Understanding and investigating the relationships between geometrical errors and lost particles in MCNP

Generating Constructive Solid Geometry (CSG) ready for MCNP input using automated programs, and the use of universe structures, may give raise to geometrical inconsistencies leading to numerical phenomena known as lost particles, which perturb the statistical reliability of the transport solution. In the context of the reference nuclear analysis models developed for ITER, there is a need to better understand these phenomena, define a sufficiently explicit way to test the models, and provide clear and uniform parameters indicating their technical quality and reliability in terms of geometrical errors. Using a source able to generate an isotropic and uniform fluence inside its closed surface, it was possible to derive a relationship between source's surface area, lost particles rate and error size. These results were verified with multiple MCNP runs and good agreement was found between simulations results and analytical predictions. This work allows defining more explicit and uniform procedures to debug for lost particles, and provides a way to grade model quality based on objective parameters, like the size of errors, instead of using others that heavily depend on the setup on which the simulation was run.