COMMUNICATIONS ON doi:10.3934/cpaa.2016.15.

871
PURE AND APPLIED ANALYSIS
Volume 15, Number 3, May 2016 pp. 871–892

TRAVELING WAVES FOR A DIFFUSIVE SEIR EPIDEMIC

MODEL

Zhiting Xu
School of Mathematical Sciences, South China Normal University
Guangzhou, 510631, P. R. China

(Communicated by Shoji Yotsutani)

In Memory of My Parents

Abstract. In this paper, we propose a diffusive SEIR epidemic model with

saturating incidence rate. We first study the well posedness of the model, and
give the explicit formula of the basic reproduction number R0 . And hence,
we show that if R0 > 1, then there exists a positive constant c∗ > 0 such
that for each c > c∗ , the model admits a nontrivial traveling wave solution,
and if R0 ≤ 1 and c ≥ 0 (or, R0 > 1 and c ∈ [0, c∗ )), then the model has no
nontrivial traveling wave solutions. Consequently, we confirm that the constant
c∗ is indeed the minimal wave speed. The proof of the main results is mainly
based on Schauder fixed theorem and Laplace transform.

1. Introduction. In order to describe the transmission of communicable diseases,

Kermack and McKendrick [10] proposed the classic deterministic Susceptible-Infected
-Removed (SIR) model
d
S(t) = −βS(t)I(t),
dt
d
I(t) = βS(t)I(t) − γI(t), (1.1)
dt
d
R(t) = γI(t),
dt
where S(t), I(t) and R(t) denote the number of the susceptible, infected and re-
moved individuals, respectively. The constant β is the transmission coefficient, and
γ is the recovery rate. Given S(0) = S 0 > 0, I(0) > 0 and R(0) = 0. It is
0
well known that, by the so-called basic reproduction number R0 := βS γ , Kermack
and McKendrick[10] obtained full information about the transmission dynamics and
epidemic potential: if R0 > 1, then I(t) first increases to its maximum and then
decreases to zero and hence an epidemic occurs; if R0 < 1, then I(t) decreases to
zero and epidemic does not happen.

2000 Mathematics Subject Classification. Primary: 35C07, 35K57; Secondary: 92D30.

Key words and phrases. Traveling wave solutions, minimal wave speed, SEIR epidemic model,
saturating incidence rate.
Research partially supported by the NSF of China (11171120).

871
872 ZHITING XU

When individuals move randomly, a modified version of model (1.1) will be stud-
ied, which takes into the following diffusive SIR model
∂
S(t, x) = d1 ∆S(t, x) − βS(t, x)I(x, t),
∂t (1.2)
∂
I(t, x) = d2 ∆I(t, x) + βS(t, x)I(x, t) − γI(t, x).
∂t
This model was first considered by Hosono and Ilyas [8], where p it is showed that
if S(0, x) = S 0 > 0 and βS 0 > γ, then for each c ≥ c∗ := 2 d2 (βS 0 − γ) there
exists a positive constant ε < S0 such that model (1.2) has a traveling wave solution
(S(x + ct), I(x + ct)) satisfying S(−∞) = S 0 , S(+∞) = ε, I(±∞) = 0. On the
other hand, there is no traveling wave solution for (1.2) when βS 0 ≤ γ.
In recent years, many researchers have attention to study the existence and non-
existence of traveling wave solutions for some diffusive epidemic models (see,e.g.
[1, 12, 20, 22, 23, 24, 25, 30, 33]). In particular, Wang and Wu [23] investigated
the existence and non-existence of traveling wave solutions for a diffusive Kermack-
McKendrick epidemic model with non-local delayed transmission. In [22], Wang,
Wang and Wu studied traveling waves of reaction-diffusion equations for a diffusive
SIR model. In a very recent paper, for a three-dimensional reaction-diffusion sys-
tems, Zhang and Wang [33] established the existence of traveling wave solutions for
influenza model with treatment.
In present paper, we will formulate a simple diffusive epidemic model incorpo-
rating exposed individuals into model (1.1). In the model considered, the total
individuals at time t ≥ 0 and position x ∈ R, denoted by a constant N , is sub-
divided into four mutually exclusive compartments of susceptible S(x, t), exposed
E(x, t), infectious I(x, t) and recovered R(x, t) individuals, so that S(x, t)+E(x, t)+
I(x, t) + R(x, t) = N . Here the model is called an SEIR model [2, 17] where in-
dividuals are susceptible, then exposed (i.e., in the latent period), then infectious,
then recovered from infectious individuals. In addition, we also assume that both
the exposed and infected individuals are infected, they are distinguished by the
absence/presence of their ability to infect other population [6, 7]. On the other
hand, it is believed that, in the modeling of infectious diseases, the incidence func-
tion plays a very important role, it can determine the rise and fall of epidemics
βSI
[3, 6, 7]. Here we will use the saturating incidence rate denoted as 1+aI (a > 0)
[3] other than the bilinear incidence rate βSI. As reported in the literature, the
saturating incidence rate is more realistic than the bilinear rate and can cause some
interesting dynamic behaviors of infectious diseases, such as limit cycle, heteroclinic
orbit, saddle-node bifurcation, transcritical bifurcation and Hopf bifurcation, see,
[7, 14, 18, 28, 29, 31], for example.
In view of the above discussion, the proposed model is given by the following
diffusive Susceptible-Exposed-Infected-Removed (SEIR) model with a saturating
incidence rate
∂
S(t, x) = d1 ∆S(t, x) − βS(t, x)g(I(t, x)),
∂t
∂
E(t, x) = d2 ∆E(t, x) + βS(t, x)g(I(t, x)) − αE(t, x),
∂t (1.3)
∂
I(t, x) = d3 ∆I(t, x) + αE(x, t) − γI(t, x),
∂t
∂
R(t, x) = d4 ∆R(t, x) + γI(t, x),
∂t
 A DIFFUSIVE SEIR EPIDEMIC MODEL 873

2
I ∂
in which g(I) = 1+aI (a > 0), ∆ = ∂x 2 , di > 0 (i = 1, 2, 3, 4) are the diffusion rates.

The constant α > 0 is the rate constant for exposed individuals becoming infectious,
β > 0 is the transmission coefficient, γ > 0 is the rate constant that the infective
individuals become recovered. In the sequel, we always assume that the initial free
equilibrium is (S 0 , 0, 0, 0) with S 0 > 0. In contrast, for the detailed epidemiological
description of the corresponding ODE model, we refer to [2, 6, 7, 17].
The purpose of the current paper is to study the existence and non-existence of
traveling wave solutions of (2.1). We will show that there exists a positive constant
c∗ > 0 such that (2.1) has a traveling wave solution if c > c∗ and βS 0 > γ. To
prove the existence theorem (see Theorem 3.1), we will employ Schauder fixed point
theorem [9, 11, 15] to the non-monotone operator used in a suitable invariant convex
set. In order to construct the appropriate invariant convex set, we also use the idea
of the iteration process [20, 22, 23, 24, 25, 30] to construct the upper-lower solutions.
One important feature of our method, which is different from the ones, is that we
need to construct the vector-value upper-lower solutions for (2.3) (see Section 2.3)
since system (2.3) consists of three equations. The ideas of such a construction is
motivated by Weng and Zhao [26], see also [5, 19] for the related works. Further, we
establish that (2.1) has no traveling wave for any c ≥ 0 and βS0 < γ (see, Theorem
3.2). For c ∈ (0, c∗ ) and βS 0 > γ, we conclude the non-existence of traveling wave
solutions for (2.1) (see, Theorem 3.3). Here the critical method of the proof of the
non-existence is based on the two-side Laplace transform. As we know that the
application of the Laplace transform requires the prior estimate of the exponential
decay of the traveling wave solutions [4, 20, 22, 23, 30]. However, it seems that the
analytical method in [4, 20, 22, 23, 30] cannot give the prior estimate due to the
three dimensional system (2.3) again. Instead, we approve the approach recently
introduced by [33] to get the prior estimate.
This paper is organized as follows. In Section 2, we first consider the well posed-
ness of system (2.1) in [0, +∞) and then derive basic properties of the waves. Sec-
tions 3 is devoted to the study of the existence and non-existence of traveling waves
for system (2.1). Finally, a brief discussion section completes the paper.

