Physics-Informed Neural Networks Enhance ITG Mode Reconstruction in Tokamaks
A recent preprint on arXiv (2608.01850) presents a physics-informed neural network (PINN) approach aimed at uncovering complex eigenfrequencies and reconstructing two-dimensional complex-valued mode fields associated with the ion-temperature-gradient (ITG) branch within the steep-gradient pedestal of high-confinement-mode tokamaks. This pedestal area plays a vital role in plasma confinement and edge transport, necessitating the simultaneous identification of eigenfrequencies and mode structures for effective analysis of ITG drift waves. The framework tackles challenges such as localized high-frequency oscillations, significant real-imaginary coupling, and nonlinear interactions between mode fields and eigenfrequencies. Utilizing Fourier feature encoding, complex-valued feature propagation, and a three-stage training regimen, it facilitates joint solutions even with limited observations and physical constraints, potentially enhancing predictive modeling in fusion devices.
Key facts
- arXiv:2608.01850
- Physics-informed neural networks (PINNs) combine sparse observations with physical equations
- Focus on ion-temperature-gradient (ITG) drift waves in tokamak pedestal
- Joint identification of complex eigenfrequencies and reconstruction of 2D complex-valued mode fields
- Challenges: localized high-frequency oscillations, strong real-imaginary coupling, nonlinear coupling
- Proposed framework: Fourier feature encoding, complex-valued feature propagation, three-stage training
- Sparse observations and physical constraints are used
- Relevant to plasma confinement and edge transport in high-confinement-mode tokamaks
Entities
Institutions
- arXiv