Description
Accurate tokamak performance prediction requires time-dependent, self-consistent modeling of free-boundary equilibrium, transport, heating/current drive, passive-structure response, and stability. We introduce Next Step Fusion simulator (NSFsim), an advanced, evolving integrated modeling solution for tokamak design, scenario development, machine optimization, and operation simulations. NSFsim solves the 2D Grad–Shafranov problem with a free boundary in external magnetic fields, coupled with a 1D kinetic/transport component on a configurable radial grid, including a non-uniform mesh for edge physics to ensure efficiency.
The electromagnetic model represents conducting structures as discretized elements, enabling calculation of induced currents and synthetic magnetic diagnostics. This supports multiple use cases within a unified framework: (i) direct discharge calculations driven by prescribed coil currents/voltages and kinetic constraints; (ii) scenario calculations with prescribed kinetic evolution; (iii) plasma-free calculations for electromagnetic model validation and diagnostic channel selection; (iv) inverse initialization to generate consistent “start points” for scenario building and machine-learning datasets; (v) disruption and vertical displacement event modeling with halo/eddy currents and force distributions; and (vi) equilibrium reconstruction using fixed- or floating-filament formulations with diagnostic validity masks.
Ongoing NSFsim development extends the computational core and interfaces to physics codes to improve physical fidelity and match experimental data. Recent advances include better scenario calculation, configurable boundary models, improved plasma and impurity treatment (including deuterium–tritium), and finer control of numerical accuracy. These updates produce realistic scenario trajectories with minimal correction, except for stabilization in diverted setups where noise is unavoidable.
NSFsim is a high-fidelity environment for reinforcement-learning plasma control with magnetic and kinetic actuators. Its precise numerical tuning and physically consistent coupling between equilibrium and transport support robust training and validation of the controller. Testing against experimental discharges confirms high control accuracy, reproducibility, and reliable generalization to unseen plasma configurations.
NSFsim is used both offline and online via an API, supporting reproducible workflows, standardizing geometry/coil configurations, and verifying results against experimental discharges, such as vacuum and plasma-shot studies. Its 'flight-simulator' operation mode aims to accelerate scenario optimization and controller development under realistic constraints.