1 Covid-19

medRxiv preprint doi: https://doi.org/10.1101/2020.07.23.20160473; this version posted July 24, 2020.
The copyright holder for this preprint

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
It is made available under a CC-BY-ND 4.0 International license .
Mathematical Modeling and Optimal Control Analysis of

COVID-19 in Ethiopia
Haileyesus Tessema Alemneh 1 and Getachew Teshome Tilahun 2

1
Department of Mathematics, University of Gondar, Gondar, Ethiopia
Email: [email protected]
2
Department of Mathematics, Haramaya University, Dire Dawa, Ethiopia
Email: [email protected]
Abstract
In this paper we developed a deterministic mathematical model of the pandemic COVID-19 trans-
mission in Ethiopia, which allows transmission by exposed humans. We proposed an SEIR model
using system of ordinary differential equations. First the major qualitative analysis, like the disease
free equilibruim point, endemic equilibruim point, basic reproduction number, stability analysis of
equilibrium points and sensitivity analysis was rigorously analysed. Second, we introduced time
dependent controls to the basic model and extended to an optimal control model of the disease.
We then analysed using Pontryagin’s Maximum Principle to derive necessary conditions for the
optimal control of the pandemic. The numerical simulation indicated that, an integrated strategy
effective in controling the epidemic and the gvernment must apply all control strategies in combat-
ing COVID-19 at short period of time.
Keywords: COVID-19; Optimal control; Modeling; Ethiopia; Numerical Simulation.
1 Introduction
Corona virus disease (COVID-19) which is caused by novel corona virus is a respiratory illness
that can spread from person to person [1, 2, 3]. People with COVID-19 have had a wide range of
symptoms reported – ranging from mild symptoms to severe illness. Symptoms may appear 2-14
days after exposure to the virus. People with these symptoms may have COVID-19: fever or chills,
cough, shortness of breath or difficulty breathing, fatigue, muscle or body aches, headache, new
loss of taste or smell, sore throat, congestion or runny nose, nausea or vomiting, diarrhea [1, 4].
NOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice.
medRxiv preprint doi: https://doi.org/10.1101/2020.07.23.20160473; this version posted July 24, 2020. The copyright holder for this preprint
The outbreak was first identified in Wuhan, China, in December, 2019, which has been spreading
worldwide [5, 6] . Following the day of the outbreak of the pandemic to present more than seven
million globally confirmed cases and half million deaths as well as four million recovered numbers
[7]. In Africa also the virus spread to 57 countries [1, 7]. Through out Africa up to Jun 14, 2020
confirmed cases are 235,707, number of deaths are 6,283 and recoveries are 36,850 [8].
In order to combat this pandemic, different preventive measures are recommended, such as
avoiding close contact with sick people, avoiding touching the eyes, nose and mouth with unwashed
hands, washing hands often with soap and water for at least 20 seconds, using an alcohol-based
hand sanitizer containing at least 60% alcohol when soap and water are not available [1, 3, 4].
Limit the number of people you have close contact with, or are visited by, to a few at a time. For
infected individuals, staying home, covering cough or sneeze with a tissue, then throw the tissue
in the trash, and clean and disinfect frequently touched objects are recommended for controlling
COVID-19, perform home quarantine for 14 days after the last contact with a patient with confirmed
COVID-19 [9, 10].
Mathematical models with optimal control analysis has become an important tool in order to
understand the dynamics of disease transmission and decision making processes regarding inter-
vention programs for the disease control. COVID-19 have been modeled by very few researchs
after the outbreak of the disease. The study by [11], used an SIR model to predict the magnitude
of the COVID-19 epidemic in Pakistan and compared the numbers with the reported cases on the
national database. They predicted that 90% of the population will have become infected with the
virus if policy interventions seeking to curb this infection are not adopted aggressively. The study
presented in [12] used SEIR compartments by considering limited parameters, from January 22,
2020 to March 3, 2020 and Prediction SEIR forecasted by using SEIR model They also looked
at the feelings, current disease trends and economic and political impacts. The article [13], pro-
posed conceptual models for the COVID-19 outbreak in Wuhan with the consideration of individual
behavioural reaction and governmental actions, e.g., holiday extension, travel restriction, hospital-
isation and quarantine. The article published by [12] also proposed SEIR model by incorporating
the intrinsic impact of hidden exposed and infectious cases on the entire procedure of epidemic,
which is difficult for traditional statistics analysis. The other study is done in China using SIR
by considering logistic growth [11]. Using their model, the study approximated future number of
cases in china and proposed possible controlling mechanisms. The study [14] simulated the on-
going trajectory of Covid 19 outbreak in Wuhan using an age-structured SEIR model for several
2
physical distancing measures. The article presented by [15], proposed a compartmental model by
dividing the quarantined individuals in to two sub-groups. They developed an SEIRU model where
R and U are quarantine-infected individuals expected to recover and meet undetectable criteria. A
more detail model is proposed by [16], and studied a mathematical modeling of the spread of the
COVID-19 taking into account the undetected infections, in China. The study in [17], proposed a
new epidemic model in the case of Italy that discriminates between infected individuals depending
on whether they have been diagnosed and on the severity of their symptoms. However, all of the
above models studied the dynamics of disease transmission, they did not study finding an optimal
control strategy using the maximum princeiple of Pontryagn.
From all the above studies we can understand that, COVID-19 have been modeled by consider-
ing different situations. In the case of Ethiopia, the proposed model can’t reflect the real situation of
the country. Currently Ethiopia is implementing different measure to control COVID-19 including
implementing state of emergency, counry border closing and mandatory quarantine of new arrivals
for 14 days. herefore, in the present paper we extend the SEIR model with the aforementioned
control measures and propose an optimal control strategy.
The paper is organized as follows. In Section 2 was devoted to the model description and
the underlying assumptions. In Section 3 we carry out mathematical analysis of this COVID-19
model. In Section 4 we propose an optimal control problem and and the optimal control analysis
are presented. In Section 5 numerical simulation and calibration of the model was implemented for
the various strategies considered in this work. The conclusion was presented in Section 6.
2 Model Description and Formulation
In this model the entire population is divided into four sub-populations: Susceptible individuals
(denoted by S) are those who are not infected by the disease pathogen but there is a possibility to
be infected. Exposed individuals (denoted by E) are individuals who are in the incubation period
after being infected and have no visible clinical signsand these individuals could infect other people
with a higher probability than people in the infected compartments. After the incubation period, the
person passes to the infected compartment; infected individuals (denoted by I) are individuals who
developed the symptom of the disease. Recovered individuals (denoted by R )are those individuals
who recovered from the disease.
Individuals are recruitment at a rate π is either through immigration or birth. The suscepti-
ble individuals got infection of COVD-19 disease at a contact rate of β either from infected with
3
probability of σ1 or from exposed individuals with probability of σ2 and move to the exposed com-
partment. The exposed individuals become infectious and join the infected compartment at τ δ and
the remaining proportion of this exposed individuals develop natural immunity and recovered from
the disease. By the treatment given, the infected individuals will recover and move to the recovered
compartment at a rate of , or may die due to the disease at a rate of ρ . The recovered individuals
become again susceptible to the disease at a rate of η. The whole population have an average death
rate of µ . Table 1 shows the description of model parameters. The flow diagram of the model is
shown in Figure 1 below.
With regards to the above asumptions, the model is governed by the following system of differ-
ential equation:
dS
dt
= π + ηR − β(σ1 I + σ2 E)S − µS
dE
dt
= β(σ1 I + σ2 E)S − (δ + µ) E
(1)
dI
dt
= τ δ E − (ε + ρ + µ) I
dR
dt
= (1 − τ )δE + ε I − (µ + η) R
With the initial condition
S(0) = S0 ≥ 0 , E(0) = E0 ≥ 0 I(0) = I0 ≥ 0 , R(0) = R0 ≥ 0 (2)
Figure 1: Compartmental flow diagram of the pandemic COVID 19 transmission
4
Table 1: Description of parameters of the model (1).
Parameter Description
π Ricuirement rate of individuals
β Contact rate of susceptible individuals
ρ Death rate due to disease
δ Proportion of exposed individuals leaving the compartment.
ε Recovery rate of individuals from the disease
τ Proportion of exposed individuals who join infected compartment
µ Natural death rate
η Proportion of recovered individuals to be susceptible.
3 Model Analysis
3.1 Invariant Region
Let us determine a region in which the solution of model (1) is bounded. For this model the total
population is N (S, E, I, R) = S(t) + E(t) + I(t) + R(t). Then, differentiating N with respect to
time we obtain:
dN dS dE dI dR
= + + + = π − ρI − µN
dt dt dt dt dt
If there is no death due to the disease, we get
dN
≤ π − µN (3)
dt
After solving equation (3) and evaluating it as t −→ ∞, we got
π
Ω = {(S, E, I, R)ε<4+ : N (t) ≤ }
µ
Which is the feasible solution set for the model (1) and all the solution set is bounded in it.
3.2 Positivity of Solutions
Theorem 3.1. If S(0) > 0, E(0) > 0, I(0) > 0, R(0) > 0 are positive in the feasible set Ω, then
the solution set (S(t), E(t), I(t), R(t)) of system (1) is positive for all t ≥ 0.
Proof. : From the first equation of the system

