The trigger system is an essential component of any collider experiment as it is responsible for deciding whether or not to keep an event from a given bunch-crossing interaction for later study. During Run 2 (2015 to 2018) of the Large Hadron Collider (LHC), the trigger system of the ATLAS experiment , shown in Figure 1, operated efficiently at instantaneous luminosities of up to 2.0 × 1034 cm−2 s−1 and primarily at centre-of-mass energies, √s, of 13 TeV, and at higher than designed number of proton–proton interactions per bunch-crossing (pile-up).
Figure 1: The ATLAS TDAQ system in Run 2 with emphasis on the components relevant for triggering (top). The L1 Topological trigger active for physics data taking since 2016 (bottom).
The ATLAS trigger system is composed of two levels: a first level trigger (L1), composed of custom made hardware, processes the input signals from the calorimeter and muon spectrometer systems within microseconds, and reduces the rate from 40 MHz to 100 KHz; a High Level trigger (HLT) farm of commodity computers utilises high granularity signals from the calorimeter and muon spectrometer, as well as from the tracking system, and runs software algorithms to reduce within hundreds of milliseconds the output rate down to an average of 1 KHz. The HLT processes mostly data from the detector regions the L1 identified as interesting; nonetheless several triggers utilising full event information are now part of the set of trigger conditions (trigger menu) evaluated to decide wether a collision event should be kept or not.
Changes to the ATLAS trigger system for Run-2
During the first long shutdown (LS1) between LHC Run 1 and Run 2 operations, the ATLAS trigger system was improved in several parts with respect to Run-1, to withstand higher input rates and pile-up.
The L1 calorimeter was upgraded with a new FPGA-based multi-chip module (nMCM) which support the use of digital autocorrelation Finite Impulse Response (FIR) filters and the implementation of a dynamic, bunch-by-bunch pedestal correction. This allows to apply bunch-by-bunch pedestal subtraction compensating for the increased trigger rates at the beginning of a bunch train caused by the interplay of in-time and out-of-time pile-up coupled with the Liquid Argon calorimeter signal pulse shape, and linearises the L1 trigger rate as a function of the instantaneous luminosity.
In the L1 muon system, additional coincidences of muon spectrometer outer chambers in the endcap regions with Tile calorimeter or inner small wheel allowed to reduce the rate of fake muons.
A new topological trigger (L1Topo) consisting of two FPGA-based (Field- Programmable Gate Arrays) processor modules was added. The modules are programmed to perform selections based on geometric or kinematic association between trigger objects received from the L1 Calorimeter or L1 Muon systems, and allowed to keep the trigger rate under control without dramatically increasing the energy threshold requirements at L1 trigger.
The Muon-to-CTP interface (MUCPTI) and the CTP were upgraded to provide inputs to and receive inputs from L1Topo, and the CTP supports up to four independent complex dead-time settings operating simultaneously, to address various detector-specific time requirements. In addition, the number of L1 trigger selections (512) and bunch-group selections (16) were doubled compared to Run 1.
During Run-1 the HLT was split into a Level-2 and Event Filter processing step, and in Run-2 these were merged into a unique HLT step, which speeded processing data up considerably. At the HLT a two stage tracking approach was introduced to speed up track reconstruction and improve efficiency for several tau, electron, b-jet triggers. The inclusion of out-of-time pile-up corrections at HLT allowed to improve the resolution of calorimeter reconstructed objects for electron, tau, jet and b-jet triggers.
The ATLAS trigger system in Run-2 allowed the efficient and stable collection of a rich data sample for physics analyses, using different streams: a Main stream promptly reconstructed at Tier0, a B-physics and Light States stream delayed in reconstruction at Tier0 or on the grid, and a Trigger Level Analysis stream saving only trigger reconstructed data at a much higher rate than what possible for the Main stream triggers. In addition, the express stream records events at a low rate for data quality monitoring; other minor streams with physics applications, such as zero-bias and background events, are also recorded. A typical stream rate profile during a run for 2018 data taking is shown in Figure 2.
Figure 2: Trigger stream rates as a function of time in a fill taken in September 2018 with a peak luminosity of L = 2.0 x 1034 cm-2s-1 and a peak average number of interactions per crossing of <μ>=56.
The ATLAS Trigger Performance
The performance of the ATLAS trigger menu is measured in data, selecting an orthogonal sample of collision events rich in the presence of di-jets, Z, W or top anti-top events, and determining the efficiency of each trigger to select offline reconstructed objects for searches or measurements. The Missing Transverse Energy trigger reconstruction based on the “pufit” method, adopted for Run-2, is shown in Figure 3. Such trigger is the key for most searches for new physics in ATLAS, both in SuperSymmetric and in Exotic models. The rate of background (left) as well as the efficiency of detecting signal (right) was extremely stable with respect to the number of pile-up interactions in Run-2, compared to Run-1, allowing very precise measurements with such triggered data.
Figure 3 : Run-2 Missing Transverse Energy trigger cross section stability (left) and trigger efficiency with respect to offline reconstructed quantity (right), versus the number of pile-up interactions.
Several new physics scenarios involve hadronic final states, with one or multiple high energetic jets. During Run-2 specific calibrations including track and calorimeter information were introduced online (Figure 4, left), as well as efficient b-tagging capabilities (Figure 4, right), to improve the efficiency for new physics.
Figure 4 : Jet Trigger efficiency with respect to offline selected jets using different calibration schemes at trigger level (left); B-tagging Trigger efficiency versus number of pile-up interactions (right).
Single lepton triggers are extensively used in ATLAS. Isolation in addition to particle identification are used online to trigger on relatively low energetic leptons, useful for most analyses. Such informations are often the convolution of several input variables and therefore stability of efficiency during a run is a non trivial, albeit highly desired feature. This was achieved in Run-2, as shown in Figure 5.
Figure 5: Trigger efficiency versus number of pile-up interactions for electron trigger (left) and muon trigger (right).
Run 3 and Beyond
The ATLAS trigger system will be upgraded  in the Long Shutdown 2 (LS2) to use higher transverse and longitudinal granularity at the L1 calorimeter system, and new muon detectors in the endcap regions, namely the New Small Wheel and the new RPC station (BIS78), to reduce rates for comparable signal efficiencies at the L1 muon system. The L1Topo will also be replaced with three new boards with improved capabilities. In addition, the HLT software will undergo major changes to become multi-thread safe, and run code very similar to the offline reconstruction. These changes will be more extensive than in LS1, and will require a thorough commissioning throughout 2021.
For the HL-LHC period, the ATLAS trigger and data acquisition system will be designed  to withstand higher input and output rates at L1 and HLT (1 MHz L1 and 10 kHz HLT output rate), benefiting to large extent from the L1 hardware upgrade already available in Run-3.
 The ATLAS Collaboration, Performance of the ATLAS trigger system in 2015, Eur. Phys. J. C (2017) 77:317
 The ATLAS Collaboration, ATLAS TDAQ Phase I Upgrade TDR https://cds.cern.ch/record/1602235/files/ATLAS-TDR-023.pdf
 The ATLAS Collaboration, ATLAS TDAQ Phase II Upgrade TDR https://cds.cern.ch/record/2285584/files/ATLAS-TDR-029.pdf