Loading [MathJax]/jax/output/SVG/jax.js
Research article

The diffusion identification in a SIS reaction-diffusion system

  • Received: 22 August 2023 Revised: 24 November 2023 Accepted: 12 December 2023 Published: 15 December 2023
  • This article is concerned with the determination of the diffusion matrix in the reaction-diffusion mathematical model arising from the spread of an epidemic. The mathematical model that we consider is a susceptible-infected-susceptible model with diffusion, which was deduced by assuming the following hypotheses: The total population can be partitioned into susceptible and infected individuals; a healthy susceptible individual becomes infected through contact with an infected individual; there is no immunity, and infected individuals can become susceptible again; the spread of epidemics arises in a spatially heterogeneous environment; the susceptible and infected individuals implement strategies to avoid each other by staying away. The spread of the dynamics is governed by an initial boundary value problem for a reaction-diffusion system, where the model unknowns are the densities of susceptible and infected individuals and the boundary condition models the fact that there is neither emigration nor immigration through their boundary. The reaction consists of two terms modeling disease transmission and infection recovery, and the diffusion is a space-dependent full diffusion matrix. The determination of the diffusion matrix was conducted by considering that we have experimental data on the infective and susceptible densities at some fixed time and in the overall domain where the population lives. We reformulated the identification problem as an optimal control problem where the cost function is a regularized least squares function. The fundamental contributions of this article are the following: The existence of at least one solution to the optimization problem or, equivalently, the diffusion identification problem; the introduction of first-order necessary optimality conditions; and the necessary conditions that imply a local uniqueness result of the inverse problem. In addition, we considered two numerical examples for the case of parameter identification.

    Citation: Aníbal Coronel, Fernando Huancas, Ian Hess, Alex Tello. The diffusion identification in a SIS reaction-diffusion system[J]. Mathematical Biosciences and Engineering, 2024, 21(1): 562-581. doi: 10.3934/mbe.2024024

    Related Papers:

    [1] Liqiong Pu, Zhigui Lin . A diffusive SIS epidemic model in a heterogeneous and periodically evolvingenvironment. Mathematical Biosciences and Engineering, 2019, 16(4): 3094-3110. doi: 10.3934/mbe.2019153
    [2] Danfeng Pang, Yanni Xiao . The SIS model with diffusion of virus in the environment. Mathematical Biosciences and Engineering, 2019, 16(4): 2852-2874. doi: 10.3934/mbe.2019141
    [3] Sebastian Aniţa, Bedreddine Ainseba . Internal eradicability for an epidemiological model with diffusion. Mathematical Biosciences and Engineering, 2005, 2(3): 437-443. doi: 10.3934/mbe.2005.2.437
    [4] Wenzhang Huang, Maoan Han, Kaiyu Liu . Dynamics of an SIS reaction-diffusion epidemic model for disease transmission. Mathematical Biosciences and Engineering, 2010, 7(1): 51-66. doi: 10.3934/mbe.2010.7.51
    [5] Xin-You Meng, Tao Zhang . The impact of media on the spatiotemporal pattern dynamics of a reaction-diffusion epidemic model. Mathematical Biosciences and Engineering, 2020, 17(4): 4034-4047. doi: 10.3934/mbe.2020223
    [6] Jinzhe Suo, Bo Li . Analysis on a diffusive SIS epidemic system with linear source and frequency-dependent incidence function in a heterogeneous environment. Mathematical Biosciences and Engineering, 2020, 17(1): 418-441. doi: 10.3934/mbe.2020023
    [7] Lin Zhao, Haifeng Huo . Spatial propagation for a reaction-diffusion SI epidemic model with vertical transmission. Mathematical Biosciences and Engineering, 2021, 18(5): 6012-6033. doi: 10.3934/mbe.2021301
    [8] Yan’e Wang , Zhiguo Wang, Chengxia Lei . Asymptotic profile of endemic equilibrium to a diffusive epidemic model with saturated incidence rate. Mathematical Biosciences and Engineering, 2019, 16(5): 3885-3913. doi: 10.3934/mbe.2019192
    [9] Chichia Chiu, Jui-Ling Yu . An optimal adaptive time-stepping scheme for solving reaction-diffusion-chemotaxis systems. Mathematical Biosciences and Engineering, 2007, 4(2): 187-203. doi: 10.3934/mbe.2007.4.187
    [10] Raimund Bürger, Gerardo Chowell, Elvis Gavilán, Pep Mulet, Luis M. Villada . Numerical solution of a spatio-temporal predator-prey model with infected prey. Mathematical Biosciences and Engineering, 2019, 16(1): 438-473. doi: 10.3934/mbe.2019021
  • This article is concerned with the determination of the diffusion matrix in the reaction-diffusion mathematical model arising from the spread of an epidemic. The mathematical model that we consider is a susceptible-infected-susceptible model with diffusion, which was deduced by assuming the following hypotheses: The total population can be partitioned into susceptible and infected individuals; a healthy susceptible individual becomes infected through contact with an infected individual; there is no immunity, and infected individuals can become susceptible again; the spread of epidemics arises in a spatially heterogeneous environment; the susceptible and infected individuals implement strategies to avoid each other by staying away. The spread of the dynamics is governed by an initial boundary value problem for a reaction-diffusion system, where the model unknowns are the densities of susceptible and infected individuals and the boundary condition models the fact that there is neither emigration nor immigration through their boundary. The reaction consists of two terms modeling disease transmission and infection recovery, and the diffusion is a space-dependent full diffusion matrix. The determination of the diffusion matrix was conducted by considering that we have experimental data on the infective and susceptible densities at some fixed time and in the overall domain where the population lives. We reformulated the identification problem as an optimal control problem where the cost function is a regularized least squares function. The fundamental contributions of this article are the following: The existence of at least one solution to the optimization problem or, equivalently, the diffusion identification problem; the introduction of first-order necessary optimality conditions; and the necessary conditions that imply a local uniqueness result of the inverse problem. In addition, we considered two numerical examples for the case of parameter identification.



    The mathematical modeling of viral infectious disease transmission is a research area that has received the attention of several researchers in mathematical biology [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]. Particularly, we refer to [16] for a recent summary of challenges in modeling the dynamics of zoonotic infectious diseases. The increasing interest of mathematicians in researching the topic of biology is motivated by several aspects; among them, there is the rapid development of advanced technology and the generation of multidisciplinary teams to understand some complex problems that originated in human body behavior or in the interaction of humans with their environment. The current technology generates numerical datasets that require advanced mathematical tools for their analysis, then there are several mathematical models. The potential contributions of multidisciplinary teams were evidenced in the solution of some specific emergencies generated in the last few years, like the Ebola virus or the COVID-19 pandemic, where the knowledge of mathematicians and epidemiologists was fundamental to the development of health public policy. However, despite the technological development achieved in recent years, there are many aspects that are not observable or measurable by contemporary technology; and, thus, models are necessary to simulate, conjecture; or predict situations where technology is not capable.

