Differential Equations

for Business and Economics

Course Notes for AMATH 350

Sue Ann Campbell

Department of Applied Mathematics
University of Waterloo

Fall 2011

c S.A. Campbell 2010

Contents

1 Introduction 1

2 First Order Ordinary Differential Equations 5

2.1 Separable, First Order Ordinary Differential Equations . . . . . . . 7
2.1.1 Application: Continuously compounded interest . . . . . . . 11
2.1.2 Dimensions and Units . . . . . . . . . . . . . . . . . . . . . 14
2.1.3 Application: Commodity Prices - The Dynamics of Supply
and Demand . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2 Linear, First Order Ordinary Differential Equations . . . . . . . . . 18
2.3 Nonlinear, First Order Ordinary Differential Equations . . . . . . . 23
2.3.1 Bernoulli Differential Equations . . . . . . . . . . . . . . . . 23
2.3.2 *A graphical/qualitative approach . . . . . . . . . . . . . . . 25

3 Higher Order Linear Ordinary Differential Equations 29

3.1 Theory of nth Order Linear ODEs . . . . . . . . . . . . . . . . . . . 30
3.2 The General Solution of the Homogeneous Equation . . . . . . . . . 34
3.3 The General Solution of the Inhomogeneous Equation . . . . . . . . 40
3.4 Constant Coefficient nth Order Homogeneous Ordinary Differential
Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.5 Finding a Particular Solution of the Inhomogeneous Equation . . . 50
3.5.1 Method of Undetermined Coefficients . . . . . . . . . . . . . 50
3.5.2 Method of Reduction of Order . . . . . . . . . . . . . . . . . 56
3.5.3 Method of Variation of Parameters . . . . . . . . . . . . . . 58

ii
4 Systems of Ordinary Differential Equations 61
4.1 Theory of Linear Systems . . . . . . . . . . . . . . . . . . . . . . . 64
4.1.1 The General Solution of the Homogeneous Equation . . . . . 65
4.1.2 The General Solution of the Inhomogeneous Equation . . . . 69
4.2 Homogeneous Systems with Constant Coefficients . . . . . . . . . . 70
4.2.1 The case n = 2 . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.2.2 The case n = 3 . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.2.3 Application . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.3 Finding a Particular Solution of the Inhomogeneous Equation . . . 84
4.3.1 Method of Variation of Parameters . . . . . . . . . . . . . . 84

5 Partial Differential Equations 87

5.1 Linear Partial Differential Equations - Operator Notation (Optional) 93
5.2 Linear, First Order Partial Differential Equations . . . . . . . . . . 96
5.3 Linear, Second Order Partial Differential Equations . . . . . . . . . 103
5.4 The Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . 112

6 Pricing of Derivatives 123

6.1 Terminology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
6.2 Pricing Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
6.2.1 Stochastic Price Model . . . . . . . . . . . . . . . . . . . . . 126
6.2.2 Deriving the Black-Scholes Equation . . . . . . . . . . . . . 127
6.2.3 Solution of the Black Scholes Equation . . . . . . . . . . . . 131

References 135

iii
Chapter 1

Introduction

Definition 1.1: A differential equation (DE) is an equation for an unknown

function involving one or more derivatives of the unknown function. The unknown
function is represented in terms of a dependent variable. The variable(s) the
unknown function depends upon are called the independent variables.

There are various types of differential equations, distinguished by the number

of dependent and independent variables involved.

Ordinary Differential Equations (ODE) : one dependent variable, one inde-

pendent variable

Partial Differential Equations (PDE) : one dependent variable, two or more

independent variables

System of Differential Equations : two or more dependent variables

Example 1.2:
Types of Differential Equations

dy
1. = 6x2 y (ODE)
dx
d2 y
2. + sin(y) = sin(x) (ODE)
dx2
dy1
3. = y1 − y2
dt
dy2
= y1 + 3y2 (ODE System)
dt
∂u ∂u
4. x +u =1 (PDE)
∂x ∂y

1
 ∂F ∂F 1 ∂ 2F
5. F = +S + S2 2 (Black-Scholes PDE)
∂t ∂S 2 ∂S
6. uxxx + uyy = vx
vxxx + vyy = u (PDE System)

To study the solutions of differential equations will classify them using the following
definitions
Definition 1.3: The order of a differential equation is the order of the highest
derivative in the equation.
Definition 1.4: A differential equation in which the dependent variable and all its
derivatives appear linearly is called linear; otherwise it is called nonlinear.

In the examples above 1, 3 and 4 are first order, 2 and 5 are second order and
example 6 is third order. Also, 1, 3, 5 and 6 are linear while 2 and 4 are nonlinear.
Definition 1.5: A solution of a differential equation is any function which satisfies
the equation.
Example 1.6:
3 dy
1. y(x) = 5e2x is a solution of the DE = 6x2 y. To check this, substitute the
dx
function into the equation.
dy
LHS =
dx
3
= 5e2x · 6x2
= 6x2 y
= RHS

2. y1 (t) = e2t , y2 (t) = −e2t is a solution of (3) above. To check this, substitute
the functions into the equations.

LHS1 = y10 = 2e2t

RHS1 = y1 − y2 = e2t − (−e2t ) = 2e2t

and

LHS2 = y20 = −2e2t

RHS2 = y1 + 3y2 = e2t − 3e2t = −2e2t

2
 ∂u ∂u
3. u(x, t) = sin(x − t) is a solution of the PDE + = 0. To check this, take
∂x ∂t
the appropriate partial derivatives:
∂u
= cos(x − t)
∂x
∂u
= − cos(x − t)
∂t
Note: In the examples above we made no restrictions on the domain of the solution
function. In some cases it is necessary to make a restriction.
Example 1.7:
∂u ∂u √
Consider again the PDE + = 0. The function u(x, t) = t − x clearly
∂x ∂t
satisfies the DE since
1 −1
ut = √ and ux = √
2 t−x 2 t−x
However u(x, y) is only defined if t − x ≥ 0, i.e., t ≥ x, and ut , ux are only defined
√
if t > x. Thus we say u = t − x is a solution of the DE for t > x.
Example 1.8:
Consider the second order linear ODE

y 00 + y = 0. (1.1)

It is easy to check that y1 = cos(x) and y2 = sin(x) are solutions of this DE. In
fact, so is any linear combination of y1 and y2 . That is, for any constants c1 and c2 ,

y = c1 sin(x) + c2 cos(x) (1.2)

is a solution of (1.1).

Now consider the following problems

(A) Find a solution of (1.1) such that y(0) = 0 and y 0 (0) = 2

(B) Find a solution of (1.1) such that y(0) = 0 and y 0 (π/2) = 0

(C) Find a solution of (1.1) such that y(0) = 1 and y 0 (π/2) = 0

It is easy to check that problem (A) has the solution y(x) = 2 sin(x), problem (B)
has solutions y(x) = c1 sin(x) for any constant c1 and that there is no solution of
the form (1.2) to problem (C).
The condition in (A) is called an initial condition, while those in (B) and
(C) are called boundary conditions. More generally, we have the following.

3
Definition 1.9: An initial value problem (IVP) is a problem which seeks to find
a solution of a differential equation subject to conditions on the unknown function
and its derivatives at one value of the independent variable. These conditions are
called the initial conditions (IC).
Definition 1.10: A boundary value problem (BVP) is a problem which seeks
to find a solution of a differential equation subject to conditions on the unknown
function and its derivatives at two or more values of the independent variable.
These conditions are called boundary conditions (BC).

There are three questions which form the basis of much of the fundamental
theory of differential equations.

1. Existence: Does a solution of a given DE/IVP/BVP exist?

2. Uniqueness: Is there more than one solution of a given IVP/BVP?

3. Method: How does one find a solution for a given DE/IVP/BVP?

The primary focus of this course is the third question, although we will briefly touch
upon questions 1 and 2.

4
Chapter 2

First Order Ordinary Differential

Equations

The most general first order, ordinary differential equation is a relationship of the
form
F (x, y, y 0 ) = 0 (2.1)
where y(x) is the (unknown) dependent function and x is the independent variable.
We will restrict our attention to equations that can be put into standard form:
dy
= f (x, y) (2.2)
dx
Definition 2.1: A solution of the first order ordinary differential equation (2.2)
is any differentiable function y = ψ(x) such that

dy
= ψ 0 (x) = f (x, ψ(x))
dx
for all x in some open interval I ⊆ R.

In other words, ψ, ψ 0 must be defined on I and satisfy the differential equation for
all x in I.
Definition 2.2: The interval I in Definition 2.1 is called the interval of existence
or interval of definition of the solution.

We will express this concept in several equivalent ways.

y = ψ(x) is a solution of (2.2) with interval of existence I.

y = ψ(x) is a solution of (2.2) on I.

5
 y = ψ(x), x ∈ I is a solution of (2.2).

Definition 2.3: An initial (or boundary) condition for (2.2) is

y(x0 ) = y0 (2.3)

where x0 and y0 are given real numbers.

Questions 1 and 2 from the end of Chapter 1 can be answered by specifying con-
ditions on the function f (x, y) in (2.2). The following is an example of a theorem
which gives sufficient but not necessary conditions for the existence of a unique
solution.

Theorem 2.4 (Existence & Uniqueness). Suppose that f (x, y) and fy (x, y) are
(real, finite, single-valued and) continuous at all points (x, y) within the rectangle
R : {(x, y) : |x − x0 | ≤ a, |y − y0 | ≤ b}. Then the initial value problem consisting of
(2.2) and (2.3) has a unique solution defined on some interval I = |x−x0 | ≤ h ≤ a.

Proof. See section 2.8 of Boyce and DiPrima [1] or Appendix 1 of Edwards and
Penney [2].

(x0,y0) R

x
I

Figure 2.1: Region of interest for the Existence-Uniqueness theorem

In the following sections we will try to answer question 3 above, by further

classifing the equation (2.2) according to the form of f (x, y).

6
2.1 Separable, First Order Ordinary Differential
Equations
A separable, first order ordinary differential equation is an equation of the form
y 0 = f (x, y) where f (x, y) is the product of a function of x only and a function of
y only, i.e.
y 0 = g(x)h(y) (2.4)
or
1 dy
= g(x) (2.5)
h(y) dx
Loosely, a separable equation is one in which we can put everything to do with y
on one side of the equation and everything to do with x on the other side.
Example 2.5:
Solve the differential equation
dy
= −xy. (2.6)
dx
Solution:
If y 6= 0 we can put this in the form
1 dy
= −x
y dx
To solve this equation we integrate both sides with respect to x:
Z Z
1 dy
dx = − x dx.
y dx
dy
Now use the change of variable formula dx dx = dy:
Z Z
1
dy = − x dx.
y
Evaluating the integrals gives:
x2
ln |y| + c1 = − + c2 ,
2
or
x2
ln |y| = − + C,
2
where c3 = c2 − c1 . This defines the solution implicitly. To find an explicit solution,
solve for y as a function of x by taking the exponential of each side.
x2
|y| = e− 2 +c3
x2
= ec3 e− 2

7
There are two cases to consider. Solutions with y > 0 are given by
x2
y = ec3 e− 2 , x ∈ R

while those with y < 0 are given by

x2
y = −ec3 e− 2 , x ∈ R.

Recall that we assumed y 6= 0 at the beginning. Returning to the original differen-

tial equation (2.6), we see that

y = 0, x ∈ R

is yet another solution.

We can represent the three types of solutions with the single formula
x2
y = Ae− 2 (2.7)

Some solutions for various values of the constant A are shown in Figure 2.2. The

2.5
A>0

A= 0
-10 -7.5 -5 -2.5 0 2.5 5 7.5 10

-2.5 A<0

-5

Figure 2.2: General solution of (2.6) for various values of the parameter A.

formula (2.7) containing all possible solutions of the differential equation is called
the general solution of the differential equation.
Definition 2.6: The general solution of a first order ordinary differential equa-
tion (2.2) is a solution containing one arbitrary constant that represents almost all
solutions of the differential equation.

8
Definition 2.7: A solution which is not represented by the general solution is
called a singular solution.
Definition 2.8: A particular solution of equation (2.2) is a solution containing
no arbitrary constants.

A particular solution may be obtained from the general solution by

(i) specifying a value for the arbitrary constant

(ii) by specifying an initial condition

Definition 2.9: An equilibrium solution of a first order ordinary differential

equation y 0 = f (x, y) is a constant solution y = k, −∞ ≤ x ≤ ∞.

Note: Any constant k such that f (x, k) = 0, for all x ∈ R defines an equilibrium
solution of (2.2).
Example 2.10:
x2
As we saw in the Example 2.5, y = Ce− 2 , x ∈ R is the general solution of equation
2 2
(2.6). Some particular solutions are y = e−x /2 , x ∈ R; y = −5e−x /2 , x ∈ R;
y = 0, x ∈ R. The only equilibrium solution of the equation is y = 0, x ∈ R.

We can generalize the approach of the previous example as follows:

Method of Separation of Variables

1. Start with equation standard form:

dy
= g(x)h(y)
dx

2. Separate:
1 dy
= g(x); h(y) 6= 0
h(y) dx
3. Integrate both sides with respect to x:
Z Z
1 dy
dx = g(x) dx
h(y) dx

dy
4. Change the variable of integration in left side using the formula dy = dx:
dx
Z Z
1
dy = g(x) dx
h(y)

9
 5. Check for constant (equilibrium) solutions y = k by looking for values of k
that satisfy:
h(k) = 0

Example 2.11:
Solve the initial value problem
dy −x
= , y 6= 0, y(1) = 2 (2.8)
dx y

Solution:
Separate the equation:
dy
y = −x.
dx
Integrate both sides with respect to x:
Z Z
dy
y dx = − x dx.
dx
Use the change of variables formula:
Z Z
y dy = − x dx.

Evaluate the integrals:

y2 x2
= − + c1 ,
2 2
or
y 2 = C − x2 ,
where C = 2c1 . Note that we must have C ≥ 0 for this expression to make sense.
Solving for y, we see that again there are two cases. Solutions with y > 0 are
given by
√
y = + C − x2
while those with y < 0 are given by
√
y = − C − x2

Unlike Example 2.5, however, there is no way to combine these into one general
solution. We can only define the general solution either for y > 0 or for y < 0.
√ √ dy
Note that in both cases y is defined for − C ≤ x ≤ C, but dx is defined only for
√ √
− C < x < C. Thus the general solution of the differential equation is
√ √ √
y = + C − x2 , − C < x < C for y > 0 (2.9)
√ √ √
y = − C − x2 , − C < x < C for y < 0 (2.10)

10
 2.4

(1,2)
y>0
1.6

0.8

-4.8 -4 -3.2 -2.4 -1.6 -0.8 0 0.8 1.6 2.4 3.2 4 4.8

-0.8

-1.6 y<0

-2.4

Figure 2.3: Family of solutions for differential equation y 0 = −xy −1 . The specific
solution of (2.8) satisfying the initial condition y(1) = 2 is given by the thick curve.

These solutions determine the two families of curves shown in Figure 2.3
Now apply the initial condition y(1) = 2. This forces us to choose general
solution (2.9). Putting the initial condition into (2.9):
√
2 = C − 1 ⇒ 4 = C − 1 ⇒ C = 5.

Thus the solution of the initial value problem (2.8) is

√ √ √
y = 5 − x2 , − 5 < x < 5.

This corresponds to the curve in Figure 2.3 that passes through the point (1, 2).

2.1.1 Application: Continuously compounded interest

Suppose that a sum of money is invested in a fund that pays interest at a constant
annual rate of r% per year. The value of the investment at any time depends on
the interest rate, but also on the frequency with which interest is compounded.
Typical frequencies of compounding are yearly, quarterly and daily. We will set up
an initial value problem to model the situation.
Variables:

V - value of the investment (measured in $)

11
t - time (measured in years)

Constants:

V0 - initial amount invested (measured in $)

r - annual interest rate (measured in %/year)

We are to determine how V depends on t.

We will make the simplifying assumption/approximation that the interest rate
is compounded continuously and evaluate at the end how good this approximation
is. Under this assumption, we can formulate a differential equation for V (t):
! ! !
rate of change of value interest value of investment
=
of investment at time t rate at time t
or
dV
= rV.
dt
The initial condition is V (0) = V0 .
The differential equation is separable, so we solve it as described above.

1 dV
= r, V 6= 0
Z V dt Z
1 dV
dt = r dt
V dt
Z Z
1
dV = r dt
V
ln |V | = rt + c1

Rearranging we see that the general solution of the differential equation is

V = Cert C = ±ec1 .

Applying the initial condition gives:

V0 = Cer·0 ⇒ V0 = C.

The particular solution of the initial value problem is thus

V = V0 ert .

12
 t(years) m=1 m = 4 m = 12 m = 365 continuous
1 1.0500 1.0509 1.0512 1.0513 1.0513
2 1.1025 1.1045 1.1049 1.1052 1.1052
5 1.2763 1.2820 1.12834 1.2840 1.2840
10 1.6289 1.6436 1.6470 1.6487 1.6487
50 11.4674 11.9951 12.1194 12.1803 12.1825

Table 2.1: Value of investment with a yearly interest rate of 0.05% if the interest
is compounded yearly, quarterly, monthly, daily and continuously.

Now we will compare with the true solution.

For an investment compounded m times per year we have the formula
 r mt
V (t) = V0 1 + (2.11)
m
Recall that  r m
lim 1+ = er .
m→∞ m
It follows that
 r mt h  r m i t
lim V0 1 + = V0 lim 1 + = V0 ert ,
m→∞ m m→∞ m
so continuous compounding corresponds to letting m → ∞.

Is this a good approximation? Table 2.1 compares the value of V (t)/V0 for an
obtained using the formula (2.11) and using continuous compounding with r =
0.05?? for various values of m and t. We can conclude from this example that
continuous compounding is a good approximation if m is large and t is not too
large (e.g. your daily interest bank account) and not good if m is small or t is
large (e.g. your Canada savings bonds).
Note: The continuous compounding approximation can also be useful in studying
investments where the interest rate is not constant but varies in time. In this case,
the initial value problem is
dV
= r(t)V, V (0) = V0
dt

r(t) is a specified function of time.

13
2.1.2 Dimensions and Units

When setting up a model such as in the last problem it is important to keep track
of the units used to measure each variable and constant. A model is only correct
if it is dimensionally homogeneous. This means that the units of each term in
an equation should be the same.
In our example

dV
has units of dollars/year
dt
r has units of 1/year

V has units of dollars

So the equation
dV
= rV
dt
is dimensionally homogeneous with each side having units of dollars/year. Also,
both V (0) and V0 have units of dollars so our initial condition

V (0) = V0

is also dimensionally homogeneous.

2.1.3 Application: Commodity Prices - The Dynamics of

Supply and Demand

Consider the price of a commodity such as oil or wheat. We wish to determine how
the price changes with time, given some basic assumptions.
Let

q(t) = quantity of commodity available at time t (# items)

p(t) = price at time t ($/item)
S = supply = rate at which commodity is supplied by producer (# items/day)
D = demand = rate at which commodity is consumed (# items/day)

The model is then

!
rate of change of
= supply − demand
quantity of commodity

14
 dq
=S−D
dt
Assumption: The rate of increase of price is proportional to the rate at which the
quantity of commodity declines, that is,
dp dq
= −α ,
dt dt
where α > 0 is a constant. This leads to
dp
= −α(S − D),
dt
or
dp
= α(D − S)
dt
In general, S and D can depend on price, time and even the rate of change of price.
This gives rise to the following first order differential equation for the price:
dp
= α(D(t, p, dp
dt
)) − S(t, p, dp
dt
))
dt
Depending on the form of the demand and supply functions, different types of
differential equations will result.
Example 2.12:
Consider a commodity where the supply and demand functions are given by

S = 100 − 20p + 2p2

D = 100 + 10p − 3p2

and the rate of change of price is twice the rate of change of the commodity.
Determine how the price varies as a function of time, given that the starting price
is p0 .
With these specifications, the model becomes:
dp
= 2[D − S]
dt
dp
= 2(100 + 10p − 3p2 − (100 − 20p + 2p2 ))
dt
= 2(30p − 5p2 )
= 10p(6 − p)
(2.12)

with initial condition

p(0) = p0 .

15
The differential equation is is separable, so we solve as in previous examples:
dp
= 10p(6 − p)
Z dt Z
1 dp
dt = 10 dt, p(6 − p) 6= 0
p(6 − p) dt
Z Z
1
dp = 10 dt.
p(6 − p)

To evaluate the integral on the left side we need to use partial fractions:
1 A B
= +
p(6 − p) p 6−p
A(6 − p) + Bp
=
p(6 − p)
6A + (B − A)p
=
p(6 − p)

Comparing the left and right sides shows that A = 61 , B = 16 .

Putting this into the original integral, we have
Z Z Z
1 dp 1 dp
+ = 10 dt
6 p 6 6−p
1 1
ln |p| − ln |6 − p| = 10t + c1
6 6
p
ln = 60t + c2 , (c2 = 6c1 )
6−p
p
= c3 e60t , (c3 = ±ec2 )
6−p
Solving for p gives the general solution
6c3 e60t
p(t) = ,
1 + c3 e60t
which can be rearranged into the form
6
p(t) = ,
Ce−60t + 1
1
where C = c3
. Note that p = 0 and p = 6 are equilibrium solutions of the differential
equation.
Applying the initial condition we have
6 6
p0 = ⇒ C= −1
C +1 p0

16
so our solution to the initial value problem is
6
p(t) =
( p60 − 1)e−60t + 1

How does the price change with time? Consider the following limits
6
lim p(t) = lim = 6, for all p0
t→∞ t→∞ ( 6 − 1)e−60t + 1
p0

lim p(t) = 0 if p0 < 6

t→−∞

lim p(t) = +∞ if p0 > 6

t→t+
A

where tA is the solution to ( p60 − 1)e−60t + 1 = 0. From the differential equation we

see that
dp
if p > 6 then < 0 ⇒ p is decreasing
dt
dp
if p < 6 then > 0 ⇒ p is increasing
dt
These results allow us to sketch the solutions, for various values of the initial price
p0 , as shown in Figure 2.4.

20

15

10

-0.025 0 0.025 0.05 0.075 0.1 0.125 0.15 0.175 0.2 0.225

Figure 2.4: Solutions to initial value problem (2.12) for various values of the initial
price p0 .

17
 Conclusion: For any starting price p0 6= 0, 6 the price will gradually approach
$6. If p0 > 6 the price will decrease to $6, if p0 < 6 the price will increase to $6.

• All solutions tend to the equilibrium solution p = 6 as t → ∞. We say p = 6

is a stable equilibrium.

• When 0 < p0 < 6, solutions tend away from p = 0 as t → ∞. We say p = 0

is an unstable equilibrium.

2.2 Linear, First Order Ordinary Differential Equa-

tions

Definition 2.13: A linear, first order ordinary differential equation is an

ordinary differential equation of the form

y 0 = f (x, y)

where f (x, y) is a linear function of y.

This equation can be written in standard form as

dy
+ P (x)y = Q(x) (2.13)
dx

One can always find an explicit solution of a linear first order ordinary differ-
ential equation. In the following pages we will develop the procedure for doing
this.
Recall from Calculus that, by the product rule of differentiation,
dv du d
u +v = [u(x)v(x)].
dx dx dx
The expression on the left side is called an exact differential as it can be directly
integrated using the Fundamental Theorem of Calculus:
Z
d
[u(x)v(x)] dx = u(x)v(x) + C
dx
Recognizing expressions which can be written as an exact differential is a key step
in solving linear first order ordinary differential equations. Here are some examples:

d
(i) 2x cos(x) − x2 sin(x) = dx
[x2 cos(x)]

18
 dy d
(ii) x dx +y = dx
[xy]
dy d
(iii) cos(x) dx − sin(x)y = dx
[cos(x) y]

If the left side of the linear differential equation (2.13) is an exact differential
then the differential equation is easy to solve.
Example 2.14:
dy
Find the general solution of x dx + y = cos(x).
Solution:
Use the product rule to rewrite the left side as an exact differential:
d
[xy] = cos(x).
dx
Integrate both sides of the equation with respect to x:
Z Z
d
[xy] = cos(x)
dx
xy = sin(x) + C.

Finally, solve for y to obtain the general solution of the equation:

sin(x) C
y= + , x 6= 0.
x x
What if the left side of the differential equation is not an exact differential? We
look for a function which, when multiplied through the equation, turns the left side
into an exact differential.
Example 2.15:
dy
Find the general solution of dx
− tan(x)y = x2 .
Solution:
The left hand side of the equation is not an exact differential, however, the list
of exact differentials above suggests that multiplying the equation by cos(x) will
turn the left side into an exact differential. In order to preserve the equation, we
multiply through both sides of the equation by cos(x)1 :
dy
cos(x) − sin(x)y = x2 cos(x).
dx
This yields an new differential equation which is equivalent to the original one, i.e.,
it describes the same unknown function. The left side can now be written as an
exact differential:
d
[cos(x)y] = x2 cos(x).
dx
1
This process is only valid if cos(x) 6= 0, however, the original equation is not defined if
cos(x) = 0 so no new restriction has been introduced.

