You are on page 1of 174

Dierential Equations I

MATB44H3F
Version September 15, 2011-1949
ii
Contents
1 Introduction 1
1.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Sample Application of Dierential Equations . . . . . . . . . . . 2
2 First Order Ordinary Dierential Equations 5
2.1 Separable Equations . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Exact Dierential Equations . . . . . . . . . . . . . . . . . . . . . 7
2.3 Integrating Factors . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4 Linear First Order Equations . . . . . . . . . . . . . . . . . . . . 14
2.5 Substitutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.5.1 Bernoulli Equation . . . . . . . . . . . . . . . . . . . . . . 17
2.5.2 Homogeneous Equations . . . . . . . . . . . . . . . . . . . 19
2.5.3 Substitution to Reduce Second Order Equations to First
Order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3 Applications and Examples of First Order odes 25
3.1 Orthogonal Trajectories . . . . . . . . . . . . . . . . . . . . . . . 25
3.2 Exponential Growth and Decay . . . . . . . . . . . . . . . . . . . 27
3.3 Population Growth . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.4 Predator-Prey Models . . . . . . . . . . . . . . . . . . . . . . . . 29
3.5 Newtons Law of Cooling . . . . . . . . . . . . . . . . . . . . . . 30
3.6 Water Tanks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.7 Motion of Objects Falling Under Gravity with Air Resistance . . 34
3.8 Escape Velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.9 Planetary Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.10 Particle Moving on a Curve . . . . . . . . . . . . . . . . . . . . . 39
iii
iv CONTENTS
4 Linear Dierential Equations 45
4.1 Homogeneous Linear Equations . . . . . . . . . . . . . . . . . . . 47
4.1.1 Linear Dierential Equations with Constant Coecients . 52
4.2 Nonhomogeneous Linear Equations . . . . . . . . . . . . . . . . . 54
5 Second Order Linear Equations 57
5.1 Reduction of Order . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.2 Undetermined Coecients . . . . . . . . . . . . . . . . . . . . . . 60
5.2.1 Shortcuts for Undetermined Coecients . . . . . . . . . . 64
5.3 Variation of Parameters . . . . . . . . . . . . . . . . . . . . . . . 66
6 Applications of Second Order Dierential Equations 71
6.1 Motion of Object Hanging from a Spring . . . . . . . . . . . . . . 71
6.2 Electrical Circuits . . . . . . . . . . . . . . . . . . . . . . . . . . 75
7 Higher Order Linear Dierential Equations 79
7.1 Undetermined Coecients . . . . . . . . . . . . . . . . . . . . . . 79
7.2 Variation of Parameters . . . . . . . . . . . . . . . . . . . . . . . 80
7.3 Substitutions: Eulers Equation . . . . . . . . . . . . . . . . . . . 82
8 Power Series Solutions to Linear Dierential Equations 85
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
8.2 Background Knowledge Concerning Power Series . . . . . . . . . 88
8.3 Analytic Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 89
8.4 Power Series Solutions: Levels of Success . . . . . . . . . . . . . . 91
8.5 Level 1: Finding a nite number of coecients . . . . . . . . . . 91
8.6 Level 2: Finding the recursion relation . . . . . . . . . . . . . . . 94
8.7 Solutions Near a Singular Point . . . . . . . . . . . . . . . . . . . 97
8.8 Functions Dened via Dierential Equations . . . . . . . . . . . . 111
8.8.1 Chebyshev Equation . . . . . . . . . . . . . . . . . . . . . 111
8.8.2 Legendre Equation . . . . . . . . . . . . . . . . . . . . . . 113
8.8.3 Airy Equation . . . . . . . . . . . . . . . . . . . . . . . . 115
8.8.4 Laguerres Equation . . . . . . . . . . . . . . . . . . . . . 115
8.8.5 Bessel Equation . . . . . . . . . . . . . . . . . . . . . . . . 116
9 Linear Systems 121
9.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
9.2 Computing e
T
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
9.3 The 2 2 Case in Detail . . . . . . . . . . . . . . . . . . . . . . . 129
9.4 The Non-Homogeneous Case . . . . . . . . . . . . . . . . . . . . 133
CONTENTS v
9.5 Phase Portraits . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
9.5.1 Real Distinct Eigenvalues . . . . . . . . . . . . . . . . . . 137
9.5.2 Complex Eigenvalues . . . . . . . . . . . . . . . . . . . . . 139
9.5.3 Repeated Real Roots . . . . . . . . . . . . . . . . . . . . . 141
10 Existence and Uniqueness Theorems 145
10.1 Picards Method . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
10.2 Existence and Uniqueness Theorem for First Order ODEs . . . . 150
10.3 Existence and Uniqueness Theorem for Linear First Order ODEs 155
10.4 Existence and Uniqueness Theorem for Linear Systems . . . . . . 156
11 Numerical Approximations 163
11.1 Eulers Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
11.1.1 Error Bounds . . . . . . . . . . . . . . . . . . . . . . . . . 165
11.2 Improved Eulers Method . . . . . . . . . . . . . . . . . . . . . . 166
11.3 Runge-Kutta Methods . . . . . . . . . . . . . . . . . . . . . . . . 167
vi CONTENTS
Chapter 1
Introduction
1.1 Preliminaries
Denition (Dierential equation)
A dierential equation (de) is an equation involving a function and its deriva-
tives.
Dierential equations are called partial dierential equations (pde) or or-
dinary dierential equations (ode) according to whether or not they contain
partial derivatives. The order of a dierential equation is the highest order
derivative occurring. A solution (or particular solution) of a dierential equa-
tion of order n consists of a function dened and n times dierentiable on a
domain D having the property that the functional equation obtained by substi-
tuting the function and its n derivatives into the dierential equation holds for
every point in D.
Example 1.1. An example of a dierential equation of order 4, 2, and 1 is
given respectively by
_
dy
dx
_
3
+
d
4
y
dx
4
+y = 2 sin(x) + cos
3
(x),

2
z
x
2
+

2
z
y
2
= 0,
yy

= 1.
1
2 CHAPTER 1. INTRODUCTION
Example 1.2. The function y = sin(x) is a solution of
_
dy
dx
_
3
+
d
4
y
dx
4
+y = 2 sin(x) + cos
3
(x)
on domain R; the function z = e
x
cos(y) is a solution of

2
z
x
2
+

2
z
y
2
= 0
on domain R
2
; the function y = 2

x is a solution of
yy

= 2
on domain (0, ).
Although it is possible for a de to have a unique solution, e.g., y = 0 is the
solution to (y

)
2
+ y
2
= 0, or no solution at all, e.g., (y

)
2
+ y
2
= 1 has no
solution, most des have innitely many solutions.
Example 1.3. The function y =

4x +C on domain (C/4, ) is a solution
of yy

= 2 for any constant C.


Note that dierent solutions can have dierent domains. The set of all
solutions to a de is call its general solution.
1.2 Sample Application of Dierential Equations
A typical application of dierential equations proceeds along these lines:
Real World Situation

Mathematical Model

Solution of Mathematical Model

Interpretation of Solution
1.2. SAMPLE APPLICATION OF DIFFERENTIAL EQUATIONS 3
Sometimes in attempting to solve a de, we might perform an irreversible
step. This might introduce extra solutions. If we can get a short list which
contains all solutions, we can then test out each one and throw out the invalid
ones. The ultimate test is this: does it satisfy the equation?
Here is a sample application of dierential equations.
Example 1.4. The half-life of radium is 1600 years, i.e., it takes 1600 years for
half of any quantity to decay. If a sample initially contains 50 g, how long will
it be until it contains 45 g?
Solution. Let x(t) be the amount of radium present at time t in years. The rate
at which the sample decays is proportional to the size of the sample at that
time. Therefore we know that dx/dt = kx. This dierential equation is our
mathematical model. Using techniques we will study in this course (see 3.2,
Chapter 3), we will discover that the general solution of this equation is given
by the equation x = Ae
kt
, for some constant A. We are told that x = 50 when
t = 0 and so substituting gives A = 50. Thus x = 50e
kt
. Solving for t gives
t = ln(x/50) /k. With x(1600) = 25, we have 25 = 50e
1600k
. Therefore,
1600k = ln
_
1
2
_
= ln(2) ,
giving us k = ln(2) /1600. When x = 45, we have
t =
ln(x/50)
k
=
ln(45/50)
ln(2) /1600
= 1600
ln(8/10)
ln(2)
= 1600
ln(10/8)
ln(2)
1600
0.105
0.693
1600 0.152 243.2.
Therefore, it will be approximately 243.2 years until the sample contains 45 g
of radium.
Additional conditions required of the solution (x(0) = 50 in the above ex-
ample) are called boundary conditions and a dierential equation together with
boundary conditions is called a boundary-value problem (BVP). Boundary con-
ditions come in many forms. For example, y(6) = y(22); y

(7) = 3y(0); y(9) = 5


are all examples of boundary conditions. Boundary-value problems, like the one
in the example, where the boundary condition consists of specifying the value
of the solution at some point are also called initial-value problems (IVP).
Example 1.5. An analogy from algebra is the equation
y =

y + 2. (1.1)
4 CHAPTER 1. INTRODUCTION
To solve for y, we proceed as
y 2 =

y,
(y 2)
2
= y, (irreversible step)
y
2
4y + 4 = y,
y
2
5y + 4 = 0,
(y 1) (y 4) = 0.
Thus, the set y 1, 4 contains all the solutions. We quickly see that y = 4
satises Equation (1.1) because
4 =

4 + 2 = 4 = 2 + 2 = 4 = 4,
while y = 1 does not because
1 =

1 + 2 = 1 = 3.
So we accept y = 4 and reject y = 1.
Chapter 2
First Order Ordinary
Dierential Equations
The complexity of solving des increases with the order. We begin with rst
order des.
2.1 Separable Equations
A rst order ode has the form F(x, y, y

) = 0. In theory, at least, the methods


of algebra can be used to write it in the form

= G(x, y). If G(x, y) can


be factored to give G(x, y) = M(x) N(y),then the equation is called separable.
To solve the separable equation y

= M(x) N(y), we rewrite it in the form


f(y)y

= g(x). Integrating both sides gives


_
f(y)y

dx =
_
g(x) dx,
_
f(y) dy =
_
f(y)
dy
dx
dx.
Example 2.1. Solve 2xy + 6x +
_
x
2
4
_
y

= 0.

We use the notation dy/dx = G(x, y) and dy = G(x, y) dx interchangeably.


5
6CHAPTER 2. FIRST ORDER ORDINARY DIFFERENTIAL EQUATIONS
Solution. Rearranging, we have
_
x
2
4
_
y

= 2xy 6x,
= 2xy 6x,
y

y + 3
=
2x
x
2
4
, x ,= 2
ln([y + 3[) = ln
_

x
2
4

_
+C,
ln([y + 3[) + ln
_

x
2
4

_
= C,
where C is an arbitrary constant. Then

(y + 3)
_
x
2
4
_

= A,
(y + 3)
_
x
2
4
_
= A,
y + 3 =
A
x
2
4
,
where A is a constant (equal to e
C
) and x ,= 2. Also y = 3 is a solution
(corresponding to A = 0) and the domain for that solution is R.
Example 2.2. Solve the ivp sin(x) dx +y dy = 0, where y(0) = 1.
Solution. Note: sin(x) dx +y dy = 0 is an alternate notation meaning the same
as sin(x) +y dy/dx = 0.
We have
y dy = sin(x) dx,
_
y dy =
_
sin(x) dx,
y
2
2
= cos(x) +C
1
,
y =
_
2 cos(x) +C
2
,
where C
1
is an arbitrary constant and C
2
= 2C
1
. Considering y(0) = 1, we have
1 =
_
2 +C
2
= 1 = 2 +C
2
= C
2
= 1.
Therefore, y =
_
2 cos(x) 1 on the domain (/3, /3), since we need cos(x)
1/2 and cos(/3) = 1/2.
2.2. EXACT DIFFERENTIAL EQUATIONS 7
An alternate method to solving the problem is
y dy = sin(x) dx,
_
y
1
y dy =
_
x
0
sin(x) dx,
y
2
2

1
2
2
= cos(x) cos(0),
y
2
2

1
2
= cos(x) 1,
y
2
2
= cos(x)
1
2
,
y =
_
2 cos(x) 1,
giving us the same result as with the rst method.
Example 2.3. Solve y
4
y

+y

+x
2
+ 1 = 0.
Solution. We have
_
y
4
+ 1
_
y

= x
2
1,
y
5
5
+y =
x
3
3
x +C,
where C is an arbitrary constant. This is an implicit solution which we cannot
easily solve explicitly for y in terms of x.
2.2 Exact Dierential Equations
Using algebra, any rst order equation can be written in the form F(x, y) dx +
G(x, y) dy = 0 for some functions F(x, y), G(x, y).
Denition
An expression of the form F(x, y) dx +G(x, y) dy is called a (rst-order) dier-
ential form. A dierentical form F(x, y) dx + G(x, y) dy is called exact if there
exists a function g(x, y) such that dg = F dx +Gdy.
If = F dx+Gdy is an exact dierential form, then = 0 is called an exact
dierential equation. Its solution is g = C, where = dg.
Recall the following useful theorem from MATB42:
8CHAPTER 2. FIRST ORDER ORDINARY DIFFERENTIAL EQUATIONS
Theorem 2.4
If F and G are functions that are continuously dierentiable throughout a
simply connected region, then F dx +Gdy is exact if and only if G/x =
F/y.
Proof. Proof is given in MATB42.
Example 2.5. Consider
_
3x
2
y
2
+x
2
_
dx +
_
2x
3
y +y
2
_
dy = 0. Let
=
_
3x
2
y
2
+x
2
_
. .
F
dx +
_
2x
3
y +y
2
_
. .
G
dy
Then note that
G
x
= 6x
2
y =
F
y
.
By Theorem 2.4, = dg for some g. To nd g, we know that
g
x
= 3x
2
y
2
+x
2
, (2.1a)
g
y
= 2x
3
y +y
2
. (2.1b)
Integrating Equation (2.1a) with respect to x gives us
g = x
3
y
2
+
x
3
3
+h(y). (2.2)
So dierentiating that with respect to y gives us
Eq. (2.1b)
..
g
y
= 2x
3
y +
dh
dy
,
2x
3
y +y
2
= 2x
3
y +
dh
dy
,
dh
dy
= y
2
,
h(y) =
y
3
3
+C
2.2. EXACT DIFFERENTIAL EQUATIONS 9
for some arbitrary constant C. Therefore, Equation (2.2) becomes
g = x
3
y
2
+
x
3
3
+
y
3
3
+C.
Note that according to our dierential equation, we have
d
_
x
3
y
2
+
x
3
3
+
y
3
3
+C
_
= 0 which implies x
3
y
2
+
x
3
3
+
y
3
3
+C = C

for some arbitrary constant C

. Letting D = C

C, which is still an arbitrary


constant, the solution is
x
3
y
2
+
x
3
3
+
y
3
3
= D.

Example 2.6. Solve


_
3x
2
+ 2xy
2
_
dx +
_
2x
2
y
_
dy = 0, where y(2) = 3.
Solution. We have
_
_
3x
2
+ 2xy
2
_
dx = x
3
+x
2
y
2
+C
for some arbitrary constant C. Since C is arbitrary, we equivalently have x
3
+
x
2
y
2
= C. With the initial condition in mind, we have
8 + 4 9 = C = C = 44.
Therefore, x
3
+x
2
y
2
= 44 and it follows that
y =

44 x
3
x
2
.
But with the restriction that y(2) = 3, the only solution is
y =

44 x
3
x
2
on the domain
_

44,
3

44
_
0.
Let = F dx + Gdy. Let y = s(x) be the solution of the de = 0, i.e.,
F +Gs

(x) = 0. Let y
0
= s(x
0
) and let be the piece of the graph of y = s(x)
from (x
0
, y
0
) to (x, y). Figure 2.1 shows this idea. Since y = s(x) is a solution
to = 0, we must have = 0 along . Therefore,
_

= 0. This can be seen


10CHAPTER 2. FIRST ORDER ORDINARY DIFFERENTIAL EQUATIONS
(x
0
,y)

1
(x
0
,y
0
)
(x,y)

y=s(x)
y
x
Figure 2.1: The graph of y = s(x) with connecting (x
0
, y
0
) to (x, y).
by parameterizing by (x) = (x, s(x)), thereby giving us
_

=
_
x
x0
F dx +Gs

(x) dx =
_
x
x0
0 dx = 0.
This much holds for any .
Now suppose that is exact. Then the integral is independent of the path.
Therefore
0 =
_

=
_
1
F dx +Gdy +
_
2
F dx +Gdy
=
_
y
y0
G(x
0
, y) dy +
_
x
x0
F(x, y) dx.
We can now solve Example 2.6 with this new method.
Solution (Alternate solution to Example 2.6). We simply have
0 =
_
4
3
2 2
2
y dy +
_
x
2
_
3x
2
+ 2xy
2
_
dx
= 4y
2
4 (3)
2
+x
3
+x
2
y
2
2
3
2
2
y
2
= 4y
2
36 +x
3
+x
2
y
2
8 4y
2
,
nally giving us x
3
+x
2
y
2
= 44, which agrees with our previous answer.
Remark. Separable equations are actually a special case of exact equations, that
is,
f(y)y

= g(x) = g(x) dx +f(y) dy = 0 =



x
f(y) = 0 =

y
(g(x)) .
2.3. INTEGRATING FACTORS 11
So the equation is exact.
2.3 Integrating Factors
Consider the equation = 0. Even if is not exact, there may be a function
I(x, y) such that I is exact. So = 0 can be solved by multiplying both sides
by I. The function I is called an integrating factor for the equation = 0.
Example 2.7. Solve y/x
2
+ 1 +y

/x = 0.
Solution. We have
_
y
x
2
+ 1
_
dx +
1
x
dy = 0.
We see that
_

x
_
1
x
_
=
1
x
2
_
,=
_
1
x
2
=

y
_
y
x
2
+ 1
_
_
.
So the equation is not exact. Multiplying by x
2
gives us
_
y +x
2
_
dx +xdy = 0,
d
_
xy +
x
3
3
_
= 0,
xy +
x
3
3
= C
for some arbitrary constant C. Solving for y nally gives us
y =
C
x

x
3
3
.

There is, in general, no algorithm for nding integrating factors. But the
following may suggest where to look. It is important to be able to recognize
common exact forms:
xdy +y dx = d(xy) ,
xdy y dx
x
2
= d
_
y
x
_
,
xdx +y dy
x
2
+y
2
= d
_
ln
_
x
2
+y
2
_
2
_
,
xdy d dx
x
2
+y
2
= d
_
tan
1
_
y
x
__
,
x
a1
y
b1
(ay dx +bxdy) = d
_
x
a
y
b
_
.
12CHAPTER 2. FIRST ORDER ORDINARY DIFFERENTIAL EQUATIONS
Example 2.8. Solve
_
x
2
y
2
+y
_
dx +
_
2x
3
y x
_
dy = 0.
Solution. Expanding, we have
x
2
y
2
dx + 2x
3
y dy +y dx xdy = 0.
Here, a = 1 and b = 2. Thus, we wish to use
d
_
xy
2
_
= y
2
dx + 2xy dy.
This suggests dividing the original equation by x
2
which gives
y
2
dx + 2xy dy +
y dx xdy
x
2
= 0.
Therefore,
xy
2
+
y
x
= C, x ,= 0,
where C is an arbitrary constant. Additionally, y = 0 on the domain R is a
solution to the original equation.
Example 2.9. Solve y dx xdy
_
x
2
+y
2
_
dx = 0.
Solution. We have
y dx xdy
x
2
+y
2
dx = 0,
unless x = 0 and y = 0. Now, it follows that
tan
1
_
y
x
_
x = C,
tan
1
_
y
x
_
= C x,
tan
1
_
y
x
_
= D x, (D = C)
y
x
= tan(D x) ,
y = xtan(D x) ,
where C is an arbitrary constant and the domain is
D x ,= (2n + 1)

2
, x ,= (2n + 1)

2
for any integer n. Also, since the derivation of the solution is based on the
assumption that x ,= 0, it is unclear whether or not 0 should be in the domain,
i.e., does y = xtan(D x) satisfy the equation when x = 0? We have y xy

2.3. INTEGRATING FACTORS 13


_
x
2
+y
2
_
= 0. If x = 0 and y = xtan(D x), then y = 0 and the equation is
satised. Thus, 0 is in the domain.
Proposition 2.10
Let = dg. Then for any function P : R R, P(g) is exact.
Proof. Let Q =
_
P(t) dy. Then d(Q(g)) = P(g) dg = P(g).
To make use of Proposition 2.10, we can group together some terms of
to get an expression you recognize as having an integrating factor and multiply
the equation by that. The equation will now look like dg + h = 0. If we can
nd an integrating factor for h, it will not necessarily help, since multiplying by
it might mess up the part that is already exact. But if we can nd one of the
form P(g), then it will work.
Example 2.11. Solve
_
x yx
2
_
dy +y dx = 0.
Solution. Expanding, we have
y dx +xdy
. .
d(xy)
yx
2
dy = 0.
Therefore, we can multiply teh equation by any function of xy without disturb-
ing the exactness of its rst two terms. Making the last term into a function of
y alone will make it exact. So we multiply by (xy)
2
, giving us
y dx +xdy
x
2
y
2

1
y
dy = 0 =
1
xy
ln([y[) = C,
where C is an arbitrary constant. Note that y = 0 on the domain R is also a
solution.
Given
M dx +N dy = 0, ()
we want to nd I such that IM dx +IN dy is exact. If so, then

x
(IN)
. .
IxN+INx
=

y
(IM)
. .
IyM+IMy
.
14CHAPTER 2. FIRST ORDER ORDINARY DIFFERENTIAL EQUATIONS
If we can nd any particular solution I(x, y) of the pde
I
x
N +IN
x
= I
y
M +IM
y
, ()
then we can use it as an integrating factor to nd the general solution of ().
Unfortunately, () is usually even harder to solve than (), but as we shall see,
there are times when it is easier.
Example 2.12. We could look for an I having only xs and no ys? For exam-
ple, consider I
y
= 0. Then
I
x
N +IN
x
= IM
y
implies
I
x
I
=
M
y
N
x
N
.
This works if (M
y
N
x
) /N happens to be a function of x alone. Then
I = e
R
MyNx
N
dx
.
Similarly, we can also reverse the role of x and y. If (N
x
M
y
) /M happens to
be a function of y alone, then
e
R
NxMy
M
dy
works.
2.4 Linear First Order Equations
A rst order linear equation (n = 1) looks like
y

+P(x)y = Q(x).
An integrating factor can always be found by the following method. Consider
dy +P(x)y dx = Q(x) dx,
(P(x)y Q(x))
. .
M(x,y)
dx + dy
..
N(x,y)
= 0.
We use the de for the integrating factor I(x, y). The equation IM dx + IN dy
is exact if
I
x
N +IN
x
= I
y
M +IM
y
.
2.4. LINEAR FIRST ORDER EQUATIONS 15
In our case,
I
x
+ 0 = I
y
(P(x)y Q(x)) +IP(x). ()
We need only one solution, so we look for one of the form I(x), i.e., with I
y
= 0.
Then () becomes
dI
dx
= IP(x).
This is separable. So
dI
I
= P(x) dx,
ln([I[) =
_
P(x) dx +C,
[I[ = e
R
P(x) dx
, e
x
> 0
I = e
R
P(x) dx
.
We conclude that e
R
P(x) dx
is an integrating factor for y

+P(x)y = Q(x).
Example 2.13. Solve y

(1/x) y = x
3
, where x > 0.
Solution. Here P(x) = 1/x. Then
I = e
R
P(x) dx
= e

R
1
x
dx
= e
ln(|x|)dx
=
1
[x[
=
1
x
,
where x > 0. Our dierential equation is
xdy y dx
x
= x
3
dx.
Multiplying by the integrating factor 1/x gives us
xdy y dx
x
2
= x
2
dx.
Then
y
x
=
x
3
3
+C,
y =
x
3
3
+Cx
on the domain (0, ), where C is an arbitrary constant (x > 0 is given).
16CHAPTER 2. FIRST ORDER ORDINARY DIFFERENTIAL EQUATIONS
In general, given y

+P(x)y = Q(x), multiply by e


R
P(x) dx
to obtain
e
R
P(x) dx
y

+e
R
P(x) dx
. .
d(ye
R
P(x) dx
)/dx
P(x)y = Q(x)e
R
P(x) dx
.
Therefore,
ye
R
P(x) dx
=
_
Q(x)e
R
P(x) dx
dx +C,
y = e

R
P(x) dx
_
Q(x)e
R
P(x) dx
dx +Ce

R
P(x) dx
,
where C is an arbitrary constant.
Example 2.14. Solve xy

+ 2y = 4x
2
.
Solution. What should P(x) be? To nd it, we put the equation in standard
form, giving us
y

+
2
x
y = 4x.
Therefore, P(x) = 2/x. Immediately, we have
I = e
R
(2/x)dx
= e
ln(x
2
)
= x
2
.
Multiplying the equation by x
2
gives us
x
2
y

+ 2xy = 4x
3
,
x
2
y = x
4
+C,
y = x
2
+
C
x
2
,
where C is an arbitrary constant and x ,= 0.
Example 2.15. Solve e
y
dy +dx + 2xdy = 0.
Solution. This equation is linear with x as a function of y. So what we have is
dx
dy
+ 2x = e
y
,
2.5. SUBSTITUTIONS 17
where I = e
R
2 dy
= e
2y
. Therefore,
e
2y
dx
dy
+ 2xe
2y
= e
y
,
xe
2y
= e
y
+C,
where C is an arbitrary constant. We could solve explicitly for y, but it is messy.
The domain is not easy to determine.
2.5 Substitutions
In many cases, equations can be put into one of the standard forms discussed
above (separable, linear, etc.) by a substitution.
Example 2.16. Solve y

2y

= 5.
Solution. This is a rst order linear equation for y

. Let u = y

. Then the
equation becomes
u

2u = 5.
The integration factor is then I = e

R
2 dx
= e
2x
. Thus,
u

e
2x
2ue
2x
= 5e
2x
,
ue
2x
=
5
2
e
2x
+C,
where C is an arbitrary constant. But u = y

, so
y =
5
2
x +
C
2
e
2x
+C
1
=
5
2
x +C
1
e
2x
+C
2
on the domain R, where C
1
and C
2
are arbitrary constants.
We now look at standard substitutions.
2.5.1 Bernoulli Equation
The Bernoulli equation is given by
dy
dx
+P(x)y = Q(x)y
n
.
Let z = y
1n
. Then
dz
dx
= (1 n) y
n
dy
dx
,
18CHAPTER 2. FIRST ORDER ORDINARY DIFFERENTIAL EQUATIONS
giving us
y
n
dy
dx
+P(x)y
1n
= Q(x),
1
1 n
dz
dx
+P(x)z = Q(x),
dz
dx
+ (1 n) P(x)z = (1 n) Q(x),
which is linear in z.
Example 2.17. Solve y

+xy = xy
3
.
Here, we have n = 3. Let z = y
2
. If y ,= 0, then
dz
dx
= 2y
3
dy
dx
.
Therefore, our equation becomes

y
3
z

2
+xy = xy
3
,

2
+xy
2
= x,
z

2xy = 2x.
We can readily see that I = e


2x dx
= e
x

. Thus,
e
x

z

2xe
x

= 2xe
x

,
e
x

z = e
x

+C,
z = 1 +Ce
x

,
where C is an arbitrary constant. But z = y
2
. So
y =
1

1 +Ce
x

.
The domain is 8

:
R, C > 1,
[x[ >
p
ln(C), C 1.
An additional solution is y = 0 on R.
2.5. SUBSTITUTIONS 21
If it is independent of y, namely, F(y

, y

, x) = 0, then it is really just a rst


order equation for y

as we saw in earlier examples.


