The ATLAS Trigger System

This paper summarizes the main features, Run-3 performance, and physics impact of the upgraded ATLAS Trigger system, which was enhanced to handle increased luminosity and pile-up conditions while reducing the event rate for offline analysis.

The Gist

Imagine the ATLAS experiment at CERN as a massive, high-speed photography studio. Every second, protons smash into each other 40 million times (40 MHz). Each collision creates a chaotic explosion of particles, generating a flood of data so huge that if we tried to save every single photo, our hard drives would fill up in a split second, and our computers would melt.

The ATLAS Trigger System is the ultra-smart, super-fast editor that sits between the camera and the hard drive. Its job is to look at this flood of 40 million photos per second and decide: "Is this picture worth keeping?" It filters out the boring snapshots and keeps only the most exciting, rare, or scientifically valuable moments.

Here is how the system works, broken down into simple steps:

1. The Two-Stage Filter: The Bouncer and The Art Critic

The system works in two distinct stages, like a two-step security check at a VIP club.

Stage 1: The Level-1 (L1) Trigger – The "Bouncer"

• Speed: This is the hardware bouncer. It's made of custom electronics that are incredibly fast but not very detailed.
• How it works: It takes a quick, blurry look at the collision. It asks simple questions: "Do I see a high-energy jet? A muon (a heavy electron)? Or a lot of missing energy?"
• The Upgrade: In the current "Run 3" era (2022–2026), the crowd is much bigger and messier (more "pile-up," meaning many collisions happening at once). To handle this, the bouncer got a new set of tools:
  • eFEX: A smart scanner that can spot electrons, photons, and tau particles even in a crowded room.
  • jFEX: A jet-spotter that can identify big, messy sprays of particles.
  • gFEX: A global observer that looks at the whole picture to find complex patterns.
• The Result: The bouncer cuts the 40 million events down to about 100,000. It's still a lot, but manageable.

Stage 2: The High Level Trigger (HLT) – The "Art Critic"

• Speed: This is a software program running on a massive farm of 60,000 computer processors. It's slower than the bouncer but much smarter.
• How it works: It takes the 100,000 events the bouncer saved and looks at them with "high-definition" vision. It reconstructs the tracks of particles just like a human scientist would, but in a fraction of a second.
• The Upgrade: The software was completely modernized to run on multiple threads (like having many workers doing tasks simultaneously) and to handle the messy "pile-up" conditions better.
• The Result: The Art Critic makes the final decision, cutting the rate down to about 3,000 events per second. These are the "keepers" that get saved forever for scientists to study later.

2. Sorting the Good Photos into Folders

Once the Art Critic says "Yes, save this," the system doesn't just dump everything into one pile. It sorts the events into different "streams" or folders, depending on what they are:

• Main Stream: The standard, high-quality photos for general physics research.
• Express Stream: A "rush job" folder. These are reconstructed immediately so scientists can get quick feedback on how the experiment is running.
• Delayed Stream: Special events (like rare particle decays) that are saved for later processing when the computers are less busy (like during the LHC's winter breaks).
• Calibration Stream: Quick, low-detail snapshots used just to tune and fix the detectors.
• TLA Stream (Trigger-Level Analysis): A clever trick. Instead of saving the whole high-definition photo (which is huge), this saves a tiny, compressed version (4.5 KB instead of 1.5 MB). This allows the system to save many more interesting events that might otherwise be too big to store.
• Debug Stream: A "trash can" for events where the system glitched, so engineers can fix the software later.

3. Why This Matters

The ATLAS Trigger is the unsung hero of the experiment. Without it, the data flow would be a firehose that drowns everything. Because of these upgrades, the system can handle the chaotic, crowded conditions of the current LHC run (where up to 60 collisions happen at once) without missing the rare, precious signals of new physics.

Looking Ahead:
As we move toward the future "High-Luminosity LHC," where the collisions will be even more intense, the Trigger system is already preparing. It's getting faster hardware, smarter algorithms, and even artificial intelligence to make split-second decisions, ensuring that when the next big discovery happens, the system is ready to catch it.

Technical Summary

Based on the provided paper, here is a detailed technical summary of the ATLAS Trigger System's performance and upgrades during Run 3.

Problem Statement

The ATLAS experiment at the Large Hadron Collider (LHC) faces the challenge of an immense data flow generated by proton-proton collisions occurring every 25 ns at a frequency of 40 MHz. The raw data rate from the subdetectors far exceeds the capacity for full recording and offline storage.

• Specific Challenge: During Run 3 (2022–2026), the LHC operates at a center-of-mass energy of $\sqrt{s} = 13.6$ TeV with significantly increased instantaneous luminosity. This results in a higher average pile-up ($\mu \sim 60$) compared to previous runs ($\mu \sim 34$ in 2018), creating a more complex environment for event selection.
• Goal: The system must reduce the event rate by a factor of approximately 10,000 while retaining interesting physics events with high efficiency and flexibility, ensuring that only the most relevant data is available for reconstruction and analysis.

