Physics-informed neural networks with caputo-fabrizio derivatives for nonlinear fractal-fractional delay equations and chaotic systems Authors: Hafsa Khan, Muhammad Sulaiman, Ghaylen Laouini, Izaz Ur Rahman, Mohammed Abdullah Salman Published: 2026 ABSTRACT Complex dynamical systems, ranging from anomalous diffusion and memristor circuits to convective fluid motion in rotating cavities, exhibit non-local memory, long-range interactions, and proportional delays governed by Fractal-Fractional Differential Equations (FFDEs). This study introduces PINN-CF, a Physics-Informed Neural Network framework that integrates Caputo-Fabrizio (CF) Fractional Derivatives with the Adam optimizer. The CF operator is accurately approximated using the L1 finite-difference scheme. Unlike standard PINN that relay on singular kernel fractional operators, PINN-CF employs a non-singular exponential kernel, which avoids initial singularities and better capture smooth memory decay. PINN-CF accurately solves four benchmark problems of FFDEs, Pantograph-type delay equations, Memristor-based neural circuits, and FFDE systems of convective fluid motion in rotating cavities. The suggested PINN-CF method achieves residual errors below \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:1{0}^{-4}$$\end{document}, and as low as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:1{0}^{-6}$$\end{document} in near classical regimes. Reproducibility and validation ensured by convergence curves, and tabulated results. Offering superior flexibility, scalability, and computational efficiency over traditional numerical solvers, PINN-CF is readily extendable to nonlinear, multi-delay, distributed-order, or stochastic models. Introduction Fractional calculus1 has been extensively explored to model various phenomena such as epidemic childhood diseases2,3, Glucose molecule4, human liver5, COVID-19 transmission6, Within-Host SARS-CoV-27. A generalization of classical differential equations with arbitrary differentiation operator orders is known as a fractional differential equation. The reference provides some insightful findings about the theoretical study of these operators8. It is vital to understand that these operators are frequently single. Thus, getting analytic solutions to problems involving fractional operators is extremely challenging. In recent years, various numerical solutions have been presented to handle these challenges, including9–13. Many researchers from practically all disciplines of science, technology, and engineering have been drawn to the notion of non-local operators of differentiation in recent years because of their ability to integrate more complicated natural phenomena into mathematical Eqs14,15. Three dominants were suggested in this field: the power law, the exponential decay law, and the extended Mittag-Leffler law. To identify them, various names were given to them, such as Caputo operators for power law or non-local and singular kernel types, Caputo-Fabrizio for non-singular local types, and Atangana-Balaena for non-local and non-singular types. The kernel Mittag-Leffler function proved to be more general than the power law and exponential decay functions16. As an outcome, Caputo and Caputo-Fabrizio are distinct Atangana-Balaena fractional operators. Atangana introduced the Fractal Fractional (FF) derivative operator with two parameters in 201716. The first parameter represents the fractional order, while the second is the fractal dimension. The fractal operators exhibit both memory effect and self-similarity qualities. Furthermore, the FF differentiation represents fractal kinetics, in which continuous time is substituted for fractal time. FF operators have an important role in many branches of science, such as the simulation of the Ebola virus17, a chaotic, different problem18, hierarchical and porous media19,20, banking and finance21,22, multi-dimensional electronic devices23, Newtonian and non-Newtonian fluids24,25, electronic circuit26–28, quantum particle29, nanofluid30, waterborne disease31. A number of articles have been published in this branch of science32,33, such as Araz et al.34 offered a numerical method for solving the FF integro-differential equation where the derivative is in the Atangana-Balaena sense, and the integral is in the Caputo sense. Sweilam et al.35 solved the FF optimal control problem via the nonstandard finite difference method. Heydari et al. in36 used the Legendre polynomials for solving nonlinear FF optimal control problems. Magin et al. in37 investigated the FF two-dimensional Emden-Fowler equation via the two-dimensional Chelyshkov polynomials. Dehestani et al.38 provided the Mott polynomials method for the numerical solution of FF two-dimensional Fredholm-Volterra integro-differential equations. Sabermahani et al. in39 introduced a numerical scheme based on the generalized Lucas wavelets method for solving nonlinear FF optimal control problems. Rahim Khani et al. in40 investigated the numerical treatment of FF 2D optimal control problems by the Müntz-Legendre polynomials. Rahim Khani et al.41 used the fractional shifted Morgan-Voyce neural networks to solve the FF pantograph differential equations. Khan et al.42 applied the two-step Lagrange polynomials method for solving a fractal-fractional type tuberculosis mathematical model. Artificial Neural Networks (ANNs) remain the central modeling tool for approximating complex nonlinear mappings across various fields of science and engineering43–45, according to Dong et al.46. Modern deep neural networks, including convolutional, recurrent, graph, and transformer architectures, have revolutionized numerous fields47. Recent comprehensive reviews highlight the rapid evolution of physics-informed deep learning and its role in next-generation scientific computing48. A landmark advancement in this direction is the development of neural-type operators that learn mappings, such as those in infinite-dimensional function spaces49. Domain-decomposition methods, including Finite Basis physics-informed neural networks (FBPINNs) have contributed greatly to the scaling of large-domain and stiff Differential Equations (DEs)50. Transfer-learning extensions of Physics-informed neural networks (PINNs) have further enabled efficient solutions to high-frequency and multi-scale problems51. Specialized splitting techniques within the PINN framework have been successfully applied to infer the dynamics of both integer- and fractional-order neuron models52. Gradient-enhanced fractional PINNs have demonstrated superior performance on multiterm time-fractional Burger-type equations53. Advanced neural architectures have also been developed for systems of coupled fractional differential equations54 and for pricing problems governed by time-fractional Black–Scholes equations55. The Caputo–Fabrizio fractional derivative with its non-singular exponential kernel was introduced as a physically consistent alternative to singular-kernel operators56. Despite progress, classical numerical schemes for fractional and fractal-fractional equations, whether implicit finite-difference57, wavelet collocation58, or spectral methods remain computationally expensive due to non-local memory terms, often scale poorly with time horizon, and struggle with stiffness and weak singularities59,60. While the Vieta-Fibonacci polynomial (VFP) collocation method61 provides accurate approximations for FF differential equations and systems using a few basic functions \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:(e.g.,\:m=\mathrm{8,10,12})$$\end{document} across Caputo, Atangana-Balaena, and Caputo-Fabrizio operators, it faces limitations compared to Physics-Informed Neural Networks tailored for Caputo-Fabrizio derivatives (PINN-CF), including reliance on structured polynomial expansions and fixed collocation points that lead to ill-conditioning, sensitivity to basis selection, convergence only for moderate m values while potentially diverging for higher-order polynomials due to numerical instability in the algebraic systems, challenges in high-dimensional or irregular domains, and limited scalability for chaotic long-term dynamics without explicit error propagation from data noise. This restricts its applicability for long-term chaotic simulations or high-dimensional problems, highlighting a research gap in handling noisy data, inverse problems, and adaptive refinement, where PINN-CF excels through mesh-free, physics-embedded neural training for robust parameter inference and complex domains in applications like epidemics or fluid dynamics. For neural-network-based solvers the comprehensive convergence of deep Galerkin methods and physics-informed for a large class of differential equations provide solid validation53. Together with the ongoing limitations of the traditional discretization methods these convergence guarantees48–52 motivate the extension of a passionate, mesh-free framework of physics-informed modeling, explicitly tailored to the CF operator. The classical Caputo derivative is defined with a power law kernel which is singular at initial time. This singularity has been used in some physical systems but can cause transient behavior. The Caputo Fabrizio (CF) operator is one of the fractional operators that utilizes the exponential kernel instead of singular kernel. For problems where the system’s memory declines efficiently, lacking an initial singularity making CF more physically reliable. Compared to the Atangana-Baleanu operator61, the CF operator is simpler computationally and still catching non-local behaviour accurately. The CF operator used in PIINN-CF also compared with PINN solutions which use the Caputo fractional operator. The comparison of both the operators (Caputo Fabrizio and Caputo fractional) shows that CF works better than simple Caputo operator. Therefore, to highlight the novelty of the study we adopt the CF derivative in this work to balance physical realism, computational efficiency and ease of integration into the PINN framework and tested with the three systems (Pantograph-type delay equations, Memristor-based neural circuits, and FFDE systems of convective fluid motion in rotating cavities). The PINN-CF, representing the neural solver that directly embeds the exact non-singular Caputo–Fabrizio derivative into the loss function via automatic differentiation. By eliminating explicit convolution, avoiding mesh dependence, and leveraging the convergence properties established in62 the proposed PINN-CF achieves higher accuracy, better stability, and significantly reduced computational cost on a range of fractional and fractal-fractional DEs with non-singular kernels. The reminder of this paper is organized as follows. Section 2 presents the fundamental concept of governing equations. Section 3 presents the proposed PINN-CF framework including the neural networks architecture and loss function formulation. Section 4 presents the numerical implementation details and discusses the results obtained for different parameter settings and comparison with PINN solution. Section 5 concludes the paper, highlights the limitations and suggests directions for future research. Fig. 1Graphical abstract of this manuscript. Graphical abstract of this manuscript. Governing equations of numerical examples In this section, mathematical formulations of the Pantograph-type delay equations (feedback control, biological growth), Memristor-based neural circuits (neuromorphic computing), FFDE systems of convective fluid motion in rotating cavities (geophysical flows) are presented. FF pantograph differential equations Delay differential equations (DDEs) have numerous applications in real problems such as economy, long transmission lines, automatic control, biology, hydraulic networks, physiological kinetics, and so forth37. One of the most prominent fractional DDEs is the fractional pantograph equation (FPE), which plays a vital role in various mathematical physics. In detail, these equations have appeared in control systems, probability, electrodynamics, quantum mechanics, and number theory. It is precisely due to this fact that several schemes have been constructed for the numerical solution of FPE, such as the fractional-order Bernoulli wavelets63, Müntz-Legendre wavelets64, fractal fractional calculus by spectral method65, piecewise fractional-order Taylor functions66, fractional