dS
= π + ηR − β(σ1 I + σ2 E)S − µS
dt
5
which can be taken as

dS
≥ −µS (4)
dt
After evaluating equation (4), we obtain
S ≥ S(0)e−µt
Similarly, we obtain
E ≥ E(0)e−(δ+µ)t
I ≥ I(0)e−(ε+ρ+µ)t
R ≥ R(0)e−(η+µ)t
Therefore, all the solution sets are positive for t ≥ 0.
3.3 Disease Free Equilibrium Point(DFEP)
When there is no infectious person of the disease in the population, I.e E = I = 0, the disease free
equilibrium occur and is obtained by taking the right side of Eq. (1) equal to zero. Therefore the
disease free equilibrium point is given by:
π
E0 = ( , 0, 0, 0) (5)
µ
3.4 Basic reproduction number
We calculate the basic reproduction number <0 of the system by applying the next generation
matrix method as laid out by [18]. The first step is rewrite the model equations, starting with
newly infective classes:
dE
dt
= β(σ1 I + σ2 E)S − (δ + µ) E
dI (6)
dt
= τ δ E − (ε + ρ + µ) I
dR
dt
= (1 − τ )δE + ε I − (µ + η) R
Then by the principle of next-generation matrix, the Jacobian matrices at DFE is given by
   
β σ2 Π β σ1 Π
0 δ + µ 0 0
 µ µ
  
   
F = 0 0 0 and V =  −τ δ
  ε+ρ+µ 0  
   
0 0 0 − (1 − τ ) δ −ε µ+η
6
 
(σ1 δ τ +σ2 (+µ+ρ))β Π β σ1 Π
µ (+ρ+µ)(δ+µ) µ (+ρ+µ)
0
 
FV −1
 
=
 0 0 0
 
0 0 0
Therefore, the basic reproduction number is given us
β Π (σ1 τ δ + σ2 ( + ρ + µ))
<0 = (7)
µ ( + ρ + µ) (δ + µ)
<0 is a threshold parameter that represents the average number of infection caused by one infectious
individual when introduced in the susceptible population [18].
3.5 Local Stability of DFEP
Theorem 3.2. The DFEP point is locally asymptotically stable if <0 < 1 and unstable if <0 > 1.
Proof. The Jacobian matrix, evaluated at the disease-free equilibrium E0 , we get:

 
−µ − β σµ2 Π − β σµ1 Π η
 
 
 0 β σ2 Π − δ − µ β σ1 Π
0 
 µ µ 
J= 


 0
 τδ − − µ − ρ 0  
 
0 (1 − τ ) δ −µ − η
The characteristic polynomial from the Jacobian matrix is
(−λ − (η + µ))(−λ − µ)(λ2 + ψ1 λ + ψ2 ) = 0 (8)
Where
−β σ2 Π + δ µ + µ + 2 µ2 + µ ρ
ψ1 =
µ
−Π β δ τ σ1 − Π β σ2 − Π β µ σ2 − Π β ρ σ2 + δ µ + δ µ2 + δ µ ρ + µ2 + µ3 + µ2 ρ
ψ2 =
µ
From the Eq.(8), we see that
−λ − (η + µ) ⇒ λ1 = −(η + µ) < 0 and − λ − µ ⇒ λ2 = −µ < 0
From the last expression, that is

