Speaker
Description
The transport of neutral atoms and molecules produced from gas puff and the wall recycling process is a critical aspect of the core-edge integration challenge. Because of this, predicting the evolution of edge plasma profiles from simulation requires coupling to self-consistent predictions of the neutral gas. The Monte Carlo method is the most widely-used kinetic method for doing so. Coupling neutral and plasma physics remains at best a performance bottleneck or even, heretofore, unfeasible when long timesteps are desired with implicit solution methods [1]. As a result, kinetic neutral physics remains often neglected in implicit plasma solvers in favor of more efficient and differentiable fluid models. Algorithmic and software enhancements continue to be developed to address this issue [2], and we present a machine learning approach to train a deterministic surrogate model from stochastic simulation data. The Monte Carlo neutral transport solver DEGAS2 is used to efficiently and robustly predict kinetic neutral profiles against a background of more than 70,000 UEDGE simulations of KSTAR [3]. A UNet model is trained on this data to predict for a wide range of plasma profiles. This model is not only very efficiently evaluated, but so are derivatives with respect to the evolving plasma profile, unlocking the feasibility of fully kinetic and implicit neutral coupling to plasma simulation. The departure of the kinetic to fluid results is examined, as is the importance of neutral-neutral scattering in the former. The path toward training a mesh-agnostic model is explored, as is direct implementation of such models in implicit plasma simulations.
This work is supported by USDoE contracts DE-AC02-09CH11466, DE-AC52-07NA27344.
[1] I. Joseph, et al. “On coupling fluid plasma and kinetic neutral physics models.” Nuc. Mat. & Energy. 12:813 (2017)
[2] G. J. Wilkie, P. K. Romano, R. M. Churchill. “Demonstration of OpenMC as a framework for atomic transport and plasma interaction.” Plas. Phys. & Cont. Fus. 67:055046 (2025)
[3] B. Zhu, et al. “Latent space mapping: Revolutionizing predictive models for divertor plasma detachment control.” Phys. Plasmas 32:062508 (2025)