next up previous
Next: Two-hit merging distance Up: Hitfinding performance Previous: Number of contributors

Hitfinding efficiency

Here, a hit is declared ``found'' if a reconstructed hit is found within 2 cm of a padrow crossing. If more than one padrow crossing is nearby, then the reconstructed hit is associated with that crossing that has the same Monte Carlo ID as the reconstructed hit. As should be clear from the measurement of spatial resolution (below), using a 1 cm cut instead of 2 cm has little effect on the results.

Two quantities relevant when discussing overall hitfinding performance are the efficiency and the purity, defined as,

where a ``lost'' padrow crossing is one which is not reconstructed (usually because two crossings were too close to resolve), and a ``ghost'' hit is a reconstructed hit which cannot be associated with any padrow crossing (usually arising from noise).

Clearly, high efficiency is desirable, but a high hit purity may also be desired if the tracking software is easily confused or overwhelmed by noise hits.

Experimentation with the cluster/hitfinding algorithms with a variety of switch settings indicates that basically only one of the parameters strongly affects the efficiency and purity of the hitfinding, and this is the peak:valley ratio (setting DECON=0 may be seen as equivalent to setting MF_PV to ). Figure 16 shows the efficiency and purity averaged for the whole TPC when hitfinding in simulated p-p and Au-Au events. Clearly, for single p-p events, deconvolution is not needed, whereas in Au-Au events, much better efficiency can be obtained with a small price to pay in purity with a peak:valley cut of 1.3.

Figure 16: Purity and efficienty as a function of the peak:valley cut used in the multipeak finder for a p-p event and a Au-Au event. Note that also shown is performance with DECON=0, which is essentially an infinite peak:valley cut

As mentioned above, the peak:valley cut basically determines fully the purity and efficiency, so one can study a correlated plot of these two variables. Such plots are shown in Figure 17, where purity vs. efficiency is shown for the inner and outer sector in a Au-Au event. The values corresponding to the default of TCLPAR.MF_PV=1.3 are also indicated. Which value of the peak:valley cut is best for a given tracking algorithm must be explored. Also, it may well be that the ability to seperately set the peak:valley cut for inner and outer sectors would be desirable.

Figure 17: Seperately for the inner and outer sectors, the reconstructed hit purity and efficiency relations obtained for simulated Au-Au events, as the peak:valley cut value is varied. The filled points show performance for the default peak:valley=1.3 setting.

next up previous
Next: Two-hit merging distance Up: Hitfinding performance Previous: Number of contributors

Michael A. Lisa
Tue Feb 6 15:49:35 EST 1996