Monte Carlo (MC) neutron transport simulations can be performed with several types of nuclear data representations; the most common ones are continuous energy (CE) and multi-group (MG) data. CE is the highest fidelity option, since it allows each cross section resonance to be represented precisely. However, due to the need to perform lengthy binary searches and expensive memory look-ups, CE MC is normally computationally expensive. Some alternatives, such as MG MC, introduce approximations but are computationally cheaper. The benefits of the two data types can be combined to speed up MC calculations while retaining as much accuracy as possible. This thesis proposes and investigates such acceleration methods for MC, based on the use of variable fidelity nuclear data.

The first method proposed consists of approximating thermal cross sections with simple analytical functions, like high order polynomials and exponentials. During a CE MC calculation, whenever a cross section in the approximated energy range is needed, a cheap function evaluation can be done instead of more expensive operations. The method, applied to thermal reactor types, produced a speed-up of up to 15%, and is extremely simple to implement. While a small bias can be introduced in the results, several ways to minimise this have been successfully tested.

The second method proposed can accelerate fission source convergence in eigenvalue calculations by simulating most of the inactive cycles with MG nuclear data. The MG cross sections needed can be generated online at the beginning of a calculation, during a few initial CE cycles. On 3D reactor models with burnt fuel composition, the acceleration provided is the largest and it is up to a factor of 4. On the other hand, convergence to a slightly different fission source can introduce a small error in the results and an increase in the standard deviation between independent runs.

Finally, a new application area for MG MC was investigated, namely neutron transport in fusion reactors: the sources of error introduced by MG MC when simulating fusion neutronics models were quantified. A simplified spherical model was simulated; the results indicated that, to contain the error introduced within acceptable limits, high order scattering anisotropy must be accounted for, and angle-dependent MG cross sections must be used to compensate for the flux separability approximation. To produce a low error with MG MC fine energy group structures and material discretisations are necessary too. This part of the thesis did not focus directly on a variable fidelity nuclear data method, but paved the way for future research in that direction.