An efficient explicit approach for predicting the Covid-19 spreading

with undetected infectious: The case of Cameroon

Eric Ngondiep
1
Department of Mathematics and Statistics, College of Science, Al-Imam Muhammad Ibn Saud
Islamic University (IMIU), 90950 Riyadh 11632, Saudi Arabia.
arXiv:2005.11279v2 [q-bio.PE] 26 May 2020

2
Hydrological Research Centre, Institute for Geological and Mining Research, 4110 Yaounde-Cameroon.

Email addresses: [email protected]/[email protected]

Abstract. This paper considers an explicit numerical scheme for solving the mathematical model of the
propagation of Covid-19 epidemic with undetected infectious cases. We analyze the stability and conver-
gence rate of the new approach in L∞ -norm. The proposed method is less time consuming. Furthermore,
the method is stable, at least second-order convergent and can serve as a robust tool for the integration of
general systems of ordinary differential equations. A wide set of numerical evidences which consider the case
of Cameroon are presented and discussed.

Keywords: mathematical model of SARS-Cov-2 spreading, a two-level explicit scheme, stabil-

ity analysis, convergence rate, numerical experiments.

AMS Subject Classification (MSC). 65M10, 65M05.

1 Introduction and motivation

Deterministic models are important decision tools that can be useful to forecasting different scenarios. The
first motivation of studying such models is based on the use of the theory of ordinary/partial differential
equations and a low computational complexity which can permit a better calibration of the model char-
acteristics. Furthermore, deterministic approaches are the only suitable methods that can be used when
modeling a new problem with few data. For more details, we refer the readers to [2, 30, 11, 49] and ref-
erences therein. The use of the mathematical models as a predictive tool in the simulation of complex
problems arising in a broad range of practical applications in biology, environmental fluid mechanic, chem-
istry and applied mathematics (for example: mathematical model in population biology and epidemiology,
mixed Stokes-Darcy model, Navier-Stokes equations, nonlinear time-dependent reaction-diffusion problem,
heat conduction equation and unsteady convection-diffusion-reaction equations) represents a good candidate
for developing efficient numerical schemes in the approximate solutions of such problems [40, 34, 9, 10, 39, 38,
36, 14, 15, 32, 41, 28, 21, 43, 45, 53]. For parabolic partial differential equations (PDEs) which present strong
steep gradients (for instance: shallow water flow and advection-diffusion equations), numerical algorithms
are needed with good resolution of steep gradients [12, 8, 20, 22, 31, 44, 27, 29].

Early in an epidemic, the quality of the data on infections, deaths, tests and other factors often are
limited by undetection or inconsistent detention of cases, reporting delays, and poor documentation, all
of which affect the quality of any model output. Simpler models may provide less valid predictions since
they cannot capture complex and unobserved human mixing patterns and other time-varying parameters
of infectious disease spread. Also, complex models may be no more reliable than simpler ones if they miss
key aspects of the biological entities (either ions, molecules, proteins or ceils) [1]. At a time when numbers
of cases and deaths from coronavirus 2019 (Covid-19) pandemic continue to increase with alarming speed,

1
accurate forecasts from mathematical models are increasingly important for physicians, politicians, epidemi-
ologists, the public and most importantly, for authorities responsible of organizing care for the populations
they serve. Given the unpredictable behavior of severe acute respiratory syndrome Covid-19, it is worth
mentioning that efficient numerical approaches are the best tools that can be used to predict the spread of
the disease with reasonable accuracy. These predictions have crucial consequences regarding how quickly and
strongly the government of a country mores to curb a pandemic. However, assuming the worst-case scenario
at state and national levels will lead to inefficiencies (such as: the competition for beds and supplies) and
may compromise effective delivery and quality of care, whereas supposing the best-case scenario can conduct
to disastrous underpreparation.

Covid-19 is a rapidly spreading infectious disease caused by the novel coronavirus SARS-Cov-2, a be-
tacoronavirus which has provided a global epidemic. Up today, no drug to treat the Covid-19 disease is
officially available (approved by the World Health Organization (WHO)) and a vaccine will not be available
for several months at the earliest. The only approaches widely used to slow the spread of the pandemic are
those of classical epidemic control such as: physical distancing, contact tracing, hygiene measure, quarantine
and case of isolation. However, the primary and most effective use of the epidemiologic models is to estimate
the relative effect of various interventions in reducing disease burden rather than to produce precise quan-
titative predictions about extent on duration of disease burdens. Nevertheless, consumers of such models
including the media, the publics and politicians sometimes focus on the quantitative predictions of infectious
and mortality estimates. Such measures of potential disease burden are also necessary for planners who
consider future outcomes in light of health care capacity. The big challenge consists to assess such estimates.

In this paper, we develop an efficient numerical scheme for solving a mathematical model well adapted
to Covid-19 pandemic subjected to special characteristics (effect of undetected infected cases, effect of dif-
ferent sanitary and infectiousness conditions of hospitalized people and estimation of the needs of beds in
hospitals) and considering different scenarios [19]. Specifically, the proposed technique should provide the
numbers of detected infected and undetected infected cases, numbers of deaths and needs of beds in hospi-
tals in countries (for example, in Cameroon) where Covid-19 is a very serious health problem. It is Worth
noticing that the model of Covid-19 considered in this work has been obtained under the asumption of ”only
within-country disease spread” for territories with relevant number of people infected by SARS-Cov-2, where
local transmission is the major cause of the disease spread (for instance: case of Cameroon). Furthermore,
the parameters of the model used in this note are taken from the literature [24, 25, 19]. Our study also
relates the disease fatality rate with the percentage of detected cases over the real total infected cases which
allows to analyze the importance of this percentage on the impact of Covid-19. In addition, to demonstrate
the efficiency and validity of the new approach when applied to the mathematical problem of coronavirus
2019 epidemic, we consider the case of Cameroon, the country of the central Africa where one can observe
the highest number of people infected by the new virus SARS-Cov-2. We compare the results produced
by the numerical method to the data obtained from this country and those provided by the World Health
Organization in its reports [46]. Finally, it important to mention that the considered area (Cameroon) in
the numerical experiments can be replaced by any territory worldwide.

This paper is organized as follows: Section 2 considers some preliminaries together with the mathematical
formulation of Covid-19 spreading. In section 3, we provide a full description of the two-level explicit scheme
for solving the problem indicated in section 2. Section 4 analyzes the stability and the convergence rate
of the new procedure while a large set of numerical experiments are presented and critically discussed in
Section 5. We draw in section 6 the general conclusion and we provided our future investigations.

2 Preliminaries and mathematical model of SARS-Cov-2 infec-

tiousness
We use a mathematical formalism [19] that describes how infectiousness varies as a function of time since
infections for a representative cohort of infected persons. We assume that transmission of SARS-Cov-2 is
contagious from person to person and not point source. Furthermore, it is also assumed that, at the initial

2
phase of Covid-19 disease, the proportion of the population with immunity to SARS is negligible [3, 16, 4, 48].
At the beginning of a contagious epidemic, a small number of infected people start passing the disease to a
large population. Individuals can go through nine states. They start out susceptible (X1 : the person is not
infected by the disease pathogen), exposed (X2 : the person is in the incubation period after being infected
by the disease pathogen, but has no clinical signs), infected (X3 : the person has finished the incubation
period, may infected other people and start developing the clinical signs. Here, people can be taken in
charge by sanitary authorities of this country (hospitalized persons) or not detected by the authorities and
continue as infectious), infectious but undetected (X4 : the person can still infect other individuals, have
clinical signs, but is not detected and reported by the authorities (these people will not die)), hospitalized or
in quarantine at home (X5 : the person is in hospital or in quarantine at home, can still infect other people,
but will recover), hospitalized but will die (X6 : the person is hospitalized and can infect other people, but
will die), recovered after being previously infectious but undetected (X8 : the person was not previously
detected as infectious, survived the disease, is no longer infectious and has developed a natural immunity to
the disease), recovered after being previously detected as infectious (X7 : the person survived the disease, is
no longer infectious and has developed a natural immunity to the virus, but she/he remains in hospital for
a convalescence period d0 days), dead by SARS-Cov-2 (X9 ).

The proposed model is based on thirteen parameters.

• R0 denotes basis reproductive number, that is, the expected number of new infectious cases per infec-
tious case,
• N is the number of persons in a considered country before the starting of the pandemic,
• µ1 ∈ [0, 1], designates the natality rate (day −1 ) in the considered country (the number of births per
day and per capita),
• µ2 ∈ [0, 1], represents the mortality rate (day −1 ) in the considered country (the number of deaths per
day and per capita),
• w(t) ∈ [w, w] ⊂ [0, 1], denotes the case fatality rate in the considered territory at time t (the proportion
of deaths compared to the total number of infectious people (detected or undetected). Here, w and w
are the minimum and maximun case fatality rates in the country, respectively),
• θ(t) ∈ [w, 1], means the fraction of infected people that are detected and reported by the authorities in
the country at time t. For the convenience of writting, we assume that all the deaths due to Covid-19
are detected and reported, so θ(t) ≥ w,
• βXj ∈ R+ , for j = 2, 3, · · · , 6, are the disease contact rates (day −1 ) of a person in the corresponding
compartment Xj , in the country (without taking into account the control measures),
• βX4 (θ) ∈ R+ , represents the disease contact rates (day −1 ) of a person in compartment X4 , in the
country (without taking into account the control measures), where the fraction of infected individuals
that are detected is θ(t),
• γX2 ∈ (0, ∞), designates the transition rate (day −1 ) from compartment X2 to compartment X3 . It’s
the same in all the countries,
• γX3 (t) ∈ (0, ∞), is the transition rate (day −1 ) from compartment X3 to compartments X4 , X5 or X6
at time t. It can change from a country to another,
• γX4 (t), γX5 (t) and γX6 (t) ∈ (0, ∞), denote the transition rate (day −1 ) from compartments X4 , X5 or
X6 to compartments X7 , X8 and X9 , respectively, in the considered country at time t,
• mXj (t) ∈ [0, 1], for j = 2, 3, · · · , 6, are functions representing the efficiency of the control measures
applied to the corresponding compartment in the considered country at time t,
• τ1 is the person infected that arrives in the territory from other countries per day. τ2 is the person
infected that leaves the territory from other countries per day. Both can be modeled following the
between-country spread part of the Be-CoDis model, see [18].

