1. ATLAS Trigger GroupPPD Staff Talk 23/5/05 1 Level-2 Trigger Overview When Level-1 accepts an event –Each part of ATLAS sends its data to the Read-Out Buffers (ROB’s) –Level-1 sends Level-2 details of why the event was accepted and where the “Regions of Interest” (RoI) are e.g. a high energy electron candidate The data for each event is ~1.5 MB –About the size of a picture in a digital camera Each event is passed to a Level-2 processor to refine Level-1 decision –Level-2 accesses the data it needs across a network – but only for the RoI’s –Only a few % of the data – reduces network size needed (~2 GB/s = 3 CD’s) –Average Level-2 decision time ~ 10 ms –Level-2 accept rate < 2 kHz (i.e. reject ~98%, but keep the important ones!) 120 GB/s Event Builder ~3 GB/s RoI data 1-2% ~2 GB/s Level-1 accept ROD Level-1 Calo MuTrCh Other detectors Level-2 accept Data requests RoI’s 120 GB/s Level-2 ROB L2Processor (75 kHz) (<2 kHz)

2. ATLAS Trigger GroupPPD Staff Talk 23/5/05 2 Level-2 Software Structure The DataFlow components handle all I/O and communication with run control RAL group previously made major contribution to the L2PU DataFlow code – but activity now nearly all in Algorithms The Selection Code (Algorithms) sees an offline like framework (Gaudi/Athena) Algorithms can be developed/tested in offline environment reading data from files Algorithms can use various off-line components – if they are fast enough The Steering Controller provides the interface between online and offline worlds Level-1 RoI’s Data Flow (L2PU) Steering Controller Algorithms Level-2 decision Request Response Data ROB

3. ATLAS Trigger GroupPPD Staff Talk 23/5/05 3 Level-2 Hardware Level-2 Infrastructure - hardware –Number of Level-2 PCs in final system ~ 500 –~30 in a rack + network switches –Rack mounted – each 1U (1.75 inches) high –Each processor has 2 network connections Data input at GbE (i.e. 10x speed of the PPD LAN) –3 GHz Dual Processor (Xeon) is ~ x3 too slow Will they be fast enough in 2-3 years ? –Cooling is a major problem - ~10 kW/rack Other Tasks of FW –TDAQ Resources –UK Level-2 Project Leader –PPD Learning and Development –Contact for articles in HI-phi

4. ATLAS Trigger GroupPPD Staff Talk 23/5/05 4 The Level-2 (HLT) Group Fred Wickens –Level-2 Infrastructure (esp h/w) TDAQ Resources Julie Kirk –Level-2 b-physics + Track fitting efficiency John Baines –Trigger Selection s/w Level-2 Tracking s/w and b-physics Dmitry Emeliyanov –Level-2 Track fitting and data preparation Bill Scott –Tracking efficiency and b-physics Monika Wielers –(Level-2) Electron/photon selection >

5. ATLAS Trigger GroupPPD Staff Talk 23/5/05 5 Trigger Selection Software John Baines The High Level Trigger (Level-2 and Event Filter) uses software running on a farm of PCs to perform: Data Preparation: Take the raw data from the detectors and prepares it for the next step. Pattern Recognition: Identify clusters of Energy in the Calorimeter, or find patterns of measurements representing a charged particles path in the Tracking detectors (Inner Detector and Muon detector). Form “Trigger Elements”: Identify characteristic features in the events which separate interesting events e.g. containing new physics from uninteresting background events The Level-2 Trigger : Reduces the trigger rate down to about 2000 events per second Works inside regions of the detector identified as interesting by the Level-1 trigger.

6. ATLAS Trigger GroupPPD Staff Talk 23/5/05 6 Data Preparation Part of Semi Conductor tracker (SCT) Wafer Part of Pixel Detector Part of SCT Module (2 wafers) See Dmitry’s Talk One particle travelling through a wafer can cause signals on several adjacent strips or wafers. Group Adjacent Strips or Pixels with charge as Cluster Find the crossing point for SCT strips with charge in the two detectors of a module to form a Space- Point (3-dimensional measurement)

7. ATLAS Trigger GroupPPD Staff Talk 23/5/05 7 RECONSTRUCTED TRACK REGION OF INTEREST HITS IN SemiConductor Tracker Level-1 Trigger finds calorimeter cluster Region Of Interest is a region of the detector centred on the calorimeter cluster position Level-2 Trigger requests data from the Inner Detector in this Region & performs Data Preparation Level-2 algorithms reconstruct track(s) Energy in Calorimeter Hits in Transition Radiation Tracker Hits in Pixel Detector Track Reconstruction See Dmitry’s & Bill’s Talks Decay of Higgs Particle to 4 electrons

