Petrov-Galerkin Variational Physics-Informed Neural Network Framework for Two-Dimensional Singularly Perturbed Problems

2026-06-15 · Source: Machine Learning · Field: Science & Research — Mathematics & Computational Sciences, Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

A new Petrov-Galerkin based Variational Physics-Informed Neural Network (VPINN) framework is proposed for efficiently solving two-dimensional singularly perturbed problems (SPPs). This method addresses SPPs involving one and two small perturbation parameters. The VPINN utilizes neural networks to construct the trial solution space, while tensor-product hat functions serve as test functions to enforce the variational form. A Petrov-Galerkin formulation is specifically implemented to accurately resolve sharp boundary layers, a common challenge in SPPs. Dirichlet boundary conditions are directly applied, and source terms are calculated via automatic differentiation. Computational experiments on standard two-dimensional problems demonstrate high accuracy in both maximum and L_2 norms, confirming the approach's efficiency and robustness in capturing multiscale features of 2D SPPs.

Key takeaway

For numerical analysts and research scientists tackling two-dimensional singularly perturbed problems, this Petrov-Galerkin VPINN offers a robust solution. You should consider integrating this approach when your simulations require high accuracy in capturing multiscale features and sharp boundary layers. Its demonstrated efficiency in both maximum and L_2 norms suggests it can improve the reliability of your computational models, especially for problems with small perturbation parameters.

Key insights

Petrov-Galerkin VPINN accurately solves 2D singularly perturbed problems by combining neural networks with a variational formulation.

Principles

• Variational forms can be enforced using test functions.
• Petrov-Galerkin formulation enhances boundary layer resolution.
• Automatic differentiation simplifies source term computation.

Method

The method constructs a trial solution space using neural networks and employs tensor-product hat functions as test functions. It applies a Petrov-Galerkin formulation for sharp boundary layers and uses automatic differentiation for source terms.

In practice

• Apply VPINN for multiscale SPP simulations.
• Use tensor-product hat functions for test spaces.
• Leverage automatic differentiation for complex source terms.

Topics

• Physics-Informed Neural Networks
• Petrov-Galerkin Method
• Singularly Perturbed Problems
• Variational Methods
• Numerical Analysis
• Automatic Differentiation

Best for: AI Scientist, Research Scientist

Related on AIssential

• EPINN: Enhanced Physics-Informed Neural Network for Solving Continuous Integral Equations — EPINN offers a robust, scalable solution for high-dimensional integral equations by integrating physical principles into neural networks.
• A Randomized PDE Energy driven Iterative Framework for Efficient and Stable PDE Solutions — This work introduces a novel "PDE energy driven framework" for efficiently and stably solving partial differential equations, addressing the limitations of traditional "matrix based discretizations" a…
• A Machine Learning Approach to the Nirenberg Problem — A physics-informed neural network can effectively solve the Nirenberg problem and distinguish between realisable and non-realisable curvatures.
• Operator learning for the 2D incompressible Navier-Stokes equations: a conformal prediction approach in the data-scarce regime — The perturbation-based conformal prediction framework quantifies uncertainty in neural operators efficiently, especially in data-scarce regimes, by comparing noisy and original label predictions.
• Overcoming the Limits of Finite Difference Method; Physics-Informed Neural Network for Noisy High-Dimensional Heat Diffusion — PINNs overcome classical numerical method limitations in noisy, high-dimensional heat diffusion, maintaining accuracy and efficiency.
• A Dirac-Frenkel-Onsager principle: Instantaneous residual minimization with gauge momentum for nonlinear parametrizations of PDE solutions — The Dirac-Frenkel (DF) principle, used for evolving nonlinear parametrizations of PDE solutions, often faces challenges like non-unique parameter dynamics and ill-conditioning, particularly during "ta…

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.