3
The control measures applied by the government to curb the Covid-19 spread are those provided by the
WHO in [7, 47]:
• isolation: infected people are isolated from contact with other persons. Only sanitary professionals
are in contact with them. Isolated patients receive an adequate medical treatment that reduces the
Covid-19 fatality rate,
• quarantine: movement of people in the area of origin of an infected person is restricted and controlled
(for instance: quick sanitary check-points at the airports) to avoid that possible infected people spread
the disease,

• tracing: the aim of tracing is to identify potential infectious contacts which may have infected an
individual or spread SARS-Cov-2 to other people. Increase the number of tests in order to increase
the percentage of detected infected persons,
• increase of sanitary resources: number of operational beds and sanitary personal available to detect and
treat affected people is increased, producing a decrease in the infectious period for the compartment
X3 .
Furthermore, the mathematical model of coronavirus 2019 epidemic considers the following assumptions:
(a1 ) the population at risk is large enough and time period of concern is short enough that over the time
period of interest, very close to 100% of the population is susceptible,
(a2 ) the pandemic is at the early stage and has not reached the point where the susceptible population
decreases so much due to death or post-infection immunity that the average number of secondary cases
falls,
(a3 ) unprotected contact results in infection,
(a4 ) the epidemic in the population of interest begins with a single host (note that the equations used in
computing cases and deaths are easily modified if this is not the case),
(a5 ) infectivity occurs during the incubation period only,
(a6 ) the models are deterministic, that is, the thirteen parameters of Covid-19 spread cited above are
constant values.

Under these assumptions, the mathematical formulation of Covid-19 disease is given by the following system
of nonlinear ordinary differential equations:
dX1 X1
=− [mX2 (t)βX2 (t)X2 + mX3 (t)βX3 (t)X3 + mX4 (t)βX4 (θ)X2 + mX5 (t)βX5 (t)X5 + mX6 (t)βX6 (t)X6 ]
dt N
− µm X1 + µn [X1 + X2 + X3 + X4 + X7 + X8 ], (1)
dX2 X1
= [mX2 (t)βX2 (t)X2 + mX3 (t)βX3 (t)X3 + mX4 (t)βX4 (θ)X2 + mX5 (t)βX5 (t)X5 + mX6 (t)βX6 (t)X6 ]
dt N
− µm X2 − γX2 (t)X2 + τ1 (t) − τ2 (t), (2)
dX3 dX4
= γX2 (t)X2 − (µm + γX3 (t)) X3 , = (1 − θ(t))γX3 (t)X3 − (µm + γX4 (t)) X4 , (3)
dt dt
dX5 dX6
= (θ(t) − w(t))γX3 (t)X3 − γX5 (t)X5 , = w(t)γX3 (t)X3 − γX6 (t)X6 , (4)
dt dt
dX7 dX8 dX9
= γX5 (t)X5 − µm X7 , = γX4 (t)X4 − µm X8 and = γX6 (t)X6 , (5)
dt dt dt
with the initial conditions
Xj (t0 ) = Xj0 ∈ (0, ∞), for j = 1, 2, · · · , 9, (6)

4
where all the unknowns Xj dependent on the time t ∈ [t0 , Tmax ]. Setting X(t) = (X1 , X2 , . . . , X9 )T and
F (t, X(t)) = (F1 (t, X(t)), F2 (t, X(t)), . . . , F9 (t, X(t)))T , where

X1
F1 (t, X(t)) = − [mX2 (t)βX2 (t)X2 + mX3 (t)βX3 (t)X3 + mX4 (t)βX4 (θ)X2 + mX5 (t)βX5 (t)X5
N
+ mX6 (t)βX6 (t)X6 ] − µm X1 + µn [X1 + X2 + X3 + X4 + X7 + X8 ], (7)
X1
F2 (t, X(t)) = [mX2 (t)βX2 (t)X2 + mX3 (t)βX3 (t)X3 + mX4 (t)βX4 (θ)X2 + mX5 (t)βX5 (t)X5
N
+ mX6 (t)βX6 (t)X6 ] − µm X2 − γX2 (t)X2 + τ1 (t) − τ2 (t), (8)
F3 (t, X(t)) = γX2 (t)X2 − (µm + γX3 (t)) X3 , F4 (t, X(t)) = (1 − θ(t))γX3 (t)X3 − (µm + γX4 (t)) X4 , (9)
F5 (t, X(t)) = (θ(t) − w(t))γX3 (t)X3 − γX5 (t)X5 , F6 (t, X(t)) = w(t)γX3 (t)X3 − γX6 (t)X6 , (10)
F7 (t, X(t)) = γX5 (t)X5 − µm X7 , F8 (t, X(t)) = γX4 (t)X4 − µm X8 and F9 (t, X(t)) = γX6 (t)X6 , (11)
the system of nonlinear equations (1)-(5) is equivalent to

dX
= F (t, X). (12)
dt

Remark. In the modeling point of view, the term w(t) θ(t) corresponds to the apparent fatality rate of the
disease (obtained by considering only the detected cases) in the considered area at time t, whereas
w(t) is the real fatality rate of coronavirus 2019 disease.
Since the mathematical model of Covid-19 provided by the system of equations (1)-(5) is too complex
and because both natality and mortality (not from SARS-Cov-2) do not seem to be useful factors for this
pandemic (at least for relatively short periods of time), we assume in the rest of this paper that

µm = µn = 0. (13)

It is worth mentioning that the aim of this paper is to compute the following Covid-19 characteristics:
1) the model cumulative of coronavirus 2019 cases at day t given by

cm (t) = X5 (t) + X6 (t) + X8 (t) + X9 (t), (14)

2) the model cumulative number of deaths (due to Covid-19) at day t, which is given by X9 (t),
3) Re (t) which is the effective reproductive number of Covid-19,
4) the number of people in hospital is estimated by the following equation

Host(t) = X6 (t) + p(t)[X5 (t) + (X7 (t) − X7 (t − d0 ))], (15)

where p(t) represents the fraction, at time t, of people in compartment X5 that are hospitalized and
d0 days is the period of convalescence (i.e., the time a person is still hospitalized after recovering from
Covid-19). This function can help to estimate and plan the number of clinical beds needed to treat all
the SARS-Cov-2 cases at time t,
5) the maximum number of hospitalized persons at the same time in the territory during the time interval
[t0 , Tmax ], which is defined as
M axHost = max Host(t). (16)
t0 ≤t≤Tmax

M axHost can help to estimate and plan the number of clinical beds needed to treat all the coronavirus
2019 cases over the interval [t0 , Tmax ],

5
6) the number of people infected during the time interval [t0 , Tmax ], by contact with people in compartments
X2 , X4 and X10 = X5 + X6 , respectively. They are defined as
Z
1 Tmax
ΓX2 (t) = mX2 (s)βX2 (s)X2 (s)X1 (s)ds, (17)
N t0
Z Tmax
1
ΓX4 (t) = mX4 (s)βX4 (s)X4 (s)X1 (s)ds, (18)
N t0
Z Tmax
1
ΓX10 (t) = (mX5 (s)βX5 (s)X5 (s) + mX6 (s)βX6 (s)X6 (s)) X1 (s)ds. (19)
N t0

We recall that the basis reproduction number R0 is defined as the number of cases an infected individual
generates on average over the course of its infectious period, in an otherwise uninfected population and
without special control measures. It depends on the considered population, but does change during the
spread of the disease, while the effective reproduction number Re (t) is defined as the number of cases one
infected person generates on average over the course of its infectious period. A part of the population
can be already infected and/or special control measures that have been implemented. It depends on the
spread of the disease. In addition, Re (t0 ) = R0 , and the evolution of the epidemic slow down when Re (t) < 1.

Now, applying the next generation method [50] to the nonlinear system (1)-(5) to get

R0 = {γX6 [((βX4 (1 − θ)γX5 + βX5 γX4 (θ − w))γX3 + βX3 γX4 γX5 )γX2 + βX2 γX3 γX4 γX5 ]

+wβX6 γX2 γX3 γX4 γX5 } (γX2 γX3 γX4 γX5 γX6 )−1 , (20)
and
X1 (t)
Re (t) = {γX6 [((mX4 βX4 (1 − θ)γX5 + mX5 βX5 γX4 (θ − w))γX3 + mX3 βX3 γX4 γX5 )γX2
N
+mX2 βX2 γX3 γX4 γX5 ] + wmX6 βX6 γX2 γX3 γX4 γX5 } (γX2 γX3 γX4 γX5 γX6 )−1 , (21)
where for the sake of simplicity of notations, all previous coefficients correspond to their particular values at
times t0 and t, respectively.

In the literature [26, 52], it is established that the observed patterns of Covid-19 are not completely
consistent with the hypothesis that high absolute humidity may limit the survival and transmission of the
virus, whereas the lower is the temperature, the greater is the survival period of the SARS-Cov-2 outside
the host. Since there is no scientific evidence of the effect of the humidity and the temperature on the
SARS-Cov-2, these factors are not included in our model.

Focusing on the application on the control strategies, the efficiency of these measures indicated in [23]
satisfies equations

 (ml − ml+1 ) exp[−kl+1 (t − λl )], t ∈ [tl , λl+1 ), l = 0, 1, · · · , q − 1,
mXj (t) := m(t) = (22)

(mq−1 − mq ) exp[−kq (t − λq−1 )], t ∈ [tq−1 , ∞),

for j = 1, 2, · · · , 6, where ml ∈ [0, 1] measures the intensity of the control measures (greater value implies
lower value of disease contact rates), kl ∈ [0, ∞) (in day −1 ) simulates the efficiency of the control strategies
(greater value implies lower value of disease contact rates) and λl ∈ [t0 , ∞), l = 1, 2, . . . , q, denotes the first
day of application of each control strategy. λ0 ∈ [t0 , ∞) is the first day of application of a control measure
that was being used before t0 , if any. In this work, q ∈ N represents the number of changes of control
strategy. In general, the values of λl are typically taken in the literature (using dates when the countries
implement special control measures). It is important to remind that some of the values of ml can be also
sometimes known. The rest of the parameters needed to be calibrated.

6
 In the following, we assume that the case fatality rate w(t), depends on the considered country, time t,
and it can be affected by the application of the control measures (such as, earlier detection, better sanitary
condition, etc...)[18]. Thus, it satisfies equation

w(t) = m(t)w + (1 − m(t))w, (23)

where w ∈ [0, 1] is the case fatality rate when no control measures are applied (i.e., m(t) = 1) and w ∈ [0, 1]
is the case fatality rate when implemented control measures are fully applied (m(t) = mq ).

Denoting by dX3 , dX4 , dX5 and dX6 , be the ”average” duration in days of a person in compartment X3 ,
X4 , X5 and X6 , respectively, without the application of control strategies, we assume as in [24, 25] that
• the transition rate from X2 to X3 depends on the disease and, therefore is considered constant, that
is γ2 (t) = α = ct ,
• the value of γX3 (t) := γ(t) can be increased due to the application of control measures (that is, people
with symptoms are detected earlier). As a consequence, the values of γX4 (t), γX5 (t) and γX6 (t) can
be decreased (i.e., persons with symptoms stay under observation during more time),
• dX6 = dX5 + δ, δ > 0,
• for the sake of readability, the infectious period for undetected individuals is the same than that
of hospitalized people that survive the disease. So dX4 = dX5 . Furthermore, we suppose that the
additional time a person is in the compartments X3 and X4 is constant, so it comes from [18] that
1
γ(t) := γX3 (t) = , (day −1 ) (24)
dX3 − g(t)
1
ρ(t) := γX4 (t) = γX5 (t) = , (day −1 ) (25)
dX4 + g(t)
1
ψ(t) := γX6 (t) = , (day −1 ) (26)
dX4 + g(t) + δ
where g(t) = dg (1 − m(t)) represents the decrease of the duration of the function dX3 due to the
application of the control measures at time t, dg is the maximum number of days that dX3 can be
decreased due to the control measures.