λ2 + ψ1 λ + ψ2 = 0 (9)
We applied Routh-Hurwitz criteria and by the principle Eq.(8) has strictly negative real root iff
ψ1 > 0 , ψ2 > 0 and ψ1 ψ2 > 0. Clearly we see that ψ1 > 0 because it is the sum of positive
7
parameters.
(ε + ρ + µ) β Πσ1 τ δ
ψ1 = + + 1 − <0 > 0
(δ + µ) µ (ε + ρ + µ) (δ + µ)
ψ2 = (ε + ρ + µ) (δ + µ) [1 − <0 ] > 0
Hence the DFEP is locally asymptotically stable if <0 < 1.
3.6 Global Stability of DFEP
Theorem 3.3. The DEFP E0 of the model (1) is globally asymptotically stable if <0 < 1.
Proof. Consider the following Lyapunov function
V = c1 E + c2 I (10)
Differentiating equation (10) with respect to t gives

dV dE dI
= c1 + c2 (11)
dt dt dt
dE dI
Substituting dt
and dt
from the model (1), we get:
dV
= c1 [β(σ1 I + σ2 E)S − (δ + µ) E] + c2 [τ δ E − (ε + ρ + µ) I]
dt
= c1 βσ1 SI − c2 (ε + ρ + µ) I + c1 (βσ2 SE − (δ + µ)E) + c2 τ δE
Here take c2 = − βσ2 S−(δ+µ)

τδ
c1 , then we have
dV βσ1 Sτ δ + βσ2 S (ε + ρ + µ) − (δ + µ) (ε + ρ + µ)
=[ ]c1 I
dt τδ
Taking c2 = 1, and substituting <0 we get
dV (δ + µ) (ε + ρ + µ)
≤ (<0 − 1)I
dt τδ
π dV dV
for S ≤ S 0 = µ
and dt
≤ 0 for <0 < 1 and dt
= 0 if and only if I = 0. This implies that the only
dV
trajectory of the system (1) on which dt
≤ 0 is E 0 . Therefore by Lasalle’s invariance principle,
E 0 is globally asymptotically stable in Ω.
3.7 Existence of endemic equilibrium (EEP)
In the presence of disease in the population,(S(t) ≥ 0; E(t) ≥ 0; I(t) ≥ 0, R(t) ≥ 0), there exist
an equilibrium point called endemic equilibrium point denoted by E ∗ = (S ∗ , E ∗ , I ∗ , R∗ ) 6= 0.
Lemma 3.4. The Zika-only model has a unique endemic equilibrium if and only if <0 > 1.
8
Proof. It can be obtained by equating each equation of the model equal to zero. i.e
dS dE dI dR
= = = =0
dt dt dt dt
Then we obtain 
(δ+µ)(ε+ρ+µ)
S∗ =




 β(τ δ+ε+ρ+µ)

µ(δ+µ)(ε+ρ+µ)2 (η+µ)[1−<0 ]

E ∗

= βκ
(12)
τ δµ(δ+µ)(ε+ρ+µ)(η+µ)[1−<0 ]
I∗ =




 βκ

δ[τ (+ρ+µ)+η+µ][1−<0 ]

R ∗

= βκ
Where
κ = δ 2 ητ 2 − (ε + ρ + µ)[δ 2 τ (ητ + µ) + δτ (ητ + µ2 )]
−(ε + ρ + µ)2 µ(δ + η + µ) − δηµτ (2µ + 3ρ) − δητ ρ2
Theorem 3.5. The EEP is locally asymptotically stable if <0 > 1 and unstable if otherwise.
Proof. The Jacobian matrix of system (1) at the DFEP is given by

 
β σ2 Π β σ1 Π
−µ − µ − µ η
 
 
 0 β σ2 Π − δ − µ β σ1 Π
0 
 µ µ 
J= 

 (13)
 0
 τδ −ε − µ − ρ 0  
 
0 (1 − τ ) δ −µ − η
To determine the local stablity of endemic equilibrium, we used the center manifold theory [19],
by taking β as a bufiracation parameter. let S = z1 , E = z2 , I = z3 and R = z4 . In addition, using
dz
vector notation z = (z1 , z2 , z3 , z4 )T , formulated as dt
= F (z), with F = (f1 , f2 , f3 , f4 , f5 )T and
Π
the disease free equilibrium is given by (z1 = µ
, z2 = 0, z3 = 0, z4 = 0). Then the value of β as a
bifurcation parameter and solve <0 = 1, which leads to
µ (ε + ρ + µ) (δ + µ)
β = β∗ =
Π (σ1 τ δ + σ2 (ε + ρ + µ))
The right eigenvector, w = (w1 , w2 , w3 , w4 )T , associated with this simple zero eigenvalue can be
9
obtained as:
−(δ + µ)(η + µ)(ε + ρ + µ) + δ(ε + ρ + µ) − δτ (ρ + µ)
w1 = w2 ,
µ(η + µ)(ε + ρ + µ)
w2 = w2 > 0,
δτ
w3 = w2 ,
ε+ρ+µ
δ(ε + ρ + µ) − δτ (ρ + µ)
w4 = w2
(η + µ)(ε + ρ + µ)
The left eigenvector, v = (v1 , v2 , v3 , v4 , v5 ) , associated with this simple zero eigenvalue is given
by
σ1 Π (δ + µ)
v1 = v4 = 0, v2 = v2 > 0, v3 = v2
Π (σ1 τ δ + σ2 (ε + ρ + µ))
Since the first and fourth component of v are zero, we don’t need the derivatives of f1 and f4 . From
the derivatives of f2 and f3 , the only ones that are nonzero are the following:
∂ 2 f2 ∂ 2 f2 ∂ 2 f3 ∂ 2 f3 ∂ 2 f2 ∂ 2 f2
= = β ∗ σ2 , = = β ∗ σ1 , = σ2 z1 , = σ1 z1
∂z2 ∂z1 ∂z1 ∂z2 ∂z3 ∂z1 ∂z1 ∂z3 ∂z2 ∂β ∂z3 ∂β
and all the other partial derivatives of are zero. The direction of the bifurcation at <0d = 1 is
determined by the signs of the bifurcation coefficients a and b, obtained from the above partial
derivatives, given respectively by
∂ 2 f2 ∂ 2 f2
a = 2v2 w1 w2 + 2v2 w1 w3
∂z1 ∂z2 ∂z1 ∂z3
= 2v2 w1 β ∗ [σ2 w2 + σ1 w3 ]
2 (δ + µ)
=− [(δ + µ)(η + µ)(ε + ρ + µ) − δ(ε + ρ + µ) + δτ (ρ + µ)] w2 < 0
Π(η + µ)(ε + ρ + µ)
and
∂ 2 f2 ∂ 2 f2

