mathematics
Article
Effects of Media Coverage on Global Stability Analysis and
Optimal Control of an Age-Structured Epidemic Model with
Multi-Staged Progression
Chao Liu 1,2, * , Peng Chen 1,2 , Qiyu Jia 3 and Lora Cheung 4
Citation: Liu, C.; Chen, P.; Jia, Q.;
Cheung, L. Effects of Media
Coverage on Global Stability
Analysis and Optimal Control of an
Age-Structured Epidemic Model with
Multi-Staged Progression.
Mathematics 2022, 10, 2712. https://
doi.org/10.3390/math10152712
Academic Editors: Guiquan Sun
and Sanling Yuan
Received: 30 June 2022
Accepted: 28 July 2022
Published: 1 August 2022
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2022 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Mathematics 2022, 10, 2712. https://doi.org/10.3390/math10152712
1 Institute of Systems Science, Northeastern University, Shenyang 110169, China; cp971121@sina.com
2 School of Mathematics and Statistics, Northeastern University at Qinhuangdao, Qinhuangdao 066004, China
3 Sydney Smart Technology College, Northeastern University at Qinhuangdao, Qinhuangdao 066004, China;
k1875796562@163.com
4 Department of Mathematics and Statistics, York University, Toronto, ON M3J 1P3, Canada;
cheulora7@gmail.com
* Correspondence: liuchao@mail.neu.edu.cn
Abstract: In this paper, a hybrid SEIAM model for infectious disease with a continuous age struc-
ture is established, where combined dynamic effects of media coverage and multi-staged infected
progression on threshold dynamics are discussed. Sufficient conditions for uniform persistence
of the solution are studied by using the basic reproduction number. By constructing appropriate
Lyapunov functions, the global stability analysis of endemic equilibrium is investigated based on
Lyapunov–LaSalle’s stability theorem. In order to minimize costs incurred due to applied controls
and infectious disease burden, an optimal cost-effective control strategy associated with disease treat-
ment and media coverage is discussed. Numerical simulations are carried out to show consistency
with theoretical analysis.
Keywords: media coverage; multi-staged infected progression; uniform persistence; global stability
analysis; optimal control strategy
MSC: 34K34; 92D25
1. Introduction
According to some recent epidemiological data collected in medical observations,
it is found that parasite levels in vectors or viral loads in an infected population have a
great influence on disease infectivity [1], and not all of the infected population behave in a
homogeneous way during the entire infectivity period [2–4]. Some infectious diseases with
a comparatively long infectivity period, such as HIV [1], bilharzia [2], HBV and HCV [3,4],
usually exhibit various infecting forces at successive discrete stages of infection. Some
staged progression models are proposed to depict disease progression with typical varying
characters in infectivity transmission [5–7]. For epidemic models with multiple infectious
stages and general distributions for the stage durations (exponential distribution [5] and
Gamma distribution [6] as special cases), it is shown that disease dynamics are determined
by the derived basic reproduction number. The global threshold dynamics and uniform
persistence of stage-processed disease models are discussed based on the basic reproduction
number in [7].
In recent decades, many research efforts have been made to discuss the potential
impact of age on disease progression, and some age-structured models have also been es-
tablished to depict the vaccination or latency phenomena in the studied epidemiology [8,9].
Although these mathematical models are usually composed of partial differential equations
and corresponding dynamical analysis is particularly challenging, global stability for some
type of age-structured models have been investigated by constructing Lyapunov functionals
https://www.mdpi.com/journal/mathematics
Mathematics 2022, 10, 2712 2 of 28

in [10–12] and the references therein. Complex bifurcation phenomenon stability analyses
are discussed in some multi-group epidemic models with age structure [13–16] and the
references therein. A novel multi-group SVEIAR epidemic model with ages of vaccination
and latency is formulated in [17], where the sufficient conditions for the occurrence of
backward bifurcation in epidemic models are investigated. In order to discuss the possible
effects of the variable infected population in each stage of disease dynamics, some epidemic
model multi-age structures are proposed in [18–20], and the uniform persistence of the
models and global stability analysis around the epidemic equilibrium are investigated.
In [21], the authors propose a reaction-diffusion SIR epidemic model with multi-group and
the nonlinear infectious force function is investigated, and global asymptotic stability for
disease-free and endemic equilibria is discussed by using the Lyapunov functions method.
It is well known that media coverage reduces the prevalence of infectious disease and
shortens the duration of infection [22]; public health alerts through social media and TV
advertisements play an important role in effectively communicating with people regarding
the spread of any infectious disease during a short period of time [23]. The impact of
public awareness and local behavioral response is studied in [24], and the awareness
among the infected population reduces the infectivity due to pharmaceutical interventions.
Since human behavior is usually changed toward the diseases through mass media, it
is constructive to prevent the spread of infectious diseases as the public relies on what
mass media projects to them [25,26]. By assuming the public health alert programs are
implemented proportional to the infected population, there has been recent progress
on cost-effective analysis for the social media coverage to promote vaccination against
infectious disease [27,28], and some recent studies were conducted to discuss the increment
in vaccination coverage due to social media and TV advertisements, such as [29–32], by
considering media alerts as a decreasing function of the infected population.
Recent theoretical research shows that vaccination and treatment are the most effective
as well as applicable preventive and control strategies during the outbreak of infectious
disease. In order to evaluate optimal strategies of social distancing and vaccination, an
age-structured dynamic system is established to describe the transmission mechanism of
seasonal influenza in [33]. By using Pontryagins’ maximum principle, the authors identify
the best way of reducing morbidity and mortality at a minimal cost. By choosing the
treatment (a pharmaceutical intervention) and constructive public health alerts (a non-
pharmaceutical intervention) as control tools, an optimal control problem is formulated
in [34], which is utilized to minimize the cost and disease fatality. In [35,36], the author
discusses the optimal distribution strategies of an infectious disease with the mixing of
two sub-groups under a resource-constrained environment. An optimal control strategy is
formulated [37] to explore cost-effective intervention strategies associated with vaccination
and water sanitation, which are designed for the prevention and intervention of cholera
epidemics. By considering information-induced vaccination and treatment as controls,
optimal control profiles for the applied controls are obtained in [38]. For some specific infec-
tious diseases (foot-and-mouth disease model in [39] and liver cirrhotic disease model [40]),
optimal vaccination and treatment measures are formulated.
To the author’s best knowledge, there are few results concerning mathematical model-
ing of the age-structured epidemic model with stage progression and media coverage. The
combined dynamic effects of multi-staged infected progression and public health alerts on
epidemiological dynamics have not been investigated before. The optimal control strat-
egy associated with treatment and media coverage has not been discussed in the related
references mentioned above.

2. Model Formulation
Based on the above analysis, some hypotheses are proposed as follows:
Mathematics 2022, 10, 2712 3 of 28

Hypothesis 1 (H1). The total population at any time t is divided into the susceptible population
S(t), exposed population E(t), time and age-dependent infected population I (t, a) and aware popu-
lation A(t). It is assumed that media campaigns induce behavioral changes among the susceptible
population, which form an isolated aware class fully protected from the infectious disease. Λ denotes
the constant recruitment rate of the susceptible population. d1 , d2 and d4 denote the per capita
death rate of S(t), E(t) and A(t), respectively. δ stands for per capita transfer rate from the exposed
population to the infected population, c denotes the dissemination rate of awareness among the
susceptible population, h0 represents the transmission rate of the aware to the susceptible population.

Hypothesis 2 (H2). In order to investigate the possible effects of multiple infectious stages and
variable infectivities in each stage on the epidemiological dynamics, the infection age in the k-th
stage is assumed to be characterized by its relationship to the corresponding adjacent stages. The
entire infectious period [0, +∞) in this paper is assumed to be partitioned into (k + 1)-th stages
defined by infection age intervals ( ak−1 , ak ] (k = 1, 2, · · · , n) such that 0 = a0 < a1 < · · · <
an < an+1 = ∞.

Hypothesis 3 (H3). It is assumed that the susceptible population is vulnerable to the infectious
disease as well as public health alerts regarding social media advertisements and TV. µk ( a) represents
the age-dependent transmission coefficient in the k-th infectious stage, which describes the varying
probability of infectiousness in the k-th stage. νk ( a) denotes the age-dependent recovery rate. d3k
stands for the per capita death rate of the infectious population at the k-th stage, β k ( a) denotes the
transmission rate of the infectious population from the k-th stage to the next (k + 1)-th stage.

Hypothesis 4 (H4). M(t) represents the cumulative number of social media and TV advertise-
ments at time t. It is plausible to assume that the cumulative number of social media and TV
advertisements proportionally increases with the number of infected individuals. It is also assumed
that the growth rate of social media and TV advertisements is proportional to the number of the in-
fected population with a decreasing function of the aware population. l represents the half-saturation
point for the impact of social media and TV advertisements on the susceptible population. M0
and η represent the baseline number and diminution rate of social media and TV advertisements,
respectively. rk ( a) denotes the age-dependent growth rate of social media and TV advertisements.
θk ( a) denotes the age-dependent decay rate in social media and TV advertisements due to an increase
A(t)
in the aware population. ωk ( a) represents the half-saturation point as rk ( a)θk ( a) ω (a)+ A(t) attains
k
half of its maximum value rk ( a)θk ( a).

Hypothesis 5 (H5). All constant parameters are nonnegative, and the age-dependent functions
are all nonnegative, bounded and integrable in their definition intervals.

Keeping all these Hypotheses (H1)–(H5) in mind, we formulate a multi-stage SEIAM

model for infectious diseases with media coverage and multi-staged infection progression,
which is governed by the following hybrid model with ODE and PDE,

dS(t) +1 a k cS(t) M(t)
= Λ + ∑nk=
R
1 ak−1 [ νk ( a ) − µk ( a ) S ( t )] Ik ( a, t )da − l + M (t) + h0 A ( t ) − d1 S ( t ),


 dt
dE(t) +1 a k

= ∑nk=
R
µk ( a)S(t) Ik ( a, t)da − (δ + d2 ) E(t),


dt 1 a

k −1


∂Ik ( a,t) ∂Ik ( a,t)
∂a + ∂t = −[νk ( a) + d3k ( a) + β k ( a)] Ik ( a, t), a ∈ ( ak−1 , ak ] (1)
dA(t) cS(t) M(t)

dt = l + M (t) − ( h0 +hd4 ) A ( t ),




 i
dM (t) n +1 a k θk ( a) A(t)

∑
R
= r ( a ) 1 − I ( a, t)da − η ( M (t) − M0 ),

k ωk ( a)+ A(t) k

dt k =1 a k −1

for k = 1, · · · , n + 1 with a0 = 0, an+1 = ∞ and β n+1 ( a) = 0 for any a ∈ ( an , an+1 ).

The boundary conditions are
Z a
k
I1 ( a0 , t) = δE(t), Ik+1 ( ak , t) = β k ( a) Ik ( a, t)da. (2)
a k −1
Mathematics 2022, 10, 2712 4 of 28

For any t > 0, k = 1, · · · , n + 1 and initial conditions

S(0) = S0 , E(0) = E0 , Ik ( a, 0) = Ik0 ( a), A(0) = A0 , M(0) = M0 , (3)

such that S0 , E0 , A0 , M0 ∈ R≥0 , Ik0 ∈ L1+ (0, ∞).

Remark 1. Recently, some epidemic models with media-induced nonlinear incidence and age-
dependent vaccination have been established in [41,42]. Uniform persistence and the sharp threshold
dynamics are investigated, and the efficacy of vaccines is discussed due to the variation in vaccination
reproduction number. However, the combined dynamic effects of multiple disease progressions and
public health alerts on epidemiological dynamics have not been investigated. Compared with the
work in [41,42], model (1) can not only describe infectious disease progression through multiple
stages but also assess the dynamic impact of social media and TV advertisements on the spread of an
infectious disease. The treatment measures generally minimize the serious consequences of infection,
and efficient media coverage act as a constructive public health alert, which reduces the infection
rate. With these proper control measures, the authorities may significantly reduce wealth loss and
mortality due to disease burden. Compared with recent references [33–40] concerning the optimal
control strategies for infectious disease, the treatment and diminution rate of media coverage are
first adopted as control variables in this paper. An optimal cost-effective control strategy is discussed
to minimize the costs incurred due to applied controls and infectious disease burden.

The rest of the sections of this paper are organized as follows. In the Section 2,
nonnegativity and the uniform persistence of the solution are studied by using the basic
reproduction number. By constructing the appropriate Lyapunov functions, the global
stability analysis of endemic equilibrium is investigated. In the Section 3, an optimal cost-
effective control strategy is discussed to minimize costs incurred due to applied controls
and infectious disease burden. In the Section 4, numerical simulations are provided to
support the theoretical analysis. Finally, this paper ends with a conclusion.

3. Qualitative Analysis of SEIAM Model

In this section, the combined dynamic effects of media coverage and multi-staged
infected progression on infectious disease dynamics are discussed. Sufficient conditions for
uniform persistence of the solution are studied by using the basic reproduction number.
By constructing appropriate Lyapunov functions, global stability analysis of the endemic
equilibrium is investigated based on the Lyapunov–LaSalle stability theorem.

