Académique Documents
Professionnel Documents
Culture Documents
Research Article
The Hopf Bifurcation Analysis and Optimal Control of
a Delayed SIR Epidemic Model
Copyright © 2014 Abdelhadi Abta et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
We propose a delayed SIR model with saturated incidence rate. The delay is incorporated into the model in order to model the latent
period. The basic reproductive number 𝑅0 is obtained. Furthermore, using time delay as a bifurcation parameter, it is proven that
there exists a critical value of delay for the stability of diseases prevalence. When the delay exceeds the critical value, the system loses
its stability and a Hopf bifurcation occurs. The model is extended to assess the impact of some control measures, by reformulating
the model as an optimal control problem with vaccination and treatment. The existence of the optimal control is also proved. Finally,
some numerical simulations are performed to verify the theoretical analysis.
effect, 𝛾 is the recovery rate of the infectious individuals, and Now, we will find the equilibria of system (2) and investigate
𝜏 is the incubation period. their dynamical features. System (2) always has two disease-
This model takes into account a number of key biological free equilibria 𝑃0 = (0, 0) and 𝑃1 = (𝐾, 0). Further, if
assumptions.
𝛽𝐾
(i) It is more reasonable to assume that the population of 𝑅0 = > 1, (3)
𝜇+𝛼+𝛾
a given region obeys logistic growth due to crowding
and limited sources. Epidemic models with logistic or system (2) admits a unique endemic equilibrium 𝑃∗ =
generalized logistic demographic structure have been (𝑆∗ , 𝐼∗ ), given by
extensively studied (see, e.g., [6, 7]).
(𝜇1 + 𝛾) (1 + 𝛼2 𝐼∗ )
(ii) Since nonlinearity in the incidence rates has been 𝑆∗ = ,
observed in disease transmission dynamics, it has 𝛽
been suggested that the standard bilinear incidence (4)
rate will be modified into a nonlinear incidence rate ∗
[𝛼2 𝑟𝛽𝐾 − 2𝛼2 𝑟 (𝜇1 + 𝛾) − 𝛽2 𝐾] + √Δ
𝐼 = ,
by many authors (see, e.g., [8, 9]). In this work we use 2𝑟𝛼22 (𝜇1 + 𝛾)
a nonlinear incidence rate of the form 𝛽𝑆𝐼/(1 + 𝛼2 𝐼).
with
The aim is to gain some insights into the best intervention for
2
minimizing the transmission of disease within the population Δ = [𝛼2 𝑟𝛽𝐾 − 2𝛼2 𝑟 (𝜇1 + 𝛾) − 𝛽2 𝐾]
and to explore the impacts of various intervention scenarios, (5)
namely, vaccination and treatment. We analyse the stability + 4𝛼22 𝑟 (𝜇1 + 𝛾) (𝑟𝛽𝐾 − 𝑟 (𝜇1 + 𝛾)) .
and the Hopf bifurcation of the model; then, we incorporate
into the model appropriate cost functions in order to study Now, let us start to discuss the local behavior of the disease-
and determine the possible impacts of these strategies on free equilibrium 𝑃0 . We linearize system (2) around 𝑃0 ; we
controlling the disease. We give the necessary conditions for have
optimal control of the disease using Pontryagin’s maximum
principle, in order to determine optimal strategies for con- 𝑑𝑆
= 𝑟𝑆 (𝑡) ,
trolling the spread of the disease. 𝑑𝑡
The organization of the paper is as follows: we examine (6)
𝑑𝐼
the existence and stability behaviour of the equilibrium = − (𝜇1 + 𝛾) 𝐼 (𝑡) .
solutions of this model and we are particularly interested in 𝑑𝑡
whether the unique endemic equilibrium can be destabilized The characteristic equation associated with system (6) is
by introducing a time delay to represent the effect of fading
of vaccination and whether stable endemic cycles can arise Δ (𝜆, 𝜏) = (𝜆 − 𝑟) (𝜆 + 𝜇1 + 𝛾) = 0. (7)
by the Hopf bifurcation from the initially stable endemic
equilibrium in Section 2. In Section 3, we formulate an Proposition 1. The disease-free equilibrium point 𝑃0 is unsta-
optimal control problem and we use Pontryagin’s maximum ble.
principle with delay given in [10] to characterize it. Our
conclusions are discussed in Section 4. Proof. It is clear that (7) has two roots 𝜆 1 = 𝑟 > 0 and 𝜆 2 =
−(𝜇1 + 𝛾) < 0. Whence 𝑃0 is unstable.
2. Stability Analysis and the Hopf We turn to study the local behavior of the disease-free
Bifurcation Occurrence equilibrium 𝑃1 .
Let 𝑥 = 𝑆 − 𝐾 and 𝑦 = 𝐼. The linearized system of (2)
In this section, local stability of each feasible equilibrium
around 𝑃1 takes the form
of the model (1) is established; conditions are found under
which the Hopf bifurcation occurs and periodic solutions 𝑑𝑥
emerge as the delay crosses some critical value. Some numer- = − 𝑟𝑥 (𝑡) − 𝛽𝐾𝑦 (𝑡) ,
ical simulations to illustrate the theoretical results are given. 𝑑𝑡
(8)
Before going into any detail, we simplify the model since 𝑑𝑦
the first two equations of (1) are independent of the third one; = 𝛽𝐾𝑦 (𝑡 − 𝜏) − (𝜇1 + 𝛾) 𝑦 (𝑡) .
𝑑𝑡
it suffices to consider the first two equations. Thus, we restrict
our attention to the following reduced model: The characteristic equation associated with system (8) is
Proof. Indeed, if 𝜏 = 0, then (9) becomes Proof. Indeed, if 𝜏 = 0, then (14) becomes
Δ (𝜆, 0) = (𝜆 + 𝑟) [𝜆 + (𝜇1 + 𝛾) (1 − 𝑅0 )] = 0. (10) Δ (𝜆, 0) = 𝜆2 + (𝑎 + 𝑏) 𝜆 + (𝑐 + 𝑑) = 0, (16)
It is easy to see that (10) has two roots 𝜆 1 = −𝑟 < 0 and with
𝜆 2 = (𝜇1 + 𝛾)(𝑅0 − 1). So, if 𝑅0 < 1, the equilibrium 𝑃1
is asymptotically stable and if 𝑅0 > 1, the equilibrium 𝑃1 is 𝑟𝑆∗ 𝛽𝑆∗
𝑎+𝑏= + (𝜇1 + 𝛾) − 2
,
unstable. 𝐾 (1 + 𝛼2 𝐼∗ )
By Rouché’s Theorem [8, p. 248], it follows that if (17)
instability occurs for a particular value of the delay 𝜏, a 𝛽2 𝐼∗ 𝑆∗ 𝑟𝑆∗ 𝛽𝑆∗
characteristic root of (9) must intersect the imaginary axis. 𝑐+𝑑= 3
+ [(𝜇1 + 𝛾) − 2
].
(1 + 𝛼2 𝐼∗ ) 𝐾 (1 + 𝛼2 𝐼∗ )
Suppose that (9) has a purely imaginary root 𝑖𝜔, with 𝜔 > 0.
Then, by separating real and imaginary parts in (9), we have 2
Since (𝜇1 + 𝛾) − 𝛽𝑆∗ /(1 + 𝛼2 𝐼∗ ) = (𝜇1 + 𝛾)[𝛼2 𝐼∗ /(1 + 𝛼2 𝐼∗ )] >
𝛽𝐾 cos (𝜔𝜏) = 𝜇 + 𝛾, 0, it follows that 𝑎 + 𝑏 > 0 and 𝑐 + 𝑑 > 0. So, according to the
(11) Hurwitz criterion, all roots of (16) have negative real parts.
𝛽𝐾 sin (𝜔𝜏) = − 𝜔. Hence, 𝑃∗ is asymptotically stable.
Hence, In the following, we treat the case of positive delay 𝜏 > 0.
Let
𝜔2 = [𝛽𝐾 + 𝜇1 + 𝛾] (𝜇1 + 𝛾) (𝑅0 − 1) . (12)
2 2 (𝜇1 + 𝛾) ∗ 𝐾 (𝜇1 + 𝛾)
So, if 𝑅0 < 1, (9) has no purely imaginary root, and as 𝑃1 (𝐻1 ): 𝑆∗ + 𝑆 − > 0,
𝛽 𝛽
is asymptotically stable for 𝜏 = 0 it remains asymptotically (18)
stable for all 𝜏 ≥ 0. ∗2 2 (𝜇1 + 𝛾) ∗ 𝐾 (𝜇1 + 𝛾)
If 𝑅0 < 1, then the disease-free equilibrium 𝑃1 is unstable (𝐻2 ): 𝑆 + 𝑆 − < 0.
𝛽 𝛽
for 𝜏 = 0. By Kuang’s Theorem [4, p. 77], it follows that 𝑃1 is
unstable for all 𝜏 ≥ 0. This concludes the proof. We have the following two results.
Next, we establish the local stability of the endemic Theorem 4. If 𝑅0 > 1 and the assumption (𝐻1 ) is satisfied,
equilibrium 𝑃∗ and we determine the conditions under then the equilibrium point 𝑃∗ is asymptotically stable for all
which the Hopf bifurcation occurs. 𝜏 ≥ 0.
