Answers are grounded in this paper and cite their sources.
Fractional partial differential equations (FPDEs) offer an effective approach to modeling complex phenomena with memory effects and nonlocal characteristics. However, due to the inherent nonlocality of fractional derivatives, their numerical solution remains extremely challenging. This paper proposes Fractional Physics-Informed Memory Networks (FPIMNs), integrating discrete fractional calculus with LSTM networks to efficiently solve FPDEs. Specifically, FPIMNs utilize the L 1 scheme for discretizing the Caputo time-fractional derivative, and the shifted Grünwald-Letnikov formula for discretizing the Riemann–Liouville space-fractional derivative. Both discretization methods are reformulated as matrix multiplication operations, enabling efficient parallel computation of physical residuals within the deep learning. The effectiveness and accuracy of FPIMNs were validated by numerical experiments on time-fractional Burgers and space-fractional convection-diffusion equations. FPIMNs provide an efficient and reliable deep learning framework for solving FPDEs.
No takes yet. Share an insight, caveat, or question.
Fang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: