Population-based variance reduction (PVR) is a novel Monte Carlo variance reduction technique with primary application in Dynamic Monte Carlo (DMC). DMC, being a method of direct calculation of time dependence of Nuclear Reactor dynamic behavior, suffers from unstable statistical properties, as neutron population size decreases with neutron escape. As a result, maintaining the neutron population is a martingale game with a high probability of extinction within practical simulation timeframes. To counteract this, frequent maintenance of the population is paramount, a population tailoring (combing) process may take place 1 million times within a single reactor transient simulation. Tactics of population combing vary, recent publications offer several choices. Some techniques developed at BME NTI rely on optimizing combing parameters with respect to transport simulation variance on top of the conventionally used importance function. Moreover, it has been shown that the so-called Weight Window (WW) technique, the most wide-spread, half a century old variance reduction tool shares the optimization framework with PVR, while its theoretical grounding is false, as demonstrated by many experts of the field. The research topic involves the theoretical founding of WW and PVR based on an existing mathematical framework developed at NTI that needs to be verified by numerical simulations. Based on the correct mathematical assumptions proper optimization of PPVR should show that the variance can be further decreased. PVR should be extended from DMC application to fixed-source source-detector problems, where the population is combed at collision events. The techniques should be implemented into the GPU-based DMC code GUARDYAN being developed at NTI. Extension GUARDYAN to handle coupled neutron-photon problems should also be carried out and PVR techniques should be tested against coupled multiparticle simulations as well.
Population-based Variance Reduction for dynamic Monte Carlo and fixed-source calculations
Fizikai Tudományok Doktori Iskola
Deep understanding of Monte Carlo methods, background in variance reduction method development, excellent numerical mathematical knowledge, sound programming skills, good command of English language
PhD project for standard admission