Simulations in Detector Design and Future Prospects Harry W. K. Cheung, Fermilab Muon Collider Works

1
Simulations in Detector Design and Future
ProspectsHarry W. K. Cheung, FermilabMuon
Collider Workshop April 21-25, 2008

• OUTLINE
• The Need for Simulation Studies
• Geant4
• Detailed Studies are Needed?
• Examples of Simulations in Detector Design
• MARS
• Conclusion

2
Introduction Why Simulations?

• Simulations are used throughout a physics
  experiment
• Design of the experiment
• Design and optimization of individual detectors
• Predicting backgrounds
• Understanding the data (simulations numeric
  calculations)
• including during physics data analysis
• Correcting for detector effects on data
  (acceptance, resolution)
• For simulation based corrections to the data
  (variations of response, due to linearity,
  cracks)
• Correcting the data vs. simulation of the effect
• Extracting signals from data (backgrounds,
  shapes)
• So simulations must be validated
• Validation of the detector geometry and detector
  response
• Validation of physics processes particle
  interaction simulations
• What is good enough?

3
Introduction The Simulation Process
Simulation
Real Life
Machine ? collisions
Event Generator
Produce events
Detector simulation
Detector, DAQ, Trigger
Observe store
Event reconstruction
Determine particles and properties
Generator analysis
Determine physics (and compare simulations with
data)
Physics Analysis
4
Introduction The Simulation Process
Simulation
Real Life
Machine ? collisions LHC
Event Generator PYTHIA CMS FW
Produce events
Detector simulation GEANT4 CMS Geom Add.
Detector, DAQ, Trigger CMS
Observe store
Event reconstruction CMS FW Recon
Determine particles and properties
Generator analysis
Determine physics (and compare simulations with
data)
Physics Analysis ROOT ( C)
5
Particle Interactions Other physics processes

• Besides the initial collision there are other
  interactions
• Interaction of generated particles with the
  detector and other material, secondaries
  generated will also interact and/or decay
• This is handled in e.g. GEANT4 as part of the
  detector simulation, includes electromagnetic and
  hadronic interactions, and decays
• Some processes are well known or calculable but
  still a large choice of processes, and over large
  energy ranges, e.g. for electromagnetic
• Photon processes (compton scattering, gamma
  conversion, photoelectric effect, muon pair
  production)
• Electron/positron processes (ionization and delta
  ray production, bremsstrahlung, positron
  annihilation, synchrotron radiation)
• Muon processes (Ionization and delta ray
  production, Bremsstrahlung, ee- pair creation)
• Hadron/ion processes (ionization)
• Multiple scattering
• Need to choose thresholds for secondary particle
  production
• Cannot do each calculation, need cross section
  tables and approximations

6
Geant4 and Physics Tables

• More complicated for poorly known processes like
  hadron interactions
• Implementation of physics processes using
  data-driven or theory-driven models, or
  parameterized cross-sections (in Geant4 the user
  is meant to select these!)
• Geant4 Physics Tables to make things simpler
• Geant 4 is set up to easily be tailored for more
  types of physics (astroparticle, nuclear physics,
  medical physics,)
• Gives for each particle type which physics
  processes to use and how the cross sections are
  computed for each, E.g. for LHC
• LHEP GHEISHA ported from Geant3 (parameterized
  inelastic)
• QGSP GHEISHA (Elt25 GeV), q-g string model (Egt25
  GeV)
• QGSC as QGSP but chiral invariant PS decay
  fragmentation
• FTFP as QGSP but with fragmentation ala FRITJOF
• No single physics table for all HEP experiments
  (or astroparticle, etc.), still an area of
  activity for development and validation with data
  
• Geant4 Users Guide physics processes, physics
  reference manual, physics table description, and
  example physics tables.
• SLAC Geant4 group physics tables.

