7

Differential Equations*

*This chapter is partly based on Calculus for Biology and Medicine, by Claudia
Neuhauser.

A differential equation is a relationship between some (unknown) function

and its derivatives. Examples of differential equations were encountered
in Math 154 in the context of population growth, temperature of a cooling
object, and speed of a moving object subjected to friction. In Section 4.1, we
reviewed an example of a differential equation for velocity, Eqn. (4.8), and
discussed its solution. Here we present a more systematic approach to solving
such equations using a technique called separation of variables. In this
chapter, we apply the tools of integration to finding solutions to differential
equations.
In this course, since we are concerned only with functions that depend
on a single variable, we discuss ordinary differential equations (ODE’s),
(whereas later, after a multivariate calculus course where partial derivatives
are introduced, a wider class, of partial differential equations (PDE’s) can
be studied). Most laws of physics are of this form; for example, applying the
familiar Newton’s law, F = ma, links the position of a pendulum’s mass to its
acceleration (second derivative of position).
Note: Newton’s law states that force is proportional to acceleration. For
a pendulum, the force is due to gravity, and the acceleration is a second
derivative of the x or y coordinate of the bob on the pendulum.
Many biological processes are also described by differential equations.
The rate of growth of a population dN/dt depends on the size of that popula-
tion at the given time N (t ).
Constructing the differential equation that adequately represents a system
of interest is an art that takes some thought and experience. In this process,
which we call “modeling”, many simplifications are made so that the essen-
tial properties of a given system are captured, leaving out many complicating
details. For example, friction might be neglected in “modeling” a perfect
pendulum. The details of age distribution might be neglected in modeling
168 I N T E G R A L C A L C U L U S W I T H A P P L I C AT I O N S F O R L I F E S C I E N C E S

a growing population. Now that we have techniques for integration, we can

devise a new approach to computing solutions of differential equations.
Given a differential equation and a starting value, the goal is to make a
prediction about the future behaviour of the system. This is equivalent to
identifying the function that satisfies the given differential equation and
initial value(s). We refer to such a function as the solution to the initial
value problem (IVP). In differential calculus, our exploration of differential
equations was limited to those whose solution could be guessed, or whose
solution was supplied in advance.
Now that we have techniques of integration, we can find the analytic
solution to a variety of simple first-order differential equations (i.e. those
involving the first derivative of the unknown function). We describe the
technique of separation of variables. This technique works for examples
that are simple enough that we can isolate the dependent variable (e.g. y) on
one side of the equation, and the independent variable (e.g. time t) on the
other side.

7.1 Solving differential equations

Section 7.1 Learning goals

1. Use method of separation of variables to solve initial value problems.

2. Identify pure-time and autonomous equations.

3. Consider simple biological and physical applications and solve corre-

sponding initial value problems.

Throughout this chapter, we will restrict ourselves to first-order differen-

tial equations of the form
dy
= f (x)g(y). (7.1)
dx
The right-hand side of Eqn. (7.1) is the product of two functions, one de-
pending only on x, the other only on y. Such equations are called separable
differential equations. This type of differential equations includes two spe-
dy
cial cases: dx = f (x) (we call them pure time differential equations) and
dy
dx = g ( y ) (so-called autonomous equations).

Separation of variables
We begin with a general method for solving separable differential equations.
We divide both sides of Eqn. (7.1) by g(y) [assuming that g(y) = 0]:

1 dy
= f (x), (7.2)
g(y) dx
 D I F F E R E N T I A L E Q UAT I O N S 169

and then multiply both sides of Eqn. (7.2) by differential dx:

1 dy
dx = f (x)dx. (7.3)
g(y) dx

Now we can integrate both sides of Eqn. (7.3) with respect to x (recall that y
is a function of x):  
1 dy
dx = f (x)dx. (7.4)
g(y) dx
dy
Since dy = dx dx (we introduced differential notation in Section 6.1 in
Math 154), Eqn. (7.4) can be written in a simpler form
 
dy
= f (x)dx. (7.5)
g(y)

If one can evaluate the integrals, a general solution to differential equation

Eqn. (7.1) can be readily found.

Pure-time differential equations

In many applications, the independent variable (which we call here x) repre-
sents time. If the rate of change of a function depends only on time, we call
the resulting differential equation a pure-time differential equation. Such a
differential equation is of the form

dy
= f (x), x ∈ I, (7.6)
dx
where I is an interval and x represents time. Multiplying both sides of
dy dy
Eqn. (7.6) by dx: dx dx = f (x)dx and using differential notation dy = dx dx,
we can write the differential equation in the form

dy = f (x)dx,

and then integrate both sides:

 
dy = f (x)dx,

or simply 
y= f (x)dx + C. (7.7)

Example 7.1 Suppose that the volume V (t ) of a cell at time t changes

according to
dV
= sin t with V (0) = 3. V(t)
dt V(t)  4  cos t
6
Find V (t ). 5
4
Solution. We can use general solution Eqn. (7.7) where we need to replace 3
variable x by t and function f (x) by sin t. Then 2
1
0
p 2p t

Figure 7.1: The solution V (t ) = 4 − cos t in

Example 7.1.
170 I N T E G R A L C A L C U L U S W I T H A P P L I C AT I O N S F O R L I F E S C I E N C E S


V (t ) = sin t dt + C,

or
V (t ) = − cos t + C. (7.8)

We still need to evaluate the constant of integration C. Plugging in given

initial condition V (0) = 3 into Eqn. (7.8) leads to

3 = −1 + C ⇒ C = 4.

Finally, the solution of the initial value problem can be found as V (t ) =

4 − cos t (See Figure 7.1).


Autonomous differential equations

Many biological situations can be modelled using differential equations of the
form
dy
= g(y), (7.9)
dx
where the right-hand side does not explicitly depend on independent variable
x. These equations are called autonomous differential equations. Examples
of such equations were considered in Math 154 (exponential growth/decay
among them). We begin by dividing both sides of Eqn. (7.9) by g(y) and then
by multiplying both sides by dx, to obtain

dy
= dx.
g(y)

Integrating both sides then gives

 
dy
= dx.
g(y)

Note that we request g(y) = 0, otherwise we cannot use the method. We will
discuss two cases: g(y) = a − by and g(y) = k(y − a)(y − b). Both cases were
briefly addressed in Math 154 when studying the unlimited growth model
and the logistic growth model, correspondingly. But since we did not study
integration yet, our analysis was lacking an analytical part.

Case 1: g(y) = a − by. We wish to solve the initial value problem (IVP)

dy
= a − by, y(0) = y0 , (7.10)
dx
where a and b are constants. Separating variables then yields

dy
= dx, (7.11)
a − by
 D I F F E R E N T I A L E Q UAT I O N S 171

where we need to assume a − by = 0. Integrating both sides of Eqn. (7.11)

results in  
dy
= dx,
a − by
or
1
− ln |a − by| = x + C1 .
b
(We call the integration constant C1 here for convenience.) Exponentiating
both sides then leads to
|a − by| = e−b(x+C1 ) ,

or
|a − by| = e−bC1 e−bx .

Removing absolute value signs, we find

a − by = ±e−bC1 e−bx .

Now we can rename the constant by setting C = ±e−bC1 (this allows for a
more readable form of equation). We get

a − by = Ce−bx ,

or
a C −bx
y= − e .
b b
Recall that y(0) = y0 . Then y0 = ab − Cb and we arrive at the general solution
of IVP (7.10)
a a
y = + (y0 − )e−bx . (7.12)
b b

To obtain Eqn. (7.12), we divided both sides of Eqn. (7.10) by (a − by).
dy
We can do this only as long as y = ab . If y = ab though, then dx = 0 and the
constant function y = b is a solution of IVP (7.10). It means that we lost
a

this solution. This kind of constant solutions describes the situation when
dy
there is no change ( dx = 0) – the so-called steady state. They are also called
equilibrium solutions. We studied steady states in Math 154 in Chapter 14.
We address them again in Section 7.2.
We can make another important observation. When renaming the con-
stants by letting C = ±ebC1 , we assumed that C = 0 (exponential term can
never be equal to zero for any real numbers C1 and b). But we can still allow
C = 0, in addition to all other values. Then the solution Eqn. (7.12) will com-
bine all solutions obtained by direct solving of IVP (7.10) and will include
the equilibrium solution y = ab .

Example 7.2 Solve

dy
= 2 − 3y, if y(1) = 1.
dx
172 I N T E G R A L C A L C U L U S W I T H A P P L I C AT I O N S F O R L I F E S C I E N C E S

Solution. We can certainly use the solution (7.12), where a = 2 and b = 3.

But it is easier to solve the equation directly and there is no need to memorize
the formula in general.
Again, we separate variables and then integrate
 
dy
= dx.
2 − 3y

Recall that an antiderivative of 1

2−3y is (− 13 ) ln |2 − 3y|. Then

1
(− ) ln |2 − 3y| = x + C1 ,
3
ln |2 − 3y| = −3x − 3C1 ,
|2 − 3y| = e−3x−3C1 ,
2 − 3y = ±e−3C1 e−3x . y
10 y 2
3  13 e 3  3x
9
Setting a new constant C = ±e−3C1 will simplify the above equation to 8
7
6
2 − 3y = Ce−3x . 5
4
3
(1, 1)
To determine C, we use the initial condition y0 = 1 when x0 = 1. That is, 2
1
0
0 1 2x
2 − 3 = Ce−3 , or C = −e3 .
Figure 7.2: The solution curve in Exam-
Hence ple 7.2.
2 − 3y = −e3 e−3x ,

or
2 1 3−3x
y= + e .
3 3


Example 7.3 (Free fall) Assume that a body of mass m is dropped from
a great height above the ground. Find velocity of the body v(t ) given that
v(0) = 0, i.e. the body starts from the rest.
dv(t )
Solution. Let a(t ) = dt be the acceleration of the falling body and constant
γ denote the frictional coefficient due to air resistance. Then a force balance
equation for the falling object leads to

ma(t ) = mg − γv(t ).

For an object of constant mass, we can divide through by m, so

γ
a(t ) = g − v(t ).
m
Let k = γ/m. Then, the velocity at any time satisfies the differential equation
and initial condition
dv
= g − kv, v(0) = 0. (7.13)
dt
 D I F F E R E N T I A L E Q UAT I O N S 173

For convenience, we use just v in the equation although we remember that

velocity is a function of time t. We can find the solution to this differential
equation and predict the velocity at any time t using separation of variables.
Consider a time interval 0 ≤ t ≤ T , and suppose that, during this time
interval, the velocity changes from an initial value of v(0) = 0 to the final
value, v(T ) at the final time, T . Then using separation of variables and
integration, we get
dv
= g − kv,
dt
dv
= dt,
g − kv
 v(T )  T
dv
= dt.
0 g − kv 0

Note that in this example we use definite integrals. In solving this way we
will not need to find constant of integration. Substitute u = g − kv for the
integral on the left-hand side. Then du = −kdv, dv = (−1/k)du, so we get an
integral of the form 
1 1 1
− du = − ln |u|.
k u k
After replacing u by g − kv and applying the limits of integration, we arrive at
v ( T ) T
1  
− ln |g − kv| = t  ,
k 0 0

1
− (ln |g − kv(T )| − ln |g|) = T ,
k
  
1  g − kv(T ) 
− ln   = T,

k g
 
 g − kv(T ) 
ln   = −kT .

g
We are finished with the integration step, but the function we are trying
to find, v(T ) is still inside an expression involving the natural logarithm.
Extricating it involves some subtle reasoning about signs because there is an
absolute value to contend with. As a first step, we exponentiate both sides to
remove the logarithm.
 
 g − kv(T ) 
  = e−kT ⇒ |g − kv(T )| = ge−kT .
 g 

Because the constant g is positive, we could remove absolute value signs

from it. To simplify further, we have to consider the sign of the term inside
the absolute value in the numerator. In this case, v(0) = 0. This means that the
quantity g − kv(T ) will always be non-negative (i.e. g − kv(T ) ≥ 0). We verify
this fact shortly. For the moment, supposing this is true, we can write

|g − kv(T )| = g − kv(T ) = ge−kT ,

174 I N T E G R A L C A L C U L U S W I T H A P P L I C AT I O N S F O R L I F E S C I E N C E S

and finally solve for v(T ) to obtain our final result,

g
v(T ) = (1 − e−kT ).
k

Here we note that v(T ) can never be larger than g/k since the term (1 − e−kT )
is always ≤ 1. Hence, we were correct in assuming that g − kv(T ) ≥ 0.
As before, the above formula relating velocity to time holds for any choice
of the final time T , so we can write, in general,
g
v(t ) = (1 − e−kt ). (7.14)
k
This is the solution to the IVP (7.13). It predicts the velocity of the falling
object through time. Note that we have arrived once more at the result
obtained in Eqn. (4.11), but using the technique of separation of variables (it
often happens that a differential equation can be solved using several different
20
methods). velocity v
We graph the expression given in Eqn. (7.14) in Figure 7.3. Note that as t 15 terminal velocity
increases, the term e−kt decreases rapidly, so that the velocity approaches a
constant whose value is 10
g
v(t ) → .
k 5
We call this the terminal velocity.
time, t
Note: a similar plot of the solution of the Differential Equation (7.13) could 2 4 6 8 10
be assembled using Euler’s method, as studied in differential calculus. That
is the numerical method alternative to the analytic technique discussed in Figure 7.3: The velocity v(t ) as a function
of time given by Eqn. (7.14). Note that as
this chapter. The student may wish to review results obtained in a previous
time increases, the velocity approaches some
semester (see Ch. 12 of Math 154) to appreciate the correspondence. constant terminal velocity. The parameters
used were g = 9.8 m/s2 and k = 0.5.
Example 7.4 (Blood alcohol)

Let y(t ) be the level of alcohol in the blood of an individual during a

party. We consider two phases: “drinking” and “recovery”. These phases
are described by two different differential equations with their own initial
conditions. Suppose that the average rate of drinking is gradual and constant
(i.e. small sips are continually taken, so that the rate of input of alcohol is
approximately constant). Further, assume that alcohol is detoxified in the
liver at a rate proportional to its blood level. Then an equation of the form
Eqn. (7.10) would describe the blood level over the first stage, period of
drinking, and y(0) = 0 would signify the absence of alcohol in the body at the
beginning of the evening.
The constant a would reflect the rate of intake per unit volume of the
individual’s blood: larger people take longer to “get drunk” for a given
amount consumed. We are here assuming a constant intake rate, as though the
alcohol is being continually sipped all evening at a uniform rate. Most people
do not drink this way, instead quaffing a few large drinks over some hour(s).
It is possible to describe this, but we do not do so in this chapter.
 D I F F E R E N T I A L E Q UAT I O N S 175

The constant b represents the rate of decay of alcohol per unit time due
to degradation by the liver, assumed constant - this is also a simplifying
assumption, as the rate of metabolism can depend on other factors, such as
food intake; young healthy drinkers have a higher value of b than those who
can no longer metabolize alcohol as efficiently.
The solution in Eqn. (7.12) has several features of note: it illustrates
the fact that alcohol would increase from the initial level, but only up to a
maximum of a/b, where the intake and degradation balance. Indeed, the
level y = a/b represents a steady state level (as long as drinking continues).
Of course, this level could be toxic to the drinker, and the assumptions of the
model may break down! 1
Blood alcohol level
Next, in the phase of “recovery”, after drinking stops, the above differ- 0.8
ential equation no longer describes the level of blood alcohol. Instead, the
0.6
process of recovery is represented by
0.4
dy
= −by, y(0) = y0 . (7.15)
0.2
dt
The level of blood alcohol then decays exponentially with rate b from
1 2 3 4 5 6
its level at the moment that drinking ends. We show this typical pattern in
Figure 7.4. Recall that exponential decay equation was studied in Math 154 Figure 7.4: The level of alcohol in the blood
in Chapter 12. is described by Eqn. (7.10) for the first two
hours of drinking. At t = 2h, the drinking
Example 7.5 (Restricted Growth: von Bertalanffy Equation) stopped (so a = 0 from then on). The level
of alcohol in the blood then decays back to
zero, following Eqn. (7.15).
This example describes the simplest form of restricted growth and can be
used to describe the growth of fish. We denote by L(t ) the length of the fish
at age t and assume that L(0) = L0 . Then
dL
= k (A − L), (7.16)
dt
where A is a positive constant that we will interpret shortly. This equation
is also of the form of Eqn. (7.10), where a = kA and b = k. We assume that
L0 < A and explain this restriction subsequently as well. The constant k is
also positive; the equation says that the growth rate dL/dt is proportional to
A − L, so k should be interpreted as a constant of proportionality. We see that
the growth rate dL/dt is positive and decreases linearly with length as long
as L < A and that the growth stops (i.e., dL/dt = 0) when L = A. To solve
Eqn. (7.16), we separate variables and integrate, yielding
 
dL
= k dt.
A−L
Note that it was convenient to leave constant k on the right-hand side. Hence,

− ln |A − L| = kt + C1 .

After multiplying this equation by −1 and exponentiating, we obtain

|A − L| = e−C1 e−kt ,
176 I N T E G R A L C A L C U L U S W I T H A P P L I C AT I O N S F O R L I F E S C I E N C E S

or
A − L = Ce−kt
with C = ±e−C1 . Since L(0) = L0 , it follows that

A − L0 = C.

The solution is then given by

A − L(t ) = (A − L0 )e−kt ,
L(t)
A
or    
L0
L(t ) = A 1 − 1 − e−kt , (7.17)
A
(See Figure 7.5). L0

Eqn. (7.17) is called the von Bertalanffy equation. Since

t
lim L(t ) = A,
t→∞
Figure 7.5: The graph of the von Bertalanffy
the parameter A denotes the asymptotic length of the fish. Mathematically, equation.
there are no restrictions on L0 ; biologically, however, we require that 0 <
L0 < A. Otherwise, the growth rate would be negative, meaning that the fish
shrinks in size. Note that A is an asymptotic length that is never reached,
since there is no finite age T with L(T ) = A if L(0) < A.
We can now interpret the differential equation Eqn. (7.16): The growth
rate is proportional to the difference of the current length and the asymp-
totic length, with k representing the constant of proportionality. Since this
difference is decreasing over time, Eqn. (7.16) also shows that the growth
rate decreases over time, implying that juveniles grow at a faster rate than
adults. Moreover, the growth rate is always positive. This means that fish
grow throughout their lives, which is indeed the case.

