Geant4 version 10.0.

p01

Event Biasing
Geant4 Tutorial: version 10.0.p01
Michael Kelsey, Thu 6 Mar 2014
 Geant4 version 10.0.p01

Outline

• Motivation
• What Is Event Biasing?
• Geometric Biasing in GEANT4
• Physics Biasing in GEANT4
• User Defined Biasing
• NEW General Biasing Interface
• Bremsstrahlung Splitting
• Summary
2
 Geant4 version 10.0.p01

Motivation

If you want a fully realistic simulation, you need to generate,

model, and track as much as possible of what would happen in
real life.

“Too many events get rejected by my cuts.”

“Why track secondaries in my support structure?”
“I don’t care about 1 MeV photons!”
“GEANT4 is too slow.”

If you need fast results, or high statistics, maybe give up some

realism (or low-probability tails) in exchange for getting right
answers on average, and worry about tails elsewhere.

3
 Geant4 version 10.0.p01

What is Event Biasing?

Method of accelerating simulation of useful events at the expense of

accurate fluctuations

Analogue simulation uses natural PDFs N(x) to generate correct

mean and correct fluctuations, including far-off tails
• Often includes a significant fraction of events/particles outside final
• acceptance (physical or phase space) or interest
• Poor net efficiency

Biased simulation replaces N(x) with an artificial PDF B(x), which

enhances production of interesting events/particles
• Increases MC efficiency: more events survive detector model, cuts, etc.
• Distribution and fluctuations are not correct: must apply weight
correction, but still won’t get tails

4
Geant4 Built-In Biasing (Application Guide, Sec
Geant4 version 10.0.p01

3.7)
GEANT4 does analogue simulation by default
For expert/specialty users, provides methods to manipulate processing
stages and apply B(x) bias early

Geometric or acceptance biasing

• Uses a combination of scoring and biasing functions
• Suppress/enhance events on the basis of coordinates or angles

Physics biasing
• Changes production of primary or secondary particles
• Changes relative branching fractions

User-defined biasing
• Available through the G4WrapperProcess interface
• For situations not covered by the built-in algorithms
5
 Geant4 version 10.0.p01

Geometric Sampling (3.7.1)

Intended for non-active material, such as shielding

• Uses scoring to assign a weight to a generated track, and

accept/reject algorithms to evaluate that weight

• Importance sampling uses geometrical splitting plus

“Russian roulette” to select tracks through each cell

• Weight roulette uses windows (upper and lower bounds) or

a simple cutoff to keep or remove tracks

6
 Geant4 version 10.0.p01

Geometric Sampling (3.7.1)

Scoring done with G4MultiFunctionalDetector

• Scorers may themselves be biased; see

G4MultiFunctionDetector (different class!)

• Scoring and importance sampling apply to particle types

chosen by the user, not globally.
- examples/extended/biasing
- examples/advanced/Tiara

7
 Geant4 version 10.0.p01

Geometry Models (3.7.1.1)

Biasing generally requires a parallel geometry equivalent to

the detector model (mass geometry) used for simulation

• World volumes for parallel and mass must be identical

• Divide large non-sensitive volumes into cells
(G4GeometryCell)

• Cells and parameters collected into a store (G4IStore, etc.)

• Volume must be fully populated with cells (no holes!)
• Cells must not share boundaries with world volume

8
 Geant4 version 10.0.p01

Geometry Models (3.7.1.1)

class B02ImportanceDetectorConstruction : public G4VUserParallelWorld
{ ... };

G4VPhysicalVolume* ghostWorld = pdet->GetWorldVolume();

G4GeometrySampler pgs(ghostWorld,"neutron");
pgs.SetParallel(true);
...
pgs.PrepareImportanceSampling(&aIstore, 0);
pgs.Configure();

9
 Geant4 version 10.0.p01

Importance Sampling (3.7.1.3)

Must have good understanding of problem physics

• Which particle types require importance sampling?
• Define the cells appropriately (size, location)
• Assign importance values to the cells
If not done properly, results cannot be interpreted as
describing real experiment
Importance store used to store values related to cells
• User creates an object which inherits from G4VIstore: built-
in G4IStore may be used
• Constructed with reference to the world-volume of the
geometry (mass or parallel) used for sampling
• User fills the store with cells and their importance values
10
 Geant4 version 10.0.p01

