East Tennessee State University

Digital Commons @ East

Tennessee State University

Electronic Theses and Dissertations Student Works

12-2010

The SIR Model When S(t) is a Multi-Exponential Function.

Teshome Mogessie Balkew
East Tennessee State University

Follow this and additional works at: https://dc.etsu.edu/etd

Part of the Non-linear Dynamics Commons

Recommended Citation
Balkew, Teshome Mogessie, "The SIR Model When S(t) is a Multi-Exponential Function." (2010). Electronic
Theses and Dissertations. Paper 1747. https://dc.etsu.edu/etd/1747

This Thesis - unrestricted is brought to you for free and open access by the Student Works at Digital Commons @
East Tennessee State University. It has been accepted for inclusion in Electronic Theses and Dissertations by an
authorized administrator of Digital Commons @ East Tennessee State University. For more information, please
contact digilib@etsu.edu.
 The SIR Model When S(t) is a Multi-Exponential Function

A thesis presented to

the faculty of the Department of Mathematics

East Tennessee State University

In partial fulfillment

of the requirements for the degree

Master of Science in Mathematical Sciences

by

Teshome Mogessie Balkew

December 2010

Jeff Knisley, Ph.D., Chair

Ariel Cintron-Arias, Ph.D.

Robert Gardner, Ph.D.

Fredrick Norwood, Ph.D.

Keywords: SIR model, Relative Removal Rate, Phase Plane Analysis, R0 .

 ABSTRACT

The SIR Model When S(t) is a Multi-Exponential Function

by

Teshome Mogessie Balkew

The SIR can be expressed either as a system of nonlinear ordinary differential equa-

tions or as a nonlinear Volterra integral equation. In general, neither of these can be

solved in closed form. In this thesis, it is shown that if we assume S(t) is a finite
Pn −σk t
multi-exponential, i.e. function of the form S(t) = a + k=1 rk e or a logistic func-

tion which is an infinite-multi-exponential, i.e. function of the form S(t) = c + b+ea wt ,

then we can have closed form solution. Also we will formulate a method to determine

R0 the basic reproductive rate of an infection.

2
Copyright by Teshome Mogessie Balkew 2010

3
 ACKNOWLEDGMENTS

First and foremost I would like to extend my deepest gratitude to the gracious,

merciful Heavenly Father Almighty God, Who helped me in every way in my stay

here in ETSU. My special thanks goes to Dr. Robert B. Gardner. I feel that words

may not be enough to express how deeply grateful I am for all help he rendered

to me. Thank you Dr. Bob!!. I also acknowledge the support and encouragement

accorded to me by Dr. Fredrick Norwood, specially in the first two or so months of

my stay here. My great appreciation and thanks go to my thesis advisor Dr. Jeff

Knisley. He has been instrumental at all stages of my thesis development. He gave

me all necessary advising for my thesis to go in the right direction. From his excellent

coaching I learned a lot. I like to say ”thank you” to all my teachers: Dr. Debra

Knisley, Dr. Robert M. Price and Dr. Ariel Cintron-Arias who inspired me one or

the other way in my stay in ETSU. I acknowledge the help I got from Dr. Yared

Nigussie on several occasions. I am so grateful to my friend Azamed Gezahagne and

his wife Lidya Hailesilassie. Finally I would like to say ”thank you” to all my friends

and classmates here in ETSU who influenced me one or the other way. Thank you

and May God bless you all.

4
 CONTENTS

ABSTRACT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

ACKNOWLEDGMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2 SIR EPIDEMIOLOGICAL MODEL . . . . . . . . . . . . . . . . . . 11

2.1 Two Formulations of the SIR Model . . . . . . . . . . . . . . 12

2.1.1 The Classical Differential Equation Form of the SIR

Model . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.1.2 The Integral Equation Form of the SIR Model . . . . 13

2.1.3 Equivalence of the Differential and Integral Form of

the SIR Model . . . . . . . . . . . . . . . . . . . . . 14

2.2 Phase Plane Analysis . . . . . . . . . . . . . . . . . . . . . . . 15

2.2.1 Phase Plane in SI Plane . . . . . . . . . . . . . . . . 15

2.2.2 Phase Plane Analysis in RS Plane . . . . . . . . . . 17

2.2.3 The Basic Reproductive Rate of Infection R0 . . . . 19

3 THE SIR MODEL WHEN S(t) IS A MULTI-EXPONENTIAL FUNC-

TION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3.1 Finite Multi-Exponential Form of S(t) . . . . . . . . . . . . . 26