3.1. Basic Reproduction Number and Endemic Equilibrium

In order to describe the age-specific survival probability of an infected person in the
age interval ( ak−1 , ak ], we define
Ra
− ak −1 [ νk ( τ )+ d3k ( τ )+ β k ( τ )]dτ
gk ( a ) = e , a ∈ ( a k −1 , a k ] , k = 1, · · · , n + 1, (4)

and some mathematical notations are also defined for k = 1, · · · , n + 1,

Z a Z a Z a
k k k
Vk = β k ( a) gk ( a)da, Qk = µk ( a) gk ( a)da, Uk = νk ( a) gk ( a)da. (5)
a k −1 a k −1 a k −1

Remark 2. It should be noted that Vk (k = 1, 2, · · · , n + 1) represents the probability that an

infected individual survives during the k-th infection age and proceeds to the (k + 1)-th infection age.
Λ
d denotes the expected new infectious individual generated by an infected individual during the k-th
1
δ(h0 +d4 )(l + M0 )
infection age from ( ak−1 , ak ]. Furthermore, (δ+d2 )[cM0 +(h0 +d4 )(l + M0 )]
is the fraction of individuals
−1
surviving through the exposed stage and reaching the first infection age. Qk Πkj= 1 Vj stands for
the part of secondary infections generated from the k−th infection age, when a typical infected
Mathematics 2022, 10, 2712 5 of 28

k −1
+1
individual contacts a susceptible individual. Hence, the sum of n + 1 terms ∑nk= 1 Qk Π j=1 Vj
denotes secondary infections from all n + 1 infection ages.

By using Remark 2 and simple computations, it is easy to show the basic reproduction
number is as follows:
n +1
δΛ(h0 + d4 )(l + M0 )
R0 =
d1 (δ + d2 )[cM0 + (h0 + d4 )(l + M0 )] ∑ Qk Πkj=−11 Vj . (6)
k =1

Based on system (1), if R0 > 1, then it follows from simple computations that the
endemic equilibrium P∗ = (S∗ , E∗ , I1∗ ( a), · · · , In∗+1 ( a), A∗ , M∗ ) is as follows,

S = d ΛR ,
 ∗
1 0

 
∗ = Λ 1

E 1 − R0 ,




 d2 + δ
−1
I1 ( a0 ) = δE , Ik∗ ( a) = I1∗ ( a0 ) gk ( a)Πkj=
 ∗ ∗ V (k = 2, · · · , n + 1),
 n + k −
 1 j n +1 k −1
(7)
∗ (a ) 1 1 ∗
∗ = Λ 1 − 1 1 0 k=1 i=1 i k + I1 ( a0 ) ∑k=1 (Πi=1 Ui ) Qk ,
d + I ∑ ( Π V ) Q

A




 h0 d1 R0 h0
l ( h0 + d4 ) A ∗

∗

M = cS∗ −(h +d ) A∗ .

0 4

3.2. Nonnegativity and Uniform Persistence of Solution

In order to facilitate the following proof, vector w is defined as follows,

w = ( w1 , w2 , u 1 , · · · , u n +1 , w3 , w4 ) T
 
= (S, E, I1 ( a, t), · · · , In+1 ( a, t), A, M)T ∈ R4 × L1 (0, +∞), Rn+1

it is a solution vector, which is endowed with the following norm,

n +1 Z a k
kwk = |S(t)| + | E(t)| + ∑ | Ik ( a, t)|da + | A(t)| + | M(t)|.
k =1 a k −1

Lemma 1. For system (1) with any nonnegative initial values, all of the solutions of system (1)
are nonnegative.

Proof. Based on the standard theory for age-structured models [43], boundary conditions
(2) and initial conditions (3), by integrating Ik ( a, t) on ( ak−1 , ak ] it yields,
(
Ik ( ak−1 , t + ak−1 − a) gk ( a), t + ak−1 > a,
Ik ( a, t) = 0 gk ( a ) a ∈ ( a k −1 , a ] . (8)
Ik ( a − t − ak−1 ) g (a−t−a ) , t + ak−1 ≤ a,
k k −1

According to the classical existence and uniqueness results for functional differential
equations, by substituting (8) into system (1), it is easy to show that Ik ( a, t) is nonnegative
for any nonnegative initial value, and there exists a unique solution for system (1).
If there exists t1 such that S(t1 ) = 0 and S(t) > 0 holds for 0 < t < t1 , then it implies
dS(t1 )
that dt > 0 and S(t) ≥ 0 holds for t ≥ 0. By using similar arguments, the nonnegativity
of other variables, E(t), R(t), A(t) and M (t), can be shown for any nonnegative initial
values. This completes the proof.

In order to guarantee the existence of the semiflow solution for system (1), which
admits a compact global attractor, system (1) can be rewritten as the following Cauchy
problem defined on boundary condition (2),

dz
= A z + F ( z ), (9)
dt
Mathematics 2022, 10, 2712 6 of 28

→ R4 × L1 (0, +∞), Rn+1 × Rn+1 , z = (w, 0, · · · , 0)T ∈ R4 ×

where A : Dom(A)
L1 (0, +∞), Rn+1 × {0}n+1 take the following form,

 
− d 1 w1

  −( d 2 + δ ) w2 


− ∂a + ν1 ( a) + d31 ( a) + β 1 ( a) u1
∂
 
 
 
 .. 

  . 


 
Az = 
 − ∂a + νn+1 ( a) + dn+1 ( a) + β n+1 ( a) un+1
∂
,

(10)
− u1 (0)
 
 
 .. 
.
 
 
− u n +1 (0 )
 
 
−(h0 + d4 )w3
 
 
−ηw4

with z(0) = z0 ∈ B+ , Dom(A) = z ∈ B+ uk (·) ∈ Z1,1 [0, +∞), k = 1, · · · , n + 1 , Z1,1



represents a Sobolev space and B+ is defined as follows,

n   o n o
B+ = R4 × L1 (0, +∞), Rn+1 × {0}n+1 ∩ R4+ × L1+ ((0, +∞), Rn+1 ) × Rn++1 . (11)

Furthermore, F is defined
as follows,

F : R4 × L1 (0, +∞), Rn+1 × {0}n+1 → R4 × L1 ((0, +∞), Rn+1 ) × Rn+1 ,
R ak ew1 w4
+1
Λ + ∑nk=
 
1 ak−1 [ νk ( a ) − Rw1 µk ( a )] uk ( a, t )da − l +w4 + h 0 w3
+1 a k
w1 ∑nk= 1 ak−1 µk ( a ) uk ( a, t )da
 
 
 

 0 L1 

 .. 

 . 


 0 L1 

 
 δw2 
F (z) = 
 .. .

(12)
 R ak . 

β k ( a)uk ( a, t)da

a k −1
 
 
 .. 

 R an . 


 a n −1 β n ( a)un ( a, t)da 

 cw1 w4 
l + w4
 
 R ak h i 
θ k ( a ) w3
+1
∑nk= 1 a k −1 r k ( a ) 1− ωk ( a)+w3
uk ( a, t)da + η M0

It follows from Theorem 2.1 and Theorem 2.2 in [44] that there exists a unique solution
semiflow Z (t) : B+ → B+ defined byR system (9).
+1 a k
Let N (t) = S(t) + E(t) + ∑nk= 1 ak−1 Ik ( a, t )da + A ( t ) + M ( t ), it follows from simple
computations and system (1) that

dS(t) dE(t) n+1 d ak

Z 
dN (t) dA(t) dM(t)
= + +∑ Ik ( a, t)da + +
dt dt dt k =1
dt a k −1 dt dt
n +1
≤ Λ + ηM0 − d1 S(t) − d2 E(t) − d4 A(t) − η M(t) − ∑ Ik ( ak , t)
k =1
n +1 Z a k
+ ∑ [νk ( a) + rk ( a)(1 − θk ( a)) − d3k ( a)] Ik ( a, t)da
k =1 a k −1
≤ Λ + η M0 − αN (t), (13)

where α = min{d1 , d2 , d4 , η, d3k ( a) − νk ( a) − rk ( a)(1 − θk ( a))} (k = 1, 2, · · · , n + 1).

Mathematics 2022, 10, 2712 7 of 28

By using (13) and the comparison principle [45], it can be obtained that

Λ + η M0
lim sup N (t) ≤ , (14)
t→∞ α

which reveals that all solutions of system (1) and the corresponding system (9) are uniformly
bounded. Hence, the solution semiflow Z (t) is point dissipative. According to Theorem
3.6.1 [45], it can be obtained that Z (t) is compact for all t > 0, and it is easy to show that
Z (t) has a compact global attractor in B+ based on Theorem 1.1.3 [46]. Based on the above
analysis, Lemma 2 can be concluded as follows.

Lemma 2. System (9) has a unique solution semiflow Z (t) : B+ → B+ . Furthermore, there exists
a compact global attractor for Z (t).

Lemma 3. The subset ∂I is positively invariant under the semiflow { Z (t)}t≥0 and Z (t)∂I ⊂ ∂I .
When t is large enough, Z (t)v → v̄ holds for each x ∈ ∂I , where I , ∂I and v̄ are defined as follows,
 Z a 
k
I = ( I1 , · · · , In+1 )T ∈ L1+ ((0, +∞), Rn+1 ) Ik ( a)da > 0, ∃k ∈ [1, · · · , n + 1] ,
a k −1
 
Λ
∂I = B+ × {0}n+1 \ { R4+ × I × {0}n+1 }, v̄ =  , 0, 0 L1 , · · · , 0 L1 , 0, · · · , 0, 0, 0.

d1 | {z } | {z }
n +1 n +1

Λ+η M0
Proof.
 Based on (14), it is easyto show that S(t) ≤ α , for any initial point in ∂I and

S0 , I 0 , · · · , I 0 , 0, · · · , 0, 0, 0∂I , we consider the following auxiliary system

1 n+1 | {z }
n +1

dEe(t) Λ+ηM0 n+1 R ak


 dt = α ∑ k =1 a k −1 µ k ( a ) e Ik ( a, t)da − (δ + d2 ) E e(t)

∂ Ik ( a,t) ( a,t)
+ ∂ Ik∂t

= −[νk ( a) + d3k ( a) + β k ( a)] e Ik ( a, t), a ∈ ( ak−1 , ak ]

 e e

 ∂a
Λ+η M0 c M

 d A(t) e (t)
dt = e ( t ) − ( h0 + d4 ) A ( t ),
 e
e
α l+M (15)
h i
d M(t) n +1 a k θk ( a) A
e(t)
∑
R
= r ( a ) 1 − e(t) Ik ( a, t )da − η ( M ( t ) − M0 ),

 e
k
e e


 dt k =1 a k −1 ωRk ( a)+ A
 a


 I1 ( a0 , t) = δ E
e e(t), eIk+1 ( ak , t) = a k β k ( a) e Ik ( a, t)da,
 k −1
E(0) = E , Ik ( a, 0) = Ik ( a), A(0) = A , M(0) = M0 .
0 0 0

 e e e e

By using the comparison principle [46], it can be obtained that E(t) ≤ E e(t), A(t) ≤
Ae(t), M (t) ≤ M e (t) and Ik ( a, t) ≤ e Ik ( a, t) (k = 1, 2, · · · , n + 1), where ( E I1 ( a, t), · · · , e
e(t), e In+1
( t ), A
e(t), M
e (t)) is a solution of system (15).
Based on (8) and system (15), the e Ik ( a, t) is obtained as follows,
(
ξ k ( a k −1 + t − a ) g k ( a ), t + ak−1 > a,
Ik ( a, t) =
e 0 gk ( a ) a ∈ ( a k −1 , a k ]
Ik ( a − t − ak−1 ) g (a−t− a ) , t + ak−1 ≤ a,
k k −1

where ξ k (k = 1, · · · , n + 1) can be obtained from the following expressions,

 R t −(d +δ)(t−τ )  n+1 R ak−1 +τ



ξ 1 (t) = δΛ
d1 0 e
2 ∑ k =1 a k −1 νk (τ1 )( ak−1 + τ − τ1 ) gk (τ1 )dτ1

n +1 a k
+ ∑k=1 a +τ νk (τ1 ) Ik (τ1 − τ − ak−1 ) g (τ g−k (ττ−1 )a ) dτ1 dτ,

0

 R
R a + t k −1 k 1 k −1 (16)

 ξ k+1 (t) = a k−1 β k (τ )ξ k (t + ak−1 − τ ) gk (τ )dτ
Rk−a 1


+ a k +t β k (τ ) Ik0 (τ − t − ak−1 ) g (s−gkτ(−τ )a ) dτ.


k −1 k k −1
Mathematics 2022, 10, 2712 8 of 28

Ra
Since ( I10 , I20 , · · · , In0+1 ) ∈ L+ ((0, +∞), Rn+1 )\I , it is easy to show a k Ik0 ( a)da = 0
k −1
holds for any k ∈ [1, 2, · · · , n + 1], which implies that

 ∑ n +1 a k 0 gk ( τ )
R
k =1 ak−1 +τ1 νk ( τ ) Ik ( τ − τ1 − ak −1 ) gk (τ −τ1 − ak−1 ) dτ = 0,
R ak gk (τ1 ) (17)

a k −1 + t β k (τ1 ) I 0 (τ1 − t − ak−1 )
k dτ1 = 0.
gk (τ1 −t− ak−1 )

It follows from (16) and (17) that ξ k (t) is the unique solution of Volterra’s Equation (18),
 R t −(d +δ)(t−τ )  n+1 R ak−1 +τ 
 ξ 1 (t) = δΛ
d1 0 e
2 ∑ k =1 a k −1 νk (τ1 )( ak−1 + τ − τ1 ) gk (τ1 )dτ1 dτ,
 ξ k+1 (t) = ak−1 +t β k (τ )ξ k (t + ak−1 − τ ) gk (τ )dτ.
R
a k −1
(18)
Ik ( a, t) = 0 holds for t > a − ak−1 .
which implies that e
On the other hand, for t ≤ a − ak−1 , it yields that

gk ( a )
Ik ( a, t)k L1 = Ik0 ( a − t − ak−1 )
ke ≤ e−et k Ik0 k L1 ,
g k ( a − t − a k −1 ) L1

where e = min{µk ( a), d3k ( a), β k ( a)} (k = 1, 2, · · · , n + 1). Hence, it can be concluded that
Ik ( a, t) → 0 for t → ∞.
e
By utilizing the comparison theorem [46], it is easy to show that Z (t)v → v̄ ∈ ∂I holds
for each v ∈ ∂I . This completes the proof.