    Although the current list of works dedicated to the modeling of virus transmission is voluminous, from the mathematical modeling cycle approach; we can distinguish some common steps that are used in the mathematical epidemiology [17]: the analysis of the experimental data for the precise disease; the election of the appropriate mathematical framework that models the quantitative data; the mathematical analysis to characterize the behavior of the model; the parameter calibration of the mathematical model; the validation model and improvement of the model if it is necessary, since the process is cyclic. Moreover, in the context of modeling, there are articles that are specialized or are interested in a single step of the cycle; for instance, the articles covering the topic of the well-posedness of the ordinary differential systems that are deduced as mathematical models. In this paper, we are interested in the calibration of a model of the dynamics of two populations during infectious diseases. Then, in order to present the identification problem, we begin by precisely stating the mathematical model.

    The epidemiological model considered in this article is based on partial differential equations, where the main variables model the dynamics of two populations: The population of healthy persons who are susceptible to contracting an infection and the population of infected persons who can transmit the disease to the population of healthy ones. More precisely, we consider the following initial boundary value problem:

    tSdiv(d1(x)S)=β(x)SIS+I+γ(x)I,in QT:=Ω×]0,T], (1.1)
    tIdiv(d2(x)I)=β(x)SIS+Iγ(x)I,in QT, (1.2)
    Sν=Iν=0,in ΓT:=Ω×]0,T], (1.3)
    (S,I)(x,0)=(S0,I0)(x),on Ω, (1.4)

    where S and I are the unknowns of the system, denoting the density of susceptible and infective populations; ΩRd (d=1,2,3) is the spatial domain where the total population lives; ν is the outward normal to the boundary Ω; β the transmission rate coefficient and γ is the recovery rate coefficient; dij,i,j=1,2, defined from Ω to R+, are the diffusion functions; and S0 and I0 denote the initial densities of susceptible and infective populations. In a broad sense, the system (1.1)–(1.4) is deduced considering the following assumptions: A population living in the spatial domain Ω is partitioned into two sets of individuals [18,19]: Susceptible and infective; the healthy susceptible individuals can contract the disease from cross contacts with infected ones (modeled by the term βSI/(S+I)); there is no immunity. In that sense, the infected individuals who are recovered can contract the diseases. That fact is modeled by the term γI; the spreading of the disease is influenced by the movement of individuals on the domain (modeled by the diffusion terms dij); and we consider that the boundary Ω is closed to emigration or immigration, which is modeled by the flux boundary condition (1.3).

    In order to precisely describe the problem and the main results of the paper, we present the function framework notation and the regularity assumptions on the domain, coefficients, and initial conditions of the problem (1.1)–(1.4). We consider the following function spaces that are standardly used in the analysis of parabolic equations [20,21,22]: Ck,α(¯Ω), kN, α]0,1] denotes the Hölder ktimes continuously differentiable functions whose kth-partial derivatives are Hölder continuous with exponent α; Lp(Ω), p1; denotes the space of all functions from Ω to R, which are p-integrable in the sense of Lebesgue; Wm,p(Ω) denotes the usual Sobolev spaces of functions that have weak derivatives up to order m and belong to Lp(Ω); Hm(Ω)=Wm,2(Ω) and Cα(¯Ω)=C0,α(¯Ω). Our analysis, in the present study, is conducted by considering the following set of assumptions:

    (D0) The spatial environment Ω is an open bounded and convex set with boundary Ω of C1 class.

    (D1) The functions S0 and I0 defining the initial conditions belonging to C2,α(¯Ω) and satisfying

    (S0,I0)(x)[0,Smax]×[0,Imax],ΩI0(x)dx>0,(S0+I0)(x)[ϕ0,[,

    on Ω, for some positive constant ϕ0;

    (D2) The coefficients of reaction have the regularity (β,γ)Cα(¯Ω) and (β,γ)(x)[b_,¯b]×[r_,¯r] on Ω for some b_,¯b,r_,¯r]0,1[.

    (D3) The functions Sobs and Iobs define the observation belonging to C2+α,1+α/2(Ω).

    We notice that the assumptions (D0)–(D2) are necessary to study the identification problem in the context of classical solutions of the direct problem (1.1)–(1.4), and more weak conditions can be considered to study the problem in the context of weak solutions; see, for instance, [23] for the particular case of the identity diffusion matrix.

    In this paper, we are interested in the model calibration, or specifically, the main aim is the identification of the diffusion matrix D=diag(d1,d2) from the observation of susceptible and infective population densities over all spatial domain Ω in a fixed time T, i.e., Sobs,Iobs:ΩR are known at time T. More precisely, let us consider the functions α,γ,S0,I0,Sobs and Iobs from Ω to R+ are known, and we want to determine D the solution of the following constrained optimization problem

    infDUad(Ω)J(D),J(D)=J(SD,ID), (1.5)
    subject to (SD,ID) solution of the system (1.1)-(1.4), (1.6)

    where J and Uad(Ω) are defined as follows

    J(S,I):=12S(,T)Sobs2L2(Ω)+12I(,T)Iobs2L2(Ω)+Γ2D2L2(Ω)2, (1.7)
    (1.8)
    (1.9)

    and Γ>0 is an appropriate constant. Hereinafter, we consider the notation D2L2(Ω)2=d12L2(Ω)+d22L2(Ω). We note that the functional J defined on (1.7) is the more pertinent for the determination of D, since the first and second terms of J are a comparison of the state solution profiles (S(,T),I(,T)) and the observation (Sobs,Iobs) in the L2 norm and the third term is a regularization term, where Γ should be appropriately selected in order to get a unique solution of the optimization problem.

    SIR (susceptible–infectious–removed) SIS (susceptible–infectious–susceptible)

    The analysis for the calibration of compartmental models (susceptible–infectious–susceptible and susceptible–infectious–removed shortly as SIS and SIR, respectively) was recently developed [17,19,23,24,25,26,27]. We observe that in all of those works, the authors identify the reaction coefficients; for instance, in [19] the authors get results for the identification of reaction term coefficients in the one-dimensional spatial domain (d=1), and it is extended to the higher dimensions (d2) in [17,24]. Moreover, in those works, the matrix modeling the diffusion is the identity matrix. However, to the best of our knowledge, the identification of matrix diffusion in epidemiological compartmental models has not yet been conducted. However, we must recognize that there are some works on the identification of the diffusion matrix in linear elliptic and parabolic problems whose results are not directly extensible to matrix diffusion identification in nonlinear systems of reaction-diffusion [28,29,30,31,32].

    The main results, which are the contributions of this paper, are given by the following five results: (ⅰ) The introduction of the necessary conditions to establish the existence and uniqueness of a positive solution to the direct problem (1.1)–(1.4) (see Section 2); (ⅱ) the existence of optimal solutions for (1.5)–(1.9) (see Section 3); (ⅲ) the introduction of an adjoint system with classical bounded solution (see Section 4); (ⅳ) the definition of a first-order optimality condition that characterizes the optimal solution in terms of direct and adjoint state solutions (see Section 5); and (ⅴ) a local uniqueness of identification problem (see Section 6). Furthermore, we present two numerical examples on and state some main conclusions (see Sections 7 and 8, respectively).

