second level trigger pixel detector for the ATLAS A track finding ... · 2000. 7. 6. · Pixel2000 - Genova - 6/6/00 P. Morettini 1 A track finding algorithm based on pixel detector

Pixel2000 - Genova - 6/6/00P. Morettini 1

A track finding algorithm based on A track finding algorithm based on pixel detector for the ATLAS pixel detector for the ATLAS

second level triggersecond level triggerAndrea Baratella – Mauro Dameri - Paolo Morettini – Fabrizio Parodi

INFN Genova

Outline• ATLAS trigger strategy for the B-physics.• PixTrig: a pixel based Level 2 track finding algorithm.• Efficiency and performances.• Application to the selection of rare B decays.• Application to the b-tagging in multi-b jets events.• Conclusions


The ATLAS Level 2 TriggerThe ATLAS Level 2 Trigger

Network switch

CommercialProcessors

R/OBuffers

R/OBuffers

R/OBuffers

The ATLAS second level trigger refines the decision taken at Level 1 using the data from the different sub-detectors, including the trackers. It has to take its decision in 10-20 ms, reducing the overall rate by a factor of 10 (from 100 KHz to 10 KHz).

The implementation of this trigger will be based on a farm of commercial processors connected to the R/O buffers trough a switch. Hardware co-processors, both at buffer and farm level, are also under study.


LVL2 Processing ModesLVL2 Processing ModesThe ATLAS trigger is mostly based on “Regions of Interest” (RoI). This means that every trigger component analyses only the portions of the detectors identified by the previous components; in the case of the LVL2, only the region where a muon or a jet where found at LVL1 are reconstructed.In the case of the rare b decays reconstruction however, the interesting jet is usually not energetic enough to be seen at LVL1, so the event is triggered by a high pT µ in the second jet.In this case a “Full Scan” is necessary at LVL2 to reconstruct the charged tracks in the whole inner tracker, and select the vertexes using invariant mass and impact parameter information.


PixTrigPixTrig: The Basic Idea : The Basic Idea PixTrig is a track finding algorithm designed to be used in Full Scan and RoI guided mode. Thanks to the high resolution and low noise of the pixel detector it can provide full 3d track reconstruction and impact parameter measurement to be used in stand-alone mode or as seed for other algorithms.PixTrig is a pure combinatorial algorithm using the ATLAS pixel detector space points: combinations in the two innermost layers are extrapolated to the third if pointing to the primary vertex. If a point close to the extrapolation is

∆ Z 3∆ φR 3

∆ φR ext Ro I ∆ Z e xt

found in layer 3, the points are kept as a candidate track.Overlaps are solved on the basis of the residuals in the third layer.


Logical Layers Definition Logical Layers Definition The original implementation of PixTrig was based on the three phisical layers of the barrel pixel detector. To adapt the algorithm to the end-caps geometry (where we have five Link L1-L2 Link L1-L3disks) we moved, in the present C++ implementation within the ATLAS trigger simulation, to the definition of “logical layers”.A logical layer is an arbitrary collection of detector modules. Logical first layers (every module in the barrel B-layer of the detector) have a link to the modules in the corresponding second and third logical layers.This approach increase the flexibility of the algorithm and its robustness to the pixel inefficiencies. However, we have to keep in mind that

More complexity in theLogical Layers More combinations

Longer processing timeand more fake tracks→→→→→→→→→→→→→→→→


Reconstruction Efficiency in jets Reconstruction Efficiency in jets

0

0.2

0.4

0.6

0.8

1

-2 0 20

0.2

0.4

0.6

0.8

1

-2 0 2

η

Eff

icie

ncy

η

Fak

e T

rks

Fra

ctio

n

0

0.2

0.4

0.6

0.8

1

-2 0 20

0.2

0.4

0.6

0.8

1

-2 0 2

η

Eff

icie

ncy

η

Fak

e T

rks

Fra

ctio

n

BB--jets jets –– No pileNo pile--upupThe The reconstruction efficiencyreconstruction efficiencyin jet in jet RoI RoI is close to is close to 90%90% and and flatflat over over η

η and and ppTT..At At low luminositylow luminosity the fraction the fraction of of fake tracksfake tracks (combination of (combination of points coming from different points coming from different particles) is always particles) is always below 20%below 20%(it’s higher in the end(it’s higher in the end--caps caps because more combinations are because more combinations are possible there).possible there).At At design luminositydesign luminosity the the fraction of fakes reaches fraction of fakes reaches 60%60%in the endin the end--caps (spurious caps (spurious combinations lying on a straight combinations lying on a straight line can be preferred to good line can be preferred to good tracks). tracks).