Finally, the disease contact rate βX4 (θ) is defined by



 β , if θ = w,
 X3


βX4 (θ) = nonincrease, (27)




 β , if θ = w,
X 3

where β X and β X3 are suitable lower and upper bounds, respectively. For the convenience of writing, we
3
assume that β X3 = βX3 := β. In addition, the people in compartments X2 , X4 , X5 and X6 are less infectious
than people in compartment X3 (due to their lower virus load or isolation measures). This fact results in

βX2 = cX2 βX3 , βX5 = βX6 = cX10 (t)βX3 , β X = cu βX3 , (28)

3

where cX2 , cX10 , cu ∈ [0, 1].

7
3 Construction of the two-level explicit numerical scheme
In this section, we develop the robust two-level explicit scheme for solving the mathematical problem (1)-(5)
modeling the spread of Covid-19 with undetected cases.

Let h := ∆t = Tmax M
−t0
be the step size, M is a positive integer. Set tn = t0 + nh, tn− 21 = tn +t2n−1
for n = 1, 2, ..., M and Ωh = {tn , 0 ≤ n ≤ M } be a regular partition of [t0 , Tmax ]. Let Fh = {Xin , n =
0, 1, ..., M ; 1 ≤ i ≤ 9}, be the grid functions space defined on Ωh × R9 ⊂ I × R9 := [t0 , Tmax ] × R9 .

Define the following norms

M
! 12
X
n
kX k∞ = max |Xin | and k|X|kL2 (I) = h kX n k2∞ (29)
1≤i≤9
n=0

where | · | designates the norm defined on the field of complex numbers C. Furthermore, denote
(i)
X
Pj (t, X(t)) = Fi (tl , X(tl ))Ll (t), (30)
l

where the function Ll (t) is given by

Y t − tq
Ll (t) = , (31)
q
tl − tq
q6=l

be a polynomial of degree j interpolating the function Fi (t, X(t)) at the node points (tl , Fi (tl , X(tl ))). Ac-
(i)
cording to equations (30)-(31), it’s important to remind that Pj (t, X(t)) is not necessarily the interpolation
polynomial of degree j of the function Fi (t, X(t)) at the node points (tl , Fi (tl , X(tl ))).

Now, integrating both sides of equation (12) at the node points tn and tn+ 12 , this yields
Z tn+ 1
2
X(tn+ 12 ) − X(tn ) = F (t, X)dt,
tn

which is equivalent to Z tn+ 1

2
X(tn+ 12 ) = X(tn ) + F (t, X)dt. (32)
tn
(i)
For j = 1, P1 (t, X(t)) is a linear polynomial approximating the function Fi (t, X(t)) at the points (tn , Fi (tn , X(tn )))
and (tn+ 12 , Fi (tn+ 12 , X(tn+ 21 ))). Using equations (30) and (31), it is easy to observe that

t − tn+ 12   t − tn 2 h  
(i)
P1 (t, X(t)) = Fi (tn , X(tn )) + Fi tn+ 12 , X(tn+ 21 ) = Fi tn+ 12 , X(tn+ 21 ) −
tn − tn+ 21 tn+ 21 − tn h
 i
Fi (tn , X(tn ))) t + tn+ 21 Fi (tn , X(tn )) − tn Fi tn+ 12 , X(tn+ 21 ) , (33)

where the error term is given by

(i) 1 d2 Fi
Fi (t, X(t)) − P1 (t, X(t)) = (t − tn )(t − tn+ 12 ) 2 (tǫ , X(tǫ )) := O(h2 ), (34)
2 dt
where tǫ (respectively, each component Xi (tǫ ) of X(tǫ )) is an unknown function which is between the max-
imum and the minimum of the numbers tn , tn+ 12 and t (respectively: Xi (tn ), Xi (tn+ 21 ) and Xi (t)). But
equation (34) can be rewritten as
(i)
Fi (t, X(t)) = P1 (t, X(t)) + O(h2 ), for i = 1, 2, ..., 9. (35)

8
Substituting approximation (35) into the ith equation of the system (32), we obtain
Z t 1
n+
(i)
P1 (t, X)dt + O(h3 ), for i = 1, 2, ..., 9,
2
Xi (tn+ 12 ) = Xi (tn ) + (36)
tn

which is equivalent to the following system

Z tn+ 1
Q1 (t, X)dt + O(h3 ),
2
X(tn+ 21 ) = X(tn ) + (37)
tn
 T
(1) (2) (9)
where Q1 (t, X(t)) = P1 (t, X(t)), P1 (t, X(t)), ..., P1 (t, X(t)) and O(h3 ) = (O(h3 ), O(h3 ), ..., O(h3 ))T .

The integration of both sides of equation (33) provides

Z t 1
n+
(i) 1 hh   i h
Fi tn+ 12 , X(tn+ 21 ) − Fi (tn , X(tn )) (t2n+ 1 − t2n ) + 2 tn+ 21 Fi (tn , X(tn ))−
2
P1 (t, X)dt =
tn h 2

 i i h   i h
tn Fi tn+ 12 , X(tn+ 21 ) (tn+ 12 − tn ) = Fi tn+ 21 , X(tn+ 21 ) − Fi (tn , X(tn )) (tn + )
4
  hh   i
+ tn+ 12 Fi (tn , X(tn )) − tn Fi tn+ 21 , X(tn+ 12 ) = Fi tn+ 21 , X(tn+ 21 ) + Fi (tn , X(tn )) , (38)
4
where the last two equalities come from the identities t2n+ 1 − t2n = (tn+ 12 − tn )(tn+ 21 + tn ), tn+ 12 − tn = h2
2
and tn+ 21 + tn = 2tn + h2 . To get the desired first-level of the new algorithm, we should approximate the sum
 
Fi tn+ 12 , X(tn+ 21 ) + Fi (tn , X(tn )) by the term a1 Fi (tn , X(tn )) + a2 Fi (tn + p1 h, X(tn ) + p2 hF (tn , X(tn ))),
in which the coefficients a1 , a2 , p1 and p2 , are real numbers and are chosen so that the Taylor expansion
Xi (tn+ 21 ) − Xi (tn ) 1h   i
− Fi tn+ 12 , X(tn+ 21 ) + Fi (tn , X(tn )) = O(h2 ).
h/2 2
The application of the Taylor series expansion for Xi and Fi about tn and (tn , X(tn )), respectively, with
step size h/2 using forward difference representations gives
" 9
#
h h2 X
Xi (tn+ 12 ) = Xi (tn ) + Fi (tn , X(tn )) + ∂t Fi (tn , X(tn )) + Fk (tn , X(tn ))∂k Fi (tn , X(tn )) + O(h3 ),
2 8
k=1
(39)
and
Fi (tn + p1 h, X(tn ) + p2 hF (tn , X(tn ))) = Fi (tn , X(tn )) + hp1 ∂t Fi (tn , X(tn ))+
9
X
hp2 Fk (tn , X(tn ))∂k Fi (tn , X(tn )) + O(h2 ), (40)
k=1
∂Fi ∂Fi
where ∂t Fi denotes ∂t and ∂k Fi represent ∂Xk , for k = 1, 2, ..., 9.

Combining equations (39)-(40), direct calculations yield

1h 
Xi (tn+ 12 ) − Xi (tn )  i  a1 + a2

− Fi tn+ 12 , X(tn+ 12 ) + Fi (tn , X(tn )) = 1 − Fi (tn , X(tn ))+
h/2 2 2
"   X9
#
h 1 1
− a2 p1 ∂t Fi (tn , X(tn )) + − a2 p 2 Fk (tn , X(tn ))∂k Fi (tn , X(tn )) + O(h2 ),
2 2 2
k=1
1
2
which equals O(h ) if and only if a1 + a2 = 2, a2 p1 = 2 and a2 p2 = 21 . But the last two equations require
a2 6= 0, p1 6= 0 and p2 6= 0. Take for instance
1
p1 = p2 = , so a1 = a2 = 1. (41)
2

9
  
Plugging equation (40) and relation (41), it is not hard to observe that the sum Fi tn+ 21 , X(tn+ 21 ) +
Fi (tn , X(tn )) is approximate as
  h
Fi tn+ 21 , X(tn+ 12 ) + Fi (tn , X(tn )) = 2Fi (tn , X(tn )) + [∂t Fi (tn , X(tn ))+
2
9
#
X
Fk (tn , X(tn ))∂k Fi (tn , X(tn )) + O(h2 ). (42)
k=1

A combination of equations (38) and (42) results in

Z t 1 " 9
#
n+
(i) h h2 X
Fk (tn , X(tn ))∂k Fi (tn , X(tn )) + O(h3 ).
2
P1 (t, X)dt = Fi (tn , X(tn )) + ∂t Fi (tn , X(tn )) +
tn 2 8
k=1
(43)
Substituting equation (43) into (36), this provides
" 9
#
h h2 X
Xi (tn+ 21 ) = Xi (tn ) + Fi (tn , X(tn )) + ∂t Fi (tn , X(tn )) + Fk (tn , X(tn ))∂k Fi (tn , X(tn )) + O(h3 ).
2 8
k=1
(44)
Tracking the infinitesimal term O(h3 ), equation (44) can be approximate as
" 9
#
n+ 12 h h 2 X
Xi = Xin + Fi (tn , X n ) + ∂t Fi (tn , X n ) + Fk (tn , X n )∂k Fi (tn , X n ) , for i = 1, 2, ..., 9. (45)
2 8
k=1

The difference equations provided by relation (45) represent the first-level of the new approach.

