Log entry time 14:52:37 on November 02, 2005

Entry number 158846

Dithering widths are larger than uncorrected widths in a few slugs (15,17 19) and we have not tracked down why. Here are a few plots indicating the problem in slug 19.

Sorry for the annoying format.

First figure, panels 1-6. Panel 1: 12x position difference vs. pair number during slug. I've zoomed in on the one HUGE spike in 12x. There is another period with instabilities in 12x, but ignore that for now. I focus on the period near the 12x spike.

Panel 2: normalized, uncorrected det1 asymmetry for +/-2k events of the spike.

Panel 3: regression corrected det1 asymmetry for that time. Panel 4: dither corrected region for that time.

You see that the dither correction width is totally ruined by the outlier (ok, I should have put that on log plot, but trust me). Regression appears to be ok... but you should be suspicious. We know that det1 is very sensitive to energy, but there is no big outlier in the raw asymmtry? That means the 12x measurement is false. Dithering is right to put a big outlier... regression is wrong because the large outlier more-or-less defines the energy slope for this minirun.

Panel 5: regressed asymmetry for all pairs more than 2k pairs away from the spike (and with a 10 sigma cut on diff_bpm12x). Panel 6: dither det1 asymmetry for the same. Notice how the dither width seems quite reasonable now.

Figure 2:

Panel 1 : Reg_det1 - Dit_det1, NEAR the bpm12x spike. Sigma ~290 ppm Panel 2: reg_det1 - dit_det1 AWAY from the 12x spike. Sigma ~206 ppm. Both plots have the 10 sigma cut on diff_12x differences.

My point is just that regression and dithering aren't agreeing as well here, and I'm suggesting this is because regression is ignoring the energy jitter.

Panel 3: reg_det1 - dit_det1 vs. diff_bpm12x, AWAY from the spike. Pretty flat near the center... I could say more about the tails, but this is long enough already. Panel 4: reg_det1 - dit_det1 vs. diff_bpm12x, NEAR the spike.

See? The disagrement between these methods is due to energy. This isn't a watertight argument, but it is at least consistent with my hypothesis that the bpm screws up, which screws up regression. Dithering hangs tough...

Actually, somewhere we have the regression slope recorded for this pair. Maybe later I'll check on that. But for now, let me conclude by pointing out that in a final analysis, we will use a position burp cut as we did last year, to get rid of this kind of junk.

Figure 1

Figure 2