| 1 |
The TaNC algorithm has been optimized for the early stages of LHC operation.
|
| 2 |
The quality selections applied by the decay mode reconstruction algorithm will
|
| 3 |
need to be re--tuned and the neural networks retrained after each significant
|
| 4 |
change in instantaneous luminosity to remove the effects of multiple
|
| 5 |
interactions in the same bunch crossing, and after collision energy changes due
|
| 6 |
to the increased energy of the multiple scattering events.
|
| 7 |
|
| 8 |
The dependence of the algorithm on the robustness of Monte
|
| 9 |
Carlo simulations can be minimized by using real events as the background
|
| 10 |
training sample. As the QCD di--jet production rate is many orders of magnitude
|
| 11 |
larger than that of tau leptons, it is relatively straightforward to obtain a
|
| 12 |
pure background training sample with early CMS datasets.
|
| 13 |
|
| 14 |
The relatively long lifetime of the tau lepton ($c\tau = 87\mu$m) permits the
|
| 15 |
possibility that using displaced decay vertex information could provide
|
| 16 |
discrimination power against quark and gluon jets, which are expected to be
|
| 17 |
produced at the primary vertex. A feasibility study is planned to determine if
|
| 18 |
adding vertex related observables to the neural networks could improve the
|
| 19 |
performance of the TaNC.
|
