Journal article
Rethinking network depth: Physics-aware deformable convolutions for strategic learning of nonlinear flow dynamics
APL machine learning, Vol.4(4), 046104
10/19/2026
DOI: 10.1063/5.0350030
Abstract
Physics-aware deep learning (PADL) enables rapid prediction of complex physical systems, yet current convolutional neural network (CNN) architectures struggle with highly nonlinear flows. While scaling model size addresses complexity in broader AI, this approach yields diminishing returns for physics modeling. Drawing inspiration from hybrid Lagrangian–Eulerian numerical methods, we introduce deformable physics-aware recurrent convolutions (D-PARC) to overcome the rigidity of CNNs. Across Burgers’ equation, Navier–Stokes, and reactive flows, D-PARC achieves superior fidelity compared to substantially larger architectures. Analysis reveals that kernels display anti-clustering behavior, evolving into a learned “active filtration” strategy distinct from traditional h- or p-adaptivity. Effective receptive field analysis confirms that D-PARC autonomously concentrates resources in high-strain regions while coarsening focus elsewhere, mirroring adaptive refinement in computational mechanics. This demonstrates that physically intuitive architectural design can outperform parameter scaling, establishing that strategic learning in lean networks offers a more effective path forward for PADL than indiscriminate network expansion.
Details
- Title: Subtitle
- Rethinking network depth: Physics-aware deformable convolutions for strategic learning of nonlinear flow dynamics
- Creators
- Jack T. Beerman - University of VirginiaShobhan Roy - University of IowaH. S. Udaykumar - University of IowaStephen S. Baek - University of Virginia
- Resource Type
- Journal article
- Publication Details
- APL machine learning, Vol.4(4), 046104
- DOI
- 10.1063/5.0350030
- ISSN
- 2770-9019
- eISSN
- 2770-9019
- Publisher
- AIP Publishing LLC
- Grant note
- National Nuclear Security Administration: DE-NA0004239 National Science Foundation: 2203580
This work was supported by the U.S. National Science Foundation under Grant No. 2203580 and partially supported by the U.S. Department of Energy under the National Nuclear Security Administration (NNSA) Stewardship Science Academic Alliances (SSAA) Program (Grant No. DE-NA0004239).
- Language
- English
- Electronic publication date
- 10/19/2026
- Academic Unit
- Engineering Administration; IIHR--Hydroscience and Engineering; Injury Prevention Research Center; Chemical and Biochemical Engineering; Mechanical Engineering
- Record Identifier
- 9985247712902771
Metrics
1 Record Views