Case 2: g(y) = k(y − a)(y − b). We now turn to differential equations of the
form
dy
= k(y − a)(y − b), (7.18)
dx
where k, a, and b are constant. We assume that k = 0. Separating variables and
integrating both sides yields
 
dy
= k dx, (7.19)
(y − a)(y − b)
provided that y = a and y = b. When a = b, we must find an antiderivative of
1
(y−a)2
, which is − y−a
1
. In this case,

1
− = kx + C,
y−a
or
1
y = a− .
kx + C
 D I F F E R E N T I A L E Q UAT I O N S 177

The constant C can then be determined from the initial condition. When y = a
or y = b, dy/dx = 0 and, consequently, y is equal to a constant.
In general case, to find the solution when a = b, we must use the partial-
fraction method that we introduced earlier in Section 6.6. We want to sim-
plify the integrand on the left-hand side of equation (7.19). We break it down
into a sum of simpler rational functions that we know how to integrate; that is,
we write the integrand in the form

1 A B
= + , (7.20)
(y − a)(y − b) y − a y − b
where A and B are constants that we must find. Next, we consider adding
back together two simple fractions on the right-hand side of Eqn. (7.20),

A B A(y − b) + B(y − a)
+ =
y−a y−b (y − a)(y − b)
(7.21)
y(A + B) − (Ab + Ba)
= .
(y − a)(y − b)
Comparing the last expression in Eqn. (7.21) with the left-hand side of
Eqn. (7.20), we conclude that

1 = y(A + B) − (Ab + Ba).

Therefore, the constants A and B must satisfy

A+B = 0 and Ab + Ba = −1.

Substituting B = −A from the first equation into the second equation above,
we find that
1
Ab − Aa = −1, or A(b − a) = −1, or A= .
a−b
Accordingly,
1
B=− .
a−b
We can put back evaluated constants A and B into Eqn. (7.20)

1 1 1 1 1
= · − · ,
(y − a)(y − b) a − b y − a a − b y − b
 
1 1 1 1
= − . (7.22)
(y − a)(y − b) a − b y − a y − b
Eqn. (7.22) allows us to evaluate the integral in the left-hand side of Eqn. (7.19)
   
dy 1 1 1
= − dy
(y − a)(y − b) a − b y−a y−b
1
= [ln |y − a| − ln |y − b|] + C1 .
a−b
178 I N T E G R A L C A L C U L U S W I T H A P P L I C AT I O N S F O R L I F E S C I E N C E S

Integrating the right-hand side of Eqn. (7.19) as well, and combining the
constants of integration into a new constant, C2 , we find that

1
[ln |y − a| − ln |y − b|] = kx + C2 ,
a−b
or  
y−a

ln   = k(a − b)x + C2 (a − b).
y−b
Exponentiating the latter equation yields
 
y−a
  C2 (a−b) k(a−b)x
y−b = e e ,

or
y−a
= ±eC2 (a−b) ek(a−b)x .
y−b
Defining a new constant C = ±eC2 (a−b) , we obtain the solution of Eqn. (7.18):
y−a
= Cek(a−b)x .
y−b

We solve this for y as follows:

y − a = (y − b)Cek(a−b)x ,
y = a + (y − b)Cek(a−b)x ,
y(1 −Cek(a−b)x ) = a − bCek(a−b)x ,

or
a − bCek(a−b)x
y= . (7.23)
1 −Cek(a−b)x
The constant C can then be determined from the initial condition. The
general solution given in Eqn. (7.23) is too complicated to memorize it.
We simply showed the procedure. When solving a particular problem, it is
more practical to solve it following the steps above.

Example 7.6 Solve initial value problem

dy
= 2(y − 1)(y + 2) with y0 = 2 when x0 = 0.
dx
Solution. Note that constant values y = 1 and y = −2 are equilibrium
solutions, when given differential equation describes a system in a steady
dy
state, with no change, i.e. dx = 0. Initial condition y(0) = 2 differs from
equilibrium solutions. Thus, we apply the process of separation of variables.
We can safely divide both sides of differential equation by (y − 1)(y + 2),
dy
multiply by dx and use dy = dx dx:
 
dy
=2 dx.
(y − 1)(y − 2)
 D I F F E R E N T I A L E Q UAT I O N S 179

We use partial fractions to integrate the left-hand side:

1 A B
= +
(y − 1)(y + 2) y − 1 y + 2
A(y + 2) + B(y − 1)
=
(y − 1)(y + 2)
(A + B)y + 2A − B
= .
(y − 1)(y + 2)

Therefore,

A+B = 0 and 2A − B = 1,

which yields

A = −B and 2A − B = 3A = 1.

Thus,
1 1
A= and B=− .
3 3
Using the partial-fraction decomposition, we must integrate
   
1 1 1
− dy = 2 dx.
3 y−1 y+2

This produces
1
[ln |y − 1| − ln |y + 2|] = 2x + C1 .
3
Simplifying the latter equation results in
 
y−1

ln   = 6x + 3C1 ,
y+2
 
y−1
 
y+2 = e e ,
3C1 6x

y−1
= ±e3C1 e6x ,
y+2
y−1
= Ce6x .
y+2

Recall that y0 = 2 when x0 = 0. Thus, we can find

1
C= .
4

The solution is therefore

y−1 1
= e6x .
y+2 4
180 I N T E G R A L C A L C U L U S W I T H A P P L I C AT I O N S F O R L I F E S C I E N C E S

If we want the solution in the form y = f (x), we must solve for y:

1
y − 1 = (y + 2) e6x ,
  4
1 6x 1 6x
y 1− e = e + 1,
4 2
2e +1
1 6x
y= ,
1 − 14 e6x
2e6x + 4 y
y= . 100 y 2e6x  4
4 − e6x 4  e6x
80
60
The solution curve is presented in Figure 7.6. 40
 20
0
0 0.05 0.1 0.15 0.2 x
7.2 Second look at steady states and stability
Figure 7.6: The solution for Example 7.6.

Section 7.2 Learning goals

1. Define a steady state.

2. Use qualitative and analytical approaches to analyze stability of steady

states.

We revisit the idea of steady states and their equilibria introduced in

Math 154 in Chapter 14.

Definition 7.1 (Steady state) A steady state is a state in which a system is

not changing. An equilibrium solution is a solution ŷ to differential equation
dy
dx = g(y) used in the model, which corresponds to a steady state, i.e. g(ŷ) =
0.

Definition 7.2 (Stability) We say that a steady state is stable, if states that
are initially close enough to that state will get closer to it with time. We
say that a steady state is unstable, if states that are initially very close to it
eventually move away from that steady state.

A physical analogue of steady states is provided in Figure 7.7. On the left

side, a ball rests on top of a hill; on the right side, a ball rests at the bottom of Figure 7.7: Stability illustrated with a ball
a valley. In either case, the ball is in equilibrium because it does not move. If on a hill and in a valley.

we move the ball out of its equilibrium slightly, the ball of the left side will
roll down the hill and not return to the top, whereas the ball on the right side
will return to the bottom of the valley. Thus, the left ball is in unstable and
the right ball is in stable equilibrium.
In general case, a differential equation may have a few steady states. Then
we classify equilibrium solutions as locally stable or locally unstable.
 D I F F E R E N T I A L E Q UAT I O N S 181

dy
In this section we consider autonomous differential equation dx = g(y). We
discuss its equilibrium solutions ŷ, g(ŷ) = 0 using graphical and analytical
approach. Study of equilibria is a powerful tool in predicting long-term be-
haviour of solutions. It is especially useful when we cannot solve differential
equation directly. g (y)

Graphical approach
We used this approach in Section 14 of Math 154. A brief summary is given 0 y1 y2 y
below.
Suppose that function g(y) is of the form given in Figure 7.8. Then 0, y1 ,
and y2 , points of intersection with horizontal axis, correspond to equilibrium
solutions ŷ = 0, ŷ = y1 , and ŷ = y2 . If we consider 0 < y < y1 , then g(y) > 0. Figure 7.8: The function g(y). The arrows
Positive derivative implies that y(x) is an increasing function. We draw an close to the equilibria indicate the type of
stability.
arrow on axis x pointing to the right, towards y1 . For y1 < y < y2 , g(y) < 0.
Since the derivative is negative, y(x) is a decreasing function. Then arrows
are pointing leftward, in the direction of y1 . Similarly, for y < 0 we draw the
arrow to the left and for y > y2 the arrow points to the right, away from y2 . We
examine each equilibrium separately.
ŷ = 0: arrows are pointing outward, away from this point of intersection.
Local equilibrium is unstable.
ŷ = y1 : arrows are directed toward it. Local equilibrium is stable.
ŷ = y2 : again, arrows are pointing away from this solution. Local equilib-
rium is unstable.

Analytical approach
Stability Criterion. Consider the differential equation
dy
= g(y),
dx
where g(y) is a differentiable function. Assume that ŷ is an equilibrium
solution; that is, g(ŷ) = 0. Then

ŷ is locally stable if g (ŷ) < 0,

ŷ is unstable if g (ŷ) > 0.

Proof of this criterion is rather complicated and it is beyond the scope of the
course.
Note that we require that function g(y) on the right side of differential
equation is differentiable. Also, we cannot make a conclusion about stability
of equilibrium solution in case when g (ŷ) = 0.
In the next two examples we use both approaches.

Example 7.7 Show that the differential equation

dy
= 1−y
dx
182 I N T E G R A L C A L C U L U S W I T H A P P L I C AT I O N S F O R L I F E S C I E N C E S

has a locally stable equilibrium at ŷ = 1. g(y)

Solution.
If we use a graphical approach (see Figure 7.9), we can see that arrows are
1
pointing toward ŷ = 1, hence it is a locally stable equilibrium. For analytical
approach, we need to find the derivative g (y) = −1. It is always negative, then
the Stability Criterion confirms that the steady state is locally stable. Also, we
can solve the equation exactly by separation of variables: y
0 1
 
dy
= dx,
Figure 7.9: The graph of g(y) in Exam-
1−y
ple 7.7.
− ln |1 − y| = x + C1 ,
1 − y = ±e−C1 e−x ,
1 − y = Ce−x .

Note that we set ±e−C1 = C. Solving for y yields

y(x) = 1 −Ce−x .

If we set y(0) = y0 , then y0 = 1 −C, and the solution is

y(x) = 1 − (1 − y0 )e−x .

We see that, for any initial value y0 ,

lim y(x) = 1.
x→∞ Rate q
Thus, our solution supports previous findings, reached by stability analysis.


Example 7.8 (Single compartment or pool)

Rate q
Compartment models are frequently used to model the flow of matter (nu-
trients, energy, and so forth). The simplest such model consists of one
compartment—for instance, a fixed volume V of water (such as a tank or
Figure 7.10: Input and output rates are the
lake) containing a solute (such as phosphorus). Assume that water enters the same: The water in the tank remains at the
compartment at a constant rate q and leaves the compartment at the same same level.
rate. (See Figure 7.10) (Having the same input and output rate keeps the
volume of the pool constant.) We will investigate the effects of different input
concentrations of the solution on the concentration of the solution in the pool.
We denote by C (t ) the concentration of the solution in the compartment at
time t. Then the total mass of the solute is C (t )V , where V is the volume of
the compartment. For instance, if the concentration of the solution is 2 grams
per liter and the volume of the compartment is 10 liters, then the total mass of
the solute in the compartment is 2 g liter-1 times 10 liters, which is equal to
20 g.
If CI is the concentration of the incoming solution and q is the rate at
which water enters, then qCI , the input loading, is the rate at which mass
 D I F F E R E N T I A L E Q UAT I O N S 183

enters. For instance, if the concentration of the incoming solution is 5 g

liter-1 and the rate at which the solution enters is 0.1 liter s-1 , then the input
loading—that is, the rate at which mass enters—is 5 g liter-1 times 0.1 liter
s-1 , which is equal to 0.5 g s-1 .
If we assume that the solution in the compartment is well mixed, so that
the outflowing solution has the same concentration as the solution in the CI C(t)
compartment—namely, C (t ) at time t—then qC (t ) is the rate at which mass
leaves the compartment at time t.
These different processes can be schematically illustrated in a flow dia-
gram, as shown in Figure 7.11. Because mass is conserved in the system, we Figure 7.11: Flow diagram for the single-
compartment model.
can use the law of conservation of mass to derive an equation that describes
the flow of matter in this system:
⎡ ⎤
rate of change    
⎢ ⎥ rate at which rate at which
⎣ of mass of ⎦= −
mass enters mass leaves
solute in pool

Writing C for C (t ) (and being careful not to confuse C with a constant),

we obtain
d
(CV ) = qCI − qC. (7.24)
dt
Since V is constant, we can write Eqn. (7.24) as
dC q
= (CI −C ). (7.25)
dt V
This is a linear differential equation of the type discussed in Section 7.1, Case
1, Eqn. (7.10). It can be solved by separation of variables. We skip the steps
C0  CI
and instead concentrate on the discussion of the system. If C (0) = C0 , then the
y
C0  CI
CI
solution of the differential equation is
   
C0 −(q/V )t
C (t ) = CI 1 − 1 − e . (7.26) CI

CI
Solution curves for different values of C0 are shown in Figure 7.12. From
Eqn. (7.25), we conclude that CI is the only equilibrium. Looking at the 0 x

solution C (t ) in Eqn. (7.26), we see that

Figure 7.12: The solution curves for the
lim C (t ) = CI , single-compartment model for different
t→∞ values of C0 .

regardless of the initial concentration C0 in the compartment. This shows that

CI is globally stable, which implies that no matter how much we perturb the
equilibrium, the system will always return to it.
We can also obtain this result by using the analytical approach. We can
write Eqn. (7.25) as
dC q
= g(C ), g(C ) = (CI −C ).
dt V
Then Ĉ = CI , g (C ) = − Vq < 0. Since the derivative is always negative,
g (Ĉ ) < 0, therefore it is a stable equilibrium.
184 I N T E G R A L C A L C U L U S W I T H A P P L I C AT I O N S F O R L I F E S C I E N C E S

dy
Example 7.9 Suppose that dx = y(4 − y). Find equilibria of this differen-
tial equation and discuss their stability, using both the graphical and the
analytical approach.

Solution. To find the equilibria, we set dy/dx = 0. That is,

g(y)
g(y)  y(4  y)
y(4 − y) = 0, 5
6

4
3
which yields 2

y1 = 0 and y2 = 4. 1

4 y

One the one hand, we set g(y) = y(4 − y) and graph g(y) (see Figure 7.13);
3
we see that y1 = 0 is a locally unstable equilibrium, while y2 = 4 is a locally
stable equilibrium. If we use the analytical approach, we need to find the
Figure 7.13: The graph of g(y) in Exam-
derivative of g(y):
ple 7.9.
g (y) = 4 − 2y.

Then g (0) = 4 > 0 and g (4) = −4 < 0, meaning that ŷ = 0 is a locally

unstable equilibrium and ŷ = 4 is a locally stable equilibrium, which confirms
our findings obtained with the help of the graphical approach.


7.3 Population growth models

Section 7.3 Learning goals

1. Re-visit the exponential growth model.

2. Learn the basic logistic growth model and its extensions in case of preda-
tion and threshold.

3. Use the qualitative approach when studying the Levins model.

We start with unlimited growth population model that was treated thor-
oughly in Math 154 and proceed to more realistic population growth models.

A simple model for exponential growth of population

Let N (t ) represent the size of a population at time t. We assume that at time
t = 0, the population level is specified, i.e. N (0) = N0 is some given constant.
Then
N (t ) = N (0)ert , t ≥ 0, r > 0.

Differentiating N (t ) with respect to time t, we found that the growth rate is

proportional to the current population size

dN
= rN (t ), t ≥ 0.
dt
 D I F F E R E N T I A L E Q UAT I O N S 185

Now we can solve this differential equation using integration technique of

separation of variables. Note that the constant r = N1 dN
dt is per capita growth
rate of the population and we can define it as a difference between per capita
birth rate k and per capita mortality rate m, that is

r = k − m.
N(t) r0
In case k > m there are more births than deaths and the population grows r0
r0
unlimited; k < m means that there are more deaths than births, so that the
population eventually goes extinct. There is also a marginal case when k = m,
for which r = 0, where the population does not change at all.
N0
We know already that unlimited growth can be found when individuals
are not limited by the availability of food or by competition. If we start a t
bacterial colony on a nutrient-rich substrate by inoculating the substrate with
Figure 7.14: Solution curves for dN/dt =
a few bacteria, then the bacteria initially can grow and divide unrestrictedly. rN.
Subsequently, when the substrate becomes more crowded and the food
source depleted, the growth will be restricted and a different differential
equation will be required to describe this situation. Also, exponential decay
in a population can be seen when the death rate exceeds the birth rate (for
instance, in cases of starvation).
The unrestricted growth model is sometimes referred to as Malthusian
growth, named after Thomas Malthus (1766–1834), a British clergyman and
economist. Malthus wrote about the consequences of unrestricted growth
on the welfare of humans. He claimed that populations and food show two
fundamentally different growth patterns: Populations grow exponentially and
food grows linearly. He concluded that, since exponential growth ultimately
overtakes linear growth, populations would eventually experience starvation.
As we know already, this is an unreasonable assumption since there are
factors that prevent a population from exceeding some limiting value. One
such model, the von Bertalanffy restricted growth model, was studied in
Example 7.5.

Density dependent growth

To correct for unrealistic unlimited explosive exponential growth, a common
assumption is that the growth rate is density dependent. The simplest such
model is called the logistic growth model. The size of population cannot grow
without bound. Other courses
The logistic growth model is studied in
The logistic equation was originally developed around 1835 by Pierre-
BISC 101 and 102: General Biology.
François Verhulst, who used the term logistic to describe the equation. His
work was completely forgotten until 1920, when Raymond Pearl and Lowell
J. Reed published a series of papers on population growth. They used the
same equation as Verhulst. After discovering Verhulst’s work, Pearl and
Reed adopted the name logistic for the equation. Pearl and Reed (1920) used
the logistic equation to predict future growth of the U.S. population on the
186 I N T E G R A L C A L C U L U S W I T H A P P L I C AT I O N S F O R L I F E S C I E N C E S

basis of census data from 1790 to 1920. Their equation would have predicted
about 185 million people in the United States in the year 2000, which is a
gross underestimate of the actual population size (over 260 million people).
Although the logistic equation does not seem to fit actual populations very
well, it is a useful model for analyzing growth under limiting resources.
The logistic equation describes the change in size of a population for
which per capita growth is density dependent. If we denote the population
size at time t by N (t ), then the change in growth is given by the initial-value
1 dN
1 dN N
problem   N dt
N dt
r 1 K
dN N
= rN 1 − with N (0) = N0 , (7.27)
dt K r
where r and K are positive constants. This is the simplest way of incorpo-
rating density dependence in the per capita growth rate, namely, having it
decrease linearly with population size (see Figure 7.15):
 
1 dN N K N
= r 1− .
N dt K
We can interpret the parameter r as follows: The per capita growth rate
Figure 7.15: The per capita growth rate in
is equal to r when N = 0. Therefore, one way to measure r is to grow the the logistic equation is a linearly decreasing
organism at a very low density (i.e., when N is much smaller than K), so that function of population size.

the per capita growth rate is close to r.

The quantity K is called the carrying capacity. Looking at Figure 7.15,
we see that the per capita growth rate is 0 when the population size is at
the carrying capacity. Since the per capita growth rate is positive below K
and negative above K, the size of the population will increase below K and
decrease above K. The number K thus determines the population size that
can be supported by the environment. To show this claim mathematically, we
need to solve Eqn. (7.27). From its solution we will then see that, starting
from any positive initial population size, the size of the population will
eventually reach K [i.e., limt→∞ N (t ) = K if N (0) > 0].
If we write right-hand side of the differential equation as
r
g(N ) = − (N − 0)(N − K ), (7.28)
K
and compare Eqn. (7.28) with the right-hand side of Eqn. (7.18), we find that
r
k = − , a = 0, b = K. (7.29)
K
The solution of Eqn. (7.27) can be found using Eqn. (7.23), where the
constants a, b, and k are given by Eqn. (7.29)
0 − KCe−(r/K )(0−K )t
N (t ) = ,
1 −Ce−(r/K )(0−K )t
or
−CKert CK
N (t ) = =
1 −Cert C − e−rt
(7.30)
K K
= −rt
= ,
1 − e /C 1 −C1 e−rt
 D I F F E R E N T I A L E Q UAT I O N S 187

where C1 = C1 .
While we used the shortcut to find the solution, it was mainly the demon-
stration how useful can be the general solution. We can solve the Eqn. (7.27)
following the same procedure as before, using the separation of variables.
Then re-writing Eqn. (7.27) as
dN r
= − N (N − K ),
dt K
and separating variables, we get
1 dN r
=− . (7.31)
N (N − K ) dt K

Note that it is more efficient to leave the constants in the right-hand side
of Eqn. (7.31). Multiplying both sides of Eqn. (7.31) by dt and using the
differential dN = dN
dt dt, we can integrate
 
1 r
dN = − dt. (7.32)
N (N − K ) K

Next we use the partial fractions

1 A B
= + ,
N (N − K ) N N −K

1 = A(N − K ) + BN,
1 = (A + B)N − AK,
A + B = 0, 1 = −AK.
Finally,
1 1
A=− , B= .
K K
The left-hand side of Eqn. (7.32)
   
1 1 1 1 1
= − · + · dN
N (N − K ) K N K N −K
 
1 1 1  N − K 
=− ln |N| + ln |N − K| = ln  + C1 . (7.33)
K K K N 
The right-hand side of Eqn. (7.32) can be re-written as

r r
− dt = − t + C2 . (7.34)
K K
Using Eqn. (7.33) and Eqn. (7.34) and combining the constants of integration,
we get  
1  N − K  r
ln = − t + C3 ,
K  N  K
or  
N −K 

ln   = −rt + C3 K,
N 
188 I N T E G R A L C A L C U L U S W I T H A P P L I C AT I O N S F O R L I F E S C I E N C E S

 
N −K 
  −rt C3 K
 N  = e ·e ,
N −K
= ±eC3 K · e−rt ,
N
N −K
= Ce−rt ,
N
N − K = NCe−rt ,
N − NCe−rt = K,
N (1 −Ce−rt ) = K,
and lastly
K
N= . (7.35)
1 −Ce−rt
Our constructed solution of the differential equation (7.27), Eqn. (7.35), is
exactly the same as the solution Eqn. (7.30), derived above by means of the
general solution.
Since N (0) = N0 , we have
K
N0 = .
1 −C
Solving this equation for C yields

N0 (1 −C ) = K or −CN0 = K − N0 , CN0 = N0 − K.

Thus,
N0 − K
C= . (7.36)
N0
Substituting Eqn. (7.36) into Eqn. (7.35), we find that
K
N (t ) = ,
1 − N0N−K
0
e−rt

or
K
N (t ) =   . (7.37)
1+ K
N0 − 1 e−rt N(t)
It follows from Eqn. (7.37) that N0  K

lim N (t ) = K. K
t→∞
K
Some typical solution curves of the logistic equation are shown in Fig- 2
0  N0  K/2
ure 7.16. We can see that the solution curves approach K, the carrying 0
0 t
capacity, as t → ∞, from above, if N0 > K or from below, if N0 < K. Recall
that we could proceed with our solution technique only under the assumption Figure 7.16: Solution curves for different
that N = 0 and N = K. It is worth noting that the solution (7.37) contains initial values N0 .

the constant solution N = K if we choose N0 = K. If N0 = 0, then nothing

happens; that is, N (t ) stays equal to 0 for all later times. That makes sense:
If there aren’t any individuals to begin with, then there won’t be any later
 D I F F E R E N T I A L E Q UAT I O N S 189

on. We can show that if 0 < N0 < K2 , then the solution curve is S-shaped, as
seen in Figure 7.16. An S-shaped curve is characteristic of populations that
show this type of density-dependent growth. At low densities, the growth is
dN
almost like unrestricted growth. At higher densities, the growth is restricted
dt
and the curve bends around and eventually levels off at the carrying capacity.
If the population size is initially greater than the carrying capacity K, the
population size decreases and becomes asymptotically (i.e., when t → ∞) equal
to K. Sometimes S-shaped curves are called sigmoidal.
We can also achieve the same conclusions about behavior of solutions
using qualitative analysis. Then N = 0 and N = K are equilibrium solutions.
0 K N
As we can see from the Figure 7.17, N = 0 is a locally unstable equilibrium
(low density populations are growing), while N = K is a locally stable equilib-
dN
Figure 7.17: The graph of versus N.
rium (eventually, a population will be either increasing or decreasing towards dt

N = K in the long run). Also, to practice our knowledge of analytical approach

in steady states, we find the derivative of the right-hand side function in
differential equation, Eqn. (7.28)
r
g (N ) = − (2N − K ),
K
and calculate it in the equilibrium solutions N̂ = 0 and N̂ = K:
g (0) = r > 0, g (K ) = −r < 0,
thus confirming our conclusions about their stability we made using graphical
approach.
It is worth mentioning that if N0 > K2 , then the solution curve is no longer S-
shaped. It resembles then the hill functions, introduced in Math 154, Chapter
2, without an inflection point on its graph. Still, limt→∞ N (t ) = K in all cases
considered, irrespective of the value of N0 > 0.
In summary, we have shown that the behavior of the logistic equation
solutions describe the population growth in a more realistic way than the
unlimited growth model. Yeast, a microscopic fungus used to make bread and
alcoholic beverages, exhibits the classical S-shaped curve when grown in a
test tube. Once the population depletes available nutrients, its growth levels
off. Obviously, if we plot experimental data, some fluctuations occur. Certain
techniques need to be used to approximate the data with the smooth curve.
Still, even with oscillations, the logistic model is confirmed.
Example 7.10 (A simple model of predation) Suppose that N (t ) denotes
the size of a population at time t. The population evolves according to the
logistic equation, but in addition, predation reduces the size of the population
so that the rate of change is given by
 
dN N 9N
= N 1− − ,
dt 50 5+N
where the first term of the right-hand side describes the logistic growth and
the second term–effect of predation. Find all equilibria and determine their
stability.
190 I N T E G R A L C A L C U L U S W I T H A P P L I C AT I O N S F O R L I F E S C I E N C E S

Solution. Set  
N 9N
g(N ) = N 1 − − .
50 5+N
Then to find equilibria, we must solve g(N ) = 0, that is
 
N 9N
N 1− − = 0,
50 5+N
 
N 9
N 1− − = 0,
50 5 + N
N 9
N̂ = 0, 1− − = 0,
50 5 + N
50(5 + N ) − N (5 + N ) − 450 = 0,

250 + 50N − 5N − N 2 − 450 = 0,

−N 2 + 45N − 200 = 0,

N 2 − 45N + 200 = 0,

N̂ = 5, N̂ = 40.

We found three equilibrium solutions. We use analytical approach to discuss

their stability because graphing g(N ) is not as straightforward as in the
9(5+N )−9N
previous example. At first we calculate g (N ) = 1 − 25 N
− (5+N )2 , or after
3 −15N 2 −225N +500
simplifications, g (N ) = − 25
1 N
· (5+N )2
. Then g (0) = − 45 < 0,
g (5) = 20
7
> 0, and g (40) = − 2845 < 0. We conclude that N̂ = 0 and N̂ = 40
are locally stable equilibria, while N̂ = 5 is a locally unstable equilibrium.


The Levins model

The ecological importance of spatial structure to the maintenance of pop-
ulations was pointed out by Andrewartha and Birch (1954) on the basis of
studies of insect populations. They observed that, although local populations
frequently become extinct, their patches of habitat subsequently become
recolonized by migrants from other patches occupied by individuals from
the same kind of population, thus allowing the population to persist globally.
Fifteen years later, Richard Levins introduced the concept of metapopula-
tions (Levins, 1969). A major theoretical advance, the concept provided a
framework for studying spatially structured populations.
A metapopulation is a collection of subpopulations. Each subpopulation
occupies a patch, and different patches are linked via the migration of individ-
uals between patches. (See Figure 7.18.) In this setting, we keep track only
Figure 7.18: A schematic description of a
of what proportion of patches is occupied by subpopulations. Subpopulations metapopulation model. The shaded patches
go extinct at a constant rate, denoted by m, which stands for mortality. Vacant are occupied; arrows indicate migration
events.
patches can be colonized at a rate that is proportional to the fraction of occu-
pied patches; the constant of proportionality is denoted by c, which stands
 D I F F E R E N T I A L E Q UAT I O N S 191

for colonization rate. If we denote by p(t ) the fraction of patches that are
occupied at time t, then writing p = p(t ), we have

dp
= cp(1 − p) − mp. (7.38)
dt
The first term on the right-hand side describes the colonization process. Note
that an increase in the fraction of occupied patches occurs only if a vacant
patch becomes occupied–hence the product p(1 − p) in the first term on
the right-hand side. The minus sign in front of m shows that an extinction
decreases the fraction of occupied patches. We will not solve Eqn. (7.38);
instead, we focus on its equilibria. We set

cp(1 − p) − mp = 0.

