Epidemic dynamics prediction using fractional SIRD and deep learning Authors: Ramsha Shafqat, Kinda Abuasbeh, Salma Trabelsi, Marwa Balti Published: 2025 ABSTRACT This paper introduces, analyzes, and numerically investigates a fractional-order SIRD epidemic model employing the normalized Caputo-Fabrizio (NCF) derivative. The model incorporates memory effects and explicitly tracks disease-induced mortality through a deceased compartment, offering a more comprehensive framework for epidemic dynamics. We establish existence, uniqueness, positivity, and population conservation for the proposed system, and present a robust numerical scheme. We demonstrate the influence of the memory parameter and kernel normalization through simulations, and highlight their significance for epidemic forecasting and real-world applications. Deep neural network (DNN) was used to approximate SIRD dynamics with fractional order and improve the accuracy of predictions. The model successfully extracted nonlinear time-dependent patterns from simulated epidemic data through its learning process. The proposed framework demonstrated strong predictive ability, achieving mean square error values as low as 0.00027 and root mean square errors below 0.17 across different compartments. By integrating fractional calculus with deep learning techniques, the model establishes a reliable approach for representing complex disease dynamics and provides meaningful insights that can support public health decision-making. Introduction Mathematical modeling is a cornerstone for analyzing, predicting, and managing the transmission dynamics of infectious diseases1,2. Among various frameworks, compartmental models such as the susceptible-infectious-recovered (SIR) and its extensions, including the susceptible-infectious-recovered-deceased (SIRD) model, are fundamental tools for capturing the spread of epidemics and informing public health interventions3. However, classical integer-order models often fail to account for the memory and hereditary properties present in real outbreaks such as variable infectious periods, delayed effects of control measures, and non-exponential waiting times4,5. Fractional-order differential equations have thus gained prominence in epidemiological modeling for their ability to incorporate memory effects, whereby the system’s current state depends on its historical trajectory5–7. The Caputo derivative, with its power-law kernel, is widely used but suffers from singularities, while the Caputo-Fabrizio (CF) derivative, featuring a non-singular exponential kernel, offers both computational advantages and physically realistic fading memory8–10. Nonetheless, the standard CF kernel is not normalized, which can introduce inconsistencies in the interpretation and scaling of memory effects. The NCF derivative addresses this limitation by ensuring the kernel integrates to unity, thereby preserving the operator’s interpretation as a true weighted average11. Despite theoretical and practical advances, fractional epidemic models with explicit mortality compartments have not been systematically explored with the NCF derivative. The SIRD model extends the classic SIR framework by introducing a deceased class, capturing fatalities due to infection, a key metric for assessing public health impact and evaluating intervention strategies (e.g. COVID-19, SARS, Ebola). Fractional operators and related numerical schemes have proved decisive for capturing memory and anomalous transport in complex systems, from porous-media diffusion and nonlinear waves to biological propagation12. Recent work combines fractional modeling with machine learning to enhance stability analysis, parameter inference, and forecasting across rabies and hepatitis B13,14. Meanwhile, new SIQR frameworks with the NCF operator and deep-learning surrogates extend these ideas to broader epidemic settings15. Fractional approaches have also been applied to poliomyelitis using neural networks, highlighting data-integrated strategies for control policy design16. Complementary evidence from vaccine safety research covering HLH, Sweet’s syndrome, and field-wide bibliometrics provides empirical context and motivates risk-aware modeling pipelines17–19. This study presents, analyzes, and numerically investigates a fractional SIRD epidemic model governed by the NCF operator. We provide a rigorous mathematical analysis of the model’s properties (existence, uniqueness, positivity, and boundedness), derive a robust numerical scheme, and compare the NCF-SIRD model with its classical and unnormalized fractional counterparts (Fig. 1). In addition to classical numerical analysis, we leverage DNNs to approximate the compartmental solutions and perform model validation. Recent research demonstrates that deep learning algorithms, particularly DNNs, can effectively complement fractional-order models by offering enhanced predictive accuracy, robustness to noise, and superior capability in capturing complex, nonlinear relationships within simulation or empirical data20–22. In our approach, DNN techniques are employed to fit and forecast the solution trajectories of the SIRD compartments, with model performance assessed using mean squared error, regression analysis, and error distribution metrics. The primary objectives of this work are as follows: Develop and analyze a fractional SIRD epidemic model using the NCF derivative.Prove the essential mathematical properties of the model.Propose and implement a reliable numerical scheme.Demonstrate the effectiveness of both numerical and DNN-based approaches through comprehensive simulations and comparative analyses. Fig. 1Schematic diagram of the SIRD model. Develop and analyze a fractional SIRD epidemic model using the NCF derivative. Develop and analyze a fractional SIRD epidemic model using the NCF derivative. Prove the essential mathematical properties of the model. Prove the essential mathematical properties of the model. Propose and implement a reliable numerical scheme. Propose and implement a reliable numerical scheme. Demonstrate the effectiveness of both numerical and DNN-based approaches through comprehensive simulations and comparative analyses. Demonstrate the effectiveness of both numerical and DNN-based approaches through comprehensive simulations and comparative analyses. Schematic diagram of the SIRD model. Section "Preliminaries" reviews the necessary preliminaries on fractional derivatives and machine learning concepts. Section "Model formulation" presents the model formulation. Section "Theoretical analysis" details the theoretical analysis. Section 'Numerical method" introduces the numerical method. Section "Numerical simulations and discussion" reports the simulation and DNN results. Section "Conclusion" provides concluding remarks and outlines directions for future research. Preliminaries Before proceeding, we present the following definitions, which will be fundamental in the development of our main results. Definition 2.1 Reference8 For \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$0< \alpha < 1$$\end{document}0<α<1, the CF derivative of a function \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$u(t)$$\end{document}u(t) is\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} (^{CF}D_0^\alpha u)(t) = \frac{1}{1-\alpha } \int _0^t e^{-\mu _\alpha (t-s)} u^\prime (s) ds, \end{aligned}$$\end{document}(CFD0αu)(t)=11-α∫0te-μα(t-s)u′(s)ds,where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _\alpha = \frac{\alpha }{1-\alpha }$$\end{document}μα=α1-α. Definition 2.2 Reference23 The NCF derivative is defined by\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} (^{NCF}D_0^\alpha u)(t) = \frac{1}{(1-\alpha )C_\alpha (t)} \int _0^t e^{-\mu _\alpha (t-s)} u^\prime (s) ds, \end{aligned}$$\end{document}(NCFD0αu)(t)=1(1-α)Cα(t)∫0te-μα(t-s)u′(s)ds,where\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} C_\alpha (t) = \frac{1}{\alpha } (1 - e^{-\mu _\alpha t}). \end{aligned}$$\end{document}Cα(t)=1α(1-e-μαt).This normalization ensures the kernel integrates to 1 over \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$[0,\upsilon ]$$\end{document}[0,υ]. Throughout let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$0<\alpha <1$$\end{document}0<α<1 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _\alpha =\alpha /(1-\alpha )$$\end{document}μα=α/(1-α). We assume \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$u\in W^{1,1}(0,T)$$\end{document}u∈W1,1(0,T) (i.e. u is absolutely continuous with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$u\prime \in L^1(0,T)$$\end{document}u′∈L1(0,T)). For the SIRD system we work in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X=[C([0,T])]^4\cap [W^{1,1}(0,T)]^4$$\end{document}X=[C([0,T])]4∩[W1,1(0,T)]4 with nonnegative data. The normalized Caputo–Fabrizio operator is\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} (^{NCF}\,D_0^\alpha u)(t) =\frac{1}{(1-\alpha )C_\alpha (t)}\int _0^t e^{-\mu _\alpha (t-s)}u^\prime (s)\,ds, \qquad C_\alpha (t)=\frac{1}{\alpha }\bigl (1-e^{-\mu _\alpha t}\bigr ). \end{aligned}$$\end{document}(NCFD0αu)(t)=1(1-α)Cα(t)∫0te-μα(t-s)u′(s)ds,Cα(t)=1α(1-e-μαt).. Lemma 2.2.1 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_\alpha \in C^\infty (0,\infty )$$\end{document}Cα∈C∞(0,∞), \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_\alpha (0)=0$$\end{document}Cα(0)=0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_\alpha ^\prime (t)=\frac{\mu _\alpha }{\alpha }e^{-\mu _\alpha t}>0$$\end{document}Cα′(t)=μααe-μαt>0, and\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} C_\alpha (t)\sim \frac{1}{1-\alpha }\,t\quad (t\downarrow 0), \qquad \lim _{t\rightarrow \infty }C_\alpha (t)=\frac{1}{\alpha }. \end{aligned}$$\end{document}Cα(t)∼11-αt(t↓0),limt→∞Cα(t)=1α.In particular \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(1-\alpha )C_\alpha (t)\sim t$$\end{document}(1-α)Cα(t)∼t as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t\downarrow 0$$\end{document}t↓0. Model formulation We consider the normalized fractional SIRD model:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} (^{NCF}D_0^\alpha S)(t)&= -\beta S(t) I(t), \nonumber \\ (^{NCF}D_0^\alpha I)(t)&= \beta S(t) I(t) - \gamma I(t) - \delta I(t),\nonumber \\ (^{NCF}D_0^\alpha R)(t)&= \gamma I(t), \nonumber \\ (^{NCF}D_0^\alpha D)(t)&= \delta I(t), \nonumber \\ S(0)= S_0 \ge 0,~ I(0) = I_0 \ge 0,&~ R(0) = R_0 \ge 0,~ D(0) = D_0 \ge 0,~ \end{aligned}$$\end{document}(NCFD0αS)(t)=-βS(t)I(t),(NCFD0αI)(t)=βS(t)I(t)-γI(t)-δI(t),(NCFD0αR)(t)=γI(t),(NCFD0αD)(t)=δI(t),S(0)=S0≥0,I(0)=I0≥0,R(0)=R0≥0,D(0)=D0≥0,where \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}β is the infection rate, \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}γ is the recovery rate, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta$$\end{document}δ is the disease-induced mortality rate. The total initial population is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_0 = S_0 + I_0 + R_0 + D_0$$\end{document}N0=S0+I0+R0+D0. Within this framework, the functions S(t), I(t), R(t), and D(t) represent, respectively, the fractions of the population that are susceptible, infected, recovered, and deceased from the disease at any given time t. The operator \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{NCF}D_0^\alpha$$\end{document}NCFD0α stands for the NCF fractional 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}$$\alpha \in (0,1)$$\end{document}α∈(0,1). This operator introduces memory effects into the dynamics through a normalized, non-singular kernel, enabling the model to incorporate the influence of past states on present epidemic behavior. The first equation describes the depletion of susceptible individuals as they become infected through contact with infectious individuals at a rate \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}β. The second equation accounts for the change in the infectious population, with increases due to new infections and decreases due to recovery (\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}γ) and disease-induced death (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta$$\end{document}δ). The third equation captures the accumulation of recovered individuals. The fourth equation models the cumulative deaths due to infection. 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}$$S_0$$\end{document}S0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_0$$\end{document}I0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0$$\end{document}R0 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D_0$$\end{document}D0 specify the initial distribution across the compartments, assumed non-negative and summing to the total population \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_0$$\end{document}N0. By extending the classical SIRD model with the NCF derivative, this formulation incorporates memory effects commonly observed in real epidemics, such as delays in infection or removal, temporary immunity, or persistent behavioral change. When \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha \rightarrow 1$$\end{document}α→1, the NCF-SIRD model reduces to the standard SIRD system with integer-order derivatives. The normalized kernel ensures that the memory contribution is consistently scaled, mitigating the mathematical singularities found in traditional fractional operators and thus enhancing the model’s ability to capture and forecast realistic epidemic phenomena. Theoretical analysis We establish the existence and uniqueness of solutions to the NCF-SIRD system, providing a foundation for the model’s mathematical validity. The demonstration relies on fixed-point theory and essential properties of the NCF fractional operator. Existence and uniqueness of solutions Before analyzing the dynamics, we establish that the NCF-SIRD model admits a unique global solution. This guarantees the mathematical consistency and reliability of the system. Under the standing assumptions of Sect. "Preliminaries" (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U\in [C([0,T])]^4\cap [W^{1,1}(0,T)]^4$$\end{document}U∈[C([0,T])]4∩[W1,1(0,T)]4 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$0<\alpha <1$$\end{document}0<α<1), the NCF-SIRD system admits a unique solution on [0, T]. Theorem 4.1 There exists a unique global solution (S, I, R, D) to the NCF-SIRD system on [0, T]. Proof We rewrite the NCF-SIRD model using properties of the NCF operator (see Ref.9):\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} S(t)&= S_0 + (1-\alpha )C_\alpha (t)\left[ -\beta S(t)I(t)\right] + \alpha \int _0^t C_\alpha (\tau )\left[ -\beta S(\tau )I(\tau )\right] \,d\tau , \\ I(t)&= I_0 + (1-\alpha )C_\alpha (t)\left[ \beta S(t)I(t) - \gamma I(t) - \delta I(t)\right] + \alpha \int _0^t C_\alpha (\tau )\left[ \beta S(\tau )I(\tau ) - \gamma I(\tau ) - \delta I(\tau )\right] \,d\tau , \\ R(t)&= R_0 + (1-\alpha )C_\alpha (t)\left[ \gamma I(t)\right] + \alpha \int _0^t C_\alpha (\tau )\left[ \gamma I(\tau )\right] \,d\tau , \\ D(t)&= D_0 + (1-\alpha )C_\alpha (t)\left[ \delta I(t)\right] + \alpha \int _0^t C_\alpha (\tau )\left[ \delta I(\tau )\right] \,d\tau , \end{aligned}$$\end{document}S(t)=S0+(1-α)Cα(t)-βS(t)I(t)+α∫0tCα(τ)-βS(τ)I(τ)dτ,I(t)=I0+(1-α)Cα(t)βS(t)I(t)-γI(t)-δI(t)+α∫0tCα(τ)βS(τ)I(τ)-γI(τ)-δI(τ)dτ,R(t)=R0+(1-α)Cα(t)γI(t)+α∫0tCα(τ)γI(τ)dτ,D(t)=D0+(1-α)Cα(t)δI(t)+α∫0tCα(τ)δI(τ)dτ,where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_\alpha (t)$$\end{document}Cα(t) is the normalization function of the kernel. Let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X = [C([0,T])]^4$$\end{document}X=[C([0,T])]4, the Banach space of continuous vector-valued functions on [0, T] with norm\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Vert U\Vert = \max \{\Vert S\Vert _\infty , \Vert I\Vert _\infty , \Vert R\Vert _\infty , \Vert D\Vert _\infty \}. \end{aligned}$$\end{document}‖U‖=max{‖S‖∞,‖I‖∞,‖R‖∞,‖D‖∞}.Define \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f_1 = -\beta S I$$\end{document}f1=-βSI, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f_2 = \beta S I - \gamma I - \delta I$$\end{document}f2=βSI-γI-δI, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f_3 = \gamma I$$\end{document}f3=γI, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f_4 = \delta I$$\end{document}f4=δI. Define the operator \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal {T}$$\end{document}T on X by:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \mathcal {T} \begin{pmatrix} S \\ I \\ R \\ D \end{pmatrix} (t) = \begin{pmatrix} S_0 + (1-\alpha )C_\alpha (t)f_1(S(t),I(t),R(t),D(t)) + \alpha \int _0^t C_\alpha (\tau )f_1(S(\tau ),I(\tau ),R(\tau ),D(\tau ))\,d\tau \\ I_0 + (1-\alpha )C_\alpha (t)f_2(S(t),I(t),R(t),D(t)) + \alpha \int _0^t C_\alpha (\tau )f_2(S(\tau ),I(\tau ),R(\tau ),D(\tau ))\,d\tau \\ R_0 + (1-\alpha )C_\alpha (t)f_3(S(t),I(t),R(t),D(t)) + \alpha \int _0^t C_\alpha (\tau )f_3(S(\tau ),I(\tau ),R(\tau ),D(\tau ))\,d\tau \\ D_0 + (1-\alpha )C_\alpha (t)f_4(S(t),I(t),R(t),D(t)) + \alpha \int _0^t C_\alpha (\tau )f_4(S(\tau ),I(\tau ),R(\tau ),D(\tau ))\,d\tau \end{pmatrix} \end{aligned}$$\end{document}TSIRD(t)=S0+(1-α)Cα(t)f1(S(t),I(t),R(t),D(t))+α∫0tCα(τ)f1(S(τ),I(τ),R(τ),D(τ))dτI0+(1-α)Cα(t)f2(S(t),I(t),R(t),D(t))+α∫0tCα(τ)f2(S(τ),I(τ),R(τ),D(τ))dτR0+(1-α)Cα(t)f3(S(t),I(t),R(t),D(t))+α∫0tCα(τ)f3(S(τ),I(τ),R(τ),D(τ))dτD0+(1-α)Cα(t)f4(S(t),I(t),R(t),D(t))+α∫0tCα(τ)f4(S(τ),I(τ),R(τ),D(τ))dτWe will show \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal {T}$$\end{document}T is a contraction on a suitable subset of X by proving each \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f_j$$\end{document}fj is Lipschitz continuous on bounded sets. Let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D_M = \{(S,I,R,D)\in X: \Vert S\Vert ,\Vert I\Vert ,\Vert R\Vert ,\Vert D\Vert \le M\}$$\end{document}DM={(S,I,R,D)∈X:‖S‖,‖I‖,‖R‖,‖D‖≤M} for some \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M > 0$$\end{document}M>0. For all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(S_1,I_1,R_1,D_1), (S_2,I_2,R_2,D_2)$$\end{document}(S1,I1,R1,D1),(S2,I2,R2,D2) in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D_M$$\end{document}DM:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned} |f_1(S_1,I_1,R_1,D_1) - f_1(S_2,I_2,R_2,D_2)|&= |\!