Data-driven tokamak density limit boundary identification

The density limit (DL) is a critical stability limit for future magnetic fusion devices; for example, ITER and DEMO plan to operate close to or above the Greenwald limit. DLs could jeopardize machine health on these devices by triggering either H-to-L back-transitions or disruptions. In this study, we assemble a database of DL events from DIII-D to evaluate various approaches for predicting the onset of H-mode and L-mode DLs, such as machine learning-based methods and theoretical scalings. We find that an edge collisionality-like scaling derived from the database is a more effective predictor of both types of DL than either the line-averaged or edge/pedestal Greenwald fraction. Our findings are also consistent with a power scaling of the density limit. These results point towards a potentially more reliable control solution for density limit avoidance and suggest that the edge/pedestal Greenwald limit may be too conservative for burning plasmas with low edge collisionality. We also present initial results from a preliminary multi-machine analysis including DL events at AUG, C-Mod, EAST, and TCV.