Ordinary Differential Equations Notes

MA3220 Ordinary Differential Equations
Wong Yan Loi
November 20, 2007

2
Contents
1 First Order Differential Equations 5

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Exact Equations, Integrating Factors . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 First Order Linear Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4 First Order Implicit Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2 Higher Order Linear Equations 17

2.1 General Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2 Linear Equations with Constant Coefficients . . . . . . . . . . . . . . . . . . . . . 22
2.3 Operator methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.4 Exact 2nd order Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.5 The adjoint differential equation and integrating factor . . . . . . . . . . . . . . . 32
3 Linear Differential Systems 35

3.1 Linear Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2 Homogeneous Linear Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.3 Non-Homogeneous Linear Systems . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.4 Homogeneous Linear Systems with Constant Coefficients . . . . . . . . . . . . . . 40
3.5 Higher Order Linear Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.6 Appendix 1: Proof of Lemma 3.12 . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4 Power Series Solutions 57

4.1 Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.2 Series Solutions of First Order Equations . . . . . . . . . . . . . . . . . . . . . . 60
4.3 Second Order Linear Equations and Ordinary Points . . . . . . . . . . . . . . . . . 61
4.4 Regular singular points and the method of Frobenius . . . . . . . . . . . . . . . . 65
4.9 Bessel’s equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5 Fundamental Theory of ODEs 77

5.1 Existence-Uniqueness Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.2 The method of successive approximations . . . . . . . . . . . . . . . . . . . . . . 78
5.3 Convergence of the successive approximations . . . . . . . . . . . . . . . . . . . . 81
5.4 Non-local Existence of Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3
4 CONTENTS
5.5 Gronwall’s Inequality and Uniqueness of Solution . . . . . . . . . . . . . . . . . . 87

5.6 Existence and Uniqueness of Solutions to Systems . . . . . . . . . . . . . . . . . 90
Chapter 1
First Order Differential Equations
1.1 Introduction
1. Ordinary differential equations.
An ordinary differential equation (ODE for short) is a relation containing one real variable x, the
real dependent variable y, and some of its derivatives y 0 , y 00 , · · · , y (n) , · · · , with respect to x.
The order of an ODE is defined to be the order of the highest derivative that occurs in the equation.
Thus, an n-th order ODE has the general form
F (x, y, y 0 , · · · , y (n) ) = 0. (1.1.1)
We shall always assume that (1.1.1) can be solved explicitly for y (n) in terms of the remaining n + 1
quantities as
y (n) = f (x, y, y 0 , · · · , y (n−1) ), (1.1.2)
where f is a known function of x, y, y 0 , · · · , y (n−1) .

An n-th order ODE is linear if it can be written in the form
a0 (x)y (n) + a1 (x)y (n−1) + · · · + an (x)y = r(x). (1.1.3)
The functions aj (x), 0 ≤ j ≤ n are called coefficients of the equation. We shall always assume
that a0 (x) 6≡ 0 in any interval in which the equation is defined. If r(x) ≡ 0, (1.1.3) is called a
homogeneous equation. If r(x) 6≡ 0, (1.1.3) is said to be a non-homogeneous equation, and r(x) is
called the non-homogeneous term.
2. Solutions.
A functional relation between the dependent variable y and the independent variable x that satisfies
the given ODE in some interval J is called a solution of the given ODE on J.
A general solution of an n-th order ODE depends on n arbitrary constants, i.e. the solution y
depends on x and n real constants c1 , · · · , cn .
A first order ODE may be written as
F (x, y, y 0 ) = 0. (1.1.4)
5
6 CHAPTER 1. FIRST ORDER DIFFERENTIAL EQUATIONS
In this chapter we consider only first order ODE. The function y = φ(x) is called an explicit solution
of (1.1.4) in the interval J provided
F (x, φ(x), φ0 (x)) = 0 for all x in J. (1.1.5)
A relation of the form ψ(x, y) = 0 is said to be an implicit solution of (1.1.4) provided it determines
one or more functions y = φ(x) which satisfy (1.1.5). The pair of equations
x = x(t), y = y(t) (1.1.6)
is said to be a parametric solution of (1.1.4) if

ẏ(t)
F (x(t), y(t), ) = 0.
ẋ(t)
Example. Consider the ODE x + yy 0 = 0 for x ∈ (−1, 1).

x2 + y 2 = 1 is an implicit solution while x = cos t, y = sin t, t ∈ (0, π) is a parametric solution.
3. Integral curves.
The solutions of an ODE
y 0 = f (x, y) (1.1.7)
represent a one-parameter family of curves in the xy-plane. These are called integral curves.
In other words, if y = y(x) is a solution to (1.1.7), then vector field F(x, y) = h1, f (x, y)i is tangent
to the curve r(x) = hx, y(x)i at every point (x, y) since r0 (x) = F(x, y).
4. Elimination of constants: formation of ODE.

0000
Example. The family of functions y = Aex + B sin x satisfies the ODE: y − y = 0 when the
constants A and B are eliminated using the derivatives.
5. Separable equations.
Typical separable equation can be written as
f (x)
y0 = , or g(y)dy = f (x)dx. (1.1.8)
g(y)
The solution is given by Z Z
g(y)dy = f (x)dx + c.
Example. Solve y 0 = −2xy, y(0) = 1.

2
Ans: y = e−x .
The equation y 0 = f ( xy ) can be reduced to a separable equation by letting u = y

x, i.e. y = xu. So
f (u) = y 0 = u + xu0 , Z Z
du dx
= + c.
f (u) − u x
Example. Solve 2xyy 0 + x2 − y 2 = 0.
1.1. INTRODUCTION 7
Ans: x2 + y 2 = cx.
6. Homogeneous equations.
A function is called homogeneous of degree n if f (tx, ty) = tn f (x, y) for all x, y, t.
p
For example x2 + y 2 and x + y are homogeneous of degree 1, x2 + y 2 is homogeneous of degree
2 and sin(x/y) is homogeneous of degree 0.
The ODE M (x, y) + N (x, y)y 0 = 0 is said to be homogeneous of degree n if both M (x, y) and
N (x, y) are homogeneous of degree n.
If we write the above DE as y 0 = f (x, y), where f (x, y) = −M (x, y)/N (x, y). Then f (x, y) is
homogeneous of degree 0. To solve the DE
y 0 = f (x, y),
where f is homogeneous of degree 0, we use the substitution y = zx. Then
dy dz
=z+x .
dx dx
Thus the DE becomes
dz
z+x = f (x, zx) = x0 f (1, z) = f (1, z).
dx
Consequently, the variables can be separated to yield
dz dx
= ,
f (1, z) − z x
and integrating both sides will give the solution.
x+y
Example. Solve y 0 = x−y .
p
Ans: tan−1 (y/x) = ln x2 + y 2 + c.
Example. An equation in the form
a1 x + b1 y + c1
y0 = .
a2 x + b2 y + c2
can be reduced to a homogeneous equation by a suitable substitution x = z + h, y = w + k when

a1 b2 6= a2 b1 , where h and k are solutions of the system of linear equations a1 h + b1 k + c1 =
0, a2 h + b2 k + c2 = 0.
Example. Solve y 0 = x+y−2

x−y .

y−1
p
Ans: tan−1 x−1 = ln (x − 1)2 + (y − 1)2 + c.
Example. Solve (x + y + 1) + (2x + 2y + 1)y 0 = 0.

Ans: x + 2y + ln |x + y| = c, x + y = 0.
1.2 Exact Equations, Integrating Factors

1. Exact equations.
We can write a first order ODE in the following form
M (x, y)dx + N (x, y)dy = 0. (1.2.1)
(1.2.1) is called exact if there exists a function u(x, y) such that

∂u ∂u
M (x, y)dx + N (x, y)dy = du = dx + dy.
∂x ∂y
Once (1.2.1) is exact, the general solution is given by
u(x, y) = c.
Theorem 1.1 1.1 Assume M and N together with their first partial derivatives are continuous in
the rectangle S: |x − x0 | < a, |y − y0 | < b. A necessary and sufficient condition for (1.2.1) to be
exact is
∂M ∂N
= for all (x, y) in S. (1.2.2)
∂y ∂x
When (1.2.2) is satisfied, a general solution of (1.2.1) is given by u(x, y) = c, where
Z x Z y
u(x, y) = M (s, y)ds + N (x0 , t)dt (1.2.3)
x0 y0
and c is an arbitrary constant.

Rx
Proof. The first part is by Green’s Theorem. For the second part, we have u(x, y) = x0
M (s, y)ds+
du(x0 ,y)
φ(y). Substituting x = x0 we have u(x0 , y) = φ(y). Therefore, N (x0 , y) = dy = φ0 (y) and
Ry
φ(y) = y0 N (x0 , t)dt.
Remark. In Theorem 1.1, the rectangle S can be replaced by any region which does not include any
“hole”.
Example. Solve (x3 + 3xy 2 )dx + (3x2 y + y 3 )dy = 0.
Ans: x4 + 6x2 y 2 + y 4 = c.
2. Integrating factors.
A non-zero function µ(x, y) is an integrating factor of (1.2.1) if the equivalent differential equation
µ(x, y)M (x, y)dx + µ(x, y)N (x, y)dy = 0 (1.2.4)
is exact.
If µ is an integrating factor of (1.2.1) then (µM )y = (µN )x , i.e.
N µx − M µy = µ(My − Nx ). (1.2.5)
One may look for an integrating factor of the form µ = µ(v), where v is a known function of x and
y. Plugging into (1.2.5) we find
1 dµ My − N x
= . (1.2.6)
µ dv N vx − M vy
1.2. EXACT EQUATIONS, INTEGRATING FACTORS 9
My −Nx
If N vx −M vy is a function of v alone, say, φ(v), then
Rv
φ(v)dv
µ=e
is an integrating factor of (1.2.1).
My −Nx
Rx
φ1 (x)dx
Let v = x. If N is a function of x alone, say, φ1 (x), then e is an integrating factor
of (1.2.1).
My −Nx
Ry
φ2 (y)dy
Let v = y. If − M is a function of y alone, say, φ2 (y), then e is an integrating factor
of (1.2.1).
M −N
R xy
φ3 (v)dv
Let v = xy. If yNy−xMx is a function of v = xy alone, say φ3 (xy), then e is an integrating
factor of (1.2.1).
Example Solve (x2 y + y + 1) + x(1 + x2 )y 0 = 0.

Ans: xy + tan−1 x = c.
Example Solve (y − y 2 ) + xy 0 = 0
Ans: y = (1 − cx)−1 .
Example Solve (xy 3 + 2x2 y 2 − y 2 ) + (x2 y 2 + 2x3 y − 2x2 )y 0 = 0

Ans: exy (1/x + 2/y) = c.
3. Find integrating factors by inspection.

The following are some differential formulas that are often useful.
x ydx − xdy
d( ) =
y y2
d(xy) = xdy + ydx
d(x2 + y 2 ) = 2xdx + 2ydy
x ydx − xdy
d(tan−1 )=
y x2 + y 2
x ydx − xdy
d(log ) =
y xy
We see that the very simple ODE ydx−xdy = 0 has 1/x2 , 1/y 2 , 1/(x2 +y 2 ) and 1/xy as integrating
factors.
1.3 First Order Linear Equations

1. Homogeneous equations.
A first order homogeneous linear equation is of the form
y 0 + p(x)y = 0, (1.3.1)
Rx
where p(x) is a continuous function on an interval J. Let P (x) = a
p(s)ds. Multiplying (1.3.1)
by eP (x) , we get
d P (x)
[e y] = 0,
dx
so eP (x) y = c. The general solution of (1.3.1) is given by
y(x) = ce−P (x) , where

Z x (1.3.2)
P (x) = p(s)ds.
a
2. Non-homogeneous equations.
Now consider a first order non-homogeneous linear equation
y 0 + p(x)y = q(x), (1.3.3)

Rx
where p(x) and q(x) are continuous functions on an interval J. Let P (x) = a
p(s)ds. Multiplying
(1.3.3) by eP (x) we get
d P (x)
[e y] = eP (x) q(x).
dx
Thus Z x
P (x)
e y(x) = eP (t) q(t)dt + c.
a
The general solution is given by

Z x
y(x) = e−P (x) [ eP (t) q(t)dt + c], where
a
Z x (1.3.4)
P (x) = p(s)ds.
a
Example. Solve y 0 − y = e2x .

Ans: y = cex + e2x .
3. The Bernoulli equation.

An ODE in the form
y 0 + p(x)y = q(x)y n , (1.3.5)
where n 6= 0, 1, is called the Bernoulli equation. The functions p(x) and q(x) are continuous
functions on an interval J.
Let u = y 1−n . Substituting into (1.3.5) we get
u0 + (1 − n)p(x)u = (1 − n)q(x). (1.3.6)

1.3. FIRST ORDER LINEAR EQUATIONS 11
This is a first order linear ODE.
Example. Solve xy 0 + y = x4 y 3 .
Ans: y12 = −x4 + cx2 .
4. The Riccati equation.

An ODE of the form
y 0 = P (x) + Q(x)y + R(x)y 2 (1.3.7)
is called the Riccati equation. The functions P (x), Q(x), R(x) are continuous on an interval J. In
general, the Riccati equation cannot be solved by a sequence of integrations. However, if a particular
solution is known, then (1.3.7) can be reduced to a linear equation, and thus is solvable.
Theorem 1.2 Let y = y0 (x) be a particular solution of the Riccati equation (1.3.7). Set
Z x
H(x) = [Q(t) + 2R(t)y0 (t)]dt,
x0
h Z x i (1.3.8)
−H(x)
Z(x) = e c− eH(t) R(t)dt ,
x0
where c is an arbitrary constant. Then the general solution is given by

1
y = y0 (x) + . (1.3.9)
Z(x)
Proof. In (1.1.7) we let y = y0 (x) + u(x) to get
y00 + u0 = P + Q(y0 + u) + R(y0 + u)2 .
Since y0 satisfies (1.3.7), we have
y00 = P + Qy0 + Ry02 .
From these two equalities we get
u0 = (Q + 2Ry0 )u + Ru2 . (1.3.10)
This is a Bernoulli equation with n = 2. Set Z = u−1 and reduce (1.3.10) to
Z 0 + (Q + 2Ry0 )Z = −R. (1.3.11)
(1.3.11) is a linear equation and the solution is given by (1.3.8).
Example. Solve y 0 = y/x + x3 y 2 − x5 . Note y1 = x is a solution.

5
Ans: ce2x /5 = y−x
y+x .
From (1.3.8), (1.3.9), the general solution y of the Riccati equation (1.3.7) can be written as
cF (x) + G(x)
y= , (1.3.12)
cf (x) + g(x)
where
f (x) = e−H(x) ,
Z x
−H(x)
g(x) = −e eH(t) R(t)dt,
x0
F (x) = y0 (x)f (x), G(x) = y0 g(x) + 1.
Given four distinct functions p(x), q(x), r(x), s(x), we define the cross ratio by
(p − q)(r − s)
.
(p − s)(r − q)
Property 1. The cross ratio of four distinct particular solutions of a Riccati equation is independent
of x.
Proof. From (1.3.12), the four solutions can be written as
cj F (x) + G(x)
yj (x) = .
cj f (x) + g(x)
Computations show that
(y1 − y2 )(y3 − y4 ) (c1 − c2 )(c3 − c4 )

= .
(y1 − y4 )(y3 − y2 ) (c1 − c4 )(c3 − c2 )
The right hand is independent of x.
As a consequence we get
Property 2. Suppose y1 , y2 , y3 are three distinct particular solutions of a Riccati equation (1.3.7).
Then the general solution is given by
(y1 − y2 )(y3 − y)
= c, (1.3.13)
(y1 − y)(y3 − y2 )
where c is an arbitrary constant.
Property 3. Suppose that y1 and y2 are two distinct particular solutions of a Riccati equation (1.3.7),
then its general solution is given by
Z
y − y1
ln
= [y1 (x) − y2 (x)]R(x)dx + c, (1.3.14)
y − y2
where c is an arbitrary constant.

Proof. y and yj satisfy (1.3.7). So
y 0 − yj0 = (y − yj )[Q + R(y + yj )],

y 0 − yj0
= Q + R(y + yj ).
y − yj
1.4. FIRST ORDER IMPLICIT EQUATIONS 13
Thus
y 0 − y10 y 0 − y20
− = R(y1 − y2 ).
y − y1 y − y2
Integrating yields (1.3.14).
Example. Solve y 0 = e−x y 2 . Note y1 = ex and y2 = 0 are 2 solutions.

Ans: y = cex /(c + ex ).
1.4 First Order Implicit Equations

In the above we discussed first order explicit equations, i.e. equations in the form y 0 = f (x, y). In
this section we discuss solution of some first order explicit equations
F (x, y, y 0 ) = 0 (1.4.1)
which are not solvable in y 0 .
1. Method of differentiation.
Consider an equations solvable in y:
y = f (x, y 0 ). (1.4.2)
Let p = y 0 . Differentiating y = f (x, p) we get
[fx (x, p) − p]dx + fp (x, p)dp = 0. (1.4.3)
This is a first order explicit equation in x and p. If p = φ(x) is a solution of (1.4.3), then
y = f (x, φ(x))
is a solution of (1.4.2).
Example. Clairaut’s equation

y = xy 0 + f (y 0 ), (1.4.4)
where f has continuous second order derivative and f 00 (p) 6= 0.

Let p = y 0 . We have y = xp + f (p). Differentiating we get
[x + f 0 (p)]p0 = 0.
When p0 = 0 we have p = c and (1.4.4) has a general solution
y = cx + f (c).
When x + f 0 (p) = 0 we get a solution of (1.4.4) given by parameterized equations
x = −f 0 (p), y = px + f (p).
2. Method of parameterization.
This method can be used to solve equations where either x or y is missing. Consider
F (y, y 0 ) = 0, (1.4.5)
where x is missing. Let p = y 0 and write (1.4.5) as
F (y, p) = 0.
It determines a family of curves in yp plane. Let y = g(t), p = h(t) be one of the curves, i.e.
F (g(t), h(t)) = 0. Since
dy dy g 0 (t)dt
dx = 0 = = ,
y p h(t)
R t 0 (t)
we have x = t0 gh(t) dt + c. The solutions of (1.4.5.) are given by
t
g 0 (t)
Z
x= dt + c, y = g(t).
t0 h(t)
This method can also be applied to the equations
F (x, y 0 ) = 0,
where y is missing.
Example. Solve y 2 + y 02 − 1 = 0.
Ans: y = cos(c − x).
3. Reduction of order.
Consider the equation
F (x, y 0 , y 00 ) = 0, (1.4.6)
where y is missing. Let p = y 0 . Then y 00 = p0 . Write (1.4.6) as
F (x, p, p0 ) = 0. (1.4.7)
It is a first order equation in x and p. If p = φ(x, c1 ) is a general solution of (1.4.7), then the general
solution of (1.4.6) is Z x
y= φ(t, c1 )dt + c2 .
x0
Example. Solve xy 00 − y 0 = 3x2 .

Ans: y = x3 + c1 x2 + c2 .
Consider the equation

F (y, y 0 , y 00 ) = 0, (1.4.8)
1.4. FIRST ORDER IMPLICIT EQUATIONS 15
dp dp dy dp
where x is missing. Let p = y 0 . Then y 00 = dx = dy dx = dy p. Write (1.4.8) as
dp
F (y, p, p ) = 0. (1.4.9)
dy
It is a first order equation in y and p. If p = ψ(y, c1 ) is a general solution of (1.4.9), then we solve
the equation
y 0 = ψ(y, c1 )
to get a general solution of (1.4.8).
Example. Solve y 00 + k 2 y = 0, where k is a positive constant.

Ans: y = c1 sin(kx) + c2 cos(kx).
Chapter 2
Higher Order Linear Equations
2.1 General Theory

Consider n-th order linear equation
y (n) + a1 (x)y (n−1) + · · · + an−1 (x)y 0 + an (x)y = f (x), (2.1.1)

k
d y
where y (k) = dx k . Throughout this section we assume that aj (x)’s and f (x) are continuous func-
tions defined on the interval (a, b). When f (x) 6≡ 0, (2.1.1) is called a non-homogeneous equation.
The associated homogeneous equation is
y (n) + a1 (x)y (n−1) + · · · + an−1 (x)y 0 + an (x)y = 0. (2.1.2)
Let us begin with the initial value problem:


(n) (n−1)
 y + a1 (x)y

 + · · · + an (x)y = f (x),

y(x0 ) = y0 ,





y 0 (x0 ) = y1 , (2.1.3)


·········





 y (n−1) (x ) = y

0 n−1 .
Theorem 2.1 Assume that a1 (x), · · · , an (x) and f (x) are continuous functions defined on the
interval (a, b). Then for any x0 ∈ (a, b) and for any numbers y0 , · · · , yn−1 , the initial value
problem (2.1.3) has a unique solution defined on (a, b).
Especially if aj (x)’s and f (x) are continuous on R then for any x0 and y0 , · · · , yn−1 , the initial
value problem (2.1.3) has a unique solution defined on R.
Proof of this theorem will be given in later chapter.
Corollary 2.2 Let y = y(x) be a solution of the homogeneous equation (2.1.2) in an interval (a, b).
Assume that there exists x0 ∈ (a, b) such that
y(x0 ) = 0, y 0 (x0 ) = 0, · · · , y (n−1) (x0 ) = 0. (2.1.4)
Then y(x) ≡ 0 on (a, b).
17
18 CHAPTER 2. HIGHER ORDER LINEAR EQUATIONS
Proof. y is a solution of the initial value problem (2.1.2), (2.1.4). From Theorem 2.1, this problem
has a unique solution. Since φ(x) ≡ 0 is also a solution of the problem, we have y(x) ≡ 0 on (a, b).
In the following we consider the general solutions of (2.1.1) and (2.1.2).

Given continuous functions aj (x), j = 0, 1, · · · , n and f (x), define an operator L by
L[y] = a0 (x)y (n) + a1 (x)y (n−1) + · · · + an (x)y. (2.1.5)
Property 1. L[cy] = cL[y] for any constant c.

Property 2. L[u + v] = L[u] + L[v].
Proof. Let’s verify Property 2.
L[u + v]
= a0 (x)(u + v)(n) + a1 (x)(u + v)(n−1) + · · · + an (x)(u + v)

= a0 (x)u(n) + a1 (x)u(n−1) + · · · + an (x)u + a0 (x)v (n) + a1 (x)v (n−1) + · · · + an (x)v
= L[u] + L[v].
Definition. An operator satisfying Properties 1 and 2 is called a linear operator.
The differential operator L defined in (2.1.4) is a linear operator.

Note that (2.1.1) and (2.1.2) can be written as
L[y] = f (x), (2.1.1’)
and
L[y] = 0. (2.1.2’)
From Properties 1 and 2 we get the following conclusion.
Theorem 2.3 (1) If y1 and y2 are solutions of the homogeneous equation (2.1.2) in an interval
(a, b), then for any constants c1 and c2 ,
y = c1 y1 + c2 y2
is also a solution of (2.1.2) in the interval (a, b).

