J. Kissel

I picked up another set of full-band sensing function measurements to add to the collection that supports the newest configuration of the DARM loop and the corresponding model parameter set (see 51784 and 51782). Of course, even though we believe we've mitigated the detuned part of the low-frequency sensing function response with the addition of the requested digital SRCL offset, there's no reason to believe it will become time-independent, and thus we continue to regularly measure with a ~weekly cadence to ensure we see what's happening comparing uncorrelated nominal-low-noise segments.

While these full-frequency-band sweeps provide a complete-picture snapshot, they don't provide the entire story. We are still tracking the time-dependence of the low-frequency end of the sensing function continuously with calibration lines, and indeed: the summary page report for the time-dependent correction factors (TDCFs) shows that beginnings of nominal low noise segments show that the response -- computed by the ~17 Hz calibration line and cast as a detuned spring frequency squared, f_s^2 (maybe errantly, maybe not) -- does indeed evolve with time. 
The values are typically, and repeatably, dropping from +35 to -5 Hz^2, indicating (if it is the detuning that's changing) that we're transitioning from an anti-spring to a pro-spring over the first ~1 hour of arriving at the nominal-low-noise (see  2019-09-13 2019-09-15 2019-09-16 examples from the bottom right most plot of the TDCF summary page).
Perhaps one bit of comfort is that -- while we're in these thermalization periods -- an anti-spring (like we had during O1 & O2), does not have a high-Q-like response, and doesn't impact the phase nearly as much as a pro-spring (like we had in mid-O3 during and after the spot-move mishap).

All this being said, I think we're fine with our current uncertainty budget tactics -- MCMC fitting the data only above 20 Hz, using a model parameter set informed by the MAP of that fit (and forcing the detuned spring to be non-existent; consistent with the MCMC fit), and treating this low frequency response as an unknown systematic error with associated uncertainty.

Supporting plots attached are (in attachment order): 
    (1) the comparison between all data sets that we claim to be represented by the 2019-09-09 model parameter set (uncorrected for time-dependence),
    (2) The MCMC corner plot of last 4 data sets that have not been processed until now
    (3) The Comparison between those same measurements' MCMC fit and the respective data for that day.

The raw data for these plots live in
/ligo/svncommon/CalSVN/aligocalibration/trunk/Runs/O3/H1/Measurements/FullIFOSensingTFs/
    2019-09-09_H1_PreModelChange_DARM_OLGTF_LF_SS_5to1100Hz_15min.xml
    (Forgot to save PreModelChange PCAL TF in chaos of the model change, but I did export and commit the data, e.g. 2019-09-09_H1_PreModelChange_PCAL2DARMTF_LF_SS_A_PCALYRX_B_DARMIN1_tf.txt)

    2019-09-09_H1_PostModelChange_PCALY2DARMTF_LF_SS_5t1100Hz_10min.xml
    2019-09-09_H1_PostModelChange_DARM_OLGTF_LF_SS_5to1100Hz_15min.xml

    2019-09-16_H1_DARM_OLGTF_LF_SS_5to1100Hz_15min.xml      # Today's measurements
    2019-09-16_H1_PCALY2DARMTF_LF_SS_5t1100Hz_10min.xml 
    2019-09-16_H1_PCALX2DARMTF_LF_SS_5t1100Hz_10min.xml     # not yet processed