A Learnable Multigrid Framework via Graph Convolutions

DOI10.48550/arXiv.2608.29082MaRDI QIDQ7285104arXivFDO

Authors Xiuping Wang, Li Luo

Publication date 29 August 2026

Published in arXiv

Tldr This paper develops a novel framework that integrates learnable graph convolutions into the geometric multigrid method for solving PDEs; The proposed framework embeds graph convolutions within the multigrid cycle, transforming it into a specialized neural network architecture that adapts to problem characteristics and achieves improved convergence rates compared to traditional methods.

Full work available at URL https://doi.org/10.48550/arXiv.2608.29082

Mathematics Subject Classification ID

Preconditioners for iterative methods (65F08) Iterative numerical methods for linear systems (65F10) Finite element, Rayleigh-Ritz and Galerkin methods for boundary value problems involving PDEs (65N30) Multigrid methods; domain decomposition for boundary value problems involving PDEs (65N55) Artificial neural networks and deep learning (68T07)

arXiv classification

math.NA

This page was built for publication: A Learnable Multigrid Framework via Graph Convolutions

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7285104)