FATRAS. A Novel Fast Track Simulation Engine for the ATLAS Experiment. Sebastian Fleischmann on behalf of the ATLAS Collaboration

Transcription

1 A Novel Fast Track Engine for the ATLAS Experiment on behalf of the ATLAS Collaboration Physikalisches Institut University of Bonn February 26th 2010 ACAT 2010 Jaipur, India

2 1 The ATLAS detector Fast detector simulations Why? 2 Track simulation in Digitisation Combined simulation: Tracking and calorimetry Comparison to other simulation 3 Comparison with full simulation Comparison with collision data Special use-cases and applications 4 22

3 The ATLAS detector Muon Spectrometer Hadronic Calorimeter Electromagnetic Calorimeter Solenoid 3 Tracking System TRT Pixel/SCT

4 The ATLAS Inner Detector TRT drift tubes straw tubes 130 µm resolution (Rφ) Xe/CO 2 /O 2 about 30 measurements per track SCT silicon strips 6 million Si strips resolution: 17 µm (Rφ)/580 µm (Z) 4 (double) measurements / track silicon Pixel 80 million Si pixels resolution: 10 µm (Rφ)/115 µm (Z) 3 measurements / track 2 T solenoidal field 44

5 Why do we need fast detector simulations? Monte Carlo simulation of detector response is needed to compare theoretical predictions (by Monte Carlo event generators) to data Detailed simulation of particles penetrating the detector material is CPU-time consuming of a single t t event in full Geant4 simulation takes about 30 ksi2kminutes Fast simulation techniques can increase the amount of simulated events 55

6 Track simulation 4-momentum, particle type (PDG ID) si standard high energy physics libraries (e.g. PYTHIA) Geant4 Detector, Full physics list Digitisation Reconstruction c Track Material effects Particle decay Photon conversions Digitisation ri ATLFAST Track represen h smearing Track detector simulation simulates passage of particles through the detector, charge deposition in sensitive detectors, showers, material interactions digitisation emulates readout reconstruction pattern recognition, track finding, track fitting 66

7 The simulation scheme of uses the component model of the ATLAS software take extrapolation engine of the ATLAS track reconstruction reconstruction modules, such as the estimation of energy loss, are replaced by Monte Carlo implementations event data objects identical to full simulation or real data Where feasible, Geant 4 modules are used, such as particle decays Nearly all effects are estimated from first principles (Bethe-Bloch, etc.) no parametrisations used, despite hadronic interactions 77

8 ATLAS Inner Detector Geometry Geant 4 r [mm] Tracking Geometry r [mm] Geant z [mm] 200 TrackingGeometry z [mm] Track reconstruction in ATLAS uses greatly simplified detector description Sensitive elements identical to full Geant 4 description! uses the same! 88

9 Track simulation in MC Event Generator 99

10 Track simulation in MC Event Generator µ 99

11 Track simulation in MC Event Generator K 0 S µ 99

12 Track simulation in MC Event Generator e K 0 S µ 99

13 Track simulation in MC Event Generator π + e K 0 S µ 99

14 Track simulation in MC Event Generator γ π + e K 0 S µ 99

15 Track simulation in MC Event Generator e γ π + e K 0 S µ 99

16 Track simulation in MC Event Generator γ π + e KS 0 µ e Particle unstable? Simulate decay length 99

17 Track simulation in MC Event Generator γ π + e KS 0 µ Extrapolation to next surface Particle unstable? Simulate decay length 99

18 Track simulation in MC Event Generator γ γ secondaries, brem photons, conversion electrons Material Effects Extrapolation to next surface π + e K 0 S µ Particle unstable? Simulate decay length 99

19 Track simulation in MC Event Generator γ γ secondaries, brem photons, conversion electrons Material Effects Extrapolation to next surface Decay point reached? π + e K 0 S µ Particle unstable? Simulate decay length 99

20 Track simulation in MC Event Generator γ γ secondaries, brem photons, conversion electrons Material Effects Extrapolation to next surface no Decay point reached? π + e K 0 S µ Particle unstable? Simulate decay length 99