3.1.1 Susceptibles as a Single Exponential . . . . . . . . . 28

3.1.2 Susceptibles as a sum of two Exponentials . . . . . . 30

3.2 Susceptible as a Logistic Function . . . . . . . . . . . . . . . . 32

3.2.1 The Basic Reproductive Rate of Infection R0 . . . . 35

5
4 CONCLUSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

BIBLIOGRAPHY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

APPENDICES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

VITA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

6
 LIST OF FIGURES

1 One Phase Curve in SI Plane . . . . . . . . . . . . . . . . . . . . . . 16

2 Phase Curves in SI Plane . . . . . . . . . . . . . . . . . . . . . . . . . 18

3 Phase Curves in RS Plane . . . . . . . . . . . . . . . . . . . . . . . . 19

4 Phase Curves in SI Plane . . . . . . . . . . . . . . . . . . . . . . . . . 24

5 Phase Curves in RS Plane . . . . . . . . . . . . . . . . . . . . . . . . 24

6 Data Fit Curve for “Susceptible” for One Exponential . . . . . . . . . 29

7 SIR Curves for One Exponential S(t) . . . . . . . . . . . . . . . . . . 30

8 Data Fit Curve for the “Susceptible” for Two Exponential . . . . . . 31

9 SIR Curves for Two Exponential S(t) . . . . . . . . . . . . . . . . . . 32

10 Best Fit Curve for Logistic Function . . . . . . . . . . . . . . . . . . 34

11 SIR Model Curves for Logistic Function . . . . . . . . . . . . . . . . 35

12 Data Fit Curve for “Recovery” Data . . . . . . . . . . . . . . . . . . 36

13 The Graph of β as a Function of Time . . . . . . . . . . . . . . . . . 37

7
 1 INTRODUCTION

From pre-history to the present day, there have been many waves of epidemics

causing death and suffering of human beings. Most of these epidemics seemingly came

from nowhere, suddenly changed the normal demography of the affected population,

and ultimately disappeared before affecting the entirety of a population. This was a

puzzle for human beings for centuries, and there were many attempts to understand

these frequent epidemiological occurrences and to create a mechanism to control them

[2, 18].

• The first known result in mathematical epidemiology is due to Daniel Bernoulli.

In 1760 he formulated and solved a model for a smallpox epidemic [18].

• In 1906 William Hamer formulated a discrete time model for measle epidemics

[18].

• A significant contributor to compartmental-based epidemiological models was

the public physician R.A. Ross. In 1911 he proposed a differential equation

model for malaria as a host-reactor disease and won the second Nobel Prize in

Medicine [18].

• The classical SIR model as we know it today was first introduced by A.G.

Mckendrick and W.O. Kermack in 1927. These two public health physicians

extended the work of Ross and developed the concept of an epidemic threshold

[18].

8
 Most epidemiological models start from the same basic premise that the popula-

tion can be subdivided into a set of distinct classes (compartments) depending upon

their experience with respect to the disease [18]. One line of investigation is to di-

vide the population under study into the three compartments of [18] “Susceptibles”,

“Infecteds” and “Recovereds.” The number of individuals in each compartment is

denoted by S(t), I(t) and R(t) respectively, where

• S(t) denotes the number of individuals who are “Susceptible” to the disease at

time t. That is, S(t) is the number of people in the population who are not

“infected” at the given time t but who are vulnerable to the disease.

• I(t) signifies the number of “infected” individuals who are infectious and able

to spread the disease to a susceptible.

• R(t) counts those in the population who are “recovered” from the disease and

are no longer “susceptible”. Thus they do not contribute to the spread of the

disease.

Such a model is called an SIR model.

A simple mathematical model of a natural phenomenon is meant to highlight

the existing qualitative behaviors of the phenomena, so that we omit most details.

As natural phenomena are influenced by a variety of internal and external factors,

taking all aspects of it and try to model as it is makes the model too complicated,

bulky, and is generally unwise, as it is common to end up with equation(s) which

are difficult or impossible to solve. So it is common to take certain assumptions in

modeling.
9
 The SIR model is formulated under the following assumptions [2, 4, 18]:

• (I) The community size is constant and is sufficiently large over the duration of

the epidemic. We denote the number of population by N , where

N = S(t) + I(t) + R(t)

• (II) The susceptibles become infected at a percentage rate proportional to the

number of “infecteds”.

• (III) It is also assumed that the “infecteds” are uniformly distributed among

the “susceptibles”.

Remark

• Assumption (I) says that since the demography of a population does not change

significantly within a short period of time we ignore birth and immigration.

• We clearly know that the rate of leaving compartment at the beginning, middle

and end of the epidemic varies greatly. What assumption (II) says is that we

take the average (mean) number leaving compartments as a constant. This

assumption is reasonable if the population size is large, but it is not good if the

size of the population is small, since the random contact rate plays a significant

role for spreading the disease in small size population [1].

• In assumption (III) we ignored such factors as age, sex, social background,

seasons, geographical location, etcetera, which may affect the epidemic signifi-

cantly.

10
 2 SIR EPIDEMIOLOGICAL MODEL

In this chapter we will study two equivalent ways of formulating the SIR model –

the ordinary differential equation form and the integral equation form. The integral

equation form of the SIR model is more general than the ordinary differential equation

form for the following two reasons [1]:

1. We can deduce the classical ordinary differential equation form of the SIR model

by assuming P (t) is exponentially distributed.

2. Different functional forms of the probability density function P (t) gives us al-

ternative types of the SIR model.

Even though the more realistic probability density function is the gamma distribution,

which presupposes the probability of leaving the class is a function of the time spent

within the class, the classical SIR model is based on the premise that the probability

of infection is exponentially distributed, which implies that the rate of transfer from

one compartment to the other is independent of the time spent within the class [1].

With this consideration, we recover the classical differential equation form of the SIR

model from the integral form. We will explore the relation between two variables

with the absence of the third one. What we do here is examine the relation between

“S” and “I” in the SI coordinate plane and the relation between “R” and “S” in the

RS coordinate plane. Such a study of the relation between two variables with the

absence of the third one is called Phase Plane Analysis for the model [18].

11
 2.1 Two Formulations of the SIR Model

We will see two formulations of the SIR model. First we consider the differential

equation formulation and then we will see the integral equation formulation of the

SIR model.

2.1.1 The Classical Differential Equation Form of the SIR Model

In addition to the assumptions we made in Chapter 1, the SIR model also assumes

the following (see [2, 3, 4, 18] for additional discussion of these assumptions):

• f (I) = βI, where f (I) is an increasing function of “I” called the force of infection

and β is a constant called the infection rate. The infection rate tells us the rate

at which the “susceptibles” transfer into the “infecteds” group.

