You are on page 1of 16

Physica A 509 (2018) 921–936

Contents lists available at ScienceDirect

Physica A
journal homepage: www.elsevier.com/locate/physa

A stochastic differential equation SIS epidemic model


incorporating Ornstein–Uhlenbeck process

Weiming Wang a , , Yongli Cai a , Zuqin Ding a , Zhanji Gui b
a
School of Mathematical Science, Huaiyin Normal University, Huaian 223300, PR China
b
Software Department, Hainan College of Software Technology, Qionghai, 571400, PR China

highlights

• A stochastic SIS epidemic model with mean-reverting process is developed.


• The stochastic extinction and persistence of the SDE model are given.
• The smaller speed of reversion or bigger intensity of volatility can suppress the disease outbreak.
• To control the disease, one need increase the intensity of volatility or decrease the speed of reversion.

article info a b s t r a c t

Article history: In this paper, based on the results of Gray et al. (2011), we propose a new SDE SIS model
Received 29 April 2018 incorporating mean-reverting Ornstein–Uhlenbeck process, and prove that the stochastic
Received in revised form 31 May 2018 basic reproduction number Rs0 can be used to identify the stochastic extinction and
Available online 22 June 2018
persistence for the SDE mode: if Rs0 < 1 under mild extra conditions, the disease will
Keywords: be extinct a.s., while if Rs0 > 1, the disease will persist a.s. Epidemiologically, we find
Ornstein–Uhlenbeck process that smaller speed of reversion or bigger intensity of volatility can suppress the disease
Stochastic basic reproduction number outbreak. Thus, in order to control the spread of the disease, we must decrease the speed
Extinction of reversion or increase the intensity of volatility.
Persistence © 2018 Elsevier B.V. All rights reserved.
Intensity of volatility
Speed of reversion

1. Introduction

According to the report of World Health Organization (WHO), in 2015, infectious diseases are the second leading cause
of death worldwide, after heart disease, responsible for a quarter to a third of all deaths worldwide, and three (i.e., lower
respiratory infections, diarrhoeal disease and tuberculosis) of the top 10 causes of death were due to infectious diseases, and
caused 5.95 million deaths worldwide [1]. Understanding the mechanism that underlies the spread of an infectious disease
can give important insights to help in the fight against the disease itself [2,3]. It is now believed that mathematical models
have been important tools in analyzing the spread and control of infectious diseases, qualitatively and quantitatively [4–7].
Some diseases, such as some sexually transmitted and bacterial diseases, do not have permanent immunity. For these
diseases individuals start of susceptible, at some stage catch the disease and after a short infectious period become
susceptible again, and in addition, there is no protective immunity [8]. For these diseases, SIS (susceptible–infected–
susceptible) models are appropriate [9–11].

∗ Corresponding author.
E-mail addresses: wangwm_math@hytc.edu.cn (W. Wang), caiyongli@hytc.edu.cn (Y. Cai), dingding@hytc.edu.cn (Z. Ding), zhanjigui@sohu.com
(Z. Gui).

https://doi.org/10.1016/j.physa.2018.06.099
0378-4371/© 2018 Elsevier B.V. All rights reserved.
922 W. Wang et al. / Physica A 509 (2018) 921–936

For the sake of learning the disease transmission dynamics in a random environment, Gray et al. [8] studied the following
stochastic differential equations (SDE) SIS model:
dS(t) = ( µN − µS − β̄ SI
) + γ I dt − σ SIdB(t),
{ ( )
(1.1)
dI(t) = β̄ SI − µI − γ I dt + σ SIdB(t),
with initial values S(0) + I(0) = N. Here S(t) and I(t) are the numbers of the susceptible and the infectious at time t,
respectively. µ is the per capita birth/death rate, γ the recovery rate, β the disease transmission coefficient, B(t) standard
Brownian motion, σ the standard deviation of the noise. And the corresponding deterministic version of SDE model (1.1) is
dS(t) = ( µN − µS − β̄ SI
) + γ I dt ,
{ ( )
(1.2)
dI(t) = β̄ SI − µI − γ I dt .
Obviously, the SDE model (1.2) is obtained randomly by

β̄ → β̄ + σ dB(t) (1.3)
from the deterministic model (1.2).
Since S(t) + I(t) = N (N is a constant), it reduces to studying the following one-dimensional SDE:
([ )
β̄ (N − I(t)) − µ − γ dt + σ (t)[N − I(t)]dB(t) .
]
dI(t) = I(t) (1.4)

In [8], the authors defined the basic reproduction number in SDE SIS model (1.4):
β̄ N σ 2N 2
RS := − , (1.5)
µ+γ 2(µ + γ )
and obtained their main results: if RS < 1 and σ 2 ≤ β̄/N (see Theorem 4.1 in [8]), or σ 2 > max{β̄/N , β̄ 2 /2(µ + γ )} (see
Theorem 4.3 in [8]), the disease dies out with probability one; while if RS > 1, the disease persists with probability one (see
Theorem 5.1 in [8]), and model (1.1) has a unique stationary distribution (see Theorem 6.2 in [8]). The similar summary and
further study on SDE model (1.4) can be found in [12].
The so-called parameter perturbation method, such as (1.3), is a classical method of deriving an SDE model from its
deterministic counterpart, which is firstly proposed by Beddington and May [13], and applied widely [14–26]. Such
environmental noise reflected in specific parameters of the stochastic epidemic models may indicate whether severity of
disease increases or decreases [12]. Indeed this is a well-established way of introducing stochastic environmental noise into
biologically realistic population dynamic models [27].
It should be noted that in the parameter perturbation method above, parameter β̄ is assumed by linear functions of
Gauss white noise. Besides, there is another approach assumes that the parameters satisfy mean–reverting stochastic
processes [10,28–32].
More precisely, for model (1.1), the second possible model for β̄ in a randomly-varying environment is the mean-reverting
process (i.e., the Ornstein–Uhlenbeck process), which has the form:
( )
dβ̄ (t) = θ βe − β̄ (t) dt + ξ dB(t), (1.6)

where θ and ξ are all positive constants, θ the speed of reversion, ξ the intensity of volatility. And in the theory of Financial-
Economics, this format of mean-reverting process was studied by Dixit and Pindyck [33] firstly, and is also known as Dixit
& Pindyck model.
Through the stochastic integral format for the arithmetic Ornstein–Uhlenbeck process (1.6), we can obtain the following
explicit form solution:
∫ t
β̄ (t) = βe + (β0 − βe )e−θ t + ξ e−θ (t −s) dB(s), (1.7)
0

where β0 := β (0). Easy to know, the expected value of β (t) is:

E [β̄ (t)] = βe + (β0 − βe )e−θ t , (1.8)


and the variance of β (t) is:
ξ2 ( )
Var [β̄ (t)] = 1 − e−2θ t . (1.9)

∫t
Considering (1.7) again, from the results in (1.8) and (1.9), one can know that the term ξ 0
e−θ (t −s) dB(s) follows the normal
ξ2 (
( ))
1 − e−2θ t
∫t
distribution N 0, , and leads to that ξ 0 e−θ (t −s) dB(s) equal to

ξ √ dB(t)
√ 1 − e−2θ t , a.s.
2θ dt
W. Wang et al. / Physica A 509 (2018) 921–936 923

where B(t) is a standard Brownian motion. Thus, (1.7) can be almost surely written as follows:
dB(t)
β̄ (t) = βe + (β0 − βe )e−θ t + σ (t) , (1.10)
dt
where
ξ √
σ (t) = √ 1 − e−2θ t . (1.11)

Submitting (1.10) into model (1.2), we can obtain the following stochastic model:
dS(t) = ( µN − µS − (βe + (β0 − βe )e−θ t )SI
) + γ I dt − σ (t)SIdB(t)
{ ( )
(1.12)
dI(t) = (βe + (β0 − βe )e−θ t )SI − µI − γ I dt + σ (t)SIdB(t).
Given that S(t) + I(t) = N, it is sufficient to study the SDE for I(t),
[( ]
(βe + (β0 − βe )e−θ t )(N − I(t)) − µ − γ dt + σ (t)(N − I(t))dB(t) ,
)
dI(t) = I(t) (1.13)

with initial value I(0) = I0 ∈ (0, N). It should be noted that when θ → ∞, SDE model (1.13) will tend to ODE model (1.2).
Thanks to the insightful work of [8], in this paper, we will study the stochastic disease dynamics of the SDE model (1.13).
In Section 2, we will give the existence of the unique positive solution. In Section 3, we will study the stochastic extinction
and persistence of the disease. In Section 4, we give some numerical examples to show the complicated stochastic dynamics
of the model. And in the last section, Section 5, we provide a brief discussion and the summary of our main results.

2. Existence and uniqueness of the global positive solution

Although SDE model (1.13) is similar to the SDE model (1.4) in [8], but the existence and uniqueness theorem (Theorem
3.1 in [8]) is not applicable to (1.13) directly. It is therefore necessary to establish such a new theorem as follows.

Theorem 2.1. For any given initial value I0 ∈ (0, N), the SDE (1.13) has a unique global positive solution I(t) ∈ (0, N) for all
t ≥ 0 with probability one, namely,
{ }
P I(t) ∈ (0, N) : ∀t ≥ 0 = 1.

Proof. Consider the following model


ξ2 (
( )
u(t) 2
du(t) = βe N − µ − γ − βe eu(t)
dt + σ (t)(N − eu(t) )dB(t),
)
− N −e (2.1)

with initial value u(0) = log I0 . It is clear that the coefficients of model (2.1) satisfy the local Lipschitz condition, and there is
a local solution u(t), t ∈ [0, τe ) of model (2.1), where τe is the explosion time. Therefore, by Itô’s formula, it is easy to check
that I(t) = eu(t) is the positive solution of model (1.13) with the initial value I0 , and to show that the solution is global, we
only need to show that τe = ∞ a.s.
Let k0 > 0 be sufficiently large so that I0 ∈ (1/k0 , N − 1/k0 ). For each integer k ≥ k0 , define the stopping time

τk = inf {t ∈ [0, τe ) : I(t) ̸∈ (1/k, N − 1/k)} ,


where throughout this paper, we set inf ∅ = ∞ (as usual ∅ denotes the empty set). Clearly, τk is increasing as k → ∞. Set
τ∞ = limk→∞ τk , hence τ∞ ≤ τe a.s. If we can show that τ∞ = ∞ a.s., then τe = ∞ and I(t) ∈ (0, N) a.s. for all t ≥ 0. In
other words, to complete the proof all we need to show is that τ∞ = ∞ a.s. For if this statement is false, then there are a
pair of constants T > 0 and ε ∈ (0, 1) such that

P{τ∞ ≤ T } > ε.
Hence there is an integer k1 ≥ k0 such that

P{τk ≤ T } ≥ ε, ∀k ≥ k1 .
Define a function V : (0, N) → R+ by
1 1
V (I) = + .
I N −I
By the Itô’s formula, we have, for any t ∈ [0, T ] and k ≥ k1 ,
∫ t ∧τk
EV (I(t ∧ τk )) = V (I0 ) + E LV (I(s))ds,
0
924 W. Wang et al. / Physica A 509 (2018) 921–936

where LV (x) : (0, N) → R is defined by


( )(
1 1 )
LV (I) =I − 2 + (βe + (β0 − βe )e−θ t )(N − I) − µ − γ
I (N − I)2
( )
1 1
+σ 2 (N − I)2 I 2 + .
I3 (N − I)3
It is easy to come to
( )(
1 I )
LV (I) = − + (βe + (β0 − βe )e−θ t )(N − I) − µ − γ
I (N − I)2
(N − I)2 I2
( )
+σ 2 N 2 +
N 2I N 2 (N − I)
µ+γ βe + (β0 − βe )e−θ t
( )
1 1
≤ + + σ 2N 2 + .
I N −I I N −I
If β0 ≤ βe , we have βe + (β0 − βe )e−θ t ≤ βe ; while β0 > βe , we have βe + (β0 − βe )e−θ t ≤ β0 . Hence, one can get
LV (I) ≤ CV (I),

where C = (µ + γ ) ∨ (βe ∨ β0 ) + σ 2 N 2 . The remained proof is similar to that in [8]. And we omit it here. □

3. Stochastic dynamics of the disease

3.1. Stochastic extinction of the disease

Following [34,35] and [8,14], we define the basic reproduction number of the corresponding deterministic model of (1.13),
i.e., ξ = 0 in (1.13):
βe N
R0 := , (3.1)
µ+γ
and the stochastic basic reproduction number for the SDE model (1.13):
βe N ξ 2N 2
Rs0 := − . (3.2)
µ+γ 4θ (µ + γ )
With the help of Rs0 , we give the property of the stochastic extinction of model (1.13) as follows:

Theorem 3.1. If
2βe θ
Rs0 < 1 and ξ 2 ≤ , (3.3)
N
or
2βe θ β 2θ
{ }
ξ > max
2
, e , (3.4)
N µ+γ
then for any given initial value I0 ∈ (0, N), the solutions of the SDE model (1.13) obey
{ }
P lim I(t) = 0 = 1.
t →∞

Namely, the disease will die out with probability one.

