Theory of ODE - Hu

Theory of
Ordinary Differential Equations

Existence, Uniqueness and Stability
Jishan Hu and Wei-Ping Li
Department of Mathematics
The Hong Kong University of Science and Technology
ii
c
Copyright 2005
by Department of Mathematics, HKUST
Contents
Existence and Uniqueness

1.1 Some Basics
1.2 Uniqueness Theorem
1.3 Continuity
1.4 Existence Theorem
1.5 Local Existence Theorem and The Peano Theorem
1.5.1 Local Existence Theorem
1.5.2 The Peasno Theorem
1.6 Linear Systems
1.7 Continuation of Solutions
1.8 Miscellaneous Problems
1
1
6
8
11
18
18
19
22
25
27
Plane Autonomous Systems

2.1 Plane Autonomous Systems
2.2 Linear Autonomous Systems
2.3 Complete Classification for Linear Autonomous Systems
2.4 Liapunov Direct Method
2.5 A Test For Instability
2.6 Nonlinear Oscillations
2.6.1 Undamped Oscillations
31
31
37
41
48
56
60
60
iii
iv
CONTENTS
2.7
2.6.2 Damped Oscillations

Miscellaneous Problems
61
62
1
Existence and Uniqueness
1.1
SOME BASICS
A normal system of first order ordinary differential equations (ODEs) is
dx1
dt
dxn
dt
X1 (x1 , . . ., xn; t),
=
..
.
(1.1)
Xn (x1 , . . ., xn; t).
Many varieties of ODEs can be reduced to this form. For example, consider an n-th order ODE
dn x
=F
dtn
Let x1 = x, x2 =

dx
dn1x
t, x, , . . . , n1 .
dt
dt
dx
dn1x
, . . ., xn = n1 . Then the ODE can be changed to
dt
dt
dx
1
dt
dxn1
dt
dxn
dt
x2 ,
..
.
=
xn ,
F (t, x1, . . . , xn).

1
1 EXISTENCE AND UNIQUENESS
Let us review some notations and facts on vectors and vector-valued functions. The normal system (1.1) can be written as its vector form:
dx
= X(x, t).
dt
p
For a vector x = (x1 , . . ., xn), define kxk = x21 + + x2n. The inner product is defined to be
x y = x1 y1 + + xnyn where y = (y1 , . . . , yn). We have the triangle inequality
kx + yk kxk + kyk,
and the Schwarz inequality
|x y| kxk kyk.
For a vector valued function x(t) = (x1 (t), . . . , xn(t)), we have
x0(t) = (x01 (t), . . . , x0n(t)) ,
and
Z
x(t)dt =
a
x1(t)dt, . . .,
xn(t)dt .
a
We have the following useful inequality

Z
Z
b

b

x(t)dt
kx(t)k dt.

a

a

A vector field X(x) = X1 (x1, . . . , xn), . . . , Xn(x1 , . . ., xn) is aid to be continuous if each Xi is a
continuous function of x1, . . . , xn.
Since only a few simple types of differential equations can be solved explicitly in terms of known
elementary function, in this chapter, we are going to explore the conditions on the function X such
that the differential system has a solution. We also study whether the solution is unique, subject
some additional initial conditions.
Example 1.1.1 Show that the differential equation
dy
1
= y1
dt
2
does not have solution satisfying y(0) = 0 for t > 0.
Solution Acturally, the general solution of this differential equation is
y2 = t + C,
where C is an arbitrary constant. The initial condition implies C = 0. Thus, we have y2 = t, that shows
there exists no solution for t > 0.
2
1.1 SOME BASICS
Example 1.1.2 Show that the differential equation x0 = x2/3 has infinitely many solutions satisfying
x(0) = 0 on every interval [0, b].
Solution Define
if 0 t < c;
0,
3
xc (t) =
(t c) , if c t b.
27
It is easy to check for any c, the function xc satisfies the differential equation and xc(0) = 0.
Definition 1.1.1 A vector-valued function X(x, t) is said to satisfy a Lipschitz condition in a region
R in (x, t)-space if, for some constant L (called the Lipschitz constant), we have
kX(x, t) X(y, t)k Lkx yk,
(1.2)
whenever (x, t) R and (y, t) R.1

Lemma 1.1.2 If X(x, t) has continuous partial derivatives on a bounded closed convex domain D,
then it satisfies a Lipschitz condition in R.2
Proof Denote
M =
sup
|Xi /xj | .
xR
1i,jn
Since X has continuous partial derivatives on the bounded closed region R, we know that M is finite. For
each component Xi we have, for fixed x, y, t,
n
X
Xi
d
(x + sy, t)yk .
Xi (x + sy, t) =
ds
xk
k=1
By integrating both sides from 0 to 1 and using the mean value theorem, we have
Xi(x + y, t) Xi (x, t) =
n
X
Xi
k=1
1 It
xk
(x + i y, t)yk ,
is worth to point out the following:

(i) The left hand side of the inequality (1.2) is evaluated at (x, t) and (y, t), with the same t for these two points.
(ii) The constant L is independent of x, y and t. However, it depends on R. In other words, for a given function
X(x, t), its Lipschitz constant may change if the domain R is different. In fact, for the same function X(x, t),
it can be a Lipschitz function in some regions, but not a Lipschitz function in some other regions.
2 Here
are some explanation of some concepts and terminologies in the Lemma.
(i) A bounded closed domain is also called compact. It has the following property that any continuous function
on a compact set has an absolute maximum and absolute minimum.

Xi
(ii) That X(x, t) = X1 (x1 , . . . , xn , t), . . . , Xn (x1 , . . . , xn , t) has continuous partial derivatives means that
xj
Xi
and
are continuous for all i, j. Sometimes we use X(x, t) C 1 (D) for this.
t
(iii) A convex domain D means that for any two points x and y in D, (1 t)x + ty D for 0 t 1.
for some i between 0 and 1. The assumption that R is convex implies that (x + i y, t) R. The Schwarz
inequality gives

!1/2
2 !1/2 X
n
n
n

X
X

Xi
Xi

2

(x + i y, t)yk
|yk |

xk (x + i y, t)

xk
k=1
k=1
k=1
!1/2
n
X
M2
kyk = nM kyk.
k=1
Thus,
kX(x + y, t) X(x, t)k
n
X
n
X
2
nM kyk
|Xi (x + y, t) Xi (x, t)|2
i=1
!1/2
!1/2
= nM kyk.
i=1
The Lipschitz condition follows, with the Lipschitz constant nM .
Example 1.1.3 Determine whether the function

X(x, t) =
x2 + 1
t
x
satisfies a Lipschitz condition in the domains: (1) R1 = [1, 2] [0, 1]; (2) R2 = (1, 2) [0, 1]; (3)
R3 = [1, 2] [0, +); (4) R4 = [1, +) [0, T ]; (5) R5 = (0, 1) [0, 1].
Solution (1) Since the function X(x, t) is continuously differentiable in the bounded closed convex domain
R = [1, 2] [0, 1], by Lemma 1.1.2, we know that the function satisfies a Lipschitz conditio in this region.
(2) Since the function X(x, t) satisfis a Lipschitz condition in R1 , and R2 R1 , we know that the function
satisfis the same Lipschitz inequality in R2 . Hence the function X(x, t) satisfis a Lipschitz condition in R2 .
(3) Since R3 = [1, 2] [0, +) is not a bounded region, we cannot apply Lemma 1.1.2 in this case. Since

|X(x, t) X(y, t)| xy + 1
=
|t| > |t| ,
|x y|
xy
as t +, there exists no contant L, independent of x, y and t, such that
|X(x, t) X(y, t)| L |x y| .
Hence, the function X is not a Lipschitz function in R3 .
(4) Again, R4 = [1, +) [0, T ] is not bounded and Lemma 1.1.2 does not apply in this case. For
(x, t), (y, t) R4 ,

xy + 1
|t| |x y|
|X(x, t) X(y, t)| =
xy

1
T |x y|
1+
xy
2T |x y|.
Thus, the function X is a Lipschitz function in R4 , with L = 2T .
1.1 SOME BASICS
(5) Since R5 = (0, 1) [0, 1] is not a closed region, Lemma 1.1.2 does not apply in this case. In fact, the
function X is continuously differentiable in R, with
X 0 (x) =
x2 1
.
x2
But the derivative is not bounded, since X 0 (x) as x 0+.

Actually, we can show that the function X does not satisfy any Lipschitz condition in R5 . To see that,
1
2
we take x = , y =
and t = 1, and have
n
n
|X(x, t) X(y, t)| =
n
1
,
2
n
n 1
1
for n 2. Obviously, there exists no constant L such that L = L|x y|. Therefore, the function
2 n
n
2
X does not satisfy any Lipschitz conditions in the domain R5 .
d3x
Example 1.1.4 Reduce the ODE 3 + x2 = 1 to an equivalent first-order system and determine
dt
in which domain or domains the resulting system satisfies a Lipschitz condition.
Solution Let x1 = x, x2 = x0 , x3 = x00 , we get the equivalent first order system
dx1
dt
dx2
dt
dx3
dt
x2 ,
x3 ,
1 x21 .
Denote x = (x1 , x2 , x3 ), y = (y1 , y2 , y3 ) and X(x, t) = (x2 , x3 , 1 x21 ). Then, X(x, t) X(y, t) = (x2
y2 , x3 y3 , x21 + y12 ), which implies
p
kX(x, t) X(y, t)k = (x2 y2 )2 + (x3 y3 )2 + (x1 y1 )2 (x1 + y1 )2 .
We can see that whenever x1 and y1 are bounded, to be precise, x1 + y1 is bounded, the Lipschitz condition
holds. In fact, consider the domain

o
n

R = (x1 , x2 , x3 , t) |x1 | M
p
max{4M 2 , 1}, then whenever (x, t), (y, t) R, we have

p
(x2 y2 )2 + (x3 y3 )2 + 4M 2 (x1 y1 )2
kX(x, t) X(y, t)k
p
L2 (x2 y2 )2 + L2 (x3 y3 )2 + L2 (x1 y1 )2 = L kx yk .

p
Therefore X(x, t) satisfies the Lipschitz condition with the Lipschitz constant L = max{4M 2 , 1} in the
domain R.
2
for some constant M . Take L =
1.2
UNIQUENESS THEOREM
In this section, we prove the following uniqueness theorem.

Theorem 1.2.1 (Uniqueness Theorem) If the vector field X(x, t) satisfies a Lipschitz condition
in a domain R, then there is at most one solution x(t) of the differential system
dx
= X(x, t),
dt
(1.3)
that satisfies a given initial condition x(a) = c in R.

Proof Assume that X(x, t) satisfies the Lipschitz condition
kX(x, t) X(y, t)k L kx yk ,
for any (x, t), (y, t) R. Let x(t) = (x1 (t), . . . , xn (t)) and y(t) = (y1 (t), . . . , yn (t)) be two solutions satisfying
the same initial condition x(a) = c = y(a).
n
P
Denote (t) = kx(t) y(t)k2 =
[xk (t) yk (t)]2 0. Then,
k=1
0 (t)
n
X

2 x0k (t) yk0 (t) [xk (t) yk (t)]
k=1
n
X
2 [Xk (x(t), t) Xk (y(t), t)] [xk (t) yk (t)]
k=1
2 [X(x(t), t) X(y(t), t)] [x(t) y(t)] .
=
By the Schwarz inequality, we have
0 (t)
0
(t) = 2 |(X(x, t) X(y, t)) (x y)|
2 kX(x, t) X(y, t)k kx yk
2L kx yk2 = 2L(t).
The last inequality implies

d
(t)e2Lt = 0 (t) 2L(t) e2Lt 0.
dt
Hence (t)e2Lt is a decreasing function. Therefore for t > a, (t)e2Lt (a)e2La = 0. Since (t) 0,
we have (t) = 0 for t a, i.e., x(t) = y(t) for t a.
To argue for t < a, as above we again have
0 (t) |0 (t)| 2L(t),
which implies

d
(t)e+2Lt = 0 (t) + 2L(t) e+2Lt 0.
dt
So (t)e+2Lt is an increasing function. Therefore for t < a, (t)e+2Lt (a)e+2La = 0. Again we have
(t) = 0 for t a, i.e., x(t) = y(t) for t a.
2
1.2 UNIQUENESS THEOREM
Example 1.2.1 Find a region where the differential equation x0 = x + 3x1/3 has a unique solution,
i.e., find (x0, t0) such that the solution x(t) of the differential equation with x(t0 ) = x0 is unique in
a neighborhood of (x0, t0).
Solution We will show the following
(1) the differential equation has a unique solution with x(t0 ) = x0 , x0 6= 0;
(2) there are more than one solution satisfying x(t0) = 0.
(1) For any given (x0 , t0 ), with x0 6= 0, we choose a small > 0 such that 0 6 [x0 , x0 + ]. By
Lemma 1.1.2, we know that the function X(x, t) = x + 3x1/3 satisfies a Lipschitz condition in the region

n
o

R = (x, t) |x x0 | , |t t0 | T = [x0 , x0 + ] [t0 T, t0 + T ],
where T > 0 is any fixed constant.3 By Theorem 1.2.1, we conclude that the differential equation has a
unique solution with x(t0) = x0 , x0 6= 0.
(2) It is easy to see that x(t) 0 is one solution of the differential equation with x(t0 ) = 0. We only need
to show that there exists another solution which also satisfies x(t0 ) = 0.
Consider the improper integral
Z x
du
.
1/3
0 u + 3u
For any c > 0, we know that
Z c
Z c
1
du
du
<
= c2/3 .
0<
1/3
1/3
u
+
3u
3u
2
0
0
Hence the improper integral converges for c > 0. This allows us to define an implicit function x(t) by
Z x
du
= t t0 .
u
+
3u1/3
0
We can further assume that x(t0 ) = 0, since the last equation becomes an identity when setting t = t0 .
Obviously, this function x(t) 6 0, otherwise we will have t t0 , a contradiction. This function x(t) certainly
satisfies the differential equation, which can be seen easily by differentiating both sides of the last equation. 2
3 We
can make the following argument. Let (x, t), (y, t) R = [x0 , x0 + ] [t0 T, t0 + T ]. Then, by the mean
value theorem, we have

X

|X(x, t) X(y, t)| =
(, t) |x y| = 1 2/3 |x y| ,
x
X
(x, t) = 1 + x2/3 is continuous in x
x
in the compact set [x0 , x0 + ], since 0 6 [x0 , x0 + ]. This indicates that there is a number L (might depend
on x0 and ) such that

1 2/3 L.
where is between x and y. This implies [x0 , x0 + ]. The function
Thus, the function X(x, t) satisfies a Lipschitz condition in [x0 , x0 + ] R.
1.3
CONTINUITY
Theorem 1.3.1 (Continuity Theorem) Let x(t) and y(t) be any two solutions of the differential
equation (1.3) in T1 t T2 , where X(x, t) is continuous and satisfies the Lipschitz condition (1.2)
in some region R that contains the region where x(t) and y(t) are defined. Then
kx(t) y(t)k eL|ta| kx(a) y(a)k ,
for any a, t [T1, T2 ].
Proof Let us first assume that t a. Then, for the function (t) = kx(t) y(t)k2 , as in the proof of
Uniqueness Theorem 1.2.1, we have
0 (t) 2L(t),
which implies

d
(t)e2Lt 0.
dt
Integrating the last inequality from a to t gives
(t)e2Lt (a)e2La ,
which yields the desired inequality.

In the case t a, again as in the proof of Uniqueness Theorem 1.2.1, we obtain
0 (t) 2L(t),
which implies

d
(t)e+2Lt 0.
dt
Now we integrate the last inequality from t to a to have
(t)e2Lt (a)e2La .
This implies the desired inequality in the case t a.
Corollary 1.3.2 Let x(t) be the solution of the differential equation (1.3) satisfying the initial
condition x(a, c) = c. Let the hypotheses of Continuity Theorem 1.3.3 be satisfied, and let the
function x(t, c) be defined for kc c0 k K and |t a| T . Then
(1) x(t, c) is a continuous function of both variables;
(2) if c c0 , then x(t, c) x(t, c0 ) uniformly for |t a| T .
Proof For (a), it is obvious that x(t, c) is continuous in t since x(t, c) is a solution of the differential
equation
To see that x(t, c) is continuous in c, we take c1 and c2 , with
1
(1.3), which
is differentiable.

c c0 K and c2 c0 K. Since x(t, c1 ) and x(t, c2 ) are solutions of (1.3), by Continuity Theorem 1.3.3, we get

x(t, c1 ) x(t, c2 ) eL|ta| x(a, c1 ) x(a, c2 ) = eL|ta| c1 c2 .

, whenever c1 c2 < , we have
eL|ta|

x(t, c1 ) x(t, c2 ) < eL|ta| = .
Let t be fixed. Hence, for any given > 0, take =
1.3 CONTINUITY
Therefore, x(t, c) is continuous in c.
For (b), in above, we can take
=

min
eL|ta|
t[aT,a+T ]
.
1
> 0 is a finite number, hence, > 0 is independent of t. Therefore, x(t, c) is
eL|ta|
continuous in c uniformly for |t a| T .
2
Since
min
t[aT,a+T ]
Theorem 1.3.3 (Strong Continuity Theorem) Assume that

(1) x(t) and y(t) satisfy the differential equations
dx
= X(x, t)
dt
and
dy
= Y(y, t),
dt
respectively, on T1 t T2 .
(2) The functions X and Y be defined and continuous in a common domain R, and satisfy
kX(z, t) Y(z, t)k ,
for any (z, t) R, with T1 t T2 .
(3) X(x, t) satisfies the Lipschitz condition (1.2) in R.4
Then
kx(t) y(t)k kx(a) y(a)k eL|ta| +

L|ta|
1 ,
e
L
for any a, t [T1, T2 ].

