the performance of the atlas inner detector … – the performance of the atlas inner detector...

M.Sutton – The performance of the ATLAS Inner Detector Trigger

X11 Topical Seminar IPRD, Siena - 7-10th June 2010

Mark SuttonUniversity of Sheffield

on behalf of the ATLAS Collaboration

9th June 2010

The performance of the ATLAS Inner Detector Trigger Algorithms in

pp collisions at the LHC



Preface

LHC took it first 900 GeV collisions in November 2009

L=~9 μb-1 at ~5×1026 cm-2s-1

Since March 2010, 7 TeV collisions L~15 nb-1 at ~2×1029 cm-2s-1

Bunch crossing every 25 ns – 40 MHz rate Data storage capability ~200 Hz Reduction of around 200000 to 1 required

Eventually expect up to ~25 (soft) pp interactions per bunch crossing

Interesting high pT interactions complicated by “pile-up”

Essentially all interesting signatures contain leptons or high momentum tracks

2



The ATLAS Detector

3



The Inner Detector Transition Radiation Tracker (TRT)

Radius 2 mm straw tubes Semiconductor Tracker (SCT)

80 μm pitch Silicon strip detectors, double layered, 40 mrad stereo

Pixel Detector Silicon pixel detector 50×400 μm pixels

4



5

The ATLAS Semiconductor Tracker



The ATLAS Trigger System Level 1

Hardware based, pipelined trigger, coarse granularity, largely Calorimeter and Muon system based

Identifies “Regions of Interest” 2.5 microsecond latency

High Level Trigger - Level 2 and Event Filter

Software based, farms of commodity CPU’s and ethernet

Level 2 Seeded by Level 1 in Regions of

Interest Full detector granularity, all detector

subsystems Event Filter

Seeded by Level 2 Access to complete event,

reconstruction similar to Offline

6

CalorimeterTrigger

MuonTrigger

CTP

ROIB L2SV

L2NL2P

EFP

µ

RoI

1PByte/s

120 GByte/s

3 GByte/s

Full events

data

requestsRoI

40 MHz

~ 40 ms

~ 4 s

EF Acc

L2 Acc

L1 Acc

75 kHz

~200Hz

320 MByte/s

~3kHz

< 2.5 s

(Design quantities)



Collected data at 7TeV

LHC provided the first collisions at 7 TeV at the end of March 2010 So far, over 800 million events written to tape with stable beams

7

!"#"$"%$#" &'()*+,(-.'(/0(12345,/6(($((789*:;5<3/+(/0(=*<(>:/?@4<3/+(5<(A(B*C D

!"#"$%"&'()

! !"#"$%"&'()$%#*+,)+$)-./&'"%%)%$"''0/1,&"#)(2$3$-453$/6$+"#"$./(().#)+$#70/*87$9'0,($::

" 3;$-453$+)(,<)0)+$42$=>?$"%$/6$@"2$:A

" B)"C$,-%#"-#"-)/*%$(*&,-/%,#2$D$:$1$3E:$F $.&5:$%53



8

Tracking through the Silicon detectors



Track Selection

Good Offline tracks were selected from 900GeV or 7TeV collisions |η| < 2.5, pT > 1GeV (no pT cut for distributions versus pT) Number of pixel hits > 0, Number of SCT clusters > 5 consistent with Offline vertex

The performance of the HLT Level 2 and EventFilter tracking was evaluated by matching with these selected Offline tracks.

9



Efficiency versus Offline track pT

Efficiencies with algorithms “out of the box”

Good performance considering no additional tuning for data

Excellent agreement with Monte Carlo

EventFilter plateaus near 100%

10

[GeV]T

Offline track p

1 2 3 4 5 6 7 8 9 10

L2 tra

ck e

ffic

iency [

%]

0

20

40

60

80

100

ATLAS PreliminaryPreliminary

=900GeVs

Data 2009

Min. Bias Monte Carlo

[GeV]T

Offline track p

0 1 2 3 4 5 6 7 8

EF

Tra

ck E

ffic

iency [%

]

60

65

70

75

80

85

90

95

100

105

ATLAS preliminary

=900 GeVs

Data 2009

EventFilter

Level 2

!Offline track

-3 -2 -1 0 1 2 3

L2 tra

ck e

ffic

iency [%

]

70

75

80

85

90

95

100

105

>1GeV)T

Data 2009 (Offline p

>2GeV)T



=900GeVs

TrigInDet 7TeV plots to be approved

Efficiency - L2

Tracking efficiency for L2 as a function of the ! of the offline tracks, measured on good offline tracks with various PT thresholds.

Trigger tracking has been run in the whole of the detector (ie. the region-of-interest is the whole of the detector). Considered tracks were identified in the Si detectors.

4

!Offline track

-3 -2 -1 0 1 2 3

L2 tra

ck e

ffic

iency [%

]

70

75

80

85

90

95

100

105


=7 TeVs

>1 GeV)T


>2 GeV)T


The threshold behaviour of the efficiency is expected and comes from usage of specialised algorithms optimised for L2 requirements on the track reconstruction & timing constraints.

TrigInDet 7TeV plots to be approved

Efficiency - EF

5

!Offline track

-3 -2 -1 0 1 2 3

EF

tra

ck e

ffic

iency [%

]

88

90

92

94

96

98

100

102


=7 TeVs

>1 GeV)T


>1.2 GeV)T


>2 GeV)T


Tracking efficiency for EF as a function of the ! of the offline tracks, measured on good offline tracks with various PT thresholds.

Trigger tracking has been run in the whole of the detector (ie. the region-of-interest is the whole of the detector). Considered tracks were identified in the Si detectors.

The threshold behaviour of the efficiency is expected and comes from usage of a configuration tuning optimised for EF requirements on the track reconstruction & timing constraints.



Efficiency versus Offline track η Well reproduced by the Monte Carlo,

except at very large |η|

Essentially flat efficiency versus η for EventFilter ~ 100% for pT>1.2 GeV

Higher efficiency for 7 TeV due to harder track pT spectrum

11

!Offline track

-3 -2 -1 0 1 2 3

L2

tra

ck e

ffic

ien

cy [

%]

70

75

80

85

90

95

100

105


=900GeVs

Data 2009

Min. Bias Monte Carlo

Level 2

EventFilter

Level 2



Efficiency vs time

Good stability of trigger over time with changing beam conditions

12

Time of day (CET)

Apr 06 03:00 Apr 06 06:00 Apr 06 09:00 Apr 06 12:00

Tra

ck f

ind

ing

eff

icie

ncy [

%]

70

80

90

100

>2GeV), Level 2 trackingT

Data 2010 (p>2GeV), EventFilter tracking

TData 2010 (p


=7 TeVs

Time of day (CET)

Apr 06 03:00 Apr 06 06:00 Apr 06 09:00 Apr 06 12:00

Me

an

tra

ck z

po

sitio

n [

mm

]

-5

0

5

10

15



TData 2010 (p


=7 TeVs

Time of day (CET)

Apr 06 03:00 Apr 06 06:00 Apr 06 09:00 Apr 06 12:00

Tra

ck f

ind

ing

eff

icie

ncy [

%]

70

80

90

100



TData 2010 (p


=7 TeVs

Time of day (CET)

Apr 06 03:00 Apr 06 06:00 Apr 06 09:00 Apr 06 12:00

Tra

ck f

ind

ing

eff

icie

ncy [

%]

70

80

90

100



TData 2010 (p


=7 TeVs



Transverse beam position

Accurate online measurement of the luminous region Available both to the HLT algorithms and the LHC machine

Easily see reduced size of luminous region for 7 TeV collisions

13

May 5, 2010 – 20 : 40 DRAFT 12

Vertex x [mm]-4 -3 -2 -1 0 1 2 3 4

mµ

Num

ber

of vert

ices p

er

100

0

1000

2000

3000

4000

5000

6000

7000

8000

9000Gaussian fit:

0.002) mm± = (-0.193 µ

0.002) mm± = (0.256 !

ATLAS Preliminary12 Dec 2009 (LHC Fill 911)

= 900 GeVs

Online primary vertex

2" per vertex trackN

| < 2.5# > 1.0 GeV, |T

p

(a) Primary vertex x (900 GeV).