    Theorem 2.1. Consider that the hypotheses (D0)–(D2) are satisfied. If (d1,d2)Cα(¯Ω)2, there is a unique positive pair of functions (S,I)C2+α,1+α/2(¯QT)2 that satisfies the direct problem defined by the initial boundary value problem (1.1)–(1.4), which admits a unique positive classical solution (S,I). Moreover, S and I are bounded on ¯QT, i.e., the estimate

    S(,t)L(Ω)+I(,t)L(Ω)C,t[0,T]; (2.1)

    is satisfied for any given TR+.

    Proof. If we assume the existence of the solution of (1.1)–(1.4), we deduce some a priori estimates. We can prove the nonnegativity behavior of S and T by applying the maximum principle. From (1.1), (1.2), the positivity of S and T, the relation S/(S+I)<1; and the bounded behavior of reaction coefficients, we deduce that

    tSdiv(d1(x)S)¯rImax,in QT,tIdiv(d2(x)I)¯bImax,in QT,Sν=Iν=0,in ΓT,(S,I)(x,0)=(S0,I0),on Ω.

    Moreover, we can observe that (Smax,Imax) is a supersolution of the following linear system

    tWdiv(d1(x)W)=¯rImax,in QT,tZdiv(d2(x)Z)=¯bImax,in QT,Wν=Zν=0,in ΓT,(Z,W)(x,0)=(Smax,Imax),on Ω.

    Thus, the upper a priori estimate

    S(x,t)W(x,t)SmaxI(x,t)Z(x,t)Imaxon QT. (2.2)

    can be deduced by applying the well-known comparison principle for parabolic equations.

    We can follow the local existence of classical solutions of (1.1)–(1.4) by the standard results given in [33,34,35] and we can deduce the Hölder regularity of the local solution by modifying appropriately the arguments used in [18]. Thus, we have that there is a pair of nonnegativity functions (ˆS,ˆI)(x,t) that are the local solutions of (1.1)–(1.4); or, equivalently, there is Tmax>0 (the maximal existence time), such that (ˆS,ˆI)(x,t) is the following initial boundary value problem:

    tˆSdiv(d1(x)ˆS)=β(x)ˆSˆIˆS+ˆI+γ(x)ˆI,in QTmax:=Ω×]0,Tmax[, (2.3)
    tˆIdiv(d2(x)ˆI)=β(x)ˆSˆIˆS+ˆIγ(x)ˆI,in QTmax, (2.4)
    ˆSν=ˆIν=0,in ΓTmax:=Ω×]0,Tmax[, (2.5)
    (ˆS,ˆI)(x,0)=(S0,I0)(x),on Ω. (2.6)

    Consequently, the proof of existence and uniqueness of the global solution is reduced to guaranteeing the L estimations of ˆS and ˆT and applying similar arguments to those given in [18,36] (see also Theorem 2 in [37] and Lemma 1.1 in [38]).

    We observe that the positive constant p0 defined on the relation (1.6) of [37] should be selected such that p0>dmax{0,1}/2>3/2 [37, Theorem 1], then to obtain the L estimations is enough to obtain estimates in Lp0 for some p0>3/2. Indeed, we select p0=2; and we derive L2 estimations of (ˆS,ˆT). Multiplying (2.3) by ˆS, integrating on Ω; and (2.2), we have that

    12ddtΩˆS2dx+δ1_Ω|ˆS|2dx12ddtΩˆS2dx+Ωd1(x)|ˆS|2dxΩβ(x)ˆS2ˆIˆS+ˆIdx+Ωγ(x)ˆIˆSdx¯rΩˆIˆSdx¯rSmaxImax|Ω|. (2.7)

    Similarly, multiplying (2.4) by ˆI; integrating on Ω, and using the fact that ˆS/(ˆS+ˆI)1, we have that

    12ddtΩˆI2dx+δ2_Ω|ˆI|2dx12ddtΩˆI2dx+Ωd2(x)|ˆI|2dxΩβ(x)ˆSˆI2ˆS+ˆIdxΩγ(x)ˆIˆSdx¯b2ΩˆI2dx¯b2I2max|Ω|. (2.8)

    The estimates (2.7) and (2.8) implies that

    S(,s)2L2(Ω)+I(,s)2L2(Ω)S02L2(Ω)+I02L2(Ω)+(2¯rSmax+¯bImax)Imax|Ω|(S2max+2¯rSmaxImax+(1+¯b)I2max)|Ω|,s]0,Imax[,

    and, consequently, with the application of [37, Theorem 1], we deduce the existence and uniqueness of the global solution and, particularly, the estimate (2.1) is satisfied.

    Theorem 3.1. Consider the assumptions (D0)–(D3) are satisfied, then the optimization problem (1.5)–(1.9) has at least one solution.

    Proof. We note that the admissible set is not empty and J(D) is bounded for any DUad(Ω). The first assertion, i.e., Uad(Ω), follows by considering the diffusion matrix D(x)=diag(δ1_+¯δ1,δ2_+¯δ2)/2Uad(Ω). Meanwhile, we can prove that the cost function J is bounded by analyzing the boundedness of each term: The first two terms are bounded as consequence of the bounded behavior of the direct problem as result of Theorem 2.1, and the regularity of the observation functions is given on hypothesis (D3); the third term is bounded as consequence of the fact that DUad(Ω) and the definition of the admissible set. Consequently, we can deduce the existence of as a minimizing sequence of J, where M is a bounded and closed set of H|[d/2]|+1(Ω)2. We observe that the following compact embedding H|[d/2]|+1(Ω)Cα(Ω) is satisfied for all α]0,1/2] and the convexity of Ω is assumed on (D0). This kind of inclusion is the consequence of two results: H|[d/2]|+1(Ω) is continuous embedding in C1/2(Ω) (see Theorem 6 [39, pp 270]), and C1/2(Ω) is compact embedding in Cα(Ω) for all α]0,1/2], and Ω is a convex set (see Theorem 1.3.1 [40, pp 11]). Thus, clearly, H|[d/2]|+1(Ω)C1/2(Ω)Cα(Ω) for all α]0,1/2] implies that the embedding H|[d/2]|+1(Ω) in Cα(Ω) is compact for all α]0,1/2], and Ω is a convex set.

    The compact embedding of H|[d/2]|+1(Ω) in Cα(Ω) for α]0,1/2] and Ω convex, implies that {Dn} is bounded in the strong topology of Cα(¯Ω)2 for all α]0,1/2], since

    C>0:DnCα(¯Ω)2CDnH|[d/2]|+1(Ω)2,α]0,1/2],

    where C is independent of d1,d2 and n. Here, we remark that the righthand side is bounded by the fact that with M as a bounded and closed set of H|[d/2]|+1(Ω)2.

    Let us consider the notation (Sn,In) to the solution of the direct problem (1.1)–(1.4) corresponding to Dn, then, by considering the fact that {Dn}Cα(¯Ω)2 for all α]0,1/2], by Theorem 2.1, we have that (Sn,In)C2+α,1+α2(¯QT)2. Also {(Sn,In)} is a bounded sequence in the strong topology of C2+α,1+α2(¯QT)2 for all α]0,1/2].