Π δτ
b = v2 w2 + v2 w3 = σ2 + σ1 > 0 (14)
∂z1 ∂β ∗ ∂z3 ∂β ∗ µ ε+ρ+µ
Do to the sign of the coefficient a < 0 and b > 0 at β ∗ , by the theorem 4.1 stated in [19] sys-
tem (11) undergo forward bifurcation at <0 = 1 and the unique endemic equilibruim is locally
assymptotically stable for <0 > 1.
3.8 Sensitivity Analysis
We carried out sensitivity analysis, on the basic parameters, to identify their effect to the transmi-
tion of the disease. To go through sensitivity analysis, we used the normalized sensitivity index
definition as defined in [20]. The Normalized forward sensitivity index of a variable, <0 , that de-
∂<0 p
pends differentiably on a parameter, p, is defined as: Λ<
p
0
= ∂p
× <0
for p represents all the
10
β Π(σ1 τ δ+σ2 (+ρ+µ))

basic parameters. Here we have <0 = µ (+ρ+µ)(δ+µ)
. For the sensitivity index of <0 to the
parameters:
∂<0 β
Λ<
β =
0
× =1>0
∂β <0
∂<0 σ1 σ1 δτ ρ
Λ<0
σ1 = × = >0
∂σ1 <0 σ1 δτ + σ2 (ε + ρ + µ)
∂<0 σ2 σ2 (ε + ρ + µ)
Λ<
σ2 =
0
× = >0
∂σ2 <0 σ1 δτ + σ2 (ε + ρ + µ)
∂<0 τ σ1 τ δ
Λ<
τ =
0
× = >0
∂τ <0 σ1 δτ + σ2 (ε + ρ + µ)
∂<0 δ δ[σ1 δτ µ − σ2 (ε + ρ + µ)]
Λ<
δ =
0
× = >0
∂δ <0 (ε + ρ + µ)[σ1 δτ + σ2 (ε + ρ + µ)]
∂<0 ρ σ1 δτ ρ
Λ<
ρ =
0
× =− <0
∂ρ <0 (ε + ρ + µ)[σ1 δτ + σ2 (ε + ρ + µ)]
∂<0 ε σ1 τ δε
Λ<
ε =
0
× =− <0
∂ε <0 (ε + ρ + µ)[σ1 δτ + σ2 (ε + ρ + µ)]
∂<0 µ βΠ[(δτ + ρ)(ρ + µ)(δ + µ) + (δτ + ρ)(δ − ρ)]
Λ<
σ =
0
× =− <0
∂µ <0 µ2 (ρ + µ)2 (δ + µ)2
The sensitivity indices of the basic reproductive number with respect to main parameters are
found in Table 2. Those parameters that have positive indices (Π, β, σ1 , σ1 , and τ ) show that
they have great impact on expanding the disease in the community if their values are increasing.
Also those parameters in which their sensitivity indices are negative (δ, ρ, ε, and µ) have an effect
of minimizing the burden of the disease in the community as their values increase. Therefore,
research advice for stakholders to work on decreasing the positive indeces and increasing negative
indices parameters
Table 2: Sensitivity indecies table.
Parameter symbol Sensitivity indecies

β +ve
σ1 +ve
σ2 +ve
τ +ve
ε -ve
δ -ve
ρ -ve
µ -ve
11
4 Extension to an Optimal Control Model
Here we extend the basic model in (1) in to optimal control. As we indicated in section 1, Ethiopian
goverment have started a lot of activities to combat COVID-19. Some of preventive activities are;
forced isolation for 14 days for new arrivals, closing country border, quarantining and supporting
infected individuals with medication, restricting number of sits by half in all transport system,
convincing all media outlets to campaigning on COVID-19, announcing state of emergency and
others. As we observe the started activities are not done in optimal level. Therefore, we want to
show the concerned body the effectiveness of those activities if they are implemented in an optimal
level. To perform optimal control, we categorize those activities in to three broad control strategies
listed below.
(i) All rounded prevention strategies including social distancing and personal hygiene (by this
strategy we aimed to block susceptible from contacting the virus)
(ii) Supporting infectives with medication (Optimal support of infected individual in quarantine
center)
(iii) Awareness creation through all Media outlets.
After incorporating the three controls in model (1) gives the following optimal control model in
equation (15) below.
dS
dt
= π + ηR − (1 − u3 ) (1 − u1 ) β(σ1 I + σ2 E)S − µS
dE
dt
= (1 − u3 ) (1 − u1 ) β(σ1 I + σ2 E)S − (u2 + u3 + δ + µ) E
(15)
dI
dt
= (1 − u2 ) τ δ E − (u2 + u3 + ε + ρ + µ) I
dR
dt
= (1 − u2 ) (1 − τ )δE + (1 − u2 ) ε I − (µ + η) R
The purpose of introducing controls in the model is to find the optimal level of the intervention
strategy required to reduce the spreads the pandemic in the population. Here we want to find the
optimal values u1 , u2 and u3 that minimizes the objective functional subject to the differential
equations (15). The objective functional is given as
Ztf
1 2 2 2
J = min a1 E + a2 I + (w1 u1 + w2 u2 + w3 u3 ) dt (16)
u1 ,u2 ,u3 2
0
where tf is the final time, a1 and a2 are weight costs of the Exposed humans and infected humans
respectively while w1 , w2 and w3 are weight costs for each individual control measure. In this
paper, a quadratic function which satisfies the optimality conditions is considered for measuring
12
the cost of the controls as applied by [21, 22, 23]. The goal is to find the optimal control (u∗1 ,u∗2 ,
u∗3 ) such that.
J(u∗1 , u∗2 , u∗3 ) = min{J(u1 , u2 , u3 )|(u1 , u2 , u3 ) ∈ U }
where the control set
U = {(u1 , u2 , u3 ) | ui (t) is lebesgue measurable on [0, tf ], 0 ≤ ui (t) ≤ 1, i = 1, 2, 3}
4.1 Hamiltonian and optimality equation
We used Pontryangin’s Maximum Principle [24] to drive the necessary conditions that an optimal
control must satisfy. This principle converts equation (15) and (16) into a problem of minimizing
point-wise Hamiltonian (H), with respect to u1 (t), u2 (t) and u3 (t) as:
1 1 1
H = H = a1 E + a2 I + w1 u1 2 + w2 u2 2 + w3 u3 2
2 2 2
+ λ1 [Π + η R − (1 − u3 ) (1 − u1 ) β (σ2 E + σ1 I) S − µ S]
+ λ2 [(1 − u3 ) (1 − u1 ) β (σ2 E + σ1 I) S − (δ + u2 + u3 + µ) e]
+ λ3 [(1 − u2 ) τ δ E − (ε + u2 + u3 + ρ + µ) I]
+ λ4 [(1 − u2 ) (1 − τ ) δ E + (1 − u2 ) ε I − (µ + η) R]
Where λi , i = 1, 2, 3, 4 are the adjoint variable associated with S, E, I, and R to be determined

