Speaker
Description
Achieving reliable power production in future tokamak reactors requires operating
regimes that simultaneously deliver high fusion plasma performance and maintain
divertor and first-wall particle and heat-fluxes within material limits. The plasma core
and the scrape-off layer (SOL) regions cannot be decoupled for predictive scenario
development. Integrated modelling frameworks self-consistently resolve core–edge
interactions but still require some external information for example on anomalous SOL
transport. Within the JINTRAC suite [1] such simulations remain computationally
constraint by the numerical cost of detailed edge/SOL solvers such as EDGE2DEIRENE. This limitation prohibits systematic parameter exploration, uncertainty
quantification, and the broad optimisation studies required for advanced scenario
development.
A promising strategy to overcome this bottleneck is the use of fast surrogate models
trained on SOL simulations [2]. In this work, we integrate SOLPS-NN—a neural-network
surrogate trained on a uniformly sampled multi-parameter SOLPS-ITER simulation
database in JET geometry [3,4]—into JINTRAC for the first time. This approach replaces
the conventional edge fluid solver with a rapid predictor of SOL and divertor plasma
conditions, enabling time-dependent core-edge simulations at drastically reduced
computational cost, effectively reducing the problem to core-solver timescales.
We present the technical implementation of the investigated JINTRAC–SOLPS-NN
coupling approaches, including the choice of exchanged quantities, coupling interface,
and numerical procedures used to connect the surrogate-based SOL model to the core
solver. The resulting performance is assessed through a series of initial tests on a JET
scenario that is consistent with the SOLPS-NN training dataset, providing a controlled
setting to evaluate the behaviour of the coupled system. Simulations obtained with the
different coupling variants are compared against reference JINTRAC/EDGE2D-EIRENE
results to examine the ability of the surrogate-based framework to reproduce expected
physical trends. These studies provide initial insight into the accuracy, robustness, and
limitations of this accelerated integrated-modelling approach.
Based on these findings, we outline the foreseen developments required to enhance the
physics fidelity of the coupling, such as extensions of the communicated quantities and
further validation, and discuss the prospective applications enabled by the achieved
speed-ups.
[1] M. Romanelli et al., “JINTRAC: A System of Codes for Integrated Simulation of Tokamak Scenarios”,
Plasma Fusion Res. (2014).
[2] S. Wiesen et al., “Data-driven models in fusion exhaust”, Nucl. Fusion (2024).
[3] S. Dasbach, “Surrogate models for particle and power exhaust in divertor plasmas”, PhD thesis (2025).
[4] S. Dasbach, S. Wiesen, “Surrogate models for interpolation of tokamak edge plasmas”, Nucl. Mater.
Energy (2023).