Monte-Carlo neoclassical transport coefficient calculations in 3D tokamak plasmas via GPU-accelerated particle simulations

Understanding neoclassical particle and heat transport in the presence of 3D magnetic perturbations is essential for the operation of tokamaks. Traditionally, these coefficients have been calculated by directly solving the drift-kinetic equation. Monte-Carlo methods offer a more direct approach but are limited by computational constraints due to the inherent statistical noise of the method, requiring a large number of particles for convergence. Our work introduces a GPU-accelerated version of the VENUS-LEVIS particle transport code that addresses these constraints through massively parallel simulations, thus achieving statistically acceptable uncertainty levels.

The new implementation of VENUS-LEVIS is architecture independent and can run on a laptop CPU or a large scale GPU cluster. The high-performance code allows for full-f Monte Carlo calculations of neoclassical coefficients. The method naturally captures the effects of different collisionality regimes from banana to Pfirsch-Schlüter, and the influence of non-axisymmetric perturbations. This method is crucial for extending the transport code beyond ideal MHD, such as Neoclassical Tearing Modes.

The computational improvements offer direct benefits for investigating heavy impurity transport phenomena, which are critical for ITER [6] and future reactor designs incorporating tungsten plasma-facing components. As tungsten interactes with the background plasma, accurate modelling of transport coefficient are required.