Geant4 Event Biasing

1
Geant4 Event Biasing
May 2007, Geant4 v8.2.p01

• Jane Tinslay, SLAC

2
Outline

• Introduction
• Built in biasing options
• Primary particle biasing
• Radioactive decay biasing
• Mars hadronic leading particle biasing
• General hadronic leading particle biasing
• Hadronic cross section biasing
• Geometrical biasing
• Importance sampling
• Weight windows weight cutoff
• User defined biasing
• G4WrapperProcess
• Uniform bremsstrahlung splitting example
• Recent developments
• Summary

3
Introduction

• What is analogue simulation ?
• Sample using natural probability distribution,
  N(x)
• Predicts mean with correct fluctuations
• Can be inefficient for certain applications
• What is non-analogue/event biased simulation ?
• Cheat - apply artificial biasing probability
  distribution, B(x) in place of natural one, N(x)
• B(x) enhances production of whatever it is that
  is interesting
• To get meaningful results, must apply a weight
  correction
• Predicts same analogue mean with smaller variance
• Increases efficiency of the Monte Carlo
• Doesnt predict correct fluctuations
• Should be used with care

4

• Geant4 simulation
• Analogue regular processing
• Non-analogue/event biased simulation
  manipulated processes and/or process list
• I.e, manipulate processing to effectively apply
  B(x) in place of N(x)
• Geant4 provides
• Several built-in general use biasing techniques
• Utility class, G4WrapperProcess to support user
  defined biasing
• Expect biasing to be used by experienced users
• Should understand what a particular biasing
  technique does, its constraints and side effects
• Understand how processing works in Geant4

5
Built in Biasing Options
6
Primary Particle Biasing

• Use case
• Increase number of high energy particles in
  cosmic ray spectrum
• Increase number of primary particles generated in
  a particular phase space region of interest
• Weight of primary particle modified as
  appropriate
• General implementation provided by
  G4GeneralParticleSource class
• Bias position, angular and energy distributions

7

• G4GeneralParticleSource is a concrete
  implementation of G4VPrimaryGenerator
• Instantiate G4GeneralParticleSource in your
  G4VUserPrimaryGeneratorAction class
• Configure biasing to be applied to sampling
  distributions through interactive commands

8

• Online documentation examples
• http//reat.space.qinetiq.com/gps/
• Geant4 examples
• examples/extended/eventgenerator/exgps

9
Radioactive Decay Biasing

• G4RadioactiveDecay simulates decay of radioactive
  nuclei
• Implements the following biasing methods
• Increase sampling rate of radioculides within
  observation times
• User defined probability distribution function
• Nuclear splitting
• Parent nuclide is split into user defined number
  of nuclides
• Branching ratio biasing
• For a particular decay mode, sample branching
  ratios with equal probability

10

• G4RadioactiveDecay is a process
• Register with process manager
• Biasing can be controlled in compiled code or
  through interactive commands

11

• Online documentation examples
• http//reat.space.qinetiq.com/septimess/exrdm/
• Geant4 examples
• examples/extended/radioactivedecay/exrdm

12
Mars Hadronic Leading Particle Biasing

• Inclusive event generator for hadron(photon)
  interactions with nuclei
• Translated from Mars13(98) version of Mars code
  system
• http//www-ap.fnal.gov/MARS
• Useful for punch through studies
• Generates fixed number of particles at each
  vertex with appropriate weights assigned
• Valid with energies lt 5GeV with ?, ?-, K, K-,
  K0L, K0S, proton, neutron, anti-proton, gamma

13

• To use, create a G4Mars5GeV object and register
  with inelastic process
• More examples provided in the LHEP_LEAD,
  LHEP_LEAD_HP etc physics lists
• Depreciated