-\beta S_1 I_1 + \beta S_2 I_2| \\&= \beta |S_2 I_2 - S_1 I_1| \\&= \beta |S_2(I_2 - I_1) + (S_2 - S_1)I_1| \\&\le \beta (|S_2||I_2 - I_1| + |I_1||S_2 - S_1|) \\&\le 2\beta M \Vert (S_1,I_1,R_1,D_1)-(S_2,I_2,R_2,D_2)\Vert \end{aligned} \end{aligned}$$\end{document}|f1(S1,I1,R1,D1)-f1(S2,I2,R2,D2)|=|-βS1I1+βS2I2|=β|S2I2-S1I1|=β|S2(I2-I1)+(S2-S1)I1|≤β(|S2||I2-I1|+|I1||S2-S1|)≤2βM‖(S1,I1,R1,D1)-(S2,I2,R2,D2)‖\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned} |f_2(S_1,I_1,R_1,D_1) - f_2(S_2,I_2,R_2,D_2)|&= |\beta S_1 I_1 - \gamma I_1 - \delta I_1 - [\beta S_2 I_2 - \gamma I_2 - \delta I_2]| \\&= |\beta (S_1 I_1 - S_2 I_2) - (\gamma + \delta )(I_1 - I_2)| \\&\le \beta (|S_1||I_1 - I_2| + |I_2||S_1 - S_2|) + (\gamma + \delta )|I_1 - I_2| \\&\le (2\beta M + \gamma + \delta )\Vert (S_1,I_1,R_1,D_1)-(S_2,I_2,R_2,D_2)\Vert \ \\ |f_3(S_1,I_1,R_1,D_1) - f_3(S_2,I_2,R_2,D_2)| = |\gamma I_1 - \gamma I_2|&= \gamma |I_1 - I_2| \le \gamma \Vert (S_1,I_1,R_1,D_1)-(S_2,I_2,R_2,D_2)\Vert \\ |f_4(S_1,I_1,R_1,D_1) - f_4(S_2,I_2,R_2,D_2)| = |\delta I_1 - \delta I_2|&= \delta |I_1 - I_2| \le \delta \Vert (S_1,I_1,R_1,D_1)-(S_2,I_2,R_2,D_2)\Vert \end{aligned} \end{aligned}$$\end{document}|f2(S1,I1,R1,D1)-f2(S2,I2,R2,D2)|=|βS1I1-γI1-δI1-[βS2I2-γI2-δI2]|=|β(S1I1-S2I2)-(γ+δ)(I1-I2)|≤β(|S1||I1-I2|+|I2||S1-S2|)+(γ+δ)|I1-I2|≤(2βM+γ+δ)‖(S1,I1,R1,D1)-(S2,I2,R2,D2)‖|f3(S1,I1,R1,D1)-f3(S2,I2,R2,D2)|=|γI1-γI2|=γ|I1-I2|≤γ‖(S1,I1,R1,D1)-(S2,I2,R2,D2)‖|f4(S1,I1,R1,D1)-f4(S2,I2,R2,D2)|=|δI1-δI2|=δ|I1-I2|≤δ‖(S1,I1,R1,D1)-(S2,I2,R2,D2)‖Let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X_T=[C([0,T])]^4$$\end{document}XT=[C([0,T])]4 (sup norm) and, on the closed ball \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbb {B}_M=\{U\in X_T:\Vert U\Vert _\infty \le M\}$$\end{document}BM={U∈XT:‖U‖∞≤M}, note that\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} F(S,I,R,D)=(-\beta SI,\ \beta SI-(\gamma +\delta )I,\ \gamma I,\ \delta I) \end{aligned}$$\end{document}F(S,I,R,D)=(-βSI,βSI-(γ+δ)I,γI,δI)is locally Lipschitz with\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} |f_1(U)-f_1(V)|\le 2\beta M\,\Vert U-V\Vert _\infty ,\quad |f_2(U)-f_2(V)|\le (2\beta M+\gamma +\delta )\,\Vert U-V\Vert _\infty , \end{aligned}$$\end{document}|f1(U)-f1(V)|≤2βM‖U-V‖∞,|f2(U)-f2(V)|≤(2βM+γ+δ)‖U-V‖∞,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} |f_3(U)-f_3(V)|\le \gamma \,\Vert U-V\Vert _\infty ,\quad |f_4(U)-f_4(V)|\le \delta \,\Vert U-V\Vert _\infty . \end{aligned}$$\end{document}|f3(U)-f3(V)|≤γ‖U-V‖∞,|f4(U)-f4(V)|≤δ‖U-V‖∞.Hence, with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_M:=\max \{2\beta M,\ 2\beta M+\gamma +\delta ,\ \gamma ,\ \delta \}$$\end{document}LM:=max{2βM,2βM+γ+δ,γ,δ},\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Vert T[U]-T[V]\Vert _\infty \le L_M\,K_\alpha (T)\,\Vert U-V\Vert _\infty , \end{aligned}$$\end{document}‖T[U]-T[V]‖∞≤LMKα(T)‖U-V‖∞,where\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} K_\alpha (T):= \sup _{t\in [0,T]}(1-\alpha )C_\alpha (t)\;+\;\alpha \!\int _0^T\!C_\alpha (\tau )\,d\tau . \end{aligned}$$\end{document}Kα(T):=supt∈[0,T](1-α)Cα(t)+α∫0TCα(τ)dτ.Using \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1-e^{-x}\le x$$\end{document}1-e-x≤x for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$x\ge 0$$\end{document}x≥0,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} C_\alpha (t)=\frac{1-e^{-\mu _\alpha t}}{\alpha }\le \frac{\mu _\alpha t}{\alpha }=\frac{t}{1-\alpha }, \quad \Rightarrow \quad \sup _{t\in [0,T]}(1-\alpha )C_\alpha (t)\le T, \end{aligned}$$\end{document}Cα(t)=1-e-μαtα≤μαtα=t1-α,⇒supt∈[0,T](1-α)Cα(t)≤T,and thus\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \int _0^T C_\alpha (\tau )\,d\tau \le \int _0^T \frac{\tau }{1-\alpha }\,d\tau =\frac{T^2}{2(1-\alpha )}. \end{aligned}$$\end{document}∫0TCα(τ)dτ≤∫0Tτ1-αdτ=T22(1-α).Therefore\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} K_\alpha (T)\ \le \ T+\frac{\alpha }{2(1-\alpha )}T^2, \end{aligned}$$\end{document}Kα(T)≤T+α2(1-α)T2,so T can be chosen so that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_M\,K_\alpha (T)<1$$\end{document}LMKα(T)<1, making T a contraction on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbb {B}_M$$\end{document}BM. By Banach’s fixed-point theorem, there is a unique fixed point (solution) in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D_M$$\end{document}DM on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$[0, T_0]$$\end{document}[0,T0] for some \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_0 > 0$$\end{document}T0>0. Since the solutions remain non-negative and bounded by the total population (which is conserved in SIRD models), the local solution can be extended step by step to the whole interval [0, T]. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\square$$\end{document}□ Let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X_T=[C([0,T])]^4$$\end{document}XT=[C([0,T])]4 with the sup norm and define, for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U=(S,I,R,D)$$\end{document}U=(S,I,R,D),\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \mathcal {T}[U](t)=U(0)+(1-\alpha )C_\alpha (t)\,F\!\left( U(t)\right) +\alpha \int _0^t C_\alpha (\tau )\,F\!\left( U(\tau )\right) \,d\tau , \end{aligned}$$\end{document}T[U](t)=U(0)+(1-α)Cα(t)FU(t)+α∫0tCα(τ)FU(τ)dτ,where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F(S,I,R,D)=(-\beta SI,\ \beta SI-(\gamma +\delta )I,\ \gamma I,\ \delta I)$$\end{document}F(S,I,R,D)=(-βSI,βSI-(γ+δ)I,γI,δI). Assume F is locally Lipschitz; then, on a closed ball \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbb {B}_M\subset X_T$$\end{document}BM⊂XT,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Vert \mathcal {T}[U]-\mathcal {T}[V]\Vert _\infty \le L_M\Big (\sup _{t\in [0,T]}(1-\alpha )C_\alpha (t) +\alpha \int _0^T C_\alpha (\tau )\,d\tau \Big )\,\Vert U-V\Vert _\infty . \end{aligned}$$\end{document}‖T[U]-T[V]‖∞≤LM(supt∈[0,T](1-α)Cα(t)+α∫0TCα(τ)dτ)‖U-V‖∞.Since \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$0\le C_\alpha (\cdot )\le 1/\alpha$$\end{document}0≤Cα(·)≤1/α, the bracket is finite and O(T). For T sufficiently small, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal {T}$$\end{document}T is a contraction on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbb {B}_M$$\end{document}BM, yielding a unique solution \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U\in X_T\cap [W^{1,1}(0,T)]^4$$\end{document}U∈XT∩[W1,1(0,T)]4. Standard continuation and the population bound extend the solution on [0, T]. All integral manipulations are justified by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$u\in W^{1,1}$$\end{document}u∈W1,1 and the bounded, integrable kernel \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e^{-\mu _\alpha (t-s)}$$\end{document}e-μα(t-s) on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{0\le s\le t\le T\}$$\end{document}{0≤s≤t≤T}. Positivity of solutions Before proceeding, it is essential to verify that the solutions of the NCF-SIRD model remain non-negative for all time, as required by the biological interpretation of the compartments. Theorem 4.2 If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_0, I_0, R_0, D_0 \ge 0$$\end{document}S0,I0,R0,D0≥0, then \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S(t), I(t), R(t), D(t) \ge 0$$\end{document}S(t),I(t),R(t),D(t)≥0 for all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t \ge 0$$\end{document}t≥0. Proof Suppose, for contradiction, there exists a first time \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}t∗>0 at which \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S(t^*)=0$$\end{document}S(t∗)=0 or \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I(t^*)=0$$\end{document}I(t∗)=0 or \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R(t^*)=0$$\end{document}R(t∗)=0 or \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D(t^*)=0$$\end{document}D(t∗)=0, after being non-negative for all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t < t^*$$\end{document}t0$$\end{document}Δ>0 with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_M\,K_\alpha (\Delta )<1$$\end{document}LMKα(Δ)<1. Iterating the local fixed-point solution on successive subintervals of length \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta$$\end{document}Δ covers [0, T], which yields a unique global solution on [0, T]. Equilibria and disease-free invariance Setting all derivatives to zero, the NCF-SIRD system becomes:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} 0&= -\beta S^*I^*, \\ 0&= \beta S^*I^*- (\gamma +\delta ) I^*, \\ 0&= \gamma I^*, \\ 0&= \delta I^*. \end{aligned}$$\end{document}0=-βS∗I∗,0=βS∗I∗-(γ+δ)I∗,0=γI∗,0=δI∗.Setting derivatives to zero yields \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I^*= 0$$\end{document}I∗=0 and conservation \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S^*+ R^*+ D^*= N_0$$\end{document}S∗+R∗+D∗=N0; hence the model admits a continuum of disease-free equilibria\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \{(S^*, 0, R^*, D^*):\ S^*, R^*, D^*\ge 0,\ S^*+ R^*+ D^*= N_0\}. \end{aligned}$$\end{document}{(S∗,0,R∗,D∗):S∗,R∗,D∗≥0,S∗+R∗+D∗=N0}. Stability of the disease-free equilibrium We analyze the stability of the disease-free equilibrium (DFE) of the NCF-SIRD model. Theorem 4.4 The disease-free equilibrium \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(S^*, I^*, R^*, D^*) = (N_0, 0, 0, 0)$$\end{document}(S∗,I∗,R∗,D∗)=(N0,0,0,0) of the NCF-SIRD model is locally asymptotically stable if \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal {R}_0 = \frac{\beta N_0}{\gamma + \delta } < 1$$\end{document}R0=βN0γ+δ<1, and unstable if \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal {R}_0 > 1$$\end{document}R0>1. Proof Recall the NCF-SIRD system:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} (^{NCF}D_0^\alpha S)(t)&= -\beta S(t) I(t), \\ (^{NCF}D_0^\alpha I)(t)&= \beta S(t) I(t) - (\gamma +\delta ) I(t), \\ (^{NCF}D_0^\alpha R)(t)&= \gamma I(t), \\ (^{NCF}D_0^\alpha D)(t)&= \delta I(t). \end{aligned}$$\end{document}(NCFD0αS)(t)=-βS(t)I(t),(NCFD0αI)(t)=βS(t)I(t)-(γ+δ)I(t),(NCFD0αR)(t)=γI(t),(NCFD0αD)(t)=δI(t).The DFE is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(S^*, I^*, R^*, D^*) = (N_0, 0, 0, 0)$$\end{document}(S∗,I∗,R∗,D∗)=(N0,0,0,0). Introduce small perturbations near the equilibrium:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} S(t) = N_0 + s(t), \quad I(t) = 0 + i(t), \quad R(t) = 0 + r(t), \quad D(t) = 0 + d(t), \end{aligned}$$\end{document}S(t)=N0+s(t),I(t)=0+i(t),R(t)=0+r(t),D(t)=0+d(t),where |s(t)|, |i(t)|, |r(t)|, |d(t)| are small. The linearized system becomes:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} (^{NCF}D_0^\alpha s)(t)&= -\beta N_0 i(t), \\ (^{NCF}D_0^\alpha i)(t)&= \beta N_0 i(t) - (\gamma +\delta ) i(t), \\ (^{NCF}D_0^\alpha r)(t)&= \gamma i(t), \\ (^{NCF}D_0^\alpha d)(t)&= \delta i(t). \end{aligned}$$\end{document}(NCFD0αs)(t)=-βN0i(t),(NCFD0αi)(t)=βN0i(t)-(γ+δ)i(t),(NCFD0αr)(t)=γi(t),(NCFD0αd)(t)=δi(t).Writing the NCF equation \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\textrm{NCF}\,D_0^\alpha i)(t)=\lambda \,i(t)$$\end{document}(NCFD0αi)(t)=λi(t) in its integral form gives\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} i(t)=i_0+(1-\alpha )C_\alpha (t)\,\lambda \,i(t)\;+\;\alpha \int _0^t C_\alpha (\tau )\,\lambda \,i(\tau )\,d\tau . \end{aligned}$$\end{document}i(t)=i0+(1-α)Cα(t)λi(t)+α∫0tCα(τ)λi(τ)dτ.Set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y(t):=|i(t)|$$\end{document}y(t):=|i(t)|. Using \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_\alpha \ge 0$$\end{document}Cα≥0 and the triangle inequality,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} y(t)\le |i_0|+|\lambda |\,(1-\alpha )C_\alpha (t)\,y(t)+|\lambda |\,\alpha \int _0^t C_\alpha (\tau )\,y(\tau )\,d\tau . \end{aligned}$$\end{document}y(t)≤|i0|+|λ|(1-α)Cα(t)y(t)+|λ|α∫0tCα(τ)y(τ)dτ.For any \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T>0$$\end{document}T>0 with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1-|\lambda |(1-\alpha )\,C_\alpha (T)>0$$\end{document}1-|λ|(1-α)Cα(T)>0,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} y(t)\le \frac{|i_0|}{1-|\lambda |(1-\alpha )C_\alpha (T)} +\frac{|\lambda |\,\alpha }{1-|\lambda |(1-\alpha )C_\alpha (T)}\int _0^t C_\alpha (\tau )\,y(\tau )\,d\tau . \end{aligned}$$\end{document}y(t)≤|i0|1-|λ|(1-α)Cα(T)+|λ|α1-|λ|(1-α)Cα(T)∫0tCα(τ)y(τ)dτ.By Grönwall’s inequality for Volterra integrals,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} y(t)\le \frac{|i_0|}{1-|\lambda |(1-\alpha )C_\alpha (T)} \exp \!