Fast simulation in ATLAS - PDF Free Download

Fast simulation in ATLAS A. Salzburger, CERN for the ATLAS collaboration 1

Simulation load & requirements typical MC campaign in ATLAS 2011 - O( 9) events << recorded data - future precision measurements will need more fast MC is needed to cope with present/future needs - higher statistics raises additional questions: I/O, disk space, reconstruction time - see talk of R. Brun on simulation R&D this afternoon ATLAS simulation, 20 2

Simulation hierarchy high CPU CONSUMPTION full library alternative/fast low parametric HIERARCHY ACCURACY 3

Simulation hierarchy high full library alternative/fast CPU CONSUMPTION event reconstruction (efficiency/fakes) low parametric HIERARCHY ACCURACY physics object creation 3

Simulation in ATLAS 20+ years of simulation full library Geant4 / Fluka,Flugg / Geant3 Frozen Showers alternative/fast parametric AFII (Atlfast2) / AFIIF (Atlfast2F ) Atlfast(1) 4

Full Simulation Geant4 full talk of J. Chapman talks at this WS geometry model: library Geant4 (translated from ATLAS GeoModel), very detailed, O( 6 ) nodes physics alternative/fast engine: Geant4 tuning possibilities: via geometry/material physics list parametric step length process cuts 5

Geant4 tuning the simulation adapting ID material (+ /20 %) investigate effects on ID tracking - to keep geometry structure done by changing material density of certain elements fitted mean ratio 0 K S 1.01 1.008 1.006 1.004 1.002 1 0.998 Minimum Bias Events ( s=900 GeV) Data / MC (nominal) ATLAS Preliminary fitted mean ratio 0 K S 1.01 1.008 1.006 1.004 1.002 1 0.998 Minimum Bias Events ( s=900 GeV) MC (%) / MC (nominal) MC (20%) / MC (nominal) ATLAS Preliminary 0.996 0.994 0.992 0.99 < 1.5 Beam Pipe Pixel Layer-0 Pixel Layer-1 Pixel Layer-2 2 SCT Layers 3 Radius [mm] 0.996 0.994 0.992 0.99 < 1.5 Beam Pipe Pixel Layer-0 Pixel Layer-1 Pixel Layer-2 2 SCT Layers 3 Radius [mm] 6

Geant4 tuning the simulation N changing the geometry description see talk of Olivier Arnaez electron shower shape dependence on geometry model, changing blended material into individual layers changing the physics list ATLAS calorimeter response to anti-neutrons (0.1 < pt < 50 GeV) 7

The library approach Frozen Showers geometry model: Geant4 (translated from GeoModel) physics engine: Geant4 (library, pre-simulated showers) tuning possibilities: limited to shower shapes in library full library alternative/fast parametric 8

Fast simulation AFII & AFIIF AFII - parametric cell response of the ATLAS calorimeter: FastCaloSim - Inner Detector & Muon System: Geant4 AFIIF full - parametric cell response of the ATLAS calorimeter: FastCaloSim - Inner Detector & MuonSystem: fast Monte Carlo track simulation (FATRAS) library alternative/fast parametric 9

Fast simulation Ways to speed up simulation approximate geometry optimise transport and navigation π 3 full approximate models library parameterisations alternative/fast ctrl C parametric V take shortcuts use new technologies

Fast simulation simplified geometry model ATLAS TrackingGeometry (a) - Inner Detector & Calorimeter: simplification to layers and cylindrical volumes keeping the exact description of sensitive elements sensors and chips (b) cooling tubes navigation through the geometry is only done using the layers and volume boundaries, modules are found by intersection with layer material is mapped onto layers using Geant4 description and geantinos thermal management tile (b) sensors and chips cooling tubes mounting socket (c) (d) t [X0] 0.05 Surface 2 Surfac e1 0.04 lx 0.03 0.02 thermal management tile (c) mounting socket 0.01 (d) t [X0] La ye Lay r1 er TrackingGeometry Geant4 module overlap 2 0-8 -6-4 -2 0 2 4 6 8 l x [mm] 0.05 11

Fast simulation simplified geometry model - Example Inner Detector: O(0) layers and detector boundaries r [mm] η=0.0 0.25 0.5 0.75 1.0 1.5 2.0 η=0.0 0.25 0.5 0.75 1.0 1.5 2.0 00 800 2.5 2.5 sensitive modules are identical 600 400 3.0 3.0 200 0 Geant4-4000 -3000-2000 -00 0 00 2000 3000 4000 z [mm] TrackingGeometry -4000-3000 -2000-00 0 00 2000 3000 4000 z [mm] - Muon System: simplification of chambers & exact transcript into TrackingGeometry classes 12

