The failure of the current CCPID to pick out CC events that hand scan true from the U of M analysis stems mainly from high angle, shorter events. Tracks that have a dcos > 0.6, but just barely, as well as only extending 12-18 planes. Taking this into note I did a further scan using the 180861 R1.18.2 Carrot MC events. I first pulled out events that failed the CCPID, but were indeed true CC events and then went through to hand scan and see how many under the class of higher angle or shorter track. Out of the first 70 that I scanned I found ~20 that I thought would hand scan clearly as CC, but were excluded using the curent CCPID. I believe the culprit to be the EventLength variable, because events shorter than ~75 planes are weighted largely NC, so the short stubby tracks need to be very, very CC in the other two parameters to escape this pitfall.

The idea:
    Hand scan from U of M as well as another independent hand scan of the MC has identified a class of events that have CC identifiable aspects and show a deficiency in one of the current CCPID parameters, the EventLength. A method being examined is to create a new CCPID parameter that looks at quantifying how much the track 'pokes' out of the shower. The first bit is of the procedure is projecting all the shower hits onto the line of the track vertex.

Fig. 1


The jtrk parameter is then track length divided by the largest distance projection hit. NC events should have a value of 1 for this parameter, because the reconstruction is putting a track through the shower, and CC events should have a value of >1 because the track should exist farther into the FD than the shower. Also for the above mentioned selection of 'missed' events that look very, very CC this parameter should be an improvement over the EventLength parameter.

As a quick comparison below is the Log plot of the NC and CC distributions for the EventLength variable as well as the new jtrk variable. Mine is so far  inferior... but tunable.

Fig. 2

    The jtrk variable has the possibility of cutting on the pulse height of the strips in the shower to get rid of superfluous cross-talk or other small PH hits that get inlcluded in the shower, but would never be included as a track hit.

    The following plots show different CC and NC probablities after cutting on the PH of the shower hits. The plot on the right has no cut, so all hits in the shower are included when calculating the jtrk parameter, while the plot on the left has a cut to only calculate the parameter for shower hits greater than 500 ADC from sigcor.

Fig. 3


    So the previous is all crap, and the next movement is that of using the charge weighted mean of the shower hits, which is from the hits projected onto the track direction. This value, the charge weighted mean, divided by either the track length gives a variable with some CC/NC separating power. One of the tantalizing prospects of the new variable is that the highest density of NC events does not occur at the same value as the highest density of CC events. For the EventLength parameter the largest number of CC and NC events occur in the 10-25 plane region. The left plot is a histogram of the of the charge weighted/track length events while the right plot is the probability plot of that variable.

Fig. 4