Global Convergence of DGM and PINN Algorithms for Solving Nonlinear PDEs

DOI10.48550/arXiv.2607.24726MaRDI QIDQ7284404arXivFDO

Authors Konstantinos Spiliopoulos, Justin Sirignano, Samuel Cohen

Publication date 27 July 2026

Published in arXiv

Tldr We analyze the convergence properties of deep Galerkin methods and physics informed neural networks for solving nonlinear partial differential equations by employing tools from optimization theory and dynamical systems. We show that under certain assumptions, these methods are guaranteed to globally converge to the exact solution when using gradient descent for minimization.

Full work available at URL https://doi.org/10.48550/arXiv.2607.24726

arXiv classification

cs.LGmath.NA

Cites work

• DGM: a deep learning algorithm for solving partial differential equations
• Elliptic partial differential equations of second order
• Introduction to Partial Differential Equations
• Neural Q-learning for solving PDEs
• Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations

This page was built for publication: Global Convergence of DGM and PINN Algorithms for Solving Nonlinear PDEs

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7284404)