Speaker
Description
We present KINDNES (KINetic Deterministic NEutral Solver), a tool for solving the kinetic Boltzmann equation applied to neutral particle dynamics in the boundary plasma of tokamaks and stellarators. Originally developed as the neutral model within the GBS turbulence code, KINDNES is now developed as a standalone library.
Neutral particles play a central role in edge plasma physics, governing key processes such as recycling, momentum and energy exchange, and the onset of plasma detachment. While existing neutral models typically rely either on fluid approximations or on Monte Carlo kinetic methods, KINDNES provides a fully kinetic, deterministic framework, free of statistical noise. The model discretizes the Boltzmann equation for each neutral species and integrates it along its characteristics, reducing the problem to the inversion of a discrete linear system.
The fundamental physical mechanisms and numerical implementation of KINDNES are first presented, followed by its first benchmark against the Monte Carlo code EIRENE. Using a deuterium plasma background derived from a SOLPS simulation, the two codes are compared in an attached divertor configuration, with EIRENE configured to match the physical assumptions currently available in KINDNES. Within this common physics framework, the two codes show good agreement in atomic deuterium density, temperature and velocities, representing an encouraging validation of KINDNES as a deterministic alternative for neutral particle modeling.
Building on this validation, the most recent developments of KINDNES are then presented: the extension to molecular deuterium dynamics, enabling the simulation of detached divertor conditions; the introduction of flexible wall geometry support, broadening the range of device configurations that can be modeled; the development of a full 3D solver, that includes the proper tratment of toroidal velocities; the relaxation of the adiabatic approximation, rendering the solver aware of the plasma evolution in time and the adoption of a Hierarchical Matrices method to substantially accelerate the solution of the linear system in terms of both memory and computational cost.