19
This equation can be solved as in the last example to find the general solution
(exercise):
y(x) = x2 tan(x) + 2x − 2 tan(x) + C sec(x).
The solution is defined on any interval where cos(x) 6= 0, for example, x ∈ (− π2 , π2 ).

The function cos(x) is called an integrating factor for the differential equation
in the previous example. It turns out we can find an integrating factor for any linear
first order differential equation. We will now describe how to find it.
Consider the just the left side of a linear first order differential equation in
standard form:
dy
+ P (x)y. (2.14)
dx
Let µ(x) be the integrating factor. Multiply the expression (2.14) by µ(x)
dy
µ(x)
+ µ(x)P (x)y.
dx
Now assume this can be written as an exact differential, i.e.,
dy d
µ(x) + µ(x)P (x)y = [µ(x)y].
dx dx
Expanding the right hand side then gives
dy dy dµ
µ(x) + µ(x)P (x)y = µ(x) + y .
dx dx dx
Comparing the two sides, we see µ(x) will make (2.14) and exact differential only
if
dµ
µ(x)P (x) = .
dx
This is a separable differential equation for µ that can be solved as described in the
previous section.
1 dµ
= P (x)
µ dx
Z Z
1
dµ = P (x) dx
µ
Z
ln |µ| = P (x)dx + c1

Rearranging give the general solution

R
P (x)dx
µ = Ce .

Note that C is an arbitrary constant, so this process yields an entire family of

integrating factors. Any member of the family will do, we usually take C = 1 for
simplicity.

20
 R
Definition 2.16: The function µ(x) = e P (x)dx is called the integrating factor
(IF) for the first order, linear ordinary differential equation

dy
+ P (x)y = Q(x).
dx

With this information in hand, we can now describe a general procedure for
solving first order, linear ordinary differential equations.

Method of Solution for Linear First Order Differential Equations

1. Start with equation standard form:

dy
+ P (x)y = Q(x)
dx

2. Find the integrating factor

R
P (x) dx
µ(x) = e

3. Multiply through the equation by the integrating factor

dy
µ(x) + µ(x)P (x)y = µ(x)Q(x)
dx

4. Write the left side as an exact differential

d
[µ(x)y] = µ(x)Q(x)
dx

5. Integrate both sides with respect to x

6. Solve for y

Example 2.17:
Find the general solution of the differential equation xy 0 + 2y = 2x2 .
Solution:
Rewrite the equation in standard form
2
y 0 + y = 2x.
x
2
Here P (x) = x
so the integrating factor is:
2 2
R R
µ(x) = e P (x)dx
=e x
dx
= e2 ln |x| = eln(x ) = x2 .

21
Multiply through the differential equation (in standard form):

x2 y 0 + 2xy = 2x3 .

The left side can be now be rewritten as an exact differential:

d 2
[x y] = 2x3 .
dx
Integrate both sides with respect to x and solve for y to obtain the general solution
Z Z
d 2
[x y]dx = 2x3 dx
dx
1 4
x2 y = x +C
2
1 2 C
y = x + 2, x > 0
2 x
Example 2.18:
Consider a commodity where the supply and demand curves are given by

S(p) = a0
D(p) = a1 + b1 p

where a0 > 0, a1 > 0, b0 are constants and p is the price. If p0 is the initial price
and the rate of change of the price is equal to the negative of the rate of change of
the quantity of the commodity, determine how the price varies in time.
Solution:
Start with the general model described in subsection 2.1.3:
dp dq
= − = −(S − D) = D − S.
dt dt
Using the functions D and S as above gives:
dp
= a1 + b 1 p − a0 ,
dt
which can be rewritten:
dp
− b 1 p = a1 + a0 . (2.15)
dt
Defining the initial condition:
p(0) = p0 (2.16)
gives the initial value problem to be solved.
The differential equation is linear with integrating factor
R
µ(t) = e− b1 dt
= e−b1 t .

22
Multiply through the equation by µ(t):
dp
e−b1 t − e−b1 t b1 p = (a1 + a0 )e−b1 t .
dt
Rewrite the left hand side:
d −b1 t
[e p] = (a1 + a0 )e−b1 t .
dt
Integrate with respect to t:
a1 − a0 −b1 t
e−b1 t p = e +C
−b1
Solve to find the general solution:
a0 − a1
p(t) = + Ceb1 t
b1
For simplicity denote p∗ = a0b−a
1
1
. Note that p∗ is an equilibrium solution of the
equation. Now apply the initial condition to solve for C:

p0 = p∗ + C ⇒ C = p0 − p∗ .

So the solution of the initial value problem is

p(t) = p∗ + (p0 − p∗ )eb1 t

How the solution behaves depends on the values of the parameters a0 , a1 , b1 , p0 .

2.3 Nonlinear, First Order Ordinary Differential

Equations

2.3.1 Bernoulli Differential Equations

Definition 2.19: An equation of the form

dy
+ P (x)y = Q(x)y n
dx
where n is a constant is called a Bernoulli differential equation.

Such equations can always be transformed into a linear differential equation by

the substitution
v = y 1−n

23
Example 2.20:
In Example 2.12 we considered a supply and demand problem where the model was
the following differential equation
dp
= 10p(6 − p) = 60p − 10p2 ,
dt
which we solved as a separable equation. Show that this equation is also a Bernoulli
differential equation and use the transformation about to solve it.
Solution:
Rewriting the equation in the form
dp
− 60p = −10p2
dt
we see that it is a Bernoulli differential equation with n = 2. To solve, let

v = p1−2 = p−1

then
dv dp
= −p−2
dt dt
Using the differential equation we get
dv dp
= −p−2
dt dt
−2
= −p [60p − 10p2 ]
= −60p−1 + 10
= −60v + 10
dv
⇒ + 60v = 10
dt
This is a linear differential equation for v(t) with integrating factor
R
60dt
µ(t) = e = e60t ,

hence we may solve it as described in the previous section.

dv
e60t + 60e60t v = 10e60t
dt
d 60t
[e ] = 10e60t
dt
1 60t
e60t v = e + c1
6
1
v = + c1 e−60t
6
1 + Ce−60t
v =
6

24
 1 6
⇒ p(t) = =
v 1 + Ce−60t
This is exactly the general solution we found by solving the differential equation as
a separable equation (see Example 2.12).

2.3.2 *A graphical/qualitative approach

For many systems we model with differential equations we are most interested in
the qualitative behaviour of the solution, that is

1. Are there equilibrium solutions?

2. Are the equilibrium solutions stable or unstable?

3. What is the behaviour of all solutions as t → ∞?

For some differential equations we can answer these equations without solving the
differential equation, just by using the properties of the differential equation itself.
We will restrict our attention to separable differential equations of the form

y 0 = f (y) (2.17)

where f is a continuously differentiable function of y (that is, f 0 (y) is continuous

and f has no dependence on the independent variable). This guarantees that for
any initial condition y(x0 ) = y0 there is a unique solution.
The answer to question 1 we’ve seen before: The equilibrium solutions of (2.17)
are the constant functions y = k, −∞ < x < ∞ where k satisfies f (k) = 0. To
answer questions 2 and 3 we use the properties of the function f (y) given below.

Property 1: The equilibrium solutions of y 0 = f (y) are the zeros of the function
f (y).
dy
Property 2: Since dx
= f (y) we have

• y(x) is increasing when f (y) > 0

• y(x) is decreasing when f (y) < 0

Property 3: Observe that

d2 y
 
d dy d dy
2
= = [f (y)] = f 0 (y) = f 0 (y)f (y)
dx dx dx dx dx
thus we have

25
 If f (y) is then y(x) is
positive and increasing increasing, concave up
positive and decreasing increasing, concave down
negative and increasing decreasing, concave down
negative and decreasing decreasing, concave up

Table 2.2: How the graph of the solutions y(x) of (2.17) is related to the properties
of f (y).

range f (p) f 0 (p) properties of f (p) properties of p(t)

p<0 <0 >0 negative, increasing decreasing, concave down
0<p<3 >0 >0 positive, increasing increasing, concave up
3<p<6 >0 <0 positive, decreasing increasing, concave down
p>6 <0 <0 negative, decreasing decreasing, concave up

Table 2.3: Properties of solutions for Example 2.21.

• y(x) is concave up when f 0 (y)f (y) > 0

• y(x) is concave down when f 0 (y)f (y) < 0

Properties 1 and 2 are summarized in Table 2.2 The information from Property 1
together with that in Table 2.2 is enough to give a qualitative sketch of the solution
of the differential equation.
Example 2.21:
Consider the differential equation
dp
= 10p(6 − p) = f (p).
dt
We’ve solved this in two ways already (see Examples 2.12 and 2.20). We will now
show how to apply the qualitative approach to this equation. Note the equilibrium
points are the zeros of f (p), that is p = 0 and p = 6. Now compute f 0 (p)

f 0 (p) = 60 − 20p.
2
Since ddt2p = f (p)f 0 (p) we can use Table 2.2 to determine how the properties of p(t)
relate to the values of p. This information is summarized in Table 2.3. We use this
information to qualitatively sketch the graph of p(t) in the various regions as in
Figure 2.5.

26
 p

p=6

p=0
t

Figure 2.5: Qualitative sketches of various solutions to the differential equation

dp
dt
= 10p(6 − p). Note how the behaviour matches the description in Table 2.3.

The power of this method is that it works on equations that we can’t solve or
that are hard to solve.
Example 2.22:
Give a qualitative sketch of the solutions of
dy
= (y − 1)2 (y − 2).
dx

Solution:
In this problem
f (y) = (y − 1)2 (y − 2)
and

f 0 (y) = 2(y − 1)(y − 2) + (y − 1)2

= (y − 1)(3y − 5)

Thus the equation has two equilibrium solutions: p = 1 and p = 2. The properties
of the solutions are as given in Table 2.4. Using the information in Table 2.4 we
can give a qualitative sketch of the solutions as shown in Figure 2.6. From this
sketch it is clear that the equilibrium points are both unstable.

27
 Range in y f (y) is so y(x) is
y<1 negative, increasing decreasing, concave down
1 < y < 53 negative, decreasing decreasing, concave up
5
3
<y<2 negative, increasing decreasing, concave down
y>2 positive, increasing increasing, concave up

Table 2.4: Properties of solutions for Example 2.22.

2.5

y=5/3

1.5

0.5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

Figure 2.6: Qualitative sketches of various solutions to the differential equation

y 0 = (y − 1)2 (y − 2). Note how the behaviour matches the description in Table 2.4.

28
Chapter 3

Higher Order Linear Ordinary

Differential Equations

The most general nth order ordinary differential equation is a relationship of the
form
G(x, y, y 0 , y 00 , . . . , y (n) ) = 0 (3.1)
where
dy 00 d2 y dn y
y0 = , y = 2 , . . . , y (n) = n .
dx dx dx
We will focus on linear differential equations. These are equations where G is a
linear function of y, y 0 , . . . , y (n) . Any linear nth order differential equation can be
written in the form

an (x)y (n) + an−1 (x)y (n−1) + · · · + a1 (x)y 0 + a0 (x)y = F (x). (3.2)

Definition 3.1: If F (x) ≡ 0 then equation (3.2) is called homogeneous otherwise

it is called inhomogeneous.
Definition 3.2: If the functions ai (x) are constant functions, i.e., ai (x) = Ai , x ∈
R, then we call equation (3.2) a constant coefficient, linear ordinary differential
equation, otherwise we call it a variable coefficient, linear ordinary differential
equation.

Two examples of higher order linear ordinary differential equations are:

cos(x)y 00 + sin(x)y 0 + 3y = tan(x) : 2nd order, variable coefficient

d3 y dy
3
+ 10 + 5y = 0 : 3rd order, constant coefficient
dx dx

29
3.1 Theory of nth Order Linear ODEs

We now define various types of solutions of the differential equation (3.2).

Definition 3.3: A solution of equation (3.2) is any n times differentiable function
y = ψ(x) such that ψ, ψ 0 , ψ 00 , . . . , ψ (n) exist and satisfy equation (3.2) for all x in
some open interval I.
Definition 3.4: The interval I in Definition 3.3 is called the interval of existence
or interval of definition of the solution.
Definition 3.5: The general solution equation (3.2) is a function ψ(x; c1 , c2 , . . . , cn )
involving n arbitrary constants which represents almost all solutions of the equa-
tion.
Definition 3.6: A solution which is not represented by the general solution is
called a singular solution.
Definition 3.7: A particular solution of (3.2) is a solution which contains no
arbitrary constants.

Note: An initial value problem for equation (3.2) must have n initial conditions:

y(x0 ) = p0 , y 0 (x0 ) = p1 , . . . , y (n−1) (x0 ) = pn−1 . (3.3)

Theorem 3.8 (Existence and Uniqueness). If there is an open interval I such that

1. the functions a0 (x), a1 (x), . . . , an (x) and F (x) are continuous on I,

2. an (x) 6= 0 on I,

then, for any x0 ∈ I, the initial value problem consisting of equations (3.2)–(3.3)
has a unique solution on I.

Proof. See Chapter 6 of Coddington [3, 4] or Appendix 2 of Edwards and Penney

[2].

Example 3.9:
Consider the initial value problem

cos(x)y 00 + sin(x)y 0 + 3x2 y = 29, y(0) = 4, y 0 (0) = 3.

Here we have n = 2,

a2 (x) = cos(x), a1 (x) = sin(x), a0 (x) = 3x2 , F (x) = 29,

30
so the aj (x) and F (x) are all continuous on IR. Further a2 (x) = cos(x) is nonzero
on any interval not containing one of the points (2k + 1)π/2, k an integer. Since
x0 = 0, the conditions of Theorem 3.8 are satisfied on the interval I = (− π2 , π2 ).
Thus we can expect there will be a unique solution of the initial value problem on
this interval.

In order to facilitate our discussion of the theory of nth order ordinary differential
equations, we will introduce the following concept.
Definition 3.10: An operator is a transformation which maps functions to func-
tions

Notation: Let T be an operator that maps the function f to the function g, then
we write
T [f (x)] = g(x), or T f (x) = g(x), or T f = g.
The domain and range of an operator are sets of functions. Some important oper-
ators we will use include the following.

1. The derivative operator, D, takes as input a differentiable function and gives

as output the derivative of the input function.
D[sin(x)] = cos(x)
De2x = 2e2x
Df (x) = f 0 (x)
In mapping notation we have
D : C 1 (R) → C(R)
where C j denotes the set of functions that are j times continuously differen-
tiable.

2. The identity operator, I, maps every function to itself.

I[f (x)] = f (x)

We can define new operators by adding or composing operators as in the following

example.
Example 3.11:
Let M = D + I then
M [f (x)] = (D + I)[f (x)]
= D[f (x)] + I[f (x)]
= f 0 (x) + f (x).

31
Let D2 = D ◦ D then

D2 [f (x)] = D[D[f (x)]] = D[f 0 (x)] = f 00 (x).

Similarly
dn f
Dn [f (x)] = D[D[· · · D[f (x)] · · · ]] = f (n) (x) = .
dxn
Using these ideas we may write equation (3.2) in operator notation as

an (x)Dn y + an−1 (x)Dn−1 y + an−2 (x)Dn−2 y + . . . + a1 (x)Dy + a0 (x)y = F (x)

or

(an (x)Dn + an−1 (x)Dn−1 + an−2 (x)Dn−2 + . . . + a1 (x)D + a0 (x)I) y = F (x) (3.4)

Thus, if we define the operator

φ(D) = an (x)Dn + an−1 (x)Dn−1 + an−2 (x)Dn−2 + . . . + a1 (x)D + a0 (x)I

we can write equation (3.2) in the very compact form

φ(D)y = F (x). (3.5)

We can then interpret solving the differential equation as finding the functions y
that are mapped to the function F by the operator φ(D). Note that the operator
φ(D) maps n times differentiable functions to continuous functions, i.e.,
φ(D) : C n (R) → C(R).
Example 3.12:
The equation
cos(x)y 00 + sin(x)y 0 + 3y = tan(x),
can be written in operator notation as φ(D)y = tan(x) where
φ(D) = cos(x)D2 + sin(x)D + 3I.
The equation
d3 y dy
3
+ 10 + 5y = 0,
dx dx
can be written in operator notation as φ(D) = 0 where φ(D) = D3 + 10D + 5I.
Definition 3.13: An operator, T , is called linear if it has the following property.
Given two functions f1 (x) and f2 (x) (in the domain of T ) and two constants c1 ,
c2 ∈ R
T [c1 f1 (x) + c2 f2 (x)] = c1 T [f1 (x)] + c2 T [f2 (x)].

32
Exercise 3.14:
Show that Dj is a linear operator for any j = 1, 2, . . .
Exercise 3.15:
Show that φ(D) is a linear operator.

The following exercise illustrates the property that operators can be factored to
yield equivalent operators.
Exercise 3.16:
Verify that the operators

(D − 1)(D − 2) and D2 − 3D + 2

are equal by applying them to an arbitrary function f ∈ C 2 (R).

To study the solutions of equation (3.2) or equivalently equation (3.5), we con-

sider the related differential equation

an (x)y (n) + an−1 (x)y (n−1) + · · · + a1 (x)y 0 + a0 (x)y = 0 (3.6)

or
φ(D)y = 0
This is called the complementary equation or the associated homogeneous
equation of equation (3.5).

Theorem 3.17. If u(x) is a solution of equation (3.5) on I1 and v(x) is a solution

of equation (3.6) on I2 then y = u(x) + v(x) is a solution of equation (3.5) on
I = I1 ∩ I2 .

Proof. Apply the operator φ(D) to u(x) + v(x) with x ∈ I:

φ(D)[u(x) + v(x)] = φ(D)[u(x)] + φ(D)[v(x)] (by linearity)

= F (x) + 0 (using (3.5) and (3.6))

It follows that u(x) + v(x) is a solution of (3.5) on I

Let yp (x) be a particular solution of equation (3.5) and yh (x) be the general
solution of the associated homogeneous equation (3.6), then, by Theorem 3.17
y(x) = yp (x) + yh (x) is a solution of (3.5) containing n arbitrary constants. To
show it is the general solution, we need to show every nonsingular solution of (3.5)
can be written in this form. To do this we first obtain some results about the
general solution of (3.6).

33
3.2 The General Solution of the Homogeneous
Equation

Theorem 3.18 (Principle of Superposition). Given k solutions, y1 (x), . . . , yk (x), of

the homogeneous equation (3.6) on the interval I and k constants, c1 , c2 , . . . , ck ∈ R,
the linear combination

c1 y1 (x) + c2 y2 (x) + · · · + ck yk (x)

is also a solution of equation (3.6) on I.

Proof. The proof follows from the linearity of the operator φ(D). For x ∈ I we
have

φ(D)[c1 y1 (x) + c2 y2 (x) + · · · + ck yk (x)] = c1 φ(D)[y1 (x)] + · · · + ck φ(D)[yk (x)]

= c1 · · · 0 + c2 · · · 0 + · · · + ck · · · 0
= 0

From Theorem 3.18, it follows that if y1 (x), . . . , yn (x) are n solutions on I of

φ(D)[y] = 0 and c1 , . . . , cn are n constants then

c1 y1 (x) + · · · + cn yn (x)

is a solution on I with n arbitrary constants. Is it the general solution? To answer

this we need another concept.
Definition 3.19: The functions f1 (x), . . . , fn (x) are said to be linearly depen-
dent on the interval I if there exist n constants α1 , . . . , αn such that not all of
the αj are zero and

α1 f1 (x) + α2 f2 (x) + · · · + αn fn (x) = 0

for all x in I. Otherwise, they are said to be linearly independent on I.

Example 3.20:
The functions e2x + ex , 5e2x + 4ex , ex − e2x are linearly dependent since

9(e2x + ex ) − 2(5e2x + 4ex ) + (ex − e2x ) = 0

34
Example 3.21:
The functions cos x, sin(x) are linearly independent on R. To check this consider

α1 cos(x) + α2 sin(x) = 0. (3.7)

When x = 0 we must have α1 = 0 to satisfy equation (3.7). Similarly, when x = π2

we must have α2 = 0 to satisfy equation (3.7). Therefore the only possible choice
of α1 and α2 to satisfy equation (3.7) for all values of x is α1 = α2 = 0.
Exercise 3.22:
Show that the functions cos x, cos(2x) are linearly independent on R.
Exercise 3.23:
Show that the functions 1, cos2 (x), cos(2x) are linearly dependent on R.

The definition of linear independence is not always easy to check, so we will

consider an alternate way of checking linear independence.
Definition 3.24: The Wronskian of n functions, f1 (x), . . . , fn (x) is the n × n
determinant
 
f1 (x) f2 (x) · · · fn (x)
 f10 (x) f20 (x) · · · fn0 (x)
 

W (f1 , . . . , fn ) = W (x) = det  .. .. .. .
. . .
 
 
(n−1) (n−1) (n−1)
f1 (x) f2 ··· fn (x)

Example 3.25:

1. Let f1 = cos(x), f2 = sin(x). Then

" #
cos(x) sin(x)
W (f1 , f2 ) = det = cos2 (x)+sin2 (x) = 1, for all x ∈ R.
− sin(x) cos(x)

2. Let f1 = 1, f2 = cos2 (x), f3 = cos(2x). Then

1 cos2 (x) cos(2x)

W (f1 , f2 , f3 ) = 0 −2 cos(x) sin(x) −2 sin(2x) = 0, for all x ∈ R.
2 2
0 2 sin (x) − 2 cos (x) −4 cos(2x)

3. Let f1 = cos(x), f2 = cos(2x). Then

" #
cos(x) cos(2x)
W (f1 , f2 ) = det = sin(x)(1 + 2 cos2 (x)).
− sin(x) −2 sin(2x)

35
 The following result shows how the Wronskian and linear independence are
related for functions which are solutions of the linear ODE (3.6).

Proposition 3.26. Let y1 (x), y2 (x), . . . , yn (x) be n solutions on the interval I of

the differential equation (3.6). Then y1 (x), y2 (x), . . . , yn (x) are linearly independent
on I if and only if W (y1 , y2 , . . . , yn ) 6= 0 for all x ∈ I.

Proof. Suppose that W (y1 , y2 , . . . , yn ) 6= 0 for all x ∈ I. Consider the linear system
    
y1 (x) y2 (x) ··· yn (x) α1 0
0 0 0
 y1 (x) y2 (x) ··· yn (x)   α2   0 
    
 .  .  =  . . (3.8)

 .
.
 .   . 
 .   . 
(n−1) (n−1) (n−1)
y1 (x) y2 (x) · · · yn (x) αn 0

The determinant of the coefficient matrix of the system is W (y1 , y2 , . . . , yn ), which

is nonzero for all x ∈ I. Hence, for each x ∈ I the linear system (3.8) has only one
solution: α1 = α2 = · · · = αn = 0. In particular, the first equation of the system,

α1 y1 (x) + α2 y2 (x) + · · · + αn yn (x) = 0,

is satisfied for all x ∈ I only if α1 = α2 = · · · = αn = 0. Thus y1 (x), y2 (x), . . . , yn (x)

are linearly independent.
Now suppose y1 (x), y2 (x), . . . , yn (x) are linearly independent. We need to show
W (y1 , y2 , . . . , yn ) 6= 0 on I. Suppose the opposite, i.e., there is at least one point
x0 ∈ I where the Wronskian is zero. At this point the linear system (3.8) has a
nonzero solution, i.e.,

α1 = ᾱ1 , α2 = ᾱ2 , · · · , αn = ᾱn (3.9)

where not all of the ᾱj are zero. Let

u(x) = ᾱ1 y1 (x) + ᾱ2 y2 (x) + · · · + ᾱn yn (x).

Then u(x) satisfies the initial value problem consisting of (3.6) and the initial
conditions
y(x0 ) = 0, y 0 (x0 ) = 0, · · · , y (n−1) (x0 ) = 0
on the interval I. (The initial conditions come from putting (3.9) into (3.8) and
evaluating at x0 .) However, the function y(x) = 0, x ∈ I also satisfies this initial
value problem. By Theorem 3.8 the initial value problem has a unique solution,
hence we must have u(x) = 0, x ∈ I, i.e.,

ᾱ1 y1 (x) + ᾱ2 y2 (x) + · · · + αn yn (x) = 0, for all x ∈ I.

36
Since not all of the ᾱj are zero, this implies that y1 (x), y2 (x), . . . , yn (x) are linearly
dependent. This is a contradiction, so we must have W (y1 , y2 , . . . , yn ) 6= 0 on I.

We can now state and prove the main theorem.

