Received: 8 September 2020

DOI: 10.1002/mma.6935

S P E C I A L I S S U E PA P E R

The analysis of a time delay fractional COVID-19 model via

Caputo type fractional derivative

Pushpendra Kumar1 Vedat Suat Erturk2

1
Department of Mathematics and
Statistics, School of Basic and Applied Novel coronavirus (COVID-19), a global threat whose source is not correctly yet
Sciences, Central University of Punjab, known, was firstly recognised in the city of Wuhan, China, in December 2019.
Bathinda, India
2
Now, this disease has been spread out to many countries in all over the world. In
Department of Mathematics, Ondokuz
Mayis University, Samsun, Turkey this paper, we solved a time delay fractional COVID-19 SEIR epidemic model via
Caputo fractional derivatives using a predictor–corrector method. We provided
Correspondence
numerical simulations to show the nature of the diseases for different classes.
Pushpendra Kumar, Department of
Mathematics and Statistics, School of Basic We derived existence of unique global solutions to the given time delay frac-
and Applied Sciences, Central University tional differential equations (DFDEs) under a mild Lipschitz condition using
of Punjab, Bathinda 151001, India.
Email: kumarsaraswatpk@gmail.com
properties of a weighted norm, Mittag–Leffler functions and the Banach fixed
point theorem. For the graphical simulations, we used real numerical data based
Communicated by: M. Efendiev on a case study of Wuhan, China, to show the nature of the projected model
with respect to time variable. We performed various plots for different values of
time delay and fractional order. We observed that the proposed scheme is highly
emphatic and easy to implementation for the system of DFDEs.

K E Y WO R D S
Caputo fractional derivative, COVID-19 epidemic, fixed point theory, predictor–corrector scheme,
SEIR model, time delay

M S C C L A S S I F I C AT I O N
26A33; 34C60; 92C60; 92D30

1 I N T RO DU CT ION

Novel coronavirus (COVID-19), a global threat whose source is not correctly yet known, was firstly recognised in the city
of Wuhan, China, in December 2019.1,2 Now, this disease has been spread out to many countries in all over the world.
From the date of its origin, it grows exponentially in mankind and infected more than 20,254,685 with 738,930 deaths and
13,118,618 recoveries on 11 August throughout the globe. For the preventions to this disease, social control measures have
been extended up by increased public awareness such as through social or physical distancing measures, good hygiene
and not walking out in the open environment. Billions of people have been infected by this virus, and in some particular
cases, the virus has also been recognised in animals. The symptoms of the coronavirus are fever, sneeze, a runny nose,
fatigue, dry cough, bilateral lung infiltration to severely ill and breathing problems,3 in which fever and dry cough are
two most common symptoms.4 So many newly reported cases have been associated to the travel history from an epidemic
region or a meeting history to people from the region in the early outbreak phases. Because a person who was infected
by COVID-19 can travel to any other region and spread the virus. This coronavirus outbreak has received considerable
global attention when, since 31 Dec 2019, the Wuhan Municipal Health Commission reported 27 cases of viral pneumo-
nia, with 7 critically ill cases. Many researchers and doctors have suggested their specialisation to the study of this virus.

Math Meth Appl Sci. 2020;1–14. wileyonlinelibrary.com/journal/mma © 2020 John Wiley & Sons, Ltd. 1
2 KUMAR AND ERTURK

COVID-19 has been studied from different points of view, in many fields of study, including epidemiology, microbiology,
environmental and occupational health, virology, veterinary science, economics, sociology and media studies. In reac-
tion to early outbreaks of the virus, the United States, Korea and China come out as the leading countries in COVID-19
research. Other research groups focused on the improvement of healthcare systems and virus prevention propounding
that the way to restraining virus transmission is the reduction of human contact and social distance by dint of the enclose
of public transport, school closures, the adjournment of common activities, etc.5 To date, there is no specific vaccine or
clinically established treatment for coronavirus. Nevertheless, according to the suggestions of different prudent infectious
disease specialists, several countries have started using chloroquine combined with azithromycin as alternative drugs.
But still there are some countries that successfully controlled the pandemic. These include Iceland, New Zealand, Fiji,
Vatican City, Tanzania, Montenegro, Seychelles, Papua New Guinea and Mauritius. On 28 February 2020, New Zealand
reported its first coronavirus case and then enforced the one of the strictest lockdowns in the world. The United States,
Brazil, India and Peru are the countries which are most infected by this virus. Particularly, the United States is one of the
most infected country of this virus with 5,251,446 cases, 166,192 deaths and 2,715,934 recoveries on 11 August. In Brazil,
3,057,470 cases with 101,857 deaths and 2,163,812 recoveries have been recorded. In India, there are 2,267,153 cases
with 45,353 deaths and 1,581,640 recoveries founded on August 11. Because of this pandemic, lots of social programmes
and activities have been cancelled or postponed. T-20 cricket world cup 2020 in Australia, Summer Olympics which was
planned to be held in Tokyo, has been postponed. One of the most popular cricket tournament, Indian Premier League
(IPL), has been shifted from India to UAE.
Mathematical models estimate the number of cases in best and worst case scenarios when disease dynamics are unclear
and can also be used to estimate the effects of preventive (healthcare) measures. A lot of research articles have been
come in literature to analyse the effects of COVID-19 via mathematical modelling.6-10 In Khan et al,11 authors formulated
a new mathematical model for the dynamics of COVID-19 with quarantine and isolation. They studied the dynamics
of the COVID-19 using fractal–fractional Atangana–Baleanu derivative. A study on the mathematical modelling of the
impact of nonpharmaceutical interventions on the dynamics of COVID-19 with optimal control analysis is given in Ullah
and Khan.12 A study on solution of a COVID-19 model via new generalised fractional derivative (FD) is done by Erturk
and Kumar.13 They used a modified predictor–corrector scheme to perform the numerical simulations with the real data
of Wuhan, China. The applications of fractional calculus have been received in various branches of Science and Engi-
neering.14-18 Unlike an ordinary derivative (OD) operator, an FD operator is non-local in nature. Due to this non-local
behaviour of the FD operator, it can formulate processes having memory and hereditary properties. Recently, an atten-
tion to the delay fractional differential equations (DFDEs) has considered cause of their applications in the mathematical
modelling of real-world problems. The delay differential equation is a differential equation in which the derivative of
the function at any time depends on the solution at previous time. It is well known that an ordinary delay differential
equation has a unique local solution under some Lipschitz conditions; furthermore, using continuation property, one can
derive global solutions as well, but in the noninteger, the existence of unique solutions (local and global) is more intricate
because of the noninteger order feature of the equation which implies history dependence of the solutions; hence, among
others, the continuation property is not applicable.
The aim of this article is to study the time delay fractional COVID-19 epidemic model using a real numerical data
of a case study of Wuhan, China, from the literature to show the nature of the given model. We also performed some
analysis to show the role of time delay parameter 𝜏 by the help of Caputo FD with predictor–corrector algorithm. The
paper is formulated as follows. In Section 2, we recall some important definitions of the FDs. Section 3 is devoted for
the description of the ordinary differential equations (ODE) model following the fractional order model. Existence and
uniqueness analysis of the problem are performed in Section 4. Solution of the projected model is done in Section 5.
Simulation results are performed in Section 6. A conclusion completes the paper.

