Published in Advances in Continuous and Discrete Models, this specialized computational architecture establishes a mesh-free numerical framework tailored specifically for this class of coupled systems.
The development addresses key operational hurdles associated with conventional physics-informed neural networks applied to partial differential equations. Standard physics-informed neural networks predominantly rely on strong-form residual minimization during the equation solving process.
To overcome these limitations, the investigators developed a variational weak formulation anchored by an appropriate weak-form energy functional. Within this mathematical construct, Kolmogorov–Arnold networks, or KANs, are employed directly to approximate key unknown physical variables. Specifically, the mesh-free framework successfully targets the approximation of unknown velocity, pressure, and hydraulic head variables across this class of coupled Stokes–Darcy systems.
By shifting to this variational weak-form strategy, the WKAN framework substantially mitigates the traditional reliance on demanding high-order derivatives. It also relieves the stringent mathematical regularity requirements that typically restrict strong-form methods. Alongside the practical network architecture, the authors also established theoretical convergence guarantees, proving that both the loss function and the WKAN approximation converge directly to the exact weak solution under the stated analytical assumptions.
To assess the performance and operational viability of the proposed method, the researchers carried out comprehensive numerical experiments spanning both two and three spatial dimensions. The experimental results confirmed the effectiveness and robustness of the WKAN framework across these tests, highlighting its clear promise for enabling high-accuracy multiphysics simulations in complex physical systems.
Subject of Research: Variational weak-form Kolmogorov–Arnold networks for steady Stokes–Darcy coupled problems
Article Title: WKAN: a variational weak-form Kolmogorov–Arnold network (KAN) framework for steady Stokes–Darcy problems
Article References: Song, Y., Li, S., Zeng, J., & Chen, J. (2026). WKAN: a variational weak-form Kolmogorov–Arnold network (KAN) framework for steady Stokes–Darcy problems. Advances in Continuous and Discrete Models. https://doi.org/10.1186/s13662-026-04151-7
Image Credits: AI Generated
DOI: 10.1186/s13662-026-04151-7
Keywords: Kolmogorov–Arnold networks, WKAN, Stokes–Darcy problems, variational weak form, mesh-free methods, physics-informed neural networks, partial differential equations, energy functional, fluid velocity, pressure approximation, hydraulic head, solution regularity, loss function convergence, multiphysics simulations, three-dimensional simulations, coupled systems
Cite Scienmag News
Reid Dalton. (November 19, 2026). Variational Kolmogorov–Arnold Network Framework Effectively Solves Coupled Stokes–Darcy Problems. Scienmag. https://scienmag.com/variational-kolmogorov-arnold-network-framework-effectively-solves-coupled-stokes-darcy-problems/
Reid Dalton. "Variational Kolmogorov–Arnold Network Framework Effectively Solves Coupled Stokes–Darcy Problems." Scienmag, 19 November 2026, https://scienmag.com/variational-kolmogorov-arnold-network-framework-effectively-solves-coupled-stokes-darcy-problems/. Accessed 19 November 2026.
Reid Dalton. "Variational Kolmogorov–Arnold Network Framework Effectively Solves Coupled Stokes–Darcy Problems." Scienmag. November 19, 2026. https://scienmag.com/variational-kolmogorov-arnold-network-framework-effectively-solves-coupled-stokes-darcy-problems/








