An AI-driven reduced order model for edge tokamak turbulence

We present an AI-driven reduced order model (ROM) for drift-wave turbulence dynamics based on proper orthogonal decomposition (POD) combined with Galerkin projection. Using the Hasegawa–Wakatani model as a proof of concept, our framework integrates neural ordinary differential equations (NODEs) with POD. In this formulation, a neural network learns the time derivative of the reduced order state, effectively modeling the underlying vector field that governs the system's dynamics. This allows direct access to the phase portrait of the system and provides insights into trends and patterns by capturing its continuous temporal evolution. Results show that the proposed method reconstructs key features of the system and holds promise for significant computational speed-up.