Online Event Selection at the CMS experiment

Triggering in the high‐rate environment of the LHC is a challenging task. The CMS experiment has developed a two‐stage trigger system. The Level‐1 Trigger is based on custom hardware devices and is designed to reduce the 40 MHz LHC bunch‐crossing rate to a maximum event rate of ∼ 100 kHz. The further reduction of the event rate to O(100 Hz), suitable for permanent storage, is performed in the High‐Level Trigger (HLT) which is based on a farm of commercial processors. The methods used for object identification and reconstruction are presented. The CMS event selection strategy is discussed. The performance of the HLT is also given.


F Summary of Level-1 Trigger
The Level-1 Trigger System [F-1] is organized into three major subsystems: the Level-1 calorimeter ger, the Level-1 muon trigger, and the Level-1 global trigger.The muon trigger is further organized subsystems representing the 3 different muon detector systems, the Drift Tube Trigger in the barrel Cathode Strip Chamber (CSC) trigger in the endcap and the Resistive Plate Chamber (RPC) trigger ering both barrel and endcap.The Level-1 muon trigger also has a global muon trigger that combines trigger information from the DT, CSC and RPC trigger systems and sends this to the Level-1 global ger.A diagram of the Level-1 Trigger system is shown in Figure F-1.Requirements driven by LHC discovery physics: • Identify high-p T leptons (including taus) and photons.Single and Combined triggers.
• All trigger thresholds and conditions must be programmable (large uncertainties in backgrounds and signals) • Need to include overlapping and min-bias triggers to well understand efficiencies • Large rejections factors needed: 40MHz (× ∼ 20 ev/bx) → 100 kHz.
• Level-1 uses muon and calorimeter detector data only • Special-purpose hardware (ASICS) but also FPGAs • Data stored on detector during fixed Level-1 latency.128BX = 3.2µs Electron/Photon imum E T tower of its four broad side neighbours.This summed transverse energy provides a shar ciency turn-on with the true E T of the particles.
The non-isolated candidate requires passing of two shower profile vetoes, the first of which is b the fine-grain ECAL crystal energy profile.The second is based on HCAL to ECAL energy com e.g.H/E less than 5% (HAC veto).
The isolated candidate requires passing of two additional vetoes, the first of which is based on the of FG and HAC Vetoes on all eight nearest neighbours, and the second is based on there being at l quiet corner, i.e., one of the five-tower corners has all towers below a programmable threshold, GeV.Each candidate is characterized by the (η,φ) indexes of the calorimeter region where the hit located.

F.1.2 Electron/Photon Triggers
In each calorimeter region (4×4 trigger towers) the highest E T non-isolated and isolated electron The nominal electron/photon algorithm allows both non-isolated and isolated stream stream uses only the hit tower information except for adding in any leakage energy f neighbour tower.This stream will be used at low luminosity to provide the B-electron tion and shower shape trigger cuts are programmable and can be adjusted to the runn example, at high luminosity the isolation cuts could be relaxed to take into account h gies.The specification of the electron/photon triggers also includes the definition of th it is applicable.In particular, it is possible to define different trigger conditions (energ tion cuts) in different rapidity regions.
The efficiency of the electron/photon algorithm, as a function of the electron transve different thresholds applied at Level-1, is shown in   From this result, the rate for electron/photon triggers as a func point at which the trigger is 95% efficient, can be computed trons as a function of the E T of the electron (95% point).Level-1 E T (95%) [GeV] Rate

[kHz]
Level  gers are defined by the number of jets (τs) and their transverse energy threshold tion in (η,φ), as well as by a prescaling factor.The global trigger accepts the different multi jet (τ) triggers conditions.
The four highest energy central and forward jets, and central τs in the calorimete occurring in a calorimeter region where an electron is identified are not conside four highest energy central and forward jets and of the four highest energy τs pr for the definition of combined triggers.• Tau jet identified by τpattern shape.