Proof Assume that a, t [T1 , T2 ]. Let us first consider the case t a. Define (t) = kx(t) y(t)k2 =
n
P
[xk (t) yk (t)]2 0. We have
k=1
0 (t)

2 x0 (t) y0 (t) [x(t) y(t)]
2 [X(x(t), t) Y(y(t), t)] [x(t) y(t)]
2 [X(x(t), t) X(y(t), t)] [x(t) y(t)] + 2 [X(y(t), t) Y(y(t), t)] [x(t) y(t)] .
The Schwarz inequality implies

0 (t) 0 (t)
2 kX(x(t), t) X(y(t), t)k kx(t) y(t)k + 2 kX(y(t), t) Y(y(t), t)k kx(t) y(t)k
2L kx(t) y(t)k2 + 2 kx(t) y(t)k

p
2L(t) + 2 (t).
4 The
function Y(y, t) may not satisfy a Lipschitz condition.
10

(1) Let us assume that (a) > 0. Consider the initial value problem
du = 2Lu + 2u,
t a,
dt
u(a) = (a).
Since initially u(a) > 0, the right hand side of the differential equation is non-negative. So, u is an
p increasing
function. Hence, for t a, u(t) u(a) > 0. Thus, we can introduce the substitution v(t) = u(t). This
gives the equivalent differential equation
2v
dv
= 2Lv2 + 2v.
dt
Since v(t) > 0, we can divide both sides of the equation by v(t) to obtain a linear differential equation for
v. This leads to the following initial value problem
dv Lv = ,
t a,
dt
p
v(a) = u(a).
The solution of the initial value problem is

p
p
L(ta)
e
u(t) = v(t) = u(a)eL(ta) +
1 .
L
Since u(a) = (a) > 0, we have
[(t) u(t)]0
2L[(t) u(t)] + 2[
2L[(t) u(t)] + 2 p
which implies
(t)
u(t)]
(t) u(t)
p
(t) + u(t)
(t) u(t)
,
2L[(t) u(t)] + 2 p
u(a)
oi0
h
n
p
0.
[(t) u(t)] exp 2L + 2/ u(a) t
Integrating the last inequality from a to t, we have

n
o
n
o
p
p
[(t) u(t)] exp 2L + 2/ u(a) t [(a) u(a)] exp 2L + 2/ u(a) a = 0.
Thus,
(t) u(t),
which implies
kx(t) y(t)k =
(t)
=
=
=
u(t)
u(a)eL(ta) +

L(ta)
e
1
L

p
L(ta)
L(ta)
e
(a)e
+
1
L

L(ta)
e
1 .
kx(a) y(a)k eL(ta) +
L
1.4 EXISTENCE THEOREM
11
(2) For (a) = 0, we have to modify the discussion above. For each positive integer n, we consider the
following initial value problem
du = 2Lu + 2u,
t a,
dt
u(a) = 1/n.
The discussion above can be applied to this problem, and we have the solution
h
i2
L|ta|
e
un (t) = n1/2 eL|ta| +
1
.
L
If we can show that (t) un (t) for t a, then, after taking n , we obtain the desired inequality in
the case (a) = 0. The last inequality can be proved by contradiction. In fact, if
(t1 ) > un (t1 ),
for some t1 > a, then there exists t0 to be the largest t in the interval a < t t1 such that (t0 ) un (t0 ).
Obviously, (t0 ) = un (t0 ) > 0 and (t) > un (t) for t0 < t t1 . But this is impossible according to the
discussion in the case (a) = 0.
We can use the inequality
0 (t) |0 (t)| 2L(t) + 2
(t)
and a similar argument to prove the following inequality

kx(t) y(t)k kx(a) y(a)k eL(at) +

L(at)
e
1 .
L
2
1.4
EXISTENCE THEOREM
In this section, we study existence of the differential equation (1.3). The idea is to establish an
equivalent integral equation for any given initial value problem. Then we show that the iteration of
the integral operator converges to a solution.
Theorem 1.4.1 Let X(x, t) be a continuous function. Then a function x(t) is a solution of the
initial value problem
dx = X(x, t),
dt
(1.4)
x(a) = c
if and only if it is a solution of the integral equation
Z t
x(t) = c +
X(x(s), s)ds.
(1.5)
Proof Let us assume that x(t) is a solution of the initial value problem (1.4). The Fundamental Theorem
of Calculus implies that
Z t
x0k (s)ds.
xk (t) = xk (a) +
a
Using (1.4), we have the integral equation (1.5).
12
Conversely, if x(t) is a solution of the integral equation (1.5), then x(a) = c and, by the Fundamental
Theorem of Calculus, we have
k = 1, . . . , n.
x0k (t) = Xk (x(t), t),
dx
= X(x, t).
2
These imply that x(t) satisfies
dt
For a given X(x, t), if it is defined for all x in |t a| T , and is continuous, then we can define
an operator U by
Z t
U (x) = c +
X(x(s), s)ds.
(1.6)
a
For this operator, its domain is

x(t) x(t) is continuous in the interval |t a| T
and its range is

y(t) y(t) is continuously differentiable in the interval |t a| T and y(a) = c .
Thus, a solution of the integral equation (1.5) is a fixed point of the operator U :
x = U (x).
Theorem 1.4.1 can be re-stated as the following:
Theorem 1.4.2 Let X(x, t) be a continuous function. Then a function x(t) is a solution of the
initial value problem (1.4) if and only if the operator U defined in (1.6) has a fixed point in C[a
T, a + T ].
Now we can use the operator U to generate a sequence of functions {xn } from the given initial
data by the successive iteration:
x0 (t) c,
xn = U (xn1 ) = U n (x0 ),
n = 1, 2, . . ..
(1.7)
This is call a Picard iteration.5

Now we show that, under some conditions, a Picard iteration converges to a solution of the initial
value problem (1.4).
Lemma 1.4.3 Assume that X(x, t) is continuous and satisfies the Lipschitz condition (1.2) on the
interval |t a| T for all x, y. Then the Picard iteration (1.7) converges uniformly for |t a| T .
Proof Let M =
sup |X(c, t)| < +. Without loss of generality we can assume that a = 0 and t a.
|ta|T
In other words, we prove the lemma on the interval 0 t T . The proof for the general a and t < a can
be deduced from this case by the substitutions t t + a and t a t.
We first prove by induction that
n

n
x (t) xn1 (t) (M/L)(Lt) ,
n!
5 The
Picard iteration is started from x0 = c, given by the initial data.
n = 1, 2, . . . .
(1.8)

In fact, for n = 1,

1
x (t) x0 (t)

Z t

0

X(x
(s),
s)
ds

0
Z t

X(x0 (s), s) ds
0
Z t
(M/L)(Lt)1
.
M
ds = M t =
1!
0
For n = 2,
2

x (t) x1 (t)
13

Z t

1
0

(s),
s)
X(x
(s),
s)
ds
X(x

0
Z t

X(x1 (s), s) X(x0 (s), s) ds
0
Z t

1
x (s) x0 (s) ds
L
L
M s ds =
0
LM t2
(M/L)(Lt)2
=
.
2
2!
Assume that the desired inequality holds for n = k:

k
(M/L)(Lt)
k
k1
(t)
.
x (t) x
k!
Then,

k+1
(t) xk (t)
x
Z t h
i

k
k1

X(x (s), s) X(x
(s), s) ds

0
Z t

k
k1
(s), s) ds
X(x (s), s) X(x
0
Z t

k
k1
L
(s) ds
x (s) x
L
t
0
(M/L)(Ls)k
(M/L)(Lt)k+1
ds =
.
k!
(k + 1)!
Hence, the estimates (1.8) hold.

Next, we show that the sequence {xn (t)} converges uniformly for 0 t T . Indeed, for 0 t T , since
(M/L)(LT )k
(M/L)(Lt)k
,
k!
k!
and the positive series
X
(M/L)(LT )k
LT
is convergent to (M/L) e
k!
k=1

1 , by the Comparison Test, the series
x0 (t) +
h
i
X
xk (t) xk1 (t)
k=1
14
is also convergent. Actually, the convergence is uniform for 0 t T , since

t. The n-th partial sum,
n h
i
X
x0 (t) +
xk (t) xk1 (t) = xn (t).
(M/L)(LT )k
is independent of
k!
k=1
Therefore, the sequence of functions {xn (t)} converges uniformly.
Theorem 1.4.4 (Existence Theorem) Assume that X(x, t) is continuous and satisfies the Lipschitz condition (1.2) on the interval |t a| T for all x, y. Then the initial value problem (1.4)
has a unique solution on the interval |t a| T .
Proof The uniqueness is a direct consequence of Uniqueness Theorem 1.2.1. We only need to prove the
existence.
By Lemma 1.4.3, the sequence {xn (t)} defined by the Picard iteration with x0 (t) c is uniformly
convergent. Denote x (t) the limit function. We show that x (t) is a solution of the integral equation (1.5).
By the definition,
Z
t
xn+1 (t) = c +
X(xn (s), s) ds.
The left hand side is uniformly convergent to x (t). By the Lipschitz condition,
kX(xm (s), s) X(xn (s), s)k L kxm (s) xn (s)k ,
and so the integral on the right hand side is also uniformly convergent. Since X(x, s) is continuous, we know
that
X(xn (s), s) X(x (s), s).
Hence, we obtain
x (t) = c +
X(x (s), s) ds.

0
Finally, by Theorem 1.4.1, we conclude that the function x (t) is a solution of the initial value problem (1.4).
2
Example 1.4.1 Solve the initial value problem

dx
= x,
dt
x(0) = 1.
by the Picard iteration.
Solution For this initial value problem, the integral operator U is defined as
U (x(t)) = 1 +
x(s) ds
0
15
Thus, the Picard iteration gives

x0 (t)
1,
x1 (t)
U (x0 (t)) = 1 +
U (x1 (t)) = 1 +
x2 (t)
1 ds = 1 + t,
0
t
(1 + s) ds
0
1+t+
t2
,
2
..
.
xn (t)
xn+1 (t)
=
=
=
tn
t2
+ ... + ,
2
n!
U (xn (t))

Z t
sn
s2
+ ... +
ds
1+s+
1+
2
n!
0
1+t+
1+t+
tn
tn+1
t2
+ ... +
+
.
2
n!
(n + 1)!
We see that the sequence {xn (t)} converges uniformly to the function et . Hence we get the solution x(t) = et
by the Picard iteration.
2
Example 1.4.2 Verify the Taylor series for sin t and cos t by applying the Picard iteration to the
first order system corresponding to the second order initial value problem
x00 = x,
x(0) = 0,
x0 (0) = 1.
Solution The associated first order system is
dx1
= x2 ,
dt
dx2 = x1,
dt
with the initial condition
x(0) = (x1 (0), x2 (0)) = (0, 1).
The corresponding X(x, t) = (x2 , x1 ), and the initial value c = (0, 1). The Picard iteration yields
x0 (t)
(0, 1),
x1 (t)
(0, 1) +
(1, 0) ds
0
=
x2 (t)
=
=
(0, 1) + (t, 0) = (t, 1),

Z t
(0, 1) +
(1, s) ds
0

t2
t2
(0, 1) + t,
= t, 1
,
2
2
16

x3 (t)
=
=
=
x4 (t)
=
=
=

s2
, s ds
2
0

3
t2
t
(0, 1) + t ,
3!
2

t2
t3
,
t ,1
3!
2

Z t
s3
s2
(0, 1) +
1 , s +
ds
2
3!
0

t2
t4
t3
(0, 1) + t , +
3!
2
4!

3
2
4
t
t
t
+
.
t ,1
3!
2!
4!
(0, 1) +
Z t
We claim that for n = 0, 1, 2, . . .,

x2n+1 (t)
x2n+2 (t)
t2n+1
t2n
, 1 + + (1)n
t + + (1)
(2n + 1)!
(2n)!
n
t + + (1)n
t2n+1
t2n+2
, 1 + + (1)n+1
(2n + 1)!
(2n + 2)!
We prove this Claim by mathematical induction. When n = 0, they hold due to the calculations above.
Suppose they hold when n = k. Then we have

Z t
s2k+2
s2k+1
, s (1)k
ds
1 + + (1)k+1
x2k+3 (t) = (0, 1) +
(2n + 2)!
(2k + 1)!
0

t2k+3
t2k+2
=
t + + (1)k+1
, 1 + + (1)k+1
,
(2k + 3)!
(2k + 2)!

Z t
s2k+2
s2k+3
, s (1)k+1
ds
1 + + (1)k+1
x2k+4 (t) = (0, 1) +
(2k + 2)!
(2k + 3)!
0

t2k+3
t2k+4
=
t + + (1)k+1
, 1 + + (1)k+2
.
(2k + 3)!
(2k + 4)!
That is, the claim is also true for n = k + 1.
It can be easily shown that the associated initial value problem has a unique solution
x(t) = (sin t, cos t).
The uniqueness can be verified by checking that the conditions of Uniqueness Theorem 1.2.1 hold for the
associated problem. Since {xn (t)} converges to the unique solution by Existence Theorem 1.4.4, we have
sin t
t2n1
t3
+ + (1)n1
+ ,
3!
(2n 1)!
cos t
t2
t2n
+ + (1)n
+ .
2!
(2n)!
2
17
Example 1.4.3 Let f(x, t) be a function satisfying the Lipschitz condition |f(x, t) f(y, t)|
L|x y| for all 0 t T and all x and y. Suppose f(x, t) is continuous and bounded. Let
M =
sup |f(x, t)|. Let x(t) be a solution of x0 = f(x, t) with initial value x(0) = c and xk (t)
<x<
0tT
be the k-th term in the Picards approximation method. Prove

|x(t) xk (t)|
M Lk k+1
t ,
(k + 1)!
for t 0.
Solution: Use mathematical induction. When k = 0, we have
|x(t) x0(t)| = |x(t) x(0)| = |x0()|
for some 0 < < t by the Mean Value Theorem. Since |x0()| = |f(, t)| M , the desired inequality
holds for k = 0.
Suppose the inequality holds for k = n. Since
Z t

Z t

|x(t) xn+1(t)| = f(x(s), s)ds
f(xn (s), s)ds
0
|f(x(s), s) f(xn (s), s)| ds

0
L|x(s) xn(s)|ds
0
L
0
M Ln n+1
ds
s
(n + 1)!
M Ln+1
tn+2
,
(n + 2)!
we know that the inequality holds when k = n + 1. By mathematical induction, the inequality holds
for all k 0.
2
Exercise 1.4
1. For the initial value problem x0 = tx, x(0) = 1, obtain the n-th approximation of the Picard iteration. Use
mathematical induction to justify the formula.
2. For the initial value problem x0 = t + x, x(0) = 0, obtain x0 (t), x1 (t), x2 (t) and the n-th term of the sequence
of the Picard approximation. Use mathematical induction to justify the formula.
3. Assume that X(x, t) is continuous and satisfies the Lipschitz condition (1.2) on the interval |t a| T for all
x, y. For any function f (t), continuous on |t a| T , define the sequence {xn (t)} by
Z t
xn (t) = c +
X(xn1 (s), s)ds,
n = 1, 2, . . . .
x0 (t) = f (t),
a
Show that {xn (t)} is uniformaly convergent to the unique solution of the initial value problem (1.4).
18

4. Show that the initial value problem
00
2
y + y 1 = 0,
y(0) = y0 ,
0
y (0) = y1
is equivalent to the integral equation

y(t) = y0 + y1 t

(t s) y 2 (s) 1 ds.
5. Solve the integral equation

y(t) = cos t +
sin( t) y( )d.
0
6. (1) Show that the initial value problem

(
y 00 + (1 + t2 )y = 0,
y0 (0)
y(0) = 1,
t > 0,
=0
is equivalent to the integral equation

y(t) = cos t +
sin( t) 2 y( )d.
(2) Show that the successive approximations {n (t)} defined by
0 (t) = 0,
Z t
sin( t) 2 n1 ( )d,
n (t) = cos t +
n = 1, 2, . . .
converges uniformly to the solution in (1) for 0 t < T . Here T is any fixed constant.
1.5
1.5.1
LOCAL EXISTENCE THEOREM AND THE PEANO THEOREM

Local Existence Theorem
In the Existence Theorem 1.4.4, the function X(x, t) satisfies a Lipschitz condition (1.2) for all x.
This condition is quite strong and many functions may fail this condition.
Example 1.5.1 Show that the conclusion of Theorem 1.4.4 fails for the initial value problem
dx = ex ,
dt
x(0) = c.
Solution The solution of the initial value problem is given implicitly by
ec ex = t.
It is obvious that the function is defined only in < t < ec . Hence, there is no > 0 such that the
differential equation has a solution defined on all of |t| < for every initial value, since we can always take
a sufficient large c > such that ec < . Thus, the conclusion of Theorem 1.4.4 fails for this initial value
problem.
1.5 LOCAL EXISTENCE THEOREM AND THE PEANO THEOREM
19
The cause of this failure is that the function X(x) = ex does not satisfy a Lipschitz condition for all x.
In fact,
ex 1
|X(x, t) X(0, t)|
=
|x 0|
x
is unbounded for large values of x.
2
However, if the function X(x, t) satisfies a Lipschitz condition in a bounded domain, then a
solution exists in a limited region.
Theorem 1.5.1 (Local Existence Theorem) Assume that X(x, t) is continuous and satisfies the
Lipschitz condition (1.2) in the closed domain kx ck K, |t a| T . Then the initial value
problem (1.4) has a unique solution in the interval |t a| min{T, K/M }, where
M =
sup
kX(x, t)k .
kxckK
|ta|T
Proof The existence can be proved as in the proof of Theorem 1.4.4, except that we have to modify the
estimates for the sequence {xn (t)} by showing that if x(t) is defined and continuous on |ta| min{T, K/M }
satisfying
(a) x(t) is defined and continuous on |t a| min{T, K/M };
(b) x(a) = c;
(c) kx(t) ck K on |t a| min{T, K/M },
then one iteration y = U (x) still satisfies these three conditions. Indeed, (a) and (b) are obvious. To show
(c), we have the following
Z t

ky(t) ck =
X(x(s), s) ds

a

Z t

kX(x(s), s)k ds

a
M |t a| M
K
= K.
M
1.5.2
The Peasno Theorem
The Lipschitz condition plays an important role in the proof of the Local Existence Theorem.
However, if we drop it, we are still able to prove the existence, by a more sophisticated argument.
Theorem 1.5.2 (Peano Existence Theorem) Assume that X(x, t) is continuous in the closed
domain kx ck K, |t a| T . Then the initial value problem (1.4) has at least one solution in
the interval |t a| min{T, K/M }, where
M =
sup
kX(x, t)k .
kxckK
|ta|T
To prove this theorem, we need the following definition and the famous Arzel
a-Ascoli Theorem.
20
Definition 1.5.3 A family of functions F is said to be equicontinuous on [a, b] if for any given
> 0, there exists a number > 0 such that
kx(t) x(s)k <
whenever |t s| < for every function x F and t, s [a, b].
Arzel
a-Ascoli Theorem Assume that the sequence {xn (t)} is bounded and equicontinuous on [a, b].
Then there exists a subsequence {xni (t)} that is uniformly convergent on [a, b].
Now we are ready to prove Peano Existence Theorem 1.5.2.
Proof Denote T1 = min{T, K/M }. As argued at the beginning of the proof of Lemma 1.4.3, we only need
to prove the theorem on 0 t T1 , with a = 0.
We first construct a sequence of bounded equicontinuous functions {xn (t)} on [0, T1 ]. For each n, define
for 0 t T1 /n,
c,
Z tT1 /n
xn (t) =
X (xn (s), s) ds, for T1 /n < t T1 .

