I am running medical linear accelerator simulation and using phase space as a particle generator. I have been studying some optimization techniques which not only can reduce run time but also provide better particle statistics about interaction with matter per voxels.
One of the techniques that I used is geometrical biasing. I collimated the particles in certain positions. This provides me a reduction of time but gives me worse statistics because of that less particle in the voxels. I am going to sum more phase spaces in order to get enough particles for the collimated region. But I wonder how many particles are required per voxel in order to get better statistics on distribution?