Geometry based biasing

This example illustrates a use of generic biasing classes to implement a technique near to "geometry importance biasing".

The geometry is the same than in EM tests, with the sampling calorimeter made of a series of layers of absorber and gap.

The biasing applies to neutrons only.

Instead of explicitely assigning "importance" values to the layers, we split neutrons moving forward and kill the ones moving backward, when they reach the exit of an absorber volume.

The splitting factor can be controlled by command line, eg:

/GB03/biasing/setSplittingFactor 2

which also determines the killing probability : 1/(splitting factor).

It can be seen than when defining 10 layers (see exampleGB03.in), a splitting factor 2 works fine : we don't suffer from under- or over-splitting. If going to 20 layers, then a splitting with a factor 2 is too large, and the biasing suffers from over-splitting. (And we can not go lower than "2", which would mean "1" and hence, no biasing...)

To alleviate the over-splitting, we introduce a probability to apply the splitting (and killing) (this is one solution, others can be considered), that can be changed as:

/GB03/biasing/setApplyProbability 0.5

With above value, we can see that we recover a satisfactory biasing scheme, with neutrons penetrating the entire setup, without over-splitting.

The classes involved are:

GB03BOptnSplitOrKillOnBoundary : which is the biasing operation making the splitting and killing;
GB03BOptrGeometryBasedBiasing : which is the biasing operator, making decision to use above operation, and configuring it, passing it the splitting factor and probability to apply the biasing.