7
Geant4 Some Issues

• Geant4 uses a particle propagator with geometries
  defined as volumes
• The smaller the volumes the smaller the step size
  and the more accurate but also the slower the
  simulation gets
• The Simulation geometry can be too realistic - if
  too many volumes are defined, the simulation gets
  too slow
• Approximations are still needed in Geant, e.g.
  the detector response
• For a pixel sensor we dont define every pixel as
  a Geant4 volume, the sensor is one single volume
  and we save the entry, exit, ELOSS for each
  particle and parameterize/simulation the charge
  deposition separately
• For detectors with light collection, no
  simulation (and fluctuations) of light
  collection, parameterize separately
• Complex (non-uniform) Magnetic fields can slow
  down the simulation significantly
• GFlash shower parameterization for regions that
  are homogeneous can speed up the simulation
  significantly
• Geant4 unlikely to match data exactly, what to
  do? Dont want to tune Geant4 as Geant4 Physics
  Tables can change in next version (Important for
  physics analysis stage)

8
The Need for Detailed Studies

• Why not just simple/toy simulations? Can learn a
  lot, e.g. efficiencies for tracking, resolutions,
  but e.g. studying a tracking trigger is more
  complex
• Offline tracking algorithms vs L1 tracking
  trigger algorithms
• Tracking trigger algorithms have less time and
  information to work with and thus more sensitive
  to the details
• Rely on geometry constraints to reduce
  combinatorics or processing time, e.g.
  simplifying hit matching, or selecting higher pT
  tracks
• Rely on looking at only limited information, e.g.
  regional tracking, larger segmentation, or binary
  hit information
• Likely to want to investigate a wider range of
  possibilities
• E.g. in BTeV, tracking trigger design can drive
  tracker hardware decisions
• Include looking at benchmark physics signals in
  optimizing tracker design choices (c.f. toy
  simulations)
• Want to see effect of otherwise similar design
  choices on physics
• Look at trigger efficiencies on data passing
  physics analysis cuts

9
The Need for Detailed Studies

• Sometimes details matter
• Noise, accuracy of MCS or other interactions,
  non-Gaussian tails, primary vertex smearing,
  alignment, overlaps,

More Relevant Example
10
The Need for Detailed Studies

• Examples of impact due to simulations of L1
  tracking trigger design influencing the tracker
  hardware design from BTeV 8 years of RD
• Complicated set of triplet planes to uniform set
  of doublet planes

5 cm
Inner pixel region
4.25 cm
11
The Need for Detailed Studies

• Examples of impact due to simulations of L1
  tracking trigger design influencing the tracker
  hardware design from BTeV 8 years RD
• Complicated set of triplet planes to uniform set
  of doublet planes
• Smaller non-bend plane

Half-plane
12
Planned Geant4 Development

• Some planned Geant4 developments
• Geometry
• Field performance/tuning, multiple
  navigators/geometry
• Hadronic Physics
• Improving matching of simulations with data
• New production cascade models, work on existing
  production cross section or fragmentation models,
  and include new ones
• Validating models with thin target beam tests
• Fermilab CD is part of the Geant4 team and
  contributing to these tasks, especially the
  hadronic physics and improving code performance
• Low Energy Electromagnetic Physics
• Materials, Generic Processes and
  Parameterisations
• Standard Electromagnetic physics optical
  processes
• Visualization and Graphics Representations
• etc.
• A surprisingly large list of planned developments
  tasks just for 2008!

13
Simulations in Detector Design

• Historical

Desired


42
Reviewers/Approvers seem to want
14
Example BTeV

• Started with MCFAST (FORTRAN) a software tool
  specially for heavy flavour decay experiments,
  with emphasis on tracking and vertexing
• Hit based tracking and vertexing with material
  effects
• Cheated on tracking in MCFAST (no full track
  reconstruction), however full track pattern
  recognition and vertexing in L1 Track and vertex
  trigger code
• Easy to configure geometry with ASCII file and
  fast (signal in large bkgd)
• MCFAST evolved to include new features
  parameterized calorimetry and particle
  identification but not hit based, muon simulation
  too simple
• As the detector design became more stable BTeV
  transitioned to using only Geant3 (later ECAL
  used Geant4)
• Had to redo the Geometry for Geant, and geometry
  changes took time
• Slow to run, limited in statistics and had to
  compromise in various areas E.g. only studied a
  few select signals, only generate selected
  background thought to be the most important,
  generate b-quarks with non-optimal parameters in
  Pythia
• Software structure simulation turned out to be
  not really an issue due to having only a
  relatively small focused group (not all going in
  different directions)

15
Example International Linear Collider

• Historically software based on four separate
  efforts, each with separate data formats, and
  framework - issues with interoperability

16
Example Linear Collider Software
17
Example Linear Collider

• One proposed solution for interoperability

In practice May be hard to do, or just not get
done in a timely manner
18
Example Upgraded CMS at SLHC