Theorem 3.27. Suppose that the conditions of Theorem 3.8 are satisfied on some
interval I and let y1 (x), y2 (x), . . . , yn (x) be n linearly independent solutions on I
of the differential equation (3.6). Then the general solution is

y(x) = c1 y1 (x) + c2 y2 (x) + · · · + cn yn (x), x∈I (3.10)

where c1 , c2 , . . . , cn are arbitrary constants.

Proof. Let ψ(x) be any solution of equation (3.6) on I, let x0 ∈ I and consider the
initial value problem consisting of (3.6) and the initial conditions

y(x0 ) = ψ(x0 ), y 0 (x0 ) = ψ 0 (x0 ), . . . , y (n−1) (x0 ) = ψ (n−1) (x0 ). (3.11)

Clearly, ψ(x) satisfies this initial value problem. Further y(x) defined by (3.10)
satisfies the ODE (3.6) for any constants c1 , . . . , cn . Applying the initial conditions
(3.11) to y(x) yields a linear system of equations for c1 , . . . , cn :

c1 y1 (x0 ) + · · · + cn yn (x0 ) = ψ(x0 )

c1 y10 (x0 ) + · · · + cn yn0 (x0 ) = ψ 0 (x0 )
..
.
(n−1)
c1 y 1 (x0 ) + · · · + cn yn(n−1) (x0 ) = ψ (n−1) (x0 )

or, in matrix form:

    
y1 (x0 ) y2 (x0 ) ··· yn (x0 ) c1 ψ(x0 )
0 0
 y1 (x0 ) y2 (x0 ) ··· yn0 (x0 ) c2 ψ 00 (x0 )
    
   
 ..  .. = ..  . (3.12)
. . .
    
    
(n−1) (n−1) (n−1)
y1 (x0 ) y2 (x0 ) · · · cn y n (x0 ) cn ψ (n−1) (x0 )

Now the determinant of the coefficient matrix of this system is W (y1 , y2 , . . . , yn )

evaluated at x0 , which is nonzero since y1 , y2 , . . . , yn are linearly independent. Hence
the system (3.12) has a unique solution

c1 = c̄1 , c2 = c̄2 , . . . , cn = c̄n .

Thus the initial value problem consisting of (3.6) and (3.11) also has the solution

c̄1 y1 (x) + c̄2 y2 (x) + · · · + c̄n yn (x).

37
But, by Theorem 3.8 the initial value problem has a unique solution on I, therefore

ψ(x) = c̄1 y1 (x) + c̄2 y2 (x) + · · · + c̄n yn (x).

Thus every solution of (3.6) on I can be written in the form (3.10), i.e., (3.10) is
the general solution of (3.6).

Note that this proof shows that every solution of (3.6) can be written in the
form (3.10). This implies that, under the conditions of the theorem, there are no
singular solutions of (3.6). Finally, we state a result that gives a useful formula for
the Wronskian.

Proposition 3.28 (Abel’s Formula). Suppose that conditions 1 and 2 of Theo-

rem 3.8 are satisfied and let y1 (x), y2 (x), . . . , yn (x) be n solutions on the interval I
of the nth order linear ODE (3.6). Then, for any x0 ∈ I, the Wronskian of these
functions satisfies
Rx an−1 (x)
− dx
W (y1 , y2 , . . . , yn ) = W (x0 )e x0 an (x)

for all x ∈ I.

Proof. To begin, consider the derivative of W with respect to x:

y10 (x) ··· yn0 (x) y1 (x) ··· yn (x)

y10 (x) ··· yn0 (x) y100 (x) ··· yn00 (x)
y100 (x) ··· yn00 (x) y100 (x) ··· yn00 (x)
W 0 (x) = .. .. + .. ..
. . . .
(n−2) (n−2) (n−2) (n−2)
y1 (x) · · · yn (x) y1 (x) · · · yn (x)
(n−1) (n−1) (n−1) (n−1)
y1 (x) · · · yn (x) y1 (x) · · · yn (x)

y1 (x) ··· yn (x) y1 (x) ··· yn (x)

y10 (x) ··· yn0 (x) y10 (x) ··· yn0 (x)
y100 (x) ··· yn00 (x) y100 (x) ··· yn00 (x)
+··· + .. .. + .. ..
. . . .
(n−1) (n−1) (n−2) (n−2)
y1 (x) · · · yn (x) y1 (x) · · · yn (x)
(n−1) (n−1) (n) (n)
y1 (x) · · · yn (x) y1 (x) · · · yn (x)

38
From a standard property of determinants, the first n − 2 determinants are zero
and thus
y1 (x) ··· yn (x)
0
y1 (x) ··· yn0 (x)
y100 (x) ··· yn00 (x)
W 0 (x) = .. .. .
. .
(n−2) (n−2)
y1 (x) · · · yn (x)
(n) (n)
y1 (x) · · · yn (x)
Since each yj (x) satisfies (3.6) we have
n−1
!
(n) 1 X (i)
yj (x) = − ai (x)yj (x) ,
an (x) i=0

and hence

y1 (x) ··· yn (x)

y10 (x) ··· yn0 (x)
y100 (x) ··· yn00 (x)
W 0 (x) = .. .. .
. .
(n−2) (n−2)
y1
P (x)  ··· yn
P (x) 
n−1 (i) n−1 (i)
− an1(x) i=0 ai (x)y1 (x) ··· − an1(x) i=0 ai (x)yn (x)

Using standard properties of determinants this simplifies to

y1 (x) ··· yn (x)

y10 (x) ··· yn0 (x)
an−1 (x) y100 (x) ··· yn00 (x)
W 0 (x) = − .. .. .
an (x) . .
(n−2) (n−2)
y1 (x) · · · yn (x)
(n−1) (n−1)
y1 (x) · · · yn (x)

Using the definition of W this equation may be written

an−1 (x)
W0 = − W.
an (x)

Solving this equation with initial condition at x = x0 ∈ I yields the result.

For the purposes, the principal significance of Abel’s Formula is that it implies the
following: if W (x0 ) = 0, for any x0 ∈ I, then W (x) = 0 on I

39
3.3 The General Solution of the Inhomogeneous
Equation
Theorem 3.29. If yh (x) is the general solution of equation (3.6) and yp (x) is a
particular solution of equation (3.5) then yh (x) + yp (x) is the general solution of
the equation (3.5).

Proof. If follows from Theorem 3.17 that yh (x) + yp (x) is a solution of equation
(3.5). Further, since yh (x) is the general solution of the associated homogeneous
equation (3.6), it contains n arbitrary constants i.e., yh (x) = v(x; c1 , . . . , cn ). So
yh (x) + yp (x) is a solution of equation (3.5) containing n arbitrary constants.
To show that yh (x) + yp (x) is the general solution of equation (3.5) we need to
show that every solution of equation (3.5) can be written in the form yh (x) + yp (x)
for some particular values of the constants.
Let ψ(x) be any solution of (3.5) and let w(x) = ψ(x) − yp (x). Then

φ(D)[w(x)] = φ(D)[ψ(x) − yp (x)]

= φ(D)[ψ(x)] − φ(D)[yp (x)]
= F (x) − F (x) = 0.

So w(x) is a solution of the associated homogeneous equation. Thus w(x) = yh (x)

for some particular choice of the constants, e.g., w(x) = v(x; K1 , . . . , Kn ). Thus we
have

ψ(x) = w(x) + yp (x)

= v(x; K1 , . . . , Kn ) + yp (x).

In other words, ψ(x) can be written in the form yh (x) + yp (x) for some appropriate
choice of the arbitrary constants.

Summary - Solving the inhomogeneous equation (3.4)

Theorems 3.27 and 3.29 implies that the general solution of a linear, nth order
ordinary differential equation can be obtained using the following steps.

1. Find n linearly independent solutions y1 , y2 , . . . , yn on I, i.e. n solutions such

that
W (y1 , y2 , · · · , yn ) 6= 0 on I.
Note that Proposition 3.28 implies that one need only check the Wronskian
at one value of x in I.

40
 Form the general solution of (3.6):

yh (x) = c1 y1 (x) + c2 y2 (x) + · · · + cn yn (x), x∈I

2. Find a particular solution, yp (x) of the equation (3.5)

3. Form the general solution of (3.5):

y(x) = yh (x) + yp (x)

We will consider methods for step 1 in the following section. Approaches for step
2 will be considered in section 3.5.

3.4 Constant Coefficient nth Order Homogeneous

Ordinary Differential Equations

It is difficult to find solutions of linear, variable coefficient, higher order ordinary dif-
ferential equations. Thus we will now specialize to the case of constant coefficients.
The general form of a constant coefficient nth order linear ordinary differential
equation is
An y (n) + An−1 y (n−1) + · · · + A1 y 0 + A0 y = F (x) (3.13)
where An 6= 0, Aj ∈ R, j = 0, . . . , n. In operator form we write this as

φ(D)[y] = F (x)

where
φ(D) = An Dn + An−1 Dn−1 + · · · + A1 D + A0 I. (3.14)
The associated homogeneous equation is

An y (n) + An−1 y (n−1) + · · · + A1 y 0 + A0 y = 0, (3.15)

or
φ(D)[y] = 0.
To motivate our method of solution for (3.15), consider the case when n = 1:
dy
A1 + A0 y = 0.
dx
Rewriting this as
dy A0
= − y,
dx A1

41
 A0
− x
we can solve this as a linear or separable equation to find y = Ce A1 . Thus the
solution of any first order, linear, constant coefficient ODE will be an exponential
function. This motivates us to try this same form of solution for higher order,
linear, constant coefficient ODEs.
Example 3.30:
Find the general solution of the differential equation y 00 − y 0 − 2y = 0.
Solution:
Let y = eλx where λ is a constant to be determined. Taking the derivatives of y(x):

y 0 = λeλx
y 00 = λ2 eλx

we substitute them into the differential equation to find

λ2 eλx − λeλx − 2eλx = 0

or
(λ2 − λ − 2)eλx = 0
Since eλx 6= 0 for all x we require λ to satisfy

λ2 − λ − 2 = 0.

This is called the auxiliary or characteristic equation of the differential equation.

Any root of this equation will determine a solution of the differential equation.
Using the quadratic formula it is easy to check that the roots of auxiliary equation
are λ = 2 and λ = −1. Thus two solutions of the differential equation are

y1 (x) = e2x , y2 (x) = e−x .

Calculating the Wronskian, we have

" #
e2x e−x
W (y1 , y2 ) = det 2x −x
= −ex − 2ex = −3ex 6= 0, for all x.
2e −e

So the solutions are linearly independent on (−∞, ∞) and the general solution is

y(x) = c1 e2x + c2 e−x , −∞ < x < ∞.

Now consider a general nth order, linear, constant coefficient ODE:

(An Dn + An−1 Dn−1 + · · · + A1 D + A0 I)[y] = 0. (3.16)

42
Substituting y = eλx into the differential equation and using the fact that Dk [eλx ] =
dk eλx
dxk
= λk eλx yields the auxiliary equation

An λn + An−1 λn−1 + · · · + A1 λ + A0 = 0 (3.17)

Since the right hand side of this equation is an degree n polynomial in λ we know
from a result from Linear Algebra that the equation will have n roots. This leads
to the following.
Summary - Solving the constant coefficient equation (3.16)

1. Find the auxiliary equation (3.17).

2. Find the n roots of the auxiliary equation.

3. Find n linearly independent solutions of (3.16) using the n roots.

Step 3 is not always trivial. In the following we will develop several theorems to
handle this. The first theorem deals with the simplest situation.

Proposition 3.31. Let λ1 , λ2 , λ3 , . . . , λn be n real, distinct roots of the auxiliary

equation (3.17) (i.e., λ1 6= λ2 6= · · · 6= λn ). Then the functions eλ1 x , eλ2 x , . . . , , eλn x
are linearly independent solutions of equation (3.16) on R and the general solution
of (3.16) is

yh (x) = c1 eλ1 x + c2 eλ2 x + · · · + cn eλn x , −∞ < x < ∞,

where c1 , c2 , . . . , cn are arbitrary constants.

Proof. Consider first the case where n = 2 and suppose that λ1 6= λ2 are the two
roots of the auxiliary equation. Calculating the Wronskian we have
" #
λ1 x λ2 x
e e
W (eλ1 x , eλ2 x ) = det λ1 x λ2 x
= (λ2 − λ1 )e(λ1 +λ2 )x = −(λ1 − λ2 )e(λ1 +λ2 )x
λ1 e λ2 e

Clearly W 6= 0 for all x ∈ R, since λ1 6= λ2 .

Now consider the case of general n. One can show using Mathematical Induction
that

W (eλ1 x , eλ2 x , . . . , eλn x ) = ±(λ1 − λ2 )(λ1 − λ3 ) · · · (λ1 − λn )(λ2 − λ3 ) · · · (λ2 − λn )

· · · (λn−1 − λn )e(λ1 +λ2 +···+λn )x .

Clearly W 6= 0 for all x ∈ R since the λj are distinct. The result then follows from
Theorem 3.27.

43
 Two issues may arise that make step 3 more complicated.

1. The roots may not be distinct, i.e., the auxiliary equation may have repeated
roots (roots of multiplicity > 1).

2. The roots may be complex.

The following propositions show how to deal with the first issue.

Proposition 3.32 (Method of Reduction of Order). If y1 (x) is one nonzero solution

of the homogeneous equation (3.6) then a second, linearly independent solution may
be obtained by the substitution y2 (x) = u(x)y1 (x).

Proof. (For the case n = 2) Let y2 (x) = u(x)y1 (x) where y1 (x) is a solution of
equation (3.6). Substituting y2 (x) into (3.6) yields

a2 (x)[u00 y1 + 2u0 y10 + uy100 ] + a1 (x)[u0 y1 + uy10 ] + a0 (x)uy1 = 0.

Rearranging this becomes

a2 (x)y1 u00 + [2a2 (x)y10 + a1 (x)y1 ]u0 + [a2 (x)y100 + a1 (x)y10 + a0 (x)y1 ]u = 0.

Now the last term is zero, since y1 (x) is solution of equation (3.6). Hence that y2 (x)
will be a solution of equation (3.6) if u(x) is a solution of

a2 (x)y1 (x)u00 + [2a2 (x)y10 (x) + a1 (x)y1 (x)]u0 = 0.

But this equation can always be solved as we now describe. Substituting v(x) =
u0 (x), v 0 (x) = u00 (x) yields a first order, linear equation for v(x):

a2 (x)y1 (x)v 0 + [2a2 (x)y10 (x) + a1 (x)y1 (x)]v = 0.

This equation can be solve for v(x) using the method of section 2.2. u(x) may then
be obtained by integration.
To see that the two solutions are linearly independent consider the Wronskian
" #
y1 (x) u(x)y1 (x)
W (y1 , y2 ) = det 0 0 0
= −u0 (x)y1 (x)2 .
y1 (x) u (x)y1 (x) + u(x)y1 (x)

Now y1 (x) 6= 0 by assumption and, since u(x) is not a constant, u0 (x) 6= 0. Thus
W 6= 0 and the two solutions are linearly independent.

44
Proposition 3.33 (Repeated Roots). If λ = b is a repeated root (root of multiplic-
ity 2) of the auxiliary equation (3.17), then two linearly independent solutions on
(−∞, ∞) of equation (3.16) are y1 (x) = ebx and y2 (x) = xebx .

Proof. Since b is a root of multiplicity 2, the auxiliary equation can be written

p(λ)(λ − b)2 = 0

or
p(λ)(λ2 − 2bλ + b2 ) = 0,
where p(λ) is a polynomial of degree n − 2 with real coefficients and p(b) 6= 0. This
means that the differential equation can be written

p(D)[(D2 − 2bD + b2 I)[y]] = 0.

We know that y1 (x) = ebx is one solution of the differential equation. We will use
the Proposition 3.32 to find another.
Let y = u(x)ebx . Substituting into the differential equation we obtain

p(D)[(D2 − 2bD + b2 I)[uebx ]] = 0

p(D)[D[u0 ebx + buebx ] − 2b[u0 ebx + buebx ] + b2 uebx ] = 0
p(D)[u00 ebx + 2bu0 ebx + b2 uebx − 2bu0 ebx − 2b2 uebx + b2 uebx ] = 0
p(D)[u00 ebx ] = 0.

Clearly this last equation will be satisfied, and hence u(x)ebx will be a solution
of equation (3.16), if u00 (x) = 0. Integrating two times with respect to x yields
u(x) = c1 x + c2 , where c1 , c2 are arbitrary constants. Hence

u(x)ebx = c1 xebx + c2 ebx .

The second term on the right-hand side is just a multiple of y1 and so the simplest
form of the second solution is
y2 (x) = xebx .

Calculating the Wronskian,

" #
ebx xebx
W (y1 , y2 ) = det bx bx bx
= e2bx + bxe2bx − bxe2bx = e2bx 6= 0 ∀x ∈ R,
be e + bxe

shows that the two solutions are linearly independent.

45
Exercise 3.34:
Check y2 = xebx is a solution of

p(D)[(D2 − 2bD + b2 I)[y]] = 0.

where p(D) is a degree n − 2 polynomial in D with real coefficients.

Example 3.35:
Find the general solution of y 000 − 3y 0 + 2y 0 = 0.
Solution:
First, write in equation operator form

(D3 − 3D + 2I)[y] = 0.

Then the auxiliary equation is

λ3 − 3λ + 2 = 0.

A simple check shows that λ = −2 is one root. Factoring the polynomial yields

(λ − 1)2 (λ + 2) = 0.

The roots are therefore λ = 1, 1, −2 with corresponding solutions

y1 = ex , y2 = xex , y3 = e−2x .

One can verify the linear independence of these solutions using the Wronskian or
the definition (exercise). So the general solution is

yh = c1 ex + c2 xex + c3 e−2x , −∞ < x < ∞.

Proposition 3.36. If λ = b is a root of multiplicity m > 1 of the auxiliary equation

(3.17), a set of m linearly independent solutions is given as follows

Multiplicity Solutions
2 ebx , xebx
2
3 ebx , xebx , x2 ebx
.. ..
. .
2 m−1
m x
ebx , xebx , x2 ebx , · · · , (m−1)! ebx

Proof. Similar to Proposition 3.33. Use the Method of Reduction of Order to the
find solutions and the Wronskian to show that they are linearly independent.

46
Note: If one has several distinct roots of different multiplicities one can use the
Wronskian to show that the solutions determined by Proposition 3.36 are linearly
independent.
Example 3.37:
Find the general solution of y 0000 − 10y 000 + 37y 00 − 60y 0 + 36y = 0.
Solution:
In operator form the differential equation is

(D4 − 10D3 + 37D2 − 60D + 36I)[y] = 0,

so the auxiliary equation is

λ4 − 10λ3 + 37λ2 − 60λ + 36 = 0.

This may be factored as

(λ − 2)2 (λ − 3)2 = 0.
So the roots are λ = 2, 2, 3, 3 and the corresponding solutions are

y1 = e2x , y2 = xe2x , y3 = e3x , y4 = xe3x .

One can verify the linear independence of these solutions using the Wronskian or
the definition (exercise). So the general solution is

y(x) = c1 e2x + c2 xe2x + c3 e3x + c4 xe3x , −∞ < x < ∞.

The next proposition tells us how to deal with complex roots of the auxiliary
equation.

Proposition 3.38. If the auxiliary equation (3.17) has a root λ = α + iβ with

α, β ∈ R and β 6= 0 then

1. it also has a root λ = α − iβ

2. two linearly independent solutions of the differential equation on −∞ < x <

∞ are
eαx cos βx and eαx sin βx.

Proof. Result 1 follows from the fact that the coefficients of the auxiliary equation
are real and the Complex Conjugate Root Theorem.

47
 To prove 2, let

u(x) = eλx = e(α+iβ)x = eαx eiβx = eαx (cos βx + i sin βx)

and
v(x) = eλx = e(α−iβ)x = eαx e−iβx = eαx (cos βx − i sin βx).
Now u(x) and v(x) are solutions on R of the differential equation since λ, λ are
roots of the auxiliary equation.
Recall from Theorem 3.18 that any linear combination of two solutions of the
differential equation is also a solution, thus
1 1
y1 (x) = u(x) + v(x) = eαx cos βx
2 2
and
1 1
y2 (x) = u(x) − v(x) = eαx sin βx
2i 2i
are also solutions on R of the differential equation. Further
" #
eαx cos βx eαx sin βx
W (y1 , y2 ) = det
αeαx cos βx − βeαx sin βx αeαx sin βx + βeαx cos βx
= αe2αx cos βx sin βx + βe2αx cos2 βx − [αe2αx cos βx sin βx − βe2αx sin2 βx]
= βe2αx 6= 0, for all x ∈ R

Thus y1 (x) and y2 (x) are linearly independent on R.

Example 3.39:
Solve the initial value problem y 00 − 2y 0 + 5y = 0, y(π/2) = 0, y 0 (π/2) = 2.
Solution:
Write differential equation in operator form:

(D2 − 2D + 5I)[y] = 0.

The auxiliary equation is then

λ2 − 2λ + 5 = 0.

Using the quadratic formula, the roots are λ = 1 + 2i, λ = 1 − 2i. Applying
Proposition 3.38 with α = 1, β = 2, shows that two linearly independent solutions
are y1 = ex cos 2x, y2 = ex sin 2x and the hence general solution of the differential
equation is
y(x) = c1 ex cos 2x + c2 ex sin 2x, −∞ < x < ∞.

48
Applying the first initial condition, y(π/2) = 0, we have
π π
0 = c1 e 2 cos π + c2 e 2 sin π
π
0 = −c1 e 2
⇒ c1 = 0.
Therefore y(x) = c2 ex sin 2x and y 0 = c2 ex sin 2x+2c2 ex cos 2x. Applying the second
initial condition, y 0 (π/2) = 2, we obtain
π π
2 = c2 e 2 sin π + 2c2 e 2 cos π
π
2 = −2c2 e 2
π
⇒ c2 = −e− 2 .
Thus the solution of the initial value problem is
π
y(x) = −e− 2 ex sin 2x
π
= −ex− 2 sin 2x, −∞ < x < ∞.
Example 3.40:
Find the general solutions of y 0000 − 3y 00 − 4y = 0.
Solution:
The operator form of the differential equation is
(D4 − 3D2 − 4I)[y] = 0,
hence the auxiliary equation is
λ4 − 3λ2 − 4 = 0.
Let µ = λ2 , then µ satisfies the quadratic equation
µ2 − 3µ − 4 = 0,
which has solutions µ = 4 and µ = −1. Thus we have
λ2 = 4 ⇒ λ = ±2
or
λ2 = −1 ⇒ λ = ±i
The corresponding solutions of the differential equation are therefore:
y1 (x) = e2x , y2 (x) = e−2x , y3 (x) = cos(x), y4 (x) = sin(x).
One can show (exercise) that these solutions are linearly independent. The general
solution is thus
y(x) = c1 e2x + c2 e−2x + c3 cos(x) + c4 sin(x), −∞ < x < ∞.

49
3.5 Finding a Particular Solution of the Inhomo-
geneous Equation

Recall we found that the most general solution to the differential equation φ(D)[y] =
F (x) can be written as
y(x) = yh (x) + yp (x)
where

• yh (x) is the general solution to the homogeneous equation φ(D)[y] = 0

• yp (x) is particular solution to the equation φ(D)[y] = F (x)

In the previous section we learned how to obtain solutions yh (x). In this section
we will see several approaches for finding a particular solution yp (x).

3.5.1 Method of Undetermined Coefficients

To motivate this method, we will begin with two examples.

Example 3.41:
Find the general solution of

(D2 − D − 2I)[y] = cos(x) (3.18)

Solution:
We saw in Example 3.30 that the general solution of the homogeneous equation
(D2 − D − 2I)[y] = 0 is

yh (x) = c1 e2x + c2 e−x , −∞ < x < ∞.

We need to find a particular solution of equation (3.18) i.e., to find a function that
is mapped to cos(x) by the operator (D2 − D − 2I). Recall that

D[sin(x)] = cos(x), D2 [cos(x)] = − cos(x), I[cos(x)] = cos(x)

So, we look for a solution in the form

yp (x) = M1 cos(x) + M2 sin(x)

50
where M1 , M2 are to be determined. Taking the derivative of yp (x)

D[yp ] = −M1 sin(x) + M2 cos(x)

D2 [yp ] = −M1 cos(x) − M2 sin(x)

and substituting into (3.18), we obtain

−M1 cos(x)−M2 sin(x)−(−M1 sin(x)+M2 cos(x))−2(M1 cos(x)+M2 sin(x)) = cos(x),

or
(−3M1 − M2 ) cos(x) + (−3M2 + M1 ) sin(x) = cos(x).
For this equality to hold for all values of x, the constants M1 and M2 must satisfy

−3M1 − M2 = 1
M1 − 3M2 = = 0.