2 PRELIMINARIES

Here, we remind some basic definitions and properties.

Definition 1 (Podlubny17 ). The Riemann–Liouville (R-L) definition of noninteger order integral of order 𝜁 > 0 of a
function G ∶ (0, ∞) → R is defined by

t
1
It𝜁 G (t) = (t − 𝜃)𝜁 −1 G(𝜃)d𝜃. (1)
Γ(𝜁 ) ∫0
KUMAR AND ERTURK 3

Definition 2 (Podlubny17 ). The R-L definition of noninteger order derivative of order 𝜁 > 0 of a function G ∶ (0, ∞) →
R is defined by
( )n t
𝜁 d 1
Dt G (t) = (t − 𝜃)n−𝜁 −1 G(𝜃)d𝜃, (2)
dt Γ(n − 𝜁) ∫0
where n = [𝜁] + 1 and [𝜁 ] is the integer part of 𝜁.

Definition 3 (Podlubny17 ). The Caputo definition of noninteger order derivative of order 𝜁 > 0 of a function G ∶
(0, ∞) → R is defined by
t
1
D𝜁t G (t) = (t − 𝜃)n−𝜁 −1 Gn (𝜃)d𝜃, (3)
Γ(n − 𝜁) ∫0
where n = [𝜁 ] + 1 and [𝜁 ] is the integer part of 𝜁.

Definition 4 (Podlubny17 ). The Mittag–Leffler type function with one parameter is defined as follows E𝜇 (z) =
∑∞ zn
n=0 Γ(𝜇n+1) , 𝜇 > 0, z ∈ C.

3 MODEL D ESCRIPTION

3.1 Description of the ODE model

There are so many mathematical models have been introduced in the literature to study the outbreaks of COVID-19. So
many researchers have analysed various type of models to study the dynamics of COVID-19 with the case study of differ-
ent particular countries. In this section, we analyse the time delay ordinary model studied by Cakan.19 to introduce the
COVID-19 epidemic. The considered model consists of four compartments with individuals of susceptible S(t), exposed
E(t), infectious I(t) and recovered R(t). The author presented and derived the projected model in the sense of ordinary
time delay differential equations as follows:

⎧S′ (t) = b − 𝛽S (t) I (t) − dS (t) ,

⎪
⎪E′ (t) = 𝛽S (t) I (t) − 𝛾𝛽S (t − 𝜏) I (t − 𝜏) e−d𝜏 − dE (t) − 𝛿E (t) ,
⎨′ (4)
⎪I (t) = 𝛾𝛽S (t − 𝜏) I (t − 𝜏) e − [𝜈1 + 𝜈2 (1 − c (t))] I (t) − [𝜃1 + 𝜃2 c (t)] I (t) − dI (t) ,
−d𝜏

⎪R′ (t) = [𝜃 + 𝜃 c (t)] I (t) + 𝛿E (t) − dR (t) .

⎩ 1 2

In the given ordinary model (4), b is the birth rate, 𝛽 is the contact rate of susceptible to infected persons and natural
death rate denoted by d. 𝛾 and 𝛿 are the progression rate of exposed humans into infectious population and rate of exposed
to removed, respectively. The brief description of all other parameters with the numerical values is given in Table 1.
The function c(t) presents the available opportunities level by healthcare system to public who are infected at time t. It
can be observed that all hospital facilities are almost consumed away when c(t) tends to zero and all hospital facilities
(opportunities) can be used fully when c(t) = 1 with respect to the time variable t. 𝜏 is a time delay corresponding the
latent period of the COVID-19.

