1. Parameterisation of EM showers in the ATLAS LAr Calorimeter
Tom Atkinson - The University of Melbourne
On behalf of
Elisabetta Barberio - The University of Melbourne
Anthony Waugh - The University of Sydney

2. Introduction
Fast Simulation of ATLAS LAr calorimeter implemented in package:
LArCalorimeter/LArG4/LArG4FastSimulation
Class LArFastShower derived from G4 class FastSimModel
Depends on LArCalorimeter/LArG4/LArG4Barrel
EMBParticleBounds
EMBShowerParameters
Simulation testing done with ATHENA
G4AtlasApps
New tag LArG4FastSimulation
Hacked version of LArG4Barrel/src/EMBParticleBounds.cc
NB: New tag produced LArG4Barrel but not backwards compatible against

3. Original simulation scheme
Timing performance previously significantly worse in full ATLAS simulation vs stand-alone reconstruction.
Original simulation scheme:
Energy range
e+ / e-
Photons
0.0GeV < E < 0.1GeV
Full sim
0.1GeV < E < 0.5GeV
0.5GeV < E < 100GeV
Kill + Fast param
This scheme was fine for stand-alone testing of the calorimeter - no showering!

4. Original simulation scheme - Timing results
But showering occurs in ATLAS when particles generated at nominal interaction point.
Large numbers of low energy shower particles also entering the calorimeter
Result is a large increase in execution time.
ATLAS scan range -0.8 <  < 0.8 (NB: Full simulation in crack)
100 events per sample
Energy
Fast Param (CALO)
Fast Param (ATLAS)
10GeV
0.13 sec
2.2 sec
50GeV
0.64 sec
5.3 sec
100GeV
1.08 sec
9.4 sec
Average time / event

5. Timing vs.  - original scheme

6. New simulation scheme
This scheme is not final. Cuts on “kill” limits will change as we optimise performance vs. timing…
Energy range
e+ / e-
Photons
0.0GeV < E < 0.1GeV
Kill
0.1GeV < E < 0.5GeV
Full sim
0.5GeV < E < 100GeV
Kill + Fast param

7. New simulation scheme - Timing results
Electrons generated at nominal vertex
-0.8 <  < 0.8 (NB: Full simulation in crack)
100 events per sample
Energy
Full Simulation
Original scheme
( CALO Only)
New scheme
10GeV
3.7 sec
2.2 sec
0.13 sec
0.6 sec
50GeV
16.9 sec
5.3 sec
0.64 sec
1.3 sec
100GeV
34.5 sec
9.4 sec
1.08 sec
2.1 sec
Average time / event

8. Number of Parameterisation method calls
Energy
IsApplicable()
ModelTrigger()
DoIt()
10GeV
7400
200
48000
1100 4
100GeV
120000
1500
470000
8500 22
Original scheme / new scheme

9. Timing tests in the ATLAS endcap
Electrons generated at nominal vertex
1.5 <  < 2.5
New particle simulation scheme used
Energy
Full Simulation
Fast Parameterisation
10GeV
7.8 sec
3.3 sec
50GeV
38.2 sec
14.5 sec
100GeV
 95 sec
23 sec
Average time / event
Under investigation - more complex geometry!

10. For discussion… A word on the direction of electrons…
LArFastShower::ModelTrigger() Get Momentum Direction:( , , )
LArFastShower::ElectronDoIt() Get Momentum Direction: ( , , )
Change of direction between ModelTrigger and DoIt seen for electrons ONLY
(Positrons and photons seem to be ok)
Consistent with effects seen by CMS
Does anybody know what causes this?

11. Conclusion
Significant improvements in timing when low energy electrons and photons from the shower are “killed”.
Impact on physics performance of new scheme is currently beginning.
“Kill” cuts will be set to optimise performance vs. timing
The End…