Introduction to Ordinary and Partial Differential Equations

Wen
c Shen, Spring 2015. All rights reserved
Contents

1 Introduction 4
1.1 Classification of Differential Equations . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Directional Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 First Order Differential Equations 11

2.1 Linear equations; Method of integrating factors . . . . . . . . . . . . . . . . . . 11
2.2 Separable Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3 Differences between linear and nonlinear equations; existence and uniqueness of solutions 20
2.4 Modeling with first order equations . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.5 Autonomous equations and population dynamics . . . . . . . . . . . . . . . . . 30
2.6 Exact Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.7 Euler’s method* . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

3 Second Order Linear Equations 42

3.1 Homogeneous equations with constant coefficients . . . . . . . . . . . . . . . . . 42
3.2 Solutions of Linear Homogeneous Equations; the Wronskian . . . . . . . . . . . 46
3.3 Complex Roots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.4 Repeated roots; reduction of order . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.5 Non-homogeneous equations; method of undetermined coefficients . . . . . . . . 60
3.6 Variation of Parameters* . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.7 Mechanical Vibrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.8 Forced Vibrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

4 Higher Order Linear Equations 78

4.1 General Theory of n-th Order Linear Equations . . . . . . . . . . . . . . . . . . 78
4.2 Homogeneous Equations with Constant Coefficients. . . . . . . . . . . . . . . . 80
4.3 Higher Order Linear Non-homogeneous Differential Equations . . . . . . . . . . 83

6 The Laplace Transform 86

6.1 Definition of the Laplace transform . . . . . . . . . . . . . . . . . . . . . . . . . 86
6.2 Solution of initial value problems . . . . . . . . . . . . . . . . . . . . . . . . . . 88
6.3 Step functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
6.4 Differential equations with discontinuous forcing functions . . . . . . . . . . . . 103
6.5 Impulse functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6.6 Convolution* . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

1
7 Systems of Two Linear Differential Equations 114
7.1 Introduction to systems of differential equations . . . . . . . . . . . . . . . . . . 114
7.2 Review of matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
7.3 Eigenvalues and eigenvectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
7.4 Basic theory of systems of first order linear equation . . . . . . . . . . . . . . . 119
7.5 Homogeneous systems of two equations with constant coefficients. . . . . . . . . 119
7.6 Complex eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
7.7 Fundamental Matrices* . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
7.8 Repeated eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
7.9 Summary of Stabilities and types of critical points for linear systems . . . . . . 140

10 Fourier Series 152

10.1 Introduction and Basic Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . 152
10.2 Even and Odd Functions; Fourier sine and Fourier cosine series. . . . . . . . . . 161
10.3 Properties of Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
10.4 Two-Point Boundary Value Problems; Eigenvalue Problems . . . . . . . . . . . 168

11 Partial Differential Equations 176

11.1 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
11.2 Heat Equation in 1D; Solution by Separation of Variable and Fourier series . . 177
11.3 Solutions of Wave Equation by Fourier Series . . . . . . . . . . . . . . . . . . . 184
11.4 Laplace Equation in 2D (probably skip) . . . . . . . . . . . . . . . . . . . . . . 187
11.5 D’Alembert’s Solution of Wave Equation . . . . . . . . . . . . . . . . . . . . . . 190
11.6 Method of Characteristics; Classification of 2nd order linear PDEs. . . . . . . . 193

12 Homeworks 196
12.1 Homework 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
12.2 Homework 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
12.3 Homework 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
12.4 Homework 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
12.5 Homework 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
12.6 Homework 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
12.7 Homework 7. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
12.8 Homework 8. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
12.9 Homework 9. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
12.10Homework 10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
12.11Homework 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
12.12Homework 12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
12.13Homework 13 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
12.14Homework 14 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220

13 Answers to Homework Problems 221

13.1 Answer/keys for homework 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
13.2 Answer/keys to homework 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
13.3 Answer/keys to homework 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
13.4 Answer/keys to homework 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
13.5 Answer/keys to homework 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225

2
13.6 Answer/keys to homework 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
13.7 Answer/keys to homework 7. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
13.8 Answer/keys to homework 8. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
13.9 Answer/keys to homework 9. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
13.10Answer/keys to homework 10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
13.11Answer/keys to homework 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
13.12Answer/keys to homework 12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
13.13Answer/keys to homework 13 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233

3
Chapter 1

Introduction

1.1 Classification of Differential Equations

Definition: A differential equation is an equation which contains derivatives of the unknown.

(Usually it is a mathematical model of some physical phenomenon.)
Two classes of differential equations:
• O.D.E. (ordinary differential equations): linear and non-linear;
• P.D.E. (partial differential equations). (not covered in math250, but in math251)
Some concepts related to differential equations:
• system: a collection of several equations with several unknowns.
• order of the equation: the highest order of derivatives.
• linear or non-linear equations: Let y(t) be the unknown. Then,
a0 (t)y (n) + a1 (t)y (n−1) + · · · + an (t)y = g(t), (∗)
is a linear equations. If the equation can not be written as (∗), then it’s non-linear.
Two things you must know: identify the linearity and the order of an equation.

Example 1. Let y(t) be the unknown. Identify the order and linearity of the following
equations.
(a). (y + t)y ′ + y = 1,
(b). 3y ′ + (t + 4)y = t2 + y ′′ ,
(c). y ′′′ = cos(2ty),
√
(d). y (4) + ty ′′′ + cos t = ey .

Answer.
Problem order linear?
(a). (y + t)y ′ + y = 1 1 No
(b). 3y ′ + (t + 4)y = t2 + y ′′ 2 Yes
(c). y ′′′ = cos(2ty) 3 No
√
(d). y (4) + ty ′′′ + cos t = ey 4 No

4
 What is a solution? A solution is a function that satisfies the equation, the boundary
conditions (if any), the initial conditions (if any), and whose derivatives exist.

Example 2. Verify that y(t) = eat is a solution of the IVP (initial value problem)

y ′ = ay, y(0) = 1.

Here y(0) = 1 is called the initial condition.

Answer. Let’s check if y(t) satisfies the equation and the initial condition:

y ′ = aeat = ay, y(0) = e0 = 1.

They are both OK. So it is a solution.

Example 3. Verify that y(t) = 10 − ce−t with c a constant, is a solution to y ′ + y = 10.

Answer.

y ′ = −(−ce−t ) = ce−t , y ′ + y = ce−t + 10 − ce−t = 10. OK.

Let’s try to solve one equation.

Example 4. Consider the equation

(t + 1)y ′ = t2

We can rewrite it as (for t 6= −1)

t2 t2 − 1 + 1 (t + 1)(t − 1) + 1 1
y′ = = = = (t − 1) +
t+1 t+1 t+1 t+1
To find y, we need to integrate y ′ :

t2
Z  
1
Z
y = y ′ (t)dt = (t − 1) + dt = − t + ln |t + 1| + c
t+1 2

where c is an integration constant which is arbitrary. This means there are infinitely many
solutions.
Additional condition: initial condition y(0) = 1. (meaning: y = 1 when t = 0) Then

t2
y(0) = 0 + ln |1| + c = c = 1, so y(t) = − t + ln |t + 1| + 1.
2
So for equation like y ′ = f (t), we can solve it by integration: y = f (t)dt.
R

5
 Review on integration:
1
Z
xn dx = xn+1 + c, (n 6= −1)
n+1
1
Z
dx = ln |x| + c
x
Z
sin x dx = − cos x + c
Z
cos x dx = sin x + c
Z
ex dx = ex + c
ax
Z
ax dx = +c
ln a
Integration by parts:
Z Z Z b b Z b
′
u dv = uv − v du, u(x)v (x) dx = u(x)v(x) − v(x)u′ (x) dx

a a a

Chain rule:
d
(f (g(t)) = f ′ (g(t)) · g ′ (t)
dt

1.2 Directional Fields

Directional field: for first order equations y ′ = f (t, y).
Interpret y ′ as the slope of the tangent to the solution y(t) at point (t, y) in the y − t plane.

• If y ′ = 0, the tangent line is horizontal;

• If y ′ > 0, the tangent line goes up;

• If y ′ < 0, the tangent line goes down;

• The value of |y ′ | determines the steepness.

1
Example 5. Consider the equation y ′ = (3 − y). We know the following:
2
• If y = 3, then y ′ = 0, flat slope,

• If y > 3, then y ′ < 0, down slope,

• If y < 3, then y ′ > 0, up slope.

6
See the directional field below (with some solutions sketched in red):

4.5

3.5

2.5

1.5

0.5

0
0 1 2 3 4 5 6

We note that, if y(0) = 3, then y(t) = 3 is the solution.

Asymptotic behavior: As t → ∞, we have y → 3.

Remarks:

(1). For equation y ′ (t) = a(b − y) with a > 0, it will have similar behavior to Example 5,
where b = 3 and a = 12 . Solution will approach y = b as t → +∞.

(2). Now consider y ′ (t) = a(b − y), but with a < 0. This changes the sign of y ′ . We now have

– If y(0) = b, then y(t) = b;

– If y(0) > b, then y → +∞ as t → +∞;
– If y(0) < b, then y → −∞ as t → +∞.

Example 6: Let y ′ (t) = (y − 1)(y − 5). Then,

• If y = 1 or y = 5, then y ′ = 0.

• If y < 1, then y ′ > 0;

• If 1 < y < 5, then y ′ < 0;

• If y > 5, then y ′ < 0.

Directional field looks like:

7
 8

−1
−0.5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

What can we say about the solutions?

• If y(0) = 1, then y(t) = 1;

• If y(0) = 5, then y(t) = 5;

• If y(0) < 1, then y → 1 as t → +∞;

• If 1 < y(0) < 5, then y → 1 as t → +∞;

• If y(0) > 5, then y → +∞ as t → +∞.

Remark: If we have y ′ (t) = f (y), and for some y0 we have f (y0 ) = 0, then, y(t) = y0 is a
solution.

Example 7: Given the plot of a directional field,

8
 6

−1

−2
−0.5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

which of the following ODE could have generated it?

(a). y ′ (t) = (y − 2)(y − 4)

(b). y ′ (t) = (y − 1)2 (y − 3)

(c). y ′ (t) = (y − 1)(y − 3)2

(d). y ′ (t) = −(y − 1)(y − 3)2

We first check the constant solution, y = 1 and y = 3. Then (a) can not be. Then, we
check the sign of y ′ on the intervals: y < 1, 1 < y < 3, and y > 3, to match the directional
field. We found that (c) could be the equation.

Example 8. Consider a more complicated situation where y ′ depends on both t and y.

Consider y ′ = t + y. To generate the directional field, we see that:

• We have y ′ = 0 when y = −t,

• We have y ′ > 0 when y > −t,

• We have y ′ < 0 when y < −t.

One can sketch the directional field along lines of y = −t + c for various values of c.

• If y = −t , then y ′ = 0;

• If y = −t − 1 , then y ′ = −1;

• If y = −t − 2 , then y ′ = −2;

• If y = −t + 1 , then y ′ = 1;

• If y = −t + 2 , then y ′ = 2;

9
 Below is the graph of the directional field, with some solutions plotted in red.
2

−1

−2

−3

−4

−5
−1 0 1 2 3 4 5

What can we say about the solutions?

The solution depends on the initial condition y(0).

• We see first that if y(0) = −1, the solution is y(t) = −t − 1;

• If y(0) > −1, then y → +∞ as t → +∞;

• If y(0) < −1, then y → ∓∞ as t → +∞.

We can also discuss the asymptotic behavior as t → −∞:

• If y(0) > −1, then y → +∞ as t → −∞.

• If y(0) < −1, then y → +∞ as t → −∞.

We see that, for general y ′ = f (t, y) where the function f depends on t, the directional
fields could be rather tedious to generate by hand. There are computer softwares that will
do this for you, as I did with these plots. They are generated in Matlab, a powerful software
that can solve many math problems.

10
Chapter 2

First Order Differential Equations

We consider the equation

dy
= f (t, y)
dt
Overview:

• Two special types of equations: linear, and separable;

• Linear vs. nonlinear;

• modeling;

• autonomous equations and population model;

• a little bit on Euler’s method.

2.1 Linear equations; Method of integrating factors

The function f (t, y) is a linear function in y, i.e, we can write

f (t, y) = −p(t)y + g(t).

So we will study the equation

y ′ + p(t)y = g(t). (A)
We introduce the method of integrating factors (due to Leibniz): We multiply equation
(A) by a function µ(t) on both sides

µ(t)y ′ + µ(t)p(t)y = µ(t)g(t)

The function µ is chosen such that the equation is integrable, meaning the LHS (Left Hand
Side) is the derivative of something. In particular, we require:

µ(t)y ′ + µ(t)p(t)y = (µ(t)y)′ , ⇒ µ(t)y ′ + µ(t)p(t)y = µ(t)y ′ + µ′ (t)y

11
which requires
dµ dµ
µ′ (t) = = µ(t)p(t), ⇒ = p(t) dt
dt µ
Integrating both sides Z
ln µ(t) = p(t) dt

which gives a formula to compute µ

Z 
µ(t) = exp p(t) dy .

Therefore, this µ is called the integrating factor.

Note that µ is not unique. In fact, adding an integration constant, we will get a different
µ. But we don’t need to be bothered, since any such µ will work. We can simply choose one
that is convenient.
Putting back into equation (A), we get

d
Z
(µ(t)y) = µ(t)g(t), µ(t)y = µ(t)g(t) dt + c
dt
which gives the formula for the solution
Z  Z 
1
y(t) = µ(t)g(t) dt + c , where µ(t) = exp p(t) dt .
µ(t)

Example 1. Solve y ′ + ay = b (a 6= 0).

Answer. We have p(t) = a and g(t) = b. So

Z 
µ = exp a dt = eat

so  
b at b
Z
−at at −at
y=e e b dt = e e + c = + ce−at ,
a a
where c is an arbitrary constant. Pay attention to where one adds this integration constant!

Example 2. Find the general solution of y ′ + y = e2t .

Answer. We have p(t) = 1 and g(t) = e2t . So

Z 
µ(t) = exp 1 dt = et

and  
1 3t 1
Z Z
y(t) = e−t et e2t dt = et e3t dt = e−t e + c = e2t + ce−t .
3 3
Can you discuss the behavior of this solution, as t → ±∞?

12
 Example 3. Solve the IVP

−2
(1 + t2 )y ′ + 4ty = 1 + t2 , y(0) = 1.

Answer. First, let’s rewrite the equation into the normal form
4t
−3
y′ + 2
y = 1 + t2 ,
1+t
so
4t
−3
p(t) = , g(t) = 1 + t2 .
1 + t2
Then
Z  Z  Z 
4t 2 2
µ(t) = exp p(t) dt = exp dt = exp d(t )
1 + t2 1 + t2

2
= exp(2 ln(1 + t2 )) = exp(ln(1 + t2 )2 ) = 1 + t2 .

Then
(1 + t2 )2 (1 + t2 )−3 dt (1 + t2 )−1 dt
R R
arctan t + c
y= = = .
(1 + t2 )2 (1 + t2 )2 (1 + t2 )2
By the IC: y(0) = 1:
0+c arctan t + 1
y(0) = = c = 1, ⇒ y(t) = .
1 (1 + t2 )2

Example 4. Find the general solution for ty ′ − y = t2 e−t , (t > 0).

Answer. Rewrite it into normal form

1
y ′ − y = te−t
t
so
p(t) = −1/t, g(t) = te−t .
We have Z 
1
µ(t) = exp (−1/t)dt = exp (− ln t) =
t
and
1 −t
Z Z
e−t dt = t −e−t + c = −te−t + ct.


y(t) = t te dt = t
t

Example 5. Solve y − 13 y = e−t , with y(0) = a, and discuss the behavior of y as t → ∞,

as one chooses different initial values for y(0) = a.

Answer. Let’s solve it first. We have

1
µ(t) = e− 3 t

13
so  
3 −4t
Z Z
1
t − 13 t −t 1
t − 34 t 1
t
y(t) = e 3 e e dt = e 3 e dt = e 3 − e 3 +c .
4
Plug in the IC to find c
3 3
y(0) = e0 (− + c) = a, c = a +
4 4
so    
1
t 3 −4t 3 3 −t 3 t/3
y(t) = e 3 − e 3 +a+ =− e + a+ e .
4 4 4 4
To see the behavior of the solution, we see that it contains two terms. The first term e−t
goes to 0 as t grows. The second term et/3 goes to ∞ as t grows, but the constant a + 34 is
multiplied on it. So we have
3
• If a + 4 = 0, i.e., if a = − 43 , we have y → 0 as t → ∞;
3
• If a + 4 > 0, i.e., if a > − 43 , we have y → ∞ as t → ∞;
3
• If a + 4 < 0, i.e., if a < − 43 , we have y → −∞ as t → ∞;

On the other hand, as t → −∞, the term e−t will blow up to −∞, and will dominate.
Therefore, y → −∞ as t → −∞ for any values of a.
See plot below:

Example 6. Solve ty ′ + 2y = 4t2 , y(1) = 2.

Answer. Rewrite the equation first

2
y ′ + y = 4t, (t 6= 0)
t

14
So p(t) = 2/t and g(t) = 4t. We have
Z 
µ(t) = exp 2/t dt = exp(2 ln t) = t2

and Z
−2
4t · t2 dy = t−2 t4 + c = t2 + ct−2 .


y(t) = t

We see that the solution has 2 terms, t2 and t−2 . With different initial value, we will get
different values of c, and the solution will be very different.
By our IC y(1) = 2, we get

y(1) = 1 + c = 2, c=1

we get the solution:

1
y(t) = t2 +
, t > 0.
t2
Note the condition t > 0 comes from the fact that the initial condition is given at t = 1, and
we require t 6= 0.
In the graph below we plot several solutions in the t − y plan, depending on initial data.
The one for our solution is plotted with a dashed line where the initial point is marked with
a ‘×’.

10

−5
−4 −3 −2 −1 0 1 2 3 4

15
Bernoulli Equations*. This is an example of solving nonlinear first order ODE by a vari-
able change and turn it into a linear equation. Consider the Bernoulli differential equa-
tions
y ′ (t) + p(t)y = q(t)y n
where n is an integer.
If n = 0 or n = 1, this equation is linear and we can solve it by the method integrating
factor. Otherwise, the equation is non-linear. Such equations arises in applications such as
population models and models of one-dimensional motion influenced by drag forces.
We now make a variable change, and call v(t) = y(t)m , for some m to be determined.
Then, y = v 1/m . By the chain rule, we have
1 1/m−1 ′
y′ = v v.
m
Put these into the differential equation, we get
1 1/m−1 ′
v v + p(t)v 1/m = q(t)v n/m .
m
Multiply both sides by mv 1−1/m , we get

v ′ + mp(t)v = mq(t)v (m+n−1)/m .

This may look more complicated than the original equation, but remember we can choose the
value m in a smart way. Now let m = 1 − n, we get

v ′ + (1 − n)p(t)v = (1 − n)q(t),

which is linear, and we can solve it! Once we get v(t), we can then easily recover y(t).

Example 7. Solve the IVP

y ′ + y = ty 3 , y(0) = 2.

Answer. We make the variable change and let v = y 1−n = y 1−3 = y −2 . By the derivation
we did, we have
v ′ − 2v = −2t, v(0) = y(0)−2 = 1/4.
The general solution is
1
v(t) = Ce2t + (t + ).
2
Imposing the initial condition, we get
1 1 1
C=− , v(t) = − e2t + (t + ).
4 4 2
Finally, we go back to y(t) = v −1/2 , and get

1 −1/2
 
1 2t
y(t) = − e + (t + ) .
4 2

16
 Remarks on the definition of linear and nonlinear equations:
(1) For any linear equation
y ′ + p(t)y = g(t)
if we make a non-linear variable change, say y = f (v) for some nonlinear function f (for
example, f (v) = v 2 ), we turn the linear equation for y into a nonlinear equation for v.
(2) However, the other way around is not always possible. Only for some special nonlinear
equations, a suitable (carefully chosen) variable change could turn it into a linear equation in
the new variable. Therefore, such special equations are worth mentioning.

2.2 Separable Equations

Let y(x) denote the unknown. We study first order equations that can be written as
dy M (x)
= f (x, y) =
dx N (y)
where M (x) and N (y) are suitable functions of x and y only. Then we have
Z Z
N (y) dy = M (x) dx, ⇒ N (y) dy = M (x) dx

and we get implicitly defined solutions of y(x).

Example 1. Consider
dy sin x
= .
dx 1 − y2
We can separate the variables:
1
Z Z
(1 − y 2 ) dy = sin x dx, ⇒ y − y 3 = − cos x + c.
3
If one has IC as y(π) = 2, then
1 3 5
2− · 2 = − cos π + c, ⇒ c=− ,
3 3
so the solution y(x) is implicitly given as
1 5
y − y 3 + cos x + = 0.
3 3

Example 2. Find the solution in explicit form for the equation

dy 3x2 + 4x + 2
= , y(0) = −1.
dx 2(y − 1)
Answer. Separate the variables

Z Z
2(y − 1) dy = (3x2 + 4x + 2) dx , ⇒ (y − 1)2 = x3 + 2x2 + 2x + c

17
Set in the IC y(0) = −1, i.e., y = −1 when x = 0, we get

(−1 − 1)2 = 0 + c, c = 4, (y − 1)2 = x3 + 2x2 + 2x + 4.

In explicitly form, one has two choices:

p
y(t) = 1 ± x3 + 2x2 + 2x + 4.

To determine which sign is the correct one, we check again by the initial condition:
√
y(0) = 1 ± 4 = 1 ± 2, must have y(0) = −1.

We see we must choose the ‘-’ sign. The solution in explicitly form is:
p
y(x) = 1 − x3 + 2x2 + 2x + 4.

On which interval will this solution be defined?

x3 + 2x2 + 2x + 4 ≥ 0, ⇒ x2 (x + 2) + 2(x + 2) ≥ 0

⇒ (x2 + 2)(x + 2) ≥ 0, ⇒ x ≥ −2.

We can also argue that when x = −2, we have y = 1. At this point |dy/dx| → ∞, therefore
solution can not be defined at this point.
The plot of the solution is given below, where the initial data is marked with ‘x’. We also
include the solution with the ‘+’ sign, using dotted line.

10

−2

−4

−6

−8
−3 −2 −1 0 1 2 3

Example 3. Solve y ′ = 3x2 + 3x2 y 2 , y(0) = 0, and find the interval where the solution is
defined.

Answer. Let’s first separate the variables.

dy 1
Z Z
2 2
= 3x (1 + y ), ⇒ dy = 3x2 dx, ⇒ arctan y = x3 + c.
dx 1 + y2

18
Set in the IC:
arctan 0 = 0 + c, ⇒ c=0
we get the solution
arctan y = x3 , ⇒ y = tan(x3 ).
Since the initial data is given at x = 0, i.e., x3 = 0, and tan is defined on the interval (− π2 , π2 ),
we have
π π h π i1/3 h π i1/3
− < x3 < , ⇒ − <x< .
2 2 2 2

Example 4. Solve
1 + 3x2
y′ = , y(0) = 1
3y 2 − 6y
and identify the interval where solution is valid.

Answer. Separate the variables

Z Z
(3y − 6y)dy = (1 + 3x2 )dx
2
y 3 − 3y 2 = x + x3 + c.

Set in the IC: x = 0, y = 1, we get

1 − 3 = c, ⇒ c = −2,

Then,
y 3 − 3y 2 = x3 − x − 2.
Note that solution is given in implicitly form.
To find the valid interval of this solution, we note that y ′ is not defined if 3y 2 − 6y = 0,
i.e., when y = 0 or y = 2. These are the two so-called “bad points” where you can not define
the solution. To find the corresponding values of x, we use the solution expression:

y=0: x3 + x − 2 = 0,

⇒ (x2 + x + 2)(x − 1) = 0, ⇒ x=1

and
y=2: x3 + x − 2 = −4, ⇒ x3 + x + 2 = 0,
⇒ (x2 − x + 2)(x + 1) = 0, ⇒ x = −1
(Note that we used the facts x2 + x + 2 6= 0 and x2 − x + 2 6= 0 for all x.)
Draw the real line and work on it as following:

✛ ✲
× × ✲ x
−2 −1 0 1 2

Therefore the interval is −1 < x < 1.

19
2.3 Differences between linear and nonlinear equations; exis-
tence and uniqueness of solutions
We discuss here some fundamental differences between linear and nonlinear equations regard-
ing existence and uniqueness of solutions.
For a linear equation
y ′ + p(t)y = g(t), y(t̄) = ȳ,
if we require the solution y(t) to be differentiable functions, then we have the following exis-
tence and uniqueness theorem.

Theorem . If p(t) and g(t) are continuous and bounded on an open interval containing
t0 , then it has an unique solution on that interval.
A brief proof. The existence of a solution is obvious, since we can write it out using
the method of integrating factors. For the uniqueness, let y1 and y2 be two solutions of the
problem, i.e.,

y1′ + p(t)y1 = g(t), y1 (t̄) = ȳ, y2′ + p(t)y2 = g(t), y2 (t̄) = ȳ.

Define the error e(t) = y1 − y2 . Then, e(t) solves the equation

e′ (t) + p(t)e(t) = 0, e(t̄) = 0.

From the directional field, we see that e(t) ≡ 0, proving y1 (t) = y2 (t), which implies unique-
ness.

One can use this Theorem to identify intervals where the solution of the IVP could be
defined.

Example 1. Find the largest interval where the solution can be defined for the following
problems.
(A). ty ′ + y = t3 , y(−1) = 3.
(B). ty ′ + y = t3 , y(1) = −3.
(C). (t − 3)y ′ + (ln t)y = 2t, y(1) = 2
(D). y ′ + (tan t)y = sin t, y(π) = 100.

Answer. (A). Rewrite: y ′ + 1t y = t2 , so t 6= 0. Since t0 = −1, the interval is t < 0.

(B). The equation is same as (A), so t 6= 0. t0 = 1, the interval is t > 0.
ln t 2t
(C). Rewrite: y ′ + t−3 y = t−3 , so t 6= 3 and t > 0 for the ln function. Since t0 = 1, the
interval is then 0 < t < 3.
(D). Since t0 = π, and for tan t to be defined we must have t 6= 2k+1 2 π, k = ±1, ±2, · · · .
π 3π
So the interval is 2 < t < 2 .

Remark: The conditions in the Theorem guarantees the uniqueness. If the conditions fail,
the uniqueness might still hold, but not granted. Also, in many cases one will have discontin-
uous functions of p(t) and g(t). One can relax the restriction on the solution, and require y(t)
to be only continuous. Then, the conditions reduces to: p, q are integrable functions.

20
 Example (With discontinuous coefficient functions). Consider
(
1, 0≤t<1
y ′ − y = g(t), y(0) = 0, g(t) =
−2, 1 ≤ t ≤ 2.

Find a solution on the interval 0 ≤ t ≤ 2.

Answer. The function g(t) has a jump at t = 1. We can solve the equation using 2 steps.
(1) For 0 ≤ t < 1, we have the IVP

y ′ − y = 1, y(0) = 0.

Solving this, we obtain

y(t) = et − 1.
At t = 1, we have y(1) = e − 1.
(2) For 1 ≤ t ≤ 2, we have the IVP

y ′ − y = −2, y(1) = e − 1.

Note that the initial condition is given at t = 1, and we use the solution in step 1 and evaluate
it at t = 1.
This can be easily solved, and we get

y(t) = (1 − 3/e)et + 2, 1 ≤ t ≤ 2.

Put them together, we obtain the solution, piecewise defined

(
et − 1, 0≤t<1
y(t) =
y(t) = (1 − 3/e)et + 2, 1 ≤ t ≤ 2.

See graphs of the functions g and y:

We see that y(t) is continuous at t = 1, but the graph has a kink, since y ′ (t) is discontinuous
at t = 1.
In general, ODEs with discontinuous coefficients have their own sets of Theorems.

21
 For non-linear equation
y ′ = f (t, y), y(t̄) = ȳ,
we have a much weaker theorem, for differentiable functions y(t).

Theorem . If f (t, y), ∂f ∂y (t, y) are continuous and bounded on an rectangle (α < t <
β, a < y < b) containing (t̄, ȳ), then there exists an open interval around t̄, contained in (α, β),
where the solution exists and is unique.
We note that the statement of this theorem is not as strong as the one for linear equation.
The proof uses the Picard iteration, and is much more complicated. It is outlined in the
textbook, you may read it if you are curious.
Some remarks*:
(1). If f (t, y), ∂f
∂y (t, y) are continuous at (t0 , y0 ), then in a small neighborhood around the
point (t0 , y0 ), one can linearize f (t, y), and f (t, y) ≈ a+by will be a good approximation. Then
by the Theorem for linear equation, solution exits and is unique. This is a rather standard
technique for nonlinear problems, studying the local linearized problem.
(2). A rough (not rigorous) argument to see how the bound on ∂f ∂y (t, y) helps the unique-
ness: Let y1 , y2 be two solutions, then

(y1 − y2 )′ = f (t, y1 ) − f (t, y2 ), y1 (t̄) − y2 (t̄) = 0.

This implies Z t 
y1 (t) − y2 (t) = f (s, y1 (s)) − f (s, y2 (s)) ds
t̄
∂f
The bound on ∂y implies that there exist a constant M , such that

|f (t, y1 ) − f (t, y2 )| ≤ M |y1 − y2 | .

We now have
Z t Z t
|y1 (t) − y2 (t)| ≤ |f (s, y1 (s)) − f (s, y2 (s))| ds ≤ M |y1 (s) − y2 (s)| ds.
t̄ t̄

Write now E(t) = |y1 (t) − y2 (t)| > 0. We have

Z t
E(t) ≤ M E(t) ds, E(t̄) = 0.
t̄

We now use a comparison argument. Let w(t) be the solution for

Z t
w(t) = M w(t) ds, w(t̄) = E(t̄) = 0.
t̄

By comparison we have E(t) ≤ w(t). For w(t), is satisfies the equation w′ = M w with
w(t̄) = 0, which implies w(t) ≡ 0. Therefore E(t) ≡ 0, leading to uniqueness.

Below we give several counter examples.

Example 1. Loss of uniqueness. Consider

dy t
= f (t, y), f (x, y) = − , y(−2) = 0.
dt y

22
We first note that at y = 0, which is the initial value of y, we have y ′ = f (t, y) → ∞. So the
conditions of the Theorem are not satisfied, and we expect something to go wrong.
Solve the equation as an separable equation, we get
Z Z
y dy = − t dt, y 2 + t2 = c,

and by IC we get c = (−2)2 + 0 = 4, so y 2 + t2 = 4. In the y − t plan, this is the equation for a

circle, centered at the origin, with radius 2. The initial condition is given at t0 = −2,√ y0 = 0,
where the tangent
√ line is vertical (i.e., with infinite slope). We have two solutions: y = 4 − t2
2
and y = − 4 − t . We lose uniqueness of solutions.

Example 2. Loss of uniqueness; another example. Consider the IVP

y ′ = y 1/3 , y(0) = 0.

We see a trivial solution is y(t) ≡ 0. Now consider the following function:

(
(2(t − c)/3)3/2 , t > c
yc (t) =
0, t≤c

where c ≥ 0 is any constant. We can easily check that yc (t) is a solution to the IVP, for any
c ≥ 0. Indeed, we have, for t ≥ c,
3 2
y′ = (2(t − c)/3)1/2 = (2(t − c)/3)1/2 = y 1/3 ,
2 3

and for t < c, y ′ = 0 = y 1/3 . The initial condition is also satisfied.

Therefore, there exist infinitely many solutions! See graphs of these solutions for c =
0, 0.5, 1.

This is no surprise, since f (t, y) = y 1/3 , then fy = 13 y −2/3 is not bounded at y = 0, which
is where the initial condition is given.

Example 3. Blow-up of solution. Consider a simple non-linear equation:

y′ = y2, y(0) = 1.

23
Note that f (t, y) = y 2 , which is defined for all t and y. But, due to the non-linearity of f ,
solution can not be defined for all t.
This equation can be easily solved as a separable equation.
1 1 −1
Z Z
2
dy = dt, − = t + c, y(t) = .
y y t+c
By IC y(0) = 1, we get 1 = −1/(0 + c), and so c = −1, and
−1
y(t) = .
t−1
We see that the solution blows up as t → 1, and can not be defined beyond that point.
This kind of blow-up phenomenon is well-known for nonlinear equations.

2.4 Modeling with first order equations

General modeling concept: derivatives describe “rates of change”.
Model I: Exponential growth/decay.
Q(t) = amount of quantity at time t
Assume the rate of change of Q(t) is proportional to the quantity at time t. We can write
dQ
(t) = r · Q(t), r : rate of growth/decay
dt
If r > 0: exponential growth
If r < 0: exponential decay
Differential equation:
Q′ = rQ, Q(0) = Q0 .
Solve it: separable equation.
1
Z Z
dQ = r dt, ⇒ ln Q = rt + c, ⇒ Q(t) = ert+c = cert
Q
Here r is called the growth rate. By IC, we get Q(0) = C = Q0 . The solution is

Q(t) = Q0 ert .

Two concepts:

• For r > 0, we define Doubling time TD , as the time such that Q(TD ) = 2Q0 .

ln 2
Q(TD ) = Q0 erTD = 2Q0 , erTD = 2, rTD = ln 2, TD = .
r

• For r < 0, we define Half life (or half time) TH , as the time such that Q(TH ) = 21 Q0 .

1 1 1 ln 2
Q(TH ) = Q0 erTH = Q0 , erTH = , rTH = ln = − ln 2, TH = .
2 2 2 −r
Note here that TH > 0 since r < 0.

24
NB! TD , TH do not depend on Q0 . They only depend on r.

Example 1. If interest rate is 8%, compounded continuously, find doubling time.

ln 2
Answer. Since r = 0.08, we have TD = 0.08 .

Example 2. A radio active material is reduced to 1/3 after 10 years. Find its half life.
dQ
Answer. Model: dt = rQ, r is rate which is unknown. We have the solution Q(t) =
Q0 ert . So
1 1 − ln 3
Q(10) = Q0 , Q0 e10r = Q0 , r= .
3 3 10
To find the half life, we only need the rate r
ln 2 10 ln 2
TH = − = − ln 2 = 10 .
r − ln 3 ln 3

Model II: Interest rate/mortgage problems.

Example 3. Start an IRA account at age 25. Suppose deposit $2000 at the beginning
and $2000 each year after. Interest rate 8% annually, but assume compounded continuously.
Find total amount after 40 years.

Answer. Set up the model: Let S(t) be the amount of money after t years

dS
= 0.08S + 2000, S(0) = 2000.
dt
This is a first order linear equation. Solve it by integrating factor

S ′ − 0.08S = 2000, µ = e−0.08t

e−0.08t
 
2000
Z
S(t) = e0.08t 2000 · e−0.08t dt = e0.08t 2000 +c = + ce0.08t = −25000 + ce0.08t .
−0.08 −0.08
By IC,
S(0) = −25000 + c = 2000, C = 27000,
we get
S(t) = 27000e0.08t − 25000.
When t = 40, we have
S(40) = 27000 · e3.2 − 25000 ≈ 637, 378.
Compare this to the total amount invested: 2000 + 2000 ∗ 40 = 82, 000.

Example 4: A home-buyer can pay $800 per month on mortgage payment. Interest rate is
r annually, (but compounded continuously), mortgage term is 20 years. Determine maximum
amount this buyer can afford to borrow. Calculate this amount for r = 5% and r = 9% and
observe the difference.

25
 Answer. Set up the model: Let Q(t) be the amount borrowed (principle) after t years

dQ
= rQ(t) − 800 ∗ 12
dt
The terminal condition is given Q(20) = 0. We must find Q(0).
Solve the differential equation:

Q′ − rQ = −9600, µ = e−rt

e−rt
 
9600
Z
Q(t) = ert (−9600)e−rt dt = ert −9600 +c = + cert
−r r
By terminal condition
9600 9600
Q(20) = + ce20r = 0, c=−
r r · e20r
so we get
9600 9600 rt
Q(t) = − e .
r r · e20r
Now we can get the initial amount
9600 9600 9600
Q(0) = − 20r
= (1 − e−20r ).
r r·e r
If r = 5%, then
9600
Q(0) = (1 − e−1 ) ≈ $121, 367.
0.05
If r = 9%, then
9600
Q(0) = (1 − e−1.8 ) ≈ $89, 034.
0.09
We observe that with higher interest rate, one could borrow less.

A few words on compounding of interests rate. Let r be the annual interest rate, and
Q0 be the amount of deposit initially. If the interest is compounded continuously, we see that
it gives exponential growth. After t years and we have

Q(t) = Q0 ert .

If the interest is compounded quarterly, we get

Q4 (t) = Q0 (1 + r/4)4t .

If the interest is compounded monthly, we get

Q12 (t) = Q0 (1 + r/12)12t .

If the interest is compounded daily, we get

Q365 (t) = Q0 (1 + r/365)365t .

26
One can show that
Q(T ) > Q365 (t) > Q12 (t) > Q4 (t).
Indeed, consider the function
Qn (t) = (1 + r/n)nt .
Write it out by binomial formula (Taylor expansion)
1 1
Qn (t) = 1 + nt(r/n) + nt(nt − 1)(r/n)2 + nt(nt − 1)(nt − 2)(r/n)3 · · ·
2 6
1 nt − 1 1 nt − 1 nt −2
= 1 + rt + (rt)2 + (rt)2 + · · ·
2 nt 6 nt nt
The Taylor expansion for the exponential function is
1 1
Q(t) = 1 + rt + (rt)2 + (rt)3 + · · ·
2 6
Comparing term-by-term, we have

Q(t) > Qn (t), Qn (t) > Qm (t) (n > m), lim Qn (t) = Q(t).
n→+∞

Model III: Mixing Problem.

Example 5. At t = 0, a tank contains Q0 lb of salt dissolved in 100 gal of water. Assume

that water containing 1/4 lb of salt per gal is entering the tank at a rate of r gal/min. At the
same time, the well-mixed mixture is draining from the tank at the same rate.

(1). Find the amount of salt in the tank at any time t ≥ 0.

(2). When t → ∞, meaning after a long time, what is the limit amount QL ?

Answer. Set up the model:

Q(t) = amount (lb) of salt in the tank at time t (min)
Then, Q′ (t) = [in-rate] − [out-rate].
r
In-rate: r gal/min × 1/4 lb/gal = lb/min
4
Q(t)
concentration of salt in the tank at time t =
100
Q(t) r
Out-rate: r gal/min × lb/gal = Q(t) lb/min
100 100
r r
Q′ (t) = [In-rate] − [Out-rate] = − Q(t), I.C. Q(0) = Q0 .
4 100
(1). Solve the equation
r r
Q′ + Q= , µ = e(r/100)t .
100 4
 
r (r/100)t −(r/100)t r (r/100)t 100
Z
−(r/100)t
Q(t) = e e dt = e e + c = 25 + ce−(r/100)t .
4 4 r

27
By IC
Q(0) = 25 + c = Q0 , c = Q0 − 25,
we get
Q(t) = 25 + (Q0 − 25)e−(r/100)t .
(2). As t → ∞, the exponential term goes to 0, and we have

QL = lim Q(t) = 25lb.

t→∞

We can also observed intuitively that, as time goes on for long, the concentration of salt in the
tank must approach the concentration of the salt in the inflow mixture, which is 1/4. Then,
the amount of salt in the tank would be 1/4 × 100 = 25 lb, as t → +∞.

Example 6. Tank contains 50 lb of salt dissolved in 100 gal of water. Tank capacity is
400 gal. From t = 0, 1/4 lb of salt/gal is entering at a rate of 4 gal/min, and the well-mixed
mixture is drained at 2 gal/min. Find:
(1) time t when it overflows;
(2) amount of salt before overflow;

(3) the concentration of salt at overflow.

Answer. (1). Since the inflow rate 4 gal/min is larger than the outflow rate 2 gal/min,
the tank will be filled up at tf :
400 − 100
tf = = 150min.
4−2
(2). Let Q(t) be the amount of salt at t min.
In-rate: 1/4 lb/gal × 4 gal/min = 1 lb/min
Q(t) Q(t)
Out-rate: 2 gal/min × lb/gal = lb/min
100 + 2t 50 + t
Q(t) 1
Q′ (t) = 1 − , Q′ + Q = 1, Q(0) = 50
50 + t 50 + t
Z 
1
µ = exp dt = exp (ln(50 + t)) = 50 + t
50 + t
 
1 1 1 2
Z
Q(t) = (50 + t)dt = 50t + t + c
50 + t 50 + t 2
By IC:
Q(0) = c/50 = 50, c = 2500,
We get
50t + t2 /2 + 2500
Q(t) = .
50 + t
(3). The concentration of salt at overflow time t = 150 is
Q(150) 50 · 150 + 1502 /2 + 2500 17
= = lb/gal.
400 400(50 + 150) 64

28
 Model IV: Air resistance

Example 7. A ball with mass 0.5 kg is thrown upward with initial velocity 10 m/sec from
the roof of a building 30 meter high. Assume air resistance is |v|/20. Find the max height
above ground the ball reaches.

Answer. Let S(t) be the position (m) of the ball at time t sec. Then, the velocity is
v(t) = dS/dt, and the acceleration is a = dv/dt. Let upward be the positive direction. We
have by Newton’s Law:
v v dv
F = ma = −mg − , a = −g − =
20 20m dt
Here g = 9.8 is the gravity, and m = 0.5 is the mass. We have an equation for v:
dv 1
= − v − 9.8 = −0.1(v + 98),
dt 10
so
1
Z Z
dv = (−0.1)dt, ⇒ ln |v + 98| = −0.1t + c
v + 98
which gives
v + 98 = c̄e−0.1t , ⇒ v = −98 + c̄e−0.1t .
By IC:
v(0) = −98 + c̄ = 10, c̄ = 108, ⇒ v = −98 + 108e−0.1t .
To find the position S, we use S ′ = v and integrate
Z Z
S(t) = v(t) dt = (−98 + 108e−0.1t )dt = −98t + 108e−0.1t /(−0.1) + c

By IC for S,

S(0) = −1080 + c = 30, c = 1110, S(t) = −98t − 1080e−0.1t + 1110.

At the maximum height, we have v = 0. Let’s find out the time T when max height is reached.

v(T ) = 0, −98 + 108e−0.1T = 0, 98 = 108e−0.1T , e−0.1T = 98/108,

−0.1T = ln(98/108), T = −10 ln(98/108) = 10 ln(108/98).

So the max height SM is
108
SM = S(T ) = − 980 ln − 1080e−0.1(10) ln(108/98) + 1110
98
108
= −980 ln − 1080(98/108) + 1110 ≈ 34.78 m.
98

Other possible questions:

• Find the time when the ball hit the ground.

Solution: Find the time t = tH for S(tH ) = 0.

29
 • Find the speed when the ball hit the ground.
Solution: Compute |v(tH )|.

• Find the total distance traveled by the ball when it hits the ground.
Solution: Add up twice the max height SM with the height of the building.

Example 8. (Nonlinear air resistance). A rocket sled with initial speed 100 m/s is slowed
by a channel of water. We assume that the acceleration satisfies a = −0.01v 2 where v is the
velocity.
(1) Find the velocity v(t) at any time t > 0.
(2) Find the distance traveled s(t) at any time t > 0.
(3) Find the distance traveled when the speed is reduced to 10 m/s.

Answer. (1). We set up the equation

v ′ = −0.01v 2 , v(0) = 100.

This is a separable equation:

1 1 1
Z Z
− 2 dv = 0.01 dt, → = 0.01t + c, → v(t) = .
v v 0.01t + c
Use the initial condition:
1
v(0) = = 100, → c = 0.01.
c
So the solution is
1 100
v(t) = = .
0.01t + 0.01 t+1
(2). We have s′ (t) = v(t) and s(0) = 0. Then
Z t Z t
100 t
s(t) = v(τ ) dτ = dτ = 100 ln(τ + 1) = 100 ln(t + 1).

0 0 τ +1 τ =0

(3). Let t̄ be the time such that v(t̄) = 10. Then

100
= 10, → t̄ = 9 s.
t̄ + 1
Thus
s(t̄) = 100 · ln(9 + 1) ≈ 230.3 m

2.5 Autonomous equations and population dynamics

Definition: An autonomous equation is of the form y ′ = f (y), where the function f for the
derivative depends only on y, not on t.
Simplest example: y ′ = ry, exponential growth/decay, where solution is y = y0 ert .
Definition: Zeros of f where f (y) = 0 are called critical points or equilibrium points, or
equilibrium solutions.

30
 Why? Because if f (y0 ) = 0, then y(t) = y0 is a constant solution. It is called an equilib-
rium.
Question: Is an equilibrium stable or unstable?

Example 1. y ′ = y(y − 2). We have two critical points: y1 = 0, y2 = 2.

3 4

2.5
3

2
1.5

y
1
f

0.5 + +
0

−1
−0.5 _

−1 −2
−1 −0.5 0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
y t

We see that y1 = 0 is stable, and y2 = 2 is unstable.

Example 2. For the equation y ′ = f (y) where f (y) is given in the following plot:

0
f

−1

−2

−3

−4

−5
0 1 2 3 4 5 6
y

(A). What are the critical points?

(B). Are they stable or unstable?

(C). Sketch the solutions in the t − y plan, and describe the behavior of y as t → ∞ (as it
depends on the initial value y(0).)

31
Answer. (A). There are three critical points: y1 = 1, y2 = 3, y3 = 5.
(B). To see the stability, we add arrows on the y-axis:

2 +

1 +

f
−1

−2 _ _

−3

−4

−5
0 1 2 3 4 5 6
y

We see that y1 = 1 is stable, y2 = 3 is unstable, and y3 = 5 is stable.

(C). The sketch is given below:

3
y

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t

Asymptotic behavior for y as t → ∞ depends on the initial value of y:

• If y(0) < 1, then y(t) → 1,

• If y(0) = 1, then y(t) = 1;

• If 1 < y(0) < 3, then y(t) → 1;

• If y(0) = 3, then y(t) = 3;

• If 3 < y(0) < 5, then y(t) → 5;

• If y(0) = 5, then y(t) = 5;

32
 • if y(t) > 5, then y(t) → 5.

Stability: is not only stable or unstable.

