Fractional partial differential equations provide an effective approach for modeling complex physical phenomena characterized by memory effects and nonlocal dynamics. However, obtaining numerical solutions to these equations remains extremely challenging due to the inherent nonlocality of fractional derivatives.
To overcome these numerical challenges, the proposed framework integrates discrete fractional calculus directly with LSTM networks to solve fractional partial differential equations efficiently. In this architecture, the researchers utilized the L1 scheme for discretizing the Caputo time-fractional derivative. Furthermore, the team applied the shifted Grünwald-Letnikov formula for discretizing the Riemann–Liouville space-fractional derivative within the network.
Crucially, both numerical discretization methods are reformulated as matrix multiplication operations. This mathematical reformulation enables the efficient parallel computation of physical residuals within deep learning algorithms.
The effectiveness and accuracy of the Fractional Physics-Informed Memory Networks were validated through systematic numerical experiments. Specifically, the researchers evaluated the architecture on time-fractional Burgers equations and space-fractional convection-diffusion equations, demonstrating reliable performance across both time-dependent and space-dependent fractional problems.
Ultimately, the numerical results demonstrate that Fractional Physics-Informed Memory Networks provide an efficient and reliable deep learning framework for solving fractional partial differential equations.
Subject of Research: Deep learning framework for solving fractional partial differential equations
Article Title: Fractional Physics-Informed Memory Networks for Solving Fractional Partial Differential Equations
Article References: Fang, Y., Dong, H., & Sun, J. (2026). Fractional Physics-Informed Memory Networks for Solving Fractional Partial Differential Equations. Journal of Nonlinear Mathematical Physics. https://doi.org/10.1007/s44198-026-00480-z
Image Credits: AI Generated
DOI: 10.1007/s44198-026-00480-z
Keywords: fractional partial differential equations, fractional calculus, physics-informed neural networks, fractional physics-informed memory networks, long short-term memory, LSTM networks, Caputo fractional derivative, Riemann-Liouville fractional derivative, L1 scheme, Grünwald-Letnikov formula, matrix multiplication, parallel computation, physical residuals, Burgers equation, convection-diffusion equations, numerical methods, memory effects, nonlocal characteristics
Cite Scienmag News
Katie Riggs. (November 19, 2026). Physics-Informed Memory Networks Efficiently Solve Complex Fractional Differential Equations. Scienmag. https://scienmag.com/physics-informed-memory-networks-efficiently-solve-complex-fractional-differential-equations/
Katie Riggs. "Physics-Informed Memory Networks Efficiently Solve Complex Fractional Differential Equations." Scienmag, 19 November 2026, https://scienmag.com/physics-informed-memory-networks-efficiently-solve-complex-fractional-differential-equations/. Accessed 19 November 2026.
Katie Riggs. "Physics-Informed Memory Networks Efficiently Solve Complex Fractional Differential Equations." Scienmag. November 19, 2026. https://scienmag.com/physics-informed-memory-networks-efficiently-solve-complex-fractional-differential-equations/