Proof. By the Itô’s formula, we have,


∫ t ∫ t
log(I(t)) = log I0 + f (I(s))ds + ϕ (t) + σ (s)(N − I(s))dB(s), (3.5)
0 0

where f : R → R is defined by
ξ2
f (x) = βe N − µ − γ − βe x − (N − x)2 . (3.6)

and
ξ 2 e−2θ s
∫ t( )
ϕ (t) = (β0 − βe )e−θ s (N − I(s)) + (N − I(s))2 ds.
0 4θ
W. Wang et al. / Physica A 509 (2018) 921–936 925

It is easy to show that


ξ 2 N 2 t −2θ s
t
∫ ∫
ϕ (t) ≤ |β0 − βe |N e−θ s ds +
e ds
0 4θ 0
|β0 − βe |N ξ 2N 2
= (1 − e−θ t ) + (1 − e−2θ t ).
θ 8θ 2
and
ξ2 2 ξ 2N ξ 2N 2
( )
f (I(s)) = − I (s) − βe − I(s) + βe N − µ − γ − .
4θ 2θ 4θ
2βe θ
If ξ 2 < , we have
N
ξ 2N 2 βe N ξ 2N 2
( )
f (I) ≤ βe N − µ − γ − = (µ + γ ) − − 1 = (µ + γ )(Rs0 − 1)
4θ µ+γ 4θ (µ + γ )
for I(s) ∈ (0, N). By virtue of (3.5), we have
ϕ (t) t

1 log S(0) 1
log I(t) ≤ + + σ (s)(N − I(s))dB(s). (3.7)
t t t t 0

ϕ (t) ∫t ξ ∫t √
Obviously, limt →∞ = 0. Noting that 0 σ (s)(N − I)dB(s) = √ 0
1 − e−2θ s (N − I(s))dB(s) is a local martingale, it
t 2θ
follows from the strong law of large number of the local martingale that
∫ t
1
lim σ (s)(N − I)dB(s) = 0, a.s.
t →∞ t 0

Taking upper limit on both sides of (3.7) and making use of Rs0 < 1 yields
log I(t)
lim sup ≤ (µ + γ )(Rs0 − 1) < 0, a.s.
t →∞ t
Under condition (3.4),
)2
ξ2 ξ 2 N − 2βe θ β 2θ
(
f (I(s)) =− I(s) − + e2 − (µ + γ )
4θ ξ 2 ξ
βe2 θ
≤ 2 − (µ + γ )
ξ
for I(s) ∈ (0, N), which implies, in the same way as the above and by condition (3.4), that
log I(t) βe2 θ
lim sup ≤ − ( µ + γ ) < 0 a. s .
t →∞ t ξ2
as required. □

3.2. Stochastic persistence of the disease

Theorem 3.2. If Rs0 > 1, then for any given initial values I0 ∈ (0, N), the solutions of the SDE model (1.13) obey

lim sup I(t) ≥ ρ a.s. (3.8)


t →∞

and
lim inf I(t) ≤ ρ a.s., (3.9)
t →∞

where
N ξ 2 − 2θ βe + 2 θ (θβe2 − ξ 2 (µ + γ ))

ρ= (3.10)
ξ2
is the unique root in (0, N) of
ξ2
f (ρ ) = β e N − µ − γ − β e ρ − (N − ρ )2 = 0. (3.11)

That is, I(t) will rise to or above the level ρ infinitely often with probability one. Namely, the disease will persist with probability
one.
926 W. Wang et al. / Physica A 509 (2018) 921–936

Proof. In the view of Rs0 > 1, it is easy to see that the equation f (x) = 0 has a positive root and a negative root. The positive
one is defined by (3.10). Noting that
f (0) = (µ + γ )(Rs0 − 1) > 0, f (N) = −µ − γ < 0,
we see that ξ ∈ (0, N) and
ξ 2 N − 2βe θ
( )
f (x) > 0 is strictly increasing on, x ∈ 0, 0 ∨ , (3.12)
ξ2

ξ 2 N − 2βe θ
( )
f (x) > 0 is strictly decreasing on x ∈ 0∨ , ρ , (3.13)
ξ2
while
f (x) < 0 is strictly decreasing on x ∈ (ρ, N). (3.14)
We now begin to prove assertion (3.8). If it is not true, then there is a sufficiently small ε ∈ (0, 1) such that
P(Ω1 ) > ε,
where Ω1 = lim supt →∞ I(t) ≤ ρ − 2ε . Hence, for every ω ∈ Ω1 , there is a T = T (ω) > 0 such that
{ }

I(t , ω) ≤ ξ − ε, whenever t ≥ T (ω). (3.15)


Clearly we may choose ε so small (if necessary reducing it) that f (0) > f (ρ − ε ). It therefore follows from (3.12), (3.13), and
(3.15) that
f (I(t , ω)) ≥ f (ρ − ε ), whenever t ≥ T (ω). (3.16)
Moreover, by the large number theorem for martingales, there is an Ω2 ⊂ Ω with P(Ω2 ) = 1 such that for every ω ∈ Ω2 ,
∫ t
1
lim σ (s)(N − I(s, ω))dB(s, ω) = 0, a.s. (3.17)
t →∞ t 0

(i) If β0 > βe , we have


ξ 2 e−2θ s
∫ t( )
ϕ (t) = (β0 − βe )e −θ s
(N − I(s)) + (N − I(s)) ds ≥ 0.
2

0 4θ
for I(s) ∈ (0, N).
Now, fix any ω ∈ Ω1 ∩ Ω2 . It then follows from (3.16) and (3.17) that, for t ≥ T (ω),
∫ T (ω )
log(I(t , ω)) ≥ log I0 + f (I(s, ω))ds + f (ρ − ε )(t − T (ω))
0
∫ t
+ σ (s)(N − I(s, ω))dB(s, ω),
0

which yields
1
lim inf log(I(t , ω)) ≥ f (ρ − ε ) > 0,
t →∞ t
hence,
lim I(t , ω) = ∞.
t →∞

However, this contradicts (3.15). We therefore must have the desired assertion (3.8).
(ii) If β0 < βe , we have
t
(βe − β0 )N

