jeffrey.kissel@LIGO.ORG - posted 16:08, Thursday 27 October 2022 - last comment - 16:54, Thursday 27 October 2022(65497)
Calibration Progress: Reconciling Fudges; Comparing 4 months of Sensing Measurements -- Low-Frequency Sensing Function Non-sense Returns
J. Kissel, L. Dartez We've pick up the "what's the status of the calibration?" story after many moons of not paying any attention to it. Last we left off, we had fudged the calibration 4 "primary" gain parameters in the front-end, in order to make the DELTAL_EXTERNAL / PCAL measure of the systematic error flat -- and after correcting for the known low-frequency impact of super-Nyquist response we see "the answer" is good, as shown in the green traces of the attachments to LHO:64311. This aLOG attempts to reconcile those fudges with what sensing and actuation measurements we've had to inform the model. %%% EXECUTIVE SUMMARY %%% I understand all fudges that have been necessary to make achieve the low-systematic error that we had just before we vented in Oct 2022. - The CAL-CS, 1.08 fudge factor on the optical gain in the 1/C bank, is consistent with the MCMC fit of the 2022-05-25 measurement results, and 2022-05-27 and 2022-06-14 model. - We'd just never updated foton filter bank from some previous optical gain value from 2022-04-17 (LHO:58656), and thus needed the "superficial" correction. - In fact, I suggested this update in May 2022, LHO:63405, but I just never installed it. - RECONCILED. - The cavity pole frequency, CAL-CS, 450.8 Hz is also consistent with the MCMC fit of the 2022-05-25 measurement results (450 (+1.843,-2.47) Hz), and the 2022-05-27 and 2022-06-14 model. - No change needed here. - RECONCILED. - The actuator results are *also* consistent -- given that I was tuning the fudges to the DELTAL_EXTERNAL / PCAL measure of the systematic error *without* correcting for the low-frequency impact of the super-Nyquist effects. - RECONCILED. - Thus, we should install the values from the 20220527 values and continue forward with that model, "until our issues with the sensing function have been solved" - That includes pushing "EPIC records" -- model transfer function values at calibration line frequencies that are used to turn calibration line transfer functions into time-dependent correction factors. "until our issues with the sensing function have been solved" because the sensing function continues to not agree with our simplistic "phenomenological" model below 80 Hz. The sensing function measurement taken on 2022-06-02 was a particularly good anomaly but is not representative of how the sensing function has behaved. Thus, I was duped into expanding the MCMC fit region down to 20 Hz, and got a value that is not representative of the in-general behavior of the sensing function. As such, the aggregate of 4 months of data seem to have less residual -- and the DELTAL / PCAL transfer function agrees -- that the 20220527 model informed by the 80 Hz and above MCMC fit of the 2022-05-25 measurement. %%% DATA TABLE COMPARISON %%% Sensing Optical Gain: Model (ct/m) (m/ct) yyyymmdd / CALCS reference aLOG CAL-CS 3.473e+06 2.879e-07 -- LHO:58656 20220527 3.206e+06 3.119e-07 1.083 LHO:63405 20220614 3.138e+06 3.187e-07 1.107 LHO:63429 Fudge correction applied: 1.08 Sensing Cavity Pole: Model (Hz) Uncertainty Range CALCS within uncertainty range Y/N? CAL-CS: 450.8 20220527: 450.0 [447.53, 451.84] Y 20220614: 457.4 [456.36. 458.45] N Fudge correction applied: None UIM: Model (N/ct) (N/A) CAL-CS 7.650e-8 20220527 7.615e-8 1.6222 0.995 20220614 7.552e-8 1.609 0.987 Fudge correction needed: 1.0 (none) (we've proven over and over that even a 10% swing in the UIM gain doesn't impact the overall response function. So, even though no fudge is required, we might as well just install the MCMC fit result so we have self-consistent uncertainty to accompany it while estimating the response function uncertainty budget.) PUM: Model (N/ct) (N/A) yyyymmdd / CALCS CAL-CS 6.054e-10 20220527 6.150e-10 0.03003 1.015 20220614 6.244e-10 0.03048 1.030 Fudge correction applied: 1.05 TST: Model (N/ct) (N/V^2) yyyymmdd / CALCS CAL-CS: 4.751e-12 20220527: 4.985e-12 4.669e-11 1.050 20220614: 4.941e-12 4.627e-11 1.04 Fudge correction applied: 1.07