2. Preliminaries. In this section. we should give some preliminary results such as

the basic reproduction ratio and the well posedness of system (2.1), the eigenvalue
problems for the wave profile (2.3), constructing a pair of upper and lower solutions
for system (2.3) and verifying the conditions of the Schauder fixed point theorem.
2.1. The basic reproduction ratio and the well-posedness. Because R(t, x)
does not appear in the first three equations of (1.3), and for the simplicity of nota-
tion, let (S, E, I) = (u1 , u2 , u3 ), we consider the following new system
∂
u1 (t, x) = d1 ∆u1 (t, x) − βu1 (t, x)g(u3 (t, x)),
∂t
∂
u2 (t, x) = d2 ∆u2 (t, x) + βu1 (t, x)g(u3 (t, x)) − αu2 (t, x), (2.1)
∂t
∂
u3 (t, x) = d3 ∆u3 (t, x) + αu2 (x, t) − γu3 (t, x),
∂t
u
where g(u) = 1+au . Accompanied with (2.1), we consider the initial value conditions
ui (0, x) = ϕi (x) ≥ 0, x ∈ R, i = 1, 2, 3, but not identically zero. (2.2)
In this subsection, we focus on the existence, uniqueness, invariance of solutions
for the Cauchy problem to system (2.1) in R (see, Theorem 2.1). Firstly, we define
874 ZHITING XU

the basic reproduction ratio R0 for system (2.1). By similar arguments to those
in [21, Theorem 2.3], we can show that the basic reproduction ratio R0 equals the
spectral radius of the following 2 × 2 matrix
βS 0
 
0
B :=  γ .
1 0
q
0
Hence, R0 = βS γ . For the definition of the basic reproduction ratio R0 for the
reaction-diffusion models and its biological interpretation, we refer the readers to
[21] for details. In the following, we always assume that R0 > 1.
Throughout this paper, the cone R3+ denotes the subset of R3 with vectors x ≥
0. Let X = C(R, R3 ) be Banach space with the supremum norm k · kX . Define
X+ = C(R, R3+ ), then (X, X+ ) is a strongly ordered space. Moreover, we shall use
the standard partial order in R3 or X. Let
XM = {ϕ = (ϕ1 , ϕ2 , ϕ3 ) ∈ X+ : 0 ≤ ϕ(x) ≤ M, ∀x ∈ R},
in which 0 := (0, 0, 0) and
 γ βS 0
1 βS 0

M := S 0 , −1 , −1 .
aα γ a γ
For any ϕ = (ϕ1 , ϕ2 , ϕ3 ) ∈ XM and ∀x ∈ R, we define F = (F1 , F2 , F3 ) : XM → X
by
F1 (ϕ)(x) = −βϕ1 (x)g(ϕ3 (x)),
F2 (ϕ)(x) = βϕ1 (x)g(ϕ3 (x)) − αϕ2 (x),
F3 (ϕ)(x) = αϕ2 (x) − γϕ3 (x).
Then F is locally Lipschitz continuous in any bounded subset of XM . We now
reformulate (2.1)-(2.2) as the following abstract differential equation
du
= Au + F (u), t > 0,
dt
u0 = ϕ ∈ XM ,
where u = (u1 , u2 , u3 ), Au := (d1 ∆u1 , d2 ∆u2 , d3 ∆u3 )T and ϕ = (ϕ1 , ϕ2 , ϕ3 ).
Theorem 2.1. For any given initial data ϕ = (ϕ1 , ϕ2 , ϕ3 ) ∈ XM , system (2.1)-
(2.2) admits a unique nonnegative solution u(t, x, ϕ) which exists globally on [0, +∞)
such that u(0, ·, ϕ) = ϕ and u(t, ·, ϕ) ∈ XM for all t ≥ 0.
Proof. For any ϕ = (ϕ1 , ϕ2 , ϕ3 ) ∈ XM and any sufficient small k ≥ 0, we get
 
ϕ1 (x) 1 − kβg(ϕ3 (x)))
ϕ(0, x) + kF (ϕ)(x) =  (1 − kα)ϕ2 (x) + kβϕ1 (x)g(ϕ3 (x)) 
(1 − kγ)ϕ3 (x) + kαϕ2 (x)
βk
 
1− )ϕ1 (x)
≥  (1 − kα)ϕ a
 ≥ 0.
 
2 (x)
(1 − kγ)ϕ3 (x)
Note that g(x) is increasing for x ≥ 0. Then
γ βS 0

0 ≤ g(ϕ3 (x)) ≤ 0
−1 .
aβS γ
 A DIFFUSIVE SEIR EPIDEMIC MODEL 875

Consequently,
 
S0
 γ βS 0

 −1 
ϕ(0, x) + kF (ϕ)(x) ≤ 
 aα γ  = M.

 1 βS 0

−1
a γ
Thus, it follows ϕ(x) + kF (ϕ)(x) ∈ XM . This in turn implies that
1
lim dist(ϕ(x) + kF (ϕ)(x), XM ) = 0, ∀ ϕ ∈ XM .
k→0+ k
It then follows from [16, Corollary 4] (taking the delay as zero) that (2.1)-(2.2) admit
a unique nonnegative mild solution u(t, x, ϕ) on [0, ∞) × R with u(0, ·, ϕ) = ϕ, and
u(t, ·, ϕ) ∈ XM for t ≥ 0. Moreover, by [27, Corollary 2.2.5], the mild solution is
classic on [0, ∞).

In view of Theorem 2.1, the dynamics of system (2.1) can be analyzed in the
bounded feasible region XM . Furthermore, the region XM is positively invariant
with respect to model (2.1) and the model is well posedness.

2.2. Eigenvalue problem. This subsection deals with the eigenvalue problems for
the wave profile, which is obtained by substituting ui (t, x) = Ui (x + ct) (i = 1, 2, 3)
in (2.1). Here (U1 , U2 , U3 ) is called the wave profile, ξ := x + ct the wave coordinate
and c the speed. For the sake of convenience, we still use ui , t instead of Ui , ξ. And
for technical reasons, we let d2 = d3 =: d, and then consider the following wave
profile equation
d1 u001 (t) − cu01 t) − βu1 (t)g(u3 (t)) = 0,
du002 (t) − cu02 (t) − αu2 (t) + βu1 (t)g(u3 (t)) = 0, (2.3)
du003 (t) − cu03 (t) − γu3 (t) + αu2 (t) = 0.
We now consider the eigenvalue problem at (S 0 , 0, 0). Linearizing of the last two
equations of (2.3) at (S 0 , 0, 0) implies that
du002 (t) − cu02 (t) − αu2 (t) + βS 0 u3 (t) = 0,
du003 (t) − cu03 (t) − γu3 (t) + αu2 (t) = 0.
Plugging ui (t) = vi eλt , i = 2, 3, into the above equations, we get the following
eigenvalue problem
 v 
2
detA(λ) = 0, A(λ) = 0,
v3
where
 h (λ) βS 0 
1
A(λ) =
α h2 (λ)
with h1 (λ) = dλ2 − cλ − α and h2 (λ) = dλ2 − cλ − γ.
Clearly, the fact R0 > 1 implies detA(0) = αγ(1 − R20 ) < 0. Then the equation
detA(λ) = 0 has at least one positive root.
When R0 > 1, we set
r q
∗


c := 2d (α + γ)2 + 4αγ(R20 − 1) − (α + γ) .
876 ZHITING XU

About the distribution of eigenvalue and their eigenvectors, we have the following
results.
Lemma 2.1. Assume that R0 > 1. Then we have the following statements.
(1) If c > c∗ , then the equation detA(λ) = 0 has three positive roots 0 < λ1 <
c
λ2 < λ3 , and a negative root λ4 . Moreover, for any  ∈ (0, 2d − λ1 ),
h1 (λ1 + ) < h1 (λ1 ) < 0, h2 (λ1 + ) < h2 (λ1 ) < 0.
(1) (1)
And, the eigenvector (v2 , v3 ) corresponding to the smallest eigenvalue λ1 is
(1) (1)
strongly positive, i.e., (v2 , v3 )  0.
(2) If c = c∗ , then the equation detA(λ) = 0 has two different positive roots, and
a negative root.
(3) If 0 < c < c∗ , then the equation detA(λ) = 0 has a positive root, a negative
root, and a pair of complex roots with positive real parts.
Proof. Let ρ = dλ2 − cλ. Then
detA(λ) = ρ2 − (α + γ)ρ − αγ(R20 − 1) =: H(ρ).
Consequently, it follows from the fact R0 > 1 that the equation H(ρ) = 0 has a
positive root ρ+ and a negative root ρ− , where
(c∗ )2 (c∗ )2
ρ+ = α + γ + , ρ− = − .
4d 4d
Clearly, the equation dλ2 − cλ − ρ+ = 0 has a positive root λ3 and a negative root
λ4 , here
p p
c + c2 + (c∗ )2 + 4d(α + γ) c − c2 + (c∗ )2 + 4d(α + γ)
λ3 = > 0, λ4 = < 0.
2d 2d
If c > c∗ , then the roots of the equation dλ2 − cλ − ρ− = 0 are real and positive,
i.e., p p
c − c2 − (c∗ )2 c + c2 − (c∗ )2
λ1 = > 0, λ2 = > 0.
2d 2d
c
Obviously, 0 < λ1 < λ2 < λ3 and 0 < λ1 < 2d . Note that h1 (λ) is decreasing in
c c c2
(0, 2d ), and h1 ( 2d ) = − 4d − α < 0. Therefore, h1 (λ1 + ε) < h1 (λ1 ) < 0 for any
c
ε ∈ (0, 2d − λ1 ). By the similar discussions, we also get h2 (λ1 + ε) < h2 (λ1 ) < 0.
On the other hand, in view of the fact that h1 (λ1 ) < 0 and h2 (λ1 ) < 0, then
(1) (1) (1) (1)
there exists a pair of (v2 , v3 )  0 such that A(λ1 )(v2 , v3 )T = 0. This shows
that the conclusion (1) holds.
The proofs of the conclusions (2) and (3) are easy and omitted the details.
2.3. Construction of the upper and lower solutions. To establish existence
of traveling wave solutions of (2.1), we will construct a convex invariant set. For
this, we will use the iteration process to construct a pair of vector-value upper-lower
solutions for (2.3). Namely, we first choose the upper solution u1 (t) = S 0 for the u1 -
component (Lemma 2.2), which is then used to build the upper solutions u2 (t) and
u3 (t) for the u2 and u3 -components (Lemma 2.3). The upper solutions u3 (t) is in
turn employed to produce the lower solutions u1 (t) and u3 (t) for the u1 -component
(Lemma 2.4). Finally, the lower solutions u1 (t) and u3 (t) are further builded to
construct the lower solution u2 (t) for the u3 -components (Lemma 2.5). Here we
also would like to stress that, different from that in [23], the pair of upper-lower
solutions constructed in this paper are vector-value type and bounded on R.
 A DIFFUSIVE SEIR EPIDEMIC MODEL 877