Let 𝑥 = 𝑆 − 𝑆∗ and 𝑦 = 𝐼 − 𝐼∗ . The linearized system of
(2) around 𝑃∗ = (𝑆∗ , 𝐼∗ ) takes the form Proof. Indeed, it follows that if instability occurs for a par-
ticular value of the delay 𝜏, a characteristic root of (14) must
𝑑𝑥 𝑟𝑆∗ 𝛽𝑆∗
= − 𝑥 (𝑡) − 2
𝑦 (𝑡) , intersect the imaginary axis. If (14) has a purely imaginary
𝑑𝑡 𝐾 (1 + 𝛼2 𝐼∗ ) root 𝑖𝜔, with 𝜔 > 0, then, by separating real and imaginary
parts in (14), we have
𝑑𝑦 𝛽𝐼∗ 𝛽𝑆∗ (13)
= 𝑥 (𝑡 − 𝜏) + 𝑦 (𝑡 − 𝜏)
𝑑𝑡 1 + 𝛼2 𝐼∗ (1 + 𝛼2 𝐼∗ )
2 𝑑 cos (𝜔𝜏) + 𝑏𝜔 sin (𝜔𝜏) = 𝜔2 − 𝑐,
(19)
− (𝜇1 + 𝛾) 𝑦 (𝑡) , 𝑏𝜔 cos (𝜔𝜏) − 𝑑 sin (𝜔𝜏) = −𝑎𝜔.
Therefore, if 𝑅0 > 1 and if the assumption (𝐻1 ) is satisfied, So the conditions of the Hopf bifurcation theorem are
then 𝑐2 − 𝑑2 > 0 and 𝑎2 − 𝑏2 − 2𝑐 > 0. Hence, according satisfied, which completes the proof.
to the criterion of Hurwitz, (20) has no positive root. Thus,
the characteristic equation (14) admits no purely imaginary We conclude this section by giving some numerical
root and, as 𝑃∗ is asymptotically stable for 𝜏 = 0, it is simulations to illustrate the theoretical results. We show
asymptotically stable for all 𝜏 ≥ 0. numerically that the system (2) has a family of periodic
solutions.
Theorem 5. If 𝑅0 > 1 and if the assumption (𝐻2 ) is satisfied,
then there exists 𝜏0 > 0 such that, for all 𝜏 ∈ [0, 𝜏0 ), the Proposition 6. Let 𝑟 = 0.1, 𝛽 = 0.1, 𝜎 = 0.96, 𝛾 = 0.5,
equilibrium point 𝑃∗ is asymptotically stable and, for all 𝜏 > 𝜏0 , 𝜇1 = 0.5, and 𝐾 = 70. Then, the system (2) has an equilibrium
the equilibrium point 𝑃∗ is unstable, and when 𝜏 = 𝜏0 , a Hopf point 𝑃∗ = (10.45, 0.875) which is asymptotically stable if 0 ≤
bifurcation of periodic solutions of system (2) occurs at 𝑃∗ . 𝜏 < 1.509 and unstable if 𝜏 > 1.509 and when 𝜏 = 1.509, the
With system has a family of periodic solutions (see Figure 1).
1 2 1 2 1/2
𝜔02 = (𝑏 + 2𝑐 − 𝑎2 ) + [(𝑏2 + 2𝑐 − 𝑎2 ) − 4 (𝑐2 − 𝑑2 )] ,
2 2 3. The Optimal Control Problem
1 (𝑑 − 𝑎𝑏) 𝜔02 − 𝑐𝑑
𝜏0 = arccos , Generally, the eradication of the disease may be too costly
𝜔0 𝑏2 𝜔02 + 𝑑2 when constant controls are considered as it requires treat-
(22) ment/vaccination at higher levels all the time. For eradication
to be achievable in a finite time, we need to consider time-
where 𝑎, 𝑏, 𝑐, and 𝑑 are defined in (14).