Isolating the factor cp, we obtain

 m 
cp 1 − − p = 0,
c
which has the two solutions
m
p1 = 0 and p2 = 1 − .
c
We call the solution p1 = 0 a trivial solution, because it corresponds to the
situation in which all patches are vacant. Since individuals are not created
spontaneously, a vacant patch can be recolonized only through migration
from other, occupied patches. Therefore, once a metapopulation is extinct, dp
cp(1  p)  mp
dt
it stays extinct. The other equilibrium, p2 = 1 − m/c, is relevant only when
0
p2 ∈ (0, 1], because p represents a fraction that is a number between 0 and 1p
1. Since m and c are both positive, it follows immediately that p2 < 1 for all
choices of m and c. To see when p2 > 0, we check

m
1− > 0,
c
which holds when Figure 7.19: Graph of d p/dt versus p, the
case m > c.
m < c.
dp
That is, the nontrivial equilibrium p2 = 1 − m/c is in (0, 1] if the extinction dt cp(1  p)  mp

rate m is less than the colonization rate c. If m ≥ c, then there is only one 0
equilibrium in [0, 1], namely, p1 = 0. We illustrate this scenario in Figures 7.19 1
m
c 1p
and 7.20; looking at the graphs, we can analyze the stability of the equilibria.
Case 1: m > c. There is only the trivial equilibrium p1 = 0. For any p ∈ (0, 1],
d p/dt < 0; hence, the fraction of occupied patches declines. The equilibrium
is locally and globally stable.
Case 2: m < c. There are two equilibria: 0 and 1 − m/c. The trivial equilib- Figure 7.20: The case m < c.
rium p1 = 0 is now unstable, since, if we perturb p1 = 0 to some value in
(0, 1 − m/c), then d p/dt > 0, which implies that p(t ) increases. The system
will therefore not return to 0.
192 I N T E G R A L C A L C U L U S W I T H A P P L I C AT I O N S F O R L I F E S C I E N C E S

The other equilibrium, p2 = 1 − m/c, is locally stable. After a small pertur- p(t)
Levins Model
bation of this equilibrium to the right of p2 , d p/dt < 0; a small perturbation to 2
the left of p2 gives d p/dt > 0. Therefore, the system will return to p2 . 1.6
We can also use the analytical approach to analyze the stability of the 1.2
equilibria. In addition, this approach will allow us to obtain information on 0.8
how quickly the system returns to the stable equilibrium. We set 0.4
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 t
g( p) = cp(1 − p) − mp.
We must find Figure 7.21: Solution curve when m > c:
 m = 2, c = 1, and p(0) = 2. The solution
g ( p) = c − 2cp − m. approaches the locally stable equilibrium
p1 = 0.
Now, if p1 = 0, then
g (0) = c − m, Levins Model
p(t)
whereas if p2 = 1 − m/c, then 0.5
 m  m
g 1 − = c − 2c 1 − − m = c − 2c + 2m − m = m − c. 1
c c
We find that 0.5

if c − m < 0, then p1 = 0 is locally stable, and 0

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 t
if c − m > 0, then p1 = 0 is unstable and p2 = 1 − mc is locally stable.
Figure 7.22: Solution curves when m < c:
To summarize our results, if m > c, then p1 = 0 is the only equilibrium in m = 1, c = 2, and p(0) = 0.1, 0.5, 1.5. When
p(0) = 0.1, the solution approaches the
[0, 1] and p1 = 0 is locally stable. (In fact, the graphical analysis showed that
locally stable equilibrium p2 = 1 − mc = 0.5.
p1 = 0 is globally stable.) Figure 7.21 shows a solution curve when m = 2 When p(0) = 0.5, the solution remains
and c = 1, starting at p(0) = 2. We see that the solution approaches p1 = 0. at the locally stable equilibrium p2 =
1 − m/c = 0.5. When p(0) = 1.5, the
If m < c, then there are two equilibria in [0, 1]. The equilibrium p1 = 0 is solution approaches the locally stable
now unstable, and the equilibrium p2 = 1 − m/c is locally stable. Figure 7.22 equilibrium p2 = 1 − m/c = 0.5.
shows solution curves when m = 1 and c = 2, starting from different initial
conditions: p(0) = 0.1, 0.5, and 1.5. All solution curves eventually approach
p2 = 1 − m/c = 0.5. When p(0) = p2 = 0.5, the solution curve stays at
p2 = 0.5.

The Allee effect

A sexually reproducing species may experience a disproportionately low
recruitment rate when the population density falls below a certain level, due
to lack of suitable mates. This phenomenon is called an Allee effect (Allee,
1931). A simple extension of the logistic equation incorporates the effect.We
denote the size of a population at time t by N (t ); then, writing N = N (t ), we
have  
dN N
= rN (N − a) 1 − , (7.39)
dt K
where r, a, and K are positive constants. We assume that 0 < a < K. We
will see that, as in the logistic equation, K denotes the carrying capacity. The
constant a is a threshold population size below which the recruitment rate
 D I F F E R E N T I A L E Q UAT I O N S 193

is negative, meaning that the population will shrink and ultimately go to dN dN

dt  g(N )
extinction. The equilibria of Eqn. (7.39) are given by N̂ = 0, a, and K. We set dt

   
N a 2 N3
g(N ) = rN (N − a) 1 − =r N + N −
2
− aN . 0
K K K a K N
A graph of g(N ) is shown in Figure 7.23. Differentiating g(N ) yields
 
 2a 3N 2 r
g (N ) = r 2N + N − − a = (2NK + 2aN − 3N 2 − aK ).
K K K
Figure 7.23: The graph of g(N ) illustrating
We can compute the Allee effect.

r
g (0) = (−aK ) < 0,
K
r
g (a) = a(K − a) > 0, y
K
r 7
g (K ) = K (a − K ) < 0. 6

Population Size
K 5

Since g (0) < 0, it follows that N̂ = 0 is locally stable. Likewise, since 4

3
g (K ) < 0, it follows that N̂ = K is locally stable. The equilibrium N̂ = a is 2
unstable, because g (a) > 0. This instability is also evident from Figure 7.23. 1
0
The Allee effect is an example in which both stable equilibria are locally, but 0 1 2 3 4 5 6 7 8 9 10 x
Time
not globally, stable.
We see from Figures 7.23 and 7.24 that if 0 ≤ N (0) < a, then N (t ) → 0 as Figure 7.24: Solution curves when r = 0.5,
t → ∞. If a < N (0) ≤ K or N (0) ≥ K, then N (t ) → K as t → ∞. To interpret a = 2, and K = 5. When the initial condition
N (0) is between 0 and 2, the solution curve
our results, we observe that if the initial population N (0) is too small [i.e.,
approaches the locally stable equilibrium
N (0) < a], then the population goes extinct, and if the initial population N̂ = 0. When the initial condition N (0) is
is large enough [i.e., N (0) > a], then the population persists. That is, the greater than 2, the solution curve approaches
the locally stable equilibrium N̂ = K = 5.
parameter a is a threshold level. The recruitment rate is large enough only The approach is from below when 2 <
when the population size exceeds this level. N (0) < 5 and from above when N (0) > 5.

7.4 Systems of autonomous equations

Section 7.4 Learning goals

1. Extend the analytical approach in the case of the system of two au-
tonomous differential equations modelling the spread of an infectious
disease.

2. Use the qualitative approach to study a two-compartment model.

In Section 7.3, we discussed models that could be described by a single

differential equation. If we wish to describe models in which several quanti-
ties interact, such as a competition model in which various species interact,
more than one differential equation is needed. We call this model a system
of differential equations. We will restrict ourselves again to autonomous
systems–that is, systems whose dynamics do not depend explicitly on the
independent variable (which typically is time).
194 I N T E G R A L C A L C U L U S W I T H A P P L I C AT I O N S F O R L I F E S C I E N C E S

A simple model of epidemics

We begin our discussion of systems of autonomous differential equations
with a classical model of an infectious disease: the Kermack–McKendrick
model (Kermack & McKendrick, 1927, 1932, 1933). We consider a popu-
lation of fixed size N that, at time t, can be divided into three classes: the
susceptibles, S(t ), which can get infected; the infectives, I (t ), which are
infected and can transmit the disease; and the removed class, R(t ), which
have recovered and are immune to the disease. The flow among these classes
can be described by
S → I → R.
We assume that the infection spreads according to the mass action law that
we encountered in the discussion of chemical reactions. Each susceptible be-
comes infected at a rate that is proportional to both the number of infectives
I and the number of susceptibles S. Each infected individual recovers at a
constant rate. A gain in the class of infectives is a simultaneous loss in the
class of susceptibles. Likewise, a gain in the class of recovered individuals is
a loss in the class of infectives. We can therefore describe the dynamics by
dS
= −bSI, (7.40)
dt
dI
= bSI − aI, (7.41)
dt
dR
= aI. (7.42)
dt
Note that
dS dI dR
+ + = −bSI + bSI − aI + aI = 0.
dt dt dt
Since
dS dI dR d
+ + = (S + I + R),
dt dt dt dt
it follows that S(t ) + I (t ) + R(t ) is a constant that we can identify as the
population size N which is constant by assumption (nobody dies from this
disease) and the model reproduces this. To analyze the system, we assume
that, at time 0,
S(0) > 0, I (0) > 0, R(0) = 0.
A question of interest is whether the infection will spread. We say that the
infection spreads if
I (t ) > I (0) for some t > 0.
Equation (7.41) allows us to answer the question: If
dI
= I (bS − a) > 0 at t = 0,
dt
then I (t ) increases at the beginning, and hence the infection can spread. This
condition can be written as
bS(0)
> 1.
a
 D I F F E R E N T I A L E Q UAT I O N S 195