BB--jets jets –– 24 24 evev. pile. pile--upup


Timing (jet Timing (jet RoIRoI))The processing time on a commercial Pentium III 500 MHz system is of 3.7 ms per jet.As expected, this time scales with the number of clusters in the RoI following a cubic law; however the coefficient of the third power is small, so the behavior is approximately quadratic.At design luminosity, the average time needed to process a jet is close to 90 ms.

0

20

40

60

0 10 20 30

a)

IDEntriesMeanRMSUDFLWOVFLW

800 679

3.656 3.113

0.0000E+00 3.000

Processing time (ms)

Eve

nts

b)

0.6536 / 11P1 0.0000E+00P2 0.6945E-01P3 0.7165E-03P4 0.6402E-04

Mean number of clusters per layer

Pro

cess

ing

time

(ms)

0

5

10

15

20

0 10 20 30 40 50

0

10

20

30

40

50

0 100 200 300 400

a)


800 719

86.32 81.04

0.0000E+00 5.000


Eve

nts

b)

2.218 / 15P1 0.0000E+00P2 0.2935E-01P3 0.1855E-02P4 0.2302E-04


Pro

cess

ing

time

(ms)

0

50

100

150

200

0 50 100 150

BB--jets jets –– No pileNo pile--upup

BB--jets jets –– 2424 evev. pile. pile--upup


ZZvertexvertex Reconstruction with a LVL1 Reconstruction with a LVL1 µµµµµµµµTo reduce the processing time we have to decrease the number of extrapolation to the third layer. This could be achieved using a hard cut on the Z impact parameter of the lay1-lay2 combinations, but the position of the primary vertex need to be known with some precision.In events triggered at LVL1 by a muon, it is possible to use PixTrig itself with a high pT threshold, to find the muon and reconstruct its Z0 (which is close to Zvertex).In 95.7% of the events one candidate is found in the µ

0

200

400

600

0 1 2 3 4

Reconstructed Tracks in µ RoI

EntriesMeanRMSUDFLWOVFLW

718 0.1667E-02 0.6718E-01

5.000 2.000

Zrec - Zvertex

0

50

100

150

-0.4 -0.2 0 0.2 0.4

RoI, and in 99% of the cases its Z0 is less than 5 mm apart from the Z of vertex .The reconstruction of the vertex takes 0.5 ms.


ZZvertexvertex Reconstruction Inside a JetReconstruction Inside a JetInside a jet RoI, the position of the vertex can be reconstructed as the barycenter of the Z0 of the tracks in 55% of the events.This does not save processing time, but allows a clean-up of the sample “a posteriori”.Using this method, the efficiency and purity achieved at design luminosity are similar to the low luminosity ones.

1

10

10 2

10 3

-20 -10 0 10 20

Entries 1644

zrec-ztrue

Num

ber

of e

vent

s

Good qualitysample

1

10

10 2

10 3

-20 -10 0 10 20

Entries 1374

zrec-ztrue

Num

ber

of e

vent

s

Bad qualitysample

0

0.2

0.4

0.6

0.8

1

1.2

1.4

0 2 4 6 8 10

pT

Efficiency

N(Fake Tracks)/Ntot

η

0

0.2

0.4

0.6

0.8

1

1.2

1.4

-3 -2 -1 0 1 2 3

BB--jets jets –– 2424 evev. pile. pile--upup


Full Scan Efficiency and TimingFull Scan Efficiency and TimingFull-scan on b eventswith at least one µwith pT > 5 GeV/c.Low luminosity.Track reconstruction pT cut at 0.5 GeV/c.CPU: PIII 500 MHz

ε: 95 %Fakes : 5% barrel

30% end-capsTotal time: 35.3 ms, almost linear with the number of clusters.

