Randomized Quasi-Monte Carlo Sampling in The Random Ray Method for Neutron Transport Simulations

TL;DR

This paper introduces a randomized Quasi-Monte Carlo sampling method in neutron transport simulations, reducing variance and improving computational efficiency compared to traditional pseudorandom sampling.

Contribution

It demonstrates the application of Quasi-Monte Carlo sampling in neutron transport, achieving lower error and faster runtimes in a practical simulation environment.

Findings

Over 10% reduction in pin-cell error

Up to 8% overall speedup

Improved variance reduction in simulations

Abstract

In a random ray method of neutral particle transport simulation, each iteration begins by sampling a set of rays before proceeding to solve the characteristic transport equation along the linear paths the rays follow. Historically, traditional pseudorandom number generators have always been used when sampling ray starting points and directions. In the present work, we experiment with using a randomized Quasi-Monte Carlo sampling technique in place of the traditionally fully random technique with the goal of reducing the variance of simulation results. To evaluate the numerical and runtime performance of the new sampling scheme, it is implemented in the random ray solver of the OpenMC particle transport application and tested on the 2D C5G7 benchmark simulation. The new scheme delivered over a 10% reduction in the average pin-cell error and up to an 8% overall speedup compared to the state-of-the-art, owing both to numerical reductions in uncertainty as well as practical reductions in shared memory contention.