[J. Kissel, A. Viets]

Starting at 19 UTC on April 25, 2019, I produced 48 hours of calibrated data with "fast kappas," that is, the time-dependent correction factors (TDCFs) were computed and averaged using a smaller amount of input data.  10 s of data were used to demodulate the calibration lines, and the computed TDCFs were smoothed with a 20 s running median and a 5 s running mean.  (Normally, 20 s are used for demodulation, the median is 128 s and the mean is 10 s.)

I was trying to find evidence of the "2-minute period oscillations" noted in several aLOGs in late April and early May, e.g. https://alog.ligo-wa.caltech.edu/aLOG/index.php?callRep=48948, https://alog.ligo-wa.caltech.edu/aLOG/index.php?callRep=48902, https://alog.ligo-wa.caltech.edu/aLOG/index.php?callRep=48857, and https://alog.ligo-wa.caltech.edu/aLOG/index.php?callRep=48846.  It appears that the oscillation as measured using kappa_C is small compared to the noise when using a shorter median + mean for the TDCFs.  So, my suspicion is that these oscillations would be difficult to correct in the calibration.  If we use a short average, noise in kappa_C increases to the 1% level, and our usual few minutes of averaging is relatively insensitive to 2-minute oscillations.

Several plots of the results are attached:
1. All 48 hours of "fast" kappa_C.  I *think* that the SR3 heater change occurred near the end of the day on 4-26, near the beginning of the last long lock stretch shown in this plot.  The noise in kappa_C does not increase noticeably, as one might expect if the 2-minute oscillations had a significant impact on kappa_C and increased at that time.  However, a small increase in the average value of kappa_C is seen at that time, which is also visible on the summary page for April 27 near the beginning of the day: https://ldas-jobs.ligo-wa.caltech.edu/~detchar/summary/day/20190427/cal/time_varying_factors/.  I also computed the correlation coefficient between fast kappa_C and one of the ASC channels (ASC-CHARD_Y_OUT_DQ) before and after the SR3 heater change:
Using 15 hours starting at 00:00 UTC on April 26 (first long lock stretch in the plot):  -0.0045891
Using 15 hours starting at 03:00 UTC on April 27 (last long lock stretch in the plot):  0.01318592
The correlation did increase, but remains small.

2. The first 15 hours noted above, with ASC-CHARD_Y_OUT_DQ also shown

3. The last 15 hours noted above, with ASC-CHARD_Y_OUT_DQ also shown

4. A zoom-in of 50 minutes of data during the last lock stretch shown in the first plot.

5. Even more zoomed in: ~17 minutes.  It is difficult to see any correlation by eye, or to see any periodic oscillations in kappa_C that are distinguishable from noise.

It is possible that I have not used the best time for this investigation, or that I should analyze a different front-end channel instead of ASC-CHARD_Y_OUT_DQ.  If this is the case, I hope someone else can point me in the right direction.  Otherwise, it looks like it is not worth trying to compensate for fast fluctuations in the calibration, but rather to stick with what we've done in the past.