c+
0
The above formula defines the value of xn (t) recursively in terms of the previous values of xn (t).
We can use mathematical induction to show that
kxn (t) ck K
on [0, T1 ]. Indeed, on [0, T1 /n], it is trivial since xn (t) = c. If we assume that the inequality holds on
[0, k T1 /n] (0 k < n), then on [k T1 /n, (k + 1) T1 /n],
Z

tT1 /n

n
n
X (x (s), s) ds
kx (t) ck =
0

M |t T1 /n| M T1 K.
Hence, the sequence {xn (t)} is uniformly bounded on [0, T1 ]:

kxn (t)k kck + K.
The equicontinuity of the sequence {xn (t)} on [0, T1 ] can be
t1 , t2 [0, T1 ],
0,

Z t T 1/n
2

n
X(x
(s),
s)ds

,

Z t1 T 1/n
kxn (t1 ) xn (t2 )k =

n
X(x
(s),
s)ds
,

t2 T 1/n
X(x (s), s)ds ,

t1 T 1/n
M |t s|.
proven by the following estimates: for any
if t1 , t2 [0, T1 /n],
if t1 [0, T1 /n] and t2 (T1 /n, T1 ],
if t2 [0, T1 /n] and t1 (T1 /n, T1 ],
if t1 , t2 (T1 /n, T1 ]
1.5 LOCAL EXISTENCE THEOREM AND THE PEANO THEOREM
21
By Arzel
a-Ascoli Theorem, we know that there exists a uniformly convergent subsequence {xni (t)} that
converges to a continuous function x (t) on [0, T1 ] as ni . We can show that the function x (t) is
actually a solution of the initial value problem (1.4). Indeed, for any fixed t (0, T1 ], we take ni sufficiently
large such that T1 /ni < t. Thus, by the definition of {xn (t)}, we have
Z t
Z t
xni (t) = c +
X (xni (s), s) ds
X (xni (s), s) ds.
0
tT1 /ni
As ni , since X (x, t) is uniformly continuous, we have
X (xni (s), s) ds
0
second integral of the last equation tends to zero, since

Z
Z
t

t
T1

ni
X (x (s), s) ds
M ds = M
0.

tT1 /ni

ni
tT1 /ni
X (x (s), s) ds; the

0
Hence, we know that the function x (t) satisfies the integral equation
Z t
x (t) = c +
X (x (s), s) ds.
0
Without the Lipschitz condition, it is known that solution of initial value problem is not necessarily
unique. For instance, it can be earily see that the initial value problem
dx = 3 x1/3,
t 0,
dt
2
x(0) = 0,
is not unique, with the following two solutions
x1(t) = 0,
x2(t) = t3/2.
Exercise 1.5
1. Determine the existence region of the solution to the initial value problem
dx = t2 + x2 ,
dt
x(0) = 0,
where the equation is defined in the region R = {(x, t)||x| 1, |t| 1}. Find the first four approximations
x0 (t), x1 (t), x2 (t), x3 (t) of the Picard iteration.
2. Determine the existence region of the solution to the initial value problem
dx = t2 x2 ,
dt
x(0) = 0,
where the equation is defined in the region R = {(x, t)||x| 1, |t| 1}. Find the first four approximations
x0 (t), x1 (t), x2 (t), x3 (t) of the Picard iteration.
22

3. Assume that the functions F (x1 , x2 , . . . , xn , t) is continuous and satisfies a Lipschitz condition in |t a| T ,
kx ck K. Then the initial value problem
( (n)
u
= F (u, u0 , u0 , . . . , u(n1) , t),
u(i) (a) = ci ,
0 i n 1,
has a unique solution on |t a| min{T, K/M }, where

M=
1/2
.
x22 + + x2n + F 2 (x1 , x2 , . . . , xn , t)
sup
kxckK
|ta|T
1.6
LINEAR SYSTEMS
In this section, we apply the theorems obtained in the previous sections to linear differential systems
of the form
n
dxi X
aij (t)xj (t) + bi (t),
1 i n,
=
(1.9)
dt
j=1
Or, in a matrix form, we write

dx
= A(t)x + b(t),
dt
where A(t) = (aij (t)) and b(t) = (b1 (t), . . . , bn(t)).
(1.10)
Theorem 1.6.1 Assume that the functions aij (t) and bi (t) are continuous for |t a| T , 1
i, j n. The the initial value problem (1.9) with x(a) = c = (c1 , . . . , cn) has a unique solution on
|t a| T .
Proof The initial value problem can be re-written in terms of the vector form (1.4), with
X(x(t), t) =
n
X
a1j (t)xj (t) + b1 (t), . . . ,
j=1
n
X
anj (t)xj (t) + bn (t) ,
j=1
and
x(a) = (c1, . . . , cn ).
The function X(x, t) is continuous and satisfies the Lipschitz condition (1.2) on the interval |t a| T for
all x, y. Indeed, by the Schwarz inequality, we have
kX(x, t) X(y, t)k2
=
n
X
!2
a1j (t)(xj yj )
j=1
n
X
|a1j (t)|
j=1
|aij (t)|
sup
i,j |ta|T
+ +
!2
anj (t)(xj yj )
j=1
n
X
|xj yj |
j=1
|aij (t)|
+ +
n
X
j=1
kx yk2
i,j
n
X
kx yk2 .
|anj (t)|
n
X
j=1
|xj yj |
1.6 LINEAR SYSTEMS
23
Hence the function X(x, t) satisfies the conditions in Existence Theorem 1.4.4. This gives the existence of
the solution. The uniqueness follows from Uniqueness Theorem 1.2.1.
2
If bi(t) 0, 1 i n, the linear differential system (1.9) is called homogeneous, for which we
can construct its general solution.
Assume that x1 , . . . , xn are n solutions of a homogeneous linear differential system. Obiously, the
matrix
1
x1 (t) xn
1 (t)
.. ,
(t) = x1 (t), . . . , xn(t) = ...
(1.11)
.
x1n (t) xn
n (t)
satisfies
d(t)
= A(t)(t).
dt
If these n solutions are linearly independent at every point t, (t) is called a fundamental matrix of
the homogeneous linear differential system.
Theorem 1.6.2 Assume that the functions aij (t) are continuous for |t a| T , 1 i, j n. If
x1 , . . . , xk are k solutions of the homogeneous linear differential system, then the constant vectors
x1 (t0 ), . . . , xk (t0) are linearly independent for some t0 if and only if they are linearly independent
for every t on |t a| T .
Proof Assume that x1 (t0 ), . . . , xk (t0 ) are linearly independent for some t0 . Let t be another point on
|t a| T other than t0 . We need to show that if
1x1 (t) + + n xk (t) = 0,
then, 1 = . . . = k = 0. In fact, since x1 , . . . , xk are solutions, we have two solutions of the homogeneous
system, 0 and
1 x1 + + n xk ,
both vanishing at t. By the uniqueness of Theorem 1.6.1, we conclude that
1 x1 (t) + + n xk (t) = 0
for every t on |t a| T . In particular, we have
1 x1 (t0 ) + + n xk (t0 ) = 0.
The linear independence of x1 (t0 ), . . . , xk (t0 ) implies that all i = 0, 1 i k.
The other direction is obvious.
Theorem 1.6.3 Assume that the functions aij (t) are continuous for |t a| T , 1 i, j n. Let
xi (t), 1 i n, be n solutions of the homogeneous linear differential system
(1.9) and assume that
they are linearly independent at some t0 . Define (t) = x1 (t), . . . , xn(t) . Then the solution of
the homogeneous linear differential system satisfying the initial condition x(a) = c = (c1 , . . . , cn) is
given by
x(t) = (t)1 (a)c.
24
In particular, if xi (t) be the solution of the homogeneous linear differential system (1.9) that satisfies
the initial condition xik (a) = 0, i 6= 0, xii (a) = 1. The the solution satisfying the initial condition
x(a) = c = (c1 , . . . , cn) is given by
x(t) = c1x1 (t) + + cn xn(t).
Since x1 (t0 ), . .. , xn (t0 ) are linearly independent, by Theorem 1.6.2, we know that the matrix
(t) = x1 (t), . . . , xn (t) is non-singular for every t on |t a| T and hence, (t) is a fundamental matrix.
Let x(t) be the unique solution of the homogeneous system satisfying the initial condition x(a) = c. Its
existence is guaranteed by Theorem 1.6.1. Consider the function
Proof
x(t) (t)1 (a)c.

Since (t) is a fundamental matrix, we know this function is again a solution of the homogeneous system.
Now we have two solutions of the homogeneous system, 0 and x(t) (t)1 (a)c, both vanishing at t = a.
By the uniqueness of Theorem 1.6.1, we have
x(t) (t)1 (a)c = 0,
That is, x(t) = (t)1 (a)c.
In particular, if xi (t) satisfies the initial condition xik (a) = 0, i 6= 0, xii (a) = 1, then (a) is the identity
matrix. Therefore
x(t) = (t)c = c1 x1 (t) + + cn xn (t).
2
Theorem 1.6.4 Let A(t) = (aij (t)) and b(t) be continuous on |t a| T . The solution of the
non-homogeneous linear system (1.10), with initial condition x(t0) = x0 , is given by
Z t
x(t) = (t)1 (t0)x0 + (t)
1 (s)b(s)ds.
t0
where (t) is any fundamental matrix of the corresponding homogeneous system. In particular, if
A is a constant matrix, then
Z t
1
x(t) = (t) (t0 )x0 +
(t s + t0)1 (t0)b(s)ds.
t0
Furthermore, if (t) is the fundamental matrix satisfying (0) = I, then

Z t
x(t) = (t)x0 +
(t s)b(s)ds.
t0
Proof For any given fundamental matrix (t) of the homogeneous system, we postulate the solution x(t)
of the nonhomogeneous system satisfying the initial condition x(t0 ) = x0 to be in the form
x(t) = (t)1 (t0 ) {x0 + (t)} .
Then, (t) satisfies the initial condition (t0 ) = 0. To find the equation satisfied by (t), we substitute the
expression into the nonhomogeneous system to have
0 (t)1 (t0 ) {x0 + (t)} + (t)1 (t0 )0 (t) = A(t)(t)1 (t0 ) {x0 + (t)} + b(t).
1.7 CONTINUATION OF SOLUTIONS
25
Since 0 (t) = A(t)(t), the last equation gives

(t)1 (t0 )0 (t) = b(t),
whose solution satisfying (t0 ) = 0 is
(t) = (t0 )
1 (s)b(s)ds.
t0
Thus
x(t) = (t)1 (t0 )x0 + (t)
1 (s)b(s)ds.
t0
If A is a constant matrix, to show

x(t) = (t)1(t0 )x0 +
(t s + t0 )1 (t0 )b(s)ds,
t0
it is sufficient to show that

(t)1 (s) = (t s + t0 )1 (t0 ),
for any t, s and t0 . Indeed, let U (t) = (t)1 (s) and V (t) = (t s + t0 )1 (t0 ), with s and t0 being
two parameters. It is obvious that U (s) = V (s) = I. We know that they also satisfy the same differential
system, since
U 0 (t) = 0 (t)1 (s) = A(t)1 (s) = AU (t),
V 0 (t) = 0 (t s + t0 )1 (t0 ) = A(t s + t0 )1 (t0 ) = AV (t).
By the uniqueness of Theorem 1.6.1, the corresponding columns of U and V are identical since they satisfy
the same equation with the same initial conditions.
2
1.7
CONTINUATION OF SOLUTIONS
Theorem 1.7.1 Assume that X(x, t) is continuously differentiable in an open region R of (x, t)space. For any point (c, a) R, the initial value problem (1.4) has a unique solution x(t) defined
over an interval a t < b (b is a finite number or infinite) such that if b < +, either x(t) is
unbounded as t b or (x(t), t) approaches the boundary of the region R.6
Proof By the Local Existence Theorem 1.5.1, we know that there exist solutions of the initial value
problem (1.4) in some interval [a, T ). Given two solutions x(t) and y(t), defined in [a, T1 ) and [a, T2 )
respectively, we define a new function z(t) to be either x(t) or y(t) wherever either is defined. Then z(t) is
again a solution of (1.4) defined on [a, max{T1 , T2 }). Thus, we can denote
n
o

b = sup T the initial value problem (1.4) has a solution in [a, T ) .
6 The
conclusion of Theorem 1.7.1 indicates that there are only three possible outcomes
1. b = + or
2. b < + and x(t) is unbounded as t b or
3. b < + and (x(t), t) approaches the boundary of R.
26
We can define a single solution, denoted again by x(t), called the maximal solution, defined in [a, b). The
above construction indicates the existence of the maximal solution. It is also unique, by Uniqueness Theorem 1.2.1.
Let us consider the limiting behavior of x(t) as t b. Then there are only possibilities:
Case 1: b = +.
Case 2: b < + and x(t) is unbounded as t b.
Case 3: b < + and x(t) is bounded as t b.
We only need to show that in the third case, x(t) approaches the boundary of the region R as t b.
Indeed, let {tn } be any sequence such that tn b. Since the sequence of points {(x(tn ), tn )} is bounded,
there exists at least one limit point, say (d, b). We now show that the point (d, b) is on the boundary of R.
In fact, if it is an interior point, then there exists a closed neighborhood D : kx dk , |t b| of (d, b)
also in R. Let M = maxD kX(x(t), t)| k. Take < min{, /2M } and let G D be the open set

n
o

G = (x, , t)kx dk < , |t b| < .
We can take k large enough such that (x(tk ), tk ) G. Applying Local Existence Theorem 1.5.1 to the
dy
= X(y(t), t), we know that there exists a unique solution in the interval |t b| < satisfying
equation
dt
y(tk ) = x(tk ). If we define
(
x(t), a t < b,
z(t) =
y(t), b t < b + ,
then clearly z(t) is a solution of the initial value problem (1.4) over the interval [a, b + ), contradicting to
the maximality of b.
2
Example 1.7.1 Find the maximal solutions for the following initial value problems:
dx = x,
dx = x2,
dt
dt
(1)
(2)
x(0) = c;
x(0) = c;
dx
1
1
= 2 cos ,
dt
t
t
(3)
x(t0) = c, t0 6= 0.
Solution
(1) The function X(x, t) = x is continuously differentiable at any point (x, t) (, +) (, +).
The differential system has a unique solution x(t) = cet, which is the maximal solution defined for t
[0, +).
(2) The function X(x, t) = x2 is continuously differentiable at any point (x, t) (, +) (, +).
If c = 0, the maximal solution is x(t) = 0, defined for t [0, +). If c 6= 0, the maximal solution is
1
x(t) =
, which is defined for t [0, +) if c < 0; or t [0, c) if c > 0. Obviously, in the later case, x(t)
ct
becomes unbounded as t c.
1
1
(3) The function X(x, t) = 2 cos is continuously differentiable at any point (x, t) (, +)
t
t
1
1
(, +) with t 6= 0. The maximal solution is x(t) = c sin + sin , which is defined for t [t0 , 0) if
t
t0
t0 < 0; or for t [t0 , +) if t0 > 0. In the former case, the function x(t) is still bounded, but t = 0 is the
boundary of

n
o

R = (x, t) t 6= 0 .
1.8 MISCELLANEOUS PROBLEMS
27
2
1.8
MISCELLANEOUS PROBLEMS
1. Show that the following initial value problem
dy
dx = x2 + y cos s,
= y 2 + sin s,
dt
dt
x(0) = s + 1,
y(0) = s2 /(s2 1)
has a unique solution (xs (t), ys (t)) in |t| T , |s| 1/2 for some T . Prove
lim (xs (t), ys (t)) = (1, 0).
s0
Solution For any K1 > 0 and T1 > 0, in the closed bounded region
R=

p
(x, y, t, s) x2 + y2 K1 , |t| T1 , |s| 1/2

the function x2 +ycos s, y2 +sin s is continuously differentiable for all its variables (x, y, t, s). Thus,
we know that it is a Lipschitz function in this region. By the Local Existence Theorem, for any fixed
s with |s| 1/2, the given initial value problem has a unque solution for |t| Ts = min{T1 , K1/Ms },
where
p
[x2 + y cos s]2 + [y2 + sin s]2 .
Ms = sup
x2 +y 2 K1
|t|T1
It is obvious that
M = max Ms < +.
|s|1/2
Thus, for any s with |s| 1/2, the given initial value problem has a unque solution for |t| T =
min{T1 , K1 /M }.

