1. Introduction
With the emergence of various infectious diseases in the development of human society, mathematical models have become an important tool for understanding the transmission dynamics of infectious diseases. In past decades, great attention has been paid to investigate the evolution of diseases through different mathematical models. For example, many different reaction-diffusion and integro-differential continuous models were established under the assumption that the populations disperse among continuous spaces, see [Reference Bai and Zhang2, Reference Cui, Lam and Lou8, Reference Ducrot, Magal and Ruan9, Reference Hu and Zou13–Reference Lou and Salako16, Reference Thieme and Zhao19, Reference Wang, Qiao and Wu21, Reference Wang, Lin and Ruan22, Reference Xu and Xiao25, Reference Xu27, Reference Xu29–Reference Yang, Li and Wang31, Reference Zhang, Wang and Zhao33, Reference Zhou, Li, Wei and Tian37]. However, due to the development of modern transportation, the mobility of population in real life usually has the characteristics of large scale and span. When the infected persons travel by buses, trains or airplanes, many epidemics (such as plague, SARS, COVID-19, H1N1 flu, etc.) can be easily spread between different discrete spaces, such as cities, countries or regions. Therefore, it is more realistic and important to consider the impact of population diffusion in patchy environments on the spread of disease.
In order to investigate the dynamics of disease transmission under the influence of a population dispersal among patches, Wang and Zhao [Reference Wang and Zhao20] proposed the following epidemic dynamical systems with population dispersal among $n$ patches:
where $S_{j}(t)$ and $I_{j}(t)$ denote the densities of susceptible individuals and infected individuals respectively in patch $ j$ at time $ t\geq 0$ ;
$a_{ii}\leq 0$ represents the emigration rate of susceptible individuals in $i-$ th patch;
$a_{ji}\geq 0\,(j\neq i)$ represents the immigration rate of susceptible individuals from $i-$ th patch to
$j-$ th patch;
$b_{ii}\leq 0$ represents the emigration rate of infective individuals in $i-$ th patch;
$N_{j}=S_{j}+I_{j}$ represents the total number of population in patch $j$ ;
$B_{j}({\cdot})$ represents the birth rate of population in the $j-$ th patch.
One can see from [Reference Wang and Zhao20] that system (1) admits a threshold dynamics for the uniform persistence and extinction of disease, provided the birth rate satisfies certain assumptions.
Later, to clarify the effects of habitat connectivity and movement rates on the disease transmission dynamics, Allen et al. [Reference Allen, Bolker, Lou and Nevai1] proposed the following frequency-dependent $SI$ metapopulation model which consists of $n$ patches:
where $\Omega =\{1,2,\cdot \cdot \cdot, n\}$ ;
$d_{S}$ and $d_{I}$ are positive diffusion rates for the susceptible and infected subpopulations (resp.);
$L_{ji}$ describes the degree of movement from patch $i$ to patch $j$ ;
$ \beta _{j}$ and $ \gamma _{j}$ are positive rates of disease transmission and recovery (resp.) in patch $j$ .
They proved the existence and stability of disease-free and endemic equilibria and established some threshold type results which can predict whether the disease will persist or die out. Their results link the spatial heterogeneity, habitat connectivity and rates of movement to disease persistence and extinction.
Note that both results of [Reference Allen, Bolker, Lou and Nevai1, Reference Wang and Zhao20] characterized the global dynamics of the disease equilibria in terms of the values of basic reproduction numbers for the considered models. It has been realized that the persistence and extinction of disease are related to whether the infectious source can spread between patches as a wave. This fact prompts some researchers to investigate the propagation phenomena of travelling waves for different epidemic patch models. Guo et al. [Reference Chen, Guo and Hamel6, Reference Fu, Guo and Wu11, Reference Wu23] considered comprehensively the travelling wave solutions for a class of epidemic patch model of the form (1), under the assumption that the population is distributed on infinite patches and spreads only in adjacent patches. The recent works [Reference Xu, Tan and Hsu28, Reference Zhang and Wu32, Reference Zhang, Wang and Liu34, Reference Zhou, Song and Wei36] made further generalization and development by introducing different types of nonlinear incidence rates. In these works, some threshold type results were established for the existence and nonexistence of travelling waves connecting two different equilibrium states. Let’s point out that all the aforementioned works considered the models that the population disperses between its adjacent patches, i.e., the population in the $j$ patch only interacts with those in the $j+1$ and $j-1$ patches. Such a characteristic in mathematics makes the wave equations of the models second-order difference equations. This brings some conveniences for the mathematical analysis to the models that considered in the aforementioned works. However, as mentioned above, the population in real world may spread over a large span due to the development of modern transportation.
On the other hand, it is known that for some diseases, such as the recent epidemic outbreak of COVID-19, the incubation period usually fluctuates in certain range due to the differences in immunologic tolerance between individuals. Therefore, the latent period from infection to onset of symptoms is often a variable. Inspired by the works [Reference Beretta, Hara, Ma and Takeuchi3, Reference Connell McCluskey7], it is more realistic to describe the incubation period via a weight function, which specifies the probability that an individual from uninfected to infection in a certain time interval.
To explore the spatial dynamics of disease spreads under the effects of large span diffusion and variable latency, we consider the following formulation of epidemic patch model:
where $ \tau \gt 0$ represents the maximum latent period from infection to onset of symptoms. Different to system (2), we assume that the disease transmission rate and recovery rate are isotropic (i.e., $\beta _i\equiv \beta$ , $\gamma _i\equiv \gamma$ ), and the population in patch $ j$ can dispersal to $j-i$ patch with probability $J_{k}(i)\,(k=1,2)$ for $i\in \mathbb{Z}$ . Similar to [Reference Beretta, Hara, Ma and Takeuchi3, Reference Connell McCluskey7], we assume that the variation of incubation period is described by a probability function $f({\cdot})$ satisfying
Note that the term $ S_{j}\displaystyle \int _{0}^{\tau }f(s)I_{j}(t-s)ds$ measures the infection force of disease in patch $j$ and at time $t$ . Actually, system (3) is a $SIR$ (Susceptible-Infectious-Removed) type epidemic model that the recovered individuals are not involved in the transmission of the disease as they will not be re-infected due to the protection of antibodies. This model describes a closed system without births and deaths. Our goal is to explore the propagation phenomena in system (3), especially travelling wave solutions. Although there have been many researches on the travelling waves of different discrete systems modelling epidemic dynamics, the analysis of systems with distributed delay and global interactions should be relatively more difficult. As we know, there seems to have no results on the wave propagation for this type of epidemic dynamical systems.
A travelling wave solution of system (3) means a solution propagating with a constant speed $c$ and a fixed profile. Mathematically, one can consider the ansatz
for some wave functions $S({\cdot})$ and $I({\cdot})$ defined on $\mathbb{R}$ , where $\xi =j+ct$ means the moving coordinate. Substituting the transformations of (4) into (3), we can obtain the profile equation
In order to explain the evolution process of disease from outbreak to extinction, we are interested in finding the positive solutions of system (5) that connect from the initial disease-free state to the final disease-free state, i.e., $(S(\xi ),I(\xi ))$ satisfies the following conditions:
The constant $S_{0}\gt 0$ represents the density of susceptible individuals before the onset of epidemics, while $ S_{\infty }\in [0,S_{0})$ represents the density of susceptible individuals after the onset of epidemics. Condition Reference Chen, Guo and Hamel(6) means that the travelling wave solutions are of mixed type, i.e., $S-$ component is front type and $I-$ component is pulse type. Biologically, the mixed type travelling wave indicates that the number of infected individuals increases first, and then decreases gradually until extinction.
In past years, there have been many literature working on the travelling waves of different discrete $SI$ type epidemic dynamical systems, see e.g., [Reference Chen, Guo and Hamel6, Reference Fu, Guo and Wu11, Reference San, Wang and Feng18, Reference Wu23, Reference Wu, Zhao, Zhang and Hsu24, Reference Xu, Tan and Hsu28, Reference Zhou, Song and Wei36, Reference Zhou, Yang and Hsu38] and so on. However, little is known so far for the models like (3) with non-adjacent diffusion and distributed delay. The main difficulty arises from the fact that the solutions of (3) have no priori upper bound when the basic reproduction number
is greater than one, which leads to the construction of bounded travelling wave solution becomes very difficult. To overcome the difficulty and for mathematical convenience, throughout this paper, we assume that $J_{k}({\cdot})$ $(k=1,2)$ are compactly supported so that
-
1. (J) $J_{k}(i)=0$ for $ |i|\gt 2$ and $J_{k}(i)=J_{k}({-}i)\gt 0$ for $ |i|\leq 2$ .
Focusing on the dispersal operator, one can see that system (2) is a particular case of system (1). In addition, focusing on the susceptible population equations, the dispersal operator in system (3) is also a particular case of that in system (2), e.g., we may take $L_{j,i}=L_{i,j}=J_1(|j-i|)$ . This fact also provides the reason why we impose the symmetry property on $J_k(i)$ in assumption (J). Moreover, the symmetry of the kernel functions can ensure that the travelling wave solutions propagate forward(i.e., wave speed $c\gt 0$ , cf. Lemma3.1), which is of particular interest in applications.
Based on the above assumption, we first establish some properties of solutions for the linearized equation of the profile equation (5) around the disease-free equilibrium $(S_0,0)$ . With the help of these properties, when the basic reproduction number $R_0$ is greater than one, we may apply the truncation method and develop some novel analytical techniques to establish the travelling wave solutions of (3) that satisfy the condition (6). Moreover, we consider the minimal speed problem of travelling waves, which is important in epidemiology since it is usually the speed at which the disease spreads. Our main results can be summarized as the following theorem.
Theorem 1.1. There exists a $c^\ast \gt 0$ such that the following statements are valid.
From Theorem1.1, one can see that $R_0$ is a threshold value in determining the occurrence of wave propagation of system (3). The critical speed $c^\ast$ is the minimal wave speed of travelling waves when $R_{0}\gt 1$ . Moreover, it is interesting to see that the population moves at different speed the disease will go extinct. Indeed, if we fix an initial time, then the $I$ component goes to zero as time goes to infinity.
To deal with the problem of minimal wave speed, we make use of some priori estimates and suitable limiting arguments. Let’s point out that the limiting arguments were used in many works to study the existence of front type minimal travelling waves (i.e., the travelling waves with minimal speed connecting a zero equilibrium and a certain positive equilibrium) for various evolution equations, see [Reference Brown and Carr4, Reference Chen and Guo5, Reference Fang and Zhao10, Reference Hsu and Yang14, Reference Thieme and Zhao19, Reference Xu and Xiao26, Reference Zhao and Xiao35] and so on. However, for $SI$ epidemic systems, little works have been done for the existence of mixed type minimal travelling waves connecting two disease-free equilibria. The difficulty comes from proving the non-triviality ( $S-$ component is non-constant, and $I-$ component is non-zero) of the limiting function and showing its asymptotic behaviour that connects two disease-free equilibria, cf. [Reference Wu23]. To overcome the difficulty, we use some limiting arguments and establish a crucial lemma (see Lemma 3.3) to prove the existence of minimal travelling wave of system (3) that connects two disease-free equilibria. Further, the nonexistence of travelling waves of system (3) is derived by using some priori estimates and the properties of solutions of the linearized profile equation.
Let’s remark that Theorem1.1 provides a complete characterization of the existence, nonexistence and minimal speed of travelling waves. To the best of our knowledge, this is the first result on the propagation dynamics of epidemic patch model with large span diffusion and variable incubation period.
The remainder of this paper is organized as follows. In Section 2, we establish some properties of the solutions for the linearized profile equation around the disease-free equilibrium. Some crucial priori estimates on wave profiles and wave speeds are given in section 3. In section 4, we first establish the existence of solutions for the profile system (5) over large finite domains. Then, we apply the truncation method via some different limiting arguments to prove the results of Theorem1.1.
2. Some properties of the linearized profile equation
Linearizing the second equation of (5) around the disease-free equilibrium $(S_{0}, 0)$ yields to the linear equation
By the assumption (J), it is clear that $\tilde{\Omega }=\{-2,-1,1,2\}$ . To establish a more general theoretical framework, we embed (7) into the following general form:
where $d,c\gt 0$ , $b_{0}\geq 0$ , $b\in \mathbb{R}$ and $J$ satisfies the assumption (J). Clearly, the equation (7) is a special form of (8) with $d\,:\!=\,d_I$ , $b_0\,:\!=\,\beta S_0$ , $b\,:\!=\,-(d_I\sum _{i\in \tilde{\Omega }}J_{2}(i)+\gamma )$ and $J(i)\,:\!=\,J_2(i)$ .
The characteristic equation of (8) is defined by
It is easy to verify that $ \Psi (d, b_{0},b,c, \cdot )=0$ has at most two real roots since it is convex with respect to $\lambda$ . Especially, when $R_{0}={\beta S_{0}}/{\gamma }\gt 1$ and $J(i)=J_2(i)$ , it is obvious that $\Psi (d_{I}, \beta S_{0},-d_{I}\sum _{i\in \tilde{\Omega }}J_{2}(i)-\gamma, c, \lambda )$ is decreasing in $c$ with $\Psi (d_{I}, \beta S_{0},-d_{I}\sum _{i\in \tilde{\Omega }}J_{2}(i)-\gamma, c, 0) \gt 0$ and $\Psi (d_{I}, \beta S_{0},-d_{I}\sum _{i\in \tilde{\Omega }}J_{2}(i)-\gamma, 0, \lambda )\gt 0$ for all $\lambda \gt 0$ . Thus, the constant $c^{*}$ , defined by
is well-defined. In addition, we have the following properties:
-
∘ if $c\gt c^{*}$ , $\Psi (d_{I}, \beta S_{0}, -d_{I}\sum _{i\in \tilde{\Omega }}J_{2}(i)-\gamma, c, \lambda )=0$ has two distinct positive roots $\lambda _{1}=\lambda _{1}(c)\lt \lambda _{2}=\lambda _{2}(c)$ ;
-
∘ if $c=c^\ast$ , $\Psi (d_{I}, \beta S_{0}, -d_{I}\sum _{i\in \tilde{\Omega }}J_{2}(i)-\gamma, c, \lambda )=0$ has a unique positive real root $\lambda ^\ast$ ;
-
∘ if $c\lt c^\ast$ , $\Psi (d_{I}, \beta S_{0}, -d_{I}\sum _{i\in \tilde{\Omega }}J_{2}(i)-\gamma, c, \lambda )=0$ has no positive real root.
Let $ \varphi (\xi )$ be a positive solution of (8) with $c\gt 0$ , it is clear that $\phi (\xi )\,:\!=\,{\varphi ^{\prime}(\xi )}/{\varphi (\xi )}$ satisfies the equation
Boundedness and smoothness for solutions of (10) are established in the following lemma.
Lemma 2.1. If $ \phi ({\cdot})$ is a solution of equation (10), then $ \phi ({\cdot}) \in L^{\infty } (\mathbb{R})\cap C^{\infty } (\mathbb{R})$ .
Proof. Let ’s denote
Then, $ u(\xi )$ satisfies the equation
which implies that $u(\xi )$ is strictly increasing on $\mathbb{R}$ . Thus, for any $p\in \tilde{\Omega }$ with $ p\gt 0$ , we have
Integrating the inequality (12) from $ \xi -\displaystyle \frac{p}{2}$ to $ \xi$ gives
which implies
and hence
Since $u(\xi )$ is strictly increasing on $\mathbb{R}$ , it follows from (11) and (13) that ${u^{\prime}(\xi )}/{u(\xi )}\leq M_{0},$ for some $M_{0}\gt 0.$ Note that $\phi (\xi )={u^{\prime}(\xi )}/{u(\xi )}-v_{1}$ . Thus, $\phi (\xi )$ is uniformly bounded. Moreover, by (10), it is easy to see $ \phi ({\cdot}) \in C^{\infty } (\mathbb{R})$ . The proof is complete.
In addition, we prove that any non-constant solution of (10) has no global extrema.
Lemma 2.2. Let $\phi (\xi )$ be a solution of (10) that attains its global maxima or minima, then it must be a constant function.
Proof. The proof of the case $b_{0}=0$ is similar to that of [Reference Guo and Lin12, Lemma 2.8], so we only consider the case $b_{0}\gt 0$ . Differentiating equation (10) gives
Suppose that $ \phi (\xi )$ admits a global maxima at $\xi _*$ , then (14) gives
which implies
By induction arguments, we can conclude that $ \phi (\xi )=\phi (\xi _{\ast })$ for all $ \xi \in \mathbb{R}$ . Similarly, $\phi (\xi )$ is a constant function provided that it admits a global minima. The proof is complete.
Next, we investigate the asymptotic behaviour of solutions for the equation (10).
Lemma 2.3. Assume that $\phi (\xi )$ is a solution of (10). Then we have the following statements.
-
(1) $\phi (\pm \infty )\,:\!=\,\lim \limits _{\xi \rightarrow \pm \infty }\phi (\xi )$ exist and $ \Psi (d, b_{0},b,c, \phi (\pm \infty ))=0$ .
-
(2) If $\phi (\xi )$ is a non-constant solution, then
\begin{align*} \phi (\xi )-\phi ({-}\infty )\in C^{1}(\mathbb{R},(0,\infty ))\cap L^{1}(\mathbb{R}_{-})\ \text{ and } \phi ({+}\infty )-\phi (\xi )\in C^{1}(\mathbb{R},(0,\infty ))\cap L^{1}(\mathbb{R}_{+}). \end{align*}
Proof. (1) Let $\{y_{j}\}_{j=1}^\infty$ be a sequence such that
Denote $\eta _j(\xi )\,:\!=\, \phi ( y_{j}+\xi )$ for $j\in \mathbb{N}$ . Clearly, $\{ \eta _j( \xi )\} _{j=1}^\infty$ is uniformly bounded and equicontinuous. Then it follows from the Arzela-Ascoli theorem that $\{ \eta _j( \xi )\} _{j=1}^\infty$ has a subsequence, still written as $\{ \eta _j( \xi )\} _{j=1}^\infty$ , such that $\eta _j(\xi )\rightarrow \eta (\xi )$ in $ C^{1}_{loc} (\mathbb{R})$ as $j\rightarrow +\infty$ . Hence, $\eta (\xi )$ is a solution of (10) with $\eta (0)=w^{\ast } =\max _{\xi \in \mathbb{R}} \eta (\xi )$ .
By Lemma 2.2, we have $ \eta (\xi )=w^{\ast }$ for all $ \xi \in \mathbb{R}$ , which implies
We first show that $\phi ({+}\infty )$ exists. Suppose $ w^{\ast } \gt w_{\ast }\,:\!=\,\liminf \limits _{\xi \rightarrow +\infty }\phi (\xi )$ , by (15), there exists a sufficiently large $ j$ such that
Let $ \hat{y}$ be the point such that $ \phi (\hat{y})= \min \limits _{\xi \in [y_{j}, y_{j+1}]}\phi (\xi )$ . According to (16), we have $ \hat{y}\in (y_{j}+2+c\tau, y_{j+1}-2-c\tau )$ . Then, for $ s\in [0,\tau ]$ , it holds that $\phi ^{\prime}(\hat{y})=0$ ,
Taking $\xi = \hat{y}$ in (14) yields
which leads to a contradiction. Hence $ w^{\ast }=w_{\ast }$ , i.e., the limit $\phi ({+}\infty )$ exists.
In the same way, we can obtain that $\phi ({-}\infty )$ exists. Letting $ \xi \rightarrow \pm \infty$ in (10), it follows that $ \phi (\pm \infty )$ satisfy the equation $ \Psi (d, b_{0},b,c, \phi (\pm \infty ))=0$ .
(2) Since $\phi (\xi )$ is a non-constant solution, according to Lemma 2.2, $\phi (\xi )$ cannot attain its global extrema. Thus, $ \phi ({-}\infty )\neq \phi ({+}\infty )$ and
We first claim that $\phi ({-}\infty )\lt \phi ({+}\infty )$ . If the claim is false, i.e., $\phi ({-}\infty )\gt \phi ({+}\infty )$ , one has $ \phi ({+}\infty )={ \lambda }_{1}$ . Thus, given any small $ \epsilon \gt 0$ , by translation if necessary, we may assume that
According to (10) and (17), we have
which implies
However, the inequality (18) contradicts to the fact that $ \Psi (d, b_{0},b,c,{ \lambda }_{1}+\epsilon )\lt 0$ for every small $ \epsilon \gt 0$ with $\lambda _1+\epsilon \lt \lambda _2$ . Therefore, ${ \lambda }_{1}=\phi ({-}\infty )\lt \phi ({+}\infty )={ \lambda }_{2}$ .
Next we show that $\displaystyle \int _{-\infty }^{0}(\phi (\xi )-{ \lambda }_{1})d\xi \lt +\infty$ . Combining the equations $ \Psi (d, b_{0},b,c,{ \lambda }_{1})=0$ and (10), we have
where $ B_{1}(\xi, i)\in [{-}i{ \lambda }_{1}, -i\max _{y\in [\xi, \xi -i]}{\phi (y)}] \mbox{ for } i\in \{-2,-1\}$ ;
Let’s denote $ R(\xi )\,:\!=\,\phi (\xi )-\lambda _{1}$ . Integrating (19) from $M$ to $ 0$ with $ M\lt -2-c\tau$ gives
Changing the integration order in (20) yields
Let ${\varepsilon }\gt 0$ be small enough, by translation if necessary, we may assume that $\phi (\xi )\lt{ \lambda }_{1}+{\varepsilon }$ for $ \xi \lt \theta$ and ${ \lambda }_{1}+{\varepsilon }\leq \phi (\xi )$ for $ \xi \geq \theta$ , where $\theta =2+c\tau$ . Note that there exists some constant $K\gt 0$ such that $ |R(\xi )|\leq K$ for all $ \xi \in \mathbb{R}$ . Then we can obtain
It is also easy to verify that
Then it follows from (21) that
where $ \mathcal{O}(1)$ is uniformly bounded and
Note that
for small $\varepsilon \gt 0$ . Let $M\rightarrow -\infty$ in (23), we have
Thus, $\displaystyle \int _{-\infty }^{0}R(y)dy\lt +\infty$ , that is $\displaystyle \int _{-\infty }^{0}(\phi (\xi )-\phi ({-}\infty ))d\xi \lt +\infty$ . By the same way, we can obtain that $ \displaystyle \int _{0}^{+\infty }(\phi ({+}\infty )-\phi (\xi ))d\xi \lt +\infty$ . The proof is complete.
Based on Lemma 2.3, we can represent solutions of (10) explicitly in the following lemma.
Lemma 2.4. Assume that $\phi (\xi )$ is a solution of (10), then $\phi (\xi )$ takes the form
with some $l\in [0,1]$ , where $\lambda _1,\lambda _2$ are two real roots of $ \Psi (d, b_{0},b,c, \cdot )=0$ . Specially, when $l\ne 0$ or $1$ , then $\phi (\xi )$ is a non-constant solution of (10) which is strictly increasing on $\mathbb{R}$ .
Proof. If $\phi (\xi )$ is constant solution of (10), by Lemma 2.3, we have $\phi (\xi )={ \lambda }_{1}$ or $\phi (\xi )={ \lambda }_{2}$ for $\xi \in \mathbb{R}$ , i.e., (24) holds with $l=1$ or $0$ respectively. Note that $\lambda _1$ may equal to $\lambda _2$ .
If $\phi (\xi )$ is a non-constant solution of (10), according to Lemma 2.3, we have ${ \lambda }_{1}=\phi ({-}\infty )\lt \phi (\xi )\lt \phi ({+}\infty ){ =\lambda }_{2}$ . Then we consider the functions
It’s easy to verify that
Note that $\displaystyle \int ^{\xi }_{0}(\phi (z)-{ \lambda }_{1})dz\gt -\int ^{0}_{-\infty }(\phi (z)-{ \lambda }_{1})dz$ for any $\xi \in \mathbb{R}$ . We have
Based on (26), we further consider the function $\varphi (\xi )\,:\!=\,w^{\prime}_{2}(\xi )/w_{2}(\xi )$ . By simple computations, $\varphi (\xi )$ satisfies the equation (10). We claim that $\varphi (\xi )$ is a constant solution of (10). If false, that is $\varphi (\xi )$ is not a constant function, by Lemma 2.3, we have
As $\xi \rightarrow -\infty$ , it follows that
which contradicts to (26). Hence, by Lemma 2.3, $\varphi (\xi )$ is a constant function which equals to ${\lambda }_{1}$ or $\lambda _{2}$ for all $\xi \in \mathbb{R}$ . If $\varphi (\xi )={ \lambda }_{1}$ , we have $w(\xi )=ae^{{ \lambda }_{1}\xi }$ for some constant $a$ . Then $\phi (\xi )$ is constant function, which gives a contradiction. Therefore $\varphi (\xi )={ \lambda }_{2}$ . In view of the definition of $\varphi (\xi )$ , we have
Since $w^{\prime}(\xi )=\phi (\xi )w(\xi )$ , the solution form (24) holds obviously. According to (24), it is easy to verify that $\phi (\xi )$ is strictly increasing on $\mathbb{R}$ . The proof is complete.
Remark 1. From the proof of Lemma 2.4 , it can be easily seen that the conclusion of Lemma 2.3 (1) also holds for more general form of (10) by replacing the constant $b$ as any continuous function $b(\xi )$ whose limits $b_{\pm }\,:\!=\,b(\pm \infty )$ exist. That is, if $\phi (\xi )$ is a solution of (10) with $b$ replaced by such continuous function $b(\xi )$ , then $\phi (\pm \infty )$ exist and $ \Psi (d, b_{0},b_{\pm },c, \phi (\pm \infty ))=0$ .
By Lemma 2.4, we have the following results on solutions of the linearized equation (8).
Proposition 2.5. Suppose that $\varphi (\xi )$ is a nonnegative solution of the linear equation (8), then
where ${ \lambda }_{1},{ \lambda }_{2}$ are two real roots of the characteristic equation $ \Psi (d, b_{0},b,c, \cdot )=0$ .
Proof. The result can be proved in the following two cases.
Case 1: $\varphi (\xi _{0})=0$ for some $\xi _{0}\in \mathbb{R}$ . In this case, it can be easily deduced from (8) that $\varphi (\xi _{0}\pm 1)=0$ . An induction argument shows that $\varphi (\xi _{0}\pm k)=0$ for any $ k\in \mathbb{N_{+}}$ . On the other hand, by (8), one can see that
which implies that $ \varphi (\xi )e^{\frac{|b|}{c}\xi }$ is non-decreasing on $\mathbb{R}$ . Thus, it follows that $\varphi (\xi )=0$ on $\mathbb{R}$ .
Case 2: $\varphi (\xi )\gt 0$ for all $\xi \in \mathbb{R}$ . Dividing the equation (8) by $\varphi (\xi )$ , we have
where $q(\xi )\,:\!=\,{\varphi ^{\prime}(\xi )}/{\varphi (\xi )}$ . By Lemma 2.4, $q(\xi )$ admits the form
Integrating the above equality gives
for some $p\gt 0.$ Hence, (27) holds by letting $C_{1}=p{ l}$ and $C_{2}=p(1-{ l})$ .
3. Priori estimates on wave profiles and wave speeds
To establish some priori estimates on positive travelling waves and wave speeds, we first provide the necessary condition for the existence of positive travelling waves that satisfy (6).
Lemma 3.1. If $(S(\xi ),I(\xi ))$ is a positive solution of (5) satisfying (6), then $R_{0}\gt 1$ and $c\gt 0$ .
Proof. Suppose the assertion is false, that is (5) admits a positive solution $ (S(\xi ),I(\xi ))$ satisfying (6) for $R_{0}\leq 1$ and some $c\in \mathbb{R}$ . We first claim that $ \displaystyle \int _{\mathbb{R}} I (\xi )d\xi \lt +\infty$ . Since $I(\pm \infty )=0$ , there exists $K\gt 0$ such that $I(\xi )\leq K$ for $\xi \in \mathbb{R}$ . Integrating the first equation of (5) from $y$ to $ x$ for any $ x,y\in \mathbb{R}$ , we have
Then, it follows from the second equation of (5) that
Thus, it follows that $ \displaystyle \int _{\mathbb{R}} I (\xi )d\xi \lt +\infty$ by the arbitrariness of $ x$ and $ y$ .
By the second equation of (5), we have
Note that $R_{0}={\beta S_{0} /}{\gamma }.$ If $ R_{0}\lt 1$ , by (29), we have
which gives a contradiction. If $ R_{0}= 1$ , the second equation of (5) also gives
Since $ 0\lt S(\xi )\leq S_{0}$ and $ I(\xi )\gt 0$ on $ \mathbb{R}$ , (31) implies that $ S(\xi )\equiv S_{0}$ on $ \mathbb{R}$ , which contradicts to $ S({+}\infty )=S_{\infty }\lt S_{0}$ . Therefore, $R_0\gt 1$ .
Next, we show that $c\gt 0$ . Suppose, by contradiction, that $(S(\xi ),I(\xi ))$ is a positive solution of (5) satisfying (6) with $c\leq 0$ . Since $R_{0}\gt 1$ and $ S({-}\infty )=S_{0}$ , there exists a $\hat{\xi }\lt 0$ such that
Then, for $\xi \lt \hat{\xi }$ , we have
Integrating the above inequality over $({-}\infty, \xi )$ with $\xi \lt \hat{\xi }$ , one can obtain
where $ \Theta (\xi ) \,:\!=\,\displaystyle \int ^{\xi }_{-\infty }I(y)dy\gt 0$ . Since
it follows that
which also gives a contradiction. Hence the assertion of the lemma holds.
In addition, we have the following limiting results of the wave profiles.
Lemma 3.2. Assume that $( S(\xi ), I(\xi ))$ is a solution of (5) with $ c\gt 0$ , and $\{\xi _j\}_{j\in \mathbb{N}}$ is a sequence satisfying $\lim \limits _{j\rightarrow +\infty }I(\xi _{j})= +\infty$ , then $\lim \limits _{j\rightarrow +\infty }S(\xi _{j})= 0$ .
Proof. Let’s prove the result by using the contradiction argument. Suppose, for some constant $ \epsilon \gt 0$ , there exists a subsequence of $\{ \xi _{j}\} _{j\in \mathbb{N}}$ , still written as $\{ \xi _{j}\}_{j\in \mathbb{N}}$ , such that $ S(\xi _{j})\geq \epsilon$ for each $ j\in \mathbb{N}$ . Let $d_{1}=d_{S}\sum _{i\in \tilde{\Omega }}J_{1}(i)$ and $d_{2}=d_{I}\sum _{i\in \tilde{\Omega }}J_{2}(i)$ . By the first equation of (5), we have $ S^{\prime}(\xi )\leq{d_{1}S_{0}}/{c}$ which implies $ S(\xi )\geq{\epsilon }/{2}$ for $ \xi \in [\xi _{j}- \xi _{0}, \xi _{j}]$ , where $ \xi _{0}\,:\!=\,{c\epsilon }/{(2d_{1}S_{0})}.$ Note that
So, $I (\xi )e^{\frac{d_{2}+\gamma }{ c}\xi }$ is strictly increasing in $\mathbb{R}$ . Then,
Moreover, for any $p\in \tilde{\Omega }$ with $ p\gt 0$ , we have
which implies
Since $I (\xi )e^{\frac{d_{2}+\gamma }{ c}\xi }$ is strictly increasing, integrating (33) from $ \xi -\displaystyle \frac{p}{2}$ to $ \xi$ gives
It follows that
Dividing the second equation of (5) by $I(\xi )$ , and using (32) and (34), we have
Thus, there exists a constant $\rho \gt 0$ such that $ \left |{I^{\prime}(\xi )}/{I(\xi )} \right | \leq \rho$ , and hence
Then it follows that
Furthermore, by the first equation of (5), one can see
as $j\rightarrow +\infty .$ Let $j$ be sufficiently large such that $ S^{\prime}(\xi )\leq{-2S_{0}}/{ \xi _{0}}$ for $ \xi \in [\xi _{j}- \xi _{0}, \xi _{j}]$ , then
which is impossible since $ 0\lt S(\xi )\leq S_{0}$ on $\mathbb{R}$ . Therefore, $S(\xi _{j})\to 0$ as $j\rightarrow +\infty$ .
We further establish the following limiting lemma, which is crucial for proving the existence of travelling wave solutions with minimal speed.
Lemma 3.3. Assume $R_{0}\gt 1$ . Let $(S_{k}(\xi ), I_{k}(\xi ))$ $(k\in \mathbb{N})$ be the positive solutions of (5) satisfying (6) with $ c=c_{k}$ such that $\lim \limits _{k\rightarrow +\infty }c_{k}=c_{0}\gt 0$ . Then (5) admits a positive solution $(S(\xi ), I(\xi ))$ satisfying (6) with $c= c_{0}$ .
Proof. Without loss of generality, we assume $\left \lbrace c_{k}\right \rbrace _{k=1}^{+\infty }$ is a strictly decreasing sequence with $ \lim _{k\rightarrow +\infty } c_{k}= c_0\gt 0$ . Let $ (S_{k}(\xi ),I_{k}(\xi ))$ be a positive solution of (5) satisfying (6) for $c= c_{k}$ . We first claim that the sequence $\{I_{k}(\xi )\} _{k\in \mathbb{N}}$ is uniformly bounded on $\mathbb{R}$ . If not, there exists a sequence $\{ z_{k}\}_{k\in \mathbb{N}}$ such that $\lim \limits _{k\rightarrow +\infty } I_{k}(z_{k})=+\infty$ . Then, it follows from Lemma 3.2 that $\lim \limits _{k\rightarrow +\infty } S_{k}(z_{k})=0$ . Since $ I_{k} (\pm \infty )=0$ and $ I_{k}(\xi )\gt 0$ , without loss of generality, we may assume that $ I_{k}(z_{k})=\max \limits _{\xi \in \mathbb{R}} I_{k}(\xi )$ . By the second equation of (5), we have
However, the inequality (36) contradicts to the properties $\lim \limits _{k\rightarrow +\infty } I_{k}(z_{k})=+\infty$ and $\lim \limits _{k\rightarrow +\infty } S_{k}(z_{k})=0$ , when $k$ is large enough. Therefore, $\{I_{k}(\xi )\} _{k\in \mathbb{N}}$ is uniformly bounded on $\mathbb{R}$ .
Since $\{S_{k}(\xi )\} _{k\in \mathbb{N}}$ and $\{ I_{k}(\xi )\}_{k\in \mathbb{N}}$ are uniformly bounded on $\mathbb{R}$ , it follows from (5) that $||S_{k}||_{C^{2}(\mathbb{R})}$ and $||I_{k}||_{C^{2}(\mathbb{R})}$ are both uniformly bounded on $\mathbb{R}$ . Then, by Arzela-Ascoli theorem, there exists subsequences of $\{S_{k}(\xi )\}_{k\in \mathbb{N}}$ and $\{ I_{k}(\xi )\}_{k\in \mathbb{N}}$ , still written as $\{S_{k}(\xi )\}_{k\in \mathbb{N}}$ and $\{ I_{k}(\xi )\}_{k\in \mathbb{N}}$ , such that $ S_{k} (\xi ) \rightarrow S(\xi )$ and $ I_{k} (\xi ) \rightarrow I (\xi )$ in $ C^{1}_{loc} (\mathbb{R})$ as $k\rightarrow +\infty$ . By Lebesgue dominated convergence theorem, it is easy to see the limiting function $ ( S(\xi ), I(\xi ) )$ is a solution of (5) with $c= c_{0}$ , i.e., satisfies the following system
Since $\{I_{k}\} _{k\in \mathbb{N}}$ is uniformly bounded on $\mathbb{R}$ , there exists a constant $I_{0}\gt 0$ such that $ I(\xi )\leq I_{0}$ on $\mathbb{R}$ . Hence, the solution $ ( S(\xi ), I(\xi ) )$ satisfies $ 0\leq S(\xi )\leq S_{0}$ and $ 0\leq I(\xi )\leq I_{0}$ for all $\xi \in \mathbb{R}.$
Next, we claim that $ S(\xi )$ and $I(\xi )$ are non-trivial, and they satisfy the asymptotic boundary conditions of (6).
Claim 1: $I(\pm \infty )=0.$
Integrating the first equation of (37) from $y$ to $ x$ for any $ x,y\in \mathbb{R}$ , we have
Thus, by the second equation of (37), we can obtain
Due to the arbitrariness of $ x$ and $ y$ , it follows from (39) that $ \displaystyle \int _{\mathbb{R}} I (\xi )d\xi \lt +\infty$ . In addition, it is easy to see from (37) that $ I^{\prime} (\xi )$ is bounded on $\mathbb{R}$ . Therefore, we have $ I (\pm \infty )=0$ .
Claim 2: $ S(\xi ) \gt 0$ on $\mathbb{R}$ and $S({-}\infty )=S_0$ .
Assume that $S({\xi }_0) =0$ for some ${\xi }_0\in \mathbb{R}$ . From the first equation of (37), we have $S({\xi }_0\pm 1) =0$ , and inductively that $S({\xi _0}\pm k) =0$ for any $k\in \mathbb{N}$ . Moreover, the first equation of (37) gives
where $\delta =(d_{S}\sum _{i\in \tilde{\Omega }}J_{1}(i)+\beta I_{0})$ , which implies that $ S (\xi )e^{\frac{\delta }{ c_{0}}\xi }$ is non-decreasing over $\mathbb{R}$ . Then it follows that $ S(\xi ) \equiv 0$ on $ \mathbb{R}$ , and hence, $ S_{k}(\xi ) \rightarrow 0$ as $k\rightarrow +\infty$ for any $ \xi \in \mathbb{R}$ . On the other hand, let’s write
then the first equation of (5) gives
As $k\rightarrow +\infty$ , it follows from (40) that $ c_{0}S_{0}=0$ , which leads to a contradiction. Hence $ S(\xi ) \gt 0$ on $\mathbb{R}$ .
To prove $ S({-}\infty )=S_{0}$ , it is sufficient to show that $\underline{S}\,:\!=\, \liminf \limits _{\xi \rightarrow -\infty } S(\xi )= S_{0}$ . Suppose that $ \underline{S}\lt S_{0}$ , then there exists a sequence $\{ \zeta _{q}\}_{q\in \mathbb{N}}$ such
Denote
Obviously, $ \lim \limits _{q\rightarrow +\infty } \tilde{I}_{q}(\xi )=0$ locally uniformly on $ \mathbb{R}$ . Since $ ||\tilde{S}_{q} ||_{C^{2}(\mathbb{R})}$ is uniformly bounded on $ \mathbb{R}$ , there exists a subsequence of $\{\tilde{S}_{q}(\xi )\} _{q\in \mathbb{N}}$ , still denote as $\{\tilde{S}_{q}(\xi )\} _{q\in \mathbb{N}}$ , such that $\tilde{S}_{q}(\xi ) \rightarrow \tilde{S} (\xi )$ in $ C^{1}_{loc} (\mathbb{R})$ as $q\rightarrow +\infty$ . Hence, the first equation of (37) gives
It is easy to note that zero is the root of the characteristic equation of (41). Since $\tilde{S}(\xi )$ is nonnegative and bounded with $ \tilde{S}(0)=\underline{S}$ , it follows that $ \tilde{S}(\xi )=\underline{S}$ on $\mathbb{R}$ according to Proposition 2.5. Hence, we get that $ \tilde{ S}_{q}(\xi )\to \underline{S}$ in $ C^{1}_{loc} (\mathbb{R})$ as $q\rightarrow +\infty$ . Since
we have
Note that
As $k\rightarrow +\infty$ , one can see
As $q\rightarrow +\infty$ , it follows that $ c_{0}(\underline{S} -S_{0})= 0$ , which contradicts to $ \underline{S} \lt S_{0}$ . Hence, $ S({-}\infty )=S_{0}$ .
Claim 3: $ I(\xi )\gt 0$ on $\mathbb{R}$ .
Assume that $I({\eta })=0$ for some ${\eta }\in \mathbb{R}$ , then it is easy to see from the second equation of (37) that $ I(\xi )\equiv 0$ on $\mathbb{R}$ . Then, the first equation of (37) gives
Since $S(\xi )$ is nonnegative and bounded, by Proposition 2.5, it follows that $S(\xi )$ is a constant function, i.e., $ S(\xi )\equiv \widehat{S}$ for some constant $ \widehat{S}\in [0,S_{0}]$ . Note that
By the second equation of (5), we have
Note that $ \displaystyle \int _{0}^{\tau }f(s)I_{k}(\xi -c_{k}s)ds\gt 0$ for any $\xi \in \mathbb{R}$ . There exists some ${\eta }_{k}$ such that $\gamma = \beta S_{k}({\eta }_{k}) .$ Since $ (S_{k}(\xi ), I_{k}(\xi ))$ is translation invariant, we may assume ${\eta }_{k}=0$ . As $k\to +\infty$ , we have
which implies $\widehat{S}=\gamma /\beta$ . Since $R_{0}={\beta S_{0} }/{\gamma }\gt 1$ , then $ S (\xi )\equiv \widehat{S}\lt S_{0},$ which contradicts to $ S({-}\infty )=S_{0}$ . So, $ I(\xi )\gt 0$ on $\mathbb{R}$ .
Claim 4: $ S({+}\infty )=S_{\infty }\lt S_{0}$ .
We first show that $ \bar \alpha \,:\!=\,\liminf \limits _{\xi \rightarrow +\infty } S (\xi )= \limsup \limits _{\xi \rightarrow +\infty } S (\xi )=:\hat \alpha$ . If $\bar{\alpha }\lt \hat \alpha$ , one can find two sequences $\{ \xi _{k}\}_{k\in \mathbb{N}}$ and $\{ \eta _{k}\}_{k\in \mathbb{N}}$ , with $\xi _{k}\lt \eta _{k}$ , $\lim \limits _{k\rightarrow +\infty } \xi _{k}=+\infty$ and $\lim \limits _{k\rightarrow +\infty } \eta _{k}=+\infty$ , such that
Denote
Obviously, $ \hat{I}_{k}(\xi )\to 0$ locally uniformly on $ \mathbb{R}$ as $k\rightarrow +\infty .$ Since $ ||\hat{S}_{k} ||_{C^{2}(\mathbb{R})}$ is uniformly bounded on $ \mathbb{R}$ , there exists a subsequence of $\{ \hat{S}_k(\xi )\}_{k\in \mathbb{N}}$ , still written as $\{ \hat{S}_k(\xi )\}_{k\in \mathbb{N}}$ , such that $\hat{S}_{k}(\xi ) \rightarrow \hat{S} (\xi )$ in $ C^{1}_{loc} (\mathbb{R})$ as $k\rightarrow +\infty$ . Hence, the first equation of (37) yields
By Proposition 2.5, it follows that $ \hat{S}(\xi )$ is constant function. Since $ \hat{S}(0)=\bar{\alpha }$ , we have $ \hat{S}(\xi )\equiv \bar{\alpha }$ on $\mathbb{R}$ . Hence, we get that $ S(\xi +\xi _{k})\rightarrow \bar{\alpha }$ in $ C^{1}_{loc} (\mathbb{R})$ as $ k\rightarrow +\infty$ . By the same way, we also obtain that $ \lim _{k\rightarrow +\infty } S(\xi +\eta _{k})=\hat{\alpha }$ in $ C^{1}_{loc} (\mathbb{R})$ . Then, integrating the first equation of (37) from $ \xi _{k}$ to $ \eta _{k}$ , we have
which contradicts to $\bar{\alpha }\lt \hat \alpha$ . Thus, $\hat{\alpha }=\bar{\alpha }$ , and then $S_{\infty }\,:\!=\, S ({+}\infty )$ exists.
Next, we further show that $S_{\infty }\lt S_{0}$ by proving $ \liminf \limits _{\xi \rightarrow +\infty } S (\xi )\lt S_{0} .$ Suppose, by contradiction, that $ \liminf \limits _{\xi \rightarrow +\infty } S (\xi )= S_{0}$ , then $ S ({+}\infty ) =S_{\infty }=S_{0}$ and the first equation of (37) gives
which gives a contradiction. Thus, $ S_{\infty }\lt S_{0}$ . The proof is complete.
4. Proof of the main results
To prove the existence result of travelling wave solutions by using the truncation method, we first establish the existence of solutions for the profile system (6) over large finite domains.
Lemma 4.1. Assume $R_0\gt 1$ . For any $c\gt c^{\ast }$ and large $\mathcal{X}\gt 0$ , the bounded domain problem
has a solution $(S(\xi ), I(\xi ))$ satisfying $ 0\leq S_{-}(\xi )\leq S(\xi )\leq S_{0}$ and $0\leq I_{-}(\xi )\leq I(\xi )\leq I_{+}(\xi )$ with
when $\nu \in (0,\lambda _{1})$ , $\varsigma \in \left ( 0, \min \{ \nu, \lambda _{2}-\lambda _{1} \}\right )$ are small enough; and $\delta \gt S_{0}$ , $K\gt 1$ are sufficiently large.
Proof. The idea of proof is similar to that of [Reference Xu, Tan and Hsu28, Proposition 3.1]. However, due to the consideration of distributed latent period, the computations are more complicated than those of [Reference Xu, Tan and Hsu28].
Firstly, for large $Y\gt 0$ , we define the set
Obviously, $ \Phi _{Y}$ is a closed and convex set. In addition, we extend any function $ (\tilde{\Omega }({\cdot}),\sigma ({\cdot}))$ to $ ( \tilde{\omega }({\cdot}), \tilde{\sigma } ({\cdot}))\in C(\mathbb{R},\mathbb{R}^{2})$ in the way
For the convenience of statement, without loss of generality, we next assume that $\sum _{i\in \tilde{\Omega }}J_{k}(i)=1$ ( $k=1,2$ ). Define an operator $\Gamma$ on $ \Phi _{Y}$ by
where $(S(\xi ),I(\xi ))$ satisfies the following initial value problem of ODE:
where $l$ is a constant satisfying $l\geq \beta e^{\lambda _{1}Y}\displaystyle \int _{0}^{\tau }f(s)e^{-c\lambda _{1}s}ds$ . Then we claim that $\Gamma$ is completely continuous which maps from $ \Phi _{Y}$ to $ \Phi _{Y}$ .
We first show that $ \Gamma [\Phi _{Y}]\subseteq \Phi _{Y}$ , i.e.,
It is easy to see that $0$ is a lower solution of (48). By comparison principle, we have
Since $0\leq \tilde{\omega }(\xi )\leq S_{0}$ on $ \mathbb{R}$ , for $ \xi \in [{-}Y,Y]$ , one can obtain
which implies that $ S_{0}$ is an upper solution of (48). Then, by comparison principle again,
Hence, the left part of (51) holds.
Now we prove the right part of (51). According to the choice of $l$ , it is easy to see that
is non-decreasing in $\omega$ . Then, for $ \xi \in [{-}Y,\displaystyle \frac{1}{\nu }\ln \frac{S_{0}}{\delta })$ , one can see $S_{-}(\xi )= S_{0}-\delta e^{\nu \xi }$ and
Since $I_{+}(\xi )=e^{\lambda _{1}\xi }$ and $S_{-}(\xi )\geq S_{0}-\delta e^{\nu \xi }$ for $\xi \in \mathbb{R}$ , direct calculation gives
for small $\nu$ and large $\delta$ . Then it follows from (52) that $ S_{-}(\xi )$ is a lower solution of (48). By comparison principle, one can obtain
On the other hand, when $ \xi \in [\displaystyle \frac{1}{\nu }\ln \frac{S_{0}}{\delta },Y],$ we have $S_{-}(\xi )=0$ . Thus,
Similarly, we can obtain
Hence, the inequalities of (51) hold.
Next, we show that $ \Gamma [\cdot, \cdot ]$ is continuous on $ \Phi _{Y}$ . Assume that $ (\omega _{i}(\xi ),\sigma _{i}(\xi ))\in \Phi _{Y}(i=1,2)$ and $ \Gamma _{2}[\omega _{i}, \sigma _{i}](\xi )=I_{i}(\xi ) (i=1,2)$ for $ \xi \in [{-}Y,Y]$ , one can verify that
where
Thus,
Note that for $ x\in [{-}Y,Y]$ , one has
and
Then, for $ x\in [{-}Y,Y]$ , it follows that
Combine (53) and (54), we have
where
Thus, $\Gamma _{2}$ is continuous on $ \Phi _{Y}$ . Similarly, $\Gamma _{1}$ is continuous on $ \Phi _{Y}$ . Furthermore, in view of (48) and (49), $ S^{\prime}(\xi )$ and $ I^{\prime}(\xi )$ are bounded on $ [{-}Y,Y]$ . Then, it can be deduced by the Arzela-Ascoli theorem that $\Gamma$ is compact. So, $\Gamma$ is completely continuous which maps from $ \Phi _{Y}$ to $ \Phi _{Y}$ .
Finally, by Schauder’s fixed point theorem, one can see that there exists a fixed point $ (S(\xi ),I(\xi ))\in \Phi _{Y}$ such that
Clearly, $(S(\xi ),I(\xi ))$ satisfies (47) with $\mathcal{X}=Y-2-c\tau$ ,
The proof is complete.
Based on the previous lemmas, we are ready to prove the main results.
Proof of Theorem1.1
(1) Let $ \{\mathcal{X}_{m}\} _{m\in \mathbb{N}}$ be an increasing sequence satisfying $\mathcal{X}_{m}\to +\infty$ as $m\rightarrow +\infty$ . According to Lemma 4.1, when $m$ is large enough (say $m\ge m_0\gg 1$ ), we denote $ (S_{m}(\xi ), I_{m}(\xi ))$ as the solution of (47) over $[{-}\mathcal{X}_{m},\mathcal{X}_{m}]$ . Note that $ \{ S(\xi )\} _{m\geq m_{0}}$ and $ \{ I_{m}(\xi )\} _{m\geq m_{0}}$ are uniformly bounded on $[{-}\mathcal{X}_{m_0},\mathcal{X}_{m_0}]$ , it follows from (47) that $\{S^{\prime}_{m}(\xi )\}_{m\geq m_{0}}$ and $\{ I^{\prime}_{m}(\xi )\}_{m\geq m_{0}}$ are uniformly bounded on $[{-}\mathcal{X}_{m_0}+a,\mathcal{X}_{m_0}-a]$ , where $a=2+c\tau$ . Then, for any $ \xi _{1},\eta _{1}\in [{-}\mathcal{X}_{m_0}+2a,\mathcal{X}_{m_0}-2a]$ , we have
which implies that $\{ S^{\prime}_{m}(\xi )\} _{m\geq m_{0}}$ and $\{ I^{\prime}_{m}(\xi )\} _{m\geq m_{0}}$ are equicontinuous on $[{-}\mathcal{X}_{m_0}+2a,\mathcal{X}_{m_{0}}-2a]$ . Moreover, for any compact set $ \Lambda$ of $\mathbb{R}$ , there is some $q_{0}\in \mathbb{N_{+}}$ such that $ \Lambda \subset [{-}\mathcal{X}_{m}+2a,\mathcal{X}_{m}-2a]$ for any $m\geq q_{0}$ . Then, it follows from the Arzela-Ascoli theorem that there exists a subsequence $\{(S_{m_{k}}(\xi ), I_{m_{k}}(\xi ))\}_{m_k\ge m_0}$ of $\{(S_{m}(\xi ), I_{m}(\xi ))\}_{m\ge m_0}$ such that $ S_{m_{k}}(\xi )\rightarrow S(\xi )$ and $ I_{m_{k}}(\xi )\rightarrow I(\xi )$ in $ C^{1}_{loc} (\mathbb{R})$ as $ k \rightarrow +\infty$ . It is clear that $ (S(\xi ),I(\xi ))$ is a solution of (5) that satisfies
Furthermore, we claim that $S (\xi )\gt 0$ and $ I (\xi )\gt 0$ on $\mathbb{R}$ . Suppose that $ S(\overline{\eta })=0$ for some $ \overline{\eta }\in \mathbb{R}$ , then $ S^{\prime}(\overline{\eta })=0$ and the first equation of (5) implies that $S(\overline{\eta }\pm 1)=0$ . By induction argument, we have $S(\overline{\eta }\pm k) =0$ for any $ k\in \mathbb{N}$ . Let $ k$ be sufficiently large such that $ \overline{\eta } - k\lt \frac{1}{\nu }(\ln{S_{0}}-\ln{\delta })$ , one can see $ S(\overline{\eta }-k)\geq S_{-}(\overline{\eta }-k)\gt 0$ , which gives a contradiction. Hence, $S (\xi )\gt 0$ on $\mathbb{R}$ . Similarly, one can obtain $ I (\xi ) \gt 0$ on $\mathbb{R}$ .