Example 3. For y ′ = y 2 , we have only one critical point y1 = 0. For y < 0, we have
y′> 0, and for y > 0 we also have y ′ > 0. So solution is increasing on both intervals. So on
the interval y < 0, solution approaches y = 0 as t grows, so it is stable. But on the interval
y > 0, solution grows and leaves y = 0, and it is unstable. This type of critical point is called
semi-stable. This happens when one has a double root for f (y) = 0.

Example 4. For equation y ′ = f (y) where f (y) is given in the plot

1.5

0.5
f

−0.5

−1
−1 −0.5 0 0.5 1 1.5 2 2.5 3 3.5 4
y

(A). Identify equilibrium points;

(B). Discuss their stabilities;

(C). Sketch solution in y − t plan;

(D). Discuss asymptotic behavior as t → ∞.

Answer. (A). y = 0, y = 1, y = 2, y = 3 are the critical points.

(B). y = 0 is stable, y = 1 is semi-stable, y = 2 is unstable, and y = 3 is stable.
(C). The Sketch is given in the plot:

33
 4

3.5

2.5

1.5

y 1

0.5

−0.5

−1
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
t

(D). The asymptotic behavior as t → ∞ depends on the initial data.

• If y(0) < 1, then y → 0;

• If 1 ≤ y(0) < 2, then y → 1;

• If y(0) = 2, then y(t) = 2;

• If y(0) > 2, then y → 3.

Application in population dynamics: let y(t) be the population of a species.

Typically, the rate of change in the population depends on the population y, at any time t.
This means y ′ (t) typically does NOT depend on t, and we end up with autonomous equations.
Model 1. The simplest model is the exponential growth, with growth rate r:

y ′ (t) = ry, y(0) = y0 .

The solution can be written explicitly as

y(t) = y0 ert .

If initially there is no life, i.e., y0 = 0, then it remains that way and y(t) ≡ 0. Otherwise, if
only a very small amount of population exists, i.e., y0 > 0, then y(t) will grow exponentially
in time.
Of course, this model is not realistic. In nature there is limited natural resource that can
support only limited amount of population.
Model 2. The more realistic model is the “logistic equation”:
dy
= (r − ay)y.
dt
or
dy  y r
=r 1− y, k= ,
dt k a

34
r=intrinsic growth rate,
k=environmental carrying capacity.
critical points: y = 0, y = k. Here y = 0 is unstable, and y = k is stable.
If 0 < y(0) < k, then y → k as t grows.
If y(0) > k, then y → k as t grows.
Im summary, if y(0) > 0, then
lim y(t) = k.
t→+∞

Remark*. Since this autonomous equation is separable, it is possible to solve it and

obtain an explicit expression for the solution. Consider now the IVP
dy  y
=r 1− y, y(0) = y0 ,
dt k
after separating the variables, we have
1
Z Z

dy = rdt
1 − ky y

The tricky part is the integration of the left hand side. We apply a technique called partial fraction,
and get
1 k 1 1
y
= = + .
1− k y (k − y) y y k−y
Then Z    
1 1  y 
+ dy = ln |y| − ln |k − y| = ln  .
y k−y k − y
This yields
     
 y   y  rt
 y0 
ln 
  = rt + c, →  k − y  = ce ,
  where c = 
 .
k − y k − y0 

If 0 < y0 < k, so does 0 < y(t) < k, and we can remove the absolute value signs in the
solution, and after some manipulation we get

ckert ky0 ert ky0

y(t) = rt
= rt
= .
1 + ce (k − y0 ) + y0 e y0 + (k − y0 )e−rt

We see that y → k as t → ∞.
On the other hand, if y0 > k, we have
y y0
= cert , c= .
y−k y0 − k
After some manipulation we get

kcert ky0 ert ky0 ky0

y(t) = rt
= rt
= −rt
=
ce − 1 y0 e − (y0 − k) y0 − (y0 − k)e y0 + (k − y0 )e−rt

We see in both cases the solution takes the same form!

35
 Model 3. An even more detailed model is the logistic growth with a threshold:
 y  y
y ′ (t) = −r 1 − · 1− y, r > 0, 0 < T < K.
T K
We see that there are 3 critical points: y = 0, y = T, y = K, where y = T is unstable, and
y = 0, y = K are stable.
Let y(0) = y0 be the initial population. We discuss the asymptotic behavior as t → +∞.

• If 0 < y0 < T , then y → 0.

• If T < y0 < K, then y → K.

• If y0 > K, then y → K.

We see that y0 = T work as a threshold. We have

lim y(t) = 0 if y(0) < T

t→+∞
lim y(t) = K if y(0) > T.
t→+∞

Some populations follow this model, for example, the some fish population in the ocean. If we
over-fishing and make the population below certain threshold, then the fish will go extinct.
That’s too sad.

Remark*. One can also solve this equation and obtain an explicit expression, but the
computation is lengthy!

2.6 Exact Equations

We will review: partial derivatives and Chain Rules for functions of 2 variables.
Consider a function f (x, y). We use these notations for the partial derivatives:

∂f ∂f
fx = , fy =
∂x ∂y
and correspondingly the higher derivatives:

∂2f ∂2f
fxx = , fyy =
∂x2 ∂y 2
and the cross derivatives
∂2f ∂2f
   
∂ ∂f ∂ ∂f
fxy = = , fyx = = ,
∂x∂y ∂y ∂x ∂y∂x ∂x ∂y

where we have the identity

fxy = fyx .
Now, consider x = x(t) and y = y(t), and we form a composite function as f (x(t), y(t)).
We see that f now depends only on t.

36
 Chain Rule:
df ∂f dx ∂f dy
= · + · = f x x′ + f y y ′ .
dt ∂x dt ∂y dt

Example 1. Let f (x, y) = x2 y 2 + ex , and x(t) = t2 , y(t) = et , and consider the composite
function f (x(t), y(t)). Compute df
dt .

Answer. We first compute the derivatives

fx = 2xy 2 + ex , fy = 2x2 y, x′ (t) = 2t, y ′ (t) = et

By the Chain Rule, we compute

df
= (2xy 2 + ex )2t + 2x2 yet
dt  
2
= 2t 2t2 e2t + et + 2t4 et et

Special case: If y = y(x), then the composite function f (x, y(x)) will follow this form of
Chain Rule
df ∂f dx ∂f dy
= · + · = fx + fy y ′ (x).
dx ∂x dx ∂y dx

Exact Equations. We will start with an Example.

Example 2. Let y(x) be the unknown. Consider the equation

6x + ex y 2 + 2ex yy ′ = 0

We see that the equation is NOT linear. It is NOT separable either. None of the methods we
know can solve it.
However, define the function

ψ(x, y) = 3x2 + ex y 2

We notice that
ψx = 6x + ex y 2 , ψy = 2ex y
and the equation can be written as

ψx (x, y) + ψy (x, y)y ′ = 0.

Since y = y(x), we apply the Chain Rule to the composite function ψ(x, y(x)) and get

dψ
= ψx + ψy y ′ (x)
dx
which is the left-hand side of the equation. By the differential equation, we now have
dψ
= 0, → ψ(x, y) = C
dx

37
where C is an arbitrary constant, to be determined by initial condition. We have found the
solution in an implicit form:
3x2 + ex y 2 = C.
In this example, we are even able to write out the solution in an explicit form by algebraic
manipulation
y 2 = e−x (C − 3x2 ),
p
y = ± e−x (C − 3x2 ).
Here, the choice of + or − sign should be determined by initial condition.

Definition of an exact equation. En equation in the form

M (x, y) + N (x, y)y ′ = 0

is called exact if there exists a function ψ(x, y) such that ψx = M and ψy = N .

How to check if an equation is exact? By the identity ψxy = ψyx , we must have

My (x, y) = Nx (x, y).

How to solve it? Need to find the function ψ, then we get implicit solution

ψ(x, y) = C.

How to find ψ? By using the facts that

ψx = M (x, y), ψy = N (x, y)

and integrate. We will see this through an example.

Example 3. Check if the following equation is exact

(2x + 3y) + (x − 2y)y ′ = 0.

Answer. Here we have

M (x, y) = 2x + 3y, N (x, y) = x − 2y.

Then
My = 3, Nx = 1, My 6= Nx ,
so the equation is not exact.

Remark 1. Consider a separable equation:

f (x)
y′ = .
g(y)

Multiply both sides by g(y) and re-arranging terms, we get

f (x) − g(y)y ′ = 0.

38
Now we check if this equation is exact. Clearly, we have fy = 0 and gx = 0, so it is exact. We
may conclude that all separable equations can be rewritten into an exact equation.
Remark 2. However, the exactness of an equation is up to manipulation. Consider the
separable equation in Remark 1, and rewrite the equation now into
f (x)
− y ′ = 0.
g(y)
So now we have
f (x)
M (x, y) = , N (x, y) = 1.
g(y)
Since Nx = 0 and My 6= 0 in general, the equation is not exact.
So, be careful. When you say an equation is exact, you must specify in which form you
present your equation.

Example 4. Consider the equation

(2x + y) + (x + 2y)y ′ = 0

(1) Is it exact? (2) If yes, find the solution with the initial condition y(1) = 1.

Answer. (1). We have M = 2x + y and N = x + 2y, so My = 1 and Nx = 1, so the

equation is exact.
(2). To solve it, we need to find the function ψ. We have

ψx = 2x + y, ψy = x + 2y. (A)

Integrating the first equation in x:

Z Z
ψ(x, y) = ψx dx = (2x + y) dx + h(y) + x2 + xy + h(y).

(Review: To perform a partial integration in x, we treat y as a constant. Therefore, the

integration constant could depend on y since it is a constant. That’s why we add h(y), a
function of y, to the anti-derivative.)
To determine h(y), we use the second equation in (A).

ψy = x + h′ (y) = x + 2y, h′ (y) = 2y, h(y) = y 2 .

Therefore
ψ = x2 + xy + y 2
and the implicit solution is
x2 + xy + y 2 = C.
Finally, we determine the constant C by initial condition. Plug in x = 1, y = 1, we get
C = 3, so the implicit solution is
x2 + xy + y 2 = 3.
Remark. This procedure is rather lengthy, and could easily cause many mistakes. An
alternative (shorter) method is described below. We integrate the two equations in (A), with

39
respect to the partial derivatives. This means, for the first equation we integrate in x and the
second equation in y. We do not take the integration constants, we only find terms in the
anti-derivatives. This gives two expressions of ψ:
Z Z
2
ψ = (2x + y) dx = x + xy, ψ = (x + 2y) dy = xy + y 2 .
x y

Now we comb through the two expressions, and collect all different terms. We get

ψ = x2 + xy + y 2 .

Note that we get the same answer.

Example 5. Given equation

(xy 2 + bx2 y) + (x + y)x2 y ′ = 0.

(1) Find the values of b such that the equation is exact. (2) Solve it with that value of b.

Answer. (1). We have

M (x, y) = xy 2 + bx2 y, N (x, y) = x3 + x2 y

so
My = 2xy + bx2 , Nx = 3x2 + 2xy
We see that we must have b = 3 to ensure My = Nx which would make the equation exact.
(2). We now set b = 3. To solve the equation, we need to find the function ψ. We have

ψx = xy 2 + 3x2 y, ψy = x3 + x2 y.

Integrating the first equation in x:

1
Z
ψ(x, y) = (xy 2 + 3x2 y)dx + h(y) = x2 y 2 + x3 y + h(y)
2

To find h(y), we use ψy :

ψy = x2 y + x3 + h′ (y) = N = x3 + x2 y

We must have h′ (y) = 0, so we can use h(y) = 0.

The implicit solution is
1 2 2
x y + x3 y = C,
2
where C is arbitrary, to be determined by initial condition.
(We remark that, if we use the alternative method in the previous Remark, we would get
the same answer. Try this yourself.)

We make some observations.

40
 (1). The exactness of an equation is up to manipulation. Consider an exact equation
M (x, y) + N (x, y)y ′ = 0 where My = Nx . If we multiply the equation by some function f (x, y)
on both sides, we get
f (x, y)M (x, y) + f (x, y)N (x, y)y ′ = 0
which in general is not exact for an arbitrary choice of f .
(2). Then, the other way around is also possible. Consider an equation M̂ (x, y) +
N̂ (x, y)y ′ = 0, which is not exact as they appear. If the conditions are favorable, it might be
possible to find some function µ(x, y), such that
µ(x, y)M̂ (x, y) + µ(x, y)N̂ (x, y)y ′ = 0
becomes exact. If this is true, then µ(x, y) is called an integrating factor.
(3). Consider now an separable equation y ′ = f (x)/g(y). Write it as
f (x)
− y′ = 0
g(y)
the equation is not exact. But if we multiply both sides by g(y), we get
f (x) − g(y)y ′ = 0
which is exact. This shows that: any separable equation could be written into an exact equation.

2.7 Euler’s method*

In general, for an equation
y ′ (t) = f (t, y), y(t̄) = ȳ,
it might not be possible to find an analytic expression for the solution y(t). If one wants
to know how the solution behaves, one way would be using directional field and sketch the
solutions. Another way is to use numerical methods to compute some approximate solution.
We choose a grid size, i.e., a small value h > 0, and make the grid, i.e., a set of points for
t, such that
t0 = t̄, tk = t0 + kh, k = 1, 2, 3, · · ·
We seek approximate solution only at these grid points, i.e.,
yk ≈ y(xk ), k = 1, 2, 3, · · ·
The values of yk are computed through iteration. The main idea here is, at t = tk , one
checks the value for the gradient y ′ (tk ) = f (tk , yk ), and use this as the slope and take a step
forward in time. This gives the following algorithm: Given the starting point t0 = t̄, y + 0 = ȳ,
perform the following computation:
yk+1 = yk + h · f (tk , yk ), for k = 0, 1, 2, · · ·
Then, one can plot the graph y(t) using these grid points, in a programing language.
One can illustrate this idea through a graph, together with directional field.
Numerical methods for ODEs are a well-studied topic. There are many methods developed,
for various models. Better methods (better than Euler) are available , would would give higher
order accuracy. Here we don’t have time to go into details. If you are interested in this topic,
you could take the course CMPSC/Math 451, or an extended version of it, CMPSC/Math 455
+ 456.

41
Chapter 3

Second Order Linear Equations

In this chapter we study linear 2nd order ODEs. The general form of these equations is

a2 (t)y ′′ + a1 (t)y ′ + a0 (t)y = b(t),

where
a2 (t) 6= 0, y(t0 ) = y0 , y ′ (t0 ) = ȳ0 .
If b(t) ≡ 0, we call it homogeneous. Otherwise, it is called non-homogeneous.

3.1 Homogeneous equations with constant coefficients

This is the simplest case: a2 , a1 , a0 are all constants, and g = 0. Let’s write:

a2 y ′′ + a1 y ′ + a0 y = 0.

We start with an example.

Example 1. Solve y ′′ − y = 0, (we have here a2 = 1, a1 = 0, a0 = 1).

Answer. Let’s guess an answer of the form y1 (t) = et .

Check to see if it satisfies the equation: y ′′ = et , so y ′′ − y = et − et = 0. So it is a solution.
Guess another function: y2 (t) = e−t .
Check: y ′ = −e−t , so y ′′ = e−t , so y ′′ − y = et − et = 0. So it is also a solution.
Claim: Another function y = c1 y1 + c2 y2 for any arbitrary constants c1 , c2 (this is called
“a linear combination of y1 , y2 ”) is also a solution.
Check if this claim is true:
y(t) = c1 et + c2 e−t ,
then
y ′ = c1 et − c2 e−t , y ′′ = c1 et + c2 e−t , ⇒ y ′′ − y = 0.

Actually this claim is a general property. It is called the principle of superposition.

Theorem (The Principle of Superposition) Let y1 (t) and y2 (t) be solutions of

a2 (t)y ′′ + a1 (t)y ′ + a0 (t)y = 0

42
Then, y = c1 y1 + c2 y2 for any constants c1 , c2 is also a solution.

Proof: If y1 solves the equation, then

a2 (t)y1′′ + a1 (t)y1′ + a0 (t)y1 = 0. (I)

If y2 solves the equation, then

a2 (t)y2′′ + a1 (t)y2′ + a0 (t)y2 = 0. (II)

Multiple (I) by c1 and (II) by c2 , and add them up:

a2 (t)(c1 y1 + c2 y2 )′′ + a1 (t)(c1 y1 + c2 y2 )′ + a0 (t)(c1 y1 + c2 y2 ) = 0.

Let y = c1 y1 + c2 y2 , we have

a2 (t)y ′′ + a1 (t)y ′ + a0 (t)y = 0

therefore y is also a solution to the equation.

How to find the solutions of a2 y ′′ + a1 y ′ + a0 y = 0?

We seek solutions in the form y(t) = ert . Find r.

y ′ = rert = ry, y ′′ = r 2 ert = r 2 y

a2 r 2 y + a1 ry + a0 y = 0
Since y 6= 0, we get
a2 r 2 + a1 r + a0 = 0
This is called the characteristic equation.
Conclusion: If r is a root of the characteristic equation, then y = ert is a solution.
If there are two real and distinct roots r1 6= r2 , then the general solution is y(t) =
c1 er1 t + c2 er2 t where c1 , c2 are two arbitrary constants to be determined by initial conditions
(ICs).

Example 2. Consider y ′′ − 5y ′ + 6y = 0.

(a). Find the general solution.

(b). If ICs are given as: y(0) = −1, y ′ (0) = 5, find the solution.

(c). What happens to y(t) when t → ∞?

Answer. (a). The characteristic equation is: r 2 − 5r + 6 =, so (r − 2)(r − 3) = 0, two

roots: r1 = 2, r2 = 3. General solution is:

y(t) = c1 e2t + c2 e3t .

(b). y(0) = −1 gives: c1 + c2 = −1.

y ′ (0) = 5: we have y ′ = 2c1 e2t + 3c2 e3t , so y ′ (0) = 2c1 + 3c2 = 5.

43
 Solve these two equations for c1 , c2 : Plug in c2 = −1 − c1 into the second equation, we get
2c1 + 3(−1 − c1 ) = 5, so c1 = −8. Then c2 = 7. The solution is

y(t) = −8e2t + 7e3t .

(c). We see that y(t) = e2t · (−8 + 7et ), and both terms in the product go to infinity as t
grows. So y → +∞ as t → +∞.

Example 3. Find the solution for 2y ′′ +y ′ −y = 0, with initial conditions y(1) = 0, y ′ (1) =
3.

Answer. Characteristic equation:

1
2r 2 + r − 1 = 0, ⇒ (2r − 1)(r + 1) = 0, ⇒ r1 = , r2 = −1.
2
General solution is:
t
y(t) = c1 e 2 + c2 e−t .
The ICs give
1
y(1) = 0 : c1 e 2 + c2 e−1 = 0. (A)
1 1 1 1
y ′ (1) = 3 : y ′ (t) = c1 e 2 t − c2 e−t , c1 e 2 − c2 e−1 = 3. (B)
2 2
(A)+(B) gives
3 1 1
c1 e 2 = 3, c1 = 2e− 2 .
2
Plug this in (A):
1 1 1
c2 = −ec1 e 2 = −e2e 2 e 2 = −2e.
The solution is
1 1 1
y(t) = 2e− 2 e 2 t − 2ee−t = 2e 2 (t−1) − 2e−(t−1) ,
and as t → ∞ we have y → ∞.

Remark. Note that the initial data is given at t = 1, not t = 0. It appears that this
causes difficulties in computing the constants c1 , c2 . Note that we write the final answer in
the form such that it becomes a function of t − 1. Indeed, this suggests a better form for the
general solution that would lead to simpler computation when one plugs in the initial data.
We take the general solution as

y(t) = c̄1 e(t−1)/2 + c̄2 e−(t−1) .

Note that we replaced t with t − t0 , where t0 = 1 is the time when the initial conditions are
given. We have
1
y ′ (t) = c̄1 e(t−1)/2 − c̄2 e−(t−1)
2
and the initial conditions give us
1
y(1) = c̄1 + c̄2 = 0, y ′ (1) = c̄1 − c̄2 = 3.
2

44
This leads to easy computation to find c̄1 = 2, c̄2 = −2, and the solution is the same
1
y(t) = 2e 2 (t−1) − 2e−(t−1) .

Summary of receipt:
1. Write the characteristic equation; y ′′ → r 2 , y ′ → r, y → 1.

2. Find the roots;

3. Write the general solution;

4. Set in ICs to get the arbitrary constants c1 , c2 .

Example 4. Consider the equation y ′′ − 4y = 0.

(a). Find the general solution.

(b). If the initial conditions are given as y(0) = 1 and y ′ (0) = a, then, for what values of a
would y remain bounded as t → +∞?

Answer. (a). Characteristic equation

r 2 − 4 = 0, ⇒ r1 = −2, r2 = 2.

General solution is
y(t) = c1 e−2t + c2 e2t .
(b). If y(t) remains bounded as t → ∞, then the term e2t must vanish, which means we must
have c2 = 0. This means y(t) = c1 e−2t . If y(0) = 1, then y(0) = c1 = 1, so y(t) = e−2t . This
gives y ′ (t) = −2e−2t which means a = y ′ (0) = −2.

Example 5. Consider the equation 2y ′′ + 3y ′ = 0. The characteristic equation is

3
2r 2 + 3r = 0, ⇒ r(2r + 3) = 0, ⇒ r 1 = − , r2 = 0
2
The general solution is
3 3
y(t) = c1 e− 2 t + c2 e0t = c1 e− 2 t + c2 .
As t → ∞, the first term in y vanishes, and we have y → c2 .

Example 6. Find a 2nd order equation such that c1 e3t + c2 e−t is its general solution.

Answer. From the form of the general solution, we see the two roots are r1 = 3, r2 = −1.
The characteristic equation could be (r − 3)(r + 1) = 0, or this equation multiplied by any
non-zero constant. So r 2 − 2r − 3 = 0, which gives us the equation

y ′′ − 2y ′ − 3y = 0.

45
3.2 Solutions of Linear Homogeneous Equations; the Wron-
skian
We consider some theoretical aspects of the solutions to a general 2nd order linear equations.

Theorem . (Existence and Uniqueness Theorem) Consider the initial value problem

y ′′ + p(t)y ′ + q(t)y = g(t), y(t0 ) = y0 , y ′ (t0 ) = ȳ0 .

If p(t), q(t) and g(t) are continuous and bounded on an open interval I containing t0 , then
there exists exactly one solution y(t) of this equation, valid on I.

Example 1. Given the equation

(t2 − 3t)y ′′ + ty ′ − (t + 3)y = et , y(1) = 2, y ′ (1) = 1.

Find the largest interval where solution is valid.

Answer. Rewrite the equation into the proper form:

t t+3 et
y ′′ + y′ − y= ,
t(t − 3) t(t − 3) t(t − 3)

so we have
t t+3 et
p(t) = , q(t) = − , g(t) = .
t(t − 3) t(t − 3) t(t − 3)
We see that we must have t 6= 0 and t 6= 3. Since t0 = 1, then the largest interval is I = (0, 3),
or 0 < t < 3. See the figure below.
t0
✛ ✲
❄
× × ✲ x
0 1 2 3

Definition. Given two functions f (t), g(t), the Wronskian is defined as

˙ f g′ − f ′ g.
W (f, g)(t) =

Remark: One way to remember this definition could be using the determinant of a 2 × 2
matrix,  
 f g 
W (f, g)(t) =  ′ ′  .
f g

Main property of the Wronskian:

46
 • If W (f, g) ≡ 0, then f and g are linearly dependent.

• Otherwise, they are linearly independent.

If two functions are linearly dependent, they are essentially the same function. In fact, one
function could be written as a scalar multiple of the other. To be precise, W (f, g) ≡ 0 (linearly
dependent) if and only if f (t) = k · g(t) for some constant k. To see this, assume W (f, g) ≡ 0,
i.e., f g′ − f ′ g ≡ 0, then
 ′
f f ′ g − f g′ −W (f, g) f
= = = 0, → = k(constant).
g g2 g2 g
On the other hand, if f (t) = kg(t) for some constant k, then

W (f, g) = kg(t) · g ′ (t) − kg′ (t)g(t) ≡ 0.

Example 2. Compute the Wronskian of the following function pairs.

(a). f = et , g = e−t .

Answer. We have
W (f, g) = et (−e−t ) − et e−t = −2 6= 0
so they are linearly independent.

(b). f (t) = sin t, g(t) = cos t.

Answer. We have

W (f, g) = sin t(sin t) − cos t cos t = −1 6= 0

and they are linearly independent.

(c). f (t) = t + 1, g(t) = 4t + 4.

Answer. We have
W (f, g) = (t + 1)4 − (4t + 4) = 0
so they are linearly dependent. (In fact, we have g(t) = 4 · f (t).)
(d∗ ). f (t) = 2t, g(t) = |t|.

Answer. Note that g ′ (t) = sign(t) where sign is the sign function. So

W (f, g) = 2t · sign(t) − 2|t| = 0

(we used t · sign(t) = |t|). So they are linearly dependent for all t 6= 0.

Theorem . Suppose y1 (t), y2 (t) are two solutions of

y ′′ + p(t)y ′ + q(t)y = 0.

Then

47
 (I) We have either W (y1 , y2 ) ≡ 0 or W (y1 , y2 ) never zero;

(II) If W (y1 , y2 ) 6= 0, the y = c1 y1 + c2 y2 is the general solution. They are also called to form
a fundamental set of solutions. As a consequence, for any ICs y(t0 ) = y0 , y ′ (t0 ) = ȳ0 ,
there is a unique set of (c1 , c2 ) that gives a unique solution.

Proof of part (II): Plug in the initial conditions, we get

c1 y1 (t0 ) + c2 y2 (t0 ) = y0
c1 y1′ (t0 ) + c2 y2′ (t0 ) = ȳ0 .

Writing this into matrix-vector form:

     
y1 (t0 ) y2 (t0 ) c y0
· 1 = .
y1′ (t0 ) y2′ (t0 ) c2 ȳ0

We see that there exist a uniqueness solution for c1 , c2 if the determinant of the coefficient
matrix is not zero. This determinant is precisely the Wronskian.

√
Example . Show that y1 (t) = t and y2 (t) = 1/t form a fundamental set of solution for

2t2 y ′′ + 3ty ′ − y = 0, (t > 0).

Answer. One needs to check two things: (1) plug in y1 , y2 to see that they are solutions.
(skip details). (2) Compute the Wronskian W (y1 , y2 ) to see if it is not 0. In fact, one get
W (y1 , y2 ) = − 23 t−3/2 < 0 for t > 0.

The next Theorem is probably the most important one in this chapter.

Theorem (Abel’s Theorem) Let y1 , y2 be two (linearly independent) solutions to

y ′′ + p(t)y ′ + q(t)y = 0

on an open interval I. Then, the Wronskian W (y1 , y2 ) on I is given by

Z
W (y1 , y2 )(t) = C · exp( −p(t) dt),

for some constant C depending on y1 , y2 , but independent on t or on I.

Proof. Since y1 , y2 are solutions, we have

y1′′ + p(t)y1′ + q(t)y1 = 0

y2′′ + p(t)y2′ + q(t)y2 = 0

Multiply the first equation by −y2 and the second one by y1 , and add them up, we get

(y1 y2′′ − y1′′ y2 ) + p(t)(y1 y2′ − y1′ y2 ) = 0.

48
Note that W (y1 , y2 ) = y1 y2′ − y1′ y2 and W ′ = y1 y2′′ − y1′′ y2 , we get

W ′ + p(t)W = 0

which is a first order linear equation which we can solve. We get

 Z 
W (y1 , y2 )(t) = C exp − p(t)dt

Completing the proof. Note that this also proves part (I) of the previous Theorem. The
Wronskian W (y1 , y2 ) is either identically 0 (when C = 0) or never 0 (when C 6= 0).

Example 3. Given
t2 y ′′ − t(t + 2)y ′ + (t + 2)y = 0.
Find W (y1 , y2 ) without solving the equation.

Answer. We first find the p(t)

t+2
p(t) = −
t
which is valid for t 6= 0. By Abel’s Theorem, we have
t+2
Z Z
W (y1 , y2 ) = C · exp( −p(t) dt) = C · exp( dt) = Cet+2 ln |t| = Ct2 et .
t

NB! The solutions are defined on either (0, ∞) or (−∞, 0), depending on t0 .

From now on, when we say two solutions y1 , y2 of the equation, we mean two linearly
independent solutions that can form a fundamental set of solutions.

Example 4. If y1 , y2 are two solutions of

ty ′′ + 2y ′ + tet y = 0,

and W (y1 , y2 )(1) = 2, find W (y1 , y2 )(5).

Answer. First we find that p(t) = 2/t. By Abel’s Theorem we have

 Z 
2
W (y1 , y2 )(t) = C · exp − dt = C · e− ln t = Ct−2 .
t

If W (y1 , y2 )(1) = 2, then C = 2. So we have

2
W (y1 , y2 )(5) = (2)5−2 = .
25

Example 5. If W (f, g) = 3e4t , and f = e2t , find g.

49
 Answer. By definition of the Wronskian, we have

W (f, g) = f g ′ − f ′ g = e2t g′ − 2e2t g = 3e4t ,

which gives a 1st order equation for g:

g ′ − 2g = 3e2t .

Solve it for g, by method of integrating factor:

Z
µ(t) = e−2t , g(t) = e2t e−2t 3e2t dy = e2t (3t + c).

We can choose c = 0, and get g(t) = 3te2t .

Next example shows how Abel’s Theorem can be used to solve 2nd order differential
equations.

Example 6. Consider the equation y ′′ + 2y ′ + y = 0. Find the general solution.

Answer. The characteristic equation is r 2 + 2r + 1 = 0, which given double roots r1 =

r2 = −1. So we know that y1 = e−t is a solutions. How can we find another solution y2 that’s
linearly independent?
By Abel’s Theorem, we have
Z 
W (y1 , y2 ) = C exp −2 dt = Ce−2t ,

and we can choose C = 1 and get W (y1 , y2 ) = e−2t . By the definition of the Wronskian, we
have
W (y1 , y2 ) = y1 y2′ − y1′ y2 = e−t y2′ − (−e−t y2 ) = e−t (y2′ + y2 ).
These two computation must have the same answer, so

e−t (y2′ + y2 ) = e−2t , y2′ + y2 = e−t .

This is a 1st order equation for y2 . Solve it:

Z
t −t
µ(t) = e , y2 (t) = e et e−t dt = e−t (t + c).

Choosing c = 0, we get y2 = tet . The general solution is

y(t) = c1 y1 + c2 y2 = c1 e−t + c2 te−t .

This is called the method of reduction of order. We will study it more later in chapter 3.4.

50
3.3 Complex Roots
We start with an example.

Example Consider the equation y ′′ + y = 0, find the general solution.

Answer. By inspection, we need to find a function such that y ′′ = −y. We see that
y1 = cos t and y2 = sin t both work. By the Wronskian W (y1 , y2 ) = −2 6= 0, we see that these
two solutions are linearly independent. Therefore, the general solution is

y(t) = c1 y1 + c2 y2 = c1 cos t + c2 sin t.

Let’ try to connect this with the characteristics equation:

r 2 + 1 = 0, r 2 = −1, r1 = i, r2 = −i.

The roots are complex. In fact, they are complex-conjugate pair. We see that the imaginary
part seems to give sin and cos functions.

In general, the roots of the characteristic equation can be complex numbers. Consider the
equation
ay ′′ + by ′ + cy = 0, → ar 2 + br + c = 0.
The two roots are √
−b ±b2 − 4ac
r1,2 = .
2a
If b2 − 4ac < 0, the root are complex, i.e., a pair of complex conjugate numbers. We will write
r1,2 = λ ± iµ. There are two solutions:

y1 = e(λ+iµ)t = eλt eiµt , y2 = y1 = e(λ−iµ)t = eλt e−iµt .

To deal with exponential function with pure imaginary exponent, we need the Euler’s Formula:

eiβ = cos β + i sin β, e−iβ = cos β − i sin β.

Then
y1 = eλt (cos µt + i sin µt), y2 = eλt (cos µt − i sin µt).
But these solutions are complex-valued. We want real-valued solutions! To achieve this, we
use the Principle of Superposition. If y1 , y2 are two solutions, then c1 y1 +c2 y2 is also a solution
for any constants c1 , c2 . In particular, the functions 12 (y1 + y2 ), 2i
1
(y1 − y2 ) are also solutions.
Write
1 1
˙ (y1 + y2 ) = eλt cos µt,
ỹ1 = ỹ2 =
˙ (y1 − y2 ) = eλt sin µt.
2 2i
We need to make sure that they are linearly independent. We can check the Wronskian,

W (ỹ1 , ỹ2 ) = µe2λt 6= 0. (home work problem).

So y1 , y2 are linearly independent, and we have the general solution

y(t) = c1 eλt cos µt + c2 eλt sin µt = eλt (c1 cos µt + c2 sin µt).

51
 Remark: We note here that if a complex valued function is a solution, then the real
part and the imaginary part are each a solution. This is a more general result. Now let
y(t) = u(t) + iv(t) be a solution of

y ′′ + p(t)y ′ + q(t)y = 0

Then, using y ′ = u′ + iv ′ and y ′′ = u′′ + iv ′′ , we get

(u′′ + iv ′′ ) + p(t)(u′ + iv ′ ) + q(t)(u + iv) = 0

Collect the real part and the imaginary part, we have

   
u′′ + p(t)u′ + q(t)u + i v ′′ + p(t)v ′ + q(t) = 0

which implies
u′′ + p(t)u′ + q(t)u = 0, v ′′ + p(t)v ′ + q(t) = 0,
proving that u and v are both solutions.

Example 1. (Perfect Oscillation: Simple harmonic motion.) Solve the initial value prob-
lem
π π
y ′′ + 4y = 0, y( ) = 0, y ′ ( ) = 1.
6 6

Answer. The characteristic equation is

r 2 + 4 = 0, ⇒ r = ±2i, ⇒ λ = 0, µ = 2.

The general solution is

y(t) = c1 cos 2t + c2 sin 2t.
Find c1 , c2 by initial conditions: since y ′ = −2c1 sin 2t + 2c2 cos 2t, we have
√
π π π 1 3
y( ) = 0 : c1 cos + c2 sin = c1 + c2 = 0,
6 3 3 2 2
√
′ π π π 3 1
y ( ) = 1 : −2c1 sin + 2c2 cos = −2c1 + 2c2 = 1.
6 3 3 2 2
√
3
Solve these two equations, we get c1 = − 4 and c2 = 14 . So the solution is
√
3 1
y(t) = − cos 2t + sin 2t,
4 4
which is a periodic oscillation. This is also called perfect oscillation or simple harmonic motion.

Example 2. (Decaying oscillation.) Find the solution to the IVP (Initial Value Problem)

y ′′ + 2y ′ + 101y = 0, y(0) = 1, y ′ (0) = 0.

52
 Answer. The characteristic equation is

r 2 + 2r + 101 = 0, ⇒ r1,2 = −1 ± 10i, ⇒ λ = −1, µ = 10.

So the general solution is

y(t) = e−t (c1 cos 10t + c2 sin 10t),
so
y ′ (t) = −e−t (c1 cos t + c2 sin t) + e−t (−10c1 sin t + 10c2 cos t)
Fit in the ICs:
y(0) = 1 : y(0) = e0 (c1 + 0) = c1 = 1,
y ′ (0) = 0 : y ′ (0) = −1 + 10c2 = 0, c2 = 0.1.
Solution is
y(t) = e−t (cos t + 0.1 sin t).
The graph is given below:

0.8

0.6

0.4

0.2

−0.2

−0.4

−0.6

−0.8

−1
0 0.5 1 1.5 2 2.5 3 3.5 4

We see it is a decaying oscillation. The sin and cos part gives the oscillation, and the e−t
part gives the decaying amplitude. As t → ∞, we have y → 0.

Example 3. (Growing oscillation) Find the general solution of y ′′ − y ′ + 81.25y = 0.

Answer.

r 2 − r + 81.25 = 0, ⇒ r = 0.5 ± 9i, ⇒ λ = 0.5, µ = 9.

The general solution is

y(t) = e0.5t (c1 cos 9t + c2 sin 9t).
A typical graph of the solution looks like:

53
 20

15

10

−5

−10

−15

−20
0 1 2 3 4 5 6

We see that y oscillate with growing amplitude as t grows. In the limit when t → ∞, y
oscillates between −∞ and +∞.

Conclusion: Sign of λ, the real part of the complex roots, decides the type of oscillation:

• λ = 0: perfect oscillation;

• λ < 0: decaying oscillation;

• λ > 0: growing oscillation.

−b
We note that since λ = 2a , so the sign of λ follows the sign of −b/a.

3.4 Repeated roots; reduction of order

For the characteristic equation ar 2 + br + c = 0, if b2 = 4ac, we will have two repeated roots
b
r1 = r2 = r̄ = − .
2a
We have one solution y1 = er̄t . How can we find the second solution which is linearly inde-
pendent of y1 ?
From experience in an earlier example, we claim that y2 = ter̄t is a solution. To prove
this claim, we plug it back into the equation. If r̄ is the double root, then, the characteristic
equation can be written
r 2 − 2r̄r + r̄ 2 = 0
which gives the equation
y ′′ − 2r̄y ′ + r̄ 2 y = 0.

54
We can check if y2 satisfies this equation. We have

y ′ = er̄t + r̄ter̄t , y ′′ = 2r̄er̄t + r̄ 2 ter̄t .

Put into the equation, we get

2r̄er̄t + r̄ 2 ter̄t − 2r̄(er̄t + r̄ter̄t ) + r̄ 2 ter̄t = 0.

Finally, we must make sure that y1 , y2 are linearly independent. We compute their Wronskian

W (y1 , y2 ) = y1 y2′ − y1′ y2 = er̄t (er̄t + r̄ter̄t ) − r̄er̄t ter̄t = e2r̄t 6= 0.

We conclude now, the general solution is

y(t) = c1 y1 + c2 y2 = c1 er̄t + c2 ter̄t = er̄t (c1 + c2 t).

Example 1. (not covered in class) Consider the equation y ′′ + 4y ′ + 4y = 0. We have

r2 + 4r + 4 = 0, and r1 = r2 = r = −2. So one solution is y1 = e−2t . What is y2 ?
Method 1. Use Wronskian and Abel’s Theorem. By Abel’s Theorem we have
Z
W (y1 , y2 ) = c exp(− 4 dt) = ce−4t = e−4t , (let c = 1).

By the definition of Wronskian we have

W (y1 , y2 ) = y1 y2′ − y1′ y2 = e−2t y2′ − (−2)e−2t y2 = e−2t (y2′ + 2y2 ).

They must equal to each other:

e−2t (y2′ + 2y2 ) = e−4t , y2′ + 2y2 = e−2t .

Solve this for y2 , Z

2t −2t
µ=e , y2 = e e2t e−2t dt = e−2t (t + C)

Let C = 0, we get y2 = te−2t , and the general solution is

y(t) = c1 y1 + c2 y2 = c1 e−2t + c2 te−2t .

Method 2. This is the textbook’s version. We guess a solution of the form y2 = v(t)y1 =
v(t)e−2t , and try to find the function v(t). We have

y2′ = v ′ e−2t + v(−2e−2t ) = e−2t (v ′ − 2v), y2′′ = e−2t (v ′′ − 4v ′ + 4v).

Put them in the equation

e−2t (v ′′ − 4v ′ + 4v) + 4e−2t (v ′ − 2v) + 4v(t)e−2t = 0.

Cancel the term e−2t , and we get v ′′ = 0, which gives v(t) = c1 t + c2 . So

y2 (t) = vy1 = (c1 t + c2 )e−2t = c1 te−2t + c2 e−2t .

55
Note that the term c2 e−2t is already contained in cy1 . Therefore we can choose c1 = 1, c2 = 0,
and get y2 = te−2t , which gives the same general solution as Method 1. We observe that this
method involves more computation than Method 1.
A typical solution graph is included below:

2.5

1.5

0.5

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

We see if c2 > 0, y increases for small t. But as t grows, the exponential (decay) function
dominates, and solution will go to 0 as t → ∞.

One can show that in general if one has repeated roots r1 = r2 = r, then y1 = ert and
y2 = tert , and the general solution is

y = c1 ert + c2 tert = ert (c1 + c2 t).

Example 2. Solve the IVP

y ′′ + 2y ′ + y = 0, y(0) = 2, y ′ (0) = 1.

Answer. This follows easily now

r 2 + 2r + 1 = 0, ⇒ r1 = r2 = −1, ⇒ y(t) = (c1 + c2 t)e−t .

The ICs give

y(0) = 2 : c1 + 0 = 2, ⇒ c1 = 2.
y ′ (t) = (−c1 + c2 t)e−t + c2 e−t , y ′ (0) = −c1 + c2 = 1, ⇒ c2 = 1 + c1 = 3.
So the solution is y(t) = (2 + 3t)et . How would the graph of y(t) look like? We have y → 0 as
t → ∞.

56
 Example 2b. If the initial conditions in Example 2 are changed into

y(2) = 2, y ′ (2) = 1,

it would be more convenient to write the general solution in the form

y(t) = c1 e−(t−2) + c2 (t − 2)e−(t−2) .

Note that we substitute t with t − 2 in the general solution form in Example 2. Now we have

y ′ (t) = −c1 e−(t−2) + c2 e−(t−2) − c2 (t − 2)e−(t−2) .

The initial conditions now give:

y(2) = c1 + 0 = 2, y ′ (2) = −c1 + c2 = 1, → c1 = 2, c2 = 3.

Note that the computation remains the same! The solution is

y(t) = [2 + 3(t − 2)]e−(t−2) .

Of course you can write the same general solution as in Example 2 and work out the detail.
Try that, and you will be convinced that it takes much more work!

Example 2c. Let’s look at even another way of solving Example 2b. Note that the
equation is autonomous, i.e., no t appears in the equation. Let the initial data be given at
t0 = 2. Introduce a new time variable τ = t − t0 = t − 2. Note this is just shift the time by 2
units into the future. We have dτ = dt, and the IVP becomes

y ′′ (τ ) + 2y ′ (τ ) + y(τ ) = 0, y(0) = 2, y ′ (0) = 1.

which is same as Example 2. We can therefore write out the solution

y(τ ) = (2 + 3τ )e−τ .

Plug back in τ = t − 2, we have the solution

y(t) = [2 + 3(t − 2)]e−(t−2) ,

which is of course the same as in Example 2b.

Summary: For ay ′′ + by ′ + cy = 0, and ar 2 + br + c = 0 has two roots r1 , r2 , we have

• If r1 6= r2 (real): y(t) = c1 er1 t + c2 er2 t ;

• If r1 = r2 = r̄ (real): y(t) = (c1 + c2 t)er̄t ;

• If r1,2 = λ ± iµ complex: y(t) = eλt (c1 cos µt + c2 sin µt).

57
 On reduction of order: This method can be used to find a second solution y2 if the first
solution y1 is given for a second order linear equation.

Example 3. For the equation

2t2 y ′′ + 3ty ′ − y = 0, t > 0,

given one solution y1 = 1t , find a second linearly independent solution.

Answer. Method 1: Use Abel’s Theorem and Wronskian. By Abel’s Theorem, and
choose C = 1, we have
 Z   Z   
3t 3
W (y1 , y2 ) = exp − p(t) dt = exp − dt = exp − ln t = t−3/2 .
2t2 2
By definition of the Wronskian,
1 ′ 1
W (y1 , y2 ) = y1 y2′ − y1′ y2 = y2 − (− 2 )y2 = t−3/2 .
t t
Solve this for y2 :
2√
Z 
1 1 12 3
Z
3
µ = exp dt = exp(ln t) = t, ⇒ y2 = t · t− 2 dt = t2 = t.
t t t3 3
√
Drop the constant 32 , we get y2 = t.

Method 2: This is the textbook’s version. Guess solution to be y2 (t) = y1 (t)v(t), and
find v(t) by solving a first order equation!
Comment 1: This method is more complicated in computation than the 1st method.
However, it is a more general method. In the case one does not have something like the Abel’s
Theorem, such a method will always work, and will always reduce the order by 1.
Comment 2: A similar situation occurs in finding roots (or factorizing) a polynomial. Let
Pn (t) be a polynomial of degree n. If we found that r1 is a root, then it has a factor t − r1 , so
Pn (t) = (t − r1 )Q(t), where Q(t) is a polynomial of degree n − 1.

Let’s introduce another method that combines the ideas from Method 1 and Method 2.
Method 3. We will use Abel’s Theorem, and at the same time we will seek a solution of
the form y1 = vy1 .
3
By Abel’s Theorem, we have ( worked out in M1) W (y1 , y2 ) = t− 2 . Now, seek y2 = vy1 .
By the definition of the Wronskian, we have

W (y1 , y2 ) = y1 y2′ − y1′ y2 = y1 (vy1 )′ − y1′ (vy1 ) = y1 (v ′ y1 + vy1′′ ) − vy1 y1′ = v ′ y12 .

Note that this is a general formula:

W (y1 , y2 ) = v ′ y12 , if y2 = vy1 .

Now putting y1 = 1/t, we get

1 2 3
Z
3 1 1
′
v 2 = t− 2 , ′
v =t , 2 v= t 2 dt = t2 .
t 3

58
Drop the constant 32 , we get
3 1 1
y2 = vy1 = t 2 = t2 .
t
We see that Method 3 is the most efficient one among all three methods. We will focus on
this method from now on.

Example 4. Consider the equation

t2 y ′′ − t(t + 2)y ′ + (t + 2)y = 0, (t > 0).
Given y1 = t, find the general solution.

Answer. We have
t(t + 2) t+2 2
p(t) = − 2
=− = −1 − .
t t t
Let y2 be the second solution. By Abel’s Theorem, choosing c = 1, we have
Z   
2
W (y1 , y2 ) = exp 1+ dt = exp{t + 2 ln t} = t2 et .
t
Let y2 = vy1 , the W (y1 , y2 ) = v ′ y12 = t2 v ′ . Then we must have
t2 v ′ = t2 et , v ′ = et , v = et , y2 = tet .
(A cheap trick to double check your solution y2 would be: plug it back into the equation and
see if it satisfies it.) The general solution is
y(t) = c1 y2 + c2 y2 = c1 t + c2 tet .
We observe here that Method 3 is very efficient.
√
3
Example 5. Given the equation t2 y ′′ − (t − 16 )y = 0, (t > 0), and y1 = t(1/4) e2 t ,
find y2 .