Importance Store
class G4IStore : public G4VIStore {
public:
explicit G4IStore(const G4VPhysicalVolume &worldvolume);
virtual ~G4IStore();
virtual G4double GetImportance(const G4GeometryCell &gCell) const;
virtual G4bool IsKnown(const G4GeometryCell &gCell) const;
virtual const G4VPhysicalVolume &GetWorldVolume() const;
void AddImportanceGeometryCell(G4double importance,
const G4GeometryCell &gCell);
void AddImportanceGeometryCell(G4double importance,
const G4VPhysicalVolume &, G4int aRepNum=0);
void ChangeImportance(G4double importance, const G4GeometryCell &gCell);
void ChangeImportance(G4double importance, const G4VPhysicalVolume &,
G4int aRepNum=0);
G4double GetImportance(const G4VPhysicalVolume &, G4int aRepNum=0) const;
private: .....
};

11
 Geant4 version 10.0.p01

Importance Sampling Algorithm

Value must be assigned to every cell

Store must be fully occupied (no missing cells)

User creates class inheriting from G4VImportanceAlgorithm

• Built-in G4ImportanceAlgorithm will be used if none passed to
sampler
- Cell is not in store: causes an exception (job terminates)
- Importance = 0: Tracks of the chosen particle type will be killed
- Importance > 0: Normal allowed values
- Importance < 0: Not allowed!
• User-defined algorithm must handle all cases above, perhaps with
different behaviour
12
 Geant4 version 10.0.p01

Weight Windows (3.7.1.5)

Alternative to importance sampling

• Applies splitting and Russian roulette depending on space (cells)
and energy
• User defines weight windows in contrast to defining importance
values as in importance sampling
• Importance sampling is “weight blind,” this technique uses particle
weight in evaluation

Apply in combination with other variance reduction techniques

• cross-section biasing, implicit capture
• A weight window may be specified for every geometric cell and for
different energy ranges
• Space-energy cell, using same G4GeometryCell as importance
sampling
• Cells with no window will pass all particles through
13
 Geant4 version 10.0.p01

Weight Window Definition

• Global lower bound WL (all cells)

• Scale factors CS and CU for each
cell, or globally
• Upper bound WU = CUWL
• Survival cut WS = CSWL
• CS = CU = 1 equivalent to
importance sampling
• User may apply at boundaries,
on collisions or both

Cells stored in subclass of G4VWeightWindowStore

•Built-in implementation G4WeightWindowStore available
•Tracks below WL may be killed, or weight reset to WS
•Tracks above WU are split (replicated n times, nWS ≤ WU)
14
 Geant4 version 10.0.p01

Weight roulette (3.7.1.6)

Also called weight cutoff

• Weight of particle may become so low that no result can change
significantly: propagating all tracks wastes computing time.
• Usually applied if importance sampling and implicit capture are used
together.
• Implicit capture reduces particle’s weight at every collision, instead
of killing outright with some probability.

Weight roulette scales particle’s weight by importance ratio

R = Is/Ic of current cell (Ic) and original source (Is)
• If weight falls below a lower bound, Russian roulette is applied
• If particle survives, weight is reset to a specified survival weight

15
 Geant4 version 10.0.p01

Physics Based Biasing (3.7.2)

Replaces the natural distribution of some process with

“fake” PDFs that limit events to what is useful for your
simulation

• Primary particle biasing: e.g., cosmic ray experiments

• Radioactive decay biasing: e.g., shielding or underground
detectors
• Hadronic leading-particle biasing: only the highest-energy
secondary(ies) at each step of a shower are kept
• Hadronic cross-section biasing: cross-sections or branching
ratios can be arbitrarily rescaled

16
 Geant4 version 10.0.p01

Primary Particle (3.7.2.1.1)

Increases number of primary particles generated

in particular phase space region of interest
• Primary particle’s weight is modified as appropriate
• Implemented in G4GeneralParticleSource
• Possible to bias position, angular and/or energy distributions
Use in subclass of G4VUserPrimaryGeneratorAction

MyPrimaryGeneratorAction::MyPrimaryGeneratorAction() {
generator = new G4GeneralParticleSource;
}

void MyPrimaryGeneratorAction::GeneratePrimaries(G4Event* anEvent) {

generator->GeneratePrimaryVertex(anEvent);
}

17
 Geant4 version 10.0.p01

Radioactive decay (3.7.2.1.2)