Denote the Lipschitz constant of the function x2 + y + t cos s, y2 + sin s in R to be L. It is easy
to see that
2

(x + y cos s, y2 + sin s) (x2 + y 1, y2 ) = k(1 cos s, sin s)k .
By the Strong Continuity Theorem, the solutions (xs (t), ys(t)) and (x0 (t), y0 (t)) = (1, 0), of the
following initial value problems
dy
dx = x2 + y cos s,
= y2 + sin s,
dt
dt
x(0) = s + 1,
y(0) = s2 /(s2 1),
and
dy
dx = x2 + y 1,
= y2 ,
dt
dt
x(0) = 1,
y(0) = 0,
28

respectively, satisfy
k(xs(t), ys (t)) (1, 0)k

(s + 1, s2 /(s2 1)) (1, 0) eL|t| + 1 k(1 cos s, sin s)k eL|t| 1 .
L
Let s 0, since the right hand side of the last inequality approaches zero, we have
lim (xs(t), ys(t) = (1, 0).
s0
2. Use the Local Existence Theorem to show that the initial value problem
dx = 1 + x2,
dt
x(0) = 0,
has a unique solution for |t| 1/2. Determine the region where the true solution is defined
by solving this initial value problem. What is the limiting behavior of the true solution as t
approaches the end point(s) of the maximal interval of existence?
Solution Consider the given initial value problem in the region

(x, t) |x| 1, |t| 1 .
The function X(x, t) = 1 + x2 satisfies the following Lipschitz condition

|X(x, t) X(y, t)| = |x + y| |x y| 2|x y|
for |x| 1 and |y| 1. Hence, by the Local Existence Theorem, there exists a unique solution of the
given initial value problem for |t| min{1, 1/M }, where

M = sup kX(x, t)k = sup 1 + x2 = 2.
|x|1
|t|1
|x|1
|t|1
That is, there exists a unique solution for |t| 1/2.

The unique true solution of the given initial value problem is
x(t) = tan t,
whose maximal interval of existence is (/2, /2). As t /2, x(t) becomes unbounded.
3. Assume that the function f(x, t) is continuously differentiable in

R = (, +) (a, b),
and satisfies the inequality
|f(x, t)| A(t)|x| + B(t),
29
for some non-negative continuous functions A(t) and B(t). Show that any solution of
dx = f(x, t),
dt
t0 (a, b),
x(t0 ) = c,
has a maximal interval of existence (a, b).
Proof Let x = x(t) be a solution of the initial value problem . We only show that it can be extend
to the interval [t0, b). The continuation of the solution to (a, x0 ] can be proved similarly.
We prove it by contradiction. Suppose that the maximal interval of existence is [t0 , ), with < b.
Select t1 and t2 such that
t0 < t1 < < t2 < b and t2 t1 < t1 t0 .
Denote T = t2 t1 > 0. Let AM and BM be positive upper-bounds of A(t) and B(t) in the interval
[t0 , t2 ], respectively. Thus, by the condition, we have
|f (x, t)| AM |x| + BM ,
1
. Now we will see
AM
that the solution x = x(t) can be extended to the interval [t0 , t2 ), a contradiction.
for (x, t) (, +) [t0 , t2 ]. We assume AM is large enough such that T <
In fact, since t1 (t0 , ) and the solution x = x(t) exists on [t0 , ), for any positive number K, the
region

n
o

R1 = (x, t)|x x(t1 )| K, |t t1 | T
is a bounded closed subset of R. In R1 , since
|f (x, t)| AM |x| + BM AM (|x(t1)| + K) + BM = M,
by the Local Existence Theorem, the solution curve (x(t), t) exists and remains in the region

n
o

R2 = (x, t)|x x(t1)| K, |t t1 | h ,
where
h = min {T, K/M } .
Since R2 is a bounded closed region, by Theorem 1.7.1, the solution curve (x(t), t) can be extended
to the boundary of R2 . That is, the solution exists in [t0, t1 + h). Since
lim
K+
1
K
=
> T,
M
AM
we know that for sufficient large K,

h = min {T, K/M } = T.
Thus, the solution x = x(t) can be extended to [t0 , t1 + T ) = [t0 , t2 ).
This contradiction implies = b.
2
Plane Autonomous Systems
2.1
PLANE AUTONOMOUS SYSTEMS
Recall a normal form of a system of first-order ordinary differential equations:
dx1
= X1 (x1, . . . , xn; t),
dt
..
.
dx
n
= Xn (x1, . . . , xn; t).

dt
If n = 2 and the functions Xi are independent of t, we have a so-called plane autonomous system.
In this case, we write the autonomous system as
dx
= X(x, y),
dt
dy
= Y (x, y).
dt
(2.1)
Let (x(t), y(t)) be a solution of system (2.1), we get a curve in R2, the phase plane, which is called
the solution curve of the system. The differential system gives the tangent vector of the solution
curve, (X(x(t), y(t)), Y (x(t), y(t)). A R2 plane together with all solution curves is called the phase
plane of the differential system.
The orientation of a solution curve is the direction
of the movement of points on the curve when
t increases. For example, the solution curve e2t , et of the system
dx
= 2x,
dt
dy = y,
dt
31
32
2 PLANE AUTONOMOUS SYSTEMS

is a parabola in Fig. 2.1(a). The solution curve e2t, et of the system
dx
= 2,
dt
dy = y,
dt
is the same parabola in Fig. 2.1(b), with the opposite orientation.
y.........
...
...
...
...
....
..
....
..
...
....
..
.......................
...............
....
..........
..
.........
... .............
... ......
..
........................................................................................................................................................................................
....
..
....
...
....
..
...
....
...
y.........
...
...
...
...
....
..
....
..
...
....
. .......
..
..................
...........
....
.........
..
.........
... ..............
... ......
..
..........................................................................................................................................................................................
....
..
....
...
....
..
...
....
...
(a)
(b)
Fig. 2.1 Solution curves with orientations
Example 2.1.1 Consider the differential system
dx
= 2x,
dt
dy = y,
dt
Sketch its oriented solutions curves in (x, y)-plane.
Solution The general solution is given by (x(t), y(t)) = (c1 e2t , c2 et ), where c1 and c2 are arbitrary
constants. For different values of c1 and c2 , we have one solution curve. They are all shown in Fig. 2.2. 2
In Example 2.1.1, the point (0, 0) is a solution curve of the differential system even though it
is not a curve in the ordinary sense. In Fig. 2.2, the point (0, 0) is a special point, at which the
direction is indeterminate. Such points are of particular importance for the study of autonomous
systems.
Definition 2.1.1 A point (x0, y0) is called a critical point of the autonomous system (2.1), if
X(x0 , y0 ) = y(x0 , y0 ) = 0.
If a point is not a critical point, we call it an ordinary point. Locally, the solution curves near
an ordinary point are a family of parallel lines. However, a local structure near a critical point can
be very complicated. If (x1 , y1) is a neighboring point of a critical point (x0, y0 ), the solution curve
2.1 PLANE AUTONOMOUS SYSTEMS
33
..
......
........
......
....
....
....
....
....
....
...
...
...
...
.
.
.....
.
.
.........
........
....
...
...
...
..........
...
...
......
...
...........
...
...........
...
.........
.
.
..........
.
..
...........
..
.............
..
....
..........
..
.........
.
.
.......
.
.
.
.
.
.
.
.
.
... .. ..
.......
.....
. .
......
......
...... .... .... .... .........
................ ...
... .................
.... .. .. .. .....
..................
.... .. ... .. ....
..................
............
.
.
.
............
...........
......... ......... ... ......... ...................
........ ............ ........
........... .....
...........
.........................................................................................................................................................................................................
.........
................................
.
.
.
.
.
.
. .
.
......... ......... ... ......... ..................
..........
............
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... ... .....
................
...
..... .. .. .. ....
...................
........................
.... .... .... .... .........
................ ..
.
.
.
.
......
. .. ..
....
.
.
.
.
.
.
.
.
.
.
.......
...
. ...
.
.
.
.
.
.
.
.
.
.
.
.
.
..........
...
.
........
.. ...........
..
..
....
...........
..........
..
.........
...
...
..........
..
..........
...
......
...
...
.........
..
.
.
.
.
....
.
.......
....
...........
.
.
.
.
.
.
...
.
...
....
...
....
....
...
....
....
....
.....
...
.....
...
Fig. 2.2 Solution curves with orientations
passing through (x1 , y1) can either stay close to the solution (x(t), y(t)) (x0, y0 ), or move away
from it, as t increases. In the former case, we say the critical point (x0 , y0) is stable; in the later
case, it is unstable. In a rigorous mathematical language, we introduce the following definition.
Definition 2.1.2 Let x0 = (x0, y0 ) be a critical point of the autonomous system (2.1), The point
(x0, y0 ) is called
(i) stable when, given > 0, there exists a > 0 such that kx(t) x0 k < for all t > 0 and all
solution x(t) = (x(t), y(t)) of the system (2.1) satisfying kx(0) x0 k < ;
(ii) attractive when, for some > 0, lim kx(t) x0 k = 0 for all solutions x(t) = (x(t), y(t)) of
t+
the system (2.1) satisfying kx(0) x0 k < ;

(iii) strictly stable when it is stable and attractive;
(iv) neutrally stable when it is stable but not attractive;
(v) unstable when it is not stable.
For linear homogeneous autonomous systems, we will see that an attractive critical point must be
a strictly stable point, but in general, this is not true. An example given by T. Brown demonstrates
this. The phase diagram is given in Fig. 2.3. Obviously, for any solution x(t) = (x(t), y(t)), the
critical point (0, 0) is attractive since limt kx(t)k = 0. However, it is not stable (so not strictly
stable), since for any > 0, no matter how small a neighborhood of the origin is, we can find an
initial point in the neighborhood such that kx(t)k for some t.
Theorem 2.1.3 Let V (x, y) be a continuously differentiable function.
tonomous system
dx
= Vy (x, y),
dt
dy
= Vx (x, y).
dt
Consider the plane au-
(2.2)
34
.
.. ............ ...
...
....
... ... ... ... ..
..
...
... ... ... ... ...
....
...
... ... .. ... ...
...
.
....
... ... .. ... ...
..
..
.
.
. .. .. .. ..
...
.
.
.
.
.
.
.
.
....
..
. .. .. .. ...
.....................................
.
.
.
.
.
.
.
.
.
.........
.
.
.
.
.
...................................................
.
.
.
.
.
......
. .. .. ... ...
.....
..
.
.
.
........
........
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
................
...... .. .. .. ......
.......
...................
.....
.
.
.
.
.
.
.....
.
.
.
.
. . . .
.
....
.....
......................
... ......... .... ......... .....
....
.....................
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
....
..........
..........
.. ... ... ... ....
......
.......
.
.
.
.
...
.
...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
....
......
... .. ... ..
....
...
.....
.
.
.
.
.
.
.
.
.
.
.
.
....
.
.
....... ..
.. .. ...
....
............
..
.
.
.
.
.
.
.
.
.
.. .. ..
....
...................
..........................
.
.
.
.
.
.
...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.. .. ...
........
.......
...
......
..
.
.......
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
....
..... .. .. .
.
...
.
.... . .. .. ........ .
....
...
....
...
...
.....
.. . . . ..
.....
....
..
...
...
................................... ..... .... ..... ...................... ...................
...
..
..
...
.....
....
...
..
..
........................... ... ..................................
.
.
.
.
.
.
....
.
.
.
.
...
...
.
..
.
........ .. ..........
.....
..
...
.
....
.
.
.
...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
....
.
.
.
.
.
.
.
.
.
.
.
.
...
..
..
.
....... ......................................... ..........
...
...
.
.
.
...
....
.
...
.
.
..
..
..
.........................
...
...
..
.
.
.
.
.
.
.
.
.
..
.
.
..
...
.
.
.
.
.
...
......
..
...
..
..
..
.
..
..
.
.
.
.
.
.
.
.
.
.
.
.............................................................................................................................................................................................................................................................................................................................................................................................
.
...
.
..
.
...
..
.
.
.
.
.
.
.
.
..... ................ .....
...
..
.
.
.
...
.
.
.
..
.
.
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...
....
...
.
.
............................
...
.
...
.
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
................... ....................................
...
...
..
.
..
.
..
..
.
.
.
.....
. .......................... .
.
..
...
..
.
.
.
.
.
.................................. ........................................
...
....
...
...
..
..
...
...
.....
.. . .. . .
..
...
...
..
...
.................. ............................. .... .............................. ...................
...
..
...
...
.........
....
..... ... ............ .... ...........
..
...
...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
......
.
.. .
.
...
.
......... .
....... ..... .. .......... ..............
.......
....
...
........................................ .......... ....
........................................
....
... ....
....
..
..
....
... .....
....
... ...
....
.....
... ......
..... .....
....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.......
...
. .
.... ........
...
....
....
..........
....... ...
...
....
....
........
....
.................................................. .......
....
..... .......................................................
...
....
....
....
....
....
....
.....
...
......
.....
.....
.......
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
........
........
..
........
..........
...........
........
....
...........................................................
.......................................................
...
...
...
...
...
..
.....
..
....
Fig. 2.3 The origin is attractive but not stable
Then each solution curve of the system lies on some level curve V (x, y) = C, where C is a constant.
Proof Let (x(t), y(t)) be a solution curve in (x, y)-plane. Then
dV
(x(t), y(t))
dt
=
=
=
V
(x(t), y(t))
x
V
(x(t), y(t))
x
0.
dx
V
dy
(t) +
(x(t), y(t))
(t)
dt
y
dt
V
V
V
(x(t), y(t)) +
(x(t), y(t)) (1)
(x(t), y(t))
y
y
x
Hence V (x(t), y(t)) = constant. That is, the solution curve (x(t), y(t)) lies on some level curve V (x, y) = C. 2
It is easy to show that an autonomous system (2.1) is in the form of (2.2) if and only if
X
Y
+
= 0,
x
y
provided both X(x, y) and Y (x, y) are continuously differentiable.
Example 2.1.2 Sketch the solution curves in the phase plane for the equation
d2 x
= x3 x.
dt2
2.1 PLANE AUTONOMOUS SYSTEMS
35
Solution The equation can be re-written as an autonomous system:

dx
= y,
dt
dy
= x3 x.
dt
It is easy to check that for X(x, y) = y and Y (x, y) = x3 x,

Y
X
+
= 0 + 0 = 0.
x
y
To find the function V (x, y) as in (2.2), we set
Vx = (x3 x).
Vy (x, y) = y,
Solving the first equation in the last system gives V (x, y) = 12 y2 + h(x), with h being an arbitrary function
in x. Substituting the expression of V into the second equation, we have h0 (x) = (x3 x). Solving for h,
we obtain
1
1
1
V (x, y) = y2 x4 + x2 + C,
2
4
2
with C a arbitrary constant. Now, by Theorem 2.1.3, we know that each solution curve of the differential
system lies on some level curve
1 2 1 4 1 2
y x + x = C,
2
4
2
where C is a constant. For different values of C, we sketch the level curves as in Fig. 2.4, which gives the
phase diagram of solution curves. Notice that there exist three critical points: (0, 0), (10) and (1, 0). It
can be shown that (0, 0) is neutrally stable and the others are unstable.
2
..
...
.. .... ...
.. ... ..
.. ... ...
........
... .... ...
.. .. ..
..
. .
....
... .. ...
.. .... ....
.. ... ....
..
.
.
..
.. ..
.
...
.. ...
.. ..
..
..
.. ..
.. ..
..
...
..
... ..
.. ...
..
....
.. ...
... ...
...
.
.
..
.
.
.
.
.
....
..
..
..................................................................................
...
..
..
..
..
.
..
...
...
..
..
...
..
...
..
..
..
.
.
.
.
.
.
.
..
.
...
..
...
..
....
..
..
...
..
..
...
..
..
..
...
...
.........
..
..
.
.
.
..
.
.
.
.
........
.........
..
...............................
...
...
... ........................................... ....
...
...
.
.. .... .... ... ... .... ...
...
.......................................................................................................................................................
...............................................................................................
.
...
..
...
....
.. ... .. ... ... ....
.. ...... ........................... ..... ....
..
..
......... .. ......
..
....
.
..
...........................
...
.......
.......
.
....
.
.
.
.
..
.
..
.
........
...
...
..
...
..
...
..
....
..
..
..
..
..
..
..
....
..
.
..
.
.
.
..
.
..
..
.
.
.
.
..
..
..
..
...
.
.
.
..
..
.
...........................................................................
.
.
...
.
...
.
.
.
.
.... ..
...
..
....
.. .....
.
... ..
..
. ...
..
..
.
.
.
.
... ..
.
..
. ...
.
.
...
.
... ..
..
..
..
... ....
.
.
.
..
.. ...
.
. ..
.
...
.
.
.
.. .. ....
.. ...
..
..
.
.
.
.
.. .. ...
..
... ... ...
.
.
.
.. ... ...
...
.... ... ..
.
.
.. .. ..
..
..
Fig. 2.4 Solution curves with three critical points
Example 2.1.3 Sketch the solution curves in the phase plane for the equation
d2
= k2 sin ,
dt2
k2 = g/l
which governs the motion of a simple pendulum of length l.
36
Solution The equation of a simple pendulum can be re-written as an autonomous system:

dv
= k2 sin .
dt
d
= v,
dt
If we write X(, v) = v and Y (, v) = k2 sin , then

Y
X
+
= 0 + 0 = 0.
v
Thus, we can write the autonomous system as in (2.2). Indeed, for
V (, v) = k2 sin ,
Vv (, v) = v,
we solve the first equation to have V (, v) = 12 v2 + h(). The second equations yields h() = k2 cos + C.
Thus, we have V (, v) = 12 v2 k2 cos + C. By Theorem 2.1.3, we know that each solution curve of the
differential system lies on some level curve
1 2
v k2 cos = C,
2
where C is a constant. In the phase plane, the (v, )-plane, the last equation indicates that if (v(t), (t)) is a
solution curve, so is (v(t), (t) + 2). We sketch its solution curves in Fig. 2.5 for the case k2 = 1. Obviously,
there are infinitely many critical points at (0, n), n = 0, 1, 2, . . ..
2
..
.......
...
...
..
..............
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...........
...........
.............................................
.......
........
.
.
.
.
.
.
.
.
.......
.
.
.
.
.
.
.
....
.
.
.........
.
.
.
.....
.......
.......
.....
......
.......
.
......
.
.
...
.
.
.......
......
.......
...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
......
.
.......
......
......
.......
.......
....
.......
.
.......
.
.
.
......... .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
................................
.............................
.......................
..............
...............................
..
..
...........
.......
........... .... .................
........
......
.......
.......
.....
.....
....
......
....
.....
..
....
.....
....
.....
....
....
.....
.
.
....
.
.
.
.
.
....
.
..
.
....
.
.
.
.
.
.
.
.
.
.
.
.
.
.......
.......
....
.... .
.
.
.
.
.
.
.
.
.
.
.........
.........
.
...
....
...
...
....
...
...
...
...................
...
........................
.........................
...
...
...
..... ... ........
.....
...
.....
....
...
...
....
...
...
...
.
...
...
..
.
...
.
.
.
.
.
.
.
.
.
.
.
.
.
...
...
...
.
..
...
.
...
.
.
.
.
...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...
..
...
.
..
...
.......
.......
.......
.
..
.
.
.
.
.
.
.
..
.
.
.
.
...
...
..
.
...
.
.
.
.
.
.
...
.
.
....
.
..
.
.
.
.
.
.
... ..
...
... ..
.
.
.
.
..
.
.
.
.
.
..
.
.
.
.
.
.
.
...................................................................................................................................................................................................................................................................................................
.............................................
......................................................................................
.
.
.
. ..
.
..
.
.
.
..
. .....
.. ....
..
.
.
.
.
.
.
.
..
.
.
.
.
.
.
..
.
..
. ....
..
.
.. ....
..
.
.
.
.
.
...
.
.
...
.
.
.
...
...
..
.
...
..
..
.
........
.
..
......
.
.
.
..
.
.
.
.
.
.
.
.
.
...
...
...
..
..
.
..
..
..
...
...
....
...
...
...
...
....
...
...
...
...
..
..
...
...
...
.
...
...
...
....
...
......
...
...... ... ........
....
...
......
...
..
...
........................
...
....................
......................
...
...
...
.
.
.
.
.
.
....
.....
.
..
........
....
....
.........
...
. ...
.....
......
....
...
....
........
....
.......
....
....
...
...
....
.
.
.
....
.
.
.
.
.
.....
.....
..
..
.....
.
.
.
..
.
.
.
.
.
.
.
.
.
.
.
......
......
......
....
...
.
...
.
.
.
.
.
.
.........
.
.
.
.........
......
.
.
.
.
.
.
.
.
.
............. . ............
......
.....
...........................
.
.............................
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.. ..........
. ..........
......
..
.......
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.......
.......
.....
......
......
......
....
......
......
......
.......
...
.......
......
......
.......
.......
.......
........
........
........
...
.......
.......
.
.
..........
.......
.
.
.
..........
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.....................................
........................................
................................
...
..
....
..
...
.
Fig. 2.5 Solution curves for a simple pendulum
2.2 LINEAR AUTONOMOUS SYSTEMS
2.2
37
LINEAR AUTONOMOUS SYSTEMS
In this section, we investigate two dimensional autonomous systems of the form

x0 = ax + by,
y0 = cx + dy,
where a, b, c and d are real numbers. It can be written in the following vector form

dx
x(t)
a b
= Ax,
x(t) =
,
A=
.
y(t)
c d
dt
(2.3)
(2.4)
It is obvious that the origin (0, 0) is a critical point of the system. If det A = ad bc 6= 0, there
are no other critical points. If det A = 0, then any nontrivial solution of the system ax + by = 0
and cx + dy = 0 is another critical point. The main purpose of this section is to make a complete
classification for the critical point (0, 0).
Theorem 2.2.1 (1) Let 1 and 2 be two real linearly independent eigenvectors of the matrix A,
with 1 and 2 being the corresponding eigenvalues. Then a fundamental matrix of the homogeneous
system (2.4) is given by

e1 t1 , e2t 2 .
(2) Let 1 = r + ii be a complex eigenvector, with 1 = r + ii the corresponding eigenvalue
(i 6= 0). Then a fundamental matrix is given by

er t (r cos i t i sin i t), er t(i cos i t + r sin i t) .
(3) If 1 = 2 = and is the only linearly independent eigenvector, then there exists another
vector , being linearly independent to and satisfying (A I) = , such that

et , tet + et
is a fundamental matrix.1
Proof (1) By Theorem 1.6.2, we know that
e1 t 1 , e2 t 2
1 When the eigenvalues are repeated and there is only one linearly independent eigenvector, the matrix A is not
diagonalizable. Or equivalently, A is similar to the matrix
!
1
.
0
In other words, there is an invertible matrix P such that

A=P
P.
38
are linearly independent, since 1 , 2 are. It remains to prove both are solutions of (2.4). Indeed,

d 1 t
e 1 = e1 t 1 1 = e1 tA1 = A e1 t 1 ,
dt
i.e., e1 t 1 is a solution. It can be proved for e2 t 2 similarly.
(2) By (1), we know that e(r +ii )t (r + ii ) is a solution of the homogeneous system (2.4). That is,

d (r +ii )t
e
(r + ii ) = A e(r +ii )t (r + ii ) .
dt
Equating the real and imaginary parts respectively in the last equation, we have

d r t
e (r cos i t i sin i t)
= A er t (r cos i t i sin i t) ,
dt

d er t (i cos i t + r sin i t)
= A er t (i cos i t + r sin i t) .
dt
Thus, er t (r cos i t i sin i t) and er t (i cos i t + r sin i t) are solutions. To show they are linearly
independent, by Theorem 1.6.2, we only need to show that they are linealy independent at t = 0. In other
words, it is sufficient to show that r and i are linearly independent.
Assume that there are constants c1 and c2 such that
c1 r + c2 i = 0.
Since
(2.5)
Ar
r r i i ,
Ai
i r + r i ,
which can be obtained by equating the real and imaginary parts of A(r + ii ) = (r + ii )(r + ii ),
multiplying A to both sides of equation (2.5), we obtain
(r c1 + i c2 )r + (i c1 + r c2 )i = 0.
Multiplying r to equation (2.5) and adding it to the last equation, we have
i c2 r i c1 i = 0.
Since i 6= 0, the last equation gives
c2 r c1 i = 0.
Combining this equation with (2.5), we can easily have
(c21 + c22 )r = (c21 + c22 )i = 0.
These imply c1 = c2 = 0. That is, r and i are linearly independent.
(3) Let us first show the existence of such vector . Since 1 = 2 = and satisfies
2 (a + d) + (a d b c) = 0,
we know that
(a d)2 + 4b c = 0 and =
a+d
.
2
2.2 LINEAR AUTONOMOUS SYSTEMS

These imply
(A I)2
a
c
b
d
39
2
(a )2 + b c
c(a ) + c(d )

b(d ) + b(a )
= 0.
2
(d ) + b c
Since is the only linearly independent eigenvector corresponding to the eigenvalue , there must be a
vector such that (A I) 6= 0. But (A I)2 = 0 implies that there exists a non-zero constant c such
1
that (A I) = c . Dividing by c and renaming to be , we have
c
(A I) = .
The linear independence of and can be shown as follows. Assume
+ = 0.
Multiplying A I to the left of the last equation, we have = 0, which implies = 0. This can be used
further to get = 0.
By (1), we know that et is a solution of (2.4). By a direct verification:

d t
te + et
dt
et + tet + et
tet () + et ( + )
tet A + et A

A tet + et ,
we know that tet + et is also a solution of (2.4). By Theorem 1.6.2, these two solutions are linearly
independent, since at t = 0, they are 1 , 2 which are linearly independent.
2
It is convenient to introduce the following concept.

dx
du
= Ax and
= Bu
dt
dt
are said linearly equivalent if there exists a non-singular matrix K such that x(t) = Ku(t).
Definition 2.2.2 Two first-order linear homogeneous autonomous systems
It can be seen easily that if two autonomous systems are linearly equivalent, then the coefficient
matrices are similar. In fact, since
du
dx
= K1
= K1 Ax = (K1 AK)u,
dt
dt
we conclude that
B = K1 AK,
i.e., A and B are similar.
Two linearly equivalent autonomous systems have similar structures at the critical point (0, 0),
since u(t) = Kx(t) indicates that solution curves only undergo stretching and rotating transformations.
40
dx
= Ax if and
dt
du
= Bu.
only if it is a stable (attractive) point for a linearly equivalent differential system
dt
Theorem 2.2.3 The origin is a stable (attractive) point for differential system
Proof By symmetry, it suffices to show that if (0, 0) is stable (attrative) for
dx
= Ax, then it is stable
dt
du
= Bu.
(attractive) for
dt
dx
du
Since
= Ax and
= Bu are linearly equivalent, there is a non-singular matrix P, such that
dt
dt

p11 p12
u(t)
x(t)
x(t)
=P
=
.
v(t)
y(t)
p21 p22
y(t)
Thus, for any t, we have
ku(t)k
=
=
u2 (t) + v2 (t)
(p11 x + p12 y)2 + (p21 x + p22 y)2
(p211 + p221 )x2 + (p212 + p222 )y2 + 2(p11 p12 + p21 p22 )xy
(p211 + p221 )x2 + (p212 + p222 )y2 + |p11 p12 + p21 p22 |(x2 + y2 )
p
M x2 + y2 = M kx(t)k ,
where
M =
2 max {p211 + p221 , p212 + p222 , |p11 p12 + p21 p22 |}.
Similarly, since x(t) = P1 u(t), there is a constant N such that

kx(t)k N ku(t)k .
dx
= Ax. For any > 0, there exists 1 > 0 such that if kx(0)k <
dt
1 , we have kx(t)k < /M for all t > 0. Take = 1 /N . If ku(0)k < , then kx(0)k N ku(0)k < N = 1 .
du
Hence ku(t)k M kx(t)k < M /M = . Therefore (0, 0) is stable for the system
= Bu.
dt
dx
= Ax. Then there exists a 1 such that whenever kx(0)k <
Suppose (0, 0) is attractive for the system
dt
1 , we have lim kx(t)k = 0. Take = 1 /N . If ku(0)k < , we have kx(0)k N ku(0)k < N = 1 . Hence
Suppose (0, 0) is stable for the system
t+
we have
0 lim ku(t)k lim M kx(t)k = 0,
t+
t+
which implies lim ku(t)k = 0. Hence (0, 0) is attrative for the system
t+
du
= Bu as well.
dt
By Theorem 2.2.3 and Definition 2.1.2, we know that linear equivalence does not change the
qualitative nature of the phase portrait for linear systems. For instance, if the origin is a neutrally
stable point for a linear differential system, then it will be a neutrally stable point for all its linearly
equivalent linear differential systems.
2.3 COMPLETE CLASSIFICATION FOR LINEAR AUTONOMOUS SYSTEMS
2.3
41
COMPLETE CLASSIFICATION FOR LINEAR AUTONOMOUS SYSTEMS
Theorem 2.3.1 For different cases, we have the following complete classification:
(1) If A has two linearly independent real eigenvectors, with 1 , 2 being the eigenvalues, then
(2.4) is linearly equivalent to the canonical form:
du
= 1 u,
dt
(2.6)
dv = v.
2
dt
(a) If 1 and 2 are both negative, then the origin is a stable nodal point, as in Fig. 2.6(a).
Furthermore, if 1 = 2 , then the origin is a stable star point as in Fig. 2.7(a).
(b) If 1 and 2 are both positive, then the origin is an unstable nodal point, as in Fig. 2.6(b).
Furthermore, if 1 = 2 , then the origin is an unstable star point as in Fig. 2.7(b).
(c) If 1 and 2 are of opposite sign, then the origin is an unstable saddle point, as in Fig. 2.8.
(d) If 1 < 0 and 2 = 0, then the origin is neutrally stable, as in Fig. 2.9(d).
(e) If 1 > 0 and 2 = 0, then the origin is unstable, as in Fig. 2.9(e).
(f) If 1 = 2 = 0, then the origin is neutrally stable, as in Fig. 2.10. In this case, A is the
zero matrix.
...
......
..
.. ....
..
..
... ..
..
..
..
.....
...
.
..
.....
.
..
...
.
......
........
..
.
.
.
.
.
...
......
..
.......
.
...
..
.
......
...
.
...
.....
.
..
...
...
..
....
.. .... ...
.
.
.
.
.
.
.
...
.. .. ..
. .
...........
...
.. .. ...
.. .. ..
. ....
...
...
.. ... .. .... .... ....
..........
...
.....
.. .. . .. .. ..
......
...
........
.. .. .. .. .. ...
....
......
...
.
.
....
.
.
.
.
.
.
.
....
.
.
...
.... .. .. .. ... .......
....
....
..
..... .. ... ........ ........
....
.......
..... ......
.. .. .. ....... .......
........
..... ....
.....
.... ... ...................................................................................
..... .....
......
....... .
...... ..... .................................................................................................
........
..........
.
.
.
.
.
.
.......................... ............................................................................................................................................................................................................................................
.
.
.
.......
.
.
.
.
.
.
.
.
.
..................
.
........................................... ...... ........ .......... ............
......
....
.
............................................................................................ ......
.........
... .........
.
.. ........ ... ... ..
.
.
.
.
.
.
.
.
.
.
....
....
. .
. ....
....
...... ... .. .. . ....
....
...
...... ... .. .. .. ...
....
....
...
..
....... ..... .... .... .... .... ....
.......
...
......
.....
...
.. .... .... ... .... ....
...
......
.
.
.
.
.
.
.
.
.
.
.
......
.. .. ..
...
.. ... ....
.......
.
.
.
.
.
.
.....
..
.. ... ..
.. .... ..
.
.
...
.. .. ...
.
.
...
...
....
..........
...
..
...
.......
......
........
...
..
....
..
..
.......
..
..
.........
...
..
...
.
...
........
..
..
...
... ...
.
..
... .
.
...
..
Case (a): stable nodal point if 1 , 2 < 0
.
...
......
...
.. ....
..
.. ..
...
..
.. ..
..
...
....
.
..
..
..
....
...
..
..
.
.
.
.
.
...
...
....
.....
..
.
.
.
.
.
.
...
.
.
...
.
.
.
.
....... .....
...
... ......... .....
.. ... ...
.
.
.
.
.
.
.
.
.
.
.
...
.. .. .
...
.. .. ...
.. . ..
...
.......
.. ... .. .... .... ....
.....
...
.....
..............
....
...
.. .. .. ... ... ...
.. .. . . .. ..
.......
....
....
...... .
.
.
.
.
.
.
.
.....
.
.
.
.
.
.
.
.
.
....
.
.
...
.... .. .. .. ... .......
....
....
...
..... .. ... ... ... .......
....
.....
..... ......
.........
.. .. .................
.
.....
.
.......
.... .. ...........................................................................
. ...
......
.......
....... ......... ........ .....................................................................................
. .
....... ..... ........................................
........
...........
............................. ................................... ........................... ....................... ........ ................... .....................................................................................
.
.
.
.
.
.
.
.
.
.
.
.
..................
.......
............................................ ..... ........ ......... ............
........
.... .....
............................................................................................. ......
...
.
..
..........
..... ...... .. .. .
.... ........
....
...... ......... .... ... .......
.
.
.
.
.
.
.
.
.
.
.
....
.
.
.... .
..
. . .
...
.
....
....... ..... .... ... .... .... ...
....
..........
..
.
.
.......
...
...
.. .... .... ... .... ...
.
..........
.
.
.
.
.
.
.
.
.
.
.
.
...
........
.. ... .... .... ... ....
......
.
.
.
..
...
. .
.. ... ...
...
.
......... ..
.. ...
.
... ....... ...
...
.. ..........
.
...
...
.....
..
..
..........
..
...
.
.
...
..
.
..
...
....
..
..
...
.
..
..
.......
..
...
..
... ...
.
...
.. ..
.
.
...
..
Case (b): unstable nodal point if 1 , 2 > 0
Fig. 2.6 A has two linearly independent real eigenvectors and 1 and 2 are distinct and of the same sign
(2) If A has two linearly independent complex eigenvectors, with 1 , 2 = r ii being the
eigenvalues, then (2.4) is linearly equivalent to the canonical form:
du
= r u i v,
dt
(2.7)
dv = u + v.
i
r
dt
42
....
.......
..
...
..
.....
..
..
....
...
.
...
.
...
...
.
.......
..
...
...
........
...........
.
...
...
......
....
...
.....
...... .
.
..
. .......
......
.. .... ....
..........
............
...... .... ... ... ..........
...... ... ... .. .......
...... .. . .. .....
... ... ...
......................................................................................................................................................................................................................................
............ ......
...... .... ..... .... ...........
.
.
.
.
.
......
..
..........
... .... ....
.......
.........
..........
... ... ....
......
...... .
..
..
...
..
.......
..
.
.
.
.
.
.
.
.
.
.......
......
..
.
.
.
.
...
......
..
.
.
...
.
....
.
...
...
..
...
....
...
...
...
....
..
Case (a): stable star point if 1 = 2 < 0
....
.......
..
...
..
.....
..
..
....
...
.
...
.
..
..
.
...
..
..
..
...
....
.
....
.
.
....... ...... ........
......
..
....
.......
.
..
......
......
.. .... ....
......
.........
.
......
. ....... .... .... ... ................
...... .. .. ... ......
...... .. .. .. ......
.. . ..
..................................................................................................................................................................................................................................
..... ...........
...... ... .... ... ...........
.
.
.
.
.
.
.
........
. ..
..
... ... ..
..........
............
.......
......
.. .... ....
......
......
....
...
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
.......
....
.......
.
..
....
..
.
.
.
...
..
....
.
...
.
.
.
.
.
.
.
....
....
...
...
...
...
...
....
..
Case (b): unstable star point if 1 = 2 > 0
Fig. 2.7 A has two linearly independent eigenvector but repeated eigenvalues
....
........
...
...
......
...
......
.......
.
...........
........
....
...
............. .... ........................
....
.
.
..... ...
.
.
................... .
.... .....
...
.
.
.
.
.....
.
..... ....... ...
... .....
.
.
.
.
.
.
.
.
.
.
.........
....
...
...
... .....
...
..............
.........
...
... ...... ......
.......................................
..
.... ....
....
...
...
..
...
.....
....
..
...
..
....
..
.
..
.
...
.
....
.........
..
.
..
..
.
.
.
.
...
.
.....
.
.
.
.
.
.
.
.
.
.
...
..
.
......
......
..
....
....
....
..
..
...
.........
....
.
....
..
...
..........
.... .... ......
..
..
..
...
.... .. ....
..
..
..
.
.
.
.
.
.
.. . .
.
.
.
...................................................................................................................................................................................................................
...
.. .. ..
...
.
.
..
..
.... .. ....
..
..
..
.... .... ......
..
.
.......
.
..
.
.
..
....
.
...
..
.......
.
.
...
.
.
.
.
.
.
.
.
....
.......
.
..
.
...
.
.
.
.
.
.
........
.
.
.
.
...
....
.....
..
..
..
.
..
.
.
.
.
.
.
.
.
.
.
......
...
..
.
..
.
..
.
.
.
.
.
.
..
.
.
.
.
....
...
.
......
..
..
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...
..
...
..
................. ... ......................
..
..
.
.
.
.
.
.
.
.
.
.
.
.
.... ...
.........
...
.
.
.. .....
.
.
.
.
.
.
.
.
.
.
.
.
......
... ...
.
....
... .....
.
.
.
.
.
.
.
.
.
.
.....
.
..
.... ...
..................................... .
.... .......
.
.....
.............
........
....
. ........
....
.
.........
......
..
........
...
......
...
....
..
...
.
Case (c): unstable saddle point if 1 and 2 are of opposite sign