Answer. We will always use method 3. We see that p = 0. By Abel’s Theorem, setting
c = 1, we have Z 
W (y1 , y2 ) = exp 0dt = 1.
1 √
Seek y2 = vy1 . Then, W (y1 , y2 ) = y12 v ′ = t 2 e4 t v ′ . So we must have
1
√ 1
√ Z
1
√
4 t ′ ′ − −4 t
t 2 e v = 1, ⇒ v = t 2 e , ⇒ v = t− 2 e−4 t dt.
√ 1
Let u = −4 t, so du = −2t− 2 dt, we have
Z
1 1 1 √
v = − eu du = − eu = − e−4 t .
2 2 2
So drop the constant − 12 , we get
√ 1
√ 1
√
y2 = vy1 = e−4 t t 4 e2 t
= t 4 e−2 t .
The general solution is
1
 √ √ 
y(t) = c1 y1 + c2 y2 = t 4 c1 e2 t + c2 e−2 t .

59
3.5 Non-homogeneous equations; method of undetermined co-
efficients
Want to solve the non-homogeneous equation

y ′′ + p(t)y ′ + q(t)y = g(t), (N )

Steps:

1. First solve the homogeneous equation

y ′′ + p(t)y ′ + q(t)y = 0, (H)

i.e., find y1 , y2 , linearly independent of each other, and form the general solution

yH = c1 y1 + c2 y2 .

2. Find a particular/specific solution Y for (N), by MUC (method of undetermined coeffi-

cients);

3. The general solution for (N) is then

y = yH + Y = c1 y1 + c2 y2 + Y.

Find c1 , c2 by initial conditions, if given.

Key step: step 2.

Why y = yH + Y ?
A quick proof: If yH solves (H), then
′′ ′
yH + p(t)yH + q(t)yH = 0, (A)

and since Y solves (N), we have

Y ′′ + p(t)Y ′ + q(t)Y = g(t), (B)

Adding up (A) and (B), and write y = yH + Y , we get y ′′ + p(t)y ′ + q(t)y = g(t).
Main focus: constant coefficient case, i.e.,

ay ′′ + by ′ + cy = g(t).

Example 1. Find the general solution for y ′′ − 3y ′ − 4y = 3e2t .

Answer. Step 1: Find yH .

r 2 − 3r − 4 = (r + 1)(r − 4) = 0, ⇒ r1 = −1, r2 = 4,

so
yH = c1 e−t + c2 e4t .

60
Step 2: Find Y . We guess/seek solution of the same form as the source term Y = Ae2t , and
will determine the coefficient A.
Y ′ = 2Ae2t , Y ′′ = 4Ae2t .
Plug these into the equation:
1
4Ae2t − 3 · 2Ae2t − 4Ae2t = 3e2t , ⇒ −6A = 3, ⇒ A=− .
2
So Y = − 12 e2t .
Step 3. The general solution to the non-homogeneous solution is
1
y(t) = yH + Y = c1 e−t + c2 e4t − e2t .
2
Observation: The particular solution Y take the same form as the source term g(t).
But this is not always true.

Example 2. Find general solution for y ′′ − 3y ′ − 4y = 2e−t .

Answer. The homogeneous solution is the same as Example 1: yH = c1 e−t +c2 e4t . For the
particular solution Y , let’s first try the same form as g, i.e., Y = Ae−t . So Y ′ = −Ae−t , Y ′′ =
Ae−t . Plug them back in to the equation, we get
LHS = Ae−t − 3(−Ae−t ) − 4Ae−t = 0 6= 2e−et = RHS.
So it doesn’t work. Why?
We see r1 = −1 and y1 = e−t , which means our guess Y = Ae−t is a solution to the
homogeneous equation. It will never work.
Second try: Y = Ate−t . So
Y ′ = Ae−t − Ate−t , Y ′′ = −Ae−t − Ae−t + Ate−t = −2Ae−t + Ate−t .
Plug them in the equation
(−2Ae−t + Ate−t ) − 3(Ae−t − Ate−t ) − 4Ate−t = −5Ae−t = 2e−t ,
we get
2
−5A = 2, ⇒ A=− ,
5
so we have Y = − 52 te−t .

Summary 1. If g(t) = aeαt , then the form of the particular solution Y depends on r1 , r2
(the roots of the characteristic equation).

case form of the particular solution Y

r1 6= α and r2 6= α Y = Aeαt

r1 = α or r2 = α, but r1 6= r2 Y = Ateαt

r1 = r2 = α Y = At2 eαt

61
 Example 3. Find the general solution for

y ′′ − 3y ′ − 4y = 3t2 + 2.

Answer. The yH is the same yH = c1 e−t + c2 e4t .

Note that g(t) is a polynomial of degree 2. We will try to guess/seek a particular solution
of the same form:

Y = At2 + Bt + C, Y ′ = 2At + B, Y ′′ = 2A

Plug back into the equation

2A − 3(2At + b) − 4(At2 + Bt + C) = −4At2 − (6A + 4B)t + (2A − 3B − 4C) = 3t2 + 2.

Compare the coefficient, we get three equations for the three coefficients A, B, C:
3
−4A = 3 → A=−
4
9
−(6A + 4B) = 0, → B =
8
1 55
2A − 3B − 4C = 2, → C = (2A − 3B − 2) = −
4 32

So we get
3 9 55
Y (t) = − t2 + t − .
4 8 32

But sometimes this guess won’t work.

Example 4. Find the particular solution for y ′′ − 3y ′ = 3t2 + 2.

Answer. We see that the form we used in the previous example Y = At2 + Bt + C won’t
work because Y ′′ − 3Y ′ will not have the term t2 .
New try: multiply by a t. So we guess Y = t(At2 + Bt + C) = At3 + Bt2 + Ct. Then

Y ′ = 3At2 + 2Bt + C, Y ′′ = 6At + 2B.

Plug them into the equation

(6At + 2B) − 3(3At2 + 2Bt + C) = −9At2 + (6A − 6B)t + (2B − 3C) = 3t2 + 2.

Compare the coefficient, we get three equations for the three coefficients A, B, C:
1
−9A = 3 → A=−
3
1
(6A − 6B) = 0, → B = A = −
3
1 8
2B − 3C = 2, → C = (2B − 2) = −
3 9

62
So Y = t(− 31 t2 − 13 t − 98 ).

Summary 2. If g(t) is a polynomial of degree n, i.e.,

g(t) = αn tn + · · · + α1 t + α0

the particular solution for

ay ′′ + by ′ + cy = g(t)
(where a 6= 0) depends on b, c:

case form of the particular solution Y

c 6= 0 Y = Pn (t) = An tn + · · · + A1 t + A0
c = 0 but b 6= 0 Y = tPn (t) = t(An tn + · · · + A1 t + A0 )
c = 0 and b = 0 Y = t2 Pn (t) = t2 (An tn + · · · + A1 t + A0 )

Example 5. Find a particular solution for

y ′′ − 3y ′ − 4y = sin t.

Answer. Since g(t) = sin t, we will try the same form. Note that (sin t)′ = cos t, so we
must have the cos t term as well. So the form of the particular solution is

Y = A sin t + B cos t.

Then
Y ′ = A cos t − B sin t, Y ′′ = −A sin t − B cos t.
Plug back into the equation, we get

(−A sin t − B cos t) − 3(A cos t − B sin t) − 4(A sin t + b cos t)

= (−5A + 3B) sin t + (−3A − 5B) cos t = sin t.

So we must have
5 3
−5A + 3B = 1, −3A − 5B = 0, → A= , B= .
34 34
So we get
5 3
Y (t) = − sin t + cos t.
34 34

We observe that: (1). If the right-hand side is g(t) = a cos t, then the same form would
work; (2). More generally, if g(t) = a sin t + b sin t for some a, b, then the same form still work.
However, this form won’t work if it is a solution to the homogeneous equation.

Example 6. Find a general solution for y ′′ + y = sin t.

63
 Answer. Let’s first find yH . We have r 2 + 1 = 0, so r1,2 = ±i, and yH = c1 cos t + c2 sin t.
For the particular solution Y : We see that the form Y = A sin t + B cos t won’t work
because it solves the homogeneous equation.
Our new guess: multiply it by t, so

Y (t) = t(A sin t + B cos t).

Then
Y ′ = (A sin t + B cos t) + t(A cos t + B sin t),
Y ′′ = (−2B − At) sin t + (2A − Bt) cos t.
Plug into the equation
1
Y ′′ + Y = −2B sin t + 2A cos t = sin t, ⇒ A = 0, B = −
2
So
1
Y (y) = − t cos t.
2
The general solution is
1
y(t) = yH + Y = c1 cos t + c2 sin t − t cos t.
2

Summary 3. If g(t) = a sin αt + b cos αt, the form of the particular solution depends on
the roots r1 , r2 .

case form of the particular solution Y

(1). r1,2 6= ±αi Y = A sin αt + B cos αt
(2). r1,2 = ±αi Y = t(A sin αt + B cos αt)
Note that case (2) occurs when the equation is y ′′ + α2 y = a sin αt + b cos αt.

We now have discovered some general rules to obtain the form of the particular solution
for the non-homogeneous equation ay ′′ + by ′ + cy = g(t).

• Rule (1). Usually, Y take the same form as g(t);

• Rule (2). Except, if the form of g(t) provides a solution to the homogeneous equation.
Then, one can multiply it by t.

• Rule (3). If the resulting form in Rule (2) is still a solution to the homogeneous equation,
then, multiply it by another t.

Next we study a couple of more complicated forms of g.

Example 7. Find a particular solution for

y ′′ − 3y ′ − 4y = tet .

64
 Answer. We see that g = P1 (t)eat , where P1 is a polynomial of degree 1. Also we see
r1 = −1, r2 = 4, so r1 6= a and r2 6= a. For a particular solution we will try the same form as
g, i.e., Y = (At + B)et . So

Y ′ = Aet + (At + b)et = (A + b)et + Atet ,

Y ′′ = · · · = (2A + B)et + Atet .

Plug them into the equation,

[(2A + B)et + Atet ] − 3[(A + b)et + Atet ] − 4(At + B)et = (−6At − A − 6B)et = tet .

We must have −6At − A − 6B = t, i.e.,

1 1 1 1
−6A = 1, −A − 6B = 0, ⇒ A = − ,B = , ⇒ Y = (− t + )et .
6 36 6 36

However, if the form of g is a solution to the homogeneous equation, it won’t work for a
particular solution. We must multiply it by t in that case.

Example 8. Find a particular solution of

y ′′ − 3y ′ − 4y = te−t .

Answer. Since a = −1 = r1 , so the form we used in Example 7 won’t work here. (Can
you intuitively explain why?) Try a new form now

Y = t(At + B)e−t = (At2 + Bt)e−t .

Then
Y ′ = · · · = [−At2 + (2A − B)t + B]e−t ,
Y ′′ = · · · = [At2 + (B − 4A)t + 2A − 2B]e−t .
Plug into the equation

[At2 + (B − 4A)t + 2A − 2B]e−t − 3[−At2 + (2A − B)t + B]e−t − 4(At2 + Bt)e−t

= [−10At + 2A − 5B]e−t = tet .

So we must have −10At + 2A − 5B = t, which means

1 1
−10A = 1, 2A − 5B = 0, ⇒ A=− , B=− .
10 25
Then  
1 2 1
Y = − t − t e−t .
10 25

Summary 4. If g(t) = Pn (t)eat where Pn (t) = αn tn + · · · + α1 t + α0 is a polynomial of

degree n, then the form of a particular solution depends on the roots r1 , r2 .

65
 case form of the particular solution Y

r1 6= a and r2 6= a Y = P̃n (t)eat = (An tn + · · · + A1 t + A0 )eat

r1 = a or r2 = a but r1 6= r2 Y = tP̃n (t)eat = t(An tn + · · · + A1 t + A0 )eat

r1 = r2 = a Y = t2 P̃n (t)eat = t2 (An tn + · · · + A1 t + A0 )eat

Other cases of g are treated in a similar way: Check if the form of g is a solution to the
homogeneous equation. If not, then use it as the form of a particular solution. If yes, then
multiply it by t or t2 .
We summarize a few cases below.
Summary 5. If g(t) = eαt (a cos βt + b sin βt), and r1 , r2 are the roots of the characteristic
equation. Then
case form of the particular solution Y
r1,2 6= α ± iβ Y = eαt (A cos βt + B sin βt)
r1,2 = α ± iβ Y = t · eαt (A cos βt + B sin βt)

Summary 6. If g(t) = Pn (t) cos βt + P̃n (t) sin βt) where Pn (t) and P̃n (t) are polynomials
of degree n, and r1 , r2 are the roots of the characteristic equation. Then
case form of the particular solution Y
r1,2 6= ±iβ Y = (An t + · · · + A0 ) cos βt + (Bn tn + · · · + B0 ) sin βt
n

r1,2 = α ± iβ Y = t · [(An tn + · · · + A0 ) cos βt + (Bn tn + · · · + B0 ) sin βt]

Summary 7. If g(t) = Pn (t)eαt (a cos βt + b sin βt) where Pn (t) is a polynomial of degree
n, and r1 , r2 are the roots of the characteristic equation. Then
case form of the particular solution Y
r1,2 6= α ± iβ Y = e [(An tn + · · · + A0 ) cos βt + (Bn tn + · · · + B0 ) sin βt]
αt

r1,2 = α ± iβ Y = t · eαt [(An tn + · · · + A0 ) cos βt + (Bn tn + · · · + B0 ) sin βt]

More terms in the source. If the source g(t) has several terms, we treat each separately
and add up later. Let g(t) = g1 (t) + g2 (t) + · · · gn (t), then, find a particular solution Yi for
each gi (t) term as if it were the only term in g, then Y = Y1 + Y2 + · · · Yn . This claim follows
from the principle of superposition. (Can you provide a brief proof?)

In the examples below, we want to write the form of a particular solution.

Example 9. y ′′ − 3y ′ − 4y = sin 4t + 2e4t + e5t − t.

Answer. Since r1 = −1, r2 = 2, we treat each term in g separately and the add up:

Y (t) = A sin 4t + B cos 4t + Cte4t + De5t + (Et + F ).

66
 Example 10. y ′′ + 16y = sin 4t + cos t − 4 cos 4t + 4.

Answer. The char equation is r 2 + 16 = 0, with roots r1,2 = ±4i, and

yH = c1 sin 4t + c2 cos 4t.

We also note that the terms sin 4t and −4 cos 4t are of the same type, and we must multiply
it by t. So
Y = t(A sin 4t + B cos 4t) + (C cos t + D sin t) + E.

Example 11. y ′′ − 2y ′ + 2y = et cos t + 8et sin 2t + te−t + 4e−t + t2 − 3.

Answer. The char equation is r 2 − 2r + 2 = 0 with roots r1,2 = 1 ± i. Then, for the term
et cos t we must multiply by t.

Y = tet (A1 cos t + A2 sin t)+ et (B1 cos 2t + B2 sin 2t)+ (C1 t + C0 )e−t + De−t + (F2 t2 + F1 t + F0 ).

3.6 Variation of Parameters*

This is a general method to find a particular solution for the non-homogeneous equation, 2nd
order and linear. Consider
y ′′ + p(t)y ′ + q(t) = 0, (H)
and
y ′′ + p(t)y ′ + q(t) = g(t), (N )
Assume that we have found the general solution for (H), given by

yH (t) = c1 y1 (t) + c2 y2 (t),

where c1 , c2 are arbitrary constants.

We now guess that a particular solution for (N) as

yp (t) = u1 (t)y1 (t) + u2 (t)y2 (t).

Our goal is to find some functions u1 , u2 that will make yp a particular solution.
Note that we have a lot of freedom here. If we plug in yp in (N), we will get one constraint.
But we have 2 functions. This means there is one constraint that is free-choice for us! We
should use it wisely, to get 1st order equations for u1 , u2 .
We compute the derivative of yp

yp′ = u′1 y1 + u1 y1′ + u′2 y2 + u2 y2′ .

If we differentiate one more time, the expression gets large. The term with u′1 and u′2 would
give u′′1 , u′′2 , which we should avoid. We see it is a good place to use our free choice and require

y1 u′1 + y2 u′2 = 0. (A)

Then, we have

yp′ = y1′ u1 + y2′ + u2 , yp′′ = y1′′ u1 + y2′′ u2 + y1′ u′1 + y2′ u′2 .

67
Plug into (N), we get
h i h i h i
y1′′ u1 + y2′′ u2 + y1′ u′1 + y2′ u′2 + p(t) y1′ u1 + y2′ + u2 + q(t) y1 u1 + y2 u2 = g(t)

Cleaning up, we get

h i h i
y1′′ + p(t)y1′ + q(t)y1 u1 + y2′′ + p(t)y2′ + q(t)y2 u2 + [y1′ u′1 + y2′ u′2 ] = g(t)

Since y1 , y2 are solutions of (H), the two brackets are 0, and we get

y1′ u′1 + y2′ u′2 = g(t). (B)

Note now (A) and (B) are two linear equations for the unknowns u′1 , u′2 . Write it into
matrix-vector form:   ′   
y1 y2 u1 0
=
y1′ y2′ u′2 g(t)
Note the determinant of the coefficient matrix is W (y1 , y2 ) 6 0, which implies unique solution
for u′1 , u′2 :  ′  ′  
u1 1 y2 −y2 0
=
u′2 W −y1′ y1 g(t)
which gives
1 1
u′1 = − y2 (t)g(t), u′2 = y1 (t)g(t). (C)
W W
We recover u1 , u2 by integrating
1 1
Z Z
u1 (t) = − y2 (t)g(t) dt, u2 (t) = y1 (t)g(t)dt.
W (t) W (t)
The particular solution is
1 1
Z Z
yp (t) = u1 y1 + u2 y2 = −y1 (t) y2 (t)g(t) dt + y2 (t) y1 (t)g(t)dt.
W (t) W (t)

Remark 1: In general, finding the general solution for (H) is not trivial. We don’t have a
general algorithm yet.
Remark 2: However, if one can find one solution of (H), call it y1 , by reduction of order
we can find y2 . Then, by variation of parameter, we can find yp , and therefore the general
solution of (N).

Example Find the general solution for

y ′′ − 2y ′ + y = et ln t, t > 0.

Answer. We first find the two linearly independent solutions y1 , y2 for the homogeneous
equation. Note that it has constant coefficients. The char eqn r 2 − 2r + 1 = 0 gives two
repeated roots r1 = r2 = 1, so
y1 = et , y2 = tet .

68
Then
y1′ = et , y2′ = (1 + t)et , W (y1 , y2 ) = (1 + t)e2t − te2t = e2t .
Write the particular solution as yp = u1 y1 + u2 y2 , by (C) we have

tet et ln t
u′1 = − = −t ln t, u′2 = ln t
e2t
which gives

t2 t2
Z Z
u1 (t) = − t ln tdt = − ln t + , u2 (t) = ln tdt = t ln t − t.
2 4

(by integration-by-parts). We get

1 h 3i
yp = t2 et ln t −
2 2
The general solution is
1 h 3i
y(t) = c1 y1 + c2 y2 + yp (t) = c1 et + c2 tet + t2 et ln t − .
2 2

Example Given that y1 (t) = t is a solution of the homogeneous equation

t2 y ′′ − ty ′ + y = 0, t>0

find the solution for the IVP

t2 y ′′ − ty ′ + y = t, y(1) = 1, y ′ (1) = 4.

Answer. . Sketch: by reduction of order, we find y2 = t lnt (see previous example).

By variation of parameter, we get yp = 21 t(ln t)2 (details...)
General solution
1
y(t) = c1 t + c2 t ln t + t(ln t)2 .
2
Find the constants c1 , c2 by initial conditions (details...)

c1 = 1, c2 = 3,

so the solution to the IVP is

1
y(t) = t + 3t ln t + t(ln t)2 .
2

69
3.7 Mechanical Vibrations
In this chapter we study some applications of the IVP

ay ′′ + by ′ + cy = g(t), y(0) = y0 , y ′ (0) = ȳ0 .

The spring-mass system: See figure below.

(A) (B) (C)

l l l+L

extra
stretch

Figure (A): a spring in rest, with length l.

Figure (B): we put a mass m on the spring, and the spring is stretched. We call length L
the elongation.
Figure (C): The spring-mass system is set in motion by stretch/squeeze it extra, with
initial velocity, or with external force.
Force diagram at equilibrium position: mg = Fs .
Fs

mg
Hooke’s law: Spring force Fs = −kL, where L =elongation and k =spring constant.
So: we have mg = kL which give
mg
k=
L
which gives a way to obtain k by experiment: hang a mass m and measure the elongation L.
Model the motion: Let u(t) be the displacement/position of the mass at time t, assuming
the origin u = 0 is at the equilibrium position, and downward is the positive direction.
Total elongation: L + u
Total spring force: Fs = −k(L + u)
Other forces:
* damping/resistent force: Fd (t) = −γv = −γu′ (t), where γ is the damping constant, and v

70
is the velocity
* External force applied on the
P mass: F (t), given function of t
Total force on the mass: f =Pmg + Fs + Fd + F .
Newton’s law of motion ma = f gives
X
ma = mu′′ = f = mg + Fs + Fd + F, mu′′ = mg − k(L + u) − γu′ + F.

Since mg = kL, by rearranging the terms, we get

mu′′ + γu′ + ku = F

where m ia the mass, γ is the damping constant, k is the spring constant, and F is the external
force.

Next we study several cases.

Case 1: Undamped free vibration (simple harmonic motion). We assume no damping
(γ = 0) and no external force (F = 0). So the equation becomes

mu′′ + ku = 0.

Solve it
r r
2 2 k k k
mr + k = 0, r =− , r1,2 = ± i = ±ω0 i, where ω0 = .
m m m
General solution
u(t) = c1 cos ω0 t + c2 sin ω0 t.
Four terminologies of this motion: frequency, period, amplitude and phase, defined
below. r
k
Frequency: ω0 =
m
2π
Period: T =
ω0
Amplitude and phase: We need to work on this a bit. We can write
!
c1 c2
q
2 2
u(t) = c1 + c2 p 2 cos ω0 t + p 2 sin ω0 t .
c1 + c22 c1 + c22

Now, define δ, such that tan δ = c2 /c1 , then

c2 c1
sin δ = p , cos δ = p
c21 + c22 c21 + c22

so we have
q q
u(t) = c21 + c22 (cos δ · cos ω0 t + sin δ · sin ω0 t) = c21 + c22 cos(ω0 t − δ).
p c2
So amplitude is R = c21 + c22 and phase is δ = arctan .
c1

71
 A trick to memorize the last term formula: Consider a right triangle, with c1 and c2 as the
sides that form the right angle. Then, the amplitude equals to the length of the hypotenuse,
and the phase δ is the angle between side c1 and the hypotenuse. Draw a graph and you will
see it better.

A few words on units:

force (f ) weight (mg) length (u) mass (m) gravity (g)
lb lb ft lb · sec2 /ft 32 ft/sec2
newton newton m kg 9.8 m/sec2

Problems in this part often come in the form of word problems. We need to learn the skill
of extracting information from the text and put them into mathematical terms.

Example 1. A mass weighing 10 lb stretches a spring 2 in. We neglect damping. If the

mass is displaced an additional 2 in, and is then set in motion with initial upward velocity of
1 ft/sec, determine the position, frequency, period, amplitude and phase of the motion.

Answer. We see this is free harmonic oscillation. The equation is

mu′′ + ku = 0

with initial conditions (Pay attention to units!)

1
u(0) = 2in = ft, u′ (0) = −1.
6
We need find the values for m and k. We have
10 10 5
mg = 10, g = 32, m= = = .
g 32 16
To find k, we see that the elongation is L = 2in = 16 ft if the mass m = 5
16 . By Hooke’s law,
we have
kL = mg, k = mg/L = 60.
Put in these values, we get the equation
1
u′′ + 192u = 0, u(0) = , u′ (0) = −1.
6
√
So the frequency is ω0 = 192, and the general solution is

u(t) = c1 cos ω0 t + c2 sin ω0 t

We can find c1 , c2 by the ICs:

1 1 1
u(0) = c1 = , u′ (0) = ω0 c2 = −1, c2 = − = −√ .
6 ω0 192
(Note that c1 = u(0) and c2 = u′ (0)/ω0 .) Now we have the position at any time t
1 1
u(t) = cos ω0 t − √ sin ω0 t.
6 192

72
The four terms of the motion are
r
√ 2π π
q
19
ω0 = 192, T = =√ , R= c21 + c22 = ≈ 0.18,
ω0 48 576

and √
c2 6 3
δ = arctan = arctan − √ = − arctan .
c1 192 4

Case II: Damped free vibration. We assume that γ 6= 0(> 0) and F = 0.

mu′′ + γu′ + ku = 0

then p
2 −γ ± γ 2 − 4km
mr + γr + k = 0, r1,2 = .
2m
We see the type of root depends on the sign of the discriminant ∆ = γ 2 − 4km.
√
• If ∆ > 0, (i.e., γ > 4km, large damping,) we have two real roots, and they are both
negative. The general solution is u = c1 er1 t + c2 er2 t , with r1 < 0, r2 < 0.
Due to the large damping force, there will be no vibration in the motion. The mass will
simply return to the equilibrium position exponentially. This kind of motion is called
overdamped.
√
• If ∆ = 0, (i.e., γ = 4km) we have double roots r1 = r2 = r < 0. So u = (c1 + c2 t)ert .
Depending on the sign of c1 , c2 (which is determined by the ICs), the mass may cross the
equilibrium point maximum once. This kind of motion is called critically damped,
and this value of γ is called critical damping.
√
• If ∆ < 0, (i.e., γ < 4km, small damping) we have complex roots
p
γ 4km − γ 2
r1,2 = −λ ± µi, λ= , µ= .
2m 2m
So the position function is

u(t) = e−λt (c1 cos µt + c2 sin µt) .

This motion is called damped oscillation. We can re-write it as

c2
q
u(t) = e−λt R · cos(µt − δ), R = c21 + c22 , δ = arctan .
c1

Here the term e−λt R is the amplitude, and µ is called the quasi frequency, and the quasi
period is 2π
µ . The graph of the solution looks like the one for complex roots with negative
real part.

73
 Summary: For all cases, since the real part of the roots are always negative, u will go to
zero as t grow. This means, if there is damping, no matter how big or small, the motion will
eventually come to a rest.

Example 2. A mass of 1 kg is hanging on a spring with k = 1. The mass is in a medium

that exerts a viscous resistance of 6 newton when the mass has a velocity of 48 m/sec. The
mass is then further stretched for another 2m, then released from rest. Find the position u(t)
of the mass.
6
Answer. We have γ = 48 = 81 . So the equation for u is

1
mu′′ + γu′ + ku = 0, u′′ + u′ + u = 0, u(0) = 2, u′ (0) = 0.
8
Solve it √ √
2 1 1 255 255
r + r + 1 = 0, r1,2 =− ± i, ω0 =
8 16 16 16
1
u(t) = e− 16 t (c1 cos ω0 t + c2 sin ω0 t) .
By ICs, we have u(0) = c1 = 2, and
1 1
u′ (t) = − u(t) + e− 16 t (−ω0 c1 sin ω0 t + ω0 c2 cos ω0 t),
16
1 2
u′ (0) = − u(0) + ω0 c2 = 0, c2 = √ .
16 255
So the position at any time t is
 
−t/16 2
u(t) = e 2 cos ω0 t − √ sin ω0 t .
255

3.8 Forced Vibrations

In this chapter we assume the external force is F (t) = F0 cos ωt. (The case where F (t) =
F0 sin ωt is totally similar.)
The reason for this particular choice of force will be clear later when we learn Fourier
series, i.e., we represent periodic functions with the sum of a family of sin and cos functions.

Case 1: With damping.

mu′′ + γu′ + ku = F0 cos ωt.

Solution consists of two parts:

u(t) = uH (t) + U (t),
uH (t): the solution of the homogeneous equation,
U (t): a particular solution for the non-homogeneous equation.
From discussion is the previous chapter, we know that uH → 0 as t → +∞ for systems
with damping. Therefore, this part of the solution is called the transient solution.

74
 The appearance of U is due to the force term F . Therefore it is called the forced response.
The form of this particular solution is U (t) = A1 cos ωt + A2 sin ωt. As we have seen, we can
rewrite it as U (t) = R cos(ωt − δ) where R is the amplitude and δ is the phase. We see it is a
periodic oscillation for all time t.
As time t → ∞, we have u(t) → U (t). So U (t) is called the steady state.

Case 2: Without damping. The equation now is

mu′′ + ku = F0 cos ωt
p
Let w0 = k/m denote the system frequency (i.e., the frequency for the free oscillation). The
homogeneous solution is
uH (t) = c1 cos ω0 t + c2 sin ω0 t.
The form of the particular solution depends on the value of w. We have two cases.
Case 2A: if w 6= w0 . The particular solution is of the form

U = A cos wt.

(Note that we did not take the sin wt term, because there is no u′ term in the equation.) And
U ′′ = −w2 A cos wt. Plug these in the equation

m(−w2 A cos wt) + kA cos wt = F0 cos wt,

F0 F0 F0
(k − mw2 )A = F0 , A= 2
= 2
= 2 .
k − mw m(k/m − w ) m(w0 − w2 )
Note that if w is close to w0 , then A takes a large value.
General solution
u(t) = c1 cos w0 t + c2 sin w0 t + A cos wt
where c1 , c2 will be determined by ICs.
Now, assume ICs:
u(0) = 0, u′ (0) = 0.
Let’s find c1 , c2 and the solution:

u(0) = 0 : c1 + A = 0, c1 = −A

u′ (0) = 0 : 0 + w0 c2 + 0 = 0, c2 = 0
Solution
u(t) = −A cos w0 t + A cos wt = A(cos wt − cos w0 t).
We see that the solution consists of the sum of two cosine functions, with different frequen-
cies. In order to have a better idea of how the solution looks like, we apply some manipulation.
Recall the trig identity:
b−a a+b
cos a − cos b = 2 sin sin .
2 2
We now have
w0 − w w0 + w
u(t) = 2A sin t · sin t.
2 2

75
Since both w0 , w are positive, then w0 + w has larger value than w0 − w. Then, the first term
2A sin w02−w t can be viewed as the varying amplitude, and the second term sin w02+w t is the
vibration/oscillation.
One particular situation of interests: if w0 6= w but they are very close wo ≈ w, then we
have |w0 − w| << |w0 + w|, meaning that |w0 − w| is much smaller than |w0 + w|. The plot
of u(t) looks like (we choose w0 = 9, w = 10)

1.5

0.5

−0.5

−1

−1.5

−2
0 5 10 15 20 25

This is called a beat. (One observes it by hitting a key on a piano that’s not tuned, for
example.)

Case 2B: If w = w0 . The particular solution is

U = At cos w0 t + Bt sin w0 t

A typical plot looks like:

76
 5

−1

−2

−3

−4

−5
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

This is called resonance. If the frequency of the source term ω equals to the frequency of
the system ω0 , then, small source term could make the solution grow very large!
One can bring down a building or bridge by small periodic perturbations.
Historical disasters such as the French troop marching over a bridge and the bridge col-
lapsed. Why? Unfortunate for the French, the system frequency of the bridge matches the
frequency of their foot-steps.

Summary:

• With damping:
Transient solution uH plus the forced response term U (t) (steady state),

• Without damping:
if w = w0 : resonance.
if w 6= w0 but w ≈ w0 : beat.

77
Chapter 4

Higher Order Linear Equations

4.1 General Theory of n-th Order Linear Equations

The general form of a linear equation of n-th order, with y(t) as the unknown, is

y (n) + pn−1 (t)y (n−1) + · · · + p1 (t)y ′ + p0 (t)y = g(t). (A)

Let L denote the linear differential operator

˙ (n) + pn−1 (t)y (n−1) + · · · + p1 (t)y ′ + p0 (t)y,

L(y)=y L(y) = g(t).

We need to assign n initial/boundary conditions. Let t0 be the initial time. Normally, the
lower derivatives are given at t0 , i.e.,

y(t0 ) = yo , y ′ (t0 ) = yo′ , · · · , y (n−1) (t0 )yo(n−1) .

Theoretical aspects are very similar to those for 2nd order linear equations, with suitable
extensions.
Existence and Uniqueness. If the coefficient functions p0 (t), p1 (t), · · · , pn−1 (t) are con-
tinuous and bounded on an open interval I containing t0 , then equation (A) has a uniqueness
solution on the interval I.
Typical problem types: Find the largest interval where solution is valid.
Linear dependency of n functions: The Wronskian determinant for n functions (y1 , y2 , · · · , yn )
is defined as  
 y1 y2 ... yn 
 y1′ y2′ yn′

... 
W (y1 , y2 , · · · , yn ) =  .. .. ..
 


 (n−1). . . 
(n−1) (n−1)

 y y . . . y 
1 2 n

The determinant is lengthy to compute for general n × n matrices. The simpler cases are
when n = 2 and n = 3, which we recall here
 
   a11 a12 a13 
 a b 
 a21 a22 a23  = a11 a22 a33 + a21 a32 a13 + a31 a12 a23
 
 c d  = ad − bc,
 

 a31 a32 a33 
 −a31 a22 a13 − a21 a12 a33 − a11 a32 a23 .

78
 One can use the Wronskian determinant to check if a set of functions are linearly dependent
or not.

Homogeneous equations when g(t) ≡ 0: There are n solutions, y1 , y2 , · · · yn , linearly

independent, with W (y1 , y2 , · · · , yn ) 6= 0 , that forms a set of fundamental solutions, whose
linear combination gives the general solution
yH (t) = c1 y1 + c2 y2 + · · · + cn yn .
Here the constants c1 , c2 , · · · , cn are determined by the n initial conditions.

Remark: We observe now how the property W 6= 0 leads to unique solutions for the
constants c1 , · · · , cn . Pugging in the n initial conditions, we have
c1 y1 (t0 ) + c2 y2 (t0 ) + · · · + cn yn (t0 ) = yo
c1 y1′ (t0 ) + c2 y2′ (t0 ) + · · · + cn yn′ (t0 ) = yo′
..
.
(n−1) (n−1)
c1 y1 (t0 ) + c2 y2 (t0 ) + · · · + cn yn(n−1) (t0 ) = yo(n−1)

Writing it into matrix-vector form, we get

y1 (t0 ) y2 (t0 ) ··· yn (t0 ) yo
    
c1
 y1′ (t0 ) ′
y2 (t0 ) ··· yn′ (t0 )     ′ 
  c2   yo 
..   ..  =  .. 


 .  .   . 
(n−1) (n−1) (n−1) (n−1)
y1 (t0 ) y2 (t0 ) · · · yn (t0 ) cn yo
The system has a uniqueness solution if the determinant of the coefficient matrix is not 0, i.e.,
W (y1 , · · · , yn ) 6= 0.

Abel’s Theorem.* Let y1 (t), · · · , yn (t) be n linearly independent solutions to L(y) = 0,

i.e.,
y (n) + pn−1 (t)y (n−1) + · · · + p1 (t)y ′ + p0 (t)y = 0, a<t<b
Let W (t) be the Wronskian of y1 .·, yn . Then, W (t) is a solution of the first order linear
equation
n Z o
′
W − pn−1 (t)W = 0, W (t) = C exp − pn−1 (t) dt , a < t < b.

The proof could be found in most advanced texts on differential equations. The basic argument
is similar to that of the case n = 2. The computations are more involved, however, since one
needs to compute the derivatives of an (n × n) determinant of functions. Here is a brief proof.
By product rule, the derivative of the determinant is:
 ′  
 y1 y2′ ··· yn′   y1 y2 ··· yn   y1 y · · · y
  
2 n

 
 ′
 y1 ′ ′   ′′ ′′ ′′ 
y1 ′ y2′ ··· ′
yn 
y 2 · · · y n
  y1 y 2 · · · y n
 

 ′′
 y1 y2′′ ··· yn′′   y1′′
 
y2′′ ··· yn′′ 
  y ′′ y2′′ ··· yn′′ 
′
 1
W = y
 ′′′ ′′′
y2 ··· ′′′ +
yn   y1 ′′′ ′′′
y2 ··· ′′′
yn  +· · ·+  ..
 . 
 1 
 ..   ..  
 (n−2) (n−2) (n−2) 

 .   .  y y · · · y n
 1 (n) 2
 (n−1) (n−1)  
(n−1)   (n−1) (n−1) (n−1) 
  
y (n) (n) 
1 y2 · · · yn y1 y2 · · · yn  y1 y2 ··· yn 

79
Here in the first matrix we differentiate the first row, in the the second matrix we differentiate
the second row and so on. There are totally n terms. We observe now that all the terms,
except the last, have two identical rows in the matrix. Then, their determinants are all 0,
except the last term. Therefore, we have
 
 y1 y2 ··· yn 

 y′ y2′ ··· yn′ 
 1
 y ′′ ′′
y2 ··· yn′′ 
′
 1
W =  .. . (A)
 .


 (n−2) (n−2) (n−2) 
y1 y2 · · · yn 
 (n) (n) (n) 
 y1 y2 ··· yn

Now, since yi ’s are solutions, we have

(n) (n−1) (n−2)
yi = −pn−1 yi − pn−2 yi − · · · − p1 yi′ − po yi

Now, multiply first row by p0 , the second row by pi , and so on, and the (n − 1)-th row by
pn−2 , and add them all to the last row. Remember that this operation does not change the
(n−1)
determinant. Then the last row becomes −pn−1 yi for the i-th entry. We now have
 
 y1 y2 ··· yn 
y1′ y2′ yn′
 
 ··· 
y1′′ y2′′ yn′′
 
W′ = 
 ···  = pn−1 (t)W ,

 .. 
 . 
(n−1) (n−1) (n−1) 
 
p
n−1 y1 pn−1 y2 ··· pn−1 yn

completing the proof.

Solution for the non-homogeneous equation: y(t) = yH (t) + Y (t), where yH is the
homogeneous solution, and Y is a particular solution.

4.2 Homogeneous Equations with Constant Coefficients.

This follows in the same setup as that for 2nd order equations. Consider the equation

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

Characteristic equation
an r n + · · · a1 r + a0 = 0
In general, one find n roots, (counting multiplicity)

(r − r1 )(r − r2 ) · · · (r − rn ) = 0.

From these roots one can find n solutions. It follows the same rules as for 2nd order
equations, with some extensions (marked with * in the following table).

80
 root type solution(s)
r is real, un-repeated ert
r is real, double root ert , tert
(*) r is real, triple root ert , tert , t2 ert
(*) r is real, repeated with multiplicity m ert , tert , · · · tm−1 ert
r = λ ± iµ complex eλt cos µt, eλt sin µt
(*) r = λ ± iµ complex and double roots eλt cos µt, eλt sin µt and teλt cos µt, teλt sin µt
(*) r = λ ± iµ complex, repeated m times similar ...

The hard work is to find the roots, i.e., factorization of polynomials!

Example 1. (a). Find the general solution of y (4) − 4y ′′ = 0.

(b). Find the solution with initial conditions

y(0) = 0, y ′ (0) = 0, y ′′ (0) = 8, y ′′′ (0) = 0.

Answer. (a). Write out the characteristic polynomial

r 4 − 4r 2 = 0, r 2 (r 2 − 4) = 0, r 2 (r − 2)(r + 2) = 0

We find the roots

r1 = −2, r2 = 2, r3 = r4 = 0
The corresponding solutions

y1 = e−2t , y2 = e2t , y3 = e0t = 1, y4 = te0t = t

General solution

y(t) = c1 y1 + c2 y2 + c3 y3 + c4 y4 = c1 e−2t + c2 e2t + c3 + c4 t.

(b). We now determine the constants by initial data. It’s useful to work out the derivatives
first:

y ′ (t) = −2c1 e−2t + 2c2 e2t + c4 .

y ′′ (t) = 4c1 e−2t + 4c2 e2t
y ′′′ (t) = −8c1 e−2t + 8c2 e2t

Then, the 4 ICs give

y(0) = 0 : c1 + c2 + c3 = 0
′
y (0) = 0 : −2c1 + 2c2 + c4 = 0
′′
y (0) = 0 : 4c1 + 4c2 = 8, c1 + c2 = 2
′′′
y (0) = 0 : −8c1 + 8c2 = 0, c1 = c2

81
From the last two equation, we get c1 = c2 = 1. Putting these back into the first 2 equations,
we get c3 = −2 and c4 = 0.
The solution is
y(t) = e−2t + e2t − 2.

In the next example, we will focus on finding general solutions.

Example 2. Find the general solution for the following equations:

(I) : y (4) + 4y ′′ = 0
(II) : y ′′′ − y = 0
(III) : y (4) + 8y ′′ + 16y = 0
(VI) : y ′′′ + 3y ′′ + 3y ′ + y = 0
(V)* : y (4) + 8y = 0

Answer. (I): Characteristic equation and the roots:

r 4 + 4r 2 = 0, r 2 (r 2 + 4) = 0, r1 = r2 = 0, r3,4 = ±2i

General solution
y(t) = c1 + c2 t + c3 cos 2t + c4 sin 2t.
(II): Characteristic equation and the roots:
√
3 2 1 3
r − 1 = 0, (r − 1)(r + r + 1) = 0, r1 = 1, r2 = − ±
2 2
General solution √ √ !
t − 21 t 3 3
y(t) = c1 e + e c2 cos t + c3 sin t .
2 2
(III): Characteristic equation and the roots:

r 4 + 8r 2 + 16 = 0, (r 2 + 4)2 = 0, r1 = r2 = 2i, r3 = 34 = −2i.

We see that we have double complex roots. General solution is

y(t) = c1 cos 2t + c2 sin 2t + c3 t cos 2t + c4 t sin 2t.

(VI): Characteristic equation and the roots:

r 3 + 3r 2 + 3r + 1 = 0, (r + 1)3 = 0, r1 = r2 = r3 = −1

This is a triple root! General solution is

y(t) = c1 e−t + c2 te−t + c3 t2 e−t = e−t (c1 + c2 t + c3 t2 ).

(V)*: Characteristic equation and the roots:

r 4 + 8 = 0, r 4 = −8 = 8eiπ , rk = 81/4 ei(π+2(k−1)π)/4 , k = 1, 2, 3, 4.

82
so we have 4 roots

r1 = 81/4 eiπ/4 , r2 = 81/4 ei3π/4 , r3 = 81/4 ei5π/4 , r4 = 81/4 ei7π/4 .

√
We now write out the real and imaginary parts, using Euler’s formula. Let a = 81/4 / 2, we
can now write
r1 = a + ia, r2 = −a + ia, r3 = −a − ia, r4 = a − ia.
We see that r1 , r4 are conjugate pairs, so are r2 , r3 . The general solution is
   
y(t) = eat c1 cos(at) + c2 sin(at) + e−at c3 cos(at) + c4 sin(at) .

NB! Factorizing a polynomial is non-trivial!

4.3 Higher Order Linear Non-homogeneous Differential Equa-

tions
We focus on how to find a particular solution.
In the simple case with constant coefficients, and we have found a set of fundamental
solutions y1 , · · · , yn , and the right-hand-side g(t) is a combination of eat , sin(αt), cos(αt) and
polynomials, we follow the same pattern as for 2nd order equation, to get the correct form of
a particular solution.
In the more general case, we will use the method called: variation of parameters. Consider
the equation
y (n) + pn−1 (t)y (n−1) + · · · + p1 (t)y ′ + p0 (t)y = g(t),
and let y1 , y2 , · · · , yn be the set of fundamental solutions for the corresponding homogeneous
equation.
We seek a particular solution of the form
Y = u1 (t)y1 + u2 (t)y2 + · · · + un (t)yn
The n functions u1 , u2 , · · · , un are to be determined. We see now we have (n − 1) free choices
to impose addition constraints these functions. We choose
y1 u′1 + y2 u′2 + · · · + yn u′n = 0
y1′ u′1 + y2′ u′2 + · · · + yn′ u′n = 0
y1′′ u′1 + y2′′ u′2 + · · · + yn′′ u′n = 0
..
.
(n−2) ′ (n−2) ′
y1 u1 + y2 u2 + · · · + yn(n−2) u′n = 0
By these constraints, we now have simple forms for the derivatives of Y .
Y ′ = y1′ u1 + y2′ u2 + · · · + yn′ un
Y ′′ = y1′′ u1 + y2′′ u2 + · · · + yn′′ un
..
.
(n−1) (n−1)
Y (n−1) = y1 u1 + y2 u2 + · · · + yn(n−1) un

83
 The n-th derivative is
h i h i
(n) (n) (n−1) ′ (n−1) ′
Y (n) = y1 u1 + y2 u2 + · · · + yn(n) un + y1 u1 + y2 u2 + · · · + yn(n−1) u′n

Plug all these into the equation (A), and collect like terms, we get
h i
(n) (n−1)
LHS = u1 y1 + pn−1 (t)y1 + · · · + p1 (t)y1′ + p0 (t)y1
h i
(n) (n−1)
+u2 y2 + pn−1 (t)y2 + · · · + p1 (t)y2′ + p0 (t)y2 + · · ·
h i
+un yn(n) + pn−1 (t)yn(n−1) + · · · + p1 (t)yn′ + p0 (t)yn
h i
(n−1) ′ (n−1) ′
+ y1 u1 + y2 u2 + · · · + yn(n−1) u′n .

Here all the terms are 0 except the last one. We get
(n−1) ′ (n−1) ′
y1 u1 + y2 u2 + · · · + yn(n−1) u′n = g(t).

Combine the last equation with the (n − 1) constraints we imposed earlier, we get a system
of n linear equations to solve for u′i ’s. Indeed, in matrix-vector form, it can be written as
 ′  
y1 y2 ... yn
 
u1 0
 y1′ y2′ ... yn′   ′ 
  u2   0 

 .. .. ..   ..  =  ..  . (∗)

 . . .  .   . 
(n−1) (n−1) (n−1)
y1 y2 . . . yn u′n g(t)

Note that the coefficient matrix is the Wronskian, and is never 0. There this system has a
unique solution for u′i . One can the recover ui by integration.

Example Consider the non-homogeneous differential equation

t3 y ′′′ − 3t2 y ′′ + 6ty ′ − 6y = t, (t > 0)

Show that
y1 (t) = t, y2 (t) = t2 y3 (t) = t3
form a fundamental set of solution for the corresponding homogeneous equation. Then, find
a particular solution for the non-homogeneous equation.

Answer. (1) One can easily plug these 3 function and verify that they are solutions for
the homogeneous equation.
(2) To check if they are linearly independent, we compute the Wronskian
 t t2 t3 
 

W (t) = 1 2t 3t2  = 2t3 .

