Wavelet-enhanced physics-informed neural networks for a class of delay differential equations
Physica Scripta · 19.06.2026 · 101 (24) · 246003
DOI: 10.1088/1402-4896/ae77b4 ↗Kurz gefasst
Fachartikel in der Zeitschrift Physica Scripta, online erschienen am 16.06.2026. Verfasst von Kuruva Vinay und 2 weiteren Autor:innen. Die Arbeit nennt 46 Literaturangaben.
Abstract (Originalsprache)
This paper addresses the challenge of accurately solving delay differential equations, which often exhibit complex temporal dependencies and localized features that are difficult to capture using conventional numerical and learning-based methods. We propose a novel grid-free deep learning framework for solving a class of delay differential equations. Specifically, we introduce a wavelet-enhanced physics-informed neural network (NN) that extends standard physics-informed NNs by embedding wavelet-based representations into the network architecture. Instead of relying on conventional activation functions, the proposed model employs multi-resolution wavelet features to capture localized irregularities and long-range temporal dependencies induced by delays. By combining wavelet NNs with a hybrid collocation sampling strategy incorporating adaptive residual-based sampling and a two-stage optimization process using AdamW followed by the Limited-memory Broyden–Fletcher–Goldfarb–Shanno algorithm, the proposed method achieves high accuracy and robust convergence. The effectiveness of the approach is demonstrated on a diverse collection of problems, including scalar and coupled systems, higher-order equations, and models with constant and proportional delays. Numerical experiments show that the proposed methodology consistently attains better accuracy and convergence behaviour. Lastly, the method is applied to the delayed logistic equation and the generalized Nicholson’s blowflies delay equation to illustrate its practical applicability.
Zitieren
APA
Vinay, K., Kumar, V., & Sharma, A. (2026). Wavelet-enhanced physics-informed neural networks for a class of delay differential equations. Physica Scripta, 101(24), 246003. https://doi.org/10.1088/1402-4896/ae77b4
Vancouver
Vinay K, Kumar V, Sharma A. Wavelet-enhanced physics-informed neural networks for a class of delay differential equations. Physica Scripta. 2026;101(24):246003. doi:10.1088/1402-4896/ae77b4
BibTeX
@article{Vinay2026,
author = {Vinay, Kuruva and Kumar, Vipin and Sharma, Anshika},
title = {Wavelet-enhanced physics-informed neural networks for a class of delay differential equations},
journal = {Physica Scripta},
year = {2026},
volume = {101},
number = {24},
pages = {246003},
doi = {10.1088/1402-4896/ae77b4}
}
EUVIMED