• Alternatively, f (I) = β (I/N ), where I/N is the “infecteds” density in a popu-

lation of size N .

The relation between the compartments is assumed to be as follows:

• As the epidemic spreads those in “susceptibles” compartment transfer to “in-

fecteds” compartment at a rate of β , i.e. S 0 = −f (I)S.

• The “infecteds” compartment receives the outflow from the first , and it also

loses “infecteds” to the “recovereds” one at a rate of β, i.e. I 0 = f (I)S − αI

– The parameter α is called the removal rate of infection. It is the rate

by which the “infected” transfer into the “recovery”. The probability of

12
 Rt
remaining “infectious” through time t is given by by P (t). 1/α = 0
P (t)dt

gives us the average infectious period.

• The third one implies R0 = αI.

From the above premises we have

β β
S 0 + I 0 + R0 = − SI + SI − αI + αI
N N
S 0 + I 0 + R0 = 0

which implies that S(t) + I(t) + R(t) is constant. In fact, from our assumptions

we know that at any given time “t”, S(t) + I(t) + R(t) = N. Then the classical

differential equation form of the SIR model can be summarized as

dI β
= S(t)I(t) − αI(t)
dt N
dS β
= − S(t)I(t) (1)
dt N
dR
= αI(t)
dt

2.1.2 The Integral Equation Form of the SIR Model

In this section we explore the Integral equation form of the SIR model (1). To be-

gin with, we derive the classic SIR model using an alternative approach. Specifically,

we begin with a modeling paradigm of the form

The number of “Infecteds” Initial number of “Infecteds” Those infected in [0, t]

= +
at time t still “Infecteds” at t who remain infectious

If we take P (t) to be the probability that an “infected” at time t0 = 0 remains

13
infectious at time t, then the idea above is given mathematically by
Z t
I(τ )
I(t) = I(0)P (t) + βS(τ ) P (t − τ )dτ (2)
0 N

Equation (2) is a non-linear Volterra Integral equation [7].

2.1.3 Equivalence of the Differential and Integral Form of the SIR Model

We said that selecting different forms of P (t) gives us alternative types of the SIR

model. Let us revisit the result in (1) by showing that if we assume P (t) as expo-

nentially distributed that is, P (t) = e−αt , then we recover the classical exponentially

distributed SIR model. Substituting P (t) = e−αt into (2) we have

Z t
−αt β
I(t) = I(0)e + S(τ )I(τ )e−α(t−τ ) dτ (3)
N 0

Differentiating (3) with respect to t, we have

β t
 Z 
dI β −αt −α(t−τ )
= SI − α I(0)e + S(τ )I(τ )e dτ
dt N N 0

which implies that

dI β
= SI − αI.
dt N

The probability of being recovered at time t is given by Q(t) = 1−P (t). Differentiating

it we have

Q0 (t) = αP (t)

which implies
dR
= αI
dt
14
 Finally, since S + I + R is constant, we know that S 0 + I 0 + R0 = 0, which in turn

implies that S 0 = −I 0 − R0 . This results in

dS β
= − SI
dt N

which completes the equivalence of the two forms of the SIR model formulation.

2.2 Phase Plane Analysis

In this section, we will study the relationship between our variables, S and I in the SI

plane, and S and R in the RS plane. This is called phase plane analysis. Moreover,

we will also study one of the very important elements of epidemiology called the

reproductive rate of infection, commonly denoted by R0 .

2.2.1 Phase Plane in SI Plane

The first two equations in (1) involves only S(t) and I(t). We examine them to study

what the solution looks like in the SI plane. Right at the beginning of the epidemic;

there is no “recovery”, i.e. R(0) = 0 , and let the number of “susceptibles” be given

by S0 and the number of “infecteds” be given by I0 . Also let S∞ be the number of

people who will not be “infected” throughout the epidemic. Thus S0 + I0 = N .

Dividing the first equation by the second in (1) we have

dI βSI − αI
=
dS −βSI
α
= −1
βS
dI ρ
= −1 (4)
dS S

15
 α
where ρ = β
is the relative removal rate. Equation (4) is called the Phase portrait

for the epidemic, and solutions to the phase portrait are called Phase curves [18] and

can be obtained as follows

Z 
ρ 
I= − 1 dS
S

= ρ ln(S) − S + C1

I = ρ ln(S) − S + N − ρ ln(S0 ) (5)

Figure 1: One Phase Curve in SI Plane

From Fig (1), we observe that

• the curve starts at (S0 , I0 ), rises to a maximum, and proceeds to S∞ . That

means the epidemic stops because of the lack of “infecteds”, not because every-

one will be sick.

16
 • the epidemic obtains a maximum somewhere between S0 and S∞ .

We can determine the maximum point as follows

dI ρ
= −1
dS S

Then, the critical value will be the solution of the equation

dI
=0
dS

which implies

ρ=S

To show that we have maximum value at ρ, we use the second derivative test. The

second derivative of I with respect to S is

d2 I ρ
2
=− 2 <0
dS S

which is negative. Hence, a maximum occurs at S = ρ as shown in Fig (2).