\Big (\frac{|\lambda |\,\alpha }{1-|\lambda |(1-\alpha )C_\alpha (T)}\int _0^t C_\alpha (\tau )\,d\tau \Big ). \end{aligned}$$\end{document}y(t)≤|i0|1-|λ|(1-α)Cα(T)exp(|λ|α1-|λ|(1-α)Cα(T)∫0tCα(τ)dτ).Using \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_\alpha (\tau )\le \tau /(1-\alpha )$$\end{document}Cα(τ)≤τ/(1-α) yields the explicit bound\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} |i(t)|\le \frac{|i_0|}{1-|\lambda |(1-\alpha )C_\alpha (T)} \exp \!\Big (\frac{|\lambda |}{2(1-\alpha )}\,t^2\Big ). \end{aligned}$$\end{document}|i(t)|≤|i0|1-|λ|(1-α)Cα(T)exp(|λ|2(1-α)t2).If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda <0$$\end{document}λ<0 (i.e. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0<1$$\end{document}R0<1), then the same steps applied to the lower inequality give a decaying bound, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i(t)\rightarrow 0$$\end{document}i(t)→0 as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t\rightarrow \infty$$\end{document}t→∞. If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda >0$$\end{document}λ>0 (i.e. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0>1$$\end{document}R0>1), the upper bound grows super-exponentially in t, hence the DFE is unstable.If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda < 0$$\end{document}λ<0 (i.e. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta N_0 < \gamma +\delta$$\end{document}βN0<γ+δ or \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal {R}_0 < 1$$\end{document}R0<1), i(t) decays to zero as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t \rightarrow \infty$$\end{document}t→∞; the DFE is stable.If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda > 0$$\end{document}λ>0 (i.e. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal {R}_0 > 1$$\end{document}R0>1), i(t) grows; the DFE is unstable.Thus, the basic reproduction number for the SIRD model is:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \mathcal {R}_0 = \frac{\beta N_0}{\gamma + \delta } \end{aligned}$$\end{document}R0=βN0γ+δand the DFE is locally asymptotically stable if \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal {R}_0 < 1$$\end{document}R0<1, and unstable if \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal {R}_0 > 1$$\end{document}R0>1. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\square$$\end{document}□ If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda < 0$$\end{document}λ<0 (i.e. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta N_0 < \gamma +\delta$$\end{document}βN0<γ+δ or \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal {R}_0 < 1$$\end{document}R0<1), i(t) decays to zero as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t \rightarrow \infty$$\end{document}t→∞; the DFE is stable. If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda < 0$$\end{document}λ<0 (i.e. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta N_0 < \gamma +\delta$$\end{document}βN0<γ+δ or \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal {R}_0 < 1$$\end{document}R0<1), i(t) decays to zero as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t \rightarrow \infty$$\end{document}t→∞; the DFE is stable. If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda > 0$$\end{document}λ>0 (i.e. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal {R}_0 > 1$$\end{document}R0>1), i(t) grows; the DFE is unstable. If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda > 0$$\end{document}λ>0 (i.e. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal {R}_0 > 1$$\end{document}R0>1), i(t) grows; the DFE is unstable. Remark 1 For the NCF operator the kernel depends on the observation time via \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_\alpha (t)$$\end{document}Cα(t), so classical eigenfunction representations based on Mittag–Leffler functions are not directly applicable. Our proof uses the equivalent Volterra form, positivity, and Grönwall-type estimates, which are sufficient to establish the DFE stability/instability criterion without invoking special-function solutions. Remark 2 We have \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lim _{\alpha \uparrow 1} C_\alpha (t)=1$$\end{document}limα↑1Cα(t)=1 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(1-\alpha )C_\alpha (t)\rightarrow 0$$\end{document}(1-α)Cα(t)→0 for fixed \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t>0$$\end{document}t>0. Hence, the NCF integral form reduces to the usual first-order integral equation, and the model recovers the classical SIRD system as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha \rightarrow 1^{-}$$\end{document}α→1-. Numerical method Let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T > 0$$\end{document}T>0 be the final simulation time and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N \in \mathbb {N}$$\end{document}N∈N the total number of time steps. Define a uniform time grid with step size \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h = T/N$$\end{document}h=T/N and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_n = n h$$\end{document}tn=nh, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n = 0,1,\ldots ,N$$\end{document}n=0,1,…,N. Discretization of the NCF derivative The NCF derivative at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_n$$\end{document}tn is approximated by\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} (^{NCF}D_0^\alpha u)(t_n) \approx \frac{1}{(1-\alpha ) C_\alpha (t_n)} \sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)} [u_{k+1} - u_k], \end{aligned}$$\end{document}(NCFD0αu)(tn)≈1(1-α)Cα(tn)∑k=0n-1e-μα(tn-tk)[uk+1-uk],where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_\alpha (t_n) = \frac{1}{\alpha }(1 - e^{-\mu _\alpha t_n})$$\end{document}Cα(tn)=1α(1-e-μαtn) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _\alpha = \frac{\alpha }{1-\alpha }$$\end{document}μα=α1-α. Numerical scheme for the SIRD model For each variable \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$u \in \{S, I, R, D\}$$\end{document}u∈{S,I,R,D}, the update at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_n$$\end{document}tn is given by\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} u_n = u_0 + \frac{1}{(1-\alpha )C_\alpha (t_n)} \sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)} [u_{k+1} - u_k], \end{aligned}$$\end{document}un=u0+1(1-α)Cα(tn)∑k=0n-1e-μα(tn-tk)[uk+1-uk],where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$u_0$$\end{document}u0 is the initial value. The right-hand sides at each step depend on the current values of S, I, R, D. Step-by-step implementation: Initialization: Set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_0$$\end{document}S0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_0$$\end{document}I0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0$$\end{document}R0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D_0$$\end{document}D0, and model parameters \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta$$\end{document}β, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document}γ, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta$$\end{document}δ, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α. Compute \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _\alpha$$\end{document}μα.Time stepping: For \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n=1$$\end{document}n=1 to N: (i)Compute \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_\alpha (t_n) = \frac{1}{\alpha }(1 - e^{-\mu _\alpha t_n})$$\end{document}Cα(tn)=1α(1-e-μαtn).(ii)For each \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k=0$$\end{document}k=0 to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n-1$$\end{document}n-1, calculate the kernel \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e^{-\mu _\alpha (t_n - t_k)}$$\end{document}e-μα(tn-tk).(iii)Evaluate increments: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned} S_{k+1} - S_k&= -h \beta S_k I_k, \\ I_{k+1} - I_k&= h [\beta S_k I_k - (\gamma +\delta ) I_k], \\ R_{k+1} - R_k&= h \gamma I_k, \\ D_{k+1} - D_k&= h \delta I_k. \end{aligned} \end{aligned}$$\end{document}Sk+1-Sk=-hβSkIk,Ik+1-Ik=h[βSkIk-(γ+δ)Ik],Rk+1-Rk=hγIk,Dk+1-Dk=hδIk.