order Legendre wavelets54, spectral-collocation67, Legendre multiwavelets collocation68, and Hermite wavelets methods69. This study introduces PINN-CF, consider the following FF pantograph differential equation61.1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:}^{FF}{D}^{\alpha\:,\beta\:}{\Delta\:}\left(t\right)=\alpha\:{\Delta\:}\left(t\right)+\beta\:{\Delta\:}\left(\gamma\:t\right)$$\end{document} subject to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\Delta\:}\left(0\right)={{\Delta\:}}_{0}$$\end{document}, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:,\:\beta\:,\:{{\Delta\:}}_{0}$$\end{document} are the real constants, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\Delta\:}\left(\gamma\:t\right)$$\end{document} is the stretched function and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:}^{FF}{D}^{\alpha\:,\beta\:}$$\end{document} is the FF derivative of order\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\:(\alpha\:,\:\beta\:)$$\end{document}. Derivation of the governing FF differential equations system of the memristor A rectified electrical component, known as a memristor, in a circuit it regulates the flow of electrical current while also remembering the amount of charge. To stores memory without power is the main advantage of a memristor is that it. This electrical component is passive, nonlinear and two-terminal, with the ability to link magnetic flux with electric charge using nonvolatile memory. The memristor’s essential capability is that it increases the flow of current in one direction while in the opposite direction decreasing the flow of current using resistance switching. The chaotic circuit depicted in (Fig. 2 see Abro et al.70 design with four dynamic components (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{i}_{1}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{i}_{2},\:{v}_{c},$$\end{document} and q), and two inductors, a capacitor, and an emulator that includes a negative conductance and a resistor. The circuit in this figure is studied using the volt-ampere characteristic of its elements and Kirchhoff’s principles to generate the state equations, which include nonlinear terms, as explained below70.2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left.\begin{array}{rr}&\:\frac{dq}{dt}={i}_{1}\\\:&\:\frac{d{v}_{c}}{dt}+\frac{{i}_{1}}{C}-\frac{{i}_{2}}{C}=0\\\:&\:\frac{d{i}_{2}}{dt}+\frac{{v}_{c}}{{L}_{2}}+\frac{{i}_{2}}{G{L}_{2}}=0\\\:&\:\frac{d{i}_{1}}{dt}-\frac{{v}_{c}}{{L}_{1}}+\frac{{v}_{m}\left(q\right)}{{L}_{1}}+\frac{R{i}_{1}}{{L}_{1}}=0\end{array}\right\}$$\end{document} In this context, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{v}_{m}\left(q\right)$$\end{document} is the terminal voltage. Applying the voltage across the memristor and by substituting the terminal voltage, the state Eq. 2, which contain nonlinear terms, become70. 3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left.\begin{array}{rr}&\:\frac{dq}{dt}-{i}_{1}=0\\\:&\:\frac{d{v}_{c}}{dt}+\frac{{i}_{1}}{C}-\frac{{i}_{2}}{C}=0\\\:&\:\frac{d{i}_{2}}{dt}+\frac{{v}_{c}}{{L}_{2}}+\frac{{i}_{2}}{G{L}_{2}}=0\\\:&\:\frac{d{i}_{1}}{dt}-\frac{{v}_{c}}{{L}_{1}}+\frac{R{i}_{1}}{{L}_{1}}+\frac{1}{{L}_{1}}\left(\frac{{R}_{2}{R}_{0}^{2}q}{10{C}_{1}{R}_{3}{R}_{1}}-{R}_{3}^{-1}{R}_{2}{R}_{0}\right)=0\end{array}\right\}$$\end{document} Fig. 2Geometry of the problem. Geometry of the problem. To derive the system’s dynamic equation, a timescale transformation is performed, considering the following transformations70.4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left.\begin{array}{rr}&\:{L}_{2}{L}_{1}^{-1}=a,{L}_{2}{C}^{-1}=b,C=R,f={G}^{-1},e=-{R}_{3}^{-1}{R}_{2}{R}_{0},d={R}_{2}{R}_{0}^{2}{\left({C}_{1}{R}_{3}{R}_{1}\right)}^{-1}\\\:&\:{{\Delta\:}}_{4}={L}_{2}^{-1}q,{{\Delta\:}}_{3}={i}_{2},t=d\tau\:{L}_{2},{{\Delta\:}}_{1}={i}_{1},{{\Delta\:}}_{2}={v}_{c}\end{array}\right\}$$\end{document} The parameters used in Eq. (4) to implement the memristor are specified as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{R}_{0}=1f{\Omega\:},{R}_{1}=500{\Omega\:},{R}_{2}=1f{\Omega\:},{R}_{3}=1f{\Omega\:},{C}_{1}=100dF$$\end{document}. By applying Eq. (4) to (3), to obtain the evolutionary differential equations simplified system with nonlinear terms70.5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left.\begin{array}{rr}&\:\frac{d}{dt}{{\Delta\:}}_{1}\left(t\right)-a\left(d{{\Delta\:}}_{4}\left(t\right){{\Delta\:}}_{1}\left(t\right)-{{\Delta\:}}_{2}\left(t\right)+e{{\Delta\:}}_{1}\left(t\right)+c{{\Delta\:}}_{1}\left(t\right)\right)=0\\\:&\:\frac{d}{dt}{{\Delta\:}}_{2}\left(t\right)+b{{\Delta\:}}_{1}\left(t\right)-b{{\Delta\:}}_{3}\left(t\right)=0\\\:&\:\frac{d}{dt}{{\Delta\:}}_{3}\left(t\right)-f{{\Delta\:}}_{3}\left(t\right)+{{\Delta\:}}_{2}\left(t\right)=0\\\:&\:\frac{d}{dt}{{\Delta\:}}_{4}\left(t\right)={{\Delta\:}}_{1}\left(t\right)\end{array}\right\}$$\end{document} with the initial conditions6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{{\Delta\:}}_{1}\left(0\right)={{\Delta\:}}_{2}\left(0\right)={{\Delta\:}}_{3}\left(0\right)={{\Delta\:}}_{4}\left(0\right)={{\Delta\:}}_{i}$$\end{document} The phase pictures and chaotic phenomena can be obtained by defining the embed the parameters in Eq. (5), as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:2=a,b=\mathrm{1,0.2}=c,e=-0.002,d=0.04,f=0.92$$\end{document}. By reformulating the system of evolutionary DEs integrating nonlinear variables, as in (5) and using the concept of the FF differential operator, to change the governing nonlinear DEs of the memristor system given in (7) as follows70:.7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left.\begin{array}{rr}&\:{\:}^{FF}{D}^{\alpha\:,\beta\:}{{\Delta\:}}_{1}\left(t\right)-a\left(d{{\Delta\:}}_{4}\left(t\right){{\Delta\:}}_{1}\left(t\right)+e{{\Delta\:}}_{1}\left(t\right)+c{{\Delta\:}}_{1}\left(t\right)\right)={{\Delta\:}}_{2}\left(t\right)\\\:&\:{\:}^{FF}{D}^{\alpha\:,\beta\:}{{\Delta\:}}_{2}\left(t\right)+b{{\Delta\:}}_{1}\left(t\right)-b{{\Delta\:}}_{3}\left(t\right)=0\\\:&\:{\:}^{FF}{D}^{\alpha\:,\beta\:}{{\Delta\:}}_{3}\left(t\right)-f{{\Delta\:}}_{3}\left(t\right)+{{\Delta\:}}_{2}\left(t\right)=0\\\:&\:{\:}^{FF}{D}^{\alpha\:,\beta\:}{{\Delta\:}}_{4}\left(t\right)-{{\Delta\:}}_{1}\left(t\right)=0\end{array}\right\}$$\end{document} with the initial conditions8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{{\Delta\:}}_{1}\left(0\right)={{\Delta\:}}_{2}\left(0\right)={{\Delta\:}}_{3}\left(0\right)={{\Delta\:}}_{4}\left(0\right)={{\Delta\:}}_{i}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:a,\:b,\:c,\:d,\:e,\:f$$\end{document} are the real constants. Bao et al. in71 introduced a circuit with two memristors and dimensionless mathematical model based on a fifth-order chaotic. They investigated dynamical analysis schemes, stability analysis, and the memristor initial states. Abro et al. in70 presented a numerical scheme based on FF differential operator in the Atangana-Balaena, Caputo-Fabrizio, and Caputo senses for solving this system. Also, this problem was discussed in references72–75. Governing FF differential equations system of convective fluid motion in rotating cavity By creating rapid pressure changes in fluids, such a situation leads to cavitation, which is called a cavity. The cavity has found different applications in references76–78 as. Kidney stones destruction with shock wave lithotripsy,For mixing and breaking down use of homogenized and colloidal fluids,Adhesion forces between particle and substrate,Egg pasteurization,Surface of tidal stream turbines. Kidney stones destruction with shock wave lithotripsy, Kidney stones destruction with shock wave lithotripsy, For mixing and breaking down use of homogenized and colloidal fluids, For mixing and breaking down use of homogenized and colloidal fluids, Adhesion forces between particle and substrate, Adhesion forces between particle and substrate, Egg pasteurization, Egg pasteurization, Surface of tidal stream turbines. Surface of tidal stream turbines. By analyzing the convection of an incompressible, viscous fluid within an ellipsoid under conditions of non-uniform external heating79. The ellipsoid types are \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\left(\frac{{x}_{1}}{{a}_{1}}\right)}^{2}+{\left(\frac{{x}_{2}}{{a}_{2}}\right)}^{2}+{\left(\frac{{x}_{3}}{{a}_{3}}\right)}^{2}=1$$\end{document} under the geometrical condition given as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{1}>{a}_{2}>{a}_{3}>0$$\end{document}. An ellipsoid’s heat sources rotate around its axis with a constant velocity \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\Omega\:}$$\end{document}. The gravity vector is anticipated to correspond to the ellipsoid’s orientation in space. The gravitational field’s impact is ignored because centrifugal forces are present. Around axis \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{x}_{3}\:\:$$\end{document} the ellipsoid spins with a constant angle \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{I}_{0}=\left({a}_{3}\mathrm{c}\mathrm{o}\mathrm{s}\left(\alpha\:\right),0,{a}_{1}\mathrm{s}\mathrm{i}\mathrm{n}\left(\alpha\:\right)\right)$$\end{document} at gravity vector \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:g(\Vert\:g\mid\:=g)$$\end{document}, as shown in Fig. 2. The vector g is placed in the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{x}_{1},{x}_{2}$$\end{document} plane. Based on the above assumptions, Glukhovsky and Dolzhansky76, in a rotating cavity a nonlinear mathematical model of convective fluid motion developed as follows79.9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left.\begin{array}{rr}&\:\frac{d}{dt}{{\Delta\:}}_{1}\left(t\right)+\left(b{{\Delta\:}}_{1}\left(t\right)-C{{\Delta\:}}_{3}\left(t\right)-A{{\Delta\:}}_{2}\left(t\right){{\Delta\:}}_{3}\left(t\right)\right)=0\\\:&\:\frac{d}{dt}{{\Delta\:}}_{2}\left(t\right)+\left({{\Delta\:}}_{1}\left(t\right){{\Delta\:}}_{3}\left(t\right)+{{\Delta\:}}_{2}\left(t\right)-{R}_{0}\right)=0\\\:&\:\frac{d}{dt}{{\Delta\:}}_{3}\left(t\right)+\left({{\Delta\:}}_{3}\left(t\right)\right)={{\Delta\:}}_{1}\left(t\right){{\Delta\:}}_{2}\left(t\right)\end{array}\right\}$$\end{document} The rheological parameters in a rotating cavity of convective fluid motion for the nonlinear mathematical model, as given in Eq. (10), are as follows79.10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left.\begin{array}{rr}&\:A={T}_{0}^{-1}{\mathrm{cos}}^{2}\alpha\:\left({a}_{1}^{2}-{a}_{2}^{2}\right){\left({a}_{1}^{2}+{a}_{2}^{2}\right)}^{-1},{\mu\:}^{-1}\lambda\:=b,2b{a}_{2}{a}_{1}^{2}{a}_{3}^{-1}\mathrm{sin}\alpha\:{\left({a}_{1}^{2}+{a}_{2}^{2}\right)}^{-1}=C,\\\:&\:{T}_{a}={{\Omega\:}}_{0}^{2}{\lambda\:}^{-2},{{\Delta\:}}_{1}\left(t\right)=\frac{{q}_{3}\left(t\right)g\beta\:{a}_{3}\mathrm{c}\mathrm{o}\mathrm{s}\alpha\:}{2{{\Omega\:}}_{0}{a}_{2}{a}_{1}}+{\omega\:}_{3}\left(t\right),{{\Delta\:}}_{2}\left(t\right)=\frac{{q}_{1}\left(t\right)g\beta\:{a}_{3}}{2{a}_{2}{a}_{1}\mu\:\lambda\:}\\\:&\:{{\Delta\:}}_{3}\left(t\right)=\frac{{q}_{2}\left(t\right)g\beta\:{a}_{3}}{2{a}_{2}{a}_{1}\mu\:\lambda\:},{R}_{a}=\frac{{q}_{0}g\beta\:{a}_{3}}{2{a}_{2}{a}_{1}\mu\:\lambda\:},{\omega\:}_{1}\left(t\right)=\frac{{q}_{1}\left(t\right)\mathrm{c}\mathrm{o}\mathrm{s}\alpha\:\beta\:{a}_{3}}{2{{\Omega\:}}_{0}{a}_{2}{a}_{1}},{\omega\:}_{2}\left(t\right)=\frac{{q}_{2}\left(t\right)\mathrm{c}\mathrm{o}\mathrm{s}\alpha\:\beta\:{a}_{3}}{2{{\Omega\:}}_{0}{a}_{2}{a}_{1}}.\end{array}\right\}$$\end{document} The parameters carry their usual meanings in Eq. (11), for instance, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\uplambda\:}$$\end{document} represents viscosity coefficients, volume expansion denoted by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\upbeta\:}$$\end{document} and heat conduction signified by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\mu\:$$\end{document}. The fluid angular velocity vectors projection onto the axes \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{x}_{1},{x}_{2}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{x}_{3}$$\end{document} are identified by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{{\upomega\:}}_{1}\left(\mathrm{t}\right),{{\upomega\:}}_{2}\left(\mathrm{t}\right)$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{{\upomega\:}}_{3}\left(\mathrm{t}\right)$$\end{document}. Along the principal axes of the ellipsoid (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{1},{a}_{2},{a}_{3}$$\end{document}) the temperature differences are denoted by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{q}_{1}\left(t\right),{q}_{2}\left(t\right)$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{q}_{3}\left(t\right)$$\end{document}, respectively. By applying the transformation on Eq. (11) as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{{\Delta\:}}_{3}\to\:\frac{{{\Delta\:}}_{2}b{C}^{-1}}{1+R{a}_{0}},{{\Delta\:}}_{2}\to\:R-\frac{{{\Delta\:}}_{3}b{C}^{-1}}{1+R{a}_{0}},{{\Delta\:}}_{1}\to\:{{\Delta\:}}_{1}$$\end{document}, the system derived as70. 