In the following, we always assume that R0 > 1, i.e., βS 0 > γ, and c > c∗ . Let
(1) (1)
λ1 be the smallest eigenvalue defined as in Lemma 2.1(1) and (v2 , v3 )  0 its
associating eigenvector. For practical reasons, it will be convenient to choose the
(1) (1) (1) (1)
vector (v2 , v3 ) = (v2 , 1)  0, where v2 satisfies
(1) (1)
h1 (λ1 )v2 + βS 0 = 0 or αv2 + h2 (λ1 ) = 0,
for the calculations in the proofs of Lemmas 2.3 and 2.5.
For t ∈ R, we define six continuous functions as follows
u1 (t) =S 0 ,
n
(1) γ βS 0
o
u2 (t) = min v2 eλ1 t , −1 ,
aα γ
n 1 βS 0
o
u3 (t) = min eλ1 t , −1 ,
a γ
and
n 1 o
u1 (t) = max S 0 − eσt , 0 ,
n σ o
(1) λ1 t
u2 (t) = max v2 e 1 − M et ), 0 ,
n o
u3 (t) = max eλ1 t 1 − M et ), 0 ,
in which σ, M,  are positive constants determined in the following lemmas.
It is easy to see the following Lemma 2.2 holds.
Lemma 2.2. The function u1 (t) satisfies the inequality
d1 u001 (t) − cu01 (t) − βu1 (t)g(u3 (t)) ≤ 0,
for all t ∈ R.
Lemma 2.3. The functions u2 (t) and u3 (t) satisfy the inequalities
1 γ βS 0
du002 (t) − cu02 (t) − αu2 (t) + βS 0 g(u3 (t)) ≤ 0, t 6= t1 :=


ln (1)
−1 ,
λ1 aαv γ
2
(2.4)
0
1 1 βS
du003 (t) − cu03 (t) − γu3 (t) + αu2 (t) ≤ 0, t 6= t2 :=


ln −1 . (2.5)
λ1 a γ
Proof. We only show the inequality (2.4) holds, since the proof of (2.5) is similar
(1)
to that of (2.4). Indeed, when t < t1 , u2 (t) = v2 eλ1 t . Note that u3 (t) ≤ eλ1 t for
all t ∈ R and g(x) ≤ x for all x ≥ 0. Then
du002 (t) − cu02 (t) − αu2 (t) + βS 0 g(u3 (t)) ≤ du002 (t) − cu02 (t) − αu2 (t) + βS 0 u3 (t)
(1)
≤ eλ1 t (h1 (λ1 )v2 + βS 0 ) = 0.
γ βS 0 1 βS
0



When t > t1 , u2 (t) = aα γ − 1 . It follows from the fact that u3 (t) ≤ a γ −1
for all t ∈ R, and the function g(x) is increasing in x > 0. Then
γ βS 0
du002 (t) − cu02 (t) − αu2 (t) + βS 0 g(u3 (t)) ≤ du002 (t) − cu02 (t) − αu2 (t) +


( −1
a γ
γ βS 0
γ βS 0

=− −1 + ( − 1 = 0.
a γ a γ
878 ZHITING XU

Thus, the inequality (2.4) holds. This completes the proof.

Lemma 2.4. Let

 λ1 1 c
σ < min , 0, . (2.6)
2 S d1 + β(S 0 )2
Then the functions u1 (t) and u3 (t) satisfy the inequality
1
d1 u001 (t) − cu01 (t) − βu1 (t)g(u3 (t)) ≥ 0, t 6= t3 := ln(σS0 ). (2.7)
σ
Proof. If t > t3 , then the inequality (2.7) holds immediately since u1 (t) = 0 on
[t3 , ∞). If t < t3 , then u1 (t) = S0 − σ1 eσt . Note that u3 (t) ≤ eλ1 t for all t ∈ R. One
gets
d1 u001 (t) − cu01 (t) − βu1 (t)g(u3 (t)) ≥ d1 u001 (t) − cu01 (t) − βu1 (t)u3 (t)
1

= −d1 σeσt + ceσt − β S 0 − eσt eλ1 t
σ
≥ eσt − d1 σ + c − βS0 e(λ1 −σ)t ,

which follows from the fact e(λ1 −σ)t < (σS 0 )(λ1 −σ)/σ for t < t3 that
λ1 −σ
d1 u001 (t) − cu01 (t) − βu1 (t)g(u3 (t)) ≥ eσt − d1 σ + c − βS 0 (σS 0 )


σ .
λ1 −σ
Noting that (2.6), we see (σS 0 ) σ ≤ σS 0 . Then, by (2.6) again,
d1 u001 (t) − cu01 (t) − βu1 (t)g(u3 (t)) ≥ eσt − d1 σ + c − βσ(S 0 )2 > 0,


t < t3 ,
which completes the proof.

Lemma 2.5. Let

 c
0 <  < min σ, λ1 , − λ1 .
2d
Then, for
n β(1 + aσS 0 ) o
M > max 1, − (1)
,
σ(h1 (λ1 + )v2 + βS 0 )
the functions u1 (t), u2 (t) and u3 (t) satisfy the inequalities
1 1
du002 (t) − cu02 (t) − αu2 (t) + βu1 (t)g(u3 (t)) ≥ 0, t 6= t4 := ln , (2.8)
 M
du003 (t) − cu03 (t) − γu3 (t) + αu2 (t) ≥ 0, t 6= t4 . (2.9)

Proof. By Lemma 2.1(1), h1 (λ1 + ) < h1 (λ1 ) < 0 and h2 (λ1 + ) < h2 (λ1 ) < 0.
Then
(1) (1)
h1 (λ1 + )v2 + βS 0 < h1 (λ1 )v2 + βS 0 = 0, (2.10)
(1) (1)
αv2 + h2 (λ1 + ) < αv2 + h2 (λ1 ) = 0. (2.11)
Now we first show (2.8) holds. In fact, the inequality (2.8) holds immediately
(1)

since u2 (t) = 0 on [t4 , ∞). For t < t4 < 0, u2 (t) = v2 eλ1 t 1 − M et . In view of
the fact g(x) ≥ x(1 − ax) for all x ≥ 0, and
1 σt
e ≤ u1 (t) ≤ S 0 , eλ1 t 1 − M et ≤ u3 (t) ≤ eλ1 t , ∀ t ∈ R,


S0 −
σ
 A DIFFUSIVE SEIR EPIDEMIC MODEL 879

we get

du002 (t) − cu02 (t) − αu2 (t) + βu1 (t)g(u3 (t))

≥ du002 (t) − cu02 (t) − αu2 (t) + βu1 (t)u3 (t)(1 − au3 (t))
≥ du002 (t) − cu02 (t) − αu2 (t)
1

+ β S 0 − eσt eλ1 t − M e(λ1 +)t − aβS 0 e2λ1 t


σ

(1) β 
= −e(λ1 +)t M (h1 (λ1 + )v2 + βS 0 ) + e(σ−)t + aβS 0 e(λ1 −)t .
σ
Note that  < min{σ, λ1 }. Then e(σ−)t < 1 and e(λ1 −)t < 1 for t < t4 < 0.
Therefore, by (2.10), for t < t4 < 0,

du002 (t) − cu02 (t) − αu2 (t) + βu1 (t)g(u3 (t))


(1) β 
≥ −e(λ1 +)t M (h1 (λ1 + )v2 + βS 0 ) + + aβS 0 > 0.
σ
Consequently, (2.8) holds.
Next, we show (2.9) holds. Clearly, (2.9) holds since u3 (t) = 0 for t > t4 . For
(1)
t < t4 , u3 (t) = eλ1 t (1 − M et ). Note that u2 (t) ≥ v2 eλ1 t (1 − M et ) for all t ∈ R.
Hence, by (2.11),
(1)
du003 (t) − cu03 (t) − γu3 (t) + αu2 (t) ≥ −M e(λ1 +)t αv2 + h2 (λ1 + ) > 0.

which follows (2.9) holds for t < t4 < 0. Hence, we complete the proof.

