Distributed Memory Algorithms for Weight Cancellation in Monte Carlo Particle Transport Simulations

Abstract

Recent literature has demonstrated use cases for Monte Carlo transport simulations where particles can have statistical weights that are positive or negative. There are even examples which require particles to have complex statistical weights, and the real and imaginary components can be positive or negative. In such cases, weight cancellation algorithms can be very efficient at reducing the variance, or might even be required for a simulation to converge. Previous works that have employed weight cancellation in distributed memory simulations required that all fission particles be sent to a single node for the cancellation operation. This work examines possible implementations of distributed memory weight cancellation algorithms that do not require the transfer of the fission source to a single node.