Solving this system we obtain

3 1
M1 = − , M2 − .
10 10
So a particular solution of (3.18) is
3 1
yp (x) = − cos(x) − sin(x).
10 10
The general solution of equation (3.18) is therefore
3 1
y(x) = yh (x) + yp (x) = c1 e2x + c2 e−x − cos(x) − sin(x), −∞ < x < ∞.
10 10
Example 3.42:
Find the general solution of

(D2 − D − 2I)[y] = e2x (3.19)

Solution:
We will try the same approach as the last example. Noting that

D[e2x ] = 2e2x , D2 [e2x ] = 4e2x ,

we take yp = Le2x , and substitute into equation (3.19) to obtain

4Le2x − 2Le2x − 2Le2x = 0

51
or
0 = e2x .
This can never be true and so the method fails.
Why does it fail? Recall yh (x) = c1 e2x + c2 e−x so e2x is a solution of the
homogeneous equation, i.e., e2x is mapped to 0 by D2 − D − 2I so it can’t be
mapped to e2x . What other function generates e2x when you take its derivative?
Note that
D[xe2x ] = e2x + 2xe2x .
So we take yp = Lxe2x . Then D[yp ] = L(e2x + 2xe2x ), D2 [yp ] = L(4 + 4x)e2x and
substituting into equation (3.19) we obtain

L(4 + 4x)e2x − L(1 + 2x)e2x − 2Lxe2x = e2x .

Simplifying gives
L(3e2x ) = e2x
which implies
1
L= .
3
1
The particular solution is therefore yp = 3 xe2x and the general solution is

1
y(x) = yh (x) + yp (x) = c1 e2x + c2 e−x + xe2x , −∞ < x < ∞.
3

The approach used in Examples 3.41 and 3.42 is called the Method of Unde-
termined Coefficients. The idea is to use the specific form of F (x) to determine
a form for yp (x). A little thought reveals that this method will only work on the
constant coefficient equation (3.13) when F (x) consists of functions with have a
finite number of derivatives. The general procedure is outlined below.
Method of Undetermined Coefficients:
This method applies to equation (3.13) if F (x) consists of products or sums of sines,
cosines, exponentials and non-negative, integer powers of x.

1. Given F (x) choose the form for yp (x) according to the Table 3.1 where
Kj , Mj , L are unknown constants which will be determined.

2. If F (x) is a product of terms in column 1, choose yp (x) to be the product of

the corresponding terms in column 2.

52
 Term in F (x) Form of yp (x)
eγx Leγx
xn Kn xn + Kn−1 xn−1 + · · · + K1 x + K0
cos ωx M1 cos ωx + M2 sin ωx
sin ωx

Table 3.1: Form of yp (x) for the Method of Undetermined Coefficients

3. If any term in your choice for yp (x) is a solution of the associated homogeneous
equation (3.15), then multiply yp (x) by the lowest power of x so that it is no
longer a solution.

4. If F (x) is a sum of functions

F (x) = F1 (x) + F2 (x) + · · · + Fm (x)

choose an appropriate ypj for each Fj (x), following steps 1-3, and then add
them together:
yp (x) = yp1 (x) + yp2 (x) + · · · + ypm (x)

5. Substitute the form you have found for yp (x) into the equation (3.13) and
determine the values for the unknown coefficients so that yp (x) satisfies the
equation.

Example 3.43:
Find the solution of the initial value problem

y 00 + y = x2 + cos(x), y(0) = 1, y 0 (0) = 1.

Solution:
The associated homogeneous equation is

y 00 + y = 0,

hence the auxiliary equation is

λ2 + 1 = 0,
which has roots λ = ±i. So the general solution of the associated homogeneous
equation is
yh (x) = c1 cos(x) + c2 sin(x), −∞ < x < ∞.

53
 To find yp (x) we identify the terms of the the RHS of the differential equation

F (x) = x2 + cos(x)
= F1 (x) + F2 (x).

Then the particular solution will be of the form

yp (x) = yp1 (x) + yp2 (x).

The terms in yp (x) are chosen according to Table 3.1

yp1 (x) = K2 x2 + K1 x + K0
yp2 (x) = M1 cos(x) + M2 sin(x).

However, the choice for yp2 is a solution of the associated homogeneous equation so
we must multiply it by x

yp2 (x) = M1 x cos(x) + M2 x sin(x).

Therefore we have

yp = K2 x2 + K1 x + K0 + M1 x cos(x) + M2 x sin(x).

Taking derivatives

yp0 = 2K2 x + K1 + M1 cos(x) − M1 x sin(x) + M2 sin(x) + M2 x cos(x)

yp00 = 2K2 − 2M1 sin(x) − M1 x cos(x) + 2M2 cos(x) − M2 x sin(x)

and substituting into the differential equation we obtain

2K2 − 2M1 sin(x) − M1 x cos(x) + 2M2 cos(x) − M2 x sin(x) +

K2 x2 + K1 x + K0 + M1 x cos(x) + M2 x sin(x) = x2 + cos(x)

Equating the coefficients of like terms we obtain

x2 : K 2 = 1
x : K1 = 0
1 : K0 + 2K2 = 0 ⇒ K0 = −2
x cos(x) : 0 = 0
x sin(x) : 0 = 0
1
cos(x) : 2M2 = 1 ⇒ M2 =
2
sin(x) : −2M1 = 0 ⇒ M1 = 0.

54
Putting these values into the form for the particular solution yields
1
yp (x) = x2 − 2 + x sin(x), −∞ < x < ∞.
2
Thus the general solution is
1
y(x) = c1 cos(x) + c2 sin(x) + x2 − 2 + x sin(x), −∞ < x < ∞.
2
Applying the first initial condition, y(0) = 1:

c1 − 2 = 1 ⇒ c1 = 3.

Putting this into y(x) and differentiating gives

1 1
y 0 (x) = −3 sin(x) + c2 cos(x) + 2x + sin(x) + x cos(x).
2 2
Applying the second initial condition, y 0 (0) = 1, implies

c2 = 1.

The solution of the initial value problem is therefore

1
y(x) = 3 cos(x) + sin(x) + x2 − 2 + x sin(x), −∞ < x < ∞.
2
Example 3.44:
Find the form of the particular solution to y (4) − 3y 00 − 4y = e2x sin(x) + 2x cos(2x).
Solution:
We showed in Example 3.40 that the solution of the associated homogeneous equa-
tion is

yh (x) = c1 e2x + c2 e−2x + c3 cos(x) + c4 sin(x), −∞ < x < ∞.

Identifying the parts of the RHS

F (x) = e2x sin(x) + 2x cos(2x)

= F1 (x) + F2 (x),

we take yp (x) = yp1 (x) + yp2 (x). Following Table 3.1, the form for yp1 is

yp1 (x) : (Le2x )(M1 cos(x) + M2 sin(x)).

Multiplying this out and noting that there are just two distinct terms, we take

yp1 (x) = N1 e2x cos(x) + N2 e2x sin(x).

55
Similarly the form for yp2 from Table 3.1 is

(K1 x + K0 )(R1 cos(2x) + R2 sin(2x)),

so we take

yp2 (x) = A1 x cos(2x) + A0 cos(2x) + B1 x sin(2x) + B0 sin(2x).

Thus the form for the particular solution is

yp (x) = N1 e2x cos(x)+N2 e2x sin(x)+A1 x cos(2x)+A0 cos(2x)+B1 x sin(2x)+B0 sin(2x).

3.5.2 Method of Reduction of Order

Proposition 3.32 gave a method for using one solution of the homogeneous equa-
tion (3.6) to generate a second linearly independent solution. In fact this approach
can also be applied to find a particular solution of the inhomogeneous equation
(3.2).

Theorem 3.45 (Method of Reduction of Order). If y1 (x) is one nonzero solution

of the homogeneous equation (3.6) with n = 2, then the general solution of the
corresponding inhomogeneous equation (3.2) may be obtained by the substitution
y(x) = u(x)y1 (x).

Proof. Using Proposition 3.32 we know that we can generate a second linearly inde-
pendent solution of the homogeneous equation (3.2) of the form y2 (x) = u1 (x)y1 (x).
Now consider the inhomogeneous equation. Proceeding as in the proof of Propo-
sition 3.32, one can show that y(x) = u(x)y1 (x) is a solution of equation (3.2) if
u(x) is a solution of

a2 (x)y1 (x)u00 + [2a2 (x)y10 (x) + a1 (x)y1 (x)]u0 = F (x).

But this equation can always be solved, just as in the proof of Proposition 3.32. Let
up (x) be the solution. Then yp (x) = up (x)y1 (x) is a solution of the inhomogeneous
equation (3.2). Thus the general solution of equation (3.2) is y(x) = c1 y1 (x) +
c2 y2 (x) + yp (x).

Note that this method works for any second order linear differential equations
(variable coefficient or constant coefficient) with any function, F (x), on the right
hand side.

56
Example 3.46:
Find the general solution of the differential equation y 00 + y = tan(x).
Solution:
The general solution of the associated homogeneous equation is

yh (x) = c1 cos(x) + c2 sin(x), −∞ < x < ∞.

The Method of Undetermined Coefficients won’t work here since the right hand
side is tan(x). So we try the Method of Reduction of Order, with y1 (x) = cos(x).
We then have

y = uy1 = u(x) cos(x),

y 0 = u0 cos(x) − u sin(x),
y 00 = u00 cos(x) − 2u0 sin(x) − u cos(x).

Substituting into the differential equation

u00 cos(x) − 2 sin(x)u0 − cos(x)u + cos(x)u = tan(x)

and simplifying gives

cos(x)u00 − 2 sin(x)u0 = tan(x).
Let v = u0 , v 0 = u00 , then we obtain a first order linear differential equation for v(x):

cos(x)v 0 − 2 sin(x)v = tan(x).

Converting to standard form

sin(x) sin(x)
v0 − 2 v= (cos(x) 6= 0)
cos(x) cos2 (x)
we see that the integrating factor is
R sin(x)
e−2 cos(x)
dx
= e2 ln(cos(x)) = cos2 (x).

Multiplying through the differential equation and solving in the usual way deter-
mines v(x):

cos2 (x)v 0 − 2 sin(x) cos(x)v = sin(x)

(cos2 (x)v)0 = sin(x)
(Integrate) cos2 (x)v = − cos(x) + C1
1 C1
v = − +
cos(x) cos2 (x)
= − sec(x) + C1 sec2 (x)

57
Thus
u0 = sec(x) + C1 sec2 (x)
which may be integrated to find u(x):

u = − ln | sec(x) + tan(x)| + C1 tan(x) + C2 .

Our general solution is therefore

y(x) = c1 cos(x) + c2 sin(x) + u(x) cos(x)

y(x) = c1 cos(x) + c2 sin(x) − cos(x) ln | sec(x) + tan(x)| + cos(x)C1 tan(x) + cos(x)C2
y(x) = c∗1 cos(x) + c∗2 sin(x) − cos(x) ln | sec(x) + tan(x)|

where we have defined c∗1 = c1 + C2 , c∗2 = c2 + C1 . Note that we could omit the
arbitrary constants in the integration of v(x) and u(x) as these just replicate the
general solution of the associated homogeneous equation. The interval of existence
of this solution is any open interval on which cos(x) 6= 0, e.g., (− π2 , π2 ).

3.5.3 Method of Variation of Parameters

This method was first developed by Joseph Lagrange in 1774. The idea is similar
to the Method of Reduction of Order, but instead of just using one solution of the
associated homogeneous equation, we will use all of them. We will develop the
method for the case n = 2. The approach for higher order equations is similar.
Consider a second order inhomogeneous equation

a2 (x)y 00 + a1 (x)y 0 + a0 (x)y = F (x). (3.20)

Suppose that the associated homogeneous equation

a2 (x)y 00 + a1 (x)y 0 + a0 (x)y = 0 (3.21)

has general solution yh (x) = c1 y1 (x)+c2 y2 (x). We will look for a particular solution
of (3.20) by “varying the parameters” in this solution, that is, by replacing the
constants c1 and c2 with arbitrary (unknown) functions of x:

yp (x) = u1 (x)y1 (x) + u2 (x)y2 (x). (3.22)

Substituting this form into equation (3.20) will give one condition on the unknown
functions u1 (x) and u2 (x). We will arbitrarily impose another condition, which will
give two equations to solve for u1 (x) and u2 (x).

58
 To begin, take the derivative of yp (x)

yp0 (x) = u01 (x)y1 (x) + u1 (x)y10 (x) + u02 (x)y2 (x) + u2 (x)y20 (x).

Now we impose a condition to simplify this expression:

u01 (x)y1 (x) + u02 (x)y2 (x) = 0. (3.23)

Then we have

yp0 (x) = u1 (x)y10 (x) + u2 (x)y20 (x),

yp00 (x) = u01 (x)y10 (x) + u1 (x)y100 (x) + u02 (x)y20 (x) + u2 (x)y200 (x).

Substituting this into equation (3.20) gives

a2 (x)(u01 y10 + u1 y100 + u02 y20 + u2 y100 ) + a1 (x)(u1 y10 + u2 y20 ) + a0 (x)(u1 y1 + u2 y2 (= F (x).

Collecting terms in u1 , u01 , u2 , u02 gives

(a2 (x)y100 +a1 (x)y10 +a0 (x)y1 )u1 +(a2 (x)y100 +a1 (x)y10 +a0 (x)y1 )u2 +a2 (x)(u01 y10 +u02 y20 ) = F (x).

But the first two terms are zero since y1 and y2 are solutions of the associated
homogeneous equation (3.21). Thus we obtain the following condition on u1 and
u2
a2 (x)u01 y10 + a2 (x)u02 y20 = F (x). (3.24)
Note that equations (3.23) and (3.24) are a set of linear algebraic equations for u01
and u02 with coefficient matrix
" #
y1 (x) y2 (x)
M=
a2 (x)y10 (x) a2 (x)y20 (x)

Now det(M ) = a2 (x)W (y1 , y2 ). Assuming a2 (x) 6= 0, then det(M ) 6= 0 since y1 , y2

are linearly independent solutions of equation (3.21). Thus we will always be able
to solve (3.23) and (3.24) to find u01 and u02 as functions of x, which can then be
integrated to find u1 (x) and u2 (x) and hence yp (x).
Example 3.47:
Find the general solution of the differential equation y 00 + y = tan(x).
Solution:
Recall that the general solution of the associated homogeneous equation is

yh (x) = c1 cos(x) + c2 sin(x), −∞ < x < ∞.

59
We will use the Method of Variation of Parameters to find a particular solution of
the inhomogeneous equation. To begin, we assume a particular solution of the form
yp (x) = u1 (x) cos(x) + u2 (x) sin(x).
The first derivative is
yp0 (x) = u01 cos(x) − u1 sin(x) + u02 sin(x) + u2 cos(x).
Imposing the condition
u01 cos(x) + u02 sin(x) = 0 (3.25)
this becomes
yp0 (x) = −u1 sin(x) + u2 cos(x).
So the second derivative is
yp00 (x) = −u01 sin(x) − u1 cos(x) + u02 cos(x) − u2 sin(x).
Substituting into the differential equation, we obtain
−u01 sin(x) − u1 cos(x) + u02 cos(x) − u2 sin(x) + u1 cos(x) + u2 sin(x) = tan(x)
which simplifies to
−u01 sin(x) + u02 cos(x) = tan(x) (3.26)
Solving equations (3.25)–(3.26) yields
sin2 (x)
u01 = −
cos(x)
u02 = sin(x).
Integrating these equations gives1
u1 = − ln | sec(x) + tan(x)| + sin(x)
u2 = − cos(x).
Thus the particular solution is
yp (x) = − cos(x) ln | sec(x) + tan(x)| + cos(x) sin(x) − sin(x) cos(x)
= − cos(x) ln | sec(x) + tan(x)|,
and the general solution of the inhomogeneous equation is
y(x) = c1 (x) cos(x) + c2 (x) − cos(x) ln | sec(x) + tan(x)|,
with interval of existence (− π2 , π2 ). Note that this is the same solution as was
obtained using the Method of Reduction of Order in Example 3.46.
1
We omit the constants of integration as they will just regenerate the solution of the homoge-
neous equation.

60
Chapter 4

Systems of Ordinary Differential

Equations

Definition 4.1: An n dimensional system of ordinary differential equations

is a set of n equations for n unknown functions of one variable which involve one
or more derivatives of the unknown functions.

The order of the system is determined by the order of the highest derivative
that appears in the set of equations.
The system is linear the equations depend linearly on the unknown functions
and their derivatives, otherwise it is called nonlinear.
Here are some examples.

1. First order, nonlinear two dimensional system

dy1
= y12 + x2
dx
dy2
= y1 y2
dx

2. Second order, linear two dimensional system

d2 r dr
= + φ + sin(t)
dt dt
dφ
= 5r − 3φ
dt

In some cases, systems of differential equations can be reduced to scalar, higher

order differential equations. The following example illustrates this.

61
Example 4.2:
Consider a commodity where the supply and demand functions are related to price
through their derivatives:
dS
= F1 (t, P ) (4.1)
dt
dD
= F2 (t, P ) (4.2)
dt
adding in the equation for the price as described in subsection 2.1.3
dP
= α(D − S), where α > 0 is a constant (4.3)
dt
gives a 3 dimensional system of first order, ordinary differential equations for the
unknown functions S(t), D(t), P(t).
This system can be reduced to one equation for the price as follows. From (4.3)

d2 P dD dS
= α − α = α[F2 (t, P ) − F1 (t, P )] (4.4)
dt2 dt dt
which is a second order differential equation for P (t). If (4.4) can be solved for P (t)
(e.g., if F1 , F2 are linear in P ), then we can put the solution for P (t) into (4.1) and
(4.2) and solve for S(t) and D(t) by integration.

The approach of Example 4.2 for solving systems is called the method of
elimination. It does not always apply and is quite ad hoc, so we will develop a
more systematic approach to solving systems. First we state an important result.

Theorem 4.3. Any higher-order, ordinary differential equation or system of higher

order, ordinary differential equations can be expressed as an equivalent system of
first-order, ordinary differential equations.

Proof. Suppose we have an nth order ordinary differential equation

y (n) = g(x, y, y 0 , . . . , y (n−1) )

where
dk y
y (k) = .
dxk
Introducing the following variables

y1 = y, y2 = y 0 , y3 = y 00 , · · · , yn = y (n−1) ,

62
we have
dy1
= y2 ,
dx
dy2
= y3 ,
dx
..
.
dyn
= g(x, y1 , y2 , · · · , yn ).
dx
This is an n dimensional system of first order, ordinary differential equations for the
unknown functions y1 (x), y2 (x), . . . , yn (x). Any solution of this system will generate
a corresponding solution of the original nth order ordinary differential equation.
The proof for systems of higher-order ordinary differential equations is similar.

Due to Theorem 4.3 we will only consider first order systems. The most general
dimensional system of first order, ordinary differential equations in standard form
is
dy1
= f1 (x, y1 , · · · , yn )
dx
dy2
= f2 (x, y1 , · · · , yn )
dx
..
. (4.5)
dyn
= fn (x, y1 , · · · , yn )
dx
We will focus on the case where the fj are linear functions of the yk . In this case
(4.5) may be written

dy1
= a11 (x)y1 + · · · + a1n (x)yn + f1 (x)
dx
..
. (4.6)
dyn
= an1 (x)y1 + · · · + ann (x)yn + fn (x)
dx
To simplify the development of the theory, we introduce the following notation.

63
 Let
     
y1 a11 (x) · · · a1n (x) F1 (x)
 .  .. ..  . 
y =  ..  , A(x) =  , F =  ..  .
 
. .
yn an1 (x) · · · ann (x) Fn (x)

Then (4.6) can be written

dy
= A(x)y + F(x). (4.7)
dx
Note that F : R → Rn is a vector valued function while A : R → Rn×n is a matrix
valued function.

4.1 Theory of Linear Systems

Definition 4.4: Equation (4.7) is called homogeneous if F(x) ≡ 0, otherwise it

is called inhomogeneous.
Definition 4.5: A solution of (4.7) is any differentiable function, y, that satisfies
(4.7) for all x in on some open interval I.

Note: y is differentiable on I if each of its component functions is differentiable

on I.
Definition 4.6: The interval I in Definition 4.5 is called the interval of existence
or interval of definition of the solution.
Definition 4.7: The general solution of (4.7) is a solution containing n arbitrary
constants that represents all nonsingular solutions of (4.7).
Definition 4.8: An initial condition for (4.7) is an equation which specifies y
at a particular value of x
y(x0 ) = y0 , (4.8)
or, in component form

y1 (x0 ) = y01 , y2 (x0 ) = y02 , · · · , yn (x0 ) = y0n . (4.9)

Theorem 4.9 (Existence and Uniqueness Theorem). Suppose A(x) and F(x) are
continuous on the open interval I and x0 ∈ I. Then, given any vector
y0 = (y01 , · · · , y0n )T ∈ Rn the initial value problem consisting of (4.7) and (4.8)
has a unique solution on I.

Proof. See Appendix 2 of Edwards and Penney [2].

64
 It follows from Theorem 4.3 that any nth order linear ordinary differential equa-
tion can be written as an n dimensional first order linear system (4.7). Thus the
theory for solving (4.7) will be similar to that for nth order ordinary differential
equations.
Definition 4.10: The associated homogeneous equation for equation (4.7) is
dy
= A(x)y (4.10)
dx
Theorem 4.11. If u(x) is a solution of equation (4.10) on I1 and v(x) is a solution
of equation (4.7) on I2 then y(x) = u(x) + v(x) is a solution of equation (4.7) on
I = I1 ∩ I2 .

Proof. Let x ∈ I and take the derivative of y(x) = u(x) + v(x):

dy du dv
= +
dx dx dx
= A(x)u + A(x)v + F(x)
= A(x)(u + v) + F(x)
= A(x)y + F(x)

Thus y(x) satisfies (4.7) on I.

As for nth order linear ODEs, a candidate for the general solution of the inhomo-
geneous equation (4.7) is the sum of the general solution of (4.10) and a particular
solution of (4.7). Before proving this, we first derive some results about the general
solution of (4.10).

4.1.1 The General Solution of the Homogeneous Equation

Theorem 4.12 (Principle of Superposition). If y1 , y2 , . . . , yk are solutions of the

homogeneous equation (4.10) on I then

y = c1 y 1 + c2 y 2 + · · · + ck y k

is also a solution of (4.10) on I for any constants c1 , . . . , ck .

Proof. The proof follows from direct evaluation of the derivative:

dy dy1 dy2 dyk
= c1 + c2 + · · · + ck
dx dx dx dx
= c1 A(x)y1 + c2 A(x)y2 + · · · + ck A(x)yk
= A(x)[c1 y1 + c2 y2 + · · · + ck yk ]
= A(x)y.

65
 To define the general solution of (4.10) we need the following
Definition 4.13: Let f1 (x), f2 (x), . . . , fk (x) be real n-dimensional vector-valued
functions defined on the interval I, i.e., fi : I → Rn . The functions f1 , f2 , . . . , fk
are linearly dependent on I if there exist k constants α1 , . . . , αk , which are not
all zero, such that

α1 f1 (x) + α2 f2 (x) + · · · + αk fk (x) = 0

for all x ∈ I. Otherwise the functions are linearly independent.

Example 4.14:
Show that the following functions are linearly independent on R.
! !
sin(x) cos(x)
f1 = , f2 = .
cos(x) cos(x)

Solution:
Consider
α1 f1 + α2 f2 = 0
In component form this becomes

α1 sin(x) + α2 cos(x) = 0 (4.11)

α1 cos(x) + α2 cos(x) = 0 (4.12)

Clearly (4.12) is satisfied for any α1 , α2 with α1 = −α2 . Equation (4.11) however
requires α2 = 0 when x = 0 and α1 = 0 when x = π/2. Therefore, the only choice
that satisfies (4.11)–(4.12) for all x ∈ R is α1 = α2 = 0 and the functions f1 , f2 are
linearly independent.
Definition 4.15: The Wronskian of the functions f1 (x), . . . , fn (x) is defined as
 
f11 (x) f12 (x) · · · fn1 (x)
 f12 (x) f22 (x) · · · fn2 (x) 
 
W (f1 , f2 , · · · , fn ) = det  .. .. .. 
. . .
 
 
f1n (x) f2n (x) · · · fnn (x)
Example 4.16:
From the previous example
" #
sin(x) cos(x)
W (f1 , f2 ) = det = sin(x) cos(x)−cos2 (x) = cos(x)(sin(x)−cos(x))
cos(x) cos(x)

66
Proposition 4.17. Let A(x) be a continuous function from I to IRn×n and let
y1 , y2 , . . . , yn be solutions on I of (4.10). Then y1 , y2 , . . . , yn are linearly indepen-
dent on I if and only if W (y1 , y2 , . . . , yn ) 6= 0 for all x ∈ I

Proof. Suppose that W (y1 , y2 , . . . , yn ) 6= 0 for all x ∈ I. Consider the linear system
  
y11 (x) · · · yn1 (x) α1
.. ..   .. 
  .  = 0. (4.13)

 . .
y1n (x) · · · ynn (x) αn

The determinant of the coefficient matrix of the system is W (y1 , y2 , . . . , yn ), which

is nonzero for all x ∈ I. Hence, for each x ∈ I the linear system (4.13) has only
one solution: α1 = α2 = · · · = αn = 0. Note that (4.13) can be rewritten as

α1 y1 (x) + · · · + αn yn (x) = 0. (4.14)

Thus the only constants α1 , α2 , . . . , αn satisfying (4.14) for all x ∈ I are α1 = α2 =

· · · = αn = 0, i.e., y1 , y2 , . . . , yn are linearly independent on I.
Now suppose y1 , y2 , . . . , yn are linearly independent solutions on I of (4.10).
We need to show W (y1 , y2 , . . . , yn ) 6= 0 on I. Suppose the opposite, i.e., there is
at least one point x0 ∈ I where the Wronskian is zero. At this point the linear
system (4.13) has a nonzero solution, i.e.,

α1 = ᾱ1 , α2 = ᾱ2 , · · · , αn = ᾱn (4.15)

where not all of the ᾱj are zero. Let

u(x) = ᾱ1 y1 (x) + ᾱ2 y2 (x) + · · · + ᾱn yn (x).