 
0 2 6t 

(3). Let the particular solution take the form

Y = t u1 (t) + t2 u2 (t) + t3 u3 (t).

84
By formula (*), we have
t t2 t3
  ′   
u1 0
1 2t 3t2  u′2  =  0 
0 2 6t u3 t−2
We get  ′      1 
u1 1/(2t) u1 2 ln t
u′2  =  −1/t2  , u2  =  1/t  .
u′3 1/(2t3 ) u3 −t−2 /4
A particular solution is
1 t 3
Y = t ln t + t2 (1/t) − t3 t−2 /4 = ln t + t, (t > 0).
2 2 4
We see that the last term is already in y1 , so we can drop it and simply take
t
Y = ln t, (t > 0)
2
We can now form the general solution for the non-homogeneous equation
t
Y = c1 t = c2 t2 + c3 t3 + ln t.
2

Remark. In general case, find the solution for the homogeneous equation with variable
coefficient is not easy! We can only handle the constant coefficient case so far.

85
Chapter 6

The Laplace Transform

Laplace transform is mainly used to handle piecewise continuous or impulsive force.

6.1 Definition of the Laplace transform

Topics:
• Definition of Laplace transform,
• Compute Laplace transform by definition, including piecewise continuous functions.

Definition: Given a function f (t), t ≥ 0, its Laplace transform is defined as

Z ∞ Z A
. . .
F (s) = L{f (t)} = e−st f (t) dt = lim e−st f (t) dt .
0 A→∞ 0

We say the transform converges if the limit exists, and diverges if not.
Next we will give examples on computing the Laplace transform of given functions by
definition.

Example 1. f (t) = 1 for t ≥ 0.

Answer.
A
1 −st A
Z 
−st
F (s) = L{f (t)} = lim e dt = lim − e 
A→∞ 0 A→∞ s
 t=0 
1  −sA 1 1 1
− 1 = lim − e−sA + = ,

= lim − e (s > 0)
A→∞ s A→∞ s s s
Note that the condition s > 0 is needed to ensure that the limit exists, and it is 0.

Example 2. f (t) = eat .

Answer.
A A
1 −(s−a)t A
Z Z 
−st at −(s−a)t
F (s) = L{f (t)} = lim e e dt = lim e dt = lim − e
A→∞ 0 A→∞ 0 A→∞ s−a 
0
1  −(s−a)A  1
= lim − e −1 = , (s > a)
A→∞ s−a s−a

86
Note that the condition s − a > 0 is needed to ensure that the limit exists.

Example 3. f (t) = tn , for n ≥ 1 integer.

Answer. Review integration-by-parts:

Z Z
u(t)v (t) dt = uv − u′ (t)v(t) dt.
′

For here, we have

( A Z A n−1 −st )
A
e−st  nt e
Z
−st n n
F (s) = lim e t dt = lim t − dt
A→∞ 0 A→∞ −s 0 
0 −s
A
n n
Z
= 0+ lim e−st tn−1 dt = L{tn−1 }.
s A→∞ 0 s
So we get a recursive relation
n
L{tn } = L{tn−1 }, for all n,
s
which means
n−1 n−2
L{tn−1 } = L{tn−2 }, L{tn−2 } = L{tn−3 }, ···
s s
By induction, we get
n n (n − 1) n (n − 1) (n − 2)
L{tn } = L{tn−1 } = L{tn−2 } = L{tn−3 }
s s s s s s
n (n − 1) (n − 2) 1 n! 1 n!
= ··· = · · · L{1} = n = n+1 , (s > 0)
s s s s s s s

Example 4. Find the Laplace transform of sin at and cos at.

Answer. Method 1. Compute by definition, with integration-by-parts, twice. (lots of

work...)
Method 2. Use the Euler’s formula

eiat = cos at + i sin at, L eiat = L{cos at} + iL{sin at}.


⇒

By Example 2 we have
1 1(s + ia) s + ia s a
L eiat =

= = 2 = 2 +i 2 .
s − ia (s − ia)(s + ia) s + a2 s + a2 s + a2
Comparing the real and imaginary parts, we get
s a
L{cos at} = , L{sin at} = , (s > 0).
s2 + a2 s2 + a2
R∞ RA
Remark: Now we will use 0 instead of limA→∞ 0 , without causing confusion.

87
 For piecewise continuous functions, Laplace transform can be computed by integrating
each integral and add up at the end.

Example 5. Find the Laplace transform of


1, 0 ≤ t < 2,
f (t) =
t − 2, 2 ≤ t.

We do this by definition:
Z ∞ Z 2 Z ∞
−st −st
F (s) = e f (t) dt = e dt + (t − 2)e−st dt
0 0 2
1 −st 2 1 −st ∞
  Z A
1 −st
= e  + (t − 2) e  − e dt
−s t=0 −s t=2 2 −s
1 1 −st ∞

1 −2s 1 −2s 1
= (e − 1) + (0 − 0) + e  = (e − 1) + 2 e−2s
−s s −s t=2 −s s

Remark. Later in Ch 6.3 we will use a different method to deal with discontinuous
(piecewise continuous) functions.

6.2 Solution of initial value problems

Topics:

• Properties of Laplace transform, with proofs and examples

• Inverse Laplace transform, with examples, and review of partial fraction,

• Solution of initial value problems, with continuous source terms, with examples covering
various cases.

Properties of Laplace transform:

1. Linearity: L{c1 f (t) + c2 g(t)} = c1 L{f (t)} + c2 L{g(t)}.

2. First derivative: L{f ′ (t)} = sL{f (t)} − f (0).

Second derivative: L{f ′′ (t)} = s2 L{f (t)} − sf (0) − f ′ (0).
Higher order derivative:

L{f (n) (t)} = sn L{f (t)} − sn−1 f (0) − sn−2 f ′ (0) − · · · − sf (n−2) (0) − f (n−1) (0).

3. L{−tf (t)} = F ′ (s) where F (s) = L{f (t)}. This also implies L{tf (t)} = −F ′ (s).

4. Shift Theorem 1: L{eat f (t)} = F (s − a) where F (s) = L{f (t)}.

This implies eat f (t) = L−1 {F (s − a)}.

Remarks:

88
 • Note properties 2 are useful in differential equations. It shows that each derivative in t
caused a multiplication of s in the Laplace transform.

• Property 3 is the counter part for Property 2. It shows that each derivative in s causes
a multiplication of −t in the inverse Laplace transform.

• Property 4 is the first Shift Theorem. A counter part of it will come later in chapter 6.3.

Proof:
1. This follows by definition.

2. By definition
Z ∞ ∞ Z ∞
′ −st ′ −st
L{f (t)} = e f (t)dt = e f (t) − (−s)e−st f (t)dt = −f (0) + sL{f (t)}.

0 0 0

The second derivative formula follows from that of the first derivative. Set f to be f ′
we get

L{f ′′ (t)} = sL{f ′ (t)} − f ′ (0) = s(sL{f (t)} − f (0)) − f ′ (0) = s2 L{f (t)} − sf (0) − f ′ (0).

For high derivatives, it follows by induction.

3. The proof follows from the definition:

Z ∞ Z ∞ Z ∞
′ d −st ∂ −st
F (s) = e f (t)dt = (e )f (t)dt = (−t)e−st f (t)dt = L{−tf (t)}.
ds 0 0 ∂s 0

4. This proof also follows from definition:

Z ∞ Z ∞
at −st at
L{e f (t)} = e e f (t)dt = e−(s−a)t f (t)dt = F (s − a).
0 0

By using these properties, we could find more easily Laplace transforms of many other
functions.

Example 1.
n! n!
From L{tn } = , we get L{eat tn } = .
sn+1 (s − a)n+1

Example 2.
b b
From L{sin bt} = , we get L{eat sin bt} = .
s 2 + b2 (s − a)2 + b2

Example 3.
s s−a
From L{cos bt} = , we get L{eat cos bt} = .
s2 + b2 (s − a)2 + b2

89
 Example 4.
3! 1 1
L{t3 + 5t − 2} = L{t3 } + 5L{t} − 2L{1} = 4
+5 2 −2 .
s s s

Example 5.
3! 1 1
L{e2t (t3 + 5t − 2)} = 4
+5 2
−2 .
(s − 2) (s − 2) s−2

Example 6.
2 4 s+1
L{(t2 + 4)e2t − e−t cos t} = 3
+ − ,
(s − 2) s − 2 (s + 1)2 + 1

because
2 4 2 4
L{t2 + 4} = + , ⇒ L{(t2 + 4)e2t } = + .
s3 s (s − 2)3 s − 2

Next are a few examples for Property 3.

Example 7.
 ′
at 1 at 1 1
Given L{e } = , we get L{te } = − =
s−a s−a (s − a)2

Example 8.  ′
b 2bs
L{t sin bt} = − = .
s 2 + b2 (s2 + b2 )2

Example 9. ′
s 2 − b2

s
L{t cos bt} = − = ··· = .
s + b2
2 (s2 + b2 )2

Inverse Laplace transform. Definition:

L−1 {F (s)} = f (t), if F (s) = L{f (t)}.

Technique: find the way back.

Some simple examples:

Example 10.
     
−1 3 −1 3 2 3 2 3
L 2
=L · 2 = L−1 = sin 2t.
s +4 2 s + 22 2 s + 22
2 2

90
 Example 11.
       
−1 2 −1 1 6 1 −1 3! 1 −5t −1 3! 1
L 4
=L · 4
= L 4
= e L 4
= e−5t t3 .
(s + 5) 3 (s + 5) 3 (s + 5) 3 s 3

Example 12.
     
−1 s+1 −1 s 1 −1 2 1
L 2
=L 2
+ L 2
= cos 2t + sin 2t.
s +4 s +4 2 s +4 2

Example 13.
     
−1 s+1 −1 s+1 −1 3/4 1/4 3 1
L 2
=L =L + = e2t + e−2t .
s −4 (s − 2)(s + 2) s−2 s+2 4 4

Here we used partial fraction to find out:

s+1 A B
= + , A = 3/4, B = 1/4.
(s − 2)(s + 2) s−2 s+2

Solutions of initial value problems.

We will go through one example first.

Example 14. (Two distinct real roots.) Solve the initial value problem by Laplace
transform,
y ′′ − 3y ′ − 10y = 2, y(0) = 1, y ′ (0) = 2.

Answer. Step 1. Take Laplace transform on both sides: Let L{y(t)} = Y (s), and then

L{y ′ (t)} = sY (s) − y(0) = sY − 1, L{y ′′ (t)} = s2 Y (s) − sy(0) − y ′ (0) = s2 Y − s − 2.

Note the initial conditions are the first thing to go in!

2
L{y ′′ (t)} − 3L{y ′ (t)} − 10L{y(t)} = L{2}, ⇒ s2 Y − s − 2 − 3(sY − 1) − 10Y = .
s
Now we get an algebraic equation for Y (s).
Step 2: Solve it for Y (s):

2 s2 − s + 2 s2 − s + 2
(s2 − 3s − 10)Y (s) = +s+2−3= , ⇒ Y (s) = .
s s s(s − 5)(s + 2)

Step 3: Take inverse Laplace transform to get y(t) = L−1 {Y (s)}. The main technique
here is partial fraction.

s2 − s + 2 A B C A(s − 5)(s + 2) + Bs(s + 2) + Cs(s − 5)

Y (s) = = + + = .
s(s − 5)(s + 2) s s−5 s+2 s(s − 5)(s + 2)

91
Compare the numerators:

s2 − s + 2 = A(s − 5)(s + 2) + Bs(s + 2) + Cs(s − 5).

The previous equation holds for all values of s. We will now choose selected values of s such
that only one of the constants A, B, C will be non-zero, so we can solve for it.
1
s=0: −10A = 2, ⇒ A=−
5
22
s=5: 35B = 22, ⇒ B=
35
4
s = −2 : 14C = 8, ⇒ C=
7
Now, Y (s) is written into sum of terms which we can find the inverse transform:
1 1 1 1 22 4
y(t) = AL−1 { } + BL−1 { } + CL−1 { } = − + e5t + e−2t .
s s−5 s+2 5 35 7
NB! Pay attention to the roots of the denominators for F (s). Note that the factors (s − 5)
and (s + 2) come from the characteristic equation, and the term s comes from the source term.

Algorithm for finding solutions:

• Take Laplace transform on both sides. You will get an algebraic equation for Y (s).

• Solve this equation to get Y (s).

• Take inverse transform to get y(t) = L−1 {Y }.

Example 15. (Distinct real roots, but one matches the source term.) Solve the initial
value problem by Laplace transform,

y ′′ − y ′ − 2y = e2t , y(0) = 0, y ′ (0) = 1.

Answer. Take Laplace transform on both sides of the equation, we get

1
L{y ′′ } − L{y ′ } − L{2y} = L{e2t }, ⇒ s2 Y (s) − 1 − sY (s) − 2Y (s) = .
s−2
Solve it for Y :
1 s−1 s−1 s−1
(s2 −s−2)Y (s) = +1 = , ⇒ Y (s) = 2
= .
s−2 s−2 (s − 2)(s − s − 2) (s − 2)2 (s + 1)
Use partial fraction:
s−1 A B C
2
= + + .
(s − 2) (s + 1) s + 1 s − 2 (s − 2)2
Compare the numerators:

s − 1 = A(s − 2)2 + B(s + 1)(s − 2) + C(s + 1)

92
Set s = −1, we get A = − 92 . Set s = 2, we get C = 31 . Set s = 0 (any convenient values of s
can be used in this step), we get
8 1 1 1 4 2
−1 = 4A − 2B + C, −1 = − − 2B + , 2B = + = , B= .
9 3 9 3 9 9
So
2 1 2 1 1 1
Y (s) = − + +
9 s + 1 9 s − 2 3 (s − 2)2
and
2 2 1
y(t) = L−1 {Y } = − e−t + e2t + te2t .
9 9 3

Discussion: We compare this to the method of undetermined coefficient. General solution

of the equation should be y = yH + Y , where yH is the general solution to the homogeneous
equation and Y is a particular solution. The characteristic equation is r 2 − r − 2 = (r + 1)(r −
2) = 0, so r1 = −1, r2 = 2, and yH = c1 e−t + c2 e2t . Since 2 is a root, so the form of the
particular solution is Y = Ate2t . This discussion concludes that the solution should be of the
form
y = c1 e−t + c2 e2t + Ate2t
for some constants c1 , c2 , A. This fits well with our result.

Example 16. (Complex roots.) Solve

y ′′ − 2y ′ + 10y = e−t , y(0) = 0, y ′ (0) = 1.

Answer. Before we solve it, let’s use the method of undetermined coefficients to find out
which terms will be in the solution.

r 2 − 2r + 10 = 0, (r − 1)2 + 9 = 0, r1,2 = 1 ± 3i,

yH = c1 et cos 3t + c2 et sin 3t, Y = Ae−t ,

so the solution should have the form:

y = yH + Y = c1 et cos 3t + c2 et sin 3t + Ae−t .

The Laplace transform would be

s−1 3 1 c1 (s − 1) + c2 (3) A
Y (s) = c1 2 2
+ c2 2 2
+A = 2 2
+ .
(s − 1) + 3 (s − 1) + 3 s+1 (s − 1) + 3 s+1
This gives us some idea on which terms to look for in partial fraction.
Now let’s use the Laplace transform:

Y (s) = L{y}, L{y ′ } = sY − y(0) = sY,

L{y ′′ } = s2 Y − sy(0) − y(0) = s2 Y − 1.

1 1 s+2
s2 Y (s) − 1 − 2sY (s) + 10Y (s) = , ⇒ (s2 − 2s + 10)Y (s) = +1=
s+1 s+1 s+1

93
 s+2 s+2 A B(s − 1) + 3C
Y (s) = 2
= 2 2
= +
(s + 1)(s − 2s + 10) (s + 1)((s − 1) + 3 ) s+1 (s − 1)2 + 32
Compare the numerators:

s + 2 = A((s − 1)2 + 9) + (B(s − 1) + 3C)(s + 1).

1
Set s = −1: 13A = 1, A = 13 .
2 1
Compare coefficients of s -term: A + B = 0, B = −A = − 13 .
Set any value of s, say s = 0, we get
15 5
2 = 10A − B + 3C, 3C = 2 − 10A + B = , C= .
13 13
Taking inverse transform, we get the solution
1 −t 1 5
y(t) = Ae−t + Bet cos 3t + Cet sin 3t = e − et cos 3t + et sin 3t.
13 13 13
We see that this fits our prediction.

Example 17. (Pure imaginary roots.) Solve

y ′′ + y = cos 2t, y(0) = 2, y ′ (0) = 1.

Answer. Again, let’s first predict the terms in the solution:

r 2 + 1 = 0, r1,2 = ±i, yH = c1 cos t + c2 sin t, Y = A cos 2t

so
y = yH + Y = c1 cos t + c2 sin t + A cos 2t,
and the Laplace transform would be
s 1 s
Y (s) = c1 + c2 2 +A 2 .
s1 +1 s +1 s +4
Now, let’s take Laplace transform on both sides:
s
s2 Y − 2s − 1 + Y = L{cos 2t} =
s2 +4
s 2s3 + s2 + 9s + 4
(s2 + 1)Y (s) = + 2s + 1 =
s2 + 4 s2 + 4
2s3 + s2 + 9s + 4 As + B Cs + D(2)
Y (s) = 2 2 2
= 2 + .
(s + 2 )(s + 1) s +1 s2 + 22
Comparing numerators, we get

2s3 + s2 + 9s + 4 = (As + B)(s2 + 4) + (Cs + 2D)(s2 + 1).

One may expand the right-hand side and compare terms s3 , s2 , s, 1 to find A, B, C, D, but that
takes more work.

94
 Let’s try by setting s into complex numbers.
Set s = i, and remember the facts i2 = −1 and i3 = −i, we have

−2i − 1 + 9i + 4 = (Ai + B)(−1 + 4),

which gives
7
3 + 7i = 3B + 3Ai, ⇒ B = 1, A = .
3
Set now s = 2i:
−16i − 4 + 18i + 4 = (2Ci + 2D)(−3),
then
1
0 + 2i = −6D − 6Ci, ⇒ D = 0, C=− .
3
So
7 s 1 1 s
Y (s) = + 2 −
3 s + 1 s + 1 3 s2 + 4
2

and
7 1
y(t) = cos t + sin t − cos 2t.
3 3

We see that the method can be applied to higher order equations with constant coefficients.
The hard part in the computation is to factorize the characteristic polynomial.

Example 18. Solve the initial value problem

y (4) − 16y = 0, y(0) = 0, y ′ (0) = 4, y ′′ (0) = 0, y ′′′ (0) = 0.

Answer. We use

L{y (4) } = s4 Y − s3 y(0) − s2 y ′ (0) − sy ′′ (0) − y ′′′ (0) = s4 Y − 4s2 .

Taking Laplace transform of the equation, we get

4s2 4s2
s4 Y − 4s2 − 16Y = 0, Y (s) = = .
s4 − 16 (s2 − 4)(s2 + 4)

By Partial fraction, we fit in

as + 2b cs + 2d
Y (s) = + 2 (s2 + 4)(as + 2b) + (s2 − 4)(cs + 2d) = 4s2 .
s2 − 22 s + 22
Set s = 2 and s = −2, we get

8(2a + 2b) = 4(4), 8(−2a + 2b) = 4(4), a = 0, b = 1.

By checking the s3 -term, we get c = −a = 0. By checking the constant term, we get d = b = 1.

So
2 2
Y (s) = 2 2
+ 2 .
s −2 s + 22

95
 Take the inverse transform, we get the solution

y(t) = sinh(2t) + sin(2t).

A very brief review on partial fraction, targeted towards inverse Laplace trans-
form.

Goal: rewrite a fractional form PPm

n (s)
(s) (where Pn is a polynomial of degree n) into sum of
“simpler” terms. We assume n < m.

The type of terms appeared in the partial fraction is solely determined by the denominator
Pm (s). First, we factorize Pm (s) and write it into product of terms of

(s − a), (s2 + a2 ), (s − λ)2 + µ2 .

The following table gives the terms in the partial fraction and their corresponding inverse
Laplace transform.

term in PM (s) from where? term in partial fraction inverse L.T.

real root, or
A
s−a g(t) = eat Aeat
s−a
double roots,
A B
(s − a)2 or r = a and g(t) = eat + Aeat + Bteat
s − a (s − a)2
double roots,
A B C C 2 at
(s − a)3 and g(t) = eat + 2
+ Aeat + Bteat + t e
s − a (s − a) (s − a)3 2
imaginary roots or
As + Bµ
s 2 + µ2 g(t) = cos µt or sin µt A cos µt + B sin µt
s 2 + µ2
complex roots, or
2 2 A(s − λ) + Bµ
(s − λ) + µ g(t) = eλt cos µt or eλt sin µt eλt (A cos µt + B sin µt)
(s − λ)2 + µ2

In summary, this table can be written as

Pn (s)
(s − a)(s − b)2 (s − c)3 ((s − λ)2 + µ2 )
A B1 B2 C1 C2 C3 D1 (s − λ) + D2 µ
= + + + + + + .
s − a s − b (s − b)2 s − c (s − c)2 (s − c)3 (s − λ)2 + µ2

96
 Remark. As we have mentioned before, Laplace transform is basically used to deal with
discontinuous (piecewise continuous) functions. The examples we have seen so far are all with
continuous functions. These example serve as a way for us to get familiar with the method
and the basic techniques involved. Next, we will study discontinuous functions.

6.3 Step functions

Topics:

• Definition and basic application of unit step (Heaviside) function,

• Laplace transform of step functions and functions involving step functions (piecewise
continuous functions),

• Inverse transform involving step functions.

We use steps functions to form piecewise continuous functions.

Unit step function(Heaviside function):

0, 0 ≤ t < c,
uc (t) =
1, c ≤ t.

for c ≥ 0. A plot of uc (t) is below:

✻uc
1

✲
c t
0

Note: It is common to write u(t) = u0 (t) where the step occurs at t = 0, and uc (t) is then
shifted by c units in t axis, i.e., uc (t) = u(t − c).

For a given function f (t), if it is multiplied with uc (t), then


0, 0 < t < c,
uc (t)f (t) =
f (t), c ≤ t.

This says that uc (t)f (t) equals to f (t) on the interval [c, ∞), and is 0 everywhere else.
We say uc (t) picks up the interval [c, ∞).

Example 1. Consider 
1, 0 ≤ t < c,
1 − uc (t) =
0, c ≤ t.
A plot of this is given below

97
 ✻1 − uc
1

✲
c t
0

Similarly, the function (1− uc (t))f (t) equals to f (t) on the interval [0, c), and 0 everywhere
else.
We say that the function (1 − uc (t)) picks up the interval [0, c).

Example 2. Rectangular pulse. Let 0 < a < b < ∞ and consider the function ua (t)−ub (t).
The plot of the function looks like

✻ua (t) − ub (t)

1

✲
a t
0 b

The function (ua (t) − ub (t))f (t) equals to f (t) on the interval [a, b), and 0 everywhere else.
We say that the function ua (t) − ub (t) picks up the interval [a, b).

We can now express discontinuous functions in terms of step functions.

Example 3. Consider the function

(
t2 , 0 ≤ t < a
g(t) =
0, a ≤ t

We can rewrite it in terms of the unit step function as

 
g(t) = t2 · 1 − ua (t) .

If we have another function

 2
 t ,
 0≤t<a
h(t) = sin t a ≤ t < b

0, b≤t


We can rewrite it in terms of the unit step function as

 
h(t) = t2 · 1 − ua (t) + sin t · (ua (t) − ub (t)).

98
 If we add another function in it:
 2
 t ,
 0≤t<a
k(t) = sin t a ≤ t < b
 t

e, b≤t

We can rewrite it in terms of the unit step function as

 
k(t) = t2 · 1 − ua (t) + sin t · (ua (t) − ub (t)) + et ub (t).

It shall be apparent to us now, that the functions t2 , sin t, et are “dummies”. We could
replace them with any function, so

 f1 (t), 0 ≤ t < a

g(t) = f2 (t) a ≤ t < b

f3 (t), b ≤ t


can be written as
 
g(t) = f1 (t) · 1 − ua (t) + f2 (t) · (ua (t) − ub (t)) + f3 (t)ub (t).

Note that in the final form, each function fi (t) is multiplied by the step function that “picks
up” the corresponding interval. Then we add them all up.

Example 4. One also needs to understand how to go the back way, i.e., if a function is
given using step functions, we must understand its meaning. Now consider

f (t) = u3 (t)t2 − u(5)(t)et .

Can you evaluate this function at various values of t, say t = 2, 4, 6?

Recalling the definition of uc (t), this shall be simple:

f (2) = u3 (2)22 − u(5)(2)e2 = 0, (because u3 (2) = 0, u5 (2) = 0).

Once this is apparent, we can speed up

f (4) = 42 = 16, f (6) = 62 − e6 .

Laplace transform of uc (t): by definition

Z ∞ Z ∞ ∞
−st −st e−st  e−sc e−sc
L{uc (t)} = e uc (t) dt = e · 1 dt = = 0 − = , (s > 0).
0 c −s t=c −s s

Shift of a function: Given f (t), t > 0, then

(
f (t − c), c ≤ t,
g(t) =
˙ uc (t) · f (t − c) =
0, 0 ≤ t < c,

99
is the shift of f by c units. See figure below.
f g
✻ ✻

✲ ✲
t c t
0 0

Let F (s) = L{f (t)} be the Laplace transform of f (t). Then, the Laplace transform of g(t)
is Z ∞ Z ∞
−st
L{g(t)} = L{uc (t) · f (t − c)} = e uc (t)f (t − c) dt = e−st f (t − c) dt.
0 c
Make a variable change, and let τ = t − c, so t = τ + c, and dt = dτ , and we continue
Z ∞ Z ∞
L{g(t)} = e−s(τ +c) f (τ ) dτ = e−sc e−sτ f (τ ) dτ = e−cs F (s).
0 0

So we conclude
L{uc (t)f (t − c)} = e−cs L{f (t)} = e−cs F (s) ,
which is equivalent to
L−1 {e−cs F (s)} = uc (t)f (t − c) .
Note now we are only considering the domain t ≥ 0. So u0 (t) = 1 for all t ≥ 0.
This is the famous second shift Theorem:

L{uc (t)f (t − c)} = e−cs F (s) .

Remark. Be careful that

L{uc (t)f (t − c)} =

6 L{uc (t)} · L{f (t − c)} .

In general, the transform of the product is NOT the product of the transform:

L{f (t)g(t)} =
6 L{f (t)} · L{g(t)} .

One would need to use convolution, which is discussed in Chapter 6.6.

In the remaining part of this Chapter, we will deal with equations with discontinuous (or
impulsive) source terms. We shall be convinced by now that the key steps in the computation
are taking Laplace transform and inverse Laplace transforms for discontinuous (piecewise
continuous) functions, using the second shift Theorem. Once we are fluent with these two
steps, we can deal with any equations!
Laplace transform with piecewise continuous functions. In following examples we
will compute Laplace transform of piecewise continuous functions with the help of the unit
step function and the second shift Theorem.

100
 Example 5. Given
 π
 sin t, 0≤t< ,
f (t) = 4
 sin t + cos(t − π ), π
≤ t.
4 4
It can be rewritten in terms of the unit step function as
π
f (t) = sin t + u π4 (t) · cos(t − ).
4
(Or, if we write out each intervals
 π  π
f (t) = sin t(1 − u π4 (t)) + sin t + cos(t − ) u π4 (t) = sin t + u π4 (t) · cos(t − ).
4 4
which gives the same answer.)
And the Laplace transform of f is
n π o 1 π s
F (s) = L{sin t} + L u π4 (t) · cos(t − ) = 2 + e− 4 s 2 .
4 s +1 s +1

Example 6. Given (
t, 0 ≤ t < 1,
f (t) =
1, 1 ≤ t.
It can be rewritten in terms of the unit step function as

f (t) = t(1 − u1 (t)) + 1 · u1 (t) = t − u1 (t) · (t − 1) .

The Laplace transform is

1 1
L{f (t)} = L{t} − L {u1 (t) · (t − 1)} = 2
− e−s 2 .
s s

Example 7. Given (
0, 0 ≤ t < 2,
f (t) =
t + 3, 2 ≤ t.
We can rewrite it in terms of the unit step function as

f (t) = (t + 3)u2 (t) = ((t − 2) + 5)u2 (t) = u2 (t) · (t − 2) + 5u2 (t) .

The Laplace transform is

1 1
L{f (t)} = L{u2 (t) · (t − 2)} + 5L{u2 (t)} = e−2s 2
+ 5e−2s .
s s

Example 8. Given (
1, 0 ≤ t < 2,
g(t) =
t2 , 2 ≤ t.

101
We can rewrite it in terms of the unit step function as

g(t) = 1 · (1 − u2 (t)) + t2 u2 (t) = 1 + (t2 − 1)u2 (t) .

Observe that

t2 − 1 = (t − 2 + 2)2 − 1 = (t − 2)2 + 4(t − 2) + 4 − 1 = (t − 2)2 + 4(t − 2) + 3 ,

we have
g(t) = 1 + (t − 2)2 + 4(t − 2) + 3 u2 (t) .

The Laplace transform is

 
1 2 4 3
L{g(t)} = + e−2s 3
+ 2+ .
s s s s

Example 9. Given 
 0, 0 ≤ t < 3,

f (t) = et , 3 ≤ t < 4,

0, 4 ≤ t.


We can rewrite it in terms of the unit step function as

f (t) = et (u3 (t) − u4 (t)) = u3 (t)et−3 e3 − u4 (t)et−4 e4 .

The Laplace transform is

1 1 1 h −3(s−1) i
L{g(t)} = e3 e−3s − e4 e−4s = e − e−4(s−1) .
s−1 s−1 s−1

Inverse transform: We use two properties:

1
L{uc (t)} = e−cs , and L{uc (t)f (t − c)} = e−cs · L{f (t)} .
s
In the following examples we want to find f (t) = L−1 {F (s)}.

Example 10.
1 − e−2s 1 1
F (s) = = 3 − e−2s 3 .
s3 s s
We know that L−1 { s13 } = 21 t2 , so we have

1 2


 t , 0 ≤ t < 2,
−1 1 2 1 2
 2
f (t) = L {F (s)} = t − u2 (t) (t − 2) =
2 2  1 2 1
t − (t − 2)2 , 2 ≤ t.


2 2

102
 Example 11. Given
e−3s
 
1 A B
F (s) = 2 = e−3s = e−3s + .
s + 7s + 12 (s + 4)(s + 3) s+4 s+3
By partial fraction, we find A = −1 and B = 1. So
h i h i
f (t) = L−1 {F (s)} = u3 (t) Ae−4(t−3) + Be3(t−3) = u3 (t) −e−4(t−3) + e3(t−3)

which can be written as a p/w continuous function


 0, 0 ≤ t < 3,
f (t) =
 −e−4(t−3) + e3(t−3) , 3 ≤ t.

Example 12. Given

se−s
 
−s s + 2 − 2 −s s+2 −2
F (s) = 2 =e =s + .
s + 4s + 5 (s + 2)2 + 1 (s + 2)2 + 1 (s + 2)2 + 1
So h i
f (t) = L−1 {F (s)} = u1 (t) e−2(t−1) cos(t − 1) − 2e−2(t−1) sin(t − 1)
which can be written as a p/w continuous function

 0, 0 ≤ t < 1,
f (t) =
 −2(t−1)
e [cos(t − 1) − 2 sin(t − 1)] , 1 ≤ t.

6.4 Differential equations with discontinuous forcing functions

Topics:
• Solve initial value problems with discontinuous force, examples of various cases,

• Describe behavior of solutions, and make physical sense of them.

Next we study initial value problems with discontinuous force. We will start with an
example.

Example 1. (Damped system with force, complex roots) Solve the following initial value
problem

′′ ′ 0, 0 ≤ t < 1,
y + 2y + 2y = g(t), g(t) = , y(0) = 1, y ′ (0) = 0 .
2, 1 ≤ t,

Answer. Let L{y(t)} = Y (s), so L{y ′ } = sY − 1 and L{y ′′ } = s2 Y − s. Also we have

L{g(t)} = 2L{u1 (t)} = e−s 2s . Then
2
s2 Y − s + 2sY − 2 + 2Y = e−s ,
s

103
which gives
2e−s s+2
Y (s) = + 2 .
s(s2 + 2s + 2) s + 2s + 2
Note that the first term is caused by the source (forced response), and the second term is from
the solution of the homogeneous equation, without source.
Now we need to find the inverse Laplace transform for Y (s). We rewrite

2 (s + 1) + 1
Y (s) = e−s + .
s((s + 1) + 1) (s + 1)2 + 1
2

We have to do partial fraction first. We have

2 A B(s + 1) C
2
= + + .
s((s + 1) + 1) s (s + 1) + 1 (s + 1)2 + 1
2

Compare the numerators on both sides:

2 = A((s + 1)2 + 1) + (B(s + 1) + C) · s

Set s = 0, we get A = 1.
Set s = −1, we get 2 = A − C, so C = A − 2 = −1.
Compare s2 -term: 0 = A + B, so B = −A = −1.
We now have  
−s 1 (s + 1) + 1 (s + 1) + 1
Y (s) = e − + .
s (s + 1)2 + 1 (s + 1)2 + 1
We now take the inverse Laplace transform. The second term is easy, we have
 
−1 (s + 1) + 1
L = e−t (cos t + sin t).
(s + 1)2 + 1

For the first term, we need to apply the 2nd shift Theorem because of the e−s term. We get
h i
y(t) = u1 (t) 1 − e−(t−1) (cos(t − 1) + sin(t − 1)) + e−t (cos t + sin t).

Remark: There are other ways to work out the partial fractions.
Extra question: What happens when t → ∞?
Answer: We see all the terms with the exponential function will go to zero, so y → 1 in the
limit. We can view this system as the spring-mass system with damping. Since g(t) becomes
constant 1 for large t, and the particular solution (which is also the steady state) with 1 on
the right hand side is 1, which provides the limit for y.
Further observation:

• We see that the solution to the homogeneous equation is

e−t [c1 cos t + c2 sin t] ,

and these terms do appear in the solution.

• Actually the solution consists of two part: the forced response and the homogeneous
solution.

104
 • Furthermore, the g has a discontinuity at t = 1, and we see a response in the solution
also for t = 1, in the term u1 (t).

Example 2. (Undamped system with force, pure imaginary roots) Solve the following
initial value problem

 0,
 0 ≤ t < π,
′′
y + 4y = g(t) = 4, π ≤ t < 2π, y(0) = 1, y ′ (0) = 0 .

0, 2π ≤ t,


Rewrite
4 4
g(t) = 4(uπ (t) − u2π (t)), L{g} = e−πs − e−2πs .
s s
So
4 −π
s2 Y − s + 4Y = e − e−2π .


s
Solve it for Y :
4 s 4e−π 4e−2π s
Y (s) = e−π − e−2π


2
+ 2
= 2
− 2
+ 2 .
s(s + 4) s + 4 s(s + 4) s(s + 4) s + 4

Work out partial fraction

4 A Bs + C
= + 2 , A = 1, B = −1, C = 0.
s(s2 + 4) s s +4

So
4
L−1 { } = 1 − cos 2t .
s(s2+ 4)
Now we take inverse Laplace transform of Y

y(t) = uπ (t) (1 − cos 2(t − π)) − u2π (t) (1 − cos 2(t − 2π)) + cos 2t
= (uπ (t) − u2π (t))(1 − cos 2t) + cos 2t
(
1 − cos 2t, π ≤ t < 2π,
= cos 2t +
0, otherwise,
= homogeneous solution + forced response

105
 Example 3. In Example 2, let

 0, 0 ≤ t < 4,
g(t) = et , 4 ≤ 5 < 2π,
0, 5 ≤ t.


Find Y (s).

Answer. Rewrite

g(t) = et (u4 (t) − u5 (t)) = u4 (t)et−4 e4 − u5 (t)et−5 e5 ,

so
1 1
G(s) = L{g(t)} = e4 e−4s − e5 e−5s .
s−1 s−1
Take Laplace transform of the equation, we get
1 s
(s2 + 4)Y (s) = G(s) + s, Y (s) = e4 e−4s − e5 e−5s


2
+ 2 .
(s − 1)(s + 4) s + 4
Remark: We see that the first term will give the forced response, and the second term is from
the homogeneous equation.
The students may work out the inverse transform as a practice.

Example 4. (Undamped system with force, example 2 from the book p. 334)

 0, 0 ≤ t < 5,
′′ ′
y + 4y = g(t), y(0) = 0, y (0) = 0, g(t) = (t − 5)/5, 5 ≤ 5 < 10,
1, 10 ≤ t.


Let’s first work on g(t) and its Laplace transform

t−5 1 1
g(t) = (u5 (t) − u10 (t)) + u10 (t) = u5 (t)(t − 5) − u10 (t)(t − 10),
5 5 5
1 −5s 1 1 1
G(s) = L{g(t)} = e 2
− e−10s 2
5 s 5 s
Let Y (s) = L{y(t)}, then

G(s) 1 1 1 1
(s2 + 4)Y (s) = G(s), Y (s) = 2
= e−5s 2 2 − e−10s 2 2
s +4 5 s (s + 4) 5 s (s + 4)
Work out the partial fraction:

. 1 A B Cs + 2D
H(s) = = + 2+ 2
s2 (s2 + 4) s s s +4

one gets A = 0, B = 14 , C = 0, D = − 18 . So
   
. −1 1 −1 1 1 1 2 1 1
h(t) = L 2 2
=L · 2− · 2 2
= t − sin 2t.
s (s + 4) 4 s 8 s +2 4 8

106
Go back to y(t)
1 1
y(t) = L−1 {Y } = u5 (t)h(t − 5) − u10 (t)h(t − 10)
 5 5   
1 1 1 1 1 1
= u5 (t) (t − 5) − sin 2(t − 5) − u10 (t) (t − 10) − sin 2(t − 10)
5 4 8 5 4 8


 0, 0 ≤ t < 5,

 1 1
= 20 (t − 5) − 40 sin 2(t − 5), 5 ≤ 5 < 10,


 1 − 1 (sin 2(t − 5) − sin 2(t − 10)), 10 ≤ t.

4 40

Note that for t ≥ 10, we have y(t) = 14 + R · cos(2t + δ) for some amplitude R and phase δ.
The plots of g and y are given in the book. Physical meaning and qualitative nature of
the solution:
The source g(t) is known as ramp loading. During the interval 0 < t < 5, g = 0 and initial
conditions are all 0. So solution remains 0. For large time t, g = 1. A particular solution is
Y = 14 . Adding the homogeneous solution, we should have y = 14 + c1 sin 2t + c2 cos 2t for t
large. We see this is actually the case, the solution is an oscillation around the constant 41 for
large t.

6.5 Impulse functions

Definition of the unit impulse function δ(t):
Z τ
δ(t) = 0, (t 6= 0), δ(t) dt = 1, (τ > 0)
−τ

One can think of this function as the limit of a rectangular wave with area equals to 1:
1
δ(t) = lim · (u(t + τ ) − u(t − τ )) .
τ →0+ 2τ
Recall u(t) is the unit step function. One can visualize this with graphs.
The impulse function can be shifted:
Z a+τ
δ(t − a) = 0, (t 6= a), δ(t − a) dt = 1, (τ > 0)
a−τ

The most useful property of δ(t − a) in integration:

Z a+ǫ
f (t)δ(t − a) dt = f (a), ǫ>0
a−ǫ

In fact, this can be easily proved, using the limit. We have

Z a+ǫ Z a+τ Z a+τ
1 1
f (t)δ(t − a) dt = lim f (t) dt = lim f (t) dt
a−ǫ τ →0 a−τ 2τ τ →0 2τ a−τ

= lim {average of f (t) on interval [a − τ, a + τ ]}

τ →0
= f (a).

107
This mean: Integrating f (t)δ(t − a) over any interval that contains t = a, one gets the value
of f evaluated at t = a.

Laplace transform of the unit impulse δ(t − a) with a > 0:

Z ∞
L{δ(t − a)} = e−st δ(t − a) dt = e−as .
0

Example 1. Solve

y ′′ + 4y ′ + 5y = δ(t − π), y(0) = 0, y ′ (0) = 0.

Physical interpretation of the equation: Think of the spring-mass system, initially at rest.
Then at time t = π, it gets a hit. BANG!

Answer. Take Laplace transform on both sides of the equation, we get

(s2 + 4s + 5)Y (s) = e−πs

which gives
e−πs 1
Y (s) = 2
= e−πs .
s + 4s + 5 (s + 2)2 + 1
Taking the inverse transform, we get

−2(t−π) 0, 0<t<π
y(t) = uπ (t) · e sin(t − π) = −2(t−π)
e sin(t − π) t ≥ π.

We see clearly the response to the impulsive hit at t = π!

6.6 Convolution*
We observe that we often need to find

L−1 {F (s)G(s)}.

In general, we don’t have h(t) = f (t)g(t). We need the convolution Theorem.

Theorem (Convolution): Let

H(s) = F (s)G(s), f (t) = L−1 {F (s)}, g(t) = L−1 {G(s)}, h(t) = L−1 {H(s)}.

Then Z t
−1
L {F (s)G(s)} = (f ∗ g)(t)=
˙ f (t − τ )g(τ ) dτ, 0 ≤ t < ∞, (A)
0
provided that the integral exists. The integral in (A) is known as the convolution of f and
g.

108
 Properties of the convolution integral:

f ∗g = g∗f (commutative law)

f ∗ (g1 + g2 ) = f ∗ g1 + f ∗ g2 , (distributive law)
(f ∗ g) ∗ h = f ∗ (g ∗ h) (associative law)
f ∗ (cg) = c(f ∗ g) (linear operator)
f ∗ 0 = 0 ∗ f = 0.

These properties could be easily proved by the definition of the convolution integral. We skip
the proofs.
Watch out! In general, (f ∗ 1)(t) 6= f (t). For example, let f (t) = cos t, then
Z t τ =t
(f ∗ 1)(t) = cos(t − τ ) dτ = − sin(t − τ ) = sin t.

0 τ =0

This is because Z t
(f ∗ 1)(t) = f (t − τ ) dτ 6= f (t).
0

Example 1. Calculate h(t) = (f ∗ g)(t) where f (t) = t and g(t) = e−t .

Answer. By definition, we have

Z t Z t Z t Z t
−τ −τ −τ
h(t) = (f ∗ g)(t) = f (t − τ )e dτ = (t − τ )e dτ = te dτ − τ e−τ dτ
0 0 0 0
t t
= −te−τ  − (−τ − 1)e−τ  = t + e−t − 1.
 
τ =0 τ =0

We could verify the Convolution Theorem in this example. We have

1 1 1 1 1 1
F (s) = , G(s) = , H(s) = + − = 2 = F (s)G(s).
s2 s+1 s 2 s+1 s s (s + 1)

Example 2. Let g(t) = δ(t) be the impulse function. Then G(s) = 1. Following the
convolution Theorem, we have
Z t
−1 −1
L {F (s)} = L {F (s)G(s)} = (f ∗ g)(t) = f (t − τ )δ(τ ) dτ = f (t),
0

which verifies the Theorem.

Example 3*. Calculate f ∗ g where


0, 0≤t<2
( 
t, 0 ≤ t < 1
f (t) = g(t) = 1, 2≤t<3
0, 1 ≤ t < ∞. 
0, 3 ≤ t < ∞.


109
 Answer. The piecewise definition of these functions, in particular, the property of local
support of these functions, provides a great opportunity to illustrate the graphical aspects of
the convolution integral. The value of the function (f ∗ g)(t) depends on the value of t in a
piecewise way.
If 0 ≤ t < 2, the graphs of f (t − τ ) and g(τ ) do not have common support, therefore the
product is 0, and so is f ∗ g.
If 2 ≤ t ≤ 3, the functions f (t − τ ) and g(τ ) are both non-zero on the interval 2 ≤ τ < t.
Then f ∗ g is the area of the overlapped triangle, i.e., (f ∗ g)(t) = 12 (t − 2)2 .
If 3 ≤ t ≤ 4, the functions f (t − τ ) and g(τ ) are both non-zero on the interval t ≤ τ < 4.
Then f ∗ g is the area of the overlapped trapezoid, i.e., (f ∗ g)(t) = 12 − 12 (t − 3)2 .
If 4 ≤ t < ∞, the graphs of f (t − τ ) and g(τ ) do not have common support, therefore the
product is 0, and so is f ∗ g.
The graph of (f ∗ g)(t) is given in plot (f) in Figure 6.1.

We now prove that

L{f ∗ g} = F (s)G(s), i.e. (f ∗ g)(t) = L−1 {F (s)G(s)}.

Proof. We have
Z ∞ Z ∞
−sξ
F (s) = e f (ξ) dξ, G(s) = e−sτ g(τ ) dτ,
0 0

Then
Z ∞ Z ∞ Z ∞ Z ∞ 
−sξ −sτ −sτ −sξ
F (s)G(s) = e
f (ξ) dξ · e g(τ ) dτ = e g(τ ) e f (ξ) dξ dτ
0 0 0 0
Z ∞ Z ∞  Z ∞ Z ∞ 
= g(τ ) e−sτ e−sξ f (ξ) dξ dτ = g(τ ) e−s(τ +ξ) f (ξ) dξ dτ.
0 0 0 0

We now make a variable change. For fixed τ , we let t = ξ + τ , so dt = dξ, and t = τ when
ξ = 0. We have
Z ∞ Z ∞  Z ∞ Z ∞ 
−st −st
F (s)G(s) = g(τ ) e f (t − τ ) dt dτ = e f (t − τ )g(τ ) dt dτ.
0 τ 0 τ

We now switch the order of integration, (add a picture), and get

Z ∞ Z t  Z ∞ Z t 
−st −st
F (s)G(s) = e f (t − τ )g(τ ) dτ dt = e f (t − τ )g(τ ) dτ dt
0 0 0 0
Z ∞
= e−st (f ∗ g)(t) dt = L{f ∗ g}
0

completing the proof.

Immediately, one can extend this result to multiple convolutions. We have

L{f1 ∗ f2 ∗ · · · ∗ fn } = F1 (s)F2 (s) · · · Fn (s), L{fi } = Fi (s).

