The ultimate goal of fusion research is to provide the scientific and technological basis for the development of fusion power plants. In the past, both aspects have often been addressed
separately and sequentially, but over time, it has become increasingly clear that they are intimately coupled. This led to the goal of fully integrated designs, where plasma physics, materials science, and fusion technology are dealt with in the context of an integrated design.
A good example is the development of MFE designs based on the stellarator principle. While state-of-the-art devices like Wendelstein 7-X have been optimized primarily in terms of plasma performance, the next generation of such machines should ideally also include various reactor-relevant aspects (like magnet buildability or the magnet-magnet and magnet-plasma spacing) right from the start. This, in turn, requires fast, but accurate ways to quantitatively model many features of plasma physics, materials science, and fusion technology.
In this context, there exists a fundamental tension between efficiency and reliability of a given (sub-)model. While the integration into optimization loops calls for fast low-fidelity models, the need to provide robust predictions (even in experimentally unexplored regimes) hangs on the availability and use of high-fidelity models. Multi-fidelity methods provide a unique way of combining models of different fidelity, so that an optimal balance between accuracy and cost is achieved. In many cases, speed-ups of several orders of magnitude are to be expected, which amounts to being a true gamer changer, allowing for studies which would be impossible otherwise. This workshop will explore the utilization of multi-fidelity methods to enable fusion device design and engineering, and discuss the mathematical challenges in characterizing the uncertainty induced by incorporating lower-fidelity models into complex plasma physics systems.
Federico Felici
(Google DeepMind)
Lise-Marie Imbert-Gerard
(University of Arizona)
Orso Meneghini
(General Atomics)
Elizabeth Paul
(Columbia University)
Copyright. All Rights Reserved.