The maximum value is

 
ρ
Imax = ρ ln −ρ+N (6)
S0

2.2.2 Phase Plane Analysis in RS Plane

If we divide the second equation by the third one in (1), then we have

dS βSI
=− = −ρS
dR αI

17
 Figure 2: Phase Curves in SI Plane

which leads to
R
S(R) = S0 e− ρ (7)

Equation (7) implies that the RS phase curve exponentially decays, which means

that the number of “susceptibles” decays exponentially as a function of the number

of “recovereds” as shown in Fig (3) . To show mathematically our claim before that

in any epidemic there is always a part of the population which will not be “infected”

throughout the epidemic we use equation (7) as follows

R N
S(R) = S0 e− ρ ≤ S0 e− ρ > 0

which give us

0 < S∞ ≤ N

18
We know that if t → ∞, then R → R∞ and I∞ = 0 implies R∞ = N − S∞ . From

Figure 3: Phase Curves in RS Plane

this it follows that

R∞
S∞ = S0 e− ρ

N −S∞
S∞ = S0 e− ρ (8)

We can use this relation to calculate the value of S∞ and/or ρ if we know either of

the two.

2.2.3 The Basic Reproductive Rate of Infection R0

The basic reproductive rate, commonly denoted by R0 , is also called the basic repro-

ductive number or the basic reproductive ratio [2, 18]. It is basically the expected

(average) number of secondary cases produced by a typical primary case in an entirely

19
“susceptible” population. The reproductive rate of infection R0 is mathematically ex-

pressed as
S0
R0 =
ρ
which results in
βS0
R0 =
α
The reproductive rate R0 is important for prediction of whether there will be epidemic

or not [2, 18].

Theorem 2.1 (Threshold theorem of Epidemiology)

If S0 ≤ αβ , then I(t) is monotonically decreasing. If S0 > αβ , then an epidemic occurs

in that the number of “infected” will increase to a maximum before dropping to 0 over

time.

The threshold theorem can be interpreted in the following way

• if S0 ≤ ρ which gives us R0 ≤ 1, then there is no epidemic.

• if S0 > ρ which implies R0 > 1, then there will be epidemic.

Thus, stopping an epidemic is usually related to reducing R0 to be below 1.

Consider equation (5) with I∞ = 0 and S(∞) = S∞ . We have

I = ρ ln(S) − S + N − ρ ln(S0 ) (9)

I + S − ρ ln(S) = N − ρ ln(S0 ) (10)

If we take the limit as t goes to ∞, then we have

S∞ − ρ ln(S∞ ) = N − ρ ln(S0 )
20
Which results in
N − S∞
ρ=  
ln SS∞0
Alternatively, suppose that the only type of data available is from hospital reports

of relevant admissions and releases. This type of data commonly reports the number

of recovered. We can also determine the average rate of “recovery” from the report.

So it is reasonable to find R0 using the hospital data.

R
Using S(R) = S0 e− ρ , dR
dt
= αI and I + S + R = N ;

dR
= α(N − R − S)
dt
R
= α(N − R − S0 e− ρ )

from which we obtain

R
R0 − αN + αR = −αS0 e− ρ
R αN − αR − R0
e− ρ =
αS0

Taking the logarithm on both sides and simplifying, we have

−R
ρ=   (11)
αN −αR−R0
ln αS0

If we take the limit as t goes to infinity, then (10) will be equal to (9) since R0 ap-

proaches 0.

Let us consider the following example to illustrate the validity of our discussion.

We are going to use this example several times in the future discussion too [18].
21
EXAMPLE

A certain flu outbreak in English boarding school lasted for 14 days. The school has

a total of 763 students, of which 512 contracted the flu during that period. One boy

is known to have been the initial infected person thus giving us a given data set for

I(t).

We can calculate the value of S(t) numerically using the data. Thus, we have the

following values:

Table 1: The Number Of “Infected” And “Susceptible” Students

Days Infected Susceptible

0 1 762
1 3 758.99
2 7 749.55
3 25 721.20
4 72 645.4
5 222 493.48
6 282 306.89
7 256 171.12
8 233 99.21
9 189 64.18
10 123 46.51
11 70 36.99
12 25 31.54
13 11 28.27
14 4 26.23

Taking the following values for β, α and ρ as

α
β = 0.0021, α = 0.441, and ρ = = 210
β

22
we obtain a reproductive rate of infection

S0
R0 = = 3.63 > 1,
ρ

matching the fact that an epidemic did occur. Using (9),

763 − 25
ρ= = 215.9745.
ln 762
25

The phase curves in SI and RS are as follows:

1. The Phase curve in the SI plane is given by substituting, N = 763 and S0 = 762

in I = ρ ln(S) − S + N − ρ ln(S0 ).

Then, we have, I = 210 ln(S) − S + 763 − 210 ln(762) whose graphs is shown in

Fig (4).

2. Phase curves in the RS plane are obtained by taking the second and the third

equation in (1), and assuming S(0) = S0 = 762 and R(0) = 0.

we have the following:

S(R) = S0 e−R/ρ = 762e−R/210

Which gives us Figure (5).

23
Figure 4: Phase Curves in SI Plane

Figure 5: Phase Curves in RS Plane

24
 3 THE SIR MODEL WHEN S(t) IS A MULTI-EXPONENTIAL FUNCTION

As stated earlier, the non-linear Volterra integral equation (2) has no solution in

closed form. With appropriate assumptions on P (t), we have shown that the integral

equation is equivalent to a system of non-linear ordinary differential equations in (1).

Unfortunately, the system of non-linear ODE also has no solution in closed form.

The main reason that the equations in (1) and (2) cannot be solved in closed form is

because of the nonlinearity of S(t)I(t).