(2) If yp is a solution of (2.1.1) and yh is a solution of (2.1.2) on an interval (a, b), then
y = yh + yp
is also a solution of (2.1.1) in the interval (a, b).
Proof. (1) As L[c1 y1 + c2 y2 ] = c1 L[y1 ] + c2 L[y2 ] = 0, we see that c1 y1 + c2 y2 is also a solution

of (2.1.2).
(2) L[yh + yp ] = L[yh ] + L[yp ] = 0 + f (x), we see that y = yh + yp is a solution of (2.1.1).
2.1. GENERAL THEORY 19
In order to discuss structures of solutions, we need the following definition.
Definition. Functions φ1 (x), · · · , φk (x) are linearly dependent on (a, b) if there exists constants c1 ,
· · · , ck , not all zero, such that
c1 φ1 (x) + · · · + ck φk (x) = 0
for all x ∈ (a, b). A set of functions are linearly independent on (a, b) if they are not linearly
dependent on (a, b).
Lemma 2.4 Functions φ1 (x), · · · , φk (x) are linearly dependent on (a, b) if and only if the following
vector-valued functions
   
φ1 (x) φk (x)
 φ0 (x)   φ0 (x) 
 1
, ··· ,  k (2.1.6)
  
 ···   ··· 
 
(n−1) (n−1)
φ1 (x) φk (x)
are linearly dependent on (a, b).
Proof. “⇐=” is obvious. To show “=⇒”, assume that φ1 , · · · , φk are linearly dependent on (a, b).
There exists constants c1 , · · · , ck , not all zero, such that, for all x ∈ (a, b),
c1 φ1 (x) + · · · + ck φk (x) = 0.
Differentiating this equality successively we find that
c1 φ01 (x) + · · · + ck φ0k (x) = 0,

··················
(n−1) (n−1)
c1 φ1 (x) + · · · + ck φk (x) = 0.
Thus    
φ1 (x) φk (x)
 φ0 (x)   φ0 (x) 
c1  1  + · · · + ck  k =0
   
 ···   ··· 
(n−1) (n−1)
φ1 (x) φk (x)
for all x ∈ (a, b). Hence the k vector-valued functions are linearly dependent on (a, b).
n
Recall that, n vectors in R are linearly dependent if and only if the determinant of matrix formed
by these vectors is zero.
Definition. The Wronskian of n functions φ1 (x), · · · , φn (x) is defined by

φ1 (x)
··· φn (x)
W (φ1 , · · · , φn )(x) = · · · ··· ··· . (2.1.7)

(n−1) (n−1)
φ
1 (x) · · · φn (x)
Note that Wronskian of φ1 , · · · , φn is the determinant of the matrix formed by the vector-valued
functions given in (2.1.6).
Theorem 2.5 Let y1 (x), · · · , yn (x) be n solutions of (2.1.2) on (a, b) and let W (x) be their Wron-
skian.
(1) y1 (x), · · · , yn (x) are linearly dependent on (a, b) if and only if W (x) ≡ 0 on (a, b).
(2) y1 (x), · · · , yn (x) are linearly independent on (a, b) if and only if W (x) does not vanish on
(a, b).
Corollary 2.6 (1) The Wronskian of n solutions of (2.1.2) is either identically zero, or nowhere
zero.
(2) n solutions y1 , · · · , yn of (2.1.2) are linearly independent on (a, b) if and only if the set of vectors
   
y1 (x0 ) yn (x0 )
 y 0 (x )   y 0 (x ) 
 1 0   n 0 
, ··· , 
··· ···
 
   
(n−1) (n−1)
y1 (x0 ) yn (x0 )
are linearly independent for some x0 ∈ (a, b).
Proof of Theorem 2.5. Let y1 , · · · , yn be solutions of (2.1.2) on (a, b), and let W (x) be their
Wronskian.
Step 1. We first show that, if y1 , · · · , yn are linearly dependent on (a, b), then W (x) ≡ 0.
Since these solutions are linearly dependent, from Lemma 2.3, n vector-valued functions
   
y1 (x) yn (x)
 y 0 (x)   y 0 (x) 
 1  n
, ··· , 
 
 ···   ··· 
 
(n−1) (n−1)
y1 (x) yn (x)
are linearly dependent on (a, b). Thus for all x ∈ (a, b), the determinant of the matrix formed by
these vectors, namely, the Wronskian of y1 , · · · , yn , is zero.
Step 2. Now, assume that the Wronskian W (x) of n solutions y1 , · · · , yn vanishes at x0 ∈ (a, b).
We shall show that y1 , · · · , yn are linearly dependent on (a, b).
Since W (x0 ) = 0, the n vectors
   
y1 (x0 ) yn (x0 )
 y 0 (x )   y 0 (x ) 
 1 0   n 0 
, ··· , 
··· ···
 
   
(n−1) (n−1)
y1 (x0 ) yn (x0 )
are linearly dependent. Thus there exist n constants c1 , · · · , cn , not all zero, such that
   
y1 (x0 ) y1 (x0 )
 y 0 (x )   y 0 (x ) 
0  n 0 
c1  1  + · · · + cn  =0 (2.1.8)
 
 ···   ··· 
(n−1) (n−1)
y1 (x0 ) yn (x0 )
2.1. GENERAL THEORY 21
Define
y0 (x) = c1 y1 (x) + · · · + cn yn (x).
From Theorem 2.3, y0 is a solution of (2.1.2). From (2.1.8), y0 satisfies the initial conditions
y(x0 ) = 0, y 0 (x0 ) = 0, · · · , y (n−1) (x0 ) = 0. (2.1.9)
From Corollary 2.2, we have y0 ≡ 0, namely,
c1 y1 (x) + · · · + cn yn (x) = 0
for all x ∈ (a, b). Thus y1 , · · · , yn are linearly dependent on (a, b).
Example. Consider the differential equation y 00 − x1 y 0 = 0 on the interval (0,∞). Both φ1 (x) = 1
1 x2
and φ2 (x) = x2 are solutions of the differential equation. W (φ1 , φ2 )(x) = = 2x 6= 0

0 2x
for x > 0. Thus φ1 and φ2 are linearly independent solutions.
Theorem 2.7 Let a1 (x), · · · , an (x) and f (x) be continuous on the interval (a, b). The homoge-
neous equation (2.1.2) has n linearly independent solutions on (a, b).
Let y1 , · · · , yn be n linearly independent solutions of (2.1.2) defined on (a, b). The general solution
of (2.1.2) is given by
y(x) = c1 y1 (x) + · · · + cn yn (x), (2.1.10)
where c1 , · · · , cn are arbitrary constants.
Proof. (1) Fix x0 ∈ (a, b). For k = 1, 2, · · · , n, let yk be the solution of (2.1.2) satisfying the initial
conditions 
(j)
0 if j 6= k − 1,
yk (x0 ) =
1 if j = k − 1.
The n vectors
   
y1 (x0 ) yn (x0 )
 y 0 (x )   y 0 (x ) 
 1 0   n 0 
, ··· , 
··· ···
 
   
(n−1) (n−1)
y1 (x0 ) yn (x0 )
are lineally independent since they form the identity matrix. From Corollary 2.6, y1 , · · · , yn are
linearly independent on (a, b). From Theorem 2.3, for any constants c1 , · · · , cn , y = c1 y1 + · · · +
cn yn is a solution of (2.1.2).
(2) Now let y1 , · · · , yn be n linearly independent solutions of (2.1.2) on (a, b). We shall show that
the general solution of (2.1.2) is given by
y = c1 y1 + · · · + cn yn . (2.1.11)
Given a solution ỹ of (2.1.2), and fix x0 ∈ (a, b). Since y1 , · · · , yn are linearly independent on
(a, b), the vectors
   
y1 (x0 ) yn (x0 )
 y 0 (x )   y 0 (x ) 
 1 0   n 0 
, ··· , 
··· ···
 
   
(n−1) (n−1)
y1 (x0 ) yn (x0 )
are linearly independent vectors. They form a basis for Rn . Thus the vector
 
ỹ(x0 )
 ỹ 0 (x ) 
 0 
···
 
 
ỹ (n−1) (x0 )
can be represented as a linear combination of the n vectors, namely, there exist n constants c̃1 , · · · ,
c̃n such that
     
ỹ(x0 ) y1 (x0 ) yn (x0 )
 ỹ 0 (x )   y 0 (x )   y 0 (x ) 
0  1 0   n 0 
 = c̃1   + · · · + c̃n  .
 
··· ··· ···

     
(n−1) (n−1) (n−1)
ỹ (x0 ) y1 (x0 ) yn (x0 )
Let
φ(x) = ỹ(x) − [c̃1 y1 (x) + · · · + c̃n yn (x)].
φ(x) is a solution of (2.1.2) and satisfies the initial conditions (2.1.4) at x = x0 . By Corollary 2.2,
φ(x) ≡ 0 on (a, b). Thus
ỹ(x) = c̃1 y1 (x) + · · · + c̃n yn (x).
So (2.1.11) gives a general solution of (2.1.2).
Any set of n linearly independent solutions is called a fundamental set of solutions.
Now we consider the non-homogeneous equation (2.1.1). We have
Theorem 2.8 Let yp be a particular solution of (2.1.1), and y1 , · · · , yn be a fundamental set of

solutions for the associated homogeneous equation (2.1.2). The general solution of (2.1.1) is given
by
y(x) = c1 y1 (x) + · · · + cn yn (x) + yp (x). (2.1.12)
Proof. Let y be a solution of the non-homogeneous equation. Then y − yp is a solution of the

homogeneous equation. Thus y(x) − yp (x) = c1 y1 (x) + · · · + cn yn (x).
2.2 Linear Equations with Constant Coefficients

Let us begin with second order linear equation with constant coefficients
y 00 + ay 0 + by = 0, (2.2.1)
2.2. LINEAR EQUATIONS WITH CONSTANT COEFFICIENTS 23
where a and b are constants. We look for a solution of the form y = eλx . Plugging into (2.2.1) we
find that, eλx is a solution of (2.2.1) if and only if
λ2 + aλ + b = 0. (2.2.2)
(2.2.2) is called the auxiliary equation or characteristic equation of (2.2.1). The roots of (2.2.2) are
called characteristic values (or eigenvalues):
1 p
λ1 = (−a + a2 − 4b),
2
1 p
λ2 = (−a − a2 − 4b).
2
1. If a2 − 4b > 0, (2.2.2) has two distinct real roots λ1 , λ2 , and the general solutions of (2.2.1) is
y = c1 eλ1 x + c2 eλ2 x .
2. If a2 − 4b = 0, (2.2.2) has one real roots λ (we may say that (2.2.2) has two equal roots λ1 = λ2 ).
The general solution of (2.2.2) is
y = c1 eλx + c2 xeλx .
3. If a2 − 4b < 0, (2.2.2) has a pair of complex conjugate roots
λ1 = α + iβ, λ2 = α − iβ.
The general solution of (2.2.2) is
y = c1 eαx cos(βx) + c2 eαx sin(βx).
Example. Solve y 00 + y 0 − 2y = 0, y(0) = 4, y 0 (0) = −5.

Ans: λ1 = 1, λ2 = −2, y = ex + 3e−2x .
Example. Solve y 00 − 4y 0 + 4y = 0, y(0) = 3, y 0 (0) = 1.

Ans: λ1 = λ2 = 2, y = (3 − 5x)e2x .
Example. Solve y 00 − 2y 0 + 10y = 0.

Ans: λ1 = 1 + 3i, λ2 = 1 − 3i,
y = ex (c1 cos 3x + c2 sin 3x).
Now we consider n-th order homogeneous linear equations with constant coefficients
y (n) + a1 y (n−1) + · · · + an−1 y 0 + an y = 0, (2.2.3)
where a1 , · · · , an are real constants.

y = eλx is a solution of (2.2.3) if and only if λ satisfies
λn + a1 λn−1 + · · · + an−1 λ + an = 0. (2.2.4)
The solutions of (2.2.4) are called characteristic values or eigenvalues for the equation (2.2.3).
Let λ1 , · · · , λs be the distinct eigenvalues for (2.2.3). Then we can write
λn + a1 λn−1 + · · · + an−1 λ + an
(2.2.5)
= (λ − λ1 )m1 (λ − λ2 )m2 · · · (λ − λs )ms ,
where m1 , · · · , ms are positive integers and
m1 + · · · + ms = n.
We call them the multiplicity of the eigenvalues λ1 , · · · , λs respectively.
Lemma 2.9 Assume λ is an eigenvalue of (2.2.3) of multiplicity m.

(i) eλx is a solution of (2.2.3).
(ii) If m > 1, then for any positive integer 1 ≤ k ≤ m − 1, xk eλx is a solution of (2.2.3).
(iii) If λ = α + iβ, then
xk eαx cos(βx), xk eαx sin(βx)
are solutions of (2.2.3), where 0 ≤ k ≤ m − 1.
Theorem 2.10 Let λ1 , · · · , λs be the distinct eigenvalues for (2.2.3), with multiplicity m1 , · · · , ms
respectively. Then (2.2.3) has a fundamental set of solutions
eλ1 x , xeλ1 x , · · · , xm1 −1 eλ1 x ;

························ ; (2.2.6)
λs x λs x ms −1 λs x
e , xe , ··· , x e .
Proof of Lemma 2.9 and 2.10
Consider the n-th order linear equation with constant coefficients
y (n) + a1 y (n−1) + · · · + an−1 y 0 + an y = 0, (A1)

k
d y
where y (k) = dx k . Let L(y) = y
(n)
+ a1 y (n−1) + · · · + an−1 y 0 + an y and p(z) = z n + a1 z n−1 +
· · · + an−1 z + an . Note that p is a polynomial in z of degree n. Then we have
L(ezx ) = p(z)ezx (A2)

2 2
∂ ∂
Before we begin the proof, let’s observe that ∂z∂x ezx = ∂x∂z ezx by Clairaut’s theorem because
zx
e is differentiable in (x, z) as a function of two variable and all the higher order partial derivatives
exist and continuous. That means we can interchange the order of differentiation with respect to x
d d zx
and z as we wish. Therefore dz L(ezx ) = L( dz e ). For instance, one may verify directly that
d dk zx k zx k−1 zx dk d zx
(e ) = xz e + kz e = ( e ).
dz dxk dxk dz
Here one may need to use Leibniz’s rule of taking the k-th derivative of a product of two functions:
k
X k
(u · v)(k) = u(i) v (k−i) . (A3)
i=0
i
k k
d zx d zx
More generally, dz k L(e ) = L( dz ke ). (Strictly speaking, partial derivative notations should be
used.) Now let’s prove our results.
(1) If λ is a root of p, then L(eλx ) = 0 by (A2) so that eλx is a solution of (A1).

(2) If λ is a root of p of multiplicity m, then p(λ) = 0, p0 (λ) = 0, p00 (λ) = 0, . . . , p(m−1) (λ) = 0.
Now for k = 1, . . . , m − 1, differentiating (A2) k times with respect to z, we have
k
dk zx dk dk X k (i)
L(xk ezx ) = L( k
e ) = k
L(ezx
) = k
(p(z)ezx
) = p (z)xk−i ezx .
dz dz dz i=0
i
Thus L(xk eλx ) = 0 and xk eλx is solution of (A1).

(3) Let λ1 , · · · , λs be the distinct roots of p, with multiplicity m1 , · · · , ms respectively. Then we
wish to prove that
eλ1 x , xeλ1 x , · · · , xm1 −1 eλ1 x ;
························ ; (A4)
eλs x , xeλs x , · · · , xms −1 eλs x .
are linearly independent over R. To prove this, suppose for all x in R
c11 eλ1 x + c12 xeλ1 x + · · · + c1m1 xm1 −1 eλ1 x

+ ····································+
cs1 eλs x + cs2 xeλs x + · · · + csms xms −1 eλs x = 0.
Let’s write this as

P1 (x)eλ1 x + P2 (x)eλ2 x + · · · + Ps (x)eλs x = 0,
for all x in R, where Pi (x) = ci1 + ci2 x + · · · + cimi xmi −1 . We need to prove Pi (x) ≡ 0 for all
i. Assume that one of the Pi ’s is not identically zero. By re-labelling the Pi ’s, we may assume that
Ps (x) 6≡ 0. Dividing the above equation by eλ1 x , we have
P1 (x) + P2 (x)e(λ2 −λ1 )x + · · · + Ps (x)e(λs −λ1 )x = 0,
for all x in R. Upon differentiating this equation sufficiently many times (at most m1 times since
P1 (x) is a polynomial of degree m1 − 1), we can reduce P1 (x) to 0. Note that in this process, the
degree of the resulting polynomial multiplied by e(λi −λ1 )x remains unchanged. Therefore, we get
Q2 (x)e(λ2 −λ1 )x + · · · + Qs (x)e(λs −λ1 )x = 0,
where deg Qi = deg Pi . Cancelling the term eλ1 x we have
Q2 (x)eλ2 x + · · · + Qs (x)eλs x = 0.
Repeating this procedure, we arrive at
Rs (x)eλs x = 0,
where deg Rs = deg Ps . Hence Rs (x) ≡ 0 on R, contradicting the fact that deg Rs = deg Ps and
Ps is not identically zero. Thus all the Pi (x) are identically zero. That means all cij ’s are zero and
the functions in (A4) are linearly independent.
Remark. If (2.2.3) has a complex eigenvalue λ = α + iβ, then λ̄ = α − iβ is also an eigenvalue.

Thus both xk e(α+iβ)x and xk e(α−iβ)x appear in (2.2.6), where 0 ≤ k ≤ m − 1. In order to obtain a
fundamental set of real solutions, the pair of solutions xk e(α+iβ)x and xk e(α−iβ)x in (2.2.6) should
be replaced by xk eαx cos(βx) and xk eαx sin(βx).
In the following we discuss solution of non-homogeneous equations. For simplicity we consider
y 00 + by 0 + cy = f (x), (2.2.7)
where b and c are real constants. The associated homogeneous equation is (2.2.1), and the charac-
teristic equation of (2.2.1) is (2.2.2). We shall look for a particular solution of (2.2.7). This method
works in general even if a and b are functions of x. Also the method applies to higher order equa-
tions.
1. Methods of variation of parameters.

Let y1 and y2 be two linearly independent solutions of the associated homogeneous equation (2.2.1)
and W (x) be their Wronskian. Look for a particular solution of (2.2.7) in the form
yp = u 1 y 1 + u 2 y 2 ,
where u1 and u2 are functions to be determined. Suppose
u01 y1 + u02 y2 = 0.
Plugging yp into (2.2.7) we get

u01 y10 + u02 y20 = f.
Hence u01 and u02 satisfy

u01 y1 + u02 y2 = 0,
(
(2.2.8)
u01 y10 + u02 y20 = f.
Solving it, we find that
y2 y1
u01 = − f, u02 = f.
W W
Integrating yields Z x
y2 (t)
u1 (x) = − f (t)dt,
x0 W (t)
Z x (2.2.9)
y1 (t)
u2 (x) = f (t)dt.
x0 W (t)
Example. Solve the differential equation. y 00 + y = sec x.

Solution. A basis for the solutions of the homogeneous equation consists of y1 = cos x and y2 =
R
sin x. Now W (y1 , y2 ) = cos x cos x − (− sin x) sin x = 1. Thus u1 = − sin x sec x dx =
R
ln | cos x| + c1 and u2 = cos x sec x dx = x + c2 . From this, a particular solution is given
by yp = cos x ln | cos x| + x sin x. Therefore, the general solution is y = c1 cos x + c2 sin x +
cos x ln | cos x| + x sin x.
The method of variation of parameters can also be used to find another solution of a second order
homogeneous linear differential equation when one solution is given. Suppose z is a known solution
of the equation
y 00 + P (x)y 0 + Q(x)y = 0.
We assume y = vz is a solution so that
0 = (vz)00 + P (vz)0 + Q(vz) = (v 00 z + 2v 0 z 0 + vz 00 ) + P (v 0 z + vz 0 ) + Qvz

= (v 00 z + 2v 0 z 0 + P v 0 z) + v(z 00 + P z 0 + Qz) = v 00 z + v 0 (2z 0 + P z).
That is
v 00 z0
= −2 − P.
v0 z
R R
An integration gives v 0 = z −2 e− P dx and v = z −2 e− P dx dx. We leave it as an exercise to
R
show that z and vz are linearly independent solutions by computing their Wronskian.
Example. Given y1 = x is a solution of x2 y 00 + xy 0 − y = 0, find another solution.

Solution. Let’s write the DE in the form y 00 + x1 y 0 − x12 y = 0. Then P (x) = 1/x. Thus a second
linearly independent solution is given y = vx, where
Z Z
R 1
v = x−2 e− 1/x dx dx = x−2 x−1 dx = − 2 .
x
Therefore the second solution is y = −x−1 and the general solution is y = c1 x + c2 x−1 .
2. Method of undetermined coefficients.

Case 1. f (x) = Pn (x)eαx , where Pn (x) is a polynomial of degree n ≥ 0.
We look for a particular solution in the form
y = Q(x)eαx ,
where Q(x) is a polynomial. Plugging it into (2.2.7) we find
Q00 + (2α + b)Q0 + (α2 + bα + c)Q = Pn (x). (2.2.10)
Subcase 1.1. If α2 + bα + c 6= 0, namely, α is not a root of (2.2.2), we choose Q = Rn , a
polynomial of degree n, and

y = Rn (x)eαx .
The coefficients of Rn can be determined by comparing the terms of same power in the two sides of
(2.2.10). Note that in this case both sides of (2.2.10) are polynomials of degree n.
Subcase 1.2. If α2 + bα + c = 0 but 2α + b 6= 0, namely, α is a simple root of (2.2.2), then (2.2.10)

is reduced to
Q00 + (2α + b)Q0 = Pn . (2.2.11)
We choose Q to be a polynomial of degree n + 1. Since the constant term of Q does not appear in
(2.2.11), we may choose Q(x) = xRn (x), where Rn (x) is a polynomial of degree n.
y = xRn (x)eαx .
Subcase 1.3 If α2 + bα + c = 0 and 2α + b = 0, namely, α is a root of (2.2.2) with multiplicity 2,

then (2.2.10) is reduced to
Q00 = Pn . (2.2.12)
We choose Q(x) = x2 Rn (x), where Rn (x) is a polynomial of degree n.
y = x2 Rn (x)eαx .
Example. Find the general solution of y 00 − y 0 − 2y = 4x2 .

Ans: y = c1 e2x + c2 e−x − 3 + 2x − 2x2 .
Example. Find a particular solution of y 000 + 2y 00 − y 0 = 3x2 − 2x + 1.

Ans: y = −27x − 5x2 − x3 .
Example. Solve y 00 − 2y 0 + y = xex .

Ans: y = c1 ex + c2 xex + 16 x3 ex .
Case 2. f (x) = Pn (x)eαx cos(βx) or f (x) = Pn (x)eαx sin(βx), where Pn (x) is a polynomial of
degree n ≥ 0.
We first look for a solution of
y 00 + by 0 + cy = Pn (x)e(α+iβ)x . (2.2.13)
Using the method in Case 1 we obtain a complex-valued solution
z(x) = u(x) + iv(x),
where u(x) = <(z(x)), v(x) = =(z(x)). Substituting z(x) = u(x)+iv(x) into (2.2.13) and taking
the real and imaginary parts, we can show that u(x) = <(z(x)) is a solution of
y 00 + by 0 + cy = Pn (x)eαx cos(βx), (2.2.14)
and v(x) = =(z(x)) is a solution of
y 00 + by 0 + cy = Pn (x)eαx sin(βx). (2.2.15)
Example. Solve y 00 − 2y 0 + 2y = ex cos x.

Ans: y = c1 ex cos x + c2 ex sin x + 12 xex sin x.
The following conclusions will be useful.

2.3. OPERATOR METHODS 29
Theorem 2.11 Let y1 and y2 be particular solutions of the equations
y 00 + ay 0 + by = f1 (x)
and
y 00 + ay 0 + by = f2 (x)
respectively, then yp = y1 + y2 is a particular solution of
y 00 + ay 0 + by = f1 (x) + f2 (x).
Proof. Exercise.
Example. Solve y 00 − y = ex + sin x.