G4RadioactiveDecay simulates the decay of radioactive

nuclei, with optional biasing
• Increase sampling rate of radionclides within observation
times through user-defined probability distribution function
• Nuclear splitting, where parent nuclide is split into a user
• defined number of nuclides
• Branching ratio biasing where branching ratios are sampled
with equal probability

This is a process: register in physics list for G4GenericIon

Examples in examples/extended/radioactivedecay
18
 Geant4 version 10.0.p01

Hadronic Leading Particle (3.7.2.1.3)

Keeps only most important part of each event, as well as

representative tracks of each given particle type:
• Track with highest energy
• One of each of baryon (p, n), π0, meson (π±, K), lepton
• Appropriate weights are assigned to the particles

Implemented in G4HadLeadBias utility

Environment variable SwitchLeadBiasOn activates

19
 Geant4 version 10.0.p01

Hadronic cross-section (3.7.2.1.4)

Artificially enhances/reduces cross section(s) for some

process
• Useful for studying thin layer interactions or thick layer
shielding
• Photon inelastic, electron nuclear and positron nuclear
processes

More details can be found in a talk presented at TRIUMF

http://legacyweb.triumf.ca/geant4-03/talks/03-Wednesday-AM-1/03-
J.Wellisch/biasing.hadronics.pdf

20
 Geant4 version 10.0.p01

Cross-Section Biasing

Controlled via
G4HadronicProcess::BiasCrossSectionByFactor()

void MyPhysicsList::ConstructProcess() {
...
G4ElectroNuclearReaction *theElectroReaction =
new G4ElectroNuclearReaction;
G4ElectronNuclearProcess theElectronNuclearProcess;
theElectronNuclearProcess.RegisterMe(theElectroReaction);
theElectronNuclearProcess.BiasCrossSectionByFactor(100);
pManager->AddDiscreteProcess(&theElectronNuclearProcess);
...
}

21
 Geant4 version 10.0.p01

User Defined Biasing (3.7.2.2)

G4WrapperProcess can be used to implement user

defined event biasing
• G4WrapperProcess, which is a process itself, wraps an existing
process
• All function calls forwarded to wrapped process
• Non-invasive way to modify behaviour of existing (built-in) process

1. Create derived class inheriting from G4WrapperProcess

2. Override only the methods to be modified, e.g.,
PostStepDoIt()
3. Register this class in place of the original
4. Finally, register the original (wrapped) process with user class

22
 Geant4 version 10.0.p01

User Defined Biasing

class MyWrapperProcess : public G4WrapperProcess {
...
G4VParticleChange* PostStepDoIt(const G4Track& track,
const G4Step& step) {
// Do something interesting
}
};

void MyPhysicsList::ConstructProcess() {
...
G4LowEnergyBremsstrahlung* bremProcess =
new G4LowEnergyBremsstrahlung();
MyWrapperProcess* wrapper = new MyWrapperProcess();
wrapper->RegisterProcess(bremProcess);
processManager->AddProcess(wrapper);
}

23
 Geant4 version 10.0.p01

General Biasing Interface (new for G4 10.0)

More “toolkit-based” approach to user-defined biasing, with

pre-written tools using common interface.

G4BiasingProcessInterface
• Wrapper class for physics processes

G4VBiasingOperation
• Base class to define single biasing action

G4VBiasingOperator
• Configuration to apply one or more operations
24
 Geant4 version 10.0.p01

General Biasing: Discrete Process Example

GetPostStepPhysicalInteractionLength()
• Returns the distance at which the
process will make an interaction
75 cm • Analog exponential law is at play
G4PhotoelectricEffect
• Above analog behavior
superseded by biased behavior (if
wished)
5 cm 1. PostStepDoIt()
G4GammaConversion
• Called if the process has
responded the shortest of the
interaction distances
9 cm
G4ComptonSca2ering • Generate final state, according to
specific process analog physical
law
• Above analog behavior
superseded by biased behavior (if
wished)

25
 Geant4 version 10.0.p01

General Biasing: Discrete Process Example

GetPostStepPhysicalInteractionLength()
• Returns the distance at which the
process will make an interaction
75 cm • Analog exponential law is at play
G4PhotoelectricEffect
• Above analog behavior
superseded by biased behavior (if
e+ wished)
5 cm PostStepDoIt()
G4GammaConversion
• Called if the process has
e -
responded the shortest of the
interaction distances
9 cm
G4ComptonSca2ering • Generate final state, according to
specific process analog physical
law
• Above analog behavior
superseded by biased behavior (if
wished)

