Reproduce “Physics-constrained DeepONet for Surrogate CFD models: a curved backward-facing step case” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Anas Jnini, Harshinee Goordoyal, Sujal Dave, Flavio Vella, Katharine H. Fraser, Artem Korobenko
This paper presents a Physics-Constrained DeepONet (PC-DeepONet) that embeds the divergence-free velocity constraint directly into its architecture for surrogate modeling of fluid flow. The PC-DeepONet achieves significantly higher accuracy, with a 2.6 times lower relative L2 error, and improved physical consistency compared to a standard DeepONet, particularly when trained on sparse data for a curved backward-facing step case.
Key insights
- Embedding physical constraints, like the divergence-free condition for velocity, directly into the DeepONet architecture leads to more accurate and physically consistent predictions compared to purely data-driven models.
- The Physics-Constrained DeepONet (PC-DeepONet) demonstrates superior performance, especially when trained on sparse data, by narrowing the solution space to physically plausible outcomes.
- Combining data-driven operator learning with architectural physics constraints offers a robust path towards efficient and reliable surrogate models for complex fluid dynamics.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.