Here, to show off the story of the evolution of the sensing function over the past four months, I've resurrected the functionality of sensing function comparison scripts from O3, which now use the new pyDARM infrastructure. Script locations listed below. In the following two attachments, I compare the sensing function measurements from 2022-05-25 through 2022-09-01. The *difference* between the two attachments, is that I compare these measurements against two different models: (1) the sensing function model (20220527) whose parameters are informed by the 80 Hz-and-above fit to the 2022-05-25 measurement, and (2) the sensing function model (20220614) whose parameters are informed by the 20 Hz-and-above fit to the 2022-06-02 measurement. First, pull up sensingFunction_comparison_H1_proc20221027_model20220527_meas20220525-20220901.pdf. There's already a story here. One can see (a) From the left two panels, there continues to be low frequency behavior that looks nothing like the "optical spring" phenomenological model that we've used in the past. This continues to carry over from O3, even though we've made claims in the recent-past that "there's no evidence for optical springs after the new ITM has been swapped." Given our history of trying to understand this frequency response, I could blame length-to-angle-to-length cross coupling, because -- as far as I know -- we're still using large spot-position offsets in the arms. (b) The unknown low-frequency response is different from measurement to measurement. These measurements, taken several weeks to a month apart, were taken by a variety of people, and little attention was paid to *when* they were taken during the lock-acquisition sequence, and/or how much after power-up -- and thus after the IFO achieves thermal equilibrium. We'll need to do some digging/trending with this in mind to see if we can come up with a correlation. (c) From the upper right panel, The spread of the frequency-dependent, overall magnitude of the residual TF is about +/-2%. That's fine, and totally consistent with the past history of variations in the optical gain. We typically correct for this change in optical gain with the "kappa_C" time-dependent correction factor, or "TDCF." (d) From the lower right panel, The phase of the residual TF is flat and 0-ish deg in phase across the measurement band, implying that the 450.0 Hz pole frequency is the right frequency. This is also consistent with what we've seen in the past -- the cavity pole frequency is typical quite solid in frequency. Then, pull up sensingFunction_comparison_H1_proc20221027_model20220614_meas20220525-20220901.pdf. This adds more to the story. Namely, again, that the fit to the particularly, anomalously flat 2022-06-02 sensing function gives us the wrong answer for the 4-month collection of measurements in aggregate (e) The upper right panel shows that the 2022-06-02 measurement had a particularly low optical gain that day. So dividing the other months' data by that optical gain yields a high overall scale in the magnitude residual. (f) The lower right panel shows that -- perhaps contrary to (d) -- if we consistently divide out a cavity pole of 457 Hz rather than 450 Hz, we'll get more phase systematic error, i.e. the phase residual TF has more wiggle to it. So. What do we do about this? It took a concerted effort over many months to try and make a model of length-to-angle-to-length cross coupling during O3 -- and we were never able to reconcile model against measurement. I think it'll take *another* concerted effort this time in effort to remove it. Once we get the IFO back -- and more importantly *after* we get the IFO into a stable controls / thermal / angular / PSL power configuration, then we systematically turn knobs to see what affects what. The first options on the table should be the same knobs we've turned before: (i) SRCL Offset (ii) Arm-cavity spot positions (iii) Length-to-angle decoupling (iv) Arm-cavity loop gains to see if we can reproduce the effects we'd seen before. If so, that hopefully means we can pick up the modeling effort where we left off, rather than having to start from scratch. Scripts to produce these plots live here: /ligo/gitcommon/Calibration/ifo/scripts/fullifosensingtfs/ compare_fullifosensingtfs_model_20220527_meas20220527-20220901.py compare_fullifosensingtfs_model_20220614_meas20220527-20220901.py