-
Timing, resolution, and other issues
:
-
I have been looking in detail at runs 1 (105 events, 17 < thit < 19, max trigger jitter = 31), 6 (315 events, 2 < thit < 3, max trigger jitter = 7), and 7 (304 events, 2 < thit < 3, max trigger jitter = 7).
This plot shows the "hit time" for the HOTs on the reconstructed tracks for run 1. If our interpretation is correct, only about half of the HOTs are real (the ones in the peak).
However, if one plots hit residuals versus hit time for the HOTs, like this, it appears that even hits that are clearly out of time still have good residuals. This means that either somehow good hits are assigned wrong times, or our tracking cuts somehow bias the resolution.
The resolution we currently get is excellent: about 18 microns in phi and z, as Gerry showed. This might seem a little too good if one takes into account the multiple scattering in the aluminum pipe, from which one would expect a resolution of about 35 microns. However, apart from the possible inaccuracies in estimating the pipe's parameters and particles' momenta, one should remember that the helix fit is likely to take the resolution out with the track's curvature. These plots show the phi residuals for helix and straight line fits, and the behaviour is as expected. A more unbiased accounting for the multiple scattering can be obtained from this curvature plot, from which one can estimate that the effective scattering angle is of the order of 17 mrad, which is about what one would expect.
-
Clustering (in)efficiency
:
All the results below are for run 1. In order to find our clustering efficiency, the following method was used:
- For a given track, determine how many HOTs it should have based on its trajectory.
- If a track is expected to have 4 HOTs but only 3 are found, see which strips the 4th HOT was supposed to have and check if those strips (+- 2) are in the list of available clusters. The comparison with the strips in the clusters is done with the help of the new function SvtCluster::getStripIDs() which John Walsh just created.
- Now, let us use the following naive model. Define the probability esp. pt. to make a space point as the product of two factors, one accounting for the phi and z hits being there, and the other, for their actually being combined into a space point. If we define ephi = probability to find a Phi hit, ez = probability to find a Z hit, and ecomb = probability to combine them, then esp. pt. = (ezephi)ecomb:
- the probability to find 4 HOTs when 4 are expected ~ ephi4ez4ecomb4
- the probability to find 3 HOTs + only the Phi strip for the fourth ~ 4 ephi4ez3(1 - ez)ecomb3
- the probability to find 3 HOTs + only the Z strip for the fourth ~ 4 ez4ephi3(1 - ephi)ecomb3
- the probability to find 3 HOTs + neither Phi nor Z strips for the fourth ~ 16 ephi3ez3(1 - ez)(1 - ephi)ecomb3
- the probability to find 3 HOTs + both Phi and Z strips for the fourth which are not combined into the space point ~ 4 ephi3ez3ecomb3(1 - ecomb)
-
Taking into account the 25% probability that the track may hit an inefficient channel, we get the following (clustering only) efficiencies:
-
ephi ~ 84%
-
ez ~ 84%
-
ecomb ~ 97%
-
This translates into the probability to find a space point of ~ 68%
. This is rather low!
Where can we be losing space points?
-
Timing cuts? Currently, for a space point to be made the default time cut requires that | tphi hit - tz hit | < 2. In my studies, I used the timing cut of 4. However, if there is a problem with assigning the correct time to the hits (see above), this cut may be quite inefficient.
-
Pulse height cuts? These are still to be understood better, but from these plots showing the pulse height distributions for HOTs in and out of time one would conclude that at least the minimum pulse height cut should be quite efficient. However, one also needs to consider the effect of the relative pulse height cut.
-
Perhaps it makes sense to lower the thresholds to get more real hits? It is true that now our noise occupancies are quite low with the cosmic run settings; but under real conditions it is likely that backgrounds and not noise will predominantly raise the occupancies.
Natalia Kuznetsova natalia@charm.physics.ucsb.edu 1/20/98