2.4. The verification of Schauder fixed point theorem. In this subsection,

we will use the upper and lower solutions (u1 (t), u2 (t), u3 (t)) and (u1 (t), u2 (t), u3 (t))
to verify that the conditions of Schauder fixed point theorem hold. Here, we use the
usual Banach space B = C(R, R3 ) of bounded continuous functions endowed with
the maximum norm, see, [32],

k(u1 , u2 , u3 )k = sup(|u1 (t)| + (u2 (t)| + |u3 (t)|).

t∈R

∗
For c > c , let

Γ = (u1 , u2 , u3 ) ∈ B : ui (t) ≤ ui (t) ≤ ui (t), i = 1, 2, 3 .

Define the operator H = (H1 , H2 , H3 ) : Γ → B by

H1 (u1 , u2 , u3 )(t) : = α1 u1 (t) − βu1 (t)g(u3 (t)),

H2 (u1 , u2 , u3 )(t) : = (α2 − α)u2 (t) + βu1 (t)g(u3 (t)),
H3 (u1 , u2 , u3 )(t) : = αu2 (t) + (α3 − γ)u3 (t),
β
where each αi is a large positive number such that α1 > a , α2 > α and α3 > γ.
Then system (2.3) can be rewritten as

d1 u001 (t) − cu01 (t) − α1 u1 (t) + H1 (u1 , u2 , u3 )(t) = 0,

du002 (t) − cu02 (t) − α2 u2 (t) + H2 (u1 , u2 , u3 )(t) = 0, (2.12)
du003 (t) − cu03 (t) − α3 u3 (t) + H3 (u1 , u2 , u3 )(t) = 0.
880 ZHITING XU

Let
√ √
c−c2 + 4d1 α1 c + c2 + 4d1 α1
λ11 = , λ12 = ,
2d 2d1
√ 1 √
c − c2 + 4dαi c + c2 + 4dαi
λi1 = , λi2 = , i = 2, 3.
2d 2d
Define the operator F : Γ → B by

F (u)(t) = (F1 (u), F2 (u), F3 (u))(t),

where u = (u1 , u2 , u3 ), and

1  t λ11 (t−s)
Z Z +∞ 
F1 (u)(t) = e + eλ12 (t−s) H1 (u)(s)ds,
ρ1 −∞ t
1  t λ21 (t−s)
Z Z +∞ 
F2 (u)(t) = e + eλ22 (t−s) H2 (u)(s)ds,
ρ2 −∞ t
Z t Z +∞
1  
F3 (u)(t) = eλ31 (t−s) + eλ32 (t−s) H3 (u)(s)ds,
ρ3 −∞ t

in which
p p
ρ1 = d1 (λ12 − λ11 ) = c2 + 2d1 α1 , ρi = d(λi2 − λi1 ) = c2 + 2dαi , i = 2, 3.

It is easily verified that the operator F is well defined for u = (u1 , u2 , u3 ) ∈ Γ, and
satisfies

d1 F100 (u)(t) − cF10 (u)(t) − α1 F1 (u)(t) + H1 (u)(t) = 0,

dF200 (u)(t) − cF20 (u)(t) − α2 F2 (u)(t) + H2 (u)(t) = 0,
dF300 (u)(t) − cF30 (u)(t) − α3 F3 (u)(t) + H3 (u)(t) = 0.

Thus the fixed point of F is the solution of (2.12), which is a traveling wave solution
of (2.1). Hence the existence of solution of (2.12) is reduced to verify that the
operator F satisfies the conditions of Schauder fixed point theorem. Here we divide
the proof into the following two lemmas.

Lemma 2.6. The operator F maps Γ into Γ.

Proof. Given u = (u1 , u2 , u3 ) ∈ Γ. It is obvious that we only need to show that

ui (t) ≤ Fi (u)(t) ≤ ui (t), ∀ t ∈ R, i = 1, 2, 3.

Based on the choice of the constants α1 , α2 and α3 , it suffices to prove that, for
any t ∈ R,

u1 (t) ≤ F1 (u1 , u2 , u3 )(t) ≤ F1 (u1 , u2 , u3 )(t) ≤ F1 (u1 , u2 , u3 )(t) ≤ u1 (t), (2.13)

u2 (t) ≤ F2 (u1 , u2 , u3 )(t) ≤ F1 (u1 , u2 , u3 )(t) ≤ F1 (u1 , u2 , u3 )(t) ≤ u2 (t), (2.14)
u3 (t) ≤ F3 (u1 , u2 , u3 )(t) ≤ F3 (u1 , u2 , u3 )(t) ≤ F3 (u1 , u2 , u3 )(t) ≤ u3 (t). (2.15)
 A DIFFUSIVE SEIR EPIDEMIC MODEL 881

Firstly, we prove (2.13) holds. In fact, for t > t3 , then, by (2.7),

F1 (u1 , u2 , u3 )(t)
1  t λ11 (t−s)
Z Z +∞
eλ12 (t−s) H1 (u1 , u2 , u3 )(s)(s)ds


= e +
ρ1 −∞ t
1  t λ11 (t−s)
Z Z +∞
eλ12 (t−s) (α1 u1 (s) + cu01 (s) − d1 u001 (s))ds


≥ e +
ρ1 −∞ t
d1 λ11 (t−t3 ) 0
= u1 (t) + e (u1 (t3 + 0) − u01 (t3 − 0))
ρ1
≥ u1 (t).
Similarly, for t ≤ t3 , we also get
F1 (u1 , u2 , u3 )(t) ≥ u1 (t).
Note that H1 (u1 , u2 , u3 )(t) ≤ α1 u1 (t) for t ∈ R. Then
1  t λ11 (t−s)
Z Z +∞ 
F1 (u1 , u2 , u3 )(t) = e + eλ12 (t−s) H1 (u1 , u2 , u3 )(s)ds
ρ1 −∞ t
Z t Z +∞
1  
≤ eλ11 (t−s) + eλ12 (t−s) (α1 u1 (s))ds
ρ1 −∞ t
α1 S 0  t λ11 (t−s)
Z Z +∞ 
≤ e ds + eλ12 (t−s) ds
ρ1 −∞ t
α1 S 0  1 1 
= − = S0.
ρ1 λ12 λ11
Therefore, (2.13) holds.
The proofs of (2.14) and (2.15) are similar to that of (2.13) and are omitted.
Consequently, we complete the proof.
Lemma 2.7. The operator F = (F1 , F2 , F3 ) : Γ → Γ is continuous and compact
with respect to the norm k · k.
Proof. We first show that F = (F1 , F2 , F3 ) : Γ → Γ is continuous with respect to
the norm k · k. Indeed, for any u = (u1 , u2 , u3 ), ue = (e
u1 , u e3 ) ∈ Γ, it is easy to
e2 , u
see that there exists L > 0 such that
3
LX
|Hi (u)(t) − Hi (e
u)(t)| ≤ |uj (t) − u
ej (t)| ≤ Lku − u
ek, ∀ t ∈ R, i = 1, 2, 3.
3 j=1

Therefore,
|F1 (u)(t) − F1 (e
u)(t)|
Z t Z +∞
1  
≤ eλ11 (t−s) + eλ12 (t−s) |H1 (u)(s) − H1 (e
u)(s)|ds
ρ1 −∞ t
L  t λ11 (t−s)
Z Z +∞ 
≤ e + eλ12 (t−s) dsku − uek
ρ1 −∞ t
L 1 1

= − ku − uek
ρ1 λ12 λ11
L
= ku − uek,
α1
882 ZHITING XU

which indicates that the operator F1 is continuous with respect to the norm k · k.
Similarly, we also can show that operator Fi , : Γ → Γ, i = 2, 3, is continuous with
respect to the norm k · k.
Next, we show that F is compact with respect to the norm k·k. In fact, note that
there exists H0 > 0 such that |Hi (u(t)) ≤ H0 (i = 1, 2, 3) for u = (u1 , u2 , u3 ) ∈ Γ.
Consequently, for any given t1 , t2 ∈ R (t2 > t1 ), and u = (u1 , u2 , u3 ) ∈ Γ, keeping
in mind that λ11 < 0 < λ12 , then
1  t2 λ11 (t2 −s)
Z Z t1 
F1 (u)(t2 ) − F1 (u)(t1 ) = e − eλ11 (t1 −s) H1 (u)(s)ds
ρ1 −∞ −∞
 Z +∞ Z +∞ 
+ eλ12 (t2 −s) − eλ12 (t1 −s) H1 (u)(s)ds
t2 t1

t1 λ11 (t1 −s)
Z Z t2
H0  λ11 (t2 −t1 )
≤ 1−e e ds + eλ11 (t2 −s) ds
ρ1 −∞ t1

+∞ λ12 (t1 −s)
Z Z t2 
+ eλ12 (t2 −t1 ) − 1 e ds + eλ12 (t1 −s) ds
t2 t1
2H0  eλ11 (t2 −t1 ) − 1 1 − e−λ12 (t2 −t1 ) 
= +
ρ1 λ11 λ12
4H0
≤ |t2 − t1 |,
ρ1
since et ≥ 1 + t for all t ∈ R. Similarly, one also gets
4H0
F2 (u)(t2 ) − F2 (u)(t1 ) ≤ |t2 − t1 |,
ρ2
and
4H0
F3 (u)(t2 ) − F3 (u)(t1 ) ≤|t2 − t1 |.
ρ3
It follows that F (Γ) is equicontinuous with respect to the norm k·k. Recall that F (Γ)
is uniformly bounded (by Lemma 2.6). Then the Arzelá-Ascoli theorem implies that
F : Γ → Γ is compact with respect to the norm k · k. Therefore, we complete the
proof.

