After things started to settle down from the California eaethquake, we were hit by another mag 7.8 earthquake in the Solomon Islands. Put the SEI_CONF to LARGE_EQ_NOBRSXY. All ISI and several SUS have tripped many times. Microseism is over 300um/s.

IFO in down while we ride through the rumbling earth.  

Ops Shift Transition: 12/08/2016, Day Shift 16:00 – 00:00 (08:00 - 16:00) -  UTC (PT)

TITLE: 12/08 Owl Shift: 08:00-16:00 UTC (00:00-08:00 PST), all times posted in UTC
STATE of H1: Earthquake
INCOMING OPERATOR: Jeff
SHIFT SUMMARY: I had a few locklosses due to PIs, but I eventually wrangled them and got it locked for almost 2 hours before a 6.5M earthquake off of California knocked us out.
LOG:

• 08:15 Out of Observing to run a2l
• 09:54 Lockloss
• 10:45 Observing
• 11:07 Lockloss
• 11:45 Observing
• 12:10 Lockloss
• 12:33 As I approached DC_READOUT, there were lines in DARM at ~28hz and ~37hz, DHARD P seemed to start to run away. Bounce/roll/violin/PI seemed to all be ok. Not sure what happened there.
• 13:07 Observing
• 14:32 Out of observing to run a2l, the dtt template looks like it could us it.
• 14:37 Observing
• 14:52 Lockloss

A very nearby earthquake just tripped basically everything. It hasn't shown up on USGS yet, but the BLRMS are off the displays.

It seems like there are some PI modes that we are no longer damping that are slightly run up according to the dtt. I also wonder if the peaks at around 13000hz are modes that we had not had issues with before, or maybe something else entirely? I don't see any mention of them on the PI medm and they are currently the highest. I will try to damp some of the other ones if they get any higher.

The attached shows the PI dtt just after reaching low noise.

PI mode 27. I damped mode 28 then, thinking that I defeated it, went to get a cup of water. When I returned, mode 27 was quickly rising and I got it to level out but couldn't damp it before it broke.

It was very sudden. There was no indication form any of the FOMs that anything was, and the environment looks good. There may be the smallest bump on the BLRMS, but nothing that ever should have dropped it. SR3 lockloss template attached, doesn't seem to be the cause (I zoomed in this time unlike before). Also attached the generic lockloss plots, which I can't tell if any of it is out of the norm. Myabe MICH was a bit more active a few seconds before it dropped, but it does sometimes seem to do that normally.

IMC is having trouble staying locked, I should stop writing here and get back to that.

TITLE: 12/08 Owl Shift: 08:00-16:00 UTC (00:00-08:00 PST), all times posted in UTC
STATE of H1: Observing at 75.5915Mpc
OUTGOING OPERATOR: Travis
CURRENT ENVIRONMENT:
    Wind: 11mph Gusts, 7mph 5min avg
    Primary useism: 0.04 μm/s
    Secondary useism: 0.24 μm/s
QUICK SUMMARY: Travis delivers a locked IFO again after recovering from an earthquake in China. LLO is currently down. Snow expected in the morning sp be safe driving in.

TITLE: 12/08 Eve Shift: 00:00-08:00 UTC (16:00-00:00 PST), all times posted in UTC
STATE of H1: Observing at 78.082Mpc
INCOMING OPERATOR: TJ
SHIFT SUMMARY:  Locked for most of the shift with ~1.5 hour downtime due to EQ.
LOG:

See previous logs for pertinent events.

 

J. Kissel for S. Karki
WP #6368

I've moved the PCALX roaming line, its long duration sweep, from 4001.3 Hz to 4301.3 Hz. See schedule status below.

Since it looks like we'll have the duty cycle to just finish the schedule within a week, I'm just going to continue to re-gathering the highest frequency data. As such, Sudarshan has something consistent and straightforward with which to work instead of data with a bunch details or qualifications. 


Frequency    Planned Amplitude        Planned Duration      Actual Amplitude    Start Time                 Stop Time                    Achieved Duration
(Hz)         (ct)                     (hh:mm)                   (ct)               (UTC)                    (UTC)                         (hh:mm)
---------------------------------------------------------------------------------------------------------------------------------------------------------
1001.3       35k                      02:00                   39322.0           Nov 28 2016 17:20:44 UTC  Nov 30 2016 17:16:00 UTC         days    @ 30 W  
1501.3       35k                      02:00                   39322.0           Nov 30 2016 17:27:00 UTC  Nov 30 2016 19:36:00 UTC         02:09   @ 30 W         
2001.3       35k                      02:00                   39322.0           Nov 30 2016 19:36:00 UTC  Nov 30 2016 22:07:00 UTC         02:31   @ 30 W
2501.3       35k                      05:00                   39322.0           Nov 30 2016 22:08:00 UTC  Dec 02 2016 20:16:00 UTC         days    @ 30 W
3001.3       35k                      05:00                   39322.0           Dec 02 2016 20:17:00 UTC  Dec 05 2016 16:58:57 UTC         days    @ 30 W
3501.3       35k                      05:00                   39322.0           Dec 05 2016 16:58:57 UTC  Dec 06 2016 21:09:56 UTC         ~15:00  @ 30 W
4001.3       40k                      10:00                   39322.0           Dec 06 2016 21:09:56 UTC  Dec 07 2016 18:50:09 UTC         ~20:00  @ 30 W
4301.3       40k                      10:00                   39322.0           Dec 07 2016 18:50:09 UTC
4501.3       40k                      10:00                   39322.0           
4801.3       40k                      10:00                   39222.0           
5001.3       40k                      10:00                   39222.0         

