Results of an XGC simulation of the ASDEX-Upgrade tokamak rendered together with the device geometry. The approximate outlines of the core and edge plasma regions are indicated by the white lines on the left plasma cross-section.
AI-facilitated Global Profile Prediction in Tokamak Plasma
Project Summary
The INCITE project leverages exascale computing, machine-learned surrogates, and advanced core–edge plasma simulations to accurately predict temperature, density, and heat-flux profiles in fusion devices, providing validated workflows to guide future tokamak design.
Project Description
This INCITE project aims to enable first-principles prediction of temperature and density profiles across fusion devices, from the plasma core to plasma-facing components. It addresses the limitations of current scaling laws and reduced fluid models by combining exascale computing on Frontier and Aurora with advanced core and edge plasma algorithms. The approach couples a machine-learned surrogate—trained on PORTALS/CGYRO core simulations—with iterative edge calculations using the full-f XGC code. The surrogate provides fast inner boundary conditions, while XGC resolves scrape-off-layer physics, impurity radiation, and exhaust-heat handling.
Using techniques such as time-telescoping, adaptive turbulence seeding, and ML-accelerated convergence, the project seeks to find steady-state profiles efficiently. Deliverables include a verified workflow for Aurora, a library of core–edge simulations for future tokamak and spherical-tokamak concepts, and an open, compressed dataset for AI tools. The results will help guide pilot-plant design by providing quantitative predictions of heat-flux limits, impurity tolerances, actuator requirements, and achievable fusion gain, while establishing a scalable, mixed-fidelity framework applicable to future fusion devices.
