LHCb Alignment

LHCb Alignment

12th April 2007S. Viret

Cosener’s Forum

« LHC Startup »

1. Introduction

2. The alignment challenge

3. Conclusions

1. Introduction


3. Conclusions

LHC Startup1 S. Viret

1. Introduction2. The alignment challenge3. Conclusions

New-physics signs are expected at the LHC

Why LHCb ?

1. Physics justification2. The detector3. LHCb startup program

They will be difficult to characterize

Heavy-flavor physics will provide plenty of

discriminating observables


Example : bs

bs

ss

b s

Involves FCNC, forbidden by

Standard ModelBsdecay

b s

W±

t

New contributions could arise and affect

observable parameters (BR, ACP, Aisospin)

b s

t

~

~

Need for loops involving heavy particles




If you want to study B-physics, it’s nice to have :

A precise vertex reconstruction

A very good particle ID

An efficient trigger system

A large b quark production in the acceptance




Beam line

The LHCb experiment :




b

b

b

b

Beam

b & b quark directions highly

correlated.

b quark production cross-section larger at high

100 b

Production cross-section (Pythia)

pT 230 b 105 b-hadrons per seconde at

L=2x1032cm-2s-1 (LHCb nominal lumi.)

A large b quark production in the acceptance

A forward geometry




Beam line

A precise vertex reconstruction An efficient trigger system

VELO (VErtex LOcator)




VELO

VELO :

Innermost part of LHCb. A detector very close to the beam (~8 mm). 42 detection modules in 2 boxes.

VErtex LOcator

~1m




VELO modules

4.2 cm 8 mm

R

Module=

2 sensors (1R/1) glued

together

x

y

z (beam)




VELO is a moving detector !

During LHC beam injection, each box is retracted by 3cm from its nominal position.

Then the boxes are moved back close to the beam, and data taking starts.

VELO box (empty here)




Beam line

A very good particle ID

RICH 1&2




2 complementary detectors :


RICH 2

RICH 1



RICH design :


Photon collected by HPD detectors (484 in total RICH 1&2)

Number of mirrors:RICH 1 : 4 sphericals / 16 planesRICH 2 : 56 sphericals / 40 planes

Some advertising…



Beam line

Tracking System

An efficient trigger system

A precise vertex reconstruction



Tracking stations

Trigger

Tracker

Outer Tracker

125.6 mm

41.4

mm

Trigger Tracker (1 station) Silicon Strips 183 m pitch 128 7-sensor ladders 4 layers: X:U(5o):V(-5o):X 128 ladders to be aligned

Inner Tracker 20% of the tracks Silicon Strips 198 m pitch 1-2 sensor ladders (336 ladders) 4 layers: XUVX

Outer Tracker (3 stations)

5.0 mm Straws Double-layer straws 4 layers: X:U(5o):V(-5o):X

Overlap regions between IT/OT to facilitate relative alignment

Inner Track

er

125.6 mm





HLT

~2 kHz

HLT (10ms): purely software. Fine event selection (streaming).

L1

40 kHz

LEVEL 1 (1ms): purely software. Look for a displaced vertex in the VELO (good detector alignment mandatory here).

Rate

(in

Hz)

107

106

105

104

103

L0

1 MHz

LEVEL 0 (4s): purely hardware. Select the events containing interesting info (, di-muons, e, & hadrons with high pT). Pile-up rejection.

Trigger strategy An efficient trigger system



2007: startup– Pilot run at 450 GeV per beam– Establish running procedures, align detectors in time and space – Integrated luminosity for physics ~ 0 fb–1

2008: early phase– Complete commissioning of detector and trigger at s=14 TeV– Calibrate momentum, energy and particle ID– Start first physics data taking, assume ~ 0.5 fb–1

– Establish physics analyses, understand performance

2009–20xx: stable running– Stable running, assume ~ 2 fb–1/year– Develop full physics program– Exploit statistics, work on systematics




1. Introduction


3. Conclusions


1. What’s the problem with alignment ?2. LHCb alignment strategy3. Sub-detectors overview

The alignment problem:

A particle passes trough a misaligned

detector

What happens if track is fitted using uncorrected

geometry

With no correction, one gets a bad quality track (or even no track at all)

How could this affect LHCb results ?



Example 1 : proper-time estimation

= d · mB c · |pB|

Proper-time

Detector

Primary vertexB-decay vertexTracks

ddnew




Example 2 : trigger efficiency

BsKK events

HLT trigger efficiency

Y axis

With 0.5 mrad tilt of one VELO box, 30% less events selected

These events are definitely lost!!!




A 3 steps procedure :

Complete survey of every sub-detector and of all the structure when installed in the pit


Hardware alignment (position monitoring):

Stepping motors information during VELO boxes closing

OT larges structures positions constantly monitored (RASNIKs system)

Laser alignment for RICH mirror positioning

Software alignment

Alig

nm

en

t p

reci

sion




Software alignment strategy :

Align all sub-detectors (VELO, IT, OT, RICHs) internally

Align the sub-detectors w.r.t. the VELO (Global alignment). Start IT & OT, then TT (not alignable internally), RICH and finally Ecal, Hcal and Muon.




VELO alignment : how to proceed ?

Alignment should be designed to be FAST (few minutes)and PRECISE (<5

m precision)

Data taking

Residuals monitoring

End of run : VELO is open

Start of run : VELO is closed

If necessary…

Software alignment procedure



Step 2

Aligned VELOAlign the boxes using global fit again on primary vertices,

overlapping tracks,...

Step 1

Internally-aligned VELO

Global fit applied on tracks (classic & beam

gas/halo) in the two boxes

Step 0

Misaligned VELO


VELO: the strategy



Residuals are function of the detector resolution, but also of the misalignments

From this…

The geometry we are looking for is the one which minimizes the tracks residuals (in fact there are many of them but there are ways to solve this problem).

… to that

Global fit ?



xclus = ∑i∙i + ∑aj∙j

LINEAR sum on misalignment constants

i

LINEAR sum on track parameters i

(different for each track)


GLOBAL FIT IDEA : Express the residuals as a linear function of the

misalignments, and fit both track and residuals in the meantime:

Taking into account the alignment constants into the fit implies a simultaneous fit of all tracks (they are now all ‘correlated’):

We get the solution in only one step.

The final matrix is huge (Ntracks∙Nlocal+Nglobal)

xclus = xtrack + x

LINEAR sum on track parameters i

(different for each track)

xclus = ∑i∙i + x

LOCAL PART GLOBAL PART


But inversion by partitioning (implemented in V.Blobel’s MILLEPEDE algorithm), reduces the problem to a Nglobal x Nglobal matrix inversion !!!




VELO : MC results (STEP 1 : modules alignment)

x

y

Before After Resolution on alignment constants (with ~20000

tracks/box) are 1.2 m (x and

y) and 0.1 mrad ()

Algorithm is fast (few minutes on a single CPU)

Code integrated into LHCb software. MC tests made with different misaligned geometries.


STEP 2 results also within LHCb requirements



VELO : testbeam results (Nov.06)

The testbeam setup

10 modules installed

4 configurations (6 modules cabled) tested




VELO : testbeam results

Track residuals

Before alignment

After alignment

• +26% vertices in target 1• +10% vertices in target 2

1 2

Vertexing




Tracking system alignment

Use the same method as the VELO (global fit via Millepede) via a common alignment software framework (currently under development).

Interface with Millepede is more complex than in the VELO, due to different track shapes (parabolas inst. of straight tracks). On the other hand detectors are not moving, alignment might be less frequently processed.


IT and OT are internally aligned separately, then w.r.t. each other using the overlap areas.

Work is ongoing. Results expected soon.



