1. Tommaso Boccali SNS and INFN Pisa Vertex 2002 Kailua-Kona, Hawaii 3-8 November 2002 High Level Triggers at CMS with the Tracker

2. Vertex 2002 Kailua-Kona, Hawaii High Level Triggers at CMS with the Tracker Tommaso Boccali The Challenge Lot of b’s and tau’s from interesting Physics – Supersymmetry – Higgs decays – Top, B physics Large QCD backgrounds (factor ~1000 or more) The Challenge L1: 40 MHz input L1:100 KHz output Write to offline:100 Hz

3. Vertex 2002 Kailua-Kona, Hawaii High Level Triggers at CMS with the Tracker Tommaso Boccali How to select events? General idea: we would like to benefit from full offline analyses to discriminate events. Since all the data is available after Lvl-1, we would need “only” ~infinite CPU power. What can be done with a reasonable number of commercial CPUs?What can be done with a reasonable number of commercial CPUs? CMS HLT functionality depends on data rate and CPU resources available Traditional L2 functionality is the most challenging parameter. Does not benefits of full-granularity

4. Vertex 2002 Kailua-Kona, Hawaii High Level Triggers at CMS with the Tracker Tommaso Boccali High Level Trigger High Level Trigger The input rate is the Full Lvl-1 output rate, 100 kHz @ 10 34 The maximum output rate is limited by the bandwidth of the storage system, O(100 Hz) We need rejection factors O(1000). 40 MHz 1 TB/s 100 Hz HLT run here On exactly the same offline code

5. Vertex 2002 Kailua-Kona, Hawaii High Level Triggers at CMS with the Tracker Tommaso Boccali Timings? We have ~100 kHz of input rate We realistically can have a few thousands of (2007) CPUs This means ~50 ms per decision Use Moore’s Law: –50 ms in 2007 ~ 500 ms on a 1 GHz CPU today  Timing for a 1 GHz CPU today ~500 ms

6. Vertex 2002 Kailua-Kona, Hawaii High Level Triggers at CMS with the Tracker Tommaso Boccali Not only selecting... but rejecting unwanted ones Since we want O(1000) rejection factors, most of the time is not spent selecting good events, but rejecting unwanted ones. “how can I kill this event using the least CPU time?” So: “how can I kill this event using the least CPU time?” Two ways: you can use 1.The fastest (most approximate) reconstruction do everything, even if it is not accurate 2.The minimal amount of precise reconstruction select only a few quantities, and use nearly offline precision

7. Vertex 2002 Kailua-Kona, Hawaii High Level Triggers at CMS with the Tracker Tommaso Boccali What about the Tk? not used at all for Lvl-1The Tracker is the most precise subsystem, but is ( thought to be ) slow  not used at all for Lvl-1 Extensive studies have be performed to answer the question: what can we do in ~500 ms? The CMS Tracker has been already described by S. Schael (  Strips) and L. Cremaldi (Pixels) 6,136 Thin 300  m (1 sensor) 9,096 Thick 500  m (2 sensors) 9x10 6 strips Occupancy ~10 -2 14192 chips 39x10 6 pixels Occupancy ~10 -4

8. Vertex 2002 Kailua-Kona, Hawaii High Level Triggers at CMS with the Tracker Tommaso Boccali Baseline TK Inner Barrel (TIB) TK Outer Barrel (TOB) TK Inner Disk (TID) TK Outer disks (TEC) pixels One charged track traversing the CMS Tracker interacts with a maximum of ~14 hits in the silicon (+3 in the Pixels)

9. Vertex 2002 Kailua-Kona, Hawaii High Level Triggers at CMS with the Tracker Tommaso Boccali Tracks et al. The Tracker is well suited for Track reconstruction (P,S)Vertex reconstruction Impact parameters etc.... and hence b-tagging  -tagging precision measurements (refinements) of jets and momenta

10. Vertex 2002 Kailua-Kona, Hawaii High Level Triggers at CMS with the Tracker Tommaso Boccali General recipes Do something only when needed! reconstruction on demand: never do anything until it is requested Usually, one is not really interested in reconstructing “the event” in the trigger decision, a smaller region is sufficient regional tracking How close we have to be to full offline performance? conditional tracking In many cases, it is simple enough to recognize the trigger event from the pile-ups pixel reconstruction

11. Vertex 2002 Kailua-Kona, Hawaii High Level Triggers at CMS with the Tracker Tommaso Boccali Regional tracking Reconstruct only clusters and tracks in a certain region of the Tracker (i.e., around a jet) In principle, complexity (time!) scales with (  /4   The region searched for is defined in terms of R typical cases  R~0.4 region around a Lvl-1calo jet -2.5<  <2.5; 0<  <2 

12. Vertex 2002 Kailua-Kona, Hawaii High Level Triggers at CMS with the Tracker Tommaso Boccali Conditional tracking What about performances?After the complete solution of the combinatorial problem, the time spent for each track is proportional to the number of hits. What about performances? if at least 7 hits are used, efficiencies and fake rates are close to optimal; 5 hits are enough if we can accept a higher ghost rate. @ 2x10 33 lumi

