Please ignore the alog above, I will reproduce it here with corrected numbers with the responsivity used correctly. 

Summary: With a cold OM2 we estimate 11% unknown HAM6 losess (including OMC losses, modemismatch, DCPD QE), with hot OM2 that goes to 21%.  We can estimate the change in HAM6 throughput 4 ways, three methods that depend on the DARM offset steps agree that the HAM6 throughput with hot OM2 is 90% of what it is with cold OM2.  The optical gain suggests that the throughput only dropped to 96% of what it was for cold OM2.  

Details:

Last Tuesday we stepped up the ring heater and did a couple of DARM offset steps before and after the change.  The first attachment shows the trends of the two sets of DARM offset steps and the two hour transitent of OM2.  The next plots zoom in on the DARM offset steps, each step was for 1 minute and the steps were larger than in  69653 to help make a small change more easily measureable.  With OM2 cold, the refl diode still doesn't seem like it makes a nice measurement of the darm offset light reflected by the OMC, with OM2 hot it is clear that we can see the change in OMC refl when the DARM offset steps. 

change in HAM6 throughput estimated by AS_C

For both hot and cold OM2, we can use the power change on AS_C to estimate the HAM6 throughput.  Calibrating DCPD sum into Watts using 0.858e-3 A/W gives 86% HAM6 throughput for cold OM2 with 60W input power, and 77% throughput for the hot OM2. This can be compared to 69653, done with 75W input power (430kW in arms) where Jennie W found 82% throughput of HAM6 (Jennie also used 0.858AW for the DCPD responsivity).  We can add up known HAM6 losses from the SQZ loss wiki, 0.993(OM1)*0.985(OM3)*0.9926(OMC QPD) = 97% HAM6 transmission if OMC mode matching, losses and QE were perfect.  This means that we have 11% additional HAM6 losses for a cold OM2, and 21% for the hot OM2.  (The HAM6 throughput with OM2 hot is 90% of what it is with OM2 cold)

change power on AS_C from darm offset and  sidebands/contrast defect

The estimate of the amount of sideband/carrier junk light entering HAM6 based on AS_C is 623mW/624mW with OM2 cold and with OM2 hot it is 634mW.  This 10-11mW increase might be due to the alignment shift in the SRC that Elenna plotted 70886 which could be due to the change in gouy phase on the AS WFS when OM2 changes.  The total power incident on HAM6 increased by 15mW during the OM2 thermal transient, the light from the DARM offset increased by 6-6.6mW.  When the HAM6 throughput changes from cold (eta) to hot (eta') the darm offset has to change from cold X to hot X' = X* sqrt(eta/eta') to keep the power on the DCPDs the same, which increases the DARM offset light on AS_C by eta/eta' (so, as_darm_power_hot/as_darm_power_cold = HAM6 througput cold/ HAM6 throughput cold).  This implies that the HAM6 throughput with OM2 hot is 89% of the HAM6 throughput with a cold OM2.

OMC mode matching from OMC refl diode

The light which is insensitive to DARM in OMC reflection increased from 623.3/624mW to 634.4mW . We can estimate the OMC mode matching with an approach like llo 60885: there is 53mW of darm offset light incident on the OMC for cold OM2 and 59mW for hot OM2 (taking into account known HAM6 losses). We can estimate the mode matching of the OMC (which could include misalignment) as 1-reflected darm offset light/ incident darm offset light which gives 96% mode matching for the cold OM2 (where the measurement may be suspect) and 85.4% or 85.2% mode matching for the hot OM2. This implies that the OMC mode matching with OM2 hot is 89% of what it is with OM2 cold.

Optical gain change

The normalized optical gain (kappa c) dropped from 1 to 0.98 while OM2 was heating up.  Since the power on the DCPDs is kept constant, the DARM offset changes when the HAM6 throughput changes, and the optical gain hot/optical gain cold = sqrt(throughput hot/throughput cold).  Kappa c of 0.98 implies that HAM6 throughput with OM2 hot is 96% of the throughput with OM2 cold.

