Choice of Biasing Function for Source Biasing

Abstract

Two different descriptions have been used for Monte Carlo source biasing. One relies on a direct optimization of biasing parameters, the other on an intuitive application of the adjoint flux. But use of the adjoint flux is based on the assumption that importance sampling will be used throughout the calculation, and that source sampling will not be stratified. It is shown that if these conditions are not satisfied, use of the importance functions has no theoretical justification and, in principle, biasing parameters must be optimized directly. Source biasing is probably the simplest, and one of the oldest, of all variance-reduction techniques, and one might suppose that it is, by now, perfectly well understood.