ϕ (t) ≥ (β0 − βe )N e−θ s ds = (e−θ t − 1).
0 θ
Then, there is an Ω3 ⊂ Ω with P(Ω3 ) = 1 such that for every ω ∈ Ω3 ,
ϕ (t)
lim sup ≥ 0.
t →∞ t
Now, fix any ω ∈ Ω1 ∩ Ω2 ∩ Ω3 . By the same manner, we have limt →∞ I(t , ω) = ∞ which contradicts (3.15).
Let us now prove assertion (3.9). If it were not true, then there would be a sufficiently small δ ∈ (0, 1) such that
P(Ω4 ) > δ,
W. Wang et al. / Physica A 509 (2018) 921–936 927

where Ω4 = {lim inft →∞ I(t) ≥ ξ + 2δ}. Hence, for every ω ∈ Ω4 , there is a T1 = T1 (ω) > 0 such that

I(t , ω) ≥ ρ + δ, whenever t ≥ T1 (ω). (3.18)

By 3.1, there is an Ω5 ⊆ Ω with P(Ω5 ) = 1 such that for every ω ∈ Ω5 ,


ϕ (t , ω)
lim = 0. (3.19)
t →∞ t
Now, fix any ω ∈ Ω2 ∩ Ω4 ∩ Ω5 . It then follows from (3.5), (3.17) and (3.19), that, for t ≥ T1 (ω),
∫ T 1 (ω )
log(I(t , ω)) ≤ log I0 + f (I(s, ω))ds + f (ρ + δ )(t − T1 (ω))
0
∫ t
+ϕ (t) + σ (s)(N − I(s, ω))dB(s, ω),
0

which, together with (3.18), yields


1
lim sup log(I(t , ω)) ≤ f (ρ + δ ) < 0,
t →∞ t
hence,

lim I(t , ω) = 0.
t →∞

However, this contradicts (3.18). We therefore must have the desired assertion (3.9). The proof is hence complete. □

Remark 3.3. Theorem 3.1 gives the sufficient conditions when the solutions to model (1.13) are converging to the disease-
free dynamics a.s., that is, almost all solutions of model (1.13) tend to 0 with probability one. And Theorem 3.2 gives the
sufficient conditions when the solutions to model (1.13) are converging to the endemic dynamics a.s., that is, almost all
solutions of model (1.13) will rise to or above the positive constant level ρ with probability one. Summarily, Rs0 can be used
to govern the stochastic dynamics of SDE model.

ξ 2N 2
Remark 3.4. From (3.1) and (3.2), we know that Rs0 = R0 − < R0 . And we can easily find an example when
4θ (µ + γ )
R0 > 1 but Rs0 < 1 such that in the deterministic case, i.e., ξ = 0 in (1.13), I(t) will persist, while in the stochastic case,
i.e., ξ > 0 in (1.13), the infectious I(t) goes extinct almost surely, which implies that large environmental fluctuations can
suppress the outbreak of disease.

Next, we will focus on the effects of the speed of reversion θ and the intensity of volatility ξ on the level ρ further. Also,
we will pay attention to the limit dynamics of SDE model (1.13) with respect to ξ and θ , respectively.

Theorem 3.5. Assume that Rs0 > 1, and regard ρ defined by (3.10) as a function of ξ for
√ ( )
2 θ β e N − (µ + γ )
0<ξ < := ξ ∗ . (3.20)
N
Then ρ is strictly decreasing,
( )
1
lim ρ = N 1−
ξ →0+ R0
and
⎨0,( if 1 < R0 ≤ 2,

lim ρ =
)
1
ξ →ξ ∗− ⎩N 1 − , if R0 > 2.
R0 − 1

Proof. Compute
( )
2θ ξ 2 (µ + γ ) − 2θβe2 + 2βe θ (θβe2 − ξ 2 (µ + γ ))


=
dξ ξ 3 θ (θβe2 − ξ 2 (µ + γ ))

(√ )2
2 θ (θβe2 − ξ 2 (µ + γ )) − θβe
=− .
ξ 3 θ (θβe2 − ξ 2 (µ + γ ))

928 W. Wang et al. / Physica A 509 (2018) 921–936


Since ξ ̸ = 0, we have θ (θβe2 − ξ 2 (µ + γ )) ̸= 0. We therefore have that < 0 which implies that ρ is strictly decreasing


as ξ increases. Moreover, by the well-known L’Hopital’s rule,

1 ( )−3/2
lim ρ 2N − 4θ 2 ξ 2 (µ + γ )2 θ (θβe2 − ξ 2 (µ + γ ))
(
= lim
ξ →0+ 2 ξ →0+
( )−1/2 )
−2θ (µ + γ ) θ (θβe2 − ξ 2 (µ + γ ))
µ+γ
( ) ( )
1
=N 1− =N 1−
βe N R0

as desired. Furthermore, it is obvious that



N ξ̂ 2 − 2θβe + 2 θ (θβe2 − ξ̂ 2 (µ + γ ))
lim ρ = lim
σ →ξ̂ − σ→(ξ̂ − ξ̂ 2 )
N |βe N − 2(µ + γ )| + βe N − 2(µ + γ )
=
2(βe N − µ − γ )
( )
N |R0 − 2| + R0 − 2
= .
2(R0 − 1)
( )
1
Hence if 1 < R0 ≤ 2, we have limξ →ξ ∗− ρ = 0, while if R0 > 2, we have limξ →ξ ∗− ρ = N 1− . The proof is
R0 − 1
complete. □

Theorem 3.6. Assume that Rs0 > 1, and regard ρ defined by (3.10) as a function of θ for

ξ 2N 2
θ> ( ) := θ ∗ . (3.21)
4 β e N − (µ + γ )

Then ρ is strictly increasing. Moreover,


( )
1
lim ρ = N 1−
θ →∞ R0

and
⎨0,( if 1 < R0 ≤ 2,

lim ρ =
)
1
θ →θ ∗ ⎩N 1 − , if R0 > 2.
R0 − 1

Proof. Compute

dρ 2θ βe2 − ξ 2 (µ + γ ) − 2βe θ (θβe2 − ξ 2 (µ + γ ))



=
dθ ξ 3 θ (θβe2 − ξ 2 (µ + γ ))

(√ )2
θ (θ βe2 − ξ 2 (µ + γ )) − θβe
= .
θ ξ 2 θ (θβe2 − ξ 2 (µ + γ ))