# DARM offset OM2 scan log file, gps start, gps stop

• 4 1371902418 1371902478
• 10.94319 1371902478 1371902539
• 4 1371902539 1371902599
• 10.94319 1371902599 1371902659
• 4 1371909859 1371909919
• 10.94319 1371909920 1371909980
• 4 1371909980 1371910040
• 10.94319 1371910040 1371910100

Once I remembered that SUS_CHARGE is not one of the guardian nodes monitored for the observation intent bit (although its subordinate nodes are), I reloaded it this morning at 14:02 UTC so Camilla's changes are pushed in.

This fixed the SUS_CHARGE code and we successfully stayed locked throughout both ESD transitions on Tuesday.

Total time taken is 18 minutes, this is longer than the 15 minute target. I've reduced the slowy_ramp_on_bias() ramptime back from 60 to 20 seconds, as I've reverted some of 68987 changes I was blaming for the looklosses. ESD_EXC_{QUAD} will need to be reloaded for each quad for this to take effect.

Currently the SUS_CHARGE code is taking 16m45s. We could look at further decreasing some tramps. 

TJ noticed this morning that the L2L tramps for ITMX and ETMX weren't reverted by the code. I've added lines to SUS_CHARGE to save and revert these after the ESD transitions and also save adn reset the ITMX_L2L gain so it's not hard-coded. 

An exploration of where the 2.6 Hz peak is visible and coherent with DARM

The ETMX M0 / R0 / TMSX are interesting signals, maybe something isn't working properly in the R0 tracking loop?

The usual ASC suspects, especially CHARD_Y.

Several DC centering loops.

Some OMC signals.

No smoking gun

I have some evidence that could indicate that this peak is related to some instability in the HARD loops.

First, we noticed this 2.6 Hz peak is very prevalent in the OMC and OM3 suspensions. Evan was tracking the presence of the peak in OM3 yaw and noted it appeared sometime around April 9-14. This was a tricky time because many things happened: we powered up, we adjusted the compensation plates and OMC to reduce scatter, etc. Evan also noticed it doubled in size on June 22, which corresponds to when we went down in power. I used the OMC SUS master out channels as an indicator of when this peak appeared. I noticed it was not present April 7th and then appeared on April 8th, UTC time. More specifically, it appeared in the Lownoise ASC guardian state. This was around the time I was working on removing RPC completely from the ASC control and working with Dan and others on increasing the power. It appears that the peak pops up right when I turned off the RPC and transitioned all the HARD loops to the new high power controllers.

The fact that the peak height doubled when we went down in power also bolsters this theory: if there was something a bit marginal in the loop at higher power, the shift in the radiation pressure plant could have worsened the marginality. To further confirm, I looked at spectra of the HARD and SOFT loops, and the 2.6 Hz peak becomes prominent around the time I changed the loop control and turned off RPC.

As a test, Gabriele and I would like to make small changes in the HARD loop gains and see if we can pinpoint which loop is marginal. If we're lucky, the peak could be fixed with a gain change. If we're less lucky, maybe we need to redo one or more loop controllers.

First attachment is an ndscope screenshot. I plotted the guardian state, RPC gains, and the SWSTATs for all the HARD loops. The cursor is on the time when I hard turned off all the RPC gains and switched the loop controllers. GPStime: 1364955781. This also corresponds to when the peak appears in the OMC suspension channels.

Second attachment is a dtt screenshot of a plot comparing the OMC sus master out spectra now (blue refs) with the spectra on April 7 before the ASC change (red live).

Edit to add: Gabriele and I tested raising and lowering various ASC loop gains but we saw no difference in the peak.

We also hypothesized that this is related to the reaction chain tracking and/or damping. We turned off the reaction chain tracking of ETMX for L, P and R and saw no change in the peak.

The two simulines injections were at the following GPS times:

Trial 1 Start:
   2023-07-05 19:02:04.672 UTC 
   GPS 1372618942.672165

Trial 2 Start:
   2023-07-05 19:25:27.781 UTC
   GPS 1372620345.781883

Each simulines suite ran for approximately 23 minutes.