Fig. 2.8 A has two linearly independent real eigenvectors and 1 and 2 are of opposite sign
(a) If r < 0, then the origin is a stable focal point, as in Fig. 2.11(a).
(b) If r > 0, then the origin is an unstable focal point, as in Fig. 2.11(b).
(c) If r = 0, then the origin is a stable vortex point, as in Fig. 2.12.
.
...
.......
..
...
....
.. ...............
...............
.
.
.
.
.
..
.....
....... ....
........
......
.........
...
......
......
.
.
.
.
.
.
.
.
.
.
.
.
.
.........
......
......
........
.......
......
.........
...... ..
......
...... ....
......
......
......
......
......
.
.
.
.
...
.
.
.
.
.
.
.
.
.
.
.
.
.
..
..
. ......
...
...........
.......
..........
......
......
..........
......
.......
......
......
........
...... ....
.
.
.
.
.
.
.
.
.
.
.
.
....
.
.
.
.
.
.
.
.
..
..
....
....
. ......
..... ............
......
......
.........
......
. ...
......
......
.....................................................................................................................................................................................................................................
.
.
...
...... ..
. ......
......
......
..........
...... ....
......
......
......
......
.........
......
.......
......
..... ...........
.
.........
.....
......
.
.
.
.
.
......
......
.
.
.
.
.
.
.
.
.
..
..
........
.......
...... ...
......
........
...........
....... ...
......
......
........
. ............
......
......
.
.
.
.
.
.
.
.
.
.
......
.
.
..
.....
.. ......
.......
......
.......
............
.......
...... ..
........
...... ....
......
......
......
......
.
.
.
.
.
.
.
.
.
.
.
.
..
...
.............
............
........
.......
......
...... ..
...... ....
.
.
.
.
.
.
......
.....
........
.......
..
...
43
....
.......
.
..
.......
...
......
.. .................
...............
.....
......
.....
...........
...... ..
...... ....
........
.
.
.
.
.
.
.
.
.
.
....
..
....
.... ...........
.......
......
...
..........
......
........
.........
......
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... ...
....
.....
.....
......
.......
......
......
.... ...........
...........
........
.......
. ..
........
......
..........
.........
......
......
.
......
.
.
.
.
.
.
.
.
....
.
.
.
.
.
.
.
.
.
.
.
.
.
...
......
.... .
......
........
........
...... ....
......
........
......
.........
... ...........
...... .
.....
......
.
.
.
.
.
.
.
.
.
.
...
.
.
.
.
.
.
.
.
............................................................................................................................................................................................................................
........
..... ..
.....
.........
.......
......
....... ....
.
......
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
...
.....
.....
.......
..... ...........
.......
......
.........
......
.
..........
........
.........
......
........
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... ...
......
......
.......
......
......
......
.........
.... ...........
......
..........
. ..
......
......
......
.........
.
.
.
.
.
.
.
.
.
.......
.
..
.....
.......... ....
......
... ...........
..........
.....
...........
......
.
.
.
.
.
.
.
..... .
......... ....
......
..
......
....
..
...
Case (d): neutrally stable if 1 < 0 and 2 = 0 Case (e): neutrally unstable if 1 > 0 and 2 = 0
Fig. 2.9
A has two linearly independent real eigenvectors and one of the eigenvalues is zero
.
...
.......
...
...
....
..
...
....
..
....
..
...
....
..
...
....
..
....
....................................................................................................................................................................................................
...
..
..
.....
..
...
...
...
...
...
...
...
...
...
...
...
...
....
.
.....................
.....................
.....................
.....................
.....................
.....................
.....................
.....................
.....................
.....................
.....................
.....................
.....................
.....................
.....................
.....................
.....................
.....................
.....................
.....................
.....................
Case (f): neutrally stable point if 1 = 0 and 2 = 0

Fig. 2.10 A = 0
(3) If A has only one repeated eigenvalue and if there exists only one linearly independent
eigenvector, then (2.4) is linearly equivalent to the canonical form:
du
= u,
dt
(2.8)
dv = u + v.
dt
(a) If < 0, then the origin is a stable nodal point, as in Fig. 2.13(a).
(b) If > 0, then the origin is an unstable nodal point, as in Fig. 2.13(b).
44
....
.......
..
...
..
....
.....
...
.
.....
....
.
..
.
......
.
..
......
.
.
.
.
.
.
...
.....
....
..
.. .......
..
... .....
..
... ....
...
.
.........................................
....
.........
.......
.......
......
.......
....
....
... ... ....... ........................
...
.
.....
... .... ....... ..........
...
....
...
. .....
...
..
.
.
...
. .... ..................
..
..
.
.
...
...
. ..
...
...
..
.
.
.
...............................................................................................................................................................................................................
...
.
...........
...
..
.......
. .
....
..
...
.... ..
.........
...
...
..... ...... ........
.
..
...
. ...
.............
..
.
.
.
.
.
.
.
.
..
.....
.
.... ..... ....
.
.
.
.
..........
.
.
.
.
.
....
.
.
.....................
.. ...
.
.
.
.
.
.
.
.
.
..
..
.
.
.
.
.
.
.
.
.
.
..
......
..
....
....
.
...
.....
...
...
..
......
..
...
.............
..
.
.
.
.
.
.
.
.
.
.
.
.
..
..........
.... .....
.................
... ..
...
...
....
..
...
Case (a): stable focal point if r < 0
....
.......
..
...
..
....
.
.....
..
.....
.
....
.
..
.
.....
.
............
...
...
.
.
....
....
..
.. ......
..
..
... .....
..
.
.
.
.
.
.
.
....................................
..
......
...........
......
..
......
........
....
..
.. ..
.....
....
...
... .... ...............................................
.........
.
...
...
....
.. .......
...
.
.
...
..
. .... ..................
..
.
.
.
...
.
...
...
..
..
.
... .
..................................................................................................................................................................................................................
........ ..
..
..
...
...
..
..
.. ...
....
..
.
...
.
.
.
.
.
.
.
.
............
.
...
..
.... .. ..
....
....
..
.... ......
.
.
.
.........
.
.
.
...
.
.... .........
.
.
. .........
.
.
.
.
.
.
.
.
.
........................
.
..
.. ...
.
.
.
.
.
.
.
.
.
.
...
.... ...
..
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
....
..
..
....
. .....
..
..
............
...
...
......
... ....
.........
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..............
.... ...
...
...
....
..
...
Case (b): unstable focal point if r > 0
Fig. 2.11 A has two linearly independent complex eigenvectors

....
.......
...
..
....
....
..
.
.
.
.
............. ............................
.
.
.
.
.
.
.
.
.......
.
......
.
....
.
...
.
..
......
.
.
..
.
..
.......
.......................
...
...
........ .... ............
.
...
...
.
.
.
.
... .
.
...
.
...
..
.
.
.
.
.
.
.
.......
..
..
..
.
...
.
.
..
..
..................
..
.
.
.
.
.
.
...
.
.
..
..
.
... ... ..........
.
..
.
.
.
..
..
.
..
...
.
....
....
.
.
.
.
..
..
.
.
..
.
.
.
.
.
.
.
.
........................................................................................................................................................................................................
...
...
..
..
.
..
...
.
.
.
..
..
.
.
.
.
.
...
..
.
.
.
...
.
.
.
.
.
.
.
.
.
.
.
.
.
...
...... .. .....
..
.
..
.............
..
...
..
..
..
...
...
...
..
..
.....
...
...
...
...
.....
....
.
.
.
.
.
.
...
.
.
.
.
.
.
.
.............................
....
...
....
.....
....
....
....
.....
....
.......
...
.......
...........
.
.
.
.
.
.
.
.
.
.
.
............................
...
..
...
...
....
..
....
.
Case (c): neutrally stable vortex point if r = 0

Fig. 2.12 A has two linearly independent complex eigenvectors
(c) If = 0, then the origin is unstable, as in Fig. 2.14.

Proof
(1) Assume that 1 and 2 are two linearly independent real eigenvectors. Take K = (1 , 2 ) and denote

u
x
u=
= K1
= K1 x.
v
y
Since
AK = (A1 , A2 ) = (1 1 , 2 2 ) = K
1
0
0
2
....
.......
...
..
...
..
...
...
... ....
..
...
.
..
.
.
...
... ...
..
.
..
.. ..
...
.
..... ..
.......
........
.....
..
...... ...
.. ..
...
..
...
.. ..
..
..
...
.
.
.
.
..
... ..
.
...............................
...
...
.....
.. ...........
...
...
...
........
.....
...
........
.......
.
.....
...
.
..... ....
.
.
.
.
.
.
.
.................................. .... ....
..
..
..........
........... .. .....
...
..
......
... . .
..
..
....
.................................. .......
...
..
.....
.........
...
..
.
.
.
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
......................................................................................................................................................................................................................
..
.......
.
.
..
.. .........................
..
.
..
.
.
.
.
.
.
.
...
.
...
.
..
..
.... .... ................................
....
..
...
... ... ... ..............................................
..
..
.......
.. .. ... .....
..
......
..
.......
.....
.....
..
.......
.....
.
..
.........
.
..
.....
...
........... ...
...
....
.........................
...
...
.......
..
..
...
...
..
....
..
.........
....
..
.........
...
.
...
.......
..
...
... ...
..
..
... ..
...
..
... ...
... ...
... .
....
..
....
.......
...
..
...
..
...
...
... ....
..
...
.
..
.
.
...
... ...
..
...
..
.. ..
.
.
.
.
.
.
.
.
......
..... ..
.......
..
...
.
...
.
.. ..
..
...
...
.
.
..
.. ..
..
.
.
.
.
.
.
..
... ..
................................
..
...
..
.. ...
..........
.........
..
........
.....
...
..
.......
..
....
...
..... ....
..
..
..... .. ...
.
.
.
.........
.
.
..
............ .......................... .... ......
..
..
......
..
... . ..
.
..
....
............... .. ...
..
...
...
....... ......................................
..
..
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
..............................................................................................................................................................................................................................
.
.
. ...............
..
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
.
.
. .....
.
...
..
..
..
..
.... .... .................................
.....
...
..
... ... ... .............................................
..
..
... .. ... .....
..
..
..
......
........
.. ....
.....
..
.......
.
....
...
........
.
..
...........
.....
...
...
..........................
...
....
...
..
..
..
....
..
....
........
...
...
....
....
...
.......
.
.......
...
.........
..
...
.......
..
... ...
..
...
..
... ...
... ...
...
...
..
Case (a): stable nodal point if < 0
Case (b): unstable nodal point if > 0
Fig. 2.13 A has only one linearly independent eigenvector (repeated eigenvalues) and 6= 0
..
.......
....
..
..
....
...
...
.... ... ..........
.. ..........
. ..
....
.........
.....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
....... ... ............
..
. ......
......
.
.
.
......
.
.
.
.
..
............
.......
......
.......
...... ....
......
.
.
.
.
.
.
.
.
.
.
.
.
.
.......
.
.
.
..
...
.....
...........
... ......
......
.............
......
......
......
.
......
...... .
..........
......
.
.
.
......
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.....
.. .......
.........
......
...........
.................................................................................................................................................................................................................................
.
...
.....
....... ..
......
.............
.........
...... .... ...........
.
...... .
........
......
......
.. .....
.......
......
.......
..........
......
.
................
.
.
.
..........
.
.
.
.
.
.
.
.
... . .
.... ... ..........
.
.
..... .
.
.
.
.
.
.
.
.
...
.. .....
......
............
.......
......
.........
........
......
...... .... ...........
.
...............
.
.
.
.
.
.
.
.
.
.....
.....
.......
............... ....
........
....
..
...
....
..
....
..
Case (c): unstable point if = 0

Fig. 2.14 A has only one linearly independent eigenvector (repeated eigenvalues) and = 0
we have
dx
du
= K1
= K1 Ax = K1 AKu =
dt
dt
This is the desired canonical form.2
2 The
identity
A=K

1
0
0
2
K1 .
1
0
0
2
u.
45
46

(1a) If 1 and 2 are both negative, by solve the canonical equation, we have (u(t), v(t)) = ae1 t , be2 t .
Thus,
m = 2 /1 > 0.
u = cvm ,
In (u, v)-plane, this is a family of parabola-like curves. Since as t +, (u(t), v(t)) (0, 0), we know
the directions are towards the origin. Together with the linear transformation K, which generally gives
a stretching and/or rotating for the solution curves, we have the schematic figure of solution curves in
(x, y)-plane, as in Fig. 2.6(a).
(1b) In the case of both 1 , 2 being positive, it is the same as in (1a), except that we have to switch the
orientations of the solution curves, as in Fig .2.6(b).
A special case in (1a) and (1b) is that 1 = 2 . We have u = cv, a family of straight lines passing through
the origin. Correspondingly, as in Fig. 2.7, the phase diagram is a so-called star: the origin is stable if < 0;
unstable if < 0.
(1c) If 1 and 2 are of opposite sign, then similarly we have
uvm = c,
m = 2 /1 > 0.
For different values of c, this gives a family of hyperbola-like curves, as in Fig 2.8. Hence, in this case,
we have a saddle point. As t +, either u(t) or v(t) approaches +. Thus, a saddle point is always
unstable.

(1d) In this case, the solution of the canonical equation is (u(t), v(t)) = ae1 t , b . Thus, we have
a family of straight lines, since for fixed b, the solution curve always has v(t) = b. Since as t +,
(u(t), v(t)) (0, b), we conclude that the origin is stable but not strictly stable, as in Fig. 2.9(d).
(1e) This is the case similar to (1d), except the orientation is opposite. Thus, the origin is unstable, as
in Fig. 2.9(e).
(1f) In the case of 1 = 2 = 0, the solution of the canonical equation degenerates to a single point
(u(t), v(t)) = (a, b). The origin is stable but not strictly stable, as in Fig. 2.10.
(2) Assume r + ii is an eigenvector corresponding to the eigenvalue r + ii . In the proof of Theorem 2.2.1, we know
Ar = r r i i ,
Ai = i r + r i ,
It is also proved there that r and i are linearly independent. Take K = (r , i ) and denote u = K1 x.
Since

r i
,
AK = (Ar , Ai ) = (r r i i , i r r i ) = K
i r
similarly we have
du
dx
= K1
= K1 Ax = K1AKu =
dt
dt
This is the desired canonical form.
(2a) Using the canonical form, we have
r
i

i
u.
r
du
du
d(u2 + v2 )
= 2u
+ 2v
= 2u(r u i v) + 2v(i u + r v) = 2r (u2 + v2 ),
dt
dt
dt

1 0
which is called the Jordan canonical form.
0 2
In fact, P = (1 , 2 ), formed by the linearly independent eigenvectors.
indicates that A is similar to this diagonal matrix:
47
and
d v
dv
du
u
v
v
d
dt
u
dt
dt = (i u + r v)u (r u i v)v = .
arctan
=
v 2 =
i
dt
u
u2 + v2
u2 + v2
1+
u
Solve these two equations, in the polar coordinates, we have
r = (u2 + v2 )1/2 = er t ,
= arctan
v
= i t + ,
u
where 0 and are arbitrary constants. Thus, the phase diagram consists of contracting spirals surrounding the origin if r < 0, as shown in Fig 2.11(a). Thus, the origin is a stable focal point.
(2b) If r > 0, by the discussion in (2a), the phase diagram consists of expanding spirals from the origin,
as shown in Fig 2.11(b). The origin is an unstable focal point.
(2c) If r = 0, then by the discussion in (2a), we have a family of closed curves. In this case, the origin
is a neutrally stable vortex point, or sometimes it is called a center, as in Fig. 2.12.
(3) Assume is the only linearly independent eigenvector corresponding to the eigenvalue . By Theorem 2.2.1, there is a vector , linearly independent to , such that A = + . Take K = (, ) and
denote u = K1x. Since

0
AK = (A, A) = (, + ) = K
,
1
similarly we have
dx
du
= K1
= K1 Ax = K1 AKu =
dt
dt

0
u.
We obtain the desired canonical form.