To develop the second-level of the desired algorithm, we should integrate both sides of system (12) at
the node points tn+ 21 and tn+1 . Thus
Z tn+1
X(tn+1 ) − X(tn+ 21 ) = F (t, X)dt,
tn+ 1
2

which can be rewritten as Z tn+1

X(tn+1 ) = X(tn+ 21 ) + F (t, X)dt.
tn+ 1
2

This implies Z tn+1

Xi (tn+1 ) = Xi (tn+ 21 ) + Fi (t, X)dt, for i = 1, 2, ..., 9. (46)
tn+ 1
2

(i)
Replacing Fi (t, X) by the linear interpolating polynomial P1 (t, X) at the node points (tn , Fi (tn , X n )) and
(tn+ 21 , Fi (tn+ 21 , X n )), and using (34), equation (46) is approximate as
Z tn+1
(i)
Xi (tn+1 ) = Xi (tn+ 12 ) + P1 (t, X)dt + O(h3 ). (47)
tn+ 1
2

Plugging (47) and (33), simple computations give


2 1h   i
Xi (tn+1 ) = Xi (tn+ 12 ) + Fi tn+ 12 , X(tn+ 21 ) − Fi (tn , X(tn )) (t2n+1 − t2n+ 1 ) + [tn+1 Fi (tn , X(tn ))−
h 2 2

 i i
tn+ 21 Fi tn+ 21 , X(tn+ 12 ) (tn+1 − tn+ 21 ) + O(h3 ) = Xi (tn+ 12 )+
hh   i
3Fi tn+ 12 , X(tn+ 21 ) − Fi (tn , X(tn )) + O(h3 ). (48)
2

10
Omitting the error term O(h3 ), we obtain the second-level of the new method which is defined as

n+ 12 hh  1
 i
Xin+1 = Xi + 3Fi tn+ 12 , X n+ 2 − Fi (tn , X n ) , for i = 1, 2, ..., 9. (49)
2
Under the assumption given by equations (13), (22),(23),(25) and (28), the functions Fi (t, X), (i = 1, 2, ..., 9,)
defined by relations (7)-(11) becomes

m(t)X1
F1 (t, X(t)) = − [βX3 (t)X2 + βX3 (t)X3 + βX4 (θ)X2 + CX5 (t)βX3 (t)X5 + CX5 (t)βX3 (t)X6 ] , (50)
N
m(t)X1
F2 (t, X(t)) = [βX3 (t)X2 + βX3 (t)X3 + βX4 (θ)X2 + CX5 (t)βX3 (t)X5 + CX5 (t)βX3 (t)X6 ] − αX2 ,
N
(51)
F3 (t, X(t)) = αX2 − γ(t)X3 , F4 (t, X(t)) = (1 − θ(t))γ(t)X3 − ρ(t)X4 , (52)
F5 (t, X(t)) = (θ(t) − w(t))γ(t)(t)X3 − ρ(t)X5 , F6 (t, X(t)) = w(t)γ(t)X3 − ψ(t)X6 , (53)
F7 (t, X(t)) = ρ(t)X5 , F8 (t, X(t)) = ρ(t)X4 and F9 (t, X(t)) = ψ(t)X6 , (54)
By straightforward computations, it is not hard to observe that
−1
∂t F1 = {CX2 βX2 [(m(t)F1 + ṁ(t)X1 )X2 + m(t)X1 F2 ] + β [(m(t)F1 + ṁ(t)X1 )X3 + m(t)X1 F3 ]
N
h i
+ m(t)βX4 (θ)F1 + (ṁ(t)βX4 (θ) + m(t)β̇X4 (θ))X1 X4 + m(t)βX4 (θ)X1 F4 + β [(m(t)cX10 (t)F1 +

(ṁ(t)cX10 (t) + m(t)ċX10 (t)) X1 ) (X5 + X6 ) + m(t)cX10 (t)X1 (F5 + F6 )]} , (55)
∂t F2 = −∂t F1 − γ̇(t)X2 − γ(t)F2 , ∂t F3 = αF2 − γ̇(t)X3 − γ(t)F3 , (56)
h i
∂t F4 = (1 − θ(t))γ̇(t) − θ̇(t)γ(t) X3 + (1 − θ(t))γ(t)F3 − ρ̇(t)X4 − ρ(t)F4 , (57)
h i
∂t F5 = (θ̇(t) − ẇ(t))γ(t) + (θ(t) − w(t))γ̇(t) X3 + (θ(t) − w(t))γ(t)F3 − ρ̇(t)X5 − ρ(t)F5 , (58)

∂t F6 = (ẇ(t)γ(t) + w(t)γ̇(t)) X3 + w(t)γ(t)F3 − ψ̇(t)X6 − ψ(t)F6 , (59)

∂t F7 = ρ̇(t)X5 + ρ(t)F5 , ∂t F8 = ρ̇(t)X4 + ρ(t)F4 , ∂t F9 = ψ̇(t)X6 + ψ(t)F6 , (60)
−m(t)
∂1 F1 = {CX2 βX2 + βX3 + βX4 (θ)X4 + CX10 (t)β(X5 + X6 )} , (61)
N
−m(t) −m(t) −m(t)
∂2 F1 = CX2 βX1 , ∂3 F1 = βX1 , ∂4 F1 = βX4 (θ)X1 , (62)
N N N
−m(t)
∂5 F1 = ∂6 F1 = CX10 βX1 , ∂k F1 = 0, for k = 7, 8, 9, (63)
N
∂1 F2 = −∂1 F1 , ∂2 F2 = −∂2 F1 − γ(t), ∂k F2 = 0, for k = 3, 4, ..., 9, (64)
∂1 F3 = 0, ∂2 F3 = α, ∂3 F3 = −γ(t), ∂k F3 = 0, for k = 4, 5, ..., 9, (65)
∂3 F3 = (1 − θ(t))γ(t), ∂4 F4 = −ρ(t), ∂k F4 = 0, for k 6= 3, 4, (66)
∂3 F5 = (θ(t) − w(t))γ(t), ∂5 F5 = −ρ(t), ∂k F5 = 0, for k 6= 3, 5, (67)
∂3 F6 = w(t)γ(t), ∂6 F6 = −ψ(t), ∂k F6 = 0, for k 6= 3, 6, ∂5 F7 = ρ(t), ∂k F7 = 0, for k 6= 5, (68)
∂4 F8 = ρ(t), ∂k F8 = 0, for k 6= 4, ∂6 F9 = ψ(t), ∂k F9 = 0, for k 6= 6. (69)
To provide a full description of the two-level explicit formulation for solving the mathematical problem (1)-
(5) to predicting the spread of Covid-19 model, we should put together relations (45), (49) and the initial
condition given by equation (6). Specifically, for n = 1, 2, ..., M − 1,
" 9
#
n+ 21 h h2 X
n n n n n
Xi = Xi + Fi (tn , X ) + ∂t Fi (tn , X ) + Fk (tn , X )∂k Fi (tn , X ) , for i = 1, 2, ..., 9; (70)
2 8
k=1

11
 n+ 12 hh  1
 i
Xin+1 = Xi + 3Fi tn+ 12 , X n+ 2 − Fi (tn , X n ) , for i = 1, 2, ..., 9; (71)
2
subject to the initial condition
Xi (t0 ) = Xi0 , for i = 1, 2, ..., 9; (72)
where the functions Fi (i = 1, 2, ..., 9) and its partial derivatives are given by equations (50)-(69).

Using the tools provided in section 3 we are ready to analyze the stability and the convergence rate
of the two-level explicit procedure (70)-(72) for predicting the spread of SARS-Cov-2 epidemic modeled by
equations (1)-(6).

4 Stability analysis and convergence rate of the new algorithm

In this section we wish to examine the stability and convergence rate of the new technique (70)-(72) applied
to the initial-value problem (1)-(6).

Firstly, we define the strip S = {(t, X), t0 ≤ t ≤ Tmax , X ∈ R9 } in which both exact and computed
solutions of problem (1)-(6) should lie. It comes from equations (50)-(69) that the functions Fi and their
partial derivatives are continuous on the strip S, but the partial derivatives are unbounded on this set. Thus
it follows from the Henrici result [17] that the system of equations (1)-(6) admits a unique solution X(t)
defined in a certain neighborhood U (t0 ) ⊂ [t0 , Tmax ] of the initial point t0 . Without loss of this constraint,
we assume in the following that U (t0 ) = [t0 , Tmax ] (indeed, we are dealing with a real world problem which,
in reality should have a unique solution defined over the interval [t0 , Tmax ]). This shows the existence and
uniqueness of the solution of the initial-value problem (1)-(6).

Let now introduce the functions ∆i (tl , Xi (tl )) and δi (tl , Xi (tl )) (for l = n, n+ 12 ) be the difference quotient
of the exact solution Xi (tl ) of equations (1)-(6) at time tl and the difference quotient of the approximate
solution Xil of problem (70)-(72) obtained at time tl , respectively. Moreover, ∆i (tl , Xi (tl )) and δi (tl , Xi (tl ))
are given by 
 X (t + 12 )−Xi (tl )
 i l h/2 , if h 6= 0,
∆i (tl , Xi (tl )) = (73)

 F (t , X (t )),
i l i l for h = 0,
and  l+ 1

 Xi 2 −Xil
h/2 , if h 6= 0,
δi (tl , Xi (tl )) = (74)


Fi (tl , Xil ), for h = 0.
It is worth mentioning that δi (tl , Xi (tl )) = δi (tl , Xil ). The local discretization error at the point (tl , Xi (tl ))
of the considered scheme is defined as
1
σi (tl , Xi (tl )) = ∆i (tl , Xi (tl )) − δi (tl , Xi (tl )), l = n or n + , for i = 1, 2, ..., 9, (75)
2
indicates how well the exact solution of the differential equations (1)-(6) obeys the difference equations (70)-
(72) provided by the two-level explicit formulation.

The following result (Theorem 4.1) analyzes the stability and gives the convergence rate of the proposed
approach (70)-(72).
Theorem 4.1. (Stability analysis and convergence rate)
Let en = X(tn ) − X n be the global discretization error provided by algorithm (70)-(72), where X(tn ) is the
solution of system (1)-(6) obtained at time tn and X n is the one provided by (70)-(72) at time tn . Thus, it
holds
M
! 12
X
n n n 2
kX k∞ = max |Xi | ≤ C1 , which implies k|X|kL2(I) = h kX k∞ ≤ C1
1≤1≤9
n=0

12
where C1 is a positive constant independent of the step size h and X denotes the approximate solution.
Furthermore
M
! 21
X
ken k∞ = max |eni | ≤ C2 h2 , which implies k|e|kL2(I) = h ken k2∞ ≤ C2 h2
1≤1≤9
n=0

where C2 is a positive constant that does not depend on the step size h. In the following we represent the
analytical solution by X(·).