3. Existence and non-existence of traveling wave solutions. In this section,

we shall state precisely and prove the main results of this paper.

3.1. Existence of traveling wave solutions. Here we establish the existence of

traveling waves for system (2.1) with d2 = d3 =: d. To begin with, we first give
some propositions.
Proposition 3.1. Assume that R0 > 1. For any c > c∗ , system (2.1) with d2 =
d3 =: d admits a nontrivial traveling wave solution (u1 (x+ct), u2 (x+ct), u3 (x+ct))
satisfying
γ βS 0
γ βS 0

0 < u1 (t) < S0 , 0 < u2 (t) < − 1 , 0 < u3 (t) < − 1 , ∀ t ∈ R.
aα γ a γ
(3.1)
And, it holds.
lim (u1 (t), u2 (t), u3 (t)) = (S 0 , 0, 0), lim (u01 (t), u02 (t), u03 (t)) = (0, 0, 0). (3.2)
t→−∞ t→−∞
 A DIFFUSIVE SEIR EPIDEMIC MODEL 883

Furthermore,
u02 (t) u03 (t)
lim = λ1 , lim = λ1 .
t→−∞ u2 (t) t→−∞ u3 (t)

Proof. In view of Lemmas 2.6 and 2.7, it follows from Schauder fixed point theorem
that there exists a pair of u = (u1 , u2 , u3 ) ∈ Γ, which is a fixed point of the operator
F . Consequently, the solution (u1 (x + ct), u2 (x + ct), u3 (x + ct)) is a traveling wave
solution of system (2.1), and
γ βS 0 1 βS 0
0 ≤ u1 (t) ≤ S 0 , 0 ≤ u2 (t) ≤



− 1 , 0 ≤ u3 (t) ≤ − 1 , ∀ t ∈ R.
aα γ a γ
We claim that the strict inequalities hold. Indeed, note that u = (u1 , u2 , u3 ) ∈ Γ
is a fixed point of the operator F , then u1 (t) = F1 (u1 , u2 , u3 )(t). Consequently,
u1 (t) = F1 (u1 , u2 , u3 )(t) ≥ F1 (u1 , u2 , u3 )(t)
1  t λ11 (t−s)
Z Z +∞ 
= e + eλ12 (t−s) H1 (u1 , u2 , u3 )(s)ds
ρ1 −∞ t
β
 t λ11 (t−s)
Z Z +∞
1 
≥ α1 − e + eλ12 (t−s) u1 (s)ds
ρ1 a −∞ t
> 0,
since u1 (t) is continuous and is not identically zero, and
β

H1 (u1 , u2 , u3 )(t) = α1 u1 (t) − βu1 (t)g(u3 (t)) ≥ α1 − u (t), ∀ t ∈ R.
a 1
Similarly, we can obtain that other inequalities are also strict ones. Thus, (3.1)
holds.
On the other hand, in view of the following inequalities
1 σt (1) (1)
S0 − e ≤ u1 (t) < S 0 , v2 eλ1 t 1 − M et ≤ u2 (t) ≤ v2 eλ1 t ,


σ
and
eλ1 t 1 − M et ≤ u3 (t) ≤ eλ1 t , ∀ t ∈ R,

which follows that

lim u1 (t) = S 0 , lim u2 (t) = 0, lim u3 (t) = 0,
t→−∞ t→−∞ t→−∞

and
(1)
lim e−λ1 t u2 (t) = v2 , lim e−λ1 t u3 (t) = 1. (3.3)
t→−∞ t→−∞

Note that u = (u1 , u2 , u3 ) ∈ Γ is a fixed point of the operator of F . Applying

L’Höspital rule to the maps F1 , F2 and F3 , it is easy to see that
lim (u01 (t), u02 (t), u03 (t)) = 0.
t→−∞

Hence, we have shown that (3.2) holds.

Now, integrating both sides of the second equation of (2.3) from −∞ to t gives
Z t Z t
0
du2 (t) = cu2 (t) − β u1 (s)g(u3 (s))ds + α u2 (s)ds. (3.4)
−∞ −∞
884 ZHITING XU

Thus, recall that (3.3), by L’Höspital rule again, we get

c Z t
β
lim e−λ1 t u02 (t) = lim e−λ1 t u2 (t) − e−λ1 t u1 (s)g(u3 (s))ds
t→−∞ t→−∞ d d −∞
Z t
α 
+ e−λ1 t u2 (s)ds
d −∞
(1)
(cλ1 + α)v2 − βS 0
=
λ1 d
(1)
= λ 1 v2 ,
u02 (t)
which implies, by the first one of (3.3), lim = λ1 . Similarly, we also show
t→−∞ u2 (t)
u03 (t)
lim = λ1 . Hence, we complete the proof.
t→−∞ u3 (t)

Proposition 3.2. Assume that R0 > 1. Then, for any c > c∗ , let (u1 (x+ct), u2 (x+
ct), u3 (x+ct)) be a nontrivial traveling wave solution of system (2.1) with d2 = d3 =:
d satisfying (3.2). Then
(1) u1 (t) is monotonically decreasing in t ∈ R;
(2) lim (u1 (t), u2 (t), u3 (t)) = (S0 , 0, 0), lim (u01 (t), u02 (t), u03 (t)) = (0, 0, 0).
t→+∞ t→+∞
(3) It holds
Z +∞
c 0
u1 (s)g(u3 (s))ds = (S − S0 ),
−∞ β
and
Z +∞ Z +∞
c 0 c 0
u2 (t)dt = (S − S0 ), u3 (t)dt = (S − S0 ).
−∞ α −∞ γ
Proof. We first show the conclusion (1) holds. Indeed, in view of the fact (3.2),
integrating the two sides of the first equation of (2.3) from −∞ to t follows
Z t
d1 u01 (t) = c(u1 (t) − S 0 ) + β u1 (s)g(u3 (s))ds. (3.5)
−∞
We now claim that the integral
Z +∞
u1 (s)g(u3 (s))ds < +∞. (3.6)
−∞

Indeed, if not, note that the fact 0 < u1 (t) < S 0 for all t ∈ R, by (3.5), we then
conclude that there exists δ0 > 0 such that u01 (t) > δ0 for all large t > 0, which
implies that u1 (t) → +∞ as t → +∞, this is contradiction. Hence, the improper
R +∞
integral −∞ u1 (s)g(u3 (s))ds converges, i.e., (3.6) holds. As a result that it follows
u01 (t) is uniformly bounded for all t ∈ R. Here it is clear from the first equation of
(2.3) that
c β c
(e− d1 t u01 (t))0 = e− d1 t u1 (t)g(u3 (t)), ∀ t ∈ R.
d1
Integrating the last equality from t to +∞ yields
β − dc t +∞ − dc s
Z
0
u1 (t) = − e 1 e 1 u1 (s)g(u3 (s))ds < 0, ∀ t ∈ R,
d1 t
which, together with the fact u1 (t) > 0 and u3 (t) > 0 are continuous in t ∈ R,
implies u01 (t) < 0 for all t ∈ R. Thus, u1 (t) is monotonically decreasing in t ∈ R, that
 A DIFFUSIVE SEIR EPIDEMIC MODEL 885

is, the conclusion (1) holds. And let S0 := lim u1 (t), consequently, S 0 > S0 ≥ 0.
t→+∞
Note that u2 (t) satisfies the second equation of (2.3). Then
β  t λ021 (t−s)
Z Z +∞ 
0
u2 (t) = 0 e + eλ22 (t−s) u1 (s)g(u3 (s))ds, ∀ t ∈ R. (3.7)
ρ2 −∞ t

here,
√ √
c− c2 + 4dα c+ c2 + 4d2 α
λ021 = , λ022 = , ρ0 = d2 (λ022 − λ021 ).
2d 2d
R +∞
By (3.6), let −∞ u1 (t)g(u3 (t))dt =: A0 < ∞, it follows from Fubini’s theorem that
Z +∞ Z +∞  Z t Z +∞
β 0 0

u2 (t)dt = 0 eλ21 (t−s) + eλ22 (t−s) u1 (s)g(u3 (s))dsdt
−∞ ρ2 −∞ −∞ t
β 1 1  β
= 0 − + A0 = A0 . (3.8)
ρ2 λ21 λ22 α
Note that for (u1 , u2 , u3 ) ∈ Γ, it is clear that
S0
0 < u1 (t)g(u3 (t)) ≤ , ∀ t ∈ R.
a
Then, for any t ∈ R,
Z t Z +∞
β 0 0

|u02 (t)| ≤ 0 |λ021 | eλ21 (t−s) + λ22 eλ22 (t−s) |u1 (s)g(u3 (s))|ds
ρ2 −∞ t
0 Z t Z +∞
βS 0 λ021 (t−s) 0 0

