5

Existence and Uniqueness

of Solutions

5.1 INTRODUCTION

existence of solutions
differential equations are the
that basic questions in be mean-
Itis known
without which their qualitative properties
may not
and their
uniqueness, existence theorems for
nonlinear
is to deal with certain
in this chapter
ingful. Our
aim
some results for
uniqueness of
vector form. We also present
scalar and a
VPs in a
solutions of IVPs.

5.2 PRELIMINARIESs

the Lipschitz condition,

of functions satisfying
consider a class Its
In this section
we
theory of differential equations.
in the qualitative
which plays an important role and continuous depend-
the existence of a unique solution
applications showing sections of this chapter.
conditions are dealt
with in the ensuing
ence on initial

defined in a region
DcR is said to satisfy
Definition 5.1 A function ft, x) constant K, if the inequality
a Lipschitz
condition in the variable x with
Lipschitz (5.1)
denote f to be a member
a case we
in D. In such
holds whenever (t, x,), (1, 1,) are

of the class Lip (D, K). condition

definition, a function f(t, x) satisfies Lipschitz
AS a consequence of the
K>0 such that
exists a constant
and only if there

whenever (t, x). (t, x,) belong to D.

criterion which would
ensure

The is to find a general

question which may arise the existence of a typical
theorem shows
following closed
pschitz condition. The we assume
the region D to be
a

Such functions. For simplicity,

T
ectangle.
 Equations
Ditferential

Textbook of Ordinary
130 defined over a

5.1 Let ft. x)

be a
continuous
function

are
some
real numbers Le
positive re clangle
Theorem Here p, q
S q}
Ir- ol Sp. It- Xol
R= f0. ): satisfies the Lipschitz conditic inR
r)
c o t i n u o u s on
R. Then f(t.
defined and
he
constant A such

af exists a positive tha

continuous on R there
Proof Sineis
d af . SA (5.2
in K. 1hen by the mean
be any twO points value
Let (1. x,). (t. xy) which lies between
Tor all (t. x) in R. there exists a number s and
of differential calculus.
theorem

A such that
af
,)-0. I)= 54, -Xz.
(5.2) holds, it is seen that
R and the inequality
Since the point (t. s) lies
in

af
S
dx
A.
Hence we have
ift.x)-ft.x,)| SA\x -Xl
whenever (1. x). (t. Xz) are in R. The proof is complete.
illustrates that the existence of partial derivative of fis
The foilowing example
be Lipschitz function.
necessary for f to a
not

Example 5.1 Let ft. x) =

|x| on the unit square R around the origin, namely.
R=(.):t|S 1. t|s 1).
The partial derivative of f at (t, 0) fails to exist but f satisfies Lipschitz condidon
on R with Lipschitz constant K = 1.
The example below shows that there exist functions which do not satisfy the

Lipschitz condition.

Example 5.2 Letft. x)= " be defined on the rectangle

R= {(t.x): 1t| S 2, x|S2).
Then f does not satisfy the inequality (5.1) in R. This is because

0) X#0,
is unbounded in R, since it be made close

to zero.
can as large as possible by choosing

Gronwall Inequality

The integral inequality, due vera

to Gronwall, plays a useful part in the study o

 Existence and
Uniqueness of Solutions 131
f
p r o p e r t i e s

f ordinary
o fo m

difterential
uniqueness
equations.
of solutione
nIn particular
particular, we pose tto
propose employ it
ablish
cSstablish the ons
E2 Assume that f(0) and
eorem g{t) are
non-negative continuous functions
Theo etk>0 Let
k>0 be a
be constant. Then the
inequality
fort2lo
f0Sk+) a() fo) ds, 12
inplies the inequality

f0 skexp|, &) ds.12

hypothesis we ave
Proof By

-S g(). 121

f) 8) is the derivative of k+ 8(s)fo) ds, integration of this

Noting that
between the limits to t, leads to
inequality

logk+s)s) ds|-logks| 0
s) ds

or, in other words,

k+s0f) da Skexp|st)d
conclusion.
the desired
with the hypothesis leads to
This inequality together

