Self-Regulated Edge Dynamics and Surrogate Modeling: New Developments from the ABOUND SciDAC Project

DOE‑funded ABOUND SciDAC develops physics‑informed, high‑performance tools for predictive control of edge turbulence, transport and plasma–material interactions in reactor‑relevant regimes. Recent nonlinear BOUT++ simulations reveal a two‑stage mechanism for passive ELM suppression: self‑generated zonal flows shear turbulent eddies, while zonal magnetic fields restore J×B≈∇p balance to arrest avalanches, producing a sustained pedestal. We also find that separatrix–to–pedestal density ratio and turbulence spreading dictate transitions from bursting ELMs to continuous transport. To probe ELM–detachment interplay, a hybrid workflow couples 3D BOUT++ turbulence to 2D transport‑informed divertor plasmas; initial results confirm time‑scale separation and show that large ELMs can reattach detached divertors. GPU‑accelerated 3D solvers and multi‑rate SUNDIALS integration enable scalable simulations, and the machine‑learning surrogate ELMO, trained on linear stability outputs, predicts growth rates and mode structures in real time, offering a virtual assistant for experimental design. These physics insights and computational innovations enhance the predictive capability of BOUT++ and influence future tokamak strategies