Consider now the case where it is independent of x, namely, F(y

, y

, y) = 0.
Substitute v = dy/dx for x, i.e., eliminate x between the equations
F
_
d
2
y
dx
2
,
dy
dx
, y
_
= 0
and v = dy/dx. Then
d
2
y
dx
2
=
dv
dx
=
dv
dy
dy
dx
=
dv
dy
v.
Therefore,
F
_
d
2
y
dx
2
,
dy
dx
, y
_
= 0 F
_
dv
dy
v, v, y
_
= 0.
This is a rst order equation in v and y.
Example 2.22. Solve y

= 4 (y

)
3/2
y.
Solution. Let v = dy/dx. Then
d
2
y
dx
2
=
dv
dy
v
and our equation becomes
dv
dy
v = 4v
3/2
y,
dv

v
= 4y dy, v 0,
2

v = 2y
2
+C
1
,

v = y
2
+C
2
,
22CHAPTER 2. FIRST ORDER ORDINARY DIFFERENTIAL EQUATIONS
where C
1
is an arbitrary constant and C
2
= C
1
/2. But v = dy/dx, so we have
dy
dx
=
_
y
2
+C
2
_
2
,
dx =
dy
(y
2
+C
2
)
2
,
x =
_
dy
(y
2
+C
2
)
2
=
_

_
1
2C
3/2
2
_
tan
1
_
y

C2
_
+

C2y
y
2
+C2
_
+C
3
, C
2
> 0,

1
3y
3
+C
3
, C
2
= 0,

1
2(C2)
3/2

y
2
y
2
+C2
+C
3
, C
2
< 0.

Next consider second order linear equations. That is,


P(x)y

+Q(x)y

+R(x)y = 0.
We can eliminate y by letting y = e
v
. Then y

= e
v
v

and y

= e
v
(v

)
2
+e
v
v

.
The equation then becomes
P(x)
_
e
v
(v

)
2
+e
v
v

_
+Q(x)e
v
v

+R(x)e
v
= 0,
which is a rst order equation in v

.
Example 2.23. Solve x
2
y

+
_
x x
2
_
y

e
2x
y = 0.
Solution. Let y = e
v
. Then the equation becomes
x
2
e
v
(v

)
2
+x
2
e
v
v

+
_
x x
2
_
e
v
v

e
2x
e
v
= 0.
Write z = v

. Then
x
2
z

+x
2
z
2
+ (1 x) xz = e
2x
.
Now we are on our ownthere is no standard technique for this case. Suppose
we try u = xy. Then z = u/x and
z

=
u
x
2
+
1
x
u

.
Then it follows that
xu

+u
2
xu = e
2x
.
This is a bit simpler, but it is still nonstandard. We can see that letting u = se
x
2.5. SUBSTITUTIONS 23
will give us some cancellation. Thus, u

= s

e
x
+se
x
and our equation becomes
xs

e
x
+

xse
x
+s
2
e
2x

xse
x
= e
2x
,
xs

+s
2
e
x
= e
x
,
xs

= e
x
_
1 s
2
_
,
s

1 s
2
=
e
x
x
,
1
2
ln
_

1 +s
1 s

_
=
_
e
x
x
dx.
Working our way back through the subsitutions we nd that s = zxe
x
so our
solution becomes
1
2
ln
_

1 +zxe
x
1 zxe
x

_
=
_
e
x
x
dx.
Using algebra, we could solve this equation for z in terms of x and then integrate
the result to get v which then determines y = e
v
as a function of x. The
algebraic solution for z is messy and we will not be able to nd a closed form
expression for the antidervative v, so we will abandon the calculation at this
point. In practical applications one would generally use power series techniques
on equations of this form and calculate as many terms of the Taylor series of
the solution as are needed to give the desired accuracy.
Next consider equations of the form
(a
1
x +b
1
y +c
1
) dx + (a
2
x +b
2
y +c
2
) dy = 0.
If c
1
= c
2
= 0, the equation is homogeneous of degree 1. If not, try letting
x = x h and y = y k. We try to choose h and k to make the equation
homogeneous. Since h and k are constants, we have d x = dx and d y = dy.
Then our equation becomes
(a
1
x +a
1
h +b
1
y +b
1
k +c
1
) d x + (a
2
x +a
2
h +b
2
y +b
2
k +c
2
) d y = 0.
We want a
1
h +b
1
k = c
1
and a
2
h +b
2
k = c
2
. We can always solve for h and
k, unless

a
1
b
1
a
2
b
2

= 0.
24CHAPTER 2. FIRST ORDER ORDINARY DIFFERENTIAL EQUATIONS
So suppose

a
1
b
1
a
2
b
2

= 0.
Then (a
2
, b
2
) = m(a
1
, b
1
). Let z = a
1
x + b
1
y. Then dz = a
1
dx + b
1
dy. If
b
1
,= 0, we have
dy =
dz a
1
dx
b
1
,
(z +c
1
) dx + (mz +c
2
)
dz a
1
dx
b
1
= 0,
_
z +c
1
+
a
1
b
1
_
dx +
_
mz +c
2
b
1
_
dx = 0,
b
1
dx =
mz +c
2
z +c
1
+a
1
/b
1
dz.
This is a separable equation.
If b
1
= 0 but b
2
,= 0, we use z = a
2
x + b
2
y instead. Finally, if both b
1
= 0
and b
2
= 0, then the original equation is separable.
Chapter 3
Applications and Examples
of First Order Ordinary
Dierential Equations
3.1 Orthogonal Trajectories
An equation of the form f(x, y, C) = 0 determines a family of curves, one for
every value of the constant C. Figure 3.1 shows curves with several values of
C. Dierentiating f(x, y, C) = 0 with respect to x gives a second equation
C=1
C=4
C=9
=/4
=0
x
y
Figure 3.1: Graphs of x
2
+y
2
C = 0 with various values of C.
25
26CHAPTER 3. APPLICATIONS ANDEXAMPLES OF FIRST ORDER ODES
g(x, y, C, dy/dx) = 0, which is explicitly given by
g =
f
x
(x, y, C) +
f
y
(x, y, C)
dy
dx
.
Example 3.1. We have
x
2
+y
3
+Cx
4
= 0 implies 2x + 3y
2
dy
dx
+ 4Cx
2
= 0.
We can eliminate C between the equations f = 0 and g = 0, that is,
f = 0 = C =
x
2
+y
2
x
4
.
Therefore,
g = 0 = 2x 3y
2
dy
dx
4
x
2
+y
3
x
= 0.
This yields a rst order de, where the solution includes the initial family of
curves.
So just as a rst order de usually determines a one-parameter family of
curves (its solutions), a one-parameter family of curves determines a de.
A coordinate system (Cartesian coordinates, polar coordinates, etc.) consists
of two families of curves which always intersect at right angles. Two families of
curves which always intersect at right angles are called orthogonal trajectories
of each other. Given a one-parameter family of curves, we wish to nd another
family of curves which always intersects it at right angles.
Example 3.2. Find orthogonal trajectories of y = C sin(x).
Solution. Figure 3.2 shows the plot of y = C sin(x) with several values of C.
We have C = y/ sin(x), so it follows that
dy
dx
= C
..
y/ sin(x)
cos(x) = y
cos(x)
sin(x)
= y cot(x).
Two curves meet at right angles if the product of their slopes is 1, i.e., m
new
=
1/m
old
. So orthogonal trajectories satisfy the equation
dy
dx
=
1
y cot(x)
=
tan(x)
y
.
3.2. EXPONENTIAL GROWTH AND DECAY 27
-3p -2p - p 2p 3p
x
-5
-3
-1
1
3
5
Figure 3.2: A plot of the family y = C sin(x) (in black) and its orthogonal
trajectories (in gray).
This is a separable equation, and we quickly have
y dy = tan(x),
y
2
2
= ln([cos(x)[) +C,
y =
_
2 ln([cos(x)[) +C
1
,
where C
1
= 2C is also an arbitrary constant.
3.2 Exponential Growth and Decay
There are many situations where the rate of change of some quantity x is pro-
portional to the amount of that quantity, i.e., dx/dt = kx for some constant k.
The general solution is x = Ae
kt
.
1. Radium gradually changes into uranium. If x(t) represents the amount of
radium present at time t, then dx/dt = kx. In this case, k < 0.
2. Bank interest. The amount of interest you receive on your bank account is
proportional to your balance. In this case, k > 0. Actually, you do not do
quite this well at the bank because they compound only daily rather than
continuously, e.g., if we measure this time in years, our model predicts
that x = x
0
e
kt
, where x
0
is your initial investment and k is the interest
rate. Actually, you get
x = x
0
_
1 +
k
n
t
_
n
,
28CHAPTER 3. APPLICATIONS ANDEXAMPLES OF FIRST ORDER ODES
where n is the number of interest periods throughout the year. Typically,
n = 365 (daily interest), but the dierence is small when n = 365. Note
that lim
n
(1 +kt/n)
n
= e
kt
.
3. Population growth of rabbits. The number of rabbits born at any time
is roughly proportional to the number of rabbits present. If x(t) is the
number of rabbits at time t, then dx/dt = kx with k > 0. Obviously, this
model is not accurate as it ignores deaths.
Example 3.3. The half-life of radium is 1600 years, i.e., it takes 1600 years for
half of any quantity to decay. If a sample initially contains 50 g, how long will
it be until it contains 45 g?
Solution. Let x(t) be the amount of radium present at time t in years. Then
we know that dx/dt = kx, so x = x
0
e
kt
. With x(0) = 50, we quickly have
x = 50e
kt
. Solving for t gives t = ln(x/50) /k. With x(1600) = 25, we have
25 = 50e
1600k
. Therefore,
1600k = ln
_
1
2
_
= ln(2) ,
giving us k = ln(2) /1600. When x = 45, we have
t =
ln(x/50)
k
=
ln(45/50)
ln(2) /1600
= 1600
ln(8/10)
ln(2)
= 1600
ln(10/8)
ln(2)
1600
0.105
0.693
1600 0.152 243.2.
Therefore, it will be approximately 243.2 years until the sample contains 45 g
of radium.
3.3 Population Growth
Earlier, we discussed population growth where we considered only births. We
now consider deaths also. Take, as our model, the following.
Let N(t) be the number of people at time t. Assume that the land is intrinsi-
cally capable of supporting L people and that the rate of increase is proportional
to both N and L N. Then
dN
dt
= kN (L N) ,
3.4. PREDATOR-PREY MODELS 29
and the solution is
N =
L
1 + (L/N
0
1) e
kLt
,
where k is the proportionality constant.
3.4 Predator-Prey Models
Example 3.4 (Predator-Prey). Consider a land populated by foxes and rab-
bits, where the foxes prey upon the rabbits. Let x(t) and y(t) be the number
of rabbits and foxes, respectively, at time t. In the absence of predators, at any
time, the number of rabbits would grow at a rate proportional to the number
of rabbits at that time. However, the presence of predators also causes the
number of rabbits to decline in proportion to the number of encounters between
a fox and a rabbit, which is proportional to the product x(t)y(t). Therefore,
dx/dt = ax bxy for some positive constants a and b. For the foxes, the
presence of other foxes represents competition for food, so the number declines
proportionally to the number of foxes but grows proportionally to the number
of encounters. Therefore dy/dt = cx + dxy for some positive constants c and
d. The system
dx
dt
= ax bxy,
dy
dt
= cy +dxy
is our mathematical model.
If we want to nd the function y(x), which gives the way that the number
of foxes and rabbits are related, we begin by dividing to get the dierential
equation
dy
dx
=
cy +dxy
ax bxy
with a, b, c, d, x(t), y(t) positive.
This equation is separable and can be rewritten as
(a by) dy
y
=
(c +dx) dx
x
.
Integrating gives
a ln(y) by = c ln(x) +dx +C,
30CHAPTER 3. APPLICATIONS ANDEXAMPLES OF FIRST ORDER ODES
or equivalently
y
a
e
by
= kx
c
e
dx
(3.1)
for some positive constant k = e
C
.
The graph of a typical solution is shown in Figure 3.3.
Figure 3.3: A typical solution of the Predator-Prey model with a = 9.4, b = 1.58,
c = 6.84, d = 1.3, and k = 7.54.
Beginning at a point such as a, where there are few rabbits and few foxes, the
fox population does not initially increase much due to the lack of food, but with
so few predators, the number of rabbits multiplies rapidly. After a while, the
point b is reached, at which time the large food supply causes the rapid increase
in the number of foxes, which in turn curtails the growth of the rabbits. By
the time point c is reached, the large number of predators causes the number of
rabbits to decrease. Eventually, point d is reached, where the number of rabbits
has declined to the point where the lack of food causes the fox population to
decrease, eventually returning the situation to point a.
3.5 Newtons Law of Cooling
Let T and T
s
be the temperature of an object and its surroundings, respectively.
Let T
0
and T
s0
be initial temperatures. Then Newtons Law of Cooling states
3.6. WATER TANKS 31
that
dT
dt
= k (T
s
T) , k > 0.
As T changes, the object gives or takes heat from its surroundings. We now
have two cases.
Case 1: T
s
is constant. This occurs either because the heat given o is
transported elsewhere or because the surroundings are so large that the contri-
bution is negligible. In such a case, we have
dT
dt
+kT = kT
s
,
which is linear, and the solution is
T = T
0
e
kt
+T
s
_
1 e
kt
_
.
Case 2: The system is closed. All heat lost by the object is gained by its
surroundings. We need to nd T
s
as a function of T. We have
change in temperature =
heat gained
weight specic heat capacity
.
Therefore,
T T
0
wc
=
T
s
T
s0
w
s
c
s
,
T
s
= T
s0
+
wc
w
s
c
s
(T
0
T) ,
dT
dt
+k
_
1 +
wc
w
s
c
s
_
T = k
_
T
s0
+
wc
w
s
c
s
T
0
_
.
This is linear with dierent constants than the previous case. The solution is
T = T
0
+
_
T
s0
+
wc
wscs
T
0
1 +
wc
wscs
_
_
1 e
k(1+
wc
wscs
)t
_
.
3.6 Water Tanks
Example 3.5. A tank contains a salt water solution consisting initially of 20 kg
of salt dissolved into 10 of water. Fresh water is being poured into the tank
at a rate of 3

/min and the solution (kept uniform by stirring) is owing out at


2

/min. Figure 3.4 shows this setup. Find the amount of salt in the tank after
5 minutes.
32CHAPTER 3. APPLICATIONS ANDEXAMPLES OF FIRST ORDER ODES
Figure 3.4: Fresh water is being poured into the tank as the well-mixed solution
is owing out.
Solution. Let Q(t) be the amount (in kilograms) of salt in the tank at time t
(in minutes). The volume of water at time t is
10 + 3t 2t = 10 +t.
The concentration at time t is given by
amount of salt
volume
=
Q
10 +t
kg per litre. Then
dQ
dt
= (rate at which salt is leaving) =
Q
10 +t
2 =
2Q
10 +t
.
Thus, the solution to the problem is the solution of
dQ
dt
=
2Q
10 +t
evaluated at t = 5. We see that it is simply a separable equation.
3.6. WATER TANKS 33
To solve it, we have
dQ
Q
= 2
dt
10 +t
,
_
dQ
Q
= 2
_
dt
10 +t
,
ln([Q[) = 2 ln([10 +t[) ,
ln([Q[) = ln
_
[10 +t[
2
_
+C,
where C is a constant. Thus,
ln([Q[) = ln
_
[10 +t[
2
_
+C,
= ln
_
A[10 +t[
2
_
,
[Q[ = A[10 +t[
2
where A = e
C
. But Q 0 (we cannot have a negative amount of salt) and t 0
(we do not visit the past), so we remove absolute value signs, giving us
Q = A(10 +t)
2
.
Initially, i.e., at t = 0, we know that the tank contains 20 kg of salt. Thus,
the initial condition is Q(0) = 20, and we have
Q(20) = 0 =
A
(10 + 0)
2
= 20 =
A
100
= 20 = A = 2000.
Our equation therefore becomes
Q(t) =
2000
(10 +t)
2
.
Figure 3.5 shows a plot of the solution. Evaluating at t = 5 gives
Q(5) =
2000
15
2
=
80
225
=
80
9
8.89.
Therefore, after 5 minutes, the tank will contain approximately 8.89 kg of salt.
Additional conditions desired of the solution (Q(0) = 20) in the above exam-
ple) are called boundary conditions and a dierential equations together with
boundary conditions is called a boundary-value problem (BVP).
34CHAPTER 3. APPLICATIONS ANDEXAMPLES OF FIRST ORDER ODES
0 100 200 300 400 500 600
t
5
10
15
20
Q
Figure 3.5: The plot of the amount of salt Q(t) in the tank at time t shows that
salt leaves slower as time moves on.
A =
A
(10+t)
2
is the general solution to the DE
dQ
dt
=
2dt
10+t
. Q =
2000
(10+t)
2
is the
solution to the boundary value problem
dQ
dt
=
2dt
10+t
.; Q(0) = 200. Boundary-
value problems like this one where the boundary conditions are initial values of
various quantities are also called initial-value problems (IVP).
3.7 Motion of Objects Falling Under Gravity
with Air Resistance
Let x(t) be the height at time t, measured positively on the downward direction.
If we consider only gravity, then
a =
d
2
x
dt
2
is a constant, denoted g, the acceleration due to gravity. Note that F = ma =
mg. Air resistance encountered depends on the shape of the object and other
things, but under most circumstances, the most signicant eect is a force op-
posing the motion which is proportional to a power of the velocity. So
F
..
ma
= mg kv
n
and
d
2
x
dt
2
= g
k
m
_
dx
dt
_
n
,
3.7. MOTIONOF OBJECTS FALLINGUNDER GRAVITYWITHAIR RESISTANCE35
which is a second order de, but there is no x term. So it is rst order in x

.
Therefore,
dv
dt
= g
k
m
v
n
.
This is not easy to solve, so we will make the simplifying approximation that
n = 1 (if v is small, there is not much dierence between v and v
n
). Therefore,
we have
dv
dt
= g
k
m
v,
dv
dt
+
k
m
v = g,
The integrating factor is
I = e
R
k
m
dt
= e
kt/m
.
Therefore,
_
dv
dt
+
k
m
v
_
e
kt/m
= ge
kt/m
,
e
kt/m
v =
gm
k
e
kt/m
+C,
v =
mg
k
+Ce
kt/m
,
where C is an arbitrary constant. Note that
v(0) = v
0
= v
0
=
mg
k
+C = C = v
0

mb
k
.
So we have
v
..
dx/dt
=
mg
k
+
_
v
0

mg
k
_
e
kt/m
.
Note that
_
x
x0
dx
. .
xx0
=
_
t
0
v dt =
mg
k
t
m
k
_
v
0

mg
k
__
e
kt/m
1
_
.
Thus, we nally have
x = x
0
+
mg
k
t +
m
k
_
v
0

mg
k
__
1 e
kt/m
_
.
36CHAPTER 3. APPLICATIONS ANDEXAMPLES OF FIRST ORDER ODES
3.8 Escape Velocity
Let x(t) be the height of an object at time t measured positively upward (from
the centre of the earth). Newtons Law of Gravitation states that
F
..
ma
=
kmM
x
2
,
where m is the mass of the object, M is the mass of the Earth, and k > 0. Note
that
x

=
kM
x
2
.
We dene the constant g known as the acceleration due to gravity by x

= g
when x = R, where R = radius of the earth. So k = gR
2
/M. Therefore
x

= gR
2
/x
2
. Since t does not appear, letting v = dx/dt so that
d
2
x
dt
2
=
dv
dx
dx
dt
=
dv
dx
v
will reduce it to rst order. Thus,
v
dv
dx
=
gR
2
x
2
,
v dv =
gR
2
x
2
dx,
_
v dv =
_

gR
2
x
2
dx,
v
2
2
=
gR
2
x
+C,
for some arbitrary constant C. Note that v decreases as x increases, and if v is
not suciently large, eventually v = 0 at some height x
max
. So C = gR
2
/x
max
and
v
2
= 2R
2
g
_
1
x

1
x
max
_
.
Suppose we leave the surface of the Earth starting with velocity V . How far
3.9. PLANETARY MOTION 37
will we get? (v = V when x = R). We have
V
2
= 2R
2
g
_
1
R

1
x
max
_
,
V
2
2R
2
g
=
1
R

1
x
max
,
1
x
max
=
1
R

V
2
2R
2
g
=
2Rg V
2
2R
2
g
,
x
max
=
2R
2
g
2Rg V
2
.
That is, if V
2
< 2Rg, we will get as far as
2R
2
g
2Rg V
2
and then fall back. To escape, we need V

2Rg. Thus, the escape velocity
is

2Rg 6.9
mi
/s, or roughly 25000 mph.
3.9 Planetary Motion
Newtons law of gravitation states that
F
..
ma
=
km
r
2
r.
Suppose we let a
r
= k/r
2
and a
0
= 0. Then
a
r
= a
x
cos() +a
y
sin(), a

= a
x
sin() +a
y
cos().
Let x = r cos() and y = r sin(). Then
x

= r

cos() r sin()

,
a
x
= x

= r

cos() r

sin()

sin()

r cos() (

)
2
r sin()

,
y

= r

sin() +r cos()

,
a
y
= y

= r

sin() +r

cos()

+r

cos()

r sin() (

)
2
+r cos()

= r

sin() + 2r

cos() (

)
2
r sin() +

r cos().
38CHAPTER 3. APPLICATIONS ANDEXAMPLES OF FIRST ORDER ODES
Now a
r
becomes
a
r
= r

cos
2
() 2r

sin() cos() (

)
2
r cos
2
()

sin() cos()
+r

sin
2
() + 2r

sin() cos() (

)
2
r sin
2
() +

r sin() cos()
= r

)
2
r
and a

becomes
a

= r

sin() cos() + 2r

sin
2
() + (

)
2
r sin() cos() +

r sin
2
()
+r

sin() cos() + 2r

cos
2
() (

)
2
r sin() cos() +

r cos
2
()
= 2r

r.
Note that
r

)
2
r =
k
r
2
, ()
2r

r = 0. ()
Equation () implies that

2rr

. .
d(r
2

)
r
2
= 0 = r
2

= h.
We want the path of the planet, so want an equation involving r and . There-
fore, we eliminate t between Equation () and r
2

= h. Thus,
r

=
dr
d

=
dr
d
h
r
2
,
r

=
d
2
r
d
2

h
r
2
2
dr
d
h
r
2
r

=
d
2
r
d
2
h
r
2
h
r
2
2
dr
d
h
r
3
dr
d
h
r
2
=
h
2
r
2
_
2
r
3
_
dr
d
_
2

1
r
2
d
2
r
d
2
_
.

Note that
A =
Z

0
Z
r()
0
r dr d =
Z

0
r

()
2
2
d,
so dA/d = r()
2
/2. So in our case, A

= h/2. Therefore, A = ht/2, which is Keplers Second


Law (area swept out is proportional to time).
3.10. PARTICLE MOVING ON A CURVE 39
Let u = 1/r. Then
du
d
=
1
r
2
dr
d
,
d
2
u
d
2
=
2
r
3
_
dr
d
_
2

1
r
2
d
2
r
d
2
=
r
2
r

h
2
=
r

h
2
u
2
.
Therefore, Equation () implies that
h
2
u
2
d
2
u
d
2

_
h
r
2
_
2
r =
k
r
2
,
h
2
u
2
d
2
u
d
2
u
3
h
2
= ku
2
, u ,= 0,
d
2
u
d
2
+u =
k
h
2
.
Therefore,
u
..
1/r
= Bcos( ) +
k
h
2
= C
1
sin() +C
2
cos() +
k
h
2
,
where
r =
1
k/h
2
+Bcos( )
=
1
k/h
2
+C
1
sin() +C
2
cos()
.
Therefore,
k
h
2
r+C1x+C2y
..
k
h
2
r +C
1
r sin() +C
2
r cos() = 1,
k
h
2
r = 1 C
1
x C
2
x,
k
2
h
4
_
x
2
+y
2
_
= (1 C
1
x C
2
y)
2
,
which is the equation of an ellipse.
3.10 Particle Moving on a Curve
Let x(t) and y(t) denote the x and y coordinates of a point at time t, respectively.
Pick a reference point on the curve and let s(t) be the arc length of the piece
of the curve between the reference point and the particle. Figure 3.6 illustrates
this idea. Then
40CHAPTER 3. APPLICATIONS ANDEXAMPLES OF FIRST ORDER ODES
Reference point
Particle
Figure 3.6: A particle moving on a curve.
v =
ds
dt
, |a| = a =
d
2
s
dt
2
.
The vector v is in the tangent direction. The goal is to nd components a in
the tangential and normal directions. Let be the angle between the tangent
line and the x axis. Then
cos() =
dx
ds
, sin() =
dy
ds
.
In the standard basis, we have a = (a
x
, a
y
), where
a
x
=
d
2
x
dt
2
, a
y
=
d
2
y
dt
2
.
To get components in T, n basis, apply the matrix representing the change of
basis from T, n to the standard basis, i.e., rotation by , i.e.,
_
cos() sin()
sin() cos()
__
a
x
a
y
_
=
_
a
x
cos() +a
y
sin()
a
x
sin() +a
y
cos()
_
,
i.e.,
a
T
= a
x
cos() +a
y
sin() = a
x
dx
ds
+a
y
dy
ds
=
_
a
x
dx
dt
+a
y
dy
dt
_
dt
ds
=
a
x
v
x
+a
y
v
y
v
and
a
n
= a
x
sin() +a
y
cos() =
a
y
v
x
a
x
v
y
v
.
Also, we have v =
_
v
2
x
+v
2
y
, so
d
2
s
dt
2
=
dv
dt
=
2v
x
dvx
dt
+ 2v
y
dvy
dt
2
_
v
2
x
+v
2
y
=
v
x
a
x
+v
y
a
y
v
= a
T
.
3.10. PARTICLE MOVING ON A CURVE 41
But this was clear to start with because it is the component of a in the tangential
direction that aects ds/dt.
We now have
y

=
dy
dx
=
dy/dt
dx/dt
=
v
y
v
x
,
y

=
v
x
dvy
dx
v
y
dvx
dx
v
2
x
=
v
x
dvy
dt
dt
dx
v
y
dvx
dt
dt
dx
v
2
x
=
v
x
a
y
v
y
a
x
v
3
x
=
a
n
v
v
3
x
,
where a
n
= y

v
3
x
/v and
v
x
=
dx
dt
=
dx
ds
ds
dt
=
dx
ds
v,
so
a
n
=
y

_
dx
ds
_
3
v
3
v
=
y

v
2
(ds/dx)
3
=
y

_
1 + (y

)
2
_
3/2
v
2
.
Let
R =
_
1 + (y

)
2
_
3/2
y

(3.2)
so that a
n
= v
2
/R. The quantity R is called the radius of curvature. It is the
radius of the circle which mostly closely ts the curve at that point. This is
shown in Figure 3.7. If the curve is a circle, then R is a constant equal to the
R
Figure 3.7: The radius of curvature is the circle whose edge best matches a
given curve at a particular point.
radius.
Conversely, suppose that R is a constant. Then
Ry

=
_
1 + (y

)
2
_
3/2
.
42CHAPTER 3. APPLICATIONS ANDEXAMPLES OF FIRST ORDER ODES
Let z = y

. Then
Rz

=
_
1 +z
2
_
3/2
and
dx =
Rdz
(1 +z
2
)
3/2
,
so
x =
_
R
(1 +z
2
)
3/2
dz =
Rz

1 +z
2
+A,
x A =
R
_
(1/z)
2
+ 1
,
_
1
z
_
2
+ 1 =
R
2
(x A)
2
,
_
1
z
_
2
=
R
2
(x A)
2
1 =
R
2
(x A)
2
(x A)
2
,
z
2
=
(x A)
2
R
2
(x A)
2
,
z
..
dy/dx
=
x A
_
R
2
(x A)
2
.
Therefore,
y =
_
x A
_
R
2
(x A)
2
dx
=
_
R
2
(x A)
2
+B,
(y B)
2
= R
2
(x A)
2
,
(x A)
2
+ (y B)
2
= R
2
.
Suppose now that the only external force F acting on the particle is gravity,
i.e., F = (0, mg). Then in terms of the T, n basis, the components of F are
_
cos() sin()
sin() cos()
__
0
mg
_
=
_
mg sin()
mg cos()
_
=
_
mg
dy
ds
mg
dx
ds
_
.
3.10. PARTICLE MOVING ON A CURVE 43
Equating the tangential components gives us

m
d
2
s
dt
2
=

mg
dy
ds
,
ds
dt
= v,
d
2
s
dt
2
=
dv
dt
=
ds
dt
=
dv
ds
v,
dv
ds
v = g
dy
ds
,
v dv = g dy,
v
2
2
= g(y y
0
) = g(y
0
y),
ds
dt
= v =
_
2g(y
0
y).
To proceed, we must know y(s), which depends on the curve. Therefore,
dt =
ds
_
2g(y
0
y(s))
,
t =
_
ds
_
2g(y
0
y(s))
.
If the ma
n
ever becomes greater than the normal component of F, the particle
will y o the curve. This will happen when

m
y

v
3
x
v
=

mg
dx
ds
= g
dx
dt
dt
ds
= g
v
x
v
,
y

v
2
x
= g,
v
x
=
dx
dt
=
dx
ds
ds
dt
=
dx
ds
v.
Therefore,
_
_
_
_
_

g = y

_
dx
ds
_
2
v
2
=
y

v
2
(ds/dx)
2
=
y

v
2
1 + (y

)
2
=
y

g(y
0
y)
1 + (y

)
2
_
_
_
_
_
= 2y

(y y
0
) = 1 + (y

)
2
.
Example 3.6. A particle is placed at the point (1, 1) on the curve y = x
3
and
released. It slides down the curve under the inuence of gravity. Determine
whether or not it will ever y o the curve, and if so, where.
44CHAPTER 3. APPLICATIONS ANDEXAMPLES OF FIRST ORDER ODES
Solution. First note that
y = x
3
, y