Then u(x) satisfies the initial value problem consisting of (4.10) and the initial
condition
y(x0 ) = 0
on the interval I. (The initial condition comes from putting (4.15) into (4.14) and
evaluating at x0 .) However, the function y(x) = 0, x ∈ I also satisfies this initial
value problem. By Theorem 4.9 the initial value problem has a unique solution,
hence we must have u(x) = 0, x ∈ I, i.e.,

ᾱ1 y1 (x) + ᾱ2 y2 (x) + · · · + ᾱn yn (x) = 0, for all x ∈ I.

Since not all of the ᾱj are zero, this implies that y1 (x), y2 (x), . . . , yn (x) are linearly
dependent. This is a contradiction, so we must have W (y1 , y2 , . . . , yn ) 6= 0 on
I.

67
Theorem 4.18. Let A(x) be a continuous function from I to IRn×n . If y1 , y2 , . . . , yn
are n linearly independent solutions on I of the DE (4.10) then the general solution
of (4.10) is

y(x) = c1 y1 (x) + c2 y2 (x) + · · · + cn yn (x), x∈I (4.16)

where c1 , . . . , cn are arbitrary constants.

Proof. Let w(x) be any solution of (4.10). By Proposition 4.17, there is a point
x0 ∈ I such that W (y1 , . . . , yn ) 6= 0 at x = x0 . Define y0 = w(x0 ). Clearly w(x)
is a solution of the IVP consisting of the DE (4.10) and the IC y(x0 ) = y0 .
We now show that there is a solution to this IVP in the form (4.16). It follows
from the Theorem 4.12 that (4.16) is a solution of the DE (4.10). It remains to
show there are constants c1 , ..., cn , such that

c1 y1 (x0 ) + c2 y2 (x0 ) + · · · + cn yn (x0 ) = y0 .

Rewriting this in matrix vector form

    
y11 (x0 ) · · · yn1 (x0 ) c1 y01
.. ..   ..   .. 
 .  =  (4.17)

 . . . 
y1n (x0 ) · · · ynn (x0 ) cn y0n
or
M c = y0 .
But det M 6= 0 since W 6= 0 at x = x0 . It follows that (4.17) may be solved, i.e.,
def
c = M −1 y0 = k.

Thus the IVP also has the solution k1 y1 (x) + · · · + kn yn (x). By the Existence and
Uniqueness Theorem, the IVP has a unique solution, thus we must have w(x) =
k1 y1 (x) + · · · + kn yn (x). That is, any solution of (4.10) can be written in the form
(4.16).

Note that this proof shows that every solution of (4.10) can be written in the
form (3.10). This implies that, under the conditions of the theorem, there are no
singular solutions of (4.10). Finally, we state a result that gives a useful formula
for the Wronskian.

68
Proposition 4.19. Suppose that the conditions of Theorem 4.9 are satisfied and
let y1 (x), y2 (x), . . . , yn (x) be n solutions on the interval I of the nth order linear
ODE (4.10). Then, for any x0 ∈ I, the Wronskian of these functions satisfies
R x Pn
− aii (x) dx
W (y1 , y2 , . . . , yn ) = W (x0 )e x0 i=1

for all x ∈ I.

Proof. The proof is similar to that of Proposition 3.28. See Section 7.3 of Boyce
and DiPrima [1].

4.1.2 The General Solution of the Inhomogeneous Equation

We can now return to the problem of finding the general solution of the inhomoge-
neous equation (4.7).

Theorem 4.20. Let yh (x) be the general solution of equation (4.10) and yp (x) be
a particular solution of equation (4.7). Then yh (x) + yp (x) is the general solution
of (4.7).

Proof. It follows from Theorem 4.11 that yh (x) + yp (x) is a solution of (4.7).
Further, since yh (x) is the general solution of the associated homogeneous equation
(4.10), it contains n arbitrary constants, i.e., yh (x) = v(x; c1 , . . . , cn ).
To show that it is the general solution suppose that ŷ(x) is any solution of (4.7)
and consider
w(x) = ŷ(x) − yp (x).
Differentiating we find
dw dŷ dyp
= − = A(x)ŷ + F(x) − (A(x)yp + F (x))
dx dx dx
= A(x)(ŷ − yp )
= A(x)w,

so w is a solution of (4.10). Then, from Theorem 4.18, w(x) = yh (x) for some
particular values of the arbitrary constants, i.e., w(x) = v(x; K1 , . . . , Kn ). Thus
we have

ŷ(x) = w(x) + yp (x)

= v(x; K1 , . . . , Kn ) + yp (x).

That is, any solution of equation (4.7) can be written in the form yh (x)+yp (x).

69
 Note that this proof shows that every solution of (4.7) can be written in the
form yh (x) + yp (x). This implies that, under the conditions of the theorem, there
are no singular solutions of (4.7).
Summary - Solving the inhomogeneous equation (4.7)
It follows from Theorems 4.18 and 4.20 that the general solution of equation (4.7)
may be found using the following steps.

1. Find n linearly independent solutions of (4.10), y1 , y2 , . . . , yn , i.e., n solutions

whose Wronskian is nonzero. Note that from Proposition 4.19 we need only
check that the Wronskian is nonzero at one point of the interval of existence
of the solutions.

Form the general solution of (4.10):

yh (x) = c1 y1 + c2 y2 + · · · + cn yn

2. Find a particular solution, yp (x) of (4.7)

3. Form the general solution of (4.7):

y(x) = yh (x) + yp (x).

We will consider methods for step 1 in the following section. Approaches for step
2 will be considered in section 4.3.

4.2 Homogeneous Systems with Constant Coef-

ficients
As for nth order ordinary differential equations, to proceed further, we need to
restrict ourselves to the constant coefficient case. This can be written in component
form as
dy1
= a11 y1 + · · · + a1n yn + F1 (x),
dx
..
. (4.18)
dyn
= an1 y1 + · · · + ann yn + Fn (x),
dx
where aij ∈ R, and in vector form as
dy
= Ay + F(x) (4.19)
dx
70
where A is an n × n constant matrix. The associated homogeneous equation is
dy
= Ay. (4.20)
dx
Recall the relationship between n-dimensional systems and nth order ordinary dif-
ferential equations given in Theorem 4.3. This motivates trying a similar procedure
to solve systems with constant coefficients as was used for nth order ordinary equa-
tions with constant coefficients. Look for a solution of (4.20) of the form

y = eλx v

where v, an n-dimensional vector, and λ, a constant, are to be determined. Differ-

entiating we obtain
dy
= λeλx v.
dx
Substituting into (4.20) then gives

λeλx v = Aeλx v = eλx Av.

Since eλx 6= 0 this will only be satisfied if

λv = Av. (4.21)

We conclude that eλx v is a solution of (4.20) if λ is an eigenvalue of A with eigen-

vector v.
Now rewrite (4.21) as
(A − λI)v = 0. (4.22)
Recall from Linear Algebra: equation (4.22) has a nontrivial solution (v 6= 0) if
and only if
det(A − λI) = 0. (4.23)
Equation (4.23) is called the characteristic equation for the differential equation
(4.20). Note that det(λI − A) will be a degree n polynomial in λ since A is an n × n
matrix.

Results from Linear Algebra

1. If A is an n × n matrix it will have n eigenvalues λ1 , λ2 , . . . , λn corresponding

to the n roots of the characteristic equation (4.23).

2. If λ1 , λ2 , . . . , λk are distinct eigenvalues of A (i.e., λ1 6= λ2 6= λ3 6= · · · =

6 λk )
then the corresponding eigenvectors {v1 , v2 , . . . , vk } are linearly independent.

71
 3. If λ = α+iβ is an eigenvalue of A with eigenvector v = u+iw, then λ = α−iβ
is an eigenvalue of A with eigenvector v = u − iw, where u, w ∈ Rn and u
and w are linearly independent vectors.

4. If λ is a root of multiplicity m of (4.23) it will have at least one and not more
than m associated eigenvectors.

This leads to the following

Theorem 4.21. If {v1 , . . . , vk } are linearly independent eigenvectors associated

with the eigenvalues λ1 , λ2 , . . . , λk of A then the functions {eλ1 x v1 , eλ2 x v2 , . . . , eλk x vk }
are linearly independent solutions of equation (4.20) on (−∞, ∞)

Proof. It is clear from the previous discussion that eλi x vi , i = 0, . . . , k are solutions
of equation (4.20). To show they are linearly independent, consider

α1 eλ1 x v1 + α2 eλ2 x v2 + · · · + αk eλk x vk = 0. (4.24)

For x = 0 this becomes

α1 v1 + α2 v2 + · · · + αk vk = 0.

Then the only way this can be satisfied is if αi = 0, i = 1, . . . , k since {v1 , v2 , . . . , vk }

are linearly independent vectors. Therefore the only way that (4.24) can be satisfied
for all x ∈ (−∞, ∞) is if αi = 0, i = 1, . . . , k. Thus {eλ1 x v1 , eλ2 x v2 , . . . , eλk x vk }
are linearly independent on (−∞, ∞).

Theorem 4.22. If λ = α + iβ and λ = α − iβ, α, β ∈ R, are roots of the charac-

teristic equation (4.23) with eigenvectors v = u + iw, v = u − iw then two linearly
independent solutions of the differential equation (4.20) are

y1 (x) = (u cos(βx) − w sin(βx))eαx

y2 (x) = (u sin(βx) + w cos(βx))eαx

Proof. Since λ, λ are eigenvalues of A with eigenvectors v, v two complex valued

solutions of (4.20) are

z1 (x) = eλx v = eαx (cos βx + i sin βx)(u + iw)

= [(u cos βx − w sin βx) + i(u sin βx + w cos βx)] eαx
z2 (x) = eλx v = eαx (cos βx − i sin βx)(u − iw)
= [(u cos βx − w sin βx) − i(u sin βx + w cos βx)] eαx .

72
Thus, by the Principle of Superposition, two real valued solutions are
1
y1 (x) = (z1 (x) + z2 (x)) = (u cos βx − w sin βx)eαx ,
2
1
y2 (x) = (z1 (x) − z2 (x)) = (u sin βx + w cos βx)eαx .
2i
We can check the linear independence of these solutions using the definition. Sup-
pose
α1 y1 (x) + α2 y2 (x) = 0 ∀x ∈ R. (4.25)
For x = 0 this becomes

α1 y1 (0) + α2 y2 (0) = 0
⇒ α1 u + α2 w = 0.

But this implies α1 = 0 = α2 since u and w are linearly independent vectors. So

the only choice of α1 , α2 so that (4.25) is satisfied for all x ∈ R is α1 = 0 = α2 .
Hence the solutions are linearly independent.

Example 4.23:
Find the general solution of
" #
dy 1 1
= Ay where A = .
dx 4 1

Solution:
First we find the eigenvalues of A. The characteristic equation is:

det(A − λI) = 0
" #
1−λ 1
⇒ det = (1 − λ)2 − 4 = 0
4 1−λ
⇒ λ2 − 2λ − 3 = 0.

Applying the quadratic formula

√
2± 4 + 12 2±4
λ= = = 3, −1
2 2
shows that the eigenvalues are λ1 = 3, λ2 = −1. Next we find an eigenvector for
each eigenvalue.
For λ1 = 3
(A − 3I)v1 = 0

73
 " #" # " #
1−3 1 v11 0
=
4 1−3 v12 0
" #" # " #
−2 1 v11 0
=
4 −2 v12 0

−2v11 + v12 = 0
⇒ .
4v11 − 2v12 = 0
This system implies v12 = 2v11 so the eigenvectors are given by
!
k
k ∈ R.
2k

Choosing k = 1 gives one solution:

! !
1 e3x
y1 (x) = eλ1 x v1 = e3x =
2 2e3x

For λ2 = −1
(A + I)v2 = 0
" #" # " #
2 1 v21 0
=
4 2 v22 0

2v21 + v22 = 0
⇒ .
4v21 + 2v22 = 0
This system implies v22 = −2v21 so the eigenvector is given by
!
k
k ∈ R.
−2k

Choosing k = 1 gives the second solution:

!
1
y2 (x) = e−x .
−2

Applying Theorem 4.18 we know the general solution of the differential equation is
! ! !
3x 1 −x 1 c1 e3x + c2 e−x
y(x) = c1 e + c2 e = , x ∈ R.
2 −2 2c1 e3x − 2c2 e−x

74
Example 4.24:
Find the solution of the initial value problem
! !
dy −1 −1 1
= y, y(0) = .
dx 5 −3 2

Solution:
First we find the eigenvalues of A. The characteristic equation is

det(A − λI) = 0
!
−1 − λ −1
⇒ det =0
5 −3 − λ
⇒ (−1 − λ)(−3 − λ) + 5 = 0
⇒ λ2 + 4λ + 8 = 0.

Applying the quadratic formula

√
−4 ± 16 − 32 −4 ± 4i
λ= = = −2 ± 2i
2 2
shows that the eigenvalues are λ = −2 + 2i, λ = −2 − 2i. Next we find the an
eigenvector for each eigenvalue.
For λ = −2 + 2i
(A − λI)v = 0
! !
−1 − (−2 + 2i) −1 1 − 2i −1
=
5 −3 − (−2 + 2i) 5 −1 − 2i
! ! !
1 − 2i −1 v1 0
=
5 −1 − 2i v2 0

(1 − 2i)v1 − v2 = 0
⇒
5v1 − (1 + 2i)v2 = 0
This system implies v2 = (1 − 2i)v1 so the eigenvector is given by
!
1
k k ∈ R.
1 − 2i

Choosing k = 1 we obtain
! ! !
1 1 0
v= = +i = u + iw.
1 − 2i 1 −2

75
 5.0

2.5

y2 0.0

-2.5
5
-5.0 0 y1
-1 -5
0 1 2
x

Figure 4.1: Plot of solutions for Example 4.24. The thick curve is the solution of the
initial value problem. The thin curves are other members of the general solution.

One can check that v = u − iw is an eigenvector for λ. Then, applying Theorem

4.22 with α = −2, β = 2, the general solution is:

y(x) = c1 eαx (u cos βx − w sin βx) + c2 eαx (u sin βx + w cos βx)

" ! ! # " ! ! #
1 0 1 0
= c1 e−2x cos 2x − sin 2x + c2 e−2x sin 2x + cos 2x
1 −2 1 −2
!
c1 e−2x cos 2x + c2 e−2x sin 2x
= , x ∈ R.
c1 e−2x cos 2x + 2c1 e−2x sin 2x + c2 e−2x sin 2x − 2c2 e−2x cos 2x
!
1
Now apply the initial condition: y(0) =
2
! !
1 c1 1 = c1 c1 = 1
= ⇒ ⇒ .
2 c1 − 2c2 2 = c1 − 2c2 c2 = − 21

The solution is thus

!
e−2x cos 2x − 12 e−2x sin 2x
y(x) = , x ∈ R.
2e−2x cos 2x + 23 e−2x sin 2x

The solution is shown in Figure 4.1.

So far all our theoretical results have been for general n. To proceed further we
need to consider specific values of n or there will be too many cases to consider.

4.2.1 The case n = 2

Since A is a 2 × 2 matrix it has 2 eigenvalues, λ1 , λ2 . There are three possibilities

76
 1. The eigenvalues λ1 , λ2 are real and have linearly independent eigenvectors
v1 , v2 . Then Theorem 4.21 applies. Note the eigenvalues may be equal.

2. The eigenvectors are complex conjugates λ1 = λ, λ2 = λ. Then Theorem 4.22

applies.

3. The eigenvalues are real, equal (λ1 = λ2 = λ) with eigenvector v and no

second, linearly independent eigenvector exists. In other words, every eigen-
vector of λ is a scalar multiple of v.

How to deal with possibility 3

We know one solution is y1 = veλx . A first guess would be to look for a solution
of the form wxeλx where w is to be determined.
Exercise 4.25:
Show that if y1 = veλx is one solution of (4.10) then there is no way to choose the
vector w so that wxeλx is also a solution.

For a second guess we look for a solution of the form

y2 = ueλx + wxeλx

where u and u are to be determined. Upon substitution into the differential equa-
tion (4.10), we arrive at

λueλx + λwxeλx + weλx = A(ueλx + wxeλx ),

⇓
λu + w + λxw = Au + Awx.

Equate coefficients of like powers of x:

x0 : λu + w = Au ⇒ (A − λI)u = w (4.26)
x1 : λw = Aw (4.27)

We know that every solution of (4.27) is kv for some constant k. So take w = v.

then we need to solve (A − λI)u = v for u. It can be shown that this can always
be solved and that u and v are linearly independent vectors.
Exercise 4.26:
Let y1 = veλx and y2 = ueλx + vxeλx with u and v linearly independent vectors.
Show that y1 , y2 are linearly independent functions on R.

77
 We summarize the above discussion in the following.

Proposition 4.27. Suppose that A is a 2 × 2 matrix with one repeated, real eigen-
value, λ, and every eigenvector of λ is a scalar multiple of the vector v. Then the
general solution of the differential equation (4.20) is

y(x) = c1 y1 (x) + c2 y2 (x)

= c1 eλx v + c2 eλx (u + xv)

where u is a solution of
(A − λI)u = v.

Example 4.28:
Find the general solution of
!
dy 1 −1
= Ay where A = .
dx 1 3

Solution:
The characteristic equation is

det(A − λI) = 0
!
1 − λ −1
⇒ det =0
1 3−λ
⇒ (1 − λ)(3 − λ) + 1 = 0
⇒ λ2 − 4λ + 4 = 0
⇒ (λ − 2)2 = 0

Thus the eigenvalues are λ = 2, 2. Now solve to fine the corresponding eigenvectors.

(A − 2I)v = 0
! ! !
1 − 2 −1 v1 0
=
1 3−2 v2 0
! ! !
−1 −1 v1 0
=
1 1 v2 0

−v1 − v2 = 0
⇒ ⇒ v2 = −v1
v1 + v2

78
 !
1
Choose v = (every other eigenvector is a scalar multiple of this). One
−1
solution of the differential equation is then
!
1
y1 (x) = e2x .
−1

We know a second linearly independent solution will be

y2 (x) = e2x (u + xv)

where u satisfies
(A − 2I)u = v
! ! !
−1 −1 u1 1
=
1 1 u2 −1

−u1 − u2 = 1
⇒ ⇒ u2 = −1 − u1
u1 + u2 = −1
Choose u1 = 0 so that !
0
u= .
−1
So a second linearly independent solution is
" ! !#
0 1
y2 (x) = e2x +x ,
−1 −1

and the general solution is

y(x) = c1 y1 (x) + c2 y2 (x)

! " ! !#
1 0 1
= c1 e2x + c2 e2x +x
−1 −1 −1
!
c1 e2x + c2 xe2x
= , x ∈ R.
−(c1 + c2 )e2x − c2 xe2x

4.2.2 The case n = 3

In this case A is a 3 × 3 matrix with 3 eigenvalues, λ1 , λ2 , λ3 . There are four

possibilities.

1. λ1 , λ2 , λ3 ∈ R: 3 linearly independent eigenvectors v1 , v2 , v3 . Theorem 4.21

applies.

79
 2. λ1 , λ2 , λ3 ∈ R: 2 linearly independent eigenvectors v1 , v2 . Find third solution
as in n = 2 case of repeated roots.

3. λ1 = λ2 = λ3 = λ ∈ R: 1 linearly independent eigenvector. See next example.

4. λ1 ∈ R, λ2 = α + iβ, λ3 = α − iβ, α, β ∈ R, β 6= 0. Apply Theorem 4.21 and

Theorem 4.22.

Example 4.29:
Find the general solution of
 
−3 0 −4
y0 = Ay, A =  −1 −1 −1  .
 

1 0 1

Solution:
One 
can verify that the eigenvalues of A are λ = −1, −1, −1 with eigenvector
0
v =  1 . One solution is therefore
 

0
 
0
y1 = eλx v = e−x  1  .
 

As we showed above, a second linearly independent solution is

y2 (x) = eλx (u + xv)


−2
where u satisfies (A − λI)u = v. Solving gives u =  0 .
 

1
In a similar manner to the discussion of the n = 2 case, we can show that a third
linearly independent solution is given by

x2
y3 (x) = eλx (w + xu + v)
2
where u, v are as above and w satisfies

(A − λI)w = u.

80
  
−1
Solving gives w =  0 , so that the general solution is
 

1
y(x) = c1 y1 (x) + c2 y2 (x) + c3 y3 (x)
 
−2c2 − (1 + 2x)c3
2
= e−x  c1 + c2 x + x2 c3  .
 

c2 + c3 (1 + x)

4.2.3 Application

Example 4.30:
Suppose a commodity has a constant supply S = 100 units and the rate of change
of the demand is given by
dD
= 80 − 4P
dt
where P is the price per unit of the commodity. The rate of increase of price is
equal to the rate of decrease of quantity and the initial demand is 70 units and the
initial price is $20/unit. Determine out how the demand and price vary in time.
Solution:
The model is
dD
= 80 − 4P, D(0) = 70
dt
dP
= D − S = D − 100, P (0) = 20
dt
!
D
Let y = , then we can write this in vector form as
P
dy
= Ay + F(t), (4.28)
dt
where ! !
0 −4 80
A= , F= .
1 0 −100
The associated homogeneous equation is thus
dy
= Ay.
dt
The characteristic equation of A is found to be λ2 + 4 = 0, so that the eigenvalues
are λ = ±2i. We find the eigenvectors as usual:
For λ = 2i
(A − λI)v = 0

81
 ! ! !
−2i −4 v1 0
=
1 −2i v2 0
−2iv1 − 4v2 = 0 1 i
⇒ v2 = v1 = − v1
v1 − 2iv2 = 0 2i 2
Choose ! ! !
2 2 0
v= = +i .
−i 0 −1
! ! !
2 0 2
Then λ̄ = −2i has eigenvector v = −i = .
0 −1 i
! !
2 0
Applying Theorem 4.22 with α = 0, β = 2, u = ,w = , shows that
0 −1
the general solution of the associated homogeneous equation is:
" ! ! # " ! ! #
2 0 2 0
yh (t) = c1 cos 2t − sin 2t + c2 sin 2t + cos 2t
0 −1 0 −1
" #
2c1 cos 2t + 2c2 sin 2t
= .
c1 sin 2t − c2 cos 2t

How do we find yp (t)? Since F(t) is a constant function we will look for a constant
solution !
k1
yp = ,
k2
where k1 , k2 are to be determined. Substituting this into differential equation (4.28)
yields ! ! ! !
0 0 −4 k1 80
= +
0 1 0 k2 −100
−4k2 + 80 = 0 k1 = 100
⇒ ⇒ .
k1 − 100 = 0 k2 = 20
Therefore the particular solution is
!
100
yp (t) = .
20

Note: This is an equilibrium (constant) solution of the differential equation.

The general solution to equation (4.28) is thus

y(t) = yh (t) + yp (t)

! !
2c1 cos 2t + 2c2 sin 2t 100
= + .
c1 sin 2t − c2 cos 2t 20

82
 160

140

120

100

80

60

40

20

-1 0 1 2 3 4 5 6 7 8 9 10 11

Figure 4.2: Plot of price and demand functions vs. time. The price function, P , is
given by the solid line while the demand function, D, is represented by the dashed
line. The thick horizontals represent their equilibrium values.

! !
D(0) 70
Now apply the initial condition y(0) = = :
P (0) 20
! !
2c1 + 100 70 c1 = −15
= ⇒ .
−c2 + 20 20 c2 = 0

Thus the solution of the initial value problem is

! !
D(t) 100 − 30 cos 2t
y(t) = = .
P (t) 20 − 15 sin 2t

Note that the demand and price oscillate around their equilibrium values (see Figure
4.2.