≤ 0 |λ21 | e ds + λ22 eλ22 (t−s) ds
aρ2 −∞ t
2βS 0
= .
aρ02
So that, u02 (t) is uniformly bounded, which, together with that u2 (t) > 0 is inte-
grable on R (by (3.8)), implies lim u2 (t) = 0. In fact, we can find a number
t→+∞
ε0 > 0, a sequence tn → +∞ as n → ∞ and a number δ > 0 such that u1 (t) ≥ ε0
for all |t − tn | < δ, which contradicts the integrability of u2 (t) on R. Furthermore,
recall that lim u1 (t) = S0 exists, letting t → +∞ in (3.5), and note that (3.6),
t→+∞
we know lim u01 (t) exists. It follows from the fact u01 (t) < 0 for all t ∈ R that
t→+∞
lim u01 (t) ≤ 0. Consequently, lim u01 (t) = 0. Otherwise, lim u01 (t) < 0, which
t→+∞ t→+∞ t→+∞
implies that lim u1 (t) = −∞, a contradiction to the fact u1 (t) > 0 for all t ∈ R.
t→+∞
Recall that lim u01 (t) = 0 and lim u1 (t) = S0 . Letting t → +∞ in (3.5), we
t→+∞ t→+∞
have
Z +∞
c 0

A0 = u1 (t)g(u3 (t))dt = S − S0 . (3.9)
−∞ β
Combining (3.8) with (3.9), we derive that
Z +∞
c 0

u2 (t)dt = S − S0 . (3.10)
−∞ α
Letting t → +∞ in (3.4), by (3.9) and (3.10), we get lim u02 (t) = 0.
t→+∞
886 ZHITING XU

On the other hand, note that u3 (t) satisfies the third equation of (2.3). Then
α  t λ031 (t−s)
Z Z +∞ 
0
u3 (t) = 0 e + eλ32 (t−s) u2 (s)ds, ∀ t ∈ R. (3.11)
ρ3 −∞ t
here,
p p
c−c2 + 4dγ c + c2 + 4d2 γ
=λ031 0
, λ32 = , ρ03 = d(λ032 − λ031 ).
2d 2d
Then, by Fubini’s theorem again, by (3.11),
Z +∞
α +∞  t λ031 (t−s)
Z Z Z +∞ 
0
u3 (t)dt = 0 e + eλ32 (t−s) u2 (s)dsdt
−∞ ρ3 −∞ −∞ t
α 1 1 c 0
= 0 − 0 + 0 (S − S0 )
ρ3 λ31 λ32 α
c
= (S 0 − S0 ). (3.12)
γ
As the same as the proof of the before, we also get u03 (t) is uniformly bounded,
together with (3.12), we get lim u3 (t) = 0. Integrating the last equality of (2.3)
t→+∞
from −∞ to t gives
Z t Z t
du03 (t) = cu3 (t) + γ u3 (s)ds − α u2 (s)ds.
−∞ −∞
Letting t → +∞ in the above, which, together with (3.10) and (3.12), implies
lim u03 (t) = 0. Consequently, we have shown the conclusion (2) holds. Obviously,
t→+∞
it follows from (3.9), (3.10) and (3.12) that the conclusion (3) holds. The proof is
completed.
In view of Propositions 3.1 and 3.2, we are in a position to state and prove the
existence of traveling wave solutions of system (2.1) with d2 = d3 =: d.
Theorem 3.1. Assume that R0 > 1. For any c > c∗ , system (2.1) with d2 = d3
admits a nontrivial traveling wave solution (u1 (x + ct), u2 (x + ct), u3 (x + ct)) such
that
(1) The following boundary conditions
lim (u1 (t), u2 (t), u3 (t)) = (S 0 , 0, 0), lim (u1 (t), u2 (t), u3 (t)) = (S0 , 0, 0) (3.13)
t→−∞ t→+∞

hold. Moreover, S0 ≤ βγ .
(2) u1 (t) is monotonically decreasing in t ∈ R, and
0 < ui (t) < S 0 − S0 , ∀ t ∈ R, i = 2, 3.
Proof. To prove Theorem 3.1, by Propositions 3.1 and 3.2, it suffices to show that
γ
S0 ≤ and 0 < ui (t) < S 0 − S0 , ∀ t ∈ R, i = 2, 3.
β
To this end, we first show S0 ≤ βγ holds. If not, in the other words, S0 > βγ , note
u1 (t) u1 (t)
> 12 S0 + βγ


that lim 1+au 3 (t)
= S0 , then there exists t1 > 0 such that 1+au 3 (t)
t→+∞
βu1 (t)
for all t > t1 . Consequently, 1+au 3 (t)
− γ > 12 (βS0 − γ) > 0 for t > t1 . Adding the
second equation and the third one of (2.3) gives
 βu (t) 
1
d(u2 (t) + u3 (t))00 − c(u2 (t) + u3 (t))0 = − − γ u3 (t) < 0. (3.14)
1 + au3 (t)
 A DIFFUSIVE SEIR EPIDEMIC MODEL 887

Recall that lim ui (t) = 0 and lim u0i (t) = 0 (by Proposition 3.2(2)), i = 2, 3,
t→+∞ t→+∞
integrating (3.14) from t to +∞, t > t1 , we obtain
Z ∞
0 βu1 (s) 
d(u2 (t) + u3 (t)) − c(u2 (t) + u3 (t)) = − γ u3 (s)ds > 0, t > t1 ,
t 1 + au3 (s)
d
which yields (e− c t (u1 (t) + u2 (t))0 > 0 for t > t1 . Thus,
d d
e− c t (u1 (t) + u2 (t)) < lim e− c t (u1 (t) + u2 (t)) = 0
t→+∞

for t > t1 , a contradiction. Hence the assertion that S0 ≤ βγ holds.

Next, we show 0 < ui (t) < S 0 − S0 , ∀ t ∈ R, i = 2, 3. In fact, we let
α t α +∞ c (t−s)
Z Z
N2 (t) := u2 (t) + u2 (s)ds + ed u2 (s)ds, ∀ t ∈ R, (3.15)
c −∞ c t
which, by L’Höspital’s rule, together with (3.2), implies that
lim N2 (t) = 0, lim N2 (t) = S 0 − S0 . (3.16)
t→−∞ t→+∞

Similarly, we obtain, by differentiating (3.15) once,

γ +∞ c (t−s)
Z
0 0
N2 (t) = u2 (t) + ed u2 (s)ds, (3.17)
d t
which yields
lim N20 (t) = 0, lim N20 (t) = 0.
t→−∞ t→+∞
Furthermore, by differentiating (3.17) once, and noting that u2 (t) satisfies the sec-
ond equation of (2.3), one gets
cN20 (t) = dN200 (t) + βu1 (t)g(u3 (t)), ∀ t ∈ R,
which follows that
Z +∞
β ct c
N20 (t) = ed e d s u1 (s)g(u3 (s))ds > 0, ∀ t ∈ R.
d t
Consequently, N2 (t) is increasing in t ∈ R. Further, by (3.16),
0 < u2 (t) < N2 (t) ≤ lim N2 (t) = S 0 − S0 , ∀ t ∈ R.
t→+∞

Similarly, we also get

0 < u3 (t) < S 0 − S0 , ∀ t ∈ R.
This completes the proof.
3.2. Non-existence of traveling wave solutions. In this subsection, we will
establish the nonexistence of traveling waves for system (2.1) either R0 < 1 and
c ≥ 0, or R0 > 1 and c ∈ (0, c∗ ).
Firstly, it is easily see that a traveling wave solution (u1 (x+ct), u2 (x+ct), u3 (x+
ct)) of (2.1) satisfying (3.13). Then u2 (t) and u3 (t) satisfies (3.7) and (3.11), re-
spectively. Furthermore, an application of L’Höspital rule to u02 (t) and u03 (t), we
can easily get the following asymptotic behavior of u2 (t) and u3 (t) as follows
lim ui (t) = 0, lim u0i (t) = 0, i = 2, 3. (3.18)
t→±∞ t→±∞

Theorem 3.2. Assume that R0 ≤ 1. Then, for any c ≥ 0, system (2.1) with d2 =
d3 has no nontrivial and nonnegative traveling wave (u1 (x+ct), u2 (x+ct), u3 (x+ct))
satisfying (3.13).
888 ZHITING XU

βu1 (t)
Proof. When R0 ≤ 1, i.e., βS 0 ≤ γ, then 1+au 3 (t)
≤ βS 0 for all t ∈ R since
u1 (t) ≤ S 0 for all t ∈ R (by Theorem 2.1). Adding the second equation and third
one of (2.3), we get (3.14) holds. Consequently,
c
0 1  βu1 (t) 1
e− d t (u2 (t) + u3 (t))0 = − − γ u3 (t) ≥ − (βS 0 − γ)u3 (t) ≥ 0,


d 1 + au3 (t) d
c
which implies that the function e d t (u2 (t) + u3 (t))0 is nondecreasing in t ∈ R. It
c
follows from the fact (3.18) and lim e− d t = 0 that (u2 (t) + u3 (t))0 ≤ 0 for all
t→+∞
t ∈ R. Again by (3.18), we know that u2 (t) + u3 (t) = 0 for all t ∈ R, which is a
contradiction. This ends the proof.