(iv)Compute the sums: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned} \text {sumS}&= \sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)} [S_{k+1} - S_k], \\ \text {sumI}&= \sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)} [I_{k+1} - I_k], \\ \text {sumR}&= \sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)} [R_{k+1} - R_k], \\ \text {sumD}&= \sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)} [D_{k+1} - D_k]. \end{aligned} \end{aligned}$$\end{document}sumS=∑k=0n-1e-μα(tn-tk)[Sk+1-Sk],sumI=∑k=0n-1e-μα(tn-tk)[Ik+1-Ik],sumR=∑k=0n-1e-μα(tn-tk)[Rk+1-Rk],sumD=∑k=0n-1e-μα(tn-tk)[Dk+1-Dk].(v)Update: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned} S_n&= S_0 + \frac{1}{(1-\alpha )C_\alpha (t_n)} \text {sumS}, \\ I_n&= I_0 + \frac{1}{(1-\alpha )C_\alpha (t_n)} \text {sumI}, \\ R_n&= R_0 + \frac{1}{(1-\alpha )C_\alpha (t_n)} \text {sumR}, \\ D_n&= D_0 + \frac{1}{(1-\alpha )C_\alpha (t_n)} \text {sumD}. \end{aligned} \end{aligned}$$\end{document}Sn=S0+1(1-α)Cα(tn)sumS,In=I0+1(1-α)Cα(tn)sumI,Rn=R0+1(1-α)Cα(tn)sumR,Dn=D0+1(1-α)Cα(tn)sumD.Repeat until \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n = N$$\end{document}n=N. Initialization: Set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_0$$\end{document}S0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_0$$\end{document}I0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0$$\end{document}R0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D_0$$\end{document}D0, and model parameters \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta$$\end{document}β, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document}γ, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta$$\end{document}δ, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α. Compute \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _\alpha$$\end{document}μα. Initialization: Set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_0$$\end{document}S0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_0$$\end{document}I0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0$$\end{document}R0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D_0$$\end{document}D0, and model parameters \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta$$\end{document}β, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document}γ, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta$$\end{document}δ, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α. Compute \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _\alpha$$\end{document}μα. Time stepping: For \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n=1$$\end{document}n=1 to N: (i)Compute \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_\alpha (t_n) = \frac{1}{\alpha }(1 - e^{-\mu _\alpha t_n})$$\end{document}Cα(tn)=1α(1-e-μαtn).(ii)For each \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k=0$$\end{document}k=0 to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n-1$$\end{document}n-1, calculate the kernel \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e^{-\mu _\alpha (t_n - t_k)}$$\end{document}e-μα(tn-tk).(iii)Evaluate increments: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned} S_{k+1} - S_k&= -h \beta S_k I_k, \\ I_{k+1} - I_k&= h [\beta S_k I_k - (\gamma +\delta ) I_k], \\ R_{k+1} - R_k&= h \gamma I_k, \\ D_{k+1} - D_k&= h \delta I_k. \end{aligned} \end{aligned}$$\end{document}Sk+1-Sk=-hβSkIk,Ik+1-Ik=h[βSkIk-(γ+δ)Ik],Rk+1-Rk=hγIk,Dk+1-Dk=hδIk.(iv)Compute the sums: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned} \text {sumS}&= \sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)} [S_{k+1} - S_k], \\ \text {sumI}&= \sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)} [I_{k+1} - I_k], \\ \text {sumR}&= \sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)} [R_{k+1} - R_k], \\ \text {sumD}&= \sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)} [D_{k+1} - D_k]. \end{aligned} \end{aligned}$$\end{document}sumS=∑k=0n-1e-μα(tn-tk)[Sk+1-Sk],sumI=∑k=0n-1e-μα(tn-tk)[Ik+1-Ik],sumR=∑k=0n-1e-μα(tn-tk)[Rk+1-Rk],sumD=∑k=0n-1e-μα(tn-tk)[Dk+1-Dk].(v)Update: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned} S_n&= S_0 + \frac{1}{(1-\alpha )C_\alpha (t_n)} \text {sumS}, \\ I_n&= I_0 + \frac{1}{(1-\alpha )C_\alpha (t_n)} \text {sumI}, \\ R_n&= R_0 + \frac{1}{(1-\alpha )C_\alpha (t_n)} \text {sumR}, \\ D_n&= D_0 + \frac{1}{(1-\alpha )C_\alpha (t_n)} \text {sumD}. \end{aligned} \end{aligned}$$\end{document}Sn=S0+1(1-α)Cα(tn)sumS,In=I0+1(1-α)Cα(tn)sumI,Rn=R0+1(1-α)Cα(tn)sumR,Dn=D0+1(1-α)Cα(tn)sumD. Time stepping: For \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n=1$$\end{document}n=1 to N: (i)Compute \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_\alpha (t_n) = \frac{1}{\alpha }(1 - e^{-\mu _\alpha t_n})$$\end{document}Cα(tn)=1α(1-e-μαtn).(ii)For each \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k=0$$\end{document}k=0 to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n-1$$\end{document}n-1, calculate the kernel \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e^{-\mu _\alpha (t_n - t_k)}$$\end{document}e-μα(tn-tk).(iii)Evaluate increments: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned} S_{k+1} - S_k&= -h \beta S_k I_k, \\ I_{k+1} - I_k&= h [\beta S_k I_k - (\gamma +\delta ) I_k], \\ R_{k+1} - R_k&= h \gamma I_k, \\ D_{k+1} - D_k&= h \delta I_k. \end{aligned} \end{aligned}$$\end{document}Sk+1-Sk=-hβSkIk,Ik+1-Ik=h[βSkIk-(γ+δ)Ik],Rk+1-Rk=hγIk,Dk+1-Dk=hδIk.(iv)Compute the sums: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned} \text {sumS}&= \sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)} [S_{k+1} - S_k], \\ \text {sumI}&= \sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)} [I_{k+1} - I_k], \\ \text {sumR}&= \sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)} [R_{k+1} - R_k], \\ \text {sumD}&= \sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)} [D_{k+1} - D_k]. \end{aligned} \end{aligned}$$\end{document}sumS=∑k=0n-1e-μα(tn-tk)[Sk+1-Sk],sumI=∑k=0n-1e-μα(tn-tk)[Ik+1-Ik],sumR=∑k=0n-1e-μα(tn-tk)[Rk+1-Rk],sumD=∑k=0n-1e-μα(tn-tk)[Dk+1-Dk].(v)Update: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned} S_n&= S_0 + \frac{1}{(1-\alpha )C_\alpha (t_n)} \text {sumS}, \\ I_n&= I_0 + \frac{1}{(1-\alpha )C_\alpha (t_n)} \text {sumI}, \\ R_n&= R_0 + \frac{1}{(1-\alpha )C_\alpha (t_n)} \text {sumR}, \\ D_n&= D_0 + \frac{1}{(1-\alpha )C_\alpha (t_n)} \text {sumD}. \end{aligned} \end{aligned}$$\end{document}Sn=S0+1(1-α)Cα(tn)sumS,In=I0+1(1-α)Cα(tn)sumI,Rn=R0+1(1-α)Cα(tn)sumR,Dn=D0+1(1-α)Cα(tn)sumD. (i)Compute \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_\alpha (t_n) = \frac{1}{\alpha }(1 - e^{-\mu _\alpha t_n})$$\end{document}Cα(tn)=1α(1-e-μαtn). Compute \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_\alpha (t_n) = \frac{1}{\alpha }(1 - e^{-\mu _\alpha t_n})$$\end{document}Cα(tn)=1α(1-e-μαtn). (ii)For each \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k=0$$\end{document}k=0 to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n-1$$\end{document}n-1, calculate the kernel \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e^{-\mu _\alpha (t_n - t_k)}$$\end{document}e-μα(tn-tk). For each \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k=0$$\end{document}k=0 to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n-1$$\end{document}n-1, calculate the kernel \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e^{-\mu _\alpha (t_n - t_k)}$$\end{document}e-μα(tn-tk). (iii)Evaluate increments: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned} S_{k+1} - S_k&= -h \beta S_k I_k, \\ I_{k+1} - I_k&= h [\beta S_k I_k - (\gamma +\delta ) I_k], \\ R_{k+1} - R_k&= h \gamma I_k, \\ D_{k+1} - D_k&= h \delta I_k. \end{aligned} \end{aligned}$$\end{document}Sk+1-Sk=-hβSkIk,Ik+1-Ik=h[βSkIk-(γ+δ)Ik],Rk+1-Rk=hγIk,Dk+1-Dk=hδIk. Evaluate increments: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned} S_{k+1} - S_k&= -h \beta S_k I_k, \\ I_{k+1} - I_k&= h [\beta S_k I_k - (\gamma +\delta ) I_k], \\ R_{k+1} - R_k&= h \gamma I_k, \\ D_{k+1} - D_k&= h \delta I_k. \end{aligned} \end{aligned}$$\end{document}Sk+1-Sk=-hβSkIk,Ik+1-Ik=h[βSkIk-(γ+δ)Ik],Rk+1-Rk=hγIk,Dk+1-Dk=hδIk. (iv)Compute the sums: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned} \text {sumS}&= \sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)} [S_{k+1} - S_k], \\ \text {sumI}&= \sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)} [I_{k+1} - I_k], \\ \text {sumR}&= \sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)} [R_{k+1} - R_k], \\ \text {sumD}&= \sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)} [D_{k+1} - D_k]. \end{aligned} \end{aligned}$$\end{document}sumS=∑k=0n-1e-μα(tn-tk)[Sk+1-Sk],sumI=∑k=0n-1e-μα(tn-tk)[Ik+1-Ik],sumR=∑k=0n-1e-μα(tn-tk)[Rk+1-Rk],sumD=∑k=0n-1e-μα(tn-tk)[Dk+1-Dk]. Compute the sums: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned} \text {sumS}&= \sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)} [S_{k+1} - S_k], \\ \text {sumI}&= \sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)} [I_{k+1} - I_k], \\ \text {sumR}&= \sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)} [R_{k+1} - R_k], \\ \text {sumD}&= \sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)} [D_{k+1} - D_k]. \end{aligned} \end{aligned}$$\end{document}sumS=∑k=0n-1e-μα(tn-tk)[Sk+1-Sk],sumI=∑k=0n-1e-μα(tn-tk)[Ik+1-Ik],sumR=∑k=0n-1e-μα(tn-tk)[Rk+1-Rk],sumD=∑k=0n-1e-μα(tn-tk)[Dk+1-Dk]. (v)Update: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned} S_n&= S_0 + \frac{1}{(1-\alpha )C_\alpha (t_n)} \text {sumS}, \\ I_n&= I_0 + \frac{1}{(1-\alpha )C_\alpha (t_n)} \text {sumI}, \\ R_n&= R_0 + \frac{1}{(1-\alpha )C_\alpha (t_n)} \text {sumR}, \\ D_n&= D_0 + \frac{1}{(1-\alpha )C_\alpha (t_n)} \text {sumD}. \end{aligned} \end{aligned}$$\end{document}Sn=S0+1(1-α)Cα(tn)sumS,In=I0+1(1-α)Cα(tn)sumI,Rn=R0+1(1-α)Cα(tn)sumR,Dn=D0+1(1-α)Cα(tn)sumD. Update: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned} S_n&= S_0 + \frac{1}{(1-\alpha )C_\alpha (t_n)} \text {sumS}, \\ I_n&= I_0 + \frac{1}{(1-\alpha )C_\alpha (t_n)} \text {sumI}, \\ R_n&= R_0 + \frac{1}{(1-\alpha )C_\alpha (t_n)} \text {sumR}, \\ D_n&= D_0 + \frac{1}{(1-\alpha )C_\alpha (t_n)} \text {sumD}. \end{aligned} \end{aligned}$$\end{document}Sn=S0+1(1-α)Cα(tn)sumS,In=I0+1(1-α)Cα(tn)sumI,Rn=R0+1(1-α)Cα(tn)sumR,Dn=D0+1(1-α)Cα(tn)sumD. Repeat until \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n = N$$\end{document}n=N. Repeat until \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n = N$$\end{document}n=N. Local truncation error (LTE) Let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_n = nh$$\end{document}tn=nh and denote by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U(t)=(S,I,R,D)$$\end{document}U(t)=(S,I,R,D) the exact solution. For a generic component \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$u\in \{S,I,R,D\}$$\end{document}u∈{S,I,R,D} we write the NCF integral form\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} u(t_n) \;=\; u(0) \;+\; \frac{1}{(1-\alpha )C_\alpha (t_n)} \int _0^{t_n} e^{-\mu _\alpha (t_n-s)}\,u'(s)\,ds, \qquad \mu _\alpha =\tfrac{\alpha }{1-\alpha }. \end{aligned}$$\end{document}u(tn)=u(0)+1(1-α)Cα(tn)∫0ne-μα(tn-s)u′(s)ds,μα=α1-α.Our scheme evaluates the convolution by the left–rectangle rule and replaces \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$u^\prime (t_k)$$\end{document}u′(tk) by the model right–hand side \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f_u(U(t_k))$$\end{document}fu(U(tk)) (Sect. "Numerical scheme for the SIRD model"). Define the one-step local truncation error as follows:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \tau _u^n :=\; u(t_n) - u(0) - \frac{1}{(1-\alpha )C_\alpha (t_n)} \sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)}\!\int _{t_k}^{t_{k+1}}\!\!u^\prime (s)\,ds. \end{aligned}$$\end{document}τun:=u(tn)-u(0)-1(1-α)Cα(tn)∑k=0n-1e-μα(tn-tk)∫tkk+1u′(s)ds.Assuming \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$u\in C^2([0,T])$$\end{document}u∈C2([0,T]) and f locally Lipschitz, Taylor expansion of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$u^\prime$$\end{document}u′ on each \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$[t_k,t_{k+1}]$$\end{document}[tk,tk+1] gives\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \bigg |\int _{t_k}^{t_{k+1}}\!\!u^\prime (s)\,ds - h\,u^\prime (t_k)\bigg | \;\le \; \tfrac{1}{2}h^2\Vert u^{\prime \prime }\Vert _{L^\infty ([t_k,t_{k+1}])}, \end{aligned}$$\end{document}|∫tkk+1u′(s)ds-hu′(tk)|≤12h2‖u″‖L∞([tk,tk+1]),hence, using \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sum _{k}e^{-\mu _\alpha (t_n-t_k)}\le \sum _{m=1}^{\infty }e^{-\mu _\alpha mh} \le \frac{1}{1-e^{-\mu _\alpha h}}=O(h^{-1})$$\end{document}∑ke-μα(tn-tk)≤∑m=1∞e-μαmh≤11-e-μαh=O(h-1) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(1-\alpha )C_\alpha (t_n)\simeq t_n$$\end{document}(1-α)Cα(tn)≃tn as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_n\downarrow 0$$\end{document}tn↓0. We obtain the uniform bound\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} |\tau _u^n| \;\le \; C_\textrm{lte}\, h, \qquad 0\le t_n\le T, \end{aligned}$$\end{document}|τun|≤Clteh,0≤tn≤T,that is, the method is first–order consistent with constant \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_\textrm{lte}$$\end{document}Clte depending on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α, T, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Vert u^{\prime \prime }\Vert _\infty$$\end{document}‖u″‖∞ but not on h. Stability and convergence Set the scheme in vector form\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} U^n \;=\; U^0 \;+\; \frac{1}{(1-\alpha )C_\alpha (t_n)} \sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)}\big (U^{k+1}-U^{k}\big ), \quad U^{k+1}-U^k = h\,F(U^k), \end{aligned}$$\end{document}Un=U0+1(1-α)Cα(tn)∑k=0n-1e-μα(tn-tk)(Uk+1-Uk),Uk+1-Uk=hF(Uk),where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F(S,I,R,D)=(-\beta SI,\ \beta SI-(\gamma +\delta )I,\ \gamma I,\ \delta I)$$\end{document}F(S,I,R,D)=(-βSI,βSI-(γ+δ)I,γI,δI). Let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U(t_n)$$\end{document}U(tn) be the exact solution and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E^n:=U^n-U(t_n)$$\end{document}En:=Un-U(tn) the error. Using the Lipschitz property of F on the invariant set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$[0,N_0]^4$$\end{document}[0,N0]4 with constant L (Sect. "Theoretical analysis"), the error satisfies\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Vert E^n\Vert _\infty \;\le \; \frac{Lh}{(1-\alpha )C_\alpha (t_n)} \sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)} \Vert E^k\Vert _\infty \;+\; \frac{1}{(1-\alpha )C_\alpha (t_n)}\sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)}\Vert \tau ^k\Vert _\infty . \end{aligned}$$\end{document}‖En‖∞≤Lh(1-α)Cα(tn)∑k=0n-1e-μα(tn-tk)‖Ek‖∞+1(1-α)Cα(tn)∑k=0n-1e-μα(tn-tk)‖τk‖∞.The discrete weights are normalized: using \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_\alpha (t_n)=\frac{1}{\alpha }(1-e^{-\mu _\alpha t_n})$$\end{document}Cα(tn)=1α(1-e-μαtn),\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \frac{1}{(1-\alpha )C_\alpha (t_n)}\sum _{k=0}^{n-1} e^{-\mu _\alpha (t_n-t_k)} \;\le \; 1 + c_\alpha \,h, \end{aligned}$$\end{document}1(1-α)Cα(tn)∑k=0n-1e-μα(tn-tk)≤1+cαh,for some \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$c_\alpha =O(1)$$\end{document}cα=O(1) (Riemann–sum approximation of the exponential kernel). Applying a discrete Grönwall inequality and the consistency bound \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Vert \tau ^k\Vert _\infty \le C_\textrm{lte} h$$\end{document}‖τk‖∞≤Clteh yields\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Vert E^n\Vert _\infty \;\le \; C\, h, \qquad 0\le t_n\le T, \end{aligned}$$\end{document}‖En‖∞≤Ch,0≤tn≤T,i.e. the method is first–order convergent in the sup norm. The constant C depends on T, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α, and the Lipschitz bound for F, but is independent of h. Manufactured solution and observed order To verify the predicted order, we consider the scalar test \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\textrm{NCF}\,D_0^\alpha u)(t)=g(t)$$\end{document}(NCFD0αu)(t)=g(t) with the manufactured solution \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$u(t)=e^{\omega t}$$\end{document}u(t)=eωt (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega >0$$\end{document}ω>0). Direct calculation gives\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} (\textrm{NCF}\,D_0^\alpha e^{\omega t})(t) = \frac{\omega }{(1-\alpha )C_\alpha (t)(\mu _\alpha +\omega )} \big (e^{\omega t}-e^{-\mu _\alpha t}\big )=:g(t), \end{aligned}$$\end{document}(NCFD0αeωt)(t)=ω(1-α)Cα(t)(μα+ω)(eωt-e-μαt)=:g(t),so g is known analytically. We time–march the discrete NCF operator with step sizes \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h=\tfrac{T}{N}$$\end{document}h=TN, compute the maximum error \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E(h)=\max _n|u^n-u(t_n)|$$\end{document}E(h)=maxn|un-u(tn)|, and report the observed order \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\textrm{p}=\log _2(E(h)/E(h/2))$$\end{document}p=log2(E(h)/E(h/2)). In all runs, we observe \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\textrm{p}\approx 1$$\end{document}p≈1 for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha \in \{0.75,0.85,0.95\}$$\end{document}α∈{0.75,0.85,0.95}, confirming first–order convergence. Numerical simulations and discussion To comprehensively investigate the impact of memory effects and kernel normalization in epidemic modeling, we conducted a series of numerical experiments using the NCF fractional SIRD model. All simulations employ the discretization scheme detailed in Sect. "Numerical method", with model parameters and initial conditions set according to realistic epidemic scenarios. Unless otherwise stated, the infection rate was set to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta =0.4$$\end{document}β=0.4, the recovery rate to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma =0.15$$\end{document}γ=0.15, the rate of disease-induced mortality at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta =0.12$$\end{document}δ=0.12, and the initial populations as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S(0)=0.96$$\end{document}S(0)=0.96, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I(0)=0.02$$\end{document}I(0)=0.02, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R(0)=0.01$$\end{document}R(0)=0.01, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D(0)=0.01$$\end{document}D(0)=0.01. The final simulation time was \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T=20$$\end{document}T=20 days, with a time step \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h=0.05$$\end{document}h=0.05. The primary focus was on fractional orders \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha =0.95,\,0.80,\,0.75$$\end{document}α=0.95,0.80,0.75, representing a transition from nearly classical to strong memory regimes. Figure 2 presents the compartmental evolution for the classical SIRD model. Fig. 2The classical SIRD model 1. The classical SIRD model 1. To assess practical identifiability and quantify uncertainty, we conducted a synthetic parameter recovery experiment. We fixed ground truth parameters \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta ^\star =(\beta ^\star ,\gamma ^\star ,\delta ^\star ,\alpha ^\star )$$\end{document}θ⋆=(β⋆,γ⋆,δ⋆,α⋆), simulated model trajectories on a uniform time grid to obtain model-implied incidence, and then generated noisy synthetic observations by adding sampling noise to that series. Parameters \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta =(\beta ,\gamma ,\delta ,\alpha )$$\end{document}θ=(β,γ,δ,α) were