Next, we show that the positive solution $(S(\xi ),I(\xi ))$ satisfies the condition (6). According to the facts $ S_{-}({-}\infty ) =S_{0}$ and $ S_{-}(\xi )\leq S(\xi )\leq S_{0}$ for $\xi \in \mathbb{R}$ , it is clear that $ S({-}\infty ) =S_{0}$ . To prove $ I (\pm \infty )=0$ and $S({+}\infty )\lt S_{0}$ , we first show that $I(\xi )$ is bounded on $\mathbb{R}$ . Suppose that $ \limsup \limits _{\xi \rightarrow +\infty } I(\xi ) =+\infty$ , if $\sigma \,:\!=\,\liminf \limits _{\xi \rightarrow +\infty } I(\xi ) \lt +\infty$ then there exists a sequence $\{s_{j}\}_{j\in \mathbb{N}}$ satisfying $ s_{j}\to +\infty$ such that $ I(s_{j})\to \sigma$ as $j\rightarrow +\infty$ . Without loss of generality, we may assume that $ I(s_{j})\lt \sigma +1$ for $ j\in \mathbb{N}$ . Given any $ j$ , one can find $ \xi _{j}\in [s_{j}, s_{j+1}]$ such that
Clearly, $ \lim \limits _{j\rightarrow +\infty }I(\xi _{j}) =+\infty$ . Then, by Lemma 3.2, we have $ \lim \limits _{j\rightarrow +\infty }S(\xi _{j}) =0$ . Without loss of generality, we may assume that $ I(\xi _{j})\geq (1+\sigma )e^{2C_{0}}$ for $ j\in \mathbb{N}$ , where $ C_{0}\,:\!=\, \sup _{\xi \in \mathbb{R}}\left |{I^{\prime}(\xi )}/{I(\xi )} \right |$ (cf. (35)). Then we have
Thus,
It then follows that $ [\xi _{j}-2, \xi _{j}+2]\subset [s_{j}, s_{j+1}]$ . In fact, if $\xi _{j}-2\lt s_{j}$ and (or) $\xi _{j}+2\gt s_{j+1}$ , then (55) contradicts to $ I(s_{j})\lt \sigma +1$ . Hence, $ [\xi _{j}-2, \xi _{j}+2]\subset [s_{j}, s_{j+1}]$ . Then, by the second equation of (5) and (32), we have
which is impossible since $ I(\xi _{j})\rightarrow +\infty$ and $ S(\xi _{j})\rightarrow 0$ as $ j\rightarrow +\infty$ . Thus, we have $I({+}\infty )=+\infty$ . It then follows from Lemma 3.2 that $ S({+}\infty ) =0$ .
On the other hand, dividing the second equation of (5) by $I(\xi )$ , we have
where $ \zeta (\xi ) \,:\!=\,{I^{\prime}(\xi )}/{I(\xi )}$ . Since $ S({+}\infty ) =0$ and
by Remark 1, it follows that the limit $\lambda _{3}\,:\!=\, \zeta ({+}\infty )$ exists, and it is a real root of the characteristic equation
Note that $\lambda _{3}=\zeta ({+}\infty ) \geq 0$ due to $I({+}\infty )=+\infty$ . Moreover, $ \Delta (\lambda )$ is convex with $ \Delta (0)=-\gamma \lt 0$ . Thus, it follows that $ \lambda _{3}$ is the unique positive real root of the characteristic equation (56). On other hand, it is clear that
where $\lambda _{k}$ ( $k=1,2$ ) is the positive real root of $\Psi (d_{I}, \beta S_{0},-d_{I}\sum _{i\in \tilde{\Omega }}J_{2}(i)-\gamma, c, \lambda )=0$ . It follows that $\lambda _{1} \lt \lambda _{2}\lt \lambda _{3}$ . Since $ \lim \limits _{\xi \rightarrow +\infty }{I^{\prime}(\xi )}/{I(\xi )}=\lambda _{3}\gt \frac{\lambda _{2}+\lambda _{3}}{2}$ , there is some sufficiently large constant $ X\gt 0$ and some constant $ C_{1}(X)\gt 0$ such that
which contradicts to the fact $ I(\xi ) \leq e^{\lambda _{1}\xi }$ on $ \mathbb{R}$ . Therefore, $ \limsup \limits _{\xi \rightarrow +\infty } I(\xi ) \lt \infty$ , i.e., $I(\xi )$ is bounded on $\mathbb{R}$ . By the boundedness of $I(\xi )$ , following the same proof procedure as Lemma 3.3, we can obtain that $ I (\pm \infty )=0$ and $S({+}\infty )\lt S_{0}$ .
Now we show the existence of travelling wave with speed $c=c^{*}$ . Let $\{c_{k}\}_{k\in \mathbb{N}}\subset (c^{*},2c^{*})$ be a strictly decreasing sequence satisfying $ c_{k}\to c^{*}$ as $k\to \infty$ . According to the above proof, we denote $ (S_{k}(\xi ),I_{k}(\xi ))$ as the solutions of (5) satisfying (6) with $c= c_{k}$ . Then, as a consequence of Lemma 3.3, we can obtain that (5) admits a solution $ (S(\xi ),I(\xi ))$ satisfying (6) with wave speed $c= c^{*}$ .
(2) By Lemma 3.1, we only need to show that (5) has no positive solution satisfying (6) for $c\in (0,c^{*})$ . Assume that (5) admits a positive solution $ (S(\xi ),I(\xi ))$ satisfying (6) with $c\in (0,c^{*})$ . Let $\{ \overline{\xi }_{n}\}_{n\in \mathbb{N}}$ be a sequence satisfying $ \overline{\xi }_{n}\to -\infty$ as $n\rightarrow +\infty$ , we consider the functions
Obviously, $ \lim \limits _{n\rightarrow +\infty } \overline{S}_{n}(\xi )=S_{0}$ locally uniformly on $ \mathbb{R}$ , and $ ( \overline{S}_{n}(\xi ), \overline{I}_{n}(\xi ))$ satisfies the equation
Denote $ E(\xi )\,:\!=\,{I^{\prime}(\xi )}/{I(\xi )}$ , which is bounded on $ \mathbb{R}$ (cf.(35)). Since
it follows that $ \overline{I}_{n}(\xi )$ is locally uniformly bounded on $ \mathbb{R}$ . Further, by (57), $ \overline{I}_{n}^{\prime}(\xi )$ and $ \overline{I}_{n}^{\prime\prime}(\xi )$ are locally uniformly bounded on $ \mathbb{R}$ . Thus, there exists a subsequence of $\{ \overline{I}_{n}(\xi )\}_{n\in \mathbb{N}}$ , still written as $\{ \overline{I}_{n}(\xi )\} _{n\in \mathbb{N}}$ , such that $\overline{I}_{n}(\xi ) \rightarrow \overline{I} (\xi )$ in $ C^{1}_{loc} (\mathbb{R})$ as $n\rightarrow +\infty$ . It follows from (57) that $ \overline{I}(\xi )$ satisfies
It is clear that $ \overline{I}(0) =1$ and $ \overline{I}(\xi )\geq 0$ on $ \mathbb{R}$ . We claim that $ \overline{I}(\xi )\gt 0$ on $ \mathbb{R}$ . If the claim is false, then there exists some $ \overline{\xi }_{0}\in \mathbb{R}$ such that $ \overline{I}(\overline{\xi }_{0})=0$ and $ \overline{I}^{\prime}(\overline{\xi }_{0})=0$ . It can be further deduced from (58) that $ \overline{I}(\xi )\equiv 0$ on $\mathbb{R}$ , which is impossible since $ \overline{I}(0) =1$ . Thus, $ \overline{I}(\xi )\gt 0$ on $ \mathbb{R}$ .
Finally, we define $ \overline{E}(\xi )\,:\!=\,{\overline{I}^{\prime}(\xi )}/{\overline{I}(\xi )}$ for $ \xi \in \mathbb{R}$ . By (58), $ \overline{E}(\xi )$ satisfies
According to Lemma 2.3, the limits $\overline{E}(\pm \infty )$ exist, which are real roots of the characteristic equation $\Psi (d_{I}, \beta S_{0},-d_{I}\sum _{i\in \tilde{\Omega }}J_{2}(i)-\gamma, c, \lambda )=0$ . However, by the definition of $c^{\ast }$ , $\Psi (d_{I}, \beta S_{0},-d_{I}\sum _{i\in \tilde{\Omega }}J_{2}(i)-\gamma, c, \lambda )=0$ has no nonnegative real roots for $0\lt c\lt c^{\ast }$ . This gives a contradiction. The proof of Theorem1.1 is complete.
Acknowledgements
We are very grateful to the anonymous referees for their careful reading and helpful suggestions which led to an improvement of our original manuscript.
Funding statement
The first author was partially supported by the NNSF of China (No. 12071182), the Fundamental Research Funds for the Central Universities.
The third author was partially supported by the NSTC and NCTS of Taiwan (No. NSTC 112-2115-M-008-005-MY2).
Competing interests
No conflict of interest exists in the submission of this manuscript.