Physics-oriented optimization strategy for the energy lookup algorithm in continuous energy Monte Carlo neutron transport simulation

摘要

The continuous energy Monte Carlo method is a most high-fidelity and high-resolution method for neutron transport simulations in reactor physics with minimal approximations. However, one of the major disadvantages is that it is very time-consuming and computational-intensive for large scale whole core simulations, especially for coupling with depletion analysis for realistic reactors. Some recent researches indicate that one of the principle performance bottlenecks for the problem lies in the energy lookup algorithm during the calculation of energy-dependent material cross sections. Therefore, two physics-oriented optimization strategies, based on making use of the physical characteristics of neutron transport behaviors, are developed to optimize the run-time performance of the algorithm for accelerating the energy lookup without any loss in precision and accuracy. The first optimization strategy is called Neighbored Material Cascade Grid (NMCG) which is a hybrid approach utilizing the key features of the cascade grid and double indexing method. The second optimization strategy is called Adaptive Optimal Logarithmic Grid (AOLG) which is a variation of the conventional logarithmic energy grid method utilizing the advantages of energy hash tables. The strategies are incorporated into a continuous energy Monte Carlo neutron transport code and tested on realistic whole-core reactor systems. The computational performance as measured by memory usage, elapsed runtime and overall speedup, associated with each of the optimization strategies are demonstrated in the whole-core Monte Carlo simulations. Depending on the complexity of the models, the number of nuclides in the material compositions and the utilization of different optimization strategies, overall speedup ratios of 1.2-1.7, relative to the conventional binary lookup algorithm, are routinely observed. Furthermore, the numerical results indicate that the runtime performance of the new physics-oriented optimization st