(3a) The solution of the canonical equation is (u(t), v(t)) = aet , atet + bet , where a and b are
arbitrary constants. Thus, if < 0, we know that the origin is a stable nodal point, as in Fig 2.13(a).
(3b) If > 0, as in (3a), we know that the origin is an unstable nodal point, as in Fig. 2.13(b).
(3c) if = 0, the solution of the canonical equations degenerates to (u(t), v(t)) = (a, at + b), which is a
family of straight lines, as in Fig 2.14. This is a case that the origin is unstable.
2
By Theorem 2.3.1 and Theorem 2.2.3, we immediately have the following results.
dx
Corollary 2.3.2 The critical point (0, 0) of a linear homogeneous differential system
= Ax is
dt
attractive if and only if every eigenvalue of A has a negative real part. In this case, the origin is
strictly stable.
Example 2.3.1 Consider the system x0 = x + 3y and y0 = y. Find the canonical form of this
system such that both are linearly equivalent.
Solution The coefficient matrix
A=
1
0

3
,
1
which has two distinct real eigenvalues 1 = 1, 2 = 1. The corresponding eigenvectors

1
3
1 =
,
2 =
0
2
Take
K = (1 , 2 ) =
1
0
3
2
48
and define

1
u
1 x
=
=K
y
32
v

x
x
.
=
1
y
32 x + 12 y
2
Then the differential system for (u, v) is

d u
u
1
= K1 AK
=
0
v
dt v

u
0
,
v
1
which is a canonical form.
Example 2.3.2 Consider the system
dx
dt
dy
dt
x + 2y,
2x.
where is a constant. Find the critical values of where the change of the qualitative nature of the
phase portrait for the system occurs. Classify the autonomous system.
Solution The eigenvalues satisfy the characteristic equation
det

2
= 2 + 4 = 0,
whose solutions are
2 16
,
2
We consider the following three different cases:
1 =
2 =
2 16
.
2
(1) 2 16 < 0;
(2) 2 16 = 0;
(1) 2 16 > 0.
In the case (1), we have || < 4 and the coefficient matrix has two complex eigenvalues, with the real
part being 2 . By Theorem 2.3.1, for 4 < < 0, the origin is a stable nodal point; for 0 < < 4, the
origin is an unstable nodal point; at = 0, the origin is a stable vortex.
In the case (2), for both = 4 and = 4, the eigenvalue is repeated and there exists only one
eigenvector. By Theorem 2.3.1, the origin is an unstable critical point.
In the case (3), we have || > 4 and the eigenvalues are real, distinct and of the same sign. Their signs
are the same as that of . Thus, by Theorem 2.3.1, the origin is a stable focal point if < < 4; an
unstable focal point if 4 < < +.
2
2.4
LIAPUNOV DIRECT METHOD
The classification of critical points of linear homogeneous autonomous differential systems in the
previous section provides us a complete understanding on the local structures of such systems.
However, local structures in the phase space for general differential systems are too complicated and
it is beyond the scope of this course.
2.4 LIAPUNOV DIRECT METHOD
49
We may try to generalize the idea of so-called linear equivalence. As we knew, a linear equivalence is a simple stretching and rotating transformation. This will not change the nature of local
structures, as shown in Theorem 2.2.3. Obviously, for the purpose of preserving local structures,
there are other transformations. In the phase space, we can stretch the solution curves a bit in some
regions while squeeze in some other regions. We can also twist a little bit of the solutions curves.
Such nonlinear equivalence is called diffeomorphism. However, finding a diffeomorphism between
two systems is not trivial. Actually, it is practically useless to classify critical points for nonlinear
autonomous systems in terms of diffeomorphisms.
One practically useful tool in the study local structures for nonlinear autonomous systems is
Liapunov functions. Let us first use the following examples to demonstrate the basic ideas.
Example 2.4.1 Show that the critical point (0, 0) of the plane autonomous system
dx
= y,
dt
dy = x (x2 1)y,
dt
is stable for any < 0.
Solution Let (x(t), y(t)) be a solution of the system for t 0. Consider V (x, y) = x2 + y2 . Then
d
V (x(t), y(t))
dt
dx(t)
dy(t)
+ 2y(t)
dt
dt

2x(t) y(t) + 2y(t) x(t) (x2 (t) 1)y(t)
2(x2 (t) 1)y2 (t) 0,
2x(t)
if |x(t)| 1. That is, the function V (x(t), y(t)) is a decreasingpfunction along any solution curve with
|x(t)| 1. For any given > 0, we choose = min{1, }. Thus, if x2 (0) + y2 (0) < , then
x2 (t) + y2 (t) = V (x(t), y(t)) V (x(0), y(0)) = x2 (0) + y2 (0) 2 2 .
In other words,
kx(0)k < kx(t)k < .
Therefore, the origin is a stable point of the system.
It can be shown that, in fact, the origin is a strictly stable in the previous example. This will be
seen later.
Example 2.4.2 Show that the critical point (0, 0) of the regular plane autonomous system
dx
= y x3 ,
dt
dy = 2(x3 + y3 ),
dt
is stable.
Solution Let (x(t), y(t)) be a solution of the system for t 0. Choose V (x, y) = x4 + y2 . Consider
V (t) = x(t)4 + y(t)2 .
dV (t)
= 4x(t)3x0 (t) + 2y(t)y0 (t) = 4(x(t)6 + y(t)4 ) 0.
dt
50
Hence V (t) is a decreasing function. Therefore when t > 0, V (t) V (0).

p
2
Given > 0, take = min{1, }. Whenever x(0)2 + y(0)2 < , we have
2
x(t)4 + y(t)2 x(0)4 + y(0)2 x(0)2 + y(0)2 < 2
since x(0)2 2 < 1. Now we have
(x(t)2 + y(t)2 )2
where we used |y(t)| < < 1. Hence
x(t)4 + y(t)4 + 2x(t)2 y(t)2
2(x(t)4 + y(t)4 )
2(x(t)4 + y(t)2 ) 22 4
x(t)2 + y(t)2 < . Thus (0, 0) is a stable point.
The above examples indicates a new approach to study stability of nonlinear autonomous systems.
Before we generalize the idea of the above examples, we should clarify one point. Since existence
and uniqueness are not always guaranteed for nonlinear systems, we have to make sure that the
solutions for the discussion of stability do exist. It is convenient to introduce the following definition.
Definition 2.4.1 A plane autonomous system (2.1) is said to be regular if for any initial condition
x(t0 ) = x0 , there exists a unique solution in the neighborhood of (x0 , t0).
By Theorem 1.5.1, we know that if X(x) is continuously differentiable for all x, then the nonlinear
system is regular.
In the two examples above, the existence of the function V (x, y) is the central ingredient of the
method. Let (x(t), y(t)) be a solution of (2.1). For a given function V (x, y), denote
d
V (x(t), y(t)) = Vx (x(t), y(t))X(x(t), y(t)) + Vy (x(t), y(t))Y (x(t), y(t))
dt
to be the total derivative of V with respect to (2.1).
We introduce the following definition.
Definition 2.4.2 A function V (x, y), defined in a neighborhood of the origin, is called a weak
Liapunov function for the plane sutonomous system (2.1) if
(1) V (x, y) is continuously differentiable and V (0, 0) = 0.
(2) V (x, y) is positive definite, in other words, V (x, y) > 0 if (x, y) 6= (0, 0).
(3)
V (x, y) X(x, y) +
V (x, y) Y (x, y) 0,
x
y
for (x, y) 6= (0, 0).
It will be called a strong Liapunov function if the above condition (3) is changed to
(3)0
V (x, y) X(x, y) +
V (x, y) Y (x, y) < 0,
x
y
51
for (x, y) 6= (0, 0).
Theorem 2.4.3 If there exists a weak Liapunov function for the regular plane autonomous system
(2.1), then the critical point (0, 0) is stable. If there exists a strong Liapunov function, then the
critical point (0, 0) is strictly stable.
y..........
...
...
...
....
..
...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...............................
.
.
..........
.
.
.
.........
.
.
.
.
...
.
........
......
.
.
.
...
.
.
.....
...
.
.
.....
.
...
.
.....
....
.
.
.
.
.
....
...
..
.
.
.
....
.
..
.
.
...
.
.
.
..
...
.
.
.
.
...
......................................
..
.
.
.
.
.
.
.
.
.
.
......
...
.
....
.
.
..
.
.
.
.
.
.
.
..
....
...
..
.
.
..
.
...
.
.
.
....
..
..
...
.
...
.
.
.
..
...
.
.
.
.
.
...
..
.
...
DR...
..
..
..
.
.
.
..
..................
.
..
.
.
.
.
.
D
.
...
.
.
r...
..
..
. .....
..
D
.
.

.
.
..
.
...
.
...
...
..
....
..
...
...
.
..
.
....
.
.
.
.
.
.
.
.
.......................................................................................................................................................................................................................................................................
...
.
...
.
...
.
..
.
.
.r
.
..
.
.
.
.
R
.
.
.
.
.
.
.
...
....
..
.
. .....
.
.
.
.
.
...
.
.
....... . ......
..
..
..........
..
..
..
.
..
...
.
..
.
.
.
.
.
.
.
.
.
.
.
.
...
.
... .......... ....... ...
.
..
.. ..
....
..
..
.............
... ..
..
..... ... .........
...
....
..
... .... ........
..
.. ...
..... t0
t .=
...
....
.............
..
.
.
.
.
.
.
.
.
.
.......... ....
...
.
.....................................
...
...
...
...
...............
....
....
...
...
...
.
.
.
....
.
.
.....
...
....
....
.....
...
.......
......
........
......
....
...........
........
.
.
.
.
.
.
.
.
.
.
.
............................................
..
...
...
....
..
...
....
.
Fig. 2.15 A typical solution curve near the origin
Proof For each constant a > 0, we denote Da the circular open disc of radius a > 0 centered at the origin.
For any fixed R > 0, since the weak Liapunov function V is continuous on the boundary DR , which is a
bounded closed set, V has a positive minimum m on DR . Since V is continuous and V (0, 0) = 0, there is
a positive number r < R such that V (x, y) < m for (x, y) Dr . Let (x(t), y(t)) be any solution of (2.1) and
d
V (x(t), y(t)) 0, we know that
(x(t0 ), y(t0 )) Dr . Since
dt
V (x(t), y(t)) V (x(t0 ), y(t0 )) < m,
for t t0 .
This implies that the solution curve can never reach the boundary DR . Hence, the origin is stable.
To show that the origin is strictly stable if V is a strong Liapunov function, we only need to show
that limt+ V (x(t), y(t)) = 0. Since V (x, y) is positive definite and V (0, 0) = 0, this limit implies that
(x(t), y(t)) (0, 0), or the origin is strictly stable. In fact, since
d
V (x(t), y(t)) =
V (x(t), y(t)) X(x(t), y(t)) +
V (x(t), y(t)) Y (x(t), y(t)) < 0,
dt
x
y
we know that V (x(t), y(t)) is decreasing in t. The condition that V (x(t), y(t)) is bounded below by 0 implies
that the limit limt+ V (x(t), y(t)) exists, say L. If L > 0, since V (0, 0) = 0, we take a small > 0 is so
52
that V (x, y) < L in D . In the ring regin bounded by D and DR , the function
V (x, y) X(x, y) +
V (x, y) Y (x, y)
x
y
has a negative maximum k, for some k > 0. Since the solution curve (x(t), y(t)) remains in this ring region,
we have
Z t
d
V (x(t), y(t)) V (x(t0 ), y(t0 )) =
V (x(t), y(t))dt k(t t0 ),
t0 dt
for t t0 . This inequality implies V (x(t), y(t)) , which contradicts to the fact V (x(t), y(t)) 0.
Thus, we conclude L = 0.
2
Theorem 2.4.4 If the critical point (0, 0) of the linear plane autonomous system
dx
dt
dy
dt
ax + by,
cx + dy,
is strictly stable, then so is that of the perturbed regular system
dx
dt
dy
dt
ax + by + (x, y),
(2.9)
cx + dy + (x, y),
provided there exist k > 0 and M > 0 such that whenever x2 + y2 < k we have |(x, y)| + |(x, y)|
M (x2 + y2 ).
Proof For simplicity, since the linear homogeneous system is strictly stable at the origin, by Theorem 2.3.1,
we can assume that it is one of the types
dx
dx
= 1 x,
= r x i y,
dt
dt
1 < 0, 2 < 0;
r < 0;
(2)
(1)
dy = i x + r y,
dy = 2 y,
dt
dt
or
dx
= x,
dt
< 0.
(3)
dy = x + y,
dt
Otherwise, we can transform the linear system to its canonical form via a linear equivalence transformation.
The new perturbed system still satisfies the conditions in Theorem 2.4.4 after such transformation.
Take V (x, y) = x2 + y2 for the first two cases above and V (x, y) = x2 + 2 y2 for the last case. To show
that V (x, y) is a strong Liapunov function, we only need to show that there exists a negative constant < 0
such that
V (x, y) X(x, y) +
V (x, y) Y (x, y) V (x, y),
x
y
for sufficiently small (x, y).
53
In fact, in the case (1), we have
V (x, y) X(x, y) +
V (x, y) Y (x, y)
x
y
2x (1 x + ) + 2y (2 y + )
2(x2 + y2 ) + 2(x + y),
2
}, since
with = max{1 , 2 }. Similarly to the case (1), we can show that, for x2 + y2 < = min{k,
16M 2
and |y|
, we have
|x|
4M
4M

|x + y| 2
M (x2 + y2 ) = V (x, y),
4M
2
which implies
V (x, y) X(x, y) +
V (x, y) Y (x, y) V (x, y),
x
y
whenever x2 + y2 < .
In the case (2), we have
V (x, y) X(x, y) +
V (x, y) Y (x, y)
x
y
2x (r x i y + ) + 2y (r x + i y + )
2r (x2 + y2 ) + 2(x + y).
By the condition,
|x + y| |x| || + |y| || (|x| + |y|) M (x2 + y2 ),
r
2r
r
and |y|
, we
whenever x2 + y2 < k. Thus, when x2 + y2 < = min{k,
}, since |x|
16M 2
4M
4M
have

r
r
M (x2 + y2 ) = (x2 + y2 ).
|x + y| 2
4M
2
Hence,
V (x, y) X(x, y) +
V (x, y) Y (x, y) 2r (x2 + y2 ) + (r )(x2 + y2 ) = r V (x, y),
x
y
In the case (3), V (x, y) = x2 + 2 y2 and
V (x, y) X(x, y) +
V (x, y) Y (x, y)
x
y
=
2x (x + ) + 22 y (x + y + )
(x2 + 2 y2 ) + (x + y)2 + 2(x + 2 y)
(x2 + 2 y2 ) + 2(x + 2 y).
By the condition,
|x + 2 y| |x| || + 2 |y| || (|x| + 2 |y|) M (x2 + y2 ),
54

|x|
we have
|x + 2 y|
,
8M (1 + 2 )(1 + 2 )
2
}, since
64M 2 (1 + 2 )2 (1 + 2 )2
|y|
,
8M (1 + 2 )(1 + 2 )
(1 + 2 )(|x| + |y|)) M (1 + 2 )(x2 + 2 y2 )
(1 + 2 )(x2 + 2 y2 )
(1 + 2 ) 2
8M (1 + 2 )(1 + 2 )
2
(x + 2 y2 ).
4
Hence,
2
V (x, y) X(x, y) +
V (x, y) Y (x, y) (x2 + 2 y2 ) +
(x + 2 y2 ) = V (x, y),
x
y
2
2
Now we can apply Theorem 2.4.4 to Example 2.4.1. The system in Example 2.4.1 is perturbed
from the linear system
dx
= y,
dt
dy = x + y.
dt
The origin is strictly stable for the linear system, since the coefficient matrix

0
1
has the eigenvalues
p
2 4
,
2
which both have negative real parts for < 0.

Example 2.4.3 Show that for the differential system
dx
dt
dy
dt
y,
1
{f(x, y)y + x},
m
the critical point (0, 0) is stable, where m > 0, > 0 and f(x, y) 0 in a neighborhood of the origin
Solution In fact, take
V (x, y) =
1
(x2 + my2 ).
2

Then
=
=
55
V (x, y) X(x, y) +
V (x, y) Y (x, y)
x
y
1
xy my [f (x, y)y + x]
m
my2 f (x, y) 0.
By Theorem 2.4.3, the origin is stable.
Example 2.4.4 Discuss the dependence on the sign of the constant of the critical point at the
origin of the regular system
dx
= y + x5 ,
dt
dy = x + y5 .
dt
Solution When = 0, the system becomes x0 = y and y0 = x and the origin is a neutrally stable vortex
point.
Suppose 6= 0. Let V (x, y) = x2 + y2 . Then
V (x, y) X(x, y) +
V (x, y) Y (x, y)
x
y
=
2x(y + x5 ) + 2y(x + y5 )
2(x6 + y6 ) < 0,
for (x, y) 6= (0, 0), since < 0. Thus V (x, y) is a strong Liapunov function. By Theorem 2.4.3, the origin is
strictly stable.
d
V (x(t), y(t)). Then the previous calculation gives V 0 (t) 0. Hence V (t) is
If > 0, denote V 0 (t) =
dt
2
increasing. Therefore x(t) + y(t)2 x(0)2 + y(0)2 for t > 0. Since
(x2 + y2 )3
x6 + y6 + 3x4 y2 + 3x2 y4
x6 + y6 + 3x2 y2 (x2 + y2 )
x2 + y2 2
x6 + y6 + 3(x2 + y2 )
2
3 2
6
6
2 3
x + y + (x + y ) ,
4
=
we conclude x6 + y6
1 2
(x + y2 )3 . Now we have
4
V 0 (t) = 2(x(t)6 + y(t)6 ) 2 (x(t)2 + y(t)2 )3 (x(0)2 + y(0)2 )3 .