    The boundedness of the sequence {(Dn,Sn,In)}, implies that there exists (¯D,¯S,¯T) such that

    ¯DC1/2(Ω)2Uad(Ω),(¯S,¯T)C2+12,1+14(¯QT)2;

    and uniformly convergent subsequences, which are again labeled by {Dn} and {(Sn,In)}; to be precise

    Dn¯Duniformly on Cα(Ω)2, (3.1)
    (Sn,In)(¯S,¯I)uniformly on [Cα,α2(¯QT)C2+α,1+α2(¯QT)]2. (3.2)

    Moreover, it is straightforward to deduce that (¯S,¯I) is the solution of (1.1)–(1.4) when the diffusion matrix is given by ¯D. Hence, using the definition of the minimizing sequence, the weak lower-semicontinuity of the L2 norm, and the Lebesgue's dominated convergence theorem, we get that

    J(¯D)limnJ(Dn)=infDUad(Ω)J(D). (3.3)

    Thus, ¯D is a solution of (1.5)–(1.9).

    In order to deduce the adjoint system, we adapt the formal calculus of the adjoint equation for scalar strongly parabolic equations in [41,42]. Let us consider L, the Lagrangian associated to the optimization problem (1.5)–(1.9), defined as follows

    L(S,I,p,q)=J(S,I)E1(S,I,p)E2(S,I,q) (4.1)

    where E1 and E2 are the weak formulations of (1.1) and (1.2), respectively. More precisely

    E1=T0Ω{S(pt+div(d1(x)p))β(x)SIS+Ip+γ(x)Ip}dxdt+Ω(Sp)(x,T)dxΩS0(x)p(x,0)dx+T0ΩSd1(x)pν(x,t)dσdt,E2=T0Ω{I(qt+div(d2(x)q))+β(x)SIS+Iqγ(x)Iq}dxdt+Ω(Sq)(x,T)dxΩI0(x)q(x,0)dx+T0ΩId2(x)qν(x,t)dσdt,

    for p and q the test functions.

    Let ¯D be a solution of optimization problem (1.5)–(1.9) and (¯S,¯I) be the solution of the forward problem (1.1)–(1.4) with ¯D instead of D. By a formal calculus of the derivative of L with respect to d1 and d2 and introducing the test functions (p,q), such that the derivatives of the state variables (¯S,¯I) with respect to d1 and d2 are vanished, we get that the functions (p,q) are obtained as the solution of the backward boundary value problem:

    pt+div(¯d1(x)p)=β(x)¯I2(¯S+¯I)2(pq),in QT, (4.2)
    qt+div(¯d2(x)q)=(β(x)¯S2(¯S+¯I)2γ(x))(pq),in QT, (4.3)
    pν=qν=0,on Γ, (4.4)
    (p,q)(x,T)=(¯S(x,T)Sobs(x),¯I(x,T)Iobs(x)),in Ω. (4.5)

    The system (4.2)–(4.5) is called the adjoint system to (1.1)–(1.4).

    Theorem 4.1. Consider that Ω,S0,I0,Sobs,Iobs,β and γ; satisfy the assumptions of Theorem 3.1. Moreover, consider that ¯DUad is a solution of (1.5)–(1.9); and (¯S,¯I) is the solution of the direct problem (1.1)–(1.4) with ¯D instead of D, then, the solution of (4.2)–(4.5) satisfies the following estimates

    p(,t)2L2(Ω)+q(,t)2L2(Ω)C, (4.6)
    p(,t)H10(Ω)+q(,t)H10(Ω)C, (4.7)
    Δp(,t)L2(Ω)+Δq(,t)L2(Ω)C, (4.8)
    p(,t)L(Ω)C,q(,t)L(Ω)C, (4.9)

    for t[0,T]. Here, C denotes some positive generic constant.

    Proof. If we introduce the change of the time variable by the following relation τ=Tt for t[0,T] and the unknowns of the direct problem and the adjoint system by the identity (w1,w2,S,I)(x,τ)=(p1,p2,¯S,¯I)(x,Tτ), we can rewrite the adjoint system (4.2)–(4.5) as the following initial boundary value problem

    (w1)τdiv(¯d1(x)w1)=β(x)(I)2(S+I)2(w1w2),in QT,(w2)τdiv(¯d2(x)w2)=(β(x)(S)2(S+I)2+γ(x))(w1w2),in QT,w1ν=w2ν=0,on Γ,(w1,w2)(x,0)=(¯S(x,T)Sobs(x),¯I(x,T)Iobs(x)),in Ω.

    Next, by applying the standard arguments of energy and regularity of solutions for linear parabolic equations, we get the desired estimates (4.6)–(4.9), see [19] for d=1 and [24] for d1 for the case of the identity matrix diffusion.

    Theorem 5.1. Consider that ¯D is a solution of the optimization problem (1.5)–(1.9), (¯S,¯I) is the solution of the direct problem (1.1)–(1.4) with ¯D instead of D, and (p,q) is the solution of the adjoint system (4.2)–(4.5), then, the inequality

    QTpdiv((ˆd1¯d1)(x)¯S)+qdiv((ˆd2¯d2)(x)¯I)dxdt+ΓΩ(¯d1(ˆd1¯d1))(x)+(¯d2(ˆd2¯d2))(x)dx0,ˆDUad(Ω); (5.1)

    is satisfied and defines the first-order optimality condition.

    Proof. Let us consider the arbitrary diffusion ˆDUad(Ω), then we define the notation

    Dε=(1ε)¯D+εˆDUad(Ω),Jε=J(Dε)=12Ω(|Sε(x,T)Sobs(x)|2+|Iε(x,T)Iobs(x)|2)dx+Γ2Ω(|dε1(x)|2+|dε2(x)|2)dx,

    where (Sε,Iε) is the solution of (1.1)–(1.4) with Dε instead of D. The fact that that ¯D is an optimal solution of (1.5)–(1.9), by taking the Fréchet derivative of Jε, we deduce that the following inequality

    dJεdε|ε=0=Ω(|Sε(x,T)Sobs(x)|Sεε|ε=0+|Iε(x,T)Iobs(x)|Iεε|ε=0)dx+ΓΩ(¯d1(ˆd1¯d1))(x)+(¯d2(ˆd2¯d2))(x)dx0, (5.2)

    is satisfied. Here, εSε and εIε for ε=0 are the sensitivities of solutions for (1.1)–(1.4), with respect to the ε-perturbations of ¯D.

