Work

WKAN: a variational weak-form Kolmogorov–Arnold network (KAN) framework for steady Stokes–Darcy problems

Year: 2026
Type: article
Abstract: Conventional physics-informed neural networks for solving partial differential equations predominantly employ strong-form residual minimization, which typically necessitates the computation of high-order derivatives and assumes solution regularity. This study introduces WKAN, a variational weak-form framework using Kolmogorov–Arnold networks (KANs), specifically designed for steady Stokes–Darcy coupled problems. By formulating the appropriate weak-form energy functional and using KANs to approximate the unknown velocity, pressure, and hydraulic head variables, a mesh-free numerical framework is developed for this class of coupled systems. The variational weak formulation substantially mitigates the reliance on high-order derivatives and stringent regularity requirements. The convergence of the loss function and the WKAN approximation to the exact weak solution is established under the stated assumptions. Comprehensive numerical experiments in two and three spatial dimensions are presented to assess the performance of the proposed method. The results confirm the effectiveness and robustness of WKAN, highlighting its promise for high-accuracy multiphysics simulations.
Language: English
Open Access status: hybrid
Estimated APC paid: $1,890
OpenAlex ID: W7236861985
FWCI:
Cites:
Cited by:
Related to: 10