The proof of Theorem 4.1 requires the following intermediate result (namely Lemmas 4.1).
Lemma 4.1. Suppose the vectors qj ∈ Cl satisfy the estimates of the form

kqj+1 kL∞ (Cl ) ≤ (1 + ǫ)kqj kL∞ (Cl ) + ξ, ǫ > 0 and ξ > 0, (76)

for j = 0, 1, ..., m. Then

exp(mǫ) − 1
kqm kL∞ (Cl ) ≤ exp(mǫ)kq0 kL∞ (Cl ) + ξ. (77)
ǫ
exp(ǫ)−1
Proof. We should prove inequality (77) by mathematical induction. Since exp(ǫ) ≥ 1 + ǫ and 1 ≤ ǫ ,
for any ǫ ≥ 0, using the assumption of Lemma 4.1, it is not difficult to see that
exp(ǫ) − 1
kq1 kL∞ (Cl ) ≤ (1 + ǫ)kq0 kL∞ (Cl ) + ξ ≤ exp(ǫ)kq0 kL∞ (Cl ) + ξ.
ǫ
Now, let assume that
exp((m − 1)ǫ) − 1
kqm−1 kL∞ (Cl ) ≤ exp((m − 1)ǫ)kq0 kL∞ (Cl ) + ξ. (78)
ǫ
Combining estimates: 1 + ǫ ≤ exp(ǫ) and 1 ≤ exp(ǫ)−1 ǫ together with inequality (76) provided by the
assumption of Lemma 4.1 and (78), direct calculations give
 
exp((m − 1)ǫ) − 1
kqm kL∞ (Cl ) ≤ (1 + ǫ)kqm−1 kL∞ (Cl ) + ξ ≤ (1 + ǫ) exp((m − 1)ǫ)kq0 kL∞ (Cl ) + ξ +ξ ≤
ǫ
 
exp((m − 1)ǫ) − 1 exp(mǫ) − 1
exp(ǫ) exp((m − 1)ǫ)kq0 kL∞ (Cl ) + ξ + ξ ≤ exp(mǫ)kq0 kL∞ (Cl ) + ξ.
ǫ ǫ
The last inequality comes the estimate ǫ − exp(ǫ) ≤ −1. This completes the proof of Lemma 4.1.
Using Lemma 4.1, we are ready to prove the main result of this paper (namely Theorem 4.1).
Proof. (of Theorem 4.1).
First of all, introduce the domains

Di = {(t, x) : t ∈ [t0 , Tmax ], x ∈ R, |Xi (t) − x| ≤ υ}, S0 = {(t, x) : t0 ≤ t ≤ Tmax , x ∈ R},

and
D = {(t, X) : t ∈ [t0 , Tmax ], X ∈ R9 , kX(t) − Xk∞ ≤ υ}, (79)
where υ is a positive constant independent of the step size h and Xi (t) denotes the ith component of the
exact solution X(t) of the initial-value problem (1)-(6).
Plugging approximations (70)-(71) and equation (74), it is not hard to observe that
" 9
#
h X
n n n n
δi (tn , Xi (tn )) = Fi (tn , X ) + ∂t Fi (tn , X ) + Fk (tn , X )∂k Fi (tn , X ) , (80)
4
k=1

and
1
δi (tn+ 21 , Xi (tn+ 12 )) = 3Fi (tn+ 21 , X n+ 2 ) − Fi (tn , X n ). (81)

13
Hence, the functions δi given by equation (74) and their partial derivatives are continuous on the domain
Di . Since Di is a compact subset of the strip S0 , their partial derivatives are bounded on Di . Applying the
Mean-Value Theorem, there exists a positive constant Li which does not depend on the step size h so that

|δi (t, Xi ) − δi (t, Yi )| ≤ Li |Xi − Yi |, (82)

for every (t, Xi ) and (t, Yi ) in Di . Furthermore, a simple manipulation of (44) results in
" 9
#
Xi (tn+ 21 ) − Xi (tn ) h X
= Fi (tn , X(tn )) + ∂t Fi (tn , X(tn )) + Fk (tn , X(tn ))∂k Fi (tn , X(tn )) + O(h2 ),
h/2 4
k=1

which is equivalent to
" 9
!#
Xi (tn+ 21 ) − Xi (tn ) h X
− Fi (tn , X(tn )) + ∂t Fi (tn , X(tn )) + Fk (tn , X(tn ))∂k Fi (tn , X(tn )) = O(h2 ).
h/2 4
k=1

Utilizing equations (73), (75) and (80), this becomes

σi (tn , Xi (tn )) = ∆i (tn , Xi (tn )) − δi (tn , Xi (tn )) = O(h2 ),

which implies
|σi (tn , Xi (tn ))| ≤ Ci4 h2 , for i = 1, 2, ..., 9, (83)
where Ci4 are positive constants that do not depend on the step size h. Setting C4 = max Ci4 ; σ(tn , X(tn )) =
1≤i≤9
(σ1 (tn , X1 (tn )), ..., σ9 (tn , X9 (tn )))T and taking the maximum of both sides of estimate (83) over i, to get

kσ(tn , Xi (tn ))kL∞ ≤ C4 h2 . (84)

In a similar manner, combining approximation (49), equations (73), (74) and (75), straightforward calcula-
tions provide
Xi (tn+1 ) − Xi (tn+ 12 ) h i
− 3Fi (tn+ 12 , X(tn+ 21 )) − Fi (tn , X(tn )) = O(h2 ),
h/2
which can be rewritten as

∆i (tn+ 21 , Xi (tn+ 12 )) − δi (tn+ 12 , Xi (tn+ 21 )) = O(h2 ).

Utilizing the definition of σi (tn+ 12 , Xi (tn+ 12 )), this implies

|σi (tn+ 12 , Xi (tn+ 12 ))| ≤ Ci5 h2 ,

where Ci5 (1 ≤ i ≤ 9) are positive constants independent of h. Taking the maximum over i, this yields

kσ(tn+ 12 , X(tn+ 21 ))kL∞ ≤ C5 h2 , (85)

where C5 = max Ci5 ; σ(tn+ 12 , X(tn+ 21 )) = (σ1 (tn+ 21 , X1 (tn+ 12 )), ..., σ9 (tn+ 21 , X9 (tn+ 12 )))T .
1≤i≤9

We consider the functions δbi (t, x) defined in the strip S0 = {(t, x) : t0 ≤ t ≤ Tmax , x ∈ R}, as


 δi (t, x), if (t, x) ∈ Di ,



δbi (t, x) = δi (t, Xi (t) + υ), for t ∈ [t0 , Tmax ] and x > Xi (t) + υ, (86)





δi (t, Xi (t) − υ), if t ∈ [t0 , Tmax ] and x < Xi (t) − υ.

We remind that Xi (t) is the ith component of the exact solution X(t) of the initial-value problem (1)-(6).
Now, using relation (86), it is not hard to observe that each function δbi satisfies the ”Lipschitz requirement”
in the strip S0 , that is ,
|δbi (t, x) − δbi (t, y)| ≤ Li |x − y|, (87)

14
for all (t, x) and (t, x) in S0 , where the positive constant is given by relation (82). Thus, each δbi is continuous
on S0 . Indeed. Let (t, x) and (t, y) be two elements of S0 . If (t, x) and (t, y) lie in Di , then estimate (87)
holds thanks to inequality (82), so the function δbi is continuous and satisfies the ”Lipschitz requirement” on
Di . Otherwise, either (t, x) or (t, y) do not lie in Di . This corresponds to three cases: (a) (t, x) lies in Di
and (t, y) does not lie in Di , (b) ((t, x) does not lie in Di and (t, y) does lies in Di ) and (c) (t, x) and (t, y)
do not lie in Di . Here we should prove only one case, for instance (t, x) does not lie in Di and (t, y) does lie
in Di , the proof of the two other cases are similar.

(t, x) ∈
/ Di ⇔ |x − Xi (t)| > υ ⇔ x − Xi (t) > υ or − (x − Xi (t)) > υ ⇔ x > Xi (t) + υ or x < Xi (t)) − υ.
So, 
 |δi (t, Xi (t) + υ) − δi (t, y)|, for t0 ≤ t ≤ Tmax and x > Xi (t) + υ,
|δbi (t, x) − δbi (t, y)| =

|δi (t, Xi (t) − υ) − δi (t, y)|, if t ∈ [t0 , Tmax ] and x < Xi (t) − υ,

 Li |Xi (t) + υ − y|, if t ∈ [t0 , Tmax ] and x > Xi (t) + υ,
≤ (88)

Li |Xi (t) − υ − y|, for t0 ≤ t ≤ Tmax and x < Xi (t) − υ.
But
(t, y) ∈ Di ⇔ |Xi (t) − y| ≤ υ ⇔ Xi (t) − y ≤ υ and − Xi (t)) + y ≤ υ.
Using this, it is easy to see that

x > Xi (t) + υ ⇔ x − y > Xi (t) + υ − y ≥ 0 ⇔ |Xi (t) − y| > |Xi (t) + υ − y|,

and
x < Xi (t) − υ ⇔ x − y < Xi (t) − υ − y ≤ 0 ⇔ |Xi (t) − y| > |Xi (t) − υ − y|.
This fact together with estimate (88) results in

|δbi (t, x) − δbi (t, y)| ≤ Li |x − y|.

This ends the proof of the first case (a). Thus, δbi satisfies the ”Lipschitz condition” on the strip S0 = {(t, x) :
t0 ≤ t ≤ Tmax , x ∈ R}. In addition, δbi is also continuous on S0 .

Since, (t, Xi (t)) ∈ Di , so δbi (t, Xi (t)) = δi (t, Xi (t)). A combination of (84)-(85) provides

|∆i (tn , Xi (tn )) − δbi (tn , Xi (tn ))| = |∆i (tn , Xi (tn )) − δi (tn , Xi (tn ))| ≤ C4 h2 , (89)

and

|∆i (tn+ 21 , Xi (tn+ 12 )) − δbi (tn+ 21 , Xi (tn+ 12 ))| = |∆i (tn+ 12 , Xi (tn+ 12 )) − δi (tn+ 21 , Xi (tn+ 12 ))| ≤ C5 h2 . (90)

We recall that the two-level explicit method (70)-(72) is generated by the function δi . In fact, plugging
equations (70), (71) and (74), δi is explicitly defined as
  
 P9
Fi (tn , X n ) + h4 ∂t Fi (tn , X n ) + Fk (tn , X n )∂k Fi (tn , X n ) , if h 6= 0
δi (tn , Xi (tn )) = k=1

Fi (tn , X n ), if h = 0,

and
1
δi (tn+ 12 , Xi (tn+ 12 )) = 3Fi (tn+ 21 , X n+ 2 ) − Fi (tn , X n ).
Thus, approximations (70) and (71) become

n+ 21 h n+ 1 h
Xi = Xin + δi (tn , Xi (tn )) and Xin+1 = Xi 2 + δi (tn+ 12 , Xi (tn+ 12 )).
2 2

15
Analogously, the two-level numerical scheme generated by δbi should furnish the approximate solutions which
satisfy
b n + h δbi (tn , X
1
b n+ 2 = X
X bi (tn )) (91)
i i
2
and
b n+ 2 + h δbi (t 1 , X
1
b n+1 = X
X bi (t 1 )). (92)
i i n+ 2 n+ 2
2
In view of equation (73), we have
h 1
Xi (tl+ 21 ) = Xi (tl ) + ∆i (tn , Xi (tl )), for l = n, n + . (93)
2 2
b l , direct calculations result in
Plugging relations (91)-(93), and because ebli = Xi (tl ) − X i

n+ 12 h  h
bi (tn )) = ebni + h (δbi (tn , Xi (tn )) − δbi (tn , X
ei
b = ebni + ∆i (tn , Xi (tn )) − δbi (tn , X bi (tn )))
2 2
i
+(∆i (tn , Xi (tn )) − δbi (tn , Xi (tn ))) .
Taking the absolute value, it is easy to see that
  hhb i
 n+ 21 
bei  ≤ |beni | + |δi (tn , Xi (tn )) − δbi (tn , X
bi (tn ))| + |∆i (tn , Xi (tn )) − δbi (tn , Xi (tn ))| ≤
2
hh i  
|beni | + Li |Xi (tn ) − Xbin | + C4 h2 ≤ 1 + Lh |b eni | +
C4 3
h , (94)
2 2 2
where L = max Li . The last two estimates follows from inequalities (87) and (89). Taking the maximum
1≤i≤9
over i, estimate (94) gives
    
