=Paper=
{{Paper
|id=Vol-2023/188-197-paper-29
|storemode=property
|title=The design and performance of the atlas inner detector trigger in high pileup collisions at 13 TeV at the Large Hadron Collider
|pdfUrl=https://ceur-ws.org/Vol-2023/188-197-paper-29.pdf
|volume=Vol-2023
|authors=Callum Kilby
}}
==The design and performance of the atlas inner detector trigger in high pileup collisions at 13 TeV at the Large Hadron Collider==
https://ceur-ws.org/Vol-2023/188-197-paper-29.pdf
Proceedings of the XXVI International Symposium on Nuclear Electronics & Computing (NEC’2017)
Becici, Budva, Montenegro, September 25 - 29, 2017
THE DESIGN AND PERFORMANCE OF THE ATLAS INNER
DETECTOR TRIGGER IN HIGH PILEUP COLLISIONS AT 13
TEV AT THE LARGE HADRON COLLIDER
C.R. Kilby, on behalf of the ATLAS collaboration
Royal Holloway, University of London, Egham Hill, Egham, Surrey, TW20 0EX, U.K
E-mail: callum.kilby@cern.ch
The design and performance of the ATLAS Inner Detector (ID) trigger algorithms running online on
the high level trigger (HLT) processor farm for 13 TeV LHC collision data with high pileup are
discussed. The HLT ID tracking is a vital component in all physics signatures in the ATLAS Trigger
for the precise selection of the rare or interesting events necessary for physics analysis without
overwhelming the offline data storage in terms of both size and rate. To cope with the high expected
interaction rates in the 13 TeV LHC collisions the ID trigger was redesigned during the 2013-15 long
shutdown. The performance of the ID Trigger in the 2016 data from 13 TeV LHC collisions has been
excellent and exceeded expectations as the interaction multiplicity increased throughout the year. The
detailed efficiencies and resolutions of the trigger in a wide range of physics signatures are presented,
to demonstrate how the trigger responded well under the extreme pileup conditions. The performance
of the ID Trigger algorithms in the first data from the even higher interaction multiplicity collisions
from 2017 are presented, and illustrates how the ID tracking continues to enable the ATLAS physics
program currently, and will continue to do so in the future.
Keywords: trigger, track, HLT, tracking, vertex, vertexing, run 2, lhc, atlas, two-stage, roi,
super-roi, pileup, performance, 2017, muon, jet, b-jet, efficiency, resolution, pixel, silicon,
strip
© 2017 Callum R. Kilby for the benefit of the ATLAS collaboration
188
Proceedings of the XXVI International Symposium on Nuclear Electronics & Computing (NEC’2017)
Becici, Budva, Montenegro, September 25 - 29, 2017
1. The ATLAS Inner Detector and Inner Detector trigger system
The ATLAS detector is the largest of the four main experiments placed at interaction points
around the ring of the Large Hadron Collider (LHC). ATLAS is a general-purpose cylindrical detector
formed of a number of sub-detectors, each designed to measure different signatures of particles
produced from LHC collisions [1].
The ATLAS Inner Detector (ID) is the sub-detector dedicated to measuring the paths charged
particles take, known as tracks, as well as the positions charged particles originate from, known as
vertices. The ID is formed of three sub-systems, which are each arranged in barrel or endcap
configurations, depending on whether they are placed in the barrel or endcap regions of ATLAS. The
first sub-system is the pixel detector, formed of layers of silicon pixel modules. Three layers of pixel
modules are included in the barrel and endcap regions, with a fourth layer known as the Insertable B
Layer (IBL) included closest to the LHC beamline in the barrel region. The IBL was added for LHC
Run 2, and provides improved impact parameter resolution, which also results in improved
identification of hadrons containing b quarks. Beyond the pixel detector is the Semiconducting
Tracker (SCT), which is formed of four barrel and nine endcap layers of stereo-doublet silicon
microstrip modules. Beyond the SCT is the Transition Radiation Tracker (TRT), which is formed of
many gaseous straw tubes, and provides on average 36 hits per track. The TRT can also provide
additional particle identification information.