11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left.\begin{array}{rr}&\:\frac{d}{dt}{{\Delta\:}}_{1}\left(t\right)+a{{\Delta\:}}_{2}\left(t\right){{\Delta\:}}_{3}\left(t\right)+b\left({{\Delta\:}}_{1}\left(t\right)\right)={{\Delta\:}}_{2}\left(t\right)\\\:&\:\frac{d}{dt}{{\Delta\:}}_{2}\left(t\right)+{{\Delta\:}}_{2}\left(t\right)+{{\Delta\:}}_{1}\left(t\right){{\Delta\:}}_{3}\left(t\right)=c{{\Delta\:}}_{1}\left(t\right)\\\:&\:\frac{d}{dt}{{\Delta\:}}_{3}\left(t\right)+{{\Delta\:}}_{3}\left(t\right)=-{{\Delta\:}}_{1}\left(t\right){{\Delta\:}}_{2}\left(t\right)\end{array}\right\}$$\end{document} with the initial conditions 12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{{\Delta\:}}_{1}\left(0\right)=0.1,{{\Delta\:}}_{2}\left(0\right)=0.2,{{\Delta\:}}_{3}\left(0\right)={{\Delta\:}}_{i}$$\end{document} After adequate linear transformation, the parameters are derived for Eq. (13) as13\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:C{R}_{a}=R,{A}_{0}={C}^{-2}A,{b}^{2}{a}_{0}{\left(1+R{a}_{0}\right)}^{-2}=a,C=\left(1+R{a}_{0}\right)R{b}^{-1}$$\end{document} The conventional model of convective fluid motion in a rotating chamber, represented by Eq. (11), is not able to convey the full memory of the physical process. In a rotating cavity due to nonlinearity in the classical model of convective fluid motion, the model in (11) was expanded by substituting the recently suggested fractal fractional derivatives of Caputo-Fabrizio sensors in the non-integer form for the classical derivative. Using fractal fractional derivatives to develop the traditional model of convective fluid dynamics in a rotating cavity79. 14\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left.\begin{array}{rr}&\:{\:}^{FF}{D}^{\alpha\:,\beta\:}{{\Delta\:}}_{1}\left(t\right)+a{{\Delta\:}}_{2}\left(t\right){{\Delta\:}}_{3}\left(t\right)+b\left({{\Delta\:}}_{1}\left(t\right)-{{\Delta\:}}_{2}\left(t\right)\right)=0,\\\:&\:{\:}^{FF}{D}^{\alpha\:,\beta\:}{{\Delta\:}}_{2}\left(t\right)-c{{\Delta\:}}_{1}\left(t\right)+{{\Delta\:}}_{2}\left(t\right)+{{\Delta\:}}_{1}\left(t\right){{\Delta\:}}_{3}\left(t\right)=0,\\\:&\:{\:}^{FF}{D}^{\alpha\:,\beta\:}{{\Delta\:}}_{3}\left(t\right)-{{\Delta\:}}_{1}\left(t\right){{\Delta\:}}_{2}\left(t\right)+{{\Delta\:}}_{3}\left(t\right)=0,\end{array}\right\}$$\end{document} with the initial conditions 15\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{{\Delta\:}}_{1}\left(0\right)=0.01,{{\Delta\:}}_{2}\left(0\right)=0.02,{{\Delta\:}}_{3}\left(0\right)={{\Delta\:}}_{i}$$\end{document} where a,b,c are the real constants79. Cai et al. in80 employed the scaling transformation scheme to check the role of fractal derivative to model the viscoelastic behavior of fluids. Ji et al.81 studied the role with the fractional order derivative of the fractal reservoir in the flow of fluids by using the finite integral transform. Abro et al. in79 investigated FF mathematical model in a rotating cavity inside the ellipsoid of convective fluid motion with inhomogeneous external heating. Mathematical foundation of PINN-CF PINN-CF architecture The PINN-CF approximates the solution \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:u\left(x,t\right)$$\end{document} as in Eq. 16. The complete framework of PINN-CF is presented in Fig. 3.16\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\varDelta\:\left(t\right)\approx\:{\varDelta\:}_{\theta\:}\left(t;\theta\:\right)$$\end{document} The network receives the temporal coordinate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left(t\right)$$\end{document} as input and outputs a scalar approximation of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\varDelta\:$$\end{document}. The proposed PINN-CF approximates the solution \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varDelta\:}_{\theta\:}\left(t;\theta\:\right)\:$$\end{document}using a feed-forward neural network with three hidden layers. In operator form the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varDelta\:}_{\theta\:}\left(t;\theta\:\right)$$\end{document} is defined as in Eq. (17)17\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varDelta\:}_{\theta\:}(t;\theta\:)={\mathcal{N}}^{\left(4\right)}\left(\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{h}\left({\mathcal{N}}^{\left(3\right)}\left(\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{h}\left({\mathcal{N}}^{\left(2\right)}\left(\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{h}\left({\mathcal{N}}^{\left(1\right)}\left(t\right)\right)\right)\right)\right)\right)\right)$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\mathcal{N}}^{\left(k\right)}$$\end{document} denotes the affine transformation at layer k. While in expended form of Eq. (17) given in Eq. (18)18\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varDelta\:}_{\theta\:}\left(t\right)={\mathrm{W}}^{\left(4\right)}\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{h}\left({\mathrm{W}}^{\left(3\right)}\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{h}\left({\mathrm{W}}^{\left(2\right)}\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{h}\left({\mathrm{W}}^{\left(1\right)}\left(\mathrm{t}\right)+{\mathrm{b}}^{\left(1\right)}\right)+{\mathrm{b}}^{\left(2\right)}\right)+{\mathrm{b}}^{\left(3\right)}\right)+{b}^{\left(4\right)}$$\end{document} here, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\theta\:$$\end{document} represents the trainable parameters (weights and biases) of the neural network. The parameters set is given in Eq. (19)19\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\theta\:=\left\{{\mathrm{W}}^{\left(1\right)},{\mathrm{W}}^{\left(2\right)},{\mathrm{W}}^{\left(3\right)},{\mathrm{W}}^{\left(4\right)},{\mathrm{b}}^{\left(1\right)},{\mathrm{b}}^{\left(2\right)},{\mathrm{b}}^{\left(3\right)},{b}^{\left(4\right)}\right\}$$\end{document} The weights of the neural network are initialized using the Glorot normal scheme given in Eq. (20)20\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{W}_{ij}\sim\:\mathcal{N}\left(0,{\sigma\:}^{2}\right){\sigma\:}^{2}=\frac{2}{{n}_{\mathrm{in\:}}+{n}_{\mathrm{out\:}}}$$\end{document} The PINN-CF solution is expressed as a finite sum of Tanh activation functions, the series form given in Eq. (21).21\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varDelta\:}_{\theta\:}\left(t\right)={\sum\:}_{j=1}^{m}\:{\phi\:}_{j}\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{h}\left({w}_{j}\mathrm{t}+{\gamma\:}_{j}\right)+b$$\end{document} The proposed method employs the hyperbolic tangent (Tanh) as the activation function and its exponential form given as,22\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{h}\left(z\right)=\frac{{e}^{z}-{e}^{-z}}{{e}^{z}+{e}^{-z}}$$\end{document} Now z=\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{w}_{j}\mathrm{t}+{\gamma\:}_{j}$$\end{document} and Eq. (21) in exponential form is given in Eq. (23)23\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varDelta\:}_{\theta\:}\left(t\right)=\sum\:_{j=1}^{50}\:{\phi\:}_{j}\left(\frac{{e}^{{w}_{j}t+{\gamma\:}_{j}}-{e}^{-\left({w}_{j}t+{\gamma\:}_{j}\right)}}{{e}^{{w}_{j}t+{\gamma\:}_{j}}+{e}^{-\left({w}_{j}t+{\gamma\:}_{j}\right)}}\right)+b$$\end{document} Expanded series representation of Eq. (23) given in Eq. (24)24\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varDelta\:}_{\theta\:}\left(t\right)=\left\{\begin{array}{l}{\phi\:}_{1}\left(\frac{{e}^{{w}_{1}t+{\gamma\:}_{1}}-{e}^{-\left({w}_{1}t+{\gamma\:}_{1}\right)}}{{e}^{{w}_{1}t+{\gamma\:}_{1}+{e}^{-\left({w}_{1}t+{\gamma\:}_{1}\right)}}}\right)+{\phi\:}_{2}\left(\frac{{e}^{{w}_{2}t+{\gamma\:}_{2}}-{e}^{-\left({w}_{2}t+{\gamma\:}_{2}\right)}}{{e}^{{w}_{2}{t}^{t+{\gamma\:}_{2}}+{e}^{-\left({w}_{2}t+{\gamma\:}_{2}\right)}}}\right)\\\:+{\phi\:}_{3}\left(\frac{{e}^{{w}_{3}t+{\gamma\:}_{3}}-{e}^{-\left({w}_{3}t+{\gamma\:}_{3}\right)}}{{e}^{{w}^{t}+{\gamma\:}_{3}}+{e}^{-\left({w}_{3}t+{\gamma\:}_{3}\right)}}\right)+{\phi\:}_{4}\left(\frac{{e}^{{w}_{4}t+{\gamma\:}_{4}-{e}^{-\left({w}_{4}t+{\gamma\:}_{4}\right)}}}{{e}^{{w}_{4}t{\gamma\:}_{4}+{e}^{-\left({w}_{4}t+{\gamma\:}_{4}\right)}}}\right)\\\:+{\phi\:}_{5}\left(\frac{{e}^{{w}_{5}t+{\gamma\:}_{5}-{e}^{-\left({w}_{5}t+{\gamma\:}_{5}\right)}}}{{e}^{{w}_{5}t+{\gamma\:}_{5}}+{e}^{-\left({w}_{5}t+{\gamma\:}_{5}\right)}}\right)+{\phi\:}_{6}\left(\frac{{e}^{{w}_{6}t+{\gamma\:}_{6}-{e}^{-\left({w}_{6}t+{\gamma\:}_{6}\right)}}}{{e}^{{w}_{6}t+{\gamma\:}_{6}+{e}^{-\left({w}_{6}t+{\gamma\:}_{6}\right)}}}\right)\\\:+{\phi\:}_{7}\left(\frac{{e}^{{w}_{7}t+{\gamma\:}_{7}-{e}^{-\left({w}_{7}t+{\gamma\:}_{7}\right)}}}{{e}^{{w}_{7}t+{\gamma\:}_{7}+{e}^{-\left({w}_{7}t+{\gamma\:}_{7}\right)}}}\right)+{\phi\:}_{8}\left(\frac{{e}^{{w}_{8}t+{\gamma\:}_{8}-{e}^{-\left({w}_{8}t+{\gamma\:}_{8}\right)}}}{{e}^{{w}_{8}t+{\gamma\:}_{8}+{e}^{-\left({w}_{8}t+{\gamma\:}_{8}\right)}}}\right)\\\:+{\phi\:}_{9}\left(\frac{{e}^{{w}_{9}t+{\gamma\:}_{9}-{e}^{-\left({w}_{9}t+{\gamma\:}_{9}\right)}}}{{e}^{{w}_{9}t+{\gamma\:}_{9}+{e}^{-\left({w}_{9}t+{\gamma\:}_{9}\right)}}}\right)+{\phi\:}_{10}\left(\frac{{e}^{{w}_{10}t+{\gamma\:}_{10}-{e}^{-\left({w}_{10}t+{\gamma\:}_{10}\right)}}}{{e}^{{w}_{10}t+{\gamma\:}_{10}+{e}^{-\left({w}_{10}t+{\gamma\:}_{10}\right)}}}\right)\\\:+\cdots\:+{\phi\:}_{50}\left(\frac{{e}^{{w}_{50}t+{\gamma\:}_{50}-{e}^{-\left({w}_{50}t+{\gamma\:}_{50}\right)}}}{{e}^{{w}_{50}t+{\gamma\:}_{50}+{e}^{-\left({w}_{50}t+{\gamma\:}_{50}\right)}}}\right)+b\end{array}\right\}$$\end{document} Now compact matrix exponential form of Eq. (23) is given in Eq. (\top is only supported in math mode 25). The data of weights/neurons corresponding to matrices \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\phi\:$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\mathrm{w}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\gamma\:\:$$\end{document} for some examples are given in appendix, Table 9. 25\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \:_{{\theta \:}} \left( t \right) = \phi \:^{{{ \top }\:}} \left( {\frac{{e^{{wt + \gamma \:}} - e^{{ - (wt + \gamma \:)}} }}{{e^{{wt + \gamma \:}} + e^{{ - (wt + \gamma \:)}} }}} \right) + b$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{e}^{\mathrm{w}t+\gamma\:}$$\end{document} denotes element-wise exponential, and the terms \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\phi\:$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\mathrm{w}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\gamma\:\:$$\end{document} in matrix form given by Eq. (\top is only supported in math mode 26) 26\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{gathered} \:\phi \: = \left[ {\begin{array}{*{20}l} {\phi \:_{1} } \hfill & {\:\phi \:_{2} } \hfill & {\: \cdots \:} \hfill & {\:\phi \:_{{50}} } \hfill \\ \end{array} } \right]^{{{ \top }\:}} ,\:w = \left[ {\begin{array}{*{20}l} {w_{1} } \hfill & {\:w_{2} } \hfill & {\: \cdots \:} \hfill & {\:w_{{50}} } \hfill \\ \end{array} } \right]^{{{ \top }\:}} ,\: \hfill \\ \:\gamma \: = \left[ {\begin{array}{*{20}l} {\gamma \:_{1} } \hfill & {\:\gamma \:_{2} } \hfill & {\: \cdots \:} \hfill & {\:\gamma \:_{{50}} } \hfill \\ \end{array} } \right]^{{{ \top }\:}} ,\:\:\phi \:_{j} = W_{{1j}}^{{\left( 4 \right)}} ,\:w_{j} = W_{{j1}}^{{\left( 1 \right)}} ,\:\gamma \:_{j} = b_{j}^{{\left( 1 \right)}} ,\:b = b^{{\left( 4 \right)}} \hfill \\ \end{gathered}$$\end{document} The dimension of weights and biases in this study given as Eq. (27)27\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left.