24.2.4. Track Quality Indicator#
There are three track quality indicators based on BDTs. The CDC Quality Indicator (CDC QI) utilizes CDC level quantities.
Analogously, the VXDTF2 QI is based on VXD level quantities and the RecoTrack QI is using quantities that are available on RecoTrack level.
All of them show large potential to reject fake and clone tracks while truth tracks are kept. As BDTs tend to enhance data-MC disagreements at their border regions (very truth-like or very fake-like), the quality indicators were revisited with respect to the data-MC disagreement that they introduce.
Due to limited person power, only the CDC QI is activated, and the RecoTrack QI is available at the moment. The CDC QI was trained on MC13 with a very reduced set of quantities that showed reasonable data-MC agreements. The most important variables to separate fakes and clones from truth tracks are the number of super layers that contain hits of the track, z0 and the sz-slope.
To further reduce the risk of introducing data-MC disagreements, a very conservative cut value of 0.7 was chosen. It was shown that it would be beneficial to select a tighter cut value (like 0.9) in future. One way of avoiding to introduce data-MC discrepancies would be to choose slightly different values for data and MC that agree with respect of the ratio of tracks that are kept.
The CDC QI rejection cut is conducted just after the CDC Track Finding. Thus, less CDCTracks are forwarded
to the SVD track finding which means that less SVD hits are connected to these tracks and more SVD hits remain “free” for the
SVD Track Finding. It was shown that several truth tracks, that were rejected by the CDC QI cut, are found again
in the SVD. Due to less “noise” due to fake and clone CDCTracks, the SVD track finding seems to find even more truth tracks than without CDC QI cut.
As a consequence, the overall track finding efficiency increases slightly (by up to 0.5%). This could be improved to more than 2% while
the fake and clone rate are reduced further by choosing a tighter cut on the CDC QI as mentioned above.
The RecoTrack QI was trained with CDC QI, VXD quantities and track quantities. The current weight file is trained on MC16, using the script
combined_quality_estimator_teacher.py ,
which uses the official basf2_mva.teacher to train the BDT.
It no longer needs VXD QI as input, as it directly use VXD quantities that used to train VXD QI.
Since one track could contain two SVD parts, one before CDC and one after CDC, the current method can take this into account.
In the moment no cut on the RecoTrack QI is performed.
Instead, the RecoTrack QI is available at the analysis level by the variable :b2:var:`trackQualityIndicator`.
It is assumed that the modelling of more properties with large separation potential between fakes, clones and truth tracks improved in the last MC productions. Moreover, the activation of the VXD- and Recotrack-QI would further increase the fake- and clone rejection by 50% and 30% compared to the current value while maintaining the same track finding efficiency.
More details concerning the different QIs and how to train them can be found on XWiki: MVA Track Quality Indicator.