Detector-Targeted Weight Window Generation via Absorbing Markov Chain Reachability in Monte Carlo Transport

This work presents a detector-targeted, data-driven weight window (WW) generation procedure based on an absorbing Markov chain surrogate of neutron transport. A collision-indexed transition model is estimated from track-based tallies on a discretized state space, after which a sparse reachability solve yields the probability that a history starting from each state will eventually reach a user-specified detector region before termination. This reachability probability is used as an importance proxy to construct WW bounds for production simulations. The method is demonstrated on two-dimensional labyrinth streaming configurations derived from the Protvino benchmark, including a modified geometry with an additional bend. The resulting WW estimators remain statistically consistent with analog sampling while improving efficiency by approximately 10× in figure of merit (FOM) for the reference case and near 400× for the modified case. These results indicate that reachability-based, detector-specific importances can provide an inexpensive and effective pathway to automated WW generation for labyrinth streaming problems.