So, in some way, if we could somehow change the product S(t)I(t) to some form

which makes the solution simple, then we can solve (2) and as a result we can also

solve (1) in closed form. Because of the following two reasons we assume S(t) is a

multi-exponential function [12, 18]

• we know that the kernel of (2) is an exponential function. If we take S(t) to

be a multi-exponential function, then the expression under the integral will be

linear . Thus we can solve the integral equation easily.

• from experience we know that the “susceptible’s”’ data fits better with S(t) if

we take S(t) a multi-exponential function.

After having selected S(t) we solve (2) by changing it to an ODE. This assumption

of S(t) will make the ODE a linear ODE, which is simple to solve.

Let us first assume S(t) = a + nk=1 rk e−σk t and solve (2). Then, we use one
P

real example to illustrate two cases of S(t) – i.e., S(t) = a + be−σk t and S(t) =

a + 2k=1 rk e−σk t . Finally, we take the assumption that S(t) is a function of the form
P

a
S(t) = b+ewt
and solve (2).
25
 3.1 Finite Multi-Exponential Form of S(t)

Let us consider S(t) to be a multi-exponential function of the form S(t) = a +

Pn −σk t
k=1 rk e . With this S(t), equation (2) becomes
Z t n
!
β X
I(t) = I(0)e−αt + a+ rk e−σk τ I(τ )e−(t−τ ) dτ
N 0 k=1

Thus, the non-linear Volterra integral equation is changed to a linear one. To solve

it, we change it to an ODE of the form

n
! !
dI β X
= a+ rk e−σk t − α I(t)
dt N k=1
n
! !
dI β X
= a+ rk e−σk t − α dt
I N k=1

Integrating both sides leads to

Z Z n
! !
dI β X
= a+ rk e−σk t − α dt + C2
I N k=1
n
!
β X rk −σk t
ln |I(t)| = at − e − αt + C2
N k=1
σk

To solve for I(t), we change both sides of the above expression to exponential function,

and then we have the following

“ ”
β rk −σ t
at− n
P
e k −αt+C2
I(t) = e N k=1 σk

To determine the constant C2 , we use the initial condition I(0) = I0 .

26
 “P ”
−β n rk
+C2
I(0) = e N k=1 σk

Thus we have
“P ”
β n rk
C2
e = I0 e N k=1 σk

Hence we have the following for I(t)

“ ” “ ”
β rk −σ t β Pn rk
at− n
P
e k −αt+ N
I(t) = I0 e N k=1 σk k=1 σk

Our next step is to find the function R(t). We know that

dR(t)
= αI(t).
dt

But I(t) = N − R(t) − S(t), so that

R0 (t) = α (N − R(t) − S(t))

R0 (t) + αR(t) = α (N − S(t))

Thus, it is a non-homogeneous linear ODE which can be solved first by determining

an integrating factor µ(t) as

R
αdt
µ(t) = e = eαt

Using µ(t) = eαt , R0 (t) + αR(t) = α (N − S(t)) becomes

n
!!

0 X
eαt R(t) = αeαt N− a+ rk e−σk t
k=1

Integrating both sides of the above equation, we have

n
X rk
eαt R(t) = N eαt − aeαt − α e(α−σk )t + C3
k=1
α − σk

27
Dividing both sides by eαt we have
n
X rk
R(t) = N − a − α e−σk t + C3 e−αt
k=1
α − σk

Now it remains to find C3 . Using initial condition R(0) = R0 , we have

n
X rk
R(0) = N − a − α + C3
k=1
α − σk

which implies that

n
X rk
C3 = R0 − N + a + α .
k=1
α − σk

This completes our solution.

Let us revisit our example taking S(t) with one exponential function and sum of

two exponential functions.

3.1.1 Susceptibles as a Single Exponential

Let us assume S(t) as a function of the form S(t) = a+be−σt . Using the Fit command

in Maple software we fit the data with a curve of S(t) as shown in Fig (6). That gives

us the values of a, b, and σ. These values, as discussed and formulated in section

3.1, are used to determine the functions S(t) , I(t) and R(t). To have a better curve

which best fits our data, we divided the data into two and fit each of them with S(t)

type functions. That gives us two functions for each of S(t) , I(t) and R(t). These

functions approximate the number of “Susceptibles”, “Infecteds” and “Recovereds.”

respectively.

By what we formulated in Section 3.1, we have the following functions which give

28
 Figure 6: Data Fit Curve for “Susceptible” for One Exponential

us Fig (7)

S1 (t) = 762 − 2.948e0.904t

S2 (t) = 25 + 8820.803e−0.584t
0.904t +0.005
I1 (t) = e0.973t+0.007e
−0.584t +8.807
I2 (t) = e−0.389t−31.704e

R1 (t) = 1 + 0.967e0.904t − 1.967e−0.441t

R2 (t) = 763 + 27152.695e−0.584t − 18864.797e−0.441t

In addition, the Fit command and other statistics routines in Maple provide

measures of variance that can be used to validate the fits. Because the curves are

piecewise-defined, we did not measure confidence intervals for these fits. However,

correlations between the actual data and the predicted curves are each in excess of
29
0.98, and chi-square goodness of fit tests similarly confirm our results.

Figure 7: SIR Curves for One Exponential S(t)

3.1.2 Susceptibles as a sum of two Exponentials

Let us assume that S(t) is a function of the form

2
X
S(t) = a + rk e−σk t .
k=1

Using the Fit command in Maple software we fit the “Susceptibles” data with curves

of S(t). The fit gives us the corresponding values of a, r1 , r2 , σ1 , σ2 . Using the

discussion and formulation of section 3.1 we can determine functions S(t), I(t) and

R(t) . To have a better curve which best fits our data we divided the data into two

and fit each of them with two different S(t). This gives us two functions for each of

S(t) , I(t) and R(t) which approximates the number of “susceptibles”, “infecteds”

and “recovereds”, respectively. As a result we have Fig (8).