estimated via constrained least squares with positivity constraints on rates and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha \in (0,1)$$\end{document}α∈(0,1). Uncertainty was quantified using a nonparametric bootstrap: we resampled the time series with replacement across B replicates, refitted the model, and reported percentile 95% confidence intervals for each parameter. Identifiability was examined using profile likelihood curves and the bootstrap correlation matrix. Across noise levels typical of aggregated surveillance data, the median recovered parameters were close to the truth, and the true values lay within the reported intervals, indicating practical identifiability of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\beta ,\gamma ,\delta ,\alpha )$$\end{document}(β,γ,δ,α) for the observation model and the considered horizon. A compact summary table (median estimate, 95% CI, relative error) and corresponding plots are provided in the supplement. Figure 3 illustrate the lowering \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α flattens and delays the infectious curve I(t), indicating slower transmission and a reduced peak. The susceptible class S(t) declines more gradually as infections propagate more slowly. The recovered R(t) and deceased D(t) trajectories rise more slowly under strong memory but approach levels comparable to the near-classical case as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha \rightarrow 1$$\end{document}α→1, with sharper transitions in that classical limit. Overall, the NCF memory term damps and spreads the outbreak over time. Fig. 3Variation of the infectious, susceptible, recovered, and deceased populations under different values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α. Variation of the infectious, susceptible, recovered, and deceased populations under different values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α. Figure 4 explore the peak infectious load decreases monotonically with smaller \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α and with lower infection rate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta$$\end{document}β; large \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α and high \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta$$\end{document}β recover the tall, steep peaks typical of classical SIRD dynamics. Complementarily, the susceptible peak surface shows rapid depletion for high \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\alpha ,\beta )$$\end{document}(α,β) combinations, whereas strong memory (small \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α) sustains higher S for longer. The recovered and deceased peak levels are likewise sensitive to both parameters: stronger memory shifts peaks later and broadens them, while higher \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta$$\end{document}β amplifies peak magnitudes. Fig. 4Peak values of Susceptible (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{peak}$$\end{document}Speak), infectious (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_{peak}$$\end{document}Ipeak), recovered (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_{peak}$$\end{document}Rpeak) and deceased (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D_{peak}$$\end{document}Dpeak) populations as functions of the fractional order \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α and infection rate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta$$\end{document}β. Peak values of Susceptible (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{peak}$$\end{document}Speak), infectious (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_{peak}$$\end{document}Ipeak), recovered (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_{peak}$$\end{document}Rpeak) and deceased (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D_{peak}$$\end{document}Dpeak) populations as functions of the fractional order \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α and infection rate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta$$\end{document}β. Figure 5 examines increasing the initial infection \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_0$$\end{document}I0 advances the timing of the epidemic peak, yet memory effects (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha <1$$\end{document}α<1) still flatten it. Changes in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_0$$\end{document}S0 primarily rescale the total population at risk while preserving qualitative behavior. Variations in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0$$\end{document}R0 or \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D_0$$\end{document}D0 modulate the early transient; however, the long-term shape and timing remain dominated by the memory parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α together with the transmission and removal rates \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\beta ,\gamma ,\delta )$$\end{document}(β,γ,δ). Fig. 5Effect of varying initial proportions of susceptible (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_0$$\end{document}S0), infectious (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_0$$\end{document}I0), recovered (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0$$\end{document}R0), and deceased (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D_0$$\end{document}D0) populations on the time evolution of S(t), I(t), R(t), and D(t), respectively. Effect of varying initial proportions of susceptible (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_0$$\end{document}S0), infectious (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_0$$\end{document}I0), recovered (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0$$\end{document}R0), and deceased (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D_0$$\end{document}D0) populations on the time evolution of S(t), I(t), R(t), and D(t), respectively. The simulations confirm that the use of the NCF derivative substantially modifies epidemic dynamics compared to the classical and unnormalized fractional models. In particular, the normalization ensures a physically consistent weighting of historical states, resulting in more interpretable and realistic predictions for diseases with incubation or latent periods. The NCF-SIRD model thus offers a flexible framework for exploring a wide range of memory effects in epidemiology, providing a valuable tool for both theoretical analysis and public health forecasting. A feedforward DNN was employed to approximate the dynamics of the NCF–SIRD model. The final architecture consisted of three hidden layers with 10, 100, and 10 neurons, utilizing sigmoid activation functions in the hidden layers and a linear activation function in the output layer to perform the regression task. This configuration provided the best performance among several tested architectures, achieving a mean squared error (MSE = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$5.6e-08$$\end{document}5.6e-08) and a root mean square error (RMSE = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2.4e-4$$\end{document}2.4e-4) across the training, validation, and testing datasets, and outperforming both shallower and deeper alternatives. The chosen structure is consistent with prior studies, which demonstrate that compact feedforward networks with two to four hidden layers can effectively capture nonlinear interactions in dynamical and epidemiological systems20–22. The wide middle layer (100 neurons) captures high-dimensional nonlinear features, while the narrower outer layers enhance feature compression and generalization. Sigmoid activations were adopted due to their smoothness and proven universal approximation capability24,25, yielding more stable convergence than ReLU when combined with the Levenberg–Marquardt optimization algorithm. The model was trained for 1000 epochs with a data split of 70%, 15%, and 15% for training, validation, and testing, respectively. Hyperparameters such as learning rate, batch size, and early stopping were tuned to ensure convergence and prevent overfitting. Model trajectories generated using the fractional Adams–Bashforth numerical scheme were used to maintain consistency between the simulated dynamics and DNN predictions. The integration of DNNs with the fractional Adams-Bashforth scheme significantly enhances the model’s analytical and predictive capabilities. By refining network weights and reducing residual errors, the DNN enables practical training, validation, and testing, allowing precise modeling of dynamic interactions within the system. Regression analyses and error metrics confirm the stability and convergence of this combined approach, highlighting the synergy between data-driven methods and fractional calculus. To provide intuitive insights into the model’s performance, we include several visual diagnostics. Time-series overlays compare predicted and observed case counts for key epidemiological compartments, illustrating strong concordance over the simulation period. Residual plots highlight the distribution and magnitude of prediction errors, confirming the model’s reliability and pinpointing areas for refinement. Where applicable, uncertainty bands are displayed to indicate confidence intervals around model predictions. All figures have been enhanced to meet or exceed 300 dpi resolution standards, with enlarged axis labels and legends to ensure readability in print and electronic formats. Furthermore, a schematic graphic summarizing the data flow, neural network pipeline, and epidemiological outputs has been added to facilitate understanding of the integrated methodology. Figure 6a shows the outcomes for the I(t) class evaluated through the DNN model. The regression coefficient is very close