would be useful when adding L1. In the correlated-coherent plane, the injections tend to follow a line Isabel: When you calculate the significance, do you actually use the background?? Shourov: No. For me significance is just a number. As long as its monotonically increasing, that's all that matters here. Dave: Let me ask about S4 data. I'm looking at this outlier, and it should be followed up Shourov: Yes, it is clearly an outlier. You can find qscans linked off the S4 page ~shourov/s5qreview/s4_results If you look at the difference between S4 and the S5 playground, you'll see that S5 playground day has a very serious outlier. That may mean that S5 is not like S4. Then again, looking at the S5 1-year background, there are events louder than the outlier from the playground day. Dave: You'll do something like 200 time shifts? Shourov: No. It's too expensive. So far I've done 6. I'm hoping to do at least 10. But, with the H1H2 data set, you can't gaurentee that zero-lag is like the background. Dave: Why a nice binary sequence of time lags? Why not pick pi seconds of time lag? Shourov: That's a good idea. I'll run a weird number as well. Let's go to ~shourov/s5qreview/s5_results. Here, the red is playground day. It's one time lag in background, but it's for the whole year. Let's go to /playground_withveto/H1H2L1 H1H2 triggers are tested for time frequency coincidence with L1. Here, time lags are cheap, because you are just moving triggers around. The black have H1H2 in zero lag, just L1 is moving There's no null veto on Livingston, so the L1 significance can go quite high Vertical features in these plots are when a Hanford trigger lines up with different L triggers Dave: Are there data quality flags applied here? Shourov: Only cat. 1 data quality flags have been applied here - vetoes are not yet ready One of the main advantages of the null stream test is that it quickly cleaned up the data, without these labor issues. Dave: I'll be really interested to see how much this is going to be cleaned up when you apply vetos and category 2 DQ Isabel: These are coincident in time and frequency? Shourov: Yes, these are coincident with a window of 15 ms. The center of the tile has to overlap with some portion of the other tile. If we look at the top left plot, we can see how much the L1 test has cleaned up this plot. This is with L1 significance threshold at 10 - if you move the L1 significance up to 100. What I had hoped was to combine these statistics in one measure of significance - but that looks to be daunting. So, what I could do is threshold on L1, and then just use the ellipsoid cuts from the H1H2 to rank events. So, L1 is acting only as a veto process.