The following functions are the result of what we have discussed in Section 3.1

30
 Figure 8: Data Fit Curve for the “Susceptible” for Two Exponential

which results in Fig (9).

S1 = 762.5 − 0.596e1.323t + 2.230 × 10−7 e8.701t

5
S2 = 15 − 1000.752e−1×10 t + 3931.869e−0.446t
1.323t +1.455×10−20 e8.701t −0.454
I1 = e0.914t+0.005e
−1×105 t −18.545e−0.446t +9.352
I2 = e−0.409t+0.0002e
5
R1 = 674.679 − 0.0001e−8×10 t + 2.429e−13.370t − 11844.678e−0.569t

R2 = 13.893 + 0.115e1.323t − 2.478 × 10−19 e7.048t − 19.612e−0.445t

Once again, the goodness of fit test shows that the fit is reasonable. Also, the

correlation was again in excess of 0.97.

31
 Figure 9: SIR Curves for Two Exponential S(t)

3.2 Susceptible as a Logistic Function

a
Assuming S(t) = c + b+ewt
(i.e logistic function), we can solve the same non-linear

Volterra integral equation we have been solving in section 3.1.1 and section 3.1.2.

For this function

a
S(0) = c + , a = (S0 − c)(b + 1)
b+1
a
Let us consider (2) with S(t) = c + b+ewt
Z t 
−αt β a
I(t) = I(0)e + c+ I(τ )e−(t−τ ) dτ
N 0 b + ewτ

As we did before, we will change it to ODE and solve it. Thus,

dI β
= S(t)I(t) − αI(t)
dt N 
  
dI β a
= c+ − α dt
I N b + ewt

This gives us
 Z 
β a
ln |I| = ct + dt − αt + C4
N b + ewt
32
 du
If we late say u = b + ewt , then w(u−b)
= dt and we have the following integral:
 Z 
β a du
ln |I| = ct + − αt + C4
N w u(u − b)

With this replacement, we have

−1 1
1
= b + b
u(u − b) u u−b

from which we get

  
−1
Z Z
β a du 1 du
ln |I| = ct + + − αt + C4
N w b u b u−b

From this follows

β  a 
ln |I| = ct + (ln |u − b| − ln |u|) − αt + C4
N bw

Changing the above to exponential function and replacing the value of u back produces
“ “ wt ””
β a e
ct+ bw ln b+ewt −αt+C4
I(t) = e N

To determine the constant C4 we use the initial condition I(0) = I0

β a 1
eC4 = I0 e− N bw ln( b+1 )

Since we know I(t) and S(t) we can determine R(t) by

R(t) = N − I(t) − S(t)

which completes our solution.

Now we will consider the same example we have seen in section 3.1.1 and section

3.1.2. With c = S∞ = 25, and using command Fit in Maple we will have a function

33
 Figure 10: Best Fit Curve for Logistic Function

which best approximates our data with functions of our form as shown in Fig (10).

Then, we have the following functions which gives us Fig (11):

1.601 × 105
S(t) = 25 +
214.304 + e0.967t”
1.19e0.967t
“
−0.389t ln +0.005e0.967t
I(t) = e 214.217+e0.967t

1.601 × 105 1.19e0.967t

“ ”
−0.389t ln +0.005e0.967t
R(t) = 745.122 − 0.967t
−e 214.217+e0.967t
214.304 + e

Maple again not only reported the fit, but it also reported various measurements

of the goodness of the fit. The correlation is more than 0.98 and the chi-square

goodness of fit test confirmed our results. Also, we were able to obtain confidence

intervals for each of the parameters in the curve fits. Specifically, in S(t), we found

that the 95% confidence interval for w is [0.954, 0.989]. The 95% confidence interval

for a and b are nearly the same as the parameter values themselves. This shows that
34
 Figure 11: SIR Model Curves for Logistic Function

the logistic function form for S(t) gives good results for the SIR model.

3.2.1 The Basic Reproductive Rate of Infection R0

The data we usually have for research purposes are data we get from hospitals re-

ports. This type of data commonly reports the number of “recovered”. We can also

determine the average rate of recovery from the report. Using the following equations

from our previous discussion;

dS
= −βSI
dt
S+I +R = N

R0 = αI

35
 R0
Substituting I = α
in S = N − I − R, we have
R00 (t)
α
+ R0 (t)
β= R0 (t)
N− α − R(t)

We can use our example to show how we can apply the above equation. Fitting the

data with a function of the form

a
R(t) = − d.
b + ce−σt

we have the fit curve Fig (12)

Figure 12: Data Fit Curve for “Recovery” Data

From Fig (12) we have a = 99175.54 , b = 138.99 , c = 26540.56, σ = −0.709 and

d = −3.717. Thus,

99175.54
R(t) = − 3.717
138.99 + 26540.56e−0.709t
36
 Figure 13: The Graph of β as a Function of Time

Using,
R00 (t)
α
+ R0 (t)
β= R0 (t)
N− α − R(t)
We have β = 0.001509. The graph of β is shown in Fig (13). This leads to the

approximation ρ = 220.5. Thus R0 = 3.45578.

