Splitting by cross-section

This example illustrates a technique that uses physics cross-sections to determine the splitting [killing] rate in a shielding problem. This technique is supposed to be an invention, and this example here is not optimized. The technique is applied here to neutrons.

In the classical treatment of the shielding problem, the shield is divided in slices at the boundaries of which particles are splitted[killed] if moving forward[backward]. In the present technique, we collect the cross-section of "absorbing/destroying" processes : decay, capture, inelastic. We then use the generic biasing facilities to create an equivalent of a spitting process, that has a "cross-section" which is the sum of the previous ones. This process is competing with other processes, as a regular one. The occurence of this process is hence the same than the "absorbing/destroying" processes together. When this process wins the competition, it splits the track, with a splitting factor 2 (ie the original track is kept and a copy of it is created). This splitting is hence occuring at the same rate than the absorption, resulting in an expected maintained (unweighted) flux.

The geometry is made of a single block of concrete it. Behind it (in the +z direction) a thin empty volume is placed to print out the particles which are exiting the shield.

As in any generic biasing use, a biasing operator (taking decisions on what biasing to apply) and a biasing operation (applying these decisions) are defined. These are: GB05BOptrSplitAndKillByCrossSection for the operator, GB05BOptnSplitAndKillByCrossSection for the operation.

The operator is created in the detector construction, and receives here the names of the absorbing/destroying processes to counterbalance for. At tracking time, it collects the up to date cross-section of these processes in the ProposeNonPhysicsBiasingOperation(...) method, and passes the sum to the GB05BOptnSplitAndKillByCrossSection operation.

The operation uses the cross-section (interaction length) to sample the distance to "interaction" with a classical exponential. If it wins the race (ie it proposes the smallest of the interaction distances among all processes) its GenerateBiasingFinalState(...) method is called, and it applies splitting or killing (Russian roulette) if the track moves forward or backward.