Physics-Informed Kolmogorov-Arnold Networks for Grad-Shafranov Tokamak Equilibria
Abstract
We employ equation-driven, physics-constrained deep learning to solve the fixed-boundary Grad-Shafranov (GS) equilibrium problem, constructing axisymmetric magnetohydrodynamic equilibria with tokamak-relevant characteristics. Equilibria across linear (Solov'ev) and nonlinear profile functions are constructed using Physics-Informed Kolmogorov-Arnold Networks (KANs) that approximate GS solutions while satisfying appropriate boundary conditions. A highly nonlinear pressure profile recreating high-confinement mode phenomenology, such as pressure pedestals and significant bootstrap current components, is also considered. To enable efficient convergence, guided training schemes are employed, specifically homotopy-based continuation curriculum learning and transfer learning via pretrained networks. While computing nonlinear equilibria employing standard Multi-Layer Perceptrons under unguided physics-informed training remains an elusive or computationally inefficient task, our framework overcomes this limitation. Specifically, we demonstrate that the combination of three key elements, namely KAN architecture, guided training schemes, and the self-scaled Broyden optimization method, enables stable, efficient, and accurate equilibrium computation with simultaneous profile parameter identification in view of equilibrium constraints.
Related
- Finite basis Kolmogorov-Arnold networks: Domain decomposition for data-driven and physics-informed problems
- Physics-Informed Kolmogorov-Arnold networks for viscoelastic fluid equations
- Physics-informed Kolmogorov-Arnold networks (PIKANs) for noise-robust reconstruction of boiling heat flux from infrared measurements
- Shape effects and Shafranov shift reversal in analytical Grad--Shafranov equilibria with non-convex boundaries
- Accelerating noisy variational quantum algorithms with physics-informed denoising networks
BibTeX
@misc{kaltsas2026physicsinformed,
title = {{Physics-Informed Kolmogorov-Arnold Networks for Grad-Shafranov Tokamak Equilibria}},
author = {D. A. Kaltsas and A. Kuiroukidis and J. Liu and L. Magafas and G. N. Throumoulopoulos},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.23846},
eprint = {2609.23846},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.23846},
}
RIS
TY - UNPB
TI - Physics-Informed Kolmogorov-Arnold Networks for Grad-Shafranov Tokamak Equilibria
AU - D. A. Kaltsas
AU - A. Kuiroukidis
AU - J. Liu
AU - L. Magafas
AU - G. N. Throumoulopoulos
PY - 2026
DA - 2026/09/20
PB - arXiv
DO - 10.48550/arxiv.2609.23846
UR - https://arxiv.org/abs/2609.23846
AB - We employ equation-driven, physics-constrained deep learning to solve the fixed-boundary Grad-Shafranov (GS) equilibrium problem, constructing axisymmetric magnetohydrodynamic equilibria with tokamak-relevant characteristics. Equilibria across linear (Solov'ev) and nonlinear profile functions are constructed using Physics-Informed Kolmogorov-Arnold Networks (KANs) that approximate GS solutions while satisfying appropriate boundary conditions. A highly nonlinear pressure profile recreating high-confinement mode phenomenology, such as pressure pedestals and significant bootstrap current components, is also considered. To enable efficient convergence, guided training schemes are employed, specifically homotopy-based continuation curriculum learning and transfer learning via pretrained networks. While computing nonlinear equilibria employing standard Multi-Layer Perceptrons under unguided physics-informed training remains an elusive or computationally inefficient task, our framework overcomes this limitation. Specifically, we demonstrate that the combination of three key elements, namely KAN architecture, guided training schemes, and the self-scaled Broyden optimization method, enables stable, efficient, and accurate equilibrium computation with simultaneous profile parameter identification in view of equilibrium constraints.
ER -