In this example we essentially used the Method of Undetermined Coefficients to

find yp . It is difficult to use this method on systems other than to find equilibrium
(constant) solutions. In the next section we consider another approach for finding
particular solutions of linear systems.

83
4.3 Finding a Particular Solution of the Inhomo-
geneous Equation

4.3.1 Method of Variation of Parameters

This method can be used to find yp (x) or the general solution y(x) of equation
(4.7). The main idea used in this method is exactly same as in the Method of
Variation of Parameters for nth order equations developed in subsection 3.5.3. We
will develop the method for the case n = 2 first, and give the generalization for
higher dimensional systems at the end.
Suppose that yh (x) = c1 y1 (x) + c2 y2 (x) is the general solution of the associated
homogeneous equation (4.10). Assume that there is a solution of (4.7) of the form
y(x) = u1 (x)y1 (x) + u2 (x)y2 (x) (4.29)
where u1 (x), u2 (x) are scalar functions to be determined. Taking the derivative:
y0 = u01 y1 + u1 y1 0 + u02 y2 + u2 y2 0
and substituting into (4.7) gives
u01 y1 + u1 y1 0 + u02 y2 + u2 y2 0 = Au1 (x)y1 (x) + Au2 (x)y2 (x) + F.
Using the fact that y1 , y2 are solutions of the homogeneous equation (4.10) this
becomes
u01 y1 + u1 Ay1 + u02 y2 + u2 Ay2 = Au1 y1 + Au2 y2 + F
which simplifies to
u01 y1 + u02 y2 = F. (4.30)
In component form we see this is a linear system in u01 , u02 :
u01 y11 + u02 y21 = F1
u01 y12 + u02 y22 = F2 , (4.31)
with coefficient matrix " #
y11 y21
M= .
y21 y22
Since y1 and y2 are linearly independent on their interval of existence, I, it follows
that det(M ) 6= 0 for all x ∈ I. Thus we for any x ∈ I we can solve (4.31) to find
expression for u01 and u02 , viz.,
u01 = G1 (x)
u02 = G2 (x)

84
where the Gi (x) depend on Fi (x) and yij . These differential equations can be solved
by integrating
Z
u1 = G1 (x) dx + c1 ,
Z
u2 = G2 (x) dx + c2 . (4.32)

Putting (4.32) into (4.29) gives the general solution of (4.7).

Z  Z 
y(x) = G1 (x) dx + c1 y1 (x) + G2 (x) dx + c2 y2 (x)
Z  Z 
= c1 y1 (x) + c2 y2 (x) + G1 (x) dx y1 (x) + G2 (x) dx y2 (x)
| {z }
yh | {z }
yp

Notes:

1. If we leave out the constants of integration in (4.32) then we just get a par-
ticular solution of (4.7)

2. This procedure may be easily generalized to higher dimensional systems. For

an n-dimensional system we assume a solution of the form

y(x) = u1 (x)y1 (x) + u2 (x)y2 (x) + · · · + un (x)yn (x)

where the yi are n linearly independent solutions of the associated homoge-

neous equation. The rest of the procedure is the same.

Example 4.31:
Find the general solution of
y0 = Ay + F(x)
where " # !
1 1 x
A= and F(x) = .
4 1 ex

Solution:
We already showed in Example 4.23 that the solution of the associated homogeneous
equation is

yh (x) = c1 y1 (x) + c2 y2 (x)

! !
e3x e−x
= c1 + c2 .
2e3x −2e−x

85
We will apply Variation of Parameters to find the general solution of the given
inhomogeneous equation. Let y(x) = u1 (x)y1 (x) + u2 (x)y2 (x). As shown above,
this implies
u01 y1 + u02 y2 = F
or, in components form,

u01 e3x + u02 e−x = x (4.33)

u01 (2e3x ) + u02 (−2e−x ) = ex . (4.34)

Solve for u01 and u02 by elimination and then integrate:

4e3x u01 = 2x + ex
1 −3x 1 −2x
⇒ u01 = xe + e
2Z 4 Z
1 −3x 1
⇒ u1 = xe dx + e−2x dx + c1
2 4
1 1 1
= − xe−3x − e−3x − e−2x + c1 ,
6 18 8

4e−x u02 = 2x − ex
1 x 1 2x
⇒ u02 = xe − e
2Z 4 Z
1 x 1
⇒ u2 = xe dx − e2x dx + c2
2 4
x x 1 x 1 2x
= e − e − e + c2 .
2 2 8
Thus the general solution of the inhomogeneous differential equation is

y(x) = u1 y1 (x) + u2 y2
  !
1 −3x 1 −3x 1 −2x 1
= − xe − e − e + c1 e3x
6 18 8 2
  !
x x 1 x 1 2x 1
+ e − e − e + c2 e−x
2 2 8 −2
! !
e3x e−x
= c1 + c 2
2e3x −2e−x
  !   !
1 1 1 x 1 1 1 1 x 1
+ − x− − e + x− − e
6 18 8 2 2 2 8 −2
! !
c1 e3x + c2 e−x 1
3
x − 59 − 41 ex
= +
2c1 e3x − 2c2 e−x − 43 x + 89
= yh (x) + yp (x).

86
Chapter 5

Partial Differential Equations

Partial Differential Equations (PDEs) are differential equations where the unknown
function depends on two or more independent variables. Such equations will
involve partial derivatives. Recall our notation for partial derivatives:
∂u ∂ 2u ∂ 2u
= ux = uxx = uxy .
∂x ∂x2 ∂x∂y
Here are some examples
∂u
= ye5x 1st order, linear PDE for u(x, y)
∂x
uxxx + cos(y)uxy + x2 y 2 uy = x2 u 3rd order, linear PDE for u(x, y
∂ 2u
= 8x + sin(y) 2nd order, linear PDE for u(x, y)
∂x∂y
φ + x2 φ φ + φ + t2 φy φt = 0 3rd order nonlinear PDE for φ(x, t)
ttt 2 tx 2xx
∂y ∂y
+ = cos(xt) 1st order, nonlinear PDE for y(x, t)
∂x ∂t
We will now restate some definitions for partial differential equations.
Definition 5.1: The order of a partial differential equation is the order of the
highest derivative that appears in the equation.

The most general 2nd order partial differential equation for u(x, y) is

H(x, y, u, ux , uy , uxx , uxy , uyy ) = 0. (5.1)

Definition 5.2: The partial differential equation (5.1) is linear if H is a linear

function of u and all of its partial derivatives, otherwise it is called nonlinear.

The most general linear, 2nd order partial differential equation for u(x, y) is

a(x, y)uxx +b(x, y)uxy +c(x, y)uyy +d(x, y)ux +e(x, y)uy +f (x, y)u = g(x, y). (5.2)

87
Recall: For a function F (x, y), ∂F
∂x
means “differentiate F with respect to x holding
y fixed”. We can use this idea to solve very simple partial differential equations as
in the next two examples.
Example 5.3:
∂u
Solve the following partial differential equation ∂x
= ye5x .
Solution:
Integrating both sides with respect to x holding y fixed gives
y
u(x, y) = e5x + F (y),
5
where F (y) is an arbitrary function of y only. This means that the functions y5 e5x ,
y 5x 2
5
e + y 2 , y5 e5x + y sin(y)ey are all solutions of the partial differential equation.
We can check that the solution is correct by differentiating:
∂u ∂  y 5x  ∂
= e + (F (y))
∂x ∂x 5 ∂x
y 5x
= (5e ) + 0
5
= ye5x

Example 5.4:
∂2u
Solve the following partial differential equation ∂x∂y
= 8x + sin(y).
Solution:
Start by integrating with respect to x holding y fixed, to obtain
∂u
= 4x2 + x sin(y) + F (y),
∂y
where F (y) is an arbitrary function of y only. Now integrate with respect to y
holding x fixed, to obtain
Z
2
u(x, y) = 4x y − x cos(y) + F (y) dy + G(x),

where G(x) is an arbitrary function of x only. Since F (y) is an arbitrary function

of y, then so is its antiderivative, so we may write the solution in the simpler form

u(x, y) = 4x2 y − x cos(y) + H(y) + G(x)

where G and H are arbitrary functions of x and y, respectively.

Note: Example 5.3 is a 1st order partial differential equation and has one
arbitrary function in the solution. Example 5.4 is a 2nd order partial differential
equation and has two arbitrary functions in the solution.

88
Definition 5.5: The general solution of an nth order partial differential equation
is an expression containing n arbitrary functions that represents all nonsingular
solutions of the differential equation
Definition 5.6: A particular solution of a partial differential equation is a so-
lution containing no arbitrary functions.
Example 5.7:
The general solution of the PDE in Example 5.4 is

u(x, y) = 4x2 y − x cos(y) + H(y) + G(x).

Some particular solutions of this PDE are:

u(x, y) = 4x2 y−x cos(y), u(x, y) = 4x2 y−x cos(y)+ey u(x, y) = 4x2 (y+1)−cos(y)(x+1).

A particular solution may be obtained from the general solution by specifying

the arbitrary functions, as in the previous example, or by applying boundary
conditions (BCs).
Examples of boundary conditions

u(x0 , y) = f (y)
• ∂u
where f, g are known functions
|
∂x (x,y0 )
= g(x)

u(x, y)|x2 +y2 =1 = u0

•
limx→±∞ u(x, y) = 0

The number of boundary conditions needed depends on the order of the partial
differential equation.
Definition 5.8: A differential equation together with a boundary condition is
called boundary value problem (BVP).

Not all boundary value problems have solutions as the following example illus-
trates.
Example 5.9:
Find the solution (if any) of the boundary value problems consisting of the partial
differential equation ∂u ∂x
= ye5x and each of the following boundary conditions:
(i) u(0, y) = y 3 , (ii) u(1, y) = 0, (iii) u(x, 0) = x2 , (iv) uy (0, y) = 1.
Solution:
Recall from Example 5.3 that the general solution of the partial differential equation
is u(x, y) = y5 e5x + F (y). Now apply each of the boundary conditions.

89
(i) u(0, y) = y 3 :
y y
+ F (y) = y 3 ⇒ F (y) = y 3 − .
5 5
So the boundary value problem has the unique solution
y y y
u(x, y) = e5x + y 3 − = y 3 + (e5x − 1).
5 5 5
(ii) u(1, y) = 0:
y 5 y
e + F (y) = 0 ⇒ F (y) = −e5 .
5 5
So the boundary value problem has the unique solution
y y y
u(x, y) = e5x − e5 = (e5x − e5 ).
5 5 5

(iii) u(x, 0) = x2 :
F (0) = x2
There is no way to choose F to satisfy this equation, so there is no solution
to the boundary value problem.

(iv) uy (0, y) = 1:
1 5x
uy (x, y) = e + F 0 (y)
5
uy (0, y) = 1
1
⇒1 = + F 0 (y)
5
4
⇒ F 0 (y) =
5
4
⇒ F (y) = y + C,
5
where C is an arbitrary constant. Thus there is a one parameter family of
solutions to the boundary value problem
y 4
u(x, y) = e5x + y + C,
5 5
i.e., the boundary value problem has a solution but it is not unique.

Example 5.10:
∂2u
Find the solution of the boundary value problem ∂x∂y
= 8x + sin(y), u(x, 0) =
x2 , uy (0, y) = e2y .
Solution:
From Example 5.4, the general solution of the differential equation is

u(x, y) = 4x2 y − x cos(y) + H(y) + G(x).

90
Applying the boundary condition u(x, 0) = x2 we have

−x + H(0) + G(x) = x2 ⇒ G(x) = x2 + x − H(0),

which gives the solution thus far as

u(x, y) = 4x2 y − x cos(y) + H(y) + x2 + x − H(0).

To apply the second condition we first calculate the partial derivative with respect
to y:
uy (x, y) = 4x2 + x sin(y) + H 0 (y).
Then the boundary condition uy (0, y) = e2y implies

H 0 (y) = e2y
1 2y
H(y) = e + C.
2
We thus have a final solution of
1
u(x, y) = 4x2 y − x cos(y) + e2y + C + x2 + x − H(0),
2
which simplifies to
1
u(x, y) = 4x2 y − x cos(y) + (e2y − 1) + x2 + x,
2
upon noting that H(0) = C + 12 .

Geometric Interpretation

A particular solution u(x, y) = f (x, y) of a partial differential equation with 2

independent variables can be drawn by letting z = f (x, y). Then one can visualize
it as a surface in R3 . The general solution will be a family of surfaces. Figure 5.1
shows plots of the solutions of the boundary value problems of Example 5.9. (i)
and (ii).
The following example gives another illustration of the solution of a boundary
value problem for a second order partial differential equation.
Example 5.11:
Find the solution of the boundary value problem

∂ 2u ∂u
=2 + y, u(x, 0) = x2 , uy (0, y) = 0.
∂x∂y ∂y

91
 4 4

3 3

2 2

1 1
u
0 2 0 2
-1 1.5 -1 1.5

-2
-1 1 -2
-1 1
-0.5 y -0.5 y
0 0.5 0 0.5
0.5 0.5
x 1 x 1
1.5 0 1.5 0
2 2

Figure 5.1: Plot of the solutions to the boundary value problems of Example 5.9
(i) and (ii).

Solution:
To begin, rewrite the differential equation as
 
∂ ∂u
− 2u = y.
∂y ∂x
Next, integrate with respect to y holding x fixed
∂u y2
− 2u = + F (x).
∂x 2
Thinking of y as fixed, this is a linear ordinary differential equation in x with inte-
grating factor e−2x . Multiplying through by the integrating factor and rearranging
the left hand side we obtain
∂  −2x  y2
e u = e−2x + e−2x F (x).
∂x 2
Now integrate with respect to x holding y fixed to obtain
e−2x y 2
Z
−2x
e u=− + e−2x F (x) dx + G(y),
4
where F (x) and G(y) are arbitrary functions. Solving for u(x, y) and simplifying
the arbitrary functions as in Example 5.10, we obtain the general solution
y2
Z
u(x, y) = − + e 2x
e−2x F (x) dx + e2x G(y)
4
y2
= − + H(x) + e2x G(y),
4
92
Applying the boundary condition u(x, 0) = x2 we have

H(x) + e2x G(0) = x2 ⇒ H(x) = x2 − e2x G(0),

which gives the solution thus far as

y2
u(x, y) = − + x2 − e2x G(0) + e2x G(y).
4
To apply the second boundary condition we first calculate
y
uy (x, y) = − + e2x G0 (y).
2
Then the boundary condition uy (0, y) = 0 implies
y
0 = − + G0 (y)
2
0 y
G (y) =
2
y2
G(y) = + C.
4
We thus have a final solution of
y2 y2
u(x, y) = − + x2 − e2x G(0) + e2x ( + C)
4 4
y2 e2x 2
y
= − + x2 +
4 4
2
y
= x2 + (e2x − 1)
4
where we have used the fact that G(0) = C. See the graph in figure 5.2.

5.1 Linear Partial Differential Equations - Oper-

ator Notation (Optional)

Recall that we showed any second order linear partial differential equation for
u(x, y) can be written in the form

a(x, y)uxx +b(x, y)uxy +c(x, y)uyy +d(x, y)ux +e(x, y)uy +f (x, y)u = g(x, y). (5.3)

Consider the following operators:

The operator Dx takes the partial derivative with respect to x of the input function:
∂u
Dx u(x, y) = = ux .
∂x

93
 y 2 2x
Figure 5.2: 3-D plot of function x2 + 4
(e − 1)

94
 The operator Dy takes the partial derivative with respect to y of the input function:
∂u
Dy u(x, y) = = uy .
∂y

By extension, we have
∂ 2u
Dx2 u = Dx (Dx u) = = uxx
∂x2
∂ 2u
Dy2 u = Dy (Dy u) = 2 = uyy .
∂y
We can now rewrite (5.3) in operator notation:

[a(x, y)Dx2 + b(x, y)Dx Dy + c(x, y)Dy2 + d(x, y)Dx + e(x, y)Dy + f (x, y)I]u = g(x, y).
(5.4)
Note that the operator acting on u is a “polynomial” of degree 2 in Dx and Dy
with coefficients which are functions of x and y. More generally, we can write any
nth order, linear partial differential equation for u(x, y) in the form

φ(Dx , Dy )u = g(x, y) (5.5)

where φ is a polynomial of degree n in Dx and Dy with coefficients that are functions

of x and y. Even more generally, any nth order, linear partial differential equation
for u(x1 , x2 , · · · , xk ) which can be written in the form

φ(Dx1 , Dx2 , · · · , Dxk )u = g(x1 , x2 , · · · , xk ) (5.6)

where φ is a polynomial operator of degree n in Dx1 , Dx2 , . . . , Dxk with coefficients

that are functions of the independent variables x1 , x2 , . . . , xk .
The complementary equation of (5.6) is

φ(Dx1 , Dx2 , · · · , Dxk )u = 0. (5.7)

Note: φ(Dx1 , Dx2 , · · · , Dxk ) is a linear operator, i.e., φ(a1 u1 + a2 u2 ) = a1 φ(u1 ) +

a2 φ(u2 ) for any constants a1 , a2 and functions u1 , u2 .
This leads to the following

Theorem 5.12. The general solution of (5.6) is

u = uh + up

where uh is the general solution of (5.7) and up is a particular solution of (5.6).

95
Example 5.13:
The PDE in Example 5.11 can be written in operator notation as

Φ(Dx , Dy ) = y

where Φ(Dx , Dy ) = Dx Dy − 2Dy . The complementary equation is

Φ(Dx , Dy ) = 0,

with general solution uh (x, y) = H(x) + e2x G(y). Thus the general solution of the
original PDE can be written

u(x, y) = uh + up

where up (x, y) = −y 2 /4.

Exercise 5.14:
Rewrite the PDEs in Examples 5.3 and 5.4 in operator form. Identify uh and up in
the general solutions given in those examples.

5.2 Linear, First Order Partial Differential Equa-

tions

The most general first order partial differential equation with two independent
variables is given by
G(x, y, u, ux , uy ) = 0. (5.8)
Definition 5.15: The first order PDE (5.8) is linear if G is a linear function of
u, ux and uy .

The most general linear, first order partial differential equation with two inde-
pendent variables is given by

a(x, y)ux + b(x, y)uy + c(x, y)u = f (x, y). (5.9)

When a(x, y) 6= 0 and b(x, y) = 0 equation (5.9) becomes

a(x, y)ux + c(x, y)u = f (x, y). (5.10)

This equation can always be solved by fixing y and treating it as a first order linear
differential equation in x. (Note that we did this in Example 5.11)Similarly, one
can always solve equation (5.9) in case a(x, y) = 0, b(x, y) 6= 0.

96
 To deal with the general equation (5.9) when a(x, y) 6= 0 and b(x, y) 6= 0 we
will make a change of variables to try to put (5.9) into a form like (5.10). We will
illustrate the idea with an example and discuss the general equation.
Example 5.16:
Solve ux + uy = 0.
Solution:
Consider the change of variables

ξ = x, η = x − y,

and define û(ξ, η) via u(x, y) = û(ξ(x, y), η(x, y)). Then, using the chain rule, we
have

ux = ûξ ξx + ûη ηx = ûξ + ûη ,

uy = ûξ ξy + ûη ηy = −ûη .

Substituting these expressions into the differential equation gives

ûξ = 0.

This equation is easily solved by integrating with respect to ξ holding η fixed, to

yield
û(ξ, η) = F (η),
where F is an arbitrary function of η. The general solution of the original differential
equation is
u(x, y) = F (x − y),
where F is an arbitrary function of one variable. Note that F must be a differen-
tiable function so that the partial derivatives of u are well defined.

Note that solution in Example 5.16 does not depend on x and y independently
but only in the combination x−y. This means that along the lines x−y = constant
the solution is constant. The lines will thus give structure to the solution and are
called the characteristics of the partial differential equation. A particular
solution is shown in Figure 5.3. The characteristics are the thick lines.
How do we find the change of variables to simplify a partial differential equation
as in Example 5.16? We will show how this can be done for any linear, first order
equation (5.9). To do this we will need the following.

97
 1

0.5

–0.5

–1
–2
0
y 2 –1 –2 –3
2 1 0
3
x

Figure 5.3: Plot of the particular solution of Example 5.16, u(x, y) = sin(x − y).

Lemma 5.17. Let φ(x, y) = k be the general solution of the first order ordinary
dy
differential equation dx = f (x, y). Then φ satisfies

φx
= −f (x, y).
φy

Proof. Consider the general solution φ(x, y) = k. Differentiating implicitly with

respect to x yields
dy
φx + φy = 0.
dx
Using the differential equation this becomes

φx + φy f (x, y) = 0,

which, upon rearranging, yields the result:

φx
= −f (x, y).
φy

98
 Consider equation (5.9), where a(x, y) 6= 0. We will apply the change of variables

ξ = ξ(x, y) η = η(x, y).

We require this transformation to be one-to-one. Thus we need the determinant of

the Jacobian of the transformation to be nonzero, i.e.,
" #
ξx ξy
det J(ξ, η) = det = ξx ηy − ξy ηx 6= 0.
ηx ηy

This means that the transformation is invertible and we can write x = x(ξ, η), y =
y(ξ, η). Define
û(ξ, η) = u(x(ξ, η), y(ξ, η))
or, equivalently,
u(x, y) = û(ξ(x, y), η(x, y)).
As we saw in Example 5.16

ux = ûξ ξx + ûη ηx
uy = ûξ ξy + ûη ηy ,

so the differential equation becomes

a(x, y)(ûξ ξx + ûη ηx + b(x, y)(ûξ ξy + ûη ηy ) + c(x, y)û = f (x, y)

or
! !
a(x, y)ξx + b(x, y)ξy ûξ + a(x, y)ηx + b(x, y)ηy ûη + c(x, y)û = f (x, y).

Thus, if we choose η(x, y) so that

a(x, y)ηx + b(x, y)ηy = 0, (5.11)

then the differential equation becomes

(a(x, y)ξx + b(x, y)ξy )ûξ + c(x, y)û = f (x, y). (5.12)

Note that (5.11) can be rewritten (assuming ηy 6= 0)

ηx b(x, y)
=− . (5.13)
ηy a(x, y)

99
Thus it follows from Lemma 5.17 that if we choose η = φ(x, y) where φ(x, y) = k
is the general solution of the ordinary differential equation

dy b(x, y)
=
dx a(x, y)

that it will satisfy equation (5.13).

How to choose ξ? The only other constraint is that the transformation be
one-to-one, i.e.,
det J(ξ, η) = ξx ηy − ξy ηx 6= 0.
We have already assumed ηy 6= 0 so a simple choice is ξ = x, for then ξx = 1, ξy = 0
and det J(ξ, η) = ηy 6= 0.
Summary: Given the first order linear partial differential equation (5.9), the
change of variables
ξ = x η = φ(x, y) (5.14)
where φ(x, y) = k is the general solution of the ordinary differential equation

dy b(x, y)
= (5.15)
dx a(x, y)

transforms equation (5.9) into

â(ξ, η)ûξ + ĉ(ξ, η)û = fˆ(ξ, η)

where â(ξ, η) = a(x(ξ, η), y(ξ, η)), ĉ(ξ, η) = c(x(ξ, η), y(ξ, η)) and fˆ(ξ, η) = f (x(ξ, η), y(ξ, η)).
Definition 5.18: Equation (5.15) is called the characteristic equation of the
partial differential equation (5.9). The curves defined by η(x, y) = k, i.e., φ(x, y) =
k are called the characteristics of the partial differential equation (5.9).
Exercise 5.19:
Find the characteristic equation for the partial differential equation of Example 5.16
and show that the method outlined above leads to the change of variables used in
the example.
Example 5.20:
Find and sketch the characteristics of the partial differential equation

ux − xuy = 4,

then find the solution with the boundary condition u(0, y) = sin(y).

100
 3

2
k>0

-5 -4 -3 -2 -1 0 1 2 3 4 5

k=0

-1

k<0
-2

-3

Figure 5.4: Characteristics of partial differential equation ux − xuy = 4.

Solution:
We have a(x, y) = 1 and b(x, y) = −x so the characteristic equation is
dy x
=− .
dx 1
The general solution of this is
x2
y=− + k,
2
which defines the characteristics for the equation. The characteristics are shown in
Figure 5.4.
Making the change of variables
x2
ξ=x η=y+ ,
2
puts the differential equation in the form

ûξ = 4.

Integrating with respect to ξ holding η fixed gives

û(ξ, η) = 4ξ + F (η)

which upon conversion to our original variables yields

x2
u(x, y) = 4x + F (y + ).
2
Applying the boundary condition u(0, y) = sin(y) implies

F (y) = sin(y)

101
 3

2 k>0
k<0

k=0
-5 -4 -3 -2 -1 0 1 2 3 4 5

-1

-2 k<0
k>0

-3

Figure 5.5: Characteristics of partial differential equation xux − yuy − 2x2 u = x3 y.

so that the solution of the initial value problem is

x2
u(x, y) = 4x + sin(y + ).
2
Example 5.21:
Find and sketch the characteristics of the partial differential equation

xux − yuy − 2x2 u = x3 y, x 6= 0,

then find the general solution.