Solution. A particular solution for y 00 − y = ex is given by y1 = 12 xex . Also a particular solution
for y 00 − y = sin x is given by y2 = − 12 x sin x. Thus 12 (xex − sin x) is a particular solution of
the given differential equation. The general solution of the corresponding homogeneous differential
equation is given by c1 e−x + c2 ex . Hence the general solution of the given differential equation is
c1 e−x + c2 ex + 21 (xex − sin x).
2.3 Operator methods

Let x denote independent variable, and y dependent variable. Introduce
d dn
Dy = y, Dn y = y = y (n) .
dx dxn
Pn
We define D0 y = y. Given a polynomial L(x) = j=0 aj xj , where aj ’s are constants, we define
a differential operator L(D) by
Xn
L(D)y = aj Dj y.
j=0
Then the equation

n
X
aj y (j) = f (x) (2.3.1)
j=0
can be written as
L(D)y = f (x). (2.3.2)
Let L(D)−1 f denote any solution of (2.3.2). We have
D−1 D = DD−1 = D0 ,
L(D)−1 L(D) = L(D)L(D)−1 = D0 .
However, L(D)−1 f is not unique.
To see the above properties, first recall that D−1 f means a solution of y 0 = f . Thus D−1 f =
R
f.
Hence it follows that D−1 D = DD−1 = identity operator D0 .
For the second equality, note that a solution of L(D)y = L(D)f is simply f . Thus by definition
of L(D)−1 , we have L(D)−1 (L(D)f ) = f . This means L(D)−1 L(D) = D0 . Lastly, since
L(D)−1 f is a solution of L(D)y = f (x), it is clear that L(D)(L(D)−1 f ) = f . In other words,
L(D)L(D)−1 = D0 .
More generally, we have:

Z
1. D−1 f (x) = f (x)dx + C,
Z
−1
2. (D − a) f (x) = Ceax
+eax
e−ax f (x)dx, (2.3.3)
ax ax
3. L(D)(e f (x)) = e L(D + a)f (x),
4. L(D)−1 (eax f (x)) = eax L(D + a)−1 f (x).
Proof. Property 2 is just the solution of the first order linear ODE. To prove Property 3, first observe
that (D − r)(eax f (x)) = eax D(f (x)) + aeax f (x) − reax f (x) = eax (D + a − r)(f (x)). Thus
(D − s)(D − r)(eax f (x)) = (D − s)[eax (D + a − r)(f (x))] = eax (D + a − s)(D + a − r)(f (x)).
Now we may write L(D) = (D − r1 ) · · · (D − rn ). Then L(D)(eax f (x)) = eax L(D + a)f (x).
This says that we can move the factor eax to the left of the operator L(D) if we replace L(D) by
L(D + a).
To prove Property 4, apply L(D) to the right hand side. We have
L(D)[eax L(D + a)−1 f (x)] = eax [L(D + a)(L(D + a)−1 f (x))] = eax f (x).
Thus L(D)−1 (eax f (x)) = eax L(D + a)−1 f (x).
Let L(x) = (x − r1 ) · · · (x − rn ). The solution of (2.3.2) is given by
y = L(D)−1 f (x) = (D − r1 )−1 · · · (D − rn )−1 f (x). (2.3.4)
Then we obtain the solution by successive integration. Moreover, if rj0 s are distinct, we can write
1 A1 An
= + ··· + ,
L(x) x − r1 x − rn
where A0j s can be found by the method of partial fractions. Then the solution is given by
y = [A1 (D − r1 )−1 + · · · + An (D − rn )−1 ]f (x). (2.3.5)
Next consider the case of repeated roots. Let the multiple root be equal to m and the equation to be
solved is
(D − m)n y = f (x) (2.3.6)
To solve this equation, let us assume a solution of the form y = emx v(x), where v(x) is a function
of x to be determined. One can easily verify that (D − m)n emx v = emx Dn v. Thus equation (2.3.6)
reduces to
Dn v = e−mx f (x) (2.3.7)
2.3. OPERATOR METHODS 31
If we integrate (2.3.7) n times, we obtain

Z Z Z Z
v= ··· e−mx f (x) dx · · · dx + c0 + c1 x + · · · + cn−1 xn−1 (2.3.8)
Thus we see that

ZZ ZZ
−n mx −mx n−1
(D−m) f (x) = e ··· e f (x)dx · · · dx + c0 + c1 x + · · · + cn−1 x (2.3.9)
Example Solve (D2 − 3D + 2)y = xex .

1 1 1
Solution First D 2 −3D+2 = D−2 − D−1 . Therefore
y = (D2 − 3D + 2)−1 (xex )

= (D − 2)−1 (xex ) − (D − 1)−1 (xex )
= e2x D−1 (e−2x xex ) − ex D−1 (e−x xex )
= e2x D−1 (e−x x) − ex D−1 (x)
= e2x (−xe−x − e−x + c1 ) − ex ( 12 x2 + c2 )
= −ex ( 12 x2 + x + 1) + c1 e2x + c2 ex .
Example. Solve (D3 − 3D2 + 3D − 1)y = ex .

Solution. The DE is equivalent to (D − 1)3 y = ex . Therefore,
Z Z Z
−3 x x −x x 2 x 1 3 2
y = (D − 1) e =e e e dx + c0 + c1 x + c2 x =e x + c0 + c1 x + c2 x .
6
If f (x) is a polynomial in x, then (1 − D)(1 + D + D2 + D3 + · · · )f = f . Thus (1 − D)−1 (f ) =

(1+D +D2 +D3 +· · · )f . Therefore, if f is a polynomial, we may formally expand (D −r)−1 into
power series in D and applying it to f . If the degree of f is n, then it is only necessary to expand
(D − r)−1 up to Dn .
Example. Solve (D4 − 2D3 + D2 )y = x3 .

Solution. We have
y = (D4 − 2D3 + D2 )−1 f = 1

D 2 (1−D)2 x
3
−2
= D (1 + 2D + 3D + 4D + 5D + 6D5 )x3
2 3 4
= D−2 (x3 + 6x2 + 18x + 24)

4
= D−1 ( x4 + 2x3 + 9x2 + 24x)
5 4
= x20 + x2 + 3x3 + 12x2 .
x5 x4
Therefore, the general solution is y = (A + Bx)ex + (C + Dx) + 20 + 2 + 3x3 + 12x2 .
2.4 Exact 2nd order Equations

The general 2nd order linear differential equation is of the form
p0 (x)y 00 + p1 (x)y 0 + p2 (x)y = f (x) (A1)
The equation can be written as
(p0 y 0 − p00 y)0 + (p1 y)0 + (p000 − p01 + p2 )y = f (x) (A2)
It is said to be exact if
p000 − p01 + p2 ≡ 0. (A3)
In the event that the equation is exact, a first integral to (A1) is

Z
p0 (x)y 0 − p00 (x)y + p1 (x)y = f (x) dx + C1 .
Example. Find the general solution of the DE

1 00 1 2 1 2
y + ( − 2 )y 0 − ( 2 − 3 )y = ex .
x x x x x
Solution. Condition (A3) is fulfilled. The first integral is
1 0 1 1 2
y + 2 y + ( − 2 )y = ex + C1 .
x x x x
That is
1
y 0 + (1 −)y = xex + C1 x.
x
From the last equation, the general solution is found to be
1 x
y= xe + C1 x + C2 xe−x .
2
2.5 The adjoint differential equation and integrating factor

If (A1) is multiplied by a function v(x) so that the resulting equation is exact, then v(x) is called an
integrating factor of (A1). That is
(p0 v)00 − (p1 v)0 + p2 v = 0. (A4)
This is a differential equation for v, which is, more explicitly,
p0 (x)v 00 + (2p00 (x) − p1 (x))v 0 + (p000 (x) − p01 (x) + p2 (x))v = 0. (A5)
Equation (A5) is called the adjoint of the given differential equation (A1). A function v(x) is thus
an integrating factor for a given differential equation, if and only if it is a solution of the adjoint
2.5. THE ADJOINT DIFFERENTIAL EQUATION AND INTEGRATING FACTOR 33
equation. Note that the adjoint of (A5) is in turn found to be the associated homogeneous equation
of (A1), thus each is the adjoint of the other.
In this case, a first integral to (A1) is
Z
v(x)p0 (x)y 0 − (v(x)p0 (x))0 y + v(x)p1 (x)y = v(x)f (x) dx + C1 .
Example. Find the general solution of the DE
(x2 − x)y 00 + (2x2 + 4x − 3)y 0 + 8xy = 1.
Solution. The adjoint of this equation is
(x2 − x)v 00 − (2x2 − 1)v + (4x − 2)v = 0.
By the trial of xm , this equation is found to have x2 as a solution. Thus x2 is an integrating factor
of the given differential equation. Multiplying the original equation by x2 , we obtain
(x4 − x3 )y 00 + (2x4 + 4x3 − 3x2 )y 0 + 8x3 y = x2 .
Thus a first integral to it is

Z
(x4 − x3 )y 0 − (4x3 − 3x2 )y + (2x4 + 4x3 − 3x2 )y = x2 dx + C.
After simplification, we have

2x 1 C
y0 + y= + .
x−1 3(x − 1) x3 (x − 1)
An integrating factor for this first order linear equation is e2x (x − 1)2 . Thus the above equation
becomes Z 2x
e (x − 1)
Z
1
e2x (x − 1)2 y = (x − 1)e2x dx + C dx + C2 .
3 x3
That is
e2x

1 x 3 2x
e2x (x − 1)2 y = − e + C 2 + C2 .
3 2 4 2x
Thus the general solution is

1 x 1 C1
y= − + 2 + C2 e−2x .
(x − 1)2 6 4 x
Exercise. Solve the following differential equation by finding an integrating factor of it.
4x 8x − 8
y 00 + y0 + y = 0.
2x − 1 (2x − 1)2
C1 C2 x −2x
[Answer: y = 2x−1 + 2x−1 e .]
Solution. The adjoint equation is

4x 4
v 00 − v0 + v = 0,
2x − 1 2x − 1
or equivalently
(2x − 1)v 00 − 4xv 0 + 4v = 0.
An obvious solution is v = x. Therefore v = x is an integrating factor of the original differential

equation. Thus
4x2 0 (8x − 8)x
xy 00 + y + y=0
2x − 1 (2x − 1)2
is exact. The first integral is
4x2
xy 0 − y + y = C1 ,
2x − 1
or equivalently,
4x2 − 2x + 1 C1
y0 + y= .
x(2x − 1) x
That is
1 2 C1
y0 + 2 − + y= .
x 2x − 1 x
1 2
R
Thus e 2− x + 2x−1 dx = e2x 2x−1

x is an integrating factor of this first order equation. Multiplying
by this factor, we have
e2x (2x − 1)

2x − 1
Z
e2x y = C1 dx + C2 .
x x2
That is
e2x

2x − 1
e2x y = C1 + C2 ,
x x
or equivalently
C1 C2 x −2x
y= + e .
2x − 1 2x − 1
Chapter 3
Linear Differential Systems
3.1 Linear Systems

The following system is called a linear differential system of first order in normal form:

x0 = a11 (t)x1 + · · · + a1n (t)xn + g1 (t),
 1


······························

 0

xn = an1 (t)x1 + · · · + ann (t)xn + gn (t),
where aij (t) and gj (t) are continuous functions of t and 0 denotes differentiation with respect to t.
Denote
     
x1 (t) g1 (t) a11 (t) · · · a1n (t)
x(t) =  · · ·  , g(t) =  · · ·  , A(t) =  · · · ··· ··· .
     
xn (t) gn (t) an1 (t) · · · ann (t)
We call g(t) a continuous vector field, or a continuous vector-valued function, if all its components
are continuous functions. We call A(t) a continuous matrix, or a continuous matrix-valued function,
if all its entries are continuous functions. Define
   R 
x01 (t) Z x1 (t)dt
x0 =  · · ·  , x(t)dt =  ··· .
   
x0n (t)
R
xn (t)dt
Then the linear system can be written in the matrix form:
x0 = A(t)x + g(t). (3.1.1)
When g ≡ 0, (3.1.1) is reduced to

x0 = A(t)x. (3.1.2)
(3.1.2) is called a homogeneous differential system, and (3.1.1) is called a non-homogeneous system
if g(t) 6≡ 0. We shall also call (3.1.2) the homogeneous system associated with (3.1.1), or the
associated homogeneous system.
35
36 CHAPTER 3. LINEAR DIFFERENTIAL SYSTEMS
Example.
x01 = 2x1 + 3x2 + 3t,
x02 = −x1 + x2 − 7 sin t
is equivalent to
!0 ! ! !
x1 2 3 x1 3t
= + .
x2 −1 1 x2 −7 sin t
Example. Given a second order system

 2
d x
 2 = x + 2y + 3t,


dt
2
 y = 4x + 5y + 6t,

 d
dt2
it can be expressed into an equivalent first order differential system by introducing more variables.
For this example, let u = x0 and v = y 0 . Then we have
x0 = u
u0 = x + 2y + 3t
.
y0 = v
v0 = 4x + 5y + 6t
Next, we begin with the initial value problem
x0 = A(t)x + g(t),
(
(3.1.3)
x(t0 ) = x0 ,
where x0 is a constant vector. Similar to Theorem 2.1 we can show the following theorem.
Theorem 3.1 Assume that A(t) and g(t) are continuous on an open interval a < t < b containing
t0 . Then, for any constant vector x0 , (3.1.3) has a solution x(t) defined on this interval. This
solution is unique.
Especially, if A(t) and g(t) are continuous on R, then for any t0 ∈ R and x0 ∈ Rn , (3.1.3) has a
unique solution x(t) defined on R.
3.2 Homogeneous Linear Systems

In this section we assume A = aij (t) is a continuous n by n matrix-valued function defined on
the interval (a, b). We shall discuss the structure of the set of all solutions of (3.1.2).
Lemma 3.2 Let x(t) and y(t) be two solutions of (3.1.2) on (a, b). Then for any numbers c1 , c2 ,
z(t) = c1 x(t) + c2 y(t) is also a solution of (3.1.2) on (a, b).
3.2. HOMOGENEOUS LINEAR SYSTEMS 37
Definition x1 (t), · · · , xr (t) are linearly dependent in (a, b), if there exists numbers c1 , · · · , cr , not
all zero, such that
c1 x1 (t) + · · · + cr xr (t) = 0 for all t ∈ (a, b).
x1 (t), · · · , xr (t) are linearly independent on (a, b) if they are not linearly dependent.
Lemma 3.3 A set of solutions x1 (t), · · · , xr (t) of (3.1.2) are linearly dependent on (a, b) if and
only if x1 (t0 ), · · · , xr (t0 ) are linearly dependent vectors for any fixed t0 ∈ (a, b).
Proof. Obviously “=⇒” is true. We show “⇐=”. Suppose that, for some t0 ∈ (a, b), x1 (t0 ), · · · ,
xr (t0 ) are linearly dependent. Then there exists constants c1 , · · · , cr , not all zero, such that
c1 x1 (t0 ) + · · · + cr xr (t0 ) = 0.
Let y(t) = c1 x1 (t) + · · · + cr xr (t). Then y(t) is the solution of the initial value problem
x0 = A(t)x, x(t0 ) = 0.
Since x(t) = 0 is also a solution of the initial value problem, by the uniqueness we have y(t) ≡ 0
on (a, b), i.e.
c1 x1 (t) + · · · + cr xr (t) ≡ 0
on (a, b). Since cj ’s are not all zero, x1 (t), · · · , xr (t) are linearly dependent on (a, b).
Theorem 3.4 (i) (3.1.2) has n linearly independent solutions.

(ii) Let x1 , · · · , xn be any set of n linearly independent solutions of (3.1.2) on (a, b). Then the
general solution of (3.1.2) is given by
x(t) = c1 x1 (t) + · · · + cn xn (t), (3.2.1)
where cj ’s are arbitrary constants.
Proof. (i) Let e1 , · · · , en be a set of linearly independent vectors in Rn . Fix t0 ∈ (a, b). For each
j from 1 to n, consider the initial value problem
x0 = A(t)x, x(t0 ) = ej .
From Theorem 3.1, there exists a unique solution xj (t) defined on (a, b). From Lemma 3.3,
x1 (t), · · · , xn (t) are linearly independent on (a, b).
(ii) Now let x1 (t), · · · , xn (t) be any set of n linearly independent solutions of (3.1.2) on (a, b).
Fix t0 ∈ (a, b). From Lemma 3.3, x1 (t0 ), · · · , xn (t0 ) are linearly independent vectors. Let x̃(t)
be any solution of (3.2.1). Then x̃(t0 ) can be represented by a linear combination of x1 (t0 ), · · · ,
xn (t0 ), namely, there exists n constants c̃1 , · · · , c̃n such that
x̃(t0 ) = c̃1 x1 (t0 ) + · · · + c̃n xn (t0 ).
As in the proof of Lemma 3.3, we can show that
x̃(t) = c̃1 x1 (t) + · · · + c̃n xn (t).

Thus c1 x1 (t) + · · · + cn xn (t) is the general solution of (3.1.2).
Recall that, n vectors

   
a11 a1n
a1 =  · · ·  , ······ , an =  · · · 
   
an1 ann
are linearly dependent if and only if the determinant

a11 · · · a1n

··· ··· ··· = 0.

an1 · · · ann
In order to check whether n solutions are linearly independent, we need the following notation.
Definition. The Wronskian of n vector-valued functions

   
x11 (t) x1n (t)
x1 (t) =  · · ·  , · · · , xn (t) =  · · · 
   
xn1 (t) xnn (t)

is the determinant

x11 (t) x12 (t)
··· x1n (t)

x (t) x (t)
12 22 ··· x2n (t)
W (t) ≡ W (x1 , · · · , xn )(t) = .

··· ··· ··· ···

xn1 (t) xn2 (t) ··· xnn (t)
Using Lemma 3.3 we can show that
Theorem 3.5 (i) The Wronskian of n solutions of (3.1.2) is either identically zero or nowhere zero
in (a, b).
(ii) n solutions of (3.1.2) are linearly dependent in (a, b) if and only if their Wronskian is identically
zero in (a, b).
Definition. A set of n linearly independent solutions of (3.1.2) is called a fundamental set of solu-
tions, or a basis of solutions. Let
   
x11 (t) x1n (t)
x1 (t) =  · · ·  , · · · , xn (t) =  · · · 
   
xn1 (t) xnn (t)

be a fundamental set of solutions of (3.1.2) on (a, b). The matrix-valued function
 
x11 (t) x12 (t) · · · x1n (t)
 x (t) x (t) · · · x (t) 
 12 22 2n
Φ(t) = 

 ··· ··· ··· ··· 

xn1 (t) xn2 (t) · · · xnn (t)

3.3. NON-HOMOGENEOUS LINEAR SYSTEMS 39
is called a fundamental matrix of (3.1.2) on (a, b).
Remark. (i) From Theorem 3.5, a fundamental matrix is non-singular for all t ∈ (a, b).
(ii) A fundamental matrix Φ(t) satisfies the following matrix equation:
Φ0 = A(t)Φ. (3.2.2)
(iii) Let Φ(t) and Ψ(t) are two fundamental matrices defined on (a, b). Then there exists a constant,
non-singular matrix C such that
Ψ(t) = Φ(t)C.
Theorem 3.6 Let Φ(t) be a fundamental matrix of (3.1.2) on (a, b). Then the general solution of
(3.1.2) is given by
x(t) = Φ(t)c, (3.2.3)
 
c1
where c =  · · ·  is an arbitrary constant vector.
 
cn
3.3 Non-Homogeneous Linear Systems

In this section we consider the solutions of non-homogeneous system (3.1.1), where A = aij (t)
is a continuous n by n matrix-valued function and g(t) is a continuous vector-valued function, both
defined on the interval (a, b).
Theorem 3.7 Let xp (t) be a particular solution of (3.1.1), and Φ(t) be a fundamental matrix of the
associated homogeneous system (3.1.2). Then the general solution of (3.1.1) is given by
x(t) = Φ(t)c + xp (t), (3.3.1)
where c is an arbitrary constant vector.
Proof. For any constant vector c, x(t) = Φ(t)c + xp (t) is a solution of (3.1.1). On the other hand,
let x(t) be a solution of (3.1.1) and set y(t) = x(t) − xp (t). Then y0 = A(t)y. From (3.2.3), there
exists a constant vector c̃ such that y(t) = Φ(t)c̃. So x(t) = Φ(t)c̃ + xp (t). Thus (3.3.1) gives a
general solution of (3.1.1).
Method of variation of parameters.

Let Φ be a fundamental matrix of (3.1.2). We look for a particular solution of (3.1.1) in the form
 
u1 (t)
x(t) = Φ(t)u(t), u(t) =  · · ·  .
 
un (t)
Plugging into (3.1.1) we get

Φ0 u + Φu0 = AΦu + g.
From (3.2.2), Φ0 = AΦ. So Φu0 = g, and thus u0 = Φ−1 g.

Z t
u(t) = Φ−1 (s)g(s)ds + c. (3.3.2)
t0
Choosing c = 0, we get a particular solution:

Z t
xp (t) = Φ(t) Φ−1 (s)g(s)ds.
t0
So we obtain the following:
Theorem 3.8 The general solution of (3.1.1) is given by

Z t
x(t) = Φ(t)c + Φ(t) Φ−1 (s)g(s)ds, (3.3.3)
t0
where Φ(t) is a fundamental matrix of the associated homogeneous system (3.1.2).
(
x01 = 3x1 − x2 + t
Example. Solve .
x02 = 2x1 + t
3.4 Homogeneous Linear Systems with Constant Coefficients

Consider a homogeneous linear system
x0 = Ax, (3.4.1)
where A = (aij ) is a constant n by n matrix.

Let us try to find a solution of (3.4.1) in the form x(t) = eλt k, where k is a constant vector, k 6= 0.
Plugging it into (3.4.1) we find
Ak = λk. (3.4.2)
Definition. Assume that a number λ and a vector k 6= 0 satisfy (3.4.2), then we call λ an eigenvalue
of A, and k an eigenvector associated with λ.
Lemma 3.9 λ is an eigenvalue of A if and only if
det (A − λI) = 0 (3.4.3)
(where I is the n × n unit matrix), namely,

a11 − λ a12 ··· a1n

a
21 a22 − λ · · · a2n
= 0.