= 3x
2
, y

= 6x, y
0
= 1.
Therefore,
2y

(y y
0
) = 1 (y

)
2
,
12x
_
x
3
1
_
= 1 + 9x
4
,
12x
4
12x = 1 + 9x
4
,
3x
4
12x 1 = 0.
Solving for the relevant value of x gives us x 0.083, which corresponds to y
0.00057. Therefore, the particle will y o the curve at (0.083, 0.00057).
Figure 3.8 shows the graph.
H1,1L
H-0.083,-0.00057L
-2 -1 1 2
x
-8
-6
-4
-2
2
4
6
8
Figure 3.8: A particle released from (1, 1), under the inuence of gravity, will
y o the curve at approximately (0.083, 0.00057).
Chapter 4
Linear Dierential
Equations
Although we have dealt with a lot of manageable cases, rst order equations
are dicult in general. But second order equations are much worse. The most
manageable case is linear equations. We begin with the general theory of linear
dierential equations. Specic techniques for solving these equations will be
given later.
Denition (nth order linear dierential equation)
An nth order linear dierential equation is an equation of the form

d
n
y
dx
n
+P
1
(x)
d
n1
y
dx
n1
+P
2
(x)
d
n2
y
dx
n2
+ +P
n
(x)y = Q(x).
An nth order linear equation can be written as a linear system (see Chapter
9) as follows. Let
y
1
(x) = y(x), y
2
(x) =
dy
dx
, y
3
(x) =
d
2
y
dx
2
, . . . , y
n
(x) =
d
n1
y
dx
n1
.
Then
dy
1
dx
= y
2
,
dy
2
dx
= y
3
, . . . ,
dy
n1
dx
= y
n

If the coecient of d
n
y/dx
n
is not 1, we can divide the equation through this coecient
to write it in normal form. A linear equation is in normal form when the coecient of y
(n)
is
1.
45
46 CHAPTER 4. LINEAR DIFFERENTIAL EQUATIONS
and we have
dy
n
dx
=
d
n
y
dx
n
= P
1
(x)
d
n1
y
dx
n1
P
n
(x)y +Q(x)
= P
1
(x)y
n
P
n
(x)y
1
+Q(x).
Therefore,
dY
dx
= A(x)Y +B(x),
where
A(x) =
_

_
0 1 0 0
0 0 1 0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0 0 0 1
P
n
(x) P
n1
(x) P
n2
(x) P
1
(x)
_

_
and
B(x) =
_

_
0
0
.
.
.
0
Q(x)
_

_
.
Theorem 4.1
If P
1
(x), . . . , P
n
(x), Q(x) are continuous throughout some interval I and x
0
is an interior point of I, then for any numbers W
0
, . . . , W
n1
, there exists
a unique solution

throughout I of
d
n
y
dx
n
+P
1
(x)
d
n1y
dx
n1
+ +P
n
(x)y = Q(x)
satisfying
y(x
0
) = w
0
, y

(x
0
) = w
1
, . . . , y
(n1)
(x
0
) = w
n1
.
Proof. The proof is given in Chapter 10.
4.1. HOMOGENEOUS LINEAR EQUATIONS 47
4.1 Homogeneous Linear Equations
To begin, consider the case where Q(x) 0 called the homogeneous case, where
y
(n)
+P
1
(x)y
(n1)
+P
2
(x)y
(n2)
+ +P
n
(x)y = 0. (4.1)
Proposition 4.2
1. If r(x) is a solution of Equation (4.1), then so is ar(x) for all r R.
2. If r(x) and s(x) are solutions of Equation (4.1), then so is r(x) +s(x).
Proof.
1. Substituting y = ar into the lhs gives
ar
(n)
+P
1
(x)ar
(n1)
+ +P
n
(x)ar = a
_
r
(n)
+ +P
n
(x)r
_
.
But since r is a solution, we have
a
_
r
(n)
+ +P
n
(x)r
_
= a 0.
Therefore, ar is a solution.
2. Substituting y = r +s in the lhs gives
_
_
_
r
(n)
+s
(n)
+P
1
(x)
_
r
(n1)
+ +s
(n1)
_
+
+P
n
(x) (r +s)
_
_
_ =
_
r
(n)
+ +P
n
(x)r
+s
(n)
+ +P
n
(x)s
_
= 0 + 0 = 0.
Therefore, r +s is a solution.
Corollary 4.3
The set of solutions to Equation (4.1) forms a vector space.
48 CHAPTER 4. LINEAR DIFFERENTIAL EQUATIONS
Proof. Let V be the set of solutions to Equation (4.1) and let y
1
, y
2
, . . . , y
n
V .
Further, let x
0
R. Set
W(y
1
, y
2
, . . . , y
n
) (x) =

y
1
(x) y
2
(x) y
n
(x)
y

1
(x) y

2
(x) y

n
(x)
.
.
.
.
.
.
.
.
.
.
.
.
y
(n1)
1
(x) y
(n1)
2
(x) y
(n1)
n
(x)

,
the Wronskian of y
1
, y
2
, . . . , y
n
. Then
dW
dx
=
two equal rows
..

1
(x) y

n
(x)
y

1
(x) y

n
(x)
.
.
.
.
.
.
.
.
.
y
(n1)
1
(x) y
(n1)
n
(x)

+
two equal rows
..

y
1
(x) y
n
(x)
y

1
(x) y

n
(x)
y

1
(x) y

n
(x)
.
.
.
.
.
.
.
.
.
y
(n1)
1
(x) y
(n1)
n
(x)

+
+

y
1
(x) y
n
(x)
.
.
.
.
.
.
.
.
.
y
(n2)
1
(x) y
(n2)
n
(x)
y
(n)
1
(x) y
(n)
n
(x)

y
1
(x) y
n
(x)
.
.
.
.
.
.
.
.
.
y
(n2)
1
(x) y
(n2)
n
(x)
y
(n)
1
(x) y
(n)
n
(x)

y
1
(x) y
n
(x)
.
.
.
.
.
.
.
.
.
y
(n2)
1
(x) y
(n2)
n
(x)

n
k=1
P
k
(x)y
(nk)
1

n
k=1
P
k
(x)y
(nk)
n

=
n

k=1
_
_
_
_
_
_
P
k
(x)

y
1
(x) y
n
(x)
.
.
.
.
.
.
.
.
.
y
(n2)
1
(x) y
(n2)
n
(x)
y
(n)
1
(x) y
(n)
n
(x)

_
_
_
_
_
_
= P
1
(x)W.
Therefore, W = Ae

R
P1(x) dx
. This is Abels formula. There are two cases:
A = 0 = W(x) 0,
A ,= 0 = W(x) ,= x
for all x.
4.1. HOMOGENEOUS LINEAR EQUATIONS 49
Theorem 4.4
Let V be the set of solutions to Equation (4.1) and suppose that
y
1
, y
2
, . . . , y
n
V . Then y
1
, y
2
, . . . , y
n
are linearly independent if and only
if W ,= 0.
Proof. Let V be the set of solutions to Equation (4.1) and suppose that y
1
, y
2
, . . . , y
n

V . Suppose that y
1
, y
2
, . . . , y
n
are linearly dependent. Then there exists con-
stants c
1
, c
2
, . . . , c
n
such that
c
1
y
1
(x) +c
2
y
2
(x) + +c
n
y
n
(x) = 0,
c
1
y

1
(x) +c
2
y

2
(x) + +c
n
y

n
(x) = 0,
.
.
.
c
1
y
(n1)
1
(x) +c
2
y
(n1)
2
(x) + +c
n
y
(n1)
n
(x) = 0.
Since there exists a nonzero solution for (c
1
, c
2
, . . . , c
n
) of the equation, we have
M
_

_
c
1
c
2
.
.
.
c
n
_

_
= 0,
so it follows that [M[ = 0 = W.
Now suppose that W = 0. Select an x
0
R. Since W(x) 0, we have
W(x
0
) = 0. Thus, there exists constants c
1
, c
2
, . . . , c
n
not all zero such that
M
_

_
c
1
c
2
.
.
.
c
n
_

_
,
where
M =
_

_
y
1
(x
0
) y
n
(x
0
)
.
.
.
.
.
.
.
.
.
y
(n1)
1
(x
0
) y
(n1)
n
(x
0
)
_

_
,
50 CHAPTER 4. LINEAR DIFFERENTIAL EQUATIONS
that is,
c
1
y
1
(x
0
) +c
2
y
2
(x
0
) + +c
n
y
n
(x
0
) = 0,
c
1
y

1
(x
0
) +c
2
y

2
(x
0
) + +c
n
y

n
(x
0
) = 0,
.
.
.
c
1
y
(n1)
1
(x
0
) +c
2
y
(n1)
2
(x
0
) + +c
n
y
(n1)
n
(x
0
) = 0.
Let y(x) = c
1
y
1
(x) + c
2
y
2
(x) + + c
n
y
n
(x). Since y
1
, y
2
, . . . , y
n
are solutions
to Equation (4.1), so is f, and f satises
f(x
0
) = 0, f

(x
0
) = 0, f

(x
0
) = 0, . . . , f
(n1)
(x
0
) = 0.
But y = 0 also satises these conditions, and by the uniqueness part of Theorem
4.1, it is the only function satisfying them. Therefore, f 0, that is,
c
1
y
1
(x
0
) +c
2
y
2
(x
0
) + +c
n
y
n
(x
0
) 0.
Therefore, y
1
, y
2
, . . . , y
n
are linearly dependent.
Corollary 4.5
Let V be the set of solutions to Equation (4.1). Then we have dim(V ) = n.
Proof. Let V be the set of solutions to Equation (4.1). Pick a point x
0
R.
Then by Theorem 4.1, given any numbers w
1
, w
2
, . . . , w
n
, there exists a unique
solution of Equation (4.1) satisfying y
(j1)
(x
0
) for j = 1, 2, . . . , n. Let y
1
be
such a solution with (w
1
, w
2
, . . . , w
n
) = (0, 0, . . . , 0), let y
2
be such a solution
with (w
1
, w
2
, . . . , w
n
) = (0, 1, . . . , 0), etc., nally letting y
n
be such a solution
with (w
1
, w
2
, . . . , w
n
) = (0, 0, . . . , 1).
We claim that y
1
, y
2
, . . . , y
n
are linearly independent. Suppose that
c
1
y
1
(x) +c
2
y
2
(x) + +c
n
y
n
(x) 0.
Then dierentiating gives
c
1
y
(j)
1
(x) +c
2
y
(j)
2
(x) + +c
n
y
(j)
n
(x) 0
4.1. HOMOGENEOUS LINEAR EQUATIONS 51
for all j. In particular,
c
1
y
(j)
1
(x
0
) +c
2
y
(j)
2
(x
0
) + +c
n
y
(j)
n
(x
0
) 0
for all j. Therefore, y
1
, y
2
, . . . , y
n
are linearly independent and dim(V )) n.
Conversely, let y
1
, . . . , y
n
be linearly independent. Then W(y
1
, . . . , y
n
) (x) ,=
0 for all x. Suppose that y V and let
M =
_

_
y
1
(x
0
) y
2
(x
0
) y
n
(x
0
)
y

1
(x
0
) y

2
(x
0
) y

n
(x
0
)
.
.
.
.
.
.
.
.
.
.
.
.
y
(n1)
1
(x
0
) y
(n1)
2
(x
0
) y
(n1)
n
(x
0
)
_

_
.
Therefore, [M[ = [W(x
0
)[ ,= 0. So there exists constants (c
1
, c
2
, . . . , c
n
) such
that
M
_

_
c
1
c
2
.
.
.
c
n
_

_
=
_

_
y(x
0
)
y

(x
0
)
.
.
.
y
(n1)
(x
0
)
_

_
,
that is,
y
(j)
(x
0
) = c
1
y
(j)
1
+c
2
y
(j)
2
+ +c
1
y
(j)
1
for 0 i n 1. Let f(x) = c
1
y
1
(x) + c
2
y
2
(x) + + c
n
y
n
(x). Then f is
a solution to Equation (4.1) and y
(j)
(x
0
) = f
(j)
(x
0
) for j = 0, . . . , n 1. By
uniqueness, y = f, i.e., y(x) c
1
y
1
(x) + c
2
y
2
(x) + + c
n
y
n
(x). This means
that y is a linear combination of y
1
, . . . , y
n
. So y
1
, . . . , y
n
forms a basis for V .
Therefore, dim(V ) = n.
We conclude that, to solve Equation (4.1), we need to nd n linearly in-
dependent solutions y
1
, . . . , y
n
of Equation (4.1). Then the general solution to
Equation (4.1) is c
1
y
1
+c
2
y
2
+ +c
n
y
n
.
Example 4.6. Solve y

+ 5y

+ 6y = 0.
Solution. It is easy to check that y = e
3x
and y = e
2x
each satises y

+ 5y

+
6y = 0. Therefore, the general solution is c
1
e
3x
+ c
2
e
2x
, where c
1
and c
2
are
arbitrary constants.
52 CHAPTER 4. LINEAR DIFFERENTIAL EQUATIONS
4.1.1 Linear Dierential Equations with Constant Coe-
cients
The next problem concerns how to nd the n linearly independent solutions. For
n > 1, there is no good method in general. There is, however, a technique which
works in the special cases where all the coecients are constants. Consider the
constant coecient linear equation
a
0
y
(n)
+a
1
y
(n1)
+ +a
n1
y

+a
n
y = 0. (4.2)
We rst look for solutions of the form y = e
x
. If y = e
x
is a solution, then
a
0

n
e
x
+a
1

n1
e
x
+ +a
n
e
x
= 0.
Now since e
x
,= 0, we have
a
0

n
+a
1

n1
+ +a
n
= 0.
Let
p() = a
0

n
+a
1

n1
+ +a
n
. (4.3)
If Equation (4.3) has n distinct real roots r
1
, r
2
, . . . , r
n
, then we have n linearly
independent solutions e
r1x
, e
r2x
, . . . , e
rnx
. We would like one solution for each
root like this.
Suppose that Equation (4.3) has a repeated real root r. Then we do not
have enough solutions of the form e
x
. But consider y = xe
rx
. Then
y

= re
rx
+rxe
rx
= (rx + 1) e
rx
,
y

= re
rx
+r (rx + 1) e
rx
=
_
r
2
x + 2r
_
e
rx
,
y

= r
2
e
rx
+r
_
r
2
x + 2r
_
e
rx
=
_
r
3
x + 3r
2
_
e
rx
,
.
.
.
y
(n)
=
_
r
n
x +nr
n1
_
e
rx
.
4.1. HOMOGENEOUS LINEAR EQUATIONS 53
Substituting y = xe
rx
in Equation (4.2) gives us
n1

k=0
a
k
_
r
nk
x + (n k) r
nk
_
=
_
n

k=0
a
k
r
nk
_
xe
rx
+
_
n1

k=0
a
k
(n k) r
nk1
_
e
rx
= p(r)xe
rx
+p

(r)e
rx
.
Since r is a repeated root of p, both p(r) = 0 and p

(r) = 0, that is,


p() = ( r)
2
q() = p

() = 2 ( r) q() + ( r)
2
q

()
= p

(r) = 0.
Therefore, xe
rx
is also a solution of Equation (4.2).
What about complex roots? They come in pairs, i.e., r
1
= a + ib and
r
2
= a ib, so e
(a+ib)x
and e
(aib)x
satisfy Equation (4.2). But we want real
solutions, not complex ones. But
e
(a+ib)x
= e
ax
(cos(bx) +i sin(bx)) ,
e
(aib)x
= e
ax
(cos(bx) i sin(bx)) .
Let
z
1
= e
ax
(cos(bx) +i sin(bx)) ,
z
2
= e
ax
(cos(bx) i sin(bx)) .
If z
1
and z
2
are solutions, then
z
1
+z
2
2
= e
ax
cos(bx),
z
1
z
2
2
= e
ax
sin(bx)
are solutions. So the two roots a + ib and a ib give two solutions e
ax
cos(bx)
and e
ax
sin(bx). For repeated complex roots, use xe
ax
cos(bx), xe
ax
sin(bx), etc.
So in all cases, we get n solutions.
The expression c
1
e
ax
cos(bx) +e
ax
c
2
sin(bx) can be rewritten as follows. Let
A =
_
c
2
1
+c
2
2
. Then
= cos
1
_
c
1
_
c
2
1
+c
2
2
_
,
54 CHAPTER 4. LINEAR DIFFERENTIAL EQUATIONS
that is,
cos() =
c
1
_
c
2
1
+c
2
2
, sin() =
_
1 cos
2
() =
c
2
_
c
2
1
+c
2
2
.
Therefore,
c
1
e
ax
cos(bx) +e
ax
c
2
sin(bx) = Ae
ax
_
c
1
A
cos(bx) +
c
2
A
sin(bx)
_
= Ae
ax
(cos() cos(bx) + sin() sin(bx))
= Ae
ax
(cos(bx )) .
Example 4.7. Solve y
(5)
4y
(4)
+ 5

+ 6

36y

+ 40y = 0.
Solution. We have

5
4
4
+ 5
3
+ 6
2
36 + 40 = 0,
( 2)
2
( + 2)
_

2
2 + 5
_
= 0.
Therefore, = 2, 2, 2, 1 2i and
y = c
1
e
2x
+c
2
xe
2x
+c
3
e
2x
+c
4
e
x
cos(2x) +c
5
sin(2x),
where c
1
, c
2
, c
3
, c
4
, c
5
are arbitrary constants.
4.2 Nonhomogeneous Linear Equations
A nonhomogeneous linear equation is an equation of the form
y
(n)
+p
1
(x)y
(n1)
+p
2
(x)y
(n2)
+ +p
n
(x)y = q(x). (4.4)
Let Equation (4.1) (p. 47) be the corresponding homogeneous equation. If u
and v are solutions of Equation (4.4), then uv is a solution of Equation (4.1).
Conversely, if u is a solution of Equation (4.4) and v is a solution of Equation
(4.1), then u +v is a solution of Equation (4.4).
Let y
1
, y
2
, . . . , y
n
be n linearly independent solutions of Equation (4.1) and
let y
p
be a solution of Equation (4.4). If y is any solution to Equation (4.4),
then y y
p
is a solution to Equation (4.1), so
y y
p
= c
1
y
1
+c
2
y
2
+ +c
n
y
n
.
4.2. NONHOMOGENEOUS LINEAR EQUATIONS 55
Therefore,
y = c
1
y
1
+c
2
y
2
+ +c
n
y
n
+y
p
is the general solution to Equation (4.4).
Example 4.8. Solve y

y = x.
Solution. The corresponding homogeneous equation is y

y = 0. It then
follows that
2
1 = 0 = 1. Therefore,
y
1
= e
x
, y
2
= e
x
.
By inspection, y
p
= x solves the given equation. Therefore, the general solution
is
y = c
1
e
x
+c
2
e
x
x,
where c
1
and c
2
are arbitrary constants.
We will consider more systematically how to nd y
p
later.
56 CHAPTER 4. LINEAR DIFFERENTIAL EQUATIONS
Chapter 5
Second Order Linear
Equations
A second order linear equation is an equation of the form
y

+P(x)y

+Q(x)y = R(x). (5.1)


To nd the general solution of Equation (5.1), we need
Two solutions: y
1
and y
2
of the corresponding homogeneous equation
y

+P(x)y

+Q(x) = 0.
One solution: y
p
of Equation (5.1).
Then the general solution is y = c
1
y
1
+c
2
y
2
+y
p
, where c
1
and c
2
are arbitrary
constants.
Except in the case where P(x) and Q(x) are constants (considered earlier
on p. 52), there is no general method of nding y
1
, but
1. given y
1
, there exists a method of nding y
2
.
2. given y
1
and y
2
, there exists a method of nding y
p
.
We will discuss (1) rst.
5.1 Reduction of Order
As mentioned earlier, we can nd a second solution from the rst. We rst
develop the tools we need.
57
58 CHAPTER 5. SECOND ORDER LINEAR EQUATIONS
Lemma 5.1
Let f(x) be a twice dierentiable function on a closed bounded interval
J. If x J : f(x) = 0 is innite, then there exists an x
0
J such that
f(x
0
) = 0 and f

(x
0
) = 0.
Proof. Let S = x J : f(x) = 0. If [S[ is innite, then by the Bolzano-
Weierstrass Theorem, S has an accumulation point x
0
, i.e., there exists an
x
0
S such that every interval about x
0
contains another point of S. Therefore,
by Rolles Theorem, every interval about x
0
contains a point where f

is 0.
Therefore, by continuity, f

(x
0
) = 0.


Corollary 5.2
Let y
1
(x) be a solution of y

+ P(x)y

+ Q(x)y = 0 on a closed bounded


interval J. If y
1
is not the zero solution, then y
1
(x) = 0 for at most nitely
many x in J.
Proof. Let V be the set of solutions of y

+ P(x)y

+ Q(x)y = 0 on J. Then
V is a two-dimensional vector space. So if y
1
, 0, then there exists a y
2
V
such that y
1
and y
2
are linearly independent. Therefore, W(y
1
, y
2
) ,= 0, i.e.,
W(y
1
, y
2
) is never zero (the Wronskian is either always 0 or never 0). But if
y
1
(x) = 0 for innitely many x in J, by Lemma 5.1, there exists an x
0
J such
that y
1
(x
0
) = 0 and y

1
(x
0
) = 0. Therefore,
W(y
1
, y
2
)(x
0
) =

y
1
(x
0
) y
2
(x
0
)
y

1
(x
0
) y

2
(x
0
)

0 y
2
(x
0
)
0 y

2
(x
0
)

= 0.
This is a contradiction. Therefore, y
1
(x) = 0 for at most nitely many x in J.
Let y
1
a solution to y

+ P(x)y

+ Q(x)y = 0 on a closed bounded interval


J. We want to nd a linearly independent second solution y
2
.
We break J into a nite collection of subintervals so that y
1
(x) ,= 0 on the
interior of each subinterval. We will describe how to nd the second solution y
2

Note that f

is necessarilybut certainly not sucientlycontinuous if f

is dieren-
tiable.
5.1. REDUCTION OF ORDER 59
on the interior of each interval. This will dene y
2
throughout J except for a
few isolated points.
Consider a subinterval I on which y
1
(x) is never zero. So y
1
(x) > 0 or
y
2
(x) < 0 throughout I. We have
W(y
1
, y
2
) =

y
1
y
2
y

1
y

.
If y
2
is also a solution to y

+ P(x)y

+ Q(x)y = 0, then W = Ae

R
P(x) dx
for
some constant A (Albels formula).

Suppose we choose A = 1. Then


y
1
y

2
y

1
y
2
= e

R
P(x) dx
,
where y
1
is known. This is a linear rst order equation for y
2
. In standard form,
we have
y

2

y

1
y
1
y
2
=
e

R
P(x) dx
y
1
,
where y
1
,= 0 throughout I. The integrating factor is
e
R

y

1
y
1
= e
ln(|y1|)+C
=
C

[y
1
[
,
where C

= 1, depending on the sign of y


1
. We use 1/y
1
. Therefore,
y

2
y
1

y
2
1
y
2
=
e

R
P(x) dx
y
2
1
,
y
2
y
1
=
_
e

R
P(x) dx
y
2
1
dx,
nally giving us
y
2
= y
1
_
e

R
P(x) dx
y
2
1
dx. (5.2a)
This denes y
2
on J except at a few isolated points where y
1
= 0. At those
points, we have
y
1
y

2
y

1
y
2
= e

R
P(x) dx
,
so
y
2
=
e

R
P(x) dx
y

1
. (5.2b)
This denes y
2
are the boundary points of the subintervals.

Dierent As correspond to dierent y


2
s.
60 CHAPTER 5. SECOND ORDER LINEAR EQUATIONS
Example 5.3. Observe that y = x is a solution of
y


x + 2
x
y

+
x + 2
x
2
y = 0
and solve
y


x + 2
x
y

+
x + 2
x
2
y = xe
x
.

Solution. Using the fact that y


1
= x, we have
y
2
= x
_
e

R
(
x+2
x
)dx
x
2
dx = x
_
e
R
(1+
2
x
)dx
x
2
dx = x
_
e
x+2 ln(|x|)
x
2
dx
= x
_
e
x
e
2 ln(|x|)
x
2
dx = x
_
e
x
[x[
2
x
2
dx = x
_
e
x
dx = xe
x
.
Therefore,
W =

x xe
x
1 e
x
+xe
x

= xe
x
+x
2
e
x
xe
x
= x
2
e
x
.
From this, we have
v

1
=
y
2
R
W
=
xe
x
xe
x
x
2
e
x
= e
x
= v
1
= e
x
,
v

2
=
y
1
R
W
=
xxe
x
x
2
e
x
= 1 = v
2
= x.
Therefore,
y
p
= v
1
y
1
+v
2
y
2
= xe
x
+x
2
e
x
,
and the general solution is
y = C
1
x +C
2
xe
x
xe
x
+x
2
e
x
= C
1
x +C

2
xe
x
+x
2
e
x
, x ,= 0,
where C
1
, C
2
, and C

2
are arbitrary constants.
5.2 Undetermined Coecients
Suppose that y
1
and y
2
are solutions of y

+ P(x)y

+ Q(x)y = 0. The general


method for nding y
p
is called variation of parameters. We will rst consider
another method, the method of undetermined coecients, which does not always
work, but it is simpler than variation of parameters in cases where it does work.
5.2. UNDETERMINED COEFFICIENTS 61
The method of undetermined coecients has two requirements, that
1. we have constant coecients, i.e., P(x) = p and Q(x) = q for some con-
stants p and q.
2. R(x) is nice.
Example 5.4. Solve y

y = e
2x
.
Solution. We have
2
1 = 0. So y
1
= e
x
and y
2
= e
x
. Since R(x) = e
2x
, we
look for a solution of the form y = Ae
2x
. Then
y

= 2Ae
2x
, y

= 4Ae
2x
.
Then according to the dierential equation, we have
4Ae
2x
Ae
2x
. .
3Ae
2x
= e
2x
= A =
1
3
.
Therefore, the general solution is
y = c
1
e
x
+c
2
e
x
+
1
3
e
2x
,
where c
1
and c
2
are arbitrary constants.
Undetermined coecients works when P(x) and Q(x) are constants and
R(x) is a sum of the terms of the forms shown in Table 5.1.
Table 5.1: The various forms of y
p
ought to be given a particular R(x).
R(x) Form of y
p
Cx
n
A
0
x
n
+A
1
x
n1
+ +A
n1
x +A
n
Ce
rx
Ae
rx
Ce
rx
cos(kx)
Ce
rx
sin(kx)
Ae
rx
cos(kx) +Be
rx
sin(kx)
The success of undetermined coecients depends on the derivative of R(x)
having some basic form as R(x). There is an easy way to deal with a complica-
tion: whenever a term in the trial y
p
is already part of the general solution to the
homogeneous equation, multiply it by x. For example, if we have y

y = e
x
,
as before, we have y
1
= e
x
and y
2
= e
2x
. Thus, we try y
p
= Axe
2x
and nd A.
62 CHAPTER 5. SECOND ORDER LINEAR EQUATIONS
Example 5.5. Solve y

+ 3y

+ 2y = e
x
3.
Solution. We have
2
+ 3 + 2 = 0. Factoring gives ( + 2) ( + 1) = 0, so
1, 2, and we have y
1
= e
x
and y
2
= e
2x
. Let y = Ae
x
+ B. Then
y

= Ae
x
and y

= Ae
x
, giving us
y

+ 3y

+ 2y = Ae
x
+ 3Ae
x
+ 2Ae
x
+ 2B = 6Ae
x
+ 2B.
Comparing coecients with e
x
3, we have
6A = 1 = A =
1
6
,
2B = 3 = B =
3
2
.
Therefore, y
p
= e
x
/6 3/2, and we have
y = c
1
e
x
+c
2
e
2x
+
e
x
6

3
2
,
where c
1
and c
2
are arbitrary constants.
Example 5.6. Solve y

+y = sin(x).
Solution. We have
2
+ 1 = 0, so = i, and it follows that y
1
= cos(x) and
y
2
= sin(x). Let y = Axsin(x) +Bxcos(x). Then
y

= Asin(x) +Axcos(x) +Bcos(x) Bxsin(x)


= (ABx) sin(x) + (Ax +B) cos(x),
y

= Bsin(x) + (ABx) cos(x) +Acos(x) (Ax +B) sin(x)


= (Ax 2B) sin(x) + (2ABx) cos(x).
Therefore,
y

+y = (Ax 2B) sin(x) + (2ABx) cos(x) +Axsin(x) +Bxcos(x).


