Abstract
Neutron clustering is a recently identified problem with Monte Carlo eigenvalue calculations which can produce significantly erroneous results. Previous work by Sutton & Mittal (2017) considered neutron clustering as a problem of maintaining ‘genetic diversity’ within the neutron population. This paper proposes reducing the extent of neutron clustering by replacing fission neutrons in the source bank with uncorrelated neutrons, sampled from a uniform source distribution – effectively adding new neutron genealogies to the population. The efficacy of the method is demonstrated on a number of simple problems, showing improved behaviour of the Shannon entropy and neutron centre-of-mass. Although currently limited in scope, this paper intends to provide a route to reducing clustering effects in more general problems.
Highlights
Monte Carlo eigenvalue solvers are increasingly popular within nuclear engineering and are regularly applied to the simulation of full-core reactor problems
Neutron clustering is problematic for Monte Carlo in that it can cause significantly erroneous solutions to be obtained, while being difficult to detect [2,3]
For large problems, or even those which have extreme aspect ratios, sufficiently large population sizes might be impractical. This could be due to i) the memory footprint of simulating many millions of particles per generation, and ii) the nature of the power iteration algorithm necessitating that a large number of generations be simulated to adequately converge the source – given that source convergence is nearly independent of the number of particles per generation, this can incur unnecessary computational expense
Summary
Monte Carlo eigenvalue solvers are increasingly popular within nuclear engineering and are regularly applied to the simulation of full-core reactor problems. For large problems, or even those which have extreme aspect ratios (e.g., a fuel pin), sufficiently large population sizes might be impractical This could be due to i) the memory footprint of simulating many millions of particles per generation, and ii) the nature of the power iteration algorithm necessitating that a large number of generations be simulated to adequately converge the source – given that source convergence is nearly independent of the number of particles per generation, this can incur unnecessary computational expense. By itself, this problem is not significant provided that said family of neutrons has spread itself across the geometry of interest. This paper will demonstrate some preliminary applications of the injection algorithm to some relatively simple problems with further developments pursued in future work
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