suitably by applying Pontryagin’s maximal principle [24] and also using [25], the existence of an
optimal control is guaranteed.
Theorem 4.1. For an optimal control set u1 , u2 , u3 that minimizes J over U, there are adjoint
variables, λ1 , ..., λ4 such that:

dλ1
= λ1 [(1 − u3 ) (1 − u1 ) β (σ2 E + σ1 I) + µ] − λ2 [(1 − u3 ) (1 − u1 ) β (σ2 E + σ1 I)]




 dt

 dλ2
= −a1 + λ1 [(1 − u3 ) (1 − u1 ) β σ2 S] − λ2 [(1 − u3 ) (1 − u1 ) β σ2 S − (δ + u2 + u3 + µ)]




 dt


−λ3 [(1 − u2 ) τ δ] − λ4 [(1 − u2 ) (1 − τ ) δ]


dλ3
= −a2 + λ1 [(1 − u3 ) (1 − u1 ) β σ1 S] − λ2 [(1 − u3 ) (1 − u1 ) β σ1 S]




 dt


+λ3 [ε + u2 + u3 + ρ + µ] − λ4 [(1 − u2 ) ε]







 dλ4

= −λ1 [η] + λ4 [µ + η]
dt
With transversality conditions, λi (tf ) = 0, i = 1, ..., 4. Furthermore, we obtain the control set
(u∗1 , u∗2 , u∗3 ) characterized by
u∗1 = max{0, min(1, ψ1 )}
13
u∗2 = max{0, min(1, ψ2 )}
u∗3 = max{0, min(1, ψ3 )}
Where
β (1 − u3 ) S (σ2 E + σ1 I) (λ2 − λ1 )
ψ1 =
w1
λ2 E + λ4 ((1 − τ ) δ E + ε I) + λ3 (δ τ E + I)
ψ2 =
w2
β (1 − u1 ) S (σ2 E + σ1 I) (λ2 − λ1 )
ψ3 =
w3
Proof. The adjoint equation and transversality conditions are standard results from Pontryagin’s
maximum principle [24]. We differentiate Hamiltonian with respect to states S, E, I and R respec-
tively, and then the adjoint system is written as

dλ1
= − ∂H = λ1 [(1 − u3 ) (1 − u1 ) β (σ2 E + σ1 I) + µ] − λ2 [(1 − u3 ) (1 − u1 ) β (σ2 E + σ1 I)]




 dt ∂S


 dλ2


 dt
= − ∂H
∂E
= −a1 + λ1 [(1 − u3 ) (1 − u1 ) β σ2 S] − λ2 [(1 − u3 ) (1 − u1 ) β σ2 S − (δ + u2 + u3 + µ)]



−λ3 [(1 − u2 ) τ δ] − λ4 [(1 − u2 ) (1 − τ ) δ]


dλ3
= − ∂H = −a2 + λ1 [(1 − u3 ) (1 − u1 ) β σ1 S] − λ2 [(1 − u3 ) (1 − u1 ) β σ1 S]




 dt ∂I


+λ3 [ε + u2 + u3 + ρ + µ] − λ4 [(1 − u2 ) ε]







 dλ4

= − ∂H = −λ1 [η] + λ4 [µ + η]
dt ∂R
With transversality conditions, λi (tf ) = 0, i = 1, ..., 4. Similarly by following the approach of

Pontryagin et al. [24], the characterization of optimal controls u∗1 , u∗2 , u∗3 , that is, the optimality
∂H
equations are obtained based on the conditions: ∂ui
, f or i = 1, .., 3, which gives,
∂H
= w1 u1 + λ1 [(1 − u3 ) β (σ2 E + σ1 I) S] + λ2 [− (1 − u3 ) β (σ2 E + σ1 I) S]
∂u1
∂H
= w2 u2 + λ2 [−E] + λ3 [−τ δ E − I] + λ4 [− (1 − τ ) δ E − ε I]
∂u2
∂H
= w3 u3 + λ1 [(1 − u1 ) β (σ2 E + σ1 I) S] + λ2 [− (1 − u1 ) β (σ2 E + σ1 I) S]
∂u3
Setting ∂H
∂ui
= 0 at u∗i , the results are
β (1 − u3 ) S (σ2 E + σ1 I) (λ2 − λ1 )
u∗1 =
w1
λ2 E + λ4 ((1 − τ ) δ E + ε I) + λ3 (δ τ E + I)
u∗2 =
w2
β (1 − u1 ) S (σ2 E + σ1 I) (λ2 − λ1 )
u∗3 =
w3
14
When we write by using standard control arguments involving the bounds on the controls, we
conclude
  



 ψ1 , if 0 < ψ1 < 1; 


 ψ2 , if 0 < ψ2 < 1; 


 ψ3 , if 0 < ψ3 < 1;
  
u∗1 = 0, if ψ1 ≤ 0; , u∗2 = 0, if ψ3 ≤ 0; , u∗3 = 0, if ψ3 ≤ 0;