Comparing coecients with sin(x), we see that
2B = 1 = B =
1
2
,
2A = 0 = A = 0.
5.2. UNDETERMINED COEFFICIENTS 63
Therefore, y
p
= xcos(x)/2, and the general solution is
y = c
1
sin(x) +c
2
cos(x)
xcos(x)
2
,
where c
1
and c
2
are arbitrary constants.
Example 5.7. Solve y

4y = x
2
.
Solution. We have
2
4 = 0, so = 2, and it follows that y
1
= e
2x
and
y
2
= e
2x
. Let y = Ax
2
+Bx+C. Then y

= 2Ax+B and y

= 2A. Therefore,
y

4y = 2A 4Ax
2
4B 4C = 4Ax
2
+ 0x + (2A 4B 4C) .
Comparing coecients with x
2
, we immediately see that
4A = 1 = A =
1
4
,
4B = 0 = B = 0,
2A 4C = 0 = C =
A
2
=
1/4
2
=
1
8
.
Therefore, y
p
= x
2
/4 1/8, and the general solution is
y = c
1
e
2x
+c
2
e
2x

x
2
4

1
8
,
where c
1
and c
2
are arbitrary constants.
Example 5.8. Solve y

= x.
Solution. We have
2
= 0, so 1, 0, and it follows that y
1
= e
x
and
y
2
= 1. Let y = Ax
2
+Bx. Then y

= 2Ax +B and y

= 2A. Therefore,
y

y = 2A 2Ax B.
Comparing coecients with x, we immediately see that
2A = 1 = A =
1
2
,
2AB = 0 = B = 1.
Therefore, y
p
= x
2
/2 x, and the general solution is
y = c
1
e
x
+c
2

x
2
2
x,
64 CHAPTER 5. SECOND ORDER LINEAR EQUATIONS
where c
1
and c
2
are arbitrary constants.
5.2.1 Shortcuts for Undetermined Coecients
Suppose we have
y

+ay

+by = R(x).
Let p() =
2
+a +b. We now consider two cases for R(x).
Case I: R(x) = ce
x
. It follows that y
p
= Ae
x
, so y

= Ae
x
and
y

= Ae
x
. Then
y

+ay

+by = ce
x
_

2
+a +b
_
Ae
x
= p()Ae
x
.
Therefore, A = c/p(), unless p() = c.
What if p() = 0? Then e
x
is a solution to the homogeneous equation. Let
y = Axe
x
. Then
y

= Ae
x
+Axe
x
,
y

= Ae
x
+Ae
x
+A
2
xe
x
= 2Ae
x
+A
2
xe
x
.
Thus,
y

+y

+y = A
__

2
x +ax +b
_
+ 2 +a

e
x
= A(p()xe
x
+p

()e
x
)
= Ap

()e
x
.
Therefore, A = c/p

(), unless p

() = 0 too.
If p() = 0 and p

() = 0, then e
x
and xe
x
both solve the homogeneous
equation. Letting y = Ax
2
e
x
, we have A = c/p

().
Case II: R(x) = ce
rx
cos(bx) or ce
rx
sin(bx). We use complex numbers. Set
= r +ib. Then
ce
x
= ce
rx
cos(bx) +ice
rx
sin(bx).
Let z = y +iw. Consider the equation
z

+az

+bz = ce
x
. (5.3)
5.2. UNDETERMINED COEFFICIENTS 65
The real part and imaginary parts are
y

+ay

+by = ce
rx
cos(bx), (5.4a)
w

+aw

+bw = ce
rx
sin(bx), (5.4b)
respectively. The solutions of Equation (5.3,5.4a,5.4b) are then
z =
ce
x
p()
, y = Re
_
ce
x
p()
_
, w = Im
_
ce
x
p()
_
,
respectively.
Example 5.9. Solve y

+ 8y

+ 12y = 3 cos(2x).
Solution. We have p() =
2
+ 8 + 12 = ( + 2) ( + 6). Therefore,
2, 6 and it follows that y
1
= e
2x
and y
2
= e
6x
. So
z =
3e
2ix
p(2i)
=
3e
2ix
(2i)
2
+ 16i + 12
=
3e
2ix
4 + 16i + 12
=
3e
2ix
8 + 16i
=
3
8

1
1 + 2i
e
2ix
=
3
8

1 2i
(1 2i) (1 + 2i)
e
2ix
=
3
8

1 2i
1 + 2i
e
2ix
=
3
40
(1 2i) e
2ix
=
3
40
(1 2i) (cos(2x) +i sin(2x))
=
3
40
[cos(2x) + 2 sin(2x) +i (2 cos(2x) + sin(2x))] .
Therefore,
Re(z) =
3
40
(cos(2x) + 2 sin(2x)) , Im(z) =
3
40
(2 cos(x) + sin(2x)) .
Hence, the solution is
y = c
1
e
2x
+c
2
e
6x
+
3
40
(cos(2x) + 2 sin(2x)) ,
where c
1
and c
2
are arbitrary constants.
We conclude that there is no shortcut when R(x) is a polynomial or any
multiple of a polynomial.
66 CHAPTER 5. SECOND ORDER LINEAR EQUATIONS
5.3 Variation of Parameters
Consider the equation
y

+P(x)y

+Q(x)y = R(x), (I)


where P(x) and Q(x) are not necessarily constant. Let y
1
and y
2
be linearly
independent solutions of the auxiliary equation
y

+P(x)y

+Q(x)y = 0. (H)
So for any constants C
1
and C
2
, C
1
y
1
+C
2
y
2
satises Equation (H).
The idea of variation of parameters is to look for functions v
1
(x) and v
2
(x)
such that v
1
y
1
+ v
2
y
2
satises Equation (I). Let y
1
and y
2
. If we want y to
be a solution to Equation (I), then substitution into Equation (I) will give one
condition on v
1
and v
2
.
Since there is only one condition to be satised and two functions to be
found, there is lots of choice, i.e., there are many pairs v
1
and v
2
which will
do. So we will arbitrarily decide only to look for v
1
and v
2
which also satisfy
v

1
y
1
+ v

2
y
2
= 0. We will see that this satisfaction condition simplies things.
So consider
y = v
1
y
1
+v
2
y
2
,
y

= v
1
y
1
+v
2
y

2
+v

1
y
1
+v

2
y
2
= v
1
y

1
+v
2
y

2
,
y

= v
1
y

1
+v
2
y

2
+v

1
y

1
+v

2
y

2
.
To solve Equation (I) means that
_
v
1
y

1
+v
2
y

2
+v

1
y

1
+v

2
y

2
+P(x)v
1
y

1
+P(x)v
2
y

2
+Q(x)v
1
y
1
+Q(x)v
2
y
2
_
= R(x),
_
_
_
v
1
(y

1
+P(x)y

1
+Q(x)y
1
)
+v
2
(y

2
+P(x)y

2
+Q(x)y
2
)
+v

1
y

1
+v

2
y

2
_
_
_ = R(x).
Therefore,
v

1
y

1
+v

2
y

2
= R(x),
v

1
y

1
+v

2
y

2
= 0,
5.3. VARIATION OF PARAMETERS 67
and it follows that
v

1
=
y
2
R
W
, v

2
=
y
1
R
W
,
where
W =

y
1
y
2
y

1
y

,= 0,
i.e., it is never zero regardless of x, since y
1
and y
2
are linearly independent.
Example 5.10. Solve y

5y

+ 6y = x
2
e
3x
.
Solution. The auxiliary equation is
y

5y

+ 6y = 0
2
5 + 6 = 0 = = 3, 2.
Therefore, the solutions are y
1
= e
3x
and y
2
= e
2x
. Now,
W =

y
1
y
2
y

1
y

e
3x
e
2x
3e
3x
2e
2x

= 2e
5x
3e
5x
= e
5x
.
Therefore,
v

1
=
y
2
R
W
=
e
2x
x
2
e
3x
e
5x
= x
2
= v
1
=
x
3
3
.
We can use any function v
1
with the right v

1
, so choose the constant of integra-
tion to be 0.
For v
2
, we have
v

2
=
y
1
R
W
=
e
3x
x
2
e
3x
e
5x
= x
2
e
x
,
so
v
2
=
_
x
2
e
x
dx
. .
by parts
= x
2
e
x
+ 2
_
xe
x
dx
. .
by parts
= x
2
e
x
+ 2xe
x
2
_
e
x
dx
= x
2
e
x
+ 2xe
x
2e
x
=
_
x
2
2x + 2
_
e
x
.
Finally,
y
p
=
x
3
3
e
3x

_
x
2
2x + 2
_
e
x
e
2x
= e
3x
_
x
3
3
x
2
+ 2x 2
_
,
68 CHAPTER 5. SECOND ORDER LINEAR EQUATIONS
and the general solution is
y = C
1
e
3x
+C
2
e
2x
+e
2x
_
x
3
3
x
2
+ 2x 2
_
= e
3x
_
x
3
3
x
2
+ 2x 2 +C
1
_
+C
2
e
2x
=
_
x
3
3
x
2
+ 2x +C
1
_
e
3x
+C
2
e
2x
,
where C
1
and C
2
are arbitrary constants.
Example 5.11. Solve y

+ 3y

+ 2y = sin(e
x
).
Solution. We immediately have
2
+ 3 + 2 = 0, which gives 1, 2.
Therefore, y
1
= e
x
and y
2
= e
2x
. Now,
W =

e
x
e
2x
e
x
2e
2x

= 2e
3x
+e
3x
= e
3x
.
Therefore,
v

1
=
e
2x
sin(e
x
)
e
3x
= e
x
sin(e
x
) = v
1
= cos(e
x
) .
For v
2
, we have
v

2
=
e
x
sin(e
x
)
e
3x
= e
2x
sin(e
x
) .
Therefore,
v
2
=
_
e
2x
sin(e
x
) dx
. .
substitute t = e
x
=
_
t sin(t) dt
. .
by parts
=
_
t cos(t) +
_
cos(t) dt
_
= (t cos(t) + sin(t))
= e
x
cos(e
x
) sin(e
x
) .
5.3. VARIATION OF PARAMETERS 69
Finally,
y
p
= v
1
y
1
+v
2
y
2
= e
x
cos(e
x
) +e
2x
(e
x
cos(e
x
) sin(e
x
))
= e
x
cos(e
x
) +e
x
cos(e
x
) e
2x
sin(e
x
)
= e
2x
sin(e
x
) ,
and the general solution is
y = C
1
e
x
+C
2
e
2x
e
2x
sin(e
x
) ,
where C
1
and C
2
are arbitrary constants.
70 CHAPTER 5. SECOND ORDER LINEAR EQUATIONS
Chapter 6
Applications of Second
Order Dierential
Equations
We now look specically at two applications of rst order des. We will see that
they turn out to be analogs to each other.
6.1 Motion of Object Hanging from a Spring
Figure 6.1 shows an object hanging from a spring with displacement d.
d
Figure 6.1: An object hanging from a spring with displacement d.
The force acting is gravity, spring force, air resistance, and any other external
71
72CHAPTER 6. APPLICATIONS OF SECONDORDER DIFFERENTIAL EQUATIONS
forces. Hookes Law states that
F
spring
= kd, (6.1)
where F is the spring force, d is the displacement from the spring to the natural
length, and k > 0 is the spring constant. As we lower the object, the spring
force increases. There is an equilibrium position at which the spring and grav-
ity cancel. Let x be the distance of the object from this equilibrium position
measured positively downwards. Let s be the displacement of the spring from
the natural length at equilibrium. Then it follows that ks = mg.
If the object is located at x, then the forces are
1. gravity.
2. spring.
3. air resistance.
4. external.
Considering just gravity and the spring force, we have

F
gravity
+F
spring
= mg k (x +s) = kx.
Suppose air resistance is proportional to velocity. Then
F
air
= rv, r > 0.
In other words, the resistance opposes motion, so the force has the opposite sign
of velocity. We will suppose that any external forces present depend only on
time but not on position, i.e., F
external
= F(t). Then we have
F
total
. .
ma
= kx rv +F(t),
so it follows that
m
d
2
x
dt
2
+r
dx
dt
+kx = F(t). (6.2)
Therefore, m
2
+r +k = 0 and we have
=
r

r
2
4mk
2m
=
r
2m

_
_
r
2m
_
2

k
m
.

Intuitively, when x = 0, they balance. Changing x creates a force in the opposite direction
attempting to move the object towards the equilibrium position.
6.1. MOTION OF OBJECT HANGING FROM A SPRING 73
We now have three cases we need to consider.
Case 1: 4mk > r
2
. Let = r/2m, let =
_
k/m, and let
=
_
k
m

_
r
2m
_
2
=
_

2
.
Then = i and
x = e
t
(C
1
cos(t) +C
2
sin(t)) = Ae
t
cos(t ), (6.3)
where C
1
and C
2
are arbitrary constants. Figure 6.2a shows how this solution
qualitatively looks like.
Damped vibrations is the common case. The oscillations die out in time,
where the period is 2/. If r = 0, then = 0. In such a case, there is no
resistance, and it oscillates forever.
Case 2: 4mk = r
2
. In such a case, we simply have
x = e

r
2m
t
(C
1
+C
2
t) , (6.4)
where C
1
and C
2
are arbitrary constants. Figure 6.2b shows how this solution
qualitatively looks like.
Case 3: 4mk < r
2
. Let a = r/2m and
b =
_
_
r
2m
_

k
m
.
Then the solution is
x = C
1
e
(ab)t
+C
2
e
(a+b)t
, (6.5)
where C
1
and C
2
are arbitrary constants. This is the overdamped case. The
resistance is so great that the object does not oscillate (imagine everything
immersed in molasses). It will just gradually return to the equilibrium position.
Figure 6.2c shows how this solution qualitatively looks like.
Consider Equation (6.2) again. We now consider special cases of F(t), where
F(t) =
_
_
_
F
0
cos(t),
F
0
sin(t).
74CHAPTER 6. APPLICATIONS OF SECONDORDER DIFFERENTIAL EQUATIONS
t
(a) 4mk > r
2
t
(b) 4mk = r
2
t
(c) 4mk < r
2
Figure 6.2: The various cases for a spring-mass system.
Use undetermined coecients to get the solution of the form
x
p
= Bcos(t) +C sin(t).
For F(t) = F
0
cos(t), we get
x
p
=
F
0
(k m
2
)
2
+ (r)
2
[(k m) cos(t) +r sin(t)] = Acos(t ) ,
where
A =
F
0
_
(k m
2
)
2
+ (r)
2
,
cos() =
k m
2
_
(k m
2
)
2
+ (r)
2
,
sin() =
r
_
(k m
2
)
2
+ (r)
2
.
For F(t) = F
0
sin(t), we get x
p
= Asin(t ), where A and is as above.
6.2. ELECTRICAL CIRCUITS 75
In any solution
e
t
() +Asin(t ) ,
the rst term eventually dies out, leaving Asin(t ) as the steady state so-
lution.
Here we consider a phenomenon known as resonance. The above assumes
that cos(t) and sin(t) are not part of the solution to the homogeneous equa-
tion. Consider now
m
d
2
x
dt
2
+kx = F
0
cos(t),
where =
_
k/m (the external force frequency coincides with the natural
frequency). Then the solution is

x
p
=
F
0
2

km
t sin(t).
Another phenomenon we now consider is beats. Consider r = 0 with near
but not equal to
_
k/m. Let =
_
k/m. Then the solution, with x(0) = 0 =
x

(0), is
x =
F
0
m(
2

2
)
cos(t t)
=
2F
0
( +) ( )
sin
_
( ) t
2
_
sin
_
( + ) t
2
_

F
0
2m
sin(t) sin(t),
where = ( ) /2.
6.2 Electrical Circuits
Consider the electrical circuit shown in Figure 6.3.
Let Q(t) be the charge in the capacitor at time t (Coulombs). Then dT/dt
is called the current, denoted I. The battery produces a voltage (potential
dierence) resulting in current I when the switch is closed. The resistance R
results in a voltage drop of RI. The coil of wire (inductor) produces a magnetic
eld resisting change in the current. The voltage drop created is L(dI/dt). The

The amplitude of the vibrations is


F
0
2

km
t,
which increases with time.
76CHAPTER 6. APPLICATIONS OF SECONDORDER DIFFERENTIAL EQUATIONS
battery
resistor
switch
capacitor
inductor I amperes
+
Figure 6.3: An electrical circuit, where resistance is measured in Ohms, capac-
itance is measured in Farads, and the inductance is measured in Henrys.
capacitor produces a voltage drop of Q/C. Unless R is too large, the capacitor
will create sine and cosine solutions and, thus, an alternating ow of current.
Kirkhos Law states that the sum of the voltage changes around a circuit is
zero, so
E(t) +RI +L
dI
dt
+
Q
C
,
so we have
L
d
2
Q
dt
2
+R
dQ
dt
+
Q
C
= E(t). (6.6)
The solution is the same as in the case of a spring. The analogs are shown in
Table 6.1.
Table 6.1: Analogous terms between a spring-mass system and an electrical
circuit.
Circuit Spring
Charge Q Displacement x
Inductance I Mass m
Resistance R Friction r
Capacitance inverse 1/C Spring constant k
Voltage generated by battery E(t) External force F(t)
Amperes I Velocity v
An example of this application is choosing a station on an old-fashioned
radio. The radio has a capacitor with two charged metal bars. When the
user turns the tuning dial, it changes the distance between the bars, which in
6.2. ELECTRICAL CIRCUITS 77
turn changes the capacitance C. This changes the frequency of the solution
of the homogeneous equation. When it agrees with the frequency E(t) of some
incoming signal, resonance results so that the amplitude of the current produced
from this signal is much greater than any other.
78CHAPTER 6. APPLICATIONS OF SECONDORDER DIFFERENTIAL EQUATIONS
Chapter 7
Higher Order Linear
Dierential Equations
In this section we generalize some of the techniques for solving second order
linear equations discussed in Chapter 5 so that they apply to linear equations
of higher order. Recall from Chapter 4 that an nth order linear dierential
equation has the form
y
(n)
+P
1
(x)y
(n1)
+P
2
(x)y
(n2)
+ +P
n
(x)y = Q(x). (7.1)
The general solution is
y = C
1
y
1
+C
2
y
2
+ +C
n
y
n
+y
p
,
where y
1
, y
2
, . . . , y
n
are linearly independent solutions to the auxiliary homoge-
neous equation and y
p
is a solution of Equation (7.1).
So given y
1
, y
2
, . . . , y
n
, how do we nd y
p
? We again consider the two meth-
ods we have looked at in 5.2 and 5.3 (p. 60 and p. 66, respectively).
7.1 Undetermined Coecients
If P
1
(x), P
2
(x), . . . , P
n
(x) are constants and R(x) is nice, we can use undeter-
mined coecients.
Example 7.1. Solve y
(4)
y = 2e
2e
+ 3e
x
.
79
80CHAPTER 7. HIGHER ORDER LINEAR DIFFERENTIAL EQUATIONS
Solution. We have
p() =
4
1 = ( 1) ( + 1)
_

2
+ 1
_
.
Therefore,
y
1
= e
x
, y
2
= e
x
, y
3
= cos(x), y
4
= sin(x).
Let y
p
= Ae
3x
+Bxe
3x
. Then by the short cut method (5.2.1, p. 64), we have
A =
2
p(2)
=
2
16 1
=
2
15
,
B =
3
p

(1)
=
3
4 1
3
=
3
4
.
Therefore, the general solution is
y = C
1
e
x
+C
2
e
x
+C
3
cos(x) +C
4
sin(x) +
2
15
e
2x
+
3
4
xe
x
.

7.2 Variation of Parameters


Suppose that y
1
, y
2
, . . . , y
n
are solutions of the auxiliary homogeneous equation.
We look for solutions of Equation (7.1) of the form
y = v
1
y
1
+v
2
y
2
+ +v
n
y
n
,
where v
1
, v
2
, . . . , v
n
are functions of x. One condition on v
1
, v
2
, . . . , v
n
is given
by substituting into Equation (7.1). We choose n 1 conditions to be
v

1
y
1
+v

2
y
2
+ +v

n
y
n
= 0, (1)
v

1
y

1
+v

1
y

2
+ +v

n
y

n
= 0, (2)
.
.
.
.
.
.
v

1
y
(n2)
1
+v

2
y
(n2)
2
+ +v

n
y
(n2)
n
= 0. (n 1)
7.2. VARIATION OF PARAMETERS 81
If conditions (1), . . . , (n 1) are satised, then
y

=
n

i=1
(v
i
y
i
)

=
n

i=1
v
i
y

i
+
n

i=1
v

i
y
i
=
n

i=1
v
i
y

i
,
y

=
n

i=1
(v
i
y

i
)

=
n

i=1
v
i
y

i
+
n

i=1
v

i
y

i
=
n

i=1
v
i
y

i
,
.
.
.
y
(n1)
=
n

i=1
v
i
y
(n1)
i
+
n

i=1
v

i
y
(n2)
=
n

i=1
v
i
y
(n1)
i
,
y
(n)
=
n

i=1
v
i
y
(n)
i
+
n

i=1
v

i
y
(n1)
i
.
Therefore,
y
(n)
+P
1
(x)y
(n1)
+ +P
n
(x)y =
n

i=1
v
i
y
(n)
i
+
n

i=1
v

i
y
(n1)
i
+P
1
(x)
n

i=1
v
i
y
(n1)
i
+ +P
n
(x)
n

i=1
v
i
y
i
=
n

i=1
v

i
y
(n1)
i
+
n

i=1
v
i
_
y
(n)
i
+P
1
(x)y
(n1)
i
+
+P
n
(x)y
i
_
=
n

i=1
v

i
y
(n1)
i
+

i=1
v
i
0
=
n

i=1
v

i
y
(n1)
i
.
Therefore, Equation (7.1) becomes the rst condition
n

i=1
v

i
y
(n1)
i
y = R(x). (n)
82CHAPTER 7. HIGHER ORDER LINEAR DIFFERENTIAL EQUATIONS
Conditions (1), . . . , (n 1) can be written as
M
_

_
v

1
v

2
.
.
.
v

n
_

_
=
_

_
0
0
.
.
.
R(x)
_

_
,
where
M =
_

_
y
1
y
n
y

1
y

n
.
.
.
.
.
.
.
.
.
y
(n1)
1
y
(n1)
n
_

_
.
Therefore, [M[ = W ,= 0.
7.3 Substitutions: Eulers Equation
Consider the equation
x
n
y
(n)
+a
1
x
n1
y
(n1)
+ +a
n1
y

+a
n
y = R(x), (7.2)
where a
1
, a
2
, . . . , a
n
are constants. Consider x > 0. Let x = e
u
so that u = ln(x)
(for x < 0, we must use u = ln(x)). Then
dy
dx
=
dy
du
du
dx
=
dy
du
1
x
,
d
2
y
dx
2
=
d
2
y
du
2
1
x
2

dy
du
1
x
2
=
_
d
2
y
du
2

dy
du
_
1
x
2
,
d
3
y
dx
3
=
_
d
3
ydx
3

d
2
y
du
2
_
1
x
3
2
_
d
2
y
du
2

dy
du
_
1
x
3
=
_
d
3
y
du
3
3
d
2
y
du
2
+ 2
dy
du
_
1
x
3
,
.
.
.
d
n
y
dx
n
=
_
d
n
y
du
n
+C
1
d
n1
y
du
n1
+ +C
n
dy
du
_
1
x
n
7.3. SUBSTITUTIONS: EULERS EQUATION 83
for some constants C
1
, C
2
, . . . , C
n
. Therefore, Equation (7.2) becomes
d
n
y
du
n
+b
1
d
n1
y
du
n1
+ +b
n1
dy
du
+b
n
y = R(e
u
)
for some constants b
1
, b
2
, . . . , b
n
.
Example 7.2. Solve x
2
y

+xy

y = x
3
for x > 0.
Solution. Let x = e
u
so that u = ln(x). Then
dy
dx
=
dy
du
du
dx
=
dy
du
1
x
,
d
2
y
dx
2
=
d
2
y
du
2
1
x
2

dy
du
1
x
2
=
_
d
2
y
du
2

dy
du
_
1
x
2
.
Therefore,
_
d
2
y
du
2

dy
du
_
+
dy
du
y = e
3u
,
d
2
u
du
2
y = e
3u
.
Considering the homogeneous case, we have
2
1 = 0 = 1. Therefore,
y
1
= e
u
and y
2
= e
u
. Using undetermined coecients, let y
p
= Ae
3u
. Then
y

= 3Ae
3u
and y

= 9Ae
3u
. Substitution gives
9Ae
3u
Ae
3u
. .
8Ae
3u
= e
3u
.
We can immediately see that A = 1/8. Therefore, the general solution is
y = C
1
e
u
+C
2
e
u
+
1
8
e
3u
= C
1
e
ln(x)
+C
2
e
ln(x)
+
1
8
e
3 ln(x)
= C
1
x +
C
2
x
+
x
3
8
,
where C
1
and C
2
are arbitrary constants.
We can also consider a direct method. Solving
d
2
y
dx
2
+p
dy
du
+qy = 0
84CHAPTER 7. HIGHER ORDER LINEAR DIFFERENTIAL EQUATIONS
involves looking for solutions of the form y = e
u
= (e
u
)

and nding . Since


x = e
u
, we could directly look for solutions of the form x

and nd the right .