8. ATLAS Trigger GroupPPD Staff Talk 23/5/05 8 Hypothesis Algorithms We identify features that are characteristic of interesting events. e.g. One or more isolated high energy electrons could signify a Higgs decay (as shown on previous slide) See Monika’s Talk An electron and positron pair, satisfying certain constraints, could signify a decay of the J/  particle - used to select events for CP violation measurements. Recall Julie’s Talk B d J/  K s e  e      SCT TRT Pixels >

9. ATLAS Trigger GroupPPD Staff Talk 23/5/05 9 Dmitry Emeliyanov Brief Biography –Graduated from Moscow Institute of Physics and Technology (M.Sc. in Applied Physics, 1992) –Ph.D. in Control and Data Processing Systems, 1998 –Worked on tracking and vertexing for the HERA-B experiment (was DESY Fellow, 2000-2004) –Joined PPD ATLAS group in 2004 as a fixed-term contract RA My work at ATLAS PPD group: –Development and optimization of the track reconstruction and fitting algorithms for the Level 2 Trigger –Currently working on timing measurements and optimization of the Level2 Trigger data preparation algorithms

10. ATLAS Trigger GroupPPD Staff Talk 23/5/05 10 Level 2 Tracking in Inner Detector Inner Detector comprises Pixel, Semiconductor (SCT) and Transition Radiation (TRT) trackers The tracking is always RoI-based: done only inside the Region-Of- Interests coming from Level 1 Trigger. Includes two major stages: Data preparation: –raw data from the detector are decoded and converted into a form usable for the reconstruction – hits in Pixel, SCT and TRT –As each SCT module is a pair of single-sided silicon wafers, hits from both sides are combined into the points in 3D space (“spacepoints”) where the particle passed through the detector Track reconstruction: –track finding – combining spacepoints into tracks – “joining the dots” – quite challenging task because the number of “dots” is high – 50-300 –track fitting – determine the particle parameters – direction of flight, curvature of the trajectory in the magnetic field (to determine a momentum)

11. ATLAS Trigger GroupPPD Staff Talk 23/5/05 11 Track finding and fitting A number of software packages have been developed in ATLAS for Level 2 tracking. One of them is IDSCAN: –developed and supported by RAL/UCL collaboration –fast track finding in Pixel and SCT: ~ 0.5 ms per track on single Pentium-4 –track fitting by dedicated algorithm based on the Kalman filtering mathematics, fast implementation takes about 0.1 ms per typical track Recently the IDSCAN has been extended to do track finding in the Transition Radiation Tracker: –tracks found in Pixel/SCT are extrapolated out to the TRT –TRT hits are associated to the tracks using the “probabilistic data association filter” (PDAF) – an algorithm widely used in radar/sonar tracking applications –takes about 1 ms per track to search the whole TRT –calculates the number of transition radiation hits – used to distinguish electrons from pions in the trigger

12. ATLAS Trigger GroupPPD Staff Talk 23/5/05 12 Timing optimizations Level 2 algorithms are implemented in Athena – general ATLAS offline software framework As a result, many offline software components and solutions have been adopted by Level 2 Some of them are not completely adequate to the Level 2 strict timing requirements – optimization/re-development needed Example: data preparation for Pixel and SCT: –due to the offline storage of clusters and spacepoints the overall timing was around 60 ms/RoI – too slow for Level 2 –development and implementation of the alternative caching algorithm reduced this time down to ~ 6 ms/RoI The timing measurements provide an important input for assessment of the Level 2 software and further optimization. >

13. ATLAS Trigger GroupPPD Staff Talk 23/5/05 13 Julie Kirk (Daphne Jackson Fellow) Daphne Jackson fellowships: –“Returning engineers and scientists to work after career breaks.” –Fellowship is for two years, half-time with dual aspect of re-training plus new reasearch. –The lab provides office/computer (“bench fees”), Daphne Jackson Trust cover salary and some extra expenses (e.g. conferences) After working at CERN on OPAL I took a career break for 8 years. In November 2004 I started a Daphne Jackson fellowship working on ATLAS level 2 trigger

14. ATLAS Trigger GroupPPD Staff Talk 23/5/05 14 B-physics triggers Triggering – finding and keeping events of interest (signal) to physicist while discarding the rest (background). ~1 per 100 interactions produce a bb pair. Rates for channels of interest to physicists are usually LOW. Finite cpu resources at level 2 mean that there is not enough CPU time to reconstruct the whole event. Strategy: use a level 1 trigger to indicate the area of the detector (Region of Interest or RoI) in which to do level 2 reconstruction. Example – bb μX B d (J/ψ(ee)K) Use high momentum muon from one b to trigger the event at level1. Look for electromagnetic ROI to seed track reconstruction and find electrons

15. ATLAS Trigger GroupPPD Staff Talk 23/5/05 15 Tracks show simulated particles in the detector. Yellow tracks - electrons from J/ψ Example J/ψ→e + e - event x y 8 m White cone illustrates a level1 EM RoI. At level2 we only look inside the ROI to reconstruct tracks. Saves lots of cpu time by only considering a relatively small portion of the detector.