110
 ✻ ✻ Rt
f (t − τ )g(τ )dτ = 0
0

f (τ ) g(τ ) f (t − τ ) g(τ )
1 1

✲τ ✲τ
1 2 3 4 1 t 2 3 4
(a) (b) 0 ≤ t < 2

Rt Rt
✻ 0 f (t − τ )g(τ )dτ = 0.5(t − 2)2 ✻ 0 f (t − τ )g(τ )dτ = 0.5 − 0.5(t − 3)2

f (t − τ ) g(τ ) g(τ ) f (t − τ )
1 1

✲τ ✲τ
1 2 t 3 4 1 2 3 t 4
(c) 2 ≤ t < 3 (d) 3 ≤ t < 4

Rt f ∗g
✻ 0 f (t − τ )g(τ )dτ = 0 ✻
0.5
f (t − τ ) g(τ )
1

✲τ ✲t
1 2 3 4 t 1 2 3 4
(e) 4 ≤ t < ∞ (f) plot of (f ∗ g)(t)

Figure 6.1: Convolution of locally supported functions.

111
 Example Find the Laplace inverse transform of
a
H(s) = .
s2 (s2 + a2 )

Answer. We see that we can rewrite

1 a2
H(s) = F (s)G(s), F (s) = , G(s) =
s2 s 2 + a2
and
f (t) = t, g(t) = sin(at).
By the convolution Theorem, we have
t
at − sin(at)
Z
L−1 {H} = (f ∗ g)(t) = (t − τ ) sin(aτ ) dτ = .
0 a2
Of course we also can do
t
at − sin(at)
Z
L−1 {H} = (g ∗ f )(t) = τ sin a(t − τ ) dτ = ,
0 a2
which gives the same result.

Example 3. Consider the IVP

y ′ = ay + g(t), y(0) = y0 .

We can solve it by the method of integrating factor, where µ(t) = e−at . We have
 ′
(µ(t)y(t))′ = µ(t)g(t), e−at y(t) = e−at g(t).

Integrate both sides from 0 to t, we get

Z t Z t
−aτ ′
e y(τ )) dτ = e−aτ g(τ ) dτ,
0 0
so Z t
−at
e y(t) − y(0) = e−aτ g(τ ) dτ.
0
So the solution is
Z t 
at
y(t) = e e−aτ g(τ ) dτ + y0
0
Z t
= y0 eat + eat e−aτ g(τ ) dτ
0
Z t
= y0 eat + ea(t−τ ) g(τ ) dτ
0
= y0 eat + eat ∗ g(t)

112
The Laplace transform of y(t), computed from this expression, is

y0 G(s)
Y (s) = y0 L{eat } + L{eat }L{g(t)} = + .
s−a s−a
We can now compare. Taking Laplace transform directly on the differential equation, we
get
y0 + G(s)
sY (s) − y(0) = aY (s) + G(s), Y (s) = ,
s−a
which is of course gives us the same result.

Example 4. Consider the second order equation

au′′ + bu′ + cu = δ(t), u(0) = 0, u′ (0) = 0. (A)

Here δ(t) is the impulse function. Let U (s) = L{u(t)}. Then,

1
as2 U (s) + bsU (s) + cU (s) = 1, U (s) = , u(t) = L−1 {U (s)}.
as2 + bs + c
Now consider a more general equation

ay ′′ + by ′ + cy = g(t), y(0) = y0 , y ′ (0) = y0′ . (B)

We have
(as2 + bs + c)Y (s) − (as + b)y0 − ay0′ = G(s)
so
(as + b)y0 + ay0′ G(s)
Y (s) = 2
+ 2 =Φ(s)
˙ + Ψ(s).
as + bs + c as + bs + c
Take inverse transform, we get

y(t) = φ(t) + ψ(t), φ(t) = L−1 {Φ(s)}, ψ(t) = L−1 {Ψ(s)}.

We see clearly now that φ(t) is the solution caused by the non-zero initial condition, and ψ(t)
is the system’s response to the source term g(t). We observe that we can write

Ψ(s) = U (s)G(s).

where U (s) solution for the problem in Example 3. This function H is known as the transfer
function, and we can express the solution using convolution
Z t
ψ(t) = (u ∗ g)(t) = u(t − τ )g(τ ) dτ.
0

113
Chapter 7

Systems of Two Linear Differential

Equations

7.1 Introduction to systems of differential equations

Write out the general form of a system of first order ODE, with x1 , x2 as unknowns.
Given
ay ′′ + by ′ + cy = g(t), y(0) = α, y ′ (0) = β
we can do a variable change: let

x1 = y, x2 = x′1 = y ′

then
x′1 = x2
( 
x1 (0) = α
1 ′′
x′2
= y = (g(t) − bx2 − cx1 ) x2 (0) = β
a
Observation: For any 2nd order equation, we can rewrite it into a system of 2 first order
equations.

Example 1. Given

y ′′ + 5y ′ − 10y = sin t, y(0) = 2, y ′ (0) = 4

Rewrite it into a system of first order equations: let x1 = y and x2 = y ′ = x′1 , then
 ′ 
x1 = x2 x1 (0) = 2
I.C.’s:
x′2 = y ′′ = −5x2 + 10x1 + sin t x2 (0) = 4

We can do the same thing to any high order equations. For n-th order differential equation:

y (n) = F (t, y, y ′ , · · · , y (n−1) )

define the variable change:

x1 = y, x2 = y ′ , ··· xn = y (n−1)

114
we get
x′1 = y ′ = x2


x′2 = y ′′ = x3




..

.
x′ = y (n−1) = xn


 n−1′



xn = y (n) = F (t, x1 , x2 , · · · , xn )
with corresponding source terms.

(Optional) Reversely, we can convert a 1st order system into a high order equation.

Example 2. Given

x′1 = 3x1 − 2x2

 
x1 (0) = 3
x′2 = 2x1 − 2x2 x2 (0) = 12

Eliminate x2 : the first equation gives

3 1
2x2 = 3x1 − x′1 , x2 = x1 − x′1 .
2 2
Plug this into second equation, we get
 ′
3 1
x1 − x′1 = 2x1 − 2x2 = −x1 + x′1
2 2

3 ′ 1
x − x′′ = −x1 + x′1
2 1 2 1
x′′1 − x′1 − 2x1 = 0
with the initial conditions:

x1 (0) = 3, x′1 (0) = 3x1 (0) − 2x2 (0) = 8.

This we know how to solve!

Definition of a solution: a set of functions x1 (t), x2 (t), · · · , xn (t) that satisfy the differential
equations and the initial conditions.

7.2 Review of matrices

A matrix of size m × n:
 
a1,1 · · · a1,n
 ..
A= .  = (ai,j ), 1 ≤ i ≤ m, 1 ≤ j ≤ n.


am,1 · · · am,n

We consider only square matrices, i.e., m = n, in particular for n = 2 and 3.

Basic operations: A, B are two square matrices of size n.

115
 • Addition: A + B = (aij ) + (bij ) = (aij + bij )

• Scalar multiple: αA = (α · aij )

• Transpose: AT switch the ai,j with aji . (AT )T = A.

• Product: For A · B = C, it means ci,j is the inner product of (ith row of A) and (jth
column of B). Example:
     
a b x y ax + bu ay + bv
· =
c d u v cx + du cy + dv

We can express system of linear equations using matrix product.

Example 1.
      
 x1 − x2 + 3x3 = 4 1 −1 3 x1 4
2x1 + 5x3 = 0 can be expressed as:  2 0 5  ·  x2  =  0 
x2 − x3 = 7 0 1 −1 x3 7


Example 2.
′
x′1 = a(t)x1 + b(t)x2 + g1 (t)
       
x1 a(t) b(t) x1 g1 (t)
⇒ = · +
x′2 = c(t)x1 + d(t)x2 + g2 (t) x2 c(t) d(t) x2 g2 (t)

Some properties:

• Identity I: I = diag(1, 1, · · · , 1), AI = IA = A.

• Determinant det(A):  
a b
det = ad − bc,
c d
 
a b c
det  u v w  = avz + bwx + cuy − xvc − ywa − zub.
x y z

• Inverse inv(A) = A−1 : A−1 A = AA−1 = I.

• The following statements are all equivalent: (optional)

– (1) A is invertible;
– (2) A is non-singular;
– (3) det(A) 6= 0;
– (4) row vectors in A are linearly independent;
– (5) column vectors in A are linearly independent.
– (6) All eigenvalues of A are non-zero.

116
7.3 Eigenvalues and eigenvectors
Eigenvalues and eigenvectors of A (only for A as a 2 × 2 real matrix)
λ: scalar value, ~v : column vector, ~v 6≡ 0.
Definition: If A~v = λ~v , then (λ, ~v ) is the (eigenvalue, eigenvector) of A.
They are also called the eigen-pair of A.
Remark: If ~v is an eigenvector, then α~v for any α 6= 0 is also an eigenvector, because

A(α~v ) = αA~v = αλ~v = λ(α~v ).

How to find (λ, ~v )?

A~v − λ~v = ~0, (A − λI)~v = ~0, det(A − λI) = 0.

We see that det(A − λI) is a polynomial of degree 2 (if A is 2 × 2) in λ, and it is also called
the characteristic polynomial of A. We need to find its roots.

Example 1: Find the eigenvalues and the eigenvectors of A where

 
1 1
A= .
4 1

Answer. Let’s first find the eigenvalues.

 
1−λ 1
det(A − λI) = det = (1 − λ)2 − 4 = 0, λ1 = −1, λ2 = 3.
4 1−λ

Now, let’s find the eigenvector ~v1 for λ1 = −1: let ~v1 = (a, b)T
     
~ 1 − (−1) 1 a 0
(A − λ1 I)~v1 = 0, ⇒ · = ,
4 1 − (−1) b 0
     
2 1 a 0
⇒ · = ,
4 2 b 0
so  
1
2a + b = 0, choose a = 1, then we have b = −2, ⇒ ~v1 = .
−2
Finally, we will compute the eigenvector ~v2 = (c, d)T for λ2 = 3:
     
1−3 1 c 0
(A − λ1 I)~v2 = ~0, ⇒ · = ,
4 1−3 d 0
     
−2 1 c 0
⇒ · = ,
4 −2 d 0
so  
1
2c − d = 0, choosec = 1, then we have d = 2, ⇒ ~v2 = .
2

117
 Example 2. Eigenvalues can be complex numbers.
 
2 −9
A= .
4 2
Let’s first find the eigenvalues.
 
2 − λ −9
det(A − λI) = det = (2 − λ)2 + 36 = 0, ⇒ λ1,2 = 2 ± 6i
4 2−λ
We see that λ2 = λ̄1 , complex conjugate. The same will happen to the eigenvectors, i.e.,
~v1 = ~v¯2 . So we need to only find one. Take λ1 = 2 + 6i, we compute ~v = (v 1 , v 2 )T :
   1 
~ −i6 −9 v
(A − λ1 I)~v = 0, · = ~0,
4 −i6 v2
2
−6iv 1 − 9v 2 = 0, choose v 1 = 1, so v 2 = − i,
3
so    
1 1
~v1 = , ~v2 = ~v¯1 = .
− 23 i 2
3i

Example 3. Eigenvalues can be repeated, i.e., λ1 = λ2 . For example, let

 
1 −1
A= .
1 3
Then
 
1 − λ −1
det(A−λI) = det = (1−λ)(3−λ)+1 = λ2 −4λ+3+1 = (λ2 )2 = 0, ⇒ λ1 = λ2 = 2.
1 3−λ
Now we try to find the corresponding eigenvector ~v = (v 1 , v 2 ),
   1 
~ −1 −1 v
(A − 2I)~v = 0, · = ~0, v 1 + v 2 = 0.
1 1 v2
Therefore only one eigenvector could be found
 
1
~v1 = .
−1

Example 4. But, sometimes repeated eigenvalues could have more than 1 corresponding
eigenvectors. Consider the identity matrix A = I, then we obviously have λ1 = λ2 = 1. In
order to find the eigenvector, we compute
   1 
~ 0 0 v
(A − I)~v = 0, · = ~0.
0 0 v2
This is automatically satisfied, without any constraint on v 1 , v 2 . We can then choose any two
linearly independent vectors, for example
   
1 0
~v1 = , ~v2 = .
0 1
We say that, in this case, the double eigenvalue has two linearly independent eigenvectors.
Note this behavior is essentially different from that in Example 3!

118
7.4 Basic theory of systems of first order linear equation
General form of a system of first order equations written in matrix-vector form:

~x′ = P(t)~x + ~g .

If ~g = 0, it is homogeneous. We only consider this case, so

~x′ = P(t)~x.

Superposition: If ~x1 (t) and ~x2 (t) are two solutions of the homogeneous system, then any
linear combination c1 ~x1 + c2 ~x2 is also a solution.
Wronskian of vector-valued functions are defined as
 
W [~x1 (t), ~x2 (t), · · · , ~xn (t)] = det X(t)

where X is a matrix whose columns are the vectors
~x1 (t), ~x2 (t), · · · , ~xn (t).
If det X(t) 6= 0, then ~x1 (t), ~x2 (t), · · · , ~xn (t) is a set of linearly
independent functions.
A set of linearly independent solutions ~x1 (t), ~x2 (t), · · · , ~xn (t) is said to be a fundamental
set of solutions.
The general solution is the linear combination of these solutions, i.e.

~x = c1 ~x1 (t) + c2 ~x2 (t) + · · · + cn ~xn (t).

7.5 Homogeneous systems of two equations with constant co-

efficients.
We consider the following initial value problem:
 ′ 
x1 = ax1 + bx2 x1 (0) = x̄1
I.C.’s:
x′2 = cx1 + dx2 x2 (0) = x̄2

In matrix vector form:

     
′ x1 x̄1 a b
~x = A~x, where ~x = , ~x(0) = A= .
x2 x̄2 c d

Claim: If (λ, ~v ) is an eigen-pair for A, then ~z = eλt~v is a solution to ~x′ = A~x.

Proof.
~z′ = (eλt~v )′ = (eλt )′~v = λeλt~v
A~z = A(eλt~v ) = eλt (A~v ) = eλt λ~v
Therefore ~z′ = A~z so ~z is a solution.
Steps to solve the initial value problem:

• Step I: Find eigenvalues of A: λ1 , λ2 .

• Step II: Find the corresponding eigenvectors ~v1 , ~v2 .

119
 • Step III: Form two solutions: ~z1 = eλ1 t~v1 , ~z2 = eλ2 t~v2 .

• Step IV: Check that ~z1 , ~z2 are linearly independent: the Wronskian

W (~z1 , ~z2 ) = det(~z1 , ~z2 ) 6= 0.

(This step is usually OK in our problems.)

• Step V: Form the general solution: ~x = c1 ~z1 + c2 ~z2 .

• If initial condition ~x(0) is given, then use it to determine c1 , c2 .

We will start with an example.

Example 1. Solve  
′ 1 1
~x = A~x, A= .
4 1
First, find out the eigenvalues of A. By an example in 7.3, we have
   
1 1
λ1 = −1, λ2 = 3, ~v1 = , ~v1 = ,
−2 2

So the general solution is

   
λ1 t λ2 t −t 1 3t 1
~x = c1 e ~v1 + c2 e ~v2 = c1 e + c2 e .
−2 2

Write it out in components:

x1 (t) = c1 e−t + c2 e3t


.
x2 (t) = −2c1 e−t + 2c2 e3t

Qualitative property of the solutions:

• What happens when t → ∞?

If c2 > 0, then x1 → ∞, x2 → ∞.
If c2 < 0, then x1 → −∞, x2 → −∞.
x1
Asymptotic relation between x1 , x2 : look at x2 :

x1 c1 e−t + c2 e3t
= .
x2 −2c1 e−t + 2c2 e3t
As t → ∞, we have
x1 c2 e3t 1
= 3t
= .
x2 2c2 e 2
This means, x1 → 21 x2 asymptotically.

120
 • What happens when t → −∞?
x1
Looking at x2 , we see as t → −∞ we have

x1 c1 e−t 1
= −t
=− ,
x2 −2c1 e 2

which means, x1 → − 21 x2 asymptotically as t → −∞.

Phase portrait. is the trajectories of various solutions in the x2 − x1 plane.
• Since A is non-singular, then ~x = ~0 is the only critical point such that ~x′ = A~x = 0.

• If c1 = 0, then
x1 c2 e3t 1
= = ,
x2 2c2 e3t 2
so the trajectory is a straight line x1 = 12 x2 .
Note that this is exactly the direction of ~v2 .
Since λ2 = 3 > 0, the trajectory is going away from 0.

• If c2 = 0, then
x1 c1 e−t 1
= −t
=− ,
x2 −2c1 e 2
so the trajectory is another straight line x1 = − 12 x2 .
Note that this is exactly the direction of ~v1 .
Since λ2 = −1 < 0, the trajectory is going towards 0.

• For general cases where c1 , c2 are not 0, the trajectories should start (asymptotically)
from line x1 = − 12 x2 , and goes to line x1 = 12 x2 asymptotically as t grows.
x2
✻
~v2
~v1 ~v2
❑ ✕

❯ ✕
✲

❄ ✲ x1
✻

✛
☛ ❑

~v1

Definition: If A has two real eigenvalues of opposite signs, the origin (critical point) is
called a saddle point.

121
 Notion of stability: (in layman’s term). For solutions nearby a critical point, as time
goes,
(1). If the solutions go away: then it is unstable;
(2). If the solutions approach the critical point: it is asymptotically stable;
(3). If the solution stays nearby, but not approaching the critical point: it is stable, but not
asymptotically.

A saddle point is unstable.

Tips for drawing phase portrait for saddle point: only need the eigenvalues and eigenvec-
tors!
General case: If two eigenvalues of A are λ1 < 0 and λ2 > 0, with two corresponding
eigenvectors ~v1 , ~v2 . To draw the phase portrait, we follow these guidelines:

• The general solution is

~x = c1 eλ1 t~v1 + c2 eλ2 t~v2 .

• If c1 = 0, then the solution is ~x = c2 eλ2 t~v2 . We see that the solution vector is a scalar
multiple of ~v2 . This means a line parallel to ~v2 through the origin is a trajectory. Since
λ2 > 0, solutions |~x| → ∞ along this line, so the arrows are pointing away from the
origin.

• The similar other half: if c2 = 0, then the solution is ~x = c1 eλ1 t~v1 . We see that the
solution vector is a scalar multiple of ~v1 . This means a line parallel to ~v1 through the
origin is a trajectory. Since λ1 < 0, solutions approach 0 along this line, so the arrows
are pointing toward the origin.

• Now these two lines cut the plane into 4 regions. We need to draw at least one trajectory
in each region. In the region, we have the general case, i.e., c1 6= 0 and c2 6= 0. We need
to know the asymptotic behavior. We have

t → ∞, => ~x → c2 eλ2 t~v2

t → −∞, => ~x → c1 eλ1 t~v1

We see these are exactly the two straight lines we just made. This means, all trajectories
come from the direction of ~v1 , and will approach ~v2 as t grows. See the plot below.

122
 x2
✻
~v2
~v1
❑ ~v2
✒
❯ ✒
q

✍ ✲ x1
✌

✐
✠ ❑

~v1

Example 2. Suppose we know the eigenvalues and eigenvectors of A:

   
1 1
λ1 = 3, ~v1 = , λ1 = −3, ~v2 = .
−1 0

Then the phase portrait looks like this:

✻
x2
~v1
■ ✲ ~v2
■

❨
✻
~v2
✲ ✛ ✲ x1

❥ ❄

~v1

If the two real distinct eigenvalue have the same sign, the situation is quite different.

Example 3. Consider the homogeneous system

 
′ −3 2
~x = A~x, A= .
1 −2

123
Find the general solution and sketch the phase portrait.

Answer.

• Eigenvalues of A:
 
−3 − λ 2
det(A−λI) = det = (−3−λ)(−2−λ)−2 = λ2 +5λ+4 = (λ+1)(λ+4) = 0,
1 −2 − λ

So λ1 = −1, λ2 = −4. (Two eigenvalues are both negative!)

• Find the eigenvector for λ1 . Call it ~v1 = (a, b)T ,

         
−3 + 1 2 a −2 2 a 0
(A − λ1 I)~v1 = · = · = .
1 −2 + 1 b 1 −1 b 0

This gives a = b. Choose it to be 1, we get ~v1 = (1, 1)T .

• Find the eigenvector for λ2 . Call it ~v2 = (c, d)T ,

         
−3 + 4 2 c 1 2 c 0
(A − λ2 I)~v1 = · = · = .
1 −2 + 4 d 1 2 d 0

This gives c + 2d = 0. Choose d = 1, then c = −2. So ~v2 = (−2, 1)T .

• General solution is
   
λ1 t λ2 t −t 1 −4t −2
~x(t) = c1 e ~v1 + c2 e ~v2 = c1 e + c2 e .
1 1

Write it out in components:

x1 (t) = c1 e−t − 2c2 e−4t


.
x2 (t) = c1 e−t + c2 e−4t

Phase portrait:

• If c1 = 0, then ~x = c2 eλ2 t~v2 , so the straight line through the origin in the direction of ~v2
is a trajectory. Since λ2 < 0, the arrows point toward the origin.

• If c2 = 0, then ~x = c1 eλ1 t~v1 , so the straight line through the origin in the direction of ~v1
is a trajectory. Since λ1 < 0, the arrows point toward the origin.

• For the general case, when c1 6= 0 and c2 6= 0, we have

t → −∞, => |~x| → ∞, ~x → c2 eλ2 t~v2

t → ∞, => |~x| → 0, ~x → c1 eλ1 t~v1

So all trajectories come into the picture in the direction of ~v2 , and approach the origin
in the direction of ~v1 . See the plot below.

124
 ✻
x2
~v1
~v2 ~v1
❨ ✒
✠
❄
❥ ✛

✲ x1

✲ ❨

✻ ~v2
✒

In the previous example, if λ1 > 0, λ2 > 0, say λ1 = 1 and λ2 = 4, and ~v1 , ~v2 are the same,
then the phase portrait will look the same, but with all arrows going away from 0.

Definition: If λ1 6= λ2 are real with the same sign, the critical point ~x = 0 is called a
node.
If λ1 > 0, λ2 > 0, this node is called a source.
If λ1 < 0, λ2 < 0, this node is called a sink.
A sink is stable, and a source is unstable.

Example 4. (Source node) Suppose we know the eigenvalues and eigenvectors of A are
   
1 1
λ1 = 3, λ2 = 4, ~v1 = , ~v2 = .
2 −3

(1) Find the general solution for ~x′ = A~x, (2) Sketch the phase portrait.

Answer. (1) The general solution is simple, just use the formula
   
λ1 t λ2 t 3t 1 4t 1
~x = c1 e ~v1 + c2 e ~v2 = c1 e + c2 e .
2 −3

(2) Phase portrait: Since λ2 > λ1 , then the solution approach ~v2 as time grows. As
t → −∞, ~x → c1 eλ1 t~v1 . See the plot below.

125
 ✻
x2
~v1 ~v2 ~v1
▼ ✕
▼
✕
✻
✲

✲ x1

❄
☛
◆

~v2

Summary:
(1). If λ1 and λ2 are real and with opposite sign: the origin is a saddle point, and it’s unstable;
(2). If λ1 and λ2 are real and with same sign: the origin is a node.
If λ1 , λ2 > 0, it’s a source node, and it’s unstable;
If λ1 , λ2 < 0, it’s a sink node, and it’s asymptotically stable;

7.6 Complex eigenvalues

If A has two complex eigenvalues, they will be a pair of complex conjugate numbers, say
λ1,2 = α ± iβ, β 6= 0.
The two corresponding eigenvectors will also be complex conjugate, i.e,
¯ 2.
~v1 =~v

We have two solutions

~z1 = eλ1 t~v1 , ~z2 = eλ2 t~v2 .
They are complex-valued functions, and they also are complex conjugate. We seek real-valued
solutions. By the principle of superposition,
1 1
~y1 = (~z1 + ~z2 ) = Re(~z1 ), ~y2 = (~z1 − ~z2 ) = Im(~z1 )
2 2i
are also two solutions, and they are real-valued.
One can show that they are linearly independent, so they form a set of fundamental
solutions. The general solution is then ~x = c1 ~y1 + c2 ~y2 .
Now let’s derive the formula for the general solution. We have two eigenvalues: λ and λ̄,
two eigenvectors: ~v and ~v¯, which we can write

λ = α + iβ, ~v = ~vr + i~vi .

126
One solution can be written

~z = eλt~v = e(α+iβ)t (~vr + i~vi ) = eαt (cos βt + i sin βt) · (~vr + i~vi )
= eαt (cos βt · ~vr − sin βt · ~vi ) + ieαt (sin βt · ~vr + cos βt · ~vi ) .

The general solution is

~x = c1 eαt (cos βt · ~vr − sin βt · ~vi ) + c2 eαt (sin βt · ~vr + cos βt · ~vi ) .

Notice now if α = 0, i.e., we have pure imaginary eigenvalues. Then ~x is a harmonic

oscillation, which is a periodic function. This means in the phase portrait all trajectories are
closed curves.

Example 1. (pure imaginary eigenvalues.) Find the general solution and sketch the phase
portrait of the system:  
′ 0 −4
~x = A~x, A= .
1 0

Answer. First find the eigenvalues of A:

det(A − λI) = λ2 + 4 = 0, λ1,2 = ±2i.

Eigenvectors: need to find one ~v = (a, b)T for λ = 2i:

     
−2i −4 a 0
(A − λI)~v = 0, · = .
1 −2i b 0

a − 2ib = 0, choose b = 1, then a = 2i,

then      
2i 0 2
~v = = +i .
1 1 0
The general solution is
         
0 2 0 2
~x = c1 cos 2t · − sin 2t · + c2 sin 2t · + cos 2t · .
1 0 1 0

Write out the components, we get

x1 (t) = −2c1 sin 2t + 2c2 cos 2t

x2 (t) = c1 cos 2t + c2 sin 2t.

Phase portrait:
• ~x is a periodic function, so all trajectories are closed curves around the origin.

• They do not intersect with each other. This follows from the uniqueness of the solution.

• They are ellipses. Because we have the relation:

(x1 /2)2 + (x2 )2 = constant.

127
 • The arrows are pointing either clockwise or counter clockwise, determined by A. In
this example, take ~x = (1, 0)T , a point on the x1 -axis. By the differential equations, we
get ~x ′ = A~x = (0, 1)T , which is a vector pointing upward. So the arrows are counter-
clockwise.
See plot below.
x2

x1

Definition. The origin in this case is called a center. A center is stable (b/c solutions
don’t blow up), but is not asymptotically stable (b/c solutions don’t approach the origin as
time goes).

If the complex eigenvalues have non-zero real part, the situation is still different.

Example 2. Consider the system

 
′ 3 −2
~x = A~x, A= .
4 −1
First, we compute the eigenvalues:
det(A − λI) = (3 − λ)(−1 − λ) + 8 = λ2 − 2λ + 5 = 0,
λ1,2 = 1 ± 2i, ⇒ α = 1, β = 2.
Eigenvectors: need to compute only one ~v = (a, b)T . Take λ = 1 + 2i,
     
2 − 2i −2 a 0
(A − λI)~v = · = ,
4 −2 − 2i b 0
(2 − 2i)a − 2b = 0.
Choosing a = 1, then b = 1 − i, so
     
1 1 0
~v = = +i .
1−i 1 −1
So the general solution is:
         
t 1 0 t 1 0
~x = c1 e cos 2t · − sin 2t · + c2 e sin 2t · − cos 2t ·
1 −1 1 −1
   
cos 2t sin 2t
= c1 et + c2 et .
cos 2t + sin 2t sin 2t − cos 2t

128
 Phase portrait. Solution is growing oscillation due to the et . If this term is not present,
(i.e., the eigenvalues would be pure imaginary), then the solutions are perfect oscillations,
whose trajectory would be closed curves around origin, as the center. But with the et term,
we will get spiral curves. Since α = 1 > 0, all arrows are pointing away from the origin.
To determine the direction of rotation, we need to go back to the original equation and
take a look at the directional field.
Consider the point (x1 = 1, x2 = 0), then ~x′ = A~x = (3, 4)T . The arrow should point up
with slope 4/3.
At the point ~x = (0, 1)T , we have ~x′ = (−2, −1)T .
Therefore, the spirals are rotating counter clockwise. We don’t stress on the exact shape
of the spirals. See plot below.

2.5

1.5

0.5
x2

−0.5

−1

−1.5

−2

−2.5
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
x1

In this case, the origin (the critical point) is called the spiral point. The origin in this
example is an unstable critical point since α > 0.
Remark: If α < 0, then all arrows will go towards the origin. The origin will be a stable
critical point. An example is provided in the text book. We will go through it here.

Example 3. Consider
− 21 1
 
′
~x = ~x.
−1 − 21
The eigenvalues and eigenvectors are:
     
1 1 1 0
λ1,2 = − ± i, ~v = = ±i .
2 ±i 0 1

Since the formula for the general solution is not so “friendly” to memorize, we use a different
approach.

129
 We know that one solution is
   
λ1 t −( 12 +i)t 1 0
~z = e ~v1 = e ±i .
0 01

This is a complex values function. We know the real part and the imaginary part are both
solutions, so work them out:
        
− 12 t 1 0 1 0
~z = e cos t − sin t + i sin t + i cos t .
0 1 0 1

The general solution is:

         
− 12 t 1 0 − 12 t 1 0
~x = c1 e cos t − sin t + c2 e sin t + cos t ,
0 1 0 1

and we can write out each component

1
x1 (t) = e− 2 t (c1 cos t + c2 sin t)
1
x2 (t) = e− 2 t (−c1 sin t + c2 cos t)

Phase portrait: If c1 = 0, we have

1 1
x21 + x22 = (e− 2 t )2 c22 (sin2 t + cos2 t) = (e− 2 t )2 c22 .

If c2 = 0, we have
1
x21 + x22 = (e− 2 t )2 c21 .
In general, if c1 6= 0 and c2 6= 0, we can show:
1
x21 + x22 = (e− 2 t )2 (c21 + c21 ).

The trajectories will be spirals, with arrows pointing toward the origin. To determine with
direction they rotate, we check a point on the x1 axis:
   1 
1 ′ −2
~x = , ~x = A~x = .
0 −1

So the spirals rotate clockwise. And the origin is a stable equilibrium point. See the picture
below.

130
 50

40

30

20

10

−10

−20

−30

−40

−50
−50 −40 −30 −20 −10 0 10 20 30 40 50

Summary: For complex roots: r1,2 = α ± iβ.

(i) If α = 0: the origin is a center. It’s stable, but not asymptotically stable.
(ii) If α > 0: the origin is a spiral point. It’s unstable.
(iii) If α < 0: the origin is a spiral point. It’s asymptotically stable.

Connections. At this point, we could make some connections between the 2nd order
equations and the 2 × 2 system of 1st order equations. Consider a 2nd order homogeneous
equation with constant coefficients
ay ′′ + by ′ + cy = 0. (7.1)
The characteristic equation is
ar 2 + br + c = 0. (7.2)
We know that the solutions depend mainly on the roots of the characteristic equation. We
had detailed discussions in Chapter 3.
We can perform the standard variable change, and rewrite this into a system. Indeed, let
x1 = y, x2 = y ′
we get
x′1 = x2

(7.3)
x′2 = − ab x2 − ac x1
In matrix-vector notation, this gives
 
′ 0 1
~x = A~x, A= (7.4)
− ac − ab
We now compute the eigenvalues of A. We have
b c
det(A) = −λ(− − λ) + = 0, ⇒ aλ2 + bλ + c = 0.
a a
We see that the eigenvalues are the same as the roots for the characteristic equation in (7.2).

131
7.7 Fundamental Matrices*
The concepts discussed here are a bit abstract. it requires some fluent knowledge on linear
algebra. We will make the discussions in connection with linear homogeneous system with
constant coefficients, i.e., ~x′ = A~x, where A is an n × n matrix. (A more general result is
possible, for variable coefficients A(t), but we will not go into that.)
Let ~x(i) (t) for i = 1, · · · , n be a set of fundamental solutions, i.e.,
d (i)
~x (t) = A~x(i) (t).
dt
We define a matrix Ψ(t) whose ith-column is the vector ~x(i) (t), for i = 1, · · · , n. This matrix
is called the fundamental matrix for the system. Note that the fundamental matrix is not
unique.

One interesting property of the fundamental matrix Ψ(t) is that it satisfies the same
ODE, i.e.,
Ψ′ (t) = AΨ(t).
This is true because each column of Ψ(t) satisfies the same equation,
  h
′ d (1) d (n) i h i
Ψ (t) = ~x (t); · · · ; ~x (t) = A~x(1) (t); · · · ; A~x(n) (t) = A ~x(1) (t); · · · ; ~x(n) (t) = AΨ(t).
dt dt

The general solution takes the form

n
X
~x = c1 ~x(1) (t) + · · · + cn ~x(n) (t) = ci ~x(i) (t)
i=1

which could be written in matrix-vector form, such as

~x = Ψ(t)~c, ~c = (c1 , · · · , cn )T .

Now, if we are given an initial condition,

~x(0) = ~x0 ,

then the constant vector ~c could be solved:

Ψ(0)~c, ~c = ~x0 , ~c = Ψ−1 (0)~x0 .

And the solution can be written

~x = Ψ(t)Ψ−1 (0)~x0 = Φ(t)~x0 ,

where Φ(t) is a special fundamental matrix, defined as

˙ Ψ(t)Ψ−1 (0),
Φ(t) =

which clearly has the property

Φ(0) = I.

132
 The Matrix exponential exp(At): Recall the scalar equation and its solution

x′ = ax, x(0) = x0 , x(t) = x0 eat .

Comparing this to the solution of ~x′ = A~x, i.e.,

~x = Φ(t)~x0 ,

it suggests that the matrix-valued function Φ(t) might have an exponential character. We will
explorer this aspect.
Recall the power expansion (Taylor series) for the exponential function
∞
X ak tk
exp(at) = 1 + ,
k!
k=1

which converges for all t.

Let’s replace the value a with our matrix A, and consider the series
∞
X Ak t k A2 2 Ak t k
I+ = I + At + t + ··· + + ··· .
k! 2! k!
k=1

Each term is an n × n matrix. Assuming that the series converges for all t, (which could
actually be proved!), then the series defines a matrix-valued function. We now define
∞
X Ak t k
exp(At) =
˙ I+ .
k!
k=1

Some properties of the matrix exponential function, in correspondence to regular

exponential function: (1) First, we have

exp(At) = I.

t=0

(2) If A is a diagonal matrix, i.e., A = D = diag(d1 , · · · , dn ), then each term in the series is
a diagonal matrix, and we have
∞ ∞ ∞
!
X diag(dk1 , ·, dkn )tk X dk1 tk X dkn tk  
exp(Dt) = I+ = diag 1 + ,··· ,1 + = diag ed1 t , · · · , edn t .
k! k! k!
k=1 k=1 k=1

(3) Differentiating the series:

∞ ∞ ∞ ∞
" #
Ak t k kAk tk−1 Ak tk−1 Ak tk
 
d X d X X X
exp(At) = = = =A I+ = A exp(At)
dt dt k! k! (k − 1)! k!
k=1 k=1 k=1 k=1

which gives
d
exp(At) = A exp(At).
dt
Recall that the fundamental matrix Φ(t) satisfies

Φ′ = AΦ, Φ(0) = I.

133
 Thus, Φ(t) and exp(At) solves the same IVP. By uniqueness, we have

Φ(t) = exp(At). (A1)

The solution can now be written

~x = exp(At)~x0 .

Diagonalizable matrices. If A is not a diagonal matrix, then the system ~x′ = A~x is
called coupled, meaning that different xi ’s effect each other, and they must be solved simulta-
neously. This presents some difficulty. For some matrices A, one could decouple it through a
variable change, and make it a diagonal system (which is uncoupled, such that each equation
could be solved separately.)
Assume now A has n eigenvalues λi and n linearly independent eigenvectors ~v (i) . We form
a matrix T whose columns are eigenvectors,
 
T= ˙ ~v (1) ; · · · ; ~v (n) .

Since the eigenvectors are linearly independent, the matrix T is non-singular and invertible.
We have    
AT = A~v (1) ; · · · ; A~v (n) = λ1~v (1) ; · · · ; λn~v (n) = TD, (∗)

where D is a diagonal matrix with the eigenvalues on the diagonal

D = diag(λ1 , · · · , λn ).

Multiply both side of (*) by T−1 on the left, we get

T−1 AT = D,

so we transformed A into a diagonal matrix. This process is called diagonalization, and such
a matrix A is called diagonalizable.

We now consider the system of ODEs

~x′ = A~x, ~x(0) = ~x0 . (Ix)

where A is diagonalizable such that

T−1 AT = D, ⇒ A = TDT−1 .

The system can be written as

~x′ = TDT−1 ~x, ⇒ T−1 ~x′ = DT−1 ~x.

We now make a variable change, and let

~y = T−1 ~x, ⇒ ~x = T~y.

134
Then ~y solves the diagonal (decoupled) system

y~′ = D~y , ~y (0) = ~y0 = T−1 ~x0 , (Iy)

which could be easily solved

 
~y (t) = exp(Dt)~y0 = diag eλ1 t , · · · , eλn t ~y0 = Q(t)~y0 , Q(t) = exp(Dt),

where Q(t) is the fundamental matrix for system (Iy). The solution of ~x is then recovered as

~x(t) = T~y (t) = T exp(Dt)T−1 ~x0 = Φ(t)~x0 , Φ(t) = T exp(Dt)T−1 (A2)

where Φ(t) is the corresponding fundamental matrix for system (Ix).

Remark. Note that, combining the equations (A1) and (A2), we accidentally proved that,
if A is diagonalizable then it holds

exp(At) = T exp(Dt)T−1 , where A = TDT−1 .

7.8 Repeated eigenvalues

Here we study the case where the two eigenvalues are the same, say λ1 = λ2 = λ. This can
happen, as we will see through our first example. We have two cases, depending on whether we
could find 2 linearly independent eigenvectors. We will demonstrate this through examples.

Example 1. (Repeated eigenvalues, with 2 linearly independent eigenvectors) Let

 
1 0
A= .
0 1

We see that the two eigenvalues and eigenvectors are

λ1 = λ2 = 1, ~v (1) = (1, 0)T , ~v (2) = (0, 1)T .

The general solution is

~x(t) = c1 ~z1 + c2 ~z2 , ~z1 = et~v (1) , ~z2 = et~v (2) .

Writing out, we have

x1 (t) = c1 et , x2 (t) = c2 et .
Note that the system is uncoupled!
We see that,
x2 (t) c2 et c2
= t
= =c
x1 (t) c1 e c1
where c is arbitrary. Therefore, the trajectories of the solutions are straight lines through the
origin in the phase plane, forming a star shape. Such a critical point is called a proper node
or a star node.

135
 Example 2. (Repeated eigenvalues, with one eigenvector) Let
 
1 −1
A= .
1 3
Then
 
1 − λ −1
det(A − λI) = det = (1 − λ)(3 − λ) + 1 = λ2 − 4λ + 3 + 1 = (λ − 2)2 = 0,
1 3−λ
so λ1 = λ2 = 2. And we can find only one eigenvector ~v = (a, b)T
   
−1 −1 a
(A − λI)~v = · = ~0, a + b = 0.
1 1 b
 
1
Choosing a = 1, then b = −1, and we find ~v = . Then, one solution is:
−1
 
λt 2t 1
~z1 = e ~v = e .
−1
We need to find a second solution. Let’s try ~z2 = teλt~v . We have
~z2′ = eλt~v + λteλt~v = (1 + λt)eλt~v
A~z2 = Ateλt~v = teλt (A~v ) = teλt λ~v = λteλt~v
If ~z2 is a solution, we must have
~z2′ = A~z2 → 1 + λt = λt
which doesn’t work.
Try something else: ~z2 = teλt~v + η~eλt . (here ~η is a constant vector to be determined later).
Then
~z2′ = (1 + λt)eλt~v + λ~η eλt = λteλt~v + eλt (~v + λ~η )
A~z2 = λteλt~v + A~η eλt .
Since ~z2 is a solution, we must have ~z2′ = A~z2 . Comparing terms, we see we must have
~v + λ~η = A~η, (A − λI)~η = ~v .
This is what one uses to solve for ~
η . Such an ~η is called a generalized eigenvector corresponding
to the eigenvalue λ.
Back to the original problem, to compute this ~η , we plug in A and λ, and get
     
−1 −1 η1 1
· = , η1 + η2 = −1.
1 1 η2 −1
 
0
We can choose η1 = 0, then η2 = −1, and so ~η = .
−1
So the general solution is
~x = c1 ~z1 + c2 ~z2 = c1 eλt~v + c2 (teλt~v + eλt ~η )
      
1 1 0
= c1 e2t + c2 te2t + e2t .
−1 −1 −1
Phase portrait:

136
 • As t → ∞, we have |~x| → ∞ unbounded.

• As t → −∞, we have ~x → 0.

• If c2 = 0, then ~x = c1 eλt~v , so the line through the origin in the direction of ~v is a

trajectory. Since λ > 0, the arrows point away from the origin.

• If c1 = 0, then ~x = c2 (teλt~v + eλt ~η ). For this solution, as t → ∞, the dominant term in

~x is teλt~v . This means the solution approach the direction of ~v . On the other hand, as
t → −∞, the dominant term in ~x is still teλt~v . This means the solution approach the
direction of ~v . But, due to the change of sign of t, the ~x will change direction and point
toward the opposite direction as when t → ∞.
How does it turn? We need to go back to the system and check the directional field. At
~x = (1, 0), we have ~x′ = (1, 1)T , and at ~x = (0, 1), we have ~x′ = (−1, 3)T . There it turns
kind of counter clockwise. See figure below.

• For the general case, with c1 6= 0 and c2 6= 0, a similar thing happens. As t → ∞, the
dominant term in ~x is teλt~v . This means the solution approach the direction of ~v . As
t → −∞, the dominant term in ~x is still teλt~v . This means the solution approach the
direction of ~v . But, due to the change of sign of t, the ~x will change direction and point
toward the opposite direction as when t → ∞. See plot below.

1
x2

−1

−2

−3 z2

−4 z1

−5
−5 −4 −3 −2 −1 0 1 2 3 4 5
x1

Remark: If λ < 0, the phase portrait looks the same except with reversed arrows.

Summary: Let A has repeated eigenvalues.

Case (1). If there are 2 linearly independent eigenvectors, the origin is called a proper node or
a star point.

137
Case (2). If there is only one eigenvector, then the origin is called a improper node.
The critical point is stable if λ < 0, and unstable if λ > 0.

Recipe for solutions of ~x′ = A~x where A has repeated eigenvalues.

1. Find the eigenvalue λ, by det(A − λI) = 0;

2. Find the eigenvector ~v , by (A − λI)~v = ~0;

3. Find the generalized eigenvector ~η , by (A − λI)~η = ~v ;

4. Form the general solution:

~x = c1 eλt~v + c2 (teλt~v + eλt ~η ).

5. Discuss stability of the critical point: Asymptotically stable if λ < 0, unstable if λ > 0.

 
−2 2
Example 2. Find the general solution to the system ~x′ = ~x.
−0.5 −4
We start with finding the eigenvalues:

det(A − λI) = (−2 − λ)(−4 − λ) + 1 = λ2 + 6λ + 8 + 1 = (λ + 3)2 = 0, λ1 = λ2 = λ = −3

We see we have double eigenvalue. The corresponding eigenvector ~v = (a, b)T

       
−2 + 3 2 a 1 2 a
(A − λI)~v = · = · = ~0
−0.5 −4 + 3 b −0.5 −1 b
 
2
So we must have a + 2b = 0. Choose a = 2, then b = −1, and we get ~v = . To find
−1
the generalized eigenvector ~ η , we solve
     
1 2 η1 2
(A − λI)~ η = ~v , · = .
−0.5 −1 η2 −1
 
0
This gives us one relation η1 +2η2 = 2. Choose η1 = 0, then we have η2 = 1, and so ~η = .
1
The general solution is
      
2 2 0
~x = c1 eλt~v + c2 (teλt~v + eλt ~η ) = c1 e−3t + c2 te−3t + e−3t .
−1 −1 1

The origin is an improper node which is asymptotically stable.

Fundamental matrices and Jordan Forms. (skip).

More on eigenvalues. Consider a 2 × 2 matrix

 
a b
A= .
c d

138
Define
p = a + d = trace(A), q = ad − bc = det(A).
Then, the eigenvalues are the roots of the characteristic polynomial
det(A − λI) = (λ − a)(λ − b) − bc = λ2 − (a + d)λ + ad − bc = λ2 − pλ + q
Let λ1 , λ2 be the two eigenvalues. Then we must have
λ1 + λ2 = p, λ1 λ2 = q. (∗)
Let ∆ = p2 − 4q denote the discriminant of the characteristic polynomial. Then, the two roots
can be written as √
p± ∆
λ1,2 = .
2
We now discuss several cases.
• If ∆ < 0 and p = 0, we have pure imaginary eigenvalues. The equilibrium is a center.
• If ∆ < 0 and p > 0, we have complex eigenvalues with positive real parts. The equilib-
rium is a spiral point, which is unstable.
• If ∆ < 0 and p < 0, we have complex eigenvalues with negative real parts. The equilib-
rium is a spiral point, which is asymptotically stable.
• If ∆ > 0, then the two eigenvalues are real and distinct. Using the relation (*), we see
that:
– If q < 0, then λ1 , λ2 have the opposite sign, and we have a saddle point which is
unstable.
– If q > 0, then λ1 , λ2 have the same sign, and we have a proper node. If p > 0, they
are positive, so the node is unstable. Otherwise, if p < 0, they are negative and we
have a stable node.
• Finally, if ∆ = 0, the eigenvalues are repeated, and we have either a proper node or an
improper node. If p > 0, it is unstable, and if p < 0, it is asymptotically stable.
See graph below for an illustration.

139
 7.9 Summary of Stabilities and types of critical points for lin-
ear systems
For the 2 × 2 system
~x′ = A~x
we see that ~x = (0, 0) is the only critical point if A is invertible.
In a more general setting: the system

~x′ = A~x − ~b

would have a critical point at ~x = A−1 b.