\begin{array}{rrrrr}{\mathrm{W}}^{\left(1\right)}&\:\:\in\:{\mathbb{R}}^{50\times\:1},&\:\:\:\:\:\:\:\:\:&\:{\mathrm{b}}^{\left(1\right)}&\:\:\in\:{\mathbb{R}}^{50\times\:1}\\\:{\mathrm{W}}^{\left(2\right)}&\:\:\in\:{\mathbb{R}}^{50\times\:50},&\:\:\:\:\:\:\:\:\:&\:{\mathrm{b}}^{\left(2\right)}&\:\:\in\:{\mathbb{R}}^{50\times\:1}\\\:{\mathrm{W}}^{\left(3\right)}&\:\:\in\:{\mathbb{R}}^{50\times\:50},&\:\:\:\:\:\:\:\:\:&\:{\mathrm{b}}^{\left(3\right)}&\:\:\in\:{\mathbb{R}}^{50\times\:1}\\\:{\mathrm{W}}^{\left(4\right)}&\:\:\in\:{\mathbb{R}}^{1\times\:50},&\:\:\:\:\:\:\:\:\:&\:{b}^{\left(4\right)}&\:\:\in\:\mathbb{R}\end{array}\right\}$$\end{document} Now for single term28\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varDelta\:}_{\theta\:}\left(t\right)={\sum\:}_{j=1}^{50}\:{\phi\:}_{j}^{\left(1\right)}\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{h}\left({w}_{j}^{\left(1\right)}t+{\gamma\:}_{j}^{\left(1\right)}\right)+{b}^{\left(1\right)}$$\end{document} Neural network approximation Let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:N(t;\:\varDelta\:)$$\end{document} be a fully connected feedforward neural network mapping \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:t\:\to\:$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{R}^{m}$$\end{document} with trainable parameters \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\varDelta\:$$\end{document}. The hard initial condition \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{{\Delta\:}}_{\mathrm{j}}\left(\mathrm{t};{\Delta\:}\right)={{\Delta\:}}_{0,\mathrm{j}}+\mathrm{t}\cdot\:{\mathrm{N}}_{\mathrm{j}}\left(\mathrm{t};{\Delta\:}\right)$$\end{document} is used in single equation while the soft initial conditions \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{{\Delta\:}}_{\mathrm{j}}(\mathrm{t};{\Delta\:})={{\Delta\:}}_{0,\mathrm{j}}+\mathrm{t}\cdot\:{\mathrm{N}}_{\mathrm{j}}(\mathrm{t};{\Delta\:})$$\end{document} are used in coupled systems of differential equations. Caputo fractional differentiation approximation The CFFD combines a non-singular exponential kernel with fractional integration, defined as29\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:}^{CF}{D}_{t}^{\alpha\:,\beta\:}\varDelta\:\left(t\right)=\beta\:{t}^{(\beta\:-1)}\frac{d\varDelta\:}{dt}\left(t\right)\frac{d\varDelta\:}{dt}\left(\mathrm{t}\right)\mathrm{*}{t}^{(-\alpha\:)}=\frac{1}{{\Gamma\:}(1-\alpha\:)}{\int\:}_{0}^{t}\:(t-\tau\:{)}^{(-\alpha\:)}\frac{d\varDelta\:}{d\tau\:}(\tau\:)d\tau\:$$\end{document} Equivalently, in L1-discretized form used in the proposed method. 30\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:}^{CF}{D}_{t}^{\alpha\:,\beta\:}\varDelta\:\left({t}_{n}\right)\approx\:\beta\:{\left({t}_{n}+\epsilon\:\right)}^{(\beta\:-1)}\frac{1}{{h}^{\alpha\:}}\sum\:_{k=0}^{n-1}{c}_{k}^{\alpha\:}\left[\varDelta\:\left({t}_{n-k}\right)-\varDelta\:\left({t}_{n-k-1}\right)\right]\:$$\end{document} 31\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{c}_{k}^{\lambda\:}\left(\alpha\:\right)=\left[\frac{{\Gamma\:}\left(\alpha\:+1\right)}{{\Gamma\:}\left(k+1\right){\Gamma\:}\left(\alpha\:-k+1\right)}\right](-1{)}^{k}k,\:\:\epsilon\:>0$$\end{document} Design of loss function for PINN-CF The physics residual is the Error Estimation (EE) which has a different construction based on the problem, be it a scalar or coupled. The physics residual is the Error Estimation (EE) which has a different construction based on the problem, be it a scalar or coupled. 32\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{R_j}(t,\Delta \:) = \:{|^{FF}}D_t^{{\alpha _j},{\beta _j}}\Delta {\:_j}\left( t \right) - {{\cal F}_j}\left( {t,\Delta \:\left( t \right),\Delta \:\left( {{R_1}t} \right), \ldots \:} \right)|,j = 1,2,{\rm{ }}n$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\mathcal{R}}_{\theta\:}(t,{\Delta\:})$$\end{document} is the residual of the differential equation predicted by PINN-CF given in Eq. (31)33\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{R}_{\mathrm{j}}\left(t,{\Delta\:}\right)\approx\:{R}_{\theta\:\mathrm{j}}\left(t,{\Delta\:}\right)$$\end{document} In PINN-CF the neural network trained so the residual errors become small as given in Eq. (35)34\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:R_{\theta \:j}(t,\Delta \:) = \:|^{FF}D_t^{\alpha_{\:j},\beta_{\:j}}\Delta_{\:\theta \:j}\left( t \right) - \mathcal{F}_j\left( t,\Delta_{\:\theta \:}\left( t \right),\Delta_{\:\theta \:}\left( t \right)\left( R_1t \right), \ldots \: \right)|, j = 1,2,\ n$$\end{document} 34\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:R_{\theta \:j}(t,\Delta \:) = \:|^{FF}D_t^{\alpha_{\:j},\beta_{\:j}}\Delta_{\:\theta \:j}\left( t \right) - \mathcal{F}_j\left( t,\Delta_{\:\theta \:}\left( t \right),\Delta_{\:\theta \:}\left( t \right)\left( R_1t \right), \ldots \: \right)|, j = 1,2,\ n$$\end{document} Sample collocation points in the given systems are represented by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\mathcal{T}}_{\mathrm{c}}=\{{\mathrm{t}}_{\mathrm{i}}^{\mathrm{c}}{\}}_{\mathrm{i}=1}^{{\mathrm{N}}_{\mathrm{c}}}\sim\:\mathcal{U}(0,\mathrm{T})$$\end{document}, while the collocation points in the delayed form are presented as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\mathcal{T}}_{\mathrm{d}}^{\mathrm{k}}=\left\{{\widehat{{\upgamma\:}}}_{\mathrm{k}}{\mathrm{t}}_{\mathrm{i}}^{\mathrm{c}}\right\}$$\end{document}. The physics loss enforces the fractal-fractional differential equation at a set of randomly sampled collocation points \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\mathrm{N}}_{\mathrm{c}}$$\end{document}uniformly distributed in the domain \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:[0,\:T]$$\end{document}. Using automatic differentiation and the L1-discretized Caputo-Fabrizio fractal-fractional derivative, the residual of the governing equation is computed as given in Eq. (32).35\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\mathcal{L}}_{\mathrm{O}\mathrm{D}\mathrm{E}}\left({\Delta\:}\right)=\frac{1}{\mathrm{m}{\mathrm{N}}_{\mathrm{j}}}\sum\:_{\mathrm{j}=1}^{\mathrm{m}}\:\sum\:_{\mathrm{i}=1}^{\mathrm{N}}\:{R}_{\theta\:\mathrm{j}}{\left({\mathrm{t}}_{\mathrm{i}}^{\mathrm{c}};{\Delta\:}\right)}^{2}$$\end{document} To satisfy the initial conditions exactly and improve training stability, especially in stiff and highly non-local systems, the initial condition loss is introduced. It penalizes any deviation of the neural network output from the prescribed initial values at t=0.36\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\mathcal{L}}_{\mathrm{I}\mathrm{C}}\left({\Delta\:}\right)=\frac{1}{\mathrm{m}}\sum\:_{\mathrm{j}=1}^{\mathrm{m}}\:{\left({{\Delta\:}}_{\theta\:\mathrm{j}}(0;{\Delta\:})-{{\Delta\:}}_{\theta\:\mathrm{j}}\right)}^{2}$$\end{document} Although PINN-CF is primarily unsupervised, a small, supervised data loss term is optionally added using high-accuracy reference solutions at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\mathrm{N}}_{\mathrm{d}}\:$$\end{document}selected data points. This hybrid strategy significantly accelerates convergence and enhances accuracy, particularly for chaotic and long-time integration problems.37\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\mathcal{L}}_{\mathrm{data\:}}\left({\Delta\:}\right)=\frac{1}{\mathrm{m}{\mathrm{N}}_{\mathrm{d}}}\sum\:_{\mathrm{j}=1}^{\mathrm{m}}\:\sum\:_{\mathrm{k}=1}^{{\mathrm{N}}_{\mathrm{d}}}\:{\left({{\Delta\:}}_{\theta\:\mathrm{j}}\left({\mathrm{t}}_{{\mathrm{i}}_{\mathrm{k}}};{\Delta\:}\right)-{{{\Delta\:}}_{\theta\:}}_{\mathrm{j},\mathrm{k}}\right)}^{2}$$\end{document} The total training is a weighted combination of the physics, initial condition, and data losses. 38\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\mathcal{L}\left({\Delta\:}\right)={\mathcal{L}}_{\mathrm{I}\mathrm{C}}+{\mathcal{L}}_{\mathrm{O}\mathrm{D}\mathrm{E}}+{\mathcal{L}}_{\mathrm{data\:}}$$\end{document} Adam optimization algorithm PINN is trained by minimizing the total loss \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\mathcal{L}\left(\theta\:\right)$$\end{document} using the Adam optimizer with a learning rate of 0.001 over 10,000 epochs. The Adam updated rules given in Eqs. (40, 41) 39\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{m}_{t}={\beta\:}_{1}{m}_{t-1}+\left(1-{\beta\:}_{1}\right){\nabla\:}_{\theta\:}{\mathcal{L}}_{t}$$\end{document} 40\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{v}_{t}={\beta\:}_{2}{v}_{t-1}+\left(1-{\beta\:}_{2}\right){\left({\nabla\:}_{\theta\:}{\mathcal{L}}_{t}\right)}^{2}$$\end{document} Since \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{m}_{t}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{v}_{t}$$\end{document} are initialized as zero vectors, they are biased toward zero. The bias correction is applied. 41\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\stackrel{\prime }{m}}_{t}=\frac{{m}_{t}}{1-{\beta\:}_{1}^{t}}$$\end{document} 42\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\stackrel{\prime }{v}}_{t}=\frac{{v}_{t}}{1-{\beta\:}_{2}^{t}}$$\end{document} The network parameters \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\theta\:$$\end{document} are updated using the bias-corrected estimates given in Eq. (35) 43\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\theta _{t + 1}} = {\theta _t} - \eta \frac{{{{\hat m}_t}}}{{\sqrt {{{\hat v}_t} - \in } }}$$\end{document} The unknown solutions are approximated by a fully connected feed-forward neural network with trainable parameters Δ Fig. 3. 44\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\Delta\:}=\mathrm{a}\mathrm{r}\mathrm{g}\underset{{\Delta\:}}{\mathrm{m}\mathrm{i}\mathrm{n}}\:\mathcal{L}\left({\Delta\:}\right)$$\end{document} Fig. 3Framework of PINN-CF. Framework of PINN-CF. Numerical experiments This section introduces a number of benchmark Examples that would prove the accuracy, robustness and general applicability of the proposed PINN-CF framework. Example 1 Consider Eq. (1), the FF pantograph differential equation of the form. 45\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:}^{FF}{D}^{\alpha\:,\beta\:}{\Delta\:}\left(t\right)=-{\Delta\:}\left(0.99t\right)+0.9{\Delta\:}\left(t\right) subject to\:{\Delta\:}\left(0\right)=1$$\end{document} The convergence curve of the total loss for PINN-CF at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:\:=\:0.99$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:\:=\:0.9$$\end{document} shown in Fig. 4. The loss drops from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${10^ \circ }$$\end{document} to below\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${10^{ - 2}}$$\end{document} and continues to decrease steadily, reaching approximately \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:5\times{10^{ - 3}}$$\end{document} by 10,000 epochs. The smooth, monotonic decay by the absence of stagnation or oscillations demonstrates excellent optimization behavior of the Adam optimizer for this nearly classical regime, confirming that the physics-informed loss, exact initial condition, and L1-based Caputo Fabrizio approximation are well-balanced and provide effective even with typical training parameters. Fig. 4Total loss graph for various values of β=0.9 and α=0.99 for Example 1 of PINN-CF solution. Total loss graph for various values of β=0.9 and α=0.99 for Example 1 of PINN-CF solution. The comparison of PINN-CF and PINN methods in solving fractal-fractional pantograph delay differential equations and their corresponding residual errors given in Figs. 5 and 6. With fixed \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:\:=\:0.9$$\end{document}, decreasing α from 0.99 to 0.3strengthens memory, shifting the solution from fast classical decay to slow relaxation with a distinct overshoot at low. With \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:\:=\:0.9$$\end{document} fixed, the residual errors of PINN-CF remain below\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${10^{ - 3}}$$\end{document} even for the most memory-intensive case \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:\:=\:0.3$$\end{document}, and rapidly drop to\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${10^{ - 5}} - {10^{ - 6}}$$\end{document} as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:\:\mathrm{a}\mathrm{p}\mathrm{r}\mathrm{o}\mathrm{a}\mathrm{c}\mathrm{h}\mathrm{e}\mathrm{s}\:\mathrm{t}\mathrm{o}\:1$$\end{document}, while for same conditions the residual errors of PINN drop up to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{10}^{-2}.