Taylor-Accelerated Neural Network Interpolation Operators on Irregular Grids: Higher-Order Approximation and Derivative-Informed Field Reconstruction
math.NA, cs.NA, math.FA
Submitted: 2026-02-05
Updated: 2026-11-01
Comments: 35 pages, 13 figures
License: http://creativecommons.org/licenses/by-nc-nd/4.0/
The gist: In this paper, a new class of Taylor-accelerated neural-network interpolation operators is introduced for quasi-uniform…
Terminology
Abstract
In this paper, a new class of Taylor-accelerated neural-network interpolation operators is introduced for quasi-uniform, irregularly distributed grids to reconstruct derivative-informed physical fields in both univariate and multivariate settings. The framework incorporates Taylor expansions within the neural network operators framework, allowing operators to efficiently utilize the smoothness of the target function. We determine the analytical properties of the proposed operators, including well-posedness, uniform boundedness, and exact interpolation at the sampling nodes. Additionally, we establish polynomial reproduction to a specified degree and provide Jackson-type error estimates based on higher-order moduli of smoothness, thereby improving convergence rates on quasi-uniform, irregular grids. The applicability of the method is demonstrated by reconstructing physical fields such as temperature, electromagnetic fields, and fluid properties, including pressure and velocity. The numerical results show that the proposed operators provide reliable, accurate estimates in the presence of noise and irregular sampling. The enhanced accuracy of the proposed method is validated by comparing it with current interpolation systems on irregular grids, including Lagrange-type methods.
Related papers
- Do physics-informed neural networks (PINNs) need to be deep? Shallow PINNs using the Levenberg-Marquardt algorithm
- A Neural-preconditioned Poisson Solver for Mixed Dirichlet and Neumann Boundary Conditions
- Advanced Linear Algebra with Applications - Part I (Numerical linear algebra for PDEs, machine learning, and data assimilation)
- Operator learning meets inverse problems: A probabilistic perspective
- Second-order consistency for learning chaotic dynamics via randomized Jacobian matching
- Windowed thinning and query complexity for the bouncy particle and Zigzag samplers