Vertex y [mm]-4 -3 -2 -1 0 1 2 3 4

mµ

Num

ber

of vert

ices p

er

100

0

1000

2000

3000

4000

5000

6000

7000 Gaussian fit:

0.002) mm± = (0.990 µ

0.003) mm± = (0.328 !


= 900 GeVs



| < 2.5# > 1.0 GeV, |T

p

(b) Primary vertex y (900 GeV).

1

10

210

310

Vertex x [mm]-5 -4 -3 -2 -1 0 1 2 3 4 5

Vert

ex y

[m

m]

-5

-4

-3

-2

-1

0

1

2

3

4

5<X> = -0.19 mm

<Y> = 0.98 mm

ATLAS Preliminary

(c) Primary vertex x vs. y (900 GeV).

Vertex x [mm]-4 -3 -2 -1 0 1 2 3 4

mµ

Num

ber

of vert

ices p

er

100

0

2

4

6

8

10

12

14

16

18

20

310!

Gaussian fit:

0.001) mm± = (-0.370 µ

0.001) mm± = (0.120 "

ATLAS Preliminary30 Mar 2010 (LHC Fill 1005)

= 7 TeVs


2# per vertex trackN

| < 2.5$ > 0.5 GeV, |T

p

(d) Primary vertex x (7 TeV).

Vertex y [mm]-4 -3 -2 -1 0 1 2 3 4

mµ

Num

ber

of vert

ices p

er

100

0

2

4

6

8

10

12

14

16

18

310!

Gaussian fit:

0.001) mm± = (0.628 µ

0.001) mm± = (0.132 "


= 7 TeVs



| < 2.5$ > 0.5 GeV, |T

p

(e) Primary vertex y (7 TeV).

1

10

210

310

410

Vertex x [mm]-5 -4 -3 -2 -1 0 1 2 3 4 5

Vert

ex y

[m

m]

-5

-4

-3

-2

-1

0

1

2

3

4

5<X> = -0.36 mm

<Y> = 0.62 mm

ATLAS Preliminary

(f) Primary vertex x vs. y (7 TeV).

Figure 4: Primary-vertex distributions in the transverse plane as produced by the ATLAS HLT, beforecorrecting for vertexing resolution. The data were recorded with unsqueezed optics (! ! = 11 m). Gaus-sian fits are used to extract the mean position and observed width; the fitting range is limited to±1 RMSso as to reduce the sensitivity to resolution tails.

Vertex z [mm]-200 -150 -100 -50 0 50 100 150 200

Num

ber

of vert

ices p

er

1 m

m

0

100

200

300

400

500

600Gaussian fit:

0.183) mm± = (-8.067 µ

0.128) mm± = (40.615 !


= 900 GeVsOnline primary vertex


> 1.0 GeVT

p

| < 2.5#|

(a) Primary vertex z (900 GeV).

Vertex z [mm]-200 -150 -100 -50 0 50 100 150 200

Num

ber

of vert

ices p

er

1 m

m

0

0.2

0.4

0.6

0.8

1

1.2

310!

Gaussian fit:

0.097) mm± = (1.030 µ

0.066) mm± = (22.143 "


= 7 TeVsOnline primary vertex


> 0.5 GeVT

p

| < 2.5$|

(b) Primary vertex z (7 TeV).

Figure 5: Longitudinal primary-vertex distributions as produced by the ATLAS HLT. The impact of thevertexing resolution is negligible. A Gaussian fit is used to extract the luminous length and the positionof the longitudinal luminous centroid.

May 5, 2010 – 20 : 40 DRAFT 12

Vertex x [mm]-4 -3 -2 -1 0 1 2 3 4

mµ

Num

ber

of vert

ices p

er

100

0

1000

2000

3000

4000

5000

6000

7000

8000

9000Gaussian fit:

0.002) mm± = (-0.193 µ

0.002) mm± = (0.256 !


= 900 GeVs



| < 2.5# > 1.0 GeV, |T

p

(a) Primary vertex x (900 GeV).

Vertex y [mm]-4 -3 -2 -1 0 1 2 3 4

mµ

Num

ber

of vert

ices p

er

100

0

1000

2000

3000

4000

5000

6000

7000 Gaussian fit:

0.002) mm± = (0.990 µ

0.003) mm± = (0.328 !