With this in mind and considering Example 7.2 again, we have
y = x

,
y

= x
1
,
y

= ( 1) x
2
.
Therefore,
x
2
y

+xy

y = 0 = ( 1) x

+x

= 0
= ( 1) + 1 = 0
=
2
1 = 0
= = 1,
as before.
Chapter 8
Power Series Solutions to
Linear Dierential
Equations
8.1 Introduction
Consider the equation
y
(n)
+P
1
(x)y
(n1)
+ +P
n
(x)y = R(x). (8.1)
Given solutions y
1
, . . . , y
n
to the homogeneous equation
y
(n)
+P
1
(x)y + +P
n
y = 0,
we know how to solve Equation (8.1), but so far there is no method of nding
y
1
, . . . , y
n
unless P
1
, . . . , P
n
are constants. In general, there is no method of
nding y
1
, . . . , y
n
exactly, but we can nd the power series approximation.
Example 8.1. Solve y

xy

2y = 0.
Solution. Let
y = a
0
+a
1
x +a
2
x
2
+ +a
n
x
n
+ =

n=0
a
n
x
n
.
85
86CHAPTER 8. POWER SERIES SOLUTIONS TOLINEAR DIFFERENTIAL EQUATIONS
Then it follows that
y

= a
1
+ 2a
2
x + +na
n
x
n1
+ (n + 1) a
n+1
x
n
+
=

n=1
na
n
x
n1
=

k=0
(k + 1) a
k+1
x
k
,
where k = n 1. Relabeling, we have
y

n=0
(n + 1) a
n+1
x
n
.
By the same procedure, we have
y

n=0
n(n + 1) a
n+1
x
n1
=

n=0
(n + 1) (n + 2) a
n+2
x
n
.
Now considering y

xy

2y = 0, we have
y

..

n=0
(n + 1) (n + 2) a
n+2
x
n
x
y

..

n=0
na
n
x
n1
2
y
..

n=0
a
n
x
n
= 0,

n=0
[(n + 1) (n + 2) a
n+2
na
n
2a
n
] x
n
= 0.
This gives us the recurrence relation
(n + 1) (n + 2) a
n+2
na
n
2a
n
= 0,
(n + 1) (n + 2) a
n+2
(n + 2) a
n
= 0,
(n + 1) a
n+2
= a
n
,
a
n+2
=
a
n
n + 1
for all n 0. Therefore,
a
2
=
a
0
1
, a
4
=
a
2
3
=
a
0
1 3
, a
6
=
a
4
5
=
a
0
1 3 5
, . . . , a
2n
=
a
0

n
k=1
(2k 1)
,
a
3
=
a
1
2
, a
5
=
a
3
4
=
a
1
2 4
, a
7
=
a
5
6
=
a
1
2 4 6
, . . . , a
2n+1
=
a
1

n
k=1
2k
.
8.1. INTRODUCTION 87
Therefore,
y = a
0
+a
1
x +a
2
x
2
+a
3
x
3
+ +a
n
x
n
+
= a
0
+a
1
x +a
0
x
2
+
a
1
2
x
3
+ +
a
0

n
k=1
(2k 1)
x
2n
+
a
1

n
k=1
2k
x
2n+1
+
= a
0
_
1 +x
2
+ +
x
2n

n
k=1
(2k 1)
+
_
+a
1
_
x +
x
3
2
+ +
x
2n+1

n
k=1
2k
+
_
= a
0

n=0
x
2n

n
k=1
(2k 1)
+a
1

n=0
x
2n+1
2
n
n!
.
That is, we have y = C
1
y
1
+C
2
y
2
, where C
1
= a
0
and C
2
= a
1
such that
y
1
=

n=0
x
2n

n
k=1
(2k 1)
, y
2
=

n=0
x
2n+1
2
n
n!
.
Can we recognize y
1
and y
2
as elementary functions? In general, no. But in
this case, we have
y
2
=

n=0
x
2n+1
2
n
n!
= x

n=0
x
2n
2
n
n!
= x

n=0
_
x
2
_
n
2
n
n!
= x

n=0
_
x
2
/2
_
n
n!
= xe
x
2
/2
.
Having one solution in closed form, we can try to nd another (see 5.1, p. 57)
by
y
1
= y
2
_
e

R
P(x) dx
y
2
2
= xe
x
2
/2
_
e

R
P(x) dx
x
2
e
x
2
dx
= xe
x
2
/2
_
e
x
2
/2
x
2
e
x
2
dx = xe
x
2
/2
_
1
x
2
e
x
2
/2
dx,
which is not an elementary function.
88CHAPTER 8. POWER SERIES SOLUTIONS TOLINEAR DIFFERENTIAL EQUATIONS
8.2 Background Knowledge Concerning Power
Series
Theorem 8.2
Consider the power series

n=0
a
n
(z p)
n
.
1. There exists an R such that 0 R < called the radius of con-
vergence such that

n=0
a
n
(z p)
n
converges for [z p[ < R and
diverges for [z p[ > R. The radius of convergence is given by
R = lim
n

a
n
a
n+1

(8.2)
if the limit exists (ratio test).
2. The function f(z) =

n=0
a
n
z
n
dened for [z p[ < R can be dier-
entiated and integrated term-wise within its radius of convergence.
Proof. Proof is given in MATA37 or MATB43.
Remark. The series may or may not converge when [z p[ = R.
Using (2) of Theorem 8.2, f

(z) exists for [z p[ < R with f

(z) =

n=0
na
n
z
n1
.
Similarly,
f(z) =

n=0
a
n
(z p)
n
, z = p = a
0
= f(p),
f

(z) =

n=0
na
n
(z p)
n1
, z = p = a
1
= f

(p),
f

(z) =

n=0
n(n 1) a
n
(z p)
n2
, z = p = a
2
=
f

(p)
2
,
.
.
.
f
(n)
(z) =

n=0
n! a
n
, z = p = a
n
=
f
(n)
(p)
n!
.
8.3. ANALYTIC EQUATIONS 89
Conversely, if f is innitely dierentiable at p, we dene a
n
= f
(n)
(p)/n!. Then

n=0
a
n
(z p)
n
is called the Taylor series of f about p.
From the above, if any power series converges to f near p, it must be its
Taylor series. But
1. the Taylor series might not converge.
2. if it does converge, it does not have to converge to f.
Denition (Analytic function, ordinary/singular point)
A function f is called analytic at p if

n=0
a
n
(z p)
n
converges to f(z) in
some neighbourhood of p (i.e., if there exists an r > 0 such that it converges for
[z p[ < r). If f is analytic at p, then p is called an ordinary point of f. If f
is not analytic at p, then p is called a singular point of f.
Theorem 8.3 (Theorem from complex analysis)
Let f(z) be analytic at p. Then the radius of convergence of the Taylor
series of f at p equals the distance from p to the closest singularity of f in
the complex plane.
Proof. Proof is given in MATC34.
Example 8.4. Consider f(x) = 1/
_
x
2
+ 1
_
. The singularities are i. There-
fore, the radius of convergence of the Taylor series about the origin is 1, with
the series given by
f(x) =

n=0
(1)
n
x
2n
.
The radius of convergence of the Taylor series about x = 1 is

2.
8.3 Analytic Equations
Consider y

+ P(x)y

+ Q(x)y = 0. If both P(x) and Q(x) are analytic at p,


then p is called an ordinary point of the de. Otherwise, p is called a singular
90CHAPTER 8. POWER SERIES SOLUTIONS TOLINEAR DIFFERENTIAL EQUATIONS
point or a singularity of the de. When looking at a de, do not forget to write
it in the standard form. For example,
x
2
y

+ 5y

+xy = 0
has a singularity at x = 0, since the standard form is
y

+
_
5
x
2
_
y

+
_
1
x
_
y = 0.
Theorem 8.5
Let x = p be an ordinary point of y

+ P(x)y

+ Q(x)y = 0. Let R be the


distance from p to the closest singular point of the de in the complex plane.
Then the de has two series y
1
(x) =

n=0
a
n
x
n
and y
2
(x) =

n=0
b
n
x
n
which converge to linearly independent solutions to the de on the interval
[x p[ < R.
Remark. The content of the theorem is that the solutions are analytic within
the specied interval. The fact that solutions exist on domains containing at
least this interval is a consequence of theorems in Chapter 10.
Proof. The proof is given in Chapter 10.
Example 8.6. Consider
_
x
2
+ 1
_
y

+ 2xy

+ 2y = 0. Dividing by x
2
+ 1, we
have
y

+
3x
x
2
+ 1
y

+
2
x
2
+ 1
y = 0.
The singularities are i. If p = 0, then r = [i 0[ = 1. Therefore, analytic
solutions about x = 0 exist for [x[ < 1. If p = 1, then r = [i 1[ =

2. There-
fore, there exist two linearly independent series of the form

n=0
a
n
(x 1)
n
converging to the solutions for at least [x 2[ <

2.
Note that the existence and uniqueness theorem (see Chapter 10) guarantees
solutions from to (since x
2
+ 1 has no real zeros), but only in a nite
subinterval can we guarantee that the solutions are expressible as convergent
power series.
Example 8.7. Consider y

xy

2y = 0. It has no singularities. Therefore,


a series solution converging on R exists about any point.
8.4. POWER SERIES SOLUTIONS: LEVELS OF SUCCESS 91
8.4 Power Series Solutions: Levels of Success
When using power series techniques to solve linear dierential equations, ideally
one would like to have a closed form solution for the nal answer, but most often
one has to settle for much less. There are various levels of success, listed below,
and one would like to get as far down the list as one can.
For simplicity, we will assume that we have chosen to expand about the
point x = 0, although the same considerations apply to x = c for any c.
Suppose 0 is an ordinary point of the equation L(x)y

+M(x)y

+N(x)y = 0
and let y =

n=0
a
n
x
n
be a solution. One might hope to:
1. nd the coecients a
n
for a few small values of n; (For example, nd the
coecients as far as a
5
which thereby determines the 5th degree Taylor
approximation to y(x));
2. nd a recursion formula which for any n gives a
n+2
in terms of a
n1
and a
n
;
3. nd a general formula a
n
in terms of n;
4. recognize the resulting series to get a closed form formula for y(x).
In general, level (1) can always be achieved. If L(x), M(x), and N(x)
are polynomials, then level (2) can be achieved. Having achieved level (2), it
is sometimes possible to get to level (3) but not always. Cases like the one
in Example 8.1 where one can achieve level (4) are rare and usually beyond
expectation.
8.5 Level 1: Finding a nite number of coe-
cients
Suppose c is an ordinary point of Equation (8.3). One approach is to nding
power series solutions is as follows. Expand P(x) and Q(x) into Taylor series
P(x) =

n=0
p
n
(x c)
n
, Q(x) =

n=0
q
n
(x c)
n
.
and set y =

n=0
a
n
(x c)
n
solves the equation. Then
y

n=0
na
n
(x c)
n1
, y

n=0
n(n 1) a
n
x
n2
,
92CHAPTER 8. POWER SERIES SOLUTIONS TOLINEAR DIFFERENTIAL EQUATIONS
and Equation (8.3) becomes
y

..

n=0
n(n 1) a
n
x
n2
+
P(x)
..
_

n=0
p
n
(x c)
n
_
y

..
_

n=0
na
n
(x c)
n1
_
+
_

n=0
q
n
(x c)
n
_
. .
Q(x)
_

n=0
a
n
(x c)
n
_
. .
y
= 0.
Since p
n
s and q
n
s are known (as many as desired can be computed from deriva-
tives of P(x) and Q(x)), in theory we can inductively nd the a
n
s one after
another.
In the special case where P(x) and Q(x) are polynomials, we can get a
recursion formula by writing a
n+2
in terms of a
n
and a
n1
(level 2) and, under
favourable circumstances, we can nd the general formula for a
n
(level 3) as in
Example 8.1. But in general, computation gets impossibly messy after a while
and, although a few a
n
s can be found, the general formula cannot.
But if you only want a few a
n
s, there is an easier way. We have a
n
=
y
(n)
(c)/n!, so
a
2
=
y

(c)
2!
=
P(x)y

(c) +Q(c)y(c)
2
=
a
1
P(c) a
0
Q(c)
2
,
a
3
=
y

(c)
3!
=
1
6
(P

(c)y

(c) P(c)y

(c)y(c) Q(c)y

(c))
=
1
6
_
P

(c)a
1
+
a
1
P(c)
2
+a
0
P(c)Q(c)
2
a
0
Q

(c) a
1
Q(c)
_
,
a
4
= ,
and so on.
Example 8.8. Compute the Taylor expansion about 0 as far as degree 4 for
the solution of y

e
7x
y

+xy = 0, which satises y(0) = 2 and y

(0) = 1.
Solution. Rearranging the equation, we have y

= e
5x
y

xy. Therefore
y

= e
5x
y

+ 5e
5x
y

xy

y
y
(4)
= e
5x
y

+ 5e
5x
y

+ 5e
5x
y

xy

= e
5x
y

+ 10e
5x
y

xy

2y

.
8.5. LEVEL 1: FINDING A FINITE NUMBER OF COEFFICIENTS 93
Substituting gives
y

(0) = e
0
y

(0) 0 = 1
y

(0) = e
0
y

(0) + 5e
0
y

(0) 0 y(0) = 1 + 5 2 = 4
y
(4)
(0) = e
0
y

(0) + 10e
0
y

(0) 0 2y

(0) = 4 + 10 2 = 12
Therefore,
y = 2 +x +
x
2
2!
+
4x
3
3!
+
12x
4
4!
+
= 2 +x +
1
2
x
2
+
2
3
x
3

1
2
x
4
+ .
Example 8.9. Find the series solution for y

+ (x + 3) y

2y = 0, where
y(0) = 1 and y

(0) = 2.
Solution. Rearranging the equation, we have y

= (x + 3) y

+ 2y. Then
y

(0) = (0 + 3) y

(0) + 2y(0) = 3 2 + 2 1 = 4,
y

= y

(x + 3) y

+ 2y

= (x + 3) y

+y

,
y

(0) = 3y

(0) +y

(0) = (3) (4) + 2 1 = 12 + 2 = 14,


y
(4)
= = 34.
Therefore,
y = 1 + 2x
4x
2
2!
+
14x
3
3!

34x
4
4!
+
= 1 + 2x 2x
2
+
7
3
x
3

17
12
x
4
+ .
In this case, since the coecients were polynomials, it would have been possible
to achieve level 2, (although the question asked only for a level 1 solution).
However even had we worked out the recursion formula (level 2) it would have
been too messy to use it to nd a general formula for a
n
.
94CHAPTER 8. POWER SERIES SOLUTIONS TOLINEAR DIFFERENTIAL EQUATIONS
8.6 Level 2: Finding the recursion relation
Example 8.10. Consider
_
x
2
5x + 6
_
y

5y

2y = 0, where y(0) = 1 and


y

(0) = 1. Find the series solution as far as x


3
and determine the lower bound
on its radius of convergence from the recursion formula.
Solution. We have
2y =

n=0
2a
n
x
n
,
5y

n=0
5na
n
x
n1
=

n=0
5 (n + 1) a
n+1
x
n
=

n=0
5 (n + 1) a
n+1
x
n
,
6y

n=0
6n(n 1) a
n
x
n2
=

n=0
6 (n + 2) (n + 1) a
n+2
x
n
,
5xy

n=0
5n(n 1) a
n
x
n1
=

n=0
5 (n + 1) na
n+1
x
n
,
x
2
y

n=0
n(n 1) a
n
x
n
.
Therefore,

n=0
_
_
_
_
_
n(n 1) a
n
5 (n + 1) na
n+1
+6 (n + 2) (n + 1) a
n+2
5 (n + 1) a
n+1
2a
n
_
_
_
_
_
x
n
=

n=0
_
_
_
6 (n + 2) (n + 1) a
n+2
5 (n + 1) (n + 1) a
n
+
_
n
2
n 2
_
a
n
_
_
_x
n
,
so it follows that
6 (n + 2) (n + 1) a
n+2
5 (n + 1)
2
a
n+1
+ (n + 1) (n 2) a
n
= 0,
a
n+2
=
5
6

n + 1
n + 2
a
n+1

1
6

n 2
n + 2
a
n
.
With a
0
= y(0) = 1 and a
1
= y

(0) = 1, we have
a
2
=
5
6

1
2
a
1

1
6

2
2
a
0
=
5
12
+
2
12
=
7
12
,
a
3
=
5
6

2
3

1
6

1
3
a
1
=
5
9

7
12
+
1
6

1
3
=
35
108
+
6
108
=
41
108
.
Therefore,
y = 1 +x +
7
12
x
2
+
41
108
x
3
+ .
8.6. LEVEL 2: FINDING THE RECURSION RELATION 95
Notice that although we need to write our equation in standard form (in
this case dividing by x
2
5x + 6 to determine that 0 is an ordinary point of
the equation), it is not necessary to use standard form when computing the
recursion relation. Instead we want to use a form in which our coecients are
polynomials. Although our recursion relation is too messy for us to achieve
nd a general formula for a
n
(level 3) there are nevertheless some advantages
in computing it. In particular, we can now proceed to use it to nd the radius
of convergence.
To determine the radius of convergence, we use the ratio test R = 1/ [L[,
where L = lim
n
(a
n+1
/a
n
), provided that the limit exists. It is hard to prove
that the limit exists, but assuming that
a
n+2
a
n+1
=
5
6

n + 1
n + 2

1
6

n 2
n + 2

a
n
a
n+1
,
taking the limit gives
L =
5
6
1
1
6

1
L
=
5
6

1
6

1
L
.
So 6L
2
= 5L 1, and solving for L gives L = 1/2 or L = 1/3. So R = 2 or
R = 3. The worst case is when R = 2. Therefore, the radius of convergence is
at least 2. Theorem 8.3 also gives this immediately.
Example 8.11. Consider
_
x
2
5x + 6
_
y

5y

2y = 0. Find two linearly


independent series solutions y
1
and y
2
as far as x
3
.
Solution. Let y
1
(x) be the solution satisfying y
1
(0) = 1; y

1
(0) = 0 and let y
2
(x)
be the solution satisfying y
2
(0) = 0; y

2
(0) = 1. Then y
1
(x) and y
2
(x) will be
linearly independent solutions. We already computed the recursion relation in
Example 8.10. For y
1
(x), we have a
0
= y(0) = 1 and a
1
= y

(0) = 0. Thus,
a
2
=
5
6

1
2
a
1

1
6
_

2
2
_
a
0
=
1
6
,
a
3
=
5
6

2
3
a
2

1
6

1
3
a
1
=
5
54
.
Therefore,
y
1
= 1 +
x
2
6
+
5
54
x
3
+ .
96CHAPTER 8. POWER SERIES SOLUTIONS TOLINEAR DIFFERENTIAL EQUATIONS
For y
2
, we have a
0
= y(0) = 0 and a
1
= y

(0) = 1, so
a
2
=
5
6

1
2
a
1

2
2
a
0
=
5
12
,
a
3
=
5
6

2
3
a
2

_

1
3
_
a
1
=
5
6

2
3

5
12
+
1
6 3
=
25
108
+
6
108
=
31
108
.
Therefore,
y
2
= x +
5
12
x
2
+
61
108
x
3
+ .
With
y = 1 +x +
7
12
x
2
+
41
108
x
3
+ ,
Looking at the initial values, the solution y(x) of Example 8.10 ought to be
given by y = y
1
+y
2
, and comparing are answers we see that they are consistent
with this equation.
Summarizing: Let p be a regular point of the equation
y

+P(x)y

+Q(x)y = 0, (8.3)
Let D be the distance from p to the closest singularities of P or Q in the complex
plane. Then Equation (8.3) has two linearly independent solutions expressible
as a convergent power series

n=0
a
n
(x p)
n
with [x p[ < D. Theorem 8.5
guarantees that the radius of convergence is at least D, but it might be larger.
Given a
0
= y(p) and a
1
= y

(p), to nd a
n
, we have a
n
= y
(n)
(p)/n!. We
can nd as many a
n
s as desired, but in general we cannot nd the formula.
Since we have y

= P(x)y

Q(x)y, it can be dierentiated to give y

(p)
and higher derivatives can be found by further dierentiation. Substituting the
values at p gives formulas for the a
n
s.
In the special case where P(x) and Q(x) are polynomials, we can instead
nd a recurrence relation. From the recurrence relation there is a chance to get
the formula for a
n
(if so, then there is a chance to actually recognize the series).
Even without the formula, we can determine the radius of convergence from the
recurrence relation (even if P(x) and Q(x) are not polynomials, we could at
least try to get a recurrence relation by expanding them into Taylor series,but
in practice this is not likely to be successful).
8.7. SOLUTIONS NEAR A SINGULAR POINT 97
8.7 Solutions Near a Singular Point
A singularity a of y

+ P(x)y

+ Q(x)y = 0 is called a regular singularity if


(x a) P(x) and (x a)
2
Q(x) are analytic at x = a. To simplify notation, we
assume that a = 0 (we can always change the variable by v = x a to move
the singularity to the origin). We assume that x > 0. For x < 0, we substitute
t = x and apply below to t.
Our primary goal in this section is to consider the Method of Frobenius. We
look for solutions of the form
y = x

(a
0
+a
1
x + +a
n
x
n
+ ) = a
0
x

+a
1
x
+1
+ +a
n
x
+n
+
for some , where a
0
,= 0. Then
y

= a
1
x
1
+a
1
( + 1) x

+ +a
n
( +n) x
+n1
+ ,
y

= a
1
( 1) x
2
+a
1
( + 1) x
1
+
+a
n
( +n) ( +n 1) x
+n2
+ ,
xP(x) = p
0
+p
1
x + +p
n
x
n
+ ,
x
2
Q(x) = q
0
+q
1
x + +q
n
x
n
+ .
So if x ,= 0, then
a
0
( 1) x
2
+a
1
( + 1) x
1
+ +a
n
( +n) ( +n 1) x
+n2
+
+x
2
(a
0
+a
1
( + 1) x + ) (p
0
+p
1
x + +p
n
x
n
+ )
+x
2
(a
0
+a
1
x + +a
n
x
n
) (q
0
+q
1
x + +q
n
x
n
+ ) = 0.
We have
a
0
( 1) +a
0
p
0
+a
0
q
0
= 0, (0)
a
1
( + 1) +a
1
( + 1) p
0
+a
0
p
1
+a
0
q
1
+a
1
q
0
= 0, (1)
.
.
.
.
.
.
(n)
Note that Equation (0) implies
( 1) +p
0
+q
0
= 0,

2
+ (p
0
1) +q
0
= 0. (8.4)
98CHAPTER 8. POWER SERIES SOLUTIONS TOLINEAR DIFFERENTIAL EQUATIONS
Equation (8.4) is known as an indicial equation. Solving for gives us two
solutions. Then Equation (1) gives a
1
in terms of a
0
, Equation (2) gives a
2
in
terms of a
1
, etc. We will get the solution y = a
0
x

(). Other gives a second


solution y = b
0
x

(). The plan is to nd a


n
from Equation (n) for n 1.
Equation (n) looks like
() a
n
+ () a
n1
+ + () a
0
= 0,
where the coecients are in terms of , p
0
, . . . , p
n
, q
0
, . . . , q
n
. There is one pos-
sible problem: what happens if the coecient of a
n
is zero? Note that
y

+P(x)y

+Q(x)y = y

+xP(x)
y

x
+x
2
Q(x)
y
x
2
,
so we have
y

+xP(x)
y

x
+x
2
Q(x)
y
x
2
=

n=0
a
n
( +n) ( +n 1) x
+n2
+
_
p
0
+p
1
x +p
2
x
2
+
_

n=0
a
n
( +n) x
+n2
+
_
q
0
+q
1
x +q
2
x
2
+
_

n=0
a
n
x
+n2
.
In Equation (n), the coecient of a
n
is
( +n) ( +n 1) +p
0
( +n) +q
0
.
Suppose that
( +n) ( +n 1) +p
0
( +n) +q
0
= 0.
Let = +n. Then
( 1) +p
0
+q
0
= 0
is the indicial equation again, i.e., is the other solution to the indicial equation.
We conclude that if two solutions of the indicial equation do not dier by an
integer, the method of Frobenius provides two solutions.

If the two solutions


do dier by an integer, then the larger one will give a solution by this method,
but the smaller one may fail.
Let F() = ( 1) + p
0
+ q
0
. Suppose the solutions to F() = 0 are r

This includes the case where the solutions are complex, i.e., in this case, we get a complex
solution z = x
a+ib
(c
0
+ c
1
x + ), so y
1
= Re(x) and y
2
= Im(z) give two real solutions.
8.7. SOLUTIONS NEAR A SINGULAR POINT 99
and r +k, where k Z
+
. Given f(x), set
L(f)(x) = f

(x) +P(x)f

(x) +Q(x)f(x).
Let w =

n=0
a
n
x
n
, where a
n
is to be chosen. For any given , we have
L(x

w) =

n=0
c
n
x
2+n
,
where c
n
= F( + n)a
n
+ () a
n1
+ + () a
0
. Chose a
0
= 1. Then for
n 1, we can always solve c
n
= 0 for a
n
in terms of its predecessors unless
F(+n) = 0. Provided that ,= s 1, s 2, . . ., we have F(+n) ,= 0 for any
n, so we can solve it. Suppose that we call the solution a
n
(), i.e., a
0
() = 1 for
any and a
n
is chosen to make c
n
= 0 for this . Set W(, x)

n=0
a
n
()x
n
.
Then
L(x

w(, x)) =

n=0
c
n
x
2+n
= c
0
x
2
.
But since c
n
= 0 for n 0 by the choice of a
n
(), we have
L(x

w(, x)) = F()x


2
.
Since F(s) = 0, y
1
(x) = x
s
w(s, x) is the solution to the de. We now need
a second solution. Unfortunately, we cannot set = r since our denition of
a
n
() depends on ,= s (positive integer). Instead, set
L(x

w(, x)) = F()x


2
and dierentiated with respect to . Note that for g(, x), we have

x
(Lg) = L
_
g
x
_
since dierentiation with respect to x commutes with dierentiation with respect
to (cross derivatives equal). Therefore,
L
_
x

ln(x)w(, x) +x

_
= F

()x
2
+F()x
2
ln(x).
100CHAPTER 8. POWER SERIES SOLUTIONS TOLINEAR DIFFERENTIAL EQUATIONS
But note that
= s = L
_
ln(x)x
s
w(s, x) +x
s
_
w

=s
__
=
k
..
F

(s) x
s2
+
0
..
F(s) x
s2
ln(x)
= L
_
ln(x)y
1
+
_
w

=s
_
x
s
_
= kx
s2
.
If k = 0, then
y
2
= ln(x)y
1
+
_
w

=s
_
x
s
is the second solution to our de.
If k > 0, consider
L
_
x
r

n=0
a
n
x
n
_
= L
_
x
r

n=0
c
n
x
r+n2
_
,
where a
n
s are to be chosen and c
n
= F(r +n)a
n
+ . We have c
0
= F(r) = 0,
where r is the root of the indicial equation. Select a
0
= 1 and inductively choose
a
1
, . . . , a
n
such that c
1
, . . . , c
n
= 0. With this choice, we have
L
_
x
r

n=0
a
n
x
n
_
=

n=k
c
n
x
r+n2
= c
k
x
s2
+

n=k+1
c
n
(r)x
r+n2
,
where c
k
is independent of the choice of a
k
. Set A = c
k
/k. Then
L
_
x
r

n=0
a
n
x
n
A
_
ln(x)y
1
(x) +x
s
_
w

=s
__
_
=

n=k+1
c
n
x
r+n2
,
L
_
_
_
_
_
_
_
_
x
r
_
_
a
0
+a
1
x + +a
k1
x
k1
+
_
a
k
A
w

=s
_
x
k
_
_
+

n=k+1
a
n
x
n
Aln(x)y
1
(x)
_
_
_
_
_
_
_
_
=

n=k+1
c
n
x
r+n2
.
The choice of a
k
does not matterwe can choose any. Then inductively solve
for a
k+1
, a
k+2
, . . . in terms of predecessor a

s to make c
n
= 0 for n > k. This
gives a solution to the de of the form
y
2
= x
r

n=0
a
n
x
n
Aln(x)y
1
(x).
8.7. SOLUTIONS NEAR A SINGULAR POINT 101
We conclude that if 0 is a regular singularity, then there are always two
linearly independent solutions (converging near 0)
y
1
= x
s

n=0
a
n
x
n
, y
2
= x
r

n=0
b
n
x
n
+B(ln(x))y
1
,
where r and s are the two solutions to the indicial equation. We have B = 0,
unless s = r +k for some k 0, 1, 2, . . ..
Example 8.12. Solve 4xy

+ 2y

y = 0.
Solution. Let y =

n=0
a
n
x
+n
. Then
y

n=0
( +n) a
n
x
+n1
, y

n=0
( +n) ( +n 1) a
n
x
+n2
.
Our dierential equation now becomes

n=0
4 ( +n) ( +n 1) a
n
x
+n1
+

n=0
2 ( +n) a
n
x
+n1

n=0
a
n
x
+n
= 0,

n=0
[4 ( +n) ( +n 1) + 2 ( +n)] a
n
x
+n1

n=1
a
n1
x
+n1
= 0.
Note that
4 ( +n) ( +n 1) + 2 ( +n) = 2 ( +n) (2 + 2n 2 + 1)
= 2 ( +n) (2 + 2n 1) .
Therefore,
2(2 1) a
0
x
1
+

n=1
[2 ( +n) (2 + 2n 1) a
n1
] x
+n1
.
Therefore, the indicial equation is 2( 1) 0, 1/2 and the recursion
relation is
2 ( +n) (2 + 2n 1) a
n
a
n1
= 0
for n 1.
For = 0, we have
2n(2n 1) a
n
a
n1
= 0 = a
n
=
1
2n(2n 1)
a
n1
.
102CHAPTER 8. POWER SERIES SOLUTIONS TOLINEAR DIFFERENTIAL EQUATIONS
Thus,
a
1
=
1
2 1
a
0
,
a
2
=
1
4 3
a
1
=
1
4!
a
0
,
.
.
.
a
n
=
1
(2n) (2n 1)
a
n1
=
1
2n(2n 1) (2n 2)
=
1
(2n)!
a
0
.
Therefore,
y = x
0
a
0
_
1 +
x
2!
+
x
2
4!
+ +
x
n
(2n)
! +
_
= a
0
_
1 +
x
2!
+
x
2
4!
+ +
x
n
(2n)
! +
_
= a
0
cosh
_
x
_
.
For = 1/2, we have
2
_
1
2
+n
_
(1 + 2n 1) b
n
= b
n1
= b
n
=
1
(2n + 1) (2n)
b
n1
.
Thus,
b
1
=
1
3 2
b
0
,
b
2
=
1
5 4
b
1
=
1
5!
b
0
,
.
.
.
b
n
=
1
(2n + 1)!
b
0
.
8.7. SOLUTIONS NEAR A SINGULAR POINT 103
Therefore,
y = x
1/2
b
0
_
1 +
x
3!
+
x
2
5!
+ +
x
n
(2n + 1)!
+
_
= b
0
_
x
1/2
+
x
3/2
3!
+
x
5/2
5!
+ +
x
(2n+1)/2
(2n + 1)!
+
_
= b
0
sinh
_
x
_
.
In general, however, it is possible to only calculate a few low terms. We cannot
even get a general formula for a
n
, let alone recognize the series.
As mentioned, the above methods work near a by using the above technique
after the substitution v = x a. We can also get solutions near innity, i.e.,
for large x, we substitute t = 1/x and look at t = 0. Then
dy
dx
=
dy
dt
dt
dx
=
dy
dx
_

1
x
2
_
=
1
x
2
dy
dt
= t
2
dy
dt
,
d
2
y
dx
2
=
2
x
3
dy
dt
=
1
x
2
d
2
y
dt
2
dt
dx
=
2
x
3
dy
dt

1
x
2
d
2
y
dt
2
_

1
x
2
_
= 2t
3
dy
dt
+t
4
d
2
y
dt
2
.
Then
d
2
y
dx
2
+P(x)
dy
dx
+Q(x)y = 0 = t
4
d
2
y
dt
2
+ 2t
3
dy
dt
P
_
1
t
_
t
2
dy
dt
+Q
_
1
t
_
y = 0.
Example 8.13 (Legendres equation). Find the behaviour of the solutions
for large x of
_
1 x
2
_
y

2xy

+n(n + 1) y = 0, where n is a constant.