Geant4 / TrackingGeometry material comparison Geant4 material TrackingGeometry representation TrackingGeometry Geant4 example for TrackingGeometry layer 13

Fast simulation optimised transport model ATLAS TrackingGeometry is a fully connective geometry with interlinked nodes - built from one common surface class to be used with extrapolation engine - result of propagation to current node guides to next node Volume A Volume B n CB t 2 Surface AB Surface CB Surface AC t 1 n AC Volume C 14

Fast simulation optimised transport model ATLAS TrackingGeometry is a fully connective geometry with interlinked nodes - built from one common surface class to be used with extrapolation engine - result of propagation to current node guides to next node Volume A Volume B t2 Surface AB ncb Surface AC Surface CB t1 nac Volume C 5 The ATLAS TrackingGeometr 14

Fast simulation AFII (FastCaloSim) geometry model: ATLAS calorimeter reconstruction geometry physics engine: ATLAS extrapolation engine (transport) shower shape parameterisation tuning possibilities: shape modification, energy response scaling full library alternative/fast parametric ctrl C V 15

Fast simulation AFII high G4 CPU CONSUMPTION 1 ATLFAST2 x 0.1? low HIERARCHY ACCURACY 16

Fast simulation AFII - AFII setup: Step 1: Inner Detector with Geant4 Calorimeter, Muon System with Geant4 for µ, all other particles are killed at Calo entry Step 2: Calorimeter cell response from parameterisation applied to particles from ID ID Calorimeter Muon System - Shower parameterisation: parameterisation of shower shape variables, allows tuning to data distributions N arbitrary units extensive validation performed for Geant4/AFII 0.07 0.06 0.05 0.04 Single Photons: E TOT Fullsim Atlfast-II (E GEN = 20 GeV : 1.00 < η < 1.05) 0.03 0.02 0.01 0 ATLAS Preliminary Simulation 0.92 0.94 0.96 0.98 1 1.02 E TOT / E GEN 17

Fast simulation AFII comparison to Geant4 Entries Entries AFII / G4.9.2! 450 AFII / G4.9.2 400 350 300 250 200 150 0 50 Not reviewed, for internal circulation only 0 1.2 1.1 1 0.9 0.8 0.7 5 4 3 Not reviewed, for internal circulation only 3 ATLAS Preliminary Simulation G4.9.2 AFII 50 0 150 200 250 jet p [GeV] T 50 0 150 200 250 ATLAS Preliminary Simulation G4.9.2 AFII miss E T jet p T 1.1-1 0 1 2 3 4 1 1.05 0.95 SUSY signal events [GeV] E relative resolution -1 0 1 2 3 4 miss E T relative resolution extensive validation in the context of SUSY analyses for summer 2011 AFII was found to be accurate enough within systematic of jet energy scale (5%) part of the SUSY signal grid simulation of ~ 60 mio events done with AFII Entries pt distribution of electron fakes AFII / G4.9.2 300 250 200 150 0 50 0 1.2 0.8 ATLAS Preliminary Simulation G4.9.2 AFII 0 20 40 60 80 0120140160180200 p [GeV] 1.1 1 0.9 0 20 40 60 80 0 120 140 160 180 200 T p T [GeV] 18

Fast simulation AFII comparison to Geant4 ATLFAST2 validation in context of top physics analyses - number of jets, jet properties well reproduced - reproduce mw and mtop within statistical uncertainties Not reviewed, for internal circulation only AFII jets being scaled by 1.0.1 19

Fast simulation in ID/MS FATRAS geometry model: physics engine: tuning possibilities: material scaling full library alternative/fast parametric ATLAS TrackingGeometry ATLAS extrapolation (transport) material effect integration parameterised Geant4 for particle decay (and more?) interaction probabilites smearing factors π 3 ctrl C V 20

Fast simulation AFIIF high G4 CPU CONSUMPTION 1 ATLFAST2F x0.01? low HIERARCHY ACCURACY 21