··· ··· ··· ···

an1 an2 ··· ann − λ
3.4. HOMOGENEOUS LINEAR SYSTEMS WITH CONSTANT COEFFICIENTS 41
Remark. Let A be an n by n matrix and λ1 , λ2 , · · · , λk be the distinct roots of (3.4.3). Then there
exist positive integers m1 , m2 , · · · , mk , such that
det(A − λI) = (−1)n (λ − λ1 )m1 (λ − λ2 )m2 · · · (λ − λk )mk ,
and
m1 + m2 + · · · + mk = n.
mj is called the algebraic multiplicity (or simply multiplicity) of the eigenvalue λj . The number of
linearly independent eigenvectors of A associated with λj is called the geometric multiplicity of the
eigenvalue λj and is denoted by µ(λj ). We always have
µ(λj ) ≤ mj .
If µ(λj ) = mj then we say that the eigenvalue λj is quasi-simple. Especially if mj = 1 we say that
λj is a simple eigenvalue. Note that in this case λj is a simple root of (3.4.3).
Theorem 3.10 If λ is an eigenvalue of A and k is an associated eigenvector, then
x(t) = eλt k
is a solution of (3.4.1).
Let A be a real matrix. If λ is a complex eigenvalue of A, and k is an eigenvector associated with
λ, then
x1 = <(eλt k), x2 = =(eλt k)
are two linearly independent real solutions of (3.4.1).
In the following we always assume that A is a real matrix.
Theorem 3.11 If A has n linearly independent eigenvectors k1 , · · · , kn associated with eigenval-

ues λ1 , · · · , λn respectively, then
Φ(t) = (eλ1 t k1 , · · · , eλn t kn )
is a fundamental matrix of (3.4.1), and the general solution is given by
x(t) = Φ(t)c = c1 eλ1 t k1 + · · · + cn eλn t kn , (3.4.4)

 
c1
where c =  · · ·  is an arbitrary constant vector.
 
cn
Proof. We only need to show det Φ(t) 6= 0. Since k1 , · · · , kn are linearly independent, so
det Φ(0) 6= 0. From Theorem 3.5 we see that det Φ(t) 6= 0 for any t. Hence Φ(t) is a funda-
mental matrix.
Remark. Under the conditions of Theorem 3.11, the eigenvalues λ1 , · · · , λn of A need not to be
distinct. In fact we only assume that all the eigenvalues of A are quasi-simple.
If A has n distinct eigenvalues λ1 , · · · , λn , and let k1 , · · · , kn be the associated eigenvectors, then
they are linearly independent. Hence the general solution is given by (3.4.4).
!
0 −3 1
Example. x = x.
1 −3
!
−3 1
A= has eigenvalues λ1 = −2, and λ2 = −4.
1 −3
!
1
For λ1 = −2 we find an eigenvector k1 = .
1
!
1
For λ2 = −4 we find an eigenvector k2 = .
−1
! !
−2t 1 −4t 1
x(t) = c1 e + c2 e .
1 −1
Example. Solve the system

! !
0 −3 1 −2t −6
x = x+e .
1 −3 2
We first solve the associated homogeneous system

!
0 −3 1
x = x
1 −3
! !
e−2t e−4t
and find two linearly independent solutions x1 (t) = , x2 (t) = . The fun-
e−2t −e−4t
damental matrix is !
e−2t e−4t
Φ = (x1 (t), x2 (t)) = .
e−2t −e−4t
! !
−1 1 −e−4t −e−4t 1 e2t e2t
Φ = = ,
−2e−6t −e−2t e−2t 2 e4t −e4t
!
−6
Let g(t) = e−2t . Then
2
!
−1 −2
Φ (t)g(t) = ,
−4e2t
!
t
−2t
Z
−1
u(t) = Φ (s)g(s)ds = ,
0 −2e2t + 2
! !
−2t −t − 1 −4t 1
Φ(t)u(t) = 2e + 2e .
−t + 1 −1
The general solution is
! ! ! !
1 1 1 −1
x = c1 e−2t + c2 e−4t − 2te−2t + 2e−2t .
1 −1 1 1
!
0 0 1
Example. Solve x = x.
−4 0
!
0 1
A= has eigenvalues ±2i.
−4 0
!
1
For λ = 2i we find an eigenvector k = .
2i
! ! ! !
2it 1 1 cos 2t sin 2t
e =(cos 2t + i sin 2t) = +i .
2i 2i −2 sin 2t 2 cos 2t

! !
cos 2t sin 2t
x(t) = c1 + c2 .
−2 sin 2t 2 cos 2t
Example. Solve 

 x0 = −3x + 4y − 2z,

y 0 = x + z,

 0

z = 6x − 6y + 5z.
 
−3 4 −2
A= 1 0 1  has eigenvalues λ1 = 2, λ2 = 1, λ3 = −1.
 
6 −6 5
 
0
For λ1 = 2 we find an eigenvector k1 =  1  .
 
2
 
1
 
0
 
1
For λ3 = −1 we find an eigenvector k3 =  0  .
 
−1
       
x 0 1 1
2t  t −t 
 y  = c1 e  1  + c2 e  1  + c3 e  0  ,
    
z 2 0 −1
namely
x(t) = c2 et + c3 e−t ,
y(t) = c1 e2t + c2 et ,
z(t) = 2c1 e2t − c3 e−t .
Example. Solve x0 = Ax, where

 
2 1 0
A= 1 3 −1  .
 
−1 2 3
A has eigenvalues λ1 = 2, λ2 = λ3 = 3 ± i. 
1
 
1
 
1
For λ2 = 3 + i we find an eigenvector k2 =  1 + i  .
 
2−i
We have  
cos t + i sin t
e(3+i)t k2 = e3t  cos t − sin t + i(cos t + sin t)  ,
 
2 cos t + sin t + i(2 sin t − cos t)

 
cos t
<(e(3+i)t k2 ) = e3t  cos t − sin t 
 
2 cos t + sin t
 
sin t
=(e(3+i)t k2 ) = e3t  cos t + sin t  .
 
2 sin t − cos t
     
1 cos t sin t
x(t) = c1 e2t  0  + c2 e3t  cos t − sin t  + c3 e3t  cos t + sin t  .
     
1 2 cos t + sin t 2 sin t − cos t

 
1 −2 2
A =  −2 1 2 .
 
2 2 1
We have
det(A − λI) = −(λ − 3)2 (λ + 3).
A has eigenvalues λ1 = λ2 = 3, λ3 = −3 (We may say that, λ = 3 is an eigenvalue of algebraic

multiplicity 2, and λ = −3 is a simple eigenvalue).
For λ = 3 we solve the equation Ak = 3k, namely
  
−2 −2 2 k1
 −2 −2 2   k2  = 0.
  
2 2 −2 k3
     
v−u 1 −1
The solution is k =  u  . So we find two eigenvectors k1 =  0  and k2 =  1  .
     
v 1 0
 
1
For λ3 = −3 we find an eigenvector k3 =  1  .
 
−1
     
1 −1 1
x(t) = c1 e3t  0  + c2 e3t  1  + c3 e−3t  1  .
     
1 0 −1
Now we consider the solutions of (3.4.1) associated with a multiple eigenvalue λ, with geometric
multiplicity µ(λ) less than the algebraic multiplicity.
Lemma 3.12 Assume λ is an eigenvalue of A with algebraic multiplicity m > 1. Then the following
system
(A − λI)m v = 0 (3.4.5)
has exactly m linearly independent solutions.
By direct computations we can prove the following theorem.
Theorem 3.13 Assume that λ is an eigenvalue of A with algebraic multiplicity m > 1. Let v0 6= 0
be a solutions of (3.4.5). Define
vl = (A − λI)vl−1 , l = 1, 2, · · · , m − 1, (3.4.6)
and let
t2 tm−1

x(t) = eλt v0 + tv1 + v2 + · · · + vm−1 . (3.4.7)
2 (m − 1)!
Then x(t) is a solution of (3.4.1).
(1) (m)
Let v0 , · · · , v0 be m linearly independent solutions of (3.4.5). They generate m linearly inde-
pendent solutions of (3.4.1) via (3.4.6) and (3.4.7).
Remark. In (3.4.6), we always have
(A − λI)vm−1 = 0.
If vm−1 6= 0 then vm−1 is an eigenvector of A associated with the eigenvalue λ.
In practice, to find the solutions of (3.4.1) associated with an eigenvalue λ of multiplicity m, we first
solve (3.4.5) and find m linearly independent solutions
(1) (2) (m)
v0 , v0 , ··· , v0 .
(k)
For each of these vectors, say v0 , we compute the iteration sequence
(k) (k)
vl = (A − λI)vl−1 , l = 1, 2, · · ·
(k)
There is an integer 0 ≤ j ≤ m − 1 (j depends on the choice of v0 ) such that
(k) (k)
vj 6= 0, (A − λI)vj = 0.
Thus vj is an eigenvector of A associated with the eigenvalue λ. Then the iteration stops and yields
a solution
t2 (k) tj (k)

(k) (k)
x(k) (t) = eλt v0 + tv1 + v2 + · · · + vj . (3.4.8)
2 j!

 
−1 1 0
A =  0 −1 4  .
 
1 0 −4
From det(A − λI) = −λ(λ + 3)2 = 0 we find eigenvalues λ1 = −3 with multiplicity 2, and λ2 = 0
simple.
For the double eigenvalue λ1 = −3 we solve
 
4 4 4
(A + 3I)2 v =  4 4 4  v = 0,
 
1 1 1
   
1 0
(1) (2) (1)
and find two linearly independent solutions v0 =  0  , v0 =  1  . Plugging v0 ,
   
−1 −1
(2)
v0 into (3.4.6), (3.4.7) we get
 
2
(1) (1)
v1 = (A + 3I)v0 =  −4  ,
 
2

  
1 2
(1) (1)
x(1) = e−3t (v0 + tv1 ) = e−3t  0  + t  −4  ,
   
−1 2
 
1
(2) (2)
v1 = (A + 3I)v0 =  −4  ,
 
1
   
0 1
(2) (2)
x(2) = e−3t (v0 + tv1 ) = e−3t  1  + t  −4  .
   
−1 1

4
For the simple eigenvalue λ2 = 0 we find an eigenvector k3 =  4  .
 
1
So the general solution is
x(t) = c1 x(1) + c2 x(2) + c3 k3

         
1 2 0 1 4
−3t  −3t 
= c1 e  0  + t  −4  + c2 e  1  + t  −4  + c3  4  .
       
−1 2 −1 1 1
Example. Solve the system



 x0 = 2x + y + 2z,

y 0 = −x + 4y + 2z,

 0

z = 3z.
 
2 1 2
A =  −1 4 2 .
 
0 0 3
The eigenvalue is λ = 3 with multiplicity 3. Solving
 
0 0 0
(A − 3I)3 v =  0 0 0  v = 0,
 
0 0 0
we obtain 3 obvious linearly independent solutions
     
1 0 0
(1) (2) (3)
v0 =  0  , v0 =  1  , v0 =  0  .
     
0 0 1
(j)
Plugging v0 into (3.4.6), (3.4.7) we get
 
−1
(1) (1)
v1 = (A − 3I)v0 =  −1  ,
 
0
 
0
(1) (1)
v2 = (A − 3I)v1 =  0 ,
 
0
   
1 −1
(1) (1)
x(1) = e3t (v0 + tv1 ) = e3t  0  + t  −1  ;
   
0 0
 
1
(2) (2)
v1 = (A − 3I)v0 =  1  ,
 
0
 
0
(2) (2)
v2 = (A − 3I)v1 =  0  ,
 
0
   
0 1
(2) (2)
x(2) = e3t (v0 + tv1 ) = e3t  1  + t  1  ;
   
0 0
 
2
(3) (3)
v1 = (A − 3I)v0 =  2  ,
 
0
 
0
(3) (3)
v2 = (A − 3I)v1 =  0  ,
 
0
   
0 2
(3) 3t (3) (3) 3t 
x = e (v0 + tv1 ) = e  0  + t  2  .
  
1 0
x(t) = c1 x(1) + c2 x(2) + c3 x(3)
       
1 −1 0 1
3t  3t 
= c1 e  0  + t  −1  + c2 e  1  + t  1 
     
0 0 0 0
   
0 2
3t 
+ c3 e  0  + t  2  .
  
1 0
Remark. It is possible to reduce the number of constant vectors in the general solution of x0 = Ax
by using a basis for the Jordan canonical form of A. We will not go into the details of the Jordan
canonical form. However the following algorithm usually works well if the size of A is small.
Consider an eigenvalue λ of A with algebraic multiplicity m.

Start with r = m. Let v be a vector such that (A − λI)r v = 0 but (A − λI)r−1 v 6= 0. [v is called
a generalized eigenvector of rank r associated with the eigenvalue λ. If no such v exists, reduce r
by 1.] Then
ur = v, ur−1 = (A − λI)v, ur−2 = (A − λI)2 v, . . . , u1 = (A − λI)r−1 v,
form a chain of linearly independent solutions of (3.4.5) with u1 being the base eigenvector corre-
sponding to the eigenvalue λ. This gives r independent solutions of x0 = Ax:
x1 (t) = u1 eλt ,
x2 (t) = (u1 t + u2 )eλt ,
x3 (t) = ( 21 u1 t2 + u2 t + u3 )eλt ,
·
·
·
1 1
xr (t) = ( (r−1)! u1 tr−1 + · · · + 2! ur−2 t
2
+ ur−1 t + ur )eλt .
Repeat this procedure by finding another v which is not in the span of the previous chains of vectors.
Also do this for each eigenvalue of A. Results of linear algebra shows that
(1) Any chain of generalized eigenvectors constitutes a linearly independent set of vectors.
(2) If two chains of generalized eigenvectors are based on linearly independent eigenvectors, then
the union of these vectors is a linearly independent set of vectors (whether the two base eigenvectors
are associated with different eigenvalues or with the same eigenvalue).
 
3 0 0 0
 0 3 0 0 
Example. Solve x0 = Ax, where A =  .
 
 0 1 3 0 
0 0 1 3
 
0 0 0 0
 0 0 0 0 
A has an eigenvalue λ = 3 of multiplicity 4. Direct calculation gives (A−3I) =  ,
 
 0 1 0 0 
0 0 1 0
 
0 0 0 0
 0 0 0 0 
(A − 3I)2 =   , (A − 3I)3 = 0, and (A − 3I)4 = 0.
 
 0 0 0 0 
0 1 0 0
   
1 0
 0   0 
It can be seen that v1 =   and v4 =   are two linearly independent eigenvectors of A.
   
 0   0 
0 1
   
0 0
 1   0 
Together with v2 =   and v3 =  , they form a basis of {v | (A − 3I)4 v = 0} = R4 .
   
 0   1 
0 0
Note that (A−3I)v2 = v3 , and (A−3I)v3 = v4 . Hence {v4 , v3 , v2 } forms a chain of generalized
eigenvectors associated with the eigenvalue 3. {v1 } alone is another chain. Therefore the general
solution is
t2

3t
x(t) = e c1 v1 + c2 (v2 + tv3 + v4 ) + c3 (v3 + tv4 ) + c4 v4 .
2
That is
c1 e3t
 
 c2 e3t 
x(t) =  .
 
 (c2 t + c3 )e3t 
2
( c22t + c3 t + c4 )e3t
Exercise. Solve x0 = Ax, where

 
7 5 −3 2
 0 1 0 0 
A= .
 
 12 10 −5 4 
−4 −4 2 −1
       
1 1 −1 −t
 0   −2   0   0 
Ans: x(t) = c1 e−t   + c2 et   + c3 et   + c4 et  .
       
 2   0   −2   1 − 2t 
−1 2 0 1
3.5 Higher Order Linear Equations

Consider n-th order linear equation
y (n) + a1 (t)y (n−1) + · · · + an−1 (t)y 0 + an (t)y = f (t), (3.5.1)

k
where y (k) = ddtky . Throughout this section we assume that aj (t)’s and f (t) are continuous functions
defined on the interval (a, b). When f (t) 6≡ 0, (3.5.1) is called a non-homogeneous equation. The
associated homogeneous equation is
y (n) + a1 (t)y (n−1) + · · · + an−1 (t)y 0 + an (t)y = 0. (3.5.2)
The general theory of solutions of (3.5.1) and (3.5.2) can be established by applying the results in
the previous sections to the equivalent systems.
We begin with the initial value problem



 y (n) + a1 (t)y (n−1) + · · · + an (t)y = f (t),

y(t ) = y0 ,


 0


y 0 (t0 ) = y1 , (3.5.3)


·········





 y (n−1) (t ) = y

0 n−1 .
3.5. HIGHER ORDER LINEAR EQUATIONS 51
Theorem 3.14 Assume that a1 (t), · · · , an (t) and f (t) are continuous functions defined on the in-
terval (a, b). Then for any t0 ∈ (a, b) and for any numbers y0 , · · · , yn−1 , the initial value problem
(3.5.3) has a unique solution defined on (a, b).
Especially if aj (t)’s and f (t) are continuous on R, then for any t0 and y0 , · · · , yn−1 , the initial
value problem (3.5.3) has a unique solution defined on R.
Next we consider the structure of solutions of the homogeneous equation (3.5.2).
Definition. Functions φ1 (t), · · · , φr (t) are linearly dependent on (a, b) if there exists constants c1 ,
· · · , cr , not all zero, such that
c1 φ1 (t) + · · · + cr φr (t) = 0
for all t ∈ (a, b). A set of functions are linearly independent on (a, b) if they are not linearly
dependent on (a, b).
Lemma 3.15 Functions φ1 (t), · · · , φr (t) are linearly dependent on (a, b) if and only if the following
vector-valued functions
   
φ1 (t) φr (t)
 φ0 (t)   φ0 (t) 
1 r
, ··· , 
   
··· ···
 
   
(n−1) (n−1)
φ1 (t) φr (t)
are linearly dependent on (a, b).
Proof. “⇐=” is obvious. To show “=⇒”, assume that φ1 , · · · , φr are linearly dependent on (a, b).
There exists constants c1 , · · · , cr , not all zero, such that
c1 φ1 (t) + · · · + cr φr (t) = 0
for all t ∈ (a, b). Differentiate this equality successively we find
c1 φ01 (t) + · · · + cr φ0r (t) = 0,

··················
(n−1)
c1 φ1 (t) + · · · + cr φ(n−1)
r (t) = 0.
Thus    
φ1 (t) φr (t)
 φ01 (t)   φ0r (t) 
c1   + · · · + cr  =0
   
 ···   ··· 
(n−1) (n−1)
φ1 (t) φr (t)
for all t ∈ (a, b). Hence the r vector-valued functions are linearly dependent on (a, b).
The Wronskian of n functions φ1 (t), · · · , φn is defined by

φ1 (t)
··· φn (t)

W (φ1 , · · · , φn )(t) = ··· ··· ··· . (3.5.4)

(n−1) (n−1)
φ
1 (t) · · · φn (t)
From Lemma 3.15 and Theorem 3.5 we get
Proposition 3.5.1 Let y1 (t), · · · , yn (t) be n solutions of (3.5.2) on (a, b). They are linearly in-
dependent on (a, b) if and only if their Wronskian W (t) ≡ W (y1 , · · · , yn )(t) does not vanish on
(a, b).
Theorem 3.16 Let a1 (t), · · · , an (t) be continuous on the interval (a, b). The homogeneous equa-
tion (3.5.2) has n linearly independent solutions on (a, b).
Let y1 , · · · , yn be n linearly independent solutions of (3.5.2) defined on (a, b). The general solution
of (3.5.2) is given by
y(t) = c1 y1 (t) + · · · + cn yn (t), (3.5.5)
where c1 , · · · , cn are arbitrary constants.
Any set of n linearly independent solutions is called a fundamental set of solutions.

Now we consider the non-homogeneous equation (3.5.1). We have
Theorem 3.17 Let yp be a particular solution of (3.5.1), and y1 , · · · , yn be a fundamental set of

solutions for the associated homogeneous equation (3.5.2). The general solution of (3.5.1) is given
by
y(t) = c1 y1 (t) + · · · + cn yn (t) + yp (t). (3.5.6)
From (3.3.3) we can derive the variation of parameter formula for higher order equations. Consider
a second order equation
y 00 + p(t)y 0 + q(t)y = f (t). (3.5.7)
!
x1
Let x1 = y, x2 = y 0 , x = . Then (3.5.7) is written as
x2
! !
0 0 1 0
x = x+ (3.5.8)
−q −p f
Assume y1 (t) and y2 (t) are two linearly independent solutions of the associated homogeneous equa-
tion
y 00 + py 0 + qy = 0.
We look for a solution of (3.5.7) in the form
y = u 1 y1 + u 2 y2 .
!
y1 y 2
Choose a fundamental matrix Φ(t) = . The corresponding solution of (3.5.8) is in the
y10 y20
form ! ! ! !
y y1 y 2 u1 y1 u 1 + y2 u 2
x= = = (3.5.9)
y0 y10 y20 u2 y10 u1 + y20 u2
3.5. HIGHER ORDER LINEAR EQUATIONS 53
Recall that, if ad − bc 6= 0, then

!−1 !
a b 1 d −b
= .
c d ad − bc −c a
Thus !
−1 1 y20 −y2
Φ(t) = ,
W (t) −y10 y1
where W (t) is the Wronskian of y1 (t), y2 (t). Using (3.3.3) we can derive
Z Z
y2 (t) y1 (t)
u1 (t) = − f dt, u2 (t) = f (t)dt. (3.5.10)
W (t) W (t)
Note that, (3.5.9) implies
y10 u1 + y20 u2 = y 0 = y10 u1 + y20 u2 + y1 u01 + y2 u02 .
Hence
y1 u01 + y2 u02 = 0. (i)
Plugging y 0 = y10 u1 + y20 u2 into (3.5.7) we find
y10 u01 + y20 u02 = f. (ii)
Solving (i) (ii) we find

y2 y1
u01 = − f, u02 = f.
W W
Again we get (3.5.10).
Now we consider linear equations with constant coefficients
y (n) + a1 y (n−1) + · · · + an−1 y 0 + an y = f (t), (3.5.11)
and the associated homogeneous equation
y (n) + a1 y (n−1) + · · · + an−1 y 0 + an y = 0, (3.5.12)
where a1 , · · · , an are real constants. Recall that (3.5.12) is equivalent to a system
x0 = Ax,
where  
0 1 ··· 0 0
0 0 ··· 0 0
 
 
 
A=
 ··· ··· ··· ··· ··· .

0 0 ··· 0 1
 
 
−an −an−1 ··· −a2 −a1
The equation for the eigenvalues of A is
det(λI − A) = λn + a1 λn−1 + · · · + an−1 λ + an = 0. (3.5.13)

The solutions of (3.5.13) are called characteristic values or eigenvalues for the equation (3.5.12).
Let λ1 , · · · , λs be the distinct eigenvalues for (3.5.12). Then we can write
λn + a1 λn−1 + · · · + an−1 λ + an
= (λ − λ1 )m1 (λ − λ2 )m2 · · · (λ − λs )ms ,
where m1 , · · · , ms are positive integers and
m1 + · · · + ms = n.
We call them the multiplicity of the eigenvalues λ1 , · · · , λs respectively.
Lemma 3.18 Assume λ is an eigenvalue of (3.5.12) of multiplicity m.

(i) eλt is a solution of (3.5.12).
(ii) If m > 1, then for any positive integer 1 ≤ k ≤ m − 1, tk eλt is a solution of (3.5.12).
(iii) If λ = α+iβ, then tk eαt cos(βt), tk eαt sin(βt) are solutions of (3.5.12), where 0 ≤ k ≤ m−1.
Theorem 3.19 Let λ1 , · · · , λs be the distinct eigenvalues for (3.5.12), with multiplicity m1 , · · · , ms
respectively. Then (3.5.12) has a fundamental set of solutions
eλ1 t , teλ1 t , · · · , tm1 −1 eλ1 t ;

························ ; (3.5.14)
eλs t , teλs t , · · · , tms −1 eλs t .
Proof. One way to prove Theorem 3.19 is to find a fundamental matrix of the equivalent system,
such that the first row is given by the functions in (3.5.14).
Another way to prove Theorem 3.19 is to show that each function in (3.5.14) is a solution of (3.5.12),
and they are linearly independent.
Remark. If (3.5.12) has a complex eigenvalue λ = α + iβ, then λ̄ = α − iβ is also an eigenvalue.