21 Track simulation in MC Event Generator γ γ secondaries, brem photons, conversion electrons Material Effects Extrapolation to next surface no Decay point reached? π + e K 0 S µ Particle unstable? Simulate decay length 99

22 Track simulation in MC Event Generator γ γ secondaries, brem photons, conversion electrons Material Effects Extrapolation to next surface no Decay point reached? π + e K 0 S µ Particle unstable? Simulate decay length 99

23 Track simulation in MC Event Generator γ γ π + secondaries, brem photons, conversion electrons Material Effects Extrapolation to next surface no Decay point reached? yes Create decay products e K 0 S µ Particle unstable? Simulate decay length 99

24 Track simulation in Postprocessing 10 10

25 Track simulation in Postprocessing Extract measurements from simulated tracks 10 10

26 Track simulation in Postprocessing Extract measurements from simulated tracks 10 10

27 Track simulation in Postprocessing Extract measurements from simulated tracks Add noise

28 Track simulation in Postprocessing Extract measurements from simulated tracks Add noise and merge clusters

29 Digitisation in Silicon clusterisation with geometric approach d track intersection with surface exit of sensor material ( ) pixel position (vetoed) cluster position A B C p y p x A B C Without Lorentz angle: Q 1 Q 2 Q 3 ixeldetector With Lorentz angle: h of the track Q1 Trackel is o the charge Q2 Track θl u θl Lorentz Angle ring tion is random landau mbers shift to θl

30 Combined and FastCaloSim 0 Y (cm) 100 Calo cells and FastCaloSim interfaced: Conversion point Electron tracks LAr Calorimeter Presampler Brem point Solenoid coil Elect simulates the Inner Detector including secondaries Calorimeter deposits simulated by FastCaloSim Muon System simulated by 0 X (cm) 12 12

31 The Fast and the Furious Atlfast I Atlfast II Geant4/full Atlfast IIF Atlfast II ID track per FatrasID digitisatio reconstruc digitisation reconstructio Calo cluster FastCaloS muons: fu digitisation reconstructio MS track perig FatrasM digitisatio reconstruc digitisation reconstructio rel. gain

32 Material effects Comparison of and Geant Multiple scattering Energy loss 10 Geant4 2 GeV Single µ - Gaussian traversing 1mm (Highland) Si 1 (~1.06 Gaussian % X Mixture 0) (Frühwirth et al.) Radiation of brem photons Entries/bin Hadronic interactions Entries Geant4 Mean Entries Mean Entries/bin p [GeV] 100 Energy 120of secondary particles

33 Digitisation in Tuning of the silicon clusterisation Pixels and SCT clusterisation tuned by adapting minimal required path length in the Pixels cell strength of Landau smearing Tuning can be done within 24h Cluster width in Pixels Measurement residual in Pixels

34 Track parameter resolutions Single muon events with p T =1 GeV, 5 GeV, 100 GeV In general good agreement, but still some parameters to tune in the digitisation In particular tails better described than in ultra-fast sim Entries/bin offline/newt 3 ATLFAST GeV 5 GeV 100 GeV Geant 4 1 GeV 5 GeV 100 GeV rec 1 d true -d (mm) 0 0 rec true d 0 - d 0 [mm] d 0 -d 0 (mm) Transverse impact parameter d 0 Inverse track momentum q/p T

35 Reconstructed tracks in minimum bias events at s = 900 GeV on minimum bias Monte Carlo (Pythia) at the center of mass energy s = 900 GeV Still some discrepancies, but not yet tuned to data and misalignments of detector modules in data 17 17

36 Reconstructed tracks in minimum bias events Average number of Pixels hits per track Pixels hits / track vs η Average Number of Pi ATLAS Preliminary Fatras sinoidal structure due to inactive Pixels modules in b-layer folded with the z-position of the primary vertex Pixels hits / track vs φ Average Number of Pixe Fatras ATLAS Preliminary Detector conditions like inactive modules automatically taken into account in Precise description of the detector geometry

