ATLAS Egamma Trigger Overview Xin Wu University of

ATLAS Egamma Trigger Overview Xin Wu University of Geneva X. Wu, March 2006 1

Outline § § § § Introduction LVL 1 EM Trigger LVL 2 EM Trigger EF EM Trigger Overall Performance Online Integration Conclusion X. Wu, March 2006 2

Introduction Ø Egamma Trigger: online selection of electrons and photons – LVL 1: hardware processors to reconstruct (isolated) EM cluster – LVL 2: Seeded fast Athena clustering and tracking algorithms – EF: (seeded) offline clustering and tracking algorithms Ø Responsible for a large fraction of data for ATLAS physics – Inclusive electron, dielectron (e 25 i, 2 e 15 i) • Main triggers for W, Z, dibosons, top, Higgs, SUSY, Exotics – Inclusive photon, diphoton ( 60 i, 2 20 i) • Main triggers for direct photon, H , Exotics – Exclusive (combination and topological) triggers Ø Dominant contributor to the trigger rate TDAQ TDR – ~65% of LVL 1 rate at L=2 E 33 • Total LVL 1: 25 KHz; EM 25 I: 12 k. Hz; 2 EM 15 I: 4 k. Hz – ~35% of EF rate at L=2 E 33 • Total EF: 200 Hz; e 25 i+2 e 15 i: 41 Hz; 60 i+2 20 i: 27 Hz X. Wu, March 2006 3

LVL 1 Calorimeter Trigger System Calorimeters (LAr, Tile) S 0. 1 x 0. 1 Ro. I identification e/ /t classification Threshold count analogue ~75 m Rx Cluster Processor Ro. I Builder Pre. Processor Timing alignment 10 -bit FADC FIR filter BCID LUT Sum 2 x 2 BC-MUX 400 Mb/s Jet/Energy Processor Sum Em+Had ET Ex, Ey Jet identification Threshold count SET, ET 0. 1 x 0. 1 0. 2 x 0. 2 X. Wu, March 2006 L 1 CTP DAQ 4

LVL 1 EM Ro. I Reconstruction Ø Ro. I EM Core: a 0. 2 x 0. 2 local EM Et maximum Ø EM Cluster: most energetic of the four Ro. I Core Trigger. Tower 0. 1 x 0. 1 2 -tower EM clusters in th Ro. I Cluster – Et : LVL 1 EM cluster Et Em Cluster Ø EM isolation – Total Et of the 12 EM towers around the Ro. I Cluster EM Isolation Ø Hadronic core isolation – Total Et of the 4 hadronic towers HAD core behind the Ro. I Core Isolation Ø Hadronic ring isolation – Total Et of the 12 hadronic towers around the Ro. I Core X. Wu, March 2006 HAD ring Isolation 5

LVL 1 Calorimeter Simulation Software Ø Analog tower sum simulation – Need to be run at digitization stage – LAr. L 1 Sim : make LAr. TTL 1 objects from hits (Fabienne Ledroit) – Tile. Hit. To. TTL 1 : make Tile. TTL 1 from hits Ø Trig. T 1 Calo : trigger tower digitization and Ro. I building – Use either TTL 1 or Cells as input – Can be run at digitization or reconstruction stage – Make Trigger. Tower, Em. Tau. ROI, Jet. ROI, Energy. Ro. I objects – Provide simulated input (Ro. I’s) to HLT • Starting point for all efficiency/rate numbers ! Ø CTPsim : make L 1 decisions for a given L 1 menu Ø EDM in ESD/AOD – Trigger. Towers – L 1 EMTau. Object. Container: collection of LVL 1 EM clusters – LVL 1_ROI: collection of LVL 1 Ro. Is ( , , threshold passed) X. Wu, March 2006 6

LVL 1 Egamma Performance Ø Benchmark numbers frequently updated with MC production and reconstruction releases – Eg. EM 25 i (M. Wielers) • Rome data: eff=96. 7%, rate 5. 6 k. Hz (L=1 E 33) • CSC validation: eff=96. 5%, rate 6. 0 k. Hz (L=1 E 33) Ø Detailed studies will be done with CSC data – Efficiency turn-on, noise effects, algorithm bias, dependence of isolation on event topology, … Ø Full characterization of LVL 1 with data has high priority at the beginning of data taking – Tower noise threshold: 250 Me. V steps – Isolation cut: HAD core, HAD ring, EM ring – Energy scale: 1 Ge. V or 500 Me. V or 250 Me. V – Efficiency turn-on – Clustering algorithm tuning, … X. Wu, March 2006 7

L 2 Egamma Calorimeter Algorithm 4 Processing steps of T 2 Calo. Egamma at each step data request is made and accept/reject decision is possible Rcore= E 3 x 7/E 7 X 7 in EM Sampling 2 Eratio=(E 1 -E 2)/(E 1+E 2) in EM Sampling 1 p 0 g Et. Em=Total EM Energy (add sampling 0 and 3) Et. Had=Hadronic Energy (Tile or HEC) X. Wu, March 2006 8

L 2 Egamma Cluster Reconstruction Ø Samp 2 Fex : in sampling 2 – – Find seed cell: hottest cell in the 0. 2 x 0. 2 window around LVL 1 Ro. I sum E in 3*7 and 7*7 cells windows around seed Rcore Cluster center = E weighted eta, phi in a 3 x 7 window around seed Cluster is a 3 x 7 window around the new cluster center Ø Samp 1 Fex: in sampling 1 (strips) – Update cluster energy – Find max E and second max E strips in a window of 0. 125 x 0. 196 around cluster center Eratio Ø Sam. En. Em. Fex – Update cluster energy with sampling 0 and 3 cells – Energy correction applied Et. Em Ø Sam. En. Had. Fex – Calculate sum E of HEC or Tile in 0. 1*0. 1 window around cluster center Et. Had X. Wu, March 2006 9