Thus both tk e(α+iβ)t and tk e(α−iβ)t appear in (3.5.14), where 0 ≤ k ≤ m − 1. In order to obtain a
fundamental set of real solutions, the pair of solutions tk e(α+iβ)t and tk e(α−iβ)t in (3.5.14) should
be replaced by tk eαt cos(βt) and tk eαt sin(βt) respectively.
3.6 Appendix 1: Proof of Lemma 3.12
Lemma 3.12 Let A be an n×n complex matrix and λ an eigenvalue of A with algebraic multiplicity
m. Then
dim {x ∈ Cn | (λI − A)m x = 0} = m.
Proof. The proof consists of several steps. Let T = {x ∈ Cn | (λI − A)m x = 0}. The space T is
called the generalized eigenspace corresponding to the eigenvalue λ.
3.6. APPENDIX 1: PROOF OF LEMMA 3.12 55
Step 1. T is a subspace of Cn . This is just direct verification.

Step 2. T is invariant under A meaning A[T ] ⊆ T . This is because if we take a vector x in T , then
(λI − A)m x = 0 so that A(λI − A)m x = 0, which is the same as (λI − A)m (Ax) = 0. Therefore,
Ax is also in T .
Step 3. By Step 2 which says that A[T ] ⊆ T , we may consider A as a linear transformation on
the subspace T . In other words, A : T −→ T . Let µ be an eigenvalue of A : T −→ T . That
is Av = µv. Then v ∈ T implies that (λI − A)m v = 0. Since Av = µv, this simplifies to
(λ − µ)m v = 0. Being an eigenvector, v 6= 0 so that µ = λ. Therefore, all eigenvalues of
A : T −→ T are equal to λ.
Step 4. Let dim T = r. Certainly r ≤ n. Then by Step 3, the characteristic polynomial of
A : T −→ T is (λ − z)r . Since T is an invariant subspace of A : Cn −→ Cn , one can choose a
basis of T and then extend it to a basis of Cn so that with respect to this new basis of Cn , the matrix
A is similar to a matrix where the upper left hand r × r submatrix represents A on T and the lower
left hand (n − r) × r submatrix is the zero matrix. From this, we see that (λ − z)r is a factor of the
characteristic polynomial of A : Cn −→ Cn . Hence r ≤ m.
We also need the Cayley-Hamilton Theorem in the last step of the proof.
Cayley-Hamilton Theorem Let p(z) be the characteristic polynomial of an n × n matrix A. Then
p(A) = 0.
Step 5. Let p(z) = (λ1 − z)m1 · · · (λk − z)mk be the characteristic polynomial of A, where
λ1 , λ2 , . . . , λk are the distinct eigenvalues of A : Cn −→ Cn . For each i from 1 to k, let
Ti = {x ∈ Cn | (λi I − A)mi x = 0}.
By Step 4, we have dimTi ≤ mi .

Let fi (z) = (λi − z)mi and gi (z) = f1 (z) · · · fˆi (z) · · · fk (z), where fˆi (z) means the polynomial
fi (z) is omitted. Note that fi (z)gi (z) = p(z) for all i.
1
Resolving (λ1 −z)m1 ···(λ m
k −z) k
into partial fractions, we have the identity
1 h1 (z) h2 (z) hk (z)

≡ + + ··· + ,
(λ1 − z)m1 · · · (λk − z)mk (λ1 − z) m1 (λ2 − z) m2 (λk − z)mk
where h1 (z), . . . , hk (z) are polynomials in z. Finding the common denominator of the right hand
side and equate the numerators on both sides, we have 1 ≡ g1 (z)h1 (z) + · · · + gk (z)hk (z). Substi-
tuting the matrix A into this polynomial identity, we have
g1 (A)h1 (A) + · · · + gk (A)hk (A) = I,
where I is the identity n × n matrix.

Now for any x ∈ Cn , we have
g1 (A)h1 (A)x + · · · + gk (A)hk (A)x = x.
Note that each gi (A)hi (A)x is in Ti because fi (A)[gi (A)hi (A)x] = p(A)hi (A)x = 0 by the
Cayley-Hamilton Theorem. This shows that any vector in Cn can be expressed as a sum of vectors
where the i-summand is in Ti . In other words,
Cn = T 1 + T 2 + · · · + T k .
Consequently, m1 + · · · mk = n ≤ dimT1 + · · · + dimTk ≤ m1 + · · · + mk so that dimTi = mi .
Remarks
1. In fact
Cn = T 1 ⊕ · · · ⊕ T k .
2. If A is a real matrix and λ is a real eigenvalue of A of algebraic multiplicity m, then T is a real

vector space and the real dimension of T is m. This is because for any set of real vectors in Rn , it is
linearly independent over R if and only if it is linearly independent over C.
3. If A is a real matrix and λ is a complex eigenvalue of A of algebraic multiplicity m, then λ is
also an eigenvalue of A of algebraic multiplicity m. In this case, if {v1 , . . . , vm } is a basis over
C of Tλ , where Tλ is the generalized eigenspace corresponding to λ, then {v1 , . . . , vm } is a basis
over C of Tλ . It can be shown that the 2m real vectors <v1 , . . . , <vm , =v1 , . . . , =vm are linearly
independent over R and form a basis of (Tλ ⊕ Tλ ) ∩ Rn .
Chapter 4
Power Series Solutions
4.1 Power Series

An infinite series of the form
X∞
an (x − x0 )n = a0 + a1 (x − x0 ) + a2 (x − x0 )2 + · · · (4.1.1)
n=0
is a power series in x − x0 . In what follows, we will be focusing mostly at the point x0 = 0. That is
∞
X
an xn = a0 + a1 x + a2 x2 + · · · (4.1.2)
n=0
m
X
(4.1.2) is said to converge at a point x if the limit lim an xn exists, and in this case the sum of
m→∞
n=0
the series is the value of this limit. It is obvious that (4.1.2) always converges at x = 0. It can be
showed that each power series like (4.1.2) corresponds to a positive real number R, called the radius
of convergence, with the property that the series converges if |x| < R and diverges if |x| > R. It
is customary to put R equal to 0 when the series converges only at x = 0, and equal to ∞ when it
converges for all x. In many important cases, R can be found by the ratio test as follow.
If each an 6= 0 in (4.1.2), and if for a fixed point x 6= 0 we have
an+1 xn+1

= lim an+1 |x| = L,

lim n
n→∞ an x n→∞ an
then (4.1.2) converges for L < 1 and diverges if L > 1. It follows from this that

an
R = lim
n→∞ an+1
if this limit exists (we put R = ∞, if |an /an+1 | −→ ∞)

The interval (−R, R) is called the interval of convergence in the sense that inside the interval the
series converges and outside the interval the series diverges.
Consider the following power series

∞
X
n!xn = 1 + x + 2!x2 + 3!x3 + · · · (4.1.3)
n=0
57
58 CHAPTER 4. POWER SERIES SOLUTIONS
∞
X xn x2 x3
=1+x+ + + ··· (4.1.4)
n=0
n! 2! 3!
∞
X
xn = 1 + x + x2 + x3 + · · · (4.1.5)
n=0
It is easy to verify that (4.1.3) converges only at x = 0. Thus R = 0. For (4.1.4), it converges for all
x so that R = ∞. For (4.1.5), the power series converges for |x| < 1 and R = 1.
Suppose that (4.1.2) converges for |x| < R with R > 0, and denote its sum by f (x). That is
∞
X
f (x) = an xn = a0 + a1 x + a2 x2 + · · · (4.1.6)
n=0
Then one can prove that f is continuous and has derivatives of all orders for |x| < R. Also the series
can be differentiated termwise in the sense that
∞
X
f 0 (x) = nan xn−1 = a1 + 2a2 x + 3a3 x2 + · · · ,
n=1
∞
X
f 00 (x) = n(n − 1)an xn−2 = 2a2 + 3 · 2a3 x + · · · ,
n=2
and so on. Furthermore, the resulting series are still convergent for |x| < R. These successive
differentiated series yield the following basic formula relating an to with f (x) and its derivatives.
f n (0)
an = (4.1.7)
n!
Moreover, (4.1.6) can be integrated termwise provided the limits of integration lie inside the interval
of convergence.
If
∞
X
g(x) = bn xn = b0 + b1 x + b2 x2 + · · · (4.1.8)
n=0
is another power series with interval of convergence |x| < R, then (4.1.6) and (4.1.8) can be added
or subtracted termwise:
∞
X
f (x) ± g(x) = (an ± bn )xn = (a0 ± b0 ) + (a1 ± b1 )x + (a2 ± b2 )x2 + · · · (4.1.9)
n=0
They can also be multiplied like polynomials, in the sense that

∞
X
f (x)g(x) = cn xn ,
n=0
where cn = a0 bn + a1 bn−1 + · · · + an b0 .
Suppose two power series (4.1.6) and (4.1.8) converge to the same function so that f (x) = g(x) for
|x| < R, then (4.1.7) implies that they have the same coefficients, an = bn for all n. In particular, if
f (x) = 0 for all |x| < R, then an = 0, for all n.
4.1. POWER SERIES 59
Let f (x) be a continuous function that has derivatives of all orders for |x| < R. Can it be represented
by a power series? If we use (4.1.7), it is natural to expect
∞
X f (n) (0) n f 00 (0) 2
f (x) = x = f (0) + f 0 (0)x + x + ··· (4.1.10)
n=0
n! 2!
to hold for all |x| < R. Unfortunately, this is not always true. Instead, one can use Taylor’s
expansion for f (x):
n
X f (k) (0) k
f (x) = x + Rn (x),
k!
k=0
where the remainder Rn (x) is given by
f (n+1) (x) n+1
Rn (x) = x
(n + 1)!
for some point x between 0 and x. To verify (4.1.6), it suffices to show that Rn (x) −→ 0 as
n −→ ∞.
Example The following familiar expansions are valid for all x.

∞
X xn x2 x3
ex = =1+x+ + + ···
n=0
n! 2! 3!
∞
X x2n+1 x3 x5
sin x = (−1)n =x− + − ···
n=0
(2n + 1)! 3! 5!
∞
X x2n x2 x4
cos x = (−1)n =1− + − ···
n=0
(2n)! 2! 4!
A function f (x) with the property that a power series expansion of the form
∞
X
f (x) = an (x − x0 )n (4.1.11)
n=0
is valid in some interval containing the point x0 is said to be analytic at x0 . In this case, an is
necessarily given by
f (n) (x0 )
an = ,
n!
and (4.1.11) is called the Taylor series of f (x) at x0 .
Thus ex , sin x, cos x are analytic at all points. Concerning analytic functions, we have the following
basic results.
1. Polynomials, ex , sin x, cos x are analytic at all points.

2. If f (x) and g(x) are analytic at x0 , then f (x) ± g(x), f (x)g(x), and f (x)/g(x) [provided
g(x0 ) 6= 0] are also analytic at x0 .
3. If f (x) is analytic at x0 , and f −1 (x) is a continuous inverse, then f −1 (x) is analytic at f (x0 ) if
f 0 (x0 ) 6= 0.
4. If g(x) is analytic at x0 and f (x) is analytic at g(x0 ), then f (g(x)) is analytic at x0 .
5. The sum of a power series is analytic at all points inside the interval of convergence.
4.2 Series Solutions of First Order Equations

A first order differential equation y 0 = f (x, y) can be solved by assuming that it has a power series
solution. Let’s illustrate this with two familiar examples.
Example. Consider the differential equation y 0 = y. We assume it has a power series solution of the
form
y = a0 + a1 x + a2 x2 + · · · + an xn + · · · (4.2.1)
that converges for |x| < R. That is the equation y 0 = y has a solution which is analytic at the origin.
Then
y 0 = a1 + 2a2 x + · · · + nan xn−1 + · · · (4.2.2)
has the same interval of convergence. Since y 0 = y, the series (4.2.1) and (4.2.2) have the same
coefficients. That is
(n + 1)an+1 = an , all for n = 0, 1, 2, . . .
1 1 1
Thus an = n an−1 = n(n−1) an−2 = ··· = n! a0 . Therefore
x2 x3 xn

y = a0 1 + x + + + ··· + + ··· ,
2! 3! n!
where a0 is an arbitrary constant. In this case, we recognize this as the power series of ex . Thus the
general solution is y = a0 ex .
Example. The function y = (1 + x)p , where p is a real constant satisfies the differential equation
(1 + x)y 0 = py, y(0) = 1. (4.2.3)
As before, we assume it has a power series solution of the form
y = a0 + a1 x + a2 x2 + · · · + an xn + · · ·
with positive radius of convergence. Then
y0 = a1 + 2a2 x + 3a3 x2 + · · · + (n + 1)an+1 xn + · · · ,

xy 0 = a1 x + 2a2 x2 + · · · + nan xn + · · · ,
2
py = pa0 + pa1 x + pa2 x + · · · + pan xn + · · · ,
Using (4.2.3) and equating coefficients, we have
(n + 1)an+1 + nan = pan , for all n = 0, 1, 2, . . .
That is
p−n
an+1 = an ,
n+1
so that
p(p − 1) p(p − 1)(p − 2) p(p − 1) · · · (p − n + 1)
a1 = p, a2 = , a3 = , ..., an = .
2 2·3 n!
4.3. SECOND ORDER LINEAR EQUATIONS AND ORDINARY POINTS 61
In other words,
p(p − 1) 2 p(p − 1)(p − 2) 3 p(p − 1) · · · (p − n + 1) n

y = 1 + px + x + x + ··· + x + ··· .
2 2·3 n!
By ratio test, this series converges for |x| < 1. Since (4.2.3) has a unique solution, we conclude that
p(p − 1) 2 p(p − 1)(p − 2) 3 p(p − 1) · · · (p − n + 1) n

(1 + x)p = 1 + px + x + x +···+ x +··· ,
2 2·3 n!
for |x| < 1, This is just the binomial series of (1 + x)p .
4.3 Second Order Linear Equations and Ordinary Points

Consider the homogeneous second order linear differential equation
y 00 + P (x)y 0 + Q(x)y = 0 (4.3.1)
Definition The point x0 is said to be an ordinary point of (4.3.1) if P (x) and Q(x) are analytic at
x0 . If at x = x0 , P (x) and/or Q(x) are not analytic, then x0 is said to be a singular point of (4.3.1).
A singular point x0 at which the functions (x − x0 )P (x) and (x − x0 )2 Q(x) are analytic is called
a regular singular point of (4.3.1). If a singular point x0 is not a regular singular point, then it is
called an irregular singular point.
Example. If P (x) and Q(x) are constant, then every point is an ordinary point of (4.3.1).
Example. Consider the equation y 00 + xy = 0. Since the function Q(x) = x is analytic at every
point, every point is an ordinary point.
Example. In the Cauchy-Euler equation y 00 + ax1 y 0 + xa22 y = 0, where a1 and a2 are constants, the
point x = 0 is a singular point, but every other point is an ordinary point.
Example. Consider the differential equation
1 8
y 00 + y0 + y = 0.
(x − 1)2 x(x − 1)
The singular points are 0 and 1. At the point 0, xP (x) = x(1 − x)−2 and x2 Q(x) = −8x(1 − x)−1 ,
which are analytic at x = 0, and hence the point 0 is a regular singular point. At the point 1, we
have (x − 1)P (x) = 1/(x − 1) which is not analytic at x = 1, and hence the point 1 is an irregular
singular point.
To discuss the behavior of the singularities at infinity, we use the transformation x = 1/t, which con-
verts the problem to the behavior of the transformed equation near the origin. Using the substitution
x = 1/t, (4.3.1) becomes
d2 y

2 1 1 dy 1 1
2
+ − 2P( ) + 4 Q( )y = 0 (4.3.2)
dt t t t dt t t
We define the point at infinity to be an ordinary point, a regular singular point, or an irregular singular
point of (4.3.1) according as the origin of (4.3.2) is an ordinary point, a regular singular point, or an
irregular singular point.
Example. Consider the differential equation
d2 y

1 1 1 dy 1
+ + + y = 0.
dx2 2 x2 x dx 2x3
The substitution x = 1/t transforms the equation into
d2 y

3 − t dy 1
+ + y = 0.
dt2 2t dt 2t
Hence the point at infinity is a regular singular point of the original differential equation.
Theorem 4.1 Let x0 be an ordinary point of the differential equation
y 00 + P (x)y 0 + Q(x)y = 0,
and let a0 and a1 be arbitrary constants. Then there exists a unique function y(x) that is analytic
at x0 , is a solution of the differential equation in an interval containing x0 , and satisfies the initial
conditions y(x0 ) = a0 , y 0 (x0 ) = a1 . Furthermore, if the power series expansions of P (x) and
Q(x) are valid on an interval |x − x0 | < R, R > 0, then the power series expansion of this solution
is also valid on the same interval.
Example. Find two linearly independent solutions of y 00 − xy 0 − x2 y = 0

1 4 1 6 3
Ans. y1 (x) = 1 + 12 x + 90 x + 1120 x8 + · · · and y2 (x) = x + 16 x3 + 403 5
x + 13
1008 x
7
+ ···.
Example. Using power series method, solve the initial value problem (1 + x2 )y 00 + 2xy 0 − 2y = 0,
y(0) = 0, y 0 (0) = 1.
Ans. y = x.
Legendre’s equation
(1 − x2 )y 00 − 2xy 0 + p(p + 1)y = 0,
where p is a constant called the order of Legendre’s equation.

2x p(p+1)
That is P (x) = − 1−x 2 and Q(x) = 1−x2 . The origin is an ordinary point, and we expect a
n
P
solution of the form y = an x . Thus the left hand side of the equation becomes
∞
X ∞
X ∞
X
(1 − x2 ) (n + 1)(n + 2)an+2 xn − 2x (n + 1)an+1 xn + p(p + 1) an xn ,
n=0 n=0 n=0
or
∞
X ∞
X ∞
X ∞
X
(n + 1)(n + 2)an+2 xn − (n − 1)nan xn − 2nan xn + p(p + 1)an xn .
n=0 n=2 n=1 n=0
4.3. SECOND ORDER LINEAR EQUATIONS AND ORDINARY POINTS 63
The sum of these series is required to be zero, so the coefficient of xn must be zero for every n. This
gives
(n + 1)(n + 2)an+2 − (n − 1)nan − 2nan + p(p + 1)an = 0,
for n = 2, 3, . . . In other words,
(p − n)(p + n + 1)
an+2 = − an .
(n + 1)(n + 2)
This recursion formula enables us to express an in terms of a0 or a1 according as n is even or odd.

In fact, for m > 0, we have
p(p − 2)(p − 4) · · · (p − 2m + 2)(p + 1)(p + 3) · · · (p + 2m − 1)

a2m = (−1)m a0 ,
(2m)!
(p − 1)(p − 3) · · · (p − 2m + 1)(p + 2)(p + 4) · · · (p + 2m)

a2m+1 = (−1)m a1 .
(2m + 1)!
With that, we get two linearly independent solutions
∞
X ∞
X
y1 (x) = a2m x2m and y2 (x) = a2m+1 x2m+1 ,
m=0 m=0
and the general solution is given by

p(p + 1) 2 p(p − 2)(p + 1)(p + 3) 4
y = a0 1 − x + x
2! 4!

p(p − 2)(p − 4)(p + 1)(p + 3)(p + 5) 6
− x + ···
6!

(p − 1)(p + 2) 3 (p − 1)(p − 3)(p + 2)(p + 4) 5
+a1 x − x + x
3! 5!

(p − 1)(p − 3)(p − 5)(p + 2)(p + 4)(p + 6) 7
− x + ··· .
7!
When p is not an integer, the series representing y1 and y2 have radius of convergence R = 1. For
example,
a2n+2 x2n+2 (p − 2n)(p + 2n + 1) 2

= − |x | −→ |x|2
a2n x2n (2n + 1)(2n + 2)
as n −→ ∞, and similarly for the second series. In fact, by Theorem 4.1, and the familiar expansion
1
= 1 + x2 + x4 + · · · , R = 1,
1 − x2
an xn must be
P
that R = 1 for both P (x) and Q(x). Thus, we know any solution of the form y =
valid at least for |x| < 1.
The functions defined in the series solution of Legendre’s equation are called Legendre functions.
When p is a nonnegative integer, one of these series terminates and becomes a polynomial in x.
For instance, if p = n is an even positive integer, the series representing y1 terminates and y1 is a
polynomial of degree n. If p = n is odd, y2 again is a polynomial of degree n. These are called
Legendre polynomials Pn (x) and they give particular solutions to Legendre’s equation
(1 − x2 )y 00 − 2xy 0 + n(n + 1)y = 0,
where n is a nonnegative integer. It is customary to choose the arbitrary constants a0 or a1 so that

the coefficient of xn in Pn (x) is (2n)!/[2n (n!)2 ] so that Pn (1) = 1. Then
bn/2c
X (−1)k (2n − 2k)!
Pn (x) = xn−2k .
2n k!(n − k)!(n − 2k)!
k=0
The six Legendre polynomials are
P0 = 1, P1 (x) = x
P2 (x) = 21 (3x2 − 1), P3 (x) = 12 (5x3 − 3x)
P4 (x) = 18 (35x4 − 30x2 + 3), P5 (x) = 18 (63x5 − 70x3 + 15x)
There is also a Rodrigues’ formula for the Legendre polynomial given by
1 dn 2
Pn (x) = (x − 1)n .
n!2n dxn
Hermite’s equation y 00 − 2xy 0 + 2py = 0, where p is a constant. The general solution of Hermite’s
equation is y(x) = a0 y1 (x) + a1 y2 (x), where
2p 2 22 p(p − 2) 4 23 p(p − 2)(p − 4) 6

y1 (x) = 1 − x + x − x + ··· ,
2! 4! 6!
2(p − 1) 3 22 (p − 1)(p − 3) 5 23 (p − 1)(p − 3)(p − 5) 7

y2 (x) = x − x + x − x + ··· .
3! 5! 7!
By Theorem 4.1, both series for y1 and y2 converge for all x. Note that y1 is a polynomial if p is an
even integer, whereas y2 is a polynomial if p is an odd integer.
The Hermite polynomial of degree n denoted by Hn (x) is the nth-degree polynomial solution of
Hermite’s equation, multiplied by a suitable constant so that the coefficient of xn is 2n . The first six
Hermite’s polynomials are
H0 (x) = 1, H1 (x) = 2x,

H2 (x) = 4x2 − 2, H3 (x) = 8x3 − 12x,
H4 (x) = 16x4 − 48x2 + 12, H5 (x) = 32x5 − 160x3 + 120x
A general formula for the Hermite polynomials is
2 dn −x2
Hn = (−1)n ex e .
dxn
4.4. REGULAR SINGULAR POINTS AND THE METHOD OF FROBENIUS 65
4.4 Regular singular points and the method of Frobenius

Consider the second order linear homogeneous differential equation
x2 y 00 + xp(x)y 0 + q(x)y = 0, (4.4.1)
where p(x) and q(x) are analytic at x = 0. In other words, 0 is a regular singular point of (4.4.1).
Let p(x) = p0 + p1 x + p2 x2 + p3 x3 + · · · , and q(x) = q0 + q1 x + q2 x2 + q3 x3 + · · · . Suppose
(4.4.1) has a series solution of the form
∞
X ∞
X
y = xr an xn = an xn+r (4.4.2)
n=0 n=0
An infinite series of the form (4.4.2) is called a Frobenius series, and the method that we are going
to describe is called the method of Frobenius. We may assume a0 6= 0 because the series must have
a first nonzero term. Termwise differentiation gives
∞
X
y0 = an (n + r)xn+r−1 , (4.4.3)
n=0
and
∞
X
00
y = an (n + r)(n + r − 1)xn+r−2 . (4.4.4)
n=0
Substituting the series of y, y 0 and y 00 into (4.4.1) yields
[r(r − 1)a0 xr + (r + 1)ra1 xr+1 + · · · ] + [p0 x + p1 x2 + · · · ] · [ra0 xr−1 + (r + 1)a1 xr + · · · ]