Solution. Let t = 1/x so that x = 1/t. Then our dierential equation becomes
_
1
1
t
2
__
2t
3
dy
dt
+t
4
d
2
y
dt
2
_

2
t
_
t
2
dy
dt
_
+n(n + 1) y = 0,
_
t
4
t
2
_
d
2
y
dt
2
+
_
2t
3

2t + 2y
_
dy
dt
+n(n + 1) y = 0,
d
2
y
dt
2
+
2t
t
2
1
dy
dt
+
n(n + 1)
t
4
t
2
y = 0.
We see that we have a regular singularity at t = 0. Therefore, there exists a
solution y = t

n=0
a
n
t
n
near t = 0, or equivalently, since t = 1/x,
y =
1
x

n=0
1
x
n
104CHAPTER 8. POWER SERIES SOLUTIONS TOLINEAR DIFFERENTIAL EQUATIONS
for large x.
Example 8.14. Find the solutions of y

+xy

+y/x = 0 for positive x near 0.


Solution. Let
y = x

n=0
a
n
x
n
=

n=0
a
n
x
n+
.
Then
y

n=0
a
n
(n +) x
n+1
,
y

n=0
a
n
(n +) (n + 1) x
n+2
,
xy

n=0
a
n
x
n+
=

n=2
a
n2
x
n+2
,
y
x
=

n=0
a
n
x
n+1
=

n=1
a
n1
x
n+2
.
Note that
Indicial equation
..
a
0
( 1) = 0, (0)
a
1
( + 1) +a
0
= 0, (1)
.
.
.
.
.
.
a
n
(n +) (n + 1) +a
n1
+a
n2
= 0. (n)
Note that Equation (0) implies that = 0, 1. The roots dier by an integer.
The largest always gives the solution. For = 1, Equation (1) implies that
2a
1
+a
0
= 0, so a
1
= a
0
/2. Equation (n) then becomes
a
n
(n + 1) n +a
n1
+a
n2
= 0 = a
n
=
a
n1
a
n2
n(n + 1)
.
Equation (2) implies that
6a
2
= a
0
a
1
= a
0
_
1 +
1
2
_
=
a
0
2
= a
2
=
a
0
12
;
8.7. SOLUTIONS NEAR A SINGULAR POINT 105
Equation (3) implies that
12a
3
= a
0
_
1
2
+
1
12
_
= a
0
7
12
= a
3
= a
0
7
144
.
Therefore,
y
1
= a
0
x
_
1
1
2
x
1
12
x
2
+
7
144
x
3
+
_
.
As for the second solution, we have
y = x
0
_
b
0
+b
1
x +b
2
x
2
+
_
+C ln(x)y
1
,
y

= b
1
+ 2b
2
x + +
C
x
y
1
+C ln(x)y

1
,
y

= 2b
2
+ 6b
3
x +
C
x
2
y
1
+
C
x
y

1
+
C
x
y

1
+C ln(x)y

1
.
Therefore, our dierential equation becomes
y

+xy

+
y
x
= 0,
_
_
_
_
_
_
_
2b
2
+ 6b
3
x + 24b
4
x
2
+
_

C
x
2
y
1
+
2C
x
y

1
+C ln(x)y

1
+
_
b
1
x + 2b
2
x
2
+
_
+Cy
1
+C ln(x)xy

1
+
_
b
0
x
+b
1
+b
2
x +
_
+C ln(x)
y
1
x
_
_
_
_
_
_
= 0,
_
_
_
_
_
_
_
_
_
_
2b
2
+ 6b
3
x + 24b
4
x
2
+
_

C
x
2
y
1
+
2C
x
y

1
+
_
b
1
x + 2b
2
x
2
+
_
+Cy
1
+
_
b
0
x
+b
1
+b
2
x +
_
+C ln(x)y

1
+C ln(x)xy

1
+C ln(x)
y
1
x
. .
0
_
_
_
_
_
_
_
_
_
= 0,
_
_
_
_
_
_
_
_
_
_
_
_
_
2b
2
+ 6b
3
x + 24b
4
x
2
+
_

C
x
_
1
1
2
x
1
12
x
2
+
7
144
x
3
+
_
+
2C
x
_
1 x +
1
4
x
2
+
7
36
x
3
+
_
+
_
b
1
x + 2b
2
x
2
+
_
+C
_
x
1
2
x
2

1
12
x
3
+
7
144
x
4
+
_
+
_
b
0
x
+b
1
+b
2
x +
_
_
_
_
_
_
_
_
_
_
_
_
_
= 0.
Looking at the coecient of x
1
, we have b
0
C + 2C = 0 C = b
0
.
Note that b
1
is arbitrary, so choose b
0
= 0. Then the coecient of x
0
implies
that 2b
2
+ C/2 2C = 0 b
2
= (3/4) b
0
; the coecient of x
1
implies that
106CHAPTER 8. POWER SERIES SOLUTIONS TOLINEAR DIFFERENTIAL EQUATIONS
6b
3
+C/12 +C/2 +C +b
2
= 0 b
3
= (7/18) b
0
. Therefore,
y
2
= b
0
__
1
3
4
x
2
+
7
18
x
2
+
_
ln(x)
_
x
1
2
x
2

1
12
x
3
+
7
144
x
4
+
__
.

Example 8.15. Solve x


2
y

xy

+ 10e
x
y = 0.
Solution. Dividing out by x
2
, we have
y

x
+
10e
x
x
2
y = 0.
We have a singularity at x = 0. Note that xP(x) = 1 and x
2
Q(x) = 10e
x
=
10

n=0
x
n
/n!. Therefore,
y =

n=0
a
n
x
+n
,
y
x
2
=

n=0
a
n
x
+n2
,
y

n=0
( +n) a
n
x
+n1
,
y

x
=

n=0
( +n) a
n
x
+n2
,
y

n=0
( +n) ( +n 1) a
n
x
+n2
.
With these components in place, our dierential equation becomes

n=0
( +n) ( +n 1) a
n
x
+n2

n=0
( +n) a
n
x
+n2
+10
_

n=0
x
n
n!
__

n=0
a
n
x
+n2
_
= 0.
We now have
( 1) a
0
a
0
+ 10a
0
= 0.
Therefore, F() = ( 1) + 10 = 0 = 1 3i. For 1 + 3i, we have
( + 1) a
1
( + 1) a
1
+ 10 (a
0
+a
1
) = 0,
F( + 1)a
1
+ 10a
1
= 0.
Therefore,
a
1
=
10
F( + 1)
a
0
=
10
F(2 + 3i)
a
0
8.7. SOLUTIONS NEAR A SINGULAR POINT 107
and
F(2 + 3i) = (2 + 3i)
2
2 (2 + 3i) + 10
= 4 + 12i 9 4 6i + 10
= 1 + 6i.
Therefore,
a
1
=
10
1 + 6i
a
0
=
10 (1 6i)
(1 + 6i) (1 6i)
a
0
=
10 60i
1 + 36
a
0
=
10 60i
37
a
0
.
Also note that
( + 2) ( + 1) a
2
( + 2) a
2
+ 10
_
a
0
2!
+a
1
+a
2
_
= 0,
and we have
F( + 2)a
2
+ 10a
1
+ 5a
0
= F( + 2)a
2

100 600i
37
a
0
+
175
37
a
0
= F( + 2)a
2
+
75 + 600i
37
a
0
.
Therefore,
a
2
=
75 + 600i
37F( + 2)
a
0
and we have
z = x
1+3i
a
0
_
1
_
10 60i
37
_
x +
_
.
But
x
1+3i
= xx
3i
= xe
3i ln(x)
= x[cos (3 ln(x)) +i sin(3 ln(x))] ,
so
z = a
0
xcos(3 ln(x)) +a
0
i sin(3 ln(x))
+a
0
_

10
37
x
2
cos(3 ln(x))
60
37
x
2
sin(3 ln(x)) +
60
37
sin(3 ln(x))
_
+
60
37
ix
2
cos(3 ln(x))
10
37
ix
2
sin(3 ln(x)) + .
108CHAPTER 8. POWER SERIES SOLUTIONS TOLINEAR DIFFERENTIAL EQUATIONS
Therefore,
Re(z) = a
0
_
_
_
_
xcos(3 ln(x))
+x
2
_

10
37
cos(3 ln(x))
60
37
sin(3 ln(x))
_
+x
3
[ cos(3 ln(x)) + sin(3 ln(x))]
_
_
_
_
+

Example 8.16. Solve xy

+ (1 x) y

y = 0.
Solution. Note that
p
0
= lim
x0
xP(x) = lim
x0
x
1 x
x
= 1
and
q
0
= lim
x0
x
2
Q(x) = lim
x0
x
2
_

1
x
_
= 0.
Therefore, 0 is a regular singular point. The indicial equation is
( 1) +p
0
+q
0
= 0,
( 1) + = 0,

2
+ = 0,

2
= 0.
Therefore, = 0 is a double root and
y
1

n=0
x
n
, y
2
=

n=0
b
n
x
n
+C ln(x)y
1
, x > 0.
To nd y
1
, we have
y =

n=0
a
n
x
n
,
y

n=0
na
n
x
n1
=

n=0
(n + 1) a
n+1
x
n
,
y

n=0
(n + 1) nx
n1
.
8.7. SOLUTIONS NEAR A SINGULAR POINT 109
Therefore, our dierential equation becomes
xy

+ (1 x) y

y = 0,

n=0
[(n + 1) na
n+1
+ (n + 1) a
n+1
na
n
a
n
] x
n
= 0.
From this, we have
(n + 1) na
n+1
+ (n + 1) a
n+1
na
n
a
n
= 0,
(n + 1)
2
a
n+1
= (n + 1) a
n
,
a
n+1
=
a
n
n + 1
.
Therefore,
a
1
=
a
0
1
, a
2
=
a
1
2
=
a
0
2
, a
3
=
a
2
2
=
a
0
3!
, . . . , a
n
=
a
0
n!
.
So we have
y = a
0

n=0
x
n
n!
= a
0
e
x
.
Immediately, y
1
= e
x
. To nd y
1
(see 5.1, p. 57), we have
y
2
= y
1
_
e

R
P(x) dx
y
2
1
, dx = e
x
_
e

R
1x
x
dx
e
2x
dx
= e
x
_
e

R
(
1
x
1)dx
e
2x
dx = e
x
_
e
(ln(x)x)
e
2x
dx
= e
x
_
e
x
/x
e
2x
dx = e
x
_
e
x
x
dx,
which is not an elementary function.
Example 8.17. Solve
_
x
2
+ 2x
_
y

2
_
x
2
+ 2x 1
_
y

+
_
x
2
+ 2x 2
_
y = 0.
Solution. Let y =

n=0
a
n
x
+n
. Then
y

n=0
( +n) a
n
x
+n1
, y

n=0
( +n) ( +n 1) a
n
x
+n2
.
110CHAPTER 8. POWER SERIES SOLUTIONS TOLINEAR DIFFERENTIAL EQUATIONS
The de then becomes
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_

n=0
( +n) ( +n 1) a
n
x
+n
+ 2

n=0
( +n) ( +n 1) a
n
x
+n1
2

n=0
( +n) a
n
x
+n+1
4

n=0
( +n) a
n
x
+n
+2

n=0
( +n) a
n
x
+n1
+

n=0
a
n
x
+n+2
+ 2

n=0
a
n
x
+n+1
2

n=0
a
n
x
+n
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
= 0,
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_

n=1
( +n 1) ( +n 2) a
n1
x
+n1
+2

n=0
( +n) ( +n 1) a
n
x
+n1
2

n=2
( +n 2) a
n2
x
+n1
4

n=1
( +n 1) a
n1
x
+n1
+ 2

n=0
( +n) a
n
x
+n1
+

n=3
a
n3
x
+n1
+ 2

n=2
a
n2
x
+n1
2

n=1
a
n1
x
+n1
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
= 0,
_

n=0
_
_
_
[2 ( +n) ( +n 1) + 2 ( +n)] a
n
+[( +n 1) ( +n 2) 4 ( +n 1) 2] a
n1
+[2 ( +n 2) + 2] a
n2
+a
n3
_
_
_
_

_x
+n1
= 0,

n=0
_
_
2 ( +n)
2
a
n
+
_
( +n)
2
7 ( +n) + 4
_
a
n1
+[2 ( +n) + 6] a
n2
+a
n3
_
_
= 0.
Note that 2
2
= 0 = 0, giving us a double root. Therefore we get the
recursion relation
2n
2
a
n
+
_
n
2
7n + 4
_
a
n1
+ (2n + 6) a
n2
+a
n3
= 0.
Substituting gives
2a
1
2a
0
= 0 = a
1
= a
0
,
8a
2
6a
1
+ 2a
0
= 0 = 8a
2
= 6a
1
2a
0
= (6 2) a
0
= 4a
0
= a
0
=
1
2
,
18a
3
+ (8) a
2
+ (0) a
1
+a
2
= 0 = 18a
3
= 8
1
2
1 = 3 = a
3
=
1
6
,
32a
4
8a
3
(2) a
2
+ (0) a
1
+a
2
= 0 = 32a
3
= 8
1
6
2
1
2
1 =
4
3
= a
3
=
1
24
.
8.8. FUNCTIONS DEFINED VIA DIFFERENTIAL EQUATIONS 111
If we our intuition leads us to guess that perhaps a
n
= 1/n! be can use induction
to check that this guess is in fact correct. Explicitly, suppose by induction that
a
k
= 1/k! for all k < n. Then the recursion relation yields
2n
2
a
n
=
n
2
7n + 4
(n 1)!
+
2n 6
(n 2)!

1
(n 3)!
=
n
2
+ 7n 4 + (2n 6)(n 1) (n 1)(n 2)
(n 1)!
=
n
2
+ 7n 4 + 2n
2
8n + 6 n
2
+ 3n 2
(n 1)!
=
2n
(n 1)!
and so a
n
= 1/n!, completing the induction.
This gives the solution y
1
=

n=0
x
n
/n! = e
x
. (If we had guessed initially
that this might be the solution, we could easily have checked it by direct subsi-
tution into the dierential equation.) Using reduction of order, one then nds
that the solutions are
y
1
= e
x
, y
2
= xe
x
+ 2 ln(x) y
1
.

8.8 Functions Dened via Dierential Equations
8.8.1 Chebyshev Equation
The Chebyshev equation is given by
_
1 x
2
_
y

xy

+
2
y = 0 (8.5)
112CHAPTER 8. POWER SERIES SOLUTIONS TOLINEAR DIFFERENTIAL EQUATIONS
of which 0 is an ordinary point. The series solution is guaranteed to converge
for [x[ < 1. We have
y =

n=0
a
n
x
n
,
y

n=0
na
n
x
n1
,
y

n=0
n(n 1) a
n
x
n2
=

n=0
(n + 2) (n + 1) a
n+2
x
n
,
and our dierential equation becomes

n=0
_
(n + 2) (n + 1) a
n+2
n(n 1) a
n
na
n
+
2
a
n

x
n
= 0.
From this, we see that
a
n+2
=
n(n 1) +n
2
(n + 2) (n + 1)
a
n
=
n
2

2
(n + 2) (n + 1)
a
n
.
Therefore,
a
0
= 1, a
1
= 0, a
2
=

2
2
, a
3
= 0,
a
4
=

2
4 3
=
_
4
2
_ _

2
_
4!
, a
5
= 0,
a
6
=
_
16
2
_ _
4
2
_ _

2
_
6!
,
.
.
.
a
2n
=
_
4 (n 1)
2

2
__
4 (n 2)
2

2
_

_

2
_
(2n)!
a
1
.
8.8. FUNCTIONS DEFINED VIA DIFFERENTIAL EQUATIONS 113
Also,
a
0
= 0, a
1
= 1, a
2
= 0, a
3
=
_
1
2
_
3 2
a
1
,
a
4
= 0, a
5
=
_
9
2
_ _
1
2
_
5!
a
1
,
.
.
.
a
2n+1
=
_
(2n 1)
2

2
_

_
1
2
_
(2n + 1)!
a
1
.
Note that if is an integer, then one of these solutions terminates, i.e.,
a
k
= 0 beyond some point. In this case, one solution is a polynomial known as
a Chebyshev polynomial.
8.8.2 Legendre Equation
The Legendre equation is given by
_
1 x
2
_
y

2xy

+( + 1) y = 0 (8.6)
of which 0 is an ordinary point. The series solution is guaranteed to converge
for [x[ < 1. Expressed as a power series, the equation becomes

n=0
= [(n + 2) (n + 1) a
n+2
n(n 1) a
n
2na
n
+( + 1) a
n
] x
n
= 0.
From this, we have
a
n+2
=
n(n 1) + 2n ( + 1)
(n + 2) (n + 1)
a
n
=
n(n + 1) ( + 1)
(n + 2) (n + 1)
a
n
.
If is an integer, then a
k
= 0 for k gives the th Legendre polynomial.
Instead, expanding around x = 1 gives
p(x)
x
1 x
2
.
114CHAPTER 8. POWER SERIES SOLUTIONS TOLINEAR DIFFERENTIAL EQUATIONS
Then
p
0
= lim
x1
(x 1) P(x) = lim
x1
x
x + 1
=
1
2
,
q
0
= lim
x1
(x 1)
2
Q(x) = lim
x1
(x 1)
2

2
x
2
1
= 0.
Therefore, x = 1 is a regular singular point. The indicial equation is
( 1) +p
0
+q
0
= 0,
( 1) +
1
2
+q
0
= 0,

2
+
1
2
= 0,

1
2
= 0.
Therefore, = 0 or = 1/2, so the solutions look like
y
1
=

n=0
a
n
(x 1)
n
, y
2
= (x 1)
1/2

n=0
b
n
(x 1)
n
for x > 1.
For a Legendre equation, around x = 1, we have
p
0
= lim
x1
(x 1)
_

2x
1 x
2
_
= 1,
q
0
= lim
x1
(x 1)
2
( + 1)
x
2
1
= 0.
Therefore, x = 1 is a regular singular point. The indicial equation is
( 1) +p
0
+q
0
= 0,
( 1) + = 0,

2
= 0.
Therefore, = 0 is a double root and the solutions look like
y
1
=

n=0
a
n
(x 1)
n
, y
2
=

n=0
b
n
(x 1)
n
+C ln(x 1) y
1
for x > 1.
8.8. FUNCTIONS DEFINED VIA DIFFERENTIAL EQUATIONS 115
8.8.3 Airy Equation
The Airy equation is given by
y

xy = 0 (8.7)
of which 0 is an ordinary point. The solution converges for all x. Expressed as
a power series, we have

n=0
[(n + 2) (n + 1) a
n+2
a
n1
] x
n
= 0.
From this, we see that
a
n+2
=
a
n1
(n + 2) (n + 1)
,
therefore,
a
3
=
a
0
3 2
, a
6
=
a
0
6 5
, . . . , a
3n
=
a
0
2 3 5 6 8 9 (3n 1) (3n)
,
a
4
=
a
1
4 3
, . . . , a
3n+1
=
a
1
3 4 6 7 (3n) (3n + 1)
,
a
2
=
a
1
2
= 0, 0 = a
2
= a
5
= a
8
= .
8.8.4 Laguerres Equation
Laguerres equation

is given by
xy

(1 x) y

+y = 0 (8.8)
of which 0 is a regular singular point. Note that
p
0
= lim
x0
x
_
(1 x)
x
_
= 1,
q
0
= lim
x0
x
2

x
= 0.

Applicable in the context of the hydrogen atom.


116CHAPTER 8. POWER SERIES SOLUTIONS TOLINEAR DIFFERENTIAL EQUATIONS
Therefore, x = 0 is a regular singular point. The indicial equation is
( 1) = 0,

2
= 0,

2
2 = 0,
( 2) = 0.
Therefore, = 0 or = 2, so the solutions look like
y
1
= x
2

n=0
a
n
x
n
, y
2
=

n=0
b
n
x
n
+C ln(x)y
1
for x > 0.
8.8.5 Bessel Equation
The Bessel equation is given by
x
2
y

+xy

+
_
x
2

2
_
y = 0. (8.9)
We have
p
0
= lim
x0
x
x
x
2
= 1,
q
0
= lim
x0
x
2
x
2

2
x
2
= .
The indicial equation is then
( 1) +
2
= 0,

2
= 0,
( +) ( ) = 0.
Therefore, = . If ,= k/2 for some integer k, then we get
y
1
= x

n=0
a
n
x
n
, y
2
= x

n=0
a
n
x
n
8.8. FUNCTIONS DEFINED VIA DIFFERENTIAL EQUATIONS 117
for x > 0. We encounter trouble if = k/2. Consider = 1/2. Then
y
1
= x
1/2

n=0
a
n
x
n
, y
2
= x
1/2

n=0
b
n
x
n
+C ln(x)y
1
.
For y
1
, we have
y =

n=0
a
n
x
n+1/2
,
y

n=0
_
n +
1
2
_
a
n
x
n1/2
,
y

n=0
_
n +
1
2
__
n
1
2
_
. .
n
2
1/4
a
n
x
n3/2
.
Therefore, the equation x
2
y

+xy

+
_
x
2
1/4
_
y = 0 becomes

n=0
__
n
2

1
4
_
a
n
+
_
n +
1
2
_
a
n
+a
n2

1
4
a
n
_
x
n+1/2
= 0.
From this, we see that
_
n
2

1
4
+n +
1
2

1
4
_
a
n
+a
n2
= 0,
a
n
=
a
n2
n
2
+n
=
a
n2
n(n + 1)
.
Therefore,
a
2
=
a
0
3 2
, a
4
=
a
2
5 4
=
a
0
5!
, a
6
=
a
0
7!
, . . . , a
2n
= (1)
n
a
0
(2n + 1)!
.
Therefore
y
1
=

n=0
(1)
n
x
2n+1/2
(2n + 1)!
=
1

n=0
(1)
n
x
2n+1
(2n + 1)!
=
sin(x)

x
.
118CHAPTER 8. POWER SERIES SOLUTIONS TOLINEAR DIFFERENTIAL EQUATIONS
Example 8.18. Identify
y =

n=0
x
3n
(3n)!
= 1 +
x
3
3!
+
x
6
6!
+
x
9
9!
+ .

Solution. We seek to solve y

= y. From this, we see that

3
1 = 0,
( 1)
_

2
+ + 1
_
= 0.
Therefore, = 1 or
=
1

1 4
2
=
1
2

3
2
i.
Therefore, the general solution is
y = C
1
e
x
+C
2
e
x/2
sin
_

3
2
x
_
+C
3
e
x/2
cos
_

3
2
x
_
,
where C
1
, C
2
, C
3
are arbitrary constants.
To nd C
1
, C
2
, C
3
, we note that
y(0) = 1, y

(0) = 0, y

(0) = 0.
Therefore,
C
1
+ 0C
2
+C
3
= 1,
C
1
+

3
2
C
2

1
2
C
3
= 0,
C
1

3
2
C
2

1
2
C
3
= 0.
8.8. FUNCTIONS DEFINED VIA DIFFERENTIAL EQUATIONS 119
Solving this system, we have
_

_
1 0 1 1
1

3
2

1
2
0
1

3
2

1
2
0
_

_
_

_
1 0 1 1
0

3
2
3
2
1
0

3
2
3
2
1
_

_
_

_
1 0 1 1
0

3
2
3
2
1
0 0 3 2
_

_
_

_
1 0 1 1
0
1

3
1
2
3
0 0 1
2
3
_

_
_

_
1 0 0
1
3
0
1

3
0 0
0 0 1
2
3
_

_
_

_
1 0 0
1
3
0 1 0 0
0 0 1
2
3
_

_.
Therefore, C
1
= 1/3, C
2
= 0, and C
3
= 2/3, so
y =
1
3
e
x
+
2
3
e
x/2
cos
_

3
2
x
_
.

120CHAPTER 8. POWER SERIES SOLUTIONS TOLINEAR DIFFERENTIAL EQUATIONS


Chapter 9
Linear Systems
9.1 Preliminaries
Consider
dY
dx
= AY +B,
where
A =
_
a b
c d
_
, Y =
_
y
1
y
2
_
such that
dy
1
dx
= ay
1
+by
2
,
dy
2
dx
= cy
1
+dy
2
.
We will rst consider the homogeneous case where B = 0.
Only in the case when the entries of A are constants is there an algorithmic
method of nding the solution. If A a is a constant, then we know that
y = Ce
ax
is a solution to dy/dx = ay. We will dene a matrix e
A
such that the
general solution of dY/dx = AY is Y = Ce
xA
, where C = [C
1
, C
2
, . . . , C
n
] is
a vector of arbitrary constants. We dene e
T
for an arbitrary n n matrix T
and apply T = xA. Indeed, we know that
e
t
= 1 +t +
t
2
2!
+
t
3
3!
+
t
4
4!
+ +
t
n
n!
+ .
Therefore, let us analogously dene
e
T
I +T+
T
2
2!
+
T
3
3!
+ +
T
n
n!
+ . (9.1)
121
122 CHAPTER 9. LINEAR SYSTEMS
We will consider systematically how to compute e
T
later, but rst we consider
an ad hoc example.
Example 9.1. Solve
dY
dx
=
_
1 1
0 1
_
Y.

