Hybrid Monte Carlo-Deterministic Methods for Nuclear Reactor-Related Criticality Calculations [report]

Edward W. Larson
2004 unpublished
The overall goal of this project is to develop, implement, and test new hybrid Monte Carlo-deterministic (or simply hybrid) methods for the more efficient and more accurate calculation of nuclear engineering criticality problems. These new methods will make use of two (philosophically and practically) very different techniques -the Monte Carlo technique, and the deterministic technique -which have been developed completely independently during the past 50 years. The concept of this proposal is
more » ... f this proposal is to merge these two approaches and develop fundamentally new computational techniques that enhance the strengths of the individual Monte Carlo and deterministic approaches, while minimizing their weaknesses. The Monte Carlo method does not suffer from the space-angle-energy grid truncation errors that affect deterministic schemes and is particularly advantageous for complex 3-D geometries. However, Monte Carlo criticality solutions have statistical errors, are error-prone in weakly coupled systems, and can be difficult to bias efficiently. This proposal aims to confront these difficulties by developing new "hybrid" schemes that use deterministic techniques to obtain more accurate Monte Carlo solutions. Two kinds of hybrid schemes have been developed along these lines for fixed-source neutron transport problems: (i) schemes that determine biasing parameters for the Monte Carlo simulation from approximate deterministic (usually, adjoint) calculations, and (ii) schemes that use more accurate functionals for desired responses, which make use of both deterministic (adjoint) and Monte Carlo (forward) information. Extending, testing, and optimizing these hybrid algorithms for criticality problems is the theoretical goal of this project. The practical goal is to achieve methods that can solve difficult, realistic Monte Carlo criticality problems more efficiently and reliably, and with significantly less user-input.
doi:10.2172/821706 fatcat:35uz35o7z5fxrdnyagv4yk25um