37 Reconstructed tracks in minimum bias events at s = 900GeV Pixel cluster size mean cluster size (Δ η Q 1 Fatras Without Lorentz angle: Q 2 Q 3 u ATLAS Preliminary θ L Track incident angl Cluster size depends on incident angle, because of sensor thickness With Lorentz angle: ing Q 1 Q 2 Track ion is random landau mbers θ L shift to Track θ L Lorentz Angle Mean cluster size (in η direction) vs incident angle (η) on the Pixel module Very sensitive test of the clusterisation model 19 19

38 Reconstructed tracks in minimum bias events at s = 900GeV Transverse impact parameter wrt. primary vertex Arbitrary Fatras ATLAS Preliminary d 0 [m Longitudinal impact parameter Arbitrary 10 Fatras ATLAS Preliminary Position and size of the beam spot in the simulation taken from detector conditions data base 20 20

39 for Super-LHC upgrade studies Occupancy Pix Pix Pix Pix 0.2 was used for SLHC upgrade studies of the ATLAS tracker Allows easy testing of various geometries at a reasonable time scale Detector occupancies can be derived reliably Reconstruction effects are included in momentum resolutions, etc.

40 Stress tests for track fitters simulation with high noise levels Detailed truth information by allows to evaluate performance of track fitters Quick simulation of arbitrary noise levels ) [MeV] T σ(p Kalman DAF Silicon-only TRT noise occupany GeV single muon events Example: Study of adaptive track fitter (Deterministic Annealing Filter) High detector occupancy Solution of left-right ambiguities in the TRT

41 Conclusions is a new track simulation concept between full Geant4 simulation and conventional fast detector simulations The full reconstruction chain can be run on output Speed improvement mostly due to simplified Tracking Geometry and extrapolation (Nearly) no parametrisations needed All important physics effects included, like multiple scattering, brem, conversions, particle decays, hadronic interactions Allows studies to be performed that cannot easily be done either with full simulation or conventional fast simulations Currently in the tuning phase Validation with collision data has started 23 23

42 Acknowledgements Special thanks to Andreas Salzburger (CERN), Sharka Todorova (Tufts University) and Simone Zimmermann (Bonn) and PHYSICS AT THE TERA SCALE Helmholtz Alliance 24 24

43 25 25

44 The ATLAS Muon System Monitored Drift Tubes (MDT) 354k straw tubes barrel and forward region 80 µm straw resolution (Z) 20 measurements / track Resistive Plate Chambers (RPC) barrel region chamber resolution: 10 mm (Z)/10 mm (φ) 6 measurements / track trigger (+ 2 nd coordinate) Thin Gap Chambers (TGC) end-cap chamber resolution: 2 6 mm (R)/3 7 mm (φ) 9 measurements / track trigger (+ 2 nd coordinate) barrel toroid: Tm bending power, end-cap toroids: Tm Cathode-Strip Chambers (CSC) forward region multi-wire prop. chambers plane resolution: 60µm (R)/5mm (φ) 4 measurements / track Muons with momenta of 4 GeV and 20 GeV in the bending plane of the barrel muon spectrometer.

45 The green simulation time/event, ksi2kseconds Minimum Bias t t Jets W ± e ± ν e Heavy Ion Full Sim , 000 Fast G4 Sim , 700 ATLFAST-II ATLFAST-IIF ATLFAST-I Hard to estimate how much CO 2 is emitted for one CPU second by the grid Taking 0.01 grams/s: Simulating 100k t t events with ATLFAST-IIF ( + FastCaloSim) instead of Full saves about 2 tons of CO

46 Digitisation in of hits with transition radiation in the Transition Radiation Tracker transition radiation in the TRT produces hits with stronger signal and is used for particle ID probability of transition radiation depends on relativistic γ factor measured with test beams and cosmic ray data fit of turn-on curve has been fed into γ

47 Combined simulation Comparison of reconstructed muon p T resolution T central muons ( η < 1.2) in Z µ + µ events Muon System stand-alone Resolution = σ Muon p A F F Combined Muon Spectrometer / Inner Detector reconstruction

48 Reconstruction efficiencies Single electrons and muons with transverse momentum p T = 5 GeV Shape roughly reproduced, but electrons Needs some extra fudge factors too perfect for Efficiency