= 900 GeVs



| < 2.5# > 1.0 GeV, |T

p

(b) Primary vertex y (900 GeV).

1

10

210

310

Vertex x [mm]-5 -4 -3 -2 -1 0 1 2 3 4 5

Ve

rte

x y

[m

m]

-5

-4

-3

-2

-1

0

1

2

3

4

5<X> = -0.19 mm

<Y> = 0.98 mm

ATLAS Preliminary

(c) Primary vertex x vs. y (900 GeV).

Vertex x [mm]-4 -3 -2 -1 0 1 2 3 4

mµ

Num

ber

of vert

ices p

er

100

0

2

4

6

8

10

12

14

16

18

20

310!

Gaussian fit:

0.001) mm± = (-0.370 µ

0.001) mm± = (0.120 "


= 7 TeVs



| < 2.5$ > 0.5 GeV, |T

p

(d) Primary vertex x (7 TeV).

Vertex y [mm]-4 -3 -2 -1 0 1 2 3 4

mµ

Num

ber

of vert

ices p

er

100

0

2

4

6

8

10

12

14

16

18

310!

Gaussian fit:

0.001) mm± = (0.628 µ

0.001) mm± = (0.132 "


= 7 TeVs



| < 2.5$ > 0.5 GeV, |T

p

(e) Primary vertex y (7 TeV).

1

10

210

310

410

Vertex x [mm]-5 -4 -3 -2 -1 0 1 2 3 4 5

Ve

rte

x y

[m

m]

-5

-4

-3

-2

-1

0

1

2

3

4

5<X> = -0.36 mm

<Y> = 0.62 mm

ATLAS Preliminary

(f) Primary vertex x vs. y (7 TeV).

Figure 4: Primary-vertex distributions in the transverse plane as produced by the ATLAS HLT, beforecorrecting for vertexing resolution. The data were recorded with unsqueezed optics (! ! = 11 m). Gaus-sian fits are used to extract the mean position and observed width; the fitting range is limited to±1 RMSso as to reduce the sensitivity to resolution tails.

Vertex z [mm]-200 -150 -100 -50 0 50 100 150 200

Num

ber

of vert

ices p

er

1 m

m

0

100

200

300

400

500

600Gaussian fit:

0.183) mm± = (-8.067 µ

0.128) mm± = (40.615 !


= 900 GeVsOnline primary vertex


> 1.0 GeVT

p

| < 2.5#|

(a) Primary vertex z (900 GeV).

Vertex z [mm]-200 -150 -100 -50 0 50 100 150 200

Num

ber

of vert

ices p

er

1 m

m0

0.2

0.4

0.6

0.8

1

1.2

310!

Gaussian fit:

0.097) mm± = (1.030 µ

0.066) mm± = (22.143 "


= 7 TeVsOnline primary vertex


> 0.5 GeVT

p

| < 2.5$|

(b) Primary vertex z (7 TeV).

Figure 5: Longitudinal primary-vertex distributions as produced by the ATLAS HLT. The impact of thevertexing resolution is negligible. A Gaussian fit is used to extract the luminous length and the positionof the longitudinal luminous centroid.

7 TeV Data 7 TeV Data

1

10

210

310

Vertex x [mm]-5 -4 -3 -2 -1 0 1 2 3 4 5

Ve

rte

x y

[m

m]

-5

-4

-3

-2

-1

0

1

2

3

4

5<X> = -0.19 mm

<Y> = 0.98 mm

ATLAS Preliminary

900 GeV Data

!L2 track

-3 -2 -1 0 1 2 3

Nu

mbe

r o

f pix

el spa

ce p

oin

ts p

er

L2 tra

ck

2.4

2.6

2.8

3

3.2

3.4

3.6

3.8

4 = 900 GeVs

Data 2009

Min. Bias MC

ATLAS Preliminary



Pixel hit multiplicities

Mean number of SCT and Pixel clusters on tracks versus offline track η

Data well described by the Monte Carlo, including small number of dead Pixel modules