I2 1o»
Corollary 5.1 If, for

for 12 1.
Theorem 5.2 hen, f{)
=

wneref andk are as given in

t21 and any e>0
clear that for
Proof From the hypothesis it is

theorem yields
ne application of the above
<E exp k(t- 1),
12
f)
f) =0 for t21o
Lett 0 . This leads to the fact that
5.3 SUCCESSIVE APPROXIMATIONs
We begin with an initial value problem

=ft, x). xt)=o (5.3)

where f(t. x) is defined and continuous in (t, x) on an open connected set D. Also
iett, o) be in D. Geometrically speaking, solving (5.3) is to find a functonx the
graph of which passes through (t.xo) and the slope of x coincides with fi, x)
whenever (1, x) belongs to some neighbourhood of( ). This initial value problem
lurtishing a solution around (o o) is -called the local existence probiem for an
Lal value problem. At this stage it is to be remarked that usual elementary
PrOocedures of determining solutions may not work for a general class of problems
he (5.3). It appears that a general type of infinite process is required to construct

UOn X Of (5.3). This is where the method of successive approximations finds

a application.

Ihe procedure that is adopted in solving (5.3) is important and involves analytic

COnsiderations. The key

Craions. to is very much dependent on the
the general theory
equivalent representation of (5.3) by the integral equation
(5.4)
a()o ,40) ds.
EAuation (5.4) is called an integral equation
because the unknown function x{)
 Equations
Ditferential

Toxtbook of Ordinary
134 the

sign. The lenmma

below
establishes

valence o
equival.

occurs
under the integral
(5.3) and (5.4).
interval 7 if and only iiff .x) ina
on s o m e
solution of (5.3)
5.1 x(1) is
a
Lemma
solution of (5.4).
show ihat x{) satisfies (e
(5.4
then it is casy to

Proof If ( ) is a
solution of (5.3) Dilerentiating both side
Pmof
solution of (5.4).
Obviously x{)=o
Let x() be a it is seen that x(t)=f .
continuous in (1, x),
x) is
(5.4), and noting thatl f(1,
the proof. solution of (5.3). Firest of
which completes t0 a
define certain approximations
Now we are set to
solution and improve
it by iteration .It is
with an approximation to a
in the limit
all we start
converge to
a solution of (5.3) he
expected that these iterations mention th
In this connection, we
(5.4) now springs up.
importance of Equation rather than derivatives
be handled casily with integrals
the estimates can
to a solution of (5.3) 1s Just the constant function
A rough approximation
in the right hand
a better approximation by substituting xg)
( ) =*oWe may get
a new approximation x() given by
sides of (5.4), thus obtaining

0=o+ f6.Mp) ds.

To get a still beter approximation we repeat the process thereby defining

)= *+ fo.x,()) d.
In general,

0)+ fs.,-())ds, n= 1,2,. (5.5)

This procedure is known in the literature as "Picards' method of successive

approximations". We conclude this section with a few examples. In the next section
we show that the sequence {x,()} does converge to a unique solution of (5.3)
provided f(, x) satisfies the desired condition.

Example 5.3 Consider the IVP x = -x, x (0) = 1, t20. It is equivalent to the
integral equation
x(t)= 1 -J xls) ds.
The first
approximation is given by x(0) = 1. The second approximation is
x)=1-J.xgs)ds= 1-t.
By the definition of successive approximations, it follows that

,0)=1- (1-s) ds =1 -|
 Existence and
Uniqueness of Solutions 135
ral, the (n +1) th apprOximationis
Inpcner

,(1) = I.
(-1
i ue here that x,() 1s the (n + 1
)th
We ecogtnize

ecthate is the solution of partial sum of the series for e'. It

the is
IVP under consideration.
y lo note

5.4 Consider the

Example 5,.4

the integral equatio

IVPr=x,x (0) =1. The cquation is equivalent to
()= 1+) r() ds.
he first approximation is gt) = 1. Now

)=1+J 1ds = 1+
(0

O=1+0 +9 ds =1+1+f+
3
3
)=1+ 1+s++ds=1+t+P+t+
All 1,0, n =0, 1, 2,. . are polynomials.
Ohserve that the IVP can be solved explicitly by the method of
variables. Here separation of

isa solution existing on -o<t<1.