The ID trigger is the part of the ATLAS High Level Trigger (HLT) system which performs
fast online track and vertex reconstruction, using measurements from the ID. The HLT itself is a
software-based trigger system which uses information from a hardware-based Level 1 (L1) trigger,
which has run prior to the HLT, and measurements from all ATLAS sub-detectors to select collision
events of interest. The HLT reduces the number of events to process and save to disk from a peak
input rate of 100 kHz to an output rate of approximately 1 kHz.
The ID trigger is an essential part of this rate reduction for nearly all physics signatures. The
tracks and vertices provided by the ID trigger allow the HLT to reconstruct physics-objects such as
muons and hadronic jets with enough spatial precision to make effective HLT trigger decisions. This
use of tracks and vertices becomes more important as the number of interactions per bunch crossing,
known as pileup, increases; however, this increase in pileup also makes track and vertex
reconstruction more difficult. In addition, track and vertex reconstruction are highly CPU intensive,
which could produce a bottleneck within the HLT due to having to wait for track and vertex
collections to be produced before running event reconstruction. This could result in “deadtime” in the
HLT, where the processing rate is slower than the input rate, resulting in no computing resources
being available during data taking.
The ID Trigger employs various methods to maintain accurate track and vertex reconstruction
while keeping runtimes down. Firstly, tracking is split between two stages: the Fast Track Finder
(FTF) which performs fast pattern recognition on hit clusters in the ID to produce an initial track
collection; this is followed by precision tracking (PT) which takes the tracks found by the FTF and
their hit clusters and improves the track quality while applying tighter requirements and handling hits
shared between multiple tracks. The performance of both of these stages is comparable to the full
offline track reconstruction algorithms. The second method used by the ID trigger is performing
tracking and vertexing in reduced detector volumes known as Regions of Interest (RoIs). These RoIs
are positioned based on information from the L1 trigger, and are sized according to the physics
signature of interest. These RoI volumes reduce the amount of detector measurements that must be
processed in each event. The third approach used by the ID trigger is employing multi-stage RoI
methods which use multiple RoIs to further reduce the detector volume to be processed, and can
further tailor RoI sizes to specific tasks within the ID trigger. This also allows reconstruction to be
performed iteratively.
189
Proceedings of the XXVI International Symposium on Nuclear Electronics & Computing (NEC’2017)
Becici, Budva, Montenegro, September 25 - 29, 2017
2. Multi-stage RoI methods
Two-stage tracking uses two RoIs in sequence, which allows reduced RoI volumes compared
to using a single RoI. In this method, an initial first-stage RoI is defined which is long along the
beamline, but narrow in phi and pseudorapidity1. The FTF is run in this first-stage RoI to produce an
initial track collection; from this, a track or vertex of interest can be determined. A second-stage RoI
can then be positioned centred on this track or vertex of interest. This second-stage RoI is much
shorter along the beamline, but wider in phi and pseudorapidity. The FTF is then run again in the
second-stage RoI, followed by the precision tracking to produce the final track collection. A cartoon
comparing the sizes of the RoIs used in the two-stage tracking method to the standard one-stage
method is shown in Figure .
Figure 1. Illustration of RoIs from the one-stage tracking (pink) and two-stage tracking (blue - first
stage, green - second stage) [2]
It can be seen that the total volume of the first- and second-stage RoIs of the two-stage
tracking is less than the volume of the RoI used in the one-stage tracking. Two-stage tracking is
employed in triggers for hadronic jets and hadronically decaying tau leptons. In these events, the
position of interest along the beamline is not known from L1 trigger information; two-stage tracking
allows this to be determined from the first-stage track collection. It has been seen from 2015 data that
two-stage tracking results in a runtime improvement per-RoI compared to the one-stage method of
approximately 20 ms for the FTF2 on average, and approximately 7 ms for the PT on average.
