Jan Dufek

Research interests

I develop new numerical methods for linear and non-linear Monte Carlo neutron transport simulations. The methods are applicable in research and development of nuclear reactors. Below is a brief summary of my major research interests:

Coupled simulations

Conventional Monte Carlo reactor physics codes are designed for fixed-source and criticality calculations that analyse fissile systems with fixed properties, i.e. systems without feedbacks. To allow for realistic simulations of nuclear power reactors, Monte Carlo codes need to be coupled to other solvers representing various feedbacks. For conventional reactors, simulation of the thermal-hydraulic feedback is crucial. We focus on the development of stable and efficient numerical schemes that couple the Monte Carlo and thermal-hydraulic solvers.

Monte Carlo burnup simulations

Monte Carlo burnup codes allow simulations of nuclear reactors during the whole fuel cycle, while the fuel is being depleted. While the first codes of this kind appeared decades ago, their extensive computing cost has not allowed their application to full-core whole-cycle simulations. We focus on the development of stable and efficient numerical schemes that couple the Monte Carlo and fuel depletion solvers.

Fission source convergence in Monte Carlo criticality calculations

The extensive computing cost applies even to pure Monte Carlo criticality (not coupled) calculations alone. One of the reasons is the problematic convergence of the fission source; a large number of so-called inactive cycles need to be simulated before the fission source is converged. Inactive cycles represent a lost computing time since no results can be collected over these cycles. We develop techniques that accelerate the convergence of the fission source.

Fission matrix based Monte Carlo simulations

While the fission matrix has long been a quantity of interest to Monte Carlo physicists, its full potential has not been utilised until recently. We develop new concepts of fission matrix based Monte Carlo calculations that can achieve a much-improved efficiency in serial and parallel computing modes.

Nodal nuclear data models

We also develop deterministic methods. Deterministic reactor simulators require a large amount of nodal nuclear data to be pre-computed and to be available for various combinations of state parameters. Storing the nodal data in a table form becomes impractical when the number of state parameters grows above about a dozen or so as the table size grows exponentially with the number of parameters. We develop efficient methods for building polynomial-regression based nodal nuclear data models that do not limit the number of parameters.