fast simulation for atlas: atlfast-ii and...
TRANSCRIPT
Fast simulation for ATLAS: Atlfast-II and ISF
Wolfgang LUKAS*, University of Innsbruck (Austria) / CERN, for the ATLAS CollaborationA.Salzburger, M. Duehrssen, E. Ritsch, R.D. Harrington, A. Schaelicke, F.M. Garay Walls, P. Sherwood, J. Chapman
CHEP 2012 Poster Session, 21-25 May 2012, New York (USA)
Each particle processed by the ISF is passed by the ParticleBroker through configurable routing chains for the current sub-detector. The particle is associated to a simulator according to the first filter rule that applies to its current properties.
Prior to the simulation launch, users canconfigure these filter rules according to their specific physics analysis requirements. They can find an optimal balance between precision and execution time by selecting faster simulation flavors for all parts that do not require full detail.
The Integrated Simulation Framework (ISF)The fast simulation Atlfast-II with the calorimeter simulation FastCaloSim has been developed in order to reduce the simulation time in the ATLAS calorimeter system. It can be tuned against data, has been validated against the Geant4 based full simulation and has been used since 2011 for large-scale MC production.
In order to meet the increasing physics demands on MC samples, the Integrated Simulation Framework (ISF) is being developed. It allows the flexible combination of full and fast simulation strategies in a single event to provide an optimal balance between precision and execution time, depending on the required accuracy.
The ISF is expected to be ready for production in 2013/2014 and will be used for all simulations and large-scale MC productions from 2015 onwards, when the LHC is expected to deliver a much higher integrated luminosity than today.
Conclusion
[1] The ATLAS Collaboration, The ATLAS Simulation Infrastructure, Eur. Phys. J. C 70 (2010) 823–874
[2] S. Agostinelli et al., Geant4 - A Simulation Toolkit, Nucl. Instr. Methods Phys. Res. A 506 (2003) 250–303
[3] Jörg Mechnich (on behalf of the ATLAS Collaboration), FATRAS - the ATLAS Fast Track Simulation project, J. Phys.: Conf. Ser. 331 (2011) 032046
References
The ISF is fully embedded in the ATLAS framework (Gaudi-Athena). The general simulation flow is steered by the SimKernel, which is a single algorithm that holds Simulator services for the sub-detectors, a ParticleBroker service and the truth service. The SimKernel retrieves particles from the ParticleBroker and routes them to their associated simulation engines. All simulators fill a common set of hit collections which are then processed in the same digitization and reconstruction chain.
Simulation Requirements for the ATLAS Detector MC samples vs. dataMonte Carlo simulations of physics events, including detailed detector simulation, are indispensable for every analysis of high energy physics experiments. Increasing the recorded luminosity at the LHC, and hence the amount of ATLAS data to be analyzed, leads to a steadily rising demand for simulated MC statistics. These MC requirements for more refined physics analyses can only be met through the implementation of fast simulation strategies which enable faster production of large MC samples. [1]Timing Issues
Full and Fast G4 SimulationThe total CPU time spent for the simulation of a typical t t event (seetable above) differs by a factor ~3 between the Full Simulation with G4 and the Fast G4 Simulation, which uses pre-calculated “frozen shower” libraries for the electro-magnetic particles in the forward calorimeter.
Geant4 provides the most accurate simulation, which is essential for several but not all parts of physics analyses. The pie chart below shows the typical fraction of full simulation CPU time spent in each subdetector.
speed
accuracy
1 %
95 %4 %
IDCaloMuon
Atlfast-II and FastCaloSim Atlfast-IIF and FatrasThe simulation time can be reduced by more than an order of magnitude by using the Atlfast-II fast simulation, which uses FastCaloSim in the calorimeter. The energy of single particle showers is deposited directly using parameterizations of the longitudinal and lateral energy profile. It is intrinsically less accurate, but can be tuned against data.
This fast simulation approach has been used since 2011 for the production of large MC samples needed for new physics searches as well as precision measurements. It has been validated against the Geant4 based full simulation for electrons, jets and missing ET. Validation plots of the calorimeter shower shapes of high Et electrons are below.
speed
accuracyAnother factor of ~10 in CPU time can be gained in Atlfast-IIF using the Fast ATLAS Tracking Simulation (Fatras) for the Inner Detector and Muon System. Fatras uses simplified physics parameterizations and thesimpler reconstruction geometry instead of the full geometry [3]. Fatras can be tuned against data and is useful for validation studies, fast material calibration and rapid large-scale MC production.
speed
accuracy
Many studies require large MC samples, but are interested in high precision simulation only for certain particles and regions. These current and future physics demands can only be supported by an Integrated Simulation Framework that flexibly combines different simulation strategies.