14
General Hadronic Leading Particle Biasing

• Built in utility for hadronic processes
• Implemented in G4HadLeadBias class
• Keep only the most important part of the event,
  and representative tracks of given particle types
• Keep track with highest energy
• I.e, the leading particle
• Of the remaining tracks, select one from each of
  the following types if they exist
• Baryons, ?0s, mesons, leptons
• Apply appropriate weight
• To activate, set SwitchLeadBiasOn environment
  variable

15
Hadronic Cross Section Biasing

• Built in cross section biasing in hadronics for
  PhotoInelastic, ElectronNuclear and
  PositronNuclear processes
• Artificially enchance/reduce cross section of a
  process
• Useful for studying
• Thin layer interactions
• Thick layer shielding

16

• Hadronic cross section biasing controlled through
  BiasCrossSectionByFactor method in
  G4HadronicProcess
• More details at http//www.triumf.ca/geant4-03/tal
  ks/03-Wednesday-AM-1/03-J.Wellisch/biasing.hadroni
  cs.pdf

17
Geometrical Biasing

• Geometry based biasing implemented within common
  framework in Geant4
• Importance sampling
• Weight windows
• Weight cutoff
• Process based approach
• Process list is modified behind the scenes to
  apply biasing
• Implements own scoring scheme
• Depreciated - future releases will use scorers
• Applicable in mass or parallel geometries
• Define physical volumes named cells
• Limitations with biasing parallel geometries
• Cant bias in fields or with charged particles
• Improved in future releases using new general
  parallel navigation
• Currently in Beta stage

18
Importance Sampling

• Aim
• Increase number of particles in geometries that
  are interesting or important
• Decrease number of particles in geometries that
  are not so interesting or important
• Method
• Divide mass or parallel geometries up into
  importance cells
• Assign an importance value to each cell
• Importance value reflects the relative importance
  of that cell
• When a particle crosses a boundary between two
  importance cells, apply biasing algorithm based
  on relative importance between two cells
• Either split(duplicate) track, or play Russian
  Roulette (kill track with certain probability)

19

• Default algorithm
• When crossing from cell A with importance IA, to
  cell B with importance IB, define relative
  importance R
• If R 1 continue transport
• If R lt 1 play Russian Roulette
• Reduce tracks when passing from more important
  to less important cell
• Kill tracks with probability 1-R
• If R gt 1 split track
• Increase tracks when passing from less
  important to more important cell
• Split into R dupicate tracks
• Apply appropriate weights

20
Importance Cell Example

• Splitting

Russian Roulette
21
Weight Window

• Weight based enhancement to importance sampling
• Particles either split or Russian Roulette played
  based on space-energy cells
• User defines a weight window for each space cell,
  and optionally for different energies
• Can help control weight fluctuations introduced
  by other variance reduction techniques

22

• User defines
• Lower weight bound
• Upper weight factor
• Survival weight factor
• Upper weight bound lower weight bound upper
  weight factor
• Survival weight bound lower weight bound
  survival weight factor

23
Samplers

• Samples (G4VSampler) are user tools which setup
  the geometrical biasing
• Handles process list manipulation behind the
  scenes
• Two samplers provided with Geant4 distribution
• G4MassGeometrySampler
• Biasing in mass geometry
• G4ParallelGeometrySampler
• Biasing in parallel geometry
• Samplers provide methods for configuring
• Importance sampling
• Weight window
• Weight roulette
• Scoring
• Each particle type that is geometrically biased
  should have its own sampler

24
G4VSampler Structure

• Use the prepare and configure methods of
  G4VSampler to prepare biasing

25
Importance Sampling Configuration

• Prepare a collection of cells
• Importance value - volume associations
• Create, prepare configure sampler

26
Weight Window Configuration

• Prepare a collection of volume-weight windows
• Assign a lower weight bound and upper energy
  bound to a volume
• Create, prepare configure sampler

27
Biasing Example B01

• examples/extended/biasing/B01
• Study punch through of 10 MeV neutrons incident
  upon thick concrete cylinder
• Demonstrates importance sampling weight windows
  technique in mass geometry
• Geometry consists of an 80 cm high concrete
  cylinder divided into 18 slabs
• Importance value for slab n 2n