The type and stability of the critical point is solely determined by the eigenvalues of A.
Below is a summary of what we learned in Chapter 7:

Summary of types and stabilities of critical points:

λ1,2 eigenvalues type of C.P. stability

real λ1 · λ2 < 0 saddle point unstable

real λ1 > 0, λ2 > 0, λ1 6= λ2 node (source) unstable

real λ1 < 0, λ2 < 0, λ1 6= λ2 node (sink) A.S.(asymptotically stable)

real λ1 = λ2 , 2 eigenvectors proper node/star point A.S. if λ1 < 0, unstable if λ1 > 0

real λ1 = λ2 , one eigenvector improper node A.S. if λ1 < 0, unstable if λ1 > 0

imaginary λ1,2 = ±iβ center stable but not asymptotically

complex λ1,2 = α ± iβ spiral point A.S. if α < 0, unstable if α > 0

As long stability is concerned, the sole factor is the sign of the real part of the eigenvalues.
If any of eigenvalue shall have a positive real part, the it is unstable.

140
9.2: Autonomous systems and their critical points
Let x(t), y(t) be the unknowns, we consider the system
 ′ 
x (t) = F (x, y) x(t0 ) = x0 ,
y ′ (t) = G(x, y) y(t0 ) = y0 .

for some functions F (x, y), G(x, y) that do not depend on t. Such a system is called au-
tonomous. Typical examples are in population dynamics, which we will see in our examples.
Using matrix-vector form, one could also write an autonomous system as

~x′ (t) = F~ (~x), ~x(t0 ) = ~x0 .

A critical point is a point such that the righthand-side is 0, i.e.,

F (x, y) = 0, G(x, y) = 0

or in the vector notation

F~ (~x) = 0.
Note that, since now the functions are non-linear, there could be multiple critical points.
Finding zeros for a nonlinear vector-valued function could be a non-trivial task.
We first go through some examples on how to find the critical points.

Example 1. Find all critical points for

 ′
x (t) = −(x − y)(1 − x − y)
y ′ (t) = x(2 + y)

Answer. We see that the righthand-sides are already in factorized form, which makes our
task easier. We must now require

x = y, or x−y =1

and
x=0 or y = −2.
We see that we have 4 combinations.
 
x=y x=0
(1) ⇒
x=0 y=0
 
x=y x = −2
(2) ⇒
y = −2 y = −2
 
x+y =1 x=0
(3) ⇒
x=0 y=1
 
x+y =1 x=3
(4) ⇒
y = −2 y = −2

141
 General strategy.
(1) Factorize the righthand as much as you can.
(2) Find the conditions for each equation.
(3) Make all combinations and solve.

Example 2. Find all critical points for

 ′
x (t) = xy − 6x
y ′ (t) = xy − 2x + y − 2

Answer. The righthand-sides are not factorized, so we need to do it first. We get

 ′
x (t) = x(y − 6)
y ′ (t) = (xy − 2x) + (y − 2) = x(y − 2) + (y − 2) = (x + 1)(y − 2)

The conditions are:

x=0 or y=6
and
x = −1 or y = 2.
In principle we could make 4 combinations, but only 2 of them would give us a critical points.
We end up with 2 critical points:

(x, y) = (0, 2), (x, y) = (−1, 6).

Example 3. Find all critical points for

 ′
x (t) = x2 − xy
y ′ (t) = xy − 3x + 2

Answer. We factorize first, and get

 ′
x (t) = x(x − y)
y ′ (t) = xy − 3x + 2

The conditions are

x=0 or x=y
and
xy − 3x + 2 = 0.
We have 2 combinations. One is

x=0 and xy − 3x + 2 = 0,

which gives no answer. The second combination is

x=y and xy − 3x + 2 = 0,

142
which gives
x2 − 3x + 2 = 0, (x − 1)(x − 2) = 0, x = 1 or x = 2.
This gives us two critical points

(x, y) = (1, 1), (x, y) = (2, 2).

Example 4. (Competing species) Let x(t), y(t) be the population densities of two species
living in a common habitat, using the same natural resource. They are not in a prey-predator
relation. They simply compete with each other for the resources. The model is
 ′
x (t) = x(M − ax − by)
y ′ (t) = y(N − cx − dy)

Here a, b, c, d, M, N are positive constants, with physical meanings.

If species x lives alone, i.e., y = 0, we get x′ = x(M − ax). The growth rate is M − ax. At
x = M/a, the rate will be 0. So one can view M/a as the max population density of x that
the habitat could support in absence of y. Adding y, the rate of growth would decrease.
A similar argument holds for the 2nd equation, where y = N/d would be the max density
of y in absence of x.
We search for critical points. They must satisfies

x=0 or M − ax − by = 0

and
y=0 or N − cx − dy = 0.
We now have 4 combinations:

(1) (x, y) = (0, 0)

(2) x = 0 and N − cx − dy = 0, → (x, y) = (0, N/d)
(3) M − ax − by = 0 and y = 0, → (x, y) = (M/a, 0)
(4) M − ax − by = 0 and N − cx − dy = 0.

In case (4) we need to solve a linear equation for (x, y).

Example 5. Prey-predator model of Lotka-Volterra. Let x(t) be the population density of

a prey which lives together with a predator, represented by y(t) as its population density. The
predator lives on eating the prey, while the prey lives on natural resource which we assume is
abundant. We have the model
 ′
x (t) = x(a − by)
y ′ (t) = y(−c + dx)

Physical meaning of the model:

(1) If there is no predator, i.e, y = 0, then x′ = at, so the prey grows exponentially with rate a.
The existence of the predator has a negative effect on the growth rate of the prey (represented
by the term −by). (2) If there is no prey, then, y ′ = −cy, the predator decays exponentially

143
(dies out). The existence of the prey has a positive effect on the growth rate of the predator
(represented by the term dx).
Find the critical points: We have the conditions

x=0 or a − by = 0

and
y=0 or − c + dx = 0.
Only 2 out of the 4 combinations give the critical point. The 2 critical points are

(x, y) = (0, 0), (x, y) = (d/c, b/a).

9.3: Stability of Critical points; local linearization

Moral: In a small neighborhood of the critical point, the nonlinear system behaves in a similar
way to a linearized system.
Linearization. Consider a scalar function f (x), and let x0 be a root such that f (x0 ) = 0.
Assuming that f is smooth near x = x0 , we see that we can approximate f with a straight
line through x0 , with the slope equals to f ′ (x0 ). Draw a graph to see the idea. Then, we have

f (x) ≈ f ′ (x0 )(x − x0 )

in a small neighborhood of x = x0 .
The idea can be extended to vector valued functions. Here, the derivative of the vector-
valued function, however, takes a more complicated form. Using our notation, consider the
system  ′
x (t) = F (x, y)
(A)
y ′ (t) = G(x, y)
Let (xo , yo ) be a critical point such that F (xo , yo ) = 0, G(xo , yo ) = 0.
We introduce the concept of the Jacobian Matrix, defined as
 
. Fx (x, y) Fy (x, y)
J(x, y) =
Gx (x, y) Gy (x, y)

This matrix serves as the derivative of the vector-valued function on the RHS of the system.
We say that we linearize the system (A) at the point (xo , yo ) as
 ′  
x x − xo
= J(xo , yo )
y y − yo

The type and stability of the critical point (xo , yo ) is determined by the eigenvalues of the
Jacobian matrix J(xo , yo ), evaluated at the critical point (xo , yo ).

Example 1. We revisit Example 1 previous chapter,

 ′
x (t) = −(x − y)(1 − x − y)
y ′ (t) = x(2 + y)

144
where the 4 critical points are

(0, 0), (−2, −2), (0, 1), (3, −2).

We now determine their type and stability. We first compute the Jacobian matrix. We have

Fx = −1 + 2x, Fy = 1 − 2y, Gx = 2 + y, Gy = x

so  
−1 + 2x 1 − 2y
J(x, y) = .
2+y x
At (0, 0), we have
 
−1 1
J(0, 0) = , λ1 = 1, λ2 = −2, saddle point, unstable.
2 0

At (−2, −2), we have

 
−5 5
J(−2, −2) = , λ1 = −5, λ2 = −2, nodal sink, asymp. stable.
0 −2

At (0, 1), we have

 
−1 1
J(0, 1) = , λ2 + λ + 3 = 0.
3 0

The eigenvalues are complex with negative real parts. This is a spiral point. It is asymptoti-
cally stable.
At (3, −2), we have
 
5 5
J(3, −2) = , λ1 = 5, λ2 = 3, nodal source, unstable.
0 3

Example 2. Consider
x′ (t) = xy − 6x


y ′ (t) = xy − 2x + y − 2
whose critical points are
(0, 2), (−1, 6).
To check their type and stability, we compute the Jacobian matrix
 
y−6 x
J(x, y) = .
y−2 x+1

At (0, 2), we have

 
−4 0
J(0, 2) = , λ1 = −4, λ2 = 1, saddle point, unstable.
0 1

145
 At (0, 2), we have
 
0 −1
J(−1, 6) = , λ1,2 = ±2i, center, stable but not asymp..
4 0

Example 3. We now consider again the prey-predator model, and set in values for the
constants. We consider1  ′
x (t) = x(10 − 5y)
y ′ (t) = y(−6 + x)
which has 2 critical points (0, 0) and (6, 2). The Jacobian matrix is
 
10 − 5y −5x
J(x, y) = .
y −6 + x

At (0, 0) we have
 
10 0
J(0, 0) = , λ1 = 10, λ2 = −6, saddle point, unstable.
0 −6

At (6, 2) we have
√
 
0 −30
J(6, 2) = , λ1,2 = ±i 60, center, stable but not asymp..
2 0

To see more detailed behavior of the model, we compute the two eigenvector for J(0, 0),
and get ~v1 = (1, 0) and ~v2 = (0, 1). We sketch the trajectories of solution in (x1 , x2 )-plane in
the next plot, where the trajectories rotate around the center counter clock wise.

One can interpret these as “circles of life”.

In particular, the big circles can be interpreted as: When there are very little predators,
the prey grows exponentially, very quickly. As the population of the prey becomes very large,
there is a lot of food for the prey, and this triggers an sudden growth of the predator. As
1
We remark that in real models the coefficients are much smaller than the ones we use here. We choose
these numbers, so we can have friendly numbers in the computation. One can also view this as a rescaled model
where one stretches the time variable.

146
the predators increase their numbers, the prey population shrinks, until there is very little
prey left. Then, the predators starve, and its population decays exponentially (dies out). The
circle continuous in a periodic way, forever!

Example 4*. As a final example, we consider the model of two competing species.
Suppose that in some enclosed environment there are two species that are competing for
natural resource, and let x(t) and y(t) be their populations at time t. Assume that, if living
alone, each species grow following the logistic equation.

x′ (t) = x(a1 − b1 x)
y ′ (t) = y(a2 − b2 y)

where (a1 , a2 ) are the growth rates and (a1 /b1 , a2 /b2 ) are the habitat’s capacities, for x and
y, respectively. When they are competing, then each species will have a negative effect on the
other species. Thus, the system is modified into

x′ (t) = x(a1 − b1 x − c1 y)
y ′ (t) = y(a2 − b2 y − c2 x)

Here all the coefficients are positive constants.

The Jacobian matrix is
 
a1 − 2b1 x − c1 y −c1 x
J(x, y) = .
−c2 y a2 − 2b2 y − c2 x

We now find all the critical points. We have

x = 0, or a1 − b1 x − c1 y = 0

and
y = 0, or a2 − b2 y − c2 x = 0.
We have 4 critical points. The first three are
   
a2 a1
(x1 , y1 ) = (0, 0), (x2 , y2 ) = 0, , (x3 , y3 ) = ,0 ,
b2 b1

and the last one is the solution of the system

a1 = b1 x + c1 y, a2 = b2 y + c2 x.

Depending on the constants, it might or might not have a solution in the first quadrant where
x > 0, y > 0. Assuming that b1 b2 − c1 c2 6= 0, the 4th critical point is (x4 , y4 ) where

a2 b2 − a1 c1 b1 a2 − a1 c2
x4 = , y4 = .
b1 b2 − c1 c2 b1 b2 − c1 c2
If b1 b2 − c1 c2 = 0, then there might be no solutions or infinitely many, depending on the values
of a1 , a2 .

147
 We see that the critical points depend on the values of the coefficients. We now consider
two different sets of constants, which would lead to very different dynamics of the populations.

Case (1). Consider the following model:

x′ (t) = x(1 − x − y)
y ′ (t) = y(0.75 − y − 0.5x)

The 4 critical points are

(x1 , y1 ) = (0, 0), (x2 , y2 ) = (0, 0.75) , (x3 , y3 ) = (1, 0) , (x4 , y4 ) = (0.5, 0.5).

For CP1, the Jacobian matrix is:

 
1 0
J1 = J(0, 0) = .
0 0.75

which gives the two eigenpairs

λ1 = 1, ~v (1) = (1, 0)T , λ2 = 0.75, ~v (2) = (0, 1)T .

This critical point is unstable.

For CP2, the Jacobian matrix is:
 
0.25 0
J2 = J(0, 0.75) = .
−0.375 −0.75

which gives the two eigenpairs

λ1 = 0.25, ~v (1) = (8, −3)T , λ2 = −0.75, ~v (2) = (0, 1)T .

It is a saddle point.
For CP3, the Jacobian matrix is:
 
−1 −1
J3 = J(1, 0) = .
0 0.25

which gives the two eigenpairs

λ1 = −1, ~v (1) = (1, 0)T , λ2 = 0.25, ~v (2) = (4, −5)T .

It is also a saddle point.

For CP4, the Jacobian matrix is:
 
−0.5 −0.5
J4 = J(0.5, 0.5) = .
−0.25 −0.5

which gives the two eigenpairs

√ √
λ1 = −0.146, ~v (1) = ( 2, −1)T , λ2 = −0.854, ~v (2) = ( 2, 1)T .

This is asymptotically stable proper node.

148
 One can sketch a phrase portrait based on these information. See Figure 9.4.2 on p. 524
in textbook.
It is apparent now that CP4 is the attractor for the first quadrant. This means, if x >
0, y > 0 initially, then after a long time they will approach (x4 , y4 ). This is an example of
co-existence of two species.

Case (2). Consider the following model:

x′ (t) = x(1 − x − y)
y ′ (t) = y(0.75 − y − 0.5x)

The 4 critical points are

(x1 , y1 ) = (0, 0), (x2 , y2 ) = (0, 2) , (x3 , y3 ) = (1, 0) , (x4 , y4 ) = (0.5, 0.5).

Straight computation gives now their eigenvalues and eigenvectors

(1) (0, 0) : λ1 = 1, ~v (1) = (1, 0)T , λ2 = 0.75, ~v (2) = (0, 1)T

(2) (0, 2) : λ1 = −1, ~v (1) = (1, 3)T , λ2 = −0.5, ~v (2) = (0, 1)T
(3) (1, 0) : λ1 = −1, ~v (1) = (1, 0)T , λ2 = −0.25, ~v (2) = (4, −3)T
(4) (0.5, 0.5) : λ1 = 0.16, ~v (1) = (1, −1.32)T , λ2 = −0.78, ~v (2) = (1, 0.57)T

We see both CP1 and CP4 are unstable, but both CP2 and CP3 are stable.
One can use these info to sketch the phase portrait. See Figure 9.4.4 on p. 528 in textbook.
This implies that, for all initial data in the first quadrant, some of them will approach
CP2, while some others will approach CP3. There is a curve that separates these two. We
call this curve a separatrix. For initial data on this separatrix, the solution will approach CP4,
but however, such an solution is unstable with respect to small perturbations. This predicts
that one of the species will die out, while the other could survive and live up to the habitat’s
capacity. No co-existence is possible. Note that this dynamics is completely different from
Case (1).

We observe that the dynamics depends on the property of the two lines

(L1) : a1 − b1 x − c1 y = 0, (L2) : a2 − b2 y − c2 x = 0.

The intersection of these two line gives the 4th critical point, if it lies in the first quadrant.
Along L1, we have x′ = 0. Above L1, we have x′ < 0. Below L1, we have x′ > 0.
Along L2, we have y ′ = 0. Above L2, we have y ′ < 0. Below L2, we have y ′ > 0.
These lines are called the x− and y−nullclines.
There are 4 possible configurations. See Figure 9.4.5 on p 529 in textbook. We will take
a close look at each, considering only the first quadrant.
Case (a): L2 lies above L1. There are 3 critical points of interests. We see that along L1,
we have x′ = 0, y ′ > 0. Along L2, we have x′ < 0, y ′ = 0. The directional fields now indicates
that critical point on the y-axis with coordinate (0, a2 /b2 ) is asymptotically stable. Here is
the phase portrait:

149
 1.6

1.5

y 1.4

1.3

1.2

1.1
L2 1

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

L1 0.1

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x

Case (b): L2 lies below L1. There are 3 critical points of interests. We see that along L1,
we have x′ = 0, y ′ < 0. Along L2, we have x′ > 0, y ′ = 0. The directional fields now indicates
that critical point on the x-axis with coordinate (a1 /b1 , 0) is asymptotically stable. Here is
the phase portrait:

0.9
y
0.8

0.7

L1
0.6

0.5

0.4

0.3

0.2

0.1
L2
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
x

Case (c): L1 crosses L2, with the L2 above L1 for small x values, and L2 below L1 for
large x values. There are 4 critical points. Below is the graph of the vector field (on the left)
and the phase portrait (on the right).

150
 2.4

y 2.2

1.8

L2 1.6

1.4

1.2

0.8

0.6

0.4

0.2
L1
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
x

Case (d): L1 crosses L2, with the L2 below L1 for small x values, and L2 above L1 for
large x values. There are 4 critical points. Below is the graph of the vector field (on the left)
and the phase portrait (on the right). We see that this is the only case where co-existence is
possible.

0.9
y
0.8

0.7

L1
0.6

0.5

0.4

0.3

0.2

0.1
L2
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
x

In the textbook there are detailed discussion analyzing the values of the constants a1 , b1 , c1
etc. In conclusion, if the competition is weak, i.e., if b1 b2 > c1 c2 , then coexist is possible.
Otherwise, if the competition is strong, i.e., if b1 b2 < c1 c2 only one species can survive.

151
Chapter 10

Fourier Series

Fourier series will be useful in series solutions for linear 2nd order partial differential equations.

10.1 Introduction and Basic Fourier Series

Objective: representing periodic functions as a series of sine and cosine functions.
Let f (x) be a periodic function with period P , i.e.,

f (x + P ) = f (x), ∀x

Note: If P is a period, so are 2P, 3P, 4P, · · · ,.

The smallest period is called the fundamental period.
Observation: If f (x) and g(x) are both periodic with period P , so will any linear combi-
nation af (x) + bg(x) for arbitrary constants a, b. Also, the product f (x)g(x) is periodic with
the same period.
Known examples of periodic functions: trig functions.
With period 2π:

sin x, sin 2x, sin 3x, · · · , cos x, cos 2x, cos 3x, · · ·

With period 2L:

πx 2πx 3πx πx 2πx 3πx
sin , sin , sin , · · · cos , cos , cos , ···
L L L L L L
We have the trig set:
n mπx mπx o
1, sin , cos , , m = 1, 2, · · ·
L L
Plots of some sine functions are given in Figure 10.1.
Definition. Let f (x) be periodic with period 2L. Fourier series for f (x) is:
∞
a0 X  mπx mπx 
f (x) = + am cos + bm sin . (∗)
2 L L
m=1

Here the constants a0 , am , bm are called: Fourier coefficients.

152
 Figure 10.1: Some sine functions.

How to compute the Fourier coefficients? Use the orthogonality of the trig set!
Definition: Given two functions u(x), v(x), define the inner product as
Z b
(u, v) =
˙ u(x)v(x) dx.
a

In our case we will choose a = −L, b = L, i.e.,

Z L
(u, v) =
˙ u(x)v(x) dx.
−L

Definition: The functions u and v are orthogonal if (u, v) = 0.

Claim: The trig set is mutually orthogonal, i.e, any two distinct functions in the set are
orthogonal to each other. This means
mπx mπx
(1, sin ) = 0, (1, cos ) = 0, ∀m
L L
mπx nπx mπx nπx
(sin , sin ) = 0, (cos , cos ) = 0, ∀m 6= n
L L L L
mπx nπx
(cos , sin ) = 0, ∀m, n
L L
Proof. By direct integration. For example:
Z L
mπx mπx
(1, sin )= sin dx = 0.
L −L L
This integration is 0 because it integrates a sine function over several complete periods.
For m 6= n, we compute
Z L
1 L
Z  
mπx nπx mπx nπx (m − n)πx (m + n)πx
(sin , sin )= sin sin dx = cos − cos dx = 0.
L L −L L L 2 −L L L

153
Both integrals are 0 because it integrates over several periods of cosine function, when m 6= n.
All other identities are proven in a similar way. We skip the details.
Useful identities.
mπx mπx mπx mπx
(sin , sin ) = (cos , cos ) = L.
L L L L
Proof: Direct computation gives:
L  
mπx mπx 1 2mπx 1
Z
(sin , sin )= 1 − cos dx = (2L) = L.
L L 2 −L L 2

Similarly
mπx mπx
(cos
, cos ) = L.
L L
One may also observe that, by the periodic property, we have
mπx mπx mπx mπx
(sin , sin ) = (cos , cos ).
L L L L
Then, use the trig identity sin2 x + cos2 x = 1, we have
Z Lh
mπx mπx mπx mπx mπx mπx i
(sin , sin ) + (cos , cos )= sin2 + cos2 dx = 2L
L L L L −L L L
which gives the same answer.

Back to Fourier series. We now multiply equation (*) by cos nπx

L and integrate over [−L, L],

L L ∞ Z L
nπx a0 nπx mπx nπx
Z Z X
f (x) cos dx = cos dx + am cos cos dx
−L L −L 2 L m=1 −L L L
∞ Z L
X mπx nπx
+ bm sin cos dx
−L L L
m=1

All the terms are 0 except one:

Z L Z L
nπx nπx nπx
f (x) cos dx = an cos cos dx = an L
−L L −L L L
This gives us the formula to compute an :

1 L nπx
Z
an = f (x) cos dx, n = 1, 2, 3, · · ·
L −L L

Deriving in a completely similar way, we get

1 L nπx
Z
bn = f (x) sin dx, n = 1, 2, 3, · · ·
L −L L
and
L
1
Z
a0 = f (x) dx.
L −L

154
Note that a0 /2 (the constant term in Fourier series) is the average of f (x) over a period. The
formula for a0 fit the one for an with n = 0.
These formulas for computing the Fourier coefficients are called Euler-Fourier formula.
If the period is 2π, i.e., L = π in the formulas, we get simpler looking formulas

1 π
Z
an = f (x) cos nx dx, n = 0, 1, 2, · · ·
π −π
1 π
Z
bn = f (x) sin nx dx, n = 1, 2, 3, · · · .
π −π

We now take some examples in computing Fourier series.

Example 1. Find the Fourier series for a periodic function f (x) with period 2π

−1, if − π < x < 0
f (x) = , f (x + 2π) = f (x)
1, if 0 < x < π

Answer. We use the Euler-Fourier formulas with L = π:

1 π
Z
a0 = f (x) dx = 0.
π −π

We note that f (x) is an odd function, i.e., f (−x) = −f (x). Therefore integrating over a
period, one get 0.
For n ≥ 1, we have
1 π
Z
an = f (x) cos nx dx
π −π
1 0 1 π
Z Z
= − cos nx dx + cos nx dx
π −π π 0
1 1 11
= (− ) sin nx|0x=−π + sin nx|πx=0 = 0
π n πn
Actually, we could get this integral quickly by observing the following: f (x) is an odd function,
and cos nx is an even function. Then, the product f (x) cos nx is an odd function. Therefore,
the integral over an entire period is 0.
Finally, we compute bn as

1 π 1 0 1 π
Z Z Z
bn = f (x) sin nx dx = − sin nx dx + sin nx dx
π −π π −π π 0
11 11
= − (− cos nx)|0x=−π + (− cos nx)|πx=0
πn πn
1 1
= (cos 0 − cos(−nπ)) − (cos nπ − cos 0)
nπ nπ
1 1 2
= (1 − cos nπ) + (1 − cos nπ) = (1 − cos nπ).
nπ nπ nπ

155
The actual computation could be shortened by observing the following: sin nx is an odd
function, so f (x) sin nx is an even function. The integrals on [−π, 0] and [0, π] are the same.
So one needs to do only one integral, and multiply the result by 2.
We observe

cos π = −1, cos 2π = 1, cos 3π = −1, · · · , ⇒ cos nπ = (−1)n

Then
4

2 nπ , n odd,
bn = (1 − (−1)n ) =
nπ 0, n even.
We can now write out the Fourier series. Since all an ’s are 0, we will only have sine
functions. Also, bn is non-zero only for odd n, and we can write out the first few
4 4 4 4
b1 = , b3 = , b5 = , , b7 = ,···
1π 3π 5π 7π
Note that 4/π is a common factor, which we can take out. This gives:
∞  
X 4 1 1 1
f (x) = bn sin nπ = sin x + sin 3x + sin 5x + sin 7x + · · · .
n=1
π 3 5 7

Partial sum of a series: the sum of the first few terms.

We can write yn (x) to be the sum of the first n term in the Fourier series. For our example,
we have
4
y1 (x) = sin x
π  
4 1
y2 (x) = sin x + sin 3x
π 3
 
4 1 1
y3 (x) = sin x + sin 3x + sin 5x
π 3 5
···

Then, the limit limn→+∞ yn (x) (if it converges) gives the whole Fourier series.
The partial sums of Fourier series and the original function f (x) for this example are
plotted together in Fig 10.2.

Example 2. Find the Fourier series of the function


 0, −2 < x < −1
f (x) = K, −1 < x < 1 , period = 4.
0, 1<x<2


Answer. Since the period is 4, we have 2L = 4 so L = 2. We compute the Fourier

coefficients by Euler-Fourier formulas. We have

1 2 1
Z
a0 = f (x) dx = 2K = K,
2 −2 2

156
 Figure 10.2: Fourier series, the first few terms, Example 1.

and
2
nπx 1

1 nπx K nπx K 2 2K nπ
Z Z
1
an = f (x) cos dx = −1 cos dx = sin = sin .
2 −2 2 2 2 2 nπ 2 x=−1 nπ 2
The function sin nπ
2 takes only values 0, 1, 0, −1 in a periodic ways, depending on n. We have


 0, n even

2K
an = nπ , n = 1, 5, 9, 13, 17, · · ·

 2K

− nπ , n = 3, 7, 11, 15, 19, · · ·
For the bn , note that f (x) is an even function, and sin nπx
2 is an odd function, so the product
is an odd function. Integrating over a whole period gives 0, i.e,
1 2 nπx
Z
bn = f (x) sin dx = 0.
2 −2 2
Note that in this example, there will be no sine functions in the Fourier series!
We can now write out the Fourier series:
∞  
1 X nπx K 2K πx 1 3πx 1 5πx 1 7πx
f (x) = a0 + an cos = + cos − cos + cos − cos + ···
2 n=1
2 2 π 2 3 2 5 2 7 2

157
 The partial sums of Fourier series and the original function f (x) are plotted together in
Fig 10.3, for K = 1.

Figure 10.3: Fourier series, the first few terms, Example 2.

Observation.
If f (x) is an odd function, then there are no cosine functions in the Fourier series.
If f (x) is an even function, then there are no sine functions in the Fourier series.
We will see later that this is a general rule!

On tabular method for integration by parts. We see that if f (x) is some kind of
polynomial, we will end up with integrations of the form
nπx nπx
Z Z
f (x) sin dx, f (x) sin dx.
L L
In this case, we need to perform integration by parts multiple times, which could be quite
time consuming. The tabular method makes this computation a bit easier.
In general, this method works best when one of the two functions in the product is a
polynomial, and the other is a function that you can easily find antiderivatives (such as
ex , sin ax, cos ax). Then after differentiating the polynomial several times one obtains zero.
The method may also be extended to work for functions that will repeat themselves, such as
eax sin bx and eax cos bx.

158
 To fix the idea, let’s take an example. Consider the integration
Z
x3 cos nx dx.

Let u = x3 and v = cos nx. Begin with these functions and we set up a table. In column A
we put derivatives of u until it becomes zero. In column B we put in integrals of v. We get
the following result
Derivatives of u (column A) Integrals of v (column B)
x3 cos nx

1
3x2 sin nx
n
1
6x − cos nx
n2
1
6 − sin nx
n3
1
0 cos nx
n4

Now take the product of the 1st entry of column A with the 2nd entry of column B, the
2nd entry of column A with the 3rd entry of column B, in this diagonal way, until you hit the
zero term. These products will be multiplied with alternative signs, starting with the positive
sign, and then added up. We get
1 1 1 1
Z
x3 cos nx dx = x3 sin nx − 3x2 (− 2 cos nx) + 6x(− 3 sin nx) − 6 4 cos nx
n n n n
x 3 3x 2 6x 6
= sin nx + 2 cos nx − 3 sin nx − 4 cos nx.
n n n n

Example 3. Find the Fourier series of


t + π/2, −π < t < 0
r(t) = , period = 2π.
−t + π/2, 0<t<π

Answer. We seek Fourier series for r(t), i.e.,

∞
1 X
r(t) = a0 + an cos nt + bn sin nt.
2
n=1

We first note that r(t) is an even function, we can immediately conclude that bn = 0 for all n.
Furthermore, r(t) cos nt will be an even function. To integrate over a period, we only need
to integrate over half period, and multiply the answer by 2. We now compute an .
1 π
Z
a0 = r(t) dt = 0, (Look at the graph of r over a period!)
π −π

159
and π π
1 2 π
Z Z
an = r(t) cos nt dt = (−t + ) cos nt dt
π −π π 0 2
Let’s try the tabular method for integration by parts:
−t + π/2 cos nt
−1 (1/n) sin nt
0 −(1/n2 ) cos nt
We have now
 π
2 π 1 1
an = (−t + ) sin nt − 2 cos nt
π 2 n n t=0

2

1 1

2  0, n even,
n
= − 2 cos nπ + 2 = (1 − (−1) ) =
π n n πn2  4 , n odd.
n2 π
We can now write out the Fourier series
∞  
X X 4 4 1 1
r(t) = an cos nt = cos nt = cos t + cos 3t + cos 5t + · · · .
n=1
n2 π π 9 25
n odd
The plots of several partial sums and their error are included in Figure 10.4.
We make some observations:
(1). In general, the error decreases as we take more terms.
(2). For a fixed partial sum, the error is larger at the point where the function r(t) has a kink,
and smaller in the region where r(t) is smooth.
(3). After taking 3 terms, the partial sum is already a very good approximation to r(t).
(4). It seems like yn (t) converges to r(t) at every point t.

Next example is a dummy one, but might be useful.

Example 4. Find the Fourier coefficients for f (x), periodic with p = 2π, given as

f (x) = 2 + 4 sin x − 0.5 cos 4x − 99 sin 100x.

Answer. Since the function f here is already given in terms of sine and cosine function,
there is no need to compute the Fourier coefficients. We just need to figure out where each
term would fit, by comparing it with a Fourier series. We have

a0 = 4, a4 = −0.5, an = 0, ∀n 6= 0, 4

and
b1 = 4, b100 = −99, bn = 0 ∀n 6= 1, 100.

160
 Figure 10.4: Fourier series, the first few terms and the errors, Example 3.

10.2 Even and Odd Functions; Fourier sine and Fourier cosine
series.
Through examples we have already observed that, for even and odd functions, the Fourier
series takes simpler forms. We will summarize it here.
A function f (x) is even if
f (−x) = f (x).
The graph of the function is symmetric about the y-axis. Examples include f (x) = 1 and
f (x) = cos nx for any integer n.
A function f (x) is odd if
f (−x) = −f (x).
The graph of the function is symmetric about the origin. Examples include f (x) = sin nx for
any integer n.

161
 Properties:

• Product of two even functions is even;

• Product of two odd functions is even;

• Product of an even and an odd function is odd.

• Integration of an odd function over [−L, L] is 0.

• Integration of an even function over [−L, L] is twice the integration over [0, L].

We have already observed that, if f (x) is an even function, then its Fourier series will NOT
have sine functions. If f (x) is an odd function, then its Fourier series will NOT have cosine
functions.
This fits the instinct: One can not represent an odd function with the sum of some even
functions, and visa versa.
The formulas for the Fourier coefficients could be simplified, as we have already observed.

• If f (x) is an even, periodic function with p = 2L, it has a Fourier cosine series
∞
a0 X nπx
f (x) = + an cos
2 L
n=1

where
L
2 nπx
Z
an = f (x) cos dx, n = 0, 1, 2, · · ·
L 0 L

• Correspondingly, if f (x) is an odd, periodic function with p = 2L, it has a Fourier sine
series
∞
X nπx
f (x) = bn sin
L
n=1

where
L
2 nπx
Z
bn = f (x) sin dx, n = 1, 2, 3, · · ·
L 0 L

Note now we only need to integrate over half period, i.e., over [0, L], because the product
is an even function.

Half-range expansion. If a function f (x) is only defined on an interval [0, L], we can
extend/expand the domain into the whole real line by periodic expansion. There are two ways
of doing this:

• Extend f (x) onto the interval [−L, L] such that f is an even function, i.e., f (−x) = f (x),
then extend it into a periodic function with p = 2L;

• Extend f (x) onto the interval [−L, L] such that f is an odd function, i.e., f (−x) =
−f (x), then extend it into a periodic function with p = 2L.

162
 even extension
1

0.8

0.6

0.4

0.2

0
−3 −2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5 3

odd extension
1

0.5

−0.5

−1
−3 −2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5 3

Figure 10.5: Even and odd extension, 3 periods plotted.

These are called even/odd periodic extensions of f , or half-range expansions.

Example 1. Let f (x) = x be defined on the interval x ∈ [0, L]. Sketch 3 periods of the
even and odd extension of f , and then compute the corresponding Fourier sine or cosine series.

Answer. The graph of even and odd extensions are given in Fig 10.5.
The odd periodic extension turns out to be the “sawtooth wave”. We have Fourier sine
series, with the following coefficients

2 L 2 L 2n
 
nπx nπx nπx nπx oL
Z
bn = x sin dx = sin − cos
L 0 L L nπ L L L

x=0
2L
= (−1)n+1 , n = 1, 2, 3, · · ·
nπ
therefore
∞
2L X (−1)n+1
 
nπx 2L πx 1 2πx 1 3πx 1 4πx
fodd (x) = sin = sin − sin + sin − sin + ··· .
π n L π L 2 L 3 L 4 L
n=1

The even extension gives triangle waves (similar to Example 3 in ch 1.1). It will have a

163
Fourier cosine series, with coefficients
L L
2 2 L2
2
Z Z
a0 = f (x) dx = ( ) = L, x dx =
L 0 0 L
L 2
(  )L
2 L nπx 2 L 2 nπx xL nπx 
Z
an = x cos dx = cos + sin
L 0 L L nπ L nπ L 

x=0
2L 2L
= (cos nπ − 1) = 2 2 ((−1)n − 1), n = 1, 2, 3, · · ·
n2 π 2 n π
Therefore an = 0 for n even, and an = − n4L
2 π 2 for odd n. We have the Fourier cosine series

 
L X 4L nπx L 4L πx 1 3πx 1 5πx 1 7πx
feven (x) = − cos = − 2 cos + cos + 2 cos + 2 cos + ··· .
2 n2 π 2 L 2 π L 9 L 5 L 7 L
n odd

Include the plots of partial sums, and the errors, for even and odd expansions are included
in Fig 10.6 and Fig 10.7.

10.3 Properties of Fourier Series

Linearity. Let f (x) and g(x) be 2 periodic functions withe same period, and each has a
Fourier series with coefficients (an , bn ) for f (x) and (ān , b̄n ) for g(x). Then, the followings
hold.
(1). The function f (x)+g(x) will have Fourier coefficients (an + ān , bn + b̄n ). (2). The function
αf (x) for some constant α will have Fourier coefficients (αan , αbn ).
Convergence Theorem. Let f (x) be a periodic function and let it have a Fourier series
F (x). Assume f (x) is piecewise continuous. Then,
(1). Fourier series converges to f (x) at all points x where f is continuous;
(2). At a point x where f is discontinuous, Fourier series converges to the mid value of the
left and right limit, i.e., 12 [f (x−) + f (x+)].

This is confirmed by our example, see Example 1 and 2 in previous section, and Figure
10.2 and Figure 10.3.

Example 1. Find the Fourier series for:


0, −π < x ≤ 0
h(x) = , h(x + 2π) = h(x).
K, 0<x≤π

Indicate the function that the Fourier series of h(x) converges to.

Answer. Instead of working out the Fourier coefficients by direct computation, we will
use the linear property. Let now
g(x) = K/2
The Fourier coefficients for g(x) are simply

ā0 = K/2, ān = b̄n = 0, ∀n ≥ 1.

164
 odd extention even extention
1 1

0.8
0.5
0.6
0
0.4
−0.5
0.2

−1 0
−3 −2 −1 0 1 2 3 −3 −2 −1 0 1 2 3

1 1

0.8
0.5
0.6
0
0.4
−0.5
0.2

−1 0
−3 −2 −1 0 1 2 3 −3 −2 −1 0 1 2 3

1 1

0.8
0.5
0.6
0
0.4
−0.5
0.2

−1 0
−3 −2 −1 0 1 2 3 −3 −2 −1 0 1 2 3

1 1

0.8
0.5
0.6
0
0.4
−0.5
0.2

−1 0
−3 −2 −1 0 1 2 3 −3 −2 −1 0 1 2 3

Figure 10.6: Even and odd extensions, first 4 partial sums, over 3 periods.

Then,
 
−K/2, −π < x ≤ 0 K −1, −π < x ≤ 0 K
h(x) − g(x) = = = f (x),
K/2, 0<x≤π 2 1, 0<x≤π 2
where f (x) is the same function as in Example 1 in previous chapter, for which we have already
computed the Fourier coefficients, i.e,
2
a0 = 0, an = 0, bn = (1 − (−1)n ).
nπ
By linearity, for h(x) = (K/2)f (x) + g(x), will have Fourier coefficients
K
ãn = (K/2)an + ān = 0, b̃n = (K/2)bn + b̄n = (1 − (−1)n ),
nπ
which gives
 
K 2K 1 1 1
h(x) = + sin x + sin 3x + sin 5x + sin 7x + · · · .
2 π 3 5 7

165
 odd extention, error even extention, error
1
0.5
0.5

0 0

−0.5
−0.5
−1
−3 −2 −1 0 1 2 3 −3 −2 −1 0 1 2 3

1 0.1

0.5 0.05

0 0

−0.5 −0.05

−1 −0.1
−3 −2 −1 0 1 2 3 −3 −2 −1 0 1 2 3

1
0.04
0.5
0.02

0 0

−0.02
−0.5
−0.04
−1
−3 −2 −1 0 1 2 3 −3 −2 −1 0 1 2 3

1 0.04

0.5 0.02

0 0

−0.5 −0.02

−1 −0.04
−3 −2 −1 0 1 2 3 −3 −2 −1 0 1 2 3

Figure 10.7: Even and odd extensions, errors for the first 4 partial sums, over 3 periods.

166
The Fourier series of h(x) converges to h(x) where-ever the function is continuous, and to the
mid value at discontinuities, i.e.,


 K/2, x = −π,
0, −π < x < 0,

h̄(x) = h̄(x + 2π) = h̄(x).

 K/2, x = 0,
K, 0 < x < π,


One can sketch a graph to see it more clearly.

Choice of the half range expansion with concerns on convergence. We note

that the Fourier cosine series, i.e, the even expansion seems to have smaller error for the
same number of terms in the partial sum. This is because the even extension is a continuous
function, while the odd extension is a piecewise continuous function with discontinuity points
at x = ±1, ±3, ±5, · · · . All sine and cosine functions are smooth. Using smooth functions to
represent discontinuous function would give larger error.
From the convergence Theorem, we know that, at a discontinuous point, the Fourier series
converges to the mid value of the left and right limits. This implies an error that is equal to
half of the size of the jump at this point. This error will not become smaller by taking more
terms in the partial sum.
In practice, when one has the choice, it would always be recommended to choose the ex-
pansion that does NOT has discontinuities, if possible. So even expansions should be preferred
for accuracy.

We give another example, on convergence of Fourier series, in connection with even and
odd periodic extensions.

Example 2. Let f (x) = 2x2 − 1 be defined on the interval x ∈ [0, 1]. Sketch 3 periods of
it even and odd periodic extension. Where do their Fourier cosine and sin series converge to
at the points x = 0, 0.5, 1, 100, 151.5?

Answer. Even and odd extensions are plotted below:

167
 even extension
1

0.5

−0.5

−1
−3 −2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5 3

odd extension
1

0.5

−0.5

−1
−3 −2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5 3

We see that the even extension is a continuous function, so Fourier cosine series converges
to the function value, and the convergence is faster.
However, the odd extension is discontinuous at x = 0, ±1, ±2, · · · , and the Fourier sine
series will converge to the mid value of the left and right limits.
We put these value in a table.
x 0 0.5 1 100 151.5
even -1 -0.5 1 -1 -0.5
odd 0 -0.5 0 0 0.5

10.4 Two-Point Boundary Value Problems; Eigenvalue Prob-

lems
Let y(x) be the unknown, and x is the space variable. We consider 2nd order linear ODE

y ′′ + p(x)y ′ + q(x)y = g(x)

over the interval x ∈ [x1 , x2 ], with the boundary conditions

y(x1 ) = y1 , y(x2 ) = y2 .

Since now the conditions are given at the two boundary points, this is called a two-point
boundary value problem.
If y1 = y2 = 0, we called this homogeneous boundary conditions.

168
 NB! Boundary conditions could of other forms, such as y ′ (x1 ) = 0 etc.
Solution strategy: Find the general solution (as in Chapter 3), then, use the boundary
conditions to determine the constants c1 , c2 .

Example 1. Solve the boundary value problem

y ′′ + y = 0, y(0) = 1, y(π/2) = 0.

Answer. By characteristic equation r 2 + 1 = 0, r1,2 = ±i, we get the general solution

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

We now put in the boundary conditions, and get

c1 = 1, c2 = 0

so the solution is y(x) = cos x.

Example 2. Boundary conditions could change the solution. In the previous example, we
now assume the boundary conditions

y(0) = 0, y(π) = 2.

From y(0) = 0 we get c1 = 0. From y(π) = 2, we get c1 = 2, which is contradictory to the

first condition. Therefore, there is no solution.
Now we assume different boundary conditions

y(0) = 0, y(π) = 0.

Then, c1 = 0, and c2 can be arbitrary, so y(x) = c2 sin x is a solution for any c2 .

Example 3. Solve

y ′′ + 4y = cos x, y ′ (0) = 0, y ′ (π) = 0.

Answer. Since r 2 + 4 = 0, and r1,2 = ±2i, the general solution for the homogeneous equation

is
yH (x) = c1 cos 2x + c2 sin 2x.
We now find a particular solution for the non-homogeneous equation. We guess the form
Y = A cos x. Then Y ′′ = −A cos x, so
1
−A cos x + 4A cos x = cos x, ⇒ 3A = 1, A= .
3
This gives the general solution
1
y(x) = c1 cos 2x + c2 sin 2x + cos x.
3

169
To check the boundary condition, we first differentiate y
1
y ′ (x) = −2c1 sin 2x + 2c2 cos 2x − sin x.
3
Then,
y ′ (0) = 0 ⇒ 2c2 = 0, ⇒ c2 = 0
and
y ′ (π) = 0, ⇒ 2c2 = 0, ⇒ c2 = 0.
Then, c1 remain arbitrary. We conclude
1
y(x) = c cos 2x + cos x
3
is the solution, for arbitrary c.

Important observation: For two-point boundary value problems, the existence and
uniqueness theorem is automatic! This is very different from the initial value problems we
studied in Chapter 3. For linear two-point boundary value problems, there are 3 possibilities:
(i) There is one uniqueness solution, (ii) There are no solutions at all; or (iii) There are
infinitely many solutions.
Comparing this to the solution of two linear equations with two unknowns, do we see the
similarity?

Eigenvalue problems. (compare to eigenvalues of a matrix.) Consider the problem:

y ′′ + λy = 0, y(0) = 0, y(L) = 0. (∗)

We are interested in non-trivial solutions (i.e., y ≡ 0 is not considered because it is trivial),

and possible values of λ that would give us non-trivial solutions.
Note that it is important to have homogeneous boundary conditions for eigenvalue
problems!
This type of problems is an important building block in series solutions of partial differential
equations. For example, this is part of the solution for a vibrating string, where u(x, t) is the
vertical displacement of the string at position x, and the string is horizontally placed. There,
it turns out (as we will study later), that the solution takes the form u(x, t) = y(x)G(t), where
y(x) satisfies the eigenvalue problem.
The eigenvalue problem is a two-point boundary value problem. Depending on the bound-
ary condition, it might or might not have non-trivial solutions.
If for certain value λn we find a nontrivial solution yn (x), then, λn is called an eigenvalue,
and yn is the corresponding eigenfunction.

Example 1. We now attempt to solve the problem in (*). The general solution depends
on the roots, i.e., on the sign of λ. We have 3 situations:
(1). If λ < 0, we write λ = −k2 , where k > 0. Then, r 2 = k2 , so r1 = −k, r2 = k, and the
general solution is
y(x) = c1 ekt + c2 e−kt

170
By boundary conditions, we must have

c1 + c2 = 0, c1 ekL + c2 e−kL = 0

which gives the solution c1 = c2 = 0. Then y(x) = 0, which is a trivial solution. We discard
it.
(2). If λ = 0, then y ′′ = 0, and y(x) must be a linear function. With the zero boundary
conditions, we conclude y(x) = 0, therefore√trivial.
(3). If λ > 0, we write λ = k2 , for k = λ > 0. Then, r 2 = −k2 , and r1,2 = ±ik, and the
general solution is
y(x) = c1 cos kx + c2 sin kx
We now check the boundary conditions. By y(0) = 0, we have

y(0) = c1 = 0

which means y = c2 sin kx. Then, by y(L) = 0, we get

y(π) = c2 sin kL.

We can either require c2 = 0 or sin kL = 0. If we require c2 = 0, then y(x) = 0 which is trivial.

So we must require c2 6= 0 and sin kL = 0. This gives a constraint on the values of k (i.e., λ).
Indeed, we must have
nπ
kL = nπ, ⇒ k= , n = 1, 2, 3, · · · ,
L
We see that we have found a family (infinite size) of eigenvalues and eigenfunctions! Using n
as the index, they are
 nπ 2 nπx
λn = , yn (x) = sin , n = 1, 2, 3, · · ·
L L
Note that we let the arbitrary constant c2 = 1, since for any arbitrary c2 will yield an eigen-
function.

