I looked more closely at run 7534, to try to figure out which BCM to trust.
First off, just glancing at the RMS of the charge asymmetry measured in each BCM (all in ppm): Aq (rms): bcm 1 = 854, bcm 2 = 1267, bcm5=158, bcm6 = 158.
They can't all be right! At least bcm5 and 6 are consistent, but those numbers are too small... we don't see charge asymmetry widths that narrow unless the source and injector are very clean, and we know they are not right now.
Looking more at these bcms, (attachment 1), there are some large excursions in all bcms, well correlated, so everything is approximately measuring the same thing. Getting the scale wrong implies a non-linearity or pedestal error, but the resolution of monitor is implied in the width of these correlations. Clearly, bcm5/6 have the best correlation, bcms 1/5 are pretty correlated, and bcm2 looks to have the worst correlation of any of them.
My guess is that bcms 5 and 6 have very mis-set pedestals, scaling them badly. I note that the DD of asym_bcm1 = 5.4051*asym_bcm5 has a width of 315 (the scale factor was simply a ratio of the charge asymmetry RMS measured in each monitor). This is an improvement over asym_bcm1 - asym_bcm2 (950 ppm). I can't get such a narrow width for a bcm2/bcm5 double difference, no matter the scale factor... I guess this is obvious from attachment 1.
Fine, is bcm1 approximately correct? Well, our main detector is a pretty good charge monitor, too. If I take asym_qwk_mdallbars + (asym_bcm1+asym_bcm2)/2.0, I have effectively un- charge-normalized the MD asymmetry. I can then normalize vs. some other monitor... like bcm1 by itself, or bcm5.
Attachment 2 shows this, from clockwise from top left: - asym_mdallbars in standard charge normalization - asym_mdallbars after removing charge normalization - asym_mdallbars(unnormalized) - asym_bcm1 - asym_mdallbars(unnormalized) - 5.405*asym_bcm5
Normalizing by bcm1 gives the best width: 289 ppm! Normalizing by bcm5 (with the scale factor) gives a width of 393. I didn't show bcm2, but it was not possible to get a good width when normalizing to that monitor (which is again implied by attachment 1).
So: bcm 1 is our most trustworthy monitor. bcm 2 is just plain bad... noisy and also wrong. BCM 5 (and presumably 6) is wrong but reasonably precise, probably due to pedestal error or possibly non-linearity.
Action items:
* the bcm 1/2 double difference must be fixed. We can't live with just one reliable bcm, and right now we don't know what the resolution of bcm1 actually is. Probably the issue lies somewhere in the electronics for bcm2. bcm1 might not be optimal either, but we have no specific indication that it has a problem. bcm1 still seems to be our best charge monitor.
* The calibration of bms 5 and 6 must be revisited (although maybe Ramesh's newest calibration wasn't incorporated into the run 7534 analysis yet, so maybe this is already done?)
* We should check these conclusions against another run, to make sure that they are stable... after all, the double difference is all over the map in various runs, so maybe the problem is moving around. If it seems a reliable trend that bcm1 gives the best detector width, we should re-analyze some number of recent runs normalizing to bcm1 alone, to see if our widths become more stable over time.
One part of this really doesn't make sense to me. The bcm correlation plots seem to suggest that bcm5 is cleaner than bcm1, and the bcm5-bcm6 double-difference (when scaled by the apparent factor of 5.4) is only 224ppm, which is smaller than the bcm 1 - bcm5 DD width (315 ppm). So I would have guessed that bcm5 would provide a more narrow detector width. One possible explanation is that bcm5 and 6 have a common mode noise, which would not be revealed by their double difference but would show up as noise when compared to an uncorrelated monitor (such as bcm1, or the mdall asymmetry). Something to keep in mind.
Figure 1
Figure 2