28
Example B01 - 10 MeV neutrons, thick concrete
cylinder
29
Geometrical Biasing Documentation

• Detailed examples can be found at
• examples/advanced/Tiara
• examples/extended/biasing
• Documentation on all geometrical biasing
  techniques at
• http//geant4.web.cern.ch/geant4/UserDocumentation
  /UsersGuides/ForApplicationDeveloper/html/ch03s07.
  html

30
User Defined Biasing G4WrapperProcess

• Implement user defined biasing through
  G4WrapperProcess
• A process itself, I.e, inherits from G4VProcess
• Wraps an existing process
• By default, function calls are forwarded to
  existing process
• Non-invasive way to manipulate the behaviour of a
  process
• To use
• Subclass G4WrapperProcess and override
  appropriate methods, e.g, PostStepDoIt
• Register sublcass with process manager in place
  of existing process
• Register existing process with G4WrapperProcess

31
G4WrapperProcess Structure
32
Example Uniform Bremsstrahlung Splitting

• In this example, only interesting in scoring
  bremsstrahlung photons
• Want to increase Monte Carlo efficiency by
  reducing computing time spent tracking electrons
• Example of biasing through enhancing production
  of secondaries

33

• When a bremsstrahlung interaction occurs
• Instead of sampling photon energy angular
  distributions just once, sample N times to
  generate N unique secondaries
• Multiple secondary generation termed splitting
  in this case
• Not to be confused with importance sample
  splitting, where N identical copies are created
• Electron energy reduced by energy of just one
  photon
• Remove bias introduced in photon energy and
  angular distributions by assigning a statistical
  weight to each secondary
• N splitting factor ( of secondary photons)

34
Implementation

• Create user class inheriting from
  G4WrapperProcess
• Override PostStepDoIt method of G4WrapperProcess

35

• Implement overridden PostStepDoIt method
• Register wrapped process with process manager

36
(No Transcript)
37
Recent Developments
http//geant4.slac.stanford.edu/EBMS/
38

• Three components
• Scoring for biasing Tsukasa Aso (TNCMT)
• Geometrical based biasing Alex Howard (CERN)
• Physics based biasing Jane Tinslay (SLAC)
• Evaluate current status of each component
• Identify missing functionality
• Comparisons with other Monte Carlo codes (eg, EGS
  family, MCNP, Penelope, Fluka)
• Investigate interaction between components
• Work plan developed with short, medium and long
  term goals

39
Workshop Brief Summary Work Plan

• Detailed summary and work plan at
  http//geant4.slac.stanford.edu/EBMS/material/Sum
  mary_EBminiworkshop.ppt
• Geometrical biasing
• Updated to use parallel navigation developments
• With release v9.0 should be able to do full
  geometrical biasing in parallel worlds
• At the moment limited to neutrals
• Biasing examples to be updated
• Producing validation examples
• Scoring for biasing
• Use in place of depreciated G4VScorer used in
  geometrical biasing
• Development of new scorers

40
(No Transcript)
41

• Physics biasing
• Existing physics based biasing fragmented
• Identify missing biasing methods variations
  between methods in other Monte Carlo codes
• Implicit capture
• General cross section biasing
• Interaction forcing
• Path length biasing
• Advanced bremsstrahlung splitting
• Leading particle biasing
• Look at developing dedicated framework to provide
  general physics biasing in analogy with
  geometrical biasing
• Manipulating physics processes/lists

42
Summary

• Number of popular event biasing techinques built
  into Geant4
• User defined biasing supported through
  G4WrapperProcess
• Ongoing developments aim to improve exiting
  Geant4 biasing, and provide new event biasing and
  scoring methods
• Documentation at
• http//geant4.web.cern.ch/geant4/UserDocumentation
  /UsersGuides/ForApplicationDeveloper/html/ch03s07.
  html