KDSource, a tool for the generation of Monte Carlo particle sources using kernel density estimation

TL;DR

KDSource leverages kernel density estimation to generate synthetic Monte Carlo particle sources, improving statistical accuracy in complex radiation transport simulations, demonstrated through analytical and neutron beam modeling.

Contribution

Introduces KDSource, a novel tool using adaptive KDE for modeling and sampling particle sources in Monte Carlo simulations, enhancing accuracy and usability.

Findings

Good agreement with experimental neutron beam data

Effective modeling of complex particle distributions

Improved statistical sampling in simulations

Abstract

Monte Carlo radiation transport simulations have clearly contributed to improve the design of nuclear systems. When performing in-beam or shielding simulations a complexity arises due to the fact that particles must be tracked to regions far from the original source or behind the shielding, often lacking sufficient statistics. Different possibilities to overcome this problem such as using particle lists or generating synthetic sources have already been reported. In this work we present a new approach by using the adaptive multivariate kernel density estimator (KDE) method. This concept was implemented in KDSource, a general tool for modelling, optimizing and sampling KDE sources, which provides a convenient user interface. The basic properties of the method were studied in an analytical problem with a known density distribution. Furthermore, the tool was used in two Monte Carlo simulations that modelled neutron beams, which showed good agreement with experimental results.