EXERCISE 5.2
1.Calculate the successive approximations for the IVP X = 8), x(0) = 0.
2 Solve the IVP X =x, x(0) 1 by the method of successive approximations.
=

Compute the first three successi ve approximations for the solution of the
following
equations
9 = , x(0) = 1;
i) X=e, z(0)
=0;
m)= . {0)
x0)== 1,1.
I+2'

5.4 PICARD'S THEOREM

With all
ethe
effect
remarks ano
remarks and examples, the reader may have doubts about the

SSand utility of Picard's method in practice. However we mention that

136 Toxtbook of Ordinary Differential Equations

this method has made a landmark in the theory of differential equations, I

not only a method to determine an approximate solution subject to a givene
but also establishes the existence of a unique solution of initial value pro err
under general conditions. roblems
In all of what follows we assume that the function f(1, x) is bounded by .
satisfies the Lipschitz. condition with the Lipschitz constant K on the closed recta
R= (0. ) e R' : | - ol Sa.lx - xol S b, a > 0, b> 0).
tangle
Before proceeding further, we have to show that the successive approximations
given by (5.5) are well-defined on an interval I. That is, to define on it
is necessary to show that (s, x,(5) lies in R for cach s in I and j z I.

Lemma 5.2 Let h = min |a. Then the successive apprOXimations given by

(5.5) are valid on l=| !- 10lSh. Further

-ol SL|t- tgl Sb, j=1,2,....te l. (5.6)

Proof The method of induction is used to prove the lemma. Since we start with
any point (to» x) in R, it is clear that xot) = *o satisfies (5.6). Now assume that.
by induction hypothesis, for any j=n >0, x, is defined on I and satisfies (5.6),
Hence (s,,(s)) is in R for all s in I. Therefore is defined on I. Because of
the definition, we have

10)=+] .x,())ds, te.

Using the induction hypothesis, it is seen that

0 - l = |, s. x,()) ds| s, if6. x,()|dssL[r-%1Sh Sb.

Thus x satisfies (5.6). This completes the proof.
We now state and prove the Picard's Theorem, a fundamental result dealing with
the problem of existence of a unique solution for a class of nonlinear initial value
problems.

Theorem 5.3 (Picard's Theorem) Let ft. x) be continuous and be bounded by

condition with Lipschitz constant K on the closed
L and satisfy
R. Then he
Lipschitz rectange
successive approximations x, n=1, 2,.... given by (5.5) convc
uniformly on an interval I = || t - t% ll Sh, h = min (a, biL), to a solution oftheiv
(5.3). In addition, this solution is unique.
Our
Proof We know that the IVP (5.3) is equivalent to the integral Equation ( .
tion
aim is to show that the successive approximations x, converge to the unique so
of (5.4) and hence to the unique solution of the IVP (5.3). First, note that
 Existence and
Uniququeness of Solutions 137
x,0)=xol)+ 2 x{0-x,-(01
artial sum of the series
t h en t h

ol0)+2 k)-x,- ,(91

(5.7)
H e n e

anvergence of the sequence (X,is

t h e c o n v e r p

equivalent to the convergence of

7). We complete the proof by
t h es e r i e s

showing that
ceries (5.7) converges unitormly to a continuous
a) function x():
s a t i s f i e s the integral Equation (5.4);
(b)
Is the unique solution of (5.3).
(c)
with we fix
a itive number h min
(a, b/L). Because of Lemma 5.1 the
=

tart
Tos
Successive
ximations x(), n = 1,2,... in (5.5) are well defined on
approxim

we stick to the interval

sj=-4ol h. Henceforth, SiS+h.. The proof on
Ihe proof on the

h , o] is milar except for minor changes.

jnterval l- us denote
We estimate )-x{0) on the interval o,o+h]. Let condition,
1,2,... Since f(t. x) satisfies Lipschitz .
by
0=1510-x{0kj=

successive approximations, we obtain

of
the definition
m00=LS(6.s40)-f6.4-0Md|
SK E)--16)|ds
inother words, (5.8)
m)SR m- (6) d
By direct computation,
*o0=|f6.o) ds
ma0 =la6)-

sI6omd (5.9)

SUt-)
(5.10)
We assert that -4) induction.

m ns L K '
j +1)! follows by assertion
assertion
the
proofof
the integerj=p2I

The
thatforan
for j=0, 1,2, and io
S I s i n + h .