• CMS has formed working groups to look at upgrades
  for CMS for the Super SLHC Phase 1 (2?1034
  cm-2s-1) and Phase 2 (1035 cm-2s-1)
• For Phase 2, the whole tracking system would be
  replaced, and some tracking information is needed
  in the L1 trigger

19
Some Upgraded Tracker Design ideas

• No single strawman tracking system or tracking
  trigger strategy/design, need comparison studies

Square pixels, long pixels, super-layers
Extra pixel layer, bigger pixels, long
pixels/short strips, 1-2 triggering layers
?
Minimal Change Approach just
increase(/decrease) granularity to the minimum
needed.
3 Super-layers of stacked doublets
Elliptical - 60 of the area
20
Example Upgraded CMS at SLHC

• CMS has formed working groups to look at upgrades
  for CMS for the Super SLHC Phase 1 (2?1034
  cm-2s-1) and Phase 2 (1035 cm-2s-1)
• For Phase 2, the whole tracking system would be
  replaced, and some tracking information is needed
  in the L1 trigger
• Need a common set of simulation and software
  tools to simulate different tracking systems for
  comparisons
• CMS has a Geant4 simulation, but is too slow for
  rapid feedback of trying out various tracking
  system layouts, etc.
• CMS has a Fast Simulation, but it cheats on
  tracking, and goes straight from simhits to
  rechits where granularities do not affect
  occupancy
• Both the Geant and Fast Simulations partly shared
  the detector geometry in XML/algo for simulation
  and reco, but the geometry is hardwired
• The Fast Simulation has its own particle
  propagator that uses its own interaction
  geometry for navigation. This geometry must be
  tuned to the one in XML.
• CMS has a working group to put features of the
  CMS Geant simulation into the Fast Simulation,
  and also make the geometry have limited
  configurability - limited manpower, tried to
  reuse code
• Still a large phase space to cover, start with
  two examples for Phase 2

21
Barrel Strawman A and B
Strawman A
Strawman B
Short strips
Superlayer
Pixel layers
Strixel layers, could be doublets
Mini-strip or Pixel doublets
(Layers simplified as rings for illustration!)
22
Strawman A Rechits from Simulation

• Strawman A rechit r-phi view showing rechits
  from Geant simulation

Can run both Fast Simulation and Geant Simulation
with both Strawman geometries and perform full
track pattern recognition and reconstruction
4 TOB short strips
2 TOB strixels
2 TIB short strips
2 TIB strixels
4 inner pixels
23
Strawman B Rechits from Simulation

• Strawman B rechit r-phi and r-z views from Fast
  simulation
• Layout and granularity configurable via XML file

Still work needed on correcting material budget,
but ready for doing studies after one year with a
group of about 6 people each with 0.2-0.5 FTE
r-z view
24
Experience from CMS Upgrade Work

• Benefited enormously from existing CMS code, and
  existing expertise for consultation
• However lack of manpower still need to get CMS
  working!

• Getting early simulation tools ready with
  (limited) flexibility is key to having a common
  set of software tools
• The CMS Trigger upgrade group is just starting
  the real work and can use our simulation tools
• However some development is still needed, but it
  helps that we can have input to standard CMS
  software development
• Redesign of FastSimulation geometry code to make
  it more easily configurable
• Redesign of tool for simulating pileup from
  multiple interactions/crossing to handle SLHC
  case
• Extending Fast Simulation to get types of (muon)
  hits that we need
• We can run Fast simulations and Geant simulations
  with the same geometry, and use the same reco code

Iguana Image of a Strawman B layer
25
Simulations Validation

• The value of a simulation depends on how well the
  output mimics real data.
• Validation of the physics processes (e.g. cross
  sections)
• Physics Table, and Range cutoffs or
  approximations used
• Validation of the geometry, including all
  materials
• Validation of the detector responses (test beam
  or real data)
• Misc. - e.g. environment, like beam backgrounds
  (cant take out)
• Beam related, pileup, cosmic
• Validation of the actual computer code and
  updates
• Deciding on what is good enough!
• Simulation Validation is hard and takes
  resources, a lot more than just setting up the
  initial common simulation and software tools
• Using established code helps a lot

