SMARTS: Surrogate Models for Accurate and Rapid Transport Solutions

The 2022 Bold Decadal Vision for Commercial Fusion Energy challenges us to demonstrate a path to economically attractive fusion power. Doing so will require cost-effective modeling and simulation tools that can provide accurate, timely, and actionable predictions. In particular, the full design and assessment of a fusion pilot plant (FPP) will require a whole facility modeling (WFM) capability to address the myriad engineering challenges beyond 'simply' confining and sustaining a burning plasma. Yet it remains the case that accurate descriptions of the burning plasma are central to the FPP's design, as that plasma produces the heat and neutron fluxes which are to be transformed into electricity. To date, the tokamak is the most well-developed FPP confinement concept, in both experimentally demonstrated performance and our ability to accurately predict the plasma's dynamics. But significant uncertainties remain for predictions of many key quantities, especially when extrapolating from current-day facilities to future burning plasmas. Some of the most uncertain high-leverage quantities are the densities of various charged particle species across the plasma. Predicting these densities at different locations across the plasma is vital to optimizing FPP performance, and a validated predictive capability for multi-species particle transport and confinement is essential for a viable WFM capability.

The ultimate objective of the SMARTS (Surrogate Models for Accurate and Rapid Transport Solutions) project is to provide the modeling and simulation capabilities needed to resolve this gap by substantially advancing our ability to accurately predict multi-species particle and thermal confinement in tokamak burning plasmas. To do so, we will combine demonstrated advances in GPU acceleration of gyrokinetic codes, applications of Bayesian optimization and surrogate models to transport in fusion plasmas, and tokamak integrated modeling capabilities. Through this approach, we will deliver:

• SMARTSsolver, a next-generation open-source HPC-compatible transport solver which will support a hierarchy of transport model fidelities, and use surrogate models to provide the accurate and rapid transport solutions needed for practical WFM design and optimization.
• Verification and validation of transport predictions against currently used production transport modeling capabilities and experimental data from U.S. tokamaks.
• Extensive predictions of multi-species density and temperature profiles in burning plasmas, examining how reactor-relevant actuators such as radiofrequency heating and deep pellet fueling will impact confinement of different particle species in a variety of pulsed and steady-state FPP-relevant operating scenarios, and thereby overall plasma confinement.
• An open, publicly accessible, curated database of gyrokinetic simulations available for the entire community to use for future studies, including model development, benchmarking, and verification.

To achieve this objective and provide these deliverables, the SMARTS research plan is organized into a set of four connected thrusts:

Thrust 1: Building SMARTS computational infrastructure

Thrust 2: Advancing SMARTS physics capabilities

Thrust 3: Utilizing SMARTS to further our predictive capabilities for burning plasmas

Thrust 4: Connecting SMARTS to the broader community

By carrying out this work, the SMARTS team will deliver significant new modeling and simulation capabilities to public and private fusion efforts, advance our understanding of burning plasma physics, help develop the next generation of fusion researchers, and support building a more diverse, equitable, and inclusive research community.