Since ξ ̸ = 0 we have θ (θβe2 − ξ 2 (µ + γ )) ̸= 0. We therefore have that > 0 which implies that ρ is strictly increasing


as θ increases. Next, we obtain
√ )2
ξ 2 (µ + γ ) ξ 4 (µ + γ ) 2
(
N ξ − 2θ βe + 2 βe2
2
θ− −
2βe2 4βe2
( )
1
ρ= ≤N 1− .
ξ2 R0

Hence, ρ is strictly monotonically increasing with θ and bounded. Moreover, we calculate


( )
1
lim ρ = N 1−
θ →∞ R0
W. Wang et al. / Physica A 509 (2018) 921–936 929

Fig. 1. The relations between RS , Rs0 with θ .

and

N ξ 2 − 2θ̂βe + 2 θ̂ (θ̂βe2 − ξ 2 (µ + γ ))
lim ρ = lim∗
θ →θ ∗ θ →θ
( ξ2 )
N |βe N − 2(µ + γ )| + βe N − 2(µ + γ )
=
( 2(βe N − µ
) − γ)
N |R0 − 2| + R0 − 2
= ,
2(R0 − 1)
as desired. □

Remark 3.7. Although the definition of the stochastic reproduction number Rs0 in (3.2) for SDE model (1.13) is similar to
that in (1.5) for model (1.4) by [8]. For the sake of showing the differences between Rs0 with RS , we adopt the same parameter
values as Example 5.2 in [8]:
βe = β̄ = 0.5, N = 200, µ = 20, γ = 25, β0 = 0.45. (3.22)

Example 3.7.1. In Fig. 1, we adopt ξ = σ = 0.03, and show the relations between RS and Rs0 with θ . As shown in Example
5.2 in [8], RS = 1.01111 > 1, the disease of SDE model (1.4) will persist a.s. But this is not true for our SDE model (1.13)
forever. We can see that if θ ≤ θ ∗ , then Rs0 ≤ 1, the disease of SDE model (1.13) will die out a.s.; while if θ > θ ∗ , then
Rs0 > 1, the disease of SDE model (1.13) will persist a.s.

Example 3.7.2. In Fig. 2, we adopt ξ = σ , and show the relations between RS and Rs0 with ξ with different values of θ . In
1
Fig. 2(a), θ < , when ξ > ξ ∗ , the disease will go extinct a.s. for SDE model (1.13); while if ξ > ξ∗ (here ξ∗ is the root of ξ
2
for RS = 1), the disease will go extinct a.s. for Gray’s model (1.4). Noting that ξ ∗ < ξ∗ , that is to say, when ξ ∗ < ξ < ξ∗ , the
1
disease will die out a.s. for SDE model (1.13), but will persist a.s. for Gray’s model (1.4). In Fig. 2(b), θ = , RS = Rs0 , SDE
2
1
model (1.13) has the same stochastic dynamics as Gray’s model (1.4). And in Fig. 2(c), θ > , noting that ξ ∗ > ξ∗ , that is to
2
say, when ξ∗ < θ < ξ ∗ , the disease will persist a.s. for SDE model (1.13), but die out a.s. for Gray’s model (1.4).

4. Numerical results

In this section, we give some numerical results to show complex disease dynamic outcomes of the SDE model (1.13) by
using the Milstein’s method mentioned in Higham [36]. In this way, the numerical scheme for the Milstein’s method applied
930 W. Wang et al. / Physica A 509 (2018) 921–936

Fig. 2. The relations between RS , Rs0 with ξ .

to the stochastic model (1.13) under consideration is given by


[( ]
βe + (β0 − βe )e−θ (k∆t) (N − Ik ) − µ − γ ∆t
)
Ik+1 = Ik + Ik

ξ 1 − e−θ (k∆t) √ ξ 2 (1 − e−θ (k∆t) )
+ √ (N − Ik )Ik ηk ∆t + (N − Ik )2 Ik2 (ηk2 − ∆t),
2θ 4θ
where ηk (k = 1, 2, . . . , n) are independent Gaussian random variables N(0, 1), ∆t is the time step size.
The main goal of this section is to further study the effects of the speed of reversion θ and the intensity of volatility ξ on
the stochastic disease dynamics of SDE model (1.13). For the sake of learning the differences between perturbation (1.6) (or
(1.7) ) and (1.3) in [8], the parameters set (3.22) are taken the same as in [8], and the initial value I(0) = 10.

4.1. The effect of the intensity of volatility ξ

In this subsection, we will study the effect of the intensity of volatility ξ on the disease dynamics of model (1.13).
We firstly adopt θ = 0.75. If we choose ξ = σ = 0.055, simple computations show that: RS = 0.8778 < 1, which
implies that the disease in SDE model (1.4) will die out with probability one [8]. But this is not true for SDE model (1.13)
because Rs0 = 1.3259 > 1. From Theorem 3.2, we know that the disease will persist a.s. In addition, we compute

N ξ 2 − 2θ βe + 2 θ (θβe2 − ξ 2 (µ + γ ))

ρ= = 81.8471,
ξ2
and therefore conclude, by Theorem 3.2, that for I(0) = 10, the solutions of (1.13) obey
lim inf I(t) ≤ 81.8471 ≤ lim sup I(t) a.s..
t →∞ t →∞

The computer simulations in Fig. 3(a) support these results clearly, illustrating fluctuation around the level 81.8471. While
if we choose ξ = 0.066, then Rs0 = 0.9316 < 1. From Theorem 3.1, we know that the disease will die out a.s. (see Fig. 3(b)),
which is similar to that in [8].
From Fig. 3, we can conclude that lower intensity of volatility ξ can benefit the disease outbreak, while higher intensity
of volatility ξ can suppress the disease outbreak.
Furthermore, from the proof of Theorem 3.5, we know that ρ is strictly decreasing as ξ increases. If we choose ξ sufficiently
small, e.g., ξ = 0.0001 → 0+ , then we compute

N ξ 2 − 2θ βe + 2 θ (θβe2 − ξ 2 (µ + γ ))

ρ= = 110.0,
ξ2
which is the maximum of ρ . We can therefore conclude, by Theorem 3.2, that for I(0) = 10, the solutions of (1.13) obey
lim inf I(t) ≤ 110.0 ≤ lim sup I(t) a.s..
t →∞ t →∞
( 1 )
And it is easy to know that I ∗ := N 1 − = 110.0 is the positive equilibrium point of the corresponding autonomous
R0
system
( )
dI(t) = I(t) βe (N − I(t)) − µ − γ dt . (4.1)
W. Wang et al. / Physica A 509 (2018) 921–936 931

Fig. 3. The evolution of a single path of I(t) for the SDE model (1.13) and its corresponding deterministic version (i.e., ξ = 0 in (1.13)) and all other
parameters are taken in (3.22) and θ = 0.75. The initial value of all solution is I(0) = 10. The time unit is day. (a) Stochastic persistence a.s. with ξ = 0.055;
(b) Stochastic extinction a.s. with ξ = 0.066.

