chep 2006 d. elvira, m. stavrianakou, fnal 1 the cms object-oriented simulation v. daniel elvira and...
TRANSCRIPT
CHEP 2006D. Elvira, M. Stavrianakou, FNAL 1
The CMS Object-Oriented Simulation
V. Daniel Elvira and Maya Stavrianakou, FNALOn behalf of the CMS collaborationCHEP 2006
Outline
Overview Interfaces and services CMS detector simulation and validation Parameterized simulations Production Summary and outlook
H->eeμμ at 300 GeV Luminosity 1034
CHEP 2006D. Elvira, M. Stavrianakou, FNAL 2
Detector Simulation
Generation Digitization
Pool HepMC data file
Pool SimHit data file
Pool Digi data file
Reconstruction
Generation – MC truth information from particle gun or physics generator about vertices and particles. Stored in HepMC format.
Detector Simulation – Hit objects with timing, position, energy loss information. Based on the Geant4 tool kit.
Digitization – Constructs Digi objects which include realistic modeling of electronic signal.
Reconstruction – Physics Objects: vertices, photons, e, , jets,……
Simulation Software in CMS
CHEP 2006D. Elvira, M. Stavrianakou, FNAL 3
Simulation Tools and the Framework Geant4
physics processes describing electro-magnetic and hadronic interactions
tools for detector geometry and sensitive element response interfaces for tuning and monitoring particle tracking
CMS framework Manages application control, incl. module scheduling,
action-on-demand where indicated, non-event data handling, data persistency, I/O,…
Interfaces to common services and tools such as magnetic field, generators, MC truth handling, infrastructure for hits, event mixing, digitization,…
Ensures provenance tracking and event immutability
CHEP 2006D. Elvira, M. Stavrianakou, FNAL 4
Simulation Tools and Framework cont’d
CMS switched from GEANT3 to GEAN4-based simulation end 2003 So far the CMS OSCAR simulation package has delivered
~100 M physics events for CMS 2004 Data Challenge and on-going Physics TDR studies
OSCAR is being adiabatically replaced by a new suite (referred to as here as SimG4), based on the new Event Data Model Software Framework
CPU: SimG4 1.5 x SimG3 - with lower production cuts! Memory: ~110 MB/evt for pp in SimG4 vs. ~100 MB in SimG3 Robustness: from ~1/104 crashes in pp events (mostly in hadronic
physics) in DC04 to < 1/106 crashes in latest productions
CHEP 2006D. Elvira, M. Stavrianakou, FNAL 5
Interfaces and services (I)
Application control using customized Event Data Producer module and RunManager interfacing Geant4 to the CMS Event Data Model Framework; persistency, I/O etc as provided by the Framework
Geometry provided by the Framework EventSetup, using the Detector Description package to convert the XML input files
Sensitive detector construction with configurable registration of sensitive elements
Hit collection and formatting, event mixing, and digitization based on common SimDataFormats classes and services
CHEP 2006D. Elvira, M. Stavrianakou, FNAL 6
Interfaces and services (II)
Magnetic field provided by the Framework EventSetup with configurable field selection, propagation tuning and optional user-defined local field managers
Physics with configurable physics lists and production cuts, cross section table storage and retrieval options
User actions for application monitoring and tuning, based on dispatcher-observer pattern (using BOOST signal library) for observable entity (run, event, track, step etc)
CHEP 2006D. Elvira, M. Stavrianakou, FNAL 7
Interfaces and services (III)
Event generation and Monte Carlo truth Choice of specific generator (Pythia, Herwig, gun etc) and event
format (ASCII, Pool, etc) run-time configurable HepMC::GenEvent converted to G4Event Navigable Monte Carlo truth record with decay/interaction
history for all generated as well as selected simulated particles
CHEP 2006D. Elvira, M. Stavrianakou, FNAL 8
22 m long & 15 m in diameter
Solenoid Magnet: 4 Tesla Field
Electromagnetic Calorimeter (Ecal)
Hadronic Calorimeter (Hcal)
Silicon Tracker
Muon System
More than 1 Million Geometrical volumes
SVS modular structure follows the detector sub-systemsThe CMS Detector
CHEP 2006D. Elvira, M. Stavrianakou, FNAL 9
Magnetic FieldField Map - TOSCA calculation
Designed to optimize simulation and reconstruction
Based on dedicated geometry of “magnetic volumes”
Decouple volume finding and interpolation within a volume
Time spent in magnetic field query (P4 2.8 GHz) for 10 minimum bias events 13.0 vs 23.6 s fornew field ~1.8-2 times faster than FORTRAN/G3
CHEP 2006D. Elvira, M. Stavrianakou, FNAL 10
Tracker• Key role in the development and
optimization of simulation infrastructure and validation
• Tracker material budget - estimated with detailed description of all active and passive components - directly affects e/m calorimeter physics; places stringent requirements on detector description and geometry construction accuracy
• Correct, navigable Monte Carlo truth, for correct decay tree reconstruction and proper treatment of hard electron bremmsstrahlung - crucial for B-t studies
• Tracker performance extensively validated in terms of tracking and hit distributions for single particles, minimum bias, and physics events
tracker geometry, hits and tracks (green - electrons)
CHEP 2006D. Elvira, M. Stavrianakou, FNAL 11
Electromagnetic Calorimeter (ECAL)• G4/test-beam comparisons
excellent representation of electromagnetic showers by G4
• Overall ECAL performance, in terms of energy and position resolution dominated by effects not included in shower simulation (electronics noise, photostatistics, longitudinal uniformity of light collection and crystal inter-calibration) only gross errors are identified by energy and position resolution comparisons