37
 4 CONCLUSION

The classical SIR epidemiological model is one of the earliest and most important

results of mathematical biology. It contains the most important features of epidemi-

ology namely “Susceptible”, “Infected” and “Recovered”. The basic SIR model has

its own limitation, but it is robust and can easily be extended. Depending on the

type of the disease under study, one can modify it to include some more aspects of

the disease. As a result we do have many variations of the model: SIS, SEIR, MSIR,

SIRS, SQIR.

Experiments with infectious disease in human populations is impossible and un-

ethical. Data are some times available from the natural occurrence of the epidemics.

Usually these data are incomplete and under reported. From experience, we know

that the data with all its weakness fits better with the results of this thesis’s functions

; S(t), I(t) and R(t). This gives us a way to closely study and interpret the data so

that we can make decisions to control a similar epidemic.

• As observed before, to find the appropriate multi-exponential S(t) = a +

Pn −σk t
k=1 rk e to fit our data better, we may need to partition the data to two

or more than two parts and fit each to a function and then we combine them.

a
• The other result discussed here is that of assuming S(t) = b+ewt
which is com-

monly called the logistic function. In this case, to find the function which best

fits the data, one should give initial values for variables a and b when using the

Fit command with Maple software.

38
 • After selecting S(t), one can use all the results of this thesis to study the

important features of the disease. As a result, it helps us to design the analysis

of epidemiological surveys, suggest crucial data that should be collected, identify

trends, and forecast and estimate uncertainty.

While we are working with this thesis we have observed that the data collected from

an epidemic fits nicely with the function of a difference of two logistic functions for

the “infectious” data. In case somebody has “infected” population data and wants

to know whether there was epidemic or not, he/she can start from that and do what

has been done in this thesis. In reality, most data reported from sources indicates

the “recovered” population number. One can also start from “recovered” populations

data and study the SIR model.

39
 BIBLIOGRAPHY

[1] Helen J.Wearing, Pejman Rohani and Mat J.Keeling, Appropriate Model For

The Management of Infectious Disease, PLoS Med 2, No.7, e174, 2005.

[2] Teri Johnson, Mathematical Modeling of Disease. Susceptible-Infected-Recovered

(SIR) model, Math 409 Senior Seminar, University of Minnesota , 2009.

[3] Vrushali A Boki, Lecture Note, Introduction to the Mathematics of Infectious

Disease, Oregon State University, Corvallis, Oregon, August 1, 2007

[4] David Smith and Lang Moore, The SIR model for spread of Disease,

www.math.duke.edu/education/ccp/materials/diffcalc/sir/index.html, Duke Uni-

versity, Last Accessed July 30, 2010.

[5] Modification of the SIR model, www.math.lsa.umich.edu/ tjacks/sirlecs.ppt.,

Last Accessed July 6, 2010.

[6] Abdul-Majid Wazwaz, A First Course in Integral Equation, World Scientific

Publishing, 15-20, 1997

[7] F. G. Tricomi, Integral Equation, Dover Publication Inc., 1-42, 1985

[8] Rainer Kress, Linear Integral Equation, Second Edition, Springer-Verlag Inc.,

1-13, 1999

[9] Ram P.Kanwal, Linear Integral Equations, Theory and Techniques, Second Edi-

tion, The Random house Birkhauser Mathematics Series, 25-36, 1997

40
[10] Ida Del Prete, Efficient Numerical Methods For Volterra Integral Equation of

Hammer Stein type, Universita Di Napoli Federico II, 1-16, 2006

[11] Herbert W.Hetthcote, Quantitative Analysis of Communicable Disease Model,

Mathematical Biosciences 28, 335-356 , 1976

[12] Christopher D. Green, Integral Equation Methods, Barnes and Noble Inc., 81-90,

1969

[13] M.Krasnov, A.Kieslev and G.Mkarenko, Problems and Exercises in Integral

Equations, Mir Publisher, 15-62, 1971

[14] Gustav Doestsch, Guide to the Application of the Laplace Transform and Z

Transform, Second Edition, Van Nostrand Reinhold Company, 118-121, 1971.

[15] Joel L.Schiff, The Laplace Transform Theory and Application, Springer-Verlag

Inc., 1-34, 1999.

[16] S.Patrak, A.Maiti, G.P. Samanata, Rich Dynamics of an SIR epidemic model,

Nonlinear Analysis, Modeling and Control 15, 71-81, 2010

[17] Michael Holile, Population Dynamics I, The SIR model, Department of Animal

Science and Health, Royal Veterinary and Agriculture University, 2002

[18] Fred Brauer and Carlos Castillo-Chavez, Mathematical Models in Population

Biology and Epidemiology, Springer-Verlag Inc., 273-332, 2001

41
 APPENDICES

Appendix MAPLE COMMANDS

1. TO PRODUCE FIGURE 6

dataplot := ScatterPlot(A, B, symbol = diamond, symbolsize = 20, color = blue);

S0 := Fit(762+C[1]*exp(-aa[1]*t), A[1 .. 6], B[1 .. 6], t, initialvalues = [C[1] = -

1.52181363299841910, aa[1] = 1.0852]);

S1 := Fit(25+C[1]*exp(-aa[1]*t), A[6 .. 13], B[6 .. 13], t, initialvalues = [C[1] = 100,

aa[1] = 10−5 ]);

l0 := plot(S0, t = 0 .. 5, color = green);

l1 := plot(S1, t = 5 .. 14, color = green);

d1 := display(l0, l1);

display(dataplot,lo,l1)

2. TO PRODUCE FIGURE 7

a1 := 762; b1 := -2.94757082275819516; sigma1 := -.903717448872219053;

a2 := 25; b2 := 8820.8029307937304; sigma2 := .584262909390753205;