TABLE 1 Parameter values for simulations

Parameter Description Value/range Reference
b Birth rate 3,210 Estimated
𝛽 Contact rate susceptible to infected 0.62 × 10−8 Yang and Wang24
d Natural death rate 3.57 × 10−5 Khan et al11
𝛾 Rate of exposed to infected 0.143 Yang and Wang24
𝛿 Rate of exposed to removed 0.006 Fitted
c Level of available opportunities by health care systems [0, 1] Yang and Wang24
𝛼1 Natural recovery rate of the infectious class 0.0005 Fitted
𝛼2 Recovery rate of the infectious class 0.0667 Yang and Wang24
𝜇1 Minimum disease-induced death rate 0.01 Yang and Wang24
𝜇1 + 𝜇2 Maximum disease-induced death rate 0.02 Fitted
4 KUMAR AND ERTURK

3.2 Description of the Caputo fractional model

The theory of the time DFDEs is a well known phenomena of fractional calculus. Now, we generalise the above ODE
model in the Caputo FD sense. Here, we are including the death equation also. In this generalisation, we replace the OD
operator by the Caputo fractional operator C D𝜁t . So the generalisation of the given ordinary time delay differential equation
system into the time DFDEs system is as follows:

⎧C D𝜁 S (t) = b − 𝛽S (t) I (t) − dS (t) ,

⎪ t
⎪C D𝜁t E (t) = 𝛽S (t) I (t) − 𝛾𝛽S (t − 𝜏) I (t − 𝜏) e−d𝜏 − dE (t) − 𝛿E (t) ,
⎪C 𝜁
⎨ Dt I (t) = 𝛾𝛽S (t − 𝜏) I (t − 𝜏) e−d𝜏 − [𝜈1 + 𝜈2 (1 − c (t))] I (t) − [𝜃1 + 𝜃2 c (t)] I (t) − dI (t) , (5)
⎪C 𝜁
⎪ Dt R (t) = [𝜃1 + 𝜃2 c (t)] I (t) + 𝛿E (t) − dR (t) ,
⎪C D𝜁 D(t) = [𝜈1 + 𝜈2 (1 − c (t))] I (t) ,
⎩ t

where C D𝜁t denotes the Caputo FD operator.

By means of the next generation matrix method,20 here, we find the basic reproduction number 0 for the fractional
model (5) as follows:
Let us assume Y = (I, S)T . Then, the model (5) can be written as

C
D𝜁t Y =  (Y ) − (Y ), (6)

where
[ ]
𝛾𝛽S (t − 𝜏) I (t − 𝜏) e−d𝜏
 (Y ) =
0
and
[[ ] ]
𝜈1 + 𝜈2 (1 − c (t)) + 𝜃1 + 𝜃2 c (t) + d I (t)
(Y ) = .
𝛽S (t) I (t) + dS (t) − b,
( )T
b
Y0 = ,0 is the unique disease-free equilibrium point of the model (6).
d
( )
b
Then, Jacobian matrices at the disease-free equilibrium point P∗ = (S∗ , I ∗ ) = d
,0 of F(Y) and V(Y) by looked to the
FDs with respect to I and S are obtained as
[ ]
𝛾𝛽S∗ e−d𝜏 𝛾𝛽I ∗ e−d𝜏
  (P ) =
∗
,
0 0

and
[ ]
𝜈1 + 𝜈2 (1 − c (t)) + 𝜃1 + 𝜃2 c (t) + d 0
 (P∗ ) = 𝛽S ∗
𝛽I + d ,
∗

respectively. So F and V, which are the new infection terms and the remaining transfer terms of the model, respectively,
are determined as
[ ]
b𝛽𝛾e−d𝜏
F =  1×1 =
d
and
[ ]
V =  1×1 = 𝜈1 + 𝜈2 (1 − c(t)) + 𝜃1 + 𝜃2 c (t) + d .
Also, the characteristic polynomial of
[ ]
−1 b𝛽𝛾e−d𝜏
FV =
d (𝜈1 + 𝜈2 (1 − c (t)) + 𝜃1 + 𝜃2 c (t) + d)
KUMAR AND ERTURK 5

is
( ) b𝛽𝛾e−d𝜏
det 𝜆I1 − FV −1 = 𝜆 − .
d (𝜈1 + 𝜈2 (1 − c (t)) + 𝜃1 + 𝜃2 c (t) + d)

It follows from Van den Driessche and Watmough20 that the basic reproduction number of the model (5) is the spectral
radius of the next generation matrix FV−1 . In that case,

b𝛽𝛾e−d𝜏
0 = . (7)
d (𝜈1 + 𝜈2 (1 − c (t)) + 𝜃1 + 𝜃2 c (t) + d)

The equivalent compact form of the above system (5) is as follows:

⎧C D𝜁 S(t) = G (t, S(t), S(t − 𝜏)),

⎪ t 1
⎪C D𝜁t E(t) = G2 (t, E(t), E(t − 𝜏)),
⎪C 𝜁
⎨ Dt I(t) = G3 (t, I(t), I(t − 𝜏)), (8)
⎪C 𝜁
⎪ Dt R(t) = G4 (t, R(t), R(t − 𝜏)),
⎪C D𝜁 D(t) = G5 (t, D(t), D(t − 𝜏)),
⎩ t

with the initial conditions taken as S(0) = k1 , E(0) = k2 , I(0) = k3 , R(0) = k4 and D(0) = k5 .