Fast simulation in ID/MS FATRAS Parameterisation of material interactions (a) multiple scattering (b) ionisation energy loss Entries 3 2 Entries 3 2 Geant4 FATRAS Entries 25000 MPV 1.069 +- 0.001 Sigma 0.05743 +- 0.00043 MPV 1.071 +- 0.001 Sigma 0.05797 +- 0.00041 1 2 GeV μ - traversing 250mm Si (~0.267 % X 0 ) -0.002-0.001 0 0.001 0.002 θ proj (c) brem photon radiation 1 5 GeV π traversing 3mm Al (~3.37 % X 0 ) 0.5 1 1.5 2 2.5 3 3.5 4 Δ [MeV] (d) brem photon conversion Entries/bin 4 3 2 Geant4 Entries 13055 Mean 4.132 RMS 7.484 Geant4 (cut) Entries 5068 Mean 9.305 RMS 9.725 FATRAS Entries 4620 Mean 9.088 RMS 9.371 Entries/bin 3 2 Geant4 FATRAS 13.2.0 Entries 19583 Mean 19274 Entries 17597 Mean 18656 e e <p FATRAS >/<p G4 > 1.1 1 0.9 0.8 0.7 0.6 0.5 1 0 5 15 20 25 30 35 40 45 p [GeV] T 1 0 20 40 60 80 0 120 140 p e [GeV] 0.4 0.3 0 20 40 E γ T [ 22

Fast simulation in ID/MS FATRAS (e) nuclear interactions (parametric model implemented) n particles, energy distributions, parameterised from Geant4 X Entries/bin Entries/bin 4 3 2 4 1 Geant4 (a) outgoing particle energies from hadr. Interaction 0 20 30 40 50 E out [GeV] (g) angular distribution of outgoing particles Entries/bin Entries/bin 400 350 300 250 200 150 0 50 400 4 3 FATRAS (b) outgoing particle energy of highest order from hadr. Interaction 0 20 30 40 50 60 1st E out [GeV] (h) angular distribution of outgoing particles 3 2 pion 0 1 2 3 4 5 6 0 1 2 3 4 0.5 1.5 2.5 3.5 Δφ phase space restrictions Δη 23

threshold. (b) Comparison of the mean cluster size in the ATLAS pixel detector in 900 GeV collision data (black points) and MC simulated with FATRAS (shaded histogram). Fast simulation in ID/MS FATRAS FATRAS in comparison to data - ID reconstruction, tracks with pt > 500 MeV - using exact same sensitive detector elements: conditions data being fully integrated Arbitrary Units -1-2 by FATRAS. (a) Data, Number Run 142383 of pixels hits versus Fatras Arbitrary Units -1-2 (b) Data, Number Run 142383 of pixels hits versus Fatras Figure 5: Comparison of the geometric distribution of pixel detector hits in (a) and (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 account -3 the tails of the distributions. -2-1.5-1 -0.5 0 0.5 1 1.5 2 [mm] -3-2 -1.5-1 -0.5 0 0.5 1 1.5 2 z 0 sin [mm] 7. Summary d 0 Since the development of FATRAS was started, it has proven to be a useful tool, not only for debugging the track reconstruction algorithms and the simplified reconstruction geometry 24of

Fast simulation in ID/MS AFII/AFIIF - ATLAS standalone Muon System and combined (ID/MS) muon reconstruction standalone combined 25

Fast simulation ATLFAST Fully parametric simulation - smearing approach to create physics objects (no reconstruction) - has been used in the TDR phase of ATLAS full library alternative/fast parametric x0.001 ctrl C V 26

Fast simulation ATLFAST Long experience in ATLAS using parametric simulation - extremely fast - no efficiencies/fakes - no pile-up effects - however, often requested by theorists/externals, generic applications on the market (e.g. Delphes) Is parametric smearing something we (as a community) want to provide? ATLAS prefers to publish unfolded fiducial measurements to a parameteric public simulation. 27

Recap full library Geant4 / Fluka (Flugg) / Geant3 Frozen showers alternative/fast parametric AFII / AFIIF Atlfast(1) 28

Recap full library Geant4 Frozen showers used for public results alternative/fast parametric AFII / AFIIF Atlfast(1) current MC11 campaign in forward calorimeter 29

Recap All simulation setups run in Athena G4 run in G4 setup: G4 run/event handling G4 particle stack Frozen AFII / F ATLFAST run in G4/FATRAS setup: requires 2/1 simulation job runs in ATLFAST setup 30

Schematic: AtlfastII-F 31

The Integrated Simulation Framework ISF 32

The nutshell ISF vision DefaultFlavorCalo: fast MC FlavorFilterID: use full MC in cone around electron FlavorFilterID: use full within jet containing b- hadron DefaultFlavorID: use fast MC FlavorFilter: process µ with full MC 33

The nutshell ISF design ISimService fast MC ISimService full MC ID fast MC Calo full MC Particle Stack Simulation Kernel MS else fast MC full MC GeoFilter FlavorFilter misc. 34