I1 := exp(beta*a1*t/(1.132)+beta*b1*exp(-sigma1*t)/sigma1-alpha*t+beta*b1/(1.35*sigma1));

I2 := exp(beta*(a2*t-b2*exp(-sigma2*t)/sigma2)-alpha*t+beta*b2/(3.6*sigma2));

P1 := plot(I1, t = 0 .. 3.81, color = blue);

P2 := plot(I2, t = 3.81 .. 14, color = blue);

42
R2 := N-alpha*b2*exp(-sigma2*t)/(alpha-sigma2)+(-N+alpha*b2/(3/2*(alpha-sigma2)))*exp(-

alpha*t);

R12 := plot(R2, t = 6.1 .. 14);

R11 := plot(R1, t = 0 .. 6.2);

d := ScatterPlot(A, R, symbol = diamond, symbolsize = 20, color = blue);

display(d1, d2, d3, linestyle = solid, color = [red, blue, green]);

3. TO PRODUCE FIGURE 8

S011 := Fit(762.5+C11*exp(-a11*t)+C21*exp(-a21*t), A[1 .. 6], B[1 .. 6], t,

initialvalues = [C11 = -1.52181363299841910, a11 = 10.0852, C21 = -2.1, a21 = -10]);

S11 := Fit(15+C12*exp(-a11*t)+C22*exp(-a21*t), A[5 .. 14], B[5 .. 14], t,

initialvalues = [C12 = -1000.75181363299841910, a11 = 100000.0732, C22 = 3000,

a21 = -.9]);

dtplot := ScatterPlot(A, B, symbol = diamond, symbolsize = 20, color = blue);

l01 := plot(S011, t = 0 .. 4.2, color = green); l11 := plot(S11, t = 4.2 .. 14,

color = green); se := display(l01, l11);

g := 15; r11 := -1000.75181363299998; r21 := 3931.86878378646998;

σ 11 := 1.00000073199999999*105 ; σ 12 := .445677297008923612;

f := 762.5; r1 := -.595872937335464426; σ 21 := -1.32299589086598157;

r2 := 2.23037891409369903*10−17 ; σ 22 := -8.70076794567668976;

I21:=exp(.901*β*(f*t-2*r1*exp(-σ21*t)/σ 12-3*r2*exp(-σ 22*t)/σ 22)

-1.2*α*t+β*(r1/σ 21+r2/σ 22))-.45399;

43
I22 := exp(1.001*β*(g*t-r11*exp(-σ 11*t)/σ 11-r21*exp(-σ 12*t)/σ12)

-α*t+5.87*β*(r11/σ21+1/σ12));

u1 := plot(I22, t = 2.5 .. 14, color = blue);

u2 := plot(I21, t = 0 .. 2.5, color = blue);

Ie := display(u1, u2);

display(dtplot, l01, l11);

4. TO PRODUCE FIGURE 9

R77 := 1.0022*N-6*g-29*α*r11*exp(-8*σ11*t)/(2*N*(α-σ11))-α*r21*exp(-30*σ

12*t)/(200*N*(α-σ12))+(-(19/2)*N+(35/2)*g+(6*9)*α

r11/(2*N*(α-σ11))+10*α

r21/(N*(α-σ12)))*exp(-1.29*α*t);

R21 := 1.011*N-f-α*r1*exp(-.999987*σ21*t)/(1.3*(α-σ21))-α*

r2*exp(-.81*σ22*t)/(22.5*(α-σ21))+(-1.025*N+f+α*r1/(4*(α-σ21))

+α*r2/(α-σ22))*exp(-1.01*α*t)+5;

R9 := plot(R21, t = 0 .. 5.56, color = red);

R99 := plot(R77, t = 5.6 .. 14, color = red);

Ree := display(R9, R99);

display(Ie, Ree, se, linestyle = solid);

5. TO PRODUCE FIGURE 10

S4 := Fit(25+a/(b+exp(w*t)), A, B, t, initialvalues = [a = 200000, b = 1, w = .3]);

ScatterPlot(A, B, plot(S4, t = 0 .. 14));

44
 6. TO PRODUCE FIGURE 11

a1 := 1.60081881732383365∗105 ; b1 := 214.217265807576496; w1 := 0.967296187971025124;

I4 := exp(β*(25*t+a1*ln(1.19*exp(w1*t)/(b1+exp(w1*t)))/b1)-alpha*t-β*a1*ln(1/(b1+1))/b1);

ScatterPlot(A, Dat, plot(I4, t = 0 .. 14));

R4 := N-S4-I4+7.122246889;

plot([S4, I4, R4], t = 0 .. 14, linestyle = [solid, solid, solid], color = [red, blue, green]);

45
 VITA

TESHOME MOGESSIE BALKEW

Education: B.S. Mathematics, Addis Ababa University in Ethiopia,

Addis Ababa, Ethiopia, 1989
M.S. Mathematics, Addis Ababa University
Addis Ababa, Ethiopia, 1998
M.S. Mathematical Sciences, East Tennessee State University
Johnson City, Tennessee, 2010

Professional Experience: High School Teacher, Wachemo Cop.Sec.School In Ethiopia

Hossana, Ethiopia, 1989–1995
Lecturer, Alemaya University
Alemaya, Ethiopia, 1998-2002
Lecturer, Addis Ababa University
Addis Ababa, Ethiopia, 2002-2008
Graduate Assistant, East Tennessee State University
Johnson City, Tennessee, 2008-2010

46