Theorem 1. When R0 > 1, the semiflow { Z (t)}t≥0 is uniformly persistent with respect to
(R4+ × I × {0}n+1 , ∂I), i.e., there is a ζ > 0 such that solution of system (9) with initial value
(S0 , E0 , I10 , · · · , In0+1 , 0, · · · , 0, A0 , M0 ) ∈ R4+ × I × {0}n+1 , and

lim inf E(t) > ζ, lim inf k Ik (·, t)k > ζ, lim inf A(t) > ζ, lim inf M(t) > ζ.
t→∞ t→∞ t→∞ t→∞

Furthermore, there exists a compact subsect of I0 ⊂ I , which is also a global attractor for
{ Z (t)}t≥0 in I .

Proof. According to Lemma 3, it is easy to show that v̄ is globally asymptotically stable in

∂I. In order to investigate the asymptotic behavior of the solution to system (15) from the
neighborhood of v̄, where v̄ is defined in Lemma 3.
In the following part, we will show that there exists T0 ≥ 0 such that kv̄ − Z ( T0 )vk > ε,
i.e., W s ({v̄}) ∩ {R4+ × I × {0}n+1 } = ∅ holds for v = (S0 , E0 , I10 , · · · , In0+1 , 0, · · · , 0, A0 , M0 ) ∈
{λ ∈ R4+ × I × {0}n+1 |kv̄ − λk ≤ ε } and any ε > 0.
If W s ({v̄}) ∩ {R4+ × I × {0}n+1 } 6= ∅, then it yields that
 
n +1 1
λJ ∈ λ∈ R4+ × I × {0} kv̄ − λk ≤
J+1

holds for any J ≥ 0.  

For any t ≥ 0, λ J = S0J , E0J , I1,J
0 , · · · , I0
n+1,J , 0, · · · , 0, A 0 , M0 satisfies
J J

1
kv̄ − Z (t)λ J k ≤ . (19)
J+1


Let Z (t)λ J = S J (t), E J (t), I1,J (·, t), · · · , In+1,J (·, t), 0, · · · , 0 , it follows from (19) that

Λ 1
S J (t) − ≤
d1 J+1

holds for any J ≥ 0 and t ≥ 0.

Mathematics 2022, 10, 2712 9 of 28

For the following auxiliary system (20) with k = 1, · · · , n + 1 and J ≥ 0,

dE J (t) +1 a k

= S J (t) ∑nk=
R
dt 1 ak−1 µk ( a ) Ik,J ( a, t )da − ( d2 + δ ) E J ( t )



 ∂Ik,J ( a,t) ∂I ( a,t)
+ k,J∂t = −(µk ( a) + d3k ( a) + β k ( a)) Ik,J ( a, t), a ∈ ( ak−1 , ak ]



 ∂a
dA J (t) cS (t) M (t)

= lJ+ M (Jt) − (h0 + d4 ) A J (t)


dt J (20)
h i
dM J (t) n +1 a k θk ( a) A J (t)
∑
R



 dt = r
k =1 a k −1 k ( a ) 1 − I ( a, t)da − η ( M J (t) − M0 )
ωk ( aR)+ A J (t) k,J
 ak
I1,J ( a0 , t) = δE J (t), Ik+1,J ( ak , t) = a β k ( a) Ik,J ( a, t)da,




 k −1
E J (0) = E0J , Ik,J (·, 0) = Ik,J 0 , A (0) = A0 , M (0) = M 0 .

J J J J

According to the comparison principle [46], it yields that

E J (t) ≥ E
eJ (t), Ik,J (·, t) ≥ e
Ik,J (·, t), A J (t) ≥ A
e J ( t ), M J ( t ) ≥ M
e J ( t ), (21)
 
I1,J (·, t), · · · , e
eJ ( t ) , e
where E In+1,J (·, t), A
e J ( t ), M
e J (t) is the solution of Equation (22),

  
dEeJ ( t )
= dΛ1 − J +1 1 ∑nk= +1 a k
R
1 ak−1 µk ( a ) Ik,J ( a, t )da − ( d2 + δ ) E J ( t )
 e e


 dt
 Ik,J ( a,t)
∂e ∂eI ( a,t)
+ k,J∂t = −(µk ( a) + d3k ( a) + β k ( a)) e Ik,J ( a, t), a ∈ ( ak−1 , ak ]



 ∂a
   e
 dAe J (t) Λ 1 c M J (t)
= d1 − J +1 e J ( t ) − ( h0 + d4 ) A J ( t )

 e
dt l+M   (22)
dMe J (t) n +1 a k θk ( a) A
e J (t)
= ∑ k =1 a r k ( a ) 1 −
R
e (t) Ik,J ( a, t )da − η ( M J ( t ) − M0 )

 e e
dt

 k −1 ωk ( a)+ A
R ak J



I1,J ( a0 , t) = δ EeJ ( t ) , e
Ik+1,J ( ak , t) = a β k ( a) e Ik,J ( a, t)da,

 e

 k −1

0
eJ ( 0 ) = E , e
E 0 0 0
J Ik,J (·, 0) = Ik,J , A J (0) = A J , M J (0) = M J .
 e e

By similar arguments in [18,47], the following notations are defined to facilitate dis-
cussing the characteristic equation of system (22),
Ra
− a [v+νk (τ )+d3k (τ )+ β k (τ )]dτ


 g̃ k ( v, a ) = e k −1 ,
a
 R

 Vek (v) = k β k ( a) g̃k (v, a)da,
Rak−1 (23)
 Qe k (v) = ak µk ( a) g̃k (v, a)da,
R a k −1


e (v) = ak ν ( a) g̃ (v, a)da,

 U
k a k −1 k k

where a ∈ ( ak−1 , ak ], k = 1, · · · , n + 1.
When R0 > 1 and there is a sufficiently large J > 0, it follows from simple computations
that the dominant eigenvalues of system (22) satisfy the following characteristic equation
k −1 e
Λ +1 e
δ ∑nk= 1 Qk ( v J ) Πi =1 Vi ( v J )
 
1
H(v J ) := − = 1. (24)
d1 J+1 v J + d2 + δ

It follows from (23) that H(0) = R0 = 1 and Vk (v J ) (k = 1, 2, · · · , n) are decreasing

functions with respect to v J . Hence, it is easy to show that there exists a unique root v∗J > 0
provided that R 0 > 1. 
Let ϕ(t) = E I1,J (·, t), · · · , e
eJ ( t ) , e In+1,J (t), A e J (t) and ∆(v J ) = 1 − H(v J ), then
e J ( t ), M
system (22) can be linearized as follows:

dϕ(t)
= B ϕ ( t ), (25)
dt
Mathematics 2022, 10, 2712 10 of 28

where
 
−(d2 + δ) EeJ (t)
deI (·,t)
− 1,Jda − (ν1 ( a) + β 1 ( a) + d1 ( a)) e
 
 I1,J (·, t) 
..
 
 
B ϕ(t) = 
 . ,

 deIn+1,J (·,t)
 − − (νn+1 ( a) + β n+1 ( a) + dn+1 ( a)) e

da In+1,J (·, t) 
 
 −(h0 + d1 ) A e J (t) 
−η Me J (t)

and the closed linear operator B is as follows,

n 
D(B) = I1,J (t), · · · , e
eJ ( t ) , e
E In+1,J (t), A e J (t) ∈ R × W 1,1 ((0, +∞), Rn+1 )
e J ( t ), M

  
eJ ( 0 ) = Λ − 1 ∑ n + 1 a k µ k ( a ) e
R
E d1 J +1 k =1 a k −1 Ik,J ( a, t)da, 


R ak 
I1,J (0) = δ E Ik+1,J (0) = a β k ( a) e
eJ , e Ik,J ( a, t)da,
e 


k −1 
,
  e
e J (0) = Λ − 1 c M J
A d1 J +1 l + M eJ , 

  

e J (0 ) = ∑ n +1 a k r k ( a ) 1 − θ k ( a ) A J e
R e 
M e Ik,J ( a, t )da + ηM0 .


k =1 a k −1 ω ( a)+ A
k J


1
where R(v, B) is the resolvent operator of B with k R(z, B)k ≤ Rev .
According to the Hille–Yosida theorem [48], a strongly continuous semigroup { Z e (t)}t≥0
4
can be generated by the operator (B , D(B)). For any v J ∈ R+ × I × {0} n + 1 from the neigh-
borhood of v̄, Φ J denotes the specific projector on the eigenspace with respect to the
dominant eigenvalue v∗J , it follows from the similar arguments in [18,49], it yields
 T
ΦJ Z e ( t ) E0 , ( I 0 ), · · · , ( I 0
J 1,J
0
n+1,J ), A J , M J
0

∗
 T
= ev J t Φ J E0J , ( I1,J
0 ), · · · , ( I 0 0 0
n+1,J ), A J , M J
T (26)
∗

= ev J t limv→v∗J (v − v∗J )(vΞ1×n − B)−1 E0J , ( I1,J 0 ), · · · , ( I 0 0 0
n+1,J ), A J , M J
∗
= e v J t ( ϕ 1 , · · · , ϕ n +4 ) T ,

where Ξ1×n represents the 1 × n identity matrix.

By the similar arguments and computations in [8], it can be obtained that
"   #
  0
Ik,J g̃k (v∗J ,a)
Λ 1 n +1 ∗ n + 1
R ak Ra
ϕ1 = −
d1 ∑
J +1
e k (v ) + ∑
ϕ k +1 Q
k =1 J µk ( a)
k =1 a k −1 ∗ dτda .
a k −1 g̃k (v J ,τ )
(27)

According to the recurrence relationships of ϕ(t), it yields that

ϕ2 = δϕ1 g̃1 (v∗J , a),



  
0 g̃1 (v∗J ,a)

 a a I1,J
∗
 R 1
R
 ϕ3 = ϕ2 V1 (v J ) + a0 β 1 ( a) a0 g̃1 (v∗ ,τ ) dτda,


J
.. (28)


 .  
g̃n (v∗J ,a)
 0
In,J
 R an Ra
∗

 ϕn+2 = ϕn+1 Vn (v ) +

β n ( a) ∗ dτda,
J a n −1 a n −1 g̃n (v J ,τ )

and
Mathematics 2022, 10, 2712 11 of 28

  
Λ 1 cϕn+4
 ϕn+3 = d1 −" J +1 l + ϕn+4 ,


  #
R ak 
θ k ( a ) ϕ n +3
R
a
0
Ik,J g̃k (v∗J ,a) (29)


+1
 ϕn+4 = ∑nk= e ∗
1 ϕ k +1 Q k ( v J ) + a k −1 r k ( a ) 1 − ωk ( a)+ ϕn+3 a k −1 g̃k (v∗J ,τ )
dτda + η M0 .

By substituting (28) into (27) and (29), ϕ1 , ϕn+3 and ϕn+4 can be obtained as follows,

 dM(v∗ ) −1
"  
 Ik0−2,J g̃k−2 (v∗J ,a)
Λ
 1 n +1 ∗
R ak Ra
∑ k =1 ϕ k +1 Q k ( v J ) a k −1 β k ( a ) a k −1
J
ϕ 1 = d − J +1 dτda


g̃k−2 (v∗J ,τ )
e


 1 dv J

  
Ii0−1,J g̃i−1 (v∗L ,a)


∑ n +1 k −1 e
Π ∗ )Q e k ( v ∗ ) a i −1 β i −1 ( a ) a
 R R

 + i =2 Vi − 1 ( v L J a g̃i−1 (v∗L ,τ )
dτda
k =3 # a i −2

  i −2


0 ∗

 Ik,J g̃k (v J ,a)
+1 a k
Ra
+ ∑nk=
 R
µk ( a) a dτda ,

g̃k (v∗J ,τ )



 1 a k − 1 k − 1

  
ϕn+3 = dΛ − J +1 1 l + ϕn+4 ,
cϕ
(30)
 1 " n +4
− 1
 
0 ∗
  
dM(v∗J ) Ik−2,J g̃k−2 (v J ,a)
  
e k ( v ∗ ) a k r k ( a ) 1 − θ k ( a ) ϕ n +3
Ra
∑nk= +1
 R
ϕ n +4 = ϕ k +1 Q dτda

g̃k−2 (v∗J ,τ )



 dv J 1 J a k − 1 ω k ( a )+ ϕ n + 3 a k − 1

  
Ii0−1,J g̃i−1 (v∗L ,a)

  R
n +1 k −1 e
Π ∗ )Q ∗ ) a i −1 r θ i −1 ( a ) ϕ n +3 a
∑
 R

 + k = 3 i = 2 Vi −1 L ( v e (
k Jv a i − 1 ( a ) 1 − ω ( a )+ ϕ a g̃i−1 (v∗L ,τ )
dτda
  i −2 i −2

 i −1 n +3
 #
0 g̃k (v∗J ,a)

 Ik,J
+1 a k
Ra
+ ∑nk=
 R
1 a k −1 µ k ( a ) a k −1 dτda + η M0 .

g̃ (v∗ ,τ )


 k J

dM(v∗J )
 
0 , · · · , I0
Since dv J > 0 and I1,J n+1,J ∈ I , it follows from (28) and (30) that

lim E J = +∞, lim k Ik,L (·, t)k = +∞, lim A J = +∞, lim M J = +∞. (31)
t→+∞ t→+∞ t→+∞ t→+∞

Hence, it follows from the above analysis that (31) contradicts (28). Based on Theorem 4.2
in [50], it is easy to show that the semiflow { Z (t)}t≥0 is uniformly persistent with respect to
(R4+ × I × {0}n+1 , ∂I), i.e., there is a ζ > 0 such that there is a solution of system (9) with an
initial value (S0 , E0 , I10 , · · · , In0+1 , 0, · · · , 0, A0 , M0 ) ∈ R4+ × I × {0}n+1 , and

lim inf E(t) > ζ, lim inf k Ik (·, t)k > ζ, lim inf A(t) > ζ, lim inf M(t) > ζ.
t→∞ t→∞ t→∞ t→∞

Furthermore, according to the above analysis, it is easy to show that there exists a
compact subsect of I0 ⊂ I , which is also a global attractor for { Z (t)}t≥0 in I .