4
2
(x(0)2 + y(0)2 )3 > 0. Consider the function V (t) kt. Since (V (t) kt)0 = V (t)0 k 0, hence
2
V (t) kt is an increasing function. For t > 0, we have V (t) kt V (0) = x(0)2 + y(0)2 . Therefore
lim V (t) lim (kt + x(0)2 + y(0)2 ) = +. Given = 1, for any > 0, if x(0)2 + y(0)2 < ,
Take k =
t+
since
t+
lim V (t) = +, there exists N depending only upon x(0) and y(0) such that whenever t > N ,
t+
2
V (t) = x(t) + y(t)2 1. Therefore (0, 0) is unstable.
56
We can generalize the idea in the second part of the last example to get a test for instability of a
critical point. This will be discussed in the next section.
Exercise 2.4
1. Show that the function ax2 + bxy + cy2 is positive definite if and only if a > 0 and b2 4ac < 0.
2. Show that the critical point is strictly stable for the following system
dx
= 4x y,
dt
dy = 2x 5y 2y sin x.
dt
3. Show that for any a < 0, the critical point (0, 0) of the system
dx = ax + y + x2 y,
dt
dy
= x + ay cos x,
dt
is strictly stable.
4. Show that the origin (0, 0) is strictly stable for the following system
dx
= xy 2x2 y 3 x3 ,
dt
dy
1
= y x2 + x3 y 2 .
dt
2
5. Show that the the origin (0, 0) is strictly stable for the following system
dx
= 2(x3 + y 3 ),
dt
dy
= x 2y 3 .
dt
6. (1) Show that the critical point (0, 0) is stable to the following regular system
dx
= y,
dt
dy
= 2x 3(1 + y 2 )y
dt
by constructing a weak Liapunov function.
(2) Show that the critical point (0, 0) is attractive to the system.
2.5
A TEST FOR INSTABILITY
Theorem 2.5.1 Assume that there exists a function, U (x, y) defined in x2 +y2 k2 , a neighborhood
of the origin, satisfying that
(1) U (x, y) is continuously differentiable and U (0, 0) = 0;
(2)
d
U (x(t), y(t)) > 0,
dt
2.5 A TEST FOR INSTABILITY
57
for any solution (x(t), y(t)) of the system (2.1) with (x(t), y(t)) 6= (0, 0);
(3) in every neighborhood of the origin, there exists at least one point (x, y) such that U (x, y) > 0.
Then the critical point (0, 0) is unstable for the regular plane autonomous system (2.1) .
Proof By the condition, for each > 0, 0 < < k, there exists a point (x , y ) such that 0 < x2 + y2 < 2
and U (x , y ) > 0. Denote (x (t), y (t)), t t0 , to be the solution curve satisfying (x (t0 ), y (t0 )) = (x , y ).
d
U (x (t), y (t)) > 0, we know that U (x (t), y (t)) U (x , y ) > 0, for t t0 . Thus the solution
Since
dt
curve cannot enter the origin as t +, since U (0, 0) = 0. Let D0 be the circular disc in which is solution
d
U (x (t), y (t)) has
curve (x (t), y (t)) is bounded away. Then in the ring region Dk D0 , the function
dt
a positve minimum m:
d
U (x (t), y (t)) m > 0,
dt
for t t0 . This implies
Z t
d
U (x (t), y (t))dt m(t t0 ).
U (x (t), y (t)) U (x , y ) =
dt
t0
Let t +, we conclude that U (x (t), y (t)) is unbounded. Therefore the solution curve (x (t), y (t))
cannot remain in the disc Dk . So this solution curve must reach the boundary Dk .
Therefore, for any fixed , 0 < < k, for any > 0, there exists at least one solution (x (t), y (t))
satisfying
x2 (t0 ) + y2 (t0 ) < 2 but x2 (t) + y2 (t) > 2 for some t.
In other words, the origin is not a stable critical point.
Theorem 2.5.2 If the critical point (0, 0) of the linear plane autonomous system
dx
= ax + by,
dt
dy = cx + dy,
dt
is unstable, where the eigenvalues of the coefficient matrix are different, non-zero, and at least one
has positive real part, then the perturbed regular system
dx
= ax + by + (x, y),
dt
dy = cx + dy + (x, y),
dt
is unstable at the origin (0, 0), provided there exist k > 0 and M > 0 such that whenever x2 + y2 < k
we have |(x, y)| + |(x, y)| M (x2 + y2 ).
Proof Since the eigenvalues of the coefficient matrix of the linear system are different, non-zero, and at
least one has positive real part, for simplicity, by Theorem 2.3.1, we can assume that it is one of the types
dx
dx
= r x i y,
= 1 x,
dt
dt
r > 0;
1 > 0.
(1)
(2)
dy = i x + r y,
dy = 2 y,
dt
dt
58
We will show that there exist functions U (x, y) satisfying the conditions in Theorem 2.5.1, repectively.
In the case (1), consider the function U (x, y) = x2 + y2 . We only need to show that the function U
satisfies the condition (2) in Theorem 2.5.1.
Let (x(t), y(t)) be a solution of the perturbed system and write U (t) = U (x(t), y(t)). For any solution of
the perturbed system, since
dU (t)
dt
dx(t)
dy(t)
+ 2y(t)
dt
dt
2x (r x i y + ) + 2y (i x + r y + )
2r (x2 + y2 ) + 2(x + y),
2x(t)
By the condition,
|x + y| |x| || + |y| || (|x| + |y|) M (x2 + y2 ),
|x + y| 2
r
4M
2r
r
r
}, since |x|
and |y|
, we have
16M 2
4M
4M
M (x2 + y2 ) =
r 2
(x + y2 ).
2
Hence,
dU (t)
2r (x2 + y2 ) + (r )(x2 + y2 ) = r U (t),
dt
whenever x2 (t) + y2 (t) < .
In the case (2), consider the function U (x, y) = x2 /1 + y2 /2 . Since 1 > 0, this function obviously
satisfies the condition (3) in Theorem 2.5.1. Again, we only need to show that the function U satisfies the
condition (2) in Theorem 2.5.1.
For any solution (x(t), y(t)) of the perturbed system,
dU (t)
dt
dx(t)
dy(t)
/1 + 2y(t)
/2
dt
dt
2x (1 x + )/1 + 2y (2 y + )/2
2(x2 + y2 ) + 2(x/1 + y/2 ),
2x(t)
By the condition,
|x/1 + y/2 | |x| ||/|1 | + |y| ||/|2 | (|x| + |y|)(min{|1 |, |2 |})1 M (x2 + y2 ),
|y|
min{21 , 22 }
min{|1 |, |2 |}
and
}, since |x|
16M 2
4M
min{|1 |, |2 |}
, we have
4M
|x/1 + y/2 | 2
min{|1 |, |2 |}
4M
(min{|1 |, |2 |})1 M (x2 + y2 ) =
1 2
(x + y2 ).
2
Hence,
dU (t)
2(x2 + y2 ) + (1)(x2 + y2 ) = x2 (t) + y2 (t),
dt
whenever x2 (t) + y2 (t) < .
2.5 A TEST FOR INSTABILITY
59
Example 2.5.1 Determine the stability of the critical points of
dx
dt
dy
dt
y(x + 1),
x(1 + y3 ).
Solution There are two critical points for this system: (0, 0) and (1, 1). At (0, 0), the linearization is
dx
dt
dy
dt
y,
x.
The eigenvalues of the coefficient matrix are 1. Hence the origin is a saddle point for the linearized system.
By Theorem 2.5.2, the origin is unstable.
Near the other critical point (1, 1), we shift it to the origin by making
x = 1 + u,
Then (u, v) satisfies the following system
du
dt
dv =
dt
y = 1 + v.
(v 1)u,

(u 1) 1 + (v 1)3 .
The linearization at (0, 0) of this system is
du
dt
dv
dt
u,
3v,
whose eigenvalues of the coefficient matrix are 1 and 3. By Theorem 2.4.4, this perturbed nonlinear
system is strictly stable at (0, 0). So the original system is strictly stable at (1, 1).
2
Exercise 2.5
1. Show that for any constants a > 0 and b > 0, the origin is an unstable critical point for the system
dx
= x2 y + ax3 ,
dt
dy
= 2x3 + by7 .
dt
2. Let be a real number. Discuss the stability of the critical point (0, 0) for the system
dx
= y + x3 ,
dt
dy = x + y 3 .
dt
60

3. Assume that f (x, y) is continuously differentiable. For the system
dx
= y xf (x, y),
dt
dy
= x yf (x, y),
dt
show that
(a) the origin (0, 0) is strictly stable if f (x, y) > 0 in a neughborhood of the origin;
(b) the origin (0, 0) is unstable if f (x, y) < 0 in a neughborhood of the origin.
2.6
NONLINEAR OSCILLATIONS
In mechanics, differential equations of the form

x
+ p(x, x)
x + q(x) = 0
(2.10)
often appear in many different occasions. Generally, they describe the displacement x of a particle
of unit mass under a force system containing a conservative element, q(x), and a dissipative or
energy-generating component, p(x, x).
In this section, we consider the stability for two special but
common cases of this type of nonlinear oscillators.
2.6.1
Undamped Oscillations
When p(x, x)
= 0, the nonlinear oscillator (2.10) is said to be undamped. Then the above equation
is equivalent to the plane autonomous system
dx
dt
dv
dt
v,
q(x).
(2.11)
We will study stability about a critical point (x0 , 0), where q(x0) = 0. Without loss of generality, we
may assume x0 = 0, otherwise a translation of coordinates can move (x0 , 0) to (0, 0). If the origin
(0, 0) is stable, we expect that the restoring force q(x) must act in the opposite direction to the
displacement x in a neighborhood of x = 0, that is, xq(x) > 0 for x 6= 0 sufficiently small.
Theorem 2.6.1 If q C 1 and if xq(x) > 0 for small nonzero x, then the critical point (0, 0) of the
system x
+ q(x) = 0 is a vortex point.
Proof We define the potential energy integral:
V (x) =
q()d.
0
For any given positive constant E, the locus v2 /2 + V (x) = E is an integral curve, since we differentiate it to
dv
yield an identity: E = x[
x + q(x)] = 0. Obviously, the locus is symmetric in the x-axis. Since
= q(x)/v,
dx
2.6 NONLINEAR OSCILLATIONS
61
dv
dv
< 0; in the second and fourth quadrants,
> 0. For any
dx
dx
small value of E, the function E V (x) has a maximum value E and decreases on both sides of x = 0.
Since E is very small, as the value of |x| increasing, E V (x) eventually becomes zero. Hence, there are
two points x = B and x = A at which the function E V (x) is zero, where A, B are small and positive.
These imply that each locus is a simple closed curve, symmetric about the x-axis.
2
2
We can show that the integral curves of (2.11) resemble a distorted
p family of circles u + v = 2E for very
small E. To see this, we change of the coordinates by (x, v) ( 2V (x), v), according as x is positive or
negative. The transformation is of class C 1 and the Jacobian determinant
p

p
q(x)/ 2V (x) 0
= q(x)/ 2V (x) 6= 0,
det
0
1
p
if x 6= 0. As E 0, we know |v| =
2[E V (x)] 2E 0, so the closed integral curves shrink
monotonically towards the origin. Hence, the integral curves resemble a distorted family of circles u2 + v2 =
2E.
2
in the first and third quadrants, we have
d2 x
= x3 x, studied in Example 2.1.2. In
dt2
this special case, the restoring force q(x) = x x3 and the origin (0, 0) is a vortex point of this
system. Theorem 2.6.1 can also be applied to Example 2.1.3 for the simple pendulum equation
d2
= k2 sin . A direct application of the theorem implies that the origin is a vortex point. A
dt2
translation k + can be used to show that all critical points at (k, 0) are vortex points, where
k = 2, 4, . . ..
Now we can apply Theorem 2.6.1 to the system
2.6.2
Damped Oscillations
For the noninear oscillator (2.10), if p(x, x)

is not identically zero, we have a damped nonlinear
oscillator. Let us consider the case q(x) = xh(x) in this subsection. Once again we assume that
the restoring force tends to restore equilibrium under static conditions by assuming h(0) > 0. The
nonlinear oscillator is equivalent to
dx
= v,
dt
(2.12)
dv = vp(x, v) xh(x).
dt
The origin (0, 0) is always a critical point of this plane autonomous system.
Theorem 2.6.2 If p and h are of class C 1 , and if p(0, 0) and h(0) are positive, then the origin is a
strictly stable critical point of the damped nonlinear oscillator (2.12).
Proof The system (2.12) can be rewritten as
dx
= v,
dt
dv
= h(0)x p(0, 0)v + O(x2 + v2 ).
dt
Now the linearized system is

dx
= v,
dt
dv
= h(0)x p(0, 0)v,
dt
62
whose coefficient matrix is
h(0)
p(0, 0)
The eigenvalues of the matrix are
p2 (0, 0) 4h(0)
2
that have negative real parts if p(0, 0) > 0 and h(0) > 0. Hence, by Theorem 2.4.4, the origin is a strictly
stable critical point; the solution curves tend to the origin in the vicinity of the origin.
2
p(0, 0)
If p(0, 0) < 0, the system is said to be negatively damped, and the origin unstable. In fact, if we
reverse the time, the differential equation becomes

d2 x
dx
dx
p x,
+ h(x)x = 0.
d(t)2
d(t) d(t)
Hence, if p(0, 0) < 0, all solution curves spiral outward near the origin.
It is worth to notice that there exists a weak Liapunov function for the system (2.12). In fact, for
Z x
1
V (x, v) = v2 +
sh(s)ds,
2
0
it is positive definite for small (x, v) since h(0) > 0. Moreover,
V (x, v) X(x, v) +
V (x, v) Y (x, v)
x
v
xh(x) v + v (vp(x, v) xh(x))
v2 p(x, v) 0,
in a neighborhood of the origin. Hence, by Theorem 2.4.3, we know that the origin is stable.
However, this Liapunov function is not capable of detecting that the origin is strictly stable.
2.7
MISCELLANEOUS PROBLEMS
1. Construct a strong Liapunov function for the following regular system
dx
= y,
dt
dy
= 2x 3(1 + y 2 )y
dt
and justify your result.
Solution To construct a strong Liapunov function, we first try to make a linear equivalent transform
!
x
y
=K
!
u
v
where K is the matrix formed by two linearly independent eigenvectors of the coeficient matrix:
!
1
1
K=
.
2 1
63
Then we have
d
dt
!
u
v
=K
d
dt
!
x
y
=K
!
x
y
AK
!
u
v
!
!
u
v
Under the same linear transform, the perturbed nonlinear system has the form
du
= 2u + f (u, v),
dt
dv = v + g(u, v),
dt
where f (u, v) and g(u, v) are polynomials of u and v, with the lowest terms from at least the second
order. For the later system, by the discussion in the proof of Theorem 2.4.4, the function u2 + v2 is
a strong Liapunov function. Since the inverse transform gives u = x y and v = 2x + y, thus, we
have a strong Liapunov function of the original nonlinear system:
V (x, y) = (x y)2 + (2x + y)2 = 5x2 + 6xy + 2y2 .
To see that it is indeed a strong Liapunov function, we only need to check that
V (x, y) X(x, y) +
V (x, y) Y (x, y)
x
y
=
=
(10x + 6y)y + (6x + 4y) 2x 3(1 + y2 )y

2 6x2 + 8xy + 3y2 + 9xy3 + 6y4

6x2 + 8xy + 3y2 < 0.
Here the inequalities hold since 9xy3 + 6y4 is of higher order than 2 and 6x2 + 8xy + 3y2 is positive
definite. A similar argument can be made as in the proof of Theorem 2.4.4.
2

Theory of ODE - Hu

Uploaded by

Copyright:

Available Formats

Theory of ODE - Hu

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Theory of ODE - Hu

Uploaded by

Copyright:

Available Formats

Theory of

Ordinary Differential Equations

Jishan Hu and Wei-Ping Li

Existence and Uniqueness

Plane Autonomous Systems

2.6.2 Damped Oscillations

A normal system of first order ordinary differential equations (ODEs) is

X1 (x1 , . . ., xn; t),

F (t, x1, . . . , xn).

1 EXISTENCE AND UNIQUENESS

We have the following useful inequality

1.1 SOME BASICS

whenever (x, t) R and (y, t) R.1

is worth to point out the following:

are some explanation of some concepts and terminologies in the Lemma.

1 EXISTENCE AND UNIQUENESS

|Xi (x + y, t) Xi (x, t)|2

The Lipschitz condition follows, with the Lipschitz constant nM .

Example 1.1.3 Determine whether the function

Thus, the function X is a Lipschitz function in R4 , with L = 2T .

1.1 SOME BASICS

But the derivative is not bounded, since X 0 (x) as x 0+.

Solution Let x1 = x, x2 = x0 , x3 = x00 , we get the equivalent first order system

max{4M 2 , 1}, then whenever (x, t), (y, t) R, we have

L2 (x2 y2 )2 + L2 (x3 y3 )2 + L2 (x1 y1 )2 = L kx yk .

1 EXISTENCE AND UNIQUENESS

In this section, we prove the following uniqueness theorem.

that satisfies a given initial condition x(a) = c in R.

2 [Xk (x(t), t) Xk (y(t), t)] [xk (t) yk (t)]

2 [X(x(t), t) X(y(t), t)] [x(t) y(t)] .

2 kX(x, t) X(y, t)k kx yk

The last inequality implies

1.2 UNIQUENESS THEOREM

Thus, the function X(x, t) satisfies a Lipschitz condition in [x0 , x0 + ] R.

1 EXISTENCE AND UNIQUENESS

which yields the desired inequality.

This implies the desired inequality in the case t a.

Let t be fixed. Hence, for any given  > 0, take =

Theorem 1.3.3 (Strong Continuity Theorem) Assume that

for any a, t [T1, T2 ].

[xk (t) yk (t)]2 0. We have

2 [X(x(t), t) Y(y(t), t)] [x(t) y(t)]

The Schwarz inequality implies

2L kx(t) y(t)k2 + 2 kx(t) y(t)k

function Y(y, t) may not satisfy a Lipschitz condition.

1 EXISTENCE AND UNIQUENESS

2L[(t) u(t)] + 2[

Integrating the last inequality from a to t, we have

1.4 EXISTENCE THEOREM

and a similar argument to prove the following inequality

Using (1.4), we have the integral equation (1.5).

1 EXISTENCE AND UNIQUENESS

For this operator, its domain is

This is call a Picard iteration.5

Picard iteration is started from x0 = c, given by the initial data.

1.4 EXISTENCE THEOREM

Assume that the desired inequality holds for n = k:

Hence, the estimates (1.8) hold.

1 EXISTENCE AND UNIQUENESS

is also convergent. Actually, the convergence is uniform for 0 t T , since

Let t be fixed. Hence, for any given > 0, take =

2L kx(t) y(t)k2 + 2 kx(t) y(t)k

2L[(t) u(t)] + 2[