16. ATLAS Trigger GroupPPD Staff Talk 23/5/05 16 Event selection True e+/e- tracks from J/ψ events Now use the reconstructed tracks to identify signal events. Assuming that they came from the decay of a particle we can combine pairs of reconstructed tracks to form a mass (plots). Make cuts on the mass (shown by arrow) to select signal events and reject most of background. Background events 3 GeV10 GeV Reconstructed J/ψ mass 10 GeV3 GeV

17. ATLAS Trigger GroupPPD Staff Talk 23/5/05 17 Bill Scott Measure HLT Track Reconstruction Efficiencies for Low Energy Electrons using Simulated Data –Tracks Found / Tracks Generated –Two Algorithms: IDSCAN, Si-Track. Using Events that were already Generated –Single electrons: pT = 25 GeV, pT=15 GeV, –Single electrons: pT=10 GeV Using Events Generated (by me) at RAL –Generate, Simulate and Digitise on CSF at RAL: –Single electrons: pT=5 GeV –Single muons: pT=5 GeV……..

18. ATLAS Trigger GroupPPD Staff Talk 23/5/05 18 Efficiency for High Energy Electrons

19. ATLAS Trigger GroupPPD Staff Talk 23/5/05 19 Efficiency for Low Energy Electrons

20. ATLAS Trigger GroupPPD Staff Talk 23/5/05 20 Unitarity Triangle/CP-Violation

21. ATLAS Trigger GroupPPD Staff Talk 23/5/05 21 Unitarity Triangle/CP-Violation

22. ATLAS Trigger GroupPPD Staff Talk 23/5/05 22 Unitarity Triangle/CP-Violation >

23. ATLAS Trigger GroupPPD Staff Talk 23/5/05 23 Monika Wielers Started last October at RAL Worked on electron and photon selection for many years now in ATLAS –Convenor of ATLAS e/gamma trigger group Long experience on calorimetry –Matches well with tracking activities here, since you need both to identify electrons and photons

24. ATLAS Trigger GroupPPD Staff Talk 23/5/05 24 Electron and Photon Triggers Number of events we ‘see’ per second: ~10 9 Challenge to efficiently select interesting events and reject the enormous number of ‘uninteresting’ events –Not easy to find the interesting events –We can only save ~1 out of 10 million events for further analysis –Sometimes the events we are interested are very rare (H  ), so we have to make sure to collect all of those –electrons and photons (and muons) are ‘easy’ to select Important physics events containing electrons/photons cover –whole range of physics topics we want to explore at the LHC From the known (Standard Model) to the unknown (Higgs, Susy, …) –Whole spectrum of possible energies ‘Low’ pT: 5-15 GeV (B-physics) ‘High’ pT: 20-100 GeV (Higgs, Susy, top…) ‘Very high’ pT: 100-1000 GeV (new heavy W/Z bosons, excited electrons…)

25. ATLAS Trigger GroupPPD Staff Talk 23/5/05 25 Event Display H  ZZ*  2e2  Electron and photon trigger selection  find interesting regions with lots of energy in the calorimeter at LVL1 (coarse granularity)  Now look in more detail at the information from the different detectors at LVL2 (calorimeter, inner detector)  if the object looks ‘like’ an electron go on to the EF  do something very similar but more ‘sophisticated’ at the EF (you have more time at this stage)  If you still think it’s an electron/photon save the event for the physicists to look at

26. ATLAS Trigger GroupPPD Staff Talk 23/5/05 26 Trigger Strategy Reject ‘uninteresting’ events as fast as possible –Achieved by sequential selection: after each step we look if we still want to identify a given object as electron or photon HLT algorithms optimised in terms of: –Physics performance Signal efficiency Background rejection In the end we have to find the best compromise among all these constraints –Signal performance Execution time Data requirements Signal efficiency  Physics reach Rate  latency  No. of computers needed Rate  data size  Size of network needed £

27. ATLAS Trigger GroupPPD Staff Talk 23/5/05 27 We are slowly getting there … 2007