4 MATHEMATICAL A NA LYSIS O F THE FRACTIONAL MODEL

4.1 Existence and uniqueness analysis

In this section, we give the existence of unique solution for the projected fractional time delay COVID-19 model by the
help of the consequences of fixed point theory. In this regard, many results have been given in the literature, and here, we
are following the procedure proposed by Cong and Tuan.21 We show the analysis for I(t), and for other equations of the
system (8), it will be similar. Let us consider the fractional time delay equations

C
D𝜁t I(t) = G3 (t, I(t), I(t − 𝜏)), t ∈ [0, T], 0 < 𝜁 ≤ 1, (9)

with the initial condition

I(t) = k3 , t ∈ [−𝜏, 0], (10)

where I ∈ Rn , T > 0, & G3 ∶ [0, T] × Rn × Rn → Rn is continuous.

(Rn be the n-dimensional Euclidean space defined with a norm ||.||)
Lemma 1 (Cong and Tuan21 ). The function 𝜙 ∈ C([−𝜏, T]; Rn ) ( space of continuous functions 𝜙 ∶ [−𝜏, T] → Rn
with the sup norm ||.||∞ ) is a solution of the initial value problem (IVP) (Equations 9–10) on the interval [− 𝜏, T] if it is a
solution of the delay integral equation

t
1
I(t) = I(0) + (t − 𝜉)𝜁 −1 G3 (𝜉, I(𝜉), I(𝜉 − 𝜏))d𝜉, ∀t ∈ [0, T] (11)
Γ(𝜁 ) ∫0

with the initial condition

I(t) = k3 , t ∈ [−𝜏, 0]. (12)

Note:21 In this particular way to show the existence of unique global solutions, we don't need to require Lipschitz
property of G3 with respect to the time delay variable of G3 , but only the Lipschitz property of G3 with respect to the
nondelay (second) variable.
6 KUMAR AND ERTURK

Theorem 1 (Existence and uniqueness of global solutions). Assume that G3 ∶ [0, T]× Rn × Rn → Rn is continuous and
agree with the following Lipschitz condition with respect to the nondelay variable: there exists a non-negative continuous
function L ∶ [0, T] × Rn → R≥0 , such that

||G3 (t, I, z) − G3 (t, I1 , z)|| ≤ L(t, z)||I − I1 || (13)

∀ t ∈ [0, T], I, z, I1 ∈ Rn . Then, the IVP (Equations 9–10) has a unique global solution 𝜙 on the interval [− 𝜏, T].

Proof. According to Lemma (1), Equation (9) with Equation (10) is equivalent to the initial value problem (11)–(12).
First, we take the case 0 < T ≤ 𝜏. In this case, Equation (11) has the form:

t
1
I(t) = I(0) + (t − 𝜉)𝜁 −1 G3 (𝜉, I(𝜉), k3 )d𝜉, ∀t ∈ [0, T].
Γ(𝜁 ) ∫0

For this integral equation, by Tisdell [22 Theorem 6.4, p. 310], there exists a unique solution on the interval [0, T].
Identify that solution by 𝜅𝜏∗ and put
{
k3 , t ∈ [−𝜏, 0],
𝜙T (t, k3 ) ∶= 𝜅 ∗ (t), t ∈ [0, T]. (14)
𝜏

Then, 𝜙T (t, k3 ) is the unique solution of the problem (11)–(12) on [− 𝜏, T].

Now, in another case when T > 𝜏, we break the interval [0, T] into [0, 𝜏] ∪ … ∪ [(m0 − 1)𝜏, m0 𝜏] ∪ [m0 𝜏, T], where
m0 ∈ N and 0 ≤ T − m0 𝜏 < 𝜏. On the interval [− 𝜏, 𝜏], using the same arguments as above, we can derive a unique
solution of the IVP (Equations 11–12), which is identified by 𝜙𝜏 . We will prove the existence of the unique solution
on the interval [− 𝜏, m0 𝜏] by induction. Let us assume that the problem (11)–(12) has a unique solution on the inter-
val [− 𝜏, m𝜏] for some 1 ≤ m < m0 . We denote that solution by 𝜙m𝜏 (., k3 ). On [m𝜏, (m + 1)𝜏], we define an operator
F(m+1)𝜏,k3 ∶ C([m𝜏, (m + 1)𝜏]; Rn ) → C([m𝜏, (m + 1)𝜏]; Rn ) as follows:

m𝜏
1
(F(m+1)𝜏,k3 𝜅)(t) ∶= I(0) + (t − 𝜉)𝜁 −1 G3 (𝜉, 𝜙m𝜏 (𝜉, k3 ), 𝜙m𝜏 (𝜉 − 𝜏, k3 ))d𝜉
Γ(𝜁 ) ∫0

t
1
+ (t − 𝜉)𝜁 −1 G3 (𝜉, 𝜅(𝜉), 𝜙m𝜏 (𝜉 − 𝜏, k3 ))d𝜉 ∀t ∈ [m𝜏, (m + 1)𝜏].
Γ(𝜁 ) ∫m𝜏