+[q0 + q1 x + · · · ] · [a0 xr + a1 xr+1 + · · · ] = 0.
(4.4.5)
r
The lowest power of x in (4.4.5) is x . If (4.4.5) is to be satisfied identically, the coefficient r(r −
1)a0 + p0 ra0 + q0 a0 of xr must vanish. As a0 6= 0, it follows that r satisfies the quadratic equation
r(r − 1) + p0 r + q0 = 0. (4.4.6)
This is the same equation obtained with the Cauchy-Euler equation. Equation (4.4.6) is called the
indicial equation of (4.4.1) and its two roots (possibly equal) are the exponents of the differential
equation at the regular singular point x = 0.
Let r1 and r2 be the roots of the indicial equation. If r1 6= r2 , then there are two possible Frobenius
solutions and they are linearly independent. Whereas r1 = r2 , there is only one possible Frobenius
series solution. The second one cannot be a Frobenius series and can only be found by other means.
Example. Find the exponents in the possible Frobenius series solutions of the equation
2x2 (1 + x)y 00 + 3x(1 + x)3 y 0 − (1 − x2 )y = 0.
Solution. Clearly x = 0 is a regular singular point since p(x) = 32 (1 + x)2 and q(x) = − 12 (1 − x)
are polynomials. Rewrite the equation in the standard form:
3
2 (1 + 2x + x2 ) 0 − 12 (1 − x)
y 00 + y + y = 0.
x x2
3
We see that p0 = 2 and q0 = − 12 . Hence the indicial equation is
3 1 1 1 1
r(r − 1) + r − = r2 + r − = (r + 1)(r − ) = 0,
2 2 2 2 2
1
with roots r1 = 2 and r2 = −1. The two possible Frobenius series solutions are of the forms
∞ ∞
1
X X
y1 (x) = x 2 an xn and y2 (x) = x−1 an xn .
n=0 n=0
Once the exponents r1 and r2 are known, the coefficients in a Frobenius series solution can be found
by substitution of the series (4.4.2),(4.4.3) and (4.4.4) into the differential equation (4.4.1). If r1 and
r2 are complex conjugates, we always get two linearly independent solutions. We shall restrict our
attention for real solutions of the indicial equation and seek solutions only for x > 0. The solutions
on the interval x < 0 can be studied by changing the variable to t = −x and solving the resulting
equation for t > 0.
Let’s work out the recursion relations for the coefficients. By (4.4.3), we have
∞
!" ∞ #
1 X X
1 0
x p(x)y = x pn xn an (n + r)xn+r−1
n=0 n=0
∞
!" ∞ #
X X
r−2 n n
=x pn x an (n + r)x
n=0 n=0
∞
" n
#
X X
= xr−2 pn−k ak (r + k) xn
n=0 "k=0
∞ n−1
#
X X
r−2
=x pn−k ak (r + k) + p0 an (r + n) xn .
n=0 k=0
Also we have
∞ ∞
! !
1 X X
1
x2 q(x)y = x2 qn xn an xr+n
n=0 !n=0∞
∞
!
1 X
n
X
n
= qn x an x
xr−2 n=0 n=0
∞ n
!
X X
= xr−2 qn−k ak xn
n=0 k=0
∞ n−1
!
X X
r−2
=x qn−k ak + q0 an xn .
n=0 k=0
Substituting these into the differential equation (4.4.1) and cancelling the term xr−2 , we have
∞
( n−1
)
X X
an [(r + n)(r + n − 1) + (r + n)p0 + q0 ] + ak [(r + k)pn−k + qn−k ] xn = 0.
n=0 k=0
4.4. REGULAR SINGULAR POINTS AND THE METHOD OF FROBENIUS 67
Thus, equating the coefficients to zero, we have for n ≥ 0,

n−1
X
an [(r + n)(r + n − 1) + (r + n)p0 + q0 ] + ak [(r + k)pn−k + qn−k ] = 0. (4.4.7)
k=0
When n = 0, we get r(r − 1) + rp0 + q0 = 0, which is true because r is a root of the indicial
equation. Then an can be determined by (4.4.7) recursively provided
(r + n)(r + n − 1) + (r + n)p0 + q0 6= 0.
This would be the case if the two roots of the indicial equation do not differ by an integer. Suppose
r1 > r2 are the two roots of the indicial equation with r1 = r2 + N for some positive integer
N . If we start with the Frobenius series with the smaller exponent r2 , then at the N -th step the
process breaks off because the coefficient aN in (4.4.7) is zero. In this case, only the Frobenius
series solution with the larger exponent exists. The other solution cannot be a Frobenius series.
Theorem 4.2 Assume that x = 0 is a regular singular point of the differential equation (4.4.1) and
that the power series expansions of p(x) and q(x) are valid on an interval |x| < R with R > 0. Let
the indicial equation (4.4.6) have real roots r1 and r2 with r1 ≥ r2 . Then (4.4.1) has at least one
solution
X∞
y1 = xr1 an xn , (a0 6= 0) (4.4.8)
n=0
on the interval 0 < x < R, where an are determined in terms of a0 by the recursion formula (4.4.7)
P∞
with r replaced by r1 , and the series n=0 an xn converges for |x| < R. Furthermore, if r1 − r2 is
not zero or a positive integer, then equation (4.4.1) has a second independent solution
∞
X
y1 = xr2 an xn , (a0 6= 0) (4.4.9)
n=0
on the same interval, where an are determined in terms of a0 by the recursion formula (4.4.7) with
P∞
r replaced by r2 , and again the series n=0 an xn converges for |x| < R.
Remark. (1) If r1 = r2 , then there cannot be a second Frobenius series solution. (2) If r1 − r2 = n
is a positive integer and the summation of (4.4.7) is nonzero, then there cannot be a second Frobenius
series solution. (3) If r1 − r2 = n is a positive integer and the summation of (4.4.7) is zero, then
an is unrestricted and can be assigned any value whatever. In particular, we can put an = 0 and
continue to compute the coefficients without difficulties. Hence, in this case, there does exist a
second Frobenius series solution. In many cases of (1) and (2), it is possible to determine a second
solution by the method of variation of parameters. For instance a second solution for the Cauchy-
Euler equation for the case where its indicial equation has equal roots is given by xr ln x.
Example. Find two linearly independent Frobenius series solutions of the differential equation
2x2 y 00 + x(2x + 1)y 0 − y = 0.
1
Ans. y1 = x(1 − 52 x + 35 x + · · · ), y2 = x− 2 (1 − x + 12 x2 + · · · ).
4 2
Example. Find the Frobenius series solutions of xy 00 + 2y 0 + xy = 0.

Solution. Rewrite the equation in the standard form x2 y 00 + 2xy 0 + x2 y = 0. We see that p(x) = 2
and q(x) = x2 . Thus p0 = 2 and q0 = 0 and the indicial equation is r(r − 1) + 2r = r(r + 1) = 0
so that the exponents of the equation are r1 = 0 and r2 = −1. In this case, r1 − r2 is an integer
and we may not have two Frobenius series solutions. We know there is a Frobenius series solution
corresponding to r1 = 0. Let’s consider the possibility of the solution corresponding to the smaller
X∞ X∞
exponent r2 = −1. Let’s begin with y = x−1 cn xn = cn xn−1 . Substituting this into the
n=0 n=0
given equation, we obtain
∞
X ∞
X ∞
X
(n − 1)(n − 2)cn xn−2 + 2 (n − 1)cn xn−2 + cn xn = 0,
n=0 n=0 n=0
or equivalently
∞
X ∞
X
n(n − 1)cn xn−2 + cn xn = 0,
n=0 n=0
or
∞
X ∞
X
n(n − 1)cn xn−2 + cn−2 xn−2 = 0.
n=0 n=2
The cases n = 0 and n = 1 reduce to 0 · c0 = 0 and 0 · c1 = 0. Thus c0 and c1 are arbitrary and
we can expect to get two linearly independent Frobenius series solutions. Equating coefficients, we
obtain the recurrence relation
cn−2
cn = − , for n ≥ 2.
n(n − 1)
It follows from this that for n ≥ 1.
(−1)n c0 (−1)n c1
c2n = and c2n+1 = .
(2n)! (2n + 1)!
Therefore, we have
∞ ∞ ∞
X c0 X (−1)n 2n c1 X (−1)n 2n+1
y = x−1 cn xn = x + x .
n=0
x n=0 (2n)! x n=0 (2n + 1)!
We recognize this general solution as

1
y= (c0 cos x + c1 sin x).
x
If we begin with the larger exponent, we will get the solution (sin x)/x.
4.9 Bessel’s equation

The second order linear homogeneous differential equation
x2 y 00 + xy 0 + (x2 − p2 )y = 0, (4.5.1)
where p is a constant is called Bessel’s equation. Its general solution is of the form
y = c1 Jp (x) + c2 Yp (x). (4.5.2)

4.9. BESSEL’S EQUATION 69
The function Jp (x) is called the Bessel function of order p of the first kind and the Yp (x) is the
Bessel function of order p of the second kind. These functions have been tabulated and behave
√
somewhat like the trigonometric functions of damped amplitude. If we let y = u/ x, we obtain
d2 u p2 − 14

+ 1− u = 0. (4.5.3)
dx2 x2
In the special case in which p = ± 21 , this equation becomes
d2 u
+ u = 0.
dx2
Hence u = c1 sin x + c2 cos x and
sin x cos x
y = c1 √ + c2 √ . (4.5.4)
x x
Also we see that as x −→ ∞ in (4.5.3), and p is finite, we would expect the solution of (4.5.1) to
behave as (4.5.4).
It is easy to see that x = 0 is a regular singular point of Bessel’s equation. Here p(x) = 1 and
q(x) = −p2 + x2 . Thus the indicial equation is r(r − 1) + r − p2 = r2 − p2 = 0. Therefore, the
∞
X
exponents are ±p. Let r be either −p or p. If we substitute y = cm xm+r into Bessel’s equation,
m=0
we find in the usual manner that c1 = 0 and that for m ≥ 2,
[(m + r)2 − p2 ]cm + cm−2 = 0 (4.5.5)
The case r = p ≥ 0. If we use r = p and write am in place of cm , then (4.5.5) yields the recursion
formula
am−2
am = − (4.5.6)
m(2p + m)
As a1 = 0, it follows that am = 0 for all odd values of m. The first few even coefficients are
a0 a0
a2 = − =− 2 ,
2(2p + 2) 2 (p + 1)
a2 a0
a4 = − = 4 ,
4(2p + 4) 2 · 2(p + 1)(p + 2)
a4 a0
a6 = − =− 6 .
6(2p + 6) 2 · 2 · 3(p + 1)(p + 2)(p + 3)
In general, one can show that
(−1)m a0
a2m = .
22m m!(p + 1)(p + 2) · · · (p + m)
Thus we have a solution associated with the larger exponent p

∞
X (−1)m
y1 = a0 2m m!(p + 1)(p + 2) · · · (p + m)
x2m+p .
m=0
2
If p = 0, this is the only Frobenius series solution. In this case, if we choose a0 = 1, we get a
solution of Bessel’s equation of order 0 given by
∞
X (−1)m x2m x2 x4 x6
J0 (x) = 2m 2
=1− + − + ··· .
m=0
2 (m!) 4 64 2304
This special function J0 (x) is called the Bessel function of order zero of the first kind. A second
linearly independent solution can be obtained by other means, but it is not a Frobenius series.
The case r = −p < 0. Our theorem does not guarantee the existence of a Frobenius solution
associated with the smaller exponent. However, as we shall see, it does have a second Frobenius
series solution so long as p is not an integer. Let’s write bm in place of cm in (4.5.5). Thus we have
b1 = 0 and for m ≥ 2,
m(m − 2p)bm + bm−2 = 0 (4.5.7)
Note that there is a potential problem if it happens that 2p is a positive integer, or equivalently if p
is a positive integer or an odd integral multiple of 12 . Suppose p = k/2 where k is an odd positive
integer. Then for m ≥ 2, (4.5.7) becomes
m(m − k)bm = −bm−2 (4.5.8)
Recall b1 = 0 so that b3 = 0, b5 = 0, · · · , bk−2 = 0 by (4.5.8). Now in order to satisfy (4.5.8) for

m = k, we can simply choose bk = 0. Subsequently all bm = 0 for all odd values of m. [If we let
bk to be arbitrary and non-zero, the subsequent solution so obtained is just bk y1 (x). Thus no new
solution arises in this situation.]
So we only have to work out bm in terms of b0 for even values of m. In view of (4.5.8), it is possible
to solve bm in terms of bm−2 since m(m − k) 6= 0 as m is always even while k is odd. The result is
the same as before except we should replace p by −p. Thus in this case, we have a second solution
∞
X (−1)m
y2 = a0 2m m!(−p + 1)(−p + 2) · · · (−p + m)
x2m−p .
m=0
2
Since p(x) = 1 and q(x) = x2 −p2 are just polynomials. The series representing y1 and y2 converge
for all x > 0. If p > 0, then the first term in y1 is a0 xp , whereas the first term in y2 is b0 x−p . Hence
y1 (0) = 0, but y1 (0) −→ ±∞ as x −→ 0, so that y1 and y2 are linearly independent. So we have
two linearly independent solutions as long as p is not an integer.
If p = n is an nonnegative integer and we take a0 = 2n1n! , the solution y1 becomes
∞
X (−1)m x 2m+n
Jn = .
m=0
m!(m + n)! 2
Jn is called the Bessel function of the first kind of integral order n.
Remarks.
1. If p is not an integer, the factorials in Jp can be replaced by the so called Gamma functions and
the general solution is Y = c1 Jp + c2 J−p . If p is an integer, (4.5.7) can still be used to get a solution
J−p , but it turns out it is just (−1)p Jp , so there is only one Frobenius series solution. A second
solution can be obtained by considering the function
Jp (x) cos pπ − J−p (x)

Yp (x) = .
sin pπ
If p is not an integer, Yp is a solution of Bessel’s equation of order p as it is a linear combination of

Jp and J−p . If p approaches to an integer, the expression of Yp gives an indeterminate form as both
the numerator and denominator approach zero. To get a second solution when p = n is an integer,
we take limit as p tends to n to get a solution Yn .
Jp (x) cos pπ − J−p (x)

Yn (x) = lim .
p→n sin pπ
Yn is called a Bessel function of the second kind, and it follows that y = c1 Jp + c2 Yp is the general
solution of Bessel’s equation in all cases, whether p is an integer or not.
2. The case r1 = r2 . Let L(y) = x2 y 00 + xp(x)y 0 + q(x)y. We are solving L(y) = 0 by taking
P∞
a series solution of the form y(x) = xr n=0 an xn . If we treat r as a variable, then an ’s are
P∞
functions of r. That is y(x, r) = xr n=0 an (r)xn . Substituting this into L(y) and requires it to be
a solution, we get (4.4.7), which can be used to determine an (r) recursively provided
(r + n)(r + n − 1) + (r + n)p0 + q0 6= 0.
When r is near the double root r1 = r2 , this expression is nonzero so that all an can be determined
from (4.4.7). This means
L(y(x, r)) = a0 (r − r1 )2 xr .
So if a0 6= 0, we take r = r1 , we get one Frobenius series solution y1 (x). Now let’s differentiate
the above equation with respect to r. We get
∂y ∂
L( )= L(y) = a0 [(r − r1 )2 xr ln x + 2(r − r1 )xr ].
∂r ∂r
Evaluating at r = r1 , we obtain

∂y ∂
L( )= L(y) = 0.
∂r r=r1 ∂r r=r1
Consequently, we have the second solution
∞ ∞ ∞
∂y X X X
y2 (x) = (x, r1 ) = xr1 ln x an (r1 )xn +xr1 a0 n (r1 )xn = y1 (x) ln x+xr1 a0 n (r1 )xn .
∂r n=0 n=0 n=1
Note that the sum in the last expression starts at n = 1 because a0 is a constant and a00 = 0.
If we apply this method to Bessel’s equation of order p = 0, we get by choosing a0 = 1 the solutions
∞
X (−1)n x 2n
y1 (x) = , and
n=0
(n!)2 2
∞
X (−1)n H(n) x 2n
y2 (x) = y1 (x) ln x − ,
n=1
(n!)2 2
Pn 1
where H(n) = k=1 k .
3. The case r1 − r2 is a nonnegative integer

Consider
x2 y 00 + xp(x)y 0 + q(x)y = 0, x > 0,
P∞ P∞
where p(x) = n=0 pn xn and q(x) = n=0 qn xn . Let r1 and r2 be the roots (exponents) with
r1 ≥ r2 of the indicial equation r(r − 1) + p0 r + q0 = 0.
P∞
Write r1 = r2 + m, where m is a nonnegative integer. Let y1 = xr1 n=0 an xn be a Frobenius
series solution corresponding to the larger exponent r1 . For simplicity, we take a0 = 1.
P∞
Let u = xr2 n=0 bn xn and make a change of variable:
y(x) = u(x) − bm y1 (x) ln x.
We get
x2 u00 + xp(x)u0 + q(x)u = bm [2xy10 + (p(x) − 1)y1 ].
P∞
Now let’s substitute u = xr2 n=0 bn xn to see if we can determine the bn ’s. Note that the first term
in the power series expansion of bm [2xy10 + (p(x) − 1)y1 ] is mbm , with m ≥ 0.
Hence after substituting the power series of u into the above equation, we have
(r2 (r2 − 1) + p0 r2 + q0 )b0 xr2 + A1 xr2 +1 + · · · + Am xr2 +m + · · · = mbm xr1 + · · · . (4.5.9)
The first term on the left hand side is 0 as r2 is a root of the indicial equation. This means b0 can
be arbitrary. The coefficients A1 , A2 , . . . are given by the main recurrence relation (4.4.7). Thus by
equating A1 , . . . , Am−1 to 0, one can determine b1 , . . . bm−1 . The next term on the left hand side of
(4.5.9) is the coefficient Am of xr1 . In the expression of Am given by (4.4.7), the coefficient of bm
is 0. Previously, this forbids the determination of bm and possibly runs into a contradiction. Now
on the right hand side of (4.5.9), if m > 0, then one can determine bm by equating the coefficients
of xr1 on both sides. From then on, all the subsequent bn ’s can be determined and we get a solution
of the form y(x) = u(x) − bm y1 (x) ln x. Note that if bm = 0 in this determination, then a second
Frobenius series solution in fact can be obtained with the smaller exponent r2 .
Example. Consider x2 y 00 + xy = 0. Here p(x) = 0, q(x) = x. The exponents are 0 and 1. Hence
m = 1. Corresponding to the exponent 1, the recurrence relation is n(n + 1)an + an−1 = 0 for
n ≥ 0.
We have the solution
∞
X (−1)n−1 n n 1 1
y1 = 2
x = x − x2 + x3 − · · · .
n=1
(n!) 2 12
Now b1 [2xy10 + (p(x) − 1)y1 ] = b1 (2x(1 − x + 14 x2 − · · · ) − (x − 12 x2 + 1 3

12 x − · · · )] =
b1 [x − 32 x2 + 12
5 3
x − · · · ].
P∞
Substituting u = x0 n=0 bn x
n
into x2 u00 + xu = b1 [2xy10 + (p(x) − 1)y1 ], we get
3 5
0·(0−1)b0 +[(1)(0)b1 +b0 ]x+[(2)(1)b2 +b1 ]x2 +[(3)(2)b3 +b2 ]x3 +· · · = b1 [x− x2 + x3 −· · · ].
2 12
Comparing coefficients, we have b0 = b1 , 2b2 +b1 = − 32 b1 and 6b3 +b2 = 12 5
b1 , · · · . Thus b1 = b0 ,
5 5 5 2 5 3
b2 = − 4 b0 , b3 = 18 b0 , . . .. Therefore u = b0 (1 + x − 4 x + 18 x − · · · ). By taking b0 = 1, we
get the solution y = (1 + x − 54 x2 + 18 5 3
x − · · · ) − y1 (x) ln x.
If m = 0, then r1 = r2 and the first terms on both sides of (4.5.9) are 0. Thus we can continue to
determine the rest of bn ’s. In this case, the ln term is definitely present.
Exercise. Find the general solution of the differential equation
x2 (1 + x2 )y 00 − x(1 + 2x + 3x2 )y 0 + (x + 5x2 )y = 0.
[Answer. y1 = x2 (1 + x + 21 x2 + · · · ), y2 = (1 + x + 2x2 + 83 x3 + · · · ) − 2y1 ln x.]

Appendix 2 Some Properties of the Legendre Polynomials
The Legendre polynomial Pn (x) is a polynomial of degree n satisfying Legendre’s equation
(1 − x2 )y 00 − 2xy 0 + n(n + 1)y = 0,
where n is a nonnegative integer. It is normalized so that the coefficient of xn is (2n)!/[2n (n!)2 ].