The nutshell ISF design encapsulation of particle stack from flavors - one central stack - one simulation kernel ISimService ID Calo ISimService fast MC full MC fast MC full MC Particle Stack Simulation Kernel MS else fast MC full MC GeoFilter FlavorFilter - division of ATLAS simulation into sub-geometries misc. 35

The nutshell ISF design flavor filtering for each sub detector - allows refinement of simulation to optimise speed/accuracy balance ISimService ID Calo ISimService fast MC full MC fast MC full MC Particle Stack Simulation Kernel MS else fast MC full MC GeoFilter FlavorFilter - flexibility to integrate new simulation flavors - prepare for parallel particle processing misc. 36

The nutshell ISF design & prototype encapsulation of common services defined by interfaces - event handling, stack handling, truth filling, hit recording, barcode creation aim to feed central hit collection from ISF flavors - to allow for common pile-up digitization FastCaloSim & FATRAS have been imported to the prototype - first simulation tests running - we want to gather experience how to mix simulation flavors Geant4 interfacing on the way - Geant4 can/will still keep an internal particle stack - learn how to feed a future parallel simulation flavor 1st Z -> ee event with FastCaloSim in ISF 37

ISF benefits mix and play A common hit service in the ISF - AFII creates already SimHit objects similar to Geant4 hits plan to update FATRAS in Inner Detector - implemented parameterized material effects ISF: use Geant4 - did not deploy ATLAS digitization model ISF: create hits like Geant4 track SimHit Layer 1 Layer 0 SimHit Geant4 interaction ISimHitSvc 38

ISF benefits extend easily new parameterisation of hadronic leakage into the Muon System 39

ATLAS fast simulaton today & tomorrow ATLAS has extremely profited from Geant4 simulation - close collaboration with Geant4 community is extremely profitable ATLAS has a long-standing experience with fast simulation - starting from ATLFAST in TDR times - first AFII used in public results for summer conferences 2011 - bulk production of AFII samples in new MC11 campaign Currently fast/full simulation require different setups Huge effort started to develop an integrated simulation framework - steer all simulation flavors from one single framework - vision: mix fast and full MC techniques in one single event - feedback this experience into simulation R&D for the future (long-term) 40

Backup section 41

Timing Tables Sample Full G4 Sim Fast G4 Sim Atlfast-II Atlfast-IIF Minimum Bias 551. 246. 31.2 2.13 t t 1990 757. 1. 7.41 Jets 2640 832. 93.6 7.68 Photon and jets 2850 639. 71.4 5.67 W ± e ± ν e 1150 447. 57.0 4.09 W ± µ ± ν µ 30 438. 55.1 4.13 Table 1: Simulation times per event, in ksi2k seconds, for the full Geant 4 simulation, fast Geant 4 simulation, Atlfast-II, Atlfast-IIF [4]. Atlfast-II uses the full simulation for the inner detector and muon system and FastCaloSim in the calorimetry. Atlfast-IIF uses FastCaloSim for the calorimetry and FA- TRAS for the inner detector and muon system. All times are averaged over 250 events. TABLE I: Average time required for production of simulated hits for single muon samples in ATLAS MuonSpectrometer. Simulated hit creation Geant4 FATRAS (MSonly) [s/event] [s/event] p T =GeV 0.13±0.01 0.015±0.002 p T =0GeV 0.17±0.02 0.015±0.002 42

ATLAS/FATRAS extrapolation engine IN IExtrapolator Extrapolator OUT INavigator Navigator IPropagator StraightLinePropagator HelixPropagator RungeKuttaPropagator STEP_Propagator IMaterialEffectsUpdator MaterialEffectsUpdator McMaterialEffectsUpdator IMultipleScatteringUpdator MultipleScatteringUpdator IMagneticFieldTool MagneticFieldTool MagneticFieldTool_xk IEnergyLossUpdator EnergyLossUpdator McEnergyLossUpdator McConversionCreator 43

ATLAS/FATRAS extrapolation engine IN IExtrapolator Extrapolator OUT INavigator Navigator IPropagator StraightLinePropagator HelixPropagator RungeKuttaPropagator STEP_Propagator IMagneticFieldTool MagneticFieldTool MagneticFieldTool_xk IMaterialEffectsUpdator MaterialEffectsUpdator McMaterialEffectsUpdator IMultipleScatteringUpdator MultipleScatteringUpdator G4MaterialEffectsUpdator IEnergyLossUpdator EnergyLossUpdator McEnergyLossUpdator McConversionCreator 43