The calibration review team has been working very hard in the past months to assess the calibration of h(t) frames for S5 V3 specifically for the analysis talks in the upcoming AAS meeting on 07/Jan/2008.
This special review process converged recently, and as a result we'd like to make two announcements, one for the new data flag and the other for the error of h(t).
Please understand that we're releasing h(t) S5 V3 specifically for the upcoming AAS talks. Even though we believe that there's no serious error in our analyses, this time we do not endorse h(t) S5 V3 data for any other purpose.
We have concluded that there were two short periods in S5, one for H1 and the other for L1, where there was a problem in h(t) production.
H1: GPS 835045153 to 835503749 (Segments 1657-1693), about 5 days in Jun 2006 L1: GPS 822656471 to 822883666, about 3 days in Jan/Feb 2006
For H1 this is the exact time of the event, and we know exactly what happened and why.
For L1 this is a very conservative estimate in that we know that the entire event was inside this window.
We decided to flag the bad segments for V3 and fix this in V4.
Since V3 h(t) is in wide use, regenerating the h(t) frames for V3 would probably cause more confusion than necessary. Instead, we decided to flag the bad segments. We will fix this in V4.
The nature of the error is a big, frequency dependent systematic that goes as large as 30% for H1 and about a factor of 2 for L1. The two documents attached should give you some idea about the systematic error of the bad frames: H1 bad and good, L1 bad.
The parties that might be affected by this regarding the AAS talks are the CW group (Crab F-stat) and the stochastic group. There should be no impact on the burst talk because of the time of the event.
The flag is yet to be released, but considering the urgency of the matter, several persons in the search groups (CW, stochastic and burst) were already notified of this.
For now we'd like to recommend to put a 5 % 1-sigma error in quadrature to the error budget for h(f) that was announced by Michael Landry.
This estimate is a conservative one based on our broadband comparison of h(f) and h(t): Summary plot. (Note that we haven't excluded the bad frames in this plot.)
In this analysis, only 10 % of the entire triple coincidence in S5 longer than 1024 sec was sampled. Both h(t) and h(f) in the sampled segments were FFT-ed with 1/6 Hz BW, and 100 frequency points were randomly chosen to be analyzed.
Even though we believe that the above statement about the calibration is correct, the analysis groups should be aware of the fact that the discrepancy in the noise level between h(t) and h(f) can become quite large in the close proximity of the lines (i.e. 60Hz harmonics, calibration lines, and violins).
Attached is the zoomed plot of the H1 near these lines: H1peaks_10.pdf
We believe that this is a noise issue and not the calibration issue. In other words, we believe that the response function of h(t) is correct within the error bar, even at these frequencies. As such, as far as this discrepancy is concerned, we don't believe that the search groups should do anything in addition to what they are already doing.
As we understand, one of the Crab analyses uses h(t). For this reason, we're preparing to provide a detailed noise comparison at around 60 Hz line for the CW group.
We gratefully appreciate the inputs from all four search groups concerning the hardware injection analyses. They helped us to gain more confidence in h(t) calibration. Nothing was terribly wrong!!