Learnable Splines for Unstructured Neural Operators

Abstract

Neural operators approximate solution operators of partial differential equations at a fraction of the cost of numerical solvers. Efficient grid-based architectures such as the Fourier neural operator (FNO) assume uniform rectangular grids, whereas engineering simulations use large, irregular, and locally refined meshes. We introduce L-SplineFNO, a memory-efficient neural operator that couples an FNO backbone with a sparse, learnable tensor-product B-spline transfer between unstructured meshes and a latent Cartesian grid. A local normalised scatter avoids the geometry-dependent solve of analytic spline transformation while preserving compact support, partition of unity, and spline smoothness by construction. The neural operator, conditioned on global design and operating parameters, learns the mapping of geometry to the field solution. Across six three-dimensional engineering benchmarks, L-SplineFNO achieves strong predictive accuracy while requiring substantially less memory and compute than baseline operators, scaling to 26.6 million nodes in training and 46 million in inference on a single A100 GPU. It also supports zero-shot super-resolution and efficient closed-form spatial derivatives. Uniquely, the analytic spline decoder yields rigorous prediction bounds over continuous space and uncertain parameters, via Bernstein convex hulls and Taylor models, demonstrated on a 29.2-million-node MAST-U tokamak model.

PINN-LPCDD: Physics-Informed Neural Network with Learnable and Physically Consistent Domain Decomposition for Compressible Multi-Medium Flows
Weiheng Zeng, Hao Wang, Jiaxin Zhu +3
arXiv16 Oct 2026PreprintarXiv licenceFull textNumerical analysis & scientific computing Computational physicsarXiv
Neural Network Fractional Quantum Hall States with Learnable Orbital Pair Matrix Factors: Balancing Physical Priors and Flexibility
Yubing Qian, Weiluo Ren, Ji Chen
arXiv20 Oct 2026PreprintarXiv licenceFull textCondensed matterarXiv
Pretrain finite element method: A pretraining and warm-start framework for PDEs via physics-informed neural operators
Yizheng Wang, Zhongkai Hao, Mohammad Sadegh Eshaghi +4
Journal of the Mechanics and Physics of Solids15 May 2026Peer-reviewedLicence not statedFull text7 citedStatistical & nonlinear physicsOpenAlex
Lecture notes on Physics Informed Neural Networks, Neural Operators, and their applications
Alessandro Bombini
arXiv15 Sep 2026PreprintCC BY 4.0Full textComputational physicsarXiv
Do Better Scores Mean Better Physics? Physics-Grounded Explanations for Sim2Real Neural Operators
Somyajit Chakraborty, Xizhong Chen
arXiv1 Oct 2026PreprintCC BY 4.0Full textFluid dynamicsarXiv

Cite

BibTeX RIS

BibTeX

@misc{gopakumar2026learnable,
  title = {{Learnable Splines for Unstructured Neural Operators}},
  author = {Vignesh Gopakumar and Zhuoyuan Wang and Ander Gray and Valentin Duruisseaux and Josh Williams and William Michael Edward Ellis and Nuno Nobre and Matthew Field and James Panayis and Alexander I. Blair and Andrew Davis and Stanislas Pamela and Robert Akers and Marc Peter Deisenroth and Anima Anandkumar},
  howpublished = {arXiv},
  year = {2026},
  month = oct,
  doi = {10.48550/arxiv.2610.22446},
  eprint = {2610.22446},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.22446},
  note = {CC BY 4.0},
}

RIS

TY  - UNPB
TI  - Learnable Splines for Unstructured Neural Operators
AU  - Vignesh Gopakumar
AU  - Zhuoyuan Wang
AU  - Ander Gray
AU  - Valentin Duruisseaux
AU  - Josh Williams
AU  - William Michael Edward Ellis
AU  - Nuno Nobre
AU  - Matthew Field
AU  - James Panayis
AU  - Alexander I. Blair
AU  - Andrew Davis
AU  - Stanislas Pamela
AU  - Robert Akers
AU  - Marc Peter Deisenroth
AU  - Anima Anandkumar
PY  - 2026
DA  - 2026/10/15
PB  - arXiv
DO  - 10.48550/arxiv.2610.22446
UR  - https://arxiv.org/abs/2610.22446
AB  - Neural operators approximate solution operators of partial differential equations at a fraction of the cost of numerical solvers. Efficient grid-based architectures such as the Fourier neural operator (FNO) assume uniform rectangular grids, whereas engineering simulations use large, irregular, and locally refined meshes. We introduce L-SplineFNO, a memory-efficient neural operator that couples an FNO backbone with a sparse, learnable tensor-product B-spline transfer between unstructured meshes and a latent Cartesian grid. A local normalised scatter avoids the geometry-dependent solve of analytic spline transformation while preserving compact support, partition of unity, and spline smoothness by construction. The neural operator, conditioned on global design and operating parameters, learns the mapping of geometry to the field solution. Across six three-dimensional engineering benchmarks, L-SplineFNO achieves strong predictive accuracy while requiring substantially less memory and compute than baseline operators, scaling to 26.6 million nodes in training and 46 million in inference on a single A100 GPU. It also supports zero-shot super-resolution and efficient closed-form spatial derivatives. Uniquely, the analytic spline decoder yields rigorous prediction bounds over continuous space and uncertain parameters, via Bernstein convex hulls and Taylor models, demonstrated on a 29.2-million-node MAST-U tokamak model.
ER  -

Abstract

Related

Cite