The performance of a tokamak strongly depends on the turbulent dynamics in the plasma boundary region, which separates the core plasma from the vessel's plasma facing components. In fact, the boundary region regulates the refueling of the main plasma as well as its impurity contamination, while simultaneously determining the peak particle and heat fluxes impinging on material surfaces. Improving the understanding of the boundary region is therefore of critical interest to interpret existing experiments and support the design of future devices and, ultimately, fusion power plants.
Plasma turbulence in the boundary region, which hosts significant density and temperature gradients, requires the consideration of strongly non-linear, spatially anisotropic dynamics over a wide range of temporal and spatial scales. Owing to its proximity with colder material surfaces, the boundary is also host to neutral particles, whose dynamics decouple from the surrounding magnetic field. Modeling such an environment therefore represents a complex exercise calling for numerical simulations.
Turbulence simulation in the boundary of fusion devices is normally carried by means of three-dimensional drift-reduced Braginskii models. Because of their substantial computation cost, state of the art codes only recently have approached the simulation of realistic scenarios in present-day devices. Particularly relevant to the exhaust problem foreseen in future reactors is the accurate simulation of divertors.
In this thesis, the GBS simulation code for the plasma dynamics and the KINDNES module for the neutral dynamics are extended from a Cartesian simulation mesh to realistic first wall geometry. In the interest of modeling divertor shaping in GBS, an approach based on curvilinear (structured) meshes is devised, in particular avoiding immersed boundary conditions and retaining the possibility to discretize the guiding equations using high-order finite differences. KINDNES is then extended specifically imposing that plasma and neutral simulation volumes match exactly. A hybrid mesh is employed to simultaneously capture fine-grained first wall features while the volume interior is coarsely discretized to limit memory consumption.
L-mode simulations of baffled and unbaffled TCV configurations are then carried out, both excluding and including neutral dynamics, and compared to existing literature where applicable. The baffled configurations, where divertor closure is increased, are notably shown to retrieve an increased neutral compression factor and present evidence of reduced plasma plugging when the clearance between plasma and baffles is increased.