Solution. First, we write


A =
_
1 1
0 1
_
= I +N,
where
N =
_
0 1
0 0
_
so that NI = IN = N and N
2
= 0. Note that
A
n
=
Binomial exp.
..
(I +N)
n
= I
n
+
_
n
1
_
I
n1
N+
_
n
2
_
I
n1
N
2
+ +
_
n
n 1
_
IN
n1
+N
n
= I +nN =
_
1 n
0 1
_
.
Therefore, with A
n
properly dened, Equation (9.1) gives
e
Ax
= I +
_
1 1
0 1
_
x +
_
1 2
0 1
_
x
2
2
+ +
_
1 n
0 1
_
x
n
n!
+
=
_
1 +x +
x
2
2!
+ +
x
n
n!
+ 0 +x + 2
x
2
2!
+ 3
x
3
3!
+ +n
x
n
n!
0 1 +x +
x
2
2!
+ +
x
n
n!
+
_
=
_
_
e
x
x
_
1 +x +
x
2
2!
+ +
x
n
n!
+
_
0 e
x
_
_
=
_
e
x
xe
x
0 e
x
_
.
Therefore, the solution is
_
y
1
y
2
_
=
_
e
x
xe
x
0 e
x
__
C
1
C
2
_
=
_
C
1
e
x
+C
2
xe
x
C
2
e
x
_
,
9.2. COMPUTING E
T
123
that is,
y
1
= C
1
e
x
+C
2
xe
x
, y
2
= C
2
e
x
.
Is this in fact the solution of the equation? We have
d
dx
e
Ax
=
d
dx
_
I +Ax +A
2
x
2
2!
+ +A
n
x
n
n!
+
_
= A+A
2
x + +A
n
x
n1
(n 1)! +
= Ae
Ax
.
So if Y = Ce
Ax
, then
dY
dx
= ACe
Ax
= AY,
as required. Therefore, Y = Ce
Ax
is indeed the solution.
9.2 Computing e
T
The easiest case to do is a diagonal matrix. If
D =
_

2
.
.
.

n
_

_
,
then
e
D
=
_

_
e
1
e
2
.
.
.
e
n
_

_
.
First, we consider the eect of changing bases. Suppose T is diagonalizable, that
is, there exists a P such that P
1
TP is diagonalizable. Therefore, T = PDP
1
and
T
2
=
_
PDP
1
_ _
PDP
1
_
= PD
2
P
1
.
124 CHAPTER 9. LINEAR SYSTEMS
So in general, we have T
n
= PD
n
P
1
. It then follows that
e
T
= I +T+
T
2
2!
+ +
T
n
n!
= P
_
I +D+
D
2
2!
+ +
D
n
n!
+
_
P
1
= Pe
D
P
1
.
So if T is diagonalizable, we can compute e
T
.
Example 9.2. Compute e
T
, where
T =
_
4 1
3 2
_
.

Solution. First note that

4 1
3 2

= 0,
(4 ) (2 ) 3 = 0,
8 6 +
2
3 = 0,

2
6 + 5 = 0,
( 5) ( 1) = 0.
Therefore, = 1 or = 5. For = 1, we have
_
4 1 1
3 2 1
__
a
b
_
=
_
0
0
_
,
_
3 1
3 1
__
a
b
_
=
_
0
0
_
,
_
3a +b
3a +b
_
=
_
0
0
_
.
We see that b = 3a, so
_
a
b
_
=
_
a
3a
_
= a
_
1
3
_
.
9.2. COMPUTING E
T
125
For = 5, we have
_
4 5 1
3 2 5
__
a
b
_
=
_
0
0
_
,
_
1 1
3 3
__
a
b
_
=
_
0
0
_
,
_
a +b
3 3b
_
=
_
0
0
_
.
We see that a = b, so
_
a
b
_
=
_
a
a
_
= a
_
1
1
_
.
So let
P =
_
1 1
3 1
_
.
Therefore,
P
1
=
1
[P[
_
1 1
3 1
_
=
1
4
_
1 1
3 1
_
.
So
PTP =
1
4
_
1 1
3 1
__
4 1
3 2
__
1 1
3 1
_
=
1
4
_
1 1
3 1
__
1 5
3 5
_
=
1
4
_
1 0
0 20
_
.
What is P? For each eigenvalue
j
, we nd its corresponding eigenvector
v
j
. Then for P = [v
1
, v
2
, . . . , v
n
], which is an n n matrix, we have
A = P
_

2
.
.
.

n
_

_
P
1
.
126 CHAPTER 9. LINEAR SYSTEMS
For Y

= AY, we have
Y = e
Ax

C = e
PDP
1
x
C = Pe
xD
constant
..
P
1

C
= P
_

_
e
1x
e
2x
.
.
.
e
nx
_

_
C
= [v
1
, v
2
, . . . , v
n
]
_

_
e
1x
e
2x
.
.
.
e
nx
_

_
C
= C
1
e
1x
v
1
+C
2
e
2x
v
2
+ +C
n
e
nx
v
n
,
where C = P
1

C.
Example 9.3. Solve
Y

=
_
4 1
3 2
_
Y.

Solution. First note that

4 1
3 2

= 0,
(4 ) (2 ) 3 = 0,
8 6 +
2
3 = 0,

2
6 + 5 = 0,
( 5) ( 1) = 0.
Therefore, = 1 or = 5. For = 1, we have
_
4 1 1
3 2 1
__
a
b
_
=
_
0
0
_
,
_
3 1
3 1
__
a
b
_
=
_
0
0
_
,
_
3a +b
3a +b
_
=
_
0
0
_
.
9.2. COMPUTING E
T
127
We see that b = 3a, so
_
a
b
_
=
_
a
3a
_
= a
_
1
3
_
.
For = 5, we have
_
4 5 1
3 2 5
__
a
b
_
=
_
0
0
_
,
_
1 1
3 3
__
a
b
_
=
_
0
0
_
,
_
a +b
3 3b
_
=
_
0
0
_
.
We see that a = b, so
_
a
b
_
=
_
a
a
_
= a
_
1
1
_
.
Therefore,
P =
_
1 1
3 1
_
, D =
_
1 0
0 5
_
and we have
Y =
_
1 1
3 1
__
e
x
0
0 e
5x
__
C
1
C
2
_
=
_
e
x
e
5x
3e
x
e
5x
__
C
1
C
2
_
=
_
C
1
e
x
+C
2
e
5x
3C
1
e
x
+C
2
e
5x
_
= C
1
e
x
_
1
3
_
+C
2
e
5x
_
1
1
_
,
that is,
y
1
= C
1
e
x
+C
2
e
5x
, y
2
= 3C
1
e
x
+C
2
e
5x
.

If T does not have n distinct eigenvalues, computing e
T
requires greater
knowledge of linear algebra.

Property 9.4
1. e
PAP
1
= Pe
A
P
1
.
2. If AB = BA, then e
A+B
= e
A
e
B
= e
B
e
A
.

MATB24.
128 CHAPTER 9. LINEAR SYSTEMS
3. If
D =
_

2
.
.
.

n
_

_
,
then
e
D
=
_

_
e
1
e
2
.
.
.
e
n
_

_
.
4. If A
n+1
= 0, then
e
A
= I +A+
A
2
2!
+ +
A
n
n!
.
Proof. We have already proved (1), and (3) and (4) are trivial. What remains
is (2). So note that
e
A
e
B
=
_
I +A+
A
2
2!
+ +
A
n
n!
__
I +B+
B
2
2!
+ +
B
n
n!
_
= I +A+B+
1
2
_
A
2
+ 2AB+B
2
_
+ .
Also
e
A+B
= I +A+B+
(A+B)
2
2!
+
(A+B)
3
3!
+
= I +A+B+
1
2
_
A
2
+AB+BA+B
2
_
+ ,
which is the same as the above if AB = BA.
Theorem 9.5
Given T over C, there exists matrices P, D, and N with D diagonal,
N
k+1
= 0 for some k, and DN = ND such that T = P(D+N) P
1
.
9.3. THE 2 2 CASE IN DETAIL 129
Proof. Proof is given in MATB24.
Corollary 9.6
We have e
T
= Pe
D
e
N
P
1
(e
D
and e
N
can be computed as above).
Proof. Proof is given in MATB24.
The steps involved in nding P, N, and D are essentially the same as those
involved in putting T into Jordan Canonical form (over C).
9.3 The 2 2 Case in Detail
Theorem 9.7 (Cayley-Hamilton)
Let p() = [AI[. Then p(A) = 0.
Proof. Proof is given in MATB24.
The eigenvalues of A are the solutions of p() = 0. We now consider three
cases for the solutions of p() = 0.
Case 1: p() has two distinct real roots
1
and
2
. This is the diagonalizable
case considered earlier. We nd eigenvectors v and w for the eigenvalues
1
and

2
. Let
P =
_
v
1
w
1
v
2
w
2
_
.
Then
P
1
TP =
_

1
0
0
2
_
,
so
e
T
= P
_
e
1
0
0 e
2
_
P
1
.
The solution of Y

= TY would be
Y = P
_
e
1x
0
0 e
2x
__
C
1
C
2
_
130 CHAPTER 9. LINEAR SYSTEMS
in this case.
Case 2: p() has a double root r. Let S = TrI so that T = S +rI. Since
SI = IS, we have
e
T
= e
S
e
rI
= e
s
_
e
r
0
0 e
r
_
.
Then
p(T) = 0,
(TrI)
2
= 0,
S
2
= 0.
Therefore,
e
S
= I +S = I +TrI = T+ (1 r) I
and
e
T
= (T+ (1 r) I)
_
e
r
0
0 e
r
_
.
Case 3: p() has complex roots r iq, where q ,= 0. Let S = TrI so that
T = S +rI. Then
e
T
= e
S
_
e
r
0
0 e
r
_
and
p() =
2
2r +
_
r
2
+q
2
_
= ( r)
2
+q
2
.
We then have
p(T) = 0,
(TrI)
2
+ (qI)
2
= 0,
S
2
+ (qI)
2
= 0,
S
2
+q
2
I = 0.
9.3. THE 2 2 CASE IN DETAIL 131
Therefore, S
2
q
2
I and it follows that
e
S
= I +S +
S
2
2!
+
S
3
3!
+
= I +S
q
2
I
2!

q
2
S
3!
+
q
4
I
4!
+
q
4
S
5!

q
6
I
6!
+
= I
_
1
q
2
2!
+
q
4
4!

q
6
6!
+
_
+S
_
1
q
2
3!
+
q
4
5!

_
= I cos(q) +
S
q
sin(q).
Therefore,
e
T
=
_
cos(q)I +
_
TrI
q
_
sin(q)
_
_
e
r
0
0 e
r
_
.
Example 9.8 (Double root). Solve
Y

=
_
3 9
4 9
_
Y.

Solution. We have
p() = 0,

3 9
4 9

= 0,
(3 ) (9 ) + 36 = 0,
27 6 +
2
+ 36 = 0,

2
6 + 9 = 0,
( 3)
2
= 0.
Therefore, = 3 is a double root and we have Y = e
Ax
C. For A, the eigenvalue
132 CHAPTER 9. LINEAR SYSTEMS
is 3. Therefore, for T, the eigenvalue is r = 3x. Hence,
Y = (T+ (1 r) I)
_
e
r
0
0 e
r
_
= (Ax + (1 3x) I)
_
e
3x
0
0 e
3x
_
C
=
__
3x 9x
4x 9x
_
+
_
1 3x 0
0 1 3x
___
e
3x
0
0 e
3x
_
C
=
_
1 6x 9x
4x 1 + 6x
__
e
3x
0
0 e
3x
_
C
=
_
e
3x
6xe
3x
9xe
3x
4xe
3x
e
3x
+ 6xe
3x
__
C
1
C
2
_
=
_
C
1
_
e
3x
6xe
3x
_
9C
2
xe
3x
4C
1
xe
3x
+C
2
_
e
3x
+ 6xe
3x
_
_
= C
1
e
3x
_
1
4
_
+C
1
xe
3x
_
6
1
_
+C
2
e
3x
_
9
1
_
+C
2
xe
3x
_
9
6
_
.
Example 9.9 (Complex roots). Solve
Y

=
_
1 4
1 1
_
Y.

Solution. We have
p() = 0,

1 4
1 1

= 0,
(1 ) (1 ) + 4 = 0,
( + 1)
2
+ 4 = 0.
Therefore, = 1 2i. For A, the roots are 1 2i, so for T, the roots are
(1 2i) x, i.e., r = x and q = 2x. Note that
TrI
q
=
1
2x
_
0 4x
x 0
_
=
_
0 2
1
2
0
_
.
9.4. THE NON-HOMOGENEOUS CASE 133
Thus,
Y = e
Ax
C
=
_
cos(2x)I + sin(2x)
_
0 2
1
2
0
___
e
x
0
0 e
x
_
C
=
_
cos(2x) 2 sin(2x)
1
2
sin(2x) cos(2x)
__
e
x
0
0 e
x
_
C
=
_
e
x
cos(2x) 2e
x
sin(2x)
1
2
e
x
sin(2x) e
x
cos(2x)
__
C
1
C
2
_
=
_
C
1
e
x
cos(2x) + 2C
2
e
x
sin(2x)
1
2
C
1
e
x
sin(2x) +C
2
e
x
cos(2x)
_
.
9.4 The Non-Homogeneous Case
The general solution of Y

= AY + B is Y = Y
h
C + Y
p
, where Y
h
C is the
general solution of Y

= AY. Therefore, Y

h
C = AY
h
C for all C, giving us
Y

h
= AY
h
, (9.2a)
and Y
p
is a particular solution to Y

= AY +B, giving us
Y

p
= AY
p
+B. (9.2b)
To nd Y
p
, we use variation of parameters. Thus, we have
Y
p
= Y
h
V, (9.2c)
where
V =
_
v
1
v
2
_
.
134 CHAPTER 9. LINEAR SYSTEMS
To nd V, we have
Y
p
= Y
h
V,
Y

p
..
(9.2b)
= Y

h
..
(9.2a)
V+Y
h
V

,
AY
p
+B = AY
h
V
. .
(9.2c)
+Y
h
V

= AY
p
+Y
h
V

.
Therefore, Y
h
V

= B V

= Y
1
h
B.
Example 9.10. Solve
Y

=
_
4 1
3 2
_
Y +
_
e
5x
e
2x
_
.

Solution. As before, we have


1
= 1 and
2
= 5. Thus,
V
1
=
_
1 3
_
, V
2
=
_
11
_
.
Therefore,
P =
_
1 1
3 1
_
, D =
_
1 0
0 5
_
,
and the solution of Y

= AY is Y = Pe
Dx
C = Y
h
C. So we have
Y
h
= Pe
Dx
=
_
1 1
3 1
__
e
x
0
0 e
5x
__
e
x
e
5x
3e
x
e
5x
_
,
Y
1
h
= e
Dx
P
1
=
1
[P[
_
e
x
0
0 e
5x
__
1 1
3 1
_
=
1
4
_
e
x
e
x
3e
5x
e
5x
_
.
It follows that
V

= Y
1
h
B =
1
4
_
e
x
e
x
3e
5x
e
5x
__
e
5x
e
2x
_
=
1
4
_
e
4x
e
x
3 +e
3x
_
,
V =
1
4
_
1
4
e
4x
e
x
3x
1
3
e
3x
_
9.5. PHASE PORTRAITS 135
and
Y
p
= Y
h
V =
1
4
_
e
x
e
5x
3e
x
e
5x
__
1
4
e
4x
e
x
3x
1
3
e
3x
_
=
1
4
_
1
4
e
5x
e
2x
+ 3xe
5x

1
3
e
2x

3
4
e
5x
+ 3e
2x
+ 9xe
5x

1
3
e
2x
_
= e
5x
_
1
16

3
16
_
+e
2x
_

1
4
3
4
_
+xe
5x
_
3
4
9
4
_
+e
2x
_

1
12

1
12
_
.
Therefore,
Y = Y
h
C+Y
p
= C
1
e
x
_
1
3
_
+C
2
e
5x
_
1
1
_
+e
5x
_
1
16

3
16
_
+e
2x
_

1
4
3
4
_
+xe
5x
_
3
4
9
4
_
+e
2x
_

1
12

1
12
_
,
where C
1
and C
2
are arbitrary constants.
9.5 Phase Portraits: Qualitative and Pictorial
Descriptions of Solutions of Two-Dimensional
Systems
Let V : R
2
R
2
be a vector eld. Imagine, for example, that V(x, y) represents
the velocity of a river at the point (x, y).

We wish to get the description of


the path that a leaf dropped in the river at the point (x
0
, y
0
) will follow. For
example, Figure 9.1 shows the vector eld of V(x, y) =
_
y, x
2
_
. Let (t) =
(x(t), y(t)) be such a path. At any time, the leaf will go in the direction that
the river is owing at the point at which it is presently located, i.e., for all t, we
have

(t) = V(x(t), y(t)). If V(F, G), then


dx
dt
= F(x, y)
. .
y
,
dy
dt
= G(x, y)
. .
x
2
.

We are assuming here that V depends only on the position (x, y) and not also upon time
t.
136 CHAPTER 9. LINEAR SYSTEMS
-4 -2 0 2 4
-4
-2
0
2
4
x
y
(a) The vector eld plot of
`
y, x
2

.
- 3 - 2 - 1 0 1 2 3 4
x
- 3
- 2
- 1
0
1
2
3
4
y
(b) The trajectories of
`
y, x
2

Figure 9.1: The vector eld plot of


_
y, x
2
_
and its trajectories.
In general, it will be impossible to solve this system exactly, but we want to
be able to get the overall shape of the solution curves, e.g., we can see that in
Figure 9.1, no matter where the leaf is dropped, it will head towards (, ) as
t .
Before considering the general case, let us look at the linear case where we
can solve it exactly, i.e., V = (ax +by, cx +dy) with
dx
dt
= ax +by,
dy
dt
= cx +dy,
or X

= AX, where
X =
_
x
y
_
, A =
_
a b
c d
_
.
Recall from Theorem 10.12 (p. 157) that if all the entries of A are continuous,
then for any point (x
0
, y
0
), there is a unique solution of X

= AX satisfying
x(t
0
) = x
0
and y(t
0
) = y
0
, i.e., there exists a unique solution through each
point; in particular, the solution curves do not cross.
The above case can be solved explicitly, where
X = e
At
_
x
0
y
0
_
is a solution passing through (x
0
, y
0
) at time t = 0. We will consider only cases
9.5. PHASE PORTRAITS 137
where [A[ , = 0.
9.5.1 Real Distinct Eigenvalues
Let
1
and
2
be distinct eigenvalues of Aand let v and wbe their corresponding
eigenvectors. Let P = (v, w). Then
P
1
AP =
_

1
0
0
2
_
. .
D
.
Therefore, At = P(Dt) P
1
and we have
X = e
At
_
x
0
y
0
_
= Pe
Dt
P
1
_
x
0
y
0
_
= (v, w)
_
e
1t
0
0 e
2t
__
C
1
C
2
_
= (v, w)
_
C
1
e
1t
C
2
e
2t
_
= C
1
e
1t
v +C
2
e
2t
w.
Dierent C
1
and C
2
values give various solution curves.
Note that C
1
= 1 and C
2
= 0 implies that X = e
1t
v. If
1
< 0, then
the arrows point toward the origin as shown in Figure 9.2a in which contains a
stable node. Note that
X = C
1
e
1t
v +C
2
e
2t
w = e
2t
_
C
1
e
(12)t
v +C
2
w
_
.
The coecient of v goes to 0 as t , i.e., as t , X (0, 0), approaching
along a curve whose tangent is w. As t , X = e
1t
_
C
1
v +C
2
e
(21)t
w
_
,
i.e., the curves get closer and closer to being parallel to v as t .
We have the degenerate case when
1
<
2
= 0, in which case X = C
1
e
1t
v+
C
2
w.
The case when
1
< 0 <
2
gives us the phase portrait shown in Figure 9.2b
in which contains a saddle point. This occurs when [A[ < 0. The case when
0 <
1
<
2
gives us the phase portrait shown in Figure 9.2c in which contains
an unstable node. We have
X = C
1
e
1t
v +C
2
e
2t
w = e
1t
_
C
1
v +C
2
e
(21)t
w
_
.
Therefore, as t , X approaches parallel to w asymptotically.
Note that in all cases, the origin itself is a xed point, i.e., at the origin,
x

= 0 and y

= 0, so anything dropped at the origin stays there. Such a point


138 CHAPTER 9. LINEAR SYSTEMS
0
0
0
0
(a)
1
<
2
< 0
0
0
0
0
(b)
1
< 0 <
2
0
0
0
0
(c) 0 <
1
<
2
Figure 9.2: The cases for
1
and
2
.
9.5. PHASE PORTRAITS 139
is called an equilibrium point; in a stable node, if it is disturbed, it will come
back in an unstable node; if perturbed slightly, it will leave the vicinity of the
origin.
9.5.2 Complex Eigenvalues
Complex eigenvalues come in the form = i, where ,= 0. In such a case,
we have
X = C
1
Re
_
e
t
v
_
+C
2
Im
_
e
t
v
_
,
where v = p +iq is an eigenvector for . Then
e
t
v = e
t
e
it
(p +iq)
= e
t
(cos(t) +i sin(t)) (p +iq)
= e
t
(cos(t)p sin(t)q +i cos(t)q +i sin(t)p) .
Therefore,
X = e
t
[(C
1
cos(t)p C
1
sin(t)q) +C
1
cos(t)q +C
2
sin(t)q]
= e
t
_
k
1
cos(t) +k
2
sin(t)
k
3
cos(t) +k
4
sin(t)
_
= e
t
_
cos(t)
_
k
1
k
3
_
+ sin(t)
_
k
2
k
4
__
.
Note that tr(A) = 2.

So
= 0 = tr(A) = 0,
> 0 = tr(A) < 0,
< 0 = tr(A) > 0.
Consider rst = 0. To consider the axes of the ellipses, rst note that
X =
_
p
1
q
1
p
2
q
2
_
. .
P
_
C
1
C
2
C
2
C
1
_
. .
C
_
cos(t)
sin(t)
_
.

Recall that the trace of a matrix is the sum of the elements on the main diagonal.
140 CHAPTER 9. LINEAR SYSTEMS
Except in the degenerate case, where p and q are linearly dependent, we have
_
cos(t)
sin(t)
_
= C
1
P
1
X.
Therefore,
cos(t) + sin(t)
_
cos(t)
sin(t)
_
= X
t
_
P
1
_
t
_
C
1
_
t
C
1
P
1
X,
cos
2
(t) + sin
2
(t) = X
t
_
P
1
_
t
_
C
1
_
t
C
1
P
1
X,
1 = X
t
_
P
1
_
t
_
C
1
_
t
C
1
P
1
X.
Note that C = C
t
, so
CC
t
=
_
C
1
C
2
C
2
C
1
__
C
1
C
2
C
2
C
1
_
C
2
=
_
C
2
1
+C
2
2
0
0 C
2
1
+C
2
2
_
=
_
C
2
1
+C
2
2
_
I.
Therefore,
C
1
=
C
C
2
1
+C
2
2
,
_
C
1
_
t
= C
1
=
C
C
2
1
+C
2
2
,
_
C
1
_
t
C
1
=
_
C
1
_
2
=
_
C
2
1
+C
2
2
_
I
(C
2
1
+C
2
2
)
2
=
I
C
2
1
+C
2
2
.
Therefore,
1 = X
t
_
P
1
_
t
_
C
1
_
t
C
1
P
1
X =
1
C
2
1
+C
2
2
X
t
_
P
1
_
t
P
1
X.
Let T =
_
P
1
_
t
P
1
. Then X
t
TX = C
2
1
+ C
2
2
and T = T
t
(T is symmetric).
Therefore, the eigenvectors of T are mutually orthogonal and form the axes of
the ellipses. Figure 9.3 shows a stable spiral and an unstable spiral.
9.5. PHASE PORTRAITS 141
0
0
0
0
(a) < 0
0
0
0
0
(b) > 0
Figure 9.3: The cases for , where we have a stable spiral when > 0 and an
unstable spiral when > 0.
9.5.3 Repeated Real Roots
We have N = AI, where N
2
= 0 and A = N+I. So
e
At
= e
Nt+tI
= e
Nt
e
tI
= (I +Nt)
_
e
t
0
0 e
t
_
.
Therefore,
X = e
At
_
C
1
C
2
_
= (I +Nt)
_
e
t
0
0 e
t
__
C
1
C
2
_
= e
t
(I +Nt)
_
C
1
C
2
_
.
Note that N
2
= 0 [N[
2
= 0 [N[ = 0. Therefore,
N =
_
n
1
n
2
n
1
n
2
_
.
Also, N
2
= 0 tr(N) = 0 n
1
+n
2
= 0. Let
V =
_
1
_
.
Then
Nv =
_
n
1
+n
2
(n
1
+n
2
)
_
=
_
0
0
_
.
142 CHAPTER 9. LINEAR SYSTEMS
So Av = (N+I) v = v, i.e., v is an eigenvector for . Therefore,
X = e
t
(I +Nt)
_
C
1
C
2
_
= e
t
_
C
1
+ (n
1
C
1
+n
2
C
2
) t
C
2
+(n
1
+n
2
C
2
) t
_
= e
t
__
C
1
C
2
_
+ (n
1
C
1
+n
2
C
2
) t
_
1

__
= e
t
__
C
1
C
2
_
+ (n
1
C
1
+n
2
C
2
) tv
_
.
If < 0, we have
_
x
y
_

_
0
0
_
as t . If > 0, then
_
x
y
_

_
0
0
_
as t . What is the limit of the slope? In other words, what line is
approached asymptotically? We have
lim
t
y
x
= lim
t
C
2
+ (n
1
C
1
+n
2
C
2
) tv
2
C
1
+ (n
1
C
1
+n
2
C
2
) tv
1
=
v
2
v
1
,
i.e., it approaches v. Similarly,
lim
t
y
x
=
v
2
v
1
,
i.e., it also approaches v as t . Figure 9.4 illustrates the situation.
We encounter the degenerate case when N = 0. This does not work, but
then A = I, so
X = e
At
_
C
1
C
2
_
=
_
e
t
0
0 e
t
__
C
1
C
2
_
= e
t
_
C
1
C
2
_
,
which is just a straight line through
_
C
1
C
2
_
.
Figure 9.5 illustrates this situation.
9.5. PHASE PORTRAITS 143
0
0
0
0
(a) < 0
0
0
0
0
(b) > 0
Figure 9.4: The cases for , where we have a stable node when < 0 and an
unstable node when > 0.
0
0
0
0
(a) < 0
0
0
0
0
(b) > 0
Figure 9.5: The degenerate cases for when N = 0, where we have a stable
node when < 0 and an unstable node when > 0.
144 CHAPTER 9. LINEAR SYSTEMS
Chapter 10
Existence and Uniqueness
Theorems
10.1 Picards Method
Consider the general rst order ivp
_
_
_
dy
dx
= f(x, y),
y(x
0
) = y
0
.
()
Picards Method is a technique for approximating the solution to Equation ().
If y(x) is a solution of Equation (), then
_
x
x0
dy
dt
dt =
_
x
x0
f(t, y(t)) dt,
_
y
y0
dy = y y
0
,
giving us
y = y
0
+
_
x
x0
f(t, y(t)) dt. ()
Conversely, if y satises (), then dierentiating gives
dy
dx
= 0 +f(x, y(x)) = f(x, y)
145
146 CHAPTER 10. EXISTENCE AND UNIQUENESS THEOREMS
and y(x
0
) = y
0
+ 0 = y
0
. Therefore, the ivp () is equivalent to the integral
equation ().
If p(x) is a function, dene a new function P(p) by
(P(p)) (x) = y
0
+
_
x
x0
f(t, p(t)) dt.
So a solution to () is a function y(x) such that P(y) = y.
Picards Method is as follows. Start with a function y
0
(x) for which y
0
(x
0
) =
y
0
. Let y
1
= P(y
0
), y
2
= P(y
1
), . . . , y
n
= P(y
n1
), . . ., and so on. Under the
right conditions (discussed later), the function y
n
will converge to a solution y.
Intuitively, we let y = lim
n
y
n
. Since y
n+1
= P(y
n
), taking limits as n
gives y = P(y). Later, we will attempt to make this mathematically precise.
Example 10.1. Solve y

= y, where y(0) = 1.
Solution. We begin with y
0
(x) 1. With x
0
= 0, y
0
= 1, f(x, y) = y, we
have
y
1
(x) = y
0
+
_
x
x0
f
_
1
x
, y
0
(t)
_
dt
= 1 +
_
x
0
(1) dt = 1 x,
and
y
2
(x) = y
0
+
_
x
x0
f(t, y
1
(t)) dt
= 1 +
_
x
0
(1 t) dt = 1 x +
x
2
2
,
and
y
3
(x) = 1 +
_
x
0

_
1 t +
t
2
2
_
dt = 1 x +
x
2
2

x
3
3!
.
So in general, we have
y
n
(x) = 1 x +
x
2
2!
+ + (1)
n
x
n
n!
.
Therefore, we nally have
y(x) = lim
n
y
n
(x) =

n=0
(1)
n
x
n
n!
.
10.1. PICARDS METHOD 147
In this case, we recognize the solution as y = e
x
.
Example 10.2. Solve y

= x +y
2
, where y(1) = 1.
Solution. Let y
0
(x) 1. With x
0
= 1, y
0
= 1, and f(x, y) = x +y
2
, we have
y
1
(x) = 1 +
_
x
1
_
t + 1
2
_
dt = 1 +
_
x
1
[(t 1) + 2] dt
= 1 +
_
(t 1)
2
2
+ 2 (t 1)
_
x
1
= 1 +
(x 1)
2
2
+ 2 (x 1)
= 1 + 2 (x 1) +
(x 1)
2
2
and
y
2
(x) = 1 +
_
x
1
_
_
t +
_
1 + 2 (t 1) +
(t 1)
2
2
_
2
_
_
dt
= 1 + 2 (x 1) +
5
2
(x 1)
2
+
5
3
(x 1)
3
+
1
4
(x 1)
4
+
1
20
(x 1)
5
.
Picards Method is not convenient for actually nding solutions, but it is
useful for proving that solutions exist under the right conditions.
Theorem 10.3
If f : X R is continuous, where X R
n
is closed and bounded (compact),
then f(X) is closed and bounded (compact). In particular, there exists an
M such that [f(x)[ M for all x X.
Proof. Proof is given in MATB43.
Corollary 10.4
Let R be a (closed) rectangle. Let f(x, y) be such that f/y exists and is
continuous throughout R. Then there exists an M such that
[f(x, y
2
) f(x, y
1
)[ M [y
2
y
1
[
. .
Lipschitz condition with respect to y
for any points (x, y
1
) and (x, y
2
) in R.
148 CHAPTER 10. EXISTENCE AND UNIQUENESS THEOREMS
Proof. Let R be a (closed) rectangle. Let f(x, y) be such that f/y exists and
is continuous throughout R. By Theorem 10.3, there exists an M such that

f
y
(r)

M
for all r R. By the Mean Value Theorem, there exists a c (y
1
, y
2
) such that
f(x, y
2
) f(x, y
1
) =
f
y
(x, c)(y
2
y
1
).
Since R is a rectangle, we have (x, c) R, so

f
y
(x, c)

M.
Therefore, [f(x, y
2
) f(x, y
1
)[ M [y
2
y
1
[.
Theorem 10.5
Let f be a function. Then
1. f is dierentiable at x implies that f is continuous at x.
2. f is continuous on B implies that f is integrable on B, i.e.,
_
B
f dV
exists.
Proof. Proof is given in MATB43.
Denition (Uniform convergence)
A sequence of functions f
n
dened on B is said to converge uniformly to a
function f if for all > 0, there exists an N such that [f(x) f
n
(x)[ < for all
n N.