1 Lh C4 3 Lh C4 3
en+ 2 k∞ ≤ max
kb 1+ eni | +
|b h ≤ 1+ en k∞ +
kb h . (95)
1≤i≤9 2 2 2 2

In a similar way, utilizing relations (92), (93), (87) and (90), one easily shows that
 
n+1 Lh 1 C5 3
kb
e k∞ ≤ 1 + kben+ 2 k∞ + h . (96)
2 2

Substituting estimate (95) into (96) to obtain

 2   
n+1 Lh n Lh C4 C5 3
kb
e k∞ ≤ 1 + kb
e k∞ + 1 + + en k∞ + α2h h3 ,
h = (1 + α1h )kb (97)
2 2 2 2


C C5
where α1h = L 1 + Lh h and α2h = 1 + Lh 2 + 2 . To guarantee the convergence
4
of the algorithm, the
4 2

step size should satisfy 0 < h ≤ 1. This restriction allows to write: α2h ≤ 1 + L2 C24 + C25 := α2 . This fact,
together with inequality (97) yield
 2
n+1 Lh
kb
e k∞ ≤ 1+ en k∞ + α2 h3 .
kb
2
Applying Lemma 4.1, it holds
exp(nα1h ) − 1
en k∞ ≤ exp(nα1h )kb
kb e0 k∞ + α2 h3 . (98)
α1h
But it comes from the initial condition that eb0 = 0. Using this, relation (98) becomes

exp[nLh(1 + Lh
4 )] − 1
en k∞ ≤
kb L
α2 h2 .
L(1 + 4 )

16
Since h = Tmax
M
−t0
and tn = t0 + nh, then nh = tn − t0 ≤ Tmax . Furthermore, 1 < 1 + Lh
4 ≤1+ L
4 (indeed,
0 < h ≤ 1)). Utilizing this fact, we have
L
exp[LTmax(1 + 4 )] −1
en k∞ ≤
kb α2 h2 , (99)
L
which can be rewritten as
   
|Xi (tn ) − Xb n | ≤ α exp[LTmax (1 + L )] − 1 h2 < α exp[LTmax (1 + L )] − 1 , for i = 1, 2, ..., 9, (100)
i
L 4 L 4
 
since 0 < h ≤ 1. Setting υ = L α b n ) ∈ Di ,
exp[LTmax(1 + L4 )] − 1 > 0, estimate (100) indicates that (tn , X i
for i = 1, 2, ..., 9.

Now, according to the definition of δbi , we should have Xin = X

b n , en = ebn and δi (tn , Xi (tn )) =
i i i
δbi (tn , X
bi (tn )). This fact and estimates (99)-(100) provide

kX(tn ) − X n k∞ ≤ υ and ken k∞ ≤ υh2 . (101)

It comes from the inequality kuk∞ − kvk∞ ≤ ku − vk∞ (for any u, v ∈ R9 ) and estimate (101) that

kX n k∞ ≤ kX(tn )k∞ + υ. (102)

The analytical solution X(·) is bounded on the interval [t0 , Tmax ] because (t, X(t)) ∈ D, where the domain
D is given by relation (79). It follows from relation (102) that the approximate solution X is also bounded
over the interval [t0 , Tmax ]. Hence, the proposed approach (70)-(72) for solving the initial-value problem
(1)-(6) is stable.

Finally, the second estimate in relation (101) suggests that the new method is at least second-order
convergent. This completes the proof of Theorem 4.1.

5 Numerical experiments and Convergence rate

In this section, we use MatLab R2007b and we present a broad range of numerical evidences to illustrate
and demonstrate the efficiency of the proposed approach applied to the mathematical model of the Covid-19
spreading. We stress that in this situation, we obtain satisfactory results, so our algorithm performances are
not worse for multidimensional problems. We consider the particular case of the SARS-Cov-2 in Cameroon
where the data are available in [6] and we analyze and discuss the obtained results. The other real data used
in this study are taken from the literature [24, 25, 6]. More precisely, the number of people in this country is
approximately N = 25.000.000, the fraction at time t, of people in compartment X5 that are hospitalized is
assumed equals 1 (p(t) = 1, because of the decision made by WHO it was decided to hospitalized all detected
cases to reduce the transmission of the SARS-Cov-2 virus [51]), dX2 = 5.5 days, dg = 6 days, dX4 = 7.3
days (that is, γX4 = 1/7.3 (day −1 )), Cu = 0.3 days, Tmax = 55 days, t0 = 0 (which corresponds to 06 March
2020), h = 1 (the step size), w = 0.00133, w = 0.0667, βX2 = 0.1125, βX3 = 0.375, d0 = 14 (the period
of convalescence, we recall that this corresponds to the time a person is still hospitalized after recovering
from Covid-19), γX2 = 0.1818, γX3 = 0.78895, CX2 = 0.3, k = 0.13 (efficiency of control measures), λ = 12
(first day of the application of the control measures which corresponds to 17 March 2020 in Cameroon),
β X = Cu × βX3 = 0.1125 and β X3 = βX3 = 0.375.
3

To prevent the transmission of the Coronavirus 2019 in the other cities of the country, the authorities
decided to suspend all fights in the country and to restrict the number of passengers in all public transporta-
tion and outbound trains in the cities of Yaounde, Douala, Bafoussam and municipalities on 17 March 2020
[6]. Hence, the following implementation of the control measures is considered in order to indicate the real
situation of the measured imposed

1, if t ∈ [05 March 2020, λ);
m(t) =
exp[−k(t − λ)], if t ∈ [λ, Tmax ),

17
where λ corresponds to 17 March 2020. Furthermore, we assume that the fraction of detected infected people
(θ(t)) is a linear function and the disease contact rate of a person in compartment X4 denoted w(t) (without
taking into account the control measures) in this territory is a continuous function and are given by

1−w 1 − θ(t)
θ(t) = w + t and βX4 (θ(t)) = β X + (βX3 − β X ) .
Tmax 3 3 1 − w(t)

We recall that the initial data used in our experiments are taken in [6]. Specifically, as on 06 March 2020,
there were two cases in which the first one was imported on 24 February 2020 and the other case was oc-
curring by contact with the first case reported. By devoting resources, on 08 March 2020, 108 of the 176
identified individuals of local transmission were traced back to their presumed exposure, either to a known
case or to a location linked to spread [6]. From this observation, we set X1 (t0 ) = N − 176, X2 (t0 ) = 172,
X3 (t0 ) = X4 (t0 ) = 1, X5 (t0 ) = 2, Xj (t0 ) = 0, for j = 6, 7, 8, 9. Using the above tools together with the data
provided in [6], we perform a wide set of numerical experiments to demonstrate the robustness of the new
approach. More specifically:

In Table 1 we show the evolution of the model cumulative number of cases in Cameroon from 06 March
2020 to 30 April 2020 subjects to various values of the fraction of detected infected people that are docu-
mented (θ), whereas Figure 1 (Figure 1) deals with the evolution of the predicted number of cases in the
same country during the considered time interval (here, Tmax = 55 days), taking into account different values
of θ. The model cumulative number of reported deaths and computed ones are presented in both Table 2
and Figure 1 (Figure 2). Also in this case, we consider different values of θ. Table 3 and Figure 1 (Figure
3) indicate the model cumulative number of hospitalized people using various values of θ while both Table
4 and Figure 1 (Figure 4) suggest the number of infected individuals, who are not expected to be detected
yet, may infect other persons and start to developing clinical signs. In both Table 5 and Figure 2 (Figure 5),
we present the expected number of people that will recover, but can still infect other persons. In Table 6 and
Figure 2 (Figure 6), we show the evolution of the cumulative number of people who recovered after being
previously infected but were undetected and documented by the government and are no longer infectious.
The model cumulative number of individuals exposed to the Covid-19 disease is indicated in both Table 7
and Figure 2 (Figure 7). Both Table 8 and Figure 2 (Figure 8) present the evolution of the number of
persons infected by contact with people in compartment X2 , while Table 9 together with Figure 3 (Figure
9) show the model cumulative number of people infected by contact with individuals in compartment X4 .
In Table 10 and Figure 3 (Figure 10), we present the number of persons infected by contact with people in
compartment X10 = X5 +X6 . Each analysis described above considers various values of θ. Finally, we draw in
Table 11 and sketch in Figure 3 (Figure 11) the effective reproduction number of SARS-Cov-2 for Cameroon.

Now, focusing on the values taken by θ, we observe from the tables and figures that for a greater value of
θ (considered in this analysis) the predicted results reproduce quite accurately the evolution of the number
of cases, number of deaths and number of hospitalized people. In particular, for θ = 0.3212 the predicted
values on 30 April 2020 of cases, deaths and people in hospital are approximately 2236.0, 111.2786 and
47.4915, respectively, while these values become overestimated: 3260.6 cases, 166.1457 deaths and 90.1832
hospitalized people when θ = w = 0.0667 (the smallest value taken by θ in [w, 1]). Furthermore, for various
values of the parameter θ we observe that the maximum number of undetected cases: 116.0502 people (for
θ = 0.0667), 85.1727 people (for θ = 0.1515), 74.3269 people (for θ = 0.3212) and 56.5865 people (for
θ = 0.7455) estimated by the new approach represent: 5.5% (for θ = 0.0667), 4.2257% (for θ = 0.1515),
4.0439% (for θ = 0.3212) and 3.6912% (for θ = 0.7455) of the number of total cases obtained on 30 March
2020. Interestingly, the results provided by our method suggest that, despite the relative control of Covid-19
pandemic in Cameroon, they may still exist an undetected source of infected persons that could cause the
increase of the disease in a near future, if the implementation of the control measures are significantly relaxed
like the government decided at the beginning of May 2020. In addition, Figure 1 (Figure 3) indicates that
the peak of the persons hospitalized in this country at the same time should be reached around 05 April 2020,
with approximately 800 hospitalized patients. This number is associated to the smallest value of θ = 0.0667
considered in this study. However, the obtained results slightly overestimate the observed values by around
693. Focusing on the recovered people, the considered technique suggests a maximum number of 3037.7
people, 2300.8 people, 2073.5 people, and 1686.0 people corresponding to various values of θ : 0.0667, 0.1515,

18
0.3212 and 0.7455, respectively. For θ = 0.1515, the final number of 241.0999 people is very close to the real
observation (around 244 hospitalized individuals on 15 April 2020). This shows that the proposed method
is a reasonable decision tool to estimate the number of beds in hospital during a pandemic or an epidemic.
Also, it is worth mentioning that our approach is able to detect early a reasonable expected date of this peak.