3.3. Stability Analysis of Endemic Equilibrium

Theorem 2. When R0 > 1, the endemic equilibrium P∗ is locally stable.

Proof. According to (7), there exists an endemic equilibrium P∗ when R0 > 1, by defining
some following new variables with exponential form,

Ik ( a, t) = evt Ī ( a) + Ik∗ ( a), (32)

and
S(t) = evt S̄ + S∗ , E(t) = evt Ē + E∗ ,

(33)
A(t) = evt Ā + A∗ , M (t) = evt M̄ + M∗ .
By substituting (32) into system (1), it yields that

∂Ik ( a, t) ∂Ik ( a, t) ∂evt Ī ( a) ∂evt Ī ( a) dIk∗ ( a)

+ = + +
∂a ∂t ∂a ∂t da
= −(νk ( a) + d3k ( a) + β k ( a))(evt Ī ( a) + Ik∗ ( a)). (34)
Mathematics 2022, 10, 2712 12 of 28

By separating and solving Equation (34), Ik∗ ( a) and Ī ( a) are obtained as follows,

Ik∗ ( a) = I1∗ ( a0 )(Πik=−11 Vi ) gk ( a), Īk ( a) = Ī1 ( a0 )(Πik=−11 Vi ) g̃k (v, a), (35)

where gk ( a) and Uk are defined in (5), and g̃k (v, a) is defined in (23).
By substituting (33) into system (1), S̄, Ē, Ā and M̄ are obtained as follows,
+1 a k n +1 a k cS∗ M̄
∑kn= ∗
1 ak −1 νk ( a ) Īk ( a,t )da − ∑k =1 ak−1 S µk ( a ) Īk ( a,t )da + h0 Ā + l + M̄
 R R
S̄ = ,

cM∗

+1 a k
v+d1 +∑nk= ∗
 R
1 ak−1 µk ( a ) Ik ( a )da + l + M∗


 Ra Ra
n + 1 ∗ n + 1 ∗

∑k=1 a k S̄µk ( a) Ik ( a)da+∑k=1 a k S µk ( a) Īk ( a)da


k −1 k −1
Ē = ,

∗ ∗
v + d2 + δ
∗
(36)
c [ M S̄ ( l + M̄ )− S M̄ ( l + M )]


 Ā = (l + M∗ )(l + M̄)(v+h +d ) ,
 0 4
A∗ θk ( a)
 
+1 a k +1 a k Āθk ( a)
∑nk= da−∑nk=
 R R ∗
rk ( a) Īk ( a) 1− ω ( a)+ 1 ak−1 rk ( a ) Ik ( a ) ωk ( a)+ Ā da

 1 a A ∗
k −1 k

M̄ = .

v+η

According to Ī1 ( a0 ) = δ Ē, the characteristic equation is expressed as follows:

(
n +1 Z a k
δ  
1 = ∑
v + d 2 + δ k =1 a k −1
S∗ µk ( a) Πik=−11 Vi g̃k (v, a)da
 
+1 a k k −1
∑nk= ∗
1 ak−1 Πi =1 Vi gk ( a )( νk ( a ) − S µk ( a ))da
R
− cM∗ n +1 a k ∗
M∗ + ∑k =1 ak−1 µk ( a ) Ik ( a )da
R
v + d1 + l +
 ∗
M∗ h0 S̄(l + M̄ )

+1 a k S (d4 +v) M̄

∑nk= ∗
R
1 ak−1 cIk ( a ) µk ( a ) d4 +h0 +v + (l + M∗ )(h0 +d4 +v) da

− h i . (37)
cM∗ n +1 a k ∗ ( a )da 
∑
R
Ī1 ( a0 )(l + M̄ ) v + d1 + l + M ∗ + k =1 a µ k ( a ) Ik k −1

In order to investigate the possibility of a solution to Equation (37) having a nonnega-

tive real part, i.e., Re(v) ≥ 0, some discussions are provided as follows.
If v = 0, then Equation (37) is rewritten as
 
n +1 a k

k −1
∑ Π gk ( a)(νk ( a) − S∗ µk ( a))da
R
δ  n +1   k =1 a k −1 i =1 V i
1 = S∗ ∑ Πik=−11 Vi Qk − cM∗ n +1 a k ∗
M∗ + ∑k =1 ak−1 µk ( a ) Ik ( a )da
R
d 2 + δ  k =1 d1 + l +
∗
∗ ( a ) µ ( a ) S∗ d4 M̄ + M h0 S̄(l + M̄) da 
 
+1 a k

∑nk=
R
cI
1 a k −1 k k d4 + h0 ∗
(l + M )(h0 +d4 )
− h i ,
cM∗ n +1 a k
Ī1 ( a0 )(l + M̄ ) d1 + l + M∗ + ∑k=1 a µk ( a) Ik∗ ( a)da 
R
k −1

which follows that

 
+1 a k

k −1
∑nk= 1 ak−1 Πi =1 Vi gk ( a ) νk ( a )da
R
δ  ∗ n +1  k −1 
d2 + δ  k∑
1 < S Πi=1 Vi Qk − cM∗ n +1 a k ∗
M∗ + ∑k =1 ak−1 µk ( a ) Ik ( a )da
R
=1 d1 + l +

+1 a k
∑kn= ∗ ∗
R
1 ak−1 cIk ( a ) µk ( a ) M h0 S̄ ( l + M̄ )da

− h i
cM∗ n +1 a k ∗
Ī1 ( a0 )(h0 + d4 )(1 + M∗ )(l + M̄ ) d1 + l + M∗ + ∑k =1 ak−1 µk ( a ) Ik ( a )da
R 
 
n +1 a k

k −1
∑ Π
R
δ  k =1 a k −1 i =1 V i gk ( a)νk ( a)da
= 1− ∗ a
d2 + δ  d1 + cM ∗ + ∑n+1 k µk ( a) I ∗ ( a)da
R
l+M k =1 a k −1 k

+1 a k
∑kn= ∗ ( a ) µ ( a ) M ∗ h S̄ ( l + M̄ )da
R
1 a k −1 kcI k 0 
+ h i ,
cM∗ n +1 a k ∗ ( a )da 
Ī1 ( a0 )(h0 + d4 )(1 + M∗ )(l + M̄ ) d1 + l + ∑
R
M ∗ + k =1 a µ k ( a ) Ik k −1

which leads to a contradiction.

Mathematics 2022, 10, 2712 13 of 28

Furthermore, if Re(v) > 0, then it is easy to show

n +1  n +1 
δ  δ Λ 
1< S∗ ∑ Πik=−11 Vi Qk = ∑ Πik=−11 Vi Qk = 1, (38)
d 2 + δ k =1 d2 + δ d1 R0 k =1

which also leads to a contradiction.

Hence, it can be concluded that Equation (37) only has eigenvalues with negative real
parts and the endemic equilibrium P∗ is locally stable.

Theorem 3. When R0 > 1, if C1 > C2 and C3 > 0, then the endemic equilibrium P∗ is globally
asymptotically stable.

clα[αM∗ S∗ +ζ (Λ+η M )] cα2 ζ 2

 h i
 C1 = (αl +Λ+η M0 )(Λ+η M00 ) + Λ+η M0 αh0 S∗ + d4 (Λ + η M0 ) + αl +Λ+ηM0 ,
 ζ





cl [αζ M∗ +S∗ (Λ+η M0 )] cα2 ζ 2
h i
A∗ ∗ + d (Λ + η M ) + (39)
 C 2 = ∗
αζ (l + M )(l +ζ )
+ Λ+η M0 αh 0 S 4 0 αl +Λ+η M0 ,




C3 = α(Λ + η M0 )(S∗ + Ik∗ ( a)) − (Λ + η M0 )2 − α2 S∗ Ik∗ ( a).



Proof. First, we define the following Wi (t) (i = 1, 2, · · · , 5) as follows:

S(t) E(t)
W1 (t) = S(t) − S∗ − S∗ ln , W2 (t) = E(t) − E∗ − E∗ ln ,
S∗ E∗
A(t) M(t)
W3 (t) = A(t) − A∗ − A∗ ln ∗ , W4 (t) = M(t) − M∗ − M∗ ln ,
A M∗
Z a  
k I ( a, t)
W5 (t) = ψk ( a) Ik ( a, t) − Ik∗ ( a) − Ik∗ ( a) ln k ∗ da, k = 1, · · · , n + 1,
a k −1 Ik ( a)

where
Z a
k gk ( τ )
ψk ( a) = [µk (τ )S∗ + ψk+1 β k (τ )] dτ, k = 1, · · · , n,
a gk ( a )
Z a ∗
n +1 β
n +1 ( τ ) S g n +1 ( τ )
ψn+1 ( a) = dτ.
a gn +1 ( a )
By calculating the time derivatives of Wi (t) (i = 1, 2, 3, 4), it yields that

S ∗ )2

dW1 (t) +1 a k
= − d1 (S(St()− + ∑nk= ∗ ∗ ∗
R
ak−1 µk ( a )[( S − S ( t )) Ik ( a, t ) + S Ik ( a )]da



 dt t ) 1

cl (S(t)−S∗ )( M∗ − M (t))
 ∗2 +1 a k
− SS(t) ∑nk= µk ( a) Ik∗ ( a)da + (l + M∗ )S(t)(l + M(t))

 R



 1 a k −1

 h i
∗ n +1 a k
νk ( a)( Ik ( a,t)− Ik∗ )da+h0 ( A(t)− A∗ )

(S(t)−S ) ∑k=1 a
 R

k −1
+ ,


S(t)





+1 a k
( E(t)− E∗ ) ∑nk=
 R
dW2 (t) 1 ak−1 µk ( a ) S ( t ) Ik ( a,t )da (40)
 dt = E(t)
− (d2 + δ)( E(t) − E∗ ),


(t)( A(t)− A∗ )

dW3 (t)
= cS(tA) M − (h0 + d4 )( A(t) − A∗ ),





 dt ( t )( l + M (t))


+1 a k
( M(t)− M∗ ) ∑nk= ∗
 R
η ( M (t)− M∗ )2 1 ak−1 rk ( a )( Ik ( a,t )− Ik )da

 dW4 (t)



 dt = − M(t)
+ M(t)


t)− M∗ ) n+1 R ak rk ( a)θk ( a) A∗ Ik∗ ( a)
  
− ( M(M rk ( a)θk ( a) A(t) Ik ( a,t)

∑ − da.


