1. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 1 Overview of the CMS Trigger and Plans for LHC Start-up Overview of this talk: Trigger Challenges at LHC and Goals The CMS Trigger System Pilot run Triggers (2007) Physics Triggers (2008)

2. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 2 Minimum Bias Events At full LHC Luminosity we have 22 events superimposed on any discovery signal. First Level Event Selection requires considerable sophistication to limit the enormous data rate. Typical event size: 1-2 Mbytes. 22 70 mb deep inelastic component

3. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 3 Higgs -> 4  Trigger Challenge at LHC We want to select this type of event (for example Higgs to 4 muons) which are superimposed by this…… +30 MinBias

4. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 4 Challenge 2: Pileup In-time pile up: Same crossing different interactions Out-of-time pile up: Tails from previous event New events come every 25 nsec  7.5 m radial separation. Out-of-time pile up: Due to events from different crossings. Need a to identify the bunch crossing that a given event comes from. P.Sphicas

5. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 5 Trigger Goals at LHC At LHC we want to select events that have: (1) Isolated leptons and photons, (2)  -, central- and forward-jets (3) Events with high E T (4) Events with missing E T. The QCD-  are orders of magnitude larger than any exotic channel . QCD events must be rejected early in the DAQ chain and selecting them using high E T cuts in the trigger will simply not work.  Need to select events at the 1:10 11 level with almost no dead-time. HLT must then be able to run full blown reconstruction software and selection filters Indicative event rates:  Inelastic: 10 9 Hz; (2) W  l : 100 Hz  t-tbar:10Hz (4) H(100 Gev): 0.1 Hz  H(500 GeV): 0.01 Hz

6. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 6 The CMS Trigger System 40 MHz input 100 KHz FLT rate 3.2  sec Latency 100 Hz written at the output Event Size 1-2 Mbytes The requirements on the Level-1 Trigger are demanding. Level-1 Trigger: Custom made hardware processor. High Level Trigger: PC Farm using reconstruction software and event filters similar to the offline analysis.

7. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 7 CMS Trigger and DAQ The First Level decision is distributed to the Front-end as well as the readout units. Front-end and readout buffers take care of Poisson fluctuations in the trigger rate. Hand-shaking using back-pressure guarantees synchronization Detector Frontend Computing Services Readout Systems Filter Systems Event Manager Builder Networks Level-1 Trigger Run Control

8. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 8 The CMS L1 Calorimeter Trigger FE: Front End P: Pipeline RCT: Regional Calorimeter Trigger GCT: Global Calorimeter Trigger GT: Global Trigger Y/N TPG RCT GT GCT P FE  Detector data stored in Front End Pipelines.  Trigger decision derived from Trigger Primitives generated on the detector.  Regional Triggers search for Isolated e/  and  and compute the transverse, missing energy of the event.  Event Selection Algorithms run on the Global Triggers 128x25ns=3.2 µsec later i.e. 128 bunch- crossings latency

9. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 9 Level-1 Strategy Selecting events using physics filters at the High Level Trigger level (HLT CPU farms) will not do. The rate must be cut earlier before the HLT is overwhelmed by MHz of background QCD jet events. It follows that the first level of selection, the First Level Trigger, should include algorithms of considerable sophistication which can find Isolated Electrons, Jets and detect specific event topologies. This is a challenging task because we only have 15x25 ns = 375 ns to accomplish it for all sub-triggers. Jets take longer: 24x25 nsec = 600 nsec; which is many orders of magnitude faster than offline. An example of this is CMS Global Calorimeter Trigger (GCT)

10. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 10 CMS Level 1 Trigger Components Level-1 Decision is based on Calorimeter and Muon information (same actually in ATLAS) 128 CC 24 CC

11. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 11 CMS L1 Latencey Budget Total Latency = 128 Bx or 3.2  sec

12. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 12 The CMS GCT Task Jet Triggers: Central, Tau and Forward jet finding and sorting. Jet Counters: Count Jets in 12 different regions of the detector or 12 different thresholds within the detector. Electron/  triggers: Select and Sort the e/  candidates from Regional Calorimeter Trigger Total Transverse, Total Missing Transverse and Total Jet Transverse Energy calculation Receive the Muon data and send them to the Global Muon Trigger. Luminosity Monitoring and readout all the RCT and GCT data for every L1A.

13. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 13 Jet Finders: A Summary Particles strike the detectors and deposit their energy in the calorimeters. Energy deposits in the calorimeters need to be recombined to reconstruct the transverse energy and direction of the original parton. This is done using tools that are called Jet finders.  jet =(-1)ln(tan(  jet /2)) 

14. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 14 Cone Jet Finders Searches for high transverse energy seeds and a cone in the  -  space is drawn around each seed. Energy depositions within a cone are combined and the Et weighted  is calculated: The new cone is drawn and the process is repeated until the cone transverse energy does not change R0R0

15. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 15 Example Algorithms  Electron (Hit Tower + Max) –2-tower  ET + Hit tower H/E –Hit tower 2x5-crystal strips >90% ET in 5x5 (Fine Grain)  Isolated Electron (3x3 Tower) –Quiet neighbors: all towers pass Fine Grain & H/E –One group of 5 EM ET < Thr.  Jet or  E T –12x12 trig. tower  E T sliding in 4x4 steps w/central 4x4 E T > others   : isolated narrow energy deposits –Energy spread outside  veto pattern sets veto –Jet   if all 9 4x4 region  vetoes off Electrons/photon finderJet Finder

16. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 16 The GCT Design Concentrator Card (1/1) Leaf Card (1/8) Wheel Card (1/2)

17. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 17 The GCT Design 63 Source cards 8 Leaf cards 2 Wheel cards 1 Concentrator 31 Source Cards 32 Source Cards 3 Jets Leafs Wheel Concentrator e/  Leafs GTI

18. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 18 CMS GCT Cards

19. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 19 The Leaf Card (e ±, Jets, E T ) Main processing devices: Xilinx Virtex II Pro P70 32 x 5 Gbit/sec Links with Serializers/Deserializers Each serves 1/6 of the detector in Jet finding mode. Virtex-II Pro-P70 3x12 Channel 5 Gbit/s Optical Links (eventually)

20. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 20 Data Sharing Scheme Each Jet Leaf Card Serves 3 Regional calorimeter crates or 1/3 of half Barrel calorimeter (forward calorimeters have been included as edges of the barrel). Each Leaf Searches for Jets using a 3x3 region sliding window. Each Leaf has access to boundary data from neighbours via data duplication at the input of each Leaf η-η- η+η+

21. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 21 CMS GCT Status 63 Source cards 8 Leaf cards 2 Wheel cards 1 Concentrator 31 Source Cards 32 Source Cards 3 Jets Leafs Wheel Concentrator e/  Leafs By March 07 By March 06 GTI

22. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 22 Trigger Commissioning and Testing without Beam : Patterns Tests Install, integrate trigger chain and connect TTC system. Propagate patters from the Trigger front end all the way to HLT and DAQ. GCT is given here as an example but other systems will perform similar tests, GCT: Electron Patterns Tests (March 07) (1) The Source Cards will be loaded with events containing 4 electrons in various parts of the detector. (2) Empty crossings will be loaded in between the electron events. Each Source Card can store half and orbit worth of data (~1500 thousand crossings) which can be either empty or test events. (3) The data will be propagated from the Source Cards via the optical links, to the two electron Leaf Cards and from there to the concentrator all the way to the Global Trigger and also to the DAQ. (4) The data will also be processed by the GCT emulator and the results of the emulator and the hardware will be compared. Goals: (a) Exercise and validate a given trigger path. (b) Establish synchronization: 4 electrons should arrive at GT at the correct crossings with the correct energy, rapidity and phi. (c) Establish agreement between software and firmware

23. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 23 Trigger Commissioning and Testing without Beam : Cosmic Ray Tests Take Cosmic Ray (CR) runs. Trigger using the muon detectors (RPC,DT,CSC) However be aware that CR do not come synchronously with the clock and do not necessarily go through the interaction point where the muon systems are optimized to trigger. Raw rate estimated ~ 1.8 KHz for muon momentum above 10 GeV. This should decrease a lot after cuts on timing and muon direction are folded in. Goals: (a) Exercise and validate the data taking system. (b) Establish coarse synchronization. (c) Start aligning the detectors. Almost no Level-1 cuts; HLT runs Level-1 simulation to validate the Level-1 trigger; Muon reconstruction at HLT but no momentum cuts.

24. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 24 Pilot Run in 2007 (900 GeV)

25. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 25 900 GeV Beam Settings kbkb 43 156 i b (10 10 )24410  * (m) 11 intensity per beam8.6 10 11 1.7 10 12 6.2 10 12 1.6 10 13 beam energy (MJ).06.12.451.1 Luminosity (cm -2 s -1 ) 2 10 28 7.2 10 28 2.6 10 29 1.6 10 30 event rate (kHz)0.42.810.364 W rate (per 24h)0.531170 Z rate (per 24h)0.050.31.17 1.Assuming 450GeV inelastic cross section 40 mb 2.Assuming 450GeV cross section W → lν 1 nb 3.Assuming 450GeV cross section Z → ll 100 pb

26. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 26 Particle Distributions: Particles go forward and have energy below 1 GeV. Need to be able to Trigger forward at low energy. Obviously you do not want a transverse energy trigger. Will not be better at 900 GeV

27. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 27 Trigger for Pilot Run 2007 CMS Trigger Mode of Operation: Ideas on how to do this: (1) Random Level-1 triggers at 1% level. (2) CMS will be using the OPAL scintillators mounted at the front face of the Hadron Forward Calorimeter (HF). (3) Energy/Et over threshold from the first 2 rings around the beams pipe (both sides) in coincidence. (4) Feature Bits from the forward regions will be used to count trigger towers over threshold. L1T identifies collisions and accepts all events HLT verifies L1T bits, and stream events to calibration, e, , jets.. In other words we need a Beam Bias Trigger just to ‘see’ beams.

28. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 28 Beam Hallo Trigger (2007)

29. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 29 Scintillator Trigger 2007 To be installed at the front faces of HF Useful for : (a) Commissioning (b) Calibration (may be) (c) Alignment OPAL Scintillators

30. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 30 Beam Pipe Rings Trigger 2007 Simple Activity Triggers Useful for: (a) Commissioning (b) Calibration (may be) (c) Alignment OR AND GCT Ring-Energy Over Threshold GT Logic and Trigger Decision GCT can compute the energy or transverse energy in rings around the beam pipe form both sides of the calorimeter; energy is better. Global Trigger can set threshold on energy or transverse energy. Forward and Rear in Coincidence

31. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 31 Feature Bits Trigger 2007 Feature bits are derived from the energy of a trigger tower after applying programmable thresholds. These bits end up also on GCT along with the jet data. GCT can count number of towers over threshold around the beam and place cuts such as N>10 on both calorimeters. It is obvious from the second plot that Et will not do but we need energy.

32. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 32 Pilot Run 2007 HLT Minimum selection at Level-1. Validate Level-1 Triggers using the Level-1 emulators running in HLT. Migrate algorithms to Level-1 as soon as they are understood. Main rate reduction at HLT. CMS (CSC): Halo Muons for alignment. We should be able to time the detectors. Validate detector and data taking concepts A course alignment will be possible.

33. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 33 Physics Event rates in2007

34. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 34 Expectations for 2007 Pilot Run To gain the first experience with LHC beams validate as much as possible the detectors and prepare for the 2008 Physics run. Some calibration studies may be possible from phi-symmetry However, we should also see :

35. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 35 Staged commissioning plan for protons@7TeV III No beam Beam Shutdown Machine checkout 7TeV Beam setup 25ns ops I Install Phase II and MKB Stage I IIIII No beamBeam Hardware commissioning 7TeV Machine checkout 7TeV Beam commissioning 7TeV 43 bunch operation 75ns ops25ns ops IShutdown 2008 2009 Widely varying conditions in 2008 Steady state operation in 2009 From Mike Lamont talk at CMS week

36. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 36 Sub-phaseBunchesBun. Int.beta*LuminosityTimeInt lumi First Collisions1 x 14 x 10 10 17 m1.6 x 10 28 12 hours0.6 nb -1 Repeat ramp - same conditions----2 days @ 50%1.2 nb -1 Multi-bunch at injection & through ramp - collimation ----2 days- Physics12 x 123 x 10 10 17 m1.1 x 10 29 2 days @ 50% in physics6 nb -1 Physics43 x 433 x 10 10 17 m4.0 x 10 29 2 days @ 50% in physics30 nb -1 Commission squeeze – single beam then two beams, IR1, IR5 ----2 days- Measurements squeezed----1 day- Physics43 x 433 x 10 10 10 m7 x 10 29 3 days - 6 hr t.a. - 70% eff.75 nb -1 Commission squeeze to 2m collimation etc. ----3 days- Physics43 x 433 x 10 10 2 m3.4 x 10 30 3 days - 6 hr t.a. - 70% eff.0.36 pb -1 Commission 156 x 156----1 day Physics156 x 1562 x 10 10 2 m5.5 x 10 30 2 days - 6 hr t.a. - 70% eff.0.39 pb -1 Physics156 x 1563 x 10 10 2 m1.2 x 10 31 5 days - 5 hr t.a. - 70% eff.2.3 pb -1 28 days total 3.1 pb -1 Pilot physics Month in 2008 Rapidly changing conditions, with collision rate below 50kHz till 156x156 From Mike Lamont talk at CMS week

37. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 37 Plans for 2008 Physics Run At 10 31 commissioning of the LHC algorithms can be tried. Algorithm cuts will be relaxed. Level-1 Rate will be up to 50 KHz. Redundancy between triggers will be used to compute the trigger efficiency using data. 200 Hz ‘on tape’ Emphasis: To understand the Trigger and the detector at LHC running conditions.

38. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 38 Level-1 Trigger Arsenal Minimum Bias Hadron Calorimeter Feature bits Program HB, HE & HF feature bits to ID towers with energy greater than noise HF E T rings Implemented on GCT - but to get minbias efficiently one needs to use HF E rather than E T Beam Scintillation counters Any TOTEM elements available (doubt it..) Normal L1 triggers Can operate eGamma, jet, … triggers with low thresholds (above noise) and muons with no threshold (any muon segment found) -- & no isolation

39. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 39 Electron Trigger Data vs Trigger Emulator I

40. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 40 Electron Triggers vs Trigger Emulator II

41. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 41 Electron Trigger Data vs Trigger Emulator III

42. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 42 Lowest Nominal L1 Trigger Thresholds Electron/  trigger –A trigger tower pair (50 crystals) over threshold - so 5  above noise (40 MeV) implies about 2 GeV minimum –We will use non-isolated e/  path Jet trigger –A jet is composed of 288 trigger towers with nominal noise floor of ~250 MeV per tower which implies a minimum threshold of 10 GeV if we stay above 3  Muon trigger –Muon will not make it until it gets to 3 GeV –Accept poor quality and possibly when any segment is seen in the DTTF or CSCTF or RPC

43. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 43 DAQ Configurations From Sergio Cittolin, Monday Plenary

44. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 44 2008: Startup Trigger Take all minimum bias identified at L1T to HLT There will be sufficient bandwidth 20-50 kHz Validate L1T and run simple HLT algorithms (1) HLT algorithms could be calorimeter and muon based Minimal use of tracking Apply thresholds (none applied at L1) Stream data by trigger type (2) Calibration triggers ECAL:  0 Jets:  + Jet Tracks: J/  →  isolated   (3) Prescale minbias as needed Output bandwidth limit 1 GB/s Rate limit 500-1000 Hz full events, 1-2 MB/events

45. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 45 Rates at 10 31-33 cm –2 s –1 10 33 10 32 10 31 10 8 10 7 10 6 10 5 10 4 10 3 10 2 10 1 10 0 10 1 10 0 10 -1 We cannot trigger on all minbias Jet and soft lepton triggers need to be operational at 10 31 cm –2 s –1 Isolated electron triggers also need to be operational at 10 32 cm –2 s –1 All triggers need to be operational at 10 33 cm –2 s –1 Assume in 2008: L1T Out < 5x10 4 Hz HLT Out < 1.5x10 2 Hz

46. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 46 2008 Operations @ 75ns, 25ns For luminosities above 10 31 cm –2 s –1 we need to set thresholds at L1 and refine object ID at HLT 75ns operation possible till we see a luminosity of 10 33 cm –2 s –1 Average of 5 interactions per crossing at the peak. Only in-time pileup relevant Going to 25ns - 1 operation, i.e. at 33% bunch intensity, keeps the luminosity about the same but pileup goes down: Now about 1-2 interactions per crossing Pileup plays a less significant role

47. UCL Seminar, 16/2/07Costas Foudas, Imperial College London 47 CMS Trigger Overview CMS is gearing up for the first data. Preparations of almost two decades come to a conclusion and I am sure it will be a very exciting time. However, the trigger and DAQ systems at LHC are orders of magnitude more complicated than before but also more sophisticated producing samples of purity not seen before. Understanding the first samples will not be easy, it will take time and requires a methodical and systematic approach. But you can be sure that he who understands his detector first will be closer to discovery using the data after 2008.