 
 


 
 

1, if ψ1 ≥ 1 1, if ψ3 ≥ 1 1, if ψ3 ≥ 1
In compact notation
u∗1 = max{0, min(1, ψ1 )}, u∗2 = max{0, min(1, ψ2 )}, u∗3 = max{0, min(1, ψ3 )}
15
The optimality system is formed from the optimal control system (the state system) and the
adjoint variable system by incorporating the characterized control set and initial and transversal
condition

dS



 dt
= π + ηR − (1 − u3 ) (1 − u1 ) β(σ1 I + σ2 E)S − µS


 dE
 dt = (1 − u3 ) (1 − u1 ) β(σ1 I + σ2 E)S − (u2 + u3 + δ + µ) E





 dI



 dt
= (1 − u2 ) τ δ E − (u2 + u3 + ε + ρ + µ) I


 dR



 dt
= (1 − u2 ) (1 − τ )δE + (1 − u2 ) ε I − (µ + η) R


 dλ1



 dt
= λ1 [(1 − u3 ) (1 − u1 ) β (σ2 E + σ1 I) + µ] − λ2 [(1 − u3 ) (1 − u1 ) β (σ2 E + σ1 I)]


 dλ2

 = −a1 + λ1 [(1 − u3 ) (1 − u1 ) β σ2 S] − λ2 [(1 − u3 ) (1 − u1 ) β σ2 S − (δ + u2 + u3 + µ)]
 dt


 −λ3 [(1 − u2 ) τ δ] − λ4 [(1 − u2 ) (1 − τ ) δ]



 dλ3



 dt
= −a2 + λ1 [(1 − u3 ) (1 − u1 ) β σ1 S] − λ2 [(1 − u3 ) (1 − u1 ) β σ1 S]






 +λ3 [ε + u2 + u3 + ρ + µ] − λ4 [(1 − u2 ) ε]


 dλ4



 dt
= −λ1 [η] + λ4 [µ + η]


1 I)(λ2 −λ1 )
u∗1 = β (1−u3 )S(σ2 E+σ



 w1



u∗ = λ2 E+λ4 ((1−τ )δ E+ε I)+λ3 (δ τ E+I)

2

 w2



u∗ = β(1−u1 )S(σ2 E+σ1 I)(λ2 −λ1 )

3 w3
(17)
λi (tf ) = 0, i = 1, ..., 4 S(0) = S0 , E(0) = E0 , I(0) = I0 , R(0) = R0
4.2 Uniqueness of the Optimality System
Due to the a priori boundedness of the state, adjoint functions and the resulting Lipschitz structure
of the ODEs, we can obtain the uniqueness of solutions of the optimality system for the small time
interval. Hence the following theorem
Theorem 4.2. For t ∈ [0, tf ], the bounded solutions to the optimality system are unique. For the
proof of the theorem [26].
16
5 Numerical Simulations
In this section, we examine the COVID 19 model and studied the effects of combined strategies
on controlling the transmission of the disease. The optimal control set was obtained by solving
the optimality system, consisting of the state and adjoint systems. An iterative scheme was used
for solving the optimality system. We start to solve the state equations with an initial guess for
the controls over the simulated time using the forward fourth order Runge–Kutta scheme. Be-
cause of the transversality conditions, the adjoint equations were solved by a backward fourth
order Runge–Kutta scheme using the current iterated solutions of the state equation. Then the con-
trols are updated by using a convex combination of the previous controls and the value from the
characterizations. This process is repeated and iterations are stopped if the values of the unknowns
at the previous iteration are very close to the ones at the present iteration. For numerical simulation
purpose, we have used empirical data of COVID-19 cases in Ethiopia for estimation of parameters
in Table 3.
Table 3: Parameter values for COVID-19 model.
Parameter symbol Value Source

π 13.5 Estimated
β 0.0143 Estimated
σ1 0.0001 Assumed
σ2 0.02 Assumed
δ 0.07 Estimated
µ 0.016 Estimated
ρ 0.0004 Estimated
τ 0.7 Assumed
0.15 Estimated
η 0.15 Estimated
We investigated numerically the effect of the following optimal control strategies on the spread of
COVID 19 in a population. Considering strategies that implement one intervention only is not guar-
anteed to reduce and/or eradicate the disease totally from the community. So that those strategies
which incorporate more than one intervention are ordered below and compared pairwise:
• Strategy A: Preventive measures (u1 ) & supporting infectives with medication (u2 )
• Strategy B: Preventive measures (u1 ) & media campaign (u3 )
17
• Strategy C: Supporting infectives with medication (u2 ) & media campaign (u3 )
• Strategy D: Using all control techniques (u1 ,u2 & u3 )
In addition to those parameter values in Table 3, we used a1 = 1, a2 = 5, w1 = 40, w2 = 150 and

