A Stable Spectral Bernstein Physics-Informed Neural Network for Solving Fractional Allee Population Models with Caputo Derivatives
International Journal of Pure, Applied and Computational Mathematics (IJPACM) · 30.06.2026 · 1 (1) · 31-83
DOI: 10.68219/f26md935 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Pure, Applied and Computational Mathematics (IJPACM), online erschienen am 30.06.2026. Verfasst von Pappu Verma und 2 weiteren Autor:innen.
Abstract (Originalsprache)
Fractional-order population models have attracted considerable attention because they effectively capture memory and hereditary effects that are absent in classical integer-order models. Among these, the fractional Allee effect model provides a realistic framework for describing population persistence and extinction by incorporating a critical population threshold. In this work, we develop a Stable Spectral Bernstein Physics-Informed Neural Network (SBNN-PINN) to solve the Caputo fractional-order Allee effect model. The proposed framework employs a trainable Bernstein polynomial expansion with a logarithmically stable basis and an analytical Bernstein differentiation matrix to evaluate the Caputo fractional derivative efficiently through matrix--vector multiplication, thereby eliminating repeated numerical discretization during training. Chebyshev--Gauss--Lobatto collocation points are used to improve numerical stability, while a hybrid Adam--L-BFGS optimization strategy accelerates convergence. Stability of the equilibrium points is investigated using the Jacobian matrix and Matignon's stability theorem, confirming that the extinction and carrying-capacity equilibria are locally stable, whereas the Allee threshold equilibrium is unstable. The proposed method is validated using the Method of Manufactured Solutions over fractional orders $0.1 \leq \alpha \leq 0.9$. Numerical experiments demonstrate excellent agreement with analytical solutions, achieving residuals of 3.04 \times 10-29, maximum absolute errors of $3.55\times 10^{-15}$, and relative $L_2$ errors of $9.51\times 10^{-17}$ for $\alpha=0.9$. The results demonstrate that SBNN-PINN is an accurate, stable, and computationally efficient framework for solving nonlinear fractional population models.
Zitieren
APA
Verma, P., Kumar, S., & Varshney, P. (2026). A Stable Spectral Bernstein Physics-Informed Neural Network for Solving Fractional Allee Population Models with Caputo Derivatives. International Journal of Pure, Applied and Computational Mathematics (IJPACM), 1(1), 31-83. https://doi.org/10.68219/f26md935
Vancouver
Verma P, Kumar S, Varshney P. A Stable Spectral Bernstein Physics-Informed Neural Network for Solving Fractional Allee Population Models with Caputo Derivatives. International Journal of Pure, Applied and Computational Mathematics (IJPACM). 2026;1(1):31-83. doi:10.68219/f26md935
BibTeX
@article{Verma2026,
author = {Verma, Pappu and Kumar, Sachin and Varshney, Paras},
title = {A Stable Spectral Bernstein Physics-Informed Neural Network for Solving Fractional Allee Population Models with Caputo Derivatives},
journal = {International Journal of Pure, Applied and Computational Mathematics (IJPACM)},
year = {2026},
volume = {1},
number = {1},
pages = {31--83},
doi = {10.68219/f26md935}
}
EUVIMED