(t) k =1 a ω ( a)+ A(t) ω ( a)+ A ∗
k −1 k k

According to (8), W5 (t) can be rewritten as follows,

Mathematics 2022, 10, 2712 14 of 28

h i
R a k −1 + t I (a ,t+ a − a ) gk ( a )
W5 (t) = a k −1 ψk ( a) Ik ( ak−1 , t + ak−1 − a) gk ( a) − Ik∗ ( a) − Ik∗ ( a) ln k k−1 I ∗ k(−a)1 da
k
 
R ak Ik0 ( a−t− ak−1 ) gk ( a) 0
∗ ( a ) − I ∗ ( a ) ln Ik ( a−t− ak−1 ) gk ( a) da.
+ a k −1 ψ
+t k ( a ) g ( a−t− a )
− Ik k I ∗ ( a) g ( a−t− a )
k k −1 k k k −1

Based on mathematical formulations of ψk ( a) defined above and Vk and Qk defined in

(5), it follows from simple computations that

n +1
I1∗ ( a0 )ψ1 ( a0 ) = S∗ ∑ Ik∗ ( a0 ) Qk . (41)
k =1

By differentiating W5 (t) with respect to t, it yields that

h i
dW5 (t) I (a ,t)
dt = ψk ( ak−1 ) Ik ( ak−1 , t) − Ik∗ ( ak−1 ) − Ik∗ ( ak−1 ) ln Ik∗ (ak−1 )
k k −1
R ak h dψk (a) ih
I (a ,t)
i
+ a dt − ( β k ( a) + νk ( a))φk ( a) Ik ( ak−1 , t) − Ik∗ ( a) − Ik∗ ( a) ln k I ∗k(−a1) da
k −1 k
h i (42)
I ( a ,t )
= ψk ( ak−1 ) Ik ( ak−1 , t) − Ik∗ ( ak−1 ) − Ik∗ ( ak−1 ) ln Ik∗ (ak−1 )
k k −1
h i
R ak I (a ,t)
− a [S∗ µk ( a) + β k ( a)ψk+1 ( ak )] Ik ( ak−1 , t) − Ik∗ ( a) − Ik∗ ( a) ln k I ∗k(−a1) da.
k −1 k

According to (42), some expressions are obtained as follows,

(i) when k = 1, it follows from I1 ( a0 , t) = δE(t) and I1∗ ( a0 ) = δE∗ that
h i
I1 ( a0 ,t)
ψ1 ( a0 ) I1 ( a0 , t) − I1∗ ( a0 ) − I1∗ ( a0 ) ln I1∗ ( a0 )

d2 + δ ∗ ∗ I1 ( a0 ,t)
= δ [ I1 ( a0 , t ) − I1 ( a0 )] − ψ1 ( a0 ) I1 ( a0 ) ln I1∗ ( a0 ) (43)
R ak I1 ( a0 ,t)
+1
= (d2 + δ)( E(t) − E∗ ) − ∑nk= 1 a k −1 µk ( a)S∗ Ik∗ ( a) ln I1∗ ( a0 )
da,

(ii) when k = n + 1, it follows from (42) that

h i
dW5 (t)
dt = ψn+1 ( an ) In+1 ( an , t) − In∗+1 ( an ) − In∗+1 ( an ) ln IIn∗+1 ((aann,t))
n +1
R a n +1 ∗ h i (44)
− an S µn+1 ( a) In+1 ( a, t) − In+1 ( a) − In+1 ( a) ln IIn∗+1 ((a,t
∗ ∗
a)
)
da.
n +1

(iii) For k = 1, 2, · · · , n + 1, it follows from (43) and (44) that

 
∗ ∗ Ik+1 ( ak ,t)
ψk+1 ( ak ) Ik+1 ( ak , t) − Ik+1 ( ak ) − Ik+1 ( ak ) ln I ∗ (a )
k +1 k
h i
R ak I (a ,t)
− a [S∗ µk ( a) + β k ( a)ψk+1 ( ak )] Ik ( ak−1 , t) − Ik∗ ( a) − Ik∗ ( a) ln k I ∗k(−a1) da
k −1 k
  (45)
R ak I ( a,t ) I ( a ,t )
= a ψk+1 ( ak ) β k ( a) Ik∗ ( a) ln Ik∗ (a) − ln Ik∗+1 (ak ) da
k −1 k k +1 k
R ak ∗ h i
− a S µk ( a) Ik ( a, t) − Ik∗ ( a) − Ik∗ ( a) ln IIk∗((a,t a)
)
da.
k −1 k

Let W (t) = ∑5i=1 Wi (t), it follows from (40) and (42)–(45) that
Mathematics 2022, 10, 2712 15 of 28

d1 ( S ( t ) − S ∗ )2 η ( M ( t ) − M ∗ )2 n +1 a k E∗ S(t) Ik ( a, t)
Z  
Ẇ (t) = − − +∑ ∗ ∗
µk ( a) S Ik ( a) − da
S(t) M(t) k =1 a k −1
E(t)
n +1 Z a k
S∗
 
I ( a , t) I ( a, t)
+∑ µk ( a)S∗ Ik∗ ( a) 1 − − ln 1 ∗ 0 + ln k ∗ da
k =1 a k −1
S(t) I1 ( a0 ) Ik ( a)
cl ( M∗ − M(t))(S(t) − S∗ ) h0 S ∗
 
cS(t) M(t)
+ − + d4 + ( A(t) − A∗ )
(l + M∗ )S(t)(l + M(t)) S(t) A(t)(l + M (t))
" #
n Z ak
Ik ( a, t) Ik+1 ( a, t)
+∑ ∗
ψk+1 ( ak ) β k ( a) Ik ( a) ln ∗ − ln ∗ da
k =1 a k −1
Ik ( a) Ik+1 ( a)
n +1 Z a k 
νk ( a)(S(t) − S∗ ) rk ( a)( M (t) − M∗ )

+∑ + ( Ik ( a, t) − Ik∗ ( a))da
k =1 a k − 1 S ( t ) M ( t )
( M(t) − M∗ ) n+1 ak rk ( a)θk ( a) A(t) Ik ( a, t) rk ( a)θk ( a) A∗ Ik∗ ( a)
Z  
−
M(t) ∑ ωk ( a ) + A ( t )
−
ωk ( a ) + A∗
da
k =1 a k −1

d ( S ( t ) − S ∗ )2 η ( M ( t ) − M ∗ )2 n +1 a k I ∗ ( a0 )S(t) Ik ( a, t)
Z  
= − 1 − +∑ µk ( a)S∗ Ik∗ ( a) 1 − 1 ∗ ∗ da
S(t) M(t) k =1 a k −1
S Ik I1 ( a0 , t)
n +1 Z a k
S∗
 
I1 ( a0 , t) Ik ( a, t)
+∑ ∗ ∗
µk ( a)S Ik ( a) 1 − − ln ∗ + ln ∗ da
k =1 a k −1
S(t) I1 ( a0 ) Ik ( a)
cl ( M∗ − M(t))(S(t) − S∗ ) h0 S ∗
 
cS(t) M(t)
+ − + d 4 + ( A(t) − A∗ ) (46)
(l + M∗ )S(t)(l + M(t)) S(t) A(t)(l + M (t))
Ik∗+1 ( ak ) Ik ( a, t)
" #
n Z ak
Ik ( a, t) Ik+1 ( a, t)
+∑ ∗
ψk+1 ( ak ) β k ( a) Ik ( a) 1 − ∗ + ln ∗ − ln ∗ da
k =1 a k −1
Ik ( a) Ik+1 ( ak , t) Ik ( a) Ik+1 ( a)
n +1 Z a k 
νk ( a)(S(t) − S∗ ) rk ( a)( M (t) − M∗ )

+∑ + ( Ik ( a, t) − Ik∗ ( a))da
a
k =1 k −1
S ( t ) M ( t )
( M(t) − M∗ ) n+1 ak rk ( a)θk ( a) A(t) Ik ( a, t) rk ( a)θk ( a) A∗ Ik∗ ( a)
Z  
−
M(t) ∑ ωk ( a ) + A ( t )
−
ωk ( a ) + A∗
da
k =1 a k −1

d1 ( S ( t ) − S ∗ )2 η ( M ( t ) − M ∗ )2 n +1 a k I1∗ ( a0 )S(t) Ik ( a, t)
Z  
= − − +∑ ∗ ∗
µk ( a)S Ik ( a) 1 − da
S(t) M(t) k =1 a k −1
S∗ Ik∗ I1 ( a0 , t)
n +1 Z a k
S∗
 
I ( a , t) I ( a, t)
+∑ µk ( a)S∗ Ik∗ ( a) 1 − − ln 1 ∗ 0 + ln k ∗ da
k =1 a k −1
S(t) I1 ( a0 ) Ik ( a)
cl ( M∗ − M(t))(S(t) − S∗ ) h0 S ∗
 
cS(t) M(t)
+ − + d4 + ( A(t) − A∗ )
(l + M∗ )S(t)(l + M(t)) S(t) A(t)(l + M (t))
Ik∗+1 ( ak ) Ik ( a, t)
" #
n Z ak
Ik ( a, t) Ik+1 ( a, t)
+∑ ∗
ψk+1 ( ak ) β k ( a) Ik ( a) 1 − ∗ + ln ∗ − ln ∗ da
k =1 a k −1
Ik ( a) Ik+1 ( ak , t) Ik ( a) Ik+1 ( a)
n +1 Z a k 
νk ( a)(S(t) − S∗ ) rk ( a)( M (t) − M∗ )

+∑ + ( Ik ( a, t) − Ik∗ ( a))da
k =1 a k − 1 S ( t ) M ( t )
( M(t) − M∗ ) n+1 ak rk ( a)θk ( a) A(t) Ik ( a, t) rk ( a)θk ( a) A∗ Ik∗ ( a)
Z  
−
M(t) ∑ ωk ( a ) + A ( t )
−
ωk ( a ) + A∗
da
k =1 a k −1
Mathematics 2022, 10, 2712 16 of 28

d1 ( S ( t ) − S ∗ )2 η ( M ( t ) − M ∗ )2 n +1 a k S∗ S∗
Z  
= − − +∑ µk ( a)S∗ Ik∗ ( a) 1 − + ln da
S(t) M(t) k =1 a k −1
S(t) S(t)
n +1 Z a k S∗ I1∗ ( a0 ) Ik ( a, t) S∗ I1∗ ( a0 ) Ik ( a, t)
 
+∑ ∗ ∗
µk ( a)S Ik ( a) 1 − + ln da
k =1 a k −1
S(t) Ik∗ ( a) I1 ( a0 , t) S(t) Ik∗ ( a) I1 ( a0 , t)
cl ( M∗ − M(t))(S(t) − S∗ ) h0 S ∗
 
cS(t) M(t)
+ − + d 4 + ( A(t) − A∗ )
(l + M∗ )S(t)(l + M(t)) S(t) A(t)(l + M (t))
Ik∗+1 ( ak ) Ik ( a, t) Ik∗+1 ( ak ) Ik ( a, t)
" #
n Z ak
+∑ ∗
ψk+1 ( ak ) β k ( a) Ik ( a) 1 − ∗ + ln ∗ da
k =1 a k −1
Ik ( a) Ik+1 ( ak , t) Ik ( a) Ik+1 ( ak , t)
n +1 Z a k 
νk ( a)(S(t) − S∗ ) rk ( a)( M (t) − M∗ )

+∑ + ( Ik ( a, t) − Ik∗ ( a))da
a
k =1 k −1
S ( t ) M ( t )
( M(t) − M∗ ) n+1 ak rk ( a)θk ( a) A(t) Ik ( a, t) rk ( a)θk ( a) A∗ Ik∗ ( a)
Z  
−
M(t) ∑ ωk ( a ) + A ( t )
−
ωk ( a ) + A∗
da.
k =1 a k −1

Since it is easy to show that

S∗ I1∗ ( a0 ) Ik ( a, t) S∗
> 0, > 0,
S(t) Ik∗ ( a) I1 ( a0 , t) S(t)

which follows that

S∗ ∗

 1−
S(t)
+ ln SS(t) ≤ 0,
S∗ I1∗ ( a0 ) Ik ( a,t) S∗ I ∗ ( a ) I ( a,t) (47)
 1−
S(t) Ik∗ ( a) I1 ( a0 ,t)
+ ln S(t) 1I ∗ (a0) I k(a ,t) ≤ 0.
k 1 0

According to (47), Lemma 1 and Theorem 1, it follows from simple computations that
Equation (46) can be rewritten as follows,

dW5 (t) d1 ( S ( t ) − S ∗ )2 η ( M ( t ) − M ∗ )2 clα[αM∗ S∗ + ζ (Λ + η M0 )]

≤ − − −
dt S(t) M(t) (l + M∗ )(αl + Λ + η M0 )(Λ + η M0 )
cl [αζ M∗ + S∗ (Λ + η M0 )] n+1 ak (rk ( a) + µk ( a))(Λ + ηM0 )(S∗ + Ik∗ ( a))
Z
+ −∑ da
αζ (l + M∗ )(l + ζ ) k =1 a k −1
αζ
αh0 S∗ + d4 (Λ + ηM0 ) cα2 ζ 2
 
+ + ( A∗ − ζ ) (48)
Λ + η M0 (Λ + ηM0 )(αl + Λ + η M0 )
n +1 Z a k (rk ( a) + µk ( a))[(Λ + η M0 )2 + α2 S∗ Ik∗ ( a)]
+ ∑ α2 ζ
da
k =1 a k −1
n +1 Z a k
rk ( a)θk ( a)( M∗ − ζ ) αζ 2 A∗ Ik∗ ( a)
 
+ ∑ ζ αωk ( a) + Λ + η M0
−
ωk ( a ) + A∗
da.
k =1 a k −1

If C1 > C2 and C3 > 0, where Ci (i = 1, 2, 3) is defined in (39), then it is easy to show

dW5 (t)
that dt < 0.
Hence, it follows from the Lyapunov–LaSalle stability theorem [45] that the endemic
equilibrium P∗ is globally asymptotically stable.