Let 𝛽 m be a positive constant satisfying 𝛽m > 2maxt∈[m𝜏,(m+1)𝜏] L(t, 𝜙m𝜏 (t − 𝜏, k3 )). On the space C([m𝜏, (m + 1)𝜏]; Rn ),
we define a new metric d𝛽m by

||𝜅(t) − 𝜅1 (t)||
d𝛽m (𝜅, 𝜅1 ) ∶= sup , ∀𝜅, 𝜅1 ∈ C([m𝜏, (m + 1)𝜏] ∶ Rn ),
t∈[m𝜏,(m+1)𝜏] E𝜁 (𝛽m t𝜁 )

where E𝜁 ∶ R → R is the Mittag–Leffler function which is defined in Definition 4. Then, the space C([m𝜏, (m +
1)𝜏]; Rn ) equipped the metric d𝛽m is complete. We will show that the operator F(m+1)𝜏,k3 is contractive on (C([m𝜏, (m +
1)𝜏]; Rn ), d𝛽m ). Indeed, for any 𝜅, 𝜅1 ∈ C([m𝜏, (m + 1)𝜏]; Rn ) and any t ∈ [m𝜏, (m + 1)𝜏], we have

||(F(m+1)𝜏,k3 𝜅)(t) − (F(m+1)𝜏,k3 𝜅1 )(t)||

max L(t, 𝜙m𝜏 (t − 𝜏, k3 )) t
t∈[m𝜏,(m+1)𝜏]
≤ (t − 𝜉)𝜁 −1 ||𝜅(𝜉) − 𝜅1 (𝜉)||d𝜉
Γ(𝜁 ) ∫m𝜏 (15)
max L(t, 𝜙m𝜏 (t − 𝜏, k3 )) t
t∈[m𝜏,(m+1)𝜏] ||𝜅(𝜉) − 𝜅1 (𝜉)||
≤ (t − 𝜉)𝜁 −1 E𝜁 (𝛽m 𝜉 𝜁 ) d𝜉.
Γ(𝜁 ) ∫m𝜏 E𝜁 (𝛽m 𝜉 𝜁 )
KUMAR AND ERTURK 7

This implies that

||(F(m)𝜏,k3 𝜅)(t) − (F(m)𝜏,k3 𝜅1 )(t)||

E𝜁 (𝛽m t𝜁 )
max L(t, 𝜙m𝜏 (t − 𝜏, k3 )) t
t∈[m𝜏,(m+1)𝜏] 1
≤ d𝛽m (𝜅, 𝜅1 ) (t − 𝜉)𝜁 −1 E𝜁 (𝛽m 𝜉 𝜁 )d𝜉
E𝜁 (𝛽m t𝜁 ) Γ(𝜁 ) ∫m𝜏
max L(t, 𝜙m𝜏 (t − 𝜏, k3 )) t
t∈[m𝜏,(m+1)𝜏] 1
≤ d𝛽m (𝜅, 𝜅1 ) (t − 𝜉)𝜁 −1 E𝜁 (𝛽m 𝜉 𝜁 )d𝜉 (16)
E𝜁 (𝛽m t𝜁 ) Γ(𝜁 ) ∫0
max L(t, 𝜙m𝜏 (t − 𝜏, k3 )) ( ( ))
t∈[m𝜏,(m+1)𝜏] 𝜁 C 𝜁 E𝜁 (𝛽m t𝜁 )
≤ d𝛽m (𝜅, 𝜅1 )I0 D0
E𝜁 (𝛽m t𝜁 ) 𝛽m
max L(t, 𝜙m𝜏 (t − 𝜏, k3 ))
t∈[m𝜏,(m+1)𝜏]
≤ d𝛽m (𝜅, 𝜅1 ),
𝛽m

for all t ∈ [m𝜏, (m + 1)𝜏]. Therefore,

max L(t, 𝜙m𝜏 (t − 𝜏, k3 ))

t∈[m𝜏,(m+1)𝜏]
d𝛽m (F(m+1)𝜏,k3 𝜅, F(m+1)𝜏,k3 𝜅1 ) ≤ d𝛽m (𝜅, 𝜅1 )
𝛽m (17)
1
≤ d𝛽m (𝜅, 𝜅1 ),
2

for all 𝜅, 𝜅1 ∈ C([m𝜏, (m + 1)𝜏]; Rn ). By a statement of the Banach fixed point theorem, there exists a unique fixed
point 𝜅(m+1)𝜏
∗
of F(m+1)𝜏,k3 in C([m𝜏, (m + 1)𝜏]; Rn ). Put

{
𝜙m𝜏(t,k3 ) , ∀t ∈ [−𝜏, m𝜏],
𝜙(m+1)𝜏 (t, k3 ) ∶= 𝜅(m+1)𝜏
∗
(t), ∀t ∈ [m𝜏, (m + 1)𝜏]. (18)

Then, 𝜙(m + 1)𝜏(t, k3 ) is the unique solution of the problem (11)–(12) on [− 𝜏, (m + 1)𝜏].
Finally, on the interval [m0 𝜏, T], we derive an operator Fk3 ∶ C([m0 𝜏, T]; Rn ) → C([m0 𝜏, T]; Rn ) by

m0 𝜏
1
(Fk3 )(t) ∶= I(0) + (t − 𝜉)𝜁 −1 G3 (𝜉, 𝜙m0 𝜏 (𝜉, k3 ), 𝜙m0 𝜏 (𝜉 − 𝜏, k3 ))d𝜉
Γ(𝜁 ) ∫0

t
1
+ (t − 𝜉)𝜁 −1 G3 (𝜉, 𝜅(𝜉), 𝜙m0 𝜏 (𝜉 − 𝜏, k3 ))d𝜉 ∀t ∈ [m0 𝜏, T].
Γ(𝜁 ) ∫m0 𝜏

Let 𝛽m0 be a positive constant satisfying 𝛽m0 > 2maxt∈[m0 𝜏,T] L(t, 𝜙m0 𝜏 (t−𝜏, k3 )). On the space C([m0 𝜏, T]; Rn ), we define
a new metric d𝛽m0 by

||𝜅(t) − 𝜅1 (t)||
d𝛽m0 (𝜅, 𝜅1 ) ∶= sup ,
t∈[m0 𝜏,T] E𝜁 (𝛽m0 t𝜁 )
8 KUMAR AND ERTURK

and repeating arguments as above, we can show that the operator Fk3 has a unique fixed point 𝜅 ∗ on [m0 𝜏, T]. Define
a function
{
𝜙m0 𝜏 (t, k3 ), ∀t ∈ [−𝜏, m0 𝜏],
𝜙T (t, k3 ) ∶= 𝜅 ∗ (t), ∀t ∈ [m 𝜏, T]. (19)
0

