Development of vectorized Monte Carlo calculation and application of stratified sampling.

Abstract

In order to improve efficiency of Monte Carlo calculation, the authors developed some calculational schemes for multiple particle processing that enables further improvement in vectorization. Significant effort is directed toward achieving (1) an algorithm to supply particles consecutively when they escape out of the cell, (2) vectorization techniques on hardware equipped with multiple pipe lined processors and (3) development of a simple algorithm of the stratified sampling for the vector Monte Carlo calculation. As a sample problem, fast neutron transport in a graphite slab was calculated using 4 group constant. The authors succeeded in achieving 98.1% among the total process steps in vector mode. As a result of (1) and (2), the calculational time is reduced to about 1/11 relative to the time for the equivalent scalar Monte Carlo processing, carried out on the same hardware, HITAC S8lO. On the other hand, (3) the stratified sampling was found to reduce not only the calculational time but also the standa...