w3 = 75 for simulation of COVID 19 pandemic disease model.
5.1 Strategy A: Control with preventive measures & supporting infectives with medication
Here we have investigated the impact of optimal prevention methods and supporting infected indi-
viduals in quarantine center with medication treatment. From the simulation results of Figure 2, we
see that the combination of the two method is effective in controlling COVID-19 in the specified
period. Moreover, we can see that after implementing this strategy the number of exposed and
infected human population goes to zero after 15 months.
Figure 2: Simulations of the COVID 19 model with preventive measure & supporting infectives with medication .
5.2 Strategy B: Control with preventive measures & media campaign
Here also we experimented by combining preventive measure and awareness creation through me-
dia campaign. From the numerical result depicted in Figure 3 below (the left hand side) show as
it is possible to combat the exposed human population of COVID-19 to zero after 8 months and it
will also start to relapse again after 10 months. The right hand side of Figure 3 also indicate that
as it is possible to minimize infected-COVID-19 humans but unable to eliminate by this strategy in
the stated time. Also this method is not effective once the infectious individuals are entered in the
country.
18
Figure 3: Simulations of the COVID 19 model with Social distancing with personal hygiene & Media campaign .
5.3 Strategy C: Control with supporting infectives by medication & media campaign
Here we experimented the model considering optimal support of infected individuals in quaran-
tine center and creating awareness through media. Any activities from preventive technique of
the disease is not considered here. From the numerical result depicted in Figure 4 shows that as
COVID-19 will relapse for second time after it goes to zero in the specified time.
Figure 4: Simulations of the COVID 19 model with Supporting infectives with medication & Media campaign .
5.4 Strategy D: Using All Control Strategies
Here we have applied allstrategies to optimlize the objective function. The results from Figure 5
shows that bringing down the exposed and infected population in short period of time and it is best
19
compared to all the above three control technique. Therefore, government decision makers and all
stakeholders consider in applying all strategies to combat COVID-19 in the specified time.
Figure 5: Simulations of the model with all control strategies.
6 Discussions and Conclusions
In this paper an SEIR deterministic model for the transmission dynamics of the pandemic COVID-
19 in the case of Ethipia was formulated. Model analysis demonstrates that its solutions are positive
and bounded, and there is a region where the model is well-posed mathematically and epidemiolog-
ically meaningful. The basic reproduction number <0 was computed and the stability of equilibria
points was investigated. Through Lyapunov’s theory, the disease free equilibrium point globally
asymptotically stable whenever the <0 < 1 was proven. Using center mainfold theory, bifurcation
analysis of the model was proven and the model exhabts forward bifurication at <0 = 1.
Second, by adding three times-dependent controls, we extend the basic model in to an optimal
control. By using Pontryagin’s Maximum Principle necessary conditions for the optimal control of
the transmission of COVID-19 were derived. From the numerical simulation it was found that the
integrated control strategy, strategy D: using all technique, is very efficient in short period of time.
Therefore, applying all rounded technique by the EFDRE , decision makers and stakeholders have
a significant contribution in combating this pandemic in very short period of time. The result also
shows, if the stakeholders could not do all of the intervention strategies together, the pandemic may
come agian and will transmit over the country.
This paper is still an ongoing research as many more investigations regarding his disease can be
20
carried out. Yet, it serves as the starting phase to research more in depth on questions that COVID-
19 is spreading with incredible speed and have severe consequences. Moreover, to identify those
exposed individuals who doesn’t develop clinical sign, mass screening in any cost is recommended
to combat COVID-19.
Data Availability
All data relevant to this publication will be provided when needed.
Conflicts of Interest
Authors have declared that no competing interests exist.
Authors’ contribution
All authors contributed equally.
References
[1] World Health Organization (WHO). Coronavirus disease 2019 (covid-19): situation report,
68. 2020.
[2] World Health Organisation (WHO). Coronavirus disease (covid-19) outbreak

2020 [internet]. Geneva: WHO; 2020 [cited 2020 27 March], Available from:
https://www.who.int/emergencies/diseases/novelcoronavirus-2019.
[3] European Centre for Disease Prevention and Control (ECDC). Infection pre-
vention and control in the household management of people with suspected or
confirmed coronavirus disease (covid-19). Stockholm: ECDC; 2020, Available
from:https://www.ecdc.europa.eu/en/publications-data/infection-prevention-control-
household-management-covid-19.
[4] Hussin A Rothan and Siddappa N Byrareddy. The epidemiology and pathogenesis of coron-
avirus disease (covid-19) outbreak. Journal of autoimmunity, page 102433, 2020.
21
[5] Weier Wang, Jianming Tang, and Fangqiang Wei. Updated understanding of the outbreak of
2019 novel coronavirus (2019-ncov) in wuhan, china. Journal of medical virology, 92(4):
441–447, 2020.
[6] Shi Zhao, Qianyin Lin, Jinjun Ran, Salihu S Musa, Guangpu Yang, Weiming Wang, Yijun
Lou, Daozhou Gao, Lin Yang, Daihai He, et al. Preliminary estimation of the basic repro-
duction number of novel coronavirus (2019-ncov) in china, from 2019 to 2020: A data-driven
analysis in the early phase of the outbreak. International journal of infectious diseases, 92:
214–217, 2020.
[7] World Health Organization(WHO). Coronavirus disease 2019 (covid-19): situation report,
92. 2020.
[8] World Health Organization (WHO). Coronavirus disease 2019 (covid-19): situation report,
72. 2020.
[9] Centers for Disease Control and Prevention (CDC). Infection prevention and con-
trol in the household management of people with suspected or confirmed coron-
avirus disease (covid-19). Atlanta: CDC;2020[cited 2020 22 Aprile], Available from:
https://www.cdc.gov/coronavirus/2019-ncov/downloads/2019-ncov-factsheet.pdf.
[10] Annelies Wilder-Smith and DO Freedman. Isolation, quarantine, social distancing and com-
munity containment: pivotal role for old-style public health measures in the novel coronavirus
(2019-ncov) outbreak. Journal of travel medicine, 27(2):taaa020, 2020.
[11] Milan Batista. Estimation of the final size of the covid-19 epidemic. Preprint.] medRxiv,
2020.
[12] Liangrong Peng, Wuyue Yang, Dongyan Zhang, Changjing Zhuge, and Liu Hong. Epidemic
analysis of covid-19 in china by dynamical modeling. arXiv preprint arXiv:2002.06563, 2020.
[13] Qianying Lin, Shi Zhao, Daozhou Gao, Yijun Lou, Shu Yang, Salihu S Musa, Maggie H
Wang, Yongli Cai, Weiming Wang, Lin Yang, et al. A conceptual model for the coronavirus
disease 2019 (covid-19) outbreak in wuhan, china with individual reaction and governmental
action. International journal of infectious diseases, 93:211–216, 2020.
[14] Kiesha Prem, Yang Liu, Timothy W Russell, Adam J Kucharski, Rosalind M Eggo, Nicholas
Davies, Stefan Flasche, Samuel Clifford, Carl AB Pearson, James D Munday, et al. The effect
22
of control strategies to reduce social mixing on outcomes of the covid-19 epidemic in wuhan,
china: a modelling study. The Lancet Public Health, 2020.
[15] Alexander Victor. Mathematical predictions for covid-19 as a global pandemic. Available at
SSRN 3555879, 2020.
[16] B Ivorra, MR Ferrándezb, M Vela-Pérez, and AM Ramos. Mathematical modeling of the