dependent controls. We use the optimal control strategies
Proof. The demonstration of stability of 𝑃∗ under the in the form of vaccination and treatment to decrease the
assumption (𝐻2 ) is similar to the previous demonstration. It number of both susceptible and infectious individuals and
suffices to remark that if 𝑅0 > 1 and if the assumption (𝐻2 ) is increase the total number of recovered individuals with
satisfied, then we deduce that minimum investment in disease control. This problem is
formulated as an optimal control problem by introducing
𝑟𝑆∗ 𝛽𝑆∗ 𝛽2 𝐼∗ 𝑆∗ two controls 𝑢1 and 𝑢2 , which represents the percentage of
𝑐+𝑑= [(𝜇1 + 𝛾) − 2
] + 3
> 0, susceptible and infected individuals being vaccinated and
𝐾 (1 + 𝛼2 𝐼∗ ) (1 + 𝛼2 𝐼∗ )
treated, respectively, per unit of time. Hence, (1) becomes
𝑟𝛽 2 2 (𝜇1 + 𝛾) ∗ 𝐾 (𝜇1 + 𝛾) 𝑑𝑆 𝑆 (𝑡) 𝛽𝑆 (𝑡) 𝐼 (𝑡)
𝑐−𝑑= ∗
[𝑆∗ + 𝑆 − ] = 𝑟 (1 − ) 𝑆 (𝑡) − − 𝑢1 (𝑡) 𝑆 (𝑡) ,
𝐾 (1 + 𝛼2 𝐼 ) 𝛽 𝛽 𝑑𝑡 𝐾 1 + 𝛼2 𝐼 (𝑡)
< 0. 𝑑𝐼 𝛽𝑆 (𝑡 − 𝜏) 𝐼 (𝑡 − 𝜏)
(23) = − (𝛼 + 𝜇 + 𝛾) 𝐼 (𝑡) − 𝑢2 (𝑡) 𝐼 (𝑡) ,
𝑑𝑡 1 + 𝛼2 𝐼 (𝑡 − 𝜏)
Thus, the characteristic equation (14) has only one purely 𝑑𝑅
imaginary root with positive imaginary part for the values of = 𝛾𝐼 (𝑡) − 𝜇𝑅 (𝑡) + 𝑢1 (𝑡) 𝑆 (𝑡) + 𝑢2 (𝑡) 𝐼 (𝑡) ,
𝑑𝑡
𝜏 given by (28)
1 (𝑑 − 𝑎𝑏) 𝜔02 − 𝑐𝑑 2𝑛𝜋 𝑆 (0) = 𝑆0 , 𝐼 (0) = 𝐼0 , 𝑅 (0) = 𝑅0 . (29)
𝜏𝑛 = arccos + . (24)
𝜔0 𝑏2 𝜔02 + 𝑑2 𝜔0
It is easy to show that there exists a unique solution
Since 𝑃∗ is asymptotically stable for 𝜏 = 0, by Kuang’s (𝑆(𝑡), 𝐼(𝑡), 𝑅(𝑡)) of system (28) with initial data (𝑆0 , 𝐼0 , 𝑅0 ) ∈
Theorem [4, p. 77], the equilibrium 𝑃∗ is asymptotically stable 3
(𝐶+ ) .
for 𝜏 ∈ [0, 𝜏0 ) and is unstable for all 𝜏 > 𝜏0 . In addition, for biological reasons, we assume that the
Lastly, to complete the proof, it remains to verify the initial data for system (28) satisfy
transversality condition
𝑆0 (𝑡) ≥ 0, 𝐼0 (𝑡) ≥ 0, 𝑅0 (𝑡) ≥ 0, 𝑡 ∈ [−𝜏, 0] . (30)
𝑑 Re(𝜆)
[ ] ≠ 0. (25)
𝑑𝜏 𝜏=𝜏0
The problem is to minimize the objective (cost) functional
given by
We have
𝐽 (𝑢1 , 𝑢2 )
𝑑 Re(𝜆)
Sign[ ] = Sign [𝑤4 − 𝑐2 + 𝑑2 ] . (26) 𝑡𝑓
𝑑𝜏 𝜏=𝜏0 1 1
= ∫ [𝐴 1 𝑆 (𝑡) + 𝐴 2 𝐼 (𝑡) + 𝐵1 𝑢12 (𝑡) + 𝐵1 𝑢22 (𝑡)] 𝑑𝑡.
Consequently, 0 2 2
(31)
𝑑 Re(𝜆)
[ ] > 0. (27) Subject to the differential equations (28), where the first two
𝑑𝜏 𝜏=𝜏0 terms in the functional objective represent benefit of 𝑆(𝑡) and
International Journal of Analysis 5
4.5 18 4.5
4 4
16
3.5 3.5
14
3
3
12 2.5
2.5
I(t)
I(t)
S(t)
10 2
2
1.5
1.5 8
1
1
6
0.5
0.5
0 4 0
5 10 15 20 0 200 400 600 800 1000 0 200 400 600 800 1000
S(t) t t
9 60 9
8 8
50
7 7
6 40 6
5 5
S(t)
I(t)
30
I(t)
4 4
3 20 3
2 2
10
1 1
0 0 0
0 10 20 30 40 50 60 0 50 100 150 200 250 300 0 50 100 150 200 250 300
S(t) t t
4.5 25 4.5
4 4
3.5 20 3.5
3
3
15
2.5
2.5
S(t)
I(t)
I(t)
2
2 10
1.5
1.5
5 1
1
0.5
0.5
0 0 0
0 5 10 15 20 25 0 100 200 300 400 500 0 100 200 300 400 500
S(t) t t
Figure 1: The equilibrium point 𝑃∗ = (10.45, 0.875) is asymptotically stable if 𝜏 = 0.2 and unstable if 𝜏 = 3 and when 𝜏 = 1.509, the system
has a family of periodic solutions.
