Silicon Stub Finding Algorithm

This page shows current results from an algorithm in development which attempts to find silicon "stubs" in the SVX and compare them to EM clusters in the calorimeter. Presence of a stub at the correct phi indicates an electron, and the precise phi0 of the stub can be used to determine the electron charge.

VERY OLD talk from June 6, 2001 collaboration meeting.

The current version of this algorithm uses only SVX layers of silicon and fits the hits to a straight line rather than a circle.

April 15, 2002 The following explanations go with these plots. The data sample is the stripped tight Z sample available at fcdfsgi2:/cdf/data40a/s0/top/HighPtElectrons/stream_a/

• The top left plots shows, for each stub found, the maximum residual for hits on the stub vs. d0 (relative to the beam position). For the rest of the plots, we cut on abs(d0)<.04 and maxres<.007.
• The top right plot shows the number of hits on each stub. There are few three-hit stubs because they are only accepted if the hits are on the first three layers. The rest of the plots require at least four hits.
• The bottom left plot shows the E/p-like variable, delta(phi)/delta(phi)_{expected}. Delta(phi)=phi_strip-phi_stub. (Here strip means generically the showermax position, central or plug.) As expected the distribution is peaked around +/- 1 for each sign of charge. The yellow stubs are for central EM clusters, and the pink and blue are west and east plug (or vice versa).
  • There are more central electrons found, as expected from the statistics in the sample.
  • The plug electrons are not recognizeably peaked at +/- 1. We assume this is due to misalignment between the SVX and PES at the level required for our analysis.
• The bottom right plot gives the efficiency of the algorithm vs. eta in this tight Z sample. The black denotes the efficiency to find any stub; the green is the efficiency to find a stub that passes the cuts on d0, maxres, and nhits as described above.

We can attempt to align the plug with the SVX by using the stubs found here. When a central electron exists, we take its charge from its COT track. The plug electron is assumed to have the opposite sign of charge. For stubs in each plug which pass stringent quality cuts, we fit the expected vs. measured position to extract parameters for an overall small rotation in phi (alpha), and a translation (a,b).

The statistics are still pretty low, but the results are promising.

• The top two plots show the results of the fit. The details are a bit messy. The values and errors for the rotation and translation are at least reasonable.
• The bottom two plots show the delta(phi)/delta(phi)_{expected} after correcting for the misalignment extracted from the fit above. Red are negative tracks (positive central track) and green are positive tracks (negative central track). In the east the separation is good, but the two peaks are not centered at 1/-1. In the west it's a bit messier but the peaks are sort of in the right place.

April 27, 2002 Recent modifications to the algorithm:

• Added a consistency check in r-z: Still use only r-phi hits, but make sure the r-z extents of the ladders they come from are consistent with |z0|<150 and the (z,r) position of the smx cluster. Greatly reduces combinatorial background, and saves time by not fitting inconsistent sets of hits.
• From the above check, a range of possible z0 is extracted. This gives a rough estimate of the z vertex of the stub, usually to within 5 or 10 cm. This average of this range is used to calculate the Et of the EM cluster, which improves the resolution of delta(phi)/delta(phi)_{expected}.
• A few various improvements to the code that fits for the alignment parameters.

• First batch of plots
  • The descriptions are same as above, except the top right plot now shows the number of stubs found by seed. The black histogram is for "all" stubs (though some loose cuts are required to make it into the ntuple) and the green histogram is for "good" stubs (tight cuts).
  • The tight cuts are outlined in a green box in the top left plots.
• Second batch of plots
  • Same as second set of April 15 plots above.

To do list:

• Use PlugW sample for better statistics. (Disadvantage: no a priori charge information.)
• Check on behavior/performance of region finder.
• Check for PES geometry or algorithm problems.
• Improve PES position (with Lorentzian fit?).