Multi-Fidelity Predictive Core Performance Modeling for an ARC Power Plant

ARC is a compact, high-field tokamak being designed by Commonwealth Fusion Systems with the goal of producing net electricity [1]. Here, we present a multi-fidelity approach to predict the fusion performance in ARC and find a robust operational point. The operational point is first selected using empirical 0D models (POPCONs), which allow for rapid scoping studies but lack realistic physics. To bridge this gap, we have developed a time-dependent profile prediction workflow using ASTRA [2], using an EPED [3] neural network model for self-consistent pedestal stability and TGLF-SAT2 [4] for quasilinear turbulence modeling. The time-dependent solution can be validated using nonlinear gyrokinetic profile predictions with PORTALS-CGYRO [5]. The converged solution is then used to update the transport parameters used as inputs to the POPCON, producing a transport-physics informed 0D model. We first analyze the ARC V2A design point using this methodology. Using the integrated modeling workflow described above, we show that small increases in heating power, elongation, and plasma current lead to broader operational windows and higher performance. A scan of key assumptions, such as separatrix density, effective ion charge, and impurity content has been performed to identify the physics assumptions with the greatest impact on performance.

[1] J. Hillesheim et al., APS-DPP (2023)

[2] G.V. Pereverzev, P.N. Yushmanov, IPP Report 5/98 (2002).

[3] P.B Snyder et al., Nuclear Fusion 51 103016 (2011)

[4] G.M. Staebler et al., Plasma Physics and Controlled Fusion 63 015013 (2020)

[5] P. Rodriguez-Fernandez et al., Nuclear. Fusion 64 076034 (2024)