Theorem 3.3. Assume that R0 > 1 and any c ∈ (0, c∗ ). Then system (2.1)
with d2 = d3 has no nontrivial and nonnegative traveling wave (u1 (x + ct), u2 (x +
ct), u3 (x + ct)) satisfying (3.13).
Proof. Assume that there exists a nontrivial and nonnegative traveling wave so-
lution (u1 (x + ct), u2 (x + ct), u3 (x + ct)) of system (2.1) satisfying (3.2). Then
(u1 (t), u2 (t), u3 (t)) satisfies
du002 (t) − cu02 (t) − αu2 (t) + βu1 (t)g(u3 (t)) = 0,
(3.19)
du003 (t) − cu03 (t) − γu3 (t) + αu2 (t) = 0.
Set u02 = v2 , u03 = v3 . Then (3.19) can be rewritten as
y 0 = By + f (t, y),
where y = (u1 , u2 , u3 , u4 )T , and
0 1 0 0
 
0
 
 α c βS 0  βS 0
β
−

 0  u3 − u1 g(u3 ) 

B= d d d  , f (t, y) =  .
 0d d
 0 
0 0 1  
α γ c
 
− 0 0
d d d
Since (u1 (t), u2 (t), u3 (t)) satisfies (3.13), we have lim y(t) = 0. It is easy to show
t→−∞
that the characteristic equation for B is given by detA(λ) = 0. Since lim u1 (t) =
t→−∞
S 0 , for any small constant  > 0 there exists t0 ∈ R such that
kf (t, y1 ) − f (t, y2 )k ≤ ky1 − y2 k, ∀ t < t0 .
Recall that the translation invariance of the traveling wave solution, without loss
of generality, we take t0 = 0. It follows from Lemma 2.1(3) that the equation
detA(λ) = 0 have one positive eigenvalue, one negative eigenvalue and a pair of
complex conjugate eigenvalue with positive real parts. Hence we get the similar
required version as Case (I) in [33, page 488]. Then following the steps in the proof
of [33, Lemma 3.1], we can obtain
sup{ui (t)e−α0 t } < +∞, sup{|u0i (t)|e−α0 t } < +∞, i = 2, 3, (3.20)
t∈R t∈R

for some α0 > 0. Also, by the second equation and the third one of (2.3), we get
sup{|u00i (t)|e−α0 t } < +∞, i = 2, 3. (3.21)
t∈R
 A DIFFUSIVE SEIR EPIDEMIC MODEL 889

Next we consider S 0 u3 (t) − u1 (t)g(u3 (t)). Clearly,

S 0 u23 (t) S 0 u3 (t)
S 0 u3 (t) − u1 (t)g(u3 (t)) = + (S 0 − u1 (t)) .
1 + αu3 (t) 1 + αu3 (t)
u23 (t)
Then, note that 1+αu3 (t) ≤ u23 (t) for all t ∈ R, by (3.20),
n u2 (t) o
sup 3
e−2α0 t < +∞. (3.22)
t∈R 1 + αu3

Let U1 (t) = S 0 − u1 (t) ≥ 0 for all t ∈ R. Integrating the first equation of (2.3) from
−∞ to t yields
d1 U10 (t) − cU1 (t) = −f (t), (3.23)
in which Z t
f (t) = β u1 (s)g(u3 (s))ds ≤ C0 eα0 t , ∀ t ∈ R
−∞
for some constant C0 > 0. That is, f (t) = O(eα0 t ) as t → −∞. Solving (3.23)
follows
1 dc t 0 − dc s
Z
c
d1 t
U1 (t) = U1 (0)e + e 1 e 1 f (s)ds, ∀ t ∈ R,
d1 t
which, together with the fact f (t) = O(eα0 t ) as t → −∞, follows that U1 (t) =
0
O(eα0 t ) as t → −∞, where α00 = min{α0 , dc1 }. In view of the fact 0 ≤ U1 (t) ≤ S 0
for all t ∈ R, we have o
0
sup{U1 (t)e−α0 t < +∞. (3.24)
t∈R
Thus, by (3.22) and (3.24), we get
0
sup |u1 (t)g(u3 (t)) − S 0 u3 (t)|e−(α0 +α )t < +∞.

(3.25)
t∈R

Now we rewrite system (3.19) as follows

du002 (t) − cu02 (t) − αu2 (t) + βS 0 u3 (t) = −β u1 (t)g(u3 (t)) − S 0 u3 (t) ,

du003 (t) − cu03 (t) − γu3 (t) + αu2 (t) = 0.

Then, we introduce two-side Laplace transform on the above system and get
Z +∞
0
e−λt u1 (t)g(u3 (t)) − S 0 u3 (t) dt,


h1 (λ)L2 (λ) + βS L3 (λ) = −β
−∞
h2 (λ)L3 (λ) + αL2 (λ) = 0,
R +∞
where Li (λ) = −∞ e−λt φi (t)dt, λ > 0. Multiplying the first equation and the
second one of the above system yields
Z +∞
e−λt u1 (t)g(u3 (t)) − S 0 u3 (t) dt.


detA(λ)L3 (λ) = αβ (3.26)
−∞
Note that both sides of (3.26) is defined for λ ∈ C with 0 < Reλ < α0 . Furthermore,
the two Laplace integrals can be analytically continued to the whole right half line;
otherwise, the integral on the left of (3.26) has singularity at λ = λ∗0 and it is
analytic for all λ < λ∗0 , cf. [4, 20, 22, 23]. But, by (3.25), the integral on the right
of (3.26) is actually analytic for all λ ≤ λ∗0 + α0 , a contradiction. Thus, (3.26) holds
for all λ > 0, and can rewritten as
Z +∞  αβu1 (t) 
e−λt u3 (t) detA(λ) + − S 0 dt = 0. (3.27)
−∞ 1 + au3 (t)
890 ZHITING XU

Here, in view of the fact that detA(λ) = 0 is always positive for all λ > λ0 , here
λ0 is the unique positive root of detA(λ) = 0 when c ∈ (0, c∗ ) (by Lemma 2.1(3)).
Then
αβu1 (t)
detA(λ) + − S 0 → ∞ as λ → +∞,
1 + au3 (t)
which lead to a contradiction to (3.27). Finally, we conclude that if R0 > 1 and
c ∈ (0, c∗ ), then there does not exist a non-trivial and non-negative traveling wave
solutions satisfying the boundary condition (3.13). Thus we conclude the proof.

4. Discussion. There have been intensive studies about the existence and non-
existence of traveling waves for some diffusive SIR model (see, for instant, [1, 8,
12, 20, 22, 23, 30]). In this paper, we have incorporated the exposed individuals
and the population diffusions into the classic SIR model (1.1), and then presented
a diffusive SEIR epidemic model (1.3) with a saturating incidence rate. By an
abstract treatment, we studied the well posedness of system (2.1). Also, applying
[21, Theorem 2.3], we gave the explicit formula of the basic reproduction number
R0 for system (2.1).
For the model under consideration, the traveling wave solutions describes the
disease propagation into the susceptible individuals from an initial disease-free equi-
librium to the final, also disease-free equilibrium. In this paper, we first use the iter-
ation process to construct the vector-value upper-lower solutions for (2.3). Together
with the Schauder fixed theorem, we can establish existence of such a traveling wave
solution. Second, we use the two-sides Laplace transform to the non-existence of
such a traveling wave solution. These results could formulate the possible propaga-
tion modes of the disease.
Theorem 3.1 gives some asymptotic behaviors of the traveling wave solution
(u1 (x + ct), u2 (x + ct), u3 (x + ct)) of system (2.1) with speed c > c∗ . Note that,
in the condition, S 0 > 0 is a constant representing the number of the susceptible
individuals before being infected. Clearly, at any fixed x ∈ R, Theorem 3.1(1) and
(2) describe that all the individuals were susceptible a long time ago (t → −∞) and
all the susceptible individuals will be decreasing to S0 after a long time (t → +∞).
In particular, if S0 = 0 (due to S0 may be zero), then all the susceptible individuals
will become the removed individuals also after a long time (t → +∞). Hence, the
natural question arises. Can we know the value of S(+∞) = S0 ? As pointed in [22],
for general system such as (1.2) with nonzero diffusion terms, it seems impossible
to obtain the value of S0 , see [22] for a brief discussion on related works for this
problem. We conjecture S0 = 0, but unfortunately, we do not know to prove it.
Here it follows from Theorem 3.1(1) we only provide an upper bound for S0 , that
is, S0 ≤ βγ .
Further, by Theorems 3.1 and 3.3, we see that the constant c∗ defined by Lemma
2.1 is indeed the minimal wave speed c∗ . Hence, the natural question arises, is the
constant c∗ the asymptotic speed of propagation? However, we have no way to
prove the c∗ is the asymptotic speeds of spread for system (2.1). We have noticed
that Liang and Zhao [13] obtained some general results on the asymptotic speeds
of spread for monotone semiflows, which two speeds indeed coincide. Since system
(2.1) cannot generate monotone semiflows, the theory [13] is invalid for the system
(2.1). How to overcome technical problems that prevent a full analysis of relations
of them should be a challenging work and leave it as our further project.
 A DIFFUSIVE SEIR EPIDEMIC MODEL 891