The new Integrated Simulation Framework is based on the requirement to allow to run all simulation types in the same job, even within the same sub-detector, for different particles.
This framework is designed to be extensible to new simulation types as well as the application of parallel computing techniques in the future.
x10 x100
Energy ratio Rη in a Δη*Δφ = 3x7 cells cluster with respect to a 7x7 cells cluster size in the bulk EM calorimeter layer 2. Pixel hits in η with Fatras and 900 GeV collision data.
Visualization of an ISF event with VP1.
kSI2
K s
econ
dspe
r eve
nt
R0.9 0.92 0.94 0.96 0.98 1
Entri
es /
0.00
5
0
200
400
600
800
1000
1200
1400
1600
R0.9 0.92 0.94 0.96 0.98 1
Entri
es /
0.00
5
0
200
400
600
800
1000
1200
1400
1600
eeZDataG4.9.2G4.9.4, new geo.AFII
ATLAS Preliminary=7 TeV,sData 2010, -140 pbtdL
R0.9 0.92 0.94 0.96 0.98 1
Entri
es /
0.00
5
0
200
400
600
800
1000
1200
1400
1600
eeZDataG4.9.2G4.9.4, new geo.AFII
stotw0.5 1 1.5 2 2.5 3 3.5 4
Entri
es /
0.03
0
500
1000
1500
2000
2500
3000
stotw0.5 1 1.5 2 2.5 3 3.5 4
Entri
es /
0.03
0
500
1000
1500
2000
2500
3000
eeZDataG4.9.2G4.9.4, new geo.AFII
ATLAS Preliminary=7 TeV,sData 2010, -140 pbtdL
stotw0.5 1 1.5 2 2.5 3 3.5 4
Entri
es /
0.03
0
500
1000
1500
2000
2500
3000
eeZDataG4.9.2G4.9.4, new geo.AFII
For many physics analyses, like the W mass precision measurement, large MC samples are required to study systematics and background processes. The full Geant4 simulation [2] can become too time-consuming for this purpose.
1.9x109
15 fb-1
expected
2.0*109
Atlfast-II
Full Sim (G4)
data 1.6x109
2.8x1091.4x109
5 fb-1
0.05 fb-1
MC data MC data MC data
2010 2011 2012
ATLAS simulation, 2010
ISF estimated
pxp
y
A
B
CA B C
track
pixel position (vetoed)( )
intersection with surface
exit of sensor material
cluster position
(a) Schematic display of the pixel cluster creationin FATRAS
(b) Mean size of pixel clusters versus the η of theassociated track
Figure 4: (a) In FATRAS, pixel clusters are created by calculating the relative path length ofa track to a pixel volume and counting all pixels as hits where this quantity passes a tunablethreshold. (b) Comparison of the mean cluster size in the ATLAS pixel detector in 900 GeVcollision data (black points) and MC simulated with FATRAS (shaded histogram).
(a) Number of pixels hits versus η (b) Number of pixels hits versus φ
Figure 5: Comparison of the geometric distribution of pixel detector hits in η (a) and φ (b)in 900 GeV collision data (black points) and MC simulated with FATRAS (shaded histogram).The distribution is shaped by the existence of inactive pixel modules that are taken into accountby FATRAS.
the tails of the distributions.
7. SummarySince the development of FATRAS was started, it has proven to be a useful tool, not only fordebugging the track reconstruction algorithms and the simplified reconstruction geometry ofthe ATLAS detector, but also as a fast simulation engine. Comparisons with real collision datashow that the description of the physics processes and the material distribution are modeled ina realistic way. The speed increase with respect to a detailed detector simulation which alsouses a much more complex description of the detector is significant. This implies that FATRASis a perfect tool for investigating questions that are related to tracking and also for simulating
Shower width Wstot determined in a window corresponding to the cluster size in the high granularity strip layer 1.
mixing of different simulators(G4, FastCaloSim, Fatras)
in one ISF event
Day in 201228/03 04/04 11/04 18/04 25/04 02/05 09/05
]-1To
tal I
nteg
rate
d Lu
min
osity
[fb
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6 = 8 TeVs ATLAS Online Luminosity
LHC Delivered
ATLAS Recorded
-1Total Delivered: 1.23 fb-1Total Recorded: 1.14 fb