Improving BSM Efficiencies of the PrLongLivedTracking with separate SM and high mass tunings
Requires the MR: LHCb!5182 (merged) lhcb-datapkg/ParamFiles!153 (merged) Moore!5031 (merged)
This MR aims to increase the low Downstream Tracking efficiencies for BSM particles (seen in HNLs and Dark Higgs) that are most likely introduced due to the small x window size for the preselection hits, the correction to the estimated y slope in the UT, and thus also the Ghost Killer MLP.
To address this issue, the following changes are included:
- the preselection window in x for the preselection hits is 1.1x larger
- the y tolerance has been tightened
- the y slope estimate is changed to be the estimate coming from the Seed Tracks
- the Ghost Killer is trained again with these changes included and the cut is set to 0.65. The training sample consists of 2025 expected MinBias MC and B0 -> (K*(892)0 -> K+ pi-)(H-> mu mu) signal MC (for 4 masses: 500, 1500, 2500, 3500 MeV)
Additionally, a change in the reconstruction logic is proposed, in order to not affect the SM performance: The PrLongLivedTracking algorithm will be run twice:
- as is
- with track-flagged seed container and hit-flagged UT container, as well as the above mentioned tuning changes.
The track output container of both runs is then merged to make up the Downstream Track Container (after Pr Kf).
Both the minPt cut as well as the GhostProbabilityTool Cut (after the PrKF) are tightened.
Efficiency and ghost rate checks on the proposed changes were run on a Dstar->(D0 -> Ks Ks)pi sample, a Lambda_b->Lambda_0Gamma sample, a B_d->Jpsi Ks sample as well as the Dark Higgs signal MC and an HNL sample (B_c→μ(N→μπX)). Additionally, also the performance on HLT1Filtered Data and a Sim10f MinBias sample was tested.
The logs and plots can be accessed here: https://cernbox.cern.ch/s/iC5UD7EgHFjB8dq
also closes Moore#981 (closed)
FYI: @clangenb