Theorems 3.1 and 3.3 combined establish a threshold condition for the existence
and non-existence of traveling wave solutions in terms of the basic reproduction
number R0 of system (2.1). Namely, R0 > 1 implies c∗ > 0. It follows from
Theorems 3.1 that the propagation of the pathogen as a wave with a fixed shape
and a fixed speed for any c > c∗ from the initial disease-free equilibrium (S 0 , 0, 0)
to another disease-free equilibrium (S0 , 0, 0). For every c < c∗ , Theorem 3.3 implies
that system (2.1) has no nontrivial and nonnegative traveling wave connecting the
two equilibria (S 0 , 0, 0) and (S0 , 0, 0). The study of traveling wave solutions provides
important insight into the spatial patterns of invading diseases.
As observed, the proofs of Theorem 3.1(1) and Theorem 3.2 depend heavily
on the choice of the same diffusion rate d for exposed individuals and infected
ones, i.e., the last two equations of system (2.1) have the same diffusion rate d.
Hence, some results obtained here can not be extrapolated for a more general class
of population models. However, in realistic communities, the infectious disease
transmits in the spatial movement of the region by the different diffusion rates
for exposed and infected individuals. From the view of mathematical biology, it
is important problem to deal with this case and we will consider the topic in our
further study.

Acknowledgements. The author is grateful to the editors and anonymous referees

for the careful reading of the original manuscript and useful comments that helped
to improve the manuscript.

REFERENCES

[1] Z. Bai and S. Zhang, Traveling waves of a diffusive SIR epidemic model with a class of
nonlinear incidence rates and distributed delay, Commun. Nonlinear Sci. Numer. Simul.,
(1–3) (2015), 1370–1381.
[2] F. Braner and C. Castillo-Chávez, Mathematical Models in Population Biology and Epidemi-
ology, Springer, New York, 2001.
[3] V. Capasso and G. Serio, A generalization of the Kermack-Mckendrick deterministic epidemic
model, Math. Biosci., 42 (1978), 43–61.
[4] J. Carr and A. Chmaj, Uniquence of the travling waves for nonlocal monostabe equations,
Proc. Amer. Math. Soc., 132 (2004), 2433–2439.
[5] J. Fang and X.-Q. Zhao, Monotone wavefronts for partially degenerate reaction-diffusion
systems, J. Dyn. Diff. Equat., 21 (2009), 663–680.
[6] H. W. Hethcote, The mathematics of infectious diseases, SIAM Review , 42 (2000), 599–653.
[7] H. W. Hethcote and van den Driessche, Some epidemiological models with nonlinear incidence,
J. Math. Biol., 29 (1991), 271–287.
[8] Y. Hosono and B. Ilyas, Traveling waves for a simple diffusive epidemic model, Math. Models.
Methods Appl. Sci., 5 (1995), 935–966.
[9] J. Huang and X. Zou, Existence of traveling wavefronts of delayed reaction diffusion systems
without monotonicity, Discrete Contin. Dyn. Sys., 9 (2003), 925–936.
[10] W. O. Kermack and A. G. McKendrick, Contribution to the mathematical theory of epidemics,
Proc. R. Soc. Lond. (Ser. A), 115 (1927), 700–721; part II, Proc. R. Soc. Lond. (Ser. A),
138 (1932), 55–83; part III, Proc. R. Soc. Lond. (Ser. A), 141 (1933), 94–112.
[11] W.-T. Li, G. Lin and S. Ruan, Existence of travelling wave solutions in delayed reaction-
diffusion systems with applications to diffusion-competition systems, Nonlinearity, 19 (2006),
1253–1257.
[12] W.-T. Li, G. Lin, C. Ma and F.-Y. Yang, Travelling wave solutions of a nonlocal delayed SIR
model with outbreak threshold, Discrete Contin. Dyn. Sys. (Ser.B), 19 (2014), 467–484.
[13] X. Liang and X.-Q. Zhao, Asymptotic speed of spread and traveling waves for monotone
semiflows with applications, Comm. Pure. Appl. Math., 60 (2007) 1–40.
[14] Y. Lv, R. Yuan and Y. Pei, The imact of predation on the coexistence and competitive
exclusion of pathogens in prey, Math. Biosci., 251 (2014), 16–29.
892 ZHITING XU

[15] S. Ma, Traveling wavefronts for delayed reaction-diffusion systems via a fixed point theorem,
J. Differential Equations, 171 (2001), 294–314.
[16] R. H. Martin and H. L. Smith, Abstract functional differential equations and reaction-diffusion
systems, Trans. Amer. Math. Soc., 321 (1990), 1–44.
[17] J. D. Murray, Mathematical Biology, I and II, third edn., Springer-Verlag, New York, 2002.
[18] S. Ruan and W. Wang, Dynamical behavior of an epidemic models with a nonlinear incidence
rate, J. Differential Equations, 188 (2003), 135–163.
[19] H. Wang, Spreading speeds and traveling waves for non-cooperative reaction-diffusion stsyems,
J. Nonlinear Sciences, 21 (2011), 747–783.
[20] H. Wang and X.-S. Wang, Travelling waves phenomena in a Kermack-McKendrick SIR model,
J. Dyn. Diff. Equat., DOI 10.1007/s10884-015-9506-2.
[21] W. Wang and X.-Q. Zhao, A nonlocal and time-delayed reaction-diffusion model of dengue
transmission, SIAM J. Appl. Math., 71 (2011), 147–168.
[22] X.-S. Wang, H. Wang and J. Wu, Travelling waves of diffusive predator-prey systems: disease
outbreak propagation, Discrete Contin. Dyn. Sys., 32 (2012), 3303–3324.
[23] Z.-C. Wang and J. Wu, Travelling waves of a diffusive Kermack-McKendrick epidemic model
with non-local delayed transmission, Proc. R. Soc. Lond. (Ser. A), 466 (2010), 237–261.
[24] Z.-C. Wang and J. Wu, Travelling waves in a bio-reactor model with stage-structure, J. Math.
Anal. Appl., 385 (2012), 683–692.
[25] Z.-C. Wang, J. Wu and R. Liu, Traveling waves of Avian influenza spread, Proc. Amer. Math.
Soc., 149 (2012), 3931–3946.
[26] P. Weng and X.-Q. Zhao, Spreading speed and traveling waves for a multi-type SIS epidemics
model, J. Differential Equations, 229 (2006), 270–296.
[27] J. Wu, Theory and Applications of Partial Functional Differential Equations, Springer, New
York, 1996.
[28] D. Xiao and S. Ruan, Global analysis of an epidemic model with nonmonotone incidence rate,
Math. Biosci., 208 (2007), 419–492.
[29] R. Xu and Z. Ma, Global stablity of a SIR epidemic model with nonlinear incidence rate and
time delay, Nonlinear Analysis: Real World Applications, 10 (2009), 3175–3189.
[30] Z. Xu, Traveling waves in a Kermack-Mckendrick epidemic model with diffusion and latent
period, Nonlinear Analysis, 111 (2014), 66–81.
[31] Y. Yang and D. Xiao, Influence of latent period and nonlinear incidence rate of the dynamics
of SIRS epidemiological models, Discrete Contin. Dyn. Sys. (Ser.B), 13 (2010), 195–211.
[32] L. Zhang, B. Li and J. Shang, Stability and travelling waves for a time-delayed population
system with stage structure, Nonlinear Analysis: Real World Applications, 13 (2012), 1429–
1440.
[33] T. Zhang and W. Wang, Existence of thaveling wave solutions for influenza model with
treatment, J. Math. Anal. Appl., 419 (2014), 469–495.

Received August 2015; revised December 2015.

E-mail address: [email protected]