    The calculus of the sensitivities (εSε,εIε) when ε0 is developed by considering the SIS systems of the form (1.1)–(1.4), (Sε,Iε) and (¯S,¯I), then letting ϵ0. More precisely, we have that (Sε,Iε) and (¯S,¯I) are solutions of the following initial boundary value problems

    (Sε)tdiv(dε1(x)Sε)=β(x)SεIεSε+Iε+γ(x)Iε,in QT, (5.3)
    (Iε)tdiv(dε2(x)Iε)=β(x)SεIεSε+Iεγ(x)Iε,in QT, (5.4)
    Sεν=Iεν=0,on Γ, (5.5)
    Sε(x,0)=S0(x),Iε(x,0)=I0(x),in Ω, (5.6)

    and

    (¯S)tdiv(¯d1(x)¯S)=β(x)¯S¯I¯S+¯I+γ(x)¯I,in QT, (5.7)
    (¯I)tdiv(¯d1(x)¯I)=β(x)¯S¯I¯S+¯Iγ(x)¯I,in QT, (5.8)
    ¯Sν=¯Iν=0,on Γ, (5.9)
    ¯S(x,0)=S0(x),¯I(x,0)=I0(x),in Ω, (5.10)

    respectively. Subtracting the system (5.7)–(5.10) from (5.3)–(5.6), dividing by ε and using the notation (zε1,zε2)=ε1(Sε¯S,Iε¯I), we deduce the initial boundary value problem

    (zε1)tdiv((ˆd1¯d1)(x)Sε+¯d1(x)zε1)=β(x)Sε¯S(SεIεSε+Iε¯SIε¯S+Iε)zε1β(x)Iε¯S(SεIεSε+Iε¯S¯I¯S+¯I)zε2+γ(x)zε2,in QT, (5.11)
    (zε2)tdiv((ˆd2¯d2)(x)Iε+¯d2(x)zε2)=β(x)Sε¯S(SεIεSε+Iε¯SIε¯S+Iε)zε1+β(x)Iε¯S(SεIεSε+Iε¯S¯I¯S+¯I)zε2γ(x)zε2,in QT, (5.12)
    zε1ν=zε2ν=0,on Γ, (5.13)
    zε1(x,0)=zε2(x,0)=0,in Ω. (5.14)

    Let us consider that (z1,z2) is the limit of (zε1,zε2) when ε0, from (5.11)–(5.14); we deduce straightforward answer that (z1,z2) is a solution of the following system

    (z1)tdiv((ˆd1¯d1)S+¯d1z1)=β(x)(¯S+¯I)2(¯I2z1+¯S2z2)+γ(x)z2,in QT, (5.15)
    (z2)tdiv((ˆd2¯d2)I+¯d2z2)=β(x)(¯S+¯I)2(¯I2z1+¯S2z2)γ(x)z2,in QT, (5.16)
    z1ν=z2ν=0,on Γ, (5.17)
    z1(x,0)=z2(x,0)=0,in Ω. (5.18)

    We remark that, in the context of optimization with partial differential equation constraints, the system (5.15)–(5.18) is called the sensitivity system for (1.1)–(1.4).

    Using the sensitivity system (5.15)–(5.18), we observe that the relation (5.2) can be rewritten as follows

    dJεdε|ε=0=Ω(|Sε(,T)Sobs|z1(,T)+|Iε(,T)Iobs|z2(,T))dx+ΓΩ(¯d1(ˆd1¯d1))(x)+(¯d2(ˆd2¯d2))(x)dx0. (5.19)

    Moreover, we notice two facts: First

    QTt(pz1+qz2)dxdt=Ω(p(x,T)z1(x,T)+q(x,T)z2(x,T))dx=Ω(|¯S(x,T)Sobs(x)|z1(x,T)+|¯I(x,T)Iobs(x)|z2(x,T))dx, (5.20)

    and second, by easy algebraic computations, from the systems (4.2)–(4.5) and (5.15)–(5.18), we can deduce the following identity

    t(pz1+qz2)=pdiv(¯d1(x)z1)+qdiv(¯d2(x)z2)z1div(¯d1(x)p)z2div(¯d2(x)q)+pdiv((ˆd1¯d1)(x)S)+qdiv((ˆd2¯d2)(x)I),

    which implies that

    QTt(pz1+qz2)dxdt=QTpdiv((ˆd1¯d1)(x)S)+qdiv((ˆd2¯d2)(x)I)dxdt, (5.21)

    by integration on QT. Thus, the relations (5.21) and (5.20) implies that

    QTpdiv((ˆd1¯d1)(x)S)+qdiv((ˆd2¯d2)(x)I)dxdt=Ω(|¯S(x,T)Sobs(x)|z1(x,T)+|¯I(x,T)Iobs(x)|z2(x,T))dx. (5.22)

    Hence, we can conclude the proof of (5.1) by replacing (5.22) in the first term of (5.19).

    Theorem 6.1. Let us consider that (D0)–(D3) is satisfied and let us consider the quotient set U(Ω)/ where the equivalence relation is defined as follows

    D1D2if and only if(D1D2)L(Ω)=0, (6.1)

    then, there exists ΓR+ such that the solution of the optimization problem (1.5)–(1.9) is uniquely defined (up an additive constant) on the quotient set U(Ω)/ for any regularization parameter Γ>Γ.

    Proof. Let us consider that D,ˆDU(Ω)/ are two solutions of (1.5)–(1.9). Moreover, let us consider that the sets of functions {S,I,p,q} and {ˆS,ˆI,ˆp,ˆq} are solutions to the systems (1.1)–(1.4) and (4.2)–(4.5) with diffusion matrices D and ˆD, respectively. From Theorem 5.1 and the hypothesis that D and ˆD are solutions of (1.5)–(1.9), we have that the following inequalities

    QTpdiv((¯¯d1d1)(x)S)+qdiv((¯¯d2d2)(x)I)dxdt+ΓΩ(d1(¯¯d1d1))(x)+(d2(¯¯d2d2))(x)dx0,¯¯DUad(Ω), (6.2)
    QTˆpdiv((d__1ˆd1)(x)ˆS)+ˆqdiv((d__2ˆd2)(x)ˆI)dxdt+ΓΩ(ˆd1(d__1ˆd1))(x)+(ˆd2(d__2ˆd2))(x)dx0,D__Uad(Ω), (6.3)

    are satisfied. If we choose the particular cases ¯¯D=ˆD in (6.2) and D__=D in (6.3), and then add both inequalities, we get

    Γ[ˆd1d12L2(Ω)+ˆd2d22L2(Ω)]QT{pdiv((ˆd1d1)(x)S)+qdiv((ˆd2d2)(x)I)+ˆpdiv((d1ˆd1)(x)S)+ˆqdiv((d2ˆd2)(x)I)}dxdt:=RHS. (6.4)

    We observe that, by integrating by parts two times, applying the Theorems (2.1) and (4.1), and using the fact that D,ˆDU(Ω)/, we can bound the righthand side of (6.4), as follows

    RHS=QT{Sdiv((ˆd1d1)p)ˆSdiv((ˆd1d1)ˆp)+Idiv((ˆd2d2)q)ˆIdiv((ˆd2d2)ˆq)}dxdt=QT{S(ˆd1d1)(x)ΔpˆS(ˆd1d1)(x)Δˆp+I(ˆd2d2)(x)ΔqˆI(ˆd2d2)(x)Δˆq+S(ˆd1d1)(x)pˆS(ˆd1d1)(x)ˆp+I(ˆd2d2)(x)qˆI(ˆd2d2)(x)ˆq}dxdt[SL(Ω)Δp2L2(Ω)+ˆSL(Ω)Δˆp2L2(Ω)]ˆd1d12L2(Ω)+[IL(Ω)Δq2L2(Ω)+ˆIL(Ω)Δˆq2L2(Ω)]ˆd2d22L2(Ω)+[S2L2(Ω)p2L2(Ω)+ˆS2L2(Ω)ˆp2L2(Ω)](ˆd1d1)L(Ω)+[I2L2(Ω)q2L2(Ω)+ˆI2L2(Ω)ˆq2L2(Ω)](ˆd2d2)L(Ω)Γ[ˆd1d12L2(Ω)+ˆd2d22L2(Ω)], (6.5)

    with

    Γ=max{SL(Ω)Δp2L2(Ω)+ˆSL(Ω)Δˆp2L2(Ω),IL(Ω)Δq2L2(Ω)+ˆIL(Ω)Δˆq2L2(Ω)}.