Fig. 4. The evolution of a single path of I(t) for the SDE model (1.13) and its corresponding deterministic version (i.e., ξ = 0 in (1.13)) and all other
parameters are taken in (3.22) and θ = 0.75, ξ = 0.0001. The initial value of all solution is I(0) = 10. The time unit is day.

That is to say, if ξ → 0+ , then the solutions of SDE model (1.13) get close to the limiting value of the corresponding
deterministic SIS model (4.1). This is proved by Theorem 3.5 rigorously. And the numerical results can be found in
Fig. 4.
Furthermore, if we choose ξ sufficiently big, e.g., ξ = 0.064 → ξ ∗ ≈ 0.064226. In this case, R0 = 2.2222 > 2. And we
compute ρ = 40.74046, and conclude, by Theorem 3.5, that for I(0) = 10, the solutions of (1.13) obey

lim inf I(t) ≤ 40.74046 ≤ lim sup I(t) a.s..


t →∞ t →∞

And the numerical results are similar to that in Fig. 3(a). In addition, if we increase N to 100, then 1 < R0 = 1.1111 < 2,
and the disease will go extinct a.s., the numerical results are similar to that in Fig. 3(b).
( 1 )
On the other hand, easy to know that I ∗ := N 1 − = 40.74046 is the positive equilibrium point of the
R0 − 1
corresponding autonomous system

µ+γ)
(( )
dI(t)
= I(t) βe − (N − I(t)) − µ − γ , (4.2)
dt N
932 W. Wang et al. / Physica A 509 (2018) 921–936

Fig. 5. The evolution of a single path of I(t) for the SDE model (1.13) and its corresponding deterministic version (i.e., ξ = 0 in (1.13)) and all other
parameters are taken in (3.22) and σ = 0.03. The initial value of all solution is I(0) = 10. The time unit is day. (a) Stochastic extinction a.s. with θ = 0.40;
(b) Stochastic persistence a.s. with θ = 0.55.

or the following SIS model



µ+γ)
( )
dS(t) (
= µN − µS − βe − SI + γ I ,



dt N

(4.3)
µ+γ)
(( )
⎪ dI(t)
βe − SI − µI − γ I .


⎩ =
dt N
That is to say, if ξ → ξ ∗ , then the solutions of SDE model (1.13) get close to the limiting value of the ODE model (4.2) or SIS
system (4.3).

4.2. The effect of the speed of reversion θ

Next, we will focus on the effect of the speed of reversion θ on the disease dynamics of model (1.13). We adopt
ξ = 0.03 = σ . In this case, for the SDE model (1.4) in [8], RS = 1.0111, the disease will persist with probability one.
If we choose θ = 0.40, simple computation shows that Rs0 = 0.9861 < 1. From Theorem 3.1, we know that the disease
will die out a.s. The computer simulations in Fig. 5(a) support these results clearly; while if we choose θ = 0.55, then
Rs0 = 1.7222 > 1, from Theorem 3.2, we know that the disease will persist a.s. In addition, we compute

N ξ 2 − 2θ βe + 2 θ (θβe2 − ξ 2 (µ + γ ))

ρ= = 102.1693,
ξ2
we can therefore conclude, by Theorem 3.2, that for I(0) = 10, the solutions of (1.13) obey
lim inf I(t) ≤ 102.1693 ≤ lim sup I(t) a.s..
t →∞ t →∞

The simulation results in Fig. 5(b) support these results clearly, showing fluctuation around the level 102.1693. From Fig. 5,
we can conclude that small speed of reversion θ can suppress the disease outbreak, while big speed of reversion θ can cause
the disease outbreak.
In addition, if we choose θ sufficiently small, e.g., θ = 0.1636 ≈ θ ∗ , in this case, R0 = 2.2222 > 2. And we compute

N ξ 2 − 2θ βe + 2 θ (θβe2 − ξ 2 (µ + γ ))

ρ= = 36.1989,
ξ2
we can therefore conclude, by Theorem 3.6, that for I(0) = 10, the solutions of (1.13) obey
lim inf I(t) ≤ 36.1989 ≤ lim sup I(t) a.s..
t →∞ t →∞
( 1 )
And easy to know that I ∗ = N 1 − = 36.1989 is the positive equilibrium point of the corresponding autonomous
R0 − 1
system (4.2) or SIS system (4.3). That is, if θ → θ ∗ , then the solutions of SDE model (1.13) get close to the limiting value of
the deterministic models (4.2) or SIS system (4.3). This is rigorously proved by Theorem 3.6.
W. Wang et al. / Physica A 509 (2018) 921–936 933

Fig. 6. The histograms of the values of the path I(t) for SDE model (1.13) with σ = 0.03 based on 10 000 stochastic simulations.

Furthermore, with the parameters set (3.22) and ξ = 0.03, we can obtain limθ →∞ ρ = 110.0. As discussed in Section
( 1 )
4.1 above, we know I ∗ = N 1 − = 110.0 is the positive equilibrium point of the corresponding autonomous system
R0
(4.1). That is to say, if θ → ∞, then the solutions of SDE model (1.13) get close to the limiting value of the deterministic
model (4.1). The numerical results, e.g., θ = 10 000, are similar to that in Fig. 4. This is of course not surprising, as discussed
in Section 1, when θ → ∞, SDE model (1.13) will tend to the deterministic model (1.2).