Solution:
Here a = x, b = −y so the characteristic equation is
dy y
=− .
dx x
This is a separable ordinary differential equation which may be solved in the usual
way:
Z Z
dy dx
= −
y x
ln |y| = − ln |x| + k1
k
y = (k = ±ek1 ).
x
Thus the characteristics are the hyperbolas xy = k, as shown in Figure 5.5.
Making the change of variables

ξ=x η = xy

102
puts the differential equation in the form
η
ξuξ − 2ξ 2 u = ξ 3
ξ
which simplifies to
uξ − 2ξu = ξη.
This is a first order, linear ordinary differential equation with integrating factor
2
e−ξ . Thus it can be solved in the usual way:
∂  −ξ2  2
e u = ξe−ξ η
∂ξ
2 1 2
e−ξ u = − e−ξ η + F (η)
2
η 2
u(ξ, η) = − + eξ F (η).
2
Reverting to the original variables yields the general solution
xy 2
u(x, y) = − + ex F (xy), x 6= 0.
2

5.3 Linear, Second Order Partial Differential Equa-

tions

The most general linear, second order partial differential equation with two inde-
pendent variables is given by

a(x, y)uxx +b(x, y)uxy +c(x, y)uyy +d(x, y)ux +e(x, y)uy +f (x, y)u = g(x, y). (5.16)

Assume a(x, y) 6= 0. We will attempt to simplify this equation using the same
approach that we used for first order partial differential equations, that is, using
the change of variables
ξ = ξ(x, y) η = η(x, y)
where det J(ξ, η) 6= 0, i.e., ξx ηy − ξy ηx 6= 0.
Let û(ξ, η) = u(x(ξ, η), y(ξ, η)). The first derivatives are as found in the previous
subsection:

ux = ûξ ξx + ûη ηx ,
uy = ûξ ξy + ûη ηy .

103
The second derivatives are given by

uxx = ûξξ ξx2 + ûξ ξxx + 2ûξη ξx ηx + ûηη ηx2 + ûη ηxx ,
uxy = ûξξ ξx ξy + ûξ ξxy + ûξη ξx ηy + ûξη ξy ηx + ûηη ηx ηy + ûη ηxy ,
uyy = ûξξ ξy2 + ûξ ξyy + 2ûξη ξy ηy + ûηη ηy2 + ûη ηyy .

In the new variables the differential equation becomes

A(ξ, η)ûξξ + B(ξ, η)ûξη + C(ξ, η)ûηη + D(ξ, η)ûξ + E(ξ, η)ûη + fˆ(ξ, η)û = ĝ(ξ, η),
(5.17)
where
A = aξx2 + bξx ξy + cξy2
B = 2aξx ηx + b(ξx ηy + ξy ηx ) + 2cξy ηy
C = aηx2 + bηx ηy + cηy2 (5.18)
D = aξxx + bξxy + cξyy + dξx + eξy
E = aηxx + bηxy + cηyy + dηx + eηy
and
fˆ = f (x(ξ, η), y(ξ, η)), ĝ(ξ, η) = g(x(ξ, η), y(ξ, η)).
Exercise 5.22:
Verify that the formulas given in equation 5.18 are correct.

Now we will determine how to choose η and ξ to simplify equation (5.17). To

make A = 0 we require
aξx2 + bξx ξy + cξy2 = 0,
or, assuming ξy 6= 0,
 2
ξx ξx
a +b + c = 0.
ξy ξy
This means that either √
ξx −b + b2 − 4ac
= (5.19)
ξy 2a
or √
ξx −b − b2 − 4ac
= . (5.20)
ξy 2a
Similarly to make C = 0 requires (assuming ηy 6= 0) either
√
ηx −b + b2 − 4ac
= (5.21)
ηy 2a
or √
ηx −b − b2 − 4ac
= . (5.22)
ηy 2a

104
Now ξ(x, y) and η(x, y) are real valued functions as are a(x, y), b(x, y) and c(x, y).
So whether it is possible to satisfy equations (5.19)-(5.22) depends on the functions
a, b, c. There are three different cases, which are used to classify all second order
partial differential equations as described below.
Definition 5.23: Suppose a(x, y), b(x, y), c(x, y) are continuous functions and a(x, y) 6=
0 on the region D ⊆ R2

I If b2 − 4ac > 0 on D then the partial differential equation (5.16) is called hyper-
bolic.

II If b2 − 4ac = 0 on D then the partial differential equation (5.16) is called

parabolic.

III If b2 − 4ac < 0 on D then the partial differential equation (5.16) is called
elliptic.

Note that if a, b, c are constants (a, b, c ∈ R) then the partial differential equation
will be of one class on all of R2 . If any of a, b, c is a function of x and y then the
partial differential equation may change class as x and y vary.
Example 5.24:
Consider the partial differential equation

uxx + 4yuxy + 9uyy − ux + uy = 0.

Determine the regions in R2 where the partial differential equation is hyperbolic,

parabolic and elliptic. Sketch and label the regions.
Solution:
In this case we have a(x, y) = 1, b(x, y) = 4y, c(x, y) = 9 so that

b2 − 4ac = 16y 2 − 36 = 4(4y 2 − 9).

The partial differential equation is then

Hyperbolic if 4y 2 − 9 > 0, i.e., if y > 3/2 or y < −3/2,

Parabolic if 4y 2 − 9 = 0, i.e., if y = ±3/2,
Elliptic if 4y 2 − 9 < 0, i.e., if − 3/2 < y < 3/2.

The regions are shown in Figure 5.6.

We will now show how each class of second order linear partial differential
equation can be reduced to its simplest form, which is called the canonical form.

105
 3

Hyperbolic

Parabolic

Elliptic

-5 -4 -3 -2 -1 0 1 2 3 4 5

-1

Parabolic

-2

Hyperbolic

-3

Figure 5.6: Classification of the partial differential equation uxx + 4yuxy + 9uyy −
ux + uy = 0.

Class I - Hyperbolic Partial Differential Equations

Since b2 − 4ac > 0, the ordinary differential equations

√
dy b + b2 − 4ac
= (5.23)
dx 2a
and √
dy b − b2 − 4ac
= (5.24)
dx 2a
will have real valued solutions. Suppose that (5.23) and (5.24) have general solu-
tions
φ1 (x, y) = k (5.25)
φ2 (x, y) = k (5.26)
respectively. Then from Lemma 5.17 we know that choosing

ξ = φ1 (x, y)

will guarantee √
ξx −b − b2 − 4ac
= ⇒ A = 0,
ξy 2a
while choosing
η = φ2 (x, y)

106
will guarantee √
ηx −b + b2 − 4ac
= ⇒ C = 0.
ηy 2a
Note: It is easy to check that ξx ηy − ξy ηx 6= 0 which guarantees that this transfor-
mation is one-to-one. Further, one can show that B 6= 0.
It follows that the transformation ξ = φ1 (x, y), η = φ2 (x, y) reduces the partial
differential equation (5.16) to

B ûξη + Dûξ + E ûη + fˆ(ξ, η)û = ĝ(ξ, η).

Since B 6= 0 we can divide through to get the canonical form for hyperbolic partial
differential equations
ûξη + Φ(ξ, η, û, ûξ , ûη ) = 0,
where Φ = (Dûξ + E ûη + fˆû − ĝ)/B.
Definition 5.25: Equation (5.23) and (5.24) are called the characteristic equa-
tions for a hyperbolic partial differential equation. The families of curves in the xy
plane corresponding to the general solutions (5.25) and (5.26) of these differential
equations are called the characteristics.

Class II - Parabolic Partial Differential Equations

In this case b2 − 4ac = 0, thus there is only one characteristic equation

dy b(x, y)
= (5.27)
dx 2a(x, y)
and one set of characteristics, given by the general solution of this differential
equation:
φ(x, y) = k. (5.28)
Let ξ = φ(x, y) then
ξx b
=−
ξy 2a
and it follows (noting that c = b2 /4a) that A = 0.
Now, we cannot make C = 0 by choosing η = φ(x, y) as well because then ξ = η
and the transformation would not be one-to-one. However, in this case we have

B = (2aξx + bξy )ηx + (bξx + 2cξy )ηy

b2
   
ξx ξx
= 2a + b ξy ηx + b + ξy ηy
ξy ξy 2a
= 0.

107
If we choose η(x, y) = x it can be shown that the transformation is invertible and
C 6= 0. Thus, using the substitution

ξ = φ(x, y) η = x,

the partial differential equation (5.16) becomes

C(x, y)ûηη + Dûξ + E ûη + fˆ(ξ, η)û = ĝ(ξ, η).

Dividing through by C gives the canonical form for any parabolic partial differential
equation
ûηη + Φ(ξ, η, û, ûξ , ûη ) = 0,
where Φ = (Dûξ + E ûη + fˆû − ĝ)/C.

Class III - Elliptic Partial Differential Equations

Here b2 − 4ac < 0 so the characteristic equations (5.23) and (5.24) do not have
real valued solutions. Thus the partial differential equation has no characteristics.
However, using other transformations, we may still simplify the partial differential
equation as we now show.
Noting that

B = 2aξx ηx + b[ξx ηy + ξy ηx ] + 2cξy ηy

= ξx (2aηx + bηy ) + ξy (bηx + 2cηy ),

we see that the transformation ξ = φ(x, y), η = x where φ(x, y) = k is the general
solution of
dy bηx + 2cηy
= ,
dx 2aηx + bηy
will make B = 0. Thus the partial differential equation (5.16) becomes

Aûξξ + C ûηη + Φ(ξ, η, û, ûξ , ûη ) = 0.

A second, more complicated, transformation can be used to make the coefficients

of the first two terms equal. Thus the canonical form for elliptic partial differential
equations is
ûξξ + ûηη + Φ(ξ, η, û, ûξ , ûη ) = 0.

108
 3 3

2 2

1 1

-5 -4 -3 -2 -1 0 1 2 3 4 5 -5 -4 -3 -2 -1 0 1 2 3 4 5

-1 -1

-2 -2

-3 -3

(a) (b)

Figure 5.7: Characteristics for uxx + 2uxy − 3uyy = 0: (a) y − 3x = k, (b) y + x = k.

Example 5.26:
Consider
uxx + 2uxy − 3uyy = 0.
Classify the equation and sketch the characteristics if any. Then put the equation
into its canonical form and find the general solution.
Solution:
Here we have a = 1, b = 2, c = −3, so b2 − 4ac = 4 + 12 = 16 > 0 for all (x, y) ∈ R2 .
Thus the partial differential equation is hyperbolic on R2 .
The characteristic equations are
√ √
dy b+ b2 −4ac dy b2 −4ac
dx
= 2a
and dx
= b− 2a
=3 = −1,
which have general solutions

y = 3x + k and y = −x + k.

Thus the two families of characteristics are y − 3x = k and y + x = k. These are

illustrated in Figure 5.7.
Choose ξ = y − 3x and η = y + x. Then, from the derivation for Class I above,
we know that A = 0 and C = 0. Further, since d = e = f = g = 0 and the second
partial derivatives of ξ and η are zero, from equation (5.18) we have

D = E = fˆ = ĝ = 0.

Thus the differential equation becomes

B ûξη = 0,

109
where
B = 2aξx ηx + b[ξx ηy + ξy ηx ] + 2cξy ηy = −16 6= 0.
This implies that
ûξη = 0,
or,  
∂ ∂ û
= 0.
∂ξ ∂η
Integrating with respect to ξ holding η fixed yields
∂ û
= F (η).
∂η
A further integration with respect to η holding ξ fixed then gives
Z
û(ξ, η) = F (η) dη + G(ξ)
= H(η) + G(ξ).

Thus the general solutions of the original partial differential equation is

u(x, y) = H(y + x) + G(y − 3x)

where H : R → R, G : R → R are arbitrary, twice differentiable functions.

We can check that this is correct by differentiating u(x, y).

ux = H 0 (y + x) − 3G0 (y − 3x)
uxx = H 00 (y + x) + 9G00 (y − 3x)
uxy = H 00 (y + x) − 3G00 (y − 3x)
uy = H 0 (y + x) + G0 (y − 3x)
uyy = H 00 (y + x) + G00 (y − 3x)

Substituting into the LHS of the original differential equation gives:

(1 + 2 − 3)H 00 (y + x) + (9 − 6 − 3)G00 (y − 3x),

which is 0, so the equation is satisfied.

Example 5.27:
Classify the partial differential equation

4uxx + 4uxy + uyy = 2 sin(x + y).

Then find the general solution.

110
Solution:
We have a = 4, b = 4, c = 1, so b2 − 4ac = 16 − 16 = 0 and the partial differential
equation is parabolic on R2 . The characteristic equation is
dy b 4 1
= = = ,
dx 2a 8 2
which has general solution
1
y = x + k.
2
Therefore, the characteristics are the lines y − 21 x = k.
Let ξ = y − 21 x, η = x. Then, from the derivation for Class II, we know that
A = 0 and B = 0. Further, since d = e = f = 0 and the second partial derivatives
of ξ and η are zero, from equation (5.18) we have

D = E = fˆ = 0, ĝ = 2 sin(x(ξ, η) + y(ξ, η)).

Now the transformation may be easily inverted to find x = η, y = ξ + 21 η, and the

differential equation becomes
3
C ûηη = 2 sin(ξ + η),
2
where
C = aηx2 + bηx ηy + cηy2 = 4.
Thus, in canonical form, the differential equation is
2 3 1 3
ûηη = sin(ξ + η) = sin(ξ + η).
C 2 2 2
Integrating with respect to η holding ξ fixed yields
1 3
ûη = − cos(ξ + η) + F (ξ).
3 2
Integrating again with respect to η holding ξ fixed then gives
2 3
û(ξ, η) = − sin(ξ + η) + ηF (ξ) + G(ξ).
9 2
Thus the general solution of the original equation is
2 1 1
u(x, y) = − sin(x + y) + xF (y − x) + G(y − x),
9 2 2
where F, G : R → R are arbitrary twice differentiable functions.
Exercise 5.28:
Use differentiation to verify that the solution u(x, y) found in Example 5.27 does
satisfy the given PDE.

Not all second order partial differential equations may be solved using the ap-
proach of the last two examples. Thus we consider a more general approach.

111
5.4 The Fourier Transform

An integral transform is a mapping that takes a function of one variable into a

function of another variable via an integral, i.e.,
Z β
ˆ
f (ω) = f (x)K(ω, x) dx
α

where α, β are constants. fˆ(ω) is called the transform of f (x) and K(ω, x) is
called the kernel of the transform.
Integral transforms are helpful in solving linear differential equations as they
can be used transform differential equations into simpler equations. For example,
an integral transform may change an ordinary differential equation into an algebraic
equation.
Definition 5.29: Let f (x) be defined for all x ∈ R. The Fourier Transform of
f is defined by Z ∞
F{f (x)} = fˆ(ω) = f (x)e−iωx dx
−∞

provided the improper integral converges.

Note: In evaluating improper integrals we will use the following result.

If f (x) is a complex valued function of a real variable x, i.e. f (x) =

fR (x) + ifI (x) where fR (x) and fI (x) are real valued functions, then
limx→a f (x) exists if and only if both limx→a fR (x) and limx→a fI (x) ex-
ist. In this case,

lim f (x) = lim fR (x) + i lim fI (x).

x→a x→a x→a

All the usual rules for evaluating limits can be easily extended to complex valued
functions of a real variable. We will use these rules as needed.
Example 5.30:
Let (
ebx for x < 0,
f (x) =
0 for x ≥ 0,
where b > 0. Determine if F{f (x)} exists. If it does, calculate it.

112
Solution:
Consider
Z ∞ Z 0
−iωx
f (x)e dx = ebx e−iωx dx
−∞ −∞
Z 0
= lim e(b−iω)x dx
s→−∞ s
0
e(b−iω)x

= lim
s→−∞ (b − iω)
s
1
= lim [1 − ebs e−iωs ]
s→−∞ b − iω
1 
1 − ebs (cos ωs − i sin ωs)

= lim
s→−∞ b − iω
1
=
b − iω
b + iω
=
b2 + ω 2
Since the integral converges, the Fourier transform of f (x) exists and

b + iω
F{f (x)} = .
b2 + ω 2
Example 5.31:
Let f (x) = ebx , where b is a fixed real number. Determine if F{f (x)} exists. If it
does, calculate it.

Solution:
Consider
Z ∞ Z ∞
−iωx
f (x)e dx = ebx e−iωx dx
−∞ −∞
Z 0 Z t
(b−iω)x
= lim e dx + lim e(b−iω)x dx
s→−∞ s t→∞ 0
1  1  bt
1 − ebs (cos ωs − i sin ωs) + lim
 
= lim e (cos ωt − i sin ωt) − 1
s→−∞ b − iω t→∞ b − iω

If b < 0 then the first limit diverges. If b > 0 then the second limit diverges. If
b = 0, both limits diverge. Thus the improper integral diverges in both cases and
the Fourier transform of f (x) = ebx does not exist for any b.
Can one determine if the Fourier transform of a function exists without evalu-
ating the improper integral? The next theorem gives a partial answer.

Theorem 5.32. Let f : R → R. If

113
 1. f is piecewise continuous on every interval [−M, M ] for any M > 0
R∞
2. −∞ |f (x)| dx converges (i.e, f is absolutely integrable)

then the Fourier transform of f (x) exists.

Proof. By condition 1, one can write down the improper integral in the definition
of the Fourier transform as a sum of (improper) integrals. Now
Z ∞ Z ∞
−iωx
f (x)e dx ≤ |f (x)e−iωx | dx
−∞
Z−∞
∞
= |f (x)| dx
−∞

since |e−iωx | = 1. It follows from a comparison theorem for improper integrals that
Z ∞
f (x)e−iωx dx
−∞

converges. Hence the Fourier transform of f (x) exists.

Note that this theorem gives sufficient conditions for existence of the Fourier
transform. If a function does not satisfy one or both of these conditions, then the
Fourier transform of the function may or may not exist.
Example 5.33:
Let (
−10 if |x| ≤ 1
f (x) =
0 otherwise
Determine if F{f (x)} exists.

Solution:
Clearly f (x) is piecewise continuous on every interval [−M, M ] for any M > 0.
Now consider
Z ∞ Z 1
|f (x)| dx = 10 dx
−∞ −1
= 20

Thus f (x) is absolutely integrable. Thus we can conclude, applying Theorem 5.32,
that the Fourier transform of f (x) exists.
Another question we may ask is whether the Fourier transform is invertible, i.e.,
given fˆ(ω) can we find f (x) such that F{f (x)} = fˆ(ω)? The answer lies in the
following definition and theorem.

114
Definition 5.34: Let fˆ(ω) be defined for all ω ∈ R. The Inverse Fourier Trans-
form of fˆ(ω) is given by
Z ∞
−1 ˆ 1
F {f (ω)} = fˆ(ω)eiωx dω.
2π −∞

Theorem 5.35. If f is continuous and absolutely integrable and F{f (x)} = fˆ(ω)
then
F −1 {fˆ(ω)} = f (x)

The following theorem shows that the Fourier transform is a linear transforma-
tion.

Theorem 5.36. Let f and g be two functions such that the Fourier transforms of
f and g exist and let α and β be two real numbers. Then the Fourier transform of
αf + βg exists and

F{αf (x) + βg(x)} = αF{f (x)} + βF{g(x)}.

Proof. The proof follows by using the definition.

Z ∞
F{αf (x) + βg(x)} = (αf (x) + βg(x))e−iωx dx
−∞
Z ∞ Z ∞
−iωx
= α f (x)e dx + β g(x)e−iωx dx
−∞ −∞
= αF{f (x)} + βF{g(x)}.

Note that the second step follows from the fact that the Fourier transforms of f
and g both exist, so the all the improper integrals are convergent.

The next theorem shows how the Fourier transform maps derivatives of func-
tions. It is a key result needed when using Fourier transforms to solve differential
equations.

Theorem 5.37. Let f be differentiable on R and suppose that the Fourier transform
of f exists and is given by F{f (x)} = fˆ(ω). Then the Fourier transform of f 0 exists
and is given by F{f 0 (x)} = iω fˆ(ω).
R∞
Proof. Since F{f (x)} exists, −∞
f (x)e−iωx dx converges which implies

lim f (x)e−iωx = 0 and lim f (x)e−iωx = 0.

x→∞ x→−∞

115
Using Integration by Parts we have
Z b Z b
b
f 0 (x)e−iωx dx = f (x)e−iωx a
− f (x)(−iω)e−iωx dx,
a a

for any a, b. Applying this with a = −∞ and b = ∞ gives

Z ∞ Z ∞
0 −iωx −iωt −iωs
f (x)e dx = lim f (t)e − lim f (s)e + iω f (x)e−iωx dx
−∞ t→∞ s→−∞ −∞

= iω fˆ(ω).

The result follows.

Corollary 5.38. If the Fourier transforms of f (x) and F (x) exist with F{f (x)} =
fˆ(ω) and f (x) = F 0 (x) then F{F (x)} = fˆ(ω)/(iω).

Theorem 5.37 can be generalized as follows.

Theorem 5.39. Let n be an integer. If the Fourier transform of f (x) exists with
F{f (x)} = fˆ(ω), and f is n times differentiable on IR, then the Fourier transform
of f (n) exists and is given by F{f (n) } = (iω)n fˆ(ω)

Proof. The proof follows from repeated application of Theorem 5.37.

Another useful property of the Fourier transform is given in the following the-
orem.
Theorem 5.40 (Shifting Property). Let f (x) be continuous and absolutely inte-
grable with Fourier transform fˆ(ω). Then

F −1 {fˆ(ω)e−iωt } = f (x − t).

Proof. Consider the integral

Z ∞ Z ∞
1 ˆ 1
f (ω)e−iωt iωx
e dω = fˆ(ω)eiω(x−t) dω.
2π −∞ 2π −∞
Making the change of variables, η = x − t:
Z ∞
1
= fˆ(ω)eiωη dω.
2π −∞
= F −1 {fˆ(ω)}
Applying Theorem 5.35:
= f (η)
Returning to the original variables, η = x − t:
= f (x − t)

116
Example 5.41:
Use the Fourier transform to find the general solution of the partial differential
equation
ux + uy = 0.

Solution:
Assume that u is continuous and absolutely integrable with respect to x (i.e.,
R∞
−∞
|u(x, y)| dx < ∞, ∀y ∈ R). Then the Fourier transform of u with respect
to x exists. Let
Z ∞
û(ω, y) = F{u(x, y)} = u(x, y)e−iωx dx.
−∞

Then F{ux } = iωû and

Z ∞ Z ∞
∂u −iωx ∂ ∂ û
F{uy } = e dx = u(x, y)e−iωx dx = = ûy .
−∞ ∂y ∂y −∞ ∂y
So the partial differential equation becomes

iωû + ûy = 0,

which we rewrite as
∂ û
= −iωû.
∂y
Fixing ω, this is a separable ordinary differential equation for û with respect to y.
Solving in the usual way gives
Z Z
∂ û
= − iω ∂y
û
ln |û| = −iωy + F̂ (ω)

So the general solution is

û(ω, y) = Ĝ(ω)e−iωy ,
where Ĝ is an arbitrary function of ω.
To find the solution of original partial differential equation we need to apply
the inverse Fourier transform (assuming it exists):

u(x, y) = F −1 {û(ω, y)}

= F −1 {Ĝ(ω)e−iωy }.

If we assume that Ĝ has inverse Fourier transform G(x) which is absolutely inte-
grable and continuous, then we may apply the Shifting Property (Theorem 5.40)
to conclude that the general solution of the differential equation is

u(x, y) = G(x − y),

117
where G is an arbitrary (differentiable) function. Note that this solution is the
same as what we found in Example 5.16 using the method of characteristics.

In the last example, we needed to find the inverse Fourier transform of a function
that was a product of two functions of ω, û(ω, y) = Ĝ(ω)e−iωy . We were able to
do this directly using the Shifting Property, but that may not always be the case.
The next theorem deals with this situation in a more general way. First we need a
definition.
Definition 5.42: The convolution of f and g is
Z ∞
(f ∗ g)(x) = f (x − z)g(z) dz.
−∞

If the argument is understood or unimportant the we will simply write f ∗ g for

the convolution of f and g.

Theorem 5.43. Let f (x), g(x) be bounded, continuous and absolutely integrable
with Fourier transforms given by F{f (x)} = fˆ(ω) and F{g(x)} = ĝ(ω). Then

F −1 {fˆ(ω)ĝ(ω)} = f ∗ g.

Proof. (Outline) Consider the following integral.