    Thus, from (6.5) and (6.4), we deduce the desired uniqueness result.

    In this section, we consider two numerical examples for the one-dimensional case where the identification is developed from observations that are constructed by considering synthetic data as observation of state variables. We begin by stating precisely that we introduce a small modification of the notation introduced previously. The diffusion coefficients d1 and d2 depend on a finite number of parameters denoted by e=(e1,,ek)Rk, which is explicitly denoted by di(x)=di(x;e) for i=1,2. The system (1.1)–(1.4) is modified by considering the mass action β(x)SI instead of β(x)SI/(S+I). We consider that kind of modification in the direct problem, since our aim is to apply the unconditionally stable Implicit-Explicit (IMEX) numerical method introduced by [43]. To be precise, the direct problem considered for the numerical simulations is given by the following initial boundary value problem:

    tSdiv(d1(x;e)S)=β(x)SI+γ(x)I,in QT, (7.1)
    tIdiv(d2(x;e)I)=β(x)SIγ(x)I,in QT, (7.2)
    Sν(0,t)=Sν(1,t)=0,in ΓT, (7.3)
    Iν(0,t)=Iν(1,t)=0,in ΓT, (7.4)
    (S,I)(x,0)=(S0,I0)(x),on Ω, (7.5)

    where Ω=]0,1[,Ω={0,1} and ΓT={0,1}×[0,T]. Concerning to the discretization of QT, we select M,NN such that the discretization of Ω is given by xk=kΔx for k=0,,M+1 with Δx=1/(M+1), and the discretization of [0,T] is given by tn=nΔt for n=0,,N with Δt=1/N. The approximation of a given function H:Ω×R+R at (xk,tn) is denoted by Hnk. Adapting the numerical discretization throughout a finite differences scheme introduced in [43], we deduce that the approximation of the initial boundary value problem (7.1)–(7.5) is given by

    Sn+1kSnkΔt=1Δx2[d1(xk;e)Sn+1k+1(d1(xk;e)+d1(xk1;e))Sn+1k+d1(xk1;e)Sn+1k1]β(xk)Sn+1kInk+γ(xk)In+1k, (7.6)
    In+1kInkΔt=1Δx2[d2(xk;e)In+1k+1(d2(xk;e)+d2(xk1;e))In+1k+d2(xk1;e)In+1k1]+β(xk)Sn+1kInkγ(xk)In+1k, (7.7)
    Sn1Sn0Δx=SnM+1SnMΔx=In1In0Δx=InM+1InMΔx=0, (7.8)
    S0k=S0(xk),I0k=I0(xk). (7.9)

    We remark that the numerical method considered in [43] is developed and analized when d1 and d2 are constants. However, by straightforward adaptation of the arguments given in [43], we can deduce that the implicit–explicit numerical scheme (7.6)–(7.9) has several properties, basically preserves the biological meaning (such as positivity), and is unconditionally convergent.

    For discretization of the inverse problem (1.5) and (1.6), we begin by considering the discretized cost function

    JΔ(SΔ,IΔ):=Δx2Mk=1(SNkSobsk)2+Δx2Mk=1(INkIobsk)2. (7.10)

    We observe that in (7.10) we have omitted the regularization term, i.e. Γ=0, then, in our numerical example, we consider that the inverse problem (1.5) and (1.6), is replaced by the following parameter identification problem

    infeRnJΔ(e),JΔ(e)=JΔ(SΔ,IΔ), (7.11)
    subject to (SΔ,IΔ) solution of (7.6)-(7.9). (7.12)

    In both numerical examples, we solve the optimization problem using the optimset routine of Matlab.

    We select the following coefficients on the reaction term β(x)=0.000284535 and γ(x)=0.144 and the initial condition (S0,I0)(x)=(x,2x)/2. We consider that e=(e1,e2), d1(x;e)=e1 and d2(x;e)=e2. We construct the observation profile at T=0.6 by considering a numerical simulation of the direct problem with eobs=(0.5,0.5), M=200 and N=100000 (i.e., Δx=5E3 and Δx=6E6). The state simulation on QT is shown on Figure 1(a), (b). We consider the initial guess eobs=(0.1,0.1) and get that the identified parameters are e=(0.52294,0.55149). The numerical identification is developed by considering M=100 and N=1000 or, equivalently, Δx=1.0E2 and Δx=5.¯9E4. The comparison of the observed, identified and initial guess profiles are shown in Figure 1(c)(f).

    Figure 1.  Numerical results for Example 1 given in section 7.1. In (a) and (b) we show the numerical solution. In (c) and (d) we show the comparison of inital guess, observed and identified profiles at T=0.6 for suceptibles and infective functions. In (d) and (e) we show the comparison observed and identified profiles at T=0.6 for suceptibles and infective functions.

    In this numerical example, we consider β(x)=0.000284535 and γ(x)=0.144 and the initial condition S0(x)=5 and

    I0(x)={0,x0.3,100000x30000,0.3<x0.5,100000x+70000,0.5<x0.7,0,otherwise.

    The parameters to identify are given by eR4, such that the diffusion functions are of the following parametric form

    d1(x;e)=0.1+e1x+e2x2,d2(x;e)=0.2+e3x+e4x2. (7.13)

    The independent terms in d1 and d2 are fixed to prevent the degeneration of the diffusion function. The observation profiles T=0.6 are constructed by a numerical simulation of the direct problem with eobs=(0.5,0.5,0.5,0.5), M=200 (Δx=5E3) and N=100000 (Δx=6E6), which are shown in Figure 2(a), (b). The initial guess and identified parameters are given by eobs=(0.1,0.1,0.1,0.1) and e=(0.75143,0.42256,0.95842,0.21146), respectively. For the identification, we assume that the discretization M=100 (Δx=1.0E2) and N=1000 (Δx=5.¯9E4). The comparison of observed, identified and initial guess profiles and diffusion functions are shown in Figure 2(c)(f).

    Figure 2.  Numerical results for Example 1 given in section 7.2. In (a) and (b) we show the numerical solution. In (c) and (d) we show the comparison of inital guess, observed and identified profiles at T=0.6 for suceptibles and infective functions. In (d) and (e) we show the comparison observed, initial guess and identified diffusion of the parametric form given in (7.13).

