Research on global neighbor list method in Monte Carlo code RMC

Abstract

Ray tracing algorithm is a main particle-tracking algorithm that has been employed in Monte Carlo transport codes. Inside the algorithm, a significantly time-consuming process is to find the neighbor cell when a particle crosses a surface. This paper proposed a new algorithm, called the Global Neighbor List (GNL) method, to accelerate the neighbor-finding process. The GNL method utilizes the searching directly between the cells in the bottom layer and uses the thread-shared data structure to achieve parallel acceleration as well as to reduce the average memory footprint of a single thread. The method is implemented and tested in Reactor Monte Carlo (RMC) code. The results show that, in the criticality calculation of the VERA benchmark 5 core model, the GNL method is 29.2% faster than the surface-based neighbor list method. In the case of 200 layers of fuel pin model, the neighbor-finding process can achieve an acceleration of 19 times.