spread of the coronavirus disease 2019 (covid-19) taking into account the undetected infec-
tions. the case of china. Research Gate, 2020.
[17] Giulia Giordano, Franco Blanchini, Raffaele Bruno, Patrizio Colaneri, Alessandro Di Filippo,
Angela Di Matteo, Marta Colaneri, et al. A sidarthe model of covid-19 epidemic in italy.
arXiv preprint arXiv:2003.09861, 2020.
[18] P. Van den Driessche and J. Watmough. Reproduction numbers and sub-threshold endemic
equilibria for compartmental models of disease transmission. Mathematical biosciences, 180
(1):29–48, 2002.
[19] C. Castillo-Chavez and B. Song. Dynamical models of tuberculosis and their applications.
Mathematical biosciences and engineering, 1(2):361–404, 2004.
[20] S. M. Blower and H. Dowlatabadi. Sensitivity and uncertainty analysis of complex models
of disease transmission: an hiv model, as an example. International Statistical Review/Revue
Internationale de Statistique, pages 229–243, 1994.
[21] Getachew Teshome Tilahun, Oluwole Daniel Makinde, and David Malonza. Modelling and
optimal control of typhoid fever disease with cost-effective strategies. Computational and
mathematical methods in medicine, 2017, 2017.
[22] Getachew Teshome Tilahun, Oluwole Daniel Makinde, and David Malonza. Co-dynamics of
pneumonia and typhoid fever diseases with cost effective optimal control analysis. Applied
Mathematics and Computation, 316:438–459, 2018.
[23] Getachew Teshome Tilahun, Oluwole Daniel Makinde, and David Malonza. Modelling and
optimal control of pneumonia disease with cost-effective strategies. Journal of Biological
Dynamics, pages 1–27, 2017.
[24] Lev Semenovich Pontryagin. Mathematical theory of optimal processes. CRC Press, 1987.
23
[25] Wendeil H Fleming and Raymond W Rishel. Deterministic and stochastic optimal control.
Applications of Mathematics. Volume, 1, 1976.
[26] K Renee Fister, Suzanne Lenhart, and Joseph Scott McNally. Optimizing chemotherapy in an
hiv model. Electronic Journal of Differential Equations, 1998(32):1–12, 1998.
24

1 Covid-19

Uploaded by

Copyright:

Available Formats

1 Covid-19

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

1 Covid-19

Uploaded by

Copyright:

Available Formats

medRxiv preprint doi: https://doi.org/10.1101/2020.07.23.20160473; this version posted July 24, 2020.

The copyright holder for this preprint

Mathematical Modeling and Optimal Control Analysis of

Haileyesus Tessema Alemneh 1 and Getachew Teshome Tilahun 2

Keywords: COVID-19; Optimal control; Modeling; Ethiopia; Numerical Simulation.

2 Model Description and Formulation

S(0) = S0 ≥ 0 , E(0) = E0 ≥ 0 I(0) = I0 ≥ 0 , R(0) = R0 ≥ 0 (2)

Figure 1: Compartmental flow diagram of the pandemic COVID 19 transmission

Table 1: Description of parameters of the model (1).

3.1 Invariant Region

3.2 Positivity of Solutions

Proof. : From the first equation of the system

which can be taken as

Therefore, all the solution sets are positive for t ≥ 0.

3.3 Disease Free Equilibrium Point(DFEP)

3.4 Basic reproduction number

3.5 Local Stability of DFEP

Proof. The Jacobian matrix, evaluated at the disease-free equilibrium E0 , we get:

The characteristic polynomial from the Jacobian matrix is

(−λ − (η + µ))(−λ − µ)(λ2 + ψ1 λ + ψ2 ) = 0 (8)

−λ − (η + µ) ⇒ λ1 = −(η + µ) < 0 and − λ − µ ⇒ λ2 = −µ < 0

From the last expression, that is

Hence the DFEP is locally asymptotically stable if <0 < 1.

3.6 Global Stability of DFEP

Proof. Consider the following Lyapunov function

Differentiating equation (10) with respect to t gives

Here take c2 = − βσ2 S−(δ+µ)

3.7 Existence of endemic equilibrium (EEP)

κ = δ 2 ητ 2  − (ε + ρ + µ)[δ 2 τ (ητ + µ) + δτ (ητ + µ2 )]

−(ε + ρ + µ)2 µ(δ + η + µ) − δηµτ (2µ + 3ρ) − δητ ρ2

Proof. The Jacobian matrix of system (1) at the DFEP is given by

3.8 Sensitivity Analysis

β Π(σ1 τ δ+σ2 (+ρ+µ))

Table 2: Sensitivity indecies table.

Parameter symbol Sensitivity indecies

4 Extension to an Optimal Control Model

(iii) Awareness creation through all Media outlets.

4.1 Hamiltonian and optimality equation

Where λi , i = 1, 2, 3, 4 are the adjoint variable associated with S, E, I, and R to be determined

u∗2 = max{0, min(1, ψ2 )}

u∗3 = max{0, min(1, ψ3 )}

With transversality conditions, λi (tf ) = 0, i = 1, ..., 4. Similarly by following the approach of

 −λ3 [(1 − u2 ) τ δ] − λ4 [(1 − u2 ) (1 − τ ) δ]

4.2 Uniqueness of the Optimality System

Table 3: Parameter values for COVID-19 model.

Parameter symbol Value Source

• Strategy B: Preventive measures (u1 ) & media campaign (u3 )

• Strategy D: Using all control techniques (u1 ,u2 & u3 )

In addition to those parameter values in Table 3, we used a1 = 1, a2 = 5, w1 = 40, w2 = 150 and

5.2 Strategy B: Control with preventive measures & media campaign

5.4 Strategy D: Using All Control Strategies

Figure 5: Simulations of the model with all control strategies.

6 Discussions and Conclusions

All data relevant to this publication will be provided when needed.

Authors have declared that no competing interests exist.

All authors contributed equally.

[2] World Health Organisation (WHO). Coronavirus disease (covid-19) outbreak

[16] B Ivorra, MR Ferrándezb, M Vela-Pérez, and AM Ramos. Mathematical modeling of the

You might also like

κ = δ 2 ητ 2 − (ε + ρ + µ)[δ 2 τ (ητ + µ) + δτ (ητ + µ2 )]

β Π(σ1 τ δ+σ2 (+ρ+µ))