Variance reduction 'secondary biasing' unclear... #33
Description
Hi all, I hope it's ok for me to add issues here, thought maybe it's easier for tracking than the forum...
It now appears that that geometric russian roulette is implemented an entirely separate process called DirectionalFilter. I understand the principle of this, but there must be numbers involved. If a particle is going outside the ROI, then russian roulette is applied - e.g. the particle must has a 1/10 chance of surviving, an it's weight increased by a factor of 10. But what are the numbers involved? Are these derived from the ParticleSplit? What happens if DirectionalFilter is implemented with no ParticleSplit?