1. The CMS Trigger System Chris Seez, Imperial College, London IV International Symposium on LHC Physics and Detectors Fermilab, 1 st -3 rd May 2003

2. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 2 Physics Selection at LHC

3. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 3 l Formidable task: u Bunch crossing rate  permanent storage rate for events with size ~1MB u 40MHz  O(10 2 )Hz l CMS design: Level-1 High Level Trigger u Beyond Level-1 there is a High Level Trigger running on a single processor farm The CMS Trigger

4. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 4 High Level Trigger l Advantages of using processor farm for all selection beyond Level-1: u Benefit maximally from evolution of computing technology u Flexibility: no built-in design or architectural limitations — maximum freedom in what data to access and in sophistication of algorithms u Evolution, including response to unforeseen backgrounds u Minimize in-house elements è cost è maintainability

5. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 5 l Modular DAQ: 8 x 12.5kHz units CMS DAQ and Trigger System l Event size: 1MB from ~700 front-end electronics modules l Level-1 decision time: ~3  s — ~1  s actual processing (the rest in transmission delays) l DAQ designed to accept Level-1 rate of 100kHz l HLT designed to output O(10 2 )Hz – rejection of 1000 l ~1000 processor units

6. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 6 Software for simulation and reconstruction u Full GEANT3 simulation of CMS detector u Digitization and reconstruction in C++ code u Many samples digitized at both 2x10 33 and 10 34 u Digitization includes both in- and out-of-time pileup (i.e. min-bias type events in the same and neighbouring bunch-crossings) l Results presented in full in CMS DAQ TDR (Dec 2002) l Results presented here for rates, efficiencies of the complete CMS trigger system (Level-1 + HLT) use event samples comprising ~7M events produced in 2002

7. Level-1 Trigger

8. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 8 Level-1 Trigger l Information from Calorimeters and muon detectors u Electron/photon triggers u Jet and missing E T triggers u Muon triggers l Synchronous, pipelined u Time needed for decision (+its propagation) ≈ 3  s u Bunch crossing time = 25 ns l Algorithms run on coarse local data u Only calorimeter and muon information u Special-purpose hardware (ASICs), but also wide use of FPGAs l Backgrounds are huge u Large rejection factor: is 40MHz (x20 ev/crossing)  100kHz (≈ 8,000) u Rates: steep functions of thresholds

9. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 9 Calorimeter Trigger SORT ASICs EISO u 18 Calorimeter trigger crates è ≈ 4000 Gb/s serial links è 224 inputs/crate è 18 bits/(trigger tower) u 32 towers/card è ASICs: process 8 or 16 towers

10. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 10 Electron/photon Trigger l Electromagnetic trigger based on 3x3 trigger towers u Each tower is 5x5 crystals in ECAL (barrel; varies in end-cap) u Each tower is single readout tower in HCAL Trigger threshold on sum of two towers Cuts put on: - e/h fraction - Fine shape in ECAL (acts as local isolation) - Isolation in both ECAL and HCAL sections

11. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 11 Electron/photon Trigger Response to electrons: Rate (jet background): Top 4 candidates in each category passed to global trigger

12. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 12 Jet and  Triggers l Single, double, triple and quad thresholds possible l Possible also to cut on jet multiplicities Also E T miss,  E T and  E T (jets) triggers Sliding window: -granularity is 4x4 towers = trigger region - jet E T summed in 3x3 regions ,  = 1.04 “  -like” shapes identified for  trigger

13. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 13 Muons l Issue is p T measurement of real muons L = 10 34 cm -2 s -1  < 2.1

14. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 14 Muon trigger Level-1  -trigger info from: u Dedicated trigger detector: RPCs (Resistive plate chambers) è Excellent time resolution u Muon chambers with accurate position resolution è Drift Tubes (DT) in barrel è Cathode Strip Chambers (CSC) in end-caps u Bending in magnetic field  determine p T

15. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 15 Drift tube and CSC trigger Implementation: ASICs for Trigger Primitive Generators FPGAs for Track Finder processors Extrapolation: using look-up tables Track Assembler: link track segment- pairs to tracks, cancel fakes Drift TubesCSC

16. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 16 Level-1 muon global trigger l Information from different detectors combined (RPC, CSC and DT) u Match muon candidates from different systems u Different sub-systems complement one another u Maximize efficiency, minimize rate u Identify 4 “best” muons and pass them on to the Global Trigger 10 34 cm -2 s -1

17. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 17 Level-1 Settings and Rates… l Current CMS plan is for phased installation of DAQ u Startup (L=2x10 33 cm -2 s -1 ): can handle 50kHz u High luminosity (L=10 34 cm -2 s -1 ): can handle 100kHz l Model assumes safety factor of three u To account for simulation uncertainties, and beam conditions… u Startup (L=2x10 33 cm -2 s -1 ): set thresholds for 16kHz u High luminosity (L=10 34 cm -2 s -1 ): set thresholds for 33kHz l Start iteration by allocating the rate equally between:  Electrons/photons;  Muons;  Tau-jets;  Jets and combined triggers l Priority: guarantee discovery physics l Then choose allocation between single and double objects, etc Modular DAQ

18. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 18 Choice of operating point l Example of electrons u Look at efficiency to trigger on Z  ee versus efficiency to trigger on W  e

19. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 19 Level-1 Trigger table (2x10 33 ) TriggerThreshold (GeV or GeV/c) Rate (kHz)Cumulative Rate (kHz) Isolated e/  293.3 Di-e/  171.34.3 Isolated muon142.77.0 Di-muon30.97.9 Single tau-jet862.210.1 Di-tau-jet591.010.9 1-jet, 3-jet, 4-jet177, 86, 703.012.5 Jet*E T miss 88*462.314.3 Electron*jet21*450.815.1 Min-bias0.916.0 TOTAL16.0

20. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 20 Level-1 Trigger table (10 34 ) TriggerThreshold (GeV or GeV/c) Rate (kHz)Cumulative Rate (kHz) Isolated e/  346.5 Di-e/  193.39.4 Isolated muon206.215.6 Di-muon51.717.3 Single tau-jet1015.322.6 Di-tau-jet673.625.0 1-jet, 3-jet, 4-jet250, 110, 953.026.7 Jet*E T miss 113*704.530.4 Electron*jet25*521.331.7 Muon*jet15*400.832.5 Min-bias1.033.5 TOTAL33.5

21. High-Level Trigger

22. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 22 High-Level Trigger l Runs on CPU farm l Code as close as possible to offline reconstruction code u Ease of maintenance u Able to include major improvements in offline reconstruction l Selection must meet CMS physics goals u Output rate to permanent storage limited to O(10 2 )Hz l Reconstruction on demand u Reject as soon as possible u Hence trigger “Levels”: è Level-2: use calorimeter and muon detectors è Level-2.5: also use tracker pixel detectors è Level-3: includes use of full information, including tracker u And “regional reconstruction”: e.g. tracks in a given road or region

23. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 23 HLT regional reconstruction l Slices must be of appropriate size l Need to know where to start reconstruction (seed) u Seeds from Level-1:  e/  triggers   triggers è Jet triggers l Seeds ≈ absent: u Other side of lepton u Global tracking u Global objects (  E T, E T miss ) D e t e c t o r ECAL Pixel L_1 Si L_1 Pixel L_2 HCAL 14 D e t e c t o r ECAL Pixel L_1 Si L_1 Pixel L_2 HCAL Regional rather than Global reconstruction

24. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 24 HLT requirements and operation l Boundary conditions: u Code runs in a single processor, which analyzes one event at a time u HLT has access to full event data (full granularity and resolution) u Only limitations: è CPU time è Output selection rate (~10 2 Hz) è Precision of calibration constants l Main requirements: u Satisfy physics program: high efficiency u Selection must be inclusive (to discover the unpredicted as well) u Must not require precise knowledge of calibration/run conditions u All algorithms/processors must be monitored closely

25. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 25 HLT selection: , , jets and E T miss l Muons u Successive refinement of momentum measurement; + isolation è Level-2: reconstructed in muon system; must have valid extrapolation to collision vertex; + calorimeter isolation è Level-3: reconstructed in inner tracker; + tracker isolation l  -leptons u Level-2: calorimetric reconstruction and isolation è Very narrow jet surrounded by isolation cone u Level-3: tracker isolation l Jets and E t miss u Jet reconstruction with iterative cone algorithm u E T miss reconstruction (vector sum of towers above threshold)

26. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 26 HLT selection: electrons and photons Level-2 Level-3 Level-1 Level-2.5 Photons Threshold cut Isolation Electrons Track reconstruction E/p, matching (  ) cut ECAL reconstruction Threshold cut Pixel matching u Issue is electron reconstruction and rejection u Higher E T threshold on photons u Electron reconstruction è key is recovery of radiated energy u Electron rejection è key tool is pixel detector

27. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 27 Electron selection: Level-2 l “Level-2” electron: u Search for match to Level-1 trigger è Use 1-tower margin around 4x4-tower trigger region u Bremsstrahlung recovery “super-clustering” u Select highest E T cluster l Brem recovery:  Road along  — in narrow  -window around seed u Collect all sub-clusters in road  “super-cluster” basic cluster super-cluster

28. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 28 Electron selection: Level-2.5 l “Level-2.5” selection: use pixel information  Very fast, large rejection with high efficiency (>15 for  =95%) è Before most material  before most bremsstrahlung, and before most conversions è Number of potential hits is 3: demanding  2 hits quite efficient Full pixel system Staged option

29. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 29 Electron selection: Level-3 l “Level-3” selection u Full tracking, loose track-finding (to maintain high efficiency) u Cut on E/p everywhere, plus  Matching in  (barrel) è h/e (endcap) u Optional handle (used for photons): isolation 2x10 33 cm -2 s -1

30. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 30 HLT table l Issues: u Purity of streams is not the same (e.g. electrons vs muons) è Kinematic overlap provides redundancy  To answer the sort of question, when a problem is under investigation in W  e : do we see this in the muons? u Comparisons of unlike things: è How much more bandwidth should go to lower-p T muons than to electrons? è How should one share the bandwidth between jet*E T miss and di- electrons? u Only final guidance is efficiency to all the known channels è While keeping the selection inclusive n This is online: events rejected are lost forever.

31. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 31 HLT Summary: 2x10 33 cm -2 s -1 TriggerThreshold (GeV or GeV/c) Rate (Hz)Cuml. rate (Hz) Inclusive electron2933 Di-electron17134 Inclusive photon80438 Di-photon40, 25543 Inclusive muon192568 Di-muon7472 Inclusive tau-jet86375 Di-tau-jet59176 1-jet * E T miss 180 * 123581 1-jet OR 3-jet OR 4-jet657, 247, 113989 Electron * jet19 * 45290 Inclusive b-jet237595 Calibration etc10105 TOTAL105

32. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 32 HLT performance — signal efficiency l With previous selection cuts ChannelEfficiency (for fiducial objects) H(115 GeV)  77% H(160 GeV)  WW*  2  92% H(150 GeV)  ZZ  4  98% A/H(200 GeV)  2  45% SUSY (~0.5 TeV sparticles)~60% With R P -violation~20% W  e 67% (fid: 60%) WW 69% (fid: 50%) Top  X 72%

33. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 33 l All numbers for a 1 GHz, Intel Pentium-III CPU u Total: 4092 s for 15.1 kHz  271 ms/event è Therefore, a 100 kHz system requires 1.2x10 6 SI95 u Expect improvements, additions. Time completely dominated by muons (GEANE extrapolation) – this will improve u This is “current best estimate”, with ~50% uncertainty. CPU time usage Physics objectCPU time (ms/Level-1) Level-1 rate (kHz) Total CPU time (s) Electrons/photons1604.3688 Muons7103.62556 Taus1303.0390 Jets and E T miss 503.4170 Electron + jet1650.8132 b-jets3000.5150

34. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 34 HLT summary l Today: need ~300 ms on a 1GHz Pentium-III CPU u For 50 kHz, need 15,000 CPUs u Moore’s Law: 2x2x2 times less time (fewer CPUs) in 2007 è Central estimate: 40 ms in 2007, i.e. 2,000 CPUs è Thus, basic estimate of 1,000 dual-CPU boxes in TDR è (Note: not an excess of CPU, e.g. no raw-data handling) u Start-up system of 50kHz (Level-1) and 105 Hz (HLT) can satisfy basic “discovery menu” è Some Standard Model physics left out; intend to do it, at lower luminosity and pre-scales as luminosity drops through fill n Examples: inclusion of B physics (can be done with high efficiency and low CPU cost; limitation is Level-1 bandwidth); details in TDR [see talk by Vitalliano Ciulli]. Also low-mass di-jet resonances. l Single-farm design works

35. C. Seez Imperial College 1st - 3rd May 2003 IV International Symposium on LHC Physics and Detectors 35 Overall Summary l Using a full and detailed simulation of the CMS trigger (Level-1 + HLT) a model trigger table has been developed which: u Meets target rates for Level-1 u and for final output to permanent storage u While maintaining high efficiency for signal events u and wide inclusive selection (open to the unexpected) l The system outlined has huge flexibility l This is only the beginning — there are many challenges ahead l Final tuning will clearly be done with the final event generator: LHC collisions