Correlation of the L-mode density limit with edge collisionality

A novel database study of the L-mode Density Limit (LDL) in metal- and carbon-wall devices (Alcator C-Mod, AUG, DIII-D, and TCV) identifies a two-variable, dimensionless stability boundary that predicts the LDL with significantly higher accuracy than the widely-utilized Greenwald limit. Historically, there has been broad interest in understanding the operational boundary imposed by the disruptive LDL because density is a critical lever for fusion performance. In this study, we create a multi-machine database of over 150 LDL events with 3000+ non-LDL discharges for evaluating the True and False Positive Rate. We find that data-driven models involving edge density and temperature measurements achieve significantly higher LDL prediction performance than the Greenwald fraction. Additionally, we utilize a Support Vector Machine to identify an analytic, dimensionless, stability boundary that retains the accuracy of the more sophisticated models, such as a Neural Network and Random Forest. The boundary is dominated by the effective collisionality in the plasma edge. This finding suggests that burning plasmas, with naturally low edge collisionality due to self-heating, may be able to achieve super-Greenwald densities. Additionally, in current and “next step” devices such as ITER, this collisionality boundary can also be deployed for active density limit avoidance.