L. Sun
The hourly C01 uncertainty budget plots and txt files are stored here https://ldas-jobs.ligo.caltech.edu/~ling.sun/Calibration/Uncertainty/O3C01/LHO/
LHO: from 1237827584 == Mar 28 2019 16:59:26 UTC to 1244307456 == Jun 11 2019 16:57:18 UTC
Note: If the detector is not locked or the calibration line uncertainty > 0.005, the results are not produced for that given time. (Results cover 71% of the whole duration.)
The attachment demonstrates that the variation of the envelopes is negligible.
- White dashed curve: median of the median values from all the hourly data
- White solid curve: median of the +/- 1 sigma values from all the hourly data
- Color: percentiles of the +/- 1 sigma curves (i.e., variation of the envelope)
L. Sun,
I've fixed the RRNom.py code to exclude kappa_U and kappa_P correction before 0416 in LHO. On the other hand, I found that some jobs unexpectedly failed due to cluster issues in the previous run. Some of the hourly data were missing. I've regenerated C01 statistics for the whole period. Now the coverage is 77% for LHO.
The updated percentile plot and the max mag/phase bound curves are attached. In the percentile plot, we see larger variation at low freq because kappa_U and kappa_P were not corrected before 0416.
The plots are generated using the script ^/trunk/Common/pyDARM/RRNomStat.py
Sample command: python3 RRNomStat.py --statDir=/home/ling.sun/public_html/Calibration/Uncertainty/O3C01/LHO/ --IFO=LHO --nameTag=O3_C01
L. Sun
The max bound plot in the previous comment shows the max of all statistics. Now it's updated to show the max bounds of different percentiles.
The variation in the uncertainty percentiles seen in the 95th ("2 sigma") and 99th ("3 sigma") from the 68th ("1 sigma") are a result of H1's h(t) not correcting for \kappa_P and \kappa_U during the observational stretched before April 16th, when we didn't trust them enough (see details in FRS Ticket 12997 and associated dependencies).
Why didn't we trust them? In short:
Because ETMX calibration lines used to determine \kappa_U and \kappa_P were too far apart from the PCAL absolute reference line -- and the time-dependent sensing function was varying enough (see 51115 and FRS Ticket 13012) that the approximation that "the calibration lines are close enough that the time variance of the ratio of C/(1+G) at PCAL frequency and each ETMX line is negligible" [see section 2.4 in T1700106] breaks down).
Thus, the (yes, still untrustworthy) measured values of \kappa_P and \kappa_U are applied as a *systematic error* in the uncertainty and error budget, and thus that budget is time-variant, and increasing the 95th and 99th percentile values. This creates an *over-estimate* of the systematic error, as PUM and UIM actuation strengths as measured with frequency dependent sweeps have remained entirely static over the run thus far -- see LHO aLOG 50992 and G1901479. However, we accept this poor assessment as of the time-dependence actuator systematic error as a "conservative" estimate of what it could be (where the GPR covers the unknown static frequency dependence).
After April 16th, we moved all actuator calibration line frequencies lower and much closer to each other (see LHO aLOG 48551).
Attached is supplemental material from Lilli showing what the percentiles look like if O3 uncertainty and systematic error estimates prior to April 16th are *excluded.*
One can see much less variance, and the 95th and 99th percentile curves lie right on top of the 68th percentile curve, as expected.
[Editor's note -- these were sent to calibration mailing list on Aug 27 2019, under thread "To Use or Not to Use: TDCFs for H1 until Apr 16th" -- 2019-08/msg00166.html. Unclear if they were yet put in the repo, but I've committed them to /ligo/svncommon/CalSVN/aligocalibration/trunk/Runs/O3/H1/Results/Uncertainty/, perhaps redundantly, and will remove and delete once I find out if /where they were committed elsewhere .