Improved Monte Carlo Methods with Application to Borehole Logging Simulations

Monte Carlo simulation is the most accurate method in general for nuclear particle transport calculations. The convergence properties are, however, often outstandingly poor. The efficiency can be enhanced by replacing the simulation by a statistically equivalent calculation by modifying the sampling algorithms by so-called variance reduction techniques. One way of achieving such aim is the adjoint Monte Carlo method, where particle interactions are simulated in reverse. The Midway Monte Carlo method combines a regular Monte Carlo forward simulation with an adjoint one on a virtual surface separating source and detector resulting in an enhanced convergence rate. The coupling of forward and ajoint calculations is a statistically evaluation of a bilinear surface integral of the radiation current and the adjoint function in every phase-space variable.
This doctorate thesis develops and applies the time dependent form of the Midway Monte Carlo method to a pulsed neutron-photon oil well logging tool. Full analysis is given on the accuracy and convergence properties of the coupling possibilities and of the efficiency increase for such an application. Additionally, the theoretical framework of the adjoint simulation of scintillation detectors (Pulse Height tallies) has been developed.
The thesis concludes that the Midway Monte Carlo method delivers a user-friendly general variance reduction tool with high efficiency improvement, moreover the application of conventional variance reduction techniques can further enhance the efficiency. On the downside the reliability of the delivered answer and confidence intervals are often poor for time dependent simulations. Also, the method requires high computer memory capacity for the current standards.