Global variance reduction scheme with self-adaptive weight-window mesh in the Serpent 2 Monte Carlo Code

A global variance reduction (GVR) scheme with self-adaptive weight-window mesh was recently implemented in the Serpent 2 Monte Carlo code. Importances used to define the weight-window boundaries are obtained from a built-in deterministic solver, which applies the response matrix method to the adjoint transport problem. The mesh is based on an octree-type data structure, which can be refined by recursively dividing the mesh cells. The split criteria take into account the local density of the medium and the gradient of the importance distribution. The methodology is demonstrated by photon transportsimulations in a hot-cell geometry, which consists of both heavily shielded structures and large volumes of empty space. The results show that the weight-window mesh adaptation obeys the constraints imposed by the problem geometry, and that the variance reduction scheme provides significant improvement in computational performance compared to an analog Monte Carlo simulation