I used an adpated version of the OMCscan class to fit the spectrum up to 20/02 carrier modes. The scan went through two free spectral ranges so I just used the first 60s to make the analysis easier, and assuming that within this 60s of data the third smallest clear peak was 20, and the fourth one was 10 mode.

The fitted spectra is attached.

Then I used an adapted version of fit_two_peaks.py to fit a sum of two lorentzians to the 20 and 02 carrier modes, the fit is shown in the second graph.

We expect the HOM spacing to be 0.588 MHz as per this entry and DCC T1500060 Table 25.

The spacing for the modes measured is  0.549 MHz.

From the heights of the two peaks this suggests mode-mismatch of the OMC to be C02+C20/C00 = (0.457mA+0.629mA)/(16.39mA+0.457mA+0.629mA) = 6.2% mode mis-match.

From the locked/unlocked powers on the OMC REFL PD the visibility on resonance is 1-(1.84mW/22.6mW) = 92% visibility.

If the total loss is 8%, this implies that the other non-mode-matching losses are roughly 1.4%.

To run the OMC scan code go to 

/ligo/gitcommon/labutils/omc_scan/ and run 

python OMCscan_nosidebands2.py 1371921531 60 "Sidebands off, 10W input, hot OM2" "single bounce" --verbose --make_plot -o 2

in the labutils conda environment and on git branch dev.

To do the double peak fitting run:

python fit_two_peaks_no_sidebands2.py  
in the labutils conda environment and on git branch dev.

This was a single bounce scan off ITMX, with 0.44W on the ring heater upper and lower segments, and no CO2.

Using Jennie's mode mismatch of 6.2%, we can use the ratio of locked vs unlocked reflected power to estimate the OMC losses, finesse and transmission for a perfectly mode matched beam. 

I've used a time when the fast shutter was blocked from 70409 to subtract the dark offset the refl diodes, this gives reflected power on resonance of 22.61mW, reflected power off resonance of 1.85mW.

The power in the mode mismatch is reflected_power_off_resonance * 6.2% = P_mm 1.4mW

Visibility for the 00 mode is (refl_on_resonance - P_mm)/(refl_off_resonance - P_mm) = 2.1%

The attached script uses this visibility to find the loss using:

def Refl_fraction(r_loss):
    on_res = (r1 - (t1**2 * r1 * r_loss)/(1-r1**2*r_loss))**2
    off_res = (r1 + (t1**2 * r1 * r_loss)/(1+r1**2*r_loss) )**2
    return on_res/off_res

with r1 = sqrt(reflectivity of the input output mirrors) = sqrt(1-7690e-6) from T1500060 page 143, and r_loss = sqrt(1-round trip losses).  With a visibilty of 2.1% this gives us a round trip loss of 2616ppm.  If true this level of loss would imply a finesse of 351, well lower than previous measurements: 69707.  This would imply that the transmission of the OMC cavity for a 00 mode is 73%.  

Koji pointed out that infering losses from the visibility as I did above is very sensitive to the HOM content, and including the first order modes above would have resulted in a different value of OMC losses. 

As an alternative approach, I adapted a mathematica notebook that Koji shared to use the transmitted power along with the visibility, and infer higher order mode content and cavity transmission by making an assumption about the DCPD QE. 

One confusing point about using these reflected power measurements is that we have to correctly take into account that the beam which arrives at the OMC REFL path has reflected twice off the QPD pick off beamsplitter.  (So, the incident power on the REFL diode = incident power on OMC breadboard/ R_pick_off^2)  

The results we get for cavity losses (and higher order mode content) depend on what we assume for DCPD QE with this method.  Below are the results of the attached script run with a QE of 1 and a QE of 96%, this only makes a small change in the higher order mode content we infer, and that small change in HOM also causes a small change in what we infer for the total efficiency of the OMC breadboard in the two cases. 