(5.9).
Assume

hen j = 0, (5.10) is,

ien0 In fact,

D.i0) is valid. Therefore,

138 Textbook of Ordinary Differertial Equationss

6
s K m,(s)dsSK[ LK
(P+ 1)! ds
)
(7-1,)P* 2
S LKP1 oS1So +h,
(p+ 2)!
whh shows that (5,10) holds when j =p+ 1. Thus (5.10) holds for all k> 0

K'
the series m) is dominated by the series
K (j+ )
-

which converge
Lu - DK. Hence the series (5.7) converges uniformly and absolutely on
interval % S I S + h. Let
the

= o ) + 2 I,()-x,- ()1: %Sish+h. (5.11)

n=l
Since the convergence is uniform, the limit function x(1) in (5.11) is continuous
ofohl. lt is easy to show that the points ( , x() are in the rectangle R for all
tel. This compleles the proof of (a).
We now show that the limit function x(t) satisfies the integral equation

x()+, S6. x()) ds. (5.12

By the definition of successive approximations

x,)= * +J, S6. , - ) ) ds. (5.13

In view of (5.13), we have

t)-o-J Ss, a()) ds

o

M)-,()+J f6.x,- ) ds -J fo. x) ds|

S
x()-,(DI+J Isls, x,- 1))-f(s, x(s))| ds. (5.14)

Since x,()> a() uniformly, and Jr,() - xol Sb for all n and for in [o. o+
follows that Jr(/)- l Sb on
o» to + h]. Using the Lipschitz condition in (.i
is seen that

)- S6. a(s)) ds| Sla()-x(0)1+K} Ix()--1) ds

(5.15)

S)-x,(O1+ Kh max
x(s) --1)
SsS+h of
The uniform side
convergence of x,() to x(t) now
implies that the right nai
 Existence and Uniquenoss of Solutions 139

S n , But the lett side of (5.15) is independent of n. Thus

as
n - o

Cadstozero
aMIstie
he
ategral Equation (5.4). This proves (b).
that if
ve that if ( ) and x(1) are any two solutions of the IVP (5.3). then
prove
now and
thEY&0DCde o n l. lo
+h]. ) .
satisfy (5.4). Therefore
ide

)-Ms I%.i) -f.x)|dr. (5.16)

4 r(s) lie in R for all s in | , o+ h] and hence it follows from (5. 16) that
x(s)
and
a19)

Roth

)-(01SK )-«(|d«.
safion of.
application
of the Gronwall inequality, we arrive at
the
8y )-x()1=0 on lo» o +h]

means
ans x{t)
+) ={)
= x{ on ooth). This proves (c), completing the proof of the
which

feature of Picard's theorem is that

h e o r e m .

important a bound for the error in

ner
Aof truncated computation at the n-th iteration can also be obtained. The
the case o f
thal follows is a result dealing with such a bound on the error.
theoremthat follows

The error x{)

-

x,(0) satisfies the estimate

Theorem 5.4

-,(1
LKh h 1e [to+h (5.17)
K(n+1)!

Proof Since x)=x+ 2 .)-{). we h

j=0

x)-,0)= 2 ,0-(0)
j=n
he above relation implies, in view of (5.10), that

0)-x,01 -1)-x(O1S 2 m{)

In JEn

s y LKhy*_LKh)* (Kh)
j=n
Kj+ 1)! Kn +1! n+2)... (n+j+ )|
sUKh"+
Kh te loloth
K(n +1)!
is (5.17). The proof is complete.
tample 5.5 Consider
nsafof
Cons the IVP =x, r(0) =
1; t20. Observe that all the condi-
the Picard's
theorem satisfied. To find a bound on the error
AU)-x) we have determine are
Let R be the
K and L. It is quite clear that K=I.
 Equations
140 Toxtbook of Ordinary Differential

closed rectangle around (0, 1)

i.e. R ={(1, x) :|
ST and .r - I|S) T en L=
E. The question is to find a
not to exceed
andh 1 . Suppose
nsuch that ix -x.I
the error

Se. To achieve
is

this, a sufficient condition is that number

LKh Kh < E.