    From Figure 2(c), (e), we observe that the profile Se(,0.6) fits the observation data. However, d1(x,e) is close to d1(x,eobs), but we conjecture that it can be improved by incorporating the regularization term on JΔ. A similar behavior is observed from Figure 2(d), (f) for the cases of Ie(,0.6) and d2(x,e).

    In this paper, we have introduced the functional framework to develop the identification of the diffusion matrix in a reaction-diffusion system arising from the modeling of the spread dynamics of virus propagation. We considered that the disease occurs in a spatially distributed population between two classes of individuals: The susceptible class, formed by the individuals who can catch the disease; and the infective class, formed by the individuals who are infected and can transmit the disease. The reaction-diffusion model was deduced by assuming that there are no vital dynamics, there is no migration or immigration during the epidemic disease propagation, and the coefficients (diffusion, transmission rate, and recovery rate) of the model are functions that depend on the spatial position. The diffusion matrix identification was developed by assuming that the susceptible and infective populations are known at a fixed time. Thus, we have formulated the inverse problem as an optimal control problem, where the cost function to minimize is the least squares cost function and a regularization term, and the optimization constraints are the SIS reaction-diffusion model.

    We have proved that the mathematical model is well-posed and has a global positive solution in the context of strong solutions in Hölder spaces when the initial conditions and coefficients are of Hölder class. We have demonstrated that there exists at least one solution to the identification problem, and the solution is unique under the assumption that the regularization parameter is large enough. Furthermore, we remarked that the uniqueness of the optimal control problem is deduced from an appropriate quotient set of the admissible set. Moreover, we have introduced a necessary optimal for the optimal control problem.

    We observed that our results define the appropriate framework to develop the numerical identification from available experimental data for a concrete epidemic propagation and even numerical analysis like convergence. Here, we can remark that in a practical epidemiology phenomenon, the observation data of profiles in all spatial domain are not usual, and the typical situation is a profile at a fixed point in the domain and during a time interval. The cost function considered in this paper must be modified to develop identification from experimental data. Moreover, in the case of parameter identification from laboratory or epidemic data, we should develop a study of the noise, and we can consider the recent uncertainty concepts [44].

    The authors declare they have not used Artificial Intelligence (AI) tools in the creation of this article.

    A. C. and F. H. thank the partial support of Universidad del Bío-Bío (Chile) through the research projects: 2120436 IF/R, research project INES I+D 22-14, and FAPEI, and the research project of the Postdoctoral Program as a part of the project "Instalación del Plan Plurianual UBB 2016-2020"; National Agency for Research and Development, ANID-Chile, through FONDECYT project 1230560; Universidad Tecnológica Metropolitana through the project supported by the Competition for Research Regular Projects, year 2020, Code LPR20-06, Universidad Tecnológica Metropolitana. A. T. and I. H. acknowledge the partial support of "ayudantes de investigación" at Universidad del Bío-Bío (Chile) and ANID (Chile) through the program "Becas de Doctorado".

    The authors declare there is no conflict of interest.