It is evident that 𝜙T is the unique solution of the problem (11)–(12) on the interval [− 𝜏, T].

5 SOLUTION O F T HE PROJECTED MODEL USING PREDICTO R–

CO RRECTOR ALGORITHM

We know that the numerical methods or techniques used for solving ordinary differential equations cannot be used
directly to solve noninteger order differential equations. Lots of numerical methods have been used to solve DFDEs aris-
ing in biology. In this section, we find the solution of the projected model by the help of Adams–Bashforth–Moulton
predictor–corrector scheme described in Bhalekar and Daftardar-Gejji23 for solving DFDEs.
Let us consider the fractional delay differential equation

C
D𝜁t I(t) = G3 (t, I(t), I(t − 𝜏)), t ∈ [0, T], 0 < 𝜁 ≤ 1, (20a)

I(t) = k3 , t ∈ [−𝜏, 0]. (20b)

Consider a uniform grid {tm = mh ∶ m = −n, −n + 1, … , −1, 0, 1, … , N}, where n and N are integers such that
h = T∕N and h = 𝜏∕n. Let

Ih (ti ) = k3 , i = −n, −n + 1, … , −1, 0, (21)

and note that

Ih (ti − 𝜏) = Ih (ih − nh) = Ih (ti−n ), i = 0, 1, … , N. (22)

Suppose we have already calculated the approximations Ih (ti ) ≈ I(ti ), (i = −n, −n + 1, … , −1, 0, 1, … , m) and we want
to calculate Ih (tm + 1 ) using the volterra integral equation equivalent to the Equations (20a) and (20b)

tm+1
1
I(tm+1 ) = I(0) + (tm+1 − 𝜉)𝜁 −1 G3 (𝜉, I(𝜉), I(𝜉 − 𝜏))d𝜉. (23)
Γ(𝜁 ) ∫0

We use approximations Ih (tm ) for I(tm ) in Equation (23). Further, the integral in Equation (23) is evaluated using product
trapezoidal quadrature formula. The corrector formula is thus

h𝜁
Ih (tm+1 ) = I(0) + G3 (tm+1 , Ih (tm+1 ), Ih (tm+1 − 𝜏))
Γ(𝜁 + 2)
h𝜁 ∑
m
+ ai,m+1 G3 (ti , Ih (ti ), Ih (ti − 𝜏))
Γ(𝜁 + 2) i=0
(24)
h𝜁
=I(0) + G3 (tm+1 , Ih (tm+1 ), Ih (tm+1−n ))
Γ(𝜁 + 2)
h𝜁 ∑
m
+ ai,m+1 G3 (ti , Ih (ti ), Ih (ti−n )),
Γ(𝜁 + 2) i=0
KUMAR AND ERTURK 9

where
{
m𝜁 +1 − (m − 𝜁 )(m + 1)𝜁 , i = 0
ai,m+1 = (m − i + 2)𝜁 +1 − 2(m − i + 1)𝜁 +1 + (m − i)𝜁 +1 , 1 ≤ i ≤ m
1, i = m + 1.
The unknown term Ih (tm + 1 ) present on both sides of Equation (24) and due to non-linearity of G3 Equation (24) cannot be
solved explicitly for Ih (tm + 1 ). So we replace the term Ih (tm + 1 ) on the right hand side by an approximation IhP (tm+1 ), called
predictor. Product rectangle rule is used in Equation (24) to evaluate predictor term

1 ∑
m
IhP (tm+1 ) =I(0) + bi,m+1 G3 (ti , Ih (ti ), Ih (ti − 𝜏))
Γ(𝜁 ) i=0
(25)
1 ∑
m
=I(0) + bi,m+1 G3 (ti , Ih (ti ), Ih (ti−n )),
Γ(𝜁 ) i=0

where
h𝜁
bi,m+1 = ((m + 1 − i)𝜁 − (m − i)𝜁 ).
𝜁
So from the above calculations, the corrector formulas for all five equations of system (8) are

h𝜁 ∑
m
h𝜁
Sh (tm+1 ) = S(0) + G1 (tm+1 , Sh (tm+1 ), Sh (tm+1−n )) + ai,m+1 G1 (ti , Sh (ti ), Sh (ti−n )),
Γ(𝜁 + 2) Γ(𝜁 + 2) i=0

h𝜁 ∑
m
h𝜁
Eh (tm+1 ) = E(0) + G2 (tm+1 , Eh (tm+1 ), Eh (tm+1−n )) + ai,m+1 G2 (ti , Eh (ti ), Eh (ti−n )),
Γ(𝜁 + 2) Γ(𝜁 + 2) i=0