to one, with an MSE of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$5.6e-08$$\end{document}5.6e-08 and an RMSE of 0.00024. The histogram provides ME = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$-9.2e-06$$\end{document}-9.2e-06 and VE = 0.00024, which validate the effectiveness of the proposed approach. Figure 6b depicts the training performance for I(t), where the regression result approaches unity. The corresponding error values are MSE = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$3.7e-08$$\end{document}3.7e-08 and RMSE = 0.00019. Histogram results yield ME = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.4e-08$$\end{document}1.4e-08 and VE = 0.00019, confirming the robustness of the method. The testing results for I(t), illustrated in Fig. 6c, indicate a regression coefficient close to one, with MSE = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$7.8e-08$$\end{document}7.8e-08 and RMSE = 0.00028. Histogram evaluation yields ME = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$-1.7e-05$$\end{document}-1.7e-05 and VE = 0.00028, further supporting the accuracy of the model. Lastly, the validation analysis for I(t), as shown in Fig. 6d, achieves a regression coefficient of one. The associated errors are MSE = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.3e-07$$\end{document}1.3e-07 and RMSE = 0.00035, while the histogram values (ME = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$-4.5e-05$$\end{document}-4.5e-05, VE = 0.00035) reinforce the validity of the approach. Fig. 6Dynamic response of the class I(t) using the DNN: (a) complete dataset, (b) training subset, (c) testing subset, and (d) validation subset. Fig. 7Dynamic response of the class S(t) using the DNN: (a) complete dataset, (b) training subset, (c) testing subset, and (d) validation subset. Dynamic response of the class I(t) using the DNN: (a) complete dataset, (b) training subset, (c) testing subset, and (d) validation subset. Dynamic response of the class S(t) using the DNN: (a) complete dataset, (b) training subset, (c) testing subset, and (d) validation subset. Figure 7a presents the results for the S(t) class analyzed through the DNN model. The regression coefficient is nearly equal to one, with MSE and RMSE values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$5.6e-08$$\end{document}5.6e-08 and 0.00024. The histogram further reports ME = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$-9.2e-06$$\end{document}-9.2e-06 and VE = 0.00024, supporting the effectiveness of the proposed method. Figure 7b shows the training data for S(t), where the regression outcome is close to unity, with MSE = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$3.7e-08$$\end{document}3.7e-08 and RMSE = 0.00019. Histogram analysis yields ME = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.4e-08$$\end{document}1.4e-08 and VE = 0.00019, confirming the reliability of the approach. The testing results for S(t), given in Fig. 7c , produce a regression coefficient near one, with MSE = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$7.8e-08$$\end{document}7.8e-08 and RMSE = 0.00028. The histogram evaluation yields ME = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$-1.7e-05$$\end{document}-1.7e-05 and VE = 0.00028, further verifying the model’s accuracy. Finally, the validation phase for S(t), illustrated in Fig. 7d, achieves a regression coefficient of one, with MSE = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.3e-07$$\end{document}1.3e-07 and RMSE = 0.00035. The histogram shows ME = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$-4.5e-05$$\end{document}-4.5e-05 and VE = 0.00035, which reinforces the validity of the method. Fig. 8Dynamic response of the class D(t) using the DNN: (a) complete dataset, (b) training subset, (c) testing subset, and (d) validation subset. Dynamic response of the class D(t) using the DNN: (a) complete dataset, (b) training subset, (c) testing subset, and (d) validation subset. Figure 8a presents the results for the class D(t) obtained using the DNN model. The regression coefficient is found to be close to unity, with MSE and RMSE values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$5.6e-08$$\end{document}5.6e-08 and 0.00024. The histogram further reports ME and VE values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$-9.2e-06$$\end{document}-9.2e-06 and 0.00024, which support the effectiveness of the method. Figure 8b shows the training performance for D(t), where the regression outcome approaches one, with MSE = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$3.7e-08$$\end{document}3.7e-08 and RMSE = 0.00019. The histogram analysis yields ME = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.4e-08$$\end{document}1.4e-08 and VE = 0.00019, indicating the strong reliability of the approach. The testing outcomes for D(t), illustrated in Fig. 8c , yield a regression coefficient close to one, with MSE = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$7.8e-08$$\end{document}7.8e-08 and RMSE = 0.00028. Histogram evaluation results (ME = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$-1.7e-05$$\end{document}-1.7e-05, VE = 0.00028) further demonstrate the accuracy of the model. Finally, validation data for D(t), shown in Fig. 8d , confirm a regression coefficient of one, with an MSE of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.3e-07$$\end{document}1.3e-07 and RMSE = 0.00035. The histogram reports ME = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$-4.5e-05$$\end{document}-4.5e-05 and VE = 0.00035, reinforcing the validity of the method. Fig. 9Dynamic response of the class R(t) using the DNN: (a) complete dataset, (b) training subset, (c) testing subset, and (d) validation subset. Dynamic response of the class R(t) using the DNN: (a) complete dataset, (b) training subset, (c) testing subset, and (d) validation subset. Figure 9a displays the results for the class R(t) obtained through DNN analysis. The regression coefficient is nearly equal to one, with an MSE of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$5.6e-08$$\end{document}5.6e-08 and an RMSE of 0.00024. The histogram further reports ME and VE values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$-9.2e-06$$\end{document}-9.2e-06 and 0.00024, confirming the robustness of the proposed method. Figure 9b presents the training outcomes for R(t), showing a regression coefficient close to unity, with corresponding MSE and RMSE values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$3.7e-08$$\end{document}3.7e-08 and 0.00019. The histogram results (ME = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.4e-08$$\end{document}1.4e-08, VE = 0.00019) support the method’s reliability. The testing phase for R(t), illustrated in Fig. 9c , yields a regression coefficient approaching one, along with an MSE of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$7.8e-08$$\end{document}7.8e-08 and RMSE of 0.00028. Histogram evaluation yields ME and VE values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$-1.7e-05$$\end{document}-1.7e-05 and 0.00028, respectively, further verifying the model’s accuracy. Finally, the validation outcomes for R(t), shown in Fig. 9d , yield a regression coefficient of one, with MSE and RMSE values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.3e-07$$\end{document}1.3e-07 and 0.00035. Histogram results (ME = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$-4.5e-05$$\end{document}-4.5e-05, VE = 0.00035) reinforce the validity of the approach. Conclusion This work introduced a fractional SIRD model based on the NCF derivative and provided a rigorous analysis of existence, uniqueness, positivity, and boundedness. Kernel normalization clarifies the physical role of memory and eliminates inconsistencies present in unnormalized operators. We also developed a stable first-order numerical scheme. We demonstrated, through targeted simulations, that the NCF model can reproduce epidemic dynamics more realistically than both classical and unnormalized fractional counterparts. The combination of deep neural networks with the fractional SIRD model yielded an efficient approximation of epidemic trajectories and improved predictive accuracy. The methodology demonstrates that DNNs can effectively work with fractional models, as they combine the clear understanding of mathematical frameworks with the flexibility of data-driven learning methods. The integration of these methods creates a track toward developing real-time epidemic forecasting systems that unite theoretical models with empirical evidence. The present study adopts homogeneous mixing and omits vital dynamics (births and natural deaths), choices that help isolate memory effects but limit long-horizon realism. The framework admits natural extensions that remain compatible with our analytical results: demographic terms can be added without altering the NCF normalization or regularity requirements; heterogeneous contact structure can be introduced via age groups with a contact matrix or via network models on graphs (recovering homogeneous mixing in the dense, uniform limit); and the memory parameter can vary across subgroups or locations (e.g. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha _a$$\end{document}αa by age/region, or \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha _i$$\end{document}αi on networks) and even in time. Looking ahead, priorities include parameter estimation from real-world data with uncertainty quantification, exploration of generalized fractional operators, incorporation of stochastic effects, and the development of physics-informed neural networks for greater interpretability. These advances will further strengthen the utility of NCF-based fractional models for real-time epidemic forecasting and policy design.