4. Optimal Control Strategy

In this section, we formulate and analyze an optimal control strategy considering
the diminution of media coverage and treatment to control the disease progression and
balance the economic loss generated due to infectious disease and treatment measures.
We will adopt time and age-dependent treatment ρ( a, t) on the infected population and a
time-dependent diminution rate of media coverage η (t) on the cumulative number of social
media and TV advertisements. Hence, the controlled system takes the following form,
Mathematics 2022, 10, 2712 17 of 28

 R ak
dS(t) cS(t) M (t)


 dt
+1
= Λ + ∑nk= 1 ak−1 [ νk ( a ) + ρ ( a, t ) − µk ( a ) S ( t )] Ik ( a, t )da − l + M (t) + h0 A ( t ) − d1 S ( t ),


 R ak
dE(t) +1
= ∑nk=




 dt 1 a k −1 µk ( a)S(t) Ik ( a, t)da − (δ + d2 ) E(t),



∂Ik ( a,t) ∂Ik ( a,t)
∂a + ∂t = −[νk ( a) + d3k ( a) + β k ( a) + ρ( a, t)] Ik ( a, t), a ∈ ( a k −1 , a k ] (49)



 dA(t) cS(t) M (t)
= − ( h0 + d4 ) A ( t ),




 dt l + M(t)

 h i
dM (t) +1 a k θ ( a) A(t)
= ∑nk=
 R
rk ( a) 1 − ω k(a)+ A(t) Ik ( a, t)da − η (t)( M (t) − M0 ),


dt 1 a k −1 k

where system (49) shares the same boundary and initial values with system (1), other
parameters share the same mathematical forms and biological interpretations described in
system (1).
The proper treatment measures and efficient media coverage significantly minimize the
serious consequences of infection associated with wealth loss and mortality due to disease
burden. Our aim is to optimize the use and minimize the cost of these two mentioned
measures, and the optimal control problem is formulated as follows:

H (ρ∗ , η ∗ ) = min H (ρ( a, t), η (t)), (50)

ρ,η ∈G

where G and H (ρ( a, t), η (t)) are defined as follows,

     0≤ρ≤ρ , 
G= (ρ( a, t), η (t)) ∈ L∞ ∪nk=
+1
( a ,
1 k −1 ka ] × ( 0, T ) , L ∞
( 0, T ) max
, (51)
0 ≤ η ≤ ηmax

where ρmax and ηmax are positive upper constants of ρ( a, t) and η (t), respectively, and where
Z T n +1 Z a
!
k 1
H (ρ( a, t), η (t)) = ∑ 2
γ11 Ik ( a, t) + γ12 ρ( a, t) Ik ( a, t) + γ21 ρ ( a, t)da dt
2
(52)
0 k =1 a k −1
Z T 
1 2
+ γ13 M (t) + γ22 η (t) dt,
0 2

where state variables are all subject to system (49), γ11 and γ13 are positive weights with
respect to disease infection and the cumulative amount of media coverage. γ12 represents
the positive unit cost of controls at different levels. γ21 and γ22 denote positive constants
associated with two quadratic terms accounting for control costs of disease infection and
media coverage.
In order to derive the optimal problem (50) with respect to time-dependent and
age-dependent control functions, some derivations of the adjoint system (53) and the
characterization of the optimal control functions are discussed as follows.

dS(t) +1 a k
= Λ + ∑nk=
R

dt 1 a [νk ( a) + ρ( a, t) + σ f 1 ( a, t) − µk ( a)S(t)] Ik ( a, t)da
k −1





− cSl +(tM
) M(t)




 (t)
+ h0 A ( t ) − d1 S ( t ),



dE(t) n +1 a k
= ∑ k =1
 R


 dt a k −1 µk ( a)S(t) Ik ( a, t)da − (δ + d2 ) E(t),
(53)
∂I ( a,t) ∂I ( a,t)
φ k∂a + k∂t = −[νk ( a) + d3k ( a) + β k ( a) + ρ( a, t) + σ f 1 ( a, t)] Ik ( a, t), a ∈ ( ak−1 , ak ]







 dA(t) cS(t) M (t)
dt = l + M (t) − ( h0 + d4 ) A ( t ),






 h i
 dM(t) = ∑n+1 ak r ( a) 1 − θk (a) A(t) I ( a, t)da − (η (t) + σ f 2 (t))( M (t) − M0 ),

 R
dt k =1 a k −1 k ωk ( a)+ A(t) k
Mathematics 2022, 10, 2712 18 of 28

where σ denotes a scalar, φ is a constant balancing of the different units between time and
age and f 1 ( a, t) and f 2 (t) are arbitrary functions. ρ( a, t) and η (t) in system (49) are replaced
with ρ( a, t) + σ f 1 ( a, t) and η (t) + σ f 2 (t) in adjoint system (53), respectively.
Since the solutions of system (53) are associated with σ, we will differentiate system (53)
with respect to σ at σ = 0 and the system for sensitivities are obtained as follows,
dF S (t)

+1 a k
= ∑nk= I S
R  
1 ak−1 ( νk ( a ) + ρ ( a, t ) − µk ( a ) S ( t )) Fk ( a, t ) + ( f 1 ( a, t ) − µk ( a ) F ( t )) Ik ( a, t ) da



 dt

 h S
F M (t)S(t)
i
−c Fl +(tM
) M(t)

+ + h0 F A ( t ) − d1 F S ( t ),



 (t) (l + M(t)) 2



dF E (t) R ak

+1
= ∑nk= µk ( a)[ FkI ( a, t)S(t) + F S (t) Ik ( a, t)]da − (δ + d2 ) F E (t),


dt 1 a k −1





∂F I ( a,t) ∂F I ( a,t) (54)
 φ k∂a + k∂t = −[νk ( a) + d3k ( a) + β k ( a) + ρ( a, t)] FkI ( a, t) − f 1 (t) Ik ( a, t), a ∈ ( a k −1 , a k ]



dF A (t)
h S
F M (t)S(t)
 i
F (t) M(t)
− ( h0 + d4 ) F A ( t ),




 dt = c l + M(t)
+ (l + M(t)) 2


dF M (t) A
 h  i
n +1 a k θk ( a) A(t) I ( a, t ) − θk ( a) F (t) I ( a, t ) da

∑
R
= r ( a ) 1 − F

k k
 2


 dt k =1 a k −1 ωk ( a)+ A(t) k (ωk ( a)+ A(t))


 M
− η ( t ) F ( t ) − f 2 ( t ) M ( t ),

where F S , F E , FkI , F A and F M denote the derivatives of S, E, Ik , A and M with respect to σ

and are evaluated at σ = 0 (k = 1, · · · , n + 1), respectively.
In order to facilitate deriving adjoint system (54), we introduce the adjoint variable
vectors λ( a, t) = (λS (t), λ E (t), λ1I ( a, t), · · · , λnI +1 ( a, t), λ A (t), λ M (t)) T . By projecting an
adjoint operator on both sides of system (54), it yields that
!
Z Z T n +1 ak
∑ λ( a, t)L x ( a, t)da dt
0 k =1 a k −1
!
Z T n +1 Z a k  
∂ ∂
=
0
∑ a k −1
λ( a, t) φ +
∂a ∂t
x ( a, t)da dt
k =1
!
Z T n +1 Z ak Z T
+ ∑ λ( a, t)Γ1 x ( a, t)da dt + Γ2 x ( a, t)dt, (55)
0 k =1 a k −1 0

where x ( a, t) = (S(t), E(t), I1 ( a, t), · · · , In+1 ( a, t), A(t), M (t)) T , L x ( a, t) and Γi x ( a, t) (i =

1, 2) are defined as follows,

L x ( a, t) = ( f 1 ( a, t) Ik ( a, t), 0, − f 1 ( a, t) I1 ( a, t), · · · , − f 1 ( a, t) In+1 ( a, t), 0, − f 2 (t) M(t))T ,

and
 
µk ( a) F S (t) Ik ( a, t) − [νk ( a) + ρ( a, t) − µk ( a)S(t)] FkI ( a, t)

 −µk ( a)[ FkI ( a, t)S(t) + F S (t) Ik ( a, t)] 


 −[ν1 ( a) + d31 ( a) + β 1 ( a) + ρ( a, t)] F1I ( a, t) 

 .. 
Γ1 x ( a, t) =  . ,
 
−[νn+1 ( a) + d3n+1 ( a) + β n+1 ( a) + ρ( a, t)] FnI +1 ( a, t)
 
 
 
 0 
θk ( a) F A (t)
 h  i 
θk ( a) A(t)
rk ( a) (ωk ( a)+ A(t))2 k
I ( a, t) − 1− ωk ( a)+ A(t)
FkI ( a, t)

and
Mathematics 2022, 10, 2712 19 of 28

 h F S (t) M(t) F M (t)S(t)

i 
c l + M(t) + (l + M(t))2 − h0 F A (t) + d1 F S (t)
( δ + d2 ) F E ( t )
 
 
 
 0 
..
 
Γ2 x ( a, t) =  ,
 
 . 

 0
h S


F M (t)S(t)
i
 (h0 + d4 ) F A (t) − c Fl +(tM ) M(t)
 
(t)
+ (l + M(t)) 2 
η (t) F M (t)
which derives that
!
Z T n +1 Z a k  
∂ ∂
∑ λkI φ +
∂a ∂t k
I
F ( a, t)da dt
0 k =1 a k −1
!
Z T Z T n +1 Z a k ∂λkI I
= −φ FkI (0, t)λ I (0, t)dt − φ ∑ F ( a, t)da dt
∂a k
0 0 k =1 a k −1
!
Z T n +1 Z a k ∂λkI I
−φ ∑ F ( a, t)da dt.
∂t k
(56)
0 k =1 a k −1

According to λkI (t, an+1 ) = λkI ( T, a) = 0, FkI (0, a) = 0 (k = 1, · · · , n + 1), it yields

!
Z T n +1 Z a k  
∂ ∂
∑ λ I ( a, t) φ +
∂a ∂t k
I ( a, t)da dt
0 k =1 a k −1
Z T n +1 Z a    !
k ∂ ∂
= − ∑ Ik ( a, t) φ +
∂a ∂t
I 0 I
λ ( a, t) + φIk Ik ( a, t)λ (0, t) da dt. (57)
0 k =1 a k −1

By differentiating the optimal objective functional H with respect to σ around σ = 0,

it can be obtained that
Z T n +1 Z a
!
dH k

dσ σ=0
= ∑ (γ12 Ik ( a, t) + γ21 ρ( a, t)) f 1 ( a, t)da dt
0 k =1 a k −1
!
Z Z T n +1 ak
+ ∑ (γ11 + γ12 ρ( a, t)) FkI ( a, t)da dt
0 k =1 a k −1
Z T
+ γ13 F M (t) + γ22 η (t) f 2 (t)dt, (58)
0

and
L∗ λ( a, t) = (0, 0, γ11 + γ12 ρ( a, t), 0, γ13 )T . (59)
For adjoint variable vectors λ( a, t) and optimal control variable (ρ( a, t), η (t)) ∈ G , it
follows from the above analysis of (56), (57) and (59) that the adjoint system is as follows,
 dλS (t) cM(t)(λS (t)−λ A (t))
+1 a k
= (λS (t) − λ E (t)) ∑nk= + d1 λ S ( t ),
R
dt 1 ak−1 µk ( a ) Ik ( a, t )da +


 l + M(t)
dλ E (t)

= ( δ + d2 ) λ E ( t ),


dt



 ∂λ I ( a,t ) ∂λ I ( a,t)
φ k∂a + k∂t = [νk ( a) + ρ( a, t) + d3k ( a) + β k ( a)]λRkI ( a, t) − γ11 − γ12 ρ( a, t)



+1 a k

+φIk0 u( a, t)λ I (0, t) − λS (t) ∑nk=

1 ak−1 [ νk ( a ) + ρ ( a, t )]da


S E n +1 a k
+(λ (t) − λ (t)) ∑k=1 a µk ( a)S(t)da
R (60)

 h k −1 i
M ( t ) n+1 ak r ( a ) 1 − θk ( a) A(t) da,

∑
R
− a ∈ ( a k −1 , a k ]


 λ k = 1 a k ω ( a )+ A ( t )
k −1 k


A
+1 a k r k ( a ) θ k ( a ) ω k ( a )

 dλdt(t) = −h0 λS (t) + (h0 + d4 )λ A (t) + λ M (t) ∑nk=
 R
1 ak−1 (ωk ( a)+ A(t))2 da,



 dλ M (t)

lS ( t )

dt = (λS (t) − λ A (t)) (l + M(t))2 + λ M (t)η (t) − γ13 ,
Mathematics 2022, 10, 2712 20 of 28

where the transversality conditions are as follows:

λS ( T ) = λ E ( T ) = λ A ( T ) = λ M ( T ) = 0, λkI ( a, T ) = 0,

which hold for k = 1, · · · , n + 1 and a ∈ ( ak−1 , ak ], and boundary condition λkI ( an+1 , t) = 0
holds for t ∈ (0, T ).
It is easy to show that adjoint system (60) hasa mild solution with Lipschitz
 continuity,

which takes the form (λS , λ E , λkI , λ A , λ M ) ∈ L∞ ∪nk=
+1 ∞
1 ( ak −1 , ak ] × (0, T ) , L (0, T ) .

Theorem 4. If (ρ∗ , η ∗ ) is a pair of optimal control subjects to optimal control problem (50), then
the characterization of optimal control (ρ∗ , η ∗ ) takes the following form

ρ∗ = min{max{0, ρ̃}, ρmax }, η ∗ = min{max{0, η̃ }, ηmax }, (61)

   
where (ρ∗ , η ∗ ) ∈ L∞ ∪nk=+1 ∞
1 ( ak −1 , ak ] × (0, T ) , L (0, T ) , ρ̃ and η̃ are defined in (63).

∂H (ρ,η )
Proof. By transferring the first part of ∂σ on adjoint functions, it yields
!
Z T n +1 Z a k
∂H (ρ, η )
∂σ
= ∑ λ( a, t)(0, 0, γ11 + γ12 ρ( a, t), 0, γ13 ) T da dt
0 k =1 a k −1
!
Z T n +1 Z a k Z T
+ ∑ γ12 f 1 ( a, t) + γ21 f 1 ( a, t)ρ( a, t)da dt + γ22 f 2 (t)η (t)dt
0 k =1 a k −1 0
!
Z T n +1 Z a k h i
= ∑ f 1 ( a, t) γ12 + γ21 ρ( a, t) + (λ S
(t) − λkI ( a, t)) Ik ( a, t) da dt
0 k =1 a k −1
Z T
+ f 2 (t)(γ22 η (t) − M (t)λ M (t))dt. (62)
0

Based on (62), the characterization of the optimal control is as follows,

(λkI ( a, t) − λS (t)) Ik ( a, t) − γ12 M(t)λ M (t)

ρ̃( a, t) = , η̃ (t) = . (63)
γ21 γ22

Consequently, it follows from the above analysis that a pair of optimal controls can be
obtained as follows: ρ∗ = min{max{0, ρ̃}, ρmax }, η ∗ = min{max{0, η̃ }, ηmax }.