Finally, we observe from both Table 11 and Figure 3 (Figure 11) that the value of the effective repro-
duction number (Re ) decreases since the application of the control measures and it becomes less than 1 after
the peak is attained (30 March 2020).
Tables 1. Model cumulative number of cases with various values of θ
06 March 2020 17 March 2020 30 March 2020 15 April 2020 30 April 2020
NC(real data) 2 15 142 848 2014
CM(0.0667) 2 696.3508 2149.8 3245.3 3322.0
CM(0.1515) 2 681.7603 2048.7 2467.4 2481.5
CM(0.3212) 2 647.5659 1861.4 2223.8 2236.0
CM(0.7455) 2 583.5342 1538.2 1808.4 1817.5

Tables 2. The evolution of number of deaths with different values of θ

06 March 2020 17 March 2020 30 March 2020 15 April 2020 30 April 2020
ND(real data) 0 1 6 14 61
X9 (0.0667) 0 17.8300 79.6390 141.2279 166.1457
X9 (0.1515) 0 17.5511 73.8069 108.6332 121.8038
X9 (0.3212) 0 16.8994 68.3353 99.5074 111.2786
X9 (0.7455) 0 15.6638 58.6885 83.6826 93.0934

Tables 3. The cumulative number of hospitalized people

06 March 2020 17 March 2020 30 March 2020 15 April 2020 30 April 2020
NH (real obs.) 2 8 61 244 328
Host(0.0667) 2 335.5282 668.8886 476.0395 86.9377
Host(0.1515) 2 325.1239 673.6724 241.0999 44.0635
Host(0.3212) 2 300.8178 583.1121 209.3954 47.4915
Host(0.7455) 2 255.0717 432.0405 158.2955 54.7413

These numbers should help to expect the number of beds in hospitals during an epidemic.

Tables 4. Cumulative number of infected individuals who are not expected to be detected yet, but start
developing clinical signs

06 March 2020 17 March 2020 30 March 2020 15 April 2020 30 April 2020
X3 (0.0667) 1 70.3135 116.0502 15.4822 0.2363
X3 (0.1515) 1 67.8561 85.1727 2.6167 0.0608
X3 (0.3212) 1 62.0955 74.3269 2.2831 0.0530
X3 (0.7455) 1 51.8569 56.5865 1.7375 0.0404

Tables 5. Model cumulative number of infected people that will recover but can still infect other persons

06 March 2020 17 March 2020 30 March 2020 15 April 2020 30 April 2020
X5 (0.0667) 2 307.1123 608.9473 486.4522 106.7015
X5 (0.1515) 2 297.3910 626.9163 268.6669 52.8153
X5 (0.3212) 2 274.6855 547.2620 232.7084 45.7222
X5 (0.7455) 2 231.8949 413.7484 173.1906 33.9913

19
Tables 6. Evolution of the number of undetected infected people who recovered but are no longer infectious

06 March 2020 17 March 2020 30 March 2020 15 April 2020 30 April 2020
X5 (0.0667) 0 342.9927 1402.0 2582.4 3037.7
X5 (0.1515) 0 339.0853 1299.3 2070.9 2300.8
X5 (0.3212) 0 329.8487 1202.1 1874.4 2073.5
X5 (0.7455) 0 312.7987 1030.5 1537.8 1686.0

Tables 7. Cumulative number of people exposed to Covid-19 disease

06 March 2020 17 March 2020 30 March 2020 15 April 2020 30 April 2020
X2 (0.0667) 172 317.5650 495.4672 52.1206 0.7125
X2 (0.1515) 172 306.0850 289.9048 8.6553 0.2011
X2 (0.3212) 172 279.0561 252.9417 7.5517 0.1754
X2 (0.7455) 172 231.3889 192.4997 5.7472 0.1335

Tables 8. Model cumulative number of people infected by contact with persons in compartment X2

06 March 2020 17 March 2020 30 March 2020 15 April 2020 30 April 2020
X2 (0.0667) 19.3499 35.7245 37.7351 0.6432 0.0013
X2 (0.1515) 19.3499 34.4331 0.0000 0.0000 0.0000
X2 (0.3212) 19.3499 31.3926 0.0000 0.0000 0.0000
X2 (0.7455) 19.3499 26.0303 0.0000 0.0000 0.0000

Tables 9. Model cumulative number of people infected by contact with persons in compartment X4

06 March 2020 17 March 2020 30 March 2020 15 April 2020 30 April 2020
X4 (0.0667) 0.3750 -9.1174 -12.2623 -1.5011 -0.0477
X4 (0.1515) 0.3641 -8.0160 0.0000 0.0000 0.0000
X4 (0.3212) 0.3368 -5.6330 0.0000 0.0000 0.0000
X4 (0.7455) 0.2686 -1.5867 0.0000 0.0000 0.0000

Tables 10. Model cumulative number of people infected by contact with persons in compartment X10 =
X5 + X6

06 March 2020 17 March 2020 30 March 2020 15 April 2020 30 April 2020
X10 (0.0667) 0.0044 0.7379 0.9748 0.1191 0.0038
X10 (0.1515) 0.0058 0.9354 000 000 000
X10 (0.3212) 0.0069 1.0414 000 000 000
X10 (0.7455) 0.0076 0.9745 000 000 000

Tables 11. Effective reproduction number of SARS-Cov-2 for Cameroon

06 March 2020 17 March 2020 30 March 2020 15 April 2020 30 April 2020
Re(0.0667) 1.1111 1.1111 0.7546 0.1229 0.0175
Re(0.1515) 1.1164 1.1163 0.0000 0.0000 0.0000
Re(0.3212) 1.1209 1.1208 0.0000 0.0000 0.0000
Re(0.7455) 1.1236 1.1236 0.0000 0.0000 0.0000

6 General conclusion and future works

We have developed an efficient two-level explicit method for estimating the propagation of Covid-19 dis-
ease with undetected infectious cases. The analysis has shown that the new algorithm is stable, at least

20
second-order accuracy and can serve as a fast and robust tool for integrating of general systems of ordinary
differential equations. Numerical results based on the case of Cameroon reproduce quite accurately the
evolution of the number of cases (detected or undetected), number of deaths, number of people in hospitals,
number of infected detected persons who recovered and number of infected undetected individuals who re-
covered by natural immunity from 06 March 2020 to 30 April 2020. The approach presented in this work can
help to estimate the number of beds in hospitals during a pandemic. Furthermore, the proposed technique
can be considered as a fundamental tool for detecting early a reasonable expected date of the peak during
an epidemic. Our future works will consider the numerical solution of a more complex system of ordinary
differential equations using the new two-level explicit approach.

Acknowledgment. This work has been partially supported by the deanship of scientific research of
Al-Imam Muhammad Ibn Saud Islamic University (IMSIU) under the Grant No. 331203.

References
[1] . J. Adam Kucharski. ”Complex models may be no more reliable than simple ones if they miss key
aspects of the biology. Complex models can create the illusion of of realism, and make it harder to spot
crucial omissions”, https://www.pnas.org/content/103/33/12221, April 2020.
[2] . M. Anderson. ”Population biology of infectious disease: Part 1”, Nature, 280 (1979), p. 361-367.
[3] R. M. Anderson, R. M. May. ”Infection diseases of humans: dynamics and control”, Oxford: Oxford
universty press, (1991).

[4] R. Beaglehole, R. Bonita, T. Kjellstrom. ”Basic epidemiology. Communicable disease epidemiology”,

Geneva: WHO, (1993), p. 97-105.
[5] F. Brauer, C. Castillo-Chávez. ”Mathematical models in population biology and epidemiology. Text in
applied mathematics”, Springer, (2001).
[6] Cameroon: COVID-19 Rapport de Situation No: 23, 22,· · · , 4. 05 March-30 April, (2020).
[7] . S. Chen, J. Yang, W. Yang, T. Barnighausen. ”Covis-19 control in China during mass population
morments at New Year”, The Nacet, 395(10226) (2020), p. 764-766.
[8] R. Codina, J. Principe, C. Munoz, J. Baiges. ”Numerical modelling of chlorine concentration in water
storage tanks”. Int. J. Numer. Methods Fluids, 79 (2015), pp. 84-107.
[9] W. Dai, R. Nassar. ”Compact ADI method for solving partial differential equations”. Numer. Meth.
Partial Differential Eq., 18 (2002) 129-142.
[10] S .C. R. Dennis, J. D. Hudson. ”Compact h4 finite-difference approximations to operators of Navier-
Stokes type”. J. Comput. Phys., 85 (1989) 390-418.
[11] . O. Diekmann, H. Heesterbeek, T. Britton. ”Understanding infectious disease dynamics”, Princeton
Series in Theoretical and Computational Biology”, Princeton University Press, (2013).
[12] J. Douglas, T. F. Russell. ”Numerical methods for convection-dominated diffusion problems based on
combining the method of characteristics with finite element or finite difference procedures”. SINUM, 19
(1982), pp. 871-885.
[13] European Centre for Disease Prevention and Control. ”Discharge criteria for COVID-19 cases-
When is it safe to discharge Covid-19 cases from the hospital or end home isolation?”.
https://www.ecdc.europa.eu/sites/default/files/documents/Covid-19-Discharge-criteria.pdf, March
(2020).
[14] M. M. Gupta, R. P. Manohar, J. W. Stephenson. ”A single cell high order scheme for the convection-
diffusion equation with variable coefficients”, Int. J. Numer. Meth. Fluids, 4 (1984) 641-651.

21
[15] V. Guvanasen, R. E. Volker. ”Numerical solutions for solute transport in unconfined aquifers”. Int. J.
Num. Meth. Fluids, 3 (1983) 103-123.
[16] M. E. Halloran. ”Concepts of infectious disease epidemiology. In: Rothman KJ, Greedland S. eds”,
Modern epidemiology, 2nd edn. Philadelphia, PA: Lippincott-Raven, (2003), p. 519-554.
[17] P. Henrici. ”Discrete variable methods in ordinary differential equations”. New York, John Wiley (1962).
[18] . B. Ivorra, A. M. Ramos, D. Ngom. ”Be-CoviDis: A mathematical model to predict the risk of hu-
man diseases spread between-countries. Validation and application to the 2014 Ebola Virus Disease
epidemic”, Bulletin of Mathematical Biology, 77(9), (2015), p. 1668-1704.
[19] . B. Ivorra, M. R. Ferrández, M. Vela-Pérez, A. M. Ramos. ”Mathematical modeling of the
spread of the coronavirus disease 2019 (Covid-19) taking into account the undetected infec-
tious. The case of China”. Instituto de Matematica Interdisciplinar, April (2020). DOI link:
http://www.doi.org/10.13140/RG.2.2.21543.29604
[20] V. John, J. Novo. ”On (essentially) non-oscillatory discretizations of evolutionary convection-diffusion
equations”. J. Comput. Phy., 231 (2012), 1570-1585.

