Accelerating Monte Carlo by variable fidelity: First results

In many reactor calculations, high fidelity, high accuracy results are required only in a small spatial region. In order to save computational resources in such cases, a Monte Carlo scheme utilising a variation of neutron tracking fidelity depending on a phase space position of a particle is proposed. The paper estimates a lower bound on accuracy of a simple example of such variable fidelity scheme, which determines tracking fidelity based on spatial position in a simple ID PWR lattice test case. It also estimates the magnitude of speed-up expected by the reduction of fidelity of a Monte Carlo solution. Different fidelity levels of a Monte Carlo transport calculation were obtained by combination of a continuous energy (CE) and a multi-group (MG) representation of nuclear interaction data with flux-volume weighted geometry homogenisation. Because none of the existing tools supports both the MG and the CE data in a single calculation, a two step methodology was utilised. First step consisted of a full geometry, reduced fidelity eigenvalue calculation in MONK 10A. In the second stage, partial current sampled in the first step was used in fixed source calculation mode Serpent 2, in which, part of domain was replaced with a vacuum boundary and a neutron source. Moreover, the influence of complexity of a geometry and of nuclear data representation on MONK 10A runtime was investigated. The obtained results suggest that significant acceleration of a Monte Carlo solution can be obtained by simplifications to problem geometry and nuclear data. Moreover, it was shown that a variable fidelity calculation could provide with a reasonable accuracy of fission rate distribution with error below 1% except for the row of pins next to the boundary,