Fig. 1 ( 12 plots ) Plot 1 shows the missing mass distributions before and after fitting. Note the non-gaussian tails. Plot 2 shows the cda before and after fitting. Plot 3 shows the chi**2 p-distribution together with the chi**2 distribution. There is a long tail to the chi**2 distribution that results in the the low probability peak; the size of the peak suggests that ~ 8% of events are badly measured. The probabilty distribution has been flattened by multiplying the first estimate of the covariance matrix by 0.7 . Plots 4,5 explore the effect of cuts on probability, cda and chisq on the missing mass distribution. The probability and chisq cuts remove the non-gaussian tails. The CDA cut has no effect on the tails. Plot 6: This shows the ratio of fitted error to measurement error for the six fitted variables. The two-constraint fit improves the track momentum measurement accuracy by ~ 10% and the angle measurement by about a factor two. The beam variables are less influenced by the fit. Plots 7-12 show the pulls of the 6 variables. After the covariance matrix correction, all pulls have an rms close to unity; there is no evidence for systematic errors in any of the measured variables. Non-gaussian tails are present for all variables. Notice that the pulls in the momentum are identical for the track and beam; this is not understood. The pulls of the angles differ however, as can be seen on an event-by-event basis. To check the effect of systematic errors in momentum and angles, the input track momentum has been increased by 0.5 standard deviations and dx/dz (dy/dz) have been increased (decreased) by 0.5 standard deviations.
Fig. 2 ( 12 plots ) shows the fit results. Plots 7,8 ,9 (10) show that the pulls of the changed variables are displaced by -0.29, -0.62, +0.51,( -0.29 ) from zero, respectively, so demonstrating the sensitivity of the fit to systematic errors. ~/aplcon ./clg6mmcda.sh > clg6mmcda.out3 readfile6mmcda.F readfile6mmcda.kumac