Frequency    Planned Amplitude        Planned Duration      Actual Amplitude    Start Time                 Stop Time                    Achieved Duration
(Hz)         (ct)                     (hh:mm)                   (ct)               (UTC)                    (UTC)                         (hh:mm)
---------------------------------------------------------------------------------------------------------------------------------------------------------
1001.3       35k                      02:00                   39322.0           Nov 11 2016 21:37:50 UTC    Nov 12 2016 03:28:21 UTC      ~several hours @ 25 W
1501.3       35k                      02:00                   39322.0           Oct 24 2016 15:26:57 UTC    Oct 31 2016 15:44:29 UTC      ~week @ 25 W
2001.3       35k                      02:00                   39322.0           Oct 17 2016 21:22:03 UTC    Oct 24 2016 15:26:57 UTC      several days (at both 50W and 25 W)
2501.3       35k                      05:00                   39322.0           Oct 12 2016 03:20:41 UTC    Oct 17 2016 21:22:03 UTC      days     @ 50 W
3001.3       35k                      05:00                   39322.0           Oct 06 2016 18:39:26 UTC    Oct 12 2016 03:20:41 UTC      days     @ 50 W
3501.3       35k                      05:00                   39322.0           Jul 06 2016 18:56:13 UTC    Oct 06 2016 18:39:26 UTC      months   @ 50 W
4001.3       40k                      10:00                   39322.0           Nov 12 2016 03:28:21 UTC    Nov 16 2016 22:17:29 UTC      days     @ 30 W (see LHO aLOG 31546 for caveats)
4301.3       40k                      10:00                   39322.0           Nov 16 2016 22:17:29 UTC    Nov 18 2016 17:08:49 UTC      days     @ 30 W          
4501.3       40k                      10:00                   39322.0           Nov 18 2016 17:08:49 UTC    Nov 20 2016 16:54:32 UTC      days     @ 30 W (see LHO aLOG 31610 for caveats)   
4801.3       40k                      10:00                   39222.0           Nov 20 2016 16:54:32 UTC    Nov 22 2016 23:56:06 UTC      days     @ 30 W
5001.3       40k                      10:00                   39222.0           Nov 22 2016 23:56:06 UTC    Nov 28 2016 17:20:44 UTC      days     @ 30 W (line was OFF and ON for Hardware INJ)  

As a benchmark against which to compare upcoming O2 data, I have compiled a list
of narrow lines seen in the H1 DARM spectrum up to 2000 Hz, using 107 hours of
ER10 FScan 30-minute SFTs. There are no big surprises relative to the lines and combs
Ansel and I have reported on previously from ER9 and later data, but below are some
observations. 

Attached figures show  selected band spectra, and a zipped attachment
contains a much larger set of bands. Also attached for reference is a plaintext list of combs, 
isolated lines, PEM-associated lines. etc. In the attached spectra, the red curve is the
ER10 data, and the black curve is the full-O1 data. The label annotations are keyed to
the height of the red curve, but in some cases, those labels refer to lines in the O1 data
that are not (yet) visible in accumulated current data. For the most part, lines seen in 
O1 that don't show up in ER10 nonetheless remain for now in the lines list and still
have labels on the graphs that end up in the red fuzz. If they fail to emerge in O2
data, they will be deleted from future line lists.

Observations:

As noted previously, the 8-Hz / 16-Hz combs seen in O1 are blessedly gone
Thanks to interventions to upgrade timing electronics firmware, the previously dominant low-frequency comb
with 1-Hz spacing and 0.5-Hz offset is greatly reduced (but not eliminated)
Compared to O1, the non-linear upconversion around quad violin modes and harmonics
is much worse in ER10, and it appears that at least some violin mode excitations are
affecting the low frequency band in that one sees there a multitude of sparsely distributed
doublets, triples and quadruplets of lines with a  spacing of about 0.0468 Hz.
Andy Lundgren pointed out that this spacing coincides closely with a beat between
two 2nd-harmonic violin modes he reported on last week. 
If I measure / eyeball those frequencies as seen in the ER10 data set, I get 1009.4414 and 1009.4881 Hz, giving
a beat of 0.0467 Hz (but with an uncertainty of a few tenths of a mHz), consistent with the explanation.
Further credence comes from the proliferation of such doublets / triplets / quadruplets seen in the wings of
the quad harmonics (fundamental and higher). 
There are so many upconverted lines around the quad harmonics that I didn't bother cataloguing them
in those regions on the assumption that they will go away in O2 with improved damping 
(but cataloguing can be revisited later, if needed). 
 With the exception of the violin mode regions, the higher frequency spectrum is generally free
of narrow lines (unlike the region below 100 Hz), but one disturbing feature is the presence of 
non-Gaussian broad disturbances that were not visible in O1. The infamous region around 1083 Hz 
is one example, as is the 100-200 Hz band.