The point is that the same N works for all xs. If each x has an Nx that worked for it
but there was no N working for all xs at once, then it would converge, but not uniformly.
10.1. PICARDS METHOD 149
Theorem 10.6
Suppose that f
n
is a sequence of functions that converges uniformly to f
on B. if f
n
is integrable for all n, then f is integrable and
_
B
f(x) dV = lim
n
_
B
f
n
(x) dV.
Proof. Proof is given in MATB43.
Theorem 10.7
Let f
n
be a sequence of functions dened on B. Suppose that there exists
a positive convergent series

n=1
a
n
such that [f
n
(x) f
n1
(x)[ a
n
for
all n. Then f(x) = lim
n
f
n
(x) exists for all x B and f
n
converges
uniformly to f.
Proof. Let f
n
be a sequence of functions dened on B. Suppose that there
exists a positive convergent series

n=1
a
n
such that [f
n
(x) f
n1
(x)[ a
n
for all n. Let
f
n
(x) = f
0
(x) + (f
1
(x) f
0
(x)) + + (f
n
(x) f
n1
(x))
= f
0
(x) +
n

k=1
(f
k
(x) f
k1
(x)) .
Therefore
lim
n
f
n
(x)
. .
f(x)
= f
0
(x) +

k=1
(f
k
(x) f
k1
(x))
and
[f(x) f
n
(x)[ =

k=n+1
(f
k
(x) f
k1
(x))

k=n+1
a
k
<
for n suciently large, since

n=1
a
n
converges. Therefore, f
n
converges uni-
formly to f.
150 CHAPTER 10. EXISTENCE AND UNIQUENESS THEOREMS
10.2 Existence and Uniqueness Theorem for First
Order ODEs
Theorem 10.8 (Existence and uniqueness theorem for rst order odes

)
Let R be a rectangle and let f(x, y) be continuous throughout R and satisfy
the Lipschitz Condition with respect to y throughout R. Let (x
0
, y
0
) be an
interior point of R. Then there exists an interval containing x
0
on which
there exists a unique function y(x) satisfying y

= f(x, y) and y(x


0
) = y
0
.

To prove the existence statement in Theorem 10.8, we will need the following
lemma.
Lemma 10.9
Let I = [x
0
, x
0
+] and let p(x) satisfy the following:
1. p(x) is continuous on I.
2. [p(x) y
0
[ a for all x I.
Then
_
x
x0
f(t, p(t)) dt exists for all x I and q(x) = y
0
+
_
x
x0
f(t, p(t)) dt
also satises (1) and (2).
Proof. Let I = [x
0
, x
0
+] and let p(x) satisfy the following:
1. p(x) is continuous on I.
2. [p(x) y
0
[ a for all x I.
Immediately, (2) implies that (t, p(t)) R for t I. We have
I R
f
R
t (t, p(t))
so f is continuous, hence integrable, i.e.,
_
x
x0
f(t, p(t)) dt exists for all x
I. Since q(x) is dierentiable on I, it is also continuous on I. Also, since
10.2. EXISTENCE ANDUNIQUENESS THEOREMFOR FIRST ORDER ODES151
[f(t, p(t))[ M, it follows that
[q(x) y
0
[ =

_
x
x0
f(t, p(t)) dt

M [x x
0
[ M a.

Proof (Proof of existence statement (Theorem 10.8)). Let R be a rectangle and


let f(x, y) be continuous throughout R that satisfy the Lipschitz Condition with
respect to y throughout R. Let (x
0
, y
0
) be an interior point of R. Since f is
continuous, there exists an M such that [f(r)[ M for all r R. Let a be the
distance from (x
0
, y
0
) to the boundary of R. Since (x
0
, y
0
) is an interior point,
we have a > 0. Let = min(a, a/M). We will show that y

= f(x, y), where


y(x
0
) = y
0
, has a unique solution on the domain I = [x
0
, x
0
+].
Inductively dene functions (1) and (2) of Lemma 10.9 by y
0
(x) y
0
for all
x I and
y
n
(x) = y
0
+
_
x
x0
f(t, y
n1
(t)) dt.
By Lemma 10.9, the existence of y
n1
satisfying (1) and (2) guarantees the
existence of y
n
. By hypothesis, there exists an A such that [f(x, v) f(x, w)[
A[v w[, whenever (x, v), (x, w) R. As in the proof of Lemma 10.9, we have
[y
1
(x) y
0
(x)[ = [y
1
(x) y
0
[ M [x x
0
[
for all x I. Likewise, we have
[y
2
(x) y
1
(x)[ =

_
x
x0
[f(t, y
1
(t)) f(t, y
0
(t))] dt

_
x
x0
[f(t, y
1
(t)) f(t, y
0
(t))[ dt

_
x
x0
A[y
1
(t) y
0
(t)[ dt

_
x
x0
AM [t x
0
[ dt

MA
[x x
0
[
2
2
,
152 CHAPTER 10. EXISTENCE AND UNIQUENESS THEOREMS
and
[y
3
(x) y
2
(x)[ =

_
x
x0
[f(t, y
2
(t)) f(t, y
1
(t))] dt

_
x
x0
[f(t, y
2
(t)) f(t, y
1
(t))[ dt

_
x
x0
_
AMA
[t x
0
[
2
2
_
dt

MA
2
[x x
0
[
3
3!
.
Continuing, we get
[y
n
(x) y
n1
(x)[
MA
n1

n
n!
.
We have
y
n
(x) = y
0
(x) + (y
1
(x) y
0
(x)) + (y
2
(x) y
1
(x)) + + (y
n
(x) y
n1
(x))
= y
0
(x) +
n

k=1
(y
k
(x) y
k1
(x)) .
Since
[y
k
(x) y
k1
(x)[
MA
k1

k
k!
and

k=1
MA
k1

k
k!
=
M
A
_
e
A
1
_
,
converges, we must have
lim
n
y
n
(x) = y
0
(x) +

k=1
(y
k
(x) y
k1
(x))
converging for all x I. Therefore, y(x) = lim
n
y
n
(x) exists for all x I
and y
n
converges uniformly to y.
Note that
[f(x, y
n
(x)) f(x, y
n1
(x))[ A[y
n
(x) y
n1
(x)[
10.2. EXISTENCE ANDUNIQUENESS THEOREMFOR FIRST ORDER ODES153
implies that f(x, y
n
(x)) converges uniformly to f(x, y(x)). With
y
n
(x) = y
0
+
_
x
x0
f(y, y
n1
(t)) dt,
we then have
y(x) = lim
n
y
n
(x) = lim
n
_
y
0
+
_
x
x0
f(t, y
n1
(t)) dt
_
= y
0
+
_
x
x0
_
lim
n
f(t, y
n1
(t))
_
dt = y
0
+
_
x
x0
f(t, y(t)) dt
implying that y satises y

(x) = f(x, y) and y(x


0
) = y
0
.
To prove the uniqueness statement in Theorem 10.8, we will need the fol-
lowing lemma.
Lemma 10.10
If y(x) is a solution to the de with y(x
0
) = y
0
, then [y(x) y
0
[ a for all
x I. In particular, (x, y(x)) R for all x I.
Proof. Suppose there exist an x I such that [y(x) y
0
[ > a. We now consider
two cases.
Suppose x > x
0
. Let s inf(t : t > x
0
and [y(t) y
0
[ > a). So s < x
x
0
+ x
0
+a. By continuity, we have
a = [y(s) y
0
[ = [y(s) y(x
0
)[ = [y

(c) (s x
0
)[ = [f(c, y(c)) (s x
0
)[
. .
M(sx0)
for some c (x
0
, s). However,
s I = s x
0
= M (s x
0
) M a.
Therefore,
M (s x
0
) = Ma = s x
0
= a = s = x
0
+a,
which is a contradiction because we have s < x
0
+ a. The case with x < x
0
is
similarly shown.
Proof (Proof of uniqueness statement (Theorem 10.8)). Suppose y(x) and z(x)
154 CHAPTER 10. EXISTENCE AND UNIQUENESS THEOREMS
both satisfy the given de and initial condition. Let (x) = (y(x) z(x))
2
. Then

(x) = 2 (y(x) z(x)) (y

(x) z

(x))
= 2 (y(x) z(x)) (f(x, y(x)) f(x, z(x))) .
Therefore for all x I,
[

(x)[ 2 [y(x) z(x)[ A[y(x) z(x)[


= 2A(y(x) z(x))
2
= 2A(x).
Therefore, 2A(x)

(x) 2A(x), and we have

(x) 2A(x) =

(x) 2A(x) 0
= e
2Ax
(

(x) 2A(x)) 0,

(x) =
d
_
e
2Ax
(x)
_
dx
= e
2Ax
(x) decreasing on I
= e
2Ax
(x) e
2Ax0
(x
0
)
= (x) 0
if x x
0
. Similarly,

(x) 2A(x) =

(x) + 2A(x) 0
= e
2Ax
(

(x) + 2A(x)) 0,

(x) =
d
_
e
2Ax
(x)
_
dx
= e
2Ax
(x) increasing on I
= e
2Ax
(x) e
2Ax0
(x
0
)
= (x) 0
if x x
0
. Therefore, (x) 0 for all x I. But obviously, (x) 0 for all
x from its denition. Therefore, (x) 0, i.e., y(x) z(x). Therefore, the
solution is unique.
10.3. EXISTENCE ANDUNIQUENESS THEOREMFOR LINEAR FIRST ORDER ODES155
10.3 Existence and Uniqueness Theorem for Lin-
ear First Order ODEs
Here, we consider the special case of a linear de
dy
dx
= a(x)y +b(x).
Theorem 10.11 (Existence and uniqueness theorem for linear rst order des)
Let a(x) and b(x) be continuous throughout an interval I. Let x
0
be an
interior point of I and let y
0
be arbitrary. Then there exists a unique
function y(x) with y(x
0
) = y
0
satisfying dy/dx = a(x)y + b(x) throughout
I.

Proof. Let a(x) and b(x) be continuous throughout an interval I. Let x


0
be
an interior point of I and let y
0
be arbitrary. Let A = max([a(x)[ : x I).
Inductively dene functions

on I by y
0
(x) = y
0
and
y
n+1
(x) = y
0
+
_
x
x0
(a(t)y
n
(t) +b(t)) dt.
Then
[y
n
(x) y
n1
(x)[ =

y
0
+
_
x
x0
(a(t)y
n1
(t) +b(t)) dt

y
0

_
x
x0
(a(t)y
n2
(t) +b(t)) dt

_
x
x0
[a(t) (y
n1
(t) y
n2
(t))] dt

.
Assuming by induction that
[y
n1
(x) y
n2
(x)[ A
n2
[x x
0
[
n
(n 1)!
,

Assuming by induction that yn(t) is continuous, a(t)yn(t) + b(t) is continuous through-


out I and thus integrable, so the denition of y
n+1
(x) makes sense. Then y
n+1
(x) is also
dierentiable, i.e., y

n+1
(x) = a(x)yn(x) + b(x), and thus continuous, so the induction can
proceed.
156 CHAPTER 10. EXISTENCE AND UNIQUENESS THEOREMS
we then have

_
x
x0
[a(t) (y
n1
(t) y
n2
(t))] dt

_
x
x0
_
AA
n2
[t x
0
[
n1
(n 1)!
_
dt

A
n1
[x x
0
[
n
n!
,
thereby completing the induction. Therefore,
[y
n
(x) y
n1
(x)[
A
n1

n
n!
,
where is the width of I. The rest of the proof is as before.
10.4 Existence and Uniqueness Theorem for Lin-
ear Systems
A system of dierential equations consists of n equations involving n functions
and their derivatives. A solution of the system consists of n functions having
the property that the n functional equations obtained by substituting these
functions are their derivatives into the system of equations, and they hold for
every point in some domain D. A linear system of dierential equations has the
form

dy
1
dx
= a
11
(x)y
1
+a
12
(x)y
2
+ +a
1n
(x)y
n
+b
1
(x),
dy
2
dx
= a
21
(x)y
1
+a
22
(x)y
2
+ +a
2n
(x)y
n
+b
2
(x),
.
.
.
dy
n
dx
= a
n1
(x)y
1
+a
n2
(x)y
2
+ +a
nn
(x)y
n
+b
n
(x).
We can write the system as
dY
dx
= AY +B,

The point is that y


1
, y
2
, . . . , yn and their derivatives appears linearly.
10.4. EXISTENCE ANDUNIQUENESS THEOREMFOR LINEAR SYSTEMS157
where
A =
_

_
a
11
(x) a
1n
(x)
a
21
(x) a
2n
(x)
.
.
.
.
.
.
.
.
.
a
n1
(x) a
nn
(x)
_

_
, Y =
_

_
y
1
(x)
y
2
(x)
.
.
.
y
n
(x)
_

_
, B =
_

_
b
1
(x)
b
2
(x)
.
.
.
b
n
(x)
_

_
.
Theorem 10.12 (Existence and uniqueness theorem for linear systems of n des)
Let A(x) be a matrix of functions, each continuous throughout an interval
I and let B(x) be an n-dimensional vector of functions, each continuous
throughout I. Let x
0
be an interior point of I and let Y
0
be an arbitrary
n-dimensional vector. Then there exists a unique vector of functions Y(x)
with Y(x
0
) = Y
0
satisfying
dY
dx
= A(x)Y +B(x)
throughout I.

Proof. Let A(x) be a matrix of functions, each continuous throughout an in-


terval I and let B(x) be an n-dimensional vector of functions, each continuous
throughout I. Let x
0
be an interior point of I and let Y
0
be an arbitrary
n-dimensional vector.
Let = max(|A(x)| : x I). Inductively dene Y
0
(x) = Y
0
and
Y
n+1
(x) +Y
0
+
_
x
x0
(A(t)Y
n
(t) +B(t)) dt
. .
vector obtained by
integrating componentwise
.
158 CHAPTER 10. EXISTENCE AND UNIQUENESS THEOREMS
Then
[Y
n
(x) Y
n1
(x)[ =

_
x
x0
[A(t) (Y
n1
(t) Y
n2
(t))] dt

_
x
x0
_

n1
[t x
0
[
n1
(n 1)!
_
dt

=
n
[x x
0
[
n
n!

n

n
,
where is the width of I. The rest of the proof is essentially the same as
before.
Recall Theorem 8.5 (p. 90) restated here as follows.
Theorem 10.13 (Theorem 8.5)
Let x = p be an ordinary point of y

+ P(x)y

+ Q(x)y = 0. Let R be the


distance from p to the closest singular point of the de in the complex plane.
Then the de has two series y
1
(x) =

n=0
a
n
x
n
and y
2
(x) =

n=0
b
n
x
n
which converge to linearly independent solutions to the de on the interval
[x p[ < R.

Proof. Let x = p be an ordinary point of y

+ P(x)y

+ Q(x)y = 0. Let R be
the distance from p to the closest singular point of the de in the complex plane.
We rst claim that if f(z) is analytic at p and R is the radius of convergence
of f, then for r < R, there exists an M (depending on r) such that
1
n!

f
(n)
(p)


M
r
n
()
for all n N. To prove this claim, since

n=0
1
n
f
n
(p)r
n
= f(r)
converges absolutely, let
M =

n=0
1
n!
[f
n
(p)[ r
n
.
10.4. EXISTENCE ANDUNIQUENESS THEOREMFOR LINEAR SYSTEMS159
Then
1
n!
f
(n)
(p)r
n
M,
so Equation () holds. Therefore, our claim holds.
Let w(z) be a solution of y

+ P(x)y

+ Q(x)y = 0. Being a solution to


the de, w is twice dierentiable. Furthermore, dierentiating the de gives a
formula for w

in terms of w, w

, w

. Continuing, w is n times dierentiable


for all n, so w has a Taylor series. We now wish to show that the Taylor series
of w converges to w for [z p[ < R.
Since [z p[ < R, there exists an a such that [z p[ < a < R. As above,
nd constants M such that
1
n!

P
(n)
(p)


M
a
n
,
1
n!

Q
(n)
(p)


N
a
n
for all n N. Let
A(z) =
M
1
zp
a
, B(z) =
N
1
zp
a
.
Note that
A(z) = M
_
1 +
z p
a
+
_
z p
a
_
2
+ +
_
z p
a
_
n
+
_
,
so
A(p) = M,
A

(p) =
M
a
,
A

(p) = 2
M
a
2
,
.
.
.
A
(n)
(p) = n!
M
a
n

P
(n)
(p)

.
Similarly,
B
(n)
(p) = n!
N
a
n

Q
(n)
(p)

.
Consider
y

= A(z)y

+B(z)y. ()
160 CHAPTER 10. EXISTENCE AND UNIQUENESS THEOREMS
Let v(z) be a solution of Equation () satisfying v(p) = [w(p)[ and v

(p) =
[w

(p)[. In general, w
(n)
(p) = C
1
w

(p) +C
2
w(p), where C
1
and C
2
are arbitrary
constants depending on P and Q and their derivatives at p, e.g.,
w

= P(p)w

(p) +Q(p)w(p),
w

= P

(p)w

(p) +P(p)w

(p) +Q

(p)w(p) +Q(p)w

(p)
= P

(p)w

(p) +P(p) (P(p)w

(p) +Q(p)w(p)) +Q

(p)w(p) +Q(p)w

(p)
=
_
P
2
(p) +P

(p) +Q(p)
_
w

(p) + (P(p)Q(p) +Q

(p)) w(p).
Similar formulas hold v
(n)
(p) involving derivatives of A and B at p.
Using

P
(n)
(p)

A
(n)
(p) and

Q
(n)
(p)

B
(n)
(p) gives

w
(n)
(p)

v
(n)
(p).
The Taylor series of W about p is

n=0
a
n
(z p)
n
, where a
n
= w
(n)
(p)/n!;
the Taylor series of v about p is

n=0
b
n
(z p)
n
, where b
n
= v
(n)
(p)/n!.
Since we showed that [a
n
[ b
n
, the rst converges anywhere the second does.
So we need to show that the Taylor series of w(x) has a radius of convergence
equal to a (this implies that the Taylor series for w converges for n pk, but
a < R was arbitrary). We have
v

=
M
1
zp
a
v

+
N
1
zp
a
v.
Let u = (z p) /a. Then
v

=
dv
du
1
a
dv
du
.
v

= a
d
2
v
du
2
du
dz
=
1
a
2
d
2
v
du
2
.
Then we have
1
a
2
d
2
v
du
2
=
M
1 u
1
a
dv
du
+
N
1 u
v,
(1 u)
d
2
v
du
2
= aM
dv
du
+a
2
Nv.
Write v =

n=0

n
u
n
. Using its rst radius of convergence, we have
v =
n
u
n
=
n
_
z p
a
_
n
=

n
a
n
(z p)
n
= b
n
(z p)
n
,
10.4. EXISTENCE ANDUNIQUENESS THEOREMFOR LINEAR SYSTEMS161
which implies that
n
= b
n
a
n
> 0. Note that
dv
du
=

n=1
n
n
u
n1
. .
adjust indices
=

n=0
(n + 1)
n+1
u
n
,
d
2
v
du
2
=

n=1
(n + 1) n
n+1
u
n1
. .
adjust indices
=

n=0
(n + 2) (n + 1)
n+2
u
n
.
Now we have
(1 u)

n=0
(n + 2) (n + 1)
n+2
u
n
=

n=0
aM (n + 1)
n+1
u
n
+

n=0
a
2
N
n
u
n
,
_
_
_
_
_
_

n=0
(n + 2) (n + 1)
n+2
u
n

n=0
(n + 2) (n + 1)
n+2
u
n+1
_
_
_
_
_
_
=
_
_
_
_
_
_

n=0
(n + 2) (n + 1)
n+2
u
n

n=0
(n + 1) n
n+1
u
n
.
_
_
_
_
_
_
Therefore,

n=0
(n + 2) (n + 1)
n+2
u
n
=

n=0
_
(n + 1) (n +aM)
n+1
+a
2
N
n

u
n
,
so it follows that
(n + 2) (n + 1)
n+2
= (n + 1) (aM +n)
n+1
+a
2
N
n
,

n+2
=
aM +n
n + 2

n+1
+
a
2
N
n
n + 2
,

n+2

n+1
=
n +aM
n + 2
+
a
2
N
n + 2

n+1
for all n N. Since
n
/
n+1
< 1, we have
lim
n
a
2
N
n + 2

n+1
= 0.
Therefore,
lim
n

n+2

n+1
= 1 + 0 = 1.
So by the ratio test (Equation (8.2) of Theorem 8.2, p. 88), the radius conver-
162 CHAPTER 10. EXISTENCE AND UNIQUENESS THEOREMS
gence of

n=0

n
u
n
=

n=0

n
a
n
(z p)
n
=

n=0
b
n
_
z p
a
_
n
= 1
is a.
The point of Theorem 8.5 is to state that solutions are analytic (we already
know solutions exists from early theorems). The theorem gives only a lower
bound on the interval. The actual radius of convergence may be larger.
Chapter 11
Numerical Approximations
11.1 Eulers Method
Consider y

= f(x, y), where y(x


0
) = y
0
. The goal is to nd y(x
end
). Figure
11.1 shows this idea. We pick a large N and divide [x
0
, x
end
] into segments of
actual value
approximate value
x
0
x
0
+h
= x
1
x
0
+2h
= x
2
...
x
end
= x
N
y
0
y
1
y
2
y
N
Figure 11.1: Graphical representation of Eulers method.
length h = (x
end
x
0
) /N. Then
y(x
1
) y(x
0
) +y

(x
0
) (x
1
x
0
)
. .
h
163
164 CHAPTER 11. NUMERICAL APPROXIMATIONS
is the linear/tangent line approximation of y(x
1
). Set
y
1
= y
0
+hf(x
0
, y
0
) y
1
f(x
1
),
y
2
= y
1
+hf(x
1
, y
1
) y
2
f(x
2
),
y
3
= y
2
+hf(x
2
, y
2
) y
3
f(x
3
),
.
.
.
y
answer
= y
N1
+hf(x
N1
, y
N1
) y
answer
f(x
end
).
Example 11.1. Consider y

= x 2y, where y(0) = 1. Approximate y(0.3)


using a step size of h = 0.1.
Solution. First note that x
0
= 0. With a step size of h = 0.1, we have
x
1
= 0.1, x
2
= 0.2, x
3
= 0.3.
Therefore,
y
1
= y
0
+hf(x
0
, y
0
) = 1 + 0.1 (0 2)
= 1 0.2 = 0.8,
y
2
= 0.8 +hf(0.1, 0.8) = 0.8 +h(0.1 1.6)
= 0.8 0.1 1.5 = 0.8 0.15 = 0.65,
y
3
= 0.65 +hf(0.2, 0.65) = 0.65 +h(0.2 1.3)
= 0.65 0.1 1.1 = 0.65 0.11 = 0.54.
Therefore, y(0.3) 0.54.
What about the actual value? To solve the de, we have
y

+ 2y = x,
e
2x
y

+ 2e
2x
y = xe
2x
,
ye
2x
=
_
xe
2x
dx =
1
2
xe
2x

_
1
2
e
2x
dx
=
1
2
xe
2x

1
4
e
2x
+C,
11.1. EULERS METHOD 165
where C is an arbitrary constant. Therefore
y =
1
2
x
1
4
+Ce
2x
.
Note that
y(0) = 1 =
1
4
+C = 1 = C =
5
4
,
so the general solution is
y =
1
2
x
1
4
+
5
4
e
2x
.
Therefore,
y(0.3) = 0.15 0.25 +
5
4
e
0.6
0.1 + 0.68601 0.58601,
so y(0.3) = 0.58601 is the (approximated) actual value.
11.1.1 Error Bounds
The Taylor second degree polynomial of a function y with step size h is given
by
y(a +h) = y(a) +hy

(a) +h
2
y

(z)
2
for some z [a, h], where h
2
y

(z)/2 is the local truncation error, i.e., the error


in each step. We have
h
2
y

(z)
2

M
2
h
2
if [y

(z)[ M on [a, h]. Therefore, dividing h by, say, 3 reduces the local trun-
cation error by 9. However, errors also accumulate due to y
n+1
being calculated
using the approximation of y
n
instead of the exact value of f(x
n
). Reducing h
increases N, since N = (x
end
x
0
) /h, so this eect wipes out part of the local
gain from the smaller h. The net eect is that
(overall error global truncation error)

Mh,
e.g., dividing h by 3 divides the overall error only by 3, not 9.
166 CHAPTER 11. NUMERICAL APPROXIMATIONS
11.2 Improved Eulers Method
To improve upon Eulers method, we consider
y
n+1
= y
n
h,
where, instead of using f

(x
n
), it is better to use the average of the left and
right edges, so
=
f

(x
n
) +f

(x
n+1
2
=
f(x
n
, y
n
) +f(x
n+1
, y
n+1
)
2
.
Figure 11.2 illustrates this idea. However, we have one problem in that we do
x
n
x
n+1
Figure 11.2: A graphical representation of the improved Eulers method.
not know y
n+1
(that is what we are working out at this stage). Instead, we ll
into the inside y
n+1
from the original Eulers method, i.e.,
1
2
(f(x
n
, y
n
) +f(x
n+1
, y
n
+hf(x
n
, y
n
))) ,
i.e.,
y
n+1
= y
n
+
h
2
[f(x
n
, y
n
) +f(x
n+1
, y
n
+hf(x
n
, y
n
))] .
This improves the local truncation error from h
2
to h
3
.
11.3. RUNGE-KUTTA METHODS 167
11.3 Runge-Kutta Methods
We again modify Eulers method and consider
y
n+1
= y
n
h .
At each step, for , we use some average of the values over the interval [y
n
, y
n+1
].
The most common one is x
n
+h/2. Figure 11.3 illustrates this idea. Set
x
n
x
n + h/2
x
n + 1
= x
n
+ h
Figure 11.3: A graphical representation of the Runge-Kutta Methods.
k
n1
= f(x
n
, y
n
),
k
n2
= f
_
x
n
+
1
2
h, y
n
+
1
2
hk
n1
_
,
k
n3
= f
_
x
n
+
1
2
h, y
n
+
1
2
hk
n2
_
,
k
n4
= f
_
x
n
+
1
2
h, y
n
+
1
2
hk
n3
_
.
Use
y
n+1
= y
n
+
h
6
(k
n1
+ 2k
n2
+ 2k
n3
+ 3k
n4
) .
This gives a local truncation error proportional to h
5
.
168 CHAPTER 11. NUMERICAL APPROXIMATIONS