[21] S. Karaa, J. Zhang. ”Higher order ADI method for solving unsteady convection-diffusion problems”. J.
Comput. Phys., 198 (2004) 1-9.
[22] P. D. Lax, B. Wendroff. ”Systems of conservation laws”. Comm. Pure & Appl. Math. 13 (1960) 217-237.
[23] . P. Lekone, B. Finkenstädt. ”Statistical inference in a stochastic epidemic seir model with control
intervention: Ebola as a case study”, Biometric, 62(4) (2006), p. 1170-1177.
[24] . R. Li, S. Pei, B. Chen, Y. Song, T. Zhang, W. Yang, J. Shaman. ”Substantial undocumented infection
facilitates the rapid dissemination of novel coronavirus (SARS-Cov2)”, Science (2020).
[25] . T. Liu, J. Hu, M. Kang, L. Lin, H. Zhong, J. Xiao, and et al. ”Transmission dynamics of 2019 novel
coronavirus (2019-ncov)”, bioRxiv (2020).
[26] . W. Luo, M. Majumder, D. Liu. ”The role of absolute humidity on transmission rates of the COVID-19
outbreak”, M edRxiv (2020).
[27] R. W. MacCormack. ”An efficient numerical method for solving the time-dependent compressible Navier-
Stokes equations at high Reynolds numbers”. NASA TM (1976) 73-129.
[28] R. J. MacKinnon, R. W. Johnson. ”Differential equation based representation of truncation errors for
accurate numerical simulation”, Int. J. Numer. Meth. Fluids, 13 (1991) 739-757.
[29] C. Man, C. W. Tsai. ”A high order predictor-corrector scheme for two-dimensional advection-diffusion
equation”. Int. J. Numer. Methods Fluids, 56(4), (2008) 401-418.
[30] . B. MartÍnez-López, B. Ivorra, A. M. Ramos, J. M. Sánchez-Vizcaino. ”A novel spatial and stochastic
model to evaluate the within- and between-farm transmission of classical swine fever virus. I. General
concepts and description of the model”, Veterinary Microbiology, 147(3 − 4) (2011), p. 300-309.
[31] F. T. Namio, E. Ngondiep, R. Ntchantcho, J. C. Ntonga. ”Mathematical models of complete shallow
water equations with source terms, stability analysis of Lax-Wendroff scheme”. J. Theor. Comput. Sci.,
Vol. 2(132) (2015).

[32] . E. Ngondiep. ”An efficient three-level explicit time-split scheme for solving two-dimensional unsteady
nonlinear coupled Burgers’s equations”, Int. J. Numer. Methods Fluids. November (2019), p. 1-19, 19
pages.
[33] E. Ngondiep. ”Stability analysis of MacCormack rapid solver method for evolutionary Stokes-Darcy
problem”, J. Comput. Appl. Math. 345(2019), 269-285, 17 pages.

22
[34] E. Ngondiep. ”Long Time Stability and Convergence Rate of MacCormack Rapid Solver Method for
Nonstationary Stokes-Darcy Problem”, Comput. Math. Appl., Vol 75, (2018), 3663-3684, 22 pages.
[35] E. Ngondiep. ”An efficient three-level explicit time-split method for solving 2D heat conduction equa-
tions”, submitted.
[36] E. Ngondiep. ”Long time unconditional stability of a two-level hybrid method for nonstationary incom-
pressible Navier-Stokes equations”, J. Comput. Appl. Math. 345(2019), 501-514, 14 pages.
[37] E. Ngondiep. ”Asymptotic growth of the spectral radii of collocation matrices approximating elliptic
boundary problems”, Int. J. Appl. Math. Comput., 4(2012), 199-219, 20 pages.
[38] E. Ngondiep. ” Error estimates of MacCormack rapid solver method for nonstationary incompressible
Navier-Stokes equations”, preprint available from http://arXiv.org/abs/1903.10857, (2019) 26 pages.
[39] E. Ngondiep. ”A novel three-level time-split MacCormack scheme for two-dimensional evolutionary
linear convection-diffusion-equation with source term”, Int. J. Comput. Math. (2020), 24 pages. DOI:
10.1080/00207160.2020.1726896.
[40] E. Ngondiep. ”A fourth-order two-level factored implicit scheme for solving two-dimensional unsteady
transport equation with time dependent dispersion coefficients”, submitted.
[41] E. Ngondiep. ”A three-level explicit time-split MacCormack method for 2D nonlinear reaction-diffusion
equations”, preprint available from http://arxiv.org/abs/1903.10877 (2019) 25 pages.
[42] E. Ngondiep. ”A novel three-level time-split MacCormack method for solving two-dimensional viscous
coupled Burgers equations”, preprint available online from http://arxiv.org/abs/1906.01544, 2019.
[43] E. Ngondiep, N. Kerdid, M. Abdulaziz Mohammed Abaoud, I. Abdulaziz Ibrahim Aldayel. ”A three-
level time-split MacCormack method for two-dimensional nonlinear reaction-diffusion equations”, Int J
Numer Meth Fluids, (2020) p. 1-26. https://doi.org/10.1002/fld.4844.
[44] E. Ngondiep, R. Alqahtani, J. C. Ntonga. ”Stability analysis and convergence rate of MacCormack
scheme for complete shallow water equations with source terms”. Preprint available online from
http://arxiv.org/abs/1903.11104, 2019.
[45] B. J. Noye, H. H. Tan. ”Finite difference methods for solving the two-dimensional advection-diffusion
equation”, Int. J. Numer. Methods Fluids, 9(1) (1989) 75-98.
[46] . World Organization Health. ”Coronavirus disease (2019) situation reports”.
https://www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/, March 2020.
[47] . World Organization Health. ”Report of the WHO-China joint mission on Coronavirus dis-
ease 2019”. https://www.who.int/docs/default-source/coronaviruse/who-china-joint-mission-on-covid-
19-final-report.pdf/, February 2020.
[48] S. M. Pautanen, D. E. Low, B. Henry, et al. ”Identification of severe acute respiratory syndrome in
Canada”, N. Engl J. Med., April (2003). https://www.nejm.org/
[49] H. R. Thieme. ”Mathematics in population biology. Mathematical biology series”, Princeton University
press, (2003).
[50] . P. Van den Driessche, J. Watmough. ”Reproduction numbers and sub-threshold epidemic equilibria for
compartmental models of disease transmission”, Mathematical Bioscience, 180(1 − 2), (2002), p. 29-48.
[51] . R. Verity, L. C. Okell, I. Dorigatti, P. Winskill, C. Whattaker, N. Imai, G. Guomo-Dannenburg, H.
Thompson, P. Wlker and et al.. ”Estimates of the severity of Covid-19 disease.”, medRxiv, (2020).
[52] W. Wang, M. D. Aili Jiang, Q. Qin. ”Temperature significantly change COVID-19 transmission in 429
cities”, M edRxiv, (2020).

23
[53] Z. Zlatev, R. Berkowicz, L. P. Prahm. ”Implementation of a variable stepsize variable formula in the
time-integration part of a code for treatment of long-range transport of air polluants”. J. Comput.
Phys., 55 (1984) 278-301.

24
Model cumulative number of cases, deaths, hospitalized people and infected persons who are not detected yet.
Cumulative number of cases with different values of θ Cumulative number of deaths with different values of θ
3500 180
obs. values obs. values
θ=0.0667 θ=0.0667
160
3000 θ=0.1515 θ=0.1515
θ=0.3212 140 θ=0.3212
θ=0.7455 θ=0.7455
2500
120

Number of deaths
Number of cases

2000 100

1500 80

60
1000
40

500
20

0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
0≤ t≤ 60, (in day) 0≤ t≤ 60, (in day)

Figure 1 Figure 2
Cumulative number of hospitalized people with different values of θ Cumulative number of infected people who are not detected yet with different values of θ
800 120
real obs. θ=0.0667
Number of infected people who are not detected yet

θ=0.0667 θ=0.1515
700
θ=0.1515 100 θ=0.3212
θ=0.3212 θ=0.7455
600 θ=0.7455
Number of people in hospital

80
500

400 60

300
40

200

20
100

0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
0≤ t≤ 60, (in day) 0≤ t≤ 60, (in day)

Figure 3 Figure 4

Figure 1: Number of cases, deaths, hospitalized people and undetected infected individual

25
Cumulative number of recovered infected, undetected infected who recovered, people exposed to Covid-19,
infected by contact with X2 .

Undetected infected people who recovered, are no longer infectious

Cumulative number of infected people that will recover with various values of θ Cumulative number of undetected infected people who recovered with various values of θ
800 3500
θ=0.0667 θ=0.0667
θ=0.1515 θ=0.1515
700 3000 θ=0.3212
Number of infected people that will recover

θ=0.3212
θ=0.7455 θ=0.7455
600
2500

500
2000
400
1500
300

1000
200

100 500

0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
0≤ t≤ 60, (in day) 0≤ t≤ 60, (in day)

Figure 5 Figure 6
Cumulative number of people exposed to Covid−19 with different values of θ Cumulative number of people infected by contact with X2 with various θ
600 1200
θ=0.0667 θ=0.0667
θ=0.1515 θ=0.1515
Number of people exposed to Covid−19 disease

θ=0.3212
Number of people infected by contact with X

500 1000 θ=0.3212

θ=0.7455 θ=0.7455

400 800

300 600

200 400

100 200

0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
0≤ t≤ 60, (in day) 0≤ t≤ 60, (in day)

Figure 7 Figure 8

Figure 2: Number of recovered infected, recovered undetected infected, exposed to Covid-19, infected by
contact with X2

26
Model cumulative number of people infected by contact with X4 , X10 and effective reproduction number.
Cumulative number of people infected by contact with X with various θ Cumulative number of people infected by contact with X with various values of θ
4 10
50 30
θ=0.0667
θ=0.1515

Number of people infected by contact with X10

Number of people infected by contact with X4

0
25 θ=0.3212
θ=0.7455
−50

20
−100

−150 15

−200
10

−250
θ=0.0667
θ=0.1515 5
−300
θ=0.3212
θ=0.7455
−350 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
0≤ t≤ 60, (in day) 0≤ t≤ 60, (in day)

Figure 9 Figure 10
Effective reproduction number of Covid−19 with various values of θ
1.4
θ=0.0667
θ=0.1515
Effective reproduction number of SARS−Cov−2

1.2 θ=0.3212
θ=0.7455
1

0.8

0.6

0.4

0.2

0
0 10 20 30 40 50 60
0≤ t≤ 60, (in day)

Figure 11

Figure 3: Number of infected by contact with X4 , X10 and effective reproduction number Re

27