Super RoIs combine multiple individual RoIs into a single geometric region, which avoids
processing RoIs with overlapping volumes. This avoids processing the same tracks in multiple RoIs,
which reduces processing time, and avoids double counting of tracks which can bias vertex
reconstruction. In this method, multiple RoIs are defined which are long along the beamline, but
narrow in phi and pseudorapidity, where the phi and pseudorapidity directions of each RoI are
determined by information from the L1 trigger. These RoIs are then combined into the single super
RoI volume. Tracking and vertexing can then be run in the super RoI. An example cartoon comparing
multiple RoIs to a single super RoI is shown in Figure ; the overlapping regions can be seen in Figure
a).
1 𝜃
Pseudorapidity is defined as 𝜂 = − ln (tan ( )), where θ is the angle from the beamline
2
2
Comparing the FTF runtime for the single one-stage RoI to the sum of the FTF runtimes for the two two-stage
RoIs
190
Proceedings of the XXVI International Symposium on Nuclear Electronics & Computing (NEC’2017)
Becici, Budva, Montenegro, September 25 - 29, 2017
Figure 2. Illustration of a) multiple RoIs used in the standard RoI approach, b) single combined region
used in the super RoI approach. The regions are defined in the phi-pseudorapidity plane
Triggers for hadronic jets containing b quarks, labelled as jet triggers in the performance plots
in Section 3, employ both two-stage tracking and super RoI methods. In b-jet triggers, the individual
RoIs that combine to form the super RoI use the phi and pseudorapidity of jets passing the L1 trigger
requirements. Primary vertex reconstruction is run in the super RoI, and second-stage RoIs for each jet
can be positioned around this primary vertex. The FTF and PT are run in the second-stage RoIs,
followed by secondary vertex reconstruction needed for tagging hadrons containing b quarks.
3. Performance in 2017 data
During 2017, the LHC has been colliding protons at unprecedented scales as part of Run 2,
with a collision energy of 13 TeV and instantaneous collision luminosity up to a peak of
20.5 × 1033 cm-2 s-1. Due to these high rate conditions the collision pileup has also reached previously
unseen values, with a peak average pileup of 78.1.
Performance of the ID trigger tracking has been evaluated in a subset of 2017 13 TeV data,
showing the behaviour of the ID trigger in the strenuous conditions of Run 2. Performance is
evaluated using triggers which do not use tracks in their decision making to avoid biasing the
evaluation, but are otherwise identical to triggers used for physics data taking. Performance is defined
relative to tracks found by the full offline tracking algorithms; efficiency is defined as the fraction of
offline tracks found by the ID trigger tracking, and resolution is defined as the difference in track
quantities between the offline track and corresponding ID trigger track.
For the performance evaluation of muon triggers, offline tracks matched to offline-
reconstructed muons passing a medium quality requirement [3] are used. For the performance
evaluation of jet triggers, all offline tracks within the trigger RoIs are used. Basic quality criteria on
the number of track hits are also applied to offline tracks in both muon and jet trigger performance
evaluations.
Figure shows the tracking efficiency in muon triggers as a function of average pileup. It is
seen that the efficiency is greater than 99% across the full range of average pileup. In general, the ID
trigger has been optimised to be robust up to a pileup of 80, and could maintain performance beyond
this.
Figure and Figure show the tracking efficiency in muon triggers as a function of offline
muon pseudorapidity and transverse momentum. It is seen that the efficiency is greater than 99%
across the full range of both pseudorapidity and transverse momentum.
191
Proceedings of the XXVI International Symposium on Nuclear Electronics & Computing (NEC’2017)
Becici, Budva, Montenegro, September 25 - 29, 2017
Figure and Figure respectively show the transverse impact parameter3 resolution and spatial
resolution along the beamline for muon triggers, as a function of offline muon transverse momentum.
It is seen that the resolutions are in general excellent, going down to approximately 10 μm for
transverse impact parameter and approximately 50 μm for position along the beamline resolution. It is
seen that the PT improves track resolution compared to tracks from the FTF.