4.3. Stationary distribution

For the sake of learning the effects of the speed of reversion θ and the intensity of volatility ξ on the stochastic disease dy-
namics of SDE model (1.13), we have repeated the simulation 10 000 times, keeping all parameters fixed and never observing
any extinction scenario up to t = 10. These results are respectively confirmed by the histograms in Figs. 6 and 7, showing
the stationary distributions of I(t) at t = 10 for model (1.13). And the numerical method for them can be found in [14].
In Fig. 6, we study the effect of the speed of reversion θ on the stationary distribution of I(t) at t = 10 for model (1.13).
For this purpose, we fix ξ = 0.03, and consider three different values of θ : 0.25, 0.5 and 1, and other parameters are taken
as (3.22). The corresponding values of Rs0 are 1.4222, 1.8222 and 2.0222, respectively. From Theorem 3.2, we know that, in
934 W. Wang et al. / Physica A 509 (2018) 921–936

Fig. 7. The histograms of the values of the path I(t) for SDE model (1.13) with θ = 1.0 based on 10000 stochastic simulations.

the cases above, the disease will persist with probability one. From Fig. 6, we can see that, when θ = 1.0, the distribution
appears closer to a normal distribution (see Fig. 6(c)), but as θ decreases to 0.25, the distribution is positively skewed (see
Fig. 6(a)). And Fig. 6(b), θ = 0.5, is the same as Gray’s SDE model (1.4). We can conclude that for higher θ (e.g., θ = 1.0), the
amplitude of fluctuation is slight and the oscillations are more symmetrically distributed; while for lower θ (e.g., θ = 0.25),
the amplitude of fluctuation is remarkable and the distribution of the solution is skewed.
In Fig. 7, we study the effect of ξ on the stationary distribution of I(t) at t = 10 for model (1.13). For this reason, we fix θ =
1.0, and consider three different values of ξ : 0.01, 0.03 and 0.05, and other parameters are taken as (3.22). The corresponding
values of Rs0 are 2.20, 2.1333 and 1.6667, respectively. From Theorem 3.2, we know that, in the cases above, the disease will
persist with probability one. From Fig. 7, we can see that, when ξ = 0.01, the distribution appears closer to a normal
distribution (see Fig. 7(a)), but as ξ decreases to 0.05, the distribution is positively skewed (see Fig. 7(c)). We can conclude
that for lower ξ (e.g., ξ = 0.01), the amplitude of fluctuation is slight and the oscillations are more symmetrically distributed;
while for lower ξ (e.g., ξ = 0.05), the amplitude of fluctuation is remarkable and the distribution of the solution is skewed.

5. Concluding remarks

In this paper, based on the results of [8], we propose a new SDE SIS model incorporating mean–reverting Ornstein–
Uhlenbeck process. The mean–reverting process possesses important features that better characterize environmental
W. Wang et al. / Physica A 509 (2018) 921–936 935

variability in biological systems than does a linear function of white noise [10]. Most importantly, the mean-reverting
process over a linear function of Gaussian white noise is continuity, nonnegativity, practicality, possession of asymptotic
distributions and so on [29]. We partially provide the effects of the environmental fluctuations (measured by the intensity of
volatility ξ or the speed of reversion θ ) on the disease spreading to SDE model (1.13). With the stochastic basic reproduction
number Rs0 , we can identify the stochastic extinction and persistence for (1.13): if Rs0 < 1 under mild extra conditions, the
disease will go to extinction a.s. (c.f., Theorem 3.1), while if Rs0 > 1, the disease will persist a.s. (c.f., Theorem 3.2). In addition,
considering (3.2), (3.20) and (3.21) together, we know that Rs0 > 1 implies that θ < θ ∗ or ξ > ξ ∗ , and Rs0 < 1 implies that
θ > θ ∗ or ξ < ξ ∗ . In this sense, θ ∗ and ξ ∗ can be used to govern the dynamics of SDE model (1.13). More precisely, if θ > θ ∗
or ξ < ξ ∗ , the disease will die out a.s.; while if θ < θ ∗ or ξ > ξ ∗ , the disease will persist a.s. These results show that smaller
speed of reversion θ or bigger intensity of volatility ξ can suppress the disease outbreak, while bigger speed of reversion θ
or smaller intensity of volatility ξ can benefit the disease outbreak.
And in Theorems 3.1 and 3.2, we give the limit dynamics of SDE model (1.13) with respect to ξ (see Theorem 3.1) and θ
(see Theorem
( 3.2),
) respectively. And in the different limit process, the solutions of SDE model ((1.13) get close ) to the limiting
1 1
value N 1 − R0
corresponding to the deterministic model (4.1), or to the limiting value N 1 − R0 − 1
corresponding to
µ+γ
the deterministic model (4.2). The incidence rate in model (4.1) is βe , and βe − N in model (4.2). That is, when ξ → ξ ∗
µ+γ
or θ → θ ∗ , the incidence rate decreases from βe to βe − N . Hence we can claim that increasing ξ or decreasing θ will be
useful to decrease the incidence rate, and beneficial to control the disease spread. Thus, in order to control the spread of the
disease, we must increase the intensity of volatility ξ or decrease the speed of reversion θ .
On the other hand, assuming ξ = σ , when θ = 1/2, and RS = Rs0 , SDE model (1.13) has the same stochastic dynamics
as Gray’s model (1.4). And in the case of θ < 1/2, when ξ ∗ < ξ < ξ∗ , the disease will die out a.s. for SDE model (1.13), but
persist a.s. for Gray’s model (1.4); in the case of θ > 1/2, when ξ∗ < ξ < ξ ∗ , the disease will persist a.s. for SDE model (1.13),
but die out a.s. for Gray’s model (1.4). Furthermore, with the parameters (3.22), the disease of SDE model (1.4) persists with
probability one. In contrast to [8], we find that SDE model perturbed by (1.6) or (1.7) exhibits rich dynamics, and the disease
not only persists a.s. (see Fig. 5(b) and Fig. 3(a)), but also goes extinct a.s. (see Fig. 5(a) and Fig. 3(b)). These results show that
different perturbation method can exhibit different stochastic dynamics.
Furthermore, in [12], based on the results of [8], Xu gave the global threshold dynamics of SDE model (1.4) through
investigating the Fokker–Planck equation associated with (1.4), an autonomous SDE model. Unfortunately, this method
cannot be used to study SDE model (1.13) which is a nonautonomous SDE model. Hence, the extinct results in Theorem 3.1
involve mild extra conditions. And the global threshold dynamics of autonomous SDE model, e.g., (1.13), is desirable in future
study.

Acknowledgments

This research was supported by the National Science Foundation of China (Grant No. 61672013, 11601179 and 61772017),
the Natural Science Foundation of the Jiangsu Higher Education Institutions of China (16KJB110003) and Huaian Key
Laboratory for Infectious Diseases Control and Prevention, PR China (HAP201704).