Offline tracks reproduced by EventFilter reconstruction extremely well

14!Track

-3 -2 -1 0 1 2 3

Num

ber

of pix

el hits p

er

track

2.6

2.8

3

3.2

3.4

3.6

3.8

4

4.2

4.4

Data 2009, EF tracks

Data 2009, Offline tracks

ATLAS preliminary

=900 GeVs

Level 2

!Track

-3 -2 -1 0 1 2 3

Num

ber

of pix

el hits p

er

track

2.6

2.8

3

3.2

3.4

3.6

3.8

4

4.2

4.4



ATLAS preliminary

=900 GeVs

!L2 track

-3 -2 -1 0 1 2 3

Nu

mb

er

of

pix

el sp

ace

po

ints

per

L2

tra

ck

2.4

2.6

2.8

3

3.2

3.4

3.6

3.8

4 = 900 GeVs

Data 2009

Min. Bias MC

ATLAS Preliminary

!Track

-3 -2 -1 0 1 2 3

Nu

mb

er

of

SC

T c

luste

rs p

er

tra

ck

7.5

8

8.5

9

9.5



� � � � � � � � � � � � �=900 GeVs

EventFilter

[mm]oL2-offline track d

-1.5 -1 -0.5 0 0.5 1 1.5

Num

ber

of tr

acks

310

410

510

610= 900 GeVs

Data 20090.0001)mm±=(0.0915

95%!

Min. Bias MC0.00002)mm±=(0.08165

95%!

ATLAS Preliminary



Resolution with respect to Offline track impact parameter

15

Δd0 (L2-Offline) [mm]

Δd0 (EF-Offline) [mm]

σ95%=(0.0258±0.0001) mm

EventFilter reconstruction within about 20μm of the Offline and around 5μm at high pT

Level 2 around 30μm for high pT tracks

(EF-offline) [mm]0d!

-1.5 -1 -0.5 0 0.5 1 1.5

Nu

mb

er

of

tra

cks

210

310

410

510

610 ATLAS PreliminaryPreliminary

=7 TeVsData 2010

0.00001) mm! = (0.01804 95%

"

[GeV]T

Offline track p1 10

Tra

ck im

pact para

mete

r [m

m]

95%

rms

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

Data 2010, Level 2 tracking

Data 2010, EventFilter tracking


=7 TeVs

Level 2

Muon Trigger Efficiency



Efficiency for reconstructing offline Muons with the Inner Detector Trigger in the Region of Interest identified by the L2 Muon algorithm

Efficiency better than 99% for both Level 2 and the EventFilter

16

Level 1 Muon

Level 2 Muon

Level 2 Inner

Detector

tracking

Level 2

EF

EventFilter ID

tracking

EventFilter

muon

reconstruction

Level 1 Muon

Level 2 Muon

Level 2 Inner

Detector

tracking

Level 2

EF

EventFilter ID

tracking

EventFilter

muon

reconstruction

!Offline muon

-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5

Tra

ck e

ffic

ien

cy [

%]

80

85

90

95

100

105




=7 TeVs

6M.Sutton – Inner Detector Trigger Performance plots for approval

Trigger General Meeting 19th May 2010

Tracking efficiency - for Muons

! L2 and EF Si-tracking efficiency for offline muon tracks of |!|<2.5 and that are located inside L2 muon regions-of-interest, shown as a

function of the transverse momentum of the offline muon.

[GeV]T

Offline muon p

5 10 15 20 25 30

Tra

ck e

ffic

ien

cy [

%]

80

85

90

95

100

105



� � � � � PreliminaryPreliminary

=7 TeVs



Summary and Outlook The LHC is routinely providing

collisions at 7 TeV

The HLT tracking is essential for the identification of physics objects with high transverse energy

The ATLAS HLT tracking has been running successfully during data taking since December 2009, and also providing accurate online determination of the vertex

17

Recently (since 23rd May) started running the HLT in rejection mode for the electron triggers.

The HLT performance accurately reproduces that of the Offline algorithms and is remarkably well described by the Monte Carlo.

The future will be an interesting time for us all.



And Finally….

A very big thank you to the organisers for such a gorgeous venue for the meeting…

18

the performance of the atlas inner detector … – the performance of the atlas inner detector...

Documents