L0 Confirmation Velo to HCAL

Transcription

1 L Confirmation box Matching strategy Tests Efficiencies IP-cut Timing Conclusions L Confirmation Velo to HCAL May 29, 26 Many thanks to Jose Hernando and Thomas Schietinger! May 29, 26 L Confirmation Velo to HCAL (1)

2 HLT Hadron alley The 1MHz readout requires a reorganisation of L1 and HLT. The general philosophy is a vertical trigger Emphasis is on confirmation of previously found trigger objects Alley boxes : L confirmation (new) Pre-trigger (old Level 1) Trigger (old HLT generic) Try to confirm the L hadron by matching it to a Velo track Code originally by N. Tuning May 29, 26 L Confirmation Velo to HCAL (2)

3 L Confirmation box Algorithms PatDataStore Make 2D tracks: PatVeloRTracking Make PV & IP: PatPV2D, HltHadTrack2DIPSelection Hlt/Tracks/RZVelo 2D confirmation: HadLConf2DDecision Make 3D tracks: PatVeloSpaceTracking 3D confirmation: HadLConf3DDecision Hlt/Tracks/RZVeloLCalo Hlt/Tracks/SpaceVelo Hlt/Tracks/SpaceVeloLCalo May 29, 26 L Confirmation Velo to HCAL (3)

4 Matching strategy HCAL x Et VeLo z kick Magnet slope Input: Velo track (2D or 3D); HCAL cluster (2 2 cells) Undo the magnetic field: knowing the Calo hit E T, estimate the kick with x±kick z match the track slope dx dz same in y (without kick correction) from the hit May 29, 26 L Confirmation Velo to HCAL (4)

5 Velo tracks θ y.4 2D tracks θ y.4 3D tracks D tracks: only Velo r sensors are used r φ geometry θ x D tracks: Velo r and φ sensors are used cartesian geometry θ x May 29, 26 L Confirmation Velo to HCAL (5)

6 Matching 3D tracks For a VeLo 3D track with slopes t x and t y for charge q = ±1, compute : = x + χ2 y ( ) 2 x qxkick 1 x = t x z ( y ) 2 y = z t 1 y x kick = C kick z z o E σ x and σ y take into account errors on cluster position and energy. C kick = GeV, z = 525 mm (center of magnetic field) Cut on σ 2 y σ 2 x May 29, 26 L Confirmation Velo to HCAL (6)

7 Matching 2D tracks For a VeLo 2D track with slope t r and azimuthal angle φ for charge q = ±1, compute, translating Calo coordinates to r φ : = r + χ2 φ ( ) dr 2 r = 1 dz t r Calo φ = (φ Calo φ VeLo ) 2 σ 2 φ σ 2 r ( ) y φ Calo = arctan x qx kick dr (x qxkick ) dz = 2 + y 2 Calo z σ r and σ φ take into account errors on cluster position, energy and φ sector size. Cut on May 29, 26 L Confirmation Velo to HCAL (7)

8 Testing the tool Does the matching work? Test procedure: via MCParticles, make sure the match is correct update the track with the MC momentum, and extrapolate to HCAL look at the distance between the calo hit and the extrapolated track. This distance should be smaller than the cluster half-size, and correlated to. compute MC in the same way as the tool does, but using the momentum from the MC-truth to compute the kick. MC should be correlated to χ2. May 29, 26 L Confirmation Velo to HCAL (8)

9 Tests for 2D δ [mm] 6 distance r [mm] 6 r difference 5 δ r δ φ [rad].8.7 φ difference MC correlation φ D resolution is dominated by VeLo φ component! May 29, 26 L Confirmation Velo to HCAL (9)

10 Tests for 3D δ [mm] 4 35 distance x [mm] δ 4 35 x difference x δ y [mm] 4 35 y difference MC 6 5 correlation y red: HCAL inner region (13 13 cm 2 cells); blue: HCAL outer region (26 26 cm 2 ) 3D resolution is dominated by Calo cluster size and error on kick May 29, 26 L Confirmation Velo to HCAL (1)

11 distribution 1-2D, B π + π d All values 8 Correct matches D, B π + π - d All values Correct matches D, MinBias All values Correct matches 8 6 3D, MinBias All values Correct matches B d π + π events, 6 Min Bias events (RTTC) May 29, 26 L Confirmation Velo to HCAL (11)

12 Efficiencies % 1 % D, B π + π - d Retention Purity Pur. Eff 6 3D, B π + π - d Retention Purity Pur. Eff cut cut % 1 % D, MinBias Retention Purity Pur. Eff 6 3D, MinBias Retention Purity Pur. Eff cut cut May 29, 26 L Confirmation Velo to HCAL (12)

13 Track selection average number of 2D tracks average number of 3D tracks B d π + π MinBias B d π + π MinBias cut cut Channel Retention Tracks Events 2D 3D 2D 3D B d π + π 99.9% 95.7% B s D s K 99.7% 92.6% MinBias 98.2% 78.9% both cuts set to 7. May 29, 26 L Confirmation Velo to HCAL (13)

14 Impact parameter cut Can we improve the track selection by applying an IP-cut before the 2D matching, without losing efficiency? 5% of background tracks have a negative IP (PV has z < ) Cutting out these negative IP s, we only lose 3% of B d π + π events Minimum bias retention goes down to 89.7% (98.2% without IP cut), with 3.7 tracks/event (instead of 7.2) 1 98 Signal MinBias 96 retention (%) none..5.1 IP cut [mm] May 29, 26 L Confirmation Velo to HCAL (14)

15 Timing Running 1 minimum bias events on a machine that is about.96 times faster than a 2.8 GHz Xeon: Algorithm Events Total [s] % HltHadLConfirmation % PatInitEvent % CreateRawEvent % PatVeloDecodeRaw % PatVeloRTracking % PatPV2D % HltHadTrack2DIPSelection % HadLConf2DDecision % HltHadVeloSpacePartial % HadLConf3DDecision % May 29, 26 L Confirmation Velo to HCAL (15)

16 Conclusions tool code in C++ compiles and runs (with RTTC data), transition to DC 6 should be painless tests show that the matching works matching resolution: 3D resolution is excellent in y, less in x because of the kick correction. 2D resolution is worse because of the poor φ information. HCAL cluster position can be improved by a factor 4 with the s-curve correction (under study by Malcolm John) it seems reasonable to cut around 7 8 signal selection efficiency is about 96% minimum bias retention is about 78% after 3D matching good track selection efficiency May 29, 26 L Confirmation Velo to HCAL (16)