References

[1] World Health Organization. The top 10 causes of death, 2017. http://www.who.int/mediacentre/factsheets/fs310/en/.
[2] Z. Ma, Y. Zhou, J. Wu, Modeling and Dynamics of Infectious Diseases, Higher Education Press, 2009.
[3] Y. Cai, Y. Kang, M. Banerjee, W.M. Wang, A stochastic epidemic model incorporating media coverage, Commun. Math. Sci. 14 (4) (2016) 893–910.
[4] H.W. Hethcote, The mathematics of infectious diseases, SIAM Rev. 42 (4) (2000) 599–653.
[5] F. Brauer, C. Castillo-Chavez, Mathematical Models in Population Biology and Epidemiology, second ed., Springer, 2012.
[6] Y. Cai, W.M. Wang, Fish-hook bifurcation branch in a spatial heterogeneous epidemic model with cross-diffusion, Nonlinear Anal. RWA 30 (2016)
99–125.
[7] Y. Cai, Y. Kang, M. Banerjee, W.M. Wang, Complex dynamics of a host-parasite model with both horizontal and vertical transmissions in a spatial
heterogeneous environment, Nonlinear Anal. RWA 40 (2018) 444–465.
[8] A. Gray, D. Greenhalgh, L. Hu, X. Mao, J. Pan, A stochastic differential equation sis epidemic model, SIAM J. Appl. Math. 71 (3) (2011) 876–902.
[9] H.W. Hethcote, J.A. Yorke, Gonorrhea Transmission Dynamics and Control, Springer–Verlag, 1984.
[10] Y. Cai, J. Jiao, Z. Gui, Y. Liu, W.M. Wang, Environmental variability in a stochastic epidemic model, Appl. Math. Comput. 329 (2018) 210–226.
[11] W. Guo, Y. Cai, Q. Zhang, W.M. Wang, Stochastic persistence and stationary distribution in an SIS epidemic model with media coverage, Physica A 492
(2018) 2220–2236.
[12] C. Xu, Global threshold dynamics of a stochastic differential equation SIS model, J. Math. Anal. Appl. 447 (2) (2017) 736–757.
[13] J.R. Beddington, R.M. May, Harvesting natural populations in a randomly fluctuating environment, Science 197 (4302) (1977) 463–465.
[14] Y. Cai, Y. Kang, M. Banerjee, W.M. Wang, A stochastic SIRS epidemic model with infectious force under intervention strategies, J. Differential Equations
259 (12) (2015) 7463–7502.
[15] W.M. Wang, Y. Cai, J. Li, Z. Gui, Periodic behavior in a FIV model with seasonality as well as environment fluctuations, J. Franklin Inst. 354 (16) (2017)
7410–7428.
[16] Y. Liu, W. Li, J. Feng, Graph-theoretical method to the existence of stationary distribution of stochastic coupled systems, J. Dynam. Differential
Equations (2016). http://dx.doi.org/10.1007/s10884-016-9566-y.
[17] H. Huo, F. Cui, H. Xiang, Dynamics of an SAITS alcoholism model on unweighted and weighted networks, Physica A 496 (2018) 249–262.
[18] Y. Liu, W. Li, J. Feng, The stability of stochastic coupled systems with time-varying coupling and general topology structure, IEEE T. Neur. Net. Lear. PP
(99) (2017) 1–12.
936 W. Wang et al. / Physica A 509 (2018) 921–936

[19] D. Li, J. Cui, M. Liu, S. Liu, The evolutionary dynamics of stochastic epidemic model with nonlinear incidence rate, Bull. Math. Biol. 77 (9) (2015)
1705–1743.
[20] J. Li, M. Shan, M. M. Banerjee, W.M. Wang, Stochastic dynamics of feline immunodeficiency virus within cat populations, J. Franklin Inst. 353 (16)
(2016) 4191–4212.
[21] M. Liu, C. Bai, Analysis of a stochastic tri-trophic food-chain model with harvesting, J. Math. Biol. 73 (2016) 597–625.
[22] M. Liu, M. Fan, Permanence of stochastic Lotka–Volterra systems, J. Nonlinear Sci. 27 (2) (2017) 425–452.
[23] X. Yu, S. Yuan, T. Zhang, The effects of toxin-producing phytoplankton and environmental fluctuations on the planktonic blooms, Nonlinear Dynam.
91 (2018) 1653–1668.
[24] Q. Liu, D. Jiang, T. Hayat, B. Ahmad, Analysis of a delayed vaccinated sir epidemic model with temporary immunity and Lévy jumps, Nonlinear Anal.
Hybr. 27 (2018) 29–43.
[25] Q. Liu, D. Jiang, N. Shi, Threshold behavior in a stochastic SIQR epidemic model with standard incidence and regime switching, Appl. Math. Comput.
316 (2018) 310–325.
[26] Q. Liu, D. Jiang, T. Hayat, B. Ahmad, Stationary distribution and extinction of a stochastic SIRI epidemic model with relapse, Stoch. Anal. Appl. 36 (2018)
138–151.
[27] Y. Cai, Y. Kang, W.M. Wang, A stochastic SIRS epidemic model with nonlinear incidence rate, Appl. Math. Comput. 305 (2017) 221–240.
[28] Y. Zhao, S. Yuan, J. Ma, Survival and stationary distribution analysis of a stochastic competitive model of three species in a polluted environment, Bull.
Math. Biol. 77 (2015) 1285–1326.
[29] E. Allen, Environmental variability and mean-reverting processes, Discrete Contin. Dyn. Syst. Ser. B 21 (7) (2016) 2073–2089.
[30] D. Duffie, Dynamic Asset Pricing Theory, Princeton University Press, 1996.
[31] F. Wu, X. Mao, K. Chen, A highly sensitive mean-reverting process in finance and the Euler–Maruyama approximations, J. Math. Anal. Appl. 348 (1)
(2008) 540–554.
[32] D.C. Trost, I.I. Overman, E.A. Ostroff, J.H. Xiong, A model for liver homeostasis using modified mean-reverting Ornstein–Uhlenbeck process, Comput.
Math. Methods Med. 11 (1) (2010) 27–47.
[33] A.K. Dixit, R.S. Pindyck, Investment under Uncertainty, Princeton University Press, 1994.
[34] O. Diekmann, J.A.P. Heesterbeek, J.A.J. Metz, On the definition and the computation of the basic reproduction ratio r0 in models for infectious diseases
in heterogeneous populations, J. Math. Biol. 28 (4) (1990) 365–382.
[35] P. Van den Driessche, J. Watmough, Reproduction numbers and sub-threshold endemic equilibria for compartmental models of disease transmission,
Math. Biosci. 180 (1) (2002) 29–48.
[36] D.J. Higham, An algorithmic introduction to numerical simulation of stochastic differential equations, SIAM Rev. 43 (3) (2001) 525–546.

You might also like