Note that these eigenfunctions are precisely part of the trig set used for Fourier series.
We can observe these eigenfunction by playing with the slinky.

Example 2. Consider the problem with different boundary conditions

y ′′ + λy = 0, y ′ (0) = 0, y ′ (L) = 0. (∗)

We will find very different eigenvalues and eigenfunctions. We still consider the same 3 cases.
(1). If λ = −k2 < 0, then

y(x) = c1 ekt + c2 e−kt , y ′ (x) = kc1 ekt − kc2 e−kt .

The boundary conditions give

kc1 − kc2 = 0, kc1 ekL − kc2 e−kL = 0, ⇒ c1 = c2 = 0

171
which gives only the trivial solution.
(2). If λ = 0, then y ′′ = 0, so y(x) = Ax + B, and y ′ (x) = A. By boundary condition, we
must have A = 0, but B remains arbitrary. So we found an eigenpair:

λ0 = 0, y0 (x) = 1.

(3). If λ = k2 > 0, then

y(x) = c1 cos kx + c2 sin kx, y ′ (x) = −kc1 sin kx + kc2 cos kx.

We now check the boundary conditions:

y ′ (0) = 0, ⇒ kc2 = 0, ⇒ c2 = 0

and
y ′ (L) = 0, ⇒ −kc1 sin kL = 0
If c1 = 0, we get trivial solution. So c1 6= 0. Then, we must have
nπ
sin kL = 0, ⇒ kL = nπ, ⇒ k= , n = 1, 2, 3, · · ·
L
For each k, we get a pair of eigenvalue and eigenfunction
 nπ 2 nπ
λn = , yn (x) = cos x, n = 1, 2, 3, · · ·
L L
One could combine the results in (2) and (3), and get
 nπ 2 nπ
λn = , yn (x) = cos x, n = 0, 1, 2, · · ·
L L
Note that these are also a part of the trig set used in Fourier series!

Observation.

• We notice that, different types of boundary conditions would give very different eigen-
values and eigenfunctions!

• In these two examples, the eigenfunctions are sine and cosine functions, in the same
form as the trig set we use in Fourier series. Recall that the trig set is a mutually
orthogonal set. So, for each of these two eigenvalue problems, the set of eigenfunctions
are mutually orthogonal. In fact, this is a more general property for eigenfunctions. One
can define proper inner product such that eigenfunctions for the same eigenvalue problem
would always form a mutually orthogonal set. (If you are familiar with eigenvectors of
a symmetric matrix, they also form an orthogonal basis!)

Example 3. Find all positive eigenvalues and their corresponding eigenfunctions of the
problem
y ′′ + λy = 0, y(0) = 0, y ′ (L) = 0.

172
 Answer. Since λ > 0, we write λ = k2 where k > 0, and the general solution is

y(x) = c1 cos kx + c2 sin kx, y ′ (x) = −kc1 sin kx + kc2 cos kx.

We now check the boundary conditions. First,

y(0) = 0 ⇒ c1 = 0

so the answer is simplified to

y(x) = c2 sin kx, y ′ (x) = kc2 cos kx.

The 2nd boundary condition gives

y ′ (L) = 0 ⇒ kc2 cos kL = 0, ⇒ cos kL = 0

which implies
1 π 1
kL = (n + )π ⇒ kn = (n + ), n = 1, 2, 3, · · ·
2 L 2
We get the eigenvalues λn and the corresponding eigenfunction yn as
 2
π(n + 1/2) π(n + 1/2)x
λn = kn2 = , yn = sin , n = 1, 2, 3, · · · .
L L

(Optional) Sometimes, with more complicated boundary conditions, eigenvalues could be

obtained through graphs.

Example 4. (optional) Find all positive eigenvalues λn and their corresponding eigen-
functions un (x) of the problem

u′′ + λu = 0, u′ (0) = 0, u(π) + u′ (π) = 0.

Answer. Since λ > 0, we write λ = k2 where k > 0. The general solution is

u(x) = c1 cos kx + c2 sin kx, u′ (x) = −kc1 sin kx + kc2 cos kx.

We now check the boundary conditions. First,

u′ (0) = 0 ⇒ kc2 = 0, ⇒ c2 = 0

so the general solution is simplified

u(x) = c1 cos kx, u′ (x) = −kc1 sin kx.

By the 2nd boundary condition

u(π) + u′ (π) = 0 ⇒ c1 cos kπ − kc1 sin kπ = 0, ⇒ c1 cos kπ = kc1 sin kπ.

173
Since c1 6= 0 (otherwise trivial), we must have
1
cos kπ = k sin kπ, ⇒ = tan kπ.
k
To find the values of k that satisfy this relation, we can plot the two functions
1
f1 (k) = , f2 (k) = tan kπ
k
on the same graph, for k > 0, and we look for intersection points. See plot below:

In this case, one finds infinitely many intersection points. The k-coordinates of all these
points give all the eigenvalues. Marking the p-coordinate of these intersection points as pn for
n = 1, 2, · · · , we get the eigenvalues and the eigenfunctions

λn = p2n , un = cos pn x, n = 1, 2, · · · .

We observe that, as n grows bigger, the interception point gets closer to the k-axis, so λn ≈ n−1
for large n.

Summary: It would be useful to memorize the solutions to the following simple eigenvalue
problems:

• Homogeneous Dirichlet boundary condition:

y ′′ + λy = 0, y(0) = 0, y(L) = 0.

The solutions are

 nπ 2 nπx
λn = , yn (x) = sin , n = 1, 2, 3, · · · .
L L

174
• Homogeneous Neumann boundary condition:

y ′′ + λy = 0, y ′ (0) = 0, y ′ (L) = 0.

The solutions are

 nπ 2 nπx
λn = , yn (x) = cos , n = 0, 1, 2, · · · .
L L

175
Chapter 11

Partial Differential Equations

11.1 Basic Concepts

Definition of PDE: an equation with partial derivatives of the unknown u that depends on
several variables, e.g., u(x, t) or u(x, y) etc.
Order of PDE = the highest order of derivatives
Linear PDE: the terms with u and its derivatives are in a linear form
Non-liear PDE: otherwise
Homogeneous: each term contains u or its derivatives
Non-homogeneous: otherwise
Notations:
∂u ∂u ∂2u
ut = , ux = , uxx = , etc.
∂t ∂x ∂x2
Examples of 2nd order linear PDEs:

• utt = c2 uxx , 1D wave equation

• ut = c2 uxx , 1D heat equation

• uxx + uyy = 0, 2D Laplace equation

• uxx + uyy = f (x, y), 2D Poisson equation (non-homogeneous)

• ut = uxx + uyy , 2D heat equation

Concept of solution: u is a solution if it satisfies the equation, and any boundary or initial
conditions if given.
The students should be able to verify if a given function is a solution of a certain equation.
Fundamental principle of superposition for linear PDEs:

• Homogeneous:
L(u) = 0 (11.1)

Principle of superposition: If u1 and u2 are two solutions, i.e., L(u1 ) = 0 and

L(u2 ) = 0, so is u = c1 u1 + c2 u2 with arbitrary constants c1 , c2 .

176
 • Non-homogeneous:
L(u) = f, (11.2)

– If uH solves (11.1) and up solves (11.2), then u = uH + up solves (11.2).

– If u1 solves L(u1 ) = f1 and L(u2 ) = f2 , then u = c1 u1 + c2 u2 solves L(u) =
c1 f1 + c2 f2 .

How to separate variables?

Let u(x, t) be the solution of some PDE, with suitable boundary and initial conditions.
We seek solutions of the form
u(x, t) = F (x)G(t)
where F (x) is only a function of x, and G(t) is only a function of t. Then, the partial derivatives
are

ux = F ′ (x)G(t), uxx = F ′′ (x)G(t), ut = F (x)G′ (t), utt = F (x)G′′ (t), uxt = F ′ (x)G′ (t).

We will then put these into the equation, and try to separate x and t on different sides of the
equation.

11.2 Heat Equation in 1D; Solution by Separation of Variable

and Fourier series
Consider the heat equation in 1D

ut = c2 uxx , 0 < x < L, t > 0. (11.3)

Here u(x, t) measures the temperature of a rod with length L. We first assign the boundary
conditions
u(0, t) = 0, u(L, t) = 0, t>0
This means, we fix the temperature of the two end-points of the rod to be 0. This type of
boundary condition is called Dirichlet condition.
We also have the initial condition

u(x, 0) = f (x), 0<x<L

gives the initial temperature distribution.

The main technique to find an explicit form of the solution is Separation of variables.

Step 1. Separating variables. Seek solution of the form

u(x, t) = F (x) · G(t)

Then
ux = F ′ (x)G(t), uxx = F ′′ (x)G(t), ut = F (x)G′ (t),

177
Plug these into the equation (11.3),

F ′′ (x) G′ (t)
F (x)G′ (t) = c2 F ′′ (x)G(t), → = 2 = constant = −k
F (x) c G(t)
We end up with 2 ODEs,

F ′′ (x) + kF (x) = 0, G′ (t) + c2 kG(t) = 0.

Step 2. Solve for F (x). Fit in the boundary conditions

u(0, t) = F (0)G(t) = 0, u(L, t) = F (L)G(t) = 0, t>0

which implies
F (0) = 0, F (L) = 0
We now have the following eigenvalue problem for F (x)

F ′′ + pF = 0, F (0) = 0, F (L) = 0.

This is an example we had earlier. We have

nπ
pn = wn2 , wn = , Fn (x) = sin wn x, n = 1, 2, 3, · · ·
L
Step 3. Solution for G(t). For any given n, we get a solution Gn (t), which solves

G′ (t) + c2 wn2 G(t) = 0

nπc
Call now λn = cwn = L , then
2
G′ (t) + λ2n G(t) = 0, Gn (t) = Cn e−λn t

where Cn is arbitrary.
Step 4. We now get the eigenvalues and their eigenfunctions
nπc 2
λn = , un (x, t) = Cn e−λn t sin wn x, n = 1, 2, 3, · · ·
L
The sum of them is also a solution. This gives the formal solution
∞ ∞
X X 2 nπ nπc
u(x, t) = un (x, t) = Cn e−λn t sin wn x, wn = , λn = . (11.4)
L L
n=1 n=1

Step 5. Find Cn by initial condition. Fit in the initial condition

∞
X
u(x, 0) = Cn sin wn x = f (x)
n=1

we conclude that Cn must the Fourier sine coefficients for the odd periodic half-range extension
of f (x), i.e.,
2 L nπx
Z
Cn = f (x) sin dx, n = 1, 2, 3, · · · . (11.5)
L 0 L

178
 Summary: The formal solution for initial and boundary value problem for the heat
equation

ut = c2 uxx , 0 < x < L, t > 0.

u(0, t) = 0, u(L, t) = 0, t>0
u(x, 0) = f (x), 0<x<L

is
∞ ∞
X X nπc 2 nπx
u(x, t) = un (x, t) = Cn e−( L
) t
sin
L
n=1 n=1
where
L
2 nπx
Z
Cn = f (x) sin dx, n = 1, 2, 3, · · · .
L 0 L

Discussions on solutions:
• Harmonic oscillation in x, exponential decay in t:

• Speed of decay depending on λn = nπc/L. Faster decay for larger n, meaning the high
frequency components are ’killed’ quickly. After a while, what remain in the solution
are the terms with small n.

• As t → ∞, we have u(x, t) = 0 for all x. This is called asymptotic solution or steady

state of the heat equation.

Example 1. Let c = 1 and L = 1. If f (x) = 10 sin πx, then we have C1 = 10 and all other
Cn = 0, so the solution is
2
u(x, t) = 10e−π t sin πx.
At t = 1, the amplitude of the solution is
2
max |u(x, 1)| = 10e−π ≈ 5.17 × 10−4 .
x

If now we let f (x) = 10 sin 3πx, then C3 = 10 and all other Cn = 0, and the solution is
2
u(x, t) = 10e−9π t sin 3πx.

At t = 1, the amplitude is
2
max |u(x, 1)| = 10e−9π ≈ 2.65 × 10−38 .
x

Note that this amplitude is much smaller.

If the initial temperature is

f (x) = 10 sin πx + 10 sin 3πx,

the solution would be

2 2
u(x, t) = 10e−π t sin πx + 10e−9π t sin 3πx

179
and the amplitude at t = 1 is

max |u(x, 1)| = 5.17 × 10−4 + 2.65 × 10−38 .

x

Clearly, the first term dominates.

Neumann boundary condition. We now consider a new BC: (insulated)

ux (0, t) = 0, ux (L, t) = 0

This means that the 2 ends are insulated, and no heat flows through.
Following the same setting, we get the eigenvalue problem for F (x) as

F ′′ (x) + pF (x) = 0, F ′ (0) = 0, F ′ (L) = 0

From Example 2 in the last Chapter, we have only non-negative p. Let p = w2 with w ≥ 0.
We have the eigenvalues wn and the eigenfunctions Fn (x):
nπ
wn = , Fn (x) = cos wn x, n = 0, 1, 2, · · ·
L
Note that n = 0 is permitted in this solution. The solution for G(t) remains the same
2 cnπ
Gn (t) = Cn e−λn t , λn = , n = 0, 1, 2, · · ·
L
This gives the eigenfunctions
2
un (x, t) = Cn e−λn t cos wn x, n = 0, 1, 2, · · ·

which leads to the formal solution

∞
X 2 nπ cnπ
u(x, t) = C0 + Cn e−λn t cos wn x, wn = , λn = . (11.6)
L L
n=1

Finally, we fit in the initial condition

∞
X
u(x, 0) = C0 + Cn cos wn x = f (x).
n=1

So, Cn must be the Fourier cosine coefficient for the even half-range expansion of f (x), i.e.,
L L
1 2 nπx
Z Z
C0 = f (x) dx, Cn = f (x) cos dx, n = 1, 2, 3, · · · . (11.7)
L 0 L 0 L

Conclusion: The formal solution is given in (11.6) + (11.7).

Discussion on the formal solution:

• harmonic oscillation in x,

• exponential decay in t, except the term C0 . Decay faster for larger n.

180
 • As t → ∞, we get u → C0 , which is the average of f (x) (initial temperature). This is
reasonable b/c the bar is insulated.

Steady State: As t → ∞, solution does not change in time anymore, as it reaches a

steady state. Call it U (x). Then Ut = 0, so Uxx = 0. So U (x) must be a linear function, i.e.,
U (x) = Ax + B, where A, B are determined by boundary conditions.

Example 2: (add graphs)

(1) If u(0, t) = a, u(L, t) = b, then U (0) = a, U (L) = b, we get

b−a
U (x) = a + x.
L

(2) If u(0, t) = a, ux (L, t) = 0, then U (0) = a, U ′ (L) = 0, we get

U (x) = a.

(3) If u(0, t) = a, ux (L, t) = b, then U (0) = a, U ′ (L) = b, we get

U (x) = a + bx.

Non-homogeneous boundary condition. Take for example

ut = c2 uxx , u(0, t) = a, u(L, t) = b.

b−a
We know that the steady state is U (x) = a + L x. Now define a new variable

w(x, t) = u(x, t) − U (x).

Then, we have

wt = ut , wx = ux − U ′ (x), wxx = uxx − U ′′ (x) = uxx

so w solve the heat equation:

wt = c2 wxx
Now, we check the BCs for w:

w(0, t) = u(0, t) − U (0) = a − a = 0, w(L, t) = u(L, t) − U (L) = b − b = 0

which are homogeneous. Then, one can find the solution for w by the standard separation of
variables and Fourier series. Once this is done, one can go back to u by

u(x, t) = w(x, t) + U (x).

We now take an example with non-homogeneous BCs.

181
 Example 3. Consider the heat equation ut = uxx with the following BCs

u(0, t) = 2, u(1, t) = 4,

and IC
u(x, 0) = 2 + 2x − sin πx − 3 sin 3πx.
Find the solution u(x, t).

Answer. : Step 1. We first find the steady state. Call it w(x), it satisfies the following
two-point boundary value problem

w′′ = 0, w(0) = 2, w(1) = 4,

which gives the solution

w(x) = 2 + 2x.
Step 2: Let now U be the solution of the heat equation with homogeneous boundary
condition
Ut = Uxx , U (0, t) = 0, U (1, t) = 0
and the initial condition

U (x, 0) = u(x, 0) − w(x) = − sin πx − 3 sin 3πx.

The formal solution for U is

+∞
2
X
U (x, t) = Cn e−λn t sin wn x,
n=1

where we have
nπ ncπ
L = 1, wn = = nπ, λn = = nπ
L L
so
+∞
2 π2 t
X
U (x, t) = Cn e−n sin nπx.
n=1

Here Cn are Fourier coefficients of the initial data U (x, 0). We find only two coefficients that
are not 0, namely
C1 = −1, C3 = −3.
This gives us
2 2
U (x, t) = −e−π t sin πx − 3e−9π t sin 3πx.
Step 3. Putting them together, we get the solution
2 2
u(x, t) = w(x) + U (x, t) = 2 + 2x − e−π t sin πx − 3e−9π t sin 3πx.

More on separation of variables. This technique could be applied to a more general

class of PDEs. It is not difficult to check whether an equation is separable. After separating
the variables, one needs to fit in the boundary condition.

182
 Example 1. Consider
x2 utt − (x + 1)t2 uxx = 0
with boundary conditions
u(0, t) = 0, u(L, t) = 5,
and initial condition
u(x, 0) = cos(x).
Letting u = F (x)G(t), we have

x2 F (x)G′′ (t) − (x + 1)t2 F ′′ (x)G(t) = 0, ⇒ x2 F (x)G′′ (t) = (x + 1)t2 F ′′ (x)G(t),

which is separable
G′′ (t) (x + 1)F ′′ (x)
= = −λ
t2 G(t) x2 F (x)
which gives two ODEs:

(x + 1)F ′′ (x) + λx2 F (x) = 0, F (0) = 0, F (L) = 5

and
G′′ (t) + λt2 G(t) = 0.

Example 2. Consider
uxx − 6utx + 9utt = 0
Letting u = F (x)G(t), we have

F ′′ (x)G(t) − 6F ′ (x)G′ (t) + 9F (x)G′′ (t) = 0

It is not possible to separate the variables.

Example 3. Consider
utt = 4uxx − 5ux + u
Then,
F G′′ = 4F ′′ G − 5F ′ G + F G, ⇒ F G′′ = G(4F ′′ − 5F ′ + F )
which is separable
G′′ 4F ′′ − 5F ′ + F
= = −λ
G F
We get two ODEs:
4F ′′ − 5F ′ + F + λF = 0
and
G′′ + λG = 0.
How to solve these eigenvalue problem, that is a different question.

183
11.3 Solutions of Wave Equation by Fourier Series
Consider the 1D wave equation

utt = c2 uxx , t > 0, 0 < x < L. (11.8)

Here u(x, t) is the unknown variable. If this is a model of vibrating string, then the constant
c2 has physical meaning, namely, c2 = T /ρ where T is the tension and ρ is the density such
that ρ∆x is the mass of the string segment.
For the first example, we assign the following boundary and initial conditions

(B.C.’s) u(0, t) = 0, u(L, t) = 0, t > 0. (11.9)

(I.C.’s) u(x, 0) = f (x), ut (x, 0) = g(x), 0 < x < L. (11.10)

We now solve this equation by separation of variables and Fourier series.

Step 1. Let u(x, t) = F (x) · G(t), then

utt = F (x)G′′ (t), uxx = F ′′ (x)G(t)

and we get
F ′′ (x) G′′ (t)
F (x)G′′ (t) = c2 F ′′ (x)G(t), → = 2 = −p (constant).
F (x) c G(t)
This gives us 2 ODEs

F ′′ (x) + pF (x) = 0, G′′ (t) + c2 pG(t) = 0.

Step 2. We now solve for F (x). The BCs in (11.9) gives

F (0) = 0, F (L) = 0.

We are now familiar with this eigenvalue problem, and the solutions are
nπ
p = wn2 , wn = , Fn (x) = sin wn x, n = 1, 2, 3, · · · .
L
Step 3. Now, for a given n, the ODE for G(t) takes the form,

G′′ (t) + λ2n G(t) = 0, λn = nwn .

The general solution is

Gn (t) = Cn cos λn t + Dn sin λn t, n = 1, 2, 3, · · · .

We let

un (x, t) = Fn (x)Gn (t) = (Cn cos λn t + Dn sin λn t) sin wn x, n = 1, 2, 3, · · ·

Here λn are eigenvalues, and un (x, t) are eigenfunction. The set of eigenvalues {λ1 , λ2 , · · · }
are called the spectrum.
Discussion on eigenfunctions:

184
 • Harmonic oscillation in x.

• G(t) gives change of amplitude in t, harmonic oscillation. Draw a figure (for ex. with
n = 2) and explain.

• Different n gives different motion. These are called modes. Draw figures of modes with
n = 1, 2, 3, 4. With n = 1, we have the fundamental mode. n = 2 gives an octave, n = 3
gives an octave and a fifth, n = 4 gives 2 octaves.

Step 4. We now construct solution of the wave equation (11.8). The formal solution is
∞
X ∞
X
u(x, t) = un (x, t) = (Cn cos λn t + Dn sin λn t) sin wn x. (11.11)
n=1 n=1

The coefficients Cn , Dn are chosen such that the ICs (11.10) are satisfied.
We first check the IC u(x, 0) = f (x). This gives
∞
X
Cn sin wn x = f (x),
n=1

therefore, Cn must be Fourier sine coefficient of f (x), namely

L
2
Z
Cn = f (x) sin wn x dx, n = 1, 2, 3, · · · . (11.12)
L 0

For the 2nd IC ut (x, 0) = g(x), we first differentiate the solution

∞
X
ut (x, t) = (−λn Cn sin λn t + λn Dn cos λn t) sin wn x.
n=1

Then, we have
∞
X
ut (x, 0) = λn Dn sin wn x = g(x),
n=1

which is a Fourier sine series for g(x), and we have

L
1 2
Z
Dn = · g(x) sin wn x dx, n = 1, 2, 3, · · · . (11.13)
λn L 0

In summary, the formal solution for the wave equation (11.8) with BC (11.9) and IC (11.10)
is given in (11.11) with (11.12)+(11.13).
NB! Note that here the function f (x) and g(x) are extended to the whole real line by the
odd half-range expension.

Discussion on the formal solution:

• If g(x) = 0, then Dn = 0 for all n.

• If f (x) = 0, then Cn = 0 for all n.

185
 Remark: This solution takes the form of a Fourier series. It is very hard to see what’s
going on in the solution. In particular, one can not get any intuition on wave phenomenon in
the solution.
On the other hand, we can manipulate the solution by some trig identity. Consider the
simpler case where g(x) = 0 so Dn = 0, the solution takes the form
∞
X nπ
u(x, t) = Cn cos λn t sin wn x, λn = cwn , wn = .
L
n=1

Recall the trig identity

1 1
sin A cos B = sin(A − B) + sin(A + B).
2 2
We have
∞
X 1
u(x, t) = Cn [sin wn (x + ct) + sin wn (x − ct)]
2
n=1
∞ ∞
1X 1X
= Cn sin wn (x + ct) + Cn sin wn (x − ct)
2 2
n=1 n=1
1 ∗ 1
= f (x + ct) + f ∗ (x − ct), (11.14)
2 2
where f ∗ is the odd half-range periodic expansion of f .
Note that f (0) = 0 = f (l), then f ∗ is continuous.
Wave interpretation: (make some graphs.)

• f ∗ (x − ct): f ∗ travels with speed c as time t goes. (wave travels to the right.)

• f ∗ (x + ct): f ∗ travels with speed −c as time t goes. (wave travels to the left.)

New meaning of solution of wave equation with g(x) = 0: The initial deflection f ∗ is split into
two equal parts, one travels to the left, one to the right, with the speed c, and superposition
of them gives the solution.

Remark: Such a phenomenon can also be observed when g(x) is not 0. The computation
would be somewhat more involving. Try it on your own for practice.

Example : Do the triangle wave.

First write out solution in Fourier series.
Then, use (11.14), and sketch the graphs of solutions at various time t, as follows.

• First sketch f (x) on 0 < x < L, then extend it to odd, and periodic.
L 2L 3L 4L
• Sketch for t = 0, 4c , 4c , 4c , 4c , the graph of f ∗ (x + ct) (in white) and f ∗ (x − ct) (in
color) on the same graph, then the sum of them on a separate graph.

• summarize the behavior.

186
11.4 Laplace Equation in 2D (probably skip)
Consider the Laplace equation in 2D

uxx + uyy = 0

Some applications of this equation: steady state of 2D heat equation, electrostatic potential,
minimum surface problem, etc.
To begin with, we consider a rectangular domain R: 0 < x < a, 0 < y < b.
There are different types of BCs one can assign:

• Dirichlet BC: u is prescribed on the boundary.

∂u
• Neumann BC: un = ∂n normal derivative is given on the boundary.

• Robin BC: mixed BC, combine the Dirichlet and Neumann BCs.

We now start with Dirichlet BC. Let

u(0, y) = 0, u(a, y) = 0, u(x, 0) = 0, u(x, b) = f (x). (11.15)

Note that on 3 sides we have homogeneous conditions, and only on one side the condition is
non-homogeneous.
This will take several steps.
Step 1. Separating variables. Let

u(x, y) = F (x) · G(y).

Then
uxx = F ′′ (x)G(y), uyy = F (x)G′′ (y),
so
F ′′ (x) G′′ (y)
=− = −p (constant)
F (x) G(y)
This gives us 2 ODEs

F ′′ (x) + pF (x) = 0, G′′ (y) − pG(y) = 0.

Step 2. Solve for F . From the BCs, we have

F (0) = 0, F (a) = 0.

We solve this eigenvalue problem for F . This we have done several times. We have
nπ
p = wn2 , wn = , Fn (x) = sin wn x, n = 1, 2, 3, · · · .
a
Step 3. Solve for G(y). For each n, we have

G′′ (y) − wn2 G(y) = 0

187
which gives the general solution

Gn (y) = An ewn y + Bn e−wn y .

The BC at y = 0 gives the condition for G

G(0) = 0

Then, we have
An + Bn = 0, Bn = −An
so
Gn (y) = An (ewn y − e−wn y ) = 2An sinh wn y.
Recall that
2 sinh x = ex − e−x , 2 cosh x = ex + e−x .
Since An is arbitrary so far, so is 2An , and we will simply call it An . Then

Gn (y) = An sinh wn y.

Step 4. Put together, we get the eigenvalues and eigenfunctions

nπ
λn = wn2 , wn = , un = Fn (x)Gn (y) = An sinh wn y sin wn x, n = 1, 2, 3, · · · .
a
The formal solution is
∞
X ∞
X
u(x, y) = un (x, y) = An sinh wn y sin wn x. (11.16)
n=1 n=1

Now we fit in the last BC, i.e., u(x, b) = f (x):

∞
X
An sinh wn b sin wn x = f (x)
n=1

This is a Fourier sine series for f (x), and we must have

1
2
Z
An sinh wn b = f (x) sin wn x dx
a 0

which gives the coefficients An

1
2
Z
An = f (x) sin wn x dx (11.17)
a sinh(wn b) 0

In summary, the formal solution is given in (11.16) + (11.17).

How does the solution look like? Consider the minimum surface problem. There is the
maximum principle: The max or min value only occur on the boundary.

188
 Example : with a different boundary condition: (graph..)

u(0, y) = 0, u(a, y) = 0, u(x, 0) = g(x), u(x, b) = 0. (11.18)

One can carry out the same steps, and reach

nπ
wn = , Fn (x) = sin wn x, Gn (y) = An ewn y + Bn e−wn y , n = 1, 2, 3, · · · .
a
Now, fit in the boundary condition u(x, b) = 0, which gives Gn (b) = 0, we have

An ewn b + Bn e−wn b = 0, → An ewn b = −Bn e−wn b .

We may write
An ewn b = Cn , Bn e−wn b = −Cn
so
An = Cn e−wn b , Bn = −Cn ewn b .
which gives
h i
Gn (y) = Cn e−wn b ewn y − Cn ewn b e−wn y = Cn ewn (y−b) − e−wn (y−b) = 2Cn sinh wn (y − b).

Since 2Cn is arbitrary, we call it Cn , so

Gn (y) = Cn sinh wn (y − b), n = 1, 2, 3, · · · .

This gives the formal solution

∞
X ∞
X
u(x, y) = un (x, y) = Cn sinh wn (y − b) sin wn x. (11.19)
n=1 n=1

We now fit in the last BC, i.e., u(x, 0) = g(x), and we get
∞
X
Cn sinh(−wn b) sin wn x = g(x).
n=1

This is a Fourier sine series for g(x), requiring

2 a
Z
Cn sinh(−wn b) = g(x) sin wn x dx, n = 1, 2, 3, · · · .
a 0
This gives the formula for the coefficient Cn :
Z a
2
Cn = − g(x) sin wn x dx. n = 1, 2, 3, · · · (11.20)
a sinh(wn b) 0

In summary, the formal solution is given in (11.19)+(11.20).

Example : If the BCs are now (graphs)

u(0, y) = 0, u(a, y) = h(y), u(x, 0) = 0, u(x, b) = 0. (11.21)

189
or
u(0, y) = k(y), u(a, y) = 0, u(x, 0) = 0, u(x, b) = 0. (11.22)
We can simply switch the roles of x and y, and carry out the whole procedure.

Non-homogeneous BCs: If we have boundary values assigned non-zeros everywhere, what

do we do?

u(0, y) = k(y), u(a, y) = h(y), u(x, 0) = g(x), u(x, b) = f (x). (11.23)

Let u1 , u2 , u3 , u4 be the solutions with BCs (11.15), (11.18), (11.21) and (11.22), respec-
tively. (Make a graph.) Then, set

u = u1 + u2 + u3 + u4

By superposition, u solves the Laplace equation, and satisfies the boundary condition in
(11.23).
Possible examples on Neumann BC, or an example of mixed BC. ?

11.5 D’Alembert’s Solution of Wave Equation

In this section we derive the D’Alembert’s solution of wave equation. Consider the wave
equation in 1D
utt = c2 uxx (11.24)
We claim that
u1 (x, t) = φ(x + ct), u2 (x, t) = ψ(x − ct)
are solutions for (11.24) for arbitrary functions of φ, ψ. This will then imply that

u(x, t) = φ(x + ct) + ψ(x − ct) (11.25)

is a solution. This is called D’Alembert’s solution of wave equation.

Proof. We need to plug u1 and u2 into the wave equation (11.24) and check is the equation
holds. By the Chain Rule, we have

(u1 )x = φ′ (x + ct), (u1 )xx = φ′′ (x + ct),

and
(u1 )t = cφ′ (x + ct), (u1 )tt = c2 φ′′ (x + ct),
We clearly have (u1 )tt = c2 (u1 )xx . The proof for u2 is completely similar.

We now assign the initial conditions

u(x, 0) = f (x), ut (x, 0) = g(x), (11.26)

and derive the formula for the solution, i.e., determine the functions φ, ψ by these ICs in
(11.26).

190
 By the condition u(x, 0) = f (x), we get

φ(x) + ψ(x) = f (x). (11.27)

We differentiate u in t

ut (x, t) = cφ′ (x + ct) − cψ ′ (x − ct).

Then, the IC ut (x, 0) = g(x) gives

1
cφ′ (x) − cψ ′ (x) = g(x), → (φ(x) − ψ(x))′ = g(x). (11.28)
c
Integrate (11.28) from x0 to x, where x0 is arbitrary,
Z x
1 x
Z
(φ(s) − ψ(s))′ ds = g(s) ds.
x0 c x0

we get
x
1
Z
φ(x) − ψ(x) = φ(x0 ) − ψ(x0 ) + g(s) ds.
c x0

Call M = φ(x0 ) − ψ(x0 ), we have

x
1
Z
φ(x) − ψ(x) = M + g(s) ds. (11.29)
c x0

Add (11.29) to (11.27) and divide by 2, we get

x
M 1 1
Z
φ(x) = + f (x) + g(s) ds. (11.30)
2 2 2c x0

Then, we can recover ψ

x
M 1 1
Z
ψ(x) = − + f (x) − g(s) ds. (11.31)
2 2 2c x0

Plug (11.30)-(11.31) back into (11.25), we get

u(x, t) = φ(x + ct) + ψ(x − ct)

Z x+ct
1 1 1 1 x−ct
Z
= f (x + ct) + g(s) ds + f (x − ct) − g(s) ds
2 2c x0 2 2c x0
1 1 x+ct 1 x0
Z Z
= [f (x + ct) + f (x − ct)] + g(s) ds + g(s) ds
2 2c x0 2c x−ct
1 1 x+ct
Z
= [f (x + ct) + f (x − ct)] + g(s) ds.
2 2c x−ct

In summary, the solution is

x+ct
1 1
Z
u(x, t) = [f (x + ct) + f (x − ct)] + g(s) ds. (11.32)
2 2c x−ct

191
The solution consists of two parts, where the first term is caused by the initial deflection f (x),
and the second term is from the initial velocity g(x).
Note that, if this is a vibrating string problem, then f, g here in (11.32) are the odd
half-range expansions onto the whole real line.
Remark: The solution (11.32) is more general than Fourier Series solution. In fact, there
is no requirement for f (x), g(x) to be periodic. For any f, g defined on the whole real line, the
formula (11.32) gives the solution of the wave equation.
Example . Solve the wave equation by D’Alembert’s formula.
1
utt = uxx , u(x, 0) = sin 5x, ut (x, 0) = cos x
5
1
Here we have f (x) = sin 5x and g(x) = 5 cos x. This is just a practice of using the formula
(11.32) with c = 1. We first work out the integral
Z x+t
1 x+t 1
Z
g(s) ds = cos s ds = (sin(x + t) − sin(x − t)).
x−t 5 x−t 5
Then, we get the solution
1 1 1
u(x, t) = sin 5(x − t) + sin(x + t) + (sin(x + t) − sin(x − t))
2 2 10  
1 1 1 1
= sin 5(x + t) + sin(x + t) + sin 5(x − t) − sin(x − t) .
2 10 2 10
Note the first term is a function of x + t, i.e., φ(x + t), where
1 1
φ(u) = sin 5u + sin u,
2 10
and the second term is a function of x − t, i.e., ψ(x − t) where
1 1
ψ(u) = sin 5u − sin u.
2 10
Characteristics. As we see, solutions of the wave equation can be written as
u(x, t) = φ(x + ct) + ψ(x − ct).
This implies:
–φ(x + ct) is constant along lines of x + ct = K where K is constant;
–ψ(x − ct) is constant along lines of x − ct = K where K is constant;
In the t − x plan,
– x + ct =constant: are straight lines with slope −1/c,
– x − ct =constant: are straight lines with slope 1/c
Draw a graph.
These lines are paths where information is being carried along. They are called character-
istics of this problem.
Remark:
• This is a more general property for many PDEs, including non-linear PDEs.
• Characteristics lines might not be parallel to each other, or straight lines. The situation
could be very complicated...

192
11.6 Method of Characteristics; Classification of 2nd order lin-
ear PDEs.
General form of 2nd order quasi-linear PDEs: Let u(x, y) be the unknown (NB! It could be
u(x, t) also...)
Auxx + BUxy + Cuyy = F (x, y, u, ux , uy ). (11.33)
Define the discriminant
∆ = B 2 − 4AC. (11.34)
Type of PDE is determined only by ∆

• If ∆ > 0, the PDE is hyperbolic. Example: wave equation utt − c2 uxx = 0;

• If ∆ = 0, the PDE is parabolic. Example: heat equation ut = c2 uxx ;

• If ∆ < 0, the PDE is elliptic. Example: Laplace equation uxx + uyy = 0.

If A(x, y), B(x, y), C(x, y) are functions, then the equation can changed type. This is called
mixed type. In general, mixed type equations are difficult.

We now consider the homogeneous case

Auxx + BUxy + Cuyy = 0. (11.35)

Assume that y + rx =constant are the characteristics, i.e., we assume that

u(x, y) = φ(y + rx)

is a solution for (11.35) for an arbitrary function φ. We now try to find values of r.
By the Chain Rule, we have

ux = −rφ′ (y + rx), uxx = r 2 φ′′ (y + rx),

and
uy = φ′ (y + rx), uyy = φ′′ (y + rx), uxy = −rφ′′ (y + rx).
Plug these back into (11.35), we get

Ar 2 φ′′ (y + rx) − Brφ′′ (y + rx) + Cφ′′ (y + rx) = 0.

Since φ is arbitrary, then φ′′ is not zero in general. We can drop the common factor φ′′ (y + rx),
and get
Ar 2 − Br + C = 0. (11.36)
This is called the characteristic equation. Solve it for r. The types of the roots depend on the
discriminant ∆:

• If ∆ > 0 (hyperbolic), then r1 , r2 are real and distinct. (meaningful)

• If ∆ = 0 (parabolic), then r1 = r2 are real and repeated. (kind of meaningful)

• If ∆ < 0 (elliptic), then r1 , r2 are complex conjugate pair. (not meaningful)

193
 From now on, we will focus on the hyperbolic case, with constant coefficients. Let r1 , r2
be the roots. Then, the solution is

u(x, y) = φ1 (y − r1 x) + φ2 (y − r2 x)

where φ1 , φ2 are arbitrary functions, to be determined by BCs (or ICs if it is a time depending
problem).

Example . Find the solution by method of characteristics for the following equation

uxx + 3uxy + 2uyy = 0

with BCs
u(0, y) = f (y), ux (0, y) = 0.
To solve this, we set up the characteristic equation

r 2 − 3r + 2 = 0, r1 = 1, r2 = 2,

and the solution is

u(x, y) = φ1 (y − r1 x) + φ2 (y − r2 x) = φ1 (y − x) + φ2 (y − 2x).

We now determine φ1 , φ2 by BCs. The first BC gives

φ1 (y) + φ2 (y) = f (y).

For the 2nd BC, we first differentiate the solution

ux (x, y) = −φ′1 (y − x) − 2φ′2 (y − 2x).

Then, we have
−φ′1 (y) − 2φ′2 (y) = 0, (φ1 (y) + 2φ2 (y))′ = 0.
Integrating this, we get
φ1 (y) + 2φ2 (y) = M
where M is any integration constant. We can now solve

φ2 (y) = M − f (y), φ1 (y) = f (y) − φ1 (y) = 2f (y) − M.

Putting these back into the solution, we get

u(x, y) = (2f (y − x) − M ) + (M − f (y − 2x)) = 2f (x − y) − f (y − 2x).

Example . One more example on Method of Characteristics. Consider

utt − utx = 0,

where u(t, x) is the unknown, and the ICs are

u(0, x) = f (x), ut (0, x) = g(x).

194
 To solve, we set up the char equation

r 2 + r = 0, r1 = 0, r2 = 1

Then
u(t, x) = φ(x − r1 t) + ψ(x − r2 t) = φ(x) + ψ(x − t).
By the first IC, we have
φ(x) + ψ(x) = f (x). (11.37)
For the 2nd IC, we differentiate

ut (t, x) = 0 − ψ ′ (x − t)

then
ut (0, x) = −ψ ′ (x) = g(x).
Integrating this equation from x0 (arbitrary) to x, we get
Z x
ψ(x) = ψ(x0 ) − g(s) ds.
x0

Then, we get Z x
φ(x) = f (x) − ψ(x) = f (x) − ψ(x0 ) + g(s) ds.
x0

Putting then into solution

u(t, x) = φ(x) + ψ(x − t)

 Z x   Z x−t 
= f (x) − ψ(x0 ) + g(s) ds + ψ(x0 ) − g(s) ds
x0 x0
Z x Z x0
= f (x) + g(s) ds + g(s) ds
x0 x−t
Z x
= f (x) + g(s) ds.
x−t

195
Chapter 12

Homeworks

12.1 Homework 1

Problem 1: In each problem, determine the order of the equation and state whether the
equation is linear or nonlinear.

(a) t2 y ′′ + ty ′ + 2y = sin t;

(b) y (4) + y (3) + y ′′ + y ′ + y = 1;

(c) y ′′ + sin(t + y) = sin t.

Problem 2: In each problem, verify that each given function is a solution of the differential
equation.

(a) ty ′ − y = t2 ; y(t) = 3t + t2 ;

(b) y ′′′′ + 4y ′′′ + 3y = t; y1 (t) = t/3, y2 (t) = e−t + t/3;

√
(c) 2t2 y ′′ + 3ty ′ − y = 0, t > 0; y1 (t) = t, y2 (t) = t−1 .

Problem 3: In each problem,draw a direction field for the given differential equation.
Based on the direction field, determine the behavior of y as t → ∞. Describe how this
behavior depends on the initial value of y at t = 0.

(a) y ′ = 1 + 2y;

(b) y ′ = y + 2;

(c) y ′ = y 2 ;

(d) y ′ = y(y − 2)2 ;

(e) y ′ = −2 + t − y;

(f) y ′ = t + 2y;

(g) y ′ = −t/y. (You don’t need to discuss asymptotical behavior for this problem.)

196
 Problem 4: Exam I, Spring 2013, Problem 2.
Problem 5: In each problem, solve the differential equation.

(a) y ′ = x2 /y;

(b) y ′ = (cos2 x)(cos2 y);

(c) xy ′ = (1 − y 2 )1/2 ;

Problem 6: In each problem, solve the differential equation, and determine approximately
the interval in which the solution is defined. You might use graphing tools if needed.

(a) y ′ = (1 − 2x)/y, y(1) = −2;

(b) x dx + ye−x dy = 0, y(0) = 1;

(c) y ′ = (e−x − ex )/(3 + 4y), y(0) = 1;

(d) y 2 (1 − x2 )1/2 dy = arcsin x dx, y(0) = 1;

(e) y ′ = 3x2 /(3y 2 − 4), y(1) = 0.

Problem 7: Exam I, Fall 2013, Problem 3.

A collection of previous exams could be found at the coordinator’s web:
http://www.math.psu.edu/tseng/class/M251samples.html

197
12.2 Homework 2
Problem 1. Find the general solution of the given differential equation. If an initial condition
is given, find the particular solution which satisfies this initial condition.

a. y ′ + 3y = et , y(0) = −2;

b. y ′ − 2y = e2t , y(0) = 4;

c. ty ′ + y = et , y(1) = 0;
y
d. y ′ = − + cos(t2 );
t
e. y ′ − 3y = 25 cos(4t);

f. z ′ = 2t(z − t2 );

g. y ′ + y cos t = cos t, y(0) = 0;

2 t+1
h. y ′ − y = , y(1) = −3;
t t
i. y ′ + ay = ebt , where a and b are constant and b + a 6= 0;

j. t2 y ′ + 2ty = 1, y(2) = a, where a is a constant.

Problem 2. Previous Exam Problems:

a. Fall 2013, Exam I, problem 2.

b. Fall 2012, Exam I, problem 2.

Problem 3. Without solving the problems, determine an interval in which the solution of
the given initial value problem is certain to exist.

a. (t − 3)y ′ + (ln t)y = 2t, y(1) = 2

b. t(t − 4)y ′ + y = 0, y(2) = 1

c. y ′ + (tan t)y = sin t, y(π) = 0

d. (4 − t2 )y ′ + 2ty = 3t2 , y(−3) = 1

e. (4 − t2 )y ′ + 2ty = 3t2 , y(1) = −3

f. (ln t)y ′ + y = cot t, y(2) = 3

Problem 4. Previous Exam Problems:

a. Fall 2013, Exam I, problem 9.

b. Fall 2012, Exam I, problem 4.

198
 Problem 5. A tank contains 10gal of brine in which 2 lb of salt is dissolved. Brine
containing 1 lb of salt per gallon flows into the tank at the rate of 3 gal/min, and the stirred
mixture is drained off the tank at the rate of 4 gal/min. Find the amount y(t) of salt in the
tank at any time t.
Problem 6. A 30-L container initially contains 10L of pure water. A brine solution
containing 20 g of salt per liter flows into the container at a rate of 4 L/min. The well-stirred
mixture is pumped out of the container at a rate of 2 L/min.

(a) How long does it take for the container to overflow?

(b) How much salt is in the tank at the moment the tank begins to overflow?

Problem 7. Previous Exam Problems:

a. Fall 2012, Exam I, problem 5

b. Fall 2013, Exam I, problem 5

A collection of previous exams could be found at the coordinator’s web:

http://www.math.psu.edu/tseng/class/M251samples.html

199
12.3 Homework 3
Problem 1. Spring 2013, Exam I, problem 10.
Problem 2. Find the critical points and equilibrium solutions of the given ODEs. Sketch
the graphs of the solutions with the given initial conditions. Determine on the stabilities of
these critical points.

a. y ′ = 300y − 2y 2 ; y(0) = 50; y(0) = 100; y(0) = 200

b. y ′ = 15y − y 2 /2; y(0) = 10; y(0) = 20; y(0) = 40

c. y ′ = 8y − 2y 2 − 6; y(0) = 0.5; y(0) = 1.5; y(0) = 3.5

d. y ′ = 2y 2 − 80y; y(0) = 10; y(0) = 30; y(0) = 50

e. y ′ = y 2 (y − 1); y(0) = −1; y(0) = 0.5; y(0) = 1.5

Problem 3. Consider the equation dy/dt = f (y) and suppose that y1 is a critical point,
i.e., f (y1 ) = 0, and f is a continuous function. Show that y(t) = y1 is asymptotically stable if
f ′ (y1 ) < 0 and unstable if f ′ (y1 ) > 0.
Problem 4. Previous Exam Problems.

a. Fall 2013, Exam I, Problem 8;

b. Spring 2013, Exam I, Problem 11;

c. Fall 2012, Exam I, Problem 11.

Problem 5. Verify that the following problems are exact. Then solve the IVP (Initial
Value Problems).

a. 4xy + 2x2 − 6y y ′ = 0, y(1) = 1;

 

b. 3x2 y −2 + x−2 y 2 − 2x−3 + 6y −4 − 2x3 y −3 − 2x−1 y y ′ = 0, y(1) = −1.

 

c. 2e2x − 2 sin(2x) sin y + 2y −3 + cos(2x) cos y y ′ = 0, y(0) = π/2.

 

Problem 6. Show that any separable equation written as M (x) + N (y)y ′ = 0 is exact.
Problem 7. Previous Exam Problems.

a. Fall 2013, Exam I, Problem 7 and Problem 11;

b. Spring 2013, Exam I, Problem 5;

c. Fall 2012, Exam I, Problem 3.

