Current Performance in the Velo#
The performance evaluation of the GNN-based pipeline depicted below does not account for inference time. Consequently, while the GNN-based pipeline demonstrates superior physics performance compared to the default track finding algorithm within the Velo subdetector, it is imperative to acknowledge that substantial efforts are required to tailor the pipeline to meet the demanding high-throughput constraints required for the first-level trigger of the LHCb experiment.
\(p\)-\(p\) collisions with spillover#
The default classical algorithm in Allen have the performance detailed by the 2 tables below.
Categories |
Efficiency |
Average efficiency |
Clone rate |
Average hit purity |
Average hit efficiency |
|---|---|---|---|---|---|
Velo |
94.30% |
94.43% |
3.85% |
99.36% |
93.76% |
Long |
99.12% |
99.18% |
2.66% |
99.64% |
96.37% |
Velo, no electrons |
98.22% |
98.31% |
3.10% |
99.66% |
95.36% |
Velo only, no electrons |
96.84% |
96.98% |
3.84% |
99.50% |
93.89% |
Velo only, only electrons |
67.91% |
67.49% |
10.27% |
97.35% |
79.21% |
Long, only electrons |
97.11% |
97.29% |
4.25% |
97.67% |
95.24% |
Velo only, from strange |
93.53% |
93.77% |
5.60% |
99.36% |
90.05% |
Velo only, from \(K^{0}_{S}\) |
94.83% |
94.97% |
5.40% |
99.54% |
90.25% |
Long, from \(K^{0}_{S}\) |
98.52% |
98.80% |
2.55% |
99.62% |
96.26% |
Categories |
# ghosts |
# tracks |
Ghost rate |
|---|---|---|---|
Everything |
27,209 |
1,245,631 |
2.18% |
The etx4velo algorithm performances are given by the 2 tables below.
Categories |
Efficiency |
Average efficiency |
Clone rate |
Average hit purity |
Average hit efficiency |
|---|---|---|---|---|---|
Velo |
97.26% |
97.22% |
1.66% |
99.73% |
98.02% |
Long |
99.49% |
99.46% |
1.35% |
99.83% |
98.74% |
Velo, no electrons |
98.80% |
98.78% |
0.95% |
99.89% |
98.65% |
Velo only, no electrons |
97.86% |
97.85% |
1.02% |
99.82% |
98.32% |
Velo only, only electrons |
86.69% |
86.11% |
4.97% |
98.99% |
93.88% |
Long, only electrons |
99.22% |
99.22% |
7.31% |
98.46% |
96.79% |
Velo only, from strange |
95.25% |
95.12% |
1.77% |
99.64% |
96.05% |
Velo only, from \(K^{0}_{S}\) |
96.10% |
95.95% |
1.72% |
99.75% |
96.04% |
Long, from \(K^{0}_{S}\) |
98.95% |
98.99% |
0.80% |
99.89% |
98.83% |
Categories |
# ghosts |
# tracks |
Ghost rate |
|---|---|---|---|
Everything |
10,151 |
1,258,270 |
0.81% |
5,000 events where used for this evaluation.
\(Pb\)-\(Pb\) collisions#
The default classical algorithm in Allen have the performance detailed by the 2 tables below.
Categories |
Efficiency |
Average efficiency |
Clone rate |
Average hit purity |
Average hit efficiency |
|---|---|---|---|---|---|
Velo |
91.51% |
93.36% |
4.03% |
98.08% |
92.88% |
Long |
97.60% |
98.64% |
2.88% |
98.59% |
95.70% |
Velo, no electrons |
96.03% |
97.62% |
3.51% |
98.43% |
94.30% |
Velo only, only electrons |
61.63% |
63.48% |
9.17% |
95.30% |
78.49% |
Long, only electrons |
91.70% |
94.65% |
3.76% |
96.05% |
94.55% |
Velo only, from \(K^{0}_{S}\) |
83.34% |
90.19% |
5.34% |
97.57% |
88.39% |
Long, from \(K^{0}_{S}\) |
92.27% |
95.29% |
2.91% |
98.11% |
94.53% |
Categories |
# ghosts |
# tracks |
Ghost rate |
|---|---|---|---|
Everything |
110,064 |
1,022,995 |
10.76% |
The etx4velo algorithm performances are given by the 2 tables below.
Categories |
Efficiency |
Average efficiency |
Clone rate |
Average hit purity |
Average hit efficiency |
|---|---|---|---|---|---|
Velo |
95.50% |
96.42% |
2.37% |
99.70% |
96.49% |
Long |
98.90% |
99.24% |
1.78% |
99.79% |
98.09% |
Velo, no electrons |
97.69% |
98.35% |
1.63% |
99.82% |
97.62% |
Velo only, only electrons |
80.31% |
82.15% |
6.75% |
99.12% |
88.07% |
Long, only electrons |
98.38% |
98.68% |
6.51% |
98.64% |
95.78% |
Velo only, from \(K^{0}_{S}\) |
88.66% |
92.75% |
3.50% |
99.56% |
90.15% |
Long, from \(K^{0}_{S}\) |
96.47% |
97.79% |
1.60% |
99.75% |
96.71% |
Categories |
# ghosts |
# tracks |
Ghost rate |
|---|---|---|---|
Everything |
9,617 |
983,172 |
0.98% |
1,000 events stored under LFN:/lhcb/MC/Upgrade/XDIGI/00165652/0000/00165652_00000699_1.xdigi
where used for this evaluation.