Monte Carlo neutronics is the prediction of how neutrons and photons move through a nuclear machine by tracking millions of individual simulated particle histories, each step sampled at random from measured nuclear cross-section data, in order to estimate radiation dose, nuclear heating, and material activation.
The method solves the neutron transport equation without discretizing it. A source neutron — in a 📝fusion device usually the 📝14 MeV product of the 📝D-T reaction — is born with a sampled energy and direction, travels a randomly sampled distance to its next collision, and then scatters, is absorbed, or leaks out; the histories are scored into tallies whose statistical uncertainty falls as one over the square root of the number of particles run. That is the trade against deterministic transport codes, which discretize space, angle and energy onto a mesh: deterministic methods are fast and return a smooth answer everywhere but carry mesh and energy-group approximations, while Monte Carlo handles complicated geometry and continuous-energy physics exactly and pays for it in computer time. MCNP, maintained at Los Alamos National Laboratory, and OpenMC, the open-source code originating at MIT, are the two codes most used for fusion work. What they produce is the 📝neutron flux everywhere in the machine, and from it the 📝shielding thickness a magnet needs, the 📝neutron activation inventory that governs maintenance access, and the 📝tritium breeding ratio a 📝blanket achieves.
Two practical problems dominate the fusion application. The first is deep penetration: the quantity of interest behind a thick shield can be ten or more orders of magnitude below the source, so an unbiased random walk almost never delivers a particle there. Variance reduction fixes this by deliberately biasing the walk and correcting each score with a statistical weight — importance splitting, weight windows, and schemes that derive those weights from a cheap adjoint deterministic solve. Saltos, Peterson, Napolitano, Kennedy and Hodgdon evaluated these techniques for fusion shielding in Transactions of the American Nuclear Society 127 (2022), 238–241. The second is geometry: a tokamak shield exists as engineering CAD, not as the constructive solid geometry a transport code expects, and rebuilding it by hand costs weeks and then drifts out of date. CAD-based neutronics converts the engineering model directly into transport geometry, the approach described by Carter, Johnson, Wart and Kennedy in Transactions of the American Nuclear Society 129 (2023), 280–283.
Every number a neutronics model produces for a machine still under construction is a prediction, not a measurement. Confidence in it comes from validating the codes and the underlying nuclear data against benchmark irradiation experiments performed elsewhere. Measured neutron output from an operating 📝plasma is a different discipline — see 📝neutron diagnostics.