Figure shows the tracking efficiency in a jet trigger as a function of average pileup. Similar to
muon triggers, the efficiency shows little dependence on pileup.
Figure and Figure show the tracking efficiency in a jet trigger as a function of offline track
pseudorapidity and offline track transverse momentum. The efficiency is greater than 98% for central
pseudorapidity, and is greater than 98% over most of the range of transverse momentum.
Figure and Figure respectively show the transverse impact parameter resolution and spatial
resolution along the beamline for a jet trigger, as a function of offline track pseudorapidity. Good
resolution performance is seen, going down to approximately 20 μm for transverse impact parameter
and approximately 30 μm for position along the beamline resolution. The resolutions degrade as a
Figure 3. Track finding efficiency in muon triggers as a function of mean number of interactions per
bunch crossing [2]
function of pseudorapidity due to tracks passing through more detector material, resulting in increased
multiple scattering.
3
Transverse impact parameter is defined as the distance of closest approach to the beamline, in the plane
transverse to the beamline
192
Proceedings of the XXVI International Symposium on Nuclear Electronics & Computing (NEC’2017)
Becici, Budva, Montenegro, September 25 - 29, 2017
Figure 4. Track finding efficiency in muon triggers as a function of offline muon pseudorapidity [2]
Figure 5. Track finding efficiency in muon triggers as a function of offline muon transverse momentum [2]
193
Proceedings of the XXVI International Symposium on Nuclear Electronics & Computing (NEC’2017)
Becici, Budva, Montenegro, September 25 - 29, 2017
Figure 6. Track transverse impact parameter resolution in muon triggers as a function of offline muon
transverse momentum [2]
Figure 7. Track spatial resolution along the beamline in muon triggers as a function of offline muon
transverse momentum [2]
194
Proceedings of the XXVI International Symposium on Nuclear Electronics & Computing (NEC’2017)
Becici, Budva, Montenegro, September 25 - 29, 2017
Figure 8. Track finding efficiency in a jet trigger as a function of mean number of interactions per
bunch crossing [2]
Figure 9. Track finding efficiency in a jet trigger as a function of offline track pseudorapidity [2]
195
Proceedings of the XXVI International Symposium on Nuclear Electronics & Computing (NEC’2017)
Becici, Budva, Montenegro, September 25 - 29, 2017
Figure 10. Track finding efficiency in a jet trigger as a function of offline track transverse momentum [2]
Figure 11. Track transverse impact parameter resolution in a jet trigger as a function of offline track
pseudorapidity [2]
196
Proceedings of the XXVI International Symposium on Nuclear Electronics & Computing (NEC’2017)
Becici, Budva, Montenegro, September 25 - 29, 2017
Figure 12. Track spatial resolution along the beamline in a jet trigger as a function of offline track
pseudorapidity [2]
4. Conclusion
The ID trigger is a vital component of the ATLAS trigger system, and the high trigger
efficiencies and necessary rate reduction would not be possible without it. During the strenuous
conditions of LHC Run 2, the ID trigger has employed multi-stage RoI methods to mitigate the high
collision rate and pileup: two-stage tracking allows further reduction in detector volumes processed by
the ID trigger, while super RoIs avoid multiple processing and double counting of tracks.
The performance of the ID trigger in 2017 data taking is seen to be excellent, showing high
track finding efficiencies and accurate spatial resolutions. In addition, track finding efficiency is seen
to be insensitive to the high pileup conditions of LHC Run 2. The ID trigger is expected to continue
providing excellent performance as LHC Run 2 continues.
References
[1] The ATLAS Collaboration. The ATLAS Experiment at the CERN Large Hadron Collider JINST 3
(2008) S08003.
[2] The ATLAS Collaboration. ATLAS HLT Tracking Public Results. Available at:
https://twiki.cern.ch/twiki/bin/view/AtlasPublic/HLTTrackingPublicResults
[3] The ATLAS Collaboration. Muon reconstruction performance of the ATLAS detector in
proton-proton collision data at √s = 13 TeV. Eur. Phys. J. C76 (2016) no.5, 292.
197