\:$$\end{document}The residual errors comparison demonstrates the high accuracy of PINN-CF across the entire fractal-fractional parameter space as shown in Figs. 5 and 6. Fig. 5PINN-CF solution and their corresponding residual errors for various values of α and β=0.9 of Example 1. PINN-CF solution and their corresponding residual errors for various values of α and β=0.9 of Example 1. Fig. 6PINN solution and their corresponding residual errors for various values of α and β=0.9 of Example 1. PINN solution and their corresponding residual errors for various values of α and β=0.9 of Example 1. With fixed \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:\:=\:0.9$$\end{document}, lowering \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\upbeta\:}$$\end{document} from 0.99 to 0.3 accelerates decay monotonically due to increased fractal irregularity, without inducing oscillations, see Fig. 7. In the case of fixing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:\:=\:0.9$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:$$\end{document} varies, errors stay uniformly low \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:(mostly\:{10^{ - 4}} - {10^{ - 6}})$$\end{document} over the changes in fractal dimension, with only small oscillations near \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:t\:=\:0$$\end{document} when the solution is steeper there, while the residuals of PINN are varying about \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{10}^{-3}$$\end{document}. The residual errors of PINN-CF are lower than PINN method which shows the effectiveness of PINN-CF. Across both parameters setting, the solutions of PINN-CF remain smooth and physically consistent, proving PINN-CF remarkable ability to reliably resolve diverse memory and fractal regimes using a single, robust neural network framework. The fact that the strongly anomalous regimes (low \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:$$\end{document} or low \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:$$\end{document}) have no systematic error growth or blow-up in proves that the physics-informed loss, combined with enforcing the exact initial-condition and L1-based Caputo–Fabrizio approximation, effectively constrains the neural network and provide highly faithful solutions over the entire domain for all tested parameter values. Fig. 7PINN solution and their corresponding residual errors for various values of β and α=0.9 of Example 1. PINN solution and their corresponding residual errors for various values of β and α=0.9 of Example 1. The accuracy of PINN-CF for Example 1 across the whole fractal-dimension spectrum given in Table 1. At \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:\:=\:0.9$$\end{document} and two representative fractal dimensions, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:\:=\:0.3$$\end{document} (strong fractal irregularity) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:\:=\:0.9$$\end{document} (near-classical behavior), across \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:t\:\in\:\:[0.1,\:1.0]$$\end{document}. In both cases the agreement is good, for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:\:=\:0.9$$\end{document}, residual errors remain consistently less than \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:3.51\: \times \:\:{10^{ - 4}},$$\end{document} and even with the more challenging fractal regime (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:\:=\:0.3$$\end{document}), the errors are less than \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:5.00\: \times \:\:{10^{ - 2}}$$\end{document} at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:t\:=\:0.1$$\end{document} and rapidly decrease to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:2.46\: \times \:\:{10^{ - 4}}\:$$\end{document}at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:t\:=\:1.0$$\end{document}. This shows that PINN-CF maintains high fidelity across the entire domain and over a broader spectrum of fractal dimensions, with errors decreasing systematically as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:$$\end{document} approaches to 1, confirming the accuracy and reliability of the PINN-CF framework even when strong fractal and memory effects. The PINN solutions and their corresponding residual errors for various values of β, with α = 0.9 for Example 1, are presented in Fig. 8. Table 1PINN-CF solution and residual errors for α =0.9 and various β for Example 1.β=0.3β=0.9tPINN-CFResidual errorsPINN-CFResidual errors0.10.9979185.000001e-020.9965965.000001e-020.20.9951991.016140e-030.9947783.757417e-030.40.9906071.243448e-030.9940111.921773e-040.60.9873421.260638e-040.9941582.646875e-040.80.9841716.092072e-040.9937322.569556e-0410.9807292.461731e-040.9937193.508925e-04 Table 1PINN-CF solution and residual errors for α =0.9 and various β for Example 1.β=0.3β=0.9tPINN-CFResidual errorsPINN-CFResidual errors0.10.9979185.000001e-020.9965965.000001e-020.20.9951991.016140e-030.9947783.757417e-030.40.9906071.243448e-030.9940111.921773e-040.60.9873421.260638e-040.9941582.646875e-040.80.9841716.092072e-040.9937322.569556e-0410.9807292.461731e-040.9937193.508925e-04 PINN-CF solution and residual errors for α =0.9 and various β for Example 1. Example 2 Consider Eq. (1), the fractal fractional pantograph differential equation. 46\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:}^{FF}{D}^{\alpha\:,\beta\:}{\Delta\:}\left(t\right)=-{\Delta\:}\left(0.5t\right) sunject to \:{\Delta\:}\left(0\right)=1$$\end{document} The total loss convergence for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:(\alpha\:=0.80,\:\beta\:=0.50)$$\end{document} as shown in Fig. 9. Although there are stronger memory and fractal effects, the loss still decreases from 10⁰ to below \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${10^{ - 3}}$$\end{document} within the first 2000 epochs. It then gradually reduces to 10⁻⁴ and has moderate frequency oscillations which are common for stiff, non-local problems using Adam optimization. Note that there is no divergence in this intermediate anomalous regime, which also shows that PINN-CF has stable and efficient training dynamics. This high convergence rate demonstrates how PINN-CF can handle non-classical fractal-fractional operators without the need for adaptive weighting distributions or specialized optimizers. Fig. 8PINN solution and their corresponding residual errors for various values of β and α=0.9 of Example 1. PINN solution and their corresponding residual errors for various values of β and α=0.9 of Example 1. With \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:\:=\:0.8\:$$\end{document}fixed, PINN-CF’s performance in solving Example 2 and the corresponding residuals errors, shown in Figures 10, greatly speeds up the growth process by reducing α from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:0.99\:to\:0.5$$\end{document}. By increasing fractal irregularity, lowering β from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:0.99\:to\:0.5$$\end{document} produces the same accelerating effect with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:\:=\:0.8$$\end{document} fixed. Excellent sensitivity to both parameters is confirmed by the fact that the solution curves are consistently perfectly smooth and clearly separated. While Fig. 11 presents the PINN solution and the corresponding errors for the same parameters setting. The residuals errors comparison of PINN and PINN-CF shows that PINN-CF method is highly accurate than PINN method. Fig. 9Total loss graph for Example 2 of PINN-CF solution. Total loss graph for Example 2 of PINN-CF solution. Fig. 10PINNs solution and their corresponding residual errors for various and α and β=0.8 of Example 2. PINNs solution and their corresponding residual errors for various and α and β=0.8 of Example 2. The PINN-CF solution and the corresponding errors presented by Fig. 12, while that of PINN method is given in Fig. 13. The residual error plots of PINN-CF are remarkably low \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:(mostly\:{10^{ - 3}} - {10^{ - 6}})$$\end{document} across the entire domain and for all tested parameters values, even in the area where the solution gradient is steepest near\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\:t\:=\:0\:\mathrm{a}\mathrm{n}\mathrm{d}\:t\:=1$$\end{document}. Only in the vicinity of boundaries and under extreme parameter settings do minor oscillations occur, and they never go above \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${10^{ - 2}}$$\end{document}, while the residual errors of PINN \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${10^{ - 3}} - {10^{ - 6}}$$\end{document}Figures 12 and 13. Fig. 11PINN-CF solution and their corresponding residual errors for various α and β=0.8 of Example 2. PINN-CF solution and their corresponding residual errors for various α and β=0.8 of Example 2. Fig. 12PINNs solution and their corresponding residual errors for various values of β and α=0.8 of Example 2. PINNs solution and their corresponding residual errors for various values of β and α=0.8 of Example 2. The effectiveness of PINN-CF for Example 2 over the whole fractal dimension. Remaining residual errors are remarkably low for the strong fractal case \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:=0.3$$\end{document}, beginning at just \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$5.0\: \times \:\:{10^{ - 2}}$$\end{document} at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:t=0.1\:$$\end{document}and falling to\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${10^{ - 3}}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:2.46\times\:10^{ - 2}\:$$\end{document}at t=1. Errors are consistently less than\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\:3.8\times\:10^{ - 3}\:$$\end{document}and drop to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:10^{ - 1}\:$$\end{document}for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:=0.9$$\end{document}, demonstrating that PINN-CF maintains high fidelity even as fractal irregularity increases. The solution’s smooth convergence to the integer-order limit and the systematic reductions in errors as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:$$\end{document} increases demonstrate PINN-CF’s remarkable capacity to handle severe self-similar anomalies without sacrificing accuracy. The PINN-CF solution and corresponding residual errors for α=0.9 in Example 2 are shown in Table 2. Table 2PINN-CF solution and Residual Errors (RE) for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:\:=0.9$$\end{document} and various \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\upbeta\:}$$\end{document} of Example 2.t \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:=0.3$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:=0.99$$\end{document} PINN-CFResidual errorsPINN-CFResidual errors0.1251.0616341.03e-021.1057645088.29E-030.2501.0965581.99e-021.2121026526.41e-040.3751.1230767.00e-031.3149394992.85e-030.5001.1461373.09e-031.4124281413.12e-040.6251.1664065.34e-031.5044215921.39e-030.7501.1847912.91e-031.5914838319.11e-040.8751.2022274.32e-041.6741727593.59e-0411.2192446.08e-041.7528005846.43e-04 Table 2PINN-CF solution and Residual Errors (RE) for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:\:=0.9$$\end{document} and various \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\upbeta\:}$$\end{document} of Example 2.t \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:=0.3$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:=0.99$$\end{document} PINN-CFResidual errorsPINN-CFResidual errors0.1251.0616341.03e-021.1057645088.29E-030.2501.0965581.99e-021.2121026526.41e-040.3751.1230767.00e-031.3149394992.85e-030.5001.1461373.09e-031.4124281413.12e-040.6251.1664065.34e-031.5044215921.39e-030.7501.1847912.91e-031.5914838319.11e-040.8751.2022274.32e-041.6741727593.59e-0411.2192446.08e-041.7528005846.43e-04 PINN-CF solution and Residual Errors (RE) for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:\:=0.9$$\end{document} and various \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\upbeta\:}$$\end{document} of Example 2. Example 3 Consider Eq. (7), the following FF memristor system. 47\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left.