Remark 3. Generally, it is difficult to solve the sequence of optimal controls and the associated
state variables converging with the optimal controls due to a lack of compact characters in the
age-structured first-order PDEs. Hence, Ekeland’s principle [51] will be utilized to minimize the
sequences of the approximate functionals.
According
 to the corresponding analysis
 in Chapter 14 [52] that there exists a pair of controls
(ρε , ηε ) ∈ L1 ∪nk= +1 1
1 ( ak −1 , ak ] × (0, T ) × L (0, T ), and the optimal control objective functional
is defined as follows,
√  
Hε (ρ, η ) = H (ρ, η ) + ε kρε − ρk L1 (∪n+1 (a ,a ]×(0,T )) + kηε − η k L1 (0,T ) .
k =1 k −1 k

If (ρε , ηε ) is a pair of optimal control subjects to the optimal control problem minimizing
Hε (ρ, η ), then it follows from the corresponding analysis in Chapter 14 [52], it is easy to show
   √  
(λ I ( a,t)−λSε (t)) Ikε ( a,t)−γ12 − εv1ε
 ρε ( a, t) = min max 0, kε , ,


γ21 ρ max
  √   (64)
Mε (t)λεM (t)− εv2ε
 ηε (t) = min max 0, , ,


γ22 η max
Mathematics 2022, 10, 2712 21 of 28

 
where (v1ε , v2ε ) ∈ L∞ ∪nk=
+1
( a ,
1 k −1 ka ] × ( 0, T ) × L∞ (0, T ) and |viε | ≤ 1 (i = 1, 2) hold for all
+1
( a, t) ∈ ∪nk= 1 ( ak −1 , ak ] × (0, T ).

Theorem 5. If γT (i = 1, 2) is sufficiently small, then there exists a pair of unique optimal controls
2i
(ρ∗ , η ∗ ) subject to optimal control problem (50).

Proof. First, two functions are defined as follows,

    
I S
 χ1 (ρ) = min max 0, (λk (a,t)−λ (t)) Ik (a,t)−γ12 , ρmax ,

γ21
n n M (t)
o o (65)
 χ2 (η ) = min max 0, M ( t ) λ
, ηmax .

γ22

For two pairs of controls, (ρ1 , η1 ) and (ρ2 , η2 ), according to L∞ Lipschitz estimations
for the states and adjoint functions, it yields that

G
e1 T G
e2 T
kχ1 (ρ1 ) − χ1 (ρ2 )k ≤ k ρ − ρ2 k L ∞ , kχ2 (η1 ) − χ2 (η2 )k ≤ k η − η2 k L ∞ ,
γ21 1 γ22 1

where G ei (i = 1, 2) are positive Lipschitz constants associated with the essential bounded-
ness of the state variables and adjoint functions.
Consequently, if γT (i = 1, 2) is sufficiently small, then it yields
2i

√ √ √ √
ε 1 ε ε 1 ε
kρ − ρε k ≤   = , k η − ηε k ≤   = .
γ21 1−G
e1 T γ21 − G
e1 T γ22 1−G
e2 T γ22 − G
e2 T
γ21 γ22

which reveals that (ρε , ηε ) converges to the optimal control (ρ∗ , η ∗ ). It follows from Eke-
land’s principle [51] that there exists a unique pair of optimal controls when the time period
is sufficiently small, and H (ρ∗ , η ∗ ) ≤ inf(ρ,η )∈G H (ρ, η ) as ε → 0.

5. Numerical Simulation
In this section, some numerical simulations are conducted to show the feasibility of
our analysis in the previous section regarding global stability and optimal control strategy,
where values of parameters in numerical simulations are partially taken from [18,31,37].
When progressing to the deterioration of diseases, it is well known that infectivity
during the acute state is greater than that of the chronic one, and infection awareness
among susceptible individuals increases due to obvious symptoms of chronic patients. In
order to indicate varying degrees of transmissibility at the acute stage (from 0 to 2 months)
and chronic stage (from 2 to 10 months), the age-dependent coefficients at different stages
are as follows:
 
0.2483, 0 < a ≤ 2, 0, 0 < a ≤ 2,
µ1 ( a ) = µ2 ( a ) = − 0.5 ( a − 0.5 )
0, 2 < a ≤ 10. 0.2483e , 2 < a ≤ 10.
(
0, 0 < a ≤ 2,

0.03, 0 < a ≤ 2,
d31 ( a) = d32 ( a) = 0.03 + 1
, 2 < a ≤ 10.
0, 2 < a ≤ 10. 2
1+e(11−0.17( a−0.5)+0.007( a−0.5) )

4(1 − 0.95e−0.7877a ), 0 < a ≤ 2,

 
0, 0 < a ≤ 2,
β1 ( a) = β2 ( a) =
0, 2 < a ≤ 10. 0.025e−0.7877a , 2 < a ≤ 10.
3.8e−0.7877a , 0 < a ≤ 2,
 
0, 0 < a ≤ 2,
ν1 ( a) = ν2 ( a) =
0, 2 < a ≤ 10. 0.025, 2 < a ≤ 10.
Mathematics 2022, 10, 2712 22 of 28

By using the set of parameter values, Λ = 1.7, d1 = 0.025, d2 = 0.01, δ = 1, l = 1200,

h0 = 0.025, η = 0.025, the basic reproduction number R0 due to variations of c and M0 are
plotted in Figure 1a,b, respectively. From this figure, it may be noted that R0 decreases
with the increasing dissemination rate of awareness (c), as well as the increasing baseline
number of social media and TV advertisements (M0 ). R0 becomes less than unity with the
appropriate c and M0 , which shows the importance of public health alerts in controlling
the spread of infectious disease.
By adopting Λ = 1.7, d1 = 0.025, d2 = 0.01, δ = 1, l = 1200, h0 = 0.025, η = 0.025,
c = 0.15, M0 = 5 and other parameters given as follows
 
0.006, 0 < a ≤ 2, 0, 0 < a ≤ 2,
r1 ( a ) = r2 ( a ) =
0, 2 < a ≤ 10. 0.006e(1.2−0.9a) , 2 < a ≤ 10.
(
0 < a ≤ 2,

0.005, 0 < a ≤ 2, 0,
θ1 ( a ) = θ2 ( a ) = 2
0, 2 < a ≤ 10. 0.005e(0.8−5a ) , 2 < a ≤ 10.
(
0 < a ≤ 2,

600, 0 < a ≤ 2, 0,
ω1 ( a ) = ω2 ( a ) = 2
0, 2 < a ≤ 10. 600e(0.6−0.3a−(a−0.2) ) , 2 < a ≤ 10.
It is easy to show that R0 = 1.0234 > 1 and C1 > C2 and C3 > 0, it follows from
Theorem 3 that the endemic equilibrium P∗ (12.9042, 1.4669, 0.1451, 0.5112, 0.3085, 5) is
globally asymptotically stable. Figure 2 depicts the infectious disease becoming endemic
and each population converges to the corresponding steady state.
Dynamic impacts of infection age on the two successive infectious states (acute stage
from 0 to 2 months, chronic stage from 2 to 10 months) are shown in Figures 3 and 4. The
distributions of the infected individuals with respect to both age and time are plotted in
Figure 3, and the corresponding dynamical responses of I1 and I2 with respect to age are
plotted in Figure 4, which reflects the transmission heterogeneity of the infected individuals
in acute (Figure 4a) and chronic infectious stages (Figure 4b).
In the following part, numerical simulations are carried out to show the effects of opti-
mal controls on reducing infectious disease and minimizing the costs of control strategies,
which are formulated and analyzed in the third section of this paper. By extending the
forward-backward sweep method for ODE models, an iterative process is designed for
numerical simulations of state and adjoint systems in the age-structured PDE model (60).
In order to maintain relatively good numerical algorithm stability, a simple finite difference
scheme is selected to facilitate the numerical implementations. By fixing the weight param-
eter values with respect to disease infection and cumulative number of media coverage as
follows γ11 = 3.5, γ13 = 5, and varying the control costs by discussing three scenarios as
follows, (i) low optimal cost controls: γ12 = γ21 = γ22 = 0.35; (ii) moderate optimal cost
controls: γ12 = γ21 = γ22 = 3.5; (iii) high optimal cost controls: γ12 = γ21 = γ22 = 35. The
total number of infected individuals with respect to time at the acute stage is plotted in
Figure 5a, and at the chronic stage is plotted in Figure 5b. The total number of infected
individuals with respect to age at the acute stage is plotted in Figure 6a, and at the chronic
stage is plotted in Figure 6b. It follows from the curves of Figures 5 and 6 that the infected
individual with high control costs decreases quickly at the acute stage, which implies the
spread of infectious disease can be significantly reduced with the increase in control costs.
The infected individual with high control costs finally approaches a relatively steady state
(approximately 0) at the chronic stage.
Mathematics 2022, 10, 2712 23 of 28

(a) (b)
1.1 1.1
c=0.5 M =5
0
1.09 c=0.4 1.08 M0=4
c=0.3 M0=3
1.08
c=0.2 1.06
M0=2
1.07 c=0.1
M0=1
1.04
1.06
1.02

R0
R0

1.05
1
1.04
0.98
1.03

0.96
1.02

1.01 0.94

1 0.92
0 2 4 0 0.2 0.4
M0 c

Figure 1. (a) Description of basic reproduction number R0 due to variations of M0 for different values
of c (c = 0.1, 0.2, 0.3, 0.4, 0.5). (b) Description of basic reproduction number R0 due to variations of c
for different values of M0 (M0 = 1, 2, 3, 4, 5).

(a) (b) (c)

15 3 0.2

10 2 0.18
I1
S

5 1 0.16

0 0 0.14
0 500 1000 0 500 1000 0 500 1000
Time Time Time
(d) (e) (f)
0.8 2 6

1.5
0.6 4
I2

1
A

0.4 2
0.5

0.2 0 0
0 500 1000 0 500 1000 0 500 1000
Time Time Time

Figure 2. Infectious disease becomes endemic, and each population converges to the corresponding
steady state, which shows that endemic equilibrium P∗ (12.9042, 1.4669, 0.1451, 0.5112, 0.3085, 5) is
globally asymptotically stable. (a) The dynamical responses of S. (b) The dynamical responses of E.
(c) The dynamical responses of I1 . (d) The dynamical responses of I2 . (e) The dynamical responses of
A. (f) The dynamical responses of M.
Mathematics 2022, 10, 2712 24 of 28

1.5

Infected individual 1

0.5

0
0 500
2 400
4 300
6 200
8 100
10 0
Age Time#

Figure 3. Age distributions of infected individuals with respect to time and acute stage (from 0 to
2 months) and chronic stage (from 2 to 10 months).

(a) (b)
3.5 0.5

0.45
3
0.4

2.5 0.35

0.3
2
I1

I2

0.25

1.5
0.2

1 0.15

0.1
0.5
0.05

0 0
0 0.5 1 1.5 2 2 4 6 8 10
Age Age

Figure 4. (a) Age distributions of infected individuals at the acute stage. (b) Age distributions of
infected individuals at the chronic stage.
Mathematics 2022, 10, 2712 25 of 28

(a) (b)
0.45 0.25
with low cost controls with low cost controls
with moderate cost controls with moderate cost controls
0.4 with high cost controls with high cost controls
0.2

0.35

0.15
0.3
I1

I2
0.25
0.1

0.2

0.05
0.15

0.1 0
0 20 40 60 80 100 0 20 40 60 80 100
Time Time
Figure 5. The weight parameter values with respect to disease infection and the cumulative number
of media coverage are set as γ11 = 3.5, γ13 = 5, and three scenarios are as follows, (i) low optimal
cost controls: γ12 = γ21 = γ22 = 0.35; (ii) moderate optimal cost controls: γ12 = γ21 = γ22 = 3.5;
(iii) high optimal cost controls: γ12 = γ21 = γ22 = 35. (a) The total number of infected individuals
with respect to time at the acute stage. (b) The total number of infected individuals with respect to
time at the chronic stage.

