Tracking performance in the CMS High Level trigger ( CMS DP-2018/038)
Tracking in the high level trigger (HLT) is performed iteratively, starting with tight requirements for the track seeds, which become looser for each subsequent iteration. Hits in the tracking detectors already used in a track are removed at beginning of the next iteration. The general track reconstruction in the HLT consists of three iterations. The first two require the maximum of four consecutive hits in the pixel detector to seed the tracking. These iterations target first high and then lower pT tracks and use the full volume of the pixel detector. The third iteration relaxes the requirement on the number of hits in the track seeds to three and is restricted to the vicinity of jet candidates identified from calorimeter information and the tracks reconstructed in the two previous iterations.
During the operation in 2017, several issues with the Phase-I pixel detector where identified that lead to a non-negligible fraction of non-active pixel modules in each event. Most notable was the failure of DCDC converters used in the powering of the detector, which resulted in an increasing fraction of non-active modules towards the end of the data taking period.
During the year end technical stop 2017/2018 the pixel detector was equipped with new converters and the initial performance was restored. To safeguard against a possible re-occurrence of this problem and other possible detector failures, an additional recovery iteration was added. Track seeds consisting of just two pixel detector hits are created in regions of the detector where two inactive modules overlap as seen from the interaction point. Due to the limited CPU time available for the HLT reconstruction, this iteration is
restricted to tracks with pT > 1.2 GeV.
The tracking efficiency and fake rate are measured in simulated ttbar events with a mean number of additional interaction of 50. They are defined with respect to Monte Carlo truth information, where the tracks of the simulated particles are matched to reconstructed tracks based on shared hits in the tracking detectors.
The tracking efficiency is defined as the fraction of simulated particles from the signal interaction with pT > 0.9 GeV, abs(eta)< 2.5, and longitudinal (transverse) impact parameters < 35(70) cm that are matched to a reconstructed track.
The fake rate is defined as the fraction of reconstructed tracks that could not be matched to a simulated particle.
To realistically model the effect of an imperfect pixel detector, a map of inactive modules representing the status of the real detector as of June 2018 is applied
to the simulation. As the inactive modules are not distributed uniformly throughout the detector, the tracking performance is expected to be asymmetric in track eta and phi. For reference, the efficiency that would be achieved with a perfect detector is also shown. As the doublet-seeded iteration is not run in the case of the perfect detector, in some cases the tracking efficiency with the realistic detector can be higher than with the perfect detector.
| Tracking efficiency versus pT Tracking efficiency measured with respect to Monte Carlo truth information as a function of simulated track pT. The contributions to the total efficiency from the dfferent tracking iterations are shown in the different colors. The initial three iterations are shown in shades of blue while the contribution of the doublet recovery iteration is shown in violet. A map of inactive pixel detector modules reflecting the state of the detector at the beginning of June 2018 has been applied to the simulation. The performance that would be achieved with a perfect detector is shown as a red line. In the plateau region around 20 GeV, the efficiency observed with the realistic detector conditions is about 5% lower than with the perfect detector. The doublet-seeded tracking iteration is able to recover a significant fraction of this efficiency loss above the pT threshold of 1.2 GeV. [Get pdf version] Contact: Jan-Frederik Schulte |
|
| Tracking efficiency versus eta Tracking efficiency measured with respect to Monte Carlo truth information as a function of simulated track eta. The contributions to the total efficiency from the different tracking iterations are shown in the different colors. The initial three iterations are shown in shades of blue while the contribution of the doublet recovery iteration is shown in violet. A map of inactive pixel detector modules reflecting the state of the detector at the beginning of June 2018 has been applied to the simulation. The performance that would be achieved with a perfect detector is shown as a red line. The efficiency loss compared to the perfect detector is more pronounced in the central part of the detector. The performance of the recovery procedure is not uniform across the eta range as it takes two overlapping inactive modules to trigger, making it dependent on the specific distribution of these modules. [Get pdf version] Contact: Jan-Frederik Schulte |
|
