1. J. Leonard, U. Wisconsin 1 Commissioning the Trigger of the CMS Experiment at the CERN Large Hadron Collider Jessica L. Leonard Real-Time Conference Lisbon, May 2010

2. J. Leonard, U. Wisconsin 2 CMS Commissioning: 2009 and 2010 Collisions Collision data taken at 900 GeV, 2.36 TeV, and 7 TeV Currently ~9.8 nb -1 of 7 TeV collision data recorded at CMS – Increased by factor of 3 in last 5 days – 93% efficient at data- taking – 99% of detector channels operational

3. J. Leonard, U. Wisconsin 3 CMS Trigger Principles LHC collides two beams of proton bunches at 40 MHz Event: two protons interact → produce end-product particles → end up in detector Don't have resources to record all events We want to keep “interesting” events: new physics But most events are types we've seen many, many times What are “interesting” events? Events with: High-energy particles Isolated particles “Missing energy” Trigger: system to quickly decide which events are potentially interesting based on signatures

4. J. Leonard, U. Wisconsin 4 Particle Signatures Reduces event rate from 40 MHz (collision rate) to 100 kHz Reconstructs e/g, jet, energy sum, and muon objects using custom electronics Muon systems Hadronic calorimeter Electromagnetic calorimeter Tracker

5. J. Leonard, U. Wisconsin 5 Data Flow Through Trigger 40 MHz ~100 kHz ~200 Hz Level-1 Trigger (L1) Custom electronics (ASIC, FPGA) Uses simplified detector information Electromagnetic and hadronic calorimeters Three muon systems Quick! Few  s High-Level Trigger (HLT) Computer farm Uses more detailed event data Regional unpacking of detector readout Can afford to be slower (lower rate) 10's of ms

6. J. Leonard, U. Wisconsin 6 Trigger Menus Single criterion for “interesting event”: trigger path When an object or combination of objects fulfills requirements specified by trigger path, that path “fires”: event information passed along for further processing Example: HLT_Mu5: requires a muon object with energy greater than 5 GeV Trigger menu Set of criteria for trigger-worthy events Prescales: “pass” rates reduced by some factor if rate is too high L1, HLT each has own trigger menu HLT criteria more complex than L1 An L1 “pass” for a given criterion causes related HLT paths to be run on that event

7. J. Leonard, U. Wisconsin 7 BCM 1 BCM 2 HF BS C HF Trigger Menus for Startup Zero-bias (any proton event) Beam pickup (BPTX) Minimum-bias (event with detector activity > noise) Beam scintillator counter (BSC) Detector activity triggers Make best use of low luminosity (low event frequency)

8. J. Leonard, U. Wisconsin 8 Trigger Commissioning: Trigger Menu Evolution Gradually enabled physics-object triggers electron/photon, muon, jet triggers L1 triggers “unmasked” after comparison to trigger simulation, studying rates (“accept event” signal enabled) HLT algorithms enabled after running offline on data to study time performance Minbias triggers will start getting prescaled (higher luminosity = higher event rate) Run 132440 Run 135993

9. J. Leonard, U. Wisconsin 9 L1 Commissioning: Unexpected Effects Physics-related signals in calorimeter Particles interacting with photo-diode electronics cause extraneous signals Developing algorithms to deal with and correct for it at trigger level Periodic spikes in trigger rate from resistive plate chamber (RPC) muon system Traced to specific condition: CMS magnet and cavern lights on at the same time Solution: turn off cavern lights!

10. J. Leonard, U. Wisconsin 10 L1 Commissioning: Timing Particles take longer to get to outer parts of detector Cables between detector parts and trigger have different lengths All event information needs to get combined correctly! Bunch crossing every 25 ns Particles travel 7.5 m in 25 ns Time-of-flight of order of bunch crossing interval

11. J. Leonard, U. Wisconsin 11 L1 Commissioning: Timing Experience Timing scan Scan a range of timing delays Find best alignment between subsystem trigger signal and min-bias trigger signal Cathode strip chamber (CSC) muon system successfully timed in triggers

12. J. Leonard, U. Wisconsin 12 L1 Bunch Crossing Identification Sample of min-bias events Trigger by BSC coincidence Fraction of candidates that are in time with bunch crossing (BPTX trigger) Plotted as function of L1-assigned E T Denominator is number of L1 candidates with +/- 2 bunch crossings of BPTX trigger Noise pollutes efficiency at low E T

13. J. Leonard, U. Wisconsin 13 L1 Commissioning: e/  Object Trigger Efficiency How many offline reconstructed electrons/photons are matched to electron/photon trigger objects? Trigger efficiency Number of objects found in L1 >= 2 GeV divided by number found in event reconstruction Use all reconstructed objects Includes inefficiency from masked channels, out-of-time triggers Efficiency at plateau very good

14. J. Leonard, U. Wisconsin 14 HLT Commissioning: Event Rates Rates well-understood Predicted by running algorithms on simulated data Actual rates agree well with prediction Preparing for higher luminosity Current menu: 1e28 cm -2 s -1 Menus already developed for 1e29, 2e29, 4e29... Reoptimization of 1e31 menu ongoing

15. J. Leonard, U. Wisconsin 15 Conclusions Useful trigger commissioning experience gained Level-1 Trigger performing well in collision data- taking, based on timing studies and efficiency curves High-Level Trigger running smoothly, moving from minimum-bias triggers to physics object triggers Trigger configuration will continue to evolve with the changing luminosity We look forward to more data!

16. J. Leonard, U. Wisconsin 16 Backup Slides

17. J. Leonard, U. Wisconsin 17 Level-1 Trigger 40 MHz → 100 kHz ASIC/FPGA algorithms use simplified detector information to reconstruct physics objects electron/photo n jet energy sum muon

18. J. Leonard, U. Wisconsin 18 High-Level Trigger 100 kHz → 100 Hz Computer farm combines detailed detector information Reconstructs more complex event information than L1 Algorithms optimized for fast performance – Regional unpacking of data

19. J. Leonard, U. Wisconsin 19 L1 Subdetector Synchronization CSC trigger timing – 99.3% of triggers on time (0.2% early, 0.5% late) – Will improve with more statistics and analysis RPC trigger timing – 27.3% of triggers late before corrections, improves to 1.2% after – [What are these corrections?]

20. J. Leonard, U. Wisconsin 20 L1 Trigger Synchronization L1 trigger bit timing alignment stable – Triggers fire on time with respect to bunch crossing

21. J. Leonard, U. Wisconsin 21 L1 Commissioning: e/  Object Trigger Efficiency How many offline reconstructed objects are matched to 2-GeV electron/photon trigger object? Trigger efficiency Number of objects found in L1 divided by number found in event reconstruction Exclude out-of-time triggers Exclude masked channels Require simple (small) reconstructed objects