• Various combinations of thresholds possible
• Cuts on jet multiplicities High rates or high cuts.system for distribution to the sub-systems to initiate the readout.ing logical combinations of the trigger data from the Calorimeter The Level-1 Trigger system sorts ranked trigger objects, rather t threshold.This allows all trigger criteria to be applied and varie earlier in the trigger processing.All trigger objects are accompa This allows the Global Trigger to vary thresholds based on the l lows the Global Trigger to require trigger objects to be close or o presence of the trigger object coordinate data in the trigger data, a L1A, permits a quick determination of the regions of interest Three systems are complementary: • gain in efficiency.
• gain in rate.
GMT selection based on candidate p T and quality.

HLT
• Runs on CPU farm (1 ev/processor at a time).Available CPU is a limitation (→ timing).Uses full granularity and resolution.C++.
• Must satisfy physics requirements: inclusive selection, high efficiency.
• Must not required precise knowledge of calibration/run conditions.
• Two strategies: -Fast but not accurate reconstruction -Use minimal amount of precise information.
Both ways used to optimize event rejection speed.Second is preferred.Code as close as possible to offline reconstruction.
• Reconstruction on demand: do not reconstruct until necessary • Regional reconstruction (→) • Partial and conditional track reconstruction (→). the number of seeds which need to be considered increases from 7 to about 44 in going from low to high luminosity, even after taking into account the larger P T cut which is applied at high luminosity.The execution time also depends strongly on the jet energy when using the combinatorial seeding, as shown in Figure 15-73.• Stop tracking when combinatorics is solved and/or precision is sufficient for analysis.
• Stop track reconstruction if track does not match kinematics requirements.

LowLumi
• Time used for track reconstruction increases linearly with number of hits.
-14 -HLT performance M. Konecki   A tool to reject muons from K, π, b, c decays which are often a background for discovery physics.Isolation is based on the E T or p T in a cone around the muon.
• Calorimeter Isolation E T in calorimeter towers.Can be applied already at L2. Sensitive to pile-up • L1, L2 rate (almost) saturates at very high p T s, • sharp L3 cut and steep rate curves.
p T thresholds at 30 GeV bremsstrahlung.In the endcap a cut on the energy found behind the super-cluster, in the HCAL, ex as a fraction of the super-cluster energy, H/E, is found to give useful additional rejection Figure 15-9 shows, as an example, the E/P distribution for barrel electrons, and for jet backgroun tron candidates in the barrel after selection at Level-2.5 followed by loose track finding seeded w Level-2.5 pixel matches.When the distributions are broken down according to the number of hits ated in the tracks, the width and proportion of events in the tail of the distribution for electrons is f vary (electrons which radiate little have tracks with more hits, and a better measured momentum).tio of signal to background also varies with the number of hits.So increased performance can be o by optimizing the E/P cut as a function of the number of hits in the track.• L2: Reconstruct clusters (full granularity) to match Level-1 candidates.
-18 -HLT performance M. Konecki     • tracking region defined by calorimeter jet, • regional and conditional tracking, • HLT performance close to offline.
Technical Design Report, Volume II 15

Performance and Timing
In this section HLT selections are given for two samples of event verse energies and an inclusive QCD sample.The di-jet sample |η| < 1.4 and 1.4 < |η| < 2.4, corresponding to the central and fo with E T = 50, 100 and 200 GeV were used: for E T = 50 GeV th scattering limiting the performance of the tag, while for E T = 20 high particle density.In the generation of these events, all the pp events with jets within the η and E T range in question were sele with the PYCELL routine of PYTHIA, choosing a cell size simila the QCD sample, events generated by PYTHIA 6.152 were reta of about 50,000 events in each P T hard bin were analysed.
The performance of the selection is given in terms of the efficien u-jets, as well as execution time.

Results on Inclusive Jet Samples
Tracks are reconstructed using a regional approach as explained gorithms, with a cut on P T of 1 GeV/c (2 GeV/c) at low (high) lu along the track is 7.The primary vertex is reconstructed using the only exception being that the default cut on pixel lines is 1 GeV/ The b-tag algorithm is the Track Counting method described in S it has two tracks exceeding a threshold on impact parameter sign least three (two) pixel hits for the full (staged) pixel detector con

Conclusions
Pixel detector allows for standalone global (or regional) track reconstruction with good space accuracy.Input to PV finding algorithms.

