Stub filtering: number of clusters per eta region

Number of silicon hit clusters remaining after three different stub-filtering scenarios, for 1000 events of single muon simulation in the ITk with 200 pileup interactions overlaid, in the 0.1< η <0.3, 0.3< φ <0.5 region.

Stub filtering: number of clusters per eta region, combined

Median number of silicon hit clusters remaining after three different stub-filtering scenarios, for 1000 events of single muon simulation in the ITk with 200 pileup interactions overlaid, in the 0.1 < η < 0.3, 0.7 < η < 0.9 and 2.0 < η < 2.2 regions (all with 0.3 < φ < 0.5). The ranges given are ± 1 σ from the median, i.e. including 68% of events.

Stub filtering: number of clusters per layer per eta region

Average number of silicon hit clusters remaining in each logical layer after three different stub-filtering scenarios, for 1000 events of single muon simulation in the ITk with 200 pileup interactions overlaid, in the 0.1 < η < 0.3, 0.3 < φ < 0.5 region.

Stub filtering: track candidate efficiency vs

Truth-matched track candidate efficiency as a function of muon , for three different stub-filtering scenarios followed by the Hough Transform, for 100k events of single muon simulation in the ITk with no pileup interactions, in the (a) 0.7 < η < 0.9 and (b) 0.1 < η < 0.3 regions (both with 0.3 < φ < 0.5). With no stub filtering, a tight selection requiring clusters from 7 out of 8 layers to fall in the same accumulator bin is imposed, while this can be loosened to 6 out of 8 layers in the stub filtering case to regain efficiency.

Stub filtering: radiation damage emulation

Truth-matched track-candidate efficiency as a function of muon , with and without medium stub filtering, followed by the Hough Transform, with varying levels of emulated radiation damage (0, 2 and 5% of clusters removed at random, uniformly over the detector) for 100k events of single muon simulation in the ITk with no pileup interactions, in the region 0.1 < η < 0.3, 0.3 < φ < 0.5. With no stub filtering, a tight selection requiring clusters from 7 out of 8 layers to fall in the same accumulator bin is imposed, while this can be loosened to 6 out of 8 layers in the stub filtering case, resulting in similar overall sensitivity to this emulated radiation damage.

Stub filtering: impact on Hardware Track Trigger

Truth-matched track efficiency (following the removal of duplicate track candidates) as a function of muon , for three different stub-filtering scenarios followed by the Hardware Track Trigger, for 100k events of single muon simulation in the ITk with no pileup interactions, in the 0.1 < η < 0.3, 0.3 < φ < 0.5 region.

ROC Curves

Performance of the neural network (NN) based fake and overlap removal algorithms, in terms of track candidate efficiency versus the number of track candidates remaining, in the 0.1 < η < 0.3, 0.3 < φ < 0.5 region, in single muon simulation overlaid with plus 200 pileup interactions. Efficiency is the number of track candidates (identified by the Hough Transform) matched to truth tracks, relative to the number of truth tracks, where these truth tracks are in both cases required to match offline-reconstructed tracks and can arise from all particles in the event; two tracks match if at least 50% of their hits are shared. Both offline and truth tracks have > 1 GeV. Results with and without stub filtering ("medium" level, and with the associated reduction in HT threshold as described earlier) are included.

Fake and overlap rejection vs

Average percentage of Hough Transform identified track candidates retained after the application of two neural networks (NN), as a function of , in the 0.1 < η < 0.3, 0.3 < φ < 0.5 region, in single muon events overlaid with plus 200 pileup interactions. "Hough Transform + NN Fake" refers to results after the application of an NN-based fake removal algorithm to all track candidates from the Hough Transform (with a threshold of 0.1 on the classifier). "Hough Transform + NN Fake + NN Overlap" refers to results after the further application of an NN-based overlap removal step with a threshold of 5 hits.

Number of track candidates, with and without stub-filtering

Distribution of the number of track candidates per event with and without stub filtering ("medium" level) in the 0.1 < η < 0.3, 0.3 < φ < 0.5 region, in single muon simulation overlaid with plus 200 pileup interactions. The results are after applying a neural network (NN) based fake removal algorithm and a NN based overlap removal step with a threshold of 5 hits on all track candidates from the Hough Transform step. The NNs are retrained for events with and without stub filtering, and similar track candidate efficiencies are observed for both scenarios.

Track reconstruction efficiency for a neural network (NN) based fake and overlap removal algorithm as a function of for The efficiencies are computed relative to the offline track reconstruction. “Hough Transformation (HT) only” refers to efficiency of the Hough Transformation algorithm step, while “HT & NN” refers to the efficiency after application of a neural network (NN) based fake removal algorithm (with a threshold of 0.005 on the classifier), and a NN based overlap removal step with a threshold of 5 hits. The average efficiency across the full range is provided in the legend. Detailed information on the HT and NN can be found in Ref. [1]

Distribution of the number of track candidates per event using hits for “Hough Transformation (HT)” refers to number of track candidates after the Hough Transformation (HT) algorithm step, “HT + NN Fake” refers to results after the application of a neural network (NN) based fake removal algorithm (with a threshold of 0.005 on the classifier) and “HT + NN Fake + NN Overlap” refers to results after a further application of a NN based overlap removal step with a threshold of 5 hits. The average average number of tracks per event is provided in the legend. Detailed information on the HT and NN can be found in Ref. [1]

Number of Fake tracks in the event retained after the application of two neural networks (NN), as a function of , in the , in single muon events overlaid with plus 200 pileup interactions. “Hough Transform + NN Fake” refers to results after the application of an NN-based fake removal algorithm to all track candidates from the Hough Transform (with a threshold of 0.005 on the classifier). “Hough Transform + NN Fake + NN Overlap” refers to results after the further application of an NN-based overlap removal step with a threshold of 5 hits. Detailed information on the HT and NN can be found in Ref. [1]