h𝜁 ∑
m
h𝜁
Ih (tm+1 ) = I(0) + G3 (tm+1 , Ih (tm+1 ), Ih (tm+1−n )) + ai,m+1 G3 (ti , Ih (ti ), Ih (ti−n )), (26)
Γ(𝜁 + 2) Γ(𝜁 + 2) i=0

h𝜁 ∑
m
h𝜁
Rh (tm+1 ) = R(0) + G4 (tm+1 , Rh (tm+1 ), Rh (tm+1−n )) + ai,m+1 G4 (ti , Rh (ti ), Rh (ti−n )),
Γ(𝜁 + 2) Γ(𝜁 + 2) i=0

h𝜁 ∑
m
h𝜁
Dh (tm+1 ) = D(0) + G5 (tm+1 , Dh (tm+1 ), Dh (tm+1−n )) + ai,m+1 G5 (ti , Dh (ti ), Dh (ti−n )),
Γ(𝜁 + 2) Γ(𝜁 + 2) i=0

where
{
m𝜁 +1 − (m − 𝜁 )(m + 1)𝜁 , i = 0
ai,m+1 = (m − i + 2)𝜁 +1 − 2(m − i + 1)𝜁 +1 + (m − i)𝜁 +1 , 1 ≤ i ≤ m,
1, i = m + 1
Similarly, the predictor terms are

1 ∑
m
ShP (tm+1 ) = S(0) + bi,m+1 G1 (ti , Sh (ti ), Sh (ti−n )),
Γ(𝜁 ) i=0

1 ∑
m
EhP (tm+1 ) = E(0) + bi,m+1 G2 (ti , Eh (ti ), Eh (ti−n )),
Γ(𝜁 ) i=0

1 ∑
m
IhP (tm+1 ) = I(0) + bi,m+1 G3 (ti , Ih (ti ), Ih (ti−n )), (27)
Γ(𝜁) i=0

1 ∑
m
RPh (tm+1 ) = R(0) + bi,m+1 G4 (ti , Rh (ti ), Rh (ti−n )),
Γ(𝜁 ) i=0

1 ∑
m
DPh (tm+1 ) = D(0) + bi,m+1 G5 (ti , Dh (ti ), Dh (ti−n )),
Γ(𝜁 ) i=0
10 KUMAR AND ERTURK

FIGURE 1 Nature of the given classes at time delay= 2 for different fractional order values

where
h𝜁
bi,m+1 = ((m + 1 − i)𝜁 − (m − i)𝜁 ).
𝜁

6 S I M UL ATION R E SU LT S

To perform numerical simulations, we use parameter values based on a case study of Wuhan, China, cited from literature
and with some assumptions, summarize in Table 1.
We also use the following initial conditions S(0) = 89,985,050, E(0) = 10,050, I(0) = 475, R(0) = 10 and D(0) = 2 for
321,000
Table 1. The function c(t) is taken as c(t) = 321,000+I(t) . Initial population N(0) is considered as 89,995,587, and birth rate is
b = d × N(0).
KUMAR AND ERTURK 11

FIGURE 2 Nature of the given classes for time delay= 6 at different fractional order values

Figure 1A exemplifies the behaviour of achieved results by given solution procedure for S(t) at different values of frac-
tional order 𝜁. In Figure 1B, we analyse the exposed individuals for different fractional order 𝜁 with respect to time t.
Figure 1C, D and E analyses the nature of infectious, recovered and deaths classes at 𝜁 = 0.85, 0.90, 0.98 and 1, respec-
tively. Figure 1A–E examines the nature of all given classes at time delay 𝜏 = 2. Similarly, the graphical simulations from
Figure 2A–E are for time delay 𝜏 = 6 for parameter values given in Table 1 . In Figure 2A, we show the nature of susceptible
individuals, in Figure 2B, exposed classes; in Figure 2C, infectious individuals; in Figure 2D, recovered and in Figure 2E,
the deaths, with respect to time variable t for parameter values used from Table 1 . In this same manner, we observed the
plots for time delay 𝜏 = 10 in Figure 3A–E. All graphs are computed using Mathematica software. For the all above sim-
ulations, we used time delay 𝜏 = 2, 6 and 10, respectively, and time step was 0.05. From the cited figures, we can observe
that the given model exceedingly depends on the order and gestures more degree of flexibility. Moreover, the fractional
12 KUMAR AND ERTURK

FIGURE 3 Nature of the given classes for time delay= 10 at different fractional order values

method gives more interesting results than the integer order model and permit to better examine the obtained results. We
observed that time delay variable 𝜏 plays an important role in this COVID-19 dynamics. The biological meaning of time
delay in COVID-19 is cleared from the above simulations.
In Figure 4A–C, we study the nature of basic reproductive number 0 at different fractional order values and for dif-
ferent time delay 𝜏. We can see that the maximum value of basic reproductive number for the given time range is near
about 1.038. Numerically, we have calculated that when c(t) tends to one, 0 tends to 1.032, and when c(t) tends to zero,
then 0 tends to 3.880. It is clear that the recovery rate and death rate are directly related to whether healthcare capac-
ity is exceeded or not. For the all above graphical simulations, we can observe that the given model works well to study
the dynamics of reproductive number respect to the time t. We have clearly observed that the parameter c plays a very
important role in the simulations and time delay can show the effect of healthcare opportunities.
KUMAR AND ERTURK 13

FIGURE 4 Nature of basic reproductive number for different time delay

7 CO NC LU SION S

In this paper, we solved a time delay fractional COVID-19 compartmental model via Caputo FDs applying
predictor–corrector algorithm. After preliminaries and the model description, the existence and uniqueness analysis of
the given time delay fractional system are established by the applications of fixed point theory (particularly, a mild Lips-
chitz condition using properties of a weighted norm, Mittag–Leffler functions and the Banach fixed point theorem). All
necessary graphical simulations are done to specified the nature of the achieved solutions in Caputo noninteger order
derivative sense. We analysed the role of time delay in the coronavirus epidemic by the graphical simulations for different
values. We also presented the plotting of the of basic reproductive number at different fractional order values and various
values of time delay. The projected scheme is strong and highly credible in finding the solution to delay fractional models
of biological, physical and medical importance.