L 2 Egamma Calo. Data Preparation Ø Region. Selector – Return list of cells and ROB’s in the Ro. I window • Initialization from LAr/Tile Geometry (F. Ledroit) Ø Retrieve ROB data – 2 GB/s link ROS LVL 2 Ø Byte. Stream data conversion (the main bottle beck) – Coupled tightly to ROD data format, DSP processing • Continuous optimization (B. Laforge, D. Fournier, …) – Dedicated LVL 2 Byte. Stream conversion (D. Damazio) • Cell memory allocated and geometry initialized during initialization • Organize cells in TT (Trigger Tower) • Modified decoding method – Factor of 6 faster than offline BS conversion Ø Not yet investigated – Handle dead/noise cells and timing information – Performance study with respect to zero suppression X. Wu, March 2006 10

L 2 Egamma Calo. Timing Performance D. Damazio Offline Conversion Fast Conversion Ø Fast conversion will become default for release 12 and 11. 0. 6 – Validation with physics performance Ø Further improvements – exploit the new ROD data format (B. Laforge) • fixed length block structure, hot cell index, . . . – use of faster/smaller LAr. Cell (D. Damazio) Ø A LVL 2 Egamma Calo. code review is being planned for May-July X. Wu, March 2006 11

LVL 2 Tracking Algorithms Ø Seeded with LVL 2 calo clusters – Search window 0. 2 x 0. 2 (could be narrowed by better Z position from T 2 Calo using strips) Ø 2 independent tacking algorithms with Pixel and SCT – IDScan: histogram method for pattern recognition; Kalman filter for track fitting • Total execution time ~4. 1 ms (Data. Prep ~3. 5 ms) – Si. Track: LUT method for finding triplet track segments straight line (R/Z) and circle (R/Phi) track fitting Ø Tool for track extension to TRT: Trig. TRT_Track. Extension. Tool – Use Probabilistic Data Association Filter • ~ 1 ms/track + Data. Prep Ø TRT standalone and full Inner Detector tracking – TRTx. K: wrapper for the offline tool Xkalman • Total TRT execution time ~4. 6 ms (Data. Prep ~2 ms) X. Wu, March 2006 12

EF Egamma Calorimeter Reconstruction Ø Wrap offline tools to EF environment (Cibran Santamarina) – Seeded approach, interface to trigger steering Trig. Calo. Rec X. Wu, March 2006 13

EF Egamma Tracking Reconstruction Ø Wrap offline new. Tracking tools (I. Grabowska-Bold) – All EF ID algorithms available since release 11. 0. 0 Ø The full Egamma slice is running on BS input with 11. 0. 5 nightlies X. Wu, March 2006 14

Overall Egamma Performance Ø Many studies and optimizations have been done with Rome data and are being repeated for CSC data – Eg. e 25 i for 1 E 33 from M. Wielers, crack region excluded Step LVL 1+LVL 2+EF+offline LVL 1+offline CSC validation data X. Wu, March 2006 Eff (%) 96. 7 80. 3 73. 5 76. 1 Rate 5. 6 k. Hz 42 Hz 34 Hz 73 Hz Rome data Offline = is. EM = 78% Step LVL 1 offline LVL 1+LVL 2+offline LVL 1+LVL 2+EF+offline LVL 1+EF LVL 1+LVL 2+EF Eff (%) 96. 5 83. 9 81. 8 80. 7 79. 2 81. 7 80. 7 Rate 6 k. Hz 180 Hz 78 Hz 52 Hz 33 Hz 59 Hz 40 15 Hz

Comment on Overall Performance Ø Performance numbers are only indicative due the fast evolution of software (trigger and offline) Ø Studies need to couple tightly with offline Egamma reconstruction (not always easy!) Ø Equally important and more challenging is to understand all individual variables – Geometrical, physical and topological bias – robustness against noise – efficiency calculation with data – Simplicity from the point of view of MC simulation, offline reconstruction and real data verification – correction and calibration Ø The final optimization can only be done with data – Get tools ready X. Wu, March 2006 16

HLT integration: Online vs. Online Simulaton vs. Offline GAUDI with support for multiple threads Online GAUDI Online Sim DAQ Data Flow ATHENA Environment L 2 PU/EFPT athena. MT/PT Steering Controller Algorithms ROS X. Wu, March 2006 Byte. Stream File (RDO) Offline ATHENA Environment Algorithms Byte. Stream File or Pool(RIO) File 17

Conclusions Ø Full HLT Egamma slice has been implemented – Basic functionalities and performance satisfactory – Great progresses have been made on more technical areas • LVL 2 data preparation, EDM, EF wrappers, athena. MT, … Ø Next – Validation and performance studies with CSC samples – Integration on HLT pre-series with 11. 0. 6 – Correction and calibration schemes; Monitoring – Algorithm reviews and improvements – Trigger menu for L=1 E 31 • Benchmark physics channels (W, Z, top, DY, Diboson, direct , searches, …) – “Trigger-aware” analyses (physics groups) • Startup scenario for Egamma slice • Trigger/data sample/physics channel for Egamma verification, optimization and efficiency calculation – Tools for trigger commissioning with data X. Wu, March 2006 18