K(n+)!
We have find such that
<
Eeor, in other words, n +
1)! > e.
to an n
(n+
(n+ 1)!°
1)!
be achieved since e'e is finite and (n+ 1)!>**.
1)! ->, Thus, if
Th.
This inequality can

E=1, we can choose n 2 2, so that

the error is less than 1.
 Exis and
Example 5.6 Consider the IVP x Uniqueness of Solut
Hutions 141
rst interval of existence of its
af xt =
cos 1, x(0) 0.
R (0.
solution. Lct R be
the
We try =

determine the to
x). 0SIsa, IriSb, rectangle containing (0, 0)
Clearly
1f. ) f =|+ +cos'<
=L azb>0i.
xx+cos'
function S(1. x) = x t
satisfies
= Lipschitz's condition
d 2 s 2h R. since
K. on

We find that x(t) exists for

0SIsh mina, b

Ohserve that the maximum value of b i Hence h

.ie d) exists on

the interval 0Srs

Evample 5.7 Consider the IVP x=x, x(0) =2. Let R be the rectangle
R:(0, x) Itl Sa, lx -21Sb, a>0,
b>0}
tnR.1f.x)=\|s(b+2)* =L and the interval of existence of solution is
1Sh. where h = min |a, b Hence the solution of the IVP exists on the
(b+2
T-interval-

However, we observe that this IVP can be explicitly solved. Its solution is
2
x)1-2
We find that x(t) éxists on -oo <1 < This interval of existence is much larger than

that obtained by the application of Picard's method.

drop this condition?
The Picard's theorem assumes Lipschitz condition. Can
we

The answer is no. The following examples illustrate this point.

Example 5.8 In the case of the IVP

Y 0; x{0)=0 5.18)
0, r=0.
closed rectangle
condition in any
satisfy Lipschitz
E
t, x)= does not
shows that z,(0) =0 for
Cont successive approximations
The method of possessesanother
g (0, 0). Yet the givencquation
n 0,1,2,... Hencex(/)
.
= lim x,() =0.

solution x() =? existing on t>0.

 Differential Equations
Textbook of Ordinary
142

of the IVP
In the case

Example 5.9
4 r . x(0)=0, t20,
4"" fails to satisfy Lipschitz condition. Each
ach successiv
againflt. ) =

O on [0, o]
a{/) =

tion ,() = 0 and hence

We observe that x(/) = f Is yet another solution of the
given P. approkim
0sISc; n
faca
x,0)(- c'. cSt<oo,
for each real value of c is a solution of the given IVP. Thus
Thus, we
get uncou
solutions to the IVP.

Example 5.10 Consider the TVP=X, x{0) = 1,120. Let

rectangle be the cloe
R: {(. x). It|S 1, |x-1|S 1}.
Here K= 1, L =2, h = 1/2. We determine bound on the error
function
Suppose that error is not to exceed E. Then in view of
(5.17), we have
)-x,()Is UKh"*1
x(-x,)=K(n+1)!
1.e
2 1)ee-2
Ciearly such n can be calculated. For
example if e=
10023. n23.
Thefollowing example illustrates the fact that the
be satisfied: yet uniqueness of solutions Lipschitz condition may nt
may prevail.
Example 5.11 Consider the
OSIS and 0
xS1. IVP=x+1, x(0) =0, on the recange
t is
easy to show that
We show that flt, x) =x+1 does not satisfy the
there are two
there is a
unique solution to the IVP. Lipschitz condiuotha
interval 0
solutions xf1) and Suppose
y() on 0StsI such that
on the
conuraiy
are not qual in the
m1,) 0. SSIShS1. Set m(t)
(2-y',
they
= equa
clearly m(0) =0. ASSu
m't)= (2-y2
=2-y72
y2

-y -)
Hence 0
(xy)
yielding m) S
one
a() =y() on
solution SIS
m(t)s0, Stsh
to the This is ony

given IVP. contradiction and,

a
contradiction therefore,
there i