CONFLICT OF INTEREST
This work does not have any conflicts of interest.

AU THORS CONTRIBUTION
Conceptualization, formal analysis, investigation, methodology, resources, visualization and writing—original draft:
Pushpendra Kumar. Conceptualization, investigation, software and writing—review & editing: Vedat Suat Erturk.

ORCID
Pushpendra Kumar https://orcid.org/0000-0002-7755-2837
14 KUMAR AND ERTURK

REFERENCES
1. Mission W-CJ. Report of the WHO-China joint mission on coronavirus disease 2019 (COVID-19). Geneva 2020; 2020.
2. Zhu N, Zhang D, Wang W, et al. A novel coronavirus from patients with pneumonia in China, 2019. New England J Med. 2020;382:727-733.
3. Sarkar K, Khajanchi S, Nieto JJ. Modeling and forecasting the COVID-19 pandemic in india. Chaos, Solitons & Fractals. 2020;139:110049.
4. Shaikh AS, Shaikh IN, Nisar KS. A mathematical model of COVID-19 using fractional derivative: outbreak in india with dynamics of
transmission and control; 2020.
5. Gao W, Veeresha P, Prakasha DG, Baskonus HM. Novel dynamic structures of 2019-nCoV with nonlocal operator via powerful
computational technique. Biology. 2020;9(5):107.
6. Gao W, Veeresha P, Baskonus HM, Prakasha DG, Kumar P. A new study of unreported cases of 2019-nCoV epidemic outbreaks. Chaos,
Solitons & Fractals. 2020;138:109929.
7. Khan MA, Atangana A. Modeling the dynamics of novel coronavirus (2019-nCoV) with fractional derivative. Alexandria Engr J.
2020;59(4):2379-2389.
8. Khoshnaw SHA, Shahzad M, Ali M, Sultan F. A quantitative and qualitative analysis of the COVID–19 pandemic model. Chaos, Solitons
& Fractals. 2020;138:109932.
9. Kuniya T. Prediction of the epidemic peak of coronavirus disease in Japan, 2020. J Clinical Med. 2020;9(3):789.
10. Nadim SS, Ghosh I, Chattopadhyay J. Short-term predictions and prevention strategies for COVID-2019: a model based study. arXiv
preprint arXiv:200308150; 2020.
11. Khan MA, Atangana A, Alzahrani E. The dynamics of COVID-19 with quarantined and isolation. Adv Differ Equa. 2020;2020(1):1-22.
12. Ullah S, Khan MA. Modeling the impact of non-pharmaceutical interventions on the dynamics of novel coronavirus with optimal control
analysis with a case study. Chaos, Solitons Fractals. 2020;139:110075.
13. Erturk VS, Kumar P. Solution of a COVID-19 model via new generalized Caputo-type fractional derivatives. Chaos, Solitons Fractals.
2020;139:110280.
14. Kilbas A. Theory and applications of fractional differential equations.
15. Miller KS, Ross B. An Introduction to the Fractional Calculus and Fractional Differential Equations. New York: Wiley; 1993.
16. Oldham K, Spanier J. The Fractional Calculus Theory and Applications of Differentiation and Integration to Arbitrary Order. New York:
Elsevier; 1974.
17. Podlubny I. Fractional Differential Equations: An Introduction to Fractional Derivatives, Fractional Differential Equations, to Methods of
Their Solution and Some of Their Applications. San Diego: Elsevier; 1998.
18. Rudolf H. Applications of Fractional Calculus in Physics. Singapore: world Scientific; 2000.
19. Çakan S. Dynamic analysis of a mathematical model with health care capacity for pandemic COVID-19. Chaos, Solitons & Fractals.
2020;139:110033.
20. Yang C, Wang J. A mathematical model for the novel coronavirus epidemic in Wuhan, China. Math Biosci Eng. 2020;17(3):2708-2724.
21. Van den Driessche P, Watmough J. Reproduction numbers and sub-threshold endemic equilibria for compartmental models of disease
transmission. Math Biosci. 2002;180(1-2):29-48.
22. Cong ND, Tuan HT. Existence, uniqueness, and exponential boundedness of global solutions to delay fractional differential equations.
Mediterranean J Math. 2017;14(5):193.
23. Tisdell CC. On the application of sequential and fixed-point methods to fractional differential equations of arbitrary order. J Integ Equa
Appl. 2012:283-319.
24. Bhalekar SACHIN, Daftardar-Gejji VARSHA. A predictor-corrector scheme for solving nonlinear delay differential equations of fractional
order. J Fract Calc Appl. 2011;1(5):1-9.

How to cite this article: Kumar P, Ertürk VS. The analysis of a time delay fractional COVID-19 model via
Caputo type fractional derivative. Math Meth Appl Sci. 2020;1–14. https://doi.org/10.1002/mma.6935