DC-PINNs: Physics-Informed Neural Networks for Solving Derivative-Constrained PDEs
Keywords: PINNs, Physics-Informed Neural Networks, Multi-objective Learning, Partial Differential Equations, Derivative-Constrained, Machine Learning
TL;DR: DC-PINNs: Self-Adaptive Neural Networks for Constrained PDEs
Abstract: Physics-Informed Neural Networks (PINNs) have emerged as a promising approach for solving partial differential equations (PDEs) using deep learning. However, standard PINNs do not address the problem of constrained PDEs, where the solution must satisfy additional equality or inequality constraints beyond the governing equations. In this paper, we introduce Derivative-Constrained PINNs (DC-PINNs), a novel framework that seamlessly incorporates constraint information into the PINNs training process. DC-PINNs employ a constraint-aware loss function that penalizes constraint violations while simultaneously minimizing the PDE residual. Key components include self-adaptive loss balancing techniques that automatically tune the relative weighting of each term, enhancing training stability, and the use of automatic differentiation to efficiently compute derivatives. This study demonstrates the effectiveness of DC-PINNs on several benchmark problems, from basic to complex, such as quantitative finance and applied physics, including heat diffusion, volatility surface calibration, and incompressible flow dynamics. The results showcase improvements in generating solutions that satisfy the constraints compared to baseline PINNs methods. The DC-PINNs framework opens up new possibilities for solving constrained PDEs in multi-objective optimization problems.
Primary Area: neurosymbolic & hybrid AI systems (physics-informed, logic & formal reasoning, etc.)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 3539
Loading