26
 Geant4 version 10.0.p01

General Biasing: Discrete Process Example

G4BiasingProcessInterface wrapper intercepts physics-
related function calls
GetPostStepPhysicalInteractionLength()
• Returns the distance at which the
G4BiasingProcessInterface process will make an interaction
12 cm • Analog exponential law is at play
G4PhotoelectricEffect
• Above analog behavior
superseded by biased behavior (if
G4BiasingProcessInterface wished)
10 cm PostStepDoIt()
G4GammaConversion
• Called if the process has
responded the shortest of the
G4BiasingProcessInterface interaction distances
7 cm
G4ComptonSca2ering • Generate final state, according to
specific process analog physical
law
• Above analog behavior
superseded by biased behavior (if
wished)

27
 Geant4 version 10.0.p01

General Biasing: Discrete Process Example

G4BiasingProcessInterface wrapper intercepts physics-
related function calls
GetPostStepPhysicalInteractionLength()
• Returns the distance at which the
G4BiasingProcessInterface process will make an interaction
12 cm • Analog exponential law is at play
G4PhotoelectricEffect
• Above analog behavior
superseded by biased behavior (if
G4BiasingProcessInterface wished)
10 cm PostStepDoIt()
G4GammaConversion
• Called if the process has
γ responded the shortest of the
G4BiasingProcessInterface interaction distances
7 cm
G4ComptonSca2ering • Generate final state, according to
e -
specific process analog physical
e-
e- e
-
law
• Above analog behavior
superseded by biased behavior (if
wished)

28
 Geant4 version 10.0.p01

Example: Bremsstrahlung Splitting

• Implemented via G4WrapperProcess (details shown)

• Assume only interested in scoring photon hits

• Increase MC performance by reducing tracking of

secondary electrons

• Demonstrates biasing through enhanced production of

secondaries

29
 Geant4 version 10.0.p01

Brem Splitting Algorithm

Sample photon energy, angular distributions N times

Generate N unique secondaries (vs. once per interaction)

• This is called “splitting”
• Don’t confuse with importance sample splitting, where N
identical copies are created

Reduce electron energy by just one of the chosen photons

Remove bias introduced in photon energy and angular

distributions
• Assign each secondary a statistical weight
• wsec = wparent / N
30
 Geant4 version 10.0.p01

Brem Splitting Implementation

User process inherits from G4WrapperProcess

class BremSplittingProcess : public G4WrapperProcess {

public:
BremSplittingProcess();
virtual ~BremSplittingProcess();

// Override only this method

G4VParticleChange* PostStepDoIt(const G4Track& track,
const G4Step& step);
private:
G4int fNSplit; // Number of secondaries per split
};

31
 Geant4 version 10.0.p01

Brem Splitting Implementation

G4VParticleChange*
BremSplittingProcess::PostStepDoIt(const G4Track& track,
const G4Step& step) {
G4double weight = track.GetWeight()/fNSplit;
...
std::vector<G4Track*> secondaries(fNSplit); // Secondary store

// Loop over wrapped PSDI method to generate multiple secondaries

for (G4int i=0; i<fNSplit; i++) {
particleChange = pRegProcess->PostStepDoIt(track, step);
assert (0 != particleChange);
particleChange->SetVerboseLevel(0);

// Save the secondaries generated on this cycle

for (G4int j=0; j<particleChange->GetNumberOfSecondaries(); j++) {
secondaries.push_back(new G4Track(*(particleChange->GetSecondary(j))));
}
}
...
}

32
 Geant4 version 10.0.p01

Brem Splitting Implementation

Register wrapped process and wrapper with process manager

G4LowEnergyBremsstrahlung* bremProcess =
new G4LowEnergyBremsstrahlung();

BremSplittingProcess* bremSlitting =
new BremSplittingProcess;

bremSplitting->RegisterProcess(bremProcess);
pmanager->AddProcess(bremSplitting);

33
 Geant4 version 10.0.p01

Uniform Bremsstrahlung Splitting

No Splitting Splitting Factor 100

Scoring
Surface
34