13. Vertex 2002 Kailua-Kona, Hawaii High Level Triggers at CMS with the Tracker Tommaso Boccali Conditional tracking resolutions Not only eff/fake rates are fine, but also resolutions Good resolution even with only 5 hits Full reconstruction

14. Vertex 2002 Kailua-Kona, Hawaii High Level Triggers at CMS with the Tracker Tommaso Boccali Pixel Resolution in z (cm) Pixels @ HLT @ 2x10 33, at LHC we have ~3.5 minimum bias events per bunch crossing, @10 34, ~17.5 moreover, we integrate signals from bunch crossings [-5,+3]. trigger primary vertexwe have up to 200 events contribute; it is essential to isolate the trigger primary vertex and the tracks associated to it.  = 26  m Using only the Pixels: fast ~ 50 ms/1GHz CPU good resolution ~ 30  m in z efficient ~100%

15. Vertex 2002 Kailua-Kona, Hawaii High Level Triggers at CMS with the Tracker Tommaso Boccali Physics @ HLT With these ingredients, we can do inclusive  tagging inclusive b tagging exclusive b tagging Moreover, we must take care about detector configuration: we ( yet ) don’t know what will be available at the startup, particularly about pixel configuration: 2 or 3 barrel layers?2 or 3 barrel layers? 1 or two forward disks?1 or two forward disks? All combinations checked!All combinations checked! !!

16. Vertex 2002 Kailua-Kona, Hawaii High Level Triggers at CMS with the Tracker Tommaso Boccali  tagging Regional Tracking : Look only in Jet-track matching cone Loose Primary Vertex association Conditional Tracking : Stop track as soon as Pixel seed found (PXL) / 6 hits found (Trk) If Pt<1 GeV with high C.L. Reject event if no “leading track” found Regional Tracking : Look only inside Isolation cone Loose Primary Vertex association Conditional Tracking : Stop track as soon as Pixel seed found (PXL) / 6 hits found (Trk) If Pt<1 GeV with high C.L. Reject event as soon as additional track found a region  R=0.5 around the calo jet is searched

17. Vertex 2002 Kailua-Kona, Hawaii High Level Triggers at CMS with the Tracker Tommaso Boccali The time spent to reject QCD background is ~300 ms @ low lumi and less than 400 ms @ high lumi  tagging A 0 /H 0  2  2  -jet selection with Tracker: various configurations studied; heavy use of conditional tracking (6 hits) and regional tracking Efficiency in various situations and two different Higgs masses. The working point is chosen at a QCD background rejection of ~10 3 @ 2x10 33 sec

18. Vertex 2002 Kailua-Kona, Hawaii High Level Triggers at CMS with the Tracker Tommaso Boccali inclusive b tagging Performance of HLT tagging close to offline! Regional Tracking: Look only in Jet-track matching cone Loose Primary Vertex association Conditional Tracking: Stop track as soon as Pixel seed found (PXL) / 6 hits found (Trk) If Pt<1 GeV with high C.L. timings: ~300ms low lumi ~1s high lumi bb uu

19. Vertex 2002 Kailua-Kona, Hawaii High Level Triggers at CMS with the Tracker Tommaso Boccali exclusive b tagging Specific decay channels: we have too many b’s in non interesting decay channels! Use dilepton triggers to start the selection, and apply tracker selection on invariant masses Use large tail in Lvl-1 resolution: we are mostly interested in muons with low p T wrongly selected ( Lvl-1 thresholds are 95% ) Most important: low rates on QCD backgrounds to tape. 2 Lvl-1  Primary vertex to associate tracks Reg & Cond tracking around the two jets (only tracks p T >2 GeV, max 5 hits) Hard cuts on mass peaks Tape 900 Hz p T >3 GeV @ 95% <2 Hz bkg eff ~ 10%

20. Vertex 2002 Kailua-Kona, Hawaii High Level Triggers at CMS with the Tracker Tommaso Boccali exclusive b tagging Exclusive b-tagging: Use Level 1 events with 2 muons, together with regional/conditional tracking B S  B S  J/  KK b  B S  D S  (KK)  B S  Offline HLT B S  J/  KK Everything feasible in HLT These are events in year 1 (~10 fb -1 ) eff~10% rejection~100 CDF RunIIa 0 CDF RunIIb maybe CDF RunIIa ~ 14 for now

21. Vertex 2002 Kailua-Kona, Hawaii High Level Triggers at CMS with the Tracker Tommaso Boccali Conclusions –5yThe CMS Tracker can be used in HLT on the full Lvl-1 output (100 kHz) with nearly offline performance now (startup –5y) b and  inclusive physics in good shape and ready for discovery b exclusive physics possible in parasitic mode These results are solid even if the detector will be staged, at and both 2x10 33 and 10 34 Lumi