In three previous postings I looked at observing efficiency in BICEP/Keck: 20190509_b3_obseff, 20201019_k2018_obseff and 20210202_k2013_obseff.
Sarah took some of these numbers and put them in a google spreadsheet - see tab "Observation_Efficiencies". She ends up assuming 0.195/0.191/0.136 at 90/150/220GHz.
I believe Reijo and Andrea used these when making the "Design Tool" sims described in this github.
In 20210303_dt1_vs_bk15 I scaled from BK15 \(N_\ell\)'s to PBD using ratios of ideal NET's, number of detector-years, and the relative hits map provided from the DT sims. I then compared the white noise levels and, while the agreement is not terrible, we would like to know why it is not better.
One possible problem is that the BK efficiency numbers I provided so far didn't actually come from the BK15 map set - they were some mix of single year BICEP3, Keck and BICEP2 maps. Here I extract numbers from the multi year BK15 map set directly.
The plot below shows the pair-difference hit maps for the three bands of BK15.
Fig 1:
The number of detector-years going into BK15 has "traditionally" been taken as:
\(N_{\rm det-yr, BK15}\) = 288*4, 512*11.2 and 512*2
These date back to Ben Racine in 20190829_noise_params_DSR where the 11.2 at 150GHz is justified as an effective number.
Taking the ratio of the number of pair-seconds binned into the maps to the nominal [288*4,512*11.2,512*2]*(365*24*60^2) gives efficiencies of 0.172/0.180/0.124. So that is a bit lower than Sarah chose but not a big difference.
The above efficiency factors include data cuts - and hence also array yield since non-working detectors will certainly get cut. However, even when a detector makes it past cuts it get co-added into the map with a weight taken as the inverse of the tod variance over the scan set. In an ideal world all detectors would have the same weight and these would also be constant over time. But in reality neither is true. The plot below shows the weights versus time for Keck 2015. The horizontal bluer bands come from detector wafers which underperform versus what is clearly possible (the yellower/redder bands). (There are 8/4/8 wafers at 90/150/220 in this year.) We can see that the primary effect is intrinsic to the wafers. There are about 12 cycles of deck rotation in the above plot which varies the elevation angle through which each detector pair is looking through the atmosphere by up to +/- 10 degrees - we do not see this pattern strongly.
Fig 2:
From the above we can see that the penalty from non uniform weights is considerable. It should be noted that in the DT sims this distribution is assumed to be close to a delta function at the calculated NET (with a small variation for elevation angle).
Applying absolute calibration to the weights and accumulating over all years going into the BK15 map set produces the histograms below. In an ideal world there would be a hard cliff on the high side representing "ideal" detectors on the best days. We sort of see this for the 90/220GHz but not so much for 150GHz. The dotted black lines in the plots below are the mean of the weights and the red the 90th percentiles.
Fig 3:
To estimate the penalty from the shape of the weight distribution we can take the mean of the weight divided by the some point towards the top end of the distribution. Taking the 90th percentile we get penalty factors 0.79/0.62/0.67 at 90/150/220GHz.
This is a real additional (in)efficiency factor which is currently implicitly included in the scaling calculation but not included in the DT sims.
Bottom line is that if we want the DT sims to match the "scaling from BK15 \(N_\ell\)'s" then the overall efficiency factor should be something like 0.172/0.180/0.124 multiplied by 0.79/0.62/0.67 for a total of 0.136/0.112/0.083. These numbers are 70/59/61% of the currently assumed numbers.