0

0.2

0.4

0.6

0.8

1

-2 0 20

0.2

0.4

0.6

0.8

1

-2 0 2

ηE

ffic

ienc

yη

Fak

e T

rks

Fra

ctio

n

0

20

40

60

0 100 200 300

a)


800 750

35.29 28.74

0.0000E+00 14.00


Eve

nts

b)

3.084 / 13P1 0.0000E+00P2 0.1457P3 0.2473E-02P4 -0.5960E-05


Pro

cess

ing

time

(ms)

0

50

100

150

0 50 100 150 200 250


Track Parameters ResolutionTrack Parameters ResolutionThe track parameters are calculated writing the equation of the helix connecting the three point in the space. The resolutions are all acceptable, if we consider that no fit is performed and all the points get the same weight. The error on d0 is dominated by the back extrapolation to the vertex

0

100

-0.1 -0.05 0 0.05 0.1-0.1

0

0.1

0 5 10 15

0

100

-0.2 -0.1 0 0.1 0.2-0.2

0

0.2

0 5 10 15

∆d0 (cm) pT (GeV/c)

∆d0

(cm

)

∆Z0 (cm) pT (GeV/c)

∆Z0

(cm

)

34 10-330 10-31/pT

1.9 mrad1.3 mradφ2.2 mrad1.8 mradθ324 µm135 µmZ0

85 µm68 µmD0

End-capsBarrel


Exclusive B decays at LVL2Exclusive B decays at LVL2For the selection of exclusive B decays, a complete track reconstruction in the inner tracker is performed following the direction of the seeds given by the full-scan. In this context, PixTrig provides good performances:

• it’s fast• it produces few track candidates• it gives a full 3D definition of the

tracks, limiting the volume scanned by the second stage algorithm

• it gives good overall resolutionbecause it’s less sensitive to the interactions in the material of the detector.

mππ (GeV)

num

ber

of e

vent

s

Si Kalman filterPixel-guidedTRT-guided

0

25

50

75

100

125

150

175

200

4 4.5 5 5.5 6 6.5 7

pT > 0.5

# of full totalseeds scan time

PixTrig 39 36 ms 66 msTRT scan 171 150ms 317ms


BB--tagging at LVL2tagging at LVL2The B-tagging can be performed in one or more jet RoI using PixTrig alone. The method is the classical one, based on the likelihood ratiobuilt on the d0 distributions for signal (b-jets) and background (u-jets).Since no error on d0 is calculated, we can use the same weight for all the tracks or use a pT parametrizederror estimate.The two methods give similar results:

εb = 0.8 Ru = 4εb = 0.5 Ru = 25

pT(GeV/c)

σ(d 0)(

µm)

0.002

0.004

0.006

0.008

0.01

0 10 20 30

1

10

10 2

0 0.2 0.4 0.6 0.8 1

εb

Ru

|d0|

d0/σ

σd0 vs pT


Comparison with the offComparison with the off--line bline b--tagtagIt is important to check that the on-line preselctiondoes not spoil the final off-line performances.This plot compares the pure off-line results with those obtained running the off-line algorithm over a sample pre-selected with the on-line b-tagging tuned at εb = 90%

1

10

10 2

10 3

0 0.2 0.4 0.6 0.8 1

Offline algorithm

Trigger algorithm

Offline algorithmon a preselected sample

εb

Ru


Selection of H Selection of H →→→→→→→→ hhhh→→→→→→→→ bbbbbbbbThe b-tagging can be used to increase the trigger acceptance of multi b-jets events, which are cut by high ET thresholds at LVL1.The price to pay is an increase of the jet-trigger rates which is significant at LVL1 and modest after the LVL2 btag. 1

10

10 2

10 3

0 0.2 0.4 0.6 0.8 1

ε(4b)

R(D

ijets

)

Single Jet

Combined

Single Jet Reference

19 %1.0Standard LVL1

53 %1.5LVL1 Loose + Btag tight58 %2.0LVL1 Loose + Btag loose76 %6.4LVL1 Loose

E on signalRate IncreaseSelection


ConclusionsConclusionsThe feasibility of a fast track finding algorithm based on the ATLAS pixel detector has been demonstrated. The performances are within the LVL2 specifications both for the “full-scan” and the “RoI guided” mode. They benefit from the high precision and cleanness of the pixel space points. The b-tagging could be useful at LVL2 to increase the acceptance of multi b-jet events.Development is on-going in the following sectors:

• Fine tuning of the algorithm, especially in the end-caps• Robustness to the detector inefficiency and

misalignment• Sensitivity to the clustering algorithm used

Documents

second level trigger pixel detector for the ATLAS A track finding ... · 2000. 7. 6. · Pixel2000 - Genova - 6/6/00 P. Morettini 1 A track finding algorithm based on pixel detector