Methodology

The ATLAS Trigger system employs a multi-level, hierarchical architecture designed to filter events in two distinct stages:

1. Level-1 (L1) Trigger (Hardware-Based)
The L1 system uses custom electronics to perform coarse-granularity identification of physics objects, reducing the rate from 40 MHz to approximately 100 kHz. It consists of four subsystems:

• L1 Calorimeter Trigger (L1Calo): Identifies electromagnetic objects, hadronic taus, and jets, and calculates global quantities like missing transverse energy ($E_T^{miss}$).
  • Upgrades: Introduced finer granularity from Liquid Argon (LAr) readout and new ATCA-based Feature Extractor (FEX) modules:
    • eFEX: For electrons, photons, and taus (sophisticated clustering/isolation).
    • jFEX: For jets and $E_T^{miss}$.
    • gFEX: For global event quantities and complex topologies (e.g., large-radius jets).
• L1 Muon Trigger (L1Muon): Combines signals from Resistive Plate Chambers (RPCs), Thin Gap Chambers (TGCs), and the New Small Wheel (NSW).
  • Upgrades: The NSW (covering $1.3 < |\eta| < 2.7$), composed of MicroMegas and small-strip TGCs, was added to improve fake-muon rejection. It operates in coincidence with the Tile calorimeter.
• L1 Topological Trigger (L1Topo): Performs topological selections (e.g., angular separation $\Delta R$) between objects from L1Calo and L1Muon to enhance background rejection.
• Central Trigger Processor (CTP): Aggregates inputs, applies prescales/deadtime vetoes, and issues the final L1 Accept signal.

2. High Level Trigger (HLT) (Software-Based)
The HLT refines event selection using full detector granularity and algorithms closely mirroring offline versions.

• Architecture: Runs on a computing farm of ~60,000 processor cores.
• Processing: Decisions are typically made within 300 ms.
• Upgrades:
  • Migration to a fully multithreaded framework.
  • Significant speed-up in track reconstruction enabling full-scan operations.
  • Enhanced triggers for pile-up-sensitive and unconventional topologies.
• Buffering: Event information is retained in the Read-Out-system Buffer (ROB) during the decision-making process.

3. Data Streaming Strategy
Accepted events are routed into specific data streams based on their content and purpose:

• Main Stream: Standard physics analyses (~3 kHz output).
• Express Stream: Prompt reconstruction for quick feedback.
• Delayed Streams: For specific physics (e.g., B-physics) processed during LHC shutdowns when Tier-0 capacity is free.
• Calibration Stream: High-rate, reduced-content events for detector calibration.
• TLA (Trigger-Level Analysis) Stream: Reduced event content (~4.5 kB vs. 1.5 MB for main stream) allowing higher save rates (up to 6 kHz in 2022).
• Debug Stream: For events affected by online failures (recovered offline).
• Note: The model is inclusive; a single event can populate multiple streams if it meets the criteria.

Key Contributions

• Hardware Modernization: Implementation of the NSW muon detector and new FEX modules (eFEX, jFEX, gFEX) in the L1Calo system to handle higher granularity and complexity.
• Software Evolution: Complete migration of the HLT to a multithreaded framework and optimization of track reconstruction algorithms to cope with $\mu \sim 60$ pile-up.
• Strategic Data Handling: Introduction of the TLA stream to maximize physics reach by saving more events with reduced storage overhead, and the implementation of an inclusive streaming model.

Results

• Rate Reduction: The system successfully reduced the event rate from 40 MHz to ~100 kHz at L1, and further to ~3 kHz for the main physics stream at HLT.
• Efficiency Improvements:
  • Large-R Jets: The L1 efficiency for large-radius jets in Run 3 shows a sharper turn-on and better plateau efficiency compared to legacy small-R jet triggers.
  • Muon Rejection: The inclusion of the NSW/Tile calorimeter coincidence logic reduced the L1 muon trigger rate by approximately 14 kHz while maintaining high efficiency across the full acceptance.
• Operational Stability: The system maintained high efficiency and flexibility across a wide range of physics signatures despite the challenging high-luminosity and high-pile-up conditions of Run 3.

Significance

The ATLAS Trigger system is critical for the success of the Run 3 physics program. By successfully managing the increased data load and pile-up conditions, it enables:

1. Precision Measurements: Ensuring high-quality data for standard model measurements.
2. New Physics Searches: Maintaining sensitivity to rare or unconventional signatures that might otherwise be lost in high-background environments.
3. Future Readiness: The upgrades and architectural changes (multithreading, ML-ready algorithms) serve as a foundational preparation for the High-Luminosity LHC (HL-LHC) era, where data rates and pile-up will increase even further.