RICH alignment : principle


: Expected Cherenkov angle

: Measured angle

: Distortion due to mirror tilts

Mirror tilt



RICH alignment : results


Fit those distributions for all the mirrors combinations in order to get

their individual orientation (tilt around X and Y axis).

Alignment code implemented

Minimization using MINUIT

Measured - expected

x

0.1 mrad resolution obtained, well within

requirements

RICH 2



Global alignment


Most critical step is tracking system alignment:

VELO to T-stations (IT & OT) TT to VELO/IT/OT

Strategy for step has been defined (match tracks fitted independently in both tracking systems) and successfully tested on MC. Has to be extended to step

Work is ongoing.

1. Introduction


3. Conclusions



LHCb has been designed to hunt New Physics sign in the heavy flavours sector.

But in order to reach our objectives, a perfect understanding of the detector will be necessary.

In particular, a very good alignment is required (for trigger, particle ID, tracking,…)



LHCb alignment strategy has been presented (available in CERN-LHCb-2006-035 ). It has to take into account LHCb unique specificities (RICH, moving VELO) and requirements (online vertex trigger)

Work is ongoing on many fronts, and some nice results have already been obtained (VELO testbeam alignment, RICH alignment,…). A common framework is taking shape.

We are now waiting the first beams (as we can’t play with cosmics )

Backup Slides

LHC StartupA0 S. Viret

Align with what ?

Alignment algorithm feeding has to be taken seriously !

First alignment will be determined using magnet OFF data (very important for tracking systems).

Then this first alignment will be updated with magnet ON data.

Specific trigger scheme for those events necessary after 2008 (work ongoing).

Extensive use of specific types of tracks* (beam halo/ beam gas), mass resonances (e.g. J/),…

* No cosmics in LHCb.


Methodology for the tests:

200 runs of 25000 events (5000 min.bias + 20000 pseudo-halo) were passed trough LHCb software with the following misalignments scales (all 6 degrees of freedom are taken into account at each level):

Misaligned events are produced via LHCb geometry framework. No momentum cut applied for track selection (try to rely on VELO information only)

Translations (in m)

(x, y, z)

Rotations (in mrad)

(,,)

Module 30 2

Box 100 1Misalignment scales chosen using misalignment studies and hardware information


The huge matrix we had to invert is very sparse:

Inversion by partitioning (implemented in V.Blobel’s MILLEPEDE algorithm), reduces the problem to a Nglobal x Nglobal matrix inversion !!!

As Nglobal 100 for the VELO, the problem could be solved in few sec. !!!

kCkglobal Hk

HkT

k

=0

0

Cklocal 00

0

0 …

…

……

……

kwkxk

kwkk

… ………

Nglobal Nlocal x Ntracks


How to linearize the system ?

Millepede is interesting, but linearity is a key point. Obviously VELO sensors R/ geometry is not the most linear thing in the world…

Could consider module as a rigid object and thus transform (R/) into (X,Y) point. Module is then the basic

detector element to align.

R

But R and sensors are precisely bonded together within a module (~10m precision), and also precisely surveyed (~ few m precision).


VELO : MC results (STEP 2 : boxes alignment)

PV

Overlap

Results obtained with limited statistic ( ~1500 PV’s and ~300 overlap tracks ) :

offset (PV) = 17m tilt (PV) = 99rad offset (Overlap) = 13mtilt (Overlap) = 40rad

Results could still be improved but are already well within trigger requirements.

Track fit: bi-directional Kalman fit Tracking efficiency (p>5GeV) ~94% (ghost rate

~16%) Proper time resolution ~ 40 fs B Mass resolution ~ 15 MeV

125.6 mm


p/pMomentum Resolution

p [GeV]

= 14.8m+30.4

m/p t (ip)

Impact Parameter Resolution

pT


Tracking performance:


Date post:	15-Jan-2016
Category:	Documents
Upload:	gaye
View:	40 times
Download:	0 times

LHCb Alignment

Documents