6 International Journal of Analysis
𝐼(𝑡) populations that we wish to reduce, the parameters 𝐴 1 where 𝜆 𝑖 , 𝑖 = 1, 2, 3, are the adjoint functions to be deter-
and 𝐴 2 are positive constants to keep a balance in the size of mined suitably. Next, by applying Pontryagin’s maximum
𝑆(𝑡) and 𝐼(𝑡), respectively. We use in the second term in the principle with delay given in [10] to the Hamiltonian 𝐻, we
functional objective the quadratic term (1/2)𝐵𝑖 𝑢𝑖2 , 𝑖 = 1, 2, obtain the following theorem.
where 𝐵𝑖 is a positive weight parameter which is associated
with the control 𝑢𝑖 (𝑡) and the square of the control variable Theorem 8. Given optimal controls 𝑢1∗ (𝑡) and 𝑢2∗ (𝑡) and
reflects the severity of the side effects of the vaccination and solutions 𝑆∗ (𝑡), 𝐼∗ (𝑡), and 𝑅∗ (𝑡) of the corresponding state
treatment. systems (31) and (28), there exist adjoint variables 𝜆 1 , 𝜆 2 , and
Our target is to minimize the objective functional defined 𝜆 3 that satisfy
in (31) by decreasing the number of infected and susceptible
𝑑𝜆 1 (𝑡) 2𝑆∗
individuals and increasing the number of recovered individu- = − 𝐴 1 + 𝜆 1 (𝑡) (𝑟 (1 − ) − Λ 1 − 𝑢1∗ )
als, by using possible minimal control variables (𝑢1 (𝑡), 𝑢2 (𝑡)). 𝑑𝑡 𝐾
In other words, the control variables (𝑢1 (𝑡), 𝑢2 (𝑡)) ∈ 𝑈ad rep-
resent the percentage of susceptible and infected individuals − 𝜆 3 (𝑡) 𝑢1∗ (𝑡) − 𝜒[0,𝑡𝑓 −𝜏] 𝜆 2 (𝑡 + 𝜏) Λ 1 ,
being vaccinated and treated, respectively, per unit of time
𝑑𝜆 2 (𝑡)
and 𝑈ad is the control set defined by = − 𝐴 2 + 𝜆 1 (𝑡) Λ 2 + 𝜆 2 (𝑡) (𝜇 + 𝛼 + 𝛾 + 𝑢2∗ ) (36)
𝑑𝑡
𝑈ad = {𝑢 = (𝑢1 , 𝑢2 ) | 𝑢𝑖 (𝑡) measurable,
− 𝜆 3 (𝑡) (𝛾 + 𝑢2∗ ) − 𝜒[0,𝑡𝑓 −𝜏] 𝜆 2 (𝑡 + 𝜏) Λ 2 ,
(32)
0 ≤ 𝑢𝑖 (𝑡) ≤ 𝑢𝑖max < ∞, 𝑡 ∈ [0, 𝑡𝑓 ] , 𝑖 = 1, 2} . 𝑑𝜆 3 (𝑡)
= 𝜆 3 (𝑡) 𝜇,
𝑑𝑡
3.1. Existence of an Optimal Control. The existence of the
2
optimal control pair can be obtained using a result by Fleming where Λ 1 = (𝛽𝐼∗ (1 + 𝛼2 𝐼∗ )/(1 + 𝛼2 𝐼∗ ) ) and Λ 2 = (𝛽𝑆∗ /(1+
and Rishel in [11] and by Lukes in [12]. 𝛼2 𝐼∗ )2 )
Theorem 7. There exists a control function 𝑢1∗ (𝑡), 𝑢2∗ (𝑡) so that
with transversality conditions 𝜆 𝑖 (𝑡𝑓 ) = 0, 𝑖 = 1, 2, 3.
𝐽 (𝑢1∗ (𝑡) , 𝑢2∗ (𝑡)) = min 𝐽 (𝑢1 (𝑡) , 𝑢2 (𝑡)) . (33) (37)
(𝑢1 ,𝑢2 )∈𝑈𝑎𝑑
Proof. To prove the existence of an optimal control pair, it is Furthermore, the optimal control pair 𝑢∗ (𝑡) is given by
easy to verify the following.
(𝜆 1 (𝑡) − 𝜆 3 (𝑡)) 𝑆∗ (𝑡) max
(1) The set of controls and corresponding state variables 𝑢1∗ (𝑡) = max (min ( , 𝑢1 ) , 0) ,
is nonempty. 𝐵1
(2) The admissible set 𝑈ad is convex and closed. (𝜆 2 (𝑡) − 𝜆 3 (𝑡)) 𝐼∗ (𝑡) max
𝑢2∗ (𝑡) = max (min ( , 𝑢2 ) , 0) .
(3) The right-hand side of the state system (28) is 𝐵2
bounded by a linear function in the state and control (38)
variables.
(4) The integrand of the objective functional is convex on Proof. Using Pontryagin’s maximum principle with delay in
𝑈ad . state, we obtain the adjoint equations and transversality
(5) There exists constants 𝜔1 , 𝜔2 > 0, and 𝜌 > 1 such conditions such that
that the integrand 𝐿(𝑆, 𝐼, 𝑢1 , 𝑢2 ) of the objective func- 𝑑𝜆 1 (𝑡) 𝜕𝐻 𝜕𝐻
tional satisfies 𝐿(𝑆, 𝐼, 𝑢1 , 𝑢2 ) ≥ 𝜔2 + 𝜔1 (|𝑢1 |2 + =− − 𝜒[0,𝑡𝑓 −𝜏] (𝑡 + 𝜏) , 𝜆 1 (𝑡𝑓 ) = 0,
𝑑𝑡 𝜕𝑆 𝜕𝑆𝜏
|𝑢2 |2 )𝜌/2 .
𝑑𝜆 2 (𝑡) 𝜕𝐻 𝜕𝐻
The result follows directly from [11]. =− − 𝜒[0,𝑡𝑓 −𝜏] (𝑡 + 𝜏) , 𝜆 2 (𝑡𝑓 ) = 0, (39)
𝑑𝑡 𝜕𝐼 𝜕𝐼𝜏
3.2. Characterization of the Optimal Control. Before char- 𝑑𝜆 3 (𝑡) 𝜕𝐻 𝜕𝐻
acterizing the optimal control pair, we first define the =− − 𝜒[0,𝑡𝑓 −𝜏] (𝑡 + 𝜏) , 𝜆 3 (𝑡𝑓 ) = 0,
𝑑𝑡 𝜕𝑅 𝜕𝑅𝜏
Lagrangian for the optimal control problem (28)–(31) by
1 1 and by using the optimality conditions we find
𝐿 (𝑆, 𝐼, 𝑢1 , 𝑢2 ) = 𝐴 1 𝑆 (𝑡) + 𝐴 2 𝐼 (𝑡) + 𝐵1 𝑢12 (𝑡) + 𝐵2 𝑢22 (𝑡)
2 2
𝜕𝐻
(34) = 𝐵1 𝑢∗ (𝑡) − 𝜆 1 (𝑡) 𝑆∗ + 𝜆 3 (𝑡) 𝑆∗ = 0, at 𝑢1 = 𝑢1∗ (𝑡) ,
𝜕𝑢1
and the Hamiltonian H for the control problem by
𝜕𝐻
𝑖=3 = 𝐵2 𝑢∗ (𝑡) − 𝜆 2 (𝑡) 𝐼∗ + 𝜆 3 (𝑡) 𝑆∗ = 0, at 𝑢2 = 𝑢2∗ (𝑡) ,
𝐻 (𝑆, 𝐼, 𝑅, 𝑢1 , 𝑢2 , 𝜆 𝑖 , 𝑡) = 𝐿 (𝑆, 𝐼, 𝑢1 , 𝑢2 ) + ∑𝜆 𝑖 𝑓𝑖 , (35) 𝜕𝑢2
𝑖=1 (40)
