1
ATLAS reconstruction software
David Rousseau- LAL/Orsay
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 20032
ATLAS detector
Straw tracker
Si tracker
Lar em calo
Lar had calo
Tile calo
Muon spectrometer
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 20033
Hno
pile-up
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 20034
Hhigh luminosity (L=10^34)
23 interactions per bunch crossing
1000 charged tracks in tracker acceptance
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 20035
Tracking
3-layers pixel Si detector (middle one missing at startup)
4 layers of stereo strip Si detectors Straw tracker (typically 30 straws on
track)More details in Alessia Tricomi’s talk
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 20036
Impact parameter resolution 1/pT resolution
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 20037
B-jet tagging
Impact-parameter of tracks combined in likelihood-ratio
Both transverse and z impact parameter used
Severe requirements on track quality (e.g innermost pixel cluster inambiguous)
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 20038
WH event
H bbWl lM(H)=120
GeV/c2
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 20039
u jet rejection @ b=60%
More material:IP resolution degradationMore fake high IP tracks
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200310
Impact parameter resolution of b-tracks degraded by multiple scattering
-More non-b tracksMore fake high IP tracksB-track narrow jet
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200311
ttH event
H bb tt W(l l )bW(qq)b
M(H)=120 GeV/c2
Same b tagging performance (for given pseudo-rapidity and Pt) factorization hypothesis holds
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200312
b tagging in Heavy Ion collision
Heavy Ion physics was not even considered when Atlas was designed
~10.000 tracks in tracker acceptance (10 times more than high luminosity pile-up)
But track density comparable to density in high Pt jets…soo with little extra work on tracking
algorithm, tracks can be foundo b tagging even doable:
Ru=35 @ b=50% , 11 @ b=60%
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200313
Electron bremsstrahlung:need special tracking
E/ identification
High pT (>7GeV): o Find e.m cluster (sliding window)o E.m shower shape cutso Track findingo 0 track: photon (typical jet rejection 2000)o 1 track: E/p matching + Transition Radiation hits
counting: electron (typical jet rejection 100000)
Low pT electron (b tagging, J/): o Start from the track and
search energy deposition in e.m. calorimeter
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200314
Soft electron ID
Tagging variables
epi
# of TR hits
diff betweenshower and impact position
shower isolation
fraction of E in 1st sampling
energy weighted width
Et(calo)/pt
transverseimpact parameter
fraction of E in 3rd sampling
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200315
Soft electron ID
Electron in b jets Hbb (mH=120 GeV)
without noise
with noise
Efficiency
Pio
n re
ject
ion
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200316
E/ reconstruction
E.m. clusters reconstructed with sliding window algorithm
Rectangle clusters usedo Robust against
Electronic noise Pile-up Underlying event Material effects
o Calibration/linearity
Typically 3 eta cells x 5 phi cells o Cell granularity 0.025 pseudo rapidity/radiano Larger in phi to accomodate B field
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200317
Em cluster reconstruction
E measured vs conversion radius
*
Conversionscan be reconstructed
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200318
pointing
PV
strips
middle
O z
1
2
direction
presampler
LHC: luminous region z~5cm
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200319
Mass resolution H
Primary vertex from tracks
=1.31 GeV=1.18 GeV
No pile-up, no el. noise
1 conv
mH (GeV)
mH (GeV)
Primary vertex from calo pointing
Primary vertex finding: not always possible with low Pt(H) and high lumi
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200320
Intermezzo… ATLAS reconstruction has evolved from feedback
on detector design to evaluation of detector performance and physics reach and recently:o foresee treatment of real Atlas datao migration fortran to C++ completedo keep on improving or new algorithms
Use Athena/Gaudi (with LHCb) flexible framework:o Separation between Data and Algorithmso Run-time configuration and dynamic library loading
Athena also used for MC generation, Geant4 simulation (coming), high level trigger, fast simulation, user analysis
Spring 2003 : Data Challenge 1, several millions event simulated (G3) and reconstructed world-wide (tens Terabytes of data)
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200321
Flow example
Fast DigiSimulation
MCHits
Detailed Digitization
RawByteStream
LArCellBS decoding Calo Clustering
Read Out Driveremulation
RawChannels First data reduction
TileCell
Missing ET
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200322
Work modelCirca 1500 C++ classes in 300 packages
maintained/developed by 100 people (only a few CERN based)
Structured CVS repository with one directory per detector specific software (Muon, Larg calorimeter…), and one per activity (Reconstruction, DetectorDescription…). Each directory managed by responsible person.