\begin{array}{rr}&\:{\:}^{FF}{D}^{\alpha\:,\beta\:}{{\Delta\:}}_{1}\left(t\right)-2\left(0.04{{\Delta\:}}_{4}\left(t\right){{\Delta\:}}_{1}\left(t\right)+0.002{{\Delta\:}}_{1}\left(t\right)+0.2{{\Delta\:}}_{1}\left(t\right)-{{\Delta\:}}_{2}\left(t\right)\right)=0\\\:&\:{\:}^{FF}{D}^{\alpha\:,\beta\:}{{\Delta\:}}_{2}\left(t\right)+{{\Delta\:}}_{1}\left(t\right)-{{\Delta\:}}_{3}\left(t\right)=0\\\:&\:{\:}^{FF}{D}^{\alpha\:,\beta\:}{{\Delta\:}}_{3}\left(t\right)+{{\Delta\:}}_{2}\left(t\right)-0.92{{\Delta\:}}_{3}\left(t\right)=0\\\:&\:{\:}^{FF}{D}^{\alpha\:,\beta\:}{{\Delta\:}}_{4}\left(t\right)-{{\Delta\:}}_{1}\left(t\right)=0\end{array}\right\}$$\end{document} with the initial conditions \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{{\Delta\:}}_{1}\left(0\right)={{\Delta\:}}_{2}\left(0\right)={{\Delta\:}}_{3}\left(0\right)={{\Delta\:}}_{4}\left(0\right)=0.01$$\end{document} The total loss convergence for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:\:=\:0.99$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:\:=\:0.98$$\end{document}, a case close to the integer-order limit as shown in Fig. 14. The loss decreases from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${10^ \circ }$$\end{document} to below \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:10^{ - 3}\:$$\end{document}within the first 1000 epochs and continues to decrease steadily to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:5\times\:10^{ - 4}$$\end{document} by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\mathrm{10,000}$$\end{document} epochs with only minor, high-frequency oscillations. This rapid, smooth training behavior validates that PINN-CF performs remarkably well when the fractal-fractional operator gets closer to the classical case. It also serves as a great benchmark and shows that the framework adds no artificial complexity even when memory and fractal effects are negligible. PINN-CF’s efficacy for such near-standard differential equations is validated by the attainment of low loss values. Fig. 13PINNs solution and their corresponding residual errors for various values of β and α=0.8 Example 2. PINNs solution and their corresponding residual errors for various values of β and α=0.8 Example 2. For the memristor system (Example 3) the performance of PINN-CF, with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:\:=\:0.98$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:\:=\:0.99$$\end{document} shown in Figs. 15 and 16. Across the entire domain all four state variables \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\Delta 1\left( t \right),\:\Delta 2\left( t \right),\:\Delta \:3\left( t \right),\:$$\end{document}and\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\Delta \:4\left( t \right)$$\end{document} exhibit physically consistent and accurate behavior. Conversely, with a fixed \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:$$\end{document} near 1 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:$$\end{document} varies, the fractal dimension effect is milder but consistent at lower \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:$$\end{document} accelerates relaxation in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta 1,\Delta 3$$\end{document} and marginally enhances the growth rate of Δ 4. In both sweeps, all the solution curves stay smooth, free of artificial oscillations, and show excellent separation and monotonic reliance on the varied parameter. Fig. 14Total loss graph of PINN-CF solution for Example 3. Total loss graph of PINN-CF solution for Example 3. Fig. 15PINN-CF solution for various values of α and β=0.98 Example 3. PINN-CF solution for various values of α and β=0.98 Example 3. The results of the residual error plots in the Example of 3 (memristor system) indicate the accuracy of PINN-CF in the nonlinear, multi-dimensional environment clearly shown in Figs. 17 and 18. For a fixed \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:=0.98\:$$\end{document}and varied \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:$$\end{document}, residuals of all components remain below \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$10^{ - 2}$$\end{document}and decreasing to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${10^{ - 4}} - {10^{ - 6}}$$\end{document} shortly after the initial transient, even the strongest memory (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:=0.5$$\end{document}). while for a fixed \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:\:=\:0.99$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:$$\end{document} varying, errors are even lower, typically \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${10^{ - 3}} - {10^{ - 5}},$$\end{document} with only small peaks around boundaries. The PINN solution profile and the corresponding residuals errors for a fixed \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:=0.98$$\end{document} and varying \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:$$\end{document} given in Fig. 19. And for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:$$\end{document}=0.99 and various value of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:$$\end{document} the solution profile of PINN and the corresponding residual erros given in Fig. 20. Fig. 16PINN-CF solution for various values of β and α=0.99 Example 3. PINN-CF solution for various values of β and α=0.99 Example 3. Fig. 17Residual errors for each at various values of α and β=0.98 of Example 3. Residual errors for each at various values of α and β=0.98 of Example 3. The superior accuracy and robustness of the suggested PINN-CF framework are clearly demonstrated by a thorough analysis of the PINN-CF solutions for the memristor system at α = 0.99 with β = 0.5 and β = 0.99. The PINN-CF solution Δ₄(t) consistently outperforms the three baseline CFA-based numerical schemes (Δ₁(t), Δ₂(t), Δ₃(t)) over the whole-time domain [0, 100] for both fractional orders, as Table 1 illustrates. The PINN-CF achieves residual errors on the order of 10⁻² to 10⁻³ at β = 0.5, which are significantly lower than Δ₃ and Δ₄ of the classical schemes and up to two orders of magnitude smaller than the best conventional CFA method Δ₁. At β = 0.99, where the memory effect is strongest, the PINN-CF residual errors stay remarkably low (6.2 × 10⁻¹ to 8.8 × 10⁻³). Consequently, the PINN-CF framework is firmly established as the best method for resolving very challenging fractal fractional differential equations, such as the pantograph equation. It exhibits exceptional stability and convergence behavior even when the fractional orders get closer to the classical integer limit, in addition to producing noticeably higher accuracy and smaller residual errors as compared to PINN across all entire domain, Refer to Tables 3 and 4. Table 3PINN-CF solution and Residual errors (RE) for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:\:=0.99$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:=0.5$$\end{document} of Example 3.tPINN-CFSolutionResidualsErrors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{1}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{2}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{3}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{4}\left(\mathbf{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{1}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{2}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{3}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{4}\left(\mathbf{t}\right)$$\end{document} 100.009310.0105810.009360.009831.746514e-025.847216e-051.764489e-031.859166e-03200.001050.0012930.000860.069312.982304e-032.142547e-042.065599e-036.440556e-04400.000870.0009320.000470.100652.189989e-034.084096e-041.355734e-036.831259e-04600.000920.0007810.000360.110141.909529e-035.673291e-041.105787e-037.944264e-04800.000890.0006580.000320.115021.652903e-035.813973e-049.426394e-047.822722e-041000.000720.0005570.000220.119291.384571e-035.031637e-047.548037e-046.187017e-04 Table 3PINN-CF solution and Residual errors (RE) for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:\:=0.99$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:=0.5$$\end{document} of Example 3.tPINN-CFSolutionResidualsErrors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{1}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{2}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{3}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{4}\left(\mathbf{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{1}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{2}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{3}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{4}\left(\mathbf{t}\right)$$\end{document} 100.009310.0105810.009360.009831.746514e-025.847216e-051.764489e-031.859166e-03200.001050.0012930.000860.069312.982304e-032.142547e-042.065599e-036.440556e-04400.000870.0009320.000470.100652.189989e-034.084096e-041.355734e-036.831259e-04600.000920.0007810.000360.110141.909529e-035.673291e-041.105787e-037.944264e-04800.000890.0006580.000320.115021.652903e-035.813973e-049.426394e-047.822722e-041000.000720.0005570.000220.119291.384571e-035.031637e-047.548037e-046.187017e-04 PINN-CF solution and Residual errors (RE) for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:\:=0.99$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:=0.5$$\end{document} of Example 3. Table 4PINN-CF solution and Residual errors (RE) for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:\:=0.99$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:=0.99$$\end{document} Of Example 3.tPINN-CFSolutionResidualsErrors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{1}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{2}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{3}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{4}\left(\mathbf{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{1}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{2}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{3}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{4}\left(\mathbf{t}\right)$$\end{document} 100.008840.009670.0107050.0093921.583820e-021.865035e-031.760664e-048.840238e-03200.002570.000850.0018810.0185638.507559e-044.591795e-032.709241e-032.608255e-03400.002640.000890.0018070.0167648.701375e-044.590541e-032.662794e-032.657101e-03600.002610.000880.0018540.0153668.752162e-044.599825e-032.702078e-032.638002e-03800.002650.000870.0018170.0146398.352284e-044.604310e-032.658344e-032.694651e-031000.002690.000860.0017620.0142918.026509e-044.589601e-032.600110e-032.742734e-03 Table 4PINN-CF solution and Residual errors (RE) for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:\:=0.99$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:=0.99$$\end{document} Of Example 3.tPINN-CFSolutionResidualsErrors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{1}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{2}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{3}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{4}\left(\mathbf{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{1}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{2}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{3}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{4}\left(\mathbf{t}\right)$$\end{document} 100.008840.009670.0107050.0093921.583820e-021.865035e-031.760664e-048.840238e-03200.002570.000850.0018810.0185638.507559e-044.591795e-032.709241e-032.608255e-03400.002640.000890.0018070.0167648.701375e-044.590541e-032.662794e-032.657101e-03600.002610.000880.0018540.0153668.752162e-044.599825e-032.702078e-032.638002e-03800.002650.000870.0018170.0146398.352284e-044.604310e-032.658344e-032.694651e-031000.002690.000860.0017620.0142918.026509e-044.589601e-032.600110e-032.742734e-03 PINN-CF solution and Residual errors (RE) for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:\:=0.99$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:=0.99$$\end{document} Of Example 3. Example 4 Consider Eq. (15) the following fractal fractional convective fluid motion in a rotating cavity system as. 48\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{*{20}{r}} {}&{^{FF}{D^{\alpha \:,\beta \:}}\Delta {\:_1}\left( t \right) + 0.005\Delta {\:_2}\left( t \right)\Delta {\:_3}\left( t \right) + 4\left( {\Delta {\:_1}\left( t \right) - \Delta {\:_2}\left( t \right)} \right) = 0,}\\ \:&{\:{\:^{FF}}{D^{\alpha \:,\beta \:}}\Delta {\:_2}\left( t \right) - 687\Delta {\:_1}\left( t \right) + \Delta {\:_2}\left( t \right) + \Delta {\:_1}\left( t \right)\Delta {\:_3}\left( t \right) = 0,}\\ \:&{\:{\:^{FF}}{D^{\alpha \:,\beta \:}}\Delta {\:_3}\left( t \right) - \Delta {\:_1}\left( t \right)\Delta {\:_2}\left( t \right) + \Delta {\:_3}\left( t \right) = 0,} \end{array}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\Delta\:}}_{1}\left(0\right)=0.1,\:{{\Delta\:}}_{2}\left(0\right)=0.2,{{\Delta\:}}_{3}\left(0\right)=0.3$$\end{document} The total loss convergence for Example 4 (convective flow system) at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:\:=\:0.99,\:\beta\:\:=\:0.98$$\end{document}, a near-classical case given in Fig. 21. Within the first 1000 epochs the loss drops sharply from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$10^{1}$$\end{document} to below \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$10^{ -1}$$\end{document}, and by 10,000 epochs, it has dropped to 5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\times\:$$\end{document}10⁻³ by 10,000 epochs, accompanied by the moderately high-frequency oscillations that are typical of Adam in multi-equation systems. The three coupled nonlinear fractal-fractional equations are more complex, but training is still quick and steady, with neither stagnation nor divergence. The robustness of the PINN-CF architecture, loss design, and optimization strategy when applied to large-scale fractal-fractional dynamical systems is further validated by this smooth and effective convergence in a high-dimensional, near-integer-order setting. Fig. 18Residual errors for each Δ(t) various values of β and α=0.99 Example 3. Residual errors for each Δ(t) various values of β and α=0.99 Example 3. The solution trajectories of the (Example 4) widely illustrate PINN-CF remarkable physical fidelity and physical consistency. When β = 0.98 is kept close to classical and the fractional order α is decreased from 0.99 to 0.5 initial decay rates slow, overshoots in Δ₂(t) and Δ₃(t) become more noticeable. On the other hand, while maintaining smooth long-term convergence to equilibrium, fixing α = 0.99 and reducing the fractal dimension β from 0.99 to 0.5 speeds up transients, sharpens early-time gradients, and amplifies short-scale fluctuations, precisely capturing enhanced self-similar irregularity. The system shows increasingly noticeable memory-induced damping when β = 0.98 is kept close to classical and the fractional order α is decreased from 0.99 to 0.5. On the other hand, while maintaining smooth long-term convergence to equilibrium, fixing α = 0.99 and reducing the fractal dimension β from 0.99 to 0.5 speeds up transients, sharpens early-time gradients, and amplifies short-scale fluctuations, precisely capturing enhanced self-similar irregularity. PINN-CF is without any doubt a highly powerful and reliable tool to the complicated fractal-fractional dynamical systems, as seen in this smooth and high-resolution re-creation of delicate memory and fractal dynamics in the stiff and coupled nonlinear systems, (see Figs. 22 and 23). Residual errors for Example 4 at various values of α, with β = 0.98, are presented in Fig. 24. On the other hand, the residual errors for various values of β, with α = 0.99, for Example 4 are presented in Fig. 25. Fig. 19PINN solution and Residual errors for each Δ(t) various values of α and β=0.98 Example 3. PINN solution and Residual errors for each Δ(t) various values of α and β=0.98 Example 3. Fig. 20Residual errors for each Δ(t) various values of β and α=0.99 Example 3. Residual errors for each Δ(t) various values of β and α=0.99 Example 3. The residual error histories for the highly nonlinear fractal-fractional Lorenz-type convective fluid system significantly highlight the remarkable resilience of PINN-CF against extreme chaos and stiffness. All three displayed residual components decreased from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O10^{ - 2}$$\end{document} to below \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$10^{ - 4}-10^{ - 5}$$\end{document} within the first few units and remained flat thereafter, even with β fixed at 0.98and α reduced to 0.5. Meanwhile, the hidden fourth residual remained below \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$10^{ - 4}$$\end{document} throughout, see Figs. 22 and 23,Figs. 24 and 25. The PINN solution profile and the corresponding residuals errors for a fixed \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:=0.98$$\end{document} and varying \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:$$\end{document} given in Fig. 26. And for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:$$\end{document}=0.99 and various value of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:$$\end{document} the solution profile of PINN and the corresponding residual erros given in Fig. 27. Fig. 21Total loss graph of PINN-CF solution for Example 4. Total loss graph of PINN-CF solution for Example 4. Fig. 22PINN-CF solution at various values of α and β=0.98 of Example 4. PINN-CF solution at various values of α and β=0.98 of Example 4. Fig. 23PINN-CF solution at various values of β and α=0.99 of Example 4. PINN-CF solution at various values of β and α=0.99 of Example 4. Fig. 24Residual errors at various values of α and β=0.98 of Example 4. Residual errors at various values of α and β=0.98 of Example 4. Fig. 25Residual errors for various values of β and α=0.99 of Example 4. Residual errors for various values of β and α=0.99 of Example 4. Fig. 26PINN solution and Residual errors for each Δ(t) various values of α and β=0.98 Example 4. PINN solution and Residual errors for each Δ(t) various values of α and β=0.98 Example 4. Fig. 27PINN solution and Residual errors for each β and α=0.99 of Example 4. PINN solution and Residual errors for each β and α=0.99 of Example 4. The PINN-CF solutions alongside their residual errors (RE) for Example 4 at selected time points under two fractional-order regimes with fixed \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:\:=\:0.99$$\end{document} but markedly different \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:$$\end{document} values. In Table 5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:(\beta\:\:=\:0.5)$$\end{document}, the solutions show \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta _2}\left( t \right)$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta _3}\left( t \right)$$\end{document} decaying rapidly from their initial values (0.187 and 0.300) to very small long-time levels (on the order of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$10^{ - 3}$$\end{document} or lower by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:t\:=\:100$$\end{document}), while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta _1}\left( t \right)$$\end{document} remains extremely small throughout \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(10^{ - 6}}\to{10^{ - 6}})$$\end{document}. However, the residuals reveal relatively poor satisfaction of the governing equations at early times, especially for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta _2}\left( t \right)$$\end{document}. In contrast, Table 6 (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:\:=\:0.99$$\end{document}, very close to the classical case) exhibits much smaller residuals overall, particularly for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta _3}\left( t \right)$$\end{document} (dropping to 10⁻⁵ or better by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:t\:=\:100$$\end{document}) and improved convergence for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta _1}\left( t \right)$$\end{document}, although \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta _2}\left( t \right)$$\end{document} residuals stay somewhat elevated ( 0.4-0.8 range) compared to the other components. The PINN-CF demonstrates how the network was successfully controlled by the physics-informed loss term. When taken as a whole, these findings demonstrate the accuracy of PINN-CF for stiff fractal-fractional dynamics, obtaining consistent parametric behavior and low residuals without the need for special tuning. The nomenclature of the constants used in the mathematical formulation is presented in Table 7, while the parameter settings employed in the PINN-CF methodology are provided in Table 8. The design weights generated by the PINN-CF methodology for the examples considered in this manuscript are presented in Appendix Table 9. Table 5PINN-CF solution and Residual errors (RE) for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:\:=0.99$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\upbeta\:}=0.5$$\end{document} Of Example 4. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\boldsymbol{t}$$\end{document} PINN-CFFASolutionResidualsErrors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{1}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{2}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{3}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{1}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{2}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{3}\left(\boldsymbol{t}\right)$$\end{document} 100.010730.1862990.2990837.226169e-012.361234e-012.685899e-01200.0000090.0021490.0028808.213558e-036.480302e-022.378731e-03400.0000560.0002840.0006248.319174e-045.271837e-029.532642e-04600.0000690.0001680.0000239.990388e-045.714583e-022.400263e-04800.0000480.0004320.0012822.087735e-036.314448e-021.055136e-031000.0000070.0007190.0025403.321934e-037.789300e-022.338084e-03 Table 5PINN-CF solution and Residual errors (RE) for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:\:=0.99$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\upbeta\:}=0.5$$\end{document} Of Example 4. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\boldsymbol{t}$$\end{document} PINN-CFFASolutionResidualsErrors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{1}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{2}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{3}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{1}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{2}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{3}\left(\boldsymbol{t}\right)$$\end{document} 100.010730.1862990.2990837.226169e-012.361234e-012.685899e-01200.0000090.0021490.0028808.213558e-036.480302e-022.378731e-03400.0000560.0002840.0006248.319174e-045.271837e-029.532642e-04600.0000690.0001680.0000239.990388e-045.714583e-022.400263e-04800.0000480.0004320.0012822.087735e-036.314448e-021.055136e-031000.0000070.0007190.0025403.321934e-037.789300e-022.338084e-03 PINN-CF solution and Residual errors (RE) for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:\:=0.99$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\upbeta\:}=0.5$$\end{document} Of Example 4. Table 6PINN-CF solution and Residual errors (RE) for α = 0.99 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\upbeta\:}=0.99$$\end{document} Of Example 4. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\boldsymbol{t}$$\end{document} PINN-CFSolutionResidualsErrors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{1}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{2}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{3}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{1}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{2}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{3}\left(\boldsymbol{t}\right)$$\end{document} 100.000860.1846240.3003437.416378e-017.738128e-013.005015e-01200.000660.0016400.0043113.875051e-034.502932e-016.954633e-04400.000610.0001120.0049752.875433e-034.156921e-011.028861e-03600.000620.0010820.0041526.821343e-034.274014e-012.390753e-04800.000650.0013400.0038617.947283e-034.436848e-012.647038e-051000.000660.0014100.0037768.262754e-034.496230e-019.876420e-05 Table 6PINN-CF solution and Residual errors (RE) for α = 0.99 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\upbeta\:}=0.99$$\end{document} Of Example 4. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\boldsymbol{t}$$\end{document} PINN-CFSolutionResidualsErrors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{1}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{2}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{3}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{1}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{2}\left(\boldsymbol{t}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\Delta\:}}_{3}\left(\boldsymbol{t}\right)$$\end{document} 100.000860.1846240.3003437.416378e-017.738128e-013.005015e-01200.000660.0016400.0043113.875051e-034.502932e-016.954633e-04400.000610.0001120.0049752.875433e-034.156921e-011.028861e-03600.000620.0010820.0041526.821343e-034.274014e-012.390753e-04800.000650.0013400.0038617.947283e-034.436848e-012.647038e-051000.000660.0014100.0037768.262754e-034.496230e-019.876420e-05 PINN-CF solution and Residual errors (RE) for α = 0.99 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\upbeta\:}=0.99$$\end{document} Of Example 4. Table 7Nomenclature.ParametersValue(s) \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:$$\end{document} Fractional order of the derivative (controls the degree of non-locality/memory) ranging from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:0\:<\:\alpha\:\:\le\:\:1$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:$$\end{document} Fractal dimension parameter fractal scaling exponent, ranging from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:0\:<\:\beta\:\le\:\:1$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\widehat{\gamma\:}$$\end{document} Stretch factor or proportional delay coefficient (pantograph parameter) \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:0\:<\:\widehat{\gamma\:}\:<\:$$\end{document}1(ensures delay effect; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\widehat{\gamma\:}=1$$\end{document} removes pantograph nature) \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\varDelta\:$$\end{document} Dependent variable/unknown function t Time domain a,b Real constant coefficients (control linear combination of current and delayed values) Table 7Nomenclature.ParametersValue(s) \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:$$\end{document} Fractional order of the derivative (controls the degree of non-locality/memory) ranging from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:0\:<\:\alpha\:\:\le\:\:1$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:$$\end{document} Fractal dimension parameter fractal scaling exponent, ranging from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:0\:<\:\beta\:\le\:\:1$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\widehat{\gamma\:}$$\end{document} Stretch factor or proportional delay coefficient (pantograph parameter) \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:0\:<\:\widehat{\gamma\:}\:<\:$$\end{document}1(ensures delay effect; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\widehat{\gamma\:}=1$$\end{document} removes pantograph nature) \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\varDelta\:$$\end{document} Dependent variable/unknown function t Time domain a,b Real constant coefficients (control linear combination of current and delayed values) Nomenclature. Stretch factor or proportional delay coefficient (pantograph parameter) \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:0\:<\:\widehat{\gamma\:}\:<\:$$\end{document}1 (ensures delay effect; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\widehat{\gamma\:}=1$$\end{document} removes pantograph nature) Table 8Parameters used in this work.ParametersValue(s)Input1Number of approximate points N100Hidden layers50Activation function tanh OptimizerAdamepochs10,000output1 Table 8Parameters used in this work.ParametersValue(s)Input1Number of approximate points N100Hidden layers50Activation function tanh OptimizerAdamepochs10,000output1 Parameters used in this work. Conclusion This manuscript presented a new Physics-Informed Neural Network architecture, PINN-CF, which uses Caputo-Fabrizio fractal-fractional derivatives to solve nonlinear fractal-fractional equations. Numerical experimentation on four benchmark problems, pantograph-type delay equations, memristor-based neural circuits, and convective fluid motion in rotating cavities showed the proposed methodology to have an exceptional robustness and accuracy in the fractal-fractional parameter space, with residual errors in the standard regimes of the order of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$10^{ - 4}$$\end{document} and much lower in the regimes where parameters were approaching classical integer-order behavior\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\:(\alpha\:,\:\beta\:\:\mathrm{a}\mathrm{p}\mathrm{p}\mathrm{r}\mathrm{o}\mathrm{a}\mathrm{c}\mathrm{h}\mathrm{e}\mathrm{s}\:1)$$\end{document}. Hence the PINN-CF maintained the accurate solution approximation even in strongly anomalous regimes \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left(\alpha\:,\:\beta\:\:\approx\:\:0.3,\:0.5\right)$$\end{document}. By integrating L1-discretized Caputo-Fabrizio operators with automatic differentiation the framework precisely captures non-local memory phenomena without explicit computations. While the CF operator is also tested with the PINN solution using traditionally Caputo fractional operator. The comparison of both the operators shows that the CF operator works better than Caputo fractional operator, Parametric sensitivity studies indicate that reduction in fractional orders in systematic improvements of memory driven damping and delay effects, it would reasonably predict physically consistent behavior in all the cases studied. Also, PINN-CF has shown computational performance, and it is highly scalable to larger problem classes, including multi-delay systems, distributed-order fractional operators, and stochastic differential equations, making it an effective and general computational tool developed to study anomalous dynamical systems with possible applications in epidemiologic modeling, neuromorphic computing, and geophysical flow analysis. Limitations and future directions Although there are some limitations associated with the PINN-CF framework that should be taken into consideration and further research. Its methodology demonstrates less accuracy at very low fractional orders, at which a very strong effect of the memory capability of the network poses a challenge to its approximation capability. The sensitivity on hyperparameters must be carefully tuned to keep them unreachable by non-specialist users, and the sensitivity of training in highly chaotic regimes sometimes calls adaptive optimization methods. Also, the cost of computation of long time or high dimensional problems is now greater than optimized classical methods. On this basis, a number of real-world research opportunities are proposed, such as adaptive collocation sampling and domain decomposition to scale to larger problem sizes, Bayesian neural networks to quantify uncertainties in inverse problems, hybrid algorithm employing PINN-CF and spectral or wavelet analysis to scale effects, rigorous error analysis to establish convergence rates and stability limits, domain-specific enabling epidemiology, neuromorphic engineering, and geophysical fluid dynamics. The response to these directions will gradually turn PINN-CF into a highly capable proof-of-concept into a fully grown computational platform to study next generation research of fractional dynamical systems. Appendix Table 9Neurons/weights for Example 1 and 2.WeightsExample 1Example 2S.NOΦjwjγjΦjwjγj10.028053−0.717670.152813−0.0806−1.084370.2756662−0.108530.3911310.041023−0.02682−0.773820.4771583−0.10992−0.5969−0.73395−0.14203−0.750610.65375940.112599−0.675230.429002−0.14020.2844410.8280945−0.134850.868873−0.646280.021106−0.01229−0.32646−0.07396−0.69394−0.153450.0973911.140424−0.0151670.001866−0.928830.342457−0.08558−0.44375−0.899888−0.04509−1.02622−0.75901−0.052360.2820860.86302590.0470960.1399240.614528−0.00104−0.20076−0.8157310−0.10384−0.67060.3468110.0512881.102195−0.1829111−0.10756−0.229080.6737050.000588−0.909180.10096120.0745660.376885−0.244280.0021290.339112−0.72684130.0018910.146690.3642130.088020.9216350.430377140.070815−0.624640.055019−0.04672−0.93534−0.0599415−0.13538−0.37624−0.396490.1034−1.35018−0.10702160.034652−0.225140.05733−0.095010.9168860.122526170.092315−0.252450.4020060.082134−0.79287−1.0456518−0.08332−0.113790.920643−0.10671−1.00343−0.0851619−0.12242−0.142230.712165−0.077220.619515−0.18713200.0690.383267−0.45006−0.037940.1188820.80585721−0.08013−0.65748−0.85584−0.155250.677478−0.6406422−0.066990.179206−0.362410.036199−0.22279−0.5500723−0.00926−0.409930.4133580.128462−0.70610.3271924−0.147380.6540540.8844420.0145191.1471490.09168425−0.072490.339125−0.96332−0.035780.260314−0.10952260.009538−0.85029−0.499010.029934−0.945480.75560127−0.081480.904719−0.38681−0.087340.2998660.274902280.0039690.5328360.574269−0.3019−0.9292−0.06303290.0953130.6973380.465553−0.07879−0.374020.75389330−0.09733−0.366260.067302−0.15016−0.385760.63057631−0.08106−0.25051−0.73586−0.09860.818595−0.3674332−0.09473−0.5254−0.478150.0285810.0580320.730424330.0201030.5509270.931746−0.0823−0.11696−0.2862340.128766−0.028380.5743580.151455−0.98204−0.01481350.016736−0.624520.1978210.025902−1.15114−0.2022236−0.10566−0.27703−0.57805−0.141950.9165960.04409737−0.09463−0.52059−0.959830.127858−0.12304−0.92612380.076331−0.732720.6419690.129936−0.893550.425223390.1261010.203775−0.55530.1673770.4009240.39990440−0.13954−0.972580.6191550.167619−0.900350.9121341−0.080960.8919770.6789010.095936−0.507450.704932420.059177−0.44699−0.998840.1466580.1318040.12401443−0.04380.550155−0.44803−0.11205−1.251610.2526544−0.052110.8891210.3267150.101019−0.3035−0.36756450.1037930.6690830.1078810.1041020.21348−0.8854460.0200280.494509−0.496080.0400170.0232040.998507470.0020560.4195230.528138−0.0379−0.60652−0.0101548−0.06364−0.059410.122114−0.124090.0587550.608592490.104192−0.615960.653391−0.12175−1.25537−0.31836500.125236−0.15914−0.88180.056407−0.24510.336779 Table 9Neurons/weights for Example 1 and 2.WeightsExample 1Example 2S.NOΦjwjγjΦjwjγj10.028053−0.717670.152813−0.0806−1.084370.2756662−0.108530.3911310.041023−0.02682−0.773820.4771583−0.10992−0.5969−0.73395−0.14203−0.750610.65375940.112599−0.675230.429002−0.14020.2844410.8280945−0.134850.868873−0.646280.021106−0.01229−0.32646−0.07396−0.69394−0.153450.0973911.140424−0.0151670.001866−0.928830.342457−0.08558−0.44375−0.899888−0.04509−1.02622−0.75901−0.052360.2820860.86302590.0470960.1399240.614528−0.00104−0.20076−0.8157310−0.10384−0.67060.3468110.0512881.102195−0.1829111−0.10756−0.229080.6737050.000588−0.909180.10096120.0745660.376885−0.244280.0021290.339112−0.72684130.0018910.146690.3642130.088020.9216350.430377140.070815−0.624640.055019−0.04672−0.93534−0.0599415−0.13538−0.37624−0.396490.1034−1.35018−0.10702160.034652−0.225140.05733−0.095010.9168860.122526170.092315−0.252450.4020060.082134−0.79287−1.0456518−0.08332−0.113790.920643−0.10671−1.00343−0.0851619−0.12242−0.142230.712165−0.077220.619515−0.18713200.0690.383267−0.45006−0.037940.1188820.80585721−0.08013−0.65748−0.85584−0.155250.677478−0.6406422−0.066990.179206−0.362410.036199−0.22279−0.5500723−0.00926−0.409930.4133580.128462−0.70610.3271924−0.147380.6540540.8844420.0145191.1471490.09168425−0.072490.339125−0.96332−0.035780.260314−0.10952260.009538−0.85029−0.499010.029934−0.945480.75560127−0.081480.904719−0.38681−0.087340.2998660.274902280.0039690.5328360.574269−0.3019−0.9292−0.06303290.0953130.6973380.465553−0.07879−0.374020.75389330−0.09733−0.366260.067302−0.15016−0.385760.63057631−0.08106−0.25051−0.73586−0.09860.818595−0.3674332−0.09473−0.5254−0.478150.0285810.0580320.730424330.0201030.5509270.931746−0.0823−0.11696−0.2862340.128766−0.028380.5743580.151455−0.98204−0.01481350.016736−0.624520.1978210.025902−1.15114−0.2022236−0.10566−0.27703−0.57805−0.141950.9165960.04409737−0.09463−0.52059−0.959830.127858−0.12304−0.92612380.076331−0.732720.6419690.129936−0.893550.425223390.1261010.203775−0.55530.1673770.4009240.39990440−0.13954−0.972580.6191550.167619−0.900350.9121341−0.080960.8919770.6789010.095936−0.507450.704932420.059177−0.44699−0.998840.1466580.1318040.12401443−0.04380.550155−0.44803−0.11205−1.251610.2526544−0.052110.8891210.3267150.101019−0.3035−0.36756450.1037930.6690830.1078810.1041020.21348−0.8854460.0200280.494509−0.496080.0400170.0232040.998507470.0020560.4195230.528138−0.0379−0.60652−0.0101548−0.06364−0.059410.122114−0.124090.0587550.608592490.104192−0.615960.653391−0.12175−1.25537−0.31836500.125236−0.15914−0.88180.056407−0.24510.336779 Neurons/weights for Example 1 and 2.