International Journal of Analysis 7
Step 1
for 𝑖 = −𝑚, . . . , 0, do
𝑆𝑖 = 𝑆0 , 𝐼𝑖 = 𝐼0 , 𝑅𝑖 = 𝑅0 , 𝑢1𝑖 = 0 and 𝑢2𝑖 = 0,
end for
for 𝑖 = 𝑛, . . . , 𝑛 + 𝑚, do
𝜆𝑖1 = 0, 𝜆𝑖2 = 0, 𝜆𝑖3 = 0,
Step 2
𝑆𝑖 𝛽𝑆𝑖 𝐼𝑖
𝑆𝑖+1 = 𝑆𝑖 + ℎ [𝑟 (1 − )𝑆 − − 𝑢1𝑖 𝑆𝑖 ],
𝐾 𝑖 1 + 𝛼2 𝐼𝑖
𝛽𝑆𝑖−𝑚 𝐼𝑖−𝑚
𝐼𝑖+1 = 𝐼𝑖 + ℎ [ ] − (𝛼 + 𝜇 + 𝛾 − 𝑢2𝑖 ) 𝐼𝑖 ,
1 + 𝛼2 𝐼𝑖−𝑚
𝑅𝑖+1 = 𝑅𝑖 + ℎ [𝛾𝐼𝑖 − 𝜇𝑅𝑖 + 𝑢1𝑖 𝑆𝑖 + 𝑢2𝑖 𝐼𝑖 ],
𝛽𝐼 (1 + 𝛼2 𝐼𝑖 )
Λ𝑖1 = [ 𝑖 ],
(1 + 𝛼2 𝐼𝑖 )2
𝛽𝑆𝑖
Λ𝑖2 = [ 2
],
(1 + 𝛼2 𝐼𝑖 )
2𝑆𝑖
𝜆𝑛−𝑖−1
1 = 𝜆𝑛−𝑖 𝑛−𝑖
1 + ℎ [−𝐴 1 + 𝜆 1 (𝑟 (1 − ) − Λ𝑖1 − 𝑢1𝑖 ) − 𝜆𝑛−𝑖 𝑖 𝑛−𝑖+𝑚 𝑖+1
3 𝑢1 − 𝜒[0,𝑡𝑓 −𝜏] (𝑡𝑛−𝑖 ) 𝜆 2 Λ 1 ],
𝐾
𝜆𝑛−𝑖−1
2 = 𝜆𝑛−𝑖 𝑛−𝑖 𝑖+1 𝑛−𝑖 𝑖 𝑛−𝑖 𝑖 𝑛−𝑖+𝑚 𝑖+1
2 + ℎ [−𝐴 2 + 𝜆 1 Λ 2 + 𝜆 2 (𝛼 + 𝜇 + 𝛾 + 𝑢2 ) − 𝜆 3 (𝛾 + 𝑢1 ) − 𝜒[0,𝑡𝑓 −𝜏] (𝑡𝑛−𝑖 ) 𝜆 2 Λ 2 ],
𝜆𝑛−𝑖−1
3 = 𝜆𝑛−𝑖 𝑛−𝑖
3 + ℎ [𝜇𝜆 3 ],
(𝜆𝑛−𝑖 𝑛−𝑖
1 − 𝜆 3 )𝑆𝑖+1
Θ𝑖+1
1 =
𝐵1
(𝜆𝑛−𝑖 𝑛−𝑖
2 − 𝜆 3 )𝐼𝑖+1
Θ𝑖+1
2 =
𝐵2
𝑢1𝑖 = max (min (Θ𝑖+1 max
1 , 𝑢1 ) , 0)
Step 3
for 𝑖 = 1, . . . , 𝑛, do
write
𝑆∗ (𝑡𝑖 ) = 𝑆𝑖 , 𝐼∗ (𝑡𝑖 ) = 𝐼𝑖 , 𝑅∗ (𝑡𝑖 ) = 𝑅𝑖 , 𝑢1∗ (𝑡𝑖 ) = 𝑢1𝑖 and 𝑢2∗ (𝑡𝑖 ) = 𝑢2𝑖 ,
end for
Algorithm 1
Using the property of the control space, we obtain (𝜆 1 (𝑡) − 𝜆 3 (𝑡)) 𝐼∗ (𝑡)
if 0 < < 𝑢2max ,
𝐵2
(𝜆 1 (𝑡) − 𝜆 3 (𝑡)) 𝑆∗ (𝑡) (𝜆 1 (𝑡) − 𝜆 3 (𝑡)) 𝐼∗ (𝑡)
𝑢1∗ (𝑡) = 0, if ≤ 0, 𝑢2∗ (𝑡) = 𝑢2max , if ≥ 𝑢2max .
𝐵1 𝐵2
(𝜆 1 (𝑡) − 𝜆 3 (𝑡)) 𝑆∗ (𝑡) (42)
𝑢1∗ (𝑡) = ,
𝐵1 So, the optimal control pair is characterized as (38).
∗
(𝜆 1 (𝑡) − 𝜆 3 (𝑡)) 𝑆 (𝑡) The optimal control pair and the state are found by solving
if 0 < < 𝑢1max ,
𝐵1 the following optimality system, which consists of the state
8 International Journal of Analysis
system (28), the adjoint system (36), boundary conditions Table 1: Values of the parameters.