One major release every ~6 monthsDeveloper release every three weeks (1-2
iterations allowed, some failure allowed)Automatic nightly builds with latest tagged
version of all software in view of the following release.
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200323
Jet reconstruction
Several algorithms exploredo Cone jet (seeded/unseeded + split/merge)o kT jet
Weighting technique:o w= a + b/E + c ln(E)
o Minimize resolution under constraint E=Etrue
o Bins in pseudo-rapidity because effective length of calorimeters is non uniform (mainly 1/sin() but also passive material distribution)
o « 2 » weights: one for e.m calo one for had caloo « 7 » weights : exploit calorimeters longitudinal
segmentations
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200324
Jet r
elat
ive
ener
gy r
esol
utio
n
E/E=100%/E 2%
E/E
true
E/E=80%/E 2%
E/E=65%/E 2%
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200325
Missing ET
Missing transverse energy essential tool for a wide range of physics
Computed from sum of cell transverse energy with optimised weights
Calorimeter acceptance very important: -5<pseudo-rapidity<5
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200326
Missing ET resolution
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200327
bbA m(A)=450 GeV Z
No noise
With noiseAsymetric cutE>2E
Z/A/H 1 2 X1 1 X2 2
Assume the two neutrinos are colinear to visible tau decays
System solvable using Missing ET measurement
Mass reco with missing ET
Mass resolution (GeV) Mass resolution (GeV)
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200328
Tau identification Especially important for A+- and
numerous supersymmetry channels Hadronic tau decay characterised as
a very narrow jet in particular:o use very fine granularity of first layer of
Lar em calorimeter : =0.003o reconstructed track counting
Strong pT dependence: at higher pT, tau jet is narrower, background jet fatter
Correlated tagging variables combined in bins of jet pT
(measured with weights optimised for tau’s ) Depending of analysis, different working points in
Rejection vs Efficiency plane
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200329
Tau Tagging variables
jet
Radius of em cluster Energy within 0.1<R<0.2 Number of charged tracks
Cluster eta width in em calo first layer
Number of em calo strip with energy abovethreshold
Charged tracks total charge
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200330
Tau ID efficiency
A+-l+h-A+-h+ h-
t H+b (h)b
0.5
susy searches
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200331
1. Identification of Region of Activity
2. Reconstruction of local straight track segments
3. Combination of three tracks segments
4. Global fit taking into account multiple scattering and energy loss
Muon reconstruction
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200332
S.Hassani Athens 05/03
A.Farilla Gallipoli 06/03
Muon backtracking
Backtracking from Muon System down to beam region through calorimeters taking into account E loss, multiple scattering and E loss fluctuations
E loss from parametrization (from calo measurement possible but risk of pollution from nearby particle)
Combination with inner detector track
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200333
pT(GeV)
•Pattern recognition perturbed by possible em shower accompagning high pT muon•Under study: Mu System pattern recognition redone using Inner Detector measurement
High pT efficiency
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200334
As pT decreases, the energy lost by the in calorimeters becomes comparable to its
energy, especially in the barrel
pT(GeV)
Inner Station Segments Tracks
•Under study :use of the Inner Station Segments could improve the identification efficiency up to 90%
Low pT efficiency
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200335
m(GeV)
J/→+-
m(GeV)
Z→ + -
H → Z Z→ +- + -
reconstruction Mass plots
David Rousseau, ATLAS reconstruction, HEP2003, Aachen, July 200336
OutlookA complete spectrum of algorithms are
availableOngoing developments:
o Cleaner modularization (toolbox)o Robustness (noisy/dead channels, misalignments)o Extend algorithms reach (e.g low pt, very high pt)o New algorithms
Next challenge: summer 2004, an ATLAS barrel wedge with all detectors in testbeam. Reconstruction and analysis using (almost) only atlas offline reconstruction.
Many thanks to my ATLAS colleagues, in particular for this talk: Nektarios Benekos, Frédéric Derue, Ambreesh Gupta, Michael Heldmann, Anna Kaczmarska, Jean-Francois Laporte, Jessica Leveque, Pavel Nevski, Frank Paige, Gilbert Poulard, Jean-Baptiste de Vivie, Silvia Resconi, Francesco Tartarelli, Monika Wielers