Z ∞ Z ∞ Z ∞ 
−iωx
(f ∗ g)e dx = f (x − z)g(z) dz e−iωx dx
−∞ −∞ −∞
Switching the order of integration:
Z ∞ Z ∞ 
−iωx
= g(z) f (x − z)e dx dz
−∞ −∞
Making the change of variables ξ = x − z :
Z ∞ Z ∞ 
−iωξ −iωz
= g(z) f (ξ)e e dξ dz
Z−∞
∞ Z ∞ −∞

−iωz −iωξ
= g(z)e f (ξ)e dξ dz
−∞ −∞
Z ∞  Z ∞ 
−iωz −iωξ
= g(z)e dz f (ξ)e dξ
−∞ −∞

Thus the Fourier transform of f ∗ g exists and F{f ∗ g} = fˆ(ω)ĝ(ω). Since f, g

are continuous, f ∗ g is continuous. Further, it can be shown that f, g absolutely
integrable implies f ∗ g absolutely integrable (see Theorem 3.8 of [5]). Applying
the inverse Fourier transform yields the result.

Theorem 5.44. The convolution satisfies the following properties

118
 1. f ∗ g = g ∗ f

2. (αf + βg) ∗ h = α(f ∗ h) + β(g ∗ h)

for any constants α, β.

Proof.
1. From the definition we have
Z ∞
f ∗g = f (x − z)g(z) dz
−∞
Let w = x − z
Z −∞
= − f (w)g(x − w) dw
∞
Z ∞
= g(x − w)f (w) dw
−∞
= g∗f

2. Follows from the definition (exercise).

The Heat Equation

The 1-D heat equation (diffusion equation) is given by

γuxx = ut (5.29)

where γ > 0 is a constant. It was first derived to model the temperature u at a

position x and time t in a thin rod as seen in Figure 5.8.

u(x,t)

Figure 5.8: Example rod used in solving heat equation.

The main idea behind the model is the following:

! !
heat flux through rate of change of
∝
point x at time t temperature at time t

119
 uxx ∝ ut
or
γuxx = ut .
The constant of proportionality, γ, is called the diffusivity or diffusion constant.
The most common boundary value problems involving this equation involve one
condition when t is fixed

u(x, 0) = f (x) (initial condition) (5.30)

and two conditions where x is fixed

u(x1 , t) = g1 (t) (boundary conditions)

u(x2 , t) = g2 (t). (5.31)

These are often called initial/boundary value problems. We will be interested

in problems with the boundary conditions at infinity, which are expressed as limits.
For example,

lim u(x, t) = 0
x→−∞
lim u(x, t) = 0 (5.32)
x→∞

Rewriting the differential equation as

γuxx − ut = 0,

and applying our standard theory (with y replaced by t) we see that a = γ, b =

0, c = 0 thus b2 − 4ac = 0 and the differential equation is parabolic. Since γ 6= 0
we can rewrite the equation as
1
uxx − ut = 0
γ
Exercise 5.45:
Show that the characteristics for the heat equation are given by t = k. Thus the
change of variables needed to simplify the equation to canonical form is ξ = t, η = x.
This shows that the equation is already in canonical form.

We will use the Fourier transform to solve the heat equation subject to the con-
ditions (5.30) and (5.32). Note that the boundary conditions (5.32) are necessary
conditions for u to be absolutely integrable with respect to x.

120
 Assuming that the Fourier transform of u exists (u continuous and absolutely in-
R∞
tegrable with respect to x) let û(ω, t) = F{u(x, t)} = −∞ u(x, t)e−iωx dx. Applying
the differentiation Theorem, we have

F{uxx } = −ω 2 û(ω, t),

and, as we saw in Example 5.41,

F{ut } = ût (ω, t).

Putting these into the given partial differential equation yields the simpler equation

ût + γω 2 û. (5.33)

We can transform the initial condition (5.30) as well to find

û(ω, 0) = fˆ(ω), (5.34)

assuming this exists. We will solve the initial value problem consisting of (5.33)
and (5.34).
For fixed ω, (5.33) is a separable differential equation for û with respect to t,
which may be solved in the usual way.
∂ û
= −γω 2 û
Z ∂t Z
∂ û
= − γω 2 ∂t
û
ln |û| = −γω 2 t + G(ω)
2
û(ω, t) = H(w)e−γω t .

Applying the initial condition û(ω, 0) = fˆ(ω) gives

H(ω) = fˆ(ω).

So the solution of the initial value problem is

2
û(ω, t) = fˆ(ω)e−γω t .

Now apply the inverse Fourier transform

2
u(x, t) = F −1 {û(ω, t)} = F −1 {fˆ(ω)e−γω t }

121
We know F −1 {fˆ(ω)} = f (x). What about F −1 {e−γω t }? Apply the definition of
2

the inverse Fourier transform:

Z ∞ Z ∞
1 −γω 2 t iωx 1 2 ix
e e dω = e−γt[ω − γt ω] dω
2π −∞ 2π −∞
Z ∞ 2
1 ix 2
−γt[(ω− 2γt ) + x2 2 ]
(complete the square) = e 4γ t dω
2π −∞
x2 Z
e− 4γt ∞ −γt(ω− 2γt ix 2
)
= e dω
2π −∞
√ ix √
let v = γt (ω − 2γt ), dv = γt dω
x2
e− 4γt
Z ∞
2
= √ e−v dv
2π γt −∞
x2
e− 4γt √
= √ · π
2π γt
x2
e− 4γt
= √
2 πγt

Therefore
x2

−γω 2 t e− 4γt
F −1 {e }= √ = g(x).
2 πγt
Thus we have
u(x, t) = F −1 {fˆ(ω)ĝ(ω)}
and, by Theorem 5.43,

u(x, t) = f ∗ g
Z ∞
= f (x − z)g(z) dz
−∞
z2
∞
e− 4γt
Z
= f (x − z) √ dz
−∞ 2 πγt
√
for t > 0, x ∈ R. Setting y = −z/(2 γt) this can be expressed as
Z ∞
1 √ 2
u(x, t) = √ f (x + 2 γty)e−y dy.
π −∞

Exercise 5.46:
Verify that the expression above satisfies the heat equation and the conditions
(5.30) and (5.32).

122
Chapter 6

Pricing of Derivatives

6.1 Terminology
An option is the right, but not obligation, to buy or sell a security (e.g., a stock)
for an agreed upon price at (or before) some agreed upon time in the future.

• The agreed upon price is called the strike price and will be denoted by K.

• The agreed upon time is called the strike time or expiry date and will be
denoted by T .

• An option to buy is called a call option.

• An option to sell is called a put option.

• A European option can only be exercised at the strike time.

• An American option can be exercised at or before the strike time.

We will focus on European options.

Fundamental Assumption: an option will only be used (exercised) if it can make
the owner profit.
Value of an option at the strike time
Consider a European call option on a stock, with strike time T and strike price K.
Let the price of the stock at t = T be S. There are two possibilities:

• S > K — The owner of the option uses (exercises) it to buy stock at price
K then sells stock at price S, for profit S − K. Value of option at t = T is
S − K. (Option is in the money).

123
 • S ≤ K — The owner of the option will not exercise it and makes no profit.
We say it expires unexercised. Value of option at t = T is 0. (Option is out
of the money).

Let C(t) be the value of the option. We have just shown

(
S − K if S > K
C(T ) =
0 if S ≤ K

Similarly, one can show the value of a put option at the strike time is
(
0 if S > K
P (T ) =
K − S if S ≤ K

Arbitrage
Arbitrage exists whenever financial instruments are mispriced relative to each other.
Example
Bank A charges 5% interest per year on a loan
Bank B gives 6% interest per year on a savings account
A person can take one year loan for $1000 from Bank A and put it in a savings
account at Bank B. At the end of the year they have $1060 in account and have
to repay only $1050 thus making $10 profit (without investing any of their own
money).
Assumption: Whenever arbitrage exists prices will quickly change to eliminate it.
Thus, any model for the price of an option should be set up so there is no
arbitrage. Potential for arbitrage is usually measured by comparison with a “risk
free” investment (e.g., Government bonds) continuously compounded at interest
rate r. The following (overly simple) example illustrates this.
Example 6.1:
The price of a share of stock is currently $S. At time T the price will be $200
with probability p and $50 with probability (1 − p). An investor can purchase a
European call option with price C. The strike time of the option is T and strike
price is $150. How should C be chosen?
Simple approach: compare two investments

1. Buy option for $C. At time T

- with probability p : stock price is $200 > $150. The value of the option is
thus $200 - $150 = $50.

124
 - with probability 1 − p: stock price is $50 < $150. The value of the option
is thus $0.
Expected value of option at time T is: 50p + 0(1 − p) = 50p.

2. Put $C in a risk-free investment continuously compounded at rate r. Value

of investment at time T : CerT .

In order for there to be no arbitrage we should have

CerT = 50p

⇒ C = 50pe−rT

Note: Price should really depend on the initial price of stock. More detailed
analysis shows that p, r are related to the initial price.

125
6.2 Pricing Options

6.2.1 Stochastic Price Model

Let F be the value of an option on some asset. We will assume that F depends on

• S – the current price of the asset

• t – time

We will work toward formulating a PDE for F (S, t).

To begin we need to discuss how the price of the asset varies in time. A generally
accepted model is as follows:
! !
deterministic random change due
Change in price = +
growth/decay to external factors

Let S(t) be the price at time t and S(t + ∆t) be the price a small time ∆t later.
Then the change in S is ∆S = S(t + ∆t) − S(t), and we can represent the model
above as:
∆S = fdet (S, t, ∆t) + frand (S, t, ∆W (t)),
1
where ∆W (t) is a continuous-time stochastic process. This means that S is
also a stochastic process.
We will focus on a simple model of the following form

∆S = µS∆t + σS∆W (t) (6.1)

µ and σ are constants. µ is called the growth rate or drift and σ is called
the volatility. In addition, we will assume that ∆W (t) is a normally distributed
random variable with mean 0 and variance ∆t. In this case, ∆W (t) is called a
√
Wiener process, and can be written ∆W (t) = φ ∆t where φ is a normally
distributed random variable with mean 0 and variance 1.
Exercise 6.2:
√
Let ∆W (t) = φ ∆t, where φ is a normally distributed random variable with mean
0 and variance 1. Verify the following.
1
A stochastic process can be thought of as a random variable with a time dependent probability
density function.

126
 1. ∆W (t) is a normally distributed random variable with mean 0 and variance
∆t.

2. (∆W (t))2 is a normally distributed random variable with mean ∆t and vari-
ance 2(∆t)2 .

In the limit ∆t → 0 equation (6.1) becomes a stochastic differential equa-

tion for the price S
dS = µS dt + σS dW (t) (6.2)
If dW (t) is a Wiener process (with variance dt) then we say that the stochastic
process S described by (6.2) follows a Geometric Brownian motion. Some
numerical solutions of this DE are shown in Figure 6.2.1. For our purposes, it is
not necessary to describe the solutions of this DE analytically, however, note that
(6.2) has the solution S(t) = 0, t ∈ IR. Also, if at some time the price becomes
becomes 0 then it will be 0 for all subsequent times, i.e., S(t) has the following
property:
(P1) If S(t∗ ) = 0 then S(t) = 0 for all t ≥ t∗ .

6.2.2 Deriving the Black-Scholes Equation

If the price of an asset, S(t), satisfies the stochastic DE (6.2) what is a model for
the value of an option on that asset, F (S, t)? Consider the change in F resulting
from a change ∆t in t (and a corresponding change ∆S in S):

∆F = F (S + ∆S, t + ∆t) − F (S, t)

Applying the Taylor expansion (assuming F is sufficiently differentiable)

∂F ∂F 1 ∂ 2F 2 ∂ 2F 1 ∂ 2F
∆F = ∆S + ∆t + (∆S) + (∆S)(∆t) + (∆t)2 + · · ·
∂S ∂t 2 ∂S 2 ∂S∂t 2 ∂t2
(Use (6.1))
∂F ∂F 1 ∂ 2F
= [µS∆t + σS∆W (t)] + ∆t + 2
[µS∆t + σS∆W (t)]2
∂S ∂t 2 ∂S
∂ 2F 1 ∂ 2F
+ [µS∆t + σS∆W (t)](∆t) + (∆t)2 + · · ·
∂S∂t 2 ∂t2
1
= σSFS ∆W (t) + [Ft + µSFS ]∆t + σ 2 S 2 FSS (∆W (t))2
2
1 1
+[µσS 2 FSS + σSFSt ]∆t∆W (t) + [ Ftt + µSFSt + µ2 S 2 FSS ](∆t)2 + · · ·
2 2

127
 S (dollars)
20

15

10

0
0 2 4 6 8 10
t (years)

(a)

S (dollars)
20

15

10

0
0 2 4 6 8 10
t (years)

(b)

S (dollars)
20

15

10

0
0 2 4 6 8 10
t (years)

(c)

Figure 6.1: Typical solutions of the stochastic DE (6.2) with initial condition S(0) =
10. (a) µ = −0.05, σ = 0.1; (b) µ = 0.05, σ = 0.1; (c) µ = 0.05, σ = 0.2.

128
 √
Now we will use the assumption that ∆W (t) = φ ∆t. Recall that this implies
(∆W (t))2 has mean ∆t and variance 2(∆t)2 . Thus we will approximate (∆W (t))2
by ∆t. Retaining only the largest terms in ∆t then gives
1
∆F ≈ σSFS ∆W (t) + (Ft + µSFS + σ 2 S 2 FSS )∆t (6.3)
2
Letting ∆t → 0 gives a stochastic DE for F
1
dF = σSFS dW (t) + (Ft + µSFS + σ 2 S 2 FSS )dt (6.4)
2
Note that this is in the same form as (6.2) but the coefficients are dependent on
S, F and the derivatives of F . It is much more difficult to solve (even numerically)
than (6.2).
The idea of Black and Scholes was to eliminate the stochastic term from the
DE for F in the following way. Consider a portfolio created by selling the option
and buying  units of the underlying asset. The value of this portfolio at time t is
Π = F − S. The change in this value over a small time change ∆t, assuming  is
constant over the time ∆t, is

∆Π = ∆F − ∆S
(Use (6.1) and (6.3))
1
= [σSFS − σS]∆W (t) + [Ft + µSFS + σ 2 S 2 FSS − µS]∆t.
2
Next comes the key step. Setting  = FS , eliminates the stochastic term (and part
of the deterministic term), leaving:
1
∆Π = [Ft + σ 2 S 2 FSS ]∆t.
2
Then as ∆t → 0 this gives
1
dΠ = [Ft + σ 2 S 2 FSS ]dt,
2
which is equivalent to deterministic differential equation
dΠ 1
= Ft + σ 2 S 2 FSS . (6.5)
dt 2

Now consider an alternate portfolio where the same amount of money is put in
a risk free investment with interest rate r. Assuming continuous compounding, the
risk free investment satisfies the DE
dΠ
= rΠ. (6.6)
dt
129
 Assuming that no arbitrage exists and there are no transaction fees, the returns
of the two portfolios should be equal for all time. To see why, suppose that, at
some time, the return from the first portfolio is more than that from the second.
Then a person could make a risk free profit by taking a loan at interest rate r and
using this to buy the first portfolio. Then the amount owed on the loan would
be less than the return on the investment. Similarly, if the return from the first
portfolio is less than the second, the someone could make risk free profit by selling
the portfolio and investing the money at an interest rate r.
Thus the right hand sides of (6.5) and (6.6) should be equal, i.e.,
1
rΠ = Ft + σ 2 S 2 FSS .
2
Setting Π = F − S and  = FS , gives the Black-Scholes PDE for F
1 2 2
σ S FSS + rSFS + Ft − rF = 0 (6.7)
2
where

r = interest rate of continuously compounded risk-free investment,

σ = volatility of the price.

Since the price must be nonnegative we will be interested in solutions of (6.7) on

0 ≤ t ≤ T and 0 ≤ S < ∞, where T is the strike time.
Exercise 6.3:
Verify that (6.7) is a parabolic partial differential equation. Explain why it is not
a constant coefficient partial differential equation.

We will now formulate boundary conditions for this PDE. The boundary condi-
tions determine what kind of option it is. We will consider a European call option.
One condition comes from the value of the option at the strike time. Recall that
K is the strike price. At t = T there are two possibilities

• S > K. In this case the option is exercised and the price at the strike time is
S − K, i.e., F (S, T ) = S − K.

• S ≤ K. In this case the option expires unexercised since the option is worth-
less at the strike time, i.e., F (S, T ) = 0.

This gives the following condition (sometimes called the final condition)

F (S, T ) = max{S − K, 0} = [S − K]+ . (6.8)

130
 A second condition comes from property (P1) above. If S = 0 at some time
t ≤ T then S = 0 < K at the strike time T and the option won’t be exercised,
i.e., F (0, T ) = 0. Thus, as soon as S = 0 we know option won’t be exercised so it
becomes worthless. This can be expressed as the boundary condition

F (0, t) = 0 (6.9)

A third condition comes from considering the limit of large values for the price.
If S → ∞ at some t0 < T , then S > K for all t > t0 . In particular S > K at the
strike time, T , thus F (S, T ) = S − K. Thus, soon as S → ∞ we know option will
be exercised and will have value S − K. This can be expressed as

F (S, t) = S − K as S → ∞.

Since S → ∞, S is much larger than K so S − K ≈ S. Thus we write the last

boundary condition as
F (S, t) = S as S → ∞. (6.10)

6.2.3 Solution of the Black Scholes Equation

We will solve the Black-Scholes PDE by transforming it into the heat equation and
using the solution for the heat equation derived in the last chapter.
Step I. Change of independent and dependent variables
Consider the change of variables S = Kex , t = T − σ2τ2 . It is easy to check that this
S 2
transformation is invertible and has inverse transformation x = ln K , τ = σ2 (T −t).
Let F (S, t) = Kv(x, τ ) and calculate the partial derivatives
∂ ∂v dτ σ2K
Ft = (Kv) = K =− vτ ,
∂t ∂τ dt 2
∂ ∂v dx K
FS = (Kv) = K = vx ,
∂S ∂x
 dS  S  
∂ ∂ K K K dx K
FSS = (FS ) = vx = − 2 vx + vxx = 2 (vxx − vx ).
∂S ∂S S S S dS S
So the PDE (6.7) becomes
σ2K σ2K
(vxx − vx ) + rKvx − vτ − rKv = 0
2 2
⇓
2r 2r
vxx + ( 2 − 1)vx − vτ − 2 v = 0
σ σ
or
vxx + (δ − 1)vx − vτ − δv = 0 (6.11)

131
where δ = σ2r2 > 0. Note that this is a constant coefficient parabolic partial differ-
ential equation.
Note also that we have the following correspondence of limits:
σ2T
t→0 ⇔ τ →
2
t→T ⇔ τ →0
S → 0 ⇔ x → −∞
S→∞ ⇔ x→∞
σ2 T
Thus we will consider the solution of (6.11) on 0 ≤ τ ≤ 2
and −∞ < x < ∞.
The final condition, (6.8), becomes an initial condition:

Kv(x, 0) = [Kex − K]+

⇓
v(x, 0) = [ex − 1]+ . (6.12)

The boundary condition at zero, (6.9), becomes

lim v(x, τ ) = 0. (6.13)

x→−∞

Finally, the boundary condition at infinity, (6.10), becomes

v(x, τ ) = ex as x → ∞. (6.14)

Step II. Change of dependent variable.

Using a result you will derive on your home work, the change of dependent variable
2
v(x, τ ) = eαx+βτ u(x, τ ), with α = (1−δ)
2
and β = − (δ+1)
4
gives the following PDE
for u(x, τ ):

uxx − uτ = 0. (6.15)
Note that this is the heat equation (5.29) with γ = 1. With this change of the
dependent variable, initial condition (6.12) becomes

eαx u(x, 0) = [ex − 1]+ ⇒ u(x, 0) = [e(1−α)x − e−αx ]+

(δ+1) (δ−1)
x x
⇒ u(x, 0) = [e 2 −e 2 ]+ . (6.16)
The boundary condition (6.13) becomes

lim eαx+βτ u(x, τ ) = 0. (6.17)

x→−∞

132
The boundary condition (6.14) becomes

eαx+βτ u(x, τ ) = ex , as x → ∞.

Thus we have reduced the problem of solving the Black-Scholes PDE (6.7) on
0 ≤ t ≤ T and 0 ≤ S < ∞ subject to the final condition (6.8) to the problem of
2
solving (6.15) on 0 ≤ τ ≤ σT and −∞ < x < ∞ subject to the initial condition
(6.16).
Step III. Using the solution of the heat equation
Recall (see the end of Chapter 5) that the heat equation BVP

γuxx − uτ = 0, −∞ < x < ∞

u(x, 0) = f (x)

may be solved using the Fourier Transform, and that the solution is given by
Z ∞
1 √ 2
u(x, τ ) = √ f (x + 2 γτ y)e−y dy.
π −∞

Here we have γ = 1 and

(δ+1) (δ−1)
x x
f (x) = [e 2 −e 2 ]+ ,

so the solution is
Z ∞ √ √
1 (δ+1) (δ−1) 2
u(x, τ ) = √ [e 2
(x+2 τ y)
−e 2
(x+2 τ y)
]+ e−y dy.
π −∞

To find an explicit expression for the solution, we need to evaluate the integral.
To begin, note that
(δ+1) √
(x+2 τ y)
(δ−1) √
(x+2 τ y)
√
e 2 −e 2 >0 ⇔ x + 2 τ y > 0.

Thus the lower limit of the integral may be replaced by − 2√x τ and the integral may
be split into two parts:
(δ+1) (δ−1)
x Z ∞ √ x Z ∞ √
e 2
(δ+1) τ y−y 2 e 2
τ y−y 2
u(x, τ ) = √ e dy − √ e(δ−1) dy.
π x
− 2√ π x
− 2√
τ τ

Completing the squares in the exponents of each integrand, we have

(δ+1) (δ+1) 2 (δ−1) (δ−1) 2
x+ 4 τ Z ∞ x+ 4 τ Z ∞
e 2 (δ+1) √ 2 e 2 (δ−1) √ 2
−(y− 2
u(x, τ ) = √ e τ)
dy− √ e−(y− 2
τ)
dy.
π x
− 2√ π x
− 2√
τ τ

133
 √ √
Now make the change of variables w1 = − 2(y − (δ+1) τ ) in the first integral and
√ (δ−1) √
2
w2 = − 2(y − 2 τ ) in the second. Then we have
(δ+1) (δ+1) 2 (δ+1) √ (δ−1) (δ−1) 2 (δ−1) √
x+ 4 τ Z √x + 2τ x+ 4 τ Z √x + 2τ
e 2 2τ 2
−w12 /2 e 2 2τ 2 2
u(x, τ ) = √ e dw1 − √ e−w2 /2 dw2 .
2π −∞ 2π −∞

Recall that Z z
1 2
Φ(z) = √ e−w /2 dw
2π −∞
is the cumulative distribution for a normal random variable with mean zero and
standard deviation 1.
Thus we can write
(δ + 1) √ (δ − 1) √
   
(δ+1) (δ+1) 2
x+ 4 τ x (δ−1) (δ−1) 2
x+ 4 τ x
u(x, τ ) = e 2 Φ √ + 2τ −e 2 Φ √ + 2τ .
2τ 2 2τ 2
(1−δ) (δ+1) 2
S
Returning to the original variables, F (S, t) = Ke 2 x− 4 τ u(x, τ ), x = ln K ,
2
δ = σ2r2 , τ = σ2 (T − t), gives the solution of the Black-Scholes equation for a
European call option:

ln(S/K) + (r + σ 2 /2)(T − t) ln(S/K) + (r + σ 2 /2)(T − t) √

   
−r(T −t)
F (S, t) = SΦ √ −Ke Φ √ −σ T −t .
σ T −t σ T −t

Assumptions explicitly or implicitly made in this derivation:

1. No arbitrage exists.

2. Price obeys stochastic DE (6.2) with dW (t) a normal random variable with
mean 0 and variance dt.

3. There are no transaction costs.

4. The asset pays no dividends.

5. Trading can take place continuously.

6. Short selling is permitted, i.e. we can sell assets we don’t own.

7. Assets are divisible, i.e. we do not have to buy or sell integer numbers of
assets.

134
References

[1] W. E. Boyce and R. C. DiPrima. Elementary Differential Equations. Wiley,

New York, 9th edition, 2009.

[2] C. H. Edwards and D. E. Penney. Differential Equations and Boundary Value

Problems. Prentice Hall, Upper Saddle River, NJ, 2004.

[3] E. A. Coddington. An Introduction to Ordinary Differential Equations.

Prentice-Hall, Englewood Cliffs, NJ, 1961.

[4] E. A. Coddington. An Introduction to Ordinary Differential Equations. Dover,

New York, NY, 1989.

[5] P.V. O’Neil. Beginning Partial Differential Equations. Wiley, Hoboken, NJ,
2008.

135