Power on refl diode when cavity is off resonance: 22.612 mW
Incident power on OMC breadboard (before QPD pickoff): 23.052 mW
Power on refl diode on resonance: 1.848 mW
Measured effiency (DCPD current/responsivity if QE=1)/ incident power on OMC breadboard: 81.6 %

assumed QE: 96.0 %
power in transmission (for this QE) 19.598 mW
HOM content infered: 8.069 %
Cavity transmission infered: 93.376 %
predicted efficiency () (R_inputBS * mode_matching * cavity_transmission * QE): 81.616 %
omc efficency for 00 mode (including pick off BS, cavity transmission, and QE): 88.780 %
round trip loss: 540 (ppm)
Finesse: 396.346

assumed QE: 100 %
power in transmission (for this QE) 18.814 mW
HOM content infered: 7.903 %
Cavity transmission infered: 89.479 %
predicted efficiency () (R_inputBS * mode_matching * cavity_transmission * QE): 81.616 %
omc efficency for 00 mode (including pick off BS, cavity transmission, and QE): 88.620 %
round trip loss: 886 (ppm)
Finesse: 388.021

Here are the plots of ASC-AS_C_NSUM, OMC-QPD_A_NSUM, OMC-QPD_B_NSUM and OMC-REFL_A_LF, during these measurements. ASC-AS_C_NSUM shows between 22.8 and 32.1mW, OMC-QPD_A_NSUM 23.4mW, OMC-QPD_B_NSUM 23.0mW, and OMC-REFL_A_LF 24.8mW. According to Keita OMC-REFL_A_DC has an incorrect calibration and shows 25.2mW. The average of the 2 QPDs would be 23.2mW, which is about 6.5% lower than 24.8mW.

Second screen shots shows a time when the IMC was unlocked. The DC offsets are in the 10s of uW at most.

Using data from the scan I adapted labutils/OMCscan class to plot the fitted scan and adapted labutils/fit_two_peaks.py to fit a sum of two lorentzians functions for distinguishing carrier 20/02 modes.

The first graph is the OMC scan plot, the second is the curvefit for the second order carrier modes.

We expect the HOM spacing to be 0.588 MHz as per this entry and DCC T1500060 Table 25.

The spacing for the modes measured is 0.592 MHz.

From the heights of the two peaks this suggests mode-mismatch of the OMC to be C02+C20/C00 = (0.83+1.158)/(15.32+0.83+1.158) = 11.0% mode mis-match.

From the locked/unlocked powers on the OMC REFL PD the visibility on resonance is 1-(3.51+0.03/24.73+0.03) = 85.7% visibility.

If the total loss is 14.3%, this implies that the other non mode-matching losses are roughly 1.3%.

To run the OMC scan code go to 

/ligo/gitcommon/labutils/omc_scan/ and run 

python OMCscan_nosidebands.py 1370712036 100 "Sidebands off, 10W input" "single bounce" --verbose --make_plot -o 2
in the labutils conda environment and on git branch dev.

To do the double peak fitting run:

python fit_two_peaks_no_sidebands.py  
in the labutils conda environment and on git branch dev.

These scans were done with OM2 cold.

For comparison with new OMC measurements I used Sheila's code to process the visibility, but updated dit to use nds2utils instead of gwpy as I was having trouble using it to get data.

The code is attached and should be run in the nds2utils conda environment on the CDS workstations.

Power on refl diode when cavity is off resonance: 24.757 mW

Incident power on OMC breadboard (before QPD pickoff): 25.239 mW

Power on refl diode on resonance: 3.525 mW

Measured effiency (DCPD current/responsivity if QE=1)/ incident power on OMC breadboard: 70.4 %

assumed QE: 100 %

power in transmission (for this QE) 17.760 mW

HOM content infered: 13.472 %

Cavity transmission infered: 82.111 %

predicted efficiency () (R_inputBS * mode_matching * cavity_transmission * QE): 70.367 %

omc efficency for 00 mode (including pick off BS, cavity transmission, and QE): 81.323 %

round trip loss: 1605 (ppm)

Finesse: 371.769