Explicitly it is given by
bn/2c
X (−1)k (2n − 2k)!
Pn (x) = xn−2k .
2n k!(n
− k)!(n − 2k)!
k=0
There is also a Rodrigues’ formula for the Legendre polynomial given by
1 dn 2
Pn (x) = (x − 1)n .
n!2n dxn
Note that in Rodrigues’ formula, the coefficient of xn is (2n)!/[2n (n!)2 ]. We can use Rodrigues’
formula to show that Pn (1) = 1. By this formula, we have 2n Pn (1) is the coefficient of (x − 1)n in
the Taylor polynomial expansion of (x2 − 1)n at x = 1. As (x2 − 1)n = (x − 1)n (x − 1 + 2)n =
(x − 1)n [(x − 1)n + n(x − 1)n−1 2 + · · · + 2n ], it is clear that the coefficient of (x − 1)n is 2n . Thus
Pn (1) = 1.
1
The Legendre polynomial Pn (x) has the generating function φ(Z) = (1 − 2xZ + Z 2 )− 2 = (1 +
1
Z 2 − 2xZ)− 2 . That is Pn (x) is the coefficient of Z n in the expansion of φ. To see this, let’s write
∞
X
φ(Z) = An Z n , −1 ≤ x ≤ 1 and |Z| < 1. (A.1)
n=0
Using Binomial expansion,
1 1 (− 12 )(− 12 − 1) 2
(1 + Z 2 − 2xZ)− 2 = 1 − (Z 2 − 2xZ) + (Z − 2xZ)2 + · · · ,
2 2!
it is clear that An is a polynomial of degree n. If we let x = 1, we obtain
1
φ(1) = (1 − 2Z + Z 2 )− 2 = (1 − Z)−1 = 1 + Z + Z 2 + Z 3 + · · · , |Z| < 1.
Hence An (1) = 1 for all n. Now, if we can show that An satisfies Legendre’s equation, it will be
identical with Pn (x) as the An ’s are the only polynomials of degree n that satisfy the equation and
have the value 1 when x = 1. Differentiating φ with respect to Z and x, we obtain
∂φ
(1 − 2Zx + Z 2 ) = (x − Z)φ, (A.2)
∂Z
∂φ ∂φ
Z = (x − Z) . (A.3)
∂Z ∂x
Substituting (A.1) into (A.2) and equating the coefficients of Z n−1 , we obtain
nAn − (2n − 1)xAn−1 + (n − 1)An−2 = 0 (A.4)

Also substituting (A.1) into (A.3) and equating the coefficients of Z n−1 , we obtain
dAn−1 dAn−2
x − = (n − 1)An−1 (A.5)
dx dx
In (A.5), replace n by n + 1 to get
dAn dAn−1
x − = nAn (A.6)
dx dx
Now differentiate (A.4) with respect to x and eliminate dAn−2 /dx by (A.5), we have
dAn dAn−1
−x = nAn−1 (A.7)
dx dx
We now multiply (A.6) by −x and add it to (A.7) and obtain
dAn
(1 − x2 ) = n(An−1 − xAn ) (A.8)
dx
Differentiating (A.8) with respect to x and simplifying the result by (A.6), we finally obtain
d2 An dAn
(1 − x2 ) − 2x + n(n + 1)An = 0 (A.9)
dx2 dx
This shows that An is a solution of Legendre’s equation. Using this generating function and Legen-
dre’s equation, it can be shown that Pn (x) satisfy the following orthogonal relations.
(
1
if m 6= n
Z
0
Pm (x)Pn (x) dx = 2
. (A.10)
−1 2n+1 if m = n
Chapter 5
Fundamental Theory of ODEs
5.1 Existence-Uniqueness Theorem

We consider the initial value problem
dx
= f (t, x), x(t0 ) = x0 , (5.1.1)
dt
Definition. Let G be a subset in R2 . f (t, x) : G → R is said to satisfy the Lipschitz condition with
respect to x in G if there exists a constant L > 0 such that, for any (t, x1 ), (t, x2 ) ∈ G,
|f (t, x1 ) − f (t, x2 )| ≤ L|x1 − x2 |.
L is called a Lipschitz constant.
Theorem 5.1 (Picard) Let f (t, x) be continuous on the rectangle
R : |t − t0 | ≤ a, |x − x0 | ≤ b (a, b > 0),
and let
|f (t, x)| ≤ M
for all (t, x) ∈ R. Furthermore, assume f satisfies a Lipschitz condition with constant L in R. Then
there is a unique solution to the initial value problem
dx
= f (t, x), x(t0 ) = x0
dt
on the interval I = [t0 − α, t0 + α], where α = min{a, b/M }.
Proof of the existence of solution will be given in section 5.2 and 5.3. The uniqueness of solution
will be proved in section 5.5.
2
Example 1. Let f (t, x) = x2 e−t sin t be defined on
G = {(t, x) ∈ R2 : 0 ≤ x ≤ 2}.
Let (t, x1 ), (t, x2 ) ∈ G.
77
78 CHAPTER 5. FUNDAMENTAL THEORY OF ODES
|f (t, x1 ) − f (t, x2 )|
2 2
= |x21 e−t sin t − x22 e−t sin t|
2
= |e−t sin t||x1 + x2 ||x1 − x2 |
≤ (1)(4)|x1 − x2 |
Thus we may take L = 4 and f satisfies a Lipschitz condition in G with Lipschitz constant 4.
√
Example 2. Let f (t, x) = t x be defined on
G = {(t, x) ∈ R2 : 0 ≤ t ≤ 1, 0 ≤ x ≤ 1}.
√
Consider the two points (1, x), (1, 0) ∈ G. We have |f (1, x) − f (1, 0)| = x = √1x |x − 0|.
However, as x → 0+ , √1x → +∞, so that f cannot satisfy the Lipschitz condition with any finite
constant L > 0 on G.
Proposition 5.1.1 Suppose f (t, x) has a continuous partial derivative fx (t, x) on a rectangle R =
{(t, x) ∈ R2 : a1 ≤ t ≤ a2 , b1 ≤ x ≤ b2 } in the tx-plane. Then f satisfies a Lipschitz condition on
R.
Proof. Since fx (t, x) is continuous on R, it attains its maximum value in R by the extreme value
theorem. Let K be the maximum value of |fx (t, x)| on R. By Mean Value Theorem, we have
|f (t, x1 ) − f (t, x2 )| = |fx (t, c)||x1 − x2 |,

for some c between x1 and x2 .
Therefore,
|f (t, x1 ) − f (t, x2 )| ≤ K|x1 − x2 |
for all (t, x1 ), (t, x2 ) ∈ R. Thus, f satisfies a Lipschitz condition in G with Lipschitz constant K.
Example 3. Let f (t, x) = x2 be defined on
G = {(t, x) ∈ R2 : 0 ≤ t ≤ 1}.
First
|f (t, x1 ) − f (t, x2 )| = |x21 − x22 | = |x1 + x2 ||x1 − x2 |.
Since x1 and x2 can be arbitrarily large, f cannot satisfy the Lipschitz condition on G. If we replace
G by any closed and bounded region, then f will satisfy the Lipschitz condition.
5.2 The method of successive approximations

We will give the proof of Theorem 5.1 in several steps. Let’s fix f (t, x) to be a continuous function
defined on the rectangle
R : |t − t0 | ≤ a, |x − x0 | ≤ b (a, b > 0).
The objective is to show that on some interval I containing t0 , there is a solution φ to (5.1.1). The
first step will be to show that the initial value problem (5.1.1) is equivalent to an integral equation,
namely Z t
x(t) = x0 + f (s, x(s)) ds. (5.2.1)
t0
5.2. THE METHOD OF SUCCESSIVE APPROXIMATIONS 79
By a solution of this equation on I is meant a continuous function φ on I such that (t, φ(t)) is in R
for all t ∈ I, and
Z t
φ(t) = x0 + f (s, φ(s)) ds.
t0
Theorem 5.2 A function φ is a solution of the initial value problem (5.1.1) on an interval I if and
only if it is a solution of the integral equation (5.1.2) on I.
Proof. Suppose φ is a solution of the initial value problem on I. Then
φ0 (t) = f (t, φ(t)) (5.2.2)
on I. Since φ is continuous on I, and f is continuous on R, the function f (t, φ(t)) is continuous on

I. Integrating (5.2.2) from t0 to t we obtain
Z t
φ(t) − φ(t0 ) = f (s, φ(s)) ds.
t0
Since φ(t0 ) = x0 , we see that φ is a solution of (5.2.1).

Conversely, suppose φ satisfies (5.2.1). Differentiating we find, using the fundamental theorem of
Calculus, that φ0 (t) = f (t, φ(t)) for all t ∈ I. Moreover, from (5.2.1), it is clear that φ(t0 ) = x0
and thus φ is a solution of (5.1.1).
As a first approximation to the solution of (5.2.1), we consider φ0 defined by φ0 (t) = x0 . This

function satisfies the initial condition φ0 (t0 ) = x0 , but does not in general satisfy (5.2.1). However,
if we compute
Z t Z t
φ1 (t) = x0 + f (s, φ0 (s)) ds = x0 + f (s, x0 ) ds,
t0 t0
we might expect φ1 is a closer approximation to a solution than φ0 . In fact, if we continue the

process and define successively
Z t
φ0 (t) = x0 , φk+1 (t) = x0 + f (s, φk (s)) ds, k = 0, 1, 2, . . . (5.2.3)
t0
we might expect, on taking the limit as k → ∞, that we would obtain φk (t) → φ(t), where φ would
satisfy
Z t
φ(t) = x0 + f (s, φ(s)) ds.
t0
Thus φ would be our desired solution.
We call the functions φ0 , φ1 , φ2 · · · defined by (5.2.3) successive approximations to a solution of

the integral equation (5.2.1), or the initial value problem (5.1.1).
Example. Consider the initial value problem x0 = tx, x(0) = 1.

The integral equation corresponding to this problem is
Z t
x(t) = 1 + s · x(s) ds,
0
and the successive approximations are given by

Z t
φ0 (t) = 1, φk+1 (t) = 1 + sφk (s) ds, k = 0, 1, 2, . . . .
0
Rt t2
Rt s2 t2 t4
Thus φ1 (t) = 1 + 0 s ds = 1 + 2, φ2 (t) = 1 + 0
s(1 + 2 ) ds = 1+ 2 + 2·4 , and it may be
established by induction that
2 k
t2 t2 t2

1 1
φk (t) = 1 + + + ··· + .
2 2! 2 k! 2
2
We recognize φk (x) as a partial sum for the series expansion of the function φ(t) = et /2 . We know
that this series converges for all t and this means that φk (t) → φ(t) as k → ∞, for all x ∈ R.
Indeed φ is a solution of this initial value problem.
Theorem 5.3 Suppose |f (t, x)| ≤ M for all (t, x) ∈ R. Then the successive approximations φk ,
defined by (5.2.3), exist as continuous functions on
I : |t − t0 | ≤ α = min{a, b/M },
and (t, φk (t)) is in R for t ∈ I. Indeed, the φk ’s satisfy
|φk (t) − x0 | ≤ M |t − t0 | (5.2.4)
for all t ∈ I.
Note: Since for t ∈ I, |t − t0 | ≤ b/M , the inequality (5.2.4) implies that |φk (t) − x0 | ≤ b for all
t ∈ I, which shows that the points (t, φk (t)) are in R for t ∈ I.
The geometric interpretation of the inequality (5.2.4) is that the graph of each φk lies in the region
T in R bounded by the two lines x − x0 = M (t − t0 ), x − x0 = −M (t − t0 ), and the lines
t − t0 = α, t − t0 = −α.
x0 + b
..............................................................................................................................................................................................................................................................................................................................................................................
.... ... .. .. ..
... ... ....... .. ...
... .......... .... ..
.........
. ...
... ... ........... ... . ....
.... . ... ...
... ... . .......... ... . ....
..... . . . . ... ...
... ... . . . . . .......... ... ....
.... . . . ...
. ...
... ... . . . . ......... .
. .
..
.
... . . . . ...
... ... . . . . ........... .... x − x0 = M (t − t0 ) ............ . . . . ......... ...
... ... . . . . . . . . . . . .......... ... ... ...
. . . . . . .. .
. ...
... ... . . . . . . . ......... .
. ... ......... . . . . . ..... ... ...
... ... . . . . . . . ........... ....
...
...... .
..
...... . . . . . ....... . ... ...
... ... . . . . . . . . . . . . . . . . . .......... ... ....... ........... . . . . . . . . . . . ...... . . . ... ...
... ... . . . . . . . . . . .......... ...
........ . . . . . .k .... . . ..
.
..... . . . . . . ......... . . . ... . φ . ...
... ... . . . . . . . . . . .......... ... . .... ...
... ... . . . . . . . . . . . . . . . . . . . . . . . .......... . .
. ...... . . . . . . . . . . . . ......... . . . . . . . ... ...
... T
... . . . . . . . . . . . . . ..........
.
..
. ..... ..
.... . . . . . . . ....... . . . . ...
..... . . . . . . ........... . . . . . ... ...
... ... . . . . . . . . . . . . . .......... ..
. .
..
.. ...
... ... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ........... .... ...
....... . . . . . . . . . . . . . ............ . . . . . . . . . . . . ... ...
... .. .
... ... . . . . . . . . . . . . . . . . .......... ... ............. . . . . . . . . . . ........................ . . . . . . . . . . . . . . . ... ...
... ... . . . . . . . . . . . . . . .. . ................. .................... . ........................................ . . . . . . . . . .... .... ..
t0 − a ... ... . . . . . . . . . .................... . . . . .......... .. .......... . . . . . . . . . . . . . . . . ..
•
..................................................................................................................................................................................................................................................................................................................................................................................................................................................................
...
t0 + a
t −α ... ... . . . . . . . ............. . . . . . . ......... ...... . . . . . . . . . . . . . . . ..
...... . . . . . . . . . . . . . . . .. 0 t +α ...
... 0 ... . . . . . . . . . . . . .............. . . . . . . . . . . . . .......... ...... . . . . . . . . . . . . . . ... ...
...
...
... . . . . . . ........ . . . . . . ..........
... . . . . ........... . . . . . . .........
(t , x )
0 . 0
..
.
..
........ . . . . . . . . . . . . . ..
...... . . . . . . . . . . . . .
. ...
...
... ... . . . . . . . ........ . . . . . . . . . . . . .......... ... ...... . . . . . . . . . . . . ... ...
...... . . . . . . . . . . . ..
... ... . . . ...... . . . . . . .......... ... ...... . . . . . . . . . . . .
...... . ...
... ... . . ....... . . . . . . ......... ... .
. ......... . . . . . . . . . . . . . . . . . ...
.
. ...
... ... . . . ......... . . . . . . . . . . ......... ... ...
........
.. .
...... . . . . . . . . .. ...
... ... . ...... . . . . . ........... ... . ..
....... . . . . . . . .. . ...
... ... ..... . . . . . ........ ... .... ...... . . . . . . .
...... . . . . . . ... ...
... ... ...... . . . . . . . . .......... ... ... ........ . . . . .. ...
... ... .... . . . .......... ... ...... . . . . . ...
... ....... . . . ........ ... x − x0 = −M (t − t0 ) ........... . . . .... ...
... ... . . . . .......... ... ....... . . ...
....... . . ...
... ... . .......... ... ...... . .. ...
... ... ........ ... ...... .. ...
........
... ...... ... .. ...
... ... ... ... ..
.....................................................................................................................................................................................................................................................................................................................................................................................
x0 − b
5.3. CONVERGENCE OF THE SUCCESSIVE APPROXIMATIONS 81
Proof. We prove it by induction on k. Clearly φ0 exists on I as a continuous function, and satisfies

(5.2.4) with k = 0. Now suppose the theorem has been proved for the functions φ0 , φ1 , . . . , φk ,
with k ≥ 0. We shall prove that it is valid for φk+1 . By induction hypothesis, the point (t, φk (t))
is in R for t ∈ I. Thus the function f (t, φk (t)) exists for t ∈ I and is continuous on I. Therefore,
φk+1 , which is given by
Z t
φk+1 (t) = x0 + f (s, φk (s)) ds,
t0
exists as a continuous function on I. Moreover,

Z t

|φk+1 (t) − x0 | ≤
|f (s, φk (s))| ds ≤ M |t − t0 |,
t0
which shows that φk+1 satisfies (5.2.4).
5.3 Convergence of the successive approximations

We now prove the main existence theorem
Theorem 5.4 Let f (t, x) be continuous on the rectangle
R : |t − t0 | ≤ a, |x − x0 | ≤ b (a, b > 0),
and let
|f (t, x)| ≤ M
the successive approximations
Z t
φ0 (t) = x0 , φk+1 (t) = x0 + f (s, φk (s)) ds, k = 0, 1, 2, . . .
t0
converges uniformly on the interval I = [t0 − α, t0 + α] with α = min{a, b/M }, to a solution of

dx
the initial value problem = f (t, x), x(t0 ) = x0 on I.
dt
Proof. (a) Convergence of {φk (t)}. The key to the proof is the observation that φk may be written
as
φk = φ0 + (φ1 − φ0 ) + (φ2 − φ1 ) + · · · + (φk − φk−1 ),
and hence φk (t) is a partial sum for the series
∞
X
φ0 (t) + [φp (t) − φp−1 (t)]. (5.3.1)
p=1
Therefore to show that the sequence {φk (t)} converges uniformly is equivalent to show that the
series (5.3.1) converges uniformly.
By Theorem 5.3, the functions φk all exist as continuous functions on I, and (t, φp (t)) is in R for
t ∈ I. Moreover,
|φ1 (t) − φ0 (t)| ≤ M |t − t0 |, (5.3.2)
for t ∈ I. Next consider the difference of φ2 and φ1 . We have

Z t
φ2 (t) − φ1 (t) = [f (s, φ1 (s)) − f (s, φ0 (s))] ds.
t0
Therefore Z t

|φ2 (t) − φ1 (t)| ≤
|f (s, φ1 (s)) − f (s, φ0 (s))| ds ,
t0
and since f satisfies the Lipschitz condition
|f (t, x1 ) − f (t, x2 )| ≤ L|x1 − x2 |,
we have Z t

|φ2 (t) − φ1 (t)| ≤ L
|φ1 (s) − φ0 (s)| ds .
t0
Using (5.3.2), we obtain Z t

|φ2 (t) − φ1 (t)| ≤ M L |s − t0 | ds .
t0
Thus if t ≥ t0 ,
t
M L(t − t0 )2
Z
|φ2 (t) − φ1 (t)| ≤ M L (s − t0 ) ds = .
t0 2
The same result is valid in case t ≤ t0 .
We shall prove by induction that
M Lp−1 |t − t0 |p
|φp (t) − φp−1 (t)| ≤ (5.3.3)
p!
for all t ∈ I.
We have proved this for p = 1 and p = 2. Let’s assume t ≥ t0 . The proof is similar for t ≤ t0 .
Assume (5.3.3) is true for p = m. Using the definition of φm+1 and φm , we have
Z t
φm+1 (t) − φm (t) = [f (s, φm (s)) − f (s, φm−1 (s))] ds,
t0
and thus Z t

|φm+1 (t) − φm (t)| ≤
|f (s, φm (s)) − f (s, φm−1 (s))| ds .
t0
Using the Lipschitz condition, we get

Z t

|φm+1 (t) − φm (t)| ≤ L
|φm (s) − φm−1 (s)| ds .
t0
By induction hypothesis, we obtain

t
M Lm M Lm |t − t0 |m+1
Z
|φm+1 (t) − φm (t)| ≤ |s − t0 |m ds = .
m! t0 (m + 1)!
Thus, (5.3.3) is true for all positive integer p.
Since |t − t0 | ≤ α for all t ∈ I, we can further deduce from (5.3.3) that

5.3. CONVERGENCE OF THE SUCCESSIVE APPROXIMATIONS 83
M Lp−1 αp M (Lα)p
|φp (t) − φp−1 (t)| ≤ = . (5.3.4)
p! L p!
∞
X M (Lα)p M Lα
Since the series converges to L (e − 1), we have by Weierstrass M-test that the
p=1
L p!
series
∞
X
φ0 (t) + [φp (t) − φp−1 (t)]
p=1
converges absolutely and uniformly on I. Thus the sequence of partial sum which is φk (t) converges
uniformly on I to a limit φ(t). Next we shall show that this limit φ is a solution of the integral
equation (5.2.1).
(b) Properties of the limit φ. Since each φk is continuous on I and the sequence converges uniformly
to φ, the function φ is also continuous on I. Now if t1 and t2 are in I, we have
Z t1

|φk+1 (t1 ) − φk+1 (t2 )| = f (s, φk (s)) ds ≤ M |t1 − t2 |,
t2
which implies, by letting k → ∞,
|φ(t1 ) − φ(t2 )| ≤ M |t1 − t2 |. (5.3.5)
It also follows from (5.3.5) that the function φ is continuous on I. In fact φ is uniformly continuous
on I. Letting t1 = t, t2 = t0 in (5.3.5), we see that
|φ(t) − φ(t0 )| ≤ M |t − t0 |
which implies that the points (t, φ(t)) are in R for all t ∈ I.
(c) Estimate for |φ(t) − φk (t)|. We have
∞
X
φ(t) = φ0 (t) + [φp (t) − φp−1 (t)],
p=1
and
k
X
φk (t) = φ0 (t) + [φp (t) − φp−1 (t)].
p=1
Using (5.3.4), we have

∞
X

|φ(t) − φk (t)| = [φp (t) − φp−1 (t)]
p=k+1
∞
X
≤ |φp (t) − φp−1 (t)|
p=k+1
∞
X M (Lα)p
≤
L p!
p=k+1
∞
M (Lα)k+1 X (Lα)p
≤
L (k + 1)! p=0
p!
k+1
M (Lα)
≤ eLα .
L (k + 1)!
k+1
Letting k = (Lα)
(k+1)! , we see that k → 0 as k → ∞ as k is a general term for the series e
Lα
. In
terms of k , we may rewrite the above inequality as
M Lα
|φ(t) − φk (t)| ≤ e k , and k → 0 as k → ∞ (5.3.6)
L
(d) The limit φ is a solution. To complete the proof we must show that
Z t
φ(t) = x0 + f (s, φ(s)) ds,
t0
for all t ∈ I. Note that since φ is continuous, the integrand f (s, φ(s)) of the right hand side is
continuous on I. Since Z t
φk+1 (t) = x0 + f (s, φk (s)) ds,
t0
we get the result by taking limit on both sides as k → ∞ provided we can show
Z t Z t
f (s, φk (s)) ds → f (s, φ(s)) ds, as k → ∞.
t0 t0
Z t Z t Z t

Now
f (s, φ(s)) ds − f (s, φk (s)) ds ≤
|f (s, φ(s)) − f (s, φk (s))| ds
t0 t0 tZ
0
t

≤ L |φ(s) − φk (s)| ds
t0
≤ M eLα k |t − t0 | by (5.3.6)
≤ M αeLα k → 0 as k → ∞.
This completes the proof of the Theorem 5.4.
Example. Consider the initial value problem x0 = (sin t)x2 , x(0) = 12 .