The value bS(0)/a is called the basic reproductive rate of the infection and
bS(0)
is typically denoted by R0 = a [not to be confused with the number R(0)
of recovered individuals at time 0]. The quantity R0 is of great importance
in epidemiology, because it tells us whether an infection can spread. R0 is
the key to understanding why vaccination programs work. It explains why
it is not necessary to vaccinate everyone against an infectious disease: As
long as the number of susceptibles is reduced below a certain threshold, the
infection will not spread. The theoretical threshold, based on the Kermack–
McKendrick model, is a/b. In practice, there are additional factors (such as
the spatial proximity of infected individuals) that influence whether or not an
infection will spread. But the basic conclusion is the same: As long as the
number of susceptibles is below a certain threshold, the infection will not
spread.
To find out how the infection progresses over time, we divide Eqn. (7.40)
by Eqn. (7.42). For I > 0, we obtain
dS/dt dS bSI b
= =− = − S.
dR/dt dR aI a
That is,
dS b
= − S.
dR a
Separating variables and integrating yields
 
dS b
=− dR.
S a
Since S(t ) > 0, we have
b
ln S(t ) = − R(t ) + C.
a
With R(0) = 0, we obtain S(t)
ln S(0) = C.
S(0)
Hence,
S(t ) = S(0)e−(b/a)R(t ) . (7.43)
Since R(t ) is nondecreasing (no one can leave the removed class), it
0
follows from Eqn (7.43) that S(t ) is nonincreasing. (See Figure 7.25). 0 t
Letting t → ∞, we find that
Figure 7.25: The solution S(t ) of the
Sin f = lim S(t ) = lim S(0)e−(b/a)R(t ) Kermack–McKendrick model.
t→∞ t→∞

= S(0)e−(b/a) limt→∞ R(t ) . I(t)

Since limt→∞ R(t ) ≤ N, it follows that

I(0)
Sin f ≥ S(0)e−(b/a)N > 0.

That is, not everyone becomes infected. As long as the number of suscep-
0
tibles is greater than a/b, dI/dt > 0 and the number of infected individuals 0 t

Figure 7.26: The solution I (t ) of the

Kermack–McKendrick model.
196 I N T E G R A L C A L C U L U S W I T H A P P L I C AT I O N S F O R L I F E S C I E N C E S

increases. Because the population size is constant, the infection “uses up”
susceptibles, and there will be a time when the number of susceptibles falls
below a/b. (See Figure 7.26) From then on, dI/dt is negative, and the num-
ber of infected individuals begins to decline. The infection will eventually
cease; that is,
lim I (t ) = 0.
t→∞

Since not everyone becomes infected [Sin f > 0], when the infection finally
comes to an end, it is because the population runs out of infectives, not
because of a lack of susceptibles.

Example 7.11 Assume the simple model of epidemics where a = 100,

b = 0.01, and N = 12, 000. Can the disease spread if, at time t = 0, there is
only one infected individual?

Solution. Since in the population of 12,000 individuals there is one infective,

then the initial number of susceptibles S(0) = 11, 999 and then

bS(0) 0.01 · 11, 999

R0 = = = 1.1999 > 1.
a 100
NI
Therefore, the disease can spread.

mX
Autotroph Nutrient
7.4.1 Two-compartment model X Pool N
r(N)X
In Example 7.8 we studied a single-compartment model that led to an au-
tonomous differential equation with one dependent variable. In this sub-
cX aN
section, we introduce a model with two compartments that describes the
interaction of an autotroph and its nutrient pool. An autotroph is an organism Figure 7.27: A schematic description of
that can manufacture organic compounds entirely from inorganic components. the interaction between autotrophs and the
nutrient pool.
Examples include most chlorophyll-containing plants and blue-green algae.
(The discussion below is partly adapted from DeAngelis, 1992.) A schematic
description of the interaction can be found in Figure 7.27.
We assume that the nutrient pool has an external source and denote the
input rate of nutrients by NI . That is, NI denotes the total mass per unit time
that flows into the system. Nutrients may be washed out, and we assume
that the output rate is proportional to the total mass of nutrients, N, in the
compartment, with proportionality constant a.
The autotroph feeds on the nutrients in the nutrient pool, and its growth
rate is proportional to the quantity of autotroph biomass, X. The specific rate
of growth, r (N ), depends on the amount of available nutrients.
Autotrophs can leave the autotroph compartment in two ways: (1) At rate
c, they get completely lost (e.g., through harvesting or by being washed out
of their habitat); (2) at rate m, the biomass of the autotroph gets recycled back
into the nutrient pool (e.g., after the death of an organism).
 D I F F E R E N T I A L E Q UAT I O N S 197

This system is then described by the following set of differential equa-

tions:
dN
= NI − aN − r (N )X + mX, (7.44)
dt
dX
= r (N )X − (m + c)X. (7.45)
dt
You can compare the set of differential equations with the schematic de-
scription of the model in Figure 7.27. In particular, you should pay attention
to the direction of the arrows. For instance, the term r (N )X shows up in
both equations: Because this term corresponds to the nutrient uptake by the
autotrophs, it appears as a loss to the nutrient pool, which is indicated by the
minus sign in front of the term in Eqn. (7.44); and the same term appears with
a plus sign in Eqn. (7.45), because the uptake of nutrients by the autotroph re-
sults in an increase in autotroph biomass. Note that the arrow labeled r (N )X
goes from the nutrient pool to the autotroph pool. Hence, it represents a loss
in the nutrient pool [a minus sign in Eqn. (7.44) in front of r (N )X] and a gain
in the autotroph pool [a plus sign in Eqn. (7.45) in front of r (N )X].
It is important to learn how to go from the schematic description to the
set of differential equations and back. A schematic description quickly sum-
marizes the flow of matter (such as nutrients), whereas the set of differential
equations is indispensable if we want to analyze the system.
To keep the discussion concrete, we assume that the function r (N ) is
linear; that is,
r (N ) = bN,
for some constant b > 0.
An equilibrium for the system given by Eqn. (7.44) and Eqn. (7.45) is
characterized by simultaneously requiring that
dN dX
=0 and = 0, (7.46)
dt dt
since, when the rates of change of both quantities are equal to 0, the values
for N and X no longer change.
There is a graphical method for finding equilibria: plotting the zero
isoclines. These curves are obtained by setting dN/dt = 0 and dX /dt = 0.
The curves for which dX /dt = 0 are obtained by setting the right-hand side of
Eqn. (7.45) equal to 0; that is, X (bN − (m + c)) = 0, yielding
m+c
N= or X = 0.
b
The curve for which dN/dt = 0 is obtained by setting the right-hand side of
Eqn. (7.44) equal to 0, yielding
NI − aN
X= .
bN − m
We plot the three curves in the N–X plane, as illustrated in Figure 7.28.
198 I N T E G R A L C A L C U L U S W I T H A P P L I C AT I O N S F O R L I F E S C I E N C E S

X Figure 7.28: The zero isoclines in the N–X

plane.
dX
dt 0

dN
(N2, X2) dt 0
dX
dt 0
(N1, 0)
0 m mc NI N
b b a

The zero isocline for N intersects the horizontal line X = 0 at (N1 , 0),
where N1 satisfies
NI − aN1
0=
bN1 − m
and thus N1 = NI /a. We call this a trivial equilibrium, since it corresponds to
the case in which there is no autotroph in the system. (The system reduces to
the single compartment model we discussed in Example 7.8.)
Depending on where the vertical zero isocline for X is located, there
can be another equilibrium (N2 , X2 ) for which both N2 and X2 are positive.
Looking at Figure 7.28, we see that when
m m + c NI
< < , (7.47)
b b a
the two zero isoclines N = (m + c)/b and X = (NI − aN )/(bN − m) intersect
in the first quadrant. We call this equilibrium a nontrivial equilibrium.
Since (m + c)/b > m/b when c > 0, the vertical zero isocline dX /dt = 0
is always to the right of the vertical asymptote N = m/b. The first inequality
in Eqn. 7.47 therefore always holds. If we solve the second inequality,
m + c NI
< ,
b a
for b, we find that
a
b> (m + c).
NI
That is, the growth parameter b must exceed a certain threshold in order
for the autotroph to survive. This should be intuitively clear, since m and c
are the rates at which the autotroph pool is depleted. The depletion must be
balanced by an increase in biomass of the autotroph.
Suppose that the ratio m/b is fixed. Then the smaller c (the rate at which
autotrophs get lost) is, the farther to the left is the vertical zero isocline
dX /dt = 0 and, consequently, the larger is the equilibrium value X2 . This
 D I F F E R E N T I A L E Q UAT I O N S 199

should also be intuitively clear, since losing fewer autotrophs should result in
a higher equilibrium value.
To find the nontrivial equilibrium (N2 , X2 ), we need to solve the system of
equations

0 = NI − aN2 − bN2 X2 + mX2 , (7.48)

0 = bN2 X2 − (m + c)X2 , (7.49)

where we now assume that X2 = 0. Eqn. (7.49) yields

m+c
N2 = .
b
Using this relationship in Eqn.(7.48), we find that
m+c
0 = NI − a − (m + c)X2 + mX2 ,
b
or  
1 m+c
X2 = NI − a .
c b
It can be shown that if the nontrivial equilibrium exists, it is locally stable;
that is, the system will return to this equilibrium after a small perturbation.

