DRAFT: Optimisation to reduce the ghost rate in the forward_then_matching_no_ut sequence
Superseded by !1442 (merged).
The goal of this MR is to provide an optimisation or alternative configuration to the HLT1 long track reconstruction for 2024 data taking using the forward_then_matching_no_ut sequence
From @tevans presentation, we see the quite high ghost rate impacts also the HLT1 BW division: about 40% in TwoTrackMVA and >50% in TrackMVA.
This optimisation reduces the ghost rate by more than a factor 2 accepting a loss of a couple % in tracking efficiency. More details and plots are given in presentation. The basic idea is to exploit the forward noUT to first reconstruct high-pt tracks (> 1-2GeV) with low ghost rate (~3%), meaning around 20% of all tracks, then running the matching to reconstruct all remaining tracks.
Throughput is not affected, actually improves by a couple of % as less ghosts (tracks) are reconstructed.
Bs->PhiPhi comparison plots
Using expected-2024 BsPhiPhi samples comparing master with this branch:
- Forward noUT optimisation on high-pt tracks reducing at maximum the ghost rate
- Seeding using realistic dxdy and matching changing fX value
- Seeing effect on full sequence
forward_then_matching_no_ut
- clear separation of tasks of forward noUT and seeding&matching is seen here:
