Download as pdf or txt
Download as pdf or txt
You are on page 1of 23

Journal of Biological Dynamics

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/tjbd20

Survival analysis of a stochastic predator–prey


model with prey refuge and fear effect

Yixiu Xia & Sanling Yuan

To cite this article: Yixiu Xia & Sanling Yuan (2020) Survival analysis of a stochastic
predator–prey model with prey refuge and fear effect, Journal of Biological Dynamics, 14:1,
871-892, DOI: 10.1080/17513758.2020.1853832

To link to this article: https://doi.org/10.1080/17513758.2020.1853832

© 2020 The Author(s). Published by Informa


UK Limited, trading as Taylor & Francis
Group

Published online: 03 Dec 2020.

Submit your article to this journal

Article views: 894

View related articles

View Crossmark data

Citing articles: 15 View citing articles

Full Terms & Conditions of access and use can be found at


https://www.tandfonline.com/action/journalInformation?journalCode=tjbd20
JOURNAL OF BIOLOGICAL DYNAMICS
2020, VOL. 14, NO. 1, 871–892
https://doi.org/10.1080/17513758.2020.1853832

Survival analysis of a stochastic predator–prey model with


prey refuge and fear effect
Yixiu Xia and Sanling Yuan
College of Science, University of Shanghai for Science and Technology, Shanghai, China

ABSTRACT ARTICLE HISTORY


In this paper, we propose and investigate a stochastic Holling type-II Received 14 January 2020
predator–prey model with prey refuge and fear effect. We first prove Accepted 10 November 2020
the existence and uniqueness of the global positive solution. Then KEYWORDS
we perform the survival analysis of the model, including the exis- Stochastic predator–prey
tence of a unique ergodic stationary distribution and the extinction model; fear effect; prey
of the model. Numerical simulations are carried out to validate our refuge; stationary
analytical results. Our findings indicate that the white noise is adverse distribution; extinction
to the growth of predator and prey populations, and the increase of
MSC SUBJECT
fear effect will lead to the decrease of predator density, but with no CLASSIFICATIONS
obvious effect on prey density. 60H10; 92D25

1. Introduction
Population models, as an important part of ecology, have been extensively studied and
explored for their rich dynamic behaviour with the aim to provide a theoretical guidance
for the protection, development and utilization of biological resources [2,5]. Among the
most important population models, predator–prey models play an important role in under-
standing the interaction of different species in volatile natural environments and have been
widely investigated [1,4,9,12,30,33].
As is well known, it is common for predators to hunt preys in nature. However, not all
preys are captured by the predators because the preys usually have refuges where they can
evade the predation [10,18,27]. This indicates that a prey refuge can protect and prevent
the extinction of preys, and thus, it plays an important role in the interaction between
predator and prey populations. Recently, the research of the predator–prey models with
prey refuges has attracted many mathematicians’ attention and some interesting results
have been obtained [19,23,25,32].
Notice that in most previous predator–prey models, only the direct killing of preys in
the presence of predators is considered due to its obviousness in nature. But in reality, the
preys may change their behaviour when the predator they faced is powerful [13,26,28].
Some studies have shown that all preys respond to the risk of predation and show all
sorts of anti-predator responses, including habitat changes, foraging, vigilance and dif-
ferent physiological changes. Such anti-predator behaviour can reduce the long-term cost

CONTACT Sanling Yuan sanling@usst.edu.cn


© 2020 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/
licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
872 Y. XIA AND S. YUAN

of reproduction and increase the probability of adult survival [4]. In addition, frightened
preys often forage less, so their birth rate decline and they use some survival mechanisms
like starvation [3].
Considering the effect of prey refuge and fear effect, in a recent paper [37], Zhang et al.
proposed a predator–prey model which takes the following form:

dx αx β(1 − m)xy
= − bx2 − ,
dt 1 + Ky 1 + a(1 − m)x
dy cβ(1 − m)xy
= −γ y + , (1)
dt 1 + a(1 − m)x

where x(t) and y(t) denote, respectively, the densities of prey and predator population at
time t, α is the intrinsic growth rate of prey, b is the strength of intraspecific competition of
prey, γ is the death rate of the predator, c is the conversion coefficient, K refers to the level
of fear and β/a is the maximum number of prey eaten per predator per unit of time. All
these parameters are assumed to be positive. The term βx/(1 + ax) represents the Holling
type-II functional response, and (1 − m)x is the quantity of preys available to the preda-
tors, where m ∈ [0, 1) is the protection rate of the prey refuge for prey. For this model, the
authors performed a detailed stability and bifurcation analysis. We refer the readers to Ref.
[37] for more details.
In fact, a real ecosystem is inevitably affected by environmental noise. Therefore,
the shifting environmental effects can not be neglected, which indicates that deter-
ministic prey–predator models have limitations in predicting dynamics accurately,
while stochastic models can make it [38]. Many researchers introduced random dis-
turbances into deterministic systems to reveal the influence of environmental noise
[14,16,17,22,29,31,34–36,39–41]. Ripa et al. [24] examined the impact of environmental
noise on populations and presented a general theory of environmental noise in ecological
food webs. Mao et al. [21] proved that the explosion of population dynamics can be sup-
pressed by environmental Brownian noise and obtained that even a sufficiently small noise
perturbation can suppress explosions in population dynamics. Hence, it is of great signifi-
cance to further incorporate environmental randomness into the deterministic model (1).
Applying the technique used in Ref. [6] to include stochastic effects, we can obtain the
stochastic version of model (1) as follows (see the appendix for specific proof):
 
αx β(1 − m)xy
dx = − bx2 − dt + σ1 xdB1 (t),
1 + Ky 1 + a(1 − m)x
 
cβ(1 − m)xy
dy = −γ y + dt + σ2 ydB2 (t), (2)
1 + a(1 − m)x

where B1 (t) and B2 (t) are two independent standard Brownian motions defined in a com-
plete probability space (, F , {Ft }t≥0 , P) with a filtration {Ft }t≥0 satisfying the usual
conditions, and σi2 , i = 1, 2 are the intensities of the white noises. In this paper, we will
devote ourselves to the investigation on the dynamics of the stochastic model (2).
The organization of this paper is as follows. In the next section, we present some nec-
essary notations and preliminary results of model (2). In Section 3, we prove the existence
of a unique ergodic stationary distribution of the model, and in Section 4, we perform the
JOURNAL OF BIOLOGICAL DYNAMICS 873

extinction analysis of the model. Some numerical simulations are carried out in Section 5
to illustrate our theoretical results. We close the paper with conclusion in Section 6.

2. Preliminaries
In this section, we present some notations and basic results about model (2), which are the
basis for further investigation on the dynamics of model (2).
Consider the following n-dimensional stochastic differential equation:

dx(t) = f (x(t), t)dt + g(x(t), t)dB(t) for t ≥ t0 . (3)

Denote by C2,1 (Rn × [t0 , ∞); R+ ) the family of all nonnegative functions V(x, t) defined
on Rn × [t0 , ∞) such that they are continuously twice differentiable in x and once in t. We
introduce the differential operator defined in Ref. [20]

∂ 
n
∂ 1 T
n
∂2
L= + fi (x, t) + [g (x, t)g(x, t)]ij .
∂t ∂xi 2 ∂xi ∂xj
i=1 i,j=1

If L acts on a function V ∈ C2,1 (Rn × [t0 , ∞); R+ ), then

1
LV(x, t) = Vt (x, t) + Vx (x, t)f (x, t) + trace[g T (x, t)Vxx (x, t)g(x, t)],
2

where Vt = ∂V/∂t, Vx = (∂V/∂x1 , . . . , ∂V/∂xn ) and Vxx = (∂ 2 V/∂xi ∂xj )n×n . From Itô’s
formula, if x(t) ∈ Rn , then

dV(x(t), t) = LV(x(t), t)dt + Vx (x(t), t)g(x(t), t)dB(t).

The following theorem is about the existence and uniqueness of the global positive solution
of model (2).

Theorem 2.1: For any given initial value (x(0), y(0)) ∈ R2+ , there exists a unique solution
(x(t), y(t)) of model (2) on t ≥ 0 and the solution will remain in R2+ with probability one,
that is to say, (x(t), y(t)) ∈ R2+ for all t ≥ 0 almost surely (a.s.).

Proof: Since the coefficients of model (2) satisfy the local Lipschitz condition, then there
exists a unique local solution (x(t), y(t)) of model (2) on t ∈ [0, τe ), a.s., where τe denotes
the explosion time. To show that this solution is global, we only need to prove τe = ∞, a.s.
Let k0 > 0 be sufficiently large such that both x(0) and y(0) lie within the interval [1/k0 , k0 ].
For each integer k ≥ k0 , define the stopping time
 1 
τk = inf t ∈ [0, τe ) : min{x(t), y(t)} ≤ , or max{x(t), y(t)} ≥ k ,
k
where τk is increasing as k → ∞. Set τ∞ = limk→∞ τk , which implies τ∞ ≤ τe , a.s. Hence
to complete the proof, we only need to show that τ∞ = ∞, a.s. We prove this by con-
tradiction. If τ∞ < ∞, then there exist a pair of constants ∈ (0, 1) and T > 0 such
874 Y. XIA AND S. YUAN

that
P(τ∞ < T) > .
Therefore, there exists k1 ≥ k0 such that for all k ≥ k1 ,

P(k ) ≥ ,

where k = {τk ≤ T}. Then define a C2 -function V: R2+ → R+ by

γ γ 2cβ(1 − m)x 1
V =x− − ln + (y − 1 − ln y).
2cβ(1 − m) 2cβ(1 − m) γ c

From Itô’s formula,


  σ y σ 
γ σ1 2 2
dV(x, y) = LV(x, y)dt + σ1 x − dB1 (t) + − dB2 (t), (4)
2cβ(1 − m) c c

where
αx β(1 − m)xy αγ bγ x
LV(x, y) = − bx2 − − +
1 + Ky 1 + a(1 − m)x 2cβ(1 − m)(1 + Ky) 2cβ(1 − m)
γ y γ σ12 γ β(1 − m)xy γ
+ + − y+ +
2c 1 + a(1 − m)x 4cβ(1 − m) c 1 + a(1 − m)x c
β(1 − m)x σ2
− + 2
1 + a(1 − m)x 2c
bγ x γ γ σ12 γ γ σ2
≤ αx − bx2 + + y+ − y+ + 2
2cβ(1 − m) 2c 4cβ(1 − m) c c 2c
 
bγ γ γ σ1
2 γ σ2
≤ −bx2 + α + x− y+ + + 2
2cβ(1 − m) 2c 4cβ(1 − m) c 2c
γ σ12 γ σ2
≤ + + 2 + J̃ := J.
4cβ(1 − m) c 2c
Here,
 
2 bγ γ
J̃ = sup − bx + α + x− y .
(x,y)∈(0,∞) 2cβ(1 − m) 2c
Integrating and taking the expectation of both sides of (4) from 0 to τk ∧ T, we obtain

EV x (τk ∧ T) , y (τk ∧ T) ≤ V x (0) , y (0) + JE (τk ∧ T) ≤ V x (0) , y (0) + JT.


(5)
Noting that for every ω ∈ k , x(τk , ω) or y(τk , ω) equals either k or 1/k. Hence
 
γ γ 2cβ(1 − m)k
V x (τk ) , y (τk ) ≥ k − − ln
2cβ(1 − m) 2cβ(1 − m) γ
 
1 γ γ 2cβ(1 − m)
∧ − − ln
k 2cβ(1 − m) 2cβ(1 − m) γk
JOURNAL OF BIOLOGICAL DYNAMICS 875

   
1 1 1
∧ (k − 1 − ln k) ∧ − 1 + ln k .
c c k

From (5), we have

V x (0) , y (0) + JT ≥ E Ik V x (τk ) , y (τk )


 
γ γ 2cβ(1 − m)k
≥ k− − ln
2cβ(1 − m) 2cβ(1 − m) γ
 
1 γ γ 2cβ(1 − m)
∧ − − ln
k 2cβ(1 − m) 2cβ(1 − m) γk
   
1 1 1
∧ (k − 1 − ln k) ∧ − 1 + ln k .
c c k

This leads to a contradiction as we let k → ∞,

∞ > V x (0) , y (0) + JT = ∞.

So we have that τ∞ = ∞, a.s. 

Next, we present a lemma which gives a condition for the existence of a unique ergodic
stationary distribution of system (3). Let X(t) be a homogeneous Markov process in Rn
described by the following stochastic differential equation:


k
dX (t) = b (X) dt + gr (X) dBr (t) ,
r=1

k j
and the diffusion matrix is A(x) = (aij (x)), aij (x) = r=1 gr (x)gr (x).
i

Lemma 2.2 ([7]): The Markov process X(t) has a unique ergodic stationary distribution
μ(·), if there exists a bounded domain D ⊂ Rn with regular boundary and

(B.1) there is a positive number M such that


n
aij (x) ξi ξj ≥ M|ξ |2 , x ∈ D, ξ ∈ Rn .
i,j=1

(B.2) There exists a nonnegative C2 − function V such that LV is negative for any Rn \D.
Then
 
1 T
P lim f (X (t)) dt = f (x) μ (dx) = 1
T→∞ T 0 Rn

for all x ∈ Rn , where f (·) is a function integrable with respect to the measure μ.
876 Y. XIA AND S. YUAN

3. Stationary distribution
In this section, we establish sufficient conditions for the existence of a unique ergodic
stationary distribution. First, we give the following assumptions:

1
(H1 ) γ > 2 max{σ1 , σ2 };
2 2

2αcβ(1−m) 2αcβ(1−m) 4c2 α 2 c2 ασ12 σ22


(H2 ) λ = b+2aα(1−m) − 3b[1+a(1−m)(2α/b)]2 − 3b − 2b −γ − 2 > 0, where c2 is a
positive constant satisfying that

3cβa(1 − m)2 cβ(1 − m)


c2 >  3 +  2 . (6)
b 1 + a(1 − m) 2α
b α 1 + a(1 − m) 2α
b

Theorem 3.1: Let (x(t), y(t)) be the solution of model (2) with any given initial value
(x(0), y(0)) ∈ R2+ . If assumptions (H1 ) and (H2 ) are satisfied, then model (2) admits a
unique stationary distribution μ(·) and it has the ergodic property.

Proof: By the condition (B.2) in Lemma 2.2, we need to construct a C2 -continuous func-
tion V : R2+ → R+ and a bounded closed U such that LV ≤ −1 for (x, y) ∈ R2+ \U .
To this end, let σmax = max{σ1 , σ2 } and choose a constant θ > 0 such that ρ = γ −
2 (θ + 1) > 0, and define a C2 -function Ṽ : R2 → R as follows:
(1/2)σmax +
   c β(1 − m)α 
α 2 1  y θ+2
Ṽ = M −c1 x + c2 x − ln x + y − ln y + x+ ,
b bγ θ +2 c

where c1 = (cβ(1 − m))/(3α[1 + a(1 − m)(2α/b)]2 ) + (2/3)c2 and c2 is defined in (6)


(obviously, condition (6) implies that c2 > c1 ); M = (2/λ) max{2, R1 } and R1 is to be
determined later. We can easily check that

lim inf Ṽ(x, y) = ∞,


→0,(x,y)∈R2+ \U

where U := ( , 1/ ) × ( , 1/ ), and > 0 is a sufficiently small number. Notice that


Ṽ(x, y) has a global minimum point (x0 , y0 ) in the interior of R2+ . Then, we can define
a nonnegative C2 -function V : R2+ → R+ by

V(x, y) = Ṽ(x, y) − Ṽ(x0 , y0 ) = MV1 (x, y) + V2 (x, y), (7)

where
 α  c β(1 − m)α
2
V1 (x, y) = −c1 x + c2 x − ln x + y − ln y,
b bγ
1  y θ+2
V2 (x, y) = x+ − Ṽ(x0 , y0 ).
θ +2 c
Applying Itô’s formula to V1 (x, y), we have

c1 αx c1 β(1 − m)xy c2 αx c2 β(1 − m)xy


LV1 = − + c1 bx2 + + − c2 bx2 −
1 + Ky 1 + a(1 − m)x 1 + Ky 1 + a(1 − m)x
JOURNAL OF BIOLOGICAL DYNAMICS 877

c2 α 2 c2 αβ(1 − m)y c2 ασ12 c2 β(1 − m)α


− + c2 αx + + − y
b(1 + Ky) b[1 + a(1 − m)x] 2b b
c2 cβ 2 (1 − m)2 αxy cβ(1 − m)x σ2
+ +γ − + 2
bγ [1 + a(1 − m)x] 1 + a(1 − m)x 2
α(c2 − c1 )x c2 αβ(1 − m) c2 ασ12
≤ + c1 bx2 + c1 β(1 − m)xy − c2 bx2 + c2 αx + y+
1 + Ky b 2b
c2 αβ(1 − m) c2 cβ 2 (1 − m)2 αxy cβ(1 − m)x σ2
− y+ +γ − + 2
b bγ 1 + a(1 − m)x 2
cβ(1 − m)x
≤ c2 αx − c1 αx + c1 bx2 − c2 bx2 + c2 αx −
1 + a(1 − m)x
 
c2 cβ 2 (1 − m)2 α c2 ασ12 σ2
+ c1 β(1 − m) + xy + +γ + 2
bγ 2b 2
   
b b 2α cβ(1 − m)x
= c2 αx 2 − x + c1 α x x−1 − −
α α b 1 + a(1 − m)x
 
c2 cβ (1 − m) α
2 2 2c1 α 2 c2 ασ12 σ2
+ c1 β(1 − m) + xy + + +γ + 2
bγ b 2b 2
cβ(1 − m)(2α/b) 2c1 α 2 c2 ασ12 σ2 cβ(1 − m)x
=− + + +γ + 2 + −
1 + a(1 − m)(2α/b) b 2b 2 1 + a(1 − m)x
   
cβ(1 − m)(2α/b) b b 2α
+ + c2 αx 2 − x + c1 α x x−1 −
1 + a(1 − m)(2α/b) α α b
 
c2 cβ (1 − m) α
2 2
+ c1 β(1 − m) + xy

cβ(1 − m)(2α/b) 2c1 α 2 c2 ασ12 σ2
=− + + +γ + 2
1 + a(1 − m)(2α/b) b 2b 2
 
c2 cβ (1 − m) α
2 2
+ c1 β(1 − m) + xy + F(x),

where

cβ(1 − m)x cβ(1 − m)(2α/b) b
F(x) = − + + c2 αx 2 − x
1 + a(1 − m)x 1 + a(1 − m)(2α/b) α
  
b 2α
+ c1 α x x−1 − .
α b
We can compute that
cβ(1 − m)
F (x) = − + 2c2 α − 2c2 bx + 2c1 bx − c1 α
[1 + a(1 − m)x]2
and
2cβa(1 − m)2
F (x) = − 2c2 b + 2c1 b.
[1 + a(1 − m)x]3
878 Y. XIA AND S. YUAN

Therefore,
cβ(1 − m)
F (x)|x=2α/b = − − 2c2 α + 3c1 α = 0,
[1 + a(1 − m)(2α/b)]2
due to the equality c1 = (cβ(1 − m))/(3α[1 + a(1 − m)(2α/b)]2 ) + (2/3)c2 . Moreover,
we can compute that
2cβa(1 − m)2
F (x)|x=2α/b = − 2c2 b + 2c1 b < 0,
[1 + a(1 − m)(2α/b)]3
where we have used inequality (6). Hence, we have that for all x ∈ R+ ,


F(x) ≤ F = 0.
b
Consequently,

cβ(1 − m)(2α/b) 2c1 α 2 c2 ασ12 σ2


LV1 ≤ − + + +γ + 2
1 + a(1 − m)(2α/b) b 2b 2
 
c2 cβ (1 − m) α
2 2
+ c1 β(1 − m) + xy

cβ(1 − m)(2α/b) 2αcβ(1 − m)
=− +
1 + a(1 − m)(2α/b) 3b [1 + a(1 − m)(2α/b)]2
4c2 α 2 c2 ασ12 σ2
+ + +γ + 2
3b 2b 2
 
c2 cβ 2 (1 − m)2 α
+ c1 β(1 − m) + xy

 
c2 cβ 2 (1 − m)2 α
:= −λ + c1 β(1 − m) + xy, (8)

where λ is a positive number defined in assumption (H2 ).
Similarly, we can compute that
  
y θ+1 αx 2 β(1 − m)xy γ β(1 − m)xy
LV2 = x + − bx − − y+
c 1 + Ky 1 + a(1 − m)x c 1 + a(1 − m)x
     
1 y θ 2 2 y 2
+ (θ + 1) x + σ1 x + σ22
2 c c
 y  θ+1  γ 
≤ x+ (α + γ )x − bx2 − γ x − y
c c
     
1 y θ
2 y
2
2 2
+ (θ + 1) x + σ 1 x + σ2
2 c c
 
y θ+1   y  1 2  y θ+2
≤ x+ G−γ x+ + σmax (θ + 1) x +
c c 2 c
 y θ+1  y θ+2 1 2  y θ+2
=G x+ −γ x+ + σmax (θ + 1) x +
c c 2 c
JOURNAL OF BIOLOGICAL DYNAMICS 879

  
y θ+1 1 2 y θ+2
=G x+ − γ − σmax (θ + 1) x +
c 2 c
ρ  
y θ+2
≤F− x+
2 c
 
ρ θ+2  y θ+2
≤F− x + , (9)
2 c
where

G= sup {(α + γ )x − bx2 } < ∞,


(x,y)∈R2+
 y θ+1 ρ  y θ+2
F= sup G x+ − x+ < ∞.
(x,y)∈R2+ c 2 c

From (7)–(9), we have that


 
c2 cβ 2 (1 − m)2 α
LV = L(V1 + V2 ) ≤ M − λ + c1 β(1 − m) + xy

 
ρ θ+2  y θ+2
− x + + F. (10)
2 c
Next, we prove LV ≤ −1 on R2+ \U . Take > 0 sufficiently small such that the following
inequalities hold:
λbγ
0< < , (11)
4[c1 β(1 − m)bγ + c2 cβ 2 (1 − m)2 α]
ρbγ
0< < , (12)
4Mcθ+2 [c1 β(1 − m)bγ + c2 cβ 2 (1 − m)2 α]
ρbγ
0< < , (13)
4M[c1 β(1 − m)bγ + c2 cβ 2 (1 − m)2 α]
ρ
− Mλ − θ+2 + R2 ≤ −1, (14)
4
ρ
− Mλ − θ+2 θ+2 + R3 ≤ −1, (15)
4c
where R2 and R3 are constants given by (17) and (18). Denote
   
U 1 = x, y ∈ R2+ : 0 < x < , U 2 = x, y ∈ R2+ : 0 < y < ,
1 1
U3 = x, y ∈ R2+ : x > , U4 = x, y ∈ R2+ : y > .
  
Obviously, R2+ \U = U 1 U 2 U 3 U 4 . Thus we only need to validate LV ≤ −1 in
each domain U i , where i = 1, 2, 3, 4.
Case 1. On the domain U 1 , we have xy ≤ (1 + yθ+2 ). It then follows from (10) that
 
Mλ Mλ c2 cβ 2 (1 − m)2 α
LV ≤ − + − + M c1 β(1 − m) +
4 4 bγ
880 Y. XIA AND S. YUAN

  
c2 cβ 2 (1 − m)2 α ρ
+ M c1 β(1 − m) + − θ+2 yθ+2
bγ 4c

ρ Mλ
− xθ+2 + − + R1 ,
4 2

where

ρ θ+2 yθ+2
R1 = sup − x + θ+2 + F = F. (16)
(x,y)∈R2+ 4 c

By M = (2/λ) max{2, R1 }, one can see that Mλ/4 ≥ 1. Hence we have that for all (x, y) ∈
U 1,
Mλ ρ θ +2 Mλ
LV(x, y) ≤ − − x ≤− ≤ −1,
4 4 4
where (11) and (12) have been used.
Case 2. On the domain U 2 , we have xy ≤ (1 + xθ+2 ). It then follows from (10) that
 
Mλ Mλ c2 cβ 2 (1 − m)2 α
LV ≤ − + − + M c1 β(1 − m) +
4 4 bγ
  
c2 cβ 2 (1 − m)2 α ρ θ+2
+ M c1 β(1 − m) + − x
bγ 4

ρ Mλ
− θ+2 yθ+2 + − + R1 .
4c 2

In view of (11) and (13), we get


Mλ ρ Mλ
LV(x, y) ≤ − − θ+2 yθ+2 ≤ − ≤ −1.
4 4c 4
Case 3. On U 3 , we have from (14) that for all (x, y) ∈ U 3 ,
 
ρ θ+2 c2 cβ 2 (1 − m)2 α ρ
LV ≤ −Mλ − x + M c1 β(1 − m) + xy − xθ+2
4 bγ 4
ρ θ+2
− θ+2 y +F
2c
ρ ρ
≤ −Mλ − xθ+2 + R2 ≤ −Mλ − θ+2 + R2 ≤ −1,
4 4
where
 
c2 cβ 2 (1 − m)2 α ρ ρ
R2 = sup M c1 β(1 − m) + xy − xθ+2 − θ+2 yθ+2 + F .
(x,y)∈R2+ bγ 4 2c
(17)
Case 4. On U 4 , we have from (15) that for all (x, y) ∈ U 4 ,
 
ρ θ+2 c2 cβ 2 (1 − m)2 α ρ
LV ≤ −Mλ − θ+2 y + M c1 β(1 − m) + xy − xθ+2
4c bγ 2
JOURNAL OF BIOLOGICAL DYNAMICS 881

ρ θ+2
− y +F
4cθ+2
ρ 1
≤ −Mλ − θ+2 θ+2 + R3 ≤ −1,
4c
where
 
c2 cβ 2 (1 − m)2 α ρ ρ
R3 = sup M c1 β(1 − m) + xy − xθ+2 − θ+2 yθ+2 + F .
(x,y)∈R2+ bγ 2 4c
(18)
To sum up, we can take a sufficiently small such that LV(x, y) ≤ −1 for all (x, y) ∈
R2+ \U , meaning that condition (B.2) in Lemma 2.2 holds.
In addition, we find M0 = min(x,y)∈U {σ12 x2 , σ22 y2 } such that


2
aij (x) ξi ξj = σ12 x2 ξ12 + σ22 y2 ξ22 ≥ M0 ξ 2
, x ∈ U , ξ = (ξ1 , ξ2 ) ∈ R2 ,
i,j=1

which implies that condition (B.1) in Lemma 2.2 also holds. Therefore, thanks to
Lemma 2.2, model (2) has a unique stationary distribution μ(·) and it is ergodic. This
completes the proof. 

4. Extinction
In this section, we derive sufficient criteria for the extinction of model (2) in two scenarios,
one is both the predator and the prey go extinct, the other is the predator goes to extinction
but the prey persists.

Theorem 4.1: Let (x(t), y(t)) be the solution of model (2) with any positive initial value
(x(0), y(0)) ∈ R2+ . We have

(i) if α < σ12 /2, then both x(t) and y(t) will be extinct, a.s., that is,

lim x(t) = 0, lim y(t) = 0, a.s.


t→∞ t→∞

(ii) if α > σ12 /2, cβ(1 − m)((α − (σ12 /2))/(b)) − γ − (σ22 /2) < 0, then y(t) will go to
extinction, a.s., i.e. limt→∞ y(t) = 0, a.s. Moreover, the distribution of x(t) converges
weakly to the measure which has the following density:
−2
π(x) = Cσ1−2 x(2α/σ1 2) −(2b/σ12 )x
e , x ∈ (0, ∞),
∞
where C = (σ12 (σ12 /2b)1−(2α/σ1 ) )/( ((2α/σ12 ) − 1)) is a constant such that
2
0 π(x)
dx = 1.

Proof: (i) Making use of Itô’s formula on ln x(t), we get



α β(1 − m)y(t) σ2
d ln x(t) = − bx(t) − − 1 dt + σ1 dB1 (t)
1 + Ky(t) 1 + a(1 − m)x(t) 2
882 Y. XIA AND S. YUAN


σ2
≤ α− 1 dt + σ1 dB1 (t),
2
from which we have
ln x(t) ln x(0) σ 2 σ1 B1 (t)
≤ +α− 1 + .
t t 2 t
Noticing that limt→∞ B1 (t)/t = 0, a.s. and α < σ12 /2, we have

ln x(t) σ2
lim sup ≤ α − 1 < 0,
t→∞ t 2

which means that


lim x(t) = 0, a.s. (19)
t→∞
Similarly, by using Itô’s formula to ln y(t), we obtain

cβ(1 − m)x(t) σ2
d ln y(t) = −γ + − 2 dt + σ2 dB2 (t)
1 + a(1 − m)x(t) 2

σ 2
≤ cβ(1 − m)x(t) − γ − 2 dt + σ2 dB2 (t).
2
Then,

ln y(t) ln y(0) cβ(1 − m) t σ22 σ2 B2 (t)
≤ + x(s) ds − γ − + . (20)
t t t 0 2 t
It follows from (19) and limt→∞ B2 (t)/t = 0, a.s., we get

ln y(t) σ2
lim sup ≤ −γ − 2 < 0,
t→∞ t 2

which implies that limt→∞ y(t) = 0, a.s.


(ii) For any positive initial value (x(0), y(0)) ∈ R2+ , we have

dx(t) ≤ x(t)(α − bx(t))dt + σ1 x(t)dB1 (t),

then by Lemma A.1 in Ref. [11] and the comparison theorem, we obtain

1 t α − (σ12 /2)
lim sup x(s) ds ≤ a.s. (21)
t→∞ t 0 b

Combining (20) and (21), it follows that if α > σ12 /2, cβ(1 − m)((α − (σ12 /2))/b) − γ −
(σ22 /2) < 0, then we have that

ln y(t) α − (σ12 /2) σ22
lim sup ≤ cβ(1 − m) −γ − < 0,
t→∞ t b 2

that is, limt→∞ y(t) = 0, a.s.


JOURNAL OF BIOLOGICAL DYNAMICS 883

On the other hand, consider the following one-dimensional stochastic differential


equation:
dX(t) = X(t)(α − bX(t))dt + σ1 X(t)dB1 (t), (22)
with the initial value X(0) = x(0) > 0. We know from Ref. [15] that model (22) has the
ergodic property and the invariant density is given by

π(x) = Cσ1−2 x(2α/σ1 )−2 e−(2b/σ1 )x ,


2 2
x ∈ (0, ∞),
∞
where C = (σ12 (σ12 /2b)1−(2α/σ1 ) )/( ((2α/σ12 ) − 1)) is a constant satisfying that 0 π(x)
2

dx = 1.
Since limt→∞ y(t) = 0, a.s., for any small , there are T and a set  ⊂  such that
P( ) > 1 − , (β(1 − m)xy)/(1 + a(1 − m)x) ≤ β(1 − m)xy ≤ x and α/(1 + Ky) ≥
α/(1 + K ) for t ≥ T and ω ∈  . Then, we have
  
α
x(t) − bx(t) − x(t) dt + σ1 x(t)dB1 (t)
1+K
≤ dx(t) ≤ x(t)(α − bx(t))dt + σ1 x(t)dB1 (t),

that is to say, when → 0, we obtain that the distribution of the process x(t) converges
weakly to the measure with the density of π(x). This completes the proof. 

5. Numerical simulations
In this section, we will perform some numerical simulations using MATLAB R2016b to
illustrate the effect of white noise, fear effect and a prey refuge on the dynamics of model (2).
The numerical scheme obtained through Milstein’s higher order method [8] applied on
stochastic model (2) under consideration is given by
 
α β(1 − m)yj
xj+1 = xj + xj − bxj − t
1 + Kyj 1 + a(1 − m)xj

√ 1 2 2
+ σ1 ζj t + σ1 t(ζj − 1) ,
2
  
cβ(1 − m)xj √ 1 2 2
yj+1 = yj + yj − γ t + σ2 ξj t + σ2 t(ξj − 1) ,
1 + a(1 − m)xj 2

where ζj and ξj are two independent Gaussian random variables N(0, 1) for j = 1, 2, . . . , n.
All numerical simulations reported here are carried out with the choice of time stepping
t = 0.01. In this section, we always take the following parameter values:

α = 0.6, b = 0.3, β = 0.3, c = 0.8, a = 0.3, γ = 0.1, (23)

and assume that the initial value of model (2) is (x(0), y(0)) = (0.6, 0.5).
First, in order to investigate the impact of white noise on the dynamics of model (2),
we choose all parameter values in (23) and m = 0.1, K = 0.3, σ1 = σ2 = 0.01. By cal-
culating, we get from (6) that c2 > 0.148. Now we take c2 = 0.1481, then we compute
λ = 0.0118 > 0. Thus Assumptions (H1 ) and (H2 ) are satisfied. By Theorem 3.1, there
884 Y. XIA AND S. YUAN

(a) (b)

(c) (d)

Figure 1. (a) and (c): The asymptotic behaviour of the solutions to stochastic model (2) around the
positive equilibrium of model (1) with initial value (x(0), y(0)) = (0.6, 0.5); (b) and (d): The density func-
tion diagrams of x(t) and y(t), respectively. The parameters are taken as (23) and m = 0.1, K = 0.3,
σ1 = σ2 = 0.01.

exists a unique ergodic stationary distribution for the solutions of model (2). Figure 1
confirms this.
To obtain deep insights of the influences of white noise on population dynamics, we
keep the model parameter values the same as in (23) but let σ1 = 1.1, which implies

Figure 2. Numerical simulation for model (1) and model (2) with initial value (x(0), y(0)) = (0.6, 0.5).
The parameters are taken as (23) and m = 0.1, K = 0.3, σ1 = 1.1, σ2 = 0.01.

Figure 3. Numerical simulation for model (1) and model (2) with initial value (x(0), y(0)) = (0.6, 0.5).
The parameters are taken as (23) and m = 0.1, K = 0.3, σ1 = 0.1, σ2 = 0.9.
JOURNAL OF BIOLOGICAL DYNAMICS 885

that condition α < σ12 /2 holds. Then we get that both prey and predator populations
are extinction by Theorem 4.1, shown as in Figure 2. Now we take σ1 = 0.1, σ2 = 0.9
and the other parameters remain unchanged, we can easily check that the condition α >
σ2
σ12 /2, cβ(1 − m)((α − (σ12 /2)/b) − γ − 22 < 0 are satisfied. By Theorem 4.1, we know
that the prey population is persistent and the predator population will be extinct (see
Figure 3). These show that the random disturbance will change the dynamics when the
environment noise is large enough.

Figure 4. Numerical simulation for model (1) and model (2) with initial value (x(0), y(0)) = (0.6, 0.5)
and different K, respectively. The parameters are taken as (23) and m = 0.1, σ1 = σ2 = 0.01.
886 Y. XIA AND S. YUAN

Next, in order to illustrate the influence of fear effect to model (2) through numerical
simulation, we choose different values of K, say K = 0, K = 0.1 and K = 1, σ1 = σ2 =
0.01. For the remaining parameter values, we keep them the same as in (23). We can com-
pute that these parameters satisfy Assumptions (H1 ) and (H2 ) and therefore model (2) has
a stationary distribution, which means that both prey and predator populations are persis-
tent. Moreover, from Figure 4 we find that the increase of fear effect will reduce the density
of predator but with no obvious effect on the density of prey.

Figure 5. The solutions of model (2) with the initial value (x(0), y(0)) = (0.6, 0.5) and different K,m,
respectively. The parameters are taken as (23) and σ1 = σ2 = 0.01.
JOURNAL OF BIOLOGICAL DYNAMICS 887

Finally, we numerically illustrate the impact of a prey refuge to model (2) and choose
different values of m, say m = 0, m = 0.2, m = 0.5 and m = 0.7, σ1 = σ2 = 0.01. For the
remaining parameter values, we keep them the same as in (23). We can easily compute that
these parameters satisfy assumptions (H1 ) and (H2 ). So we obtain that there is a unique
ergodic stationary distribution for model (2), which means that both prey and predator
populations are persistent. From Figure 5, we can see that with the increase of the amount
of refuges m, the density of the prey increases while the predator density decreases. This
means that the increase of the amount of refuges m is favourable to prey population but it
is adverse to the persistence of predator population. Moreover, we can also see that large
K may lead to the predator density decreasing fast. When we choose σ1 = 1.1, σ2 = 0.01
and the other parameters are the same as in Figure 5, we simply calculate that condition
α < σ12 /2 holds. Then we obtain that both prey and predator populations are extinction
by Theorem 4.1, shown in Figure 6. The results again show that the large noise is very
destructive to the persistence of the population.

Figure 6. The solutions of model (2) with the initial value (x(0), y(0)) = (0.6, 0.5) and different K,m,
respectively. The parameters are taken as (23) and σ1 = 1.1, σ2 = 0.01.
888 Y. XIA AND S. YUAN

6. Conclusion
This paper focuses on a stochastic Holling type-II prey–predator model with a prey refuge
and fear effect. Mathematically, the sufficient criteria for the existence of a unique ergodic
stationary distribution and the extinction of the model have been obtained. Ecologically,
we get the following conclusions:

(1) Random disturbance may change the dynamic behaviour of the population, espe-
cially when the noise is large, it may lead to the extinction of the prey and predator
populations.
(2) By simulations, we find that the increase of fear effect K will reduce the density of
predator, but the effect on the density of prey is not obvious. In addition, one can see
that the increase of the amount of refuges m is favourable to prey population but it is
adverse to the persistence of predator population.

There are some interesting themes worthy of further research. On the one hand, we
can incorporate some other environmental noises into model (2), such as coloured noise,
Poisson noise and so on. On the other hand, to make the model (2) more realistic, we can
further consider the factors such as the influence of impulsive perturbations and delay. We
leave these for future investigations.

Disclosure statement
No potential conflict of interest was reported by the author(s).

Funding
Research is supported by the National Natural Science Foundation of China (Grant Nos. 11671260
and 12071293).

References
[1] J.R. Beddington, Mutual interference between parasites or predators and its effect on searching
efficiency, J. Anim. Ecol. 44 (1975), pp. 331–340.
[2] F. Brauer and C. Castillo-Chvez, Mathematical Models in Population Biology and Epidemiology,
Springer, Berlin, 2001.
[3] S. Creel and D. Christianson, Relationships between direct predation and risk effects, Trends
Ecol. Evol. 23 (2008), pp. 194–201.
[4] W. Cresswell, Predation in bird populations, J. Ornithol. 152 (2011), pp. 251–263.
[5] D.L. DeAngelis, R.A. Goldstein, and R.V. O’Neill, A model for tropic interaction, Ecology, 56
(1975), pp. 881–892.
[6] R. Durrett, Stochastic Calculus: A Practical Introduction, Probability Stochastics, 1st ed., CRC
Press, Boca Raton, FL, 1996.
[7] R.Z. Has’minskii, Stochastic Stability of Differential Equations, Sijthoff Noordhoff, Alphen aan
den Rijn, The Netherlands, 1980.
[8] D. Higham, An algorithmic introduction to numerical simulation of stochastic differential
equations, SIAM Rev. 43 (2001), pp. 525–546.
[9] C.S. Holling, The functional response of predators to prey density and its role in mimicry and
population regulation, Mem. Entomol. Soc. Can. 97 (1965), pp. 5–60.
JOURNAL OF BIOLOGICAL DYNAMICS 889

[10] Y. Huang, F. Chen, and L. Zhong, Stability analysis of a prey–predator model with Holling
type III response function incorporating a prey refuge, Appl. Math. Comput. 182 (2006), pp.
672–683.
[11] C. Ji and D. Jiang, Qualitative analysis of a stochastic ratio-dependent predator–prey system, J.
Comput. Appl. Math. 235 (2011), pp. 1326–1341.
[12] D. Jia, T. Zhang, and S. Yuan, Pattern dynamics of a diffusive toxin producing phytoplank-
ton–zooplankton model with three-dimensional patch, Int. J. Bifurcat. Chaos 29 (2019), pp.
1930011.
[13] S.L. Lima, Nonlethal effects in the ecology of predator–prey interactions, Bioscience 48 (1998),
pp. 25–34.
[14] M. Liu and M. Deng, Permanence and extinction of a stochastic hybrid model for tumor growth,
Appl. Math. Lett. 94 (2019), pp. 66–72.
[15] Q. Liu, D. Jiang, T. Hayat, and A. Alsaedi, Dynamics of a stochastic predator-prey model with
stage structure for predator and Holling type II functional response, J. Nonlinear Sci. 28 (2018),
pp. 1151–1187.
[16] C. Liu, L. Wang, Q. Zhang, and Y. Yan, Dynamical analysis in a bioeconomic phytoplankton zoo-
plankton system with double time delays and environmental stochasticity, Physica A 15 (2017),
pp. 682–698.
[17] M. Liu and Y. Zhu, Stability of a budworm growth model with random perturbations, Appl. Math.
Lett. 79 (2018), pp. 13–19.
[18] Z. Ma, F. Chen, C. Wu, and W. Chen, Dynamic behaviors of a Lotka-Volterra predator–prey
model incorporating a prey refuge and predator mutual interference, Appl. Math. Comput. 219
(2013), pp. 7945–7953.
[19] Z. Ma, S. Wang, W. Li, and Z. Li, The effect of prey refuge in a patchy predator–prey system, Math.
Biosci. 243 (2013), pp. 126–130.
[20] X. Mao, Stochastic Differential Equations and Applications, Woodhead Publishing, Cambridge,
2007.
[21] X. Mao, G. Marion, and E. Renshaw, Environmental Brownian noise suppresses explosions in
population dynamics, Stochast. Proc. Appl. 97 (2002), pp. 95–110.
[22] X. Meng, F. Li, and S. Gao, Global analysis and numerical simulations of a novel stochastic eco-
epidemiological model with time delay, Appl. Math. Comput. 339 (2018), pp. 701–726.
[23] D. Mukherjee, The effect of prey refuges on a three species food chain model, Differ. Equ. Dyn.
Syst.22 (2014), pp. 413–426.
[24] J. Ripa, P. Lundberg, and V. Kaitala, A general theory of environmental noise in ecological food
webs, Am. Nat. 151 (1998), pp. 256–263.
[25] S. Sarwardi, P.K. Mandal, and S. Ray, Analysis of a competitive prey–predator system with a prey
refuge, Biosystems 110 (2012), pp. 133–148.
[26] S. Sasmal, Population dynamics with multiple Allee effects induced by fear factors induced by fear
factors – a mathematical study on prey–predator, Appl. Math. Model. 64 (2018), pp. 1–14.
[27] A. Sih, Prey refuges and predator–prey stability, Theor. Popul. Biol. 31 (1987), pp. 1–12.
[28] X. Wang, L. Zanette, and X. Zou, Modelling the fear effect in predator–prey interactions, J. Math.
Biol. 73 (2016), pp. 1–38.
[29] C. Xu and S. Yuan, Competition in the chemostat: a stochastic multi-species model and its
asymptotic behavior, Math. Biosci. 280 (2016), pp. 1–9.
[30] C. Xu, S. Yuan, and T. Zhang, Global dynamics of a predator–prey model with defence mechanism
for prey, Appl. Math. Lett. 62 (2016), pp. 42–48.
[31] C. Xu, S. Yuan, and T. Zhang, Average break-even concentration in a simple chemostat model
with telegraph noise, Nonlinear Anal. Hybrid. Syst. 29 (2018), pp. 373–382.
[32] S. Yan, D. Jia, T. Zhang, and S. Yuan, Pattern dynamic in a diffusive predator–prey model with
hunting cooperations, Chaos Solitons Fract. 130 (2020), pp. 109428.
[33] F. Yi, J. Wei, and J. Shi, Bifurcation and spatiotemporal patterns in a homogeneous diffusive
predator–prey system, J. Differ. Equ. 246 (2009), pp. 1944–1977.
[34] X. Yu and S. Yuan, Asymptotic properties of a stochastic chemostat model with two distributed
delays and nonlinear perturbation, Discrete Contin. Dyn. B 25 (2020), pp. 2373–2390.
890 Y. XIA AND S. YUAN

[35] X. Yu, S. Yuan, and T. Zhang, Asymptotic properties of stochastic nutrient-plankton food chain
models with nutrient recycling, Nonlinear Anal. Hybrid. Syst. 34 (2019), pp. 209–225.
[36] S. Yuan, D. Wu, G. Lan, and H. Wang, Noise-induced transitions in a nonsmooth predator–prey
model with stoichiometric constraints, Bull. Math. Biol. 82 (2020), pp. 167.
[37] H. Zhang, Y. Cai, S. Fu, and W. Wang, Impact of the fear effect in a prey–predator model
incorporating a prey refuge, Appl. Math. Comput. 356 (2019), pp. 328–337.
[38] S. Zhang and X. Meng, Dynamics analysis and numerical simulations of a stochastic non-
autonomous predator–prey system with impulsive effects, Nonlinear Anal. Hybrid. Syst. 26
(2017), pp. 19–37.
[39] Y. Zhao, S. Yuan, and J. Ma, Survival and stationary distribution analysis of a stochastic com-
petitive model of three species in a polluted environment, Bull. Math. Biol. 77 (2015), pp.
1285–1326.
[40] S. Zhao, S. Yuan, and H. Wang, Threshold behavior in a stochastic algal growth model with
stoichiometric constraints and seasonal variation, J. Differ. Equ. 268 (2020), pp. 5113–5139.
[41] W. Zuo and D. Jiang, Stationary distribution and periodic solution for stochastic predator–prey
systems with nonlinear predator harvesting, Commun. Nonlinear Sci. 36 (2016), pp. 65–80.

Appendix 1: Construction of the stochastic model (2)


By the methods used in Ref. [6] about stochastic effects, we first consider a discrete time Markov
chain. For x(t) and y(t) in model (2), given t > 0, we present a process

X (t) (t) = (x(t) (t), y(t) (t))T , t = 0, t, 2t, . . .


with initial valueX (t) (0)
= (x(t) (0), y(t) (0))T ∈ R2+ . Let normal distribution random variable
(t) ∞
sequence {Ri (l)}l=0 satisfies

E[R(t)
i (l)] = 0, E[R(t)
i (l)]2 = σi2 t, E[R(t)
i (l)]4 = o(t), (A1)
where i = 1, 2, l = 0, 1, 2, . . . and σi2 express the intensities of stochastic effects. On each period
[lt, (l + 1)t), we assume that X (t) (t) grows according to model (2) and is also affected by the
random (x(t) R(t)
1 (l), y
(t) R(t) (l))T . Specifically, for l = 0, 1, . . ., we get
2
(t)
x(t) ((l + 1)t) = x(t) (lt) + x(t) (lt)R1 (l)

α
+ x(t) (lt) − bx(t) (lt)
1 + Ky(t) (lt)

β(1 − m)y(t) (lt)
− t,
1 + a(1 − m)x(t) (lt)
(t)
y(t) ((l + 1)t) = y(t) (lt) + y(t) (lt)R2 (l)
 
(t) cβ(1 − m)x(t) (lt)y(t) (lt)
+ −γ y (lt) + t.
1 + a(1 − m)x(t) (lt)

Next we demonstrate that X (t) (t) converges to a diffusion process as t → 0. And we need to
determine the drift coefficient and diffusion coefficient of the diffusion process. Let P (t) (x̄, dz)
denote the transition probabilities of the homogeneous Markov chain {X (t) (lt)}, that is

P (t) (x̄, B) = Prob{X (t) ((l + 1)t) ∈ B|X (t) = x̄}


for all x̄ = (x, y) ∈ R2+ and all Borel sets B ⊂ R2+ . From (A1)
 
1 α β(1 − m)y x
(z1 − x)P (t)
(x̄, dz) = x − bx − + ER(t) (0)
t 1 + ky 1 + a(1 − m)x t 1
JOURNAL OF BIOLOGICAL DYNAMICS 891


α β(1 − m)y
=x − bx − ,
1 + ky 1 + a(1 − m)x

1 cβ(1 − m)xy y
(z2 − y)P (t) (x̄, dz) = −γ y + + ER(t) (0)
t 1 + a(1 − m)x t 2
cβ(1 − m)xy
= −γ y + .
1 + a(1 − m)x
To determine the diffusion coefficients, we consider the following moments:

(t) 1
gij (x) = (zi − xi )(zj − xj )P (t) (x, dz), i, j = 1, 2.
t
By (A1)
   
   1 α β(1 − 2 
 (t)   m)y (t) 
g11 (x) − σ12 x2  =  E t x − bx − + R1 (0)x − σ12 x2 
 t 1 + Ky 1 + a(1 − m)x 
  2
 α β(1 − m)y
= t x − bx −
1 + Ky 1 + a(1 − m)x

α β(1 − m)y
+ 2x2 − bx − ER(t)
1 (0)
1 + Ky 1 + a(1 − m)x

1 2  (t) 2 
+ x E R1 (0) − σ12 x2 
t
 2
α β(1 − m)y
= t x − bx − ,
1 + Ky 1 + a(1 − m)x
   
   1 cβ(1 − m)xy 2 
 (t) 2 2  (t) 2 2
g22 (x) − σ2 y  =  E t −γ y + + R2 (0)y − σ2 y 
 t 1 + a(1 − m)x 
  2  
 cβ(1 − m)xy cβ(1 − m)x
= t −γ y + + 2y2 −γ + ER(t)
2 (0)
1 + a(1 − m)x 1 + a(1 − m)x

1 2  (t) 2 2 2

+ y E R2 (0) − σ2 y 
t
 
cβ(1 − m)xy 2
= t −γ y + .
1 + a(1 − m)x
Therefore,
   
 (t)   (t) 
lim sup g11 (x) − σ12 x2  = 0, lim sup g22 (x) − σ22 y2  = 0
t→0+ ||x||≤L t→0+ ||x||≤L

for all 0 < L < ∞. Similarly,


 
 (t) 
lim sup g12 (x) = 0.
t→0+ ||x||≤L

In addition, from (A1), we obtain that for all 0 < L < ∞



1
lim sup ||z − x||3 P (t) (x, dz) = 0.
t→0+ ||x||≤L t

Finally, extend the definition of X (t) (t) to all t ≥ 0 by setting X (t) (t) = X (t) (lt) for all t ∈
[lt, (l + 1)t). On the basis of Theorem 7.1, Lemma 8.2 in Ref. [6], we can verify that as t → 0,
892 Y. XIA AND S. YUAN

X (t) (t) converges weakly to the solution of the stochastic system as follows:
 
αx β(1 − m)xy
dx = − bx2 − dt + σ1 xdB1 (t),
1 + Ky 1 + a(1 − m)x
 
cβ(1 − m)xy
dy = −γ y + dt + σ2 ydB2 (t),
1 + a(1 − m)x
where B1 (t) and B2 (t) are two independent Brownian motions with the intensities represented by
two positive constants σ1 and σ2 .

You might also like