Neural network models plasma boundary with geometry and uncertainty
Probabilistic and Geometry Aware Neural Surrogate of Scrape Off Layer Plasma Simulations
Machine Learning
Summary
Simulating plasma behavior near a tokamak’s edge is hard because it can behave in different ways under the same conditions. Existing fast models often average these different behaviors into one guess without showing how confident they are. The authors represent the plasma simulation mesh as images that keep their layout intact so a neural network can learn about the geometry. Their new model predicts multiple possible plasma states and shows when the simulation outcome could be hot or cold, capturing uncertainty more realistically.
What this means in practice
- •For tokamak simulation engineers: Produce faster and more reliable plasma boundary predictions that capture multiple steady states and uncertainty near detachment transitions.
- •For computational fluid dynamics developers: Incorporate geometry-preserving image representations and probabilistic neural surrogates to model sensitive flow regimes with bifurcations in other physics simulations.
Authors
Gabriele Gianuzzo, Stefan Dasbach, Fleur Hendriks, Sven Wiesen, Vlado Menkovski
Abstract
Fast surrogates for tokamak boundary-plasma simulation are typically deterministic regressors mapping a global operating point to a flattened vector of cell values. Near the divertor detachment transition the steady state is not reliably single-valued. A point estimate must average over qualitatively different plasma states, and it arrives with no statement of confidence. Moreover, the flattened vector representation discards the geometric structure of the SOLPS-ITER mesh. This work addresses both problems. We unroll the curvilinear mesh into three fixed-size image tensors whose layout preserves cell adjacency and inverts exactly, letting a convolutional network act on the geometry without loss of information. A conditional flow matching model, well suited to highly sensitive systems, is then trained on this representation. The result is an efficient, scalable surrogate that captures multiple plausible outcomes even at sensitive operating points. Along a gas-puff scan, the predictive distribution splits into a hot and a cold mode across an early regime transition. A further check on synthetic data with an injected bifurcation of known size confirms the model recovers both branches rather than their average.