Geant4 simulations of neutron flux using automated weight windows with applications to ESS

Abstract: This thesis investigates the use of two automated weight window methods for variance reduction of flux in particle transport simulations. The methods are implemented in the Geant4 simulation software. Both methods are based on pre-simulations, using flux or relative error information as input. The methods are intended for use in shielding and instrumental background applications and are tested on flux and spectrum calculations using two versions of an European Spallation Source (ESS) instrument shielding target model. For each model, the methods were implemented by defining a parallel geometry mesh grid which overlaid the mass geometry. In each mesh cell the number of tracks in the cell and the relative error of the number of tracks were scored. This information was used to update the weight windows. Two figures of merit were introduced to assess the methods. The first figure of merit is inversely proportional to the square of the relative errors and the simulation time. The second figure of merit is the standard deviation of the relative errors. Both figures of merit showed improvement for the two methods and almost all simulation times. Compared to the analog simulation, the first figure of merit increased by a factor of 2.17 to 3.10 when applying the relative error based method and by a factor ranging from 2.10 to 14.60 when applying the flux based method. The second figure of merit decreased by a factor ranging from 1.33 to 2.89 when applying the relative error based method and increased by a factor ranging from 1.58 to 3.15 when applying the flux based method. The improvement depends on the simulation time and the simulated model. The methods showed improved figures of merit, reduced the parts of the simulated geometry that saw few Monte Carlo particles and reduced the error significantly in almost all spatial regions.