| Tracking efficiency versus phi Tracking efficiency measured with respect to Monte Carlo truth information as a function of simulated track phi. The contributions to the total efficiency from the different tracking iterations are shown in the different colors. The initial three iterations are shown in shades of blue while the contribution of the doublet recovery iteration is shown in violet. A map of inactive pixel detector modules reflecting the state of the detector at the beginning of June 2018 has been applied to the simulation. The performance that would be achieved with a perfect detector is shown as a red line. The efficiency loss compared to the perfect detector is concentrated in the region around phi= 0.6, where a significant region of inactive modules is present. The doublet-seeded iteration recovers most of the lost efficiency in this region. The performance of the recovery procedure is not uniform across the phi range as it takes two overlapping inactive modules to trigger, making it dependent on the specific distribution of these modules. [Get pdf version] Contact: Jan-Frederik Schulte |
|
| Tracking efficiency versus PU Tracking efficiency measured with respect to Monte Carlo truth information as a function of the number of PU interactions. The contributions to the total efficiency from the different tracking iterations are shown in the different colors. The initial three iterations are shown in shades of blue while the contribution of the doublet recovery iteration is shown in violet. A map of inactive pixel detector modules reflecting the state of the detector at the beginning of June 2018 has been applied to the simulation. The performance that would be achieved with a perfect detector is shown as a red line. The efficiency is robust against the presence of PU, decreasing only slightly with the number of additional interactions. The performance of the doublet-seeded recovery is also independent of the PU conditions. [Get pdf version] Contact: Jan-Frederik Schulte |
|
| Tracking fake rate versus pT Tracking fake rate measured with respect to Monte Carlo truth information as a function of simulated track pT. The fake rate for the first three iterations (default tracking) is shown in dark blue while the fake rate after including the doublet recovery iteration is shown in violet. The fake rate that would be observed with the perfect detector is shown in dark red. There is no difference between the fake rates with the perfect and realistic detector conditions. Taking into account the doublet-seeded recovery iterations, a slight increase of the fake rate above the pT threshold of this iteration is observed. [Get pdf version] Contact: Jan-Frederik Schulte |
|
| Tracking fake rate versus eta Tracking fake rate measured with respect to Monte Carlo truth information as a function of simulated track eta. The fake rate for the first three iterations (default tracking) is shown in dark blue while the fake rate after including the doublet recovery iteration is shown in violet. The fake rate that would be observed with the perfect detector is shown in dark red. There is no difference between the fake rates with the perfect and realistic detector conditions. As the fake rate is integrated over all pT values, no significant increase in the fake rate is observed for the doublet-seeded recovery iteration. [Get pdf version] Contact: Jan-Frederik Schulte |
|
| Tracking fake rate versus phi Tracking fake rate measured with respect to Monte Carlo truth information as a function of simulated track phi. The fake rate for the first three iterations (default tracking) is shown in dark blue while the fake rate after including the doublet recovery iteration is shown in violet. The fake rate that would be observed with the perfect detector is shown in dark red. There is no difference between the fake rates with the perfect and realistic detector conditions. As the fake rate is integrated over all pT values, no significant increase in the fake rate is observed for the doublet-seeded recovery iteration. [Get pdf version] Contact: Jan-Frederik Schulte |
|
| Tracking fake rate versus PU Tracking fake rate measured with respect to Monte Carlo truth information as a function of the number of PU interactions. The fake rate for the first three iterations (default tracking) is shown in dark blue while the fake rate after including the doublet recovery iteration is shown in violet. The fake rate that would be observed with the perfect detector is shown in dark red. In all cases, it is increasing with the number of additional PU interactions. There is no difference between the fake rates with the perfect and realistic detector conditions. As the fake rate is integrated over all pT values, no significant increase in the fake rate is observed for the doublet-seeded recovery iteration. [Get pdf version] Contact: Jan-Frederik Schulte |
|
-- ElisabettaGallo - 2018-07-07
CMSPublic Web>PhysicsResults>HighLevelTriggerRunIIResults>HLTplots2018DataTracking (2018-07-07, ElisabettaGallo)
EditAttachPDF
CMSPublic Web
Create New Topic
Index
Search
Changes
Notifications
Statistics
Preferences
Copyright &© 2008-2023 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.