(29) and (37), and the characterization of the optimal control
pair (𝑢1∗ , 𝑢2∗ ) (38): Parameters Descriptions Values
𝑆0 Initial susceptible population 120
∗ ∗ ∗ ∗ 𝐼0 Initial infected population 50
𝑑𝑆 (𝑡) 𝑆 𝛽𝑆 𝐼
= 𝑟 (1 − ) 𝑆∗ − 𝑅0 Initial recovered population 100
𝑑𝑡 𝐾 1 + 𝛼2 𝐼∗
𝜇 Natural death of the population 0.01
(𝜆 1 (𝑡) − 𝜆 3 (𝑡)) 𝑆∗ max 𝛼 0.01
− max (min ( , 𝑢1 ) , 0) 𝑆∗ , Death rate due to disease
𝐵1 Parameter that measures the
𝛼2 0.001
∗ ∗
inhibitory effect
∗
𝑑𝐼 (𝑡) 𝛽𝑆 (𝑡 − 𝜏) 𝐼 (𝑡 − 𝜏) 𝛽
= − (𝛼 + 𝜇 + 𝛾) 𝐼∗ Transmission rate 0.0001
𝑑𝑡 1 + 𝛼2 𝐼∗ (𝑡 − 𝜏) 𝛾 Recovery rate 0.0004
(𝜆 2 (𝑡) − 𝜆 3 (𝑡)) 𝐼∗ max 𝑟 Intrinsic birth rate 0.5
− max (min ( , 𝑢2 ) , 0) 𝐼∗ , 𝐾 Carrying capacity 300
𝐵2
𝐴1 Weight parameter 1000
𝑑𝑅∗ (𝑡) 𝐴2
= 𝛾𝐼∗ − 𝜇𝑅∗ Weight parameter 1000
𝑑𝑡 𝐵1 Weight parameter 5000
(𝜆 1 (𝑡) − 𝜆 3 (𝑡)) 𝑆∗ max 𝐵2 Weight parameter 10
+ max (min ( , 𝑢1 ) , 0) 𝑆∗ 𝜏
𝐵1 Time incubation 1
300 150
200 100
S(t)
I(t)
100 50
0 0
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
Time (days) Time (days)
Control u1
R(t)
200 0.5
100
0 0
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
Time (days) Time (days)
R without control
R with control
1
Control u2
0.5
0
0 5 10 15 20 25 30 35 40 45 50
Time (days)
Figure 2: Evolution of different classes of individuals with and without control for time delay 𝜏 = 1.
and we show that the optimal strategy becomes more effective [5] A. Abta, A. Kaddar, and H. T. Alaoui, “Global stability for delay
when we combined the vaccination and treatment strategies SIR and SEIR epidemic models with saturated incidence rates,”
together. Electronic Journal of Differential Equations, vol. 2012, no. 23, pp.
1–13, 2012.
Conflict of Interests [6] F. Brauer, “Models for the spread of universally fatal diseases,”
Journal of Mathematical Biology, vol. 28, no. 4, pp. 451–462,
The authors declare that there is no conflict of interests 1990.
regarding the publication of this paper. [7] H. W. Hethcote, “A thousand and one epidemic models,” in
Frontiers in Mathematical Biology, S. A. Levin, Ed., vol. 100 of
Lecture Notes in Biology, pp. 504–515, Springer, 1995.
References
[8] V. Capasso and G. Serio, “A generalization of the Kermack-
[1] N. T. J. Bailey, The Mathematical Theory of Infectious Diseases McKendrick deterministic epidemic model,” Mathematical Bio-
and Its Applications, Hafner Press, New York, NY, USA, 2nd sciences, vol. 42, no. 1-2, pp. 43–61, 1978.
edition, 1975.
[9] A. Korobeinikov, “Global properties of infectious disease mod-
[2] J. Zhang, Z. Jin, J. Yan, and G. Sun, “Stability and Hopf bifur-
els with nonlinear incidence,” Bulletin of Mathematical Biology,
cation in a delayed competition system,” Nonlinear Analysis:
vol. 69, no. 6, pp. 1871–1886, 2007.
Theory, Methods & Applications, vol. 70, no. 2, pp. 658–670,
2009. [10] L. Göllmann, D. Kern, and H. Maurer, “Optimal control
[3] C. C. McCluskey, “Complete global stability for an SIR epidemic problems with delays in state and control variables subject to
model with delay-distributed or discrete,” Nonlinear Analysis: mixed control-state constraints,” Optimal Control Applications
Real World Applications, vol. 11, no. 1, pp. 55–59, 2010. & Methods, vol. 30, no. 4, pp. 341–365, 2009.
[4] R. Xu, “Global stability of an HIV-1 infection model with satura- [11] W. H. Fleming and R. W. Rishel, Deterministic and Stochastic
tion infection and intracellular delay,” Journal of Mathematical Optimal Control, Applications of Mathematics, no. 1, Springer,
Analysis and Applications, vol. 375, no. 1, pp. 75–81, 2011. New York, NY, USA, 1975.
10 International Journal of Analysis
International
Journal of Journal of
Mathematics and
Mathematical
Discrete Mathematics
Sciences