Summary

In this chapter we focused on solving autonomous differential equations

using separation of variables integration technique. We considered a number
of important biological models based on such differential equations, in
particular:

1. Exponential growth model.

2. Restricted growth model (von Bertalanffy model).
3. Single compartment model.
4. Logistic growth model (including threshold).
5. The Levins model.
6. Simple model of epidemics (the Kermack–McKendrick model).
7. Two-compartment model.

We analyzed local stability of steady states using qualitative (studied in Math

154) and analytical techniques.

Exercises

7.1. Differential equation. Solve the following pure-time differential

equations with initial conditions.
dy
(a) = e−3x , where y0 = 10 for x0 = 0.
dx
200 I N T E G R A L C A L C U L U S W I T H A P P L I C AT I O N S F O R L I F E S C I E N C E S

dy 1
(b) = , where y0 = 1 when x0 = 0.
dx 1 + x2
dh
(c) = 5 − 16t 2 , where h(3) = −11.
dt
7.2. Cell volume. Suppose that the volume V (t ) of a cell at time t changes
according to
dV
= 1 + cos t, with V (0) = 5.
dt
Find V (t ).
7.3. Differential equation. Solve the given autonomous differential
equations.
dx
(a) = 1 − 3x, where x(−1) = −2.
dt
dN
(b) = 5 − N, where N (2) = 3.
dt
7.4. Radioactive decay. Assume that W (t ) denotes the amount of ra-
dioactive material in a substance at time t. Radioactive decay is then
described by the differential equation
dW
= −λW (t ) with W (0) = W0 ,
dt
where λ is a positive constant called the decay constant.
(a) Solve the equation.
(b) Assume that W (0) = 123 gr and W (5) = 20 gr and that time is
measured in minutes. Find the decay constant λ and determine the
half-life of the radioactive substance.
7.5. Asymptotic length of the fish. Denote by L(t ) the length of a fish at
time t, and assume that the fish grows according to the von Bertalanffy
equation
dL
= k(34 − L(t )) with L(0) = 2.
dt
(a) Solve the differential equation.
(b) Use your solution in (a) to determine k under the assumption that
L(4) = 10. Sketch the graph of L(t ) for this value of k.
(c) Find the length of the fish when t = 10.
(d) Find the asymptotic length of the fish; that is, find limt→∞ L(t ).
7.6. Solving a differential equation. Use the partial-fraction method to
solve the following differential equations with given initial conditions.
dy
(a) = y(1 − y), y(0) = 2.
dx
dy
(b) = (y − 1)(y − 2), y(0) = 0.
dx
dy
(c) = (3 − y)(2 + y), y(0) = 0.
dx
 D I F F E R E N T I A L E Q UAT I O N S 201

7.7. Logistic equation. Suppose that the size of a population, denoted by

N (t ), satisfies  
dN N
= 0.7N 1 − .
dt 35
(a) Determine all equilibria by solving dN/dt = 0.
(b) Solve this differential equation for (i) N (0) = 10, (ii) N (0) = 35.
Find limt→∞ N (t ) for each of the two initial conditions.
7.8. Separation of variables. Consider the following differential equation,
which is important in population genetics:
1 d
a(x)g(x) − [b(x)g(x)] = 0.
2 dx
Here, b(x) > 0.
(a) Define y = b(x)g(x), and show that y satisfies

a(x) 1 dy
y− = 0.
b(x) 2 dx

(b) Separate variables in this modified equation and show that if y > 0,
then   
a(x)
y = C exp 2 dx .
b(x)
7.9. Equilibrium solutions. Suppose that
dy
= (4 − y)(5 − y).
dx
(a) Find the equilibria of this differential equation.
(b) Graph dy/dx as a function of y, and use your graph to discuss the
stability of the equilibria.
(c) Discuss the stability of the equilibria using the analytical approach.
7.10. Equilibrium solutions. Suppose that
dy
= y(2 − y)(y − 3).
dx
(a) Find the equilibria of this differential equation.
(b) Graph dy/dx as a function of y, and use your graph to discuss the
stability of the equilibria.
(c) Discuss the stability of the equilibria using the analytical approach.
7.11. Logistic equation. Assume that the size of a population evolves
according to the logistic equation with intrinsic rate of growth r = 1.5.
Assume that the carrying capacity K = 100.
(a) Find the differential equation that describes the rate of growth of
this population.
202 I N T E G R A L C A L C U L U S W I T H A P P L I C AT I O N S F O R L I F E S C I E N C E S

(b) Find all equilibria, and, using the graphical approach, discuss the
stability of the equilibria.
(c) Use the analytical approach to determine the stability of the equilib-
ria. Compare your answers with your results in (b).
7.12. Logistic equation. The logistic curve N (t ) is an S-shaped curve that
satisfies  
dN N
= rN 1 − with N (0) = N0
dt K
when N0 < K. Use the differential equation to show that the inflection
point of the logistic curve is at exactly half the saturation value of the
curve. [Hint: Differentiate the right-hand side with respect to t.]
7.13. Single compartment model. Denote the concentration of the solute at
time t by C (t ), and assume that

dC
= 3(20 −C (t )) for t ≥ 0.
dt
(a) Solve the differential equation when C (0) = 5.
(b) Find limt→∞ C (t ).
(c) Use your answer in (a) to determine t so that C (t ) = 10.
7.14. Levins model. Denote by p = p(t ) the fraction of occupied patches in
a metapopulation model, and assume that
dp
= 2p(1 − p) − p for t ≥ 0.
dt
(a) Set g( p) = 2p(1 − p) − p. Graph g( p) for p ∈ [0, 1].
(b) Find all equilibria of the differential equation that are in [0, 1]. Use
your graph in (a) to determine their stability.
(c) Use the analytical approach to analyze the stability of the equilibria
that you found in (b).
7.15. A metapopulation model with density-dependent extinction. De-
note by p = p(t ) the fraction of occupied patches in a metapopulation
model, and assume that
dp
= cp(1 − p) − p2 for t ≥ 0,
dt
where c > 0. The term p2 describes the density-dependent extinction
of patches; that is, the per-patch extinction rate is p, and a fraction
p of patches are occupied, resulting in an extinction rate of p2 . The
colonization of vacant patches is the same as in the Levins model.
(a) Set g( p) = cp(1 − p) − p2 and sketch the graph of g( p).
(b) Find all equilibria of the differential equation in [0, 1], and deter-
mine their stability.
 D I F F E R E N T I A L E Q UAT I O N S 203

(c) Is there a nontrivial equilibrium when c > 0? Contrast your findings

with the corresponding results in the Levins model.

7.16. Allee effect. Denote the size of a population at time t by N (t ), and

assume that
 
dN N
= 2N (N − 10) 1 − for t ≥ 0.
dt 100

(a) Find all equilibria of the differential equation.

(b) Use the analytical approach to determine the stability of the equilib-
ria you found in (a).
(c) Set  
N
g(N ) = 2N (N − 10) 1 − ,
100
for N ≥ 0, and graph g(N ). Identify the equilibria of the differential
equation on your graph, and use the graph to determine the stability
of the equilibria. Compare your results with your findings in (b).

7.17. Kermack–McKendrick model. We will investigate the classical

Kermack–McKendrick model for the spread of an infectious disease
in a population of fixed size N. If S(t ) denotes the number of suscep-
tibles at time t, I (t ) the number of infectives at time t, and R(t ) the
number of immune individuals at time t, then

dS
= −bSI,
dt
dI
= bSI − aI,
dt
and R(t ) = N − S(t ) − I (t ).

(a) Divide second differential equation by first one to show that when
I > 0,
dI a1
= − 1.
dS bS
Also, show that when R(0) = 0, I (0) = I0 , and S(0) = S0 , then
solution of this new differential equation satisfies

a S(t )
I (t ) = N − S(t ) + ln .
b S0

(b) Since I (t ) gives the number of infectives at time t and dI/dt =

bSI − aI, if S(0) > a/b, then dI/dt > 0 at time t = 0. Also, since
limt→∞ I (t ) = 0, there is a time t > 0 at which I (t ) is maximal.
Show that the number of susceptibles when I (t ) is maximal is given
by S = a/b. [Hint: When I (t ) attains a maximum, the derivative of
I (t ) with respect to t, dI/dt, is equal to 0.]
204 I N T E G R A L C A L C U L U S W I T H A P P L I C AT I O N S F O R L I F E S C I E N C E S

(c) In (a), you expressed I (t ) as a function of S(t ). Use your result in

(b) to show that the maximal number of infectives is given by
 
a a a/b
Imax = N − + ln .
b b S0

(d) Use your result in (c) to show that Imax is a decreasing function
of the parameter a/b for a/b < S0 (i.e., in the case in which the
infection can spread). Use the latter statement to explain how a and
b determine the severity (as measured by Imax ) of the disease. Does
this make sense?