Let f (t, x) = (sin t)x2 be defined on
1 1
R = {(t, x) : |t| ≤ 1, |x − | ≤ }.
2 2
|f (t, x)| = |(sin t)x2 | ≤ 1. Thus we may take M = 1. Therefore by Theorem 5.4 a solution exists
on [−α, α] where α = min{1, 12 } = 12 . In fact x(t) = (1 + cos t)−1 is a solution defined on the
maximal domain (−π, π).
Exercise. Consider the initial value problem x0 = tx + x10 , x(0) = 1

10 . Show that a solution of this
problem exists for |t| ≤ 12 .
5.4 Non-local Existence of Solutions

Theorem 5.5 Let f (t, x) be a continuous function on the strip S = {(t, x) ∈ R2 : |t − t0 | ≤ a},
where a is a given positive number, and f satisfies the Lipschitz condition with respect to S. Then
the initial value problem
x0 (t) = f (t, x), x(t0 ) = x0 ,
where (t0 , x0 ) ∈ S has a unique solution on the entire interval [−a + t0 , a + t0 ].
5.4. NON-LOCAL EXISTENCE OF SOLUTIONS 85
Remark. If f is bounded on S, the result can be deduced from Picard’s Theorem. If f is not
necessarily bounded, the proof is slightly different.
Proof of Theorem 5.5. First note that the given region S is not bounded above or below. Hence
f (t, x) needs not be bounded in S. However, as in Theorem 5.4, we shall consider the series
∞
X
φ0 (t) + (φp (t) − φp−1 (t))
p=1
whose n-th partial sum is φn (t) and φn (t) → φ(t) giving the solution of the initial value problem.
Since f (t, x) is not bounded in S, we adopt a different method of estimating different terms of the
series. Let M0 = |x0 | and M1 = max |φ1 (t)|. The fact that M1 exists can be seen as follows. Since
f (t, x) is continuous in S, for a fixed x0 , f (t, x0 ) is a continuous function on |t − t0 | ≤ a.
Rt
Thus φ1 (t) = x0 + t0 f (s, x0 ) ds is a continuous function in this interval so that |φ1 (t)| attains its
maximum in this interval. We take it to be M1 and let M = M0 + M1 .
Thus, |φ0 (t)| = |x0 | ≤ M and |φ1 (t) − φ0 (t)| ≤ M . If t0 ≤ t ≤ t0 + a, then we have
Z t Z t

|φ2 (t) − φ1 (t)| = [f (s, φ1 (s)) − f (s, φ0 (s))] ds ≤
|f (s, φ1 (s)) − f (s, φ0 (s))| ds
t0 Z t0
t
≤L |φ1 (s) − φ0 (s)| ds ≤ LM (t − t0 ), where L is the Lipschitz constant.
t0
Now
Z t Z t

|φ3 (t) − φ2 (t)| = [f (s, φ2 (s)) − f (s, φ1 (s))] ds ≤
|f (s, φ2 (s)) − f (s, φ1 (s))| ds
t0 Z t0
t t
L2 M
Z
≤L |φ2 (s) − φ1 (s)| ds ≤ L2 M |(s − t0 )| ds = (t − t0 )2 .
t0 t0 2
Hence, in general, we can prove by induction that

Ln−1 M (t − t0 )n−1
|φn (t) − φn−1 (t)| ≤ .
(n − 1)!
Similar argument is true for the interval t0 − a ≤ t ≤ t0 . Hence for every t with |t − t0 | ≤ a,
Ln−1 M (t − t0 )n−1 Ln−1 M n−1
|φn (t) − φn−1 (t)| ≤ ≤ a .
(n − 1)! (n − 1)!
Thus
∞ ∞
X X (La)n−1
|φ0 (t)| + |φn (t) − φn−1 (t)| ≤ M .
n=1 n=1
(n − 1)!
Hence each term on the left hand side of the above equation is less than the corresponding term
of the convergent series of positive constants. Hence, by Weierstrass M -test, the series on the left
converges uniformly on the whole interval |t − t0 | ≤ a and let’s denote its limit by φ(t).
Next we show that φ(t) is a solution of the initial value problem. We need to show that φ(t) satisfies
the integral equation Z t
φ(t) − x0 − f (s, φ(s)) ds = 0. (5.4.1)
t0
We know that Z t
φn (t) − x0 − f (s, φn−1 (s)) ds = 0. (5.4.2)
t0
Substituting the value of x0 in (5.4.2) into the left hand side of (5.4.1), we get
Z t Z t
φ(t) − x0 − f (s, φ(s)) ds = φ(t) − φn (t) − f (s, φ(s)) − f (s, φn−1 (s)) ds.
t0 t0
Thus we obtain
Rt Rt
φ(t) − x0 − t0 f (s, φ(s)) ds≤|φ(t) − φn (t)| + t0 |f (s, φ(s)) − f (s, φn−1 (s))| ds

Rt
≤|φ(t) − φn (t)| + L t0 |φ(s) − φn−1 (s)| ds (5.4.3)
Since φn (t) → φ(t) uniformly for t ∈ [t0 − a, t0 + a], the right hand side of (5.4.3) tends to zero as
n → ∞. Hence
Z t
φ(t) − x0 − f (s, φ(s)) ds = 0. (5.2.4)
t0
The uniqueness of solution will be proved in section 5.5.
Corollary 5.6 Let f (t, x) be a continuous function defined on R2 . Suppose that for any a > 0, f
satisfies the Lipschitz condition with respect to S = {(t, x) ∈ R2 : |t| ≤ a} with (t0 , x0 ) ∈ S.
Then the initial value problem
x0 (t) = f (t, x), x(t0 ) = x0
has a unique solution on the entire R.
Proof. If t is any real number, there is an a > 0 such that t is contained in [t0 − a, t0 + a]. For this
a, the function f satisfies the condition of Theorem 5.5 on the strip
{(t, x) ∈ R2 : |t − t0 | ≤ a}.
Since this strip is contained in the strip
{(t, x) ∈ R2 : |t| ≤ a + |t0 |}.
Thus there is a unique solution φ(t) to the initial value problem for all t ∈ R.
Example. Consider the initial value problem x0 = sin(tx), x(0) = 1.
Let f (t, x) = sin(tx). Let a > 0. Using the mean value theorem, we have for any t ∈ [−a, a],
|f (t, x1 ) − f (t, x2 )| = | sin(tx1 ) − sin(tx2 )| = |t cos(tζ)(x1 − x2 )| ≤ |t||x1 − x2 | ≤ a|x1 − x2 |.
Thus f satisfies a Lipschitz condition on the strip S = {t ∈ R | |t| ≤ a}, and there exists a solution
on the entire R.
x3 et
Exercise. Show that the initial value problem x0 = + t2 cos x, x(0) = 1 has a solution on
1 + x2
R.
5.5. GRONWALL’S INEQUALITY AND UNIQUENESS OF SOLUTION 87
5.5 Gronwall’s Inequality and Uniqueness of Solution

Theorem 5.7 Let f , g, and h be continuous nonnegative functions defined for t ≥ t0 . If
Z t
f (t) ≤ h(t) + g(s)f (s) ds, t ≥ t0 ,
t0
then Z t Rt
g(u) du
f (t) ≤ h(t) + g(s)h(s)e s ds, t ≥ t0 .
t0
Proof. First we are given Z t

f (t) ≤ h(t) + g(s)f (s) ds (5.5.1)
t0
Rt
Let z(t) = t0
g(s)f (s) ds. Then for t ≥ t0 ,
z 0 (t) = g(t)f (t) (5.5.2)
Since g(t) ≥ 0, multiplying both sides of (5.5.1) by g(t) and using (5.5.2), we get
z 0 (t) ≤ g(t)[h(t) + z(t)]
which gives
z 0 (t) − g(t)z(t) ≤ g(t)h(t).
This is a first order differential inequality which can be solved by finding an integrating factor
− t g(u) du
R
e t0 . Hence the solution is
Z t
− t g(u) du − s g(u) du
R R
z(t)e t0 ≤ g(s)h(s)e t0 ds
t0
Or equivalently,
Z t Rs Rt Z t Rt
− g(u) du g(u) du g(u) du
z(t) ≤ g(s)h(s)e t0
e t0
ds = g(s)h(s)e s ds (5.5.3)
t0 t0
Substituting for z(t) in (5.5.3), we get

Z t Z t Rt
g(s)f (s) ds ≤ g(s)h(s)e s g(u) du ds (5.5.4)
t0 t0
From (5.5.1), we can replace the left side of (5.5.4) by the lesser inequality to obtain
Z t Rt
f (t) − h(t) ≤ g(s)h(s)e s g(u) du ds.
t0
Theorem 5.8 (Gronwall’s Inequality) Let f and g be continuous nonnegative functions for t ≥ t0 .
Let k be any nonnegative constant. If
Z t
f (t) ≤ k + g(s)f (s) ds, for t ≥ t0 ,
t0
then Rt
g(s) ds
f (t) ≤ ke t0
, for t ≥ t0 .
Corollary 5.9 Let f be a continuous nonnegative function for t ≥ t0 and k a nonnegative constant.
If Z t
f (t) ≤ k f (s) ds
t0
for all t ≥ t0 , then f (t) ≡ 0 for all t ≥ t0 .
Proof. For any > 0, we can rewrite the given hypothesis as

Z t
f (t) ≤ + k f (s) ds,
t0
for all t ≥ t0 . Hence applying Gronwall’s inequality, we have

Rt
k ds
f (t) ≤ e t0
,
for all t ≥ t0 , which gives f (t) ≤ ek(t−t0 ) , for all t ≥ t0 . Since is arbitrary, we get f (t) ≡ 0 by
taking limit as → 0+ .
Remark. Similar results hold for t ≤ t0 when we all the integrals are integrated from t to t0 . For
example, in Corollary 5.9, if Z t0
f (t) ≤ k f (s) ds
t
for all t ≤ t0 , then f (t) ≡ 0 for all t ≤ t0 .
Corollary 5.10 Let f (t, x) be a continuous function which satisfies a Lipschitz condition on R with
a Lipschitz constant L, where R is either a rectangle or a strip. If φ and ϕ are two solutions of
x0 = f (t, x), x(t0 ) = x0 ,
on an interval I containing t0 , then φ(t) = ϕ(t) for all t ∈ I.
Proof. Let I = [t0 − α, t0 + α]. For t ∈ [t0 , t0 + α], we have

Z t
φ(t) = x0 + f (s, φ(s)) ds,
t0
and Z t
ϕ(t) = x0 + f (s, ϕ(s)) ds.
t0
Thus Z t Z t
|φ(t) − ϕ(t)| ≤ |f (s, φ(s)) − f (s, ϕ(s))| ds ≤ L |φ(s) − ϕ(s)| ds.
t0 t0
By Corollary 5.9, |φ(t) − ϕ(t)| ≡ 0 for t ∈ [t0 , t0 + α]. Thus φ(t) = ϕ(t) for t ∈ [t0 , t0 + α].
Similarly, φ(t) = ϕ(t) for t ∈ [t0 − α, t0 ].
Remark. If we only assume that f (t, x) is a continuous function, we can still show that (5.1.1) has
at least one solution, but the solution may not be unique.
5.5. GRONWALL’S INEQUALITY AND UNIQUENESS OF SOLUTION 89
Theorem 5.11 (Peano) Assume G is an open subset of R2 containing (t0 , x0 ) and f (t, x) is con-
tinuous in G. Then there exists a > 0 such that (5.1.1) has at least one solution on the interval
[t0 − a, t0 + a].
Example. Consider the initial value problem x0 = x2/3 , x(0) = 0.

1 3
We find that x(t) = 0 and x(t) = 27 t are both solutions.
Example. Suppose φ(t) is a solution to the initial value problem
x3 − x 1
x0 = , φ(0) = .
1 + t2 x2 2
Show that 0 < φ(t) < 1 for all t for which φ(t) is defined.
Solution. Let φ(t) be a solution defined on a domain J to the initial value problem. Suppose there
exists s ∈ J such that φ(s) ≥ 1. Since φ(t) is continuous and φ(0) = 1/2, we have by Intermediate
value theorem, that φ(s0 ) = 1 for some s0 ∈ (0, s). We may take s0 to be the least value in (0, s)
such that φ(s0 ) = 1. In other words, φ(t) < 1 for all t ∈ (0, s0 ) and φ(s0 ) = 1.
Now consider the initial value problem
x3 − x
x0 = , x(s0 ) = 1.
1 + t2 x2
3
x −x
The function f (t, x) = 1+t 2 x2 satisfies the conditions of the Existence and Uniqueness Theorem.
Thus there is a unique solution defined on an interval I = [s0 − α, s0 + α] for some α > 0. The
above function φ(t) defined on J is a solution to this initial value problem, and it has the property
that φ(t) < 1 for all t < s0 . However, ϕ(t) ≡ 1 is clearly a solution to this initial value problem on
I. But ϕ and φ are different solutions to the initial value problem contradicting the uniqueness of
the solution. Consequently, φ(t) < 1 for all t ∈ J. Similarly, φ(t) > 0 for all t ∈ J.
Corollary 5.12 Let f (t, x) be a continuous function which is defined either on a strip
R = {(t, x) ∈ R2 | |t − t0 | ≤ a},
or a rectangle
R = {(t, x) ∈ R2 | |t − t0 | ≤ a, |x − x0 | ≤ b}.
Assume f satisfies a Lipschitz condition on R with a Lipschitz constant L. Let φ and ϕ be solutions
defined on I = [−a + t0 , t0 + a] of x0 = f (t, x) satisfying the initial condition x(t0 ) = x0 and
x(t0 ) = x1 respectively on I, then
|φ(t) − ϕ(t)| ≤ |x0 − x1 |eL|t−t0 |
for all t ∈ I.
Remark. In particular
|φ(t) − ϕ(t)| ≤ |x0 − x1 |eLa ,
for all t ∈ I. Thus if the initial values x0 and x1 are close, the resulting solutions φ and ϕ are also
close.
The proof of this corollary is by Gronwall’s lemma and is left as an exercise.

5.6 Existence and Uniqueness of Solutions to Systems

Consider a system of differential equations


 x01 = f1 (t, x1 , · · · , xn ),


 x0 = f2 (t, x1 , · · · , xn ),

2


 ···············

 x0 = f (t, x , · · · , x ),

n n 1 n
dxj
where x0j = dt . Let us introduce notations
     
x1 x01 f1 (t, x)
x = · · · , x0 = · · · , f (t, x) =  · · ·  .
     
xn x0n fn (t, x)
Then the system can be written in a vector form:
x0 = f (t, x). (5.6.1)
Differential equations of higher order can be reduced to equivalent systems. Let us consider
dn y dn−1 y
n
+ F (t, y, y 0 , · · · , n−1 ) = 0. (5.6.2)
dt dt
Let
dy dn−1 y
x1 = y, x2 = , · · · , xn = n−1 .
dt dt
Then (5.6.2) is equivalent to the following system


 x01 = x2 ,


 x0 = x3 ,

2
 ·········



 0
 xn = −F (t, x1 , x2 , · · · , xn ).
It can be written in the form of (5.6.1) if we let

 
  x2 ,
x1  x3 , 
x = · · · , f (t, x) =  .
   
 ··· 
xn
−F (t, x1 , · · · , xn )
 
x1
 x2 
 
 .. , its magnitude |x| is defined to be
Recall that for a vector x =  
 . 
xn
|x| = |x1 | + |x2 | + · · · + |xn |.

5.6. EXISTENCE AND UNIQUENESS OF SOLUTIONS TO SYSTEMS 91
The following 2 properties of the magnitude can be proved easily.

1. Triangle Inequality |x + y| ≤ |x| + |y|.
Pn
2. If A = (aij ) is an n × n matrix and x ∈ Rn , then |Ax| ≤ |A||x| where |A| = i,j |aij |.
Definition. Let G be a subset in R1+n . f (t, x) : G → Rn is said to satisfy the Lipschitz condition
with respect to x in G if there exists a constant L > 0 such that, for all (t, x), (t, y) ∈ G,
|f (t, x) − f (t, y)| ≤ L|x − y|.
Example. Let f : R1+2 −→ R2 be given by

! !
2x2 cos t x1
f (t, x) = , where x = .
x1 sin t x2
! !
2x cos t 2y2 cos t
2
Then |f (t, x) − f (t, y)| = −

x1 sin t y1 sin t
= |2 cos t(x2 − y2 )| + | sin t(x1 − y1 )|
≤2|x2 − y2 | + |x1 − y1 |
≤2(|x2 − y2 | + |x1 − y1 |)
= 2|x − y|.
Thus f satisfies the Lipschitz condition with respect to x in R3 with Lipschitz constant 2.
Theorem 5.13 Suppose f is defined on a set G ⊂ R1+n of the form
|t − t0 | ≤ a, |x − x0 | ≤ b, (a, b > 0)
or of the form
|t − t0 | ≤ a, |x| < ∞, (a > 0).
If ∂f /∂xk (k = 1, . . . , n) exists, is continuous on G, and there is a constant L > 0 such that

∂f
∂xk ≤ L, (k = 1, . . . , n),

for all (t, x) ∈ G, then f satisfies a Lipschitz condition on G with Lipschitz constant L.
 
f1 (t, x)
 f2 (t, x) 
 
1+n
Proof. Let f (t, x) =  . 

 , where each fi (t, x) : R −→ R.
 .. 
fn (t, x)
Thus
∂f1
 
∂x
 ∂f2k 
∂f  ∂xk 
 ..  .
= 
∂xk  . 
∂fn
∂xk
Let (t, x), (t, y) ∈ G ⊆ R1+n . Define F : [0, 1] −→ Rn by
F(s) = f (t, sx + (1 − s)y) = f (t, y + s(x − y)).
The point sx + (1 − s)y lies on the segment joining x and y, hence the point (t, sx + (1 − s)y) is
in G.
 ∂f1 
k ∂x
n n  ∂f2 
X ∂f dxk
X
 ∂x. k  (xk − yk ).
Now F0 (s) =
 
=
∂xk ds  .. 
 
k=1 k=1
∂fn
∂xk
Therefore,
n n
0
X ∂f X
|F (s)| ≤ ∂xk |xk − yk | ≤ L
|xk − yk | = L|x − y|,
k=1 k=1
for s ∈ [0, 1].

Since Z 1
f (t, x) − f (t, y) = F(1) − F(0) = F0 (s) ds
0
we have |f (t, x) − f (t, y)| ≤ L|x − y|.
Theorem 5.14 (Picard) Let f (t, x) be continuous on the set
R : |t − t0 | ≤ a, |x − x0 | ≤ b (a, b > 0),
and let
|f (t, x)| ≤ M
there is a unique solution to the initial value problem
dx
= f (t, x), x(t0 ) = x0
dt
on the interval I = [t0 − α, t0 + α], where α = min{a, b/M }.
Theorem 5.15 Let f (t, x) be a continuous function on the strip S = {(t, x) ∈ Rn+1 : |t − t0 | ≤ a},
where a is a given positive number, and f satisfies the Lipschitz condition with respect to S. Then
the initial value problem
x0 (t) = f (t, x), x(t0 ) = x0 ,
where (t0 , x0 ) ∈ S has a unique solution on the entire interval [−a + t0 , a + t0 ].
Corollary 5.16 Let f (t, x) be a continuous function defined on Rn+1 . Suppose that for any a > 0,
f satisfies the Lipschitz condition with respect to S = {(t, x) ∈ Rn+1 : |t| ≤ a} with (t0 , x0 ) ∈ S.
Then the initial value problem
x0 (t) = f (t, x), x(t0 ) = x0
has a unique solution on the entire R.

5.6. EXISTENCE AND UNIQUENESS OF SOLUTIONS TO SYSTEMS 93
The proofs carry over directly from those for Theorem 5.1 and 5.5 and Corollary 5.6 using the
method of successive approximations. That is the successive approximations
Z t
φ0 (t) = x0 , φk+1 (t) = x0 + f (s, φk (s)) ds, k = 0, 1, 2, . . .
t0
converge uniformly on the interval I = [t0 − α, t0 + α] with α = min{a, b/M }, to a solution of the
dx
initial value problem = f (t, x), x(t0 ) = x0 on I.
dt
Example. Find the first 5 successive approximations to the initial value problem
x00 = −et x, x(0) = 1, x0 (0) = 0.
The initial value problem is equivalent to the following initial value problem of differential system.
!0 ! ! !
x(t) y(t) x(0) 1
= , = .
y(t) −et x(t) y(0) 0
We start with ! !
x0 (t) 1
= , for all t ∈ R.
y0 (t) 0
Then ! ! Z ! !
t
x1 (t) 1 0 1
= + ds = .
y1 (t) 0 0 −es × 1 1 − et
! ! Z ! !
t
x2 (t) 1 1 − es 2 + t − e−t
= + ds = .
y2 (t) 0 0 −es 1 − et
! ! Z ! !
t
x3 (t) 1 1 − es 2 + t − e−t
= + ds = 1 1 2t
.
y3 (t) 0 0 −es (2 + s − es ) t t
2 − e − te + 2 e
! ! ! !
t 1
− es − ses + 12 e2s 3
+ 2t − tet + 14 e2t
Z
x4 (t) 1 2 4
= + ds = 1
.
y4 (t) 0 0 −es (2 + s − es ) 2 − et − tet + 12 e2t
Example. Consider the linear differential system x0 = Ax, where A = (aij ) is an n × n constant
matrix. Let f (t, x) = Ax. For any a > 0 and for all |t| < a, we have |f (t, x1 ) − f (t, x2 )| =
Pn Pn
|A(x1 − x2 )| ≤ |A||x1 − x2 |, where |A| = i=1 j=1 |aij |, so that f satisfies the Lipschitz
condition on the strip S = {(t, x) ∈ Rn+1 : |t| ≤ a}. Therefore the system has a unique solution
for any initial value and is defined on the entire R.
Example. Let x0 = A(t)x, where A(t) = (aij (t)) is an n × n matrix of continuous functions
defined on a closed interval I. Let |aij (t)| ≤ K for all t ∈ I and all i, j = 1, . . . n.
Thus if f (t, x) = A(t)x, then
 
a1k (t)
 a2k (t) 
 
∂f
= . 

,
∂xk  .. 
ank (t)
which is independent of x.
Therefore,
n
∂f X
=
∂xk |aik (t)| ≤ nK ≡ L, for all t ∈ I and k = 1, . . . , n.
i=1
By Theorem 5.13 the function f satisfies a Lipschitz condition on the strip
S = {(t, x) ∈ R1+n | t ∈ I}
with Lipschitz constant L. Thus by Theorem 5.15, the system x0 = A(t)x has a unique solution for
any initial value in S and is defined on all of I.
Bibliography
[1] Ravi P. Agarwal and Ramesh C. Gupta, Essentials of ordianry differential equations, McGraw-
Hill (1991)
[2] Earl A. Coddington, An introduction to ordinary differential equations, Dover (1961)
[3] George F. Simmons, Differential equations with applications and historical notes, 2nd edition,
McGraw-Hill (1991)
95

Ordinary Differential Equations Notes

Enviado por

Dados do documento

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

Ordinary Differential Equations Notes

Enviado por

Direitos autorais:

Formatos disponíveis

MA3220 Ordinary Differential Equations

Wong Yan Loi

November 20, 2007

1 First Order Differential Equations 5

2 Higher Order Linear Equations 17

3 Linear Differential Systems 35

4 Power Series Solutions 57

5 Fundamental Theory of ODEs 77

5.5 Gronwall’s Inequality and Uniqueness of Solution . . . . . . . . . . . . . . . . . . 87

First Order Differential Equations

F (x, y, y 0 , · · · , y (n) ) = 0. (1.1.1)

where f is a known function of x, y, y 0 , · · · , y (n−1) .

a0 (x)y (n) + a1 (x)y (n−1) + · · · + an (x)y = r(x). (1.1.3)

F (x, φ(x), φ0 (x)) = 0 for all x in J. (1.1.5)

x = x(t), y = y(t) (1.1.6)

is said to be a parametric solution of (1.1.4) if

Example. Consider the ODE x + yy 0 = 0 for x ∈ (−1, 1).

4. Elimination of constants: formation of ODE.

Example. Solve y 0 = −2xy, y(0) = 1.

The equation y 0 = f ( xy ) can be reduced to a separable equation by letting u = y

where f is homogeneous of degree 0, we use the substitution y = zx. Then

and integrating both sides will give the solution.

Example. An equation in the form

can be reduced to a homogeneous equation by a suitable substitution x = z + h, y = w + k when

Example. Solve y 0 = x+y−2

Example. Solve (x + y + 1) + (2x + 2y + 1)y 0 = 0.

1.2 Exact Equations, Integrating Factors

M (x, y)dx + N (x, y)dy = 0. (1.2.1)

(1.2.1) is called exact if there exists a function u(x, y) such that

and c is an arbitrary constant.

µ(x, y)M (x, y)dx + µ(x, y)N (x, y)dy = 0 (1.2.4)

is an integrating factor of (1.2.1).

Example Solve (x2 y + y + 1) + x(1 + x2 )y 0 = 0.

Example Solve (xy 3 + 2x2 y 2 − y 2 ) + (x2 y 2 + 2x3 y − 2x2 )y 0 = 0

3. Find integrating factors by inspection.

d(xy) = xdy + ydx

d(x2 + y 2 ) = 2xdx + 2ydy

1.3 First Order Linear Equations

y(x) = ce−P (x) , where

y 0 + p(x)y = q(x), (1.3.3)

The general solution is given by

Example. Solve y 0 − y = e2x .

3. The Bernoulli equation.

u0 + (1 − n)p(x)u = (1 − n)q(x). (1.3.6)

This is a first order linear ODE.

4. The Riccati equation.

where c is an arbitrary constant. Then the general solution is given by

y00 + u0 = P + Q(y0 + u) + R(y0 + u)2 .

Since y0 satisfies (1.3.7), we have

y00 = P + Qy0 + Ry02 .

From these two equalities we get

u0 = (Q + 2Ry0 )u + Ru2 . (1.3.10)

This is a Bernoulli equation with n = 2. Set Z = u−1 and reduce (1.3.10) to

Z 0 + (Q + 2Ry0 )Z = −R. (1.3.11)

(1.3.11) is a linear equation and the solution is given by (1.3.8).

Example. Solve y 0 = y/x + x3 y 2 − x5 . Note y1 = x is a solution.

F (x) = y0 (x)f (x), G(x) = y0 g(x) + 1.

Computations show that

(y1 − y2 )(y3 − y4 ) (c1 − c2 )(c3 − c4 )

where c is an arbitrary constant.

y 0 − yj0 = (y − yj )[Q + R(y + yj )],

Example. Solve y 0 = e−x y 2 . Note y1 = ex and y2 = 0 are 2 solutions.

1.4 First Order Implicit Equations

which are not solvable in y 0 .