PhD in Nuclear Science

Start date: 1 October 2026/nDuration: 4 years/nApplication deadline: 31 March 2026/nFunding source: AWE Nuclear Security Technologies/nFunding information: UKRI equivalent stipend; Year 1 £21,820.00, Year 2 £22,910.00, Year 3 £24,055.00, Year 4 £25,260.00/nAbout: Neutron-nucleus cross sections can be divided into different energy ranges depending on the initial energy of the neutron. At lower energies, cross sections are characterised by peaks, called resonances, caused by the neutron being absorbed by the target nucleus. Positions and widths of these resonances cannot be predicted and must be measured. This energy range is called the “resolved resonance region” (RRR). Increasing further the energy of the neutron we reach a point where the resonances in the cross section are too close to each other to be resolved experimentally and we can only infer the average values of the widths and spacings between two adjacent resonances. This energy range is called the “unresolved resonance region” (URR)./nCurrent computational methods treat the resonances in the URR using delta functions in place of full conditional cross section probability distributions to represent the probability of individual reaction channels (capture, elastic, fission), potentially missing more complex correlations between the channels. Moreover, this method does not allow to easily include information from different sources, for example from nuclear experiments. Additionally, for many applications we also need to know the cross sections at different temperatures and, thus, we need to properly account for the thermal motion of the target nuclei./nDue to these issues, we need to develop a theoretical framework that allows us to consistently treat the URR, the available experimental information, and the target thermal motion of the cross sections of neutron-induced reactions relevant for nuclear science and applications. This project aims to develop the required formalism using modern probabilistic and machine-learning approaches, reformulating the problem in terms of conditional probabilities. Bayesian networks and related machine-learning methods will be used to calculate cross-section probability density functions in a much faster way, enabling the combination of multiple probability distributions describing various physical effects.