MCNPX-PoliMi Variance Reduction Techniques for Simulating Neutron Scintillation Detector Response.

Abstract

Scintillation detectors have emerged as a viable He–3 replacement technology in the field of nuclear nonproliferation and safeguards. MCNPX–PoliMi is a Monte Carlo Code that has been used for simulating detailed scintillation physics; however, until now, simulations have only been done in analog mode. Analog Monte Carlo simulations can take long times to run. In this thesis, two nonanalog approaches to speed up MCNPX–PoliMi simulations of neutron scintillation detector response have been studied.

In the first approach, a response matrix method (RMM) is used to efficiently calculate neutron pulse height distributions (PHDs). This method combines the neutron current incident on the detector face with an MCNPX–PoliMi–calculated response matrix to generate PHDs. The PHD calculations and their associated uncertainty are compared for three different techniques: fully analog MCNPX–PoliMi, the RMM, and the RMM with source biasing. The RMM with source biasing reduces computation time or increases the figure–of–merit on an average by a factor of 600 for polyethylene and 300 for lead shielding (when compared to the fully analog calculation). The simulated neutron PHDs show good agreement with the laboratory measurements, thereby validating the RMM.

In the second approach, MCNPX–PoliMi simulations are performed with the aid of variance reduction techniques. This is done by separating the analog and nonanalog components of the simulations. Tally mechanisms are developed for PHDs, time–of–flight curves, and cross–correlations. Three laboratory measurements (bare, lead–shielded, and polyethylene shielded) are performed with a Cf–252 source to validate the nonanalog MCNPX–PoliMi cross–correlation simulations. For the bare cross–correlation case, the nonanalog simulation speedup was a factor of 3.4; for the lead–shielded case, the speedup was a factor of 16; and for the polyethylene–shielded case, the speedup was a factor of 2.6. The agreement of simulations with laboratory measurements was good.

In summary, this thesis demonstrates that known variance reduction techniques, when properly applied to nonlinear scintillation detector response problems, can significantly increase the figure–of–merit (sometimes by 2 or 3 orders of magnitude). This can yield major reductions in computation times and analyses for important homeland security problems.