Accelerated Workflow for Advanced Kinetic Equilibria

Abstract

Kinetic equilibria are a fundamental aspect of tokamak plasma analysis, but are often highly specialized and labor intensive to produce. This has become a bottleneck to both deeper physics understandings and more sophisticated exper-ment controls. This project aims to remove these barriers by developing a rapid, fully-automated workflow to produce better-than-human, high-precision whole-discharge kinetic equilibria. The required elements in this workflow now exist separately, but what is missing is the coupling of different aspects and overall performance optimization. We have designed this workflow for the DIII-D national fusion facility with the goal of producing results quickly enough to be used for experiment planning in the 15–20 minute time window between subsequent discharges. This is made possible by on-demand availability of supercomputing resources and direct access between these systems and experimental data servers. The results will also be stored in a readily available database for more detailed follow-up analysis. Beyond the between shot application, we plan to apply the workflow offline to the full historical database of the DIII-D experiment. This will both provide scientists a wider set of data than presently available for sophisticated scoping studies and also establish a strong base for training AI/ML surrogate models capable of producing equilibria or their derivatives with similar fidelity at vastly greater speed. The workflow developed here is intended to serve as a prototype that can be replicated on other plasma experiments and provide timely and essential information for the international burning plasma experiment, ITER, as well as next stage fusion power plants. Equilibrium analysis is only the tip of the iceberg in terms of the potential which can be offered by automated superfacility workflows. The foundation of this workflow has been established and tested using of a full scale mock example on the Perlmutter system at NERSC. The results suggest that it may be possible to achieve our goal within a target 10 minute window since there is potential for performance improvement.