Momentum Reconstruction and Triggering in the ATLAS Detector

1
Momentum Reconstruction and Triggering in the
ATLAS Detector

• FermiLab, October 2000
• Erez Etzion1,
• Gideon Dror2, David Horn1, Halina Abramowicz1
• 1. Tel-Aviv University, Tel Aviv, Israel.
• 2. The Academic College of Tel-Aviv-Yaffo, Tel
  Aviv, Israel.

2
LHC _at_ CERN
3
ATLAS
S.C Solenoid
Hadron Calorimeter
S.C Air core Toroids
Inner Detector
EM Calorimeters
Muon Detectors
4
Typical ATLAS collision
4107 bunch crossing per second23 events per
bunch crossing1Mbyte per event Data rate 1015
Byte/s
5
Experimental setup
calorimeter
beam pipe
TGC
6
Trigger Chambers

• Trigger goal
• selecting 100 interesting events per second out
  of 1000 million others

7
LowPt High Pt trigger
8
Network architecture
linear output

sigmoid hidden layers
input
parameters of straight track of muon
9
Training

• Training is performed using Levenberg-Marquardt
  method.
• Early stopping methods are used (validation set /
  bayesian regularization).
• Architectures varying in the number of neurons in
  first and second layers.

10
Testing network performance
Training set 2500 events. In one octant. Test
set of 1829 events. Distribution of network
errors - approximately gaussian. compatible
with stochasticity of the data. charge is
discrete!!! 95.8 correct sign.
11
Relative error of PT vs. pseudorapidity
Small pseudorapidity - larger widths. The effect
is due to smaller magnetic field and larger
inhomogeneities
12
Network mean charge error
Larger errors in charge at high momentum
. (infinite momentum tracks do not curve!)
13
Triggering by the network
Final task discriminate between background (PTlt6
GeV) and candidates for further processing (PTgt6
GeV)

• by PT estimating network.
• by a network specifically trained for
  classification.

14
Triggering performance
Errors in event classification
PT estimating network
classification network
15
New Developments

• New preprocessing, replacing the neural
  networks/Hough transform with a simple (though
  very efficient) heuristic simple straight line
  LMS fitting. Omit Too far hits, and refit.
  Tracks with too many omitted hits are rejected.
• New training retrained with larger sample and
  better over fitting control (Use standard early
  stopping technique, using a validation set).
• The results do not change significantly but they
  are more robust.

16
Summary discussion

• The network can successfully estimate the charge
  and transverse momentum of the muon.
• Classification (triggering) is most efficient by
  specially trained network.
• The data is intrinsically stochastic giving rise
  to approximately gaussian errors.
• The simplicity of the network enables very fast
  hardware realization. (See presentation this
  workshop)

17
Future work

• Further optimize the architecture.
• Calculate the lower bound for network errors
  based on data stochasticity.
• Calculate triggering efficiencies in realistic
  environments.
• Use further data (TGC station, data from 1st
  level trigger).
• Realize in hardware.