Data-Derived Operational Boundaries and Scaling of RMP ELM Suppression

Suppression of edge-localized modes (ELMs) by application of resonant magnetic perturbation (RMP) fields has been demonstrated in many tokamaks, however, the access criteria are not fully understood. Linear discriminant analysis (LDA), a classifier that projects data onto a linear axis that maximizes the distance between class means, is performed on a dataset of discharges from the ASDEX Upgrade tokamak including ELMy and ELM-suppressed phases. Treating this analysis as a classification problem, an LDA model using equilibrium, control, and plasma parameters is trained with a predictive accuracy of >90%. The decision boundary for determining a discharge’s classification as ELMy or suppressed is derived and compared to known experimental threshold conditions. Scaling laws for confinement time are extracted from a multi-device database consisting of RMP ELM-suppressed H-mode discharges from ASDEX Upgrade, DIII-D, and KSTAR. These are compared to previously derived H-mode and L-mode laws. Including rotation data in addition to previously used quantities improved the overall goodness-of-fit, especially so for single-device data. This work provides a further understanding of the parameter space required for ELMs to be suppressed by RMPs and the confinement quality expected therein.