I've analysed the monitoring results from all runs taken during February 2004 in order to assess the quality of the calibration data and to update the calibration banks FxCy ; x = P or Q chambers ; y = T(zero) or V(elocity). Input to the analysis is from the monitoring bank FTMO. Runs where the estimated error on Tzero is greater than 1 nsec have been rejected, leaving results from 480 runs to be analysed. My first attempt was coded to use the specific monitoring/calibration banks FPMO and FQMO, which contain all of the monitored quantities required for detailed (wire-to-wire) calibration - unfortunately I found that those banks have not been written since 30th January! The reason is not yet understood.
Fortunately the global Vdrift and Tzero for each chamber type are also recorded in the FTMO bank - duplication of data isn't always mistake! The results from FTMO are displayed as history plots vs run number and as frequency distributions in figures 1 to 4 below.
The P chamber Tzeros cluster very tightly about a constant value (mode) of 139.46 nsec. There are a few far-outliers (~12 in 480 ~ 3%) where the fitting procedure has failed. The reasons for these failures are not yet know but from HERAI experience "proton satellites" would be the first guess. In the upper and lower plots the dashed line shows the result of a simple weighted mean of all runs. In the lower plot the mode of the distribution is determined by a fit, which effectively excludes the outliers. The difference between mode and weighted mean is .052 nsecs which corresponds to about 1.7 microns in drift distance so that the effect of the outliers is completely negligible in practice. Note that the vast majority of runs give a fitted Tzero within +/- 0.5 nsec of the mode. That's equivalent to +/- 15 microns in space. Conclusion is that Tzero from the P chambers is stable and well calibrated over the whole of February.
Pretty much the same comments apply to the Q Tzeros as did to the Ps. Note that the anomalous runs are the same as in the Ps - which supports the hypothesis that the cause is external and common, eg proton satellites in the beam.
Shows the history and frequency distribution of fitted drift velocity. The position of the mode +/- 1% is indicated by the green box data point. +/- 1% corresponds to +/- 150 microns at the mean drift distance of 15 mm. Comments about outliers are the same as in the Tzero plots (the same runs are at fault). From the lower plot you can see that the fitted velocities cluster well within about 0.3% of the mode or 50 microns. So we have no problem in assigning the mode velocity to all runs in February. It may be worth noting that there's some evidence for the fitted velocity rising, through about 0.3%, during the course of a fill - see the upper plot in run ranges 371200-350, 371500-600 and 373350-500. Might this be related to the rise in temperature during a fill which the CJC claim to observe?
Much the same comments apply to the Q Velocities as did to the P Velocities. Its noticeable that the spread in fitted velocities is approximately twice as wide as for the Ps. We don't understand why yet - although it is reflected in the individual errors in the upper plot, so maybe just statistics. The errors on determination of Tzero are much the same for Ps and Qs, which suggests that the back edge of the drift distribution must be less precisely determined in Qs compared to Ps since V = constant/(Tback-Tzero).
The mode values for each plot have been use to update the FPCT,FQCT,FPCV and FQCV constants on the H1 database, applied from run 370589 onwards. For the range of runs shown in the plots tracks were, of course, reconstructed by h1rec/ftrec using whatever constants were on the database at the time. Here's a comparison;
Bank Constant New Old delta delta(drift) ---- -------- --- --- ----- ------------ FPCT P Tzero 139.463 137.87 +1.59 -52 microns FQCT Q Tzero 143.048 141.59 +1.46 -47 microns FPCV P Vdrift 32.724 32.702 +0.022 +10 microns FQCV Q Vdrift 32.351 32.407 -0.056 -26 microns
Vdrifts in microns/nsec (or km/sec if you prefer) and Ts in nsecs. The last column shows the difference in reconstructed drift distance produced by changing from old to new calibration constant, averaged over all drift distances.