A collection of previous exams could be found at the coordinator’s web:

http://www.math.psu.edu/tseng/class/M251samples.html

200
12.4 Homework 4
Problem 1. Find the general solution of the given differential equations.

a. y ′′ + 2y ′ − 3y = 0

b. 6y ′′ − y ′ − y = 0

c. y ′′ + 5y ′ = 0

d. y ′′ − 9y ′ + 9y = 0

Problem 2. Find the solution y(t) of the given IVP (Initial Value Problems).

a. y ′′ + y ′ − 6y = 0, y(0) = 2, y ′ (0) = 9

b. 2y ′′ + y ′ − y = 0, y(0) = 4, y ′ (0) = −5/2

c. y ′′ + 4y ′ = 0, y(0) = −2, y ′ (0) = −4

d. 6y ′′ − y ′ − y = 0, y(0) = 2, y ′ (0) = −3/2

Problem 3. Find the differential equation whose general solution is given.

a. y(t) = c1 + c2 et

b. y(t) = c1 e2t + c2 e3t

Problem 4. Solve the IVP

y ′′ − y ′ − 2y = 0, y(0) = a, y ′ (0) = 2.

Then find a so that the solution approach 0 as t → ∞.

Problem 5. Find the Wronskian of the given pair of functions

a. e2t , e−3t/2

b. cos t, sin t

c. e−2t , te−2t

d. et sin t, et cos t

Problem 6. Without solving the equation, find the longest interval in which the given
IVP is certain to have a unique twice differential solution.

a. ty ′′ + 3y = t, y(1) = 1, y ′ (1) = 2

b. (t − 1)y ′′ − 3ty ′ + 4y = sin t, y(−2) = 2, y ′ (−2) = 1

c. y ′′ + (cos t)y ′ + 3(ln |t|)y = 0, y(2) = 3, y ′ (2) = 1

d. (x − 3)y ′′ + xy ′ + (ln |x|)y = 0, y(1) = 0, y ′ (1) = 1

Problem 7.

201
 a. If W (f, g) = 3e4t and if f (t) = e2t , find g(t).

b. If W (f, g) = t2 et and if f (t) = t, find g(t).

Problem 8. Assume that y1 and y2 are a fundamental set of solutions of y ′′ + p(t)y ′ +

q(t)y = 0 and let y3 = ay1 + by2 and y4 = cy1 + dy2 , where a, b, c, d are any constants. Show
that
W (y3 , y4 ) = (ad − bc) · W (y1 , y2 ).
Are y3 , y4 also a fundamental set of solutions? Why or why not?
Problem 9. Previous Exam Problems.

a. Fall 2013, Exam I, Problem 4, 9, 13.

b. Spring 2013, Exam I, Problem 6, 7;

c. Fall 2012, Exam I, Problem 7, 8.

A collection of previous exams could be found at the coordinator’s web:

http://www.math.psu.edu/tseng/class/M251samples.html

202
12.5 Homework 5
Problem 1. Find the general solution of the given equations.

a. 9y ′′ + 6y ′ + y = 0

b. 4y ′′ − 4y ′ − 3y = 0

c. 4y ′′ + 12y ′ + 9y = 0

Problem 2. Solve the following IVP (Initial Value Problems). Sketch the graph of the
solution and describe the asymptotic behavior as t → ∞.

a. 9y ′′ − 12y ′ + 4y = 0, y(0) = 2, y ′ (0) = −1

b. y ′′ − 6y ′ + 9y = 0, y(0) = 0, y ′ (0) = 2

c. y ′′ + 4y ′ + 4y = 0, y(−1) = 2, y ′ (−1) = 1.

Problem 3. Consider the following IVP

y ′′ − y ′ + 0.25y = 0, y(0) = 2, y ′ (0) = b.

Find the solution (which will contain the constant b). Then determine the critical value of b
that separates solutions that grow positively from those that eventually grow negatively.
Problem 4. Use the method of reduction of order to find a second solution of the given
differential equations.

a. t2 y ′′ + 2ty ′ − 2y = 0, t > 0; y1 (t) = t

b. t2 y ′′ + 3ty ′ + y = 0, t > 0; y1 (t) = t−1

c. t2 y ′′ − t(t + 2)y ′ + (t + 2)y = 0, t > 0; y1 (t) = t

d. (x − 1)y ′′ − xy ′ + y = 0, x > 1; y1 (x) = ex

Problem 5. Find the general solution of the given equations.

a. y ′′ − 2y ′ + 6y = 0

b. y ′′ + 2y ′ + 2y = 0

c. y ′′ + 4y ′ + 6.25y = 0

Problem 6. Solve the following IVP (Initial Value Problems). Sketch the graph of the
solution and describe the asymptotic behavior as t → ∞.

a. y ′′ + 4y ′ + 5y = 0, y(0) = 1, y ′ (0) = 0

b. y ′′ − 2y ′ + 5y = 0, y(π/2) = 0, y ′ (π/2) = 2

c. y ′′ + y ′ + 1.25y = 0, y(0) = 3, y ′ (0) = 1

203
Problem 7. Consider the IVP

y ′′ + 2y ′ + 6y = 0, y(0) = 2, y ′ (0) = a ≥ 0.

a. Find the solution y(t) of this problem.

b. Find a so that y = 0 when t = 1. (You may use a calculator)

c. Find, as a function of a, the smallest positive value of t for which y = 0.

d. Determine the limit of the expression found in part (c) as a → ∞.

Problem 8. Previous Exam Problems.

a. Fall 2012, Exam I, Problem 12.

b. Spring 2012, Exam I, Problem 7, 12.

c. Fall 2011, Exam I, Problem 8, 13.

A collection of previous exams could be found at the coordinator’s web:

http://www.math.psu.edu/tseng/class/M251samples.html

204
12.6 Homework 6
Problem 1. Find the general solution of the given equations.
a. y ′′ − 2y ′ − 3y = 3e2t

b. y ′′ + 2y ′ + 5y = 3 sin 2t

c. y ′′ + 9y = t2 e3t + 6
Problem 2. Solve the following IVP (Initial Value Problems). Be patient! These problems
take a lot of time!
a. y ′′ + y ′ − 2y = 2t, y(0) = 0, y ′ (0) = 1

b. y ′′ + 4y = t2 + 3et , y(0) = 0, y ′ (0) = 2

c. y ′′ + 4y = 3 sin 2t y(0) = 2, y ′ (0) = −1.

Problem 3. Determine a suitable form for Y (t) for the following non-homogeneous equa-
tions, if the method of undetermined coefficients is to be used. (Do not solve for the coefficients.
)
a. y ′′ + 3y ′ = 2t4 + t2 e−3t + sin 3t

b. y ′′ − 5y ′ + 6y = et cos 2t + e2t (3t + 4) sin t

c. y ′′ − 4y ′ + 4y = 2t2 + 4te2t + t sin 2t

d. y ′′ + 3y ′ + 2y = et (t2 + 1) sin 2t + 3e−t cos t + 4et

Problem 4. Determine the general solution of
N
X
y ′′ + w2 y = am sin mπt
m=1

where w > 0, N > 0 is a positive integer, and w 6= mπ for any m = 1, 2, · · · , N

Problem 5. Write the following functions into the form u = R cos(ω0 t − δ).
a. u = 3 cos 2t + 4 sin 2t
√
b. u = − cos t + 3 sin t

c. u = 4 cos 3t − 2 sin 3t

d. u = −2 cos πt − 3 sin πt
Problem 6. A mass weighing 3 lb stretches a spring 3 in. If the mass is pushed upward,
contracting the spring a distance of 1 in., and then set in motion with a downward velocity of
2 ft/s, and if there is no damping, find the position u of the mass at any time t > 0. Determine
the frequency, period, amplitude, and phase of the motion.
Problem 7. A mass weighing 8 lb stretches a spring 1.5 in. The mass is also attached
to a damper with coefficient γ. Determine the value of γ for which the system is critically
damped.

205
 Problem 8. The position of a certain spring-mass system satisfies the initial value problem
3 ′′
u + ku = 0, u(0) = 2, u′ (0) = ν
2
If the period and amplitude of the resulting motion are observed to be π and 3, respectively,
find the values of k and ν.
Problem 9. Previous Exam Problems.

a. Fall 2013, Exam I, Problem 10, 12;

b. Spring 2013, Exam I, Problem 12;

c. Fall 2012, Exam I, Problem 13.

A collection of previous exams could be found at the coordinator’s web:

http://www.math.psu.edu/tseng/class/M251samples.html

206
12.7 Homework 7
Problem 1. Write the given expression as a product of two trigonometric functions of different
frequencies.

a. cos 9t − cos 7t

b. sin 7t − sin 6t

c. cos πt + cos 2πt

Problem 2. Determine intervals in which solutions are sure to exist.

a. y (4) + 4y ′′′ + 3y = t,

b. ty ′′′ + (sin t)y ′′ + 3y = cos t

c. t(t − 1)y (t) + et y ′′ + 4t2 y = 0

Problem 3. Find the general solution of the given differential equation.

a. y ′′′ − y ′′ − y ′ + y = 0

b. y ′′′ − 3y ′′ + 3y ′ − y = 0

c. y (6) − y ′′ = 0

d. y (4) + 2y ′′ + y = 0

Problem 4. Previous Exam Problems.

a. Fall 2013, Exam II, Problem 1; (Note these are not Exam I).

b. Spring 2013, Exam I, Problem 8, 13;

c. Fall 2012, Exam I, Problem 10.

d. Fall 2012, Exam II, Problem 1.

e. Spring 2012, Exam II, Problem 1, 2, 5, 9a.

A collection of previous exams could be found at the coordinator’s web:

http://www.math.psu.edu/tseng/class/M251samples.html

207
12.8 Homework 8
Problem 1. Find the Laplace transform of the given functions by definition (without using
the Table 6.2.1). This means you should find a way to work out the integrals.

a. f (t) = cosh(bt), recall cosh x = 12 (ex + e−x );

b. f (t) = sinh(bt), recall sinh x = 21 (ex − e−x );

c. f (t) = eat cosh(bt);

d. f (t) = eat sinh(bt).

Problem 2. Use Euler’s formula eix = cos x + i sin x to find the Laplace transform of
f (t) = eat sin(bt) and g(t) = eat cos(bt).
Problem 3. Use Integration by parts to find the Laplace transform of the following
functions.

a. f (t) = teat

b. f (t) = t sin(at), g(t) = t cos(at)

c. f (t) = t sinh(at), g(t) = t cosh(at)

Problem 4. Recall the following property of the Laplace transform:

If F (s) = L{f (t)}, then F ′ (s) = L{−tf (t)}.

a. Show that F ′′ (s) = L{(−t)2 f (t)}, and furthermore F (n) (s) = L{(−t)n f (t)} for all inte-
ger n ≥ 1.

b. Use these observations to find the Laplace transforms of the functions

f (t) = tn eat , g(t) = teat sin(bt), h(t) = teat cos(bt).

Problem 5. Find the inverse Laplace transform of the given functions.

3
a. F (s) = ;
s2 +4
4
b. F (s) = ;
(s − 1)3
2
c. F (s) = ;
s2 + 3s − 4
3s
d. F (s) = :
s2 −s−6
2s + 2
e. F (s) = ;
s2 + 2s + 5
8s2 − 4s + 12
f. F (s) = ;
s(s2 + 4)

208
 2s − 3
g. F (s) = .
s2 + 2s + 10
Problem 6. Use Laplace transform to solve the following IVPs. Be patient! These
problems take a lot of work!

a. y ′′ − y ′ − 6y = 0; y(0) = 1, y ′ (0) = −1;

b. y ′′ − 2y ′ + 2y = 0; y(0) = 0, y ′ (0) = 1;

c. y ′′ − 4y ′ + 4y = 0; y(0) = 1, y ′ (0) = 1;

d. y (4) − 4y ′′′ + 6y ′′ − 4y ′ + y = 0; y(0) = 0, y ′ (0) = 1, y ′′ (0) = 0, y ′′′ (0) = 1;

e. y (4) − y = 0; y(0) = 1, y ′ (0) = 0, y ′′ (0) = 1, y ′′′ (0) = 0;

f. y ′′ + w2 y = cos(2t), w2 6= 4; y(0) = 1, y ′ (0) = 0;

Problem 7. Previous Exam Problems. (Note all the problems are now from Exam II. )

a. Fall 2013, Exam II, Problem 2, 5;

b. Spring 2013, Exam II, Problem 2;

c. Fall 2012, Exam II, Problem 2, 10a.

d. Spring 2012, Exam II, Problem 8a.

A collection of previous exams could be found at the coordinator’s web:

http://www.math.psu.edu/tseng/class/M251samples.html

209
12.9 Homework 9
Problem 1. Sketch the graph of the given functions and express f (t) in terms of the unit
step function uc (t).


 0, 0≤t<3 

−2, 3≤t<5 1, 0≤t<2
(a). f (t) = (b). f (t) = −(t−2)

 2, 5≤t<7 e , t≥2
1, 7 ≤ t.


Problem 2. Find the Laplace transform of the given functions.


0, 0≤t<2
a. f (t) = 2
(t − 2) , t ≥ 2

0, 0≤t<1
b. f (t) =
t2 − 2t + 2, t≥1

 0, 0≤t<π
c. f (t) = t − π, π ≤ t ≤ 2π
0, 2π < t


d. f (t) = u1 (t) + 2u3 (t) − 6u4 (t)

e. f (t) = (t − 3)u2 (t) − (t − 2)u3 (t).

f. f (t) = t − u1 (t)(t − 1), t ≥ 0.

Problem 3. Find the inverse Laplace transform of the given functions.

e−2s
a. F (s) =
s2 +s−2
2(s − 1)e−2s
b. F (s) =
s2 − 2s + 2
2e−2s
c. F (s) =
s2 − 4
(s − 2)e−s
d. F (s) =
s2 − 4s + 3
e−s + e−2s − e−3s − e−4s
e. F (s) =
s

210
 Problem 4. Suppose that F (s) = L{f (t)} exists for s > d ≥ 0.
1 s
a. Show that if c is a positive constant, then L{f (ct)} = F , s > cd.
c c
 
−1 1 t
b. Show that if k is a positive constant, then L {F (ks)} = f .
k k
c. Show that if a and b are constants with a > 0. Then
 
−1 1 t
L {F (as + b)} = e−bt/a f .
a a

d. Utilize these results to find the inverse Laplace transform of the following functions.

2n+1 n! 2s + 1 e2 e−4s
F1 (s) = , F2 (s) = , F3 (s) = .
sn+1 2
4s + 4s + 5 2s − 1

Problem 5. Find the solution of the given IVP. Be patient! These problems take long
time!

′′ ′ 1, 0 ≤ t < 3π
a. y + y = f (t); y(0) = 0, y (0) = 1; f (t) =
0, 3π ≤ t

b. y ′′ + 4y = sin t + uπ (t) sin(t − π); y(0) = 0, y ′ (0) = 0

c. y ′′ + 3y ′ + 2y = u2 (t); y(0) = 0, y ′ (0) = 1


′′ ′ t/2, 0 ≤ t < 6
d. y + y = g(t); y(0) = 0, y (0) = 1; g(t) =
3, 6≤t

Problem 6. Find the solution of the given IVP. (Expect a lot of work!)

a. y ′′ + 2y ′ + 2y = δ(t − π); y(0) = 1, y ′ (0) = 0;

b. y ′′ + 4y = δ(t − π) − δ(t − 2π); y(0) = 0, y ′ (0) = 0;

c. y ′′ + 3y ′ + 2y = δ(t − 5) + u10 (t); y(0) = 0, y ′ (0) = 1/2;

d. y ′′ + y = δ(t − 2π) cos t; y(0) = 0, y ′ (0) = 1;

Problem 7. Previous Exam Problems. (Note all the problems are now from Exam II. )

a. Fall 2013, Exam II, Problem 3, 4, 11, 13;

b. Spring 2013, Exam II, Problem 1ab, 4, 8, 9;

A collection of previous exams could be found at the coordinator’s web:

http://www.math.psu.edu/tseng/class/M251samples.html

211
12.10 Homework 10
Problem 1. In each of the problem, transform the given equation into a system of first order
equations. If ICs are given, specify these as well.

a. u′′ + 0.5u′ + 2u = 0;

b. t2 u′′ + tu′ + (t2 − 0.25)u = 0;

c. u(4) − u = 0;

d. u′′ + 0.25u′ + 4u = 2 cos 3t, u(0) = 1, u′ (0) = −2;

e. u′′ + p(t)u′ + q(t)u = g(t), u(0) = u0 , u′ (0) = u′0 .

Problem 2. Find the eigenvalues and their corresponding eigenvectors of the following
matrices.
       
5 −1 −2 1 −3 3/4 −3 4
(a). , (b). , (c). (d).
3 1 1 −2 −5 1 0 1

Problem 3. For each problem, find the general solution and sketch the phrase portrait.
You may use some of the results in Problem 2.
     
′ 5 −1 ′ −2 1 ′ −3 4
(a). ~x = ~x, (b). ~x = ~x, (c). ~x = ~x,
3 1 1 −2 0 1

Problem 4. For each problem, find the general solution and sketch the phrase portrait.
Note that some of the eigenvalues are 0.
   
4 −3 3 6
(a). ~x′ = ~x, (b). ~x′ = ~x,
8 −6 −1 −2

Problem 5. Consider the IVP ~x′ = A~x, x1 (0) = 2, x2 (0) = 3. Given the eigenvalues and
eigenvectors of A, (i) write out the general solution; (ii) solve the IVP; (iii) sketch the phase
portrait; and (iv) sketch the trajectory passing through the initial point (2, 3).
   
−1 1
(a). λ1 = −1, ~v1 = , λ2 = −2, ~v2 =
2 2
   
−1 1
(b). λ1 = 1, ~v1 = , λ2 = −2, ~v2 =
2 2

212
Problem 6. Previous Exam Problems.

a. Fall 2013, Exam II, Problem 6, 14;

b. Spring 2013, Exam II, Problem 3, 5, 6;

c. Fall 2012, Exam II, Problem 5;

d. Spring 2012, Exam II, Problem 6, 10.

A collection of previous exams could be found at the coordinator’s web:

http://www.math.psu.edu/tseng/class/M251samples.html

213
12.11 Homework 11
Problem 1. For each of the problem, find the eigenvalues and eigenvectors of the coefficient
matrix, find the general solution, state the type of the equilibrium and the stability, and sketch
the phase portrait.
     
′ 3 −2 ′ −1 −4 ′ 2 −5
(a). ~x = ~x, (b). ~x = ~x, (c). ~x = ~x.
4 −1 1 −1 1 −2

Problem 2. Find the solution for the IVPs.

       
′ 1 −5 1 ′ −3 2 1
(a). ~x = ~x, ~x(0) = , (b). ~x = ~x, ~x(0) = ,
1 −3 1 −1 −1 −2

Problem 3. For each of the problem, find the eigenvalues and eigenvectors of the coefficient
matrix, find the general solution, state the type of the equilibrium and the stability, and sketch
the phase portrait.
   
′ 3 −4 ′ 4 −2
(a). ~x = ~x, (b). ~x = ~x.
1 −1 8 −4

Problem 4. Find the solution for the IVPs.

       
′ 1 −4 3 ′ 3 9 2
(a). ~x = ~x, ~x(0) = , (b). ~x = ~x, ~x(0) = ,
4 −7 2 −1 −3 4

Problem 5. Find all the critical points.

a. x′ = x − xy; y ′ = y + 2xy;

b. x′ = 1 + 2y, y ′ = 1 − 3x2 ;

c. x′ = 2x − x2 − xy, y ′ = 3y − 2y 2 − 3xy;

d. x′ = (2 + x)(y − x), y ′ = y(2 + x − x2 ).

214
 Problem 6. For each problem: (i) Find all the critical points; (ii) Find the corresponding
linear system near each critical point; (iii) Find the eigenvalues for each system; (iv) Classify
the type of each critical point and its stability. (NB! These problems take long time!)

a. x′ = (2 + x)(y − x); y ′ = (4 − x)(y + x);

b. x′ = 1 − y, y ′ = x2 − y 2 ;

c. x′ = (2 + y)(y − 0.5x), y ′ = (2 − x)(y + 0.5x);

Problem 7. Previous Exam Problems.

a. Fall 2013, Exam II, Problem 7, 8, 9, 10, 12;

b. Spring 2013, Exam II, Problem 1cd, 6, 7, 10, 11;

c. Fall 2012, Exam II, Problem 6, 7, 8, 12;

A collection of previous exams could be found at the coordinator’s web:

http://www.math.psu.edu/tseng/class/M251samples.html

215
12.12 Homework 12
Problem 1. For each of the problem, (i) sketch the graph of the given function for three
periods, (ii) find the Fourier series, (iii) write out the first 5 non-zero terms in the series, and
(iv) describe how the Fourier series seems to be converging.
(a). f (x) = −x,
( −L ≤ x < L; f (x + 2L) = f (x);
1, −L ≤ x < 0,
(b). f (x) = f (x + 2L) = f (x);
0, 0 ≤ x < L;
(
x, −π ≤ x < 0,
(c). f (x) = f (x + 2π) = f (x);
0, 0 ≤ x < π;

0, −2 ≤ x ≤ −1,

(d). f (x) = x, −1 ≤ x < 1, f (x + 4) = f (x);

0, 1 ≤ x < 2;

(
−1, −2 ≤ x < 0,
(e). f (x) = f (x + 4) = f (x);
1, 0 ≤ x < 2;
Problem 2*. (optional) Suppose f is an integrable and differentiable periodic function
with period T .
(a). Show that f ′ is periodic with the same period T .
(b). Show that for any a, b, we have
Z a+T Z b+T
f (t) dt = f (t) dt.
a b
Rt
(c). Determine whether F (t) = 0 f (s) ds is always periodic.
Problem 3. Assume that f has a Fourier series
∞
a0 X  nπx nπx 
f (x) = + an cos + bn sin ,
2 n=1
L L

show formally that

L ∞
1 a2 X 2
Z
[f (x)] dx = 0 +
2
an + b2n .


L −L 2 n=1
This relation is known as the Parseval’s identity, and is an important result. (Hint: Recall
how we derived the formula for computing an , bn .)
Problem 4. In each of the problem, a function f is given on and interval of length L.
Sketch the graphs
( of the even and odd extension of f of period 2L.
x, 0 ≤ x < 2,
(a). f (x) =
1, 2 ≤ x < 3;
(
0, 0 ≤ x < 1,
(b). f (x) =
x − 1, 1 ≤ x < 2;
(c). f (x) = 2 − x, 0 < x < 2.
(d). f (x) = x − 3, 0 < x < 4.
Problem 5. In each of the following problem, either solve the given boundary value
problem or else show that it has no solution.

216
(a). y ′′ + y = 0, y(0) = 0, y ′ (π) = 1;
(b). y ′′ + 2y = 0, y ′ (0) = 1, y ′ (π) = 0;
′′
(c). y + y = 0, y(0) = 0, y ′ (L) = 0;
(d). y ′′ + y = x, y(0) = 0, y(π) = 0;
(e). y ′′ + 4y = cos x, y ′ (0) = 0, y ′ (π) = 0.
Problem 6. Solve the eigenvalue problems. Assume that all eigenvalues are real.
(a). y ′′ + λy = 0, y(0) = 0, y ′ (π) = 0;
′′
(b). y + λy = 0, y ′ (0) = 0, y(π) = 0;
′′
(c). y + λy = 0, y ′ (0) = 0, y ′ (π) = 0;
(d). y ′′ + λy = 0, y ′ (0) = 0, y(L) = 0;
Problem 7. Previous Exam Problems. Note that these are Final Exams!

a. Fall 2013, Final Exam, Problem 1, 11, 13, 14;

b. Spring 2013, Final Exam, Problem 2, 10, 12, 13;

c. Fall 2012, Final Exam, Problem 2, 14, 15;

A collection of previous exams could be found at the coordinator’s web:

http://www.math.psu.edu/tseng/class/M251samples.html

217
12.13 Homework 13
Problem 1. For each of the problem, determine whether the method of separation of variables
can be used to replace the given PDE by a pair of ODEs. If so, find the ODEs.
(a). xuxx + ut = 0
(b). tuxx + xut = 0
(c). uxx + uxt + ut = 0
(d). [p(x)ux ]x − r(x)utt = 0
(e). uxx + (x + y)uyy = 0
(f). uxx + uyy + xu = 0
Problem 2. Find the solution of the heat conduct problem

100uxx = ut , 0 < x < 1, t>0

u(0, t) = 0, u(1, t) = 0, t>0
u(x, 0) = sin 2πx − sin 5πx, 0 ≤ x ≤ 1.

Problem 3. Find the formal solution of the heat conduct problem

uxx = ut , 0 < x < 40, t>0

u(0, t) = 0, u(40, t) = 0, t>0
u(x, 0) = 50, 0 ≤ x ≤ 40.

Problem 4*. (Optional) The heat conduct equation in two space dimension is

α2 (uxx + uyy ) = ut .

Assuming that u(x, y, t) = X(x) Y (y) T (t), find ODEs that are satisfied by the functions
X(x), Y (y), and T (t).
Problem 5*. (Optional) The heat conduct equation in two space dimensions may be
expressed in terms of polar coordinates as

α2 urr + (1/r)ur + (1/r 2 )uθθ = ut .

 

Assuming that u(r, θ, t) = R(r) Θ(θ) T (t), find ODEs that are satisfied by R(r), Θ(θ), and
T (t).
Problem 6. In each of problem, find the steady state solution of the heat conduct equation
c2 uxx = ut that satisfies the given set of boundary conditions.
(a). u(0, t) = 10, u(50, t) = 40
(b). ux (0, t) = 0, u(L, t) = T
(c). ux (0, t) − u(0, t) = 0, u(L, t) = T
(d). u(0, t) = T, ux (L, t) + u(L, t) = 0
Problem 7. Let an aluminum rod of length 20cm be initially at the uniform temperature
of 25o C. Suppose that at time t = 0, the end x = 0 is cooled to 0O C while the end x =
20 is heated to 60o C, and both are thereafter maintained at those temperatures. Find the
temperature distribution in the rod at any time t.
Problem 8. Previous Exam Problems. Note that these are Final Exams!

a. Fall 2013, Final Exam, Problem 9, 10, 15;

218
b. Spring 2013, Final Exam, Problem 8, 9, 14, 16;

A collection of previous exams could be found at the coordinator’s web:

http://www.math.psu.edu/tseng/class/M251samples.html

219
12.14 Homework 14
Problem 1. (This problem takes a lot of work!) Consider the wave equation for an vibrating
string with length L = 10, as

utt = uxx , 0 < x < 10, t>0

The ends of the string are fixed. The initial displacement of the string is given as u(x, 0) =
f (x), where f (x) = x on 0 < x < 5, and f (x) = 10 − x on 5 < x < 10. The string is initially
at rest.
(a). Plot the initial displacement;
(b). Find the formal solution u(x, t) in term of Fourier series;
(c*) (optional). Write out the D’Alembert solution of this wave equation;
(d*) (optional). Plot u(x, t) as a function of x, for several t values, such as t = 0, t = 2.5, t =
5, t = 7.5, t = 10.
Problem 2. Consider the same problem as in Problem 1, but with a different function
f (x). Let 0 < a < 10 be given. Then, f (x) = x/a on 0 < x < a, and f (x) = (10 − x)/(10 − a).
Answer all the questions in Problem 1.
Problem 3. Previous Exam Problems. Note that these are Final Exams!

a. Fall 2013, Final Exam, Problem 16;

b. Spring 2013, Final Exam, Problem 15;

c. Fall 2012, Final Exam, Problem 11, 17;

d. Spring 2012, Final Exam, Problem 14.

A collection of previous exams could be found at the coordinator’s web:

http://www.math.psu.edu/tseng/class/M251samples.html

220
Chapter 13

Answers to Homework Problems

13.1 Answer/keys for homework 1

Problem 1: (a) 2nd order linear, (b) 4th order linear, (c) 2nd order nonlinear
Problem 5:
(a). 3y 2 − 2x3 = c; y 6= 0;
(b). 2 tan 2y −2x−sin 2x = c if cos 2y 6= 0, i.e., y 6= ±(2n+1)π/4 for any integer n; everywhere
(c). y = sin[ln |x| + c] if x 6= 0 and |y| < 1; also y = ±1.
Problem p 6:
(a) y = − 2x − 2x2 + 4, −1 < x < 2;
(b) y = [2(1 − x)ex − 1]1/2 , −1.68 < x < 0.77 approximately
3 1p
(c) y = − + 65 − 8ex − 8e−x , |x| < 2.08 approximately
4 4
(d) y = [ 32 (arcsin x)2 + 1]1/3 , −1 < x < 1;
√
(e) y 3 − 4y − x3 = −1, |x3 − 1| < 16/3 3 or −1.28 < x < 1.60;

221
13.2 Answer/keys to homework 2
Problem 1:
1 9
a. y = et − e−3t
4 4
b. y = te2t + 4e2t
et e
c. y = −
t t
sin(t2 ) c
d. y = +
2t t
e. y = 4 sin 4t − 3 cos 4t + ce3t
2
f. z = t2 + 1 + cet
g. y = 1 − e− sin t
1 3
h. y = −t − − t−2
2 2
1 bt
i. y = e + ce−at
a+b
1
j. y = + (4a − 2)t−2 .
t
Problem 3.
a. 0 < t < 3
b. 0 < t < 4
c. π/2 < t < 3π/2
d. −∞ < t < −2
e. −2 < t < 2
f. 1 < t < π

Problem 5.
8
y = (10 − t) − (10 − t)4 , for 0 ≤ t ≤ 10. After 10 min, there is no salf in the tank.
10000
Problem 6.
(a) 10 min; (b) ≈ 533.33g.

222
13.3 Answer/keys to homework 3
Problem 2.
a. y = 0 (unstable), y = 150 (asymptotically stable),
b. y = 0 (unstable), y = 30 (asymptotically stable)
c. y = 1 (unstable), y = 3 (asymptotically stable)
d. y = 0 (asymptotically stable), y = 40 (unstable)
e. y = 0 (semi-stable), y = 1 (unstable)
Problem 3. If f ′ (y1 ) < 0 and f (y1 ) = 0, then in a small neighborhood around y = y1 ,
we have y ′ = f (y) < 0 for y > y1 and y ′ = f (y) > 0 for y < y1 . Therefore nearby solutions
approach y = y1 , therefore it is asymptotically stable. On the other hand, if f ′ (y1 ) >, then
y ′ = f (y) > 0 for y > y1 and y ′ = f (y) < 0 for y < y1 , and nearby solutions go away from
y = y1 , therefore it is unstable.
Problem 5.
a. 2x2 y − 3y 2 + 1 = 0
b. x−2 − 2y −3 + x3 y −2 − x−1 y 2 = 3
c. cos(2x) sin y + e2x − y −2 = 2 − 4/π 2
Problem 6. Since My (x) = 0 and Nx (y) = 0, the equation is exact.

223
13.4 Answer/keys to homework 4
Problem 1.

a. y = c1 et + c2 e−3t

b. y = c1 et/2 + c2 e−t/3

c. y = c1 + c2 e−5t
√
r1 t r2 t 9±3 5
d. y = c1 e + c2 e where r1,2 =
2
Problem 2.

a. y(t) = 3e2t − e−3t

b. y(t) = 3e−t + et/2

c. y(t) = −3 + 4e−4t

d. y(t) = 3e−t/3 − et/2

Problem 3.

a. y ′′ − y ′ = 0

b. y ′′ − 5y ′ + 6y = 0

Problem 4. a = −2.
7
Problem 5. (a). − et/2 , (b). 1, (c). e−4t , (d). −e2t
2
Problem 6. (a) 0 < t < ∞, (b) −∞ < t < 1, (c) 0 < t < ∞, (d) 0 < x < 3
Problem 7. (a) 3te2t + ce2t , (b) tet + ct (One can choose c = 0 here.)
Problem 8. They are a fundamental set of solutions if and only if ad − bc 6= 0.

224
13.5 Answer/keys to homework 5
Problem 1.
a. y = c1 e−t/3 + c2 te−t/3
b. y = c1 e−t/2 + c2 e3t/2
c. y = c1 e−3t/2 + c2 te−3t/2
Problem 2.
7
a. y = 2e2t/3 − te2t/3 , y → −∞ as t → ∞
3
b. y = 2te3t , y → ∞ as t → ∞
c. y = 7e−2(t+1) + 5te−2(t+1) , y → 0 as t → ∞
Problem 3. y = 2et/2 + (b − 1)tet/2 ; b=1
Problem 4.
a. y2 (t) = t−2
b. y2 (t) = t−1 ln t
c. y2 (t) = tet
d. y2 (x) = x
Problem 5.
√ √
a. y = c1 et cos 5t + c2 et sin 5t
b. y = c1 e−t cos t + c2 e−t sin t
c. y = c1 e−2t cos(3t/2) + c2 e−2t sin(3t/2)
Problem 6.
a. y = e−2t cos t + 2e−2t sin t; decaying oscillation
b. y = −et−π/2 sin(2t); growing oscillation
5
c. y = 3e−t/2 cos t + e−t/2 sin t; decaying oscillation
2
Problem 7.
√ a+2 √
a. y = 2e−t cos 5t + √ e−t sin 5t
5
b. a ≈ 1.50878
" √ #
1 2 5
c. t = √ π − arctan
5 2+a
√
d. π/ 5

225
13.6 Answer/keys to homework 6
Problem 1.

a. y = c1 e3t + c2 e−t − e2t

3 12
b. y = c1 e−t cos 2t + c2 e−t sin 2t + sin 2t − cos 2t
17 17
1 2
c. y = c1 cos 3t + c2 sin 3t + (9t2 − 6t + 1)e3t +
162 3
Problem 2.
1 1
a. y = et − e−2t − t −
2 2
7 19 1 1 3
b. y = sin 2t − cos 2t + t2 − + et
10 40 4 8 5
1 3
c. y = 2 cos 2t − sin 2t − t cos 2t
8 4
Problem 4.

a. Y = t(A4 t4 + A3 t3 + a2 t2 + A1 t + a0 ) + t(B2 t2 + B1 t + B0 )e−3t + D sin 3t + E cos 3t

b. Y = et (A cos 2t + B sin 2t) + (D1 t + D0 )e2t sin t + (E1 t + E0 )e2t cos t.

c. Y = (A2 t2 + A1 t + A0 ) + t2 (B1 t + B0 )e2t + (D1 t + D0 ) sin 2t + (E1 t + E0 ) cos 2t

d. Y = (A2 t2 + A1 t + A0 )et sin 2t + (B2 t2 + B1 t + B0 )et cos 2t + e−t (D cos t + E sin t) + F et

Problem 5.

a. u = 5 cos(2t − δ), δ = arctan(4/3) ≈ 0.9273

b. u = 2 cos(t − 2π/3)
√
c. u = 2 5 cos(3t − δ), δ = − arctan(1/2) ≈ −0.4636
√
d. u = 13 cos(πt − δ), δ = π + arctan(3/2) ≈ 4.1244
1 √ 1 √ √ π
Problem 6. u = √ sin(8 2t) − cos(8 2t) ft, ω = 8 2 rad/s, T = √ s, R =
4 2 12 4 2

11/288 ≈ 0.1954 ft, δ = π − arctan √32 ≈ 2.0113

p

Problem 7. γ = 8 lb · s/ft √
Problem 8. k = 6, ν = ±2 5.

226
13.7 Answer/keys to homework 7
Problem 1. a. −2 sin(8t) sin t, b. 2 sin(t/2) cos(13t/2), c. 2 cos(3πt/2) cos(πt/2)
Problem 2. a. −∞ < t < ∞, b. t > 0 or t < 0, c. t > 1, or 0 < t < 1, or
t < 0.
Problem 3.

a. y = c1 et + c2 tet + c3 e−t

b. y = c1 et + c2 tet + c3 t2 et

c. y = c1 + c2 t + c3 e2t + c4 e−t + c5 cos t + c6 sin t

d. y = (c1 + c2 t) cos t + (c3 + c4 t) sin t

227
13.8 Answer/keys to homework 8
Problem 1.
s b
a. F (s) = , s > |b|; b. F (s) = 2 , s > |b|;
s 2 − b2 s − b2
s−a b
c. F (s) = 2 2
, s > a + |b|; d. F (s) = , s > a + |b|;
(s − a) − b (s − a)2 − b2
Problem 2.
b s−a
F (s) = (s > a) and G(s) = (s > a).
(s − a)2 + b2 (s − a)2 + b2
Problem 3.
1 2as
a. F (s) = 2
, s > a; b. F (s) = 2 , s > 0;
(s − a) (s + a2 )2
s 2 + a2
c. F (s) = , s > |a|.
(s − a)2 (s + a)2
Problem 4.
n! 2b(s − a) (s − a)2 − b2
(b). F (s) = , G(s) = , H(s) = .
(s − a)n+1 [(s − a)2 + b2 ]2 [(s − a)2 + b2 ]2
Problem 5.
3
a. f (t) = sin 2t; b. f (t) = 2t2 et ;
2
2 2 9 6
c. f (t) = et − e−4t ; d. f (t) = e3t + e−2t ;
3 5 5 5
−t
e. f (t) = 2e cos 2t; f. f (t) = 3 − 2 sin 2t + 5 cos 2t;
5
g. f (t) = 2e−t cos 3t − e−t sin 3t.
3
Problem 6.
1
a. y(t) = (e3t + 4e−2t ); b. y(t) = et sin t;
5
2
c. y(t) = e2t − te2t ; d. y(t) = tet − t2 et + t3 et ;
3
1 1
e. y(t) = cosh t = (et + e−t ); f. y(t) = 2 [(w2 − 5) cos wt + cos 2t].
2 w −4

228
13.9 Answer/keys to homework 9
Problem 1.
a. f (t) = −2u3 (t) + 4u5 (t) − u7 (t)
b. f (t) = 1 + u2 (t)[e−(t−2) − 1]
Problem 2.
a. F (s) = 2e−s /s3
b. F (s) = e−s (s2 + 2)/s3
e−πs e−2πs
c. F (s) = 2 − (1 + πs)
s s2
1 −s
d. F (s) = (e + 2e−3s − 6e−4s )
s
e. F (s) = s−2 [(1 − s)e−2s − (1 + s)e−3s ]
f. F (s) = (1 − e−s )/s2
Problem 3.
1
a. f (t) = u2 (t)[et−2 − e−2(t−2) ]
3
b. f (t) = 2u2 (t)et−2 cos(t − 2)
c. f (t) = u2 (t) sinh 2(t − 2)
d. f (t) = u1 (t)e2(t−1) cosh(t − 1)
e. f (t) = u1 (t) + u2 (t) − u3 (t) − u4 (t)
1 1
Problem 4. (d). f1 (t) = 2(2t)n , f2 (t) = e−t/2 cos t, and f3 (t) = et/2 u2 (t/2).
2 2
Problem 5.
a. y(t) = 1 − cos t + sin t − u3π (t)(1 + cos t)
1 1
b. y(t) = (2 sin t − sin 2t) − uπ (t)(2 sin t + sin 2t)
6 6
1 1
c. y(t) = e−t − e−2t + u2 (t)[ − e−(t−2) + e−2(t−2) ]
2 2
1 1 1
d. y(t) = sin t + t − u6 (t)[t − 6 − sin(t − 6)]
2 2 2
Problem 6.
a. y(t) = e−t cos t + e−t sin t + uπ (t)e−(t−π) sin(t − π)
1 1
b. y(t) = uπ (t) sin 2(t − π) − u2π (t) sin 2(t − 2π)
2 2
1 −2t 1 −t 1 1
c. y(t) = − e + e + u5 (t)[−e−2(t−5) + e−(t−5) ] + u10 (t)[ + e−2(t−10) − e−(t−10) ]
2 2 2 2
d. y(t) = sin t + u2π (t) sin(t − 2π)

229
13.10 Answer/keys to homework 10
Problem 1.
a. x1 = u, x2 = u′ , → x′1 = x2 , x′2 = −2x1 − 0.5x2
′
b. x1 = u, x2 = u , → x′1 = x2 , x′2 = −(1 − 0.25t−2 )x1 − t−1 x2
c. x1 = u, x2 = u′ , x3 = u′′ , x4 = u′′′ → x′1 = x2 , x′2 = x3 , x′3 = x4 , x′4 =
x1
d. x1 = u, x2 = u′ , → x′1 = x2 , x′2 = −4x1 − 0.25x2 + 2 cos 3t, x1 (0) =
1, x2 (0) = −2
e. x1 = u, x2 = u′ , → x′1 = x2 , x′2 = −q(t)x1 − p(t)x2 + g(t), x1 (0) =
u0 , x2 (0) = u′0
Problem 2.    
1 1
(a). λ1 = 2, ~v1 = , λ2 = 4, ~v2 =
3 1
   
1 1
(b). λ1 = −3, ~v1 = , λ2 = −1, ~v2 =
−1 1
   
3 1
(c). λ1 = −1/2, ~v1 = , λ2 = −3/2, ~v2 =
10 2
   
1 1
(d). λ1 = −3, ~v1 = , λ2 = 1, ~v2 =
0 1
Problem 3. Use the eigenpairs found in Problem 2, and the formula: (a). ~x = c1 eλ1 t~v1 +
λ2 t
c2 e ~v2 , one can simply write out the general solutions.
Problem 4.    
3 −2t 1
(a). ~x = c1 + c2 e
4 2
   
−2 t −3
(b). ~x = c1 + c2 e
1 1
Problem 5.    
−t −1 −2t 1
(a).(i). ~x = c1 e + c2 e , (ii). c1 = −1/4, c2 = 7/4
2 2
   
−1 1
(b).(i). ~x = c1 et + c2 e−2t , (ii). c1 = −1/4, c2 = 7/4
2 2

230
13.11 Answer/keys to homework 11
Problem 1.    
t cos 2t t sin 2t
a. ~x = c1 e + c2 e
cos 2t + sin 2t − cos 2t + sin 2t
   
−t 2 cos 2t −t −2 sin 2t
b. ~x = c1 e + c2 e
sin 2t cos 2t
   
5 cos t 5 sin t
c. ~x = c1 + c2
2 cos t + sin t − cos t + 2 sin t
Problem 2.  
cos t − 3 sin t
(a). ~x = e−t
cos t − sin t
 
−2t cos t − 5 sin t
(b). ~x = e
−2 cos t − 3 sin t
Problem 3.       
t 2 t 2 t 1
(a). ~x = c1 e + c2 te +e
1 1 0
      
1 1 0
(b). ~x = c1 + c2 t −
2 2 0.5
Problem 4.  
−3t 3 + 4t
(a). ~x = e
2 + 4t
   
1 3
(b). ~x = 2 + 14t
2 −1
Problem 5.
(a). (−0.5, 1), saddle point, unstable;
(0, 0), proper node, unstable
(b). (0, 0), node, unstable;
(2, 0), node, asymptotically stable;
(0, 1.5), saddle point, unstable;
(−1, 3), node, asymptotically stable
(c). (0,√0), spiral
√ point, asymptotically stable;
(1 − √2, 1 + √2), saddle point, unstable;
(1 + 2, 1 − 2), saddle point, unstable

231
13.12 Answer/keys to homework 12
Problem 1.
∞
2L X (−1)n nπx
a. f (x) = sin
π n L
n=1
∞
1 2X 1
b. f (x) = − sin[(2n − 1)πx/L]
2 π 2n − 1
n=1
∞ 
(−1)n+1

π X 2
c. f (x) = − + cos(2n − 1)x + sin nx
4 n=1 π(2n − 1)2 n
∞  
X 2 nπ 4 nπ nπx
d. f (x) = − cos + 2
sin sin
nπ 2 (nπ) 2 2
n=1
∞
4 X sin[(2n − 1)πx/2]
e. f (x) =
π 2n − 1
n=1 Z
x
Problem 2. (c). f (t) dt may not be periodic; for example, consider f (t) = 1 + sin t.
0
Problem 5.
a. y = − sin√x √ √ √
b. y = (cot 2π cos 2x + sin 2x)/ 2;
c. y = c2 sin x if cos L = 0, and y = 0 for all other L;
d. No solution
1
e. y = c1 cos 2x + cos x.
3

232
13.13 Answer/keys to homework 13
Problem 1.
a. Let u(x, t) = F (x)G(t), we have xF ′′ (x) − λF (x) = 0, G′ (t) + λG(t) = 0
b. Let u(x, t) = F (x)G(t), we have F ′′ (x) − λxF (x) = 0, G′ (t) + λtG(t) = 0
c. Let u(x, t) = F (x)G(t), we have F ′′ (x) − λ(F ′ (x) + F (x)) = 0, G′ (t) + λG(t) = 0
d. Let u(x, t) = F (x)G(t), we have [p(x)F ′ (x)]′ + λr(x)F (x) = 0, G′′ (t) + λG(t) = 0
e. Not separable.
f. Let u(x, t) = F (x)G(y), we have F ′′ (x) + (x + λ)F (x) = 0, G′′ (y) − λG(y) = 0
Problem 2.
2 2
u(x, t) = e−400π t sin 2πx − e−2500π t sin 5πx
Problem 3. ∞
100 X 1 − cos(nπ) −n2 π2 t/1600 nπx
u(x, t) = e sin
π n 40
n=1

Problem 4*.

X ′′ + µ2 X = 0, Y ′′ + (λ2 − µ2 )Y = 0, T ′ + α2 λ2 T = 0

Problem 5*.

r 2 R′′ + rR′ + (λ2 r 2 − µ2 )R = 0, Θ′′ + µ2 Θ = 0, T ′ + α2 λ2 T = 0.

Problem 6.
3
(a). U (x) = 10 + x (b). U (x) = T
5
T (1 + x) T (1 + L − x)
(c). U (x) = (d). U (x) =
1+L 1+L
Problem 7. The IVBP for the heat equation is:

ut = c2 uxx , u(0, t) = 0, u(20, t) = 60, u(x, 0) = 25.

Note that this has non-homogeneous B.C.s! Carrying out the computation, we get
∞ 20
nπx 1 nπx 1
Z
−( cnπ )2 t
X
u(x, t) = 3x+ Cn e 20 sin , Cn = (25−3x) sin dx = (70(−1)n +50)
20 10 0 20 nπ
n=1

233