• Largest sensitivity to changes in radiation and showering in tracker material, but not trivial in to validate in test beam
Tracks and energy deposition in ECAL
CHEP 2006D. Elvira, M. Stavrianakou, FNAL 12
Hadronic Calorimeter (HCAL)• HCAL studies on energy resolution and
linearity, e/ ratio, and shower profile instrumental in G4 hadronic physics validation
• Comparisons between single particle measurements in test beam: 2002-2004, different HCAL modules, preceded by ECAL prototype, to beams of , e and over large energy range - G4 hadronic physics parametric (LHEP) and microscopic (QGSP) models
energy resolution and response linearity as a function of incident energy in good agreement with the data within the large systematic uncertainties in the latter
• Transverse and longitudinal shower profiles studied in 1996 and 2004 test beam
showers predicted by G4 narrower than those by G3
• Showers predicted by QGSP (v 2.7) shorter than those by LHEP (v 3.6) list, with LHEP predictions closer to those from G3/Geisha
test beam 2004 results
CHEP 2006D. Elvira, M. Stavrianakou, FNAL 13
Muons• G3/G4 comparisons for single 10 GeV-10
TeV: good agreement for ionization but brem, e+e− production, and in particular, -nuclear significantly different, due to newer theoretical developments included in G4
• Multiple scattering significantly smaller in G4, in agreement with experimental results
• G4 results also show improvement wrt precision of propagation in detector
• Test beam setup: two chambers with and without iron slab between, to investigate effect of showers in passive material; analysis: with PT 50-300 GeV
• G4 slightly underestimates soft -ray production in cell volumes, while hard - rays and e/m showers are correctly modeled; despite this, local track reconstruction efficiency and resolution well reproduced by simulation
HZZ4, MH =150 GeV
CHEP 2006D. Elvira, M. Stavrianakou, FNAL 14
Forward Detectors
• Forward detectors, CASTOR and ZDC calorimeters, Totem telescopes: essential for diffractive and heavy ion programs.
• ZDC: Cerenkov detector designed to collect any remaining neutral fragments of the colliding nuclei - may be used as a measure of collision centrality
• Simulation underway to study energy resolution and leakage to be compared with Nov 2005 test beam data
3D view of CASTOR
ZDC: Zero Degree Calorimeter at 140 m
Castor Calorimeter at 14.37 m (5.3 <h< 6.7)
Totem Telescopesat 7.5 m < z < 13.5
CHEP 2006D. Elvira, M. Stavrianakou, FNAL 15
Parameterized Simulation (I) detailed simulation of e/m showers CPU intensive Þ parameterization of
spatial energy distribution of e/m shower, based on probability density functions, allows speed up without compromising simulation accuracy
GFlash model (G. Grindhammer, S. Peters), based on three probability density functions (originally developed and used by H1) used to parameterize electrons and positrons in CMS barrel and endcap e/m calorimeter
comparisons between GFlash-based and full simulation energy depositions in central crystal, 3x3, 5x5 matrices: agreement to ~1% transverse and longitudinal shower profiles: agreement to ~1-3% speed increases by factors 3-10 depending on event type, particle energy and
detector region
examples single e or with E=100 GeV in ECAL barrel: factor ~10 speed-up large extra dimensions full signal event, with single > 1000 GeV: factor ~4
CHEP 2006D. Elvira, M. Stavrianakou, FNAL 16
Parameterized Simulation (II)
Energy deposition in a 5x5 crystal matrix for 50 GeV electrons
Histograms = full geant4 simulation Red markers = shower parameterization
Transverse shower profile for 50 GeV
Longitudinal shower profile for 50 GeV
CHEP 2006D. Elvira, M. Stavrianakou, FNAL 17
Production 11/2003 - 10/2005:
~100 M physics events simulated by production team
Failure rate: ~1/104 - 106 events Performance for typical signal
event: ~250 MB memory ~100-200K Si2K (*) CPU ~1 MB output data CMS Computing Model: same
amount of simulated as real data, ~1.5 x 109 events/year mix of full and fast simulation will be used to manage required resources
(*) typical 3 GHz CPU is ~1.7 K Si2K, so 200K Si2K is 2 minutes on such a CPU; Si2K ratings scale more or less linearly with CPU speed
CHEP 2006D. Elvira, M. Stavrianakou, FNAL 18
Summary and Outlook
SUSY (leptons, missing ET)
visualization with IGUANACMS
In CMS, the Geant4-based OO simulation has successfully replaced its Fortran/Geant3 predecessor. It has been validated and adopted by all CMS detector and physics groups. It has proven robust and performant, easily extensible and configurable.
It has also proven modular and portable to a different software framework without loss of functionality or usability.
New features, extending functionality, such as exotics simulation, or improving performance, such as hadron shower parameterization, envisaged for 2006, can be added without disrupting standard operation/ performance.
Black Hole EventModel of Landsberg and DimopoulosPlanck scale = 1 TeVNumber of extra dimensions = 2
CHEP 2006D. Elvira, M. Stavrianakou, FNAL 19
Heavy Ion Simulation
G3: event in slices of 100 tracks run separately(ZEBRA limitations) - G4: run full HI events55K generated particles, with 97K tracks from 80K vertices…• Event in slices of 100 particles
• G3 230 min• G4 180 min
• Full event G4 210 min (P4 3.2 GHz )
hits in tracker and ECAL by particles from central Pb+Pb collision at 5.5 TeV/nucleon