Figure F- 1
Figure F-1 Overview of the Level-1 Trigger system.
Figure F-3.Also shown in Figure F as function of pseudorapidity for electrons with P T =35 GeV/c.

Figure F- 3
Figure F-3 The efficiency of the Level-1 Trigger for single electrons as a function of the

Figure F- 5
Figure F-5 The rate of the single electron/photon Level-1 Trigge

FigureFFigure F- 9
Figure F-8 shows the Level-1 rates for jet triggers at low and high luminosity.A est, which is also useful in making comparisons between different algorithms a the rate as a function of the generator-level jet E T by plotting the rate for the gives 95% efficiency for the generator-level jet E T .This is shown in FigureF-9 Figure F-16 Efficiency of the Level-1 Muon Trigger to identify single muons from W decays at high luminosity as a function of η.

Figure F- 17
Figure F-17 Level-1 muon trigger rate as a function of P T threshold for low (a) and high (b) luminosity.

R
e.g.DIGI to RHITs) each detector fully • then link detectors • then make physics objects GLOBAL: Reconstruct raw data detector by detector, link detectors to make objects.Needed when no seed given.Also: global tracking, E T , Missing E T , "other side of lepton" REGIONAL: Reconstruct data only where it is needed.Slices of appropriate size.Need to know where to start reconstruction (seeds from Level-1, Level-2).example: Track in the region of interest defined by a jet.Typical cone size: ∆R = 0.2 − 0.5 Trigger/DAQ project Data Acquisition and High-Level Trigger Technical Design Report, Volume II 15 Physics Object Selection and HLT Performance

Figure 15 -
Figure 15-72 Execution time, at low luminosity, as a function of the number of track hits used for 100 GeV di-jet events (left bb, right uu).The three components shown with different colours are described in the text.No. of hits

•-
Figure15-20 The HLT single-muon trigger rates as a function of the P T threshold for (a) low luminosity and (b) high luminosity.The rates are shown separately for Level-1, Level-2, and Level-3, with and without isolation applied at Levels 2 and 3.The rate generated in the simulation is also shown.

Figure 15 -
Figure15-21 Contributions to the Level-3 trigger rate at high luminosity from all sources of muons (a) before and (b) after all isolation criteria have been applied.
Figure15-24 Combined single and di-muon trigger rates as a function of both the symmetric di-muon P T

uu
Road along φ -in narrow η-window around seed u Collect all sub-clusters in road → "super-cluster" Very fast, large rejection with high efficiency (>15 for ε=95%) Ł Before most material ⇒ before most bremsstrahlung, and before most conversions Ł Number of potential hits is 3: demanding ≥ 2 hits quite efficient Full pixel system Staged option CMS: The Trigger/DAQ project Data Acquisition and High-Leve Technical Design Report, Volume II 15 Physics Object Selection and HLT Perf

BEAUTY 2003 ,
Figure15-39 Rates for 1, 3, and 4-jet triggers as a function of calibra for the cut that gives 95% efficiency for the generator-E T shown on t Right: high luminosity.

Figure 15 -
Figure 15-40The rates for 1-jet events, and the incremental 2-jet a 350 GeV, at low luminosity (left), and a 1-jet threshold of 650 GeV at h Narrow jet surrounded by isolation region.•L2.5/L3: Pixel reconstruction or Full Tracker reconstruction.Check for leading track in the calorimeter defined matching cone.No tracks in bigger isolation cone.

•
algorithm(s) rely on large value of the b-hadron proper time (large impact parameter comparing to tracks from u,d jets),

Figure 15 -Figure 4 :
Figure15-68 shows the performance of the algorithm for 100 Ge low and high luminosity.The difference in performance at low an cut on the track P T and the increased pixel readout inefficiency at

Figure 5 :
Figure 5: Resolution of the longitudinal impact point from the helix parametrization, as a function of η and for for p T values 1, 10 and 100 GeV/c.

Table 15 -
13 Jet rate summary table.The table gives the generato on the reconstructed jet E T gives 95% efficiency for this generatorrate of 1 kHz (Level-1) and 1 Hz (HLT).The actual value of the cu efficiency points is given in parentheses.

Table 2 :
Tagged and closest PV-finding efficiencies of the divisive method, for different samples of events at high and low luminosity.