Figures: [ER10 in red and O1 in black; labels are attached to peaks in the ER10 data]

Fig 1 - 0-1000 Hz
Fig 2 - 1000-2000 Hz
Fig 3 - 2-20 Hz
Fig 4 - 20-50 Hz
Fig 5 - 50-100 Hz
Fig 6 - 100-200 Hz
Fig 7 - 490-535 Hz 
Fig 8 - 1000-1000 Hz
Fig 9 - 1800-1900 Hz 

Other attachments:
* zipped file with full set of sub-band spectra
* Lines list with label definitions 

Keita, Rana, Evan

After refocusing the HAM5 camera, we see in full lock that there are at least two ghost beams hitting the SRM composite mass. These ghost beams move when CPY (but not CPX) is moved.

The attached plots show the DARM spectrum for different CPY alignments. There is no obvious sweet spot, but perhaps we will find one by looking at some long-term DARM BLRMS.

This image shows the HAM5 camera with the nominal CPY alignment (-150 µrad in pitch, 0 µrad in yaw). The two bright, vertically aligned spots on the left-hand side are the CPY ghost beams.

C. Cahillane

The ER10 LHO Actuation Uncertainty Budget

What has been done:
(1) Gathering all the Newtons/count actuation function measurements for the three stages (UIM, PUM, TST) from November 8th, 9th, 11th, 12th, and 18th.  
(2) Running an MCMC over the physical parameters gain and delay for all three stages.  (Only used freq < 10 Hz for UIM, freq < 100 Hz for PUM)
(3) Plot MCMC cornerplots and resulting residuals uncertainty budget from the MCMC PDFs.  
(4) Take the residuals from the MCMC physical parameters' fit, and plug them into a Gaussian Process with kernel k(x,y) = a + b * exp(-0.5*(x-y)**2/l**2), where a, b, and l are the hyperparameters of the GP.
(5) Train the GP on the residuals, and produce a mean posterior fit and an uncertainty budget from the trained kernel.

TL:DR : Take measurements -> Physical MCMC -> Unmodeled GP -> Frequency-Dependent Systematic Error +- Uncertainty Budget

The mean posterior fit is the frequency-dependent systematic error.  The uncertainty budget is also frequency-dependent.  
The Gaussian Process results are shown in Plots 1, 2, and 3 for the UIM, PUM, and TST stages.  The 1-sigma, 68% confidence intervals are plotted.

I want to call attention to how small the uncertainty is in the bucket.
The frequency-dependent uncertainty in the bucket is on the order of 0.1%

This is exciting, but is it real?  
Points in favor:
(1) We have five sets of measurements, and uncertainty roughly goes as 1/sqrt(N) where N = measurement number.  This allows us to win a lot where the measurements are nearly the same and the errorbars are small.
(2) The GP fit hits most of the measurement errorbars.
(3) Uncertainty expands where we have less information.
Points against:
(1) Overfitting might be a problem.  There seem to be unphysical wiggles that nail our data in the bucket. 
(2) This uncertainty budget is lower than ever before by a factor of 10.  This requires extraordinary proof.   
Rebuttals:
(1) We might be able to combat overfitting by lengthening the kernel length-scale and adding more noise to our measurements.  Also, the whole point of this method is to capture unmodeled wiggles.  
(2) The data analysis methods used are more advanced and designed to handle frequency-dependence.  

Plots 4, 5, and 6 are the UIM, PUM, and TST physical model MCMC fits with the measurements.  
Plots 7, 8, and 9 are the UIM, PUM, and TST physical model MCMC parameter cornerplots.

A couple of points: 
(1) The GP uncertainty doesn't take into account the uncertainty from the MCMC physical parameters.  The GP uncertainty dominates nearly everywhere, since the MCMC uncertainty is so tiny (See plots 4, 5, 6).  I may try to incorporate the MCMC uncertainty directly into the measurement errorbars that I input into the GP.
(2) A lot remains to be done.  LLO actuation, sensing at both sites, and final response uncertainty.  
(3) If these results hold, we may have uncertainty dominated by time-dependence.  Stay tuned.