Solving the Biharmonic Equation with Coupled Physics-informed Neural Networks Based on Self-adaptive Loss Balancing

Read full text Summarise with VIRAT
Authors: Jing Jiang
Cited by: 0
References: 6

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Standard Physics-informed neural networks (PINNs) encounter notable challenges in solving this equation numerically, such as inadequate approximation accuracy, considerable computational expense, and difficulties in loss function optimization. To mitigate these limitations, we propose introducing an auxiliary variable to decompose the fourth-order biharmonic equation into a coupled system of two second-order partial differential equations. By separately approximating the original solution and the intermediate variable, this approach effectively alleviates the numerical errors associated with computing high-order derivatives while also reducing computational overhead. Furthermore, we incorporate an adaptive loss weighting scheme to dynamically balance the contributions of the PDE various PDE loss terms during training, thereby addressing the issue of loss imbalance inherent in multi-constrained learning. The effectiveness of the proposed method is validated through several numerical experiments.

Fields of Study & Subjects

Model Reduction and Neural Networks Neural Networks and Applications Advanced Numerical Methods in Computational Mathematics

Publication details

Journal / source
Journal of Higher Education Research
Volume
7
Issue
1
Pages
10–10
Publication date
7 February 2026
Language
EN