26
Other Established Simulation Codes

• Other established codes besides Geant4
• MARS has been in use for 30 years
• Simulates the passage of particles through
  matter. All manner of particle interactions for
  hadrons, leptons, photons and heavy ions are
  present.
• Used in many applications, usually related to
  accelerators
• Original MARS code based on an inclusive approach
  to multiparticle interactions using weighting
  techniques
• Averaged results, cannot study a single
  interaction
• Results on direct energy deposition, fluence,
  dose, tabulated or in histograms not trivial to
  combine results from multiple runs
• No study of fluctuations in each cascade, or
  production correlations
• Exclusive code available now, but much slower,
  relatively new
• Fast CPU time increases logarithmically with
  energy, not linearly as in an exclusive method,
  ideal for really high energy cascades

27
Other Established Simulation Codes

• MARS continued
• Exists various interfaces Useful for specific
  applications
• MCNP4C code for neutron and photon production and
  transport below 20 MeV
• ANSYS code for thermal and stress analyses
• STRUCT code for multi-turn particle tracking in
  large synchrotrons and collider rings
• MAD to design accelerator lattices and beamlines
  to study beam loss, collimation, backgrounds and
  shielding
• Comparison and verification of hadronic
  interactions
• Various ongoing developments including comparison
  of hadronic shower code with Geant4, Fluka, etc.
• CMS Using MARS for beam-related background
  studies
• Studying radiation environment (pixels and
  silicon strip tracker) and accident scenarios
• Uses an earlier geometry of CMS (in Fluka), many
  changes since in the Standard CMS geometry no
  full geometry update for MARS
• No fluctuations/saving of single collision data
• Looking at possible use of Geant4 for some studies

28
Conclusions

• Need to work on common set of simulation and
  software tools
• A need for both a fast simulation and a full
  simulation based on Geant4
• Both should share as much code as possible - e.g.
  input files like geometries, data formats, etc.
• Would benefit greatly if these simulations are
  also the standard for the beamline
• If a different beamline simulation code is used,
  at least it should share as much of the other
  code as possible, e.g. input geometry files, data
  formats, production cross sections
• Simulation and analysis tools needs to be done at
  a very early stage so people dont feel the need
  to go off in (many) different direction with
  their own software tools
• Thanks to Rob Kutschke and Pushpa Bhat for help
  with my homework for this talk

29
Backup Slides
30
Intro The Generator

• A generator is needed to get events of interest,
  a general purpose generator is needed to do many
  things
• The hard scattering process, parton showers,
  resonance decays, underlying events,
  hadronization, and ordinary decays
• There are specialized generators for various
  physics
• There are many! E.g. ALPGEN, GR_at_PPA, EvtGen,
  AcerMC,
• Still need a general purpose generator, PYTHIA,
  HERWIG, ISAJET, SHERPA, each have many parameters.

31
Aside What About Decays?

• Decays can be in the generator and/or in the
  Geant4 Physics Table
• PYTHIA can decay the particles it generates,
  there are many parameters in PYTHIA you can set
  to control decays, but heavy quark (b and c)
  decays are not handled in the best way.
• Geant4 also has decay tables that can be used and
  constructed as part of making the Physics Table -
  usually for the simpler particle decays.
• There are specialized decay tables and decay
  code, e.g. sequential heavy flavour decays are
  handled with EvtGen

32
Aside Geant4 vs Geant3

• Geant4 is in C and is set up to easily be
  tailored for more types of physics
  (astroparticle, nuclear physics,)
• Geant4 geometry is done differently and there are
  new volume types, and a nice visualization tool.
  (In CMSSW, the geometry is split into a part in
  XML but also can be in algorithms in code.)
• Many other differences, see the Geant4
  introduction.
• What particle id numbers do we use inside Geant?
• CMS FW loads a particle ID table (in PDG id
  scheme) from, see also http//pdg.lbl.gov/2005/mcd
  ata/mc_particle_id_contents.html.
• Does Geant4 have the Geant3-style energy cutoffs
  for tracing?
• Not by default. One can give instead a range (in
  length) for particular particles, materials or
  volumes. This is translated internally into
  energy thresholds for production of secondaries.
  All particles produced will be traced to zero
  energy. (Production thresholds are actually more
  complicated, see this part in the Geant4 users
  manual.)
• Multiple scattering is done differently than
  Geant3 , must look at the Geant4 Physics
  reference manual