Received xx xxxxx xxxx; Revised xx xxxxx xxxx; Accepted xx xxxxx xxxx

DOI: xxx/xxxx

RESEARCH ARTICLE

New results on the asymptotic behavior of an SIS epidemiological

model with quarantine strategy, stochastic transmission, and Lévy
disturbance
Driss Kiouach*1 | Yassine Sabbar1 | Salim El Azami El-idrissi1

1 LPAIS Laboratory, Faculty of Sciences

Dhar El Mahraz, Sidi Mohamed Ben
Abdellah University, Fez, Morocco.
Correspondence
*Corresponding author.
E-mail addresses: d.kiouach@uiz.ac.ma (D.
Kiouach), yassine.sabbar@usmba.ac.ma (Y.
Sabbar),
salim.elazamielidrissi@usmba.ac.ma (S. El
Azami El-idrissi).
Summary
The spread of infectious diseases is a major challenge in our contemporary world,
especially after the recent outbreak of Coronavirus disease 2019 (COVID-19). The
quarantine strategy is one of the important intervention measures to control the
spread of an epidemic by greatly minimizing the likelihood of contact between
infected and susceptible individuals. In this study, we analyze the impact of various
stochastic disturbances on the epidemic dynamics during the quarantine period. For
this purpose, we present an SIQS epidemic model that incorporates the stochastic
transmission and the Lévy noise in order to simulate both small and massive per-
turbations. Under appropriate conditions, some interesting asymptotic properties are
proved, namely: ergodicity, persistence in the mean, and extinction of the disease.
The theoretical results show that the dynamics of the perturbed model are determined
by parameters that are closely related to the stochastic noises. Our work improves
many existing studies in the field of mathematical epidemiology and provides new
techniques to predict and analyze the dynamic behavior of epidemics.

KEYWORDS:
Epidemics, Quarantine, Dynamics, White noise, Lévy jumps, Asymptotic properties.

1 INTRODUCTION

The study of infectious diseases has long been a subject where epidemiological issues are combined with financial and social
problems 1,2,3,4,5 . The rapid spread of COVID-19 these days shows that humanity stills suffer from epidemics that may lead to
the collapse of medical and economic systems. By isolating infected individuals and quarantining the susceptible population at
home, many countries have basically controlled the outbreak of COVID-19 6,7 . In order to analyze the impact of this strategy
on the spread of epidemics and to predict their future behavior, we use different mathematical formulations according to their
characteristics 8,9,10 . In this study, we consider an SIQS epidemic model in the form of ordinary differential equations (ODEs for
short). These ODEs describe the evolution of susceptible 𝑆(𝑡), infected 𝐼(𝑡), and isolated 𝑄(𝑡) individuals as time functions. The
rates of change and the interactions between different population classes in our case are expressed by the following deterministic
2

model 11 :
( )
⎧d𝑆(𝑡) = 𝐴 − 𝜇1 𝑆(𝑡) − 𝛽𝑆(𝑡)𝐼(𝑡) + 𝛾𝐼(𝑡) + 𝑘𝑄(𝑡) d𝑡,
⎪ ( )
⎨d𝐼(𝑡) = 𝛽𝐼(𝑡)𝑆(𝑡) − (𝜇1 + 𝑟2 + 𝛿 + 𝛾)𝐼(𝑡) d𝑡, (1)
⎪ ( )
⎩d𝑄(𝑡) = 𝛿𝐼(𝑡) − (𝜇1 + 𝑟3 + 𝑘)𝑄(𝑡) d𝑡,
where the parameters appearing in this system are described as follows:

∙ 𝐴 is the recruitment rate of the susceptible individuals corresponding to new births.

∙ 𝜇1 is the natural death rate.

∙ 𝛿 is the isolation rate.

∙ 𝑟2 is the disease-related mortality rate.

∙ 𝑟3 is the death rate associated with the disease under isolation intervention. For simplicity, we denote 𝜇2 = 𝜇1 + 𝑟2 and
𝜇3 = 𝜇1 + 𝑟3 as a general mortality rates.

∙ 𝛾 and 𝑘 are the rates which individuals recover and return to 𝑆 from 𝐼 and 𝑄, respectively.

∙ 𝛽 represents the transmission rate.

All parameters are usually assumed to be positive. The schematic flow diagram of the model (1) is illustrated in Figure 1 .
Herbert et al. 11 proved that the basic reproduction number of the deterministic model (1) is expressed by 0 = 𝜇 (𝜇𝛽𝐴 . This
1 2 +𝛿+𝛾)
parameter is an essential quantity to predict whether a disease will persist or not. If 0 ≤ 1, the model (1) has only the disease-
free equilibrium 𝐸 ⊝ = (𝐴∕𝜇1 , 0, 0) which is globally asymptotically stable, and if 0 > 1, 𝐸 ⊝ becomes unstable and there
exists a global asymptotically stable endemic equilibrium 𝐸 ⊛ = (𝑆 ⊛ , 𝐼 ⊛ , 𝑄⊛ ), where
𝐴 𝐴(1 − 1∕0 ) 𝛿𝐼 ⊛
𝑆⊛ = , 𝐼⊛ = , and 𝑄⊛ = .
𝜇1 0 𝜇2 (1 + 𝛿∕(𝜇3 + 𝑘)) (𝜇3 + 𝑘)

kQ
γ

A
S I Q
μ1 μ2 μ 3Q

FIGURE 1 The transfer diagram for the deterministic SIQS epidemic model (1).

The spread of infectious diseases can undergo random disturbances and stochastic phenomena due to environmental fluc-
tuations 12,13,14 . Since disturbed models can describe many practical problems very well, many types of stochastic differential
equations have been used to analyze various epidemic models in recent years 15,16,17,18 . There are two common ways to intro-
duce stochastic factors into epidemic systems. The first one is to assume that the transmission of the diseases is subject to
some small random fluctuations which can be described by the Gaussian white noise 19,20,21 . The other one is to admit that
the model parameters are affected by massive environmental perturbations like the climate changes, earthquakes, hurricanes,
floods, etc 22,23 . For a better explaination to these phenomena, the use of a compensated Poisson process into the population
dynamics provides an appropriate and more realistic context 24 . Considering these two types of random disturbances, many
works have analyzed the asymptotic behaviors of various epidemic models, including persistence in the mean, extinction, and
ergodicity 25,22,23,26,27 . These interesting researches have served an important role in the stochastic modeling of epidemics.
But, all these models have considered either the standard white Gaussian noise or the Lévy jumps. In this work, we combine
 3

these two perturbations by treating an SIQS epidemic model that simultaneously includes the stochastic transmission and the
discontinuous Lévy process. This original idea extends the studies presented in 27,24 and gives us a general view of the disease
dynamics under different scenarios of random perturbations.

The threshold analysis of perturbed epidemic systems is very important for understanding and controlling of the disease
spread. In our case, the deterministic model (1) will be perturbed not only by white noise but also by Lévy jumps, which makes
its analysis more complicated and needs some new techniques and methods. During this study, we aim to develop a mathemati-
cal approach to prove the existence of a unique ergodic stationary distribution and persistence in the mean of the new perturbed
model. Without using the classical Lyapunov method presented in 28 , we obtain sufficient conditions for the ergodicity by
employing the Feller property and mutually exclusive possibilities lemma. Under the same conditions, we demonstrate that the
persistence in the mean of the disease occurs. To analyze properly our new model, we study the stochastic extinction case.

The organization of this paper is as follows: in section 2, we present our new stochastic system and some preliminary results.
In section 3, we focus on the stochastic characteristics of the perturbed model. Since the ergodicity is an important statistical
characteristic, the existence of a unique stationary distribution is obtained. Almost sufficient condition for the persistence is also
established. To complete our study, we give sufficient conditions for the disease extinction. Finally, we support our theoretical
results by illustrating some numerical examples.

2 THE STOCHASTIC SIQS MODEL AND SOME PRELIMINARIES

Let (Ω,  , ℙ) be a complete probability space with a filtration {𝑡 }𝑡≥0 satisfying the usual conditions, and containing all the ran-
dom variables that will be meted in this paper. We merge the stochastic transmission with a discontinuous perturbed mortality
rates. The random variability in the epidemic transmission 𝛽 and the mortality rates 𝜇𝑖 (𝑖 = 1, 2, 3) are presented by a decompo-
sition of usual white noise and the Lévy-Itô process, respectively. Under these assumptions, the evolution of an epidemic during
the quarantine strategy is modeled by the following system of stochastic differential equations:
( )
⎧d𝑆(𝑡) = 𝐴 − 𝜇1 𝑆(𝑡) − 𝛽𝑆(𝑡)𝐼(𝑡) + 𝛾𝐼(𝑡) + 𝑘𝑄(𝑡) d𝑡 + 1 (𝑡),
⎪ ( )
⎨d𝐼(𝑡) = 𝛽𝑆(𝑡)𝐼(𝑡) − (𝜇2 + 𝛿 + 𝛾)𝐼(𝑡) d𝑡 + 2 (𝑡), (2)
⎪ ( )
⎩d𝑄(𝑡) = 𝛿𝐼(𝑡) − (𝜇3 + 𝑘)𝑄(𝑡) d𝑡 + 3 (𝑡),
where
1 (𝑡) = 𝜎1 𝑆(𝑡)d1 (𝑡) + ̃(d𝑡, d𝑢) − 𝜎 𝑆(𝑡)𝐼(𝑡)d (𝑡),
𝜂1 (𝑢)𝑆(𝑡− ) 𝛽 𝛽
∫
𝑍

2 (𝑡) = 𝜎2 𝐼(𝑡)d2 (𝑡) + ̃(d𝑡, d𝑢) + 𝜎 𝑆(𝑡)𝐼(𝑡)d (𝑡),

𝜂2 (𝑢)𝐼(𝑡− ) 𝛽 𝛽
∫
𝑍

3 (𝑡) = 𝜎3 𝑄(𝑡)d3 (𝑡) + ̃(d𝑡, d𝑢).

𝜂3 (𝑢)𝑄(𝑡− )
∫
𝑍
Here, 𝛽 (𝑡) and 𝑖 (𝑡) (𝑖 = 1, 2, 3) are the mutually independent Brownian motions defined on (Ω,  , {𝑡 }𝑡≥0 , ℙ) with the
positive intensities 𝜎𝛽 and 𝜎𝑖 (𝑖 = 1, 2, 3). 𝑆(𝑡− ), 𝐼(𝑡− ) and 𝑄(𝑡− ) are the left limits of 𝑆(𝑡), 𝐼(𝑡) and 𝑄(𝑡), respectively.  is a
Poisson counting measure with compensating martingale  ̃ and characteristic measure 𝜈 on a measurable subset 𝑍 of (0, ∞)
satisfying 𝜈(𝑍) < ∞. It assumed that 𝜈 is a Lévy measure such that  ̃(d𝑡, d𝑢) =  (d𝑡, d𝑢) − 𝜈(d𝑢)d𝑡. We also assume that
𝑖 (𝑡) (𝑖 = 1, 2, 3, 𝛽) are independent of  . The bounded functions 𝜂𝑖 ∶ 𝑍 → ℝ (𝑖 = 1, 2, 3) are continuous. For the sake of
notational simplicity, we define

∙ 𝜎̄ = max{𝜎12 , 𝜎22 , 𝜎32 }.

∙ 𝜂(𝑢)
̄ = max{𝜂1 (𝑢), 𝜂2 (𝑢), 𝜂3 (𝑢)}.

∙ 𝜂(𝑢) = min{𝜂1 (𝑢), 𝜂2 (𝑢), 𝜂3 (𝑢)}.

[ ]𝑛𝑝
∙ 𝜌̂𝑛,𝑝 (𝑢) = 1 + 𝜂(𝑢)
̄ − 1 − 𝑛𝑝𝜂(𝑢).
̄
4

[ ]𝑛𝑝
∙ 𝜌̌𝑛,𝑝 (𝑢) = 1 + 𝜂(𝑢) − 1 − 𝑛𝑝𝜂(𝑢).
[ ]
∙ 𝓁𝑛,𝑝 = ∫𝑍 𝜌̂𝑛,𝑝 (𝑢) ∨ 𝜌̌𝑛,𝑝 (𝑢) 𝜈(d𝑢).

To properly study our model (2), we have the following fundamental assumptions on the jump-diffusion coefficients:

∙ (A1 ) We assume that the jump coefficients 𝜂𝑖 (𝑢) in (2) satisfy ∫𝑍 𝜂𝑖2 (𝑢)𝜈(d𝑢) < ∞, {𝑖 = 1, 2, 3}.
[ ]
∙ (A2 ) For all 𝑢 ∈ 𝑍, we assume that 1 + 𝜂𝑖 (𝑢) > 0 and ∫𝑍 𝜂𝑖 (𝑢) − ln(1 + 𝜂𝑖 (𝑢)) 𝜈(d𝑢) < ∞, {𝑖 = 1, 2, 3}.
[ ]2
∙ (A3 ) We suppose that ∫𝑍 ln(1 + 𝜂𝑖 (𝑢)) 𝜈(d𝑢) < ∞, {𝑖 = 1, 2, 3}.
[( )2 ]2
∙ (A4 ) We suppose that ∫𝑍 1 + 𝜂(𝑢)
̄ − 1 𝜈(d𝑢) < ∞.

∙ (A5 ) We suppose that for each positive integer 𝑛 there is some real number 𝑝 > 1 for which
(𝑛𝑝 − 1) 1
Γ𝑛,𝑝 = 𝜇1 − 𝜎̄ − 𝓁𝑛,𝑝 > 0.
2 𝑛𝑝
In view of the biological interpretation, the question of whether the stochastic model is well-posed is the first concern. There-
fore, to analyze the stochastic model (2), it is necessary to verify the existence of a unique global positive solution, that is,
there is no explosion in finite time for any positive initial value (𝑆(0), 𝐼(0), 𝑄(0)) ∈ ℝ3+ . The following lemma assures the
well-posedness of the stochastic model (2).
Lemma 2.1. Let assumptions (A1 ) and (A2 ) hold. For any initial value 𝑌 (0) = (𝑆(0), 𝐼(0), 𝑄(0)) ∈ ℝ3+ , there exists a unique
positive solution 𝑌 (𝑡) = (𝑆(𝑡), 𝐼(𝑡), 𝑄(𝑡)) of system (2) on 𝑡 ≥ 0, and this solution will stay in ℝ3+ almost surely.
The proof is somehow standard and classic (see for example 29,30 ), so we omit it here.
In the following, we always presume that the assumptions (A1 ) - (A5 ) hold. For reference purposes, we will prepare several
useful lemmas.
Lemma 2.2. Let 𝑛 be a positive integer and let 𝑌 (𝑡) denotes the solution of system (2) that starts from a given point 𝑌 (0) ∈ ℝ3+ .
Then, for any 𝑝 > 1 that satisfies Γ𝑛,𝑝 > 0, we have
[ ] 𝑛𝑝Γ𝑛,𝑝
2Δ
∙ 𝔼 𝑁 𝑛𝑝 (𝑡) ≤ 𝑁 𝑛𝑝 (0) × 𝑒− 2 𝑡 + .
Γ𝑛,𝑝
𝑡
1 [ ] 2Δ
∙ lim sup 𝔼 𝑁 𝑛𝑝 (𝑠) d𝑠 ≤ .
𝑡→∞ 𝑡 ∫ Γ𝑛,𝑝
0
Γ𝑛,𝑝
where Δ = sup {𝐴𝑁 𝑛𝑝−1 − 2
𝑁 𝑛𝑝 } and 𝑁(𝑡) = 𝑆(𝑡) + 𝐼(𝑡) + 𝑄(𝑡).
𝑁>0

Proof. Making use of Itô’s lemma 31 to 𝑁 𝑛𝑝 (𝑡), we obtain

{ ( ) 𝑛𝑝 ( )
d𝑁 (𝑡) = 𝑛𝑝𝑁 𝑛𝑝−1 (𝑡) 𝐴 − 𝜇1 𝑁(𝑡) − 𝑟2 𝐼(𝑡) − 𝑟3 𝑄(𝑡) + (𝑛𝑝 − 1)𝑁 𝑛𝑝−2 (𝑡) 𝜎12 𝑆 2 (𝑡) + 𝜎22 𝐼 2 (𝑡) + 𝜎32 𝑄2 (𝑡)
𝑛𝑝
2
[( )𝑛𝑝
𝑛𝑝 𝑆(𝑡) 𝐼(𝑡) 𝑄(𝑡)
+ 𝑁 (𝑡) 1 + 𝜂1 (𝑢) + 𝜂2 (𝑢) + 𝜂3 (𝑢) −1
∫ 𝑁(𝑡) 𝑁(𝑡) 𝑁(𝑡)
𝑍
( )] }
𝑆(𝑡) 𝐼(𝑡) 𝑄(𝑡)
− 𝑛𝑝 𝜂1 (𝑢) + 𝜂2 (𝑢) + 𝜂3 (𝑢) 𝜈(d𝑢) d𝑡
𝑁(𝑡) 𝑁(𝑡) 𝑁(𝑡)
( )
𝑛𝑝−1
+ 𝑛𝑝𝑁 (𝑡) 𝜎1 𝑆(𝑡)d1 (𝑡) + 𝜎2 𝐼(𝑡)d2 (𝑡) + 𝜎3 𝑄(𝑡)d3 (𝑡)
[( )𝑛𝑝 ]
𝑛𝑝 − 𝑆(𝑡− ) 𝐼(𝑡− ) 𝑄(𝑡− ) ̃(d𝑡, d𝑢).
+ 𝑁 (𝑡 ) 1 + 𝜂1 (𝑢) + 𝜂2 (𝑢) + 𝜂3 (𝑢) −1 
∫ 𝑁(𝑡− ) 𝑁(𝑡− ) 𝑁(𝑡− )
𝑍
 5

Then
( ( ) )
𝑛𝑝 𝑛𝑝−1 𝑛𝑝 𝑛𝑝 𝑛𝑝
[ ]
d𝑁 (𝑡) ≤ 𝑛𝑝𝑁 (𝑡) 𝐴 − 𝜇1 𝑁(𝑡) + (𝑛𝑝 − 1)𝑁 (𝑡)𝜎̄ + 𝑁 (𝑡) 𝜌̂𝑛,𝑝 (𝑢) ∨ 𝜌̌𝑛,𝑝 (𝑢) 𝜈(d𝑢) d𝑡
2 ∫
𝑍
( )
+ 𝑛𝑝𝑁 𝑛𝑝−1 (𝑡) 𝜎1 𝑆(𝑡)d1 (𝑡) + 𝜎2 𝐼(𝑡)d2 (𝑡) + 𝜎3 𝑄(𝑡)d3 (𝑡)
( )
+ 𝑁 𝑛𝑝 (𝑡− ) (1 + 𝜂(𝑢))
̄ 𝑛𝑝
−1  ̃(d𝑡, d𝑢). (3)
∫
𝑍
Rewriting the above inequality, one can see that
{ ( ) }
(𝑛𝑝 − 1) 1 [ ]
d𝑁 𝑛𝑝 (𝑡) ≤ 𝑛𝑝 𝐴𝑁 𝑛𝑝−1 (𝑡) − 𝜇1 − 𝜎̄ − 𝜌̂𝑛,𝑝 (𝑢) ∨ 𝜌̌𝑛,𝑝 (𝑢) 𝜈(d𝑢) 𝑁 𝑛𝑝 (𝑡) d𝑡
2 𝑛𝑝 ∫
𝑍
( )
𝑛𝑝−1
+ 𝑛𝑝𝑁 (𝑡) 𝜎1 𝑆(𝑡)d1 (𝑡) + 𝜎2 𝐼(𝑡)d2 (𝑡) + 𝜎3 𝑄(𝑡)d3 (𝑡)
( )
+ 𝑁 𝑛𝑝 (𝑡− ) (1 + 𝜂(𝑢))
̄ 𝑛𝑝
−1  ̃(d𝑡, d𝑢).
∫
𝑍
(𝑛𝑝 − 1) 1 [ ]
We choose neatly 𝑝 > 1 such that Γ𝑛,𝑝 = 𝜇1 − 𝜎̄ − 𝜌̂𝑛,𝑝 (𝑢) ∨ 𝜌̌𝑛,𝑝 (𝑢) 𝜈(d𝑢) > 0. Therefore
2 𝑛𝑝 ∫
𝑍
{ Γ𝑛,𝑝 } ( )
d𝑁 𝑛𝑝 (𝑡) ≤ 𝑛𝑝 Δ − 𝑁 𝑛𝑝 (𝑡) d𝑡 + 𝑛𝑝𝑁 𝑛𝑝−1 (𝑡) 𝜎1 𝑆(𝑡)d1 (𝑡) + 𝜎2 𝐼(𝑡)d2 (𝑡) + 𝜎3 𝑄(𝑡)d3 (𝑡)
2
( )
+ 𝑛𝑝 −
𝑁 (𝑡 ) (1 + 𝜂(𝑢))
̄ 𝑛𝑝
−1 ̃(d𝑡, d𝑢).
∫
𝑍
On the other hand, we have
𝑛𝑝Γ𝑛,𝑝 𝑛𝑝Γ𝑛,𝑝 𝑛𝑝Γ𝑛,𝑝
𝑡 𝑡 𝑡
d𝑁 𝑛𝑝 (𝑡) × 𝑒 2 = 𝑝Γ𝑛,𝑝 𝑁 𝑛𝑝 (𝑡) × 𝑒 d𝑁 𝑛𝑝 (𝑡)
2 +𝑒 2

( [ )
𝑛𝑝Γ𝑛,𝑝 𝑛𝑝Γ𝑛,𝑝
≤ 𝑛𝑝Δ𝑒 2 𝑡 + 𝑒 2 𝑡 𝑛𝑝𝑁 𝑛𝑝−1 (𝑡) 𝜎1 𝑆(𝑡)d1 (𝑡) + 𝜎2 𝐼(𝑡)d2 (𝑡) + 𝜎3 𝑄(𝑡)d3 (𝑡)
( ) ]
+ 𝑁 𝑛𝑝 (𝑡− ) (1 + 𝜂(𝑢))
̄ 𝑛𝑝
−1 ̃(d𝑡, d𝑢) .
∫
𝑍
Then, by taking the integration and the expectations, we get
𝑡
[ ] −
𝑛𝑝Γ𝑛,𝑝 𝑛𝑝 𝑛𝑝Γ𝑛,𝑝
2Δ
𝑛𝑝
𝔼 𝑁 (𝑡) ≤ 𝑁 (0) × 𝑒 𝑛𝑝 2
𝑡
+ 𝑛𝑝Δ 𝑒− 2 Γ𝑛,𝑝 (𝑡−𝑠) d𝑠 ≤ 𝑁 𝑛𝑝 (0)𝑒− 2
𝑡
+ .
∫ Γ𝑛,𝑝
0
Obviously, we obtain
𝑡 𝑡
1 [ ] 1 𝑛𝑝Γ𝑛,𝑝
2Δ 2Δ
lim sup 𝔼 𝑁 𝑛𝑝 (𝑠) d𝑠 ≤ 𝑁 𝑛𝑝 (0) × lim sup 𝑒− 2 𝑠 d𝑠 + = .
𝑡→∞ 𝑡 ∫ 𝑡→∞ 𝑡 ∫ Γ𝑛,𝑝 Γ𝑛,𝑝
0 0
This completes the proof.

Remark 2.3. Throughout this remark, 𝑋̃ is standing for the sum 𝜂1 (𝑢)𝑆 + 𝜂2 (𝑢)𝐼 + 𝜂3 (𝑢)𝑄, where 𝑢 ∈ 𝑍. In the study of
stochastic biological models driven by Lévy jumps (see for example, 23,30,31,32,33 ), the following quantity
[( ) ]
𝑛𝑝 𝑋̃ 𝑛𝑝 𝑋̃
𝑁 1+ − 1 − 𝑛𝑝 𝜈(d𝑢),
∫ 𝑁 𝑁
𝑍
is widely majorazed by
( )
𝑁 𝑛𝑝 (1 + 𝜂(𝑢))
̄ 𝑛𝑝
− 1 − 𝑛𝑝𝜂(𝑢) 𝜈(d𝑢).
∫
𝑍
6

However, the last estimation can be ameliorated by considering the following inequality
[( ) ]
𝑋̃ 𝑛𝑝 𝑋̃ [ ]
𝑁 𝑛𝑝 1 + − 1 − 𝑛𝑝 𝜈(d𝑢) ≤ 𝑁 𝑛𝑝 𝜌̂𝑛,𝑝 (𝑢) ∨ 𝜌̌𝑛,𝑝 (𝑢) 𝜈(d𝑢), (4)
∫ 𝑁 𝑁 ∫
𝑍 𝑍
which is established from the observation that the function
𝑔(𝑥) = (1 + 𝑥)𝑛𝑝 − 1 − 𝑛𝑝𝑥, 𝑛, 𝑝 ≥ 1,
is decreasing for 𝑥 ∈ (−1, 0) and increasing for 𝑥 ≥ 0. Needless to say, this last fact makes necessarily 𝑔(𝑎) ∨ 𝑔(𝑏) as the highest
value of 𝑔 on any interval [𝑎, 𝑏] ⊂ (−1, ∞). The adoption of the inequality (4) in our calculus, especially in (3), (6) and (11),
will improve many classical results presented in the above mentioned papers.
Remark 2.4. Lemma 2.2 takes into consideration the stochastic transmission and the effect of Lévy jumps, and this makes it
clearly an extended version of Lemma 2.3 presented in 34 .
Lemma 2.5. Consider the initial value problem
{ ( )
d𝑋(𝑡) = 𝐴 − 𝜇1 𝑋(𝑡) d𝑡 + ̄ 1 (𝑡) + ̄ 2 (𝑡) + 3 (𝑡),
(5)
𝑋(0) = 𝑁(0) ∈ ℝ+ ,
where
̄ 1 (𝑡) = 𝜎1 𝑆(𝑡)d1 (𝑡) + ̃(d𝑡, d𝑢),
𝜂1 (𝑢)𝑆(𝑡− )
∫
𝑍

̄ 2 (𝑡) = 𝜎2 𝐼(𝑡)d2 (𝑡) + ̃(d𝑡, d𝑢).

𝜂2 (𝑢)𝐼(𝑡− )
∫
𝑍
Let us denote by 𝑋(𝑡) and 𝑌 (𝑡) the positive solutions of systems (2) and (5) respectively. Then
𝑋 𝑛 (𝑡)
∙ lim = 0 a.s., ∀𝑛 ∈ {1, 2, ⋯}.
𝑡→∞ 𝑡
𝑡 𝑡 𝑡
∫0 𝑋(𝑠)𝑆(𝑠)d1 (𝑠) ∫0 𝑋(𝑠)𝐼(𝑠)d2 (𝑠) ∫0 𝑋(𝑠)𝑄(𝑠)d3 (𝑠)
∙ lim = 0, lim = 0, and lim = 0 a.s.
𝑡→∞ 𝑡 𝑡→∞ 𝑡 𝑡→∞ 𝑡
𝑡 ( 2
) ̃(d𝑠, d𝑢)
∫0 ∫𝑍 (1 + 𝜂(𝑢))
̄ − 1 𝑋 2 (𝑠− )
∙ lim = 0 a.s.
𝑡→∞ 𝑡

Proof. Our approach to demonstrate this lemma is mainly adapted from 30 . The proof falls naturally into three steps.

Step 1. Applying the generalized Itô’s formula 31 to (𝑋) = 𝑋 𝑛𝑝 , where 𝑛 is a fixed integer number, we derive
( )
d(𝑋) ≤ d𝑡 + 𝑛𝑝𝑋 𝑛𝑝−1 𝜎1 𝑆d1 (𝑡) + 𝜎2 𝐼d2 (𝑡) + 𝜎3 𝑄d3 (𝑡)
( )
+ 𝑋 𝑛𝑝 (𝑡− ) (1 + 𝜂(𝑢))
̄ 𝑛𝑝
−1 ̃(d𝑡, d𝑢), (6)
∫
𝑍
where
[ ( (𝑛𝑝 − 1) ) ]
1
 ≤ 𝑛𝑝𝑋 𝑛𝑝−2 𝐴𝑋 − 𝜇1 − 𝜎̄ − 𝓁𝑛,𝑝 𝑋 2 .
2 𝑛𝑝
Choose a positive constant 𝑝 > 1 such that Γ𝑛,𝑝 = 𝜇1 − (𝑛𝑝−1)
2
1
𝜎̄ − 𝑛𝑝 𝓁𝑛,𝑝 > 0. Then
( ( ) ) ( )
d(𝑋) ≤ 𝑛𝑝𝑋 𝑛𝑝−2 𝐴𝑋 − Γ𝑛,𝑝 𝑋 2 d𝑡 + 𝑛𝑝𝑋 𝑛𝑝−1 𝜎1 𝑆d1 (𝑡) + 𝜎2 𝐼d2 (𝑡) + 𝜎3 𝑄d3 (𝑡)
( )
+ 𝑋 𝑛𝑝 (𝑡− ) (1 + 𝜂(𝑢))
̄ 𝑛𝑝
−1  ̃(d𝑡, d𝑢). (7)
∫
𝑍
 7

For any constant 𝑚 satisfying 𝑚 ∈ (0, 𝑛𝑝Γ𝑛,𝑝 ), one can see that
( ) ( )
d𝑒𝑚𝑠 (𝑋(𝑠)) ≤  𝑒𝑚𝑡 (𝑋(𝑡)) + 𝑛𝑝𝑒𝑚𝑡 𝑋 𝑛𝑝−1 (𝑡) 𝜎1 𝑆(𝑡)d1 (𝑡) + 𝜎2 𝐼(𝑡)d2 (𝑡) + 𝜎3 𝑄(𝑡)d3 (𝑡)
( )
+ 𝑒𝑚𝑡 𝑋 𝑛𝑝 (𝑡− ) (1 + 𝜂(𝑢))
̄ 𝑛𝑝
−1 ̃(d𝑡, d𝑢).
∫
𝑍
Integrating both sides of the last inequality from 0 to 𝑡, we get
𝑡 𝑡
( ( ))
𝑚𝑠
d𝑒 (𝑋(𝑠)) ≤ 𝑚𝑒𝑚𝑠 (𝑋(𝑠)) + 𝑒𝑚𝑠  (𝑋(𝑠) d𝑠
∫ ∫
0 0
𝑡
( )
+ 𝑛𝑝 𝑒𝑚𝑠 𝑋 𝑛𝑝−1 (𝑠) 𝜎1 𝑆(𝑠)d1 (𝑠) + 𝜎2 𝐼(𝑠)d2 (𝑠) + 𝜎3 𝑄(𝑠)d3 (𝑠)
∫
0
𝑡
( )
+ 𝑒𝑚𝑠 𝑋 𝑛𝑝 (𝑠− ) (1 + 𝜂(𝑢))
̄ 𝑛𝑝
−1 ̃(d𝑠, d𝑢).
∫ ∫
0 𝑍
Taking expectation on both sides yields that
𝑡
{ ( ( ) }
𝑚𝑡
𝔼𝑒 (𝑋(𝑡)) ≤ (𝑋(0)) + 𝔼 𝑚𝑒𝑚𝑠 (𝑋(𝑠)) + 𝑒𝑚𝑠  (𝑋(𝑠)) d𝑠 .
∫
0
In view of (7), we can see that
( )
̄
𝑚𝑒𝑚𝑡 (𝑋(𝑡)) + 𝑒𝑚𝑡  (𝑋) ≤ 𝑛𝑝𝑒𝑚𝑡 𝐻,
{ [ ( ) ] }
𝑚
where 𝐻̄ = sup 𝑋 𝑛𝑝−2 − Γ𝑛,𝑝 − 𝑛𝑝 𝑋 2 + 𝐴𝑋 + 1 . Then, we have
𝑋>0

𝑛𝑝𝐻̄ 𝑚𝑡
𝔼𝑒𝑚𝑡 (𝑋(𝑠)) ≤ (𝑋(0)) + 𝑒 .
𝑚
Therefore, we get
[ ] 𝑛𝑝𝐻̄
lim sup 𝔼 𝑋 𝑛𝑝 (𝑡) ≤ a.s.
𝑡→∞ 𝑚
̄ such that for all 𝑡 ≥ 0,
Consequently, there exists a positive constant 𝑀
[ ]
𝔼 𝑋 𝑛𝑝 (𝑡) ≤ 𝑀. ̄ (8)
Step 2. Integrating from 0 to 𝑡 after applying the famous Burkholder-Davis-Gundy inequality 35 to (7), allows us to conclude
that for an arbitrarily small positive constant 𝑧, 𝑚 = 1, 2, ...,
[ ] [ ]𝑛𝑝 ( 1( ( )2 ))
𝔼 sup 𝑋 𝑛𝑝 (𝑡) ≤ 𝔼 𝑋(𝑚𝑧) + 𝑧1 𝑧 + 𝑧2 𝑧 2 𝑛𝑝𝜎̄ + (1 + 𝜂(𝑢))
̄ 𝑛𝑝
− 1 𝜈(d𝑢)
𝑚𝑧≤𝑡≤(𝑚+1)𝑧 ∫
𝑍
[ ]
× sup 𝑋 𝑛𝑝 (𝑡) ,
𝑚𝑧≤𝑡≤(𝑚+1)𝑧

where 𝑧1 and 𝑧2 are positive constants. Specially, we select 𝑧 > 0 such that
( ( )2 )
1
𝑛𝑝 1
𝑧1 𝑧 + 𝑧2 𝑧 2 𝑛𝑝𝜎̄ + (1 + 𝜂(𝑢))
̄ − 1 𝜈(d𝑢) ≤ .
∫ 2
𝑍
Then
[ ]
𝔼 sup ̄
𝑋 𝑛𝑝 (𝑡) ≤ 2𝑀.
𝑚𝑧≤𝑡≤(𝑚+1)𝑧
8

Let 𝜖̄ > 0 be arbitrary. By employing Chebyshev’s inequality, we derive

[ ]
{ } 𝔼 sup 𝑋 𝑛𝑝 (𝑡)
𝑚𝑧≤𝑡≤(𝑚+1)𝑧 2𝑀̄
ℙ sup 𝑋 𝑛𝑝 (𝑡) > (𝑚𝑧)1+𝜖̄ ≤ ≤ .
𝑚𝑧≤𝑡≤(𝑚+1)𝑧 (𝑚𝑧)1+𝜖̄ (𝑚𝑧)1+𝜖̄
Making use of the Borel-Cantelli lemma gives that for almost all 𝜔 ∈ Ω
sup 𝑋 𝑛𝑝 (𝑡) ≤ (𝑚𝑧)1+𝜖̄ , (9)
𝑚𝑧≤𝑡≤(𝑚+1)𝑧

verifies for all but finitely many 𝑚. Consequently, there exists a positive constant 𝑚0 (𝜔) such that 𝑚0 ≤ 𝑚 and (9) holds for
almost all 𝜔 ∈ Ω. In other words, for almost all 𝜔 ∈ Ω, if 𝑚0 ≤ 𝑚 and 𝑚𝑧 ≤ 𝑡 ≤ (𝑚 + 1)𝑧,
ln 𝑋 𝑛𝑝 (𝑡) (1 + 𝜖) ̄ ln(𝑚𝑧)
≤ = 1 + 𝜖.
̄
ln 𝑡 ln(𝑚𝑧)
Because 𝜖̄ is arbitrarily small, then
ln 𝑋 𝑛 (𝑡) 1
lim sup ≤ a.s.
𝑡→∞ ln 𝑡 𝑝
Therefore, for any small 𝑣̄ ∈ (0, 1 − 1∕𝑝), there is a constant 𝑉̄ = 𝑉̄ (𝜔), for which if 𝑡 ≥ 𝑉̄ then
( )
1
ln 𝑋 𝑛 (𝑡) ≤ + 𝑣̄ ln 𝑡.
𝑝
Hence
1
+𝑣̄
𝑋 𝑛 (𝑡) 𝑡𝑝
lim sup ≤ lim sup = 0.
𝑡→∞ 𝑡 𝑡→∞ 𝑡
This together with the positivity of the solution implies
𝑋 𝑛 (𝑡)
lim = 0 a.s.
𝑡→∞ 𝑡
Step 3. Now, we define
𝑡 𝑡
1 1
1 (𝑡) = 𝑋(𝑠)𝑆(𝑠)d1 (𝑠), 2 (𝑡) = 𝑋(𝑠)𝐼(𝑠)d2 (𝑠),
𝑡∫ 𝑡∫
0 0
𝑡 𝑡
1 1 (( )2 )̃
3 (𝑡) = 𝑋(𝑠)𝑄(𝑠)d3 (𝑠), 4 (𝑡) = 𝑋 2 (𝑠− ) 1 + 𝜂̄ − 1  (d𝑠, d𝑢).
𝑡∫ 𝑡 ∫ ∫
0 0 𝑍
In view of the Burkholder–Davis–Gundy inequality, we find that for 𝑝̄ > 2,
[ ] [ 𝑡 ] 𝑝̄ [ 𝑡 ] 𝑝̄ [ 𝑡 ] 𝑝̄
2 2 2

𝔼 sup |1 (𝑡)|𝑝̄ ≤ 𝐶𝑝̄ 𝔼 𝑋 4 (𝑠)d𝑠 ≤ 𝐶𝑝̄ 𝔼 𝑋 4 (𝑠)d𝑠 ≤ 𝐶𝑝̄ 𝔼 |𝑋 4 (𝑠)|d𝑠 , (10)
𝑚≤𝑡≤(𝑚+1) ∫ ∫ ∫
0 0 0
[ ]𝑝∕2
̄
𝑝̄𝑝+1
̄
where 𝐶𝑝̄ = 2(𝑝−1)
̄ 𝑝−1
̄
> 0. Similarly to the previous case, we find
[ ] ( ) 𝑝̄ [ 𝑡 ] 𝑝̄
(( )2 )2 2 2

𝔼 sup |4 (𝑡)|𝑝̄ ≤ 𝐶𝑝̄ 1 + 𝜂̄ −1 𝜈(d𝑢) 𝔼 |𝑋 4 (𝑠)|d𝑠 .

𝑚≤𝑡≤(𝑚+1) ∫ ∫
𝑍 0
Via (8) and (10), one can see that
[ ] 𝑝̄ 𝑝̄
𝔼 sup |1 (𝑡)|𝑝̄ ≤ 21+ 2 𝑀𝐶
̄ 𝑝̄ 𝑚 2 .
𝑚≤𝑡≤(𝑚+1)

For any arbitrary positive constant 𝜖,

̃ and by making use of Chebyshev’s inequality, we obtain
[ ]
{ } 𝔼 sup | 1 (𝑚 + 1)|𝑝̄ 𝑝̄

̃ 2𝑝̄
1+𝜖+ 𝑚≤𝑡≤(𝑚+1) 21+ 2 𝑀𝐶
̄ 𝑝̄
ℙ sup |1 (𝑡)| > 𝑝̄
𝑝̄
≤ 𝑝̄
≤ , 𝑚 = 1, 2, ...
𝑚≤𝑡≤(𝑚+1) 𝑝̄1+𝜖+
̃ 2 𝑝̄1+𝜖̃
 9

Using the Borel-Cantelli lemma, one has

( )
ln |1 (𝑡)|𝑝̄ 1 + 𝜖̃ + 2𝑝̄ ln 𝑚 𝑝̄
≤ = 1 + 𝜖̃ + .
ln 𝑡 ln 𝑚 2
Taking the limit superior on both sides of the last inequality and applying the arbitrariness of 𝜖, ̃ we deduce
ln |1 (𝑡)| 1 1
lim sup ≤ + a.s.
𝑡→∞ ln 𝑡 2 𝑝̄
( )
That is to say, for any positive constant 𝜏̄ ∈ 0, 12 − 1𝑝̄ , there exists a constant 𝑇̄ = 𝑇̄ (𝜔) such that for all 𝑡 ≥ 𝑇̄ ,
( )
1 1
ln |1 (𝑡)| ≤ + + 𝜏̄ ln 𝑡.
2 𝑝̄
Dividing both sides of the last inequality by 𝑡 and taking the limit superior, we have
1 1
|1 (𝑡)| + +𝜏̄
𝑡 2 𝑝̄
lim sup ≤ lim sup = 0.
𝑡→∞ 𝑡 𝑡→∞ 𝑡
|1 (𝑡)| | (𝑡)| 1 (𝑡)
Combining it with lim inf ≥ 0, one has lim 1𝑡
𝑡
= lim 𝑡
= 0 a.s.
𝑡→∞ 𝑡→∞ 𝑡→∞
In the same way, we prove that
 (𝑡) 3 (𝑡) 4 (𝑡)
lim 2 = 0, lim = 0, lim = 0 a.s.
𝑡→∞ 𝑡 𝑡→∞ 𝑡 𝑡→∞ 𝑡
This completes the proof.

Remark 2.6. The positivity of the solutions 𝑋(𝑡) and 𝑌 (𝑡) together with the stochastic comparison theorem 35 , leads to the fact
that 𝑁(𝑡) ≤ 𝑋(𝑡) a.s. which in turn implies that
𝑆 𝑛 (𝑡) 𝐼 𝑛 (𝑡) 𝑄𝑛 (𝑡) 𝑁 𝑛 (𝑡)
lim = 0, lim = 0, lim = 0, and even lim = 0 a.s.
𝑡→∞ 𝑡 𝑡→∞ 𝑡 𝑡→∞ 𝑡 𝑡→∞ 𝑡

Remark 2.7. By comparing our findings with those of Lemmas 3.3 and 3.4 in 30 , one can conclude that the new result 2.5
presents a modified and generalized version to these lemmas, which will be necessary to prove Lemma 2.11.
Lemma 2.8. Let 𝑌 (0) ∈ ℝ3+ be a positive given value. If 𝑌 (𝑡) denotes the positive solution of system (2) that starts from 𝑌 (0),
then
𝑡 𝑡 𝑡 𝑡
∫0 𝑆(𝑠)d𝛽 (𝑠) ∫0 𝑆(𝑠)d1 (𝑠) ∫0 𝐼(𝑠)d2 (𝑠) ∫0 𝑄(𝑠)d3 (𝑠)
∙ lim = 0, lim = 0, lim = 0, and lim = 0 a.s.
𝑡→∞ 𝑡 𝑡→∞ 𝑡 𝑡
𝑡→∞ 𝑡→∞ 𝑡
𝑡 ̃(d𝑠, d𝑢) 𝑡 ̃(d𝑠, d𝑢) 𝑡 ̃(d𝑠, d𝑢)
∫0 ∫𝑍 𝜂1 (𝑢)𝑆(𝑠− ) ∫0 ∫𝑍 𝜂2 (𝑢)𝐼(𝑠− ) ∫0 ∫𝑍 𝜂3 (𝑢)𝑄(𝑠− )
∙ lim = 0, lim = 0, lim = 0 a.s.
𝑡→∞ 𝑡 𝑡→∞ 𝑡 𝑡→∞ 𝑡

Remark 2.9. The last lemma is easily demonstrated by using an analysis similar to that in the proof of Lemma 2.5.
Remark 2.10. In the absence of Lévy noise (see for example 36 ), the stationary distribution expression is used to calculate the
time averages of the auxiliary process solution by employing the ergodic theorem 35 . Unfortunately, the said expression is still
unknown in the case of the Lévy jumps. This problem is implicitly mentioned in 25 as an open question, and the authors presented
the threshold analysis of their model with an unknown stationary distribution formula. In this article, we propose an alternative
method to establish the exact expression of the threshold parameter without having recourse to the use of ergodic theorem. This
new idea that we propose is presented in the following lemma.
Lemma 2.11. Let 𝑋(𝑡) ∈ ℝ+ be the solution of the equation (5) with any given initial value 𝑋(0) = 𝑁(0) ∈ ℝ+ . Suppose that
[ ]
𝜒 = 2𝜇1 − 𝜎̄ − ∫𝑍 𝜂̄ 2 (𝑢) ∨ 𝜂 2 (𝑢) 𝜈(d𝑢) > 0, then
𝑡
1 𝐴
lim 𝑋(𝑠)d𝑠 = a.s.
𝑡→∞ 𝑡 ∫ 𝜇1
0
10

and
𝑡
1 2𝐴2
lim 𝑋 2 (𝑠)d𝑠 ≤ a.s.
𝑡→∞ 𝑡 ∫ 𝜇1 𝜒
0

Proof. Integrating from 0 to 𝑡 on both sides of (5) yields

𝑡 𝑡 𝑡
𝑋(𝑡) − 𝑋(0) 𝜇 𝜎 1 ̃(d𝑠, d𝑢)
=𝐴− 1 𝑋(𝑠)d𝑠 + 1 𝑆(𝑠)d1 (𝑠) + 𝜂 (𝑢)𝑆(𝑠− )
𝑡 𝑡 ∫ 𝑡 ∫ 𝑡∫ ∫ 1
0 0 0 𝑍
𝑡 𝑡
𝜎2 1 ̃(d𝑠, d𝑢)
+ 𝐼(𝑠)d2 (𝑠) + 𝜂 (𝑢)𝐼(𝑠− )
𝑡 ∫ 𝑡∫ ∫ 2
0 0 𝑍
𝑡 𝑡
𝜎3 1 ̃(d𝑠, d𝑢).
+ 𝑄(𝑠)d3 (𝑠) + 𝜂 (𝑢)𝑄(𝑠− )
𝑡 ∫ 𝑡∫ ∫ 3
0 0 𝑍
Clearly, we can derive that
𝑡 𝑡 𝑡
1 𝐴 𝑋(𝑡) − 𝑋(0) 𝜎 1 ̃(d𝑠, d𝑢)
𝑋(𝑠)d𝑠 = − + 1 𝑆(𝑠)d1 (𝑠) + 𝜂 (𝑢)𝑆(𝑠− )
𝑡∫ 𝜇1 𝜇1 𝑡 𝜇1 𝑡 ∫ 𝜇1 𝑡 ∫ ∫ 1
0 0 0 𝑍
𝑡 𝑡
𝜎2 1 ̃(d𝑠, d𝑢)
+ 𝐼(𝑠)d2 (𝑠) + 𝜂 (𝑢)𝐼(𝑠− )
𝜇1 𝑡 ∫ 𝜇1 𝑡 ∫ ∫ 2
0 0 𝑍
𝑡 𝑡
𝜎3 1 ̃(d𝑠, d𝑢).
+ 𝑄(𝑠)d3 (𝑠) + 𝜂 (𝑢)𝑄(𝑠− )
𝜇1 𝑡 ∫ 𝜇1 𝑡 ∫ ∫ 3
0 0 𝑍
According to Lemma 2.8, we have
𝑡
1 𝐴
lim 𝑋(𝑠)d𝑠 = a.s.
𝑡→∞ 𝑡 ∫ 𝜇1
0
Now, applying the generalized Itô’s formula to equation (5) leads to
( ( ) )
[ ]
d𝑋 2 (𝑡) ≤ 2𝑋(𝑡) 𝐴 − 𝜇1 𝑋(𝑡) + 𝜎𝑋 ̄ 2 (𝑡) + 𝑋 2 (𝑡) 𝜂̄ 2 (𝑢) ∨ 𝜂 2 (𝑢) 𝜈(d𝑢) d𝑡
∫
𝑍
( )
+ 2𝑋(𝑡) 𝜎1 𝑆(𝑡)d1 (𝑡) + 𝜎2 𝐼(𝑡)d2 (𝑡) + 𝜎3 𝑄(𝑡)d3 (𝑡)
(( )2 )
+ 𝑋 2 (𝑡− ) 1 + 𝜂(𝑢)
̄ −1  ̃(d𝑡, d𝑢). (11)
∫
𝑍
Integrating both sides from 0 to 𝑡, yields
𝑡 ( ) 𝑡
[ 2 ]
𝑋 2 (𝑡) − 𝑋 2 (0) ≤ 2𝐴 𝑋(𝑠)d𝑠 − 2𝜇1 − 𝜎̄ − 𝜂̄ (𝑢) ∨ 𝜂 2 (𝑢) 𝜈(d𝑢) 𝑋 2 (𝑠)d𝑠
∫ ∫ ∫
0 𝑍 0
𝑡 𝑡 𝑡

+ 2𝜎1 𝑋(𝑠)𝑆(𝑠)d1 (𝑠) + 2𝜎2 𝑋(𝑠)𝐼(𝑠)d2 (𝑠) + 2𝜎3 𝑋(𝑠)𝑄(𝑠)d3 (𝑠)

∫ ∫ ∫
0 0 0
𝑡
(( )2 )
+ 𝑋 2 (𝑠− ) 1 + 𝜂(𝑢)
̄ ̃(d𝑠, d𝑢).
−1 
∫ ∫
0 𝑍
 11

[ ]
Let 𝜒 = 2𝜇1 − 𝜎̄ − ∫𝑍 𝜂̄ 2 (𝑢) ∨ 𝜂 2 (𝑢) 𝜈(d𝑢) > 0. Therefore
𝑡 𝑡 𝑡
1 2𝐴 𝑋 2 (0) − 𝑋 2 (𝑡) 2𝜎1
𝑋 2 (𝑠)d𝑠 ≤ 𝑋(𝑠)d𝑠 + + 𝑋(𝑠)𝑆(𝑠)d1 (𝑠)
𝑡∫ 𝜒𝑡 ∫ 𝜒𝑡 𝜒𝑡 ∫
0 0 0
𝑡 𝑡
2𝜎2 2𝜎
+ 𝑋(𝑠)𝐼(𝑠)d2 (𝑠) + 3 𝑋(𝑠)𝑄(𝑠)d3 (𝑠)
𝜒𝑡 ∫ 𝜒𝑡 ∫
0 0
𝑡
(( )2 )
1 ̃(d𝑠, d𝑢).
+ 𝑋 2 (𝑠) 1 + 𝜂(𝑢)
̄ −1 
𝜒𝑡 ∫ ∫
0 𝑍
By Lemma 2.5 and assumptions (A4 )-(A5 ), we can easily verify that
𝑡
1 2𝐴2
lim 𝑋 2 (𝑠)d𝑠 ≤ a.s.
𝑡→∞ 𝑡 ∫ 𝜇1 𝜒
0

Dissimilar from the Lyapunov approach, in this paper, we use the Feller property and mutually exclusive possibilities to derive
the condition for existence of the ergodic stationary distribution, which can close the gap left by using the Khasminskii method.
We begin with the definition of the stochastic Feller process.
Definition 2.12 (Feller property). Let  be a stochastic process that starts with initial value 𝑥.  is said to be a Feller process
[ ]
if, for any fixed 𝑡 ≥ 0 and any bounded, continuous and  -measurable function ℎ ∶ ℝ𝑛 → ℝ, 𝔼 ℎ(𝑡 ) depends continuously
upon 𝑥.
Now, we present a lemma which gives mutually exclusive possibilities for the existence of an ergodic stationary distribution
to the system (2).
Lemma 2.13 (Mutually exclusive possibilities lemma, 37 ). Let 𝜙(𝑡) ∈ ℝ𝑛 be a stochastic Feller process, then either an ergodic
probability measure exists, or
𝑡
1
lim sup ℙ(𝑠, 𝑥, Σ)𝜈(𝑑𝑥)d𝑠
̂ = 0, (12)
𝑡→∞ 𝜈̂ 𝑡 ∫ ∫
0
for any compact set Σ ⊂ ℝ , where the supremum is taken over all initial distributions 𝜈̂ on ℝ𝑛 and ℙ(𝑡, 𝑥, Σ) is the probability
𝑛

for 𝜙(𝑡) ∈ Σ with 𝜙(0) = 𝑥 ∈ ℝ𝑛 .

3 LONG-TERM DYNAMICS OF THE STOCHASTIC SIQS EPIDEMIC SYSTEM

3.1 Ergodicity and persistence in the mean

In the following, we aim to give the condition for the ergodicity the persistence of the disease. We suppose that 𝜒 > 0 and we
define the parameter:
( 𝜎 2 )−1 ⎛ 𝛽𝐴 𝐴2 𝜎𝛽2 ⎞
𝑠0 = 𝜇2 + 𝛿 + 𝛾 + 2 ⎜ − − 𝜂2 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢)⎟ .
2 ⎜ 𝜇1 𝜇1 𝜒 ∫ ⎟
⎝ 𝑍 ⎠
12

For simplicity, we introduce the following notations:

𝜇2 ( 𝜎 2 )( )
𝑀1 = 21 𝜇2 + 𝛿 + 𝛾 + 2 𝑠0 − 1 ,
4𝛽 𝐴 2
−(𝑝+1) ( 𝜎 2 )( )
𝑝𝜇1 Γ2,𝑝 𝛽
𝑀2 = 𝜇2 + 𝛿 + 𝛾 + 2 𝑠0 − 1 ,
8Δ 2
𝜇1 𝑞 ( 𝜎22 )( 2𝐴 )−1 ( )
𝑀3 = 𝜇2 + 𝛿 + 𝛾 + + 𝑁(0) 𝑠0 − 1 .
8𝛽 2 𝜇1
Theorem 3.1. If 𝑠0 > 1, the stochastic system (2) admits a unique stationary distribution and it has the ergodic property for
any initial value 𝑌 (0) ∈ ℝ3+ .

Proof. Motivated by the proof of Lemma 3.2 in 38 , we briefly verify the Feller property of the stochastic model (2). The main
purpose of the next step is to prove that (12) is impossible. Applying the generalized Itô’s formula to ln 𝐼 − 𝜇𝛽 (𝑋 − 𝑆), we easily
1
derive
{ } ( )
𝛽( ) 𝜎22 𝜎𝛽2 2
d ln 𝐼(𝑡) − 𝑋(𝑡) − 𝑆(𝑡) = 𝛽𝑆(𝑡) − (𝜇2 + 𝛿 + 𝛾) − − 𝑆 (𝑡) − 𝜂 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢) d𝑡
𝜇1 2 2 ∫ 2
𝑍
𝛽( )
− − 𝜇1 (𝑋(𝑡) − 𝑆(𝑡)) + 𝛽𝑆(𝑡)𝐼(𝑡) − 𝛾𝐼(𝑡) − 𝑘𝑄(𝑡) d𝑡 + 𝜎2 d2 (𝑡)
𝜇1
+ ̃(d𝑡, d𝑢) + 𝜎 𝑆(𝑡)d (𝑡) − 𝛽 𝜎 𝑆(𝑡)𝐼(𝑡)d (𝑡)
ln(1 + 𝜂2 (𝑢)) 𝛽 𝛽
∫ 𝜇1 𝛽 𝛽
𝑍
𝛽 ̄ 𝛽
−  (𝑡) − 3 (𝑡).
𝜇1 2 𝜇1
Then
{ } ( )
𝛽( ) 𝜎22 𝜎𝛽2
2
d ln 𝐼(𝑡) − 𝑋(𝑡) − 𝑆(𝑡) ≥ 𝛽𝑋(𝑡) − (𝜇2 + 𝛿 + 𝛾) − − 𝑆 (𝑡) − 𝜂2 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢) d𝑡
𝜇1 2 2 ∫
𝑍
𝛽2 ̃(d𝑡, d𝑢) + 𝜎 𝑆(𝑡)d (𝑡)
− 𝑆(𝑡)𝐼(𝑡)d𝑡 + 𝜎2 d2 (𝑡) + ln(1 + 𝜂2 (𝑢)) 𝛽 𝛽
𝜇1 ∫
𝑍
𝛽 𝛽 𝛽
− 𝜎𝛽 𝑆(𝑡)𝐼(𝑡)d𝛽 (𝑡) − ̄ 2 (𝑡) − 3 (𝑡). (13)
𝜇1 𝜇1 𝜇1
Integrating from 0 to 𝑡 on both sides of (13) yields
𝐼(𝑡) 𝛽( ) 𝛽( )
ln − 𝑋(𝑡) − 𝑆(𝑡) + 𝑋(0) − 𝑆(0)
𝐼(0) 𝜇1 𝜇1
𝑡 ( )
𝜎22 𝜎𝛽2 2
≥ 𝛽𝑋(𝑠) − (𝜇2 + 𝛿 + 𝛾) − − 𝑆 (𝑠) − 𝜂 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢) d𝑠
∫ 2 2 ∫ 2
0 𝑍
𝑡 𝑡 𝑡
𝛽2 ̃(d𝑠, d𝑢) + 𝜎
− 𝑆(𝑠)𝐼(𝑠)d𝑠 + 𝜎2 𝑊2 (𝑡) + ln(1 + 𝜂2 (𝑢)) 𝛽 𝑆(𝑠)d𝛽 (𝑠)
𝜇1 ∫ ∫ ∫ ∫
0 0 𝑍 0
𝑡 𝑡 𝑡
𝛽𝜎𝛽 𝛽𝜎2 𝛽
− 𝑆(𝑠)𝐼(𝑠)d𝛽 (𝑠) − 𝐼(𝑠)d2 (𝑠) − ̃(d𝑠, d𝑢)
𝜂 (𝑢)𝐼(𝑠− )
𝜇1 ∫ 𝜇1 ∫ 𝜇1 ∫ ∫ 2
0 0 0 𝑍
𝑡 𝑡
𝛽𝜎3 𝛽 ̃(d𝑠, d𝑢).
− 𝑄(𝑠)d3 (𝑠) − 𝜂 (𝑢)𝑄(𝑠− )
𝜇1 ∫ 𝜇1 ∫ ∫ 3
0 0 𝑍
 13

Hence
𝑡 𝑡 ( 𝜎𝛽2 )
𝜇 𝜎22
𝛽𝑆(𝑠)𝐼(𝑠)d𝑠 ≥ 1 𝛽𝑋(𝑠) − (𝜇2 + 𝛿 + 𝛾) − − 𝑆 2 (𝑠) − 𝜂2 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢) d𝑠
∫ 𝛽 ∫ 2 2 ∫
0 0 𝑍
𝑡
𝜇 𝐼(𝑡) ( ) ( ) 𝜇 𝜇 ̃(d𝑠, d𝑢)
− 1 ln + 𝑋(𝑡) − 𝑆(𝑡) − 𝑋(0) − 𝑆(0) + 1 𝜎2 𝑊2 (𝑡) + 1 ln(1 + 𝜂2 (𝑢))
𝛽 𝐼(0) 𝛽 𝛽 ∫ ∫
0 𝑍
𝑡 𝑡 𝑡 𝑡
𝜇1 𝜎𝛽
+ 𝑆(𝑠)d𝛽 (𝑠) − 𝜎𝛽 𝑆(𝑠)𝐼(𝑠)d𝛽 (𝑠) − 𝜎2 𝐼(𝑠)d2 (𝑠) − ̃(d𝑠, d𝑢)
𝜂2 (𝑢)𝐼(𝑠− )
𝛽 ∫ ∫ ∫ ∫ ∫
0 0 0 0 𝑍
𝑡 𝑡

− 𝛽𝜎3 𝑄(𝑠)d3 (𝑠) − ̃(d𝑠, d𝑢).

𝜂3 (𝑢)𝑄(𝑠− ) (14)
∫ ∫ ∫
0 0 𝑍
From Remark 2.6 and Lemma 2.8, one can derive that
𝑡
𝑋(𝑡) 𝑆(𝑡) 1 ̃(d𝑠, d𝑢)
lim = 0, lim = 0, and lim (𝜂2 (𝑢)𝐼(𝑠− ) + 𝜂3 (𝑢)𝑄(𝑠− )) a.s.
𝑡→∞ 𝑡 𝑡→∞ 𝑡 𝑡→∞ 𝑡 ∫ ∫
0 𝑍
Moreover,
𝑡 𝑡 𝑡
1 1 1
lim 𝑆(𝑠)d𝛽 (𝑠) = 0, 𝐼(𝑠)d2 (𝑠) = 0 and lim 𝑄(𝑠)d3 (𝑠) = 0 a.s.
𝑡→∞ 𝑡 ∫ 𝑡 ∫ 𝑡→∞ 𝑡∫
0 0 0
Application of the strong law of large numbers and assumption (A3 ) shows that
𝑡
𝑊 (𝑡) 1
lim 2 = 0 and lim ln(1 + 𝜂2 (𝑢))̃ (d𝑠, d𝑢) = 0 a.s.
𝑡→∞ 𝑡 𝑡→∞ 𝑡 ∫ ∫
0 𝑍
Applying similar arguments to those in the proof of Lemma 2.5, we obtain
𝑡
1
lim 𝑆(𝑠)𝐼(𝑠)d𝛽 (𝑠) = 0 a.s.
𝑡→∞ 𝑡 ∫
0

Since lim sup 1𝑡 ln 𝐼(𝑡)

𝐼(0)
≤ lim sup 1𝑡 ln 𝑁(𝑡)
𝐼(0)
≤ 0 a.s., one can derive that
𝑡→∞ 𝑡→∞
𝑡
1
lim inf 𝛽𝑆(𝑠)𝐼(𝑠)d𝑠
𝑡→∞ 𝑡 ∫
0
𝑡 ( 𝜎𝛽2 )
𝜇 1 𝜎22
≥ 1 lim inf 𝛽𝑋(𝑠) − (𝜇2 + 𝛿 + 𝛾) − − 𝑋 2 (𝑠) − 𝜂2 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢) d𝑠
𝛽 𝑡→∞ 𝑡 ∫ 2 2 ∫
0 𝑍
𝑡 𝑡
𝜇1 𝜎𝛽2 ( 𝜎2 )
1 1
= lim 𝛽𝑋(𝑠)d𝑠 − lim 𝑋 2 (𝑠)d𝑠 − 𝜇2 + 𝛿 + 𝛾 + 2 − 𝜂 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢).
𝛽 𝑡→∞ 𝑡 ∫ 2 𝑡→∞ 𝑡 ∫ 2 ∫ 2
0 0 𝑍
From Lemma 2.11, it follows that
𝑡 ( )
𝛽𝐴 𝐴 𝜎𝛽 ( 𝜎2 )
2 2
1 𝜇
lim inf 𝛽𝑆(𝑠)𝐼(𝑠)d𝑠 ≥ 1 × − − 𝜇2 + 𝛿 + 𝛾 + 2 − 𝜂 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢)
𝑡→∞ 𝑡 ∫ 𝛽 𝜇1 𝜇1 𝜒 2 ∫ 2
0 𝑍

𝜇 ( 𝜎 2 )( )
= 1 𝜇2 + 𝛿 + 𝛾 + 2 𝑠0 − 1 > 0 a.s. (15)
𝛽 2
14

To continue our analysis, we need to set the following subsets:

Ω1 = {(𝑆, 𝐼, 𝑄) ∈ ℝ3+ | 𝑆 ≥ 𝜖, and, 𝐼 ≥ 𝜖},
Ω2 = {(𝑆, 𝐼, 𝑄) ∈ ℝ3+ | 𝑆 ≤ 𝜖},
Ω3 = {(𝑆, 𝐼, 𝑄) ∈ ℝ3+ | 𝐼 ≤ 𝜖},
where 𝜖 > 0 is a positive constant to be determined later. Therefore, by (15), we get
𝑡
( )
1
lim inf 𝔼 𝛽𝑆(𝑠)𝐼(𝑠)𝟏Ω1 d𝑠
𝑡→∞ 𝑡 ∫
0
𝑡 𝑡 𝑡
( ) ( ) ( )
1 1 1
≥ lim inf 𝔼 𝛽𝑆(𝑠)𝐼(𝑠) d𝑠 − lim sup 𝔼 𝛽𝑆(𝑠)𝐼(𝑠)𝟏Ω2 d𝑠 − lim sup 𝔼 𝛽𝑆(𝑠)𝐼(𝑠)𝟏Ω3 d𝑠
𝑡→∞ 𝑡 ∫ 𝑡→∞ 𝑡 ∫ 𝑡→∞ 𝑡 ∫
0 0 0
𝑡 𝑡
𝜇1 ( 𝜎 2 )( )
1 [ ] 1 [ ]
≥ 𝜇2 + 𝛿 + 𝛾 + 2 𝑠0 − 1 − 𝛽𝜖lim sup 𝔼 𝐼(𝑠) d𝑠 − 𝛽𝜖lim sup 𝔼 𝑆(𝑠) d𝑠.
𝛽 2 𝑡→∞ 𝑡 ∫ 𝑡→∞ 𝑡 ∫
0 0
Then, one can see that
𝑡
( ) 𝜇 ( 𝜎 2 )( ) 2𝐴𝛽𝜖
1
lim inf 𝔼 𝛽𝑆(𝑠)𝐼(𝑠)𝟏Ω1 d𝑠 ≥ 1 𝜇2 + 𝛿 + 𝛾 + 2 𝑠0 − 1 − .
𝑡→∞ 𝑡 ∫ 𝛽 2 𝜇1
0
We can choose 𝜖 ≤ 𝑀1 , and then we obtain
𝑡
( ) 𝜇 ( 𝜎 2 )( )
1
lim inf 𝔼 𝛽𝑆(𝑠)𝐼(𝑠)𝟏Ω1 d𝑠 ≥ 1 𝜇2 + 𝛿 + 𝛾 + 2 𝑠0 − 1 > 0. (16)
𝑡→∞ 𝑡 ∫ 2𝛽 2
0
𝑎0 1 1
Let 𝑞 = 𝑎0 > 1 be a positive integer and 1 < 𝑝 = 𝑎0 −1
such that Γ2,𝑝 > 0 and 𝑞
+ 𝑝
= 1. By utilizing the Young inequality
𝑥𝑝 𝑦𝑞
𝑥𝑦 ≤ 𝑝
+ 𝑞
for all 𝑥,𝑦 > 0, we get

(
𝑡
) ( ) 𝑡
1 1 −1 𝑝 −1 −𝑞
lim inf 𝔼 𝛽𝑆(𝑠)𝐼(𝑠)𝟏Ω1 d𝑠 ≤ lim inf 𝔼 𝑝 (𝜂𝛽𝑆(𝑠)𝐼(𝑠)) + 𝑞 𝜂 𝟏Ω1 d𝑠
𝑡→∞ 𝑡 ∫ 𝑡→∞ 𝑡 ∫
0 0
𝑡 𝑡
( ) [ ]
1 1
≤ lim inf 𝔼 𝑞 −1 𝜂 −𝑞 𝟏Ω1 d𝑠 + 𝑝−1 (𝜂𝛽)𝑝 lim sup 𝔼 𝑁 2𝑝 (𝑠) d𝑠,
𝑡→∞ 𝑡 ∫ 𝑡→∞ 𝑡 ∫
0 0
where 𝜂 is a positive constant satisfying 𝜂 𝑝 ≤ 𝑀2 . By Lemma 2.2 and (16), we deduce that
𝑡 ( )
[ ] ( 𝜎22 )( 𝑠 ) 2𝜂 𝑝 𝛽 𝑝 Γ2,𝑝
1 𝑞 𝜇1
lim inf 𝔼 𝟏Ω1 d𝑠 ≥ 𝑞𝜂 𝜇 +𝛿+𝛾 + 0 − 1 −
𝑡→∞ 𝑡 ∫ 2𝛽 2 2 𝑝Δ
0
𝜇1 𝑞𝜂 𝑞 ( 𝜎 2 )( )
≥ 𝜇2 + 𝛿 + 𝛾 + 2 𝑠0 − 1 > 0. (17)
4𝛽 2
Setting Ω4 = {(𝑆, 𝐼, 𝑄) ∈ ℝ3+ | 𝑆 ≥ 𝜁, or, 𝐼 ≥ 𝜁} and Σ = {(𝑆, 𝐼, 𝑄) ∈ ℝ3+ | 𝜖 ≤ 𝑆 ≤ 𝜁, and, 𝜖 ≤ 𝐼 ≤ 𝜁} where 𝜁 > 𝜖 is a
positive constant to be explained in the following. By using the Tchebychev inequality, we can observe that
( )
1 1 2𝐴
𝔼[𝟏Ω4 ] ≤ ℙ(𝑆(𝑡) ≥ 𝜁) + ℙ(𝐼(𝑡) ≥ 𝜁) ≤ 𝔼[𝑆(𝑡) + 𝐼(𝑡)] ≤ + 𝑁(0) .
𝜁 𝜁 𝜇1
1
Choosing 𝜁
≤ 𝑀3 𝜂 𝑞 . We thus obtain
𝑡
1 𝜇 𝑞𝜂 𝑞 ( 𝜎 2 )( )
lim sup 𝔼[𝟏Ω4 ]d𝑠 ≤ 1 𝜇2 + 𝛿 + 𝛾 + 2 𝑠0 − 1 .
𝑡→∞ 𝑡 ∫ 8𝛽 2
0
 15

According to (17), one can derive that

𝑡 𝑡 𝑡
1 1 1
lim inf 𝔼[𝟏Σ ]d𝑠 ≥ lim inf 𝔼[𝟏Ω1 ]d𝑠 − lim sup 𝔼[𝟏Ω4 ]d𝑠
𝑡→∞ 𝑡 ∫ 𝑡→∞ 𝑡 ∫ 𝑡→∞ 𝑡 ∫
0 0 0
𝜇 𝑞𝜂 𝑞 ( 𝜎 2 )( )
≥ 1 𝜇2 + 𝛿 + 𝛾 + 2 𝑠0 − 1 > 0.
8𝛽 2
Based on the above analysis, we have determined a compact domain Σ ⊂ ℝ3+ such that
𝑡
( ) 𝜇 𝑞𝜂 𝑞 ( 𝜎 2 )( )
1
lim inf ℙ 𝑠, 𝑌 (0), Σ d𝑠 ≥ 1 𝜇2 + 𝛿 + 𝛾 + 2 𝑠0 − 1 > 0.
𝑡→∞ 𝑡 ∫ 8𝛽 2
0

Applying similar arguments to those in Theorem 5.1 of 28 , we show the uniqueness of the ergodic stationary distribution of our
model (2). This completes the proof.

Theorem 3.2. If 𝑠0 > 1, then for any value 𝑌 (0) ∈ ℝ3+ , the disease is persistent in the mean. That is to say
𝑡
1
lim inf 𝐼(𝑠)d𝑠 > 0 a.s.
𝑡→∞ 𝑡 ∫
0

Proof. From model (2) it yields

( )
d(𝑆(𝑡) + 𝐼(𝑡) + 𝑄(𝑡)) = 𝐴 − 𝜇1 𝑆(𝑡) − 𝜇2 𝐼(𝑡) − 𝜇3 𝑄(𝑡) d𝑡 + ̄ 1 (𝑡) + ̄ 2 (𝑡) + 3 (𝑡). (18)
Integrating (18) from 0 to 𝑡, and then dividing 𝑡 on both sides, we get
( )
1
(𝑆(𝑡) + 𝐼(𝑡) + 𝑄(𝑡)) − (𝑆(0) + 𝐼(0) + 𝑄(0))
𝑡
𝑡 𝑡 𝑡 𝑡 𝑡
𝜇 𝜇 𝜇 𝜎 1 ̃(d𝑠, d𝑢)
=𝐴− 1 𝑆(𝑠)d𝑠 − 2 𝐼(𝑠)d𝑠 − 3 𝑄(𝑠)d𝑠 + 1 𝑆(𝑠)d1 (𝑠) + 𝜂 (𝑢)𝑆(𝑠− )
𝑡 ∫ 𝑡 ∫ 𝑡 ∫ 𝑡 ∫ 𝑡∫ ∫ 1
0 0 0 0 0 𝑍
𝑠 𝑡 𝑡 𝑡

+
𝜎2
𝐼(𝑠)d2 (𝑠) +
1 ̃(d𝑠, d𝑢) + 𝜎3
𝜂 (𝑢)𝐼(𝑠− ) 𝑄(𝑠)d3 (𝑠) +
1 ̃(d𝑠, d𝑢).
𝜂 (𝑢)𝑄(𝑠− )
𝑡 ∫ 𝑡∫ ∫ 2 𝑡 ∫ 𝑡∫ ∫ 3
0 0 𝑍 0 0 𝑍
Taking the integration for the third equation of model (2) yields
𝑡 𝑡 𝑡 𝑡

𝑄(𝑡) − 𝑄(0) = 𝛿 𝐼(𝑠)d𝑠 − (𝜇3 + 𝑘) 𝑄(𝑠)d𝑠 + 𝜎3 𝑄(𝑠)d3 (𝑠) + ̃(d𝑠, d𝑢).

𝜂3 (𝑢)𝑄(𝑡− ) (19)
∫ ∫ ∫ ∫ ∫
0 0 0 0 𝑍
Dividing 𝑡 on both sides of equation (19), we have
𝑡 𝑡 𝑡
1 𝛿 1 𝜎3 1
𝑄(𝑠)d𝑠 = 𝐼(𝑠)d𝑠 + 𝑄(𝑠)d3 (𝑠)
𝑡∫ (𝜇3 + 𝑘) 𝑡 ∫ (𝜇3 + 𝑘) 𝑡 ∫
0 0 0
𝑡
1 ̃(d𝑠, d𝑢) − 1
+ 𝜂 (𝑢)𝑄(𝑠− ) (𝑄(𝑡) − 𝑄(0)).
(𝜇3 + 𝑘) ∫ ∫ 3 (𝜇3 + 𝑘)𝑡
0 𝑍
Then, one can obtain that
𝑡 ( ) 𝑡
1 𝐴 1 𝜇2 𝛿𝜇3
𝑆(𝑠)d𝑠 = − + 𝐼(𝑠)d𝑠 + Φ1 (𝑡), (20)
𝑡∫ 𝜇1 𝑡 𝜇1 𝜇1 (𝜇3 + 𝑘) ∫
0 0
16

where
𝑡 𝑡
𝜎3 𝜇3 𝜇3 ̃(d𝑠, d𝑢) − 1
Φ1 (𝑡) = 𝑄(𝑠)d3 (𝑠) + 𝜂 (𝑢)𝑄(𝑠− ) (𝑄(𝑡) − 𝑄(0))
𝜇1 (𝜇3 + 𝑘)𝑡 ∫ 𝜇1 (𝜇3 + 𝑘)𝑡 ∫ ∫ 3 (𝜇3 + 𝑘)𝑡
0 0 𝑍
𝑡 𝑡 𝑡
𝜎 1 ̃(d𝑠, d𝑢) + 𝜎2
+ 1 𝑆(𝑠)d1 (𝑠) + 𝜂 (𝑢)𝑆(𝑠− ) 𝐼(𝑠)d2 (𝑠)
𝜇1 𝑡 ∫ 𝜇1 𝑡 ∫ ∫ 1 𝜇1 𝑡 ∫
0 0 𝑍 0
𝑡 𝑡 𝑡

+
1 ̃(d𝑠, d𝑢) + 𝜎3
𝜂2 (𝑢)𝐼(𝑠− ) 𝑄(𝑠)d3 (𝑠) +
1
𝜂 (𝑢)𝑄(𝑠− )̃(d𝑠, d𝑢)
𝜇1 𝑡 ∫ ∫ 𝜇1 𝑡 ∫ 𝜇1 𝑡 ∫ ∫ 3
0 𝑍 0 0 𝑍
( )
1
− (𝑆(𝑡) + 𝐼(𝑡) + 𝑄(𝑡)) − (𝑆(0) + 𝐼(0) + 𝑄(0)) .
𝜇1 𝑡
Applying Itô’s formula to the second equation of (2), we get
( )
𝜎22 𝜎𝛽2 2
d ln 𝐼(𝑡) = 𝛽𝑆(𝑡) − (𝜇2 + 𝛿 + 𝛾) − − 𝑆 (𝑡) − 𝜂 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢) d𝑡
2 2 ∫ 2
𝑍

+ 𝜎2 d2 (𝑡) + ̃(d𝑡, d𝑢) + 𝜎 𝑆(𝑡)d (𝑡).

ln(1 + 𝜂2 (𝑢)) (21)
𝛽 𝛽
∫
𝑍
Integrating (21) from 0 to 𝑡 and then dividing 𝑡 on both sides, we have
𝑡 𝑡
1 𝛽 𝜎22 𝜎𝛽2
(ln 𝐼(𝑡) − ln 𝐼(0)) = 𝑆(𝑠)d𝑠 − (𝜇2 + 𝛿 + 𝛾) − − 𝑆 2 (𝑠)d𝑠 − 𝜂 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢)
𝑡 𝑡 ∫ 2 2𝑡 ∫ ∫ 2
0 0 𝑍
𝑡 𝑡
𝑊2 (𝑡) 1 𝜎
+ 𝜎2 + ̃(d𝑠, d𝑢) + 𝛽
ln(1 + 𝜂2 (𝑢)) 𝑆(𝑠)d𝛽 (𝑠).
𝑡 𝑡 ∫ ∫ 𝑡 ∫
0 𝑍 0
From (20), we get
( ) 𝑡
1 𝛽𝐴 𝛽 𝜇2 𝛿𝜇3
(ln 𝐼(𝑡) − ln 𝐼(0)) = − + 𝐼(𝑠)d𝑠 + 𝛽Φ1 (𝑡) − (𝜇2 + 𝛿 + 𝛾)
𝑡 𝜇1 𝑡 𝜇1 𝜇1 (𝜇3 + 𝑘) ∫
0
𝑡
𝜎22 𝜎𝛽2
− − 𝑆 2 (𝑠)d𝑠 − 𝜂2 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢)
2 2 ∫ ∫
0 𝑍
𝑡 𝑡
𝑊 (𝑡) 1 𝜎
+ 𝜎2 2 + ̃(d𝑠, d𝑢) + 𝛽
ln(1 + 𝜂2 (𝑢)) 𝑆(𝑠)d𝛽 (𝑠).
𝑡 𝑡∫ ∫ 𝑡 ∫
0 𝑍 0
Since 𝑆(𝑡) ≤ 𝑋(𝑡) a.s., we obtain
( ) 𝑡
1 𝛽𝐴 𝛽 𝜇2 𝛿𝜇3
(ln 𝐼(𝑡) − ln 𝐼(0)) ≥ − + 𝐼(𝑠)d𝑠 + 𝛽𝜙1 (𝑡) − (𝜇2 + 𝛿 + 𝛾)
𝑡 𝜇1 𝑡 𝜇1 𝜇1 (𝜇3 + 𝑘) ∫
0
𝑡
𝜎22 𝜎𝛽2
− − 𝑋 2 (𝑠)d𝑠 − 𝜂2 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢)
2 2 ∫ ∫
0 𝑍
𝑡 𝑡
𝑊 (𝑡) 1 𝜎
+ 𝜎2 2 + ̃(d𝑠, d𝑢) + 𝛽
ln(1 + 𝜂2 (𝑢)) 𝑆(𝑠)d𝛽 (𝑠).
𝑡 𝑡∫ ∫ 𝑡 ∫
0 𝑍 0
 17

Hence, we further have

( ) 𝑡
𝛽 𝜇2 𝛿𝜇3 1 𝛽𝐴
+ 𝐼(𝑠)d𝑠 ≥ − (ln 𝐼(𝑡) − ln 𝐼(0)) + + 𝛽𝜙1 (𝑡) − (𝜇2 + 𝛿 + 𝛾)
𝑡 𝜇1 𝜇1 (𝜇3 + 𝑘) ∫ 𝑡 𝜇1
0
𝑡
𝜎22 𝜎𝛽2
− − 𝑋 2 (𝑠)d𝑠 − 𝜂2 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢)
2 2 ∫ ∫
0 𝑍
𝑡 𝑡
𝑊 (𝑡) 1 𝜎
+ 𝜎2 2 + ̃(d𝑠, d𝑢) + 𝛽
ln(1 + 𝜂2 (𝑢)) 𝑆(𝑠)d𝛽 (𝑠).
𝑡 𝑡∫ ∫ 𝑡 ∫
0 𝑍 0
By assumption (𝐴3 ), Lemmas 2.8 - 2.11, and the large number theorem for martingales, we can easily verify that
𝑡 ( )−1
𝛿𝜇3 ( 𝜎2 )
1 1 𝜇2
lim inf 𝐼(𝑠)d𝑠 ≥ + 𝜇2 + 𝛿 + 𝛾 + 2 (𝑠0 − 1) > 0 a.s.
𝑡→∞ 𝑡 ∫ 𝛽 𝜇1 𝜇1 (𝜇3 + 𝑘) 2
0
This shows that the system (2) is persistent in the mean with probability one. This completes the proof.

3.2 The extinction of the disease

Now, we will give the result on the extinction of the disease. Define
( ( )
𝜎22 )−1 𝛽𝐴 𝜎𝛽2 𝐴2
̂ 𝑠
0 = 𝜇2 + 𝛿 + 𝛾 + − − 𝜂 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢) .
2 𝜇1 2𝜇12 ∫ 2
𝑍

Theorem 3.3. Let 𝑌 (𝑡) be the solution of system (2) with initial value 𝑌 (0) ∈ ℝ3+ .
If
𝜇𝛽
̂ 𝑠 < 1 and 𝜎 2 ≤ 1 , (22)
0 𝛽
𝐴
or
𝛽2 ( 𝜎22 )
− 𝜇 2 + 𝛿 + 𝛾 + − 𝜂 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢) < 0, (23)
2𝜎 2 2 ∫ 2
𝛽 𝑍
then the disease dies out exponentially with probability one. That is to say,
ln 𝐼(𝑡)
lim sup < 0 a.s. (24)
𝑡→∞ 𝑡

Proof. By Itô’s formula for all 𝑡 ≥ 0, we have

( )
𝜎22 𝜎𝛽2
d ln 𝐼(𝑡) = 𝛽𝑆(𝑡) − (𝜇2 + 𝛿 + 𝛾) − − 𝑆 2 (𝑡) − 𝜂2 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢) d𝑡
2 2 ∫
𝑍

+ 𝜎2 d2 (𝑡) + ̃(d𝑡, d𝑢) + 𝜎 𝑆(𝑡)d (𝑡).

ln(1 + 𝜂2 (𝑢)) (25)
𝛽 𝛽
∫
𝑍
Integrating (25) from 0 to 𝑡 and then dividing 𝑡 on both sides, we get
𝑡 𝑡
ln 𝐼(𝑡) 𝛽 ( 𝜎2 ) 𝜎𝛽2
= 𝑆(𝑠)d𝑠 − 𝜇2 + 𝛿 + 𝛾 + 2 − 𝜂2 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢) − 𝑆 2 (𝑠)d𝑠 + Φ2 (𝑡), (26)
𝑡 𝑡 ∫ 2 ∫ 2𝑡 ∫
0 𝑍 0
where
𝑡 𝑡
𝜎𝛽 𝜎 𝑊 (𝑡) 1
Φ2 (𝑡) = 𝑆(𝑠)d𝛽 (𝑠) − 2 2 + ̃(d𝑠, d𝑢) − ln 𝐼(0) .
ln(1 + 𝜂2 (𝑢))
𝑡 ∫ 𝑡 𝑡∫ ∫ 𝑡
0 0 𝑍
18

Obviously, we know that

𝑡 𝑡
( )2
1 1
𝑆 2 (𝑠)d𝑠 ≥ 𝑆(𝑠)d𝑠 .
𝑡∫ 𝑡∫
0 0
Therefore, from (20), we derive
𝑡 𝑡
ln 𝐼(𝑡) 𝛽 ( 𝜎2 ) 𝜎𝛽2 ( 1 )2
≤ 𝑆(𝑠)d𝑠 − 𝜇2 + 𝛿 + 𝛾 + 2 − 𝜂2 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢) − 𝑆(𝑠)d𝑠 + Φ2 (𝑡)
𝑡 𝑡 ∫ 2 ∫ 2 𝑡∫
0 𝑍 0
( ( ) 𝑡 )
𝜇2 𝛿𝜇3 ( 𝜎2 )
𝐴 1
=𝛽 − + 𝐼(𝑠)d𝑠 + 𝜙1 (𝑡) − 𝜇2 + 𝛿 + 𝛾 + 2 − 𝜂 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢)
𝜇1 𝑡 𝜇1 𝜇1 (𝜇3 + 𝑘) ∫ 2 ∫ 2
0 𝑍
2
( ( ) 𝑡 )2
𝜎𝛽 𝐴 1 𝜇2 𝛿𝜇3
− − + 𝐼(𝑠)d𝑠 + 𝜙1 (𝑡) + Φ2 (𝑡).
2 𝜇1 𝑡 𝜇1 𝜇1 (𝜇3 + 𝑘) ∫
0
Hence, one can see that
ln 𝐼(𝑡) 𝛽𝐴 ( 𝜎2 ) 𝐴2 𝜎𝛽2
≤ − 𝜇2 + 𝛿 + 𝛾 + 2 − 𝜂2 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢) −
𝑡 𝜇1 2 ∫ 2𝜇12
𝑍
( )(
𝐴𝜎𝛽2 ) 1
𝑡
𝜇2 𝛿𝜇3
− + 𝛽− 𝐼(𝑠)d𝑠
𝜇1 𝜇1 (𝜇3 + 𝑘) 𝜇1 𝑡∫
0
(( ) 𝑡 )2
𝜎𝛽2 𝜇2 𝛿𝜇3
− + 𝐼(𝑠)d𝑠 + Φ2 (𝑡) + Φ3 (𝑡), (27)
2𝑡2 𝜇1 𝜇1 (𝜇3 + 𝑘) ∫
0
where
𝜎𝛽2 𝜎𝛽2 𝐴Φ1 (𝑡) ( ) 𝑡
𝜇2 𝛿𝜇3
Φ3 (𝑡) = 𝛽Φ1 (𝑡) − Φ21 (𝑡) − + 𝜎𝛽2 Φ1 (𝑡) + 𝐼(𝑠)d𝑠.
2 𝜇1 𝜇1 𝜇1 (𝜇3 + 𝑘) ∫
0
Based on Lemma 2.8, one has
Φ2 (𝑡) Φ (𝑡)
= lim 3 = 0 a.s.
lim
𝑡→∞ 𝑡 𝑡→∞ 𝑡
Taking the superior limit on both sides of (27), then by condition (22), we arrive at
ln 𝐼(𝑡) ( 𝜎 2 )( 𝑠 )
lim sup ≤ 𝜇2 + 𝛿 + 𝛾 + 2  ̂ − 1 < 0 a.s.
0
𝑡→∞ 𝑡 2
Now, from (26), we have
𝑡 𝑡
ln 𝐼(𝑡) 𝛽 ( 𝜎2 ) 𝜎𝛽2
= 𝑆(𝑠)d𝑠 − 𝜇2 + 𝛿 + 𝛾 + 2 − 𝜂2 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢) − 𝑆 2 (𝑠)d𝑠 + Φ2 (𝑡)
𝑡 𝑡 ∫ 2 ∫ 2𝑡 ∫
0 𝑍 0

(
𝑡 ( )2
𝛽2 𝜎22 ) 𝜎𝛽2
1 𝛽
= − 𝜇2 + 𝛿 + 𝛾 + − 𝜂2 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢) − 𝑆(𝑠)d𝑠 − 𝑑𝑠 + Φ2 (𝑡)
2𝜎𝛽2 2 ∫ 2 𝑡∫ 𝜎𝛽2
𝑍 0

𝛽 2 ( 𝜎22 )
≤ − 𝜇2 + 𝛿 + 𝛾 + − 𝜂2 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢)) + Φ2 (𝑡).
2𝜎𝛽2 2 ∫
𝑍
By the large number theorem for martingales, Lemma 2.8 and the condition (23), our desired result (24) holds true. This
completes the proof.
 19

4 EXAMPLES

In this section, we will validate our theoretical results with the help of numerical simulation examples taking parameters from
the theoretical data mentioned in the Table 1 . To construct a discrete-time approximation of our model, we use the well-known
Euler scheme for SDE with jumps. A number of papers have been devoted to investigating the Euler scheme for the stochastic
differential equations driven by Lévy processes (see for example 39,40 ). Furthermore, we numerically simulate the solution of
system (2) with the initial values (𝑆(0), 𝐼(0), 𝑄(0)) = (0.5, 0.3, 0.1). The unit of time is one day.

180 250

160

200
140

120
150
Density

Density
100

80
100
60

40
50

20

0 0
1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
S(300) I(300)

200

180

160

140

120
Density

100

80

60

40

20

0
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08
Q(300)

FIGURE 2 Histogram of the probability density function for 𝑆, 𝐼, and 𝑄 population at 𝑡 = 300 for the stochastic model (2),
the smoothed curves are the probability density functions of 𝑆(𝑡), 𝐼(𝑡) and 𝑄(𝑡), respectively.

Parameters Description Value

𝐴 The recruitment rate 0.1
𝜇1 The natural mortality rate 0.05
𝜇2 The mortality rate of 𝐼 0.09
𝜇3 The mortality rate of 𝑄 0.052
𝛽 The transmission rate 0.075
𝛿 The isolation rate 0.03
𝛾 The recovered rate of 𝐼 0.01
𝑘 The recovered rate of 𝑄 0.04

TABLE 1 Some theoretical parameter values of the model (2).

20

2
S(t)
I(t)
1.8
Q(t)

1.6

1.4

1.2

0.8

0.6

0.4

0.2

0
0 50 100 150 200 250 300
time

FIGURE 3 The paths of 𝑆(𝑡), 𝐼(𝑡) and 𝑄(𝑡) for the stochastic model (2) with initial values (𝑆(0), 𝐼(0), 𝑄(0)) = (0.5, 0.3, 0.1).

I(t) I(t)
0.25 0.2

0.18

0.2 0.16

0.14

0.15 0.12

0.1

0.1 0.08

0.06

0.05 0.04

0.02

0 0
0 50 100 150 200 250 300 0 50 100 150 200 250 300
t t

(a) (b)

FIGURE 4 The numerical simulation of 𝐼(𝑡) in the system (2).

Example 4.1. We have chosen the stochastic fluctuations intensities 𝜎1 = 0.01, 𝜎2 = 0.03, 𝜎3 = 0.07 and 𝜎𝛽 = 0.02. Further-
more, we assume that 𝜂1 (𝑢) = 0.01, 𝜂2 (𝑢) = 0.02, 𝜂3 (𝑢) = 0.05, 𝑍 = (0, ∞) and 𝜈(𝑍) = 1. Then, 𝑠0 = 1.1756 > 1. From
Figure 2 , we show the existence of the unique stationary distributions for 𝑆(𝑡), 𝐼(𝑡) and 𝑄(𝑡) of the model (2) at 𝑡 = 300, where
the smooth curves are the probability density functions of 𝑆(𝑡), 𝐼(𝑡) and 𝑄(𝑡), respectively. It can be obviously observed that the
solution of the stochastic model (2) persists in the mean (see Figure 3 ).
Example 4.2. Now, we choose the white noise intensities 𝜎2 = 0.12 and 𝜎𝛽 = 0.1 to ensure that the condition (23) of theorem
(3.3) is satisfied. We can conclude that for any initial value, 𝐼(𝑡) obeys
𝛽2 ( 𝜎2 )
1 𝐼(𝑡)
lim sup ln ≤ 2 − 𝜇2 + 𝛿 + 𝛾 + 2 − 𝜂 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢),
𝑡→∞ 𝑡 𝐼(0) 2𝜎 2 ∫ 2
𝛽 𝑍
= −0.1374 < 0 a.s.
That is, 𝐼(𝑡) will tend to zero exponentially with probability one (see Figure 4 (a)). To verify that the condition (22) is satisfied,
we change 𝜎2 to 0.01, 𝜎𝛽 to 0.02 and 𝛽 to 0.05 and keep other parameters unchanged. Then we have
( ( )
𝜎22 )−1 𝛽𝐴 𝜎𝛽2 𝐴2
̂ 𝑠
0 = 𝜇2 + 𝛿 + 𝛾 + − − 𝜂 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢) = 0.7650 < 1,
2 𝜇1 2𝜇12 ∫ 2
𝑍
 21

and
𝜇1 𝛽
𝜎𝛽2 − = −0.0249 < 0.
𝐴
Therefore, the condition (22) of Theorem 3.3 is satisfied. We can conclude that for any initial value, 𝐼(𝑡) obeys
( 𝜎 2 )( 𝑠 )
1 𝐼(𝑡)
lim sup ln ≤ 𝜇2 + 𝛿 + 𝛾 + 2  ̂ − 1 = −0.0306 < 0 a.s.
0
𝑡→∞ 𝑡 𝐼(0) 2
That is, 𝐼(𝑡) will tend to zero exponentially with probability one (see Figure 4 (b)).

CONCLUSION AND DISCUSSION

Epidemic models provide the frameworks that allow ideas about the behavior of an infectious disease to be conceptualized and
communicated. There is a necessity for epidemic center managers and policymakers to these models in order to comprehend,
examine and evaluate requisite approaches to control the illness. All models are, by nature, the facilitation and simplification of
more complex systems. Epidemic models can be classified into diverse categories depending on their treatment of stochasticity
or variability, time, space, and the structure of the population. The approaches will differ from simple deterministic mathematical
models through to general stochastic framework. The major feature of using perturbed models in human health is as an appro-
priate tool to aid retrospective analysis of some environmental effects to gain an understanding of their impact on the disease
outbreak. By allowing various types of perturbation to be combined in a structured way, hypothetical scenarios can be sophisti-
cated to provide insights into the merits of different strategies in different real situations. Stochastic modeling can contribute to
better disease control through:
∙ Reconnaissance of different strategies in a normal and presumptive outbreak (contingency planning).
∙ Appraisal of the effectiveness of various surveillance strategies in real situations.
∙ Provision of actual scenarios for future probably situations under any sudden environmental cases.
In this study, we proposed a new version of a perturbed SIS epidemiological model with a quarantine strategy. This model
simultaneously takes into account random transmission and the effects of jumps. We have addressed possible scenarios of
the pandemic spread during unforeseen climate changes or environmental shocks. Compared with the existing literature, the
novelty of our study manifested in new analysis techniques and improvements which are summarized in the following items:

∙ Our paper is distinguished from previous works 23,30,31,32,33 by improving the majorization of the following quantity
[( ) ]
𝑛𝑝 𝑋̃ 𝑛𝑝 𝑋̃
𝑁 (𝑡) 1 + − 1 − 𝑛𝑝 𝜈(d𝑢),
∫ 𝑁 𝑁
𝑍
which raises the optimality of our calculus and results.
∙ Our results in Lemmas 2.5 and 2.8 provide an extended and generalized version of classical lemmas 3.3 and 3.4 presented
in 30 which are widely used in the literature.
∙ Our study provides an improved threshold
( 𝜎 2 )−1 ⎛ 𝛽𝐴 𝐴2 𝜎𝛽2 ⎞
𝑠0 = 𝜇2 + 𝛿 + 𝛾 + 2 ⎜ − − 𝜂2 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢)⎟ ,
2 ⎜ 𝜇1 𝜇1 𝜒 ∫ ⎟
⎝ 𝑍 ⎠
by taking into consideration the Remark 2.3. This parameter is a sufficient condition for the existence of a unique ergodic
stationary distribution and persistence of the disease under some assumptions. The last two asymptotic properties are
proven in Theorems 3.1 and 3.2, by using a new approach based on Lemma 2.11 and the mutually exclusive possibilities
lemma 2.13.
∙ Our study offers an alternative method to the gap mentioned in (Theorem 2.2, 25 ). Without using the explicit formula of the
distribution stationary 𝜇(⋅) of 𝑋 (which still up to now unknown), we gave the expression of the ergodicity and persistence
threshold.
22

∙ For the case of non-persistence, in Theorem 3.3, we proved that the following parameter
( ( )
𝜎22 )−1 𝛽𝐴 𝐴2 𝜎𝛽2
̂ 𝑠
0 = 𝜇2 + 𝛿 + 𝛾 + − − 𝜂 (𝑢) − ln(1 + 𝜂2 (𝑢))𝜈(d𝑢) ,
2 𝜇1 2𝜇2 ∫ 2
1 𝑍
is a sufficient conditions for the disappearance of the disease.

Eventually, we point out that the obtained results extend and generalize many previous works (for example, 19,20,21,24,27 ), by
analyzing the dynamics of the SIQS epidemic models with two disturbances. We believe that our article can be a rich basis for
future studies. However, some limitations in this paper should be stated. The actual application of our new stochastic model has
not yet been explored in detail, and the model has not been validated by real data. It remains just an abstract framework and it
has the ability to apply it for simulating theoretically the current diseases. More advanced statistical techniques are required to
estimate the diffusion and jump intensities in our model. Standard empirical investigations of jump dynamics and volatility are
completely complicated due to the presence of diverse potential factors in continuous time. This doesn’t make us give up and
our work can be extended in other different ways. Generally, the parameters of the model in this paper can be estimated by using
a long time series of cross-sections of real data, and the findings can be compared with the theoretical results in this paper. We
treat these ideas in our outlook works.

Author contributions
The authors declare that the study was conducted in collaboration with the same responsibility. All authors read and approved
the final manuscript.

Financial disclosure
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Conflict of interest
The authors declare that the study was conducted in collaboration with the same responsibility. All authors read and approved
the final manuscript.

References

1. Kermack W O, McKendrick A G. A contribution to the mathematical theory of epidemics. Proceedings of The Royal Society
A Mathematical Physical and Engineering Sciences. 1927;115(772):700–721.

2. Beretta E, Hara T, Ma W. Global asymptotic stability of an SIR epidemic model with distributed time delay. Nonlinear
Analysis: Theory, Methods and Applications. 2001;47:4107–4115.

3. Guo H, Li M Y, Shuai Z S. Global stability of the endemic equilibrium of multigroup SIR epidemic models. Canadian
Applied Mathematics Quarterly. 2006;14:259–284.

4. Meng X Z, Chen L S. The dynamics of a new SIR epidemic model concerning pulse vaccination strategy. Applied
Mathematics and Computation. 2008;197:528–597.

5. Roy M, Holt R D. Effects of predation on host-pathogen dynamics in SIR models. Theoretical Population Biology.
2008;73:319–331.

6. Tang B, Xia F, Tang S, et al. The effectiveness of quarantine and isolation determine the trend of the COVID-19 epidemics
in the final phase of the current outbreak in China. International Journal of Infectious Diseases. 2020;95:288-293.

7. Khan M A, Atangana A, Alzahrani E, Fatmawati . The dynamics of COVID-19 with quarantined and isolation. Advances
in Difference Equations. 2020;425.
 23

8. Safi M A, Gumel A B. Global asymptotic dynamics of a model for quarantine and isolation. Discrete and Continuous
Dynamical Systems B. 2010;14:209–231.

9. Sun C, Yang W. Global results for an SIRS model with vaccination and isolation. Nonlinear Analysis Real World
Applications. 2010;11:4223–4237.

10. Safi M A, Gumel A B. The effect of incidence function on the dynamics of a qrarantine/isolation model with time delay.
Nonlinear Analysis Real World Applications. 2011;12:215–235.

11. Herbert H, Ma Z, Liao S. Effects of quarantine in six endemic models for infectious diseases. Mathematical Biosciences.
2002;180:141–160.

12. Allen L J S. An introduction to stochastic epidemic models. Mathematical Epidemiology. 2008;144:81–130.

13. Lin Y, Jiang D, Xia P. Long-time behavior of a stochastic SIR model. Applied Mathematics and Computation. 2014;236:1–9.

14. Ji C, Jiang D, Shi N. Asymptotic behavior of global positive solution to a stochastic SIR model. Applied Mathematical
Modelling. 2011;45:221–232.

15. Ji C, Jiang D. Threshold behaviour of a stochastic SIR model. Applied Mathematical Modelling. 2014;38:5067–5079.

16. Ji C, Jiang D, Shi N. The behavior of an SIR epidemic model with stochastic perturbation. Stochastic analysis and
applications. 2012;30:755–773.

17. Tornatore E, Buccellato S, Vetro P. Stability of a stochastic SIR system. Physica A. 2005;354:111–126.

18. Christen A, Yanez MA Maulen, Valencia Y, Gonzalez E Olivares, Rial DF, Cure M. Linear incidence rate: Its influence on
the asymptotic behavior of a stochastic epidemic model. Mathematical Methods in the Applied Sciences. 2020;2020:1-17.

19. Zhang X, Huo H, Xiang H, Meng X. Dynamics of the deterministic and stochastic SIQS epidemic model with nonlinear
incidence. Applied Mathematics and Computation. 2014;243:546–558.

20. Zhang X, Huo H, Xiang H, Shi Q, Li D. The threshold of a stochastic SIQS epidemic model. Physica A. 2017;482:362–374.

21. Wei F, Chen F. Stochastic permanence of an SIQS epidemic model with saturated incidence and independent random
perturbations. Physica A. 2016;453:99–107.

22. Zhang X, Wang K. Stochastic SIR model with jumps. Applied Mathematics letters. 2013;826:867–874.

23. Zhou Y, Zhang W. Threshold of a stochastic SIR epidemic model with Levy jumps. Physica A. 2016;446:204–2016.

24. Zhang X, Shi Q, Ma S, Huo H, Li D. Dynamic behavior of a stochastic SIQS epidemic model with Levy jumps. Nonlinear
Dynamics. 2018;93:1481–1493.

25. Zhao D, Yuan S, Liu H. Stochastic dynamics of the delayed chemostat with Levy noises. International Journal of
Biomathematics. 2019;12(5).

26. Yang Q, Jiang D, Shi N, Ji C. The ergodicity and extinction of stochastically perturbed SIR and SEIR epidemic models with
saturated incidence. Journal of Mathematical Analysis and Applications. 2012;388:248–271.

27. Chen Y, Wen B, Teng Z. The global dynamics for a stochastic SIS epidemic model with isolation. Physica A. 2018;492:1604-
1624.

28. Khasminskii R. Stochastic stability of differential equations. A Monographs and Textbooks on Mechanics of Solids and
Fluids. 1980;7.

29. Kiouach D, Sabbar Y. Stability and Threshold of a Stochastic SIRS Epidemic Model with Vertical Transmission and Transfer
from Infectious to Susceptible Individuals. Discrete Dynamics in Nature and Society. 2018;(7570296).

30. Zhou Y, Yuan S, Zhao D. Threshold behavior of a stochastic SIS model with Levy jumps. Applied Mathematics and
Computation. 2016;275:255–267.
24

31. Cheng Y, Zhang F, Zhao M. A stochastic model of HIV infection incorporating combined therapy of HAART driven by
Levy jumps. Advance in Difference Equations. 2019;321.

32. Cheng Y, Li M, Zhang F. A dynamics stochastic model with HIV infection of CD4 T cells driven by Levy noise. Chaos,
Solitons and Fractals. 2019;129:62–70.

33. Gao M, Jiang D, Hayat T, Alsaedi A. Threshold behavior of a stochastic Lotka Volterra food chain chemostat model with
jumps. Chaos, Solitons and Fractals. 2019;523:191–203.

34. Zhao D, Yuan S. Sharp conditions for the existence of a stationary distribution in one classical stochastic chemostat. Applied
Mathematics and Computation. 2018;339:199-205.

35. Mao X. Stochastic Differential Equations and Applications. Chichester: Horwoodl; 1997.

36. Wang Y, Jiang D. Stationary Distribution and Extinction of a Stochastic Viral Infection Model. Discrete Dynamics in Nature
and Society. ID 6027509, 2017;2017.

37. Stettner L. On the existence and uniqueness of invariant measure for continuous-time Markov processes. Technical Report,
LCDS, Brown University, province, RI. 1986;:18–86.

38. Tong J, Zhang Z, Bao J. The stationary distribution of the facultative population model with a degenerate noise. Statistics
and Probability Letters. 2013;83(14):655–664.

39. Protter P, Talay D. The Euler scheme for Lévy driven stochastic differential equations. Ann Probab. 1997;2020:393–423.

40. Rubenthaler S. Numerical simulation of the solution of a stochastic differential equation driven by a Lévy process. Stoch
Proc Appl. 2003;103:311–349.