(a) (b)
4.5 0.2

4 0.18

with low cost controls 0.16

3.5
with moderate cost controls with low cost controls
with high cost controls with moderate cost controls
0.14
3 with high cost controls

0.12
2.5
I1

I2

0.1
2
0.08

1.5
0.06

1
0.04

0.5 0.02

0 0
0 0.5 1 1.5 2 2 4 6 8 10
Age Age

Figure 6. The weight parameter values with respect to disease infection and cumulative number
of media coverage are set as γ11 = 3.5, γ13 = 5, and three scenarios are as follows, (i) low optimal
cost controls: γ12 = γ21 = γ22 = 0.35; (ii) moderate optimal cost controls: γ12 = γ21 = γ22 = 3.5;
(iii) high optimal cost controls: γ12 = γ21 = γ22 = 35. (a) The total number of infected individuals
with respect to age at the acute stage. (b) The total number of infected individuals with respect to age
at the chronic stage.
Mathematics 2022, 10, 2712 26 of 28

6. Conclusions
It is observed that social media and TV advertisements regarding the spread of infec-
tious diseases generally have potential effects on behavioral changes among susceptible
people as well as to control the spread of diseases. The infected population usually shows
different infecting forces during various stages of progress. Hence, it seems more reason-
able for the specific purpose of disease modeling to introduce infection age structure. In
this paper, we propose a mathematical model to discuss how media coverage impacts the
global stability and threshold dynamics of an infectious disease. In order to depict varying
degrees of transmissibility at different stages of infection, the continuous age structure for
each successive infectious stage during a long infective period is taken into account. Suffi-
cient conditions associated with the basic reproduction number are derived for uniform
persistence of solution in Theorem 1. By constructing appropriate Lyapunov functions,
it follows from the Lyapunov–LaSalle stability theorem that global stability analysis of
endemic equilibrium is investigated in Theorem 3.
As public health alerts spread among the susceptible population, behavioral changes
such as protective measures and vaccination will help healthy individuals to avoid con-
tracting infection. Consequently, we quantify the information-induced vaccination function
as a control measure to vaccinate susceptible individuals so that the maximum number
of individuals get vaccinated with minimum cost involved. Since the treatment includes
diagnosis, medication, hospitalization and other related hygienic efforts depending on
limited medical resources, the treatment function is chosen as another control for optimal
treatment. The total incurred cost is determined by taking a weighted sum of the cost
of productivity loss due to disease and costs incurred in applying control interventions.
Analytical characterization and the existence of optimal control paths is carried out in
Theorems 4 and 5, respectively.

Author Contributions: Conceptualization, C.L.; methodology, C.L. and P.C.; software, Q.J.; valida-
tion, C.L., P.C. and L.C.; investigation, C.L., P.C. and Q.J.; resources, Q.J.; writing—original draft
preparation, C.L. and P.C.; writing—review and editing, Q.J. and L.C.; supervision, L.C.; funding
acquisition, C.L. and P.C. All authors have read and agreed to the published version of the manuscript.
Funding: This work is supported by the National Natural Science Foundation of China, grant
No. 12171074. Hebei Natural Science Foundation, grant No. A2020501005. Natural Sciences and
Engineering Research Council of Canada (NSERC).
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: Data are available in a publicly accessible repository.
Acknowledgments: The authors would like to express their gratitude to the editor and anonymous
reviewers for their valuable comments and suggestions, and the time and effort they have spent on
the review. Without the expert comments made by the editor and anonymous reviewers, the paper
would not be of this quality.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Hyman, J.M.; Li, J.; Stanley, E.A. The differential infectivity and staged progression models for the transmission of HIV. Math.
Biosci. 1999, 155, 77–109. [CrossRef]
2. Guo, H.; Li, M. Global dynamics of a staged progression model for infectious diseases. Math. Biosci. Eng. 2006, 3, 513–525.
[CrossRef] [PubMed]
3. Yang, Y.; Xu, D.; Feng, Z. Analysis of a model with multiple infectious stages and arbitrarily distributed stage durations. Math.
Model. Nat. Phenom. 2008, 3, 180–193. [CrossRef]
4. Guo, H.; Li, M. Global dynamics of a staged progression model with amelioration for infectious diseases. J. Biol. Dyn. 2008,
2, 154–168. [CrossRef] [PubMed]
5. Yuan, Z.; Zou, X. Global threshold property in an epidemic model for disease with latency spreading in a heterogeneous host
population. Nonlinear Anal. Real World Appl. 2010, 11, 3479–3490. [CrossRef]
Mathematics 2022, 10, 2712 27 of 28

6. Guo, H.; Li, M.; Shuai, Z. Global dynamics of a general class of multistage models for infectious disseises. SIAM J. Appl. Math.
2012, 72, 261–270. [CrossRef]
7. Wang, J.L.; Liu, S.Q. The stability analysis of a general viral infection model with distributed delays and multi-staged infected
progression. Commun. Nonlinear Sci. Numer. Simul. 2015, 20, 263–272. [CrossRef]
8. Magal, P.; Clusky, C.C.M.; Webb, G. Lyapunov functional and global asymptotic stability for an infection age model. Appl. Anal.
2010, 89, 1109–1140. [CrossRef]
9. Cluskey, C.C.M. Global stability for an SEI epidemiological model with continuous age structure in the exposed and infectious
classes. Math. Biosci. Eng. 2012, 9, 819–841.
10. Huang, G.; Liu, X.; Takeuchi, Y. Lyapunov functions and global stability for age structured HIV infection model. SIAM J. Appl.
Math. 2012, 72, 25–38. [CrossRef]
11. Shen, M.; Xiao, Y.; Rong, L. Global stability of an infection-age structured HIV-1 model linking within-host and between-host
dynamics. Math. Biosci. 2015, 263, 37–50. [CrossRef]
12. Cluskey, C.C.M. Global stability for an SEI model of infectious disease with age structure and immigration of infected. Math.
Biosci. Eng. 2016, 13, 381–400.
13. Feng, Z.L.; Huang, W.Z.; Chavez, C.C. Global behavior of a multi-group SIS epidemic model with age structure. J. Differ. Equ.
2005, 218, 292–324. [CrossRef]
14. Kuniya, T. Global stability analysis with a discretization approach for an age-structured multi-group SIR epidemic model.
Nonlinear Anal. Real World Appl. 2011, 12, 2640–2655. [CrossRef]
15. Shen, M. Xiao, Y. Global stability of a multi-group SVEIR epidemiological model with the vaccination age and infection age. Acta
Appl. Math. 2016, 144, 137–157. [CrossRef]
16. Liu, Z.J.; Hu, J.; Wang, L.W. Modelling and analysis of global resurgence of mumps: A multi-group epidemic model with
asymptomatic infection, general vaccinated and exposed distributions. Nonlinear Anal. Real World Appl. 2017, 37, 137–161.
[CrossRef]
17. Cherif, A.; Dyson, J.; Maini, P.K.; Gupta, S. An age-structured multi-strain epidemic model for antigenically diverse infectious
diseases: A multi-locus framework. Nonlinear Anal. Real World Appl. 2017, 34, 275–315. [CrossRef]
18. Zhang, S.X.; Guo, H.B. Global analysis of age-structured multi-stage epidemic models for infectious diseases. Appl. Math. Comput.
2018, 337, 214–233. [CrossRef]
19. Zhang, S.X.; Guo, H.B.; Smith, R. Dynamical analysis for a hepatitis B transmission model with immigration and infection age.
Math. Biosci. Eng. 2018, 15, 1291–1313. [CrossRef]
20. Baiya, V.P.; Tripathi, J.P.; Kakkar, V.; Wang, J.S.; Sun, G.Q. Global dynamics of a multi-group SEIR epidemic model with infection
age. Chin. Ann. Math. Ser. B 2021, 42, 833–860. [CrossRef]
21. Luo, Y.T.; Tang, S.T.; Teng, Z.D.; Zhang, L. Global dynamics in a reaction diffusion multi-group SIR epidemic model with nonlinear
incidence. Nonlinear Anal. Real World Appl. 2019, 50, 365–385. [CrossRef]
22. Liu, R.; Wu, J.; Zhu, H. Media/psychological impact on multiple outbreaks of emerging infectious diseases. Comput. Math.
Methods Med. 2007, 8, 153–164. [CrossRef]
23. Cui, J.; Tao, X.; Zhu, H. An SIS infection model incorporating media coverage. Rocky Mt. J. Math. 2008, 38, 1323–1334. [CrossRef]
24. Funk, S.; Gilad, E.; Jansen, V.A.A. Endemic disease, awareness and local behavioural response. J. Theor. Biol. 2010, 264, 501–509.
[CrossRef]
25. Nyabadza, F.; Chiyaka, C.; Mukandavire, Z.; Musekwa, S.D.H. Analysis of an HIV/AIDS model with public health information
campaigns and individual withdrawal. J. Biol. Syst. 2010, 18, 357–375. [CrossRef]
26. Misra, A.K.; Sharma, A.; Shukla, J.B. Modeling and analysis of effects of awareness programs by media on the spread of infectious
diseases. Math. Comput. Model. 2011, 53, 1221–1228. [CrossRef]
27. Kim, M.; Yoo, B.K. Cost-effectiveness analysis of a television campaign to promote seasonal influenza vaccination among the
elderly. Value Health 2015, 18, 622–630. [CrossRef]
28. Agaba, G.O.; Kyrychko, Y.N.; Blyuss, K.B. Mathematical model for the impact of awareness on the dynamics of infectious disease.
Math. Biosci. 2017, 286, 22–30. [CrossRef]
29. Xiao, Y.; Tang, S.; Wu, J. Media impact switching surface during an infectious disease outbreak. Sci. Rep. 2015, 5, 7838. [CrossRef]
30. Pharaon, J.; Bauch, C.T. The influence of social behaviour on competition between virulent pathogen strains. J. Theor. Biol. 2018,
455, 47–53. [CrossRef]
31. Misra, A.K.; Rai, R.K.; Takeuchi, Y. Modelling the control of infectious diseases: Effects of TV and social media advertisements.
Math. Biosci. Eng. 2018, 15, 1315–1343. [CrossRef] [PubMed]
32. Yang, A.J.; Song, B.J.; Yuan, S.L. Noise-induced transmitions in a non-smooth SIS epidemic model with media alert. Math. Biosci.
Eng. 2021, 18, 745–763. [CrossRef] [PubMed]
33. Shim, E. Optimal strategies of social distancing and vaccination against seasonal influenza. Math. Biosci. Eng. 2013, 10, 1615–1634.
[CrossRef] [PubMed]
34. Kumar, A.; Srivastava, P.K.; Takeuchi, Y. Modeling the role of information and limited optimal treatment on disease prevalence. J.
Theor. Biol. 2017, 414, 103–119. [CrossRef]
35. Yu, T.H.; Cao, D.Q.; Liu, S.Q. Epidemic model with group mixing: Stability and optimal control based on limited vaccination
resources. Commun. Nonlinear Sci. Numer. Simul. 2018, 61, 54–70. [CrossRef]
Mathematics 2022, 10, 2712 28 of 28

36. Zhang, H.T.; Yang, Z.W.; Pawelek, K.A.; Liu, S.Q. Optimal control strategies for a two group epidemic model with vaccination
resource constraints. Appl. Math. Comput. 2020, 371, 124956. [CrossRef]
37. Yang, J.Y.; Modnak, C.; Wang, J. Dynamical analysis and optimal control simulation for an age-structured cholera transmission
model. J. Frankl. Inst. 2019, 356, 8438–8467. [CrossRef]
38. Kumar, A.; Srivastava, P.K.; Dong, Y.P.; Takeuchi, Y. Optimal control of infectious disease: Information-induced vaccination and
limited treatment. Phys. A Stat. Mech. Appl. 2020, 542, 123196. [CrossRef]
39. Gashirai, T.B.; Musekwa-Hove, S.D.; Lolika, P.O.; Mushayabasa, S. Global stability and optimal control analysis of a foot and
mouth disease model with vaccine failure and environmental transmission. Chaos Solitons Fractals 2020, 132, 109568. [CrossRef]
40. Khatun, M.S.; Biswas, M.H.A. Optimal control strateties for preventing hepatitis B infection and reducing chronic liver cirrhosis
incidence. Infect. Dis. Model. 2020, 5, 91–110.
41. Wang, L.W.; Liu, Z.J.; Zhang, X.A. Global dynamics for an age-structured epidemic model with media impact and incomplete
vaccination. Nonlinear Anal. Real World Appl. 2016, 32, 136–158. [CrossRef]
42. Wang, L.W.; Liu, Z.J.; Xu, D.S.; Zhang, X.A. Global dynamics and optimal control of an influenza model with vaccination, media
coverage and treatment. Int. J. Biomath. 2017, 10, 1750068. [CrossRef]
43. Webb, G.F. Theory of Nonlinear Age-Dependent Population Dynamics; Marcel Dekker: New York, NY, USA, 1985; pp. 101–134.
44. Thieme, H.R. Semiflows generated by Lipschitz perturbation of non-densely defined operators. Differ. Integral Equ. 1990,
3, 1035–1066.
45. Hale, J. Theory of Functional Differential Equations; Springer: New York, NY, USA, 1977; pp. 25–76.
46. Zhao, X.Q. Dynamical Systems in Population Biology; Springer: New York, NY, USA, 2003; pp. 112–149.
47. Zhang, S.; Xu, X. A mathematical model for hepatitis B with infection-age structure. Discret. Contin. Dyn. Syst. Ser. B 2016,
21, 1329–1346. [CrossRef]
48. Brezis, H. Functional Analysis, Sobolev Spaces and Partial Differential Equations; Springer: London, UK, 2011; pp. 36–89.
49. Inaba, H. A semigroup approach to the strong ergodic theorem of the multistate stable population process. Popul. Stud. 1988,
1, 49–77. [CrossRef] [PubMed]
50. Hale, J.; Waltman, P. Persistence in infinite dimensional systems. SIAM J. Appl. Math. 1989, 20, 388–395. [CrossRef]
51. DelValle, S.Y.; Hyman, J.M.; Chitnis, N. Mathematical models of contact patterns between age groups for predicting the spread of
infectious diseases. Math. Biosci. Eng. 2013, 10, 1475–1497.
52. Fister, K.R.; Gaff, H.; Lenhart, S.; Numfor, E.; Schaefer, E.; Wang, J. Optimal Control of Vaccination in An Age-Structured Cholera
Model, Mathematical and Statistical Modeling for Emerging and Re-Emerging Infectious Diseases; Springer: New York, NY, USA, 2016;
pp. 221–248.