Assessment of 0-D L-H Power Threshold Scaling and Regression Stability in DIII-D with Applied 3D Magnetic Fields

Abstract

A database of 192 L-H transitions in DIII-D is used to assess the effects of applied three-dimensional (3D) magnetic fields on the H-mode power threshold and the stability of zero-dimensional (0-D) empirical regressions. The dataset includes nominally axisymmetric discharges and discharges with resonant or non-resonant magnetic perturbations. Filtering criteria reduce uncertainties associated with absorbed power, neutral-beam modulation, and fast-ion losses, while applied-field components are quantified using equilibrium reconstruction and spectral analysis. Measured threshold powers show substantial scatter and systematic deviations from the 2008 ITPA multi-machine scaling, including for discharges without applied perturbations. TRANSP modeling indicates that empirical estimates can substantially underpredict fast-ion losses, particularly at low plasma current and with non-axisymmetric fields. Adding global 3D-field metrics does not robustly isolate the effects of applied perturbations. A fully unconstrained regression retains a 70% residual root-mean-square error and produces nonphysical parameter dependencies, including a plasma surface-area exponent of 2.79. Extrapolations to ITER-relevant conditions consequently have broad confidence intervals. These results show that hidden-variable dependencies can strongly affect empirical threshold parameterizations even in a restricted single-machine dataset. Machine-specific conditions, local edge physics, and power-accounting uncertainties limit the predictive capability of purely 0-D scalings. Improved predictions for ITER and future devices will require better fast-ion-loss treatment and physics-based, edge-localized quantities.