    [1] R. M. Anderson, R. M. May, B. Anderson, Infectious Diseases of Humans: Dynamics and Control, Oxford University Press, 1992.
    [2] J. D. Murray, Mathematical Biology: Ⅰ. An Introduction, Springer, 2002. https://doi.org/10.1007/b98868
    [3] J. D. Murray, Mathematical Biology: Ⅱ. Spatial Models and Biomedical Applications, Springer, 2003.
    [4] N. Bacaër, A Short History of Mathematical Population Dynamics, Springer Science and Business Media, 2011.
    [5] O. Diekmann, J. A. P. Heesterbeek, Mathematical Epidemiology of Infectious Diseases: Model Building, Analysis and Interpretation, John Wiley & Sons, 2000.
    [6] V. Akimenko, An age-structured SIR epidemic model with fixed incubation period of infection, Comput. Math. Appl., 73 (2017), 1485–1504. https://doi.org/10.1016/j.camwa.2017.01.022 doi: 10.1016/j.camwa.2017.01.022
    [7] B. Armbruster, E. Beck, An elementary proof of convergence to the mean-field equations for an epidemic model, IMA J. Appl. Math., 82 (2017), 152–157. https://doi.org/10.1093/imamat/hxw010 doi: 10.1093/imamat/hxw010
    [8] J. Ge, L. Lin, L. Zhang, A diffusive SIS epidemic model incorporating the media coverage impact in the heterogeneous environment, Discrete Contin. Dyn. Syst. Ser. B, 22 (2017), 2763–2776. https://doi.org/10.3934/dcdsb.2017134 doi: 10.3934/dcdsb.2017134
    [9] Q. Ge, Z. Li, Z. Teng, Probability analysis of a stochastic SIS epidemic model, Stochastics Dyn., 17 (2017), 1750041. https://doi.org/10.1142/S0219493717500411 doi: 10.1142/S0219493717500411
    [10] M. Koivu-Jolma, A. Annila, Epidemic as a natural process, Math. Biosci., 299 (2018), 97–102. https://doi.org/10.1016/j.mbs.2018.03.012 doi: 10.1016/j.mbs.2018.03.012
    [11] X. Lu, S. Wang, S. Liu, J. Li, An SEI infection model incorporating media impact, Math. Biosci. Eng., 14 (2017), 1317–1335. https://doi.org/10.3934/mbe.2017068 doi: 10.3934/mbe.2017068
    [12] A. Nwankwo, D. Okuonghae, Mathematical analysis of the transmission dynamics of HIV syphilis co-infection in the presence of treatment for syphilis, Bull. Math. Biol., 80 (2018), 437–492. https://doi.org/10.1007/s11538-017-0384-0 doi: 10.1007/s11538-017-0384-0
    [13] C. M. Saad-Roy, P. Van den Driessche, A. A. Yakubu, A mathematical model of anthrax transmission in animal populations, Bull. Math. Biol., 79 (2017), 303–324. https://doi.org/10.1007/s11538-016-0238-1 doi: 10.1007/s11538-016-0238-1
    [14] V. M. Veliov, Numerical approximations in optimal control of a class of heterogeneous systems, Comput. Math. Appl., 70 (2015), 2652–2660. https://doi.org/10.1016/j.camwa.2015.04.029 doi: 10.1016/j.camwa.2015.04.029
    [15] A. Widder, C. Kuehn, Heterogeneous population dynamics and scaling laws near epidemic outbreaks, Math. Biosci. Eng., 13 (2016), 1093–1118. https://doi.org/10.3934/mbe.2016032 doi: 10.3934/mbe.2016032
    [16] M. Roberts, A. Dobson, O. Restif, K. Wells, Challenges in modeling the dynamics of infectious diseases at the wildlife-human interface, Epidemics, 37 (2021), 100523. https://doi.org/10.1016/j.epidem.2021.100523 doi: 10.1016/j.epidem.2021.100523
    [17] A. Coronel, L. Friz, I. Hess, M. Zegarra, On the existence and uniqueness of an inverse problem in epidemiology, Appl. Anal., 3 (2021), 513–526. https://doi.org/10.1080/00036811.2019.1608964 doi: 10.1080/00036811.2019.1608964
    [18] W. E. Fitzgibbon, M. Langlais, J. J. Morgan, A mathematical model for indirectly transmitted diseases, Math. Biosci., 206 (2007), 233–248. https://doi.org/10.1016/j.mbs.2005.07.005 doi: 10.1016/j.mbs.2005.07.005
    [19] X. Huili, L. Bin, Solving the inverse problem of an SIS epidemic reaction-diffusion model by optimal control methods, Comput. Math. Appl., 70 (2015), 805–819. https://doi.org/10.1016/j.camwa.2015.05.025 doi: 10.1016/j.camwa.2015.05.025
    [20] N. V. Krylov, Lectures on Elliptic and Parabolic Equations in Sobolev Spaces, American Mathematical Society, 2008.
    [21] O. A. Ladyzhenskaya, V. A. Solonnikov, N. N. Ural´ceva, Linear and Quasi-Linear Equations of Parabolic Type, American Mathematical Society, 1968.
    [22] G. M. Lieberman, Second Order Parabolic Differential Equations, World Scientific Publishing Co., 1996.
    [23] A. Coronel, F. Huancas, E. Lozada, M. Rojas-Medar, Results for a control problem for a SIS epidemic reaction-diffusion model, Symmetry 15 (2023), 1–14. https://doi.org/10.3390/sym15061224 doi: 10.3390/sym15061224
    [24] A. Coronel, F. Huancas, M. Sepúlveda, A note on the existence and stability of an inverse problem for a SIS model, Comput. Math. Appl., 77 (219), 3186–3194. https://doi.org/10.1016/j.camwa.2019.01.031 doi: 10.1016/j.camwa.2019.01.031
    [25] A. Coronel, F. Huancas, M. Sepúlveda, Identification of space distributed coefficients in an indirectly transmitted diseases model, Inverse Prob., 11 (2019), 115001. https://doi.org/10.1088/1361-6420/ab3a86 doi: 10.1088/1361-6420/ab3a86
    [26] A. E. Laaroussi, M. Rachik, M. Elhia, An optimal control problem for a spatiotemporal SIR model, Int. J. Dyn. Control, 6 (2018), 384–397. https://doi.org/10.1007/s40435-016-0283-5 doi: 10.1007/s40435-016-0283-5
    [27] K. Adnaoui, I. Elberrai, A. E. A. Laaroussi, K. Hattaf, A spatiotemporal SIR epidemic model two-dimensional with problem of optimal control, Bol. Soc. Parana. Mat., 40 (2022), 18. https://doi.org/10.5269/bspm.51110 doi: 10.5269/bspm.51110
    [28] M. Hinze, T. N. T. Quyen, Matrix coefficient identification in an elliptic equation with the convex energy functional method, Inverse Prob., 32 (2016), 085007. https://doi.org/10.1088/0266-5611/32/8/085007 doi: 10.1088/0266-5611/32/8/085007
    [29] S. Mondal, M. T. Nair, Identification of matrix diffusion coefficient in a parabolic PDE, Comput. Methods Appl. Math., 22 (2022), 413–441. https://doi.org/10.1515/cmam-2021-0061 doi: 10.1515/cmam-2021-0061
    [30] R. V. Kohn, B. D. Lowe, A variational method for parameter identification, RAIRO Model. Math. Anal. Numer., 22 (1988), 119–158. https://doi.org/10.1051/m2an/1988220101191 doi: 10.1051/m2an/1988220101191
    [31] M. S. Gockenbach, A. A. Khan, An abstract framework for elliptic inverse problems. Part 1: An output least-squares approach, Math. Mech. Solids, 12 (2005), 259–276. https://doi.org/10.1177/1081286505055758 doi: 10.1177/1081286505055758
    [32] B. Jadamba, A. A. Khan, M. Sama, Inverse problems on parameter identification in partial differential equations, in Mathematical Methods, Models and Algorithms in Science and Technology, World Scientific Publ., (2011), 228–258. https://doi.org/10.1142/9789814338820_0009
    [33] H. Amann, Global existence for semilinear parabolic problems, J. Reine Angew. Math., 360 (1985), 47–83. https://doi.org/10.1515/crll.1985.360.47 doi: 10.1515/crll.1985.360.47
    [34] D. Henry, Geometric Theory of Semilinear Parabolic Equations, Springer, 1981. https://doi.org/10.1007/BFb0089647
    [35] F. Rothe, Global Solutions of Reaction-Diffusion Systems, Springer, 1984. https://doi.org/10.1007/BFb0099278
    [36] Z. Du, R. Peng, A priori L estimates for solutions of a class of reaction-diffusion systems, J. Math. Biol., 72 (2016), 1429–1439. https://doi.org/10.1007/s00285-015-0914-z doi: 10.1007/s00285-015-0914-z
    [37] L. Dung, Dissipativity and global attractors for a class of quasilinear parabolic systems, Commun. Partial Differ. Equations, 22 (1997), 413–433. https://doi.org/10.1080/03605309708821269 doi: 10.1080/03605309708821269
    [38] M. Pierre, Global existence in reaction-diffusion systems with control of mass: A survey, Milan J. Math., 78 (2010), 417–455. https://doi.org/10.1007/s00032-010-0133-4 doi: 10.1007/s00032-010-0133-4
    [39] L. C. Evans, Partial Differential Equations, American Mathematical Society, 1997.
    [40] R. A. Adams, Sobolev Spaces, Academic Press, 1975.
    [41] S. Berres, R. Bürger, A. Coronel, M. Sepúlveda, Numerical identification of parameters for a strongly degenerate convection-diffusion problem modeling centrifugation of flocculated suspensions, Appl. Numer. Math., 52 (2005), 311–337. https://doi.org/10.1016/j.apnum.2004.08.002 doi: 10.1016/j.apnum.2004.08.002
    [42] A. Coronel, F. James, M. Sepúlveda, Numerical identification of parameters for a model of sedimentation processes, Inverse Prob., 19 (2003), 951–972. https://doi.org/10.1088/0266-5611/19/4/311 doi: 10.1088/0266-5611/19/4/311
    [43] X. Liu, Z. W. Yang, Numerical analysis of a reaction-diffusion susceptible-infected-susceptible epidemic model, Comput. Appl. Math., 41 (2022), 392. https://doi.org/10.1007/s40314-022-02113-9 doi: 10.1007/s40314-022-02113-9
    [44] B. Jadamba, A. A. Khan, M. Sama, H. J. Starkloff, C. Tammer, A convex optimization framework for the inverse problem of identifying a random parameter in a stochastic partial differential equation, SIAM/ASA J., 9 (2021), 119–158. https://doi.org/10.1137/20M1323953 doi: 10.1137/20M1323953
  • This article has been cited by:

    1. Jie Ding, Ziheng Xu, Shenggao Zhou, Structure-preserving numerical methods for disease transmission with chemotaxis of singular sensitivity, 2024, 0, 1531-3492, 0, 10.3934/dcdsb.2024184
  • Reader Comments
  • © 2024 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(1642) PDF downloads(114) Cited by(1)

Figures and Tables

Figures(2)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog